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!
] aapear, which depend on B in the case of a non-abelian gauge fields. The following will illustrate this procedure on concrete examples. III. QUANTUM ELECTRODYNAMICS This results scheme It is imply [<$i, 92] e
section has a methodological character. Here we show how all known of quantum electrodynamics can be obtained by means of the general of part I. well known that the Lagrangian of electrodynamics:
w h e r e « £ f - ^ / < ^ is invarrant wiih respect to the abellan gauge group:
The first recipe of Sec. I is to choose a hypersurface f = 0. There are two well known special cases: the Lorentz gauge and the Coulomb (radiation) gauge:
which gives us the equations of the "hyperplanes" in the manifold of all fields A(x). The corresponding integrals of the type (1.2)
and
36_ 218 where and □ mean the Laplace and d'Alembert operators respectively, do not depend on A(x), hence we can put without loss of generality In order to proceed to relativistic perturbation theory it is useful to find a generating functional of the form (1.4) for the non-perturbed Green's functions:
Here??, *}. ^ are the sources of the fields puted by means of a translation
that does not affect
This functional can be corn-
and is equal to
where
and
From (2.3) and (2.4) the Wick theorem follows, and hence, all usual diagrammatic techniques with well known expressions for the electron lines and vertices and purely transverse photon Green's functions (Landau gauge). In the second procedure we have to choose the additional term AS. Choosing it to be
we see the corresponding factor <J> [ A](cf. 1.5) does not depend on A and there fore we can put 4) p = Ience ew eome ence more et ohe eiagrammatic cechniques with the transverse photon Green's functions of the form
which for the special case £ = i ii equal to the Feynman function:
Both frameworks of perturbation theory obtained are well known and lead to the same results for physical quantities.
37
219
Exact relations among Green's functions of electrodynamics can be obtained by a change of variables in the functional integrals. As an illustration of this we shall derive the Ward identity6 and the relations between Green's functions in different gauges:
where
A rotation of the spinor field into:
in the numerator of (2.5) gives rise to the factor:
where
Differentiating with respect to c(z) and putting afterwards c = 0
we obtain from which the Ward identity follows:
connecting the Green's function G(p) and the irreducible vertex part' ^ f A: ?-<*,). It is clear that this identity is valid in any gauge of the photon Green's function, because in the derivation the change of variables was made only on the spinor fields. Let us look at the transition from the Coulomb gauge to the Lorentz gauge for the electron Green's function:
where
In order to accomplish the transition we put under the integral signs in the numerator and in the denominator of (2.6) the expression yj$(a C- 3^ A^)<*C0O which does not depend on A. Then we perform a transformation of the type (2.2) which does not affect the action S. Under this transformation the arguments
38
220 of the delta functions change as In the factor exp { ie [ c(x) - c(y)])which emerges in the numerator we can put for c(x) a solution of the equation Ac0O + div A - 0 , i.e.,
where
When this substitution is made the integrals over c(x) in the numerator and denominator cancel each other. The formula that results is:
It expresses the Green's function in Coulomb gauge as a power series of Green's functions im the Lorentz gauge.7 IV. THE YANG-MILLS FIELDS 1. Yang-Mills field theories8 are the simplest example of fields associated with non-abelian gauge groups. It is very convenient to describe Yang-MUls vector fields related to any simple compact Lie group, G,9'10 in terms of matrices B^ 00 which form a definite representation of the Lie algebra characterizing the group. It is clear that such a representation may be defined by
where TA are linearly independent matrices representing the Lie algebra, are normalized according to the relation T>(T„ Tk ) = SL £ay. n is the number of group parameters, and fc* („> is a function with vector index h- and isotopic index a. As is known one may represent the latter index to denote a matrix element of £ > » by means of the relatton,! '.where t a ^ c are the totally anti-symmetric structure constants of the group G. The gauge transformations are:
where &00 is an arbitrary matrix-function with the values in G. The Lagrangian
39
221 WhCre
is invariant with respect to such transformations. It is also clear that X =* X , + £. L where £, is a quadratic form and X , is a sum of trilinear and quadratic forms in the fields B(x). 2. Now we shall proceed with the formalism developed in the part 1, making use of the first recipe. As in electrodynamics the most convenient gauges seem to be the Lorentz or Coulomb gauges:
Both equations are matrix equations and represent really n conditions (n = dim G) instead of only one condition as in electrodynamics. In order to compute the factor A_ (B] we shall remark that we are interested only in its value for the transverse fields ( W = o ) . For such fields the only solution of the equation^B^ ■ O in ft will be the identity element of the group G, and so the entire contribution to the integral (1.2) is given by the neighborhood of the identity element. Therefore we can put
where u(x) is an element of the Lie algebra of the group, and retain in only the terms linear in u(x):
where D is the d'Alembert operator. The condition (1.2) reads as
where
Formally [6] is the determinant of the operator A. It is useful to have also another realization of this operator. We introduce instead of the matrices u(x) the column vector U / x ) in such a way that U =i^T«, T„ ' beinn ghe matrices so fhe adjoined representation of G. In this new representation the operator A is given by the matrix
1P_ 222
Extracting the trivial infinite factor det D we are lead to a following expression for JU A L [6]-.
where $ (x) is the Feynman Green's function for the d'Alembert operator
On the left-hand side of (3.2) the trace is understood in the operator sense, whereas on the right-hand side the trace is taken only in the matrix sense. Analogous computations in the Coulomb gauge give the factor \{B]J II is given by the expression,
where £ (£; is a Green's function for the Laplace operator:
and the indices i, . . . if| run over the values 1, 2, 3. 3 Relativistic invariant perturbation theory for computing the Green's functions in the Lorevtz cauge, viz,
where arises naturally in the Lorentz gauge when developing the functional
41
223
in a power series in the parameter e . The expression U A j 6 ]can be interpreted as an addition to the action, but this addition does not have the form j<£&<) <*x associated with some Lagrangian. The term of degree e* of this series in a striking way looks like a contribution of a closed loop along which a scakr particle is propagating and interacting with a vector field in a tri-linear way. This circumstance permits a description of the perturbation theory in terms of additional graphs containing the propagation of a fictitious scalar particle and a vertex with one vector and two scalar ends. The elements of this diagrammatic technique are shown in Fig. 1, the corresponding analytic expressions are given by:
Figure 1. In order to calculate the contribution of a given diagram we ought to integrate the product of the expressions for its elements over the independent momenta
42
224
and sum over discrete indices, and then multiply the result by ( _ / , where v is the number of vertices, i, the number of internal lines, S the number of closed scalar loops, and R is the rank of a symmetry group df the diagram concerned. In this way all the effects of the presence of a factor \ U J can be interpreted as the introduction of a fictitious scalar particle. The propagation lines of this particle should be taken into consideration only inside the diagrams so that they form closed loops. Moreover these scalar particles behave in a way as fermions The factor (-1) , indeed, where S is the number of loops is characteristic of fermions. The last fact becomes less queer if we remark that for A L [6J one can rewrite the integral representation,
in terms of the anticommuting fields u and u . Putting this expression in the integral (1.3) we are lead to the usual Feynman integral for a two field system: The transverse vector field Py , 3K 6,< = o a nd scalaa ffrmion field 4. In the second-type approach to Yang-Mills fields it is natural to choose an additional term to the action as in electrodynamics, i.e.,
Then after expanding the functional<J> [6}e*p( i(3t/J5)}in nhe eowers so f €w are led to a non-transverse Green's function, in particular, for £ -1 , ,he Feynman function. Nevertheless in this case it is no longer possible to account for all additional diagrams generated by introducing into our diagrammatic technique just one new line and one new vertex as was done in the previous case. 5. Sometimes the so-called first order formalism is more useful. This formalism can be obtained if the Lagrangian (3.1) is re-expressed as follows:
and to integrate in the functional integral over the fields *V and Fp„ as if they were independent, extracting, as usual the "orbit volume". Then for instance, for the Lorentz gauge we arrive at the integral,
Here the integral over F can be computed exactly, thus leading us to the formerly investigated formalism. However, if we do not do that, but yet in the integral (3.6) expand the expressionexp L((c]in powers of 6 , we will obtain a new variant of the diagrammatic perturbation theory with three lines, correspon ding to functions* >and one vertex describing tri-linear interaction
43
225
FBB. The elements of this diagrammatic technique are given in Fig. 2 and formulae (3.7); the field B is represented by a single line, whereas F is represented by a double line:
Figure 2. The lines and vertices describing the propagation of the fictitious scalar particles and their interaction with the vector ones are the same as in the case of the second-order formalism, because the factor depends only on B, but not on F. 6. In the first^rder formalism discussed above we have three full one-particle Green's functions which can be expressed in terms of the corresponding nonperturbed functions (3.7) and three self^nergy parts, which in second order perturbation theory take the form:
44
226
Here k0 denotes some fixed 4-momentum, with and a, b, c, d, e are some renormalization constants (indeed only first derivatives with respect to momentum of second derivatives of £ * ^ (?) and third derivatives of (f^are uniquely defined). In ■ there is also a contribution of an additional diagram and only due to its appearance can we assume the function to be purely transverse, i.e., proportional to v - ^A t f , by choosing in sorne special way the constants a, b,.and c such that b = 0, and a = - c . 7. In order to establish a connection with the canonical quantization method it is useful to start from an integral over B and F in Coulomb gauge with sources:
Following Schwinger4 we shall choose for the dynamical variable the transverse (in three-dimensional sense) components of the field Bj, and Fj 0 , i= 1,2,3. We also assume the sources present only correspond to our dynamical variables; this means that
45 227
In three-dimensional notations our Lagrangian (3.5) takes the form
The absence of the sources corresponding to F i k and BQ permits the integration in (3.8) over these fields, which is equivalent to' changing in the resulting integral over the remaining variables B., F. the matrix F.. into w
1
10
IK
and also the o -function into
Now, put into the integral (3.8) the factor <«C(x)that really does not depend on F Jo and perform a translation F{o - FF ^ 3f d • c,us inducing the transformations:
Let c 0 be a solution of the equation:
which can be expressed by a Green's function depending on B (cf. Ref. 4)
After performing a translation C-*C +C.the S -function appears asTj^(CC + 6 C B* , d-C]) and we can put c = 0 everywhere except when equal to the argument of this function. The resulting integral
cancels with
46
228 Hence the integral (3.8) can be transformed, up to an infinite factor, to the form
where
and
The relations obtained look like the usual formulae of quantization of a classical Hamiltonian system using Feynman method (cf. Ref. 11). The transverse variables £ . (x) and £>.Cx> are playing the role of canonically conjugate variables, whereas the functional H[f, b] that of the Hamiltonian function. Following the general quantization scheme, we ought to introduce the operators £* (x) £ * (jT^fulfilllng the transversality conditions: * ' °v
as well as the commutation relations
where
The energy operator /) can be optained by putting the operators ^ and f instead of the functions b and f into the functional H. The canonical quantization of Yang-Mills fields given above is equivalent to the one proposed by Schwinger in Ref. 4. It is also known that the Feynman and the canonical quantization procedure are equivalent. In particular the functional integral (3.9) is the generating functional for the Green-Schwinger function. Strictly speaking, the following relations occur:
47
229 where
The left-hand side is an average over the physical vacuum of the chronological product of the Heisenberg operators B ( x ) ) . . . F ( y ) ) . . . As a closing remark we indicate that the presence of the factor AR [ 8 ] in the primary integral (3.8) is necessary for casting it into canonical form. 8. After having established the equivalence of our quantization with more conventional one in terms of operators, we can define the unitary S-matrix using well known reduction formulae (cf. Ref. 12). For the sake of brevity we return to the second-order formalism, so that the Coulomb-gauge Green's functions are given by the following expressions:
where
Take now their Fourier transforms;
which obtain a final (i.e., to
are proportional t o 5 ( £ p ) d u e to translational invariance. In order to the transition amplitudes between an initial state containing r particles and state containing s particles, r + s = n, we have to go to the mass shell put p 2 = 0 for all the momenta p) in the product of
and n factors of the form for every particle. Her| e^p is a unit transverse polarization vector^ e v = oandu where the last sign takes account of the difference between the in- and outgoing particles; the ^normalization constant Z is defined by the residuum of the one-particle Green's function for p* =0. More exactly, for p 2 = 0 this function has the asymptotic form:
up to infrared singularities.
48_ 230 From the practical point of view it is useful to express the matrix elements of the S-matrix in terms of the Green's functions in the Lorentz gauge, for which we have developed above a covariant perturbation theory. As a matter of fact this can be done and the corresponding formulae have exactly the same form as the ones developed above in the terms of Coulomb-gauge Green's functions. We need only change the Coulomb gauge Green's functions to the Lorentz gauge Green's Ab
functions, and put instead of the factor Z R the residuum of &^ (p2) for p 2 =0. For a more detailed explanation we consider the connection between these two types of Green's functions corresponding to different gauges. We put into the denominator and numerator of (3.10) the factor not depending on B:
and then perform a translation 6~* b ° In the numerator the following integral over Si appears:
from which can be extracted a factor °,, formation changing the field B satisfying the field transverse in the three-dimensional sense, depends on B, so that 5n* is a complicated terms of this functional developed in powers of
B^ ( where Ua is a gauge trans Lorentz conditiondK 6,. = ointo a i.e., div 6 = o This transformation functional of B. Here are the first £:
Here T denotes the three-dimensional transverse part of the corresponding vector. The rest of the integral cancels with AR 16] and we are lead to the expression of the form:
that can be used to calculate the Coulomb Green's functions in terms of the Lorentz Green's functions. When expanding in terms of perturbation series we can ascribe the extra vertices to the terms of (3.11), so that any Coulomb Green's function is really an infinite series containing the integrals of Lorentz-gauge Green's functions with any number of external lines. In this way the connection between the Green's functions in different gauges appears to be very complicated. Our scheme becomes much simpler, however, if we go to the mass-shell: We ought to compare only the terms having the asymp totic singular behaviour whenf*-^ 0■ It 's clear tnat the pole f°r a given p will remain only if the whole effect of extra vertices reduced to terms of the type of
49
231 self energy insertions into a corresponding external line. In the limit />1-* 0, the resulting Lorentz and Coulomb Green's functions will differ by the factor
where for which the relativists invariant perturbation theory could be developed as well. It can be shown that the recipe given by Feynman in (2) for computing one-loop diagrams is equivalent to taking into account some of the terms appearing in the expansion of the factor <^s [fi] in powers of 6 . In the general case this expansion gives rise to the infinite number of extra vertices, whereas in the formalism of the first type it is sufficient to introduce only one extra vertex. As a closing remark: The computation of S-matrix elements in perturbation theory without taking into account factors of the type & L [ B ] or 4> [B] would lead to the formally invariant amplitudes, which would not, however, satisfy the unitarity condition. This fact was indeed revealed by Feynman in Ref. 2. V. THE THEORY OF GRAVITATION Now we will examine one proposed scheme for quantizing the theory of gravitation. Here we limit ourselves only to an outline of the perturbation theory and compute the analogue of \[B] The Lagrangian of the gravitation field,
is invariant with respect to general coordinate transformations which we write here in an infinitesimal form:
50
232
Here SA/ are arbitrary infinitesimal functions. The coordinate transformation depend on four arbitrary functions. Therefore in our scheme a hypersurface in the manifold of all possible fields should be given by four conditions, and we choose the harmonicity conditions::3
If for the variables of functional integration we c h o o s e ^ °)^and arrive at an integral of the form
V* we will
where
The notation ,(9) recalls the origin of this factor from the harmonicity conditions. In spite of the fact that the integral over H , in (4.1) can be taken explicitly we will prefer not to do it and work in the formalism. The diagrammatic perturbation technique arises in a natural way if we put
and then expand the integral (4.1) in the powers of K At that moment. we get two kinds of vertices: One of them corresponding to interaction * (and the other being generated by the factor AJ9)which we shall now evaluate. We must know ^ f g ] 0 n l y yor rhe earmonic cields. For ruch hields
and the general formula (1.2) for the factor we are seeking takes the form:
Proceeding thereafter as in the case for Yang-Mills fields we obtain the result:
51
233
A general term in the expansion (4.2) describes a fictitious vector particle propagating along a loop with n external lines and interacting with the "harmonic" field £ / v following the coupling rule K ^f 3 ^a / This paper was received by the Publishing Department July 5, 1967. REFERENCES 1Bogoliubov and Shirkov, An Introduction to the Theory of Quantized Fields, Moscow 1957 (in Russian). 2 Feynman, R. P. Acta Phys, Polonica, 24, 6 (12), (1963). 3 De Witt, B. S., Relativity, Groups and Topology, London, (1964). "Schwinger, J., Phys. Rev. 125, 1043, (1962); Phys. Rev. 127, 321, (1962). 5 Feynman, R. P., Phys. Rev. 80, 440, (1950). 6 Ward, J. C, Phys. Rev. 77, 293L, (1950); Phys. Rev. 78, 182L, (1950). 7 Schwinger, J., Phys. Rev., 115, 721, (1959). 8 Yang, C. N. and Mills, R. L., Phys. Rev. 96, 191, (1959). 'Glashow, S. L., Gell-Mann, M., Ann. Phys. 15, 437, (1961). 10 Utiyama, R.( Phys. Rev. 101, 1957, (1956). "Garrod, C, Rev. Mod. Phys. 38, 483, (1966). 12 Matthews, P. T., Relativistic Quantum Theory of the Interactions of the Elemen tary Particles, Russian translation (1959). 13 Fock, V., The Theory of Space, Time and Gravitation, issued by the State Publishing Department, Moscow, 1955 (in Russian).
52 THE FEYNMAN INTEGRAL FOR SINGULAR
LAGRANGIANS
L. D. F a d d e e v
A generalization is obtained for the continual Feynman integral, adapted for the quantization of a mechanical system describable by a singular Lagrangian. As an example, the quantization of the Yang-Mills field is considered. The quantization of classical mechanical systems using the Feynman integral [1] has not, up to now, enjoyed the popularity which it deserves. In our opinion it is the most convenient of the known methods of quantization and can be applied to situations in which the generally accepted canonical quantization encounters difficulties. In this paper we try to show this by means of the example of a mechanical system, the Lagrangian of which, L(q, q), is singular, in the sense t h a t the usual equations
cannot be solved for the q. Such s y s t e m s were investigated in general form by Dirac [2, 3 ] . For them he developed a generalized Hamiltonian formulation and discussed its application to quantization. But in his approach, in particular, in the application to a gravitational field [4, 5], a number of difficulties occur. Thus, the problem of ordering the operator factors, which is essential for the verification of the coordination of the constraints [6], is nontrivial. Moreover, the phase space is nonlinear, and the spectrum of the generalized coordinates does not coincide with the ordinary spectrum of the canonical variables (compare [7]). Finally, the formulation, in the case of field theory, is not relativistically covariant, which complicates the control of infinities. We intend to show that the difficulties enumerated above do not occur in quantization by the method of the Feynman integral and do not have the same central c h a r a c t e r . Of course, the infinities which are characteristic for relativistic field theory remain even in this approach, so that the problem of r e n o r m a l i z a tion still r e m a i n s . The fundamental proposition of the paper is formulated in terms of a mechanical system with a finite number of degrees of freedom. This makes the discussions and conclusions shorter and more t r a n s p a r e n t . In addition their general nature becomes clearer. The transition to field theory is made in the usual way. Field theory gives the most interesting examples of systems with singular Lagrangians. All fields having some kind of geometrical sense, e . g . , an electromagnetic field, the Y a n g - M i l l s field, and the Einstein gravitational field,* are described by singular Lagrangians. The general expressions obtained in this paper are illustrated by means of the example of the Y a n g - Mills field. The application to the g r a v i tational field will be described separately. 1. T h e
Hamiltonian
F o r m of t h e F e y n m a n
Integral
In his first paper in 1948 [1] Feynman introduced and studied the continual integral along trajectories in the configuration space of a mechanical system. A more convenient and more general form is given by * In talking about the "geometrical sense" of these fields we have in mind that they are all connections in some vector bundle (cf., for example, the general definition in [8]). Leningrad Branch of the V. A. Steklov Mathematics Institute, Academy of Sciences of the USSR. Translated from Teoreticheskaya i Matematicheskaya Fizika, V o l . 1 , N o . l , p p . 3 - 1 8 , October, 1969. Original article submitted February 5, 1969.
1
53 the expression obtained by Feynman in 1951 [9] for the continual integral in which integration is along tra jectories in phase space. That paper [9] has received little attention and subsequently the form of the con tinual integral described in it has been criticized several times (cf., for example, [10] and [11]). The Feynman integral gives an expression for the transition matrix element which is constructed in terms of the classical characteristics of the system. Let q = (q1 q11), p = (p b . . . , p ) be the canoni cal coordinates and momenta; Let H (q, p) be the Hamiltonian function. Then
The exponential function on the right side contains the classical action (cf., for example, [12]), computed for the trajectory q(t), p(t), t' ^ t £ t", joiningthe surfaces q = q' and q = q" in phase space. Unfortunately, up to the present time there has not been developed a general internal definition of in tegrals of the form (1). In the literature a limiting procedure is referred to in which the trajectories q(t) and p(t) are replaced by piecewise linear and piecewise constant functions respectively and the integral be comes finite-dimensional. This approach was described in detail in [11], where it is also possible to find a proof of the unitarity and a derivation of Schrodinger's equation. For applications to field theory, another use of the continual integral is interesting. Firstly, the whole expression for the S-matrix is of prime importance here, when the interval of time in (1) is infinite, -°° < t < », and the trajectories q(t) have a particular asymptotic form for t — ±°°. Secondly, the Hamil tonian function H (q, p) has the form
where H0 is a quadratic form and Hint is a form of degree higher than second in the field variables; g is a v
*
lilt
«=»
w
«=*
small parameter. If we expand in powers of g we obtain integrals of the form
where A , B , and C ik are constant matrices; Q is a polynomial in q(t) andp(t). These integrals are Gaussian and are taken in explicit form. The expressions which are thus obtained can naturally be described by Feynman diagrams. Thus the continual Feynman integral is a generating expression for Feynman dia grams. To conclude this section we make some observations on the canonical invariance of (1). Under the canonical transformations
the measure ndp dq is not changed, the Hamiltonian function suffers a change of variables and
Thus, the only change in the integrand in the new variables is in t at the ends of the interval. In the case of a linear canonical transformation, when <Ms a quadratic form, we can show that the change is equivalent to a unitary transformation. The author does not know how to deduce a similar property for the general canonical transformation, although it is quite possible that it exists. The difficulty is, of course, connected with the very definition of the integral (1). In field theory the interesting canonical transformations are linearized asymptotically as t — ±«°. In this sense we can assert that the representation of the elements of the S-matrix as continual integrals of the form (1) is canonically invariant.
2
54 2.
The P h a s e
Space
of a S y s t e m
with Singular
Lagrangian
We give h e r e the n e c e s s a r y details on the Hamiltonian formulation of the equations of motion for a system with singular Lagrangian. Apart from the above-mentioned fundamental paper by Dirac [2], much of the detail can be found in the p a p e r s of Bergman and his colleagues (cf., for example, [13], [14], and the literature cited therein). We shall give the results in a form which is convenient for us and clarify certain aspects, which, we hope, will make it unnecessary to refer to the original a r t i c l e s . In the case of interest to us the canonical variables q = (q1, . . . , op), p = (Pi, . . . , p n ) do not vary throughout the phase space I \ but satisfy the equations
The functions cpa (q, p) are called conssraints. It is natural to assume that they are independent and "irreducible" in the sense t h a t Eqs. (2) define a surface M of dimension 2n - m in T, an arbitrary function f vanishing on M, being a linear combination of the constraints
with coefficients cfl which a r e , in general, variable. We shall consider the case, which at first sight is special, in which the constraints
Here {f, g} is the usual Poisson bracket in T, c c a b and c , a are certain functions. In other words, we a s sume t h a t the Poisson brackets of the constraints with themselves and of the constraints with the Hamiltonian vanish on M. In the theory of gauge invariant fields constraints satisfy the enumerated conditions. We note t h a t for (4) to hold it is necessary that m does not exceed n. The equations of motion can be obtained from the variational principle with generalized action of the form
so that, apart from the canonical variables, they contain as independent unknown functions the X e (t), which act as Lagrangian multipliers. Thus, the equations of motion consist of canonical equations of the form
and conditions (2). Using (4) and (5) it is easy to see that conditions (2) hold for a r b i t r a r y functions Xa (t) if they hold for the initial conditions. In other words, trajectories beginning on M do not leave that surface, so that Eqs.(7) define a coordinate transformation on M, containing not only the Hamiltonian, but also m arbitrary functions * fl(t), a =1, . . . , m . Naturally not all the functions on M can be assumed to be observable quantities, but only those for which a r b i t r a r i n e s s in the choice of ^ f l (t) does not affect their variation in time. This requirement i s s a t i s fied by the functions f for which
Indeed, in the equations of motion for such functions
55
terms depending on Xa vanish on M. Equations (8) and (9) require clarification. The function f which occurs in them is an arbitrary contin uation in T of a function defined on M. Since the constraints (3) are irreducible, any two such continuations differ by a linear combination of constraints and, by (4), condition (8) is indepedent of the choice of the continuation. In what follows we shall make use of such a method of describing functions and equations on M several times. We can say that we are thus considering classes of functions on T as functions on M. We take in one class all functions which differ by a linear combination of constraints. The function f which is defined on M and satisfies (8) does not depend essentially on all its variables. Indeed, (8) can be considered as a set of m first-order differential equations on M with (4) serving as the integrability condition. Hence the function f is uniquely defined by its values in the submanifold of the ini tial conditions of the equations which is of dimension (2n - m) — m = 2(n—m)) As such a submanifold we can take the surface P*, defined by the equations
which are called the subsidiary conditions.
The functions xa must satisfy the condition
since only in this case can T* be the initial surface for Eqs.(8). It is convenient to assume also that the xa commute with each other
In this case we can simply introduce canonical variables on T*. Indeed, if (12) holds, then by a canonical transformation in r we can pass to new variables in which the xa ta^e tne simple form
where the p a = 1, . . . , m are part of the canonical momenta in the new system of variables. Let q a de note the corresponding conjugate coordinates and let q*, p* be the remaining canonical variables. In the new variables condition (11) is as follows:
so that Eqs. (2) can be solved for the q a .
As a result, the surface T* is defined in Tby the equations
and q* and p* act as independent variables on r*. It appears that these are canonical variables; the Poisson bracket of any functions f and g satisfying (8) can be computed as follows:
whore
and g* is defined similarly, the left side of (13) containing arbitrary continuations of f and g from M to r. To verify (13) it is convenient to compute the Poisson bracket { f, g} in the noncanonical coordinates rj = (<pa, q*. P fl .P*)- Then
4
56
where
By (4) and (8) the series of terms on the right side of (15) vanishes and as a result it coincides with the right side of (13), where
which is equivalent to (14). We emphasize again that the deduction that q* and p* are canonical is essential ly linked with condition (12). Thus, we have two methods of describing observable quantities on our system. In the first of these the observables are functions on M (more exactly, classes of functions on T), satisfying (8). The Poissonbrack et is defined as the value on M of the Poisson bracket on T. In the second method the observables are ar bitrary functions on T*. Tb pass to the second method we should choose the subsidiary conditions xato solve Eqs. (2) and (10) and construct f* according to (14) We can show that this procedure is independent ofthe choice of the conditions and the change in the X when (11) and (12) hold is reduced to a canonical transformation in r*. 0 In practice it is not simple to solve the constraints, so it is preferable to learn how to work with the first method of describing observables. On the other hand, in using the second method of description, we operate in ordinary phase space and can use the usual expressions of mechanics, in particular, Eq. (1) for the con tinual integral. Thus, to verify the validity of some expression in the first method of description of the observables, it is sufficient to verify that it passes over to the usual expression as we pass to the second method as described above. It is in this way that we begin the next section for the case of the continual in tegral. To conclude this section we note that the case in which the constraints do not satisfy condition (4) can be considered in a similar manner. In this case the phase space T* is of dimension 2(n + k - m), where 2k is the rank of the matrix {<pa, cpb} in M. Conditions (8) in this case are modified. In place of the Poisson brack et we have to use the Dirac bracket [2, 14]. Such cases obviously do not occur in examples from field theory. Discussion of the geometrical meaning of the constructions developed in this section is relegated to a special appendix. 3. The Modified Feynman
Integral
In this section we show the form of the continual Feynman integral for the case of the mechanical sys tem described in the previous section. Let such a system be defined by the canonical variables q = (q1, . . . , cp) and p = (pi, . . . , p n ), the Hamilton function H(q, p), and the links <pa(q, p) which satisfy conditions (4) and (5). We choose additional conditions X a (q,p) for which (11) and (12) hold. Then the expression for the matrix element of the S-matrix is
where the measure of the integration is defined as follows:
Here the trajectories q(t) coincide as t - ±°° with the solutions q in (t) and qoUt(t) of ^e equations describing the asymptotic motion and are uniquely defined by the states |in > and < out| respectively and the linearized conditions.
57 To prove this we reduce the expression to an integral of the form (1) in which integration is along trajectories in the phase space F*. To do this we pass to the description in the second section using the coordinates qa, q*, pflI and p*. The integral (16), to within nonessential boundary terms, referred to at the end of the first section, has a form similar to (16), but with a different measure:
which, obviously, can be rewritten as
Integration with respect to qa and pa is eliminated by the delta functions. As a result, the integral takes the form >
which coincides literally with (1), and we can consider that we have proved the validity of (16). We note now that the integral in this equation can be rewritten
since it is obvious that integration with respect to X can be carried out explicitly and (18) becomes of the form of (16). In (18) we integrate the functional exp{(i/h) s}, where S is the generalized action (6) with respect to some measure in the space of the trajectories q(t), p(t), X (t) lying on the surface Xa(q, P) = 0. The explicit expression for the measure in (18) is the fundamental result of the paper. In the next section we shall see that the general expression (18) can be particularized to the example of the Yang-Mills field. To conclude this section we show that the integral (16) is independent of the choice of the subsidiary conditions Xa (q, P) . Let 6 x^ be an infinitely small change in these conditions. To within a linear com bination of the links we can represent 6 Xa as the result of an infinitesimal canonical transformation in T, the generator of which is also a linear combination of the constraints. Indeed,
where as h a we can take the solution of the system of equations
which, from (11), has a unique solution. In the canonical transformation the constraints are replaced by linear combinations of themselves
and, thus, the quantities in (16) are changed in the following manner:
58 We use here an obvious abbreviated notation and omit linear combinations of the constraints which vanish on integration. As a result of the canonical transformation the integral takes the same form as (16), but with x replaced by x + 8 X, which proves that it is independent of the choice of %. 4. T h e Y a n g - M i l l s
Field
The simplest nontrivial application of the continual integral introduced in the third section is the quan tization of the Yang-Mills field [15]. We shall show how to obtain from (18) an expression for the modified Feynman integral for this field, as previously obtained by V. N. Popov and the author [16] from different considerations. As shown in [17], from this expression there follows a modified diagram technique in per turbation theory, the necessity of which was first remarked by Feynman [18] (for other conjectures on this subject see [19-21]). The Y a n g - Mills vector field is constructed for any compact group G (cf. [22] and also [8])) Let t^ be generators of the group, normed by the condition tr (t^to) = - 2 6 ^ ; l<*py the corresponding structural constants which are antisymmetrical in all three indices. The Y a n g - Mills field is conveniently described by the vectorf A^ and the antisymmetric tensor F^, where these quantities are matrices in the Lie algebra of the group G. We can put
where A® and F below.
tt
form a vector in the adjoint representaiion of G. Both notations are convenient
The Lagrange function has the form
and does not contain the time derivatives of the variables A,, and F i k . The latter occur in L quadratically and can be eliminated by putting
After the substitution (19), the action takes the form
where we have introduced the notation
and C is given by the expression
Comparing (20) and (6), we see that E^ and A^ act as canonical variables, Ao is the Lagrangian multiplier and C is a constraint. If we introduce the fundamental Poisson bracket
it is easy to verify that equations of the type of (4) and (5) hold:
and
tin this section we put h = 1, c = 1 and use the normal relativistic notation
59 Further, the equation
makes it possible to express the longitudinal part of the vector E^ in terms of the vector A^ and the transverse part of Ek, and thus reduce the dimensionality of the phase space (six functions of three variables) by one such function. Indeed, if we put
we can rewrite (21) as
which can be solved uniquely for
».
Thus, all the techniques developed in the second and third sections can be applied to the system under consideration. For the subsidiary condition we naturally take
since the corresponding Poisson bracket
coincides with the kernel introduced in (22) of the operator M, which has a unique inverse. With such a choice of the subsidiary condition, the continual integral (18) takes the form
Here S is the action (20) and by det' M we denote the regularized determinant of the operator
which only differs from det M by an uninteresting constant (and infinite) factor. The continual integral (23) is a generating expression for Feynman diagrams in the Coulomb gauge. Its essential disadvantage is that it is not relativistically covariant. However, we can rewrite the integral in explicit covariant form. To do this we integrate firstly with respect to E^, having taken the Gaussian integral. Then we obtain an integral of the form
where S[A] is the covariant action of the Y a n g - Mills field, expressed in terms of the potentials A^
Only the factor
defining the measure of the integration, is not covariant in (24). It turns out that the value of the integral (24) is not changed if in place of this factor we use the covariant expression
where N is the four-dimensional generalization of the operator M
8
60
and det' again denotes the regularized determinant.
As a result (24) can be rewritten as
The diagram technique, occurring in the computation of this integral in perturbation theory, has been d e scribed in detail in [17]. Its characteristic feature is the phenomenon of additional diagrams generated by the expansion of In det' N. Other approaches to the problem of the quantization of the Y a n g - Mills field, developed by De Witt [19], Mandelstam [20], and Veltman [21], have led to similar r e s u l t s . We note that, as De Witt [19] showed, perturbation theory is based on the integral
with a r b i t r a r y P * 0. In diagrams generated by this integral the a r b i t r a r y p a r t of the propagator of the vector field is (1//3) - 1 , while in computing (28) we can use only the Landau gauge. To prove that it is possible to replace (26) with (27), it is pertinent to clarify its geometrical meaning. The action (25) is invariant with respect to the gauge transformation
where J2(x) is an a r b i t r a r y matrix function of G. We can say that S[A] is a function on the manifold a of classes of the fields A„, where fields which differ only by a gauge transformation form one c l a s s . The in tegral (24) can be considered as an integral with respect to this manifold, since the condition
selects one representative in each c l a s s . The condition
has a similar property, so that we can look on (29) and (30) as two different means of uniquely p a r a m e t e r izing the manifold a of c l a s s e s . We note now that every measure on a is obtained from a gauge invariant m e a s u r e on the set of all fields A„. These m e a s u r e s have the form
where m[A] is a gauge invariant function. If the parameterization of a is defined by the equation f[A] = 0, the corresponding measure on a is given by the expression
where Af[A] is obtained by averaging over the group
and dft is an invariant measure on G. We can verify that (26) and (27) a r e obtained if we apply this recipe to an invariant measure of the form (31) with m = 1 for the c a s e s f = ^Afc and f = 9 A „ , respectively. To do this, repeating the considerations in [17], we note that (32) contains values of Af[A]'only for fields A„ satisfying the equation of f[A] = 0 . For such fields every contribution of (33) gives a neighborhood of a unique
61
element of G, so that the integral can be rewritten as
where we have put ft = 1 + u, and u is a matrix of the Lie algebra of G, and we have linearized the argument of the delta-function. If we apply this expression to the case of conditions (29) and (30), we a r r i v e at (26) and (27), respectively, thus proving the equivalence of the integrals (24) and (28). APPENDIX This appendix is intended for r e a d e r s who a r e disappointed by the superficial noninvariance of the description of the second section. We shall clarify here the geometrical meaning of the conditions and con siderations formulated t h e r e . At the same time we shall sharpen the conditions under which these con siderations a r e actually valid. We shall, naturally, use the invariant language of differential geometry (cf., forexample, [8]), which is the most adequate for general problems in mechanics (cf., for example, [23, 24]). The phase space of a mechanical system with n degrees of freedom is a smooth manifold T of 2n d i mensions with a nondegenerate closed differential 2-form S2
Using this form we can introduce the s t r u c t u r e of the Lie algebra into the algebra of functions a on T, i . e . , we can define the antisymmetric operation {f, g}, f <= a, g e a, satisfying Jacobi's identity and called the Poisson bracket. We find the following variant of the definition convenient. For any function f, the equation
uniquely defines the vector field Xf.
The Poisson bracket is defined as
and, obviously, is linear and a n t i s y m m e t r i c .
Jacobi's identity is a corollary of the closed nature of ft.
Thus, in the linear space of functions on the phase space, we have, apart from the usual product fg, one additional binary operation {f, g}. The second operation is differentiation with respect to the first
The algebra a of functions on I \ with the operations described, is an algebra of observable (quantities) of the mechanical s y s t e m under consideration. In local coordinates | = ( | \ . . . , £ 2n ) on T the form ft is specified by the antisymmetric nonsingular matrix Qap, satisfying the condition
The vector field Xf has the following components in a natural b a s i s :
where fta0 is the inverse of Qap.
The Poisson bracket is specified by the equation
[compare with (15)] and has the usual form
10
62_
in the canonical variables £ = (q, p), q = (q1
q11), p = (pj, . . . , p n ), in which £2 is given by
The fundamental object introduced in the second section is the phase space r * . Its description involves: 1) the phase space r with canonical coordinates (q, p) and canonical form Q; 2) )cnstraints
Let Q = ft | M be the value of Q on M. The form Q is closed, but singular. We can show that, by (4), its rank is 2(n - m). It is not l e s s than this number, since the rank of a regular matrix cannot be lowered by more than 2m when it is r e s t r i c t e d to a submanifold of codimensionality m. On the other hand, we can select m linearly independent zero vectors from this form. We note that vector fields tangential to M are restrictions to M of vector fields Y on r satisfying
which, by (A. 1), we can rewrite as
By (4), we have
and, comparing this with (A.3), we see that the vectors X a are tangential to M and a r e zero vectors of the form S, which proves our assertion. Every singular closed 2-form Q, defined on some manifold, generates its foliation. Layers a r e maximal integral manifolds of an involutive distribution (cf., forexample, [25]), formedbythe zero vectorsof a f o r m . The set of such zero vectors is denoted by P and we shall show that since X E P a n d Y e p , we have [X, Y] e P . We have
and, by the condition on X and Y, the single t e r m which does not vanish by definition is the first t e r m on the right hand side. Thus, it also vanishes and this proves that P is involutive. We note that in the case under consideration, we have described explicitly the basis for the zero space of the form Q and the involutive nature of this distribution can be verified directly using (A.4). We now assume that foliation is indeed a fibration and let T* be its base. The form Q* of the inverse image of £2 on T* is closed and nondegenerate. Thus, r* is a phase space which is the object we look for The condition, formulated in this paragraph, for the existence of a fibration imposes very strong restrictions on the constraints
can be con
which coincides with (8) and at the same time clarifies that condition. We now give an alternative algebraic definition of the algebra r . Let a be the algebra of observables on T with operations f + g, fg, and {f, g}. We denote by
63 that
and which can be called the normalizator of * in a with respect to the Lie operation. The existence of such a nontrivial (distinct from $) set is a strong condition on the constraints <pa, ccmparable with the abovee mentioned condition on fibration. The set a" is a subalgebra of the algebra 9 . Indeed, if / e a and J e= « , then, by (A.2), and the Jacobi identity
F u r t h e r , $ is, by definition, an ideal in 9 . The algebra a* = 8/d> is one further realization of the a l gebra of observables of the system under consideration. In the actual description of the space T* in the second section we introduced the subsidiary condition (10). Equations (10) define a submanifold T* of the space M and condition (11) implies that r * is a t r a n s versal integral manifold of the distribution P . We impose on cpa and Xb ^e condition that each integral manifold i n t e r s e c t s T* at only one point. In this case we have actually constructed a fibration and r * can be used as the realization of its base. r*.
Condition (12) has only technical value and simply makes it e a s i e r to choose canonical coordinates on Indeed, if it holds, then in some canonical coordinates on Tthe manifold F* is defined by the equation
The inverse image of the form in r * .
so that the variables q* and p* are canonical
In conclusion we wish to e x p r e s s our gratitude to V. I. Arnol'd, with whom discussions of the p r o b lems of nonholonomic mechanics were of great value for this appendix. Responsibility for all noninvariant expressions not eliminated from the text r e s t s entirely with the author. LITERATURE
CITED
1. 2.
R. Feynman, Rev. Mod. P h y s . , 20, 367 (1948). P . A. M. Dirac, Can. J . Math., 2 , 129(1950).
3. 4. 5.
P. A. M. Dirac, Proc. Roy. Soc, A246, 326(1958). P. A. M. Dirac, Proc. Roy. Soc, A246, 333 (1958). P. A. M. Dirac, Phys. Rev., 114, 924(1959))
6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.
12
P . A. M. Dirac, Report at the symposium on Modern Physics at T r i e s t e (June, 1968). J. Klauder, T h e s i s at the Fifth International Conference on Gravitation and Relativity, Tbilisi (1968). A. Likhnerovich, The Theory of Cohesion and Holonomy Groups [in Russian], IL (1960). R. Feynman, P h y s . R e v . , 84, 108(1951). W. Tobocman, Nuovo C i m . , 3 , 1213(1956). C. Garrod, Rev. Mod. P h y s . , 3 8 , 483 (1966). L. D. Landau and E. M. Livshits, Mechanics [in Russian], Fizmatgiz (1958). P . Bergmann, Rev. Mod. P h y s . , 33, 510(1961). P . Bergmann and I. Goldberg, P h y s . R e v . , 98, 531 (1955). C. S. Yang and R. L. Mills, P h y s . R e v . , 96, 191 (1954). L. Faddeev and V. Popov, P h y s . L e t t . , 25B, 29 (1967). V. N. Popov and L. D. Faddeev, "Perturbation theory for gauge invariant f i e l d s , " P r e p r i n t of ITF, AN UkrSSR, Kiev (1967). R. Feynman, Acta P h y s . Polonica, 24, 697(1963). B . S. DeWitt, P h y s . R e v . , 162, 1195, 1239(1967). S. Mandelstam, P h y s . R e v . , 175, 1604(1968).
64 21. 22. 23. 24. 25.
M. Veltman, Nucl. P h y s . , B7, 637 (1968). S. L. Glashow and M. Gell-Mann, Ann. Phys. , 15, 437 (1961). J. Schwinger, P h y s . Rev., 125, 1043(1962). V. I. Arnol'd, Lectures 30-42, Lectures on Classical Mechanics [in Russian], Izd. MGU (1968). R. Jost, Rev. Mod. P h y s . , 36, 572(1964). R. S. P a l a i s , Mem. A m e r . Math. S o c , 22 (1957).
13
65
Covariant quantization of the gravitational field L D. Faddeev and V. N. Popov Leningrad Section of the V. A. Steklov Mathemaiical Usp. Fiz. Nauk 111, 4 2 7 ^ 5 0 (November 1973)
Institute.
USSR Academy of Sciences
A review dedicated to the contemporary methods of quantization of the gravitational field. In view of possible applications to elementary particle theory, the authors consider only asymptotically flat gravitational fields. The basis of the exposed method of quantization is the method of quantization of gauge fields in the functional integration formalism. The main result is the formulation of covariant rules for a diagrammatic perturbation theory. Its elements are the lines representing gravitons and the vertices of graviton-graviton interaction, as well as the lines and interaction vertices of fictitious vector particles ("Faddeev-Popov ghosts") characteristic for the theory of gauge fields. The expressions for the propagators and vertex functions are given explicitly. It is shown that the presence of fictitious particles in the covariant diagram technique guarantees the unitanty of the theory and the agreement between the covariant quantization with the canonical quantization. The bibliography contains 44 entries (54 names).
CONTENTS Introduction 1. The Feynman Rules for Covariant Perturbation Theory and the Feynman Functional Integral 2. Feynman Rules for the Gravitational Fields 3. A Derivation of the Modified Perturbation Theory Rules 4. The Hamiltonian Form of Gravitation Theory 5. The Hamiltonian Form of the Functional Integral References
777 778 779 782 784 786 788
INTRODUCTION The interest in a quantum theory of gravitation is maintamed to a large extent by hopes that inclusion of gravitation into the scheme of quantum field theory will allow us to construct a self-consistent closed theory of elementary particles. In this connection there are two directions. One is related to cosmology and to the use of cosmological considerations in the theory of elementary particles (cf. the papers of Wheeler- l ] M i s n e r ^ , Markov and collaborators « ) . The other direction, which does not make use of cosmological considerations, considers the field of the particles concentrated effectively in a finite volume and vanishing at spacelike infinity. In the latter approach the gravitational field is considered on the same footing as any other field ( c f . ^ ) and the theory is a variant of the theory of gauge fields. The role of gauge transformations is played by coordinate transformations which do not affect the spacelike infinity, and the role of the gauge group is played by the Poincare group (in more detail this problem is discussed in the report by one of the a u t h o r s c 0 ) . The role of Lorentz-invarince in the general theory of relativity was underlined by Fock C 6-. .... . „■1 The interest in elementary particle theory is mainly , ' , 1 * u uTTTI v • *«. - •* <■„ 1 related to the hope that it may be just the gravitational . . , , , . , , / , .. 1 . x , 11 u 1 field which could play the role of a natural "physical r e g u l a r i z e d which removes the singularities and infinit i e sfrom quantum field theory. The first results which supported this point of view were obtained by De W i t t " and K h r i p l o v i c h ^ . At the present time this direction is actively developed by Salam and collaborators 1 1 9 ] , making use of methods developed by Efimov, Fradkin, Volkov and Filippovt — ] .
In this review article we restrict our attention only to the problem of constructing covariant rules for a diagrammatic perturbation theory. The most convenient tool for m i s purpose is Feynman's functional integral. * this formalism the physical degrees of freedom, which are quantized, and the gauge-field degrees of freedom, w n ich remain c-numbers, are treated on the same footinS. This makes it possible to use a large class of different transformations. The first correct formalism for the quantization of the gravitational field was constructed by Dirac in 1958 C»3, within the framework of the Hamiltonian m e thod. Other methods, essentially equivalent to that de veloped by Dirac have been developed by the ArnowittDeser-Misner groupE* 3 , by S c h w i n g e r ^ and also some others C l T - l 9 ] . The absence of explicit covariance in the Hamiltonian formalism makes perturbation theory very cumbersome. It is appropriate to recall the analogous situation in quantum electrodynamics. The noncovariant perturbation methods of the thirties did not allow one to g 0 beyond the first nonvanishing approximation of perfurbation theory. The creation of a covariant perturbation theory toward the end of the forties and the development of the diagram technique associated to it have sim„. , , , ^ ' , ., . , ,, 1. ^ . pilied the calculations considerably, so that at present f " \ ". * the theory can be compared to experiment up to the c / . *"~ *" *: seventh f c ^ ^ce' to the the0^ of gravitation the tech i u cal A c u l t i e s a r e so great that a noncovariant quantization scheme becomes practically useless in concrete perturbation-theoretic computations. The first attempts of constructing a covariant quantization scheme for the gravitational field were contamed in the papers of G u p t a ^ . He followed the scheme
777
Copyright © 1975 American Institute of Physics
Sov. Phys.-Usp., Vol. 16, No. 6, May-June 1974
777
66_ developed for quantum electrodynamics, and to overcome the difficulties related to the singularity of the free Lagrangian of the gravitational field, he made use of a trick analogous to the well-known Fermi method of quantization for the electromagnetic field (cf., e . g . / [ " ' 2 2 ] , .
Here are some details about the plan of our review article. Chapter 1 deals with the scheme of derivation of rules of covariant perturbation theory in the functional integration formalism, on the example of a scalar selfinteracting field theory. The Feynman rules for the gravitational field are formulated in Chap. 2. The deri-
It turned out, however, that an uncritical transfer of the Fermi quantization method (winch is justified in quantum electrodynamics to more complicated systems can lead to a violation of the unitarity condition of the theory. This was first discovered by Feynman in I M * * on the examples of the Yang-Mills field and gravitation theory. Feynman has mapped out the path to overcome the difficulties which he had pointed out. He has shown that the unitarity of a closed-loop diagram can be recovered if one subtracts from the appropriate matrix element the contribution of another diagram, also having the form of a loop and describing the propagation of a fietitious particle It was not possible to extend the Feynman method to more complicated diagrams. The solution of the problem for arbitrary diagrams was found in 1967 by De Witt [24 ] and by the authors of the present review-" 1 , using basically different m e thods. Both methods are unified by the use of the method of functional integration, which provides a scheme of covariant perturbation theory for gauge fields. In the case of a non-Abelian gauge group the method involves lines corresponding to fictitious particles and vertices describing their interactions with the quanta of the gauge
^X^lio^te^lorZ gravSnaT ££ poTJjustification for this is the possibility of f ^ ^ J u Hamiltcnianform. _ ... B . ,. „. . .. t. „ „, d l°r ^ S rea S f °n w e / * a I £ C ^ t a ^ o ' T T i the 1 ^ f ffh ^ n t " \ ^ h a t thafoni f nal ' t ^al e u n c 10 i n eg 7 ? . ' ' CJia Ptf r ^ s o w /^ ™ £ ™ * ^ useTin m e ^ r i v a t i o n of the " '"'""■**' "*»
duced by Feynman in order to recover the unitarity of the one-loop diagram for the real particles. The sign (—1) in front of the contribution of the closed IOOD shows ihat the fictitious particles (which are also known under the name of "Faddeev-Popov ghosts"-transl.) are subject to Fermi statistics. This makes a direct interpretation of the fictitious particles difficult, since they have integral spin (they are scalar for the Yang-Mills field and vectorjal for the gravitational field). The results obtained in*- ' s ] have been repeatedly rederived by v a r M a n d e l s W 2 9 } , Veltman^J Fradkin and T y u t i n ^ , K h r i p i o v i c h ^ , Boulware [30J , 't Hooft1 1 J , Altukhov and KH-^l«tf^ht3»3 ^"P10™11 ' The construction of a correct covariant quantization scheme for gauge fields has generated a series of papers where the (massless) gauge theory is compared to the corresponding massive (non-gauge) theory and the problems of the limit m - 0 are discussed (the papers of F a d d e e v a n d S l a v n o v ^ , Valnshteih and K h r i p l o v i i c ^ 1 , Z a k h a r o v ^ 3 , V e l t m a n ^ , Fradkin and T y u t i n ^ . The basic result of the majority of these papers reduces to the fact that the limit of a massive theory for m - 0 does not, in general, coincide with the corresponding massless theory As we already noted, methods of quantization of the gravitational field have been developed within the framework of the canonical HamUtonian formalism and in ex plicitly covariant form. The functional integration m e thod allows one to relate both formulations. The canonical Hamiltonian yields a unitary normalized theory and the covariant formulation allows one to construct a perturbation theory which is convenient for concrete calculations. 778
Sov. Phys.Usp., Vol. 16, No. 6, May-June 1974
+ T u c FFYMAN RUL ES FOR COVARIANT , „ „ „ . , . , - . , , - , , , - r t n v AMI-* - r u r CCVMIWIAM S E D ^ J ° » ™ ™ ,Y A N ° ™ E F E Y N M A N FUNCTIONAL INTEGRAL w « " c a U and discuss here the recipe for the derivation of the rules of covariant perturbation theory according to F e y n m a n ^ . We start with the simplest example <* a self-interacting scalar field
n'rnxl
The diagrams corresponding to the t e r m s of perturbation expansions is constructed out of two elements: the ^ G (propagator) and the vertex V: t
^
''
/
•*%. *
' "'
The scattering amplitude M n (ki, ..., k ) describing n (incoming or outgoing) particles with momenta kt, ..., k n j,s represented by a sum of contributions corresponding to diagrams with external (non-closed) lines. The contribution of a given diagram is obtained by associating to its elements the following expressions: to internal C(fc„ fcj- ' j f f i ' ,
(1.3)
:o the vertices one associates the vertex function (coupling constant) v
= u <*' + *« + * * P-*l ^ && integrating over all the momenta k of the internal elements of the diagram, the final result must be multiplied by ,p , ( L 5 ) H w ] ■ where I is the number of internal lines, v is the number of vertices and r is the order of the symmetry group of ^e diagram1'. To obtain the amplitude for a real process one has to go onto the energy shell k° = ± (k2 +■ m2)l/2; ^ sign depends on whether the particle is incoming or outgoing The described elements of the diagram technique are determined by the Lagrangian (1.1) in the following manner. We consider the action s y m f £ ( l } 9m (1 6) ' * as a functional of the Fourier transform 7j> (k) of fhe eield L. D. Faddeev and V. N. Popov
778
J>7 (p(x). This functional consists of a quadratic form in the field and a form of third degree: The function G(k t , k2) is the kernel of the integral operator inverse to the operator which defines the quad ratic functional form S 2 ; the function V is the coefficient function determining the form of third degree. This correspondence can also be seen in many other examples for which there exists a diagram technique, e.g., for the pseudoscalar theory of pions and nucleons. The only difference is the appearance of the factor ( - l ) s , in the weight (1.5) of the diagram, where s is the number of closed loops formed of fermion lines.
where the integrands have the form expfi(quadratic form)] x polynomial. Such integrals can be explicitly calculated and can be expressed in terms of the operator which is the inverse of the operator of the quadratic form as well as the co efficients of the polynomial. Indeed, let us consider an integral of this type for a finite number of variables:
It can be calculated by differentiating the generating function
The described recipe is not directly applicable to theories where the corresponding quadratic form is singular. The simplest example is provided by electro dynamics, where the action integral of the free photons
namely:
is degenerate owing to gauge invariance, and does not depend on the longitudinal component 9 AU of the poten tial A„. However, in this case it is known (cff C21 ' 22] )
The generating function can be computed by a shift of the variable, yielding
that one can u s e a«? photon p r n n a g a t o r a crpnera MzPH
inverse operator of the quadratic form (1 8V the momenturn-space representation of this propagator is the fol where the matrix B is the inverse of the matrix A. The first factor here plays the role of the normalizing con lowing stant which we have agreed to neglect. The differentia tion of the second factor leads to a result which in words can be formulated in the following manner: The integral where the constant is arbitrary. The scattering ampli J is represented by the sum of the coefficients c ^ . . . i n , tudes for real processes do not depend on this constant. contracted with products of the elements B^- of the ma However, as we have already pointed out in the intro trix B, carrying the same set of indices. duction, a direct adaptation of this recipe to theories with In our infinite-dimensional case the analogues of the nonabelian gauge groups leads to incorrect results. In coefficients c t >#>j are products of the coefficientorder to make clear the reason for these difficulties it is functions of the cubic form in the action. The role of useful to analyze the derivation of the rules of covariant the matrix A is played by the hyperbolic differential perturbation theory. In our opinion the most convenient operator (the Klein-Gordon operator) (D -m 2 2) or, in approach is the formalism of the Feynman path (func momentum space, the multiplication operator by tional) integral 2 '. (k2 - m22) A natural geneealization of our rinite-dimen We recall the main features of this method on the ex sional result leads then to the Feynman rules formulated ample of the scalar field with the Lagrangian (1.1). The above. scattering amplitude is obtained as an integral over all Two remarks are in place here. The first refers to possible fields with a given asymptotic behavior for the definition of the inverse of the operator A, i.e., of t — ± « of the functional exp(iS), where S is the action the Green's function of the operator ( □ - m22) There integral. We shall use the following notation for such are many such Green's functions: the retarded one, the integrals advanced one, the causal one, etc. The Feynman rules require the use of the causal Green's function. One can justify this within the framework of the functional integ where the symbol rxd
2. FEYNMAN RULES FOR THE GRAVITATIONAL FIELD
Expansion of the functional exp(iS) in powers of the coupling constant A leads to integrals of the form
The peculiarities of the gravitational field are rela ted mainly to its self-interaction. Therefore the major ity of this chapter will deal with the "free" self-interact ing gravitational field. The main result, namely the dia gram technique, is listed at the end of the chapter. We
779
L. D. Faddeev and V. N. Popov
Sov. Phys.-Usp., Vol. 16. No. 6, May-June 1974
779
^ _ _ _
68 also indicate there the changes made necessary by the presence of a matter field or the electromagnetic field. Among the most frequently used parametrizations of the gravitational field the following two are the most im portant: the metric tensor and the moving-frame or tetrad (Vierbein) formalism. We summarize both of these. In the metric tensor formalism the gravitational field is described by the potentials gM„(x) and the Christoffel symbols r £ (x). The latter can be considered as inde pendent dynamical variables (the Paiatini formalism) or as functions of the g u „ :
(frame) e^ a (x) and the torsion coefficients u^ ^ ( x ) = _ u D a (x). The set of e^ a (x) form a mariix with positive determinant e(x). The action functional
The contravariant g ^ - m a t r i x is the inverse of g^, denotes the determinant of the matrix g .
A variation of S with respect to w leads to equations which allow us to express w in terms of e. The solution is conveniently written in the form
is invariant with respect to coordinate transformations
and with respect to local Lorentz transformations
g
In this review we shall only consider asymptotically flat gravitational fields. In this case the space-time manifold is topologicaily equivalent to four-dimensional Euclidean space and can be parametrized by global c o ordinates x^ ( - » <xM < +», ^ = 1, 2, 3). These coor dinates shall be compatible with the conditions at spacelike infinity such that
where If necessary, one may assume that this is already done ahead of time, so that S may be considered a s a functional only of the functions e ^ a .
where r = ((x1)' + (x2)2 + (x 3 ) 2 ) l/2 , and TI is the metric tensor of Minkowski space, with the signature (+ ). ~^ ^ ., ^. The action functional has the form
We shall talk about a formalism of the first order if the variables g ^ and r p (or e^a and w ^ ) are considered as indeDendent If the r are exore'ssed in terms o u c l p K ** ** . * , ,, " „ , of the g, and the e in terms of the w we shall talk about a formalism of the second order. The descriptions of the free gravitational field in terms of the g„,, or the e ^ a are equivalent. The differ*££ of components—10 m m e first ^ s e c o n d - i s compensated by the difference £ ^ g^Q ^ ^ which ^ ^ ^ ^ ^ ^ trized bv four functions and in the second case bv ten Thetetrad formalism Lconvenient for the description of interactions with spinor fields. The equivalence between first- and second-order formalisms may disappear when the interaction with other fields is switched on. Geometrically Eq. (2.11) defines a connection without torsion. The minimal inter action of the gravitational field with the spinor field in a first-order formalism leads to the appearance of torsion (cf C3al)
where * is the Newton constant is invariant under the group of coordinate transformations acting on the quantities et»>, r j „ according to the rules
We have written here the equations for infinitesimal transformations; r, fi are the infinitesimal components of a vector field which generates the coordinate transformations
Variation of the action (2.3) with respect to the r ^ leads to equations, the solutions of which are the func tions (2.1). In this sense one may consider the r ^ „ as The remainder of the exposition of this chapter will independent variables, which is sometimes convenient to be ^ ^ on *» ^ ^ ^ <* a tensor formalism of the do. ^ second order. We set Substituting into the expression (2.3) the explicit form (2.1) of the Christoffel symbols Tp „ in terms of the metric tensor, it becomes
where for convenience we have introduced the contravariant tensor d e n s i t v d e g r
and consider uM » a tensor field describing the gravitational field. The action functional (2.6) takes on the form
In the tetrad formalism (also known as the "movingframe formalism" or "Vierbein formalism") the gravitational field is described by the components of the tetrad
where S.j « " quadratic form and S n is a form of n-th e in the variables M^V and their first derivatives. The linearization (2.12) is in many respects not a natural one. It can violate the signature of the metric tensor if u ^ " is not sufficiently small. Recently exponential parametrizations have become popular, e.g., for the tetrad matrix e^a = expflcx"*). In principle the expansion (2.13) can be computed in this parametrization. We
780
L. D. Faddeev and V. N. Popov
Sov. Phys.-Usp., Vol. 16, No. 6, May-June 1974
e
780
69 note that the quadratic form S2 does not depend on the parametrization. A Hir-^t a«r»n„„+,-„ «* «. • « ^u * i * A direct application of the recipe from Chapter 1 to the case of gravitation would lead to the following formulation of covariant perturbation theory rules: t t a q u a d ratic form S2 defines the propagator and the coefficientfunctions of the forms S give the expressions for the vertices, which in this case are S t e i n number. We note here that in a first-order formalism the linearization (2.12) and the substitution transforms the action into a sum of forms of second and third degrees, so that in this formalism the number of vertices is finite. As a consequence of the invariance of the action with respect to transformations (2.4) the quadratic form
In addition to these elements, in the internal parts of the diagrams one must make use of additional elements which can be interpreted in terms of vector particles i . „ .. ... * .. _ . . . £.. . . nte/ac'inS w l t * gravitons. Such an interpretation is int r ° f C,ed
is degenerate. It does not contain the longitudinal com ponents a uM". The example of electrodynamics suggests _ . , , . . the idea to use as propagator for the gravitons a generfh l ct l t l °us e,^ments p a r f c l Pate only m ,closed loops **» o alized inverse operator, e.g., the one-parameter family t e external hnes are always graviton lines. The contribution from a given diagram is obtained if the product of expressions of the form (2.16)-(2.18), (2.20) associated to its elements is integrated over the internal momenta, and the result is multiplied by with the elements of the S-matrix independent of the choice of the constant a. However, FeynmanC23] has shown that the matrix elements calculated according to the naive rules depend in an essential way on this constant, and the unitarity condition is violated. Feynman has also outlined the way out of this difficulty. As a result of the efforts of a large number of authors, as described in the introduction, correct rules for the perturbation theory have been obtained, rules which we describe here. Their derivation will be given in the following chapters. s * * The diagram technique contains, to be sure, all the elements of the "naive approach": graviton lines, for which the expressions have the form (2.16) and vertices generated by the forms S n + o of expansions of the type (2.13). Here is the explicit depression for the thirdorder vertex, corresponding to the linearization (2.12):
where s is the number of closed loops formed by fictitious particles. A comparison of this equation with (2.5) shows that the fictitious particles act like fermions, i.e., they violate the spin-statistics theorem. This shows that their role reduces to a subtraction of the contributions from unphysical degrees of freedom. * addition to * * described diagrams perturbation theory ^volves M i n i t e contributions of the renormalization type, contributions which a r e proportional to powers of the delta-function 6 (4,(0). The structure of these terms will be described below in Chap. 3. In the firstorder formalism the elements associated to ffctitious particles do not change. In addition to the tensor propagator (uu) the perturbation theory also involves the propagators (uy) and
where
The only graviton vertex is generated by the trilinear form
+ the sum over permutations of the pairs
781
Sov. Phys.-Usp., Vol. 16, No. 6, May-Juna 1974
L. D. Faddeev and V. N. Popov
781
70
We do not write out the elements of the diagram technique in the tetrad formalism. The reader can obtain them by himself using the methods described below. The inclusion of matter fields does not lead to the appearance of new fictitious particles and the corresponding elements of the diagram technique, as long as the quadratic forms in the action functionals of the matter fields are nondegenerate. As an example we list the interaction Lagrangians of the gravitational field with a scalar field and with a spinor field. In the first case it is convenient to use the metric formalism and in the second case one must use the tetrad formalism:
point in M and a a group element, ? a denotes the action of the group element a on the point 4. Consider the quo tient manifold M/G = M*, formed by the classes of all points of the form £ a , where £ (the representative of the class) is fixed and a runs over the whole group ( | a is also called the orbit of the point { and M» *i she erbitspace of G—Transl.). measure on m e orbit M* can be actea. ded £ view of £ b consUnt on the classe S ) to a meisure M on M whichgis m v a n a n t under the group action. Conversely, given an invariant measure u on M it is not hard to construct a measure M* on M*, which extends to p in the sense indicated. One can make the selection of representatives from the classes concrete by defining in M a hypersurface which intersects each orbit (class) once. This means that if the hypersurface is defined by the equation then for given £ the system of equations
here we have utilized the standard notations for the components of the scalar and spinor fields; y a and crcd = V,(rcrd ~ r V ) are the usual Dirac matrices. We note that if there is no mass term it is easy to choose the parametrization of the gravitational field in such a manner that these functionals generate only a finite num ber of vertices.
must have a unique solution a (depending in general on O. « one uses such a parametrization the measure M* looks as follows: where the function Af(4) is defined by the relation
3. A DERIVATION OF THE MODIFIED PERTURBATION THEORY RULES Functional integration is a convenient heuristic means for the explanation and heuristic derivation of the rules of perturbation theory for the gravitational field, rules which have been enumerated in the preceding chapter. Within the framework of this approach the additional terms mentioned there are interpreted as a consequence of the nontriviality of the measure with respect to which the functional exp(iS) is being integrated. . . ... . . . ., ,. . Let us explain this in more detail, making use of natural geometric terms. The functional exp(iS) is in variant with respect to the infinite group of coordinate transformations in the metric formalism or with respect to the semidirect product of this transformation group and the group of local Lorentz transformations in the moving frame (tetrad) formalism Thus this functional is ^function on (equivalence) classes of fields where
and da is the invariant measure on the group G. The function Aj(£) is invariant, i.e., Aj(£ a ) = Af(?)- Equation (3.3) can be explained in the following way. We go over to the new variables where a is the group element defined by Eq. (3.2) and ? T = ? a . Let the invariant measure on M have the following expression in terms of the coordinates £: & ' , ... the new variables "
m
^
where
^ S r a ^ i n t o ^ ^ S r t o n S r m a l o n s from th dicated erouD ^ ^ 6 m S V ' We shall assume that it is a class rather than an individual field which describes a concrete physical situation. This is the content of the principle of general co„,„io„„ Q ir„ar.in,r =,„-h , f n , m l n ™ , nt th it ^^„^ir>i Q v a r i a n t . Keeping such ^formulation of tois principle in mind we can consider that m quantum theory the Feynman functional exp(iS) should be integrated with respect to such classes of fields, rather than individual fields. The nontriviality of the measure which we have mentioned is related just to this circumstance. Let us discuss how one can describe measures on the set of classes of fields. Following our method, we first consider the finite-dimensional case. Mathematically we are dealing with the following situation: we are given a manifold M (in the sequel this will be the set of all fields) and a group G (in the sequel this will be the group of gauge transformations) acting on M. Let E, be a
is * " measure Senerated by *e measure d ? on the hypersurface, Df is the Jacobian of the transformation to the variables ( | T , a). The measure on the orbit space M* is obtained from (3.7) by omitting the invariant measure da in the group coordinates. „, „ . . . . . . . We show that the Jacobian Df coincides with the in f J . I( ) ( ) consider the integral
782
L. D. Faddeev and V. N. Popov
Sov. Phys.-Usp., Vol. 16, No. 6, May-June 1974
of the arbitrary invariant function *(«), integrated with tresPect to " » measure M^d^ The V a r i a n c e condi-
allows one to assume that * depends only on £ T . Indeed, since ?T = 5 , 782
71 Making use of the expression (3.7) for the measure M(£)d£ in the integral (3.9), we transform the integral to the f o r m X
and thus is expressed in terms of the integral of the 5-functional n 5 ( 6 y ( h ^ ) a - |M) over the gauge group. , |x
) where M(G) = Jda is the "volume of the group." Another expression for the integral (3.9) can be obtained by introducing into the integrand the factor (3.4) which equals one, then carrying out the substitution | a - §, with respect to which the functions *, Af and the measure Md€ are invariant:
Let us discuss the calculation of this integral. The expression An[g] enters into the integral (3.15) only on the hypersurface determined by the equations (3.16). For such gV-v the total contribution to the integral from (3.16) comes from an infinitesimal neighborhood of the unit element of the group. In this nei°hborhood the action of group transformations on the 3 $ and the measure da can be parametrized by means of the infinitesimal func tions n E w S S e d a b S E t a a 5) W WUft thisparamelUl trization ^ ^ ^ ^ ^ * ^ ^ ^ ^
In view of the arbitrariness of the function # ( | ) the measures in the integrals (3.10) and (3.11) must coin cide:
At the unit element of the group, the measure da has the simple form
which yields the equality Df = Af. Let us return to the gravitational field. From what we just said it is clear that in order to define a measure in a class of fields it suffices to define the measure on a manifold of fields which is invariant with respect to co ordinate transformations (and local Lorentz transforma tions) and to specify the equations which parametrize the c l a s s e s . These equations should be Lorentz-invariant if we wish to obtain a covariant perturbation theory. In the metric formalism we choose as such equations the harmonicity conditions of de Donder-Fock:
where Z^(x) is a prescribed vector field. The a r b i t r a r i ness in the choice of l^(x) will be useful in the sequel for formal transformations. The condition (3.12) is not generally covariant and therefore can serve for the parametrization of classes. The analog of Eq. (3.2) is a complicated nonlinear equation for the parameters of the coordinate transformation which takes a given metric into a harmonic one. Within the framework of perturba tion theory this equation has a unique solution. In the following chapters it will be shown that one must select as the invariant measure the expression3' where
This will be done on the basis of an investigation of the Hamiltonian formulation, which we shall interpret as an alternative, non-Lorentz-invariant method of p a r a m e trization of field classes. The advantage of the Hamiltonian formalism is the fact that in it the unitarity condition leads to the stan dard expression for the integration measure. Having the parametrization (3.12) of the classes and the measure (3.13) we obtain the following expression for the functional integral:
where according to (3.4) the functional A h [g] is defined by the e q u a t i o n w h e r
783
So..Phys.-Usp., Vol. 16, No. 6, May-June 1974
Consequently, the integral in which we are interested has the form
Formally this integral equals (det A)"1 where A is the operator acting on the quartet of functions r?M: Thus we have found that
For the formulation of perturbation theory it is convenient to represent det A as an integral over auxiliary fields of a functional of exponential type. These fields must be anticommuting fields, since we need an integral which yields the first power of the determinant. These requirements are satisfied by the expression
where 0 and 9 are classical anticommuting fields satisfying ^e relations and similar relations for the pairs (9, 9), (0, 9). A definition and rules of operation with integrals over anti commuting variables can be found, e.g., in the mono graph of Berezin- 3 9 : i . Returning to the integral (3.15) we can not write it in the form
whi<=h can be used directly for the formulation of p e r turbation theory. However, we still transform it, making use of the arbitrariness in the selection of u. The integral (3.24) does not depend on the choice of JH, by definition. We may therefore average it over IV- with an arbitrary weight. Let us use as a weight the exponential of the quadratic form in the fields e n is y ^ Minkowski metric tensor. The averag ing can be done explicitly and yields the expression L. D. Faddeev and V. N. Popov
783
72
where S e is the action of the free gravitational field and A[g] equals the product of the determinants
o f t e S 2 K U Mth t h f J b T t r a r v ^ J S S T ^ l S ' of the fields hM^ with the arbitrary coefficient a. It follows from our reasoning that the integral does not depend on a A method for seemg this directly has been proposed by DeWitt<> Here we have followed the simpler method of ' t H o o f t ^ . The diagram technique discussed in the preceding chapter follows from the expression (3.26) in the same manner as explained in Chap. 1 for the example of the scalar field. By introducing the "fictitious" fields 9» and 0M we have managed to make the quadratic form in the exponent nondegenerate Thus the inverse operators correspondmg to the operators of the quadratic forms in W, 9» and ?" become well defined, i.e., we obtain the propagators of the particles corresponding to the lines of the diagrams. The graviton propagator ( h ^ ^ ^ 0 ) contains the arbitrary constant a. The propagator (9^8U) of the fictitious vector particles in the k-representation is given by Eq. (2.18). The anticommutation of the fields 0V-, 9^ leads to the factor ( - l ) s for a diagram containing s closed fermion loops. The higher-degree forms in the expansion of the action (3.26) in powers of the fields h^", 8^, Bv give rise to the vertex functions of the diagram technique, as described in Chap. 2. Their concrete form depends on the choice of linearization used. . .. . ,.. . „ u -3/2, vWe note also the role of the local factor n h 5 2 fr in v x the measure. For the linearization (2.12) we have ... „.■f.* u i^u * i •* * • «. and thus this actor should be taken into account in the construc ion of perturbation theory Formally its role reduces to the appearance of a contribution of the form
where A is the operator (3.20>' The p r M e n c e of a ^ ^ thi duct shows that the in^ ^ 2 ^ £ ^ ^ J v b i c h could be introdescription of the electromagnetic field X ^ n t e r a c t s W l th the gravitational field Thus, in coS y ariant p e r ^ b a t i o n the ory for the electromagnetic and gravitational fields a fictitious neutral scalar particle participates in addition to the elements which have been described above. The reader who has understood me basic principles C 0 nstruct l 0 n of the diagram technique for gauge fields 4 approqpriate ^iculations J , '.r** *»f *• fortoemore comP il cated case. THE HAMILTONIAN FORMULATION OF RAVITATiniM THEORY r ^ A A justification of the correctness of the expression (3.13) for the invariant measure is based on the Hamiltonian formulation of gravitation theory. This formulation has been developed by D i r a c [ 1 4 ] . A series of variants for this formulation were obtained by diverse a u t h o r s ^ " 1 9 1 . The construction of an explicitly Hamiltonian form of the Einstein equations runs into the difficult problem of finding solutions to the constraint equations. For us it will be sufficient to consider a generalized Hamiltonian formulation of gravitation theory, . * . , \ . ' t where it is not necessary to solve the constraint equations, and one can restrict one's attention to a verifica tion of the commutation relations. We explain the generalized Hamiltonian formulation of the example of a system with a finite number of de^ * formulation the action funcf ^ ^ consideration ^ the form
in the action giving rise to vertices which are proportional to 6 (4) (0). The appearance of such renormalization terms is noted in many papers treating nonlinear theories (cf., e.g., C 4 0 ' 4 ^). We note that they are absent in the exponential parametrization. In this parametrization the measure (3.13) has, up to a constant factor, the simple form
Here p, q denote canonically conjugate coordinates and momenta which form a phase space of dimension 2n, Si is the Hamiltonian, <pa are the "constraints," A a are Lagrange multipliers (a = 1 m, m < n). The constraints <pa and the Hamiltonian & are in involution, i.e., satisfy the conditions
without any local additions. We have considered in detail the case of the gravitational field in vacuo. Introducing interactions with other fields does not change substantially the scheme of con struction of perturbation theory. For matter fields with
these ns usual Pois SO n brackets
SS^^i^S^^r^^sS^: s i 0 T i " a s fi\Tto,a
t i d e s and their correspondmg diagramfappear o n l / when a field with larger gauge group than the gravitational field is included, e.g., the electromagnetic field or fields of the Yang-Mills type. We shall not consider this case in detail here. We just list, as an example the expression of the functional integral corresponding to the electromagnetic and gravitational fields:
'
£°ed * » ~ £ £ * J ? " £ the m constraint equations
Tehductionof thebdimT £
£^t^rmine^t "
and the m supplementary conditions The functions x a are subject to the conditions
784
Sov. Phys.-Usp., Vol. 16, No. 6, May-June 1974
L. D. Faddeev and V. N. Popov
784
73 In addition it is convenient to assume that the funcThe variables r P „ which differ from the r ^ are not tions x a commute with one a n o t h e r : d y n a m i c a l variables. They can be excluded with the help of the constraint equations In this case it is simple to introduce canonical varia bles on the submanifold r * . Indeed, in view of (4.7), a canonical transformation in r allows one to go over to new variables, where the >•a take the form
The tem (4 14) contains the equations (4.10) together with the equations
where p a are part of the canonical momenta of the new system of variables. Let q a denote the coordinates conjugate to them and p*, q* the other canonical variables. In the new variables the condition (4.6) has the form { '
The solution of the system (4.10), (4.15) expressing the 'nondynamical" quantities r ° n , rk r k in terms of the l0 l0 1J n ,,„ T^. and h ^ " is of the form
and can be interpreted as the condition for solvability of the constraints <pa = 0 with respect to the coordinates q a . As a result of this the surface r * is defined in r by the equations
, where r . k are the three-dimensional connection coeffii]
so that the p* and q* play the roles of independent variables on r * . By construction these variables are canonical. A more detailed discussion of the generalized Hamiltonian formulation for the finite-dimensional case with applications to the theory of the Yang-Mills field is contained in a paper by one of the a u t h o r s ^ . Let us return to the gravitational field. We shall show that its action can be reduced to a form which is a fieldtheoretic analog of (4.1), with the appropriate constraints and the Hamiltonian satisfying conditions of the type (4.2). We shall follow the general method proposed by one of the authors in a form especially adapted for the gravitational field C 1 9 ] . For our purposes it is convenient to make use of a formalism of the first order. We consider the expression of the action for the gravitational field in the form (2.3) and collect in the corresponding Lagrange function all the t e r m s which involve derivatives with respect to
This expression does not contain the variables r £ which occur in y(h, r ) linearly and play the roles of Lagrange multipliers. The factors (denoted by A°°) in front of r £ are the constraints. The constraint equations
allow us to express the variables rL , (r?« - r%) in 0 l0 n HU , ^ terms of the r £ and h^ ". Then the t e r m s containing time-denvatives take the form
if one omits the terms
cients, defined by the three-dimensional metric g i k (i, k = l, 2, 3). us substitute ^ expression (4.16) for the rf 0 , k r k into the ^ g ^ e function x(h, V). After l 0 6 v MO'Mj * ' omittinS several terms of * e ^ of a divergence, which vanish when integrated over three-space when the asymptotic conditions (2.2) are taken into account, the result of the substitution reduces to the form
where
here g3 = d e t g i k , R3 is the three-dimensional curvature scalar associated to the three-dimensional metric g l k M k, = 1, 2, 3). The symbol v k in the expressions for the constraints Ti denotes the covariant derivative with respect to the metric g ^ . As pointed out by Arnowitt, Deser and Misnercis :l , the canonical variables and the expressions for the con straints have an intuitive geometric meaning. The functions qik and "ik serve as the coefficients of the first an<* second quadratic forms associated to the surface x = const, submerged in the four-dimensional spacetime with the metric g and connection r £ „ . More p r e sicely, q1* are a covariant metric density of weight +2, and the n^ form a covariant density of weight - 1 . The constraints are then the well-known Codazzi-Gauss r e lations in the theory of surfaces (cf. e.g., [ 1 3 -). .«,,,, _,, Equation 4.17) solves the problem of reducing the ac^n of the(gravitational field to the generalized Hamiltonian form, analogous to (4.1) for finite-dimensional systems with constraints The constraints T„, as is easily checked, commute with one another. In order to write explicit expressions it is convenient to introduce the quantities
whic , ,„..rat„.,bypart .■qua„„ 411) s u g g e s t X t the natural dynamical variables are the quantities
w h e r e TJ i s a v e c t o r field, pis* density of weight " U . The fo U owing relat l 0 ns hold
785
L. D. Faddeev and V. N. Popov
Sov. Phys.Usp., Vol. 16, No. 6, May-June 1974
785
74
here [T]U rj2] is the Lie bracket of the vector fields, i.e., the vector field with the components r,l8,^-vh,n^, n
where M is the total mass which can be obtained by integrating<SC(X):
Thus, one may consider that St= [$(?xxd3x plays indeed the role of energy. The integrand
and is nondegenerate if the curvature of the metric g ^ is nonzero. , „ „ T l i r . ,_. , . , , « . , « . . A . 5. THE HAMILTONIAN FORM OF THE FUNCTIONAL INTEGRAL The functional integral for the quantization of a c l a s s ** system defined «» ^generalized Hamiltonian formuW i o n has the foU<wmB form ^ >:
The integration measure is here defined by the equation
To prove this we reduce the integral (5.1) with the measure (5.2) to an integral over paths in the physical phase space r*. For this purpose we go over to the coordinates p , p*, q a , q* described in Chap. 4. In these coordinates the measure has the following form:
which can be rewritten: [] 6 ( Pa )6( ? °-<,'(/>', q-))dp*dq" [[ lilJfL
which plays the role of energy density, has the form of a sum of two quadratic forms: one in the derivatives of q1* and another in the "momenta" n^, as required for the energy density of a wave field. In our case this is the energy of the gravitational field having two polarization states in agreement with the usual counting: 2 = 6(coordinates) - 4(constraints). We remark moreover, that in the weak field approximation the Hamiltonian is represented by a quadratic form in the densities of the transverse components of the linearized field. We now discuss the selection of supplementary condi tions. It is widely accepted to use the conditions which were first proposed by D i r a c [ 1 w h e r where q = det q ^ = (det g^) 2 . These conditions have a simple geometric sense: the surface x° = const is minimal and the coordinates x 1 , x2, x3 on it are "harmonic" coordinates. For us the following supplementary conditions will be more convenient:
The integration with respect to the p a and qa is reduced b the delta-functions. As a result the integral takes the form
which is standard for the usual Hamiltonian formula tion^ 4 3 . This proves the correctness of Eq. (5.1), which has ^ advantage over (5.3) of not requiring a solution of the constraint equations The integral (5.1) can be represented in the form
e the functional exp(iS) (S is the generalized action of the system (4.1)) is integrated over all independent variables p i ( q1, X a . Indeed, the Lagrange multipliers Aa enter linearly into this action and the integral with respect to them yields a delta function in the constraints <^a. Starting with this formula we shall not write out the factors of the type of it or the volumes of the integration lattices. mfc l BB "' « Let us return to the case of the gravitational field. We select the supplementary conditions in the form (4.22) and introduce the notations
where 4> is a function with the asymptotic behavior C/r at infinity. The commutation relations (4.7) are satisfied for these conditions. The Poisson bracket matrix of the conditions (4.21) with the constraints is determined by the equations
The analog of the integral looks as follows
786
L. D. Faddeev and V. N. Popov
Sov. Phys.Usp.. Vol. 16, No. 6, May-June 1974
.
786
75
We intend to reduce this expression to a form where the integration is only over the field g^". This will allow us to identify the invariant measure we are looking for. For this purpose we must integrate with respect to the fields n^. This integration can be done explicitly, since the e x p r e s s i o n F o
depends linearly on ir^, and thus the integral turns out to be gaussian; here C is the operator defined by the r e lations (4.22). Let us explain in more detail this feature of det C. The functions ffik enter only into the coefficients Cm, of the operator C, coefficients which do not contain derivatives; moreover the dependence on the rry, is linear. Thus, the operator C can be written in the form
We now show that this integral is an integral over classes of gravitational fields in the sense of Ch. 3, the classes being parametrized by the condition (4.22), and the invariant measure having the form
r this it is sufficient to verify that det B coincides with the factor A [h] obtained according to the rules of Chap. 3 The integral in this expression can be calculated in in the same way as the integral (3.16) in Chap 3 This yieldc: A [k] = det£' (5 13) where the operator B' is defined as follows:
where the operator Cz does not contain derivatives and for each n is determined by a matrix of rank 1. It is known from linear algebra that the determinant of a ma trix A + B, where B is one-dimensional, is linear in the matrix elements of B. The analog of this assertion in our case leads to the indicated linearity of det C with respect to the 77r.. The Gaussian integration over 7rik. reduces to the sub stitution
where 7rjk(h) is an expression which follows from formulas of ffie type (2.1), expressing the Christoffel symbols in terms of the metric. After such a substitution the a c tion corresponding to the Lagrangian (4.17) turns into the initial covanant action (2.3). The determinant det C turns into the product
where B is the operator defined by the equations
Finally, the local factors in the products of differentials together with the local factor which appeared in the integration over ffik and the differentials themselves collect i to the express o
Here the factor in front of the differentials can be r e duced to the f
o
r
m
it is easy to see that
^deed, one can go over from one operator to the other by means of the triangular substitution Let us summarize. Starting from the obviously unitary Hamiltonian formulation of the functional integral, after formal changes of the integration variables we have transcribed it in the form of an integral over equivalence classes of fields with a concrete parametrization of the classes. The corresponding invariant measure has the form (5.12). This justifies the Lorentzinvariant expression for the functional integral in Chap. 3, which represents another writing of the same integral with another parametrization of the classes. With these considerations we conclude the derivation of the covariant rules of perturbation theory for the quantization of the gravitational field. The next problem which appears here is considerably more difficult. It consists in a consistent performance of the renormalization procedure based on an invariant r e g u l a t i o n . The difficulties are caused by the unwieldiness of the theory as well as by the fact that from a formal point of view the theory is non-renormalizable. We hope that symmetry and general covariance considerations will help in solving this problem.
'>For instance r = 2 for the diagram
">
.
.,, and the l a s t f a c t o r q i / 2 can be omitted owing to the c o n s t r a i n t q = exp $ . As a r e s u l t our functional i n t e g r a l t a k e s the f o r m
787
L. D. Faddeev and V. N. Popov
Sov. Phys.Usp.. Vol. 16, No. 6, May-June 1974
)
^
^ Q{ funct l0 n a| (path) integrals in vanous problems of quantum mechanics and field theory is discussed in the recent review article o1" Blokhintsev and Barbashov in Uspekhi ( ;)7b ]. ])Any measure [\h*11dH^•„invariant with respect to the group of coordinate tranlformations. Indeed, under coordinate transformations this expression acquires a factor of the type [Jtdet (&'^'4^F- This factor should be considered equal to one, since for infinitesimal trans formation it is equal to exp [v Tr In (1 + S^igl = exp (Y8"> (0)J3^**) = t,
J
787
76
l
J. A. Wheeler, Geometrodynamics, Academic Press, N. Y. 1962 (see also C. W. Misner, K. S. Thome and J. A. Wheeler, Gravitation, W. H. Freeman and Co., San Francisco, 1973; added by Transl.). 2 C. W. Misner, Phys. Rev. 186, 1319, 1328 (1969). 3 M. A. Markov and V. P. Frolov, Teor. Matem. Fiz. 3, 3 (1970); V. A. Berezin and M. A. Markov, ibid., p. 161. 4 W. E. Thirring, Ann. Phys. (N.Y.) 16, 96 (1961). 5 L . D. Faddeev, Proc. Intern. Congress of Mathematicians, Nice, September, 1970, Vol. 3, p. 35. 6 V. A. Fock, Teoriya prostranstva, vremeni i tyagoteniya (The Theory of Space, Time and Gravitation), Fizmatgiz, Moscow, 1961. 7 B. S. DeWitt, Phys. Rev. Lett. 13, 114 (1964). 8 1 . B. Khriplovich, Yad. Fiz. 3, 575 (1966) [Sov. J. Nucl. Phys. 3, 415 (1966)]. 9 C. J. Isham, A. Salam and J. Strathdee, Phys. Rev. D3, 1805 (1971). 10 E. S. Fradkin, Nucl. Phys. 49, 624 (1963); 76, 588 (1966). 11 G. V. Efimov, Zh. Eksp. Teor. Fiz. 17, 1917 (1963) [sic!] 12 M. K Volkov, Ann. Phys. (N.Y.) 49, 202 (1968). 13 A. T. Filippov, Preprint JINR R-1439, Dubna, 1964. 14 P. A. M. Dirac, Proc. Roy. Soc. A246, 333 (1958). 15 R. Arnowitt, S. Deser and C. W. Misner, Phys. Rev. 117, 1595 (1960). 16 J. Schwinger, Phys. Rev. 130, 1253; 132, 1317 (1963). 17 P. G. Bergmann, Rev. Mod. Phys. 33, 510 (1961). 18 J. L. Anderson, Rev. Mod. Phys. 36, 929 (1964). 19 L. D. Faddeev, Gamil'tonova forma teorii tyagoteniya (The Hamiltonian Form of Gravitation Theory), Abstracts of the 5-th International Conference on Gravitation and Relativity, Tbilisi, 1968. 20 S. N. Gupta, Proc. Roy. Soc. A65, 1952. 21 N. N. Bogolyubov and D. V. Shirkov, Vvedenie v teoriyu kvantovannykh polef (Introduction to the Theory of Quantized Fields), Gostekhizdat, Moscow, 1957; Engl. Transl. Wiley-Interscience, 1959. 22 A. I. Akhiezer and V. B. Berestetskii Kvantovaya elektrodinamika (Quantum Electrodynamics), Nauka, Moscow, 1969; Engl. Transl., Wiley-Interscience, 1965.
788
Sov. Phys.-Usp., Vol. 16, No. 6, May-June 1974
23
R. P. Feynman, Acta Phys. Polonica 24, 6 (1963). " B. S. DeWitt, Phys. Rev. 160, 1113; 162, 1195, 1239 (1967). M L. D. Faddeev and V. N. Popov, Phys. Lett. B25, 29 (1967); V. N. Popov and L. D. Faddeev, Teoriya vozmushchenii dlya kalibrovochno-invaraintnykh polef (Perturbation Theory for Gauge-Invariant Fields), Preprint, Inst. Theor. Fiz. Acad. Sci. Ukr. SSR, Kiev, 1967. 26S. Mandelstam, Phys. Rev. 175, 1580, 1604 (1968). 27 M. Veltman, Nucl. Phys. B7, 637 (1968); B21, 288 (1970). 28E. S. Fradkin and L V. Tyutin, Phys. Lett. B30, 562 (1969); Phys. Rev. D2, 2841 (1970). 29 I. B. Khriplovich, Yad. Fiz. 10, 409 (1969) [Sov. J. Nucl. Phys. 10, 235 (1970)]. 30 D. G. Boulware, Ann. Phys. (N.Y.) 56, 1 (1970). 31 G. 'tHooft, Nucl. Phys. B33, 173 (1971). 32 A. M. Altukhov and I. B. Khriplovich, Yad. Fiz. 11, 902 (1970) [Sov. J. Nucl. Phys. 11, 504 (1971)]. B A . A. Slavnov and L. D. Faddeev, Teor. Matem. Fiz. 3, 18 (1970). 34 A. I. Vainshtein and I. B. Khriplovich, Yad. Fiz. 13, 198 (1971) [Sov. J. Nucl. Phys. 13, 111 (1971)]. M V. I. Zakharov, ZhETF Pis. Red. 12, 447 (1970) [JETP Lett. 12, 312 (1970)]. 36 I. V. Tyutin and E. S. Fradkin, Yad. Fiz. 13, 433 (1971) [Sov. J. Nucl. Phys. 13, 244 (1971)]. 37 a) R. P. Feynman, Rev. Mod. Phys. 20, 367 (1948); Phys. Rev. 84, 108 (1951); b) D. I. Blokhintsev and B. M. Barbashov, Usp. Fiz. Nauk 106, 593 (1972) [Sov. Phys.-Uspekhi 15, 193 (1972)]. 38 H. Weyl, Phys. Rev. 77, 699 (1950). 30 F . A. Berezin, Metod vtorichnogo kvantovaniya (The Method of Second Quantization), Nauka, Moscow, 1965; 40 H. Umezawa and Y. Takahashi, Prog. Theor. Phys. 9, 14, 501 (1953). 4l T . D. Lee and C. N. Yang, Phys. Rev. 129, 885 (1962). 42 L. D. Faddeev, Teor. Matem. Fiz. 1, 3 (1969). 43 G. Favard, Cours de geometrie differentielle locale (Russian transl.: IL, 1960). 44 G. Garrod, Rev. Mod. Phys. 38, 483 (1966). Translated by Meinhard E. Mayer
L. D. Faddeev and V. N. Popov
788
77
Comments on Paper 5
This paper constitutes the lecture notes of a course given at the famous Les Houches summer school in August 1975.1 was to give an introductory course on the functional integral quantization. In these lectures, a point of special pride for me is the stress on the necessity of extra boundary terms in the action when the holomorphic quantization is used. Another technical novelty is the proper definition of the S matrix in terms of the asymptotics of the admissible trajectories in the functional integral. This approach was implicitly advocated by Feynman himself; however, I had not seen a selfcontained exposition of this idea in any written form before and was to devise it myself for the course. This text influenced the monograph on quantization of the gauge fields, which I wrote with A. A. Slavnov some time later.
79
COURSE 1 INTRODUCTION TO FUNCTIONAL METHODS
Ludwig D. FADDEEV V.A. Steklov Institute of Mathematics, Academy of Sciences, Fontanka 25, Leningrad 191011
§0. Contents
1. Functional integral and Feynman diagrams 2. Functional integral in phase space 3. Holomorphic form of the functional integral 4. Generating functional for the 5-matrix 5. The case of fermions 6. Some special features of the quantization of the Yang-Mills field References
R. Balian and J. ZinnJustin, eds., Les Houches, Session XXVII,. 1975 - Mithodes en theories des champs /Methods in field theory © North-Holland Publishing Company, 1976
3 5 11 19 28 32 39
81
Functional methods
3
1. Functional integral and Feynman diagrams We begin with the formulation of the main result of the functional ap proach. Consider for simplicity the example of the self-interacting scalar neu tral field >(*) with tth Lagrangian
where V(
The functional Z(V) of the arbitrary function rj(X) (source)
is a generating functional for the Green functions of the system in considera tion,
There exists no rigorous definition of the integral "over all fields 0(JC)" which is symbolically written in (1.1). We will understand the expression (1.1) in terms of perturbation theory. Let us expand the expression for G(xlt...,x„) which follows from (1.2) in a power series in A:
The integrand in the last expression has the form exp{/ X quadrltic form} X polyntmial , if we understand that
82
4
L.D. Faddeev
Indeed let us consider the finite dimensional integral
where the matrix Aik of the quadratic form and the coefficients C^ _. of the polynomial are given. To evaluate / it is convenient to introduce the gen erating function
We are using the natural matrix notations. It is evident that
Now differentiating Z 0 (T?) we find
so that to get the non-vanishing result when r? = 0 one must differentiate also 77 in the second factor. It is clear from this observation, that / i s obtained by summing over all possible ways of saturating the indices of Ct by the { l products of the matrix elements of the matrix A ~ such thatthe index k comes in ik times. Let us return to our infinite dimensional expression and choose the nor malizing factor Was the integral
The operator
plays now the role of the matrix A, x being the analogue of the index k. The matrix elements of the inverse matrix are just the values of the free Green function
Jtt Functional methods
5
The polynomial we integrate has a particularly simple form
It is now easy to see that the finite dimensional rules described above lead iu the Feynman diagram prescription for the calculation of the Green functions, vacuum diagrams included, if we take as G(x - y) the Feynman propagator, obtained from the underdetermined expression (1.3) by the substitution m2 -> im2 - iie All the combinattons of the chronological lnd normal products and the Wick theorem are contained in the rules of the Gaussian integration. Two questions come to mind immediately after this observation: (i) Is it possible to explain the expression (1.1) or the corresponding gen erating functional for the S-matnx without any reference to the operator methods of the textbooks of quantum field theory? (ii) Why must the kernel of the operator, inverse to ( - D - m2) be taken in the form of the Feynman propagator? The following lectures are organized to give a systematic approach to an swer both these questions. We will begin with the derivation of the representa tion of the matrix elements of the evolution operator in terms of the function al integral. The 5-matrix will be obtained further by means of the appropriate limiting procedure.
2. Functional integral in phase space The general methods of quantum theory are more easily illustrated on the example of quantum systems with a finite number of degrees of freedom rath er than in the field theoretical case. That is why in these lectures we shall quite often describe the methods on the finite dimensional example before ex plaining their realization in field theory. Let us consider the mechanical system with one degree of freedom and the simplest phase-space two-dimensional plane. Canonical coordinates p and q, -°°
84 6
L.D. Faddeev
Let h(p,q) be a Hamiltonian function and h the corresponding energy opera tor. We are interested in the evaluation of the evolution operator
Let us suppose for definiteness that when quantizing h we have chosen the or der of factors in such a way that all operators p stand to the left of q. The for mal answer we shall obtain in this lecture does not depend on such a choice. With this convention we have a simple formula:
which follows immediately from (2.1). We shall now use it to evaluate the matrix elements of the evolution operator
The idea is to approximate the operator e~^e for small e in the form
and then consider the limit
To get the matrix elements of (2.4) we can use (2.3) and obtain
85 Functional methods
7
and to evaluate (2.5) we have to make convolutions of the products of such expressions. The strategy being clear, let us proceed with the calculations. For TV = 2 we get
where the factors ( l / 2 * ) W e * M " and (l/2ir)We*i«i convert the momen tum representation to the coordinate one to get the product of the operators involved. We rewrite the last formula in the form
where we understand that
It is clear from this example that the case of general N looks as follows
The limit N -> °°, e -> 0, TVe = t" - t\ corresponds to the integration over an infinite number of variables we can formally present as p{t\ q(t), t'
86
8
L.D. Faddeev
i.e. that of the action functional for the phase-space trajectory p(t),q(t). This observation leads to the main result of this lecture: the matrix element (2.3) can be obtained by averaging the Feynman functional exp{/ X action} over all trajectories in phase space for a given time interval t < t < t" with the pre scribed initial and final values for coordinate variables
The last formula is attractive being expressed entirely in terms of the clas sical mechanical ingredients: action and Liouville measure on the phase space. The last objects are canonical invariants. This makes one suspect that with the proper definition of the functional integral one would be able to devise a canonically invariant procedure to obtain the quantum mechanical answers from the corresponding classical ones. This can not be true for all observables be cause it is well known that there exists no natural action of the group of gen eral (non-linear) canonical transformations in the space of states of the quan tum system. This shows that one must be very careful in dealing with such a formal object as the functional integral. I shall not make any attempt to devise an intrinsic definition of the r.h.s. in (2.6). Our understanding of this object will increase in the course of the for mal manipulations with it. The derivation above suggests the definition of (2.6) by means of a finite dimensional approximation of p(t) by piece-wise constant and of q(t) by piece-wise continuous functions and using the quadra ture formula to approximate the integral for the action. This procedure, how ever is by no means a natural one. Different possibilities appear if h(p,q) con tains a quadratic form - aa sn the field theoretical lxamplle One ccn then use a well-defined perturbation theory. The original formula of Feynman differs from (2.6) in the following way: the action is considered as a functional of the coordinate-space trajectory q(t) only. It is easy to get the Feynman formula in the case when the Hamiltonian is quadratic in the momentum variable
Indeed, the integration over p(t) in (2.6) is of a Gaussian type,
and after the shift of the integration variables,
87 Functional methods
9
we get instead of (2.8) the expression
Using it in (2.6) we obtain
which is the original Feynman formula. It is clear from the derivation that (2.9) is less general than (2.6); the former is valid only if the Lagrangian is quadratic in velocities. The case of any number of degrees of freedom can be treated analogously. With the appropriate choice of the vector notations
one can use formulas (2.6) and (2.9) in this case. The field theoretical applica tion is evident in the representation when the field operator 0(x) )i nome eiven time (say, x0 = 0) is diagonal,
Here His a Hamiltonian operator, corresponding to the Lagrangian £ of lec ture 1. One sees the relevance of (2.10) to understanding the role of the func tional integral in that lecture. For further use, let us note here, that one can rewrite (2.10) in the form
88 10
L.D. Faddeev
where the first factor must be understood in terms of a power series expansion in X. We see that in the perturbation approach the dynamical problem of quan tum field theory reduces essentially to that for the non-self-interacting field influenced by the external source. We shall finish this lecture with a comment about the definition of the Smatrix by means of the functional integral. Let us return to the case of one degree of freedom and consider the non-relativistic particle interacting with the short range potential. The Hamiltonian of the system is given by (2.7) and we shall rewrite it in the form
The asymptotic behavior of the solutions of the Schrodinger equation is char acterized by the operator h0 and the S-matrix is defined in the usual way,
The matrix elements of the 5-matrix between two states \j/j and \J,2 described in the momentum representation by the functions Cj(p) and C2(p) can be written in the form
where /,■(*, f) are the corresponding wave packets
The momentum representation S(p",p) of the ^-matrix is defined by
To compute the limit (2.13) it is convenient to use the approximate form of the wave packet
which can be obtained by means of the stationary phase method. Substituting (2.15) in (2.13) and changing the variables x -+ pt/m ww ebtain
89
Functional methods
11
Comparing (2.14) and (2.16) and remembering the functional integral expres sions (2.6) or (2.9) for (q",t"\q',t') we can say that to get the 5-matrix in the momentum representation one is to calculate the Feynman functional inte gral over the trajectories defined on the whole time axis -°° < t < °° having special asymptotic behavior for |r| -*•*•
This behavior is defined by the asymptotical dynamics which in our example is a free motion of the non-relativistic particle of mass m. Some factors are to be cancelled before the calculation of the limit. This observation shows that the S-matrix is defined by means of a func tional integral in a more elegant manner than the evolution operator. There is no reference to the fixed boundary conditions, they are substituted by the asymptotic behavior of the trajectories. This is especially attractive in the case of field theory because the expressions of the type (2.11) become manifestly Lorentz covariant, if the time JC0 varies on the whole axis. However the asymp totic behavior in this case is not easy to define in the (*) representation. In deed, the analogue of the operator H is not p2/2m, but rather an oscillator Hamiltonian h = p2/2m + |u2<7. This operator has a simple form in the socalled holomorphic representation, defined by Bargmann, Berezin and Segal. Having all this in mind it is natural to try to get the analogue of (2.6) in holo morphic representation before the discussion of the S-matrix for quantum field theory. This question will be discussed in the next lecture. Problem Choose some other ordering of factors for quantizing h(p,q) (i.e. such that all q are to the right of p) and repeat the derivation of (q"t"\q't'). Verify that the final formal expression for it coincides with (2.6). 3. Holomorphic form of the functional integral Let us consider once again a system with one degree of freedom but use the
90_
12
L.D. Faddeev
complex coordinates a* and a instead of the real p and q in phase space. Hav ing in mind the application to oscillator type problems we shall choose the fol lowing form for a* and a:
co being an arbitrary rral larametee which will lecome e arequency in whha follows. In quantum mechanics the variables a* and a turn into the operators a* and a with the commutation relations
(the latter having sense only if we have more than one degree of freedom). The only difference between p and q is that i has disappeared from the com mutation relations. Operators a* and a are to be adjoint to.each other. The representation
when we do not specify the nature of the variable z, clearly satisfies the com mutation relations. We face the problem to find the inner product for the functions/(z) which the operators a* and a act on such that those operators are mutually adjoint. The authors mentioned above have found such an inner product. One is to consider polynomials/(z) in the complex variable z and define the scalar pro duct in the form
where it is understood that
This product is positive definite; indeed the monomials are orthonormal with respect to it as shown by the following computation
where we used polar coordinates z = pe'* and the r.h.s. is equal to 0 if n ± m and/*! ifn = m.
91
Functional methods
13
The mutual formal adjointness of a* and a in this representation follows from the identities
and the complex form of the Cauchy-Riemann condition
The complete space of state vectors on which a* and c act must be obtained by means of closure of the linear combinations of the monomials in the norm defined by the inner product. It can be shown that this space consists of the entire functions of the type \ and that a* and a when properly defined are truly adjoint to each other. The functions
are the eigenfunctions of the occupation number operator h or oscillator Hamiltonian h
We can use this to connect the representation described with the more famil iar ones, say the coordinate representation. This will be left as an exercise. Let us turn now to the problem of representing the operators. We shall use two ways to do it. The first is based on the matrix elements of a given opera tor in the /„(z) basis. If
then we shall represent the operator^ by a kernel
It is clear that the action of A on an arbitrary vector/(z) is given by the for mula
92
14
L.D. Faddeev
and the product of two operators looks as follows
One sees that to be able to realize these formulae it is necessary to consider A(z,z) as an analytic function of two independent complex variables z and z which are not to be taken as being complex conjugate one to another. The second representation is a well-known normal product, when ,4 is writ ten as a power series in a* and a such that all a* are to the left of a,
It is convenient at this point to make some change in the notation and use the variable a* instead of z
We see that one can associate with a given operator A two analytic functions of two variables: the kernel A(a*,a) and the normal symbol
Let us show that these two functions are connected by the relation
To prove it, it is sufficient to consider a monomial
so that
Now
93
Functional methods
15
where 6{n>k) = 1 if n >k and 0 if n
QED. Formulae (3.1) and (3.2) are very similar to (2.1) and (2.2). The substitu tion
enables one to convert one set of formulae to another. This shows that we can repeat all the procedure of the previous lecture to get the expression for the kernel U(a*,a,t",t') of the evolution operator in terms of the functional inte gral over the trajectories a*(t), a{t) in the phase space. Indeed, let h(a*,a) be a Hamiltonian in a*-a variables which we quantize by choosing the normal product order
Then the kernel U(a*,a;e) of the evolution operator U(e) = exp{-/e/j} for small e is given by
For the finite-time difference t" -t'=Ne such kernels
where it is supposed that
we are to make a convolution of
94
16
L.D. Faddeev
The formal limit N-+ - , e ^ 0 cac be bepressed in tht form
or after the symmetrization in a* and a,
It is especially evident in the last expression that we have got the result equiv alent to (2.6). Indeed, the integral
is nothing but the action of a trajectory in phase space given in terms of the complex coordinates. The extra term exp \(a*{t")a(t") + a*(t')a(t')) in the integrand takes care of the fact that we are using different boundary condi tions for the trajectory at t = t" and t = t'. Two temptations are to be avoided in these formal manipulations: (i) It is wrong to consider that a{t") is the complex conjugate of a*(t") or that a*(t') is the complex conjugate of a(t') and extract the factor exp \(a*(t")a(t") +a*(t')a(t)) from under the integration symbol. It must be understood that a(t") is the value at t = t" of the function a(t) for which only the value a(r') is fixed. (ii) It is wrong to say that the normal symbol of the evolution operator U(a*,a\t" - t'))t-a*a can be computed as a functional integral over the closed trajectories going through the point a*,a in phase space in spite of the fact that the factor
seems to provide the term we lack to convert the sum in (3.3) into the integral
_95 Functional methods
of the form
17
a dr. Indeed there ii no rrason to belleve that aQ muss be
close to aN in the limit e -* 0. We finish this lecture by applying formula (3.4) to describe the evolution operator for the harmonic oscillator in the interaction with the external source. The relevance of this example to our main subject is clear from the comment at the end of the previous lecture. The Hamiltonian of the system in question looks as follows where the complex function y(t) represents the time-dependent source. It is clear that all the derivations above can be used for time-dependent Hamiltonians. The only difference is that the evolution operator is not a function of the time difference and that the action is to be defined by
The action for our system is quadratic in a*(i) and a{t) and the functional in tegral (3.4) is a Gaussian one. We can evaluate it saying that it is equal to the value of the integrand at the critical point (i.e. at the stationary trajectory). The critical point is defined in terms of the classical equation of motion. Indeed for arbitrary h(a*,a) for fixed a*(t") = a* and a(t') = a,
because the boundary term we acquire when integrating by parts is cancelled by [This is one more confirmation that the strange looking term a I is quite natural.] In our example the equations of motion look as follows:
and have a unique solution subject to the boundary conditions
namely
96
18
L.D. Faddeev
Let us note that this solution is not real in the sense that a*(t) is not a com plex conjugate of a(f). Nevertheless we are to use it to evaluate the Gaussian integral. Indeed, it is well known that the Gaussian integral is equal to the in tegrand at the critical point even if it does not belong to the space over which one integrates. We are now to substitute the trajectory found into the integrand in (3.4). The action part can be rewritten in the form
and the first term on the r.h.s. vanishes at the critical point. Collecting the terms both from the action and the boundary term we get
It is convenient to symmetrize the last term with respect to y(u) and y(u). After this is done we can write the answer for the kernel U(a*,a;t",t') we are looking for in the form
The system of several degrees of freedom, each having its own frequency u can be considered in the same way. Each degree of freedom gives its own con tribution to U(a*,a;t",t') which we are simply to multiply. This result will be used in the next lecture to get a final expression for the field theoretical 5-matrix.
97
Functional methods
19
Problems 1. Find the kernel K(x,z) such that the map
converts the holomorphic representation into the coordinate one. Describe the inverse map. 2. Show that
so that ta*a plays the role of the 5-function in the holomorphic representation. 3. Consider the integral
over the real variables z, when A is a complex matrix, Re A > 0 and b is a complex vector. Show that/= const. ew, W= -\{Az,z) + bz\z=A-\h.
4. Generating functional for the S-matrix We are now able to consider the main problem of these lectures, namely the construction of the ^-matrix by means of the functional integral. We begin with the case of a scalar field
and it leads to the Hamiltonian
which can be written in terms of a system of oscillators interacting with the time-dependent source. This well-known fact will enable us to use the result of the previous lecture to get the desired answer.
98
20
L.D. Faddeev
Let us introduce the complex coordinates a*(k), a(k) instead of the real n(x),
In terms of a*(k) and a(&) the Hamiltonian acquires the form
where
The analogy with the formulae of the previous lecture is apparent; the variable k plays the role of the label for the degrees of freedom. Generalizing the for mula (3.5) we get for the kernel of the evolution operator U(t",t') = Texp{-i / / H(t)dt} the following expression:
Let us use this to evaluate the S-matrix. We suppose that the source van ishes for | f | -*°° (note that the self-interacting terms we will be faced with in the general case effectively have such a property). Then the asymptotical dy namics are governed by the operator
so that the 5-matrix is defined by
99 Functional methods
21
It is easy to make the necessary convolutions to calculate the operator product in the r.h.s. Indeed, the kernel of the operator U0(t) = exp {-iH0r} has the form
and using the result of exercise 2 of lecture 3 we get for an arbitrary operator A with the kernel A(a*,a)
so that the multiplication by U0(f') from the right and by U~l{t") from the left reduces to the change of variables
in the kernel. The new variables are in fact the solutions of the classical equa tion of motion for the Hamiltonian HQ. As a result the kernel of the operator on the r.h.s. of (4.3) has the form
and the limit t" -> » and r'-> ^ is obtained simply by extendind the time in tegration to the whole axis. Let us rewrite the expression we should get in the limit using the source n(x,r) rather than its Fourier transform y(k,t),
100
22
L.D. Faddeev
where Sn(a*,a) stands for the kernel of the S-matrix for scattering on the source T?(JC). It is clear now that it will be more elegant to use the normal sym bol. Indeed, to do this we are to drop the first term in the exponent. The re maining terms are manifestly Lorentz covariant. To see this let us introduce the solution of the free Klein-Gordon equation
and the Green function of this equation
It is easy to see that
Now the S-matrix for the self-interacting field has the form
and the discussion analogous to that at the end of lecture 1 shows that the last expression readily leads to the Feynman rules of perturbation theory. We have finished answering the two questions posed at the end of lecture 1. We have given the full derivation of the Feynman rules using the functional in tegral description of quantum dynamics. In particular we have shown how the correct account of the asymptotical conditions leads uniquely to the use of the Feynman propagator. We have used the scalar field as a simple example. The generalization for the spinor and vector fields will be discussed in the fol lowing two lectures. We shall finish this lecture with a brief discussion of the possible modifica tions of perturbation theory which follow from the general expression for the S-matrix we have found. Let us comment that it is by no means necessary to use the trick of introducing the source T?(X) and differentiating it to define the S-matrix. An alternative expression can be given directly in terms of the func tional integral. Let S(a*,a) be a kernel of the S-matrix. Then
101 Functional methods
23
where the trajectories over which we integrate have the prescribed values
We use here the notation Hint(a*,a) for the functional of two functions a*(k) and a(k) which we get by substitution of (4.1) into the interaction functional,
All the ideas necessary to derive this result have been illustrated earlier and we shall not repeat them. If we forget for the moment about the boundary terms in (4.5) we can say that the kernel of the •S-matrix is obtained by integrating the functional exp{/ X action} over the trajectories a*(k,t), a{k,t) (or 0(x), n(x)) defined on the whole time axis with a prescribed free-motion for the incoming wave a{k,t)~a(k)t-iu>t, f->-°°and the outgoing wavea*(fc,f) ~a*(k)ei<jJt, t -*■ °°, The boundary terms take care of the subtractions necessary to make the action functional on such trajectories converge. The most natural perturbation scheme based on this definition is given by the stationary phase approximation to the functional integral. To realize it, it is convenient to restore the Planck constant h, which until now we considered to be equal to 1. The Feynman functional will acquire the form exp{(z'/fc X ac tion}. Of course ti may enter the action itself through the mass term, for ex ample. This "internal" fi will be held fixed in the following and we will pro ceed with the construction of the power series expansion only in the "exter nal"/!. The main contribution to the Feynman integral will be given by the value of the integrand in (4.5) on the critical trajectory. We already know that the boundary terms in (4.5) have such a form that the critical trajectory is given by the classical equations of motion:
102
24
L.D. Faddeev
The solutions subject to the boundary conditions (4.6) have a limit for t" -*■ °° and t' -* —°°; the asymptotic form can be found by the iteration of the follow ing integral equation:
Introducing the field 0c[(x), where
we see that it satisfies the classical field equation
and that the integral equations (4.7) are equivalent to the integral equation
where
103
Functional methods
25
The first term is to be dropped if we are interested in the normal symbol of the S-matrix. The other two can be rewritten in terms of the >cl. Performing the trivial substitutions we find the main contribution to the normal symbol of the S-matrix has the form
The notations are explained as follows. We use the argument (J)Q(X) instead of a*(k), a(k) for the functional Stiee to emphasize the-Lorentz invariance of the answer; 4>Q(X) is given by (4.4) and is uniquely connected with a*(k) and a(k). The argument 0 C | on the r.h.s. is a solution
computed on the classical solution
which is equal to the exponent of (4.9) if we agree to integrate by parts in the first two terms, use the free-motion equation to eliminate them and not care about the boundary terms. The next approximation is given by a Gaussian integration over the small deviations from the critical trajectory. Let us make the change of variables in the functional integral (4.5),
104
26
L.D. Faddeev
We can now use as well the real variables
instead of the a and a* but the boundary conditions are more easily described in the former. The quadratic form in the deviations >(*), 7r(x) is given by
where the particular choice of the first term on the r.h.s. is irrelevant due to the boundary conditions on 0 and if. The Gaussian functional integral
can be evaluated to give
where the operator ( - □ - m2) l is to be represented by the Feynman propa gator. The functional S0(>0) generates the contribution to the normal symbol
105
Functional methods
27
of the S-matrix from the diagrams with one closed loop. To check it is a sec ond exercise. It is clear now how to proceed to the higher orders in (ft) 1 / 2 . One decom poses the action written in the variables $ and if,
and performs the Gaussian integration. There appears the natural generaliza tion of the Feynman rules. The quadratic form (1/ft) ^? 2 supplies the propa gator which is a Green function <^{x,y) of the differential operator
uniquely defined by means of the integral equation
Higher orders in (4.10) give the vertices, the total number of which is equal to the order of the polynomial V minus 2. The normal symbol of the S-matrix is given in terms of exp {i 2 vacuum diagrams}. The scheme described seems to be more complicated compared to usual perturbation theory. However it turns into the latter if one uses perturbation theory to solve eqs. (4.8) and (4.11). On the other hand it can be convenient for general theoretical investigations, because the combinations of the ordinary diagrams contained in one vacuum diagram in this generalized scheme can have nice invariance properties. This will be important for the discussion of the renormalization of gauge field theories and quantum gravity. Problems 1. Check that the functional Stiee(
106
28
L.D. Faddeev
5. The case of fermions The functional methods described above explicitly take into account the Bose statistics of the fields involved. To treat the fermions one must use a dif ferent procedure. It is very desirable to have this procedure as close as possible to that used for bosons because the description of the interaction of bosons and fermions requires a unified scheme to display their dynamics. It was shown by Berezin that it is possible to devise a system of notations for fermions such that one can define the functional integral for fermions to look the same way as that for bosons. Let us begin to describe this system of notation on the example of the sys tem of fermions with one degree of freedom. The space of states for such a system is two-dimensional and two operators a* and a generate all algebra of the observables. These operators satisfy the anticommutation relations
and can be simply represented by means of 2 X 2 matrices. We shall devise an other representation for them. Let us consider two anticommuting variables a* and a,
considered as abstract algebraic symbols. In mathematics they are called the generators of the Grassman algebra. The general function f{a*,a) of these vari ables (an element of the Grassman algebra) can be written in the form
where f$,f\,f\ and/ 2 are complex numbers. We see that the linear space of such functions is four-dimensional. The "analytic" functions/(a*), i.e. the functions of the variable a* consti tute a two-dimensional space. We shall use it to represent the state space of our system. The operators a* and a are realized in the form
where the differentiation operation is defined as usual,
107
Functional methods
29
It is easy to see that the anticommutation relations (5.1) are satisfied. Indeed a*2 = 0 because a*2 = 0, and a 2 = 0 because nothing remains after differentiat ing twice. Now
which gives the first relation in (5.1). Let us now introduce the integration of the functions f(a*,a). By definition
We will get this answer if we consider da* and da as anticommuting variables between themselves as well as with a* and a and introduce the rules of integra tion, devised by Berezin,
Note that the last two formulae show that the integral of a derivative is equal to zero. The notion of integration enables us to introduce the inner product into the space of functions/(a*). Define
Then
gives a positive definite inner product. Indeed
This shows that the monomials /0 = 1 an( * *l>\ =a* constitute the orthonormal basis in the space of the functions/(a*). The operators a* and a are ad joint to each other in the inner product we have introduced. One can devise two ways of realizing the operators, either by means of a kernel or a normal symbol. The latter is defined as follows: with the normal form of a given operator A,
108
30
L.D. Faddeev
we associate a function of the variables a* and a,
The kernel ,4 (a*,a) is given by
where Amn, m,n = 0, 1 are the matrix elements of the operator A in the O 0 , i^j) basis,
It is clear that
We see that to define these formulae we are to enlarge the set of the anticommuting variables by introducing the variables of integration a*,a. By defini tion they anticommute with a* and a. The normal symbol K(a*,a) and the kernel A (a*,a) of a given operator A are connected by means of the formula
We leave the proof as an exercise. All the formulae written above can be readily generalized to the case of a system with several degrees of freedom. One must use the set of 2n anticommuting variables
and pay attention to the order of factors. Arbitrary analytic functions/(a*) constitute the space of the state vectors. Creation and annihilation operators a* and a. act as follows
109
Functional methods
31
The subscript L on the differentiation symbol means that the variable ai (if present in f{a*)) is to be pushed to the extreme left using anticommutativity before being dropped. These operators are adjoint with respect to the inner product
Here the * operation is defined by
and the integration is defined as above in the understanding that all the sym bols a*, a{, da*, dat are anticommuting. The monomials a,- ... aik are orthonormal with respect to this inner product. The kernels (a*, a) ana the normal symbol K(a*,a) of a given operator A are defined as in the case of one degree of freedom. The relations
are also satisfied. Now if we look at all the calculations in lecture 3, we realize that all the algebra is based on relations analogous to (5.3) and (5.4). This shows that all the results obtained there will be valid in the developed symbolism, including the fermion particles and fields. In particular, the kernel for the evolution op erator, generated by the Hamiltonian with the normal symbol h{a*,a,t) looks as follows
where it is supposed that
no 32
L.D. Faddeev
We shall not write the explicit formulae for the field theoretical case including the fermion fields. The education from the previous lectures suffices for one to do it him (or her) self. Problems 1. Show that
2. Obtain the rule for changing variables in the integral for an arbitrary nonhomogeneous linear substitution of the integration variables. 3. Prove (5.3). 4. Describe the generating functional for the Green function for the spinor charged field.
6. Some special features of the quantization of the Yang-Mills field This lecture is a transitional one to the course of Professor B. Lee. We shall give an introduction to the problem of quantization of gauge fields which will be discussed in his lectures in much more detail. First let me say several words about the definition and general meaning of the gauge field. We can associate it with an arbitrary compact Lie group G. Let ta,a- 1,... , n, be generators of the corresponding Lie algebra in the ad joint representation, n being the dimension of the group. We shall use the nor malization
then the structure constants fabc introduced in
are real and totally antisymmetric. Gauge field/I describes the parallel trans port of the arbitrary fields \p belonging to one or another of the representa tions of the group G. The space of this representation is called the internal space or space of charges to distinguish it from the ordinary space-time. Using
Ill
Functional methods
33
the words of H. Weyl, applied by himself to electrodynamics, one can say that the theory of gauge fields is a general relativity in the internal space. More specifically the gauge field is defined as a vector field A with the values in the adjoint representation of the Lie algebra. We shall use also the components^ of>l with respect to the basis ta,
By means of A^ one defines the covariant derivative of the field \p,
where T(ta) = Ta is a representation of the generators ta to which the field \p belongs. If the covariant derivative vanishes for some direction /,
we say that the field 1// at the point x^ + e/M is a parallel transport of the field i// at the point x^ for infinitesimal e. The parallel transport around the infinitesimal loop in general changes the value of \p. The measure of this change is called the curvature and is given by
where aM" is an infinitesimal surface bounded by the loop and
The change of the frame, with respect to which we define the components of the i// fields, can be considered as the following change in the field A^,
where e is a function e(x) belonging to the Lie algebra and describing the in finitesimal change of the frame. The transformation (6.2) is called the infinite simal gauge transformation. It can be integrated to define the gauge group
112
34
L.D. Faddeev
which is formally written as a product of the group G taken once for each point in space-time. The group Gg auge plays the role of the group of the gen eral coordinate transformations. The general relativity means that the fields A^ and A'^ which can be obtained one from another by means of the gauge trans formation are physically indistinguishable. In other words it means that there exists no physical way to define a preferred frame to label the components of the charged field \p. This principle implies important conditions for the dynamics of the field A . Not all the components of this field are dynamical variables. Some of them are simply the group parameters. More specifically, the Lagrangian of this field is to be invariant with respect to the gauge transformations. This means that the time derivatives of some components of the fields will not enter into the Lagrangian, because the gauge transformations contain arbitrary functions of time. This in turn will lead to the singularity of the Lagrangian in the sense of Hamiltonian dynamics, i.e. to the fact that the canonical coordinates and momenta for the field A are not independent. The implications for the quan tization problem are evident. After these general considerations let us face the concrete problem. We con sider for simplicity the case of the pure gauge field. The Lagrangian introduced by Yang and Mills can be written in the second order form
or in the first order form
where in the former expression it is understood that F is to be written in terms of A according to (6.1) whereas in the latter expression .4 and F are independent variables. Using the 0 + 3 notations^ = (0,k), v = (Q,l), etc., k,l= 1,2,3 we can re write the last expression in the form
where
and we have made use of the time-independent equations of motion
113
Functional methods
35
to express the Fik variables in terms of Ak. We see from (6.5) that
can be considered as canonically conjugate pairs whereas Ag has no conjugate momentum and can not be expressed in terms of other variables because it enters the Lagrangian linearly. Introducing the Poisson brackets
it is easy to check that
and
All this means that the Yang-Mills Lagrangian exemplifies the so-called gen eralized Hamiltonian form of dynamics. I shall explain this notion in the ex ample of a system with a finite number of degrees of freedom. Let us be given a phase space r 2 „ with the canonical coordinates