This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
= Y7(p))
(1.17)
(1.18)
= (27t)3<53(p' - p). = X '\x')ai(p'{x') = Rijcp^{x)ajq>{x), (4)(P3)=(o) 0)(A) and given by (3.42): DU>(A(K)) = e-' e - J0) , and %, respectively. The direct-product representation (§, 0) + (0, i) has a carrier space of four-component "Dirac bispinors" \j/ = (£) upon which the 4 x 4 matrices @(i) operate. Such freeparticle momentum-space wave functions can be formally expressed in terms of the boost as ?(p) = D(L„) = 2£(t 3 (p'-p)5 (J . ff , , must obey V p < ^ > ( p ) = V'>(p), = \ d3x >'*(xK = 2E83(p' — p), follows from (4.14b) with the replacement apK-*-*2£S3(p-Pi) *\id^ - eAjcp = i0*5„cp - 2e<j>* U (r). 2r _ r dr r r J (4.30a) , current conservation implies, for p'2 = p2 = m2, qf' = «(p')y,«(p). which is also conserved in the sense of (5.79). As we have already noted in Section 4.C, such off-diagonal momentumspace currents play an important role in the theory in their own right. Consider first a useful identity called the Gordon reduction. Defining P = \{p' + p),q = p' — P and using the y-matrix property (5.31a), it is easy to show (Problem 5.8) that K*qv = P% + Y,P~2Pfl. (p(p). (5.91) Thus Dirac Zitterbewegung is nonvanishing to zeroth order in momentum. While unphysical consequences can be avoided for free particles by choosing either a positive- or a negative-energy packet as in (5.76) or (5.80), for bound states this cannot be done. We return to this latter problem shortly. (\V) • \ = -MP))V(1 + iys)MP')- = ^(9 2 )(p' + p i - = - eF(q2)(p' + p)„, 2 = eu(p')r,(p\ PHvl r>', = ( - e ) " c ( P T > ' , P)"C(P) 1 R and cpL in (5.17), as only p changes sign in in the latter case: (E-H0)\^) = V\^y, = = [E - p 2 /2m]- *J3(p' - p). (7.7) Transforming (7.7) into coordinate space, we obtain G0(r,r';£)= , we can write the scattering amplitude/"1" as the Fourier transform of the potential /P-,P=/ , , = (id, - H) | = 0, (id,. - H0)G0(t' -t) = (id,. - H)G(t' -t) = 5(t' - t). satisfying (7.7). Its space-time Fourier transform can be found directly from (7.27), giving for the free-particle propagator (Problem 7.4) iGZ{x'; x) = 0(f - () + G0+(E)F|,MO)>„ , , |#>>}. = = <53(P' " P) = = ( = - £ - £ (21 + l)[S,(p) - 1]P,(P' • P ). imp , (8.9) and £ t + a> = E. That is, E{ = £ t + a>, Ef = E2, and energy conservation implies Ef = Ei. This time-dependent Born approximation is represented graphically in Figure 9.2(a). The potential, = 2£<53(p' — p), it is convenient to define the relativistic (but not always Lorentz-invariant) flux as & = 2£ 2 = 2£<53(p' — p) [because 33(p) has dimensions m - 3 from | d3p S3(p) = 1], we have for n-particle reactions dim S}°? = wT", (p')[(£ + m) + (E - m)a • p' = e M ^ i M , + F2(*R*Av/2m]«p, = M W K = pP{QA{q2)iy^5 + M ^ f a ^ s K + [-fe ILPMP = «„y«(l - iys9A)Un^2rnNCa = -\ til K* then has 9K* -Lp = 9K* Kn = 9p according to the SU(3) extension of CVC and vectormeson dominance (12.58). The Feynman diagram of Figure 13.5 then leads to the pv matrix element [T1 + 12 -*• 2 in (13.67)] - Kn°p | J f " 12 + > = ( - i)i{^ = (J^^Mptt!, = p ' ^ + P^P: - nM • P - ™2)> = < - p ' | T v J - p > = ,
(1.25)
To streamline the notation a bit, we parallel the convention h = h/2n and define 3 3 3 d3p = d3p/(2n)3, (1.26a) ; t (p) = (27r) «5 (p), for then
J>P* 3 (P) = J>P<5 3 (P)=1-
(1.26b)
A similar convention will be used in n dimensions, with the crossed notation referring to an appropriate factor of (2n)n associated with 8 "(p) and d"p. This notation becomes extremely useful if we also normalize plane-wave solutions for a free particle of momentum p in a box of volume V,
(L27)
8
Introduction
for then the probability of the particle being somewhere in the box is unity. We can, of course, identify (1.27) with (1.23) by setting 7 = 1 . The advantage of using (1.27) will be apparent when calculating physical rates, lifetimes, or cross sections. The normalization volume 7 cancels out of the quantity of physical interest; any unphysical quantity will vanish or become infinitely large as V -*• oo. If we had employed the continuum normalization with periodic boundary conditions, then 7 -»(2n)3, and typical scattering amplitudes would contain as many as 12 powers of 2n, sometimes a bewildering situation indeed! Use of box normalization goes a long way toward "keeping the factors of 2n straight". One further modification, covariant box normalization, will be useful for relativistic particles. It will be discussed in Chapters 3-5 and used extensively in the latter part of this book. Selection Rules. Returning to the discussion of angular-momentum transformation coefficients, we observe that if the hamiltonian is independent of Lor depends only upon L2, then (1.5) implies i ^ = [ L , t f ] = 0,
(1.28)
i.e., L is conserved. As a consequence of (1.28) we have 0 = {I'm | [H, L2] | lm} = [/'(/' + 1) - /(/ + l)]'m | H | lm}, 0 =
(1.29)
The scattering operator S, called the S-matrix, is the scattering analog to H in the sense that energy conservation (magnitude of momentum conservation in the one-body nonrelativistic scattering case) and angular-momentum conservation lead to a selection rule similar to (1.29): {p'l'm' | S | pirn} = S,(p)<5(p' - p)dndm.m.
(1.30)
We shall have much more to say about the S-matrix later. Note now that the "reduced matrix element" S,(p) is independent of the L3 eigenvalue m; this will be verified in Section 2.C. For the moment we may treat (1.30) as another transformation coefficient like (1.15) or (1.17)—(1.20) and thus compute (suppressing summation indices and integrals)
0)3«5(/ - p) 4np2
X(2/+l)S,(p)P,(p'-p).
(1.31)
The factor (2n)3 in (1.31) is consistent with our momentum normalization. When S^p) is expressed as a phase exp (2i5t(p)), the relation (1.31) is referred to as the partial-wave or phase-shift expansion. It is hoped that the reader
Unitary Operators and Transformation Theory 9 appreciates the transparent simplicity of (1.31) when the transformationcoefficient technique is applied. A similar pattern is valid for other operator eigenfunction expansions.
l.C Unitary Operators and Transformation Theory While no mention of unitary operators was made in the postulates of quantum mechanics in Section l.A, they nevertheless play a significant role in the theory. If two observers view the state of a quantum system as 11/^> and 11//>, respectively, then they must of necessity measure the same quantum probability. The "off-diagonal" version of this statement is that for |^> and |i/f'> to have the same probability overlap to the respective states <(/>| and (
I W > | 2 = |<0|*>|2.
(1-32)
Unitary Operators. There are two solutions of (1.32), the obvious one being
< W > = <0|*>.
(1-33) t_1
This leads naturally to a unitary operator U = t / (where the adjoint operation * corresponds to transposition and complex conjugation of finitedimensional matrices) which transforms the state |i//> to the state |i//> as m
= U\
(1.34)
The relation (1.33) then follows from (1.34), the unitary property of U, and the adjoint property (cf>'\ =(Ucfi\ =
(1.35)
The latter is linked to the linearity of U, [ / | a 1 ^ 1 + a 2 ^ 2 > = a1C/|iA1> + a 2 [/|^ 2 >,
(1.36)
where a l j 2 are complex numbers. Antiunitary Operators. The second solution of (1.32) is < W > = <#|*>*
(1.37a)
= <
(137b)
The solution of (1.37a) is m
= A\^y,
(1.38)
where A is again unitary, ArA = I, but now "antilinear", satisfying instead of (1.36), >4|«1^1+«2^2> = « M | ^ 1 > + « ! ^ | ^ 2 > -
(1-^)
The antilinear analog of (1.35), consistent with the solution (1.37a), is W\=(A4>\=(\A\
(1.40)
10
Introduction
where A operating to the left corresponds to <0|.A|*> = O4ty|*>*
(1.41)
and the dot in .A is sometimes omitted. In effect we can replace the unitaryantilinear operator A (henceforth referred to as antiunitary) by a purely unitary operator U (now meaning unitary-linear) and a complex conjugation operator K, i.e., A = UK. It is also possible to define an antiunitary operator in a slightly different manner, following the form (1.37b) rather than (1.37a), but we shall not pursue this possibility in detail. Transformations. The active interpretation of (1.34) or (1.38), without reference to any observers, is that U (or A) represents the physical transformation which converts | ij/} to 11//>. As such, U (or A) is the primary quantity of interest in transformation theory. In particular, we can investigate the action of U (or A) on operators by generalizing (1.33) and (1.37a) to <*'|O'|tfO = < 0 | O k >
(1-42)
for unitary transformations and <0'|O'|i/O = <>|°l'>* (!-43) for antiunitary transformations. Owing to the properties (1.35) and (1.40), (1.42) and (1.43) imply the same operator behavior WO'U = 0,
A^O'A = O.
(1.44)
The unitary property of both U and A then leads to 0' = UOU~ \
O' = AOA~1
(1.45)
in the two cases, both equations taking the form of similarity transformations from 0 to 0'. It may be recalled that such similarity transformations as (1.45) leave the form of operator equations unchanged except in the case of complex numbers in the latter equation because the antilinearity of A means complex conjugation of "c-numbers", as in (1.39). For example, the commutation relation \X, Y] = iZ would become [X', Y'] = —iZ', for antiunitary transformations. A case in point would be the behavior of the commutator [x, px] = i under various transformations, thus determining the unitary or antiunitary nature of those transformations. Next, consider the case where the operator O is left invariant by the transformation, i.e., O' = O in (1.45). One may then conclude that [[/, 0] = 0.
(1.46)
However antiunitary operators will not obey (1.46), because, as we shall see, invariance under an antiunitary operation will not mean 0' = 0. Infinitesimal Generators. Let us now consider in more detail unitary (unitary-linear) transformations dependent upon a small parameter e and
Translations in Time
11
evolving continuously from the identity. In this case we may write Ut=l-
ieG,
(1.47)
where G is called the infinitesimal generator of the transformation. Clearly a unitary Ut = Uj~l implies an hermitian G = G1. Applying (1.47) to the operator transformation 0' = UOU1 leads to 0=0-
ie[G, 0],
(1.48)
so that if G commutes with 0, then 0 is left unchanged by the transformation, 0' = 0. If 0 as well as G is hermitian, both operators could be physical observables and [G, O] = 0 would then imply that G and 0 are simultaneously measurable. For noninfinitesimal unitary transformations (e->A), (1.47) can be exponentiated: Ux = e~UG.
(1.49)
By use of the operator Taylor series expansion e-ixGOeaG = 0 - a[G, 0] + ^^~
[G, [G, 0]]
+ ^ ) 3 [ G , [ G , [ G , 0 ] ] ] + -",
(1.50)
which follows by differentiating the left-hand side of (1.50) with respect to A, the form Ux OU\ can be evaluated and 0' is thus generalized from (1.48) to (1.50) for noninfinitesimal values of the parameter L
l.D Translations in Time Time-Translation Operator. The dynamical development of states governed by the Schrodinger equation can be recast in the form of a unitary transformation. For time-dependent states (d, = d/dt) (idt-H)\+(t)>
= 0,
(1.51)
a time translation of | \j/(t)} means, according to (1.34), that we may define a time-translation operator by
|#')> = ^.0I^(0>.
(1-52)
1
where U(t', t) is unitary, U~ = U\ Moreover U obeys the fundamental "closure" property, V(f, t) = U(t', t")U(t", t),
(1.53)
stating that two successive time translations constitute a time translation. In order to determine the explicit form of U, one notes from (1.52) that
12
Introduction
U(t, t) = 1. For t' = t + St, it is therefore clear that U(t'; t) evolves continuously from the identity, and following (1.47), one writes U(t + dt,t)=l-
iH 8t,
(1.54)
where the infinitesimal parameter 5t in (1.54) is required by (1.53). Then the infinitesimal generator of time translations can be identified as the hamiltonian of (1.51). In fact, (1.54) indicates that U itself satisfies a Schrodinger equation (id, - H)U(t, t0) = 0.
(1.55)
To proceed further, one must make a dynamical assumption concerning H. If the physical states are isolated between observations, then H is independent of t and (1.54) can be exponentiated to U(t',t) = e-ii'-')H.
(1.56a)
If instead the system is not isolated between observations, then H may be explicitly time dependent. In this case the differential equation (1.55) and the boundary condition U(t0, t0) — 1 can be combined into an integral equation U(t, t 0 ) = 1 - i f' df H(t')U{t', t0), which can be iterated successively (plug U = 1 into the integral to approximate U, etc.) to form the time-ordered infinite series (t1>t2>--> t„) U(f, t ) = l + I
( - 0 " f dtt f" dt2-
(' ' dt„ H(tl)H(t2) ••• H(t„). (1.56b)
Clearly (1.56b) becomes (1.56a) if H is time independent; alternatively if H(t) commutes with H(t'), then the upper limits tu ..., tn in (1.56b) can be extended to t' (accompanied by a factor of 1/n! due to a change of integration regions—see Chapter 7), leading to U(t', f) = e x p ( - i j
dt" H(t")\.
(1.56c)
Then again (1.56a) follows as a special case of (1.56c). Time-Development Pictures. The above formulation of time development makes no mention of the time dependence of observable operators (other than H). The time evolution of states |i/'(r)>s, combined with timeindependent observables Os, is referred to as the Schrodinger picture. An alternative but equivalent dynamical scheme, called the Heisenberg picture, involves stationary states | \j/yH along with observables 0H(t) which change with time according to the Heisenberg equation of motion (1.5). More
Translations in Time 13 specifically we may choose
\Ht)>s= l"A(0X 0S = 0(t0),
I«A>H=
|#o)>
0H(t) = 0(t),
(1.57a) (1.57b)
with t0 a fixed time (t — t0). These two pictures are related by
|„=t/- 1 (Mo)l'/'(0> S <«AH 10a(t) | >AH> = < M ) | O s | <Mr)>,
(l-58a) (1.58b)
where (1.58a) follows from (1.52), and (1.58b) is in the spirit of (1.42). Combining (1.58a) and (1.58b) leads to the similarity transformation relating the observables in the two pictures, OH(t)=U-1(t,t0)OsU(t,t0).
(1.59)
Interaction Picture. There is a third picture, called the Dirac or interaction picture, which makes use of the decomposition H = H0 + V with observables 0,(t) dynamically driven by H0 and states |tHf)>/ driven by V. In particular one defines L6
I #)>/ = *""* | *(0>,
( °)
iHo
which removes the e~ ' dependence from \*l/(t)},. Given (1.51), |^(t)>j satisfies a Schrodinger equation driven only by V,
(;a,-F;(t))|>A(0>/ = o
(i.6i)
with V, in turn driven only by H0, VI(t) = eiHotVe-iH°'
(1.62)
[note that the general identity (1.50) may be applied to (1.62)]. In fact (1.62) is one example of the general similarity transformation relating any observable in the interaction and Schrodinger pictures (Vs = V), obtained from (1.60) and the connection between matrix elements in the two pictures,
(i-63)
Finally, we may define a unitary time translatior operator in the interaction picture, C/,(t', t) [similar in nature to the Schrodinger-picture operator Us(f,t)=U(f,t)]: Like (1.61), U, obeys {id, - VkWk,
t0) = 0,
(1.65)
which, like (1.56b), can be iterated to the general time-ordered form U,{f, t) = 1 + t
(-if
( dt, [ ' dt2 ••• (' ' dtn VAtjVfa)-
V,(tn). (1.66)
14
Introduction
The interaction picture will play a fundamental role in the general formulation of the scattering problem discussed in Chapter 7. Even earlier in Chapter 6, however, we will make frequent use of the scattering operator S,= [/7(f = oo, t= -oo).
(1.67)
This is a form of the S-matrix which explicitly shows, as in (1.66), that S contains all the dynamics of the hamiltonian. For systems with no hamiltonian defined, which is usually the case in elementary-particle physics, S continues to describe the dynamics of the interaction. The material presented in this chapter is intended only as a brief review of the fundamentals of quantum mechanics needed as a basis for this book. For a more detailed exposition of this subject, the reader is referred to Pauli (1933), Ludwig (1954), Kramers (1957), Mandl (1957), Dirac (1958), Landau and Lifshitz (1958), Messiah (1962), Feynman et al. (1965), Gottfried (1966), Tomanaga (1966), Matthews (1968), Schiff (1968), Merzbacher (1970), Gasiorowicz (1974), and Mathews and Venkatesan (1976).
CHAPTER 2
Transformations in Space
The invariance of the laws of physics under time translations, space translations, and rotations is naturally linked with the conservation laws of energy, linear momentum, and angular momentum, respectively. In Chapter 1 we have briefly described time translations in terms of the time development operator; spatial transformations, however, can be expressed most compactly and elegantly in the language of group theory. The rotation-group operators are the infinitesimal generators of angular momentum; the group closure property then leads to the familiar commutation relations of angular momentum. We shall also consider in detail the structure of vector and spinor wave functions, spin matrices, tensor operators, and the WignerEckart theorem. Such a review of rotations will serve as a bridge between nonrelativistic and relativistic quantum theory. In particular it will serve as a basis for our treatment of Lorentz transformations in the next chapter.
2.A Continuous Symmetry Transformations and Group Theory Continuous symmetry transformations such as translations and rotations are generated in a passive sense by two observers viewing the same point and measuring the values x and x', respectively. In the case of rotations, these measurements are related by a rotation matrix R as xj = RijXj (Einstein summation convention used). A symmetry transformation will also alter the structure of quantum states, and we have seen in Section l.C that this is most naturally described by a unitary operator transforming the state | ij/} to 1>'> 15
16 Transformations in Space in an active sense (one observer looking at the system before and after the transformation). Continuous Symmetry Operators and Matrix Representations. These two operations (on coordinates and states) must be combined for wave functions <x | if/} = iA(x), i.e., for the transformations on functions of the symmetry variable, but the notions of passive and active must then be kept clear. Since the transformed state can take on three different forms [iZ'(x'), >A'(x), and i/f'(x'), where the primed notation ip'(x) indicates a change in the functional form], it will prove convenient henceforth to apply active transformations on coordinates as well as on states. In practice this simply requires coordinate transformations of the form x' = Rx, with R3Ciive = R ^ v e - The three types of transformations on the wave functions then involve the three operators (9, 2, and
(2.1a)
iA'(x) = @{R)»l/(x)
(2.1b)
\\i' (x') = 5f(R)il/(x).
(2.1c)
The operation (9 corresponds to a differential displacement in a Taylor series expansion of the argument x: i/^(x + Ax) = i^(x) + • • •. The & operation in (2.1b) is a finite-dimensional matrix representation of the unitary operator UR in the subspace of (spin) eigenvalues as Daa(R) = <
M*) = L, •+Ax)Da.0(R). Lastly, the operator £f(R) in (2.1c) represents the combined transformation of coordinates and states with i/C(*') = ^oa(R)^a(x), corresponding to matrix multiplication in the a subspace in a manner similar to the cartesian index representation x\ = RyXj with i, j = 1, 2, 3. Irreducible Group Representations. In mathematical terminology, the collection of operators R or (9R or UR satisfying (2.1) falls under the classification of a group. Finite-dimensional representations of the group correspond to the matrices D(R) and S(R) (or R itself, called the defining representation). Reducible representations can be broken down to block-diagonal form via a similarity transformation. Block-diagonal or irreducible representations in general satisfy the Schur lemma (only multiples of the identity operator commute with all members of an irreducible representation). With reference to (2.1), irreducible representations of 3>(R), y{R), and OR1 are essentially equivalent. Reducible representations always can be decomposed into a linear combination of irreducible representations (for "compact" groups). Closure. The theory of group representations is very abstract and beautiful, but it is not our central concern. From the standpoint of transformation
Translations in Space 17 theory, which is our major interest, the crucial property of a group is that of closure, i.e., a group operation on a group element always gives rise to another group element. This group closure property applied to the operations in (2.1) means that if Rt and R2 are symmetry transformations, then so is R1R2', likewise URiUR2=URlR2 with faithful representations D(K,)D(il2) = D(Rt R2), S(R1)S(R2) = S(Rt R2), and (9RlR^(x)
= iP(R2RlX)
(note the reversed order of operation in the latter case).
2.B Translations in Space Consider the effect of translating the coordinate vector x upon the wave function i^(x). The coordinate transformation x' = x + a can be interpreted as a passive translation of the coordinate system through — a or alternatively as an active translation of the system by + a. In either case the Taylor series expansion of the wave function is i^(x + a) = i^(x) + a • V^(x) + ---,
(2.2)
which can be exponentiated to the form tfr(x') = 0 . M x ) = e - V ( x ) .
(2-3)
v
where the displacement operator Ga = e"' serves to define the translation operation. Linear Momentum as Infinitesimal Generator. The unitary translation operator Ua measures the (active) evolution of the system away from the notranslation configuration in terms of the hermitian infinitesimal generator P as Ua= 1 - M - P + ---.
(2.4)
Since the order of performing any two translations is immaterial, U„ Ua. = Ua. Ua implies from (2.4) that [P ; ,P,] = 0.
(2.5)
Moreover the product of any two translations being a translation, Ua Ua. = Ua.+a, means that the hermitian operator P is additive. It is therefore natural to identify P as the momentum operator P = — iV. To further establish this relation, note that since P and Ua commute with the spin operator S, translations do not affect the spin (angular momentum) indices of a wave function. Consequently, we may treat any i//(x) as a scalar under translations,
n*')=HA
(2.6)
18 Transformations in Space corresponding to if {a) = 1 in (2.1c). Also, since the unitary translation of states is | i/O = Ua \ if/}, we may write f ( x ) = <x|l/ B | f /f> = t / ^ ( x ) .
(2.7)
Comparing (2.3) with (2.6) and (2.7) it is clear that P = — i\ and Ua = 0-1=e-i'-F.
(2.8)
That is, the displacement operator (9a acts on adjoint bra states <x |, while the Hilbert-space operator Ua acts on ket states 11//>. Translation Group. The set of translations in space forms a group, but the commuting infinitesimal generators (2.5) and the simple transformation property (2.6) make this group almost trivial. Evidently (2.8) or ^(a) = 1 indicates that the only finite-dimensional irreducible representation of the translation group is one-dimensional. A similar interplay between transformation properties (2.1) and group closure properties also plays the central role in the study of rotations and Lorentz transformations. In these latter cases the group structure is much richer by virtue of the noncommuting nature of the infinitesimal generators analogous to (2.5). In geometric terms this is because two successive operations, such as rotations, do not commute in three dimensions.
2.C Rotations in Space Rotation Matrix. An active rotation x\ = R\?]Xj about the z-axis through an angle 6 corresponds to a passive rotation of the coordinates through — 6 and is given by the cartesian rotation matrix <3)
i?
/cos0 = sin 9 \ 0
-sin0 cos 9 0
0\ 0 . 1/
(2.9)
For infinitesimal rotations, (2.9) and similar rotations about the x- and y-axes imply x' = x + <38 x x and Ru = Su - S9k ekij,
(2.10)
where ekij is the Levi-Civita symbol, antisymmetric in all pairs of indices ki and ij and normalized so that s123 = 1. Clearly (2.9) becomes (2.10) for 9 replaced by S93 in (2.10). For noninfinitesimal rotations, (2.10) can be exponentiated to the form p . 11) R .. = e -ie<s)y where the "vector" matrix (S)0 has components (S^-fey*.
(2.12)
Rotations in Space 19
The antisymmetry of (S)0- guarantees that the inverse of Rtj is equal to the transpose Rjt = Rfj, i.e., RuRji = tij, 1
(2-13)
T
or equivalently R' = R . This relation defines an orthogonal transformation, preserving the length of any vector (x|x| = RilRimxlxm = x,x;), as any rotation must. Orthogonal transformations include more than rotations, however, because (2.13) implies (det R)2 = 1. Reflections, on the other hand, obey (2.13) but have det R — — 1. In this chapter we shall restrict the discussion to pure rotations (det R = 1), in which case not only do polar vectors tranform like the coordinates {V'{ — Ru Vj), but so do axial vectors such as (V\ x V2)| = Rij(\i x V2)j. This latter transformation implies (for det R — 1; see Problem 2.1) which can be verified directly from R(3), etc., as given by (2.9). With the aid of (2.12) and (2.13), we can write (2.14) as a matrix relation in the space of the cartesian indices, R~lSiR = RijSJ.
(2.15)
_1
Interpreting R S,R as a similarity transformation in this space to the rotated S't, we can take (2.15) as further indication that the vector (spin) matrix S as defined by (2.12) does transform as a true vector. Angular Momenta as Infinitesimal Generators. For rotated wave functions, a Taylor series expansion can be developed for infinitesimal rotations x' = x + (56 x x: i^(x + <58 x x) = i//(x) + c58 x x • Vi^(x) + • • •.
(2.16)
Since the orbital-angular-momentum operator is L = — ix x V, we may exponentiate (2.16) for noninfinitesimal rotations to iA(x') = e i 0 V ( x ) = ^«i/'(x).
(2.17)
In Hilbert space the unitary operator for rotations, UR, is expected to be of the form of (2.11), and &RX as given by (2.17). Then we write UR = e-*s,
(2.18)
where the infinitesimal generator of rotations in (2.18), J, is presumably the angular-momentum operator in Hilbert space, and f ( x ) = URil/(x) = e"' e •>(*)•
(2-19)
Finally there exists a finite-dimensional matrix S(R) [not to be confused with S as given by (2.12)] obeying (2.1) in the space of spin indices. The rotation matrix itself as given by (2.11) is the spin-1 representation of £f(R). Ab-
20 Transformations in Space stracted as an operator in Hilbert space, we expect that S(R) = e~ies
(2.20)
by analogy with (2.11) and (2.18). Then, by use of (2.1)-(2.3), il,'(x')=URiJ,(x')=UR(9R
(2.21)
which implies that J = L + S.
(2.22)
This relation lends further credence to the physical interpretation of J as the total-angular-momentum operator and S as the spin-angular-momentum operator; while (9R operates on the coordinates, UR operates in Hilbert space, and S acts in the subspace of spin indices in such a way that (2.22) is valid. Note that (2.22) depended upon keeping straight the concepts of active vs. passive and UR vs. GRl. Having tied down the relation between the various angular momenta as infinitesimal generators, we are ready to investigate the more abstract structure of the rotation group as it pertains to transformation theory.
2.D The Rotation Group 0 ( 3 ) Closure. We now exploit the closure property that the product of any two successive rotations is also a rotation. More specifically, since the group closure property R1R2{R1R2)T = 1 holds if RYR\ = 1 and R2RT2 = 1, the set of all three-dimensional orthogonal rotation matrices with RRT = 1, det R = 1 forms a group, denoted by 0 + (3). While there is a continuous infinity of group elements corresponding to the values of Qu 62, 63 in (2.11), there are only three infinitesimal generators Jl9 J2, and J3. In order to characterize the rotation group, therefore, it will be most revealing to describe the behavior of J. To begin with, the spin-1 cartesian matrices (2.12) form a closed algebra [henceforth we display explicitly the spin-1 structure of(Sj%=-iefJJ [S\l\ Sf] = mijkS?\
(2.23)
These are, of course, the familiar commutation relations of angular momentum, which are in fact satisfied by all infinitesimal generators of rotations, Sh 1^, and J ; . This is a direct consequence of the Hilbert-space analog of the group closure property (choosing the phase to be unity) URlUR2=URlR2,
(2.24a)
which, when applied twice, gives UR URUR — UR-iRR.
(2.24b)
The Rotation Group 0(3) 21 For infinitesimal rotations of R', (2.18) becomes UR, = l - i<58' • J,
(2.25)
and so UR lUR.UR = l - W\UR %UR,
(2.26)
whereas UR-IRR
= Ui-im-R-iSiii)R
= 1 — i dO'i RijJj
(2.27)
follows from the group closure property and (2.15). Comparing (2.26) and (2.27) we see that URlJiUR = RijJj.
(2.28)
Then take R in (2.28) as an infinitesimal rotation, so that (1 + i 30k J t )J ( (l - i 89, J,) = (dy - 59k ekiJ)Jj + • • •, which, for arbitrary 50k, implies the angular-momentum commutation relations [J„ Jj] = ieiJkJk.
(2.29)
Thus, the group closure property generates a closed algebra for the infinitesimal generators, called a Lie algebra, which is governed by "structure constants", in this case sijk. This completes the description of the infinitesimal generators for rotations. Irreducible Representations. Next we investigate the representations of the rotation group. Given the structure of UR as e~'° J and the properties of J as operators in the infinite-dimensional Hilbert space, we can search for faithful finite-dimensional representations of the group. Clearly the general property (2.28) is reproduced by the three-dimensional matrix relations (2.15). Similarly, the general commutation relations (2.29) are faithfully reproduced by the three-dimensional commutation relations (2.23), and likewise the general angular-momentum eigenstates J2 | jrri> = j(j + 1 ) | jm>,
J 3 1 jm> = m | jm)
(2.30)
are represented by the orbital eigenstates (1.9) and (1.10). Such eigenstates represent the "carrier space" for the rotation operators UR. For a given j , the action of UR on j jm} serves to "reshuffle" the eigenstates among the 2/ + 1 different values of m, and the reshuffling coefficients DlJ)m(R) are collectively called the jth irreducible representation of the rotation group of dimension 2j + 1. That is, for D$m(R) = (jm'\UR\jm>
= {jm' | C - i e j \jm>
(2.31)
we have for; = 0, \, 1, | , ... andy' > \m\, | jm}' = UR | jm) = £ | jm'yD$m(R),
(2.32)
22 Transformations in Space
where the finite-dimensional unitary matrices satisfy D$m(R) = DV>*(R->)
(2.33)
and transform according to the general relation (2. lb). [We shall continue to include the summation symbol in relations like (2.32).] For integer j = /, (2.32) can be converted into the transformation law for spherical harmonics by applying
(2.34)
m'
This relation has important consequences, such as the addition theorem (1.16). (See Problem 2.2.) To exploit such properties in detail, it is useful to express D$m(R) in terms of the conventional Euler angles, i?^,9,^ = R^RgR^, corresponding to three successive, active, counterclockwise rotations: First rotate by > about the z-axis (c - '* 53 ), then by 6 about the new /-axis {e~ieSl), and finally by \\> about the new z"-axis (e~'*S3'). In terms of the original fixed axes, the unitary operator has the form e-i*^e~iej2e~i'l'J\ (See Problem 2.3.) Taking 9 and > as the usual spherical angles, 0 < 6 < n, 0 <
(2.35)
so that for (Re
WM*)
(2.36)
J
dUP) = <J»*\e-» >\jtri>,
(2.37)
J
where the rotation functions d m.m satisfy the general formula (see, e.g. Wigner 1959) dj
(e] =
y ( - H ( / + w)! Q- ~ m)\ (j + m')\ (j - m')\]* n (j ~ m' - n)\ (j + m- n)\ (n + rri - m)\ n\ x (cos i0) 2 j ' + m - m '- 2 n (-sin $8)m-m+2n, w
(2.38) (1)
satisfying CSR phase conventions. We tabulate D and £> in Appendix II. Nonetheless it is instructive to derive some of the important properties of the d(9) from first principles (see Problem 2.4). Spin-1 Defining Representation. The defining representation of the rotation group is given by the cartesian transformation of the coordinates, x\ = i?(Jx7. It is also possible, however, to formulate this representation in the irreducible spherical basis. Introducing the spherical coordinates for j — 1, rm =
(2.39)
it is clear that the analog of x\ = Rtj Xj is r'm=Rmmrm-
(2.40)
The Rotation Group 0(3) 23
On the other hand, YT(f) = < r | l m > ~ r * ,
(2.41)
so a comparison of (2.40) with the transformation law (2.34) for / = 1 and using (2.33) reveals that Rm,m = D^m(R),
(2.42)
which proves that R is the faithful spin one-dimensional representation of UR. Next define a spin-1 polarization vector e|m)(p3) proportional to a spin-1 wave function with momentum dependence in the z-direction and transforming like yj1, i.e., 4 1 ''(Pa) = T - ^ (I ±i, 0),
£!
0)
(P3) = (0, 0, 1),
(2.43)
we conclude from the fact that e|m) does not change its functional dependence upon rotation (see Problem 2.5) that ei"W3) = RijePfa). a
(2.44)
{l)
Thus R acts like D \R) in (2.42) but like S (R) in (2.44). The vector e!m>(p3) is referred to as a polarization vector and will be needed to describe spin-1 relativistic wave equations in Chapter 4. Spin-j Spinor Representation. It is also possible to represent the cartesian spin-1 angular-momentum matrix (Sj1')^ = —i£iJk in the spherical basis by coupling two spin-1 states to form a spin-1 vector. In general, coupling two spin-s states with magnetic quantum numbers m and m', a (2s + 1) x (2s + 1) vector matrix can be formed as (see Problem 2.6) ( S £ U = [(2s + l)( s + l)s/3]*(-)- m <ssm', -m\ 1M>.
(2.45)
Here (ssm', — m 11M> is a Clebsch-Gordan or Wigner angular-momentum coupling coefficient (with the CSR phase convention). While (2.45) is not the only way to form higher-spin matrices, for s = 1 it is indeed the spherical analog of (Sj1*),* = — ieiJk. For s = \, the 2 x 2 spin-one matrices which are a direct consequence of (2.45) can be written as (see Problem 2.6) S(*> =
fr,
(2.46)
where at are called the Pauli spin matrices with cartesian components,
- U
-(?"4 Hi-") ™
and the matrices are in the spin space of spherical indices \ and —\, respectively. As faithful representations of J, the spin matrices (2.47) must, and do, obey the angular-momentum commutation relations, i.e., [ff„ aj\ = 2ieijkak.
(2.48)
24 Transformations in Space Furthermore, computation with (2.47) reveals that the Pauli matrices also obey the anticommutation relations {ah oj} = 2,5;,,
(2.49)
where {A, B} = AB + BA. Combining (2.48) with (2.49) we see that teijkak-
°i<*j = ^.j +
(2-50)
This decomposition holds in general for any representation of the spin-^ matrices and corresponds to the direct product of angular momenta,
i x i = 0+ 1. Given these spin-^ matrices, it is possible to describe the spinor representations of the rotation group. Since (2.50) implies a\ = 1, a\ — au etc., the spin4 representations of UR can be separated into even (cosine) and odd (sine) functions of the rotation angle according to SW(R) = e~ie•*' = cos \9 - itf • a sin # .
(2.51)
Then (2.50) leads immediately to the spin-^ analog of (2.15), [S«>(R)]-1
(2.52)
Next, identify q>a as an entry in the spin column {q>i/(p^i). Such coordinatespace wave functions transform as faithful representations of the rotation group cp'(x') = S(i)(R)
(2.53a) (2.53b)
where (2.53b) follows from (2.53a) and (2.52) and explicitly demonstrates the vector character of a. The analogs of the spin-1 polarization vectors e(m,(p) are the spin4 momentum-space spinors
and
9 <-*>(fc)=^J.
(2.54a)
Since these spinors do not change their functional forms under rotations, they transform similarly to (2.44), i.e., cp™(Rp3) = S«\R)
(2.54b)
It is also well to point out that e^Rp^) and
Tensor Operators and the Wigner-Eckart Theorem 25
2.E Tensor Operators and the Wigner-Eckart Theorem Vector Operators. To pursue the connection between transformation theory and the properties of the rotation group, note that the transformation law for spherical harmonics (2.34) can be interpreted as one for a scalar operator, because the left-hand side can be thought of as i//'(f) or alternatively as \p(R~ lr). The extension of (2.34) to vector operators Vt transforming like the coordinates x[ = RijXj must be
< * ' I W > = Ky<
(2-55)
where the left-hand side can be interpreted as the expectation value of V\ with respect to the unrotated states. Applying 11//> = UR 11/') to (2.55) then yields UllVlUR = Ri}VJ,
(2.56)
for which (2.15) is a particular group realization. This relation can be verified directly by the use of (1.50) and (2.18). (See Problem 3.3.) It should be noted that the active rotation (2.56) appears not to have the same similarity transformation structure as (1.45). The latter can be emulated by converting R to R~l in (2.56), corresponding to a passive rotation, U^U^^iR-^jVj.
(2.57)
Then proceeding as in the derivation of the fundamental group commutation relations (2.29), an infinitesimal rotation in (2.56) or (2.57) leads to [Ji,Vj] = ieiJkVk,
(2.58)
for which (2.29) is a special case with V replaced by J in (2.58). Tensor Operators and Higher-Spin Wave Functions. The generalization of a vector operator to a cartesian tensor operator is straightforward. First recall that a second-rank nonoperator (c-number) rotation tensor Tu must transform like the product of coordinates, TiJ = RikRjlTkl.
(2.59)
An irreducible cartesian tensor must be symmetric and "traceless", Tti=Tfit
- 5 ^ = 0,
(2.60)
such as X; Xj — \xk xk Stj, which has only 5 independent components and corresponds to the cartesian analog of Y^(f). Tensor wave functions describing a particle with higher spin must also be symmetric and traceless as in (2.60). The spin-2 wave function is proportional to rank-2 polarization tensors e\f) obeying (2.60) and related to the spin-1 polarization vectors (2.43) as (see Problem 2.8) e\f=
£
(2.61)
26 Transformations in Space
Likewise, a spin-| polarization vector can be constructed from spin-1 polarization vectors and spin-^ spinors as (pf) = £ < l i m'm||M>£!m'Vm).
(2.62)
m'm
While this one-index (pt cannot be constrained by (2.60), it must instead obey (see Problem 2.9) <x;(p, = 0.
(2.63)
The extension to tensors of rank higher than | or 2 is straightforward but always involves the "irreducibility conditions" (2.60) and (2.63). Cartesian tensor operators are further complicated by the noncommutativity of the operator components, such as S„ which obey (2.23). The construction then becomes quite tedious. Spherical Tensor Operators. It is more natural instead to generalize spherical representations as in (2.45). Consider then combinations of vector operators built up to form irreducible spherical tensor operators Tf. For J = 1, these tensor operators must transform like Yf, i.e., with CSR phases, Ti±1 =
_(j^±3) ;
n =K
(264)
Since YT transforms like |/w>, Tf must transform like \JM}. We recall from (2.41), however, that spherical components of vectors, in this case Vm, transform according to the complex-conjugate representation. This subtlety means that the spherical analog of the transformation law (2.57) for general J can be written as URTfURl
= D*M%{R-')Tf,
(2.65)
since for J = 1, D(1) = R as in (2.42) reduces (2.65) to the spherical form of (2.57). It is conventional to use the unitary property of D in (2.65) and rewrite this transformation law for tensor operators as (reinstating the summation sign) UnTfU^^^TfD^iR).
(2.66)
M'
With hindsight, (2.66) appears to be the operator extension of the transformation law for spherical harmonics, (2.34), with the left-hand side of both equations expressed in passive form. Given (2.66), an infinitesimal rotation leads to the general commutation relations [J±, Tf] = N ± (M)75 f ± 1 ,
[J 3 , Tf] = MTf,
(2.67)
where J± = J x + U2 and N±(m) = [j(j+l)-m(m±lj\\ J± \jm} = N±(m)\j, m ± 1>
(2.68) (2.69)
Tensor Operators and the Wigner-Eckart Theorem
27
defines the angular-momentum raising and lowering operators and determines the CSR phase convention. The commutation relations (2.67) are the spherical generalization of the vector commutation relations (2.58). Wigner-Eckart Theorem. At this point it is natural to discuss the WignerEckart theorem, which is satisfied by these tensor operators Tf. This follows immediately from the Clebsch-Gordan completeness expansion \jm'>= Z
\Jd2'nlm2yij1j2m1m2\jmy.
(2.70)
mim2
The realization that Tf transforms like |JM> means that its matrix elements are proportional to <Jm' | Tf | jm} = £
\Tf\jm}
oc £ 4,. 7 4,. M 6ri8M..m. Since the Clebsch-Gordan coefficients are taken to be real (CSR phase convention), the Wigner-Eckart theorem is usually written as
|| T3 ||j>,
(2.72)
where || Tj ||j> is the "physical part" of Tf, independent of magnetic quantum numbers, and is called the "reduced matrix element". The only magnetic-quantum-number (geometrical) dependence of the matrix elements of Tf is via the Clebsch-Gordan coefficient (2.71). An alternative derivation of (2.72) relies upon a comparison of (2.66) with the ClebschGordan series for direct-product representations with J = j t + j2,ji +72 — !>•••> \ji — 72 | ( s e e Problem 2.10), DWmi(R)DV%2(R) = X
<Jd2mlm2\JMXJ1J2m'1m'2\JM'}D^M(R).
j
(2.73)
A more transparent and useful special case of (2.72) is sometimes referred to as the "little" Wigner-Eckart theorem for scalar operators S = T%. For J = M = 0 in (2.72), the Clebsch-Gordan coefficient is proportional to djj3mm; it therefore remains to verify that (jm\S\jm} is independent of m. To this end, consider SJ + between states with m values differing by 1 (and delete the) dependence): <m+ l | S J + | m > = N + (m)<m+ l | S | m + 1>,
(2.74)
by use of (2.69). Now S being a scalar operator means [S, J] = 0, so that S commutes with J + and N+(mXm+
l\S\m+
1> = <m +
l\J+S\m}
= /Y_(m+ l)Oi|S|m>.
(2.75)
However, N+(m) = N_(m + 1), which implies that (2.75) becomes <m+ l\S\m+
l> = <m|S|m>,
(2.76)
28 Transformations in Space
meaning that the reduced matrix element, (jm \S \jm} = Sp is independent of m. Thus we write the little Wigner-Eckart theorem as
(2.77)
Assuming S to be the 5-matrix of Section l.B (also a scalar operator in the Wigner-Eckart sense), (2.77) is proof of the assertion (1.30), the latter being diagonal in energy as well. As such, (2.77) is justification for the partial-wave expansion (1.31). Another example of a scalar operator is J2 itself, since [Jh J2] = 0. On the other hand, the vector operator T, = J2St — (S • J)J ; , where S, is the spin, does not commute with J 2 ; instead [[ri,j2],j2]
= 2{ri,j2}.
(2.78)
This operator identity serves to generate selection rules when sandwiched between angular-momentum eigenstates, as in the little Wigner-Eckart theorem above or in the selection rules leading to (1.29). The vector model in atomic physics then follows from (2.78) or equivalently from the WignerEckart theorem (2.72) for Tf (see Problem 2.10). For higher-spin cases, the general Wigner-Eckart theorem certainly is easier to apply than identities like (2.78). For general references on the rotation group, see Condon and Shortley (1951), Rose (1957), Wigner (1959), Messiah (1962), Brink and Satchler (1968), Bargmann (1970), and Merzbacher (1970).
CHAPTER 3
Transformations in Space-Time
The review of translations, rotations, and the rotation group being completed, we are prepared to investigate their relativistic generalizations: translations, rotations, and velocity transformations in space-time, along with the homogeneous and inhomogeneous Lorentz groups. We shall again stress group closure properties and the resulting transformation laws and commutation relations. The new concept of a "boost" operation, a form of active Lorentz velocity transformation, is particularly useful for the construction of relativistic wave functions describing particles with spin. Helicity is another important notion which we shall develop. Spin4 helicity spinors and spin-1 helicity polarization vectors are covariantly generalized from their rotation-group form. Massive-particle helicity states are contrasted with massless-particle helicity states with an eye to their use in later chapters.
3.A Lorentz Velocity Transformations We begin at the beginning—the origins of the theory of relativity. Briefly stated, the measured aberration of starlight, the observed independence of the velocity of light upon the motion of double star sources, and the absence of fringe shifts in the Michelson-Morley experiment led to the conclusion that light traveled in a manner independent of any "aether". The "logical" consequence was the Einstein postulates (Einstein 1905): i. all physical laws are covariant (frame independent); ii. the constancy of the speed of light, vh as measured by all observers; Vi = c = 1 (recall our units h = c = 1). 29
30 Transformations in Space-Time
Velocity Transformation Matrix. In the language of transformation theory, postulate ii links time and space together by the four-vectors (note that this is not the convention used in most texts on general relativity) x„ = (x0, x),
x" = (x0, - x ) ,
(3.1)
where x 0 = t and n = 0, 1, 2, 3, such that the scalar product x„x" = XQ — x 2 = x • x = x 2
(3.2)
is invariant from frame to frame, x;x'" = x„x".
(3.3)
Postulate i then implies that there must exist a linear transformation connecting x„ and xj, obeying (3.3). For a passive Lorentz velocity transformation, vx= —v, of an observer measuring an event at x' relative to an observer measuring the event at x, (3.3) is satisfied provided (see Problem 3.1) t' = y(t + vx), x' = y(x + vt),
/ = y,
z' = z,
(3.4)
where y = (1 — i; 2 ) - *. We prefer the active interpretation of (3.4), whereby a single event is transformed through space-time from xM = (t, x) to x'p = (t', x') via the velocity vx = v. In matrix language, x^ can be taken as a column vector satisfying the transformation law *; = A/x v ,
(3.5)
where A / is the active Lorentz velocity transformation matrix A(v) with /z referring to the rows and v to the columns. For v in the x-direction, (3.4) implies 'y yv 0 0 yv y 0 0 0 0 1 0 ,0 0 0 1 Since the linear transformation (3.5) has the same form as a lengthpreserving rotation x\ = Rq Xj, it is useful to regard (3.2), or more generally the proper-time interval ds = dx^ dx*, as a squared "length" in space-time. The analog to an orthogonal rotation RRT = 1 is the length-preserving constraint obtained from (3.3) and (3.5), A/A", = S\,
(3.7)
where S\ = lifv = k{ = 0, 1, 2, 3) and 5\ = 0 for v # A. The matrix A"v, defined by x'" = A\x*,
(3.8)
is the same as A(v) except that v is replaced by — v in (3.6), consistent with (3.7), or equivalent^, (A -1 )" v = A / .
Lorentz Velocity Transformations 31 The velocity transformation also preserves the causal nature of events, transforming timelike separations (Ax)2 = (At)2 — (Ax)2 > 0 into timelike separations, because v
&.»=
0 0 0n -1
1 0 0 0 - 1 0 n n _ i 0 0 - 1 0 0 0
(3-10)
and x" = #"vxv, with g^ given by the same matrix as (3.10). With the aid of g^, it is possible to express the length-preserving constraint completely in terms of the matrix A(v):
V A ' =j", T
(3.H)
vfl
or A gA = g, since g^ = g is the inverse of g^, g"ag.v = s\.
(3.12)
In general relativity, the "metric" g^ is not constant as in (3.10), and it plays a central role in the theory. In that case the raising and lowering matrix (3.10) is relabeled as r\^ and referred to as the "flat space" limit of g^. We shall make use of this distinction only in Chapter 14. Infinitesimal Transformations. Returning to the velocity transformation (3.6) and paralleling the structure of rotations as much as possible, one gains further insight into the origin of A(v) by investigating infinitesimal velocity transformations. Starting with an infinitesimal Newtonian transformation in the x direction, x' = x + t 8v, one writes [see e.g. Bjorken and Drell (1964)]
A/ = V + CV,
(3-13)
where /0 /v_ 7
"
-
1 0 0\
i o o o
/l {/2)v = / a / v =
[
x
0 0 0'
o loo.
1 0 0 0 0 I' '" " I0 0 0 0 ' ' \0 0 0 0/ \ 0 0 0 0, and (7 3 )/ = / / . Exponentiation of (3.13) then leads to A(w) = e£/,
( 14) {
'
(3.15)
or equivalently, employing the projection-operator character of (3.14) along with the even (odd) parity of cosh CI (sinh £7), as in cosh CI = 1 — jC2I2 H—,
32 Transformations in Space-Time we find A / = (cosh C/)„v + (sinh £/)/
= V - ( /2 V + ( /2 V
cosh
C + i; sinh £.
(3.16)
2
Since (3.14) means 1 — J is nonvanishing and unity only in the yy and zz entries, it is clear that (3.16) is identical to the original form of the velocity transformation (3.6) provided y = cosh £,
yv = sinh £,
(3-17)
which implies v = tanh £ or y = (1 — v2)~i (see Problem 3.1). One can generalize (3.13)—(3.16) to include velocity transformations in an arbitrary direction by replacing £/,,„ with the sum of three terms, one corresponding to each of the three space directions. It is also possible to invert (3.17) as £ = tanh" 'v = \ log ( | ^ j ,
(3.18)
with the "Lorentz angle" £ sometimes referred to as "rapidity". It has the simple property that under a combination of Lorentz velocity transformations in the same direction (A12 = A t A2), the rapidities add (£12 = £ t + £2). Boost Transformation. The Lorentz velocity transformation is a covariant operation in that any four-vector V^ = (V0, V) must transform like the coordinates, i.e., V'p = A/F v . The only limitation to this statement is that a causal event described by ds2 > 0 can never be transformed out of the "light cone" to the spacelike region, ds2 < 0, because v < c. In particular, the energy-momentum four-vector describing a free particle, pM = (£, p), has an invariant length p2 = p„p* = E2 - p 2 = m2,
(3.19)
where m is the rest mass. Since p^ must transform like the coordinates (pj, = A/p v ), it is possible to find a Lorentz velocity transformation which "boosts" the particle from rest with four-momentum m^ = (m, 0) for m j= 0 to the momentum p; it is denoted as L/(p): p„ = V(p)m v .
(3.20)
The boost transformation L(p) is the specific Lorentz velocity transformation A(v) with y = E/m = cosh £ and yv — \ p | /m = sinh £, or equivalently from (3.20) L0°(p)=^
L/>(p) = £
(3.21a)
and then from (3.7),
MP) =-, m
W = 5ij+
PiP
J
m(E + m)
(3.21b)
Homogeneous Lorentz Group if 33 It is important to stress the covariant nature of the boost operation; in particular, its manifest covariant representation implied by (3.21) is (see Problem 3.1)
where m- (p + m) = m(E + m). It is clear that (3.22) satisfies (3.20) and (3.21) and also evolves continuously from the identity, since L„v-»0„v as Pn -* "V Note too that the third term in (3.22) accounts for the intrinsic nonsymmetric structure of the boost LMV. This boost operation will play a fundamental role in the construction of quantum-mechanical wave functions describing particles with spin, but first it will be necessary to investigate the group-theoretical structure of Lorentz transformations, a subject to which we now turn.
3.B Homogeneous Lorentz Group i? Closure. While two successive rotations are a rotation, two successive velocity transformations (or boosts) are not, in general, a pure velocity transformation. However, the notion of closure can be enlarged to include both boosts and rotations. For velocity transformations Ax and A2 obeying the "lengthpreserving" constraint ATgA = g, we have (A, A2)Tg(Al A2) = AjAfrA, A2 = AT2gA2 = g.
(3.23)
Now rotation matrices R satisfying RRT = 1 also satisfy ATgA — g and (3.23) provided we extend R to space-time as A/(K)^V=(1
R
\
(3-24)
with coordinates x„ transforming according to x' = Ax.
(3.25)
Now it can be shown that, in addition, the discrete transformations of space reflection, time reversal, and space-time reflection also obey (3.23) and (3.25). The complete set of velocity transformations A(v), rotations A(R), and the aforementioned discrete transformations all obey (3.23) and (3.25), are called in general Lorentz transformations, and form a group called if, the homogeneous Lorentz group. Infinitesimal Generators. To restrict the discussion to the continuous Lorentz transformations A(v) and A(R) which evolve continuously from the identity, we constrain A to satisfy not only (3.23) and (3.25), but also A0° > 1, det A = 1. Then the set of A(v) and A(R) make up the proper orthochronous homogeneous Lorentz group Z£\, a subgroup of <£. It is then possible to
34 Transformations in Space-Time
represent any Lorentz transformation in jzfj as a product A(v)A(R). To express this fact in the most compact manner, it is useful to combine infinitesimal transformations for rotations, (2.10), and for velocity transformations, (3.13). To this end we write for small a)„v \y = 9w +
(3-26)
where the six-compnent tensor
(3-27)
combines (ou(R) = eijkOk,
(3-28)
for rotations and velocity transformations, respectively. While it is clear that (D^R) is antisymmetric as required by (3.27), co^v) acquires its antisymmetry by virtue of the metric tensor applied to (3.14a), JMV = I*g„ = —Ivll. The six continuous parameters 6xyz and Cx>y,z in (3.28) indicate the existence of six infinitesimal generators of i?J. To explore this possibility, we follow the transformation-theory analysis of the rotation group, first displaying the exponential form of the defining representation of !£\, (3.26), A = exp (-bw^Sj,
(3.29)
where SMV = — Sv/J is the covariant analog of the spin-1 matrix S, with the cartesian representation (Problem 3.2)
(S.vV'^.V-VW,
(3-30)
analogous to (S,)jk = —iEijk. The Hilbert-space unitary operator expressing the action in Z£\, \l/'(x) = UA\l/(x), must be similar in form to (3.29): C/A = e x p ( - ^ V , V ) ,
(3.31)
where l/A = t/A \ and is a six-component hermitian angular-momentum tensor operator. Likewise the Lorentz displacement transformation, \p(x') = $ A i//(x), corresponds to a Taylor series expansion with an exponential form 0A = exp (iiw^Lj,
(3.33)
and L^v = — L^ = i{xfldv — xvdfl) is a six-component orbital-angularmomentum tensor operator. Finally, since a spin-1 wave function transforms like the coordinates [i/^(x') = A/i/^x)], the infinitesimal form of (3.29), (3.31), and (3.33) immediately leads to J„v = LMV + S^v for spin-1 wave functions, as expected. A half-integral-spin angular-momentum decomposition must await a discussion of Dirac matrices in Chapter 5. It is therefore clear that if J is a six-parameter continuous (Lie) group with six infinitesimal generators composing the angular-momentum tensor operator J To determine the Lie algebra for the group (i.e., the commuta-
Homogeneous Lorentz Group if
35
tion relations satisfied by the generators J^), we again follow the pattern of 0(3) and extend the group closure property (3.23) to Hilbert space (up to a phase), UAlUAl=UAlAl.
(3.34)
1
Applying (3.34) twice to obtain C/^ t/A- UA = t/A_1A.A, an infinitesimal Lorentz transformation for A', and UK. = 1 — fyco^J^, we then obtain (Problem 3.2) C/A 1 ^»1/A = A | 1 " A V % ,
(3.35)
similar to (2.28) except that J^v is a second-rank tensor, transforming like the product of two coordinates. An infinitesimal Lorentz transformation for A and I/A then leads to the fundamental commutation relations a form satisfied by the spin-1 tensor matrix (3.30), as must be the case for all representations of the group (see Problem 3.2). Irreducible Representations. Next we investigate the irreducible representations of if J, similar in some sense to 0(3) x 0(3) because velocity transformations can be made formally to look like rotations. Given J^, define •>i ~ lEijkJjki
K-i — Joi>
(3.37)
so that the Lie algebra (3.36) becomes (Problem 3.2)
[Jt, K.j] = iZijkKki [Kt,K^=-iBtJkJk.
(3.38)
It can be seen from (3.28) and (3.31) that the hermitian angular-momentun operator J generates rotations and the hermitian "boost" operator K generates velocity transformations, represented by the active unitary transformations UMR) = e-«J,
C/L(p) = e - V K
(3.39)
Each operator in (3.39) has the structure of an 0(3) rotation, but the analogy is not complete, because the commutation relations (3.38) are coupled together. While (3.39) are infinite-dimensional unitary representations of <£\, the finite-dimensional irreducible representations of if J are not unitary (except in one dimension). The nonhermitian generators defined by A = i[J + iK],
B = i[J - iK],
(3.40)
decouple the commutation relations (3.38) according to (Problem 3.2) [Al,Aj] = ieUkAk,
[BhBj] = ieijkBk,
[At,Bj] = 0.
(3.41)
36 Transformations in Space-Time It is then possible to label the irreducible representations of jz?J as (A, B), where the operators A and B satisfy angular-momentum commutation relations (3.41a), (3.41b), but are decoupled by (3.41c). Inverting (3.40) and substituting in (3.39) gives t/A W = e~' e(A+B) > t/,
Dw(L(p)) =
(3.43)
If instead A = 0, so that B = J 0) , the (2/ + l)-dimensional irreducible representation (0,;') with Ul°,J) written as DU)(A) satisfies SU)(A(R)) = e~ie'
,0)
,
D0)(L(p)) = e c ' J0) .
(3.44)
We see from (3.43) and (3.44) that in general D°)(A) = D (J)t (A- 1 ).
(3.45)
For mixed representations (A, B) of j£?<j>, the Clebsch-Gordan composition (2.73) applied to A + B in (3.42) leads to effective rotation-group angular momenta A + B, A + B — 1,..., \A — B\. This fact will be useful for the construction of relativistic wave functions, to be discussed shortly. Matrix Representation. It is clear from (3.42) that direct-product representations such as (j, 0) + (0, j) will often occur, and it will be useful to combine them into a 2(2/ + l)-dimensional irreducible representation corresponding to the matrix
While Z)0)(A) and D0)(A) are not equivalent irreducible representations of J?}), the larger matrices S>U) and ^ 0 ) are equivalent because of the similarity transformation ®{1\A) = ^(A-1)
= p~ 1&Ui{A)P,
(3.47)
which follows from (3.45) along with
The significance of (3.46) is further reinforced by the space reflection operation (to be discussed in Chapter 6) under which D 0) -* 5 0 ) . Therefore the 3>a) of (3.46) are the finite, nonunitary irreducible representations of the larger group 5£, including the velocity transformations, rotations, and also space, time, and space-time reflections.
Homogeneous Lorentz Group 5£ 37
It is worthwhile to consider some specific representations of these boost operators. Later we shall demonstrate that the boost operator itself [i.e., LMV(p) as given by (3.22)] can represent the spin-1 boost. For the spin-i case, the (\, 0) boost is obtained from (3.43) as (Problem 3.3) Dw(Lp) = e-iia
= cosh & - o • p sinh #
(3.48)
in a manner similar to (2.51), where cosh if =
E+m 2m
sinh if =
m 2m
(3.49)
follows from cosh f = E/m. Combining (3.48) and (3.49), we see that £>(i)(Lp) = [2m(E + m)]-*(E + m-o-
p).
(3.50a)
D(i)(Lp) = [2m(E + m)]"*(£ + m + a • p).
(3.50b)
Likewise, the (0, \) boost found from (3.44) is
Since 2(2/ + 1) = 4 for j = \, the four-dimensional spin 4 boost constructed from (3.46) and (3.50) is 0<±>(I„) = [2m(E + m)\
(E + m)
oX
0 (3.51)
These spin-i boosts will play a fundamental role in the construction of the Dirac equation in Chapter 5. Carrier Space. The finite-dimensional irreducible representations of 5£ can also be characterized by the vector space of wave functions (carrier space) on which the Lorentz transformations act (analogous to \jm) for the rotation group). For example, the one-dimensional irreducible representation (0, 0) is specified by a scalar wave function
38 Transformations in Space-Time We now turn to more specific details concerning the group theoretical structure of relativistic wave functions with spin by first investigating the inhomogeneous Lorentz group.
3.C Inhomogeneous Lorentz Group 0> Group Properties. Enlarging <£ to include translations as well as Lorentz transformations, as in xj, = L/ia, A)xv = A„vxv + a„,
(3.53)
we obtain the inhomogeneous (a =/= 0) Lorentz group, known as the Poincare group &. This group is generated in Hilbert space by the unitary operator Ua,A=UaAU0>A=UaUA (3-54) with the closure properties
uQ. ua = ua.+a,
uA uA = uA.A,
t/ a , A . [/..A = t/„,+A.fl,A,A. (3.55)
The last result follows from successive applications of (3.53): Ua', K')Ua, A)x = L(a', A')[Ax + a] = A'Ax + A'a + a'.
(3.56)
From our prior experience with translations, rotations, and boosts, it is clear that the unitary operators in (3.54) have the exponential form Ua =Uatl=
expOaT„),
UA = l/0, A = exp( - ito'V,,),
(3.57)
where JMV is again the hermitian six-component antisymmetric tensor operator including the angular-momentum generators J and the boost generators K, and Pll is the hermitian energy-momentum vector operator, which generates time and space translations according to P„ = i d „ = ( i ^ , - i v j .
(3.58)
The significance of the minus sign in the space components of (3.58), dictated by quantum mechanics, as opposed to the plus sign in x„ = (x 0 , x), is that dM transforms like d/dx11 rather than like d/dx^. As such, the minus sign in the space component of (3.58) can be regarded as due to the metric tensor g^. Following the development of rotations and homogeneous Lorentz transformations, we now combine the infinitesimal forms of (3.57) with the closure relations (3.57). The resulting commutation relations are (see Problem 3.5) [ P „ J \ ] = 0, (3.59a) [J, v , P,] = iiP.g^ - P v „„), J
Vw Jpo] = i( n<,9vP + hP9w - J^dva - Jv„9np)-
(3.59b) (3.59c)
Inhomogeneous Lorentz Group 0> 39 Therefore the 10-parameter Poincare group (4a„, 6a»„v) has 10 infinitesimal generators (£, P, J, K) obeying the Lie algebra (3.59). Little Group. The irreducible representations of & must, of course, incorporate the representations of y discussed in Section 3.B. The inclusion of the momentum operator Pfl as an infinitesimal generator of ^ introduces the notion of a "little group" (Wigner 1939), defined by a subgroup of Lorentz transformations which leave the momentum of a particle invariant, A / A = p„,
(3.60)
where pM is an eigenvalue of the operator P^. An infinitesimal Lorentz transformation in (3.60) of the form of (3.26), A„v = 0„» +
(361)
< . p v = 0.
(3.62)
implies This latter condition is manifested if we choose «&*v = hvp°e"P° = -«&»,.» (3-63) where Qa is an arbitrary four-vector and E^,pa is the covariant extension of the Levi-Civita eijk, antisymmetric in each pair of adjacent indices, with £oo* = e .j*'
£
oi23 = 1-
(3.64)
The little-group transformations (3.60) are therefore described in Hilbert space by the unitary operator U'x = exp(-iid>MV J"v) = exp(i0„ W)
(3.65)
%=-K^PvJ^
(3-66)
with referred to as the Pauli-Lubanski four-vector and playing the role of a covariant angular momentum in (3.65). The latter analogy can be sharpened by searching for operator invariants (Casimir operators) which commute with every infinitesimal generator of the group. For 0(3) the only operator invariant is J 2 = j(j + 1), while for i£, only J 2 — K2 and J • K commute with all the group generators J, and Kt. For the inhomogeneous Lorentz group J", however, neither J 2 , J 2 — K2, nor J • K is an operator invariant; instead P^P* = P2 and Wu W = W2 commute with all group generators Jh Kh P{ (see Problem 3.6) and so are the Casimir operators for &. Since these quantities are also Lorentz invariants, they can be evaluated in any frame. In particular, in the rest frame of a particle with mass m and spin s, P^m„
= (m, 0),
WJm = (0, s),
(3.67)
and so the little group of 3? associated with a massive particle is the rotation group 0(3) with eigenvalues of the Casimir invariants P2 = m2 and
40 Transformations in Space-Time
W2 = — m2s(s + 1). These operator invariants are significant because all members of an irreducible representation of & must be eigenstates of P 2 and W2 with the same eigenvalues, P2(UaJil,}) = Va,sP2m W2(Ua^})
= m2(Ua,^»,
(3.68a)
= t4,A^2l = -m2s(s + l)([/ fl>A |^».
(3.68b)
The eigenstates or carrier space of this little group can therefore be labeled as | [m, s]p, er>, where [m, s] denotes the little-group irreducible representation (usually deleted) and p, a are the specific momentum and spin components in analogy with the 0(3) carrier space labeled by |// 3 >. Next we define spin-state wave functions as the spin-state matrix elements of the boost operator in (3.52), e ) ( p ) = < ^ | P ^ > = <^'|t/Lpk>,
(3-69)
where a refers to the spin-state row index and a' corresponds to the spinorcomponent column index. In this way, one sees that the rotation-group rest-frame eigenstates |
(3-7°)
a'
where l/A both Lorentz-transforms | p> and reshuffles the spin indices a in (3.70). The matrix D(s)(/?Ap p) is the usual spin-s rotation matrix with /?Ap,P (called the Wigner rotation) corresponding to the series of Lorentz transformations 17i (Ap)AL(p) and satisfying ^ s , f f (R A p , P )=<ApW|C/ A |p, f f > = W\UL-HJir)UJ,ULp\o\ 1
(IT (Ap)AL(p))/mv = ( r '(ApMAp), = m„
(3.71a) (3.71b)
where (3.71b) is a special case of the little-group transformation (3.60). The significance of this little-group analysis is that the simple rotation-group matrices, D(S)(R), are sufficient to describe the irreducible representations of the inhomogeneous Lorentz group 3>. Covariant Normalization. Finally we note that relativistically covariant states | pa} generated in this way are most naturally normalized in a covariant sense [recall (1.25)]:
(3.72)
where the factor of E = ym explicitly compensates for the Lorentz contraction of the momentum component parallel to the velocity transformation in the normalization integral of (3.72), and the accompanying factor of 2 is
Helicity Formalism 41 included for convenience. More specifically, the identity (recall p2 = pi — p2, E = VP 2 + m2) 0(PoMP2 - m1) = 9(p0)8[(p0 - E)(p0 + £)] = (IE)' M(p0 - E) (3.73) means that the normalization integral (recall dp = dp/In)
i S = I$* ( P o -E) = l^P^P2 - m2WPo)
(3-74)
is manifestly covariant. The factors of 2£ and 2n in (3.72) and (3.74) therefore cancel, and one is again left with a delta-function (three-dimensional) integral normalized to unity.
3.D Helicity Formalism While the rest-frame z-component of spin can be used to parametrize both the spin states and spinor components of the relativistic wave function (3.69), it is not a Lorentz invariant, but transforms according to a Lorentz transformation when the particle is boosted from rest. Another drawback to the rest-frame spin eigenstates is that they cannot be used to describe massless-particle wave functions, because E = p and v = c for m = 0, so that it is not possible to transform to a rest frame. On the other hand, the component of spin along (or against) a particle's momentum has a simple meaning in any frame and is a valid concept even for a massless particle. Thus one is led to the concept of "helicity", defined as X = s p,
(3.75)
where s is the spin of a particle and p its momentum direction. Helicity is a manifest rotation invariant; helicity states also transform simply under Lorentz transformations, as we now demonstrate. Massive Particles. Consider first the construction of a general helicity state for a massive particle m ± 0. Starting with the rest-frame spin eigenstate | (M> = | X}, where X is the J3 eigenvalue, first rotate | A> along the new z-axis p 9i0 = /?p3 as |p*> =
tf«»3l*>.
<3-76)
which leaves X unchanged [an 0(3) invariant]. Next boost the particle from rest to momentum p along the p-direction, obtaining the general helicity \pX) = ULf>\pX) = ULpURPi\X},
(3.77)
where again X remains unaltered. Using ULp URp = URp ULpi, it is also possible to reverse the order of the operation in (3.76), first boosting the particle along the p3-direction and then rotating p 3 to the general state p = p0 ^.
42 Transformations in Space-Time Massless Particles. Turning next to the construction of helicity states for massless particles, a solution in this case for the general little-group equation is Kv£*=h>
£„ = (*; 0,0, ±K),
(3.78)
which corresponds to a massless particle never at rest. As in the massive case, an infinitesimal A„v = g^ + cb^v has an antisymmetric (b^ which now obeys, instead of (3.62), cy„v£v = 0.
(3.79)
This relation constrains d)^v to only three (rather than six) independent values (e.g. a>l2,6ixo = (al3, d>20 = d)2i with associated infinitesimal generators J 3 , Ki — J2, K2 + J j , and the corresponding little group is E(2), the Euclidean group in two dimensions [and not 0(3)] (Wigner 1963). Evidently the action of Kx — J2 and K2 + J ± upon physical states has no physical significance, since they can be represented by zero, whereas exp( — iu>l2J3) corresponds to a simple rotation in two dimensions about the z-axis. To proceed further, it is useful to note the relation connecting the operators P„ and W,,: W„ = 1P„.
(3.80)
This relation follows from the definition of W^, (3.66), and the m = 0 limit of (3.68), i.e., P2 = W1 = P • W = 0. The proportionality constant in (3.80) can be interpreted as helicity, because the time component of (3.66) is W0=-$e0iJkP'JJk
= V-J,
(3.81)
so that the eigenvalues of (3.80) imply (P^ -+ /c„) X = J • k//c0, which is the definition of helicity (3.75) for massless particles, k = kk0. Since the eigenmomenta must be of the form of the little-group momenta (3.78b), it is clear that A = J-£=±|J3|.
(3.82)
Thus, a massless particle with spin is specified by only two independent helicity states, no matter how large s2 = s(s + 1) may be. This result is independent of the boost operation for massless particles defined by fc„ = ^/(k)£ v ,
|kA>=l/^ (k) |A>,
(3.83)
which boosts the two helicity states from momentum | K | = K0 = K up to |k| = k0 via exp( — i[,K3) with £ = log(|k|/K) and rotates k away from the K3-direction (Problem 3.8). We shall not, however, dwell upon the details of massless-particle boosts (see Weinberg 1964b). Common Properties. Even though there are 2s + 1 helicity states for a massive particle of spin s but only two states for a massless particle, it can be shown that the m = 0 limits of the massive states (3.77) are indeed massless states (3.83), with the corresponding if irreducible representation (A, B)
Helicity Formalism
43
giving X = B — A in the massless limit (Weinberg 1964b). There are, in fact, many properties which the massive and massless helicity states have in common, such as covariant normalization, < P T|pA> = 2E<5 3 (p'-p)<W
(3.84)
which will be a necessary condition in order to pass from m =/= 0 to the m = 0 limit. It is also possible to combine a state containing two particles with helicity (either massive or massless) into a single helicity state, | Pi Aip2 A2> ~ I Q^X in the "center-of-momentum" system p : + p 2 = 0, where px = qe ^, p 2 = — q = qn_fl n+(l>, and X = Xt — X2. Furthermore, an angular-momentum helicity state can be obtained from a massive or massless helicity state as | pjmX} oc J dQ. Dll\*(ReJ | p ^ X \
(3.85)
which in turn can be used to form a covariant partial-wave expansion analogous to (1.31) without reference to the nonrelativistic decomposition J = L + S, by replacing P, with D^\ instead. Helicity states have definite drawbacks, however, not the least of which is the problem of selecting proper phases. Fortunately, we shall be able to avoid these subtleties in practice by summing over all helicity states in a coherent fashion for quantities such as £ i//wil/*w. Projection-operator techniques also will be developed to select out specific helicity states from such a helicity (spin) sum (see Section 5.C). M-Function. An important link between Lorentz invariance and Poincare invariance is provided by helicity states (or rest-frame spin states for massive particles) and the S-matrix. In Sections l.B and 2.D we have exploited the rotational invariance of the S-matrix, and in Section l.D we have alluded to its dynamical content. Now we wish to explore its behavior under Lorentz transformations. Consider the helicity-state S-matrix elements
(3.86)
the helicity matrix elements (X\ X'2\S\ Xt X2} have rather complicated transformation properties under the action of the Lorentz group, the helicity components being "shuffled" by the Wigner rotations (3.70). It is customary to display this helicity dependence by factoring out of S the helicity wave functions of each particle, the remaining "M-function" then transforming simply under J? (dropping any energy- or momentum-conserving delta functions and assuming each particle has spin 1), (X\ X'2\S\ X, X2} = ^'»*^ 2 V»*M'"'«>-'' 2 ^V^ 2 ) -
(3-87)
Thus Mfll.fl2.pilfl2 is called the M-function for the process, transforming according to the homogeneous Lorentz group (spinor transformations will
44 Transformations in Space-Time
be considered in Chapter 5), M' • . 1 J
= A .V1'A .V2'A
flC2 >if2
'Vi
'V2
Vi
V1
V2
A 'V2
M . . m
»iv2v1v2-
(3 88) y->.oo)
As such, AfM1.M2>1/12 is a covariant cartesian tensor, composed of particle momenta and spin tensors, to be constructed in detail later. In (3.87) the momentum dependence of the S-matrix and wave functions has been deleted for brevity. Restoring it corresponds to expressing initial wave functions = •A^HP) I P^> a n d final wave functions IA* with the covariant cartesian tensor index (j. acting like a bra state for a spin-1 Poincare-group state iff p. For higher-spin particles, a higher-rank traceless and symmetric covariant tensor can be constructed along the lines of the rest-frame tensor wave functions of Section 2.E. Spin4 Helicity States. It will be useful to display spin4 and spin-1 helicity eigenstates explicitly, in parallel with the rotation eigenstates of Section 2.D. The s = -j helicity rest-frame eigenspinors for massive particles,
(3.89)
where X = ±\. These wave functions must also transform according to the irreducible representation of 0(3); in particular (3.76) implies
(3-90)
where D{£\ is the rotation-group matrix, given by (2.37) and (2.38) for) = \ and m -> X = +-j (see also Problem 2.5 or Appendix II), and ^ M - U s m * * }
*
< M ( |
H
cosi/)'
<3'91>
where we have taken (for 9 =
^*)(P3)=(i),
(3-92)
according to (3.89) and the Pauli matrices (2.48). Next apply the boost operators for j = ^, obtaining [for example from (3.50a)] the (j, 0) wave function describing a free particle moving with momentum p, helicity X, and covariantly normalized:
(3.93)
Here we have explicitly removed a factor (2m) - * from the boost (3.50a) because the covariantly normalized rest-frame spinors are ^Jlm <pa){p). Then
(3.94)
Helicity Formalism 45 by virtue of (3.89). Clearly only the X = —\ left-handed state survives for the (j, 0) representation in the massless limit, because the X = \ state must vanish by (3.94). Replacing p by - p in (3.93) and X by — X in (3.94) converts the (\, 0) to the (0, \) representation. Therefore only the X = j right-handed state survives for the (0, \) representation in the massless limit. Consequently, while massive spin4 particles (e.g., electrons, positrons) can have both helicity states X = +\, massless spin4 particles (e.g., neutrinos, antineutrinos) must have their momenta lined up along or against their spin, with only one configuration allowed for each irreducible representation of i£. We shall conclude later that nature chooses the (j, 0) representation for the (left-handed) neutrino and the (0, ^) representation for the (right-handed) antineutrino. Spin-1 Helicity States. Turning to the construction of spin-1 helicity eigenstates, it is most convenient (but not required) to choose (^, | ) helicity polarization vectors ef] in parallel with 0(3) polarization vectors sSm) and S£ wave functions (f)^. Identifying £^](p3) with rest-frame helicity states |p3A> for s = 1 and X = ± 1, 0, the covariant analogs of (2.43) are (again following the CSR phase convention) < ± 1 ) ( P 3 ) = + ( 0 ; 1 , ±U0)/y/2, 40,(P3) = (0; 0, 0, 1),
(3.95a) (3.95b)
where X = + 1 (— 1) corresponds to a right-handed (left-handed), transverse, circularly polarized helicity state, and X = 0 is a longitudinal state polarized in the z-direction for massive spin-1 particles. Note that with the rest momentum m^ = (w, 0), (3.95) leads to the "subsidiary condition" for X = ± 1, 0, "•"^(Pa) = 0-
(3-96)
Next rotate p 3 to p = Rp3 according to (3.76), giving (as in Section 2.D, the functional form of the polarization vectors remains unchanged) ^(P) = K^](U
(3.97a)
since e^ is a covariant four-vector, or alternatively, ^ ) (P) = I ^ ' ) ( P 3 M 1 1 ( ^ ) ,
(3.97b)
A'
similar to (3.90). Either (3.97a) or (3.97b) generates ef 10) (p) f° r general pfl>), but to simplify these expressions it is convenient to choose pe 0 with > = 0, corresponding to an active rotation by 9 about the y-axis in the x-z plane (see Problem 2.5 or Appendix II): elr M(P) = + (0; cos 9, ± i, - sin 0)1J% eJP'tP) = (°; s i n 0> 0, cos 6).
(3.98a) (3.98b)
46 Transformations in Space-Time
Not only is the meaning of helicity preserved by such a rotation, but so too is the subsidiary condition m"£^(p) = 0,
(3.99)
since the particle is still at rest. Finally, boost the particle along p from rest to momenta p via p^ = L^v(p)mv with L^v(p) given by (3.22): « ! » = V(p)e<*>(p). 2
(3-100)
2
This gives (with E = p0 = ^Jp + m , p= | p |, and
(3.101a)
^ ' ( P ) = + ( ° ; cos 9, ±i, -sin 9)1^/2,
(3.101b)
0)
4 (P) = (P; £ sin 0, 0, E cos 0)/m,
(3.101c)
with a corresponding subsidiary condition p"<>(p) = 0.
(3.102)
Thus, at all stages in the construction of the general spin-1 helicity states, a subsidiary condition naturally arises which guarantees that only 3 out of 4 components of d*) are independent, with the spin-0 part of the (^, ^) representation of J5f projected out according to the Clebsch-Gordan decomposition by (3.102). Photon Helicity States. It is also possible to pass to the m = 0 limit for spin-1 particles (i.e., photons), with only two helicity states X = ± 1 by virtue of (3.82). The little-group standard four-momentum (3.78b) is associated with eJ, ± 1 ) (K 3 )=T(0;l, ±U 0)1^/2.
(3.103)
After a boost and rotation to k = kfl 0 (with a> = k0 = | k |), (3.78b) and (3.103) become, in analogy with (3.101), kli= {a>;a) sin 9, 0, a> cos 9), <*"«-*.•«•?'*.>
(3.104a) (3104b)
= +(0; cos 9, ±i, -sin 6)/y/2, so that k»e(ll±1)(k) = 0. (3.105) As (3.105) only reduces the number of independent components of e„ from 4 to 3, it is not sufficient for describing a massless spin-1 photon, which can only have 2 independent components in e^. In group-theoretical terms, this problem arises because e„ transforms according to the (^, ^) representation of !£ with an 0(3) little-group structure when k-e — 0. A massless spin-1 wave function, however, is supposed to have a little-group transformation structure of E(2), the Euclidean group in two dimensions. The problem can
Helicity Formalism
47
be resolved by first noting the simple transformation law satisfied by e^ for transverse polarized helicity state (3.104b), #= u(A(R)k) = RJe? "CO = e 1 8 * ^ 1}(k),
(3-106)
where R is an active rotation by 0A about the £-axis. For little-group Lorentz transformations of the form (3.78), a transformation law similar to (3.106) holds for massless polarization vectors: A/ei* "(K3) = e™h?
1
>(K3)
+
z±
£>,
(3.107)
which satisfies (3.79) for infinitesimal A^v = g^ + 01^, with 0 = d>12 and X± = +(cbio ± i
(3.108)
it is possible to recast the transformation law (3.107) in terms of a general Lorentz transformation A as (Weinberg 1964c; see Problem 3.8) A/ - A o ' M e ^ A - ' k ) = e + ' ^ ^ f k ) .
(3.109)
The nonmanifest covariant nature of the A0V term in (3.109) corresponds to the radiation or transverse "gauge" for 4 ± n > namely e^*1' = 0, which was assumed in (3.103) and is now a consequence of (3.109) as well. It is possible to choose a different gauge for ej,11) and obtain a transformation law similar but not identical to (3.109) with a modified /cM term. This ambiguity will be resolved later by demanding that a choice of gauge cannot lead to physical consequences (gauge in variance); this will be the needed constraint, along with the subsidiary condition (3.105) which reduces the four spin states of eJ,A) to two for massless photons. General references on Lorentz transformations and the Lorentz group are: Schweber (1961), Bjorken and Drell (1964), Weinberg (1964a-c), Akhiezer and Berestetskii (1965), Gasiorowicz (1966), Martin and Spearman (1970), and Bargmann (1970).
CHAPTER 4
Boson Wave Equations
The dynamical quantum-mechanical wave equations of spin-0 pions, spin-1 vector particles, and massless spin-1 photons are formulated in a consistent one-particle fashion. For the spin-0 Klein-Gordon equation, the interpretation of negative-energy states as describing antiparticles is stressed. The relativistic bound-state Coulomb problem is then solved forrc-mesicatoms. The parallel is made between the massive spin-1 and photon wave equations. The notion of currents, current conservation and gauge invariance for photon amplitudes is discussed in detail and linked to the principle of minimal replacement. Minimal coupling of photons to charged particles will be the basis of the general electromagnetic interaction to be considered in later chapters. Second-quantized field theories are briefly described, and an analogy is made between (relativistic) photons and nonrelativistic phonons.
4.A Spin-0 Klein-Gordon Equation Derivation. For a particle moving at relativistic velocities (i.e., having a kinetic energy that is a substantial fraction of its rest mass), the nonrelativistic approximation for energy, E = m + p2/2m, is no longer valid and one must use instead the exact relation £ = (p2 + m2)*. The formal quantummechanical replacement p -» — iV would then result in the Hamiltonian H = ( - V 2 + m2)* 48
(4.1)
Spin-0 Klein-Gordon Equation 49 and a free particle Schrodinger-type equation id,il/ = ( - V 2 + m2)H.
(4.2)
Since the square-root operation in (4.1) and (4.2) is difficult to interpret, it would seem more reasonable to construct a relativistic wave equation associated with the square of the Hamiltonian operator, (idt)2(p = H2
(4.3)
where pM = id^, one is naturally led to the free-particle Klein-Gordon equation ( • + m2)(t>(x) = 0,
(p2 - m2)
(4.4)
2
Such a particle is said to be "on its mass shell", p = m . Covariance. The manifest covariance of (4.4) for a quantum-mechanical (Lorentz) scalar wave function, >'(x') = >(x), i.e., for | = l/A \
(4.5a)
4>'{x) = <x |
(4.5b)
implies that (a;3'" + m2))'(x') = (d^d" + m2)
(4.6)
Due to these transformation properties, the wave function + (x)oce- ip ' x = e' p "e- , ' £ ', f ( x ) o c c i ' J = e"ipV£'.
(4.7)
These equations are special cases of the general solutions of the KleinGordon wave equation containing a possible interaction term, cP±(x) = (t>±(x)e+iEt.
(4.8) 2
It is therefore clear that the use of the operator H in forming a wave equation leads to seemingly unphysical negative-energy solutions e+,Et as well as physical positive-energy solutions e~'E' for the quantum mechanical state of the particle. While this problem led to a temporary discarding of the Klein-Gordon equation in the late 1920s and early 1930s, we have since learned to live with it, as will be discussed shortly. Probability Current. Another problem which arises with solutions of (4.4) is the construction of a positive definite probability density. Paralleling postulate v in Section LA, one searches for a covariant probability current density
50 Boson Wave Equations h — (P> J) which obeys a continuity equation d% = 8% + dlh = 8, j0 + V • j = 0,
(4.9)
where we have used (3.60), 3" = (d„ V). The obvious candidate for j^ is the hermitian form 7„(x) = 4>*iS,4> = 4>*tt>4> - ' K >*)>,
(4-10)
because this current is conserved (d • j = 0) for states > and »* obeying the Klein-Gordon equation (4.4): d-j = id"(c/>*5M>) - i5"(05>*) = i>*D<£ - *'>•>* = 0.
(4.11)
Furthermore, since the spatial part of (4.10), j , is identical in form to the nonrelativistic current density (1.6) except for normalization, we are obliged by covariance arguments to accept the timelike component of (4.10), p, as the probability density. When combined with the general solutions (4.8), this probability density becomes ? + (x) = c/>*;30(K=2£|(/> + ( x ) | 2 > 0
(4.12a)
for positive-energy solutions, and p.(x) =
(4.12b)
for negative-energy solutions, with E > 0. Clearly a negative probability density is unacceptable; we shall contend with (4.12b) shortly. Wave Packets. Construct general spin-0, free-particle wave packets in the Hilbert space of positive and negative energy states as (normalized in a box—see Section l.B) , + (x) =
I2£^a'e"""JC'
*-W-J2§* *'*"'*
(4-13)
where the complex conjugation of bp follows the usual convention. The factor of 2£ (E > 0) in (4.13) is a manifestation of our covariant normalization of states. From (4.13) it is clear that the evolution of these packets does not alter their positive- and negative-energy character. Consider then a scalar product defined over the positive-energy states (4.13a) as (see Problem 4.1) <<*>'+,
(4-14a)
= j ^ 4 %
(4.14b)
= J d*p 8(p2 - m2)9(p0)a?ap/V,
(4.14c)
where 8(p2 — m2)6(p0) in (4.14c) indicates that only positive-energy states with p2 = m2 are allowed. Note that this norm is time independent [differen-
Spin-0 Klein-Gordon Equation 51 tiate (4.14a) with respect to time and use the Klein-Gordon equation; the result is obvious for (4.14b) and (4.14c)]. Note too that the norm (4.14) is a Lorentz invariant. Also, the covariant normalization of states,
(4.15)
for a plane wave of momentum p,. Interpretation of Negative-Energy States. Historically, the resolution of the negative-energy and negative-probability-density problems led to a reformulation of the Klein-Gordon theory in a many-body context. It is possible, however, to stay within a single-particle framework (Stuckelberg 1941; Feynman 1949) by interpreting (4.12a) as the charge density of a positively charged, positive-energy particle "propagating" forward in time (t > 0, E > 0) via the plane-wave phase e~iEt. Similarly, one interprets (4.12b) as the charge density of a positively charged, negative-energy state propagating backward in time (t = — |r| < 0) via eiEt = e~im. Alternatively (4.12b) is the charge density of a negatively charged, positive-energy particle propagating forward in time via the complex conjugate of the phase, e'E' = (e~'E')*. For neutral particles which are their own antiparticles (i.e., 7t" = n°, where c refers to the "charge conjugate" antiparticle—see Section 6.A), one can choose the wave function to be purely real or imaginary. In this case the probability density (4.12) vanishes, consistent with treating p as a charge density. Unfortunately, a thorough understanding of this interpretation must await a discussion of charge conjugation in Chapter 6 and "backward propagation in time" in Chapters 7 and 10. The Stiickelberg-Feynman interpretation is ideally suited for scattering processes, where the particle is free and unlocalized before and after the collision. For bound-state wave packets, however, a particle constrained to Ax < m"l and Ap ~ 1/Ax ~ m demands a superposition of all Fourier components, negative as well as positive energy:
>(*) = J ~ i K e - i r , +
fcpV'-).
(4-16)
Then (f)*4> contains interference terms e±2lEl which produce violent oscillations, referred to as "Zitterbewegung". Since such "jittery-motion" plays a more significant role for spin4 particles, we postpone a detailed discussion of it until Chapter 5. Suffice it to say that as E -> m, such interference between positive and negative energy components could have physical consequences. Feshbach-Villars Formulation. It is possible to circumvent these interference terms by constructing a Klein-Gordon bound-state wave function which has no negative-energy component in the nonrelativistic limit (Feshbach and Villars 1958). For (p satisfying the Klein-Gordon equation with time
52
Boson Wave Equations
derivative fy, define
»"i(* + s 4
x
(417)
-i(*-i*\
Because (j> = 0 + -+ e~iml as £ -*m implies q> -*• e~imt and x -* 0 by (4.17), (p is called the "large" and x the "small" positive energy component of <j>. Moreover q> and x satisfy coupled first-order equations in time, V2 iq> = -—
((p + x) + m(p,
V2 ix ^ , (•? + *) _ =
m
*-
(418)
These equations can be unified into one matrix equation satisfied by the column vector *=(*)'
(419)
and resembling the Schrodinger form ia,
(4.20)
where H0 is the 2 x 2 matrix "hamiltonian" H0 = ^(1 + a)p2/2m + pm,
(4.21)
with
'-(1-tt —C i> While the positive- and negative-energy states are coupled in this scheme, the problems associated with negative energy are not completely eliminated, because H0 as given by (4.21) is not hermitian in that
This links the large and small components q> and x together, implying as in (4.12) that the probability density p is not positive definite. If nothing else, however, this approach indicates that a second-order equation in time can be reduced to a first-order equation by doubling the Hilbert space of states via the column vector (4.19). [From an historical standpoint, Dirac learned this fact 30 years earlier when he discovered the first-order Dirac equation for spin-j particles (Dirac 1928).] Furthermore this formalism is ideally suited for nonrelativistic reductions. We will exploit a similar pattern in the case of the Dirac equation in Chapter 5. External Fields. Next we consider the modification of the Klein-Gordon equation for spin-0 particles in the presence of an electromagnetic field, specified by the vector and scalar potentials A and A0 as the four-vector Ap = (A0, A). Following the minimal substitution procedure of classical
Spin-0 Klein-Gordon Equation 53 physics, we take #„ = p„ - p„ - *Ki (4-23) where e is the electric charge of the particle, taken as positive unless otherwise specified. The Klein-Gordon equation then becomes [(id - eAf - m2](f>(x) = 0,
(4.24a)
or (O + m2)(p(x) = -J(x),
(4.24b)
where J(x) is a scalar current "source" density J(x) = 2ieA • d
(4.25)
or J = V
(4.26)
The doubling of the A, term in (4.26) is a consequence of A^ being real. Following the procedure (4.11), one can demonstrate that (4.26) is also conserved, since in this case d • j = i
d((p*
by use of (4.24b) and (4.25). Bound-State Coulomb Atom. Finally we consider the specific bound-state Coulomb problem of a spin-0 n~ particle with charge —e bound to a heavy nucleus with charge Ze (n-mesic atom). The static potential for this configuration is (a = e2/4n) eA, = Vd^o = - ^ 8,0.
(4.28)
Writing the positive-energy wave function as (p(r, t) = cp(r)e~iE', the spatial part obeys a Klein-Gordon equation obtained from (4.24):
K)
+ V2 - m2 4>(r) = 0.
(4.29)
This latter equation is solved by the standard method of separation of variables with 0(r) =
(4.27)
54 Boson Wave Equations In the limit £ -» m, E2 — m2 -»• 2m£NR, and Ztx
1(1 + 1) r
2mZ«" <j>{r)= -2m£ NR ^(r). r
(4.30b)
We are familiar with the solutions of (4.30b) for integer values of n — / = 1, 2, ..., corresponding to the Bohr energy levels for n. = 1, 2, ..., m(Za)2
E - m = £NR =
(4.31)
Given (4.31), we can infer the Klein-Gordon energy levels by replacing m-> E, £NR -»(£ 2 - m2)/2E [by inspection of (4.30a)], and n -> n' [see Schiff (1968)]: £ 2 - m 2 = -£ 2 (Za) 2 /n' 2 .
(4.32a)
This relation can be solved explicitly for the relativistic energy as £ =m 1+
(Za)2
(4.32b)
„'2
Here ri is a new relativistic principal quantum number with ri — I' equal to the usual nonrelativistic quantum-number difference n — I, assuming only the integer values ri — l = 1,2, 3, ...; and /' is the effective relativistic orbital angular momentum, inferred from (4.30a) to be /'(/'+l)=/(/+l)-(Za)2.
(4.33)
Solving (4.33) for /' gives ' ' = - i + [(' + i ) 2 - ( Z a ) 2 ] * , „' = „ _ / + / - = „ _ ( / + -y + [(/ + 1)2 _ (Za)2]*,
(434a) (4.34b)
where the positive sign of the square root has been chosen in order that /' may be nonnegative as Za -• 0, corresponding to bound radial solutions r' regular at the origin. This form of ri, (4.34b), is to be applied to the energy levels (4.32). For Za <^ 1, both (4.32b) and (4.34) can be expanded in the form of (4.31) with n = 1, 2, ... (see problem 4.2), £ —m
J
NR
m(Za)2 ~~2rF
1+
(Za) n V+ i
An)
(4.35)
This removes the / degeneracy of £ in the 0(a 4 ) relativistic fine-structure term. For large Z, (4.34) provides the constraint that for /' and ri real, the discriminant of the square root must be positive (of course, it cannot vanish) and for s-waves (/ = 0) this means that „, Z <
1 2a
=
137 lT
(4.36)
Spin-1 Wave Equation
55
If instead Z > l/2a, the centrifugal barrier term in (4.30) becomes attractive for s-waves and the energy becomes imaginary, so that the wave function has a damped exponential part and the particle orbits become unstable. The situation is similar to the classical relativistic situation for the Coloumb potential; for L2 < (Za)2 the centrifugal barrier also becomes attractive and the particle spirals in to the origin. Multiparticle quantum states then presumably play a role, with short-distance corrections such as vacuum polarization by pair creation (see Chapter 15) modifying the single-particle wave function. In the next chapter we shall again return to this strong-field limit for the case of bound electrons.
4.B Spin-1 Wave Equation Derivation. Starting with a three-component wave function >, describing a massive spin-1 free particle in its rest frame, two possible rest-frame covariant forms exist: a covariant four-vector ^ = (0, >,) and a rank-two antisymmetric (field) tensor >MV given by (recall the angular-momentum tensor operators LMV and J//v in Chapter 3) 0 Oi = — c/>l0 = 5f <^; and cf)00 = (frjj = 0. In a general frame, the boosted form of »/JV can be obtained from (/>M as ^v = 3 ^ v - 5 v ^ .
(4.37)
The free-particle dynamical relation between <\>^ and >/lv is called the Proca equation: <^„ v = m2tfv
(4.38)
Owing to the antisymmetric structure of (j)^, the derivative of (4.38) implies the subsidiary condition 5 " ^ = 0.
(4.39)
Since 0„ must transform according to the (^, j) representation of if, we know that (4.39) is required to rule out the spin-0 component in (/>,, (see Section 3.B). This in turn provides a group-theoretical justification for the dynamical equation (4.38). Moreover, using (4.37) to eliminate (j)^ in (4.38) and applying (4.39), we are led to a Klein-Gordon equation for the vector wave function, ( • + m2)
(4.40)
which, as in the spin-0 case, guarantees the correct dynamical relation between energy and momentum for a free particle. Current Densities. Given the wave functions (^v and >„, it is possible to construct a conserved, hermitian current density analogous to (4.10) but describing spin-1 probabilities:
56
Boson Wave Equations
The field equation (4.38) along with (4.37) then manifests 8 • j = 0. As was the case for the spin-0 probability density, j0 = p obtained from (4.41) is not a positive definite quantity. Likewise, negative energy and Zitterbewegung problems again arise, to be handled in a manner similar to spin-0 particles. We therefore proceed to indicate the new aspects associated with spin-1 wave functions. In the presence of an electromagnetic field, the minimal replacement (4.23) converts the field equation (4.38) from the spin-1 free-particle Klein-Gordon equation (4.40) to a similar equation containing a vector source current (see Problem 4.3), (a+m2)
(4.44)
with the polarization vector further specified by the helicity eigenvalues, E^\p) and k= ± 1, 0. The subsidiary condition d • (f> = 0 is therefore equivalent to p • e = 0, as derived earlier in (3.102). As in the spin-0 case, a factor of (2£)~* in (4.44) has been absorbed into the covariant normalization of the states, and a one-particle volume normalization factor of V - * has been set equal to unity. The orthogonality of the wave functions (4.44) then demands eW')*(p^>(p)=-^.A,
(4.45)
a result which can be verified from the specific forms (3.98) and (3.101). Likewise, in the rest frame of the particle, the completeness property of the polarization vectors implies £#>(PMA)*(P) = <5.7. >. and since <5y can be expressed in four-dimensional ~(9w — mllmv/m2), the boosted form of (4.46a) is I^)(PKA)*(P)=-(^V-^).
(4.46a) language as
(4.46b)
Spin-1 Maxwell Equation
57
The p-meson (mp « 776 MeV) and the co-meson (m„, « 783 MeV) are examples of spin-1 particles occurring in nature. They are significantly heavier than the spin-0 7t-meson (m, x 140 MeV).
4.C Spin-1 Maxwell Equation A massless spin-1 photon obeys a quantum-mechanical wave equation somewhat similar in structure to the massive one, with 0M and <^/JV respectively replaced by the four-vector potential A^ and the antisymmetric field tensor F /lv , where (4-47)
F„v = d„4 v - M M The Maxwell free-field equation is then [setting m = 0 in (4.38)] -d*F„v = nA„-dft(d-A)
= 0.
(4.48)
We take this as the dynamical wave equation for noninteracting photons. Existence of Gauges. Note now that a subsidiary condition d A = 0 is no longer a direct consequence of the field equation itself, as was (4.39). Rather, an ambiguity now exists, and the value of d • A is correlated with the way we restrict the number of spin states of the photon to two, as required by the helicity constraint X = ± 1. For plane waves, the representation for A^ satisfying (4.48) for a noninteracting photon with k2 = co2 — k 2 = 0 has the general form Att(x) = sli(k)e-ikx.
(4.49)
In Section 3.D we chose the "transverse gauge" for ej,* 1}(k) in order to insure only two independent components in e^: £f)
±1)
(k) = 0,
k-£ ( ± 1 ) (k) = 0.
(4.50)
But we cannot separately choose both conditions (4.50) in a frameindependent manner. A Lorentz-invariant choice is the combination k • s(k) = 0, and (4.49) then corresponds to the Lorentz gauge for (4.48): d-A = 0,
D ^ = 0.
(4.51)
While (4.51) is not the only gauge decomposition of (4.48), it is the natural choice for free photons, paralleling massive spin-1 particles. The transverse or radiation gauge is d • A = 0. In any case, the orthogonality and completeness relations for photon polarization vectors are analogous to (4.45) and (4.46): (4.52) e*w')(k)-ew>(k)=-^, valid in any gauge, while X e^(k) £ f*(k) = < 5 . v - ^ x= ± i
is valid only in the transverse gauge (4.50).
(4-53)
58 Boson Wave Equations
Gauge Invariance. It is possible to transform A^ to a different gauge so that (4.50) and (4.51) no longer hold but (4.48) remains an identity. This is achieved by a gauge transformation A^A^
+ B^ix),
e„-> e„ + ?*„,
(4.54)
kx
for a plane wave;
with £(x) an arbitrary scalar function becoming i^e~' thus d • A and k • e are transformed to d-A-»nt(x),
fc-e-»0.
(4.55)
1
Consequently, while k • e must always vanish for k = 0, d • A vanishes only in the Lorentz gauge (4.51). The connection between the minimal coupling replacement id^ -> id^ — eA^ and the gauge transformation (4.54) is the principle of "gauge invariance of the second kind". This is the coordinatedependent phase transformation of a charged-particle wave function,
(4.56a)
"where £(x) generates the gauge transformation (4.54), under which the canonical momentum transforms simply: (id„ - eA,M>{x) - g" «M(i3„ - d > ( x )
(4.56b)
Observables such as the current densities (4.26) and (4.43) are built up from bilinear products as >*> and >*(«?,, — eAjcj), manifestly invariant under the phase transformations (4.56); thus current conservation is naturally extended by this principle to include interactions in the presence of electromagnetic fields. From our viewpoint this is further justification for considering minimal replacement as a fundamental principle which generates the only interaction between charged particles and photons. In the context of lagrangian field theory, the principle of gauge invariance of the second kind [the first kind corresponding to a constant phase in (4.56) and linked to charge conservation] plays the central role and is sometimes considered the raison d'etre for the existence of A^ itself and its interaction with charged matter. Gauge principles recently have been used to generate other fundamental interactions (strong and weak), but such topics are beyond the scope of this book. One final fact about gauge invariance of significant import for us later will be the manner in which the two physical spin states of the photon are realized for a general interaction with matter. This is most conveniently stated by the Lorentz-invariant S-matrix element, itself expressed in terms of the M-function of (3.87). Accordingly, we may write Sw(k) = e^(k)M"{k), w
(4.57)
where S (k) and the polarization vector ef(k) represent a photon of momentum k and helicity L While M^ transforms like a simple four-vector under £?, we recall that e^ has a slightly modified transformation law (3.109) because a photon really behaves like an E(2) [and not an 0(3)] object under a little-group transformation. Applying (3.109) to (4.57) and using
Spin-1 Maxwell Equation 59 A / = (A" 1 )^ then leads to (see Problem 4.5) Sw(k): = eikB A / - A
0
^
eiX)(A~
^M^k)
CO
= eue £ W(A- 1 k)M"(A" 1 k) - A 0 v el A) (A" 1 k)^M' i (k) (4.58) Aside from an overall phase factor (of no consequence for physical probabilities), the Lorentz-invariant S-matrix transforms as a simple scalar (Weinberg 1964c) Sw(k) = ea@Sw(\~1k). (4.59) Then comparing (4.59) with (4.57) and (4.58), we see that MM must satisfy an additional constraint, *"M„(k) = 0.
(4.60)
This condition is sometimes referred to as "on-shell" gauge invariance (onshell refers to physical, as opposed to virtual or off-shell, energy or momentum—more about this later). It guarantees that e^l) transforms properly under little group transformations. The reference to gauge invariance means that (4.57) is invariant under the gauge transformation
(4.61a)
or in terms of the dynamical (Klein-Gordon-type) equation for /4„ in the Lorentz gauge (d • A = 0), OA,(x) = ejll(x)^rir(x).
(4.61b)
Note that the sign of the spin-1 source current in (4.61b) is opposite to that of the spin-0 source current in (4.24b). As a consequence we shall show later that this sign change is linked to a fundamentally attractive spin-0 force (e.g., the pion-exchange strong force) and a fundamentally repulsive spin-1 force (e.g., the photon-exchange electromagnetic force between particles of like charge). It is clear from the divergenceless nature of the left-hand side of (4.61) that this charged-matter current must be conserved: d • j(x) = 0. As noted after (4.56), this result is also a consequence of the quantum-mechanical structure of the probability current density, as given for charged spinless particles in (4.26), coupled with the principle of gauge invariance of the second kind as
60
Boson Wave Equations
applied to such particles via (4.56). The connection between current conservation and gauge invariance takes on further significance when formulated in momentum space. For a free, spinless, charged particle absorbing the four-momentum q = p' — p from a photon, the current density constructed from positive-energy packets (4.13a) becomes, via (4.15),
= \ji\jf
and the differential operator id^ in (4.62) requires the off-diagonal momentum-space current to be
(4-63a)
2
for a free photon, q = (p' — p) = 0. On the other hand, in the presence of an external electromagnetic field, q1 = (p' — p)2 ± 0; and on grounds of Lorentz covariance and current conservation, the off-diagonal momentumspace electromagnetic current must have the form e
(4.63b)
Here the charged particle is still considered as free, as in initial- and finalstate scattering configurations. Since d • j oc ^
+ p)
= F(q2)(m2 - m2) = 0,
(4.64)
which demonstrates the absence of the only other possible covariant in (4.63b), q^ = (p' — p)^ The dimensionless Lorentz-invariant function F(q2) is called a form factor. It represents all possible interactions between the charged particle and photons, altered only by "strongly interacting" particles also interacting with the spinless particle in question. At q2 = 0, (4.63a) demands that F(0) = 1, regardless of the type of interacting particles present. The conversion of the free-particle current [(4.10) in coordinate space or (4.63a) in momentum space] to the interacting forms (4.26) or (4.63b) for q2 41 0 is a complicated dynamical process which we shall attempt to explain in the latter half of this book. At the present level of discussion the interesting connection is between the dynamical current-conservation statement (4.64) and the kinematical on-shell gauge-invariance constraint (4.60). For a free photon with momentum k and k2 = 0, (4.64) is a special case of (4.60) (with q replaced by k). An analogous situation exists for spin-1 matter currents based upon (4.43) and also spin4 matter currents to be described in the next chapter.
Second Quantization: Photons and Phonons 61
4.D Second Quantization: Photons and Phonons Thus far we have treated a photon as a "particle" having a quantummechanical wave function obeying the dynamical wave equation (4.48) for a free photon and (4.61) for a photon interacting with charged matter. Needless to say, these quantum field equations are identical in form to Maxwell's equations for the classical electromagnetic field tensor, related to physical electric and magnetic fields as -Ei = F0i = dtAt-dtA0, Bt = hijkFjk = eiJkdjAk.
(4.65a) (4.65b)
This complementarity relation between a quantum particle and a classical wave or field can be extended one step further, to a quantum field, then referred to as second quantization. Noninteracting Photons. In such a quantum field theory, the vector potential of the radiation field is scaled to the space-time-averaged energy flux (Poynting vector), which in rationalized units (restoring h and c here) is <S> = c<ExB> = £ — |Ak|2fc,
(4.66)
c
k
where Ak is the Fourier component of the vector potential A(r, t) = X (A k e*- r e -to « + Aj
(4.67)
k
[Note that we have anticipated that A will be treated as an operator and have used Af instead of A* in (4.67).] If (4.66) is to represent a flux of photons of density N/V with energy hco and angular momentum h, then <S> = hcocN/V implies |A k | 2 = ^
-
(4.68)
In a second-quantized theory, (4.68) is interpreted as a matrix element
and
= Nk.
(4.70)
We see that the second-quantized field operator Ak has the same structure as the single-photon wave function (4.49), now noncovariantly normalized, except for the second-quantized operator ak, where a[ a^ is a number operator according to (4.70).
62 Boson Wave Equations
To proceed more rigorously, one treats (4.67) as a normal-mode harmonic-oscillator expansion and expresses the field energy as E = \ \d3x(E2 ^
+ B2y = YJh(o(Nk + $),
J
(4.71)
k
where use of (4.65), (4.67), and (4.69) leads to (4.71) with the identification (now being careful to respect the order of the operators A and Af)
= 2Nk + l.
(4.72)
Then abstracting (4.71) to a hamiltonian operator expressed in terms of canonical coordinates and momenta, the commutation relation [xk, pk] = ih leads to the second-quantized commutation relations (see Problem 4.6) [ak, al] = (5k,k.,
K , Ok-] = 0.
(4.73)
It is therefore clear from (4.72) and (4.73) that ak is an annihilation operator and al a creation operator in Fock space satisfying
(4.74)
and consistent with a^ ak being the number operator as expected in (4.70). Spin and Statistics. Second quantization is a natural way to build in the Bose statistics of the photon field from the outset via the commutation relations (4.73). Likewise the Bose statistics of spin-0 or spin-l massive particles can be built in by commutation relations such as (4.73) to form a similar free quantum field theory. A field theory of fermions also can be constructed in this way, with the commutation relations (4.73) replaced by anticommutation relations in order to manifest the Fermi statistics. This connection between spin and statistics follows in a natural way from the requirement of positive definiteness for the free-field hamiltonian (Pauli 1940). Interacting quantum fields are described in terms of coordinate-space lagrangian densities. While the methods of quantum field theory are elegant and powerful, we shall follow the simpler and intuitive (one-particle) methods of Feyman for the greater part of this book. This means we will be investigating the structure of (one-particle) current and hamiltonian densities (usually in momentum space) rather than lagrangian densities (in coordinate space). The connection between spin and statistics will be invoked as a postulate, as elsewhere in quantum mechanics (postulate vi, Section l.A); no reference will be made to the second-quantized commutation relations (4.73). Phonons. Before leaving this subject, it will prove useful to consider a nonrelativistic quantum field construct, that of phonons, corresponding to boson quanta but associated with lattice vibrations in a solid. Like photons, phonons have no mass—but this is not dynamically relevant. Instead the relation between phonon energy (frequency) and momentum (wave number) depends upon a "dispersion law" resulting in cwq « constant at small q for "optical phonons" and coq = cs | q | at small | q | for "acoustical phonons",
Second Quantization: Photons and Phonons
63
where cs is the velocity of sound in a solid, cs ~ 105 cm/sec. Since phonons are induced by displacements of lattice-site ions, they can be associated with a vector "spin", having longitudinal as well as transverse components. The scale of the phonon displacement amplitude A with Fourier coefficients Aq and expansion (4.67) is set by the time-averaged displacement energy of a spring with spring constant k = Mco2, where M is the lattice ion mass, <£> = 2xiMo; q 2 |Aj 2 .
(4.75)
Since this classical energy represents only half the energy of the spring (the kinetic energy accounts for the other half, by the virial theorem), the quantum analog of (4.75) is \ha>q Nq for Nq phonons of angular frequency u>q, giving
l A 'l 2 = 2 * ^ '
<4J6)
Paralleling the photon case (4.68) and (4.69), it would appear that the second-quantized phonon displacement field operator is
with aq and aq satisfying the commutation relations (4.73) and a\ aq is the number operator similar to (4.70). In Chapter 9 we shall employ the oneparticle phonon wave function (noncovariantly normalized) A(x l) =
'
V ipoTve{q)eiq'
Xe_iM,
''
(4J8)
where the ion mass M in (4.77) is expressed as pV in (4.78) so that the amplitude displays the usual one-particle box normalization. The ion density p varies from 2 to 20 g/cm3 in a solid, a typical value being 5 g/cm3. General references for boson wave equations are Hamilton (1959), Roman (1960), Schweber (1961), Bjorken and Drell (1964), Bethe and Jackiw (1968), Schiff (1968), Baym (1969), and Berestetskii et al. (1971).
CHAPTER 5
Spin-i Dirac Equation
What the Schrodinger equation is to nonrelativistic physics, the Dirac equation is to relativistic physics. We begin this chapter by describing three alternative ways of deriving this spin4 wave equation—the more the merrier, in order to develop as much intuition as possible about this fundamental dynamical tool. Next we formulate the Dirac equation in a manifest covariant manner and emphasize the structure of y-matrix algebra and the positive and negative free-particle solutions. The Dirac equation in the presence of external fields is then generated by minimal replacement, and the resulting electron bound-state energies are obtained for the one-particle Coulomb atom and for a constant external magnetic field. We pay particular attention to the difference between the Dirac atom and the fine-structure level shifts in the Schrodinger atom. Finally, we develop free-particle Dirac equations for spin-j massless neutrinos and spin-f massive particles.
5.A Derivations of the Dirac Equation The Dirac equation plays a fundamental role in any relativistic quantum theory, not only because it circumvents many of the problems arising from an unphysical interpretation of the Klein-Gordon equation, but also because it naturally describes the basic spin4 constituents of matter at the atomic and nuclear level—electrons and nucleons (protons and neutrons). To appreciate the significance and beauty of the Dirac equation, it is well to describe three alternative derivations: the original relativistic approach of 64
Derivations of the Dirac Equation
65
Dirac, the ad hoc but elegant derivation via a Pauli spin-matrix replacement, and finally the group-theoretic derivation. Dirac's Derivation. In order to obtain a positive definite probability density for spin4 particles, Dirac searched for a relativistic differential equation which was first order in time (Dirac 1928, 1958). To this end he interpreted the Schrodinger equation, (id, — HQ)^/ = 0, as generated for spin 4 particles by a free-particle matrix hamiltonian H0, first order in momentum for relativistic considerations: H 0 = a p + /?m.
(5.1)
Here a; (i = 1, 2, 3) and /? are four matrices to be determined, and p is the usual quantum-mechanical momentum operator — iV for a particle of mass m. Since the single-component spin-0 Klein-Gordon equation leads to a rwo-component first order matrix (Feshbach-Villars) equation (4.20), it is natural to suppose that a two-component spin4 equation, second order in time, should be linked to a four-component matrix equation, first order in time. Such a "guess" was all the more impressive at the time because Dirac did not have the hindsight of the Feshbach-Villars equation. Thus one assumes that a, and ]8 are 4 x 4 matrices and that the Dirac wave function \// -»\j/„ is a four-component (Dirac) bispinor. The second-order equation in time for a free particle must, of course, be of the Klein-Gordon form, which in momentum space is obtained from the first-order equation via multiplication by E + H0: (E - H0)iP = 0 -> (E2 - H2)^ = 0.
(5.2)
Dirac then demanded that the square of the 4 x 4 matrix hamiltonian (5.1) be constrained to E2 = p 2 + m2 by (5.2), i.e., Hi = i(a,o, + ajOifapj + (/fat, + a,/*)p, + P2m2 = p 2 + m2.
(5.3)
This leads to the defining properties of the matrices a, and /?: {a„a,} = 2S y , {/J,al} = 0, a2 =/? 2 = 1.
(5.4)
Furthermore, since H0 must be an observable hermitian operator, so must a; and /? be hermitian:
a/ = «„
P = P.
(5.5)
The adjoint row bispinor i/^ can be combined with the column bispinor xf/ to form a positive definite probability density P
= ^V= I Ma,
(5-6)
now naturally linked with the hermitian probability current density j = i/^ai/f.
(5.7)
66 Spin-i Dirac Equation
A justification for the form (5.7) is the resulting continuity equation 8,p + V • j = 0, which follows from (5.6), (5.7), and the Dirac equation (id, - H0)i}> = 0,
(5.8)
where H0 obeys (5.1) and (5.3) (see Problem 5.1). From the properties (5.4), it is clear that at = — /?<x,/? and /J = — a.l^a.l, which means that these matrices are traceless (Tr ft = YZ=i PooYTr otj = Tr p = 0,
(5.9)
since Tr BA = Tr AB. It is then useful to describe at and /? by a specific representation, such as the Dirac-Pauli representation, a =
C »)• H i -")•
[which, of course, satisfies the general properties (5.4), (5.5), and (5.9)], where the elements, a and 1, are themselves the 2 x 2 Pauli and identity matrices, obeying oiOj = 8ul + isiJk(Tk. (5.11) Dirac used this representation (5.10) to obtain the first (but by no means the last) profound prediction of the Dirac equation. Making a two-component reduction of (5.8) in the presence of an electromagnetic field via the DiracPauli representation (5.10), he discovered that the electron must have the unique magnetic (dipole) moment ^
= 2X
^neX2=2me°>
<5"12>
i.e., the Lande g-factor is ge = 2, a result in almost perfect agreement with experiment and assumed ad hoc up to that time. We shall return to Dirac's method of predicting ge = 2 in Section 5.D and to the small but important corrections to this result in Chapter 15. Derivation via Pauli-Spin-Matrix Replacement. This leads naturally to the second derivation of the Dirac equation with ge = 2 as input [see, e.g., van der Waerden (1932), Sakurai (1967)]. Recall from (5.11) that the square of any vector such as p 2 is equivalent to the Pauli-spin-matrix replacement (a • p)2. The question then arises as to when this substitution is required, i.e., for all or only a few problems involving spin-^ particles. Furthermore, while (o • p)2 = p 2 , it is clear that in the presence of an electromagnetic field, the minimal replacement p -»• n = p — eA means (
(5.13)
Need nonrelativistic or relativistic momentum operators be subject to this Pauli-spinor replacement ? Consider first a nonrelativistic free-particle hamiltonian with p2/2m -»• (
Derivations of the Dirac Equation
67
field, (5.13) then leads to the Pauli hamiltonian H = ~(ap)2^^n2-~aB. (5.14) 2m 2m 2m The spin-dependent term in (5.14) has the form — u • B, with \i = gnBJ where J = a/2 and g = 2, a very desirable result. Given this success as input, we are encouraged to apply this mnemonic to all problems involving p 2 and spin \. In particular we modify the spin4 Klein-Gordon equation to [E2 - (a • p)2]> = m2<$>,
(5.15a)
which then has the time-dependent factored form (id, — a • p)(idt + a • p)> = m24>.
(5.15b)
Now we parallel the discussion of the Feshbach-Villars version of the Klein-Gordon equation (Section 4.A) by defining
(5.16)
and expressing the second-order equation (5.15) as two coupled first-order equations, (id, + a • p)(pL = mcpR,
(id,-o
-p)(pR = m
(5.17)
Further defining (p =
X =
(5-18)
and using the Dirac-Pauli representation (5.10), the first-order equations (5.17) for these two-component spinors can be put into the four-component Dirac-equation form (id, — H0)\f/ = 0, with ty = (£) and
again the Dirac free-particle hamiltonian. Setting p = 0 in (5.17) means that
#M0)=±M0),
(5-20)
stating that the Dirac-Pauli representation diagonalizes the bispinor energy states in the extreme nonrelativistic limit. Alternatively we may use another representation for a, and /?, called the Weyl representation
68 Spin-^ Dirac Equation also satisfying the defining relations (5.4), (5.5), and (5.9). Given (5.21), the two-component equations (5.17) again can be combined into the fourcomponent Dirac form (5.8) provided that
in this representation (Problem 5.1). In Section 5.E we shall show that for m = 0, (5.17) indicates that
(5.22a)
Ml>) = D(Lp)(pR(0) = [2m(E + m)]-*(£ + m + a • v)
(5.22b)
from which one can derive the coupled relations (5.17) (see Problem 5.1). Now use (5.22) along with the matrix boost obtained from (3.51) in the Weyl representation as 9{L>) = ( D ( ^ }
5(°^})
= [2m(E + m)]-*[(E + m)l + a • p], (5.23)
where 1 denotes the 4 x 4 unit matrix (sometimes deleted). Again expressing the Dirac bispinor in the Weyl representation as \j/ = (££), (5.22) and (5.23) boost i/f(0) from rest to *(p) = ®(I*M0). (5.24) Finally, to recover the Dirac equation, operate on (5.24) with id, — H0, obtaining (id, - H0)iA±(p) oc [±E - (a • p -I- j8m)][±£ + a p + m]il/±(0) = (£2-p2-ro2)>M0) + m(±E + m-<x- p)(l + P)il/±(0). (5.25)
Covariant Formulation 69 The first term on the right-hand side of (5.25) vanishes by the energymomentum relation, and the second term is zero by (5.20). Thus the Dirac equation (id, — H0)i// = 0 is equivalent to the Klein-Gordon constraint plus (5.20), which projects out the positive- or negative-energy parts of \p. Put another way, the Dirac equation is the boosted form of the rest-frame energyeigenvalue relation (5.20) subject to the energy-momentum (mass-shell) constraint E2 — p 2 = m2 for p in an arbitrary direction. In Chapter 6 we shall learn that (5.20) has invariant meaning as the spatial-inversion (parity) eigenvalue relation. Combining two-component spinors which transform into one another under spatial inversion as in (5.23) means that 3) is again equivalent to the parity transform via (3.47), P&(Lp)P = .@(L_P). This is sometimes considered the main motivation for the four-component Dirac formalism. Another is the fermion mass; if it is zero then the two-component equations (5.17) suffice.
5.B Covariant Formulation Covariant y-Matrices. It is possible to formulate this relativistic fourcomponent formalism in manifestly covariant language by expressing the Dirac matrices /} and a in terms of covariant ^-matrices y„ = (y0, y), defined as y0 =
fi,
y = /fa. 2
(5.26) 2
These matrices have the "lengths" (y? = y = y| = y .) y2o=-yf=l
w' = r p 4
(5.27)
and satisfy fundamental anticommutation relations following from (5.4): {y^y,} = y^v + y,y^ = ^9^-
(5-28)
Note that (5.28) has the form of a covariant symmetric tensor equation; this implication will be discussed shortly. Next combine the hermitian and antihermitian relations yj = y0 and y/ = — y, into a "Dirac adjoint" matrix operation iv = 7o7j7o = y„,
(5.29)
where in general the Dirac adjoint (barred operation) of any 4 x 4 matrix is A = y0 A*y0 with AB = BA. That is, (5.29) implies that y„ is "self-barred". Then define the new y-matrices 7s = 70 7172 73
(5.30a)
satisfying 75 = 7s,
7s = - 1 ,
7s 7„ = -7„7s,
(5.30b) (5.31a)
70
Spin4 Dirac Equation
satisfying a^ = a^, since i = —i and Gij — £;jfc
°i = kijk°jk>
ffoi = iyo y.- = '«,- = - 7s
(5.31b) ffi.
(5-3 lc)
where a{ are now 4 x 4 matrices, built up from the 2 x 2 Pauli matrices as (" „). In the latter case, a = iy5o follows from a t = iyoTi^ysiyz73 — yo7i, etc. Since 4 x 4 matrices have 16 possible elements, there must be 16 independent y-matrices. We may choose these independent matrices as the selfbarred set r , = 1, y„ < v , iy„y5, y5 = S(l), v(4), T(6), ,4(4), P(l),
(5.32)
where 1 is the one scalar (S) unit matrix, yM are the four vector (V) matrices,
(5.33a)
( 5 - 33b )
y5^ = i w ^ y„ yv yP = 0M. yP - gw y, + g,P y» - « w V"TS >
(5.33c)
w y V Y = -3!y,y 5 ,
(5.33d)
£,vpffy/'yYy''=-4!y5, y^/vyp/ff
CT/iv/p/ff
(5.33e) y^pivia
• yvpinto
• yparptv
yvo/n/p
+ y5 W "
(5-33f)
etc., where the y-matrix products on the left-hand sides have been expressed as linear combinations of the 16 Tt on the right-hand sides of (5.33), and we have used the fundamental properties of the Levi-Civita pseudotensor ( e 0123
=
1)> pnvpa — _ 4 |
(5.34a)
F
f '">" =
(5.34b)
P
P
F
etc.
— 3 I Jim'
P" — — "> 1
9m
9v«
9»t>
9vp
(5.34c)
Covariant Formulation 71 As regards the matrices /? and a, the y-matrices satisfying only (5.27) and (5.28) then have many possible representations. The representationindependence (Pauli-Good) theorem states that all representations of ymatrices are equivalent up to a similarity transformation, y'tl = S~lyflS. The Dirac-Pauli representation, diagonalizing energy via y0 in the extreme nonrelativistic limit, is 1 0\ . 10 1) 7o= 0 -1 ' ^ 5 = 11 0 (5.35) 0
"
\0
whereas the Weyl representation, diagonalizing helicity via y5 in the extreme relativistic limit, is 0 1\ / - l 0^ y =
°
'i o '
H-:;>
iy5
'~\
o ir
(5.36)
-[i:\
Other representations, including the Majorana and the light-plane representation, are worked out in Problem 5.3. Covariant Transformation Laws. To put the Dirac equation (id, — H0)i// = 0 on a covariant footing so as not to single out time via d„ multiply the equation on the left by y0—obtaining (iy0 d0 — y • p — m)i// = 0—and use p = — iV. This results in the covariant form of the Dirac equation, (iy • d - m)\jj(x) = 0.
(5.37)
Henceforth it will be convenient to define a "slash" operation 4 = y 'A so that (5.37) takes on the compact form (i) — m)i// = 0. To verify that (5.37) is indeed covariant under homogeneous Lorentz transformations x' = Ax, we rely upon the discussion in Chapters 2 and 3 to obtain a 4 x 4 spinor matrix ^(A) which transforms the bispinor wave functions in a fashion analogous to (2.1c), IA'(X') = ^(A)^(x). (5.38) Since the Hilbert-space operator is (7A = exp( —jo»''vJ/lv/2), the obvious identification J/iV -» <J/1V /2 in the spin4 Dirac space means we can write y(A) = exp(-ia/V„ v /4).
(5.39)
An infinitesimal transformation A„v -> g^ + OJ^ induces (5.39) to become y(A)-» 1 — ict/v<7„v /4, and the y-matrix identity (see Problem 5.2) [ff„v, yP] = 2i(gpvyli-gPflyv)
(5.40)
then leads to the transformation law for y-matrices,
(5-41)
72
Spin-j Dirac Equation
[Note that (5.41) bears the same relation to the Lorentz group as (2.15) and (2.52) do to the rotation group.] It is in the sense of (5.41) that yll transforms like a four-vector; since y^ is a constant matrix, yj, has no meaning other than (5.41). Then we establish the covariance of (5.37) by multiplying it on the left with y(A): 0 = y(A)(iy • d - m)iA(x) = [ ^ ( A ^ y ' H A K - mMA)iA(x)
(5-42)
= (iA-Vd» - mty'(x') = (iy • d' - mW{x% where we have used d'v = (dx^/dx'^d* = A'/S". Thus, (5.42) implies that the Dirac equation (5.37) is valid in any frame x' = Ax, an explicit demonstration of the meaning of covariance. Now we extend the Dirac adjoint or "bar" operation to bispinor column vectors. Define the row bispinor $ =
(5-43)
If if/ satisfies (5.37), then $/ satisfies the adjoint Dirac equation (# - m)«A = ifr{ -1? - m) = 0,
(5.44)
_1
since ytt = y^. Then using £?(A) = 5^ (A), which follows from (5.39) and d^ = a^, we have from (5.38) ^'(x') = $(x)?(A) = $(x)&-
X
(A).
(5.45)
Combining (5.38), (5.41), and (5.45), we can obtain the transformation laws for the bilinear covariants ^ r , \p (now "c-numbers" in the Dirac space, since a row times a column matrix is a pure number—see Problem 5.4), S: ^'(x')iA'(x') = #(*M*)
(5.46a)
V: ^'(x')y^'(x') = A/.A(x)y>(x)
(5.46b)
T: ^(x'K v ^'(x') = A/A v ^(x)a a/! V(x)
(5.46c)
A-
(5.46d)
^'(x')iy,y 5 *'(*') = (det A)A„>(x)iyvy5iftx)
P: ^(x')y5
(5.46e)
It is in this sense that the r ; transform like integral representations of if, indicating, for example, that the Dirac probability current (5.6) and (5.7), expressed in covariant form as ;„ = ^ ( 1 , atyr = ^(y 0 , y)* =
fa^,
(5.47)
transforms as a four-vector, j'^x') = A/;'v(x), with a continuity equation d • j = 0 which is invariant from frame to frame. Dirac Trace Algebra. The bilinear covariants just discussed can be used not only to construct the probability current;^, but also to form the S-matrix or transition-probability matrix elements for a process involving a spin-j parti-
Covariant Formulation
73
cle. In analogy with (3.87) we write <^'|S|A> = ^W'>M^W),
(5.48)
where i//A) and \j/(X"> are column and row bispinors which are eigenstates of helicity, and M is a scalar 4 x 4 matrix composed of invariant products of momenta and y-matrices. The complex conjugate (*) of (5.48) can be expressed in terms of the Dirac adjoint matrix, M = y0 M^y0, according to * = (^'>M^ W ) ) t = ^ttt>Mtj£ttt'> = iAtU)yo7oMty0'u') = VX)M\l/(X'\ The physical quantity of interest is the transition probability, which corresponds to the product of (5.48) and (5.49); summing over all possible helicity states (unpolarized spin sum), we write
X |a'|S|A>| 2 = X^u')M^A?iAu'), k',k
(5.50)
k'
where SP is a Dirac "projection" operator gp = £ ^wyw (5.51) x obeying SP1 = SP for orthonormal bispinors ij/u 'tyU) = 5XX. Rearranging the bispinors in (5.50) to the form (M^,M)(7-
|| 2 = T r M ^ M ^ ' .
(5.52)
Now we shall show shortly that the projection operators & and 3P' can be expressed in terms of y-matrices, so (5.52) represents in principle a trace over the product of many y-matrices. Consequently it will prove useful to describe the trace algebra of y-matrices. To begin with, in the Dirac space Tr 1 = 4.
(5.53a)
Inspection of the Dirac-Pauli or Weyl representation [(5.35) and (5.36)] or application of the defining relations (5.28) and (5.4) along with the property Tr AB = Tr BA gives Tr y„ = Tr y5 = 0, (5.53b) Tr yMyv = i Tr{y„, yv} = 0(IV Tr 1 = 4 ^ v ,
(5.53c)
and from (5.30b) and (5.33b), Tr y5y„ = 0,
Tr y5y„yv = -\iz^
Tr &" = 0.
(5.53d)
Next, note that (5.33c) then implies Tr y^ yv yp = 0, which can be extended to a product of any odd number of y-matrices y ^ (excluding y5, which is an even product y0 yx y2 y3) via multiplication of (5.33c) by successive pairs of y-matrices, implying that Tr yM = 0. (5.53e)
74 Spin-j Dirac Equation
For even numbers of y-matrices, (5.53a) and (5.53c) are nonvanishing; for four y's, use of (5.33c) or (5.33f) leads to Tr y»yvypy„ = Mg^Qpa - g^g*. + g^g,P\ Tr y5 y„ yv yp ya = - 4£„VP(T,
(5.53f) (5.53g)
with £0123 = 1- The l a s t identity may be verified by choosing /i = 0, v = 1, p = 2, a = 3, and using (5.30a). Multiplication of (5.33f) by y„ yx leads to the trace of six y-matrices, but at this point and thereafter it is easier to apply repeatedly the anticommutation relation (5.28) to permute y„ from one side of an even number of y-matrices to the other and then use Tr Ay^ = Tr y„ A to find in general (see Problem 5.5) Tr y„yvypyB • • • = g„v Tr ypyff • • • - g^ Tr yvyn • • • + 0^Tryvyp
etc.
(5.53h)
It is clear from the algebra (5.53) that including two additional y-matrices in a trace can lead to a considerable complication. Quite often, however, the indices of two of the y's are summed (contracted), in which case repeated applications of (5.28) results in (see Problem 5.5) ^y„f =-2y„, y*y„7vf = 40„v, y. )V ?v yPf=-
2
)V 7v y„,
y*y*y*ypy.f = 2(yffy„yvyP + yPyvy„y,x),
(5.54a) (5.54b) (5.54c) (5.54d)
etc. The identities (5.54) should be applied before computing y-traces according to (5.53). The y-matrix traces can be used to verify the completeness of the 16 Dirac covariants r, of (5.32), with Tr r , r , = 4 ^ ,
(5.55)
where */, = + 1 for T, = 1, y0, iy,y5, a, and r\{ = - 1 for T, = y5, yh iy0y5, o0i. Then any 4 x 4 matrix can be expanded as M = £ a, T; with the coefficients from (5.55) found to be 4a; = JJ, Tr Mr,. Also the completeness relation (5.55) is needed to develop Fierz "reshuffling" matrices (see Problem 5.6).
5.C Free-Particle Solutions of the Dirac Equation Positive- and Negative-Energy Spinors. In the spirit of the free-particle spin-0 analysis, we now examine the positive- and negative-energy solutions of the covariant free-particle Dirac equation ( # - m t y ± ( x ) = 0.
(5.56)
Free-Particle Solutions of the Dirac Equation
75
For the positive-energy solution corresponding to a particle of momentum p (with V = 1), we write il/+(x) = u{p)eivxe-iEt,
(5.57a)
while for the negative-energy solution with energy — E (E = ^/p2 + m2 > 0) associated with momentum — p, ^Mx) = w(p)e-' p •V £, .
(5.57b) 2
In both cases we require the spin-^ particle of mass m to obey E — p 2 = p2 = m2. Substituting (5.57) into (5.56) then leads to the momentum-space Dirac equations {p - m)u(p) = 0,
it(p)(p - m) = 0,
(p + m)v(p) = 0,
v(p)(p + m) = 0.
(5.58a) (5.58b)
It is convenient to normalize the positive-energy spinors u(p) and the negative-energy spinors v(p), choosing the signs according to (5.20), so that u(p)w(p) = 2m,
v(f)v (p) = — 2m,
(559)
while i//± are normalized covariantly in analogy with the spin-0 KleinGordon solutions (4.7). To obtain the free-particle solutions in the rest frame, we evaluate (5.58) in the limit p0 = m, p = 0, obtaining yo«(0) = u(0),
yov(0)=-v(0).
(5.60)
[Recall that the notation u(0) emphasizes the p = 0 aspect of the rest-frame wave functions; rest-frame two-component spinors will be written as >(p) as a reference for rotations.] In the Dirac-Pauli representation with y0 diagonal, (5.60) diagonalizes the energy (mass, in the rest frame) with the spinors expressed as
u(0) = y/2^fe\
r(0) = V2^(°J,
(5.61)
where q> and i are two-component spinors, both normalized to unity in order that (5.59) may remain valid. In particular, we further specify these two component spinors as helicity eigenstates
VP?W(P) = V1P)
(5-62)
and represented by the rotated spinors (3.91). The reason for the choice of phase e T '* for k = ±\ in /(p) will be explained in Chapter 6. We may also recognize (5.60) as equivalent to the boost constraint (5.20) in the Lorentz-group derivation of the Dirac equation. In this case, the
76 Spin-^ Dirac Equation
covariant version of the Dirac boost matrix (5.23) is, since E + a • p = 1>y0, ®(L,) = [2m(E + m)]-*(fiy0 + m) , x
(5-63)
so that @(L„) = y(Lp)- Application of this boost to (5.61) and using (5.60) with X = ±j then leads to
^)-»W^)--^i( , , 7 W ).
(5.64a)
^-*^>^>-7S^UA-J
(564b)
From the form of (5.64) it is clear that these bispinors satisfy the free-particle Dirac equations (5.58), since /*/> = \{p, p} = p2 = m2 implies that [p - m)wU)(p) oc (p - m){p + m) = p2-m2 w
2
= 0, 2
{p + m)v {p) az(p + m)(p - m) = p - m = 0. Further, the bispinors (5.64) are indeed helicity eigenstates, obeying by virtue of (5.62) and [a • p, p] = 0 & • p«(A)(p) = Xu^(j>),
(5.65a)
• pVA)(p) = *»w(p)
-\e
(5.65b)
for X = ±\, where a in (5.65) is now the four-component Dirac spin matrix. Thus, the advantage of the explicit boost construction in the Dirac-Pauli representation (5.64) is that it manifests both the Dirac equation and helicity constraints satisfied by positive- and negative-energy bispinors. Moreover, the connection between u(p) and r(p) for momentum states — p in the latter case is made readily apparent by (5.64). There is another use for the boost operator, that of "de-boosting" u(p) and v(p) back to the rest frame, sometimes referred to as the "nonrelativistic reduction" procedure. Recalling that — Yy0 =fys<*> w e express the boost (5.63) and its Dirac adjoint in terms of the even (diagonal) operator 1 and the odd (off-diagonal) operator y5 a • p in the Dirac-Pauli representation, converting the bispinors to the form
^(p)=^ ,y/E"::+/sm: p (
( 5 - 66a )
E + m + iy5 a
«(>) = ( /
(
1 p ) , 0 ) £ ^ t o yJE + m E + m + iy5a • p yjE + m
)
^-""^•"UA-J *«« = (0, e± V u , (-P)) £
+ m
/F|y5
(5.66b)
(Mfc) (5.66d)
Free-Particle Solutions of the Dirac Equation
77
The nonrelativistic reduction of bilinear covariants such as ur ; u can be easily obtained using (5.66) to identify the net even operators (in the DiracPauli representation) of ® r ; ® . For «r,H and vT^, only net even operators (two-component diagonal) contribute, while for ur ; v and vrt u only net odd operators (two-component off-diagonal) do (see Problem 5.7). Free-Particle Projection Operators. Recalling that the two-component spinors satisfy the completeness relation
X>WWW)(*)=1,
(5.67)
x we expect the four-component bispinors to obey some sort of analogous relation. Given (5.64) and (5.67), we may compute
(E + m)i «w(p)iiw>(p) = {i> + m) I1 °) (P + m) \U U / = i(/> + m){\ + y0)(p + m).
*
(5.68)
With the help of the anticommutation relations (5.28) and p2 = m2, we see that py0 i> + m2y0 = 2Ep, (y0 /> + py0) = IE, and (p + mf = 2m(p + m). Then the factor of (E + m) cancels from both sides of (5.68) and we obtain 2
X wa)(p)"U)(p) = /* + ™,
(5.69a)
x and similarly 2
X vU){p)vw{p) = t-m. (5.69b) x These useful relations can be verified by applying the Dirac operator (/> — m) to (5.69a) and (p + m) to (5.69b); the normalizations are easily confirmed in the rest frame via (5.61). The identities (5.69) define respectively positive and negative projection operators in the sense of (5.51). That is,
M r i - ^
("0)
are normalized energy projection operators obeying A+ = A+ and A + A_ = 0 for p2 = m2. The additional identity A + + A_ = 1, or equivalently from (5.69), X («U)(p)tiU)(p) - «W,(P)»W,(P)) = 2w (5.71) x is the Dirac analog of the two-component completeness relation (5.67). The usefulness of the energy projection operators (5.69) is that, for an S-matrix specified by positive- or negative-energy free-particle bilinear co-
78
Spin-^ Dirac Equation
variants, the general unpolarized-spin sum (5.52) becomes in particular X |u U) (p')Mu a) (p)| 2 = Tr M(p + m)M(p' + m),
(5.71a)
A',A
£
|DU)(p')Mi;U)(p)|2 = Tr M(p - m)M(p' - m),
(5.71b)
x'.x
and similarly for uMv and vMu. It is also possible to construct a covariant spin projection operator. In the rest frame of the spin4 particle, such a spin projection operator is j(l + a • s), where s is the direction of spin polarization (s -» p corresponds to a helicity projection operator). Defining sM = (0, s) and m^ = (m, 0) in the rest frame, so that m • s = 0, it is clear that in a boosted frame p • s = 0. Then, replacing a by — iysyy0 and dropping the noncovariant factor y0 (since py0 does not change sign when applied to either u or i;), we obtain the covariant spin projection operator E(s) = i(l + iy 5 /). 2
(5.72)
2
The invariant length s = — 1 means that £ (s) = Z(s), characteristic of any projection operator. We may then compute positive- or negative-energy polarized spin sums using the Dirac trace algebra by simply replacing, say, p ± m in (5.71) with Z(s)(p ± m) (or (p ± m)L(s), since [y5 £ p] = 0) for a particle whose spin is not unobserved, but pointing in the "direction" s. Specific helicity states can then be generated using s^ = + (p, Ep)/m in (5.72), whereas spin representations other than helicity follow from a different choice for sM. Finally, a third set of Dirac projection operators P±=i(l±iy5)
(5.73)
will prove useful for our later work. They too obey P\ = P±, P+ P_ = 0, and P+ + P_ = 1. One immediate application of (5.73) stems from P + 7n = y^P-> which means that for free particle bispinors \jj = m'1^^ we have P± ip = m~ iii/)P+ i//, so that ij, = P+il/ + P_il/ = m~l(i$ + m)P+il/.
(5.74)
Thus for (j) = P+ i//, ( $ - w)i//= - m _ 1 ( n + m2)
(5.75)
which says that, given > obeying the less restrictive Klein-Gordon equation (for bispinor wave functions), one may obtain \p via (5.74) which satisfies the more restrictive Dirac equation provided (j> = P+ if/. This is reminiscent of our second derivation of the Dirac equation built up from two-component spinors related by the differential operator in (5.16), here equivalent to (5.74) in the Weyl representation. In the above approach, the Dirac-Pauli representation reduces (5.75) to two identical two-component spinor equations. This is true even in the presence of electromagnetic fields, and we shall take advantage of this fact shortly.
Free-Particle Solutions of the Dirac Equation 79 Free-Particle Probability Current. We have noted in (5.6), (5.7), and (5.47) that the bilinear covariant i?y„i/' is the natural choice to represent a conserved probability current density. Consider the current density constructed from the free-particle positive-energy wave packet similar in structure to the Klein-Gordon form (4.13a), normalized in a box: «A + W = J 2 ^ 4 « P « ( P K ' P J C
(5-76)
(where we have deleted the helicity summation for clarity), giving j ; ( x ) = * + (x)y> + (x) = U l l \ w
< ° P < P U + |P>e*"' (5.77)
with q = p' — p. The off-diagonal momentum-space current is then
(5.78)
In this language, current conservation d • j(x) = 0 is equivalent to
m)uu = 0, (5.79)
by the free-particle Dirac equations (5.58a). An analogous negative-energy packet with E = >/p 2 + m2 > 0, +-(*) = j^b*v(VW*,
(5.80)
leads to a
(5.81)
Sandwiching this identity (5.81) between free-particle positive-energy bispinors, we note that p' operating to the left and f> to the right both become m in (5.81), again by the free-particle Dirac equations (5.58a). Rearranging this result, we obtain a form of the Gordon reduction, U(P% u(p) = «(P') (~+ifff)
«(P)>
(5-82)
where the P^ and a^ qv terms in (5.82) are called the convection and spin currents, respectively. Clearly, relations similar to (5.82) can be generated by sandwiching (5.81) between v and v, or u and v, or v and u.
80 Spin4 Dirac Equation As a by-product of (5.82), the diagonal matrix element can be obtained with q = 0 and P = p:
"(PK «(p) = (P„H"(PMP) = 2p„.
(5.83a)
This result is "obvious" from covariance considerations alone, because the left-hand side of (5.83a) must be a simple four-vector and the only candidate depending upon p is p^, with a normalization factor following from the contraction with p^, using p2 = m2 and the Dirac equations (5.58). In a similar manner we deduce that «(P))V«>(P) = (-Pjrn)v(p)v(p) = 2p„.
(5.83b)
We may now calculate the expectation value of the Dirac velocity operator a. Noting that the spatial integral of (5.77) generates # 3 (p' — p), which in turn picks out the diagonal matrix elements of j * , we use (5.83a) to write [a similar relation holds for the Klein-Gordon current (4.62)] \d>xj;(x,t) = \^\ap\2pJE,
(5.84)
with an analogous relation for j ~ , where ap is replaced by bp. For /j = 0we see that p0/E = 1 in (5.84) as well as for the jo integral. Since these jo integrals are j d3x \j/\{x)^i±(x), the latter integrals normalized to unity as the total probability, we learn that V) 2£ K
l
~2E | b p l ~L
'V)
(5 85)
-
Next we evaluate d3x j ± (x, t) = j d3x \j/i(x)aal/±(x) = ± ,
(5.86)
and from (5.84) conclude that <«>± = ^
+
=
(5-87)
where the latter two averages in (5.87) are weighted over the momentumspace probabilities (5.85). The important result (5.87) simply states that the positive- or negative-energy wave-packet expectation value of the current is just the relativistic group velocity vg, a conclusion paralleling the nonrelativistic situation (1.6), and expected in this case because in the Heisenberg picture, ^
= i[H, rj] = iak[pk, rj] = a,,
(5.88)
Zitterbewegung. It turns out, however, that both positive- and negativeenergy states must be included in the construction of the eigenfunctions of a.
Dirac Equation in an External Field 81
Consider then the general packet (again suppressing the helicity summation)
* ( x ) = I 2^K* k»(PK"" x + b*vW<\,
(5-89)
and computing as before leads to <«> = ( | ) + j ^
[a*b%u(P>tv(-v)e2iE'
+ apb_pD(-pMP)^2'£']-
(5.90) As was the case with the Klein-Gordon equation, the cross terms between the positive- and negative-energy states oscillate violently in time. Unlike the Klein-Gordon situation, however, the free-particle amplitudes multiplying the oscillating phases in (5.90) need not vanish as p -> 0. To see this, use a = iy5 a and the forms (5.66) to obtain the nonrelativistic reductions (Problem 5.7) i i ( p M - p ) -* 2m
5.D Dirac Equation in an External Field Covariant Electromagnetic Interactions. Following the procedure used in the Klein-Gordon analysis of a particle in the presence of an electromagnetic field, we consider for spin-^ particles the minimal replacement of id^ by id^ — eA^ in the Dirac equation (5.37), giving (# -eA-
m)i//(x) = 0, or (# - m)tfr(x) = eA\p(x).
(5.92)
We will assume that this minimal coupling eA represents the fundamental interaction between the spin4 particles and photons. As in Dirac's original approach, the bispinor equation (5.92) obeys a Klein-Gordon constraint, obtained by multiplying (5.92) on the left by the operator i$ — eA + m: [(id - eA)2 -m2-
e\a^vF^(x)
= 0,
(5.93)
where we have used the anticommutation relations of the y-matrices to write Ui> + m
= i[y„, y,]F"V + d • (Aij,).
We see from (5.93) that the Dirac wave function also obeys the KleinGordon equation in the presence of an external field (4.24a) apart from the spin-dependent factor i^vF^ = ia-E-
82 Spin4 Dirac Equation which couples E and B respectively to the electric and magnetic dipole moments of the spin4 particle. In particular, an energy eigenstate of ^(x) in (5.93) obeys, for E2 — m2 = 2mENR + • • •, (5.95)
£ N R " A = - ^ - B < A + ---,
showing once again that \i = ea/2m and g = 2 are natural consequences of the Dirac equation. It is also interesting to investigate the Dirac probability current, j^x), in the presence of an external field. In spite of the absence of a derivative term in ij/y i/f, one may consider the Gordon decomposition in coordinate space analogous to (5.82) and then make a minimal replacement to find (see Problem 5.8) h(x) = ^
^(x)a>(x) - ~ MxWWAAx)
+~
fl1f(XKi(4 (5.96)
It is clear that the first two terms in (5.96) have the same form as the nonrelativistic and Klein-Gordon current density (4.26), covariantly normalized in the latter case. The first and third terms in (5.96) correspond to the Gordon decomposition for A^ = 0. As in the spinless case (4.27), current conservation d • j(x) = 0 is an obvious consequence of (5.96). Dirac Form Factors. Following now the discussion in Section 4.C, the probability current density j„(x), whether due to nonrelativistic, Klein-Gordon, or Dirac particles, plays a dual role because ejj^x) represents a chargedmatter current density which is a source for the electromagnetic field, \3A^ = ejp (in the Lorentz gauge). Thus the spin-j analog of the momentumspace current (4.63b) in the presence of an electromagnetic field for positiveenergy scattered particles, obtained from (5.96), is e<9'\tf \p> = eufa')
2m
«(P).
(5-97)
This is the most general form consistent with Lorentz covariance and the Dirac equation. The four-momentum transferred from the photon to the spin-j particles is q = p' — p, and the dimensionless invariant form factors Flt2(q2) parallel F(q2) in (4.63b) for spinless particles. For free photons, q2 = 0 and we must have F t (0) = 1 in order that the coefficient of uy^u in (5.97) may be simply the charge e, consistent with the free-particle probability current density (5.78). Thus F^q2) is called the charge form factor. For q2 ± 0, the photon is not free, and the second term of (5.96) contributes to both Fx(q2) and F2(q2) of (5.97) in a complicated dynamical manner. We shall return to the q2 dependence of these form factors in later chapters. Concerning the magnetic form factor F2(q2) in (5.97), since the y^ current already consists of a convection and magnetic moment part as given by
Dirac Equation in an External Field 83 (5.82), with g = 2 in the latter case, the additional magnetic-moment term in (5.97) described by F2(q2) represents an anomalous correction to g = 2. That is, at q2 = 0, F2(0) = ic = i(flr - 2)
(5.98)
is the dimension less, anomalous magnetic moment of a spin4 particle. For the electron (e < 0),
or Ke ~ 0.001, due totally to electromagnetic interactions. For the proton (e > 0), ^ =^ ( 1 + ^(2.79)^,
(5.99b)
or KP ~ 1.79, primarily due to strong interactions—as is the anomalous magnetic moment of the chargeless neutron, /i„ = ^ - ( 0 + « „ ) * ( - 1 . 9 1 ) - ^ - , (5.99c) 2m„ 2m„ or K„ ~ — 1.91. In Chapter 15 we shall show that the minimal (QED) coupling in (5.92) ultimately generates Ke = cc/2n + 0(cc2), in perfect agreement with experiment. It will also turn out that we can estimate KP and K„ better than we have a right to expect. Lastly, the form factors Fl2{q2) a r e not a unique description of the dynamics of spin-| particles interacting with an external electromagnetic field. Two different but equivalent descriptions in terms of the Sachs form factors GE(q2) and GM(q2) are worked out in Problem 5.8. Constants of the Motion. For a free-particle hamiltonian a • p + fim, the fundamental commutation relations (5.4) lead to [H, r x p] = — ia x p,
[H, oj = 2ia x p,
(5.100)
and it is therefore clear that the total angular momentum J = r x p + \a and helicity ^a • p are conserved operators. In a central, spherically symmetric field V = V(r) = eA0 with A = 0, the Dirac hamiltonian becomes H = a - p + j?m+ V(r),
(5.101)
and while commutation relations similar to (5.100) become more cumbersome to apply (see, e.g., Sakurai 1967), group-theory arguments alone tell us that the total-angular-momentum operator must remain conserved. To simplify the search for constants of the motion and their eigenvalues for spin-^ particles, it will prove convenient to reduce the four-component Dirac analysis back to two-component form. We consider then the less restrictive second-order equation (5.93) satisfied by a Dirac bispinor i/r for
84
Spin-j Dirac Equation
positive-energy stationary states with n = p — eA and eE = — V V, B = 0, (t + m){i - m)il/ = [(£ - V)2 + V2 - m2 + id • VFty = 0. (5.102) Multiplying (5.102) on the left by the projection operators P± = i(l ± iy5) of (5.73), we see that the four-component wave function <> / = P+ ip satisfies (5.102), and the more restrictive Dirac wave function, obeying H\p = E\\i, is then recovered from (p by i/> = m~ x{t + m)4> = m~ l[y0(E - V) - y • p + m]>,
(5.103)
a situation similar to the free particle case (5.74). Next we work in the Dirac-Pauli representation with P+ = ^(} \), giving
(5.104)
Then the four-component equation (5.102) reduces to two identical twocomponent equations [for V = V(r)], (£ _
V{r))2
_
2
m2 +
1 3 2r - -j+ia L 2 . • ^dV f
r or
r
dr
(5.105)
equivalent to the Klein-Gordon form (4.29) except for the last spindependent term. If all we are interested in is the quantized energy levels, then we need only deal with (5.105), converting it to the form of the Schrodinger equation as in the Klein-Gordon case. We need not bother with recovering 4i from
(5.106)
This suggests that we define the two-component spin operator - A = (l + « - L ) + (Zo)w f,
(5.107)
A 2 = (1 + a • L) 2 - (Za)2,
(5.108)
which has the square
the cross terms in (5.108) dropping out because of the identity (see Problem 5.9) {(1 + a • L), o f} = 0.
(5.109)
Combining (5.106)-(5.108), we see that the coefficient of — r~2 in (5.105) has the simple form of an angular-momentum operator L 2 - (Za)ia • f - (Za)2 = A2 + A = A(A + 1).
(5.110)
Dirac Equation in an External Field 85 Next we obtain the eigenvalues of A2 and A by squaring J = L + ja, to write for states j = I ± \ =j± (<,-L)j± = (J2-V-i)j±
= {_{ll+{)\
(5.111a)
(l + c r - L ^ j ' + ^ t O + i), where the upper (lower) cases in (5.111) correspond toj=l Then we have
(5.111b) + \(j = l — j).
(A2)J± = () + i ) 2 - ( Z a ) 2 , (A)J± = + [ 0 + i ) 2 - (Za)2]* - +A,
(5.112a) (5.112b)
where the signs in (5.112b) are determined by (5.107) and (5.111b) with - A intrinsically positive for ;' = / + \ and Za -> 0. Following the Klein-Gordon analysis, we may now express the radial equation for the Dirac atom in the form of (4.30a), 1 d2 r dr
/'(/' + 1) r
2EZa q>(r) = -(E2 - m2)cp{r), r
(5.113)
with (5.110) and (5.112) giving, for) = / ± i l'±(l'± + 1) = [A(A + l)]J± = A(A + 1).
(5.114)
Solving (5.114) for the effective orbital-angular-momentum eigenvalue, we see that
r*-f-l\
(5.U5)
As before, we have chosen the nonnegative solutions of /' in (5.114) as Za -»0, for otherwise the radial solution r' will not be regular at the origin. The orbital constants of the motion now being understood for the Dirac atom, we proceed as in the Klein-Gordon atom to find the energy eigenvalues. Since the radial equations are identical in the two cases except for different values of /', we may conclude that the energy levels of the Dirac atom also are of the form (4.32b), (Za)2 E=m 1 + 2 n'
(5.116)
with ri — I' = n— I constrained to the integer values 1, 2, ... as before. Substituting /' given by (5.114) then leads to n- = n - l + r ± = n - 0 - + i) + [(/ + W ~ (^ff (5.117) for both j = I ±\. [Note that this two-component analysis eliminates the need for an auxiliary quantum number k, appearing in the four-component analysis. See e.g. Sakurai (1967).] We see that the energy levels of a Dirac
86
Spin-5 Dirac Equation
atom are identical in form to the Klein-Gordon levels, but with / in (4.34b) replaced by) in (5.117). Note too that for s-waves andy = \, the square root in (5.117) becomes imaginary unless Z < 1/a. In the strong-field limit, Z > 1/a ~ 137 and the breakdown of the boundstate Dirac solutions is similar to the Klein-Gordon case with Z > l/2a. Once again, the breakdown has a classical analog for (Za)2 > L2 with multiparticle quantum states modifying the single particle orbit at short distances. These effects are not fully understood, nor have they been detected experimentally. For Za <^ 1, one can expand (5.116) and (5.117) in powers of (Za) to find the fine-structure corrections to the Bohr formula for a single-electron atom ( n = l , 2 , ...), m(Za)2 r | (Za)22 / 1 (5.118) ^ N R — *• — j 7 m
n V+ \
now degenerate for a given value of j . We shall return to a detailed study of the hydrogen energy levels shortly. Finally, an analysis of the Dirac wave functions (proportional to associated Laguerre functions), based upon this two-component approach, depends upon (5.103). [See Auvil and Brown (1977).] Alternatively, the entire (and more complicated) four-component analysis of Hip = Exjj can be carried out in a straightforward manner [see e.g. Bethe and Salpeter (1957)]. For our purposes, the above analysis of the energy levels alone will suffice for our later work. Before leaving this topic, it is worth mentioning that the two-component form of (5.93) also can be used to obtain the constants of the motion for spin4 problems other than the one-electron Coulomb atom. Consider, for example, an electron of momentum p moving through a region of constant magnetic field, B = Be3 and A = \E x r = \B( — ye1 + xe2) with A0 = 0. Applying the projection operator (5.104) in the Dirac-Pauli representation, the four-component equation (5.93) once more decouples into two identical two component equations: [E2 - (p - eA)2 - m2 + eBo3]q> = 0.
(5.119)
Since V • A = 0 and A • p = ^B • L, we may write (5.119) in the form [p2 + p2 + (eB/2)2(x2 + y2)]q> = [E2 - p2 - m2 + eB(L3 + a3)]
(5.121)
Dirac Equation in an External Field 87 corresponding to an angular frequency of a) = y/k/^n=
\eB\/2m.
(5.122)
Since the eigenvalues of such an harmonic oscillator are known to be of the form Eh.a. = (nx + ^,)(o + (ny + \)a> = nw,
(5.123)
where n = nx + ny + 1 = 1, 2, 3, ..., we may identify the right-hand side of (5.120) as 2mEho. This leads to the constant of motion E2 = pj + m2 + n\eB\ - (*3 ± \)eB,
(5.124)
with E2 > pi + m2 implying n > | /3 | + 1. Thus we see that the energy and angular-momentum eigenvalues of a spin4 particle in a magnetic field can be found from (5.93) in a manner similar to those for the Klein-Gordon or Dirac atom. The two-component and four-component bispinor wave functions for this problem are proportional to Hermite polynomials, as might be expected from the structure of (5.120). (See Problem 5.10.) Nonrelativistic Reduction. While the exact energy eigenvalues for a relativistic spin4 particle in an external field can always be obtained as in the last section, it is nonetheless of interest to expand the Dirac equation (5.92) in powers of ENK/m, where £NR = E — m. Starting with the Dirac hamiltonian (V = eA0, n = p - eA) H = an + pm+V,
(5.125)
we may express Hi]/ = Exjt in the Dirac- Pauli representation as E
- m - V - " ' * ) ( * ) = <>. - a n E+ m-V)\x) This results in two coupled two-component spinor equations (E — m — V)q> = a • nx,
(5.126)
(E + m — V)% = a • nq>. l
(5.127)
l
For weak potentials V
(5.128a)
Assuming (£NR — V)/2m <^ 1, we expand 1 +
^NLHZr1*i_fcL^. 2m } 2m
(5.129) '
K
88 Spin~j Dirac Equation
The leading term in (5.129) then converts (5.128b) to
where we have used (5.13) to obtain the Pauli magnetic-moment term with g = 2, a result now expected. To next order in the expansion (5.129), we set A = 0 to isolate the correction terms due to a weak electrostatic potential V:
Hi
™ = hr+V
4m2H' = -
+H
(5131a)
''
P(£NR - V)a p
= - (£NR - V)p2 + io • (pK) x p + (pK) • p, (5.131b) 2
where a^ = <50- + ieijk<jk and pt Vpt = Vp + (p; V)p{ have been used to obtain (5.131b). The first term in (5.131b) represents a relativistic momentum correction, since to leading order £NR — V ~ p2/2m means that H^-iEn-V)^*-^.
(5.132)
This potential then corresponds to the usual kinetic-energy correction
The second term in (5.131b) can be written for a spherically symmetric potential V = V(r) as w " (fV) >< P = ( ^ ) ° • (* >< P) = ( l ^ y
• L-
(5134)
Consequently this term generates the familiar spin-o.bit interaction in (5.131),
Note that this relativistic derivation of H'so includes the troublesome factor of \ referred to as the Thomas precession, a factor which must be introduced in a subtle manner in the usual nonrelativistic derivation of (5.135) when transforming from the rest frame of the electron to the laboratory i.e. rest frame of the nucleus (Problem 5.9). Finally, we interpret the third term in (5.131b), but first note that (pK) • p = — (VK) • V is not an hermitian operator. To convert it tohermitian form, we sandwich this term between a spatial integral over wave functions q>* and q> in a symmetric fashion and then integrate by parts.
Dirac Equation in an External Field 89
Discarding the resulting surface term, we are led to the real integrand (pj/) . p _^ -i[jp*(VK) • V
(5.137)
While this hamiltonian does not have an obvious analog in the usual nonrelativistic formulation, it can be understood as a consequence of the relativistic effect of Zitterbewegung. To see this, account for the electron jittery motion to order of the Compton wavelength, Sr < m~\ by the expansion Hzitt. = V(r + Sr) - V(r) = \ br{ 5r} ^
+ • • • * i(<5r)2V2F, (5.138)
where symmetry dictates that the first-order term in (5.138) vanishes and the second-order derivatives are ^V2. Comparing (5.138) with (5.137), we see that (m 5r ~ 1) HLr. *!(»t &r)-2Hziu. ~ Hzitl..
(5.139)
Fine-Structure Energy Levels of Hydrogen. To cap the discussion on bound states, we return to the Coulomb energy levels (5.118) for a Dirac atom, now specializing to the hydrogen atom with Z = 1: ma2 "2M2
1+
(5.140)
To understand how (5.140) arises in the context of Schrodinger theory, we write HNR = H0 + H', where H0 is the lowest-order Schrodinger hamiltonian with V = — cc/r, which generates the Bohr energy levels for n = 1, 2, ... (structure): .2
ma n=-^2=-^T2In1 2a0n2
E
(5-141)
The small splittings of (5.141) to 0(<x2) are the fine-structure corrections caused by H' as given by (5.131b), viz., H' = H'nl + H;.0. + H'Dar,
(5.142a)
generating the first-order perturbation-theory shifts A£ft = (nl|H' |nV> = (nl\H'rel.|n/> +
90 Spin-j Dirac Equation where | n/> refers to the usual unperturbed Laguerre eigenfunction solutions To calculate the relativistic momentum-correction shift, note that (5.143a)
S J P V - - ( « + ?)« implies the square ^2
= (nl
(*« + ? ) ' nl\
(5.143b)
where one may easily verify that 1 1
(5.144)
Then combining (5.143) and (5.144) with (5.132) we find 1
4
< n / | # ; e l > / > = - ^ - 3
- ~
1
3
(5.145)
which breaks the /-degeneracy and is in fact the entire fine-structure shift for the Klein-Gordon atom (4.35). The spin-orbit shift is found from (5.111a), (5.135) with dV/dr = a/r2 and 3
<«/|r" 3 |n/> =
3
3
m a.
n /(/ + i ) ( / + l ) '
(5.146)
giving
=
~2
nl
nl
1
ma -4» 3 C/ + i)(/ + ni) (1 - <5/,o)
(5.147)
for j = I ±\, but zero for / = 0. Lastly, V2V = 47ta^3(r), and at the origin
ll»)|
m3a3
nn which converts (5.137) to the Darwin shift
(5.148)
<5|,o>
net (nl\S3(r)\nl} 2m2
= ^ma. dlt0.
(5.149)
While this Darwin shift is nonvanishing only for s-states, the important point is that it is precisely what the spin-orbit shift would be for / = 0 if 1 — <5, o were replaced by 1 in (5.147). In a sense then, the effect of Zitterbewegung on the Dirac atom corresponds to a "continuation in /" of the spinorbit energy shift down to / = 0.
Dirac Equation in an External Field 91 Next, we combine (5.145), (5.147), and (5.149) according to (5.142b) to obtain the total fine-structure energy shift ma. "2^
A£ft
4
mot. =
1
_
1
1 +
l + 2 2(j
_J
+ m + l)
_3_" 4n|
_
/ r i
_3_
~Jn* j + i
4n ' now breaking the) = I ±j degeneracy but preserving the / degeneracy. Note that (5.150) is identical to the relativistic momentum shift but with; replaced by /. Finally, adding (5.150) to £° as given by (5.141), we are led to the relativistic form (5.140)—a necessary but nonetheless satisfying conclusion. The Dirac theory therefore predicts excited states with fine structure splitting for n = 2 and different j of the order \ x 10" 4 eV, or equivalently, v = E/h ~ 104 Mc/sec. Hyperfine Splitting. Aside from fine-structure splitting, one must account for hyperfine shifts due to the interaction of the proton and electron magnetic moments via the dipole magnetic field of the proton (in rationalized units):
V
'
(5.151)
where the proton is assumed localized at the origin. Since iie\ip = — gegpe2/4memp with ge x 2, gp = 2(1 + KP) x 5.6, and since spherically symmetric s-states imply didjr'1 -+jdiJ'S71r~1, (5.151) becomes for s-states H" = ^Lae-ap53(T).
(5.152)
This splits every line of given j for / = 0 and S = \(ae + op) according to ("e • <*P)S=I ~ (««' °„)s=o = 1 - ( - 3 ) = 4.
(5.153)
Then applying (5.148) and (5.153) to the matrix elements of (5.152), the net hyperfine shift for s-states is AJ?h.f._
8„
m
e
m
e«*
/
S 1 s
.x
mp 2n with the singlet spin state shifted down three times as much as the triplet state is shifted up from the fine-structure energy level. For the ground state, (5.154) corresponds to a 6 x 10"6-eV or 1420-Mc/sec shift. In wavelength language this is the 21-cm line for which astrophysicists search in order to detect the existence of neutral hydrogen in the intergalactic medium. The story is not yet complete, however, because it was experimentally verified (Lamb and Retherford 1947) that, contrary to the Dirac theory, the
92 Spin-^ Dirac Equation
El
n=2
-v \
n>/z
2Si/2 2P,/2
2S|/2 , 2 P | / 2
Structure ma2
Fine Structure ma4
Lamb Shift ma5
Figure 5.1 The n = 2 energy-level splittings in hydrogen.
2Si and 2P± energy levels are not degenerate. Rather, the 2Si line is shifted upward relative to the 2Pi state by some 1058 Mc/sec, as pictured in Figure 5.1. While the theory of this "Lamb shift" is now well understood, a detailed explanation must be postponed until Chapter 15. Qualitatively speaking, this shift is caused by the "cloud" of photons and virtual electron-positron pairs which surround the bound electron, modifying its form factors as given in (5.97) and the resulting Coulomb interaction with the proton nucleus. Suffice it to say now that theory and experiment agree exactly to five significant figures, a truly remarkable result rarely equaled in modern science.
5.E Wave Equations for Other Fermi Particles Thus far we have explored the consequences of the Dirac equation for massive, spin4 fermions (e.g., electrons, protons). We now consider wave equations for massless spin-^ fermions (neutrinos) and also massive spin-f fermions (e.g., the A, 33 "resonance"). Massless Spin4 Particles. Recall from the discussion in Section 3.D that the two helicity states for a massive spin-^ particle, k = +-j, become reduced to just one helicity state as m -»• 0. The explicit structure of the two-component spinor boost operator in (3.94) shows that the 1 = \ right-handed massless state (with spin parallel to the momentum) survives for the (0, \) representation, while X = — \ survives for the (j, 0) irreducible representation of the homogeneous Lorentz group. Experiment alone (see Chapter 13) has determined that the massless neutrino is left handed and the antineutrino is right handed. Consequently the two-component (j, 0) free-particle wave function >Lv for a neutrino satisfies a wave equation inferred from the boost in (3.94): {id, + a • p)(pLv(x) = 0,
(5.155a)
whereas the (0, j) right-handed antineutrino wave equation is {id, -
(5.155b)
Wave Equations for Other Fermi Particles 93
While these wave equations were first obtained by Weyl (1929), they were rejected for 28 years because of noninvariance under spatial reflection {cpL -> cpR, q>R -»cpL). We shall return to this subject in Chapter 6. To link (5.155) with helicity eigenstates, note that the plane-wave solution of (5.155a) for positive-energy states, proportional to e~ipx with E= | p | , satisfies <*P
(5.156a)
in accordance with a left-handed neutrino. Likewise, the plane-wave solution of (5.155b) for positive-energy states, proportional to (e~ip '*)* with E = | p | , obeys «" P
(5.156b)
corresponding to a right-handed antineutrino. On the other hand, the planewave solution of (5.156b) for negative-energy neutrino states, also proportional to eip'x with E = | p |, satisfies « - p > , - ( - p ) = ?v-(-p)-
(5.156c)
Comparing (5.156b) and (5.156c), we see that
(5.157)
This is our first concrete example of the Feynman interpretation, identifying a negative-energy particle state with a positive-energy antiparticle state. More examples will be given in Chapter 6. It is possible, and extremely useful, to couch these two-component neutrino equations in four-component Dirac language. One way to proceed is to note that (5.155) corresponds to (5.17) with m = 0 along with the identification cpL-KpLv, (PR-+
Then we define the Weyl-representation bispinors uLv(P)-P-uLv(v)=(q>L^)\
(5.159a)
^(p) = P + M p ) = ( g - i V ° _ ( _ p ) )
(5.159b)
with the positive-energy bispinor uLv(p) describing a left-handed neutrino and the negative-energy bispinor vm(p) describing a right-handed antineutrino, and both having momentum p. As for the four-component dynamical equations that these bispinors satisfy, since P + projects out the lower negative-energy components in the Weyl representation and uLv has no such component, it is obvious that P + » L v = i(l + iy5)uLv = 0,
(5.160a)
94
Spin-i Dirac Equation
and similarly, P-vFf, = $(l-iys)v»,
= 0.
(5.160b)
We may also proceed by starting with the free-particle Dirac equations (5.58) for m = 0, M P ) = M P ) = 0,
(5-161)
then multiplying (5.161) on the left by iy5 y0 and using iy5 y0 y = a, E = | p | we obtain
a • pu(p) = - iy5 v(p).
(5.162)
Next observe that the bispinor helicity eigenstates (5.65) become in this case « ' P"ZJP) = - « £ » »
- 5 ' W P )
= ^V(P)-
(5163)
Combining (5.162) with (5.163), we are once more led to (5.160). Another useful bispinor expression for massless fermions is the probability current density j ^ — ^iy^- Applying (5.159), we may write the momentum-space current for Weyl neutrinos obeying (5.155) as (dropping the factor of \ in P_ by convention)
= tUp'Wi - iysWM
(5-164)
Since yM(l — iy5) = (1 + iys)yll, the latter y-matrix combinations used in (5.164) can also be expressed as uF- = uP+. For an antineutrino current, it turns out that the analog of (5.164) is
(5165)
The projection operator P+ in (5.165) again follows from (5.159), but the reason for reversing the momentum in the spinors in (5.165) must await the discussion of charge conjugation in Chapter 6. Finally, note that even for q2 = (p' — p)2 / 0, we assume that (5.164) and (5.165) do not develop anomalous magnetic-moment contributions or form factors, because neutrinos interact only weakly with matter (Chapter 13). Such weak interaction experiments detect two types of neutrinos, associated with electrons and muons, respectively. Spin-§ Particles. It is also possible, and again useful, to construct a Dirac bispinor wave function for a spin-| free particle. In Section 2.E we built up a spin-f two-component spinor by the Clebsch-Gordan combination (2.62) of a spin-1 polarization vector and a spin4 spinor, with the spin-! combination (1 x \ = | -(- \) removed by the condition (2.63),
(5.166)
Wave Equations for Other Fermi Particles 95 where such spin-| covariant bispinors satisfy a free-particle Dirac equation (Rarita and Schwinger 1941, Auvil and Brehm 1966) (P - m)M„(p) = 0,
(5.167)
with (2.63) replaced by (see Problem 5.11) y"H„(p) = 0.
(5.168)
Given (5.167) and (5.168), u^ automatically obeys the weaker conditions (p2 - m2)uJp) = 0,
(5.169)
/?X(P) = 0-
A polarization or spin sum can then be formed from these spin-| bispinors in analogy with the spin-1 polarization sum (4.46) and the Dirac projection operator (5.69). In the rest frame this projection operator is
^; i =i d > r = * u - ^
(5.170)
because (2.63) requires CT,^ = ^ijOj = 0, a condition obviously satisfied by (5.170), since aiai = 7>. The normalization of (5.170) is chosen so that ^ijSPjk = SPik- Then the boosted version of (5.170) is (see Problem 5.11) (5.171a)
^
+ m)+>(r, +
^_„,)(v,
+
O + my
£;)
(5.171b) (5.171c)
As long as the particles are on mass shell, this scheme can be extended to arbitrarily high spin [see e.g. Fronsdal (1958), Scadron (1968)]. There are, however, alternative formulations of spin-| wave functions. Moreover, for p2 # m2 or for wave equations involving interactions, ambiguities arise and such theories are no longer unique. In the context of lagrangian field theory, interactions involving massive particles with spin-| and higher (and sometimes even spin 1) are "nonrenormalizable" (see Chapter 15). Luckily, nature has been kind enough to see to it that most of the fundamental particles detected so far have low spin. General references on the Dirac equation are: Dirac (1958), Hamilton (1959), Schweber (1961), Bjorken and Drell (1964), Muirhead (1965), Sakurai (1967), Pilkuhn (1967), Bethe and Jackiw (1968), Schiff (1968), Baym (1969), Berestetskn et al. (1971), Jauch and Rohrlich (1976).
CHAPTER 6
Discrete Symmetries
In Chapters 2 and 3 we discussed in detail the space-time continuous symmetries: translations, rotations, and Lorentz velocity transformations. We now consider the discrete (not generated from the identity) transformations of space-time: space reflection P and time reversal T, along with the charge conjugation (particle-antiparticle) transformation C. It turns out that C and P are unitary operations which have conserved hermitian observables as in continuous-symmetry cases. While they have no classical analogs, they do generate physically significant selection rules. On the other hand, T and CPT are antiunitary operations which give rise to neither conserved quantities nor selection rules. Instead the latter transformations lead to reciprocity or phase relations, which do have important physical consequences. These discrete symmetries (and especially CPT) will be needed to build up the proper covariant (Feynman) scattering diagrams in Part III of this book.
6.A Charge-Conjugation Transformation We have seen that in the presence of an electromagnetic field, the KleinGordon and Dirac equations are modified by the minimal replacement factor idp -> id^ — eA^ and the Maxwell equation is then driven by a source current ej^. The factor e = ± | e | represents the charge of the particle in question, but its sign does not appear to be of fundamental significance. In fact, we have become accustomed to assigning negatively charged electrons e~ and positively charged protons p as particles, and positrons e+ and anti96
Charge-Conjugation Transformation
97
protons p as antiparticles, simply because of the observed preponderance of e~ and p in nature. Defining Properties. The charge conjugation operation converts particle to antiparticle. In quantum mechanics it is defined by the transformation laws UclQUc=-Q, Uc I = 1c I >Ac>,
(6.1)
Uc I p, A> = tic I p, A>,
(6.2)
generating the charge conjugate state up to a phase t]c. Since |\j/ \2 = \i//c\2, the operator Uc is unitary with | rjc \2 = 1. Obviously t/ c commutes with x and p, and therefore with the commutator [x, px] = i, implying that Uc is a linear operator in the sense that cucp^ + a2>2 -> «i >? + v-ifti ( see Section l.C). Then if Uc is applied twice to a state, it will reproduce that state. For such a linear operator it is always possible to choose phases so that U2=l,
WCUC=L
(6.3)
That is, the unitary, unimodular operator Uc can always be taken from (6.3) as hermitian, [/£ = Uc. Thus Uc is an observable operator, with eigenvalues which are simultaneously measurable along with energy eigenvalues, provided Uc commutes with H—that is, if H is invariant with respect to C. The catch is that Uc must also commute with the charge operator Q since the latter commutes with H and has measurable eigenvalues. Then the phase rjc in (6.2) can also be an observable eigenvalue, r\c = ±1. But we know from (6.1) that Uc and Q anticommute; only for neutral states, Q = 0, do Uc and Q effectively commute. Thus it is meaningful to define charge-conjugate eigenstates and a "charge parity" r\c = ± 1 for neutral states alone. It turns out, however, that this class of states is further restricted by the existence of other (additive) internal quantum numbers such as strangeness S and baryon number B which transform like Q under Uc as in (6.1). Thus only neutral "self-conjugate" states with Q = B = S = 0 have a well-defined charge parity t]c = + 1 . The choice in these cases between r\c = + 1 and r\c = — 1 is determined by experiment. We now turn to the finite-dimensional representations of Uc which have a complex-conjugation part consistent with minimal electromagnetic coupling with Ap taken as real, as well as a spin-dependent part to be determined. Spin-0 Particles. In the presence of an electromagnetic field, the KleinGordon equations for spinless particles and antiparticles are [(id - eAf - m2j>(x) = 0, 2
2
[(id + eA) - m ]4>c(x) = 0.
(6.4a) (6.4b)
It is clear that (6.4b) can be obtained from (6.4a) with A^ real provided 0cM = 1c
(6.5)
98
Discrete Symmetries
with positive- and negative-energy parts related by 4>c±(x) = fc#*( x ) a s expected. From (6.5) we see that the finite-dimensional spin-0 representation of Uc is just the unit matrix. Given (6.5), a self-conjugate particle with e = 0 and r\c = ± 1 has a vanishing probability current, >*i^ (j>. It is also clear that the spin-zero electromagnetic current density fj" = ej^ changes sign under charge conjugation (Problem 6.1), but the manner in which this relates the positive and negative energy currents in momentum space is worth pursuing for our later work. Consider, therefore, the form of 7™ for free, scattered, positive-energy states (4.63b),
(6.6)
The charge-conjugate current for positive-energy states is then
= eF(q )(-p'-p)fl,
(6.7a) (6.7b)
where (6.7a) has the form of (6.6) but with e replaced by — e for the antiparticle current. The form (6.7b) corresponds to the particle current, but for negative-energy states with p and p' replaced by — p and —p' (traveling backward in space-time). It is this latter interpretation rather than (6.7a) that will prove more useful in subsequent chapters. Spin-1 Particles. Since the charge-conjugation operation changes the sign of the electromagnetic current, the Maxwell equation C\Afl= y'*m requires that if we want A^ to transform simply under C, we must take A real (since the photon has zero charge) with 4(x)=-^(x).
(6.8)
While this choice is an empirical one, it is consistent with the C-invariant dynamical equations (6.4), as Uc converts (6.4a) to (6.4b) using (6.8). Also note that (6.8) is similar to the Klein-Gordon form (6.5), but with the photon intrinsic charge parity uniquely specified as ^c(y) = — 1. Massive spin-1 particles likewise obey a Klein-Gordon-type equation (4.42), and so #(x)=f/c*:(4 (6-9) As is the case for spinless particles, the charge parity of massive spin-1 particles must be determined from experiment. Furthermore it turns out, that the spin-1 electromagnetic current can in principle have an "anomalous" part with charge conjugation rjc = 1 (Lee 1965). Spin-2 Particles. In a similar fashion, we observe that the Dirac equations for spin-^ particles and antiparticles of the same mass in an external field are [# -eA-
m]il/(x) = 0,
[# + eA - m]\l/c(x) = 0.
(6.10a) (6.10b)
Charge-Conjugation Transformation
99
Since \j/ is the Dirac analog of an adjoint, ij/T (T means transpose) corresponds to the complex-conjugate operation (actually ij/T = y0 \j/*) which is necessary to transform ijt to i//c. Taking the "barred transpose" of (6.10a) and using ABT = ATBT and y^ = y^, we see that [-{id + eA)-yT-m]iJtT(x)
= 0.
(6.11)
It is therefore natural to search for a Dirac matrix C such that CylC-'
= -y„
(6.12)
for then, operating on (6.11) to the left by C gives [(id + eA) • y - m]C\]/T = 0,
(6.13)
which is identical to the charge-conjugate Dirac equation (6.10b) provided
M*) = fcC-FM-
(6-14)
Comparing (6.14) with (6.5), we observe that Cy0 is the Dirac [(^, 0) + (0, ^)] representation of the operator Uc. To determine further the structure of the matrix C beyond the defining property (6.12), we note that since Uc is unitary in Hilbert space, C must be unitary in the Dirac space: CC< = 1. Moreover, (6.14) can have i//c and ifr interchanged, since t// is no more fundamental than \\ic according to (6.2); so # c ) = ijcC#?(x).
(6.15)
Comparing (6.15) with (6.14), we find C" x = CT = y0C*y0. Next, evaluate (6.12) for pL = 0 and use the (representation-dependent) relation yj = To which leads to C = — y0Cy0. Combining these last three constraints gives C=-CT,
C = - C _ 1 = -Ct.
(6.16)
It is sometimes convenient to obtain a specific (but not unique) representation for C, which we choose to be real while obeying (6.16): C = iy5io2 = -iy2y0-
(6-17)
Paralleling the discussion for Klein-Gordon particles, we now write positive-energy antiparticle bispinors in terms of negative-energy particle bispinors as
(6.18)
for a free antiparticle of momentum p. The conjugate bispinor satisfies, for A, = 0, V - m)uc(p) = (p - m)u(p) = 0, (6.19) so that we may choose «c(P) = «(P)(6-20) p x Then combining (6.20) with (6.18), (6.16), and \j/_ -> ve' , we conclude that t;(p) = Cur(p),
u(V) = CvT(p).
(6.21)
100
Discrete Symmetries
As a specific check of (6.17) and (6.21), we evaluate the latter in the DiracPauli representation using (5.64):
=
(6-22)
7rriH
Note then the property of two-component spinors for k = ±\, ia2(p*a)(p) = e ¥ i V U ) ( - P ) ,
(6.23)
obtained from (3.91) with p = p9i^, — p = p„_flin+^ (also see Problem 3.7). Applying (6.23) to (6.22), we recover v = CuT from (5.64b). Finally we find the Dirac analog of the off-diagonal momentum-space current (6.7), with
2
(6.24a)
2
p) = FM )y, + F2(q )™^l2m,
(6.24b)
where we have used (5.97). The charge-conjugate positive-energy current is then
r
(6.25a)
= ^(p')C- r M (p',p)Ct) (p)
(6.25b)
= e(-v(p))r„(p',p)v(v'),
(6.25c)
where (6.25b) is a consequence of (6.21) and (6.25a). Because the transpose of a number is a number, (6.25c) follows from (6.25b) and the defining property of C, (6.12). Note that while (6.25a) is similar to (6.7a), (6.25c) is similar to (6.7b), the latter two both becoming — elp^ for p' — p. The fact that the momentum dependence of the negative-energy spinors in (6.25c) is reversed is characteristic of "backward propagation in space-time." Since uc = u, we learn from (6.25a) that^ m c = -;^ m , while (6.25c) indicates that^ m c + =^ m _ (see Problem 6.1). Note too that of the 16 fundamental ^-matrices, only y^ and a^ change sign under C: C M 1 ' V ffM»' ty.Vs. 7s)C = (1, -y„, -a„v, iy^y5, y5)T.
(6.26)
A subtle but important point is that the minus sign in \j/c = —r]*\j/C~l is associated with the sign of ( — v) in (6.25c) or the antineutrino current (5.165), not to be included in (6.26). For hamiltonian densities this sign is linked with the connection between spin and statistics; in field-theory language it occurs because of the antisymmetric nature of fermion fields. Representations for Higher Spin. The action of Uc on a single-particle helicity state of higher spin continues to follow (6.2), but the identification of charge-conjugate positive-energy states of spin ;' with negative-energy particle states still necessitates complex conjugation coupled with momentum
Charge-Conjugation Transformation 101 inversion. A further complication is that charge conjugation converts an irreducible representation (A, B) of J2? to (B, A). In particular, by analogy with the second derivation of the Dirac equation in Section 5.A, given states
(6.27)
A particular representation of C 0) is (see Problem 2.4) Ci?m = <Jm'\e-*J*\jm> = ( - T ^ - m ,
(6-28a)
which possessses the general properties C 0)t C 0) = lf
cU)*CU) = (-) 2 j '.
(6.28b)
This matrix transforms spin matrices according to a unitary rotation by n in (2.28), accompanied by complex conjugation, C U)-ijU) C 0) =
_jO')*
=
_jO)T
(629a)
It also complex-conjugates rotation matrices as well as inverting the momentum of (j, 0) boosts [following directly from (3.43)]: c u>- ' / ) 0 ) ( K ) C C 0)-
U)
= Dij)*(R),
(6.29b)
(J)
W
(6.29c)
iZ)«^(L_p)C = D *(L„),
thus justifying (6.27). Of particular interest is the construction of the Dirac C using the Weyl representation for the y matrices, with (6.12) paralleling (6.29a) (see Problem 6.1). Denning ip = (££) in this representation, we recover f/*i^c = Cij/T = Cf}\jj* with C constructed in the usual manner [exp( — ina2/2) = —ia2 = C (i) = — C(i)],
C
-(T;HT:J-^-
<-»
This is perhaps a more revealing construction of C than in the Dirac-Pauli representation (6.21)-(6.23). For more details on high-spin representations, see Weinberg (1964a). Charge-Conjugation Invariance. Applying the unitary operator Uc to the Klein-Gordon, Dirac, and Maxwell equations converts A^ to — A^ and the charged current densities (constructed from matter wave functions) f™ to —fj". Aside then from noninteracting parts of the hamiltonian density, there are interacting terms such asf^A* which remain invariant under the chargeconjugation operation. Converting such dynamical relations as the Dirac equation in an external field (6.10a) into Schrodinger-equation language, application of Uc results in Uc(id, - H)| = (id, - UCHU-Cl)\,c> = 0.
(6.31a)
102
Discrete Symmetries
Then invariance under C means that UcHUc
l
= H, or [Uc, H] = 0.
(6.31b)
For many interactions of interest, however, the form of the hamiltonian is unknown. It is therefore of interest to formulate invariance laws in terms of the general S-matrix operator [recall from (1.66) and (1.67) the relation between H7 and S,]. Charge-conjugation invariance then corresponds to [Uc, S] = 0, or S = U^CSUC
(6.32)
with matrix elements whose absolute squares are measurable, |S|»> = »J?/»ta
(6.33)
where | Ci} is the charge conjugate of the state | i>. To be more specific, for i and/corresponding to spin4 plus spin-1 (positive-energy) free particles, the S-matrix elements can be expressed in terms of M-functions as (see Section 3.D) | S | i> = s^kJu^Mjf, iMp^),
Ci)v(pfy(kf),
(6.34a) (6.34b)
and we again note the reversed order of the negative-energy wave functions and the minus sign associated with v in (6.34b). Then C-invariance leads to the relation between M-functions, Ci C Mjf> 0 = IcCMlM ) ~ '> (6-35) where rjc = r\*{ r\ci is the total charge-conjugation phase for the process. We see that (6.34) and (6.35) are a generalization of the transformation law for currents (see Problem 6.2). The obvious test for C-invariance would be to check the equivalence between particle and antiparticle cross sections and lifetimes. This is not always easy, however, because of the difficulty in preparing antiparticle targets in our particle-dominated world. Nevertheless certain tests do exist which lead us to believe that strong and electromagnetic interactions are invariant under C (but weak interactions break C-invariance). It is therefore useful to assign C-parities to neutral particles. Under C, not only the charge Q but also the baryon number B and strangeness S change sign; therefore eigenstates of C are not all neutral particles, but only those "self-conjugate" mesons with Q = B = S = 0: n°, rf, p°, a>°, (p°, etc. Given rjc(y) = - 1 , a multiplicative quantum number, the existence of the spin-0 electromagnetic decays n° -»• 2y and rf° -* 2y requires that r]c(n°) = r]c(t]°) = rjc(y) = +1. This in turn leads to the selection rules n° -f* 3y, rf -/> 7t°7r°y—decays as yet undetected. In a similar way the existence of the decays p° -» n°y, a>° -* n°y, and ($)0-+rfy requires r\c(p°) = t]c(a)°) = t]c((f>0) = — 1 for these spin-1 mesons. Particle-antiparticle (aa) configurations are also self-conjugate, possessing the C-parity (see Problem 6.2)
t]c(ad)={~r\
(6.36)
Space-Reflection Transformation
103
where / is the relative orbital angular momentum and S is the total spin of the ad system (to be derived in Section 6.B). A list of the common elementary particles, along with their masses, spins, and quantum numbers, is given in Appendix III.
6.B Space-Reflection Transformations Defining Properties. A spatial reflection of the coordinates of a system means x -* x' = — x and t-*t' = t. Such a reflection can be interpreted as an improper rotation Rn converting xfl<#) to x „ _ M + ^ but with det Rn = — 1; alternatively it can be thought of as an improper Lorentz transformation x l = (xo> _ x ) with A£v = #„v and det A„ = — 1. For true 0(3) and ¥ scalars such as p 2 or p2, or polar vectors p or p^, improper rotations or Lorentz transformation have no effect on the transformation laws different from proper transformations with det R = det A = 1. For pseudoscalars p • J or $y5 \jj, or axial vectors L = r x p or ij/y^ y5 \\i, the transformation laws contain an additional sign change corresponding to a factor of det RK or det A„ appearing in these laws [see e.g. (5.46)]. In quantum mechanics the spatial reflection is defined by the action of the unitary operator Up on operators and states [see e.g. Wick (1959)]: VplxllUP = xl = {x0, - x ) , UP\
U~PlJUP = Jn = J,
t/p|pA> = M - r ^ | - p , -A>,
(6.37) (6.38)
where the helicity phase for a particle of spin s follows from (3.77) for p e 0 . Since x and p change sign under a space reflection, the commutator [x, px] = i remains unchanged, and therefore UP is a linear unitary operator, as is Uc. If UP is applied twice to a state, it will reproduce this state; so again, like Uc, the space reflection operator is also unimodular: Uj=l,
UPUP=l.
(6.39)
As a consequence, UP (like Uc) is also hermitian, with eigenvalues which are simultaneously measurable with the energy provided UP commutes with H. This results in the phase t]Pm (6.38) becoming the intrinsic space parity (or simply the parity) of the state. Integer-Spin States. Consider first spin-0 solutions of the free Klein-Gordon equations for a particle in its normal and space-reflected state: (p2 - m2)
(p2 - m 2 )0 P (xJ = 0.
(6.40)
Since p2 is a true Lorentz scalar, these solutions are related by
(6.41)
In a similar fashion, the spin-1 wave function satisfies
(6-42)
104
Discrete Symmetries
where the minus sign associated with tjP in (6.42) denotes the rest-frame behavior of vector particles, and where g{0) = 1, gr(1 2 3) = — 1 (not summed over) distinguish between the time and space components of a four-vector, which transform differently under space reflections according to (6.37). Herein lies a key difference between C and P: applying P twice with (j>{xnn) ~
(6-44)
Note that the overall minus sign in (6.44) is similar to (6.42), while the factor ( — f~x is the same as for the helicity state in (6.38). For (integer) orbital-angular-momentum states,
= f7,i'lT(r).
(6.45a)
But - r ^ = r„_ 9> „ + , and 17(71 - 6, n + $) = (-)lY?(0, <j>) then require the phase in (6.45a) to be m = (-)'• (6.45b) This orbital parity must be included along with intrinsic parities to determine the total parity of a system. Spin-j States. While spinors change sign under a complete rotation R2K, they do not change sign under two successive space reflections. To see this, note that the spin or analog of (6.44), obtained from (3.91) by inspection is, for P = Pe,tf,>
Space-Reflection Transformation 105 these dynamical relations. Put another way, the (j, 0) irreducible representation of if transforms under a space reflection into the (0, ^) irreducible transformation. This is sometimes considered the major reason for combining (j, 0) and (0, j) two-component spinors into a Dirac (\, 0) + (0, %) bispinor \jj in the Weyl representation. Then the action of P is modified by
H::H:KX:K
<->
Alternatively, in the Dirac-Pauli representation with cp ->
(6-49)
for p = p„ 0 , as can be verified from (6.46) and (5.64a). In spite of the simple transformation law (6.48), the presence of the y0 matrix precludes a double reflection from leading to t\\ = 1. Thus rjP is in general complex for spin \ and all other fermions. While rjP is therefore not absolutely measurable for fermions, the relative parity of any two such fermions can be determined, and this is adequate for all physical processes because angular-momentum conservation requires an even number of fermions to interact. Further, baryon (nucleons, hyperons, etc.) and lepton (electrons, muons, neutrinos) number are found to be conserved independently in any reaction (particles — antiparticles = constant). We may therefore define an absolute intrinsic parity for baryons and choose by convention the normality n = t]P( — ) J ~* so that normal spin-./ baryons satisfy n = 1 and Jp = \+, | _ , j + , ... while abnormal baryons have n= — 1 and J
JP — I " — 2
1+ i i
5> 2
>
—
Space-Reflection Invariance. Paralleling the discussion for C-invariance, space-reflection invariance or parity conservation for a system driven by a hamiltonian means that H(\) = H( — x), so that UPHUp 1 = H,
or [Ur, H] = 0.
(6.50)
Alternatively, in terms of an S-matrix, parity conservation corresponds to [UP, S] = 0, or S = UpSUp,
(6.51a)
with matrix elements whose absolute squares are measurable {t]P = Jj*/f/pj), < / | S | i > = fj,
(6.51b)
For a representative process involving spins \ 4- 1 -* \ + 1, the transformation law under P for the M-function (6.34a) follows from (6.51), and employ-
106
Discrete Symmetries
ing (6.44) and (6.49), we have M„v(/. 0 = WwdMyoM^iPf, Pi)y0
(6.52)
[note that the helicity phases in (6.38), (6.44), and (6.49) cancel in (6.52)], where n is the total normality n = nfnh nf = t1Pf( — )Jf, ni = rjPi(—)Ji. This just means that if the reaction is normal (n = 1), then M must be expanded in terms of "true" scalars, vectors, and tensors. For an abnormal process with n = — 1, M must instead be expanded in terms of "pseudo" scalars, vectors, and tensors, containing an odd number of y5 or e^^ symbols, since for example in the spinor case y0 y5 y0 = — 7sGeneral tests of P-invariance include the absence of pseudoscalar terms like o • p in cross sections and specific relations between various polarization and asymmetry measurements. It now appears that strong and electromagnetic interactions are invariant under P. This in turn leads to selection rules correlated with a parity assignment for every particle—an assignment which must remain unchanged from process to process. It is conventional to choose the proton and neutron as having the same relative intrinsic parity. Then from the experimental fact that the strong interaction n~ capture in n~ + d -» n + n is in the s-state, P-invariance holds provided t]P(n') = — 1. [That is, the initial total angular momentum of the spinless n~ and spin-1 deuteron is J, = 1, since /, = 0. Identical neutrons have spins antiparallel, so Sf = 0 requires Jf = lf = I by angular-momentum conservation. Finally, the intrinsic parity of d, a pn bound state, must be the same as for nn, so that the initial and final relative parities are t]Pi = t]P(n")( — )'' and t\Pf = ( —)(/ = — 1; P-invariance (t]Pi = r\P}) then requires rfP(n~) = — 1.] In a similar fashion, the final-state photons in n° -> 2y are observed to have perpendicular polarization vectors, implying a decay matrix element proportional to the pseudoscalar E' x £ • k in the n° rest frame. Then P invariance requires n° also to be a pseudoscalar, so that (6.50) or (6.51) is valid if rjP(n°) = — 1. For strange particles, the relative ApK+ parity is observed to be negative. It is conventional to choose r]P(K+) — — 1, in agreement with pions, and also take A to have the same relative parity as nucleons. We further choose the spin-*r+ p, n, A particles to have t]P{p, n, A) = 1 and normality n = 1. The remaining intrinsic parities of the other elementary particles are then uniquely determined (see Appendix III). Selection rules for nuclear transitions such as a-particle emission are also a consequence of P-invariance, as are electromagnetic nuclear and atomic transitions with photon emission. In the latter case, electric-multipole radiation of rank J corresponds to r\P = ( — )J, an effective normal (n = 1) spin-J particle. Then for initial and final orbital-angular-momentum states (in nuclei or atoms) lt and lf, P-invariance requires (-)lf+J = ( — )'', or lf + /; + J to be even. For magnetic-multipole radiation, corresponding to abnormal r\P = — (— )J transitions, P-invariance requires lf + /; + J to be odd. The combination of C and P gives further restrictions on spatial and charge parities. In particular, the charge-conjugate (j, 0) + (0, j) wave func-
Space-Reflection Transformation 107 tion has the Dirac form (6.14) with C given by (6.30) in the Weyl representation for spin j with CU) = (— )2jCU). The space-reflected wave function is likewise given by (6.48). Then for particle a and antiparticle a, UP^c = r]p{a)nMPCV, Ucxl>P = r,c(a)^(a)CliV,
(6.53a) (6-53b)
and since (6.53a) and (6.53b) both must give the same result, the identity Cp = ( - )2j'j?C requires that r,P(a)nP(a) = (-)2J=±l
(6-54)
for bosons and fermions, respectively. The charge parity for the aa system (6.36) then follows from (6.54) along with the factors ( —)' for spatial exchange r -> — r under P and + (— ) s for spin exchange s + s -> s -I- s under C, where S is the total spin (see Problem 6.4). It is useful to construct eigenstates of CP. Since r\c(n°) = ~ / 7P( 7C °) — U it is clear that rjCP(n°) = — 1. Extending this to non-self-conjugate kaons, we choose by convention UCP\K°> = - |K°>, U_CP\K°y = - \K°>, and then define CP eigenstates KLS = 2"*(|X°> + |-K°>) for CP-conserving reactions so that ricr(n°) = flcp(KL) = -riCP(Ks) = - 1 .
(6.55)
Violation of C and P. For a long time it was thought that the discrete symmetries C, P, and T were—indeed had to be—invariant for each of the fundamental forces of nature: strong, electromagnetic, and weak interactions. (We shall postpone a discussion of the gravitational force until Chapter 14.) Due to the pioneering work of Lee and Yang (1956), however, it was discovered that P and C are not separately invariant for weak interactions, although the combination CP appears to be an approximate symmetry operation. For example, in the /?-decay of 60 Co (see Wu et al. 1957), the emitted electron has a preferred direction with respect to the spin of the 60 Co nucleus, violating P. In fact,
108
Discrete Symmetries
C. Using (6.36), it can be shown that the charged pion decays Ks — n+n~ etc. follow a similar pattern. Finally, it was discovered in 1964 that a small component of the CP-violating decay KL -* 2n does exist, about 0.1% relative to the weak decay amplitude of Ks -*• 2n. Consequently, even CP is not an exact symmetry in nature.
6.C Time-Reversal Transformation Defining Properties. The operation of time reversal (T) or motion reversal is of fundamental significance, underlying the laws of classical as well as quantum physics, because in the former case Newton's laws are invariant under T. On coordinates, time reversal means t -+ — t and x -» x or xj, = — g(fl) x„, again an improper Lorentz transformation A^ = —g^ with det A' = — 1. Also p -> — p, J -v — J, and j^ -* (j0, — j) under time reversal. The quantum-mechanical time-reversal operator
(6.57b)
<<£/1 >A,> = <>r/1 >Ar;>* = <
(6.57c)
Writing 9~ = UTK, where UT is unitary and K is a complex-conjugation operator (changing i ->• — i), a time-dependent spinless state transforms according to
^•\Ht)> = nTK\H-t)>,
(6-58)
2
where r\T is a time-reversal phase with | r\T | = 1. It is not possible, however, to interpret rjT as a "time parity", because | i^(t)> is not an eigenstate of 9~. Put another way, for 2T1 = el the antilinear property of ST prevents a rescaling of the phase e to unity, and 3~3 = E3~ = 3~s implies only that e* = E is real, so that 5" 2 = ± 1 , ^3- = 1. (6.59) Integer-Spin Particles. To acquire a feeling for how to deal with an antiunitary operator in practice, consider first a plane-wave state for a spinless particle <x |p> oc eipx. Then for r\T = 1, we have 9~ |p> = | — p> and <x|^|p> = <x|-p> = e-'px,
(6.60a)
this expression being equal to <,T t x|p>* = <x|p>* = < r i p ' \
(6.60b)
Time-Reversal Transformation
109
A similar argument applied to partial waves leads to
(6-61)
Looking then at a spinless Klein-Gordon particle in the presence of an electromagnetic field, we compare [(id0 - eA0(x))2 - (N - eA(x))2 - m2]4>(x) = 0, [ ( - id0 - eA0,T(xt))2 - (iV - eAr(xt))2 - m2]<M*,) = 0,
(6.62a) (6.62b)
and applying (6.61) to (6.62b), conclude that <M*«) = 1T4>*(X\ In a similar fashion, spin-1 wave functions transform as
(6-63) (6-64)
0M.r(*«) = >7r0M >*(*)•
With p-> — p and X -»A, the corresponding polarization vectors (3.101) transform according to (for p in the 1-3 plane,
(6-65)
consistent with the general time-reversal transformation of a single-particle helicity state of spin s and p 9 0 , • ^ | M > = >7r(-)2sK|-MX
(6-66)
and the identification UT = exp( — inJ2)- (See Problem 6.5.) Spin-£ Particles. Since the time-reversed Klein-Gordon wave function corresponds to a complex conjugation, also needed in charge conjugation, it is not surprising that the Dirac wave function and its time-reversed partner satisfying [y0(id0 - eA0(x)) - Y • (iV - eA(x)) - m]tfr(x) = 0, (6.67a) bo(-ido - eA0>T(xt)) - y • (iV - eAr(x,)) - » # r ( * ) = 0
(6.67b)
T
are related as in charge conjugation through the form \[i = y0 \j/*. That is, ^T{xt) = nTW{x),
(6.68)
where T is the Dirac matrix obeying T ^ T " 1 = <,,„, V
(6.69)
This behavior is necessary to maintain the correct transformation property of the current,^ = ij/y^i^ -> g^j^, under time reversal. The positive-energy free-particle bispinor (5.64a) then satisfies (see Problem 6.5) ul»{-p)={-)2'TuWT{p)
(6.70)
110 Discrete Symmetries
in accordance with the helicity state (6.66) for p 9 0 . The representationdependent form for T consistent with (6.17) and (6.48) is T= -io2y0
= y2lyxy0.
(6.71)
General Spiny. For angular-momentum states \jm}, the transformation law $~~1J$~ = - J leads to £T\jrriy = rjj(-)~m\j, — m>. Alternatively, from (6.71) we may infer that the (j, 0) if representation of UT is —ia2, and the (j, 0) representation exp( - nrJ^) = CU) by (6.28a). This gives ST2 = ( - ) 2 s for a particle of spin s, corresponding to the additional phase occurring in (6.66) and (6.70). Note that both operations C and T have representations related to exp( — inJ2), while P and T transform coordinates and fourvectors in a similar fashion, employing gw. We will demonstrate shortly that it is the CPT theorem that supplies the vital link between these three inversion operations. Time-Reversal Invariance. For a nonrelativistic spinless particle, 9~ = K, so that under the time-reversal operation the hamiltonian governing the particle's motion transforms as ^H{t)^~l
= H*(-t),
(6.72)
consistent with the transformation law for Schrodinger states (6.58). Further, if H is real and time independent, then the state of the system is invariant under all time translations (including time reversal) and (6.72) becomes [JT, H] = 0.
(6.73)
Because 3~ is antiunitary, no conservation law, such as exists for space reflections, follows from (6.73). Alternatively we may formulate time-reversal invariance in terms of the S-matrix which transforms in the interaction picture S, = U,(co, — oo) according to ZTS,^-1
= U,(-oo, oo) = Si
(6.74)
Since S is unitary and not hermitian, a commutation relation like (6.73) will not be valid. Nevertheless the matrix elements of (6.74) lead to dynamical constraints on "motion reversal". Referring to states i and / as initial and final states of the system and suppressing the interaction-picture index (other pictures give the same result—see Chapter 7), we find
\s |0 = (f \jrwr\ /> = nT
Time-Reversal Transformation
111
While it is difficult to verify T-invariance, at the present time observations are consistent with T-invariance for strong and electromagnetic interactions. A principal tool is the reciprocity relation (6.75) coupled with rotation invariance of the partial-wave expansion (or alternatively P-in variance), which converts the time-reversed momenta p' = — p back to p. The effect of the helicities is removed by summing the absolute square of (6.75) over all spin states. Since the latter quantity is proportional to the product of the initial spin multiplicities 2s + 1 times the initial flux [proportional to the square of the CM three-momentum] times the unpolarized differential cross section (we shall see many examples of this in Chapter 7 and thereafter), we are led to the "detailed balance" relation for two-body reactions,
For example, cross-section measurements of the strong-interaction process p + p->n+ + d and its time-reversed partner 7t+ + d-> p + p are consistent with (6.76). Originally the T-invariance of (6.76) was used to infer the spin of the pion, Sn+ = 0. An interesting application of T-invariance is Kramers's theorem, stating that the stationary states of a T-invariant system containing an odd number of fermions are pairwise degenerate. This follows because the Schrodinger states | if/} and | i^r> = ST 11//} are then simultaneously measurable with the same energy. For each fermion 2T1 = ( —)2s = — 1, a multiplicative result that remains unchanged for an odd number of fermions, so that
(6.77) Such orthogonal states cannot be proportional; they are therefore independent but degenerate. Kramers's theorem is used extensively in atomic and nuclear physics. In solid-state physics time-reversal invariance leads to inversion symmetry of energy bands associated with periodic-potential problems; i.e., £,(k) = £,( —k). Finally, in astrophysics and cosmology, time-reversal invariance plays a central role in the equilibrium dynamics of Saha's equation. This ultimately explains why primordial nucleosynthesis occurs at T ~ 8 x 108 °K rather than at the deuteron binding energy of kT ~ 2.2 MeV, T ~ 1010 °K. It also justifies the recombination of primordial hydrogen at a much lower temperature than the ionization energy kT ~ 13.6 eV. As for T-violation, one way to observe it is to detect an electric-dipole moment of an elementary particle or bound system, because T- and also P-invariance require that such a dipole moment must vanish (see Problem 6.5). Since P is known to be violated at the level of weak interactions, a nonvanishing electric dipole moment at this level then could indicate T-
112
Discrete Symmetries
violation. To date the neutron electric dipole moment vanishes to a high degree of measured accuracy, of order 10" 2A e-cm. As with CP violation, the K°-K° system is also a place to observe T-violation. A recent experiment in fact detects a T-violating amplitude of order 10" 3 relative to the weak interaction [see Kleinknecht (1976)].
6.D CPT Invariance The weakest discrete symmetry is the product CPT, which turns incoming particles into outgoing antiparticles while flipping helicities. From a theoretical standpoint, this is a "natural" symmetry because it does not appear possible to construct a "reasonable" interaction which violates CPT. It is therefore worthwhile to give special attention to this operation, first discussing the CPT theorem and then its direct consequences. CPT Theorem. In the presence of an electromagnetic field, the spinless and spin-1 Klein-Gordon and spin4 Dirac equations are left invariant under the antiunitary operation sd = CP2T. This is a direct observation from Sections 6.A-C with ^ ( x ) - > -?™{-x) and A*A-x)=
-(gw)(gw)Afl(x)=
-A^x),
(6.78a)
4>A~X) = VCPT4>{X\
(6.78b)
4>n.A -x)=- r\CPT >„(*),
(6.78c)
whereas for the Dirac particles CPT = iy5 ia2 7o{ — ^i)yo = ! 7s' s o that ^A-^^Vcpriys^x).
(6.78d)
For a general hamiltonian H(t), the CPT transformation requires stf-1H(t)*/
= H*P(-t).
(6.79)
To link HCP with HT, consider relativistically covariant hamiltonian densities Jf(x) such that H{t) = J d3x Jf(x). Then JfPT(x) -> Jf*(-x), whereas the charge-conjugate j f c (x) is built up, as in the case of external electromagnetic fields, in terms of scalar products such as f™c(x)A£(x). The complexconjugation operation due to T in (6.79) is canceled by the complex conjugation inherent in the charge-conjugate wave functions, as in (6.78), with the additional constraint that no extra factors of i occur in H; i.e., the density J^(x) is real and the operator H is hermitian. As for the spindependent part of the CPT operation, from (6.78) we see that boson wave functions are even for even-rank tensors (spin 0,2, ...) but odd for odd-rank tensors (spin 1, 3,...), as are the derivatives d-d, d^, d^dv, etc. since dp-* —3^ under CPT. Further, the spinor bilinear covariants transform in the same fashion, with the inverse of the CPT operator (ry5)~l = iy5 leading to (see Problem 6.6)
iysih y„, v 'VMTS. yshs = (l, -y„, *„„ -»%y 5 .ys\
(6.80)
CPT Invariance 113 so that again even-rank tensors are even and odd-rank tensors are odd under CPT. Note that a sign has been removed from ij/^ = — ij/iys, as was the case for the operation C in (6.25) and (6.26). [Strictly speaking, this version of the CPT theorem should be proven in a field-theory context to manifest the connection between spin and statistics and also to employ lagrangian densities J£?(X) instead of J^(x), because there can be a problem with Lorentz invariance in the latter case.] Thus for a Lorentz-invariant interaction hamiltonian density, for which an even number of tensor indices (of wave functions, derivatives, and ymatrices) must be contracted to form a scalar, we have Jf *(*) = JV{x) provided Jf (x) is local and hermitian, with phases chosen for each particle such that 1CPT=1(6-81) Then the CPT theorem follows (Liiders 1954, Pauli 1955, Lee et al. 1957): s/~1jr(x\s/ = jf(-x).
(6.82)
The S-matrix version of the CPT theorem following from (6.82) and (7.48) is < / IS |i> = < / | ^ ' S V |i> = r,CPT(CPT i\S\CPTf),
(6.83)
with incoming particles becoming outgoing antiparticles with flipped helicities, where we have used the antiunitary property of s/ as in (6.75) and t]CPT is the product of the CPT phases of all the particles in the interaction. In terms of M-functions similar to (6.34) for the representative process i + l - » i + l , the CPT relation (6.83) with p->p, A->-A, and (see Problem 6.6) «<-»(p)=(-r*»y 5 «<»(p),
e*<-«(p)= - ( - r ^ P ) .
(6-84)
(valid for p = pe 0 ) leads to the CPT condition for M-functions, M,M i) = (-YricrriVsMjCi, Cf)iy5.
(6.85)
Here J is the total number of vector indices in the reaction (two in this case). Under what assumptions is (6.85) valid? First, consider "crossing" all particles from the initial to the final state and vice versa, while preserving momentum, energy, and charge conservation as in Pi + Pi = P3 + P* -* P2 = (-Pi) + Ps + P4 etc.,
(6.86a)
ex -I- e2 = e3 + e 4 ->• e2 = ( — *i) + e3 + e 4 etc.
(6.86b)
Interpreting — px and —er as corresponding to a negative-energy particle traveling backward in time, or alternatively as an outgoing antiparticle (as opposed to an incoming particle), it is clear the kinematics of "total crossing" is the same as the CPT transformation. If the Lorentz-covariant dynamical amplitudes in MMV are in some sense "analytic" in the particle momenta— roughly linked with a local lagrangian density through "causality" (to be
114 Discrete Symmetries
discussed in detail in Chapter 15)—then they do in fact obey the dynamical total crossing relation Mjf,
i; {p}) = MjCi, Cf; {-p}),
(6.87)
where { — p} refers to all four momenta reversed. We shall see in Part III of this book that individual covariant Feynman graphs indeed obey (6.87). The general proof, however, is related to the analytic structure of "complex" Lorentz transformations (Jost 1957). Then since the mechanics of (6.85) converts y^ to — y^, etc., according to (6.80), and p^ to — ( — p j , every vector index in (6.85) has a sign change associated with it which cancels the ( — )J factor while transforming the CPT transformation law (6.85) to the total crossing relation (6.87). Thus if riCpT = 1, then the CPT theorem, proved originally for a local, hermitian hamiltonian density, is also linked with total crossing in momentum space applied to a unitary S-matrix. We shall return to the concept of crossing in later chapters. Consequences of the CPT Theorem. Considering the few and weak assumptions employed in the CPT theorem, its implications are significant indeed: i. The existence of antiparticles follows from CPT invariance even if charge-conjugation invariance is violated in nature. It is not necessary to invoke a "Dirac sea"—a ground-state vacuum filled with antiparticles—in order to justify the existence of antiparticles interacting above the sea. ii. The CPT theorem provides the ultimate justification as in (6.78) for the Stiickelberg-Feynman picture of antiparticles as negative-energy particles traveling backward in space-time. iii. The mass of an antiparticle is identical to that of the particle in question. For stable particles this is a consequence of the mass being an eigenvalue (in the rest frame) of the simultaneously commuting operators H and UCPT. A similar proof for unstable particles follows from the CPT invariance of the S-matrix. iv. The lifetime of an unstable antiparticle is identical to that of the corresponding particle, regardless of the dynamical interaction causing the decay. This follows directly from the absolute square of (6.83), accordingly proportional to the rates for the particle and antiparticle decays. Likewise, the cross sections for particle and antiparticle reactions are equal [see Lee et al. (1957)]. v. If one of the three discrete symmetries is violated, then at least one of the other two also must be. Moreover, the otherwise unknown timereversal phase for a single particle is, from (6.81), rir =
tip
(6-88)
which in turn determines the reality of scattering amplitudes (see Section 15.H). Indeed, the discrete symmetry phases play a major role in the CPT theorem and can, at times, be rather subtle [see, e.g., Feinberg and Weinberg (1959)].
CPT Invariance
115
CPT Invariance in Nature. Since it is observed that C, P, and T are separately valid for strong and electromagnetic interactions, the product CPT is also invariant for these interactions. For weak interactions it is believed that CP and T are separately conserved, so again the CPT symmetry is valid. For the K°-K° system, the small CP- and T-violating amplitude, ~ 10 - 3 relative to the weak interaction, appears to be consistent with CPT invariance [see Kleinknecht (1976)]. Thus CPT invariance appears always to be valid in nature, corresponding to our theoretical expectations as stated above. General references on the discrete symmetries are: Wigner (1959), Wick (1959), Hamilton (1959), Roman (1960), Marshak and Sudarshan (1961), Schweber (1961), Streater and Wightman (1964), Bjorken and Drell (1964), Sakurai (1964, 1967), Muirhead (1965), Gasiorowicz (1966), Pilkuhn (1967), Bernstein (1968), Rowe and Squires (1969), Bohr and Mottelson (1969), Martin and Spearman (1970), Fonda and Ghirardi (1970), and Emmerson (1972).
This page is intentionally left blank
PART II
SCATTERING THEORY
This page is intentionally left blank
CHAPTER 7
Formal Theory of Scattering
As transformation theory is to the kinematics, so scattering theory is to the dynamics of much of advanced quantum theory, for both nonrelativistic and relativistic particles. In this chapter we describe the formal theory of scattering in the three time-development pictures of Heisenberg, Schrodinger, and Dirac. All three formulations culminate in the central operator of scattering theory—the S-matrix, from which transition probabilities, decay rates, and cross sections are computed. While this approach is at times quite formal, it is worth elaborating in detail in order that we may appreciate the underlying structure of Feynman diagrams to be developed in later chapters.
7.A Formulation of the Scattering Problem We have already noted the utility of the S-matrix, as opposed to the hamiltonian, for expressing general selection rules (Chapter 1) and Lorentz transformation laws (Chapter 3), and for stating the discrete-symmetry invariance principles (Chapter 6). In particular, the S-matrix accommodates states, such as scattering states, which are not discrete eigenstates of H. At this point, therefore, it is natural to reformulate the complete quantummechanical problem in scattering language, valid for bound-state as well as scattering configurations. Bound-state wave functions and probabilities are then replaced by transition probability amplitudes—S-matrix elements— and transition probabilities, convertible either to a physical decay rate or to a cross section. Simple singularities of the S-matrix correspond to bound states. 119
120 Formal Theory of Scattering Validity of Scattering Theory. While knowledge of the S-matrix [introduced by Heisenberg (1943)] is certainly the central goal of scattering theory, we shall not tackle the most complicated scattering configurations head on. Instead it will pay to make a step-by-step transition from one-body nonrelativistic to many-body relativistic quantum-scattering diagrams. First we shall consider time-independent potential scattering, corresponding to the time-independent hamiltonian H = H0 + V for a single particle of mass m scattered off of a fixed potential V. The energy of the scattered particle is conserved in this case, but three-momentum is transferred to the potential "target". Such a formulation will be valid for any type of particle, with H0 = p2/2w for nonrelativistic spinless particles, H0 = a • p + pm for Dirac spin4 particles, and H2, = p 2 + m2 for Klein-Gordon spin-0 particles all transformed into scattering Green's functions. The latter act as kernels of scattering integral equations which take the place of the bound-state wave equations. The second level of sophistication is time-dependent potential scattering with V -» V(t). Energy will again be conserved if the time dependence is of the form V(t)cce+ia" and we interpret co as the (incoming or outgoing) energy of an additional particle, such as a photon, partaking in the scattering with A£ = Ef — E{ = +ct>. It is also possible to extend nonrelativistic potential scattering to two-body scattering, since the former is equivalent to the latter with m replaced by the reduced mass ml m2/(m1 + m2). Then the total energy and three-momentum of the two particles must be conserved. Since there is no simple CM (center of mass) transformation for relativistic particles (mass is not conserved), the final step of complexity will be twoparticle or even multiparticle scattering for relativistic particles, which we postpone until Part III. The CM system then refers to "center of momentum" or "barycentric" frame, one of the many possible reference frames from which to view relativistic scattering processes. Scattering Wave Functions. In scattering theory, particle wave functions are of particular interest far away from the scattering target, i.e., at r -* ± oo or t -> + oo. The scattered wave packets then can be replaced by simple plane waves, designating the energy and momentum of the particles in these asymptotic limits. The dynamical information about the scattering is embodied in the S-matrix, "full" Green's function, or scattering T-matrix, the latter being a generalization of V itself. We shall normalize these scattering plane waves in a manner suitable to the problem in question. For nonrelativistic scattering it will usually prove convenient to choose the noncovariant box-normalized wave functions, while for relativistic scattering we shall usually use covariant boxnormalized wave functions. In either case the box volume factors V (not to be confused with the matrix elements of a potential Vfi) must cancel in any physically measurable quantity; later in Part III we shall set V = 1 (as well as h = c = 1) in order to suppress as many irrelevant variables as possible.
Time-Independent Potential Scattering
121
Time-Development Pictures. Recall from Chapter 1 that there are three ways of describing the time development of quantum-mechanical states and operators: the Schrodinger, Heisenberg, and (Dirac) interaction pictures. Timeindependent scattering is a special case of the Heisenberg picture with operators fixed, while time-dependent scattering theory can be formulated in either the Schrodinger picture using Feynman's notion of "propagators", or in the Dirac-interaction picture with the asymptotic scattering states driven by HQ. As we shall see in this chapter, an essential difference between the three scattering pictures is the way causal boundary conditions are incorporated into each formalism. On the other hand, the physical content of the three scattering pictures is embodied in the same S-matrix, whose matrix elements are identical in all three cases.
7.B Time-Independent Potential Scattering Asymptotic Wave Functions. For a coordinate-space potential V(r) of shorter range than Coulomb [i.e., rV(r) -* 0 for r -*• oo], it is always possible to decompose nonrelativistic wave functions satisfying [E — V(r) + V2/2m]i/'(r) = 0 into incident and scattering components as ^(r) T ^>/'inc(r) + 'sca„(r).
(7.1)
As pictured in Figure 7.1(a), such a decomposition makes physical sense for an "in" solution (setting the volume V = 1 here) * ( + ) ( 0 - ^ z > e * " +/ + (P*, V)eipr/r,
(7.2a)
corresponding to an incoming plane wave and an outgoing spherical wave of momentum p. Alternatively, as shown in Figure 7.1(b), an outgoing plane wave and incoming spherical wave "out" solution obeys
'/''"V)—>** " +/-(/*, pK"7r.
(7.2b)
While the "out" solutions (7.2b) seem somewhat unphysical, they are on a par with "in" solutions, since, as we shall see, the mathematical notion of completeness treats i// +) and \j/(~) on an equal footing. Nevertheless, it is customary to concentrate o n / + instead of/", defining/ p jP = f(p, 6) s f + (pf, p) as the "scattering amplitude" for the process. Pr
^V (a)
Pi",
/SSc (b]
Figure 7.1 Pictorial definition of time-independent "in" and "out" states.
122
Formal Theory of Scattering
Integral Equation in Hilbert Space. The foregoing decomposition can be abstracted to Hilbert space by writing the general scattering solutions of the Schrodinger equation as H|^!> = £ | ^ > ,
(7.3)
where + respectively indicate "in" and "out" boundary conditions. Writing H = H0 + V, (7.3) can be expressed as an inhomogeneous equation for V^O and a homogeneous equation for V = 0, with \I/J±}->- |
(7.4a)
(E - H0)\cpE} = 0.
(7.4b)
These differential equations can then be combined into a formal Fredholm integral equation of the second kind,
I "Al > = I
(7.5)
Operation on the left in (7.5) by E — H0 and using (7.4b) again reproduces (7.4a). Boundary-Condition Prescription. To decide upon the relative weight of the q> term and the \p term in (7.5) and to incorporate the physical content of Figure 7.1, we return to the coordinate-space representation and assume the potential is diagonal,
(7.6)
This form corresponds to the general decomposition (7.1) with the integral related to i/fscatt for r -> oo. To proceed further, we specialize to nonrelativistic potential theory with E = p2/2m and \q>E) = |p>. Then, since H0 is diagonal in momentum space, so is (E — H0)~l, which is called the freeparticle Green's function G0(E):
(7.8)
where p2 = ImE. The integral in (7.8) obviously is undefined at k2 = p 2 , and we must impose the boundary condition so as to circumvent this singularity. It turns out that the boundary conditions related to the configurations of Figure 7.1 and (7.2) correspond to the "ie prescription", G0(E)^> GQ (£) = (E — H0± ie)~ l , where e is a small positive number. To verify this
Time-Independent Potential Scattering
123
connection in detail, write GQ (r, r'; E) = 2m(2n) 3I±, where J* is the threedimensional momentum-space integral with d3k = k2 dk dQk, dkk2 J
f d a «*•<—'» 2
(7.9a)
p -k ±ieiailke The solid-angle integral in (7.9a) can be evaluated via e
2
0
dQk = In sin 0 d9 = -27t d(cos 0), where 0 is defined relative to r — r', i.e., k • (r — r') = k | r — r' | cos 9. Thus,
JdO» e * <'-') =
2
"
[g*l'-''l - e"*!'-''!],
(7.9b)
so that (7.9a) becomes J 1 = 27t(i |r — r ' | ) " 1 J ± , with J* - !
kdk p - fc2 ± i£ 2
g»|r-r'|
(7.9c)
where we have changed variables k ->• —kin the second radial integral over e-iii|r-r'i j n (7 9b). Finally, to evaluate (7.9c) we extend the real variable k = | k | into the complex fe-plane with denominator p 2 - k2 ± ie = - [k - Vp 2 ± «e][fc + ^ P 2 ± ' £ ] * - [fc - (p ± ie)][fc + (p ± ie)],
(7.9d)
since 8 is small. Observing that gilc|r-r'| __
gi|r-r'|Reke-|r-r'|Imk
(7.9e)
it is clear that both the J + and J~ integrals (7.9c) are analytic in the upper half fc-plane (Im k > 0), except for the simple poles at k = p + ie and k = —p + ie, respectively, as shown in Figure 7.2. Since the contours at infinity in Figure 7.2 vanish, we may apply the Cauchy residue theorem to J±, obtaining 2ni Res J±(k = ±p + is) = —ine ± i p | r - r ' |
(7.9f)
p-i«
•p-ie
(b]
(a 2
Figure 7.2 Position of the poles of Gp(£;
124 Formal Theory of Scattering Collecting together the relevant definitions and conclusions from (7.9), we obtain the scattering "ie" Green's functions Go±(r r ;£)=
' '
-4^7|e±ipk"'
(7 10)
-
Finally, we evaluate (7.10) in the asymptotic limit r -* oo using | r — r' | = r — r' • r + • • • and defining p' = pf so that GHf,^E)-^-^e±l"e^-':
(7.11)
Substituting (7.11) into the asymptotic form of (7.6), we have ^ ( r ) - ^ * * - ~ ^ e ± i p r | d V e** -Vir'Wiir').
(7.12)
Comparing (7.12) with the "in" and "out" solutions (7.2), we see that the ie prescription indeed corresponds to the required outgoing and incoming spherical waves, respectively, along with the identification
f+-(p\ p) = - ^ I
d V e +ip
' ' • r 'no'f («•')•
(7.i3)
One may also obtain (7.13) in another manner (see Problem 7.1). /-Matrix and Full Green's Function. From the structure of (7.13) with e~v ' r replaced by
+
(P',P)=
(7.14a)
where
(7.14b)
T\p} = V\9+y,
(7.15a)
where for plane waves or in Hilbert space 7 » = F|«A
+
>.
(7.15b)
Just as the T-matrix describing the "full" interaction is a generalization of the "bare" potential V, so the bare (free) Green's function Gj (£) =(E-H0±
ie)~*
(7.16a)
has a full-Green's-function companion G±(E)=(E-H±ie)-i,
(7.16b)
Time-Independent Potential Scattering 125
where H0(or H) is the bare (or full) hamiltonian. Put another way, the differential operators D0(E) = E — H0 and D(E) = E — H imply D 0 (£)G 0 + (£)=1,
D(£)G + (£)=1,
(7.17)
or in coordinate space with D diagonal,
= 83(T-r').
(7.19)
Lippmann-Schwinger Integral Equations. The Hilbert-space integral equation (7.5) with the prescribed ie boundary condition is an example of a Lippmann-Schwinger integral equation. The name is now given to the integral equations for the scattering wave functions, for the T-matrix, and also for the Green's function. In particular, (7.5) plus the ie prescription is I
(7.20a)
Multiplying (7.20a) on the left by V and applying (7.15b), we may drop the |cp£> state on the right to obtain the operator (Lippmann-Schwinger) equation T(£) = V + VG£ (£)T(£). (7.20b) In this equation the energy variable £ is a parameter, which may or may not be equal to the energy of the plane-wave state matrix elements p'2/2m = p2/2m upon which T(£) operates. If 2m£ = p 2 = p' 2 , we say the T-matrix is "on the energy shell"; if not, T(£) is off shell. In any case, the intermediate momentum states which can be inserted between GQ (£) and T(£) in the second term of (7.20b) force this T(£) term to be off shell. But more about this point later. To continue, the Lippmann-Schwinger equation for the Green's function is obtained by noting the operator identity A'1 — B"1 = B~ *(B — A)A~l and letting A = G± and B = GQ and vice versa. This leads to G± (£) = Gi (£) + Gj {E)VG± (£)
(7.20c)
= Gj (£) + G±{E)VG^ (£).
(7.20c')
Adjoint-type relations for 11/^> and T similar to (7.20c') can also be derived (see Problem 7.2). These three types of Lippmann-Schwinger equations (7.20) are all of the form full = bare + bare x full. So if we are to "solve" (7.20), at least formally, we must remove the "full" parts on the right-hand sides of these equations. To this end, we note the simple identities (Problem 7.2): V | n > = T(£) | cpE}, G± (E)V \^>
±
= G (E)V | cpE>,
<«A£-1 V = (cpE I T(£),
(7.21a) (7.21b)
126
Formal Theory of Scattering
G + (E)V=G£(E)T{E),
VG~(E)=T(E)Go(E).
(7.21c)
Applying (7.21) to (7.20), we obtain \^>=\
+ G±(E)V\
T(E)=V+VG+(E)V, G+(E) = G0+(£) + G0+(£)T(£)G0+(£).
(7.22a) (7.22b) (7.22c)
Of course, we have not really succeeded in solving the Lippmann-Schwinger equations, because the right-hand sides of (7.22) still contain all the dynamics in the guise of another "full" scattering quantity—G, G, or T, respectively. Nevertheless an important consequence of (7.22c) is T(E) = G0+ -*(£) + G0+ - 1 (£)G + (£)G 0 + -*(£),
(7.23)
called the reduction formula. We shall return to it in Section 7.E and again briefly in Section 15.H.
7.C Time-Dependent Scattering in the Schrodinger Picture Feynman (1949) formulated time-dependent scattering in the Schrodinger picture in a manner relevant for relativistic scattering as well as for nonrelativistic potential theory. The method is based on time-dependent Green's functions and "causal propagators". Time-Dependent Green's Functions. The bare and full time-dependent wave functions and Green's functions must satisfy the Schrodinger equations (id, - H 0 ) |
(7.24a) (7.24b)
Aside from boundary conditions, the time-dependent integral equations analogous to the Lippmann-Schwinger equations (7.20a, b) are then | = I (?)> + J G0(f - t)V(t) I # ) > dt, G(t' - t) = G0(f -t) + j G0(t' - t")V(t")G(t" - t) dt".
(7.25a) (7.25b)
Causality. It is natural to impose a boundary condition which manifests the physical requirements of causality; viz G+(t' -t) = G$(t' - t ) = 0 for t'
(7.26)
This means that the system cannot respond until the signal (beam particle) arrives at the target (scatterer). A formal representation of these causal
Time-Dependent Scattering in the Schrodinger Picture 127 Green's functions for time-independent H0 and H is G + (f -t)=
-i9(t' - t)e-m,'-'\
Gj(f - r) = -i0(t' - tjg-^oc-o,
(7.27a) (7.27b)
where the step function 8(t' — t) satisfies 6(f - t) = l! t*', > f ' v ; |0 i f t ' < r , dt8(t) = S(t).
(7.28a) v ' (7.28b)
Given (7.28), it is straightforward to show that (7.27) obeys (7.24b). To see how the causal G+(t' — t) is related to the time-independent, "in" Green's function G+(E), consider the identity 0(t) w
1 i-00 date'1" _L f ^ ^ , 2JTI J_ x
<x + ie
(7.29) v
'
easily verified by contour integration similar to (7.9) (see Problem 7.3). Then substituting (7.29) into (7.27a) and changing the integration variable a to £ = a + H, we obtain G+(t'-t)
= ^- C
d£e-' £ ( r -'»G + (£),
(7.30)
J
2n -oo
where G + (£) = (£ — / / + ie) _1 . A similar relation holds for GQ . Clearly G+(t) is the Fourier transform of G + (£), and therefore the causal boundary condition in time, (7.26), corresponds to the outgoing-spherical-wave ie prescription in coordinate space. Propagators. To handle time-dependent Green's functions for the more interesting case of time-dependent interactions [V = V(t)], Feynman (1949) employs Huygens's principle. This requires each point on a propagating wave front to be a new source of propagation. Thus we write for t' > t {t, t),
(7.31a) (7.31b)
where
128 Formal Theory of Scattering Lippmann-Schwinger equation (7.20c) leads to (7.25b). For time-dependent interactions the causal propagator satisfies the obvious generalization of (7.25b): iG+{x'; x) = iG0+(x'; x) + J d V iG0+(x'; x")[-iV(x")]iG+(x"; x). (7.32)
To demonstrate that (7.31) is consistent with (7.32), we insure that the condition t' > t is satisfied by invoking the plane-wave limit as t -> — oo (also valid as t -• + oo),
t),
(7.33)
which simply states that the particle is initially free before scattering at some finite time t > - o o . Substituting (7.32) and (7.33) into (7.31a) for r-> - o o gives (Problem 7.3) •Mr', f)-rr^\
d3r iG$(f, f; r, t)(p(r, t)
+ J d3r dVG 0 + (x'; x")V(x")iG+(x"; r, t)q>(r, t).
(7.34)
Then using the fact that I'GQ propagates q> via (7.31b) while iG* propagates cp at t = — oo and converts it to the full wave function \\i via (7.31a), we see that (7.34) becomes the expected time-dependent analog of the LippmannSchwinger equation (7.25a), i^(x') =
(7.35)
Because the dynamical equations (7.32) and (7.35) have space and time on an equal footing [x = (r, t)], these results apply both to the nonrelativistic Schrodinger theory and to relativistic wave mechanics. Explicit Form of Propagators. First consider the free-particle momentumspace Green's function for a nonrelativistic particle with H0 = p2/2m, namely,
-"<''-')(M^f-('"^f)
(7.36a)
^
as can be seen by completing the square of the momentum-space Gaussian implicit in (7.36a). The causal propagators for free relativistic particles must be linked with those for antiparticles—Feynman's insight. To see this, consider the free-
Time-Dependent Scattering in the Schrodinger Picture 129 particle causal Green's function for the spin-0 Klein-Gordon equation GQ(X'; X) -> AF(x' — x) (F for Feynman), satisfying ( • ' + m2)AF(x' -x)=
-<54(x' - x).
(7.37)
Its Fourier representation with the causal ie boundary condition included is (Problem 7.4)
M*) = j>P 2!m'2 + j.£,
(7.38)
now with space-time variables x = (r, t) and p = (p, p0) on an equal footing. The implications of Feynman's ie prescription can be worked out in the complex pp-plane, with poles of (7.38) at + p0 = E — ie where E = y/p2 + m2, as shown in Figure 7.3. The situation is similar to the contour integration of (7.9) and Figure 7.2. But now for t > 0 the contour is closed in the lower half p0-plane as in Figure 7.3(a), while for f < 0 the contour is closed in the upper half p0-plane as in Figure 7.3(b). This choice guarantees that the exponential exp(t Im p0) damps the integral (7.38) to zero along the infinite contours in Figure 7.3. Each closed contour encircles a simple pole, so (7.38) becomes, using the residue theorem [P2 w + ie = (p0 - E + je)(p0 + E- ie)], iAF(x) =
ilni
+
=1
a3P
-&y
iEte*
d3P \~9(-t)eiE'e^' •IE
(7.39a)
e(-ty%
(7.39b)
where the minus sign in front of the first term of (7.39a) is because the contour of Figure 7.3(a) is clockwise, and the factor — 2£ in the second term comes from the residue of the pole encircled in Figure 7.3(b). The conversion of (7.39a) to (7.39b) follows from the replacement p-> — p in the second
(b) Figure 7.3 Position of the forward and backward propagating poles in AF(p2) in the 5 complex p0-plane for E = ^yjr+m .
130 Formal Theory of Scattering
term. This latter decomposition (7.39b) says that iAF propagates
d3x
' M * ' - x)id0(p + (x),
-0(t - r')>-(*') = f d3x iAF(x' - x)id0(j)_(x).
(7.40a) (7.40b)
A similar situation occurs for the spin4 free-particle causal propagator iSF(x'; x) — iSF(x' — x). As a free-particle Green's function it satisfies (#' -m)SF(x';x)
= 54(x' - x),
(7.41)
The p0 contours for (7.42) are the same as shown in Figure 7.3, leading to (Problem 7.4) iSpW =
1S
[d{t){P
+ m ) e ~ i p x + 0(-t)(-p
+ m)eip•X],
(7.43)
which propagates the positive-energy bispinors ip+ forward in time and the negative-energy bispinors i/^_ backward in time, as 0(t' - t)iA + (x') = J d3x iSF(x' - x)y0ty+ (x),
(7.44a)
-6(t - t')iA_(x') = J d3x iSF{x' - x)y0il/_(x).
(7.44b)
Extension of the propagator formalism to massless photons presents a slight complication, to which we shall return in Chapter 10. Suffice it now to say that for all types of relativistic particles scattering via covariant interactions, causal propagation is the key notion which links positive- and negative-energy states together in the manner suggested in earlier chapters. The CPT interpretation of antiparticles as backward-propagating negativeenergy states then allows us to treat particle-antiparticle relativistic scattering on a par with nonrelativistic scattering.
7.D Time-Dependent Scattering in the Interaction Picture Recall that in the Dirac-interaction picture, the hamiltonian is split up as H = Ho + V with H0 driving the operators and V driving the states. We have just seen how causality manifests itself in the other two pictures via the 16 prescription and the step function 9(t' — t). Now we investigate how causality enters the scattering problem in the interaction picture.
Time-Dependent Scattering in the Interaction Picture
131
t'
»
P
T2
Figure 7.4 Region of integration denned by the time-ordering operation for the timetranslation operator and S-matrix.
Time-Ordering Operation. Returning to the iterated time-ordered infinite series for the time-translation operator (1.66), we consider in detail the second-order term with time-ordered integration variables as designated by the shaded area in Figure 7.4. Then we may write f dtl fl dt2 VfcJVMz) = ^ j ' dt, {' dt2 nvfcjVfci)},
(7.45)
where T is the Dyson time-ordering operator defined by
W i M a = VfrMtiMt! - t2) + VfaWitMi - h). (7.46) For the general nth-order term in (1.66), the factor of \ in (7.45) becomes \jn!, so that the infinite series can be exponentiated to the form U,(t', t) = T exp j - i |
dt" Vj{t") j.
(7.47)
In the Schrodinger picture, we have seen that space and time are on an equal footing and that this leads naturally to covariant Feynman propagators. The analogous statement in the interaction picture follows from a condition on the covariant hamiltonian density used to build up the interaction Vj(t) = J d3x Jif^x). The S-operator in the interaction picture then appears manifestly Lorentz invariant, S, = 17,(00, - oo) = T exp j - i f d*x JT,(X) I
(7.48)
except for the nonmanifestly covariant 6(t' — i) and 6(t — t') structure of the time-ordering operator T in (7.46). On the other hand we know that all information is propagated at or at less than the speed of light, which means that causal events must be separated by timelike distances, (At)2 — (Ax)2 > 0. Then two causal operators such as Jif7(xi) and Jf 7(x2) separated by spacelike distances are simultaneously measurable, i.e., [Jf ;(*!), Jf ,(x2)] = 0 for (xj - x 2 ) 2 < 0.
(7.49)
132
Formal Theory of Scattering
The condition (7.49) compensates for the non-Lorentz-invariant nature of the time-ordered product in (7.48), with the result that S ; is indeed a Lorentz invariant. We shall return to this condition (7.49) in Chapter 10. Adiabatic Switching. To see how the ie prescription can be used in the interaction picture, we first note that the plane-wave limit (7.33) for Schrodinger states far from the scattering as r-> + oo is | iA£(t)>s -»• e~iEt \
i(t) -»• * U 0 = eiHo
(7.51)
With £ small and positive, VliC(t) -> 0 for t -* ± oo as required. In a similar fashion, Ufa', t) -> UlE{?, t) by the substitution of (7.51) in (7.47). Dressing Relation. Now apply the damped t//i£(t', t) to a full energy eigenstate at t = — oo, transforming it to t = 0: l<M0)>7= l//..(0, -oo)|<M-oo)> 7 . Iterating l/ /£ (0, — oo) to first order for V time independent, ,o C//>£(0, - oo) = 1 - i dt" ea"eiHo'"Ve-iHo'" + •••, ^
(7.52)
(7.53)
- O O
(7.52) gives, using (7.50), dt" e«"eiHo'"Ve-iEt" |
|lM0)>/ = WE> - i f J
(7.54a)
-00
= \cpEy + (E-H0
+ k)-iV\(pE}
+ ---
- |
(7.54b) (7.54c)
where (7.54c) follows from (7.54b) by iterating (7.53) to all orders in V. Since (7.54c) is just the Lippmann-Schwinger equation for "in" states (outgoing spherical waves) in the time-independent formalism, we conclude that !#>)>/= |* + >,
(7-55)
|* + > = I/,..(0, -oo)|
(7-56)
or from (7.52) valid for time-independent states of any energy. In a similar fashion one can show that \iJ,-} = ULt(0,cc)\(py,
<*" | = <
(7-57)
The S-Matrix
133
The equations (7.56) and (7.57) are called "dressing relations" for a timedependent V because they represent the "clothing" of a bare state (at t = — oo), converting it to a full state at t — 0. Mailer Operators. Define the Moller operators Q ± = l/,,£(0, +oo),
(7.58)
so that the dressing relations become (see Problem 7.5) |lA±> = n ± |>.
(7.59)
Once the state is dressed from t — — oo to t = 0, it will undergo the scattering interaction from t = 0 to t = ts, with "in" states 11/ + } converted to "out" states \ij/~} because the damping factor e~£|(| changes sign when t > 0. Finally, from t = ts to t = oo the "out" state is undressed, becoming another plane-wave state. Putting all this together, the complete scattering process is I <7>(°°)>z = tf,.«(oo, - oo) I q>(- oo)>„ S, = l//f,(oo, - oo) = t//>E(oo, 0)[//>£(0, - oo) = fil Q + .
(7.60) (7.61)
Clearly the S-operator is the collective sum of the dressing, scattering, and undressing processes. It is therefore time to take a detailed look at the S-matrix itself.
7.E The S-Matrix From the foregoing discussion of the various scattering pictures, we found a clue to their underlying unity—the ie prescription. It supplies the outgoing spherical-wave boundary condition in the time-independent formalism, is needed to build a causal propagator in the time-dependent Schrodinger picture, and is used to damp adiabatically the asymptotic interaction in the interaction picture. The complete unity of the theories is embodied in the S-matrix elements, as we now demonstrate. Overlap Matrix Elements. The merging of the three pictures (Heisenberg, Schrodinger, and interaction) occurs for the overlap S-matrix elements S^=<^-|^+>,
(7.62a)
Ss/f = lim<^ / (t)| S ,
(7.62b)
f-»ao
Stf- = lim<<MOWt)>/-
(7.62c)
t->ao
These overlap matrix elements all measure the probability amplitude for the transition i-*f and therefore must be equal: S"fr = Ssfr = Stf •.
(7-63)
134 Formal Theory of Scattering To prove this contention, we make the plane-wave-limit identification \
(7.64a)
The latter is related to the Heisenberg picture for time-independent interactions via the dressing relations and (7.61): Syr- = <
= < * ; \tf> = S"f:p,
(7.64b)
with the Heisenberg-picture definition valid even for time-dependent interactions. To go from the interaction picture to the Schrodinger picture we note that | ^,-(0)/ -» e'1*0' | i/f;(t)>s as r ->• oo for states that are adiabatically damped. Then we have
S£- = lim <M0kH°'e_'7Vl'iW>/ = <MQ°)I^(00)>S = ^fr->oo
(7.64c) Given the equivalence of the S-matrix elements, we can relate Sfi to the scattering operators characteristic of the three pictures—Tfi, G+, and U, in the H.P., S.P., and I.P., respectively. In the latter case, (7.48) is just such a relation. Relation to Tfl. Since the S-matrix embodies all possible scattering configurations, it must have a "no scattering" part. That is, we may write S = 1 + R,
(7.65)
where the identity operator indicates no scattering and the reaction matrix R denotes only the definite scattering part of S. To see how R and therefore S are related to Tfi, we investigate time-independent potential scattering (a special case of the Heisenberg picture) by subtracting | i / 0 from | ^ + > in (7.22a). The plane-wave part then drops out and we have W > - l ^ > = [G+(Ed - G-(Et)]V\
(7.66a)
Recalling the Dirac principal-value theorem (Problem 7.6), (x + ie)~ x = P/x + i7t<5(x) (a proof follows from a contour integration similar to the two already encountered), we see that the principal parts of the Green's functions in (7.66a) cancel, leaving G + (£,) -G~ (£,) = - 2mS(Ei - H).
(7.66b)
Now substitute (7.66b) into (7.66a), insert a complete set of "out" states X" I ^n X'A.T I = 1 between d(Ei — H) and V, and use H | \\i~ > = En | ifr'> as S(Et - H)V = X 8(Ei - En) | ^ X ^ |
V.
(7.66c)
n
Next multiply (7.66a) on the left by (IAEJ and use the orthogonality of scattering states (see Problem 7.6):
<*7 I "A.7 > = <«A/+ | *! > = <
(7-66d)
The S-Matrix 135 Collecting together (7.66a)-(7.66d), we have
Finally, applying the definition of the T-matrix, (7.21a), to (7.66e), we know that < f e I V|
(7.66f)
Then (7.66e) and (7.66f) give the desired result Rfi = Sfi -5fi=-
id{Ef - £,.)7},.,
(7.67)
where we delete the energy dependence of Tfi in (7.67) because the energy delta function multiplying it demands that Ef~E{. Put another way, while the T-matrix can be "off the energy shell", Tfi(E) for E not necessarily Ef or £ ; , the S-matrix must be on the energy shell by (7.67). In any case, for a real scattering process, i =fcf and 5fi = 0, so the S-matrix element in potential theory takes the form Sfi = —id(Efi)Tfi. For general time-dependent interactions, a form similar to (7.67) will also be valid, but with the energy-conserving delta function accounting for every source of energy (cf. Section 9.B). Even without a fixed potential, two-body scattering amplitudes will have a form like (7.67), but containing the energyarid momentum-conserving delta function 8(Efi)83(Pfi) — S4(Pfi) when the scattering process conserves the total four-momentum, Pf = P{. (The factor of 2n is included with the delta functions for convenience: S -»3 = 2nd.) Relation to G+. According to Feynman's propagator version of scattering in the Schrodinger picture, the overlap definition of Sfi, (7.62b), combined with Huygens's principle (7.31) and the plane-wave limit (7.33), gives Sft = <<M°°)I<M°°)>s =
lim
f{t')\iG+(f - 0|
f-»'oo,r-*' — ao
=
lim
i d3r'd3r (pj(r', t')iG+{r', t';r, ^ , ( r , t).
(7.68)
r - » o o , r - » - QO
While (7.68) directly relates Sfi to the full, causal Green's function (even for time-dependent interactions), it is cumbersome because of the infinite time limits in G + itself. It is possible, however, to transform these infinite time limits to definite integrals over all time by propagating the plane waves over time in an inverse sense (from t= +oo to finite t) of (7.31b). After a slight bit of manipulation this leads to (see Problem 7.6) Sfi = ~i\ dAx' dAx ^(x')G^ T (x')G + (x'; x)G^r(x)(Pi{x),
(7.69)
where the limits of the integrals in (7.69) are over infinite space and time for x and x'. The inverse functions GQ1(X) are just the diagonal differential operators id, + V2/2m, — ( • + m2), and iijl — m for free Schrodinger, KleinGordon, and Dirac particles, respectively. The general form (7.69) is known
136 Formal Theory of Scattering as the reduction formula. It is the time-dependent analog of the timeindependent relation (7.23) and is very useful for proving general theorems of a fairly advanced nature. For our purposes, however, a perturbative approach to (7.68) will suffice for most of the applications that follow in this book. Only in Section 15.H shall we refer to (7.69) again.
7.F Transition Probabilities Assuming that we are given an S-matrix element, Sfi, found in any of the three pictures, it is time to ask, how does one obtain a physically measurable quantity such as a cross section or decay rate? Transition Probability Rate. The absolute square | Sfi | 2 is the probability for a transition from state i -»•/. Consequently, the transition probability rate is
»/« = ^ | < W 0 I M ) > | 2 - - ^ .
(7-70)
where T is the total time over which the scattering takes place. While it is possible to work with time-dependent wave packets to calculate cofi [see e.g. Schweber (1961)], henceforth it will prove more convenient to pass to the plane-wave limit as indicated in (7.70). In this limit for i =^=/(a real scattering transition), the special case of (7.67) leads to \Sfi\2 = \Rfi\2 = \HEfi)\2\Tfi\\
(7.71)
2
where formally, \8(E)\ = S(0)8(E) and .772
dte~iE'
j{E)= lim
(7.72a)
.772
TZ-m=\
dt
=T-
(7-72b)
•'-772
Substituting (7.71) and (7.72b) into (7.70) then gives (also see Problem 7.7) ^i-l-^f
= S(Efi)3f\Tfi\2 = S(Efi)\Tfi\2.
(7.73)
The apparently tricky interchange of infinite limits in (7.72) can be justified in a rigorous fashion [cf. Schiff (1968)]. Nevertheless such a mnemonic as (7.72b) is the cleanest path to the well-known "generalized golden rule" of potential scattering, (7.73). We shall develop similar mnemonics for more complicated scattering configurations in subsequent chapters. Density of States. A quantum transition always involves going from a prepared state i to a range of final states / The number of final states in a given energy and momentum range is known from quantum statistical
Transition Probabilities 137
mechanics to have a six-dimensional phase-space uncertainty (having the dimensions of angular momentum) of h3 (see Problem 7.8), so that for the one-body phase space with pf = p',
4»,-?g.-V«~W
(7.74,
in our units of h = c = 1 (recall dp = dp/In). Here V is the three-dimensional coordinate part of phase space, which we shall eventually set equal to unity, but retain now in order to demonstrate that it cancels out of any physically measurable quantity. Given a number-of-final-states factor such as (7.74), the transition rate into a spread of final states is dTfi = a>fl dNf = I ^ J l dNf = 8(Efi)\ Tfi\2 dNf.
(7.75)
Then for energy-conserving, one-body phase-space processes (one free particle in the final state), with E, = Ef and dNf = (dNf/dEf) dEf, (7.75) gives an integrated transition probability rate of Tfi=\dYfi=\\Tfi\^5{Efi)dEf = 2np(Ef)\Tfi\2,
(7.76)
where p(Ef) = dNf/dEf is called the density-of-states factor. In general all "irrelevant" variables are to be integrated over in (7.76), so for one-body phase space, (7.74), we must integrate out the solid angle, dNf
V4np'2 dp'
Vp'2 m
where we have evaluated the "jacobian" for the case of a nonrelativistic final free particle: E' = p'2/2m, so that dp'/dE' = (dE'/dp')'1 = m/p'. More complicated jacobian factors will be encountered later. Lifetimes. In quantum mechanics anything that can happen does happen— with a certain probability governed by Vfi. For a single free particle in the final state and only a bound initial state (as for spontaneous emission of photons from atoms or nuclei, to be considered in Section 9.C), the final free-particle wave function is proportional to i//J oc 1/y/V. Then | Tfi \2 oc l/V in (7.76), which cancels the V in the single-particle phase space (7.77). Consequently I}, as given by (7.76) is then independent of normalization volume and is a physically measurable rate. The lifetime of such a state is given by *ft = h/rfh
(7.78)
and at this point we restore the proper number of h and c factors so that T has the dimensions of time when I}; is expressed in energy units.
138 Formal Theory of Scattering Cross Sections. For a two-body collision or a single-particle scattering off a fixed potential (equivalent to the two-body case in the nonrelativistic CM frame), the cross section is defined by = Gfi
scattering rate incident
flux
^' '
The scattering rate dTfi is again given by \Sfi\2 dNf/T, as in (7.75) for potential scattering, and Tfi is the number rate (transition probability rate) scattered per target particle, (NJNt) l/t. The incident number flux Ft is the number of beam particles incident upon the target per unit area per unit time:
'•-Tl-"4-
(7 80
' >
Consequently a cross section (7.79) is (NJN^A/N,), corresponding to the effective interaction area of the beam as seen by one target particle. To treat the scattering rate as a probability rate, we take Nt = 1 and JV, = 1, so that As a simple example, consider the nonrelativistic potential scattering of a single particle of momentum p scattered elastically to momentum p' (but with p' 2 = p2). Then Ft = p/mV with Tfi given by (7.76) and (7.77), resulting in
Since both particles i and / are free, i\ij, ^i oc l/y/V and | Tfi | 2 oc l/V2, so that ofi in (7.81) is independent of the box volume, as is required of any physical observable. Likewise the differential cross section da/dQ, corresponding to a measured number of particles scattered into a given solid angle is (now setting V = 1) m *
,
-
—
•
2 m
2nT»
dO. = I f(p, 9)2 | 2 dQ,
(7.82a)
since we know from (7.14b) t h a t / = —(m/2n)Tfi. This of course is the expected result, as can be seen from the structure of the outgoing spherical scattered wave in (7.2a), which gives da = i ^ p £ r2 dO. = | f(p, 0) | 2 dQ.
(7.82b)
This cross section (7.82) is useful primarily for one-body potential scattering. For general two-body elastic or inelastic scattering, either nonrelativistic or relativistic, phase-space considerations require a slight generalization of (7.76), to be considered in subsequent chapters.
Unitarity and Probability Conservation 139
7.G Unitarity and Probability Conservation For 5-Matrix. Since the overlap S-matrix elements Sfi are the transition probability amplitudes from an initial state i to a particular final state/, the total transition probability from state i to all final states is Yj | Sfi |2. Probability conservation dictates that the latter is unity: I | S / . f = l-
(7-83)
Writing \Sfi\2 as SJtSfi = SjfSfi, it is clear that (7.83) are the diagonal elements of S^S, i.e., (SrS)u = lH for various states i. The generalization of this relation to all matrix elements of S^S, both diagonal and off-diagonal, is referred to as the unitarity of S, Sts
= sst
= i.
(7.84)
The general proof that S is unitary follows in the Heisenberg picture with Sfi = (.^J I >A;+> a n d Sji = S*f =
I s}nsni = z < ^ ; | v„-><^; | = < = <5/;, n
(7.85)
n
which follows from the completeness of the "out" states 1\/~ > and the orthogonality of the "in" states | if/?"). In the interaction picture, S, = U(oo, — oo) is manifestly unitary, directly implying (7.84). For r-Matrix. First convert S to the reaction matrix R, S = 1 + R. Then the no-scattering identity matrix cancels from both sides of (7.84), giving i? + K + = -RiR.
(7.86a)
For one-body processes conserving energy, we know that Rfi = — iS(Efi)Tfh so that (7.86a) becomes - id(Efi)[Tfi - Tff] = - X d(Efn)8(Eni)T*f Tni.
(7.86b)
n
But d(Ef„)d(Eni) = 8(Efi)8(E„i), so that the delta function S(Efi) cancels from both sides of (7.86b) provided Ef = Et. That is, (7.86b) can be expressed "on the energy shell" as the unitarity relation i(Tfi - Tfr) = X 8(Eni)TtfTni.
(7.87)
n
Now the left-hand side of (7.87) is - 2 Im Tfi (since Tfi = Tif for elastic spinless one-body processes, due to time-reversal invariance) and the righthand-side summation reduces to J dNn = J dEn p(E„), where p(E„) is the single-particle nonrelativistic density-of-states factor (7.77). The scattering amplitude/= —(m/2n)Tfi then obeys the "elastic unitarity" relation (see Problem 7.9)
Im/(p', P) = | J dCK /*(k, p')/(k, p).
(7.88)
140 Formal Theory of Scattering At relativistic energies where matter can be created, new channels open up and (7.88) must be modified. We postpone a discussion of relativistic unitarity, phase-space complications, etc., until Chapters 10 and 15. Optical Theorem. The connection between elastic scattering and probability conservation is worth pursuing in detail. In general, the diagonal elements of (7.87) are - 2 Im Tu = £ d(Eni)\ Tni\2 = £ rni = Th n
(7.89)
n
where (7.75) has been used to convert the right-hand side of (7.87) for i = / t o the total transition probability rate rt. For scattering processes, rt is related to the total cross sectionCT,0,via (7.81) as T{ = F;fftot, where CT
tot = Z °ni = Gii + n
Z CTm = nfi
CT
elas +
CT
inelas-
C7-90)
Then (7.89) becomes . - - - 2 ^ .
,7.91)
This relation is a version of the optical theorem, proven first in classical optics (see Section 15.G). For one-body elastic scattering in the forward direction, (7.14b) implies
f(p,p)=f(p,e
= 0)=-(m/2n)Tii,
so that (7.91) becomes <7tot = - I m / ( p , 0 = O).
(7.92)
This familiar form of the optical theorem also follows directly from (7.88) for p' = p. Its meaning in terms of wave functions is that <7tot is the area flux reduction behind the scattering target due to diffraction scattering (see Problem 7.9).
7.H Bound States in Scattering Language For nonrelativistic scattering energies E, we have seen that the Schrodinger equation H | i/f> = E \ i//} has scattering solutions in the continuum E > 0, called "in" and "out" states 11^| >. For any particular V there may also exist stationary bound-state solutions | \j/By, with E = EB= — K2/2m < 0. As H^>H0, K->0, however, bound states may also exist, |iAB>-> \il/By for EB -* EB. The implications of such a level shift will be discussed later, but first let us see how |i/fB> is related to | ^ >. Completeness. The completeness and orthogonality of the scattering solutions, as needed in (7.66) and in (7.85), must be generalized to include bound
Bound States in Scattering Language 141 states. That is, complete sets of states for H are |in> = {|^£+>, |,„>},
(7.93a)
|out> = { | ^ > , |^ B >},
(7.93b)
|bare> = {|
(7.93c)
and for H -* H0
The completeness relation is then X |bare>
(7.94)
with orthogonality given by (7.66d) along with the mixed relations (.ifri I "AB) = WE I "AB> — 0- Then, for example, the Moller operators can be expanded as fl±=I
m><
(7.95a)
£
only when operating on continuum states, whereas "t=I
\9EK^\
(7.95b)
£
is valid for all states because Q^. | i/fB> = 0. 5-Matrix Poles. To connect the existence of bound-state wave functions to the structure of scattering operators such as the S- or T-matrix, first recall that/(p', p) is the coefficient of the outgoing spherical wave, ^ + (r)->e*-'+/(p',p)e*7r. If ij/+ is to become a bound-state wave function, it must be normalizable as E ->EB < 0; i.e., it must damp to zero as r -*• oo. The outgoing spherical wave does this for p -»• IK, E = p2/2m -* —K2/2m, eipr -* e~Kr. However, the plane-wave exponential exp( — ;cp • r) must then be suppressed, for p f < 0. This can be achieved if the spherical wave (now damped) is enhanced by / being singular at p = ix. Alternatively, since/oc Tfi, the T-matrix T(£)must have a pole in the complex £-plane at E = EB < 0,
E->EB ^
^B
That the pole is simple is due to the relation between T(E) and G(E) as given by (7.22b) or the reduction formula (7.23). A pole in T(E) then requires that there also be a pole in the S-matrix. We shall return to the pole structure of the partial-wave S-matrix in the next chapter. For other discussions of scattering theory see Feynman (1949), Mott and Massey (1949), Heitler (1954), Schweber (1961), Goldberger and Watson (1964), Bjorken and Drell (1964), Muirhead (1965), Akhiezer and Berestetski (1965), Newton (1966), Lax and Phillips (1967), and Martin and Spearman (1970).
CHAPTER 8
Simple Scattering Dynamics
In this chapter we attempt to reinforce some of the main ideas of scattering dynamics which were introduced in the last chapter. While particle spin is certainly an essential complication in working out the details of any realistic scattering model, we shall ignore it here in order to stress the essential features of scattering quantities such as the S-matrix. We shall discuss partial-wave expansions, low-energy s-wave scattering with and without the presence of bound states, resonance scattering, the Born approximation, form factors and high-energy optical-model scattering.
8.A Partial Waves and Phase Shifts Partial Wave Expansions. From the standpoint of the rotation group, the S-matrix must be invariant under spatial rotations, i.e., [S, UR] = 0 and
[S, J] = 0.
(8.1)
For spinless, one-particle elastic scattering, S is then a scalar operator with eigenstates of the orbital-angular-momentum operator L given by the little Wigner-Eckart theorem (1.30), (p'l'm' \S \pirn} = dnSm.m8{p' - p)S,(p). The S-matrix element Sfi is then given by (1.31), Sfl =
142
(2n)3 iP
* '~ <mp
P)
I (21 + 1)P,(P' " P)S,(P). i
(8-2)
Partial Waves and Phase Shifts 143 Since we also know from (1.22) that for noncovariantly normalized states hi =
{2n)3
^p2~
P)
I (2l + W
• P), (8.3)
we may subtract (8.3) from (8.2) to obtain the reaction-matrix "partial-wave expansion" Rfi =
2
" W - r t l (2/ + l)P,(p' • p)[S,(p) - 1]. (8.4)
To proceed further, recall from Section 7.F that for nonrelativistic energies, 5(p' — p) = (p/m)5(Efi) and Rfi = — i8(Efi)Tfi. This leads to the partial-wave expansion for the T-matrix [recall 8(E) = 2nS(E)], Tfi =
(8.5)
i
Next, spherical symmetry allows us to define a partial-wave expansion for the scattering amplitude/(p, 9) as f{V, P) =f(p, 0) = I (21 + l)P,(i>' • p)/,(p), HP) = \ (
d(cos 0)P,(cos d)f(p, 6).
(8.6a) (8.6b)
Comparing (8.6a) with (8.5) for the potential-theory result/= —(m/2n)Tfi, we see that UP) = ~
[S.(P) - i]-
(8.7)
Finally, the orthogonality of the Legendre polynomials (see Problem 2.2) means that the cross section a = \ dil \f\2 has the partial-wave expansion
o(p) = Z *i(p) = 4 I (2/ + 1) | Sip) - 112. /
P
(8.8)
i
Much of this formalism remains intact for inelastic scattering, for twobody scattering in the CM system, and for particles with spin provided one employs the helicity formalism. In particular, apart from overall normalization factors, the helicity partial-wave expansion for two-particle scattering in the CM frame (pt + p 2 = p't + p'2 = 0) becomes
p'2A'2| T I M , , p2A2> oc £ (2; + l)Dft(0\
where A' = k\ — A2, A = At — A2 and px = z.
144
Simple Scattering Dynamics
Turning to the partial-wave expansion for the "in" and "out" states, we write **(') = L ( r l r / m X r / m l ^ ) = £ (2/+ l)P,(f • £)
(8.10)
i
where we have assumed i//(r) has no (^-dependence so that
Then combining (8.11) with the asymptotic form for "in" states i/>+(r)-» c i P r ^j-giprjj. a n ( j comparing with (8.10) for z = p, we see that e,pr * i » -
lip
e
e~
sl(P) --(-y-
(8.12)
We thus learn that the partial-wave S-matrix, St(p), is the coefficient of the asymptotic outgoing spherical wave in ^t{r). Phase Shifts. For elastic scattering the unitarity of the S-matrix, S^S = 1, has the partial-wave component |S,(p)| 2 = 1. The latter condition is satisfied if we choose S,(p) to be a phase: S,(p) = em'(p\
(8.13)
The real quantity (for elastic scattering) <>,(p) is called a phase shift. Given (8.13), the asymptotic form (8.12) becomes
*.+M—«*"*• ^fr + W - H ,
(814)
whereas the asymptotic plane wave (8.11) has the partial-wave component i' sin[pr — \ln\lpr. The reason for the name "phase shift" is now readily apparent; the scattering only causes the asymptotic partial wave to shift its phase by an amount dt. As the interaction V -> 0, we have (5, -»0, whereas if V > 0 (repulsive), the wave is "pushed back" and c>, < 0. Likewise if V < 0, then (5, > 0, but if V is strongly attractive, then absorption can occur as well (see Problem 8.1). The elastic amplitudes and cross section have simple phase-shift forms, as substituting (8.13) into (8.7) and (8.8) lead to ft(p) = - e«™ sin dip) = [p cot SAp) - ip]~\ 471
alp) = ^(21+
1) sin2 dtp).
(8.15) (8.16)
Low-Energy Scattering and Bound States 145 Partial-Wave Unitarity. While the elastic S-matrix has a simple partial-wave form (8.13), the elastic unitarity relation (7.88) is a nonlinear integral equation for the amplitude / In terms of partial waves, this relation is "diagonalized" to the algebraic form (Problem 8.1) Im/,(p) = p|/,(p)| 2 .
(8.17)
2
Using the identity Im/, = — | /, | Im 1//,, (8.17) can be formally solved as
,m
wr~p
(818)
This result can also be obtained from the reciprocal of (8.15), fi~1 = P cot <5, — ip. We shall make use of (8.18) shortly. For elastic scattering, real phase shifts mean that sin2 (5, < 1. The partialwave cross section (8.16) then has the upper bound at(p) < 4n(2l + l)/p 2 , known as the unitarity limit. For inelastic scattering, the total cross section must be of the form CT
tot =
CT
elas + < W
(8-l9)
Now the conservation of probability (unitarity) must be maintained as the scattering energy increases and new channels open up. This is insured by the damping of the partial-wave S-matrix, or equivalently by a complex phase shift with Im <5,(p) > 0, so that S,{p) = exp[2i Re 5,(p)] exp[-2 Im <5,(p)].
(8.20)
Then (8.8) generalizes to <W(P) = 4 I ( 2 / + I ) k 2 w ' ( p , - I | 2 , P
^bs(p) = 4 E ( 2 / + l ) ( l - | e 2 i ^ | 2 ) . P
(8.21a)
i
(8.21b)
i
The optical theorem (7.92) also follows from (8.21) (see Problem 8.1).
8.B Low-Energy Scattering and Bound States 5-Wave Scattering. Classically the maximum angular momentum of a particle having momentum p that scatters off a target of dimension R is /max ~ pR. Consequently, low-energy scattering (for a finite-range potential) excites only a few low-lying partial waves. Near threshold, / = 0 s-waves and perhaps / = 1 p-waves then dominate (elastic) scattering amplitudes. Since we have just noted in (8.18) that elastic unitarity completely constrains Im 1//,, we need only find Re l/f, in order to determine the entire partial-wave amplitude. But (8.15) implies Re 1//, = p cot 5,. As p-> 0, it can be shown that <5,(p) oc p 2 , + x (see Problem 8.5). Consequently, we expand the low-energy, / = 0 s-wave amplitude Re l// 0 in a power series in the
146 Simple Scattering Dynamics three-momentum p — | p |, Re Vf0(p) = p cot 50(p) = a0~' + {rep2 + •••
(8.22)
(note that many texts choose the opposite sign convention for a0), which is even in p for parity-conserving (strong or electromagnetic) scattering processes with F, -> (— )'Pt. The first two terms in (8.22) are known as an effective-range expansion with a0 the s-wave scattering length 80(p)/p -* a0 andCT0-> 47tao a s P ~* 0- For the case of nonrelativistic potential scattering, the effective-range parameter re can be expressed directly in terms of wave functions, whereas the scattering length f0 -»a 0 is a simple measure of a weak potential [/-» — (m/27r)V},-; set box volume equal to unity and see Section 8.D]: a0 = - 2 m f dr r 2 F(r). (8.23) •'o Consequently if 7 < 0 (i.e., 80 > 0), then a0 > 0, whereas if V > 0 (i.e., (50 < 0), then a0 < 0. Shallow Bound States. The above result (8.23) only holds for a very weak potential, and the reciprocal experimental situation of a measured negative scattering length does not, however, always mean that the potential is repulsive. If a0 is very large and negative, then V will be attractive and strong enough to create at least a shallow bound state. To see this in S-matrix language, recall from (8.12) that S0(p) is the coefficient of the outgoing spherical wave for / = 0. Furthermore, from the discussion in Section 7.H, we know that the amplitude has a pole at the bound state p = IK, E= — K2/2m < 0. This means that the first term in (8.12), S0(p = iK.)e~Kr, will have the correct long-range bound-state tail provided it dominates the large second term, eKr, i.e., provided S0(p = »c) = oo.
(8.24)
Clearly (8.24) is consistent with (7.96)—but we can go further for individual partial waves. We can also require the second term in (8.12) to provide the long-range bound-state tail e~Kr for p = — I'K if the (S-matrix) coefficient of the then large first term vanishes (Heisenberg 1943), i.e., So(p=-iK) = 0.
(8.25)
Such a zero is only valid for a particular partial wave; the entire S-matrix need not vanish at bound-state energies. Partial-wave unitarity, S0(p)SJ(p) = 1, (8.24), and the analytic reflection principle SJ(p) = S0(p*) also lead to (8.25). Returning to the scattering length and the existence of a shallow bound state, evaluating (8.7) for s-waves at p = — I'K, and applying (8.25), we see that / 0 ( - H C ) = ( 0 - 1 ) / 2 I ( - H C ) = - 1/2K. On the other hand (8.15) says l//o(—i»c) = ( — ix) cot (50 — i( —wc) ~ OQ1 — K, where we have applied
Low-Energy Scattering and Bound States
147
(8.22) for a shallow bound state and neglected the reK2 term. Thus we learn that a0 « - 1/K « 0,
(8.26)
and a0 is indeed large and negative for EB= — K2/2m and K small. A case in point is the scattering of low-energy neutrons off a proton target. To first approximation we may apply the s-wave effective-range expansion (8.22) in the CM frame with reduced mass m -»\m N to the two spin-state configurations. Experimentally the incoherent low-energy cross section aQ = 7t( | as012 + 3 | a'0 | 2 ) « 20 barns (where as0 and a'0 are the total spin singlet and triplet np scattering lengths) is very large. A detailed analysis of coherent NN scattering (see, e.g., Elton 1959) gives a"0 ~ 24 fm and a'0 ~ — 5 fm. These values are much bigger than the size of the nucleons themselves, ~ 1 fm = 10 ~ 13 cm. For the triplet case, the culprit is the shallow np spin-1 bound-state deuteron with EB= — K2/mN ~ — 2.2 MeV or K ~ 50 MeV. According to (8.26), this predicts a negative scattering length a'0 ~ —hc/K ~ —4.3 fm (recall he ~ 197 MeV-fm), not far from experiment. The large positive np scattering length, as0 « 24 fm, and also the pp scattering length, a0 ~ 8 fm, indicate almost bound (virtual) states, consistent with the absence of any other nucleon bound state save the deuteron. Contrast this with the s-wave scattering lengths for pion-nucleon scattering, a0 ~ +0.4 fm for 7t ± p->7i ± p at low energy. The latter are of nuclear size and much smaller than nucleon-nucleon scattering lengths, indicating the absence of any real or virtual nN bound states. The physical interpretation of the scattering length is the radius of a hard sphere, scattering from which gives a0 = 4nal for pa 1 (see Problem 8.2). The effective range can be interpreted as the range of a square-well potential which produces a zero-energy bound state, K -> 0 [or infinite scattering length via (8.26)—see Problem 8.2]. For a shallow bound state such as the deuteron, including the re term in (8.22) modifies (8.26) and gives for K j= 0 (see Problem 8.2)
,+
'•*;( i)
(827>
Then the more accurate values of a'0 ~ — 5.4 fm and K X 46 MeV/c lead to r'e ~ 1.74 fm, close to the range of the spin triplet np square-well nuclear potential, which barely binds the deuteron (see Section 12.C). Deep Bound States. For V very strong and attractive, say V = — V0 < 0 for a square-well potential of short range R, s-waves again dominate the scattering, but (8.26) and (8.27) are no longer valid. Instead we return to the s-wave Schrodinger equation ^
+ -r j r + 2mE)^0(r) = -2mV0^0{r),
(8.28)
148 Simple Scattering Dynamics
and solve it for the scattering problem with u — ri/^0 • Inside the well, u oc sin pr with p2 = 2m(E + V0) for r < R. Outside the well we may write u oc s'm(kr + <50) with k1 = 2mE for r > R. As is the case with the boundstate configuration, continuity of u and u' at the boundary r = R then requires that p cot pR = k cot(kR + 80). (8.29) Solving (8.29) for <50, we find = PCOtpRD + *e-™". (8.30) v p cot pR - ik ' Clearly S0 has a pole at k = — ip cot pR and a zero at /c = ip cot pR with both corresponding to the bound-state energy k2 = —2m\EB\. That is, for k = IK (or — IK) where K2 = — 2m£jj > 0, the pole (or zero) of (8.30) leads to the recognized condition for square-well s-wave bound states [see e.g. Schiff (1968), and Problem 8.2] p cot pR= -K, (8.31) with p2 + K2 = 2mV0. What we learn from these examples is that the poles and zeros of the partial-wave S-matrix do indeed generate bound states. For shallow bound states p -» + IK, whereas for deep bound states k -> ± IK. In either case the pole (or zero) is in the energy variable E ->• — K2/2m (however, not in the momentum p in the latter case), as anticipated by the energy pole in the full amplitude, (7.96). So = e™°
Levinson Theorem. For nonrelativistic potential scattering, the number of bound states of a potential for each partial wave, nB,, is related to the phase shifts at zero and infinite energy via the Levinson theorem, 3,(0) - 3t(oo) = nBln.
(8.32)
The physical meaning of (8.32) is that it counts the number of states "sucked out" of the continuum scattering region as the strength of the interaction increases. That is: i. As r -+ oo the continuum wave function becomes (say for s-waves) u0 -»sin(pr + 50(p)) -> 0 on the boundary of the infinite box. Such a boundary condition is achieved for a given potential V when pR + 80(p) = m(p)n, where m(p) is an integer. ii. The argument pR can be eliminated in this boundary condition, because pR = m0(p)n for V -* 0, in which case 80 -* 0. iii. The difference m0(oo) — m0(0), the total number of continuum states for V = 0, must also be the same as m(oo) — m(0) + nB 0 , the total number of continuum plus bound states for V < 0. Combining the above statements i—iii then leads to (8.32). A more mathematical derivation of the Levinson theorem is given in Problem 8.3. The Levinson theorem puts a constraint on the number of nonrelativistic bound states (EB mB, such as the deuteron) that a potential can have.
Resonance Scattering, Formation, and Decay 149 Unfortunately, as £ -* oo, relativistic effects must become important. In particular, it is difficult to distinguish between a relativistic bound state (EB ~ mB) and a stable "elementary" particle. Thus the Levinson theorem is of limited use outside of nonrelativistic potential theory.
8.C Resonance Scattering, Formation, and Decay If two particles cannot "stick together" to form a bound state for EB < 0 (as for the triplet np bound-state deuteron), they may "linger awhile" in an excited "resonant" state at scattering energies ER > 0. Such is the case for n+p scattering, which generates the A (i.e., 33) resonance. Resonant Cross Section. One manifestation of a resonance, either classical or quantum, is a "bump" in the scattering cross section at the resonant energy ER. While an undamped classical harmonic oscillator has a resonance denominator of the form co2 — o)R, quantum cross sections a = ]T, «r,, having a bump in a particular partial wave (8.16), peak when sin2 <5, = 1, or <5, = \n, \n, etc. Noting that sin2 <5, = [cot2 8t + 1] _ 1 , it is useful to make the identification cot2 <5, = [2(£ - £ R )/r] 2 (i.e., cot 5, -> 0 when £ -» ER or <5, = \ii, etc.). Then quantum cross sections have a "Breit-Wigner" resonant form quite similar to a classical harmonic oscillator: a<es(E)_ l (t)
° ax
where of
(n/P2)(2i + i)r 2 _ oT-ir2 ~ (£ - £R)2 + (ir)2 - (£ - £K)2 + ir 2 ' 2
= 4n(2l + \)/p
{ 35)
*
is the maximum unitarity limit at resonance,
Resonant Width. It is clear from (8.33) that the resonance width T is the energy spread of the peak at half maximum, a\es{ER ± jT) = %o?ax. Moreover for £ R ^> ^r, the bump becomes a delta-function spike; that is, in the narrow-width approximation, the area under the cross-section curve must be [s(x2 + e2)~1 -»7r<5(x) as e -*• 0] f' d£ o f ' ( £ ) « ? H J ? " . •'o ^
(8.34)
The width T also has an interpretation in terms of the lifetime of an excited resonant state. For a time-dependent probability of the form | \l/(t) | 2 oze~r\ the average lifetime of the resonant state is
Since (8.35) is identical in form to (7.78), T is the transition probability rate for the decay of the resonance into its constituents. That this T is the same as the width in (8.33) can be seen by extending the energy variable £ into the
150 Simple Scattering Dynamics complex plane near resonance £ -*ER — i ^r. Then the time-dependent phase becomes il/(t)oce-iE'^e-iE,"e-ir',
\i]/(t)\2 oc e'u,
(8.36) 2
and the Fourier transform of i/^f) has an absolute square 11/>(£) | with a resonant denominator identical to (8.33). Resonant Amplitude. Given (8.36), we can now examine the scattering amplitude near resonance which gives rise to (8.33); viz., 1 fi(P) = P- cot (5 (p) (
1
i
^r
(8.37)
E •£« p ER - £ - i jl
That is, near <5,(£R) = \n, frc, etc., (8.36) dictates that 3, must increase through the resonance, which means that cot 3, = (ER — £)(^r)~'. This corresponds to a counterclockwise circle in the complex /,-plane with a resonance occurring when/, becomes pure imaginary at £ = ER. As the scattering becomes inelastic, the complex vector/, must shrink in length off the "unitarity circle" in accordance with (8.20). Owing to (8.7), the partial-wave Smatrix has a similar nonrelativistic resonance form to the amplitude (8.37), SJ es (£) =
E-ER-i±r E-(ER-iiry
(8.38)
a form like the bound-state pole except that the singularity is at positive energies below the real axis rather than at negative bound-state energies. Furthermore (8.38) implies S,(ER) = — 1, whereas S,(k = — wc) = 0 at bound states. For p-wave n+p scattering, the amplitude and cross section in the A(1232) resonance region are plotted in Figure 8.1(a) and (b). The resonance is approximately nonrelativistic because the A mass, mA x 1232 MeV, is only slightly greater than the sum of the constituent masses mn + mN ~ 140 + 940 « 1080 MeV. The resonant pion kinetic energy in the lab (nucleon at Im f 0>ot
^
87r/p2
Ref, (a)
(b) +
Figure 8.1 First resonance A(1232) region of n p scattering: (a) Argand diagram for p-wave amplitude; (b) resonance cross section and unitarity limit (dashed line).
Resonance Scattering, Formation, and Decay 151
2 Figure 8.2 Resonance-formation diagram. rest) frame is found from the invariant form PA = (P, + PN)2 = ml = ml + m> + 2mN(mn + KEJ,
(8.39)
or KE„ « 195 MeV. Relativistic resonances (with mRP mt + m2) can be treated in a fashion similar to nonrelativistic resonances, with denominators in (8.38) replaced by the covariant form (see Problem 8.4) (Pi + P2)2 - {ml - imR T). Resonance Formation. Consider the "scattering" of two spinless particles with masses mj and m2 which form a single spinless resonance "particle" at total CM energy E = mR (i.e., in the rest frame of the resonance). This is depicted in Figure 8.2 with p 1; p 2 , and pR represented in a general frame. The S-matrix element for this process must be of the energy-conserving form Sfi = — id(Efi)Hfi, where Efi = E — Ei — E2 and the effective "phenomenological" hamiltonian Hfi, also conserving three-momentum, is proportional to <53(Py;) with P / ; = PR — Pi — P2- Thus we write for noncovariantly normalized states
where g measures the strength of the interaction (g' = g/mR is a dimensionless "coupling constant" because g has the dimensions of the hamiltonian, i.e., mass) and the square-root factors (2EV)~i are the box normalization factors of relativistic wave functions (which will be absorbed into the normalization of states beginning in Chapter 10). Following the procedure of Section 7.F, the square of (8.40) can be written formally as | S / ; | 2 = ^(PfdVTgt/ZE^tEV3,
(8.41)
where we have used <53(0) = V by analogy with (7.72). The physical quantity of interest will be the cross section a(E), independent of the box normalization volumes, with E = Et + E2. In the CM frame the flux factor is v/V, with the relative velocity given by (px + p 2 = 0, p = |Pi | = IP21) 1; =
El
E2
.P&L±M=JE_. EtE2 E^E2
(8.42)
152 Simple Scattering Dynamics Then including the single-particle phase-space factor for the final-state resonance, dNf = V d3pR, we have v
_ra* _ [ F ^ i y _ ^ '
flux
pE/E1E2V
Ap
y
_
Rl
y
'
independent of V. The energy delta function in (8.43) indicates that we are working in the narrow-width approximation, with the resonance cross section peaking at E = mR in the CM frame. A finite width can be taken into account by integrating over (8.43) and applying (8.34), which leads to
£-£*-r.
<M4>
This result also follows directly from unitarity considerations (see Problem 8.4). In practice one measures the narrow width and peak cross section of the resonance, thereby inferring the strength of the interaction from (8.44). For example, for the (nonrelativistic) A(1232) resonance in n+p scattering, Figure 8.1(b) indicates that
R
•
< ^ ^ 2
Figure 8.3 Resonance-decay diagram.
Born Approximation in Potential Scattering 153 body decay rate using (8.41) as
r = j 1M F2 d3pi d*P2 = j „ ^ _ £ 8{E _ Ei _ Ei) d3p2 (8.45a) Note that this decay rate T is independent of normalization volume as expected. To evaluate the integral in (8.45a), we write d3p2 = 4npl dp2 to obtain (assuming 1 and 2 are not identical)
r
~^w¥'
(8 45b)
-
The "jacobian" dp2/dE in this expression is more complicated than the simple density-of-states factor (7.77) for one-body phase space. To work out dp2/dE we evaluate it in the rest frame of the decaying particle (resonance) with E = mR, pR = 0, and | p t | = | p21 = p. Then p2 = E\ — m\ — E\ — m\ leads to dE dEx dE2 p p pmR h T~ = J 1— = F~ + F~ = E- c- ' (8.46) dp2 dpi dp2 £t E2 E^E2 which, not by coincidence, is identical to the relative velocity (8.42) in the CM frame for the formation process. Inverting (8.46) and substituting into (8.45b) then gives the final form ,'2
= - ^ ^ = ^pg . (8.47) y %nm\ y 2 Ait ' For the case of the A(1232) decay into n and N, p™ ~ 230 MeV (as before) and T ~ 115 MeV then imply the coupling constant strength g'2/4n ~ 1, about half that found from (8.44). In fact, when we take into account the spin | of the A(1232), the agreement between (8.44) and (8.47) will be even closer. The point of the above calculation is to familiarize oneself with computational techniques and quantities such as two-body initial flux and two-body final-state phase space—concepts which we shall encounter many times in later chapters. r
8.D Born Approximation in Potential Scattering The Born approximation corresponds to keeping only the leading term in the Lippmann-Schwinger equation T = V + VG0 T x V. It is a reasonable approximation if V is "weak" in the sense that | VG0 V\ <| | V \. Alternatively the Born approximation is valid at high energies in a "peripheral" sense that the scattering interaction is of such short duration that the potential has "only time to act once", with V0 R
154
Simple Scattering Dynamics
former is valid at "high" energies with pR $> 1, whereas the latter contains only a few partial waves at low energies with pR < 1. Fourier Transform. Recall that for nonrelativistic potential theory, /(P'> P) = —(m/2n)Tfi- Then in Born approximation we have for a local potential - — / ( P ' , p ) * f y = \d3re-"
••'K(r)e*-'= V(q),
(8.48)
where K(q) is the Fourier transform of V(r), and q is called the momentumtransfer vector q = p' — p. For elastic scattering, p' 2 = p 2 = p2 then implies that q2 = 2p2(l — cos 9) = 4p2 sin2 $, where 9 is the scattering angle. If V(r) is spherically symmetric, depending only upon the magnitude r as V(r), then K(q) likewise depends only upon the magnitude q. Integrating out the angular dependence of (8.48), we obtain in Born approximation 2m /(P'. P) ~ — - J f dr v(r)r sin qr. q o Partial-wave projections of (8.49) are considered in Problem 8.5.
(8.49)
Screened Coulomb Potential. Consider the screened Coulomb (Yukawa) potential V(r) = ge-*'/r, (8.50) where g is the dimensionless strength (g = Ze2) and n is a screening parameter having the dimensions of mass. The Fourier transform of (8.50) is (q= |q|) V(q) = 4ng(q2 + n2)~\
/(p', p) * -2mg(q2 + fi2)'1,
(8.51)
and the resulting cross section is therefore a(p) = \
\f\2d£l
{2mgf 2n rn d9 sin 9 2 2 2 (4p ) J0 (sin i0 + /. 2 /4p 2 ) 2
16ng2 m2 fi2 4p2 + n2 ' (8.52)
As the screening parameter /i -»0, we have V(r) -*• g/r, the Coulomb potential. In this limit the differential cross section becomes in Born approximation
which is the classical Rutherford cross section (then not a first-order result). As in the classical case, the total cross section is infinite [set \i = 0 in (8.52)]. This is simply a statement that the long-range 1/r potential has a significant effect over all space. An exact quantum solution to all orders for nonrelativistic Coulomb scattering can be obtained in terms of phase shifts [see, e.g., Schiff (1968)]. We shall return to relativistic Coulomb scattering in Section 11.A.
Form Factors 155
8.E Form Factors Nuclear Form Factors. When electrons scatter elastically off of a composite heavy nucleus, e + Nuc -» e + Nuc, we may neglect the nuclear recoil and consider the electrons as (screened) Coulomb scattering off a continuous nuclear charge distribution p(R), where J d3R p(R) = 1,
(8.54)
integrated over the entire nucleus (R -> oo). Nonrelativistic electrons then see a potential (neglecting the electron spin) Knuc(r) = j d3R p(R)K(r - R),
(8.55)
where V(r — R) is a screened Coulomb potential such as (8.50). The convolution integral (8.55) has Fourier transforms related by r„»c(q) = F(
(8.56)
where F(q) is the Fourier transform of the charge distribution F(q) = j > . R e - , q R p ( R )
(8-57)
and V(q) is given by (8.51). For qR <£ 1, we may expand the exponential in (8.57) and assume p(R) to be spherically symmetric, so that j d3R RiP(R) = 0 and | d3R RiRjp{R) = (50
(8-58)
2
where {R } is the mean square "charge radius" of the nuclear distribution. Alternatively, given F(q), the inverse Fourier transform of (8.57) leads to p(R). To measure F(q), we apply (8.56) in Born approximation, T ~ V, obtaining
^r-=id%) dQnuc
l^l 2 -
(8 59
- >
\dQ/Born
where {da/dQ)^orn is the Rutherford cross section for spinless particles (8.53), modified by relativistic spin corrections in Chapter 11. If the electronnucleus potential V becomes very strong, the Born approximation breaks down. In this case it is sometimes possible to invoke the "impulse approximation", replacing (da/dQ^^ in (8.59) by a measured electron-constituentnucleus cross section. Electron scattering experiments at energies ~ 500 MeV have accurately probed F(q) and therefore p(R) for many nuclei, leading to charge radii of ~ 1-5 fm. Nucleon Electromagnetic Form Factors. We have seen in Chapters 4 and 5 that form factors exist for individual nucleons (and even for electrons) as
156
Simple Scattering Dynamics
higher-order modifications of "bare" electromagnetic currents due to the photon or pion cloud surrounding the particle. They differ from nuclear form factors in that they are definitely spin dependent (charge and magnetic moment form factors Fu F2 or GE,GM) and also Lorentz invariant, being a function only of the invariant four-momentum transfer q2 = (p' — p)2. Then it is possible to probe these form factors by scattering e + N -> e + N and applying a formula similar to (8.59) with spin effects an essential complication (see Section ll.F for details). The result is that (recall from Problem 5.8 that GE = Fl + F2 q2/4m2, GM = Ft+ F2) GE, &M /1 np |, and G"M /1 /z„ | for the proton and neutron all approximately fit the empirical dipole formula with q2 < 0, G(q2) = (1 - q2/b2)-2, b « 840 MeV, (8.60) for q2 up to 25 GeV2. Moreover, although GE(0) = 0 for the chargeless neutron, G'E(0) =£ 0 due to the effects of the virtual charged particles in the (on the average, neutral) hadron cloud surrounding the neutron. But more about this later. For small q2, q0 = E' — E « m — m = 0, so that q2 « — q2, and the nucleon form factors then have an interpretation as the Fourier transform of the charge and magnetic-moment distribution within the proton and neutron. Clearly exponential distributions of the form p(R) oc e~bR reproduce (8.60) via (8.57). The charge radius of the proton is then found from (8.58) to be (see Problem 8.6)-recall he » 197.3 MeV-fm / ? rm S =
W l 2 ^
0 8 7 f m
(g61)
compatible with data. Such a result is not all that different from the low-energy vrp scattering lengths or deuteron effective-range scales of 0.4 fm and 1.7 fm, respectively, all measuring in different ways the size of the nucleons. 8.F High-Energy Scattering For low-energy scattering, the dominant elastic channel usually can be described by a potential or a few low partial waves. As the scattering energy increases, the (high-energy) Born approximation for the potential can be used. Inelastic channels may also open up, and a complex absorption potential may be constructed. At extremely high energies, the equivalence between matter and energy requires all inelastic channels with the proper quantum numbers" to be excited along with the elastic channel. In the latter case a simple potential model is of little use in understanding the physics of high energy. Black-Disk Model. For p large (pR $> 1 or x <^ R), it is reasonable to assume that the geometrical-optics scattering limit is reached. Combining this with the dominance of absorption at high energies, we assume that the target appears as an absorptive "black disk" to an incoming high-energy particle.
High-Energy Scattering 157 That is, we take phase shifts <5, = 0 for / > pR and Im dt -* oo for / < pR as p-»oo. Then (8.20) becomes St(p) -> 9(1 — pR), which in turn implies via (8.21) that as p-> oo, pR
n I (2/+ 1) = nR2,
ff.
elas ~ * ~ 2
-abs^
2
(8.62a)
PR
1 ( 2 / + 1) = nil 2 .
(8.62b)
P7C 0
The limit <7abs = area is the expected classical result, but the limit (jelas = area is due to quantum (physical optics) diffraction scattering of the wave near the edge of the black-disk target. The latter result is similar to the fluxreduction version of the optical theorem (see Problem 7.9). Given (8.62), the black-disk total cross section (8.19) approaches, as p->oo,
pR
/(P. 0) = Ty- I (2/ + 1VMCOS 6) zip 0 1 rpR dx R «— 2xJo{x0) = - J^pRO). 2ip J 0
la
(8.64)
With d£l ss 2TC0 dO SB ndq2/p2, (8.64) leads to da
n
W
7
n i
* ~ i ti 0o\ 122 _ J 2 S*Sl/(P> lf{p 9)l)| ^l i(4*)| -
'
*
( 8 - 65 )
158 Simple Scattering Dynamics One also encounters Ji(qR) in the diffraction scattering of light off of a disk in physical optics. For R ~ 1 fin, the first zero of Ji(qR) is at q2 ~ 0.6 (GeV/c)2, and this is approximately where dael/dq2 "dips" for highenergy elastic n*/? scattering. Lastly the observed shrinking with (high) energy of many differential cross-section peaks can be understood in terms of an optical or "Regge" model. Further discussion of this topic, however, will take us too far from our main interest. For other general discussions of these subjects, see Blatt and Weisskopf (1952), Bethe and Morrison (1956), Elton (1959), Gasiorowicz (1966), Pilkuhn (1967), Matthews (1968), Schiff (1968), Bohr and Mottelson (1969), and Perkins (1972).
CHAPTER 9
Nonrelativistic Perturbation Theory
It is now time to expand the formal nonrelativistic scattering equations into perturbation series and apply them to simple atomic and solid-state problems. First we review time-independent perturbation expansions and extend them to the time-dependent case. The use of scattering diagrams will allow us to develop some intuition for this transition. Next we apply the theory to the interaction of radiation with matter, i.e., to the scattering of photons and bound atomic electrons. Finally we consider the scattering of phonons and electrons in solids, developing scattering diagrams which parallel the atomic diagrams as much as possible.
9.A Time-Independent Perturbation Theory Formal Iteration. The iteration of the time-independent LippmannSchwinger equations (7.20) yields the following time-independent perturbation series: |«AE
> = \
T(E) =V+
FG0+
(E)V +
FG0+
(E)VG£ (E)V + • • •,
G + (£) = G0+(£) + G0+(£)FG0+(£) + G0+(£)FG0+(£)FG0+(£) + - . 1
(9.1a) (9.1b) (9.1c)
2
All three series are of the form (1 — x)~ =l + x + x + ..., where x is the scattering kernel Go (E)V. Such a series of numbers converges absolutely if |x| < 1. Finding the radius of convergence for the operator series (9.1) is a 159
160 Nonrelativistic Perturbation Theory nontrivial matter, however, because the causal is boundary condition in GQ(E) does not keep the norm |G<J"(E)F| finite. Nevertheless this formal problem can be circumvented [see, e.g., Hunziker (1961), Scadron et al. (1964)], and so we ignore such subtleties hereafter. Instead we simply calculate each term, V, VGQ V, ..., and see if the potential is "weak enough" so that the series converges (see Problem 9.1). If V is too strong, the series will diverge. Then at x = 1, the series will blow up like a simple pole; but we know that the T-matrix must have a pole at real (virtual) bound states for a large enough attractive (repulsive) potential (recall the discussion of Section 8.B). Thus x = 1 corresponds to the eigenvalues of the (Schrodinger) equation G0(E)V\\j/} = |i/f>, and the formal iteration makes physical sense even for bound states [see e.g. Weinberg (1963)]. Old-Fashioned Perturbation Theory. Since the bare intermediate states | n) diagonalize the bare Green's function with H0 |n> = £°|n>, so that
(9.2)
it is natural to sum over such intermediate states \n) in the series (9.1). Substituting (9.2) into (9.1b) for continuum states with E° = p^/2m, the continuum matrix elements of T(E) obey T IP\
T/ x f ^ 3
TfiW=Vfi
+ \*P»E_Eo
V
fnVni
, [[
^Pn' ^Pn Vfn' K n Ki
+ ie+\j{E_Eo+ie){E_Eo
+
, iE)+----
(9.3) Such a series is represented symbolically as shown in Figure 9.1. For bound states H \ N} = EN \ N}, (9.1a) leads to the familiar Brillouin-Wigner perturbation series
l">=|«>+L-^0 + I I
l (E
%\b(EV°n
£ 0
,+(9.4)
for nondegenerate bound states. Such a series has a graphical representation with the single upward lines in Figure 9.1 replaced by double (bound-state) lines. The series (9.3) and (9.4) are sometimes referred to as "old-fashioned perturbation theory". Theyfixthe total or full energy parameter E in (9.3)
Time-Dependent Perturbation Theory 161
and £,y in (9.4) while summing over the bare intermediate energies E°. It is also possible to extend "old-fashioned perturbation theory" to the Green'sfunction expansion (9.1c). While G0 is diagonal between bare states, G is not. As we shall show in Chapter 15 (when we need it), this leads to the Dyson equation G~1 = GQ X — X, where Z is the self-energy operator which generates the energy-level shifts E — E° and the decay width of unstable states. It is worth mentioning that there exists an alternative approach to oldfashioned perturbation theory which fixes the bare energy parameter E° rather than the full parameter E. The Goldstone expansion in terms of E° then recovers the energy-level shift E — E°, and the analogous RayleighSchrodinger perturbation series recovers | N} from E° and | «> [as opposed to (9.4)]. Hereafter we shall consider only old-fashioned perturbation theory because it can be extended to time-dependent and relativistically covariant theories in a simple manner.
9.B Time-Dependent Perturbation Theory Except for the Born approximation, we shall make little direct use of timeindependent old-fashioned perturbation theory in this book. We shall need it, however, as a guide to setting up a time-dependent formulation of perturbation scattering diagrams. For a time-dependent interaction V(t), we iterate the S-matrix element in the Schrodinger picture (7.68) by expanding the time-dependent Green's function (7.32) as in (9.1c) and propagating the plane waves according to (7.31b). The result is the expected series Sfi =
5fi-i\dt
+ •••,
(9.5)
where (Pi(t) =
s
dt
-00
= -id(E2-E1-co)Vfi,
(9.7)
where Vfi = (q>f \ F|
162
Nonrelativistic Perturbation Theory
# (a)
(b)
(c)
Figure 9.2 Absorptive (a), emissive (b), and static (c) first-order interaction diagrams.
acting at the three-point vertex in Figure 9.2(a), is supplemented by wavy lines moving up along the time axis into the interaction to indicate that it is carrying energy co into the system. Similarly, for an emission potential which extracts energy from the system, we write Kmiss(t) = V*el™\
(9.8)
and the first-order term in (9.5) becomes, for El = E2 + co, SW=~i5(E2-E1+co)V*fi.
(9.9)
This is shown in Figure 9.2(b), with E, = Eu Ef = E2 + co, and £, = Ef. In general a real potential has both absorptive and emissive parts, V(t)= Ve-itttt + V*eia",
(9.10)
a form which is manifestly time-reversal invariant. For a static potential, co = 0, we may represent the Born approximation as shown in Figure 9.2(c). It is completely equivalent to the first term in Figure 9.1 if the potential represented by the wavy line and x is contracted to the point at x. Henceforth we shall not label the time axis, but it will be understood as upward. For covariant graphs with space and time on an equal footing, the time axis will be abolished altogether. But more about that beginning with the next chapter. Second Order. Using the Fourier-transform representation for GQ by analogy with (7.30) [the latter now no longer valid for G + with V(t) time dependent], Go (f " 0 = ^
f
dE" e-i£"(''-'>G0+(£"),
(9.11)
the second-order term in (9.5) becomes, for the real potential (9.10),
S£> = -^[°° dE" \f
dt'dte^'e-'^'e-^'-"
00
™ "x < / \{V*eiwr + Ve-imt')GZ{E"){Ve-im
+ V*eiu")\i}.
(9.12)
The product of the two potentials in (9.12) leads to four possible terms, with the integrals generating ^-functions in energy which force the V*V* and VV terms to vanish by energy conservation for the Compton graphs of Fig. 9.3.
Time-Dependent Perturbation Theory
163
The nonvanishing V*V term corresponds to the "direct" scattering graph depicted in Figure 9.3(a), where the energy ^-function generated by the time integrals in (9.12) forces co -> ca^ and co' -> co2. The additional wavy lines in Figure 9.3(a) again indicate energy transport into or out of the system. Then we have S^=
(9.13a)
-a(EflXf\V*G$(Et)V\i\
where £, = Er + couEf = E2 + co2, and the energy-conservation statement is £, = Ef = Ed. Here Ed is the value of the energy parameter in (9.13a) with GQ (£d) = [£i + cot — E° + is]'1, and E° = p2/2m is the intermediate-state energy with p„ integrated over in (9.13a) as in (9.3). Likewise the VV* term in (9.12) is called the "exchange" graph as shown in Figure 9.3(b), leading to S¥?=-i8(Efi)
(9.13b)
with co -> — co2, <»' ->• — a>i, Ef = Et as before, but E^ — co2 = E2— col = Ee. The exchange S-matrix element (9.13b) again contains an implied integral over p„, where E° = p^/2m. Photons as Particles. If the potential of interest is electromagnetic in origin, then in the rest frame of the heavy charged nucleus generating the interaction leading to x in Figure 9.2(c),. the static Coulomb potential is due to a "virtual" (but not transverse) photon (see Problem 9.9 and the next chapter), which does not carry energy into or out of the system. If, however, the charged source should be accelerating (with respect to the scattering "target"), £ and B radiation fields are produced, and the resulting transverse photons could be considered as free, [x deleted in Figure 9.2(c)]. We have seen in Section 4.D that a single free monochromatic photon of energy co also has a unique momentum, co2 = k2, and a real vector-potential amplitude (noncovariantly normalized) A(x, t) =
1 ItoV
[e(k)eikxe-i'°'
+ s*(k)e-
"],
P2 Pn'
Pi
(a)
(b)
Figure 9.3 Second order direct (a) and exchange (b) diagrams.
(9.14)
164
Nonrelativistic Perturbation Theory
with c(k) • k = 0. This amplitude may be treated as part of the minimal coupling nonrelativistic potential (p -»• p — e\ in p2/2m) V(t) = — p-A(f) + - A 2 ( t ) . m m
(9.15)
Thus far our perturbation series have been in terms of the entire potential, but since a = e2/47t ~ 10" 2 , it is more natural to reorder the perturbation sum as a power series in the small coupling constant (i.e., the charge) e. Consequently, first-order graphs due to the e2A2 part of (9.15) will be combined with the second-order graphs associated with the ep • A term in (9.15). We shall give explicit examples of this reordering shortly. In sum, for a time-dependent electromagnetic potential, the graphical picture puts a three-body vertex involving a photon in place of the two-body vertex at x in old-fashioned time-independent perturbation theory. All that is new is the concept of direct plus exchange scattering of the photon—and this can be viewed as a "cross-channel" version of the connection between spin and statistics. But more about that in the next chapter.
9.C Electron-Photon Interactions in Atoms Now we are prepared to investigate the interaction of radiation with electrons bound in atoms. To this end we shall employ the time-dependent perturbation diagrams just described. Induced Absorption and Emission. If we shine light on an atom, the bound electrons can make transitions between levels i and j (Et < £,) as shown in Figure 9.4(a). We can construct an external electromagnetic potential to lowest order in e, Vexl(t) = (e/m)p • Aext(t), with a monochromatic planewave amplitude Aex.(0 = A0(a>)e''(kr-M') + A$(co)
(9.16) 3
1
At transition energies a> = k ~ 10 eV, Ratm ~ A ~ 10" eV" (recall he ~ 2000 eV-A), so that (kR)Mm <§ 1. We may therefore employ the dipole
absorption J
emission
absorption t emission
(a)
(b)
(c)
Figure 9.4 Induced absorption and emission of photons from bound-state atomic electrons.
Electron-Photon Interactions in Atoms
approximation e±iK Vexl(t) are then
r
165
~ 1 in (9.16). The bound-state matrix elements of
Vfi{t) ~ -
Pji
• A 0 (co) e - tot + -
m
Pij
• A$(a>¥°*,
(9.17)
m
where the Heisenberg equation of motion dt/dt = i[H0, r] implies Pji/m = i[H0, r]fi = iE^}i, (9.18) with Eji = Ej — £, and r^,- the dipole matrix element between bound statewave functions,
r,, = J d3r^(r)#,(!•).
(9.19)
For induced absorption i -*j, the first-order S-matrix element (9.7) selects out only the first term in (9.17) by energy conservation Et + a> = Ej, Sfs = 8{EJt - a))ecoA0((o) • r}l.
(9.20a)
This is depicted in Figure 9.4(b) (double lines indicate bound states). Similarly, for induced emission j -»i, [Figure 9.4(c)], the first-order S-matrix element (9.9) in dipole approximation with £, = Ej — w is Sffiss = S(Eij +
(9.20b)
Note that (9.20a) and (9.20b) are the time-reversal transforms of one another because the dipole operator is hermitian, r 0 = r*. To proceed further, we compute the transition rates for these two processes. Since | S^ \2 = \ Stj p (microscopic reversibility), we have oij. = coij = iSjtf/T = 8(Ej -Ei-
(o)e2co2 \ A0(co) • in\2,
(9.21)
where we have again used S2(E) = 7S(£). In this case we have no density of final states, because the final electrons are always assumed bound and the final-state photons (for induced emission) contribute to the energy density of the photon "gas" with u{a>) -» 2co2 \ A0(co) \2. For a single photon with amplitude (9.14), u(a>) becomes co/V. Identifying this volume factor with u(a>) corresponds to only the photon energy, but not the three-momentum, variable, so that dF^ = co^ da> with energy density u(co) dco between a> and a) + da>. Then with a solid-angle average over dipole-transition directions, |^|A-r|
2
= i|A|2|r|2,
(9.22)
the total induced absorption and emission rates are (recall e2 = Ana) r „ = r „ = j dr;? = U2^{E3i) \ tji \2,
(9.23)
where M(£J,) is the photon energy density per unit a>, evaluated at (o = Ej — £,. Given (9.20)-(9.22) and the incoming photon flux n(w) = u(a))/a>, the absorption cross section
166 Nonrelativistic Perturbation Theory
Spontaneous Emission. Apart from the emission of a photon induced by an incoming radiation beam, an atom in an excited state j will spontaneously emit a photon of energy co = Ej - E{ when the electron jumps to a lower level i as shown in Figure 9.5(a). Here we can treat the single photon as a free particle rather than as part of an energy density u(co). The semiclassical derivation of the spontaneous emission rate is based upon the classical power emitted from an accelerated (charged) particle, P = 2aa 2 /3. The electron's radius vector, r(t) = xe-'im + rV"', (9.24) for a circular orbit then leads to the time-averaged acceleration
fa(2a}4)|riJ|2 co
4
3|
|2
(9.25) With | r | ~ 1 A (the size of an atom) and co ~ 10 eV, (9.25) gives r ~ 10" 2 x 103 (2000)" 2 ~ 10" 6 eV, corresponding to the lifetime T = h/T ~ 10" 9 sec (see Problem 9.3). Note also that the dipole (£1) selection rules /; = lj±l and mi = mj± 1, ntj are an immediate consequence of the vector nature of r,,- coupled with the Wigner-Eckart theorem (see Section 2.E). It is instructive to rederive the spontaneous-emission rate (9.25) via our quantum-scattering-diagram approach. For the graph shown in Figure 9.5(b), the matrix elements of the single-photon-emission potential pick out the second term in (9.17), withy -* i and A$(w) replaced by E*(k)(2coF)"* in dipole approximation according to (9.14): Vtj{t) = ie(2coF)"*£ij£*(k) • t ^ ' .
(9.26)
(The basis for such an emission potential, with the photon as an output rather than an input generating the interaction, is the underlying existence of the many-body quantum photon field as briefly described in Section 4.D. We shall continue, however, to interpret the photon as a single elementary particle throughout this book.) Then the first order S-matrix element (9.7) free photon
J
(a)
(b)
Figure 9.5 Lowest-order spontaneous emission of a photon from an atom.
Electron-Photon Interactions in Atoms 167 becomes, for spontaneous emission, it eiE'%(t)e-
Sy=-ij = -d(Eij
+
iEit
co)^=B*(k)-rij.
(9.27a)
Since the final-state photon is free, the density of final states is dNf = Vd3k. The differential decay rate is then independent of normalization volume, I S1 I 2
AlTH
dru = LuLdNf
=
_ _
W31 E * ( k ) . ti.
|2 dQk.
(9.27b)
To convert (9.27) to a total transition rate we first must sum (9.27b) over all possible final-state spin configurations (two in this case for the transverse photon). For photon helicity states ^ -» EW, we recall the closure relation derived in Section 4.C, YJs*w(k)sf(k)
= Sab-kakb,
(9.28)
which insures that there are only two spin states, i.e., k • e(k) = 0. While the solid-angle integral over the spherically symmetric part dab of (9.28) is J dQ, = 4n, the integral over the tensor fca£b term in (9.28) must, by symmetry considerations, be proportional to Sab, which is the only tensor independent of k available. This leads to the identity \dnkkakb
= ±4n5ab,
(9.29)
obtained by contracting both sides of (9.29) with 5ab. Given (9.29), the total spontaneous emission rate is aco3
a, r , = — Ida Z|s (k)T l 7 |
aft,347t
2n
| r 0 . | 2 ( l - i ) = faa) 3 |r,| 2 , (9.30)
identical to the semiclassical rate (9.25). There is a less general (framedependent) way to compute photon spin sums. It is worked out for various cases in Problem 9.4. It is also possible to calculate the spontaneous-emission rate for atomic spin-flip magnetic transitions in a similar manner. Combining the effective hamiltonian, in this case - | i e • (V x A), with the electron magnetic moment (-e/2m)a and the free-photon vector-potential amplitude (9.14), the Smatrix element for the lowest-order "transition, still described by Figure 9.5, is S y = 8(EtJ + to) - ^ = <«>-*•'«!./> • k x £*(k).
(9.31)
168 Nonrelativistic Perturbation Theory
Applying (9.31) to the Is hyperfine spin-flip transition [the 21-cm line— recall (5.154)] leads to a rate proportional to to3 in dipole approximation, like (9.25), but since the hyperfine level splitting is so small (~ 10" 5 eV), the resulting lifetime is fantastically long, T ~ 2 x 1014 sec (see Problem 9.5). (However the 21-cm absorption line for neutral hydrogen in the intergalactic medium has a spin-flip rate enhanced by about 104 due to thermal collisions between the hydrogen atoms.) Also, the 2 s | -> ls[ metastable Ml transition in hydrogen can be computed using (9.31), giving x ~ 2 x 107 sec (Problem 9.5), far slower than the two-photon spontaneous-emission lifetime of ~ 10 _ 1 sec which dominates this transition. Finally, given the induced rate (9.23) and the spontaneous-emission rate (9.25), it is possible to pass from the one-particle photon picture to the Bose-Einstein many-body photon distribution at temperature T, ,v>lkT u(
I]' 1 ,
by following in reverse the well-traveled path first walked by Einstein in 1917. Details are worked out in Problem 9.6. In general our scattering diagrams will correspond to zero-temperature quantum amplitudes, but it is usually possible to infer finite-temperature S-matrix elements by folding in the relevant quantum-statistical distributions. We shall do this only in Section 9.D. Bound-State Compton Scattering. If a bound electron is excited by a free photon, the possibility exists for Compton-type scattering y + ebd -> y + ebd. The time-dependent potential is again (9.15), with a free-photon amplitude given by (9.14). The resulting graphs, now ordered according to the power of the coupling constant, are the second-order analogs of Figure 9.3 for free photons, Figure 9.6(a), (b), and a first-order graph involving two photons as depicted in Figure 9.6(c) with the coupling e2A2/m of (9.15). Then the direct, exchange, and "seagull" graphs respectively contribute to Sfi as Sfi = S£d> + Sf? + S^.
(9.32)
The direct graph Figure 9.6(a) is given by (9.13a) with G0(Ed) diagonal between the various atomic bound states | n>, so that (GQ -* G0 for bound-
(a)
(b)
(c)
Figure 9.6 Direct (a), exchange (b), and seagull (c) Compton scattering of photons off bound atomic electrons.
Electron-Photon Interactions in Atoms
169
state matrix elements) C(2d)
Pbn • e*(k')Pno' e(k)
•i8(Efi)
m Vjlco'lco
„
(9.33a)
where £ d = Ea + co = Eb + co'. Likewise (9.13b) leads to the exchange Smatrix element C(2e)
with £„
i3{Efi) CO
e2
^ Ptn • e(k)P„a' e*(k')
mVy^T^?
£ e -£„
(9.33b)
- Eb — co. Finally the seagull graph yields ^ A 2 ( t ) k>e _U! -'$ba 2m
•'-00
2e 2
=
-WEji);
(9.33c) = E*(k')-E(k)^„
2mKv/2a/2a;
where the factor of 2 enters (9.33c) because there are two nonzero contributions in the product A2(r) (i.e., E* • E and E • £*), which of course are equal. The terms E • E and E* • t* are ruled out by energy conservation. In all three S-matrix elements, the argument of the ^-function, Efi, stands for overall energy conservation, which is the same in all three cases. Since only the photon is free in the final state, dNf = V d3k'. So with the incoming photon flux of c/V = V~l, the cross section is then the physical quantity independent of normalization volume. Consequently we have [lfik' = a>'2 ctco' dQk./(2n)2]
4,-W'
T 1_ m
2
$H- (£K** Pirn • £*(k')P»a • « 0 0 t*A
t'n
(9.34)
P*n • E(k)P„fl ' E*(k')
Ee-E„
dQk,
called the Kramers-Heisenberg formula, with r0 the classical electron radius r0 = tx/m ~ 3 fm. There are three limits of (9.34) worth noting. For small co, no other level can be excited, so co' = co, a = b. As co -> 0, the cross section (9.34) must of course vanish (for verification see Problem 9.7). Then expanding the energy denominators in (9.34) as (£d e = Ea ± co) (£d,e - £„)" 1 = (Em ± ft))"* * E-J [\ + ^-
+^-),
(9.35)
the first two terms give no contribution in (9.34) (the second because the direct and exchange terms then cancel). The result is that the leading lowenergy contribution in (9.34) comes from the third term in (9.35), which, being squared in (9.34), gives
170 Nonrelativistic Perturbation Theory This low-energy co4 dependence for co
° = r 2 | £ *(k')-e(k)| 2 ,
(9.37)
the Thomson differential cross section. To obtain the unpolarized cross section, we must first average over initial spin states and sum over all final spin states (not of the bound electron). This amounts to summing over all the helicity states for both the initial and final free photons and dividing by 2, the total number of initial photon spin states. That is (also see Problem 9.4),
|e*(k') • £(k)|2 ->± X |e*U,)0O • £w(k)|2 = Wu - W(*y - Mj) = «1 + (t' • t)2]- (9.38) Then integrating over the solid angle in (9.37), and using (9.29), the unpolarized Thomson cross section is tf-rhom = irg J dQk, [1 + (£' • £)2] = \r% 4«(1 + i) = fTir2.
(9.39)
This, of course, is just the classical result that probes only the (squared) charge-to-mass ratio of the target, independent of the target spin. Lastly, in the energy region where co = Ena = E„ — Ea, the contribution of the direct graph, (9.33a), becomes infinite. Clearly this is unphysical and simply means that we must include a "radiation damping" correction factor due to the spontaneous emission of photons from the intermediate states n in question. This width Tn, given by (9.25), must be incorporated into G0(Ed) so as to keep (9.34) finite. This is analogous to the Breit-Wigner resonance configuration discussed in Section 8.C; in this case it is called resonance fluorescence. It corresponds to replacing E„ in
awk. ~ r° U JU* J (Ea + co-ERr+in •
(940)
with the matrix elements of p again given by (9.18). Photoelectric Effect. If the (free) initial photon striking an atom has enough energy co to ionize the bound atomic electron, then Et = Ea + co, but Ef = p2/2m for a free nonrelativistic electron. This corresponds to co in the keV region, so that E1 <^ co <^ m ~ 1 MeV, in which case kR ~ 103 x
Electron-Photon Interactions in Atoms
171
k Figure 9.7 Lowest-order photoabsorption diagram.
(2000)"1 ~ 1, so that the dipole approximation is not valid. However, then to ~ p2/2w ~ 1 keV implies p ~ 105 eV/c and pR ~ 102 ^> 1, so that the Born approximation should be valid for the time-dependent electromagnetic potential (9.15). This is depicted in Figure 9.7. In this case, the lowest-order absorption S-matrix element (9.7) becomes (removing the V_i factor from
< / | e ' k r p | a > = -i
Jd3re-ipVkrViAa(r)
= (p - k) J d\ *-•*-"> ^a(r) = q^q),
(9.42)
where q = p — k, and iAa(q) is the Fourier transform of the bound-state wave function
TJ/v=
3{Efi)
^»r? ' ^'(q) '2 ' £(k) ' P '2 ^
(943)
Using d3p = (2n)~2p2 dp dClp and dp/dEf = m/p [recall (7.77)], the differential photoionization cross section takes the form d
°
d£lp
_ a P £ / , :. , . ^ | 2 , „2 ™ . A , 22
% co
l
For hydrogenlike atoms in the ground state, i/^ = 1(r) = {Z3/italf
exp( - Zr/a0)
leads to the Fourier transform [use (8.49)] v* na
\ ol
UZ/a0
P/^fTq 2 ]
(9-44)
172 Nonrelativistic Perturbation Theory
The average over initial photon spin states using (9.29) is (see also Problem I «W • P | 2 - i I | e(A)(k) • P | 2 = i[l - (t • P)2].
(9.46)
Finally, the angular dependence of the momentum-transfer variable q2 = 4p2 sin2 %0 makes the solid-angle integral over (9.44) somewhat difficult. So instead we specialize directly to the physically interesting limit of en > Ea, with co ~ Ef or q2 = p 2 = 2mco. Then combining (9.44)-(9.46), we obtain for the photoionization total cross section o{co)^zZ*{^al
(9.47)
where Ej = 13.6 eV and co is in the keV region. It is worth noting that the cross section for the inverse process, called radiative recombination, follows immediately from (9.47) via the principle of detailed balance (see Problem 9.8).
9.D Electron-Phonon Interactions in Solids Recall from the discussion in Section 4.D that a phonon is a quantum associated with lattice vibrations in a solid. It is like a photon in that we may assign a unit "spin" to it, corresponding to the displacement vector associated with lattice ions. Phenomenological Interactions. In a microscopic sense, a phonon is a collective sum of Coulomb attractions between an electron and the positively charged lattice ions. To construct a single-particle interaction hamiltonian between an electron and a phonon, we must make a quantum average over the Fermi sea of degenerate electrons with number density nF = 2 x 47tp|/3(27t)3 = Pf/3n2 and nonrelativistic Fermi energy EF = pj/2m oc nF*. In particular, for a low-momentum acoustical phonon, with a> = csq, we can assume that the energy of an electron is perturbed as a result of the change in the density of lattice ions when a phonon is present. Then energy = F x d, where the dimensionless dilation d is d -* V • A, with A the phonon displacement amplitude and the force (in units of energy) is given by the crystal deformation Cx = — V dE/dV = | £ F . The lowest-order quantum-interaction hamiltonian is then (set t = 0 in the electron wave functions) H(t) = Ct j d3r ^*.(r)V • A(r, t)<AP(r),
(9.48)
with the (spinless) electron and phonon single-particle wave functions given by [recall (4.78)] ^(r) = 4 = g * - ' J
Aq(x) = - ^ = ^ £ ( q y " •'£-'»'+ c.c. (9.49)
Electron-Phonon Interactions in Solids
173
Substituting (9.49) into (9.48) gives a momentum-conserving interaction hamiltonian (absorptive or emissive), C e+itot Hq{t) = + i i £ • qS3(p' - p + q). (9.50) i yJ2p(aV The static limit of (9.50) corresponds to setting q = 0 in the wave functions, so that E = E, p' = p, and <53(0) -> V. In this configuration it is also convenient to replace the lattice mass density p by NCM/V, where Nc is the number of unit cells in the crystal and M is the lattice molecular weight. The static deformation hamiltonian for acoustical phonons with a> = cs q is then fl,(Q)- + f , C l £ q . (9.51) qK ' j2NcMcsq ' On the other hand, for optical phonons coupled to electrons via a longrange dipole interaction V(r) oc r~2, the Fourier transform is proportional to q~*. So for momentum-conserving processes we write H t
,( ) = iyL-yr-J
—
5
V - p + q),
(9.52)
with only the longitudinal polarization mode of the phonon contributing (Frohlich 1954). The choice of the dimensionless coupling constant ccl is conventional (not to be confused with the fine-structure constant a). It is about a t ~ 1-10 in most ionic crystals and can be measured directly in terms of the dielectric constant for the crystal [see e.g., Kittel (1963)]. The hamiltonians (9.50)-(9.52) are phenomenological in the sense that Cx and (XJ are not predetermined fundamental constants (like e). Also (9.50)-(9.52) have been Fourier transformed into momentum space for simplicity. Their time dependence, however, still must be integrated out through the scattering formalism. Finally, there is no guarantee that the coupling constants Ct and ax are sufficiently small that the Born approximation is valid. Higher-order corrections can be interpreted as a "form-factor cloud" of phonons surrounding each electron. The combined electron-plusphonon cloud is called a polaron. But more about that when we study renormalization in Chapter 15. Here we simply look at the lowest-order scattering graphs. Resistivity in Metals. The static deformation interaction (9.51) is adequate to describe the long-wave length (q -> 0) acoustical-phonon-electron interaction in metals. The lowest-order interaction is shown in Figure 9.8. It is static because nF ~ 1023 cm" 3 in metals implies (approximating a metal by a Fermi gas of electrons) that EF ~ 10 eV and v ~ pF/m ~ 108 cm/sec, which is much greater than the speed of sound in a metal, cs ~ 105 cm/sec. This means that a> = csq <^ E, E', the initial and final electron energies. Thus the interaction is static, with no energy loss to the electron: E' = E. The S-matrix element associated with the lowest-order static graph of
174 Nonrelativistic Perturbation Theory
Figure 9.8 Lowest-order resistivity diagram. The wavy line represents a phonon.
Figure 9.8 is obtained from (9.51) as C, E • a5(E' - E)
m
'"-- *W')--mk^-
(953)
Only the final-state electron has a phase space dNf = V d3p', and with ]T |E • q \2 -> q2, the net rate of momentum loss due to both emission and absorption of phonons [providing a factor 2Ap • p/p = 2(1 — cos 6)] in a lattice at temperature T [the latter giving the Bose-Einstein fraction N' q = (e!c«lkT - l)" 1 ] is \sfi\2 „ 2 A P - P rne, = I ^ - Nq ~-=^-± dNf
Mcs\Nj)
ct3p' 8(E'-E)Nqq(l
-cos0).
(9.54)
To simplify (9.54), we assume there is but one electron per cell that scatters (as in copper), so that NJV ~ nF near the Fermi surface, where the scattering takes place. Also p' = p implies dp'/dEf = me /p' ~ \/vF and q2 = (p' - pf = 2p2 (1 - cos 9), so that dQp. = 2ra*(-cos 6) = 2nq dq/p2. Then d3p' = p2 dp' dQp = 2nq dq dE'/vF, and (9.54) becomes r
»«« = A
C2 J 2
r2pF 3
dq N«q\
(9.55)
where 0 < q < 2pF, which accounts for any value of q for electrons within the Fermi sea. Since r net is independent of the normalization volume, it is of physical significance. Alternatively we may compute the temperature-dependent "ideal" resistivity p of a metal, p = mje2nx (e2 in MKS units), where 1/T -> r n e t , since the net rate of momentum loss is what causes an electrical current to damp out. Now we reinstate factors of h and define a dimensionless variable x = (cJkT)q with xF = (cJkT)2pF. Then (9.55) leads to the ideal resistivity {k0 = 9 x 109, a = k0e2/hc) _ me(TnJh3) _ k0C2(kT/hcs)5 r* x 4 dx nFtxch2/k0 4ncmFccsvFMme J0 ex — 1' At high T, xF -»0 and the integral in (9.56) becomes xF/4, corresponding to pozT, the classical limit. For low T, xF -> oo and the integral is 4! £(5) ~ 25.
Electron-Phonon Interactions in Solids 175 Then p ccT5 (Bloch-Griineisen law), which appears to hold for most metals. While there are important corrections to (9.56) in a realistic metal (impurity and Umklapp scattering), the scale of (9.56) in the low-T quantum region is reasonable. For example, for copper at 4°K with cs ~ 105 cm/sec, C t ~ $EF ~ 5 eV and M ~ 60mp, (9.56) predicts p ~ 10" 13 Q-m (i.e., 10" 2 3 esu, which means 10" 23 sec), not inconsistent with experiment. Semiconductor Relaxation Time. Consider the emission of phonons from conduction electrons in a zero-temperature semiconductor. This is shown in Figure 9.9. The phonons carry off energy as well as momentum, and so we must consider the momentum-conserving interactions (9.50) and (9.52) for acoustical and optical phonons, respectively. For acoustical-phonon emission, (9.50) leads to the lowest-order S-matrix element
Sfi =~ifjt
rf(t)Hq(t)Ht) = J ^
8 3
*(pf')> ( 9 5 7 )
where p = p' + q and E = E + a> and a> = csq. Now, however, both the final electron and phonon are free, so we must consider the two-body phase space dNf = V2 <73p' £t3q. Then the transition rate itself is independent of normalization volume [recall (<54(P))2 = VTd*{P)], and integrating over the electron phase space, we have
«y . Vg V dV *< = i^gflfi
+ - " « M «V (9.58)
where £' in <5(£' + a; — E) is replaced by £' = (p - q)2/2m. Then applying the identity (see Problem 9.9) dq 5(Efi) = 2m \ dq d(q2 — 2p • q + 2mcsq) = 2m9"10(pq-mcs),
(9.59)
where q = | q | = 2p • q — 2mcs > 0 accounts for the step function in (9.59), with the spin sum or longitudinal phonon alone leading to | e • q j 2 —• q2, the
Figure 9.9 Lowest-order phonon-emission diagram.
176
Nonrelativistic Perturbation Theory
total acoustical emission rate becomes
Here v = p/m is the electron velocity. The step function 0(1 — cs/v) in (9.60) insures that p • q = cos 6 does not assume a value greater than unity. For optical-phonon emission co ~ const; we apply (9.52) to Figure 9.9, obtaining
with the same energy and momentum configuration as in the acousticalphonon case. The two-body phase space is also dNf = V2 d3p d3q, giving a rate similar to (9.58), but now for co = const, f dq S(Efi) = 2m|(p • q)2 - 2mco r 1/2 0((p • q)2 - 2mo))
(9.62)
[recall S(f(x)) = \f'(x0)\~18(x — x0)~\, where the reality of = p q ± ((P ' 4)2 — 2mco)i is insured by the step function in (9.62). Integrating over the solid angle, the total optical-phonon emission rate is [see Problem 9.9 and Feynman (1955, 1972)] = fi
2m^/2conci1 (2n)2p
r' \2malp^
dx [x - (2mco/p2)]* 2
= a 1 o)( > /m/£ cosh"1 ^/E/m)6(E — co). (9.63) The structure of the rates (9.60) and (9.63) resembles Cerenkov radiation in a medium in which the velocity of the electron exceeds the velocity of light. The acoustical rate (9.60) is nonvanishing only if the electron velocity v is greater than the speed of sound cs, whereas for the optical case the rate is nonvanishing only if E > co. The constraint is on the velocity in the former case but on the energy in the latter, because acoustical phonons have a characteristic velocity but no characteristic energy, with the reverse valid for optical phonons. For the case of photon emission from a free electron, Figure 9.9 vanishes identically by energy-momentum conservation; i.e., p2 = m2 = (p' + k)2 implies pk = 0orco = 0 which forces the rate to vanish. Measured values of the relaxation times z = h/T as given by (9.60) and (9.63) do not agree with the determination of Cx and a1 by other means. It turns out, however, that "renormalization" of these coupling constants and the electron (polaron) mass will save the day (see Section 15.F). Electron-Electron Scattering. Electron-phonon interactions also play an important role in electron-electron scattering in metals. Consider first two
Electron-Phonon Interactions in Solids 177
(a)
(b)
Figure 9.10 Direct electron-electron scattering via phonon exchange.
electrons scattering in a metal by exchanging an acoustical phonon as shown in Figure 9.10 (again the time axis is upward). We may regard Figure 9.10(a) as representing an emissive first-order acoustical-phonon (time-dependent) potential (9.50) interacting on electron 1 and absorbed by electron 2. Likewise Figure 9.10(b) can be thought of as a first-order absorptive potential (9.4b) interacting on electron 1 and created by electron 2. In either case the momentum transferred by the phonon is q = px — p\ = p 2 — p 2 , and the energy transferred is co = £ x — E\ = E'2 — £ 2 . Second-order time-dependent perturbation theory may be used to evaluate the graphs of Figure 9.10, both of which are of the "direct" graph type [Figure 9.3(a)]. That is, for Figure 9.10(a), only one intermediate state n contributes, and according to (9.7), (9.8), and the second-order amplitude (9.13a), S£"> = - &4{PfiK921 VI p2>(£d -E-n + is)' \ V \ | V | P l >*.
(9.64a)
Here Ed = £ t + E2 = E\ + E'2 [the Fourier integrals in (9.12) always link states existing at the same times, not particles coupled to the same vertex], with intermediate energy Ean = E\ + E2 + coq (particles 1,2, and the phonon exist between times t and t', coq being the phonon energy with coq = cs q). Similarly, the direct graph of Figure 9.10(b) gives S£» = -
tf^KPi
I V | P l >(£ d - Ebn + is)- *
with Ed the same as before, but Eb„ = Et + E'2 + a>q is the total intermediate energy between times t and t'. Another way to understand the structure of (9.64a) is to alter the firstorder result (9.9) for an emissive potential Vy™'**, (9.50), to involve a "virtual" intermediate phonon, s -> Avir(q). The latter obeys a Schrodinger equation with the Fourier transform (id, -» — coq, H0 -> E'2 — £ 2 ; set V = 1) Avir(q) = [-a>, - (£'2 - E2) + is]~l ( ^ ^ ) a 3 ( p ' 2 - p 2 - q). (9.65) The virtual displacement amplitude (9.65), generated by particle 2, has momentum q to be integrated over, with the delta functions in (9.50) and (9.65) picking out q = p'2 — p 2 = p t — p'x. Adding this to a similar result
178 Nonrelativistic Perturbation Theory found from Figure 9.10(b), we are led to an effective phonon-induced interaction for electron-electron scattering (see Problem 9.10),
VlHa) = ^ t V M)
*
1
1 —
2pa>q\-o>q-(E'2- E2)
C2q„ 2
+ •
-ojq +
(E'2-E2)j
[
(9.66)
This is the same result as obtained from the sum of (9.64a) and (9.64b) written as Sfi = — id*(Pfi)Veff. Thus, for electrons on the Fermi surface, E2 -*E2, and then
r2a2
r2
n?(q)--^r=-4 J PK pet
(9-67)
is the Fourier transform of the direct potential V(r). There is also an exchange potential Vex(r) due to two more diagrams, similar to Figure 9.10, but with the two final electrons crossed (Pauli statistics handles this case). Note that (9.67) is an attractive potential, as we might expect, since the net interaction between the negatively charged electrons and the positively charged ions is attractive. Any two electrons, whether in a metal or in free space, also have direct electromagnetic interactions. There is, of course, the direct Coulomb repulsion, V(r) = cc/r. (It is tempting to try to derive this potential from the analog diagrams to Figure 9.10 for virtual-photon exchange. The problem is that transverse photons give no contribution to such a static force—see Problem 9.10. The Coulomb potential can, however, be derived within the context of relativistic diagrams, as we shall see shortly.) In a metal, such a Coulomb force is screened by all the other electrons in the "gas", V(r) -> cte~Kr/r, with a Fourier transform [recall (8.51)] ysc. Co u ..( q ) = a ( q 2
+ K2y
1
(9 6 g ) 2
Near the Fermi surface, the (Debye) screening parameter is K D = 3anF /2EF1 (see Problem 9.10). Then in the long-wavelength, q -»• 0 limit, (9.68) becomes V7; Cou, (q) - a / 4 = 2EF/3nF.
(9.69)
For an ideal Fermi gas of electrons, the speed of sound is c2 — 2EF /3M (Problem 9.11), and combining this with nF = p/M and C1 = ^EF, the acoustical-phonon-induced potential (9.67) becomes in this "jellium limit" ^f(q)--2£F/3nF.
(9.70)
The fact that (9.69) and (9.70) are equal and opposite in sign says that the net electron-electron force is zero in this idealized limit. This is simply a statement of the charge neutrality of the electron-ion system. Cooper Pair. For shorter wavelengths (i.e., q = p' — p large), the q2 term in (9.68) dominates over K2,, and then Vnet =Veit + Vsc Coul- < 0 for electrons
Electron-Phonon Interactions in Solids 179 near the Fermi surface of a superconductor. Such a net attractive "pairing" potential between two electrons near the Fermi surface leads to a long-range "boson" bound state, called a Cooper pair (Cooper 1956). For the small momentum space net pairing potential Vnei«const < 0, the binding energy A =. Et + E2 — 2EF < 0 can be found from the LippmannSchwinger iteration T=V+
VG0V + --- = V(l
-G0V)-\
with the phonon energy cut off at £ max , •Em"
T(E) = Vnei + Vn2el [ 2E„
E
V 'net
di
~
1-
/£-2£F\ (9.71)
where N(0) = dNF/V dEF is the density-of-states factor (7.77) per unit volume on the Fermi surface. Since we expect the T-matrix to become singular at a bound state, the vanishing of the denominator in (9.71) at E = £ t + E2 = 2EF + A leads to (see, e.g., Ziman 1964) A « exp
1 A^(0)|Vnet
(9.72)
Equation (9.72) leads :ads to the low temperature BCS Superconductivity di m p n e i n n 1*=»ce rc*t\r\ mension less ratio kBTc
2ne~YE fa 3.528
for Euler constant yE « 0.577216. This highly nonanalytic dependence of the binding energy A upon the net attractive but long-range pairing potential Knet is the first step towards a superconducting state—a state which is stable against the normal resistivity process of Figure 9.8. This state also depends upon many-body interactions between Cooper pairs—e.g., the ground state has electron spins antiparallel in each Cooper pair because of the exclusion principle suppression of phase space due to the other Cooper pairs for the parallel spin configuration. Since such considerations take us beyond the realm of our one-body quantum scattering methods, we conclude our discussion of phonon-electron interactions and superconductivity at this point. For more detailed discussions on old-fashioned perturbation theory and atomic calculations see Mott and Massey (1949), Heitler (1954), Bethe and Salpeter (1957), Sakurai (1967), Schiff (1968), Bethe and Jackiw (1968), Baym (1969), and Berestetskil et al. (1971); for phonon scattering see Kittel (1963), Ziman (1964), Mattuck (1967), Baym (1969), Taylor (1970), and Feynman (1972).
This page is intentionally left blank
PART III
COVARIANT FEYNMAN DIAGRAMS
This page is intentionally left blank
CHAPTER 10
Covariant Feynman Rules
In this chapter we recast the theory of nonrelativistic scattering and oldfashioned perturbation theory into relativistic, manifestly covariant language. First we reacquaint ourselves with the relativistic kinematics of twoand three-body decays and two-body scattering processes. Next we reformulate the general dynamical scattering problem in terms of a covariant Smatrix and develop general formulae for decay rates and scattering cross sections. Then after examining the composition of covariant interactions and Feynman propagators, we enumerate a set of Feynman rules from which dynamical S-matrix elements can be constructed.
10.A Covariant Kinematics Before proceeding directly to covariant Feynman diagrams, we review and streamline the relativistic kinematics of decay and scattering processes. Unlike the situation for such nonrelativistic processes discussed in Chapter 7, mass can now be converted to energy (and vice versa) in the relativistic case. Our plan will be to work around the rest frame of decay or scattering particles, defining kinematic variables which are the covariant generalizations of effective masses. Two-Body Decays. Consider first the decay D -+ 1 + 2 in the rest frame of the decay particle D. Three-momentum conservation, pD = Pi + P2 = 0, requires that J p t J = I p21 = p. The decay energies are then easily obtained 183
184 Covariant Feynman Rules from the four-momentum scalar products pDP\ = wi D £i, p \ = m\ = {PD — Pi)2> w i t n similar relations for particle 2, giving
ml + mj-ml 1
2mD
ml
+
m\-m\ 2mD
v
'
Energy conservation, mD = Et + E2, automatically follows from (10.1), as does the magnitude of the three-momentum p= (E\ — m\f = {E\ — ml)*, leading to (2mDpf
= X{m2D, m\, m\) = [m2D - (m t + m2)2][m2D - {m^ - m 2 ) 2 ]. (10.2)
The function X is in fact symmetric under interchange of any of the three masses. As expected, two-body decay kinematics are completely determined by the three masses, as in (10.1) and (10.2); there are no variable energies or angles. In a general Lorentz frame, the two body decay energies and momenta follow from (10.1) and (10.2) via a Lorentz transformation (see Problem 10.1). Three-Body Decays. We may carry over these simple two-body decay results to the three-body decays D - > l + 2 + 3 by replacing the fixed masses in (10.1) and (10.2) with variable "effective masses" defined as the Lorentz scalars mfj = (p; + pj)2. That is, we may couple together any two of the final-state four-momenta so that the general energy-momentum relation PD — Pi + {Pi + P3)) f° r example, resembles the two-body case for m\ and m\3. There are then two independent effective-mass variables for three-body cases (none for two-body decays), since four-momentum conservation implies (see Problem 10.1) m\2 + w| 3 + ml3 = mj, + m\ + ml + m\. Finally a further parameter which has important significance for any decay is the "Q value", Q = mD — £ mf, which measures the degree to which the decay is relativistic and has decay mass available to be converted into decay-product kinetic energies. Two-Body Scattering. Next consider the scattering process 1 + 2 -> 1' + 2', which we depict in Figure 10.1, satisfying the four-momentum conservation law p + q -»• p' + q' with p2 = m2, p'2 = m'2, q2 = ft2, q'2 = n'2. Since four-
Figure 10.1 Effective-mass (Mandelstam) invariants for the "s-channel" two-body scattering process.
Covariant Kinematics
185
momentum conservation is but one constraint upon the four momenta, there are three independent combinations of these momenta (and energies), but only two independent combinations of Lorentz scalar products. Because this is the same kinematical situation as exists for three-body decays, it is convenient to define analogous effective-mass variables for the two-body scattering process (called Mandelstam invariants) as indicated in Figure 10.1: s = (P + qf = (p' + q'f,
(10.3a)
t = (p'-p)2
= {q~q')2,
(10.3b)
u = {p'- qf = (q' - pf,
(10.3c)
and we also have s — u = (p' + p) • (q' + q). Again, as for three-body decays, these variables satisfy the constraint condition s + t + u = m2 + m'2 + \a2 + n'2,
(10.4)
so that there are only two independent momentum invariants, e.g., the analogs of the scattering energy and angle variables. To be more specific, two-body scattering is commonly described in the following coordinate systems: i. CM frame: Define, in general, p = (E, p), p' = (£', p'), q = (co, q), q' — (a>', q'), K = p + q = (W, K); then in the CM frame we choose K = p + q = p' + q' = 0,
P = (£, - q ) ,
W = E + a> = E + eo\
q = (o>, q),
p' = (£', -q'),
(10.5a)
q' = (to', q'\ (10.5b)
Thus s = W2 is the square of the total CM energy, and for elastic scattering (rri = m, p.' = n, qCM = \ q | = |q' |), t= - 2 ^ M ( 1 - cos 6CM) is the momentum-transfer invariant, where q' • q = cos 0CM- The crossed momentum-transfer invariant u = — 2^CM(1 + cos 9CM) + (E — to)2 can be eliminated in favor of s and t, i.e., energy (or qCM) and angle 0CM, via (10.4). In terms of these invariants, we may use the two-body decay relation (10.2) to express qCM, qcM, and 6CM in general as (see Problem 10.2) 4CM = ls % m2, ^CM^CM
fy
cos 6CM = (t-u)
q& = ~ X{s, m'2, n'2), + ~ (m'2 - n'2)(m2 - p.2).
(10.6a) (10.6b)
ii. Lab frame: Assume the initial particle p at rest, p = 0. Then we have m + a> = E' + a>', p=(m,0),
q=(co,q),
p' = (£', p'), q' = K q'),
ttiio = p • q, mE' = p • p', mm' = p • q'.
(10.7a) (10.7b)
186 Covariant Feynman Rules
The Mandelstam invariants become s = m2 + LI2 + 2mco, t = m'2 + m2 — 2m(m + co — co'), and u = m2 + LI'2 — 2mco', with co and co' replacing the energy and angle CM variables, and conversely, the lab-frame threemomenta qL = |q|, q'L = \q'\ have the particularly simple forms d =^
[(P • if ~ ™ V ] = ~
qL2 = ^2 [(P • ')2 - m2n'2] = ~
X{s, m2, ft2), X{u, m2, LI'2).
(10.8a) (10.8b)
Note that while <JICM and qL are functions of the invariant s alone, this dependence is slightly different in the two cases, as it must be, because qCM and qL are not Lorentz invariants. If fact (10.6a) and (10.8a) lead in general to the simple result qcu = qhm/s112. The relationships between thefinal4CM and q'L or between 0CM and 0L, however, are not so simple, except for the obvious inequality 0CM > 6h. Other coordinate systems are considered in Problem 10.2. The CM and lab frames, however, are both specific examples of collinear frames with p||q, with the CM frame also having the final momenta (anti)parallel. Covariant Flux. Recall from Section 7.F that the nonrelativistic flux factor is F = N/At = Nv/V -> v for N = V = 1. Since factors of 2£ and 2co invariably appear in relativistic calculations, as in the covariant normalization of states
(10.9a)
where £ and co are the energies of the colliding particles and vR is the absolute relative velocity difference vR = | (q/co) — (p/£) |, which undergoes Lorentz distortions if q is not parallel to p. Because of this it will also prove useful to consider a total momentum flux, defined as J5-K = 4(qW -coq-K),
(10.9b)
where K = p + q = (W, K). This latter flux factor is also not a manifest Lorentz invariant, so it will be helpful to relate (10.9a) and (10.9b) to a flux-type quantity which is a Lorentz scalar, the Mailer flux (see Problem 10.3) jF M = 4[(p • qf - m V ] 1 -
(10.9c)
For collinear reference frames, all three flux factors (10.9a-c) are equivalent and Lorentz invariant, J* -»& K -> JzrM (see Problem 10.3). In the CM frame p||q, K = 0, we have # c M = J^CM = ^ C M = 49CM W, while in the lab frame p = 0, ^L = J5"* = &L = 4 4L m. These relations will be useful later. Covariant Phase Space. Because the density-of-states factor d3p Lorentzcontracts along the direction of motion, it is natural to define a covariant
Covariant Kinematics 187 density-of-final-states factor as (V = l,nf = number of final state particles) dJV7v = n i F ^ (101°) / 2E«f 3 Recalling the identity d p/2E = 8 + (p2 — m^a^p where 8 + (p2 — m2) = 9(p0)8(p2 — m2), we define the covariant, n-body phase-space factor by folding in an overall energy-momentum-conserving delta function, dp™ = S 4 {K - £ P^dNT
(10.11a)
= S + (P2f - m2) n &pnf_J2Enf^.
(10.11b)
/-i
As a specific example of the utility of (10.11), consider two-body final-state phase space, K = p' + q', K2 = s, with
= 8 + [(K-qf-m'2]^.
(10.12a)
It is easy to show that 8[(K - q'f - m'2] = \8{q' • K - %{s + p.'2 - m'2)], which, when coupled with d3q' = q'2 dq' d£lq., leads to the jacobian d(q' • K)/dq' = &$/4
{
^tdnq,,
(10.12b)
valid in any frame. This will be an extremely useful formula for two-body decays and two-body scattering processes. For three-body decays, we will need to know dp3. In the CM frame K = Pi + p 2 + P3 = 0, we integrate out p 3 using 84(K — py — p2 — p 3 ) = 8(W -E1-E2£ 3 )^ 3 (Pi + P2 + Pa), leading to dpT H3
= S*(K - Pl-p2-p3) v
1 8(2TI)5
F1
F2
F3;
# ^ ^ 2Ei 2E2 2E3
p\dpx p\ dp2 dQt dQ2 8(W -El-E2EtE2E3
E3). (10.13a)
But Pi dpi = £; dEh dQi d£l2 — An d£ll2 = Sn2 d(cos 612), where p t • p 2 = cos 912. Then p 2 = (pt + p 2 ) 2 implies p3 dp3 = E3 dE3 = px p2 d(cos 6l2), so that dp3 can be written in the simple form dPr = -^dE
(10.13b)
188 Covariant Feynman Rules Translating dE™ and dE™ back in terms of the invariant effective-mass variables (Problem 10.4), dE^ dEf = dmj3 dm213/4K2, we see that dpT is indeed a Lorentz scalar, but the form (10.13b) will suffice for our purposes.
10.B Covariant S-Matrix Covariant S-Matrix Elements. Throughout the rest of this book we shall normalize states covariantly, which means that factors (2EV)~i will be removed from the initial- and final-state relativistic wave functions, as was done in Chapters 4-6. The resulting covariant S-matrix element S/T = (f\S\i}cov must then have the simple structure for energymomentum-conserving processes [recall §A = (2n)4d4] S}7 = <5}°v + iS4(Pfi)T}°*.
(10.14)
This relation then defines the covariant T-matrix element T}°\ a Lorentzinvariant scalar function of scattering (or decay) particle momenta and spins. Henceforth we will drop the 5fi in (10.14) for decay or scattering processes, since it vanishes in these cases. The reduced "covariant dimensions" of S}°v and T}°v are worth noting. Since in units of h = c = 1 each covariant state vector has dimension m" \ consistent with
dim T}7 = m4"".
(10.15)
In particular, for two-body decays (n = 3), dim T}°v = m1, so that T}°v is an effective decay hamiltonian, while for three-body decays and two-body scattering (n = 4), dim T}7 = m°, so that T}°v is dimensionless. Covariant Rates. To convert (10.14) to a covariant transition rate, r}°\ we make use of the formal identity [5 4 (P / ,)] 2 = VTd4(Pfi), so that the covariant "golden rule" generalized from the discussion in Section 7.F is (we reinstate the normalization volume in the next two equations only) I ocov 12 l
dV}7 = ' (i
dN}ov = VdA(Pfi) | T}°? |2 diV}ov.
(10.16a)
Removing the volume factors from T}Y (i.e., | Tyjv|2 oc V~"f'ni) and from the density of states (dNf oc V"f), and combining dA(Pfi) with dNc/v into the covariant phase-space factor (10.11), we have in general drc/y = K 1 -" i |T}° v | 2 dp™.
(10.16b)
For the case of decays (nt = 1), we see that (10.16b) is independent of normalization volume and r}°v is physically measurable. While dry? is manifestly Lorentz invariant, physical decay rates are not; the former is converted to the latter by dividing by the time dilation factor 2£,, which also
Covariant S-Matrix
accounts for the normalization factor (2Et) we have for decays dYfi=
~ut;=
1
189
not included in (10.10). Thus
IE, ' T™ |2 dp-'-
(1017a)
Since dim pc„°v = m 2 "'" 4 [from (10.11) with K = 1], we see from (10.17a) that dim Vfi = m, as must be the case, since r / ; is expressed in energy units. Now we specialize further to two-body decays (n, = l,nf = 2) in the rest frame of the decaying particle which is the CM frame of the final decay products. Since the final-state kinematics is completely determined by (10.1) and (10.2), we may immediately integrate over the solid angle 4n in (10.12b) with J dp2 -* mD. Then the total two-body decay rate is
^-j& lr /Tf' 8ran D
(1017b)
which is a general result if the two final-state particles are not identical. If instead they are identical, it will prove convenient to ignore this fact in Tf? (a Feynman rule) and instead integrate over only one-half the final solid angle to avoid double-counting the two indistinguishable particles. (Alternatively, the factor of one-half can be gotten from the symmetrization of finalstate wave functions while integrating over the entire solid angle, but we shall not do this.) Thus, (10.17b) must be multiplied by an extra factor of one-half for identical particles. For three-body decays (n; = 1 , nf = 3), (10.17a) is also valid, but the final-state kinematics are not constrained, depending instead upon two variables. Since a convenient choice of these variables will depend upon the particle decay in question, we postpone further discussion of T 3 until such decays are encountered in Chapter 13. One generalization of interest, however, is the additional factor of l/nf\ needed in the integrated total decay rate (10.17a) for nf identical particles in the final state, for reasons similar to those for the two-body final state discussed above. We shall incorporate this fact into the Feynman rules stated in Section 10.E to follow. Finally, for two-body scattering processes (n; = 2,nf = 2), we define the (physical) differential cross section as dafi = -£-
= — | T}712 dpT,
(10.18a)
where the initial flux factor is given by (10.9a). Note that dafi is a manifest Lorentz invariant only in collinear reference frames, i.e., for Lorentz transformations along the beam direction. Note too that dim afi = m~2 • m° • m° = m~2, as is necessary because afi must have the dimensions of area. All factors of 2£ and V have canceled out in (10.18a); dafi is therefore physically measurable and is identical to the noncovariant cross section of Chapter 7. Then applying (10.12b) to (10.18a), we obtain the
190 Covariant Feynman Rules general differential cross section dofi^jq'/ln)2
,^ c o v2| 2 \Tf°?\ ,
(10.18b)
valid in any frame. For identical final-state particles, (10.18b) must be further divided by 2, as for decays. Note that in collinear frames with both &{ and J* * invariant, the product q'2 dilq. must be invariant (see Problem 10.4). In the CM frame with J ^ -+ 4qCM W and &\. -> Aq'CM W, s = W2, (10.18b) becomes do
""CM
1
cov
fi
(10.18c)
Then TpCOV
*•-*&
(1018d)
plays the role of the nonrelativistic scattering amplitude/in (7.82b). In the chapters to follow we shall rely heavily on these general formulae (10.17) and (10.18). Covariant Unitarity. Given the covariant decomposition (10.14), we may extend the discussion of unitarity as presented in Section 7.G by converting S^ — 1 taken between covariantly normalized states, obtaining the on-shell relation analogous to (7.87): 1 (T}°? - T?/*) = l- X | dp? T<„°f*TT.
(10-19)
The sum in (10.19) is over all possible particles which could be present in the intermediate state, and dpc„oy is the covariant phase-space factor (10.11), now for intermediate rather than final states. Using PT invariance as determined from the discussion in Chapter 6, the left-hand side of (10.19) can be converted to the imaginary part of invariant amplitudes in the scattering region. But more about this in Section 15.G.
10.C Covariant Vertices Having disposed with the reformulation of scattering kinematics in covariant language, we now return to the dynamical problem of obtaining S}°v or Tf°i for various relativistic processes. Once we construct Tcf°?, we can make an immediate identification with experiment by applying (10.17)-(10.19). We begin by building up "fundamental" covariant vertex interactions out of the elementary particles (photons, hadrons, leptons), based upon the minimalcoupling hypothesis in the case of electromagnetic interactions or based upon simple phenomenological Lorentz-invariant forms consistent with the
Covariant Vertices 191
(a)
(b)
Figure 10.2 Three-point (a) and four-point (b) spin-zero-photon vertices. discrete symmetries in the case of strong and weak interactions. A discussion of gravitational processes will be postponed until Chapter 14. Covariant Currents and Potentials. Consider first the spin-0-photon (pionphoton) vertices of Figure 10.2. From the viewpoint of the charged pion, the electromagnetic field generates an effective potential, modifying the freeparticle Klein-Gordon equation to {q2-f)<$>=V^
= J.
(10.20a)
Making the minimal replacement p ~* p — eA in (4.24), we are led to the coordinate-space current (4.25), or equivalently to the momentum-space "potential" in (10.20a): K^ = e(q' + q)- A - e2A • A.
(10.20b)
The vector potential A can also be expressed in momentum space with A • Aoz2e*(k') • e(k) as in (9.33c), but we shall not do this here. Alternatively, from the viewpoint of the photon, the pion generates a chargedmatter current, which in momentum space couples minimally to the Maxwell equation (4.61) as -k2A„ = ;° m = e<}>*(q' + q\4>- e2A„>*>.
(10.20c)
The first terms in (10.20b) and (10.20c) correspond to the "Yukawa"-type three-point vertex of Figure 10.2(a), while the second terms correspond to the "seagull"-type four-point vertex of Figure 10.2(b). Next consider the spin4-photon (e.g., electron-photon, proton-photon) Yukawa vertex of Figure 10.3. From the viewpoint of the Dirac particle, the em field A generates an effective potential, again determined by the minimal replacement, i.e., 0 - my, = V+ tfr, V> = ey- A,
(10.21a)
whereas the photon sees the minimal charged Dirac current - f c % = 7j- = g ^ > .
(10.21b)
We take as a phenomenological prototype of strong interactions the spin-^-spin-0 (nucleon-pion) Yukawa vertex of Figure 10.4. Experimentally
192 Covariant Feynman Rules
Figure 10.3 Spin-^-photon vertex. we know that the pion is a pseudoscalar 0~ particle, and P-invariance (parity) then dictates the form of the effective potential, (P - tntf = V„, ifr,
V* = gy5 >,
(10.22a)
and also the effective pion pseudoscalar current, (q2-H2)d> = J = ghs^. (10.22b) Here g is a phenomenological nN coupling constant to be determined by experiment. Such a "fundamental" vertex is not unique, however, because it is also possible to construct a "pseudovector" type of coupling \j/y5 $\j/. But more about this in Chapter 12. Notice that we have used the letter p for Dirac particles, q for pions (or other mesons), and k for photons. We shall try to adhere to this convention whenever possible. Note too that we have specifically used arrows in Figures 10.2 to 10.4 to denote the direction of the current, but have not put arrows on the accompanying (neutral) boson field. For charged bosons, however, we must also use arrows to distinguish particle from antiparticle. Finally, weak-interaction couplings are observed to have the form of a product of (usually charged) currents^)", called a "Fermi" type of four-point vertex. But more about that in Chapter 13. Interaction Hamiltonian Densities. The foregoing formulation of relativistic vertices can be restated in terms of Lorentz invariant hamiltonian densities in the interaction picture with |-»jrf.
(10.23)
Recall from the discussion in Section 7.D that Jf 7(x) in coordinate space leads directly to the interaction hamiltonian in the interaction picture,
Figure 10.4 Spin-j-spin-0 pion vertex.
Covariant Feynman Propagators 193 Hj = j d3x J^i(x), and also to the interaction-picture S-matrix operator S7 = T exp
-i JVxJf7(x)
(10.24)
Then [Jf j(x), Jf 7(x')] = 0 for spacelike distances insures that 5 7 is Lorentz invariant. Our goal will be to unravel (10.24) into a covariant perturbation expansion. Of course this requires that the expansion parameter, — iJ4?,, be small. While this is expected to be the case for electromagnetic interactions with e2/4n = 1/137 or e2 ~ 1/11, for strong interactions g2/4n ~ 10 and such a perturbation expansion is suspect. But as we shall see, the latter expansion is sometimes useful in spite of this drawback. For weak (and gravitational) interactions, however, even though the coupling strength is very small, the momentum integrals over the vertex factors can at times be infinite, with such infinities increasing in number for each higher order in the perturbation expansion. For the }>„ and y5 em and strong-coupling theories, the resulting infinities do not increase in higher orders. We shall address these "renormalization" questions in Chapter 15. One final complication concerning Jf, is worth noting. The total hamiltonian density Jf (x) is not required to be a Lorentz scalar; instead, it is the time-time component of a covariant stress-energy-density four-tensor. Indeed, it is the lagrangian density of field theory, which is always a Lorentz scalar. It can be shown, however, that for simple interactions such as in (10.23), Jf 7 (x)= -J£?J(X), so that Jf7(x) must be Lorentz invariant. We shall ignore such coordinate-space subtleties henceforth and work with V, J, Jf j , etc., directly in momentum space, where Lorentz invariance is manifest.
10.D Covariant Feynman Propagators Now we come to the key ingredient in the theory. In order to build up the sought-after dynamical amplitudes T}°\ the covariant vertices just described, — tifj, must somehow be "glued" together in a covariant manner. This glue is none other than the Feynman propagators. Green's Functions as Feynman Propagators. We have already noted in Section 7.C that the bare Green's functions for relativistic scattering resemble the causal nonrelativistic Green's function (E — H0 + ie)'1 in that they are the momentum-space inverses of the Klein-Gordon, Dirac, etc., free-particle differential operators. Furthermore, the causal propagation generated by Huygens's principle was formulated in terms of propagators iG0; i.e., for relativistic scattering in momentum space the propagators are iGt(q2) = iAF(q2) = i(q2 - \i2 + ie)' \
(10.25a)
1
iG%{p) = iSF(p) = i(p -m + ie)'
= i(f> + m)(p2 - m2 + ie)' \ iGA0{k2) = iDF(k2) = - ig^k2 + ie)-l.
(10.25b) (10.25c)
194
Covariant Feynman Rules
The factor i in iG0 of (10.25) arises from i in idt. It means that a perturbation expansion such as the Lippmann-Schwinger series iG = iG0 + (iG0)( — iV) x (iG0) + ••• has — iV as the perturbation parameter, analogous to — iJtf, in (10.24). Covariant propagators in the latter equation arise, as in the nonrelativistic case (7.54), from the time-ordering operation's cutoff of the time integrals. All of this is very formal and suggests propagators as only a technical device. They are much more; but to appreciate why, we must proceed to describe Feynman's more intuitive pictures (Feynman 1949, 1961a,b). Forward-Backward, Off-Mass-Shell Propagation. The causal is prescription in (10.25) can be interpreted as defining an integration contour in the complex p0-plane. As noted in Section 7.C, the Klein-Gordon and Dirac contours pick up two residue terms corresponding to the first two old-fashioned perturbation-theory diagrams of Figure 10.5. The graph in (a) describes the forward propagation of a particle of momentum p„ and positive energy E„ = Ep + a>q, corresponding to i(E — En + ie)~l. The graph in (b) depicts the backward propagation of the particle with momentum p„ but negative energy — (£ p + a>q) = — E„ (or equivalently, the forward propagation of an antiparticle with momentum — p„ and positive energy £„). The latter therefore corresponds to i(E + E„ + ie)"1, a result that can be understood in terms of the old-fashioned perturbation theory with a five-particle intermediate state, similar in structure to the three-particle intermediate-state Green's function used to calculate the phonon-exchange graphs of Figure 9.10. These two graphs are to be evaluated off their energy shell; all states E„ (one) are summed over, with E =fc E„. Since energy and momentum are conserved at each vertex, the sum of these nonrelativistic propagators can be written as i i 2Ei £ - £„ + ie + £ + £„ + ie ~ E2 - E2 + ie 2Ei (E2-V2n)-(E2-p2) = IE —2
\
,
p2-m2
2
2
2
+ ie
2
(10.26a) v
+ ie 2
'
where we have used E — p = m and defined p = E — p^ as the virtual four-momentum squared of the off-mass-shell intermediate state with pi =1= m2. The final form of (10.26a) is identified as the Feynman covariant propagator of Figure 10.5(c), with the energy parameter E reinterpreted as the virtual energy of the intermediate state and the factor of 2£ absorbed into the covariant normalization of states [i.e., factors of (2£)"* are removed from the top and bottom vertices of Figure 10.5(a,b), which then cancel the factor 2£ in (10.26a)]. The result is the spin-0 momentum-space propagator, (10.25a). The forward-backward propagation of a spin-| Dirac particle can be handled in a similar manner. The old-fashioned perturbation-theory spinor
Covariant Feynman Propagators 195
+
(b)
(a)
Figure 10.5 Forward (a) plus backward (b) old-fashioned propagation equals Feynman (c) covariant propagation. propagators analogous to Figure 10.5(a,b) sum to
i£
w w w (P„)u (pn)
-t £ ^>(-p> u , (-p n ) +
E - E„ + ie
^-= ;; : E + E„ + ie
i(f„ + m)
=2E
2
p„ -m1
2
.
• '
+ ie (10.26b)
where we have used the arguments associated with (10.26a) again along with the spin projection operators ]T u(p„)u(pn) = p„ + m and -X
V
(-
Pn)v(-Vn) =
~(-tn)
+
m
= Pn +
m
>
and we recall that i?(p) is labeled by the momentum opposite to the direction of motion of the negative-energy state. Again absorbing the factor 2£ into the covariant normalization of states, (10.26b) is indeed the Feynman propagator for Dirac particles, (10.25b). Needless to say, the minus sign in the second term of (10.26b) saves the day. This sign is related to the chargeconjugation properties of the spinor current (6.25c) and also plays a central role in the connection between spin and statistics. Off-Spin-Shell Propagation. For massive spin-1 propagators, we again turn to Figure 10.5 to write the sum of forward and backward propagation in old-fashioned perturbation theory as
i I #\9nttX)*(Pn)
l I 4A>*(-Pn)4A,(-Pn) _ X
X
E — En + ie
E + E„ + ie —
•*•£'
7
2
1
p*-m
:
•
(10.27a)
+ ie
To obtain (10.27a), we have used the fact that a physical spin-1 particle having four-momentum p has three spin states satisfying £ 4 W * ( P ) = £ 4 A ) *(-P)4 A ) (-P) = -fo.v X
P.PJP2)
X
(10.27b)
196 Covariant Feynman Rules
in order that the subsidiary condition p • s(p) — 0 may remain valid. However, the spin-1 particle in Figure 10.5, being off its mass shell (p2 =/= m2), cannot obey p • e = 0. Thus we must add in a fourth virtual spin state corresponding to spin 0, so that p • a =fc 0. It satisfies
44,(PnK4,*(pn) = 4 4 ) *(-p n K 4 ) (-P„) = ^ V ^
PIPI
(io.27c)
and vanishes, as it must, on the mass shell (p2 -* m2). Adding (10.27c) together with (10.27b) indeed leads to the desired covariant spin-1 propagator of (10.27a). Thus the virtual spin-1 propagator is "off its spin shell" as well as off its mass shell; all four virtual spin states must be included in the propagator-numerator spin sum of (10.27a). The photon propagator involves one more degree of complication. Replacing p„ by k„ in Figure 10.5 (because the letter k is especially reserved for photons), the analog of (10.27a) for a massless particle is
i i #>(k„)e>*(k„) —
:
j£ £ u>*(- k n ) £ a>(_ k n ) +—
co — u>„ + is
:
=2ca~^,
co + a>„ + le
(10.28a)
k„ + is
with the factors 2co absorbed into the covariant normalization of states. The form (10.28a) then corresponds to (10.25c), the Feynman propagator for photons in the Feynman gauge. To understand this result, we decompose the off-shell spin sum for k2 =fc 0 into spin-1 transverse, spin-1 longitudinal, and spin-0 parts: 4 / , &n \*J£y
trans V*V
2-i
^n^v
' ^ji^v |long "• fi/i^v |spin 0
= 8
( "-w)+{w)+{~d,io9voy
(ia28b)
While the longitudinal component appears necessary in order to make the usual identification £ 3 £M e* -* <50, it has, along with the unfamiliar spin-0 term, a deep dynamical significance that is worth pursuing in detail. Sidewise-Force Propagation. Consider a positively charged particle 1 Coulomb-scattering off a positively charged particle 2, with lowest-order S-matrix element S}°v = — ie/'J A\, where A% is taken as an external potential as depicted in Figure 10.6(a). Treating this problem symmetrically, we can think of Al as generated by the charge density ejl according to Gauss's law V • E 2 = e/'o or \i2Al = ej%. Solving for A\ in Sfi, we see that S}°v = — ie2jok~2Jo, which has the structure of a "sidewise" (static) propagator graph as shown in Fig. 10.6(b), corresponding to the instantaneous Coulomb hamiltonian
Hc„ul = ^ U 3 ^ V % M M ) . an J
r —r
(10.29)
Co variant Feynman Propagators 197
(a)
(b)
Figure 10.6 Static component of the photon-exchange force. Then the Fourier transform of (10.29) can be written as
(VU = e2f0(k)-^f0(k)
*1(k) w;'j2(k) - ^ • w
m ^^-m = g2
2
!
_> g2
!
!
!
k + is
k2 + ie (10.30)
because of electromagnetic current conservation, k • j = 0 or kQj0 = kjt. We observe that the two terms in the numerator of (10.30) respectively correspond to the longitudinal and spin-0 terms in the spin sum (10.28b). On the other hand, the dynamical, transverse electromagnetic interaction #dyn = — ei ' Ax leads to the second-order interaction
e2jmUj-^)jm (FGo+ K ) d
-
=
—fc'+fe
—
•
(1031)
Adding (10.30) to (10.31) according to Figure 10.7, we have (VGZV)dya + (VU = e2Jl(k)^^j
j2(k),
(10.32)
consistent with the decomposition (10.28). Note that Figure 10.7(a) plus the longitudinal part of. Figure 10.7(b) is analogous to the dynamical phononexchange diagrams (Figure 9.10). Now we see why in fact the latter graphs cannot generate a static Coulomb potential (recall Problem 9.10). Only the static interaction of Figure 10.7(b) and (10.30) can lead to such a force;
(a)
(b)
(c)
Figure 10.7 Transverse (a) plus static (b) photon exchanges equal covariant (c) photon exchange.
198 Covariant Feynman Rules conversely, only the sum of (10.30) and (10.31) forms a Lorentz-covariant photon exchange, a situation which must occur if a quantized theory of the covariant Maxwell equations is to make any sense at all. Two other points concerning photon exchange are worth noting. First, off-shell current conservation, k • j(k) = 0, insures that any additional term Xk^kv could be added to g^ in (10.32) without altering the physical consequences. This is linked with the (off-shell) gauge-invariance condition in coordinate space (4.56) and is a dynamical generalization of the kinematical on-shell gauge-invariance relation (4.60). For example, the photon propagator numerator is — ig^ in the Feynman gauge but — i(gr„v — /c„ kv /k2) in the Landau gauge. Second, we note the significance of the minus sign in the spin-1 equation k2Ap = — ;'*m as opposed to the plus sign in the spin-0 equation (q2 — n2)
Feynman Rules in Momentum Space 199
be uniquely pinned down off mass shell. Only on mass shell (p2 = m2) will, say, the s = f propagator numerator—the projection operator (5.171)—be unique.
10.E Feynman Rules in Momentum Space The Feynman diagram scheme should now be almost evident. A covariant S-matrix element in any order of perturbation theory can be built up from a series of covariant vertices connected by the appropriate Feynman propagators. Any such diagram is therefore composed of (a) external particles in the i and / states, (b) covariant vertices, (c) internal covariant propagators, (d) momentum-linking internal integrals and vertex energy-momentumconserving delta functions, and (e) various spin-statistics minus signs and multiplicity and statistical factors. Accordingly, we list here the various momentum-space components of Feynman diagrams as a series of rules by which to calculate any prescribed S}°\ In the chapters to follow we shall make extensive use of these rules, explaining them in further detail when necessary, in order to work out the simple consequences for the relativistic quantum theory of electromagnetic, strong, weak, and gravitational interactions. Spin-1 Feynman rules are discussed in Problem 10.5. Spin-2 masslessgraviton rules are similar to those for photons, but with much more complicated vertices and propagators, which are worked out in Chapter 14. External Particles a. Incoming "particles": 1
spin 0 w
^
q
w (p)
spin j , A = ±\ (particle)
DU)(p)
spin j , X = ±j (antipart:icle)
^'(q)
spin 1, A = ± 1 , 0
q
£tA)(k)
spin-1 photon, X = + 1 .
k
b. Outgoing "particles": 1
spin 0
« w (p)
spin j , X = ±j (particle)
vw(p)
spin j , X = ±j (antiparticle)
4A)*(q)
s
e^l)*(k)
spin-1 photon, X — ± 1.
P i n 1, A = ± 1, 0
[For brevity, we shall sometimes write up for «a)(p), etc.]
200 Covariant Feynman Rules The arrows always follow the charge, which is also the momentum direction of particles, but opposite the momentum direction of antiparticles. For fermion lines, work against the arrow when writing down products of spinors. Propagating Lines spin 0
q2 — /i2 + ie i{P + m) p2 — m2 + ie
spin j spin 1
q2 — (i2 + ie
photon (Feynman gauge).
k2 + ie
Vertex Factors. -i8A(py?*') = - i(27r)4<54(P}ejrtex) multiplied by couplings. e{q' + q)n spin-0-y { -2e2gflv
spin-i-y
97s
spin-^-7c
e[(
-9 spin-j-spin-j.
Intermediate Integrals. Integrate over every internal propagating line of momentum / by \ d*l = (27t)-4 j d4l. The net effect is that the product of vertex delta functions 34(PyiItex) will be reduced to one, that of overall energy-momentum conservation of S}°v = i54(P/j)T}°v- The only remaining integrals will be over closed internal loops. Such loops can generate
Feynman Rules in Momentum Space 201 infinities, which must be circumvented for a physically meaningful Sfi (see Chapter 15). Statistical Factors and Bonus Minus Signs a. External: —1 —1 -I-1 l/nf! (2SJ + l ) - 1
for each incoming antifermion (v) for exchange of each pair of identical fermion lines (including a fermion, say, in initial and a like antifermion in final state) for exchange of each pair of identical boson lines (need only permute final or initial boson or fermion lines—but not both) in the total rate for nf identical particles in the final state (do not symmetrize or antisymmetrize product wave functions in for i states) in the rate for each spin averaged initial particle of spin s{ (i.e., average over initial spin and sum over final spin states).
b. Internal: \ —1 ignore
for each closed boson loop containing two boson lines for each closed fermion loop all other statistical and symmetrization factors for internal lines.
Spin Sums for Rates i. Trace over each closed fermion loop. See (5.33), (5.34), (5.53), and (5.54) for the appropriate gamma matrix algebra. ii. For T=up.Mup with unobserved particles spins, J ^ | ^ | 2 - > Tr{(/ + m)M($ + m)M], with analogous formulae for antiparticles. If spin is observed, multiply associated energy projection operator by the factor j(l + iy5f) before tracing. iii. For T = e • M involving an external photon, gauge invariance (k • M = 0) then leads to the unpolarized-spin sum Yjk | T | 2 -»• — M • M*. If T = a • M refers to an external massive spin 1 vector meson, then £ A | T\2 -* —M • M* + \q • M\2/fi2. For general references on Feynman diagrams see Feynman (1949, 1961a,b), Mandl (1959), Schweber (1961), Bjorken and Drell (1964), Muirhead (1965), Sakurai (1967), Berestetskii et al. (1971), Jauch and Rohrlich (1976).
CHAPTER 11
Lowest-Order Electromagnetic Interactions
The relativistic quantum theory of interacting leptons (electrons or muons) and photons is referred to as quantum electrodynamics (QED). Applying the Feynman rules just developed, we first work our way through many of the lowest-order interactions of QED, pointing out when possible the elegant structure of Feynman graphs and their connection with CPT notions such as crossing. We then extend these techniques to electromagnetic form factors and decays of hadrons. Finally we return to the nonrelativistic concept of an electromagnetic force, i.e., static potential, and show how it can be extracted from relativistic Feynman graphs.
11.A Coulomb Scattering Before embarking upon an analysis of various two-body scattering processes in QED, we begin by reviewing the simpler one-body static Coulombscattering problem. Starting with nonrelativistic charged particles, we "dress up" the classical Rutherford cross section with relativistic and spin structure effects. Classical Rutherford Scattering. Following Fermi (1950), we consider the small-angle deflection of a beam of light charged particles with momentum p = mv and charge e scattering off of a charged (Ze) heavy nuclear target. For small scattering angle 6, the transverse momentum p ± ~ p6 can be 202
Coulomb Scattering 203
calculated from the electric impulse force Pl
= \F1dt
= e\E1dt
= ^vJE1dA,
(11.1)
where b is the impact distance of closest approach and the area dA is a "Gaussian-type" cylindrical pillbox surrounding the target nucleus. Gauss's law, <>j EL dA = Ze, converts (11.1) to b = (2Za/pu)0_1, and then da = — 2nb db leads to the Rutherford differential cross section da = dQ
bdb Ode
=
4(Za)2 1 (Za)2 p V 0 4 "" 4p V sin4 # '
[
'
where we have generalized (11.2) to larger angles via the replacement (acquired by hindsight) \Q -> sin \Q. As noted in Section 8.D, this exact classical result is identical to the lowest-order nonrelativistic quantum result; in the latter case just apply the Fourier transform of the Coulomb potential V(r) = Zoc/r to the Born-approximation differential cross section (8.53). Relativistic Spinless Coulomb Scattering. For a static, "external" Coulomb field, we have
•<"(«) = \dlx «•" •Al»{x) = ^g„0HE J
-E).
(11.3b)
q
The energy delta function in (11.3b) simply says that a static potential will not transfer energy into or out of the system with which it interacts. To lowest order, this potential "interacts once" with a charged spinless particle as shown in Figure 11.1. (We use solid lines to denote the spinless particle here because the same graph will be referred to for the spin-j case.) In terms of the covariant "potential" (10.20b) and (11.3b), the Born approximation in the Schrodinger picture is to order e, S™ =-ij
d*x cj>*.(x)V(x)(t>p(x) = -ie(p' + p) • A°«(q),
(11.4a)
where q — p' — p. This result also follows directly from the Feynman rules: — i for one vertex with coupling e(p' + p)^. The energy-momentum delta
Figure 11.1 Lowest-order Coulomb scattering diagram.
204 Lowest-Order Electromagnetic Interactions function in the Feynman rules is replaced in this case by the static interaction (11.3b), itself containing the energy-conserving delta function, giving for e>0 Ze 2 Sjs* = - i — 2E8{E' - £),
(11.4b)
where q2 = (p' - p)2 = 4p2 sin2 ifi. Now we work out the relativistic single-particle phase space. Given (11.4), the covariant rate is | OCO' cov
dr «,v =
12
I J' I ^ J V } ° V
—=r2£ q
8{E
'~E)W-
< 1L5 >
Then combining p' dp' = E' dE', d3p' = p'E' dE' dQp,, with the covariant flux & = 2Ev = 2p, we have with e2 = 47ta,
which is identical to the classical and nonrelativistic Rutherford cross sections, but with v = p/E. Relativistic Spin-j Coulomb Scattering. Again referring to the lowest-order Coulomb-scattering graph of Figure 11.1, the only difference between the spinless and spin-! cases is that the vertex current e(p' + p)M is replaced by eup.y^Up. Then (11.4) becomes for e > 0 S}7 = - ieu, fupA?(q)
= - i~
up. y0 up8(E' - E).
(11.7)
If the helicity states are unobserved, the resulting spin sum is, for E' = E and yoho = P = Po7o + P Y, I
I u^y0uf
| 2 = \ Tr {(,>' + m)(p + m)} = 4£ 2 (1 - v2 sin2 |0). (11.8)
Since this trace replaces the spinless factor (2E)2 in (11.5), we may immediately infer from (11.6) the unpolarized Mott differential cross section da _ (Za)2(l - v2 sin2 jfl) = %u,h(0)(l - v2 sin2 tf). (11.9) ^u„poi ~ 4 p V sin4 # To appreciate further the spin structure of spin-^ Coulomb scattering, we also consider the case where the spin states are observed. Recalling the structure of the Dirac boost in the Dirac-Pauli representation (5.66a), we
Coulomb Scattering 205 may express the helicity matrix elements of the probability current as "U,)(p')yo"W(p) =(E + m)" W ( i , ) (P'). O p + m - iy5
r?
x y0[E + m + iy5a • p] t(A
=
*•••»>- Q . then the; = ^ rotation matrices require [see (3.91)]
"<+)(P'K"(+)(P) =
2E
cos |0,
W ( _ ) (P')7O" ( + ) (P) =
2m
sin
#•
(11.11b) Squaring (11.11b), the coefficients of (2£)2 again modify the Rutherford differential cross section to — ^ — = <xRuth(0) cos2 |0,
- £ - =
ffRuth(0)
fe)2 sin2 |0. (11.12)
The sum of the nonflip and flip cross sections again reproduces (11.9). Note that this helicity analysis can be carried out in manifestly covariant language by using the covariant spin projection operator (5.72) (see Problem 11.1). Spinless Recoil Scattering. In the real world the charged scattering target is not infinitely heavy, but must recoil, conserving momentum, as shown in the Feynman graph of Fig. 11.2, with A = p' — p = q — q'. Now overall energy and momentum must be conserved, with the photon exchanged having dynamic (transverse) components as well as the static Coulomb (longitudinal and scalar) components, mixed together in a covariant way as described in Section 10.D. Thus, Figure 11.2 is our first example of a full-
Figure 11.2 Covariant recoil scattering via photon exchange.
206 Lowest-Order Electromagnetic Interactions fledged relativistic diagram. F o r t h e idealized case of positively charged, structureless a n d spinless particles of mass m and /i (i.e., leptons), the Feynm a n rules of Section 10.E give
5 r=
> iz^£ f^li_ii (A^T^) !^li£i 3-X^ order
vertex q
^
v ' propagator
vertex p
(1U3)
4mom. cons.
T h e connection between (11.13) a n d C o u l o m b scattering is worth stressing. Replace Ae" in (11.4a) by t h e virtual A^ satisfying • A^x) = fj°(x) o r A^A) = —gllve{p' + p)v/A2. C o m b i n i n g this with m o m e n t u m conservation at each vertex a n d integrating over the variable intermediate fourm o m e n t u m A according to f cf A 8\q - q ' - A)5 4 (A +p-p')
= 8*(q -q'
+ p-p')
=
S*(Pfi), (11.14)
we are again led to (11.13). Note that the ie prescription in the propagator in (11.13) can be neglected for such "forcelike" Feynman diagrams as Figure 11.2 with spacelike momentum transfers A2 < 0. Accordingly, we shall henceforth drop the ie for such configurations. To use the general two-body cross section formulae developed in Section 10.B for diagrams such as Figure 11.2, we express (11.13) as S}T = i84(Pfi)T}T. Then defining A2 = t, (// + p)(q' + q) = s - u, the dimensionless invariant amplitude T}°v and related differential cross section are
T
W
>" (V)'
da
{q'/2n)2(4noi)2
s—u
dn:,
(11.15)
F o r m$> [x, say p 2 ~ 0, it is n a t u r a l t o evaluate (11.15) in t h e lab frame p = 0, in which case (10.7b) says that s — u = 2m(co' + co) a n d t = 2m(a>' — co) % — 4a>'a> sin 2 j9L. T h e latter t w o expressions for t can be combined t o form the C o m p t o n relation (remember h o w difficult this formula was t o derive using noncovariant energy a n d m o m e n t u m conservation?) CO
co
2co 1 + — sin 2 | 0 L m
= 1
^sin2i0L. m
(11.16)
Moreover, from o u r discussion of covariant flux in Section 10.A, q2 = 0 implies ^ - » 4 q ' • K, a n d in t h e l a b frame, J5"; = J**. = Acorn. T h e n (11.15) becomes da dQ~L
4a>2 sin 4 \QL
1 -I- (w/m) sin 2 §0, 1 + (2
(11.17)
In t h e static, no-recoil limit, m -> oo a n d we see that (11.17) again becomes the Rutherford result (11.6) with co->pv a n d Z = 1.
Coulomb Scattering 207
One final point: The close connection between single-particle-exchange Feynman diagrams and the simple pole structure in the Mandelstam variables, such as the t"1 behavior of (11.13), has led to the shorthand descriptions of these diagrams as "pole diagrams" or simply, "poles". On occasion, we shall lapse into this language as well. Spin-4 Recoil Scattering. For the case of (light) spin-! electrons scattering electromagnetically off (massive but not infinitely heavy) spin-j muons, or alternatively off structureless (no form factor) spin-j "Dirac" protons, we again refer to Figure 11.2. All that changes between the spin-0 and -\ cases is the form of the currents, i.e., e(p' + p)v -* eup,yvup and e(q' + q\ -* — euq-?„«,. The Feynman rules then give S}7 = (~in-e)uq^uq[p^eup,yvupd*(Pfi),
(11.18a)
T}7 = {-e2lt)uq,yiluqup,y»up.
(11.18b)
v
[Note again that T}° is a Lorentz scalar and dimensionless by (10.15).] To compute the unpolarized square | TJi | 2 , we use the fact that the bilinear covariant uq. y^ uq is a number in the Dirac-spinor space which is decoupled from iip.y^Up. Then standard trace techniques (5.53) lead to (see Problem 11.1) Tr{fo' + m^p
+ m)yv} = L^p', p) = 4[/>;Pv + p„p'v - gjp'
• p - m2)], (11.19a)
I T}7 | 2 - £
L^(q', q)Ljp', p) = £
[(s - m2 - p.2)2 + st + it2]. (11.19b)
In the lab frame with p2 = m2 > q2 = p.1 ~ 0, we apply the general cross section formula (10.18b) to (11.19b) and find da dn unpol
a2 cos2 $0 [1 - (t/2m2) tan 2 tf\ 4co2 sin4 $0 [1 + (2w/m) sin2 |0] t
- ^2 tan 2 ±0 = <W#) 1 - 2m
(11.20)
whereCTNS(0)is the Mott differential cross section for the recoil scattering of a light spin-! lepton off a spin-0 pointlike (no structure) heavy nucleus (see Problem 11.1). Again we note the richer angular momentum structure of the cross section due to higher-spin particles and recoil scattering; as m-> oo, (11.20) becomes <xNS(0) for ca-> pv and Z = 1. For real hadron (strongly interacting) targets, there are even further form-factor modifications of (11.20). We shall return to such targets in Section ll.F. Ionization Loss. A practical application of these recoil scattering formulae is the ionization energy loss of charged particles in matter, dE/dx. Energetic
208 Lowest-Order Electromagnetic Interactions
protons traversing a medium lose energy Q = £ — £' = a/ — /x in collisions with atomic electrons, p + e(q ~ 0) -> p + e. A change of variables converts do IdQ. for Figure 11.2 to (see Problem 11.2) da
2na
dQ
/if 2
2
+
M'-£ ^
<->
2
where <2max « 2/ur (l — f )"* for incident protons of velocity v < c, and where the last term is absent for a beam of spinless particles. Then recalling that mean free paths are computed via the exponential damping law, e~nax, with n the number density of electron target particles, we may use (11.21) to calculate the ideal ionization loss: d£ = „ f° " Q d a ^ n dx -0 .
(log % * - A
(11-22)
for Q2 4, E2. Realistic modifications of (11.22) then lead to the Bethe-Bloch formula [see e.g. Fermi (1950)].
11. B M oiler Scattering The designation "Moller" here refers to electromagnetic identical-particle scattering in lowest order. Again we work out the Feynman graphs for various spin configurations. Potential Theory. Given the Fourier transform of the Coulomb potential in Section 8.D, Vfi = 4na(4p2 sin2 jQ)'1, identical charged bosons or fermions modify the potential to Vfi -> ^(direct) ± ^.-(exchange), where 9 -»• n — 9 for the exchange potential. Working in the CM frame with reduced mass m -* \m, the nonrelativistic Born approximation yields for particles of spin s da
U) 2
2
am 4 16p
Vfi ± V}? | 1 1 2/(25 + 1) sin4 | 0 cos4 $9 ~ sin2 $6 cos2 |6
(11.23)
The factor of (2s + l ) " 1 in the cross term of (11.23) is due to a quantumstatistical counting argument [see e.g. Schiff (1968)]. This factor will appear naturally in the examples of relativistic Feynman diagrams to follow. Spinless Mailer Scattering. For relativistic, spinless, charged identical particles (bosons), the lowest-order Feynman diagrams are shown in Figure 11.3. Note that the relative positive sign between graphs with final-state (or
Moller Scattering 209
q'
^
q
,*
* '
-p
P
p'
+
> P
Figure 11.3 Spinless Meller scattering. alternatively, initial-state) identical bosons follows from the Feynman rules. Then we may immediately write down S}7 = (-i)2e(q'
w
+ q\
pf + ie
( ^
+ H'W 1
+
^ ({q<-pgy
(q' + q) • (p' + p) fi
+
e(p' +
p)v8*(Pfi)
e
+ i)
U + P)^(pfil
(P' + q) • («' + p)
(P' ~ Pf
W - Pf
= e
(n-24a) s— u
s— t
+ (11.24b)
where t = (// — pf and u = (q' — p)2. Under the transformation t<->«, the two terms in (11.24b) become interchanged. This is an example of crossing symmetry, with the sum of the t- and w-channel poles in (11.24b) possessing, for Ty-f = A(s, t, u), the property A(s, t, u) = A(s, u, t). Bose statistics or charge conjugation of the general 0 + 0 -> 0 + 0 scattering amplitude insures that this crossing-symmetry property under t <-> u will be valid to all orders in the electromagnetic (and strong) interaction. To construct the differential cross section for Figure 11.3 in the CM frame, we use ^ i = ^^ = Aq^fs, with s = 4E2 = 4(q 2 + m2), t = - 4 q 2 sin 2 %0, and u = — 4q 2 cos 2 j6. Then we obtain (see Problem 11.3) do*"**
(q'/2n)2
ddcM
or arK
2f &% a2m4 16£V
1 \ l Si \
(87t^) :2
T>cov 12
I l fi I
lab b2 , + • 4 4 2 + cos he sin 40 cos 2 \0 sin \6
(11.25a) (11.25b)
where a = 1 + (1 + cos 2 i#)q 2 /m 2 and b = 1 + (1 + sin 2 ^0)q 2 /m 2 . As q 2 /m 2 ->0, the result (11.25) becomes equivalent to the nonrelativistic expression (11.23) for 2s + 1 = 1. Electron Mailer Scattering. For spin-j electrons, the lowest-order Feynman graphs are shown in Figure 11.4. The relative minus sign between these two graphs is due to the crossed fermion lines in the second diagram, again a
210 Lowest-Order Electromagnetic Interactions
Figure 11.4 Spin-j Moller scattering. Feynman rule. By analogy with (11.18) and (1124) we infer T?y = Td-Te
e2 e2 U — « « ' ^ " « P ' A P - — Up-y^Uf-fti,.
(11.26)
We see here that Td<-+ —Te under the crossing p'<-+q', t*->u. The unpolarized square of (11.26) is found from
|77T|2-i I
(|T d | 2 +|T e | 2 -2ReTJT e ),
(11.27)
spins
with the factor of £ due to averaging over the two spin-| electrons in the initial state, (2s + l) 2 = 4. The relevant traces are then similar to (11.19a), having the invariant forms analogous to (11.19b) (see Problem 11.3),
H\Td\2 = —2ir(q',q)L„v(p',p) 2e4 H\Te\2
=
[{s - 2m2)2 + (u- 2m2)2 + 4m2t],
(11.28a)
—2L^p\q)Ltlv(q',p)
= \[{s-
2m2)2 + (t- 2m2)2 + 4m2u],
(11.28b)
i Re £ TSTe = — Tr{(
(11.28c)
We see that the sum of (11.28) according to (11.27) remains unchanged under the crossing t<->u, s fixed. Note that the interference term (11.28c) is not the direct product of two traces LMV, each involving four y-matrices; instead it is one trace over eight y-matrices. This is because the latter juxtaposition of Td and Te keeps the spinor indices in the order required for matrix multiplication. To evaluate such a big trace, it is helpful to reduce the number of y-matrices via (5.54).
Bhabha Scattering 211 The Moller differential cross section in the CM frame is then found from (11.25a), (11.27), and (11.28) to be daunpoi
x2m4 32EV
B
sin |6
+ cos4 | 0
sin
2a2 1 + = L cos2 | 0 m
A B
4
-(••3H 1 + m
2
C= 1 -
sin
2 U
0
u
2 1
2C 6 cos2 #
4q2
(11.29)
sin2 |0,
(11.30a)
cos2 \0,
(11.30b)
m -
m2
4(q2 )\ 2
(11.30c) m As q2/m2 -• 0, we observe that A -* 2, B -»2, C -»1, so that the nonrelativistic limit (11.23) is recovered, this time with a coefficient of —1 in the interference term, corresponding to — 2/(2s + 1) for s = \. Note that dividing £ | T | 2 in (11.27) and (11.28) by the statistical factor of 4 was necessary to achieve the correct nonrelativistic limit.
11.C Bhabha Scattering If one of the identical particles in Meller scattering is replaced by its antiparticle, the process is referred to as Bhabha scattering. We analyze the various Bhabha scattering processes by invoking the notion of crossing. Spinless Bhabha Scattering. The Moller momentum configuration for n+ is pn+ + q„+ = p'n+ + q'„ + . Let us apply the CPT operation and convert particles q and q' into antiparticles: g„ + (in) -> 4*-(out), q'n+(out) -> q'n-(in). This CPT transformation and momentum conservation, pn+ + q'n- = />'„+ + qn., then require that q — —q and q' = — q'. The corresponding direct plus exchange Feynman diagrams for n+n~ scattering are shown in Figure 11.5, with the direction of the arrows for q and q' following the negative-energy particle states and therefore opposite to the sense of motion of the antipar-
y p
q \
?y
* p
i
/p'
" V /'
\ p
q>'
\
P
Figure 11.5 Spinless Bhabha scattering.
V
> p
212 Lowest-Order Electromagnetic Interactions tides n (q) and n (q1). The Feynman rules then lead to
S}7 = (-ife(-q' - q), ({p,Irf + {-ife{V ~ q\ {{_-qry^
+ iE)e(P'
+
^{--1'
+ PWV/.) +
P\S4(Pfi). (11.31a)
Notice that we have expressed the electromagnetic current at the q vertex in the negative-energy sense e( — q' — q)^, which is, of course, the same as the antiparticle current ( — e)(q' + q\. The former interpretation will manifest the crossing (CPT) nature of Feynman graphs; this too is built into the Feynman rules, even for spins s > \, where the connection between antiparticle and negative-energy currents is not so trivial (recall the discussion in Chapter 6). The corresponding covariant scattering amplitude is then T}7 = e2 (-q'-q)-(p'
[P'-Pf
= e2
t
s— t
-+
+ p) +(p'-q)-(-q'
+ P)
(-q'-pf (11.31b)
where the Moller invariants in (11.24) become s = (p + q)2 -> (p — q)2, t = (p'-p)2-+(p'-p)2 = (-q + q')\ and u = (q' - pf ^ (-q> - pf. That is, Tcfi of (11.31b) has the same invariant structure as the Moller amplitude (11.24b). Crossing. The difference between the spin-0 Meller and Bhabha amplitudes is then not in the invariant structure of the amplitude, but instead in how the invariants themselves are interpreted. For s-channel Moller scattering in the CM frame, s = 4E2 = 4(q^ + m2) is the squared energy variable, while t = — 4q^ sin2 j6s and u = — 4q2 cos2 %6S are momentum-transfer invariants. On the other hand, for Bhabha scattering in the u-channel CM frame, u = 4E2 = 4(q2 + m2) becomes the squared energy variable, while t = — 4q2 sin2 j6u and s = — 4q2 cos2 j0u are now momentum-transfer invariants. Thus, while the t*->u transformation generates the crossingsymmetry property of the Moller amplitude, the crossing of the Moller to the Bhabha process corresponds to the switch of s and u as energy and momentum-transfer invariants. Then both Feynman graphs are not only time independent, they are channel independent, having the same invariant meaning when "tumbled" on their sides [e.g., from the Moller (s) to the Bhabha (u) channel]. This crossing property of Feynman diagrams follows from their Lorentz in variance and the absence of (nonanalytic) invariants such as 6(p0) in Feynman S-matrix elements. In the same way that total crossing is linked with the CPT invariance of Sfi, crossing two particles from one channel to another holds for individual Feynman diagrams. In fact, the
Bhabha Scattering 213 simple (analytic) structure of Feynman graphs allows us to cross (CPTtransform) even one particle at a time to relate, for example, a two-body scattering amplitude to a three-body decay amplitude (substitution rule). In any case, given (11.31b), the w-channel CM cross section for spinless Bhabha scattering is then (Problem 11.4) daunpo1
dnCM
(8TT./U)
2
\w\2
arm
a
b
T 2
16Etfu ^Wu~l
+ m /qlj '
(1L32)
where a = 1 + (q 2 /m 2 )(l + cos 2 jOu) and b = (q 2 /w 2 ) cos 8U. Electron Bhabha Scattering. For e~e+ -> e~e+ scattering, we must not only superimpose the Dirac spinor algebra upon the above spinless Bhabha amplitude, but we must also explain the relative minus sign in Figure 11.6. This sign is a Feynman rule. The second graph in Figure 11.6 does not appear to be a fermion-exchange version of the first diagram, as was the case for electron Moller scattering of Figure 11.4. But imagine instead leaving the photon and electron line p fixed and permuting the other fermion and antifermion lines around, not crossing p, until Figure 11.6(b) takes the form of Figure 11.6(a), but on its side. Line p' goes by q and q' with a net plus sign, and then q' goes by q with a resulting sign change—in spite of the fact that one of these latter particles is in the initial and the other in the final state (a Feynman rule). The final configuration is equivalent to Figure 11.6(b) but viewed from the right side. Such a weird geometric contortion is just a visual analog of the algebraic bookkeeping process of anticommuting creation and annihilation operators in field theory. More to the point for our purposes, the relative sign in Figure 11.6 is a manifestation of crossing; the Moller graphs of Figure 11.4 cross to the Bhabha graphs of Figure 11.6 only if the same relative minus sign is there in both cases. The latter Bhabha minus sign is therefore a crossed version of Fermi statistics—again an ultimate consequence of the CPT theorem (and the connection between spin and statistics) combined with the absence of the invariants 9(p0) in the Feynman rules.
(a)
(b)
Figure 11.6 Spin-^ Bhabha scattering.
214 Lowest-Order Electromagnetic Interactions To proceed further with electron Bhabha scattering, the Feynman rules give Sfi =* a*(Pfi)TJ?, with 1
fi
- v?y»v«up-yf'uP - -
~
wwrfup
(11.33)
where the overall sign is a consequence of the Feynman factor of (— v) for an incoming external antifermion. Rather than compute the unpolarized spin traces of (11.33) "from scratch", we invoke crossing to deduce the equality of the unpolarized square | Tf?*\2*-* | T" p "| 2 when each is expressed in terms of the Mandelstam invariants s, t, u of (11.28). The Bhabha cross section in the w-channel CM frame is then (see Problem 11.4) d ( T unpol
<*OcM
1
cov 12 I nncov I
I
(87Tv4) 2 I fi I
cc2mA
A' 4 sin \eu 32E2rf
+ (1 +
B' m2/qu2)2
2C (1 + m2/qu2) sin2 R j
+ •
(11.34) 2 40, A' = 1 + 2 % cos 2"u m
B' = 1 + 2 % cos2 # , C=\
+ f l + ^ ) 2 - 4 4sin 2 R, +
1 + 2 ^ sin2 \9U m
+ AEilm2,
-A {El - q2 sin2 \du) m
l l . D Compton Scattering The quantum-mechanical scattering of light by free charged particles is called Compton scattering. In the classical limit u>, co' <£ m, such scattering will not change the wavelength (i.e., energy) of light (photons), but quantummechanically it will. Kinematically, Act) — a>' — a is obtained from the Compton relation, co'/co = [1 + (2(o/m) sin2 ifl] - 1 . Dynamically the Compton scattering cross section can be computed via the Feynman rules. Classical Thomson Cross Section. When an electromagnetic wave impinges on a free charged classical particle, the particle accelerates and reradiates at the same frequency, with a cross section formally equal to energy rate divided by energy flux. The scattered energy rate corresponds to the average power radiated by a charged accelerating particle, faa 2 , where a = eE/m and E is the electric-field amplitude. Combining this with the initial energy flux
Compton Scattering 215 (in rationalized units) E2, we obtain the Thomson cross section J
Thom
§ oc(eE/m)2 87ir2
energy rate energy flux
(11.35)
where r0 = ct/m is the classical charge radius of the scattered particle. For an electron, r0 ~ 2.8 x 10" 13 cm and ae ~ 10" 24 cm2, which are far larger than the classical charge radius and cross section for a target proton with (Tp ~ 10" 31 cm2. Since the classical limit is recovered for low quantum energies, the Thomson cross section must be the low-energy limit for the radiation scattered off any charged object, depending only upon the ratio of the object's squared charge to its mass. Spinless Compton Scattering. If the scattered photon is to have a different frequency than the initial photon according to the Compton relation, the lowest-order dynamical graphs must be second order in e. For a free spinless relativistic particle, the Compton graphs are shown in Figure 11.7. Since the Klein-Gordon and nonrelativistic Schrodinger equations are both quadratic in momentum, the minimal coupling replacement for the Klein-Gordon case makes Figure 11.7 similar to the nonrelativistic bound-state Compton graphs of Figure 9.6. We use the Feynman rules to write the sum of the three graphs of Figure 11.7 for q + k = q' + k' = K and q — k' = q' — k = K as
S}7 = (-ifet(k')e(q' + Kf ( ^ — i ^ — ^ J ^ + q)%(k)d*(Pfi) + (-i)\(k)e(q'
+ Kf (K2_lm2
+ ie)e(K
+
qYe*(k')8^Pfi)
+ (-«>*(k')(-2 g V v K(k)3 V/<) = iTJi^iPfi). (11.36) It is useful to express this covariant amplitude in terms of the Compton M-function, T}7 = e*"(k')MMVev(k), with iW„v = -
q
2
\(2q' + k')„(2q + k)v
|
(2q - k%(2q' - k\
\
e
*
q */
q *
/ Figure 11.7 Spin-0 Compton scattering.
\
^ ^
216 Lowest-Order Electromagnetic Interactions where sm = s — m2, um = u — m2, and the Mandelstam relation (10.4) is sm + t + um = 0. The Compton tensor M-function (11.37) exhibits two properties worth stressing. First, since the external photons have but two transverse spin states, we know from the discussion in Section 4.C that the on-shell gaugeinvariance condition is satisfied: /c'"M„v = M„vfcv = 0.
(11.38)
In fact, (11.38) is also satisfied for off-shell photons k2 =/= 0 when the dynamical coupling is of the minimal (gauge-invariant) form q -»q — eA. That this is so can be seen directly from (11.37), since sm = 2q • k + k2 and um= -2q' • k + k2 lead to M„vkv oc (2q' + k\ - (2q - k% - 2k,,, which vanishes by four-momentum conservation q + k = q' + k'. Secondly, (11.37) is crossing symmetric in that exchanging the role of the two photons in Figure 11.7 via the crossing transformation k *-* — k', ft «-> v leaves Tf? unchanged. Likewise, the spinless-boson crossing transformation q*-> —q', H~* n, v -» v also leaves T}°v unchanged. These symmetry transformations then imply the crossing relations (see Problem 11.5) Mjk',
k) = Mv„(-k,
-k'),
Mjq',
q) = M„v{-q, -q').
(11.39)
To compute the Compton differential cross section, it is convenient to work in the lab frame q = 0, with fluxes ^ = 3F% = 4ma>. The subsidiary condition e'* • k' = e • k = 0 and the special transverse gauge choice e0 = 0, leading to 6 • q = e'* • q = 0 in the lab frame, then force the first and second terms of (11.36) to vanish (the s- and u-channel poles, respectively). This is similar to the nonrelativistic bound-state Coulomb process of Section 9.C. All that remains is the "seagull"-graph contribution Tf? -*• 2e2e'* • e, which generates the lab cross section for sm = 2ma>, um = — 2ma>',
H^y ! | r j r | 2 = (£) % 5 | E '*- t | 2
(n4oa)
'
Since (11.40a) is valid only for the special transverse gauge, we shall sum over the final and average over the initial photon polarizations as in Section 9.C (see also Problem 11.5). This leads to
where cos 0L = £'• fc. Then the solid-angle integral over (11.40b) gives a = (
Compton Scattering 217 k'
P'
Figure 11.8 Spin-j Compton scattering. seagull diagram as in Figure 11.7. Now the Feynman rules give Sfi = (~i)h*(k')up,ey
+
i(fc + m) K -mz
(-i)hv(k)up,ef
z
+ is
efupsv(k)5*(Pfi)
i(£ + m) efu,e*(P)3\Pfi), Kz -mz + ie (11.41)
corresponding to the M-function
+
y v (Jt + m)y/J|
(11.42)
Note that this M „v again satisfies the k<-> —k' photon crossing-symmetry property (11.39a), but the p<-> — p' electron crossing-symmetry property is now (see Problem 11.5), (11.43) M > ' , p ) = C " 1 M j v ( - P . ~P')CThe conditions (11.39) and (11.43) are special cases of C-invariance for elastic scattering with net r]c = 1. Note too that M^v also satisfies the gaugeinvariance condition (11.38), but to prove this one must use Dirac algebra to push f terms to the left in £ = p' + ty' and p terms to the right in 1JL = j> — Ijt' in order to apply the free particle Dirac equations (problem 11.5). While the current j^ obeys k-j = up.ljtup = 0 between on-shell spinors, the off-shell currents in (11.42) are not gauge invariant—yet the overall M-function is gauge invariant—without the existence of an electron seagull-type term. With reference to this latter point, it is the backward-propagating negative-energy electron states in both the direct and exchange graphs of Figure 11.8 that simulate the seagull graph of spinless Compton scattering. To see this, make a nonrelativistic reduction of T}°v- For the first graph of Figure 11.8, we obtain in the gauge e0 = 0, with the fermion propagator decomposed as in (10.26b), - T^'/e2 -• (E'* • up,yuK)(uKyup • -
s)(-Imo)-1
(E'*-up.yvK){vKyup-£)(2m)-2
-> 0 + (e'* •
= £'* •
E
+ ia£'*
(11.44a) X £,
218
Lowest-Order Electromagnetic Interactions
where we have used uyu -* 0 and uyv -» — 2ms in the nonrelativistic limit. Similarly the second, exchange graph of Figure 11.8 becomes - T}Y/e2 " • £ ' * • £ - is • e'* x c,
(11.44b)
as might be expected from (11.44a) on the basis of crossing. Then adding (11.44a) and (11.44b) together, we obtain Ty?-> -2eh'*-t, which is the required form for the nonrelativistic Thomson amplitude. Thus we see that the backward-propagating parts of Feynman propagators not only are required for covariance of relativistic theories; for electron Compton scattering they generate the entire amplitude in the low-energy limit as well. Now we compute the unpolarized electron Compton cross section by first summing over the final and averaging over the initial electron spins according to I T}7 | 2 - » \ Tr[(/ + m)e'* • M • e{p + m)e' • M • s*]. (11.45a) Since M„v as given by (11.42) is composed of a product of three y-matrices, the trace in (11.45a) contains eight y-matrices each for the direct, exchange, and cross terms. In general this amounts to 315 separate dot products in the traces of (11.45a). To reduce the labor, we follow Feynman (e.g. 1961b) and specialize again to the transverse gauge s 0 = 0, so that e p = e ' * p = 0in the lab frame. The trace algebra then leads to (see Problem 11.5) 177)v I2 ^ 4e4 | |E'* - E I 2 - - ^ — ) .
(11.45b)
The Compton cross section in the lab frame for unpolarized electrons is therefore daunpo1
dnL
|T/i
-\SnmwJ
' ~U/ ° '
2 +4i«
'
CO
(11.45c) which is called the Klein-Nishina formula. Clearly at low energies with on' -> oo, (11.45c) assumes the spinless Thomson form (11.40). Electron Pair Annihilation. Consider now the crossed version of electron Compton scattering, that of electron pair annihilation e~ + e+ -+2y, also depicted by the Feynman diagrams of Figure 11.8, but with fcin -»• kout = — k and p^,ut -+ p'in = —p'. In effect, p' in Figure 11.8 then represents the momentum of a negative-energy electron state with p[n the momentum of an incoming positron. The Feynman rules for such a CPT configuration give 7 / 7 = -e*tyM'"Vv*. with =
_el [y,tf-f+m)y¥ \
K
+
y ^ - r + m)yJ U
\
obtained directly from (11.42). Then the Compton crossing-symmetry properties for k and p, (11.39) and (11.43) respectively, become statements of
Bremsstrahlung and Pair Production 219
Bose and generalized Fermi statistics when expressed in the annihilation channel (see Problem 11.6). To compute the unpolarized annihilation cross section, we need not perform another lengthy trace calculation for (11.46). Instead, we use the fact that the Compton trace (11.45b) is invariant in frames where £0 = 0 and e* • p = e'* • p = 0, which includes the annihilation-channel lab frame p = 0 witht = 2m(F + m),sm = — 2mcb, um = —2ma>'. The differential cross section is then (see Problem 11.6) ^unpol
dQL
^ J
— + -, + 2 - 4(E'* • s*)2
Sp'(E' + m) a>
to
(11.47)
with m + E' = a) + a>\ p' = k + k'. A useful application of the pair annihilation cross section is for the decay of positronium e~ + e+ from the singlet s-state. Recall from (6.36) that for a fermion-antifermion system to decay to an rjc = 1 state, we must have I + S = even integer, as is the case for (e~ + e+)ls-+2y. The total cross section obtained from (11.47) in the nonrelativistic limit for unpolarized photons averaged over solid angle (i.e., | e'* • e | 2 -> 0) is a3nn -*rcr2,/v', where we have included an extra factor of j in order not to double-count one of the photons when integrating over the entire solid angle 47t (a Feynman rule). This cross section is then converted to the decay rate in the singlet n = 1 state T(1S) = 41 iAis(0)|2^'o'ann> where the factor 41 ^ ls (0) | 2 is introduced to insure that only the 1S "parapositronium" state (1 of 4 possible spin configurations) is included with a nonvanishing probability overlap at the e~e+ reduced-mass origin. Borrowing the hydrogen-atom result (5.148) with m->jm, |i// ls (0)| 2 = (wa)3/87t for n = 1, the decay rate becomes r( 1 S) = 4|^ls(0)|2D'<7ann = |ma 5 * 5 x 1(T 6 eV, 1
1
or T( S) = ft/T( S) ~ 1.23 x 1 0 ment (also see Problem 11.6).
-10
(11.48)
sec, a result which agrees with experi-
ll.E Bremsstrahlung and Pair Production Bremsstrahlung is the German word for "braking" radiation emitted by charged particles (electrons) in the presence of an external field. Pair production is the crossed process, whereby a photon "splits up" into an electronpositron pair, also in the presence of an external field. Such a field must be present in both cases because we know that a free electron cannot interact solely with a free photon by energy-momentum conservation; i.e., p' + p = k and p'2 = p2 = m2 require co = 0. This is consistent with the classical statement that only an accelerated electron can radiate since dccaj2. Electron Bremsstrahlung. Consider first electron Bremsstrahlung in the static Coulomb field of a heavy nucleus with charge Ze, depicted by the Feynman graphs of Figure 11.9. Because overall three-momentum is not conserved
220 Lowest-Order Electromagnetic Interactions
•x
+
Figure 11.9 Lowest-order Bremsstrahlung diagrams. (the nuclear recoil is hot observed), this process resembles the Coulomb scattering of Section 11.A in that energy conservation is embedded in the structure of Aexi(q) as given by (11.3), where q = p' + k — p in this case. The Feynman rules yield for the two graphs of Figure 11.9 i{p" + fi + m) ey • Aexi(q)up (p' + kf - m2 + is
S}7=(-i)2u„.eery +
»(jf - Ji + m)
(-i)2up.eyAe«(q)
(P ~ k)2
m + ie
e e * • yup-
(11.49) v
II
Writing S}° = id(E' + a> — E)s*u„M up, 2
M„
Ze
(y„#-2p;)y0 2p' -k
the M-function for this process is +y0(^V
+ 2/v) 2p • k
(11.50)
This M^ is in fact gauge invariant, k • M = 0, as can be verified with a little Dirac algebra (Problem 11.7). Soft-Photon Theorem. For soft-photon emission k -> 0, the 1/t terms in the numerator of (11.50) vanish relative to the p'^ and p^ terms, and then y0 -> 1 in the latter cases. The structure of the resulting soft-photon amplitude, M,(k)-
->e
PK
/V
p' • k
p • k
M,
(11.51)
is the same as for spinless charged particles (with vertices 2pJ, + /c„ -»2p^, 2pM — k,, -> 2p„). In fact (11.51) is valid for all charged particles, regardless of the spin or hadron or lepton structure of the process. For the case considered, M is the Coulomb amplitude M->Z/|q| 2 , but in general in (11.51) it represents the M-function for the radiative process under consideration, though without the external photon leg. That is, the amplitude for the soft-photon process A -»• B + y (k - • 0) is given by (11.51) with M representing the transition A - » B . It is important to note that (11.51) is gauge invariant and singular like 0(\/k). In fact, it is the most singular part of any such radiative amplitude expanded as a power series in the photon momentum k. It turns out that even the next 0(k°) term in such an expansion can be determined by gauge invariance in a model-independent way (i.e., for hadron and lepton
Bremsstrahlung and Pair Production
221
processes). The sum total of these above general observations is known as the soft-photon theorem (Low 1958). It is the prototype of similar soft-pion and soft-graviton theorems to be discussed later. Feynman Integrals. Returning to the specific Bremsstrahlung process of Figure 11.9, i.e., M -*• M0""1 in (11.51), the soft-photon unpolarized differential cross section has the general form dasotlBrem
_daCoui
da
So
e
m^-^-i
<»->
• *
2
To obtain (11.52) from (11.51) we have used the rule for photon polarization sums, YJ I £ • M | 2 -• — M • M* (see Problem 10.6). The covariant integral in (11.52) over photon phase space is necessary for unobserved soft-photon emission. This integral behaves like j dco/a>, which diverges logarithmically as a) -*• 0. Experimental detectors, however, cannot measure an a> = 0 photon; we therefore limit the photon integral in (11.52) to comin < a) < o)max, so that dasoflBrem dO.
/d
—2]
dQ,
u>dai\dak CUmin
2p' • p
[p'-kp-k
m2
(p'-kf
m2
(p-kf\
(11.53)
As k -»0, the initial and final electron energies and velocities become the same, £' -»E, v' -> v = p/E, and the last two terms in the solid-angle integral of (11.53) have the same value, rdQk m2/E2 _ m2 1 r 1 dz 2 2 J 4TT (1 - v • £) " E 2J_ X (\~vz)2
_ m2 2 "£ (l-v2) ~
(
'
The first solid-angle integral in (11.53) is more difficult to evaluate and requires the clever trick introduced by Feynman to combine denominators: l r 1 ab = l
dx [ax + ft(l - x)] 2 "
(1L55)
Then we may write dQk 1 (-1 ^ f dnk 1 dX An (1 - v' • £)(1 - £ • v) oJ0 ' 4TC' {1 - £ • [y'x + v(l - x)]}2 1 dx o i - |v'x + v ( l - x ) | 2 '
Hi
(11.56) The last integral can be evaluated easily for electron energies in the extreme nonrelativistic limit (NR) and also in the extreme relativistic limit (ER).
222 Lowest-Order Electromagnetic Interactions
+
Figure 11.10 Lowest-order pair-production diagrams. Combining (11.56) and (11.54) with (11.53), we finally obtain for cos 0 = v' • v and q2 = (p' - pf « - 4 £ 2 sin2 jd a soft version of the Bethe-Heitler formula, do*0"8™ /d<jCoul _ 2a c u lfi>2 sin2 # + 0(v4), 0g dQ / dQ. ~ n comin\\og(-q2/m2)-l + 0(m2/q2),
NR ER (11.57)
While this result appears to be finite, we must still contend with the limit comin -> 0. It turns out that there is a higher-order photon radiative correction term which also behaves like log comin. In fact, the sum of the latter and (11.57) is independent of such divergent log comin terms. This result, called the infrared theorem, will be discussed in detail in Chapter 15. Electron Pair Production. The Feynman diagrams for production of an electron-positron pair in the presence of a Coulomb field of a heavy nucleus are shown in Figure 11.10. The S-matrix element and M-function for this process can be immediately obtained from the Bremsstrahlung amplitudes (11.49) and (11.50) by crossing the momenta koul -» —kin and pin -> —pout. A calculation not unlike the one just described, but with a more complicated final-state phase space, leads to the pair-production total cross section for high-energy electrons produced near a charged (Z) nucleus, which is [see e.g. Heitler (1954)] ZV m2
6N§4)-
<-)
This result is to be compared with the total electron Compton cross section, the solid-angle integral of (11.45c) for Z atomic electrons, I Z
Comp = ( 3 w Z
8co
K4)
o) m, (11.59) ^Thorn,
0)>m,
along with the nonrelativistic photoelectric cross section of Section 9.C,
Electromagnetic Interactions of Hadrons 223
photoelectric effect is the dominant absorption process. For moderate a>, roughly co ~ m ~ ^ MeV, electron Compton scattering controls the absorption process. Finally, at high co, co > 2m, pair production takes over as the dominant mechanism for the absorption of photons. Electron Bremsstrahlung and pair production off a free electron (rather than a heavy nucleus) can also be worked out in a similar fashion. The computations are extremely messy, however, due to the final-state threebody phase space [see e.g. Berestetskii et al. (1971)].
ll.F Electromagnetic Interactions of Hadrons In the earlier sections of this chapter we have not bothered to compare with experiment the predictions for the electromagnetic interactions of "structureless" leptons (QED) as calculated from lowest-order Feynman diagrams. This is because the theory agrees with experiment so well that we may regard the minimally coupled electron-photon vertex as "exact" and then use it as a probe to investigate the electromagnetic structure of strongly interacting hadrons, such as protons, neutrons, pions, etc. Elastic Form Factors. Recall from Section 5.D the general form of the electromagnetic current for spin-^ particles,
(11.60)
2
with A = p' — p and A = t. It is sometimes convenient to re-express the charge and magnetic form factors, FA) and F2(t) respectively, in terms of the Sachs helicity-type form factors (see Problem 5.8) GE(t) = FA) + - ^ FM
GM(t) = F^t) + FA).
(11-61)
To study the form factors of nucleons, consider the covariant Feynman graph for the lowest-order electromagnetic scattering of light structureless electrons off heavy target nucleons (protons or neutrons) as shown in Figure 11.11. The "fuzzy ball" at the nucleon vertex represents strong interactions
Figure 11.11 Electron-nucleon elastic scattering and nucleon form factors.
224 Lowest-Order Electromagnetic Interactions
to all orders, i.e., a charged cloud of virtual n±, etc., which give rise to the nucleon form factors. The Feynman rules for Figure 11.11 are similar to those for Figure 11.3, but with the nucleon electromagnetic current (11.60) replacing the structureless spin-j vertex eiip. y„up in (11.18). Then a little Dirac trace algebra modifies (11.19) to the unpolarized-spin sum (see Problem 11.8) M ft I
4e 4 ~*~a
K^4
> - m2)2 + st) + G2M \e
(11.62)
which obviously reduces to (11.19b) for Fu GM -> 1, F2 ->0, and fi2/m% -> 0. Given (11.62), it is straightforward from (11.20) to compute the unpolarized Rosenbluth differential cross section in the lab frame of the nucleon, J—Rosenbl
F\1 -~F\
Ami
2 -Pi 2m tan #
l
(11.63)
where aNS(9) is the no-structure Mott cross section defined in (11.20). To extract Fu F2 or GE, GM from the single formula (11.63), one analyzes the recoil-electron data so as to split off the tan 2 \Q term in (11.63) for fixed t = q2 (see Problem 11.8). The result is the dipole fit (1 - t/m2)~2 to the shape of the Sachs form factors G|, G&, G"M for mv zz 840 MeV, as discussed in Section 8.E. On the basis of such early eN scattering data for spacelike t < 0, it was possible to predict the existence of the p and a> vector mesons which resonate in the nucleon form factors for timelike t > 0 (Frazer and Fulco 1960). The above analysis, however, only scratches the surface of the electromagnetic structure of hadrons. Much more has been recently learned from the "deep inelastic" region of eN scattering, involving inelastic form factors referred to as structure functions when the final-state hadron is not detected (see Problem 11.8). For this kinematic configuration, the photon can probe deeply into the electromagnetic structure of the nucleon. It now appears that there is a hadron substructure called partons (Feynman 1969) which in fact like leptons, have no structure (i.e., form factor), to lowest order in e. What is even more exciting is that these partons are most probably a manifestation of fundamental particles called quarks (Gell-Mann 1964b, Zweig 1964), "bare" hadrons which appear to underlie the spectroscopic internal symmetry (SU(3)) patterns of the "hadron zoo." But further discussion of these topics is beyond the scope of this text. Radiative Decays of Hadrons. Another manner in which the electromagnetic structure of hadrons is revealed is through radiative photon decays. For example, the first order (in e) radiative hyperon decay Z° -> A + y(k) is driven by a magnetic transition-matrix element
s}°? = -aA(pfl)jrfi,
Jtrfi = — e - ^ ~ n^Vu^*.
(n.64)
Electromagnetic Interactions of Hadrons 225 This phenomenological hamiltonian density is the only form allowed by Lorentz and gauge invariance (see Problem 11.9). The effective decay coupling constant KXA is the Z-A-transition analog of the proton and neutron anomalous magnetic moments, KP = 1.79, K„= —1.91; so one might expect that | K IA | ~ 1 (consistent with the higher-symmetry SU(3) prediction). Given (11.64), it is straightforward to apply the two-body decay-rate formula (10.17b) for unpolarized S° and A to find (see Problem 11.9),
If in fact | K IA | ~ 1, then (11.65) implies TEO = fc/r(S° -> Ay) ~ 1(T 19 sec. Even though all three particles in this decay are neutral, the E° lifetime was recently measured to be T I0 * 0.6 x 10~ 19 sec, about what we have estimated from (11.65). As an aside, one might expect that the lepton em decay /i -* ey could proceed through a decay rate similar to (11.65), with T,, ~ 10" 20 sec (see Problem 11.9). While these leptons are charged, so that this decay can be easily detected, it turns out not to be seen to the level x{n -* ey) > 102 sec. Since according to quantum mechanics, anything that can happen does happen (with a predictable probability), some dynamical mechanism must be suppressing this decay. A similar circumstance occurs for other \i-e scattering transitions and so we refer to this dynamical /x-e suppression as a "selection rule". A second kind of radiative hadron decay is typified by the second-order photon decay n° ->2y. Lorentz and gauge invariance require a phenomenological matrix element of the form [q(n°) -> k + k']
sy? = - i3*{pfiprft,
jrfi = Fny^t{k')e*{ky^Kkfi.
(11.66)
The corresponding unpolarized decay rate (see Problem 11.10) lV^2y)=F2y7^|,
(11.67)
plus the experimental lifetime x(7t°) « 0.83 x 10" 16 sec and branching ratio 99% [i.e., r(7t° -> 2y) « 8 eV] lead to the hadron scale of \Fnyy/e2\ ~ 0.037m ~ 1 , where we have removed a factor of e2 from F„yy corresponding to the two photons in the final state. While there is no absolute measure of Fnyy/e2, as there was for /crA in (11.64), it would appear that the extracted Fnyy scale is anomalously small for a strong vertex. This is not the case, however, because this n° -> 2y scale will be related to the radiative a> -> ny scale in Section 12.E. The latter decay structure is in fact similar to (11.66)—see Problem 11.10. We shall return to the n° -> 2y amplitude in a different context in Section 15.B.
226
Lowest-Order Electromagnetic Interactions
ll.G Static Electromagnetic Potentials A covariant photon exchange between elastically scattered particles leads naturally to the concept of a force or, equivalently, a static electromagnetic potential. In Section 5.D we worked out the corrections to the Coulomb ep hydrogen atom due to a relativistic Dirac electron, but neglected the recoil of the proton nucleus and both the proton and electron electromagnetic form factors. We now want to correct for such effects. General Static-Potential Prescriptions. Given a relativistic Feynman diagram and the corresponding amplitude Sfi = i<54(P/i)T}°vfbr a photon exchanged between elastically scattered charged particles, we may extract a static electromagnetic potential via V(r) = - (AmnY* \ & ^ ' rT}7(NR).
(11.68)
Here T}7(NR) is the nonrelativistic limit of Tfi, m and /i are the elastically scattered masses at each vertex (required because the states are normalized covariantly) and A = p' — p = q — q' is the three-momentum transferred from the q to the p vertex. For spinless particle recoil scattering of Figure 11.2, the prescription (11.68) is the reverse of the procedure we used in Section 10.D to build up Feynman diagrams from covariant "sidewise-force propagation". More specifically, given (11.15), we have
since t = A2, - A2 = (£' - E)2 - A2 -> - A2 as £', E -»• m, A2 <^ m. Keeping only the leading A2 term in (11.69) then leads to the usual Coulomb potential from the prescription (11.68), , „,
e2 Amu r _,, A e ' A ' r
e2 1
a
w-i^r^-ir-*-,-?
lt t
_n .
,n 70a)
-
while the constant term in (11.69) has a delta-function Fourier transform AF(r)= - - — | > A e i A 4m/i J
r
= _ ^ ! ^ .
(11.70b)
mu
The total potential is then K0(r) + AF(r), with (11.70b) corresponding to an "average" of the Zitterbewegung-type anharmonic expansion of V(t + Ar) in (5.138). This "average" is the correct relativistic way to account for center-ofmass motion, rather than by a reduced nonrelativistic mass and weighted Zitterbewegung expansions at each vertex. In either case we need not worry about (11.70b) destroying the classical significance of the Coulomb interaction, because (11.70b) has physical matrix elements only for quantum states which have a nonvanishing probability at the origin (i.e., for / = 0 s-states).
Static Electromagnetic Potentials 227 Form-Factor Corrections. Further quantum corrections to static electromagnetic potentials arise from the momentum dependence of the form factors. For a spinless particle with structure (e.g., a hadron), say at the p vertex as in Figure 11.11, (11.69) becomes modified by a form factor F(t) = 1 + tF'(0) + •••-> 1 - A2F'(0) + • • •, where F(0) = dF(t)/dt |,. 0 . Then the leading correction to (11.70) is AFF(r) = -L^W
F ( 0 ) J d3A e i A r = -47raF(0>53(r),
(11.71)
which is in fact equivalent to the charge-radius correction to nonrelativistic nuclear form factors discussed in Section 8.E, with R2ms = 6F'(0). Further complications arise if one of the particles (mass m) has spin-j and structure. Folding in the electromagnetic current (11.60) with the prescription (11.68) leads to the static potential for positively charged particles (see Problem 11.11), e2 r
eiAr I
f i M o + Fi(t)io0v
^
"p
/NR
(11.72a) Taking, in the nonrelativistic limit, F^r)-* 1 — A2Fi(0), F2(t)^K, Up- 7o UP -* 2m, iip. ia0v Avup ->• — A2 + ia • A x P, where A = p' — p and 2P = p' + p, (11.72a) then becomes V(r) = -r-4na
^(0) + ^ ^ ( r ) - ^
cr • L,
(11.72b)
where we recognize the factor in braces in the s-wave central part of the potential as G'M(0), and the last term as a type of spin-orbit coupling with L = — ir x V. This potential will be needed to calculate the Lamb shift (see Section 15.E) due to the "very slight" form factors of the structureless electron. Finally, if both particles have spin-j and structure, as in proton-proton Moller scattering, the resulting static potential has a Coulomb part, along with an s-wave central part, a spin-orbit part, and a spin-spin part (see Problem 11.11). In fact, the measured spin-orbit term for proton-proton scattering suggested not only a (vector) photon electromagnetic exchange, but also a vector-meson strong-interaction exchange (Breit 1960, Sakurai 1960). But more about such strong interactions in the next chapter. For further reading on lowest-order electromagnetic graphs, see e.g. Feynman (1949, 1961a,b), Heitler (1954), Thirring (1958), Kallen (1958, 1972), Mandl (1959), Drell and Zachariasen (1961), Schweber (1961), Bjorken and Drell (1964), Akhiezer and Berestetskii (1965), Muirhead (1965), Gasiorowicz (1966), Sakurai (1967), Berestetskii et al. (1971), and Jauch and Rohrlich (1976).
CHAPTER 12
Low-Energy Strong Interactions
Because the strong-interaction coupling constants are so large, a perturbation theory of lowest-order strong-interaction Feynman diagrams does not always make sense. Yet if interpreted in the right manner (i.e., at low energies), such graphs reveal a great deal about the qualitative and sometimes quantitative nature of hadron (strong-interaction) physics. They are also the beginnings of the dynamical scheme of strong-interaction dispersion theory, a theory that we will pursue in greater detail in Chapter 15. Here we begin by investigating the basic pion-exchange, short-range Yukawa force. Next we develop the concept of isospin, isospin conservation, and isospin projection operators, which we fold into the Feynman rules. Then we use simple Feynman graphs to explain the essential features of the long-range part of the nucleon-nucleon force, and low-energy scattering. Feynman graphs are also used to explain low-energy pion-nucleon scattering and the dynamical effect of the 33 resonance, A(1232). Finally, we search for conserved and approximately conserved currents for hadronic interactions. Conserved isotopic vector currents are investigated in the context of the vector-meson dominance model, and partially conserved, isotopic axial-vector currents are probed in the context of pion pole dominance.
12.A Yukawa Force By 1935 the only strongly interacting hadrons detected were the proton and neutron of mass ~ 939 MeV. Then Yukawa (1935) predicted the existence of a light spinless meson with mass about 200 MeV which would mediate the 228
Yukawa Force 229
strong force between nucleons in the same way that the photon mediates the electromagnetic force between charged particles. We devote this section to a motivation of the ideas underlying the notion of a force created by a massive-particle exchange. Range of Nuclear Force. The Coulomb potential 1/r can be felt at very great distances, a fact expressed by the infinite Rutherford cross section (11.2) when integrated over all solid angles. We know that this behavior is connected with the massless photon, for if my =/= 0, then the 1/r potential would fall off much more quickly as e~myr/r (recall the discussion in Section 8.D), i.e., with range Rem oc m~' -> co as my -* 0. Our observations tell us, however, that the strong force between nucleons must be very short ranged, effectively damped to zero within a nuclear diameter or two. This corresponds to a range of about a fermi, 1 fm = 1 x 10" 13 cm. Assuming then a "Yukawa-type" nuclear potential V{r)cc e~m"r/r, the range can be associated with Rsi ocm~l, where the pi meson (pion) is the lightest-mass hadron analog of the electromagnetic photon exchange. Since he ss 200 MeV-fm, in our units (h = c = 1) we see that mn ~ hc/R ~ (200 MeV-fm)/(l fm) = 200 MeV, corresponding to Yukawa's predicted meson. (Actually, the first "meson" to be discovered in this mass range was the muon in 1941, a weakly interacting lepton with half-integer spin. The sought-after pion was finally found in 1947, with mn K 140 MeV.) An alternative but equivalent interpretation of this exchange-mass force is through the notion of a virtual pion cloud surrounding each nucleon, N -* N + n. The uncertainty principle says that a mass-energy AE ~ mn c2 can escape the nucleon only for times At ~ h/AE ~ h/mnc2. If two such nucleons come within the range R ~ c At ~ h/mn c, then a pion from one of the clouds can be "captured" by the other nucleon cloud without violating the uncertainty principle. For mn c2 = 140 MeV, we have R ~ hc/mn c2 ~ 1.4 fm, the approximate range of the strong force. Evidently, the heavier hadrons in the now detected elementary particle "zoo" (see Appendix HI) contribute to shorter-range and strangenesschanging components of the strong force. We shall, however, concentrate upon the longer-range (pion-exchange) aspects of the nuclear force. Sign of Nuclear Force. To formulate the mass-range and uncertaintyprinciple-cloud analogies in a quantitative fashion, we construct a covariant diagram as shown in Figure 12.1: a spinless pion propagator linking two identical nucleons, with a coupling strength g at each vertex (and assume the pion is a scalar rather than a pseudoscalar meson for the time being in order
P
q
Figure 12.1 Lowest-order spin-0 exchange force between two nucleons.
230 Low-Energy Strong Interactions to simplify the discussion). Then sc;? = id*(pfi)Tcfy S}7 = (-i)2guP'Up Tcov fi
the Feynman
rules give for
i guq.uqd*(Pfi), -ml + ie [A 2
g2upupuq,uq ~ ml-A2
(12.1)
where again we drop the ie in Tfi, since a space-like A2 prevents the denominator from vanishing. According to the general relation (11.68) between a local potential and the nonrelativistic limit of a covariant amplitude, in this case r^(NR)-(2mJV)V(A2 + m2r1, we may write ()
4m2J
A2+m2
4n
r
'
U
^
We observe that this nuclear potential is attractive, V < 0. In fact, as noted earlier, even-integer spin exchanges (s = 0, 2, 4,...) lead to fundamentally attractive forces, while odd-integer spin exchanges, like the spin-1 photon, generate fundamentally repulsive potentials, as in (11.70a). Turning the argument around, since the nuclear force must overcome the Coulomb repulsion of protons in a nucleus, it must be fundamentally attractive. Thus the spin of the pion is even. In fact, we have seen that the detailed balance relation (6.76) for n+d*-+pp leads to 2sn + 1 ~ 1 from experiment, or sn+ = 0. The decay n° -* 2y likewise requires sno = 0. Strength of Nuclear Force. If such a potential strength as (12.2) binds two nucleons at nuclear distances R oc m~1, then mng2/4n must be large enough to overcome the "uncertainty-principle repulsion" at such short distances: Ap Ar ~ h, or p~R~1~mn, p2/mN ~ m2 /mN ~ 20 MeV. That is, 2 2 m„g /4n ~ 20 MeV requires g /4n ~ j . We know, however, that the pion is a pseudoscalar 0" particle (recall the discussion in Section 6.B). Thus our estimate for (12.2) should include p-wave and not s-wave couplings at each vertex in order to conserve parity. That is, g2/4n in (12.2) should be suppressed by (p/m)2 ~ ml/ml ~ j$, or equ'ivalently the coupling must be greatly enhanced to g2/4n ~ 10. Thus, once again we warn the reader that such a large coupling means the lowest-order Feynman graphs do not always correspond to physics. When and why is the kind of intuition that we shall try to develop in this chapter.
12.B Isospin A mass-vs.-spin "spectroscopic" plot of the 100 or so particles in the hadron zoo reveals a Zeeman-type splitting with, for example, the states (n+,n°, n~), {K+,K°\(p+, p°, p~), (n, p), (S + , 2°, 2T) grouped close together with the
Isospin 231
7T l
7T
(a)
(bl
7T
7T ocg^opp g x o n n + 9 T * P I I
(c) Figure 12.2 Lowest-order charge-independence force diagrams.
same spin and almost the same mass. This suggests that there exists another symmetry operator like angular momentum, called isospin I, which is conserved by strong interactions but split by the weaker electromagnetic interactions according to the "z-component" I3, somehow related to the charge Q. Our discussion of isospin will center on its implications for the nucleonnucleon force and pion-nucleon scattering. Charge Symmetry and Independence of Nuclear Forces. The proton and neutron, although having different charge (Qp = 1, Qn = 0 in units of e), empirically interact strongly with about the same strength. An example of this is the similar nuclear energy levels of Li7 and Be7, mirror nuclei differing by one neutron or proton. This gives rise to an extra nn versus pp bond but with the same number of np bonds. We therefore deduce that nn forces are about the same as pp forces, usually referred to as charge symmetry of nuclear forces. A similar consideration of the nuclear energy levels of (B12, C 12 , N 1 2 ) or (C 14 , N 1 4 , O 14 ) requires the np bond also to be nearly the same as the nn and pp bonds. This is called charge independence of nuclear forces. It is further suggested by the approximate equivalence of the pp, nn, and singlet np s-wave scattering lengths of ~ 20 fm (once the large Coulomb repulsion effect is removed from the experimental a o(pp)—we s n a ll return to the NN scattering lengths later). In terms of lowest-order Feynman graphs, these three forces are represented in Figure 12.2 with gn+p„ = gKnp (by CPT invariance). The charge independence of nuclear forces then requires the S-matrix elements in Figure 12.2 to be equal, Spp = S„n = Spn (again when the electromagnetic effects are ignored or "turned off"), so that QnOpp = QnOnn 9n0pp
:
=
9n0pp9it0„„ + 9n + pm
~9n0„n — 9-K + pnl\l 2- — 9
(12.3a)
(12.3b)
232 Low-Energy Strong Interactions Even though we expect g to be large, g ~ 10, the relations (12.3b), being true in every order of perturbation theory, lead to Sl°lp = Sl™ = Sl°*, once electromagnetic and bound-state effects are turned off. This equality is roughly consistent with data. Isospin Conservation. In order to link charge independence to a conservation law, we note that the ratios in (12.3b) are reminiscent of the ClebschGordan angular-momentum coefficients (see Appendix II), <&|±1, ¥» - : = 1: - l : ^ / 5 (12.4) So first define an "isospinor" nucleon with /f = \, I3 = —\ and an "isovector" pion as l\* = 1, If = 0, 75" = - 1, i.e.,
N=(Pn),
*=(H
(12.5)
The pion charge states can be expressed in terms of the "cartesian states" nh where we choose the simple phase convention (not Condon-Shortley) JV = (p, h), n
±
=
7
^ ^ ,
7r° = 7t3,
(12.6)
and the antiparticle isospinors are worked out in Problem 12.1. Next define Pauli-type 2 x 2 matrices T in "isospin space":
Then construct an isospin "scalar" NxN • n, which can be expanded in terms of charge states as NxN • n = Nx^Nii! + NT2Nn2 + +
= Jl Nx+Nn
NT3NK3
+ Jl Nx-Nn-
+ Nx3Nn°
= y/l pnn+ + Jl npn' + ppn° - nnn°,
(12.8a) (12.8b)
where T ± = ^(x1 ± iz2) are isospin raising and lowering matrices. The relative signs in (12.8b) are the same as those of the Clebsch-Gordan ratios (12.4), so these isospin constructs do combine like angular momenta. Since (12.8b) is equivalent to the charge-independence statement (12.3b), it is clear that if we express the 7t AW strong coupling as proportional to the isoscalar (12.8), then it is reasonable to assume that such an isoscalar stronginteraction hamiltonian, H st , is invariant under "rotations" in isospin space. Put another way, the corresponding infinitesimal generator, the isovector
Isospin 233 operator I, is conserved by the strong interactions, i.e., [I, Hst] = 0.
(12.9)
This is a nontrivial statement, stronger than charge independence, because it predicts, for example, the selection rules d + d-/*a. + n° or A »Arc0 as strong interactions. That is, isospin cannot be conserved for these reactions, since the deuteron, the a-particle (He 4+ + ), and the lambda particle are all isosinglets, Id = Ix — / A = 0, while In = 1. Isospin Diagrams. In order to fold in isospin into the Feynman-diagram approach, we first tabulate Feynman-type rules for isospin vertices and isospin projection operators (the isospin analog of propagator numerators). These rules are listed in Table 12.1. One need only make sure that all isospin indices are saturated for a given strong-interaction Feynman graph in order to insure isospin invariance. Table 12.1 Isospin Diagram Rules Vertex / = 0, 0, 0
/ = i, i 0 1
1 3i
Rule
Projection Op.
Rule
l 1 (spinor)
lo
1 1 (spinor)
T1'
ij
Sij ieiik
I = 1', l y , 0
/ = r, v,
I*
h i\ h ii
$(5ik5jl + 5"5ik)-$5ijdkl
The 1 = \ isospinor-vector, i]/', deserves some explanation (also see Problem 5.11). It is the analog of the J = f angular-momentum wave function (2.62), satisfying the subsidiary condition (2.63). The latter corresponds here to T'I//' = 0. For the same reason the / = f projection operator 1% = £ i/^ J obeys T'I'J = I^xj = 0. Its obvious form, normalized to I2 = J, is then /jjf = 8iJ — J T V , since t • t = 3 implies T'/^ = zj — %x • XTJ = 0. Finally, using the Pauli identity T'V = 5ij + ieiJkxk converts /^ to the form given in Table 12.1. Also note that the / = 2 tensor projection operator is traceless and symmetric in the appropriate indices, as required by the usual irreducibility conditions (2.60). See also Problem 12.1. As an application of these isospin rules, consider nN scattering in the three crossing-related channels: njN -> JI'JV,
s-channel, Is = j , f,
nlnJ,
t-channel, /, = 0, 1,
NN7tW-
j
n N,
u-channel, /„ = \, | .
Envisioning simple pole (propagator) diagrams in each of the three channels, we apply the rules of Table 12.1 to find the projection operators (again
234 Low-Energy Strong Interactions normalized to I2 = I and suppressing the isovector indices) It = ir'lTJ' = &iJ + ii£ ij V, ij
(12.10a)
Jk k
/f = 5*71'*^ = id ~ W ? , 1° = diJ,
1} = i£ijk'5k\k = ie iJ V,
7* = %z}W = %6ij - Wjk?k, ij
(12.10b) (12.10c)
ijk k
7* = ld + \h x . In the t-channel, 1, x 1„. = 2 + 1 + 0, with the maximum 7 = 2 states (ruled out at the nucleon vertex jN x %N. = 1 + 0) symmetric under the interchange of isospin indices, so that 7 = 1 and 7 = 0 are respectively odd and even for i •«—>_/. Another way of deriving the projection operators (12.10) is to write Is = jx + T, where T is the pion isovector, and then squaring, Is2 = Js(Js + l) = i + T-T + 2. For Js = i x T = - 2 , while for 7S = f, x • T = 1. These conditions are equivalent to (12.10a). Such isospin techniques can also be applied to NN scattering for the three crossing-related channels N1N2-^ N\ N'2,
s-channel, 7S = 0, 1,
Ni # i -» N2 N'2,
t-channel, 7, = 0, 1,
Nt N2 -*• N\ N2,
u-channel, Iu = 0, 1.
Equivalently, the corresponding projection operators in the direct product isospin spaces can be found from Is = jxt + %x2, I2 = IS(IS + 1) = ^(3 + tj • x2), so that 7S = 0 requires x1 • x2 = — 3, and Is = 1 requires T i ' T2 = 1> etc - We are then led to (see Problem 12.1) Ils = |(3 + x, • x2),
7S° = i(l - x, • x2), 7,° = l j l 2 = 1,
1} = i ! • x2,
I°u = i(l + *i • t 2 ),
11 = |(3 - x, • x2).
(12.11a) (12.11b) (12.11c)
12.C One-Pion-Exchange Nucleon-Nucleon Force Having dispensed with the preliminaries, let us investigate in detail the implications of the simple pion-exchange Feynman diagram between two nucleons as depicted by Figure 12.3. nNN Effective Hamiltonian. Given the fact that the pion has ln = 1, sK = 0, P= - 1 [general designation 7 G (J p )C-» 1~(0~)+], the complete Lorentz scalar, isoscalar nNN hamiltonian density can be alternatively written as jenNN = gNy5x-nN = (f/mjNty5x
(12.12a) • nN.
(12.12b)
One-Pion-Exchange Nucleon-Nucleon Force 235 ir
Figure 12.3 One-pion-exchange nucleon-nucleon diagram. Both forms of Jtf are invariant under Lorentz and isospin transformations and under the discrete symmetries of C, P, T, and Y (hypercharge), consistent with our observations about strong interactions (recall Problem 6.4). On the nucleon and pion mass shells, the pseudovector coupling constant/is related to the pseudoscalar coupling constant g by f/mn = g/2mN [i.e., let id -+qn = p'N — pN, and then apply the free-particle Dirac equation to (12.12b)]. Present data constrain these dimensionless on-shell coupling constants to be g2/4n = 14.30 ± 0.08,/ 2 /4TC = 0.079 ± 0.001 (see, e.g., Nagels et al. 1976). Off mass shell, the/and g couplings differ, and one of our goals will be to determine which coupling is preferable for a given problem. One-Pion-Exchange Potential. The light-mass pion-exchange graph of Figure 12.3 controls the long-range part of the NN force. The Feynman rules for A = p' — p = q — q' and pseudoscalar coupling (12.12a) give
(12.13a) T}7 = 9%y^2uq
• u,rthlUp(n£
- A 2 )" 1 .
(12.13b)
This ^-channel (A2 = t) Moller-type pole graph has a crossed u-channel analog which we need not explicitly calculate once (12.13b) is properly anitsymmetrized. But more about this shortly. To extract a static nonrelativistic potential from (12.13b), we apply the general prescription (11.68) and make the nonrelativistic reductions [recall (5.66) and Problem 5.7] up. y5 up-^ia • (p' — p) and m2 — A2 -* m2 4- A2, to obtain 92 . Ami'1
_ f j 3 * _iA.r<*i • A
" y^'-h^^S*** 4nml
x20l-\a2-\
.
(12.14)
The form (12.14) should be understood as sandwiched between the product of two-component spinors, both in spin and isospin space, which we delete for clarity. Note that (12.14) is naturally expressed in terms of the pseudovector coupling constant even though we inputted the pseudoscalar couplings into (12.13) and (12.14). Had we instead used pseudovector couplings in (12.13), the same final result (12.14) would emerge in the nonrelativistic limit.
236 Low-Energy Strong Interactions It is customary to carry out the differentiation in (12.14), expressing V as a "central" potential Vc plus a "tensor" potential VT: V(r) = Vc(r)ai • a2x, • x2 + VT(r)Sl2Tl • x2, 2
m r
2
f e~ " 47i 3r
f t 47t \
3 mnr
(12.15a) m r
3 \ e~ " myr) sr
where the irreducible tensor-spinor operator is S12 = 3 ^ • ra2 ' f — ©! •
f2 dr r Vl=0{r) = -i— mn « - 11 MeV, 2
(12.17a)
471
ml \ dr r2Vsw(r) = - $ K U ) 3 K 0 .
(12.17b)
Equating (12.17a) to (12.17b) at the Compton-wavelength range R = m" 1 leads to V0 % 33 MeV. Given this range, we can independently compute the
One-Pion-Exchange Nucleon-Nucleon Force 237 minimum square-well depth necessary to bind the two nucleons in an s-wave configuration. This corresponds to solving the square-well bound-state condition (8.31) for a zero-energy bound state K = ^/m^ E = 0. That is, cot pminR = 0 implies pminR = n/2, or in
_ P2min _
*2
„ 50 MeV
Then at R = m~\ (12.18) requires V%in « 50 MeV. Since V0 « 33 MeV at this range, the single-pion-exchange s-wave potential (12.16), although attractive, does not appear strong enough to bind the nucleons. Folding in realistic wave functions and adjusting the range accordingly does not alter this conclusion. This is consistent with the fact that no 1S0 bound state [pp, nn, or (np)/ = J has been observed. We now believe, however, that this virtual state is very close to being bound, with | EB \ x 60-90 keV [see e.g. Elton (1959)]. Bound-State Deuteron. The deuteron is a spin-1 (J = 1), triplet (S = 1), and isoscalar singlet (7 = 0) bound state of n and p. We may inquire how many of the measured properties of the deuteron—its binding energy EB = —2.2247 MeV, electric quadrupole moment Q = 2.86 x 10" 27 cm2, and magnetic dipole moment fid = 0.85735e/2mp—can be explained by the single-pion-exchange Feynman diagram of Figure 12.3. To begin with, a positive electric quadrupole moment requires the existence of a long-range attractive tensor force, as predicted by single-pion exchange. That is, defining the deuteron wave function as iAdeut(r) = ( V r ) x [u(r) + w(r) N /|S 12 ]xi, where Xi iS the triplet spin state (f| -I- |T)/v2> the electric quadrupole moment is, for J dr (u2 + w2) = 1 (see Problem 12.2), Q = V?zz ^ =
1 f J__2I./.
2 \drr2\^/..\deM12/1^2 {r)\2{^2-r„2\ )
-
20 i 0
dr r 2 ( ^ 8 uw — w2).
(12.19)
Now solving for the wave functions u and w from coupled nonrelativistic Schrodinger equations for the single-pion-exchange potentials (12.15) is a nontrivial business. For Q, however, the factor of r2 in the integral in (12.19) justifies using the asymptotic forms for long-range pion exchange [replacing mn by K in (12.15b)], u{r) = sfmnAse-Kr,
w(r) = ^/m~„ADe-"[l
+ 3(KT)- J
+3(jcr)" 2 ], (12.20)
over a wide range of r > m~l x 1.4 fm, where K _ 1 = {mNEB)^ % 4.3 fm « 3m~1. Substituting (12.20) into (12.19) then constrains the product ASAD « 0.014 (Problem 12.2). On the other hand, we know that the shallowbound-state deuteron generates a 3S1 s-wave effective range (8.27), so that
238 Low-Energy Strong Interactions
for a0 w — 5.40 fm we have 3re « (2/K)[1 + (a0 K) _ *] « 1.74 fm. This in turn fixes the normalization, Al + Af> « 1.10 (see Problem 12.2). Then combining these two results we obtain the normalization ratio AD/AS « 0.013—if the pion-exchange force controls the long-range part of the nuclear force. A more accurate calculation gives AD/AS « 0.025 [see e.g., Iwadare et al. (1956)]. The deuteron magnetic moment also has a bearing on the D/S ratio. The deviation of the dimensionless nd = 0.8573 from the sum \ip + n„ = 2.7928 - 1.9130 = 0.8798 is due mainly to the orbital motion of the proton in the 3Di state as determined by the D-state probability PD = J dr w2. A nonrelativistic vector-model calculation (Problem 12.2) gives the estimate PD « 0.04. Virtual-meson corrections increase PD slightly, but in any case it is roughly consistent with the single-pion-exchange value of AD/AS ~ 0.02; the latter corresponds to w/u ~ £ at their peak values for r ~ 2 fm, as found from (12.19) and (12.20), so that PD ~ w2/u2 ~ 0.06. For the deuteron binding itself, we return to the s-wave square-well estimates (12.17) and (12.18). For an E = 0 s-wave bound state, the range of the potential is the same as the low-energy scattering, s-wave effective range, 3 re « 1.7 fm. For an E « —2.2 MeV bound state, however, the range of the potential must be greater than 1.7 fm, but less than the deuteron "radius" 1/JC « 4.3 fm. An effective-range fit to the 3Sl low-energy scattering data in fact finds [see e.g. Brown and Jackson (1976)] R « 2.2 fm as the square-well range of the 3S1 s-wave potential. This EB and range R in turn shift the minimum square-well depth for binding from (12.18) to (see Problem 12.2) Kg"" w
* . + | £ B | | 1 + 4 T ) ~ 2 1 + 1 1 * 3 2 MeV. 4mN R \ KR/
(12.21)
On the other hand, the volume estimate of the / = 0 pion central potential (12.17a) for a range R « 2.2 fm leads to the square-well depth of V0 * 33 x (mKR)~3 % 9 MeV, far short of (12.21). Since <Xi | S 12 | Xi) averages to zero, the contribution of the stronger tensor potential to the well depth will be suppressed, being proportional to the volume integral f. dr uwVT. The upshot is that much of the required deuteron well depth (12.21) must arise from quantum exchanges other than the single pseudoscalar pion. Two exchanged pions could resonate in an / = 0, JF = 0 + state (as perhaps inferred from nn phase shifts), called the a scalar meson (recall Problem 8.4). Since such an even-spin particle generates an attractive nucleon-nucleon force via the r-channel exchange NN -»• a -> NN, the shorter-range "fitted a" force is usually assumed as the missing 3S1 NN bond. The parameters required to bind the deuteron and explain other NN scattering constraints are daNN ~ 0JgnNN, ma ~ 550 MeV, and Ta ~ 500 MeV [see e.g. Erkelenz (1974)]. While such a neutral particle would be hard to detect in a resonance-formation experiment (since Ya ~ ma), its existence is also needed as a basis for the so-called "CT model", a field-theory model which underlies the notion of a partially conserved axial current (PCAC) to be discussed in Section 12.F.
One-Pion-Exchange Nucleon-Nucleon Force 239 AW Scattering Lengths. In Section 8.B we saw that a weak potential without bound states can account for the s-wave scattering length via the Born approximation (8.23). Incorrectly applying this approximation to the strong NN single-pion-exchange potentials (12.16) yields the s-wave scattering lengths from (8.23) (with m -»%m N being the reduced mass), a%"'n((1%So, , 33SSlX)=-m ) =-m N\NC As
expected,
a0(3si)
5fm
this
result
2 dr drrrtV V,= = L- {-^r^2 * 1 fm. (12.22a) l=0(r) 0(r) = J o 4n mn is far from reality, a 0 ( 1 So) ~ 20 fm,
-
Nevertheless (12.22a) does contain an element of truth when the effects of the 3St virtual and 'So bound states are accounted for. T o see this, first write the Born approximation for the full nonrelativistic NN scattering amplitude,
m*>--Z<*m*>-£w-Zt>+*-
which is consistent with (12.22a) for p 2 - » 0 . Then in order to modify (12.22a) to account for the bound or virtually bound state, replace V in (12.22b) by V + VG0 V + ••• and force the sum t o diverge like ± | E — EB | _ l as E -> EB. That is, first d r o p the cos 9 term in (12.22b) for s-waves, and then replace the denominator 2p 2 + m2. by + 2mN|p2/mN — EB\. Choose the sign of this latter expression to be positive for the virtual bound state, so t h a t / 0 > 0 for V < 0 as in (12.22b). F o r a shallow bound state, however, choose the sign to be negative as discussed in Section 8.B; i.e., for EB « —2.2 MeV, a0(3St) « — (mN | EB |)~ * « — 4.3 fm, not far from experiment. T o see this sign change in Feynman-diagram language, note that t- and u-channel particle exchanges generate the s-channel bound-state deuteron, so that a simpler scalar deuteron pole must be of the form Sfi = i5 4 (P / j )T}° v . with -ycov __ ifdpn fi ~m2d-s~
^
Udpn
/< j y
4mN(EB - p 2 / m N ) ' 2
2
2
JW
>
2
where in the CM frame, s = (Ep + E„) « 4p + 2(m + m ) and EB — md — mp-mnx -2.2 MeV. Then (12.23) requires T??, f< 0 for real bound states. Next, to see the effect of the real or virtual bound states upon the s-wave scattering lengths, let p 2 -»• 0 in these modified amplitudes with / = T}°?/SnW^a0 as p 2 -• 0 for cos 9 = 0. That is, convert (12.22a) to 3c \ ~. - f2 a o ( X S t ) K+4n2EB' 3
l
~ I 2 0 " 3 0 f m f o r ^ B = - 6 0 to - 9 0 keV, \ - 3 . 5 fm for E'B * - 2 . 2 MeV, (12.24)
where E"B and E'B are the singlet and triplet np "binding" energies, respectively. The approximate result (12.24) is in qualitative agreement with experiment, aoOSo) ~ 23.7 fm, a 0 ( 3 Si) ~ - 5 - 4 fm.
240
Low-Energy Strong Interactions
While (12.24) is only a plausible dynamical approximation, it is in the range of experiment, while (12.22a) is not. In any case, the primary purpose of the exercise is to demonstrate again that the single-pion-exchange Feynman diagram controls the long-range part of the NN force and also underlies the low-energy NN scattering lengths—properly interpreted and modified in both cases. Furthermore, the above method is a precursor of the more rigorous Chew-Low dynamical scheme for nN scattering, where the scattering lengths are modified not by s-channel shallow or virtual bound states, but by the nearby A(1232) resonance, also in the s-channel. We shall take up the Chew-Low model shortly. Other Single-Particle-Exchange Effects. Our present picture of the nuclear force is described by a nuclear potential which is controlled at large r > 3 fm by the attractive spin-zero single-isovector-pion exchange, and at r ~ 2 fm by a (fictitious?) attractive spin-zero isoscalar er-meson exchange with m„ ~ 550 MeV. For r ~ j fm the NN potential appears repulsive, corresponding to the spin-1 p and a> vector-meson exchanges, with m p,w ~ 780 MeV. These latter exchanges are similar in structure to the spin-1 photon exchange for the hydrogen atom or Moller scattering (see Section ll.G). In fact, the properties of the resulting (short-range) spin-orbit interaction are consistent with NN phase shifts and led Breit (1960) to postulate the existence of vector mesons. At even shorter distances, r ~ j fm, the spin-2 isoscalar/-meson exchange with mf x 1270 MeV (see Appendix III) presumably generates an additional attractive component of the 3Sl NN potential, etc. Finally, for large-angular-momentum NN scattering, /max ~ pR. Then we expect the phase shifts to be fitted by the long-range part of the NN force— again the one-pion-exchange Feynman diagram. Experimentally this is not only verified, but such data have been used to obtain a reasonable estimate for gnNN, i.e., g2/4n « 14 [see Moravcsik et al. (1959)], in good agreement with the more recent and accurate determination from nN scattering. Clearly then the NN force is a richly varied but not altogether mysterious beast once the notion of single-particle covariant (Feynman-diagram) exchanges is exploited in detail. The strategy is, the more complicated the force, the more one squeezes as much physics out of simple dynamical approximations (such as Feynman pole graphs) as is possible.
12.D Low-Energy nN Scattering Armed with this qualitative and even quantitative success in understanding limited aspects of the nuclear force and low-energy NN scattering primarily via the single pion exchange Feynman graph, we now apply similar techniques to low-energy nN scattering.
Low-Energy nN Scattering ,P'
P'
K,,
P
241
K ,, V
+ \
\
/P
Figure 12.4 Nucleon "pole" diagrams for pion-nucleon scattering.
Nucleon Pole Diagrams. The analog of the Compton diagrams of Figure 11.8 for the strong-interaction process nj(q) + N(p) -»n'(q') + N(p') is shown in Figure 12.4. The s-channel and w-channel nucleon poles (s = K2, u = K2) are added together by the Feynman rules because the crossed pions are bosons. For the case of pseudoscalar coupling Ny5 x • nN, the Feynman rules give
5}7 = (-0V"VT'y 5
i(£ + mN)
ml
is
gy5T\n^(Pfi) (12.25a)
+
(-i)2gnJup.xjy5
+ mN) W5JupitP(Pfi). u — ml + ie
Writing y5(fi + mN)y5 = p, + 4~mN = 4 and y 5 (£ + mjy)y5 = -# between free particle nucleon spinors and using T V = d'J + ieljkxk, we obtain the covariant amplitude S}°v = id*(Pf^Tf? m terms of the isotopic amplitudes r ( ± ) , where r<±) = 1 pole
Mi
+
-P-
(12.25b)
Here sm = s — ml, um = u — ml, and the isospin structure of / = \, \ scattering can be expressed in terms of the general nN decomposition T}°y = ni[diiP+) + ie'^T*^ - V-
( 12 - 26 )
Note that T(+> is even under the s-u crossing of the pions, q <-> —q', with T( ~> odd, corresponding to t-channel isospin /, = 0 and /, = 1, respectively. The connection with T* and Ti is provided by (12.10), and the relation with the possible charge states n+p->n+p, n°p -> n°p, n+n, and n~p-> n°n, n~p for proton targets is worked out in Problem 12.1. s-Wave Scattering Lengths. Returning to the Feynman pole amplitudes (12.25), we again test their validity in the low-energy limit s = (mN + mn)2, u = (mN — mn)2, t — 0 so that the Mandelstam relation s + t + u = 2ml + 2m2- is satisfied. This configuration corresponds to sm x 2mN mn and um « —2mNmn, which is as close as we can get to the singularities of
242
Low-Energy Strong Interactions
(12.25b) at sm = 0 or um = 0. Presumably at this threshold point, the breakdown of the perturbation-theory expansion in the large coupling constant g will be compensated by the "nearby singularities" at sm — 0 or um = 0 of the pole diagrams (12.25). Since there is no analog potential for nN as there is for NN scattering, we must compute the isospin-even and -odd scattering lengths for / = 0, a(0+) = i(% + 2a f ), and a(0_) = ^(a^ - a^) (see Problem 12.1) from the p -* 0 limit of (12.25b). First making the nonrelativistic reductions up4up, uP'4'up->2mNmn, we find (Problem 12.3) 2g2 _) g2m„/m « 2* , 2 T ( ' - U) / " - " *2N, . 2 (12.27) T ( + >_ 1 - m JAm N ' ~* 1 - m JAm N ' Then applying f(±) = T(±)/87iW-> a{±) to (12.27) as p-*0, we obtain the s-wave scattering lengths for pseudoscalar coupling (ps), a(o+)(ps) « - ? a - 1.8m;x « -2.6 fm, 0 ^ ' An mN + mn
(12.28a) '
2
a^Hps) x £ j ^ K 0.14m;l a 0.19 fm. (12.28b) VF ' 8TC mN(mN + mn) While the isospin-odd prediction (12.28b) is reasonably close to experiment (a(0_) % 0.09m;1 « 0.12 fm), the isospin-even s-wave scattering length (12.28a) is about 400 times larger than the measured value, a<0+) « -0.005m; 1 K -0.007 fm. If we instead use pseudovector coupling (pv) at the 7rNiV vertices Ni$y5 x • nN, then the predictions for the Feynman poles of Figure 12.4 are (see Problem 12.3), f2 1 a(n+)v(pv) « * -0.010m; 1 * -0.014 fm, (12.29a) r ' An mN + mn
«(o_,(pv) * C -T^
x * °-
(1129b)
8K mN(mN + m„) In this case, a(0+)(pv) is near experiment, but a(0~'(pv) is not. Neither coupling explains both s-wave scattering lengths. Where did we go wrong? The trick is to maintain the good predictions a(0_)(ps) and a(0+)(pv) while discarding the bad predictions a(0+)(ps) and a(0_)(pv). For many years, this problem was regarded as a true sickness of s-wave nN scattering. We shall see in Section 15.H, however, that dispersion theory provides the natural explanation of this puzzle for ps coupling: It preserves (12.28b) while modifying (12.28a) by requiring the "residue" of the pole denominator of T( + ) in (12.25b) to be evaluated at the kinematical configuration of the pole, sm um = 0. In effect, the dispersion prescription suppresses the prediction a(o+)(ps) by the factor (mn/2mN)2 at threshold, thus converting g2 t o / 2 and a(0+)(ps) to the desired a^+)(pv). An alternative explanation is provided by the PC AC constraint on the background (nonpole) part of the amplitude T ( + )in T( + ) = T^il + f , + ). We show in Section 12.F that this also converts a(0+)(ps)
Low-Energy nN Scattering
243
effectively to a^+)(pv). For pv coupling, the resolution of this s-wave puzzle is provided by the theory of current algebra (see Section 15.H), which in fact successfully predicts the complete low-energy structure of the entire nN amplitude (not only s-waves, but all partial waves). The upshot of this digression is that the simple Feynman pole diagrams of Figure 12.4, when properly interpreted, provide a reasonable qualitative description of lowenergy nN scattering, as such diagrams do for JVJV scattering. />-Wave Scattering Lengths. To prove our point, we extend the discussion to p-wave nN scattering at threshold. The 0~ pion requires the virtual s- and w-channel poles in Figure 12.4 to have / = 1 by parity and angularmomentum conservation (recall Problem 6.4), so that the p-wave scattering amplitude should be very important. Rather than making a formal partialwave expansion of the pole amplitudes (12.25b) and then specializing to / = 1, it is simpler to employ the / = 1 projection operators for J = I + j , J = I> h analogous to the isospin projection operators (12.10a),
&\ = W
q + * q * q),
&*, = W
• 4 - * • 4' * 4).
(12.30)
Then keeping momentum terms through second order in the nonrelativistic reduction of the spinors in (12.25), the lowest-order T ( ± ) are (12.27), while the next-order p-wave amplitude is (Problem 12.3) T}7(p wave) =
-3g2q'q
-I
(12.31)
The s- and u-channel projection operators in (12.31) are natural consequences of the fact that both the s- and the w-channel nucleon poles have
/ = i S = i J = i and / = 1. Since we are interested in low-energy scattering in the s-channel, it will prove useful to cross the /* and &%u projection operators back into the s-channel by way of the crossing transformation inferred from (12.10c),
it = -W + \i\,
( 12 - 32 )
&l. = - i < + M-
Then defining the combined s-channel projection operators &2i,2J &ii = I&j.,
PIZ
^31=/?^.,
4*33 = J . M , ,
= I$P*J,
as
(1233)
the u-channel projection operator for / = J = \ can be expressed in the s-channel via (12.32) as $&.
= ^ 1 1 " 1^13 - 1^31 + ^ 3 3 -
(12-34)
Thus, for low-energy p-wave scattering in the s-channel CM frame with sm % 2mN a>, um& — 2mN co, to = pion energy, q = q' = | q | = pion momentum, (12.31) becomes (for either ps or pv coupling), T}7(p wave) * - ^ - ( 4 ^ u + <*>13 + £»31 - 2^ 33 )5mNco
(12.35)
244 Low-Energy Strong Interactions
At low energies, the partial-wave amplitude goes like p21 (Problem 8.5), so that at threshold co ->• mn, the p-wave scattering lengths are «2/,2j = [87t(mN + mn)]'lT2I2J(p
wave)/q2 | thresh .
(12.36)
Then combining the coefficients of ^ 2 / , 2 J m (12.35) with (12.36), the nucleon pole predictions for the p-wave scattering lengths are (Problem 12.3)
(12.37a)
a\f = alt ~ -f^ (6m 2 mJ- x « -0.05m; 3 ,
(12.37b)
alf ~ ~ (3m^mJ- l * 0.11m;3.
(12.37c)
These values compare rather well with experiment: olY « -0.09m; 3 ,
ae3x3p x 0.22m;3.
(12.38)
ar 3 p «a e 3T~ -0.04m; 3 , It turns out that with a little bit of work we can even improve upon these already reasonable pole predictions. Chew-Low Model. While the Feynman nucleon-pole amplitudes, properly interpreted, give a good description of low-energy s- and p-wave 7tJV scattering, they fail to account for the observed bump in the cross sections at yfs ~ 1230 MeV. In this region experiment indicates o(n+p ^> n+p):a(n~p ^ n°n):o(n~p ^ n~p) x 9:2:1,
(12.39)
with angular distributions do/dQ ocl + 3 cos2 0. All of these facts point to the influence of the 33 resonance A(1232) with / = J = \ (see Problem 12.2 and Section 8.C). Chew (1954) observed that a possible dynamical clue to the origin of this resonance resides in the minus sign of ^33 relative to SPXX, 0>13, &>3i in (12.35). Recalling from (11.68) that an effective potential is V oc — Tcov, we see from (12.35) that only T33 generates an attractive interaction, perhaps strong enough to force the 33 state to "resonate". The other three T2iy2j generate repulsive interactions, consistent with the observed absence of such resonances near threshold. Rather than proceed with the quite involved nonlinear techniques used by Chew and Low (1956) to generate the A resonance, we parallel our discussion with the effects of the deuteron upon low-energy NN scattering. That is, we assume the existence of the A resonance at mA « 1232 MeV and investigate how it modifies the nucleon pole amplitudes near threshold. First, in the nonrelativistic scattering region w mN, the structure of the partial-wave amplitude (8.15) is most easily stated in terms of 1//,. For the I = J = j contribution of the nucleon pole amplitudes, we may write from
Low-Energy nN Scattering
(12.35) and
245
f33=T33/SnW,
<,140)
•"'W-'/H'W
Then forcing the phase shift to resonate by making <533 -+ n/2 (i.e., cot <533 -> 0) in a counterclockwise sense as a> increases through the resonance a> -> coR (as explained in Section 8.C), we induce the effective-rangetype expansion ^ cot 533(q) * Rcf33l(q)
-
CO
(J)
* If)
' 3m£ (l - — ) .
\47t/
\
(12.41)
C0 R /
Extrapolation of (12.41) back to threshold allows us to identify the intercept with the coupling-constant strength, giving g2/4n « 14.6. Now near the resonance, the amplitude is not purely real [as is the pole amplitude (12.40)], but must develop an imaginary part according to the unitarity condition Im l// 33 (q) = — q. Combining unitarity with (12.41), we are led to a BreitWigner form for/ 33 (see Problem 12.4),
/ 3 TM
(g2/4n)(q2coR/3m2oj)
3
1 2
(iq coR/3m^w)(g /4n)
qw-coR-
lr33
i\Y33 (12.42)
provided the width of the A resonance is r33 = rA = ^ ^ f . 4n 3mji
(12.43a)
For O>RM = (ml + m\ — w£)/2mA ~ 266 MeV, we have qR = {
227 MeV,
and then g2/4n « 14.3 implies from (12.43a) that TA « 126 MeV,
(12.43b)
very close indeed to experiment. This agreement lends credence to the Chew-Low hypothesis that the A "particle" is nothing but the unitarized u-channel nucleon pole (crossed over to the s-channel) which dynamically resonates at ofRu x 266 MeV, ^/s « 1232 MeV. In the same manner that we treated the bound-state deuteron as effectively elementary in (12.23) in the region of the bound state, we may treat the A resonance as "elementary" with (12.42) the analog of (12.23). Such a | + A-particle couples to the nN system via the effective phenomenological hamiltonian density J*^, = ^ A> V N T T ^ ) , (12.44) mN where A,, is the Rarita-Schwinger spin-f bispinor satisfying y^A^ = 0 (recall Section 5.E). A relativistic calculation of the decay A -• Nn in the narrowwidth approximation, similar to the nonrelativistic spinless estimate of Sec-
246 Low-Energy Strong Interactions tion 8.C, leads to the dimensionless coupling strength (see Problem 12.5) g*2/4n x 15 for TA « 115 MeV. Corrections due to the finite width of the A in the resonance amplitude (12.42) reduce the strength to g*2/4n ~ 11 [recall the difference between (8.44) and (8.47)]. Return to />-Wave Scattering Lengths. Because the A resonance is so close to threshold, mA — (mN + mn) x 150 MeV, its Breit-Wigner "tail" should have an effect on threshold dynamics. If we extend (12.41) down to threshold, with co/coR -* 140/266 ~ 0.53, then the 33 p-wave scattering length is modified from (12.37c) to a£ = [q3 cot d<3%s(qmlm„ * -
2
f,/4*
,
, * 0.22m;3, (12.45a)
in perfect agreement with experiment, (12.38). This parallels the deuteron modification of the a 0 ( 3 ^i) NN scattering length. A similar Chew-Low dynamical extrapolation due to the presence of the A in the determination of the other three p-wave amplitudes is based upon the projection-operator structure of (12.35). This modifies the nucleon-pole predictions (12.37a) and (12.37b) to a r«
3. _l_ 2(im2NmK)-l(\+2rtlA 47C
\
* -0.10m; 3 ,
(12.45b)
(DR }
a\
' * -0.04m; 3 ,
(12.45c)
again matching experiment (12.38) in every way (Problem 12.4).
12.E Hadronic Vector Currents While simple Feynman pole diagrams involving hadrons provide an explanation, when properly interpreted, of a variety of strong-interaction problems, they fall far short of a complete dynamical theory. Turning to electromagnetism for inspiration, we search for hadronic "currents" which might be completely or partially conserved in analogy with the electromagnetic current. Such currents can be used to supplement the Feynman-diagram approach. Isovector and Isoscalar Vector Currents. The equality of the magnitude of electron and proton charge tells us that the strong-interaction cloud surrounding the proton does not "renormalize" the proton charge; i.e., charge is conserved in strong interactions. Since isospin and hypercharge are also conserved in strong interactions, we may regard the Gell-Mann-Nishijima (1953) equation Q = / 3 + jY as relating the electromagnetic charge Q (in units of e) and the conserved strong-interaction "charges" 73 and Y, the
Hadronic Vector Currents 247 latter charges also being separately conserved in electromagnetic interactions. Now conserved charges are linked to conserved currents, d • V = 0, by Gauss's theorem, and so we decompose the electromagnetic vector current Un ->•./£ n e r e ) into its isovector and isoscalar parts as Jl=Jl+)l-
(12.46)
The corresponding charges are then related to these currents according to Q = j d3x r0(x),
73 = | d3x jv0(x),
hY = j d3x js0(x). (12.47)
Since these charges are conserved, the7„(x) in (12.47) can be evaluated at any particular time, say t = 0. The notion of CVC (conserved vector current) is to treat j ^ and jj/' (with JM = %'3) as conserved hadronic currents, d-jv'i(x)
= 0,
djs(x)
= 0,
(12.48)
having dynamical significance for strong as well as electromagnetic interactions (actually, CVC was first introduced for weak interactions—see Section 13.B). For example, the nucleon matrix elements of these conserved hadronic currents have the momentum-space structure for q = p' — p (N„,| f/(x)|
JVp> = Np. M W K
+ Fv2(q2)ia^qy2mN]Npe^"*,
*,
(12.49a) (12.49b)
2
where F\'l(q ) are isovector or isoscalar, "charge" or "magnetic-moment" hadronic form factors, respectively. Now the sum of the i = 3 component of (12.49a) plus (12.49b) must be the electromagnetic current (5.97) for protons or neutrons, according to (12.46). Then for specific isospin states in (12.49) we obtain
= m.2
+ n,2),
(12.50a)
= i ( - n , 2 + Fi2), F 1.2 = F"U2 - F1,2, 2
(12.50b) Fsli2 = Fl2 + F[,2.
(12.50c)
In particular, at q = 0, the statements F{(0) = 1, F1(0) = 0 (no charge) require F\(0) = F?(0) = 1. Likewise, Fp2(0) = KP= 1.79, F"2(0) = K„ = - 1.91, so that F^(0) = KV = 3.7, and Fs2{0) = KS = -0.12. Finally, in the same way that up 4UP = 0 for free nucleons and q^a^ qv = 0 lead to a conserved electromagnetic current
248 Low-Energy Strong Interactions In a similar manner we may extend the electromagnetic charged-pion current (4.63b) to an isovector-vector hadron current for q = p' — p, « | f/(x)
17t*p> = isiikFM2W
+ PU" •x,
(12.51)
where again FK(0) = 1. The photon current (in units of e) is J1 =j%"3 for i = 3, with no contribution coming from /£, because it does not couple to pions (due to an isospin generalization of C-parity called G-parity, G = C( — f for neutral self-conjugate members of isomultiplets). The vector current (12.51) is indeed conserved for p'2 = p2, with a possible additional term proportional to (p' — p\ absent for (ji \ d • j v , i 17t> = 0. Existence of Vector Mesons. If the nucleon charged clouds were made up completely of uncorrelated pions, then p<-m + n+ and n<->p + n~ would lead to eF\ = — eF\, since the photon would see en+ = — e„_ in the two cases. But experiment says F" ~ 0. To explain this fact, Nambu (1957) suggested the existence of a neutral / = 0, JF = 1 ~ vector meson (the co) which could couple to the photon and generate an isoscalar nucleon current such that F\ ~ F\ and F\ ~ 0 by (12.50b). Moreover, Frazer and Fulco (1960) required the existence of an J = 1, Jp = 1" vector meson (the p) in order to explain the structure of the nucleon form factor F\(q2) [i.e., Gj^q2)]. Finally, we have noted in Sections 1 l.G and 12.C that detailed aspects of the nucleonnucleon force suggested the existence of these vector mesons (Breit 1960, Sakurai 1960). Shortly afterward, in 1961, the gates to the "meson zoo" were opened with the observation of co(783) and p(776) (i.e., mffl = 783 MeV, mp = 776 MeV). Vector-Meson-Vacuum Transition. Given the existence of vector mesons p and co (and also the >, which we ignore here), both having the C-parity of the photon, C(y) = C(p°) = C(a>) = — 1, there must exist a direct p-y and co-y transition. The isovector p-to-vacuum matrix elements of the hadronic isovector current can be written as <0\f/(x)\pJ(q)y
m2 = S- e,(q)5^-^',
(12.52a)
9p
which is the only form consistent with Lorentz and isospin invariance. The "form factor" gp in (12.52a) is defined in this peculiar way for reasons which will become clear shortly. For p on shell, gp = 3P(m2) does not vary. Likewise, the isoscalar co-to-vacuum matrix element of the hadronic isoscalar current has the form m2 <01 %(x) | a>{q)> = -f zM)e-iq' *•
(12.52b)
Obviously these currents are conserved, because d -jv's oc q • e(q) = 0 by virtue of the subsidiary condition satisfied by on-shell spin-1 polarization vectors.
Hadronic Vector Currents 249 yV
^o
ys
enr^/g,
,jj
emw/g„
Figure 12.5 Vector meson-vacuum matrix elements of the isovector and isoscalar electromagnetic currents. By analogy with the nucleon matrix elements of the electromagnetic current, the vacuum-p° or vacuum-a) matrix elements of the electromagnetic current will be PYYl
<01 ;7(x) | p°, coy = ^
2
sM)e-iq'x
(12.53)
for the isovector and isoscalar parts of (12.46), respectively. The corresponding Feynman vertices are shown in Figure 12.5. The coupling constants gp, g^ can be extracted from the electromagnetic decays p° -+e+e~~, co->e + e _ via the photon pole graphs of Figure 12.6, since only the photon couples to the em current. The Feynman rules plus (12.53) and Figure 12.5 then give for q2 = m2p>(0 and Figure 12.6 S/T = (-if(-e)ue-fve+
l^f)e-^^q)8^Pfi),
iy?=-—ue-ys{q)ot+.
(12.54a) (12.54b)
Plugging (12.54b) into the general two-body decay formulae of Section 10.B, the experimental decay rates give in the narrow-width approximation (see Problem 12.6) 2
?*-* 2.1,
2
f2«i8.i.
(12.55)
Vector-Meson Dominance. If CVC is to have any dynamical significance, we must find a way to isolate the hadron vector current from other hadronic interaction mechanisms. One such "probe" of the hadronic vector current is the photon via vector-meson dominance of photon-hadron reactions (Sakurai 1960). In particular, with the hadronic pirn vertex of the form for q = p' -p,
*>„«« = <W fe^W + PYPJM)AP\
Figure 12.6 Photon-dominated electron decay modes of vector mesons.
(12-56)
250 Low-Energy Strong Interactions 4 TT<
y
V
\
7T k
Figure 12.7 Rho dominance of the isovector em current.
p° dominance of the isovector current (12.51), as shown in Figure 12.7, leads to
" -UtT-M/nt) «-1 m I K> * (- i)h gPM + Pi q2 -ml + ie iik
m2 dp
(12.57) 2
Assuming gpnn does not vary significantly from its on-shell value at q = m2 to q2 = 0, we evaluate (12.57) on the photon mass shell at q2 = 0 in order to make the comparison with (12.51) for j=3 and F„(0) = 1. Since q • (p' + p) = 0 in (12.57), this results in 9P « gP„,
(12.58)
which is, of course, why we chose the definition of gp in (12.52a) in such a peculiar fashion. Thus, to test the notion of CVC and vector-meson dominance (there is no way to separate them in this analysis), we need only compute the value of gpn„ from the physical p° ->n + n~ decay (see Problem 12.7). In the narrowwidth approximation this gives gpn„/4n « 3.0, which when compared with g2p/An K2.\ leads to the confirmation of (12.58) to within 15%. This accuracy is all we can expect, because gp presumably does vary, perhaps slightly, from t = 0 to m2p. The conservation of the hadronic vector current means that if CVC and vector-meson dominance were exact, we would expect the p (and <x>-§) to couple to all hadron states in a universal way, similar to the universal coupling of the photon to all charged particles. Then universality would require gp = gpnn = gpNN = •••, where, for example, from (12.49a), ^pNN
= 9PNN
(12.59)
and one repeats the argument leading to (12.58), but with nucleons replacing pions in Figure 12.7 and (12.57). From the energy dependence of s-wave nN scattering, it is possible to isolate the t-channel p contribution, giving gPnn gpNN /4JI » 2.4, in good agreement with universality, except possibly for the F\ form factor (Genz and Hohler 1976). In fact, with the proper interpretation (see the discussion of current algebra in Section 15.H), the p pole controls the It = 1 scattering length a{~\ again consistent with vector dominance and universality (see Problem 12.7). While vector dominance gives moderately good predictions for other photon-hadron processes such as yN -> nN vs. nN -* pN, it still cannot be
Hadronic Axial-Vector Currents 251 regarded as a complete theory. Problems exist with vector mass extrapolation, with off-shell gauge invariance of (12.53), and even with the double pole structure of the nucleon form factors of Sections 8.E and ll.F [naive vector dominance of G\M{i) would predict a single p pole denominator]. It does, however, justify the apparently small n° -> 2y amplitude of (11.66), because vector dominance relates Fnyy for n° -> 2y to Fwny of a> -> n°y via n° -* y{p -»y) + y(a> -» y) or Fnyy = 2Fn(ttpe2/gpglo vs. co->7r(p-y) or ^c»y = FnoP e/dP- The resulting prediction | FKyy /Fmny \ = lejg^ « 0.040 is in fact quite close to experiment (| Fnyy/Fc}ny \ « 0.032; see Problem 11.10 and Gell-Mann et al. 1962a). In any case, the many phenomenological successes of the CVC-vector-dominance hypothesis are a clear indication that a conserved hadronic vector current underlies, in part, strong-interaction dynamics.
12.F Hadronic Axial-Vector Currents The concept of a conserved vector 1~~ current is so useful in electromagnetic, strong, and even weak interactions, that it is compelling to search for an analog 1 + axial-vector hadronic current in strong (and also weak) interactions. Isovector Axial-Vector Hadronic Current. Consider the nucleon matrix elements of an isovector 1 + current j A , t . By analogy with (12.49a), we write for q = P' - p,
= Np.^\gA{q2)^,ys
+ h^fayjN
,<*•*,
(12.60)
2
where gA{q ) and hA(q ) are called the axial and induced pseudoscalar form factors, respectively. The complex number i in iypys and in iqlt -> — d^ guarantees that the two terms in (12.60) are hermitian if gA(q2) and hA(q2) are real for real q2. This is parallel with the hermitian vector currents (12.49). The analog to the p-vacuum 1" transition (12.52a) is the pion-vacuum 1 + transition <0\jA/(x)\nJ(q)> = ifnq^e-^\
(12.61) -
consistent with isospin and Lorentz invariance for the 0 , In = 1 pion, where fn is the "form factor" similar to gp and is a constant for an on-shell pion, q2 = m2.. Like gp, which is measured in the electromagnetic leptonic decay p° -> e+e~, the constant/, in (12.61) can be extracted from the weak leptonic decays Ti^ —> / i ± v as fn ~ 93 MeV. Moreover, the form-factor scale in (12.60) at q2 = 0 is gA(0) = gA « 1.27, as found from neutron /3 decay. But more about these weak decays in Chapter 13. This axial-vector current is represented by a jagged line, as shown for nucleon and pion-vacuum matrix elements in Figure 12.8. Note that the momentum directions of the pion and axial vector in Figure 12.8(b) are opposite, as defined by (12.61). Both directions are to be reversed for out-
252 Low-Energy Strong Interactions
s\itss\s^/\(
)
N
/vr\/s/s/\»
(a)
-«--q
(b)
Figure 12.8 Nucleon (a) and pion-vacuum (b) matrix elements of the axial-vector current (represented by the jagged line). going axial vectors with q-> — q in (12.61), consistent with the complex number i in (12.61) and the time-reversal transformation. PC AC. There is no a priori reason to suppose that the axial-vector currenty'^ is strictly conserved as are the photon and vector currents _/'£, but if we compute d • jA for the pion-vacuum transition (12.61), we find at x = 0 and <0|5-/'V>=/-^y(12-62) Since the pion is by far the lightest hadron (i.e., on the hadronic mass scale set by the nucleon ml /m^ « j$) it may be a reasonable approximation to set ml x 0 in (12.62), leading to the assumption of the partially conserved axial current (PCAC) (Nambu 1960) <0|d • / • ' > ; > %0.
(12.63)
This statement is without much dynamical content, but the intriguing possibility then exists that perhaps d • jA « 0 in an operator sense, independent of the hadronic matrix elements considered. In particular, for the nucleon transition (12.60), d-jA zzO implies (usingfay5 ^> 2mN y5 between on-shell spinors) 2mNgA(q2) + q2hA(q2)~0-
(12-64)
Pion-Pole Dominance. To proceed further, we parallel the CVC-vectormeson-dominance scheme by linking the PCAC hypothesis, d • jA « 0, with the pion-pole dominance of axial-vector, hadronic transitions, such as
Hadronic Axial-Vector Currents 253
/V^/\A(
)
N
~
/V/fs/WS.*--*-
V-X
q <
Figure 12.9 Pion pole dominance of the nucleon matrix elements of the axialvector current near q2 = 0.
hA that contains the pion pole, M
(12.66)
Finally, we combine the PCAC condition (12.64) with pion-pole dominance (12.66) and suppress the nonpole background corrections to (12.66) by setting q2 = 0 in [q2hA(q2)] = 0. Then we obtain the relation (Goldberger and Treiman 1958) LgnNN~mNgA.
(12.67)
This amazing equation links together three otherwise unrelated stronginteraction coupling constants, two of which are extracted from weak decays. It is in fact almost exactly satisfied in nature. Using the present measured values fn = 92.42 ±0.26 MeV, gA = 1.2695 ±0.0029, gnNN = 13.17 ±0.66, mN = \{mp + mn) « 938.92 MeV and the GoldbergerTreiman discrepancy A is about 2.1% A= 1 - ^
-
= (2.07±0.01)%.
(1168)
In the dispersion-theory context of Chapter 15, it will be most natural to combine the PCAC operator Ansatz, d • jA « 0, with pion-pole dominance (12.66) in terms of a single "dispersion relation" for d • j A , the latter (also referred to as PCAC) then being on a firmer footing than the operator PCAC statement. Nevertheless the general notion of PCAC remains a radical assumption and must always be tested in each individual case, as in (12.68). With hindsight, the PCAC-pion-pole-dominance approach in fact appears more reliable than the CVC-vector-meson-dominance scheme, primarily because of the smaller extrapolations in q2, from ml to 0 in the former case, but from m2 ~ 30m* to 0 in the latter case. This more than compensates for the nonexact nature of PCAC. Soft-Pion Theorem. The success of the Goldberger-Treiman relation (12.67) suggests that we ought to pursue further the consequences of PCAC and pion pole dominance. In the same manner that the pion pole in (12.66) is enhanced in the invariant form factor at q2 = 0, we expect the pion-pole contribution to general axial-vector, hadronic covariant amplitudes to dominate in the soft-pion limit q^ -> 0 along with other (nucleon) pole contribu-
254 Low-Energy Strong Interactions
tions, 0(l/q). That is, explicitly separating the pion-pole contribution M„ from an axial-vector M-function M„, as in the simple case of Figure 12.9, but also keeping the background amplitude M„, the Feynman rules give for a pion of momentum q and isospin index i K = ( - 0 ( - ' / , 9 j [q2_lm2
i
+ i^M
M + K.
(12.69)
Again assuming that ml « 0 in the pion propagator along with the PCAC statement g"M; w 0, (12.69) leads to iy,M&) * ^Mj,.
(12.70)
Now as q* -+ 0, only the axial vector "tagging" onto external nucleon lines, as designated by x in Figure 12.10(a), will generate nucleon propagators 0(l/q). The x will represent g^-type coupling of the axial vector, since the hA-type is already accounted for by the M„ term in (12.69), by analogy with (12.66). Then applying the Feynman rules to Figure 12.10(a) (see Problem 12.8), we may use (12.67) to identify the nucleon pole parts of g"Mj, with the pion-nucleon pseudoscalar interactions of Figure 12.10(b) plus a nonpole term of the form (-r'y5M0 -I- M0ysx% where M 0 is the general hadronic amplitude of Figure 12.10, but not containing the pion or axial vector. Finally, letting g" -*0 in (12.70) in order to suppress all further background parts in M,, of O(q0), we are led to the soft-pion theorem M'.{q) ~ MJ,s„poles(q) + MlK(q - 0),
(12.71a)
M & - 0) = ^ ( T ' ys M 0 + Mol5 x% (12.71b) zmN That is, PCAC and pion-pole dominance determine the non-rapidly-varying background pion amplitude, once the rapidly varying (in q) nucleon poles of Figure 12.10(b) with pseudoscalar coupling to the pion, gx'ys, have been explicitly removed. This theorem, valid for either an incoming or an outgoing soft pion, is the analog of the (CVC) soft-photon theorem (11.51).
:* •% x -t (a)
(b)
Figure 12.10 Axial-vector (a) and pseudoscalar-coupling nucleon-pole (b) contributions to the axial-vector amplitude arising in the soft-pion theorem.
Hadronic Axial-Vector Currents 255 Application to JTJV Scattering. As an application of (12.71), we return to the low-energy rciV-scattering puzzle of Section 12.D: how to maintain the nucleon-pole scattering length a(0_)(ps) while converting a(0+)(ps) to a(0+)(pv). According to (12.71), we should add the background amplitude (12.71b) to the nucleon-pole graphs of Figure 12.4 with pseudoscalar coupling, (12.25). Letting the final pion with isospin index ; become soft (q' -»0), we identify M 0 in this case as — gt'y5, so that (12.71b) becomes the "PCAC-consistency condition" (Adler 1965a), M!,V - 0 ) = -JL
(T''y50T>y5 + ^VsTst') = — 5iJ.
(12.72)
Since (12.72) contributes only to the isospin-even nN amplitude, we replace r j , 2 in (12.25b) with T$e + (g2/mN)upup, which at threshold (uu-*2mN) modifies (12.27a) to -<+)
2g2 1 - ml)Am2N 2g2(mJ2mNf 1 - m2JAm2N
2g2 2p 1 - m2JAml'
(12.73)
while leaving T{ > unchanged. This is precisely the sought-after result, because (12.73) indeed converts a'0+)(ps) to a^+)(pv) a s given by (12.29a). The a Model. Gell-Mann and Levy (1960) demonstrated that there exists a field-theory model which builds in PCAC and the Goldberger-Treiman relation in a natural way. It requires the existence of a tr-meson, a scalar 0 + particle, which couples to nucleons along with pions in the hamiltonian or lagrangian density as N(y5 x • n + a)N. This structure guarantees that the axial current is exactly conserved, d • A1 = 0, and also requires that 9ONN = 9nNN- Moreover, the background nN amplitude (12.72) is then generated by a t-channel c-meson pole, so that (Problem 12.10) = ~m(m2-m2n). (12.74) zmN This in turn determines the a width as a function of its mass, giving in the "narrow"-width approximation for gaNN = gnNN, Ta ~ 600-700 MeV for ma ~ 600 700 MeV. As noted in Section 12.B, while such a particle is hard to detect, it does appear necessary to explain the attractive NN binding force at intermediate ranges, r ~ 2 fm, with a coupling constant gaNN ~ 0.7g(„NN and mass ma ~ 550 MeV not unlike those predicted in the a model. In summary, then, the main point of this chapter is that lower-order simple Feynman diagrams, when properly interpreted and modified by "nearby" bound states and resonances, coupled together with conserved vector and partially conserved axial-vector hadronic currents, go a long way toward realistically describing low-energy strong-interaction dynamics. 9aNNg^
256 Low-Energy Strong Interactions For further general reading on low-energy strong interactions, see Blatt and Weisskopf (1952), Bethe and Morrison (1956), Elton (1959), Schweber (1961), Moravcsik (1963), Bjorken and Drell (1964), Debenedetti (1964), Muirhead (1965), Gasiorowicz (1966), Pilkuhn (1967), Bernstein (1968), Sakurai (1969), Bransden and Moorhouse (1973), and Brown and Jackson (1976).
CHAPTER 13
Lowest-Order Weak Interactions
Weak interactions are responsible for the radioactivity of nuclear elements as well as the decays of most of the stable elementary particles. A perturbation-theory phenomenological approach to low-energy weak interactions is even more successful than it is for strong interactions because the coupling is weak. But it is on an even less firm theoretical footing due to the "nonrenormalizability" of the current-current weak hamiltonian. Nevertheless, such a perturbation theory again reveals the fundamental role of (weak) currents—both of the lepton and of the hadron type. Accordingly, we discuss weak leptonic, semileptonic, and nonleptonic decays in the language of currents and lowest-order covariant (Feynman) diagrams. As a consequence we shall be able to extract the weak-interaction scale or coupling constant, along with the scales of hadronic currents which we exploited in the last chapter. We also attempt to unify strangeness-conserving and strangeness-changing weak hadronic transitions by introduction of the Cabibbo angle. Finally we look into the nonleptonic weak transitions, which are complicated by strong- as well as weak-interaction dynamics. We shall use the tools of hadronic currents, isospin, and Feynman pole graphs to analyze such nonleptonic decays.
13.A Phenomenology of Weak Decays Weak interactions are, as the name suggests, weaker than strong and electromagnetic interactions—weak enough in fact that at "normal" laboratory energies they are manifested not by scattering processes, but by spontaneous 257
258 Lowest-Order Weak Interactions
weak decays of particles which are stable with respect to strong and electromagnetic interactions. Decay Lifetimes and Quantum-Number Violation. Suppose at time l = 0we isolate in a box a strongly interacting hadron (a half-integer-spin baryon or integer-spin meson). Again the rules of quantum mechanics hold: Anything that can happen, does happen—with a definite probability. If the particle is a strong-interaction resonance, it will decay strongly into other lighter hadrons, with a typical decay width of r s t ~ 100 MeV and coupling strength g2/4n ~ 10. It does this in i st = h/Tst ~ 10" 2 3 sec, the time it takes light to travel across a hadron of size ~ 1 fm, while conserving the internal quantum numbers of charge, hypercharge, isospin, and the discrete space-time symmetries of C, P, T. The remaining stable hadrons in the box (for example the low-mass p, n, I*- 0 , A, E~-°, Q", n±-°, K±, K°, K°, r]) then try to decay electromagnetically. Only £°, n°, r\ succeed, however, via the decays Z° -> Ay and n°, r\ -* 2y, taking the longer time of Tem ~ 10" 19 , 10" 1 6 sec, respectively, and violating isospin symmetry (but not I3) in the process. Finally, the remaining particles in the box have no choice but to decay weakly (all except the proton, which is stable even against weak interactions), further violating not only I2, but also I3, and sometimes Y (or S) as well as C and P, but conserving CP and T. The weak analog of electromagnetic photon radiation is the massless neutrino "weak radiation". The neutrinos, electrons, and muons are spin-^ leptons which are weak decay products of semileptonic decays, with the muon then decaying via a purely leptonic process. Hadrons decay weakly by either semileptonic or purely nonleptonic transitions. These weak-interaction lifetimes vary over a wide range of TW ~ 10" 10 -10 3 sec. Ultimately, after 15 minutes have elapsed, only stable protons and electrons remain in our box, surrounded by photon and neutrino radiation. Only the internal quantum number of charge and the space-time CPT symmetry are absolutely conserved during this time. Weak-Decay Patterns. We observe empirically that: i. Leptons occur primarily in charged pairs (ev) and (/iv'), with the absolute lepton-number conservation apparently decoupled by the selection rules fi+l+e and v' */* v, with v the electron and v' the muon neutrino. ii. Such charged lepton pairs take part in the purely leptonic decay, H -* ev'v and, for example, the semileptonic neutron /?-decay n -> pev, pion jS-decay n+ -»7t°ev(7t,3), nl2 decay n+ ->n+v', e+v, Kn decay K+ -»/i+V, e+v, Kl3 decay K -»nev, n/iv', and hyperon /J-decays. iii. Neutral lepton pairs such as (vv), (vv) have recently been detected in scattering processes as ve -» ve, v'e -> v'e, vp -*• vp. However the strangeness-changing decay K -> nvv has not yet been seen, iv. Hadron transitions in semileptonic decays obey the empirical selection rules that AQ = 1 or 0 (but not 2) for AY = 0 and AQ = 1 for AY = 1; in the latter case, AQ = AY (but not AQ = -AY).
Current-Current Hypothesis 259
v. Hadron transitions in nonleptonic A Y = 1 decays approximately obey the M = \ rule (and not Al = f, f, ...). vi. Both P and C appear to be "maximally" violated in many weak decays (i.e., large asymmetry parameters), but CP and T are conserved to order ~ 1 0 - 3 relative to the weak interaction.
13.B Current-Current Hypothesis We have seen, in the last two chapters, the theoretical significance of the minimal three-point em vertex Jf em = je™A" and the utility of the phenomenological three-point Yukawa vertex Jt?st =jn
(13.1)
where I — iy5 = 1 + iys. Then only V — A -> ^ ( l — iy5) survives this projection operator, and 2/Pw for processes involving an ev transition has the current-current form ^ w cc v'LTinerivL ex vi,(y„ + S i y ^ s ^ W " iJs>
= i(i - Ov%(i - iy?WY(i - iy5)v,
(13.2a)
(13.2b)
where we have exploited the left-handed nature of v'L = \v'{\ + iy5) to convert (13.2a) to (13.2b). Even though there is no a priori reason for the \i\' current to be of the V — A form with £ = — 1, we see from (13.2b) that for
260
Lowest-Order Weak Interactions
any value of t, (but t, =/= 1), the V — A structure emerges all the same. Feynman and Gell-Mann (1958) then proposed that all weak currents are of the V — A form even when an ev (or vv) transition is not present (also see Sudarshan and Marshak 1958). We shall investigate this hypothesis shortly. Weak Hadronic Currents and CVC. Turning to the hadronic structure of semileptonic weak decays, the observed AQ = + 1 or 0 for the strangenessconserving decays and AQ — A Y rule for strangeness-changing decays can be combined with the Gell-Mann-Nishijima formula Q = I3 + \ Y to give (AQ=±1,0
for Ay = 0,
(13.3a)
forAy=±l.
(13.3b)
A/ 3 = A ( 2 - i A y = UAQ=±£
This suggests that AI = 1 for A Y = 0, AQ = 1 transitions (and A/ = 0 for A y = AQ = 0 transitions which we ignore hereafter). It also suggests that A/ = \ for A y = + 1 transitions. Then it is natural to associate isotopic properties with the weak hadronic currents themselves, taking AI = 1, \ for A y = 0, ± 1 charge-changing weak currents, respectively. In the latter case (13.3b) implies that the empirical Ay = AQ rule is an automatic consequence of the AI = \ structure of the Ay = ± 1 hadronic currents. The next question to ask then is: Are these weak hadronic isotopic currents just isospin rotations of the strong-interaction currents for AQ = ± 1 transitions? Put another way, is the strongly interacting cloud surrounding hadrons that participate in a weak transition the same as the cloud for strong and electromagnetic transitions? Gershtein and Zel'dovich (1956) and Feynman and Gell-Mann (1958) answered in the affirmative by suggesting that the observed near-equality of the weak coupling strength for \i decay and for the vector part of /?-decay (a fact we shall verify shortly) means, by analogy with the charge equality - ee = ep = en+ = • • •, for current-conserving electromagnetic interactions (d • Vem = 0), that the weak vector current must also be conserved, d • Vw = 0. In the case of neutron /?-decay with q2 = (p„ — pp)2 x (m„ — mp)2 « 0, this weak-vector-current conservation requires a normalized form factor F^(0) = 1 in the same way that F\m(0) = Fi(0) = 1. It is therefore natural to assume that F^(q2) = F\(q2) for all q2 and that the neutron /?-decay AQ = 1 current is just the /i + il2 component of the strongly interacting hadron current, for the /1„ part as well as the V^ part of j * e a k , where it is conventional to denote^ -» V)1 and j* -* A^ for the weak currents. (This A^ is not to be confused with the photon field.) The total AQ = 1, Ay = 0 hadronic weak current is then determined by the CVC hypothesis, for AQ = Qf - Qt = 1 (the Ay = 1 currents will be considered in Section 13.F) to be i h , a d (AQ=l,Ay = 0 ) = J i + i 2 = K ;
+i 2
-/li
+ i2
.
(13.4a)
Current-Current Hypothesis 261 From the isotopic currents discussed in Sections 12.E and F, we deduce that for q = p' - p,
+ F^(g 2 K v q v /2m N ]n p
+ K0S[^/2Fn(q2)(p' + PIK
- n;,*[j2Fn{q2){p'
(13.4b) + P)K
+ -,
+•••• (13.4c)
More general forms of these hadronic currents can be considered on the basis of Lorentz invariance alone, with restrictions determined by Tinvariance and classified by "G-parity". We shall find, however, that the "first-class" isotopic vector current (13.4b) and axial current (13.4c), whose terms have already been encountered in Chapters 11 and 12, will suffice for our purposes. [Alternative "second-class" currents, such as a term proportional to (/?' — p)„ = q^ in the pion or nucleon weak vector current, have no analog in strong and electromagnetic interactions, since they violate C- or G-parity, so we ignore them in weak currents as well.] Since q2 « 0 for both np and n+%° transitions from the decay configuration (but not for scattering such as p + e -» n + v), we may take F\{q2) « 1, F\{q2) » KV, Fn(q2) « 1, and gA(q2) ~ 9A- The experimental value of gA « 1.25 [it need not be 1 in order to justify the V — A /?-decay structure analogous to (13.2)] was first thought to be an indication that the axial current was almost conserved (PC AC), because gA ~ 1. Now we know, however, that the PC AC nature of Ap is not due to the nearness ofgA to unity, but instead due to the smallness of the pion mass relative to the hadronic mass scale. In fact, we saw in Section 12.F that PC AC and pion-pole dominance constrain gA to approximately the Goldberger-Treiman value, gA ~fKg/mN = 1.33. Universality of A Y = 0 Currents. Since CVC for electromagnetism means that fj° = e(P)VP — ^yve) + "' ( at g2 = 0). CVC for weak currents means that, at least for AY = 0 charge-changing transitions (including the axial part), h =%"+%"• (13-5) The hadronic V — A weak current is then given by (13.4), and the corresponding A<2 = 1, V — A lepton current in (13.5) is f:p = vy„(l - iy5)e + v'y„(l - iy5)fi. ad
(13.6)
This "universality" between j£ (AY = 0) and/Jf" is similar to the notion of vector-meson-dominance universality discussed in Section 12.E. It will be extended to AY = 1 weak currents in Section 13.F. Given (13.4)-(13.6), we are at last set to define the weak coupling constant
262 Lowest-Order Weak Interactions
for the V — A current-current weak hamiltonian density:
^ = ^=Ur+m
(i3.7)
The factor of y/2 in (13.7) is a convention unaltered through the years so that Gw agrees in magnitude with Fermi's original definition of Gw for nuclear j?-decay vector (Fermi) transitions. We have introduced the factor of \ and the two current-current terms in (13.7) so as to manifest the hermiticity of Jfw. For purely leptonic and semileptonic decays, however, only one current-current term in (13.7) will occur, but without this factor of \. Finally, using our dimensional analysis for covariantly normalized states (10.15), we see from any one of the various terms of (13.4) or (13.6) that dim j = m [i.e., dim i^ve = m* in (13.6)]. Furthermore, since Jfw can describe a four-point scattering process or a three-body decay, (10.15) says dim Jiffy, = m° (for our convention ignoring normalization volumes). Then (13.7) requires dim Jfw = dim Gw (dim)) 2 or dim Gw = m~2. As for the magnitude of Gw, we shall show later that the weak-decay data give a value for the raw or bare coupling constant of Gw = 1.026 x HT*m; 2 . (13.8) Intermediate Vector Bosons. While the Yukawa-type coupling constants gnNN and e are dimensionless in our units of h = c = 1, the Fermi coupling constant is not. Also, (13.8) is valid only for q2 * 0. If Gw were to remain constant for very large q2, then the crossed analogy of \i decay, V + e -*• v + fi, would have a cross section a oc G2 E2, which violates the unitarity limit [see (13.18) and Problem 13.2]. This indicates that higher-order graphs would be needed to "cut off" the singular leading term at high energies. Put another way, we might interpret (13.8) as representing a second-order force graph at very low energies for an exchange of a "charged intermediate vector boson", W+. Feynman rules would then give a structure for such a "VG0V" second-order exchange analogous to Figure 10.7(c) of 1 mw — q
H
9w mw
Gw i
/nn\
with gw the coupling at a three-point Yukawa vertex j ~ W (see Problem 13.3). It was recently proposed that W+ is the charged analog of a neutral vector boson Z° (remember vp -> vp scattering does exist), the latter being a "spontaneous symmetry-breaking" partner of the massless photon (Weinberg 1967, Salam 1968). In this case, elegant gauge-theory ideas beyond the scope of this book suggest that weak and electromagnetic interactions are really manifestations of the same force with gw(W+) ~ gw(Z°) ~ e. This in turn would pin down the W+ mass via (13.9) to be m
2
,------^^, Lr w
1U
Trip
m^-lOOGeV.
(13.10)
Muon Decay 263 While naive considerations make it hard to accept associating the massless photon in the same scheme with the superheavy W, renormalization problems concerning this theory appear to have been resolved ftHooft 1971, Lee and Zinn-Justin 1972). In any case, this gets the "sick" Fermi theory at high energies off the hook, while postponing the discovery of such a high mass as mw to future generations of accelerators. Nevertheless, at very low energies (i.e., q2 a 0), the current-current phenomenology remains valid, placing severe constraints on any ultimate theory of weak interactions. Thus we shall follow the time-honored path and analyze the current-current hamiltonian (13.7) in detail for weak decays. If nothing else, it will give us more practice in calculating lowest-order Feynman diagrams and decay rates.
13.C Muon Decay In Section ll.F we saw that the muon does not decay electromagnetically to e + y. Thus it must decay weakly. Consider then the purely leptonic weak decay n~ -> e~ + v + v' driven by the current-current graph as shown in Figure 13.1 for the momentum-conserving transition pifl) = p'(e) 4- £|5) + k{vl.
Figure 13.1 Muon decay.
Matrix Element. The charged currents are j v > and j e i = j \ e in (13.6) with the electron neutrino crossed over from the initial state to an antineutrino in the final state. The product of these currents appears twice in (13.7), and so we may write the matrix element of the hamiltonian density as
*fi = ^=JleJv>, = %*y°V ~ '"*) V v A i -
as-")
Given (13.11), we can compute the unpolarized-spin sum by averaging over the initial muon spin states and summing over the final e, v, and v' spins. Since the muon and electron currents are in different spin spaces, we may write
l*frl 2 -;El-*/ = i<*'£,J(y,r Z
pol
(13 12)
'
264 Lowest-Order Weak Interactions The tensors in (13.12) are similar to those in electron Moller scattering (11.19), but with parity-violating as well as parity-conserving parts, ^
= Tr{Q>' + m.K(l - iysW
+ 'VsW
= 8[p;/c^ + kxp'„ - gafip' • k -
( n
u&)
ly 5
iEafydp k \
v
n /» = Tr{^«(l - iy5)(p + m„)(l + iy5)yi,} = %[P*K + Kp0-
gx0p • k' + isafivsp1'k"].
Contracting the tensors (13.13) together according to (13.12) and using the identity n'yPpaP<"'n k'
p
p' • p p' -k'
k-p k-k'
(13.14)
= 2(p' • k'p • k — p' • pk • k'\ the unpolarized spin sum (13.12) becomes (see Problem 13.1) \J?fi\2^43G2Fpe-k'pu-k.
(13.15)
It is also of interest to compute the polarized-spin sum, for which it can be shown (see Problem 13.1) that p -> p — m^s and p' -» p' - mes' in (13.15) by use of the spin projection operator (5.72). Unpolarized Rate. To calculate the total (unpolarized) decay rate in the rest frame of the muon, first we express the invariants in (13.15) as £>„ • k = m^w and (p' + k')2 = (p — k~)2 or pe • k' « mjjm^ — co) since m2. m2,. Next we fold (13.15) into the general three-body decay rate and phase space (10.13) to obtain dT = 1
L ' ^ " P d P3* - ^ 1 ? «>(K - «>) d™ dEe>
(13-16)
where we have integrated over the v' momentum to eliminate d3(Pfi). The next step is to fix the electron energy Ee (with me « 0) and integrate over co. Energy conservation says m^ = Ee + a> + a>. Then the "maximum" kinematical configuration of p and k', aligned and opposite k, gives a>mm — jnip, while the "minimum" kinematical configuration of p and k, aligned and opposite k', gives, for fixed Ee, tomin = ^m„ — Ee. Integrating (13.16) over this range of to, we obtain the electron energy spectrum for muon decay, —-=\ da ——•— = —^ m2uE2 1 - —?• . 13.17a e y dEe ) i m ^ E e \d(odEe) 4TT3 " \ 3mJ ' This characteristic electron energy dependence is in fact observed in muon decay. Finally we find the total decay rate for unpolarized muons by integrating
Muon Decay 265 (13.17a) from E?" « 0 to E™ = > „ (i.e., co max ~ E?n and
d E e l ^ r ) * ^ ^ 3. JEe)
d3.17b)
~ 192TT
Then the experimental lifetime x^ = 2.199 x 10" 6 sec (or Tft = h/ttl = 2.965 x 10" 16 MeV) and m^ = 105.66 MeV lead to an approximate determination of the weak coupling constant from (13.17b), GF = 11.6637 x 1 0 - 6 GeV - 2 . The sign of GF is taken as positive as determined by the presumed underlying weak-vector-boson positive-definite form of GF as given by (13.9). To improve upon this estimate of GF, we must account for higher-order radiative corrections to the lowest-order graph of Figure 13.1 [see e.g. Section 15.D and Marshak et al. (1969)]. As expected, such corrections alter (13.17b) to 0(a2) and shift GF by 0.2% in GF = 11.6637 x 1 0 - 6 GeV - 2 . This latter result is then the "bare" weak coupling as stated in (13.8). Corrections to (13.8) due to the nonvanishing electron mass and estimates of higher-order weak interactions contribute in the fourth significant figure and approximately cancel [see e.g. Nagels et al. (1976)]. Neutrino Cross Sections. The leptonic current-current decay hamiltonian (13.11) can be crossed into the scattering configuration V + e~ -»v + //", leading in a straightforward way to the unpolarized total cross section (Problem 13.2) a(Ve~ -> v/O A
{S
"
m 2)2
'
,
(13.18a)
where s = (kv. + pe)2 = (kv + p^)2 is the usual Mandelstam squared-energy invariant. Slightly above threshold at s ~ 2m2,, corresponding to a highenergy v' laboratory beam of a>v. ~ m2, /me ~ 20 GeV, the cross section (13.18a) is very small (a ~ 10" 39 cm2) and not yet seen in high-energy accelerators. It is even smaller for elastic scattering, such as v + e~ -• v + e", recently measured, for U 2 3 5 fission-reactor antineutrinos of low energy 1-5 MeV, to be a ~ 10" 4 5 -10" 4 6 cm2 (Reines et al. 1976). The latter cross section is roughly consistent with the V — A prediction (Problem 13.2) a(ve -> ve )•
G2F(s-m2f 12s3
1+ 3
/s + m 2 \ 2
\s-m2)
(13.18b)
but is even more compatible with a W+, V — A coupling in the s-channel, combined with a Z° Weinberg-type neutral current in the t-channel (Weinberg 1967). While such a tiny cross section is very hard to measure in Earth laboratories (the ve cross section data were collected over a period of 18 years), its crossed version e~e+ -> vv is thought to be the dominant energyloss mechanism cooling superheavy stars. Given that the neutrino-lepton structureless weak current is thought to be well understood (like the photon-electron QED current), neutrino beams
266 Lowest-Order Weak Interactions are now being used to probe the structure of hadron targets, as in v' + p -* H+ + n, v' + p, n+ + anything. Results of such experiments will certainly play a decisive role in probing the structure of matter, but at this point analysis of the initial data is beyond the scope of this book.
13.D Neutron P-Decay Now we turn to the prototype semileptonic weak decay n -> p + e" + v. While the lowest-order /J-decay graph (Figure 13.2) is kinematically similar to \i decay, it differs in one major aspect. The Q-value, or momentum transferred from the muon to the neutrino, is large (~ 105 MeV), but that transferred from the neutron to the proton in /?-decay is very small, since Awjy = m„ — mp ~ 1.29 MeV. That is, muon decay has an extremely relativistic three-body phase space, while /?-decay has an extremely nonrelativistic recoil proton with Ep « mp but a recoil electron that can be relativistic, since AmN ~ 3me. Matrix Element. Like the muon-decay matrix element, Jf fi for /?-decay has the product of the lepton and hadron currents appearing twice in (13.7), so that ^fi = -j=jlXp\jl + i2\n>,
(13.19)
where we have replaced GF by G„ in order to test the universality hypothesis for these couplings. The charged weak hadron current in (13.19) is of the form Va — Aa according to (13.4), but because of the aforementioned nonrelativistic approximation, the momentum-transfer invariant in the neutron rest frame is t — (Am^)2 ~ 0. Then making a nonrelativistic reduction, we have for FftO) = L rf) = 9A,
(13.20a)
Ca = (1, -gA«)
(13.20b)
where we have taken both nucleons at rest and applied the Dirac identity To Y*75 =
2 r =
! ^Cauer(l-iy5)v-v.
•P Figure 13.2 Neutron beta decay.
(13.21)
Neutron /?-Decay 267
Given (13.21), we may proceed immediately to the unpolarized-spin sum by again using the lepton trace (13.13a) along with the two-component spin | TidC,
= 5*5,0 + g2A5u,
(13.22)
which follows from (13.20b) and Tr a,-^ = 2<50- with i, j = 1, 2, 3. Then we find in a straightforward way (Problem 13.3) \^fi\2^(2mN)2^-^TrC*C,)1Je% 1
2
(13.23) 2
= (4m„) G„ £eco[(l + ve cos 0ev) + 3^(1 - K cos 6ev)]. Correlations. To probe the structure of (13.23), it is natural to observe the angular correlation between the emitted electron and antineutrino for unpolarized neutrons. To this end we convert the three-body phase space (10.13) over dEe dEp to dEe d cos 6ei via the identity mN dEp = peu> d cos 6ei. Then we have 1 G2 -~ (1 + 2>gl)(b2pe Ee(l +a cos 6ei) dEe d cos 6ei, a = (1 - g2A)/(l + 3g2A).
(13.24)
(13.25)
Since pe • k^ oc cos 0ei is a true rotation scalar, this correlation term conserves parity, and the observed values of a distinguishes between possible S and/or T currents and the favored V — A structure. Furthermore the present observed angular distribution (13.24) gives a = —0.103 ±0.0004, or from (13.25), gA « 1.2634, quite near other determinations of gA. It is also possible to extract gA from other neutron measurements. For polarized free neutrons, the ratio of the electron asymmetry
=
1 ~gA
i-i-x~>f\
(m6)
B 7*y^rTT7/
The experimental ratio A/B = 0.1193 then gives gA = 1.2709. The combination of these two values for gA is very close to the overall present accepted value ofgA = l .2695 ± 0.0029 (Particle Data Group 2004). Unpolarized Rate. Returning to the angular distribution (13.24) for unpolarized neutrons, we integrate over cos 9ei to obtain the electron energy spectrum dV dEe
r1 , '\.J_,™ d
„ / dY \ ""[I^LOEJ~2? "\d cos 6evdEe}
G2 (1 + In
3
^''-E"(m7>
268 Lowest-Order Weak Interactions where the antineutrino energy is a) « AmN - Ee. To calculate the total decay rate, we integrate (13.27) over the electron energy with £^in = me and £™x = AmN. In terms of the dimensionless ratio £ = me/AmN = 0.3951 and dimensionless variable x = Ee/AmN, the integral over (13.27) is X = | dx x{x2 - £?(\ - xf *'4 £ (1 - £2)1/2 I* - ji - 8 + ( 1 _ , . 2 ) 1 / 2 c o s h _ 1 r 60
(13.28) = 0.01575.
The total unpolarized neutron decay rate is then r„ = H
(1 + 3gA)(AmN)5X.
(13.29)
Unfortunately, at the present time the free-neutron lifetime is only known to 2% accuracy (as we shall soon see, we want to know it well within this error): T„ = 885.7 ± 0 . 8 sec, gA = 1.2695 leads to the coupling strength Gn = 11.772 • 10~6 GeV~ 2 , very close indeed to the accepted Fermi coupling GF = 11.6637(1) • 1 0 - 6 GeV~ 2 . Due to the neglect so far of radiative corrections to neutron j8-decay and the use of an approximate value of gA in the rate formula (13.29), however, we are not yet finished with our identification of Gn with GM and our test of the universality of the weak currents in the current-current hamiltonian J^,. Nuclear P-Decay. Beta-decay transitions in nuclei also pin down the parameters in the V — A theory. One advantage of nuclear measurements is that the 0 + ->0 + "Fermi" transitions, say in 0 1 4 - > N 1 4 or Al 26 -> Mg 26 , suppress the "Gamow-Teller" gA a term in (13.20b). This allows us to zero in on the strength of Gp (G„ now called Gp for nuclear transitions), independent of gA. After taking account of Coulomb and radiative corrections, the latter in a slightly model-dependent fashion in the W+ theory (Sirlin 1974), one obtains the "bare" jS-decay weak coupling constant averaged over various nuclear 0 + -> 0 + nuclear transitions [see e.g. Nagels et al. (1976) for recent data coupled with the Weinberg-Salam value mzo « 85 GeV], Gp = (0.9996 ± 0.0009) x 10" 5 m; 2 .
(13.30)
Universality dictates that when the experimental errors on the free-neutron decay are reduced, we must find G„ = Gp as given by (13.30). In any case, this value is almost 3% less than the muon decay constant (13.8), 1.026 x 10" 5m~2, a discrepancy which has profound implications for the theory, as we shall show shortly. A second application of nuclear transitions is an accurate determination of gA for Gamow-Teller AS = 1 transitions which enhance, rather than suppress, the gA a term in the nuclear current analogous to (13.20b). Transitions such as He6(0 + )-> Li6(l + ) determine gA most accurately as 1.25 + 0.01, the present accepted value. Nuclear /J-decays also allow us to probe the heretofore suppressed
Charged-Pion Decay 269 anomalous-magnetic-moment vector form-factor part of the weak hadronic current in (13.4b) (Gell-Mann 1958). The magnetic-dipole (pure axial-vector 1 + ->0 + ) transitions in B 12 ->Ci2e~v and N 1 2 -• C 12 e + v have electron and positron energy distributions containing a term (/ip — ^„)< x (pe + k) • j l e p due to interference between the forbidden vector and allowed axial-vector amplitudes. Such "weak magnetism" terms have in fact been detected in the laboratory, further confirming the hypothesis of CVC for weak currents. There are many other interesting consequences of nuclear /?-decay studies, such as on the possible existence of "second-class" weak currents or for the theory of muon capture. Suffice it to say that such studies lend further support to the V — A current-current hypothesis. Pion P-Decay. The ne3 decay n+ ~*n°ev is quite similar in structure to nonrelativistic neutron /?-decay, with the hadron mass difference Amn = mn+ - m„0 « 4.60 MeV <mn+ x 139.6 MeV. The CVC hypothesis then guarantees that only the spinless matrix elements of the vector current °c (p' + p\ enter [and not (p' - p)J as in (13.4b), with Fn(q2) x Fn(0) = 1. It is then straightforward to obtain the total n+ decay rate analogous to (13.29), as found in Problem 13.3,
The experimental ne3 rate of 2.58 x 10" 22 MeV then yields a value for the weak coupling constant for n transitions (neglecting radiative corrections), Gn « 10.73 x 10~ 16 G e V - 2 .
(13.32)
Clearly G„ is approximately consistent with G^^n, and G^; what is needed are more accurate measurements of Gn and also G„ to see if both of them are really the same as Gp, as one would expect on the basis of the CVC hypothesis and universality.
13.E Charged-Pion Decay The ne3 decay just considered is not the dominant decay mode of n 1 ; it is suppressed by the small mass difference Amn and also by three-body phase space. Rather, n+ decays primarily into two leptons, either fi+v' or e+v, as shown in Figure 13.3. Since m„ x 105.7 MeV and me x 0.5 MeV, one would expect that n+ -*e+v has much more mass available than n+ ->/i + v to enhance the former decay rate. In fact it turns out the n^ rate is not smaller, but ~ 104 times larger than the ne2 rate. We shall show that this turnabout is a direct consequence of the V — A (in this case axial) structure of the hadronic current. n,2 Decay Rates. The hadronic transition for nn decays is the pion-tovacuum 7t+ ->0 charged-axial-current matrix element of (13.4c). Lorentz covariance and parity prevent the vector current from coupling in this case.
270 Lowest-Order Weak Interactions
H-W
•v,v Figure 13.3 Charged-pion decays {R^2, i^ei)-
Then the current-current weak hamiltonian with coupling constant G^ (replacing Gw) becomes °ti = ~ \ =
<0\Al-i2\n+(q)}uvf(l
- iy5)v{
-ifnGnUj(i-iy5)Vl
= i/,G>,ii ¥ (l + iy5)vh
(13.33)
where we have used 4 — h + ^ f° r nil)^ KPi) + V (^) a n d t n e Dirac equations i>iVl= —mlvl, uvljt = 0. Summing over the final lepton spin states according to £ K ( l + iy5)v,\2 = Tr{£(l + iy5)(P, + «,)(1 - iy5)} = 8/c • p„(13.34) we evaluate the general rate equation (10.17b) in the pion rest frame to find for pCM = {ml+ - mf)/2mn+, A 2
PCMK/JI 2 _fl(G n) \ n) = Snm o _ 22.+ = —7= 4n
Y %
mf(m2n+-mff =3 mL
•
I1335)
Branching Ratio. Now that we know the predicted V — A rates for the two lepton decays (13.35), we can divide one by the other to obtain the branching ratio Y
fre+:\=iii*
~ iS=123 * 10-.
(i3.36)
K r(n+^n+v) ml{ml+-ml)2 ' This prediction agrees very well with experiment, (1.24 + 0.25) x 10 ~4, a result which we have already suggested earlier is hard to understand for any other model of JCw. Clearly this suppression of the ne2 rate is due to the factor (me/m,,)2 in (13.36). Angular-momentum conservation in the pion rest frame requires both outgoing leptons to have the same circular polarization (helicity). Since the neutrino must be left-handed, so must the outgoing massive anrilepton. But we know that relativistic antileptons must be right-handed. Consequently the mass factor me in the ne2 amplitude acts as an angularmomentum suppressant of the normally right-handed relativistic decay positron. Put another way, the m^ factor in the n^ amplitude does not greatly inhibit the slow-decaying antimuon from being polarized left-handed.
Cabibbo Universality
271
Pion Decay Constant. Assuming the universality of the current-current weak hamiltonian along with the V — A hadron structure, we may take the axial weak coupling constant G* as equal to the vector coupling constant GK = Gp. Then the dominant n^ lifetime of xn+ = 2.60 x 10" 8 sec or Tn+ = 2.53 x 10" 14 MeV along with (13.30) allows us to solve (13.35) for/, (taken as positive by convention), v
m„(mj|+ - ml)Gp
'
Inclusion of radiative corrections and present experimental errors in Tn+ and Gp alters the result (13.37) only slightly to/* = 93.3 ± 0.1 MeV. Thus we now know the value of the hadronic form-factor scales gA = gA (0) and f„ =fn{ml) from the weak semileptonic decay formulae (13.25) or (13.26) and (13.37). We have already seen in Section 12.F the importance of these couplings for the Goldberger-Treiman relation and the PCAC hypothesis.
13.F Cabibbo Universality Hypercharge-changing weak semileptonic decays can be treated in a manner similar to nl2, nn, and neutron /?-decay. Again the weak couplings G(A Y = 1) are all about the same magnitude. The natural way to try to link the G(AY = 0) to the G(AY=1) couplings is via the internal SU(3) symmetry, an extension of SU(2) isospin symmetry linking kaons with pions and hyperons with nucleons, which manifests itself in strong interactions [see e.g. Gell-Mann and Ne'eman (1965)]. Even after SU(3) symmetry has been introduced, however, one finds that G(AY = 0) ~ 4G(AY = 1). The way out of this predicament was suggested by Cabibbo (1963), who showed that the two mismatches of G(A Y = 0)/G„ « 0.97 and G(A Y = 1)/G„ « 0.22 can be related to a single parameter, the Cabibbo angle. A Y = 1 Weak Currents. We have seen that the weak hadronic currents which occur in A Y = 0 decays transform like an / = 1 charge-raising or -lowering operator Ix ± il2. The AY = 1 transition currents from (13.3b) could have AI = j , f. For K+ ->• n°e+v and K° -> n~e+v decays, the A/ = % transitions occur in an isotopic (Clebsch-Gordan) proportion according to the Wigner-Eckart theorem (2.72),
^ V
{
'
Then comparing the A/ = \ rate prediction with experiment, r(K°-+n-e+v) r(K+ ->7t°e+v)
12, (1.93,
A/ = \, experiment,
(13.39)
we see that the AI = \ rule is approximately valid. Since the ClebschGordan coefficients and yjl ratio in (13.38) are reminiscent of the nNN
272 Lowest-Order Weak Interactions
isospin structure of Section 12.B, we may by analogy introduce an isospinor T' matrix (called u-spin) in the isospin space of the J = y kaon as the column index and the AI = \ current as the row index. The isovector index i refers to the pion, i.e., <7i' | J(AI = j) \ K} oc T'. This isospin structure immediately leads to (13.38), including sign, provided one adopts the same phase convention as in Section 12.B, that is, |7C±> =
72|7rl±f7r2>'
K= K
( K°\
KC=
(KK°)"
(13 40)
-
The antikaon Kl3 decays in this T-matrix representation also produce the Clebsch-Gordan analog of (13.38)—see Problem 13.4. Use of this t?-spin T-matrix is a poor man's version of SU(3), but it will suffice for our purposes. Now we are ready to define the axial-vector and vector-meson matrix elements of AY = ± 1 hadronic weak currents. By way of comparison we write the AQ = — 1 axial transitions as <0\Al-i2(AY
= 0)\n+(q)y = y/2ifnqfl,
< 0 | ^ ( A Y = -l)\K+(k)}
= ^2ifKk„,
(13.41a) (13.41b)
where A^AY = — 1) transforms like AI = j , AY = AQ = — 1. The internal SU(3) symmetry requires that the kaon decay constant fK be the same asfn. Likewise for AQ = — 1 vector currents we define, according to the discussion in the previous paragraph, < 7 r ^ ) | ^ - ' ' 2 ( A Y = 0)|7C"(p)>
tf(q)\V,,(AY=-l)\K(k)y
ie,-1-2-*F,(t)(p + «)„
(13.42a)
^[um + qi+f-m-qn (13.42b)
where the second form factor in (13.42b) reflects the fact that V^ is not conserved, but d • V(AY = — l)oc/ + (m£ — ml) + tf_. By analogy, in (13.42a), d • V oc Fn • 0 + tf- = 0 requires an/_ -type form factor to vanish, by CVC. The proportionality constant — 1/^/2 in (13.42b) is chosen so that the isospin vector in (13.42b), \xl, is the SU(3) extension of the isospin vector in (13.42a), (/
Cabibbo Universality 273
element and rate analogous to (13.33) and (13.35) are (see Problem 13.4) *fi=-^<0\Aa(AY=
-l)\K
+
(k)>&p = ifKGimiuv(l
+ iy5)vh (13.43a)
r
2
PCMK/,12
=fUGi)
mf
d~ K - - w?)2-
(13.43b)
Then substituting the experimental ratio F(K+ -^n+v')/r(n+ 1.34 into (13.35) and (13.43b) leads to
->/i + v') =
«« =
, ' 2 . " = ^~^~
IKGA
— £ = 0.275.
(13.44)
This suggests that the AY — I weak coupling is suppressed relative to the AY = 0 coupling by a factor of ~ 4. For the Ke3 decay K+ ->7r°e+v, the relevant charge configuration of (13.42b) is (Problem 13.4) = -l)|K+(fc)>= -^=[/+(0(fc + ^ + / . ( t ) ( f c - 9 U
{n%)\Vll{AY
(13.45) 2
2
for t = (k — q) = ml + m . — 2mK con in the kaon rest frame. Proceeding as in the case of ne3 decay, the current-current hamiltonian density folded into the three-body phase space kr.ds to an energy spectrum (see Problem 13.5) dT dco„
mK{GvK)2f\(t)ql I2n3
(13.46)
where the /_ form factor is weighted by me
(13.47)
+
(13.46) integrates to the total K ->7t°e v width (see e.g. Marshak et al. 1969 and Problem 13.5) T(Xe+3) = 0.57924(1 + 3.697/1+) ^ K ? / ^ ? m * , (13.48) 15367T where the numerical factors in (13.48) correspond to complicated functions of the ratio mn/mK. For the experimental rate r(K*3) « 0.256 x 10 ~ 14 MeV and slope X+ « 0.030 (the latter near the vector-dominance value of A+ = 0.025—see Problem 13.5), (13.48) gives a hyperchange-changing vector coupling strength (including a 0.45% reduction due to radiative corrections) of A(0)G£/G„ = 0.217 ±0.001.
(13.49)
274 Lowest-Order Weak Interactions
Thus, once again we see that the A Y = 1 weak coupling is suppressed relative to the AY = 0 (and purely leptonic) weak coupling by a factor of ~ 4. This result is further supported by AY = 1 hyperon transitions such as A -> pe~v relative to neutron /J-decay n = pe~v. Cabibbo Angle. With regard to the universality of the current-current hamiltonian, we are now faced with two problems: (1) the aforementioned suppression of AY = 1 weak transitions relative to AY = 0 transitions, and (2) the slight discrepancy between the AY = 0 coupling (13.30) and the purely leptonic bare coupling (13.8). Cabibbo (1963) suggested that these two discrepancies are manifestations of a fundamental mismatch between nonstrange and strange hadronic V — A, AQ = — 1 weak currents as measured by an angle 9C (the Cabibbo angle): f;d = cos ecJl-i2{AY
= 0) + sin 0J„(AY = - 1 ) .
(13.50)
If this is the correct form of the hadronic current to be used in the currentcurrent Jfw, then for AY = 0 transitions, (13.30) is modified to
cos ec = p - = ^ r = °- 974 ± 0 0 0 2 ' CJ M
(13-51)
l.UZO
while the A Y = 1 transitions (13.44) and (13.49) become, respectively, Y tan 9C = 0.275,
/ + (0) sin 9C = 0.217 ± 0.001.
(13.52)
Solving for 9C from the nuclear /J-decay ratio (13.51), we find 9C « 13.1°. This is very close to 12.6°, the value implied by the Ke3 rate in (13.52) if we set /+(0) = 1. A value for 9C only slightly larger follows from hyperon semileptonic decays. Furthermore, eliminating 6C from (13.52), we obtain /*//,/+(<)) =1.23 ±0.01,
(13.53)
which is a perfectly respectable measure of the breaking of SU(3) symmetry in the kaon decay constant relative to the pion decay constant for/ + (0) = 1. Actually it can be shown that/ + (0) is slightly less than unity, which serves to increase 9C closer to the A Y = 0 value and also lower the ratio of/K //„ from (13.53) toward the SU(3) limit of unity. While there is general agreement as to the existence of the Cabibbo angle, its origin is not fully understood. One scenario suggests that 6C arises from a slight "spontaneous symmetry breaking" of strangeness or hypercharge at the strong-interaction level which then forces a redefinition of these quantum numbers through a Cabibbo rotation. Weak interactions, on the other hand, conserve the original or bare hyperchange quantum number. Since it is the strong interactions which define what we mean by an S or Y, the Cabibbo-rotated hadronic current (13.50) must then be used in Jf w.
Nonleptonic Decays
275
13.G Nonleptonic Decays Finally we turn to weak decays which do not involve leptons, referred to as nonleptonic decays. We know that hadronic decays such as K -> 2n, 2>n or E -» Nn are in fact weak, because the lifetimes are of order 10" 1 0 -10" 8 sec, about the same as semileptonic weak decays and far longer than the usual strong decays, which are of order 10" 23 sec. These specific decays also violate /, I3, and Y conservation. They usually violate P and C—K3n decay, however, conserves P and C. Moreover the hypercharge-conserving transitions N -> Nn have a small weak parity-violating component. We shall comment only upon hypercharge-changing nonleptonic decays in what follows. Since these transitions also involve strong interactions of the hadrons among themselves, our strategy will be to lump together the weak effects in an Jtw treated in first order as if it were a strong vertex, and then approximate the amplitude by an "appropriate" Feynman pole diagram, as first done by Feldman et al. (1961). A/ = \ Rule. Even if we assume a current-current structure for Jf £,', we cannot automatically infer a simple isospin transformation property, because both hadronic currents have isospin. For AY = 1 decays, universality would require one charged strange and one nonstrange current in the combination C ' ( A y = 1) ~ Gw cos 9C sin 6C Jl(AY = 0, A/ = 1)/(AY = 1, A/ = i), (13.54) with no neutral currents occurring in (13.54) since K -> nvv vanishes. Since these hadron currents are not structureless probes, as are lepton electromagnetic and weak currents, we cannot pick out specific hadron states contributing to (13.54). Instead we can only infer the isospin transformation properties from (13.54) of 1 x \ = \, f; that is, JtT^(AY = 1) ~ AI = \, \. Yet experiments do suggest that 3tf"£ transforms predominantly like AI = j . This is especially easy to see in the K2n system, consisting of K° -> 7t+7t", 7r°7r° and K+ -» n+n°. (It is traditional to consider CP eigenstates of K° and K°, KL,S = (K° ± K 0 ) / ^ , but we will stick to K°, relegating KLS to Problem 13.7.) The K2n0 decays occur much more rapidly than the K2n decay, with x(K2n^) ~ 10" 10 sec, whereas x{K2^) ~ 10" 8 sec, indicating some sort of suppression in the latter case. The only difference between these decays is that the final state of K2n, n+n° must have 1 = 2, since Bose statistics rule out the 1=1 antisymmetric state and the n+ charge with / 3 = 1 rules out / = 0. On the other hand, the n+n~ and 7t°7t° final states of K2„0 could have / = 0, 1, 2 and / = 0, 2, respectively. Since IK = \, we must have JPl\AY = 1) ~ AI = f for K^K decay (to link lf = 2 with It = | ) but Jf i'(AY = 1) ~ AI = j , \ for K2na decay. The suppression of the former rate then suggests AI = \ dominance of Jf J'(AY = 1). Other predictions due to the nonleptonic Al = \ rule follow from the use
276 Lowest-Order Weak Interactions
of Clebsch-Gordan isotopic coefficients—or better still from our point of view, from the use of u-spin r-matrices for AY = ± 1 transitions, already employed in the semileptonic currents (13.42). Treating Jf^(AY = 1) as a "particle" (called a spurion) with I = %, I3= — \ in the initial state or J 3 = \ in the final state, we may combine the isotopics of K -»• Jtf"^ + nl + nj in a manner similar to the strong-interaction process of nN scattering of Section 12.C. That is, the K2n amplitude has the isotopic form adiJ + bie,jkxk in the isospinor space of the column K and row Jf^. But the isotopic extension of Bose statistics requires the two final-state pion isovector indices to be symmetric under interchange, which means that b = 0. Then, since 81'12,3 = 0 and 5 1 " i2 > 1+i2 = 2d33 = 2, we see that for K-+K+, K° <7t+7r°|jfw(A/ = ! ) | K + > = 0,
(13.55a)
+
0
<7r 7t- | Jfw(AI = i ) | K°> =
(13.55b)
The former A/ = \ relation agrees with our expectations as stated in the previous paragraph, and the latter relation leads to a rate prediction which is close to experiment,
r(K°^n+n-)^\2, r(K°->7tV)
A/= ±, |2.21,
experiment,
K
'
where the factor of 2 for AI = \ corresponds to the Feynman rule of \ in the 2n° rate. After the mass-difference effect of m,+ — mn0 « 4.6 MeV is removed from the phase-space factors in (13.56), the prediction (13.55b) deviates from experiment by about 5%. This agrees with the experimental rate enhancement of K2no over K2„ by about 500. A similar analysis can be applied to other weak nonleptonic decays such as K -> 3rc, A -> Nn, and H -> An (see Problem 13.6). For example, the AI = \ spurion [Jf W(AJ = j)] can be folded into the K3K amplitude in an isospin-invariant way, symmetric in all three pion isotopic indices, <7rV'7r* | Jf W(A/ = i) | K> = dijxka + 5ikzjb + # V c ,
(13.57)
where the amplitudes a, b, c are linear combinations of the various charged K3n decay modes (see Problem 13.6). Vector-Meson Dominance of K° -* n°n°. The nonleptonic AI = \ rule offers only a clue to the underlying dynamical structure of Jf£!(AY =1). Alternatively we may decompose M"^,(AY = 1) into parity-violating (pv) and parityconserving (pc) parts, C ' ( A y = i) = jfpv + #epc,
(13.58)
where only Jfpv connects the initial and final states of K2n decays (since both K and 7t are pseudoscalar 0" states, the spinless structure of 3tf requires (rax | Jfpc | X> = 0). We may then consider the K*(895) vector-meson dominance of this pv transition through the Feynman graphs of Figure 13.4 (Sakurai 1967a). Since IK* = \ and JK* = 1 ~ (like the p meson but with Y = 1,
Nonleptonic Decays 277 ,v
7T«
• ir
+
- +•
k ^
TTl
• *tf
pvi
q'
.q
V
TT
Figure 13.4 Vector-JC* dominance of nonleptonic K2n0 decay. S = 1), the weak pv transition K* —^->7r can be written as <7r°(q) | J*T | K*°(q)> = i ( ^ p v W
(13-59)
pv
with the scale (Jf )reK» to be determined shortly. The strong transition K -» K*7t can be expressed in terms of vector matrix elements (13.42b), again using the u-spin r-matrices, (n°(q)\ F„(K*°)|K°(/c)> = -h\f+m
+ «)„ +f-m
~ «U (13.60)
with T3 -»• — 1 in this case. The CVC, vector-meson-dominated current (13.60) corresponding to the strong coupling gK*Kn<ji\V,l\fC)K* has g2K*Kn/4n « 3.1, quite close to its SU(3) partner (12.56), g2^/4n ~ 3.0. Given (13.59) and (13.60), the standard Feynman rules applied to Figure 13.4 for K° -»7i°7r° lead to S /£ = 3*(P/i)(-0«'(-^ pv W«v +
d*(Pfi)(-i)i(Jl?n«K*q'v
= -i5\Pfi)W\K*
q2 - m2K.
(13.61a)
i{gVfi-
1
q -ml* GK.KM - ml)lml.,
(13.61b)
where we have approximated /+ -»1, /_ -» 0 and used k = q + q', q'2 = q2 = ml in order to convert (13.61a) to (13.61b) (see Problem 13.8). Then denning Sfi = -i8\Pfi)(it°n° | Jf p v |K°>, (13.61) gives <TIV
|^
\ K°) = i{^"U- g„ K« " ^ T ^
•
(13-62)
We may compute the magnitude of this matrix element from the experimental rate and (10.17), i.e., r(X°-7rV) = ^ ^ ^ ! > J !
«
L 1 5x
10" » MeV, (13.63)
where a factor of \ is included in (13.63) because the final pions are identical. For the CM momentum qn = 209 MeV, (13.63) and (13.62) give | <7t°7t° | jfpv | JC°> | * 260 eV,
| (^f p v U. | * 150 eV.
(13.64)
278
Lowest-Order Weak Interactions
We shall test this scale shortly. A similar analysis for K^n decay for Jfpv ~ A/ = j suppresses this amplitude to an order of magnitude smaller than experiment, but at least consistent with A J = \ rule (see Problem 13.8). E + ->/ra° Decay. Next we turn to the AY = 1 hyperon nonleptonic decay E + -> p + 7t°. Since in this case both Jfpv and Jfpc have nonvanishing matrix elements, we write <7r°p | Jfpv + Jf pc | Z + > = u,{iA + By5)uz
(13.65)
[the factor i is included for CPT reasons and indicates parity violation as in (13.62)], where A and B are the (real) pv and pc decay amplitudes, respectively. Experimentally, the unpolarized decay rate r ( Z + ->p7t°)«4.25 x 10" 12 MeV and the observed asymmetry distribution ~ (1 — 0.98<
\B\
2.7 x 10 - 6
AB < 0.
(13.66)
Now we must invoke dynamical arguments in order to relate (13.66) to (13.64). Vector-meson dominance as in Figure 13.5, again can only be applied to Jfpv (since (n | Jfpc | K*} = 0), i.e., to the pv amplitude A. The hyperon SU(3) analog to (13.60) is (with Z + = I 1 + i 2 />/2)
uM,
(13.67)
where we drop the anomalous magnetic form factor, since a^q" clearly will not couple to (13.59). The coupling 6fK*i:p
)nK,
c?
•i{9^-9,9jm2K,)
ml*
(13.68a)
QWLpUpfUz,
A= (^pvU^ i , K -
mp)lj2m2Kt.
(13.68b)
ir-
I
+ (pv
Figure 13.5 Vector-K* dominance of nonleptonic, parity-violating S + -> pn° decay.
Nonleptonic Decays 279 • 7T' -7TV
z+
(pc
r .
2*^-N P —-(pc)——
«
Figure 13.6 Baryon pole dominance of nonleptonic, parity-conserving S + -y pn° decay. Taking the Kln scale (13.64) and |g K *i P | ~ 6 along with mL « 1190 MeV, mKt « 895 MeV, (13.68b)then predicts \A\ « 0.2 x 10" 6 , reasonably close to experiment, (13.66). Since K* dominance only applies to the pv amplitude A, we must test the scale of the pc amplitude B in some other manner. The natural choice in this case is to pole-dominate B with the S + and p poles as shown in Figure 13.6. These graphs, like Figures 13.4, 5, can be treated in a Feynman-diagram sense if we apply
(13.69)
along with the strong couplings gnoppupy5up, 0,01+3:+ uzysuT, where SU(3) gives gnox+-L+ ~ IffnOpp- The standard Feynman rules then lead to •i =
Hh + mp)
(-i)g«oppupy5
Pi m„
uA-WU
m + (-0(^pc)p5:«p Kh + i) ysM-fy-.nOL+I+5
Pi ml
(13.70) where p | = m| and pj; = ml in the second and third graphs of Figure 13.6, respectively. Applying the free-particle Dirac equation in both terms of (13.70) and comparing with (13.65) then gives the pc pole amplitude (Problem 13.9) B
E0nOpp _ (^pcW
(
1
0*01 + 1 +
mz-mpmp \\ gn0pp Applying the experimental scale (13.66) to (13.71), we find
\(^n
pi 1
150 eV. pv
(13.71)
(13.72)
Similar contributions to A in terms of (^ ) p i are substantially suppressed, partly because mz — mp -*• mz + mp in the pole denominator and because (3f p v ) p I is also suppressed in the SU(3) limit. The fact that the nonleptonic hyperon scale (13.72) is. about the same size as the meson scale (13.64) is perhaps somewhat fortuitous, but because a V — A structure of the currents gives J*fpv ~ V • A + A • V and
280 Lowest-Order Weak Interactions jtifp* ~ V • V + A • A, this equality suggests that V — A operates in the nonleptonic as well as in the semileptonic and leptonic weak sectors. A more detailed analysis using the techniques of current algebra (see Section 15.H) roughly confirms this V — A structure of Jtf"£(AY =1). Model-dependent dynamics indicates that (13.72) may be related to the Cabibbo scale in (13.54), Gw cos 9C sin 9C [see e.g. McNamee and Scadron (1976)]. Other hyperon decay amplitudes can be explained in terms of (13.72), but this depends upon a deeper understanding of SU(3) than we have discussed in this chapter. Moreover, (13.72) also sets the proper scale for the pc Z + -• py weak radiative decay (see Problem 13.9). Interestingly enough, the one aspect of nonleptonic decays that is not satisfactorily understood is the origin of the empirical AI = \ rule. Suffice it to say that while nonleptonic weak interactions contain in principle all of the complications of weak and also strong interactions, one can partially probe their structure through the use of hadronic currents, isospin, and simple Feynman diagrams, as suggested in this section. For further reading on weak interactions, see e.g. Bjorken and Drell (1964), Muirhead (1965), Okun' (1965), Wu and Moszkowski (1966), Gasiorowicz (1966), Sakurai (1967), Bernstein (1968), Marshak et al. (1969), Commins (1973), Lifshitz and Pitaevskii (1974), and Taylor (1976).
CHAPTER 14
Lowest-Order Gravitational Interactions
Having investigated the significance of lowest-order covariant Feynman "tree" (no loop) diagrams for strong, electromagnetic, and weak interactions, it is quite natural to try to extend the approach to the fourth fundamental force, that of gravity. The problem is that the gravitational interaction is so weak that quantum gravity corrections to the classical force will, in all likelihood, never be detected. Moreover, like the current-current weak interaction, higher order corrections are usually divergent. Thus, the major justification for looking at quantum gravity is that it may give us a deeper understanding of the classical newtonian force, and if a unified theory of forces is ever developed in detail, it most certainly will have to include the gravitational interaction. Accordingly, we attempt to construct quantum "graviton" wave functions, propagators, vertices, Feynman rules and diagrams in the same spirit as for the other three forces. Then we briefly discuss the connection between this linearized quantum gravity theory with Einstein's nonlinear classical theory of general relativity.
14.A Graviton Wave Function and Propagator The principle of equivalence teaches us that in the presence of gravitational sources, space is curved as specified by the space-dependent metric g^{x). Nevertheless the quantum gravitational (Planck) length as formed from h, c, and G/c2 w 7.4 x 1(T29 cm/g, i.e., /* = (Gh/c3)1'2 ~ 1 0 " " cm, is so much smaller than any other quantum length scale that it is usually possible to work 281
282 Lowest-Order Gravitational Interactions
in a "flat Minkowski space" and approximate the quantum metric by the special-relativity form
gM - gfAx) = n,v = I
_,
.
(14.1)
With this approximation we can proceed in a manner paralleling the elementary-particle quantum description of the other three fundamental forces. Spin-2 Graviton. Since we know that Newton's law of gravity is long range like Coulomb's law, we may presume the existence of a quantum gravity force mediated by massless particles called gravitons. Furthermore we know that even-spin exchanges (like the pion) are fundamentally attractive, so that observed gravitational attraction of matter requires the graviton spin to be sg = 0, 2, Feynman (1962) suggests that a distinction between sg = 0 and sg = 2 can be made on the basis of the fact that the gravitational attraction between masses of a hot gas is greater than for a cool gas; i.e., that energy is an effective form of gravitational mass. This observation corresponds to a velocity-independent gravitational potential between two massive bodies, which, because E = ym, requires an interaction energy for spin-0 exchange to be proportional to y" l = (1 — t;2)*. This incorrectly predicts that the attraction between masses of a hot gas (i;|) is less than for a cool gas (vl) with the opposite being true for a spin-2 graviton theory. Another argument which rules out spin-0 gravitons is that they cannot generate any light bending, the latter being proportional to the trace of the em stress-energy tensor which vanishes. While it is still possible for the graviton to have a very small spin-0 component (as in the Brans-Dicke theory), we will henceforth assume it to be pure spin-2, sg = 2. Linearized Field Equation. Recall from Section 4.C that the photon field (Maxwell) equation for the covariant vector potential A^ which is independent of gauge is \3Ati-dfl(d-A)=rtr(x). (14.2) Then a conserved em source current, d • j e m = 0, is consistent with the divergenceless nature of the left-hand side of (14.2). In the same manner, the "linearized" graviton tensor potential h^ = hV)t (symmetric in the space-time indices for a spin-2 graviton) must satisfy a field equation of the form
D V - (S,dxhn + 8%h„a) + d„dvK" + tl^FfK, - UK*) = -JUX) (14.3) (up to field transformations of the form /i„v -»h^ + Xq^h*). The structure of (14.3) is the only possibility which is both symmetric in /iv and divergenceless (see Problem 14.1) with;'*rv =j% and 3*7^ = 0. The choice of the sign accompanying the conserved gravitational source current jl\ in (14.3)
Graviton Wave Function and Propagator 283 is anticipating the attractive nature of the gravitation force, but in any case this attraction will be built into the Feynman rules independent of the sign in (14.3) via the structure of the spin 2 graviton propagator. Since (14.3) is in general an algebraic mess, it is convenient to specialize to a specific gauge. Just as the Lorentz or Feynman gauge d A = 0 simplifies (14.2), so the harmonic or de Donder gauge d"/i„v = # v V
(14.4)
simplifies the form of (14.3). Defining the barred operation for tensors (not to be confused with the barred operation for Dirac bispinors),
the field equation (14.3) subject to the gauge condition (14.4) reduces to (Problem 14.1) D V = -/; V , d%v = o. (14.6) Noting that haa — —h*, the barred operation is its own inverse,
Kv = K - k» K = v - W C + k") = Kv,
(14.7)
so that (14.6) can be transformed to D V = -/8/v>
5"V = 0.
(14.8)
While both sides of (14.6) are conserved, neither side of (14.8) is divergenceless. Free-Graviton State. If no gravitational source currents are present, the free-field equation in the harmonic gauge, D^v = 0,
(14.9)
2
has traveling-wave solutions for k = 0, hflv(x) = e,v(k)e±ik
\
(14.10)
The monochromatic polarization tensor e^ = evll always satisfies the subsidiary conditions fc%v(k) = £„v(k)/cv - 0, e„" = 0 (14.11) in any gauge, but also we may choose (i, j = 1, 2, 3) e0„ = 0,
k%j(k) = 0,
e/ = 0
(14.12)
in the radiation gauge /c%v(k) = 0. Taking k = coe3, the physically significant circularly polarized waves correspond to e± oc e n + ie12 with £3„ = 0. That is, under a rotation through 6 about the z-axis (Problem 14.1) C = VRvV
e± = £ ± 2 % ,
(14.13)
which indicates that the spin-2 massless graviton travels at the speed of light with spin lined up along or against its momentum, corresponding to the two helicity states ^|1V -* e$ for X = ±2.
284 Lowest-Order Gravitational Interactions Classically such gravitational waves are hard to detect on Earth [but recently they have been inferred from the period of a binary pulsar (Taylor et al., 1979)]. Quantum-mechanically the covariantly normalized graviton wave function (14.10) carries energy E = ha> with the normalization e ^ V W = dxx.
(14.14)
Accordingly, the Feynman rules in momentum space for an incoming or outgoing graviton respectively are
e,vao
. £*v(fc) . ^ r •
R(jS sr
( 14 - 15 )
Virtual-Graviton Propagator. Following the discussion in Section 10.D on building covariant Feynman propagators, we expect the spin-2 graviton propagator to be of the form jj^Wl KSWSWm&A&MMSLQSISUm.*
(1416)
k2 + is
\
k
'
•
t
The projection operator is the sum of e^v e*fi over all ten possible virtual states (five spin 2, three spin 1, two spin 0). Recall from (Section 10.D) that the on-shell photon spin sum over two states, say in the transverse gauge £ 2 £; ef = dij — k{ kj, can be extended to the off-shell photon projection operator in the Feynman gauge £ 4 A^A* = — g^. In like fashion the two on-shell graviton states in the radiation gauge (14.12) satisfy the spin sum (Problem 14.1) 2
pol
- (sukjkm + s^kjk, - dijkfa —
(Ojmkiki + djikikm
(i4.i7)
— dtnkikj)
+ «;«,&,«„,],
so that (14.17) is symmetric in i *-+j and / *-* m and has zero contraction with any k index. On the other hand the off-shell spin sum in the numerator of the graviton propagator (14.16) is, in the harmonic gauge, =
&n*e
10
X V " 3 » = ibimivii + ntfiv* - *&•»?«*)•
(14.18)
pol
The advantage of the gauge choice (14.18) is that just as the off-shell photon field equation in the Lorentz (Feynman) gauge, HAfl = Jl™, can be inverted to Ap = { — g^Jk2)jlm, so in the graviton case the off-shell field equation in the harmonic (de Donder) gauge. J liv> can (14.18) to
, nv, = C8>
K*(k) = -'
k
2
1 ^ " ~k~2 J =
flV
(14.19)
Graviton Vertices 285
14.B Graviton Vertices While there is no simple linear substitution rule as for classical electromagnetism and QED, quantum gravity couplings may be obtained by appealing to the classical limit for spin-2 gravitons. Stress-Energy Tensor. The gravitational coupling strength is extremely weak (i.e., for G~1 « 1.5 x 1044 MeV2 in natural units h = c = 1) as measured by the dimensionless coupling Gtris ~ 10" 38 . This is conventionally expressed in terms of the rationalized coupling constant f=y/0nG,
f2m%/4nx
1.2 x 1(T 38 ,
(14.20)
which will play the role of charge for gravitational couplings. Another parallel exists between electromagnetic and gravitational classical theories, for just as the em vector potential interacts classically with all charged matter via the vector current j^ oc p^, fT = eu,
Jt^j^A^ej^A",
(14.21)
the linearized gravitational tensor potential interacts with all matter (charged or neutral, massive or massless) classically via the stress-energy tensor 7^v oc pMpv, ft=fT^,
jrI=j*tT=fTllytt"'.
(14.22)
The result (14.22) follows most elegantly from a lagrangian formulation of the problem (Utiyama 1956, Kibble 1961, Feynman 1962). We will simply apply this coupling and relate the scale of/to G, (14.20), when we calculate the quantum force in Section 14.D. What we must do here is induce the quantum-mechanical stress-energy tensors for off-diagonal matrix elements in momentum space for particles of various spins. It will be sufficient that the 7J,v(/c) = Tvfl(k) thus induced have the classical form in the forward direction p' -> p, k = p' — p -»0, where k is the graviton momentum, and are also conserved on and off the graviton mass shell for p', p on shell, k»Tjk) = T„v(fc)fc* = 0.
(14.23)
Spin-O-Graviton Coupling. The classical stress-energy tensor in coordinate space for noninteracting energetic matter is
TM » IPj0-<53(x - x„(t))-
(14-24)
Since the Fourier transform of c53(x) is unity and a factor of (2£)~ 1 can be absorbed into the covariant normalization of quantum states, we induce the off-diagonal quantum stress-energy tensor for spin-0 particles from (14.24) to be (Problem 14.2)
( 14 - 25 )
286 Lowest-Order Gravitational Interactions
which becomes equivalent to (14.24) for p' — p, p'2 = p2 = m2. Note that (14.25) is also conserved in the sense of (14.23), but only on the mass shell of the spinless particles p'2 = p2 — m2. The Feynman rule for such a spin-Ograviton vertex is therefore
fitiPv + PuPl + l^ri^),
kviMAAMAmX^
, (14.26)
where we have replaced m2 — p' • p by \k2 in (14.26) to make it obvious that such a term vanishes on the graviton mass shell. Spin-y-Graviton Coupling. The analogous off-diagonal stress-energy tensor for Dirac particles normalized to iiu -> 2m is
(14.27)
with up. y^ up -* 2ptl in the forward direction converting (14.27) to (14.25) or (14.24). Again (14.27) is conserved for on-mass-shell Dirac particles p2, p'2 -»m 2 , p, p" -* m. The absence of a direct r\^ coupling term in (14.27) as opposed to (14.25) is somewhat in the spirit of the electromagnetic spin-0 vs. spin 4 currents of Chapter 11. The corresponding Feynman rule for a spin-^-graviton vertex is then
kfW
+ pXy* + yM'+ P\]
k^ftM&mftaX^
• (14.28)
Photon-Graviton Coupling. Gravitons couple not only to massive particles, but also to massless particles having an effective relativistic mass E/c2. For em fields the classical stress energy tensor T£ = Fll'Fav+inllvF*i'Fal,
(14.29)
can be generalized off the forward direction for monochromatic plane waves of (graviton) momentum q = k' — k to (Problem 14.2) T%(q) = e^WT^q,
k', kp(k),
(14.30a)
<*'. PI T„v(q)\k, <x> = i[*i(*„ty» + fcvij*.) + kp(k'^m + k'yr\a)l) - li/iiKK + k^K) + >/„„(&' • ktj^ - kpk'a) (14.30b) - V • Htl^tly,, + tltftly,)]. This stress-energy tensor is conserved for on-shell photons (k'2 = k2 = 0) and also conserved (gauge invariant) when regarded as a photon current
Graviton Spontaneous Emission 287 (Problem 14.2) tfT^q) = 0-
WT^
= T^X
= 0.
(14.31)
The corresponding Feynman rule for this vertex is
<
k'.(3
(14.32)
k, a
Graviton-Graviton Coupling. Finally, since a graviton couples to all energetic particles, it must couple to other gravitons as well, as shown in Figure 14.1. However, it turns out that there are 18 independent terms at this graviton-graviton vertex and so we will not bother to display (g 17^v | g} (see, e.g., Feynman 1962, Duff 1975a). Nonetheless this vertex is our first clue that the theory of gravity is inherently nonlinear in the gravitational field. What complicates the theory even further is that (g\Tllv\g}= T*rvav(/j2) is not gauge invariant on its own; the »/,,„ in the matter tensor (14.25) must be replaced by g^ to 0(h2) and then the sum of the matter and gravity stressenergy tensors is gauge invariant to 0(h2) (Feynman 1962). Such a connection between gauge invariance and nonlinearity is a characteristic property of the theory as a whole and leads to interesting physical consequences such as the length contraction in curved space (Thirring 1961). 9
Figure 14.1 Graviton-graviton vertex.
14.C Graviton Spontaneous Emission Armed with this formalism, we proceed to calculate the simplest quantum effect, the spontaneous emission of a graviton from an atom as shown in Figure 14.2. Differential Decay Rate. Following the calculation of the photon spontaneous emission in Section 9.C, we compute the first-order graviton emission
288 Lowest-Order Gravitational Interactions
a; Figure 14.2 Spontaneous emission of a graviton from a bound atomic electron.
amplitude according to (Eba = Eb — Ea) Sha=-if°0
Vba(t)dt=-ifo*v(kKb\T»*(k)\a>d(Eba-co),
(14.33)
where is the off-diagonal, bound-state stress-energy tensor. In the usual way, with only the free graviton having a density of final states, the unpolarized differential decay rate is drba = ^dNf=f2
£ \elTZ\28(Eba-co)^.
*
(14.34) Z ( a
pol
Now d3k = oi2doidnk, so (14.34) becomes J 2-r l £ » i v T b o |
2
dilk
|e vita|
8* £ 1 "
—
TbZTba
2 J efivEap-
(14.35)
n
Angular Integral over Spin Sum. Since the polarization sum in (14.35) includes only the two helicity states for on-shell gravitons, we may apply the gauge conditions (14.12) and the sum (14.17) which involves zero, two, and four factors of the graviton three-momentum. The solid-angle integral for the first two cases has been used for the spin-1 photon case in Section 9.C, while the latter integral is (Problem 14.3)
J dtoMiWn, = ~ ( M * + StiSjm + Sta^i)-
(14-36)
The solid-angle integral over the spin sum (14.17) is then . J
2
g
dak X M^OO = T ftW* + 5>*si') ~ &M pol
<1437)
J
Note that the index structure of (14.37) is that of a standard rotation group or isospin irreducible tensor of rank 2, as discussed in Section 2.E and Section 12.C, Table 12.1, respectively. The total decay rate found from (14.35) and (14.37) is rba = *^-Ttf2)(k)T\?(k),
(14.38)
where the irreducible rank-2 part of the stress-energy tensor Tff has a Fourier transform (analogous to the acceleration in the photon
Quantum Corrections to the Newtonian Force Law 289 spontaneous-emission case a,•.-* — w2ri),
(14.39) 2
-> -\me(o
3
2
J d xM(x%XiX} - ir (5;j]iAa(x)
for £2 transitions. The Wigner-Eckart theorem then leads to the atomic selection rule \Jb-Ja\ <2<Jb + Ja. Substituting (14.39) into (14.38) then gives, for example, the 3d (m = 2) to Is transition in hydrogen [see, e.g., Weinberg (1972)] T(3d - Is) =
223Gm3a6 "_;« « 2.5 x KT 4 4 sec"».
(14.40)
Since photon spontaneous emission rates of order ~ 109 sec" x completely swamp (14.40), it is hard to imagine this "leading order" quantum gravity effect ever being detected in the laboratory.
14.D Quantum Corrections to the Newtonian Force Law Recall from Section ll.G that the single-photon exchange between two charged particles tells us about the underlying nature and quantum corrections to the static Coulomb force. The same is true for the single-graviton exchange between two massive particles as shown in Figure 14.3 in connection with the static newtonian force F = —Gml m2/r2. Potential for Spinless Matter. Assuming the massive external particles in Figure 14.3 have zero spin, the Feynman rules for graviton exchange gives
S}T = {-ifipTl
-£—K
T2p8*(Pfi)
(14.41)
"T" ( o
T~*l T^/XV
= ~~{Ls
- (m2 + ml)? + [u - (m2 + m 2 )] 2 - [t2 + 4m 2 m 2 ]}. (14.42)
In converting (14.41) to (14.42) we have used the usual Mandelstam invariants s = (Pi + p2)2, t = (p\ — prf, M = (Pi — p'i)2, with the spinless matrix elements of the stress-energy tensor (14.25) and the graviton propagator numerator (14.18) as given in the de Donder gauge (see Problem 14.4). In the nonrelativistic limit, t -» — k 2 ,s-> (m, + m2)2, M-> (m2 — m2)2 + k2,
290 Lowest-Order Gravitational Interactions
Figure 14.3 Lowest-order graviton-exchange force.
the Fourier transform of (14.42) leads to the static potential (recall (11.68)) V(t)= -~±—
| > k r i k ' r } 7 ( N R ) = -l^lUll
+if 2 <5 3 (r). (14.43)
We of course recognize the first term in (14.43) as the attractive newtonian potential, justifying our initial choice of/ 2 = 87rG. The second term in (14.43) is analogous to the quantum correction to the Coulomb or nuclear force. It is repulsive, independent of particle masses, and can only be measured for bound (quantum) s-states. (It would appear not to be present for a spin-0 graviton force.) As might be expected, the graviton-exchange potential between particles with spin modifies (14.43) with spin-orbit and spin-spin terms as well. Moreover, the velocity-dependent corrections to (14.43) generate the general-relativistic, "post-newtonian" modifications of the classical equation of motion of a particle in a gravitational field (Corinaldesi, 1956). Antimatter Forces. It is occasionally conjectured that by analogy with em forces, the sign of the matter-antimatter gravitational force ought to be opposite that of the matter-matter force (i.e., repulsive). This analogy is incorrect because it is the (even) spin of the exchange particle and not its mass (both photon and graviton are massless) which determines the sign of the force. We have learned in Chapters 6 and 10 that CPT symmetry requires incoming particles of momentum p to be replaced by outgoing antiparticles of momentum — p (rather than changing the sign of the coupling e in em Feynman graphs). Similarly, the matter stress-energy tensor (p' | ^iv IP> m u s t be replaced by the antimatter stress-energy tensor <~ P | ^ I V | ~ P'> m t n e Feynman rules. Then the quadratic nature of T„v, depending in the spinless matter case on the product of momentum factors such as p'p pv leads to the general relation
(14.44)
the stress-energy tensor always being symmetric in the indices p and v. Since the antimatter and matter stress-energy tensors are always of the same sign, the corresponding static potentials due to (even-spin) graviton exchange are
Gravitational Light Bending 291
always attractive; both matter-matter and matter-antimatter gravitational forces are attractive, assuming pure spin two gravitons. Since we are living in what we believe to be a matter-dominated universe, it is very hard to observe this weak matter-antimatter gravitational force in normal experiments. Yet there is an observation (Good 1961) which supports this attraction hypothesis. The measured KL — Ks mass difference of AmK ~ 3 x 1CT6 eV, presumably a second-order weak-interaction effect with mK ~ 500 MeV, gives AmK/mK ~ 10" 6 /10 8 ~ 10" 14 ~ 0(Jti),
(14.45)
which sets a (CPT-violating) upper bound on the K° — K° mass difference of the same size. / / we incorrectly assume that the K° is gravitationally attracted to the Earth while the K° is repelled, then the gravitational potential energy difference would be AEK = mK —
(-mK) —— = 2mK(j>E,
(14.46)
where
14.E Gravitational Light Bending Now consider the gravitational force between a massive, spinless object such as the Sun and a massless photon. The corresponding Feynman diagram is shown in Figure 14.4. While gravitational light bending was originally calculated via the nonlinear classical theory of general relativity (Einstein 1916), it is of interest to see how the quantum theory of gravity accounts for this effect.
k Deflected photon
Figure 14.4 Lowest order gravitational light bending due to the Sun.
292 Lowest-Order Gravitational Interactions
Differential Cross Section. As in any other quantum scattering process, we first apply the Feynman rules to calculate a scattering cross section. We may use the obvious approximation of very small grazing angles, corresponding to a graviton invariant momentum transfer of q2 « 0 in the numerator of the Feynman diagram of Figure 14.4, i.e., we set cos 6 % 1 in the numerator but not the denominator of the Feynman amplitude. Accordingly, the Feynman rules give
•KM** + MXv - ^KK - to^kPKW*frY(W*(Pfi) (14.47) with a factor of 2 in (14.47) due to two possible momentum configurations of the photons in Figure 14.4 for the coupling defined by (14.30b). Then in the static limit p'll = pp -+ MQfif^o w^ith q2 -> — 4a>2 sin2 j6, a>' -* co, (14.47) leads to (Problem 14.5) 1
f2M% E*(k') • c(k). sin2 | 0
fi
(14.48)
It is now easy to compute the spin averaged square of Ty?: f2M\
I1 T*COV 12
I fi I
sin
E|2
2 U
=
pol
f2M sin2 +0
(14.49)
The general differential-cross-section formula (10.18) for J^ = #"' = 4a>MQ then becomes for small angles (sin \Q « j6), da
1
dQ
16G 2 Mi
fi
(14.50)
%nMr.
Classical Deflection. To convert (14.50) into a measurable form, recall the classical connection between cross section and impact parameter, a — nb2, or in differential form bdb = — (do/dQ) sin 6 d6. Integrating this relation via (14.50) for small grazing angles we find rR°
1
bdb
=~l
te° da
ede
dn =-
c1 i6G2M '° d9 fl3 '
°l
(14.51)
where the minimum value of the impact parameter b is the radius of the Sun, RQ, corresponding to the maximum deflection of the photon orbit, 9Q. Solving for 0O from (14.51), we obtain the standard Einstein formula (Delbourgo and Phocas-Cosmetatos 1972) 0Q = 4GMe/R0
= 0.849 x 10" 5 rad = 1.75"
(14.52)
for GMQ= 1.475 km and RQ = 6.95 x 105 km. Needless to say, (14.52) is in agreement with experiment, with present errors at the 1 % level. But in any
Quantum Gravity and General Relativity 293
case we may regard the classical light-bending result (14.52) as a real laboratory test of a spin-2 quantum graviton exchange (a spin-0 graviton gives no such light bending—see Problem 14.5). Moreover it is also a test of the masslessness of the graviton because a massive graviton does not reproduce (14.52) no matter how small mg is (see Van Dam and Veltman 1970 and Problem 14.5). Quantum corrections to this classical deflection angle arise from a higherorder loop modification of the graviton-photon vertex in Figure 14.4. This will be briefly discussed in Section 15.B (see Figure 15.8).
14.F Connection between Quantum Theory of Gravity and General Relativity Having reproduced one of the major predictions of general relativity, it is time to point out how the linear quantum theory just developed is related to Einstein's nonlinear classical theory of general relativity. It will turn out that the former uniquely implies the latter provided that the graviton has pure spin 2, with no spin-0 contamination (Gupta, 1952). Soft-Graviton Theorem and the Equivalence Principle. Consider a graviton with soft momentum k interacting with a matter current with amplitude etlvM'"'. To leading order, the only singular parts of the amplitude 0(k~1) are generated by the graviton attaching itself to an external massive line, with pf = mf. Summing over all such massive lines as shown in Figure 14.5 and using the soft limit of all stress-energy couplings 7^v -> Ip^ pv as in (14.25) and (14.27), the leading-order term of the graviton M-function is
M =
- ? /; V^ M '
(14 53)
-
where M is the amplitude in the k = 0 limit without the graviton present. The structure of (14.53) is strikingly similar to the leading-order soft-photon amplitude of Section ll.E, K = I «i ^7~k
M,
(14-54)
where we may substitute a soft photon for a soft graviton in Figure 14.5.
Figure 14.5 Dominant soft-graviton graphs.
294 Lowest-Order Gravitational Interactions
Both of these amplitudes must be gauge invariant, (see Problem 14.6) not only to all orders in k, but to leading order, 0(k~1), k»Ml = 0,
X et = °
(14.55a)
i
kfM»v = 0,
£ / i p i = 0.
(14.55b)
The former relation corresponds to charge conservation; the latter statement, in conjunction with overall four-momentum conservation for this k = 0 process, I P J , = 0, (14.56) i
requires that / ; = / = constant,
(14.57)
universally for all particles, regardless of their spin type or mass. This is the content of the equivalence principle, implying the equality of gravitational and inertial mass (Weinberg 1964c—see also Problem 14.6). A similar analysis for soft massless particles with spin s > 2 implies that such particles cannot couple consistently in the soft limit because, for example, the s = 3 analog of (14.55), £ g{ p\ p\ = 0, cannot be satisfied along with (14.56) unless its coupling gt vanishes for all particles. The conclusion is that only spin s = 0, 1,2 massless particles couple consistently at zero energy and momentum. Alternatively this means that there can be no static component of the gravitational force for s > 2. This soft-graviton theorem does not preclude the graviton from coupling to other massless particles, including photons. Light bending is an example of this coupling, and the agreement of (14.52) with experiment certainly suggests that the graviton-photon coupling does indeed have the universal strength f2 = SnG, as anticipated in (14.32). Iteration of Linear Theory. Lastly, the soft-graviton theorem requires the graviton also to couple universally to other gravitons as depicted in Figure 14.1. This trilinear coupling f(g \ T^ | gyW is measured in part by the discrepancies between the observed and classical newtonian precessions of the planetary elliptical orbits. More specifically the first-order graviton exchange between a planet and the Sun analogous to Figure 14.3 in the static, large M 0 limit gives a contribution to the perihelion advance 0(M0). The remaining part of the classical Einstein prediction is 0(M%) and is generated by the linear graviton coupling acting twice via a VG0 V interaction, combined with the potential acting twice via the trilinear graviton vertex (and also a contact seagull term involving the graviton acting twice). Clearly the existence of the nonlinear graviton-graviton vertex makes the connection between the quantum and classical perihelion calculations much more complicated than was the case for light bending (see Duff 1974 for details). Such a trilinear graviton coupling points toward a nonlinear nature of the underlying dynamical field equation because there is now no way of com-
Quantum Gravity and General Relativity
295
pletely decoupling the stress-energy "source current" from the graviton field with which it interacts. To see this, write T„v = T?:""(h°) + T™(h°) + T*;r(h2)
(14.58)
and substitute (14.58) back into the linearized gauge-independent field equation of Section 14.A, • V -(dlld%v
+ dvd%a)=
-/T„ ¥ = -fT^h0)
-fT„v(h2). (14.59)
2
Transposing T^h ) to the left-hand side of (14.59), it turns out that this trilinear vertex of Figure 14.1, or equivalently {glT^lg} (with its 18 independent terms), is just right to allow us to express the field equation (14.59) as (Feynman 1962) R,v(h2)=-f2T„v(h°).
(14.60)
Here R^h2) is the Ricci tensor to 0(h2), inferred from (14.59) to be in the harmonic gauge
R„v(h2) = -ftf"
+ Wtd^Kf
- dfd^
- 5„a a v + d,da*U- (14-6i)
Canceling a factor of/from both sides of (14.60), it is easy to check that it is consistent with (14.59) to leading order in h given (14.61). It is then natural to define the coordinate-dependent metric as an extension of (14.1),
(14.62)
but then the inverse g^(x) must be expressed in a power series in h in order that 0"v0vp = (5%. It turns out, however, that the general form of the Ricci tensor can be taken over directly from (14.61), but only in a "locally inertial" coordinate system C, where R^ is defined as K»«v = W'lS^g^
- dpd.g^ - d^g^
+ d0dag^].
(14.63)
The field equation (14.60) then iterates to Einstein's field equation for general relativity, (where now 7^v refers only to matter and em radiation), *„v = K„» - l9,vRxX = -ZnGT^,
(14.64)
and this is the major result (Gupta 1952). It is worthwhile to point out that proceeding from (14.61) to (14.64) is a nontrivial step related to a quantum version of the principle of general covariance. This principle is very powerful and can be used to convert all "flat space" dynamical equations involving n^ and 3„ or <3V to curved space dynamical equations involving the metric tensor g^x) and covariant derivatives. Even at the classical level a satisfactory explanation of this step links up g^ with notions of differential geometry such as affine connection, and the interested reader should consult standard texts on general relativity at this point (see, e.g., Weinberg 1972). At the quantum level, we may regard the "minimal" version of the principle of general covariance—i.e., replacing rj^
296 Lowest-Order Gravitational Interactions
by g^ and d^ by covariant derivatives—as the gravitational analog of the minimal coupling (gauge) principle of QED in that transforming (14.64) back to the linear form (14.59) automatically gives the linearized (flat space) quantum coupling of /7J,„ h"v. Without assuming the existence of vector or tensor fields, a more systematic approach to the classical limit has been worked out by Weinberg for QED (1965) and by Boulware and Deser (1975) for the quantum theory of gravity. The conclusion is that as long as helicity + 1 and + 2 massless photons and gravitons couple to conserved currents both on and off the massless particle mass shells, then all tree (no loop) graphs are fixed in the low energy limit. Such tree graphs in turn lead uniquely to the classical Maxwell and Einstein theories, respectively. Word of Caution. Combining the notions of geometry and general relativity with quantum mechanics and the Lorentz group has more drawbacks than is suggested above, however. In particular, higher-order loop diagrams involving spin-2 gravitons have higher and higher degrees of divergence, (as do higher-order weak interaction current-current theories) and they are not generally covariant in certain gauges. But as a general rule the tree diagram, quantum weak current-current and graviton theories which we have discussed are valid only at low energies and in this sense they should be regarded as successful phenomenologies, not complete theories. For general references on the quantum theory of gravity see Thirring (1961), Feynman (1962), Weinberg (1964c, 1965,1972), Kibble (1965), Isham et al. (1975), Boulware and Deser (1975).
CHAPTER 15
Higher-Order Covariant Feynman Diagrams
In this final chapter we investigate the consequences of higher-order loop diagrams, first justifying their necessity and describing a pragmatic "regularization" procedure for circumventing unphysical infinities inherent in many loop integrals. Next we work out in detail the finite second-order anomalous-magnetic-moment radiative corrections for the electron and point out where other finite triangle-type loop diagrams lead to interesting results. Then we compute all of the second-order QED self-energy and vertex-modification loop integrals, showing how the resulting infinities cancel in the final form for the 0(e2) dressed-electron form factors and applying the latter to calculate the Lamb shift in atomic hydrogen. To give the reader a feeling for the rigor required to handle properly loop infinities, we briefly survey the renormalization program in field theory and also develop an alternate dispersion-theoretic interpretation of Feynman diagrams in general and QED loop graphs in particular. Finally we use dispersion theory to investigate the strong-interaction S-matrix, presumably corresponding to summing over large classes of Feynman diagrams. Dispersion relations for proton Compton and pion-nucleon scattering reveal the method, with the additional formalisms of current algebra and Regge poles describing the low- and high-energy behavior of dispersion integrals.
15.A Closed-Loop Diagrams One Feynman rule which we have not yet investigated in detail is the integration over all internal momentum lines / by | d4l. Thus far we have dealt with diagrams typically having two vertices and one propagator, leading to 297
298 Higher-Order Covariant Feynman Diagrams
one overall energy-momentum-conserving delta function and an internal virtual four-momentum which is uniquely determined by external momenta. As noted at the end of Chapter 14, such tree diagrams can have a classical limit (as the em Thomson amplitude at low energies). If, however, there are at least as many propagators as vertices in a Feynman graph, corresponding to a closed loop, there will be additional powers of h in the numerator which force these diagrams to vanish in the classical limit. The Feynman internal momentum rule for loop topologies (but with h = 1) corresponds to
s}7 oc J > P l . . . J a*?. s*(Pf - P „)... a 4 ( Pl - Pi) x d*(pfi) J a*z..., so that the internal momentum / is not constrained by the external momenta, but can vary over all possible values including / = oo and / = 0. Depending upon the number of momentum factors in each loop vertex and the geometry of the closed loop, the closed-loop integral can diverge at / = oo (referred to as an ultraviolet divergence). If one of the internal loop particles has zero mass (photons, neutrinos, gravitons), then the loop integral can also diverge at / = 0 (called an infrared divergence), but only when one of the other particles at the loop vertex is external, on its mass shell. In Chapter 10 we stressed the compactness of covariant Feynman graphs in combining forward and backward propagators. But now with potentially infinite loop integrals, we may ask if the covariant Feynman-diagram approach, although compact, is meaningful in higher orders. The answer is yes—even more so—but to see why, it will be necessary to unravel the structure of closed-loop graphs in greater detail. Loop Reduction. Before tackling the divergence problem, let us see how closed-loop "box" graphs, such as those in Figure 15.1(a), (b), reduce to a familiar form, Figure 15.1(c), in the static (not classical) limit. The external spinless particles of momenta p, p' have mass m, while lines q, q' have mass (x; the internal wavy lines represent photons, lines / have mass n, and lines p + q — I and p' + I — q have mass m. Diagram (b) is the crossed-photon version of diagram (a), equivalent to uncrossed photon lines but crossed
(a)
(b)
Figure 15.1 Loop reduction to old-fashioned perturbation theory.
(c)
Closed-Loop Diagrams 299
external lines of mass m (or p). All of the propagators in Figure 15.1(a), (b) have forward and backward propagating parts, and this leads formally to a total of 2 4 + 2 4 = 32 distinct nonrelativistic graphs. So compactness is surely a strong point in favor of Feynman covariant loop diagrams, in this case giving, according to the Feynman rules,
s^ +
s^^^p^yj^(/2- i2 /
an
+ j £ ) ( ( ^ - / ) 2 + i £ )((/-g) 2 + i£)
(q' + l).(2p' + q'-l)(q + l)-(2p + (p + q-l)2 -m2 + ie
+
q-l)
(q' + I) • (2p -q' + l)(q + I) • (2p' {p' -q + I)2 -m2 + ie
(15.1)
q + l)
Leaving aside the divergence problems of (15.1) (the ultraviolet divergence is only like a logarithm—which, as we shall see later, will not affect our argument), consider mass m to be very heavy with m P p.. In this static limit Vn,v'n^> m9fio dominate over q and / in the numerator terms, leading to Safi + S1fi
84(Pfi)4m2e4 J
^-^
+
*4l (q' + Uq + Oo ie)((q'-iy + ie)((l-qf
2p • (q — I) + ie + — 2p'
= -,3fe 0 - q0)2me j (/2 _ ^
(15.2a)
1 • (q — I) + ie +
+ ie)
. ^ _ /)2 + .g)((/ _ q)2 + .g), (15.2b)
where we have used (x ± ie)'i = P/x + in8(x) and d4(Pfl) -> 8(q'0 — q0) to convert (15.2a) to (15.2b). We then recognize the latter as just the oldfashioned perturbation-theory result Sfi = — i8(Efi)(VG0 V)fi for the static Coulomb potential V(l - q) = e2(l — q)" 2 and time-independent free Green's function in the static limit 3(1'-1) l -p? + ie 2
30'-1) 2
2
{l 0-n )-\
2
+
1_
+
ie)-1\\}. (15.3)
The factors of 2m in (15.2b) and 2p in (15.3) are to be absorbed into the redefinition of the noncovariant normalization of states. So we see that in the static limit the virtual-loop momentum integral j rf*l and Feynman's covariant is prescription are just what is needed to "sever the connection" between p and p' in Figure 15.1, allowing the static potential to act twice, but independently in the sense of Figure 15.1(c). Therefore, the loop integral is a necessary Feynman rule and its consequences must be taken seriously in other contexts as well.
300 Higher-Order Covariant Feynman Diagrams
q>
(a)
(b)
Figure 15.2 Relative-sign identification of loop diagrams. Relative Signs of Loop Diagrams. We now observe that the relative signs of higher-order loop graphs containing at least a pair of bosons can be obtained by imagining one of the boson (even photon) masses to be very large, i.e., ignoring this boson propagator altogether. For the case of the box diagrams of Figure 15.2, this limit results in the "skeleton pole graphs" of Bhabha scattering discussed in Section ll.C. Since we know from Figures 11.5 and 11.6 that the skeleton pole graphs add (subtract) for external boson (fermion) lines, we may infer the same relative sign for the box diagrams of Figure 15.2 in the two cases: Ta — Tb for external fermions. The same conclusion can be drawn from the Feynman sign rule of Section 10.E: for fermions first rotate the box of Figure 15.2(b) through 90° (say counterclockwise) so that the photon propagators are horizontal as in Figure 15.2(a). Then push lines p, q', q clockwise and line p' counterwise past p, q', q to adjacent vertices, picking up a factor of (— )3 = — 1 in the latter case in order to make Figure 15.2(b) look like Figure 15.2(a). So again we recover the relative fermion amplitudes Ta — Tb. Directional Independence of Loop Momenta. Next we observe that a virtual momentum completely integrated over is a "dummy variable" whose directional flow can be reversed without altering the physical content of the resulting S-matrix element. The degree of divergence of some loop graphs can in fact be minimized by a convenient choice of momentum labeling, but more about that later. Here we use this fact to investigate the signs of pure fermion loops; such graphs could not be treated as in Figure 15.2. In particular, consider the two- and three-fermion loops of Figure 15.3 and 15.4. For the two-point vacuum polarization loop of Figure 15.3, charge conjugation converts fermion to antifermion and amplitudes Ta<-> Tb. Since each photon vertex has a yM or yv matrix, the transformation CyJC~1 — — y^ leads to two sign changes, so that Ta = Tb = T or T = \Ta + \Tb as shown in
I "-"—2
(a)
(b)
Figure 15.3 Internal momentum symmetrization of bubble diagram.
Closed-Loop Diagrams 301
Figure 15.3, consistent with the directional independence of loop momenta. The latter is also true for the three-point vertex of Figure 15.4, but in this case there are three sign changes due to charge conjugation at each vertex, giving Ta= — Tb, or 7 = j(Ta + Tb) = 0. Clearly then all closed fermion loops containing an odd number of external photons (or other C = — 1 particles) vanish (Furry's theorem). Note that Figure 15.4 does not include the exchange graphs obtained by crossing the two (final) photons. To account for this possibility some authors drop the factors of \ in Figure 15.4, effectively letting graph (b) act as the crossed version of graph (a). We prefer not to do this, in order to uphold the Feynman rule requiring symmetrization of external (final) lines and ignoring statistics for intermediate lines. As to the relative sign of closed fermion loops vs. closed boson loops (containing an even number of external photons, etc.), it is negative. This is due, for example, to Figure 15.3(a) not having the usual minus sign associated with an incoming antifermion v2 in ulyfivlv2yvu2, because no fermion lines are crossed when relabeling the bubble as in Figure 15.3(b). This accounts for the Feynman rule of — 1 for each closed fermion loop. Pragmatic Reguiarization. This is a catch phrase for making sense out of a divergent Feynman loop integral by rendering it finite in some "physical" way (by hook or by crook). Later, in Sections 15.F and G, we shall attempt to infuse some rigor into the theory, but by way of introduction to the subject it will prove more convenient (but perhaps somewhat less convincing) to adopt the following "pragmatic" approach to unwanted loop infinities. i. Covariant separation: Pick out the finite parts of an infinite loop integral by separating it into covariant terms which depend upon external momenta and spins but transform differently under JS?. ii. Symmetrization of loop momenta: Because of the directional independence of loop momenta, a "symmetric" choice of loop momenta usually gives the configuration with the lowest degree of ultraviolet divergence. Sometimes, however, symmetrization is not possible. iii. Introduction of cutoff masses: Parametrize ultraviolet divergences by cutting off the integral at /max = M or alternatively by subtracting off
A A A (a)
(b)
Figure 15.4 Internal momentum symmetrization of triangle diagram and Furry's theorem.
302 Higher-Order Covariant Feynman Diagrams from a loop propagator (I2 — m2)~1 another propagator (I2 — M2)~i for a fictitious heavy particle M. Also, an infrared divergence can be parametrized by giving the massless virtual particle a mass k in the propagator, l~2 -* (I2 — k2)~'. Of course, at the end of the calculation we must take M -* oo and k->0, but at that point the physical amplitudes thus obtained should not diverge in either limit, iv. Forward-direction subtraction: Subtract off from a closed-loop integral A(p', p) the "forward-direction" part A(p, p) which contributes to a form factor that is a priori normalized, e.g. to F(0) = 1. In QED this corresponds to the absence of radiation from nonaccelerating charges.
15.B Electron Anomalous Magnetic Moment As stressed in Chapter 5, one of the cornerstones of the Dirac theory is that the predicted anomalous magnetic moment of the electron, n = (e/2me)a, or equivalently the Lande gf-factor gDirac = 2, is quite close to the experimental value, now measured to be (van Dyck et al. 1977, 1979) ge = 2(1 + Ke) = 2[1.001159652200(40)].
(15.4)
Can higher-order perturbation theory (i.e., Feynman diagrams) account for this last bit of discrepancy between theory and experiment? The answer is a resounding yes, to date through the first eleven significant figures in (15.4). Such fantastic agreement marks the pinnacle of man's ability to describe nature in a precise manner. To make this point in a finite number of pages, however, we will compute only the lowest 0(e2) correction to g = 2, first obtained by Schwinger (1948). Covariant Separation. The 0(e2) radiative correction A„(p', p) to the minimally coupled Dirac vertex y^ is depicted in Figure 15.5. We may compute this vertex modification (sandwiched between on-mass-shell electron spinors) according to the Feynman rules in Feynman gauge as ctk w +m)lM +m)f - «v, - - . * > , P)=(- ,r.-3( - iyf [(p' - kf ~- *m2][(p - kf~ -* m2]k2 '
(15.5a) where we have deleted the i'e factors in the propagator denominators D = (k2 - 2p • k)(k2 — 2p • k)k2 for the time being. Then commuting the
XM
A^p'.p) 2
Figure 15.5 Vertex modification to 0(e ) of electron em current.
Electron Anomalous Magnetic Moment 303 factors / , p1 to the left and right, respectively, and applying the free-particle Dirac equation, (15.5a) can be written as (Problem 15.1) \{p', p) = -
^
/ ^
[2(*2 - 4fc • P + 2p' • p% - 4mk, + 8#P„ - 4^/cJ,
(15.5b) with 2P = p' + p and D(p', p) symmetric in p' <-» p. We observe that as k -> oo or k - 0, (15.5) diverges logarithmically,
K-*y*\
d*k
rdk
(15.6)
with only the k2y^ and l/tk^ oc k2y^ terms in the numerator of (15.5b) leading to the ultraviolet divergence (D -* k6), while the p' • py,, term leads to the infrared divergence (D ->fc4).This means that in either case the unwanted singularity resides in the form-factor coefficient of yll for the on-shell em current (recall Problem 5.8) jT(p\p)=eup.
F1(t>f, + F2(t)
WavCf
2m
(15.7a)
up=eupT^(p',p)up (15.7b) m [the difference between (15.7a) and (15.7b) being the Gordon reduction (5.82), giving GM = F t -I- F2], where q = p' — p, q2 = t. Our strategy will be to circumvent the divergences in (15.7) by using the pragmatic regularization technique of covariant separation described in Section 15.A to split off the infinities in GM y^ from the wanted finite contribution of F2 PM in T^(p', p) = y» + Kip', P) + • • • of Figure 15.5. To this end we use covariance arguments and the denominator symmetry of D(p', p) = D(p, p') in (15.5b) to write for q = p' - p, IP = p' + p, = eur Gu{t)v> - F2(t)
d*k
d*k
d4k j - p - KK = a2(t)P„Pv + b2(t)q^qv + c2(t)g^.
(15.8a) (15.8b)
That is, the integration over k converts internal invariants to the only external invariant in the problem: t = q2, with P2 = m2 — %t also a function of t and 2q • P = p'2 — p2 = 0 vanishing on mass shell. Likewise the integrals over k^ and k^kv must be described by external covariants P^, P^PV, q^qv, g^v symmetric in p' *-*p and /*•<-• v. This rules out the covariants q^ and
te
4m[ai(t) - a2{t)\
(15.9a)
304 Higher-Order Covariant Feynman Diagrams Given (15.9a), we can then set t = 0 to express the electron anomalous magnetic moment to 0(e2) with e2 = 4na as 0-2 . . ice 22 = ^— — = F2(0) = -3 m - [ai(0) - a2(0)].
K
Z
(15.9b)
71
Feynman Integrals. To find the coefficients aj(0) and a2(0) in (15.9b), we use Feynman's tricks already introduced in Section ll.E: 1
l r
dx
Vb = l [ax + b(l-x)]2> =
i
I0 [ax + b{\- x)f '
(151
°a)
(151 b)
°
the latter obtained from the former by differentiation with respect to the parameter a. Then setting p' = p = P in the integrals (15.8), so that t = 0, we apply (15.10b) for a = k2 — 2p • k and b = k2 and complete the square of the resulting denominators. For a0(0) this procedure yields ao(0)=
=
J (k2 - 2p •fc)2/c2= J0 i
,
2
2X dX
^ f d*k' 2 2 [k' - m22Vx213 ]
L WrF^
J [k2 - 2p • fcxf (15.11a)
'
where k' = k — px and we have shifted the (infinite) region of integration from the variable k to k'. Likewise the integrals over k^ and /c^/cv can be obtained immediately from (15.11a) as lc\\ — [ P — f-> A i \ P*'* " " J (k2 - 2p • kfk2 ~ J0 2X dX J [k'2 - m2x2f ' (15.11b)
ai(0,P
a2(0)p,Pv + c2(0)g„v = j
_
_f S
» '
2
,* fd4fc,(fc, + Px),(fc, + px)v J
~ J0
(1511c)
[fc'2 - m 2 x 2 ] 3
Note that the trick of setting p' = p or q = 0 to obtain (15.11) cannot be applied before the covariant separation (15.9); initially taking q = 0 in (15.5) would wipe out the entire F 2 contribution in (15.7a) or blend it together with the unwanted GM contribution in (15.7b). The next step is to evaluate the four-dimensional integral common to the three forms (15.11), i.e., restoring the ie prescription in the denominators, we have in general (see Problem 15.1),
f
:d'k
J [k - C + ief 2
* 2C
(15.12)
Electron Anomalous Magnetic Moment 305
where C = m2x2 for (15.11). Since the integral (15.12) is independent of all dot products such as k • p, its first moment overfc„can be rotated via k0 -»ik 4 to an euclidian metric — k2 -> k\ + k2 and hence vanishes by symmetry. Likewise the second moment integral of (15.12) over /c„/cv can only depend upon g^ with C independent of any external momenta such as p. That is, using dAk
f
- ~'"2
'
dAk
f
J [k'2 - m2x2 + ief ~ 2m2x2 '
'
J [k'2 - m2x2 + is]3 " ~ (15.13a)
4
C
d fc'
_t
J [k'2 - m2x2 + i £ ] 3 " v
4
f
2
d / c ' fc'
4
^"v J [fc'2 - m2x2 + is]3 '
(
}
(the latter following from contraction by g"v with g^g^ = 4), we may obtain from (15.11) the desired coefficients in2 r1
0
« 1 ( ) = -2m £ l |J0 ^ 2
in2
2x • x x^ 2
a )
^ ^-2rr?\0dX-^
= -m ^ 22 . =
-2m2-
(15-14a) (1514b)
Then our final result (15.9b) becomes (Schwinger 1948) K=
- % 2 ( _ ! ^ + ^ 1 ) = ^ =0.0011614098 7i3
\
m2
2 m 2 / 2n
(15.15) V
;
for the most recent experimental value of the fine-structure constant oTJ = 137.035987(29). Comparison with Experiment. This value for K is independent of the chargedparticle mass and hence applies to both structureless leptons, the electron and the muon. In a similar, but much more difficult manner, the fourthorder corrections for the electron were first calculated by Karplus and Kroll (1950) and corrected by Sommerfield (1957). Recently the sixth-order electron corrections (involving 72 Feynman diagrams) have been calculated numerically by Levine and Wright (1973) and Cvitanovic and Kinoshita (1974) and partially checked analytically by Levine and Roskies (1976). The sum total for Ke in QED turns out to be ,QED = A _ 0.328478445 2K
(;)' + »«<»>(;)'
(15.16)
0.001159652359(282), which matches experiment, (15.4), through the first 8 significant figures in tce and 11 significant figures in ge. Likewise the muon anomalous magnetic
306 Higher-Order Covariant Feynman Diagrams
moment calculated in QED agrees quite well with experiment (Bailey et al. 1977): K? ED = 0.001165926(29),
^ x p = 0.001165922(9)
(15.17)
with estimated hadronic corrections affecting the sixth-order terms. We may conclude from these spectacular successes that the electron and muon are indeed elementary (structureless) particles and that the correct QED coupling is minimal, ey^. While it is also true that Feynman diagrams are therefore an accurate calculational tool, we may not infer that they are the only existing tool, because dispersion-theory methods (see Section 15.G) and Schwinger's mass-operator technique (see Sommerfield 1957, Schwinger 1969) also can produce these results, at least in principle. Noncovariant Description. In order to emphasize the point that the covariant calculation of loop diagrams is by far the most sensible approach to follow, we remind the reader that a simple (noncovariant) semiclassical argument suggests that the lepton anomalous magnetic moment ought to be of the other sign (Welton 1948). That is, vacuum fluctuations due to photons emitted from the charged lepton disturb the alignment of u along B, so that for H0 = — u • B we expect K = A/i//z = A(cos 6) < 0. Now in the formalism of old-fashioned perturbation theory there are six triangle diagrams which contribute to the electron anomalous magnetic moment, only one of which has all forward propagating energy denominators of the form (Ea — Eb + je)" 1 . The latter graph is singled out by going in the infinite momentum frame of the electron with pz ->• oo, in which case the electron spin-flip matrix element of the longitudinal and time-like current (jV/$)i,-t gives Ke = cc/2n (Chang and Ma 1969, Bjorken et al. 1971). The transverse current ;'*m, however, picks up the tx/2n moment in a much more subtle fashion, involving cancellations of cut-off factors log A/me which arise in part from backward-propagating energy denominators in "z-diagrams" (Foerster 1972). Nucleon Magnetic Moments. Recall that the measured anomalous magnetic moments of the proton and neutron are KP = 1.79 and K„= —1.91, far larger than the 0(e2) lepton anomalous magnetic moments. This tremendous enhancement is presumably a strong-interaction effect, due to the charged cloud of 71* and other hadrons surrounding the "bare nucleons". While we have no right to think that lowest-order 0(g2) triangle loop graphs analogous to Figure 15.5 could accurately account for KP or K„ given the large value g2/4n « 14.3, the graphs of Figure 15.6 indeed do a respectable job. Ignoring the pion mass in the Feynman denominators, since ml /m2, ~ ^ , one finds for the proton (and neutron, making the proper isospin replacements in Figure 15.6—see Problem 15.2) that Figure 15.6(a) alone gives Kp=
-K„ = ^ — * 2.28.
(15.18a)
Electron Anomalous Magnetic Moment 307
(a)
(b)
Figure 15.6 Vertex modification to 0(g2) of proton em current.
The justification for keeping Figure 15.6(a) alone is on the basis of dispersion theory and the lowest-lying intermediate 2% mass states (see Section 15.H). In a field-theory approach one may add Figure 15.6(a) to Figure 15.6(b), along with a
(15.18b)
4.55.
The experimental values for KP and K„ lie closer to the former estimate (15.18a). Other Finite Triangle Graphs. The calculations of Ke„ and KP n have given us good practice in computing simple loop diagrams. Another example is the closed-fermion-loop calculation of the n° ->2y amplitude as shown in Figure 15.7(a). Note that C invariance does not force Figure 15.7(a) to vanish as in the Furry-loop example of Figure 15.4. This is because f/c(7t°) = + 1 in this case, while in Figure 15.4, the TC° is replaced by a photon with r\c{y) = — 1. In fact the n°yy loop with pseudoscalar n°NN coupling is completely finite (Steinberger 1949), but the axial-vector-2y loop integral for Figure 15.7(b) (remember that PCAC links n° and d • A0) is infinite, diverging linearly in a non-gauge-invariant fashion unless the loop is labeled in the symmetric fashion as shown in Figure 15.7(b). For the former case of Figure 15.7(a) plus its exchange graph, if we "pretend" that the internal loop is composed of structureless protons with pseudoscalar coupling to pions and y^ coupling to the photons (i.e., Fj = 1,
(a)
(b)
Figure 15.7 Triangle diagrams for ji° -» 2y decay (a) and axial-vector -* 2y decay (b).
308 Higher-Order Covariant Feynman Diagrams
Figure 15.8 Electron triangle loop, 0(e2) modification of photon-graviton vertex. F2 = 0), then it is straightforward to obtain for w^/m^ x 0 [see Problem 15.3 and (11.66)] U - P . W = ~ie29 j > ' Tr[>5(/ + t x (/ + t - *' - mNy\{l F
nyy=
-ccg/nmN.
mN)-\ - l ' - m*)" 1 ],
(15.19a) (15.19b)
It is interesting that (15.19b) gives \Fnyy/e2\ * 0.049m"1, not far from the experimental value of 0.037m"1, suggesting that it is the (heavy) nucleon mass in (15.19b) which sets the (small) scale for Fnyy. It is presently believed, however, that the ratio g/mN in (15.19b) really corresponds to the Goldberger-Treiman scale/^ 1 for "bare" fermions, presumed to be "elementary quarks". Indeed, a detailed PCAC analysis of n° -»2y based upon the axial-vector decay amplitude of Figure 15.7(b) leads to such a structure for Fnyy (Adler 1969—also see Problem 15.3). The internal quark triangle diagram is then in almost perfect agreement with experiment if there are three quarks of the same type traversing the loop. Lastly, a finite loop graph in the quantum gravity theory is that of Figure 15.8, representing the lowest-order quantum correction to the classical lightbending of Section 14.E [see Delbourgo and Phocas-Cosmetatos (1972), Berends and Gastmans (1976)]. The dominant loop corresponds to the lightmass electrons, but unfortunately this quantum correction is much too small to be detected in present solar light-bending experiments.
15.C Self-Energy Loop Diagrams Finite triangle loop diagrams not withstanding, there is still good physics to be squeezed out of manifestly infinite loop graphs. The QED bubble graphs corresponding to photon and electron "self-energy" corrections in lowest order are cases in point. Vacuum Polarization. In the QED photon self-energy bubble of Figure 15.9, an electron-positron pair can be created seemingly out of nothing (save a virtual photon) and could in principle "polarize the vacuum". This exciting quantum concept has been realized for quite some time (Uehling 1935,
Self-Energy Loop Diagrams 309
q.^—*
-
—q-A1
—
1>v
»
CTj
'
Q'H-
a (a)
Figure 15.9 Relabeling of electron loop momenta for 0(e2) vacuum-polarization diagram. Serber 1935), but its modification of Coulomb's law is extremely hard to detect, as in the Lamb shift (see Section 15.E). For an internal electron momentum labeling of / in Figure 15.9(a), it is clear that the vacuum-polarization graph n^q) diverges like j d*l/l2 for large /. Such a quadratic divergence is not easily handled, because the shift of momentum integration regions as in (15.11) can only be carried out if the integral diverges at most like a logarithm. Surface terms must be kept for linearly divergent integrals, and quadratically divergent integrals are simply ambiguous. For a photon bubble graph, however, we also know that the vacuum-polarization amplitude must be gauge invariant q%v(<0 = 7t„v(q)
(15.20)
so that if we are clever, there ought to be a way of using (15.20) to circumvent the ambiguities associated with the quadratic divergence. There are in fact many "regularization schemes" which do the job (the simplest being the dispersion approach of Section 15.G, as we shall see). For our purposes here we follow the "pragmatic" approach suggested in Section 15.A and conveniently relabel the loop momenta as shown in Figure 15.9(b). The Feynman rules for the bubble n^q) replacing g^ in the "bare" propagator numerator then give (Feynman 1949) n 0„v
„ ln\ - -i** f /**„ n^q) }<rp [{p
Tr +
yM + hi + m)yv(p' - \j + m) ^ ) 2 _m2 + .£][{p _ k)2 _m2 + . g]
a rd*p = -»' -3 -pr [2p„P, - &„«v - 0„v(P2 - k2 ~ ™2)l
<15-21)
where in this case D = d+d_ with d± = p1 — m2 + \q2 + p • q + ie. Since D->p 4 as p-> 00, (15.21) still diverges quadradically; we have, however, eliminated the linearly divergent p„gv cross terms with this choice of momentum labeling. Then in terms of the covariant integrals 14
74
j - p - = «o(2),
j - p - P»Pv = a2{q2)q^qy + c 2 (q 2 ^ v ,
(15.22)
we may rewrite (15.21) as (Problem 15.4) «MV(«) =
~ 5 bM«v(2a2(92) - ?<*o(q2))
tA
ft
. „.
(15.23) - 0„v(2c2(42) - ma0(q2)) - g„vq2(a2(q2) -
ia0{q2))].
310 Higher-Order Covariant Feynman Diagrams
To proceed further, we note that it is the c2(q2)gfiV term in (15.22) that we expect to be quadratically divergent and ambiguous [just as it is the c2(0)<7/IV term which is the divergent part of (15.11c)]. Since n^q) must be gauge invariant according to (15.20), we are at liberty to discard the ambiguous g^ term in (15.23) in favor of the gauge-invariant combination **„»(«) = ^(q2)(q2g^ - g^v)
(15.24a)
Aq2)=2-^[a2{q2)-ka0{q% (15.24b) n It is in fact possible to regularize these integrals to show that n^ is indeed gauge invariant, but this will not be worth our trouble, since it is an automatic consequence in the dispersion-theory approach of Section 15.G. To evaluate the integrals in (15.24b), we exploit the structure of the propagator denominators by expressing the Feynman integral (15.10a) as 1 1 r1 d+d„-2\_l[2-{d++d_)
dx + W+-d-)x\2-
(
>
Then we shift the region of integration in the logarithmically divergent integral a0 in (15.22) to find
a0(q2) = i j _ i dx J [p/2+iq2{ld:Px2)_m2
+ i£?,
(15-26)
where p' = p + \qx. To separate the logarithmically divergent integral a2(q2) from the quadratically divergent integral c2(q2) in (15.22), we must be careful when shifting variables from p to p'. It turns out, however, that one can uniquely separate these two terms by formally differentiating the p„ pv integral with d2/dqp dq„ (see Problem 15.4). The result for the a2 term is, apart from a finite constant, equivalent to applying p = p' — \qx in (15.22), leading to lr1 r dV 2 «2^ )~o x2 dx \ -=—= r—j=r = TTT- (15.27) 2y v ' 8-Li J [p'2 + ^q2{l - x2) - m2 + ie]2 ' 2 2 Now we are prepared to evaluate n(q ) in (15.24). For small q , we make a Taylor series expansion in q2 of the denominators in (15.26) and (15.27), cutting off the leading logarithmically divergent term at p'2 = A2 which corresponds to the divergent analog of (15.12), f d k * ^ m feM o g ^ . J [k2 - C + is]2 C
(15.28)
We then obtain the final form for the vacuum polarization (15.24) with (Problem 15.4)
**>-£
A2 m
q2 5m
4 —2 + —— 2 + 0(q ).
(15.29)
It is also possible to obtain n(q2) for general q2, but (15.29) will be more useful for our purposes.
Self-Energy Loop Diagrams 311
Charge Renormalization. The gauge invariance of n^{q2) insures that the "dressed" photon propagator maintains a simple-pole (q2 + is)'1 structure as q2 -+ 0, i.e., that the photon remains mass less. That is, the sum of the bare and lowest-order (vacuum-polarized) corrected photon propagator in the Feynman gauge is
•
^
-
"
+ ( - ^ ) ( - ' ^ ) ) ( - ^ ) - - ^
^
[1 - «
(1530)
since a conserved current hooking on to one of the ends of the propagator satisfies q^j^q) = 0 or j^q) — Ttmf{q). Then as q2 -+ 0, the factor 1 — 7i(0) in (15.30) modifies what we mean by charge; the e+e~ loop acts as a "capacitance for the vacuum", shielding or renormalizing the charge seen at each vertex to be a , A 4 e. (15.31) e« = 1 " ^£- ll°g o g _2 37r m The infinite cutoff A is therefore reabsorbed into the definition of physical charge at q2 = 0, or equivalently 2
9nve
•
l
n
_V„2YI _„
2
l
9"v
eR
,
l
a
-J5^
„2
R<2
, rii
+
*\
0(q)
(15.32)
Note that while the infinite constant 7t(0) is not measurable, its sign in (15.29) is positive (due to the minus-sign rule for closed fermion loops), as it must be if 1 — 7t(0) is to produce a shielding effect. On the other hand, 7t'(0) = a/157tm2 can be detected. As we shall see in Section 15.E, it detracts 27 Mc/sec (2.6 %) from the Lamb shift in hydrogen. This renormalized charge is what is measured in low-energy Compton scattering, where as in (15.32) the radiative corrections which modify the Thomson amplitude vanish as k -> 0. The proof of this assertion is a nontrivial result even for QED (Thirring 1950). We shall take up similar lowenergy theorems for charged hadrons in Section 15.H. Electron Self-Energy. Consider next the electron self-energy graph of Figure 15.10(a). The photon bubble in Figure 15.10(a) modifies the bare electron
P-I (a)
+ -<
"-<
"- +
(b)
Figure 15.10 Lowest order electron self-energy (a) and Dyson summation (b)
312 Higher-Order Covariant Feynman Diagrams propagator with numerator
Classically, self-energies of charged particles diverge linearly, but in QED the divergence is "only" logarithmic as k -* oo because the leading linearly divergent part is zero by symmetry arguments. However, (15.33) has a logarithmic infrared divergence as k -* 0, so we give the photon a fictitious mass X, k2 ->k2 — X2, for the time being. Then writing in general for constant A and B, S(p) = A + B(p-m)
+ S^pK/* - m)2,
(15.34)
where ~Lf(i> = m) ^ 0, we complete the square on the propagator denominators in (15.33) and shift the region of integration in the usual manner to obtain, apart from finite constants (see Problem 15.4) Jam [-1 , A= dX ~2^\0
r
fi {l+X)
\
B
d*k' 3am/ A2 l \ Ik'2 - m2*2 - X2(l - x) + ieV = ^n ( l 0 g H? + l) (15.35a)
a 1/ A2 9 ^ , m2N = - 7 z— ( °log g ^-2^ + -7 -- 2 21 log o g 1—. y) An \ mr 4 X2
(15.35b)
near mass shell, p2 « m2, p->m. The important point is that the infinite ultraviolet constant A is independent of the infrared divergence, while B is not [see e.g. Jauch and Rohrlich (1976) for more details]. Mass Renormalization. Since A in independent of the infrared divergence, we can absorb it into the definition of the electron mass in a manner analogous to the absorption of 7t(0) into the electron charge. Summing the bare electron propagator and the lowest-order self-energy correction, we obtain from (15.34) for p « m
1 p1 — m
'
p1 — + m
Ap :— ^m ) ~f>»—^m ± p^—. m — U«6> A' 1
1
where we have iterated the "small" constant A to all orders [1 + x x (1 — x) _ 1 ] according to the Dyson summation of Figure 15.10b. Then we may take the renormalized mass as mR = m + A, so that A is of no physical significance. However, B will play a more significant role because the propagator numerator in (15.36) becomes 1 + B rather than 1; but more about that later. For p1 =/= m, the (ultraviolet-) finite, but infrared-divergent radiative correction ~Lf{p) also modifies the propagator, but we shall not have occasion to explore this effect [see e.g. Karplus and Kroll (1950)].
Free-Electron Charge Form Factor
313
15.D Free-Electron Charge Form Factor We are now prepared to sum up all of the second-order radiative corrections to the minimal QED photon coupling ey„ for an on-mass-shell electron, as shown in Figure 15.11. This procedure just generates the complete electron form factors (15.7) to order 0(e2) due to the photon cloud surrounding the elementary electron. F1 Vertex Modification. In Section 15.B we used the techniques of covariant separation to split off the finite anomalous-moment form factor F2 from the vertex modification A„(p', p) of Figure 15.11b. The resulting charge form factor F™{q2 = 0), however, diverges for k -> oo and k -»0, rather than vanishing as it ought to, since Figure 15.11(a) already gives Fx(0) = 1. Before deciding what to do with FYM(0), let us construct F\M(q2) for small 2 q . Returning to (15.5b), we use (15.8) again to identify GM = Ft + F2 and then subtract off F2 as given by (15.9a) to find FV(t) = - ^
171
[P2a2(t) + tb2(t) + 2c2(t) (15.37) + 2p' • pa0(t) - AP2a1(t) + 2m2a1(t) -
2m2a2(t)\
where q2 = t, P2 = m2 — \t, and p' • p = m2 — \t. Following our pragmatic regularization approach, we subtract off FYM(0) OC C 2 (0) from (15.37) as F\M(t) - F™(0) = tF'!VM(0) + • • •
(15.38)
and calculate the finite contribution FiVM(0) from (15.37) by differentiation (Problem 15.5): F'!VM(0) = - ^
[-oo + ai - Ui + b2 + m2{2a'0 - 2a\ - a'2) + 2c'2]t=0. (15.39)
(d)
(e)
Figure 15.11 Complete electron form-factor corrections to 0(e2).
314
Higher-Order Covariant Feynman Diagrams
To obtain the various coefficients in (15.39), we must not set t = 0 until after we combine the three propagator denominators in (15.8): D = d+d_k2, where d± = k2 — 2P • k ± q • k and 2P = p' + p, q = p' — p. Then using the Feynman tricks (15.25) and (15.10b), we complete the square using k' = k + (P + \qx)y and shift the integration regions, leading to, for example, A\J d*k 13 [k' - Z2y2 + is]
\\dxCy2dy\J-2
a1{t) =
(15.40)
'
where Z2 = m2 — ?t(l — x2). It is then straightforward to find, for the integration coefficients defined by (15.8) (see Problem 15.5), -171
«i(0) =
2
m a2(0) =
— in 2
a'2(0) =
'
2m
•in
a\(0) =
'
b 2 (0)=-c' 2 (0) =
(15.41a)
6m4 ' .2
•in
(15.41b)
12m4' 2
•in
(15.41c)
2
24m
The coefficient a0(t) is infrared divergent, so we again give the photon a propagator mass,
a0(0 = j_ i dxj o
d*k' Z2y2 - X2{\ -y) + isf '
ydy\jF2
(15.42)
which then leads to (Problem 15.5) im2ao(0)
-1
l
= f dy.
-J.
f+^-,)
A2/m2
dz
,
m
Yz
=log
V (15.43a)
, /«.
hx4)-
—In I,
m
(15.43b)
Putting all this together in (15.39), we finally obtain F7M(0) =
a
m
3
1O 8T ~OI37tm 2 I
(15.44)
Total Charge Radius. The vacuum-polarization graph in Figure 15.11(c) modifies only the y„ charge form factor. According to (15.30) this contributes a factor of —q2n'(0)/q2 = — TC'(0) for small q2 to the charge form factor. The total charge form factor for small t = q2 is then
F1(t)=l+tF'1(0)
+ -
F\(0) = F'r(0)-n'(0)
=
(15.45a) 3nm
log m
(15.45b)
Free-Electron Charge Form Factor
315
where a is now taken as the measured (renormalized) fine-structure constant a « 1/137. Before applying (15.45) to a physical problem, however, it is time to pass to the limits A -» oo, X -+ 0, and directly confront the infinite terms FyM(0) and log(m/A) as well as the B term in Figure 15.1 l(d, e). Cancellation of Ultraviolet Singularities. It turns out that the gauge invariance of QED forces the ultraviolet-divergent parts of FYM(0) and B to cancel exactly in the charge form factor. At t = 0 the sum of all the graphs of Figure 15.11 contribute to the charge form factor as
JMOK = [i + FHo)]y„ + yM - mY W -m) + mi> - «)(* - m)~ \ = [1 + FTM(0) + B]y„
(1546)
where the factor of | is included with B because the self-energy correction of Figure 15.1 l(d, e) also contributes to vertices not accounted for in (15.46). The singular part of F^M(0) is obtained from the a0(0) and c2(0) coefficients of (15.37), the latter being directly inferred from (5.42) but with integrand weighted by the additional factor \k'2 at t = 0, n =
(15.47)
itx r
~2? J0
y
y
(•
a k (K * + 4m )
J [k'2 - m2y2 - k2{\ -y) + ie]
Now both FYM(0) and — B as given by (15.35b) are ultraviolet and infrared divergent, and it can be shown that they are equal, even retaining the finite constants,
F
™< 0 >=- B -£( l o 4 + s- 2 1 o 4)-
(1548
»
Therefore, these singularities cancel in the charge form factor (15.46). This result is in fact a consequence of gauge invariance, which for a zero-energy photon says that the formal equation J 1 _ 1 1 3p" p — m p — m ** p — m
(15.49a)
can be summed to the Ward identity (Ward 1950) 5
„£(p) = A > , p )
(15.49b)
in the Dyson sense of Figure 15.10(b). This relation directly gives — B = FYM(0). [Such a cancellation is not always so clean for radiative corrections to other vertices, such as in [i decay, \x -*• e + v + v'. In the latter case both the vertex modification and self-energy graphs contribute in 0(a) to modify Gw. One must then work harder to separate out all the finite contributions from the loop integrals. See Berman and Sirlin (1962).]
316
Higher-Order Covariant Feynman Diagrams
Cancellation of Infrared Divergences. The constants A and FyM(0) + B are independent of the infrared X cutoff. But there still remains the singular factor of \og(m/X) in the charge radius (15.45). The cancellation of this divergence is even more subtle than (15.48); it follows from the incoherent sum of the modified Coulomb and soft-Bremsstrahlung cross sections, the latter integrated over soft emitted photons [Bloch and Nordsieck (1937) Schwinger (1949)]. In quantitative terms this infrared theorem states that lim
"UV°"<* = finite. dn
~daCoxs\X) _ f'
+
dQ.
A->0
(15.50)
It is based upon the observation that a photon detector cannot distinguish between free Bremsstrahlung photons and virtual photons for low energies co < comax(X) = v^max + ^2> where comax P X is the minimum photon energy detected and dNk = d3k/2co is the density-of-states factor for a free photon. To lowest order in radiative corrections, (15.50) corresponds to the graphs of Figure 15.12, with the infrared-divergent interference term in Figure 15.12(a), of order Ze2 • Ze4 \og(m/X), cancelling the integrated square of the soft-Bremsstrahlung terms of Figure 15.12(b), of order (Ze3)2 j dco/co. To show that this infrared cancellation indeed occurs in Figure 15.12, we work directly with the electron charge form factor, identifying the Bremsstrahlung contribution for soft-photon emission as (see Section ll.E) e*(k) • p'
F\Br
e*(k) • p
(15.51) p •k p'k The unpolarized spin sum for (15.51) in the nonrelativistic limit | p' |, | p | < m is m (p' • kf
I\FT\2=-e2
2p'-p p'kpk
m +' (p • kf
e\2
1
m2cj2
(q • k)2 CO
(15.52) 2
2
2
1
where q = p' — p, q % - 1 , and co = k + X . Then integrating (15.52) over the soft-photon phase space according to (15.50) with k dk = co dco, we find
dNk X | f f (t, k)|2 *
o
k2dkdQk e2 „\2 1 2w{2n)3 m2co2 _
1-*- dco^
2M
3nm2 2at 3nm
log
(q-k)2 co2 1+
co
(a)
/
d
N
(15.53)
2con
4 x •
2co'
+ x *
(b)
Figure 15.12 Pictorial representation of lowest-order infrared theorem.
Bound-State Lamb Shift 317 On the other hand, the square of the electron charge form factor (15.45) to order 0(e2) is l r M I
F
,
l ^l=[
[«
1 +
<xt /,
w(
l0g
m
3
1\1 2
t
2M /,
m
3
l\
I - 8 - 5 ) J " 1 + w( l0g I-8-5} (15.54)
Then adding together (15.53) and (15.54) according to (15.50), we find
|JM0I2 + J
dN t I|Ff(t,k)| 2
0
2at 1 + - 37tm2
(15.55) °g
m 5_3_1 2wm„+ 6 8 5
and we observe that the infrared-divergent log X terms cancel in the sum (15.55). In general this "fortuitous" cancellation occurs in all higher-order interference terms as well [see e.g. Yennie et al. (1961)].
15.E Bound-State Lamb Shift We have seen how the virtual cloud of photons surrounding a free electron generates the radiative corrections to the free-electron form factors. In a similar manner, such a photon cloud surrounding a bound electron in an atom modifies the (form-factor) static Coulomb interaction between the electron and the atomic nucleus. The 2Si-2Pi energy-level "Lamb" shift in hydrogen measures just this form-factor modification of Coulomb's law. (Recall Figure 5.1 for a pictorial representation of this shift.) Qualitative Estimate. According to the Dirac theory of an atomic electron (see Section 5.D), the hydrogenic energy shift fof n = 2, AE(LS) = A£2st — AE2Pi, vanishes because the level shifts are degenerate for a given ./-value, in this case j = \ with / = 0, 1. But in 1947 Lamb and Retherford detected such a shift of size AE(LS) ~ 4 x 10_6eV, measured in frequency units as Av(LS) = 1057.8 ± 0.2 Mc/sec. Possible causes of this shift are either the photon cloud surrounding the electron or the pion cloud surrounding the proton nucleus, either of which alters the Coulomb potential V = -Za/r to V + AV, where [recall (5.138)] AV = i(dr)2V2V = ^
(<5r)2<53(r).
(15.56)
This AV then generates a first-order perturbation-theory energy shift according to (5.148), AEnl = inl | AV| n/> = ^
{5rf \
{5rf8u0. (15.57)
318 Higher-Order Covariant Feynman Diagrams The pion cloud gives a fluctuation Sr ~ m" 1 , which corresponds to an energy-level shift A£2,o = ml^l^-ml ~ 10" 9 eV, much too small to be the cause of the Lamb shift. On the other hand, Welton (1948) estimated 5r for photon-cloud "vacuum fluctuations" from the classical equation of motion for electron oscillations, m 6'r = eE(t). The quantum Fourier component of the squared electric field is E2, = 2 x 4na)2(a>2/(2n)32a>) = a>3/2n2, so that a Fourier average of 5r is = —2 — £» = —2 — =—log^^(15-58) 2 4 2 m J cu nm J„min
of
A£ 2 0
mZ4a5 1 =— log - * 2.7 x 10" 6 Z 4 eV,
(15.59)
a
D7C
or a frequency shift of Av x 650 Mc/sec for Z = 1. So clearly the photon cloud around the electron must be the cause of the Lamb shift. For exotic atomic configurations, however, such as a %\i atom, me in (15.57) becomes the reduced mass mtlmK/{mn + m j ~ 120me, which scales up the pion-cloud shift to a value greater than the photon-cloud shift (see Problem 8.6). Connection Formula. In principle the Lamb shift can be obtained from the exact electron propagator in an external Coulomb field, dynamically linked to the proton via the Bethe-Salpeter equation [cf. (1957)], a relativistic generalization of the Lippmann-Schwinger equation [also see Erickson and Yennie (1965)]. However, for our purposes it will be sufficent to treat the external field as a perturbation, both in first and second order in Ze but to order e2 in radiative corrections, as shown in Figure 15.13. The graph in Figure 15.13(a) is just the Dirac theory of the hydrogenic electron (see Section 5.D) including all of the 0(m(Za)4) fine-structure energy-level corrections. The graphs in Figure 15.13(b) are the bound-state analogs of the vertex-modification and vacuum-polarization radiative corrections just considered in Sections 15.B-D for free electrons. They give rise to energy-shifts 0(w(Za)4a). Lastly, the graph in Figure 15.13(c) is second
x
<^f" -f x
(a)
<
+
(b) Figure 15.13 Lamb-shift (b, c) corrections to Dirac atom (a).
(c)
Bound-State Lamb Shift
319
order in the external field, generating an energy shift formally of smaller size, 0(m(Za)4aZa). However, for low-energy virtual photons and nonrelativistic bound electrons (as in hydrogen), the electron propagator in Figure 15.13(c) is dominated by its nonrelativistic forward-propagating bound-state part. This in turn contributes an additional enhancement factor of (£;„ — co)/m ~ Za. to the level shift, of order 0(m(Za)4aZ<x(Za)_1), which is the same order as graphs of Figure 15.13(b). While the graph in Figure 15.13(c) requires soft photons (a> < mZa), those in Figure 15.13(b) are clearly dominated by high-energy photons (co > m) in order that the electron-positron bubble may be produced in a nonvirtual sense. These are in fact the same frequency cutoffs used in the Welton estimate, suggesting that we divide the Lamb energy shifts A£ = £ — £ (a) into the two domains, A£ = A £ w + A£(c),
(15.60)
where AE"0 is calculated for high-energy virtual photons (co > m) and A£(c) is calculated for nonrelativistic virtual photons in the Welton range m(Za)2
Ff(t)*l+ F?(t)*^.
M
3nm
l0g
m 2a>ffl„
+
5 _3 _1 6 8 5
(15.61a) (15.61b)
It is now a simple matter to convert (15.61) to an effective static potential V" = — (Zoc/r) + AV. Following the discussion of Section ll.G, we make a nonrelativistic reduction of the electron momentum-space current and then transform back to a coordinate-space static potential (11.72). Then substituting (15.61) into (11.72b), we obtain for a hydrogenic atom with e-> — e for
320 Higher-Order Covariant Feynman Diagrams the negatively charged electron, AV =
42a2 2
lin "
m 2<„
log
1 3 + 5 8
+ •
' 6
Za 2
<53(0 + 4nm2r3 a • L, (15.62)
where the second factor of f in the s-wave part of the potential, as well as the last spin-orbit term, are due to the anomalous-magnetic-moment form factor (15.61b). This A F then leads to a first-order energy-level shift for high-energy virtual photons (co > co^\n ~ m) of A£ =
1 5
m
2
mZ 4 a 5
3 +
8
ho
(15.63)
1
+ •4™ 3 (/ + i)(j + ±) for j = l±j, where we have used the Schrodinger wave functions of (5.146)-(5.148) in going from (15.62) to (15.63). Low-Frequency Part of the Lamb Shift. Now we assume that co < co^ and follow the nonrelativistic calculation of Bethe (1947) for the low-frequency part of the shift, A£ (c) . According to second-order bound-state perturbation theory (i.e., the VG0 V term), the bound-state energy shift due to Figure 15.13(c) has the exact nonrelativistic form A£ (bd) =
'^fclJ-Afc
^ n+i
(15.64)
E,„ - co
CO
where Ein = Et — E„ and the states are normalized noncovariantly so that j = ep/m and A = sjy/2a> with e2 = 4rox. The integral in (15.64) diverges linearly like a nonrelativistic self-energy [Figure 15.13(c) is, after all, a type of self-energy graph], so we must subtract off the free-electron energy shift found by evaluating (15.64) with Ein = 0. This is another example of the pragmatic regularization scheme. Then integrating over the virtual-photon phase space, we find A£ (c) = AE (bd) - A£ (free) =
da
2a 37tm2 2a 37tm2
Z IP» co
|_
12 c
1^
1 Eln-co l^niax
V col
|fc,|£«.log(—j—)
To evaluate the sum in (15.65), we first neglect the Ein term next to con in the numerator of the logarithm in (15.65) to write A£ (c) =
2« 3nm:
(15.65)
Ein\
Z IPml2^,
log
2coif> m
+ log m 2E„
m
(15.66)
Bound-State Lamb Shift 321 In the first term the sum is independent of the logarithm, with \vni\2Ein =
I
kn\[[Pi,H],Pi]\n> (15.67)
= i | d3r |
g
(Za) 2 w = (-2.811769883(28), 2£„ ( + 0.030016697(12),
2S, IP.
K
' '
Finally then we write the low-frequency part of the level shift as A£(c) =
4mZV 37tn
,
ML
1
{Zafm
log —=^x 5,.0 + log ? —rj + log ^ r ^
(15.69)
Note that the 0(mZAcc5) order of (15.69) is the same as the Welton shift (15.59) and also the relativistic part of the shift (15.63). For / = 0 all three logarithmic terms in (15.69) can be recombined to resemble the Welton shift. In fact, if one assumes that coj^ = m (as Bethe did), then the level shift from (15.69) for n = 2, / = 0 in hydrogen is [E2 = 8.31967ma2 from (15.68)] mc2a5 . m Av(c) = -—— log -=- « 1047 Mc/sec, (15.70) onn E2 very close indeed to experiment, 1058 Mc/sec. This has prompted many "nonrelativistic physicists" to dismiss the high-frequency part of the shift (15.63) as an unessential relativistic "technical correction." If, however, one chooses an upper cutoff of c o ^ = 2m—which is just as reasonable a value, since it corresponds to the onset of e+e~ pairs—then (15.70) is altered to Av(c) x 1237 Mc/sec, no longer so close to the experimental shift. Total Lamb Shift. The essential point is that we need not choose a particular cutoff for the virtual-photon energy in (15.69); instead we connect A£(c) with A£(b) by requiring co^ax= a>j^n in (15.60). Then the log(2a>/m) (5,0 terms in (15.63) and (15.69) exactly cancel, and the total shift is independent of any cutoff energy. This is similar to the free-particle infrared theorem, with the measured differential cross section independent of the fictitious photon-mass term log(m/A). The total energy shift for / = 0 states is then (Kroll and Lamb 1949, Weisskopf and French 1949) _ 4mZ V l 0 g
1 (Z^
, + l 0
(Za)2m 5 3 1 3 8 ~2lT+6-8-5+° (15.71a)
322 Higher-Order Covariant Feynman Diagrams while for / ^= 0 states with j = / ± \ it is -,n,li=0
4mZV Inn*
,
(Za)2m 2£„ - 4(2/+ 1)(2/ + 1 )
(15.71b)
The original calculations were based upon old-fashioned perturbation theory; we have followed the more compact covariant approach of Feynman (1949) for the high-frequency part of the Lamb shift. For the case of the specific 25^.-2?^ Lamb shift in hydrogen, (15.71) gives the second-order shift A £J (2) m = A£ 2Si
A£ 2P*
5
ma. log m IE 2,0 6n
5 +
3 1 3 , 6 - 8 - 5 + 8 - l
0 8
m 2 l ^
+
1 8
(15.72a)
or in terms of numerical frequencies for each of the terms in (15.72a), Av(2) = 953.40 + 113.03 - 50.87 - 27.13 + 50.87 - 4.07 + 16.96 = 1052.19 Mc/sec.
(15.72b)
To (15.72b) we must add the second-order binding (6.76 Mc/sec, see e.g. Erickson and Yennie 1965) and the fourth-order (ma6), reduced-mass, recoil, and proton form-factor corrections [—1.04 Mc/sec, see Appelquist and Brodsky (1970)], leading to the net present prediction AvQED = 1052.19 + 6.76 - 1.04 = 1057.91 ± 0.16 Mc/sec. (15.73a) The most precise measurements of this shift to date are [Triebwasser et al. (1953), Robiscoe and Shyn (1968)] AV"P :
j 1057.77 ± 0.10 Mc/sec, i 1057.90 + 0.06 Mc/sec.
(15.73b)
Clearly the consistency between (15.73a) and (15.73b) is satisfying. The agreement between theory and experiment for the IS^-IP^ Lamb shifts in deuterium and He + and the 3Si-3Pi shift in hydrogen is almost as good as (15.73). Other accurate atomic measurements, such as the helium fine structure, the muonium hyperfine splitting, and the n = 1, 3 S 1 - 1 S 0 level shift in positronium, are also consistent with QED calculations [see e.g. Lautrup et al. (1972)]. The sum total of these bound-state successes, combined with the phenomenally accurate anomalous-magnetic-moment agreement for electrons and muons, plus the pinning down of the otherwise elusive 27-Mc/sec vacuum-polarization shift in hydrogen, leaves little doubt that QED via minimal coupling is "the" correct theory, a statement that can seldom be made with such confidence in modern science.
Renormalization in Field Theory 323
15.F Renormalization in Field Theory Considering the inherent ultraviolet and infrared infinities in closed loop graphs, the reader might well regard the pragmatic regularization procedure used in Sections 15.B-E to obtain such precise QED predictions as wizardry of the highest order. We devote the next two sections to justifying this procedure in the context of field theory and dispersion theory. Mass Renormalization. Returning to our brief discussion of second quantization and many-body fields in Section 4.D, the fundamental dynamical field construct is the lagrangian T — V rather than the hamiltonian T + V. In relativistic theories the lagrangian density, if (x), is always a Lorentz invariant, remaining so after the interactions are turned on and the bare fields are replaced by interacting fields, etc. For the case of the lagrangian density for a free electron, ^0(x) — "AoMO? — m o)>Ao(x)> t n e presence of an external em field coupling in a minimal way leaves m0 unchanged in i?(x). There is no getting around the fact that m0 is infinite in QED in some gauges, but by defining the mass shift dm in terms of the finite physical mass as 8m = m — m0, the interacting QED lagrangian density can be written as JSP(x) = 0(x)[(i0 - eA(x)) - m0ty(x) = if free (x) + JS?,(x), ^ f r e e M =
(15.74a) (15.74b)
&,(x) = -e$(x)4(x)il/(x) - (5m^(x)iA(x). (15.74c) Then the second-order self-energy bubble graph of Figure 15.10(a) due to the first term in if 7 generates a mass shift cancelled by the second term in if,, which generates a first-order "contact" graph. The cancellation occurs order by order in Figure 15.10(b); the fact that 8m is infinite is immaterial to the argument. Thus the only mass remaining in if (x) is the physical mass in if free (x). The same type of mass renormalization occurs for other relativistic theories, for example in the Klein-Gordon lagrangian density with 8m ij/ijj -* 8m2 >*>, etc. Consider the contrast between this relativistic mass renormalization with mass terms in the lagrangian-density numerator on the one hand, and on the other hand the "quasiparticle" mass shifts in nonrelativistic many-body theories with masses occuring in the kinetic-energy denominators, p2/2m -» p2/2m*. The latter mass shifts cannot be renormalized away as in relativistic theories. They are in fact finite (for the finite ultraviolet cutoffs demanded by the physical requirements of the system in question) and can, in principle, be measured. Examples are the "Hartree-Fock" dressed mass of nucleons in nuclear matter, the "random-phase" dressed mass of electrons in a (screened) electron gas, and the "polaron mass" of electrons in a phonon bath (see Problem 15.6). Renormalization Constants. Aside from the infinite mass shift 8m = A, the remaining infinite constants in QED—FYM(0), B, n(0)—can be absorbed
324 Higher-Order Covariant Feynman Diagrams
(a)
( b)
Figure 15.14 Field-theory renormalization of QED vertex.
into "renormalization constants". That is, to second order we define Z\x = 1 + F™(0), Z 2 = 1 + B, Z 3 = 1 - n(0),
(15.75)
called the vertex, wave-function, and charge renormalization constants, respectively. The higher-order QED vertex becomes dressed according to Figure 15.14(a), with the multiplicative rescaling of the dressed fields
^(x)-yz7<(x),
(15.76)
combined with the vertex infinity occuring in the dressed lagrangian vertex T, i ->Z;~ 1 r£. This rescales the renormalized charge as depicted in Figure 15.14(b), e=
(15.77) Zi
For the gauge-invariant QED theory, however, we also saw in Section 15.D that to lowest order FYM(0) = — B, a result most easily obtained from the Ward identity (15.49). In the language of renormalization constants this says that Zi
e = x/Z-,3 ee.0 -
(15.78)
The final form (15.78) then requires that all conserved em currents have their charge renormalized in the same manner (via Z 3 only), regardless of the type of interaction, be it for leptons or hadrons. Then, for example, the equality of bare charges, e0(proton) = — e0(electron), automatically insures the equality of the physical charges, e(proton) = — e(electron). Renormalizable Theories. A renormalizable theory is one in which the degree of internal ultraviolet divergence does not increase indefinitely in powers of momenta for increasingly higher-order loop graphs. For example, in QED, higher-order loop graphs diverge only logarithmically via constants directly related to A, B, TZ(0), and FyM(0). An interesting question then arises: can we determine a priori if a field theory is renormalizable or if the higher-order infinities "run away," leading
Renormalization in Field Theory 325 to a nonrenormalizable theory? For simple interactions the answer is in the affirmative. It is an easy matter to show that (see Problem 15.7) D = 4 - |F e x t - B6"' - £ Ntfj
(15.79)
j
is the degree of divergence of any particular loop graph, with T oc pD as p -> oo (D = 0 corresponds to a logarithmic divergence), where Fexl (Bext) is the number of external fermion (boson) lines at the interaction vertex and Nj is the number of vertices of typej. The internal vertex parameter r\j in (15.79) is defined by rij = 4 - iF'f - Bf - rij,
(15.80)
with rij the number of derivatives (i.e., powers of momentum) at vertex j . The significance of (15.79) and (15.80) is that n alone determines the renormalizability of the theory (say for a single type of vertex), because if n < 0, then D becomes more and more positive (divergent) as N increases, and the theory is nonrenormalizable. More specifically, the renormalizability criterion is
(
= 0, < 0, > 0,
renormalizable, nonrenormalizable, "superrenormalizable".
For the QED vertex ij/y^ i/M", we have F int = 2, Bint = 1, n = 0, so that f/ = 4 — f • 2 — 1— 0 = 0, and the theory is renormalizable, as expected. The utility of this internal divergence exponent n is that it is simply related to the dimension of the associated internal vertex coupling constant ginl as (Problem 15.7) dim ginx = m".
(15.81)
This profound rule immediately allows us to catagorize various relativistic theories considered in previous chapters: i. the coupling constants e for QED and gKNN for pseudoscalar nN coupling are dimensionless, corresponding to renormalizable field theories. ii. f/mn for pseudovector nN couplings, Gw for Fermi four-point weak vertices, and G* for newtonian gravitational coupling have dimensions m~i,m~2,m~i, respectively, corresponding to runaway, nonrenormalizable theories. iii. g„nn for scalar-pseudoscalar coupling in the cr-model has dimension m1, corresponding to a superrenormalizable theory. From the above remarks the reader may begin to sense the elegance and power of formal field theory. To proceed further, we note briefly that Dyson (1949a) has explicitly shown the connection between Feynman diagrams and the field-theoretic expansion of the S-matrix in terms of T-products as in (7.48). But even then the separation of divergences and calculation of overlapping divergences for higher-order loop graphs is a tricky business (Dyson
326
Higher-Order Covariant Feynman Diagrams
1949b, Salam 1951, Weinberg 1960). Moreover, the proof of the renormalizability of a given field theory is highly nontrivial, though it has been carried out, for example, for spinor QED (Ward 1951) and scalar QED (Salam 1952). More recently, the subjects of dimensional regularization and renormalization in nonabelian gauge theories appear to have a bearing on unification of strong, electromagnetic, and weak interactions. The interested reader is urged to consult more advanced texts on field theories. For an up-to-date summary of how to handle infinities in a field-theory context, see e.g., Delbourgo (1976).
15.G Dispersion Theory and QED The closed-loop Feynman prescription, or equivalently field theory, is not the only way of extracting the physics out of higher-order loop diagrams. Alternatively we may keep the scattering particles and couplings always on shell and renormalized by interpreting only the imaginary parts or discontinuities of Feynman diagrams as having physical content, consistent with the unitarity of the on-shell S-matrix. The real parts of scattering amplitudes are then found via dispersion relations with knowledge of subtraction constants, the latter corresponding to the infinities in field theory. Put another way, dispersion theory has no infinities; instead it has ambiguities embodied in possible subtraction constants. While the connection between scattering and dispersion theory was sensed in Heisenberg's original article on the S-matrix (1943), its application to QED for vacuum polarization and other processes was worked out piecemeal in the 1950s and early 1960s. Nevertheless, dispersion-theoretic ideas were fully utilized in the classical theory of optics as early as 1930, and so we begin our discussion of this subject at that point. Dispersion Relations in Classical Optics. The classical harmonic-oscillator model of matter consists of charged electrons being driven from "atomic equilibrium" by external time-dependent fields. For a monochromatic field E(t) = E(co)e',a1', Newton's second law leads to an electric dipole moment p(co) = -ex(co) = - - 2
4
^2°
E(co),
(15.82)
to — ^Vo '
where ma>l = k is the atomic spring constant, r0 = a/m is the classical electron radius, and e <-> r/2 measures the damping. Then with n the number of oscillating electrons per unit volume, the macroscopic polarization P(co) = np(co) = D(a>) — E(co) leads to a dielectric constant , -> D{(o) £(
4nnr0 T
rr •
(15.83)
v E(co) (to2 -tol + ie) ' While the form of this simple dielectric constant displays some very general features, its analytic structure in co is of particular significance. Recall
Dispersion Theory and QED 327 that the product of Fourier transforms £(co)£(co) corresponds to the convolution £>((') = j
e(t' - t)E(t) dt,
(15.84)
— 00
which makes physical sense for a linear medium only if s(t) is a causal response to a delta-function input, e(t) = 0 f o r * < 0 .
(15.85)
This says that the medium cannot respond until the electric-field input arrives. Then in terms of the Fourier transform e(co) for complex co, (15.85) leads to e(co)=\
dte(t)ei'Rewe-tlmc',
(15.86)
which implies that e(co) is analytic in the upper half co-plane due to the damping exponential in (15.86) for t > 0. Such a conclusion is consistent with (15.83), which is analytic in co except for poles at co = ±co0 — '£ in the lower half co-plane. To proceed further, we consider the general properties of a function F(a>) analytic in the upper half co-plane which follow from the Cauchy formula „/ ^ 1 i' dco' F(co') F((o) =— f — — ^ v ' 2ni Jc at' -co
K(15.87)
'
for any contour encircling co in the upper half co'-plane. Letting co -> coreal + ie for e > 0, the contour in (15.87) can be deformed to the real co'-axis as shown in Figure 15.15. Then we use (x + is)"x = P/x — inS(x) to express (15.87) as a Hilbert transform, or equivalently
F(m)
' f f noy) _i r
* M ? I
(1588)
2m J _ „ co — co — ie 7C J _ ^ co — co — is provided that F(co)-+0 along the infinite semicircle in Figure 15.15; otherwise this asymptotic contribution must be added to (15.88). If F(co) is not known on the infinite semicircle, a sufficient requirement for the validity of (15.88) is that the integral J00 dco [Im F(co)]/co be finite, which is equivalent to Im F(co) -> co~c vanishing as a power [whereas Re F(co) could vanish like OJJ
T£/-
(a) (b) Figure 15.15 Deformation of Cauchy contour in complex co'-plane.
328
Higher-Order Covariant Feynman Diagrams
(log a))~l] as to -* oo. Given the "dispersion relation" (15.88), the real part of F(a>) can be obtained from the knowledge of the imaginary part of F(co) for all co. Returning to the classical dielectric constant for a linear medium, e(a>), the function F(co) = e(co) — 1 obviously satisfies the analytic and also the asymptotic properties necessary for the validity of (15.88); the asymptotic behavior e(co) — 1 -• 0 as co-> oo follows from (15.83) and corresponds to D(co) -> E(co) for an infinite-frequency input field which sees the bound electrons as effectively free. Finally, the reality of e,(t) can be converted to the "crossing" condition in the transform space, e*(co) = e( — co*), so that Im e(coR) = — Im e( — coR) is an odd function of co for co real. Combining these properties into F(co) = e(a>) — 1, (15.88) becomes the Kramers-Kronig relation e(co) - 1 = -1 If
dco'
Im e(co')
% J,0
co — co — is _ 2 f °° dco' co' Im e(co' n J0 co'2 — co2 — ie
Irae(-co') — co — co — is (15.89)
For Im e(co') = £(5(co' — co0), corresponding to absorption at a single (atomic) oscillator frequency co0, (15.89) assumes the simpler form (15.83) with £ = 2n2nr0/co0. To connect up with scattering physics, [e(co) — l]co2/4nn can be identified with the classical forward-scattering (Compton) amplitude /(co) which measures the scattering of the incoming photons off the bound-electron targets. The amplitude f(co) then obeys a dispersion relation [for F(a>) = f(co)/u)2—see Problem 15.8]
/ M - ^ C ,,"M;'m-y>
(.I*,
n J0 co (co — co — IE) The optical theorem Im /(co') = (co'/47i:)crto,(co') then allows us to make practical use of (15.90). Furthermore, inputing the model-independent low- and high-frequency limits of/(co) obtained from the classical amplitude analogous to (15.83), ^2 foe co2, co->0, (15.91a) iclass^) — ~
2
2 , •
co -coo + is
r
0 Uf.o>
^
^
(1591b)
the dispersion relation (15.90) leads to the sum rules (see Problem 15.8) co2 f00 dco' crlot(co') / H - ^ I 772 as co ^ 0 , In -'0 co l r11' —/(co)-» r0 =-—2 dco'
-'n
(15.92a) (15.92b)
Dispersion Theory and QED 329 The former sum rule corresponds to Rayleigh scattering, and the latter Thomson limit is equivalent to the Thomas-Reiche-Kuhn sum rule derived from [x, p] = i in nonrelativistic quantum mechanics (recall Problem 9.7). The lesson learned from this analysis is: If we are given a dispersion relation such as (15.90) [which follows from the model-independent statements of analyticity (i.e., causality) and crossing (i.e., reality)], input modelindependent unitarity (i.e., the optical theorem), and asymptotic and threshold behavior, then a complete description of the scattering amplitude plus constraints (i.e., sum rules) on the amplitude and cross section results. Our goal now will be to develop a similar program for QED and strong interaction processes. Propagator Dispersion Relations. We follow the same spirit as the above analysis, but now apply it to the self-energy operators in QED. For simplicity we consider initially the self-energy X(p2) for spinless 0 + particles (i.e., (7-mesons) and (j)3 interactions g4>3. The problem is that we do not know a priori the analytic behavior of E(p2) in the complex p2-plane [as we did for a causal e((o)]—but we can find out with the aid of the Feynman loop diagram of Figure 15.16(a), dAk _ \ (k - m + is)[{k - Pf pf - m2 + ie]
(P ) = lQ
2
(2nf J0
2
dX
J [k'2 -m2
d4k' + p2x(l -x)
(15.93) + is]2 "
While (15.93) diverges logarithmically as k -> oo, we apply the identities 1 2
(m - a)2 to write (15.93) as
=
f
™* Jm2 (m'2 -a)3' ,2
S
.oo
,1
J^/2
(P ) = T ^ f f —2 2^, ' ^ •• (1595) 2 J 167r J0 x)-ie m2 m - P *(1 We note two properties of (15.95): (i) even though £(p 2 ) is logarithmically divergent, it is manifestly analytic in the upper half p 2 complex plane; (ii) only the real part of 2(p 2 ) is logarithmically divergent, with £(p 2 ) -»log p 2 as p 2 -» oo. These properties suggest that £(p 2 ) could be made to satisfy a once-subtracted dispersion relation.
J-^-
2
f ^L = -^(1594) J (k'2 -C + is)3 2C ( >
dX
i
o
" "
k (a) (b) Figure 15.16 Bubble-graph (a) calculation via unitarity graph (b). The broken pt and p2 lines in (b) represent on-mass-shell particles.
330 Higher-Order Covariant Feynman Diagrams To proceed further, we use (15.95) to compute the (finite) imaginary part of S(p2) as - I m X(p2) = 4r- f
dx
107t J0
167t
|o
dm'2 8(m'2 - p 2 x(l - x))
C •>
2
dx6\x(l-x)-yj,
with the 0-function insuring that the argument of the ^-function is in the region of integration. The inequality x(l — x) > m2/p2 is equivalent to (Problem 15.9) 1 f 4^ . 1 i 4m2" 1<X- l2< < x - i < - / 1 - - T (15.97)
-2^—^ - 2
for p2 > 4m2, so that (15.96) can be expressed as Im 2
1 2 (P 2 ) " fzZ 16TC JV " - p3 " %
-
4m2
)-
(1598)
This final form is a simple statement of unitarity. To see this, apply the covariant form of unitarity for two-particle intermediate states (see Section 10.B), Im TfT = l- \ dp™2 TtfTni,
(15.99)
to Figure 15.16(b) with T}7 = -Z(p 2 ),T* r = Tni = -g. The resulting two-particle phase-phase integral for on-shell intermediate momenta v\ — v\ — m2, Pi + P2~ P, can then be evaluated in the CM frame as
-Im I(p2) = \g2 \ dpT(Pl, P2) = £ ^ Jm j dQ 2 4%2 4plJp2
1 16TI yj
0(p2-4m2), pr 2
(15.100)
since pf* = ^p2 - Am2 for p2 > 4m2 from (10.2). We observe that (15.100) is precisely (15.98). In order to apply the general dispersion relation (15.88) to this problem, we must construct an F(a>) -* F(p2) such that F(p2) vanishes as p2 -*• oo. On the other hand, we also require the self-energy dispersion contribution to vanish on mass shell, p2 -*m2. We combine these two requirements by choosing F(p2) in (15.88) as p1 — mz No crossing condition is needed for the p2 dispersion integral, since unitarity requires p2 > 4m2, ruling out negative p2. Substituting (15.101) into (15.88)
Dispersion Theory and QED 331 for to -»p 2 , the final form of our dispersion relation is
^ - i K I - ^ ^ f 4^,„
y ,
...
(15.102)
J « 4m2 (p - m2){p'2 -p2 -is) 2 where Im I(p' ) is given by (15.98) or (15.100). The form (15.102) is sometimes called a once-subtracted dispersion relation, referring either to the subtraction structure of (15.101) or to the process of subtracting the formal identity (15.88) [including the (nonzero) contribution on the infinite semicircle £(oo)] from the same relation evaluated at p2 — m2 (see Problem 5.9). If we write E(p2) = A + B(p2 - m2) + £/(/>2), then the constant A is E(m2), and we can pick up the constant B by making a second subtraction in (15.102), again at p2 = m2 (Problem 15.9). From the structure of (15.102), we might claim to have "solved" the fieldtheory problem of infinities; there are no such infinities in (15.102). This is a false impression, however, because an infinity of field theory is replaced by an ambiguity in dispersion theory—we do not know the value of the real, finite subtraction constant £(w2). Returning to QED, dispersion-theory methods are a very simple way to obtain the vacuum-polarization contribution to the electron form factor. Direct application of unitarity to this problem gives, instead of (15.100) (see Problem 15.9).
I (*„v - < , ) = \ \
2i
dpTiPu Pi)<e+e- \T\yv>
= \e2 \ Trfo + m)yX-ti = (4 V
+ m)yv dpTip* Pi)
- «„«.) *« ( l + ^ )
^l- —
(15-103)
0(q2 ~ 4m2).
This result is automatically gauge invariant because the intermediate e+e~ pair are required to be on mass shell by unitarity. Then making one subtraction at q2 = 0 because Im n(q2)-* const as q2 -* oo in (15.103), the amplitude n(q2) satisfies the dispersion relation
2 „°o Am 2
dq'2fr(l + 2m2lq'2y\ ./2,v2 .2 q'2(q'2 -q2-
- 4m2/q'2
.-„, fe)
(15104)
,2 a
/i/„4\ +, 0(q*). « -o 157tw2 This is in agreement with (15.29), but certainly much easier to obtain, because there are no ambiguities associated with quadratic ultraviolet divergences in the dispersion-theory formalism.
2
«
332
Higher-Order Covariant Feynman Diagrams .
Vertex Dispersion Relations. We may extend this analysis to vertex form factors, which we expect to be analytic in the momentum-transfer invariant in some region of the complex plane. We leave it as a problem to show that for equal masses, spinless (p3 coupling and F(t) = 1 + A(t), the second-order vertex modification, A(r), as given by the Feynman rules, indeed satisfies an unsubtracted dispersion relation (Problem 15.10). Here we assume upperhalf-plane analyticity in t for the QED form factors F2(t) and F^t) = GM(t) — F2{t), and compute their imaginary parts from the unitarity condition (Tf°? written as 7}, here) - (Tfl - Tff) = -tv[y„ Im GM - P„ Im F2/m\upA't (15.105)
The extension of unitarity to a process involving particles with spin means that Tff is converted to Tj{ by the PT transformation (see Section 6.B, C) PT = — ia2 in the Dirac spin space, with the covariants in (15.105) becoming -gwgw{PT%{PT)-1
= -y„
-gwgw{PT)Pl(PT)-1
= -P„. (15.106)
Then for Tfi =
(15.107a)
GM(t)y, - F2(t) -£ uvA",
PT invariance, combined with the PT phase [determined by the CPT theorem (6.88)] r\PT = rfc = — 1, gives Tff
G%(t%
(15.107b)
F*2(t) ^ upA».
m
The difference of (15.107a) and (15.107b), as needed in the unitarity condition, indeed leads to Im GM and Im F2, which we have already anticipated in (15.105). Now we saturate the right-hand side of the unitarity condition (15.105) with the 0(e2) e+e~ two-particle state, as depicted in Figure 15.17. The vacuum-polarization contribution in Figure 15.17(b) has already been calculated in dispersion theory via (15.104). Therefore we concentrate on the vertex-modification contribution in Figure 15.17(a), with the on-shell momentum labeling A = p' - p = q
q',
P = W + P\ p + q = p' + q' = P + Q,
= p'2 = q* = c[2 = m2, A 2 = t,
P2 = Q2 = m2
(15.108a) (15.108b)
it.
Dispersion Theory and QED 333
Im
x-
+
= x
X
(b) Figure 15.17 The imaginary part of the electron em form factors to 0(e2) in the e+e~ channel. Diagrams (a) and (b) correspond to the s- and f-channel e+e~ Bhabha poles and generate the vertex and vacuum-polarization corrections, respectively.
Then ignoring the free-particle spinors but using the Dirac equation, we write (Problem 15.10)
=
i e 2 (•
d
P2{Q)
(P + Qf = e2
r <we) (P + Qf
[Tp(_
^~4
+
m
^
[ - ( i t + 2P • Qhv - AmQ, + 2P.Q-
(15.109) 2Q,Ql
where dp2(Q) refers to covariant two-body phase space, with total momentum Q = q' + q integrated over, but A = p' — p held fixed, with q = Q + jA, q' = Q- ±A. Next we invoke covariance arguments to determine the various (finite) phase-space integrals in (15.109). Since t is the only variable external invariant in the problem, we write > -2 = = aM a0(t),
dp (Q)
\ffoJ dp2(Q)
(P + Q)2
\ffwQ^^ai{t)P\ -
d p
Q„QV = a2(t)P,Pv + b2(t)(A,Av - tgj.
(15UOa) (15.110b)
Symmetry considerations rule out covariant terms odd in A,, in (15.110), and the on-shell conditions Q • A = \{q2 - q'2) = 0,
P • A = \{p'2 - p2) = 0
(15.111)
force the structure A^ Av — tg^ in (15.110b). Then using the identities (recall 15.100) f dp2(Q) - 1(0 = ^ ^ 1 - ^ - 9(t - Am2), (15.112) \dp2(Q)Qll = 0,
334 Higher-Order Covariant Feynman Diagrams
along with (P + Qf = 2(P2 + P • Q), we may contract (15.110) with P„ or g^ to obtain the simple relations (Problem 15.10) a,{t) + a2{t) = I(t)[t - 4m 2 ]" 1 ,
b2{t) = -il(t),
a2(t) = a0(t)--^^.
(15.113a) (15.113b)
Finally we substitute these results into the second-order unitarity condition (15.109) and isolate the coefficients of P^ and y„: lmF™(t) = 2m2e2{a1+a2)
=C ^ l l - ~ \
* 6{t - 4m2), (15.114a)
Im GyMM(t) = -e2[(t - 2m2)a0(t) + |/(t)].
(15.114b)
Now (15.114a) is the only contribution of 0(e2) to Im F2, and its asymptotic behavior is Im F2(t)->0(t~i), so that F2(t) should obey an unsubtracted dispersion relation
F2(t)=ic
dfim
F2{t>)
n i4m2 t' - t - ie a 26 a ~2n sin 26 ' - ° ' 2n'
=am2 c ^ ( i ~ 4 m 2 / t ' ) " i n J4m2
t'(t' - t - ie)
v(15.115);
where sin2 6 = t/4m2. Not only is the correct anomalous magnetic moment reproduced [recall (15.15)], but the t-dependence in (15.115) is the same as the field-theory result. Before writing a dispersion relation for GM or Fu we must work out a0(t) in (15.114b), but only after introducing the fictitious photon mass in (15.110a) to circumvent the infrared divergence. Evaluating this integral in the CM frame Q = 0, adding in the vacuum-polarization contribution of Figure 15.17(b), and subtracting (15.114a) from (15.114b), we obtain the complete 0(e2) result (Problem 15.10) Im F1(t) =
2JI - Am2 It (15.116)
Since in fact Im F^t) does not vanish as t -* oo, we must include one subtraction in the dispersion relation for F1, which we make at ( = 0, where Fx(0) = 1. Then we have
m.l+Lf
"JSLIM
,15,n)
Substituting (15.116) into (15.117) leads to a complicated t-dependence for Fx(t), but one in agreement with field theory. At t = 0, the derivative of
Dispersion Theory and QED 335
,, P + k22 - k" 2
(b)
(a)
Figure 15.18 Photon-photon scattering amplitude (a) and corresponding doublespectral-function diagram (b). (15.117) is trivial to evaluate:
corresponding to the Feynman loop-integral answer (15.45). Dispersion Relations for Photon-Photon Scattering. Consider the lowest-order box diagram for pure photon scattering as shown in Figure 15.18(a). In the limit of all soft photons, the internal electron box cannot be produced as viewed from any channel because it is always below the unitarity thresholds. This implies that the low-energy limit of the photon-photon amplitude is T(kt ^0,k2->
0, k\ -> 0, k'2 -* 0) = 0.
(15.119)
Classically, of course, photon-photon scattering is a nonlinear effect which does not exist at any energy for the linear Maxwell equations. To simplify the structure of the nonvanishing quantum-mechanical amplitude, we sum over all photon helicities. Then we use crossing symmetry to write the scattering amplitude in terms of the Mandelstam invariants (satisfying s + t + w = 0)as Tcov(s, t, u) = 2e*[A(s, t) + A(u, s) + A(t, u)\
(15.120)
where each invariant amplitude A represents one such box diagram of Figure 15.18(a), and the factor of 2 corresponds to direct plus exchange scattering (or equivalently to the internal loop momentum labeled in both directions). Then the Feynman integral for each box gives, for example,
A(s, t) = i I f d*P T l W - mY Vi(* POI
J
-h-m)-1 (15.121)
which diverges logarithmically as p -» oo. Regularization of this field theory integral by incorporating (15.119) is exceedingly difficult to perform
336 Higher-Order Covariant Feynman Diagrams
(Karplus and Neumann 1950). However the dispersion-theoretic calculation of (15.121) is much more transparent and manifests the crossing properties and the corresponding subtraction at (15.119) in a simple way (DeTollis 1964). Two-particle unitarity for Figure 15.18(a) takes the electron and positron having momenta p + k2 and p — /q on mass shell in (15.121) according to (15.99), which we write as Im As(s,t) = \ £ ^ pol
\d*p J
x Tr[(* - m)-1^
-h+
m)'1
mWM + %2 - & -
x fry + h+ m)t2]d+((p - ktf
- m2)d + ((p + k2f - m2). (15.122)
That is, the discontinuity of A(s, t) across the s-channel "production cut", A(s + is, t) — A(s — is, r) for s > 4m2, is proportional to a modified Feynman integral with the s-channel propagators replaced by mass-shell delta functions in (15.122). Cutkosky (1960) generalized this unitarity result by analytically continuing (15.122) into the t-channel for t positive and then applying t-channel unitarity to (15.122), so that all four internal particles are on mass shell, as shown in Figure 15.18(b). The resulting double spectral function p(s, t) = Im, Ims A(s, t) can be obtained as (Problem 15.11) p(s, t) = J £ f
x tW + h + m)t2]8(P2 - m2)8((P - *i) 2 - ™2) x 8((p + k2- k'2)2 - m2)3{(p + k2)2 - m2) (15.123) _ n2 -st + 2{m2 - sr/4(s + r)}2 ~T {st[st - 4m2(s + t)]}* ' with the denominator vanishing on the boundary of the physical region in either the s- or the t-channel. Given (15.123), the double dispersion-relation form (Mandelstam 1958) j p(s', r') ds' dt'/(s' — s)(r' — t) diverges, so we make a subtraction at s = 0, t = 0 with zero subtraction constant by (15.119), leading to the representation (dropping the is factors) s rds'Im As(s',t = 0)
^"-;J
•(,-)
t r At' Im At(t', s = 0)
+
«1
«-,)
-
(15124)
st st (r• ds' as dt' 1 p{s', t') 2 (s'-s)(t'-tY n) sW The functions Im As(s', t = 0), Im At(t', s = 0) in (15.124) can be obtained in closed form from (15.122) with, e.g. (k2 — k'2)2 ->0 in the former case (see Problem 15.11).
Dispersion Theory and Strong Interactions 337 Field Theory vs. Dispersion Theory for QED. As long as we are making a perturbation expansion in powers of a coupling constant, such as e in QED, there is no difference in the field-theory (Feynman loop integrals) and dispersion-theory results [see e.g. Chou and Dresden (1967)]. Some QED calculations are easier to carry out using field-theory techniques, some using dispersion theory. Some higher-order radiative corrections have been worked out in dispersion theory [see e.g. Petermann (1957)], but it now appears that field-theory computations are more tractable for, say, the sixthorder anomalous moment. At the very least, however, QED dispersion theory is pedagogically appealing because it does not deal with infinite quantities.
15.H Dispersion Theory and Strong Interactions For strong interactions we do not have obviously convergent perturbation series in coupling constants (g2/4n ~ 10). Therefore a rigorous derivation of dispersion relations based upon infinite sums of Feynman diagrams is not very revealing. They have been verified, however, for nn scattering and in a limited domain for nN scattering using nonperturbative techniques. For our purposes it will be more productive to postulate the smooth analyticity behavior of S-matrix elements in momentum space as the natural analog of causality in coordinate space, as in classical optics and QED. Then the hope is that "nearby" singularities will control a given strong-interaction process, allowing us to make an effective power-series expansion in momentum around a given kinematic point rather than one in the coupling constant. Owing to the amazing success of the Chew-Low p-wave analysis (see Section 12.D), which adopts this philosophy, such a scheme may be more than just wishful thinking. Even the PCAC hypothesis has an appealing dispersion-theoretic interpretation in terms of the nearby pion pole (Bernstein et al. 1960). 5-Matrix-Theory Postulates. As an alternative to a Feynman-diagram dynamical theory of strong interactions, the following properties of the stronginteraction S-matrix are assumed to be valid [see e.g., Chew (1961)]. i. Lorentz invariance: Any (two-body) scattering process has an S-matrix (i.e., M-function) which can be expanded in terms of Lorentz-invariant amplitudes At(s, t) depending only upon the Mandelstam invariants (with u determined by the Mandelstam relation s + t + u = £ mf). The associated kinematic covariants KJ,V are composed of external momenta and spin factors in such a way so as not to introduce any artificial kinematic singularities or zeros in the At (this is almost always possible except for processes involving virtual photons, and even then the resulting kinematic zeros can be taken into account): M„v... = E At(s, t)Kl....
(15.125)
338 Higher-Order Covariant Feynman Diagrams
ii. Analyticity: The invariant amplitudes At(s, t) are reasonably smooth analytic functions of the channel invariants s, t, u. This analyticity must be consistent with simple Feynman graphs, but its global extent is confined by other properties to follow. iii. Crossing: The At(s, t) can always be analytically continued from one channel (say with s > 0, t < 0) to another channel (with s < 0, t > 0) reached from the first channel via the CPT transformation. This is due to postulate ii and the absence of 9(E0) functions in the Feynman rules and in (15.125). If C-invariance is a property, say of the t-channel process, then s — u crossing symmetry is a property of the invariant amplitudes, At(s, t,u)= ± At(u, t, s). iv. Unitarity: The only nonanalytic behavior of the invariant amplitudes is due to purely real elementary-particle "poles" below a physical threshold or due to "production cuts" above threshold with absorptive (imaginary) parts determined by unitarity in the physical channel in question. Strong interactions can, in some cases, saturate unitarity in a single partial wave [recall the A(1232) resonance in Figure 8.1(b)]. v. Subtractions: We cannot determine the strong-interaction subtractions from unitarity perturbation diagrams as in QED. Instead the asymptotic behavior in the forward direction is found from the optical theorem and the finiteness of total cross sections, while off the forward direction it is assumed given by Regge-pole behavior. The subtraction constants are usually determined by low-energy theorems, such as those of current algebra. vi. Bootstrap dynamics: The "elementary" hadrons themselves are assumed to be self-consistently generated. Cross-channel resonances or elementary particle poles create forces which in turn generate direct-channel resonances in a manner consistent with analyticity and unitarity. The Chew-Low u-channel nucleon pole crossed over into the s-channel and then unitarized (see Section 12.D), thereby generating the A(1232), is an example of the bootstrap. Likewise the crosschannel p force in nn scattering self-consistently bootstraps the p resonance in the direct channel. Low-Energy Proton Compton Scattering. Consider yp -> yp scattering as a strong-interaction process (aside from the factor of e2 in the amplitude due to the external photon probes). At low photon energies we separate the entire yp amplitude into the nearby singular proton pole at threshold (Tp) and other background contributions (T) as shown in Figure 15.19. First we assume for simplicity that the proton and photon are spinless, so that the entire T-matrix is an invariant amplitude, which we write in terms of the crossing-odd variable v = (s — u)/4mp as T(v, t). Since T(v, t) is even under s <-> u crossing, it is a function only of v2, satisfying the general fixed-t dispersion relation (ignoring subtractions) 1 f dv'2 Im T(v', t) T(v, 0 = - J - ^ a Z ^ ^ - = TP+T.
(15.126)
To extract the proton poles from (15.126), we use the variables v and vB= —k'- k/2mp, which are related to the Mandelstam variables via s - m2 = 2mp(v — vB),
u — m2 = —2mp(v + vB).
(15.127)
Dispersion Theory and Strong Interactions k
k,
P
P ,N
339
(a)
• • •
(b)
Figure 15.19 Separation of proton Compton amplitude into proton-pole parts (a) and background graphs (b).
Simple one-particle unitarity then gives for the s- and u-channel proton intermediate states (Problem 15.12) Im Tp(v, t) = ^
^
[5(v - vB) + 3(v + vB)] = - ^
«5(v2 - v& (15.128a)
where e is the renormalized proton charge. Then from (15.126) we obtain 6 Tp(v, t) = - \ dV* ^ W ' ^ . = l , (15.128b) p z z nJ v - v 4mp(vg - v2) in agreement with the sum of the renormalized-field-theory poles for spinless scattering. Next we reinstate the proton and photon spins but specialize to forward scattering in the lab frame with v = a>, the photon lab energy. We then expand the covariant amplitude according to (15.125) but in the simple transverse gauge with two-component (proton) spin matrices
Tfy = 4n[/i(c0)£'* • £ +f2((o)ia • e'* x E].
(15.129)
This choice of invariant amplitudes allows us to identify ft (
as c o - 0 ,
(15.130a)
Im/^cojocco
as co->oo,
(15.130b)
340
Higher-Order Covariant Feynman Diagrams
where r0 = oc/mp is the classical proton charge radius. Since/^cu) is crossing even, we may disperse in co2, and then only one subtraction is needed for convergence, which we take at co = 0. That is, for F(co) = [fi{co) + r0]/a>2, (15.88) becomes (Thirring 1950, Gell-Mann et al, 1954a) fiH
da>'2 Im/i (to') co2 r00 da' crtM(co') + r0 = - \ —j2 2 2 n JQ co' CO (co' — co — is) In2 '0 co'2 — co2 - is ' (15.131)
This dispersion relation is reasonably well satisfied by data. For the crossing-odd amplitude/2(
(15.132)
where
E'* X E).
(15.133)
Actually the Compton low-energy theorem can be extended away from the forward direction, but (15.133) will suffice for our purposes (Problem 15.12). If we assume that