Duality in Perturbation Theory and the Quantum Adiabatic

Approximation

Marco Frasca
Via Erasmo Gattamelata, 3, 00176 Roma (Italy)
(September 18, 2018)

Abstract
arXiv:hep-th/9801069v3 23 Jul 1998

Duality is considered for the perturbation theory by deriving, given a series

solution in a small parameter, its dual series with the development parameter
being the inverse of the other. A dual symmetry in perturbation theory is
identified. It is then shown that the dual to the Dyson series in quantum
mechanics is given by a recently devised series having the adiabatic approx-
imation as leading order. A simple application of this result is given by
rederiving a theorem for strongly perturbed quantum systems.
PACS: 03.65.Bz, 02.90.+p

Typeset using REVTEX

1
 A known result of fluid mechanics [1], given the Navier-Stokes equation
in absence of any forcing
∂u
+ (u · ∇)u − ν∇2 u + ∇p = 0 (1)
∂t
where ν is the viscosity, p the pressure and u the velocity field, is that, by
fixing the 0-th order solution through the Eulerian part as [2]
∂u0
+ (u0 · ∇)u0 = 0, (2)
∂t
one obtains a perturbation series for a large Reynolds number Re while, taking
at the leading order the equation
∂w0
− ν∇2 w0 = 0 (3)
∂t
one obtains a perturbation series for small Re. So far, this was the only case
in perturbation theory, as applied to physics, with an equation generating
both a small perturbation series and its strong perturbation counterpart.
Duality in perturbation theory should then be understood in the sense
that, for a given differential equation, one has the possibility to derive pertur-
bation series both in a small development parameter and its inverse, giving
in this way the possibility to study a solution in different regions of the pa-
rameter space. However, it should be said that the method could not be of
absolutely general usefulness as some limitations can appear for the compu-
tation both of the leading and higher orders. Beside, there exists situations
where better approximations are known, as I will show. However, it is easy
to realize that a lot of problems in physics can get new insights from this
approach, so that, it is worthwhile to be exploited.
A natural question, in the light of the above defined duality in perturba-
tion theory, is what should be the dual to the well-known Dyson series for the
Schrödinger equation. The answer to this question is the main aim of this
communication. In fact, the existence of this possibility gives a new technique
to analyse quantum systems in different regions of their parameter space.
Quite recently, I showed that new solutions for the Schrödinger equation,
in time-dependent problems, can be obtained when a strong perturbation is
applied to a quantum system [3]. This approach seems to indicate that, at the
leading order, an adiabatic approximation should be used [4]. On a different
line of research, Mostafazadeh [5] was able to show that a series exists, for
the Schrödinger equation, with a well defined development parameter, having
the adiabatic approximation as leading order. Using duality in perturbation
theory the above different research lines can be merged, as I am going to
show, giving us the main result of this paper. In fact, the series derived
by Mostafazadeh is dual to the Dyson series. Then, the theory of strong
perturbations in quantum mechanics can be proved to be dual to the standard
small perturbation theory.
In order to show how a dual series can be obtained in a simple case, let
us consider the model given by the following differential equation

2
 ẍ = f0 (x) + λf1 (x) (4)

where the dots mean derivation with respect to the time and λ is an ordering
parameter. It is a well-known matter that, when λ → 0, a solution series
of the form x ∼ x0 + λx1 + λ2 x2 + O(λ3 ) can be obtained. However, as for
the Navier-Stokes equations, we are free to choose at the leading order, as an
unperturbed equation, ẍ0 = λf1 (x0 ). To show that this choice√ gives a dual
perturbation series, I rescale the time variable in eq.(4) as t → λt = τ . One
gets

λẍ = f0 (x) + λf1 (x) (5)

where now the dots mean derivation with respect to τ . It is quite easy to
verify that the series
1 1 1
 
x = x0 + x1 + 2 x2 + O (6)
λ λ λ3
is a solution of eq.(5) when

x¨0 = f1 (x0 )
x¨1 = f1′ (x0 )x1 + f0 (x0 ) (7)
1
x¨2 = f0′ (x0 )x1 + f1′ (x0 )x2 + f2′′ (x0 )x21
2
..
.

By analogy with the results in quantum mechanics [3], I take the above as
the dual method to small perturbation theory to obtain a dual perturbation
series to a given one. It is important to note that the above result is true
independently of one’s ability to solve the leading order equations.
We see that the arbitrariness in the choice of the leading order equation
gives rise to a symmetry. In fact, putting λ = 1 into eq.(4), there is no more
reason to see any difference between the perturbation and the unperturbed
system in the same way as happens in fluid mechanics. This means that the
series given by the small perturbation theory can be derived from the one given
by the dual method and vice versa by simply interchanging f0 (x) ↔ f1 (x).
That is a symmetry of the perturbation theory whose meaning can be really
understood only after the introduction of the dual series. Actually, the general
solution of eq.(4) can be written, for the one-dimensional case, as
Z x 1
t − t0 = dx′ √ q (8)
2 E + x0 f0 (x′′ )dx′′ + xx0 f1 (x′′ )dx′′
R x′ R ′
x0

with E a motion constant. It is easily seen that both the series expansions,
for small f0 or f1 , can be straightforwardly obtained. What it is interesting is
that, the small parameter in a case is the inverse of the development parameter
in the other. From the discussion above it should be clear that both series
can have the same problems as secularities or divergent terms.

3
 A more interesting example is given by the following Duffing equation

ẍ + ω02 x + βx3 = f0 cos(ωt). (9)

ω ω02 βf02
By setting τ = ω0 t, ν = ω0 , ξ = f0 x and λ = ω06
, one gets the rescaled
equation

ξ̈ + ξ + λξ 3 = cos(ντ ). (10)

where the dots mean derivation with respect to τ and λ is just a parameter
measuring the strength of the nonlinearity. Eq.(10) is generally considered,
analytically, only for small λ, but what happens in the limit of a very strong
nonlinearity? Duality can be applied and one easily realizes that, for large
values of the parameter λ, the quantity ǫ = 12 ξ̇ 2 + λ 14 ξ 4 tends to be a constant
of motion. This is due to the result that the perturbation completely drives
the system. So, we have regular periodic motion in the considered limit. This
example shows that, although the leading order equation can be easy to solve,
going to higher orders could be very involved.
A class of important problems arises from the Schrödinger equation that
I consider in the one-dimensional form
h̄2 d2 ψ
− + V0 (x)ψ + λV1 (x)ψ = Eψ (11)
2m dx2
where λ → ∞. One could apply immediately the symmetry between the dual
and small perturbation theories discussed so far and use without difficulty
the Rayleigh-Schrödinger approximation scheme. While that is a correct ap-
proach,
√ I will show how the dual
 method works in this case.
  So, let us put
1 1 1
ξ = λx, ψ = ψ0 + λ ψ1 + O λ2 and E = λE0 + E1 + O λ . This yields
the following equations

h̄2 d2 ψ0
− + V1 (ǫξ)ψ0 = E0 ψ0
2m dξ 2
h̄2 d2 ψ1
− + V1 (ǫξ)ψ1 + V0 ψ0 = E1 ψ0 + E0 ψ1 (12)
2m dξ 2
..
.

where ǫ = √1λ . At the leading order we get a well-known equation, that is, a
second order differential equation with a slowly varying coefficient due to the
perturbation. In this case we can apply the WKB approximation [7]. Thus,
the dual method yields in this case a solution that is a combination of both
Rayleigh-Schrödinger and semiclassical methods.
There are several problems where the above approximation can be applied.
A well-known example is given by the anharmonic oscillator that has a large
body of literature [8] and is a model which any approximation scheme should
address. The Hamiltonian can be cast in the form
p2 1 2 λ 4
H= + q + q . (13)
2 2 4

4
The method I discussed so far gives an unambigous answer to this problem,
i.e. the leading order approximation, when the anharmonicity is very strong,
can be obtained by solving the equation
!
p2 λ 4
+ q ψ0 (q) = Eψ0 (q). (14)
2 4

The quartic oscillator is well-known in the literature [9]. So, we can compare
our method with numerical results. To leading order of the WKB approxima-
 q  32
h̄2 λ
tion of the energy levels, normalized in unit of 2 4 , the agreement is
within 18% with the true value of the ground state energy for the anharmonic
oscillator. That agreement improves for higher excited states. However, we
know from Symanzick scaling that the quartic oscillator is the right approxi-
mation for energy levels of the anharmonic oscillator when λ → ∞ [10]. Often,
the use of semiclassical eigenfunctions can be too much involved and better
approximation schemes, as those given in ref. [8], can improve the situation.
Duality, as applied in perturbation theory, yields anyway a definite answer.
The situation is surely more interesting in time-dependent problems. By
noting that the only meaningful quantities are transition probabilities between
states of the unperturbed system, we have the initial conditions definitely fixed
breaking in this way the dual symmetry of the perturbation theory. In fact,
in ref. [3] I showed that the problem

(H0 + λV (t))|ψi = ih̄∂t |ψi (15)

∂
with ∂t = and λ → ∞, using the above dual method, has the leading order
∂t
solution
1
 
|ψi ∼ U (t)|ni + O (16)
λ
with

λV (t)U (t) = ih̄∂t U (t) (17)

and H0 |ni = En |ni. So, the unperturbed solution fixes the initial condition
as also happens in the small perturbation theory. But, in order to leave the
dual simmetry untouched, one should physically consider also systems ini-
tially prepared with the eigenstates of the perturbation, but this is not the
case for the computation of probability transitions. Then, we can conclude
that, for the time-dependent perturbation theory as usually applied in quan-
tum mechanics, the dual symmetry is broken due to the choice of the initial
conditions.
By the methods discussed above, we can obtain the main result of the
paper. Our aim is to show that the dual to the Dyson perturbation series, for
the time-dependent Schrödinger equation, is given by the series obtained by
Mostafazadeh [5] having the adiabatic approximation as leading order. So,

5
let us consider the Schrödinger equation H(t)U = ih̄∂t U being U the time
evolution operator. The Dyson series is the solution of that equation and can
be written formally as

i
Z t 
U (t) = T exp − dt′ H(t′ ) (18)
h̄ 0
with T the time-ordering operator. This is a compact writing for the series
development
t 2 Z t t′
i i
Z  Z
′ ′ ′
U (t) = I − dt H(t ) + − dt dt′′ H(t′ )H(t′′ ) + · · · (19)
h̄ 0 h̄ 0 0
A dual series to the one above is meant as a series having for develop-
ment parameter its inverse, as discussed above. To obtain it, let us con-
sider the case with the hamiltonian H being constant in time. Assuming,
for the sake of simplicity, here and in the following that the hamiltonian
has a discrete spectrum, the solution to the time-dependent Schrödinger
equation is easily obtained through the time evolution operator U (t) =
P − i En t
n e h̄ |nihn| with H|ni = En |ni, where the effect of the time deriva-
P i − i En t
tive is simply ∂t U (t) = n − h̄ En e h̄ |nihn|. Instead, for a time-
dependent hamiltonian H(t), in the case of the adiabatic approximation we
have UA (t) = n eiαn (t) |n, tihn, 0| being αn (t) = γn (t) − h̄1 0t dt′ En (t′ ) with
P R

H(t)|n, ti = En (t)|n, ti, so that En (t) gives the dynamical part of the phase
αn (t) and γn (t) = i 0t dt′ hn, t′ |∂t′ |n, t′ i the geometrical part. It is natural to
R

ask how one can define a derivative Dt to obtain the same result as for the
time-independent case, that is, Dt UA (t) = n − h̄i En (t) eiαn (t) |n, tihn, 0|.
P

It is quite easy to verify that the following definition of Dt has the requested
property
X
Dt = ∂t + i hm, t|i∂t |n, ti|m, tihn, t| (20)
n,m
n6=m

so that, the adiabatic approximation is exact for the equation

H(t)UA (t) = ih̄Dt UA (t) (21)
and it is easily verified that iDt |k, ti = γ̇k (t)|k, ti. In this way, we are a step
away from the sought result. In fact, let us introduce a generic perturbation
V (t) into eq.(21), so that
(H(t) + V (t))U (t) = ih̄Dt U (t). (22)
In this form the duality principle of perturbation theory can be ap-
plied. Having V (t) as the unperturbed hamiltonian and H(t) as the per-
turbation, we get no physical meaningful results from the leading order
equation VX(t)U (0) (t) = ih̄Dt U (0) (t) unless we choose V (t) = H(t) or
V (t) = − hm, t|ih̄∂t |n, ti|m, tihn, t|. The latter is the interesting case giv-
n,m
n6=m
ing trivially the standard Dyson series. So, when we take H(t) as the un-
perturbed hamiltonian and V (t) as the perturbation, we have at the leading
order

6
 H(t)U (0) (t) = ih̄Dt U (0) (t), (23)

but this is nothing else than eq.(21) and then U (0) (t) = UA (t) (i.e. at the
leading order we have the adiabatic approximation). To complete the identifi-
cation with the Mostafazadeh result, we have that the higher order corrections
are computed solving the equation

H ′ (t)U ′ (t) = ih̄∂t U ′ (t), (24)

where

H ′ (t) = UA† (t)V (t)UA (t) = − e−i(αm (t)−αn (t)) hm, t|ih̄∂t |n, ti|m, 0ihn, 0|
X
(25)
n,m,n6=m

as given in ref. [5]. Then, one obtains [5]


i
Z t 
U (t) = UA (t)T exp − dt′ H ′ (t′ ) (26)
h̄ 0

that completes the proof. No adiabatic hypothesis entered into this argument
as it should be.
As an application of that result, a theorem recently derived by me and,
in a rigourous way but in a different context, by Joye [4] can be obtained for
the theory of the strong perturbations in quantum mechanics [3]. In fact, it
was proved that, for a quantum system described by the Schrödinger equation
(H0 + λV (t))|ψi = ih̄∂t |ψi, in the limit λ → ∞, the adiabatic approximation,
using the eigenstates of the perturbation V (t), is a good approximation for
|ψi. So, by considering the perturbed system in the interaction picture, gives
the hamiltonian HI (t) = U (0)† (t)λV (t)U (0) (t) where H0 U (0) (t) = ih̄∂t U (0) (t).
Then, the result obtained for the Dyson series by duality can be directly
applied to HI (t). We obtain for a small λ (otherwise we miss convergence),
the Dyson series
t 2 Z t t′
i i
Z  Z
′ ′ ′
UI (t) = I − dt HI (t ) + − dt dt′′ HI (t′ )HI (t′′ ) + · · · (27)
h̄ 0 h̄ 0 0

and, for a large λ, the Mostafazadeh result applied to HI (t) having at the
leading order the adiabatic approximation as it should be. It must be noticed
that, in the latter case, the eigenstates to be considered are those of the
perturbation. In fact, we have U (0)† (t)V (t)U (0) (t)|k, tiI = vkI (t)|k, tiI but
this is equivalent to V (t)(U (0) (t)|k, tiI ) = vkI (t)(U (0) (t)|k, tiI ). Then, vkI is
an eigenvalue of V (t) and U (0) (t)|k, tiI differs just by a time-dependent phase
factor from the corresponding eigenstate of the perturbation. As a by product,
we get the confirmation that higher order corrections are those computed by
the method of strong perturbation theory as given in [4].
In summary I introduced the duality principle in perturbation theory for
differential equations. A dual method with respect to the theory of small
perturbation is yielded and a dual symmetry between the two methods arises
from the freedom in the choice of what the perturbation is. The use of dual-
ity shows that, for the time-dependent Schrödinger equation, the dual series

7
to the Dyson one is given by a perturbation series computed recently by
Mostafazadeh, with a well-defined development parameter, having the adia-
batic approximation as leading order. This enriches the possibility to analyze
quantum systems in completely different regions of their parameter space.

8
 REFERENCES
[1] D.Pnueli, C.Gutfinger, Fluid Mechanics, (Cambridge University Press, Cambridge, 1992);
V.L’vov, I.Procaccia, “Lectures Notes of the Les Houches 1994 Summer School”, chao-
dyn/9502010 Los Alamos Preprint Archive (1995)
[2] Euler equations have not second order derivatives differently from the Navier-Stokes equa-
tions: This originates the well-known boundary layer problem in perturbation theory. See i.e.
J.Kevorkian, J.D.Cole, Perturbation Methods in Applied Mathematics, (Springer-Verlag, New
York, 1985)
[3] M.Frasca, Phys. Rev. A 43, 45 (1992); 47, 2374 (1993); 56, 1548 (1997).
[4] M.Frasca, hep-th/9406075 and hep-th/9406119 (LANL preprint archive) unpublished; Nuovo
Cimento 111B, 957 (1996); 112B, 1073 (1997); A.Joye, Ann. Inst. H. Poincare’ 63, 231
(1995).
[5] A.Mostafazadeh, Phys. Rev. A 55, 1653 (1997)
[6] I.S.Gradshteyn, I.M.Ryzhik, Table of Integrals, Series, and Products, (Academic Press, Lon-
don, 1980)
[7] A.H.Nayfeh, Introduction to Perturbation Techniques, (Wiley, New York, 1981), Ch.14.
[8] I do not aim to cite all the work about this fundamental problem of quantum mechanics.
From the earlier to the latest I just cite C.M.Bender, T.T.Wu, Phys. Rev. 184, 1231 (1969);
Phys. Rev. D 7, 1620 (1973); J.Zinn-Justin, Phys. Rep. 70, 109 (1981) and references therein;
W.Janke, H.Kleinert, Phys. Rev. Lett. 75, 2787 (1995). However, this is a very active field of
physics for its implications in quantum field theory.
[9] G.Parisi, Statistical Field Theory,(Addison-Wesley, 1988, New York) pg.264ff
[10] B.Simon, Ann. Phys. (N.Y.) 58, 76 (1970)