About Heisenberg Uncertainty Relation

By E. Schrodinger
Proceedings of The Prussian Academy of Sciences
Physics-Mathematical Section. 1930. XIX, pp.296-303

Abstract
The original Schrodingers paper is translated and annotated in honour of the
anniversary of his Uncertainty Relation [will appear in Bulg.J.Phys.]. In the
annotation it is shown that the Uncertainty Relation can be written in a complete
compact canonical form.
70th

ANNOTATION

by A. Angelow

++ ,

M.-C. Batoni

+++

The main reason to publish the original Schr

odingers paper in English, is the fact that
no one of the books on Quantum Mechanics cites it (see for example [1 -15 ]). Actually,
Schr
odingers paper is chiefly based on the notes of the seminars of Physics-Mathematical
Section of The Prussian Academy, where many famous physicists worked to establish the
underlying basis of Quantum Theory. Being a kind of internal report, this work remained,
for many years at a certain marginal distance from the physicist scientific awareness.
Another argument in favour of its oblivion concerns the enthusiastic discussions, mostly
about the physical interpretation of the uncertainty principle, rather than its mathematical
straightforward derivation. After Schr
odinger, the very first appearance of the new uncertainty relation occurs in the book of Merzbacher. However, he did not pay any attention to
the new term ( the covariance) and directly derives the Heisenberg inequality [11 ]. The
same embarrassment appears in [12 ] and [13 ].
Fortunately, over the last years the scenario changed: for example, in the field of Quantum
Optics, which was avocated to demonstrate the fundamental limit of Quantum Theory with
understanding underlying on that new term, two monographs [14 ,15 ] were published, but
the authors missed to cite the original Schr
odingers paper.
Schr
odingers work, originally written in German, was translated only in russian by A.
Rogali [16 ], in 1976. However, we would like to emphasize that the sentence (*), written
in the eleventh line after equation (11), is wrongly translated and it contradicts to the end
of the same paragraph [16 ]. In that paragraph Schr
odinger established classes of states
with non-vanishing covariance.
Finally, we would like to demonstrate the essential contribution of Schr
odingers inequality presenting it in a new compact form, using a modern terminology from mathematics,
rather than that used in 1930. Let us take the three independent second-order central moments of the joint quantum distribution of two variables A and B, which are of special
interest and warrant a special notation (see, for ex. [17 ,18 ]):
E[(A E[A])2 ] (A A)2 V ar[A] (A) 2

(1 )

E[(B E[B])2 ] (B B)2 V ar[B] (B) 2

(2 )

(A E[A])(B E[B]) + (B E[B])(A E[A])

E
Cov[A, B] (A, B)
2


(3 )

where E[.] means the expectation value. Obviously, the fourth second-order moment is
Cov[B, A] = Cov[A, B], (respectively (B, A) = (A, B)). Note that when two observables X and Y dont commute the correct expression for their product is not XY , but the
X
, and we conclude that the covariance is equivalent to the new
symmetrized one XY +Y
2
term in the Schr
odingers inequality:
Cov[A, B] =

AB + BA
AB
2

(4 )

Let us construct the so-called [19 ] covariance matrix (keeping in mind the non-commutativity of the observables in contrast to [19 ]):
[A, B]

V ar[A]
Cov[B, A]

Cov[A, B]
V ar[B]

(A) 2
(B, A)

(A, B)
(B) 2

(5 )

Now, we can write the Schr

odinger Uncertainty Principle ( see below-eq.(9) ) in the canonical form
det([A, B])

1
2
|[A, B] | ,
4

or , for position and momentum

det([q, p])

h
2
,
4

(6 )

and it is easy to see that the uncertainty relation is invariant under the rotation transformation in the phase space, while the Heisenberg one is not. We would like to emphasize
that the new term in the inequality also plays an important role in the method of linear
invariants in Quantum Mechanics, where the covariance is expressed in therms of the solution of the equation of a non-stationary two-dimensional harmonic oscilator [20 ].
Translated and annotated in honour of 70 th anniversary of Schr
odinger Uncertainty Relation.
Sofia, January 1999
S
ao Paulo, March 1999.

PACS: 03.65.-w
1. Recently E. U. Condon and H. P. Robertson [1] took into consideration the generalization of the fundamental principle of the quantum mechanics - that of the uncertainty
- over an arbitrary canonical non-conjugate couple of physical variables. Trying to reach
the same, I arrived at a slightly wider generalization than the Robertsons one, which is,
in fact stronger than the original Heisenberg inequality.
First of all, let us set out what is well known. The state-of-the-art of the interpretation
question is the following: the test domain is a single specific physical system. The bases
for the system knowledge that we dispose of - the catalogue of all that we can assert
about the system - is equivalent to a complex function in the coordinate space of the
system (it changes in a regular manner in time, but is not important at the moment).
The mathematical correlate of a physical variable, i.e. of a very specific measurement
that one might apply to the system, is a very specific linear Hermitean operator that from
each -function produces an other such a -function. One can calculate the expectation
2

value of the resfpective measurement from the measure operator, say A, and the given
-function:
Z
A = Adx
(1)
( is the complex conjugate, the integration
goes over the whole coordinate space; given
R
that is constantly normalized, i. e. dx = 1).

The meaning of the expectation value is: mean value by unlimited number of measurements, while one must be sure that the system state is the same before each measurement,
not changed by the measurement itself. In general, all possible statements one can make
about the system are encoded in the expectation values. Moreover, one should keep in
mind that it is up to us to choose the marking of reference scale of our measurement
instrument. We can, for example, to set a value one only to one scale division and zero
to all other. There is a specific operator attached to this measurement one could
name it as operator in blinkers , V.Neumann named it unitary operator. The respective
expectation is obviously nothing else than the probability of the corresponding measurement value or measurement value intervals. The -function determines also the total
measurement statistics.
The average error or the mean uncertainty of the value, which belongs to the operator
A, is defined as
q
q
(2)
A = (A A)2 = A2 (A)2

(where in the first of the two expressions A should be more precise: A multiplied by
the identical operator.) It may be proven, that this definition is not only formally constructed according to the theory of errors, but A is really the average error of the
variable A, when the statistics is defined in the above given way.

To prove now, that the product of the uncertainties of two random variables A and
B satisfies the Heisenberg or yet more precise inequality we need to denote the following
mathematical statements:
1. the Hermitean character of A implies that the expectation value (1) is constantly
real;
2. for each Hermitean operator it holds
Z

f Agdx =

gA f dx,

(3)

i.e., it could be rolled over on the other factor in such an integral, in such case the operator
transforms into its conjugate form [2];
3. the product of two Hermitean operators is in general not Hermitean, but it could
be split into symmetrical product and its (half) commutator:
AB =

AB + BA AB BA
+
2
2

(4)

The first term is Hermitean,

the last one is skew Hermitean, i.e. it becomes Hermitean

multiplying by i = 1. The splitting in many aspects corresponds to the splitting of

a random (complex) number into real and imaginary part. Immediately from this one
3

might extract the splitting of the expectation value into real and imaginary parts. The
expectation value of every commutator is pure imaginary.
4. Finally, we need the so-called Schwartz inequality [3]
(a1 a1 + a2 a2 + ... + an an )(b1 b1 + b2 b2 + ... + bn bn ) |a1 b1 + a2 b2 + ... + an bn |2 ,

(5)

that we will apply in a limiting case on the continuous range of values of both functions
f and g in the coordinate space:
Z

f f dx

We assume here that specially

gg dx  f gdx .

(50 )

g = A ,

(6)

f = B

where A and B are some Hermitean operators and is an arbitrary wave function, i.e. an
arbitrary continuous and normalized function in the coordinate space. Using the equation
(3) one obtains
Z

B dx

A dx

i.e., in therms of the notation (1)

2

ABdx ,

(7)

(70 )

A2 B 2 AB  .

When we decompose the right hand side according to (4), because of the linked note with
this splitting we get
A2 B 2

AB + BA
2

!2


 AB

+


BA 
 .

2

(8)

This is already the inequality that we need to proof, but only in the special case when
A and B vanish. In order to arrive at the general case, one should apply (8) and instead
of the operators A and B rather use the following
A 1 and B 1.
First of all and must be arbitrary real constants, 1 is the (unitary) operator
multiplied by . The resulting inequality is therefore valid: 1. for an arbitrary 2.
for every real pair of constants , . Therefore, there is no limitation on the -function
to influence the choice of the pair of constants and especially to set
= A,

= B.

Finally, we end up with:

2

(A) (B)

AB + BA
AB
2

!2


 AB

+


BA 
 .

2

(9)

This is the final form. The first from the two addends on the right hand side is new (to
the best of my knowledge). (Without that term the inequality stands as the one of H. P.
4

Robertson.) So, the inequality links together three quantities: 1. the product of the
mean deviations squared, 2. the absolute value squared of half of the mean value of the
commutator, 3. a quantity which could be defined as a square of the mean deviationsproduct (the covariance) in the condition that non-commutability is taken into account,
i.e. the mean deviations-product must be define as the arithmetic mean of
(A A)(B B) and (B B)(A A)

(10)

which are the mixed expressions ( covariances, see eq.(3 ) ) , completely analogous
to (A)2 and (B)2 .
One is led to the Heisenberg inequality when the last mentioned quantities are stricken
out in order to make stronger the inequality and A, B are chosen to be canonically
conjugate:
h
AB BA =
.
2i
Then it results in
h
.
(11)
A B
4
On the other hand, it is known that the Heisenberg limit is not really too low, but for
some special -functions achieves even higher value [4]. This implies that at least for
these special -functions the (central) mean deviations-product of the canonical conjugate operators vanishes. This will be used in 2.
In the classical theory of errors or fluctuation theory it is well known that the vanishing
of the mean deviations-product is a necessary (but not sufficient) condition for two values
to fluctuate totally independent one from an other. While canonically conjugate quantum
variables have some independence that could mean that some precise knowledge about
one excludes such a knowledge about the other, so one could perhaps suppose that their
mean deviations-product, i. e. for each -function, has vanishing expectation value. But
this is not the case( ). Let us consider the two canonically conjugate operators
A=x

B=

h
,
2i x

so we get [5]
"

2i AB + BA
1
=
h
2
2

1
2

A=

x dx

2i
B=
h

dx =
x
2

+
(x) dx =
x
x
x
2

x
dx
x
x
!

ln dx
x

dx

x
x

1Z

ln dx.
2
x

Let now = rei with real and real, non-negative r, which must satisfy the normalizing
condition
Z
r 2 dx = 1
Then we get:

2
h

Z
Z
Z

AB + BA
A B = xr 2
dx xr 2 dx r 2
dx.
2
x
x

is any real function and r 2 is an absolute non-negative function (not taking into
As
x
account the normalizing condition), so we get that in general the right hand side does
not vanish. One needs for example to choose r 2 to be even and
to be odd (and not
x
identically vanishing), so the deviation product is surely positive.
As known, the canonically conjugate quantum variable is not unambiguously defined.
If B is conjugate to A this implies that B + A is as well ( is any real number). With this
change the mean deviations-product changes as well, and becomes, as one could easily
calculate, (A)2 . In the same manner, the result would be (B)2 , if A was changed
to A + B. This can always make the deviations-product equal to zero by changing one
of the operators, without changing their canonical relation. The change depends on the
above shown special -function of course. One can not reach an identical vanishing of
the deviations-product in such a manner.
2. To the discovery of the complete inequality (9) we are led, by a chance, to the
following question, which is interesting by itself. Let us consider a force free mass point,
p2
. I must undertake
mass m, coordinate q, momentum p, Hamilton-function H = 2m
simultaneous measurements of the coordinate and momentum at time zero, with highest
possible precision, i.e. so that
h
.
(12)
q0 p0 =
4
Further I must distribute the error on q0 and p0 so that for a given later time point t, it
could be achieved the most precise place. This means q to become the least possible.
We use for this purpose the very convenient q-number-method, which is in a methodical manner opposing to the wave mechanics. I would like to elucidate shortly on it here,
repeting what is well known. For the theorist working on a wave mechanics the operator,
which corresponds to a specific physical variable, does not change in time. If one wants to
know the mathematical expectation value for this variable, one calculates the -function
for this later moment from the time-dependent wave equation. Then one applies the
corresponding operator, which as already mentioned, is the same for every moment. On
the other side the q-number-theorist has to operate with one single -function at one single chosen moment, once and for all. However it is unnecessary to express any statement
for it, once the moment is totally arbitrary chosen. One assumes, instead that the operators are time dependent and we may ask: how does the operator change itself in time,
i.e. which operator should be applied on the original -function, in order to calculate
the mathematical expectation of the respective value at the time t?
Here we point out, one may calculate the operators (or q-numbers, or matrices) almost
as the usual numbers, and indeed, their change in time is determined by the equation
6

of motion of the classical mechanics. The only difference is that, occasionally, when
it is the case, one should pay special attention to an eventual non-commutability of the
operator multiplication.
So, in this present simple case, the integration of the equation of motion reads:
q = q0 +

t
p0 .
m

One can directly make from it the mean square deviation of the coordinate, (q)2 , for
every moment t:
!2

2
t
t
2
(q) = q0 + p0 q0 + p0
m
m
2t
= (q0 ) +
m
2

q0 p 0 + p 0 q0
q 0 p0
2

!2

+ (p0 )2 .

The middle term above is essentially the mean deviations-product of q0 and p0 , which
vanishes, in accordance with the prediction, when q0 and p0 are determined with optimal
precision. Then we simply have
t2
(q) = (q0 ) + 2 (p0 )2
m
2

or using (12)
h
(q) = (q0 ) +
4
2

!2

t2
1
,
2
m (q0 )2

see Ref.[6].

This expression becomes a minimum for that value of (q0 )2 , which makes both addends
on the right hand side equal, i.e. for
(q0 )2 =

ht
;
4m

(q)2 is then exactly twice the value of (q0 )2 , i. e.

q =

ht
.
2m

(13)

It seems to me, that this final result is likely to have two points of interest. First,
the proportional relation with the square root from the time, which makes allusion to
well known classical deviation principles. Secondly, that the statement has an remarkable
absolute character, namely, the precision attainable in a later moment depends only on
intermediate time and not on the initial momentum. For example, for a free electron one
might give a place prognosis for the end of the first second on the bases of already taken
measurements of position and momentum, in the most favorable case with a precision of
1cm, quite independent of whether the electron is fast or slow [7].
Of course, at a very high speed this will be changed as it should taken into consideration
the relativity theory. I believe that this could occur by the following simple considerations.
7

The equation (13) is applied to the rest reference system of a point mass. Let m be the
rest mass, tr , the internal time:
(q)r =

htr
.
2mr

(14)

This is the precision that is attainable for a moving observer when calculating the position
of the point mass of the co-moving system for the moment, called tr seconds later. When
the observer shows his knowledge through signs in the space,
to the rest observer those
signs are seeming to be nearer to each other in relation of
1 2 : 1 ; further more
he must say looking from his standpoint that the prognosis were made for a time interval
t=

tr
,
1 2

(15)

because for him the clock, with which all the statements of the moving observer were
made, run slower than his own clock. From his point of view, the mean error decreases
q =

v
u
u ht 1 2
2t

2mr

1 2

ht
.
2m

(16)

It becomes smaller and comes nearer to zero when the velocity approaches is nearing the
speed of the light. This happens not only when the mass m goes to infinity, but also for
a series of point masses moving with an ever growing speed and an ever smaller rest mass
such that all the moving masses m keep the same value m. Even in this case the maximum
precision grows unlimited with the velocity approaching the speed of the light. This is
indeed satisfying, since this is a boundary process that gives a hope to obtain an accurate
statement for a light quantum. And this is really true for light quantum because the
Maxwell waves exhibit no dispersion. They preserve indefinitely long the place precision
that they got in the beginning and it could indefinitely grow, since the strong momentum
dispersion, which is connected with it does not have a bad influence.
Reported on the 5th June 1930
Joint-Staff Meeting on the 19th June 1930
Distributed on the 16th July 1930

References
[1] E. U. Condon, Science, 31. Mai (1929);
H. P. Robertson, Phys. Rev. 34 (1929) 163.
Mr. A. Sommerfeld was so kind to point out to me these two notes, when I told him
about the following considerations.
[2] A is so defined that evidently A = (A) , and therefore A = (A ) .
[3] Vgl. H. Weyl, Gruppentheorie und Quantenmechanik, Hirzel, Leipzig, (1928) 272.
This proof is closely connected to the proof of the Heisenberg inequality given there.
8

[4] J.v.Neumann, ZS.f. Phys. 57 (1929) 34;

W. Heisenberg, Ibid. 43 (1927) 187;
G.G. Darwin, Proc. Ray. Soc. A 117 (1927) 268.
[5] All integrals are from to +.
[6] This equation was already developed by Heisenberg in his first work on the uncertainty principle (Zeitschr. f. Phys. 43 (1927) 188), but for the special case of pseudo
arbitrary character -function.
[7] Of course, there are wave functions that, in a precise given later moment, can determine the position with an arbitrary given precision. But one needs simply the chance
to be able to trace back such an approximation of a maximal function by means
of a wave equation over the corresponding time interval, and the resulting function
to be taken as original state. But it should be not less than optimal.

) Note added from the translators.

) By this Schrodinger means that the operator does not change the direction and modulus of the vector, as the horse in blinkers does not change the direction and the speed
until this is not required by the driver.

) Indeed, if one put B = A in (3 ), then Cov[A, A] = V ar[A] (A)2 .

) Reprinted in: ERVIN SCHRODINGER,

Gesammelte Abhandlungen, Band 3, Wien,

Verlag der Osterreichischen Akademie der Wissenschaften, pp.348-356 (1984)

++
) Permanent address:
Institute of Solid State Physics, Bulgarian Academy of Sciences, 72 Trackia Blvd., 1784
Sofia, Bulgaria.
+++
) Permanent address:
Instituto de Fisica Teorica, Universidade Estadual Paulista, Rua Pamplona, 145, CEP
01405-900 Bela Vista, Sao Paulo, Brazil.

References

added from translators.

[1 ] W.Heisenberg, The physical principles of the quantum theory, New York, Dover,
(1930);
[2 ] J. Von Neumann, Mathematical foundations of quantum mechanics, Princeton,
NJ, University Press, (1955);
[3 ] P.A.M. Dirac, The principles of quantum mechanics. 4th rev. ed., Oxford, Clarendon Press (1958);
[4 ] L.Landau, E.Lifshits, Quantum Mechanics - non-relativistic theory. 3rd ed., Oxford, Pergamon Press, (1977);
[5 ] L.de Broglie, Heisenberg Uncertainties and Probabilistic Interpretation of Wave
Mechanics, Dordrecht, Kluwer Academic Publishers (1995) Chapter 8, eq.(3) and (8);
[6 ] W.Price, S.Chissick, W.Heisenberg, The uncertainty principle and foundations of
quantum mechanics: a fifty years survey, New York, Wiley, (1977);
[7 ] A.Bohm, M.Loewe, Quantum mechanics: foundations and applications. 2nd rev.
9

and enl.ed., New York, NY, Springer-Verlag (1986);

[8 ] C.Cohen-Tannoudji, B.Diu, F.Laloe, Mecanique quantique, v.1, Paris, Hermann
(1973) ;
[9 ] L.Schiff, Quantum Mechanics, 3rd ed., International series in pure and applied
physics, New York, McGraw-Hill (1968);
[10 ] A. Messiah, Quantum Mechanics, v.1, North-Holland Publishing Company, Amsterdam (1961) p.300, eq.(VIII.9); actually the author had been very close to the general relation, only that he had not taken the centralized covariance, see the proof below
eq.(VIII.9);
[11 ] E.Merzbacher, Quantum Mechanics, 3rd ed.,New York, NY, Wiley (1998) p.219,
eq. (10.58); the same situation with the 1st and 2nd editions;
[12 ] A.Das, C.Melissinos, Quantum mechanics: a modern introduction, New York,
NY., Gordon & Breach, (1986);
[13 ] J.Sakurai, S.Tuan, Modern quantum mechanics. Rev. ed., Reading, MA, Addison - Wesley, (1994);
[14 ] F.Schroeck, Quantum mechanics on phase space, Dordrecht, Kluwer Academic
Publishers (1996);
[15 ] J.Perina, Z.Hradil, B.Jurco, Quantum optics and fundamentals of physics, Dordrecht, Kluwer Academic Publishers (1994);
[16 ] About Heisenberg uncertainty relation, translated in russian by A.Rogali, pp.210217, in E.Schrodinger, Izbrannie trudi po kvantovoi mehanike (Collected papers on quantum mechanics) , Moscow, Nauka (1976); the russian translation of the sentence (*) is:
However, this never happens.;
[17 ] G.Korn, T.Korn, Mathematical handbook for scientists and engineers, sec. enl.
and rev. ed., New York, N.Y., McGraw-Hill Publs. Co., (1968) p.604, eq.(18.4-10);
[18 ] H.Tucker, An introduction to probability and mathematical statistics, New York,
NY, Academic Press, (1963) p.91;
[19 ] C.Gardiner, Handbook of stochastic methods, 2nd ed., Berlin, Springer-Verlag,
(1985);
[20 ] A.Angelow, Physica A, 256 (1998) pp.485-498, the last equation in (24) and
equation (B.4).

10