Manifestation of pointer state correlations in complex weak values of quantum
observables
Som Kanjilal,1, ∗ Girish Muralidhara,2, † and Dipankar Home1, ‡
arXiv:1512.08632v3 [quant-ph] 17 Nov 2016
1
Center for Astroparticle Physics and Space Science (CAPSS), Bose Institute, Kolkata-700091, India
2
Indian Institute of Science Education and Research Pune, Pune-411 008 , India
In the weak measurement (WM) scenario involving weak interaction and postselection by projective measurement, the empirical significance of weak values is manifested in terms of shifts in the
measurement pointer’s mean position and mean momentum. In this context, a general quantitative treatment is presented in this paper by taking into account the hitherto unexplored effect of
correlations among the pointer degrees of freedom which pertain to an arbitrary multidimensional
preselected pointer state. This leads to an extension of the earlier results, showing that, for complex
weak values, the correlations among different pointer degrees of freedom can crucially affect the way
the imaginary parts of the weak values are related to the observed shifts of the mean pointer position
and momentum. The particular relevance of this analysis is discussed in the case of sequential weak
interactions followed by a projective measurement enabling postselection (called sequential WM)
which, in the special case, reduces to the usual WM scheme (involving a single weak interaction and
postseletion) modified by the effect of pointer state correlations.
PACS numbers: 03.65.Ta, 03.65.-w
I.
INTRODUCTION
Formulation of the seminal idea of weak measurement (abbreviated, WM) in quantum mechanics
by Aharanov, Albert and Vaidmann (AAV)[1] and
its subsequent clarifications as well as elaborations
[2–11] over the years have given rise to a plethora of
investigations, theoretical as well as experimental [for a
recent comprehensive review see, for example, J.Dressel
et al.[12]. This ranges from the use of WM in the
analyses of intriguing quantum effects such as Hardy’s
paradox [13], the three box paradox[14], the quantum
Cheshire Cat[15], to the application of WM in the
context of quantum entanglement [16], for verifying
the “error-disturbance uncertainty relations” [17], for
observing the evolution of a quantum system in the
semiclassical regime [18], for avoiding loopholes in
showing quantum violation of hybrid Bell-Leggett-Garg
inequalities [19], for experimentally verifying Bell’s
inequality in time [20], for shedding light on quantum
discord [21], for demonstrating quantum contextuality
[22] and for studying tunneling [23]arrival time [24], as
well as for revealing interesting effects in the physics of
telecommunication fibers [25]. Further, WM has been
studied using neutron interferometry [26] and has been
∗
†
‡
somkanji@jcbose.ac.in
girish.muralidhara@students.iiserpune.ac.in
dhome@jcbose.ac.in
invoked for high precision measurements concerning
quantum metrology [27] such as for identifying a tiny
spin Hall effect[28], for detecting very small transverse
beam deflections [29, 30], and tiny temporal delay
[31]. Interestingly, it has also been used for directly
measuring the quantum wave function [32] and for discerning signatures of the average quantum trajectories
for photons [33]. Against this backdrop, in order to
motivate our work, it will be useful to recapitulate the
essence of the standard WM scenario.
Let us consider the preparation of a given system in an
appropriate preselected state |ψi i, with the state being
subjected to a von Neumann type interaction described
by the Hamiltonian Ĥ = g(t)Â⊗ q̂ where  is the system
observable, q̂ is the measurement pointer observable and
g(t) is the coupling parameter given by the normalized
compact support around the time of measurement [5];
here the von Neumann coupling is assumed to be ‘weak’
in the following sense. Taking the initial pointer state
to be, say, a Gaussian wave function, the interaction
involved is said to be ‘weak’ if it results in the pointer
state to be a superposition of Gaussian wave functions
which are substantially overlapping. Subsequently, this
interaction is followed by an appropriate postselection
pertaining to the projective measurement of any system
observable other than  which is involved in the preceding weak interaction. Then the superposition of overlapping Gaussian wave functions, in effect, gives rise to
a single slightly shifted Gaussian wave function [5, 10].
2
The net effect is manifested in terms of the shifted probability distribution of the pointer variable corresponding
to the postselected system state |ψf i which is an eigenstate of one of the outcomes of the projective measurement in question. The key result shown by AAV is that
the weak interaction involving the system observable Â,
combined with postselection, results in the final shift of
the postselected pointer variable distribution that turns
out to be proportional to a quantity called the ‘weak
value’ of the observable Â, which is defined as
(A)w =
hψf |Â|ψi i
hψf |ψi i
(1)
Note that the weak value (A)w is, in general, a complex quantity. Its empirical significance was first pointed
out in the footnote 4 of the paper by Aharonov, Albert
and Vaidman [1] as well as in the paper by Aharonov
and Vaidman [7] using the Gaussian function for the
pointer state. later, the more general quantitative relations linking the shifts of the mean pointer position
and momentum with the real and imaginary parts of
the weak value were formulated by Jozsa [34]. To put
it precisely in the context of weak interaction involving
the Hamiltonian H = g(t)Â ⊗ q̂, the final and the initial expectation values of pointer position (hq̂if , hq̂ii )and
pointer momentum (hp̂if , hp̂ii ) are respectively mutually
related as follows
hq̂if = hq̂ii + 2λ(Im(A)w )var(q)
(2)
∂var(q)
(3)
∂t
R
where the variance var(q) = hq̂ 2 ii − hq̂i2i , g(t)dt = λ,
and m is the mass of the system in question. It is the
above results that constitute the specific starting point
of this work.
hp̂if = hp̂iin − λRe(A)w + mλ(Im(A)w )
Before proceeding further, an important basic point
to stress is that since a typical WM scheme uses at-least
two von Neumann interactions involving, in general,
two different pointer degrees of freedom (one for weak
interaction and the other for projective measurement
leading to the postselection), the scenario in general
involves the use of multidimensional pointer variable
distribution.
Here it is relevant to emphasize that, given any covariance matrix, it is possible to generate a multivariate
distribution embodying the correlations given by the
covariance matrix (Cholesky decomposition). Now,
note that while the effects of multivariate pointer state
distribution without correlations have already been discussed in the context of weak measurement [18, 35–38],
curiously, the possible effects of multidimensionality
embodying correlations among different pointer degrees of freedom have remained largely unexplored.
This holds apart from a couple of works probing the
effects of multidimensionality of pointer states in the
special case of two dimensional Hermite-Gaussian and
Laguerre-Gaussian optical modes as pointer states [39].
Against this backdrop, in this paper we seek to provide
a hitherto unexplored general framework for treating
the effects of correlations among different pointer
degrees of freedom, which has a special significance,
for example, in the context of continuous variable
entanglement as discussed in the final section of this
paper.
Note that, in the usual treatments, including in
Jozsa’s derivation of Eqs. (2) and (3)[34], the underpinning assumption is either that the multidimensional
pointer wave function is factorizable, or that the same
pointer degree of freedom which is weakly coupled to
the system degree of freedom is used for projective
measurement enabling postselection. A key aspect of
our treatment is that by taking into account a general
preselected multidimensional pointer state, we consider
the case where the pointer degree of freedom involved
in the final projective measurement is different from
that used in the preceding weak interaction or weak
interactions in the case of sequential WM. The key
feature arising from the latter aspect, as shown in
our paper for the case of sequential WM(Sec. II), as
well as for the usual WM scenario(Sec. III), is that if
the initial pointer state involves correlations between
the pointer degrees of freedom which are involved in
weak interaction(s) and postselection , the final expectation values of the pointer degrees of freedom will
contain contributions from these non-zero correlations,
depending upon whether the relevant non-vanishing
weak value(s) are complex or not. The essential result
demonstrated is the way the quantitative relations
between mean pointer position (momentum) and weak
values given by Eqs. (2) and (3) get significantly modified in the presence of correlations between different
pointer degrees of freedom.
Here we may observe that a noteworthy work using
the initial pointer wave function as a multidimensional
3
function is that by G. Mitchison [37]. In this work he
considered the joint expectation values of postselected
pointer degrees of freedom in terms of joint weak values
and generalized the treatment for weak measurement
involving arbitrary number of weak interactions.
However, the multidimensional initial pointer wave
function considered by Mitchison is factorizable. On
the other hand, in this paper, we consider essentially
the effect of nonfactorizability embodying correlations
in the initial pointer wave function including all the
pointer degrees of freedom. Another interesting line
of works using multidimensional pointer wave function
involves extracting joint weak value involving a product
of two single particle operators[35] and subsequently
extending it for joint weak values of the product of
N single particle operators [36]. Again, these studies
also essentially use factorizable initial pointer wave
function and hence the possible effect of correlations
in the multidimensional pointer wave function remains
unanalyzed.
Now, for outlining our scheme, considering sequential
WM, we use a weak interaction involving a system
variable (Â1 ) coupled with a pointer degree of freedom
(q̂1 or p̂1 ), followed by another weak interaction involving a system variable (Â2 ) coupled with a pointer
degree of freedom (q̂2 or p̂2 ). It is then found that the
postselection using a projective measurement involving
the pointer variable (q̂3 or p̂3 ) results in the individual
shifts along different axes (denoted by the lower index
i = 1, 2, 3) of the postselected three dimensional
pointer variable distribution. Each of these shifts
has contributions from both the weak values ((Â1 )w
and (Â2 )w ), arising from the two successive weak
interactions considered, apart from being dependent
on the correlations between the pointer degrees of
freedom. To compute these shifts, we evaluate the final
expectation values of the respective degrees of freedom
(hq̂1,2,3 if or hp̂1,2,3 if ) pertaining to the postselected
pointer state, relating each of these expectation values
to both the weak values ((A1 )w and (A2 )w ), along with
the correlation terms involving the pointer degrees of
freedom. For this demonstration, it suffices to consider
the strength of each weak interaction up to first
order. An important significance of the aforementioned
extension is that the correlation terms appear on the
final shifts of the pointer degrees of freedom only when
the imaginary parts of the respective weak values are
nonzero.Here we may stress that , as in Jozsa’s result,
we do not consider the effect of the time evolution of
the probe state that may occur before detection, which
has been taken into account by Lorenzo and Egues [40]
The archetypal investigation to date concerning the
sequential WM by Mitchison et al.[41],while considering
the strength of each weak interaction up to second
order, has essentially calculated the expectation value
of a product of pointer variables pertaining to the
post-selected pointer state without taking into account
the possible effects of correlations among the pointer
degrees of freedom. To be precise, Mitchison et al.[41]
obtained the joint expectation value of q̂1 q̂2 in terms
of the joint weak value (A1 A2 )w and the product of
the individual weak values (A1 )w and (A2 )w . The
fundamental difference between their work and ours as
discussed in this paper is, thus, easily evident.
In the following Section II we proceed to delineate the
mathematical details of our treatment in the case of sequential WM showing explicitly the way the empirical
signature of weak values in terms of the observed shifts
along different axes of the pointer variable distribution
involves correlations among the pointer degrees of freedom. Then, in Section III, we discuss how in the special
case of the vanishing strength of one of the two weak interactions, i.e., in the usually considered WM scenario,
our treatment reveals the effects of the pointer state
correlations on the individual shifts along different axes
pertaining to the multidimensional pointer variable distribution, thereby affecting their relation with the weak
value in question. In the concluding Section IV, we indicate some implications of this work as well as a few
directions for further studies, including possibilities of
future empirical probing.
II.
THE TREATMENT OF SEQUENTIAL WM
IN THE PRESENCE OF CORRELATIONS
BETWEEN POINTER DEGREES OF FREEDOM
Let the initial joint state of the system and the measurement pointer be given by
|Ψi = |ψi i ⊗ |φi i
(4)
where |ψi i and |φin i are respectively the system and the
pointer initial states that are taken to be three dimensional.
If φi (q1 , q2 , q3 ) 6= φi (q1 )φi (q2 )φi (q3 ), this would imply that the pointer state is not correlated between any
two of the 3 pointer degrees of freedom, i.e., the relevant covariance matrix [42] whose elements are given by
Σij = h(qˆi − hqˆi i)(qˆj − hqˆj i)i is diagonal, meaning that
4
only the variance terms of this matrix are non-vanishing.
In general, the covariance matrix is, however, not diag-
corr(ql , qm )i =
Z
φ∗i (p1 , p2 , p3 )q̂l q̂m φi (p1 , p2 , p3 )d~
p−
Z
φ∗i (p1 , p2 , p3 )q̂l φi (p1 , p2 , p3 )d~
p
where d~
p = dp1 dp2 dp3 . Before proceeding further, we
note the following properties of the correlation terms.
Consider the correlation function of momentum dis-
Z
onal, with the non vanishing off-diagonal terms, i.e., the
correlation terms with respect to the initial pointer state
being given by
Z
φ∗i (p1 , p2 , p3 )q̂m φi (p1 , p2 , p3 )d~
p
(5)
placed initial pointer wave function φi (p1 − l1 , p2 −
l2 , p3 − l3 ) (li is the amount of displacement of momentum pi ), given by
Z
− l1 , p2 − l2 , p3 − l3 )pl pm φi (p1 − l1 , p2 − l2 , p3 − l3 )d~
p − φ∗i (p1 − l1 , p2 − l2 , p3 − l3 )pl
Z
p
(6)
φi (p1 − l1 , p2 − l2 , p3 − l3 )d~
p φ∗i (p1 − l1 , p2 − l2 , p3 − l3 )pm φi (p1 − l1 , p2 − l2 , p3 − l3 )d~
Z
Z
= φ∗i (p1 , p2 , p3 )(pl + ll )(pm + lm )φi (p1 , p2 , p3 )d~
p − φ∗i (p1 , p2 , p3 )(pl + ll )φi (p1 , p2 , p3 )d~
p
Z
φ∗i (p1 , p2 , p3 )(pm + lm )φi (p1 , p2 , p3 )d~
p
corr(pl , pm )i,l1 ,l2 ,l3 =
φ∗i (p1
= corr(pl , pm )i
(7)
Similarly, it can be shown that
corr(ql , pm )i,l1 ,l2 ,l3 = corr(ql , pm )i
(8)
corr(ql , qm )i,l1 ,l2 ,l3 = corr(ql , qm )i
(9)
The above equations signify that the correlation functions are not dependent on the displacement of the momentum distribution of the pointer wave function. Note
that, corr(ql , ql ) = var(ql ) and corr(pl , pl ) = var(pl ).
Therefore, in the light of above relations it can be stated
that
var(ql )i,l1 ,l2 ,l3 = var(ql )i
(10)
var(pl )i,l1 ,l2 ,l3 = var(pl )i
(11)
In our treatment, for generality, we take the initial
pointer state to be as follows
Z
|φi (p1 , p2 , p3 )i = φi (p1 , p2 , p3 )|p1 i|p2 i|p3 id~
p (12)
If φi (p1 , p2 , p3 ) involves correlations, then the question
addressed in this paper is whether and, if so, how the
shifts of the relevant pointer degrees of freedom will capture the effect of these correlations.
The successive weak interactions in the setup considered here are taken to be of the von Neumann type
weak coupling between the system and the pointer observables, where the two Hamiltonians in question are
H1 = g1 (t)Â1 ⊗ qˆ1 and H2 = g2 (t)Â2 ⊗ qˆ2 . We assume that, apart from the von Neumann couplings used,
the system evolves freely in-between the weak interactions and before being subjected to the postselection. The postselection is performed by using projective measurement involving von Neumann type strong
coupling between Â3 and q̂3 , (the corresponding
HamilR
tonian H = gt Σk a3k Π̂3k ⊗ q̂3 with g3 (t)dt = 1, where
Π̂3k = |a3k i ha3k | is the projection operator
R corresponding to
R |a3k i with eigenvalue a3k ). Taking g2 (t)dt = λ2
and g1 (t)dt = λ1 , if we expand both the exponentials
occurring in the evolution operators of H1 and H2 up
to first order of λ1 and λ2 respectively, the joint state
l 6= m
5
of system and pointer after the strong von Neumann
interaction but before postselection can be written as
follows
Z
−iΣk a3k Π̂3k ⊗q̂3
d~
p(1 + iλ1 Â1 ⊗ q̂1 + iλ2 Â1 ⊗ q̂2 )
|ψi = e
φi (p1 , p2 , p3 ) |ψi i |p1 i |p2 i |p3 i
(13)
Using eiΣk a3k Π̂3k ⊗q̂3 = Σk eia3k q̂3 Π3k , the expression for the weak value given by Eq.(1) and
eia3k q̂3 φi (p1 , p2 , p3 ) = φi (p1 , p2 , p3 − a3k ), Eq. (13) can
be modified as follows
Z
|ψi = Σk ha3k | ψi i dp1 dp2 dp3 (1 + iλ1 (A1 )w ⊗ qˆ1 +
iλ2 (A2 ) ⊗ qˆ2 )φi (p1 , p2 , p3 − a3k ) |a3k i |p1 i |p2 i |p3 i
(14)
It is to be noted that, after successive weak interactions,
the pointer degrees of freedom q1 and q2 get entangled
with the system observables A1 and A2 . Presence
of nonzero correlation between q1,2 and q3 and/or
p3 would imply that before the strong von Neumann
interaction, the system observables A1 and A2 are
further correlated with q3 and/or p3 . Subsequently,
it can be seen from Eq. (14) that the strong von
Neumann interaction creates an entanglement between
|a3k i and |p3 i.
Postselecting the system state onto |a3l i which is
one of the eigenstates of the system variable Â3 one
can obtain the relevant pointer state as follows(taking
~ = 1 throughout our treatment)
|φf,3l i ≈ ha3l | ψi i
Z
(1 + iλ2 (A2 )w q̂2 + iλ1 (A1 )w q̂1 )i
φi (p1 , p2 , p3 − a3l ) |p1 i |p2 i |p3 i dp1 dp2 dp3 (15)
Let us consider the final expectation value of an arbitrary pointer variable M , corresponding to the postselected pointer state |φf,3l i given by
D
E
φf,3l M̂ φf,3l
hM̂ if,3l =
(16)
hφf,3l | φf,3l i
Then, writing the relevant weak values occurring in Eq.
(15) as follows
(A1 )w = a1 + ib1
(17)
(A2 )w = a2 + ib2
(18)
and, using Eq. (15), the numerator and the denominator of Eq. (16) are respectively given by
D
E
φf M̂ φf,3l ≈ | ha3l | ψi i |2 (hM̂ ii,3l + iλ1 a1 h[M̂ , qˆ1 ]ii,3l
+ iλ2 a2 h[M̂ , q2 ]ii,3l − λ1 b1 h{M̂ , q1 }ii,3l
− λ2 b2 h{M̂ , qˆ2 }ii,3l )
(19)
where,
hAii,3l =
Z
φ∗i (p1 , p2 , p3 −a3l )Âφi (p1 , p2 , p3 −a3l )dp1 dp2 dp3
(20)
Similarly
hφf,3l | φf,3l i ≈ | ha3l | ψi i |2 (1−2λ1 b1 hqˆ1 ii,3l −2λ2 b2 hqˆ2 ii,3l )
(21)
Next, using Eqs. (19) and (21), from Eq. (16) one
can obtain the value of hM̂ if,3l up to the first order in
λ1 , λ2 given by
hM̂ if,3l ≈ (hM̂ ii,3l + iλ1 a1 h[M̂ , qˆ1 ]ii,3l + iλ2 a2 h[M̂ , qˆ2 ]ii,3l
− λ1 b1 h{M̂ , qˆ1 }ii,3l − λ2 b2 h{M̂ , qˆ2 }ii,3l )
(1 + 2λ1 b1 hqˆ1 ii,3l + 2λ2 b2 hqˆ2 iin,3l )
(22)
Here [...] and {....} denote respectively the commutator
and the anti-commutator. For the specific choice of the
pointer observable M̂ = q̂1 in Eq. (22), the relevant
commutators vanish and one obtains
hqˆ1 if,3l = hqˆ1 ii,3l − 2λ1 Im(A1 )w var(q1 )i,3l
− 2λ2 Im(A2 )w corr(q1 , q2 )i,3l
(23)
Using relevant forms of Eqs. (9) and (10) we can rewrite
Eqn. (23) as follows
hqˆ1 if,3l = hqˆ1 ii − 2λ1 Im(A1 )w var(q1 )i
− 2λ2 Im(A2 )w corr(q1 , q2 )i
(24)
For M = q2 we will obtain
hq̂2 if,3l = hqˆ2 ii − 2λ2 Im(A2 )w var(q2 )i
−2λ1 Im(A1 )w corr(q1 , q2 )i
(25)
using Eqs.(17) and (18), and where the correlation term
corr(q1 , q2 )i is given by Eq. (5) for l, m = 1, 2.
The key consequence of the presence of the correlation term corr(q1 , q2 ) in the above equations can be expressed as follows. Eq. 24(25) shows that the shift of
the expectation value of the pointer degree of freedom
6
q1 (q2 ), apart from being dependent on the imaginary
part of the weak value of the system observable Â1 (Â2 )
which is coupled (a la von Neumann) with the pointer
degree of freedom q1 (q2 ), contains an additional contribution arising from the correlation term corr(q1 , q2 )i
that depends on the imaginary part of the weak value of
the system observable Â2 (Â1 ) which is, too, von Neumann coupled with the pointer degree of freedom q̂2 (q̂1 ).
For M̂ = q̂3 , one obtains from Eq. (22)
hq̂3 if,3l = hq̂3 ii,3l − 2λ1 Im(A1 )w corr(q1 , q3 )i,3l
− 2λ2 Im(A2 )w corr(q2 , q3 )i,3l
(26)
Using hq̂3 ii,3l = hq̂3 ii and Eq. (9) we can recast Eq.
(26) as follows
hq̂3 if,3l = hq̂3 ii − 2λ1 Im(A1 )w corr(q1 , q3 )i
− 2λ2 Im(A2 )w corr(q2 , q3 )i
(27)
Note that the shift in the expectation value of the
pointer degree of freedom q3 arising from sequential
weak interactions cum postselction essentially depends
upon the non-zero correlations between q1,2 and q3
present in the initial pointer wave function. Presence of
these correlations ensure that after two successive weak
interactions, not only q1,2 but also q3 gets entangled
with the system observables A1 and A2 . It is this
entanglement due to which the postselection of a
particular system state ensures that the expectation
value of q3 gets shifted.
Considering the sequential WM scenario, if the
pointer degree of freedom involved in the projective
measurement enabling postselection is the same as that
used in the preceding weak interactions, then Eq. (27)
reduces to
hq̂3 if = hq̂3 ii −2λ1 Im(A1 )w var(q3 )−2λ2 Im(A2 )w var(q3 )
(28)
which does not contain any effect of correlations
between the pointer degrees of freedom present in the
initial state.
Now, if the pointer degrees of freedom that are
involved in the weak interactions are different from
that used in postselection, then the shift of the final
expectation value of the pointer degree of freedom
(q3 ) involved in postselection will contain the effect of
correlations embodied in the initial preselected pointer
state. When these correlations vanish, such a shift will
also vanish. It is the above feature that is reflected in
Eq.(27) which essentially pertains to the case where
the pointer degree of freedom q3 used in postselection
is different from the pointer degrees of freedom (q1 , q2 )
used in the preceding weak interactions. In such cases,
the shift of the final expectation value of q3 crucially
depends on whether at least one of the correlations
corr(q1 , q3 ) or corr(q2 , q3 ) is non-vanishing.
Further, note that, even if all these correlations
vanish, the shifts of the final expectation values of the
pointer degrees of freedom q1 , q2 occurring in the weak
interactions remain non-vanishing, as can be seen from
Eqs. (24) and (25).
Similarly, for M̂ = p̂1,2 , from Eq. (22) we get
hp̂1,2 if,3l ≈ (hp̂1,2 ii,3l + iλ1 a1 h[p̂1,2 , q̂1 ]ii,3l + iλ2 a2 h[p̂1,2 , q̂2 ]ii,3l
− λ1 b1 h{p̂1,2 , q̂1 }ii,3l − λ2 b2 h{p̂1,2 , q̂2 }ii,3l )
(1 + 2λ1 b1 hq̂1 ii,3l + 2λ2 b2 hq̂2 ii,3l )
(29)
Using the canonical commutation relations [q̂1 , p̂2 ] =
0, [q̂1 , p̂1 ] = iI, [q̂2 , p̂1 ] = 0, [q̂2 , p̂2 ] = iI, Eqs. (8),(10)
and following the mathematical treatment of Jozsa [34]
we can obtain
hp̂1 if,3l = hp̂1 ii + λ1 Re(A1 )w + mλ1 Im(A1 )w
+ 2λ2 Im(A2 )w corr(p1 , q2 )i
hp̂2 if,3l = hp̂2 ii + λ2 Re(A2 )w + mλ2 Im(A2 )w
+ 2λ1 Im(A1 )w corr(q1 , p2 )i
∂var(q1 )i
∂t
(30)
∂var(q2 )i
∂t
(31)
where corr(ql , pm )i with i 6= j is given by
corr(ql , pm )i = hq̂l p̂m ii − hq̂l ii hp̂m ii
(32)
Here again, the effect of the pointer state correlation is
embodied in Eq. 30(31) in terms of the shift of the expectation value of the pointer degree of freedom p1 (p2 )
containing an additional contribution from the imaginary part of the weak value of the system observable
A2 (A1 ) which is coupled (a la von Neumann) with the
pointer degree of freedom q2 (q1 ).
For M̂ = p̂3 , from Eq. (22) one can similarly obtain
hp̂3 if,3l = hp̂3 ii,3l + 2λ1 Im(A1 )w corr(q1 , p3 )i,3l
+ 2λ2 Im(A2 )w corr(q2 , p3 )i,3l
(33)
Note that hp̂3 ii,3l = hp̂3 ii + a3l . Therefore, using Eq.
(8) we obtain
hp̂3 if,3l = a3l + hp̂3 ii + 2λ1 Im(A1 )w corr(q1 , p3 )i
+ 2λ2 Im(A2 )w corr(q2 , p3 )i
(34)
7
In the absence of weak interactions, the final expectation value of p3 will get shifted by the amount a3l due to
projective measurement. In the presence of successive
weak interactions involving q̂1 and q̂2 pointer degrees of
freedom, entanglement between p3 and the system observables A1 and A2 is created through non-zero values
of corr(q1 p3 ) and corr(q2 , p3 )- this results in the further
shift of the expectation value of p3 which can be seen
from Eq. (34).
It may be stressed here that the correlation terms
among the pointer position and momenta degrees of
freedom occurring in Eqs. (30) - (34) are non-vanishing
essentially because of the non-vanishing correlation
among the position degrees of freedom occurring in the
preselected pointer state. This can be seen by recalling that the Fourier transform of a multidimensional
probability density function, say, f (q1 , q2 , q3 ) can be
done through intermediate steps, each step comprising Fourier transform of qi to pi which ensures that
if corr(qi , qj ) 6= 0, then corr(qi , pj ) is also necessarily
non-vanishing (see Appendix A for the relevant mathematical details).
It is the results given by Eqs. (24) - (27) and (30)
- (34) which provide the extension of Jozsa’s results
(given by Eq. (2) and (3)) for the sequential WM in
the presence of pointer state correlations. Note that, if
all the correlation terms vanish in Eqs. (24) - (27) and
(30) - (34), i.e., if the three-dimensional pointer state is
separable in different pointer degrees of freedom, then
Eqs. (24) - (27) and (30) - (34) reduce to Jozsa’s results in the three dimensional case. Thus, an upshot of
our analysis is that if the strengths of both the weak
interactions are taken to be up to first order, Jozsa’s results (although originally derived for the WM scenario
using single weak interaction) remain valid for the case
of sequential WM, too, provided the pointer state correlations are ignored.
pointer state to be of the two-dimensional form
Z
|φi (p1 , p2 )i = φi (p1 , p2 ) |p1 i |p2 i d~
p
where we take corr(q1 , q2 )i 6= 0.
Here again the preselected system state is |ψi i, while
the von Neumann coupling for weak interaction is of the
usual form H = λ1 Â1 ⊗ q̂1 and the subsequent postselection is done through projective measurement using
another von Neumann coupling involving the pointer
degree of freedom q̂2 . Then, for the postselection of
the system state |a2l i (which corresponds to one of the
eigenstates of the system variable Â2 that is von Neumann coupled with q̂2 ). Going through the similar calculations leading up to Eq. (14) we can obtain the state
after the strong von Neumann interaction but before the
postselection as follows
Z
|ψi = Σk (1 + iλ1 (A1 )w q̂1 )φi (p1 , p2 − a2k )
|a2k i |p1 i |p2 i d~
p
In the case of the usually considered WM scenario involving a single weak interaction, one needs essentially
two pointer degrees of freedom, one for the von Neumann weak interaction and the other for implementing postselection via a projective measurement. Thus,
in this context, it suffices to consider the preselected
(36)
Postselecting |a2l i and writing Eq. (36) in terms of weak
values we obtain the postselected pointer state
Z
|φf i ≈ ha2l | ψi i (1+iλ1 (A1 )w q̂1 )φi (p1 , p2 −a2l ) |p1 i |p2 i d~
p
(37)
going through the similar calculations as before, one can
obtain the following results for the shifts of the expectation values of the pointer degrees of freedom
hq̂1 if = hq̂1 ii − 2λbvar(q1 )i
(38)
hq̂2 if = hq̂2 ii − 2λbcorr(q1 , q2 )i
(39)
hp̂1 if = hp̂1 ii + λa + mλb
III. THE TREATMENT OF WM WITH SINGLE
WEAK INTERACTION IN THE PRESENCE OF
POINTER STATE CORRELATIONS
(35)
∂var(q1 )i
∂t
hp̂2 if = a2l + hp̂2 ii + 2λbcorr(q1 , p2 )i
(40)
(41)
where a and b are respectively the real and the imaginary parts of the weak value of the observable  in
question.
Here also, the effect of entanglement between q1
and system observable A1 due to weak interaction is
8
manifested through non-zero correlation between the
degree of freedom used for postselection and that used
in weak interaction - this can be seen from Eqs. (39)
and (41)
Note that, Eqs. (38) and (40) do not contain any
effect of pointer state correlation and are the same as
Jozsa’s results given by (2) and 3).On the other hand,
it can be seen from Eqs. (39) and (41) that the effect
of pointer state correlation is manifested in the shift of
that pointer degree of freedom (say, position) which is
involved in the von Neumann coupling used for the projective measurement resulting in postselection, as well
as in the shift of its conjugate variable (momentum).
Then, Eqs. (39) and (41) constitute the key extension
of Jozsa’s results in the case of WM scenario involving
single weak interaction that arises essentially from
the pointer state correlations.Note that, in these Eqs.
(24) - (27), (30) - (34),(39) and (41) the correlation
terms essentially contain the imaginary part of the
weak value. This means that for the effect of the
pointer state correlation to be manifested in terms of
observable shifts of the pointer degrees of freedom, the
relevant weak value has to be necessarily complex.
Here it is relevant to mention that in the treatment
of weak measurement using orbital angular momentum(OAM) pointer states in terms of the LaguerreGaussian optical modes, it has been noted that for the
optical modes endowed with OAM, the pointer state distribution is not factorisable [43]. In this context, for extracting the joint weak values from the two-dimensional
spatial displacements, the relevant results for the shifts
of the mean pointer position degrees of freedom have
been obtained by Kobayashi et al. [44].
To put it more specifically, in the aforementioned
paper, the preselected pointer state is represented by
the two dimensional Laguerre Gauss modes with nonvanishing OAM l, and the weak interaction is taken to
be of the form
H = gδ(t − t0 )(Â ⊗ pˆx + B̂ ⊗ pˆy )
(42)
where g is the coupling parameter. The preselected
pointer state is given by
ψ(x, y, l) = N (x + isgn(l)y)|l| e−(x
2
+y 2 )/4σ 2
(43)
where N is the normalization constant and 2σ is the
beam waist. Kobayashi et. al. [44] considered the weak
measurement scheme using the Hamiltonian given by
Eq. (42) for the Laguerre Gauss mode with OAM l as
the initial pointer state. They obtained the following
relations
hx̂if − hx̂ii = g[Re(A)w + lIm(B)w ]
(44)
hŷif − hŷii = g[Re(B)w − lIm(A)w ]
(45)
Now, note that the compatibility between Eqs. (30),(31)
(obtained in our more general treatment) and Eqs. (44),
(45) requires that for the Laguerre Gauss mode with
OAM l characterized by ψ(x, y, l) given by Eq. (43)
corr(px , y) = corr(py , x) =
l
2
(46)
where l is the OAM corresponding to the Laguerre
Gauss mode. In order to check whether this condition
is indeed satisfied, we’ve calculated the values of
correlations for the OAM l = 1 Laguerre Gauss mode,
ψ(x, y, l = 1) given by Eq. 43. It is found that for the
two dimensional spatial wave function corresponding
to the OAM l = 1 Laguerre Gauss mode , using weak
interaction of the form given by Eq. (42), one obtains
corr(x, y) = 0 and corr(px , y) = corr(py , x) = 21 ,
thereby ensuring in this case the compatibility between
Eqs. (44), (45) and Eqs. (30) and (31). Similarly, it
can be checked that this compatibility holds for any
other value of l. Turek et. al.[45] later generalized the
treatment using all orders of interaction strength and
by taking specific system observables. If our treatment
is extended to include higher orders of interaction
strength. It would give rise to higher order cross
moments pertaining to pointer degrees of freedom.
This calls for further study.
Here we may stress that we have considered the
generic case of non-separable pointer state, but in the
treatment by Kobayashi et al. [44] they consider a
specific case of the non-separability of the pointer state
arising from non-vanishing orbital angular momentum
in the optical modes. A curious point to be noted is
that the observable shifts obtained in our treatment
of sequential WM by considering each of the two
weak interactions to be involving single von Neumann
coupling turn out to be the same as that obtained by
Kobayashi et al. [44] using a single weak interaction
with two von Neumann couplings given by Eq. (42).
Since both these treatments are based on considering
effects up to the first order of weak interaction, it should
be worth investigating in detail the implication of this
equivalence and how this is affected by considering
9
effects up to the second order of weak interactions. A
further implication of extending our treatment of the
sequential WM for the second order of weak interactions
would be to probe the way the results obtained in the
presence of pointer state correlations would reduce to
the results that were derived by Mitchison et al. [41]
for the second order of weak interactions but without
considering the pointer state correlations.
IV.
CONCLUDING REMARKS AND
OUTLOOK
To put it in a nutshell, in the case of sequential
WM as well as for the usually considered WM scenario
involving single weak interaction, our treatment shows
the way the effect of the pointer state correlation
is reflected in the empirical manifestation of weak
values in terms of observable shifts of the expectation
values of the pointer degrees of freedom. In both these
cases we have derived explicit forms of the pertinent
extensions (Eqs. (24) - (27), (30) - (34),(39) and (41))
of Jozsa’s original results. A key point to be noted
here is that in each of these equations, the effect of
correlations among the pointer degrees of freedom is
embodied in those individual terms which essentially
contain the imaginary part of the weak values involved.
Thus, for real weak values, the observable shifts of the
expectation values of the pointer degrees of freedom do
not contain any effect of the pointer state correlations.
Now, regarding the role of entanglement between the
pointer degrees of freedom in the pointer state |φin i,
we may stress that the existence of non-zero correlation
between the pointer degrees of freedom does not necessarily imply entanglement. It is only for the two mode
Gaussian wave function, the non-zero correlation between the pointer degrees of freedom necessarily implies
entanglement in the sense of satisfying the PPT criterion for continuous variable entanglement [46]. For the
three mode Gaussian wave function, non-zero correlations do not necessarily imply entanglement [47]. Thus,
our treatment of the standard WM scenario (Sec. III)
involving single weak interaction and the choice of |φi i
(Eq.(35)) to be two mode Gaussian involving non-zero
correlation has an interesting testable implication in the
sense that the shift of the final expectation values of the
pointer degrees of freedom will contain the effect of entanglement in terms of the relevant correlations (Eqs.
(39) and (41) of Sec. III). To elaborate on this, note
that, a general two-mode Gaussian state can be written
as follows
ψ(q1 , q2 ) = N exp[−(αq12 + βq22 + 2γq1 q2 )]
(47)
where N is the normalization constant and the above
state implies correlation given by corr(q1 , q2 ) ∝ γ. Using the criterion for any Gaussian state to be entangled
[48], the value of the determinant of the matrix C defined as follows for the two mode Gaussian determines
whether it is entangled
< q1 q2 > < q1 p2 >
C=
(48)
< p1 q2 > < p1 p2 >
When det(C) < 0, the state is entangled. Using this
criterion, it has been shown that for all non-zero values
of γ, the two mode Gaussian state given by the Eq.(47)
is entangled [46].
An important point to note here is that, using Eqs.
(39) and (41) giving the shifts of the expectation values
of the postselected pointer degrees of freedom (q2 and
p2 ), one can obtain the diagonal terms of the matrix C
from the experimentally determined shifts. Similarly,
in order to empirically determine the off diagonal
terms of the matrix C, one needs to interchange the
pointer degrees of freedom involved in weak interaction
and projective measurement used in the postselection.
Thus, by determining all the elements of the matrix
C, one can verify whether the initial pointer state in
question is entangled or not. Note that this procedure
essentially uses the scheme developed in Sec.III of our
paper.
Similarly, for the three-mode Gaussian as the initial
pointer state, one can obtain the elements of the
matrix C using the scheme presented in Sec.II for
the sequential WM. An extension of this procedure
seems possible for the initial pointer state taken to be
multi-mode Gaussian. A comprehensive investigation
of this possibility will be pursued in a sequel paper.
Another line of future study could be with respect
to the paper mentioned earlier by G. Mitchison [37]. In
that paper, joint expectation values of the postselected
pointer degrees of freedom are obtained in terms of
joint weak values for factorizable multidimensional
preselected pointer state, i.e., in the absence of the correlation between different pointer degrees of freedom.
One can thus investigate how their result would be
modified if we consider the initial preselected pointer
state to be correlated among different pointer degrees
10
of freedom.
It may also be instructive to probe how correlations
in the initial pointer state would affect the earlier analysis of ”weak trajectories” extracted from the pointers of
a series of weakly interacting devices using factorizable
N- dimensional Gaussian function for the initial pointer
state [18]. Finally, we offer a few remarks about a possible experimental test of the effect of pointer state correlations derived in the present paper. Recently, predictions made by Mitchison et al. [41] have been confirmed
by Piacentini et al. [49] in an interesting experiment by
considering sequential weak interactions undergone by
single photons using birefringence in optical crystals. In
this setup, the preselected two dimensional pointer state
is experimentally prepared to be a separable Gaussian,
by using single photon guided in a single-mode optical
fiber that is suitably collimated with a telescopic optical system. It would therefore be an interesting extension of this experimental setup to study how the shifts
[1] Y. Aharonov, D. Z. Albert, and L. Vaidman, Phys. Rev.
Lett. 60, 1351 (1988).
[2] A. J. Leggett, Phys. Rev. Lett. 62, 2325 (1989).
[3] A. Peres, Phys. Rev. Lett. 62, 2326 (1989).
[4] Y. Aharonov and L. Vaidman, Phys. Rev. Lett. 62, 2327
(1989).
[5] I. M. Duck, P. M. Stevenson, and E. C. G. Sudarshan,
Phys. Rev. D 40, 2112 (1989).
[6] Y. Aharonov and L. Vaidman, J. Phys. A 24, 2315
(1991).
[7] Y. Aharonov and L. Vaidman, Phys. Rev. A 41, 11
(1990).
[8] J. Tollaksen, J. Phys. A 40, 9033 (2007).
[9] N. Brunner and C. Simon, Phys. Rev. Lett. 105, 010405
(2010).
[10] A. K. Pan and A. Matzkin, Phys. Rev. A 85, 022122
(2012).
[11] N. D. Hari Dass, arXiv:1509.04869 [quant-ph].
[12] J. Dressel, M. Malik, F. M. Miatto, A. N. Jordan, and
W. Boyd, Robert, Rev. Mod. Phys. 86, 307 (2014).
[13] K. Yokota, T. Yamamoto, M. Koashi, and N. Imoto,
New J Phys 11, 033011 (2009).
[14] A. Matzkin and A. K. Pan, J. Phy. A: Math. Theor. 46,
315307 (2013).
[15] Y. Aharonov, S. Popescu, D. Rohrlich,
and
P. Skrzypczyk, New J Phys 15, 113015 (2013).
[16] G. S. Agarwal and P. K. Pathak, Phys. Rev. A 75,
032108 (2007).
[17] F. Kaneda, S.-Y. Baek, M. Ozawa, and K. Edamatsu,
Phys. Rev. Lett. 112, 020402 (2014).
of the postselected probability distribution functions of
the relevant pointer variables are modified by an appropriate tuning of the procedure for preparing the preselected pointer state that would entail pointer state correlations. Such an investigation would thus constitute
a critical empirical check of the results that have been
obtained and analyzed in this paper.
V.
ACKNOWLEDGEMENTS
We thank Lev Vaidman for helpful comments on the
initial version of this paper. Thanks are also due to the
anonymous referee for insightful suggestions. the visit
of GM enabling this collaboration was supported by the
DST INSPIRE Fellowship. The research of DH is supported by the Dept. of Science and Technology (DST),
Govt. of India and Center for Science, Kolkata. SK acknowledges the support of Fellowship of Bose Institute,
Kolkata.
[18] A. Matzkin, Phys. Rev. Lett. 109, 150407 (2012).
[19] J. Dressel and A. N. Korotkov, Phys. Rev. A 89, 012125
(2014).
[20] A. Palacios-Laloy, F. Mallet, F. Nguyen, P. Bertet,
D. Vion, D. Esteve, and A. Korotkov, Nat. Phys. 6,
442447 (2010).
[21] U. Singh and A. K. Pati, Ann Phys-NY 343, 141
(2014).
[22] M. F. Pusey, Phys. Rev. Lett. 113, 200401 (2014).
[23] A. M. Steinberg, Ann NY Acad Sci 755, 900 (1995).
[24] S. E. Ahnert and M. C. Payne, Phys. Rev. A 69, 042103
(2004).
[25] N. Brunner, A. Acı́n, D. Collins, N. Gisin, and
V. Scarani, Phys. Rev. Lett. 91, 180402 (2003).
[26] S. Sponar, T. Denkmayr, H. Geppert, H. Lemmel,
A. Matzkin, J. Tollaksen, and Y. Hasegawa, Phys. Rev.
A 92, 062121 (2015).
[27] L. Zhang, A. Datta, and I. A. Walmsley, Phys. Rev.
Lett. 114, 210801 (2015).
[28] O. Hosten and P. Kwiat, Science 319, 787 (2008).
[29] D. J. Starling, P. B. Dixon, A. N. Jordan, and J. C.
Howell, Phys. Rev. A 80, 041803 (2009).
[30] S. Goswami, M. Pal, A. Nandi, P. K. Panigrahi, and
N. Ghosh, Opt. Lett. 39, 6229 (2014).
[31] L. J. Salazar-Serrano, D. Janner, N. Brunner,
V. Pruneri, and J. P. Torres, Phys. Rev. A 89, 012126
(2014).
[32] J. S. Lundeen, S. Sutherland, A. Patel, C. Stewart, and
C. Bamber, Nature 474, 188191 (2011).
11
[33] S. Kocsis, B. Braverman, S. Ravets, M. J. Stevens, R. P.
Mirin, L. K. Shalm, and A. M. Steinberg, science 332,
1170 (2011).
[34] R. Jozsa, Phys. Rev. A 76, 044103 (2007).
[35] K. J. Resch and A. M. Steinberg, Phys. Rev. Lett. 92,
130402 (2004).
[36] K. J. Resch, Journal of Optics B: Quantum and Semiclassical Optics 6, 482 (2004).
[37] G. Mitchison, Phys. Rev. A 77, 052102 (2008).
[38] J. Lundeen and K. Resch, Physics Letters A 334, 337
(2005).
[39] B. de Lima Bernardo, S. Azevedo, and A. Rosas, Opt.
Commun. 331, 194 (2014).
[40] A. Di Lorenzo and J. C. Egues, Phys. Rev. A 77, 042108
(2008).
[41] G. Mitchison, R. Jozsa, and S. Popescu, Phys. Rev. A
76, 062105 (2007).
[42] N. V. Kampen, in Stochastic Processes in Physics and
Chemistry (Third Edition), North-Holland Personal Library, edited by N. V. Kampen (Elsevier, Amsterdam,
2007) third edition ed., pp. 1 – 29.
[43] G. Puentes, N. Hermosa, and J. P. Torres, Phys. Rev.
Lett. 109, 040401 (2012).
[44] H. Kobayashi, G. Puentes, and Y. Shikano, Phys. Rev.
A 86, 053805 (2012).
[45] Y. Turek, H. Kobayashi, T. Akutsu, C.-P. Sun, and
Y. Shikano, New Journal of Physics 17, 083029 (2015).
[46] R. W. Rendell and A. K. Rajagopal, Phys. Rev. A 72,
012330 (2005).
[47] G. Adesso, A. Serafini, and F. Illuminati, Phys. Rev.
A 73, 032345 (2006).
[48] R. Simon, Phys. Rev. Lett. 84, 2726 (2000).
[49] F. Piacentini, M. P. Levi, A. Avella, E. Cohen, R. Lussana, F. Villa, A. Tosi, F. Zappa, M. Gramegna,
G. Brida, I. P. Degiovanni,
and M. Genovese,
arXiv:1508.03220 [quant-ph].
Appendix A
In this section we illustrate how the correlation between pointer position variables gives rise to correlation
between position and momentum variables. Here we assume the initial pointer state distribution as a Gaussian
function represented by φ(q1 , q2 ), whence the position
distribution function f (q1 , q2 ) is given by
where hqˆ1 iin = hqˆ2 iin = 0 and |G| is the determinant of
the covariance matrix written as
G=
σ1
corr(q1 , q2 )
corr(q1 , q2 )
σ2
(A2)
Now, note that for the above mentioned Gaussian
f (q1 , q2 ), non-zero correlation between q1 and q2 implies
the non-separability of f (q1 , q2 ). Next, we show that the
non-separability of f (q1 , q2 ) implies non-separability of
f (p1 , q2 ) which in turn, entails non-zero correlation between p1 and q2 .
From Eq. (A1), taking the partial Fourier transformation from q1 to p1 , one can obtain as follows (by ignoring
the normalization constant)
f (p1 , q2 ) =
Z
e−ip1 q1 f (q1 , q2 )dq1
1
1
e−ip1 q1 exp[− σ12 q12 − σ22 q22 −
2
2
corr(q1 , q2 )q1 q2 ]dq1
=
Z
corr2 (q1 , q2 ) 2
p21
1
)q
−
+
= exp[− (σ22 −
2
2
σ12
2σ12
corr(q1 , q2 )
i
p1 q 2 ]
(A3)
σ12
Then, using Eq. (32) in the text, one can evaluate the
correlation between p̂1 and q̂2 for the function f (p1 , q2 )
given by Eq. (A3). Thus we obtain
corr(p1 , q2 ) =
Z
Z
p1 q2 f (p1 , q2 )dp1 dq2 −
Z
p1 f (p1 , q2 )dp1 dq2
q2 f (p1 , q2 )dp1 dq2
[corr(q1 , q2 )]
σ12
(A4)
f (q1 , q2 ) = |φ(q1 , q2 )|
s
|G|
1
1
=
exp[− σ12 q12 − σ22 q22 − corr(q1 , q2 )q1 q2 ]
(2π)2
2
2
From Eq. (A4) it is evident that if corr(q1 , q2 ) 6= 0,
corr(p1 , q2 ) is also necessarily non-vanishing.
(A1)
2
=i