Entanglement harvesting and divergences in quadratic Unruh-DeWitt detectors pairs

Allison Sachs,1, 2, Robert B. Mann,1, 2, 3, and Eduardo Martn-Martnez1, 4, 3,

1
Institute for Quantum Computing, University of Waterloo, Waterloo, ON, N2L 3G1, Canada
2
Dept. Physics and Astronomy, University of Waterloo, Waterloo, ON, N2L 3G1, Canada
3
Perimeter Institute for Theoretical Physics, Waterloo, ON, N2L 2Y5, Canada
4
Dept. Applied Math., University of Waterloo, Waterloo, ON, N2L 3G1, Canada
(Dated: April 26, 2017)
We analyze correlations between pairs of particle detectors quadratically coupled to a real scalar
field. We find that, while a single quadratically coupled detector presents no divergences, when one
considers pairs of detectors there emerge unanticipated persistent divergences (not regularizable via
smooth switching or smearing) in the entanglement they acquire from the field. We have character-
ized such divergences, discussed whether a suitable regularization can allow for fair comparison of
the entanglement harvesting ability of the quadratic and the linear couplings, and finally we have
found a UV-safe quantifier of harvested correlations. Our results are relevant to future studies of
the entanglement structure of the fermionic vacuum.

I. INTRODUCTION models such as electromagnetic coupling of atoms [23].

All these studies, however, analyzed entanglement har-
The vacuum state of a quantum field displays quantum vesting form bosonic fields.
correlations between observables defined in spacelike sep- It is known from fundamental studies that there are
arated regions [1, 2]. This vacuum entanglement has been differences between the entanglement structure of the
studied in quantum foundations, and has found a vari- vacuum of fermionic and bosonic fields [2441]. However,
ety of applications such as quantum energy teleportation a study of entanglement harvesting in fermionic setups
[3, 4], the black hole information loss problem [5] and has never been performed. A study of detector-based en-
firewalls, along with black hole complementarity [6, 7]. tanglement harvesting from a fermionic vacuum could re-
In a phenomenon called entanglement harvesting [8], solve ambiguities in defining entanglement measures be-
correlations in a quantum field (such as the electromag- tween disjoint regions of a fermionic field [32, 3541]. The
netic field) can be swapped to particle detectors (such as reasons why this has not been done can be traced back to
atoms or qubits). This is possible even when the particle fundamental difficulties associated with particle detector
detectors remain spacelike separated throughout the du- models for fermionic fields.
ration of their interaction with the field. This was first To analyze fermionic fields form the perspective of lo-
shown by Valentini [9] and later by Reznik et al. [10, 11]. calized particle detectors a detector model was intro-
Since then, entanglement harvesting has been shown duced that consisted of a cavity coupled to a fermionic
to be sensitive to the background geometry of space- field [42], much like Unruhs original detector was a cavity
time [1214], as well as the topology [15]. Additionally, coupled to a bosonic field [43]. Later, Takagi introduced
it has been shown that entanglement harvesting can be an UDW-like model for fermionic fields [44, 45], wherein
done sustainably and distilled into Bell pairs in a process a two level system coupled quadratically to a fermionic
called entanglement farming [16], a protocol that can be field,
adapted to create a quantum seismometer [17]. Entangle- Hf
. (1)
ment harvesting has also been studied in detail in timelike
separation contexts [18, 19] with implementation propos- However, this model contained persistent divergences
als in different testbeds from quantum key distribution that could not be regularized by an appropriate choice
based on homodyne detection [20] to strongly coupled of switching and smearing functions, as it was found in
superconducting qubits [21]. [46]. Thus, these investigations restricted themselves to
To model the entanglement-swapping interaction be- studying transition rates instead of transition probabil-
tween the detectors and field, the Unruh-DeWitt detec- ities. To track down the origin of these divergences in
tor model is used. This model utilizes a first-quantized Takagis fermionic detector, H umer et al. studied three
system (called a detector) linearly coupled to a scalar types of quadratically coupled UDW-like detectors [46].
bosonic field. While most of our knowledge of entangle- They concluded that these divergences are mainly due to
ment harvesting has been gleaned from this setup [12 the detector coupling quadratically to the field instead of
17, 22], there has been some exploration of more realistic resulting from the analytic structure (spinor vs. scalar)
or statistics (fermionic vs. bosonic) of the fields involved.
Persistent divergences in the single-detector vacuum ex-
citation probabilities (VEP) of quadratic UDW detectors
[email protected] were found to be renormalizable by the same techniques
[email protected] used in QED [46], i.e. normal ordering the interaction
[email protected] Hamiltonian.
 2

The analysis in [46], however, was limited to a single the entanglement harvesting capabilities for the linear
detector coupled to the field. It is therefore natural to ex- and quadratic models, first looking at entanglement har-
tend the studies of the quadratic coupling to fermions to vesting under suitable UV-regularization and then study-
settings of many detectors to explore, for example, entan- ing a divergence free quantifier of harvested correlations
glement harvesting from a fermionic vacuum. Neverthe- from the field: the mutual information. We present our
less, before moving to the fermionic coupling, it is impor- conclusions in section B.
tant to understand how entanglement harvesting works
for detectors quadratically coupled to a bosonic field, so
as to determine how much of any new phenomenology II. TIME EVOLUTION OF LINEAR AND
would be due to the fermionic nature of the field and QUADRATIC DETECTOR MODELS
how much of it is due to the quadratic nature of the cou-
pling. Let us introduce the two different detector models that
In this paper we extend the notion of entanglement we will analyze and compare in this paper. First, let
harvesting to a detector coupled quadratically to a us consider the well-known UDW detector model. This
bosonic field. While this does not answer open ques- model was first introduced as an operational way to study
tions about entanglement ambiguities in fermionic fields the particle content of a bosonic quantum field [43, 47].
as posed in [32], it does shed light on differences between It consists of a two-level quantum emitter (detector) cou-
linearly and quadratically coupled detectors, which is one pled linearly to a scalar quantum field along its worldline.
prominent difference between the bosonic and fermionic For a single inertial detector (labelled A) in flat space-
UDW models. A fermionic UDW model could resolve time, the UDW interaction Hamiltonian in the interac-
ambiguities in defining entanglement measures between tion picture is given by
disjoint regions of a fermionic field [32] and to this end it
Z
is important to understand how it is distinguished from (t) = a a (t ta ) t).
its bosonic counterpart. H a (t) dn xFa (x xa )(x,
We find that, despite the finite renormalized single de- (2)
tector vacuum excitation probability, the two-detector
density matrix remarkably contains persistent diver- Here, the monopole moment a eia t
a+ eia t +
a (t) =
gences at leading order in perturbation theory. These represents the two-level internal degree of freedom of the
divergences cannot be regularized by means of a smooth detector, which couples linearly to a real massless scalar
switching function or spatial profile, nor are they renor- t). 0 a (t) 1 is the switching function that
field (x,
malized by the techniques used in [46]. Instead, one must controls the time-dependence of the coupling of strength
introduce additional means of regularization (e.g., a soft A . The spatial profile F (x) carries information about
UV cutoff). It is interesting to note that these diver- the shape and size of the detector. The case of the point-
gences appear only in the non-local contributions to the like detector, commonly considered in the literature, is a
density matrix. As a result the entanglement harvested particular case of (2) where the smearing function is a
by detectors quadratically coupled to bosonic fields is delta distribution, Fa (x) = (x).
sensitive to the choice of UV cutoff and may require fur- Modifications of this model where the detector is cou-
ther regularization. pled quadratically to the field [48] allow one to compare
To tackle the problem of entanglement harvesting with on equal footing the response of bosonic and fermionic
a pair of quadratically coupled detectors, we will follow detectors (the latter necessarily being quadratic [45]).
two different avenues: a) We will analyze the nature and These models have been recently analyzed in detail in
strength of the divergences in 3+1D flat spacetime, ana- [46]. The interaction Hamiltonian for a quadratically
lyzing possible physically motivated regularization scales coupled UDW detector is given by
in entanglement harvesting and b) we will propose a mea- Z
sure of correlations between the detectors that are diver- a (t) dn x Fa (x xa ) : 2 (x, t) :,
H2 (t) = a a (t ta )
gence free and use it to further our knowledge of the
differences between the use of linear and quadratic cou- (3)
plings of particle detectors to study the entanglement
structure of quantum fields. where 2 (x, t) has been normal-ordered as prescribed by
This paper is organized as follows. In Section II we in- the analysis in [46].
troduce the linear and quadratic UDW detector models It is convenient at this point to define two different
and examine in detail their time evolution. In Section III types of UV divergences that particle detector models
we provide an overview of the single UDW detector, both may present. A regularizable divergence is one that can
quadratically and linearly coupled to a scalar bosonic be removed by use of a smooth switching function and/or
field. Section IV analyzes the two-detector entanglement- a smooth spatial profile (see, e.g., [4951]). A persis-
harvesting set up; we show in detail how persistent diver- tent divergence is one that remains even with smooth
gences emerge in the non-local terms of the quadratically switching and smearing functions (such as the diver-
coupled two-detector system. In section V we compare gences renormalized in [46]).
 3

Analysis of the detector response function [4951] and We can express T in a perturbative expansion as
a number of investigations of entanglement harvesting
and quantum communication with (linear) UDW detec- T = 0 + (1,0)
T
+ (0,1)
T
+ (1,1)
T
+ (2,0)
T
+ (0,2)
T
+ O(3 ),
tors [3, 811, 23, 5255] indicate that all leading or- (10)
der UV divergences present in the time evolution of lin-
early coupled UDW detectors are regularizable. While where (i,j) (i) 0 U
=U (j) .
T
this is not the case for quadratically coupled detectors For our purposes we take as the initial state
[42, 44, 45, 48], it has been shown that all persistent
divergences can also be renormalized for an individual 0 = |0ih0| ab,0 . (11)
quadratically coupled detector [46]. We will demonstrate
below that a straightforward application of the leading- with the field starting out in its lowest-energy (vacuum)
order prescription in [46] cannot renormalize persistent state.
leading-order divergences in more complex scenarios with After time evolution, the time evolved partial state of
several detectors. the detectors is obtained by tracing out the field degrees
of freedom:

ab,t = Tr (
T ) (12)
A. Time evolution of detector pairs

The first order term (1,0)

T
+ (0,1)
T
does not contribute
Previous studies of the quadratic UDW model focused at all to the detectors dynamics for field states whose
on the response of a single detector [42, 4446, 48]. Since one-point function is zero. This includes Fock states,
one of our goals is to analyze the model dependence of free thermal states and the vacuum state as a particular
vacuum entanglement harvesting, we will also consider case of these two categories. In fact, for the
the evolution of two particle detectors coupled to the field
vacuum
state it can be easily proved that Tr (i,j)
T
= 0 when
vacuum.
i + j is odd (see e.g., [52]). Thus, we can express the
Both the linear and quadratically coupled UDW
time-evolved density matrix of the subsystem consisting
Hamiltonians can be rewritten for the two-detector case
of the two detectors as
as
X Z ab,t = ab,0 + 2a a,t + 2b b,t + a b cor,t + O(4 ),

H = t),
(t) dn x F (x x )(x,
(t t ) (13)
{A,B}
(4) where we have separated the local contributions to time
X Z evolution (proportional to 2a and 2b at leading order)
2 =
H (t) dn x F (x x ) : 2 (x, t) :,
(t t ) from the non-local terms (responsible for the correlations
{A,B} the detectors acquire through the field) proportional to
(5) a b . Notice that, from (13), we can quickly recover the
case of the evolution of a single detector just by taking
where {A, B} is the label identifying detectors A and b = 0.
B. Note that the coupling strength in the quadratic case Let us now particularize for the case where both de-
does not have the same dimensions as in the linear case. tectors start out in the ground state:
If we let the initial state of the field-detector system be
0 , its time evolved state is T = U 0 U
, where the label ab,0 = |ga ihga | |gb ihgb | . (14)
T denotes the timescale where the switching function is
non-zero, and the time evolution operator U is given by It is convenient to pick the usual [52] 4 4 matrix repre-
the time-ordered exponential sentation for ab,t in the basis
Z 
|ga gb i = (1, 0, 0, 0) , |ea gb i = (0, 1, 0, 0) ,
U = T exp dt HI (t) . (6)
|ga eb i = (0, 0, 1, 0) , |ea eb i = (0, 0, 0, 1) . (15)
Consequently, we can express the time evolution oper- In this basis, ab,t takes the form
in terms of a Dyson expansion as
ator U

= 11 + U
(1) + U
(2) + O(3 ), 1LaaLbb 0 0 M
U (7)
0 Laa Lab 0
ab,t = +O(4 ), (16)
where 0 Lba Lbb 0
Z M 0 0 0
(1) = i
U I (t)
dt H (8)

where M and L depend on the nature of the cou-
Z Z t pling (e.g., linear versus quadratic, different switching
(2) =
U dt I (t)H
dt0 H I (t0 ). (9) and smearing functions, etc. See sections II A 1 and
II A 2).
 4

1. Linear coupling The Wightman function (21) for the linear coupling
case can be written as
Z 0 0
The matrix elements of (16) for the linear coupling ei(|k|(t t)k(x x))|k|
have been studied at length in the literature (See, for W (t, x, t0, x0 ) = dn k , (23)
2(2)n |k|
instance, [52], which sets the notation that we will follow
here) and are given by as it is easy to check, for example, through the usual
plane-wave expansion in Eq. (22). Particularizing to 3+1
Z Z t Z Z
0
dimensions, the two-point function becomes

M = a b dt dt n
d x dn x0
1
W (t, x, t0 , x0 ) = . (24)
0 0
M (t, x, t , x ) W (t, x, t , x )

0 0
(17) 4 2 (x x0 )2 (t t0 i)2

Z Z Z Z Note here that we see  takes the form of the usual pole

0 0 prescription for the Wightman function.
L = dt dt d x n
d x n
(18)


L (t, x) L (t0, x0 ) W (t, x, t0, x0 ), 2. Quadratic coupling

where, assuming and F are real, L and M (t, x, t0, x0 )

For the quadratic coupling case in (5), the elements of
are given by
the density matrix (16) take the following form
Z Z t Z Z
L (t, x) = (t t ) F (x x ) ei t (19) 2

M = a b dt dt d x dn x0
0 n
M (t, x, t0, x0 ) = La (t, x)Lb (t0, x0 )
+La (t0, x0 )Lb (t, x), (20) 2
M (t, x, t , x ) W (t, x, t0, x0 )
0 0
(25)
Z Z Z Z
2
and the Wightman function, W , is given by

L = dt dt0 dn x dn x0

2

0 0 t) (x
W (t, x, t , x ) = h0| (x, 0, t0 ) |0i . (21) L (t, x) L (t0, x0 ) W (t, x, t0, x0 ),

(26)
which is structurally the same as in the linear coupling
To find an explicit expression for the Wightman func- case. Indeed, M and L are also defined as in Eqs. (20)
tion, we will utilize a plane-wave mode expansion of the and (19), respectively. Thus, the difference between the
field operator with soft UV cutoff , usual UDW detector and the quadratically coupled UDW
2
Z n |k|/2 detector comes at the level of the functional W . For
t) = d ke 2
(x, p the quadratically coupled model, W is the vacuum ex-
2(2)n |k| pectation of the normal ordering of the square of the field
 
operator at two different points, as given by
k ei(|k|tkx) + a
a k ei(|k|tkx) . (22)
2
W (t, x, t0, x0 ) = h0| : 2 (x, t) : : 2 (x0, t0 ) : |0i . (27)
Here, a k (and a k ) are creation (and annihilation) oper- In Appendix A we show that the correlation functions
ators which satisfy the canonical commutation relations 2
W and W satisfy the following relation
[ k2 ] = (n) (k1 k2 ).
a k1 , a
2
Typically, the introduction of  could be associated W (t, x, t0, x0 ) = 2W (t, x, t0, x0 )2 , (28)
with a regularization procedure that leads to the usual 2
pole prescription, in which the limit  0 is well-defined which allows us to write W explicitly as
and eventually taken when evaluating observable quanti- Z Z
2 0 0 (2)2n
ties. However  can also be viewed as an ad hoc screening W (t, x, t , x ) = d k1 dn k2
n
(29)
of the detectors sensitivity to high frequency modes of |k1 ||k2 |
0 0
the field (soft UV cutoff). This would effectively model, ei((|k2 |+|k1 |)(t t)(k2 +k1 )(x x))(|k1 |+|k2 |) .
for example, a frequency dependent coupling strength
where a detector does not couple to frequencies much If we particularize to 3+1 dimensions, the correlation
2
larger than 1 . When giving this kind of interpretation function W is
to the -regularization one should be careful with possi- 2 2
ble non-localities introduced in the theory due to a finite W (t, x, t0, x0 ) =  2 . (30)
4 (x x ) (t t0 i)2
2 0 2
value of  [56]. Although this point will not be relevant
when the limit  0 is well defined, it must be taken into Note here that the correlator for the quadratic UDW
account when managing possibly UV divergent terms, es- detector has a higher power polynomial in x and t in its
pecially in the case of the quadratic coupling (3), as we denominator than does the usual correlator for the linear
will see below. UDW detector.
 5

III. SINGLE DETECTOR VACUUM any spacetime dimensionality, switching function and
EXCITATION PROBABILITY IN 3+1 spatial profile. For this analysis we will use smooth
DIMENSIONS, LAA switching and spatial smearing functions which are only
strongly supported in a finite region (T and respec-
tively). Smooth switching and smearing will ensure the
removal of all regularizable divergences of the kind stud-
Dimensionless ied in [49]. In particular, we choose Gaussian switching
variable Expression Physical meaning and Gaussian smearing,

T Energy gap 1 2 2
F (x x ) = n
e(xx ) / , (32)
( )
T UV cutoff

|xa xb |/T Detectors separation 2

/T 2
(t t ) = e(tt ) . (33)
t T switch-on times
As mentioned previously, in the literature UDW de-
/T Detectors size tectors are often considered to be point-like. Notice that
the Gaussian spatial profile can be particularized to the
q/T point-like case by taking the limit 0.

Table I. Collection of all the dimensionless quantities that are

used throughout this paper.
A. Linear coupling, LAA

In this section we will briefly review the vacuum ex- In this section we will calculate the vacuum excitation
citation probability (VEP) of a single detector for the probability for a single linearly-coupled, 3+1 dimensional
usual linear UDW detector model and the more recently UDW detector with Gaussian switching and smearing
studied VEP for a quadratically coupled model [46]. The functions. To do so, we first begin with Eq. (18), setting
vacuum excitation probability is the probability of exci- = = A. Then we substitute into Eq. (18) the 3+1
tation of a single UDW detector initialized in its ground dimensional Wightman function (24), the spatial profile
state in the vacuum. (32), and the switching function (33). Furthermore, to
It is well known that in 3+1 dimensions, point-like lin- further simplify the calculation, we apply the change of
early coupled UDW detectors with sharp switching func- coordinates
tions suffer regularizable UV divergences, which we re-
call can be eliminated by introducing a smooth switch- u = t1 + t2 , v = t1 t2 ,
ing or smearing function [4951]. However, quadratic
UDW models (such as the quadratic scalar model intro- p = x1 + x2 , q = x1 x2 . (34)
duced by Hinton in [48], the cavity detector coupled to a
fermionic field [42], or the fermionic UDW-like detector This results in
model introduced in [44, 45]) have VEPs that present per- Z Z
2 u2
2T p2
sistent divergences (not removable with a smooth switch- Laa = 5 6
du e 2
d3 p e 22
ing and/or smearing). These persistent divergences can, 64
however, be renormalized with techniques analogous to Z Z q2
2 v2
2 2T 2 iv
3 e
those in QED [46]. Once renormalized, the quadratically dv d q 2 . (35)
coupled single-detector UDW model is regularizable both (q (v i)2 )
in its scalar and fermionic variants.
To find the time evolved state of a single (quadratically The above integrals in u, p and the angular parts of q can
or linearly coupled) detector, we begin with the density be easily evaluated in closed form. To find the integral
matrix (16), then set b = 0. It is then simple to trace over v, we use the convolution theorem, as outlined in
out detector B to find the single-detector reduced state, the Appendix B. At this point, it is convenient to follow
[52] and rewrite these intergals in terms of dimensionless
 
1 Laa 0 parameters , , , , , and as outlined in Table I. The
a,T = Trb (
ab,T ) = +O(4a ), (31) outcome is
0 Laa
Z
in the basis |ga i = (1, 0) , |ea i = (0, 1) . The element 2 i 2 2 2
Laa = 3
d ei 22 + 2 i 2 (36)
of (31), Laa , given in Eq. (18), is the vacuum excitation 8 0
"    #
probability of detector A. + i + + i
2i(+)
The vacuum excitation probability expressed as erfc e erfc ,
Eq. (18) is quite general and can be particularized to 2 2
 6

where erfc is the complementary error function, defined The integrals over u, p, the angular part of q, and v can
in terms of the error function as follows: be evaluated in closed form (again, the last performed
Z z through a convolution product as shown in appendix B).
2 2
Once again, it is convenient to utilize the convention in
erf (z) = dtet (37)
0 [52] and recast these integrals in terms of the dimension-
erfc (z) = 1 erf (z) . (38) less parameters , , , , , and (outlined in Table I).
The result is
At this point, we can take the UV cutoff scale to in- 2 Z 2 2
finity ( 0, or in dimensioless quantities, 0, as per 2 2 e+ 2 ei 22 i 2
table I). The result is Laa = dq
32 4 3 T 3 0
" 
Z (2i 2 + 2 +)

i2
1 2
2 2e 2 (+i) + + i( 2 + 1)
lim Laa = d e 2 2
0 8 3 0
"    #


i + i + e2i(+) + + i 2 + 1
erfc e 2i
erfc , (39)  
2 2 + + i
erfc
2
which is not divergent. Fig. 1.(a) illustrates the be-  #
2

i + i +
haviour of Laa as 0 is reached. + i + i + + 1 erfi , (41)
2
We note that in previous literature closed expressionss
for (39) have been found for Gaussian switching and where erfi is the imaginary error function defined as
smearing functions in 3+1 dimensions [52]. The differ-
ence between calculations here and in [52] is that we have erfi(z) = i erf(iz). (42)
worked in the position representation. One can readily
check numerically that all elements L and M in this For this integrand, the limit of no cutoff, i.e,

paper are equivalent (after the limit 0 is taken) T = 0, at constant T , is well-defined:
to those in [52] for the linear detector. The motivation Z
behind complicating the calculation of the linear matrix 2 2 d 22 i 12 ( 12 +1)2
lim Laa = 4 3 2
e
elements by working in the position representation lies 0 32 T 0
"
in the difficulty of calculating the matrix elements of the 2 2 2
quadratic detector pairs, which is reduced by the method ie 2 2 + e 2 ie 2

described here. Moreover, there is an additional advan-

tage working in the position representation in the lin- 1 1
+ i 2 e 2 (+4i) + e 2 (+4i)
ear case: the method of computing leading order density 1 1
matrix elements in the position representation used here 2 2e 2 (+2i) + ie 2 (+4i)
  
yields results that have greater numerical stability for 2
 i
small detector gap in those terms for which we do not e 2 i 2 + 1 erf
2
have closed expressions neither in position nor in mo-
   #
mentum representation in [52], as we will show when we 1
(+4i) 2

+ i
e 2 + i + 1 erf .
present numerical results in section V. 2
(43)
2
B. Quadratic coupling, LAA This integral is convergent. How the convergent 0
limit is reached is shown numerically in Fig. 1.(d).
Similar to the linear model, in this section we calcu-
late the vacuum excitation probability for the quadratic
IV. THE TWO-DETECTOR MODEL
model, Laa . We begin with Eq. (18), setting = = A.
Then we substitute into (18) the quadratic two-point cor-
The vacuum excitation probability does not provide
relator in 3+1 dimensions (30), the spatial profile (32),
full information about the time evolution of a pair
and the switching function (33). Applying the same
of particle detectors coupled to the field. Indeed, to
change of coordinates as in the linear case, (34), yields
characterize more complicated effects, such as entangle-
Z Z ment harvesting [9, 10, 52], or quantum communication
2 2 u2
2T p2
Laa = 7 6
du e 2
d3 p e 22 [53, 54, 5660], the full time-evolved density matrix of
128
two detectors coupled to the field is necessary. The de-
Z Z q2 v2
tectors time-evolved density matrix (16) has extra terms
3 e 22 2T 2 iv
dv d q
2 . (40) in addition to the VEPs. Two different kinds of non-local
q 2 (v i)2 terms, Lab and M, now appear along with their complex
 7

conjugates. To fully characterize the two detector sys- which has a well-defined limit as 0,
tem, we need to find explicit expressions for these terms
and study the regularity of their behaviour.
As we will discuss in detail below (and as mentioned
in [52]), Lab is the term responsible for the leading order 2 1 2 Z

contribution to classical correlations (or, possibly, dis- i2 e 22 2 (a b )

lim Lab = d
cord) between the detectors, whereas M can be thought 0 8 || 0
 
of as responsible for the harvested entanglement from the || ia b 22 22
field to the detectors, as we will discuss in section V A. sinh e 2
2
We will analyze these two terms independently in the   
+ i(a b + )
next two subsections. e2(b +i) erfc
2
 
+ i( a )
e2a erfc b . (46)
A. LAB non-local term in 3+1 dimensions
2

In the following, we will find Lab for the linear and

quadratic models in 3+1 dimensions.a Not only is the integrand well-defined, but the integral
is convergent as well. We show numerically how the con-

vergent limit 0 of |Lab | is reached in Fig. 1.(b).

1. Linear coupling, Lab

For the linear UDW detector, the term Lab is given
by Eq. (18) when = A and = B. We also explicitly
write the Wightman function in 3+1 dimensions (24),
the spatial profile (32), and the switching function (33)
in equation (18). The same change of coordinates as in
the calculation of Laa , shown in Eq. (34), again simplifies
2
the calculation. This transformation yields 2. Quadratic coupling, Lab

t2
a t2
b x2
a b Z
x2
2 e T 2 T 2 2 2 ta u tb u u2
Lab = 5 6
du e+ T 2 + T 2 2T 2 2
In order to find Lab , given by Eq. (18), we set = A
64
Z Z Z and = B. We then substitute into Eq. (18) the
p2 pxa pxb
d pe 3 22 + 2 + 2
dv d3 q quadratic two-point correlator in 3+1 dimensions (30),
the spatial profile (32), and the switching function (33).
q
2
2 qxa qxb ta v tb v v 2
2 + 2 2 + T 2 T 2 2T 2 iv
As is now tradition, we will do the same change of co-
e
. (44) ordinates as in the calculation of the VEPs, shown in
q 2 (v i)2 Eq. (34). This transformation results in

The integrals over u, p, the angular part of q, and v

can be evaluated in closed form (with the same technol-
ogy shown in Appendix B). As before, we follow [52] and t2 2 x2 2
a +tb a +xb
rewrite these integrals in terms of the dimensionless pa- 2 2 e T 2 2
rameters as outlined in Table I. The result is Lab= 7 6
Z 128 Z
(ta +tb )u u2 p(xa +xb ) p2
2T
2 1 2 2 due T 2 2
d3 p e 2 22
2 e 22 2 (a b ) e+ia ib + 2
Lab =
8 ZZ q(xa xb ) v(ta tb ) q2
2 v2
Z    + T 2 2T 2 iv
 3 e
2 2
dv d q . (47)
d sinh 2
i e2(i+b +i) e2a (q 2 (v i)2 )
2
0
 
i + a b i +
+ e2(i+b +i) erfi
2
 
i a + b + i + The integrals over u, p, the angular part of q, and v can
+ e2a erfi
2 be evaluated in closed form (details in appendix B). Once
2 2 again, we follow [52] and rewrite these integrals in terms
eia b 22 i 2 , (45) of the dimensionless quantities outlined in Table I. The
 8

result is simple form under the same change of coordinates (34)

()2
22 12 (a b )2 +ia ib + 2 as in all previous calculations. In the case of M , this
2 2 e 2
change of coordinates also helps to de-nest the nested
Lab =
32 4 T 2 () time integrals. This yields
Z 2 2 Z Z

ei+a b 22 i 2 t2 t2 x2 x2
Ta2 Tb2 2a 2b
d M = e du dv
0 2
 " Z Z " qxa qxb ta v tb v
() 1 2 2 e 2 + 2 T 2 + T 2
sinh 2
2 2e 2 (+ia i(b +i+)) 3
d p d q 3
64 5 6 (q 2 (v i)2 )

#
+ i a + b + i + 2 + 1 qxa qxb ta v
2 e( 2 + 2 T 2 + T 2 )
tb v

  +
i a + b + i + 64 5 6 (q 2 (v i)2 )
erfi
2 p2 q2 2 2
 e 22 22 2
qxa
+ qx
2
b + ta u + tb u u v
T2 T2 2T 2 2T 2
+iu
. (50)
+ + ia ib + i 2 i
The integrals in u, p, and the angular parts of q can be
+ e2(ia +b +i) ( (a b i i + ) + 1) evaluated in closed form. This results in
  # t2 t2 x2 x2
i + a b i + a ta tb
2Tb2 T
2 2
+ita +itb 2a2 + xa xb b
i erfc i . (48) M = e 2T 2 + T2 2 2 2 2

2 Z Z q2 v2
2 T e 22 2T 2 q
2 2 dv dq
Taking the limit as 0 of Lab yields 2 (xa xb ) 0 (q 2 + (v i)2 )
   
2 Z   q(xa xb ) v(ta tb )
1
2 e 22 2 (a b )
2

d () sinh cosh . (51)
lim
2
Lab = sinh 2 T2
0 32 3 T 2 () 0 2 2
2 2 To obtain a closed form for the integral over v, we sim-
i+a b 2 2 2
e
plify M by choosing to switch on the detectors simul-

taneously within their co-moving frame, i.e. we make
( + i(a b ) i) the simplifying additional assumption ta tb = 0. Un-
 
der this assumption, M (in dimensionless parameters
+ i(a b )
erfc as shown in I) takes the form
2
r Z  
8 1 (+i(a b ))2 2
e 2 M ta =tb = dq sinh
16 2 0 2
"    
+ e2(ia +b ) ( + i(a b + ) + i) T iT (T iT )2
  2 erfi 2Ei
+ i(a b + ) 2T 2T 2
erfc . (49)    
2 (T iT )2 T2
+ log log
Fig. 1 (e) illustrates the behaviour as decreases of T2 (T iT )2

the result of numerical integration over . (T iT )2
4 log(T iT ) + 4 log(T ) e 2T 2
  
B. M non-local term in 3+1 dimensions
(T +iT )2 T + iT
+e 2T 2
2 erfi
2T
   
In the following, we will derive M and M , then
2 (iT + T )2 1
+ 2Ei + log
discuss the relevant differences between the two. We will 2T 2 (T + iT )2

see that M has only regularizable divergences, while #
2
M exhibits persistent UV divergences. + 4 log(T iT ) 2 log(T + iT )

2 2 T 2 4ia 2 T 2 + 2 T 2 +2 T 2


e 2 2 T 2 , (52)
1. Linear coupling, M
where is the principal value of the exponential integral
For the usual linear detector, M is given by Eq. (17). function defined as
We substitute into Eq. (17) the Wightman function in Z t
e
3+1 dimensions (24), the spatial profile (32), and the Ei(z) := P.V. dt. (53)
switching function (33). The integrals take a particualrly z t
 9

3.30
10.9 3.64
3.25

M  (103 a b )
|Lab  (103 a b )
10.7 3.63
3.20
Laa (103 2a )

3.62
10.5
3.15 3.61
10.3
3.60
3.10

 

10.1 (a) (b) 3.59 (c)

3.05
102 104 106 108 1010 1012 102 104 106 108 1010 1012 102 104 106 108 1010 1012

8.7 9.0 2.23

M  (105 a b )
8.3 8.8

|Lab  (106 a b )
Laa (105 a a )

2.13
8.1 8.6

7.8 8.4
2.03

 2 
7.5 8.2
2 
(d) (e) (f )
2

1.93
7.2 8.0
102 104 106 108 1010 1012 102 104 106 108 1010 1012 102 104 106 108 1010 1012

Figure 1. All plots illustrate the behavior of relevant quantities as decreases on a log scale. Plots on the top row are for the
usual (linearly coupled) UDW detector. Plots on the bottom row are for the quadratically coupled UDW detector. All plots
use parameters = 1, = 1, b a = 4, and = 4, where relevant. Note how all plots (a)-(e) indicate convergence, except
(f), which (in contrast to (c)) shows shows linear growth of M2 on a logarithmic scale of , and thus a logarithmic divergence
as 0.

M is well behaved in the UV limit. If we remove the The integrals over u, p, and the angular part of q can
cutoff taking the limit  0 (i.e., 0), we obtain be evaluated in closed form. We can furthermore employ
the simplifying assumption that the two detectors are
2 2
Z switched on simultaneously, which results in
2 e 2 +2ia 2 2 2 2
lim Mta =tb =
de 22 2
0 4 0 T 2 2 (xa xb )2
    2
2 T e 2 +2ita 22
i () Mta =tb =
i erfc sinh . (54) 4 4 (xa xb )
2 2  
Z Z q sinh q(xa x b)
2 q2 v2
2 2 2T 2
The integral is convergent, and how the limit is reached dq dv 2 e . (56)
as 0 is shown in Fig. 1.(c). 0 (q 2 (v i)2 )

To obtain a closed from for the integral over v, we op-

2
erate as in the linear case and simplify by choosing to
2. Quadratic coupling, M switch on the detectors simultaneously within their co-
moving frame, setting ta tb = 0. The resulting semi-
2 closed form we write as
For the quadratic detector, M is given by Eq. (17).
We substitute into Eq. (17) the quadratic two-point cor- 2 2 Z
relator in 3+1 dimensions (30), the spatial profile (32), 2
2 e 2 +2ia 22
Mta =tb = d G () (57)
and the switching function (33). The traditional change 64 4 T 3 0
2
of coordinates shown in Eq. (34) simplifies M . The
result of these substitutions is after carrying out the integral over v. The details of G ()
can be found in appendix C, concretely Eq. (C2).
t2 t2 x2 x2
Z Z
2 a b a b
M = e T 2 T 2 2 2 du dv

Z Z " qxa qxb ta v tb v
C. Divergences in the quadratic model
3 3 2 e 2 + 2 T 2 + T 2
d p d q 2 2
128 5 6 (q 2 (v i)2 ) Unlike the linear model, the non-local term M is not
qxa qxb ta v tb v
# free of UV divergences, despite the fact that the detector
2 e( 2 + 2 T 2 + T 2 )
+ 2
has a smooth switching and a Gaussian spatial smearing,
128 5 6 (q 2 (v i)2 ) and despite the renormalization process that removed the
p2 q2 qxa
+ qx b + ta u + tb u u v 2 2 single-detector divergences. Concretely, the integral in
e 22 22 2 2 T2 T2 2T 2 2T 2
+iu
. (57) is logarithmically divergent with the UV cutoff scale,
(55) as illustrated in Fig. 1.(f).
 10

2
To gain insight on the logarithmic divergence in M Negativity of a bipartite state is an entanglement
we examine the integrand G(), defined in the appendix monotone defined as the sum of the negative eigenvalues
equation (C2). of the partial transpose of [62]:
We begin by noticing that G() has the limit 0
X |i | i
( ) 2 +1 2
  N () = , (60)
4 2
lim G() = 2 e 22 sinh (58) i [A ]
0 2
"   #
2
2

i where A denotes the partial transpose of with respect
2e + i + 1 erfc
2 .
2 to subsystem A.
As seen, for instance, in [52], the negativity can be
Expanding in Laurent series and keeping the leading or- expressed in terms of the vacuum excitation probability,
der O( 1 ) results in the UV divergent term L , of each detector and the non-local term M in (57).
Concretely, it is given by
4i
lim GM2 . (59)
0 ta =tb 2 h i
N = max N (2) , 0 + O(2 ), (61)
This divergence is peculiar due to the fact that it shows
up only in the two-detector model, in spite of the fact
where
that the vacuum excitation probability for a quadratic
 q 
detector is finite [46] as discussed in section III B. Thus, (2) 1 2 2
while a single quadratically coupled detector does not N = Laa + Lbb (Laa Lbb ) + 4 |M| .
2
require additional UV regularization, a cutoff is required (62)
for certain quantities describing detector pairs, of which
the M term in (57) is an example. When both detectors are identical (i.e. they have
We would like to emphasize that these are persistent the same spatial profile, switching function, coupling
UV divergences, that is, they are present regardless of strength, and detector gap), Eq. (62) becomes
the use of smooth switching functions and spatial pro-
files (for example, in this case we have used Gaussian N (2) = |M| L (63)
functions for both). Moreover, these divergences appear
after renormalization of the zero-point energy and at the from which we can justify the usual argument that entan-
same order in perturbation theory at which the single glement emerges as a competition between the non-local
detector dynamics is regular. contribution M and the noise associated to the vacuum
excitation probability for each detector [11, 52].
Figure 3 shows the behavior of the negativity with the
V. HARVESTING CORRELATIONS spatial separation of the detectors, for the linear and
quadratic case and a range of detector cutoffs.
The analysis of the correlation terms Lab and M in Recall that the term M is UV divergent in the
both the linear and quadratic models is necessary to ex- quadratic model, therefore to compute a physically mean-
plore the entanglement structure of the field through par- ingful value for the negativity further regularization and
ticle detectors. eventual renormalization would be required. However for
In this section we will study two types of correlations a fixed UV-cutoff scale, it is possible to get an estimate
that the two detector models can harvest from the field of the quadratic model performance to harvest entangle-
vacuum: a) those measured by the mutual information, ment relative to the linear model by computing entan-
which quantifies both classical and quantum correlations glement harvesting for both models applying the same
[61], and b) the negativity, which is a faithful entangle- UV-cutoff scale. What is more, studying how negativity
ment measure for bipartite two-level systems [62]. These changes as we start increasing the cutoff scale will help us
two types of correlation harvesting were studied for the see how the UV divergence of M impacts entanglement
linear model in [52]. Here we will compare these results harvesting.
with the predictions for the quadratic model. As seen in Fig. 3, the magnitude of entanglement har-
vesting increases linearly with the logarithm of the cutoff,
which is not surprising since the two-detector quadratic
A. Entanglement Harvesting model suffers a logarithmic UV divergence. This implies
that there would always exist a value for the cutoff scale
We consider first the harvesting of entanglement from so that harvesting is possible at any distance, regardless
the vacuum and we quantify it with the negativity ac- of the detector gap. It would also imply that for large
quired between the two (initially uncorrelated) detectors enough cutoff frequencies we could always harvest more
through their interactions with the field while remaining entanglement with the quadratic model than for the lin-
spacelike separated. ear model.
 11

Linear Quadratic Inset Quadratic Quadratic Inset

8 8
2.58
Quadratic
6 8 1 6

6
T

T
4 4
T

4
2 2
2
a b c d
0 (b) 0 0 2.33
0 0 2 4 6 8 2.3 3 3.7 0 2 4 6 8 4.08 4.33
0 2d/T 4 6 8 d/T d/T d/T
d/T
Not Possible
1012 102 1012 104 1012 106 1012 108 1012 1010 1012
for 1012

Figure 2. Entanglement harvesting is possible (darker regions colored red or blue) for any detector separation d given a
sufficiently large detector gap . These plots show various detector cutoffs,  = /T , ranging from = 102 to = 1012 .
Note the cutoff does not make a significant difference for the linear model (no visible red regions), while there is a marked
increase in the harvesting region for smaller cutoffs (see inset (b) and (c)). Both plots use parameters = 1 and b a = 0.
The vertical white line shows the light cone. The dark lines in the insets are an extrapolation indicating the location of where
harvesting is no longer possible for a cutoff = 1029 (corresponding to setting the cutoff scale to the Planck frequency, as
described at the end of section V A).

One can therefore ask the following question: is there the results in Fig. 2 to the Planck scale. We show these
any finite value of the cutoff scale that we could take in results also in Fig 2, as the thin black lines. These plots
order to give some physical meaning to the finite cutoff illustrate the slow logarithmic nature of the divergences,
results? which makes the study of negativity still meaningful for
low energies with a quadratic detector and does not get
Unlike the linear modelwhich has been shown to cap-
significantly qualitatively modified even if the cutoff is
ture the fundamental features of the atom-light interac-
transplanckian.
tion [8, 16] the quadratic model does not have a direct
comparison with something as simple as the atom-light
interaction mechanism (maybe one could think of non-
linear optical media [63], but that is perhaps a stretch).
Recall, however, that we do not use the quadratic Unruh- B. Harvesting Mutual Information
DeWitt model to necessarily reproduce the physics of
a particular experimentally motivated setup. Our mo- A way around the problems associated to the UV-
tivation to explore this model is double: a) probe the divergent nature of M is to look at UV-safe quanti-
field with a different model to show model indepen- ties. Namely, it is possible to find quantifiers of correla-
dence/dependence of harvesting phenomena and b) ad- tions that are, by construction, UV-safe for the quadratic
vance towards the fermionic model (which is a quadratic model. One such figure of merit is the mutual informa-
model that does indeed have physical motivation) where tion.
the study of field entanglement remains still full of open The mutual information I(ab ) between two detectors
questions. quantifies the amount of uncertainty about one detector
The fact that this model cannot be connected with that is eliminated if some information about the state of
something as simple as an atom interacting with light, the other is revealed [61]. Thus, it constitutes a faithful
makes it difficult to motivate a choice of cutoff. How- measure of correlations (regardless if they are classical or
ever, if we were to take the result for a finite value of quantum).
the cutoff scale seriously, and thus if we were to choose In general, for composite quantum system consisting of
some physically motivated cutoff, we could compare the two subsystems A and B, the mutual information given
two models when such a cutoff is taken to be the Planck by
Frequency. In this scenario, the dimensionless cutoff pa-
rameter can be written as = kp1T , where kp is the I(ab ) = S(a ) + S(b ) S(ab ), (64)
Planck frequency. If we consider scales for the detector
gap to be commensurate with the energy of the first
transition of Hydrogen h 1015 s1 , then kp = 1029 . where = Tr ( ) is the partial trace of with re-
If we set T 1 spect to subsystem {A, B} and S is the von Neumann
h (which means that 1 represents
the case of a Hydrogen atom) the cutoff associated with entropy given by S() = Tr( log ).
the Planck time is then = 1029 . We can extrapolate For a density matrix of the form (16), the mutual in-
 12

Linear Quadratic
6 6

102ab

N2 102ab


5
5






4






4









3



3















2




2









1


1

(2)

(2)






(a)




(b)
N










0


0
























0 0 2 3 4 0 0 2 3 4

: 101 102 103 104 106 108 1010 1012

Figure 3. The magnitude of entanglement harvested is dependent on the cutoff chosen. For the same cutoff, either the linear
or quadratic models may harvest more entanglement, although the linear model quickly converges to a fixed value, while the
quadratic model grows logarithmically with decreasing cutoff. Thus, a cutoff can always be chosen sufficiently high such that
the quadratic model harvests more entanglement for a given set of parameters. Here, the top row shows plots of leading order
negativity with increasing spatial distance for the linear model, while the bottom row shows the same thing for the quadratic
model. The insets in each plot who where the negativity goes to zero. All plots use parameters = 1 and b a = 0, while
(a) and (c) show harvesting for detector gap = 0 and (b) and (d) show = 1.

8 8
log[I] both the linear and quadratic case where the other pa-
-2 rameters are the same as those used in Fig 2. First, we
Quadratic -4 observe something that was already present in previous
6 8 6
-6
-8
liteatura on linear detectors [52]: Unlike entanglement,
6 the mutual information harvesting can be performed at
T

-10
4 4
-12 any distance and detector gap, albeit less efficiently as
T

4 -14
2 2 -16
the distance (or the detector gap) increases, a feature
2 -18 that comes from the fact that the detectors are harvest-
0 (b) 0
-20 ing classical correlations as well as quantum correlations.
0 0 2 4 6 8 0 2 4 6 8 From Fig. 4 we observe that the linear detector can
0 2d/T 4 6 8 d/T
d/T
harvest more entanglement and for further distances than
Figure 4. Linear mutual information harvesting (left) is the quadratic detectors. This can in turn be used to as-
greater in magnitude that quadratic mutual entanglement sess the scale at which the soft cutoff model introduced
harvesting (right). Legend indicates Log base ten of the mu- in the study of negativity fails to capture the behaviour
tual information, I. Both plots use parameters = 1 and of UV-safe measures of correlations: as illustrated in Fig
b a = 0. The thick black line shows the light cone. 3, for cutoff scales that are of the order of & 106 ,
the linear model can harvest more entanglement than
the quadratic model in the parameter region studied.
formation is given by [52] This might suggest that comparison of negativity be-
tween the two models can be trusted only for cutoffs
I(ab ) = L+ log(L+ ) + L log(L )
above = 106 .
Laa log(Laa ) Lbb log(Lbb ) + O(4 ) (65)

where VI. CONCLUSION

 q 
1 2
L = Laa + Lbb (Laa Lbb ) + 4 |Lab | .
2
2 We have studied further the behaviour of particle de-
(66) tectors quadratically coupled to scalar fields introduced
Note how I(ab ) is not dependent on the divergent M in [44, 45] and renormalized in [46]. In particular we
term at leading order in perturbation theory. Hence, the have focused on the case of a pair of particle detectors
mutual information is finite without any further regular- harvesting entanglement from a scalar field, a case pre-
ization. Because of this, it provides a UV-cutoff inde- viously studied only for linear detectors [10, 11, 52]. Un-
pendent sense of the harvesting of correlations from the derstanding the harvesting of correlations from quadratic
vacuum, and can also be compared with previous results couplings is a necessary step in order to compare the
for linear detectors in [52], making it a relevant figure of entanglement that can be harvested from fermionic and
merit for the comparison in this article. bosonic fields, since the former only couple to particle
Fig. 4 shows the behavior of the Mutual information detectors quadratically [4446, 64]. Our motivation to
with spatial and temporal separation of the detectors for explore this model is twofold: a) probe the field with
 13

a different particle detector model to show model inde- harvested correlations that are UV-safe. In particular we
pendence/dependence of harvesting phenomena and b) showed that the harvested mutual information from the
provide a model that can be compared on equal footing field vacuum is UV safe. It therefore constitutes a better
for bosonic and fermionic fields (for which the coupling figure of merit to compare the harvesting of correlations
necessarily has to be quadratic). from the vacuum without need for further regularization.
Perhaps the most remarkable finding of our investiga- Understanding the particulars of entanglement har-
tion of harvesting with a quadratic detector is the ap- vesting with bosonic quadratic coupling is important in
pearance of a new logarithmic UV divergence at leading order to properly answer questions about fermionic fields
order in the two-detector setup. Notably, this divergence where the study of field entanglement remains full of
remains even when the Hamiltonian is normal-ordered, open questions [2441]. A comparison of bosonic and
and even when the switching functions and spatial pro- fermionic entanglement harvesting on equal footing re-
file are smooth functions. This is in stark contrast with quires knowledge of the model-dependence entanglement
the linear case where smooth smearing [50] or switching harvesting, specifically the difference between linear and
[49, 51] were enough to guarantee the UV regularity of quadratic coupling, as the latter is necessarily present in
the model. the fermionic case. The entanglement structure of the
We emphasize that a single detector, at the same or- fermionic vacuum remains an interesting open question,
der in perturbation theory, does not present this kind of one we are now prepared to address using the results we
divergence. Curiously, the UV divergence is only present have obtained.
in a particular kind of term, namely that responsible for
the entanglement of the two detectors. This divergence
is easily parametrized via a UV cutoff. ACKNOWLEDGEMENTS
Once this was established, analysis and comparison
with the linear model can be carried out. We proceeded The authors would like to thank Jose De Ram on for
in two different ways. First, using negativity to study helpful discussions. E.M-M. and R.B.M. acknowledge
entanglement harvesting. We discussed whether a finite the support of the Natural Sciences and Engineering Re-
value of the UV cutoff scale allows for fair comparison of search Council of Canada NSERC programme. E.M-M
the entanglement harvesting ability of the quadratic and also acknowledges the support of the Ontario Early Re-
the linear couplings. Following this, we found measures of search Award.

Appendix A: Calculation of the quadratic two-point function

Here we demonstrate that

 2
2
W (t, x, t0, x0 ) = 2 W (t, x, t0, x0 ) , (A1)

2
as asserted in section II A 1 and II A 2, where W and W are defined in Eqs. (27) and (21), respectively.
The relationship between an operator A and its normal ordered version is given by:
: A : = A h0| A |0i . (A2)
2
Using this identity, W can be rewritten as
2
W (t, x, t0, x0 ) = h0| : 2 (t, x) : : 2 (t0, x0 ) : |0i = h0| 2 (t, x)2 (t0, x0 ) |0i h0| 2 (t, x) |0i h0| 2 (t0, x0 ) |0i . (A3)
2
The first term of W (t, x, t0, x0 ) can be simplified. To do so, we will write the field operator as = + + , where
+ and are defined as
Z n |k|/2 Z n |k|/2
d ke d ke
+ (x, t) = p k ei(|k|tkx)
a (x, t) = p k ei(|k|tkx)
a (A4)
2(2) |k|
n 2(2)n |k|
which satisfy the commutation relation
 
(x , t ), + (x , t ) = C 11 (A5)
where C C is given by
Z
dn k e|k|/2 i(|k|(t t )k(x x ))
C = e (A6)
2(2)n |k|
 14

Using the notation (x

, t ), a scalar field vacuum four point function h0| 1 2 3 4 |0i can be rewritten as

h0|1 2 3 4 |0i = h0| ++ ++

1 2 3 4 |0i + h0| 1 2 3 4 |0i , (A7)

where, to remove all vanishing summands, we have used that

+
|0i = h0| = 0, (A8)

together with the fact that only summands with as many + as give a non-vanishing vacuum expectation.
Using (A5), we can write the first summand in Eq. (A7) as

h0| + +
1 2 3 4 |0i = C23 C14 + C13 C24 (A9)

and the second as

h0| + +
1 2 3 4 |0i = C12 C34 . (A10)

Thus (A7) can be written as

h0|1 2 3 4 |0i = C23 C14 + C13 C24 + C12 C34 . (A11)

From (A5), we see that we can rewrite the C coefficients as

 
C = h0| + +
, |0i = h0| |0i = h0| |0i . (A12)

This allows us to rewrite Eq. (A11) as

h0|1 2 3 4 |0i = h0| 1 2 |0i h0| 3 4 |0i + h0| 2 3 |0i h0| 1 4 |0i + h0| 1 3 |0i h0| 2 4 |0i . (A13)

To apply this identity to (A3), we set 1 = 2 = (t, x) and 3 = 4 = (t 0 , x0 ). Then, the first summand in (A3)
becomes
 2
h0|2 (t, x)2 (t0, x0 ) |0i = h0| 2 (t, x) |0i h0| 2 (t0, x0 ) |0i + 2 h0| (t,
x)(t 0, x0 ) |0i . (A14)

Which allows (A3) to be written as

 2
2 x)(t
0, x0 ) |0i ,
W (t, x, t0, x0 ) = 2 h0| (t, (A15)

which is (A1).

Appendix B: Convolution

In this appendix, we will find a closed form for the following integral
Z (ta tb ) v2
ev T 2 2T 2 iv

fm := dv 2 2 m
, (B1)
(q (v i) )

where T is a positive constant, ,  and v are non-negative constants, ta and tb are reaal constants, and m {1, 2}.
Introducing some basic notation that we will use in this appendix, F denotes the Fourier transform
Z
 
F a(x) () := dx a (x) eix . (B2)

We also introduce to denote the convolution product, defined as

Z
1
[a(x) b(x)] [x] := d a( )b (x ). (B3)
2
 15

The convolution theorem allows us to write fm in (B1) as

   
fm = F g(v) () F hm (v) () , (B4)

where g and hm are functions defined as

(ta tb ) v2
2T 1
g(v) := ev T2 2
, hm (v) := m . (B5)
(q 2 (v i)2 )

The Fourier transform of g(v) and hm (v) are

2
  (iT 2 ta +tb )
F g(v) () = 2T e 2T 2 (B6)

  ie(iq) h

F h1 (v) () = sgn() e2iq sgn(| Im(q)|) + sgn(| + Im(q)|) + e2iq 1
2q
i
2e2iq sgn( Im(q))(sgn( Im(q))) + 2sgn(Im(q) + )(sgn( + Im(q))) (B7)

 h
  e(iq)
F h2 (v) () = sgn() e2iq (q + i)sgn(| Im(q)|) + (q i)sgn(| + Im(q)|)
4q 3

i h
+ ie2iq + q + qe2iq i + 2 e2iq (q + i)sgn( Im(q))(sgn( Im(q)))
i
+ (q i)sgn(Im(q) + )(sgn( + Im(q))) . (B8)

Thus, using (B4) we find that fm takes the closed forms that we use to obtain equations (45) (48), i.e.,
"
! !
! #
q i T 2 itb +  + ta 2qta 2qtb 2iq q + i T 2 +  ta + tb
f1 = erfi + i e T 2 + T 2 + T 2 +2iq + erfi i
2T 2T
 
q 2 + qt2a qt2b iq2 iq+ ita2 itb2 +  2 +
2 2
e 2T T T T T T 2T (B9)
2q
"
(q+i)(q+i(2T 2 +)2ta +2tb )
1
f2 = 3 2 e 2T 2
iq 2 + qT 2 + iqta iqtb + q iT 2
4q T

! !!
(( )
2q i T 2 + ta +tb ) q i T 2

it b + 

+ ta
2 2
e T2 T + q q i T itb +  + ta erfi +i (B10)
2T

! 2
#



q + i T 2 +  ta + tb (q+i(T 2 +)ta +tb )
2 2
+ T + q q + i T +  ta + tb erfi 2 2qT e 2T 2
.
2T

Appendix C: Quadratic Non-local Term M

In this appendix we give the full-length closed expression of the integral over the variable v in Eq. (56), i.e.,

2 2 Z
2
2 e 2 +2ia 22
Mta =tb = d G () , (C1)
64 4 T 3 0
 16

The full expression of the integrand G() is

  "
1 2
 
2
G () := 2 2 sinh e 2
4 2 2 2i + 2 2
( + 2 ) 2
   
2
1
 1
+ e 2 2 + 2 2ie 2 (+2i) + i 2 + 1 Chi ( + i)2
2
     
+ i 2 1 2
+ i erf + 2 log( i) log ( + i) Shi ( + i)
2 2
    
1
(2i) 2

1 2 2


i
+e 2 2 i + + 1 Chi ( i) + 2 + i + 1 erf
2 2

h  i  
2 2
+ 4 i + + 1 log( i) 2 log( + i) + log ( i) + log ( i)2
    !#
2 2

1 2
2i( i) log ( i) + 2 i 1 Shi ( i) , (C2)
2

where Chi and Shi are the cosine and sine hyperbolic integral functions defined as
Z z
t sinh
Shi(z) := dt (C3)
0 t
Z z
t cosh 1
Chi(z) := e+ dt + z log, (C4)
0 t

and e here is the Euler-Mascheroni constant, and everything is expressed in terms of dinemsnionless variables as
detailed in table I.

[1] S. J. Summers and R. Werner, Physics Letters A 110, [18] S. J. Olson and T. C. Ralph, Phys. Rev. A 85, 012306
257 (1985). (2012).
[2] S. J. Summers and R. Werner, Journal of Mathematical [19] S. J. Olson and T. C. Ralph, Phys. Rev. Lett. 106,
Physics 28, 2440 (1987). 110404 (2011).
[3] M. Hotta, J. Phys. Soc. Jpn. 78, 034001 (2009). [20] T. C. Ralph and N. Walk, New Journal of Physics 17,
[4] M. Hotta, Phys. Lett. A 374, 3416 (2010). 063008 (2015).
[5] L. Susskind, L. Thorlacius, and J. Uglum, Phys. Rev. D [21] C. Sabn, B. Peropadre, M. del Rey, and E. Martn-
48, 3743 (1993). Martnez, Phys. Rev. Lett. 109, 033602 (2012).
[6] A. Almheiri, D. Marolf, J. Polchinski, and J. Sully, J. [22] A. Pozas-Kerstjens, J. Louko, and E. Martn-
High Energy Phys. 2013, 1 (2013). Martnez, Degenerate detectors are unable to harvest

[7] S. L. Braunstein, S. Pirandola, and K. Zyczkowski, Phys. spacelike entanglement (To appear in Phys. Rev. D -
Rev. Lett. 110, 101301 (2013). arXiv:1703.02982, 2017).
[8] G. Salton, R. B. Mann, and N. C. Menicucci, New J. [23] A. Pozas-Kerstjens and E. Martn-Martnez, Phys. Rev.
Phys. 17, 035001 (2015). D 94, 064074 (2016).
[9] A. Valentini, Phys. Lett. A 153, 321 (1991). [24] P. M. Alsing, I. Fuentes-Schuller, R. B. Mann, and T. E.
[10] B. Reznik, Found. Phys. 33, 167 (2003). Tessier, Phys. Rev. A 74, 032326 (2006).
[11] J. S. Benni Reznik, Alex Retzker, Phys. Rev. A (2005). [25] E. Martn-Martnez and J. Le on, Phys. Rev. A 80,
[12] G. V. Steeg and N. C. Menicucci, Phys. Rev. D 79, 042318 (2009).
044027 (2009). [26] E. Martn-Martnez and J. Le on, Phys. Rev. A 81,
[13] E. Martn-Martnez and N. C. Menicucci, Classical 032320 (2010).
Quant. Grav. 29, 224003 (2012). [27] E. Martn-Martnez and J. Le on, Phys. Rev. A 81,
[14] E. Martn-Martnez and N. C. Menicucci, Classical 052305 (2010).
Quant. Grav. 31, 214001 (2014). [28] I. Fuentes, R. B. Mann, E. Martn-Martnez, and
[15] E. Martn-Martnez, A. R. H. Smith, and D. R. Terno, S. Moradi, Phys. Rev. D 82, 045030 (2010).
Phys. Rev. D 93, 044001 (2016). [29] D. E. Bruschi, J. Louko, E. Martn-Martnez, A. Dragan,
[16] E. Martn-Martnez, E. G. Brown, and a. A. K. and I. Fuentes, Phys. Rev. A 82, 042332 (2010).
William Donnelly, Phys. Rev. A 88, 052310 (2013). [30] E. Martn-Martnez, L. J. Garay, and J. Le on, Phys.
[17] E. G. Brown, W. Donnelly, A. Kempf, R. B. Mann, Rev. D 82, 064006 (2010).
E. Martn-Martnez, and N. C. Menicucci, New J. Phys. [31] E. Martn-Martnez and I. Fuentes, Phys. Rev. A 83,
16, 105020 (2014). 052306 (2011).
 17

[32] M. Montero and E. Martn-Martnez, Phys. Rev. A 83, [49] J. Louko and A. Satz, Classical Quant. Grav. 25, 055012
062323 (2011). (2008).
[33] M. Montero and E. Martn-Martnez, Phys. Rev. A 84, [50] J. Louko and A. Satz, Classical Quant. Grav. 23, 6321
012337 (2011). (2006).
[34] M. Montero and E. Martn-Martnez, Journal of High [51] A. Satz, Classical Quant. Grav. 24, 1719 (2007).
Energy Physics 2011, 6 (2011). [52] A. Pozas-Kerstjens and E. Martn-Martnez, Phys. Rev.
[35] M. Montero, J. Le on, and E. Martn-Martnez, Phys. D 92, 064042 (2015).
Rev. A 84, 042320 (2011). [53] R. H. Jonsson, E. Martn-Martnez, and A. Kempf, Phys.
[36] N. Friis, P. K
ohler, E. Martn-Martnez, and R. A. Bertl- Rev. Lett. 114, 110505 (2015).
mann, Phys. Rev. A 84, 062111 (2011). [54] A. Blasco, L. J. Garay, M. Martn-Benito, and
[37] K. Bradler and R. J auregui, Phys. Rev. A 85, 016301 E. Martn-Martnez, Phys. Rev. Lett. 114, 141103
(2012). (2015).
[38] M. Montero and E. Martn-Martnez, Phys. Rev. A 85, [55] E. Martn-Martnez and J. Louko, Phys. Rev. Lett. 115,
016302 (2012). 031301 (2015).
[39] N. Friis, A. R. Lee, D. E. Bruschi, and J. Louko, Phys. [56] E. Martn-Martnez, Phys. Rev. D 92, 104019 (2015).
Rev. D 85, 025012 (2012). [57] A. Blasco, L. J. Garay, M. Martn-Benito, and
[40] M. Montero and E. Martn-Martnez, Phys. Rev. A 85, E. Martn-Martnez, Phys. Rev. D 93, 024055 (2016).
024301 (2012). [58] R. H. Jonsson, E. Martn-Martnez, and A. Kempf, Phys.
[41] N. Friis, A. R. Lee, and D. E. Bruschi, Phys. Rev. A 87, Rev. A 89, 022330 (2014).
022338 (2013). [59] R. H. Jonsson, arXiv:1512.05065 (2015).
[42] B. R. Iyer and A. Kumar, Journal of Physics A: Mathe- [60] A. G. S. Landulfo, Phys. Rev. D 93, 104019 (2016).
matical and General 13, 469 (1980). [61] M. A. N. adn Isaac L. Chuang, Quantum Computation
[43] W. G. Unruh, Phys. Rev. D 14, 870 (1976). and Quantum Information (Cambridge University Press,
[44] S. Takagi, Prog. Theor. Phys. 74, 501 (1985). 2011).
[45] S. Takagi, Prog. Theor. Phys. Supp. 88, 1 (1986). [62] G. Vidal and R. F. Werner, Phys. Rev. A 65, 032314
[46] D. Hummer, E. Martn-Martnez, and A. Kempf, Phys. (2002).
Rev. D 93, 024019 (2016). [63] M. O. Scully and M. S. Zubairy, Quantum Optics (Cam-
[47] S. DeWitt, General Relativity: An Einstein Centenary bridge University Press, 1997).
Survey, edited by S. W. Hawking and W. Israel (Cam- [64] J. Louko and V. Toussaint, Phys. Rev. D 94, 064027
bridge University Press, 1979) p. 680. (2016).
[48] K. J. Hinton, Classical Quant. Grav. 1, 27 (1984).