Ghirardi-Rimini-Weber model with massive flashes

Antoine Tilloy
Max-Planck-Institut fr Quantenoptik, Hans-Kopfermann-Strae 1, 85748 Garching, Germany
(September 13, 2017)
arXiv:1709.03809v1 [quant-ph] 12 Sep 2017

Abstract
We introduce a modification of the Ghirardi-Rimini-Weber (GRW) model in which the
flashes (or collapse space-time events) source a classical gravitational field. The resulting
semi-classical theory of Newtonian gravity preserves the statistical interpretation of quantum
states of matter in contrast with mean field approaches. It can be seen as a discrete version
of recent proposals of consistent hybrid quantum classical theories. The model seems to be
in agreement with known experimental data and introduces new falsifiable predictions: (1)
particles do not attract themselves, (2) the 1/r gravitational potential of Newtonian gravity
is cut-off at short (. 107 m) distances, and (3) gravity makes spatial superpositions decohere
at a rate inversely proportional to that coming from the vanilla GRW model. Together, the
last two predictions make the model experimentally falsifiable for all values of its parameters.

1 Introduction
Most attempts to unify gravity and quantum theory rely on the quantization (fundamental or
effective) of the former. Yet this route has proved so hard that, in the absence of experimental
evidence, it may be time to wonder if it really is the only one available. Could a classical curved
space-time and quantum matter not just cohabit at the deepest level? This latter option has
been historically dismissed on the ground that its naive instantiations yield crippling conceptual
difficulties [1], already in the Newtonian limit. However, it has been recently revisited and various
proposals have been put forward to construct consistent (non-relativistic) hybrid theories of gravity
[25].
At a formal level, these proposals implement gravity as a continuous measurement and feed-
back scheme on quantum matter. Equivalently, the proposals take the form of continuous collapse
models in which a noisy modification of the mass density is what sources the gravitational field.
The first point of view insures that the models have a transparent empirical content and the second
guarantees that they have a clear ontology and solve the measurement problem as a byproduct.
The pedagogical difficulty in discussing those models is that they require It calculus to be pre-
cisely formulated. As a consequence, many nice results (such as linearity at the master equation
level) appear rather mysteriously from the use of Its lemma. This makes the physical intuition
admittedly hard to grasp for the newcomer.
The objective of this article is to introduce a conceptually similar but technically easier model.
Starting from the simplest discrete collapse model, the Ghirardi-Rimini-Weber (GRW) model [6],
instead of its continuous refinements [79], we will implement a gravitational interaction by mak-
ing the flashes (collapse space-time events) massive. Arguably less elegant than its continuous
predecessors, the resulting model will yield a similar phenomenology while keeping technicalities
to a minimum. Thus our aim is not so much to propose a model of semi-classical gravity one
should believe in than to show a rather transparent example of a consistent approach with a clear
ontology and empirical content.
antoine.tilloy@mpq.mpg.de

1
2 The GRW model and its interpretation
2.1 The model
Let us start by introducing the standard GRW model. For that matter, we will consider an
isolated system of N distinguishable particles of coordinates {x1 , , xN }, masses {m1 , , mN },
and without spin nor other internal degrees of freedom. This restriction is not absolutely necessary
for the introduction of the GRW model (see e.g. [10]) but makes the presentation far simpler. The
N -particle system is described by the wave-function t (x1 , , xn ), with t L2 (R3N , C) which
obeys the standard Schrdinger equation between collapse events:
"N #
i i X 2k
t t = H0 t + V (x1 , , xN ) t , (1)
~ ~ 2mk
k=0

where xk is the position operator associated to particle k and V is some non-gravitational potential
(say electromagnetic). In the GRW model, this unitary evolution is interrupted by spontaneous
jumps of the wave-function that can hit each particle independently:

Lk (xf )t
t , (2)
kLk (xf )t k
with:
1 2 2
Lk (xf ) = 2 )3/4
e(xk xf ) /(2rC )
. (3)
(rC
These N types of jumps occur independently of each other and are uniformly distributed in time
with intensity . Given that a jump happens to particle k at time t, the probability density for
the center of the collapse xf is finally given by P (xf ) = kLk (xf )t k2 .
This model introduces two new parameters: the width of the collapsed wave-function rC and
the jump rate . They need to have small enough values that collapse events go unnoticed for
N small, making the standard unitary evolution (1) approximately valid, but large enough that
macroscopic superpositions (with N & 1020 ) are almost immediately collapsed in definite locations.
The typical values proposed the literature, = 1016 s1 and rC = 107 m, can be shown to achieve
this goal [6, 11] (although they might soon be experimentally falsified).
The GRW model is a quantum theory without observer in the sense that what happens in
measurement situations is derived rather than postulated. As measurement results are no longer
primitive in this approach, it brings the question of what is. What is the theory about? The
minimalist answer is that it is just about wave-functions [12]. One should understand the emergence
of tables and chairs from the fact that the wave-function of the universe becomes sharply peaked
around points in configuration space that would be classically associated to tables and chairs.
Another option is to introduce a primitive ontology [13, 14] (or local-beables [15]), that is stuff
in physical space, moving according to some laws and from which tables and chairs should emerge
via coarse graining (see [16] for an in depth discussion of the merits of this option). A possibility is to
take the space-time events or flashes (xf , tf ) naturally associated to the jumps of the wave-function
and make the commitment that they are what the theory is ultimately about. The wave-function
then only provides the probabilistic law of this pointillist fabric of reality. As we shall later see,
thinking of the flashes as real entities will be natural in the context of our gravitational extension
of the GRW model. It is however not a necessity and the reader may remain agnostic about the
ontological content of the theory.

2.2 Consistency with the statistical interpretation

An important feature of the GRW model, and of all collapse models for that matter, is that it
is consistent with the statistical interpretation of the wave-function. Once one understands how
the measurement postulate is derived from the model, one can use the wave-function exactly as in

2
orthodox quantum theory (with the Born rule and collapse in measurement situations) but for a
small modification of the dynamics. As we shall see, this is crucial to make the theory usable in
practice.
The probabilities one can compute in orthodox quantum theory are obtained as sesquilinear
forms on the state |i, i.e. they are quantities of the form a = h|A|i, where A is a Hermitian
operator. In collapse models, one does not have a direct empirical access to the random jumps and
so, actually, the empirical content of the model is contained in quantities of the form:
h i
a = E h|A|i
   (4)
= tr A E |ih| ,

where E[ ] denotes the statistical average over the random jumps. Thus, although the non-linear
random state evolution (1) + (2) is necessary to understand why the theory makes sense, all the
empirical content lies ultimately in the density matrix = E |ih| . In our N particle scenario,
the density matrix t is simply:

t : (x1 , , xN , y1 , , yN ) 7 E [t (x1 , , xN )t (y1 , , yN )] . (5)

and a straightforward computation shows that it obeys the equation:

N Z 
i X
t t = [H0 , t ] + dxf Lk (xf )t Lk (xf ) t (6)
~
k=1
N Z o
i X 1n
= [H0 , t ] + dxf Lk (xf )t Lk (xf ) Lk (xf )Lk (xf ), t , (7)
~ 2
k=1

where the second line is just meant at making the Lindblad form manifest. For notational conve-
nience we have used position operators acting on the left and on the right defined in the following
way:

xk t : (x1 , , xN , y1 , , yN ) 7 xk t (x1 , , xN , y1 , , yN ) (8)

t xk yk t : (x1 , , xN , y1 , , yN ) 7 yk t (x1 , , xN , y1 , , yN ). (9)

Importantly, equation (6) is linear. If it were not the case, the standard statistical interpretation
would necessarily break down [1719]. Indeed, if (6) were non-linear, it would mean that proper
mixtures (statistical ensembles of pure states) would no longer be empirically equivalent with
improper mixtures (decohered quantum superpositions of pure states obtained by tracing out
unknown degrees of freedom). This equivalence is what usually guarantees that Alice and Bob
cannot communicate using an EPR pair: measuring her state, Alice cannot know what Bob did
with his. With non-linearities, the statistics on Alices side would depend on Bobs measurements.
The most obvious and common objection is that it would allow faster than light signalling. The
difficulty is actually even more crippling: with non-linearities, reduced states can simply no longer
be trusted to make predictions. The only states one ever has access to in practice being reduced
states, there can be no consistent orthodox statistical interpretation1 . This does not necessarily
make a physical theory inconsistent, but requires to forgo the convenient operational toolbox of
quantum theory to rely solely on the ontology to make predictions2 . In practice, this seems like
an insurmountable task.
1 There might be alternative (non-quadratic) statistical interpretations of states for non-linear dynamics [20], but

we will not discuss this option here.

2 The linearity of the equation for the density matrix in collapse models has essentially the same role as equivari-

ance in Bohmian mechanics [21]. Without it, the theory might still make sense but its empirical content becomes
almost impossible to extract.

3
 We insist on this linearity at the master equation level because it is at the same time necessary
and unexpected. After all, we started with a non-linear collapse evolution! Yet, the jump proba-
bilities exactly simplify with the normalization factors of (2) when averaging over the noise, thus
magically removing the non-linearity for the density matrix. Of course, there is no coincidence and
the jump probabilities are fixed with this very objective in mind. One might say that everything
is fine-tuned to preserve the statistical interpretation of the quantum state. This means that if we
start messing with the GRW stochastic evolution (say by adding gravity), we would better make
sure that linearity is preserved. Unless there is some particular reason why linearity should hold,
it will typically not hold.

2.3 Orthodox reformulation and intuition

A possible understanding of why collapse models are consistent with the statistical interpretation
comes with noticing that they can be formally rewritten as purely orthodox models3 . The jump
part of the GRW evolution (2) takes the form of a positive-operator valued measure (POVM), i.e.
a generalized measurement perfectly authorized in operational quantum mechanics. Averaging
over measurement results is operationally equivalent to tracing out the degrees of freedom of the
measurement apparatus and naturally gives a perfectly legitimate linear open system evolution
(6).
Of course, the interpretation of the theory is entirely different. In the GRW model, no observer
is fuzzily measuring the particle positions and the jumps are taken as fundamental. Yet, the
fact that the formalism can be rewritten in orthodox terms can be seen as a guarantee that the
operational toolbox of quantum theory will survive.
In the orthodox picture, the flashes are measurement outcomes. Measurement outcomes have a
particularly nice status in quantum theory: they are the only classical variables one has access to.
In particular, measurement outcomes can be fed back into the evolution without causing trouble.
Some controller can apply an external field to a quantum system conditionally on a measurement
outcome without wreaking havoc on the statistical interpretation of states. If it were not so,
the operational toolbox of quantum theory would be unusable. Obviously in the GRW model,
no observer has access to the flashes. However, because they are equivalent to measurement
outcomes at a formal level, one can make the wave-function dynamics depend on them. If we are
in the business of constructing new theories, it is legitimate to use the flashes as classical sources.
This makes the flashes very different from other primitive ontologies encountered in foundations.
Neither the quantum expectation value of the mass density operator nor the Bohmian particle
trajectories can back react on the wave-function without ruining the statistical interpretation. To
my knowledge, the flashes (or their field equivalent for continuous collapse models) are the only
candidates for a primitive ontology so far proposed that need not be purely passive (i.e. determined
by the wave-function without influencing it)4 .

3 Sourcing gravity from the flashes

3.1 General setup
Guided by the previous intuition, we can use the flashes to source a classical gravitational field
(instead of the expectation value of the mass density used in [26]). We start by considering that
a flash just happened and forget the rest of the dynamics. For a flash (xf , tf ) associated to the
collapse operator Lk and thus to particle k, we source using the Poisson equation:
(x, t) = 4Gmk 1 f (t tf , x xf ), (10)
3 This is of course only a sufficient condition. For non-Markovian collapse models, there is no straightforward

real-time measurement interpretation [2225] and reformulating the model in orthodox terms is consequently not
obvious. It does not mean they violate the statistical interpretation.
4 We should insist that being passive is in no way a fundamental problem for a primitive ontology. Being able to

consistently back react on the wave-function simply offers more flexibility for theory building.

4
where f is a space-time form factor that could potentially be used to smear the flash in space-time.
It should verify d3 x dt f (t, x) = 1, f 0 and t < 0, f (t, x) = 0. As the particles flash at a
R
rate , the factor 1 insures that the field generated by the flashes approximately corresponds
on average to the field created by a classical particle in the same position. We will mainly be
interested in the singular limit where the gravitational potential is sourced by the sharp flash only
and thus take f (t tf , x xf ) (t tf ) 3 (x xf ). However, a general f may be useful for
generalizations out of the Newtonian limit where point sources would yield divergent metrics.
We now add the corresponding gravitational potential in the evolution for the wave-function
as a totally benign external potential:
Z
VG = dx (x) M (x) (11)

where M (x) is the mass density operator:

N
X
(M (x)t )(x1 , , xN ) = mk (x x )t (x1 , , xN ). (12)
=1

Inverting the Poisson equation (10) gives:

N
f (t tf , x xf )
X Z
VG (t) = G1 mk m dx t (x1 , , xN ). (13)
|x x |
=1

This is the quantum potential created by a single flash. By linearity, the total potential is simply
the sum of the potentials associated with all the past flashes.

3.2 Sharp limit

Taking the limit f (t tf , x xf ) (t tf ) 3 (x xf ) will makethe potential (13)
 created by
R +
each flash divergent but the corresponding unitary Uk (xf , k) = exp i tf dt VG (t) applied on
the wave-function instantaneous in the limit will remain finite. Indeed, in that limit we get:
N
!
G X mk m
Uk (xf ) = exp i . (14)
~ |xf x |
=1

Hence in the absence of smearing, the physical picture becomes particularly simple. Immedi-
ately after a flash, an instantaneous unitary is applied to the wave-function corresponding to the
singular gravitational pull of the flash. The evolution is thus very close to that of the original
GRW model but for a modification of the jump operators. Indeed, after particle k flashes, the
wave-function experiences the transformation:

Lk (xf )t Uk (xf )Lk (xf )t Bk (xf )t

t Uk (xf ) = . (15)
kLk (xf )t k kUk (xf )Lk (xf )t k kBk (xf )t k

As before we obtain the master equation for the density matrix:

N Z 
i X
t t = [H0 , t ] + dxf Bk (xf )t Bk (xf ) t . (16)
~
k=1

This is the central result of this article.

5
4 Phenomenology
4.1 Single particle case
We now study the evolution of a single isolated particle. Ignoring the free Hamiltonian H0 of the
particle to focus on the gravitational and collapse effects, we see that the master equation (6) is
diagonal in position:
t t (x, y) = ((x, y) 1)t (x, y), (17)
where (x, y) is an a priori complex kernel:
Gm2 (x xf )2 + (y xf )2
   
1 1 1
Z
(x, y) = 2 )3/2
dx f exp i 2 . (18)
(rC ~ |x xf | |y xf | 2rC
Changing of variable with xf = xf + (x + y)/2 and writing u = (x y)/2 gives:
Gm2 (xf u)2 + (xf + u)2
   
1 1 1
Z
(u) = 2 )3/2
dxf exp i 2 . (19)
(rC ~ |xf u| |xf + u| 2rC
This form shows that the integral on the upper half space is equal to the conjugate of the integral
on the lower half space, hence (u) is real. This is an important result as it means that the effect
of collapse and gravity on a single particle is simply a decay of the phases in the position basis. A
single particle does not attract itself in this model. This absence of self-attraction, shared with the
continuous versions of the present model, is a counter intuitive prediction markedly distinguishing
this approach from other semi-classical theories.
We further notice that the master equation (16) involves a new length scale rG = Gm2 /(~).
For = 1016 s1 , we get rG 1.8 1014 m for protons and rG 5.3 1021 m for electrons.
Hence = rG /rC < 107 is a small dimensionless parameter and we may expand around = 0:
= 0 + 1 + 2 2 + o(2 ). (20)
2
The interesting effect on decoherence can be found at second order. Indeed, we have 2 =
G2 m4 /(~2 )2 1 . This means that contains decoherence effects inversely proportional to
the collapse rate : the range of values allowed for the latter is experimentally falsifiable from
below. We will come back to this crucial point in the discussion.

4.2 Recovering the Newtonian potential

Intuitively, we constructed the model in such a way that the gravitational field generated by a large
massive object is approximately the one we would expect classically. Otherwise the model would
grossly clash with observations. To make sure that the model is indeed not empirically absurd,
we consider a test particle of coordinate x0 interacting with a lump of N particles of coordinates
x1 , , xN . The only thing we assume is that the test particle is separated from the lump by a
distance much larger than rC .
Ignoring again H0 , we expand (16) to first order in rG /rC to write the approximate master
equation for this set of N + 1 particles:
N Z
X 
t t dxf Lk (xf )t Lk (xf ) t
k=0
(21)
N X
N  
Gmk m 1 1
X Z
+i dxf Lk (xf )t Lk (xf ).
~ |x xf | |y xf |
k=0 =0

We shall be interested in the reduced density matrix of the test particle:

t (x0 , y0 ) = trN (t )(x0 , y0 ) (22)
Z
dx1 dxN t (x0 , x1 , , xN , y0 , x1 , , xN ). (23)

6
Taking the trace in (21) cancels most of the terms and one gets:
Z
t t (x0 , y0 ) = dxf L0 (xf )t L0 (xf ) t

N  
Gm0 mk 1 1
X Z
+i dx1 dxN dxf (24)
~ |x0 xf | |y0 xf |
k=1
|xk xf |2
 
1
2 )3/2
exp 2 t (x0 , x1 , , xN , y0 , x1 , , xN ).
(rC rC

We now use the fact that the test particle is far from the lump, i.e. for k [1, N ], the positions x0
and xk such that the density matrix is not vanishingly small verify |xk x0 | rC . This means
that in the previous integral:

|xk xf |2
 
1 1 1
Z
dxf 2 )3/2
exp 2 . (25)
|x0 xf | (rC rC |x0 xk |

Including this simplification in (24) yields:

N
(" #)
X Gm0 mk
t t (0 (t ) t ) i trN , t , (26)
~|x0 xk |
k=1

where 0 (t )(x0 , y0 ) = 0 (x0 , y0 )t (x0 , y0 ) contains the single particle decoherence coming from
the original GRW model. Equation (24) shows that the statistics on the test particle are approxi-
mately as if it were interacting with the N particles of the lump through a (quantum) Newtonian
pair potential. The crucial (and only) hypothesis to obtain this result is that the distance between
the test particle and the lump is much larger than rC .
If we further assume that the particles in the lump are well localized (typically because of
the collapse mechanism itself) around positions r1 , , rN on a length scale far smaller than the
lump-test particle distance, we can remove the trace and obtain the classical limit:
" N #
X Gm0 mk

t t (0 (t ) t ) i , . (27)
~ |x0 rk | t
k=1

Apart again from the decoherence coming from the collapse, this is what one would have put
by hand. A classical piece of matter approximately creates the classical gravitational field one
would naively expect, at least at a distance d rC 107 m away from it. At shorter distances,
the effective Newtonian pair potential is smoothed by the collapse operators and the 1/r law of
Newtonian gravity breaks down. However, to our knowledge this behavior is still compatible with
all the experiments carried so far where the gravitational field probed is typically sourced by large
objects away from the test mass.
Finally, one might argue that it would have been easier to just posit the existence of a quantum
pair potential to begin with (as one would have done e.g. for electromagnetism) instead of making
the flashes gravitate. However, this would have been against the spirit of fundamental semi-
classical gravity: we want a theory in which there is a classical gravitational field that mediates
the interaction. Actually, it comes almost as a surprise that the fluctuating classical gravitational
field we define in (10) ultimately yields an effective pair potential. Had we started from a pair
potential, we would have had no classical gravitational field to associate to it.

4.3 Falsifiability
Including gravity in the GRW model in the way we have suggested, i.e. by making the flashes
gravitate, has an interesting consequence on the parameter diagram of the model. Let us recall the

7
difficulty one faces in attempting to falsify collapse models, even only in principle. The problem
of the GRW model (and of collapse models in general) is that parameters leading to very unsharp
or very infrequent collapse can only ever be philosophically discarded. With a too small or rC
too large, the GRW model does not reduce macroscopic superposition and (1) is useless with
the minimalist wave-function ontology with respect to the Many-Worlds view (2) gives an a priori
inadequate picture of the world with the flash or mass density primitive ontology. However, there
is no way to experimentally falsify this fuzzily defined zone of philosophical discomfort. The widely
different values proposed by Ghirardi Rimini and Weber and by Adler for and rC can be seen as
an illustration of this rather subjective character of the collapse lower bounds.
Provided one believes in the present model, this problem is solved. As we have seen in 4.1,
the leading decoherence term is (the instrinsic GRW decoherence) but the coupling with
gravity adds decoherence 1 . This means that for a fixed value of rC , all the values of
can in principle be experimentally falsified. Intuitively, a very infrequent collapse yields very
infrequent but supermassive flashes and thus a very noisy gravitational interaction yielding strong
gravitational decoherence. Reciprocally, reducing gravitational decoherence requires very frequent
flashes and thus a strong intrinsic decoherence.
Even if for a given value of rC all values of can be falsified by probing positional decoherence,
arbitrarily large values of rC reduce both intrinsic and gravitational decoherence and are thus not
a priori eliminated. This is when the analysis of 4.2 is useful: the effective potential one obtains is
smoothed by the collapse operators and the 1/r law of gravity breaks down for distances d . rC .
Hence, small values of rC can be falsified by decoherence and large values of rC can be falsified
by probing the gravitational force at short distances. Our model is in principle experimentally
falsifiable for all values of and rC .

5 Discussion
5.1 Summary
The dynamical equations of collapse models like the GRW model are the same that the one that de-
scribe repeated unsharp measurements of particle positions (or mass density for continuous collapse
models). This formal equivalence is ultimately what guarantees that the statistical interpretation
of quantum states survives in such models. Further, it shows that the space-time events (or flashes)
labeling collapse events are formally measurement outcomes. This means that, even though they
are not known to observers, one can make them back-react explicitly on the quantum state as one
would in a feedback scheme. Such a coupling does not create the usual difficulties of mean field
couplings which yield non-linearities at the master equation level. This is a crucial insight for the-
ory building and offers a possible way to consistently couple matter with a classical gravitational
field.
Following this insight, we made the flashes themselves source the gravitational field, as infinitely
massive space-time events. The model we obtained gives the Newtonian gravitational force one
would expect between particles for distances larger than the collapse length scale rC and thus does
not clash with known experiments. However, particles do not attract themselves, contrary to what
one would have guessed from naive semi-classical theories. The model is falsifiable for all possible
values of its two parameters and rC in contrast with what happens with standard collapse models
where only large values of and small values of rC can be experimentally eliminated.

5.2 Link with continuous versions

The present model is mathematically simpler than its continuous predecessors. The continuous
models nonetheless keep several advantages. Notably, they yield master equations in which deco-
herence and unitary gravitational effects are separated and not intertwined as in (16). Further,
the theory space of continuous collapse models is better understood and it is easier to make model
independent statements about e.g. decoherence. A nice feature of continuous collapse models,

8
that gravitational decoherence can be made functionally equal to intrinsic decoherence (singling
out the Disi-Penrose model in the way), is finally absent in the discrete. It is thus important to
understand how continuous collapse models coupled to gravity are (or are not) connected with the
present model.
Unfortunately, continuous models are not simply obtained by scaling the parameters and
rC of the GRW model in an appropriate way. Actually, taking + and rC + with
/rC gives a very special collapse model, quantum mechanics with universal position localiza-
tion (QMUPL) [7, 9], which only has one free parameter. More general collapse models like the
continuous spontaneous localization model (CSL) correspond to a different stochastic unraveling5
of the same (or similar for other continuous collapse models) master equation (6). Thus there
is no straightforward way to go from the present model to the continuous ones. This was not our
objective here. The model we have presented hopefully provides a pedagogical introduction to the
continuous ones and may even be interesting for its own sake.

5.3 Should we believe in it?

Making the flashes gravitate gives a very cheap way to define a gravitational field in the GRW
model. The resulting model preserves the statistical interpretation of the state vector and the
operational toolbox of quantum theory survives. Provided one pays the price of collapse models to
solve the measurement problem, the possibility to construct consistent hybrid dynamics comes for
free. Reciprocally, insisting on some form of stochastic hybrid dynamics naturally leads to models
solving the measurement problem as they contain classical variables which are good candidates
for a primitive ontology. Those are appealing features.
That said, how much should we believe in the model introduced here? As theoretical physics
is currently drowned in wild speculations delusionally elevated to the status of truth, a bit of
soberness and distance is required. The present model most likely does not describe gravity, even
in the Newtonian approximation. It is but a toy model, a proof of principle rather than a proposal
that should be taken too seriously. Nonetheless some lessons survive its ad hoc character:
1. There is no obstacle in principle to construct consistent fundamentally semi-classical theories
of gravity.

2. Collapse models can be empirically constrained by a natural coupling with gravity.

3. A primitive ontology can have a central dynamical role and need not be only passive.
If semi-classical theories of the type presented here can be extended to general relativity in a
convincing way and if robust criteria can be found to make them less ad hoc (see e.g. [5]), then
further hope will be warranted.

Acknowledgments I thank the participants of the 5th International Summer School in Philos-
ophy of Physics that took place in July 2017 in Saig, Germany, for pushing me to write down the
model presented here. This work was made possible by support from the Alexander von Humboldt
foundation and the Agence Nationale de la Recherche (ANR) contract ANR-14-CE25-0003-01.

References
[1] K. Eppley and E. Hannah, Found. Phys. 7, 51 (1977).
[2] D. Kafri, J. M. Taylor, and G. J. Milburn, New J. Phys. 16, 065020 (2014).
5 Different unravelings of the same master equation are empirically indistinguishable: this why many different

collapse models have the same empirical content. However, once one introduces gravity, which is sourced by objects
depending on the unraveling (flashes or continuous fields), the empirical content starts to differ. For the purpose of
extending a theory, what it says about the world (and not only its empirical content) matters.

9
 [3] D. Kafri, G. J. Milburn, and J. M. Taylor, New J. Phys. 17, 015006 (2015).
[4] A. Tilloy and L. Disi, Phys. Rev. D 93, 024026 (2016).
[5] A. Tilloy and L. Disi, arXiv:1706.01856 (2017).
[6] G. C. Ghirardi, A. Rimini, and T. Weber, Phys. Rev. D 34, 470 (1986).
[7] L. Disi, Phys. Rev. A 40, 1165 (1989).
[8] G. C. Ghirardi, P. Pearle, and A. Rimini, Phys. Rev. A 42, 78 (1990).
[9] A. Bassi, K. Lochan, S. Satin, T. P. Singh, and H. Ulbricht, Rev. Mod. Phys. 85, 471 (2013).
[10] A. Kent, Mod. Phys. Lett. A 04, 1839 (1989).
[11] W. Feldmann and R. Tumulka, J. Phys. A: Math. Theor. 45, 065304 (2012).
[12] A. Ney and D. Z. Albert, The wave function: Essays on the metaphysics of quantum mechanics
(Oxford University Press, Oxford, UK, 2013).
[13] V. Allori, S. Goldstein, R. Tumulka, and N. Zangh, Brit. J. Philos. Sci. 65, 323 (2014).
[14] V. Allori, International Journal of Quantum Foundations 1, 107 (2015).
[15] J. S. Bell, Epistemological Letters 9 (1976).
[16] R. Tumulka, arXiv:1102.5767 (2011).
[17] N. Gisin, Phys. Lett. A 143, 1 (1990).
[18] J. Polchinski, Phys. Rev. Lett. 66, 397 (1991).
[19] A. Bassi and K. Hejazi, Eur. J. Phys. 36, 055027 (2015).
[20] B. Mielnik, Comm. Math. Phys. 37, 221 (1974).
[21] D. Drr and S. Teufel, Bohmian Mechanics (Springer Berlin Heidelberg, 2009).
[22] J. Gambetta and H. M. Wiseman, Phys. Rev. A 68, 062104 (2003).
[23] L. Disi, Phys. Rev. Lett. 100, 080401 (2008).
[24] L. Disi, Phys. Rev. Lett. 101, 149902 (2008).
[25] H. M. Wiseman and J. M. Gambetta, Phys. Rev. Lett. 101, 140401 (2008).
[26] M. Derakhshani, Phys. Lett. A 378, 990 (2014).

10