SPECIAL SECTION: QUANTUM MEASUREMENTS

Heisenberg’s original derivation of the

uncertainty principle and its universally valid
reformulations
Masanao Ozawa*
Graduate School of Information Science, Nagoya University, Chikusa-ku, Nagoya, 464-8601, Japan

momentum, or more generally the mean error of the

Heisenberg’s uncertainty principle was originally posed
for the limit of the accuracy of simultaneous meas- simultaneous momentum measurement, where h is
urement of non-commuting observables as stating that Planck’s constant.
canonically conjugate observables can be measured
simultaneously only with the constraint that the prod- Let  (qˆ ) [originally, q1 ] be the precision with which
uct of their mean errors should be no less than a limit the value q is known ( (qˆ ) is, say, the mean error of
set by Planck’s constant. However, Heisenberg with q), therefore here the wavelength of the light. Let  ( pˆ )
the subsequent completion by Kennard has long been [originally, p1 ] be the precision with which the value p
credited only with a constraint for state preparation is determinable; that is, here, the discontinuous change
represented by the product of the standard deviations. of p in the Compton effect (ref. 1, p. 64).
Here, we show that Heisenberg actually proved the
constraint for the accuracy of simultaneous measure- Heisenberg claimed that this relation is a ‘straightforward
ment, but assuming an obsolete postulate for quantum mathematical consequence’ (ref. 1, p. 65) of fundamental
mechanics. This assumption, known as the repeatabil- postulates for quantum mechanics. In his mathematical
ity hypothesis, formulated explicitly by von Neumann derivation of relation (1), he derived
and Schrödinger, was broadly accepted until the 1970s,
but abandoned in the 1980s, when completely general 
quantum measurement theory was established. We  (qˆ ) ( pˆ )  (2)
also survey the present author’s recent proposal for a 2
universally valid reformulation of Heisenberg’s uncer-
tainty principle under the most general assumption on for standard deviations  (qˆ ) and  ( pˆ ) of position q̂ and
quantum measurement. momentum p̂ for a class of Gaussian wave functions,
later known as minimum uncertainty wave packets. Sub-
sequently, Kennard2 proved the inequality
Keywords: Error-disturbance relations, quantum root
mean square, quantum measurement, repeatability hypo-

thesis, uncertainty principle.  (qˆ ) ( pˆ )  (3)
2
Introduction
for arbitrary wave functions. By this relation, the lower
T HE uncertainty principle proposed by Heisenberg in 1 bound of relation (1) was later set as
1927 revealed that we cannot determine both position and

momentum of a particle simultaneously in microscopic  (qˆ ) ( pˆ )  , (4)
scale as stating ‘the more precisely the position is deter- 2
mined, the less precisely the momentum is known, and
conversely’ (ref. 1, p. 64), and had overturned the deter- where   h /(2 ).
ministic world view based on Newtonian mechanics. Us- Textbooks3–6 up to the 1960s often explained that the
ing the famous -ray microscope thought experiment physical meaning of Heisenberg’s uncertainty principle is
Heisenberg1 derived the relation expressed by eq. (4), but it is formally expressed by eq.
(3). This explanation was later considered to be confus-
 (qˆ ) ( pˆ ) ~ h (1) ing. In fact, it was pointed out that eq. (4) expresses a
limitation of measurements, while the mathematically
for  (qˆ ), the ‘mean error’ of the position measurement, derived relation eq. (3) expresses a statistical property of
and  ( pˆ ), thereby caused ‘discontinuous change’ of the quantum state, or a limitation of state preparations, so
that they have different meanings 7. Thus, Heisenberg
*e-mail: [email protected] with the subsequent completion by Kennard has long

2006 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015

 SPECIAL SECTION: QUANTUM MEASUREMENTS
been credited only with a constraint for state preparation Axiom 2 (Born statistical formula). If an observable A
represented by eq. (3). is measured in a state , the outcome obeys the probability
This article aims to resolve this longstanding confu- distribution of A in  defined by
sion. It will be shown that in 1927 Heisenberg1 actually
‘proved’ not only eq. (2) but also eq. (1) from basic Pr{ A  ||}  Tr[ E A ( )  ], (5)
postulates for quantum mechanics. In showing that, it is
pointed out that as one of the basic postulates Heisenberg where    (R).
supposed an assumption called the ‘repeatability hypo-
thesis’, which is now considered to be obsolete. In Axiom 3 (Time evolution). Suppose that a system S is
fact, in the 1930s the repeatability hypothesis was explic- an isolated system with the (time-independent) Hamilto-
itly claimed by von Neumann 3 and Schrödinger8, whereas nian H from time t to t + . The system S is in a state (t)
this hypothesis was abandoned in the 1980s, when quan- at time t if and only if S is in the state (t + ) at time
tum measurement theory was established to be general t +  satisfying
enough to treat all the physically realizable measure-
ments.  (t   )  exp(i  H/ )  (t ) exp(i H/). (6)
Through those examinations it will be concluded that
Heisenberg’s uncertainty principle expressed by eq. (4) is
Under the above axioms, we can make a probabilistic
logically a straightforward consequence of eq. (3) under a
prediction of the result of a future measurement from a
generalized form of the repeatability hypothesis. In fact,
knowledge about the past state. However, such a predic-
under the repeatability hypothesis a measurement is
tion applies only to a single measurement in the future. If
required to prepare the state with a sharp value of the
we make many measurements successively, we need an-
measured observable, and hence the ‘measuremental’
other axiom to determine the state after each measure-
uncertainty relation (4) is a logical consequence of the
ment. For this purpose, the following axiom was broadly
‘preparational’ uncertainty relation (3).
accepted in the 1930s.
As stated above, the repeatability hypothesis was aban-
doned in the 1980s, and nowadays relation (4) is taken to
Axiom 4 (Measurement axiom). If an observable A is
be a breakable limit9,10 . Naturally, the problem remains:
measured in a system S to obtain the outcome a, then the
what is the unbreakable constraint for simultaneous
system S is left in an eigenstate of A belonging to a.
measurements of non-commuting observables? To answer
this question, we will survey the present author’s recent
Von Neuamann3 showed that this is equivalent to the fol-
proposal for a universally valid reformulation of Heisen-
lowing assumption called the repeatability hypothesis
berg’s uncertainty principle under the most general
(ref. 3, p. 335), posed with a clear operational condition
assumption on quantum measurement11–13.
generalizing a feature of the Compton–Simons experi-
ment (ref. 3, pp. 212–214).
Repeatability hypothesis
(R) Repeatability hypothesis. If an observable A is
The uncertainty principle was introduced by Heisenberg measured twice in succession in a system S, then we get
in a paper entitled ‘Über den anschaulichen Inhalt der the same value each time.
quantentheoretischen Kinematik und Mechanik’1 published
in 1927. In what follows we shall examine Heisenberg’s It can be seen from the following definition of measure-
derivation of the uncertainty principle following this paper. ment due to Schrödinger given in his famous ‘cat para-
Before examining the details of Heisenberg’s deriva- dox’ paper8 that von Neumann’s repeatability hypothesis
tion, we shall examine the basic postulates for quantum was broadly accepted in the 1930s.
mechanics in Heisenberg’s time, following von Neu-
mann’s formulation 3. In what follows, a positive operator The systematically arranged interaction of two systems
on a Hilbert space with unit trace is called a density op- (measured object and measuring instrument) is called a
erator. We denote by  (R) the set of Borel subsets of R measurement on the first system, if a directly sensible
and by E A the spectral measure of a self-adjoint operator variable feature of the second (pointer position) is
A, i.e. A has the spectral decomposition A   R  E A (d ). always reproduced within certain error limits when the
process is immediately repeated (on the same object,
Axiom 1 (States and observables). Every quantum sys- which in the meantime must not be exposed to any
tem S is described by a Hilbert space  called the state additional influences)8.
space of S. States of S are represented by density opera-
tors on  and observables of S are represented by self- Based on the repeatability hypothesis, von Neumann3
adjoint operators on . proved the impossibility of simultaneous measurement of

CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015 2007

SPECIAL SECTION: QUANTUM MEASUREMENTS
two noncommuting observables as follows. Suppose that observable q̂ to obtain the outcome q with mean error
two observables A, B are simultaneously measurable in  (qˆ ) and consider what relation holds between  (qˆ ) and
every state and suppose that the eigenvalues of A are non-  ( pˆ ) if the momentum observable p̂ has been measured
degenerate. Then, the state just after the simultaneous simultaneously to obtain the outcome p with mean error
measurement of A and B is a common eigenstate of A and  ( pˆ ). Then, by (AR) or eq. (8) the state  should have
B, so that there is an orthonormal basis consisting of the position standard deviation  (qˆ ) satisfying
common eigenstates of A and B, concluding that A and B
commute.  (qˆ )   (qˆ ). (9)
Since Heisenberg’s uncertainty principle concerns
measurements with errors, it is naturally expected that it Heisenberg actually supposed that the state  is a Gaus-
can be mathematically derived by extending the above sian wave function (ref. 1, p. 69)
argument to approximate measurements.
1  (q  q ) 2 i 
 (q)  exp   p(q  q ) , (10)
Approximate repeatability hypothesis ( q12 )1/4 
2
2q1  

To extend the repeatability hypothesis to approximate which was later known as a minimum uncertainty wave
measurements, we generalize the notion of eigenstates as packet, with its Fourier transform
follows. For any real number  and a positive number , a
(vector) state  is called an -approximate eigenstate be- 1  ( p  p ) 2 i 
longing to  iff the relation ˆ ( p)  exp   q( p  p)
( p12 )1/4 
2
2 p1  
|| A   ||   (7) (11)

and he proved relation (2) for the state  given by eq. (10).
holds. If  = 0, the notion of -approximate eigenstates is
Exactly this part of Heisenberg’s argument was gener-
reduced to the ordinary notion of eigenstates. A real
alized by Kennard2 to prove relation (3) for any vector
number  is called an approximate eigenvalue of an
state . Thus, Kennard2 relaxed Heisenberg’s assumption
observable A iff for every  > 0 there exists an -appro-
on the state  to the assumption that the state  after the
ximate eigenstate of A. The set of approximate eigen-
position measurement can be the arbitrary wave function
values of an observable A coincides with the spectrum of
 satisfying eq. (9). Then, if the momentum observable
A (ref. 14, p. 52).
p̂ has been measured simultaneously to obtain the out-
Now, we formulate the approximate repeatability
come p with an error  ( pˆ ), by (AR) or eq. (8) again the
hypothesis as follows.
state  should also satisfy the relation
(AR) Approximate repeatability hypothesis. If an ob-  ( pˆ )   ( pˆ ). (12)
servable A is measured in a system S with mean error 
to obtain the outcome a, then the system S is left in an Therefore, Heisenberg’s uncertainty relation (4) immedi-
-approximate eigenstate of A belonging to a. ately follows from Kennard’s relation (3).
In 1927, Heisenberg not only derived relation (1) using
Obviously, (AR) is reduced to (R) for  = 0. Since we the -ray thought experiment, but also gave its mathemati-
have cal proof. However, he supposed the repeatability hypothe-
sis or its approximate version as an additional assumption
|| A   ||  ||A   A ||   ( A)
to the standard postulates for quantum mechanics.
The approximate repeatability hypothesis (AR) has
for any real number , where A = (, A), (AR) implies not been explicitly formulated in the literature, but in
the following statement: If an observable A in a system S the following explanation on the derivation of the uncer-
is measured with mean error  (A), then the post- tainty principle, von Neumann3 (pp. 238–239) assumed
measurement standard deviation  (A) of A satisfies (AR):
 ( A)   ( A). (8)
We are then to show that if Q, P are two canonically
conjugate quantities, and a system is in a state in which
Heisenberg’s derivation of the uncertainty the value of Q can be given with the accuracy  (i.e. by
principle a Q measurement with an error range ), then P can be
known with no greater accuracy than    /(2 ). Or: a
Heisenberg’s derivation of eq. (1) starts with considering measurement of Q with the accuracy  must bring about
a state  just after the measurement of the position an indeterminacy    /(2 ) in the value of P.
2008 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015
 SPECIAL SECTION: QUANTUM MEASUREMENTS
In the above, it is obviously assumed that a state with the Pr{x  || }  Tr[  ()  ], (13)
position standard deviation  is resulted by a Q measure-
ment with an error range . This assumption is what we  ( ) 
have generally formulated in eq. (8) as an immediate log-    {x}  . (14)
Tr[  ()  ]
ical consequence of (AR).
Two inequalities (3) and (4) are often distinguished as
the preparational uncertainty relation and the measure- For any    (R), define () by
mental uncertainty relation respectively. However, under
the repeatability hypothesis such a distinction is not appar-  ( )   ()* 1, (15)
ent, since a measurement is required to prepare the state
with a sharp value of the measured observable. In fact, the where ()* is the dual map of () given by
above argument shows that there exists an immediate logi- Tr[( ()*X)] = Tr[X(  ())] for all X  (). Then,
cal relationship between these two inequalities. the map   () is a probability operator-valued meas-
ure (POVM)21, called the POVM of , satisfying
Abandoning the repeatability hypothesis
Pr{x  ||}  Tr[ ()  ] (16)
The repeatability hypothesis applies only to a restricted
class of measurements and does not generally character- for all   () and   (R).
ize the state changes caused by quantum measurements. The problem of mathematically characterizing all the
In fact, there exist commonly used measurements of dis- physically realizable quantum measurements is reduced
crete observables, such as photon counting, that do not to the problem as to which instruments are physically
satisfy the repeatability hypothesis15. Moreover, it has realizable13. To settle this problem, standard models of
been shown that the repeatability hypothesis cannot be measuring processes were introduced in ref. 16 as fol-
generalized to continuous observables in the standard lows. A measuring process for (a system described by) a
formulation of quantum mechanics16–19. In 1970, Davies Hilbert space  is defined as a quadruple (, 0, U, M)
and Lewis20 proposed abandoning the repeatability hy- consisting of a Hilbert space , a density operator 0 on
pothesis and introduced a new mathematical framework to , a unitary operator U on   , and a self-adjoint
treat all the physically realizable quantum measurements: operator M on . A measuring process (, 0, U, M) is
said to be pure if 0 is a pure state, and it is said to be
One of the crucial notions is that of repeatability which separable if  is separable.
we show is implicitly assumed in most of the axiomatic The measuring process ( , 0, U, M) mathematically
treatments of quantum mechanics, but whose abandon- models the following description of a measurement. The
ment leads to a much more flexible approach to meas- measurement is carried out by the interaction, referred to
urement theory (ref. 20, p. 239). as the measuring interaction, between the object S and
the probe P. The probe P is described by the Hilbert
Denote by c() the space of trace class operators on , space  and prepared in the state 0 just before the meas-
by () the space of density operators on , and by urement. The time evolution of the composite system
P(c()) the space of positive linear maps on c(). Da- P + S during the measuring interaction is described by the
vies and Lewis20 introduced a mathematical notion of in- unitary operator U. The outcome of the measurement is
strument as follows. A Davies–Lewis (DL) instrument for obtained by measuring the observable M called the meter
(a system S described by) a Hilbert space  is defined as observable of the probe P just after the measuring inter-
a P(c())-valued Borel measure  on R countably addi- action. We assume that the measuring interaction turns on
tive in the strong operator topology such that (R) is at time t = 0 and turns off at time t = t. In the Heisen-
trace-preserving (Tr[(R)] = Tr[]). berg picture, we write
Let A(x) be a measuring apparatus for S with the out-
put variable x. The statistical properties of the apparatus
A1 (0)  A1  1, A2 (0)  1  A2 , A12 (t )  U † A12U
A(x) are determined by (i) the probability distribution
Pr{x  ||} of the outcome x in an arbitrary state , and
(ii) the state change   {x} from the state  just be- for an observable A1 of S, an observable A2 of P, and an
fore the measurement to the state {x} just after the observable A12 of S + P.
measurement given the condition x  . The proposal of Suppose that the measurement carried out by an appa-
Davies and Lewis20 can be stated as follows. ratus A(x) is described by a measuring process (, 0,
U, M). Then, it is shown in ref. 16 that the statistical
(DL) The Davies–Lewis thesis. For every measuring properties of the apparatus A(x) are given by
apparatus A(x) with output variable x, there exists a
unique DL instrument  satisfying Pr{x  || }  Tr[ E M ( t ) ()(    0 )], (17)

CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015 2009

SPECIAL SECTION: QUANTUM MEASUREMENTS

Tr [ E M ( t ) ( )(   0 )] cess M = (, 0 , U, M) from time t = 0 to t = t. An ap-

  {x}  M ( t )
, (18) proximate simultaneous measurement of A(0) and B(0) is
Tr[ E ( )(   0 )]
obtained by direct simultaneous measurement of commut-
ing observables M(t) and B(t), where M(t) is consid-
where Tr stands for the partial trace on the Hilbert ered to approximately measure A(0), and B(t) is
space . The POVM  of the apparatus A(x) is defined considered to approximately measure B(0). In this case
by the error of the B(0) measurement is called the distur-
bance of B caused by the measuring process M, and the
 ( )  Tr [ E M ( t ) ()(1   0 )] (19) relation for the errors of the A(0) measurement and the
B(0) measurement is called the error–disturbance rela-
for any    (R). Then, the map   () is a probabil- tion (EDR). In what follows, we examine the EDR for po-
ity operator-valued measure (POVM)21 satisfying sition measurement error and momentum disturbance.
Until the 1980s, only solvable model of position meas-
Pr{x  ||}  Tr[ ()  ] (20) urement had been given by von Neumann 3. We show that
this longstanding model satisfies Heisenberg’s error–
for all   () and    (R). disturbance relation 11 , a version of Heisenberg’s uncer-
Now it is easy to see that the above description of the tainty relation (4).
measurement statistics of the apparatus A(x) is consistent Consider a one-dimensional mass S, called an object,
with the Davies–Lewis thesis. In fact, the relation with position x̂ and momentum pˆ x , described by a Hil-
bert space  = L2 (Rx), where Rx is a copy of the real line.
 ()   Tr [ E M ( t ) ( )(    0 )] (21) The object is coupled from time t = 0 to t = t with the
probe P, another one-dimensional mass with position ŷ
and momentum pˆ y , described by a Hilbert space
defines a DL instrument . In this case, we say that the
 = L2(Ry), where Ry is another copy of the real line. The
instrument  is realized by the measuring process (, 0,
outcome of the measurement is obtained by measuring
U, M).
the probe position ŷ at time t = t. The total Hamilto-
A DL instrument for  is said to be completely posi-
nian for the object and the probe is taken to be
tive (CP) if () is completely positive for every   
(R), i.e. ()idn : c()  Mn  c()  Mn is a posi- H S + P  H S  H P  KH , (22)
tive map for every finite number n, where Mn is the
matrix algebra of order n and idn is the identity map on where HS and HP are the free Hamiltonians of S and P
Mn . The following theorem characterizes the physically respectively, H represents the measuring interaction. The
realizable DL instruments by completely positivity16,22. coupling constant K satisfies Kt = 1 and it is so strong
(K  1) that HS and HP can be neglected.
Theorem 1 (Realization theorem for CP instruments). The measuring interaction H is given by
A DL instrument can be realized by a measuring process
H  xˆ  pˆ y , (23)
if and only if it is completely positive. In particular, every
CP instrument can be realized by a pure measuring proc-
ess, and if  is separable, every CP instrument for  can so that the unitary operator of the time evolution of S + P
be realized by a pure and separable measuring process. from t = 0 to t =   t is given by

Now, we have reached the following general measure-  i K 

U ( )  exp  xˆ  pˆ y  . (24)
ment axiom, abandoning Axiom 4 or the repeatability hy-   
pothesis.
Suppose that the object S and the probe P are in the
Axiom 5 (General measurement axiom). To every vector states  and , respectively, just before the meas-
measuring apparatus A(x) with output variable x, there ex- urement. We assume that the wave functions (x) and
ists a unique CP instrument  satisfying eqs (13) and (14). (y) are Schwartz rapidly decreasing functions23 . Then,
Conversely, to every instrument  there exists at least one the time evolution of S + P in the time interval (0, t) is
measuring apparatus A(x) satisfying eqs (13) and (14). given by the unitary operator U(t) = exp(ixˆ  pˆ y /).
Thus, this measuring process is represented by
( L2 (R y ), |  |, exp(ixˆ  pˆ y /), yˆ ).
von Neumann’s model of position measurement The state of the composite system S + P just after the
measurement is U(t)  . By solving the Schrödinger
Let A and B be observables of a system S described by a equation, we have
Hilbert space . Let A(x) be a measuring apparatus for S
with the output variable x described by a measuring pro- U (t )(   )( x, y )   ( x) ( y  x). (25)

2010 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015

 SPECIAL SECTION: QUANTUM MEASUREMENTS
From this, the probability distribution of output variable x yˆ (t ). Since xˆ (0) and yˆ (t ) commute by eq. (30), we
is given by have the JPD  xˆ (0), yˆ ( t ) (dx, dy ) of xˆ (0) and yˆ (t ) as

2 2
 xˆ (0), yˆ ( t ) (dx, dy )   E xˆ (0) (dx) E yˆ ( t ) (dy ), (34)
Pr{x  || }   

d y |  ( x)| | ( y  x )| d x.
R
(26)
where  stands for the mean value in the state   .
Then, by eq. (33) the rms error  ( xˆ, y ) for measuring x̂
By a property of convolution, if the probe probability dis- in state  is defined as the rms error  G ( xˆ(0), yˆ (t )) of
tribution |(y)|2 approaches the Dirac delta function  (y), yˆ (t ) for xˆ (0), so that we have
the output probability approaches the Born probability
distribution |(x)|2. 1/2
The corresponding instrument  is given by  
2 xˆ (0), yˆ ( t )
 ( xˆ ,  )     ( y  x)  (dx, dy ) 
 
 R 2

 ()    ( y1  xˆ )  ( y1  xˆ ) † dy ,
 (27)
   ( yˆ (t )  xˆ (0)) 2 1/2   yˆ (0) 2 1/2 . (35)

and the corresponding POVM is given by Since pˆ x (0) and pˆ x (t ) also commute from eq. (31), we
also have the JPD  pˆ x (0), pˆ x ( t ) (dx, dy ) of the values of
pˆ x (0) and pˆ x (t ). The rms disturbance  ( pˆ x , ) of
 ( )  | ( y1  xˆ )|2 dy,
 (28) pˆ x in state  is similarly defined as the rms error
  G ( pˆ x (0), pˆ x ()), so that we have

Solving the Heisenberg equations of motion, we have  

1/2
2 pˆ x (0), pˆ x ( t )
 ( pˆ x ,  )     ( y  x)  (dx, dy ) 
xˆ (t )  xˆ (0), (29)  
 R 2


yˆ (t )  xˆ (0)  yˆ (0), (30)

  ( pˆ x (t )  pˆ x (0)) 2 1/2   pˆ y (0) 2  1/2 . (36)

pˆ x (t )  pˆ x (0)  pˆ y (0), (31)

Then, by Kennard’s inequality (3), we have

pˆ y (t )  pˆ y (0). (32)

 ( xˆ ,  ) ( pˆ x ,  )   yˆ (0) 2  1/2  pˆ y (0) 2 1/2

Root mean square error and disturbance 

  ( yˆ (0)) ( pˆ y (0))  . (37)
2
To define the ‘mean error’ of the above position meas-
urement, let us recall classical definitions. Suppose that a Thus, the von Neumann model satisfies Heisenberg’s
quantity X = x is measured by directly observing another error–disturbance relation
quantity Y = y. For each pair of values (X, Y ) = (x, y), the
error is defined as y – x. To define the ‘mean error’ given 
the joint probability distribution (JPD) X,Y(dx, dy) of  ( xˆ ) ( pˆ x )  (38)
2
X and Y, Gauss24 introduced the root mean square (rms)
error G (X, Y) of Y for X as for  ( xˆ ) = ( xˆ,  ) and  ( pˆ x )   ( pˆ x ,  ).
By the limited availability for measurement models up
1/2
  to the 1980s, the above result appears to have enforced a
G ( X , Y )    ( y  x) 2  X ,Y (dx, dy )  , (33) prevailing belief in Heisenberg’s EDR (eq. (38)), for in-
  stance, in claiming the standard quantum limit for gravi-
 R2 
tational wave detection25–27.
which he called the ‘mean error’ or the ‘mean error to be
feared’, and has long been accepted as a standard defini- Measurement violating Heisenberg’s EDR
tion for the ‘mean error’.
In the von Neumann model, the observable xˆ (0) is In 1980, Braginsky et al.25 claimed that Heisenberg’s
measured by directly observing the meter observable EDR (eq. (38)) leads to a sensitivity limit called the

CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015 2011

SPECIAL SECTION: QUANTUM MEASUREMENTS
standard quantum limit (SQL), for gravitational wave Therefore, this model obviously violates Heisenberg’s
detectors exploiting free-mass position monitoring. Sub- EDR (eq. (38)).
sequently, Yuen28 questioned the validity of the SQL, and
Caves27 defended the SQL by giving a new formulation Universally valid error–disturbance relation
and a new proof without directly appealing to Heisen-
berg’s ERD (eq. (38)). Eventually, the conflict was rec- To derive a universally valid EDR, consider a measuring
onciled29,30 by pointing out that Caves27 still supposed process M = (, 0, U, M). If A(0) and M(t) commute,
(AR), in spite of avoiding Heisenberg’s ERD (eq. (38)). the rms error of the measuring process M for measuring
More decisively, a solvable model of a precise position A in  can be defined through the JPD of A(0) and M(t).
measurement was also constructed that breaks the Similarly, if B(0) and B(t) commute, the rms distur-
SQL29,30; later this model was shown to break Heisen- bance can also be defined through the JPD of B(0) and
berg’s EDR (eq. (38))31. B(t). In order to extend the definitions of the rms error
In what follows, we examine this model, which modi- and disturbance to the general case, we introduce the
fies the measuring interaction of the von Neumann mod- noise operator and the disturbance operator.
el. In this new model, the object, the probe, and the probe The noise operator N(A) is defined as the difference
observable, the coupling constant K, and the time dura- M(Dt) – A(0) between the observable A(0) to be measured
tion t are the same as the von Neumann model. The and the meter observable M(t) to be read and the distur-
measuring interaction is taken to be29 bance operator D(A) is defined as the change B(t) − B(0)
of B caused by the measuring interaction, i.e.

H  (2 xˆ  pˆ y  2 pˆ x  yˆ  xp
ˆ ˆ x  1  1  yp
ˆ ˆ y ). (39) N(A) = M(t) − A(0), (49)
3 3
D(B) = B(t) − B(0). (50)
The corresponding instrument is give by13
The mean noise operator n(A) and the mean disturbance
 ixpˆ x ˆ operator d(B) are defined by
 ( )   e |  |e ixp x Tr[ E xˆ (dx )  ], (40)
 n(A) = Tr [N(A)1  0], (51)

where  (x) = (−x), and the corresponding POVM is d(B)=Tr[D(B)1  0 ]. (52)

given by
The rms error (A, ) and the rms disturbance (B, ) for
qˆ observables A, B respectively, in state  are defined by
 ( )  E ( ). (41)
 ( A,  )  (Tr[ N ( A) 2    0 ])1/2 , (53)
Solving the Heisenberg equations of motion, we have
 ( B,  )  (Tr[ D ( B ) 2   0 ])1/2 . (54)
xˆ (t )  xˆ (0)  yˆ (0), (42)
An immediate meaning of  ( A,  ) and  ( B ,  ) is the
yˆ (t )  xˆ (0), (43) rms of the noise operator and the rms of the disturbance
operator respectively.
pˆ x (t )   pˆ y (0), (44) Suppose that M(t) and A(0) commute in   0, i.e.

[ E A(0) (), E M ( t ) ()]  0  0 (55)

pˆ y (t )  pˆ x (0)  pˆ y (0). (45)
for all ,    (R)32–34. In this case, the relation
Thus, xˆ (0) and yˆ (t ) commute and also pˆ x (0) and
pˆ x (t ) commute, so that the rms error and rms distur-
 A(0), M ( t ) (dx, dy )  Tr[ E A(0) (dx) E M ( t ) (dy)    0 ]
bance are well defined by their JPDs, and given by
(56)
 ( xˆ ,  )  0, (46)
defines the JPD of A(0) and M(t) satisfying
 ( pˆ x , )  ( pˆ y (0)  pˆ x (0)) 2 1/2  . (47) Tr[ p( A(0), M (t ))   0 ]

Consequently, we have
A(0), M ( t )
 ( xˆ ) ( pˆ x )  0. (48)
   p ( x, y ) 
2
(dx, dy ) (57)
R

2012 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015

 SPECIAL SECTION: QUANTUM MEASUREMENTS
for any real polynomial p(A(0), M(t)) in A(0) and M(t)32. holds for all    (R). The rms error  (A, ) satisfies
Thus, the classical rms error G(A(0), M(t)) of M(t) for that  (A, ) = 0 for all  if and only if M is probability
A(0) is well defined, and we easily obtain the relation reproducible for A in all  (refs 13, 31). Thus, the condi-
tion that  (A, ) = 0 for all  characterizes the class of
 (A, ) = G(A(0), M(t)). (58) measurements with POVM  satisfying  = EA.
Similarly, we have  (B, ) = G(B(0), B(t)) if B(0) and Busch et al.49 pointed out that there are cases where
B(t) commute in   0. (A, ) = 0 holds but M is not probability reproducible
In 2003, the present author 11,12,35 derived the relation and where M is not probability reproducible but  (A, ) = 0
holds, and questioned the reliability of the rms error
1  (A, ) as a state-dependent error measure. However,
 ( A) ( B)  |[n( A), B]  [ A, d ( B)] |  |[ A, B]|, (59) their argument lacks a reasonable definition of precise
2
measurements, necessary for discussing the reliability of
error measures. In response to their criticism, the present
where (A) = (A, ), (B) = (B, ), which is universally
author33,34 has successfully characterized the precise mea-
valid for any observables A, B, any system state , and
surements of A in a given state  and shown that the rms
any measuring process M. From eq. (59), it is concluded
error  (A, ) reliably characterizes such measurements.
that if the error and disturbance are statistically inde-
In what follows we survey those results, which have been
pendent from system state, then the Heisenberg-type EDR
mostly neglected in the recent debates50–52.
Let us start with the classical case. Suppose that a
1
 ( A) ( B )  |[ A, B]| (60) quantity X = x is measured by direct observation of an-
2 other quantity Y = y. Then, this measurement is precise
iff X = Y holds with probability 1, or equivalently the JPD
holds, extending the previous results36–39. The additional X,Y (dx, dy) of X and Y concentrates on the diagonal set,
correlation term in eq. (59) allows the error–disturbance i.e.
product  (A)(B) to violate the Heisenberg-type EDR
(eq. (60)). In general, the relation
 X ,Y ({( x, y )  R 2 |x  y})  0. (65)
1
 ( A) ( B)   ( A) ( B)   ( A) ( B )  |[ A, B ]| (61) As easily seen from eq. (33), this condition is equivalent
2 to the condition G (X, Y) = 0.
Generalizing the classical case, we say that a measur-
holds for any observables A, B, any system state , and
ing process M makes a strongly precise measurement of
any measuring process M11–13,35,40,41.
A in  iff A(0) = M(t) holds with probability 1 in the
New relation (61) leads to the following new con-
sense that A(0) and M(t) commute in   0 and that the
straints for precise measurement and non-disturbing
JPD  A(0),M(t) of concentrates on the diagonal set, i.e.
measurements:

1  A(0), M ( t ) ({( x, y )R 2 |x  y})  0. (66)

 ( A) ( B )  |[ A, B ]|, if  ( A)  0, (62)
2
On the other hand, we have introduced another opera-
1 tional requirement. The weak joint distribution
 ( A) ( B)  |[ A, B]|, if  ( B)  0. (63)
2 WA(0), M ( t ) of A(0) and M(t) in a state  is defined by
Note that if [A, B]  0, the Heisenberg-type EDR (eq.
WA(0),M ( t ) (dx, dy )  Tr[ E A(0) (dx ) E M ( t ) (dy)    0 ].
(60)) leads to the divergence of (A) or (B) in those
cases. The new error-bound eq. (63) was used to derive (67)
conservation-law-induced limits for measurements12,42–44
quantitatively generalizing the Wigner–Araki–Yanase The weak joint distribution is not necessarily positive but
theorem45–48 and was used to derive a fundamental accu- operationally accessible by weak measurement and post-
racy limit for quantum computing12. selection53. We say that the measuring process M makes a
weakly precise measurement of A in  iff the weak joint
distribution WA(0), M ( t ) in state  concentrates on the dia-
Quantum root mean square errors gonal set, i.e.

We say that the measuring process M is probability WA(0), M ( t ) ({( x, y )  R 2 |x  y})  0. (68)
reproducible for the observable A in the state  iff
This condition does not require that A(0) and M(t)
Tr[ E M ( t ) ( )   0 ]  Tr[ E A ( )  ] (64) commute, while it only requires that the weak joint
CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015 2013
SPECIAL SECTION: QUANTUM MEASUREMENTS
distribution concentrates on the event A(0) = M(t). A of measuring processes precisely measuring A in all  is
similar condition has been used to observe momentum characterized by the following equivalent conditions33,34:
transfer in a double-slit ‘which-way’ experiment54,55 . We (i) (A, ) = 0 for all   ; (ii) probability reproducible
naturally consider that strong preciseness is a sufficient for A in all   ; (iii)  = EA. The above result ensures
condition for precise measurements and weak preciseness our longstanding belief that a measurement with POVM
is a necessary condition. In the previous studies33,34, it  satisfying  = EA is considered to be precise in any
was mathematically proved that both conditions are state in the sense that the measured observable A(0) and
equivalent. Thus, either condition is concluded to be a the meter observable M(t) to be directly observed are
necessary and sufficient condition characterizing the unique perfectly correlated in any input state, not only reproduc-
class of precise measurements. As above, we say that the ing the probability distribution in any state.
measuring process M precisely measures A in  iff it We say that the measuring process M does not disturb
makes a strongly or weakly precise measurement of A in . an observable B in a state  iff observables B(0) and
To characterize the class of precise measurements in B(t) commute in the state   0 and the JPD
terms of the rms error-freeness condition, (A, ) = 0,  B (0), B ( t ) of B(0) and B(t) concentrates on the diagonal
and the probability reproducibility condition, we intro- set. The non-disturbing measuring processes defined
duce the following notions. The cyclic subspace C (A, ) above can be characterized analogously.
generated by A and  is defined as the closed subspace of From the above results, a non-zero lower bound for
 generated by {E A() |   (R),   ran()}, where (A) or (B) indicates impossibility of precise or non-
ran() denotes the range of . Then, the following theo- disturbing measurement. In particular, if  (A),  (B) < 
rem holds33,34. and [A, B]  0, then any measuring process cannot
precisely measure A without disturbing B.
Theorem 2. Let M = (, 0, U, M) be a measuring The above characterizations of precise and non-
process for the system S described by a Hilbert space . disturbing measurements suggest the following defini-
Let A be an observable of S and  a state of S. Then, the tions of the locally uniform rms error  ( A,  ) and the
following conditions are equivalent. locally uniform rms disturbance  ( B,  ) 56 :

(i) M precisely measures A in .  ( A,  )  sup  ( A,  ), (70)

(ii)  (A, ) = 0 in all    (A, ).   ( A,  )

(iii) M is probability reproducible for A in all  

 (A, ).  ( B,  )  sup  ( B,  ). (71)
  ( B ,  )

In the case where A(0) and M(t) commute, precise mea-

surements are characterized by the rms error-freeness Then, we have  ( A,  )  0 if and only if the measure-
condition, since in this case we have G (A(0), M(t)) = ment precisely measures A in , and that  ( B,  )  0 if
(A, ). However, the probability reproduciblity condi- and only if the measurement does not disturb B in . For
tion does not characterize the precise measurements even those quantities, the Heisenberg-type EDR
in this case. To see this, suppose that A(0) and M(t) are

identically distributed and independent (ref. 34, p. 763).  ( xˆ )  ( pˆ x )  (72)
Then, we have 2
is still violated by a linear position measurement56, and
2 the relation
 G ( A(0), M (t ))    ( y  x)  A(0) (d x)  M ( t ) (dy )
R 2
1
 ( A) ( B)   ( A) ( B)   ( A) ( B)  |[ A, B]| (73)
2
  ( A(0)) 2   ( M (t )) 2  ( A(0)  M (t ) ) 2 .
holds universally56, where  ( A)   ( A,  ) and
 ( B)   ( B,  ).
Since  (A(0)) =  (M(t)) and A(0) = M(t), we have
Thus, the locally uniform rms error  ( A,  ) com-
pletely characterizes precise measurements of A in  and
 G ( A(0), M ( t ))  2 ( A). (69) the locally uniform rms disturbance  ( B,  ) completely
characterizes measurements non-disturbing B in , while
Thus, M is not a precise measurement for the input state they satisfy the EDR of the same form as the rms error
 with  (A)  0. In the case where A(0) and M(t) do not and disturbance. Further investigations on quantum gen-
commute, the rms error-freeness condition well character- eralizations of the classical notion of rms error and EDRs
izes precise measurements to a similar extent to the prob- formulated with those quantities will be reported else-
ability reproducibility condition. In particular, the class where.
2014 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015
 SPECIAL SECTION: QUANTUM MEASUREMENTS

1. Heisenberg, W., The physical content of quantum kinematics and 25. Braginsky, V. B., Vorontsov, Y. I. and Thorne, K. S., Quantum
mechanics. In Quantum Theory and Measurement (eds Wheeler, nondemolition measurements. Science, 1980, 209(4456), 547–557.
J. A. and Zurek, W. H.), Princeton University Press, Princeton, 26. Caves, C. M., Thorne, K. S., Drever, R. W. P., Sandberg, V. D.
NJ, USA, 1983, pp. 62–84 [originally published: Z. Phys., 1927, and Zimmermann, M., On the measurement of a weak classical
43(3–4), 172–198]. force coupled to a quantum mechanical oscillator, I. Issues of
2. Kennard, E. H., Zur quantenmechanik einfacher bewegungstypen. principle. Rev. Mod. Phys., 1980, 52(2), 341–392.
Z. Phys., 1927, 44(4–5), 326–352. 27. Caves, C. M., Defense of the standard quantum limit for free-mass
3. von Neumann, J., Mathematical Foundations of Quantum position. Phys. Rev. Lett., 1985, 54(23), 2465–2468.
Mechanics, Princeton University Press, Princeton, NJ, 1955 [Orig- 28. Yuen, H. P., Contractive states and the standard quantum limit for
inally published: Mathematische Grundlagen der Quantenme- monitoring free-mass positions. Phys. Rev. Lett., 1983, 51(9),
chanik, Springer, 1932]. 719–722.
4. Bohm, D., Quantum Theory, Prentice-Hall, New York, 1951. 29. Ozawa, M., Measurement breaking the standard quantum limit for
5. Messiah, A., M´ecanique Quantique, Dunod, Paris, 1959, vol. I free-mass position. Phys. Rev. Lett., 1988, 60(5), 385–388.
[Quantum Mechanics, North-Holland, Amsterdam, 1959, vol. I]. 30. Ozawa, M., Realization of measurement and the standard quantum
6. Schiff, L. I., Quantum Mechanics, McGraw-Hill, New York, 1968. limit. In Squeezed and Nonclassical Light (eds Tombesi, P. and
7. Ballentine, L. E., The statistical interpretation of quantum me- Pike, E. R.), Plenum Press, New York, 1989, pp. 263–286.
chanics. Rev. Mod. Phys., 1970, 42(4), 358–381. 31. Ozawa, M., Position measuring interactions and the Heisenberg
8. Schrödinger, E., Die gegenwärtige situation in der quanten- uncertainty principle. Phys. Lett. A, 2002, 299(1), 1–7.
mechanik. Naturwissenshaften, 1935, 23(48, 49, 50), 807–812, 32. Gudder, S., Joint distributions of observables. Indiana Univ. Math.
823–828, 844–849 [English translation by Trimmer, J. D., Proc. J., 1969, 18(4), 325–335.
Am. Philos. Soc., 1980, 124(5), 323–338]. 33. Ozawa, M., Perfect correlations between noncommuting observ-
9. Braginsky, V. B. and Khalili, F. Y., Quantum Measurement, Cam- ables. Phys. Lett. A, 2005, 335(1), 11–19.
bridge University Press, Cambridge, 1992. 34. Ozawa, M., Quantum perfect correlations. Ann. Phys. (N.Y.),
10. Giovannetti, V., Lloyd, S. and Maccone, L., Quantum-enhanced 2006, 321(3), 744–769.
measurements: Beating the standard quantum limit. Science, 2004, 35. Ozawa, M., Physical content of Heisenberg’s uncertainty relation:
306(5700), 1330–1336. limitation and reformulation. Phys. Lett. A, 2003, 318(1–2), 21–
11. Ozawa, M., Universally valid reformulation of the Heisenberg 29.
uncertainty principle on noise and disturbance in measurement. 36. Arthurs, E. and Goodman, M. S., Quantum correlations: a general-
Phys. Rev. A, 2003, 67(4), 042105-1 to 042105-6. ized Heisenberg uncertainty relation. Phys. Rev. Lett., 1988,
12. Ozawa, M., Uncertainty principle for quantum instruments and 60(24), 2447–2449.
computing. Int. J. Quant. Inf., 2003, 1(4), 569–588. 37. Raymer, M. G., Uncertainty principle for joint measurement of
13. Ozawa, M., Uncertainty relations for noise and disturbance in noncommuting variables. Am. J. Phys., 1994, 62(11), 986–993.
generalized quantum measurements. Ann. Phys. (N.Y.), 2004, 38. Ozawa, M., Quantum limits of measurements and uncertainty
311(2), 350–416. principle. In Quantum Aspects of Optical Communications (eds
14. Halmos, P. R., Introduction to Hilbert Space and the Theory of Bendjaballah, C., Hirota, O. and Reynaud, S.), Springer, Berlin,
Spectral Multiplicity, Chelsea, New York, 1951. 1991, pp. 3–17.
15. Imoto, N., Ueda, M. and Ogawa, T., Microscopic theory of the 39. Ishikawa, S., Uncertainty relations in simultaneous measurements
continuous measurement of photon number. Phys. Rev. A, 1990, for arbitrary observables. Rep. Math. Phys., 1991, 29(3), 257–273.
41(7), 4127–4130. 40. Ozawa, M., Uncertainty relations for joint measurements of non-
16. Ozawa, M., Quantum measuring processes of continuous observ- commuting observables. Phys. Lett. A, 2004, 320(5–6), 367–374.
ables. J. Math. Phys., 1984, 25(1), 79–87. 41. Ozawa, M., Universal uncertainty principle in measurement opera-
17. Ozawa, M., Conditional probability and a posteriori states in quan- tor formalism. J. Opt. B: Quantum Semiclass., 2005, 7(12), S672–
tum mechanics. Publ. Res. Inst. Math. Sci., Kyoto Univ., 1985, S681.
21(2), 279–295. 42. Ozawa, M., Conservation laws, uncertainty relations, and quantum
18. Srinivas, M. D., Collapse postulate for observables with continu- limits of measurements. Phys. Rev. Lett., 2002, 88(5), 050402-1 to
ous spectra. Commun. Math. Phys., 1980, 71(2), 131–158. 050402-4.
19. Ozawa, M., Measuring processes and repeatability hypothesis. In 43. Ozawa, M., Universal uncertainty principle and quantum state
Probability Theory and Mathematical Statistics, Lecture Notes in control under conservation laws. AIP Conf. Proc., 2004, 734, 95–
Math. 1299 (eds Watanabe, S. and Prohorov, Y. V.), Springer, 98.
Berlin, 1988, pp. 412–421. 44. Busch, P. and Loveridge, L., Position measurements obeying
20. Davies, E. B. and Lewis, J. T., An operational approach to quan- momentum conservation. Phys. Rev. Lett., 2011, 106(11), 110406-
tum probability. Commun. Math. Phys., 1970, 17(3), 239–260. 1 to 110406-4.
21. Helstrom, C. W., Quantum Detection and Estimation Theory, 45. Wigner, E. P., Die messung quntenmechanischer operatoren. Z.
Academic Press, New York, 1976. Phys., 1952, 133(1–2), 101–108.
22. Ozawa, M., Conditional expectation and repeated measurements 46. Araki, H. and Yanase, M. M., Measurement of quantum mechani-
of continuous quantum observables. In Probability Theory and cal operators. Phys. Rev., 1960, 120(2), 622–626.
Mathematical Statistics, Lecture Notes in Math. 1021 (eds Itô, K. 47. Yanase, M. M., Optimal measuring apparatus. Phys. Rev., 1961,
and Prohorov, J. V.), Springer, Berlin, 1983, pp. 518–525. 123(2), 666–668.
23. Reed, M. and Simon, B., Methods of Modern Mathematical Phys- 48. Ozawa, M., Does a conservation law limit position measurements?
ics, I: Functional Analysis (Revised and Enlarged Edition), Aca- Phys. Rev. Lett., 1991, 67(15), 1956–1959.
demic Press, 1980. 49. Busch, P., Heinonen, T. and Lahti, P., Noise and disturbance in
24. Gauss, C. F., Theory of the Combination of Observations Least quantum measurement. Phys. Lett. A, 2004, 320(4), 261–270.
Subject to Errors: Part One, Part Two, Supplement, SIAM, Phila- 50. Dressel, J. and Nori, F., Certainty in Heisenberg’s uncertainty
delphia, USA, 1995 [originally published: Theoria Combinationis principle: revisiting definitions for estimation errors and distur-
Observationum Erroribus Miinimis Obnoxiae, Pars Prior, Pars bance. Phys. Rev. A, 2014, 89(2), 022106-1 to 022106-14.
Posterior, Supplementum, Societati Regiae Scientiarum Exhibita, 51. Korzekwa, K., Jennings, D. and Rudolph, T., Operational
15 February 1821]. constraints on state-dependent formulations of quantum

CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015 2015

SPECIAL SECTION: QUANTUM MEASUREMENTS
error-disturbance tradeoff relations. Phys. Rev. A, 2014, 89(5), experiment on the complementarity-uncertainty debate. New J.
052108-1 to 052108-6. Phys., 2007, 9, 287-1 to 287-11.
52. Busch, P., Lahti, P. and Werner, R. F., Colloquium: Quantum 56. Ozawa, M., Noise and disturbance in quantum measurements and
root-mean-square error and measurement uncertainty relations. operations. Proc. SPIE, 2006, 6244, 62440Q-1 to 62440Q-9.
Rev. Mod. Phys., 2014, 86(4), 1261–1281. 57. Ozawa, M., Heisenberg’s uncertainty relation: violation and
53. Lund, A. P. and Wiseman, H. M., Measuring measurement– reformulation. J. Phys. Conf. Ser., 2014, 504, 012024-1 to
disturbance relationships with weak values. New J. Phys., 2010, 012024-12.
12, 093011-1 to 093011-11.
54. Garretson, J. L., Wiseman, H. M., Pope, D. T. and Pegg, D. T., ACKNOWLEDGEMENTS. This work was supported in part by JSPS
The uncertainty relation in ‘which-way’ experiments: how to ob- KAKENHI, No. 26247016 and No. 15K13456, and the John Templeton
serve directly the momentum transfer using weak values. J. Opt. Foundation, ID #35771. Some parts of the results presented here have
B, 2004, 6(6), S506–S517. appeared in an earlier conference review article57 .
55. Mir, R., Lundeen, J. S., Mitchell, M. W., Steinberg, A. M., Garret-
son, J. L. and Wiseman, H. M., A double-slit ‘which-way’ doi: 10.18520/v109/i11/2006-2016

2016 CURRENT SCIENCE, VOL. 109, NO. 11, 10 DECEMBER 2015