See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/282602093

Heisenberg's Original Derivation of the Uncertainty Principle and its Universally

Valid Reformulations

Article  in  Current Science · July 2015

DOI: 10.18520/cs/v109/i11/2006-2016

CITATIONS READS

13 7,625

1 author:

Masanao Ozawa
Nagoya University
160 PUBLICATIONS   4,427 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Complementarity View project

All content following this page was uploaded by Masanao Ozawa on 14 July 2020.

The user has requested enhancement of the downloaded file.

 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∗

Heisenberg’s uncertainty principle was originally posed for the limit of the accuracy of simultaneous mea-
surement of non-commuting observables as stating that canonically conjugate observables can be measured
simultaneously only with the constraint that the product of their mean errors should be no less than a limit set
by Planck’s constant. However, Heisenberg with the subsequent completion by Kennard has long been credited
only with a constraint for state preparation represented by the product of the standard deviations. Here, we show
that Heisenberg actually proved the constraint for the accuracy of simultaneous measurement but assuming an
obsolete postulate for quantum mechanics. This assumption, known as the repeatability hypothesis, formu-
arXiv:1507.02010v3 [quant-ph] 11 Mar 2020

lated explicitly by von Neumann and Schrödinger, was broadly accepted until the 1970s, but abandoned in the
1980s, when completely general quantum measurement theory was established. We also survey the author’s re-
cent proposal for a universally valid reformulation of Heisenberg’s uncertainty principle under the most general
assumption on quantum measurement.
Keywords: quantum measurement, uncertainty principle, simultaneous measurement, repeatability hypothesis, instruments,
root mean square error

I. INTRODUCTION for arbitrary wave functions. By this relation, the lower bound
of relation (1) was later set as
The uncertainty principle proposed by Heisenberg1 in 1927
h̄
revealed that we cannot determine both position and momen- ε (q̂)ε ( p̂) ≥ , (4)
tum of a particle simultaneously in microscopic scale as stat- 2
ing “the more precisely the position is determined, the less
where h̄ = h/(2π ).
precisely the momentum is known, and conversely”1 (p. 64),
Text books3–6 up to the 1960s often explained that the
and had overturned the deterministic world view based on
physical meaning of Heisenberg’s uncertainty principle is ex-
the Newtonian mechanics. By the famous γ ray microscope
pressed by Eq. (4), but it is formally expressed by Eq. (3). This
thought experiment Heisenberg1 derived the relation
explanation is later considered to be confusing. In fact, it was
said that Eq. (4) expresses a limitation of measurements, while
ε (q̂)ε ( p̂) ∼ h (1)
mathematically derived relation Eq. (3) expresses a statistical
for ε (q̂), the “mean error” of the position measurement, and property of quantum state, or a limitation of state preparations,
ε ( p̂), thereby caused “discontinuous change” of the momen- so that they have different meanings7. Thus, Heisenberg with
tum, or more generally the mean error of the simultaneous the subsequent completion by Kennard has long been cred-
momentum measurement, where h is Planck’s constant: ited only with a constraint for state preparation represented by
Eq. (3).
Let ε (q̂) [originally, q1 ] be the precision with This paper aims to resolve this long standing confusion. It
which the value q is known (ε (q̂) is, say, the mean will be shown that Heisenberg1 in 1927 actually “proved” not
error of q), therefore here the wavelength of the only Eq. (2) but also Eq. (1) from basic postulates for quan-
light. Let ε ( p̂) [originally, p1 ] be the precision tum mechanics. In showing that, it is pointed out that as
with which the value p is determinable; that is, one of the basic postulates Heisenberg supposed an assump-
here, the discontinuous change of p in the Comp- tion called the “repeatability hypothesis”, which is now con-
ton effect1 (p. 64). sidered to be obsolete. In fact, in the 1930’s the repeatabil-
ity hypothesis was explicitly claimed by von Neumann3 and
Heisenberg claimed that this relation is a “straightforward Schrödinger8, whereas this hypothesis was abandoned in the
mathematical consequence”1 (p. 65) of fundamental postu- 1980s, when quantum measurement theory was establish to be
lates for quantum mechanics. In his mathematical derivation general enough to treat all the physically realizable measure-
of relation (1), he derived ments.
Through those examinations it will be concluded that
h̄ Heisenberg’s uncertainty principle expressed by Eq. (4) is log-
σ (q̂) σ ( p̂) = (2)
2 ically a straightforward consequence of Eq. (3) under a gener-
for standard deviations σ (q̂) and σ ( p̂) of position q̂ and alized form of the repeatability hypothesis. In fact, under the
momentum p̂ for a class of Gaussian wave functions, later repeatability hypothesis a measurement is required to prepare
known as minimum uncertainty wave packets. Subsequently, the state with a sharp value of the measured observable, and
Kennard2 proved the inequality hence the “measuremental” uncertainty relation (4) is a logical
consequence of the “preparational” uncertainty relation (3).
h̄ As stated above, the repeatability hypothesis was aban-
σ (q̂) σ ( p̂) ≥ (3) doned in the 1980s, and nowadays relation (4) is taken to be
2
 2

a breakable limit9,10 . Naturally, the problem remains: what generalizing a feature of the Compton-Simons experiment3
is the unbreakable constraint for simultaneous measurements (pp. 212–214).
of non-commuting observables? To answer this question, we
will survey the author’s recent proposal11–13 for a universally (R) Repeatability hypothesis. If an observable A is mea-
valid reformulation of Heisenberg’s uncertainty principle un- sured twice in succession in a system S, then we get the same
der the most general assumption on quantum measurement. value each time.

It can be seen from the following definition of measurement

II. REPEATABILITY HYPOTHESIS due to Schrödinger given in his famous “cat paradox” paper8
that von Neumann’s repeatability hypothesis was broadly ac-
The uncertainty principle was introduced by Heisenberg in cepted in the 1930s.
a paper entitled Über den anschaulichen Inhalt der quanten-
theoretischen Kinematik und Mechanik1 published in 1927. In The systematically arranged interaction of two
what follows we shall examine Heisenberg’s derivation of the systems (measured object and measuring instru-
uncertainty principle following this paper. ment) is called a measurement on the first sys-
Before examining the detail of Heisenberg’s derivation, we tem, if a directly-sensible variable feature of the
shall examine the basic postulates for quantum mechanics in second (pointer position) is always reproduced
Heisenberg’s time, following von Neumann’s formulation3. within certain error limits when the process is im-
In what follows, a positive operator on a Hilbert space with mediately repeated (on the same object, which in
unit trace is called a density operator. We denote by B(R) the meantime must not be exposed to any addi-
the set of Borel subsets of R and by E A the spectral measure tional influences)8.
of a self-adjoint operator A, i.e., A has the spectral decompo-
sition A = R λ E A (d λ ). Based on the repeatability hypothesis von Neumann3
R
proved the impossibility of simultaneous measurement of two
Axiom 1 (States and observables). Every quantum system S non-commuting observables as follows. Suppose that two ob-
is described by a Hilbert space H called the state space of servables A, B are simultaneously measurable in every state
S. States of S are represented by density operators on H and and suppose that the eigenvalues of A are non-degenerate.
observables of S are represented by self-adjoint operators on Then, the state just after the simultaneous measurement of A
H. and B is a common eigenstate of A and B, so that there is an
orthonormal basis consisting of common eigenstates of A and
Axiom 2 (Born statistical formula). If an observable A is B, concluding that A and B commute.
measured in a state ρ , the outcome obeys the probability dis- Since Heisenberg’s uncertainty principle concerns mea-
tribution of A in ρ defined by surements with errors, it is naturally expected that it can be
Pr{A ∈ ∆kρ } = Tr[E A (∆)ρ ], (5) mathematically derived by extending the above argument to
approximate measurements.
where ∆ ∈ B(R).

Axiom 3 (Time evolution). Suppose that a system S is an III. APPROXIMATE REPEATABILITY HYPOTHESIS
isolated system with the (time-independent) Hamiltonian H
from time t to t + τ . The system S is in a state ρ (t) at time t if To extend the repeatability hypothesis to approximate mea-
and only if S is in the state ρ (t + τ ) at time t + τ satisfying surements, we generalize the notion of eigenstates as follows.
For any real number λ and a positive number ε , a (vector)
ρ (t + τ ) = e−iτ H/h̄ ρ (t)eiτ H/h̄ . (6) state ψ is called an ε -approximate eigenstate belonging to λ
iff the relation
Under the above axioms, we can make a probabilistic pre-
diction of the result of a future measurement from the knowl-
kAψ − λ ψ k ≤ ε (7)
edge about the past state. However, such a prediction applies
only to a single measurement in the future. If we make many holds. If ε = 0, the notion of ε -approximate eigenstates is
measurements successively, we need another axiom to deter- reduced to the ordinary notion of eigenstates. A real number
mine the state after each measurement. For this purpose, the λ is called an approximate eigenvalue of an observable A iff
following axiom was broadly accepted in the 1930s. for every ε > 0 there exists an ε -approximate eigenstate of
Axiom 4 (Measurement axiom). If an observable A is mea- A. The set of approximate eigenvalues of an observable A
sured in a system S to obtain the outcome a, then the system S coincides with the spectrum of A14 (p. 52).
is left in an eigenstate of A belonging to a. Now, we formulate the approximate repeatability hypothe-
sis as follows.
Von Neuamann3 showed that this assumption is equiv-
alent to the following assumption called the repeatability (AR) Approximate Repeatability Hypothesis. If an ob-
hypothesis3 (p. 335), posed with a clear operational condition servable A is measured in a system S with mean error ε
 3

to obtain the outcome a, then the system S is left in an ε - or its approximate version as an additional but obsolete as-
approximate eigenstate of A belonging to a. sumption in addition to the standard postulates for quantum
mechanics.
Obviously, (AR) is reduced to (R) for ε = 0. Since we have The approximate repeatability hypothesis (AR) has not
been explicitly formulated in the literature, but in the follow-
kAψ − λ ψ k ≥ kAψ − hAiψ k = σ (A) ing explanation on the derivation of the uncertainty principle
von Neumann3 (pp. 238–239) assumed (AR):
for any real number λ , where hAi = (ψ , Aψ ), (AR) implies
the following statement: If an observable A in a system S is We are then to show that if Q, P are two canon-
measured with mean error ε (A), then the post-measurement ically conjugate quantities, and a system is in a
standard deviation σ (A) of A satisfies state in which the value of Q can be given with the
accuracy ε (i.e., by a Q measurement with an er-
σ (A) ≤ ε (A). (8) ror range ε ), then P can be known with no greater
accuracy than η = h̄/(2ε ). Or: a measurement of
Q with the accuracy ε must bring about an inde-
terminacy η = h̄/(2ε ) in the value of P.
IV. HEISENBERG’S DERIVATION OF THE In the above, it is obviously assumed that a state with the posi-
UNCERTAINTY PRINCIPLE tion standard deviation ε is resulted by a Q measurement with
an error range ε . This assumption is what we have generally
Heisenberg’s derivation of (1) starts with considering a state formulated in Eq. (8) as an immediate logical consequence of
ψ just after the measurement of the position observable q̂ to (AR).
obtain the outcome q′ with mean error ε (q̂) and consider what Two inequalities (3) and (4) are often distinguished as the
relation holds between ε (q̂) and ε ( p̂) if the momentum ob- preparational uncertainty relation and the measuremental un-
servable p̂ has been measured simultaneously to obtain the certainty relation, respectively. However, under the repeata-
outcome p′ with mean error ε ( p̂). Then, by (AR) or Eq. (8) bility hypothesis such a distinction is not apparent, since a
the state ψ should have the position standard deviation σ (q̂) measurement is required to prepare the state with a sharp
satsifying value of the measured observable. In fact, the above argument
shows that there exists an immediate logical relationship be-
σ (q̂) ≤ ε (q̂). (9) tween those two inequalities.

Heisenberg actually supposed that the state ψ is a Gaussian

wave function1 (p. 69) V. ABANDONING THE REPEATABILITY HYPOTHESIS

(q − q′)2 i ′
 
1 ′ The repeatability hypothesis explains only a restricted class
ψ (q) = exp − − p (q − q ) , (10)
(π q21 )1/4 2q21 h̄ of measurements and does not generally characterize the state
changes caused by quantum measurements. In fact, there
which is later known as a minimum uncertainty wave packet, exist commonly used measurements of discrete observables,
with its Fourier transform such as photon counting, that do not satisfy the repeatability
hypothesis15. Moreover, it has been shown that the repeata-
(p − p′)2 i ′
 
1 ′
ψ̂ (p) = exp − + q (p − p ) , (11) bility hypothesis cannot be generalized to continuous observ-
(π p21 )1/4 2p21 h̄ ables in the standard formulation of quantum mechanics16–19 .
In 1970, Davies and Lewis20 proposed abandoning the re-
and he proved relation (2) for the state ψ given by Eq. (10).
peatability hypothesis and introduced a new mathematical
Exactly this part of Heisenberg’s argument was generalized
framework to treat all the physically realizable quantum mea-
by Kennard2 to prove relation (3) for any vector state ψ . Thus,
surements:
Kennard2 relaxed Heisenberg’s assumption on the state ψ to
the assumption that the state ψ after the position measurement One of the crucial notions is that of repeatabil-
can be arbitrary wave function ψ satisfying Eq. (9). Then, if ity which we show is implicitly assumed in most
the momentum observable p̂ has been measured simultane- of the axiomatic treatments of quantum mechan-
ously to obtain the outcome p′ with an error ε ( p̂), by (AR) or ics, but whose abandonment leads to a much
Eq. (8) again the state ψ should also satisfy the relation more flexible approach to measurement theory20
(p. 239).
σ ( p̂) ≤ ε ( p̂). (12)
Denote by τ c(H ) the space of trace class operators on
Therefore, Heisenberg’s uncertainty relation (4) immediately H , by S (H ) the space of density operators on H , and
follows from Kennard’s relation (3). by P(τ c(H )) the space of positive linear maps on τ c(H ).
As above Heisenberg in 1927 not only derived relation (1) Davies and Lewis20 introduced a mathematical notion of in-
by the γ -ray thought experiment but also gave its mathemati- strument as follows. A Davies-Lewis (DL) instrument for
cal proof. However, he supposed the repeatability hypothesis (a system S described by) a Hilbert space H is defined as
 4

a P(τ c(H ))-valued Borel measure I on R countably addi- for an observable A1 of S, an observable A2 of P, and an ob-
tive in the strong operator topology such that I (R) is trace- servable A12 (0) of S + P.
preserving (Tr[I (R)ρ ] = Tr[ρ ]). Suppose that the measurement carried out by an appara-
Let A(x) be a measuring apparatus for S with the output tus A(x) is described by a measuring process (K , ρ0 ,U, M).
variable x. The statistical properties of the apparatus A(x) are Then, it is shown16 that the statistical properties of the appa-
determined by (i) the probability distribution Pr{x ∈ ∆kρ } of ratus A(x) is given by
the outcome x in an arbitrary state ρ , and (ii) the state change
ρ → ρ{x∈∆} from the state ρ just before the measurement to Pr{x ∈ ∆kρ } = Tr[E M(∆t) (∆)(ρ ⊗ ρ0)], (17)
the state ρ{x∈∆} just after the measurement given the condition TrK [(1 ⊗ E M (∆))U(ρ ⊗ ρ 0 )U † ]
ρ → ρ{x∈∆} = , (18)
x ∈ ∆. The proposal of Davies and Lewis20 can be stated as Tr[E M(∆t) (∆)(ρ ⊗ ρ0)]
follows.
where TrK stands for the partial trace on the Hilbert space
(DL) The Davies-Lewis thesis. For every measuring ap- K . The POVM Π of the apparatus A(x) is defined by
paratus A(x) with output variable x there exists a unique DL
instrument I satisfying Π(∆) = TrK [E M(∆t) (∆)(1 ⊗ ρ0)] (19)

Pr{x ∈ ∆kρ } = Tr[I (∆)ρ ], (13) for any ∆ ∈ B(R). Then, the map ∆ → Π(∆) is a probability
I (∆)ρ operator-valued measure (POVM)21 satisfying
ρ → ρ{x∈∆} = . (14)
Tr[I (∆)ρ ] Pr{x ∈ ∆kρ } = Tr[Π(∆)ρ ] (20)

for all ρ ∈ S (H ) and ∆ ∈ B(R).

For any ∆ ∈ B(R), define Π(∆) by Now it is easy to see that the above description of the mea-
surement statistics of the apparatus A(x) is consistent with the
Π(∆) = I (∆)∗ 1, (15) Davies-Lewis thesis. In fact, the relation
I (∆)ρ = TrK 1 ⊗ E M (∆) U(ρ ⊗ ρ0 )U †


where I (∆)∗ is the dual map of I (∆) given by (21)
Tr[(I (∆)∗ X)ρ ] = Tr[X(I (∆)ρ )] for all X ∈ L (H ). Then,
defines a DL instrument I . In this case, we say that
the map ∆ → Π(∆) is a probability operator-valued measure
the instrument I is realized by the measuring process
(POVM)21 , called the POVM of I , satisfying
(K , ρ0 ,U, M).
Pr{x ∈ ∆kρ } = Tr[Π(∆)ρ ] (16) A DL instrument for H is said to be completely positive
(CP) if I (∆) is completely positive for every ∆ ∈ B(R), i.e.,
for all ρ ∈ S (H ) and ∆ ∈ B(R). I (∆) ⊗ idn : τ c(H ) ⊗ Mn → τ c(H ) ⊗ Mn is a positive map
The problem of mathematically characterizing all the physi- for every finite number n, where Mn is the matrix algebra of
cally realizable quantum measurements is reduced to the prob- order n and idn is the identity map on Mn . The following theo-
lem as to which instruments are physically realizable13 . To rem characterizes the physically realizable DL instruments by
settle this problem, standard models of measuring processes completely positivity16,22 .
were introduced in16 as follows. A measuring process for
Theorem 1 (Realization theorem for CP instruments). A
(a system described by) a Hilbert space H is defined as a
DL instrument can be realized by a measuring process if and
quadruple (K , ρ0 ,U, M) consisting of a Hilbert space K , a
only if it is completely positive. In particular, every CP instru-
density operator ρ0 on K , a unitary operator U on H ⊗ K ,
ment can be realized by a pure measuring process, and if H
and a self-adjoint operator M on K . A measuring process
is separable, every CP instrument for H can be realized by a
(K , ρ0 ,U, M) is said to be pure if ρ0 is a pure state, and it is
pure and separable measuring process.
said to be separable if K is separable.
The measuring process (K , ρ0 ,U, M) mathematically Now, we have reached the following general measurement
models the following description of a measurement. The mea- axiom, abandoning Axiom 4 or the repeatability hypothesis.
surement is carried out by the interaction, referred to as the
measuring interaction, between the object S and the probe P. Axiom 5 (General measurement axiom). To every mea-
The probe P is described by the Hilbert space K and prepared suring apparatus A(x) with output variable x there exists a
in the state ρ0 just before the measurement. The time evolu- unique CP instrument I satisfying Eqs. (13) and (14). Con-
tion of the composite system P + S during the measuring in- versely, to every instrument I there exists at least one mea-
teraction is described by the unitary operator U. The outcome suring apparatus A(x) satisfying Eqs. (13) and (14).
of the measurement is obtained by measuring the observable
M called the meter observable of the probe P just after the
measuring interaction. We assume that the measuring interac- VI. VON NEUMANN’S MODEL OF POSITION
MEASUREMENT
tion turns on at time t = 0 and turns off at time t = ∆t. In the
Heisenberg picture, we write
Let A and B be observables of a system S described by a
A1 (0) = A1 ⊗ 1, A2 (0) = 1 ⊗ A2, A12 (∆t) = U † A12 (0)U, Hilbert space H . Let A(x) be a measuring apparatus for S
 5

with the output variable x described by a measuring process By a property of convolution, if the probe probability distribu-
M = (K , ρ0 ,U, M) from time t = 0 to t = ∆t. An approxi- tion |ξ (y)|2 approaches to the Dirac delta function δ (y), the
mate simultaneous measurement of A(0) and B(0) is obtained output probability approaches to the Born probability distri-
by direct simultaneous measurement of commuting observ- bution |ψ (x)|2 .
ables M(∆t) and B(∆t), where M(∆t) is considered to ap- The corresponding instrument I is given by
proximately measure A(0) and B(∆t) is considered to approx- Z
imately measure B(0). In this case the error of the B(0) mea- I (∆)ρ = ξ (y1 − x̂)ρξ (y1 − x̂)† dy, (27)
surement is called the disturbance of B caused by the mea- ∆
suring process M, and the relation for the errors of the A(0)
measurement and the B(0) measurement is called the error- and the corresponding POVM is given by
disturbance relation (EDR). In what follows, we examine the Z
EDR for position measurement error and momentum distur- Π(∆) = |ξ (y1 − x̂)|2 dy, (28)
∆
bance.
Until 1980’s only solvable model of position measure- Solving the Heisenberg equations of motion, we have
ment had been given by von Neumann3. We show that this
long-standing model satisfies Heisenberg’s error-disturbance x̂(∆t) = x̂(0), (29)
relation11, a version of Heisenberg’s uncertainty relation (4). ŷ(∆t) = x̂(0) + ŷ(0), (30)
Consider a one-dimensional mass S, called an object, with
p̂x (∆t) = p̂x (0) − p̂y(0), (31)
position x̂ and momentum p̂x , described by a Hilbert space
H = L2 (Rx ), where Rx is a copy of the real line. The object p̂y (∆t) = p̂y (0). (32)
is coupled from time t = 0 to t = ∆t with the probe P, another
one-dimensional mass with position ŷ and momentum p̂y , de-
scribed by a Hilbert space K = L2 (Ry ), where Ry is another VII. ROOT-MEAN-SQUARE ERROR AND DISTURBANCE
copy of the real line. The outcome of the measurement is ob-
tained by measuring the probe position ŷ at time t = ∆t. The To define the “mean error” of the above position measure-
total Hamiltonian for the object and the probe is taken to be ment, let us recall classical definitions. Suppose that a quan-
tity X = x is measured by directly observing another quantity
HS+P = HS + HP + KH, (22) Y = y. For each pair of values (X,Y ) = (x, y), the error is de-
fined as y − x. To define the “mean error” given the joint prob-
where HS and HP are the free Hamiltonians of S and P, respec-
ability distribution (JPD) µ X,Y (dx, dy) of X and Y , Gauss24
tively, H represents the measuring interaction. The coupling
introduced the root-mean-square (rms) error εG (X,Y ) of Y
constant K satisfies K∆t = 1 and it is so strong (K ≫ 1) that
for X as
HS and HP can be neglected.
The measuring interaction H is given by ZZ 1/2
εG (X,Y ) = (y − x)2 µ X,Y (dx, dy) , (33)
H = x̂ ⊗ p̂y , (23) R2

so that the unitary operator of the time evolution of S + P from which Gauss24 called the “mean error” or the “mean error to
t = 0 to t = τ ≤ ∆t is given by be feared”, and has long been accepted as a standard definition
for the “mean error”.
−iK τ
 
U(τ ) = exp x̂ ⊗ p̂y . (24) In the von Neumann model, the observable x̂(0) is mea-
h̄ sured by directly observing the meter observable ŷ(∆t). Since
Suppose that the object S and the probe P are in the vector x̂(0) and ŷ(∆t) commute by Eq. (30), we have the JPD
states ψ and ξ , respectively, just before the measurement. We µ x̂(0),ŷ(∆t) (dx, dy) of x̂(0) and ŷ(∆t) as
assume that the wave functions ψ (x) and ξ (y) are Schwartz
rapidly decreasing functions23. Then, the time evolution of µ x̂(0),ŷ(∆t) (dx, dy) = hE x̂(0) (dx)E ŷ(∆t) (dy)i, (34)
S + P in the time interval (0, ∆t) is given by the unitary opera-
tor U(∆t) = e−ix̂⊗ p̂y /h̄ . Thus, this measuring process is repre- where h· · · i stands for the mean value in the state ψ ⊗ ξ . Then,
by Eq. (33) the rms error ε (x̂, ψ ) for measuring x̂ in state ψ is
sented by (L2 (Ry ), |ξ ihξ |, e−ix̂⊗ p̂y /h̄ , ŷ).
defined as the rms error εG (x̂(0), ŷ(∆t)) of ŷ(∆t) for x̂(0), so
The state of the composite system S + P just after the mea-
that we have
surement is U(∆t)ψ ⊗ ξ . By solving the Schrödinger equa-
tion, we have ZZ 1/2
2 x̂(0),ŷ(∆t)
ε (x̂, ψ ) = (y − x) µ (dx, dy)
U(∆t)(ψ ⊗ ξ )(x, y) = ψ (x)ξ (y − x). (25) R2

= h(ŷ(∆t) − x̂(0))2 i1/2 = hŷ(0)2 i1/2 . (35)

From this, the probability distribution of output variable x is
given by Since p̂x (0) and p̂x (∆t) also commute from Eq. (31), we
Z Z
also have the JPD µ p̂x (0), p̂x (∆t) (dx, dy) of the values of p̂x (0)
Pr{x ∈ ∆kψ } = dy |ψ (x)|2 |ξ (y − x)|2 dx. (26) and p̂x (∆t). The rms disturbance η ( p̂x , ψ ) of p̂x in state ψ is
∆ R
 6

similarly defined as the rms error εG ( p̂x (0), p̂x (∆)), so that we Solving the Heisenberg equations of motion, we have
have
ZZ 1/2 x̂(∆t) = x̂(0) − ŷ(0), (42)
η ( p̂x , ψ ) = (y − x)2 µ p̂x (0), p̂x (∆t) (dx, dy) ŷ(∆t) = x̂(0), (43)
R2 p̂x (∆t) = − p̂y (0), (44)
= h( p̂x (∆t) − p̂x(0))2 i1/2 = h p̂y (0)2 i1/2 . (36) p̂y (∆t) = p̂x (0) + p̂y (0). (45)
Then, by Kennard’s inequality (3) we have Thus, x̂(0) and ŷ(∆t) commute and also p̂x (0) and p̂x (∆t)
commute, so that the rms error and the rms disturbance are
ε (x̂, ψ )η ( p̂x , ψ ) = hŷ(0)2 i1/2 h p̂y (0)2 i1/2 well defined by their JPDs, and given by
h̄
≥ σ (ŷ(0))σ ( p̂y (0)) ≥ . (37) ε (x̂, ψ ) = 0, (46)
2
2 1/2
Thus, the von Neumann model satisfies Heisenberg’s error- η ( p̂x , ψ ) = h( p̂y (0) + p̂x (0)) i < ∞. (47)
disturbance relation (EDR)
Consequently, we have
h̄
ε (x̂)η ( p̂x ) ≥ (38) ε (x̂)η ( p̂x ) = 0. (48)
2
for ε (x̂) = ε (x̂, ψ ) and η ( p̂x ) = η ( p̂x , ψ ). Therefore, this model obviously violates Heisenberg’s EDR
By the limited availability for measurement models up to (38).
the 1980’s, the above result appears to have enforced a pre-
vailing belief in Heisenberg’s EDR (38), for instance, in
IX. UNIVERSALLY VALID ERROR-DISTURBANCE
claiming the standard quantum limit for gravitational wave
RELATION
detection25–27 .
To derive a universally valid EDR, consider a measuring
VIII. MEASUREMENT VIOLATING HEISENBERG’S EDR process M = (K , ρ0 ,U, M). If A(0) and M(∆t) commute, the
rms error of the measuring process M for measuring A in ρ
In 1980, Braginsky, Vorontsov, and Thorne25 claimed that can be defined through the JPD of A(0) and M(∆t). Similarly,
Heisenberg’s EDR (38) leads to a sensitivity limit, called the if B(0) and B(∆t) commute, the rms disturbance can also be
standard quantum limit (SQL), for gravitational wave detec- defined through the JPD of B(0) and B(∆t). In order to extend
tors exploiting free-mass position monitoring. Subsequently, the definitions of the rms error and disturbance to the general
Yuen28 questioned the validity of the SQL, and then Caves27 case, we introduce the noise operator and the disturbance op-
defended the SQL by giving a new formulation and a new erator.
proof without directly appealing to Heisenberg’s ERD (38). The noise operator N(A) is defined as the difference
Eventually, the conflict was reconciled29,30 by pointing out M(∆t) − A(0) between the observable A(0) to be measured
that Caves27 still supposed (AR), in spite of avoiding Heisen- and the meter observable M(∆t) to be read and the disturbance
berg’s ERD (38). More decisively, a solvable model of a pre- operator D(A) is defined as the the change B(∆t) − B(0) of B
cise position measurement was also constructed that breaks caused by the measuring interaction, i.e.,
the SQL29,30 ; later this model was shown to break Heisen- N(A) = M(∆t) − A(0), (49)
berg’s EDR (38)31 .
D(B) = B(∆t) − B(0). (50)
In what follows, we examine this model, which modifies the
measuring interaction of the von Neumann model. In this new The mean noise operator n(A) and the mean disturbance op-
model, the object, the probe, and the probe observables, the erator d(B) are defined by
coupling constant K, and the time duration ∆t are the same as
the von Neumann model. The measuring interaction is taken n(A) = TrK [N(A)1 ⊗ ρ0], (51)
to be29 d(B) = TrK [D(B)1 ⊗ ρ0]. (52)
π
H = √ (2x̂ ⊗ p̂y − 2 p̂x ⊗ ŷ + x̂ p̂x ⊗ 1 − 1 ⊗ ŷp̂y ). (39) The rms error ε (A, ρ ) and the rms disturbance η (B, ρ ) for
3 3 observables A, B, respectively, in state ρ are defined by
The corresponding instrument is give by13 ε (A, ρ ) = (Tr[N(A)2 ρ ⊗ ρ0])1/2 , (53)
2 1/2
η (B, ρ ) = (Tr[D(B) ρ ⊗ ρ0 ]) . (54)
Z
I (∆)ρ = e−ix p̂x |φ ihφ |e−ix p̂x Tr[E x̂ (dx)ρ ], (40)
∆
An immediate meaning of ε (A, ρ ) and η (B, ρ ) are the rms’s
where φ (x) = ξ (−x), and the corresponding POVM is given of the noise operator and the disturbance operator.
by Suppose that M(∆t) and A(0) commute in ρ ⊗ ρ0 , i.e.,

Π(∆) = E A (∆). (41) [E A(0) (∆), E M(∆t) (Γ)]ρ ⊗ ρ0 = 0 (55)

 7

for all ∆, Γ ∈ B(R)32–34 . In this case, the relation holds for all ∆ ∈ B(R). The rms error ε (A, ρ ) satisfies that
ρ (A, ρ ) = 0 for all ρ if and only if M is probability repro-
µ A(0),M(∆t) (dx, dy) = Tr[E A(0) (dx)E M(∆t) (dy)ρ ⊗ ρ0 ] (56) ducible for A in all ρ 13,31 . Thus, the condition that ε (A, ρ ) = 0
defines the JPD of A(0) and M(∆t) satisfying for all ρ characterizes the class of measurements with POVM
Π satisfying Π = E A .
Busch, Heinonen, and Lahti49 pointed out that there are
ZZ
Tr[p(A(0), M(∆t))ρ ⊗ ρ0 ] = p(x, y) µ A(0),M(∆t) (dx, dy)
R2 cases where ε (A, ρ ) = 0 holds but M is not probability re-
(57) producible and where M is not probability reproducible but
for any real polynomial p(A(0), M(∆t)) in A(0) and M(∆t)32 . ε (A, ρ ) = 0 holds, and questioned the reliability of the rms er-
Thus, the classical rms error εG (A(0), M(∆t)) of M(∆t) for ror ε (A.ρ ) as a state-dependent error measure. However, their
A(0) is well defined, and we easily obtain the relation argument lacks a reasonable definition of precise measure-
ε (A, ρ ) = εG (A(0), M(∆t)). (58) ments, necessary for discussing the reliability of error mea-
sures. In response to their criticism, the present author33,34
Similarly, we have η (B, ρ ) = εG (B(0), B(∆t)) if B(0) and has successfully characterized the precise measurements of A
B(∆t) commute in ρ ⊗ ρ0 . in a given state ρ and shown that the rms error ε (A, ρ ) re-
In 2003, the present author11,12,35 derived the relation liably characterizes such measurements. In what follows we
survey those results, which were mostly neglected in the re-
1
ε (A)η (B) + |h[n(A), B]i + h[A, d(B)]i| ≥ |h[A, B]i| , (59) cent debates50–52 .
2 Let us start with the classical case. Suppose that a quantity
where ε (A) = ε (A, ρ ), η (B) = η (B, ρ ), which is universally X = x is measured by direct observation of another quantity
valid for any observables A, B, any system state ρ , and any Y = y. Then, this measurement is precise iff X = Y holds with
measuring process M. From Eq. (59), it is concluded that if probability 1, or equivalently the JPD µ X,Y (dx, dy) of X and
the error and the disturbance are statistically independent from Y concentrates on the diagonal set, i.e.,
system state, then the Heisenberg type EDR
µ X,Y ({(x, y) ∈ R2 | x 6= y}) = 0. (65)
1
ε (A)η (B) ≥ |h[A, B]i| (60)
2 As easily seen from Eq. (33), this condition is equivalent to
the condition εG (X,Y ) = 0.
holds, extending the previous results36–39 . The additional cor-
Generalizing the classical case, we say that a measuring
relation term in Eq. (59) allows the error-disturbance product
process M makes a strongly precise measurement of A in ρ iff
ε (A)η (B) to violate the Heisenberg type EDR (60). In gen-
A(0) = M(∆t) holds with probability 1 in the sense that A(0)
eral, the relation
and M(∆t) commute in ρ ⊗ ρ0 and that the JPD µ A(0),M(∆t)
1 concentrates on the diagonal set, i.e.,
ε (A)η (B) + ε (A)σ (B) + σ (A)η (B) ≥ |h[A, B]i| , (61)
2
holds for any observables A, B, any system state ρ , and any µ A(0),M(∆t) ({(x, y) ∈ R2 | x 6= y}) = 0. (66)
measuring process M11–13,35,40,41 .
On the other hand, we have introduced another operational
The new relation (61) leads to the following new constraints A(0),M(∆t)
for precise measurements and non-disturbing measurement: requirement. The weak joint distribution µW of A(0)
then and M(∆t) in a state ρ is defined by
1 A(0),M(∆t)
σ (A)η (B) ≥ |h[A, B]i| , if ε (A) = 0, (62) µW (dx, dy) = Tr[E A(0) (dx)E M(∆t) (dy)ρ ⊗ ρ0 ]. (67)
2
1 The weak joint distribution is not necessarily positive but
ε (A)σ (B) ≥ |h[A, B]i| , if η (B) = 0. (63)
2 operationally accessible by weak measurement and post-
Note that if h[A, B]i =
6 0, the Heisenberg type EDR (60) leads selection53 . We say that the measuring process M makes a
to the divergence of ε (A) or η (B) in those cases. The new weakly precise measurement of A in ρ iff the weak joint dis-
A(0),M(∆t)
error bound Eq. (63) was used to derive a conservation- tribution µW in state ρ concentrates on the diagonal
law-induced limits for measurements12,42–44 quantitatively set, i.e.,
generalizing the Wigner-Araki-Yanase theorem45–48 and was
A(0),M(∆t)
used to derive a fundamental accuracy limit for quantum µW ({(x, y) ∈ R2 | x 6= y}) = 0. (68)
computing12.
This condition does not require that A(0) and M(∆t) com-
mute, while it only requires that the weak joint distribution
X. QUANTUM ROOT MEAN SQUARE ERRORS concentrates on the event A(0) = M(∆t). A similar condi-
tion has been used to observe momentum transfer in a double-
We say that the measuring process M is probability repro- slit ‘which-way’ experiment54,55 . We naturally consider that
ducible for the observable A in the state ρ iff strongly preciseness is a sufficient condition for precise mea-
surements and weak preciseness is a necessary condition. In
Tr[E M(∆t) (∆)ρ ⊗ ρ0] = Tr[E A (∆)ρ ] (64) the previous investigations33,34 , it was mathematically proved
 8

that both conditions are equivalent. Thus, either condition is directly observed are perfectly correlated in any input state,
concluded to be a necessary and sufficient condition charac- not only reproducing the probability distribution in any state.
terizing the unique class of precise measurements. As above, We say that the measuring process M does not disturb an
we say that the measuring process M precisely measures A in observable B in a state ρ iff observables B(0) and B(∆t) com-
ρ iff it makes a strongly or weakly precise measurement of A mute in the state ρ ⊗ ρ0 and the JPD µ B(0),B(∆t) of B(0) and
in ρ . B(∆t) concentrates on the diagonal set. The non-disturbing
To characterize the class of precise measurements in terms measuring processes defined above can be characterized anal-
of the rms error-freeness condition, ε (A, ρ ) = 0, and the prob- ogously.
ability reproducibility condition, we introduce the follow- From the above results, a non-zero lower bound for ε (A) or
ing notions. The cyclic subspace C (A, ρ ) generated by A η (B) indicates impossibility of precise or non-disturbing mea-
and ρ is defined as the closed subspace of H generated by surement. In particular, if σ (A), σ (B) < ∞ and h[A, B]i 6= 0,
{E A (∆)φ | ∆ ∈ B(R), φ ∈ ran(ρ )}, where ran(ρ ) denotes the then any measuring process cannot precisely measure A with-
range of ρ . Then, the following theorem holds33,34 . out disturbing B.
The above characterizations of precise and non-disturbing
Theorem 2. Let M = (K , ρ0 ,U, M) be a measuring process measurements suggests the following definitions of the locally
for the system S described by a Hilbert space H . Let A be uniform rms error ε (A, ρ ) and the locally uniform rms distur-
an observable of S and ρ a state of S. Then, the following bance η (B, ρ )56 :
conditions are equivalent. ε (A, ρ ) = sup ε (A, φ ), (70)
(i) M precisely measures A in ρ . φ ∈C (A,ρ )
(ii) ε (A, φ ) = 0 in all φ ∈ C (A, ρ ). η (B, ρ ) = sup η (B, φ ). (71)
(iii) M is probability reproducible for A in all φ ∈ C (A, ρ ). φ ∈C (B,ρ )

Then, we have ε (A, ρ ) = 0 if and only if the measurement

In the case where A(0) and M(∆t) commute, precise mea- precisely measures A in ρ , and that η (B, ρ ) = 0 if and only if
surements are characterized by the rms error-freeness condi- the measurement does not disturb B in ρ . For those quantities,
tion, since in this case we have εG (A(0), M(∆t)) = ε (A, ρ ). the Heisenberg type EDR
However, the probability reproducible condition does not
characterize the precise measurements even in this case. To h̄
see this suppose that A(0) and M(∆t) are identically dis- ε (x̂)η ( p̂x ) ≥ (72)
2
tributed and independent34 (p. 763). Then, we have
is still violated by a linear position measurement56, and the
relation
ZZ
εG (A(0), M(∆t)) = (y − x)2 µ A(0) (dx)µ M(∆t) (dy)
R2 1
ε (A)η (B) + ε (A)σ (B) + σ (A)η (B) ≥ |h[A, B]i| (73)
= σ (A(0))2 + σ (M(∆t))2 + (hA(0)i − hM(∆t)i)2 . 2
holds universally56, where ε (A) = ε (A, ρ ) and η (B) =
Since σ (A(0)) = σ (M(∆t)) and hA(0)i = hM(∆t)i, we have η (B, ρ ).
√ Thus, the locally uniform rms error ε (A, ρ ) completely
εG (A(0), M(∆t)) = 2σ (A). (69) characterizes precise measurements of A in ρ and the lo-
cally uniform rms disturbance η (B, ρ ) completely character-
Thus, M is not a precise measurement for the input state ρ izes measurements non-disturbing B in ρ , while they satisfy
with σ (A) 6= 0. In the case where A(0) and M(∆t) do not the EDR of the same form as the rms error and disturbance.
commute, the rms error-freeness condition well characterizes Further investigations on quantum generalizations of the clas-
precise measurements to a similar extent to the probability re- sical notion of root-mean-square error and EDRs formulated
producibility condition. In particular, the class of measuring with those quantities will be reported elsewhere.
processes precisely measuring A in all ρ is characterized by
the following equivalent conditions33,34 : (i) ε (A, ψ ) = 0 for
all ψ ∈ H ; (ii) probability reproducible for A in all ψ ∈ H ; ACKNOWLEDGMENTS
(iii) Π = E A . The above result ensures our long-standing be-
lief that a measurement with POVM Π satisfying Π = E A is This work was supported in part by JSPS KAKENHI,
considered to be precise in any state in the sense that the mea- No. 26247016 and No. 15K13456, and the John Templeton
sured observable A(0) and the meter observable M(∆t) to be Foundation, ID #35771.

∗ ozawa@is.nagoya-u.ac.jp mechanics. In Wheeler, J. A. & Zurek, W. H. (eds.) Quantum

1 Heisenberg, W. The physical content of quantum kinematics and Theory and Measurement, 62–84 (Princeton UP, Princeton, NJ,
 9

1983). [Originally published: Z. Phys., 43, 172-98 (1927)]. 25 Braginsky, V. B., Vorontsov, Y. I. & Thorne, K. S. Quantum non-
2 Kennard, E. H. Zur Quantenmechanik einfacher Bewe- demolition measurements. Science 209, 547–557 (1980).
gungstypen. Z. Phys. 44, 326–352 (1927). 26 Caves, C. M., Thorne, K. S., Drever, R. W. P., Sandberg, V. D. &
3 von Neumann, J. Mathematical Foundations of Quantum Zimmermann, M. On the measurement of a weak classical force
Mechanics (Princeton UP, Princeton, NJ, 1955). [Originally coupled to a quantum mechanical oscillator, I, Issues of principle.
puglished: Mathematische Grundlagen der Quantenmechanik Rev. Mod. Phys. 52, 341–392 (1980).
(Springer, Berlin, 1932)]. 27 Caves, C. M. Defense of the standard quantum limit for free-mass
4 Bohm, D. Quantum Theory (Prentice-Hall, New York, 1951). position. Phys. Rev. Lett. 54, 2465–2468 (1985).
5 Messiah, A. Mécanique Quantique, vol. I (Dunod, Paris, 1959). 28 Yuen, H. P. Contractive states and the standard quantum limit
[Quantum Mechanics, Vol. I (North-Holland, Amsterdam, 1959)]. for monitoring free-mass positions. Phys. Rev. Lett. 51, 719–722
6 Schiff, L. I. Quantum Mechanics (MacGraw-Hill, New York, (1983).
1968). 29 Ozawa, M. Measurement breaking the standard quantum limit for
7 Ballentine, L. E. The statistical interpretation of quantum me- free-mass position. Phys. Rev. Lett. 60, 385–388 (1988).
chanics. Rev. Mod. Phys. 42, 358–381 (1970). 30 Ozawa, M. Realization of measurement and the standard quan-
8 Schrödinger, E. Die gegenwärtige Situation in der Quanten- tum limit. In Tombesi, P. & Pike, E. R. (eds.) Squeezed
mechanik. Naturwissenshaften 23, 807–812, 823–828, 844–849 and Nonclassical Light, 263–286 (Plenum, New York, 1989).
(1935). [English translation by J. D. Trimmer, Proc. Am. Philos. ArXiv:1505.01083 [quant-ph].
Soc. 124, 323–338 (1980)]. 31 Ozawa, M. Position measuring interactions and the Heisenberg
9 Braginsky, V. B. & Khalili, F. Y. Quantum Measurement (Cam- uncertainty principle. Phys. Lett. A 299, 1–7 (2002).
bridge UP, Cambridge, 1992). 32 Gudder, S. Joint distributions of observables. J. Math. Mech. 18,
10 Giovannetti, V., Lloyd, S. & Maccone, L. Quantum-enhanced 325–335 (1968).
measurements: Beating the standard quantum limit. Science 306, 33 Ozawa, M. Perfect correlations between noncommuting observ-
1330–1336 (2004). ables. Phys. Lett. A 335, 11–19 (2005).
11 Ozawa, M. Universally valid reformulation of the Heisenberg 34 Ozawa, M. Quantum perfect correlations. Ann. Phys. (N.Y.) 321,
uncertainty principle on noise and disturbance in measurement. 744–769 (2006).
Phys. Rev. A 67, 042105 (2003). 35 Ozawa, M. Physical content of Heisenberg’s uncertainty relation:
12 Ozawa, M. Uncertainty principle for quantum instruments and limitation and reformulation. Phys. Lett. A 318, 21–29 (2003).
computing. Int. J. Quant. Inf. 1, 569–588 (2003). 36 Arthurs, E. & Goodman, M. S. Quantum correlations: A general-
13 Ozawa, M. Uncertainty relations for noise and disturbance in ized Heisenberg uncertainty relation. Phys. Rev. Lett. 60, 2447–
generalized quantum measurements. Ann. Phys. (N.Y.) 311, 350– 2449 (1988).
416 (2004). 37 Raymer, M. G. Uncertainty principle for joint measurement of
14 Halmos, P. R. Introduction to Hilbert Space and the Theory of noncommuting variables. Am. J. Phys. 62, 986–993 (1994).
Spectral Multiplicity (Chelsea, New York, 1951). 38 Ozawa, M. Quantum limits of measurements and uncertainty
15 Imoto, N., Ueda, M. & Ogawa, T. Microscopic theory of the con- principle. In Bendjaballah, C., Hirota, O. & Reynaud, S. (eds.)
tinuous measurement of photon number. Phys. Rev. A 41, 4127– Quantum Aspects of Optical Communications, 3–17 (Springer,
4130 (1990). Berlin, 1991). ArXiv:1505.05083 [quant-ph].
16 Ozawa, M. Quantum measuring processes of continuous observ- 39 Ishikawa, S. Uncertainty relations in simultaneous measurements
ables. J. Math. Phys. 25, 79–87 (1984). for arbitrary observables. Rep. Math. Phys. 29, 257–273 (1991).
17 Ozawa, M. Conditional probability and a posteriori states in quan- 40 Ozawa, M. Uncertainty relations for joint measurements of non-
tum mechanics. Publ. Res. Inst. Math. Sci., Kyoto Univ. 21, 279– commuting observables. Phys. Lett. A 320, 367–374 (2004).
295 (1985). 41 Ozawa, M. Universal uncertainty principle in measurement op-
18 Srinivas, M. D. Collapse postulate for observables with continu- erator formalism. J. Opt. B: Quantum Semiclass. Opt. 7, S672–
ous spectra. Commun. Math. Phys. 71, 131–158 (1980). S681 (2005).
19 Ozawa, M. Measuring processes and repeatability hypothesis. In 42 Ozawa, M. Conservation laws, uncertainty relations, and quan-
Watanabe, S. & Prohorov, Y. V. (eds.) Probability Theory and tum limits of measurements. Phys. Rev. Lett. 88, 050402 (2002).
Mathematical Statistics, Lecture Notes in Math. 1299, 412–421 43 Ozawa, M. Universal uncertainty principle and quantum state
(Springer, Berlin, 1988). control under conservation laws. AIP Conf. Proc. 734, 95–98
20 Davies, E. B. & Lewis, J. T. An operational approach to quantum (2004).
probability. Commun. Math. Phys. 17, 239–260 (1970). 44 Busch, P. & Loveridge, L. Position measurements obeying mo-
21 Helstrom, C. W. Quantum Detection and Estimation Theory mentum conservation. Phys. Rev. Lett. 106, 110406 (2011).
(Academic, New York, 1976). 45 Wigner, E. P. Die Messung quntenmechanischer Operatoren. Z.
22 Ozawa, M. Conditional expectation and repeated measurements Phys. 133, 101–108 (1952).
of continuous quantum observables. In Itô, K. & Prohorov, J. V. 46 Araki, H. & Yanase, M. M. Measurement of quantum mechanical
(eds.) Probability Theory and Mathematical Statistics, Lecture operators. Phys. Rev. 120, 622–626 (1960).
Notes in Math. 1021, 518–525 (Springer, Berlin, 1983). 47 Yanase, M. M. Optimal measuring apparatus. Phys. Rev. 123,
23 Reed, M. & Simon, B. Methods of Modern Mathematical Physics, 666–668 (1961).
I: Functional Analysis (Revised and Enlarged Edition) (Aca- 48 Ozawa, M. Does a conservation law limit position measurements?
demic, New York, 1980). Phys. Rev. Lett. 67, 1956–1959 (1991).
24 Gauss, C. F. Theory of the Combination of Observations Least 49 Busch, P., Heinonen, T. & Lahti, P. Noise and disturbance in
Subject to Errors: Part One, Part Two, Supplement (SIAM, quantum measurement. Phys. Lett. A 320, 261–270 (2004).
Philadelphia, USA, 1995). [Originally published: Theoria Com- 50 Dressel, J. & Nori, F. Certainty in heisenberg’s uncertainty prin-
binationis Observationum Erroribus Miinimis Obnoxiae, Pars ciple: Revisiting definitions for estimation errors and disturbance.
Prior, Pars Posterior, Supplementum (Societati Regiae Scien- Phys. Rev. A 89, 022106 (2014).
tiarum Exhibita, Feb. 15, 1821)]. 51 Korzekwa, K., Jennings, D. & Rudolph, T. Operational
 10

constraints on state-dependent formulations of quantum error- uncertainty relation in ‘which-way’ experiments: how to observe
disturbance trade-off relations. Phys. Rev. A 89, 052108 (2014). directly the momentum transfer using weak values. J. Opt. B:
52 Busch, P., Lahti, P. & Werner, R. F. Colloquium: Quantum root- Quantum Semiclass. Opt. 6, S506–S517 (2004).
mean-square error and measurement uncertainty relations. Rev. 55 Mir, R. et al. A double-slit ‘which-way’ experiment on the
Mod. Phys. 86, 1261–1281 (2014). complementarity-uncertainty debate. New J. Phys. 9, 287 (2007).
53 Lund, A. P. & Wiseman, H. M. Measuring measurement- 56 Ozawa, M. Noise and disturbance in quantum measurements and
disturbance relationships with weak values. New J. Phys. 12, operations. Proc. SPIE 6244, 62440Q (2006).
093011 (2010).
54 Garretson, J. L., Wiseman, H. M., Pope, D. T. & Pegg, D. T. The

View publication stats