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!
0. Notice that the eigenvalues of the unperturbed operator Ho(k) coincide with yt +
E rt, so that
k121, ryt
(3.2) No(P21; k) = N[B(p, k)]. Formula (3.1) shows that (N[B(p, )]) = wd pd, where wd is the volume of the unit ball in 118d. We need lower bounds on the variation of the number N[B(p, k)] when the parameter k varies. To describe this variation, we first introduce the following integral norms describing the deviation of the above number from its average value: -Wdt1-1PdIIp' PC
ap(p) = IIN[B(P)]
[1,00),
a. (P) = sup I N[B(P, k)] - wd µ-1 pd I. k
IIp denotes the norm in LP(Ot). Another convenient quantity is the 6variation WO (A; b); see (2.2). By (3.2) for all A > 0 and arbitrary 6 : 181 < A, we Here II
-
have
Wo(A, S) = max N[B((A - 6)2'1 , k)] - min N[B((A + 6)2'1 , k)].
This 6-variation was first introduced in [40] for l = 1. The question of estimating the quantities ap(p) and Wo(A, 6) falls in the same category with the famous circle problem (see, e.g., [7, 25]). The circle problem is usually associated with estimating the number a,, from above and has been investigated quite well. We start with the classical bound initially proved by Sierpinski [36] for d = 2 and generalized to d > 3 by Landau [28]:
(P) = 0(p2,,) xl =
l
2
+
1=1,2,....
(3.3)
l + 1, Today more precise results are known. For relevant results in dimensions d = 2 and 3, 4, we refer to papers by Huxley [15] and Kratzel and Nowak [26] respectively. For dimensions d > 5, Gotze [11] (see also the preceding paper [3]) established the universal bound 2
aoo(P) = Q(Pd-2)
(3.4)
If one aims to obtain more precise estimates, one is led to differentiate between the rational and irrational lattices. It was shown in [11] (see also [3]) that for d > 5, o(Pd-2)
aoo(P) =
(3.5)
BETHE-SOMMERFELD CONJECTURE
389
if and only if the lattice r is irrational. Furthermore, in [16], examples of irrational latices were constructed, for which
N[B(P, 0)] =
Pd+C(P2
µ for any c > 0. For our purposes we need lower bounds for a1(p) and Wo(A, S). For a sufficiently large po > 0, the quantity al (p) satisfies the bound d-1 al (P) >- csp 2 -b, VP ? Po,
(3.6)
where
S=
r0, arbitrary positive,
Sl
if d 1 (mod4); if d = 1 (mod 4).
Here and below we denote by C, c (with or without indices) various positive conCp(d-1)/2 stants whose value is of no importance. As was shown in [23], a2(p) < for all d, so that (3.6) is sharp if d 1 (mod 4). For these values of d, the estimate (3.6) can be easily derived using an argument due to Dahlberg and Trubowitz; see [4] and also [13] (A similar idea was also used by Skriganov; see [39, Lemma 12.2].) This argument is so elementary that we repeat it here. Denote N(p, k) = N[B(P; k)].
Just as in [4] and [23], we easily conclude that for any y E r N(P; -Y)
f N(p, k)eZrykdk
=f
e
dkOkIP
and, in particular, N(P; 0) = (N(P)) = wd pd
Observe that 2a1(P) = 2
f
IN(P, k)-N(P; 0) Idk > (IN(P; y)I+IN(P; 2'Y)I),
dy E r\{0}. (3.7)
Computing the Fourier coefficient, we obtain that (27r)d/2y-d/2Pd/2Jd/2(P`Y),
`Y = I'YI > 0. N(p;'y) = We point out the following elementary property of Bessel functions:
2v 0 1 (mod4)
(3.8)
for all sufficiently big z > 0. Indeed, the Bessel function has the asymptotics (see [1])
Jv(z)
z g(z) + O(z-3/2),
with
g(z) = sin (z + a7r), a
= - 2v4- 1
The required estimate will be proved if we show that g(z) I + Ig(2z) I > c,
z > zo
A. V. SOBOLEV
390
for some zo > 0. The roots of g(z) and g(2z) are -a7r + 7rn and -a7r/2 + 7rm/2, m, n E Z respectively. Since a is not an integer, these roots never coincide. This proves (3.9) and (3.8). Now (3.8) implies that N(P; y) + IN(P; 2 Y) ? cp d21,
(3.10)
so that (3.7) immediately yields the required lower bound (3.6) for d 54 1 (mod 4).
For the dimensions d = 1 (mod4), (3.6) was established for the first time in [31]. In [31] it was also proved that there exists a sequence p, , oo such that d-1
Cpn2 (log
0,1(Pn)
Pn)_Q
with a > 0. This shows that (3.6) cannot be improved in the power scale. However, it can be made more precise in the log-scale: THEOREM 3.1. Let d > 2. Then for some positive constants A, M > 0, one has
a1(P) > MAPd20(P), for all sufficiently large p > 0, where 0 is defined in (2.3).
(3.11)
The above estimate for d = 1 (mod 4) was proved by Konyagin, Skriganov, and Sobolev in [24], and for d 1 (mod 4), it simply repeats (3.6). Remembering that for any bounded function f on (fit with the property (f) = 0, one has 2µmin f(k) < -11f Ill, 2pmkax f(k) > 11f ill,
we obtain from (3.11) and (3.2) that Imaxk No (P21; k) mink N o (P21; k)
>
µ-1 wd Pd + MP d 21 0(P)
(3.12)
µ-1 wd Pd - MP d z l 0(P)
From these we can easily derive the following lower bounds for the 8-variation:
COROLLARY 3.2. Let F C Rd, d > 2 be an arbitrary lattice.
Then for all
sufficiently large p > po(r) and all 6 E [0, bo(p)], where So(p) = MdWI
p21
d
d210(P),
we have the bound Wo(P21, S) > MP d 21 O(P).
Using the formulas (2.1), one immediately obtains (2.4) for the functions m(A) and ((A),.\ = p21 for the unperturbed operator Ho. For rational lattices, the lower bounds for the 6-variation can be made more precise. The next three theorems are proved in [40]. From now on we consider the case 1 = 1 only. The positive constants Ao, So, c featured in the theorems below depend only on F.
THEOREM 3.3. Let F C Rd be a rational lattice, and let d > 5,1 = 1. Then there are three positive constants So, )to and c, such that for all S E [0, So] and all A > Ao, we have Wo(A, 6) >
cAd22.
(3.13)
BETHE-SOMMERFELD CONJECTURE
391
According to (3.4), for all 6 E R, W,, (A; 6) = O(A T'),
d > 5,
(3.14)
so that (3.13) is sharp. The next theorem deals with the four-dimensional case: THEOREM 3.4. Let r C 1184 be a rational lattice, and let 1 = 1. Then there are three positive constants So, A0 and c, such that for all 6 E [0, 60] and all A > A0, we have
W0(A, 8) > cA(log log A)-1.
(3.15)
It is not yet clear whether one can get rid of the log log-factor in (3.15) for general rational lattices. However, for the case of a cubic lattice r, this can be done:
THEOREM 3.5. Let I = (2irZ)4, 1 = 1. Then for each 8 E [0, 2-15], all sufficiently large A > A0 > 0 and some c > 0, one has the bound W0(A, 8) > cA.
(3.16)
The main tool in the proof of the above theorems are asymptotic formulas for the number of representations of integers by positive definite quadratic forms with integer coefficients.
In [38] Skriganov obtained a number of conditional results which relate the behaviour of ,,(p) as p - oc, with the number of gaps in the spectrum of the
Schrodinger operator H = -A + V with a periodic perturbation V, which is not supposed to be local. In particular, Theorem 15.3 from [38] combined with the estimates (3.4), (3.5), gives the following proposition.
PROPOSITION 3.6. Let d > 5. Then there exists a number t > 0 and a (nonlocal) periodic perturbation V with the norm 11VII = t such that the number of gaps in the spectrum of H is infinite. If, in addition, the lattice is irrational, then for any t > 0, there exists a (nonlocal) periodic perturbation V such that II V II = t and the number of gaps in the spectrum of H is infinite. This shows that the lattice properties are relevant for the band-gap structure of the operator -A + V, although one should remember that the perturbation V in the above proposition is not local.
3.2. Perturbation Theory. The results on lattice points counting in the previous section lead to lower bounds for the overlap multiplicity and overlap length for the unperturbed operator. In this subsection we provide necessary perturbation-
theoretic results which allow one to extend the mentioned lower bounds to the perturbed operator. The following estimate is crucial for the proof of Theorem 2.2: PROPOSITION 3.7. Let d > 2, 21 > 1. Suppose that V E C°°(O) and that
10 V(x)dx = 0. Then II N(P21) - No(P21)1
for sufficiently large p.
Cpd+l-41 log
(3.17)
P,
(3.18)
A. V. SOBOLEV
392
The bound (3.18) was derived by Karpeshina in [20] as an intermediate result for obtaining the corresponding estimate for the integrated density of states D(p21) = (N(p21)). Indeed, the unperturbed density of states Do(p21) coincides with wd pd, so that (3.18) leads to D(p21) = wd pd + pd+1-410(log P),
P - oo. Actually, the conditions in [20] require only a finite smoothness of the potential, but we do not go into these details here and refer to the original source instead. Note also that for 1 = 1, a similar asymptotics for D(p2) was established in [13] with the remainder estimate O(pd-3+E) with arbitrary E > 0. For rational lattices, the reduction to the unperturbed operator is implemented in the following theorem:
THEOREM 3.8 ([41]). Let d > 2, 1 = 1, and let V be a continuous real-valued 1-periodic function satisfying (3.17). Then for any 6 E R, e > 0, 7o (A, 8 +
E) - O()xd-1) < W(A; 8) < WO (A; 8 - E) + O()"d-1)
(3.19)
for large A, where W is defined in (2.2) and x1 in (3.3). Note that the condition (3.17) in Proposition 3.7 and Theorem 3.8 is imposed for convenience only. For the general V, the results can be recovered in the obvious way.
To be precise, the result obtained in [41] gives a two-sided bound for m(A) of the form (3.19) with 8 = 0. However, a minor modification of the proof in [41] immediately leads to (3.19). We do not give the full proof of this theorem, but make a few comments on its pivotal points. The fact that the presence of the potential V has a relatively small effect on the 8-variation W admits a perturbation-theoretic explanation. Assume for simplicity
that
V(x)
veez0X BEO
where O C Ft is a finite subset, and ve = v=e. The eigenvalues of the unperturbed Hamiltonian Ho(k) are given by A(') (k) _ (w + k)2, w E It. The analysis of these eigenvalues under the perturbation V is dramatically different for d = 1 and d > 2. If d = 1, then the standard perturbation theory yields a complete asymptotic expansion of the eigenvalues. On the contrary, for d > 2, the unperturbed eigenvalues split in two groups behaving differently under the perturbation V, depending on whether or not the value w is close to the set A, which is the union of the hyper-
planes { E Rd : O(2 + 0) = 0}, 0 E O. The eigenvalues A(') (k) with w away from A are well separated and can be more or less completely described by the perturbation theory; see Feldman, Knorrer, and Trubowitz [8] and Karpeshina's book [19]. The eigenvalues with w close to A move by a quantity of order JIV II under the perturbation V. This effect is due to the small divisors arising when the eigenvalues A(") (k) get close to one another. In the relevant literature these exceptional eigenvalues are sometimes called resonant, unstable or singular, see [9, 19]. It was shown in [9, 19] that their behaviour can be described by means of some effective one-dimensional Schrodinger operators. When proving Theorem 3.8, one needs to estimate the number of those points in the resonant set that can affect the counting function N(A; k). Since A is a union
BETHE-SOMMERFELD CONJECTURE
393
of hyperplanes, this problem amounts to counting lattice points inside balls of the reduced dimension d - 1. This leads to the error term O(A'"d-1) by virtue of the classical bound (3.3). In view of (3.13), (3.15), and (3.16), for rational lattices and d > 4, this error is dominated by the 8-variation Wo (A; 8), and hence the bound (3.19) leads to Theorem 2.4. For the sake of comparison, note that the estimate (3.18) also requires the study of the resonant set, but it involves only the average number of lattice points, and hence one can use the formula (3.1) without appealing to number-theoretic results.
4. Proof of Theorems 2.1, 2.2, 2.4, and 2.5 4.1. Proof of Theorems 2.1 and 2.2. We prove only lower bounds for the multiplicity m(A). Introduce the notation Ti(P) = II N(P21) - No(P21)II i.
Si(P) = IIN(P2d) - (N(P21))II1,
Before proceeding with the proof, we establish a few elementary estimates. First of all, observe that S1(P) ? al (P) - Tl (P) - I (N(P21) - (No (P21)) I
> ai (P) - 2T1(P)
As in the proof of (3.12), we get for A = p21, 2p max N(A; k) > (N(A)) + Si (p) W d Pd + Sl (P) - Tl (P)
Wd Pd + al (P) - 3T1(P)
Similarly,
2p min N(A; k)
Wd Pd - al (P) + 3T1(P).
k
The two previous bounds, together with (2.1), imply that [al (P) - 3Ti (p)].
m(P21) >
(4.2)
It
We emphasize that the above bound contains only L1-norms. Assume that the conditions of Theorem 2.1 are satisfied, i.e., V is a bounded periodic perturbation. Denote v = II V II . By a straightforward perturbation argument, N(A; k) - No (A; k)
No(A+ v; k) - No (A; k),
No (A; k) - N (A; k)
IN(A;k) -No(A;k)I < No(A+v;k) -No(A-v;k), so that, by (3.2), Ti(p)
(N[B(P+)]) - (N[B(P-)]) = Wd(P+ - P ),
p± = (A ± v) z
Consequently,
Ti(P) =
O(vpd-21).
(4.3)
A. V. SOBOLEV
394
Thus the bound (4.2) in combination with (4.3) and (3.11) implies that m(p21) > cpd2ll o(P)
-
Cvpd-21
Under the condition 41 > d+ 1, the first term dominates, which immediately proves the proclaimed lower bound for Suppose now that the conditions of Theorem 2.2 are satisfied. Now we need to use the bound (3.18), which, after substitution into (4.2), leads to m(p21).
m(p21) >
cPd
1
-Cpd+1-41 loge
0(P)
The condition 81 > d + 3 ensures that the first term dominates again, which leads to the bound (2.4) for m(p21).
REMARK 4.1. The proof of the Bethe-Sommerfeld conjecture for l = 1 given in [13] uses a variation of the method described above. Namely, instead of estimating Sl (p) from below by a1(p), as in (4.1), Helffer and Mohamed find the asymptotics of the generalized density of states, i.e., of Fourier coefficients N(A; -y)
= fo N( A; k)ei7kdk
for arbitrary y c r as A -+ oo:
N(p2; ^t) = No(P2; Y) + O(pd which, in view of (3.10) implies
3+E),
IN(P 2;.y)I+IN(p2;2^t)I>cpdzl
,
VC
> 0,
d=2,3,4.
(4.4)
Then using the same argument as in the proof of (3.6) for d # 1 (mod 4) (see above), they obtain for S1 (p) the bound S, (p) > cpd21,
d= 2,3,4
(4.5)
for sufficiently large p.
4.2. Proof of Theorems 2.4 and 2.5. Since xd_1 < (d - 2)/2, the lower bounds in Theorem 2.4 follow immediately from Theorem 3.8 and estimates (3.13), (3.15), (3.16). The upper bound in (2.5) is a consequence of (3.19) and the upper bound (3.14). To prove (2.7), note that in view of (3.5), lim lim sup A- d22 VV Wo (\, c) = 0. C-0 A- cc Now, using (3.19) with 6 = 0, we obtain (3.14).
5. Results for Other Operators 5.1. Periodic Magnetic Schrodinger Operator. Consider now the operator with the magnetic potential a = (al, a2, ... , ad):
H = (-iV - a)2 + V, where a1, a2i ... , ad, V are r-periodic real-valued C°°-functions. As shown by Herbst
and Hempel in [14], under certain assumptions on the magnetic potential, there are gaps in the the spectrum of the operator H. Mohamed proved in [29] that there can be only finitely many of them if d = 2. The method is a generalization
BETHE-SOMMERFELD CONJECTURE
395
of the approach put forward in [13], and it is based on the asymptotics of the Fourier coefficients of the counting function N(A; k); see Remark 4.1. Applying microlocal techniques developed in [13], Mohamed proved for d = 2 the estimate of the form (4.4), which allowed him to deduce (4.5), leading to the justification of the Bethe-Sommerfeld conjecture. Another proof of the conjecture for d = 2 was given by Karpeshina in [21], relying on a modification of the methods described in [19] for the operator (1.1). The approach involves subtle asymptotic formulas for the Bloch eigenfunctions and eigenvalues of the perturbed operator. A different spectral picture emerges when, instead of the periodicity of the magnetic potential, one assumes that the magnetic field B = curl a is periodic. Assume
that B = const
0. Remembering that for d = 2 the spectrum of the operator
Ao = (-iV - a)2 consists of isolated equidistant eigenvalues of infinite multiplicity (Landau levels), it is clear that at least for sufficiently small perturbation V, the spectrum of A = A0 + V will have infinitely many gaps. If d = 3, the spectrum of A0 fills the half-line [IBI, oo) and it makes sense to ask whether the perturbed operator A has finitely many gaps in the spectrum. Under the assumption that the flux of the magnetic field is rational, and the potential V is sufficiently small, the number of gaps was shown to be finite by Geiler, Margulis, and Chuchaev in [10].
Later, Elton in [6] removed the restriction on the size of the perturbation. The idea follows the proof of Theorems 2.1 and 2.2 above, but the technical difficulties are more substantial. In particular, one needs a new bound of the form (3.6) for the counting function of the Floquet fibres A0(k) of the unperturbed operator A0. Another ingredient is the following estimate similar to (3.18): T1(P) = O(Pl-E),E > 0.
The estimate of the type (3.6) was found in [10], and it is sufficient to justify the Bethe-Sommerfeld conjecture for small perturbations. This is the above upper bound for T1 that allows one to extend the result to perturbations of arbitrary size, and its proof is the main focus of [6].
5.2. The Limit-Periodic Operators. We say that a sublattice A C r has
index p = 1, 2, ... , if I OA I = pI Or I. Denote the set of all sublattices of index p by 91p. We say that the potential V is limit-periodic if it has the form 00
V = E > VA, P=1 AE91p
where all VA's are A-periodic, and the series converges absolutely in the sup-norm.
We are interested in the spectrum of the perturbed polyharmonic operator (1.1) with a real-valued limit-periodic potential V. For d = 1 and 1 = 1, this problem has been studied relatively well; see, e.g., Avron and Simon [2] or Pastur and Tkachenko
[33]. It was shown that generically the spectrum is a Cantor set. In the multidimensional case, it is natural to ask whether the spectrum of the operator (1.1) contains a half-line, i.e., there is a number Ao E 11 such that [)o, oo) C a(H). (5.1) With some mild conditions on the decay of VA asp -- oc, supposing that 81 > d+3, d # 1 (mod 4) (cf. Theorem 2.2) and that V is periodic in one direction, Skriganov and Sobolev [42] showed that the spectrum indeed contains a half-line. The proof is
A. V. SOBOLEV
396
based on the study of the operators (1.1) with periodic potentials Vs, s = 1, 2, ... , approximating V in the sup-norm. As the lattice of periods for Vg grows together
with s, in general, the spectral bands of the approximating periodic operators shrink, which makes it harder to control their overlap. However, since one chooses all potentials VS to be periodic in one direction with a period independent of s, one
finds that the spectral bands and their overlap lengths stay well separated from zero as s - oc. This guarantees (5.1) for some Ao. For the operator (1.1) with l > 6 and d = 2, the inclusion (5.1) was proved by Karpeshina and Lee in [22] under the assumptions that the potential V is composed of periodic potentials with doubling periods, and that the size of VA decays superexponentially. Using the methods developed for periodic operators in [19], in combination with the ideas of the KAM method, the authors proved that for every sufficiently large A, there is a family of generalized eigenfunctions that are close to plain waves, thereby proving that the spectrum contains a semi-axis.
References [1] M. Abramowitz, I. A. Stegun, Handbook of Mathematical Functions, Dover, 1972. [2] J. Avron, B. Simon, Almost periodic Schrodinger operators. I. Limit periodic potentials, Comm. Math. Phys. 82 (1981), 101-120. [3] V. Bentkus, F. Gotze, On the lattice point problem for ellipsoids, Acta Arithm. 80 (1997), 101-125.
[4] B. E. J. Dahlberg, E. Trubowitz, A remark on two dimensional periodic potentials, Comment. Math. Helvetici 57 (1982), 130-134. [5] M. S. P. Eastham, The spectral theory of periodic differential equations, Scottish Academic Press, 1973.
[6] D. Elton, The Bethe-Sommerfeld conjecture for the 3-dimensional periodic Landau operator, Rev. Math. Phys. 16 (2004), 1259-1290. [7] P. Erdos, P. M. Gruber, J. Hammer, Lattice Points, Longman, 1987. [8] J. Feldman, H. Knhrrer, E. Trubowitz, The perturbatively stable spectrum of a periodic Schrodinger operator, Invent. Math. 100 (1990), 259-300. [9] , Perturbatively unstable eigenvalues of a periodic Schrodinger operator, Comment. Math. Helvetici 66 (1991), 557-579. [10] V. A. Geller, V. A. Margulis, I. I. Chuchaev, Spectrum structure for the three-dimensional periodic Landau operator, Algebra i Analiz 8(3) (1996); English transl. in St. Petersburg Math. J. 8 (1997), 447-461. [11] F. Gotze, Lattice point problems and values of quadratic forms, Invent. Math. 157 (2004), 195-226.
[12] E. L. Green, Spectral theory of Laplace-Beltrami operators with periodic metrics, J. Diff. Eq. 133 (1997), 15-29. [13] B. Helffer, A. Mohamed, Asymptotics of the density of states for the Schrodinger operator with periodic electric potential, Duke Math. J. 92 (1998), 1-60. [14] R. Hempel, I. Herbst, Strong magnetic fields, Dirichlet boundaries, and spectral gaps, Comm. Math. Phys. 164 (1995), 237-259. [15] M. N. Huxley, Exponential sums and lattice points, III, Proc. London Math. Soc. III 87 (2003), 591-609.
[16] V. Jarnik, Uber Gitterpunkte in mehrdimensionalen Ellipsoiden, Math. Ann. 100 (1928), 699-721.
[17] Yu. E. Karpeshina, Perturbation theory for the Schrodinger operator with a periodic potential, Trudy Mat. Inst. Steklov 188 (1990), 88-116; Engl transl. in Proc. Steklov Math. Inst., Issue 3 (1991), 109-145. [18] , Analytic perturbation theory for a periodic potential, Izv. Akad. Nauk SSSR Ser. Mat., 53 (1989), 45-65; English transl. in Math. USSR Izv. 34 (1990), 43-63.
BETHE-SOMMERFELD CONJECTURE [19] [20]
397
, Perturbation Theory for the Schrodinger operator with a Periodic Potential, Lecture Notes in Math. 1663, Springer, 1997. , On the density of states for the periodic Schrodinger operator, Ark. Mat. 38 (2000), 111-137.
[21]
,
Spectral properties of periodic magnetic Schrodinger operator in the high-energy
region. Two-dimensional case, Comm. Math. Phys. 251 (2004), 473-514. [22] Yu. Karpeshina, Y.-R. Lee, Spectral properties of polyharmonic operators with limit-periodic
potential in dimension two, preprint arXiv:math-ph/0601008.
[23] D. G. Kendall, R. A. Rankin, On the number of points of a given lattice in a random hypersphere, Quart. J. Math. Oxford 2 (1953), 178-189. [24] S. Konyagin, M. M. Skriganov, A. V. Sobolev, On a lattice point problem arising in the spectral analysis of periodic operators, Mathematika 50 (2003), 87-98. [25] E. Kratzel, Lattice Points, Kluwer, 1988. [26] E. Kratzel, W. G. Nowak, Lattice points in large convex bodies, II, Acta Arithmetica 62 (1992), 285-295.
[27] P. Kuchment, Floquet Theory for Partial Differential Equations, Birkhauser, 1993. [28] E. Landau, Zur analytischen Zahlentheorie der definiten quadratischen Formen (Uber die
Gitterpunkte in einem mehrdimensionalen Ellipsoid), Sitzber. Preuss. Akad. Wiss. 31 (1915), 458-476. [29] A. Mohamed, Asymptotics of the density of states for the Schrodinger operator with periodic
electromagnetic potential, J. Math. Phys. 38 (1997), 4023-4051. [30] L. Parnovski, On the Bethe-Sommerfeld conjecture, preprint 2005. [31] L. Parnovski, A. V. Sobolev, On the Bethe-Sommerfeld conjecture for the polyharmonic operator, Duke Math. J. 107 (2001), 209-238. , Lattice points, perturbation theory and the periodic polyharmonic operator, Ann. [32] H. Poincare 2 (2001), 573-581. [33] L. A. Pastur, V. A. Tkachenko, Spectral theory of a class of one-dimensional Schrodinger operators with limit-periodic potentials, Trans. Moscow Math. Soc. 51 (1989), 115-166.
[34] V. N. Popov, M. Skriganov, A remark on the spectral structure of the two dimensional Schrodinger operator with a periodic potential, Zap. Nauchn. Sem. LOMI AN SSSR 109 (1981), 131-133 (Russian). [35] M. Reed, B. Simon, Methods of Modern Mathematical Physics, IV. Analysis of Operators, Academic Press, 1975. [36] W. Sierpinski, 0 pwenem zagadnieniu z rachunku funkcyj asymptotychnych, Prace Mat.Fiz. 17 (1906), 77-118; in French: Sur un probleme du calcul des fonctions asymptotiques, Oeuvres Choisies, (S. Hartman, A. Schinzel, eds.), Warszawa: Editions Scientifiques de Pologne, 1974.
[37] M. Skriganov, Finiteness of the number of lacunae in the spectrum of the multidimensional polyharmonic operator with a periodic potential, Mat. Sb. 113 (155) (1980), 131-145; English transl. in Math. USSR Sb. 41 (1982), 115-125. [38]
,
Geometrical and arithmetical methods in the spectral theory of the multi-
dimensional periodic operators, Proc. Steklov Math. Inst. 171 (1984), 121 pp. [39] , The spectrum band structure of the three-dimensional Schrodinger operator with periodic potential, Invent. Math. 80 (1985), 107-121. [40] M. M. Skriganov, A. V. Sobolev, Variation of the number of lattice points in large balls, Acta Arithmetica, 120 (2005), 245-267. , Asymptotic bounds for the spectral bands of periodic Schrodinger operators, Algebra [41] i Analiz 17 (2005), 276-288; English transl. in St. Petersburg Math. J. 17 (2006), 207-216. [42] , On the spectrum of a limit-periodic Schrodinger operator, Algebra i Analiz 17 (2005), 164-189; English transl. in St. Petersburg Math. J. 17 (2006), 815-833. [43] O. A. Veliev, Asymptotic formulas for the eigenvalues of a periodic Schrodinger operator and the Bethe-Sommerfeld conjecture, Funkt. Anal. i Prilozhen. 21 (1987), 1-15; English transl. in Functional Anal. Appl. 21 (1987), 87-99. [44] , On the Spectrum of Manydimensional Periodic Operator, Functions Theory, Functional Analysis and Their Applications, Kharkov University, 49 (1988), 17-34 (Russian). , Asymptotic analysis of the periodic Schrodinger operator, preprint mp-arch 05-90. [45] [46] , On the polyharmonic operator with periodic potential, preprint mp-arch 04-204.
398
A. V. SOBOLEV
[47] N. N. Yakovlev, Asymptotic estimates of the densities of lattice k-packings and k-coverings, and the structure of the spectrum of the Schrodinger operator with a periodic potential, Dokl. Akad. Nauk SSSR, 276 (1984); English transl. in Soviet Math. Dokl. 29 (1984), 457-460. [48]
, On spectra of multi-dimensional pseudo-differential periodic operators, Vestn. Mosk. Univ. Ser. 1 Mat, Mekh, No 3 (1985), 80-81 (Russian).
SCHOOL OF MATHEMATICS, UNIVERSITY OF BIRMINGHAM, EDGBASTON, BIRMINGHAM B15 2TT, U.K.
E-mail address: [email protected]
Electric and Magnetic Fields, Semiclassical Limit
Proceedings of Symposia in Pure Mathematics Volume 76.1, 2007
Recent Developments in Quantum Mechanics with Magnetic Fields LAszlo Erdos Dedicated to Barry Simon on his sixtieth birthday ABSTRACT. We present a review of the recent developments concerning rigorous mathematical results on Schrodinger operators with magnetic fields.
CONTENTS 1.
2.
3.
4. 5.
Introduction Basic Qualitative Properties Quantitative Properties of One-Body Operators Many-Body Magnetic Systems Random Schrodinger Operators with Magnetic Fields
References
1. Introduction The mathematical formulation of quantum mechanics, given by Schrodinger, Pauli and Dirac, has posed an enormous challenge: can mathematics, with its own tools and standards, rigorously justify or even predict physical phenomena of the quantum world? Similarly to the development of the differential and integral calculus, strongly motivated by Newton's classical mechanics, new mathematical tools
have been created (most notably by von Neumann, Weyl, Wigner and later by Kato). Functional analysis, representation theory and partial differential equations would have been much poorer mathematical disciplines without quantum mechanics.
Electromagnetic fields play a central role in quantum physics; their rigorous inclusion in the theory is certainly one of the key goals of mathematical physics. 2000 Mathematics Subject Classification. 81Q10, 81Q70. Key words and phrases. Schrodinger operator, Pauli operator, magnetic fields. Partially supported by EU-IHP Network "Analysis and Quantum" HPRN-CT-2002-0027 and by Harvard University. ©2007 American Mathematical Society 401
402
L. ERDOS
Quantum electrodynamics (QED) postulates that electric and magnetic fields are to be described within a unified relativistic theory. Although the framework for QED has been clear since the 1930's, the mathematical difficulties to even formulate the theory rigorously have not yet been resolved. In the low energy regime, however, massive quantum particles can be described non-relativistically. Electric and magnetic fields, with a good approximation, can be considered decoupled. Since typical magnetic fields in laboratory are relatively weak, as a first approximation one can completely neglect magnetic fields and concentrate only on quantum point particles interacting via electric potentials. The rigorous mathematical theory of Schrodinger operators has therefore started with studying the operator H = --1- p2 + V (x) on L2 (Rd) and its multi-particle analogues. Here x E Rd is the location of the particle in the d-dimensional config-
uration space, p = -iVy is the momentum operator and m is the mass, that can be set m = with convenient units. The Laplace operator describes the kinetic energy of the2 particle and the real-valued function V (x) is the electric potential. Although both the kinetic and potential energy operators are very simple to understand separately, their sum exhibits a rich variety of complex phenomena which differ from their classical counterparts in many aspects. The mathematical theory of this operator is the most developed and most extensive in mathematical physics: the best recent review is by Simon [151]. As a next approximation, classical magnetic fields are included in the theory, but spins are neglected. The kinetic energy operator is modified from p2 to (p + A)2 by the minimal substitution rule: p r-4 p - eA and we set the charge to be e = -1.
Here A : Rd -* Rd is the magnetic vector potential that generates the magnetic field B according to classical electrodynamics. In d = 2 or d = 3 dimensions B = V x A is a scalar or a vector field, respectively. In d = 1 dimension the vector ei'P(p+A)2e-ip = p2, potential can be removed by a unitary gauge transformation, cp = f A, therefore magnetic phenomena in R1 are absent (they are present in the case of Sl).
We will call the operator (p + A)2 + V the magnetic Schrodinger operator. In general, even the kinetic energy part contains non-commuting operators, [(p + A)k, (p + A)e] # 0, and the theory of (p + A)2 itself is more complicated than that of p2 + V. The simplest case of constant magnetic field, B = const, is explicitly solvable. The resulting Landau-spectrum consists, in two dimensions, of infinitely degenerate eigenvalues at energies (2n + 1) 1 B1, n = 0, 1, .... Notice that the magnetic spectrum is characteristically different from that of the free Laplacian. The eigenfunctions are localized on a scale JBJ-1/2; this corresponds to the cyclotronic radius in classical mechanics (Landau orbits). The interaction of the spin with a magnetic field is proportional to the field strength. In the low energy regime this effect is comparable with the energy shift due to inclusion of A into p2. Since electrons are spin-! particles, the spin, in principle, should not be neglected whenever magnetic fields are considered. Nevertheless, magnetic Schrodinger operators constitute an important intermediate step to understand magnetic phenomena. The state space of a spin-! particle is L2 (Rd, C2) (in d = 2, 3) and the momentum operator is the Dirac operator, VA := o (p+A), where o = (al, 92, Q3) is the vector of the Pauli matrices. The kinetic energy is given by the Pauli operator, B,
(1.1)
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
403
and external potential may be added as before. The last identity is a special case of the Lichnerowicz formula known from spinor-geometry. In most of this review we restrict ourselves to these operators and their multi-
particle generalizations. However, we briefly mention that in dimensions higher than 3 or on configuration spaces with a non-flat Riemannian metric, the vector potential is canonically defined as a one-form, a, and the magnetic 2-form, 0 = da, is its exterior derivative. In the conceptually most general setup for the spinless case, the Hilbert space of states consists of the L2-sections of a U(1)-bundle over an orientable Riemannian manifold, M, representing the configuration space, and the momentum operator is the covariant derivative, V, on this bundle. In this formulation, the vector potential does not appear directly but the magnetic field is (i-times) the curvature 2-form of V. Proper description of the spin involves covariant derivatives on sections of a Spin-bundle with Pauli matrices replaced by Clifford multiplication [46]. In relativistic theories, electron-positron pair-creations cannot be neglected and one studies the full relativistic Dirac operator, a (p + A) + (3m, where (a, 0) is the vector of the Dirac matrices and m is the mass. Due to the lack of semiboundedness
of the Dirac operator, its definition, even without a magnetic field, is a complex issue that is not yet satisfactorily resolved in the many-body situation ("filling the Dirac sea"). We will not pursue this direction here since the current research focuses more on the non-magnetic aspects of the Dirac operator.
A consistent quantum theory requires to quantize the electromagnetic field as well. Ideally, this should be done within the framework of the Dirac operator (relativistic QED) but this problem is beyond the reach of the current techniques. A more tractable model is the non-relativistic QED, where quantized electromagnetic field is introduced in the Pauli operator, i.e., pair-creations are neglected. This overview gives an admittedly biased summary of a few recent key results involving magnetic Hamiltonians. Many people have contributed to these questions and a selection was unavoidable; the author apologizes to everyone whose work has been left out. The choices reflect the author's taste and the pressure of the editors to keep to the page limit. In Section 2 we present results related to the proper definitions of these operators. In Section 3 we discuss one-particle spectral theory, including Lieb-Thirring type bounds and semiclassical methods. In Section 4 we consider multi-particle problems, including stability of matter, large atoms and scattering. Finally, Section 5 is devoted to random Schrodinger operators with magnetic fields. Barry Simon was undoubtedly one of the initiators and most important contributors to the endeavor to put Schrodinger operators on solid mathematical ground. His work was especially pioneering in the theory of magnetic fields. Among many of his achievements in this area, I would just mention here those two that had the biggest impact on my own work. Barry was the first who systematically exploited path integral methods for magnetic fields upon an initial suggestion of Nelson (see e.g., [152]). Secondly, his seminal papers with Avron and Herbst [5] have become the classical reference "handbook" about magnetic fields. This overview is dedicated to his 60th birthday.
404
L. ERDOS
2. Basic Qualitative Properties 2.1. Definitions. Along the development of the rigorous theory of Schrodinger operators without magnetic fields, it was apparently Kato who first initiated the natural program to extend this theory to the most general magnetic fields. The unique self-adjoint extension of the operator (p+A)2+V without any growth condition on A was shown in 1962 by Ikeda and Kato [88]. This result indicated that magnetic operators should not simply be viewed as second order differential operators with variable coefficients. For most mathematical purposes it is misleading to look at (p + A)2 as p2 + A p + p A + A2. The A-field plays a special role in magnetic problems: it balances the derivative of the phase of the wave function. This effect is inherently present in the form (p + A)2. Kato proved his celebrated distributional inequality, Al-%I > Re [sgn 0V], for any 0 E L', 0I E Li , in 1973 [93]. Simon realized its connection to the semigroup inequality, leto'I < e"l'bl in 1977 [146]. A more general abstract setup was considered in [147], and independently in [80], leading to the magnetic versions of these inequalities. For regular vector potential, a simple proof of the semigroup diamagnetic inequality, le-t(P+A2,)I < l
,
(2.2)
via the Feynman-Kac formula was given in Simon's paper [146] quoting an argument of Nelson in a private communication. It was apparently Nelson who pointed out the probabilistic approach to Simon (see the history in Simon's book [152]), but the real power behind the rigorous path integral method for magnetic fields was realized in a series of papers of Simon and collaborators [5] (see also [25]). More singular vector potentials were considered with analytic methods in [94]. Finally, in his seminal paper [148], Simon gave a simple proof of the diamagnetic inequality (2.2) where the operator H = (p+A)2 was defined under the most general conditions, namely for A E Li, and for H defined as the operator associated with the maximal quadratic form. The domain of the maximal form contains all z/) E L2 with (p + A)V) E L2 in distributional sense. Using (2.2) and semigroup smoothing, Simon showed that the CO' is a form core for H. A non-negative potential V E L'10i can be added to H without any difficulty. The optimal conditions for CO' being the operator core for H are A E L I., and divA E Ll . This was conjectured by Simon and proved by Leinfelder and Simader, [102]. Leinfelder [101] has also showed the unitarity equivalence under any gauge transformation, A - A+V gyp, that stays within the above classes. Again, a non-negative potential V E L2,,i , can be added to H without any difficulty and the Leinfelder-Simader theorem extends to a certain class of negative potentials as well (V = V1 + V2 < 0, V1, V2 E Li C, Vi (x) _> -(const) 1x12, V2 bounded relative to
-A with a bound smaller than one). The most general conditions on potentials with non-trivial negative part, V_ 0, are hard to use directly. The typical argument uses the KLMN theorem (Theorem X.17 [139]) that defines self-adjoint operators by adding a relative form bounded perturbation (with bound less than 1) to a semi-bounded closable quadratic form. The boundedness of V_ relative to (p + A)2 + V+ is, however, hard to check. With the help of the diamagnetic inequality, the boundedness of V_ relative to p2 + V+
is sufficient. We recall that V_ being in the Kato class, V_ E 1C, implies even
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
405
infinitesimal boundedness. Using Stratonovich stochastic integrals, the FeynmanKac formula can be extended to A E Li , vector potentials if V_ is relative bounded by p2 + V+ ([84]). If one prefers to use Ito stochastic integrals, then, additionally, divA E L2 is also necessary for the Feynman-Kac formula. The definition of the magnetic operator with Neumann boundary conditions was carefully worked out recently in [87] and the proof of the diamagnetic inequality and Kato's inequality were extended to this case using the method of Simon [148]. The most general diamagnetic result for the Neumann case is obtained in [86] that uses no regularity assumptions on the domain and on V+. Similarly to the non-magnetic case worked out in the fundamental paper by Simon [150], with the help of the (magnetic) Feynman-Kac formula one can prove smoothing and continuity properties of the semigroup and its kernel. This work has been carried out in [17] with great care and with many fine details. To summarize the results, one assumes that the vector potential A belongs to the so-called (local) magnetic Kato class, i.e., A2, divA E and the potential is Kato decomposable (V+ E 1Clo,, V_ E 1C). Then the LP-semigroup is continuous in time and if A and V are approximated locally in the Kato-norm, then the approximating semigroups converge. Moreover, the Feynman-Kac formula defines a continuous representation of the semigroup kernel. 10,
The definition of the Pauli operator can be directly reduced to that of the magnetic Schrodinger operator using (1.1) and treating or - B as a (matrix-valued) potential term. However, the supersymmetric structure of the Pauli operator (at least in even dimensions) allows one to define the Pauli operator directly and for more general magnetic fields. On topologically trivial domains only the magnetic field has physical relevance. The weakest necessary condition on a B, if considered as a potential, is B E Ll . However, not every L1 field can be generated by an L oC vector potential, hence (p + A)2 might not be defined even as a quadratic form. Therefore it is desirable to define the Pauli operator directly, by circumventing the vector potential. This idea has been worked out in d = 2 dimensions in [49], where A was replaced by a scalar potential, h, satisfying Ah = B, and the Pauli quadratic form was given by
s
q('W),'W)
4J
la=(e-h,b+)I2e2h+4 f
la=(eh
-)12e-2h
V=
("J
(2.3)
This definition is applicable for any measure-valued magnetic field and it is consistent with the standard one for fields that can be generated by Li, vector potential. However, for singular fields the form core is not Co' any more. Strangely enough, a similar construction does not seem to apply for the magnetic Schrodinger operator and the higher-dimensional generalizations are also open.
2.2. Compact Resolvent, Essential Spectrum, Absolute Continuous Spectrum. A basic qualitative fact about magnetic fields is that their inclusion into the free spinless Laplacian, very roughly speaking, increases the bottom of the local spectrum by 1 B(x)1. This intuitive statement makes sense only if the spectrum of the localized operator can be defined and if B(x) is sufficiently regular. The key mathematical reason is the Lichnerowicz identity (1.1) that shows that (p+A)2 on
spinors is a non-negative operator plus -a B. Viewing this identity restricted to spinors with a spin direction opposite to the field, one obtains a useful lower bound on the magnetic Schrodinger operator. One can also see this by using the fact that
L. ERDOS
406
the components of (p + A)2 do not commute: the commutator is the magnetic field (up to a factor i). This trivial but crucial observation is the core of many results
throughout the next sections. We emphasize that this effect holds only for the spinless magnetic Schrodinger operator and not for the Pauli operator. This idea has been elaborated by several authors to investigate the location of the essential spectrum and resolvent compactness of the magnetic operators. For the Schrodinger operator it has been shown that for sufficiently regular fields, the condition that the strength of the magnetic field goes to infinity is equivalent to the compactness of the resolvent (see [72] and references to previous results therein, in particular [5]). The regularity assumptions were weakened in [144] by using functions belonging to the so-called reverse Holder class. For the Pauli operator without external potential it is conjectured that its resolvent is never compact. This has been shown for sufficiently "well-behaving" magnetic fields in [75]. Under stronger conditions about the magnetic field at infinity, the essential spectrum of the Pauli operator was also identified in [75]. In a recent work of Last and Simon [100] a different characterization of the essential spectrum was given in terms of the union of the spectra of certain limit operators at infinity. The localized eigenfunctions of the Landau spectrum in case of a constant magnetic field indicate that in d = 2 dimensions the magnetic field has a strong localization effect, while in d = 3 dimensions the free motion parallel with the field guarantees absolutely continuous spectrum. It is somewhat surprising that by a small change of the constant magnetic field, the spectrum can become purely absolutely continuous even in d = 2. This was first observed and proved by Iwatsuka [92] for magnetic fields that are translation invariant in one direction and tend to two different values at plus and minus infinity in the other direction. The classical analogue of this model is actually a very simple geometric picture. Since the cyclotronic radius depends on the field strength, the closed Landau orbits become spirals whose average velocity is non-zero and perpendicular to the gradient of the field.
A similar phenomenon can be created by an external potential in constant magnetic field or by Dirichlet boundary conditions along an edge of the sample that extends to infinity. Under suitable conditions the states can be classified as edge states and bulk states. The edge states are localized along the boundary and they give rise to pure absolutely continuous spectrum inside the Landau gaps. They carry non-vanishing chiral edge currents. This picture persists even under perturbations
with a small (possibly random) potential [57]. The edge states exhibit a level repulsion that is even stronger than that of the Gaussian ensembles expected for the usual Anderson model in the extended states regime [125].
2.3. Zero Modes and Multiplicity. The supersymmetric structure of the Pauli operator is responsible for the spectrally rigid and typically large kernel of Hp in d = 2. The Aharonov-Casher theorem [3, 27, 128] states that dim KerHp is given (essentially) by the total flux, divided by 27r, z f B. As a special case of the Index Theorem, for smooth data and on a compact manifold, it basically relies on algebraic identities. Still, in its most general form on R2 it was only recently proved in [49] (for finite total flux) and [141] (for non-negative field) using (2.3). For arbitrary field it is false [49]. In the strong field limit, under some regularity assumptions, the local density of Aharonov-Casher zero modes converge to B [34].
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
407
The elements of the kernel the Pauli operator (the so-called zero modes) in d = 2 are conceptually much easier to understand than in d = 3. Naive extensions of the two-dimensional constructions to three dimensions fail, and they even seem to indicate that there are no zero modes in d = 3. A fundamental observation of Loss and Yau [124] is that the equation
DA'b=0,
A,B,0EL2
does have a solution on R3, albeit quite complicated. This seemingly innocent fact implies, among others, that non-relativistic matter with a magnetic field cannot be stable unless the fine structure constant is sufficiently small [60]. The explicit construction of [124] does not shed much light on the conceptual origin of the zero modes. It turns out that two-dimensional Aharonov-Casher zero modes on S2 can be lifted to R3 using the Hopf map and spinor geometry [46] (see also [2, 32] for other examples). In particular, magnetic fields with arbitrary number of zero modes can be constructed. Although many zero modes are obtained in this conceptual way, still not all explicit zero modes of [124] are covered. On the other side, it is known that magnetic fields with zero modes form a slim set in the space of all magnetic fields [10], [33]. It is an interesting open question to connect the existence, or even the number of the zero modes with the geometry of the magnetic field. Currently we do not even have a conjecture for a general characterization of magnetic fields with zero mode. For the spinless magnetic Schrodinger operator no supersymmetric structure is available to analyze the ground states and even to compute the bottom of the spectrum is complicated, apart from the strong field regime (Section 3.2.1). Since the Perron-Frobenius theorem does not apply to the magnetic Laplacian, the ground state can be degenerate, although for generic field it is simple. Still, the strength of magnetic field restricts the possible multiplicity. Based upon similar observations by Colin de Verdiere on graphs, it was conjectured in [23] that on a two-dimensional manifold M, the total curvature of the line bundle, i.e., the total flux, fm IBI, gives an upper bound on the multiplicity of the magnetic ground state. This was proved in [42] modulo constants depending on the geometry of the base manifold. The same bound with constants depending only on the genus of M is still an intriguing open question. The proof in [42] relies on an upper bound on the ground state energy in terms of the total flux and this intermediate result necessarily depends on the geometry of M. The construction of an appropriate trial state uses the scalar
potential h (with Oh = B) instead of the vector potential in order to control the energy solely by the LI norm of B.
2.4. Magnetic Operators on the Lattice. The magnetic Schrodinger (and Pauli) operator can also be defined on the lattice. The magnetic field is defined on the plaquets, while the magnetic vector potential, Ab, is a function on the bonds. The magnetic translation operator along the bond b amounts to a multiplication by a complex phase eiA, in addition to the usual hopping. The field on each plaquet is the oriented sum of the vector potentials along the boundary. The field and the vector potential are defined only modulo 27r. Although this definition is very natural, the spectral properties of the lattice magnetic Schrodinger operator differ vastly from the continuous version. Even for a constant magnetic field B on a regular two-dimensional square lattice (Harper
L. ERDOS
408
operator), the spectrum exhibits a complex fractal behavior ("Hofstadter butterfly") depending sensitively on the Diophantine properties of a = B/2'ir. With a simple transformation, this operator can be reduced to the almost Mathieu operator; a simple prototype of a one-dimensional discrete Schrodinger operator with an almost periodic potential: H,,,A = .\ cos(27raD,,) + cos x .
(2.4)
The continuous Schrodinger operator with a constant magnetic field and periodic potential leads to a similar equation. The Cantor-like spectrum of Ha,a was first proven in [14] for a dense set of parameter values and later in a series of papers, Helffer and Sjostrand performed a detailed quantitative semiclassical analysis [76] initiated by Wilkinson [166] to identify a large set of parameter values a with Cantor spectrum if A = 1. With quite different techniques, Last obtained a similar result and he also computed the Lebesgue measure of the spectrum [99] for all A. Finally, the Cantor spectrum has recently been proven by Puig [137] (A 0, a is Diophantine) and by Avila and Jitomirskaya [4] for all conjectured values of the parameters: A 54 0, a irrational ("Ten martini problem," as it was named and popularized by Barry Simon).
2.5. Diamagnetism and Paramagnetism. Diamagnetism plays a crucial role in the analysis of the magnetic Schrodinger operators since it gives an easy a priori comparison of magnetic and non-magnetic operators, like (2.2). However, the apparent strength of the basic diamagnetic inequality is somewhat misleading when it comes to quantitative results. On one hand, it completely neglects magnetic effects; operators with two different but non-zero magnetic fields are not comparable with this method. In particular, diamagnetism in a strong sense, i.e., monotonicity of the energy in the magnetic field strength, does not hold in general because of the de Haas-van Alphen oscillation effect (see [77] for a rigorous proof in the weak field regime with a periodic external potential). On the other hand, diamagnetism is applicable only for the exponential statistics, tr e-13H = >i of the eigenvalues, Aj, in particular for the ground state (,3 -- oo). Going beyond these constraints is notoriously difficult and there are only a few results and many open questions. Loss and Thaller proved [123] that the heat kernel of a two-dimensional Schrodinger operator H = (p+A)2 with an arbitrary magnetic field B(x) can be estimated by e
_tH
B _ e - 47r sink Bt
(x, y) <
(=_v)2 4t
if B(x) > B(> 0). The right hand side is smaller than the free heat kernel and its exponential behavior Pz e- Bt correctly reflects a spectral shift at the ground state energy by at least B. However, it does not retain the full Gaussian off-diagonal decay of the magnetic heat kernel with a constant field. With the help of this inequality, sharp LP - L4 bounds were shown in [123]. The proof heavily uses the Gaussian character of the heat kernel of the constant field operator. Several counterexamples [39] show that this result is basically the best one could hope for: there is no strict diamagnetic comparison between two non-homogeneous magnetic fields or even between two homogeneous magnetic fields with a potential. The Gaussian off-diagonal decay cannot be fully recovered. Such type of decay apparently
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
409
requires real analyticity of the magnetic field and potential in the angular direction [37, 132, 159]. In the large field limit, the diamagnetic effect is so strong that the magnetic Schrodinger operator converges (in resolvent sense) to the free Laplacian with Dirichlet boundary conditions on the regime where the magnetic field vanishes [79]. In other words, strong magnetic fields act like Dirichlet walls, confining the electron motion to regimes where the field is zero. In contrast to the diamagnetism of the (spinless) magnetic Schrodinger operator, the Pauli operator tends to be paramagnetic. This issue was apparently raised first in [82] and Lieb proved (appendix of [5]) that the ground state energy of the Pauli operator with potential cannot increase as a constant magnetic field is turned on. However, paramagnetism fails for non-homogeneous fields [7, 71, 39]. For many-fermion systems, one studies the sum of the low lying eigenvalues of the one-body operator. This statistics is more singular, it is beyond the exponential statistics offered by the heat kernel and surprising phenomena occur. The magnetic Schrodinger operator on a square lattice turns out to be paramagnetic at half-filling. It is the maximal flux (7r on each plaquet) that minimizes the sum of the first A/2 magnetic eigenvalues on a torus of volume A. The result actually holds on any bipartite graph that has a periodicity at least in one direction. After some special cases presented in [104] and proved in [108], the general result was proven by Lieb [105]. The proof uses reflection positivity and seemingly it cannot be extended to other filling factors or to graphs without periodicity, leaving the general case as an intriguing open question. Diamagnetism for sums of the Schrodinger eigenvalues fails in the continuum as well. For a compact domain in Rd and for a constant magnetic field B, let A3 (B) be the j-th magnetic eigenvalue. The sum of the first N eigenvalues, E 1 Aj (B), may decrease by turning on a non-zero magnetic field, but it can never drop below the semiclassical bound [43]. The proof heavily relies on the homogeneity of the magnetic field. For this case a stronger diamagnetic inequality was proven: tr[Xf ((p + A)2)] < tr[Xf (p2 )] for an arbitrary non-negative, convex function f decaying to zero at infinity. Here X is the characteristic function of an open set, the operators are defined in the whole R'. This stronger diamagnetic inequality fails for non-homogeneous fields [43] but still the semiclassical bound for the eigenvalue sum is conjectured to hold.
2.6. One-Body Scattering. A short range magnetic field, IB(x)I does not substantially influence the non-magnetic scattering theory, in particular asymptotic completeness holds. Long range potentials can also be included. The most general result is due to Robert [140]; previously Loss and Thaller treated C(x)-3/2-E the IB(x)l < case in [121] and they also considered the Dirac operator
[122].
The borderline case, when B(x) decays as xJ-1 at large distances, is substantially more involved. In this case, there is no decay on the vector potential. In the simplest d = 2-dimensional, axially symmetric situation, the lack of decay leads to a dense point spectrum in the low energy region while the spectrum is absolutely continuous above an energy threshold [129]. To study the scattering, for simplicity one considers d = 2 and assumes that the field is homogeneous of degree -1, i.e., in
L. ERDOS
410
polar coordinates (r, 0) it is given by B = b(B)/r. The analogous problem for nonmagnetic scattering is the case of a homogeneous of degree zero electric potential, V(x) = U(x/lxl), where the generic classical trajectories are asymptotically straight and they select the directions of local extrema of U. The classical trajectory in the borderline magnetic case turns out to be a logarithmic spiral if the magnetic field has a definite sign. If the total flux is zero, J b(O)d9 = 0, then the trajectories are approximately straight lines (in the direction of the zeroes of b(9)); if the total flux is non-vanishing but b has no definite sign, then both types of behavior may occur. The corresponding quantum scattering follows these trajectories. Part of this picture has already been proven in the recent work [26]; the rest is work under preparation. Scattering in constant magnetic field was first studied in [5] where the asymptotic completeness of the one-particle scattering for short range and Coulomb potential was shown (a different proof given in [149]). The general long range potential was treated independently by Laba [97, 98] and Iwashita [91]. 2.7. Miscellaneous. In this section we mention two results whose non-magnetic counterparts are classics but their standard proofs are quite rigid and their extensions to magnetic fields were considerably more involved. The Rayleigh-Faber-Krahn isoperimetric inequality on the lowest Dirichlet eigenvalue of a domain of given area predates quantum mechanics. Its magnetic analogue asserts [38] that the lowest magnetic Dirichlet eigenvalue, A(S2, B), of a planar domain 1 C RZ with a fixed area and with a fixed homogeneous magnetic field B = const is attained exactly for the disk .\(1l, B) > ,A (D, B),
Area(1l) = Area(D),
D = disk.
(2.5)
The constant magnetic field plays the role of the homogeneity of the membrane in Rayleigh's original formulation of the problem. In the non-magnetic case, the minimal eigenvalue in any dimension is attained for the ball. The minimizing domain for constant magnetic fields in dimensions d > 3, however, is unknown. Isoperimetric results for the magnetic Neumann Laplacians are also unknown. Note that (2.5) does not hold for the Neumann case since the ground state has a tendency to favor non-spherically symmetric geometry (Section
3.2.1), but the disk geometry should be extremal for other spectral variational problems in this case as well. The standard proof of the original Faber-Krahn inequality uses rearrangement methods that are applicable for positive functions. The magnetic ground state of a general domain is genuinely complex and its amplitude, its phase, and the vector potential must be rearranged separately. A Schrodinger operator with a periodic external potential has purely absolutely continuous (AC) spectrum by a classical theorem of Thomas [161]. The periodicity of the magnetic field itself does not guarantee AC spectrum (e.g., B = const zA 0 in d = 2), but a periodic vector potential does. Note that this latter implies not only the periodicity of the magnetic field but also that the flux is zero in the unit cells.
The absolute continuity of the magnetic Schrodinger spectrum with a small periodic vector potential was first proven in [79]. The proof was reduced by perturbation to the original analyticity argument of Thomas and it could not be extended beyond the perturbative regime. In [15] a representation similar to (2.3) was used
to transform a periodic vector potential into a periodic external potential and a
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
411
modification of Thomas' argument applied. A periodic metric can also be included [131]. This approach, however, works only in d = 2 dimensions. The general case was obtained by Sobolev [158] who proved that the spectrum is purely a.c. for the magnetic Schrodinger operator with a sufficiently smooth periodic vector potential in any dimension. The proof combined Thomas' argument with a pseudodifferential technique.
3. Quantitative Properties of One-Body Operators 3.1. Lieb-Thirring Inequalities. One of the fundamental results about the standard Schrodinger operator -A + V is the Lieb-Thirring bound [119] on the moments of negative eigenvalues, Ej, in terms of integral norms of the negative parts of the potential, V_
tr [-A + V]" =
J[V]d+-y/2
jEj 1-1 < Ld,7
with a finite constant Ld,,y for d > 3, -y > 0; d=2, -y> 0 or d = 1, -Y> 1/2. This bound plays a crucial role in the proof of the stability of matter and it provides a basic a priori estimate for the semiclassical formulas and for justification of the Thomas-Fermi theory for the ground state energy of atoms and molecules. By the diamagnetic inequality, the usual proof of the Lieb-Thirring (LT) bound for the non-magnetic operator, -A+V, applies directly to the magnetic Schrodinger operator, (p + A)2 + V as well. The same holds for the Cwikel-Lieb-Rozenblum (CLR) bound on the number of eigenvalues (-y = 0). The presence of a magnetic field should, in principle, improve these estimates, but no such non-trivial result is available.
The systematic study of Lieb-Thirring bounds and semiclassics for the Pauli operator started with a series of seminal papers by Lieb, Solovej and Yngvason [116, 117, 118]. For the d = 3-dimensional Pauli operator, H = [o (p + A)]2 + V, with a constant magnetic field, B, and external potential, the following bound was proven for the sum of negative eigenvalues of H [116]
E I < (const) f[v]5/2 + (const)
f
JBI [V]3/2
(3.6)
where [V]_ = -min{0, V}. A similar bound holds in d = 2 dimensions as well [118]. The first term in (3.6) is the corresponding Lieb-Thirring estimate for -A + V. Due to the paramagnetism, the Pauli energy may be below the nonmagnetic energy and the additional term f JBI [V]3/2 is indeed necessary. The number of eigenvalues can be infinite in d = 2, 3 dimensions, so there is no CLRbound for the Pauli operator. For non-homogeneous magnetic fields, the bound (3.6) does not hold. Most importantly, the existence of the Loss-Yau zero modes shows that, in the perturbative regime, the lowest eigenvalue itself may scale linearly in [V]_. Moreover, the pointwise density of the Loss-Yau zero mode scales as max (x) I2 ,-, B3/2 Therefore, a general LT estimate in the strong field regime must contain a term that grows as the 3/2 power of B. To prove an LT estimate with the B3/2 scaling, the spin-coupling term a B in (1.1) is treated as a potential and the diamagnetic inequality is used for (p + A)2. Several papers [36, 156, 145, 19] used this idea with different assumptions on the
L. ERDOS
412
magnetic field. The most general result in this direction is due to Lieb, Loss and Solovej [113] showing that
E; I < (const)
J
[V]5/2 + (const)IIBII2/2 II[V]4.
(3.7)
The proof introduces the so-called running energy scale method. It consists of artificially scaling down the Pauli kinetic energy in an energy-dependent way to reduce the negative effect of o, B. The main advantage of this method is that it uses no other assumptions on B apart from the finiteness of its L2-norm. Note that f B2 is the field energy. Although a term growing as B3!2 (in the large field regime) is necessary for a general LT bound, a smaller power is sufficient if some control on VB is allowed. Especially, the linearity in B of the bound (3.6) reflects the basic fact that the space with a magnetic field cannot be considered isotropic: the magnetic field affects only the quantum motion in the transversal directions. Under a control on the H1 norm of B, the LT bound in [20] scales as the 17/12 power of the field. With more regularity on B and V the lower power 5/4 was obtained in [44]. Finally, the correct linear behavior in the field strength under a stronger regularity assumption was proved in [47] and [48]. The proof in [47] is shorter, but the estimate is not local: a large irregular magnetic field far away from the support of [V]_ should not influence the eigenvalue sum too much, but the estimate in [47] does not reflect this. A conceptually different proof was given in [48] that relies on a much stronger localization and approximation technique. The main difficulty behind these proofs is to control the density of Loss Yau zero modes. It is amusing to note that it was a substantial endeavor to show that zero modes may exist at all (Section 2.3). On the other hand, it is quite difficult to prove an upper bound on their number in terms of the expected first power of the magnetic field [48].
3.2. Semiclassics and Strong Fields. We have seen that magnetic fields can cause surprising effects when the magnetic lengthscale is comparable with other lengthscales in the problem. However, in the semiclassical and/or in the strong field regimes, lengthscales are typically separated, rendering simpler formulas available
in the limit. One studies the magnetic Schrodinger or Pauli operators with two parameters:
H(h, b) :_ (hp + bA)2 + V or [o,. (hp + bA)]2 + V
,
where h << 1 is the semiclassical parameter and b is the field strength; in most cases b >> 1 (assuming that A and V are fixed). The magnetic field is bB(x) = b curl A(x). Under these scalings, the magnetic field can typically be approximated by a (locally) homogeneous one, since the magnetic lengthscale (b/h)-1/2 is short. If, in addition, hb << 1, then the gap between (local) Landau levels shrinks to zero and magnetic effects usually do not contribute to the main term in the asymptotic regime. This is especially the case for the standard semiclassics when h -* 0 and b is fixed [25]. If, on the other hand, hb 74 0, then the main term is typically obtained by replacing the magnetic field by a locally constant one.
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
413
The key technical step is, therefore, to localize the problem to a sufficiently small scale, where the data (especially the magnetic field) can be considered homogeneous. Similarly to the non-magnetic theories, two basic techniques have been developed: pseudodifferential calculus and coherent states. Note that the presence of a constant magnetic field changes the structure of the phase space. For example, in three dimensions, the phase space is U' 0 R4, where v labels the Landau levels. The phase space for each level is four-dimensional; it consists of three position coordinates and only one momentum coordinate which represents the free motion parallel with the magnetic field. The momenta transversal to the field are not present due to the localization effect of the field. Accordingly, one either has to develop a pseudodifferential calculus that treats the harmonic oscillators on each Landau level exactly or one has to construct magnetic coherent states. 3.2.1. Results on individual eigenvalues. Semiclassical estimates on individual eigenvalues and eigenfunctions have mainly been carried out for the ground states.
For the magnetic Schrodinger operator without potential the ground state is localized near the minimum of the field strength ("magnetic bottles"; see [5] and [22]). The basic observation is that due to the positivity of the Pauli operator and the Lichnerowicz formula (1.1), the magnetic Schrodinger operator can always be estimated from below by B(x), at least in two dimensions and if B > 0. A similar result holds in higher dimensions, at least locally. This concentration of ground state is especially visible in the large field regime. Note that if an external potential is not present, the semiclassical limit is formally identical to the large magnetic field limit, h = 1/B. Precise analysis of this phenomenon was initiated by several groups with different methods. With the help of the Feynman-Kac formula, the magnetic ground
state energy can be turned into a question about the rate of decay of an oscillatory Wiener integral, see [126] and for a more precise bound [35], [162]. Ueki has also explored the connection with the hypoellipticity of the 8b problem [163]. Montgomery [130] has analyzed the case when the two-dimensional magnetic field
vanished along a curve. In this case min JBI = 0, hence the leading term in the large field asymptotics vanishes, and Montgomery obtained the subleading term that involved the curvature of the zero locus. This approach was later generalized in [73].
On a domain with boundary, however, the Lichnerowicz formula (1.1) does not hold, unless Dirichlet boundary conditions are imposed. In particular, the ground state energy of the magnetic Schrodinger operator with Neumann boundary condition is smaller than B even for a constant magnetic field. In this latter case the ground state is localized near the boundary, more precisely near the point with largest curvature of the boundary. The second term in the semiclassical expansion of the ground state energy is determined by the curvature, similarly to Montgomery's
result. A similar phenomenon occurs in three dimensions as well with a proper definition of an effective curvature [74]. Recently a complete expansion for the energy of the low lying eigenvalues in d = 2 dimensions was carried out in [55]. We remark that the Neumann boundary problem naturally arises at the minimization of the Ginzburg Landau energy functional describing superconducting states. Several people have contributed to these results; see [74, 55] and references therein.
L. ERDOS
414
3.2.2. Results on Cumulative Spectral Quantities. The first result on spectral statistics, where the magnetic field played a non-trivial role, is probably due to Colin de Verdiere [22] and Tamura [160] (independently) who proved a Weyl-type asymptotics for the number of eigenvalues with a non-homogeneous magnetic field and a confining potential. The magnetic field increases at infinity, ensuring that magnetic effects contribute to the large energy asymptotics. Colin de Verdiere used the magnetic extension of the classical Dirichlet-Neumann bracketing, while Tamura estimated the short time asymptotics of the magnetic heat kernel. Several authors extended these results; see [127] for references. A technically somewhat similar problem is the rate of the eigenvalue accumulation near the Landau levels due to perturbation by a decaying potential. The basic idea is that the magnetic field strongly localizes the particle and its interaction with the potential can be computed fairly precisely. The accumulation rate is explicitly given by the decay rate of the external field; see [138] and references therein. The next semiclassical question concerns the moments of negative eigenvalues in the spirit of [115]. Here we consider only the more interesting case of the Pauli operator, H(h, b) := [o . (hp+bA)]2+V. For simplicity, we work in d = 3 dimensions and we will approximate only the eigenvalue sum E(h, b) := tr [H(h, b)] The corresponding semiclassical expression is
ESc(h, b) := -h-3 fR P (hbB(x), [V (x)])dx 3
with
P(B, W) :=
82
(W3/2 + >[2vB - W]3/2)
.
(3.8)
V=1
This formula can be simply deduced from the structure of the phase space outlined above.
For a homogeneous magnetic field, the semiclassical limit lim
E(h, b)
h-o E,,(h, b)
=1
(3.9)
was proved uniformly in the field strength b [117] (the two-dimensional result was obtained in [118]). The main ingredients were the magnetic Lieb-Thirring inequality (3.6) and new magnetic coherent states. For the non-homogeneous case, a Lieb-Thirring inequality that scales as B3/2 (see Section 3.1) allows one to prove
the semiclassical formula only up to hb = 0(1) [157]. With the improved LiebThirring inequality [44] and a new construction of coherent states, the proof can be extended to b << h-3. This result is already sufficient to cover the full semiclassical regime of the large atoms [45]; see Section 4.2. Uniform semiclassics can be obtained with the help of the uniform Lieb-Thirring inequalities [47, 48]. The development using pseudodifferential calculus has focused on obtaining precise spectral asymptotics for the local traces of the form tr [ xcp(H) ], where x is a spatial cutoff function. These efforts have culminated in the book of Ivrii [89] where precise remainder estimates were proven in great generality. His recent work
investigates the same questions with irregular data [90]. A more concise result using these ideas is the proof of a certain local version of (3.9) for homogeneous
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
415
fields in [154], improved later in [155] to include Coulomb singularities. The microlocal technique also gives higher order corrections to the leading term. However, these methods require, in general, strong regularity assumptions on the data. Moreover, some non-asymptotic a priori estimate (Lieb-Thirring bound) is necessary to remove the cutoff X. In addition to the energy, other physical quantities are also of interest. Fournais
has studied the quantum current in a magnetic field and proved the corresponding semiclassical formula. Note that the current is a second order effect in the semiclassical expansion and it becomes a leading term only after non-trivial cancellations. Both microlocal techniques similar to [154] and coherent states methods similar to [117] have been tested, e.g., in [52] and [53].
3.3. Peierls Substitution and Corrections to the Semiclassics. The Bloch decomposition for a single particle Schrodinger operator with a periodic external potential can be extended to include weak electromagnetic fields. The basic idea due to Peierls is to substitute the minimally coupled magnetic momentum p+A into the band functions, E,,,(p), obtained from the non-magnetic Bloch decomposition. If the electromagnetic field varies on a much larger scale than the periodic background, then the problem is effectively semiclassical with a scale-separation parameter e (for the general theory, see [135]). The resulting pseudodifferential operator can be analyzed with well-developed mathematical tools. Algebraic methods were applied in [13]; for a systematic presentation of the pseudodifferential approach, see [77]. For example, it was shown in [65] that near a fixed energy level the original Hamiltonian is isospectral to a pseudodifferential operator with the same principal symbol as the Peierls Hamiltonian has. The detailed behavior of the density of states, in particular, the de Haas-van Alphen effect for the oscillation of the magnetization, was shown in [78]. The de Haas-van Alphen oscillation is due to a subleading effect in a semiclassical type expansion. However, it determines the current to leading order. The electromagnetic field is weak, but on the long time scale, t - e-1, it yields an order one change in the dynamics. To describe these effects on the dynamics correctly, Panati, Spohn and Teufel [136] developed a time dependent version of the Peierls substitution. The classical equations are corrected by an effective magnetic moment (Rammal-Wilkinson term) and an "anomalous velocity" term due to the curvature of the Berry connection. The latter, in particular, provides a simple semiclassical explanation of the quantum Hall current.
We remark that an oscillation similar to the de Haas-van Alphen effect is exhibited for the Harper operator (2.4) for magnetic fluxes (per unit cells) that are near a rational number; see [62] and references therein. The Harper operator itself can also be viewed as a Peierls substitution by quantizing the classical symbol
cost + cos ij with [,
21rica.
In this case the distance of a to a
nearby rational number with small denominator plays the role of the semiclassical parameter [76].
L. ERDOS
416
4. Many-Body Magnetic Systems 4.1. Magnetic Stability of Matter. In a system of charged fermionic quantum particles subject to a Coulomb interaction the ground state energy per particle is uniformly bounded, independently on the number of particles. This fundamental fact is called the stability of matter. For an excellent review of the progress in the last 35 years, see [106]. The first proof of the stability of matter with the non-relativistic kinetic energy, -A = p2, is due to Dyson and Lenard. A simpler proof was given later by Lieb and Thirring using the Lieb-Thirring inequality. Stability of matter also holds if the non-relativistic kinetic energy operator, p2, is replaced by JpJ = v"--0 ("relativistic" kinetic energy) and the fine structure constant, a, is sufficiently small. The first proof was given by Conlon, improved by Fefferman-de la Llave and finally the optimal bound was obtained by Lieb and Yau; see also a recent improvement and references in [112]. In the relativistic case, the kinetic energy and the potential energy both scale as [length]-1, therefore the energy per particle can be bounded from below only if the Hamiltonian is non-negative, i.e., N
1: Ipjl+aVc>0
(4.10)
j=1
where V. stands for the Coulomb potential of N electrons and K nuclei with charges Z. This inequality was proven by Lieb and Yau in [120] (Theorem 2) for a < 1/94
and Za < 2/7r, the second condition being optimal. Theorem 1 of [120] has a weaker result (a < 0.016, Za < 1/7r) but its proof can easily be generalized to the magnetic case as well, where the kinetic energy JpI can be replaced by its magnetic counterpart lp + Al. This follows from a simple application of the diamagnetic inequality, as pointed out in [113]. With the Pauli kinetic energy, however, even the hydrogen atom is unstable because in a strong magnetic field the electron can be strongly localized around the nucleus without a penalty in the kinetic energy. The ground state energy of the hydrogen in a constant magnetic field B diverges as (log B)2, if B is large [5]. If the energy of the magnetic field is added to the total energy, then stability is restored: N
(p+A)j]2+V+ j=1
1
87ra2
JB2 > -C(Z)(K + N)
(4.11)
where the constant depends only on the charges of the nuclei. The parameter a (fine structure constant) must be sufficiently small in order for (4.11) to hold. The existence of a Loss-Yau zero mode [124] shows that the total energy can be negative if a is large. Actually, an absolute upper bound on a and an upper bound on (maxZj)a2 are both necessary, where Zj are the nuclear charges. The bound (4.11) is called the stability of matter interacting with a classical magnetic field. It was first proven for atoms [60] and single electron molecules [107]. The general case was settled by Fefferman [50] for a very small a. Lieb, Loss and Solovej [113] gave a much shorter proof that also holds for the physical value of the fine structure constant. The backbone of this proof is the magnetic Lieb-Thirring inequality (3.7).
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
417
Using the Birman-Koplienko-Solomyak trace inequality and the magnetic version of the Lieb-Yau bound (4.10), Lieb, Siedentop and Solovej [114] also proved
the stability of matter for the Dirac operator. The particles must be restricted to the positive energy subspace of the one-particle Dirac operator a (p + A) +,3 (filling the "Dirac sea"). It is important to note that the Dirac sea must be defined via the gauge invariant Dirac operator. Restriction onto the positive energy subspace of the free Dirac operator (as it is often done in perturbation theory) leads to instability of matter for any a. Ultimately, the electromagnetic field must also be quantized. Imposing an ultraviolet cutoff and using results from [113], Bugliaro, Frohlich and Graf proved stability of matter for the Pauli operator with a quantized electromagnetic field [21]. The essential observation is that to restore the magnetic stability for the Pauli operator with a classical field, it is sufficient to add the field energy only near the nuclei. On a finite volume and with an ultraviolet cutoff, the classical and quantized field energies can be compared. Lieb and Loss [109] showed stability of matter for the Dirac operator with a quantized electromagnetic field and with a suitable one-body spectral projection, similar to [114]. The quantization of the electromagnetic field poses several complications, such as ultraviolet cutoff and mass renormalization, and little is known about how to rigorously include them into a fully consistent theory. However, one problem has been settled satisfactorily: the existence of atoms in non-relativistic QED. Because of the quantized field, the ground state of the total system (atom and photons) is at the bottom of a continuous spectrum, and it is not at all obvious that it is an eigenvalue. Moreover, the so-called binding condition also has to be satisfied, i.e., the energy of a system of N electrons is actually lower than that of a system with fewer electrons, otherwise the ground state may contain no electron at all. For small values of the parameters (ultraviolet cutoff parameter and fine structure constant) the existence of atoms was shown in [8] and for arbitrary parameter values in [70, 110]. No infrared cutoff was needed, unlike for the scattering problem (Section 4.3). More recently, the thermodynamic limit for non-relativistic Coulomb matter with quantized electromagnetic field was investigated and the lower bound was proved [111].
4.2. Large Atoms. One of the main motivations to study semiclassical spectral asymptotics for -h2A+V originates in the seminal paper of Lieb and Simon, [115], where the exactness of the Thomas-Fermi theory for the ground state energy of atoms with large nuclear charge, Z >> 1, was proven with semiclassical methods. In the presence of magnetic fields the Thomas-Fermi theory is more complex.
Depending on the strength of the magnetic field compared with Z, there are five different regimes.
Within the classical Thomas-Fermi theory, the kinetic energy as a functional of the density is always given by the Legendre transform of the pressure (3.8). This classical magnetic Thomas-Fermi theory, however, holds only for weak and moderate magnetic fields (B << Z3). More precisely, for B << Z4/3 the magnetic effects are absent in the leading term of the large Z asymptotics. For B _ Z4/3 the full magnetic Thomas-Fermi theory is needed. If Z4/3 << B << Z3, the usual Thomas-Fermi theory still applies, but only the first summand in the pressure function (3.8) is needed. The atom is spherical in all these cases, but the energy is
L. ERDOS
418
affected by the magnetic field. These results were proven in [117] with the help of the magnetic Lieb-Thirring inequality and the semiclassical result (3.9). For a strong magnetic field, B > (const)Z3 with a positive constant, the electrons are confined to the lowest Landau band and the shape of the atom is a long
cylinder along the field. For B - Z3 the atoms in the transversal direction have a non-trivial structure that can be described by a new density functional theory relying on minimizing density matrices instead of density functions [116]. Finally, if B >> Z3, the atom becomes effectively one-dimensional and a one-dimensional
Thomas-Fermi caricature applies. Analogous results in d = 2 were obtained in [118].
If the magnetic field goes to infinity, but Z is fixed, then the ground state energy
diverges as -4Z2(logB/2)2. For one-electron atoms this has been established in [5]. The energy of the many-body system, after factoring out the divergent (log B)2 term, is given by the ground state energy of an effective one-dimensional bosonic Hamiltonian with Dirac delta interactions [11]. This idea has been extended to prove resolvent convergence and explore other effective Hamiltonians in [18] and references therein. The correctness of the Thomas-Fermi theory in the semiclassical regime (B Z3) for non-homogeneous magnetic fields was proven in [45] after extending the Lieb-Thirring inequality [44] and constructing appropriate coherent states. The uniform magnetic Lieb-Thirring inequality [47] should also allow the extension of the strong field regime from [116] to non-homogeneous magnetic fields. The asymptotic behavior of the total magnetization and of the current for large atoms in homogeneous fields was obtained in [53, 54]. A bound on the maximal ionization is proven in [143].
4.3. Multiparticle Scattering in a Magnetic Field. A detailed presentation of scattering theory and asymptotic completeness is given in the contribution of C. Gerard in this Festschrift; here we just briefly mention the most important results involving magnetic fields. N-body asymptotic completeness is well understood for non-relativistic particles without magnetic field. Scattering in the presence of a constant magnetic field is a much more delicate question since a classical charged particle moves on circles. Therefore, charged subsystems can scatter only parallel with the field, while neutral systems may move out to infinity in all directions. The general theory has been developed and asymptotic completeness has been proved by Gerard and Laba for the case when all possible subsystems are charged (the best reference is their book [64]). The zero charge case in general is still open. The special case of three particles with Coulomb forces was solved in [63] and one charged particle was considered in [1]. A constant electric field can also be included. Skibsted [153] reduced this prob-
lem to scattering of non-interacting subsystems all having the same charge/mass ratio. In the presence of a quantized electromagnetic field, the scattering of photons on a bound electron (Rayleigh scattering) and the scattering of electrons dressed with photons (Compton scattering) have been studied. The mathematical framework is non-relativistic quantum mechanics with a quantized field with ultraviolet cutoff to ensure that the Hamiltonian is well defined. For total energies below the electron-
positron pair creation threshold, this non-relativistic caricature of QED is well
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
419
justified. One also has to cope with the infrared (IR) problem ("soft bosons"): there could be infinitely many photons with a finite total energy. For technical simplicity, scalar photons are considered. This set of problems was initiated by Frohlich who first investigated the infrared problem and constructed wave operators (without completeness) [56]. Rayleigh scattering was first tackled in [28] where the photons were massive to avoid the IR problem and the potential was confining. Later it was extended to the physically more realistic massless photons, but with an IR-cutoff, and instead of a confining potential, the total energy was set below the ionization threshold which also guarantees spatial localization of the electrons [58]. During scattering, photons are absorbed by the bound electrons, lifting them to higher excited states,
but no electron can escape. Thus, after some time, the electron cloud relaxes to the ground state and emits photons that propagate essentially freely to infinity in space. Note that the analysis of Bach et al. guarantees that all excited states are unstable [8, 9]. For Compton scattering [59], the total energy is assumed to be sufficiently small so that the speed of the massive electron is less than 1/3 of the speed of light. This technical assumption safely separates free photons from the dressed electron after long time evolution. The asymptotic completeness in this model means that
the long time evolution of the state is a linear combination of asymptotic states consisting of a freely moving electron dressed by a photon cloud plus freely moving excess photons.
5. Random Schrodinger Operators with Magnetic Fields Since the proof of the Anderson localization with the powerful multiscale method of Frohlich and Spencer [61], random Schrodinger operators have become one of the main research directions in mathematical physics. The typical problems concern the self-averaging properties (deterministic spectrum, existence of density of states), the asymptotics of the density of states (Lifshitz tail), and establishing the dense point spectrum in the localization regime. These questions have been recently studied with magnetic field as well; most results are in the most relevant d = 2 dimensions.
5.1. Constant Magnetic Field. First we consider the problems where the magnetic field is constant and only the external potential is random, i.e., the random Landau Hamiltonian (for a recent survey, see [103] and references therein). In
this case, the standard self-averaging techniques work to establish deterministic spectrum, the translation must simply be replaced by the magnetic translation. The Lifshitz tail in the low energy regime for Gaussian random potential is very universal and it is insensitive to the magnetic field (it even holds for certain random magnetic
fields). For Gaussian randomness even the density of states is bounded and the Wegner estimate holds [87]. Localization with algebraically decaying eigenfunctions was proved in [51], exponential localization in [163]. These proofs are valid only at very low energies, using the deep wells of the Gaussian randomness; in particular, this regime is far from the analogue of the band-edge localization.
For repulsive Poissonian obstacles, the precise Lifshitz tail is more delicate, similar to the classical vs. quantum dichotomy in the non-magnetic setup. If the single-site potential has a slow decay, then classical effects dominate and the Lifshitz tail can be computed from a simple mean-field argument. Otherwise the quantum
420
L. ERDOS
localization energy competes with the entropy of the large domains free of impurities. For magnetic fields, the threshold decay is not algebraic but Gaussian. The classical regime was investigated in [16] (in three dimensions [85]), the main result in the quantum regime was obtained in [40] using Sznitman's coarse-graining probabilistic method (see also [83, 41]). Since in the Poisson model the energy is not monotone in the random variable so Wegner estimates are much harder to obtain, localization for a constant magnetic field was investigated for Anderson-type i.i.d. random potentials. Wang has given a full asymptotic expansion of the density of states away from the Landau bands [165]. More explicit quantitative results are known on the Lifshitz tails, including the double logarithmic asymptotics at each band edge [96]. The pure point spectrum at a certain distance away from the Landau levels was proven independently by Wang [164], Combes and Hislop [24] and in a single-band approximation by Dorlas, Macris and Pule [29]. Since d = 2 is the borderline dimension for the necessary large distance decay of the Green's function of the non-magnetic Lapla-
cian, an additional decay must be extracted from the presence of the magnetic field. This eventually required basic results from two-dimensional bond percolation theory. Precise estimates on the localization length and dynamical localization near the Landau band edges was obtained by Germinet and Klein [67] using their extension of the Frohlich-Spencer multiscale analysis [66]. In contrast to the localization regime, the presumed regime of delocalization in Anderson-type models is poorly understood and there are almost no mathematical results. Apart from the Bethe lattice, there is only one model, where the existence of the mobility edge has been rigorously proven: the random Landau Hamiltonian [68]. More precisely, it is shown that there exists at least one energy E near each
Landau band, so that the the local transport exponent, O(E), is positive. The local transport exponent measures the extension of the wave packet in a suitable averaged sense for large times. (Dynamical) localization is characterized by O(E) = 0. Moreover, there is an important dichotomy for /3(E): it is either zero or at least 1/2d.
The key quantity in the proof in [68] is the Hall conductance. In the regime of dynamical localization, the Hall conductance is constant in the mobility gap (see [12], strengthened later in [31]). On the other hand, the Hall conductivity jumps by one at each Landau level for the free Landau Hamiltonian and it is also constant, as a function of the disorder parameter, in the gaps. Therefore, it must jump somewhere inside the bands, at least for sufficiently small disorder, but then the complete band cannot belong to the dynamically localized regime, completing the argument of delocalization in [68]. The Hall conductance for quantum Hall systems (two-dimensional disordered samples subject to a constant perpendicular magnetic field) at energies falling into
the mobility gap, A, can actually be defined in two different ways. The bulk conductance is defined on R2 by the Kubo-Streda formula (see [6]) QB (A) = -itrPP [ [PA, A,], [Pa, A2] ]
where PA is the spectral projection onto (-oo, A] and Ai is the characteristic function of {xi < 0}, i = 1, 2. The edge conductance is defined in a half-plane sample (x2 > -a, eventually
a -ioo)by UE = -itrO (H) [H, Al]
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
421
(modulo non-trivial technicalities), where p is a smooth spectral cutoff function that is one below A and zero above A. This quantity gives the derivative of the current
flowing through the line x1 = 0 with respect to lowering the chemical potential along the edge. These two conductances are the same. For intervals A falling into the spectral gap, this was proven in [142] with K-theoretical methods and later in [30] by basic functional analysis. The proof given recently in [31] is also valid for intervals that contain strongly localized spectrum only (such intervals are often called mobility gaps).
5.2. Random Magnetic Field. Since a magnetic field itself enhances localization, one expects that a random magnetic field is localizing even stronger than a random potential. Technically, however, random magnetic fields are harder to fit into the multiscale analysis, mainly because the vector potential is non-local. There is a substantial difference between the zero and non-zero flux cases, the former being easier. In particular, stationary random vector potentials always generate magnetic fields that have zero flux on average.
The existence of the integrated density of states (IDS) and its independence of the boundary conditions in the thermodynamic limit (uniqueness) has first been shown by Nakamura for both the discrete and continuous Schrodinger operator with a random magnetic field with non-zero flux [133, 134]. In both cases, Lifshitz tails were also obtained. Recently, Hundertmark and Simon gave a short proof for the existence and uniqueness of the IDS [86]. Anderson localization has been shown for a Gaussian vector potential by Ueki in [163]. The Germinet-Klein multiscale analysis has been extended to include very general random magnetic fields, but the Wegner estimate requires a random vector potential (which implies zero average flux) in addition to other technical conditions. For the discrete random magnetic Schrodinger operator, the method of Nakamura [133] has been extended to obtain the Wegner estimate and localization [95]. However, the zero flux condition is enforced in a strong sense: neighboring cells are paired and the magnetic flux is opposite in these pairs. A deterministic background magnetic field atop of the local random vector potential is allowed in [81] where the Wegner estimate was proved in the continuous model. A small stationary random vector potential was included in a Schrodinger operator with a periodic background potential in [69], where Lifshitz tails were proven under a special non-resonance condition. Band-edge localization for Schrodinger operators with random magnetic field is widely believed to hold in the most general case. The additional assumptions
on the zero flux (which, in one form or another, is present in all papers so far) seems to be only technical. However, there is no agreement in the physics literature about the possible existence of the continuous spectrum for such operators, unlike in the non-magnetic case, where the existence of the extended states is universally accepted by physicists and "only" the mathematical proof is missing. The
Landau orbits and their quantum counterparts, the strongly localized magnetic eigenfunctions, are characteristic only to the constant magnetic field and they are not universal. There is a competition between a possible weaker form of the Landau localization, that may still hold for random fields, and the resonance effects that enhance delocalization.
L. ERDOS
422
Acknowledgements. The author is indebted to Jan Philip Solovej for his help in preparing the manuscript. Special thanks to Christian Gerard, Gian Michele Graf, Dirk Hundertmark, Frederic Klopp, Michael Loss, Benjamin Schlein, Stefan Teufel,
Jakob Yngvason, Simone Warzel and to the referee for their critical reading and corrections. Part of this work was done at Harvard University. References
[1] T. Adachi: On spectral and scattering theory for N-body Schrodinger operators in a constant magnetic field. Rev. Math. Phys. 14, 199-240 (2002)
[2] C. Adam, B. Muratori and C. Nash: Multiple zero modes of the Dirac operator in three dimensions, Phys. Rev. D (3) 62, 085026 (2000) [3] Y. Aharonov and A. Casher: Ground state of spin-1/2 charged particle in a two-dimensional magnetic field. Phys. Rev. A19, 2461-2462 (1979) [4] A. Avila and S. Jitomirskaya: The ten Martini problem. To appear in Ann. of Math. [5] J. Avron, I. Herbst and B. Simon: Schrodinger operators with magnetic fields. I. General interactions. Duke Math. J. 45 (1978), 847-883; II. Separation of the center mass in homogeneous magnetic fields. Ann. Phys. 114, 431-451 (1978); III. Atoms in homogeneous magnetic fields. Commun. Math. Phys. 79, 529-572 (1981) [6] J. Avron, R. Seiler and B. Simon: Charge deficiency, charge transport and comparison of dimensions. Commun. Math. Phys. 159, 399-422 (1994) [7] J. Avron and B. Simon: A counterexample to the paramagnetic conjecture. Phys. Lett. A75, 41-42 (1979) [8] V. Bach, J. Frohlich and I. M. Sigal: Spectral analysis for systems of atoms and molecules coupled to the quantized radiation field. Commun. Math. Phys. 207, 249-290 (1999)
[9] V. Bach, J. Frohlich, I. M. Sigal and A. Soffer: Positive commutators and the spectrum of Pauli-Fierz Hamiltonian of atoms and molecules. Commun. Math. Phys. 207, 557-587 (1999)
[10] A. Balinsky and W. D. Evans: On the zero modes of Pauli operators, J. Funct. Anal. 179, 120-135 (2001)
[11] B. Baumgartner, J. P. Solovej and J. Yngvason: Atoms in strong magnetic fields: the high field lmit at fixed nuclear charge. Commun. Math. Phys. 212, 703-724 (2000) [12] J. Bellissard, A. van Elst and H. Schulz-Baldes: The noncommutative geometry of the quantum Hall effect. J. Math. Phys. 35, 5373-5451 (1994) [13] J. Bellissard and R. Rammal: An algebraic semi-classical approach to Bloch electrons in a magnetic field. J. Physique France, 51, 1803 (1990) [14] J. Bellissard and B. Simon: Cantor spectrum for the almost Mathieu equation. J. Funct. Anal. 48, 408-423 (1982) [15] M. Sh. Birman and T. A. Suslina: The two-dimensional periodic magnetic Hamiltonian is absolutely continuous. St. Petersburg Math. J. 9, 21-32 (1998) [16] K. Broderix, D. Hundertmark, W. Kirsch and H. Leschke: The fate of Lifshits tails in magnetic fields. J. Stat. Phys. 80 1-22 (1995) [17] K. Broderix, D. Hundertmark and H. Leschke: Continuity properties of Schrodinger semigroups with magnetic fields. Rev. Math. Phys. 12, 181-225 (2000) [18] R. Brummelhuis and P. Duclos: Effective Hamiltonians for atoms in very strong magnetic fields.xxx.lani.gov/math-ph/0502050 [19] L. Bugliaro, C. Fefferman, J. Frohlich, G. M. Graf and J. Stubbe: A Lieb-Thirring bound for a magnetic Pauli Hamiltonian, Commum. Math. Phys. 187, 567-582 (1997) [20] L. Bugliaro, C. Fefferman and G. M. Graf: A Lieb-Thirring bound for a magnetic Pauli Hamiltonian, II, Rev. Mat. Iberoamericana, 15, 593-619 (1999) [21] L. Bugliaro, J. Frohlich and G. M. Graf: Stability of quantum electrodynamics with nonrelativistic matter, Phys. Rev. Lett. 77, 3494-3497 (1996) [22] Y. Colin de Verdiere: L'asymptotique de Weyl pour les bouteilles magnetiques. Commun. Math. Phys. 105, 327-335 (1986) [23] Y. Colin de Verdiere and N. Torki: Opdrateurs de Schrodinger avec champs magndtique. Seminaire de theorie spectrale et geometrie (Grenoble) 11, 9-18 (1992-1993)
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
423
[24]
J. M. Combes and P. Hislop: Landau Hamiltonians with random potentials: localization and the density of states. Commun. Math. Phys. 177, 603-629 (1996)
[25]
J. M. Combes, R. Schrader and R. Seiler: Classical bounds and limits for energy distributions of hamaltonaan operators an electromagnetic fields. Ann. Physics 111, 1-18 (1978)
[26] H. Cornean, I. Herbst and E. Skibsted: Spiraling attractors and quantum dynamics for a class of long-range magnetic fields. Preprint. Univ. Texas Mathematical Physics Archive, 02-452 (2002)
[27] H. L. Cycon, R. G. Froese, W. Kirsch and B. Simon: Schrodinger Operators with Application to Quantum Mechanics and Global Geometry. Springer (1987)
J. Derezinski and C. Gerard: Asymptotic completeness in quantum field theory: Massive Pauli-Fierz Hamiltonians. Rev. Math. Phys. 11, 383-450 (1999) [29] T. C. Dorlas, N. Macris and J. V. Pule: Localisation in a single-band approximation to random Schrodinger operators in a magnetic field. Helv. Phys. Acta 68, 329-364 (1995) [30] P. Elbau and G. M. Graf: Equality of bulk and edge Hall conductance revisited. Commun. Math. Phys. 229, 415-432 (2002) [31] A. Elgart, G. M. Graf and J. H. Schenker: Equality of the bulk and edge Hall conductances in a mobility gap. Commun. Math. Phys. 259, 185-221 (2005) [32] D. Elton: New examples of zero modes. J. Phys. A 33 (41), 7297-7303, (2000) [33] D. Elton: The local structure of zero mode producing magnetic potentials. Commun. Math. Phys. 229, 121-139 (2002) [34] L. Erdos: Ground state density of the Pauli operator in the large field limit. Lett. Math. Phys. 29, 219-240 (1993) [35] L. Erdos: Estimates on stochastic oscillatory integrals on the heat kernel of the magnetic Schrodinger operator. Duke Math. J. 76, 541-566 (1994) [36] L. Erdos: Magnetic Lieb-Thirring inequalities. Commun. Math. Phys. 170, 629-668 (1995) [37] L. Erdos: Gaussian decay of the magnetic eigenfunctions. Geom. Funct. Anal. 6, 231-248 [28]
(1996)
[38] L. Erdos: Rayleigh-type isoperimetric inequality with a homogeneous magnetic field. Calc. Var. and PDE 4, 283-292 (1996) [39] L. Erdos: Dia- and paramagnetism for nonhomogeneous magnetic fields. J. Math. Phys. 38, 1289-1317 (1997)
[40] L. Erdos: Lifshitz tail in a magnetic field: the non-classical regime. Probab. Theor. Relat Fields 112, 321-371 (1998) [41] L. Erdos: Lifshitz tail in a magnetic field: coexistence of classical and quantum behavior in the borderline case. Probab. Theor. Relat. Fields 121, 219-236 (2001) [42] L. Erdos: Spectral shift and multiplicity of the first eigenvalue of the magnetic Schrodinger operator in two dimensions. Ann. Inst. Fourier (Grenoble) 52, 1833-1874 (2002) [43] L. Erdos, M. Loss and V. Vougalter: Diamagnetic behavior of sums of Dirlchlet eigenvalues. Ann. Inst. Fourier (Grenoble) 50, 891-907 (2000) [44] L. Erdos and J. P. Solovej: Semiclassical eigenvalue estimates for the Pauli operator with strong non-homogeneous magnetic fields. I. Non-asymptotic Lieb-Thirring type estimate. Duke Math. J. 96, 127-173 (1999) [45] L. Erdos and J. P. Solovej: Semiclassical eigenvalue estimates for the Pauli operator with strong non-homogeneous magnetic fields. II. Leading order asymptotic estimates. Commun. Math. Phys. 188, 599-656 (1997) [46] L. Erdos and J. P. Solovej: The kernel of Dirac operators on S3 and R3. Rev. Math. Phys. 13, 1247-1280 (2001) [47] L. Erdos and J. P. Solovej: Magnetic Lieb-Thirring inequalities with optimal dependence on the field strength. J. Stat. Phys. 116, 475-506 (2004)
[48] L. Erdos and J. P. Solovej: Uniform Lieb-Thirring inequality for the three-dimensional Pauli operator with a strong non-homogeneous magnetic field. Ann. Inst. H. Poincare 5, 671-741 (2004)
[49] L. Erdos and V. Vougalter: Pauli operator and Aharonov-Casher theorem for measure valued magnetic fields. Commun. Math. Phys. 225, 399-421 (2002) [50] C. Fefferman: Stability of Coulomb systems in a magnetic field. Proc. Nat. Acad. Sci. USA 92, 5006-5007 (1995) [51] W. Fischer, H. Leschke and P. Muller: Spectral localization by Gaussian random potentials in multi-dimensional continuous space. J. Stat. Phys. 101, 935-985 (2000)
L. ERDOS
424
[52] S. Fournais: Semaclassacs of the quantum current an a strong constant magnetic field. Comm. Partial Diff. Eq. 26, 1427-1496 (2001) [53] S. Fournais: The magnetisation of large atoms in strong magnetic fields. Commun. Math. Phys. 216, 375-393 (2001) [54] S. Fournais: Confinement to lowest Landau band and application to quantum current. Rev. Math. Phys. 15, 1219-1253 (2003) [55] S. Fournais and B. Helffer: Accurate estimates for magnetic bottles in connection with
superconducivity. To appear in Ann. Inst. Fourier. Preprint in Univ. Texas Mathematical Physics Archive, 04-385 (2004)
[56] J. Frohlich: On the infrared problem in a model of scalar electrons and massless scalar bosons. Ann. Inst. H. Poincare 19, 1-103 (1973); and Existence of dressed one electron states in a class of persistent models. Fortschr. Phys. 22, 159-198 (1974) [57] J. Frohlich, G. M. Graf and J. Walcher: On the extended nature of edge states of quantum Hall Hamiltonians. Ann. Henri Poincare 1, 405-442 (2000) [58] J. Frohlich, M. Griesemer and B. Schlein: Asymptotic completeness for Rayleigh scattering. Ann. Henri Poincare 3, 107-170 (2002) [59] J. Frohlich, M. Griesemer and B. Schlein: Asymptotic completeness for Compton scattering. Commun. Math. Phys. 252, 415-476 (2004) [60] J. Frohlich, E. H. Lieb and M. Loss, Stability of Coulomb systems with magnetic fields. I. The one-electron atom. Commun. Math. Phys. 104, 251-270 (1986) [61] J. Frohlich and T. Spencer: Absence of diffusion in the Anderson tight binding model for large disorder or small energy. Commun. Math. Phys. 88, 151-184 (1983) [62] O. Gat and J. Avron: Semiclassical analysis and the magnetization of the Hofstadter model. Phys. Rev. Lett. 91, 186801 (2003) [63] C. Gerard and I. Laba: Scattering theory for 3-particle systems in constant magnetic fields: dispersive case. Ann. Inst. Fourier (Grenoble) 46, 801-876 (1996)
[64] C. Gerard and I. Laba: Multiparticle Quantum Scattering in Constant Magnetic Fields. Math. Surveys and Monographs 90, AMS, Providence, RI (2002) [65] C. Gerard, A. Martinez and J. Sjostrand: A mathematical approach to the effective Hamiltonian in perturbed periodic problems. Commun. Math. Phys. 142, 217-244 (1991) [66] F. Germinet and A. Klein: Bootstrap multiscale analysis and localization in random media. Commun. Math. Phys. 222, 415-448 (2001) [67] F. Germinet and A. Klein: Explicit finite volume criteria for localization in continuous random media and applications. Geom. Funct. Anal. 13, 1201-1238 (2003) [68] F. Germinet, A. Klein and J. Schenker: Dynamical delocalization in random Landau Hamiltonians. Preprint.xxx.lanl.gov/math-ph/0412070 [69] F. Ghribi: Asymptotique de Lifshitz pour des operateurs de Schrodinger magnetiques aleatoires. Ph.D. Thesis. Universite Paris 13 (2005) [70] M. Griesemer, E. H. Lieb and M. Loss: Ground states in non-relativistic quantum electrodynamics. Invent. Math. 145, 557-595 (2001) [71] B. Helffer: Effet d'Aharonov Bohm sur un etat borne de l'equation de Schrodinger. Commun. Math. Phys. 119, 315-329 (1988) [72] B. Helffer and A. Mohamed: Sur le spectre essential des operateurs de Schrodinger avec champ magnetique. Ann. Institut Fourier 38, 95-113 (1988) [73] B. Helffer and A. Mohamed: Semiclassical analysis for the ground state energy of a Schrodinger operator with magnetic wells. J. Funct. Anal. 138, 40-81 (1996) [74] B. Helffer and A. Morame: Magnetic bottles for the Neumann problem: curvature effects in the case of dimension 3 (general case). Ann. Sci. Ecole Norm. Sup. 37, 105-170 (2004) [75] B. Helffer, J. Nourrigat and X. P. Wang: Sur le spectre de l'equation de Dirac (dans R2 ou R3) avec champ magnetique. Ann. Sci. Ecole Norm. Sup. 4e serie t. 22, 515-533 (1989) [76] B. Helffer and J. Sjostrand: Analyse semi-classique pour l'equation de Harper I-III. Mem. Soc. Math. France (N.S.) 34 Tome 116 (1989), 39 Tome 117 (1990), 40 Tome 118 (1990) [77] B. Helffer and J. Sjostrand: Equation de Schrodinger avec champ magnetique at equation de Harper. Schrodinger Operators (Sonderborg, 1988), pp. 118-197, Lecture Notes in Physics 345, Springer (1989) [78] B. Helffer and J. Sjostrand: On diamagnetism and de Haas-van Alphen effect. Ann. Inst. H. Poincare Phys. Theor. 52, 303-375 (1990)
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
425
[79] R. Hempel and I. Herbst: Strong magnetic fields, Dirzchlet boundaries and spectral gaps. Commun. Math. Phys. 164, 237-259 (1995) [80] H. Hess, R. Schrader and D. A. Uhlenbrock: Domination of semzgroups and generalization of Kato's inequality. Duke Math. J. 44, 893-904 (1977) [81] P. Hislop and F. Klopp: The integrated density of states for some random operators with nonsign definite potentials. J. Funct. Anal. 195, 12-47 (2002) [82] H. Hogreve, R. Schrader and R. Seiler: A conjecture on spinor functional determinant. Nucl. Phys. B 142, 525-534 (1978) [83] T. Hupfer, H. Leschke, and S. Warzel: Poissonian obstacles with Gaussian walls discriminate between classical and quantum Lifshits tailing in magnetic fields. J. Stat. Phys. 97, 725-750 (1999)
[84] D. Hundertmark: Zur Theorie der magnetischen Schrodingerhalbgruppe. Ph.D. Thesis, Ruhr-Universitat Bochum (1997) [85] D. Hundertmark, W. Kirsch and S. Warzel: Classical magnetic Lifshitz tails in three space dimensions: impurity potential with slow anisotropic decay. Markov Proc. Rel. Fields 9, 651-660 (2003)
[86] D. Hundertmark and B. Simon: A diamagnetic inequality for semigroup differences. J. Reine. Angew. Math. 571, 107-130 (2004) [87] T. Hupfer, H. Leschke, P. Miiller and S. Warzel: The absolute continuity of the integrated density of states for magnetic Schrodinger operators with certain unbounded random potentials. Commun. Math. Phys. 221, 229-254 (2001) [88] T. Ikebe and T. Kato: Uniqueness of the self-adjoint extension of singular elliptic differential operators. Arch. Rat. Mech. Anal. 9, 77-92 (1962) [89] V. Ivrii: Microlocal Analysis and Precise Spectral Asymptotics. Springer (1998) [90] V. Ivrii: Sharp spectral asymptotics for operators with irregular coefficients. III, IV. V. Preprints 2005.xxx.lanl.gov/math.AP/0510326, 05103267, 0510328 [91] H. Iwashita: On the long-range scattering for one- and two-particle Schrodinger operators with constant magnetic fields. Tsukuba J. Math. 19, 369-376 (1995) [92] A. Iwatsuka: Examples of absolutely continuous Schrodinger operators in magnetic fields. Publ. RIMS 21, 385-401 (1985) [93] T. Kato: Schrddinger operators with singular potentials. Israel J. Math. 13, 135-148 (1973) [94] T. Kato: Remarks on Schrodinger operators with vector potentials. Integral Eq. Operator Theory 1, 103-113 (1978) [95] F. Klopp, S. Nakamura, F. Nakano and Y. Nomura: Anderson localization for 2D discrete Schrodinger operators with random magnetic fields. Ann. Henri Poincare 4, 795-811 (2003)
[96] F. Klopp and C. Raikov: [97] [98]
Lifshitz tails in constant magnetic fields. Preprint 2005,
xxx.lanl.gov/math-ph/0509022 I. Laba: Long-range one-particle scattering in a homogeneous magnetic field. Duke Math. J. 70, 283-303 (1993) I. Laba: Scattering for hydrogen-like systems in a constant magnetic field. Comm. in P.D.E. 20, 741-762 (1995)
[99] Y. Last: Zero measure spectrum for the almost Mathieu operator. Commun. Math. Phys. 164, 421-432 (1994) [100] Y. Last and B. Simon: The essential spectrum of Schrodinger, Jacobi, and CMV operators. J. d'Analyse Math. 98, 183-220 (2006) [101] H. Leinfelder: Gauge invariance of Schrodinger operators and related spectral properties, J. Op. Theory 9, 163-179 (1983) [102] H. Leinfelder and C. Simader: Schrodinger operators with singular magnetic vector potentials. Math Z. 176, 1-19 (1981) [103] H. Leschke, P. Muller and S. Warzel: A survey of rigorous results on random Schrodinger operators for amorphous solids. Markov. Proc. Rel. Fields 9, 729-760 (2003) [104] E. H. Lieb: The flux phase problem on planar lattices. Hely. Phys. Acta 65, 247-255 (1992) [105] E. H. Lieb: The flux phase of the half-filled band. Phys. Rev. Lett. 73, 2158-2161 (1994)
[106] E. H. Lieb: Quantum mechanics, the stability of matter and quantum electrodynamics. Jahresber. Deutsch. Math. Verein. 106, 93-110 (2004) [107] E. H. Lieb and M. Loss: Stability of Coulomb systems with magnetic fields, II. Commun. Math. Phys. 104, 271-282 (1986)
L. ERDOS
426
[108] E. H. Lieb and M. Loss: Fluxes, Laplacaans, and Kasteleyn's theorem. Duke Math. J. 71, 337-363 (1993) [109] E. H. Lieb and M. Loss: Stability of a model of relativistic quantum electrodynamics. Commun. Math. Phys. 228, 561-588 (2002)
[110] E. H. Lieb and M. Loss: Existence of atoms and molecules in non-relativistic quantum electrodynamics. Adv. Theor. Math. Phys. 7, 667-710 (2003) [111] E. H. Lieb and M. Loss: The thermodynamic limit for matter interacting with Coulomb forces and with the quantized electromagnetic field: I. The lower bound. Commun. Math. Phys. 258, 675-695 (2005)
[112] E. H. Lieb, M. Loss and H. Siedentop: Stability of relativistic matter via Thomas-Fermi theory. Helv. Phys. Acta 69, 974-984 (1996) [113] E. H. Lieb, M. Loss and J. P. Solovej: Stability of matter in magnetic fields, Phys. Rev. Lett. 75, 985-989 (1995) [114] E. H. Lieb, H. Siedentop and J. P. Solovej: Stability and instability of realativistic electrons in classical electromagnetic fields. J. Stat. Phys. 89, 37-59 (1997) [115] E. H. Lieb and B. Simon: The Thomas-Fermi theory of atom, molecules and solids. Adv. in Math. 23, 22-116 (1977) [116] E. H. Lieb, J. P. Solovej and J. Yngvason: Asymptotics of heavy atoms in high magnetic fields: I. Lowest Landau band region, Commun. Pure Appl. Math. 47, 513-591 (1994) [117] E. H. Lieb, J. P. Solovej and J. Yngvason: Asymptotics of heavy atoms in high magnetic fields: IT Semiclassical regions. Commun. Math. Phys. 161, 77-124 (1994) [118] E. H. Lieb, J. P. Solovej and J. Yngvason: Ground states of large quantum dots in magnetic fields. Phys. Rev. B 51, 10646-10665 (1995) [119] E. H. Lieb and W. Thirring: Inequalities for moments of the eigenvalues of the Schrodinger Hamiltonian and their relation to Sobolev inequalities. in "Studies in Mathematical Physics" (E. Lieb, B. Simon, A. Wightman, eds.), pp. 269-330, Princeton University Press (1975) [120] E. H. Lieb and H.-T. Yau: The stability and instability of relativistic matter. Commum. Math. Phys. 118, 177-213 (1988) [121] M. Loss and B. Thaller: Scattering of particles by long-range magnetic fields. Ann. Physics 176, 159-180 (1987) [122] M. Loss and B. Thaller: Short range scattering in long-range magnetic fields: the relativistic case. J. Diff. Equ. 73, 225-236 (1988) [123] M. Loss and B. Thaller: Optimal heat kernel estimates for Schrodinger operators with magnetic fields in two dimensions. Commun. Math. Phys. 86, 95-107 (1997) [124] M. Loss and H.-T. Yau: Stability of Coulomb systems with magnetic fields: III. Zero energy bound states of the Pauli operator. Commun. Math. Phys. 104, 283-290 (1986) [125] N. Macris: Spectral flow and level spacing of edge states for quantum Hall Hamiltonians. J. Phys. A: Math. Gen. 36, 1565-1581 (2003) [126] P. Malliavin: Minoration de l'etat fondamental de l'dquation de Schrodinger du magndtisme et calcul des variations C. R. Acad. Sci. 302, 481-486 (1986) [127] H. Matsumoto and N. Ueki: Spectral analysis of Schrodinger operators with magnetic fields. J. Funct. Anal. 140, 218-255 (1996) [128] K. Miller: Bound States of Quantum Mechanical Particles in Magnetic Fields. Ph.D. Thesis, Princeton University (1982) [129] K. Miller and B. Simon: Quantum magnetic Hamiltonians with remarkable spectral properties. Phys. Rev. Lett. 44, 1706-1707 (1980) [130] R. Montgomery: Hearing the zero locus of a magnetic field. Commun. Math. Phys. 168, 651-675 (1995) [131] A. Morame: Absence of singular spectrum for a perturbation of a two-dimensional LaplaceBeltrami operator with periodic electromagnetic potential. J. Phys. A 31, 7593-7601 (1998) [132] S. Nakamura: Gaussian decay estimates for the eigenfunctions of magnetic Schrodinger operators. Comm. Partial Diff. Eq. 21, 993-1006 (1996) [133] S. Nakamura: Lifshitz tail for 2D discrete Schrodinger operator with random magnetic field. Ann. Henri Poincare 1, 823-835 (2000) [134] S. Nakamura: Lifshitz tail for Schrodinger operator with random magnetic field. Commun. Math. Phys. 214, 565-572 (2000) [135] G. Nenciu: Dynamics of band electrons in electric and magnetic fields: rigorous justification of the effective Hamiltonians. Rev. Mod. Phys. 63, 91-127 (1991)
RECENT DEVELOPMENTS IN MAGNETIC FIELDS
427
[136] G. Panati, H. Spohn and S. Teufel: Effective dynamics for Bloch electrons: Pe%erls substitution and beyond. Commun. Math. Phys. 242, 547-578 (2003) [137]
J. Puig: Cantor spectrum for the almost Mathieu operator. Commun. Math. Phys. 244,
297-309 (2004) [138] G. Raikov and S. Warzel: Quasi-classical versus non-classical spectral asymptotics for magnetic Schrodinger operators with decreasing electric potentials. Rev. Math. Phys. 14, 10511072 (2002), and Spectral asymptotics for magnetic Schrodinger operators with rapidly decreasing electric potentials. C. R. Math. Acad. Sci. Paris 335, 683-688 (2002) [139] M. Reed and B. Simon: Methods of Modern Mathematical Physics. I-IV. Academic Press (1975, 1978, 1980)
[140] D. Robert: On scattering theory for long range perturbations of Laplace operators. J. Anal. Math. 59, 189-203 (1992) [141] G. Rozenblum and N. Shirokov: Infiniteness of zero modes for the Pauli operator with singular magnetic field. Preprint 2005, xxx.lanl.gov/math-ph/0501059 [142] H. Schulz-Baldes, J. Kellendonk and T. Richter: Simultaneous quantization of edge and bulk Hall conductivity. J. Phys. A: Math. Gen. 33, L27-L33 (2000) [143] R. Seiringer: On the maximal ionization of atoms in strong magnetic fields. J. Phys. A 34, 1943-1948 (2001)
[144] Z. Shen: Eigenvalue asymptotics and exponential decay of eigenfunctions for Schrodinger operators with magnetic fields. Trans. Amer. Math. Soc. 348, 4465-4488 (1996) [145] Z. Shen: On the moments of negative eigenvalues for the Pauli operator, J. Diff. Eq. 149, 292-327 (1998); 151, 420-455 (1999) [146] B. Simon: An abstract Kato's inequality for generators of positivity preserving semigroups. Indiana Univ. Math. J. 26, 1067-1073 (1977) [147] B. Simon: Kato's inequality and the comparison semigroups. J. Funct. Anal. 32, 97-101 (1979)
[148] B. Simon: Maximal and minimal Schrodinger forms. J. Op. Theory 1, 37-47 (1979) [149] B. Simon: Phase space analysis of simple scattering systems: extension of some work of Enss. Duke Math. J. 46, 119-168 (1979) [150] B. Simon: Schrodinger semigroups. Bull. Amer. Math. Soc. 7, 447-526 (1982) [151] B. Simon: Schrodinger operators in the twentieth century. J. Math. Phys. 41, 3523-3555 (2000)
[152] B. Simon: Functional Integration and Quantum Physics. second edition, AMS Chelsea Publishing (2005) [153] E. Skibsted: Asymptotic completeness for particles in combined constant electric and magnetic fields. II. Duke Math. J. 89, 307-350 (1997) [154] A. Sobolev: The quasi-classical asymptotics of local Riesz means for the Schrodinger operator in a strong homogeneous magnetic field. Duke Math. J. 74, 319-429 (1994) [155] A. Sobolev: Two-term asymptotics for the sum of eigenvalues of the Schrodinger operator with Coulomb singularities in a homogeneous magnetic field. Asymptotic Anal. 13, 393-421 (1996)
[156] A. Sobolev: Lieb-Thirring inequalities for the Pauli operator in three dimensions, IMA Vol. Math. Appl. 95, 155-188 (1997) [157] A. Sobolev: Quasiclassical asymptotics for the Pauli operator, Commun. Math. Phys. 194, 109-134 (1998)
[158] A. Sobolev: Absolute continuity of the periodic magnetic Schrodinger operator. Invent. Math. 137, 85-112 (1999) [159] V. Sordoni: Gaussian decay for the eigenfunctions of a Schrodinger operator with magnetic field constant at infinity. Comm. Partial Diff. Eq. 23, 223-242 (1998) [160] H. Tamura: Asymptotic distribution of eigenvalues for Schrodinger operators with magnetic fields. Nagoya Math. J. 105, 49-69 (1987) [161] L. E. Thomas: Time dependent approach to scattering from impurities in a crystal. Commun. Math. Phys. 33, 335-343 (1973) [162] N. Ueki: Lower bounds for the spectra of Schrddinger operators with magnetic fields. J. Funct. Anal. 120, 344-379 (1994). Erratum: J. Funct. Anal. 127, 257-258 (1995)
[163] N. Ueki: Estimates on the heat kernel of the Pauli Hamiltonian and its application to problems on hypoellipticity of the 8b-Laplacian. Math. Z. 239, 69-97 (2002)
428
L. ERDOS
[164] W.-M. Wang: Microlocalizataon, percolation and Anderson localization for the magnetic Schrodinger operator with a random potential. J. Funct. Anal. 146, 1-26 (1997) [165] W.-M. Wang: Supersymmetry and density of states of the magnetic Schrodinger operator with a random potential revisited. Comm. Part. Diff. Eq. 25, 601-679 (2000) [166] M. Wilkinson: An exact renormalisation group for Bloch electrons in a magnetic field. J. Phys. A 20, 1791 (1987)
MATHEMATISCHES INSTITUT, LMU, THERESIENSTRASSE 39, D-80333 MUNICH, GERMANY E-mail address: [email protected]
Proceedings of Symposia in Pure Mathematics Volume 76.1, 2007
Aspects of the Integer Quantum Hall Effect Gian Michele Graf To Barry, with admiration ABSTRACT. We review some of the concepts which shaped the understanding
of the integer quantum Hall effect, as well as the mathematical results they led to. Also described are the underlying physical motivations. Emphasis is placed on the equivalence of different approaches.
CONTENTS 1.
2. 3.
4.
Introduction Results Motivations Some Snapshots from the Proofs
References
1. Introduction The integer quantum Hall effect (IQHE) has been at the crossroads of several developments in mathematical physics, such as non-commutative geometry, index theory, localization, and the adiabatic theorem. Some of the most illuminating contributions to these topics are coauthored by Barry Simon. The fractional quantum Hall effect, which we shall not review here, further ties to conformal and topological field theory [21] and to the classification of certain integral lattices [19]. The IQHE appears in some two-dimensional samples at temperatures close to zero and in a strong, transverse magnetic field. The current density j and the stationary electric field E, both lying in the plane of the sample, are empirically related by the Hall-Ohm law
j=crE,
(1)
2000 Mathematics Subject Classification. 81V70 (primary); 82D30 (secondary). Key words and phrases. quantum Hall effect, index, conductance, localization. ©2007 American Mathematical Society 429
G. M. GRAF
430
which introduces the conductivity tensor a. The phenomenon is that, under appropriate conditions, transport is dissipationless, j E = 0, which means that the tensor is antisymmetric, or = I
0 CH
O)
(2)
Even more remarkably, the Hall conductance OH is quantized in multiples of e2/h, where e is the electron charge and h is Planck's constant (i.e., of 1/27r if natural
units are used); and the quantization of o H is accurate to within 10-8 when the magnetic field or the density of electrons are varied over a sizeable range, a fact called a plateau. Its width may be comparable to the separation between plateaus. Three basically distinct explanations have been proposed for the IQHE. Common to all, at least in their original formulation, is that they treat the electrons as independent particles, except for the Pauli principle. (i) The Hall conductance may be identified with the charged pumped across a closed ribbon when the magnetic flux threading it is increased by one flux quantum [26]. The Hamiltonians before and after that change are unitarily conjugate and the state has evolved adiabatically in between because the transport is dissipationless. The occupation of states follows the spectral flow as a function of the flux and that charge is the number of occupied eigenvalues having crossed the Fermi energy in the process. Clearly, that number is an integer. (ii) Linear response theory computes the current j induced by a weak electric field E in the bulk of the sample, which yields the Kubo formula for the Hall conductance. That expression can be related to a Chern number [33], which is an integer.
(iii) The Hall current is ascribed to states flowing at the edge of the sample. In special cases it may be identified with the number of edge channels [23, 10], which is an integer. The Hall conductance as defined on the basis of (i) has been linked [26, 23] to the current flowing along the ribbon when an electric potential is applied across it. That current may be interpreted as flowing either in the bulk [26] or at the edges [23] of the ribbon. In real experiments it is a combination of both possibilities [23]. Following their original formulation, approaches (i)-(iii) have not only gained in mathematical precision, but the physical concepts involved have been clarified as well. For instance, the argument [26] for (i) depended on the fact that eigenvalues moving down (resp. up) under the spectral flow are associated with the inner (resp. outer) edge. If the Hamiltonian is symmetric against rotations along the ribbon, as it was first assumed, then the eigenvalues are monotone in the flux throughout its variation, permitting the conclusion on charge transport stated above. If it is not, as for example implied by the presence of weak disorder, eigenvalue crossings formerly protected by symmetry may now be avoided, destroying monotonicity. This issue, which reflects the possibility of eigenstates tunnelling between edges, admittedly
remained to be investigated in [26]. It can be avoided altogether if the outer edge is pushed to infinity [4], which turns the ribbon into a punctured plane. In that geometry, the Hall conductance can be identified with the relative index [4] of a pair of projections. As another example, the argument [33] for (ii) assumes a periodic Hamiltonian and hence rational magnetic flux per unit cell and no disorder. This is however not needed, since the unit cell may be replaced by a torus of boundary
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
431
conditions [29] or fluxes [5] (in the latter case the approach extends to interacting particles). Another generalization [6] of the Chern character, more appropriate to the thermodynamic limit, is by means of non-commutative geometry. Disorder is crucial for the formation of plateaus. There, the Fermi energy varies within an interval where bulk states are localized, a so-called mobility gap. Such a variation changes the occupation of states (and hence the electron density), but not that of those participating in transport. By contrast, a variation over the extended states spectrum, or over a spectral gap, changes both (resp. neither). Disorder thus affects the analysis in two main ways: First, it destroys a symmetry, like the above mentioned rotations or periodicity. This applies even if the simplifying assumption is made that (a) the Fermi energy falls in a spectral gap. Second (b), the Fermi energy actually falls in a mobility gap. It it therefore of utmost importance that definitions of Hall conductance and their equivalence be compatible with disorder. For (i, ii) this was achieved in [6, 4]; for (iii, a) in [31, 24] and for (iii, b) in [16].
2. Results We shall momentarily present three definitions of Hall conductance related to the above pictures (i)-(iii). A mathematical setting, in which they are conveniently placed, is that of discrete Schrodinger operators [14]. The bulk is represented by the lattice Z2 E) x = (X1,X2) with Hamiltonian HB = HB on £2(Z2). Its matrix elements HB(x, x'), (x, x' E Z2), are of short-range in the sense that
IHB(XIx')I (eµl"I - 1) =: C1 < oo
sup
(3)
xEZ2 x'EZ2
for some It > 0, where xj = Ix11 + Ix21. A bounded, open interval A C IR, which shall contain the Fermi energy, is assumed to lie (a) in a spectral gap or, more generally, (b) in a mobility gap: (a)
Ano,(HB)=o;
(4)
(b) For some v > 0,
sup E jg(HB)(x, x')I(1 +
xj)-veWlx-x'1
< oc ,
gEBi(A) x x'EZ2
(5)
where B1(0) denotes the set of Borel measurable functions g which are constant in {AIA < Al and in JA IA > Al with jg(A)l < 1 for all A E R. In particular, the spectrum is pure-point in A [25]. Denoting by EM the characteristic function of M C IR, the assumption is completed by dimE{a}(HB) < oo
,
(A E A)
(6)
,
i.e., no eigenvalue in A is infinitely degenerate. Condition (5) is basically a statement about dynamical localization. It has been established in [1] and more explicitely in [30], where the above property is related to the SULE property, as well as in [2], where g is allowed to be constant, rather than zero, outside of A. The condition holds true almost surely for ergodic Schrodinger operators whose Green's function G(x, x'; z) = (HB -z)-1(x, x') satisfies a moment condition [3] of the form
limsup E(IG(x,x';E+irl)I')
EEO,,q-0
G. M. GRAF
432
for some 0 < s < 1, but can also be proved [22] using multi-scale analysis [20]. Condition (6) appears to be essential for a plateau, in view of the fact that for the Landau Hamiltonian (though defined on the continuum rather than on the lattice) QH jumps as the Fermi energy crosses an infinitely degenerate Landau level. The condition, in fact, simple spectrum, follows almost surely from the arguments in [32]. We stress however that the hypotheses (a) and (b) themselves are deterministic. Translation covariance or ergodicity of HB are not assumed here. The three definitions of Hall conductance shall now be associated with a Flux, the Bulk and the Edge. The physical motivations relating them to approaches (i)-(iii) mentioned before will be given in the next section. (i) The definition is based on the index of a pair of projections and depends on some unitaries associated with gauge transformations. Let P and Q be two orthogonal projections on a Hilbert space, so that P - Q is compact. Then
Ind(P, Q) = dim{o I Po = b, Q' = 0} - dim{ '% 1 Pz,b = 0, QO = '}
.
(7)
Let U(x), (x e Z2), satisfy IU(x)l = 1 and JU(x) - U(y)I < Cl Il
-x
,
(Ix - YI <-
(8)
for some C1, C2. Along a large loop encircling the origin counterclockwise,
the phases of U(y)/U(x), which are small for single bonds (x, y), add up to a multiple, N(U) E Z, of 27r. We assume that the winding number N(U) equals 1. Then aF(A)
where PA =
21Ind(PA, UPAU*)
(9)
,
(HB).
(ii) The definition makes use of switch functions: Let A(n), (n E Z), be a function which equals 0 for large negative n (resp. 1) for large positive n. Then 5B (A) = i tr PA [[PA, Al], [PA, A2]]
(10)
,
where Al and A2 are switch functions of x1 (resp. X2). (iii) The sample with an edge is modeled as a half-plane Z x 7Gq,, where Za _
{n E Z I n < a}, with the height a of the edge eventually tending to oo. The Hamiltonian Ha, = Ha on £2(Z X Za) is obtained by restriction of HB under some largely arbitrary local boundary condition, described as follows. Denoting by Ja : £2(Z X Za) --f £2(Z2) the extension by 0, we assume that
E
IEa(x,x')JeN'(Jx2+aI+Ix1-xiI)
< 00 .
xEZ aEZ2 x'EZxZ
We then set
vE = i lim lim trp'(Ha)[Ha,Al]A,?,a(A2) , ,7-0 a-+oc
(12)
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
433
where p E C°° (I[8) satisfies
P(A) =
1
(A
0,
(A > A)
(13)
and
A" .(X) = r f0
e-77te'x°tXe- Hatdt
is the average over a time - 17-1 of a bounded operator X with respect to the Heisenberg evolution generated by Ha. The main results are as follows: First of all, these quantities are well-defined for A E A. Second, they are independent of various auxiliary objects, such as U, A1, A2 and Ea. In particular, they do not change when U, A1, or A2 are replaced by some translates. For concreteness only, the reader may think of Up(x) = U(x - p) with U(x) = IX
(14)
X1
and p = (pl,p2) C Z2* = Z2 +
as well as Ai,P = O(xi - pi); for Dirichlet boundary conditions, Ho, = Ja HB Ja, the locality condition (11) holds true by (3). Third, the conductances are independent of A C 0 (resp. p) with (13), which is the manifestation of a plateau. Fourth, QF =QB =QE Finally, 27rO'F is manifestly an integer.
Some of these definitions were first formulated under slightly different assumptions. For instance, under assumption (a), the definition (12) of the edge Hall conductance can be replaced by the simpler QE = i tr p'(Ha) [Ha, Ai]
(15)
for any a c Z. In an ergodic setting, Al and A2 in (10) may be replaced by xl (resp. X2) and the trace by the trace per unit area, UB = i try Pa [[PA, xl], [Pa, x2]]
,
(16)
which takes an almost sure value [7]. The same replacement can be made in (15), with the trace now becoming trace per unit length [31]. It is now appropriate to review the genesis of these results. The definition (16) was proposed in [6] and identified as a 2-cocycle. It was shown to be an integer by relating it to a Fredholm index, through a formula of Connes [13]. In [4] that index was formulated as the index (7) of a pair of projections and endowed with independent physical motivation. It was proved to be equal to the expectation of (10), again in the ergodic setting and using Connes' formula, but without explicit reference to non-commutative geometry. Translation invariance of (9) was shown and used in [6, 4]. It was noted in [16] that the same is true for (10), and not just for its expectation, which allowed to turn QF = QB into a deterministic statement. In [31], which is placed in the ergodic setting and under assumption (a), QE was displayed as a winding number, which provided an independent reason for its quantization. It was also shown to be a 1-cocycle, from which the conclusion vE = vB was drawn using K-theory. The equality OE = QF, again for (a) but in a deterministic setting, was shown in [15, 27]. In the more general case (b), QE = trB
G. M. GRAF
434
is due to [16], with a special case due to [12]. In all this, the pioneering role of non-commutative geometry [6, 31] is manifest.
3. Motivations We provide the physical pictures underlying the definitions (i)-(iii) of Hall conductances. The discussion is largely heuristic. (i) In one of its guises the Laughlin argument relates the Hall conductance with the charge pulled from infinity when a flux quantum is slowly added near the origin. By the continuity equation, the charge Q inside a loop C changes at the rate dQ dt
_
J - v s,
where v is the outward normal. By Faraday's law, a change in the flux 4) is accompanied by an electric field,
i E Tds=where T is the tangent vector. If the change is slow, the field is nearly stationary, so that (1, 2) apply, resulting in AQ = aHO4). This is to be compared with the quantum mechanical computation of OQ. The state of the system is U(t, 0)PAU(t, 0)*,
where U(t, 0) is the evolution generated by the Hamiltonian, now depending on time through the flux. By the time, t = to, the flux has increased by 04) = 2ir the Hamiltonian becomes unitarily conjugate to H through U(x) with winding number 1. Its Fermi projection is UPaU*. In the limit of a large loop and of a slow process, the change OQ equals the excess number of electrons in the evolved many-body state, as compared with the ground state for the same flux: OQ = Ind(U(to, O)PAU(to, 0)*, UPAU*) = Ind(PA, UPAU*)
The second equation follows because of the additivity (Ind(P, R) = Ind(P, Q) + Ind(Q, R)) and the continuity (11P - QJJ < 1 Ind(P, Q) = 0) of the index, implying
Ind(U(t, 0)PaU(t, 0)*, PA) = 0 (ii) The Kubo formula for conductance is derived by adiabatically switching .
an electric field and considering the linear response of the system. The function -A2 can be seen as an electric potential of unit drop for a field pointing in the positive x2-direction, while the operator i[H, A1] stands for the total current in the x1-direction. The expectation of the latter in the state perturbed by the former yields o12i i.e., aH in eq. (2). The time-dependent Hamiltonian is
H(t) = H - A2f(t)
,
(t<0),
where f (t) slowly interpolates between 0 and 1, e.g., f (t) = e7t for some small 77 > 0. As in (i) the unperturbed density matrix is PA. The perturbed density matrix p(t) satisfies the initial value problem
dtp(t) = -i[H(t), p(t)]
,
t
li
e,Htp(t)e-'Ht = Pa
To first order, in the electric field the solution is
P(0) - Pa = i
f0 1-00
dt e'lte'Ht [A2 PA]e-'Ht
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
435
and we obtain, after an integration by parts, aB = lim ni tr[H, AI] (p(0)
PA) = li o irl tr
/0
J-00
dt e0t (e-,xtAie'xt - A1) [A2, PA]
(17)
Since PA is a projection, we have [A2, PA] = PA[A2, PA](1-PA)+(1-PA)[A2, PAPA.
The substitution of this into (17) amounts, by cyclicity, to the substitution of AI there by the expression (1 - PA)A1PA + PAAI (1 - PA) _ [[A,, PA], Pa]
The l.h.s. contributes two terms to (17) containing ef;Ht similar to one another, one of which is /0
irltr J
dte'tte-Wt(1
-00
- PA)AlPAe'Ht[A2, PA]
As we now sketch, it vanishes for rl -4 0. Representing the propagators as e-'xt = f e-'t`tdPN,, we are led to irl J
dt
ente-'(tt+-`)t
-00
=-
77
µ + irl
(18)
with µ+ > A, p_ < A. Since this quantity vanishes pointwise as rl -4 0 for (p+, p_) in the stated region, one is tempted to conclude that QB = -i tr [[AI, Pa], Pa] [A2, PA] = i tr PA [[AI, Pa], [A2, Pa]]
which is (10). The passage from (17) to (10), or rather its analogue in the ergodic setting, cf. (16), has been put on a firm basis; see [7], but also [2, 9]. In the present setting the result can be obtained using methods of [17]. In both cases it is crucial that A lies in a mobility gap, which allows one to control the small denominator in (18).
(iii) For a simpler start let us first discuss the definition (15) for QE valid in the case of a spectral gap. We interpret p(Ha,) as the 1-particle density matrix of a stationary quantum state. Though some current is flowing near the edge, we should discard it, as it is supposed to be canceled by current flowing at an opposite edge
located at x2 = -oc. The idea is that the two opposite edges of a macroscopic sample are infinitely separated from a microscopic perspective, and we focus on one of them. The drop in electrical potential used in (ii) is now given a different physical realization: The Fermi energy is lowered by 6 at the first edge, but not at
the second. Then a net current I = itr((p(Ha + 6) - p(Ha))[Ha,AI]) is flowing, resulting in QE = lira
I
i tr p'(Ha) [Ha, All
The current operator i[Ha, Al] is relevant only on states along a strip near xI = 0, and p'(Ha) only near the edge x2 = a, because p'(HB) = 0 due to (4). The intersection of the two strips is compact, which is basically why the trace exists. In the presence of a mobility gap, however, this property of p'(Ha) fails. In search of a proper definition of QE for this case, we consider only the current flowing across the line xI = 0 within a finite window 0 5 x2 < a next to the edge. This
G. M. GRAF
436
amounts to modifying the current operator to be i[Ha, A1]A2 (or a symmetrized version thereof), with which one may tentatively use lim itrp'(Ha)[Ha,A1]A2 a-oo
(19)
as a definition for QE. Though this limit exists, it is not the physically correct choice. States in the range of p'(Ha) supported far away from the edge are close to bound states of the bulk Hamiltonian, HB'OA = Ana, or linear combinations thereof. Such states may carry persistent currents (whence the operator in (15) is no longer trace class), but no current across the line xl = 0, since (OA, [HB, A1] 5A) = 0.
This cancellation is the rationale for ignoring the part x2 < 0 of the line x1 = 0 by means of the cutoff A2 in (19). However, the cancellation is not achieved on states located near the end point x = (0, 0). The contribution missed by (19) is (0A, i[HB, A1] (1 - A2)6) = -(z/'A, i[HB, A1]A2ba) from each bound state. By weighting them with p'(A), we amend the definition (19) of the edge conductance:
QE= lim itrP'(Ha)[Ha,A1]A2-i E p'(A)trE{a}[HB,A1]A2E{a} a-.oo
(20)
.
AEE&
The sum, which is over the eigenvalues in A, happens to be absolutely convergent,
but there is no general reason for it to vanish. In fact, it can be shown to be non-zero for the Harper Hamiltonian with a Cauchy distributed random potential. Alternatively, one may use in (19) and instead of A2 a cutoff which commutes with the dynamics generated by Ha, at least in some limit. Its use will not create spurious contributions which call for compensation. Such a possibility is realized by the time average of A2 and leads to definition (12). Unlike (20), it is stated purely in terms of the edge Hamiltonian Ha. Nevertheless, the two definitions agree [16]. We also remark that the time averages in eqs. (12) and (17), though of different physical origin, are mathematically related, which is instrumental to the proof of QE =QB.
4. Some Snapshots from the Proofs 4.1. The Equality QF = 0B in Case (b). In the case of finite-dimensional projections P and Q, the index can be computed as Ind(P, Q) = tr(P - Q). The generalization [4] to the infinite-dimensional case is
Ind(P, Q) = tr(P - Q)2-+1 if P - Q E .72,+1 for some odd integer 2n + 1, where JP, (1 < p < oo), are the trace ideals. In particular, tr(P - Q)3 = tr(P - Q) if P - Q E J1, which can be seen from the identity
(P - Q) - (P - Q)3 = [PQ, QP] = [PQ, [Q, P - Q]]
(21)
In the application to the QHE, where P = PA, Q = UP,\ U*, the difference luckily is not trace class, since the contrary would imply tr(PA - UPAU*) = 0 by evaluating the trace in the position basis. However, the third power of this difference is trace class, which yields QF =
tr(PA - UPAU*)3 = 2,7r
P, (x, y)PA(y, z)PA(z, x)S(x, y, z) x,y,zEZ2
S(x y, z) = - 2'(1- U(y)) (1 - U(z)) (1 - U(-))
,
(22)
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
437
To show its equality with
vB=i
Pa(x,y)P (y,z)Pa(z,x)[(Ai(y)
Ai(x))(A2(z)-A2(y))
(1 H 2)]
,
x,y,zEZ2
(23)
the authors of [4] assumed that the projection is ergodic (or covariant) with respect to magnetic translations. Translation invariance of QF then implies that (22) is a translation invariant function of the randomness and almost surely equal to its expectation. After taking the expectation of both equations (22), (23), the expressions IE(PA(x, y)PP11(y, z)PA(z, x)) are constant under a common shift a of the
summation variables x, y, z. By trading one them against a, the sums over the latter involve only U (resp. Ai). Moreover, for U = xl lxl, they are related through a formula of Connes, to be discussed below. The result is UF = IE(UB). A slightly different use of translation invariance, which does not depend on ergodicity, was made in [16]. Let U = Up and Ai = Bi,p as in (14). Since in fact both OF and aB are independent of p, averaging of (22), (23) over AL = {p E Z2* pJ < L} results I
in OT = 7rL 2
>
Px(x, y)Pa(y, z)Px(z, x)S(p, x, y, z)
,
(24)
,
PEAL
x,y,zEZ2
S(p, x, y, z) = sin L(x, p, y) + sin L(y, p, z) + sin L(z, p, x)
(25)
,
respectively in QB =
L2
/
.
P (x 71)P1(91 z)P" (z x). ,
,
,
PEAL
x,y,zEZ2
.[(B(yl -PI) - 9(x1 - pi))(0(z2 - P2) - O(y2 -P2)) - (1 H 2)]
.
(26)
These two expressions, which do not depend on L by derivation, will be shown to be equal in the limit L -p oc. Because the decay (5) applies to PA, the summation ranges x E Z2, P E AL can then be replaced by x E AL, P E Z2*. At this point, Connes' formula [13] S(p, x, y, z) = 2 Area(x, y, z) 11
(27)
pEZ2.
may be used in (24), where Area(x, y, z) is the triangle's oriented area, namely a (x - y) A (y - z). On the other hand, the corresponding sum in (26) also yields 2 Area(x, y, z), since E (O(yi
- pi) - 0(xi - pi)) = xi - yi
p,EZ*
The proof of QF = aB is completed by P,\L (y, z) = 6yz - PA(y, z).
4.2. Connes' Formula. The role of the sine function in (27), (25) is less special than one might think, as noted by [11]: For a fixed triplet u(I), u(2), u(3) E Z2, let cxi(p) = L(u(i+I), p, u(i+2)) E (-7r, 7r) be the angle of view from p c 7G2* of u(i+2) relative to 0+I) (with cxi(p) = 0 if p lies between them). Let g(a) be a bounded function satisfying g(-a) = g(a) and
g(a) = a + 0(03)
(28)
G. M. GRAF
438
near a = 0. Then, g(ai(p)) = 27rArea(u(') u(2) u(3))
.
(29)
PE7Z2* i=1
The proof is as follows. We may assume the triangle to be positively oriented. The statement (29) is true for g(a) = a. Indeed, for each p E 7G2*, 3
i=1
inside
1
ai(p) = 27r
1/2
on the boundary of outside
for p
0
the triangle.
(30)
Thus, for g(a) = a the l.h.s. of (29) is 27rx the number of dual lattice sites within the triangle (counting a boundary site with weight 1/2). This number equals the triangle's area. The above observation reduces (29) to the statement that for f (a) = g(a) - a 3
E f(ai(p)) = 0 . PEZ2 i=1
(31)
A significant difference between f and g is that the individual terms f (ai(p)) are summable in p E Z2, since by (28) f (ai(p)) = O(IpJ-3) for JpJ -i oo. However, each of the three individual sums changes sign under the reflection with respect to the midpoint of the corresponding edge, (u(z+1) + u(i+2))/2 E (Z/2)2 (which is a symmetry of the lattice Z2). Thus, even the individual sums (at given i) vanish.
4.3. The Equality QF = QE in Case (a). In this setting QE is given by (15), without the need to pass to the limit a -f oo. We may thus take Ho,=o on $2(Z x Z0) as the edge Hamiltonian. Unlike for HB, the interval A is not a spectral gap for HO. Its spectrum in A may actually be absolutely continuous [28, 8, 18], which is a manifestation of states extending along the edge. As a result, the matrix elements of a spectral projection (Ho), (A E A), will no longer decay rapidly away from the diagonal, which is the property that in the bulk case ensured
P -UPU* E .73
(32)
for P = PA = E(_,,,A) (HB) and U as in (8). By contrast, we also have PA = p(HB) with p as in (13) due to assumption (a), and the property (32) extends to P = p(Ho). The price to pay is that P is no longer a projection, but that does not seem to be a crucial aspect of the Laughlin argument. Morally, we may identify OF with (27r)-1 tr(P - UPU*), while eq. (21) is cautioning us that the correct computation of the trace is, in the case of projections, by discarding the trace of the commutator on the r.h.s. which, though not defined, is formally zero. That identity reads [(1- P)(1- Q), (1 - Q)(1 - P)] (P - Q) - (P - Q)3 = 2 [PQ, QP] 2 + (1 - 2P)(P - P2) - (1 - 2Q)(Q - Q2)
+2{P-Q,P-P2+Q-Q2} if it is not restricted to projections, where
denotes the anticommutator. It
suggests considering the expression
K(U) := tr(2 {P - Q, (P - P2) + (Q - Q2)} + (P - Q)3)
(33)
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
439
as a replacement for tr(P - Q) when P and Q = UPU* are unitarily conjugated. More precisely, we consider unitaries U which are multiplication operators with respect to some fixed basis, like the position basis of P2(Z x Zo), and operators P P* such that (P - Q)3, (P - Q) (P - P2), p(P) - p(Q) E 3 for p(A) = A - A2 and p(A) = (1 - 2A) (A - A2). We remark that these properties are satisfied for P = p(Ho) and for U as in (8), if U = 1 on all but a finite piece of the edge. For
instance, (P - Q)(P - P2) is then associated with such a piece, because p - P2 vanishes away from the edge. Hence it is trace class. The important property is that K(U) is unaffected by changes of U which are trace class. This is used as follows: Let U - 1 be supported in a cone whose rays point into the lower half-plane Z x Z0, and let its curl, i.e., the magnetic flux, be concentrated near the vertex. Moving the vertex without changing the fan of the cone is an example of such a change of U. If the vertex too is placed well inside the lower half-plane, the cone does not intersect the edge and the first term in (33) is negligible. In this limit K(U) reduces to 27rOF. If, on the other hand, the vertex is pulled across the edge and well into the upper half-plane, then the second term in (33) is associated with the intersection of the cone with the lower half-plane, which in the limit is a negligible (though infinite) tail. Moreover, and still inside the lower half-plane, U may be represented as an exponential, which near the edge is of the form e2'riA' W. The remaining first term in (33) can then be computed as
3 tr(P - Q) ((P - P2) + (Q - Q2))
6 tr(P - Q) (p - P2) f27r lzz 6
J
dcp- tr(P - ei`°A1 Pe i"oA1) (P - P2)
= 6 27ri tr[P(Ho), A1] (p(Ho)
- P(Ho)2)
= 6 2iri tr[Ho, Al]P'(Ho) (P(Ho) - P(Ho)2) = 27ritr[Ho, AI]p'(Ho) = 27raE , where p = 3p2 - 2p3 satisfies the same assumption (13) as p does.
4.4. The Equality QE = 0B in Case (b). We sketch some of the steps towards this identity when OE is defined by (20). If the other definition (12) is chosen, the argument runs along similar lines. The statement may be rephrased as
lim itrp'(Ha)[Ha,A1]A2=UB+i > p'(A)trE{A}[HB,AI]A2E{a}
a-.oo
(34)
AEE&
The operator on the l.h.s. is geometrically associated with the finite but growing portion 0 < x2 < a of the the line x1 = 0. It therefore has no chance to converge in trace class norm as a oo. To see that its trace nevertheless does, we look for an operator Z(a) E J with tr Z(a) = 0, and replace the operator with iP (Ha)[Ha,A1]A2 - Z(a)
,
(35)
hoping that convergence in that norm now holds true. A first attempt is Z(a) = i[p(Ha), AI]A2i which satisfies the two requirements; in particular, its trace is seen to vanish by computing it in the position basis. A partial cancellation between the
G. M. GRAF
440
two terms is made manifest using the Helffer-Sjostrand representations
P(Ha) =
2' f d2zazp(z)Ra(Z) P (Ha) = i fd2Z3p(Z)Ra(Z)2, where Ra(z) = (Ha -
z)-1
,
and p(z) on the r.h.s. is a quasi-analytic extension of
p(x). It yields in fact [P(Ha),A1]A2 = -27r f d2zazp(z)Ra(z)[Ha,A,]Ra(z)A2
,
(36)
P (Ha)[Ha,A1]A2 = -27r fd2zap(z)Ra(z)2[Ha,Ai]A2.
The two expressions would look even more similar if, in the second line, one power of the resolvent could be moved to the end of the expression. This however is just a commutator which may be absorbed into a redefinition Z(a). Then (35) reads
- i f d2za2P(z)Ra(z)[Ha,A1](A2Ra(z)-Ra(z)A2) _ --L f d2zazp(z)Ra(z)[Ha,A1]Ra(z)[Ha,A2]Ra(Z) 27r
27r
(37)
This expression is geometrically associated with the intersection of the lines x1 = 0 and x2 = 0, which is independent of a. It is therefore reasonable that it has a limit as a ---> oo, which is indeed obtained by replacing the subscript a with B. It remains to show that the trace of the bulk quantity T so obtained equals the r.h.s. of (34). To this end we use (37), (36) in reverse, but now with a - B, and obtain T = -i[p(HB), A1]A2
- 27r1 fd 2 zazP(z)R(z)[Hs,A1]A2R(z)
.
Unlike for a < oc, the two terms are not separately trace class. We next compare the expression with (17): While the first term, e-;xtAleiH1 is necessary to ensure that the whole expression is trace class, it is formally only the second one which contributes to the trace, as explained there: QB = -i tr Al [A2i Pa], or O'B = i tr[A1 i PA]A2. These expressions are not well defined, but the following is a correct representation for QB:
aB(Ao) = -itrE_[Paa,A1]A2E_ -itrE+[Paa,A1]A2E+ - E i tr E{a} [Pao, A1]A2E{a} ,
(38)
AEE&
where E_, E+ are the spectral projections for HB onto {A A < Al (resp. {A A > 0}). Since QB(Ao) is independent of A0 E 0, we may replace PAO by p(HB) in (38). We then frame T similarly with Ej_, E{A}, without changing its trace. The first term is then just 0B; the contributions with E± from the second vanish because E+R(z) and R(z)E+ are analytic on the support of p(z) or of p(z) - 1. The remaining contribution is I
-1 27r
f d2za2P(z)(A - z)-2 trE{A}[HB, A1]A2E{A} , AEE&
which equals the last term in (34).
ASPECTS OF THE INTEGER QUANTUM HALL EFFECT
441
Acknowledgements. This contribution would not have been possible without all I learned from collaborations with M. Aizenman, Y. Avron, P. Elbau, A. Elgart, J. Frohlich, and J. Schenker, to all of whom I am indebted. I thank A. Elgart for a critical reading of the manuscript.
References [1] M. Aizenman. Localization at weak disorder: some elementary bounds. Rev. Math. Phys., 6:1163-1182, 1994.
[2] M. Aizenman and G. M. Graf. Localization bounds for an electron gas. J. Phys. A, 31:67836806, 1998.
[3] M. Aizenman and S. Molchanov. Localization at large disorder and at extreme energies: an elementary derivation. Comm. Math. Phys., 157:245-278, 1993. [4] J. E. Avron, R. Seiler, and B. Simon. Charge deficiency, charge transport and comparison of dimensions. Comm. Math. Phys., 159(2):399-422, 1994. [5] J. E. Avron, R. Seiler, and L. G. Yaffe. Adiabatic theorems and applications to the quantum Hall effect. Comm. Math. Phys., 110:33-49, 1987. [6] J. Bellissard, K-theory of C*-algebras in solid state physics, in Statistical Mechanics and Field Theory: Mathematical Aspects. pp. 99-156. Lecture Notes in Physics 257. Edited by T. Dorlas, M. Hugenholtz, and M. Winnink, Springer-Verlag, Berlin, 1986. [7]
J. Bellissard, A. van Elst, and H. Schulz-Baldes. The noncommutative geometry of the
quantum Hall effect. J. Math. Phys., 35:5373-5451, 1994. S. De Bievre and J. V. Pule. Propagating edge states for a magnetic Hamiltonian. Math. Phys. Electr. J. 5, 1999. [9] J.-M. Bouclet, F. Germinet, A. Klein, and J. H. Schenker. Linear response theory for magnetic Schrodinger operators in disordered media. J. Funct. Anal. 226:301-372, 2005. [10] M. Biittiker. Absence of backscattering in the quantum Hall effect in multiprobe conductors. Phys. Rev. B, 38:9375-9389, 1988. [11] Y. Colin de Verdiere. Private communication, reported by R. Seiler. [12] J.-M. Combes and F. Germinet. Edge and impurity effects on quantization of Hall currents. Comm. Math. Phys., 256:159-180, 2005. [13] A. Connes. Noncommutative differential geometry. Inst. Hautes Etudes Sci. Publ. Math., [8]
62:257-360, 1985. [14] H. L. Cycon, R. G. Froese, W. Kirsch, and B. Simon. Schrodinger Operators with Application to Quantum Mechanics and Global Geometry. Texts and Monographs in Physics. SpringerVerlag, Berlin, 1987.
[15] P. Elbau and G. M. Graf. Equality of bulk and edge Hall conductance revisited. Comm. Math. Phys., 229:415-432, 2002. [16] A. Elgart, G. M. Graf, and J. H. Schenker. Equality of the bulk and edge Hall conductances in a mobility gap. Comm. Math. Phys., 259:185-221, 2005. [17] A. Elgart and B. Schlein. Adiabatic charge transport and the Kubo formula for Landau-type Hamiltonians. Comm. Pure Appl. Math., 57:590-615, 2004. [18] J. Frohlich, G. M. Graf, and J. Walcher. On the extended nature of edge states of quantum Hall Hamiltonians. Ann. H. Poincare, 1:405-442, 2000. [19] J. Frohlich, T. Kerler, U. M. Studer, and E. Thiran. Structuring the set of incompressible quantum Hall fluids. Nucl. Phys. B, 453:670-704, 1995. [20] J. Frohlich and T. Spencer. Absence of diffusion in the Anderson tight binding model for large disorder or low energy. Comm. Math. Phys., 88:151-184, 1983. [21] J. Frohlich and U. M. Studer. Gauge-invariance and current-algebra in nonrelativistic manybody theory. Rev. Mod. Phys., 65:733-802, 1993. [22] F. Germinet and S. De Bievre. Dynamical localization for discrete and continuous random Schrodinger operators. Comm. Math. Phys., 194:323-341, 1998. [23] B. I. Halperin. Quantized Hall conductance, current carrying edge states, and the existence of extended states in a two-dimensional disordered potential. Phys. Rev. B, 25:2185-2190, 1982. [24]
J. Kellendonk, T. Richter, and H. Schulz-Baldes. Edge current channels and Chern numbers in the integer quantum Hall effect. Rev. Math. Phys., 14:87-119, 2002.
G. M. GRAF
442
[25] H. Kunz and B. Souillard. Sur le spectre des operateurs aux differences finies aleatoires. Comm. Math. Phys., 78:201-246, 1980/81. [26] R. B. Laughlin. Quantized Hall conductivity in two-dimensions. Phys. Rev. B, 23:5632-5733, 1981.
[27] N. Macris. On the equality of bulk and edge conductance in the integer Hall effect: microscopic analysis. Unpublished.
[28] N. Macris, P. A. Martin, and J. V. Pule. On edge states in semi-infinite quantum Hall systems. J. Phys. A, 32:1985-1996, 1999. [29] Q. Niu, D. J. Thouless, and Y.-S. Wu. Quantized Hall conductance as a topological invariant. Phys. Rev. B, 31:3372-3377, 1985.
[30] R. del Rio, S. Jitomirskaya, Y. Last, and B. Simon. Operators with singular continuous spectrum. IV. Hausdorff dimensions, rank one perturbations, and localization. J. Anal. Math., 69:153-200, 1996.
[31] H. Schulz-Baldes, J. Kellendonk, and T. Richter. Simultaneous quantization of edge and bulk Hall conductivity. J. Phys. A, 33(2):L27-L32, 2000. [32] B. Simon. Cyclic vectors in the Anderson model. Rev. Math. Phys., 6:1183-1185, 1994. [33] D. J. Thouless, M. Kohomoto, M. P. Nightingale, and M. den Nijs. Quantized Hall conductance in a two-dimensional periodic potential. Phys. Rev. Lett., 49:405-408, 1982.
THEORETISCHE PHYSIK, ETH-HONGGERBERG, CH-8093 ZURICH, SWITZERLAND
E-mail address: [email protected]
Proceedings of Symposia in Pure Mathematics Volume 76.1, 2007
Barry Simon's Work on Electric and Magnetic Fields and the Semi-Classical Limit Ira W. Herbst Dedicated to Barry Simon on his 60th birthday ABSTRACT. We review some highlights of Barry Simon's work on Schr6dinger operators with external electric and magnetic fields. In addition, we summarize Barry's work on the spectrum of -A+A2 V (x) as A -> oo where V has multiple
wells. In particular, we outline one of his arguments giving a formula for the asymptotics of the splitting of the ground state eigenvalue in the case of a double well.
CONTENTS
Magnetic Bottles The Zeeman Effect in Hydrogen Enhanced Binding Translation Invariance Universal Diamagnetism Schrodinger Operators with Constant Electric Field The Rayleigh-Schrodinger Perturbation Series One-Dimensional Electric Fields Semi-Classical Analysis References 1.
2. 3. 4. 5. 6. 7. 8. 9.
It was an exciting time back in the seventies when I was a postdoc at Princeton, and perhaps the most exciting and rewarding experience for me was my interaction with Barry Simon and Yosi Avron. This began with a three-person collaboration on Schrodinger operators with magnetic fields. We wrote a series of six ([5] -p [10]) papers, and I will first try to give you some highlights of these works with some additional comments about Barry's work on magnetic fields in which I was not involved. 2000 Mathematics Subject Classification. 46N50, 35P05. Key words and phrases. Schrodinger operator, external field, double well. ©2007 American Mathematical Society 443
L W. HERBST
444
I will then present some of Barry's work on Schrodinger operators with electric fields and finally turn to highlights of his work on the semi-classical limit. I apologize in advance for not including an up-to-date list of references for the material considered here. For a more complete set of references on magnetic fields, please refer to the contribution of Laszlo Erdos in this Festschrift.
1. Magnetic Bottles The usual intuitive notion of compactness for bounded operators in L2(Rn) goes back to the idea that if the symbol of a pseudo-differential operator, p(x, ), tends to zero as jxj + oo, then the operator should be compact. Thus, it may have surprised some to learn that although for no magnetic field B = da does
(( - a(x))' + i)_1 , 0 xj -- oo, there are many magnetic fields for which the resolvent of H = (-iV -a(x))2 is compact. In [5] it is shown that if B(x)j - oc as xj -- oo in such
as
a way that B(x)/IB(x)l does not twist too wildly, then H has a compact resolvent. This is based on the simple commutator estimate (,P, H
THEOREM 1 ([5]). Suppose B is continuous and {oa} is a covering of R by cubes of fixed size with A, centered at a c 1R' . Suppose for each a there are two unit vectors ea and e', such that inf B(x)(e,,, ea) --* oc as jal -+ oo. xEO Then H = (-iV - a(x) )2 with B = da has a compact resolvent. Theorem 1 is a precise expression of the consequence of a localized zero point energy becoming large as jx - 3 oo. This zero point energy is absent with the Pauli operator (o- (p - a(x) ))2, p = -iV, but bound states can occur there for other reasons [37, 21].
Magnetic fields such that H has a compact resolvent are dubbed magnetic bottles of the third kind in [5]. A simple example in three dimensions is B (x = (x 1i x2, -2x3). See [29] for further developments in understanding under what conditions H = (-iV - a(x) )2 + V (x) has a compact resolvent. There is an example of Miller and Simon [38] in two dimensions which shows
that 0(,) behavior at infinity of the magnetic field B is borderline for certain spectral behavior: They consider magnetic fields essentially given by b/jxjry with b a non-zero constant and prove THEOREM 2 ([38]).
(a) If 0 < y < 1, H has dense point spectrum in [0, oo). (b) If 1 < 'y, H has purely absolute continuous spectrum. (c) If ry = 1, the spectrum of H is dense point in [0, b2] and absolutely continuous in (b2, oo).
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
445
In arbitrary dimension n > 2, if the 1-form B(x)(x, ) is o(1) at infinity, then H has no eigenvalues [20]; thus indeed, behavior is borderline. The latter result is independent of even or odd dimension. There is a conjecture in [5] which is still open as far as I know:
In R3, if B(x) is continuous and o(1) at infinity, then H has some continuous spectrum. Intuitively, one expects the particle to be able to follow some field lines out to infinity.
Let us indicate how Miller and Simon prove (c) above because there are two interesting ideas there: We ignore the singularity at x 0 and write
H=(p-a)z=-0+b2_2L, r=xJ, r L = -i8/89 in polar coordinates. The first step is to write H as the direct sum of operators
H=®°°-_ (-A + b 2
2rm
where 'H n is the kernel of (L - m). General considerations thus show the spectrum of H is absolutely continuous in (b 2, oc) and pure point in [0, bz). The second ingredient is the general statement that no matter how a(x) behaves at infinity, aegg((p -a) 2) = [0, oo) if lim IB(x)l = 0. IxHoo
The proof of this statement uses a Weyl sequence fin, and the ability to build a gauge transformation a -- a + V fn = an, into 0,,, where a,,, is small on the support of 0n. In the case at hand, this shows the spectrum is dense pure point in [0, b2].
2. The Zeeman Effect in Hydrogen The Zeeman effect refers to the effect of a constant magnetic field on the structure of the spectrum of an atom. Consider the Hamiltonian for the hydrogen atom in a constant magnetic field:
H=2
(-i0
ea(x))z
- e2
B=V x a= Be3i
with B a non-zero constant. If a(x) = a B x x, then z
z
H=-Z O- BL3+e8B (x12 +x2)1
e2
Here L3 is the third component of the angular momentum. THEOREM 3 ([7, 9]).
(1) The ground state of H has L3 = 0(2) The binding energy for the ground state is monotone increasing in B. (3) The Rayleigh-Schrodinger series for the eigenvalues of H (a power series in Bz) is Borel-summable to the correct eigenvalue.
I. W. HERBST
446
The first two results were proved using FKG [22] inequalities and the third using dilation analyticity at a fixed L3. The FKG inequalities imply the following: Suppose V and W are functions of xI j, ... , Ixnl and in the region {x : xj > 0 for all j}. aia,V(x) < 0 if i j and
aiW > 0 all i. Suppose 1 C Rn is such that 1p depends only on ixI I,-, Ixnj and is decreasing in each xjj. Then the normalized ground states, 'v and "v+w, of -A + V and -A + V + W satisfy
f f IOv+w(x)12dnx >
J
IOv(x)I2dnx.
(1)
For (1), see [7] and [32]. Here is an outline of the proof of (1) and (2) of Theorem 2: We add cz2, z = x3i to H, go to a subspace of fixed L3 = £, and make a simple unitary transformation to give the operator he,, = (2m)
1
( 1\
a2
aa2z2
t2
ape
41
e2 B2
eB P - e 2 (p2 +z2)-12 +Ez - pt
p2
+ 8m
p2
2
(2)
in L2 (R+ x ]R, dp dz). Here p = xl -+x22. Assume B > 0. The ground state ')e of he,, will have energy E(2, e) satisfying
at
\
(mp
B
The ground state of H + cz 2 is infinitely degenerate when the Coulomb potential
is removed so it is no surprise that '9E = 0 for 2 > 0 with no Coulomb term. According to the inequality (1), if we add back in the Coulomb term p-2 t) increases. Thus, taking e 10, we obtain the result. Similarly, to prove monotonicity of the binding energy with JBI, differentiate es - E(0, c) with respect to CBI. This derivative is zero without the Coulomb term '2-M and increases when the Coulomb term is added back in. For additional material on the ordering of the energy levels in this system, see [12] and [26]. A formal series (usually not convergent for x
0)
00
E anxn n=0
is said to be Borel-summable to f (x) for small positive x if the Borel transform
B(x) _
anxn/n!
has an analytic continuation to a neighborhood of [0, oc) and
f (x) = J
B(tx)e-tdt
0 00
for small positive x. Sufficient conditions for this to be true (see [27] for Watson's
theorem and [39] or [51] for a more refined result quoted here) require f to be analytic in {z : Iz - R/21 < R/2} for some R > 0 and a bound
f(z) - Eanznl < ICzIN+I(N+ 1)! n=0
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
447
in this region. In [7] these conditions are proved using dilation analyticity and a simple formula for the eigenvalue involving contour integrals of the resolvent of the Hamiltonian. For additional work on the perturbation series for the eigenvalues in the Zeeman effect in hydrogen (in particular on Bender-Wu formulas), see [2] and [3].
3. Enhanced Binding Note that in two dimensions H = (-iV-a(x))2 with V x a = B = constant 0 has pure point spectrum with each eigenvalue infinitely degenerate. (The infinite degeneracy is because any Landau orbit can be translated anywhere in the plane without changing its energy. See below.) It seems then that by adding a scalar potential, binding in three dimensions should be easier in constant B field since it is only in the direction along B that a particle can escape to infinity. In one dimension an arbitrarily small attractive V will produce a discrete bound state ([42, 14]). This is not true in three dimensions if B = 0 but is true with B r 0 ([7, 10]). Consider first the one-body problem in three dimensions. Take
H = (-iV - 2 B x x)2 + V (x),
B = be3i
where b > 0 is constant. THEOREM 4 ([7, 10]). Suppose V is 0-compact and for large enough R, V (x)
0 for I xH > R. In addition, suppose V is not zero a.e. outside the cylinder x2 + x2 < R2. Then H has at least one eigenvalue below the continuum, and if V is azimuthally symmetric, H has infinitely many eigenvalues below the continuum.
We sketch a proof in the case V is also bounded with limb,,I-,, V(x) = 0. Choose a trial state 0(x) = f (x3)cpe(x1i x2), where L3W = fWe, f > 0 and cpe is a normalized ground state of the corresponding two-dimensional purely magnetic Hamiltonian which thus has energy b. One easily sees that Qess(H) = [b,oc). By the min-max principle, we can increase V, if necessary, to make V integrable. Note
that 2
(
,
(H - b) b) = Cf, (- dx3 + V )
f)
where Ve(x3)
=
I Ve(xl, x2)12V (XI, x2, x3)dxldx2.
J It is easy to see that f Vi (X3) dX3 < 0 for all large 2 (since Ve I2 has most of its weight far from the origin for these $). If we take f (X3) = fo(ax3) with fo E Co (R), and fo (0) = 1, we find
1li o(O, (H - b)O) = f Ve(x3)dx3 < 0.
This completes our outline of the proof. More physically relevant perhaps is a result in [7] about negative ions of multielectron atoms. We take units where the Hamiltonian of an ion of nuclear charge n - 1 and n > 2 electrons is n
H(b) j=1
i<j
I. W. HERBST
448
where B = be3, S is the sum of the electron spins, and H(b) acts in H, the antisymmetric subspace of ®;1 1L2(II83; C2).
THEOREM 5 ([7, 10]). Fix b > 0. Then the operator H(b) has infinitely many bound states in 7-l below the physical continuum. For small b, the binding energy of each such state is at least as large as cb3, (c > 0 depends on the eigenvalue of L3).
We give some ideas about the proof of the first statement of the theorem. Using an inductive proof, we first show that the neutral atom has a ground state rl of definite L3. It is important to have an estimate for the decay of rl but this is not difficult. The idea is to construct a trial function on the space of fixed large L3, where the nth electron is far from the neutral atom. We put this electron in a state yn)((an), rn = (xn, yn, zn), for large £ and small a and ( (1) = 1, ((-1)z = 0. But if we leave the other n - 1 electrons in state rl, the potential due to the shielded nucleus is not enough to give binding because of exchange terms whose signs are not known. We take into account an induced dipole moment of the neutral atom, due to the far away electron. The state we choose is b = Pcp, P = antisymmetrization, and n-1
cP=rl (1+Y(r
-al Z''I cc (xn, yn)S (an), j=1
where
g(r) = -j3 sgn(z)7(z)/(1 + Ir12),
and ry is a smooth even function, zero in a neighborhood of zero, but one in a neighborhood of infinity. We take $ large, a _ f 2 , C3 a positive constant and estimate
(0, (H(b) - E)0) Here E arises from moving the nth electron off to infinity in the z-direction with zero z-kinetic energy and putting it in a ground state Landau orbit with spin state ((a). For large £, this expectation is negative.
4. Translation Invariance Consider an N-body problem in constant B for example, with two-body potentials:
0 where the particles interact,
1( <j
The Hamiltonian is not invariant under xj - xj + c all j, but since V x a(x + c) = V x a(x), we can follow the usual translation by a gauge transformation to obtain invariance of H. The resulting unitary group {U(c)}cEa3 satisfies U(ci)U(c2) = ezQ"U(e2)U(el),
where Q = >j ej is the total charge and (P is the flux of B through the parallelogram with sides c1 and c2i 4 = B B. (c1 x c2). Thus, if Q = 0,
H = f H(k)d3k,
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
449
where k is a quasimomentum. In fact, H(k) = H(0) + 2M - l x B R [7], R = E eixi, M = E mi. (It turns out that in a quantized radiation field it is the quasimomentum + photon momentum which is conserved.) If Q 54 0, then as with a single particle in two dimensions, the generators of translations perpendicular to B satisfy the CCR. And thus, because of von Neumann's theorem on the canonical commutation relations, there is an infinite degeneracy in the strongest sense:
H=h®I, where magnetic translations perpendicular to B act in the second factor. If
0,
U(c) = 10 u(c). Note that the Hilbert space associated with the second factor must be infinitedimensional. This material can be found in [6].
5. Universal Diamagnetism Barry was the originator of some far-reaching inequalities involving magnetic Schrodinger operators. The first leads to universal diamagnetism for spinless particles:
THEOREM 6. Suppose a E Li c(Rn) and V-, the negative part of V, is bounded.
Let E(a) = inf a((-iV - a)2 + V). Then E(a) > E(0). PROOF. Suppose u E Co (JPn) (a form core for our Hamiltonian by definition). Then VIUI2
= 2 Re(u(V - ia)u),
so
V IuF + E2 _
Re(u(V - ia) u) u2 + E2
< I (V - ia)ul. Taking c 10 and integrating, we obtain 11 (V - ia)u112 + (u, Vu).
II V IU1112 + (u, Vu)
The result follows from the variational principle.
In fact, Theorem 6 follows from the more general result (here p = -iV) e-t((p-a)2+V)
f
<
e-t(-o+V)
If
(3)
which Barry first conjectured in [48]. Upon hearing Barry's conjecture, Ed Nelson supplied the path integral proof which appears in the same paper. Barry gave another very elegant proof [50] which simply uses the Trotter product formula [41] e-t(p-a)2
= s - hm M-00
(e-t- (pi-ai)2 ... e- -ML (per.-a. )2
and the fact that for j = 1, .... n, 2
for some (real) fj with ajfj = aj.
, /1
2
I. W. HERBST
450
There is much additional work on comparison of semigroups which Barry has been involved in (see [36, 49]). In [49], Barry proves the equivalence of a semigroup comparison such as (3) with a generalized Kato inequality for the generators of the two semigroups. In [36], Barry and Dirk Hundertmark prove the pointwise bound Ct((P a)N,,+V)
-e
t((p a)e
D,A+V) f <
(e-t(-Div,n+V)
- e-t(-OD,A+V) (4)
where the subscripts N, A indicate Neumann boundary conditions on the boundary of the open set A and similarly D, A indicate Dirichlet boundary conditions on M.
Here A is an arbitrary open set and a and V are very general. The inequality (4) is very useful in proving uniqueness of the integrated density of states under rather general conditions (see [36]). In contrast to (3), the proof [36] is not trivial, basically because there does not (yet?) exist a sufficiently useful path integral representation for the semigroup involving Neumann boundary conditions. An interesting conjecture of universal paramagnetism for fermions [35] which for the Pauli operator reads
inf spec((p - a)2 + o, B + V) < inf spec(-A + V)
(5)
was later shown to be incorrect by Avron and Simon in [11]. However, as is shown in [35], (5) is true if B = constant.
6. Schrodinger Operators with Constant Electric Field After some initial work on resonances for the two-body Stark problem ([24, 4, 30]), Barry and I followed this up with the more involved N-body problem, in particular, the Stark effect in atoms and molecules. For definiteness, consider Oj
H(f)=E j
2m7 .
+fe Eejxj+EVj(xi-xj) i<j
and H(f), the same Hamiltonian with center of mass motion removed. We assume the Vij are dilation analytic so that resonances can be defined as poles of matrix elements of the resolvent of H(f), continued to the second sheet. These matrix elements are, of course, between dilation analytic vectors. We proved the following stability result: THEOREM 7 ([33, 34]). Suppose E0 is a discrete eigenvalue of H(O) of multi-
plicity m. Then, given e > 0, there is an fE > 0 so that if 0 < f < fE, H(f) has m resonances (counting multiplicity) in the disk {A : JA - E0 < e}, and they all converge to E0 as f 10.
Unexplored problem. Are resonances of H(0) stable under perturbation to H(f )? This seems to be a very difficult problem in the framework of dilation analyticity.
7. The Rayleigh-Schrodinger Perturbation Series To deal with degeneracy, define an eigenvalue of H(0) as "normal" if on all symmetry subspaces the degeneracy, if any, is removed to first order in f (see below for a precise definition). Then
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
451
THEOREM 8 ([33, 34]). Suppose Eo is a normal discrete eigenvalue of H(O)
Then for any b > 0, there is an R5 > 0 and t functions E1(f),...,EQ(f) analytic for f in with multiplicity Q. {f
:
If I < Ra, -Min (p, 2) + 6 < arg f < 7 + Min cp,
2
where the V3 are dilation analytic in {B Im01 < (p}, which for real f > 0 are the resonances of H(f). The Rayleigh-Schrodinger series for the putative perturbed eigenvalues of H(f) are Borel-summable to the resonances of H(f).
The region of analyticity of the resonances has its origins in the region of analyticity in the two complex variables (f, 0) of the operator
-e-290 + f6 xee, which is
{(f,0): f 40, 0<argf+3Im0
FIG. 1
In contrast to the resonances of H(f), the coefficients in the Rayleigh-Schrodinger series are real; thus the recovery of the resonances from the Borel transform for real f has a slightly different form from that given in the discussion after Theorem 3 (see [33] and [51]).
I. W. HERBST
452
We also show a relationship between the asymptotics of the width of the resonance and the asymptotic behavior of the Rayleigh-Schrodinger coefficients an. For simplicity, for a non-degenerate eigenvalue of H(O) in a state of definite parity (assuming V3 (u) = Vj(-u))
E(f) ,,,
a2nf2n
with
F(f) + O(R-2n). a2= - II IR f2.ndf In joint work ([28, 17]), Barry and his co-authors derived rigorously the Oppenheimer formula, r(f)=4f-'e_ 3f
1
(I+0(f)))
for the width F(f) = -2 Im E(f) of the ground state resonance of hydrogen, but as far as we know, for multielectron atoms the leading behavior of r(f) is not rigorously known.
In most cases of interest the spectrum of the complex dilated operator H(f, 0) in the region of analyticity, f 0, 0 < 3 Im 0 + arg f < ir, I Im 0 1 < co, is purely discrete, consisting of resonances only. This is true unless for some non-trivial decomposition of the particles into clusters, the total charge to total mass ratios of all the clusters are all equal. Definition of normal eigenvalue of H(0): An eigenvalue E0 of H(0) is normal if there exist mutually orthogonal projections Qj commuting with the dilation gen-
erator, H(0), and X = e >j ej (xj - R) where R = Ej mj xj /(Y:j mj) such that with P = orthogonal projection onto the eigenstates of H(0) with eigenvalue Eo:
(i) Range >j Qj > Range P; (ii) P(QjXQj)P has non-degenerate eigenvalues on Range P. In practice the Qj are orthogonal projections onto eigenspaces of a set of generators of symmetry groups which commute with X, H(0), and the dilations. The proofs of Theorems 7 and 8 depend on being able to analyze the behavior of the resolvent (z - H(f, B))-1 for z near a discrete eigenvalue of H(0, 0) and for appropriate (f, 0) with If small. This presents interesting difficulties even in determining the spectrum of operators of the form HC1(f, 0) ® I + I ® H 2 (f, 0), where C1 and C2 are non-interacting subsystems. We used Weinberg-van Winter equations for appropriate semigroups, and a theorem of Gearhart [23] which allows one to determine the spectrum of a contraction semigroup e-tA on a Hilbert space from the behavior of the resolvent of A. See [31] for a review of this work where the analysis is somewhat simplified in that only the usual Weinberg-van Winter equations for resolvents are used. / In [25], Graffi and Grecchi treated Stark resonances in atoms obtaining similar results using different methods.
8. One-Dimensional Electric Fields There are three very interesting papers which Barry co-authored involving operators of the form z
H = - dx2 - fx+V(x), where V could be random. The first shows:
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
453
THEOREM 9 ([13]). Suppose f > 0, V C C2(II8) with V, V', and V" bounded. Then a(H) = Qa,c(H).
For me, the interesting thing about this result is that there is no restriction on the absolute size of V'(x) so that the force due to V could overwhelm the force due to the electric field, at least pointwise. Of course, on the average V'(x) is small in the sense that sup
1
a+r
aE2r I-,
V'(x)dx
0,
as r
oc,
and this is what is used in [13]. A similar result was proved previously in the half-line case in [52]. But if V is not smooth, the situation can be quite different as the other two papers co-authored by Barry show: THEOREM 10 ([18, 19]). Suppose 00
V(x,w) _
an(w)6(x - n), n=-oo
where {an} is a sequence of iid random variables with bounded probability density and zero mean. Then, for small f , a(H) = aPP (H) while for large f , app (H) = 0, almost surely.
In their proof of the above theorem, the authors note that the problem reduces
to a discrete one which has a lot in common with the model described by the Hamiltonian
-A + Ann-2',
A, random and i.i.d., A = discrete Laplacian in 1 dimension, which they also analyze. It is unknown in their electric field model whether for large f the continuous spectrum corresponds to diffusion. Although rather singular, the potential of Theorem 10 is still Laplacian form bounded with form bound zero. Thus it is interesting to ask how smooth a bounded potential needs to be to get a result such as Theorem 9. This question is answered in [16] (see also [40]).
9. Semi-Classical Analysis Beginning in 1983, Barry began a series of papers ([43] --- [47]) studying the low lying eigenvalues of the operator H(A) = - z A + A2V(X)
in the limit A T oc, where V is a smooth non-negative function with isolated zeros, and H(A) acts in L2(W'). He proved some beautiful results.
THEOREM 11 ([43]). Suppose the zeros of V are finite in number and nondegenerate in the sense that at each zero, a, [c5z3 V (a)] is positive definite. Suppose
in addition that V(x) > 6 > 0 for large jxj. Let Eo(A) < EI(A) < E2(A)...
be the eigenvalues of H(A) counting multiplicity (the number of such eigenvalues oc as A T oo). Order the eigenvalues eo < el < e2...
I. W. HERBST
454
(counting multiplicity) in the union of the spectra of the operators
+
2(aiajV(a))Xix31
as a varies over the zeros of V. Then
A 'Ej(A) = ej. This result is not far from intuition since if V(O) = 0, A-1H(.X) is unitarily equivalent to - a A + AV (x/ f) and the Taylor series of V at 0 gives
AV (x/V) _ EaaV(0)xixj +O(A-2). i,j
But Barry goes much further to give asymptotic expansions to all orders of the Ej (A) and their corresponding eigenfunctions. A much finer result is THEOREM 12 ([44, 45]). With the assumptions of Theorem 11 and in addition that V has only two zeros, a and b, let Sto (A) be the normalized ground state of H(A) and suppose that for any small e > 0,
liminf
f
x-aIGE
Aloo
IQo(A)(x)12dx > 0.
IQo(A)(x)12dx fx
(6)
Then
lim -A-' log(El(A) - Eo(A)) = p(a, b),
ATCO
where
f
T
p(x, y) = inf
f
T
a
T
V(ry(s))ds : -1(-T) = x, 'y(T) = y, T > 0 T (7)
Thus, p(a, b) is the infimum of the action for an orbit going from a to b with the potential -V. There is an orbit minimizing the action if we allow T = oo. This orbit is called an instanton. It is a solution of Newton's equation with potential -V, which goes from a to b in infinite time. Actually p(x, y) is better known to some as the Agmon metric p(x, y) = inf { f 1
2V('y(s)) l'Y(s)ids : 'Y(0) = x, 'y(1) = y1
(8)
The equality of (7) and (8) was shown by Carmona and Simon [15]. The assumption (1) is satisfied, for example, if the double well arises because of a Euclidean symmetry R of V such that Ra = b. Then for small E > 0, lim ATE
f
I
= lim f
p(, ()`)(x)I2dx = 2
ATE
po(A)(x)I2dx.
x-b)GE
Part of the proof of Theorem 12 uses large deviation results (a la DonskerVaradhan), but Barry also shows how to obtain the result using "standard" PDE methods. (We give a sketch below.) Another theorem of the same genre concerns periodic V.
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
455
THEOREM 13 ([46]). Suppose V is non-negative, smooth, and periodic with only one (non-degenerate) zero per unit cell. Let A(A) be the width of the ground state band. Then urn -A-' log 0(A) = min{p(a, b)
:
a and b are distinct zeros of V J.
Finally, Barry examines the effect of a seemingly negligible potential W in the situation of Theorem 11, but where V has only two zeros, a and b, connected by a Euclidean symmetry of order 2. W is assumed smooth, non-negative, and zero in a neighborhood of a and b but does not necessarily respect the symmetry of V. He is interested in differences between four different energies:
E0(A), El (A), the first two eigenvalues of - !A+ A2V (x), and
Eo (A), El (A), the first two eigenvalues of - a O
) 2 (V (x) + W (x) ).
Of course, because of the tunneling from a to b, W will have an effect, but a less complicated one than one might expect. THEOREM 14 ([47]). Let d = p(a, b),
dl = min{2p(a, supp W), 2p(b, supp W)}, d2 = max{2p(a, supp W), 2p(b, supp W)},
where p(x, S) = inf{p(x, y) : y c S}. Let eo and el be the first two eigenvalues of e-ad 0
( e- ad
0
/
and eo and e"1 the first two eigenvalues of e-adl C
e- Ad
e- ad e-ad2
Then the leading behavior of the differences of any of the E's and E's is given by the leading behavior of the differences of the corresponding e's and Vs in any of the three cases: (i)
d < dl < d2;
(ii)
dl < d < d2;
(iii)
dI < d2 < d.
Thus one can compute lim -A-I log(E1(.X) - Eo(A)) = d = lim
-A-I log(E1(.X)
aToo
AToo
- Eo(.\))
if d < dl, but if dl < d < d2,
lim -) I log(El(A) - Eo(A)) = dl.
XToo
Notice that in cases (i) through (iii) the size of W is not relevant, just the Agmon distance of its support from the zeros of V. This shows why certain cases (iii). of equality have been omitted in (i)
I. W. HERBST
456
Let us give a very brief sketch of the beautiful result lim A-' log(Ei(A) ATOO
Eo(A)) _ -p(a, b),
under the assumptions of Theorem 12. For simplicity we also assume the existence of a Euclidean symmetry R of the potential such that Ra = b. We first need the behavior of SZo (A, x) for large A:
PROPOSITION 15 ([44]). Let S2o(A,x) be the positive normalized ground state of H(A) and d(x) = min{p(x, a), p(x, b)}. Then with the hypotheses above
lim -A I log Qo (A, x) = d(x),
ATOO
uniformly on compact sets. Barry gives two proofs of this result, one relying on large deviations estimates of path integrals and the other based on PDE methods of Agmon. We sketch Barry's PDE proof. First we note that Barry's work in [43] shows that as A T oo, + (b)
SZ0
(9)
where (a is the ground state for -
1
A + 2
1:
aiajV(a)(x - a)i(x - a)j, 2
and similarly, bb.
This is a very rough result but a bit more work gives pointwise estimates in any neighborhood of the form I x - al < CAA, , which show, for example, that with Pa (x) =
A-,./4
SZo (A, A- 2 x +a);
ka (x) = A-v'4S}a (A-2' x + a);
lim O (x)/1a(x) = 1,
(10)
uniformly on compact sets and similarly with a replaced by b. We will use this later. We now sketch a proof of the upper bound half of Proposition 15: First note
that Id(x) - d(y)j
p(x, y) : Ix - yI
J0
1
2V-(1 - t)x + ty) dt,
where the second inequality comes from taking a straight line path from x to y. If follows that Iod(x)I2 < 2V(x), a.e. Given real f : ][8v - ll , let Hf(A) = f(x)H(A)e-f(x)
= H(A) -
2IVf(x)I2+(V.Vf+Vf.V)/2,
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
457
and note that lief (H(A) - Eo(A))V)l = II (Hf(A) - Eo(A))ef*ll
> Re(ef0, (Hf(A) >- (ef0, (A2V
- Eo(A))efV))/ll efVGll
-1
- Eo(A))ef*) /Ilef
Vfl2
If we take f (x) = A(1 - c)2' d(x), the usual Agmon analysis [1] shows efgo E L2 if A is large enough since
A2V-2lofl2-Eo(A)>6>0,
(12)
for large x. If we take zb _ j o with q(x) = 0 if Ix - al < coA-2 or Ix - bI < coA-2, then, in fact, for large A (12) is true on the support of ri and thus (11) implies Ilef (H(A) - Eo(A))nQoll >- 6 Ief,gQoll,
which (with (10) and a similar result with a -+ b) leads to eX[(I-E)d-E]S2o
< const.
(13)
Now 12o is subharmonic outside a neighborhood of {a, b} of the form min{Ix al, Ix - bI} < c1A-2. Thus the L2 estimate (13) along with (10) gives a similar pointwise estimate and the upper bound lim sup A-' log 12o (A, x) < -d(x). Atoo
The lower bound of Proposition 15 relies on the following comparison result based on unpublished work of Agmon: LEMMA 16. Suppose 2 A11 = W12 where W and 12 are non-negative in the cylinder Do = {(xi, x1) : 0 < xl < all + 6), Ix1 < R}. Here x1 E I1 Y
.
Define a > 0 by
2 = sup { R + W (x) : 2
x E Do } ,
where eo is the lowest eigenvalue of -20 in L2 of the ball of radius 1 in Rv-1 with Dirichlet boundary conditions. Then, if xjj < 2R,
12(a, x1) > 0e-,a(1 - e-26aa) min{12(0, y1) : Iy1I < R}, = min{rl(y) : lyl < 2} and q > 0 is the eigenfunction corresponding to eo normalized with IHII,,, = 1. where
The proof of Lemma 16 is based on subharmonic comparison which compares 12 to
4' = '1'rl (x1/R) l e-axl - e-2aa(1+5)eaxil where
ry = min{12(0,x1)
Note that 12 > 0 on 8Do which implies 12 >
Ix 1I < R}.
in Do. We do not give details.
I. W. HERBST
458
The idea behind the lower bound for SZo is to approximate a geodesic from a to x by a piecewise linear path connecting the points xo = a, x1i X2.... , xn = x with segments so small that n
i=I
where vi is sup{V(x) : x c Di} and Di is a cylinder of radius along xi - xi-1 and length (1 + 2S)Ixi - xi_11. If we take
2-'A-2
with axis
zaj(A)2 = and skip a few of the initial cylinders until A2vj > Eo(A), we can apply Lemma 16 to obtain n SZo(.\ x) > e- ra,Ix. -x3 _11on fl(1 e26a.(A)) (14)
-
i=1
where we use (10) to infer that in any region Ix - aI < CA-2, SZo(A,x) > 1 for large A. From (14), we obtain lim i f A-1 log S2o (A, x) > -p(x, a). ATOO
A similar argument with a - b gives the lower bound. We now sketch the proof of the upper bound on E1(A) - Eo(A): Clearly, if (f Q0, 0o) = 0, then E1(A) _ Eo(A) < (.f1o, (H(A) - Eo(A))f1o) 1
(15)
1 f Qo 112
To make sure (f Q0, Qo) = 0, define
f =g-
2
dog dx,
where P(x,a) - P(x,b)
g(x) = h
p(a, b)
)
'
and h is a smooth, odd function with Ih(t) I = 1 outside a small neighborhood of 0. Actually, we should smooth out the argument of g, but for clarity we refrain from doing so in this sketch. Because of our assumption of symmetry, actually f Slog dx = 0. Note that since Q0 concentrates near a and b, 11gS2o II - 1 The numerator of (15) is z f Qo(Vg)2dx, and is supported in a small neighborhood of {x : p(x, a) = p(x, b) } = B. But min{d(x) : x c B} = a p(a, b),
which leads to lim sup A-' (El (A) - Eo (A)) < -P(a, b) ATc
Finally, we sketch the proof of the lower bound on El (A) - Eo(A): Let
Il(A,x)
9a (x) = S20 (A x)
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
459
where S21 is the first excited state of H(A) normalized so that IIQ1II2 = 1 and S21(A, b) > 0. Let -y be a geodesic from a to b and notice that p(a,
Max {Min{p(a, y(t)), p(b, y(t))} : t c [0, 1]} =
b)
2
since, of course, p(a, -y(t)) + p(b, -y(t)) = p(a, b). Find a tubular neighborhood Tb of -y small enough so that for large A and x E T5, 1o (A, x)2 >
e-A(p(a,b)+E).
Again from Barry's work in proving Theorem 11 [43], he obtains (since Sb - (a is orthogonal to Sb + (a) lim
=0
1 - 72= ((b - Sa)
,TOO
(see (9)), and for the same reason that (10) is true,
ga(x) = T1,
(16)
uniformly in a neighborhood of the form Ix - al < A- 2 (respectively, Ix - bI < a- 2 ).
Choose coordinates (t, u), u E Ri-1 in Tb, so that -y(t) has u = 0. Shrink Tb, if necessary, so that
{x: t=O,Iul
Tb={x: 0
El(i) -Eo(n) = z fIVg2dx f IVgal21odx
a/T6
(2 J
\
\
vgdx I/
(17)
a
To estimate fTb I Vga l2 from below in Tb, notice that 9,\(t = 1, u) - 9a (t = O,u)IZ <
f0 l a at gA
2
dt.
Integrating over u and using (16) gives
(v
a 2 dt du, at gA
1)/2 <
changing variables from (t, u) to x introduces only another constant so that for some c" > 0, c"A-(v-1)12 <
IT5 IV ga2dx.
This and (17) give the desired lower bound.
I. W. HERBST
460
References [1] Agmon, S., Lectures on exponential decay of solutions of second-order elliptic equations: Bounds on eigenfunctions of N-body Schrodinger operators, Princeton University Press and University of Tokyo Press, 1982. [2] Avron, J. E., Bender-Wu formulas for the Zeeman effect in hydrogen, Ann. Phys. 131, no. 1 (1981), 73-94.
[3] Avron, J. E., Adams, B. G., Cizek, J., Clay, M., Glasser, M. L., Otto, P., Paldus, J., and Vrscay, E., Bender-Wu formula, the SO(4, 2) dynamical group, and the Zeeman effect in hydrogen, Phys. Rev. Lett. 43 (1979), 691-693. [4] Avron, J. E. and Herbst, I., Spectral and scattering theory of Schrodinger operators related to the Stark effect, Commun. Math. Phys. 52 (1977), 239-254. [5] Avron, J., Herbst, I., and Simon, B., Schrodinger operators with magnetic fields, I. General interactions, Duke Math. J. 45 (1978), 847-883. [6] Avron, J., Herbst, I., and Simon, B., Schrodinger operators with magnetic fields, II. Separation of center of mass in homogeneous magnetic fields, Ann. Phys. 114 (1978), 431-451. [7] Avron, J., Herbst, I., and Simon, B., Schrodinger operators with magnetic fields, III. Atoms in homogeneous magnetic field, Commun. Math. Phys. 79 (1981), 529-572. [8] Avron, J., Herbst, I., and Simon, B., Schrodinger operators with magnetic fields, IV. Strongly bound states of hydrogen in intense magnetic field, Phys. Rev. A20 (1979), 2287-2296. [9] Avron, J., Herbst, I., and Simon, B., The Zeeman effect revisited, Phys. Lett. 62A (1977), 214-216.
[10] Avron, J., Herbst, I., and Simon, B., Formation of negative ions in magnetic fields, Phys. Rev. Lett. 39 (1977), 1068-1070.
[11] Avron, J. and Simon, B., A counterexample to the paramagnetic conjecture, Phys. Lett. A 79 (1979/80), no. 1-2, 41-42. [12] Baumgartner, B., and Seiringer, R., On the ordering of energy levels in homogeneous magnetic fields, Lett. Math. Phys. 54, no. 3 (2000), 213-226. [13] Bentosela, F., Carmona, R., Duclos, P., Simon, B., Souillard, B., and Weder, R., Schrodinger
operators with electric field and random or deterministic potential, Commun. Math, Phys. 88 (1983), 387-397.
[14] Blankenbecker, R., Goldberger, M. L., and Simon, B., The bound states of weakly coupled long-range one dimensional quantum Hamiltonians, Ann. Phys. 108 (1977), 69-78. [15] Carmona, R. and Simon, B., Pointwise bounds on eigenfunctions and wave packets in N-body quantum systems, V. Lower bounds and path integrals, Commun. Math. Phys. 80 (1981), 59-98.
[16] Christ, M. and Kiselev, A., Absolutely continuous spectrum of Stark operators, Ark. Mat. 41 (2003), 1-33. [17] Corngold, M., Harrell, E., and Simon, B., The mathematical theory of resonances whose widths are exponentially small, II, J. Math. Appl. 99 (1984), 447-457. [18] Delyon, F., Simon, B., and Souillard, B., From power localized to extended states in a class of one dimensional disordered systems, Phys. Rev. Lett. 52 (1984), 2187-2189. [19] Delyon, F., Simon, B., and Souillard, B., From power pure point to continuous spectrum in disordered systems, Ann. Inst. H. Poincare 42 (1985), 285-309. [20] Eastham, M. S. P. and Kalf, H., Schrodinger Operators with Continuous Spectrum, Pitman, Boston, 1982. See Theorem 6.3.1. [21] Erdos, L. and Solovej, J. P., The kernel of Dirac operators on S3 and R3, Rev. Math. Phys. 13, no. 10 (2001), 1247-1280. [22] Fortuin, C., Kasteleyn, P., and Ginibre, J., Correlation inequalities on some partially ordered sets, Commun. Math. Phys. 22 (1971), 89-103.
[23] Gearhart, L., Spectral theory for contraction semigroups on Hilbert space, Trans. Amer. Math. Soc. 236 (1978), 385-394.
[24] Graffi, S. and Grecchi, V., Resonances in the Stark effect and perturbation theory, Commun. Math. Phys. 62 (1978), 83-96. [25] Graffi, S. and Grecchi, V., Resonances in the Stark effect of atomic systems, Com-
mun. Math. Phys. 79 (1981), 91-109. [26] Grosse, H. and Stubbe, J., Splitting of Landau levels in the presence of external potentials, Lett. Math. Phys. 34, no. 1 (1995), 59-68.
ELECTRIC AND MAGNETIC FIELDS AND THE SEMI-CLASSICAL LIMIT
461
[27] Hardy, G. H., Divergent Series, Oxford Univ. Press, Oxford, 1949, page 191. [28] Harrell, E. and Simon, B., The mathematical theory of resonances whose widths are exponentially small, Duke Math. J. 47 (1980), 845-902. [29] Helffer, B. and Mohamed, A., Caracterisation du spectre essentiel de l'operator de Schrodinger avec un champ magnetique, Ann. Inst. Fourier (Grenoble) 38 (1988), 95-112.
[30] Herbst, I., Dilation analyticity in constant electric field I: The two-body problem, Commun. Math. Phys. 64 (1979), 279-298. [31] Herbst, I., Schrodinger operators with external homogeneous electric and magnetic fields, in Rigorous Atomic and Molecular Physics (G. Velo and A. Wightman, eds.), pp. 131-183, Plenum Press, NY, 1981. [32] Herbst, I. and Pitt, L., Diffusion equation techniques in stochastic monotonicity and positive correlations, Prob. Theory Related Fields 87, no. 3 (1991), 275-312. [33] Herbst, I. and Simon, B., Dilation analyticity in constant electric field, II. The N-body problem, Borel summability, Commun. Math. Phys. 80 (1981), 181-216. [34] Herbst, I. and Simon, B., The Stark effect revisited, Phys. Rev. Lett 41 (1978), 67-69. [35] Hogreve, H., Schrader, R., and Seiler, R., A conjecture on the spinor functional determinant, Nuclear Phys. B 142, no. 4 (1978), 525-534. [36] Hundertmark, D. and Simon, B., A diamagnetic inequality for semigroup differences, J. Reine Angew. Math. 571 (2004), 107-130. [37] Loss, M. and Yau, H.-T., Stability of Coulomb systems with magnetic fields, III. Zero energy bound states of the Pauli operator, Commun. Math. Phys. 104 (1986), 283-290. [38] Miller, K. and Simon, B., Quantum magnetic Hamiltonians with remarkable spectral properties, Phys. Rev. Lett. 44 (1986), 1706-1707.
[39] Nevanlinna, F., Zur Theorie der asymptotischen Potenzreihen. (Finnish) Ann. Acad. Sc. Fennicae (A) 12, Nr. 3, 81 S. (1916). [40] Perelman, G., On the absolutely continuous spectrum of Stark operators, Commun. Math. Phys. 234 (2003), 359-381. [41] Reed, M. and Simon, B., Methods of Modern Mathematical Physics I: Functional Analysis, Revised and Enlarged Edition, Academic Press, San Diego, 1980, page 275. [42] Simon, B., The bound state of weakly coupled Schrodinger operators in one and two dimensions, Ann. Phys. 97 (1976), 279-288. [43] Simon, B., Semiclassical analysis of low lying eigenvalues, I. Non-degenerate minima: Asymptotic expansions, Ann. Inst. H. Poincare 38 (1983), 295-307. [44] Simon, B., Semiclassical analysis of low lying eigenvalues, II. Tunneling, Ann. of Math. 120 (1984), 89-118.
[45] Simon, B., Instantons, double wells, and large deviations, Bull. Amer. Math. Soc. 8 (1983), 323-326. [46] Simon, B., Semiclassical analysis of low lying eigenvalues, III. Width of the ground state band in strongly coupled solids, J. Funct. Anal. 63 (1985), 123-136.
[47] Simon, B., Semiclassical analysis of low lying eigenvalues, IV. The flea on the elephant, J. Funct. Anal. 63 (1985), 123-136. [48] Simon, B., An abstract Kato's inequality for generators of positivity improving semigroups, Indiana Univ. Math. J. 26, no. 6 (1977), 1067-1073. [49] Simon, B., Kato's inequality and the comparison of semigroups, J. Funct. Anal. 32, no. 1 (1979), 97-101.
[50] Simon, B., Maximal and minimal Schrodinger forms, J. Operator Theory 1, no. 1 (1979), 37-47.
[51] Sokal, A., An improvement of Watson's theorem on Borel summability, J. Math. Phys. 21 (1980), 261-263.
[52] Walter, J., Absolute continuity of the essential spectrum of -d2/dx2+q(x) without monotony of q, Math. Z. 129 (1972), 83-94.
DEPARTMENT OF MATHEMATICS, UNIVERSITY OF VIRGINIA, CHARLOTTESVILLE, VA 22903,
U.S.A. E-mail address: iwhmvirginia. edu
Proceedings of Symposia in Pure Mathematics Volume 76.1, 2007
Some Bound State Problems in Quantum Mechanics Dirk Hundertmark Dedicated to Barry Simon on the occasion of his 60th birthday ABSTRACT. We give a review of semi-classical estimates for bound states and their eigenvalues for Schrodinger operators. Motivated by the classical results, we discuss their recent improvements for single particle Schrodinger operators as well as some applications of these semi-classical bounds to multi-particle systems, in particular, large atoms and the stability of matter.
CONTENTS
Semi-Classical Bounds for Single-Particle Schrodinger Operators Multi-Particle Coulomb Schrodinger Operators More on Bound States for Atoms References 1.
2. 3.
In this survey, we focus on results for bound states of Schrodinger operators related to the semi-classical limit and Coulomb potentials. We will not discuss a large part of the existing literature on the general theory of bound states for Schrodinger operators. With no attempt at completeness, we nevertheless would like to mention at least some part of this literature: For one-particle Schrodinger operators, see, for example, [19, 151]. Two-body cluster results are discussed in [4, 84, 145, 153, 154, 175]; finiteness results of the discrete spectrum for Nparticle systems can be found in [1, 41, 166]; and for results on the Efimov effect, see, for example, [3, 37, 125, 157, 162, 172].
1. Semi-Classical Bounds for Single-Particle Schrodinger Operators The origin of semi-classical estimates can be traced back to the dawn of quantum mechanics in the beginning of the last century. Around 1912, Hermann Weyl 2000 Mathematics Subject Classification. 35J10, 81Q10. Key words and phrases. bound states, semi-classical inequalities, stability of matter, Coulomb Hamiltonian. Supported in part by the National Science Foundation grant DMS-0400940. ©2007 by the author. Faithful reproduction of this article, in its entirety, by any means is permitted for non-commercial purposes. 463
464
D. HUNDERTMARK
published a series of papers [168, 169, 170] (see also [171]) on the frequencies of an oscillating membrane and the radiation inside a cavity, verifying a conjecture of Jeans and Lorenz on the connection between the asymptotic behavior of these frequencies and the volume of the cavity. While in itself being a classical problem, this work was the starting point of a substantial branch of analysis and mathematical physics, especially in quantum mechanics. Consider a bounded domain A C Rd and the eigenvalue problem for the Dirichlet Laplacian -DA cp = Ecp,
that is, the partial differential equation d
a2
j
andcplOA=0.
(1)
Furthermore, let E1 (A) < E2 (A) _< E3 (A) < ... be an ordering of the eigenvalues of (1) and define the counting function
ND(E):=
1
E3 (A)<E
which counts the number of eigenvalues of -AD below E. Weyl showed that
ND(E) =
CAI Ed/2 + o(Ed/2)
as E
oc.
(2)
d/2
Here wd = r(1+(d/2)) is the volume of the unit ball in Rd. Weyl's formula holds for all bounded domains A C Rd.
The origin of Weyl's equality is easy to see: If A is a centered cube of side length a, then the eigenfunctions and eigenvalues of the Dirichlet Laplacian are known explicitly and given by d
u(x) = 11 sin("Q"
n , E N` and
=1
E=
2
a2
2
In2
d
nv,
a2 =1
that is, ND (E) is precisely the number of points n E N d within the ball of radius E'/2, which behaves asymptotically as wd (27r)d adEd/2 + o(Ed/2) for E ---> oc.
(3)
Similarly, the counting function NN (E) of the Neumann Laplacian has the same asymptotics as the Dirichlet Laplacian. They differ by a surface term which is of lower order in the high energy asymptotics, at least for domains with a nice boundary A. Weyl's crucial idea was to approximate a general domain A C Rd by cubes and to use (3). Using variational arguments, he showed wd(a )d#{disjoint
lim E-d/2ND(E) = lim cubes of side length a in A}. a-.0 E-oo 27r Since #{disjoint cubes of side length a in Al = ad (JAI + o(1)), one obtains (2).
BOUND STATES IN QUANTUM MECHANICS
465
1.1. P61ya's Conjecture. In 1961, Polya [127] conjectured that the asymptotic result (2) holds as a uniform bound on ND(E) for all E > 0 with some constant P(d), that is, ND (E) < P(d) JAI Ed/2 for all E > 0,
(4)
or, equivalently,' En > (P(d) I AI)-2/d n2/d for all n E N.
Polya also conjectured that the sharp constant in (4) is given by Weyl's asymptotic result, P(d) = (2 dr)d . He was able to prove this conjecture for the special class of tiling domains A, that is, disjoint congruents of A are assumed to cover 1(8d. The argument is rather simple. Scaling AT := rA, one gets another tiling domain with ND(E) = ND (r-2E) (by scaling of the kinetic energy). So with B = ball of unit volume in 1[8d, we get D( ND (r 2E) < NB (r
2
E) T
where AT = #{disjoint congruents of rA which are subsets of B}. Fix E > 0 and let r -* 0. By Weyl's asymptotic we know
NB (r 2E) = r -d
((27rWd)d
Ed/2
and, obviously, A, = r-d(IAI-1 + 0(1)) as r ---> 0. Together these estimates give the bound D( r-2
ND (E) < lim NB
AT
CJd
E)
T2--j-)d Ink
Ed/2
for a tiling domain A. Unfortunately, the sharp result is not known for general bounded domains. The best result is due to Li and Yau [93], who showed
()d/2
N (E)
(27r)d
IAI Ed/2
(5)
More precisely, they proved the sharp bound n
> j=1
d
d+2
2/d 10d
(27r)JAI
n'+2 forallnENd
and deduced (5) from this simply by observing En > n E; I E. Laptev [88] gave a much simpler argument than the original one by Li and Yau in 1996. Moreover, he showed that if Polya's conjecture holds for a domain A, C 1[8d1, then it holds for all domains A = Al x A2 for all d2 E N and domains A2 C 1[842. This paper was the first instance where the idea of "stripping off' dimensions appeared, which later turned out to be the key for a refined study of semi-classical inequalities for moments of eigenvalues of Schrodinger operators, see Section 1.7.
'Indeed, if N(E) < CE', then putting E = E one sees n = N(En) < CE,', that is, E, > C-l/-nl/a. Conversely, N(E) = FEj<E 1 1 < CEa.
D. HUNDERTMARK
466
1.2. Weyl Asymptotics for Schrodinger Operators. In the early 1970's, Birman-Borzov, Martin, and Tamura [17, 122, 161] proved semi-classical asymptotics for the number of the negative eigenvalues2 of Schrodinger operators with a
Holder continuous and compactly supported potential V. Let El < E2 < E3 < < 0 be a counting of the negative eigenvalues of -A + V on L2 (R) and set N(V) := #{negative eigenvalues of - A + V}. Assume that V is non-positive for the moment and introduce a coupling constant A. We want to study the large A asymptotics of N(AV). Assuming the DirichletNeumann bracketing technique developed by Courant and Hilbert, a short argument
can be given as follows; for details, see [128]. We let Aa = {-a < xj < 2, j = 1, ... , d} and define Aa,j = j + A. for j E aZd. Assume that V is constant on the cubes Aa,j, that is, V(x) = Va,j for x E Aa,j. Using Dirichlet-Neumann bracketing, one gets
Nn ,, (AVa,j).
I NA , (AVaj) < N(AV) < jEaZd
jEaZd
Applying Weyl's asymptotic result (2) with E
N(AV) =
1:
Wd
(21r)d Adl2ad
IVa,j
_
AIVa,j
,
I d/2 + o(Ad/2) as A - oo.
jEaZd
So if V is continuous, non-positive, and of compact support, N(AV) = (21d
Ad/2
fd
V(x)dl2 dx + o(Ad/2)
(6)
This formula is, in fact, a semi-classical asymptotic. Let be the so-called classical symbol associated with the operator -A + AV. Using the Fubini-Tonelli
theorem and scaling, the volume of the negative energy region in phase space is given by
ffI2+av(x)
wdAd/2
f V(x)d]2 dx. d
Thus, formula (6) says that the number of negative eigenvalues of -A+AV asymptotically, for large A, behaves like the classical allowed phase-space volume divided by (27r)d,
lira A-d/2N(AV) = (21 ^-00
d
ff(2 +V(x))° d dx,
where we set ro = 1 for r < 0 and 0 for r _> 0. So each eigenfunction corresponding to a negative eigenvalue occupies a volume (2rr)d in phase space. This is in perfect agreement with the Heisenberg uncertainty principle, according to which an electron occupies a volume of at least (27r)d in phase space; see also [45]. One should also note that Weyl's original result fits very well into this more general framework. Indeed, and again somewhat formally, for a bounded open set
A, one can recover the Weyl asymptotic by setting V = 001Ac - 1A (with the 2For a probabilistic proof, see [78] or [152].
BOUND STATES IN QUANTUM MECHANICS
467
convention oo 0 = 0). Then (6) gives
NA (E) = N(EV) 1
(27r)d fL2+EV(x)<0 Wd
(27r)
d JAI
E
d dx + o(Ed/2)
+ o(Ed/2)
which is Weyl's asymptotic.
1.3. The Birman-Schwinger Principle. Around 1961, Birman [16] and Schwinger [139] independently found a way to reduce an estimate on the number of negative bound states to the study of a bounded integral operator. The idea of their argument is as follows: First assume that V = -U is non-positive. Let cp be
an eigenvector of -0 - U for the negative energy E. That is, (-0 - U)cp = Ep, with E < 0. Rearranging gives (-A - E)cp = Ucy, which in turn is equivalent to cp = (-A-E)-1Ucp, since for E < 0, (-O-E) is boundedly invertible. Multiplying this with VU and setting V) = v"U-W, we see -0 - U has a negative eigenvalue E if and only if the Birman-Schwinger operator
)-I U KE _ /U(-A - E)-'V-U-
(7)
has eigenvalue 1. Moreover, since the map E -> nth eigenvalue of KE is (strictly) monotone increasing, a careful analysis shows that the corresponding eigenspaces
have the same dimension. In short, denoting Al(KE) > A2(KE) > ... > 0 the <_ 0 eigenvalues of the Birman-Schwinger operator KE and by El < E2 < E3 < the negative eigenvalues of -A - U, the Birman-Schwinger principle is the statement that 1 = An (KE.) for all n. (8)
If V is not non-positive, one uses the min-max principle to see that NE(V) < NE(-V_). Here NE(V) = #{eigenvalues of - 0 + V < E}. Thus, setting U = -V-, using the Birman-Schwinger principle, and the monotonicity of A (KE), one sees
NE(V) < #{eigenvalues of KE > 1}. In particular, NE(V) < tr[KE] for all n E N. On 1883, KE is a Hilbert-Schmidt operator with integral kernel e-./=EIx yl KE(x,y) = V_(x) V/ _(y)
-yJ
implying the bound
V-(x)V (y) x - yl As E -* 0, we get the Birman-Schwinger [16, 139] bound NE(V) < tr[KE] =
1
(47r)2
N(V) = No(V) C
ff
(y) (47r)2
X- yl 2 ff V (x)
dxdy.
dxdy.
(9)
Note that (the negative part of) a potential V_ is in the Rollnik class [148] if and only if the right-hand side of (9) is finite. On the other hand, this simple bound does not have the right large coupling behavior since it only shows that N(AV) < (const))2 which grows much faster than
D. HUNDERTMARK
468
the asymptotic growth A3/2 shown by large coupling asymptotic. Nevertheless, the Birman-Schwinger principle is at the heart of all proofs of semi-classical eigenvalue bounds.
1.4. The CLR and Lieb-Thirring Bounds. The semi-classical Weyl-type asymptotics for the number of bounds states leads naturally to the question whether there is a robust bound of the form
f N(V) < (2ir)d Co'd JJ 1
1 d dx 12+v
10 )
=Lpdf V-(x)d/2dx d
for arbitrary potentials V for which the right-hand side is finite.3 Here Lo,d = Co,dwd/(27r)d, where wd is the volume of ball of radius one in d dimensions.
More generally, one can ask a similar question for higher moments4 of the negative eigenvalues. That is, whether there is a semi-classical bound for
S"(V) = tr[(-A + V)'-] of the form Sry(V) <
C1, d (21r)d
f
+V(x))' dd x.
(11) (12)
Or, equivalently, whether S-'(V) < Lry d with
L ry,d = C.y,d L i d, w here L..,d =
V(x)ry+dl2 dx
J 2 - 1)y d = (a i )d f (
(13) 2d7rdi 2r(ry+1 +d/2)
is
the so-called classical Lieb-Thirring constant. In particular, Lc d = wd/(21r)d and
Lild = d+a
Id)d
Of course, one has to assume that the negative part of the
potential V_ E L'Y+d/2 (1Rd) One recovers N(V) from Sry (V) by N(V) = S° (V) _ lim.y_o S7(V). Of course, the physically most interesting cases are y = 0, the counting function for the number of bound states, and y = 1, which gives a bound for the total energy of a system of non-interacting fermions in an external potential given by V. Also a simple approximation argument (see, e.g., [149]) shows that a bound of the form
(10) allows one to extend the semi-classical asymptotic (6) to all potentials in Ld/2 (Rd).
It is easy to see that bounds of the type (12) can only hold for y > 0 in d < 2 and y > 0 in d > 3. In fact, since any non-trivial attractive potential has at least one bound state [150, 83] and [87, pp. 156-157], there can be no semi-classical bound of the form (13) for y = 0 in d = 2; and in one dimension, this bound is even impossible5 for 0 < y < z 3More precisely, V+ should be locally integrable and V_ E Ld/2(Rd) 4Often called Riesz' moments.
5Let c > 0, b be the Dirac measure at 0, and note that, by a one-dimensional Sobolev embedding, the operator -82 - c6 is well-defined as a sum of quadratic forms. Take a sequence of approximate delta-functions 6 converging weakly to 6. Then -82 - c6 converges to -82 - c6 in strong resolvent sense. If -y < Z, the right-hand side of (13) goes to zero, but the ground state
of -82 - c6 stays bounded away from zero (it converges, in fact, to -c2/4 = single negative eigenvalue of -82 - c6).
BOUND STATES IN QUANTUM MECHANICS
469
The inequalities (12) and (13) were proven by Lieb and Thirring [117, 118] in 1975-76 in the cases ry > 2 in d = 1 and y > 0 for d > 2. The proof of (10), the famous Cwikel-Lieb-Rosenblum bound, is considerably more complicated than the proof for ry > 0. It also has an interesting history: Rozenblum announced his proof of (10), which is based on an extension of a machinery developed by Birman and Solomyak [18], in 1972 in [130]. This announcement went unnoticed in the West. Independently of Rozenblum, Simon established
in [149] a link between the bound (10) and the then conjectured fact that the Birman-Schwinger operator K° given in (7) for E = 0 is a certain weak trace ideal6
for d > 3. This conjecture by Simon on the asymptotic behavior of the singular values of Ko motivated Cwikel and Lieb for their proofs of the CLR bound. In [31], Cwikel proved Simon's conjecture and Lieb, [94], used semigroup methods to bound tr[F(Ko)] for suitable functions F. Rozenblum's proof appeared in 1976 in [131], Lieb's was announced in 1976 in [94], and Cwikel's proof was published in 1977. Of the three methods, Lieb's gives by far the best estimates for the constants Co,d. A very nice and readable discussion of Lieb's method can be found in Chapter 9 of [152] and Chapter 3.4 of Ropstorff's book [129]. In particular, Ropstorff discusses the fact that an extension of Lieb's method to higher moments y > 0 gives the upper bound Cy,d < 27(y+d/2)(1+O((y+d/2)-1)) as ry+d/2 -> oc. Later proofs of the CLR bound were given by Li and Yau [93] and Conlon [27]; see also [92, 1321. The Lieb-Thirring inequalities (12) fit beautifully into the large coupling asymptotics. At least on a formal level, it is easy to lift the asymptotics for N(AV) _ S°(AV) to moments y > 0 by the following observation: For any -Y > 0, )ry = y fo,3 t7-1 dt
(S
=y
100
Jo
(s + t)°
dt
for all real s (here s° = 1 if s < 0 and zero if s > 0). Freely interchanging integrals and traces gives f S'Y (AV) = tr(-A + AV)" = y tr(-A + AV + t)°
J
=ryL°°(
0
(27r)d
1
0 0,0
ff(2 + V(x) + t)° d dx +o(Ad/2) C-ldt
ff(2 + AV (x))ry
Thus urn
A-(7+d/2)S-y(AV)
_
(2x)d
o(A'r+d/2).
ff(2 + V(x))I (14)
Ly,d f V
(x)y-+dl2
dx.
Of course, to make this sketch rigorous, one needs to handle the error terms more carefully, which we skip. This large coupling asymptotics shows that the best possible constants L y,d in the Lieb-Thirring inequality have the natural lower bound L,y,d > L' Id, or, equivalently, C.y,d > 1. 6A compact operator is in a trace ideal SP if its singular values are in the space 1P(N) and it is in the weak trace ideal Sw if its singular values are in the weak-lP space l,P (N); see, for example, [156].
D. HUNDERTMARK
470
1.5. A Sobolev Inequality for Fermions. Besides being mathematically very appealing, the y = 1 version of the Lieb-Thirring bound gives a Sobolev inequality for fermions whose d = 3 version has a nice application to the stabilityof-matter problem. For notational simplicity, we will not take the spin of the particles into account. The following gives a duality between a Lieb-Thirring type bound and a lower bound for the kinetic energy of an N-particle fermion system. Thus the kinetic energy inequality for fermions is an immediate corollary of the Lieb-Thirring bound for y = 1. THEOREM 1. The following two bounds are equivalent for non-negative convex functions G and F : 1E8+ -> R+ which are Legendre transforms of each other: The Lieb-Thirring bound,
EjJ = trL2(Ed)(Ho +V)_ < f G((V(x))_) dx,
(15)
(usually with Ho = -0, but this does not matter in the following) and the ThomasFermi bound, Ho
\
? fF(p(x))dx,
L2(Rd) / for all antisymmetric states 0C AN,L12(Rd) with norm one. Here
(16)
n=1
P,G(x):=N f
aN-1d
IZ/J(x,x2,...,xN)I2dx2...dxN
is the so-called one particle density associated with the antisymmetric N-particle
state.
For an explicit relation between F and G, see (18) and (19). Using the traditional Lieb-Thirring bound with y = 1, one immediately gets the following COROLLARY 2. For normalized
E
AN L2 (R d),
N
, - E Ojv j=1
L2(1RNd)
Kd F
with TF
Kd
d
d+
2(d+2 2
cl
Ld)
f P, (x) d2dx d+
C1,d2/d
>
2/d
-
ad
d
47r2
d+ 2 U)2/d'
One should note that the right-hand side of this bound is exactly the ThomasFermi prediction for the kinetic energy of N fermions and Kd F is the Thomas-Fermi
constant; see section 2.2. In particular, if C1,d is equal to one, then the ThomasFermi ansatz for the kinetic energy, a priori only supposed to be asymptotically correct for large N, should be a true lower bound for all N. This is a situation very much similar in spirit to Polya's conjecture. REMARK 3. Taking the spin of electrons into account, that is, assuming that b E A N (L2 (Rd, (Cq)) is normalized (and q = 2 for real electrons), one has the lower bound
N
E-Ojo) j=1
> (qCl
d)-2/dKdTF f
(x)(d+2)/ddx.
(17)
BOUND STATES IN QUANTUM MECHANICS
471
PROOF OF THEOREM 1: This proof is certainly known to the specialist, but we include it for completeness. In fact, the reverse implication is the easy one, (15) = (16): Fix N E N and let El < E2 < ... < EN < 0 be the first N negative eigenvalues of the one-particle Schrodinger operator H = Ho+V. Usually,
Ho = -0, but this does not really matter. By the min-max principle, we can assume without loss of generality that V is non-positive, V = -V_ = -U. If H has only J < N negative eigenvalues, then we put Ej = 0 for j = J + 1, . , N. Consider HN = EN1(Hoj - U(xj)) on AN L2(W1) be the sum of N independent copies of H. More precisely, one should write HN = N1 Hj with . .
®H®10 ... 01.
H3 =1
N-j-1 times
j times
Let cpl, ... , cON be the normalized eigenvectors corresponding to the eigenvalues
Ej (if J < N, pick any orthonormal functions for j > J) and put N
cpl A ... AWN GA L2 (R d),
the normalized antisymmetric tensor product of the cpj's, that is, a Slater determinant. Then N
N
J:IEnI n=1
N
-EEn -V),EHnV n=1 n=1 (0, N
N
\\
//,,//''
n=1
\
n=1
Since En 1 Un is a sum of one-body (multiplication) operators, we have (V), E'1 UnV)) f U(x)p,0(x) dx, by the definition of the one-particle density. Thus, taking (16) into account, one gets N
dx -
IEnj < f n=1
r
J
F(pV, (x)) dx
=
f(u(x)p(x) dx - F(p.,(x))) dx
<
f sup(U(x)t - F(t)) dx t>o
G(U(x)) dx,
where we were forced to put
G(s) := sup(st - F(t))
(18)
t>o
since we only know that pp (X) > 0. (16): This is certainly standard-the argument in the original case goes (15) through nearly without change. By min-max and the Lieb-Thirring inequality (15), we know that for any non-negative function U and any normalized antisymmetric N-particle V),
\
n=1
(Ho - U)z/i) > - tr(Ho - U) > /
f G(U(x)) dx.
D. HUNDERTMARK
472
Thus N n=1
HoV)) > K
_
UnV))
- fc U(x)) dx (
n=1 f
U(x)p,,(x) dx -
J
G(U(x)) dx
= J [U(x)p, (x) - G(U(x))] dx again by the definition of the one-particle density. Hence N
n=1
Ho>
>
U>p
[U(x)P,G (x) - G(U(x))] dx
J
G(U(x))] dx
sup U(x)>o
= f sup [sp(x) - G(s)] dx = fF(P(x)) dx, s>O
where, of course, we put
F(t) := sup(st - G(s)). s>o
REMARK 4. Since F and G are Legendre transforms of each other and since the double Legendre transform of a convex function reproduces the function (under suitable semi-continuity and convexity assumptions), we see that the Lieb-Thirringtype inequality (15) and the Thomas-Fermi-type kinetic energy bound (16) are dual to each other. In particular, one implies the other with the corresponding optimal
constants. This could be interesting in the hunt for sharp constants, since Eden and Foias gave in [36] a direct and rather simple proof of the kinetic energy lower bound in one dimension which, via the duality result of Theorem 1, still gives the best estimate for C1,1.
Following Lieb and Thirring, the bound in Theorem 1 has a beautiful application to the stability-of-matter problem which we will discuss a little bit in Section 2.3.
The Lieb-Thirring inequalities also found other applications, especially in the theory of non-linear evolution equations, as a tool to bound the dimension of attractors [29, 55, 59, 100, 133, 164].
1.6. Classical Results for the Lieb-Thirring Constants. The moment inequalities due to Lieb and Thirring are an important tool in the theory of Schrodinger operators since they connect a purely quantum mechanical quantity with its classical counterpart. Moreover, as we already saw, a dual version of it, the Sobolev inequality for fermions, is related to the theory of bulk matter. So a good understanding of the Lieb-Thirring coefficients is of some importance for our understanding of quantum mechanics. In general dimensions d c N, one now knows the following properties of C,y,d: Cy,d > 1, which follows from the Weyl-asymptotics. Monotonicity in -y: C.y,d < C7,,,d for all ry > yo (Aizenman and Lieb [2]).
BOUND STATES IN QUANTUM MECHANICS
473
Cy,d > 1 as soon as 'y < 1 (Helffer and Robert [66]7). The best bounds on C,y,d are due to Lieb [100], but they are explicitly dimension dependent and grow like Cy,d = 27ra as a = y + d/2 -> oo; see [129, Chapter 3.4]. Some special bounds in small dimensions: The bounds CI,1 < 27r
CI,2 < 6.03388 C1,3 < 5.96677
are due to Lieb [100] and, after twelve years of additional work, were slightly improved by Blanchard and Stubbe [20] in 1996 C1,1 < 5.81029
C,2 < 5.17690 C1,3 < 5.21809.
Unnoticed by the Schrodinger operator community, Eden and Foias [36] gave a simple and direct proof for the Thomas-Fermi-type kinetic energy inequality in one dimension. Using the duality between the ThomasFermi-type bound and the Lieb-Thirring inequality, their approach shows
that C1,1 < 7r/./ < 1.82, which is still the best available bound in this case.
There is a natural lower bound on Co,d using the fact that the CLR bound implies a Sobolev inequality8 [58], [118, eq. (4.24)], see also the discussion in [152, pp. 96-97], In three dimensions, it gives 4.6189 < Co,3 < 6.869
where the upper bound is from Lieb [94]. In particular, this shows that in dimension 3, Lieb's result is at most 49% off the best possible. In fact, the above lower bound is conjectured to be the correct value [58, 118, 152]. The monotonicity in -y is probably easiest to understand in the phase-space picture (12) of the Lieb-Thirring bounds: Let s- = (s)_ = (JsJ - s) be the negative part a of s. For any 0 < yo < y, there exists a positive (!) measure y on ll + such that9 (s)ry =
f(s + t) o
(dt).
71n particular, this disproved part of a conjecture of Lieb and Thirring made in [118]. 8More surprisingly, the Sobolev inequality together with the fact that -0 generates a Markov semigroup implies the CLR bound, see [92]. 91n fact, p.(dt) = ctry-YO-1 dt for an explicit constant c; just do the integral on the right-hand side by scaling.
D. HUNDERTMARK
474
With this we have
tr(-A + V)-' = <
j(tr(_+ V + t)°
2d(ISI2+V(x)+t)-'°d dxp(dt)
-,.,dd f f fa dd
(2 (21r)d
(dt)
dx
ff fLd(
12+V(x))ry d dx,
by freely interchanging the integrations and the trace. In particular, this shows C.y d <
Cy0,d.
Thus monotonicity in y is just a simple consequence of the Fubini-Tonelli theorem and scaling. In one dimension, much more is known: C3/2,1 = 1 (Gardner, Greene, Kruskal, and Miura [57]; Lieb and Thirring [118]). By monotonicity, this implies, C.y,1 = 1 for y > 2.
For 1 < y
2, an explicit solution10 of the variational problem
supV# ° }v7+vz El (V) I leads to C'7,1 > 2
(y- 1/2ry-1/2 y + 1/2)
(Keller [79], later rediscovered by Lieb and Thirring [118]). In particular, C.y,1 > 1 for y < 2 and C1,1 > 2/V' > 1.154, which should be compared to the Eden-Foias upper bound on C1,1. C1/2,1 < 00 (Weidl [167]). The sharp result C3/2,1 = 1 follows from the lower bound Cy,d > 1 and a sum rule for one-dimensional Schrodinger operators from the theory of the KdV equation. It reads 3 16
foo V(x)2 dx =
jEj I3/2
+ "scattering data".
oo
As noted on page 115 of [57], the contribution from the scattering data is nonnegative, so one can drop it to get an inequality for the moment of the negative 2 eigenvalues.11 It remains to note that L ,2,1 = 1s Lieb and Thirring [118] did not settle the critical case ry = 2 in one dimension. The question whether C112,1 is finite or not was open for twenty years until Weidl [167] showed that C1/2,1 < 4.02. But, despite some considerable interest, and in contrast to other results on sharp inequalities (see, e.g., [24, 25, 99] on Sobolev 10See also the very nice discussion in [15].
"It might be amusing to note that dropping the contribution of the eigenvalues gives an upper bound for the contribution of the scattering data which was the key to proof of a conjecture of Kiselev-Last-Simon [82] on the ac spectrum of one-dimensional Schrodinger operators with L2 potentials by Deift and Killip [33]. Sum rules have also turned out to be instrumental in the study of other related spectral problems; see [80, 81, 85, 124, 136, 137].
BOUND STATES IN QUANTUM MECHANICS
475
inequalities), the only sharp bound for the Lieb-Thirring inequalities for more than twenty years was the original result by Lieb and Thirring. This was especially tantalizing since, depending on the dimension, there are obvious conjectures for the sharp Lieb-Thirring constants,
CONJECTURE 5. In dimension d > 3, C1,d = 1. In particular, the ThomasFermi-type bound for the kinetic energy of N fermions should hold with the Thomas-
Fermi constant. In one dimension, 'y-y
for2<7<2
Cry,l=2(y
\y+
2
(with C1/2,1 = lim,y_..,l/z C.y,1 = 2) and the extremizers in the one-bound-state variational problem studied by Keller [79] and by Lieb and Thirring [118] should also be
extremizers in the Lieb-Thirring inequality. That is, up to scaling and translations, the extremizing potentials V in (13) for d = 1 and 2 < y < a are of the form
X1/4 ))-2.
V(x) _ -yz 11/4 (cosh)
Yz
In particular, for -y = 2, CI/z,I = 2 and the extremizing potential should be a multiple and translate of a delta function. For the first moment one should have C1,1 = 2/v < 1.155. Shortly after Weidl's result, Hundertmark, Lieb, and Thomas proved the y = 2 part of the one-dimensional Lieb-Thirring conjecture in [70]. THEOREM 6 (Hundertmark-Lieb-Thomas [70]). Suppose v is a non-negative measure with v(ll) < oo and let El < Ez < E3 < . < 0 be the negative eigenvalues counting multiplicity of the Schrodinger operator -ate - v (if any) given by the corresponding quadratic form. Then 00
< 2 v(l ) %=1
with equality if and only if the measure v is a single Dirac measure. Since Li,2,1 = 4, this shows C112,1 = 2, confirming the left endpoint of the one-dimensional Lieb-Thirring Conjecture 5. It might be interesting to note that Schmincke, [138], proved a corresponding sharp lower bound for one-dimensional Schrodinger operators with a potential. That is, for -8,2,, + V, he showed 00
-4J V(x)dx<E. 1
i=1
Schmincke's proof uses Schrodinger's factorization method (see also [30, 32]) and has been extended in [126] to some higher moments.
1.7. The Laptev-Weidl Extension of the Lieb-Thirring Bounds. The last few years saw a dramatic increase in our understanding of the Lieb-Thirring inequalities: C.y,d = 1 for y > 2 all d (Laptev and Weidl [91]).
C1,d < 2 for all d (Hundertmark, Laptev, and Weidl [69]). Hence also Cy,d < 2 for 1 < y < a , by monotonicity in y.
D. HUNDERTMARK
476
Cod < 81 all d > 3 (Hundertmark [68]). The key observation of Laptev and Weidl [91], which was already noted in Laptev [88], was to do something seemingly crazy: Extend the Lieb-Thirring inequalities from scalar to operator-valued potentials. They considered Schrodinger operators
of the form -A ® lg + V on the Hilbert space L2 (Rd, C), where V is now an operator-valued potential with values V(x) in the bounded self-adjoint operators on the auxiliary Hilbert space 9, and asked whether a bound of the form
trL2(Rd,g)(-® lg + V)ry
cop (2x)d
ffddx
+V(x))ry
(20)
RdRd
holds. Or, again doing the t; integral explicitly with the help of the spectral theorem and scaling, trL2(ad,g)(-A ® lg + V)ry < LOP d J d dx trg(V(x)ry+d/2).
(21)
This seemingly purely technical extension of the Lieb-Thirring inequality turned out to be the key in proving at least a part of the Lieb-Thirring conjecture. In fact, using this type of inequality, Laptev and Weidl noticed the following12 monotonicity properties of the operator-valued Lieb-Thirring constants Cop in the dimension:
1
and
op op 1
(22)
Assume that one knows C3j21 = 1. Then using the first part of (22) repeatedly, we get
1GC3/22
- C3/21C3/21 -1 1=1
that is, C3/22,2 = 1 and hence
1 G(3/23-c3/21c3/22=1 also. Thus an obvious induction in d shows 1 < C3/2 d - C3/2 103/2 d-1
and with the monotonicity in y, one concludes
Cry1 for y>3/2and all dEN. The beauty (and simplicity!) of this observation is that C3/211 = 1 is well-known (already in [118]) and Laptev and Weidl were able to prove that C3j21 = 1 by extending the Buslaev-Faddeev-Zakharov sum rules for the KdV equation [23, 173] to matrix-valued potentials. THEOREM 7 (Laptev-Weidl [91]). One has op C3/2 1 - 1.
121n fact, they observed that Lry d = Lc11Lry+2,d-1 by explicitly multiplying F-functions. So if (21) holds with the classical constant for some -y, then using monotonicity in -y, one can start an induction in the dimension argument. We prefer to avoid multiplying P-functions and to think of this in terms of the Cry d's, using a Fubini-type argument on the operator side instead.
BOUND STATES IN QUANTUM MECHANICS
477
For a nice alternative proof, which avoids the use of sum rules, see [14]. However, as beautiful as this result is, it sheds no light on the physically most important constants C1,d. Recently, Benguria and Loss [15] developed a new viewpoint on the Lieb-Thirring conjecture in one dimension. They connected a simplified "2 bound state version" with an isoperimetric problem for ovals in the plane. However, no progress has been made so far using this viewpoint but it has led to to very interesting mathematical problems; see, for example, [22]. The second submultiplicativity property of Cry d, together with the sharp bound due to Laptev and Weidl, shows that Clop < C1 P C3j2 1 = C1 P . Again by induction, this implies C1d
In particular, if the one-dimensional version of the Lieb-Thirring conjecture (see Conjecture 5) were true for operator-valued potentials, the uniform bound C1 d < 1.16 would follow.
Using ideas from [70], Hundertmark, Laptev, and Weidl extended the sharp bound in the critical case in one dimension to operator-valued potentials [69]. Together with the Aizenman-Lieb monotonicity in 'y, this gives the uniform bound Cop < C°pl = 2, for 1 < -y < 2. Together with the sharp result from Laptev and Weidl, this gives ,
-
THEOREM 8 (Hundertmark-Laptev-Weidl [69]). The bounds C.y,d < 2
for l < ry and all d > 1,
C.y,d :54
for! < y < 1 and all d > 2
on the Lieb-Thirring constants hold.
Moreover, the same estimates for the Lieb-Thirring constants for magnetic Schrodinger operators hold. This follows from the proof of these estimates: One strips off one dimension, but in one dimension there are no magnetic fields, since any "vector" potential in one dimension is gauge equivalent to the zero potential. Thus, by induction, in the dimension, the magnetic vector potential simply drops out, see [69] for details. It soon became obvious that one is not restricted to stripping off only one dimension at a time. More precisely, one has the following two general submultiplicativity properties of C" THEOREM 9 (Hundertmark [68]). For 1 < n < d,
1
-
pp
7,n
and
op
C +n/2 d-n'
In particular, stripping off n = 3 dimensions, and using that Co+3/2,d-3 = 1 for d > 3 by the Laptev-Weidl result, one immediately sees that Cop < Cry 3
for ally > O and d > 3.
(23)
Laptev asked the question [89] (see also [90]) whether, in particular, the CwikelLieb-Rozenblum estimate holds for Schrodinger operators with operator-valued potentials. The proof of this fact was also given in [68]:
D. HUNDERTMARK
478
THEOREM 10 (Hundertmark [68]). Let be some auxiliary Hilbert space and V a potential in Ld/2(Rd Sd/2(g)) with Sd/2 the Schatten-von Neumann operator ideal on G. Then the operator -0 ® 1g + V has a finite number N of negative eigenvalues. Furthermore, one has the bound
N < Lo d L trg(V(x)dl2) dx with
Lo,d < (21rKd)dLp d,
where the constant Kq is given by 8 1-2/q (27r)_d/(T)
Kq =
2
\1+ g-
l/
1/q
This shows Co,d < (27r Kd)d. The constant Kd is exactly the one given by Cwikel [31]. The proof of the above theorem is by extending Cwikel's method to an operator-valued setting. Thereby one recovers Cwikel's bound Co 3 < 81, which is 17 times larger than Lieb's estimate (however, for the scalar case). Nevertheless, using the submultiplicativity, it gives the uniform bound Cod < Co s < 81 which, by monotonicity in ry, also extends to moments 0 < ry < 2. We will not discuss the by now rather big literature on Lieb-Thirring inequalities for the Pauli operator (see, e.g., the review article by Erdos [40]), nor the quite extensive literature on quantum graphs (see [86]), nor the results on quantum wave guides (see, e.g., [39, 42, 43]), but we would like to mention one more recent and, at least for us, rather surprising result on Lieb-Thirring inequalities.
1.8. The Ekholm-Frank Result. It is well known that an attractive potential does not necessarily produce a bound state in three and more dimensions. This follows, for example, from Hardy's inequality, which says that in dimension three and more, the sharp operator inequality (d - 2)2 < 41x12
(24)
holds. Using this, one can refine the usual Lieb-Thirring inequalities in the following way: Using Hardy's inequality, for any e E (0, 1), one has
_
-0+V>-e0+(1-e)(4I
x12)2
+V.
Thus with the known Lieb-Thirring inequality, one gets
tr(-0 +V)ry < =
(d 41x12)2 C )a,d/2 JfRdXRd (1512 + (1 - e)
+V(x)J7
(2
L.y,de_
d/2 d
fm
C(1
- e)
(d - 2)2 41x12
+ V(X) )
'y+d/2
dx.
Of course, as a -> 0, the constant in front of the integral diverges. Very recently, Ekholm and Frank established that one can nevertheless take the limit e -> 0. They proved the rather surprising result
BOUND STATES IN QUANTUM MECHANICS
479
THEOREM 11 (Ekholm-Frank [38]). For moments -y > 0, the inequality
tr(-0 + V)ry <
2) liEdxRd
(41x12)2 +V(x))ry d dx
(2
= LEd f d (
(d - 2)2 4IxI2
\'y+d/2
+V(x) I
dx
/
holds in dimension three and more.
Thus, as far as moments are concerned, only the part of the potential below the critical Hardy potential is responsible for bound states. This amounts to an infinite phase-space renormalization on the level of the Lieb-Thirring inequality. Note that the Ekholm-Frank bound cannot hold for ry = 0. Also, effective bounds on CEd (resp. LEd) are not known.
2. Multi-Particle Coulomb Schrodinger Operators 2.1. The Coulomb Hamiltonian. The Hamiltonian for N electrons in the field of M nuclei is given by
H=HN,z,R=T+VC=T+VC,
(26)
where T = EN1 -Aj is the kinetic energy of N electrons, and (27)
Vc = Vee. + Vee
with
NM j=1 a=1
Z
Ix -Ra
the electron-nucleus interaction, 1
Vee = i
Ixi
<j
the electron-electron repulsion. Sometimes, one also considers Vnn =
za_ E Ra - RI' 0<0 I
the repulsion of the M nuclei at positions R = (R1i ... , RM) E R3M. We keep the position of the nuclei fixed, for simplicity. The electrons have spin q. In real life q = 2. Thus the N-electron operator HN,Z,R is defined on /feermions. ^ N (L2 (R2), C4), the antisymmetric subspace of L2([83N,CgN), since electrons are Note that in the units we use, the ground state energy of Hydrogen is 4. Zhislin's theorem [174] (see also [147] for a simple proof) guarantees the existence of a ground state if N < Ea Za + 1. One of the problems is to compute the ground state and the ground state energy, that is, N
EQ (N, Z, R) = inf
((VI, HN,z,R'O) 10 E / \(L2(R3, C4)),
11011
= 1)
(28)
and the minimizing groundstate wave function. The catch is that, although the Schrodinger equation is a linear equation, even the problem of two electrons in the field of one nucleus is not exactly solvable. Moreover, due to the exponential scaling
D. HUNDERTMARK
480
of the degrees of freedom, there are no efficient methods to solve the Schrodinger equation approximately, even for moderate numbers of electrons. Nevertheless, one can give a rather complete answer in the limit of large atoms or molecules. As shown by Lieb and Simon, Thomas-Fermi theory becomes exact in this case.
2.2. Thomas-Fermi Theory. Thomas-Fermi theory is a simplification of the usual Schrodinger equation. For an excellent review and proofs of results, see [97]. A fundamental object in this theory (or approximation) is not the wave function any more, but the so-called single-particle density. Given a normalized wave function O c AN L2 (R', Co), recall that its single particle density is given by
p,p (x) = N
J
10(x, a1, x2, a2, ... , XN, QN)I2 dx2 ... dxN.
(29)
The electron nucleus interaction is easily expressed in terms of the density, M
(V, Ven) _ -
Z
j=1 JR-
Ix
- Rj I pp (x) dx.
(30)
The kinetic energy T = (0, Hoo) cannot be expressed by density p, but it can be approximately expressed by the density. For a free electron gas in a container of sidelength L in d dimensions, with periodic boundary conditions, say, the ground state energy is given by
(2)2 7r
T= (n,o)EZdx{1,..., q}
I<no
In the limit N - oo, one has
N=
do = gwdn0
g
1
111n I<no
(, ,a)EZdx{1,...,q} InI<"o
with wd the volume of the sphere of radius one. Also, again in the limit N z T = (Lv)
g
n2 do = Inl<no
d
(27r)2
2
d d +2
W dd
L
oo,
nd+2 0
N(d+2)/d
d + 2 (gwd)2/d
L2
Thus the kinetic energy density should be given by
T Ld
_
d
(2ir)2
d + 2 (gwd)2/d
N (d+2)/d Ld
q-2/dKa Fp(d+2)/d
with the density p = N/Ld and the Thomas-Fermi constant Ka F = dd+z (id d For slowly varying densities, the above suggest that the kinetic energy of N electrons (in the large N limit) is well-approximated by f TF = d with Kd - d + 2 /d2 (31) T = q 2/d K TF J d p(x) (d+2)/d dx d
BOUND STATES IN QUANTUM MECHANICS
481
In fact, the duality of the Lieb-Thirring bound and the kinetic energy bound for fermions shows that, as an inequality, this holds with the Thomas-Fermi constant, if the Lieb-Thirring inequality is true with the classical constant, that is, if C1,d = 1. Now let us concentrate on the physical relevant case of three space dimensions. Similarly as for the kinetic energy, the Coulomb repulsion of the electrons cannot be expressed by the density alone, but, keeping fingers crossed, it should be wellapproximated by the Coulomb integral. In fact, up to a small error, the Coulomb integral is a lower bound; see Lieb-Oxford [111] and Lieb [96]:
2<j x
xa
>2
fj
y) 3xR3
dxdy - cLOJ 3 PG(x)413 dx, (32)
where 1.234 < cLO < 1.68. For an alternative derivation, with a slight loss in the constants, see, for example, [121].
Ignoring the error terms and putting the above together, one is led to the Thomas-Fermi energy functional
S(P) = S(P, Z, R) = 4 2/3x3
F
f
/ P(x)5/3 dx - J 3 V(x)p(x) dx 3
P, (x)P,G (y) dxdy
+a
(33)
x-y with V(x) = EM1 ZRn for Coulomb matter. N(p) = fE3 p(x) dx = N is the J R3XR3
number of electrons (although in this formulation it does not have to be an integer). The functional is well-defined on the spaces S = {pI p > 0, p E L5/3(R3) n LI (R3)} Sa = {p E SI f p(x) dx < A}, and Saa = {p E SI f p(x) dx = Al. The Thomas-Fermi energy is given by ET F (N, Z, R) = inf {E(p, Z, R) I p E SaN }.
(34)
P
The catch with this definition is that Saa is not a convex set. But for N < Zn, it is shown in [116] that the minimum over the convex set SN exists and gives the same energy as ET F (N, R, Z). This follows from the strict convexity of the Thomas-Fermi functional. One can show (see [116, 97]) that for Coulomb matter, the function A --3 ETF (A, Z, R) is strictly convex on [0, E Z,,,] and constant if A is bigger than the total charge of the nuclei. Moreover, for A less or equal to the total charge, the minimizing density of the TF functional exists and is unique and is a solution of the so-called Thomas-Fermi equations. The Thomas-Fermi energy functional and the Thomas-Fermi energy have a natural scaling. For any 6 > 0, let p6 (X) = 62P(61/3x). Then S(P8, 6Z, 6-1/3R) = 6713E(p, Z, R).
(35)
So for a Coulomb system with nuclear charges Z = (Z1,. .. , ZM) at positions R = (R1,. .. , RM), this implies the scaling law ETF(6N, 6Z, 6-113R) = 67/3ETF(N, Z, R).
(36)
The main quantum mechanical limit theorem, proved by Lieb and Simon in [116] is that asymptotically, the Thomas-Fermi energy gives the correct answer for the ground state energy of atoms or molecules with a large nuclear charge.
D. HUNDERTMARK
482
THEOREM 12 (Lieb-Simon [116]). lim b
EQ (SZ)
1/3 R)
6713
00
= ETF(.A, Z, R).
Moreover, the quantum m-point densities, suitably rescaled, converge to a product p(x1) ... p(xm,) of Thomas-Fermi densities. Originally, the above result was proved using a decomposition of the space into boxes together with DirichletNeumann decoupling techniques. Motivated by a result of Thirring [165], Lieb used coherent states techniques in [97]. In addition, much more is known by now about lower order terms in the asymptotic for large atoms and in part for molecules. The energy has the asymptotic EQ(Z, R) = -aTF 1: Z,7/3 + 2 n
ZN + CiZ513 + O(Z5/3-E) n
(here aTF is the Thomas-Fermi prediction). The second term in this asymptotic is the Scott correction (see [140]); for molecules it was formulated in [97]. The second term asymptotic was established in a series of papers for atoms by Hughes [67] and Siedentop and Weikard [143, 142, 144]. For molecules it was established by Ivrii and Sigal [77]; see also [160]. There are also some results on the convergence of suitably scaled ground states, the so-called strong Scott conjecture; see [73, 74, 75]. The next higher correction due to Dirac and Schwinger was established for atoms in a monumental series of papers by Fefferman and Seco [54]. The Fefferman-Seco proof is a two-step proof. First, one reduces the problem, with an error less than O(Z5/3-E), to the study of an effective one-particle Hamiltonian which is given by the mean-field approximation. The second step then
consists of a very detailed analysis of the bound states of this mean-field Hamiltonian. Based on his work on error bounds for the Hartree-Fock approximation in [5], Bach gave in [6] a much simpler proof of the first step, the reduction to the mean-field Hamiltonian. For a variation on Bach's proof, see [63]. Bach's proof also works for molecules and not only atoms, but the spectral properties of the meanfield Hamiltonian have been established to the needed accuracy only for atoms. In this case, the mean-field Hamiltonian is rotationally symmetric and the study of its bound states, within accuracy of can be done by a quite delicate WKB O(Z5/3-E)
analysis. The abovementioned landmark paper of Simon and Lieb established many more
interesting results for the Thomas-Fermi theory of matter, for example, a rigorous treatment of the Thomas-Fermi theory of solids, also Teller's no-binding result for Thomas-Fermi theory. It says that, taking the nucleus-nucleus repulsion into account, the Thomas-Fermi energy of a molecule is always bigger than the sum of the energies of the individual atoms. Thus binding, which is due to the outermost electrons in atoms, is not correctly described by TF theory. For any nuclear charge vector Z = (Z1, ... , ZM) and positions of nuclei R = (R1, ... , RM), define
eTF(A,Z,R) =ETF(A,Z,R)+ E n-1
the energy with nucleus-nucleus repulsion. Then
ZnZm
A, - Rm l
BOUND STATES IN QUANTUM MECHANICS
483
THEOREM 13 (Teller's no-binding theorem). For any strictly positive Z = (Z1,.. ZM), that is, Zj > for all j, and R and A > 0, one has . ,
eTF(A, Z, R) > min [eTF(A', ZA, RA) + eTF(A - A', ZB, RB) 0<1\1<1\
for any decomposition ZA = (Z1, ... , Zk), ZB = (Zk+1, ... , Zn), and similarly for RA and RB .
This theorem is at the heart of the proof of stability of matter by Lieb and Thirring. The no-binding lower bound and the behavior of the Thomas-Fermi energy in terms of A together with scaling implies, in particular, that K
K
ETF(Zn, Zn) = ETF(1, 1) j ZZ/3.
eTF(A, Z, R) >n=1
(37)
n=1
By scaling, one has
E (1,1)=- K with K = q-2/3K3 F. Numerically, it is known that ETF = 2.21. So with the classical value of KTF and q = 2 for real electrons, one has ETF(1,1) > -0.385. On the other hand, it is known by now that molecules do bind in modifications of the Thomas-Fermi model, for example, the Thomas-Fermi-von Weizsacker model, as was established by Catto and Lions [26].
2.3. Stability of Matter. Following Lieb and Thirring, Teller's no-binding theorem for Thomas-Fermi theory and the kinetic energy lower bound in Theorem 1
has a beautiful application to the stability-of-matter problem. A basic fact of astrophysics claims that bulk matter undergoes gravitational collapse in the absence of nuclear forces. Onsager asked in the 1930's why bulk matter does not undergo an electrostatic collapse, too. This is due to the Pauli principle. The first proof of this fact is due to Dyson and Lennard in the 1960's [34, 35]. However their proof was rather complicated. On the other hand, this defect of Thomas-Fermi theory is in turn a chance to give a physically motivated and very appealing proof of stability of quantum mechanical matter, as soon as one can give a lower bound of the quantum mechanical energy of an antisymmetric N-fermion wave function in terms of the Thomas-Fermi energy of its one-particle density pp. The lower bound provided in Theorem 1 was the main tool of Lieb and Thirring [117], leading to a much simpler proof of stability of matter than the original one by Dyson and Lenard. At the same time, by relating the stability-of-matter question to the Thomas-Fermi model, it provided considerable new physical insight in the mechanism which prevents this collapse. (See [95, 102, 103] for a nice presentation of the physical and mathematical aspects of this problem.) Also, it turned out that not only the existence of these bounds but good estimates on the constants in these inequalities are of considerable importance for a rigorous understanding of the properties of matter [28, 46, 47, 62]. We will only sketch this application here. Note that the kinetic energy lower bound and the Lieb-Oxford bound give the following lower bound for the quantum
D. HUNDERTMARK
484
mechanical energy (V),
(HC + Vnn)
£KF
(PV))
-c LO Jf
P'P (x)4/3 dx.
Here £KF is the Thomas-Fermi functional with KTF in the kinetic energy term Recall that q = 2 for real electrons and that the replaced by equality C1,d = 1 is a long-standing open conjecture. The best known bounds so far say that C1,d < 2. (gC1,d)-2/3KTF.
Using Holder's inequality, one has f PO (x)4/3 dx <
(f(x) dx l
1/2
1/2
)
\f Pp (x)5/3 dx
< 4 N + ry
I
fp(x)/3 dx
Thus, replacing y by cLOy
(HC + Vnn)V)) > £KF(p) -
)2
(C
4
yN - fp(x5/3dx
(cL0)z
yN.
4
Taking the no-binding result of Thomas-Fermi theory into account, the right-hand side is bound below by e
K
LO 2
TF
1/y E Zn/3 n
- (c 4) yN =
TFN CK
1 1/y
7/3 C L. N n - c }Nl
l
with c = (cLC)2/(4ETF) < 0.3193. Maximizing w.r.t. y leads to the bound (0, (HC + Vnn)W)
-(gC1,3)2/3 KTF N (
\z E 7/3/NJt + n
Together with the easy bound 2ab < a2 + b2, one gets
/
ETF
(0, (HC + Vnn) `V) >- -(gC1,3)2/3 KTF (1 + C) I N + \
Znl3 l 1: n
which, in the case of real fermions q = 2 and the estimates C1,3 < 2 and c < 0.3193, gives (
, (HC + Vnn)O) ? -0.804 I N +
\
Zn7/3 f . n
/
(38)
As shown by Thirring [165], it is also known that, asymptotically, ThomasFermi theory is, indeed, the correct lower bound for the stability-of-matter result, K
Zn7/3(1 + O(Z 2/33),
(0, (Hc + Vnn,)4b) > -0.385 n=1
2/33) depend on the number of electrons. where the constants in the error term O(Zz Somewhat better error estimates were established in [123]. In particular, the robust estimate in (38) is not far from the truth. Note also that, following the route leading to (38), the best possible lower estimate for the Coulomb energy is -0.5078(N + Zn7/3) (using the conjectured value C1,3 = 1 and q = 2). En
BOUND STATES IN QUANTUM MECHANICS
485
REMARK 14. On can also use Thomas-Fermi theory to get a lower bound for the electron-electron repulsion: Reshuffling the terms in the stability result for Thomas-Fermi theory,13 1
J
-
p5/3 dx
1
x - yl
7=1
p(y) dy + D(p, p) + i
1
1
>
ET FyN
gives
?
1
1
Noting (,0,
ENI
b, E
1 x' -1 Y1 p(y) dy - D(p, p) - ETFyN - I/
j=1
f -s'
y1
1
i<j Ixi -x.71
fP5/3dx.
p(y) dyb) = 2D(pp, p) and choosing p= pp, one sees
b > D(pV,, pp) - TFyN - 1
y
/
f
P5/3
dx.
This way one gets the lower bound
(Hc + Vnn))
>-
JF K
)2
N 1+
Zn/3 /N n
Again, using Cauchy-Schwartz, a similar lower bound as before follows from this, with the factor 1 + c replaced by 2. The "best possible" lower bound for the energy of real matter this way is
/
('0, (HC + Vnn)0) >- -0.771 N + E Za/3
\
n
(see, e.g., [20, 122]), which should be compared with (38). Of course, this is not best possible, since, as demonstrated above, using the Lieb-Oxford bound improves the estimate quite a bit. We will not discuss other approaches to this circle of problems [60, 61, 44], nor the extension of the stability results to other models of real matter, for example, the Pauli operator [56, 106, 110, 48, 49], or relativistic models [52, 119, 120, 109, 113, 112], and, more recently, matter interacting with quantized electromagnetic fields [50, 51, 107, 108], but point the interested reader to the excellent reviews by Lieb [104, 105]. There is also a growing literature on stability/instability of the relativistic electron-positron field in Hartree-Fock approximation; see, for example, [7, 8, 9, 21, 64, 65, 711.
3. More on Bound States for Atoms It is known that an N-electron Coulomb system has a bound state as long as the number of electrons is lower than the total charge of the nuclei (plus one). This is known as Zhislin's theorem [174]. In nature, one can observe once negatively charged free atoms, but not twice or more negatively charged atoms. Formulated conservatively, one expects the maximal negative ionization to be independent of 13This was the route originally taken by Lieb and Thirring in [117].
D. HUNDERTMARK
486
the nucleus charge Z. This is known as the ionization conjecture (see, e.g., the review article by Simon [155]). For a more precise form of this conjecture, let
E(-)+ :: Z
HN,
j=1
1
1 (39)
Nlxi-xjI
be the Hamiltonian for N electrons in the field of an atom of charge Z. For fermions, the Hilbert space is AN L2(ll 3 (C2) Define
E(N, Z) = inf o,(HN,Z)
(40)
I = I(N, Z) = inf Oess(HN,Z) - info (HN,Z).
(41)
and the ionization energy By the HVZ theorem, Uess(HN, Z) = [E (N - 1, Z), oo), so
I(N, Z) = E(N - 1, Z) - E(N, Z). Note that HN,Z has a bound state, a discrete eigenvalue below its essential spectrum, if and only if I(N, Z) > 0. Due to electrostatic reasons, one expects that there should be an Ncr(Z) such
that
E(N, Z) = E(N - 1, Z) for all N > Ncr(Z), that is, the atom can bind only a finite number of electrons. The picture is not that simple, however. Heuristically, ignoring all many-body effects, the potential felt by the Nth electron is given by the effective potential Ueff(x)
-ICI +
f
x
1 ylpG(y)dy,
(42)
where p,, is the single-particle density of the other N - 1 electrons. The physics of the system should be somehow described by the effective Hamiltonian (43) Heff = -0 + Ueff. When N increases, pp increases and hence Ueff should increase at least in some average sense. Moreover, when N > Z, Newton's theorem implies Ueff(x) > 0 for large x. But for x near zero, Ueff is very negative since, due to the uncertainty principle, the electrons cannot concentrate too much close to zero. Hence the attraction of the nucleus is never fully screened. Nevertheless, eventually, Ueff will cease to have a bound state. Note that there is a big difference for fermions and bosons from this viewpoint: Whereas in order to bind N bosonic "electrons" the effective one-particle Hamiltonian should have at least one bound state, for fermionic systems, due to the Pauli principle, if one wants to bind N electrons the effective one particle Hamiltonian needs to have at least N bound states. For real atoms, this Nth bound state seems to disappear precisely when N - Z + 1. Bosons, however, should be much more easily bound to an atom than fermions. Thus, one expects the atomic Hamiltonian (39) not to have a fermionic bound
state if N > Ncr = Z + Q, where Z is the total charge of the nucleus and Q > 0 some fixed positive number, hopefully of the order of one. This innocent-looking conjecture, known as the ionization conjecture (see, e.g., the review article by Simon [155]), has withstood all attempts to prove it, even for very large but fixed excess charge Q. Only partial results are known.
BOUND STATES IN QUANTUM MECHANICS
487
3.1. Ruskai-Sigal-type Results. We are far from understanding the ionization conjecture rigorously, at least for fermions. One of the first results in this direction is the result by Ruskai and Sigal.
THEOREM 15 (Ruskai [134, 135]; Sigal [145, 146]). For any Z, there exist Ncr(Z) such that
I(N, Z) = E(N - 1, Z) - E(N, Z) = 0 for all N > Ncr(Z). In addition, for fermions, one has the bound lim sup
2
Z--oc
This theorem was proven independently by Ruskai [134, 135] and Sigal [145, 146]. A huge improvement of this theorem is due to Lieb. He showed THEOREM 16 (Lieb [101]). Independently of the statistics of the particles,
Ncr(Z) < 2Z + 1. In fact, Lieb also treats molecules and other refinements like magnetic fields and relativistic kinetic energies; see also Ichinose [76]. Another improvement of the Ruskai-Sigal bound is given by Lieb, Sigal, Simon, and Thirring [114]. They showed that large atoms are asymptotically neutral, lim Ncr(Z) = 1. Z
(44)
Z-00
Unfortunately, their proof used a compactness argument in the construction of suitable localizing functions and did not give any quantitative information or error estimates for finite nuclear charges Z. The first quantitative result seems to have been given by Fefferman and Seco in 1990. They proved THEOREM 17 (Fefferman-Seco [53]).
Ncr(Z) < Z + O(Z')
with a =
47 56
Shortly afterwards, this approach was simplified by by Sigal, Seco, and Solovej [141] who gave the estimate THEOREM 18. One has the bound
I(N, Z) < C1Z4o,13 - C2(N - Z)Zc,/3 for the ionization energy and as a consequence also
Ncr(Z) < Z + CZ-Here a = 47/56 as in the Fefferman-Seco theorem. Nevertheless, these bounds are still far from the (expected) truth, since Ncr(Z)Z and I(Z, Z) should be bounded in Z.
D. HUNDERTMARK
488
3.2. Existence of Highly Negative Ions for Bosonic Atoms. As discussed above, for an atom to bind N electrons, the effective one particle Hamiltonian Heff should have at least N bound states. For bosons, however, it is enough
that it has at least one bound state. Thus a "bosonic" atom should be able to bind many more particles than one might naively expect from purely electrostatic reasons. This is indeed the case, THEOREM 19 (Benguria-Lieb [12]). For bosonic atoms, lim inf N,ZZ) z--oo
> 1+7.
Here 0 < -y < 1 is defined via the ground state of the Hartree functional: Let /' be the unique positive solution of the non-linear Hartree equation
-off - (iT - I,0I2*0 = 0. Then 1,0(x) I2
(1 +'Y)Z = f
dx.
R3
Benguria (see [97] for a reference) showed that 0 < ry < 1. Moreover, numerically it is known that ry = 0.21; see [11]. Later, Solovej proved a corresponding upper bound using similar ideas as in [141],
THEOREM 20 (Solovej [158]). For bosonic atoms,
N,, (Z) < (1 + 7)Z + CZ719. In particular, NrZ
lim
Z-oo
Z)
= 1+7.
3.3. Solution of the Ionization Conjecture in Hartree-Fock Theory. In Hartree-Fock theory (or better in the Hartree-Fock approximation), one does not consider the full N-body Hilbert space AN L2 ([83, C2), but restricts the attention to pure Slater determinants
0=vln...AVN
(45)
where the single-particle orbitals v? E L2(ll3,(C2). In other words, the density matrix
Y_ J ...
P(X,Ce,X2, 0'2,
.
, xN, 0'N) b (y, )3, x2) Q2 ... , xn,, ON) dx2... dxN (46)
is assumed to be a projection operator. Note that, for arbitrary
CAL 2 (]E83, C2),
the associated density matrix -y p is a bounded operator with 0 < rye < 1 on L2(R3
C2)
The Hartree-Fock energy of an atom of charge Z with N electrons is then defined to be
E HF = EHE (N, Z) =
inf (0, HN,z')
(47)
BOUND STATES IN QUANTUM MECHANICS
A little bit of calculation shows that for Slater determinants
489
,
(V), HN,Zb) = EHF('y )
with EHF(y) = trL2(a3,c2)
((-A - xI )y) + D(y) - Ex(7)
(48)
I
where trc2 (-Y (x, x)y(y, y)) dxdy
D(y) = z ff Ix - yI is the direct part of the Coulomb energy and
Ex(-y) = 2 ff
trC2I Iry(, )I2
dxdy
the exchange term. In particular, EHF(N, Z) = inf {,,HF(,y)I'y*'y ='y, try = N}.
The existence of a minimizing projection operator y for N < Z + 1 was proven by Lieb and Simon [115]. Moreover, given the Hartree-Fock functional, one can relax the assumption that y is a projection operator, as was shown by Lieb [98] EHF(N, Z) = inf {,,1F(y)10 < y < 1, tr(y) = N}. For a simple proof of this see [5]. For the Hartree-Fock approximation, Solovej gave a proof showing the ionization energy and maximum surcharge an atom can bind to be bounded uniformly in Z.
THEOREM 21 (Solovej [159]). For a neutral atom, the ionization energy in Hartree-Fock approximation is bounded, that is, IHF(Z, Z) = EHF(Z - 1, Z) - EHF(Z, Z) = O(1) as Z - oo.
Moreover, there exists a finite Q > 0 such that for all N > Z + Q, there are no minimizers for the Hartree-Fock functional among N-dimensional projections.
A related result for the Thomas-Fermi-von Weizsacker model was shown in [13]. Unfortunately, all results of this precise form are, so far, for models which shed no light on the Schrodinger case.
References [1] F. Ahia: Finiteness of the discrete spectrum of (N > 3) N-body Schrodinger operators, which have some determinate subsystems that are virtual at the bottom of the continuum. J. Math. Phys. 33 (1992), no. 1, 189-202. [2] M. Aizenmann and E. H. Lieb: On semi-classical bounds for eigenvalues of Schrddinger operators. Phys. Lett. 66A (1978), 427-429. S. Albeverio, R. Hoegh-Krohn, and T. S. Wu: A class of exactly solvable three-body quantum mechanical problems and the universal low energy behavior. Phys. Lett. A 83 (1981), no. 3, 105-109. [4] M. A. Antonec, G. M. 2hislin, and I. A. §eresevskii: Discrete spectrum of the Hamiltonian [3]
of a quantum N-particle system. (Russian) Teoret. Mat. Fiz. 16 (1973), no. 2, 235-246. English translation: Theoret. and Math. Phys. 16 (1973), no. 2, 800-809 (1974). [5] V. Bach: Error bound for the Hartree-Fock energy of atoms and molecules. Comm. Math. Phys. 147 (1992), no. 3, 527-548. [6] V. Bach: Accuracy of mean field approximations for atoms and molecules. Comm. Math. Phys. 155 (1993), no. 3, 295-310.
D. HUNDERTMARK
490
[7] V. Bach, J.-M. Barbaroux, B. Helffer, and H. Siedentop: Stability of matter for the HartreeFock functional of the relativistic electron-positron field. Doc. Math. 3 (1998), 353-364 (electronic).
[8] V. Bach, J.-M. Barbaroux, B. Helffer, and H. Siedentop: On the stability of the relativistic electron-positron field. Comm. Math. Phys. 201 (1999), no. 2, 445-460. [9] J.-M. Barbaroux, W. Farkas, B. Helffer, and H. Siedentop: On the Hartree-Fock equations of the electron-positron field. Comm. Math. Phys. 255 (2005), no. 1, 131-159. [10] B. Baumgartner: The Thomas-Fermi theory as result of a strong-coupling-limit. Comm. Math. Phys. 47 (1976), no. 3, 215-219. [11] B. Baumgartner: On Thomas-Fermi-von Weizsacker and Hartree energies as functions of the degree of ionisation. J. Phys. A 17 (1984), no. 8, 1593-1601. [12] R. Benguria and E. H. Lieb: Proof of the stability of highly negative ions in the absence of the Pauli principle. Phys. Rev. Lett. 50 (1983), 1771-1774. [13] R. Benguria and E. H. Lieb: The most negative ion in the Thomas-Fermi-von Weizsacker theory of atoms and molecules. J. Phys. B 18 (1985), no. 6, 1045-1059. [14] R. Benguria and M. Loss: A simple proof of a theorem of Laptev and Weidl. Math. Res. Lett. 7 (2000), no. 2-3, 195-203. [15] R. Benguria and M. Loss: Connection between the Lieb-Thirring conjecture for Schrodinger operators and an isoperimetric problem for ovals on the plane. Partial Differential Equations and Inverse Problems, pp. 53-61, Contemp. Math., 362, American Mathematical Society, Providence, RI, 2004. [16] M. S. Birman: The spectrum of singular boundary problems. Mat. Sb. 55 (1961), no. 2 125-174. English translation: Amer. Math. Soc. Trans. (2) 53 (1966), 23-80. [17] M. S. Birman and V. V. Borzov: The asymptotic behavior of the discrete spectrum of certain singular differential operators. (Russian) Problems of Mathematical Physics, No. 5: Spectral Theory, pp. 24-38. Izdat. Leningrad. Univ., Leningrad, 1971. [18] M. Sh. Birman and M. Z. Solomyak: Piecewise polynomial approximations of functions of classes WY'. (Russian) Mat. Sb. (N.S.) 73 (115) (1967), 331-355. [19] M. Sh. Birman and M. Z. Solomyak: Schrodinger operator. Estimates for number of bound states as function-theoretical problem. Spectral Theory of Operators (Novgorod, 1989), pp. 1-54. Amer. Math. Soc. Transl. Ser. 2, 150, American Mathematical Society, Providence, RI, 1992. [20] Ph. Blanchard and J. Stubbe: Bound states for Schrodinger Hamiltonians: phase space methods and applications. Rev. Math. Phys. 8 (1996), no. 4, 503-547. [21] R. Brummelhuis, N. Rohrl, and H. Siedentop: Stability of the relativistic electron-positron field of atoms in Hartree-Fock approximation: heavy elements. Doc. Math. 6 (2001), 1-9 (electronic).
[22] A. Burchard and L. E. Thomas: On an isoperimetric inequality for a Schrodinger operator depending on the curvature of a loop. J. Geom. Anal. 15 (2005), no. 4, 543-563. [23] V. S. Buslaev and L. D. Faddeev: Formulas for traces for a singular Sturm-Liouville differential operator. (Russian) Dokl. Akad. Nauk SSSR 132 13-16. English translation: Soviet Math. Dokl. 1 (1960), 451-454. [24] E. A. Carlen and M. Loss: Extremals of functionals with competing symmetries. J. Funct. Anal. 88 (1990), no. 2, 437-456. [25] E. A. Carlen and M. Loss: Competing symmetries of some functionals arising in mathematical physics. Stochastic Processes, Physics and Geometry (Ascona and Locarno, 1988), pp. 277-288, World Scientific, Teaneck, NJ, 1990. [26]
I. Catto and P.-L. Lions: Binding of atoms and stability of molecules in Hartree and Thomas-Fermi type theories. Part 1: A necessary and sufficient condition for the stability of general molecular systems, Comm. Partial Differential Equations 17 (1992), no. 7/8, 1051-1110. Part 2: Stability is equivalent to the binding of neutral subsystems, Comm. Partial Differential Equations 18 (1993), no. 1/2, 305-354. Part 3: Binding of neutral subsystems, Comm. Partial Differential Equations 18 (1993), no. 3/4, 381-429. Part 4: Binding of neutral systems for the Hartree model, Comm. Partial Differential Equations 18 (1993), no. 7/8, 1149-1159.
BOUND STATES IN QUANTUM MECHANICS
491
[27] J. G. Conlon: A new proof of the Cwikel-Lieb-Rosenbljum bound. Rocky Mountain J. Math. 15 (1985), no. 1, 117-122.
[28] J. G. Conlon, E. H. Lieb, and H.-T. Yau: The Coulomb gas at low temperature and low density. Comm. Math. Phys. 125 (1989), no. 1, 153-180. [29] P. Constantin and C. Foias: Navier-Stokes Equations. Chicago Lectures in Mathematics. University of Chicago Press, Chicago, 1988. [30] M. M. Crum: Associated Sturm-Liouville systems. Quart. J. Math. Oxford Ser. (2) 6 (1955), 121-127.
[31] M. Cwikel: Weak type estimates for singular values and the number of bound states of Schrodinger operators. Ann. of Math. 106 (1977), no. 1, 93-100. [32] P. Deift: Applications of a commutation formula. Duke Math. J. 45 (1978), no. 2, 267-310. [33] P.
Deift and R. Killip:
On the absolutely continuous spectrum of one-dimensional
Schrodinger operators with square summable potentials. Comm. Math. Phys. 203 (1999), no. 2, 341-347. [34] F. Dyson and A. Lenard: Stability of matter I. J. Math. Phys. 8 (1967), 423-434. [35] F. Dyson and A. Lenard: Stability of matter H. J. Math. Phys. 9 (1967), 698-711. [36] A. Eden and C. Foias: A simple proof of the generalized Lieb-Thirring inequalities in onespace dimension. J. Math. Anal. Appl. 162 (1991), no. 1, 250-254. [37] V. Efimov: Energy levels arising from resonant two-body forces in a three-body system. Phys. Lett. B 33 (1970), 563-564. [38] T. Ekholm and R. Frank: On Lieb-Thirring inequalities for operators with virtual level. To appear in Commun. Math. Phys. [39] T. Ekholm and H. Kovarik: Stability of the magnetic Schrodinger operator in a waveguide. Commun. Partial Differential Equations 30 (2005), no. 4, 539-565. [40] L. Erdds: Recent developments in quantum mechanics with magnetic fields. Preprint 2006. [41] W. D. Evans and R. T. Lewis, N-body Schrodinger operators with finitely many bound states. Trans. Amer. Math. Soc. 322 (1990), no. 2, 593-626. [42] P. Exner, H. Linde, and T. Weidl: Lieb-Thirring inequalities for geometrically induced bound states. Lett. Math. Phys. 70 (2004), no. 1, 83-95. [43] P. Exner and T. Weidl: Lieb-Thirring inequalities on trapped modes in quantum wires. Proc. XIII Internat. Congr. Math. Phys. (London, 2000), pp. 437-443. International Press, Boston, 2001.
[44] P. Federbush: A new approach to the stability of matter problem. I, H. J. Math. Phys. 16 (1975), 347-351; ibid. 16 (1975), 706-709. [45] C. Fefferman: The uncertainty principle. Bull. Amer. Math. Soc. (N.S.) 9 (1983), no. 2, 129-206
.
[46] C. Fefferman: The N-body problem in quantum mechanics. Frontiers of the Mathematical Sciences: 1985 (New York, 1985). Comm. Pure Appl. Math. 39 (1986), no. S, suppl., S67S109.
[47] C. Fefferman: The atomic and molecular nature of matter. Rev. Mat. Iberoamericana 1 (1985), no. 1, 1-44. [48] C. Fefferman: Stability of Coulomb systems in a magnetic field. Proc. Nat. Acad. Sci. U.S.A. 92 (1995), no. 11, 5006-5007. [49] C. Fefferman: On electrons and nuclei in a magnetic field. Adv. Math. 124 (1996), no. 1, 100-153.
[50] C. Fefferman, J. Frohlich, and G. M. Graf: Stability of nonrelativistic quantum mechanical matter coupled to the (ultraviolet cutoff) radiation field. Proc. Nat. Acad. Sci. U.S.A. 93 (1996), no. 26, 15009-15011 (electronic). [51] C. Fefferman, J. Frohlich, and G. M. Graf: Stability of ultraviolet-cutoff quantum electrodynamics with non-relativistic matter. Comm. Math. Phys. 190 (1997), no. 2, 309-330. [52] C. Fefferman and R. de la Llave: Relativistic stability of matter. I. Rev. Mat. Iberoamericana 2 (1986), no. 1-2, 119-213. [53] C. Fefferman and L. A. Seco: Asymptotic neutrality of large ions. Comm. Math. Phys. 128 (1990), no. 1, 109-130. [54] C. Fefferman and L. A. Seco: On the energy of a large atom. Bull. Amer. Math. Soc. (N.S.) 23 (1990), no. 2, 525-530. Aperiodicity of the Hamiltonian flow in the Thomas-Fermi potential. Rev. Mat. Iberoamericana 9 (1993), no. 3, 409-551.
D. HUNDERTMARK
492
On the Dirac and Schwinger corrections to the ground-state energy of an atom. Adv. Math. 107 (1994), no. 1, 1-185. The density in a one-dimensional potential. Adv. Math. 107 (1994), no. 2, 187-364. The eigenvalue sum for a one-dimensional potential. Adv. Math. 108 (1994), no. 2, 263335.
The density in a three-dimensional radial potential. Adv. Math. 111 (1995), no. 1, 88-161. Weyl sums and atomic energy oscillations. Rev. Mat. Iberoamericana 11 (1995), no. 1, 165-226.
The eigenvalue sum for a three-dimensional radial potential. Adv. Math. 119 (1996), no. 1, 26-116.
[55] C. Foias, 0. Manley, and R. Temam: Attractors for the Benard problem: existence and physical bounds on their fractal dimension. Nonlinear Anal. 11 (1987), no. 8, 939-967. [56] J. Frohlich, E. H. Lieb, and M. Loss: Stability of Coulomb systems with magnetic fields. I. The one-electron atom. Comm. Math. Phys. 104 (1986), no. 2, 251-270. [57] C. S. Gardner, J. M. Greene, M. D. Kruskal, and R. M. Miura: Korteweg-deVries equation and generalization. VI. Methods for exact solution. Comm. Pure Appl. Math. 27 (1974), 97-133.
[58] V. Glaser, H. Grosse, A. Martin, and W. Thirring: A family of optimal conditions for the absence of bound states in a potential. Studies in Mathematical Physics, Essays in Honor of Valentine Bargmann, Princeton University Press, Princeton, NJ, 1976. [59] J.-M. Ghidaglia, M. Marion, and R. Temam: Generalization of the Sobolev-Lieb-Thirring inequalities and applications to the dimension of attractors. Differential Integral Equations 1 (1988), no. 1, 1-21. [60] G. M. Graf: Stability of matter through an electrostatic inequality. Hely. Phys. Acta 70 (1997), no. 1-2, 72-79. [61] G. M. Graf: Stability of matter in classical and quantized fields. Proc. Internat. Congress of Math. (Berlin, 1998). Doc. Math. 1998, Extra Vol. III, pp. 153-162 (electronic). [62] G. M. Graf and D. Schenker: On the molecular limit of Coulomb gases. Comm. Math. Phys. 174 (1995), no. 1, 215-227. [63] G. M. Graf and J. P. Solovej: A correlation estimate with applications to quantum systems with Coulomb interactions. Rev. Math. Phys. 6 (1994), no. 5A, 977-997. [64] C. Hainzl, M. Lewin, and E. Sere: Existence of a stable polarized vacuum in the BogoliubovDirac-Fock approximation. Comm. Math. Phys. 257 (2005), no. 3, 515-562. [65] C. Hainzl, M. Lewin, and J. P. Solovej: Mean-field approximation in quantum electrodynamics. The no-photon case. To appear in Comm. Appl. Math. [66] B. Helffer and D. Robert: Riesz means of bound states and semi-classical limit connected with a Lieb-Thirring conjecture. I, II. Asymptotic Anal. 3 (1990), no. 2, 91-103. Ann. Inst. H. Poincare Phys. Theor. 53 no. 2 (1990), 139-147. [67] W. Hughes: An atomic energy lower bound that agrees with Scott's corrections. Adv. Math. 79 (1990), no. 2, 213-270. [68] D. Hundertmark: On the number of bound states for Schrddinger operators with operatorvalued potentials. Ark. Mat. 40 (2002), 73-87.
[69] D. Hundertmark, A. Laptev, and T. Weidl: New bounds on the Lieb-Thirring constant. Invent. Math. 140 (2000), no. 3, 693-704. [70] D. Hundertmark, E. H. Lieb, and L. E. Thomas: A sharp bound for an eigenvalue moment of the one-dimensional Schrddinger operator. Adv. Theor. Math. Phys. 2 (1998), no. 4, 719-731.
[71] D. Hundertmark, N. Rohrl, and H. Siedentop: The sharp bound on the stability of the relativistic electron-positron field in Hartree-Fock approximation. Comm. Math. Phys. 211 (2000), no. 3, 629-642.
[72] D. Hundertmark and B. Simon: Lieb-Thirring inequalities for Jacobi matrices. J. Approx. Theory 118 (2002), no. 1, 106-130. [73] A. Iantchenko, The electron density in intermediate scales. Comm. Math. Phys. 184 (1997), no. 2, 367-385.
[74] A. lantchenko, E. H. Lieb, and H. Siedentop: Proof of a conjecture about atomic and molecular cores related to Scott's correction. J. Reine Angew. Math. 472 (1996), 177-195. [75] A. Iantchenko and H. Siedentop: Asymptotic behavior of the one-particle density matrix of atoms at distances Z-1 from the nucleus. Math. Z. 236 (2001), no. 4, 787-796.
BOUND STATES IN QUANTUM MECHANICS
493
[76] T. Ichinose: Note on the kinetic energy inequality leading to Lieb's negative ionization upper bound. Lett. Math. Phys. 28 (1993), no. 3, 219-230.
[77] V. Ja. Ivrii and I. M. Sigal: Asymptotics of the ground state energies of large Coulomb systems. Ann. of Math. (2) 138 (1993), no. 2, 243-335. [78] M. Kac: On the asymptotic number of bound states for certain attractive potentials. Topics in Functional Analysis (I. Gohberg and M. Kac, eds.), pp. 159-167. Academic Press, New York, 1978.
[79] J. Keller: Lower bounds and isoperimetric inequalities for eigenvalues of the Schrddinger equation. J. Math. Phys. 2 (1961), 262-266. [80] R. Killip and B. Simon: Sum rules for Jacobi matrices and their applications to spectral theory. Ann. of Math. 158 (2003), 253-321. [81] R. Killip and B. Simon: Sum rules and spectral measures of Schrddinger operators with L2 potentials. To appear in Ann. of Math. [82] A. Kiselev, Y. Last, and B. Simon: Modified Prefer and EFGP transforms and the spectral analysis of one-dimensional Schrddinger operators. Comm. Math. Phys. 194 (1998), 1-45. [83] M. Klaus: On the bound state of Schrddinger operators in one dimension. Ann. Physics 108 (1977), no. 2, 288-300. [84] M. Klaus and B. Simon: Coupling constant thresholds in nonrelativistic quantum mechanics. II. Two-cluster thresholds in N-body systems. Comm. Math. Phys. 78 (1980/81), no. 2, 153168.
[85] S. Kupin: Spectral properties of Jacobi matrices and sum rules of special form. J. Funct. Anal. 227 (2005), no. 1, 1-29. [86] P. Kuchment: Quantum graphs: I. Some basic structures. Waves Random Media 14 (2004), S107-S128.
[87] L. D. Landau and E. M. Lifshitz: Quantum Mechanics. Non-Relativistic Theory. Pergamon Press, London, 1958. [88] A. Laptev: Dirichlet and Neumann eigenvalue problems on domains in Euclidian spaces. J. Funct. Anal. 151 (1997), 531-545. [89] A. Laptev: Private communication. [90] A. Laptev and T. Weidl: Recent results on Lieb-Thirring inequalities. Journees "Equation aux derivees partielles" (La Chapelle sur Erdre, 2000), Exp. No. 10, 14 pp., Univ. Nantes, Nantes, 2000. [91] A. Laptev and T. Weidl: Sharp Lieb-Thirring inequalities in high dimensions. Acta Math. 184 (2000), no. 1, 87-111. [92] D. Levin and M. Z. Solomyak: The Rozenblum-Lieb-Cwikel inequality for Markov generators. J. Anal. Math. 71 (1997), 173-193.
[93] P. Li and S.-T. Yau: On the Schrddinger equation and the eigenvalue problem. Comm. Math. Phys. 88 (1983), 309-318. [94] E. H. Lieb: Bounds on the eigenvalues of the Laplace and Schrddinger operators. Bull. Amer. Math. Soc., 82 (1976), 751-753. See also The number of bound states of one-body Schrddinger operators and the Weyl problem. Geometry of the Laplace Operator (Honolulu, 1979), pp. 241-252, Proc. Sympos. Pure Math., 36, American Mathematical Society, Providence, RI, 1980. [95] E. H. Lieb: The stability of matter. Rev. Modern Phys. 48 (1976), no. 4, 553-569. [96] E. H. Lieb: A lower bound for Coulomb energies. Phys. Lett. 70A (1979), 444-446. [97] E. H. Lieb: Thomas-Fermi and related theories of atoms and molecules. Rev. Modern Phys. 53 (1981), no. 4, 603-641. Erratum: Ibid. 54 (1982), no. 1, 311. [98] E. H. Lieb: Variational principle for many-fermion systems. Phys. Rev. Lett. 46 (1981), no. 7, 457-459. [99] E. H. Lieb: Sharp constants in the Hardy-Littlewood-Sobolev and related inequalities. Ann. of Math. (2) 118 (1983), no. 2, 349-374. [100] E. H. Lieb: On characteristic exponents in turbulence. Comm. Math. Phys. 92 (1984), no. 4, 473-480.
[101] E. H. Lieb: Bound on the maximum negative ionization of atoms and molecules. Phys. Rev. A29 (1984), 3018-3028. [102] E. H. Lieb: Kinetic energy bounds and their application to the stability of matter. Schrddinger Operators (Sonderborg, 1988), pp. 371-382, Lecture Notes in Phys., 345, Springer, Berlin, 1989.
D. HUNDERTMARK
494
[103] E. H. Lieb: The stability of matter: from atoms to stars. Bull. Amer. Math. Soc. (N.S.) 22 (1990), no. 1, 1-49. [104] E. H. Lieb: The stability of matter and quantum electrodynamics. Milan J. Math. 71 (2003), 199-217.
[105] E. H. Lieb: Quantum mechanics, the stability of matter and quantum electrodynamics. Jahresber. Deutsch. Math.-Verein. 106 (2004), no. 3, 93-110. [106] E. H. Lieb and M. Loss: Stability of Coulomb systems with magnetic fields. II. The manyelectron atom and the one-electron molecule. Comm. Math. Phys. 104 (1986), no. 2, 271282.
[107] E. H. Lieb and M. Loss: Stability of a model of relativistic quantum electrodynamics. Comm. Math. Phys. 228 (2002), no. 3, 561-588.
[108] E. H. Lieb and M. Loss: The thermodynamic limit for matter interacting with Coulomb forces and with the quantized electromagnetic field: I. The lower bound. Comm. Math. Phys. 258 (2005), no. 3, 675-695. [109] E. H. Lieb, M. Loss, and H. Siedentop: Stability of relativistic matter via Thomas-Fermi theory. Helv. Phys. Acta 69 (1996), no. 5-6, 974-984. [110] E. H. Lieb, M. Loss, and J. P. Solovej: Stability of matter in magnetic fields. Phys. Rev. Lett. 75 (1995), no. 6, 985-989. [111] E. H. Lieb and S. Oxford: Improved lower bound on the indirect Coulomb energy. Intl. J. Quant. Chem. 19 (1981), 427-439. [112] E. H. Lieb, H. Siedentop, and J. P. Solovej: Stability and instability of relativistic electrons in classical electromagnetic fields. J. Statist. Phys. 89 (1997), no. 1-2, 37-59. [113] E. H. Lieb, H. Siedentop, and J. P. Solovej: Stability of relativistic matter with magnetic fields. Phys. Rev. Lett. 79 (1997), no. 10, 1785-1788. [114] E. H. Lieb, I. M. Sigal, B. Simon, and W. Thirring: Approximate neutrality of large-Z ions. Comm. Math. Phys. 116 (1988), no. 4, 635-644. [115] E. H. Lieb and B. Simon: The Hartree-Fock theory for Coulomb systems. Comm. Math. Phys. 53 (1977), no. 3, 185-194. [116] E. H. Lieb and B. Simon: The Thomas-Fermi theory of atoms, molecules and solids. Advances in Math. 23 (1977), no. 1, 22-116. [117] E. H. Lieb and W. Thirring: Bound for the kinetic energy of fermions which proves the stability of matter. Phys. Rev. Lett. 35 (1975), 687-689. Erratum: Ibid. 35 (1975), 1116. [118] E. H. Lieb and W. Thirring: Inequalities for the moments of the eigenvalues of the Schrodinger Hamiltonian and their relation to Sobolev inequalities. Studies in Mathematical Physics, Essays in Honor of Valentine Bargmann, Princeton University Press, Princeton, NJ, 1976. [119] E. H. Lieb and H.-T. Yau: Many-body stability implies a bound on the fine-structure constant. Phys. Rev. Lett. 61 (1988), no. 15, 1695-1697. [120] E. H. Lieb and H.-T. Yau: The stability and instability of relativistic matter. Comm. Math. Phys. 118 (1988), no. 2, 177-213. [121] P. Mancas, A. M. K. Miiller, and H. Siedentop: The optimal size of the exchange hole and reduction to one-particle Hamiltonians. Theoretical Chemistry Accounts: Theory, Computation, and Modeling (Theoretica Chimica Acta) 111 (2004), no. 1, 49-53. [122] A. Martin: Bound states in the strong coupling limit. Helv. Phys. Acta 45 (1772), 140-148. [123] P. B. Matesanz and J. P. Solovej: On the asymptotic exactness of Thomas-Fermi theory in the thermodynamic limit. Ann. Henri Poincare 1 (2000), no. 2, 281-306. [124] S. Molchanov, M. Novitskii, and B. Vainberg: First KdV integrals and absolutely continuous spectrum for 1-D Schrodinger operator. Comm. Math. Phys. 216 (2001), no. 1, 195-213. [125] Yu. N. Ovchinnikov and I. M. Sigal: Number of bound states of three-body systems and Efimov's effect. Ann. Physics 123 (1979), no. 2, 274-295. [126] N. Pavlovic: Bounds for sums of powers of eigenvalues of Schrodinger operators via the commutation method. Advances in Differential Equations and Mathematical Physics (Birmingham, AL, 2002), pp. 271-281, Contemp. Math., 327, American Mathematical Society, Providence, RI, 2003. [127] G. Pdlya: On the eigenvalues of a vibrating membrane. Proc. London Math. Soc. 11 (1961), 419-433.
[128] M. Reed and B. Simon: Methods of Modern Mathematical Physics IV: Analysis of Operators. Academic Press, New York, 1978.
BOUND STATES IN QUANTUM MECHANICS
495
[129] C. Roepstorff: Path Integral Approach to Quantum Physics. Springer, Berlin, 1994. [130] G. V. Rozenblum: Distribution of the discrete spectrum of singular differential operators. Dokl. AN SSSR 202 (1972), 1012-1015; English translation: Soviet Math. Dokl. 13 (1972), 245-249.
[131] G. V. Rozenblum: Distribution of the discrete spectrum of singular differential operators. Izv. VUZov Matematika 1 (1976), 75-86; English translation: Soviet Math. (Iz. VUZ) 20 (1976), no. 1, 63-71. [132] C. V. Rozenblum and M. Z. Solomyak: The Cwikel-Lieb-Rozenblum estimator for generators of positive semigroups and semigroups dominated by positive semigroups. Algebra i Analiz 9 (1997), no. 6, 214-236; English translation: St. Petersburg Math. J. 9 (1998), no. 6, 1195-1211.
[133] D. Ruelle: Large volume limit of the distribution of characteristic exponents in turbulence. Comm. Math. Phys. 87 (1982), 287-302. [134] M. B. Ruskai: Absence of discrete spectrum in highly negative ions. Comm. Math. Phys. 82 (1981/82), no. 4, 457-469. [135] M. B. Ruskai: Absence of discrete spectrum in highly negative ions. II. Extension to fermions. Comm. Math. Phys. 85 (1982), no. 2, 325-327. [136] A. Rybkin: On the absolutely continuous and negative discrete spectra of Schrodinger operators on the line with locally integrable globally square summable potentials. J. Math. Phys. 45 (2004), no. 4, 1418-1425. [137] A. Rybkin: On the spectral L2 conjecture, 3/2-Lieb-Thirring inequality and distributional potentials. J. Math. Phys. 46 (2005), no. 12, 123505, 8 pp. [138] U.-W. Schmincke: On Schrodinger's factorization method for Sturm-Liouville operator. Proc. Roy. Soc. Edinburgh Sect. A 80 (1978), 67-84. [139] J. Schwinger: On the bound states of a given potential. Proc. Nat. Acad. Sci. U.S.A. 47 (1961), 122-129.
[140] J. M. C. Scott: The binding energy of the Thomas-Fermi atom. Phys. Rev. 43 (1952), 859-867.
[141] L. A. Seco, I. M. Sigal, and J. P. Solovej: Bound on the ionization energy of large atoms. Comm. Math. Phys. 131 (1990), no. 2, 307-315. [142] H. Siedentop and R. Weikard: On the leading energy correction for the statistical model of the atom: interacting case. Comm. Math. Phys. 112 (1987), no. 3, 471-490. [143] H. Siedentop and R. Weikard: Upper bound on the ground state energy of atoms that proves Scott's conjecture. Phys. Lett. A 120 (1987), no. 7, 341-342. [144] H. Siedentop and R. Weikard: On the leading correction of the Thomas-Fermi model: lower bound. Invent. Math. 97 (1989), no. 1, 159-193. [145] I. M. Sigal: Geometric methods in the quantum many-body problem. Nonexistence of very negative ions. Comm. Math. Phys. 85 (1982), no. 2, 309-324. [146] I. M. Sigal: How many electrons can a nucleus bind? Ann. Physics 157 (1984), no. 2, 307-320.
[147] B. Simon: On the infinitude or finiteness of the number of bound states of an N-body quantum system, I. Helv. Phys. Acta 43 (1970), 607-630. [148] B. Simon: Quantum Mechanics for Hamiltonians Defined as Quadratic Forms. Princeton Series in Physics, Princeton University Press, Princeton, NJ, 1971. [149] B. Simon: Analysis with weak trace ideals and the number of bound states of Schrodinger operators. Trans. Amer. Math. Soc. 224 (1976), 367-380 [150] B. Simon: The bound state of weakly coupled Schrodinger operators in one and two dimensions. Ann. Physics 97 (1976), no. 2, 279-288. [151] B. Simon: On the number of bound states of two-body Schrodinger operators: A review. Studies in Mathematical Physics, Essays in Honor of Valentine Bargmann, pp. 305-326, Princeton University Press, Princeton, NJ, 1976. [152] B. Simon: Functional Integration and Quantum Physics. Academic Press, New York, 1977. [153] B. Simon, Geometric methods in multiparticle quantum systems. Comm. Math. Phys. 55 (1977), no. 3, 259-274. [154] B. Simon, N-body scattering in the two-cluster region. Comm. Math. Phys. 58 (1978), no. 2, 205-210.
[155] B. Simon: Schrodinger operators in the twenty-first century. Mathematical Physics 2000, pp. 283-288, Imperial College Press, London, 2000.
D. HUNDERTMARK
496
[156] B. Simon: Trace Ideals and Their Applications. Second edition, Mathematical Surveys and Monographs, 120, American Mathematical Society, Providence, RI, 2005. [157] V. A. Sobolev: The Efimov effect. Discrete spectrum asymptotics. Comm. Math. Phys. 156 (1993), no. 1, 101-126. [158] J. P. Solovej: Asymptotics for bosonic atoms. Lett. Math. Phys. 20 (1990), no. 2, 165-172.
[159] J. P. Solovej: The ionization conjecture in Hartree-Fock theory. Ann. of Math. (2) 158 (2003), no. 2, 509-576. [160] J. P. Solovej and W. Spitzer: A new coherent states approach to semiclassics which gives Scott's correction. Comm. Math. Phys. 241 (2003), no. 2-3, 383-420. [161] H. Tamura: The asymptotic eigenvalue distribution for non-smooth elliptic operators. Proc. Japan Acad. 50 (1974), 19-22. [162] H. Tamura: The Efimov effect of three-body Schrddinger operators: asymptotics for the number of negative eigenvalues. Nagoya Math. J. 130 (1993), 55-83. [163] E. Teller: On the stability of molecules in the Thomas-Fermi theory. Rev. Mod. Physics 34 (1962), 627-631.
[164] R. Temam: Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Second edition, Springer, Berlin, 1997.
[165] W. Thirring: A lower bound with the best possible constant for Coulomb Hamiltonians. Comm. Math. Phys. 79 (1981), no. 1, 1-7. [166] S. A. Vugal'ter and G. M. Zhislin: The spectrum of the Schrddinger operators of multiparticle systems with short-range potentials. (Russian) Trudy Moskov. Mat. Obshch. 49 (1986), 95-112, 239.
[167] T. Weidl: On the Lieb-Thirring constants L.y,l for ry > 1/2. Comm. Math. Phys. 178 (1996), no. 1, 135-146.
[168] H. Weyl: Uber die Abhdngigkeit der Eigenschwingungen einer Membrane von deren Begrenzung. J. Reine Angew. Math. 141 (1912), 1-11. [169] H. Weyl: Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen. Math. Ann. 71 (1912), 441-479. [170] H. Weyl: Uber das Spektrum der Hohlraumstrahlung. J. Reine Angew. Math. 141 (1912), 163-181.
[171] H. Weyl: Ramifications, old and new, of the eigenvalue problem. Bull. Amer. Math. Soc. 56 (1950), 115-139. [172] D. R. Yafaev: On the theory of the discrete spectrum of the three-particle Schrddinger operator. (Russian) Mat. Sb. (N.S.) 94(136) (1974), 567-593, 655-656. [173] V. E. Zakharov and L. D. Faddeev: The Korteweg-de Vries equation is a fully integrable Hamiltonian system. Funkcional. Anal. i Priloaen. 5 (1971), no. 4, 18-27; English translation: Functional Anal. Appl. 5 (1971), 280-287. [174] G.-M. Zhislin: Discussion of spectrum of Schrddinger operators for systems of many particles. (Russian) Trudy Moskov. Mat. Obshch. 9 (1960), 81-120. [175] G.-M. Zhislin: Finiteness of the discrete spectrum in the quantum problem of n particles. Teoret. Mat. Fiz. 21 (1974), 60-73. English translation: Theoret. and Math. Phys. 21 (1974), no. 1, 971-980 (1975).}
SCHOOL OF MATHEMATICS, WATSON BUILDING, UNIVERSITY OF BIRMINGHAM, BIRMINGHAM
B15 2TT, U.K.; ON LEAVE FROM THE DEPARTMENT OF MATHEMATICS, ALTGELD HALL, UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN, IL 61801, U.S.A.
E-mail address: [email protected]
Titles in This Series 76
Fritz Gesztesy (Managing editor), Percy Deift, Cherie Galvez, Peter Perry, and Wilhelm Schlag, Editors, Spectral theory and mathematical physics: A Festschrift in honor of Barry Simon's 60th birthday, Parts 1 and 2 (California Institute of Technology, Pasadena, CA, March 27-31, 2006)
75 Solomon Friedberg (Managing editor), Daniel Bump, Dorian Goldfeld, and Jeffrey Hoffstein, Editors, Multiple Dirichlet series, automorphic forms, and analytic number theory (Bretton Woods, New Hampshire, July 11-14, 2005)
74 Benson Farb, Editor, Problems on mapping class groups and related topics, 2006
73 Mikhail Lyubich and Leon Takhtajan, Editors, Graphs and patterns in mathematics and theoretical physics (Stony Brook University, Stony Brook, NY, June 14-21, 2001)
72 Michel L. Lapidus and Machiel van Frankenhuijsen, Editors, Fractal geometry and applications: A jubilee of Benoit Mandelbrot, Parts 1 and 2 (San Diego, California, 2002 and Ecole Normale Superieure de Lyon, 2001)
71 Gordana Matic and Clint McCrory, Editors, Topology and Geometry of Manifolds (University of Georgia, Athens, Georgia, 2001)
70 Michael D. Fried and Yasutaka Ihara, Editors, Arithmetic fundamental groups and noncommutative algebra (Mathematical Sciences Research Institute, Berkeley, California, 1999)
69 Anatole Katok, Rafael de la Llave, Yakov Pesin, and Howard Weiss, Editors, Smooth ergodic theory and its applications (University of Washington, Seattle, 1999)
68 Robert S. Doran and V. S. Varadarajan, Editors, The mathematical legacy of Harish-Chandra: A celebration of representation theory and harmonic analysis (Baltimore, Maryland, 1998)
67 Wayne Raskind and Charles Weibel, Editors, Algebraic K-theory (University of Washington, Seattle, 1997)
66 Robert S. Doran, Ze-Li Dou, and George T. Gilbert, Editors, Automorphic forms, automorphic representations, and arithmetic (Texas Christian University, Fort Worth, 1996)
65 M. Giaquinta, J. Shatah, and S. R. S. Varadhan, Editors, Differential equations: La Pietra 1996 (Villa La Pietra, Florence, Italy, 1996)
64 G. Ferreyra, R. Gardner, H. Hermes, and H. Sussmann, Editors, Differential geometry and control (University of Colorado, Boulder, 1997)
63 Alejandro Adem, Jon Carlson, Stewart Priddy, and Peter Webb, Editors, Group representations: Cohomology, group actions and topology (University of Washington, Seattle, 1996)
62 JAnos Kollar, Robert Lazarsfeld, and David R. Morrison, Editors, Algebraic geometry-Santa Cruz 1995 (University of California, Santa Cruz, July 1995)
61 T. N. Bailey and A. W. Knapp, Editors, Representation theory and automorphic forms (International Centre for Mathematical Sciences, Edinburgh, Scotland, March 1996)
60 David Jerison, I. M. Singer, and Daniel W. Stroock, Editors, The legacy of Norbert Wiener: A centennial symposium (Massachusetts Institute of Technology, Cambridge, October 1994)
59 William Arveson, Thomas Branson, and Irving Segal, Editors, Quantization, nonlinear partial differential equations, and operator algebra (Massachusetts Institute of Technology, Cambridge, June 1994)
58 Bill Jacob and Alex Rosenberg, Editors, K-theory and algebraic geometry: Connections with quadratic forms and division algebras (University of California, Santa Barbara, July 1992)
57 Michael C. Cranston and Mark A. Pinsky, Editors, Stochastic analysis (Cornell University, Ithaca, July 1993)
56 William J. Haboush and Brian J. Parshall, Editors, Algebraic groups and their generalizations (Pennsylvania State University, University Park, July 1991)
TITLES IN THIS SERIES
55 Uwe Jannsen, Steven L. Kleiman, and Jean-Pierre Serre, Editors, Motives (University of Washington, Seattle, July/August 1991)
54 Robert Greene and S. T. Yau, Editors, Differential geometry (University of California, Los Angeles, July 1990)
53 James A. Carlson, C. Herbert Clemens, and David R. Morrison, Editors, Complex geometry and Lie theory (Sundance, Utah, May 1989)
52 Eric Bedford, John P. D'Angelo, Robert E. Greene, and Steven G. Krantz, Editors, Several complex variables and complex geometry (University of California, Santa Cruz, July 1989)
51 William B. Arveson and Ronald G. Douglas, Editors, Operator theory/operator algebras and applications (University of New Hampshire, July 1988)
50 James Glimm, John Impagliazzo, and Isadore Singer, Editors, The legacy of John von Neumann (Hofstra University, Hempstead, New York, May/June 1988)
49 Robert C. Gunning and Leon Ehrenpreis, Editors, Theta functions - Bowdoin 1987 (Bowdoin College, Brunswick, Maine, July 1987)
48 R. O. Wells, Jr., Editor, The mathematical heritage of Hermann Weyl (Duke University, Durham, May 1987)
47 Paul Fong, Editor, The Arcata conference on representations of finite groups (Humboldt State University, Arcata, California, July 1986)
46 Spencer J. Bloch, Editor, Algebraic geometry - Bowdoin 1985 (Bowdoin College, Brunswick, Maine, July 1985)
45 Felix E. Browder, Editor, Nonlinear functional analysis and its applications (University of California, Berkeley, July 1983)
44 William K. Allard and Frederick J. Almgren, Jr., Editors, Geometric measure theory and the calculus of variations (Humboldt State University, Arcata, California, July/August 1984)
43 Francois Treves, Editor, Pseudodifferential operators and applications (University of Notre Dame, Notre Dame, Indiana, April 1984)
42 Anil Nerode and Richard A. Shore, Editors, Recursion theory (Cornell University, Ithaca, New York, June/July 1982) 41 Yum-Tong Siu, Editor, Complex analysis of several variables (Madison, Wisconsin, April 1982)
40 Peter Orlik, Editor, Singularities (Humboldt State University, Arcata, California, July/August 1981)
39 Felix E. Browder, Editor, The mathematical heritage of Henri Poincare (Indiana University, Bloomington, April 1980)
38 Richard V. Kadison, Editor, Operator algebras and applications (Queens University, Kingston, Ontario, July/August 1980)
37 Bruce Cooperstein and Geoffrey Mason, Editors, The Santa Cruz conference on finite groups (University of California, Santa Cruz, June/July 1979)
36 Robert Osserman and Alan Weinstein, Editors, Geometry of the Laplace operator (University of Hawaii, Honolulu, March 1979)
35 Guido Weiss and Stephen Wainger, Editors, Harmonic analysis in Euclidean spaces (Williams College, Williamstown, Massachusetts, July 1978)
34 D. K. Ray-Chaudhuri, Editor, Relations between combinatorics and other parts of mathematics (Ohio State University, Columbus, March 1978)
'pr a complete list of titles in this series, visit the
AMS Bookstore at www.ams.org/bookstore/.