The origin of the algebra of quantum operators in the stochastic

formulation of quantum mechanics∗

Mark Davidson†
arXiv:quant-ph/0112099v1 18 Dec 2001

Abstract. The origin of the algebra of the non-commuting operators of quantum mechanics
is explained in the general Fényes-Nelson stochastic models in which the diffusion constant is
a free parameter. This is achieved by continuing the diffusion constant to imaginary values,
a continuation which destroys the physical interpretation, but does not affect experimental
predictions. This continuation leads to great mathematical simplification in the stochastic
theory, and to an understanding of the entire mathematical formalism of quantum mechanics.
It is more than a formal construction because the diffusion parameter is not an observable
in these theories.

1 Introduction
The complementarity view of quantum mechanics dominates in most circles of modern
physics. The wave-particle dualism of this interpretation, and the assertion that a wave
function describes the state of a particle completely has been a source of concern and mys-
tery for many who have studied the subject. The stochastic formulation offers a possible
alternative to the complementarity view. Quantum particles move along continuous trajec-
tories in the stochastic picture, and quantum averages are ensemble averages over the space
of trajectories. The subject is in a state of development, and it is too early to tell whether
the present stochastic models will lead to a satisfactory explanation of quantum mechanical
laws, but this prospect justifies further exploration.
A history of stochastic models of quantum mechanics has been given by Jammer [1]. Fritz
Bopp, whose numerous contributions are reviewed by Jammer [1], has laid a philosophical
and physical foundation for this theory. Imre Fényes [2] is credited with the first proof that
Schrödinger’s equation could be understood as a kind of diffusion equation for a Markov
process. Edward Nelson [3, 4] greatly elaborated on the work of Fényes, and put the theory
on a much more rigorous footing. At about the same time, Favella [5] made strides in
understanding the mathematics of quantum mechanics in terms of diffusion processes. Since
then, de la Pena-Auerbach [6, 7] has made numerous contributions. An excellent review has
∗ c 1979 by D. Reidel Publishing Company,
Letters in Mathematical Physics 3 (1979) 367–376. Copyright
Dordrecht, Holland, and Boston, U.S.A.
†
Current Address: Spectel Research Corporation, 807 Rorke Way, Palo Alto, CA 94303
Email: [email protected], Web: www.spectelresearch.com

1
been given by Diner and Claverie [8]. Related models have been proposed by Wiener and
Siegel [9], Della-Riccia and Wiener [10], Bohm and Vigier [11], Moyal [12], and many others.
A continuum of Fényes-Nelson type models, each with different diffusion parameter,
are possible [13]. The diffusion parameter is constrained to be positive, but is otherwise
undetermined. This freedom allows one to chose the diffusion parameter at will. In this
paper, this freedom is exploited, and the diffusion parameter is continued to the complex
plane. Although the mathematics of the diffusion theory can be continued in this way,
the physical interpretation of the theory is lost. The mathematics simplify greatly at one
special complex value of the diffusion parameter, and this is the reason for considering this
continuation.
Within the Fényes-Nelson framework, it is possible to construct a Hilbert space which
plays essentially the same role as the Hilbert space of states in quantum mechanics. It is also
possible to introduce non-commuting operators for position, velocity, acceleration, etc. [14].
For real values of the diffusion parameter, the commutation rules for these operators are
different from the corresponding quantum operators. However, when the continuation to the
complex plane is performed, one finds that the algebra of the familiar quantum operators may
be recovered. The various quantum expectations may be considered as analytic continuations
to complex diffusion parameters of stochastic expectations. The hypothesis that measurable
quantities are independent of the diffusion parameter allows this transformation, and makes
it more than just a formal construction.

2 The algebra of quantum operators

Consider a diffusion process defined by:
d~x = ~b(x, t)dt + dW
~ (1)
~ is a Wiener process satisfying:
where ~x is the position of a particle, and W
E(dWi (t)dWj (t)) = 2νδij dt, (2)
where ν is a positive constant, the diffusion parameter, with dimensions (length)2 /time. The
process in eqn. (1) was studied by Nelson [3, 4] who established a strong connection with
quantum theory.
~x is known to be a Markov process for sufficiently regular ~b, but the most general regular-
ity conditions are not known. Therefore, complete mathematical rigor shall not be attempted
here. It will be assumed that ~b is sufficiently well behaved so that ~x is a Markov process.
Let (Ω, Σ, P ) be the underlying probability space for ~x, and for ω ∈ Ω, x(t, ω) is a sample
trajectory as t varies with ω fixed. These sample trajectories are continuous almost surely
if ~b is sufficiently regular, and this property shall be assumed. The process ~x satisfies:
!
~x(t + h) − ~x(t) 

lim E ~
x(t) = x = b(x, t) (3)
h→0+ h
where the notation E(·|·) denotes conditional expectation. Also,
1
lim E((xi (t + h) − xi (t))(xj (t + h) − xj (t))|x(t) = x) = 2νδij . (4)
h→0+ h

2
Since ~x is a Markov process, a Markov transition function may be defined:
1
P (x, t; y, s) = 3lim P (~x(t) ∈ d3 x|x(s) = y), t>s (5)
d x→0 d3 x
which satisfies a Chapman-Kolmogorov equation:
Z
P (x, t; y, s) = d3 zP (x, t; z, u)P (z, u; y, s), t > u > s. (6)

Continuity of the sample paths requires:

lim P (x, t; y, s) = δ 3 (x − y). (7)

t↓s

The backward equation of Kolmogorov may be derived, formally, by choosing u in eqn. (6)
to be close to s. Then one can expand:
~ y P (x, t; y, u) +
P (x, t; z, u) = P (x, t; y, u) + (~z − ~y ) · ∇
1 ∂2
+ (z − y)i (z − y)j P (x, t; y, u) + . . . . (8)
2 ∂yi ∂yj

Substitution of eqn. (8) into eqn. (6), using eqns. (3) and (4) yields
" #
∂ ~ ~ y + ν∆y P (x, t; y, s) = 0
+ b(y) · ∇ (9)
∂s

which is the backward equation. Differentiating eqn. (6) with respect to u, using eqn. (9),
integrating by parts, and taking u → t, one finds the forward equation:
" #
∂ 
~ x · ~b + ~b · ∇

~ x − ν∆x P (x, t; y, s) = 0.
+ ∇ (10)
∂t

Nelson [3, 4] also considers the time reversed process:

d~x = ~b∗ dt + dW
~ ∗,
!
x(t) − x(t − h)

~b∗ = lim E 
x(t) = x ,
h→0+ h 

E(dW∗i dW∗j ) = 2νδij dt. (11)

He shows that, independent of any dynamical assumption, one has

(~b − ~b∗ )/2 = ν ∇

~ ln (ρ), ρ = probability density. (12)

He defines forward and backward time derivatives, D and D∗ , by

1
Df (x, t) = lim
E(f (x(t + h)) − f (x(t))|x(t) = x) (13)
h→0+ h
1
D∗ f (x, t) = lim E(f (x(t), t) − f (x(t − h), t − h)|x(t) = x) (14)
h→0+ h

3
and D and D∗ are found to be:
∂ ~ ~ ∂ ~ ~
D= + b · ∇ + ν∆, D∗ = + b∗ · ∇ − ν∆. (15)
∂t ∂t
The mean acceleration is defined by
1
~a = [DD∗ + D∗ D]~x. (16)
2
Equating m~a to the force leads to Schrödinger’s equation
2
" #
~ → − h̄ ∆ + V exp(R + iSN ) = ih̄ ∂ exp(R + iSN ),
m~a = −∇V (17)
2m ∂t
where
~ × ~b = 0, ν = h̄ , ρ = exp(2R), ~b = 2ν ∇(R
∇ ~ + SN ). (18)
2m
The dynamical assumption, eqn. (17), is not unique [13]. If
~
m~a + m(β/8)(D − D∗ )2~x = −∇V, (19)

then one obtains [13]:

h̄2
" #
∂
− ∆ + V exp(R + izSN ) = ih̄ exp(R + izSN ), (20)
2m ∂t
provided that
~ × ~b = 0,
∇ ~b = 2ν ∇(R
~ + SN ), ρ = exp(2R),
h̄ q
ν=z , z = 1/ 1 − β/2. (21)
2m
In eqn. (20), zSN should be considered fixed as z and ν range over their possible values.
Since β is a free parameter, ν can take on any real value from 0 to ∞. In order that z be
real, we must restrict β < 2, from (21).
Next, a Hilbert space is introduced, with operators corresponding to dynamical variables
[14] (note that ν in Ref. 14 differs from ν here by a factor of 2). Let Ht be the Hilbert space
of complex functions, f (x), x ∈ R3 , with inner product given by:
Z
(f, g) = E(f ∗ (x)g(x)) = d3 xρ(x, t)f ∗ (x)g(x). (22)

Operators for ~ẋ, ~ẍ, etc., shall now be defined. First ~ẋ:

~ẋf = lim ∂ E(~x(u)f (x(t))|x(s) = x), t > u > s. (23)

u ↑ t ∂u
s↑u

This definition is, for practical purposes, equivalent to the one used in Ref. 14. The domain
and range of ~ẋ will depend on the properties of the Markov process in question. A detailed

4
investigation of these will not be made. Instead, regularity assumptions will be pointed out
along the way to a formal derivation of ~ẋ. ~ẋ may be evaluated, for a certain class of functions
f , by using the Markov transition function:
∂ ∂
Z
E(~x(u)f (x(t))|x(s) = x) = d3 yd3zf (z)~y
P (y, u; x, s) P (z, t; y, u)
∂u Z ∂u
~ y · ~b(y, t) − ~b(y, t) · ∇
~ y + ν∆y P (y, u; x, s)+
n h   i
= d3 yd3 xf (z)~y P (z, t; y, u) − ∇

+P (y, u; x, s) −~b(y, t) · ∇
h i
~ y − ν∆y P (z, t : y, u) } (24)

where it has been assumed that the order of integration and differentiation can be freely
interchanged in (24), and the forward and backward equations (eqns. (10) and (9)) have
been used to reexpress the u derivative above. Integration by parts yields:
Z
~ y · ~b(y, t) + ~b(y, t) · ∇
~ y − ν∆y P (z, t; y, u)
h  i
= d3 yd3zf (z)P (z, t; y, u) ∇ (25)

where [·, ·] denotes commutator. In the limit u ↑ t, P (z, t; y, u) → δ 3 (z −y), and therefore,
assuming that this limit can be taken inside the integral:
Z
~ y · ~b(y, t) + ~b(y, t) · ∇
h  i
~ẋf (x) = lim d3 yf (y) ∇ ~ y − ν∆y , y P (y, t; x, s). (26)
s↑t

Another integration by parts yields:

Z
d3 yP (y, t; x, s) ~y , −~b(y, t) · ∇ ~b + ν ∇
h i h  i
~ẋf (x) lim ~ y − ν∆y f (y) = ~ · ∇,
~ ~x f (x) (27)
s↑t

so that
~ẋ = ~b + 2ν ∇.
~ (28)
Operators for higher time derivatives are calculated in a similar fashion. One finds

~b + ν ∇ ~ ~xn + ∂ ~xn
~ · ∇,
h  i
~xn+1 = (29)
∂t
where ~xn denotes the nth time derivative:
∂n
~xn f (x) = lim E(~x(u)f (x(t))|x(s) = x). (30)
u↑t ∂un
s↑u

In particular, for ~ẍ, one finds

~
~b + ν ∇ ~ ~b + 2ν ∇
~ · ∇, ~ + ∂b
h
 i
~ẍ = (31)
∂t
~ẍ = ∂ ~b + ν ∆~b +
  1  
~ × ~b) × (~b + 2ν ∇).
~ ~b2 + (∇ ~
∇ (32)
∂t 2
~ × ~b = 0, as in eqn. (21), then
If ∇
~
~ẍ = ∂ b + ν∆~b + 1 (∇
~ ~b2 ). (33)
∂t 2
5
It is desirable to express ~ẍ in terms of the potential V . To do this, using eqn. (21), rewrite
(33) as: " !#
∂
 
~ exp(−R − SN ) 2ν + 2ν ∆ exp (R + SN ) .
~ẍ = ∇ 2
(34)
∂t
Now, using eqns. (20) and (21), and after some tedious algebra, one finds:
! √
~ ~ ~ h̄2 2 ∆ ρ
mẍ = −∇V + ∇ + 2mν √ . (35)
2m ρ

The operators for ~x and ~ẋ do not commute.

[ẋi , xj ] = 2νδij ; [xi , xj ] = 0; [ẋi , ẋj ] = 2ν(∂i bj − ∂j bi ), (36)

where all velocities do commute if ∇ ~ × ~b = 0, as in eqn. (21).

If F (x, ẋ) is any ordered polynomial function of the operators ~x and ~ẋ, then one has
Z
(g, F (x, ẋ)h) = d3 xρ(x, t)g ∗ (x)F (x, ~b + 2ν ∇)h(x)
~
Z
= ~
d3 x exp(R − SN )g ∗ (x)F (x, 2ν ∇)h(x) exp(R + SN ). (37)

~ if the Hilbert space

This last expression shows that ẋ takes a particularly simple for (2ν ∇)
Ht is mapped onto a new Hilbert space, call it It , by the mapping T : T f = exp(R + SN )f ,
and where the inner product on It , denoted by ((·, ·)) preserves the norm, i.e.:

(f, g) = ((T f, T g)). (38)

Clearly, this requires Z

((f, g)) = d3 x exp(−2SN )f ∗ (x)g(x). (39)

From (37), it is clear that T maps ~ẋ onto the operator:

~
T ~ẋ = 2ν ∇, or ~ f.
T (~ẋf ) = 2ν ∇T (40)
~ × ~b = 0, because ~ẍ is simply
The acceleration operator, ~ẍ, remains invariant under T if ∇
a multiplicative function on x in this case (see eqn. (33)). Higher time derivatives may be
calculated by use of the formula:
∂ 1
T ~xn+1 = T ~xn + [H, T ~xn ] ;
∂t 2mν ! √
1 (2mν)2 h̄2 2 ∆ ρ
H = ∆+V − + 2mν √ (41)
2 m 2m ρ

which follows by considering eqn. (29) under the mapping T . T ~xn denotes the image of ~xn
under T , i.e.: T (~xn f ) = (T ~xn )T f . From here on, the following notation shall be used:
~ n = T ~xn
X (42)

6
and it follows from the above that
(g, F (x, ẋ)h) = ((T g, F (X, Ẋ)T h)). (43)
~ + s) by:
It is possible to define an operator X(t
∞
~ + s) = ~ n (sn /n!),
X
X(t X (44)
h=0

and it may be shown that

E(x(t+s1 )×. . . ×x(t+sn )) = ((exp(R+SN ), X(t+s1 )×. . . ×X(t+sn ) exp(R+SN )), (45)
where si ≤ si+1 . Note that the operators of eqn. (44) do not commute for different s,
and therefore the ordering matters for the right-hand side of eqn. (45). Equation (45) is
essentially the Feynman-Kac formula. Its derivation is tedious and will be omitted, but it
can be proved by expressing the left-hand side in terms of the Markov transition function
and using the forward and backward equations.
H in eqn. (41) depends on ρ, and this fact complicates the expression for X(t + s). Note
that if ν is allowed to be imaginary, then
h̄ 1 h̄2
ν = ±i →H= ∆+V (46)
2m 2m
and the term involving ρ in H becomes zero. This is a major simplification. Although an
imaginary value of ν is physically meaningless, we may allow it to be imaginary without
affecting measurable results, since ν is a free parameter, and all measurable quantities are
independent of ν, or at least this is the hypothesis.
Consider eqn. (46). If ν is allowed to take on this imaginary value, then the commutation
rules become: h i
[ẋi , xj ] = Ẋi , Xj = ±ih̄δij . (47)
From eqn. (21) one has in this case z = ±i, and since zSN is fixed, SN becomes imaginary:
zSN = S; SN = ∓iS; S real. (48)
The fact that SN is imaginary affects the inner product on the space It . Equation (39) is no
longer applicable, but rather one finds:
Z
((f, g)) = d3 xf ∗ g (49)

~ becomes:
so that in this case, It is just L2 . The operator Ẋ
~ = ±ih̄/m∇;
Ẋ ~ ~ = ±ih̄∇
P~ = mẊ ~ (50)
and this agrees with the familiar quantum mechanical result if the minus sign is chosen
above. The equation for time differentiation, eqn. (42), becomes:

~ n+1 = ∂ X~ n ∓ i H, X
h
~n ;
i h̄2
X H=− ∆+V (51)
∂t h 2m
7
which agrees with the quantum mechanical result for the analogous operators.
The operators X(t + s) are found to be the familiar Heisenberg position operators:
i i
   
X(t + s) = exp ∓ Hs X exp ± Hs , (52)
h̄ h̄
where it has been assume that V does not depend explicitly on time in arriving at (52). The
Feynman-Kac formula (eqn. (45)) becomes:
Z
Ec (x(t + s1 ) × . . . × x(t + sn )) = d3 x exp(R ± iS) × . . .
. . . × X(t + s1 ) × . . . × X(t + sn ) exp(R ∓ SN ), (53)
where si ≤ si+1 , and where Ec denotes a continuation, to complex ν = ±ih̄/2m (either sign is
possible), of the Markov expectation. Ec is in general complex, but not directly measurable.
If all of the times in (53) are the same time, then Ec becomes real, and equal to the real
Markov expectation, which in this case is independent of ν.
It is clear that the mathematical formalism of ordinary quantum mechanics is recovered
if the minus sign of eqn. (46) is chosen. The interpretation is similar also. Care must be
taken to compare the theory with measurable quantities, thereby avoiding the ambiguities
of an interpretation of complex expectations in eqn. (53). The underlying reason for the
algebra of quantum operators is explained, however, in the stochastic formulation of quantum
mechanics. It has been derived here from the postulate that quantum mechanics is equivalent
to a class of Markov processes with diffusion constant a free parameter. The continuation
to imaginary ν which leads to complex expectations is simply a convenient artifice, which
facilitates calculations without affecting measurable results.

3 Conclusion
In the stochastic formulation of quantum mechanics, simplicity is achieved by exploiting the
indeterminate nature of the diffusion constant, and choosing it to be a particular imaginary
value. Although the physical interpretation is lost, or at least obscured, by this, experimen-
tally measurable averages are not affected, or at least this is the postulate which justifies
the continuation, and it does not seem obviously false. Perhaps a way to measure ν will
someday be found, but this would require a measurement which goes beyond the ordinary
predictions of quantum theory.
The central question, not discussed here, is that of the origin of the diffusion laws under-
lying quantum theory. Several models have been proposed [15]–[19]. Although the definitive
explanation has not yet been found, one possibility seems to stand out: quantum mechanics
may arise out of the interaction of charged particles with random forces in the vacuum, and
with radiative forces playing an important role. This random force may be due, at least in
part, to the existence of the random radiation of stochastic electrodynamics [20].

References
[1] Jammer, M., The Philosophy of Quantum Mechanics Wiley, New York, 1974.

8
 [2] Fényes, I., Zeitschrift für Physik. 132, 81–106, (1952).
[3] Nelson, E., Phys. Rev. 150, 1079–1085 (1966).
[4] Nelson, E., Dynamical Theories of Brownian Motion (Princeton University Press,
Princeton, 1967).
[5] Favella, L.F., Ann. Inst. Henri Poincaré 7, 77–94 (1967).
[6] de la Pena-Auerbach, L., Phys. Lett. 24A, 603–604 (1967); Ibid . 27A, 594–595 (1968);
Revista Mexicana de Fisica 19, 133–145 (1970); Phys. Letters 31A, 403–404 (1970).
[7] de la Pena-Auerbach, L., and Cetto, A.M., Phys. Letters 29A, 562–563 (1969); Revista
Mexicana de Fisica 18, 253–264 (1969); Phys. Rev. D3, 795–800 (1971) .

[8] Diner, S., and Claverie, P., ‘Statistical and Stochastic Aspects of the Delocalization
Problem in Quantum Mechanics’. In O. Chalvet et al . (Eds.), Localization and Delocal-
ization in Quantum Chemistry, Reidel, Dordrecht, 1976.

[9] Wiener, N., Siegel, A., Rankin, B., and Martin, W.T., Differential Space, Quantum
Systems, and Prediction, M.I.T. Press, Cambridge, 1966.

[10] Della-Riccia, G., and Wiener, N., J. Math. Phys. 7, 1372 (1966).
[11] Bohm, D., and Vigier, J.P., Phys. Rev . 96, 208 (1954).
[12] Moyal, J.E., Proc. Camb. Phil. Soc. 45, 99–124 (1949).
[13] Davidson, M., ‘A Generalization of the Fényes-Nelson Stochastic Model of Quantum
Mechanics,’ Lett. Math. Phys. 3, 271–277.
[14] Davidson, M., ‘A Dynamical Theory of Markovian Diffusion’, to appear in Physica A,
1979.
[15] de la Pena-Auerbach, L., and Cetto, A.M., J. Math. Phys. 18, 1612–1622 (1977), and
references therein.

[16] Santos, E., Nuovo Cimento B19, 57 (1974); Ibid . B22, 201 (1974).
[17] Braffort, P., Surdin, M., and Taroni, T., C. R. Acad. Sci. Paris 261, 4339 (1965).
[18] Marshall, T.W., Proc. Roy. Soc. 276A, 475 (1963); Proc. Cambridge Philos. Soc 61,
537 (1965); Nuovo Cimento 38, 206 (1965).
[19] Davidson, M., ‘A Model for the Stochastic Origins of Schrödinger’s Equation’, to appear
in Journal of Mathematical Physics, 1979.

[20] Boyer, T. H. Phys. Rev. D11, 790, 809 (1975), and references therein.

(Received June 1, 1979)