PH YSI CAL R EVI EW VOLUME 122, NUMBER 8 JUNE |,

Time in the Quantum Theory and the Uncertainty Relation for Time and Energy
Y. AHARowov* Axe D. BoHM
II. H. IVills Physics Laboratory, Bristol, England
(Received September 7, 1960)

Because time does not appear in Schrodinger's equation as an operator but only as a parameter, the time-
energy uncertainty relation must be formulated in a special way. This problem has in fact been studied by
many authors and we give a summary of their treatments. We then criticize the main conclusion of these
treatments; viz. , that in a measurement of energy carried out in a time interval, ht, there must be a minimum
uncertainty in the transfer of energy to the observed system, given by 6(E' E) &~h/n—i We s. how that this
conclusion is erroneous in two respects. First, it is not consistent with the general principles of the quantum
theory, which require that all uncertainty relations be expressible in terms of the mathematical formalism,
i.e., by means of operators, wave functions, etc. Secondly, the examples of measurement processes that were
used to derive the above uncertainty relation are not general enough. We then develop a systematic presenta-
tion of our own point of view, with regard to the role of time in the quantum theory, and give a concrete
example of a measurement process not satisfying the above uncertainty relation.

1. HISTORICAL SUMMARY OF THE STATE OF THE where 1ta(x) is the eigenfunction of the Hamiltonian H
PROBLEM OF TIME MEASUREMENT IN of the system belonging to the eigenvalue E, and Cz is
THE QUANTUM THEORY
an arbitrary coefficient. If we consider a wave packet of
width DE in energy space (i.e., AE is the range in which
A S is well known, the uncertainty relations in
quantum mechanics can be regarded in two closely ~CE~ is appreciable), it immediately follows from the
related ways. First of all, they are a direct mathematical properties of Fourier analysis that (DE)r&~h, where r
consequence of the replacement of classical numbers by is the time during which the wave packet does not
operators, and of adding the basic principle that the change significantly (r may be regarded as the mean life
statistical distributions of the corresponding observables of the state in question').
can be obtained by means of the usual formulas from The above is a discussion in terms of the Schrodinger
the wave function and its probability interpretation. ' representation. Mandelstamm and Tamm4 have formu-
Secondly, however, it can be shown by analyses such lated what is, in essence, the same point of view but it is
as that of the Heisenberg microscope experiment that expressed in the Heisenberg representation. They con-
they are also limitations on the possible accuracy of sider a dynamical variable 3, which is a function of the
measurements. ' time (e.g. , the location of the needle on a clock dial or
These considerations apply to observables such as the position of a free particle in motion) and which can
x, p, and H. With regard to the measurement of time, therefore be used to indicate time. If D.4 is the un-
however, a further problem appears, because time certainty in 3, then the uncertainty in time is
. enters into Schrodinger's equation, not as an operator
Sf = ZA/ (A), (,
(i.e. , and "observable" ) but rather, as a parameter, ~

which is a "c" number that has a well dined value. provided that A does not change significantly during
Nevertheless, the uncertainty principle, AEht~&h, is the time, At, and that DA/~ (A. ), is negligible. From
~

generally accepted as valid, even though it is not the relation

deduced directly from commutation relations in the
way described above.
we obtain
aA aH& ~((A, H)). =h~(A). (,.( .
The justification of the time-energy uncertainty
relationship has been attempted in several ways. (We AA AH/(A). = 2 f AH ~&Ir.
shall restrict ourselves here entirely to a discussion of Since II represents the Hamiltonian of the isolated
the nonrelativistic case, since the theory of relativity system, AII is also equal to AE, the uncertainty in
has no essential relationship to the measurement energy of that system.
problems that we are going to treat in this paper. ) It should be noted that the method proposed by
First, one can begin with the wave function Mandelstamm and Tamm can actually lead to a deter-
'"'i" mination of time, only when the system is not in a
1f (x, f) =g@ Ca/a(x)e (1) stationary state; i.e. , only when the wave function takes
*Now at Brandeis University, Waltham, Massachusetts. the form of a packet, consisting of a linear superposition
'The uncertainty relations are obtained in this way using
Schwarz's inequality with the expressions for of stationary states. In other words, the AII appearing
((A — .
A. )'). ((8 — ,
8)s). = (nA)'(AB) s.
See, for example, D. Bohm, Quuntuns Theory (Prentice-Hall, Inc
in equation (2) is determined by the range of energies
s V. Fock and N. Kry1ov, J. Phys. (U. S.S.R. 11, 112 (1947))
)
Englewood Cli6s, New Jersey, 1951), pp. 205 —206. present a more detailed account of the lifetime of a state.
2 W.
Heisenberg, The Physical Princi ples of the Quantum Theor y 4I. Mandelstamm and I. Tamm, J. Phys. (U. S.S.R.) 9, 249
(Dover Publications, New York, 1930), Chap. 2. (194S).
f649
1650 Y. AHA RONOV AN D D. BOHM
in the wave packet. In this way, the relation of the is still a minimum uncertainty in energy transfer, be-
Mandelstamm and Tamm treatment to the Schrodinger cause the change of velocity will then become very
representation is made clear. large, if At is finite. )
The above is then a discussion of the relation Landau and Peierls therefore conclude that there is an
AE ht~&h insofar as this has been obtained from the uncertainty relation between the energy transferred to
ma, thematical formalism of the quantum theory (i.e. , the system and the time at which the energy is
the wave function, operators, and probability inter- measured. This means that the energy of the observed
pretation). Naturally, as is necessary in the case of system cannot be measured in a short time, without
observables such as x and p, this uncerta, inty relation changing it in an unpredictable and uncontrollable way.
must also be analyzed in terms of the interaction of the In other words, energy meuslremeets curried out in short

and Peierls, "

measuring apparatus with the observed system. Landau
for example, do this by considering a
special example, in which the momentum of a free
periods of time are not reproducible.
Fock and Krylov' criticize the derivation of the above
results, but come to essentially the same conclusion. In
particle is measured by means of a collision with a heavy effect, they do not accept the definition of the time of
test particle (also free). To simplify the problem they measurement by means of a time-dependent potential
consider a case in which the measuring particle is a of interaction between the two particles. This is because
perfectly rejecting mirror, and discuss only the move- in a real collision, there is no such time-dependent
ment in one dimension (perpendicular to the mirror). potential. Rather, the time of collision is determined by
They then apply the laws of conservation of energy and the movement of the particles themselves, in such a way
momentum, which are that one of them serves as a clock. Let us suppose that
p'+P' (p+P) =—0, (3a) it is the test particle which fulfills this function. This
particle defines the time, t, as that at which it passes a
E'+e' (E+e) =0, — (3b) definite point, X, by the equation, t= X/V. The
time, as defined in this way, has an uncertainty
where lower case letters refer to the observed particle,
LU= d, X/V (provided, as will actually be the case in our
capitals to the test particle, unprimed quantities to
example, that AV/V((l).
values before collision, and primed quantities to values
after collision. Because E=P'/2M, e= p'/2m, one can In order to define the time of collision, we must have
solve for the momentum of the observed particle before some information about the initial location of the ob-
and after collision, in terms of the corresponding served particle, as well as that of the test particle. For
momenta of the test particle. simplicity, let us suppose that the initial velocity of the
In order to define the time of measurement, Landau test particle is so much higher than that of the observed
and Peierls' (and also Landau and Lifshitz)' consider particle that the latter can be regarded as essentially at
the case of a time-dependent interaction between the rest until the collision. The mean initial position x of the
particle and the mirror, which lasts for some known observed particle will be taken to be at the origin, while
the uncertainty in this position is represented by Ax.
period of time, At. This period At, which is the un-
certainty in the time of measurement, then implies Evidently we must choose Ax & Ax, if the location of the
(e.g. , according to perturbation theory) an uncertainty test particle is to serve as a definition of the time of
in the energy of the combined system consisting of collision. Therefore
observed particle and mirror, of magnitude h/ht,
resulting from the time-dependent interaction. Instead d, t= AX/V)~A/(dP) V =i'i/5E.
of Eq. (3b) for the exact conservation of energy, we must
therefore write (This is just the well-known uncertainty relation be-
i
e'+E' —(e+E) i
& h/Dt. (3c) tween the energy of the test particle and the time that is
defined by the movement of its coordinate. )
Evidently the momentum of the test particle before Fock and Krylov then point out that in this case, the
and after collision can be measured with arbitrary laws of conservation of energy and momentum are both
accuracy, so that AP=AP'=0. As a result, we obtain satisfied exactly, so that Eqs. (3a) and (3b) can be used
from Eq. (3a), 6p=hp', and from (3c), we have directly, while the approximate form (3c) for the con-
2 (e' —e) &~h/At servation of energy is not to be applied here (since
perturbation theory no longer has any relevance to the
Since e= p'/2m, we can also write the above result as
("—v)ap& h/at.
(Note that although Ap itself may be very small, there
problem. ) From (3a), we obtain

~(p p') = ~(P P')— —

If V is chosen large enough, we can for a given
5 L. Landau and R. Peierls, Z. Physik 69, 56 (1931). dd&~h/VDP make DP and hP' arbitrarily small, and,
6 See also L. Landau and E. Lifschitz, Qnantlnz 3feclzanics
(Pergamon Press, New York, 1958), pp. 150—153. a, s a result, we can likewise make h(p —
p') as small as
 QUANTUM THEORY FOR TIME AND ENERGY 165i
we please. From Eq. (3b), it then follows that and Peierls, ' as well as by Landau and Lifshitz') which
applies to individual measurements, and which refers to
the relation between the error in the measurement of
energy and the duration of the measurement process.
While criticizing some of the methods of Landau and
where we have used the result that A(p' — p) is negli- Peierls, they agree with the essential conclusion that
gible. TlIe above is exactly the same Eq. (5) as that energy cannot be measured in arbitrarily short periods
obtained by Landau and Peierls, but the uncertainty in of time, without introducing uncertainties, according to
energy transfer to the observed system is now deduced the relation A(e' — e) ~& h jhow.
on the basis of the fact that the (time-dependent) posi-
tion of one of the particles is used to define the time of 2. CRITICISM OF COMMONLY ACCEPTED
collision. INTERPRETATIONS OF THE TIME-ENERGY
UNCERTAINTY RELATION
Fock and Krylov then go on to criticize the approach
of Mandelstamm and Tamm, suggesting that it is in- The main conclusion of Landau and Peierls, ' Landau
complete. They assert that by means of the wave func- and Lifschitz, ' and Fock and Krylov' (given in Sec. 1),
tion and the operators of the observed system, one can viz. , that energy cannot be measured in a short time
discuss only the statistical features of any measure- without introducing an uncertainty in its value,
ment. In order to discuss an ieCkviduu/ measurement represents a very widely accepted interpretation of the
process, they refer to what they call "Bohr's uncertainty time-energy uncertainty relation. This conclusion is, as
relation," A(e' — e)At& h, where e and e' are the actual we shall show, erroneous, the error being based in part
values of the energy of an individual observed system on an inadequate formulation of Bohr's point of view
before and after measurement. concerning measurement, and in part on the use of an
To clarify this distinction between the statistical illustrative example of a measurement process, that was
uncertainty relations discussed by Mandelstamm and not sufficiently typical of the general case of such a
Tamm, and the Bohr relation, they point out that, for measurement. (In fact, in Sec. 4 we shall give a counter-
example, in observation of a state with lifetime r (as example, in which the energy of a particle is measured to
described by its wave function), one can make measure- arbitrary accuracy in as short a time as we please).
ments in times much shorter than r. Therefore, it is With regard to Bohr's point of view concerning
necessary to distinguish between the time intervals measurements, it is important to stress here his con-
defined by the wave function of the observed system, tinual insistence that the minimum ambiguities in the
and the time interval representing the actual duration results of any individual measurement process (in the
of an individual measurement. The time interval defined sense of what Fock and Krylov called the "Bohr un-
by the wave function has in measurements generally certainty relations" ) are always exactly the same as the
only a statistical significance. minimum ambiguities in the possibility of definition in
Even if one treated the measurement process by the mathematical theory of the observables that are
means of a many-body Schrodinger equation, including being measured. ' (These latter ambiguities are, of course,
the apparatus coordinates, the same distinction would the "statistical" uncertainty relations referred to by
arise. For it would the necessary to observe the com- Fock and Krylov. )
bined system by means of additional apparatus; and The ambiguities in the results of individual measure-
here too, there will be a "Bohr uncertainty principle" ments are regarded as originating in the indivisible
for the individual observation and a statistical un- quantum connections of the object under investigation
certainty principle following from the wave function, to the apparatus (and indeed to the whole universe),
which applies to an ensemble of cases. To treat the which give rise to a minimum ambiguity in the degree to
apparatus by quantum theory is, in effect, to push back which well-defined classical properties (e.g. , position
the well known "cut" between classical and quantum and momentum) can be assigned to the object as a
sides another stage. While it is always permissible and result of any such measurement. ' In addition, however,
sometimes convenient to do this, it cannot change the the result of each measurement defines a "quantum
content of the theory. state" of the observed system, specified by its wave
Let us now sum up the problem. Mandelstamm and function. This wave function is, of course, not a repre-
Tamm propose a mathematical operator uncertainty sentation of an individual system, but it implies, in
relation between energy and time, as determined by general, only a set of statistical predictions concerning
the wave function. Fock and Krylov regard such a treat- the results of possible measurements. Nevertheless, if
ment as incomplete, because it applies only statistically these predictions were such that the minimum ambi-
to a large number of measurements, and because within guity in the definition of the results of an individual
it one cannot even consider the question of the interval measurement were less than the minimum statistical
of time needed to carry out an individual measurement.
' N.Bohr, Die Naturwissenschaften, 251 (1928).
To complete the treatment, they call attention to the W. Heisenberg, Physics and Philosophy (Allen and Unwin,
"Bohr uncertainty relation" (discussed also by Landau London, 1959), Chap. 3.
i652

fluctuation implied by the mathematical theory, then ent of the statistical relations obtainable from the
there would be a contradiction. Vice versa, if they were mathematical formalism. As we have already pointed
such that the minimum ambiguity in the result of an out, however, such a procedure is arbitrary and there-
individual measurement were greater than the minimum fore subject to the continual danger of being contradicted
statistical fluctuation as described above, then this by the development of new examples of measurement
would lead to an arbitrary restriction, not related in any processes, which reduce the ambiguity down to the
general way to the mathematical formalism, a restric- minimum allowed by the formalism. (For, as is quite
tion that evidently has no place in a coherent over-all evident, there is no way to be sure that conclusions
framework of theory. Moreover, one could, in general, obtained from an illustrative example have universal
expect that with sufficient effort, it would be possible to validity).
find an example of an individual measurement process It follows from the above discussion that to complete
with the same minimum ambiguity as that implied by the treatment of the time-energy uncertainty relation, it
the formalism; and if such a process is found, then the is necessary to develop a method of showing how the
supposedly greater minimum ambiguity in the results of time of measurement and the energy transferred in this
an individual measurement will be contradicted. For measurement are to be expressed in terms of suitable
these zeasons, it is necessary to consider the statistical operators. The method that we shall use in this paper
and individual uncertainty relations as two equally starts from our discussion of the example first treated by
essential sides of what is basically the same limitation Landau and Peierls, and then by Fock and Krylov;
on the precise definability and measurability of the viz. , the one in which the energy of a free particle is
state of any system. In other words, as Bohr has measured by collision with another particle. As we saw
stressed, there car be eo Hmitatioe oe individual measure- in Sec. 1, in such an interaction, the time of collision is
ments that cannot also be obtained from the mathematical determined physically by the state of some system
formalism and the statistical interpretation. which serves as a clock. In the example of Fock and
There is no question that all the above considerations Krylov, it is determined by the position of the test
apply for common examples of the uncertainty principle particle (which was taken to be free). Now, for any
(e.g. , x and p). However, as we pointed out in Sec. 1, system, one can define a Hermitian operator represent-
time enters into Schrodinger's equation only as a ing such a time. In the case of a free particle, this is
parameter, so that there is no straightforward way to
1
apply these ideas to the time-energy uncertainty rela-
tion. Of course, we can, with Mandelstamm and Tamm,
t, = —
y
= M/ y
—',
p
—
1
+— y f,
pw
obtain an uncertainty relation between the lifetime of
a state of the observed system and its energy. But let us where y and p„are respectively position and momentum
recall here that (as pointed out by Fock and Krylov), of the particle in question. "
the operators of the observed system have no connection The commutation relations between the above opera-
whatsoever with the duration of the measurement tor and the Hamiltonian, H„of the "clock" in question
process (which is evidently determined, in general, by are (as can easily be verified for the case of a free
the apparatus). Keeping this fact in mind, let us now particle, for which H, = p„'( 352),
raise the question of whether there can be a genuine un-
certainty relation between the energy transferred to the
observed system and the time at which the measure- This procedure is evidently very similar to that of
ment took place (as has been suggested by I.andau and Mandelstamm and Tamm. However, they discussed
Peierls, Fock and Krylov, and other authors). only the operators of the observed system, and obtained
In accordance with Bohr's point of view on the sub- an uncertainty relation (2) between the energy of this
ject, as we have described it above, we are led to point system and the "inner" time as defined by dynamical
out that one cannot safely regard any given uncertainty variables in this system (e.g. , the lifetime of a state).
relations as representing a real limitation on the On the other hand, we are applying the relations (8) to
accuracy of all possible measurements of the quantities the energy, E„of the "clock" in the apparatus, and to
under discussion unless the relationship has been shown the time, t„of measurement as determined by this clock.
to follow from the mathematical formalism. On the Since the time of measurement can be represented by
other hand, all of the authors referred to above seem to an operator, t„belonging only to the observing appara-
be satisfied to establish the time-energy uncertainty tus, it follows that this time must commute with every
relations as applying to individual measurements by operator of the observed systeni and, in particular, with As
what Fock and Krylov called "illustrative examples. " Hamil/oriana. There is therefore no reason inherent in the
Such a point of view would imply, of course, that the principles of the quantum theory why the energy of a
uncertainty relations applying to individual measure- system cannot be measured in as short a time as we
ments could in principle, have a basis that is independ- ' There is a singularity for p„=0, but it is easily shown that this
will be unimportant if p„ is large enough, as will be the case in
N. Bohr (private communication). our applications.
 QUANTUM THEORY FOR TIME AND ENERGY 1653
/

please. (Recall, however, that in accordance with the momentum. It is evident that the uncertainty in this
treatment of Mandelstamm and Tamm, any such a momentum has no essential relation to the size of the
measurement of the energy of the observed system to an atom, but only to the lifetime of the excited state. The
accuracy AE must leave the "inner" time undefined to momentum has therefore been measured by an appara-
at& A/z E). tus, which is as localized in space as we please. (Of
In view of the above discussion, it is evident that the course, the position of the photon after the measure-
usual treatment of the energy-time uncertainty relation ment is over is indeterminate, just as happens with
(e.g. , as discussed by Fock and Krylov) must be in some "inner" time variables in the analogous case of time
way erroneous. Since the particular illustrative example measurement. )
chosen by all the authors cited here (which is, in fact,
the one usually given) has in fact been treated correctly, 3. TREATMENT OF TIME OF MEASUREMENT IN
it follows that the mistake must be that this example is TERMS OF THE MATHEMATICAL FORMALISM
OF THE QUANTUM THEORY
not suKciently typical of the general case. And, indeed,
as we shall see in Sec. 4, one can suggest more general We saw in Sec. 1 that (as pointed out by Fock and
methods of measurement of energy, which do not lead Krylov') there is a need to make a careful distinction
to the above limitation. In this way, we confirm our between the time at which a measurement takes place
conclusion, based on general considerations regarding and the time as defined by the wave function and
the principles of the quantum theory, that it is always operators of the observed system (e.g. , the lifetime of an
possible to obtain true limitations on the measurability excited state). In Sec. 2, we showed how such a dis-
of any observable from the mathematical formalism, tinction can be represented within the mathematical
and that any other limitations that are added to these formalism of the quantum theory by considering as
are arbitrary restrictions, which can eventually be operators certain variables that have hitherto usually
contradicted, if further examples of measurement been associated with the observing apparatus; viz. ,
processes are sought. " those variables determining the time at which inter-
Finally, it is instructive to point out that problems action between the apparatus and the observed system
similar to those connected with the time-energy un- takes place. This implies, of course, that the wave func-
certainty relations arise in the more familiar example tion must now be extended, so as to depend on these
of the position momentum relationship, hp, Ax~&A. To latter variables, It is equivalent to placing the "cut"
bring out the analogy, we can ask ourselves whether the between observing apparatus and observed system at
momentum of a particle can be measured to arbitrary a di6erent point.
accuracy by means of an apparatus, which is localized in It is well known that while there is a certain kind of
space. (Here, p, takes the place of E, while the region of arbitrariness in the location of this cut, it is not com-
space in which the apparatus is located takes the place pletely arbitrary. For example, in the treatment of the
of the duration, d, t, of the measurement. ) At first sight, energy levels of a hydrogen atom, one can, in a certain
it may seem that if the apparatus is localized in a very approximation, regard the nucleus as a classical particle
small region of size AX, the momentum of a particle in a well-defined position. If, however, the treatment
cannot be measured to an accuracy greater than aims at being accurate enough to take the reduced mass
hp =h/AX. This, however, is not the case, because, into account, both electron and nucleus must be treated
what is defined here is the coordinate, X, of the appara- quantum-mechanically, and the cut is introduced
tus, and not that of the measured particle, x. Since p, instead between the atom as a whole and its environment.
commutes with X, there is no inherent limitation on The place of the cut therefore depends, in general, on
how accurately it can be measured, if X is defined. how accurate a treatment is required for the problem
To illustrate such a possibility, consider the measure- under discussion. Of course, it follows that once a given
ment of the momentum of a photon, by measuring its place of the cut is justified, then it can always be moved
energy and using the relation p = E/c. (This is analogous further toward the classical side without changing the
to the measurement of energy of a particle by measuring results significantly.
its momentum and using the relation E= p'/2M. ) We If we are interested only in discussing what Fock and
can do this by means of an atom which is very highly Krylov called the "Bohr uncertainty relation" (the one
localized, provided that this atom has a sharp level, referring to an Asdieidla/ measurement of energy and
excited above the ground state by the amount 8=
pc. time), then we are justified in placing the physical
If the photon has the appropriate energy, it will be variables that determine the time of the measurement
absorbed and eventually reemitted (being delayed and on the classical side of the cut. For, as is quite evident,
perhaps scattered). It is observable whether this in this aspect of the uncertainty relations, these varia-
happens or not. If it does happen, then this provides a bles are by de6nition regarded as classical, in the sense
measurement of the energy, and through this, of the that their uncertainty represents only an inherent
ambiguity in the possibility of defining the state of an
"This conclusion, the validity of which is fairly evident, will
be obtained again in Sec. 3 from a more detailed discussion of the iedkidmul system. We have seen in Sec. 2, however, that
mathematical formalism. according to the Bohr point of view, every uncertainty
 Y. AHARONOV AN D D. BOHM
relation that appears in this way must also be able to this is indeed basically the reason why Fock and Krylov
appear as a statistical fiuctuation in a corresponding were led to postulate such an uncertainty relation
operator, which must, of course, be calculated from an independently, and to try to justify this relationship by
appropriate wave function. To discuss this side of the means of illustrative examples of measurement processes
uncertainty relations, it is clear that we must change (see Sec. 1).
the position of the cut, so that the corresponding Let us now go on to consider the time-energy un-
variables now fall on the quantum-mechanical side. certainty relation from the other aspect, in which the
In the subsequent discussion of how the uncertainty variables determining the time of measurement are
relations appear in the mathematical formalism, we shall placed on the quantum-mechanical side of the cut. In
begin with the case in which the time determining this case, we must introduce these variables into the
variables are placed on the classical side of the cut. In wave function, so that we are in this way led to a many-
this case, the time variables can be rejected in the body Schrodinger equation. Let us recall, however, that
Schrodinger equation only in the time parameter 3 which the "cut" has not been abolished, but merely pushed
can, of course, have an arbitrarily well-defined value. back another stage. Thus, as was pointed out in the
This time parameter is related to measurement in discussion of the treatment of Fock and Krylov given
several ways. in Sec. 1, there is implied an additional observing
First of all such a relation comes about in the prepara- apparatus on the classical side of the cut, with the aid
tion of a system in a definite quantum state, and in of which the many-body system under discussion can
observations carried out later on that system. Consider, be observed. The probabilities for the results of such
for example, a quantum state prepared at a time deter- observations are determined by the wave functions,
mined by means of a shutter (which we are, of course, which take the form
now regarding as being on the classical side of the cut). %=%(x,y, , t),
There must be some relationship between the time, t„
where y;, represents the apparatus variables on the
at which the shutter functions and the time parameter,
quantum-mechanical side of the cut (which include
t appearing the Schrodinger's equation. Indeed, if the
those that describe the time of measurement). The
Hamiltonian of the observed system does not depend on
time parameter t here plays a role similar to that which
time, then it is easily seen that the wave function takes
the form
it had in the one-body problem; viz. , through it the time
frame on the large-scale classical side of the cut is
P=P(x, t —t, ), brought into relationship with the quantum-mechanical
where 1b is a solution of Schrodinger's equation for the formalism by means of suitable observations.
system in question. The form of f
is determined by
We shall now consider as an example of the approach
described above, the measurement of energy and time
choosing that solution which at f=t, becomes equal to
the function, Po(x), representing the quantum state in by means of a collision of two particles, as treated in
which the "preparing" measurement leaves the system. Sec. 1. The initial wave function of the combined
Then, when an observation is made, the time t, of the system is, for this case, a product of two packet func-
tions, one representing the test particle coming in with
measurement is likewise determined by suitable varia-
bles on the classical side of the cut. The probability of a very high velocity, U, and the other representing the
observed particle, essentially at rest (with a velocity
any given result is, of course, computed in the well-
known way from the wave function, P=P(x, t — t, ). that is negligible in comparison to V) and with its
It is clear that as far as this one-body treatment is center at the origin. After the collision, it is well
concerned, there is certainly nothing in the formalism
known"" that the wave function becomes a sum of
which would prevent the system under discussion from products of such packets, correlated in such a way that
being either prepared or observed in a state of definite an observation of the properties of the test particle can
yield information about the particle under discussion.
energy, when f and 3, are as well defined as we please.
Thus, if the system is in a state of definite energy, E As far as this particular example is concerned, it will
(so that the uncertainty, hE, in its energy is zero, while not be relevant here to go into a more detailed discussion
the lifetime r) A/AE of the state is infinite), its wave of the problem of solving Schrodinger's equation. All
function, /=pe(x)e 'e"" )where p~(x) is the eigen- that is important here is that as we saw in Sec. 2, the

y-
function of the Hamiltonian operator belonging to the time of collision is given essentially by the operator,
eigenvalue, Eg is evidently able to represent such a
state, no matter what value is given to t. A wave func- t, = ,'nfl y —
1
+
1

tion of this kind is therefore evidently compatible with

the statement that at some time, t=t„ the system was
prepared in the eigenstate of the energy represented by (where y and p„refer to the position and momentum of
fe(x). Thus, in the one body treatmen-t alone, no reason '2 See reference 1, Chap. 22.
for an uncertainty relation between the energy of the
'3 J.
von Neumann, 3fathematica/ P'oundations of QNantum
M'echanics (Print:&top tJn&varsity Press, Princeton, Nevr Jersey,
system and the time of measurement can be found. And j.955), Chap. 6.
 QUANTUM THEORY FOR TIME AND ENERGY
the test particle, respectively), so that the operator, t„ Z=Z(t). To the extent that this happens, we obtain, as
commutes with the Hamiltonian, H„of the observed a good approximation,
system. As a result, there is, as we have already stated
in Sec. 2, no uncertainty relation between the time of V(x, y, s) = V(x, y, s(t))
measurement and the energy of the observed system. To treat this problem mathematically, we begin with
But in the treatment that we are now using, we have the Schrodinger equation for the whole system.
obtained this result by allowing the variables deter-
mining the time of measurement to fall on the quantum-
mechanical side of the cut. (As was to be expected, of iA (x,y, s, t) = PI„+H„+H, V(x, y, s)]%'(x,y, s, t), (14)
+
course, the actual physical consequences of the theory 8t
do not depend on which side of the cut these variables where JI„represents the Hamiltonian of the observed
are placed. )
particle, H„ that of the apparatus, H, that of the time-
In the example given above, the same apparatus was determining variable, s. We must then show the equiva-
used to determine both the time of measurement and
lence of this treatment with that obtained by placing
the momentum of the particle. However, it is possible,
the time-determining variable on the classical side of the
and in fact frequently advantageous, to consider a more
cut. To do this, it is sufficient to demonstrate that in a
general situation, in which the time determining
suitable approximation, equation (14) leads to the time-
variables are separated from those which are used to
dependent Schrodinger equation for the x and y
measure other quantities (such as momentum). This, in
variables alone, viz. ,
fact, would be the correct way of completing the treat-
ment of the example given by Landau and Peierls" (see
Sec. 1), in which the time of measurement was deter- ik —
8$
(x,y, t) = [H„+H„+V(x, y, s(t)))P(x, y, t). (15)
mined by an interaction between the test particle and Bt
the observed particle which was assumed to last for
We shall simplify the problem" by letting the time
some interval At.
determining variable be represented by a heavy free
If there is a time-dependent interaction between
particle mass M, for which we have H, =p, '/2M. We
apparatus and observed system which lasts for an
suppose that the initial state of the time-determining
interval At, then the Schrodinger equation will have to
variable can be represented by a wave packet narrow
have a corresponding potential, which represents this
enough in s space, so that Dt= As/ i~ can be made as
interaction. The form of this potential will depend on ~

small as is necessary. This packet is

where we place the cut.
If the apparatus determining the time of interaction
is taken to be on the classical side, then the potential
will be a certain well defined function of time, which is
C, (s, t)=g C„exp — sp, — —z

2M
2

nonzero only in the specified interval of length At. We

may write this potential as
Because M is very large, the wave packet will spread
very slowly, and to a good approximation, we shall have
V= V(x,y, t),
(I )'
where x represents the coordinate of the observed 4p(s, t) =4 (s —v, t) exp — p, s—
2M
particle, and y that of the test particle. This is indeed.

""
the usual way by which measurements are represented where s, = p, /M is the mean velocity, and C (s — s, t) is
in the mathematical formalism
just a form factor for the wave packet which is, in
In the next section we shall apply this method in general, a fairly regular function which varies slowly in
order to treat a specihc example, in which it will be comparison with the wavelength, X=h/P, .
shown in detail that the energy of a system can be If the interaction, V(x, y, s) is neglected, a solution
measured in an arbitrarily short time. for the whole problem will be
If, on the other hand, the variables determining the
time of interaction are placed on the quantum mechani- +(x,y, s, t) =Cp(s, t)gp(x, y, t),
cal side of the cut, then we cannot regard the potential
as a well-defined function of time. Instead, we must where 1t p(x, y, t) is a solution of the equation
write iABPp(x, y, t)/Bt= (H„+H„)gp(x, y, t). (19)
V= V(x, y, s), (12)
When this interaction is taken into account, the
where s is the variable that determines the time of solution will, in general, take the form
interaction.
If the particles determining the time of interaction @(x,y, s, t) = Q„C „(s,t)f„(x,y, t)C„,
are heavy enough, then they will move in an essentially '40ur procedure is along lines similar to those developed by
classical way, very nearly following a definite orbit, H. L. Armstrong, Am. J. Phys. 22, 195 (1957).
1656 Y. AHA RONOV AN D D. BOHM

where the sum is taken over the respective eigenfunc- result is

tions, C „(z,t) and f„(x,y, t) of H, and (H„+H„). If such
a sum is necessary, there will be correlations between 8$'(x, y, 0,t)
'It = $H„+H„+U(x, y, ,t) ]tp'(x, y, 0,t). (22)
the time-determining variable and the other variables,
with the result that there will be no valid approximation
in which an equation such as (15) involving only x and y In this way, we have obtained the Schrodinger equa-
canbe separated out. However, if the mass, M, of the tion for x, y, with the appropriate time-dependent
time determining particle is great enough, so that the potential V(x, y, z, t), the relationship between the time
potential V(x, y, z) does not vary significantly in a, wave- parameter 3 and the time determining variable s being,
length, X=h/p„ then, as is well known, the adiabatic in this case, t=z/z, . We have therefore completed the
approximation will apply. In this case, one can obtain demonstration of the equivalence of the two treatments,
a simple solution, consisting of a single product, even in which the time-determining variables are placed on
when interaction is taken into account. To show in more different sides of the cut.
detail how this comes about, we first write the solution
in the form 4. EXAMPLE OF A REPRODUCIBLE MEASUREMENT
4 I
= o(z, t) I//(x, y, z, t). OF ENERGY IN A WELL-DEFINED TIME

When this function is substituted into Schrodinger s We saw in Secs. 2 and 3 that there is no reason
equation (15), the result is inherent in the principles of the quantum theory why a
reproducible and exact measurement of energy cannot
be made in an arbitrarily short period of time. Since
ik =H„+—
H p+ V (x,y, z) Landau and Peierls, ' Pock and Krylov, ' and many
Bf
8 in@'p l9 IE
others have considered examples leading to a contrary
(2o) conclusion, it is necessary to complete the discussion by
M Bs Bs 2M Bs' giving a specific example of a method of measuring
energy precisely in as short a time as we please. This we
If M is large, and if the potential does not vary too shall do in the present section. Following the develop-
rapidly as a function of s, the last term on the right-hand
side of (20) can. be neglected. Moreover, " ment of our example, it will become clear in what
way the previous treatments of this problem were
8 lnCp i 8 ink(z —n, t) inadequate.
= —[p, +It As a preliminary step, we discuss the treatment of the
Bs Bs measurement of energy by means of the Schrodinger
equation for the apparatus and the observed system
Because C (z — v, t) does not vary significantly in a wave- together. The Hamiltonian of the combined system is
length, this term too can be neglected in the above
equation, and we obtain H= Hp(p, x)+Hp'(p, y)+Hr(p*, x; p„,y, t), (23)
where Hp(p„x) is the Hamiltonian of the observed
i7i
BP(x,y, z, t)
= H„+Hp+U(x, y, z) i'm, P(x, y, z—
, t).
—
system, Hp'(p„, y), that of the apparatus, and H&(p„x;
83 '
Bs
p„,y, t) is the interaction, which is zero except during a
certain interval of time between tp and tp+At. (Here we
We then make the substitution, s—v, t=N, and
are adopting the point of view described in Sec. 3, in
ip'(x, y, u, t) =ip(x, y, z, t) =tp(x, y, u+z, t, t). which we regard the time determining variables as
being on the classical side of the cut, so that they do not
With the relation
appear explicitly in Schrodinger's equation. ) It will be
BP'/Bt= (rtiiP/Bt)+ (n, 8$/rlz), adequate for our purposes to assume that both the
observed system and the apparatus are free particles,
we have with respective Hamiltonians
t', ABQ'/Bt= t H„+H„+V (x,y, u+ti, t)]P'(x, y, u, t). (21) '= p—
—
Hp p, '/2M H „'/2M. (24)
Note that this equation does not contain derivatives of
I
I, so that can be given a definite value in it. To simplify the problem,
a measurement
we consider the ideal case of
of p, which does not change p, . This
The complete wave function is, of course, obtained
by multiplying iP'(x, y, u, t) by 4p(z —
will happen if Hr is not a function of x. (The satisfaction
z, t) =Cp(u). Now,
of this condition will evidently guarantee that repro-
this was assumed to be a narrow packet centering at
+=0, such that the spread of can be neglected. As a I ducible measurements of p„and therefore H p= p, '/2M
will be possible). The Hamiltonian of the whole system
result, we can write m=0 in the above equation. The
will then be taken as
's If V(s, y, s) varies too rapidly, then (As/2N)8 P/ass will not be'
negligible, even when Jtt/I is large. H = (p, '/2M)+ (p„'/2M)+yp, g (t), (25)
 QUANTUM THEORY FOR TIME AND ENERGY 1657

where g(t) is everywhere zero, except between to and Y direction which is much greater than their thickness,
to+At, where it is constant. (The interaction Hamil- d, in the X direction. Therefore, they will produce a
tonian is similar to a vector potential in its effects). uniform electric Geld in the X direction, except for edge
With the Hamiltonian (25), p, is, of course, a constant effects which can be neglected when /))d. Each con-
of the motion. The equations of motion for the remain- denser will go by the particle at a velocity, V„ in the Y
ing variables are then direction, which is assumed to be so great that the
electric field acts for a very short time, t/U„, with the
result that the field approximates the one cited in our
mathematical example, where the period of action was
On solving for p„, we obtain (using infinitesimal. If the two condensers follow each other,
p, =constant) one at t=tp, the other at t= t~, then we shall approach
(2&) the case treated in Eq. (25).
As in the case of the collision treated in Secs. 1 and 3,
This equation implies a correlation between p„— P„' the time of measurement is defined as the time at which
and p„such that if p„— p„' is observed, we can the condenser passes the observed particle. (This means
calculate p, .
that we are now shifting to the point of view in which
It is important also to consider the behavior of i.
Although p, is constant, i shifts suddenly at t= to from
the time-determining variables are on the quantum-
mechanical side of the cut, but as we saw in Sec. 3, both
p /rN to P,/rN+g(t)y, and remains at this value until both points of view are equivalent and can be used
t=t;, after which it returns to its initial value. (In a As in the case of collision, the un-
interchangeably).
similar way, the velocity and rnornentum differ in the
certainty Dt in this time will be given by Dy/V„, pro-
case of a vector potential. )
vided that the observed particle is initially localized in
The above behavior of the velocity is, as a simple
the Y direction, with a velocity v„much smaller than
calculation shows, just what is needed to produce the
V~. This will imply an uncertainty in the energy of the
uncertainty in position, which is required by the im-
condenser, hF, = V„AP» A/At.
proved definition of the momentum resulting from the
measurement.
It is easily seen that if g(t) is large enough, the meas- transfer of I
In the interaction between particle and condenser, the

edge effects) is
component of the momentum (neglecting
urement described above can be carried out in as short a
time as we please. In order that a given accuracy, Ap„ Ap =p O=Fr=eSr=ehl/Uy, (28)
be possible, the change of deflection of the apparatus,
where 7 is the time taken by the condenser to pass the
A(p„—p„') due to the shift AP„must be greater than
particle. (Note that r and At are different quantities).
the uncertainty, h(p„) in the initial state of the This transfer is independent of initial conditions, and is
apparatus. This means that we must have
calculable to arbitrary accuracy. (Since V„ is as large as
DP g(t)ht=hP„', you please, the Y component of the velocity of the par-
ticle can be neglected in the above calculation. ) The
and if g(t) is large enough, both At and Ap, can be made
above transfer of momentum implies a transfer of
arbitrarily small for a given hp„'.
energy to the particle,
This hypothetical example confirms our conclusion
once again that accurate energy measurements can be
reproduced in an arbitrarily short time. We shall now (29)
show how to carry out such a measurement by means of
a concrete experiment. To guide us in the choice of this
By conservation, this transfer must be equal to the
experiment, we note that the essential feature of the —
interaction described in Eq. (25) is that it implies a energy loss Ep E& of the condenser. Since the initial
force that is independent of the x coordinate of the momentum I', of the condenser in the X direction is
partide, and which alters the velocity suddenly at zero, and since the mass M of the condenser is large, the
"
t= tp to bring it back to its original value at t= ti. This term (Ap, )'/2M which represents the energy of the
condenser due to momentum transfer in the X direction,
force is therefore equivalent in its effect to a pair of
will be negligible. The energy change must therefore be
equal and opposite pulses in a uniform electric 6eld, the
first at tp and the second at ti. In order to approximate the result of alteration in the Y component of the
condenser momentum, so that it will be equal to
such pulses, we shall consider two condensers, the fields
of which cross the observed particle at the times tp and
— —
Ep Eg Vyd I'y.
t~. The condensers are assumed to have a length, /, in the
Equation (29) can now be used to permit p, ' to be
— —
measured if Ep E=6 6p is known. For since ~ — 6p de-

"In the hypothetical example of Eq. (25), this force resulted pends on P,' and since AP, can be obtained from
from the time-dependent interaction, which was equivalent to a Eq. (28), p,' can be calculated in terms of Eo I:'i.
corresponding vector potential, which would produce a field,
8= —(1/c)BA/8t, that is nonzero only when A(t) changes; i.e., at There is, however, a limitation on how accurately
the beginning and the end of the interval. —
Ep E& can be measured because we require that the
 Y. AHA RONOV AN D D. BOH M

V coordinate of the condenser shall serve as a clock to be avoided by charging an insulator instead of a metal
an accuracy, bt, with the result that the uncertainty plate. The field will still be uniform, but the charges will
relation, 8E&A/8 will hold for the energy of the con- not be mobile, so that no currents will be induced in the
denser. By conservation, the same uncertainty relation condenser.
must hold for the energy c — eo transferred to the particle. The error in the treatments of Landau and Peierls,
By evaluating 8(e —eo) from Eq. (29), we obtain Fock, and Krylov, and others, as discussed in Sec. 1, is
now evident. For in all of these treatments, the example
Ap, used was that of a single collision of a pair of particles.
8(eo bt=8p, '
e)— Q=bp, 'An, Q=bp. '(n, ' —n, )Q)~h. For this case, our own treatment also gave the result
that energy transfer in a short time must be uncertain.
This is exactly the same relation as was obtained in But as shown in our general canonical treatment of the
the collision example given in Sec. 1. In other words, problem [see Eqs. (25)—(28)7, it is clear that this is not
the measurement on the first condenser alone, must the correct way to measure momentum and energy
satisfy the condition that if it is carried out in a time without changing them. To accomplish this purpose we
defined as 0t, there will be an uncertain energy transfer, need an interaction of the kind described in the above
hE) h/Q. It is at this point, however, that the second equations, which changes the velocity only while
condenser plays an essential role. For immediately interaction is taking place, but which brings it back to
after the interaction with the first condenser is over, it the initial value after interaction is over. And, as we
will bring about a transfer of X component of the have seen, it is possible to realize such a measurement
momentum, which is equal and opposite to that trans- in a concrete example.
ferred to the first condenser. As a result, the velocity of
S. SUMMARY AND CONCLUSIONS
the observed particle will return to its initial value, just
as happened in the mathematical example discussed in There has been an erroneous interpretation of un-
the beginning of this section. Thus, the momentum and certainty relations of energy and time. It is commonly
the energy have been measured without their being realized, of course, that the "inner" times of the
changed. There is, therefore, no limitation on the observed system (defined as, for example, by Mandel-
accuracy with which the energy of the particle can be stamm and Tamm') do obey an uncertainty relation
measured, regardless of the value of 6t, which can be AEht&~ h, where AE is the uncertainty of the energy of
made as small as we please by making V„very large. the system, and At is, in effect, a lifetime of states in that
A similar two-stage interaction can be carried out in system. It goes without saying that whenever the energy
the collision example described in Sec. 1. To do this, we of any system is measured, these "inner" times must
recall that the uncertainty in energy transfer, 8(e' — e) become uncertain in accordance with the above relation,
= (n —n'~8p is large because (n n') is large. Nev— erthe- and that this uncertainty will follow in any treatment of
less v — v' can be determined with arbitrary accuracy the measurement process. In addition, however, there
from the results of the measurement. After this is done, has been a widespread impression that there is a further
one can then send in a second test particle, with initial uncertainty relation between the dlratioe of measure-
momentum calculated to be such as to change v' back ment and the energy transfer to the observed system.
to n. After the two collisions, there will be, as in the Since this cannot be deduced directly from the operators
case of the condensers, no uncertainty in energy of the of the observed system and their statistical fluctuation,
observed particle. In the collision experiment, the it was regarded as an additional principle that had to
change of velocity depends on the value of the momen- be postulated independently and justified by suitable
tum of the observed particle, so that the initial condi- illustrative examples. As was shown by us, however,
tions in the second collision must be arranged, in this procedure is not consistent with the general
accordance with this value, which is learned from the principles of the quantum theory, and its justification
first collision. On the other hand, in the condenser was based on examples that are not general enough.
experiment, e, — v, is independent of initial conditions Our conclusion is then that there are no limitations
so that the second condenser can be prearranged to on measurability which are not obtainable from the
cancel out this shift of velocity without any information mathematical formalism by considering the appropriate
from the results of the first interaction. operators and their statistical Quctuation; and as a
At first sight, one might raise the question as to special case we see that energy can be measured
whether our conclusions could be invalidated by eGects reproducibly in an arbitrarily short time.
of radiation, or by currents which might be induced in
the condenser. Since we are discussing only the problem ACKNOWLEDGMENTS
of nonrelativistic quantum mechanics, we can assume Ke are gra, teful to Professor M. H. L. Pryce and to
that the velocity of light is infinite. In this case, radia- Dr. G. Carmi for helpful discussions, and one of us
tion and relativistic effects, in general, can be made (Y. Aharonov) wishes to acknowledge aid from a
negligible, no matter how sudden the shift of potential grant provided by the D. S.I.R. at the University of
is. As for currents induced in the condenser, these can Bristol, while this work was being done.