Volume 7, number 1 CHEMICAL PHYSICS LETTERS 1 October 2970

A QUANTUM MECHANICAL CURVATURE THEOREM

J. N. SILVERMAN* and J. C. :‘AN LEUVEN**

Battelle Institute, Admmed Studies Center, Carouge-Genetur. Switzerluad

Received 30 July 1970

A general quantum mechanical curvature theorem of the form d2E/d$ ,C (9 ld2H/dA2I@) is estab-
lished by means of perturbational-variational theory: here. 9 and E are the exact eigenfunctions and
eigenvalues of the Schr’bdinger time-independent equation, H is the Hamilton operator. and A is
any real parameter occurring in H. The theorem is also es’ablished for arbitrary optimum varia-
tiortil solutions Zo the SchrMinger equation. Several applicaticns of the curvature theorem are dis-
cz-zsed in conjunction with perturbation theory and the stability of approximate uave functions.

1. INTRODUCTION Feynman theorem [l]. Hutley [2] has shown that

the Hellmann-Feynman theorem is also appli-
Consider the time-independent Schradinger cable to properly optimized variaticnzl approxi-
equation mations to J/Jand E, and HaIl[3] has used the
satisfaction of (3) as a criterion of stabiiity of
(I-I-E)@/=O, (I) both exact and approximate wave functions.
where a is the Hamilton operator, and 3/ and It does not appear to have been noted, how-
E are the exact e&e&unctions and eigenvalues, ever, that there exists a generally valid and ex-
respectively. It is often of interest to examine tremely useful second-or&??- theorem. Thus, the
the behaviour of 3/and E as functions of some further differentiation of (3b) in respect to h
real parameter A occurring in the hamiltonian. yields
With this in mind, we write explicitly H=H(X),
IJ.J=+(A), and E= E(A). (When, for example, A can
be treated as a perturbing parameter, this for-
mulation will be recognized as the starting point + (dJ//a 1dH/dx I%+ +-(9 1c!W dk 1d@,;dA) ,
of Rayleigh-SchrMiier (RS) perturbation theo- which is not useful in this form. But, as wilL be
ry.) For normalized +, we then have from eq. shown,
(I),
(44
E(X) = ($(A) IH(X) IWQ) * (2)
Differentiation of eq. (2) in respect to ,Xyields
ci2E/dX2 c (II/] d2iY/cU2 I@) ; (4bl
dE/dh = (IL(cW/&l@)+(d$/dxjHIrCI)
further, it will be shown that this second-order
+ (9 iEF\d$/dh) , +heorem is also applicable to optimum varia-
tional counterparts to I&and E. We teqn (4a) and
but, as is well known, (4b) the curvature theorem.
In this paper we establish the curv&ure theo-
(d~/~lal~}+(J/(~ld~/dh) = 0, (34
rem by perturbational-variational (PV) theory
whence [4], first treating the more general case of op-
timum variational functions, and then extending
dE/dx = (J,IdR/dxIJ/) ; (3b)
the analysis by a simple artifice to the exact
thesefi7sf-mder relationships, (3a) and (3b), solutions of (1); in no case, however, do we
are usually termed collectively the Hellmann- make specific use of (1). A still more general
treatment of the curvature theorem in cormec-
3 Reprint requests should be addressed to this author. tion with the extrema of arbitrary perturbed
** On leave of absence from the Choate School, variational functionak is presented elsewhere
Wallingford, Connecticut, USA. 151. The curvature theorem provides consider-
Volumei’. number1 CHEMICAL PHYSiCS LETTERS 1 October 1970

able insight in a variety of problems, and some tam from (61, (?a), and PV theory,
examples are given here of the type of informa-
tion that can be obtained by ~p~~~ation of the
theorem in the case of a perturbed hamiltonian.

Let $ = Cp(a,>be an arbitrary nokalized an-

alytic ansatz to JI, where the as are any number @a3
and type of adjustable variational parameters.
We specifically take C#as having no explicit de-
pendence on X; this is required for the derivation d+j/dx = aqax = (6 fdwax [ $1 t (3b’I
of (3) and (4) but involves no Ioss in generality of the variational counterpart of (3); this result was
(3, as will snbsequently become evident. From first derived by Hurley [3] from similar con-
the variational prZnciple, which is assumed ap- siderations.
plicable throughout, we have_ We proceed to second order by differentiation
of (8) in respect to X, yielding
JI/$s;X] = (QMsI I&Q I #@St, 3 E , w
where the v?riational energy 7 is an upper bound
to E for arbitrary values of the as. For the
chosen #, the optimum variational quantities
are obtained by soking the customary set of
variztionaf equations
(9)
a7/aas = 0. N>
The salient point to note is that the optimum au apparently intra&ible result. When (9) is
parameters ~2sobtained by solution of (6) are evakated at the optimum C&(X), it is clear from
necessarily explicit functions of _k, (6) that the fir& summation over s vanishes; but
additionaXPV analysis is required to determine
as =5&x> . (7a) the remaining summations. TQ wit, we note that
The superbar here and thro~hout denotes an the partial derivatives ~~~/~~~~, when evaluated
energetically optimized quantity. The other at the &(A), are themselves functions only of A
optimum quantities are then which vanish identically; hence their total h-
derivatives must also vanish identically, imply-
CT= @[~&II , (761
ing
77= n@.$fxM] = (Q ]r-l(x)I@ * UC)
Thus, all optimum variational quantities are at
least implicit functions of X through the opt~um
parameters. Recognition of this X-dependency
is crucial to the development and application of
PV theory [4], and hence to the proof presented
below of (3) and (4); moreover, PV theory goes Es0 l (101
beyond the formal. statement (7a) by providing
a procedure for explicitty calcniating all *the Therefore, at the Z&Q, (9) reduces by (6) and
optimum variational quantities of (7) in the form (LO) to
of pe~rbation expansions analogous to the RS
expansions of the exact quantities.
We now return to (5) and form the total X-
derivative of 9:
Were

0x1 evaluating (8) at the optimum Es(A), we ob-

38
Volume 7. number 1 CHEMICAL PHYSICS LETTERS I October I970

and small h,, the inequality would cIearLy be vio-

lated, and the point i&(&f) would be a maximum
“2”=((#p~ UN
or a saddle point, contrary to supposition. We
conclude that a necessary condition for (16b),
aA2
anci hence for a relative minimum at q[&(Xi) ;Xi],
Further, from (lo), (12a) can be transformed to is

for sufficiently small 5,. To complde the proof,

we evaluate (13) at Xj, multiply it by (A# Ear
arbitrarily small AX, and note Erorn (14) that the
magnitude of the resultant expression on the
r.h.s. of (13) differs from the quadratic form
Q(fi,) defined in (17) only by infinitesimals of
higher order than (8X)2. Thus, from (LL), (12),
(13), and (17), we obtain the necessary condition
for optimum Q and ?j.

(4a’)
where we have introduced the usual operator
notation. whence
We now prove that for arbitrary analytic 4,
the r.h.s. of (13) is non-positive. Let Xi and 3
(4b ‘)
be two closely adjacent but otherwise arbitrary
Values of x with A7t= hi - .; the corresponding
optimum Z_s are Zs(+) an% Zs(3) where which is the variational equivalent of the cur-
vature theorem.
The same analysis appiies to the exact solu-
zse a&j) - ir,(%) = (%), AA+q(Ax)2] . (1-V tions of (1) if we parameterize 9 so as to con-
i
vert it into a variational ansatz. This is done
Compare the energies q[E?&-);hf] and most simply by replaciug h in q with a varia-
T&.(~~);X~]; the former is the optimum, i.e., tional parameter a. Then Q(a) = $(Q),
minimum, energy corresponding to Xi while the q[a;h] =(@(a) II?(A) 1$‘(a)), and it is evident that
latter is not an optimum energy for any A. As is the optimization of 77in respect to a would yield,
well known from the theory of extrema [6], a among other optimum solutions in the event of
necessary and sufficient condition for a relative multiple extrema,
minimum is that
Aq = @s(s(hi) +I$;+] -s[ff&‘&+] ~6 (15) a = ii(a) = a,
for mbihary hs in a sufficiently small neighbor- 6 = @(ii) = $(A) )

hood of +he optimal point &s(hi). Hence, for the i-j = @;A] = E(a) ,
pa?-tictdm Es of (14), it is necessary that
Aij q q[7r,($) +%,;Af] - tl[Zs(+);$] > 0 , (16a) which, together with (3a’) and (3b’), and (4a’)
for sufficient small AA. Treating the & as de- and (4b’), immediately Lead to (3a) and (Sb), and
viations from the optimum ;i,(%), we may ex- (4a) and (4b). This procedure implicitly justifies
pand (Isa) in a Taylor series to obtain the previous remark that an ansatz of the form
@(a,;A) is actually less general than +(a,). For
we can always pass from the former to the latter
by replacing h with another variational param-
eter, whose optimum form woulcJ be either X as
where the inequality must hold for sufficiently in the special case of the exact solution above.
small hs. But if the first term with brackets in or a more complicated function OEK in the gen-
(16b) were negative definite for sufficiently eral case of an arbitrary variationaL solution.

39
 Volume 7. number 1 CHEMICAL PHYSICS LETTERS 1 October 1970

3. DISCUSSION to a truncated perturbed hamiltonian

H=H,+-HIA, (25)
Byers Brown [7] has previously derived the
classical analogue of (4b) for the special case of where H1 is positive definite. Such a hamiltonian
the exact solutions of a microcanonicai ensem- arises in the Hylleraas [lo] treatment of the
ble. The quantum mechanical curvature theorem non-relativistic energy levels of ati N-electron
is also implicit in his treatment but only for the atomic isoelectronic sequence, where Z?l is
exact solutions of the ground state of the sys- taken 23 the interelectronic repuision potential
tems considered; our results are applicable to and A is the inverse nuclear charge. Then, from
both exact and optimum variational solutions for the Hellmann-Feynman theorem, we obtain
all states. 0
We now touch upon several applications of the
curvature theorem in connection with RS and PV
theory. Let k be a real perturbing parameter,
and HA) a general perturbed hamiltonian with
the infinite expansion dij/dA =mgl mijmAmel =($[H1/@ 2 0, (26b)

(18) ~1 =(JlOlH&LOb 0, ‘?I =(6O~H~~&b 0,

(2% 4
From RS perturbation theory*, we have the as- and from the curvature theorem
sociated RS expansions of the exact solutions,
0
00 d2e/dA2 = mc- 2 m(m - 1)~ mXm-2 c 0 , (27a)
E=m~OEmAm* (1% b)

and from PV theory for arbitrary $I, the op- d2ij/dA2=m~2m(?n-I)&~m-2~0, (27b)
timum PV expansions t27c,d)
.
f2 = (3/ojH, ]!bl) ,c 0 , f& = (i& p1> s 0 ;
(20%b) for Category I wave functions, (24) also applies,
Thus, we deduce that for alI states of all sys-
For normalized $o, q, 60, 6, and hermitian Hl, tems described by the II of (25) with positive
we obtain for this general H, definite HI (e.g., all atomic isoelectrcnic se-
quences), the E -X and ?j -A curves have every-
r2 = G&O[a1 1ti) +(IcIoIH2 I ad , (214 where positive slopes, and are everywhere con-
cave downward, the last fact requiring ~2 and ij2
7i2 =<iolH~I~1~+~Qo~H2l~o~, (2W to be non-positive; these predictions are con-
where, without loss of generality, the +,. and qr firmed in all cases where exact and variational
are taken as real [9]. It follows immediately energy data as functions of h are available for
from the curvature theorem that such systems*. Our conclusions are more gen-
eral than tho&e of Stillinger [12] who deduced in
(lclOiHl]Gl) s O, ~2 “(+oiH2\*0) , WJa,b) another manner that the exact solutions E(X) for
non-degenerate ground states have positive
(&iHlI&) d 0, 172 ~&1lH2)60) ; Wa,W slope and negative curvature. It sf;ould be noted
further, for Catsgory I variational wave func- that for negative *finite Hi, the inequalities of
tions [4] where @O=qO, one has ?j2 3 ~2 so that (26) are reversed; on the other hand, the rela-
tions (27), like (22) -(24), are valid for a11HI.
(@OIHI I$$ s ($OlHl j&-l) G 0 l (24) Finally, we discuss the concept of the stabil-
These results hold for all hermitian HI, whether ity [3] of a variational wave function under a
positive definite, negative definite, or indefi- perturbation in terms of the Hellmann-Feyn-
nite. man and curvature theorems. We regard stabil-
It is of interest to specialize the above case
$ For a review of and numerous references to such
l For n review of and extensive references to RS per- energy data for atomic isoelectronic sequences. see
turbation theory, see ref. [S]. ref. [ill.

40
Volume 7, number 1 CHEMICAL PHYslCS LETTER3 1 October 1970

ilty as being synonymous with the requiremeni of (17! vanish identically, one must go to higher-
that the optimum energy q[&(h);X] be at a rela- order derivatives to examine the critical point.
tive minimum for aH h for which it is defined; Thus, if cases (1) and (4) are observed over a
this requirement is by no means automatically wide range oE X, one may conclude with reason--
fulfilled for variationally obtained energies, par- able certainty that the point in question is a
ticularly in cakulations exhibiting multiple ex- minimum.
trema. In the above, we have derived the Hell- These stability criteria are applicable to any
mann-Feynman and curvature theorems by ex- wave function yield@ energy data for seLected
amining the behaviour of arbitrary variational values of the perturbation. Such data may be dif-
n in the vicinity of a supposed minimum. Con- ferenced to obtain the 1.h.s. of eqs. (3b’} and
versely, we now wish to determine if these (4b’), whose r-h-s. are computed in the same
theorems furnish simple criteria for judging manner as other expectation values for the given
the nature of a critical (maximum, minimum, or wave function. Since, however, the accuracy of
saddle) point. Since satisfaction of the Hellmann- this method is limited by the spacing of data, it
Feynman theorem is a necessary condition for is preferable to test stability wfien possible by
establishing the curvature theorem, the use of means of PV perturbation expansions for the ex-
the latter must be preceded b-j verification of pectation values required Optimum PV energy
the former. It is clear, however, that satisfac- expansions automatically satisfy the Heknann-
tion of the Hellmann-Feynman theory, which Feynman theorem order by order in A, and the
guarantees stability in Hall’s sense [3], does not curvature theorem can be tested with PV ex-
guarantee stability in the sense of a minimum, pansions for continuous k within the range of
as it merely proves that ?&(A) is a critical point. validity imposed by their usually finite order,
Satisfaction of the curvature theorem is also on- PV expansions are also easier to obtain in some
ly a necessary con_dition for a minimum, essen- cases than are total expectation values, and may
tially because the ha of (14) are not entirely ar- in principle be adapted to computer cakuIation
bitrary. The most general way of determining for a large number of parameters as easiry as
the nature of the critical point would be to ex- the normal variational method;
pand (15) as in (16b), and to examine the qua-
dratic form
4. CONCLUSIONS

The curvature theorem provides insight into

the behaviour of exact and optimum variational
as a function of X for small arbitrary h,; meth- solutions to the Schrtidinger equation, when
ods of do;- this are described, fcr example, these solutions are treated as functions of a reaL
by Courant [S], and more generally by Birkhoff parameter occurring in the hamiltonian. F’ur-
and lMacLane [13]. Since this approach is es- ther, the theorem serves as a useful cornpIe-
pecially arduous in the case of many param- ment to other relationships, such as the Hell-
eters, it is fortunate that the curvature theorem mann-Feynman and virial theorems, known to
provides a much simpler although less exhaus- be satisfied by the exact solution to (I), and
tive test. By reversing the logic of section 2, which can be satisfied by variational wave func-
the following cases can be distinguished: (1) if tions if properly optimized. Since fulfillment of
the inequality of the curvature theorem, (4a3 or all these theorems is necessary, but not suf-
(4b’), is satisfied for any X, a maximum is im- ficient, for exact solution of the eigenvalue
mediately excluded, and the critical point is problem, they furnish additional criteria to that
either a minimum for all X or a saddle point; of the variational principle for judging the qual-
(2) analogously, if the curvature theorem is ity and stability of approximate wave functions.
violated for any A, a minimum is excluded, and Finally, it should be noted that the curvature
the point is either a maximum for all X or a theorem applies analogously to the expectation
saddle point; (3) if both cases (1) and (2) occur value of any operator other than the hamiltonian
for different X, the point is a saddle point; (4) if which contains a real parameter X, and which is
the emliiy of the curvature theorem is also subject to a minimum principle, such as the
satisfied for some h in cases (l), (2), and (3), statistical variance; if a maximum principIe
the above conclusions remain valid; and (5) if should apply, the inequality of the curvature
the equality is satisfied trivially for all A, as_ theorem would be reversed [5].
wouId occur when Q(h,) of (28), and hence Q(h,)
Vohme 7, number 1 CHEMICAL PHYsrcs LETTERS 1 October 1970

’ REFERENCES [7] W. Byers Brown, Proc. Cambridge Phil. Sot.

54 (1958) 251.
[l] IX.Heilmann. Einfuhrung in die Quantenchamie [S] J. ti. Hirschfelder, W. Byers Brown and S. T. Ep-
(Deutteke, Vienna, 1937); stein, Advan. Quantum Chem. 1 (l964) 255.
RP.Feynman, Phys. Rev. 56 (1939) 340: [9 ] II. U. Con&m and G. II. Shortley. The theory of
J,I.Musher. Am. J. Phys. 34 (1966) 261. atomic spectra {Cambridge Univ. Press, Cam-
(21 A. C. Hurley. Pros. Roy. Sot. A226 (1954) 179. bridge; England, 1957) p. 33:
f3] G.G.Hatl, PhiI. &lag. 6 (1961) 249. S.T.Epstemn. Am. J. Phys. 22 (1954) 613.
[4] J. N. Silverman and J. C. van Leuven, Phys. Rev. [lo] E.A.Hylteraas. 2. Physik 65 (1930) 209.
162 (1967) 1175. [ll] J.N. Silvarmsn and G. II. Brigman, Rev. Mod.
[S] ksf;;an Leuven and J.N.Silverman, to be pub- Phys. 39 (1967) 228.
[12] F_H.Stiltinger Jr.. J. Chem. Phys. 45 (l966) 3623.
[S] R. Co&ant, Differential and integral calculus, [13] G. Birkhoff and 5. MaoLane. A survey of modern
vol. II (Blackie, London, 1947) p. 294. algebra (IZacmiiIan, New York, 1958) p. 271.