Backreaction and Unruh effect:

New insights from exact solutions of uniformly accelerated detectors

Shih-Yuin Lin∗
Physics Division, National Center for Theoretical Sciences, P.O. Box 2-131, Hsinchu 30013, Taiwan

B. L. Hu†
Maryland Center for Fundamental Physics, Department of Physics,
University of Maryland, College Park, Maryland 20742-4111, USA
(Dated: 25 July 2007)
Using nonperturbative results obtained recently for an uniformly accelerated Unruh-DeWitt detec-
tor, we discover new features in the dynamical evolution of the detector’s internal degree of freedom,
and identified the Unruh effect derived originally from time-dependent perturbation theory as oper-
ative in the ultra-weak coupling and ultra-high acceleration limits. The mutual interaction between
arXiv:gr-qc/0611062v2 13 Sep 2007

the detector and the field engenders entanglement between them, and tracing out the field leads to a
mixed state of the detector even for a detector at rest in Minkowski vacuum. Our findings based on
this exact solution shows clearly the differences from the ordinary result where the quantum field’s
backreaction is ignored in that the detector no longer behaves like a perfect thermometer. From a
calculation of the evolution of the reduced density matrix of the detector, we find that the transition
probability from the initial ground state over an infinitely long duration of interaction derived from
time-dependent perturbation theory is existent in the exact solution only in transient under special
limiting conditions corresponding to the Markovian regime. Furthermore, the detector at late times
never sees an exact Boltzmann distribution over the energy eigenstates of the free detector, thus in
the non-Markovian regime covering a wider range of parameters the Unruh temperature cannot be
identified inside the detector.

PACS numbers: 04.62.+v, 42.50.Lc 04.70.Dy

I. INTRODUCTION

The Unruh effect states that an observer while undergoing uniform acceleration in the Minkowski vacuum feels
as if it lives in a thermal state at the Unruh temperature. It is usually demonstrated by way of time-dependent
perturbation theory (TDPT) for a “particle detector” over an infinitely long duration of interaction [1, 2, 3]. Consider
the detector as a quantum mechanical object with internal degree of freedom Q coupling to a quantum field Φ by the
interaction Hamiltonian

HI = λ0 Q(τ )Φ(z µ (τ )) (1)

where λ0 is the coupling constant, τ is the proper time for the detector and z µ (τ ) is the trajectory of the uniformly
accelerated detector (UAD) with proper acceleration a. Suppose initially the detector-field system is in a product
state, i.e., they are uncorrelated,

| τ0 → −∞i = | E0 i ⊗ | 0M i , (2)

where | E0 i is the ground state of the free detector and | 0M i is the Minkowski vacuum of the free field. From TDPT,
the transition probability P0→n from the ground state to the n-th excited state | En i of the detector is given to first
order in γ ≡ λ20 /8πm0 , by [3]
 Z ∞ 2
1 X  i(En −E0 )τ /h̄

P0→n = 2 hEn , nk | dτ HI (τ )e | E0 , 0M i
h̄ nk 
−∞
2
 
λ0 2 En − E0
= 2 |hEn |Q(0)| E0 i| R , (3)
h̄ h̄

∗ Electronic address: sylin@phys.cts.nthu.edu.tw

† Electronic address: blhu@umd.edu
 2

which is nonzero. Here

∞
h̄ κη
Z
′
R(κ) = dτ dτ ′ e−iκ(τ −τ ) h 0M |Φ(z(τ ))Φ(z(τ ′ ))|0M i = 2πκ/a
. (4)
−∞ 2π e −1
R∞
is the response function with η ≡ −∞ dτ being the duration of interaction [20]. In particular, for a harmonic
oscillator with renormalized natural frequency Ωr , all the transition probabilities with n > 1 are O(γ 2 ), so the only
non-vanishing P of O(γ) is
λ20 η
P0→1 = (5)
4πm0 e2πΩr /a − 1
From the Planck factor in (5) one can identify the Unruh temperature TU = h̄a/2πkB . When a → 0, the transition
probability per unit time P0→1 /η vanishes, which implies that there is no excitation in an inertial detector initially
prepared in its ground state [2].
According to the above calculation of the transition probability one may expect that after the coupling is switched
on, the reduced density matrix of the detector would evolve continuously from the ground state to higher excited
states, with the detector finally ending up in equilibrium with a Boltzmann distribution ∼ exp[−En /(kB TU )] at the
Unruh temperature TU . Then one can read off the Unruh temperature from the late-time distribution inside the
detector.
In this note we use the exact solutions to an Unruh-Dewitt detector [1, 2] in uniform acceleration interacting
with a quantum field to show that the above described conventional scenario is good only in the Markovian regime
corresponding to the limits of ultra-high acceleration or ultra-weak coupling. The transition probability calculated
from the infinite-time TDPT is valid only in transient under restricted conditions. The evolution of the detector under
general conditions is quite different from the above picture.
The difference between this new understanding and the conventional picture originates from the fact that the full
interplay between the detector and the field is included here whereas the conventional approach based on perturbation
theory operative for infinite time ignores the backreactions – the use of TDPT over indefinite time amounts to invoking
a Markovian approximation which imposes rather severe limitations. The conventional wisdom is built upon conditions
which cannot reflect the most general features in the full dynamics of the detector-field system made possible here by
the nonperturbative solutions.
This paper is organized as follows: In Sec.II we consider a moving Unruh-DeWitt detector in 4-D Minkowski space
coupled to a quantum scalar field Φ and calculate the reduced density matrix after tracing out the field. In Sec.III
we calculate the transition probability of the detector from the ground state to an excited state. We elaborate on the
physical meaning of the two constants Λ0 , Λ1 in the theory. In Sec.IV we calculate the purity and the von Neumann
entropy or the entropy of entanglement of the detector-field system and identify an effective temperature Teff which
is ostensibly different from the Unruh temperature. We identify the range of validity of the conventional results to
that which corresponds to making a Markovian approximation with its strong limitations. In Sec.V we summarize
our main results with remarks pertaining to the issues raised in this paper and conclude with a comment on how our
nonperturbative results may bear on a new understanding of the Hawking effect of black hole radiance.

II. REDUCED DENSITY MATRIX FOR DETECTOR

Consider an Unruh-DeWitt (UD) detector moving in (3+1) dimensional Minkowski space. The total action is given
by[4]
m0 h 1
Z i Z Z Z
S = dτ (∂τ Q)2 − Ω20 Q2 − d4 x ∂µ Φ∂ µ Φ + λ0 dτ d4 xQ(τ )Φ(x)δ 4 (xµ − z µ (τ )) , (6)
2 2
where Q is the internal degree of freedom of the detector, assumed to be a harmonic oscillator with bare mass m0
and bare natural frequency Ω0 . The scalar field Φ is assumed to be massless, and λ0 is the coupling constant.
This UD detector theory behaves like the quantum Brownian motion (QBM) of a harmonic oscillator interacting
with an Ohmic bath provided by the 4-D scalar quantum field [5]. The QBM model is a useful comparison because it
shows clearly the dissipative and stochastic behavior of the dynamics arising from the interplay between the system
and the environment, and the influence functional treatment incorporates the backreaction of the environment on
the system (which could be either the quantum field or the harmonic oscillator depending on what one is after) in a
self-consistent way. Since the QBM model indicates that there are nontrivial activities at zero temperature [5, 6, 7],
we caution that even for the a = 0 case the detector is not just laying idle but has interesting physical features due
to its interaction with the vacuum fluctuations in the quantum field.
 3

Unruh and Zurek [7] have studied an exactly solvable QBM model where a harmonic oscillator interacts with a
massless scalar field in 2-D. They derived the exact master equation for the reduced density matrix of the system
(oscillator) at a temperature determined by the initial state of the field, and observed some general features different
from the conventional Markovian results (valid for an ohmic bath at ultra-high temperature). One feature is the
dependence of the UV cut-off in the master equation and the reduced density matrix, and thus also in the von
Neumann entropy of the system. When the interaction is switched on, the factors in the master equation and
the entropy have initial “jolts” over a very short time scale corresponding to the UV cut-off frequency, which cause
significant change in the coherence of the quantum state of the system, while the coherence residing in each subsystem
is rapidly transferred into the correlation between them. This effect is discussed in detail in [6]. Below we will see a
similar behavior occurring in UD detector theory beyond the Markovian regime.
Suppose the initial state of the system at τ0 is given by (2), a direct product of the ground state for Q and the
Minkowski vacuum for Φ. In the Schrödinger representation, this initial state is a product of Gaussian functions,

d3 k d3 k ′ kk′
 Z 
ψ0 (τ0 ) = N exp −AQ2 − B Φ k k .
Φ ′ (7)
(2π)3 (2π)3

Owing to the linearity of the model, this state must evolve into a general Gaussian form
h ′
i
ψ0 (τ ) = N (τ ) exp −A(τ )Q2 − B kk (τ )Φk Φk′ − 2C k (τ )Φk Q , (8)

after the coupling is switched on. Here we have used the DeWitt notation where each upper-lower indices pairing
indicates an integration (or summation). So the density matrix for this pure state in the (Q, Φk ) representation can
be written as ρ[Q, Φk , Q′ , Φ′k ; τ ] = ψ0 [Q, Φk ; τ ]ψ0∗ [Q′ , Φ′k ; τ ], and the reduced density matrix defined by tracing out
the field Φk reads
Z
ρ (Q, Q ; τ ) = DΦk ψ0 [Q, Φk ; τ ]ψ0∗ [Q′ , Φk ; τ ] = exp −Gij (τ )Qi Qj − F (τ ) ,
R ′
 
(9)

where i, j = 1, 2, Qi = (Q, Q′ ). The factors Gij and F could be obtained by solving the master equations [8]. But
here we do not need to solve them directly, because the two-point functions of the detector have been obtained in
the Heisenberg picture in Ref.[4]. (These results are placed in Appendix A for convenience.) One can reconstruct
the complete evolution of the reduced density matrix using those two-point functions of the detector by solving the
simple algebraic relations
1
G11 + G22 + 2G12 = , (10)
2 h Q2 i
2 h i
G11 + G22 − 2G12 = 2 h P 2 i h Q2 i − h P, Q i2 , (11)
h̄ h Q2 i
i h P, Q i
G11 − G22 = − , (12)
h̄ h Q2 i

where h P, Q i ≡ 12 h (P Q + QP ) i = (m0 /2)(d/dτ ) h Q2 i (and h [P, Q] i = −ih̄). Substituting (A1-A4) gives the
values of Gij at every moment, and the factor F in (9) will be determined by a normalization condition.
To compare with the transition probability (5), the reduced density matrix (9) has to be further transformed to
the representation in the basis of energy-eigenstate for the “free” harmonic oscillator Q:
X
ρR (Q, Q′ ) = ρR ′
m,n φm (Q)φn (Q ) (13)
m,n≥0

where the wave function for the n-th excited state is

α
r
2 2
φn (Q) = √ Hn (α Q)e−α Q /2 (14)
2n n! π

with the Hermite polynomial Hn (x) and the real constant

r
m 0 Ωr
α= . (15)
h̄
 4

Here Ωr is the renormalized natural frequency [4]. Following the method given by Ruiz [9], the matrix elements ρR
m,n
could be extracted by comparing the coefficients of sm n
1 s2 on both sides of the following equation,
r
X 2m+n α 1  11 2
ρR m n
g̃ s1 + g̃ 22 s22 + 2g̃ 12 s1 s2 ,

m,n s1 s2 = q exp (16)
m!n! G̃
m,n 2G̃ h Q2 i

where
α4
∗ h P2 i iα2 h P, Q i
g̃ 11 = g̃ 22
= − 2 + , (17)
4 4h̄ h Q2 i 2h̄ h Q2 i
" #
2
α2 1 h P2 i h P, Q i
g̃ 12 = = − − 2 + 2 , (18)
2 4 h Q2 i h̄ h̄ h Q2 i
" #
2
α4 α2 1 h P2 i h P, Q i h P2 i
G̃ = + + − + . (19)
4 2 4 h Q2 i h̄2 h̄2 h Q2 i 4h̄2 h Q2 i

III. TRANSITION PROBABILITY

The transition probability from the initial ground state to the first excited state is the m = n = 1 component of
the reduced density matrix in energy eigenstate representation (32):
h i
h̄ h P 2 i h Q2 i − h P, Q i2 − (h̄2 /4)
ρR1,1 = n o3/2 . (20)
2
h̄−1 α−2 h P 2 i + (h̄/2) [h̄α2 h Q2 i + (h̄/2)] − h P, Q i


Expanding the two-point functions of the detector (A1-A4) in terms of γ, h Q2 i and h P 2 i look like

h̄
h Q2 i ≈ + γ h Q2 i(1) + O(γ 2 ),
2m0 Ωr
h̄
h P 2 i ≈ m0 Ωr + γ h P 2 i(1) + O(γ 2 ), (21)
2
and h P, Q i ∼ O(γ), which yield
γ h i
ρR
1,1 ≈ h P 2 i(1) + m20 Ω2r h Q2 i(1) + O(γ 2 ). (22)
2h̄m0 Ωr

When η ≡ τ − τ0 ≫ a−1 the approximate value up to the first order of γ ≡ λ20 /8πm0 becomes

λ20
 
η≫a−1 η Λ1 + Λ0 − 2 ln(a/Ωr )
ρR
1,1 | γη→0 −→ + (23)
4πm0 e2πΩr /a − 1 2πΩr

from (A1-A4). Here Λ0 and Λ1 are large constants in (A3) and (A4): Λ1 denotes the time resolution/frequency cut-off
of this detector theory, while Λ0 denotes the time scale of switching on the interaction.

A. Range of validity in conventional results

We see that the first term of (23) gives the conventional transition probability (5) from TDPT over infinite time.
Only when Ωr η ≫ Λ1 , Λ0 , or a is extremely large, can the second term in (23) be neglected and (5) recovered. Hence
the conventional transition probability (5) is valid only in the limits of (a) ultra-high acceleration (a ≫ Ωr and
Λ1 ≪ aη ≪ aγ −1 ) or (b) ultra-weak coupling (a−1 , Ω−1r Λ1 ≪ η ≪ γ
−1
).
−1 −1
Note that, in obtaining (23), we have assumed a ≪ η ≪ γ , when the system is still in transient. If a < γ,
the conventional transition probability (5) has no chance to dominate at all. Mathematically, the first term in (23) is
contributed by the poles at ±Ω + iγ or ±Ω − iγ in the κ-integrations of two-point functions (see Eqs.(60) and (67) in
[4]). But when a < γ the poles at ±ia would be closer to the real κ axis than the poles at ±Ω ± iγ, while the poles on
 5

the imaginary axis κ = ±ina, n ∈ N become very dense for small a, so their contributions dominate the result and
the first term of (23) becomes unimportant.
In particular, the a = 0 case is beyond the reach of TDPT over infinite time shown in Section I, and the conventional
wisdom from perturbation theory that no transition occurs in an inertial detector is untenable. In contrast, our
result indicates that the evolution of ρR 1,1 with a = 0 behaves qualitatively similar to those cases with nonzero
acceleration [4]. This agrees with the expectation from the observation that the UD detector theory is a special case
of quantum Brownian motion [5], where there is nontrivial interplay between the oscillator and the quantum field at
zero temperature.

B. Essential roles of Λ0 and Λ1

We see the presence of two constants Λ0 , Λ1 and may wonder whether they have some real physical meaning or
are just part of a calculational tool or artifact. We will address these concerns here. To begin with, the presence
of large constants corresponding to frequency cut-offs in the coincidence limit of two-point functions of the detector
is a common feature for detector-field theories and quantum Brownian motion. For example, the Raine-Sciama-
Grove(RSG) model in (1+1)D [18], in which the detector acts like a harmonic oscillator in a sub-Ohmic bath, also
has two-point functions of the detector with dependence of large constants due to the infrared cut-off.
Let us explore the physical meaning of Λ0 and Λ1 . Since Λ0 corresponds to the time scale of switching on the
interaction, it could be finite in real processes, and for every finite value of Λ0 , the Λ0 terms in all two-point functions
vanish at late times (see (A3) and (A4)). Hence Λ0 will not be present in the late-time results. On the other hand,
Λ1 is a constant of time, appearing from the very beginning and never decays. One way to see that it is a quantity
of real physical
p meaning is that if Λ1 was subtracted naively, the uncertainty principle will be violated, namely,
∆P ∆Q = h P 2 i h Q2 i < h̄/2 at late times for a is small enough, as shown in FIG.1.
Actually (22) is formally identical to the first-order transition probability from TDPT for a UAD with finite duration
of interaction (τ0 , τ ) [10], so is the result (23). To verify this, first note that the O(γ) term in (22) is the O(λ20 ) term
of
1
hE0 , 0M |ĤQ (η)| E0 , 0M i . (24)
h̄Ωr
Here ĤQ (η) is defined by
i
Rτ Rτ
dsĤ(s) − h̄i dsĤ(s)
ĤQ (η) ≡ e h̄ τ0
ĤQ e τ0
, (25)
where Ĥ(τ ) is the total Hamiltonian for the combined system, ĤQ is the Hamiltonian for the “free” detector so that
| E0 i is an eigenstate of ĤQ . Then in the interaction picture it is straightforward to show that
Z τ Z τ
R λ20  2 i
ρ1,1 ≈ dτ1 dτ2 (E1 − E0 ) hE1 |Q̂(0)| E0 i e− h̄ (E1 −E0 )(τ1 −τ2 ) h 0M |Φ(z(τ1 ))Φ(z(τ2 ))|0M i , (26)
 
h̄Ωr τ0 τ0

from which one recovers the finite-time transition probability from TDPT (cf. Eq.(3)) since E1 − E0 = h̄Ωr . Math-
ematically the large constants Λ0 and Λ1 are formally the same as the divergences found in Ref. [10]. In Ref. [11]
it has been shown that these divergences can be tamed if one switches on and off the interaction smoothly, so can
Λ0 . Nevertheless, the physical meaning of Λ1 here is totally different from those divergences of the finite-time UAD.
Here we are looking at the real-time causal evolution problem (“in-in” formulation) rather than a scattering transition
amplitude (“in-out” formulation) problem. Also in our set-up we never turn off the coupling and Λ1 is present at
every moment.
As was already found in Ref. [7] the constant Λ1 changes the scenario of the evolution of the system in an essential
way. Λ1 is present whenever the coupling is on, so right after the initial moment the values of ρR m,n , m, n > 0 jump
from zero to large numbers depending on Λ1 , in a time scale indistinguishable from zero at the level of precision of
this theory. This means that the initial distribution of ρR R
m,n peaked at the element ρ0,0 would upon the switch-on of
the coupling collapse rapidly (rather than smoothly diffuse) into a distribution widely spread (if Λ1 is large) over the
whole density matrix. After that the density matrix begins to redistribute itself and finally settles into a steady state.

IV. PURITY, ENTROPY AND EFFECTIVE TEMPERATURE IN DETECTOR

After substituting the late-time two-point functions (A7) and (A8) with Λ1 kept in h P 2 i into (16), we find that, in
steady state, the diagonal terms of ρR
m,n in the free eigenstate representation do not assume the form of a Boltzmann
 6

DPDQ
€€€€€€€€€€€€€
12

1.5

1.25

0.75

0.5

0.25

a
2 4 6 8 10

FIG. 1: Plot
p of (∆P ∆Q)/(h̄/2) at late times with Λ1 = 0. One can see that there exists a range of a (less than about 4) where
∆P ∆Q ≡ h P 2 i h Q2 i is smaller than h̄/2, violating the uncertainty principle. So Λ1 cannot be naively subtracted. Other
parameters in this plot are taken to be γ = 0.1, Ω = 2.3, m0 = 1, and h̄ = c = 1.

distribution, while the off-diagonal terms of ρR

m,n do not vanish at late times. Therefore the free eigenstates are no
longer orthogonal when interactions are introduced.
To define an effective temperature and the entropy of entanglement, one should choose another set of orthogonal
states which have a similar form as the energy eigenstate of the free detector but with α in (15) replaced by
1/4
h P2 i

α= 2 . (27)
h̄ h Q2 i
Since our initial state is Gaussian, one can further apply a canonical transformation introduced by Unruh and Zurek
[7] to diagonize the reduced density matrix into
n
(U − h̄/2)
ρ̃R
m,n = h̄δmn n+1 , (28)
(U + h̄/2)
where
q
2
U≡ h P 2 i h Q2 i − h P, Q i . (29)
is the uncertainty function and U ≥ h̄/2 is the Robertson-Schrödinger uncertainty relation. From (28) one immediately
obtains the purity P for the reduced density matrix (9),
h̄/2
P ≡ T r[ρ̃R ρ̃R ] = . (30)
U
In this case P −1 is the ratio of the uncertainty function U to its minimal value h̄/2, hence one always has P ∈ (0, 1)
because of the uncertainty principle. Also the von Neumann entropy or the entropy of entanglement reads
       
R R U 1 U 1 U 1 U 1
S ≡ −T rρ̃ ln ρ̃ = + ln + − − ln − (31)
h̄ 2 h̄ 2 h̄ 2 h̄ 2
which is a measure of the entanglement between the detector and the quantum field.
Further, in steady state, h P, Q i → 0, so (28) becomes
hp in
h P 2 i h Q2 i − h̄/2
γη≫1
ρ̃R
m,n −→ h̄δmn hp
R
in+1 ≡ δmn ρ0,0 × e
−nh̄Ωr /kB Teff
. (32)
h P 2 i h Q2 i + h̄/2

Now the diagonal terms assume a Boltzmann distribution with the effective temperature
" p !#−1
kB h P 2 i h Q2 i + h̄/2
Teff = ln p . (33)
h̄Ωr h P 2 i h Q2 i − h̄/2
This is not surprising: Even for such a simple system containing only two coupled harmonic oscillators, the ground
state of the total system also looks like a thermal state with some effective temperature in view of the reduced density
matrix for one of the oscillator [12].
 7

Η
5 10 15 20 25 30

FIG. 2: The evolution of the entropy of entanglement S(η) in (31) for a detector initially in the ground state. At late times,
the entropy converges to a large number if Λ1 is large. Here the parameters are taken to be a = 2, γ = 0.1, Ω = 2.3, m0 = 1,
Λ1 = Λ0 = 10000, and h̄ = c = 1.

A. Realistic cases: ΩΛ1 ≫ a, γ

Since Λ1 corresponds to the cut-off frequency of the theory, in real processes it is most likely that ΩΛ1 ≫ a, γ.
The evolution of the entropy of entanglement in this case is shown in FIG.2. At the initial moment the entropy of
entanglement has a sudden jump from zero to a large number ∼ O(ln Λ1 ) while the entropy converges to a number of
the same order at late times. Indeed, for Λ1 ≫ 1, the late-time entropy of entanglement has approximately the value
     
1 1 γ a i γ − iΩ i γ + iΩ
S ≈ ln Λ1 + 1 + ln + ψ 1 + − ψ 1 + + O(Λ−11 ), (34)
2 2 π 2 Ω2r Ω a Ω a
which is dominated by the ln Λ1 term (the Λ0 term dies out at late times as long as Λ0 is finite). This indicates that
the constant Λ1 is due to the entanglement between the detector and the infinitely many degrees of freedom of the
field.
For Λ1 ≫ 1, the effective temperature is approximately
    1/2
h̄Ωr p a i γ − iΩ i γ + iΩ −1/2
Teff ≈ γΛ1 + ψ 1 + − ψ 1 + + O(Λ1 ). (35)
πkB Ω2r Ω a Ω a
which, as we can see, is totally different from the Unruh temperature: It is determined not only by the proper
acceleration a, but also by the properties of√the detector (Ωr ), the interaction (γ), and the frequency cutoff of this
theory (Λ1 ). Moreover Teff is very large (O( Λ1 )) for large Λ1 and non-vanishing even when a → 0.

B. Inertial detector: a = 0

Substituting (A11) and (A12) into (30), (31) and (33), one can write down in closed form the purity, entropy of
entanglement and the effective temperature of an inertial detector (a = 0). These quantities are still nontrivial. In
particular, in the ultra-weak coupling limit γΛ1 ≪ 1, one has
 
γ(Λ1 − 1) γ(Λ1 − 1)
S ≈ 1 − ln + O(γ 2 ), (36)
πΩr πΩr
h̄Ωr
Teff ≈ h i (37)
γ(Λ1 −1)
−kB ln πΩr + O(γ 2 )

[Note that limǫ→0+ ln(−1 + ǫ) = −iπ).] Both quantities go to zero as γ → 0, when the description of the combined
system returns to the nearly free theory.

C. Conventional result: the Markovian regime

As we mentioned in Section III.A, the transition probability calculated from TDPT over infinite time is valid only in
the ultra-high acceleration (temperature) limit or the ultra-weak coupling limit, both corresponding to the Markovian
 8

regime for this model (with the quantum field as an Ohmic bath). In these limits the entropy of entanglement and
effective temperature do have the approximate values expected in the conventional picture:
(a) Ultra-high acceleration (temperature) limit, a ≫ γΛ1 , Ω. In this limit, the late-time entropy of entanglement
reads
a
S ≈ ln + 1 + O(a−1 ), (38)
2πΩ
and the effective temperature
h̄γ  a
Teff ≈ TU + Λ1 + γe − ln + O(a−1 ), (39)
kB π Ω
is very close to the Unruh temperature TU = h̄a/2πkB .
(b) Ultra-weak coupling limit, γΛ1 ≪ a, Ω. In this limit, the late-time entropy of entanglement and effective
temperature read
       
πΩ πΩ πΩ 2γ a iΩ iΩ (1) iΩ
S ≈ coth − ln 2 sinh + Λ1 − ln − Re ψ + ψ + O(γ 2 ), (40)
a a a a Ω a a a
h̄γa2
     
2 πΩ a iΩ iΩ (1) iΩ
Teff ≈ TU + sinh Λ 1 − ln − Re ψ + ψ + O(γ 2 ). (41)
kB π 3 Ω2 a Ω a a a

Again the effective temperature of the detector is very close to the Unruh temperature.
It is well-known that the quantum mechanical time-dependent perturbation theory invokes the equivalent of a
Markovian approximation (e.g., as in the derivation of the Pauli master equation [13]). Indeed, both the above
limits correspond to the Markovian regime of the quantum Brownian motion, its dynamics being described by the
master equation of Caldeira and Leggett [14] (in contradistinction to the non-Markovian master equation of Hu, Paz,
and Zhang [6]). There is admittedly a fluctuation-dissipation relation between the fluctuations in the field and the
dynamics of the detector and thus backreaction is present, but with ultra-weak coupling it is restricted to the linear
response regime (which has limited domain of applicability compared to the full nonequilibrium dynamics of the
combined system). Here, the thermal bath is only slightly affected by the back reaction from the detector to the field,
namely, the detector acts essentially as a test particle in the field. It is only under these special assumptions that the
detector can come to equilibrium with an approximate thermal field at the Unruh temperature. Beyond this regime,
the dynamics of the detector-field system has a totally different character.

V. REMARKS

We conclude with a few remarks:

a. Working range of Unruh-DeWitt detectors. Conventional results from time-dependent perturbation theory
over an infinitely long duration of interaction are trustworthy only in the ultra-high acceleration (or temperature)
limit and the ultra-weak coupling limit, under the Markovian regime. In the non-Markovian regimes, the large
constants Λ1 and Λ0 alter the scenario fundamentally. With the presence of the large constant Λ1 , which reflects
the entanglement between the detector and the infinitely many degrees of freedom of the field, the evolution of the
reduced density matrix of the detector collapses (sudden rapid change), as shown here, rather than diffuses (smooth
gradual change), as conjured in the conventional picture. Furthermore, the detector at late times never sees an exact
Boltzmann distribution over the energy eigenstates of the free detector, let alone the Unruh temperature. The strong
interplay between the detector and the field makes the late-time quantum states for the detector-field system highly
entangled and never factorizable (an analogy in quantum optics is the photon-atom bound state [15].)
In equilibrium conditions, √
one can diagonalize the reduced density matrix and define an effective temperature (35),
which is usually large (∼ O( Λ1 )) and does not vanish even for an inertial detector in Minkowski vacuum as long as
the coupling is on. Only in the Markovian regime indicated above could this effective temperature get very close to
the Unruh temperature.

b. Correspondence with QBM. For a uniformly accelerated UD detector in (3+1)D with proper acceleration a
the Unruh effect [1, 2, 3] attests that it should behave the same way as an inertial UD detector in contact with a
thermal bath at Unruh temperature TU , or more precisely, as an inertial harmonic oscillator in contact with an Ohmic
 9

bath at TU [8]. This is clear from an examination of the integral in the derivation of the two-point functions of the
detector, for example, Eq.(60) in Ref.[4]:

λ20 h̄ κdκ
Z
′
h Q(η)Q(η ) iv ∼ 2 2 [. . .], (42)
(2π) m0 1 − e−2πκ/a
Therefore for an inertial UD detector in contact with an Ohmic bath at temperature T , one can simply substitute
2πkB T /h̄ for a in the above results to get its purity, entropy and effective temperature.
However, examining this from the vantage point of the exact solutions we obtained, we see the above statements
are accurate only at the initial moment. After the coupling is switched on, the quantum state of the field will have
been changed by the detector, so the field is no longer in the Minkowski vacuum and it does not make exact sense to
say that the detector is immersed in a thermal state (or any state defined in the test-field description, i.e., where the
field is assumed not to be modified by the presence of the detector).
A theorem by Bisognano and Wichmann (BW) [16] states that the Minkowski vacuum, which is uniquely charac-
terized by its invariance under all Poincaré translations, is a Kubo-Martin-Schwinger (KMS) state with respect to
all observables confined to a Rindler wedge. One may wonder why it does not apply here. The reason is that the
BW theorem refers to the vacuum state of a quantum field alone, not the combined detector-field system. Even when
the combined system is in a steady state, the quantum state of the interacting field is not invariant under spatial
translations in Minkowski space, hence is not covered under the assumption of the BW theorem pertaining to Poincaré
invariance.
Actually the Planck factor in (42) is a consequence of the BW theorem. Nevertheless, it is derived from only the
free-field-solution part of the complete interacting field (see Eqs. (28), (56) and (58) in Ref. [4]). Here the factor is
not distorted by the interaction simply because the field is linear and the coupling is bilinear. For nonlinear fields or
couplings it would have a nonPlanckian spectrum and the departure from the conventional picture would be more
pronounced.

c. Backreaction and memory effects. It is common knowledge in nonequilibrium statistical mechanics [17] that
for two interacting subsystems the two ordinary differential equations governing each subsystem can be written as
an integro-differential equation governing one such subsystem, thus rendering its dynamics non-Markovian, with the
memory of the other subsystem’s dynamics registered in the nonlocal kernels (which are responsible for the appearance
of dissipation and noise should the other subsystem possesses a much greater number of degrees of freedom and are
coarse-grained in some way). Thus inclusion of back-action self-consistently in general engenders non-Markovian
dynamics. For our problem the two subsystems are the detector and the quantum field. Combining Eqs. (28), (30)
and (10) in Ref.[4], we can write down the equation of motion for the detector evolution functions,
 Z ∞ 
2 2 (+) λ0 (+) ′ ′ (+) ′
(∂τ + Ω0 )q (τ ; k) = f (z(τ ); k) + λ0 dτ Gret (z(τ ); z(τ ))q (τ ; k) , (43)
m0 0 τ0

which is an integro-differential equation. The backreaction to the detector is registered through the retarded Green’s
function Gret of the field. Various approximations are usually invoked to solve this equation, amongst which the
most common is the Markovian or memoryless approximation. This is what enters in the conventional derivation of
the Unruh effect, but as we see above, is a very special and nongeneric subcase.

d. Hiding Λ1 ? One may wonder why the large constant Λ1 cannot be absorbed by any parameter of this theory
so one can renormalize Teff . To begin with, as we mentioned in Sec.III B, the UD detector theory is not a fundamental
theory to meet the renormalizability requirement, and the presence of cut-offs as physical parameters is an expected
feature which characterizes the range of validity of this semiclassical theory, just like the Compton wavelength of
the electron acting as a cut-off in quantum optics. Second, the large constant Λ1 is not present in the renormalized
stress-energy tensor of the field induced by the detector [4], thus Teff may not be a directly measurable quantity. The
interference between the vacuum fluctuations and their back reaction cancels this cut-off dependence so that Λ1 is
not observable outside of the detector.
In some cases, though, Λ1 term can be subtracted from the physical quantities of the detector after these quantities
are worked out. For example, one can subtract Λ1 from the detector energy defined in Eq.(82) of Ref.[4] because
during a physical process only the difference of energy matters.

e. Hawking effect. It is tempting to see if analogous descriptions can be made and implications drawn for the
Hawking effect. For a UD detector at rest in a static gravitational field, the response of the detector is similar. The
simplest example is the UD detector fixed at radius r far from a Schwarzschild black hole, while the scalar field
is in Hartle-Hawking (HH) vacuum (which is the counterpart of Minkowski vacuum for the uniformly accelerated
 10

detector in Minkowski space [3]). The response function per unit time for a massless scalar field in the Schwarzschild
background is given by Candelas (Section V in Ref.[19]). According to [19], when r → ∞, the response function
for HH vacuum is exactly the same as the one in the uniform acceleration case (Eq.(58) in [4]) with the Unruh
temperature a/2πkB replaced by the Hawking temperature TH ≡ 1/8πkB M for the Schwarzschild black hole with
mass M . Therefore the results in this paper can be directly applied to the case of UD detector fixed at r → ∞ outside
a Schwarzschild black hole by neglecting its back reaction to spacetime. The effective temperature Teff read off from
such a detector in steady state is given by (33) with a replaced by 1/4M rather than the Hawking temperature TH .
For the case with the detector sitting very close to the event horizon, due to the effective potential barrier in the
radial equation for the field, one has to consider the back-scattering of the retarded field induced by the detector in
addition to the vacuum fluctuations described by the response function. We expect that the effective temperature
read off from the detector in steady state would be even more complicated but definitely different from TH . We hope
to report on this investigation in a later paper.

Acknowledgement We wish to thank Chris Fleming and Albert Roura for discussions, Bill Unruh for pointing out
his paper with Zurek, and Ted Jacobson and Ralf Schutzhold helpful suggestions in presenting our results. This work
is supported in part by NSF grant PHY-0601550.

APPENDIX A: TWO-POINT FUNCTIONS OF INERTIAL DETECTORS IN MINKOWSKI VACUUM

Recall that for a uniformly accelerated UD detector with proper acceleration a [4], once we choose the factorized
initial state (2), the two-point functions split into two part, h . . . i = h . . . ia + h . . . iv . The coincidence limits of the
two-point functions of the detector with respect to its ground state (with q0 in Ref.[4] being zero) read

h̄θ(η) −2γη  2
h Q(η)2 ia = Ωr − γ 2 cos 2Ωη + γΩ sin 2Ωη ,

2
e (A1)
2Ω Ωr m0
h̄Ωr
h Q̇(η)2 ia θ(η)e−2γη Ω2r − γ 2 cos 2Ωη − γΩ sin 2Ωη ,
 
= 2
(A2)
2Ω m0
which are independent of a, while the coincidence limits of the two-point functions of the detector with respect to
Minkowski vacuum read
1
h Q(η)2 iv = lim h {Q(η), Q(η ′ )} iv
′ η →η 2
2h̄γ n a  −2γη 2
= θ(η)Re Λ 0 − ln e sin Ωη
πm0 Ω2 Ω
a −(γ+a)η Fγ+iΩ (e−aη ) iΩ −iΩη F−γ−iΩ (e−aη )
     
iΩ iΩη −iΩη
+ e − e + 1+ e −e
2 γ + iΩ + a γ γ + iΩ − a γ
  
1 iΩ iΩ
− + e−2γη + 1 − e−2iΩη (ψγ+iΩ + ψ−γ−iΩ )
4 γ γ
   
iΩ −2γη iΩ −2iΩη π
− − +e +1−e iπ coth (Ω − iγ) , (A3)
γ γ a
2h̄γ n a 2  a  −2γη
h Q̇(η)2 iv = 2
θ(η)Re Λ1 − ln Ω + Λ0 − ln e (Ω cos Ωη − γ sin Ωη)2
πm0 Ω Ω Ω
Fγ+iΩ (e−aη ) iΩ −iΩη F−γ−iΩ (e−aη )
     
a iΩ iΩη
+ (γ + iΩ)2 e−(γ+a)η e + 1− e − e−iΩη
2 γ + iΩ + a γ γ + iΩ − a γ
  
1 iΩ iΩ
+ (γ + iΩ)2 + e−2γη − 1 + e−2iΩη (ψγ+iΩ + ψ−γ−iΩ )
4 γ γ
   
iΩ −2γη iΩ −2iΩη π
− − +e −1+e iπ coth (Ω − iγ) . (A4)
γ γ a
p
Here again η ≡ τ − τ0 is the duration of interaction, γ ≡ λ20 /8πm0 , Ω ≡ Ω2r − γ 2 , Fs (y) is defined by the hyper-
geometric function as
 s s 
Fs (y) ≡ 2 F1 1 + , 1, 2 + ; y , (A5)
a a
 11

and
 s
ψs ≡ ψ 1 + (A6)
a
is the digamma function. The large constant Λ0 ≡ − ln Ω|τ0 − τ0′ | − γe with the Euler’s constant γe corresponds to
the time scale of switching-on the interaction, so Λ0 could be finite in real processes, and for every finite value of Λ0 ,
the terms containing Λ0 in (A3) and (A4) vanish as γη → ∞. The other large constant Λ1 ≡ − ln Ω|τ − τ ′ | − γe
corresponds to the time-resolution or the cut-off frequency of this theory. Note that here we use slightly different
definitions of Λ1 and Λ0 from those of Ref.[4] to make the arguments in logarithm functions dimensionless. Note also
that the above results are valid when Ω|τ − τ ′ | ≪ 1 and Ω|τ0 − τ0′ | ≪ 1. Beyond this regime, the form of the above
“coincidence” limit of two-points functions should be modified.
At late times (γη ≫ 1), the two-point functions of the detector with respect to the initial ground state (h. . .ia ) die
away, and the two-point functions of the detector saturate to
 
2 h̄ ia
hQ i → Re − 2iψγ+iΩ , (A7)
2πm0 Ω γ + iΩ
  
h̄m0 ia 2  a
h P 2 i = m20 h Q̇2 i → Re (Ω − iγ)2 − 2iψγ+iΩ + h̄m0 γ Λ1 − ln , (A8)
2πΩ γ + iΩ π Ω
which are identical to the results of the quantum Brownian motion of a harmonic oscillator in contact with an Ohmic
bath at the Unruh temperature initially [8]. To satisfy the uncertainty principle for all a (see FIG.1), here we keep
the constant Λ1 , which was subtracted in Ref.[4] because of the observation that Λ1 will not be seen outside of the
detector.
When a → 0, the regular terms in the two-point functions diverge as ln a and cancel the ones following Λ0 and Λ1 ,
so the two-point functions remain well-behaved. In this limit the two-point function (A3) continuously approaches
h̄γ
h Q(η)2 iv |a=0 = θ(η)Re 2Λ0 e−2γη sin2 Ωη

πm0 Ω2
 
iΩ iΩ
− Γ [0, (γ + iΩ)η] − e−2γη + 1 − e−2iΩη Γ [0, −(γ + iΩ)η]
γ γ
     
iΩ iΩ γ  Ω
− + e−2γη + 1 − e−2iΩη ln + i − πe−2γη + sin 2Ωη , (A9)
γ γ Ω γ
where Γ is the incomplete gamma function, while (A4) goes to
h̄γ n h
2
i
h Q̇(η)2 iv |a=0 = 2
θ(η)Re 2 Λ1 Ω2 + Λ0 e−2γη (Ω cos Ωη − γ sin Ωη)
πm Ω
 0  
iΩ iΩ
+ Γ [0, (γ + iΩ)η] + e−2γη − 1 + e−2iΩη Γ [0, −(γ + iΩ)η]
γ γ
    
iΩ iΩ γ
+ + e−2γη − 1 + e−2iΩη ln + i (γ + iΩ)2
γ γ Ω
 
−2γη Ω 2 2 2 2
+ πe − (Ω + γ ) + 2γΩ cos 2Ωη + (Ω − γ ) sin 2Ωη . (A10)
γ
At late times, one has
h̄i γ − iΩ
h Q2 i |a=0 → ln , (A11)
2πm0 Ω γ + iΩ
γ2
   
h̄m0 i γ − iΩ
h P 2 i |a=0 → 2 2
(Ω − γ ) ln + γ 2Λ1 − ln 1 + 2 . (A12)
π 2Ω γ + iΩ Ω
Substituting this into (30), (31) and (33), one obtains the purity, entropy of entanglement and effective temperature
for the inertial detector in the Minkowski vacuum.

[1] W. G. Unruh, Phys. Rev. D 14, 870 (1976).

 12

[2] B. S. DeWitt, in General Relativity: an Einstein Centenary Survey, edited by S. W. Hawking and W. Israel (Cambridge
University Press, Cambridge, 1979).
[3] N. D. Birrell and P. C. W. Davies, Quantum Fields in Curved Space (Cambridge University Press, Cambridge, 1982).
[4] S.-Y. Lin and B. L. Hu, Phys. Rev. D 73, 124018 (2006).
[5] B. L. Hu and A. Matacz, Phys. Rev. D 49,6612 (1994)
[6] B. L. Hu, J. P. Paz and Y. Zhang, Phys. Rev. D 45, 2843 (1992).
[7] W. G. Unruh and W. H. Zurek, Phys. Rev. D 40, 1071 (1989).
[8] C. H. Fleming, B. L. Hu and A. Roura, arXiv: 0705.2766 [quant-ph].
[9] M. J. Ruiz, Phys. Rev. D 10, 4306 (1974).
[10] B. F. Svaiter and N. F. Svaiter, Phys. Rev. D 46, 5267 (1992); 47 4802 (1993).
[11] A. Higuchi, G. E. A. Matsas and C. B. Peres, Phys. Rev. D 48, 3731 (1993).
[12] D. Han, Y. S. Kim and M. E. Noz, Am. J. Phys. 67, 61 (1999).
[13] F. Reif, Fundamentals of Statistical and Thermal Physics (McGraw-Hill, New York, 1965) Chapter 15.
[14] A. O. Caldeira and A. J. Leggett, Physica A 121, 587 (1983).
[15] S. John and T. Quang, Phys. Rev. A 50, 1764 (1994).
[16] J. J. Bisognano and E. H. Wichmann, J. Math. Phys. 16, 985 (1975); 17, 303 (1976).
[17] R. Zwanzig Nonequilibrium Statistical Mechanics (Oxford University Press, London, 2001)
[18] D. J. Raine, D. W. Sciama, and P. G. Grove, Proc. R. Soc. Lond. A435, 205 (1991).
[19] P. Candelas, Phys. Rev. D 21, 2185 (1980).
[20] By definition, Eq.(58) in [4] with κ′ = κ is actually the response function R(−κ)/(2π)2 with η ≡ 2πδ(0).