Home Search Collections Journals About Contact us My IOPscience

Schrödinger equation from an exact uncertainty principle

This article has been downloaded from IOPscience. Please scroll down to see the full text article.

2002 J. Phys. A: Math. Gen. 35 3289

(http://iopscience.iop.org/0305-4470/35/14/310)

View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 128.42.202.150
The article was downloaded on 09/06/2013 at 07:57

Please note that terms and conditions apply.

INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS A: MATHEMATICAL AND GENERAL
J. Phys. A: Math. Gen. 35 (2002) 3289–3303 PII: S0305-4470(02)31884-5

Schrödinger equation from an exact uncertainty

principle
Michael J W Hall1 and Marcel Reginatto2,3
1Theoretical Physics, IAS, Australian National University, Canberra, ACT 0200, Australia
2US Department of Energy, Environmental Measurements Laboratory, New York,
NY 10014-4811, USA

Received 10 December 2001

Published 29 March 2002
Online at stacks.iop.org/JPhysA/35/3289

Abstract
An exact uncertainty principle, formulated as the assumption that a classical
ensemble is subject to random momentum fluctuations of a strength which is
determined by and scales inversely with uncertainty in position, leads from the
classical equations of motion to the Schrödinger equation.

PACS number: 03.65.Ta

1. Introduction

The uncertainty principle is generally considered to be a fundamental conceptual tool for

understanding differences between classical and quantum mechanics. As first argued by
Heisenberg in 1927 [1], the fact that quantum states do not admit simultaneously precise
values of conjugate observables, such as position and momentum, does not necessarily imply
an incompleteness of the theory, but rather is consistent with not being able to simultaneously
determine such observables experimentally to an arbitrary accuracy.
Corresponding uncertainty relations such as
x
p
h̄/2 ‘give us that measure of
freedom from the limitations of classical concepts which is necessary for a consistent
description of atomic processes’ [2]. The uncertainty principle provides the basis of the
Copenhagen interpretation of quantum mechanics, famously used by Bohr in defending the
completeness of the theory against critics such as Einstein [3].
If regarded as merely asserting a physical limit on the degree to which classical concepts
can be applied, the uncertainty principle is not sufficiently restrictive in content to supply
a means for moving from classical mechanics to quantum mechanics. Thus Landau and
Lifschitz write that ‘this principle in itself does not suffice as a basis on which to construct
a new mechanics of particles’ [4]. In particular, uncertainty relations expressed as imprecise
inequalities are not enough to pin down the essence of what is nonclassical about quantum
mechanics. Authors have tended to point instead, for example, to the commutation relation
3 Present address: Physikalisch-Technische Bundesanstalt, Bundesallee 100, 38116 Braunschweig, Germany.

0305-4470/02/143289+15$30.00 © 2002 IOP Publishing Ltd Printed in the UK 3289

3290 M J W Hall and M Reginatto

[x̂, p̂] = ih̄ for quantum observables [5], or to the principle of superposition of wavefunctions
[4, 6], in this regard.
However, it will be shown here that an exact form of the uncertainty principle may in
fact be formulated, which provides the single key element in moving from the equations of
motion of a classical ensemble to those of a quantum ensemble. In particular, if it is assumed
that a classical ensemble is subject to random momentum fluctuations, where the strength of
these fluctuations is precisely determined by and scales inversely with uncertainty in position
(as characterized by the position probability density), then the resulting modified equations
of motion are equivalent to the Schrödinger equation. Thus, surprisingly, there is an exact
formulation of the uncertainty principle which does in fact capture the essence of what is
‘quantum’ about quantum mechanics.
In the following section we recall the description of a classical ensemble in terms of a
pair of equations in configuration space (the Hamilton–Jacobi equation and the continuity
equation), and provide the corresponding Lagrangian from which these equations follow.
In section 3 we show that the above exact uncertainty principle leads to a modification
of this Lagrangian (essentially incorporating the kinetic energy of the random momentum
fluctuations), the form of which yields equations of motion equivalent to the Schrödinger
equation. Further, an exact uncertainty relation for position and momentum uncertainties is
derived, corresponding to the exact uncertainty principle, from which the usual Heisenberg
inequality follows as a consequence.
Of course, equations of motion equivalent to the Schrödinger equation do not in themselves
imply the full quantum formalism. Accordingly, in section 4 a Hamiltonian formulation
is provided for the equations of motion, which leads naturally to the usual wavefunction
representation as corresponding to the normal modes of the modified system. Conclusions are
presented in section 5.

2. Classical mechanics

For simplicity, we limit ourselves to the case of a single particle, described in a configuration
space of n dimensions. In the Hamilton–Jacobi formulation of classical mechanics, the
equation of motion takes the form [7]
∂S 1
+ ∇S · ∇S + V = 0. (1)
∂t 2m
The velocity field u(x, t) that describes the motion of the particle is related to the momentum
potential S(x, t) by
1
u= ∇S. (2)
m
We assume that the initial conditions are not known exactly, and that the probability of
finding the particle in a given volume of the configuration space is described by a probability
density P (x, t). The probability density must satisfy the following two conditions: it must be
normalized,


P dn x = 1

and it must satisfy a continuity equation,

 
∂P 1
+ ∇ · P ∇S = 0. (3)
∂t m
Schrödinger equation from an exact uncertainty principle 3291

Equations (1) and (3), together with (2), completely determine the motion of the classical
ensemble. Equations (1) and (3) can be derived from the Lagrangian

 
∂S 1
LC = P + ∇S · ∇S + V dn x dt (4)
∂t 2m
by fixed end-point variation (δP = δS = 0 at the boundaries) with respect to S and P.

3. The transition from classical mechanics to quantum mechanics

3.1. Momentum fluctuations

Consider now the possibility that the classical Lagrangian is not quite right, because ∇S is
actually an average momentum: one also has a fluctuation N about ∇S. Thus the physical
momentum is
p = ∇S + N .
No particular underlying physical model will be assumed for the momentum fluctuation N .
Indeed, one could instead regard the fluctuations as fundamentally nonanalysable, being
introduced as a simple device to remove the notion of individual particle trajectories (since u
in equation (2) is no longer ascertainable).
Since the momentum fluctuation N may conceivably depend on position, the average over
such fluctuations for a given quantity A at point x will be denoted by Ā, while the average
over fluctuations
 and position will be denoted by A. One hence has the general relation
A = PĀ dn x. A physically very reasonable randomness assumption for the momentum
fluctuation N is that it vanishes on average everywhere, i.e. N̄ ≡ 0. However, here only two
weaker assumptions will be made:
N  = 0 ∇S · N  = 0. (5)
The first of these states that the fluctuations are unbiased and the second that the fluctuations
are linearly uncorrelated with the average momentum ∇S.
 1 It follows
 that when the momentum fluctuations are significant, the kinetic energy term
2m ∇S·∇S in the Lagrangian should be replaced by 1
2m (∇S + N ) · (∇S + N ) , yielding the
modified Lagrangian

 
∂S 1
L= P + (∇S + N ) · (∇S + N ) + V dn x dt

 ∂t 2m 

∂S 1 1
= P + ∇S · ∇S + V dn x dt + P N · N dn x dt
∂t
2m 2m
1
≡ LC + (
N)2 dt
2m (6)
where
N is the average rms momentum fluctuation, given by N · N 1/2 . Thus the
consequence of taking into consideration the momentum fluctuations is to add a positive
term to the Lagrangian, arising from the additional kinetic energy due to the fluctuations.

3.2. Exact uncertainty principle

How can we estimate the magnitude of this additional term, if we do not know anything
else about the system except the probability density P and the average momentum ∇S? To
estimate the magnitude of the momentum spread, we will assume that an exact uncertainty
principle holds, in the sense that the strength of the momentum fluctuations at a given time is
3292 M J W Hall and M Reginatto

inversely correlated with uncertainty in position at that time, where the uncertainty of position
is characterized by P. Clearly, this assumption is an additional hypothesis that is independent
of classical mechanics.
To make this assumption precise, consider the general case of an n-dimensional space and
a one-parameter family of probability distributions (which we label with a parameter k > 0)
at time t0 , related by a rescaling of variables
P (x) → Pk (x) ≡ k n P (kx).
These transformations preserve the normalization,




P (x) d x → k P (kx) d x = P (y) dn y
n n n

where we have introduced the change of variables y = kx. We also have

∇P (x) · ∇P (x) → k 2n+2 ∇y P (y) · ∇y P (y)

x · ∇P (x) → k n y · ∇y P (y).

Under such a transformation, any direct measurement of position uncertainty δx such as

the rms uncertainty
x changes according to the rule
1
δx → δxk ≡ (δx).
k
Thus probability densities with different values of k represent physical systems that only differ
in how well we know the location of the particle, since the shape of the probability densities is
the same except for the rescaling. The exact uncertainty principle that we want to make use of
corresponds roughly to the assumption that such a scaling of position by a factor 1/k scales
the momentum fluctuation by a factor k.
More precisely, the exact uncertainty principle is equivalent to the statement that the
momentum fluctuation
N is determined by the uncertainty in position, where the latter is
characterized by the probability density P, and where

N → k
N (7)
under k transformations. Note that the uncertainty product δx
N is thus preserved under k
transformations, for any direct measure of position uncertainty δx.
To apply the exact uncertainty principle,we assume that the Lagrangian formalism remains
applicable to L. Hence the additional term in equation (6) must be a spacetime integral over
a scalar function of x, P and S and their derivatives. Moreover, since (
N)2 is determined
solely by position uncertainty (where the latter is characterized by P), then this additional term
is in fact independent of S. Finally, for causality to be preserved (i.e. the equations of motion
require only P and S to be specified on an initial surface), second and higher order derivatives
of P must be excluded 4. Hence, the additional term in the Lagrangian (6) can be written in
the form

(
N)2 dt = Pf (x, P , x · ∇P , ∇P · ∇P ) dn x dt. (8)

4 Requirements of causality do not exclude a term linear in ∇ 2 P , but since it can be shown that such a term does not
lead to a different result we will not consider it here.
Schrödinger equation from an exact uncertainty principle 3293

The exact uncertainty principle requires f to transform under k transformations as follows:

P (x)f [x, P (x), x · ∇P (x), ∇P · ∇P (x)] dn x dt

→ P (y)f k −1 y, k n P (y), k n y · ∇y P (y), k 2n+2 ∇y P (y) · ∇y P (y) dn y dt

≡ k 2 P (y)f (y, P (y), y · ∇y P (y), ∇y P (y) · ∇y P (y)) dn y dt.

This leads to the homogeneity condition

f (k −1 x, k n u, k n v, k 2n+2 w) = k 2 f (x, u, v, w) (9)
where we have introduced the more compact notation
u=P v = x · ∇P w = ∇P · ∇P . (10)
From this requirement we derive the first-order partial differential equation
n
∂f ∂f ∂f ∂f
− xi + nu + nv + (2n + 2)w = 2f. (11)
i=1
∂x i ∂u ∂v ∂w
The problem of finding the general integral of such an equation is equivalent to the problem
of finding the general integral of a system of ordinary differential equations [8], which in our
case is given by
dx1 dxn du dv dw df
− = ··· = − = = = = .
x1 xn nu nv (2n + 2)w 2f
This system of ordinary differential equations has (n + 3) independent integrals, which can be
chosen as

u−1 w1/2 xi = const

u−1 v = const
u2/n x · x = const
u2 w−1 f = const.
and the solution of the first-order partial differential equation (11) is then of the general form
f = (u−2 w)g(u−1 w1/2 x, u−1 v, u2/n x · x) (12)
where g is an arbitrary function.

3.3. Independent subsystems

To determine f completely, we need to fix the form of g in equation (12). We therefore
introduce a natural independence condition, by the requirement that the Lagrangian L
decomposes into additive subsystem contributions for the case of a system composed of
independent subsystems. This is equivalent to the condition that the momentum fluctuations
N1 and N2 are linearly uncorrelated for two such subsystems, and hence can equivalently be
interpreted as a further randomness assumption for the momentum fluctuations.
To investigate the requirements imposed on f by the independence condition, it will be
sufficient to consider the case where we have a system consisting of two uncorrelated particles
of mass m that do not interact, one particle described by a set of coordinates x1 and the other
by x2 . Thus P is of the form
P (x1 , x2 ) = P1 (x1 )P2 (x2 ) (13)
3294 M J W Hall and M Reginatto

and hence from equations (6) and (8) the independence condition requires
Pf = P1 P2 (f1 + f2 ) (14)
where subscripts 1 and 2 refer to subsystems 1 and 2, respectively.
Equation (13) immediately implies the relations
u = u1 u2
v ≡ u−1 v = u−1 −1
1 v1 + u2 v2 = v1 + v2
−2 −2
w ≡ u w = u1 w1 + u2 w2 = w1 + w2
−2

and hence equation (12) becomes



f = (w1 + w2 )g w1 + w2 x, v1 + v2 , (u1 u2 )2/n x · x

where x = (x1 , x2 ). From equation (14), this form of f must decompose into the sum of
a function of u1 , v1 , w1 and x1 , and a function of u2 , v2 , w2 and x2 . Since the factor that
multiplies g, and the second and third arguments of g, are such functions (with respect to w ,
v , and x respectively), these terms cannot be mixed by the functional form of g. It follows
that g must be of the form
g (a, b, c) = C + g0 (a) + bg1 (a) + cg2 (a)
where C is a constant, and the functions gj satisfy the condition
gj (λa) = λ−2 gj (a) j = 0, 1, 2 (15)
(to allow cancellation of the factor w1 + w2 that multiplies g). Hence
f = C(w1 + w2 ) + g0 (x1 , x2 ) + (v1 + v2 )g1 (x1 , x2 )
+ (u1 u2 )2/n (x1 · x1 + x2 · x2 )g2 (x1 , x2 ).
The independence condition equation (14) places further conditions on the gj . First, g0
is required to be a sum of a function of x1 and a function of x2 . Hence it only represents a
classical additive potential term (that satisfies the homogeneity condition (15) above), and so
can be ignored as having no nonclassical role (it can be absorbed into the classical potential
V in the Lagrangian). Secondly, to avoid subsystem cross terms, g1 must be constant. But the
homogeneity condition (15) is then only satisfied by the choice g1 = 0. Thirdly, cross terms
in u1 and u2 can only be avoided by choosing g2 = 0.
The form of f thus reduces to the first term, C(w1 + w2 ). From equation (14) this term
is to be identified with the sum of f1 and f2 , thus yielding the final form
1
f = Cw = C 2 ∇P · ∇P (16)
P
for f , where C is a universal constant. Note from equation (8) that C must be positive.

3.4. Equations of motion

The modified Lagrangian follows from equations (6), (8) and (16) as

 
∂S 1 C 1
L= P + ∇S · ∇S + ∇P · ∇P + V dn x dt. (17)
∂t 2m 2m P 2
Fixed end-point variation with respect to S leads again to (3), while fixed end-point variation
with respect to P leads to

∂S 1 C 1 2 2
+ ∇S · ∇S + ∇P · ∇P − ∇ P + V = 0. (18)
∂t 2m 2m P 2 P
Schrödinger equation from an exact uncertainty principle 3295

Equations (3) and (18) are identical to the Schrödinger equation

ih̄∂ψ/∂t = −(h̄2 /2m)∇ 2 ψ + V ψ
provided the wavefunction ψ(x, t) is written in terms of S and P by
 
√ S
ψ = P exp i
h̄
and the constant C is set equal to
 2
h̄
C= .
2
Just why one would introduce the wavefunction ψ at all is considered in section 4 below.
Note that the classical limit of the Schrödinger theory is not the Hamilton–Jacobi equation
for a classical particle, but equations (1) and (3) which describe a classical ensemble.

3.5. Exact uncertainty relation

The Schrödinger equation has been derived above using an exact uncertainty principle to
fix the strength of random momentum fluctuation in terms of the uncertainty in position.
Note that no specific measure of position uncertainty was assumed; it was required only that
the momentum fluctuations scale inversely with position uncertainty under k transformations.
However, having obtained a unique form, equation (16), for the function f in (8) we are now
in a position to write down an exact uncertainty relation relating position and momentum
uncertainties.
For simplicity we consider the case of one dimension (n = 1), and define

  −1/2
1 dP 2
δx = P dx .
P dx
For the case of a Gaussian probability density with rms uncertainty σ one has δx = σ .
More generally, this measure has units of position, scales appropriately with x (δy = λδx for
y = λx) and vanishes in the limit that P approaches a delta function. Hence it represents a
direct measure of uncertainty in position. From equations (8) and (16) one has
√ h̄
δx
N = C = . (19)
2
Thus we have an exact uncertainty relation between position and momentum. A quantum-
operator form of this relation has been derived elsewhere, in which
N is replaced by the rms
deviation of a nonclassical momentum operator [9].
The usual Heisenberg uncertainty relation can be derived from the above exact uncertainty
relation. From the Cramer–Rao inequality of statistical estimation theory [10] one has

x
δx, while the randomness assumptions in equation (5) imply
(
p)2 = Var(dS/dx + N) = Var(dS/dx) + (
N)2
(
N)2
and hence it follows immediately from equation (19) that
x
p
h̄/2.

4. Hamiltonian formulation and wavefunction representation

4.1. Hamiltonian formulation

In the previous section, we derived an extension of the classical Lagrangian which yields
equations of motion equivalent to the Schrödinger equation. The Lagrangian field formalism
3296 M J W Hall and M Reginatto

was conveniently used because it is well known. However, one can in fact obtain equivalent
results using the Hamiltonian form of field theory, with no essential differences in the
assumptions and manipulations used.
The Hamiltonian formalism does provide one important advantage: the concept of
canonical
√ transformations.

In the previous section, the wavefunction representation ψ =
P exp i h̄S was simply ‘magicked out of thin air’, to obtain the Schrödinger equation written
in terms of the wavefunction ψ instead of the hydrodynamical variables P and S. In contrast,
in the Hamiltonian formalism this complex combination of P and S arises immediately from
asking a natural question about canonical transformations.
The Hamiltonian form corresponding to Lagrangian (17) is given by

 

1 h̄2 1
H = P ∇S · ∇S + ∇P · ∇P + V d x ≡ H dn x
n
(20)
2m 8m P 2
and is derived similarly. The field P plays the role of a field coordinate and S the role of the
momentum canonically conjugate to P. The equations of motion are given by [7]
∂P δH
= {P , H } =
∂t δS
∂S δH
= {S, H } = −
∂t δP
where the Poisson bracket of two functions F and G is defined by


δF (x) δG(x ) δF (x) δG(x ) n
{F (P (x), S(x)), G(P (x ), S(x ))} = − d x . (21)
δP (x ) δS(x ) δS(x ) δP (x )
To simplify the formulae, we will sometimes use the notation P ≡ P (x), P ≡ P (x ), etc
which allows us to write equation (21) in the concise form


δF δG δF δG n
{F, G } = − d x .
δP δS δS δP
From
δP δS
= = δ n (x − x ) (22)
δP δS
we derive the Poisson bracket of the canonically conjugate fields as
{P , S } = δ n (x − x ).
The equations of motion that correspond to H are
 
∂P δH 1
= = −∇ · P ∇S
∂t δS m
  
∂S δH 1 h̄2 1 2 2
=− =− ∇S · ∇S + ∇P · ∇P − ∇ P + V .
∂t δP 2m 8m P 2 P
These equations are of course identical to (3) and (18) which were derived using the Lagrangian
formalism.

4.2. Wavefunctions and normal modes

The Hamiltonian form H has been expressed in (20) in terms of fields which represent important
physical quantities: P has the physical interpretation of a position probability density and S
that of an average momentum potential. However, H can be rewritten in terms of any pair of
Schrödinger equation from an exact uncertainty principle 3297

fields φ and χ without changing the physical content provided they are related to P and S by
a canonical transformation, i.e.


δP δS δP δS n
{P , S } = − d x = {φ, χ }. (23)
δφ δχ δχ δφ
Of course, such a transformation is generally only of interest if the new fields have some
particular physical significance.
One transformation of obvious physical interest, when it exists, is to two fields φ and
χ which have uncoupled equations of motion. Such fields label two independent physical
degrees of freedom in the system, and hence have fundamental physical significance as the
‘normal modes’ of the system. It is therefore natural to ask whether such a transformation
exists for H, i.e. whether there is a one–one mapping
P = P (φ, χ) S = S(φ, χ)
such that the fields φ and χ are uncoupled. √
It will be
seen

that this question is sufficient to
derive the wavefunction representation ψ = P exp i h̄S and its complex conjugate from the
Hamiltonian H, as corresponding to the physical fields describing the ‘normal modes’ of the
system.
To examine the question of whether there is a canonical transformation that will lead to
uncoupled equations of motion for φ and χ we first need to establish the following lemma.

Lemma. A necessary condition for two conjugate fields φ and χ to be uncoupled is that the
corresponding Hamiltonian density H has the form
H = F (x, φ, χ) + Ak (x, φ, χ)∂k φ + Bk (x, φ, χ)∂k χ + Gj k (x, φ, χ)(∂j φ)(∂k χ)
where k = 1, . . . , n, repeated indices are summed over, and ∂k denotes the partial derivative
with respect to xk . Furthermore, the symmetric part of Gj k is independent of φ and χ, i.e.
Gj k (x, φ, χ) + Gkj (x, φ, χ) = 2Gj k (x)
where Gj k (x) is symmetric with respect to j and k.

Proof. For a Hamiltonian

H = H (x, φ, χ, ∂k φ, ∂k χ) dn x

the equations of motion are given by

∂φ ∂ H ∂ 2 H ∂ 2 H ∂ 2 H
= − − ∂k φ − ∂k χ
∂t ∂χ ∂xk (∂k χ) (∂φ) ∂(∂k χ) (∂χ) ∂(∂k χ)
∂ 2 H ∂ 2 H
− ∂kl φ − ∂kl χ.
∂(∂l φ) ∂(∂k χ) ∂(∂l χ) ∂(∂k χ)

A similar expression is obtained for ∂χ

∂t
. Since we assume that φ evolves independently of χ,
then in particular no second derivatives of χ can appear in the above equation of motion for φ,
and similarly, no second derivatives of φ can appear in the corresponding equation of motion
for χ. Hence H must be linear in both ∂k φ and ∂k χ, and so H has the general form given in
the statement of the lemma. Substituting this form into the equation of motion for φ gives
 
∂φ ∂F ∂Ak ∂Bk ∂Gj k
= − ∂k Bk + − − ∂j Gkj ∂k φ − (∂j φ)(∂k φ) − Gj k ∂kl φ
∂t ∂χ ∂χ ∂φ ∂φ
3298 M J W Hall and M Reginatto

and a similar equation for ∂χ

∂t . Hence, since Gj k is the coefficient of ∂kl φ and ∂kl χ in the
respective equations of motion, the fields are uncoupled only if the symmetric part of Gj k is
independent of both φ and χ. 

If we now express the Hamiltonian density H (20) in terms of the new variables, we find
            
1 ∂S 2 h̄ 2 ∂ln P 2 1 ∂S 2 h̄ 2 ∂ln P 2
H=P + ∇φ · ∇φ + +
2m ∂φ 2 ∂φ 2m ∂χ 2 ∂χ
  2  
1 ∂S ∂S h̄ ∂ln P ∂ln P
× ∇χ · ∇χ + + ∇φ · ∇χ + V .
m ∂φ ∂χ 2 ∂φ ∂χ
It follows immediately from the lemma that the first two terms of H must vanish, implying
∂S h̄ ∂ln P ∂S h̄ ∂ln P
= iα = iβ (24)
∂φ 2 ∂φ ∂χ 2 ∂χ
where α, β = ±1. Since P and S are real, the normal modes are therefore complex fields.
Substituting equations (24) into the third term of H, the lemma further implies that
 2  
h̄ 1 − αβ ∂ln P ∂ln P
Gj k (x) = P δj k = Gδj k
2 m ∂φ ∂χ
where G is a constant (the last equality follows since the second term has no explicit x
dependence). Now, if α = β, the Hamiltonian density reduces to H = P (φ, χ)V , and the
inverse transformation to the fields P and S then yields a Hamiltonian density proportional to
V , which is inconsistent with the form of H. Therefore
 2 
h̄ 1 ∂ln P ∂ln P
α = −β 2P = G. (25)
2 m ∂φ ∂χ
From equations (24) and (25) the Hamiltonian density H in terms of φ and χ is
H = G∇φ · ∇χ + P (φ, χ)V
and the equations of motion take the simple form
∂φ ∂P
= V − G∇ 2 φ
∂t ∂χ
∂χ ∂P
− = V − G∇ 2 χ.
∂t ∂φ
For these equations to be uncoupled P must be of the form P = W + Xφ + Y χ + Zφχ, which
when substituted into (25) yields
 2
2 m
P = G (φ + K) (χ + L) (26)
h̄ 2
where K and L are constants (related to X, Y and Z ). The general form of S(φ, χ) is found by
substitution of (26) in (24) with α = −β, which leads to a pair of differential equations with
solution
h̄ φ + K
S = b + iα ln (27)
2 χ +L
where b is an arbitrary complex constant.
Schrödinger equation from an exact uncertainty principle 3299

Equations (26) and (27) establish the functional forms of P (φ, χ) and S(φ, χ) that will
permit uncoupled equations of motion for φ and χ. We now have to check that these functional
forms lead to a canonical transformation. This requires


δP δS δP δS n
{P , S } = − d x
δφ δχ  δχ δφ
2 n
= −mG iα δ (x − x )
h̄
= δ n (x − x )
and thus the value of G is fixed to be
G = iαh̄/(2m).
Recalling that P and S are real (and using the property that P is positive), one can show that
the inverse transformation follows from (26) and (27) as
√
φ = a P exp(−iαS/h̄) − K (28)
αh̄ √
χ= P exp(iαS/h̄) − L (29)
ia
where a is an arbitrary complex constant (related to b).
Thus, an essentially unique canonical transformation to uncoupled fields φ and χ indeed
exists, given by equations (28) and (29). We recognize
√that these

fields are, up to a scale factor
and additive constant, the usual wavefunction ψ = P exp i h̄S and its complex conjugate,
and hence the wavefunction has a fundamental physical significance as a ‘normal mode’ of
the system. The corresponding Hamiltonian density H follows as
iαh̄ iα
H= ∇φ · ∇χ + V (φ + K) (χ + L)
2m h̄
h̄2
= |∇ψ| + V |ψ|2
2
2m
for all choices of a, K and L, and leads directly to the Schrödinger equation and its conjugate.
We point out a quantization condition that follows from equations (27), (28) and (29).
If
 4(x) is a single-valued complex function, ln 4 is a multi-valued function which satisfies
C d ln 4 = ±i2πn, where n is an integer. If we make the assumption that the fields φ and χ
describing the ‘normal modes’ of the system are single-valued functions, then
 
h̄
dS = −iα d(ln ψ − ln ψ ∗ ) = ±2πh̄n.
C 2 C
This is precisely the quantization condition that was introduced by Takabayasi as ‘a new
postulate’ in his hydrodynamic interpretation of quantum mechanics [11]. This subsidiary
condition is of course compatible with the equations of motion.

4.3. Expectation values

While the wavefunction and the corresponding wave equation have been obtained, these do
not represent the full quantum formalism. For example, the nature of the assumptions about
momentum fluctuations in section 3.1 provides recipes for calculating the first two moments
of the momentum distribution in terms of integrals that can now be expressed in terms of the
wavefunction, since
3300 M J W Hall and M Reginatto

h̄
p = P ∇S dn x = ψ ∗ ∇ψ dn x
i

  2 

h̄ 1 n
p  = P ∇S · ∇S +
2
∇P · ∇P d x = h̄2
|∇ψ|2 dn x.
2 P2

However, it is not immediately clear how within this framework higher order moments are
to be calculated, nor expectation values of functions of position and momentum. We briefly
note here a possible approach to this problem, based on a symmetry in the representation
of position and momentum displacements, which leads to the usual relations assumed in the
Hilbert space formulation of quantum mechanics.
Under a position displacement Ta : x → x + a, the fields P and S transform as
P (x) → P (x − a), S(x) → S(x − a). Hence in the wavefunction representation one has
Ta : ψ(x) → ψ(x − a). (30)
Under a momentum displacement Mq : p → p + q, the position distribution (which is
given by the field P) should be unaffected, while the average momentum must change
by q, ∇S → ∇S + q. Therefore, the fields P and S transform as P (x) → P (x),
S(x) → S(x) + q · x (where an arbitrary additive constant added to S has been ignored,
as it has no effect on the equations of motion). Hence in the wavefunction representation one
has
Mq : ψ(x) → exp(iq · x/h̄)ψ(x). (31)
Comparing (30) and (31), one recognizes that the transformations Ta and Mq are Fourier-
pairs. In particular, if one defines the Fourier transform of ψ(x) by


1
ϕ( p) = ψ(x) exp(ix · p/σ ) dn x
(2πσ )−n/2
where σ is a constant with units of action, then one has

Ta : ϕ( p) → exp(ia · p/σ )ϕ( p) Mq : ϕ( p) → ϕ( p − σ q/h̄).

Comparing with equations (30) and (31), there is a direct symmetry between ψ and ϕ under
position and momentum translations, provided one sets σ = h̄.
In light of this symmetry, it is natural to postulate that, in analogy to P (x) = |ψ(x)|2 ,
the momentum probability density is given by P̃ (p) = |ϕ(p)|2 . Under this postulate one
finds that


 
h̄
f ( p) = P̃ ( p)f ( p) dn p = ψ ∗ (x)f ∇ ψ(x) dn x
i
which then leads to the natural generalization

 
h̄
f (x, p) = ψ (x)f x, ∇ ψ(x) dn x
∗
(32)
i
as per standard quantum theory (where in general an operator ordering must be specified for
the expectation value to be well defined).
Finally, we point out another approach that is also natural within this framework. Since
the equations of motion in the variables ψ and ψ ∗ are linear, it is natural to investigate the
Schrödinger equation from an exact uncertainty principle 3301

group of canonical transformations that preserve the linearity of these equations. This leads
us to consider the group of functional transformations of the form


ρ(y) = K(x, y)ψ(x) dn x


ρ (y) = K ∗ (x, y)ψ ∗ (x) dn x.
∗

Using (22), a simple calculation leads to the following condition for the transformation to be
canonical:


∗ δρ δρ ∗ δρ δρ ∗ n
{ρ, ρ } = − d x
δψ δψ ∗ δψ ∗ δψ


= K(x , y)K ∗ (x , y ) dn x = δ(y − y ).

This is the condition for a transformation to be unitary. Arguments similar to those discussed
above can then be used to single out the choice
1
K(x, y) = exp(ix · y/h̄)
(2πh̄)−n/2
which corresponds to the transformation that leads to the momentum space representation.

5. Conclusions

We have shown that an exact uncertainty principle, formulated in the form that the strength
of the momentum fluctuations is inversely correlated with the uncertainty in position, leads
from the classical equations of motion to the Schrödinger equation. The assumptions that we
used for this fall into three main categories: maximal randomness (equations (5) and (14)), an
exact uncertainty principle (equation (7)) and causality (equation (8)).
The additional term in the Lagrangian is essentially the Fisher information, originally
introduced by Fisher [12] as a measure of ‘intrinsic accuracy’ in statistical estimation theory.
This Fisher information term was derived using an information theoretical approach in [13].
The connection between Fisher information and quantum mechanics has been developed
further in [9], where it was shown that the Fisher information is proportional to the difference
of a classical and quantum variance (thus providing a measure of nonclassicality) and to the
rate of entropy increase under Gaussian diffusion (thus providing a measure of robustness).
The operator formulation of the exact uncertainty relation in equation (19) is studied in detail
in [14] including its extension to entangled systems. We point out that in all of these references
the Fisher information is defined in the usual way, that is, as a functional of the probability
distribution—and therefore, one should not confuse it with the quantity by the same name
that appears in Frieden [15], which is essentially a generalized Fisher information defined for
wavefunctions and proportional to the quantum kinetic energy.
It is worth noting that the approach here, based on exact uncertainty, is rather different from
other approaches which assign physical meaning to fields P and S related to the wavefunction.
For example, in the de Broglie–Bohm approach [16], there are no momentum fluctuations
and the classical equations of motion for P and S are instead modified by adding a mass-
dependent ‘quantum potential’, Q, to the classical potential term in the Hamilton–Jacobi
equation. The form of this quantum potential is left unexplained and is interpreted as
arising from the influence of an associated wave acting on the system. Similarly, while
Bohm and Vigier generalize Bohm’s original formalism to permit fluctuations of momentum
3302 M J W Hall and M Reginatto

about ∇S, this is merely to ensure that an ensemble of such particles will quickly evolve
to have a stable distribution given by the modulus-squared of the associated wave [17]. In
contrast, in the exact uncertainty approach ∇S is an average momentum, the form of an
additional kinetic energy term arising from random momentum fluctuations is derived, and
no associated wave is assumed. The formal connection between the two approaches is the
relation


δ(L − LC ) = dt dn xQδP .

The exact uncertainty approach is also very different from the stochastic mechanics
approach [18]. The latter postulates the existence of a classical stochastic process in
configuration space, with a drift velocity assumed to be the gradient of some scalar, and
defines an associated time-symmetric ‘mean acceleration’ a in terms of averages over both
the stochastic process and a corresponding time-reversed process, which is postulated to obey
Newton’s law ma = −∇V . In contrast, the exact uncertainty approach does not rely on a
classical model of fluctuations, nor on a new definition of acceleration, nor on properties of
stochastic processes running backwards in time. Indeed, as remarked in section 3.1, one may
view the introduction of N as a means of effectively eliminating the notion of trajectories,
differentiable or otherwise, from the classical hydrodynamical formulation of section 2. The
formal connections between the approaches are
∇S = mu (
N)2 = m2 v · v
where u + v and u − v are the drift velocities of the forward-in-time and backward-in-time
processes, respectively. It should be noted that u · v = 0, and hence, noting equation (5),
one cannot identify mv with the random momentum fluctuation N .
In [13] it was suggested that the Fisher information term represented an ‘epistemological’
contribution to the action, which in the context of the present analysis can be interpreted
as reflecting a lack of detailed knowledge of nonclassical momentum fluctuations. In our
approach we do not attempt to provide a ‘realistic’ model of such fluctuations, which
would at any rate require a whole new (and nonlocal) theory that goes beyond quantum
mechanics. Our approach to understanding quantum mechanics is therefore different from
other descriptions based on the postulate of an underlying stochastic process, such as stochastic
mechanics [18]. What our analysis primarily offers is a new way of viewing the uncertainty
principle as the key concept in quantum mechanics. While it is true that no one before
quantum mechanics would think of taking an uncertainty principle as a fundamental principle,
our analysis is valuable in that it enforces the importance of the uncertainty principle in
distinguishing quantum mechanics from classical mechanics—in a sense, it says that the
uncertainty principle is the fundamental element that is needed for the transition to quantum
mechanics.

References

[1] Heisenberg W 1927 Z. Phys. 43 172

[2] Heisenberg W 1930 The Physical Principles of the Quantum Theory (New York: Dover) p 4
[3] Bohr N 1958 Atomic Physics and Human Knowledge (New York: Wiley) pp 32–66
[4] Landau L D and Lifschitz E M 1977 Quantum Mechanics 3rd edn (Oxford: Pergamon) p 2
[5] Born M and Jordan P 1925 Z. Phys. 34 858
Born M, Heisenberg W and Jordan P 1926 Z. Phys. 35 557
[6] Dirac P A M 1958 The Principles of Quantum Mechanics 4th edn (Oxford: Clarendon)
[7] Corben H C and Stehle P 1994 Classical Mechanics 2nd edn (New York: Dover)
[8] Schuh J F 1968 Mathematical Tools for Modern Physics (Eindehoven: Philips Technical Library)
Schrödinger equation from an exact uncertainty principle 3303

[9] Hall M J W 2000 Phys. Rev. A 62 012107

[10] Cox D R and Hinkley D V 1974 Theoretical Statistics (London: Chapman and Hall) ch 8
[11] Takabayasi T 1952 Prog. Theor. Phys. 8 143
Takabayasi T 1953 Prog. Theor. Phys. 9 187
[12] Fisher R A 1925 Proc. Camb. Phil. Soc. 22 700
[13] Reginatto M 1998 Phys. Rev. A 58 1775
[14] Hall M J W 2001 Phys. Rev. A 64 052103
[15] Frieden B R 1999 Physics from Fisher Information (Cambridge: Cambridge University Press)
[16] Bohm D 1952a Phys. Rev. 85 166
Bohm D 1952b Phys. Rev. 85 180
Bohm D and Hiley B J 1993 The Undivided Universe: An Ontological Interpretation of Quantum Theory
(London: Routledge)
[17] Bohm D and Vigier J-P 1954 Phys. Rev. 96 208
[18] Nelson E 1966 Phys. Rev. 150 1079
Nelson E 1967 Dynamical Theories of Brownian Motion (Princeton, NJ: Princeton University Press)