Classical-quantum scattering

Daniel Carney1, ∗ and Akira Matsumura2, †

1
Physics Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
2
Department of Physics, Kyushu University, Fukuoka, 819-0395, Japan
(Dated: December 9, 2024)
arXiv:2412.04839v1 [hep-th] 6 Dec 2024

Abstract
We analyze the framework recently proposed by Oppenheim et al. to model relativistic quantum
fields coupled to relativistic, classical, stochastic fields (in particular, as a model of quantum matter
coupled to “classical gravity”). Perhaps surprisingly, we find that we can define and calculate
scattering probabilities which are Lorentz-covariant and conserve total probability, at least at tree
level. As a concrete example, we analyze 2 → 2 scattering of quantum matter mediated by a classical
Yukawa field. Mapping this to a gravitational coupling in the non-relativistic limit, and assuming
that we can treat large objects as point masses, we find that the simplest possible “classical-
quantum” gravity theory constructed this way gives predictions for 2 → 2 gravitational scattering
which are inconsistent with simple observations of, e.g., spacecraft undergoing slingshot maneuvers.
We comment on lessons learned for attempts to couple quantum matter to “non-quantum” gravity,
or more generally, for attempts to couple relativistic quantum and classical systems.

∗
[email protected]
†
[email protected]

1
 CONTENTS

I. Introduction and motivation 2

II. Classical-quantum framework 4

A. Classical-quantum states 5
B. Time evolution of the full CQ state 6
C. Time evolution of the reduced quantum state 9

III. Classical-quantum scattering 10

A. Non-relativistic 2 → 2 Yukawa scattering 12
B. Relativistic 2 → 2 Yukawa scattering 19

IV. Comparison to astrophysical scattering 22

V. Comments and outlook 24

Acknowledgements 25

References 26

A. Classical equations of motion 29

B. Feynman-Vernon derivations 30

C. Lorentz invariance of scattering supermatrix 34

D. Conservation of probability 36

I. INTRODUCTION AND MOTIVATION

When the curvature of spacetime is small compared to the Planck scale R/Mpl2 ≪ 1,
gravity can be quantized perturbatively as an effective quantum field theory, with action
!
√ c1 c2
Z
S= Mpl2 4
d x −g R + Lmatter + 2 R2 + 2 Rµν Rµν + · · · (1)
Mpl Mpl

where the dots represent an infinite series in powers of curvature in Planck units [1–3].
The theory is non-renormalizable: when R/Mpl2 & 1, this infinite series of terms becomes
dominant over the Einstein-Hilbert term R. Because the coefficients ci are unknown, the
theory is non-predictive in this regime. However, in a laboratory experiment involving solid
objects, R/Mpl2 ≈ 10−91 , so only the Einstein-Hilbert term contributes meaningfully. Thus,
this effective quantum field theory treatment of the weak gravitational field is extremely

2
well-defined in settings other than near a singularity, such as the deep interior of a black
hole or the very early universe.
While quantizing gravity this way is theoretically self-consistent at ordinary scales, it
is interesting to consider the possibility that gravity is not actually quantized this way in
nature. Investigating this notion is particularly motivated by the possibility of experimen-
tal tests that could distinguish standard quantized gravity from alternative models [4–15].
Recently, a number of models in which the gravitational field is “classical” have been con-
structed to make this suggestion more concrete [16–20]. Generally, these models have been
in the non-relativistic regime, and “classical” means that the Newtonian interaction cannot
generate entanglement between two massive systems. This would be in stark contrast to
standard quantized gravity as in Eq. (1), which definitely predicts entanglement genera-
tion through virtual graviton exchange, or equivalently, through the Newtonian two-body
operator obtained when integrating out the gravitons in the low-velocity limit [21].
Very recently, Oppenheim and collaborators have suggested an interesting relativistic
generalization of these “classical” gravity models [22–26]. They refer to their general setting
as “classical-quantum” (CQ) gravity and we will follow this terminology. In their CQ model,
the gravitational field is a bone fide classical field. It undergoes stochastic time evolution
and is coupled to quantum matter in a way made precise below. It is very interesting to
determine if this kind of CQ framework is mathematically self-consistent, and if it can be
used to formulate a physically viable model of gravity. Our purpose in this paper is to make
some steps towards answering these questions.
The general CQ framework is independent of gravity and we choose to specialize to the
simpler case of a Yukawa model, where a classical Yukawa field φ couples to some quantized
matter, which avoids any complications coming from gauge symmetry. On the other hand,
for non-relativistic questions, a Yukawa interaction in the massless limit generates a 1/r
potential, so it can be used to analyze the Newtonian limit of gravity by identifying the
coupling constants appropriately.
As a straightforward observable, we attempt to formulate scattering theory for quantum
matter coupled through this CQ Yukawa interaction. This proves to be instructive in many
ways. For one thing, it raises some basic concerns about the purely diffusive evolution law
assumed for the classical field [23, 25], which makes it hard to even formulate the scattering
problem because the theory does not have a stable ground state.1 Nevertheless, allowing for
some choices and approximations which we make clear, we formulate the 2 → 2 scattering
problem for two massive bodies. The CQ framework has two free functions, called D0 and D2 ,
and we analyze only the simplest case where these are assumed to be constants. We find that
1
Penington [27] has estimated that, in the context of gravity, the stochastic generation of gravitational
waves from this diffusive evolution is quantitatively inconsistent with non-observation of a stochastic
gravitational wave background in LIGO [28].

3
noise effects in the resulting CQ model produce large deviations in the scattering distributions
of even non-relativistic particles. Once the gravitational parameters are inserted, the CQ
model predicts effects which are in direct conflict with observations. We discuss the generality
of this conclusion in Sec. V.
Beyond the specifics of these gravity models, it is of substantial interest to understand if
classical systems can be coupled to quantum ones in some way compatible with special rela-
tivity, let alone general relativity [29–32]. One might hope that the quantitative difficulties
found here come more from specific choices in the construction rather than the overall idea
of a relativistic framework for coupling classical systems to quantum ones. Conversely, this
construction appears perhaps more robust than one might have anticipated. For example, it
appears capable of producing Lorentz-covariant scattering probabilities, at least at leading
order in a perturbative expansion. We end with a few comments on this, with the goal of
deriving some general lessons for possible relativistic classical-quantum evolution laws.

II. CLASSICAL-QUANTUM FRAMEWORK

To motivate the general idea of coupling quantum systems to classical ones, and to high-
light some of the features that arise in the model of Oppenheim et al., consider a quan-
tum computer operating under quantum error correction. This fundamentally consists of
quantum systems (qubits) which are periodically being measured (to determine the error
syndromes). These measurements are done by a classical apparatus: some readout system
which ultimately takes qubit states and records measurement outcomes as classical data.
This classical data is then used to determine the appropriate feedback unitary that needs to
be applied to the qubits. This feedback unitary is also implemented by an effectively classical
system, like a control laser or microwave drive. Thus, the qubits and classical control system
together constitute a dynamically coupled quantum-classical system.
This picture has already been used to construct models of non-relativistic matter coupled
through “semi-classical” gravitational interactions [16]. See. Fig. 1. These models have the
property that they induce noise on the quantum system, coming from a combination of the
measurements being done (which have random outcomes as usual in quantum mechanics)
and the feedback (which has noise due to the random nature of the control data). In these
models, however, the classical system is implicit; it is taken for granted that something is
doing the classical feedback control, but this is not invoked as an explicit degree of freedom.
Oppenheim et al.’s proposal is essentially to extend this paradigm by introducing a clas-
sical field (e.g., the gravitational field) as an explicit dynamical system which performs the
readout and control. This field has noise for the same reasons just discussed. In the next
subsections, we briefly review this construction.

4
 |ψ1 i UFB
Uent
|0i

|0i
Uent
|ψ2 i UFB

FIG. 1. Measurement-and-feedback gravity [16, 20]. Shown here is one timestep in a Markovian
process describing the gravitational interaction of two massive systems i = 1, 2. Each system is
coupled to an ancilla, which performs a position measurement to obtain an estimate of the two
positions hxi i. This information is then fed back through a one-body operator UFB = eiVsc ∆t ,
with Vsc = GN m1 m2 / |xi − hxj i|. This generates a noisy open channel on the two masses which
reproduces semiclassical Newtonian gravity. Figure reproduced from [4].

A. Classical-quantum states

A quantum system is generally described by a density matrix, while a classical stochastic

system is described by a classical probability distribution. Let q denote the canonical position
variables of a quantum system, and Q, P some complete set of canonical classical variables
on a classical system. In practice these will be position and momentum type variables as
indicated by the q, Q, P notation. We are using one symbol for each but the generalization
to many degrees of freedom (e.g., q → q1 , q2 , . . .) is completely straightforward. The basic
state variable in CQ theory is an object encoding the state of both of these:
Z
̺(Q, P ) := dqdq ̺(q, q; Q, P ) |qi hq| . (2)

We are writing this as an density matrix-valued function of the classical variables.

The way to interpret this is as follows. For a fixed classical configuration (Q, P ), this
object is the density matrix of the quantum system, conditioned on that classical configura-
tion. If we are not measuring the classical system at all, we can also find the reduced state
after averaging over the classical system,
Z
ρ := dQdP ̺(Q, P ). (3)

Conversely, the state (probability distribution) of the classical variables can similarly be
considered as either conditioned on or averaged over the quantum system. The probability
distribution for the classical variables, conditioned on the quantum system being in state

5
|qi, is
P(Q, P |qi ) = ̺(q, q; Q, P ). (4)

If we are not measuring the quantum system, the reduced state of the classical variable is
just this with an average (trace) over the quantum state:
Z
P(Q, P ) = dq ̺(q, q; Q, P ). (5)

In what follows we will be particularly interested in the reduced state of just the quantum
matter, with the classical degrees of freedom traced out, as in Eq. (3).

B. Time evolution of the full CQ state

The CQ evolution law is defined by a path integral of the following form:

Z
̺(qf , q f , Qf , Pf , tf ) = DqDqDQDP eICQ [q,q,Q,P ]̺(q0 , q0 , Q0 , P0 , t0 ), (6)

with the boundary conditions set as usual by the arguments on the left-hand side. We will
give the form of ICQ shortly. First let us motivate it by considering, separately, the path
integral for a quantum system and that for a stochastic classical system.
First consider a quantum system. The density matrix obeys

ρ̇ = −i[H, ρ]. (7)

Assuming H is at most quadratic in momenta, this equation famously admits a path integral
solution (a particular “Schwinger-Keldysh” path integral)
Z
ρ(qf , qf , tf ) = DqDqei(S[q]−S[q]) ρ(q0 , q0 , t0 ). (8)

Here, the first exponential evolves the ket, while the second evolves the bra, in the sense of
ρ → UρU † . The action S is defined as usual. The generalization beyond theories quadratic
in momenta is standard, but we will not need it in what follows.
Now consider a classical system undergoing stochastic evolution. As a concrete example,
consider a particle satisfying the simple Fokker-Planck equation

P ∂ D2 ∂ 2
Ṗ = − P+ P, (9)
m ∂Q 2 ∂P 2

where P = P(Q, P ) is the probability distribution on (Q, P ). This equation describes pure

6
diffusion in the position variable Q, where D2 is a coefficient with units of M 3 that sets the
rate of diffusion. This equation similarly admits a path integral solution [33],

P − D1 R dt [mQ̈]2
Z  
P(Qf , Pf , tf ) = N DQDP δ Q̇ − e 2 P(Q0 , P0 , t0 ), (10)
m

where N is a normalization factor. A good way to understand what this path integral does
is as follows.2 The Fokker-Planck equation can equivalently be described by the Langevin
equations of motion for a particle being stochastically kicked,
√
P p D2
Q̇ = , Ṗ = D2 ξ =⇒ Q̈ = ξ, (11)
m m

where ξ = ξ(t) is a Gaussian-distributed random white noise variable

R
dt ξ 2 (t)
P[ξ(t)] = Ne− . (12)

Thus, the probability distribution of a given particle configuration Qf at tf can be obtained

√
by just solving ξ = mQ̈/ D2 and inserting it into this distribution, yielding
Z R 2
− D1 dt [mQ̈]
P (Qf , tf ) = N DQe 2 P (Q0 , t0 ). (13)

Notice that this path integral is in real (Lorentzian) time. It is very different from the
usual quantum form: the equations of motion mQ̈ = 0 (averaged over the Gaussian noise
hξi = 0) appear explicitly in the integral weight. In this sense the path integral represents
time evolution on the space of configurations, with the dominant contribution from the ξ = 0
solution, and contributions away from this average solution distributed according to the noise
ξ 6= 0. Note that this path integral is quartic in derivatives, but unlike a quantum-mechanical
path integral this does not imply anything about acausal behavior.

Putting these ideas together, Oppenheim et al. define a total CQ path integral by evolving
the quantum parts with a Schwinger-Keldysh path integral and the classical part with a
Fokker-Planck path integral. They then make an ansatz for coupling terms. We refer the
reader to [22–26] for the general construction. Here we will consider quantum matter, which
we model as a pair of distinguishable scalar fields χ1,2 , with Yukawa couplings to a classical

2
We thank Shivaji Sondhi for explaining this to us.

7
scalar field φ. The CQ action for this model is

ICQ [χ, χ, φ, π] = i(S0 [χ] + Sint [χ, φ] − S0 [χ] − Sint [χ, φ]) + ln(δ[φ̇ − π])
2
D0 tf 4 δSint

δSint
Z
− dx [χ, φ] − [χ, φ]
2 t0 δφ δφ (14)
Z tf  2
1 4 2 2 1 δSint 1 δSint
− d x π̇ − (∇ − mφ )φ − [χ, φ] − [χ, φ] .
2D2 t0 2 δφ 2 δφ

Here χ = (χ1 , χ2 ) is a multiplet of the two scalars, and π is the canonical momentum for
the classical φ field. The log term enforces the canonical variable delta function like that
appearing in Eq. (10).
Let us describe the role of each term in Eq. (14). In the first line, S0 is just the usual
kinetic term for the quantum scalar fields. The term
Z
Sint = λ d4 x φχ2 (15)

is an ordinary Yukawa coupling. Since we are using scalar fields, λ has dimensions of mass.
The second line, which involves
δSint
= λχ2 (16)
δφ
generates decoherence of the quantum matter. Specifically, in the field configuration basis
|χi, this term causes off-diagonal terms in the density matrix (i.e., the elements |χi hχ| with
χ 6= χ) to decay at a rate proportional to D0 λ2 . Here D0 is called the decoherence parameter
and in this model has units of D0 ∼ 1/M 2 .
Finally, the third line generates the diffusive, stochastic evolution of the classical field
φ. The overall coefficient D2 , called the diffusion parameter, has units of D2 ∼ M 2 in
this model. Just like the Brownian particle example above, the diffusion is such that the
most likely path satisfies a classical equation of motion without noise, and then there are
fluctuations around this. The highly novel part is that the classical φ equations of motion
are sourced by the quantum field, through the last terms. Essentially, the classical fields
become sourced by expectation values of the quantum fields. In this example, we have
p
hπ̇i = [∇2 − m2φ ] hφi + λ hχ2 i + D2 /2 hξi. We explain this in detail in Appendix A. What
this equation shows is that the classical fields are sourced by two systems: the quantum field,
and some implicit external driving field which exerts a white noise force.
The construction given above is supposed to have three key features [22–26]. The obvious
one is that it gives a coupling of quantum matter to classical matter. Another is that the
time evolution generated by this path integral maps density matrices to density matrices
and in particular preserves total probability for the whole system. For this to hold, one

8
requires [22–26].
D0 D2 ≥ 1, (17)

which Oppenheim et al. refer to as a “decoherence-diffusion tradeoff”. Finally, this path

integral is supposed to be Lorentz covariant, and in the gravitational context, can be made
generally covariant. In general, D0,2 = D0,2 (q, Q, P ) are allowed to be functions of the
canonical path integral variables. In this paper, we are only analyzing the simplest case
where these are taken to be simple constants.
The meaning of Lorentz covariance in this framework deserves some discussion. It is clear
that the action here is Lorentz invariant in the simple sense of having no free spacetime
indices. What is less clear is how to define Lorentz or general coordinate transformations on
the CQ states. Since Oppenheim et al. have not given a prescription for this, we will not
generally attempt it here. However, in the context of scattering theory in Sec. III, we will
define Lorentz transformation properties of the asymptotic states for the quantum matter
using the standard rules of quantum field theory. Interestingly, we will find that the scattering
probabilities constructed with these states are Lorentz invariant in the sense of being given
by Lorentz-invariant functions. However, there is an important twist: the probabilities
violate energy-momentum conservation. These two facts are consistent because this model
is fundamentally an open quantum system, and thus there is no Noether’s theorem [32, 34].
Concretely, energy and momentum can be injected both by the classical field, and by the
white noise ξ driving the classical field.

C. Time evolution of the reduced quantum state

The CQ path integral generates time-local dynamics on the joint classical and quantum
system. However, if we are only interested in the dynamics of the quantum system, we can
obtain a non-local path integral by simply doing the integration over the classical system.
The result is the standard Feynman-Vernon path integral, representing an open quantum
system [35, 36]. This will be particularly useful for us in order to study scattering of quantum
matter, because the path integral constructed this way averages over the final configuration
of the classical field, which we will assume is unmeasured in the examples below.
More concretely, assume that the joint CQ state at a time t0 has a product form
̺(q0 , q0 , Q0 , P0 , t0 ) = ρ(q0 , q0 , t0 )P(Q0 , P0 , t0 ). The time evolution of the quantum system at
a later time tf is given by the Feynman-Vernon path integral
Z
ρ(qf , qf , tf ) = DqDq eiS[q]−iS0[q]+iSIF [q,q] ρ(q0 , q0 , t0 ) (18)

9
 |βi hβ|

··· ··· ··· ··· ··· ···

time

= × +

··· ··· ··· ··· ··· ···

|αi hα|

FIG. 2. Schwinger-Keldysh-Feynman-Vernon diagrams for CQ theories. Here we show a general

process with many incoming particles in both the ket |αi and bra |αi variables. Physical time runs
from bottom to top and the arrows represent phase time evolution as either forward or “backward”
(i.e., like e+iHt acting on the bra). On the right hand side, the first diagram represents time
evolution in a closed quantum system: the ket and bra lines talk to themselves but not to each
other. The last term is unique to open systems like the CQ model studied here, which allows for
correlations to develop between the ket and bra variables.

with the influence functional SIF [q, q] defined through the following equation,
Z
iS0 [q]−iS0 [q]+iSIF [q,q]
e = DQDP eICQ [q,q,Q,P ]P(Q0 , P0 , t0 ). (19)

Here S0 [q] is again the free action of quantum system (which in a more general model would
include any self-interactions). The influence functional SIF [q, q] contains all effects from the
classical system coupled with the quantum system of interest. We give explicit computations
for SIF in our Yukawa model in Appendix B.

III. CLASSICAL-QUANTUM SCATTERING

Scattering theory in quantum mechanics and quantum field theory is defined in terms of
states of well-separated particles freely propagating in from t → −∞, interacting near t = 0,
and then evolving into (superpositions of) states of well-separated particles propagating
freely out to t → +∞. We will attempt to do the same in our CQ Yukawa model, but this
will require a few assumptions and modifications to the usual quantum treatment.
In unitary quantum mechanics, one usually calculates a scattering matrix Sα→β =
hβ|U|αi, where U generates time evolution from past to future infinity. Here α = p1 , σ1 , p2 , σ2 , . . .
are the quantum numbers of the initial state, and β = p′1 , σ1′ , p′2 , σ2′ , . . . are the quantum
numbers of the final state, in the notation of [37, 38]. The probability of transition |αi → |βi

10
is then given by
P (α → β) ∼ |Sα→β |2 , (20)

where the overall proportionality involves correctly normalizing the states to a flux of in-
coming particles, as we discuss below.
In CQ theory, there are two twists to this story. One is that the asymptotic states now also
need to include the probability distribution of the classical field at past and future infinity.
The other is that the model is fundamentally an open system, and thus there is no unitary
S-matrix. Instead, one has to work directly with probabilities. To circumvent the first issue,
here we are going to just assume a certain initial state for the classical field—namely, the
distribution with the field completely turned off P[φ, π] = δ[φ]δ[π]—and then average over
the final φ states by using a Feynman-Vernon path integral (see Sec. II C). This will be used
to compute what we will call a scattering supermatrix element

/ αα→ββ = hβ|M [|αi hα|] |βi ,

M (21)

where M is the channel on the quantum system generated by the Feynman-Vernon path
/ one obtains the probability for a transition |αi → |βi by
integral. From M

/ αα→ββ .
P (α → β) ∼ M (22)

We will give a much clearer and more explicit formula below in the examples. The pro-
portionality factor again has to do with normalization to a flux of incoming particles. The
slashed notation is in homage to Hawking’s infamous “dollar matrix” [39]. The M / elements
can be computed by a simple set of Feynman rules, which we depict schematically in Fig. 2.
A remarkable result in our explicit calculations is that the scattering supermatrix elements
defined by Eq. (21) transform covariantly under the Poincaré group. We give some more
precise statements to this effect below and in Appendix C. In particular, note that we are
computing transition elements only between states of the quantum matter, with the φ field
averaged over. It is not at all obvious that this result should be covariant — for example,
one might have imagined that the classical field evolved into some state which spontaneously
broke Lorentz invariance by setting a direction. As a sanity check, we show in Appendix
C that an analogous statement holds if we treat φ as a quantum field, initialize it to the
vacuum |0i, and trace over the final states. Even allowing for processes where φ quanta are
emitted into the final state, we find that the trace produces a channel on the χ fields which
is Lorentz covariant. The key seems to be that the initial state is Lorentz invariant.
One major conceptual difference between CQ theory and ordinary quantum scattering is
that the asymptotic states in CQ theory are necessarily unstable. There are two reasons for

11
this. One is that there is no mass gap for the classical field, even if it has a non-zero mass,
because one can always find solutions to the wave equation with arbitrarily small energy.
The other is that the classical field is being continuously stochastically driven. In particular,
this means that the initial state we will use, P[φ, π] = δ[φ]δ[π], is completely unstable.
In fact, at sufficiently long times, any initial classical field configuration will asymptote
to P[φ, π] ≡ 1, because the evolution law is purely diffusive with no friction. This is a
highly pathological state, especially once extrapolated to gravity. It would represent a state
where the universe is equally likely to be in flat spacetime or a highly chaotic assortment of
gravitational excitations. In particular, the CQ model does not have a vacuum state for the
classical fields. Quite similarly, momentum eigenstates for the quantum particles are also
unstable, because the D0 term is trying to decohere them into the field configuration basis.
One could attempt to deal with these problems by adiabatically switching off the D0,2
terms. However, perhaps a more straightforward interpretation is to recognize that a scat-
tering description in this model only makes sense for finite times T such that the D0,2 effects
are small. In particular, we are going to treat this model in perturbation theory, so T D0
and T D2 cannot be too large anyway. Thus in what follows we will regulate everything with
an overall time duration T < ∞. This is similar to the use of scattering theory to compute
transitions and decay rates with quasistable particles in quantum field theory, for example
muon scattering.
With these assumptions and caveats in mind, we begin by calculating 2 → 2 scattering
of the χ particles in the non-relativistic limit. This is an instructive warmup to show how
standard path integral scattering calculations generalize to this setting. We then move on
to a relativistic calculation to analyze the Lorentz invariance of the scattering process.

A. Non-relativistic 2 → 2 Yukawa scattering

To begin, we will analyze χ1 χ2 → χ1 χ2 scattering by starting directly in the non-

relativistic limit for the massive particles. That is, we assume that pi /mi ≪ 1 and work
with single-particle states only. We continue to use a continuum description of the Yukawa
field φ. The interaction takes the form

λ X
Z
Vint = d3 x φ(x)δ 3 (x − xi ), (23)
m i=1,2

using standard matching methods. The free Lagrangian for the particles is L0 [x1 , x2 ] =
m i ẋ2i /2. Our main goal here is to show how one can apply elementary scattering methods
P

to the CQ framework. We will extend the beautiful treatment of non-relativistic quantum

scattering given in [40].

12
 As discussed in Sec. II C, the reduced dynamics of the particles can be calculated by first
performing the path integral integral over φ. This produces a Feynman-Vernon type path
integral for the quantum particles,
Z
ρ(x1,f , x2,f , x1,f , x2,f , tf ) = Dx1 Dx2 Dx1 Dx2 ei(S0 [x1 ,x2 ]−S0 [x1 ,x2 ]) eiSIF [x1 ,x2 ,x1 ,x2 ]
(24)
× ρ(x1,0 , x2,0 , x1,0 , x2,0 t0 ).

In this simple example, we have

λ2
Z X
iSIF =− dt F1 (x1 − x2 ) + F1∗ (x1 − x2 ) − F2 (xi − xj ) (25)
m1 m2 i6=j

and the boundary values of all the integrals are taken at tf and t0 , which will later become
±T /2. The functions here are

i
F1 (x1 − x2 ) = − [GR (x1 − x2 ) + GR (x2 − x1 )] + D2 GC (x1 − x2 ) + D0 δ 3 (x1 − x2 )
2 (26)
F2 (x1 − x2 ) = D2 GC (x1 − x2 ) + D0 δ 3 (x1 − x2 )

with

d3 k eik·x d3 k eik·x d3 k ik·x

Z Z Z
GR (x) = , GC (x) = , δ 3 (x) = e . (27)
(2π)3 k2 + m2φ 3 2
(2π) (k + mφ ) 2 2
(2π)3

The subscript R on the first Green’s function stands for retarded; this notation is just an
artifact of the full relativistic calculation shown in the next section. The subscript C on the
second Green’s function stands for classical; it comes from the Fokker-Planck path integral
over φ, which is quartic in derivatives as discussed above.

The expression in Eq. (25) is a highly simplified form of the fully relativistic expression.
We give a detailed derivation of SIF in Appendix B. To go from the relativistic case to Eq.
(25), we are assuming that all momenta are non-relativistic, and dropping all self-interaction
terms. This latter assumption is particularly helpful, although to justify it would require a
detailed study of renormalization of propagators in this theory, which goes beyond the scope
of what we are trying to do here. We make some further comments in Appendix B.

While the full relativistic Feynman-Vernon functional is non-local in time, our non-
relativistic limit Eq. (24) is local in time. Thus it can be used to do scattering calculations
in nearly the exact same fashion as ordinary non-relativistic quantum mechanics. The only
R
new things are that the “potential” defined by iSIF = dt VIF (t) is complex and it can
connect bra and ket variables.

13
 p′1 p′2 p′1 p′2

p1 p2 p1 p2
=
( )( )( ) × + × + + ···

FIG. 3. Schwinger-Keldysh-Feynman-Vernon diagrams for 2 → 2 scattering in CQ Yukawa theory,

in the non-relativistic limit, to O(λ2 ). The dotted line indicates interaction through the Yukawa
field. The first term on the right hand side represents no interaction at all, i.e., the λ0 term. The
second term gives an example of ket-ket interaction (∼ MF1 ). The third gives an example gives
an example of ket-bra interaction (∼ MF2 ). The dots represent the other permutations and higher
order terms.

Consider 2 → 2 scattering. We let pi , pi denote the initial momenta in the ket and
bra respectively, and put primes for final state momenta. The full scattering supermatrix
element is computed by evolving an initial set of plane wave states

Y ei(p·x−Ep t)
ρ(x1 , x2 , x1 , x2 , t0 ) = ψpi (xi,0 , t0 )ψp∗ (xj,0, t0 ), ψp (x, t) = (28)
i,j=1,2
j (2π)3/2

and then projecting the result onto a similar set of plane waves at tf . Concretely,
Z
/ p1 p2 p
M p →p′1 p′2 p′1 p′2 = dx1,f dx2,f dx1,f dx2,f dx1,0 dx2,0 dx1,0 dx2,0
1 2

× ψp∗ ′1 (x1,f , tf )ψp∗ ′2 (x2,f , tf )ψp′1 (x1,f , tf )ψp′2 (x2,f , tf )

(29)
/ 1,f , x2,f , x1,f , x2,f , tf |x1,0 , x2,0 , x1,0 , x2,0 , t0 )
× K(x
× ψp1 (x1,0 , t0 )ψp2 (x2,0 , t0 )ψp∗ (x1,0 , t0 )ψp∗ (x2,0 , t0 ).
1 2

/ from t0 to tf is given by
Here, the density matrix propagator K

/ 1,f , x2,f , x1,f , x2,f , tf |x1,0 , x2,0 , x1,0 , x2,0 , t0 )

K(x
Z (30)
= Dx1 Dx2 Dx1 Dx2 ei(S0 [x1 ,x2 ]−S0 [x1 ,x2 ]) eiSIF [x1 ,x2 ,x1 ,x2 ] ,

with the limits on the path integral given by the argument of K./ To compute this perturba-
iSIF
tively, we expand e as a power series in λ/m, which gives an expression of the schematic

14
form Z
/ f |t0 ) = K
K(t / 0 (tf |t0 ) + / 0 (tf |t)VIF (t)K
dt K / 0 (t|t0 ) + · · · (31)

/ 0 is the free propagator.

where K

The free propagator is standard: it is just a product over each particle,

Y
/ 0 (x1,f , x2,f , x1,f , x2,f , tf |x1,0 , x2,0 , x1,0 , x2,0 , t0 ) =
K K0 (xi,f , tf |xi,0 , t0 )K0∗ (xj,f , tf |xj,0 , t0 ),
i,j

(32)

where K0 is the ordinary propagator for a single non-relativistic particle in quantum me-
/ 2 of order λ2 /(m1 m2 ) is more
chanics (see [40] for the explicit expression). The term K
complicated:

λ2
Z
/2 =−
K dtdy1 dy2 dy1 dy2 K / 0 (x1,f , x2,f , x1,f , x2,f , tf |y1 , y2 , y , y , t)
m m 1 2
" 1 2 #
(33)
X
× F1 (y1 − y2 ) + F1∗ (y1 − y2 ) − F2 (yi − yj )
i6=j

/ 0 (y1 , y2 , y , y , t|x1,0 , x2,0 , x1,0 , x2,0 , t0 ).

×K 1 2

Here the y are the variables at the time t where the interaction occurs. While this equation
is long, it is easy to understand in terms of Feynman-type diagrams. See Fig. 3. To leading
order (O(λ2 )) in perturbation theory, we have a simple factorization of the supermatrix
elements:

/ p1 p2 p
M p →p′1 p′2 p′1 p′2 = δ 3 (p1 − p′1 )δ 3 (p2 − p′2 )δ 3 (p1 − p′1 )δ 3 (p2 − p′2 )
1 2

+ MF1 (p1 p2 → p′1 p′2 )δ 3 (p1 − p′1 )δ 3 (p2 − p′2 ) + δ 3 (p1 − p′1 )δ 3 (p2 − p′2 )MF1∗ (p1 p2 → p′1 p′2 )
+ δ 3 (p2 − p′2 )MF2 (p1 p2 → p′1 p′2 )δ 3 (p1 − p′1 ) + δ 3 (p1 − p′1 )MF2 (p2 p1 → p′2 p′1 )δ 3 (p2 − p′2 ).
(34)

The term in the first line represents no interaction at all. The following terms contain the
effects of either the F1 interaction which connects ket-ket or bra-bra, and the F2 interaction
which connects kets to bras. We now turn to explicit calculation of these terms.

First consider the term with F1 . This clearly factors in the sense that it affects only
the ket, not the bra. Inserting it into Eq. (29), we have a contribution to the scattering

15
supermatrix

λ2
Z
MF1 (p1 p2 → p′1 p′2 ) =− dtdy1 dy2 dx1,f dx2,f dx1,0 dx2,0 ψp∗ ′1 (x1,f , tf )ψp∗ ′2 (x2,f , tf )
m1 m2
× K0 (x1,f , x2,f , tf |y1 , y2 , t)F1 (y1 − y2 )K0 (y1 , y2 , t|x1,0 , x2,0 , t0 )ψp1 (x1,0 , t0 )ψp2 (x2,0 , t0 ).
(35)

Performing the initial and final state integrals moves the exponentials to the interaction
vertex at t,

λ2
Z
MF1 (p1 p2 → p′1 p′2 ) =− dtdy1 dy2 F1 (y1 − y2 )ψp∗ ′1 (y1 , t)ψp∗ ′2 (y2 , t)ψp1 (y1 , t)ψp2 (y2 , t),
m1 m2
(36)

where the free propagator gave the free evolution of the plane waves,
Z
ψp1 (y1 , t)ψp2 (y2 , t) = dx1,0 dx2,0 K0 (y1 , y2 , t|x1,0 , x2,0 , t0 )ψp1 (x1,0 , t0 )ψp2 (x2,0 , t0 )
Z (37)
ψp∗ ′1 (y1 , t)ψp∗ ′2 (y2 , t) = dx1,f dx2,f ψp∗ ′1 (x1,f , tf )ψp∗ ′2 (x2,f , tf )K0 (x1,f , x2,f , tf |y1 , y2 , t).

In Eq. (36), the integral over dt produces (2π)δ(Ep′1 + Ep′2 − Ep1 − Ep2 ), i.e., total energy
conservation. Now we insert the expressions in Eq. (26) for F1 . Every term in F1 can be
expressed as an integral over d3 k with a factor e±ik·(y1 −y2 ) . The integrals over the y variables
then produce a pair of momentum delta functions (2π)6 δ 3 (p′1 − p1 ± k)δ 3 (p′2 − p2 ∓ k).
Writing this as an overall momentum conservation delta function, the k integral can now be
performed by setting k = p′1 − p1 . This produces the simple answer
" #
λ2 i D2
MF1 (p1 p2 → p′1 p′2 ) = −δ 4 (p1 + p2 − p′1 − p′2 ) + 2 + D0 .
(2π)2 m1 m2 k + mφ (k + m2φ )2
2 2
k=p′1 −p1
(38)

The first term is ordinary scattering in quantum mechanics. The next two terms are open
system corrections and take a specific form in CQ theory. We have written the total energy-
momentum delta function in a Lorentz-invariant notation, but it should be remembered that
in this limit the energies do not actually transform correctly to make four-vectors. The term
with F1∗ is clearly the same except we replace the ket variables with bra variables.

Now we analyze the F2 terms, which mix the bra and ket variables. This no longer
factorizes between the bra and ket. However, one particle in the ket and one in the bra will

16
still evolve trivially. Using the same manipulations as above that led to Eq. (36), we obtain

λ2
Z
MF2 (pi pj → p′i p′j ) =− dtdyi dyj F2 (yi − yj )ψp∗ ′i (yi , t)ψp′j (yj , t)ψpi (yi , t)ψp∗ (yj , t).
m1 m2 j

(39)

There is a critical difference between this and Eq. (36): the phases on the bra variables
are flipped. When we perform the integral over the vertex variables, this now produces a
different set of delta functions: a momentum conservation factor ∼ δ 3 (p′i − pi − (p′j − pj ))
and an energy conservation factor ∼ δ(Ep′i − Epi − (Ep′j − Epj )). In total, we get
" #
2
λ D 2
MF2 (pi pj → p′i p′j ) = −δ 4 (p′i − pi − (p′j − pj )) + D0 . (40)
(2π) m1 m2 (k + m2φ )2
2 2
k=p′i −pi

We have written the arguments of the delta functions in a suggestive order: MF2 represents
a process where the scalar φ imparts an equal kick k = ∆p to both the ket and the bra
particle.

Finally, we can use these results to calculate the actual transition probabilities, using
Eqs. (22) and (34). To do this we are going to look at diagonal elements pi = pi and
p′i = p′i . To order λ2 , the only non-trivial process is forward scattering pi = p′i ; anything
else vanishes due to the various delta functions. As a non-trivial check on the CQ framework,
we show in Appendix D that these forward scattering terms sum appropriately to conserve
total probability (i.e., the open systems version of the optical theorem holds). However, to
get a differential cross section for non-trivial scattering angles, we are therefore going to need
to go to order λ4 .

The λ2 terms in MF1 and MF1∗ multiply to produce non-trivial scattering at this order,
namely
M/ p1 p2 p1 p2 →p′ p′ p′ p′ ⊃ |MF1 (p1 p2 → p′1 p′2 )|2 , (41)
1 2 1 2

with MF1 given in Eq. (38). Other than the CQ corrections, this works like ordinary scatter-
ing. There is also a contribution to forward scattering at this order, where the evolution acts
to order λ4 but only on the ket, or only on the bra — i.e., the one-loop forward scattering
contribution, which we will ignore following the above discussion.

What about the F2 terms? To order λ4 , we can have two kinds of contributions. One is
the process where |p1 i interacts with hp2 | and |p2 i interacts with hp1 |, which is a product
of two tree diagrams. The other possibility is a loop diagram, where e.g. |p1 i interacts with
hp2 | twice. It is easy to check that the loop diagrams only contribute to forward scattering

17
again to this order, so we will continue to ignore them. Thus the F2 term contributes as

/ p1 p2 p1 p2 →p′ p′ p′ p′ ⊃ MF2 (p1 p2 → p′1 p′2 )MF2 (p2 p1 → p′2 p′1 ) = [MF2 (p1 p2 → p′1 p′2 )]2 ,
M 1 2 1 2

(42)

where in the last equality we used Eq. (40).

Putting the above results together we obtain the total probability. Both the F1 and F2
contributions contain squares of energy-momentum conservation delta functions. To regulate
this, we can do the usual process of converting one of the delta functions into a spacetime
volume V T /(2π)4 [37, 38]. Using this, we have, for non-forward scattering,

2 2
P (p1 p2 → p′1 p′2 ) = |MF1 (p1 p2 → p′1 p′2 )| + [MF2 (p1 p2 → p′1 p′2 )]
 !2 !2 
4
VT 4 ′ λ 1 D2
= δ (p1 + p′2 − p1 − p2 )  + + D0 
(2π)4 (2π)4 m21 m22 (p′1 − p1 )2 + m2φ [(p′1 − p1 )2 + m2φ ]2
!2
VT 4 ′ ′ λ4 D2
+ δ (p1 − p1 − (p2 − p2 )) + D0 .
(2π)4 (2π)4 m21 m22 [(p′1 − p1 )2 + m2φ ]2
(43)

We see that the two momentum transfers are the same p′1 − p1 , but the energy-momentum
delta functions are different. In particular, this is a Galilean-invariant expression, but in-
cludes non-trivial probabilities for processes which violate momentum conservation. As
mentioned above, this is consistent because this is calculated in an open quantum system,
and so there is no Noether theorem connecting conservation laws to symmetries. Note that
large violations of momentum conservation are suppressed like 1/∆p4 .
Finally, we can derive a differential cross section from Eq. (43) following standard meth-
ods [37]. Let u be the relative velocity of the initial state momenta, e.g., u = |p|/µ in
the center-of-momentum frame, where µ is the reduced mass. Then the differential rate of
particles scattered into the final states in a phase space volume d3 p′1 d3 p′2 is given by

dσ (2π)2 λ4
=
d3 p′1 d3 p′2 u (2π)4 m21 m22
(  !2 !2 
1 D2
× δ 4 (p′1 + p′2 − p1 − p2 )  + + D0 
(p′1 − p1 )2 + m2φ [(p′1 − p1 )2 + m2φ ]2
!2 )
D2
+ δ 4 (p′1 − p1 − (p′2 − p2 )) + D0 .
[(p1 − p1 )2 + m2φ ]2
′

(44)

18
In Sec. IV, we show how to evaluate this in a frame in an explicit example, the scattering
of a light particle off a heavy one.

The non-relativistic example here demonstrates the basic structure of scattering in CQ

theory. It is also sufficient to extract useful predictions, as discussed in Sec. IV. However, it
is of considerable formal interest to understand the full relativistic version of this calculation,
to which we now turn.

B. Relativistic 2 → 2 Yukawa scattering

Let us move on to the analysis of χ1 χ2 → χ1 χ2 scattering without taking the non-

relativistic limit. We use the relativistic interaction of Eq. (15) and consider the massive
scalar fields χ = (χ1 , χ2 ) in ordinary canonical quantization. We find the following Feynman-
Vernon type path integral for the fields,
Z
ρ(χ1,f , χ2,f , χ1,f , χ2,f , tf ) = Dχ1 Dχ2 Dχ1 Dχ2 ei(S0 [χ]−S0 [χ]) eiSIF [χ,χ] ρ(χ1,0 , χ2,0 , χ1,0 , χ2,0 , t0 ),
(45)

where S0 [χ] = − 12 d4 x 2
+ m2i χ2i ] is the free action of the fields and
R P
i [(∂µ χi )

Z X
2
iSIF = −λ d4 xd4 y F1 (x − y)χ21 (x)χ22 (y) + F1∗(x − y)χ21 (x)χ22 (y) − F2 (x − y)χ2i (x)χ2j (y).
i6=j
(46)
is the influence functional. Again see Appendix B for details. Note that unlike the non-
relativistic case, this is a non-local functional. Here, we again only wrote the cross terms
between the two fields, dropping the self-interaction terms. These only contribute to loops
in the forward scattering process and are beyond the scope of this paper. The F1,2 functions
here are

i
F1 (x − y) = − [GR (x − y) + GR (y − x)] + D2 GC (x − y) + D0 δ 4 (x − y), (47)
2
i
F2 (x − y) = [GR (x − y) − GR (y − x)] + D2 GC (x − y) + D0 δ 4 (x − y) (48)
2

19
with the relativistic forms

d4 k eik·x
Z
GR (x) = ,
(2π)4 −(k 0 + iǫ)2 + k2 + m2φ
d4 k eik·x
Z
GC (x) = , (49)
(2π)4 [−(k 0 + iǫ)2 + k2 + m2φ ][−(k 0 − iǫ)2 + k2 + m2φ ]
d4 k ik·x
Z
4
δ (x) = e .
(2π)4

The retarded Green’s function describes the exchange of energy and momentum between
χ1 and χ2 . One can reproduce the non-relativistic counterpart (25) by taking χ2i (x) →
1 3
m
δ (x − xi ) and ignoring k 0 ± iǫ in the functions GR and GC .
In this relativistic setting, the scattering supermatrix element is given by the path integral,
Z
/ p1 p2 p
M p →p′1 p′2 p′1 p′2 = Dχ1,f Dχ2,f Dχ1,f Dχ2,f Dχ1,0 Dχ2,0 Dχ1,0 Dχ2,0
1 2

× Ψ∗p′1 (χ1,f , tf )Ψ∗p′2 (χ2,f , tf )Ψp′1 (χ1,f , tf )Ψp′2 (χ2,f , tf )

(50)
/ 1,f , χ2,f , χ , χ , tf |χ1,0 , χ2,0 , χ , χ , t0 )
× K(χ 1,f 2,f 1,0 2,0

× Ψp1 (χ1,0 , t0 )Ψp2 (χ2,0 , t0 )Ψ∗p (χ1,0 , t0 )Ψ∗p (χ2,0 , t0 ),

1 2

p
where Ψpi (χi , t) = e−iωpi t hχi |pi i with ωpi = p2i + m2i is the wave functional of momen-
tum eigenstate |pi. Replacing the field χi and the relativistic energy ωpi with a particle
position xi and the non-relativistic one Epi = p2i /2mi , we have the plane wave ψpi (xi , t) =
e−iEpi t hxi |pi i = ei(pi ·xi −Epi t) /(2π)3/2 . Further adopting the density matrix propagator in
Eq. (30), we rederive the non-relativistic scattering supermatrix given in Eq. (29).
The density matrix propagator K / in the field configuration basis is
Z
/ 1,f , χ2,f , χ , χ , tf |χ1,0 , χ2,0 , χ , χ , t0 ) =
K(χ 1,f 2,f 1,0 2,0
Dχ1 Dχ2 Dχ1 Dχ2 ei(S0 [χ]−S0 [χ]) eiSIF [χ,χ]
(51)
/ As in Sec. III A,
with the boundary condition of the path integral given by the argument of K.
iS
/ by expanding e IF as a series in the
we can perturbatively evaluate the propagator K
coupling strength λ. The ingredients MF1 and MF2 needed for M / up to the second order of
λ are given by
Z
2
MF1 (p1 p2 → p′1 p′2 ) = −λ d4 y1 d4 y2 F1 (y1 − y2 )
Z
× dχ1 dχ2 χ21 (y1)χ22 (y2 )Ψ∗p′1 (χ1 , y10 )Ψ∗p′2 (χ2 , y20 )Ψp1 (χ1 , y10)Ψp2 (χ2 , y20),
(52)

20
and
Z
2
MF2 (pi pj → p′i p′j ) =λ d4 y1 d4 y2 F2 (y1 − y2 )
Z (53)
2 2 ∗ 0 0 0 ∗ 0
× dχi dχj χi (y1 )χj (y2 )Ψp′i (χi , y1 )Ψp′j (χj , y2 )Ψpi (χi , y1 )Ψp (χj , y2 ),
j

where the integrals of field variables χ1 and χ2 are performed on time slices y10 and y20 ,
respectively. The field integrals are evaluated as
′
e−i(p −p)·y
Z
2
dχ χ (y)Ψ∗p′ (χ, y 0)Ψp (χ, y 0) ′ 2
= hp |χ̂ (y)|pi = √ , (54)
(2π)3 ωp′ ωp

where p 6= p′ and χ̂(y) is the solution of free field. After the field integrals and the remaining
integrals of y1 and y2 , we have the relativistic version of MF1 and MF2 ,
" #
2 4 ′ ′
λ δ (p 1 + p 2 − p 1 − p 2 ) i D 2
MF1 (p1 p2 → p′1 p′2 ) = − √ + + D0 ,
(2π)2 ωp1 ωp2 ωp′1 ωp′2 k 2 + m2φ (k 2 + m2φ )2 ′
k=p1 −p1
2 4 ′ ′
" #
λ δ (pi − pi − (pj − pj )) D2
MF2 (pi pj → p′i p′j ) = + D0 .
2
(2π) ωpi ωpj ωpi ωp′j (k + m2φ )2
2
p ′
′ k=pi −pi
(55)

Here, we took the limit ǫ → 0 since there is no pole in the denominator because of (p′i −
pi )2 + m2φ > 0. In the non-relativistic limit, the above MF1 and MF2 are consistent with our
non-relativistic results, Eqs. (38) and (40).

At this stage, we can discuss the Lorentz invariance of the resulting scattering supermatrix
elements. In unitary S-matrix theory, scattering is described via scattering amplitudes,
which are covariant under Lorentz transformations, and appropriate combination of the
matrix elements with overall energy factors are Lorentz invariant functions of the external
momenta. We remind the reader of the standard results and notation in Appendix C. In the
CQ results shown here, it is easy to check that pi → Λpi and similarly the underlined (bra)
and primed (final state) variables, we have that

√
ωp1 ωp2 ωp′1 ωp′2 MF1 , ωpi ωpj ωp′i ωp′j MF2 (56)
p

are both invariant. These are just the straightforward analogues of the S-matrix element
transformations. In particular, this means that the full supermatrix element is Lorentz
invariant. In this sense, the χ1 χ2 → χ1 χ2 scattering in the CQ model is consistent with
Lorentz symmetry.

21
 Finally, let us look at the transition probability in the full relativistic setting,

2 2
P (p1p2 → p′1 p′2 ) = |MF1 (p1 p2 → p′1 p′2 )| + [MF2 (p1 p2 → p′1 p′2 )]
 !2 !2 
4 4 ′ ′
V T λ δ (p1 + p2 − p1 − p2 )  1 D2
= 4 4
Q ′ 2 2
+ ′ 2 + m2 ]2
+ D0 
(2π) (2π) ω ω
i=1,2 pi pi
′ (p 1 − p 1 ) + mφ [(p 1 − p 1 ) φ
!2
V T λ4 δ 4 (p′1 − p1 − (p′2 − p2 )) D2
+ + D0
[(p1 − p1 )2 + m2φ ]2
′
Q
(2π)4 (2π)4 i=1,2 ωpi ωp′i
(57)

where δ 4 (0) = V T /(2π)4 . The first term is proportional to δ 4 (p′1 + p′2 − p1 − p2 ) and hence
this follows the four-momentum conservation, on the other hand, the last term does not. As
discussed above, the χ1 χ2 → χ1 χ2 scattering process in the CQ model is consistent with the
Lorentz symmetry, but it does not follow the four-momentum conservation law.

IV. COMPARISON TO ASTROPHYSICAL SCATTERING

To compare this model to the real world, let’s consider scattering a small body, like a
spacecraft of mass m, against a large body, like a planet of mass M, as in a slingshot maneuver
to accelerate a spacecraft. See Fig. 4. We choose this because it is a clear scattering problem,
and because there is extensive and precise data from real spacecraft. Another option could
be to analyze asteroid scattering, for example.
We can analyze this problem in the non-relativistic setting for simplicity. To match to

p′

FIG. 4. Heavy-light scattering, for example a spacecraft of mass m performing a slingshot maneuver
around a planet of mass M .

22
the Newton potential, we identify the Yukawa coupling as
p
λ= GN mM. (58)

Also, we send the Yukawa mass mφ → 0. Since M ≫ m, the reduced mass µ ≈ m. We will
work in the rest frame of M. Let p = pẑ be the initial momentum of the small object. We
now compute the effects of the three terms in Eq. (44).

The first term,

" 2 #
(2π)2 λ4 4 ′

dσ 1 D 2
⊃ δ (p + P ′ − p − P ) + + D0 , (59)
d3 p′ d3 P′ u m2 M 2 (p′ − p)4 (p′ − p)4

has the usual overall energy-momentum conservation delta function and thus can be analyzed
like a standard scattering problem. First we integrate δ 3 (p′ +P′ −p)d3 p′ d3 P′ = d3 p′ , and set
P′ := p−p′ everywhere. We then integrate the energy delta function δ(Ep′ +EP′ −Ep )d3 p′ ≈
mpdΩ, using m/M ≪ 1, and set |p′ | = |p| = p everywhere. In particular, we then have
p′ = pn̂ where n̂ is a unit three-vector. The cross-section becomes
"  2 #
1 D2
dσ ⊃ 4π 2 G2N M 2 m4 + + D0 dΩ, (60)
16p4 sin4 (θ/2) 16p4 sin4 (θ/2)

where ẑ · n̂ = cos θ and we used u = p/µ. The first term here is just the ordinary for-
mula for Rutherford/Newtonian scattering. The subsequent terms involving D0,2 are new
contributions from the CQ theory. We comment on their effects below.

Now consider the second term in Eq. (44),

2
(2π)2 λ4 4 ′

dσ D2
⊃ δ (p − p − (P ′ − P )) + D0 . (61)
d3 p′ d3 P′ u m2 M 2 (p′ − p)4

The phenomenology of this term is very different. The energy-momentum delta function
does not enforce the usual conservation rule but rather says that both particles receive equal
kicks, as opposed to equal and opposite kicks. These kicks come from the stochastic field.
The three-momentum delta function now correspondingly says that P′ = p′ − p, and the
corresponding integral gives δ 3 (p′ − p − P′ )d3 p′ d3 P′ = d3 p′ as before. The energy delta now
enforces
|p′ |2 − |p|2
0 = Ep′ − Ep − EP′ ≈ , (62)
2m
where we used m ≪ M, which here means we neglect the energy change in the large mass. So
this again sets p′ = pn̂ where n̂ is any unit three-vector. We can then reduce the remaining

23
volume element in the same way as the previous paragraph, and we obtain
 2
2 D2
dσ ⊃ 4π G2N M 2 m4 + D0 dΩ. (63)
16p4 sin4 (θ/2)

This is identical to the CQ contribution in Eq. (60). We emphasize that this is only the case
in our approximation that m ≪ M.
Putting these results together, we obtain the differential cross-section
"  2 #
dσ 1 D 2
= 4π 2 G2N M 2 m4 4 +2 + D0 . (64)
dΩ 4
16p sin (θ/2) 16p sin4 (θ/2)
4

Again, the first term here is the usual formula obtained for Newtonian/Rutherford scattering.
Thus this equation represents a deviation from the usual predictions of classical gravitational
scattering, encoded in the second term. This term comes from the noise added in the CQ
model. In particular, its total contribution is lower-bounded by the CQ diffusion-decoherence
constraint D0 D2 ≥ 1. Moreover, since this term has a nearly identical angular dependence
as the usual Newtonian term, this appears to change gravitational scattering distributions at
O(1), which is in conflict with basic observations of these kinds of processes. We emphasize
again that this conclusion is obtained only in the simplest case of D0,2 constant, although it
holds for any values of these constants subject to the CQ constraint D0 D2 ≥ 1.

V. COMMENTS AND OUTLOOK

It is of elementary interest to know if and how relativistic quantum fields can couple to
classical ones. In the context of gravity, while our experience with the other gauge fields of
nature strongly suggests that the metric should be similarly quantized, it is also of elementary
interest to determine if there is any self-consistent way that the metric can be treated as a
classical stochastic variable.
In this paper, we have analyzed the recent classical-quantum (“CQ”) proposal of Oppen-
heim et al., which provides an attempt at an affirmative, constructive answer to both of these
questions. Perhaps surprisingly, we have found that a version of tree-level scattering theory
can be formulated in an appropriate limit of this model. As advertised by Oppenheim et
al., the scattering probabilities are Lorentz invariant. Counter-intuitively, the probabilities
do not obey energy-momentum conservation. These two facts are consistent because the
model is fundamentally an open system, and there is no analogue of Noether’s theorem.
Moreover, although the model is an open system and thus not unitary, total probability is
still conserved, so the results appear to be self-consistent at this basic level.

24
 Quantitatively, we have analyzed scattering of two massive objects in this CQ framework,
both relativistic and non-relativistic. Taken at face value, the non-relativistic result is in
direct conflict with scattering of astrophysical objects. The scattering probabilities over-
predict small-angle scattering, as well as the total cross-section, by an O(1) amount. This
may suggest that the detailed construction is ruled out.
There are three main caveats to this conclusion. The obvious one is that we are working
with a simple Yukawa interaction, not full general relativity, although for our non-relativistc
calculations this should not pose any problems. Another is that we are only working in the
simplest case where the diffusion and decoherence functions D0,2 are constants. We make
no claims about what happens when these are allowed to be field-dependent functionals.
Perhaps the trickiest caveat is that we are assuming that we can use this formalism treating
a large composite object as a point mass. In ordinary classical or perturbative quantum
gravity, this works automatically because of the Gauss law. Here, we should really do a more
careful study of the renormalization of the model to understand how to go from microscopic
interactions to macroscopic interactions. This is an interesting direction that we leave to
future work.
Generally speaking, our conclusion is that the Oppenheim et al. construction is an in-
teresting and non-trivial attempt to couple relativistic quantum and classical systems. At a
structural level it seems to produce basic features that any such coupling should have, namely
Lorentz invariance and conservation of probability. However, as a model of quantum gravity,
the simplest case (D0,2 = constant) seems to be in conflict with the real world. Moreover,
the lack of friction in the diffusive evolution of the classical fields may be fundamentally
problematic. These issues warrant further study.

ACKNOWLEDGEMENTS

We thank Isaac Layton, Jonathan Oppenheim, Geoff Penington, Shivaji Sondhi, Jacob
Taylor, and Zach Weller-Davies for valuable discussions. D.C. is supported by the U.S. DOE,
Office of High Energy Physics, under Contract No. DEAC02-05CH11231, by DOE Quan-
tum Information Science Enabled Discovery (QuantISED) for High Energy Physics grant
KA2401032, and by the Heising-Simons Foundation grant 2023-4467 “Testing the Quantum
Coherence of Gravity”. A.M. is supported by JSPS KAKENHI (Grant No. JP23K13103 and
No. JP23H01175).

25
 [1] J. F. Donoghue, “Introduction to the effective field theory description of gravity,” in
Advanced School on Effective Theories. 6, 1995. arXiv:gr-qc/9512024.
[2] C. P. Burgess, “Quantum gravity in everyday life: General relativity as an effective field
theory,” Living Reviews in Relativity 7 (2004) 1–56.
[3] J. F. Donoghue, Quantum General Relativity and Effective Field Theory. 2023.
arXiv:2211.09902 [hep-th].
[4] D. Carney, P. C. E. Stamp, and J. M. Taylor, “Tabletop experiments for quantum gravity: a
user’s manual,” Class. Quant. Grav. 36 no. 3, (2019) 034001,
arXiv:1807.11494 [quant-ph].
[5] N. H. Lindner and A. Peres, “Testing quantum superpositions of the gravitational field with
Bose-Einstein condensates,” Phys. Rev. A 71 (2005) 024101, arXiv:gr-qc/0410030.
[6] D. Kafri and J. M. Taylor, “A noise inequality for classical forces,”
arXiv:1311.4558 [quant-ph].
[7] S. Bose, A. Mazumdar, G. W. Morley, H. Ulbricht, M. Toroš, M. Paternostro, A. A. Geraci,
P. F. Barker, M. Kim, and G. Milburn, “Spin entanglement witness for quantum gravity,”
Physical review letters 119 no. 24, (2017) 240401.
[8] C. Marletto and V. Vedral, “Gravitationally induced entanglement between two massive
particles is sufficient evidence of quantum effects in gravity,” Phys. Rev. Lett. 119 no. 24,
(2017) 240402.
[9] A. Matsumura and K. Yamamoto, “Gravity-induced entanglement in optomechanical
systems,” Phys. Rev. D 102 no. 10, (2020) 106021, arXiv:2010.05161 [gr-qc].
[10] D. Carney, H. Müller, and J. M. Taylor, “Using an Atom Interferometer to Infer
Gravitational Entanglement Generation,” PRX Quantum 2 no. 3, (2021) 030330,
arXiv:2101.11629 [quant-ph]. [Erratum: PRX Quantum 3, 010902 (2022)].
[11] A. Datta and H. Miao, “Signatures of the quantum nature of gravity in the differential
motion of two masses,” Quantum Sci. Technol. 6 no. 4, (2021) 045014,
arXiv:2104.04414 [gr-qc].
[12] J. Oppenheim, C. Sparaciari, B. Šoda, and Z. Weller-Davies, “Gravitationally induced
decoherence vs space-time diffusion: testing the quantum nature of gravity,”
Nature Commun. 14 no. 1, (2023) 7910, arXiv:2203.01982 [quant-ph].
[13] L. Lami, J. S. Pedernales, and M. B. Plenio, “Testing the Quantumness of Gravity without
Entanglement,” Phys. Rev. X 14 no. 2, (2024) 021022, arXiv:2302.03075 [quant-ph].
[14] D. Carney, V. Domcke, and N. L. Rodd, “Graviton detection and the quantization of
gravity,” Phys. Rev. D 109 no. 4, (2024) 044009, arXiv:2308.12988 [hep-th].

26
[15] D. Carney, “Comments on graviton detection,” 7, 2024. arXiv:2408.00094 [gr-qc].
[16] D. Kafri, J. M. Taylor, and G. J. Milburn, “A classical channel model for gravitational
decoherence,” New J. Phys. 16 (2014) 065020, arXiv:1401.0946 [quant-ph].
[17] R. Penrose, “On Gravity’s role in Quantum State Reduction,”
Gen. Relativ. and Gravt. 28 (1996) 581–600.
[18] A. Tilloy and L. Diósi, “Sourcing semiclassical gravity from spontaneously localized quantum
matter,” Physical Review D 93 no. 2, (2016) 024026.
[19] A. Großardt, “Three little paradoxes: Making sense of semiclassical gravity,”
AVS Quantum Sci. 4 no. 1, (2022) 010502, arXiv:2201.10452 [gr-qc].
[20] D. Carney and J. M. Taylor, “Strongly incoherent gravity,” arXiv:2301.08378 [quant-ph].
[21] D. Carney, “Newton, entanglement, and the graviton,”
Phys. Rev. D 105 no. 2, (2022) 024029, arXiv:2108.06320 [quant-ph].
[22] J. Oppenheim, “A post-quantum theory of classical gravity?,”
Phys. Rev. X 13 (2023) 041040, arXiv:1811.03116 [hep-th].
[23] J. Oppenheim and Z. Weller-Davies, “Covariant path integrals for quantum fields
back-reacting on classical space-time,” arXiv:2302.07283 [gr-qc].
[24] I. Layton, J. Oppenheim, A. Russo, and Z. Weller-Davies, “The weak field limit of quantum
matter back-reacting on classical spacetime,” JHEP 08 (2023) 163,
arXiv:2307.02557 [gr-qc].
[25] Z. Weller-Davies, Classical-quantum dynamics with applications to gravity. PhD thesis,
University Coll. London, 2024.
[26] A. Grudka, T. Morris, J. Oppenheim, A. Russo, and M. Sajjad, “Renormalisation of
postquantum-classical gravity,” arXiv:2402.17844 [hep-th].
[27] G. Penington, March, 2024.
https://twitter.com/quantum_geoff/status/1765886036956299595.
[28] LIGO Scientific, Virgo Collaboration, B. P. Abbott et al., “Upper Limits on the
Stochastic Gravitational-Wave Background from Advanced LIGO’s First Observing Run,”
Phys. Rev. Lett. 118 no. 12, (2017) 121101, arXiv:1612.02029 [gr-qc]. [Erratum:
Phys.Rev.Lett. 119, 029901 (2017)].
[29] T. Banks, L. Susskind, and M. E. Peskin, “Difficulties for the evolution of pure states into
mixed states,” Nuclear Physics B 244 no. 1, (1984) 125–134.
[30] W. G. Unruh and R. M. Wald, “Evolution laws taking pure states to mixed states in
quantum field theory,” Physical Review D 52 no. 4, (1995) 2176.
[31] L. Diósi, “Is there a relativistic Gorini-Kossakowski-Lindblad-Sudarshan master equation?,”
Physical Review D 106 no. 5, (2022) L051901.

27
[32] A. Matsumura, “Reduced dynamics with Poincaré symmetry in an open quantum system,”
Physical Review A 108 no. 4, (2023) 042217.
[33] H. Risken, “The Fokker-Planck equation,” 1996.
[34] I. Marvian and R. W. Spekkens, “Extending noether’s theorem by quantifying the
asymmetry of quantum states,” Nature communications 5 no. 1, (2014) 3821.
[35] R. P. Feynman and F. L. Vernon Jr, “The theory of a general quantum system interacting
with a linear dissipative system,” Annals of physics 24 no. 118, (1963) .
[36] A. O. Caldeira and A. J. Leggett, “Path integral approach to quantum Brownian motion,”
Physica A: Statistical mechanics and its Applications 121 no. 3, (1983) 587–616.
[37] S. Weinberg, Lectures on Quantum Mechanics. Cambridge University Press, 2015.
[38] S. Weinberg, The Quantum Theory of Fields. Cambridge University Press, 1995.
[39] S. W. Hawking, “Breakdown of predictability in gravitational collapse,” Physical Review D
14 no. 10, (1976) 2460.
[40] L. H. Ryder, Quantum Field Theory. Cambridge University Press, 1996.

28
 Appendix A: Classical equations of motion

In this appendix, we consider the time evolution of classical field φ, averaging over the
quantum matter. In particular, we want to show in what sense the CQ framework produces a
“semiclassical” equation of motion where φ is sourced by expectation values of the quantum
matter, e.g., by λ hχ2 i. To this end, we begin with the following CQ path integral,
Z
̺(χf , χf , φf , πf , tf ) = DχDχDφDπeICQ [χ,χ,φ,π] ̺(χ0 , χ0 , φ0 , π0 , t0 ), (A1)

where Dχ = Dχ1 Dχ2 , Dχ = Dχ1 Dχ2 , and the CQ action is

Z
ICQ [χ, χ, φ, π] = iS0 [χ] − iS0 [χ] + iλ d4 x(χ2 − χ2 )φ + ln(δ[φ̇ − π])
(A2)
D0 λ 2 1 λ
Z Z h i2
4 2 2 2
− d x(χ − χ ) − d4 x π̇ − (∇2 − m2φ )φ − (χ2 + χ2 ) .
2 2D2 2

This CQ action is computed from Eq. (14) by adopting the Yukawa interaction Sint of
Eq. (15).

Since the the term with D2 generates the diffusive and stochastic evolution of the classical
field φ, we expect that its time evolution is governed by Langevin equations. To see this
explicitly, we can use the identity
r
λ D2 i
Z h
1= Dξδ π̇ − (∇2 − m2φ )φ − (χ2 + χ2 ) − ξ . (A3)
2 2

Here ξ = ξ(x) is again a white noise random variable [c.f. Eq. (11)], which generates the
diffusive evolution of the classical field. Concretely, we can use this identity and Eq. (A2)
to rewrite Eq. (A1) as follows:

̺(χf , χf , φf , πf , tf )
r
R 4 2 λ D2 i
Z Z h
= Dξe− d x ξ DχDχDφDπδ[φ̇ − π]δ π̇ − (∇2 − m2φ )φ − (χ2 + χ2 ) − ξ
2 2
R D0 λ2 R
d4 x(χ2 −χ2 )φ − d4 x(χ2 −χ2 )2
× eiS0 [χ]−iS0 [χ]+iλ e 2 ̺(χ0 , χ0 , φ0 , π0 , t0 ).
(A4)

The delta functionals condition the classical trajectory of (φ, π) on

r
λ D2
φ̇ = π, π̇ = (∇2 − m2φ )φ + (χ2 + χ2 ) + ξ, (A5)
2 2

29
and these are the Langevin equations of classical field.

With this rewriting of the path integral, it is easy to see how the classical field evolves.
Consider the expectation value of π̇, for example. This is given by
Z
hπ̇i = DχDχDφDπ eICQ [χ,χ,φ,π] × π̇ × ̺(χ0 , χ0 , φ0, π0 , t0 )
r
R 4 2 λ D2 i
Z Z h
= Dξe− d xξ DχDχDφDπ δ[φ̇ − π]δ π̇ − (∇2 − m2φ )φ − (χ2 + χ2 ) − ξ
2 2
R 4 2 2 D0 λ2 R 4 2 2 2
× eiS0 [χ]−iS0 [χ]+iλ d x(χ −χ )φ e− 2 d x(χ −χ )
r
n λ D2 o
× (∇2 − m2φ )φ + (χ2 + χ2 ) + ξ × ̺(χ0 , χ0 , φ0 , π0 , t0 )
2 r 2
D2
= (∇2 − m2φ )hφi + λhχ2 i + hξi.
2
(A6)

The first line is the usual expression for an expectation value in a path integral. The second
line was obtained by inserting Eq. (A4) and using the delta function to replace π̇ with the
equation of motion. To obtain the third line, we used hχ2 i = hχ2 i∗ = hχ2 i. Note that we
should take the boundary condition χ = χ in the upper limit of the DχDχ path integral,
which is called the closed-time-path condition.

The result in Eq. (A6) tells us that the classical field is excited by the quantum field
through the source term λhχ2 i, i.e., by the expectation value of the quantum matter. It is
also excited by the white noise ξ, which generally in the main text we have taken to have
zero mean hξi = 0.

Appendix B: Feynman-Vernon derivations

Here we derive the Feynman-Vernon path integrals used in the main text. Our starting
point is the path integral for giving the influence functional,
Z
iS0 [χ]−iS0 [χ]+iSIF [χ,χ]
e = DφDπeICQ[χ,χ,φ,π]P[φ0 , π0 , t0 ], (B1)

30
where S0 [χ] the free action of massive scalar fields χ = (χ1 , χ2 ) and P[φ0 , π0 , t0 ] is the initial
distribution of classical field. The CQ action is
Z
ICQ [x1 , x2 , x1 , x2 , φ, π] = ln(δ[π − φ̇]) + iS0 [χ] − iS0 [χ] + i d4 x(J − J )φ
D0 1 1
Z Z h i2
4 2 4 2 2
− d x(J − J) − d x π̇ − (∇ − mφ )φ − (J + J)
2 2D2 2
(B2)

with the currents of quantum particles,

X X
J(x) = Ji (x), Ji (x) = λχ2i (x), J(x) = J i (x), J i (x) = λχ2i (x). (B3)
i i

Substituting this CQ action into Eq. (B1), we have the following expression of the influence
functional SIF ,
Z R D0 R 1
R  2
iSIF i d4 x(J−J)φ− d4 x(J−J)2 − 2D d4 x π̇−(∇2 −m2φ )φ− 21 (J+J)
e = DφDπδ[π − φ̇]e 2 2 P[φ0 , π0 , t0 ].
(B4)

Using the identity,

1
Z h i
2 2
Dξδ π̇ − (∇ − mφ )φ − (J + J) − ξ = 1, (B5)
2

we then arrive at
Z R 4 2 D0 R 4
iSIF − 1 d xξ − 2 d x(J−J)2
e = Dξe 2D2
 i R d4 x(J−J)φ (B6)
1
Z
2 2

× DφDπδ[π − φ̇]δ π̇ − (∇ − mφ )φ − (J + J) − ξ e P[φ0 , π0 , t0 ].
2

Let us evaluate the path integral with respect to φ and π. The delta functionals δ[π − φ̇]
and δ π̇ − (∇2 − m2φ )φ − 21 (J + J ) − ξ suggest that the classical field follows the Langevin
 

equations,
1
φ̇ = π, π̇ = (∇2 − m2φ )φ + (J + J ) + ξ, (B7)
2
where the white noise ξ obeys the Gaussian statistics,
Z
1
R Z
1
R
− 2D d4 x ξ 2 − 2D d4 x ξ 2
Dξe 2 ξ(x) = 0, Dξe 2 ξ(x)ξ(y) = D2 δ 4 (x − y). (B8)

31
The initial condition of the Langevin equations is determined by the initial distribution
P[φ0 , π0 , t0 ]. Here, we adopt the initial vacuum state of classical field with P[φ0 , π0 , t0 ] =
δ[φ(t0 )]δ[π(t0 )] at the initial time t0 . This means that the classical field φ is initially zero and
it is sourced by the currents Ji , J i of quantum particles and the white noise ξ. Explicitly,
the solution φ is
 
1
Z
4
φ(x) = d y GR (x − y) {J(y) + J(y)} + ξ(y) , (B9)
2

where

d4 k eik·(x−y)
Z
GR (x − y) = , (B10)
(2π)4 −(k 0 + iǫ)2 + k2 + m2φ

is the retarded Green’s function. It satisfies (−∂x2 +m2φ )GR (x−y) = δ 4 (x−y) and GR (x−y) =
0 for x0 − y 0 < 0. Hence, after performing the integration with respect to φ and π, we just
should replace φ with the solution given in Eq. (B9). Following this rule, we can get the
exact form of eiSIF as
Z
1
R 4 2 D0 R 4
iSIF − 2D d xξ − 2 d x(J−J)2
e = Dξe 2

1  R 4
Z
× DφDπδ[π − φ̇]δ π̇ − (∇2 − m2φ )φ − (J + J) − ξ ei d x(J−J)φ P[φ0 , π0 , t0 ]

2
Z R 4 2 D0 R 4 R
− 1
d xξ − 2 d x(J−J)2 +i d4 xd4 y(J−J )GR [ 21 (J+J)+ξ ]
= Dξe 2D2
(B11)
1
Z
2 2
 
× DφDπδ[π − φ̇]δ π̇ − (∇ − mφ )φ − (J + J) − ξ δ[φ(t0 )]δ[π(t0 )]
2
D0 R 4 R 4 4
Z R 4 2 R 4 4
2 1 − 1 d xξ +i d xd y(J−J)GR ξ
= e− 2 d x(J−J) +i 2 d xd y(J−J )GR (J+J) Dξe 2D2
D0 R R D2 R
d4 x(J−J)2 +i 21 d4 xd4 y(J−J )GR (J+J)− d4 xd4 y(J−J)GC (J−J)
= e− 2 2

In the second equality, the field φ of the exponential was replaced with the solution given
in Eq. (B9), and in the third equality the φ and π path integrals were performed. In the
fourth equality, we did the Gaussian integration with respect to ξ. The two point function
GC = GC (x − y) is
Z
GC (x − y) = d4 z GR (x − z)GR (y − z)

d4 k eik·(x−y)
Z
= . (B12)
(2π)4 [−(k 0 + iǫ)2 + k2 + m2φ ][−(k 0 − iǫ)2 + k2 + m2φ ]

32
The influence functional SIF is thus

1
Z
iSIF = − d4 xd4 yJ(x)F1 (x − y)J(y) − 2J(x)F2 (x − y)J(y) + J(x)F1∗ (x − y)J(y), (B13)
2

where
i
F1 (x − y) = D2 GC (x − y) − [GR (x − y) + GR (y − x)] + D0 δ 4 (x − y),
2 (B14)
i 4
F2 (x − y) = D2 GC (x − y) + [GR (x − y) − GR (y − x)] + D0 δ (x − y).
2
P
Since the current is J = i Ji , in the influence functional, we have the self interaction term
Ji2 and J 2i . For the scattering process of interest, we pick up the cross terms Ji Jj , J i J j and
Ji J j with i 6= j. The influence functional only with the cross terms is
Z X
iSIF = − d4 xd4 yJ1(x)F1 (x − y)J2 (y) + J 1 (x)F1∗ (x − y)J 2 (y) − Ji (x)F2 (x − y)J j (y).
i6=j
(B15)
Substituting the form of each current (B3) for this, we can produce the influence functional
given in Eq. (46).

To get the non-relativistic limit, note that the above procedure for deriving Eq. (B15)
does not rely on the concrete form of the currents. To get the non-relativistic version (25),
we can substitute the current Ji (x) = mλi δ 3 (x − xi ) for the influence functional. The non-
relativistic propagators are reduced by noting that the momentum transfer k µ = p′µ − pµ ,
so k 0 ≈ (p′2 − p2 )/2m ≪ |k|. This means we can drop the k 0 terms in the denominators,
0 0 0
which in particular produces dk 0 eik (x −y ) = 2πδ(x0 − y 0 ), which is why the non-local SIF
R

collapses to a time-local expression.

As discussed here and in the main text, we are only focusing on terms of the form J1 J2 ,
rather than allowing for any self-interactions like J1 J1 . This is because we are looking at
tree-level scattering to lowest order in perturbation theory. To this order, the self-interaction
terms only contribute as loop diagrams in the forward scattering process, so as long as we
consider non-forward scattering this is sufficient. However, these self-energy diagrams are
definitely of importance in the CQ model, and will behave differently than in a unitary
QFT. In particular, the self-energy of the matter will necessarily have finite width coming
from the D0,2 interactions. Understanding how to deal with these loops via some kind of
renormalization procedure is a very interesting problem which we leave to future work (see,
e.g., [26] for some further comments).

33
 Appendix C: Lorentz invariance of scattering supermatrix

In Sec. III B, we have observed the Lorentz invariance of the scattering supermatrix el-
ements in our CQ model. Here, to help our understanding of why this invariance appears,
we discuss the Lorentz invariance of the scattering supermatrix in the quantum counterpart
of our CQ model, where all the fields including φ are quantized. In particular, one might
have wondered by the classical φ field does not provide a reference frame; we show here why
the analogous fully quantum process does not provide a reference frame. This also gives
us the opportunity to review the basic transformation properties of transition probabilities,
and show how Lorentz-invariant probabilities that do not preserve four-momentum can arise
once one drops the constraint of a unitary S-matrix.
In the CQ calculations in the main text, we defined a channel M which gives time
evolution on just the χ fields, see Eq. (21). This was defined by assuming an initial φ
distribution P[φ, π] = δ[φ]δ[π] and averaging over the final φ configurations. The analogous
channel in the fully quantized setting is
h i
M [|αi hα|] = Trφ S |αi hα| ⊗ |0i h0|φ S † , (C1)

where S = limT →∞ eiH0 T e−i2HT e−iH0 T is the usual S-operator defined with the full Hamilto-
nian H and the free Hamiltonian H0 . The scattering supermatrix element computed from
the channel M is given by the same formula we used in the CQ case, viz.

/ αα→ββ = hβ|M [|αi hα|] |βi .

M (C2)

In the following, we will analyze how M / αα→ββ acquires Lorentz invariance, despite the fact
that we are tracing out the final φ states. In particular this means that the φ field does not
generate some preferred frame for the χ fields, even though φ can be excited. The proof
does not rely on perturbation theory.
Let U = V ⊗ W be a Poincaré transformation U = U(Λ, a). Following the standard
formulation of scattering theory presented in [38], we assume that this factors into a Poincaré
transformation V acting on the asymptotic χ states and a Poincaré transformation W acting
on the asymptotic φ states. The vacuum of φ is invariant

W |0iφ = |0iφ , (C3)

and the full transformation commutes with the S-matrix

[S, U] = 0. (C4)

34
Using these, we can show that M is covariant under V :
h n o i
M[V |αi hα| V † ] = Trφ S V |αi hα| V † ⊗ |0i h0|φ S †
h n o i
= Trφ S V |αi hα| V † ⊗ W |0i h0|φ W † S †
h i
† †
= Trφ U S |αi hα| ⊗ |0i h0|φ S U
h i (C5)
= V Trφ W S |αi hα| ⊗ |0i h0|φ S † W † V †
h i
= V Trφ S |αi hα| ⊗ |0i h0|φ S V †
†

= V M[|αi hα|]V † .

The first line is a definition, the second used Eq. (C3), and to get from the fourth to fifth
lines we used that W only acts on the φ Hilbert space so it cycles under the partial trace.

We can be a bit more explicit about the implications of Eq. (C5), which is instructive
for comparing to the CQ model. Since we are interested in scalar fields χ = (χ1 , χ2 ), we
can take the Fock states |αi and |βi of χ particles with definite momenta α = p1 , p2 , . . .
and β = p′1 , p′2 , . . ., respectively. In our scattering process, the following matrix element is
relevant, h i
M/ p1 p2 p p →p′ p′ p′ p′ = hp′1 p′2 |M |p1 p2 i hp p | |p′ p′ i . (C6)
1 2 1 2 1 2 1 2 1 2

These states transform as usual [38],

s
(Λp1 )0 (Λp2 )0 −i(Λp1 +Λp2 )µ aµ
V (Λ, a)|p1 p2 i = e |Λp1 Λp2 i. (C7)
p01 p02

Using V M[|αi hα|]V † = M[V |αi hα| V † ], or equivalently M[|αi hα|] = V † M[V |αi hα| V † ]V
shown in (C5), we have
" s # " #
Y (Λp′i )0 (Λpi )0 (Λp′i )0 (Λpi )0 X
/ p1 p2 p
M p →p′1 p′2 p′1 p′2 = exp i (Λp′i − Λpi − Λp′i + Λpi )µ aµ
1 2
i=1,2
p′ 0i p0i p′ 0i p0i i=1,2

/ Λp1 Λp2 Λp
×M Λp2 →Λp′1 Λp′2 Λp′1 Λp′2 .
1

(C8)

This is the generalization of the usual Lorentz invariance of S-matrix elements, which reads
" s # " #
Y (Λp′i )0 (Λpi )0 X
Sp1 p2 →p′1 p′2 = exp i (Λp′i − Λpi )µ aµ SΛp1 Λp2 →Λp′1 Λp′2 (C9)
i=1,2
p′ 0i p0i i=1,2

35
Notice, however, the more general condition can mix the bra and ket variables in the energy-
momentum conservation exponential.
To be even more explicit, first consider the case with no spacetime translation aµ = 0.
Then Eq. (C8) implies
hYq i
p′ 0i p0i p′ 0i p0i M/ p1 p2 p p →p′1 p′2 p′1 p′2
1 2
i
hYq i (C10)
= (Λp′i )0 (Λpi )0 (Λp′i )0 (Λpi )0 / Λp1 Λp2 Λp
M Λp2 →Λp′1 Λp′2 Λp′1 Λp′2 .
1
i

hQ q i
Hence, p ′ 0 p0 p′ 0 p0 M/ p1 p2 p p →p′ p′ p′ p′ is Lorentz invariant under pi → Λpi , p′i → Λp′i ,
i i i i i 1 2 1 2 1 2

pi → Λpi and p′i → Λp′i . This is again a straightforward generalization of the usual S-matrix
result, and it also seems to hold for the scattering supermatrix computed for our CQ Yukawa
model in Sec. III B. On the other hand, with a non-trivial translation aµ 6= 0 and Λµ ν = δνµ ,
Eq. (C8) gives
" #
X
/ p1 p2 p
M p →p′1 p′2 p′1 p′2 = exp i (p′i − pi − p′i + pi )µ aµ M
/ p1 p2 p p →p′1 p′2 p′1 p′2 . (C11)
1 2 1 2
i=1,2

Notice that this involves both the bra and ket variables. In the usual S-matrix constraint,
the exponential is a product of two exponentials, one enforcing δ 4 (p′1 + p′2 − p1 − p2 ) and the
other enforcing δ 4 (p′1 +p′2 −p1 −p2 ). In the more general case shown here, there are also more
general delta function combinations allowed, such as those in the CQ terms of Eqs. (43) or
(57), which mix the bra and ket variables.

Appendix D: Conservation of probability

In this appendix we show how to check that the total probability is conserved in this
model, order by order in perturbation theory. The condition we need is that the final state
density matrix has unit trace. In terms of the scattering supermatrix elements, the final
state is
XX
ρf = / αα→ββ ρ0,αα |βi hβ| ,
M (D1)
ββ αα

where ρ0 is the initial state. Normalization of the final state requires

XX
1 = tr ρf = / αα→ββ ρ0,αα .
M (D2)
β αα

36
We assume this initial state has unit trace,
X
1 = tr ρ0 = ρ0,αα (D3)
α

in which case our requirement on the scattering elements becomes

X
/ αα→ββ = δαα .
M (D4)
β

Eq. (D4) is our elementary requirement on the scattering process which ensures that total
probability is conserved.

Perturbatively, we can be a little more explicit. In perturbation theory we can write

M/ αα→ββ = δαβ δαβ + T/ αα→ββ , where the first term represents no scattering and the T/ is the
open systems version of the usual scattering T matrix (e.g., the generalization of S = 1 + iT
in S-matrix theory). In this setting, Eq. (D4) becomes simply
X
T/ αα→ββ = 0. (D5)
β

This is the open systems analogue of the optical theorem.

In our Yukawa model, we can check Eq. (D5) order by order in λ. The first non-trivial
check in 2 → 2 scattering is at O(λ2 ). In terms of scattering probabilities, only forward
scattering contributes at this order, but note that Eq. (D5) is more general. It is a statement
about amplitudes for superpositions |αi hα|, i.e., involves the coherence of the state. More
explicitly, Eq. (D5) reads
Z
d3 p′1 d3 p′2 T/ p1 p2 p p →p′1 p′2 p′1 p′2 = 0. (D6)
1 2

Following the same logic as in Sec. III A, we have processes involving MF1 and MF2 . Con-
cretely we have, for the general process,

T/ p1 p2 p p →p′1 p′2 p′1 p′2

1 2

= MF1 (p1 p2 → p′1 p′2 )δ 3 (p1 − p′1 )δ 3 (p2 − p′2 ) + δ 3 (p1 − p′1 )δ 3 (p2 − p′2 )MF1∗ (p1 p2 → p′1 p′2 )
+ δ 3 (p2 − p′2 )MF2 (p1 p2 → p′1 p′2 )δ 3 (p1 − p′1 ) + δ 3 (p1 − p′1 )MF2 (p2 p1 → p′2 p′1 )δ 3 (p2 − p′2 ).
(D7)

These are the terms on the right hand side of Eq. (34), except for the trivial no-scattering
term. Now we set p′1 = p′1 , p′2 = p′2 , and integrate over the final states, to compare to

37
Eq. (D5). The result is
Z
d3 p′1 d3 p′2 T/ p1 p2 p p →p′1 p′2 p′1 p′2
1 2

= MF1 (p1 p2 → p1 p2 ) + MF1∗ (p1 p2 → p1 p2 ) + MF2 (p1 p2 → p1 p2 ) + MF2 (p2 p1 → p2 p1 )

" #
2
λ i D 2
= −δ 4 (p1 + p2 − p1 − p2 ) + + D0
(2π)2 m1 m2 k2 + m2φ (k2 + m2φ )2
k=p1 −p1
" #
λ2 −i D2
− δ 4 (p1 + p2 − p1 − p2 ) + 2 + D0
(2π) m1 m2 k + mφ (k + m2φ )2
2 2 2
k=p1 −p1
" #
λ2 D2
+ δ 4 (p1 − p2 − p1 + p2 ) + D0
(2π) m1 m2 (k + m2φ )2
2 2
k=p −p1
" # 1
λ2 D2
+ δ 4 (p2 − p1 − p2 + p1 ) + D0
(2π) m1 m2 (k + m2φ )2
2 2
k=p2 −p2

= 0,
(D8)

where pµ1 + pµ2 = pµ1 + pµ2 holds for each term and note that the delta function of MF2 is
different from that of MF1 . This is because MF2 bridges the ket line and the bra line and
because the flow of momentum for each line is opposite to each other. Hence at least to
O(λ2 ), the trace is preserved under CQ evolution.
It would be very interesting to generalize this to O(λ4 ), but this would require the analysis
of loop integrals and we leave it to future work. We also note that even at O(λ2 ), we have
simplified our analysis by ignoring self-energy terms. We have also checked that including
self-energy diagrams at O(λ2 ) does not change the conclusion that probability is conserved.
The proof is somewhat tedious so we omit it here.

38