Annals of Physics 315 (2005) 505531

www.elsevier.com/locate/aop

Computing the wavefunction from

trajectories: particle and wave pictures
in quantum mechanics and their relation
Peter Holland
Green College, University of Oxford, Woodstock Road, Oxford OX2 6HG, England, United Kingdom
Received 25 May 2004; accepted 16 September 2004
Available online 16 December 2004

Abstract
We describe the particle method in quantum mechanics which provides an exact scheme to
calculate the time-dependent wavefunction from a single-valued continuum of determinstic
trajectories where two spacetime points are linked by at most a single orbit. A natural language for the theory is oered by the hydrodynamic analogy, in which wave mechanics corresponds to the Eulerian picture and the particle theory to the Lagrangian picture. The
Lagrangian model for the quantum uid may be developed from a variational principle.
The EulerLagrange equations imply a fourth-order nonlinear partial dierential equation
to calculate the trajectories of the uid particles as functions of their initial coordinates using
as input the initial wavefunction. The admissible solutions are those consistent with quasi-potential ow. The eect of the superposition principle is represented via a nonclassical force on
each particle. The wavefunction is computed via the standard map between the Lagrangian
coordinates and the Eulerian elds, which provides the analogue in this model of Huygens
principle in wave mechanics. The method is illustrated by calculating the time-dependence
of a free Gaussian wavefunction. The Eulerian and Lagrangian pictures are complementary
descriptions of a quantum process in that they have associated Hamiltonian formulations that
are connected by a canonical transformation. The de BroglieBohm interpretation, which employs the same set of trajectories, should not be conated with the Lagrangian version of the
hydrodynamic interpretation. The theory implies that the mathematical results of the de

E-mail address: peter.holland@green.ox.ac.uk.

0003-4916/$ - see front matter
2004 Published by Elsevier Inc.
doi:10.1016/j.aop.2004.09.008

506

P. Holland / Annals of Physics 315 (2005) 505531

BroglieBohm model may be regarded as statements about quantum mechanics itself rather
than about its interpretation.

2004 Published by Elsevier Inc.
PACS: 03.65.Bz
Keywords: Quantum mechanics; Hydrodynamics; Particle trajectories; Variational principle; Canonical
transformation

1. Introduction
The notion that the concept of a continuous material orbit is incompatible with a
full wave theory of microphysical systems was central to the genesis of wave mechanics [1]. Early attempts to justify this assertion using Heisenbergs relations were subsequently shown to be awed, and indeed no credible proof forbidding the treatment
of quantum processes in terms of precisely dened spacetime trajectories has ever
been oered. The idea nevertheless entered the folklore of the subject and even
now is invoked to highlight alleged paradoxical implications of quantum mechanics
(e.g., Schrodingers cat). So great was the philosophical bias that not only was the
material orbit ruled out as an aid to comprehension, but the possibilities of using
the trajectory as a computational tool, or even as the basis of an alternative representation of the quantum theorythe twin subjects of this paperwere foregone.
In the path-integral method the path concept is integral to the computation of the
wavefunction: the propagator linking two spacetime points is calculated by linearly
superposing the elementary amplitudes associated with all the paths connecting the
points [2]. The wavefunction at a point is then found from Huygens principle by
superposing the contributions coming from all the other points (weighted by the initial wavefunction). The technique gives the impression that the trajectory concept
works in this context as a computational device only because it is allied with a simultaneous application of the superposition principle to the path. Thus, the propagation
of a particle somehow involves the simultaneous traversal of multiple equally likely
paths. The method therefore rather reinforces the view that a description of the
propagation between two points using a single spacetime track is incompatible with
the quantum description of the process.
Of course, de Broglie and Bohm demonstrated long ago the falseness of this conclusionno rule of quantum theory is broken by employing a single trajectory to
specify the state of a system more nely than does the wavefunction [3,4]. But, being
just an interpretative element rather than an intrinsic ingredient of quantum theory,
it has been possible to dismiss the de BroglieBohm trajectory as a superuous
ideological superstructure [5]. Indeed, the very method of computing the trajectory, which relies on rst knowing the wavefunction whose determination is independent of the path, appears to conrm the de BroglieBohm trajectory as a redundant
addendum.
The question whether the concept of a single path linking two spacetime points,
such as that employed by de Broglie and Bohm, can be made the basis of an alternative computational scheme in quantum mechanics has received scant attention in

P. Holland / Annals of Physics 315 (2005) 505531

507

the foundations of physics literature. Indeed, it appears to be unknown that the seeds
of an alternative view of the de BroglieBohm trajectories were sown more than 30
years ago when it was shown that they could be instrumental in solving numerically
the time-dependent Schrodinger equation [69]. Rather than being merely interpretative, the trajectories acquired a use-value. Further work was done on this particle
(or trajectory) method [10,11] and recently it has ourished, principally in the
chemical physics literature, with several signicant developments in the technique
being reported (for a selection of references see [1233]).
The approach arises from the well-known analogy between quantum mechanics
and hydrodynamics. The similarity in form between the single-body Schrodinger
equation in the position representation (wave mechanics), suitably expressed as
two real equations, and the equations of uid mechanics was rst pointed out by
Madelung [34]. In this analogy, the probability density is proportional to the uid
density and the phase of the wavefunction is a velocity potential. A novel feature
of the quantum uid is the appearance of quantum stresses, usually represented
through the quantum potential. To achieve full mathematical equivalence of the
models, the hydrodynamic variables have to satisfy conditions inherited from the
wavefunction. These in turn provide physical insight into the original conditions.
For example, the single-valuedness requirement on the wavefunction corresponds
to the appearance of quantized vortices in the uid [3537].
The analogy prompts the following observation. For a classical uid one can
adopt the Eulerian picture, in which one observes the ow from xed space points
and to which the Schrodinger uid-like equations correspond in the analogy just
mentioned, or the Lagrangian picture, in which one follows individual uid particles. The idea explored in the numerical work cited above is to adapt elements of
the Lagrangian method of the classical theory to quantum hydrodynamics as a
means of solving the wave equation. For example, the uid may be approximated
by a nite collection of representative particles and their equations of motion
solved simultaneously with those of the phase and amplitude functions to generate
expressions for the wave function along each trajectory. These techniques generally
have advantages over methods employing just the Eulerian equations for the computational eort is naturally concentrated in regions where the density of trajectories is highest.
The aim of this paper is to elaborate on the general theory of the Lagrangian
method in quantum mechanics to bring out its full signicance. The Lagrangian
model for the quantum uid may be developed systematically from (variational) rst
principles, in much the way one proceeds in classical hydrodynamics (Section 2). The
EulerLagrange equations provide an exact formula (a version of Newtons second
law) to calculate the trajectories of the uid particles as functions of their initial coordinates using as input the initial wavefunction. The latter data restricts the admissible solutions to those consistent with quasi-potential ow (Section 3). Using these
solutions we obtain a quantum analogue of the classical construction of the time-dependent Eulerian uid functions purely from the uid particle trajectories (Section
4). The key step in the computation is to express the initial coordinates as functions
of space and time. One thereby achieves the same end as the path-integral method

508

P. Holland / Annals of Physics 315 (2005) 505531

computation of the wavefunction given its initial valueusing a quite dierent and
conceptually simpler trajectory method in which two spacetime points are connected
by at most a single continuous orbit. The two steps in Feynmans approach (propagator, then Huygens) are thus condensed into one, the wavefunction as a whole
being generated from its initial form by a single-valued continuum of trajectories.
The superposition principle enters at the level of the path not in the sense that many
dierent routes through space are available to a particle starting from one point but
indirectly in that on the unique path that the particle takes between two points it is
subject to nonclassical eects due to an additional quantum term in Newtons force
law. However, unlike the path-integral method this is not so far a method of quantization since the functional form of the uid internal potential energy, the origin of
the quantum eects, is unexplained.
The fact that we can derive the time-dependent wavefunction from the trajectories
(an example is given in Section 5) indicates that we are dealing here not just with an
interpretation (hydrodynamic), but rather with an alternative mathematical representation, or picture, of quantum mechanics. The hydrodynamic analogy is a particularly suggestive and natural conceptual resource for presenting the theory but the
mathematical development is prompted by rather than reliant upon it.
The full hydrodynamic model of quantum mechanics therefore provides an interpretation of two picturesthe wave-mechanical (Eulerian) and the particle
(Lagrangian), and the latter is just as valid a representation of quantum processes
as the former. The word picture seems appropriate, for we shall see that the
two versions of quantum mechanics have associated Hamiltonian formulations that
are connected by a canonical transformation, much as the familiar pictures of quantum mechanics (e.g., Schrodinger and Heisenberg) are connected by unitary transformations (Section 6). This mapping therefore gives a new and mathematically
precise meaning to the notion of wave-particle duality. Indeed, the Lagrangian
particle coordinates and their canonical momenta have the character of c-number
analogues of Heisenberg-picture operators.
Without going into the intricacies of what it means to interpret quantum
mechanics [38], it is emphasized that, although it is closely related mathematically,
and is cited as inspiration for the numerical studies mentioned above, the de BroglieBohm interpretation should not be conated with the Lagrangian version of the
hydrodynamic interpretation. In particular, it is necessary to distinguish the de BroglieBohm corpuscle from a uid particle (Section 7).
The work we report is a simple generalization and combination of several strands
of classical Lagrangian uid theory, which came to fruition in the 1980s. Background material is given by Lamb [39] and the history and elements of the theory
are recorded in the beautiful book of Truesdell [40] (see also [41]). Classical variational treatments are described in [4256] and relevant work on the hydrodynamic
approach to quantum mechanics in [3,3537,5767]. A variational treatment similar
to that presented here has been given by Rylov [68] who has considered a generalization of the classical theory to include dependence of the internal energy on higher
derivatives of the density of the kind needed in the quantum case, and has written
down a special case of one version of the equation of motion for the uid coordi-

P. Holland / Annals of Physics 315 (2005) 505531

509

nates. Rylov approaches the Lagrangian theory in a somewhat dierent spirit,

regarding it as descriptive of an ensemble of stochastic processes, and does not discuss the benets of the approach in providing an alternative method of computation,
or other aspects such as the canonical mapping. Other examples of the mapping between Eulerian eld equations and Lagrangian coordinates are discussed in [69].

2. Lagrangian picture
The Lagrangian picture in hydrodynamics is a eld theory but its discourse is
drawn from particle mechanics. A uid is a continuum of elements (particles) distributed throughout Euclidean space, each of which retains its identity throughout the
evolution of the whole. The history of the system is encoded in the state variables
q (a, t), the positions of all the distinct uid elements at time t, each particle being distinguished by a continuously variable vector label a. For convenience we take a to be
the particle position at t = 0. The motion is continuous in that the mapping from aspace to q-space is single-valued and dierentiable with respect to a and t to whatever
order is necessary, and the inverse mapping a (q, t) exists and has the same properties.
These assumptions are in accord with the properties of the velocity eld implied by
quantum mechanics (i.e., the ratio of current to density), in particular its single-valuedness. The entire set of motions for all a is termed a ow. The vectors q and a are
referred to the same set of Cartesian space axes but they may also be regarded as
related to one another by a time-dependent coordinate transformation. To each particle there is associated an elementary volume whose mass is conserved by the ow.
The whole is structured by an internal potential derived from the density that represents a certain kind of particle interaction, and each particle responds to the potential via a force whose action is described by a form of Newtons second law. For all
these reasons, we shall regard the model as providing a particle picture.
Let q0 (a) be the initial quantal probability density. In the hydrodynamic model
q0 (a) is identied with the initial number density (which is normalized:
q0 (a) d3a = 1). Then, introducing a mass parameter m, the mass of an elementary
volume d3a attached to the point a is given by mq0 (a) d3a. The signicance of the
parameter m, conventionally described as the mass of the quantum system, is that
it is the total mass of the uid, since mq0 (a) d3a = m. In this picture, the conservation of the mass of a uid element in the course of its motion is expressed through the
relation
mqqa; td3 qa; t mq0 ad3 a

2:1

qa; t J
1 a; tq0 a;

2:2

or

where J is the Jacobian of the transformation between the two sets of coordinates
J

1
oq oqj oqk
eijk elmn i
;
3!
oal oam oan

0 < J < 1:

2:3

510

P. Holland / Annals of Physics 315 (2005) 505531

Here eijk is the completely antisymmetric tensor with e123 = 1 and the indices i, j,
k, . . . = 1, 2, 3. Summation over repeated indices is always assumed.
Let V be the potential of an external (classical) conservative body force and U the
internal potential energy of the uid due to interparticle interactions. We assume
that the Lagrangian has the same form as in the classical theory of ideal uids except
for the functional dependence of U: this depends on q (q) and its rst derivatives, and
hence from (2.2) on the second-order derivatives of q with respect to a, and is independent of other variables such as entropy. The Lagrangian is then
Z
Lq; oq=ot; t lq; oq=ot; oq=oa; o2 q=oa2 ; t d3 a

Z

1
2
mq aoqa; t=ot
q0 aU q
q0 aV qa d3 a:

2 0
2:4
Here q0 (a) and V are prescribed functions and we substitute for q from (2.2). We
assume that q0 and its derivatives vanish at innity, which ensures that the surface
terms in the variational principle vanish. If in addition to the equations for the coordinates q we wished to derive (2.2) from a variational principle we could add
k (Jq
q0) to the integrand in (2.4), where k is an undetermined multiplier, and vary
independently with respect to k, q, and q. This construction is unnecessary for our
purposes here (we shall however absorb the (Eulerian version of the) conservation
equation into a canonically related action principle in Section 6).
It is the action of the conservative force derived from U on the trajectories that
represents the quantum eects in this theory. As we shall see, these eects are characterised by the following choice for U (motivated by the well known Eulerian
expression for the internal energy):
h2 1 oq oq


h2 1


o q0  o q0 
U

J
J
;
2:5
ij
ik
2
oaj J oak J
8m q2 oqi oqi 8m q0
where we have substituted from (2.2) and used
o
o
J
1 J ij
;
oqi
oaj

2:6

where
J il

oqj oqk
oJ
1
eijk elmn
ooqi =oal 2
oam oan

2:7

is the cofactor of oqi/oal. The latter satises

oqk
J ki J dij :
oaj

2:8

Clearly, U has a local dependence on q and its derivatives and the coordinates q enter
only through the deformation gradients oqi/oaj and their derivatives with respect to
a. This dependence implies an interaction between each uid particle and its neighbours. The fact of interaction in the quantum case does not signal a conceptual

P. Holland / Annals of Physics 315 (2005) 505531

511

departure from classical uid dynamics where likewise the internal forces depend on
the deformation gradients. The dierence here is that the order of the derivative-coupling of the particles is higher than is customary in a classical equation of state, and
implies a considerably more complex and subtle mutual particle dependence. Other
features of the quantum ow, such as non-crossing of the paths, are also anticipated
in the behaviour of classical ideal uids but note that not all features arethere is a
notable dierence in the vortex structures, for example (see Section 3).
The EulerLagrange equations for the coordinates,
o
oL
dL


0;
ot ooqi a; t=ot dqi a

2:9

where
dL
ol
o
ol
o2
ol

dqi oqi oaj ooqi =oaj oaj oak oo2 qi =oaj oak

2:10

give
mq0 a

o2 qi a
oV oW ij

q0 a


;
ot2
oqi
oaj

where
oU
o

W ij
q0 a
ooqi =oaj oak

!
oU
q0 a
:
oo2 qi =oaj oak

2:11

2:12

They obviously have the form of Newtons second law. We shall not give the explicit
form of the quantity Wij but will utilise instead a more useful related tensor, rij, dened by Wik = Jjkrij. The latter is the analogue in this theory of the classical pressure
tensor, pdij. Using (2.8) we can invert to obtain
rij J
1 W ik

oqj
:
oak

2:13

Evaluating (2.12) and using the relation (2.8) and its derivatives with respect to oqm/
oan the pressure tensor may be written


 oq0 o2 qm
h2


oq0 oq0 
1
o2 q0
J q
1
J J jl J mn
J jm ln

J jl
rij
0 J jl
3 ik
oak oal
oal oak oan
oak oal
4mJ
2

 o qr o2 qm
q0 J
1 J mn J jrls J
1 J jl J mrns
2J
2 J jl J mn J rs
oak oas oal oan

3
o
q
m
q0 J
1 J jl J mn
;
2:14
oak oal oan
where
J jm ln

oJ jl
oq
ejmk eln r k :
ooqm =oan
oar

2:15

512

P. Holland / Annals of Physics 315 (2005) 505531

It is easy to check that this tensor is symmetric. The equation of motion of the ath
uid particle moving in the eld of the other particles and the external force is now
mq0 a

o2 qi a
oV
orik

q0 a

J kj
;
ot2
oqi
oaj

2:16

where we have used the identity

oJ ij
0:
2:17
oaj
This (second order in t, fourth order in a) local nonlinear partial dierential equation
(or its equivalent forms (2.18), (3.6), and (3.11) below) is the principal analytical result of the quantum Lagrangian method. For we shall see that from its solutions
qi (a,t), subject to specication of oqi0/ot whose determination is discussed in the next
section, we may derive solutions to Schrodingers equation. The remainder of the paper is mainly concerned with the various facets of the relationship between the two
dynamical equations.
Other forms of (2.16) will be useful. Multiplying by oqi/oak we obtain the
Lagrangian form
o2 qi a oqi
oV
oq
orik

q0 a

i J kj
:
2:18
ot2 oak
oak oak
oaj
In addition, it is straightforward to develop the Hamiltonian formalism for the quantum uid. The canonical eld momenta are
oL
oq a
2:19
mq0 a i
pi a
ooqi a=ot
ot
so that, making the usual Legendre transformation, the Hamiltonian is
Z
oq a
H q; p; t pi a i d3 a
Lq; oq=ot; t
ot
#
Z "
2
pa

1
q0 a U J q0 q0 aV qa d3 a:
2:20

2mq0 a
mq0 a

Hamiltons equations can be written in terms of the usual Poisson brackets, where
the basic relations are
fqi a; qj a0 g fpi a; pj a0 g 0;

fqi a; pj a0 g dij da
a0 :

2:21

We obtain
oqi a
dH

;
ot
dpi a

opi a
dH


ot
dqi a

2:22

which when combined reproduce (2.16).

3. Quasi-potential ow
To obtain a ow that is representative of quantum mechanics we need to restrict
the initial conditions of (2.16) to those that correspond to what we shall term quasi-

P. Holland / Annals of Physics 315 (2005) 505531

513

potential ow. This means that the initial velocity eld is of the form (we introduce
the mass factor for later convenience)
oqi0 a 1 oS 0 a

ot
m oai

3:1

but the ow is not irrotational everywhere because the potential S0 (a) (the initial
quantal phase) obeys the quantization condition
I
I
oqi0 a
1 oS 0 a
nh
dai
3:2
dai ; n 2 Z;
ot
m oai
m
C

where C is a closed curve composed of material particles. If it exists, vorticity occurs

in nodal regions (where the density vanishes) and it is assumed that C passes through
a region of good uid, where q0 0. To show that these assumptions imply motion
characteristic of quantum mechanics we rst demonstrate that they are preserved by
the dynamical equation. To this end, we use the method based on Webers transformation applied to the law of motion in the Lagrangian form (2.18) [39,40] together
with a quantum analogue of Kelvins circulation theorem. Other methods are available, such as Cauchys vorticity formula [40], but we shall not discuss them here.
We rst get (2.18) into a more convenient form. Expressed in terms of the dependent variables q (a), and using (2.2), the stress tensor (2.14) takes a simpler form:
!

h2 1 oq oq
o2 q
rij


:
3:3
4m q oqi oqj oqi oqj
Using the easily proved but important result
1 orij oV Q

;
q oqj
oqi

3:4

where
VQ




2
h
1 oq oq
o2 q


4mq 2q oqi oqi oqi oqi

3:5

is the de BroglieBohm quantum potential [3,4],1 we see that (2.18) may also be
simplied
m


o2 qi oqi
o 


V VQ :
oak
ot2 oak

3:6

We now integrate this equation between the time limits (0, t):
m

oqi oqi
oq
ova; t
m k0
;
oak
ot oak
ot

3:7

1
Madelung [34] was one of the rst to recognize the importance of this quantity. The earliest use of the
expression quantum potential is apparently due to de Broglie [70].

514

P. Holland / Annals of Physics 315 (2005) 505531

where
va; t

t
0

!
 2
1
oq
m

V
V Q dt:
2
ot

3:8

Then, substituting (3.1),

oqi oqi
1 oS

;
ot oak m oak

Sa; t S 0 a va; t;

3:9

with initial conditions q = a, v0 = 0. The left-hand side of (3.9) gives the velocity at
time t with respect to the a-coordinates and this is obviously a gradient. To obtain
the q-components we multiply by J
1 Jik and use (2.6) and (2.8) to get
oqi 1 oS

;
m oqi
ot

3:10

where S = S (a (q, t), t). Thus, for all time the velocity of each particle is the gradient
of a potential with respect to the current position.
Eq. (3.10) is a form of the law of motion. Corresponding to this, it is notable that
we may write (3.6) in terms of the current variables as well:
m

o2 qi
o


V V Q :
oqi
ot2

3:11

This puts the uid-dynamical law of motion (2.16) in the form of Newtons law for a
particle of mass m moving in the potential V + VQ, a point we discuss in Section 7. In
the case where we substitute (2.2) in (3.5), a special case of (3.11) (corresponding to
q0 = 1) has been written down by Rylov [68].
To complete the demonstration, we note that the motion is quasi-potential since
the value (3.2) of the circulation is preserved following the ow:
I
o
oqi
dqi 0;
3:12
ot
ot
Ct

where C (t) is the evolute of the material particles that compose C. This theorem has
been stated previously in the quantum context [35] and is proved in the usual way
[44,71]. We conclude that each particle retains forever the quasi-potential property
if it possesses it at any moment.
To obtain the equation obeyed by the potential S, we use the chain rule applied to
a function F (a, t) with a = a (q, t):


oF
oF
oq oF
i
:
3:13

ot a
ot q ot oqi
Then, since from (3.9) v = S
S0,
!



ov
oS
oqi oS
oS 0
oqi oS 0

:
ot a
ot q ot oqi
ot q ot oqi

3:14

P. Holland / Annals of Physics 315 (2005) 505531

515

Using (3.13) the two terms in the brackets sum to oS0 (a)/ot = 0. Using (3.10), (3.14)
therefore becomes


 2
ov
oS
oq

m
:
3:15


ot a
ot q
ot
Thus, since from (3.8)

 2
ov
1
oq
m

V
V Q;

ot a 2
ot
we obtain nally, using (3.10),
 2
oS
1 oS

V V Q 0:
ot 2m oq

3:16

3:17

This is the quantum Hamilton-Jacobi equation,2 one half of Schrodingers equation, which will be derived again in Section 4. We have therefore shown that the set
of equations comprising the conservation law (2.2) and the EulerLagrange equations (2.16), together with the initial condition (3.1), are equivalent to the ve equations (2.2), (3.9) and (3.17). These equations serve to determine the functions qi, q, S,
which are now the unknowns. The completion of the demonstration that the collective particle motion generates Schrodinger evolution is given in Section 4. Note that
although the particle velocity is orthogonal to a moving surface S = constant, the
surface does not keep step with the particles that initially compose it and hence it
is not a material surface.
There is a noteworthy dierence between quantum and classical vortex lines.
One of the classical Helmholtz laws states that vortex lines are material lines,
i.e., they are composed of the same material particles for all time, and so move
with the uid [73]. In contrast, in the quantum case vorticity occurs only where
q = 0, i.e., where there are no material particles, so the vortex lines (strictly, line
vortices) cannot be material lines. Indeed, oq/ot = m
1 oS/oq is singular at nodes
so the material velocity does not give a description of the vortex motion. Rather,
the spacetime behaviour is given directly by the (two real) condition(s) w (x, t) = 0
(in Eulerian terms; see Section 4). The vortex motion is nevertheless closely related
to the surrounding local ow, as is obvious from the circulation theorem (3.12).
Thus, the loop C (t) can be chosen arbitrarily small as long as it does not pass
through a node. Then a vortex line initially inside C (t) cannot cross it because this
would imply a change in the circulation, contradicting the theorem. Therefore, the
vortex is trapped inside, and moves with, the surrounding good uid. The
quantum analogue of the classical theorem is therefore that vortex lines move with
the uid, but are not material lines. The motion of quantum vortices is discussed in
[3,36,37,66,67].

The validity of this epithet is discussed in [72].

516

P. Holland / Annals of Physics 315 (2005) 505531

4. Eulerian picture: Schrodingers equation

The fundamental link between the particle (Lagrangian) and wave-mechanical
(Eulerian) pictures is dened by the following expression for the Eulerian
density:
Z
qx; t dx
qa; t q0 a d3 a:
4:1
This relation may be regarded as a component of a transformation from the three
coordinate functions qi (a) to a new system of coordinate functions one of which is
q (x)indeed, as we shall see in Section 6, (4.1) is a building block of the canonical
transformation linking the two pictures. The corresponding formula for the Eulerian
velocity v is contained in the expression for the current:
Z
oqi a; t
dx
qa; t q0 a d3 a:
qx; tvi x; t
4:2
ot
Relations (4.1) and (4.2) play an analogous role in our approach to the Huygens
formula
Z
4:3
wx; t Gx; t; a; 0 w0 a d3 a
in Feynmans method, for from them we can construct the wavefunction, up to a global phase. In view of the similar role it plays to G in (4.3), we may refer to the delta
function d (x
q (a, t)) as the propagator of the motion in time. Unlike the manyto-many mapping embodied in the formula (4.3), the quantum evolution is described
here by a local point-to-point development. Using the result

dx
qa; t J
1 ax;t da
a0 x; t; x
qa0 ; t 0;
4:4
and evaluating the integrals, (4.1) and (4.2) are equivalent to the following local
expressions (we give also the inverse relations):

qx; t J
1 ax;t q0 ax; t; qxa; t; t J
1 a; tq0 a
4:5

oqi a; t
vi x; t
;
ot ax;t

vi x; tjxa;t

oqi a; t
:
ot

4:6

Relations (4.5) restate the conservation equation (2.2), and (4.6) gives the relations
between the velocities in the two pictures. We may term J
1 the local propagator.
Note that, unlike the Feynman propagator, the uid propagator, which is determined by solutions to (2.16), depends on w0 (a).
From the rst relation (4.6) we may deduce the following relations between the
accelerations in the two pictures:



ovi
ovi o2 qi a; t
ovi
ovi
o2 qi a; t
vj

:
4:7
j


2
ot
ot2
ot
oxj
ot
oxj xa;t
ax;t

P. Holland / Annals of Physics 315 (2005) 505531

517

These formulas enable us to translate the Lagrangian ow equations into Eulerian

language. Dierentiating (4.1) with respect to t and using (4.2) we easily deduce the
continuity equation
oq
o

qvi 0:
ot oxi

4:8

Next, dierentiating (4.2) and using (3.11) and (4.8) we get the quantum analogue of
Eulers equation:
ovi
ovi
1 o
vj


V V Q :
m oxi
ot
oxj

4:9

Finally, the quasi-potential condition (3.10) becomes

vi

1 oSx; t
:
m oxi

4:10

The rst members in (4.5) and (4.6) give the general solutions of the continuity equation (4.8) and Eulers equation (4.9), respectively, in terms of the paths and initial
density.
To establish the connection between the Eulerian equations and Schrodingers
equation, we note that, using (4.10) and (4.9) can be written


o oS
1 oS oS

V V Q 0:
4:11
oxi ot 2m oxi oxi
The quantity in brackets is thus a function of time. Since the addition of a function
of time to S does not aect the velocity eld, we may absorb the function in S, i.e.,
set it to zero. Then
oS
1 oS oS

V V Q 0:
4:12
ot 2m oxi oxi
p
Combining (4.8), (4.10) and (4.12), the function wx; t q expiS=
h obeys Schrodingers equation:
ow
h2 o2 w




V w:
4:13
ot
2m oxi oxi
We have deduced this from the Lagrangian particle equation (2.16) subject to the
quasi-potential requirement. The quantization condition (3.12) becomes here
I
oSx; t
dxi nh; n 2 Z;
4:14
oxi
i
h

C0

where C0 is a closed curve xed in space that does not pass through nodes. This is a
consistent subsidiary condition on solutions since it is easy to see that the value of
(4.14) is preserved in time as long as nodes do not cross C0 (an Eulerian version
of the circulation theorem).
p
Given the initial wavefunction w0 a q0 expiS 0 =
h we can compute the wavefunction for all x, t, up to a global phase, as follows. First, solve (2.16) subject to the
initial conditions q0 (a) = a, oqi0 (a) /ot = m
1 oS0 (a)/oa to get the set of trajectories

518

P. Holland / Annals of Physics 315 (2005) 505531

for all a, t. Next, substitute q (a, t) in (4.5) to nd q and oq/ot in (4.6) to get oS/ox.
This gives S up to an additive function of time, f (t). To x this function, apart from
an additive constant, use (4.12). We obtain nally the following formula for the
wavefunction:

Z

q

1 
i

wx; t
J q0 ax;t exp
4:15
moqi a; t=otjax;t dxi f t :
h


Note the key role played by the initial coordinates in this formula.
The Eulerian Eqs. (4.8) and (4.9) form a closed system of four rst-order coupled partial dierential equations to determine the four independent elds
q (x), vi (x) and do not refer to the material paths. We shall call them the basic
Eulerian equations. The erasure of the particle variables is part of the reason
why the Eulerian language is particularly suited to represent the wave-mechanical
formalism which likewise, of course, makes no reference to the trajectory concept
(another reason is that classical hydrodynamics already exhibits the key feature of
mutual coupling between the density and velocity transport equations). It will be
noted, however, that the Lagrangian theory from which we derived the Eulerian
system comprises seven independent elds q, q (a), p (a). (In the case of quasi-potential ow there are two (ve) independent Eulerian (Lagrangian) elds.) Depending
on the information one wants about the physical system this may be regarded as
indicating a redundancy in the Lagrangian description, or an incompleteness in
the Eulerian one. In the sense that the Eulerian view refers only to the variables
of interest, namely, the wavefunction, then the Lagrangian method may seem profligate. The situation is similar to that of the Feynman paths where likewise the
bones of the calculus are washed out in the nal answer. On the other hand, from
the perspective of establishing the mathematical equivalence of the pictures we
must adopt the second option. The necessity of supplementing the Eulerian picture
with variables that do not appear in the basic equations is well known in the classical variational theory [4556].
A simple solution is to append to the Eulerian description the three particle-label
functions ai (x, t). Their equations of evolution state that the labels are conserved by
the ow implied by the Eulerian velocity eld:
oai
oai
vj
0:
ot
oxj

4:16

Additional structure of this type arises naturally in the canonical theory described in
Section 6. Eq. (4.16) is equivalent to the second version of the equation of motion
(4.6). To see this, multiply (4.16) by oxk/oai to get
vk

oxk oai
:
oai ot

4:17

Dierentiating the relation x (a (x, t), t) = x with respect to t we therefore obtain

vi x; tjxa;t

oxi a; t
:
ot

4:18

P. Holland / Annals of Physics 315 (2005) 505531

519

Starting from the Eulerian approach the trajectories become derived quantities
(which we emphasize by using the label x in place of q). We can pass to the Lagrangian picture using the conversion formulas (4.5) and (4.6) above.

5. Example
The special case of a one-dimensional ow may be extracted from the three-dimensional theory when V (x) = V1 (x1) + V2 (x2) + V3 (x3), for solutions to the Eulerian
equations exist for which q (x) = q1 (x1)q2 (x2)q3 (x3) and S (x) = S1 (x1) + S2 (x2) +
S3 (x3). In these circumstances the internal potential decomposes:

2 
2 
2 !
h2


1 oq1
1 oq2
1 oq3
:
5:1
U

q1 oq1
q2 oq2
q3 oq3
8m
In the Lagrangian picture, (3.10) implies that this condition translates as independence of the orthogonal coordinates: q1 = q1 (a1, t), etc., Eq. (2.16) for q1, say, then
becomes, dropping the index,
 2
o2 q
oV
h2 o


oJ
oJ oq0
o2 J

2q0 J
5

q0 J
4 2
mq0 2
q0

J
4
ot
oq 4m oa
oa
oa oa
oa
!


2
o2 q
1 oq0
J
3 20
J
3
5:2
q0 oa
oa
with J = oq/oa.
To illustrate the method, we use formula (5.2) to compute the free (V = 0) wavefunction at time t that is initially a Gaussian at rest:
2 =2r2
0

q0 a 2pr20
1=2 e
a

S 0 a 0:

5:3

Let us try a solution of the form q (a, t) = A (a) T (t). The initial conditions
q0 (a) = a, oq0 (a)/ot = 0 imply:

6 0
A a; T 0 1; oT 0 =ot 0 a
5:4
A 0; or T 0 oT 0 =ot 0; or both a 0:
When a 0, (5.2) becomes, noting that J = T and hence oJ/oa = 0,
 2 !
o2 T

h2 o o2 q0 1 oq0


:
mq0 a 2
ot
4mT 3 oa oa2 q0 oa
Substituting from (5.3), this reduces to

2
o2 T
a
h


3; a
:
ot2
2mr20
T

5:5

5:6

The solution of (5.6) respecting the stated initial conditions is T = (1 + at2)1/2 and
the paths are therefore

520

P. Holland / Annals of Physics 315 (2005) 505531


1=2
qa; t a 1 at2
:

5:7

When a = 0, A = 0 implies q = 0 for all t. When A 0, it is easy to see that the only
solution obeying T0 = oT0/ot = 0 is T (t) = 0 so that again q = 0. Hence, (5.7) holds
for all a.
The wavefunction at time t may now be found by substituting (5.7) into the onedimensional analogues of (4.5) and (4.6) (with (4.10)):


1
qx; t oq=oa
q0 ax; t;
5:8
ax;t

oSx; t=ox moqa; t=otjax;t :

We nd
2
1=2
x2 =2r2

qx; t 2pr



1

h

ht
2 2
2

1
Sx; t matx r0 r
tan
;
2
2
2mr20

5:9

5:10

where r = r0 (1 + at2)1/2 and the time-dependent addition in S is obtained by putting

(5.9) in (4.12). This is the correct time-dependent free Gaussian wavefunction.
6. Canonical transformation linking the particle and wave pictures
The absence of the material paths in the basic Eulerian equations suggests that
q (x) and S (x) may be regarded as collective coordinatesfunctions that describe
the bulk properties of the system without depending on the complex details of the
particulate substructure. Indeed, (4.1) is a continuum analogue of the microscopic
particle density of discrete particle mechanics,
X
qx; t
dx
qa t;
6:1
a

which in certain contexts [7476] is a collective coordinate. This idea has been examined previously in the classical uid-dynamical context [77,78].
The notion that q (x) and S (x) should be regarded as (canonical) coordinates for
each x is of course the basis of the standard variational treatment of the Schrodinger
equation (where the alternative canonical variables w, w* are commonly used; see below), and the coordinate interpretation nds its most natural expression in that context. Here we shall show how to pass from the Hamiltonian theory of the Lagrangian
picture to a corresponding Hamiltonian theory of the Eulerian picture via a suitably
chosen canonical transformation in which qi (a), pi (a) are the old phase space coordinates and q (x), S (x) are related to a new set of phase space coordinates denoted
by Qi (x), Pi (x).
Apart from applying the quasi-potential restriction, the quantum treatment follows closely the corresponding classical analysis for vortical ows given by van Saarloos [52]. We suppose that the generating function of the canonical transformation is
a time-independent functional of the old coordinates and the new momenta:
W [q (a),P (x)]. The transformation formulas are therefore

P. Holland / Annals of Physics 315 (2005) 505531

Qi x

dW
;
dP i x

pi a

dW
;
dqi a

KQ; P ; t H q; p; t:

521

6:2

The W we seek is a generalization to continuous variables of the generating function

of an arbitrary coordinate transformation:
Z
W dx
qa q0 a P 1 x AaP 2 x BaP 3 x d3 a d3 x:
6:3
Here the functions A and B are arbitrary (e.g., we may choose A = a2, B = a3) subject
to certain conditions such as that the Jacobian of the transformation is unity. Our
transformation equations are then
Z

Q1 x dx
qa q0 a d3 a J
1 q0 ax qx;
6:4
Q2 x qxAax; Q3 x qxBax;
where we have substituted q using (4.5), and


oP 1 x
oP 2 x
oP 3 x
A a
Ba
pi a q0 a
:
oxi
oxi
oxi
xa

6:5

The transformation is therefore of the type sought in which q becomes a coordinate.

Combining these formulas and substituting in (2.20) the new Hamiltonian is

2
Z "
Q1 x oP 1 x Q2 x oP 2 x Q3 x oP 3 x

KQ; P ; t
2m
ox
Q1 x ox
Q1 x ox

Q1 x U Q1 ; rQ1 Q1 xV x d3 x:
6:6
Using (2.19) and (4.6) the expression (6.5) for the momentum is just an extended Clebsch representation of the Eulerian velocity (any vector eld can be expressed in this
form [54]):
oP 1 x
oP 2 x
oP 3 x
mvi x
Aax
Bax
:
6:7
oxi
oxi
oxi
Using this shorthand along with Q1 = q the Hamiltonian can be written in terms of
the non-canonical variables q, v, as in classical hydrodynamics

Z 
1
mqv2 qU q; rq qV d3 x:
K
6:8
2
Hamiltons equations for the new variables are:
oQ1
dK
o

Q vj ;
dP 1
oxj 1
ot

6:9

oQ2
dK
o

Q vj ;
dP 2
oxj 2
ot

6:10

oQ3
dK
o

Q vj ;
dP 3
oxj 3
ot

6:11

522

P. Holland / Annals of Physics 315 (2005) 505531

oP 1
dK
1
oP 1


mv2
vi

VQ
V;
dQ1 2
ot
oxi
Z
d
Q1 U Q1 ; rQ1 d3 x;
VQ
dQ1

6:12

oP 2
dK
oP 2



vi
;
dQ2
ot
oxi

6:13

oP 3
dK
oP 3



vi
:
dQ3
ot
oxi

6:14

It is convenient to replace the equations for Q2 and Q3 by equations obeyed by the

functions A and B given in (6.4):
oA
oA
vi
0;
ot
oxi

oB
oB
vi
0:
ot
oxi

6:15

Hamiltons equations therefore become (6.9) and (6.12) together with the four
Eqs. (6.13), (6.14) and (6.15) which state that each of the functions A, B, P2, P3 is a
constant following the ow. Eq. (6.9) is just the Eulerian conservation equation
(4.8) and it is straightforward to derive Eulers equation (4.9) from (6.7) using
(6.12)(6.15).
The coordinates Pi are dened by relation (6.7) only up to a certain transformation, an analogue of the corresponding result for the usual Clebsch or Euler representation of a vector eld [42,44,79]. For, suppose (6.7) is satised by another set
of parameters A 0 , B 0 , P 01 , P 02 , P 03 . Then, subtracting the new relation from the old, we
have
0

oP 1
P 01
oP 2
oP 0
oP 3
oP 0
A

A0 2 B

B0 3 :
oxi
oxi
oxi
oxi
oxi

6:16

This dierential relation implies that

P 01
P 1 f ;

of
;
oP 2

of
;
oP 3

A0

of
;
oP 02

B0

of
oP 03

6:17

for some function f P 2 ; P 02 ; P 3 ; P 03 . The variables A, B, P2, P3 and A 0 ,B 0 , P 02 , P 03 are

therefore connected by a canonical transformation generated by f. The functions
A, B being xed parameters of the transformation (6.3), we have A = A 0 , B = B 0
and the generating function has the form f P 2
P 02 ; P 3
P 03 .
To complete the canonical mapping linking the pictures we must state the conditions obeyed by the new variables corresponding to the initial quasi-potential
requirement in the old variables. At time t, the latter condition reads
pi a oqi
oS

6:18
q0 a oak oak
(obtained by combining (2.19) with (3.9)). The relations (6.18) imply that the initial
data for Hamiltons equations is given by the four independent functions S0, qi0 (= ai)
(q0 being regarded as a prescribed function). From (4.10) we have initially

P. Holland / Annals of Physics 315 (2005) 505531

523

mvi0 = oS0/oxi which is connected with the initial variables Qi0, Pi0 via (6.7). At time t
we then have
o
oP 2
oP 3
S
P 1 A
B
:
oxi
oxi
oxi

6:19

These relations x the functions A and B in terms of the gradients of S and Pi. Following (6.17), (6.19) implies that
S
P 1 g;

og
;
oP 2

og
oP 3

6:20

for some function g (P2, P3). This establishes the required relations between the Qs
and Ps. To conrm this, we derive the equation obeyed by S. Thus, substituting
for P1 in (6.12) and using (6.13) and (6.14) we obtain (4.12), as required. In addition,
Eqs. (6.13) and (6.14) imply (6.15) via the second and third relations in (6.20). The
initial data is therefore given by S0, Q10 (= q0), P20, P30.
We conclude that Hamiltons equations (6.9)(6.14)) for Q, P, subject to the
restrictions (6.20) imposed at t = 0, imply the two Eulerian equations that make
up Schrodingers equation together with two equations determining constants of
the motion.
Some words of clarication on the canonical theory may be useful. In the usual
variational approach to the Schrodinger equation the Hamiltonian is written

Z  2
h


2


0
K w; w 
rw  rw V jwj d3 x;
6:21
2m
where w, w* are conjugate variables. Alternatively, we may make a canonical transformation to the conjugate variables q (= Q 0 ), S (= P 0 ) and get the Hamiltonian

Z 
1
2
00
rS U q; rq V d3 x:
K q; S q
6:22
2m
Hamiltons equations for Q 0 , P 0 lead to the required self-contained equations for
q, S. The point of the demonstration above is that we can achieve the same set of
self-governing equations using an extended Hamiltonian formalism where q is one
of three coordinate functions and S enters not as its conjugate quantity but through
an initial functional relationship that is maintained by Hamiltons equations. It is the
enhanced canonical version of wave mechanics, rather than the conventional one,
that maps to the particle Hamiltonian theory.
Finally, we give the Lagrangian in the Eulerian picture:
Z
oP i x 3
LP ; oP =ot; t Qi x
d x
KQ; P ; t
ot


Z
oP 1
oP 2
oP 3 1 2
A
B

mv
U q; rq
V d3 x:
q
2
ot
ot
ot
6:23
Naturally, the EulerLagrange equations for Pi, q, A, and B reproduce Hamiltons
equations (6.9)(6.14), after some rearrangement.

524

P. Holland / Annals of Physics 315 (2005) 505531

7. Relation with the de BroglieBohm theory and related hydrodynamic approaches

It will be noted that the law of motion (4.18) (or (3.11)) for the uid elements coincides with that postulated in the de BroglieBohm interpretation of spin 0 quantum
mechanics [3,4]. It is important, however, to distinguish the present approach from
the general issue of interpretation, and to discriminate its hydrodynamic interpretation from the de BroglieBohm interpretation. This is particularly important as
the uid and de BroglieBohm approaches are sometimes discussed as if they are
identical.
Regarding the general problem, we emphasize that, in the rst instance, the purpose of the present theory is to provide an alternative method of computation that,
in itself, is neutral regarding interpretation (in this regard it has a status analogous to
that of the path-integral method). Of course, this mathematical formulation is particularly suited to a hydrodynamic discourse and hence may be regarded as having a
bearing on the meaning of quantum theory. In this model, the uid is composed of a
continuum of ctitious probability elements each propagating deterministically
along a uid path (we have used mass elements to present the theory as fully as
possible in the hydrodynamic language). The probability here refers to the distribution of measurement outcomes. The mass of a uid particle is xed by the initial density on the relevant path. All the uid particles are present together and their
collective motion is an alternative representation of quantum evolution. Unlike
the original Madelung (Eulerian) formulation, the Lagrangian picture adds variables
(a) to the quantum formalism but none is singled out as special.
In contrast, we recall that the purpose of the de BroglieBohm approach is to
underpin the quantum statistical description with a theory of individual causally
connected physical processes, a problem that is not addressed by the hydrodynamic
approach in its primitive form. For this purpose, a corpuscle of mass m is introduced
and the wavefunction (strictly, gauge-invariant functions of it [72]) is attributed
ontological status.3 The statistical aspect of the wavefunction is now a secondary
property and has a dierent meaning in referring to the likely current location of
the particle rather than just the measured location. Although mathematically similar,
the de BroglieBohm theory is logically independent of and involves elements additional to the hydrodynamic model. Within the context of the hydrodynamic model,
the de BroglieBohm corpuscle should be regarded as a passive foreign body immersed in the uid, or tracer, a lone uid path being singled out as its propagation
track.4 The mass point may lie on any path for which the probability is nite but the
continuum of paths is realized only over an (innite) ensemble of trials in each of

It is sometimes overlooked that these features do not fully characterise the de BroglieBohm model,
or rather, it does not full its potential as a physical theory just with these concepts. Its explanations draw
freely upon the full range of physical concepts (such as energy and force). This allows a signicant
improvement in the clarity and consistency with which language is used in quantum theory, where
concepts are now no longer just words attached to symbols.
4
There is debate about where mass is located in the pilot-wave theory [80]. If, as here, the corpuscle is
attributed a mass, the uid model suggests the further question as to what xes its numerical value.

P. Holland / Annals of Physics 315 (2005) 505531

525

which just one is traversed. In the original form of the theory, the law of motion of
the corpuscle was appended to the Schrodinger equation in its Eulerian form but it
can equally well be appended to the Lagrangian form (where it coincides with that of
one of the uid particles). The de BroglieBohm theory involves a further level of interpretation of quantum mechanics beyond the hydrodynamic interpretation; this includes
both the Eulerian and Lagrangian pictures (or any other formulation) and it should
not be identied with either. We attribute deviations in the motion of the corpuscle
to local action of the quantum potential in the former case and to interactions between uid particles in the latter; in both cases the corpuscle is a passive entity that
does not act back upon the guiding agent. The corpuscle is therefore to be distinguished from a uid particle both in its mass and in its dynamical behaviour. That
the de BroglieBohm approach relates to a dierent physical problem to the hydrodynamic model becomes clear when we consider, for example, the phase space theory
of the eld-corpuscle system where it is necessary to introduce contributions for both
the Schrodinger eld (or set of uid elements) and the corpuscle in the total Hamiltonian to achieve a consistent theory of their interaction [72].
Having said this, the uid approach indicates that the paths employed in the de
BroglieBohm theory, when shorn of an attached corpuscle, play a key role in quantum mechanics. We can, therefore, invoke the large de BroglieBohm-related oeuvre
as a resource descriptive not of an ensemble of possible tracks of a single particle but
of an alternative model of quantum evolution that provides insight not available in
other approaches (e.g., the trajectories corresponding to two-slit interference
[81,82]). In particular, the unambiguous solutions given by the de BroglieBohm
interpretation to problems that have been dicult to solve in other ways can be reinterpreted as answers given by quantum mechanics itself. These tend to be cases
requiring denitions of particle-like concepts such as speed and separation that
are ambiguous in a pure-wave context. Examples include analyses of tunnelling times
(e.g., [83]) and chaos (e.g., [3,4,8494]). In the latter case it is important to note that
we can formulate the notion of Lagrangian chaos (to be distinguished from the Eulerian variety [9597]) in which initially adjacent uid elements diverge exponentially
in time. This concept is absent in wave mechanics where, as is well known, the linearity of the wave equation precludes the appearance of (Eulerian) chaos. In contrast, many examples of chaotic behaviour have been found in the (highly
nonlinear) Lagrangian description. This may be characterised by positive quantum
Lyapounov exponents, the same denition being used for an exponent as for classical dynamical systems [95].
Likewise, what may appear as somewhat arcane unresolved problems in de
BroglieBohm-like theoriessuch as whether they admit relativistic formulationsmay be recast as open problems within the particle picture of quantum
theory.
Finally, we remark on the relation between the Lagrangian picture as presented
here and the numerical techniques mentioned in Section 1. An important example
of the latter is the quantum trajectory method [14]. The idea is to write the Eulerian Eqs. (4.8) and (4.12) in terms of the time derivative along a uid particle track,
d/dt = o/ot + vjo/oxj:

526

P. Holland / Annals of Physics 315 (2005) 505531

dq
ovi

q
;
dt
oxi

7:1

 2
dS
1 oS

V
V Q:
dt 2m ox

7:2

The uid is approximated as a discrete set of particles of mass m and the coupled
Eqs. (7.1) and (7.2) are solved together with the particle Eq. (4.18) to give the variables x, S, and q at a point on a trajectory at time t. The solution for the wavefunction evaluated at the location of the nth particle is then


 Z t

Z
1 t
i
wxn t w0 xn 0 exp

ovi =oxi jxn t dt exp
LQ xn t dt ;
2 0

h 0
7:3
2

where LQ = (1/2m)(oS/ox)
V
VQ is the Lagrangian of a particle of mass m

moving in the potential V + VQ. This and related methods of solution dier from
the full Lagrangian technique of Sections 2 and 4, where the EulerLagrange
equations for the paths are solved independently of the computation of the wavefunction (the latter being deduced from formula (4.15)), and the paths are not populated by particles of mass m.
In fact, between the extremes of what we might call the pure Lagrangian formalism (in which the wavefunction appears only in the initial conditions) and the
pure Eulerian formalism (in which the trajectories appear only in extraneous equations), many intermediate pictures, or combinations of equations, are possible. An
example similar to that just described was given in Section 3 where the Lagrangian
equations subject to the initial quasi-potential condition were shown to be equivalent
to a set of equations mixing q, S, and q. The application of mixtures of equations is
well known in numerical hydrodynamics (e.g., [98]).

8. Discussion
We have established a formal phase space relation between the wave-mechanical
formalism for a single body and a (continuously-)many-particle theory possessing an
interaction potential of a certain kind. This is achieved by mapping the particle
Hamilton equations together with certain initial conditions into a corresponding
Hamiltonian theory that implies the Schrodinger equation together with extraneous
eld equations. We thereby obtain complementary descriptions of a single quantum
process of a quite dierent kind to the usual notion of complementarity (the application of concepts in mutually incompatible experimental contexts). The two pictures entail dierent concepts of quantum state, and consequently tell dierent
stories about the history of a system. Thus, consider a xed space point x and the
passage of time t :0 t. Then w (x, t) describes how the amplitude at the point
evolves from w0 (x) during the time interval, and similarly for derived quantities such
as velocity. On the other hand, for a particle that occupies the point at time

P. Holland / Annals of Physics 315 (2005) 505531

527

t, x = q (a, t) describes how the particle travelled there along a unique track through
space from an initial point a. That is, w encodes the temporal history of a space
point, while q encodes its spatial history. Given the initial wavefunction, the time
dependence of either state function can be computed and implies the other. The theory brings the wave equation within the orbit of rational mechanics but of course
we do not suggest that quantum mechanics can be incorporated in some classical
model. Those who would dismiss the trajectory concept when presented in the de
BroglieBohm context can no longer do so without risk of rejecting quantum
mechanics itself.
Whether the exact computational scheme embodied in (2.16) and (4.15) will prove
of practical value is uncertain. The little that is known about the solutions to (2.16)
comes essentially from studies of the de BroglieBohm model. That wave mechanics
may be so constructed that it can be mapped into a particle-like theory is not surprising since it is a Hamiltonian system. It is nevertheless notable that it is connected to a
deterministic particle system of Newtonian type. The wave-mechanical formulation
is undoubtedly simpler but it is arguable that the detailed model provided by the particle approach makes it the more fundamental description [54]. Since the initial particle positions a do not appear in the nal computed wavefunction (just as
Feynmans paths leave no trace in w), they invite comparison with, and interpretation as, hidden variables. Indeed, the step to the de BroglieBohm model looks
more compelling from this viewpoint but it is necessary to distinguish the de BroglieBohm and hydrodynamic interpretations.
We conclude with some further remarks and suggestions for extensions of the
theory:
1. Our results constitute an additional reason to formulate hydrodynamical versions
of quantum evolution equations. Although hydrodynamics is a natural language
in this context, the mathematics is compatible with other theories of continuum
mechanics, for example, elasticity. In that case vortex lines may be regarded as
(wavefront) dislocations, the circulation (4.14) being a quantized Burgers vector
[99,100].
2. It is straightforward to include an external vector potential in the particle-picture
Lagrangian using the standard minimal coupling procedure. The extension to an
N-body quantum system is also easy in principleone just extends the range of
values taken by the Latin indices. The theory is considerably more complicated,
though, as the Jacobian J becomes a sum of products of 3N deformation gradients. The uid analogy also becomes less compelling as a uid particle has
3N components, and the Eulerian functions are dened on 3N-dimensional space.
3. The uid particle paths may be regarded as analogues in the full wave theory of
rays in the geometrical optics limit. The latter (classical) limit is obtained in circumstances where the internal potential and force are negligible in comparison
with the external body potential and force, respectively. The result is a uid-dynamical representation of the classical HamiltonJacobi theory (subject to the
caveats discussed in [101]). Note that this limit describes a continuous ensemble
of non-interacting trajectories moving in the potential V, rather than just one.

528

P. Holland / Annals of Physics 315 (2005) 505531

4. It is anticipated that other eld equations of physics will be similarly susceptible to

an analysis in terms of uid paths (e.g., [69]). Indeed, Eulerian pictures for the
Pauli, Dirac, Weyl, and Maxwell wave equations are well developed [3,102
104]. However, the equations in the standard Eulerian approach to systems with
spin are complicated and dicult to interpret, at least in comparison with the simplicity of the spin 0 Madelung model, and there are ambiguities in identifying suitably dened phases of the relevant wavefunctions (and the associated vortex
structures) [105]. It has been suggested that the origin of these problems is that
the standard approach works with the wrong (angular momentum) representation of the quantum theory, in which the rotational freedoms appear as discrete
indices in the wavefunction [3]. The local uid quantities (density, velocity, spin
vector, . . .) are dened by averaging over these indices, and reproducing
through them all the information in the wavefunction requires introducing ever
more complex quantities and combinations of quantities (especially in the
many-particle case; see, e.g., [106]). The alternative procedure advocated in [3]
is to start from the angular coordinate representation in which the spin freedoms
are represented as continuous parameters a (Euler angles) in the wavefunction, on
the same footing as the spatial variables x :w (x, a, t). This implies a physically
clearer and simpler hydrodynamic-like model, in both its Eulerian and Lagrangian guises. The phase S of the wavefunction is immediately identiable and the
equations for oa/ot and ox/ot are dened in terms of the gradients of S with respect to the respective coordinates, obvious generalizations of the spin 0 theory.
The approach also provides a natural framework to study vortices, which occur
in nodal regions in (x,a)-space, and the extension to many-body systems is
straightforward.
5. The issue of the uniqueness of the Lagrangian model associated with given Eulerian equations has several aspects. For example, without altering the dynamics
one can modify the denition of the quantum pressure tensor (2.14) by adding
a divergence-free tensor [3,57]. A further point is whether the Lagrangian (2.4)
may be modied so that it is compatible with a law of motion dierent to that implied by the quasi-potential requirement, in accordance with the alternative guidance equation implied by relativistic constraints [82,107109].
6. Although the full translation of the quantum formalism into the Lagrangian language remains to be investigated, some preliminary remarks may be made. For
the mean position and momentum in the two pictures we have:
Z
Z
h^xi i xi qx; t d3 x qi a; tq0 a d3 a;
8:1
h^
pi i

oSx; t
qx; t d3 x
oxi

Z
m

oqi a; t
q0 a d3 a:
ot

8:2

It is clear from these formulas that, as mentioned previously, the uid coordinates
and momenta are playing the role of Heisenberg c-numbers, that is, they carry
the time variation while the (Schrodinger) quantum state is static. Although the
values of the new variables coincide with the spectra of the position and momen-

P. Holland / Annals of Physics 315 (2005) 505531

529

tum operators, in themselves they give no clue as to which result will be obtained
in a measurement of position or momentum. Their connection with observables
requires careful consideration, as does the connection between the canonical and
unitary mappings. The simultaneous attribution of position and momentum variables to each uid element does not contradict Heisenbergs relations for reasons
already discussed in detail in connection with the de BroglieBohm theory [3].
7. The Lagrangian picture exhibits a new quantum symmetry, viz., a continuous particle-relabelling transformation with respect to which the Eulerian functions are
invariant [68]. This is an analogue of the classical symmetry connected with vorticity conservation [110].
8. We may try to develop the quantum/uid analogy further and enquire whether
the Schrodinger uid may be modelled in terms of some microscopic structure,
much as a classical uid is ultimately composed of atoms, and if so what this
structure must be in order that the bulk or collective behaviour of the atoms
is governed by Schrodingers equation. We shall show elsewhere that the Schrodinger uid may be so modelled if the atoms are governed by a Hamiltonian
dynamics involving a particular set of point-like interactions. This throws light
on the oft-made claim that the quantum potential acts nonlocally in the single-particle case (of course we expect the N subparticles that constitute a uid
particle in N-body quantum theory to interact nonlocally, as is well understood
from the de BroglieBohm theory [3,4]).

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]

E. Schrodinger, Phys. Rev. 28 (1926) 1049.

R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integrals, McGraw-Hill, New York, 1965.
P.R. Holland, The Quantum Theory of Motion, Cambridge University Press, Cambridge, 1993.
D. Bohm, B.J. Hiley, The Undivided Universe, Routledge, London, 1993.
W. Heisenberg, in: W. Pauli (Ed.), Niels Bohr and the Development of Physics, Pergamon Press,
Oxford, 1955, p. 12.
J.H. Weiner, Y. Partom, Phys. Rev. 187 (1969) 1134.
J.H. Weiner, A. Askar, J. Chem. Phys. 54 (1108) (1971) 3434.
A. Askar, J.H. Weiner, Am. J. Phys. 39 (1971) 1230.
H.Y. Kim, J.H. Weiner, Phys. Rev. B 7 (1973) 1353.
G. Terlecki, N. Grun, W. Scheid, J. Phys. B 17 (1984) 3719.
M. Zimmermann, P. Zimmerer, N. Grun, W. Scheid, J. Phys. B 25 (1992) 2969.
B.K. Dey, A. Askar, H. Rabitz, J. Chem. Phys. 109 (1998) 8770.
F. Sales Mayor, A. Askar, H.A. Rabitz, J. Chem. Phys. 111 (1999) 2423.
R.E. Wyatt, Chem. Phys. Lett. 313 (1999) 189.
C.L. Lopreore, R.E. Wyatt, Phys. Rev. Lett. 82 (1999) 5190.
E.R. Bittner, J. Chem. Phys. 112 (2000) 9703.
E.R. Bittner, R.E. Wyatt, J. Chem. Phys. 113 (2000) 8888.
R.E. Wyatt, E.R. Bittner, J. Chem. Phys. 113 (2000) 8898.
D. Nerukh, J.H. Frederick, Chem. Phys. Lett. 332 (2000) 145.
X.-G. Hu, et al., Phys. Rev. E 61 (2000) 5967.
R.G. Brook, et al., Int. J. Quant. Chem. 85 (2001) 263.

530
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]

P. Holland / Annals of Physics 315 (2005) 505531

R.E. Wyatt, K. Na, Phys. Rev. E 65 (2001) 016702.
C.L. Lopreore, R.E. Wyatt, J. Chem. Phys. 116 (2002) 1228.
K. Na, R.E. Wyatt, Phys. Lett. A 306 (2002) 97.
X.-G. Hu, et al., Comput. Math. Appl. 43 (2002) 525.
R.G. Brook, et al., J. Mol. Struct. (Theochem) 592 (2002) 69.
C.J. Trahan, R.E. Wyatt, J. Chem. Phys. 118 (2003) 4784.
C.J. Trahan, K. Hughes, R.E. Wyatt, J. Chem. Phys. 118 (2003) 9911.
E.R. Bittner, J. Chem. Phys. 119 (2003) 1358.
B.K. Kendrick, J. Chem. Phys. 119 (2003) 5805.
J.B. Maddox, E.R. Bittner, J. Chem. Phys. 119 (2003) 6465.
R.K. Vadapalli, et al., Int. J. Quant. Chem. 94 (2003) 1.
B. Wu, et al., IEEE Trans. Nanotech. 2 (2003) 291.
E. Madelung, Z. Phys. 40 (1926) 322.
I. Bialynicki-Birula, Z. Bialynicki-Birula, Phys. Rev. D 3 (1971) 2410.
J.O. Hirschfelder, C.J. Goebel, L.W. Bruch, J. Chem. Phys. 61 (1974) 5456.
J.O. Hirschfelder, J. Chem. Phys. 67 (1977) 5477.
M. Jammer, The Philosophy of Quantum Mechanics, Wiley, New York, 1974.
H. Lamb, Hydrodynamics, sixth ed., Cambridge University Press, Cambridge, 1932.
C. Truesdell, The Kinematics of Vorticity, Indiana University Press, Bloomington, 1954.
C. Truesdell, R.A. Toupin, in: S. Flugge (Ed.), Handbuch der Physik, Band III/I, Springer, Berlin,
1960, p. 226.
J.M. Ziman, Proc. Roy. Soc. A 219 (1953) 257.
J.W. Herivel, Proc. Camb. Phil. Soc. 51 (1955) 344.
C. Eckart, Phys. Fluids 3 (1960) 421.
C.C. Lin, in: D.J. Benney, et al. (Eds.), Selected Papers of C.C. Lin, World Scientic, Singapore,
1987, p. 429.
R.L. Seliger, G.B. Whitham, Proc. Roy. Soc. A 305 (1968) 1.
F.P. Bretherton, J. Fluid Mech. 44 (1970) 19.
L.J.F. Broer, Physica 76 (1974) 364.
L.J.F. Broer, J.A. Kobussen, Appl. Sci. Res. 29 (1974) 419.
W. Yourgrau, S. Mandelstam, Variational Principles in Dynamics and Quantum Theory, Dover,
New York, 1979.
W. van Saarloos, D. Bedeaux, P. Mazur, Physica A 107 (1981) 109.
W. van Saarloos, Physica A 108 (1981) 557.
A. Gria, Physica A 127 (1984) 265.
R. Salmon, Ann. Rev. Fluid Mech. 20 (1988) 225.
P.J. Morrison, Rev. Mod. Phys. 70 (1998) 467.
R. Jackiw, Lectures on Fluid Dynamics, Springer, Berlin, 2002.
T. Takabayasi, Prog. Theor. Phys. 8 (1952) 143.
M. Schonberg, Nuovo Cimento 12 (1954) 103.
F. Halbwachs, Theorie Relativiste des Fluides a` Spin, Gauthier-Villars, Paris, 1960.
L. Janossy, Z. Phys. 169 (1962) 79.
K.-K. Kan, J.J. Grin, Phys. Rev. C 15 (1977) 1126.
E.A. Spiegel, Physica. D 1 (1980) 236.
S.K. Ghosh, B.M. Deb, Phys. Rep. 92 (1982) 1.
T. Takabayasi, Prog. Theor. Phys. 69 (1983) 1323.
R.T. Skodje, H.W. Rohrs, J. VanBuskirk, Phys. Rev. A 40 (1989) 2894.
I. Bialynicki-Birula, Z. Bialynicki-Birula, C. Sliwa, Phys. Rev. A 61 (2000) 032110.
I. Bialynicki-Birula, et al., Acta Phys. Pol. A 100 (Suppl.) (2001) 29.
Y.A. Rylov, J. Math. Phys. 40 (1999) 256;
physics/0210003/v4 (2003).
A.M. Meirmanov, V.V. Pukhnachov, S.I. Shmarev, Evolution Equations and Lagrangian
Coordinates, Walter de Gruyter, Berlin, 1997.

P. Holland / Annals of Physics 315 (2005) 505531

[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]

531

L. de Broglie, Introduction a` lEtude de la Mecanique Ondulatoire, Hermann, Paris, 1930.

L.D. Landau, E.M. Lifschitz, Fluid Mechanics, Pergamon Press, Oxford, 1959.
P. Holland, Nuovo Cimento B 116 (1043) (2001) 1143.
P.G. Saman, Vortex Dynamics, Cambridge University Press, Cambridge, 1992.
D. Pines, D. Bohm, Phys. Rev. 85 (1952) 338.
J.K. Percus, G.J. Yevick, Phys. Rev. 110 (1958) 1.
D. Bohm, B. Salt, Rev. Mod. Phys. 39 (1967) 894.
J.P. Boon, S. Yip, Molecular Hydrodynamics, Dover, New York, 1991.
Ph. Choquard, Physica A 279 (2000) 45.
D.P. Stern, Am. J. Phys. 38 (1970) 494.
H.R. Brown, A. Elby, R. Weingard, in: J.T. Cushing et al. (Eds.), Bohmian Mechanics and
Quantum Theory: An Appraisal, Kluwer, Dordrecht, 1996, p. 309.
C. Philippidis, C. Dewdney, B. Hiley, Nuovo Cimento B 52 (1979) 15.
P. Holland, C. Philippidis, Phys. Rev. A 67 (2003) 062105.
J.G. Muga, C.R. Leavens, Phys. Rep. 338 (4) (2000) 353.
U. Schwengelbeck, F.H.M. Faisal, Phys. Lett. A 199 (1995) 281.
F.H.M. Faisal, U. Schwengelbeck, Phys. Lett. A 207 (1995) 31.
R.H. Parmenter, R.W. Valentine, Phys. Lett. A 210 (1995) 1; Phys. Lett. A 213 (1996) 319; Phys.
Lett. A 227 (1997) 5.
G. Iacomelli, M. Pettini, Phys. Lett. A 212 (1996) 29.
G.G. de Polavieja, Phys. Rev. A 53 (1996) 2059.
H. Wu, D.W.L. Sprung, Phys. Lett. A 261 (1999) 150.
A.J. Makowski, P. Peploswski, S.T. Dembinski, Phys. Lett. A 266 (2000) 241.
A.J. Makowski, M. Frackowiak, Acta Phys. Pol. B 32 (2001) 2831.
P. Falsaperla, G. Fonte, Phys. Lett. A 316 (2003) 382.
D.A. Wisniacki, F. Borondo, R.M. Benito, Europhys. Lett. 64 (2003) 441.
J.A. de Sales, J. Florencio, Phys. Rev. E 67 (2003) 016216.
J.M. Ottino, The Kinematics of Mixing: Stretching, Chaos, and Transport, Cambridge University
Press, Cambridge, 1989.
A. Crisanti, et al., Riv. Nuovo Cimento 14 (12) (1991) 1.
J.M. Finn, D. del-Castillo-Negrete, Chaos 11 (2001) 816.
J.R. Wilson, G.J. Mathews, Relativistic Numerical Hydrodynamics, Cambridge University Press,
Cambridge, 2003.
J.F. Nye, M.V. Berry, Proc. Roy. Soc. Lond. A 336 (1974) 165.
P.R. Holland, Found. Phys. 17 (1987) 345.
P.R. Holland, in: J.T. Cushing, et al. (Eds.), Bohmian Mechanics and Quantum Theory: An
Appraisal, Kluwer, Dordrecht, 1996, p. 99.
T. Takabayasi, Suppl. Prog. Theor. Phys. 4 (1957) 1.
I. Bialynicki-Birula, Acta Phys. Pol. B 26 (1995) 1201.
I. Bialynicki-Birula, in: E. Infeld, et al. (Eds.), Nonlinear Dynamic, Chaotic and Complex Systems,
Cambridge University Press, Cambridge, 1997.
I. Bialynicki-Birula, Z. Bialynicki-Birula, Phys. Rev. A 67 (2003) 0621141.
P.R. Holland, Phys. Rep. 169 (1988) 293.
P. Holland, Phys. Rev. A 60 (1999) 4326.
P. Holland, Ann. Phys. (Leipzig) 12 (2003) 446.
M.M. Ali, A.S. Majundar, D. Home, S. Sengupta, Phys. Rev. A 68 (2003) 042105.
N. Padhye, P.J. Morrison, Phys. Lett. A 219 (1996) 287.