Experimental demonstration of an analogy between optical

non-coherence and irreversibility of heat transport

Aleksandr Meilakhs and Claudio Pastorino

Departamento de Fı́sica de Materia Condensada, GIYA, CAC-CNEA and
Instituto de Nanociencia y Nanotecnologı́a, INN-CONICET-CNEA

Miguel Larotonda
CITEDEF-CONICET
(Dated: April 30, 2025)

1
 Abstract
We probe experimentally a connection between coherence in the context of optical physics and the
irreversibility present in heat transfer through an interface separating two media. The robustness
of the experiment on one hand, and the theoretical description taken from the statistical-mechanics
treatment of the heat tranfer problem, allows for the study of arrow of time’s problem within the
clean and robust framework of optical physics’ experiments. The central aspect of the experiment is
a light beam incident and split at an interface to produce two beams. These beams are subsequently
reflected in two mirrors to produce a second incidence of both beams at the interface, at the same
spot. We used two light sources that are only different by their coherence length. In the case of
the highly coherent light source, a laser, we were able to combine the previously split parts of the
light back into a single beam. This is indicative of the reversibility of the process of coherent light
transmission through the interface. The light source with low coherence length, on the contrary,
does not allow for such a recombination, thus producing an irreversible process. We explain how
the latter case is analogous to the process of heat transport through the interface, thus establishing
an important connection between optical non-coherence and irreversibility of transport phenomena.
Our finding paves the way for the study of fundamental processes in heat transfer and the surge
of irreversibility in the realm of optical physics.

I. INTRODUCTION

One of the most intriguing problems in modern physics is the irreversibility of macroscopic
processes. As it was first noticed in the context of thermodynamics, the transport of heat
from a hot body to a cold one is accompanied with an entropy production. Since entropy can
only grow and never decrease such processes are irreversible [1, 2]. Later, it was found that
the Boltzmann equation, along with other kinetic equations governing the behavior of large
ensembles of particles are irreversible [3]. Thismarks a drastic contrast with the equations
describing individual particles, whether in classical or quantum framework.
The most widespread approach for the derivation of the Boltzmann equation from classical
mechanics is the so-called BBGKY hierarchy [4]. A serious drawback of this method is the
postulate of molecular chaos, as a model for the two-particle distribution function, whose
fundamentals are not properly explained. Another approach to explain entropy growth

2
postulates an initial low entropy state for every process. Equivalently it postulates low-
entropy initial states of the universe [5, 6], which is an ad hoc hypothesis [7] and hence, not
completely satisfactory. Some authors [8] propose modifications of the physical laws, to
explicitly incorporate irreversibility.
Most modern approaches propose the emergence of irreversibility from a time-asymmetric
process, which is an inherent feature of quantum physics [9]. The most studied of such
mechanisms is decoherence [10, 11]. However, the decoherence theory have serious intrinsic
difficulties [12].
In one of our previous works [13], the problem of irreversibility of heat transport through
an interface between two media was addressed. It is known that a temperature jump occurs
at the interface when heat flows through two different media. The proportionality coefficient
between the heat flux and the temperature jump is usually called the Kapitza resistance.
Currently, the Kapitza resistance has become a wide area of research, not only for its
theoretical interest, but also for its technological relevance in a wide variety of thermal
management applications. Some works investigate the dynamics of the crystal lattice at
the interface through computer simulations[14–19], while there is also a great number of
works with analytical approaches [20–22]. Other studies deal with phonon kinetics at the
interface with Boltzmann theory [23–26] or the nonequilibrium Green’s function method
[27–29]. Measurements of values of Kapitza resistance for different pairs of materials were
also performed [30, 31]. Additionally, the problem of Kapitza resistance was studied in a
broader context than phonon transport, such as electron transport between semiconductors
[32] and molecular transport at a liquid-vapor interface [33].
After the discovery of the Kapitza resistance, it was very soon realized that the temper-
ature jump is due to the reflection of phonons at the interface [34]. Since phonon reflection
at a solid-solid interface is a linear process, the Kapitza resistance effect can be used as a
testbed to shed light on the mechanism that leads to irreversibility. In contrast with the
time-dependent and highly non-linear Boltzmann equation, its analog at the interface be-
tween two materials, the matching equation for the distribution functions, [35, 36] is linear
and static
In Ref. [13] the interfacial phenomenom of Kapitza resistance was related to the concept
of non-coherence. It was suggested that if particles incident on the interface from different
sides are phase-correlated (coherent regime) then their transmission through the interface

3
occurs reversibly. On the contrary, if there is no phase correlation between particles at both
sides of the interface (non-coherent regime), which is the naturally occurring situation, the
transmission becomes irreversible. In that work, an optical experiment was proposed to
clearly establish this connection between non-coherence and ireversibility. In the present
manuscript we perform the suggested experiment and we empirically show the validity of
this connection and the , prediction of the theory [13]. The time irreversibility shown
by non-coherent transmission through an interface, is put to test with photons incident
on an air-glass interface with a mirror arrangement and two types of light sources. In
section II we present the theoretical framework within the context of light transport, an ideal
experiment and a feasible experimental realization. In section III, the used experimental
setup is described in detail. Finally, we devote section IV to present the results and a final
discussion is provided in section V.

II. THEORY

A. Preambula

We studied the light transport through an interface of two materials under two differ-
ent coherence conditions of the waves, as sketched in Fig. 1. A light beam is split at a
partially reflecting interface S. The reflected part of the light travels through one of the
optical media and bounces at a high reflectivity surface M1 , which is parallel to the inter-
face. Concurrently, the refracted (transmitted) part of the light travels through a different
medium and hits a second highly reflective surface M2 . After their respective reflections,
both beams are recombined at the interface. At this point, the coherence condition of the
light imposes a limit on the reversibility of the process, i.e., whether the light can get to-
tally or partially recombined into a single output beam. Since the materials have different
refractive indices, their relative thicknesses must be adjusted to compensate for the differing
propagation lengths caused by the refraction and reflection angles.

4
Figure 1. General scheme for the study of coherence conditions that lead to reversible and irre-
versible processes of photons at an spatialinterfacial boundary. Light incident on an interface S
between two optical media of different refraction indexes n1 and n2 gets partially reflected. The
refracted and reflected beams are re-routed back to the interface by means of the mirrorsreflecting
surfaces M1 and M2 , where they interact in different ways depending on the coherence conditions
of the light.

We consider linearly polarized waves, with a polarization perpendicular to the plane of

incidence. If the light is incident on the interface from one side, the amplitudes of transmitted
(refracted) and reflected light are given by the Fresnel formula[Hecht? Miguel, a word here?]:

sin(ψ − ϕ) 2 sin ψ cos ϕ

rS = , tS = (1)
sin(ϕ + ψ) sin(ϕ + ψ)

Here, ϕ and ψ are respectively the angles between the propagation direction of the wave
and the axis perpendicular to the interface (Fig. 1). The interface is a boundary between
two optical materials of refractive indices n1 and n2 . The Snell’s law of refraction imposes
the condition for the incident and refracted angles ϕ and ψ: n1 sin ϕ = n2 sin ψ. For the
particular condition of an interface between air and optical glass, n1 ≈ 1 and n2 = n, hence
sin ϕ = n sin ψ.
If light is incident at the interface from both sides with amplitudes AI1 and AI2 , and the
angles of incidence are matched by Snell’s law, the amplitudes of departing waves AD1 , AD2 ,
can be found by
   
AD1 AI1
  = A . (2)
AD2 AI2

5
where A is the matrix of transfomation of amplitudes defined by
 
1 sin(ψ − ϕ) 2 sin ϕ cos ψ
A=  . (3)
sin(ϕ + ψ) 2 sin ψ cos ϕ sin(ϕ − ψ)

The values of the columns in matrix A are both obtained from equation (1), with the second
column assuming that the light is incident from the right direction.
In [13] it is shown that, if we normalize the amplitudes such that its squares at a given
side and direction are equal to the intensity at a given side and direction, i.e. |A|2 = I, the
transformation matrix becomes unitary and it is given by the following expression
 √ 
1 sin(ψ − ϕ) 2 sin ψ cos ψ sin ϕ cos ϕ
U= . (4)
sin(ϕ + ψ) 2√sin ψ cos ψ sin ϕ cos ϕ

sin(ϕ − ψ)

It is well known that unitary matrices form a group, and for any unitary matrix there exists
an inverse matrix, which is also unitary. Throughout the remainder of the text, we will use a
system of units in which squares of amplitudes represent intensities, and the transformations
of amplitudes are given by unitary matrices.
If the transmission is non-coherent, the transformation of intensities is given by the
matrix[13]
 
2
1 sin (ψ − ϕ) 4 sin ψ cos ϕ sin ϕ cos ψ
T = 2
 . (5)
sin (ϕ + ψ) 4 sin ψ cos ϕ sin ϕ cos ψ 2
sin (ϕ − ψ)

Each element of this matrix is obtained by taking a square of a module of an element of the
unitary matrix (4): |Uij |2 = Tij , i, j takes the values of one or two. Matrices of this type
are called bistochastic. Such matrices lack inverses within this class of matrices. That is the
mathematical essence of the phenomenon that is to be demonstrated: the correspondence
of loss of reversibility with the loss of coherence.
In the presented notation, the problem can be written as follows. We initially have a
light beam at only one side of the interface. Let us define its amplitude as one:
   
A 1
 I1  =   . (6)
AI2 0

Upon incidence at the interface, it undergoes a transformation given by matrix U (Eq.

4). Then the light is split into two parts, each of which evolves freely so that, before the next

6
incidence on the interface, their phases will undergo the transformation exp(ikl). Here k is
a corresponding wave vector and l is the distance that light travels between two instances
of incidence at the interface. When the beams meet at the interface for the second time,
their amplitudes are again transformed by U (see Fig. 1). So we have the following final
amplitudes:
     
ik1 l1
A e 0 1
 F 1 = U  U  . (7)
AF 2 0 eik2 l2 0

The intensities are given by the squares of the amplitudes |AF 1 |2 , |AF 2 |2 . These intensities
are ultimately measured in our experiment.
If we multiply both amplitudes on the same quantity with module 1, the resulting inten-
sities would not change. Therefore, the important quantity that characterizes the matrix
of free propagation is exp i(k1 l1 − k2 l2 ), i.e, the phase difference between the two paths of
propagation.
Since U, given by Eq. (4), is symmetric then U † = U. Here the dagger symbol represents
the Hermitian conjugation. We recall that the defining property of unitary matrices is
U † U = E, where E is the identity matrix. Combining these two properties we find that
U 2 = E.
If k1 l1 − k2 l2 = 2πN , for any natural value of N (N ∈ N), the matrix of free propagation
is equivalent to the identity matrix. This means that the product of the three matrices in
the expression (7) is equal to the identity matrix: UEU = E. Therefore,the system that went
through these series of transformations will come back to its initial state. More concretely,
the two beams initially split at the interface, would merge back into a single one. That is
the physical essence of the transformation being reversible.
In the non-coherent case, the intensities are given by
   
I 1
 F 1 = T 2   . (8)
IF 2 0

The intensities are not dependent on the optical path, they are strictly greater than zero,
and lower than one. It means that non-coherent beams cannot be combined back into
a single beam, which is the physical manifestation of the irrreversibility of non-coherent
transformations .

7
 B. Scheme of the ideal experiment

Our goal is to distinguish the coherent regime from the non-coherent one. The most
convenient way of observing the coherence is by measuring the visibility of the optical fringes,
defined by the expression
Imax − Imin
ν= . (9)
Imax + Imin
Experimentally, these maximum and minimum intensity values can be obtained by
slightly varying the size of the air gap.
We will show experimentally that the visibility ν is zero in a non-coherent, irreversible
regime, whereas it can reach values close to unity for a reversible, coherent case. In real
conditions the visibility of a coherent light beam may be ultimately limited by experimental
factors such as a distorted spatial mode, or scattering in the reflecting surfaces. Before
reaching this condition, several aspects of the particular setup must be considered. In what
follows, we describe the experimental conditions needed to obtain the highest possible degree
of reversibility (maximum visibility) with the available light source.
In an “ideal” experiment, a light beam incident at a boundary between two isotropic
media is split into a reflected beam and a transmitted one. These two beams are totally
reflected on parallel surfaces and re-routed to a common point at the boundary surface. In
order to maximize visibility, the amplitudes of both beams must be equal. These amplitudes
are given by the reflection and transmission coefficients that can be obtained from Eq. (1)
Let us denote the elements of the unitary transformation matrix U given by the expression
in Eq. (4) as
 
a b
U = . (10)
b −a

The minimum intensity is obtained when the waves are incident on the interface exactly in
antiphase. Therefore, the full transformation matrix (7) reads:
     
2 2
a b 1 0 a b a −b 2ab
   = . (11)
b −a 0 −1 b −a 2ab −a2 + b2

The minimum is then obtained for a2 = b2 , which means that the intensities are split equally
in both directions.

8
 This results in the expression sin2 (ψ − ϕ) = 4 sin ψ cos ψ sin ϕ cos ϕ. We use the substitu-
tion sin ψ = n1 sin ϕ/n2 and, after some manipulations, we find
s p
n22 + n21 − 9/8(n22 − n21 )
sin ϕ = . (12)
2
This gives the incidence angle for a balanced reflection and refraction amplitudes, as a
function of the refractive indices on both sides of the interface.

C. Scheme of the real experiment

However, since mirrors do not have perfect reflectivities, part of the light is absorbed by
them. In order to include these losses, a more realistic transformation of the amplitudes
between incident beams on the interface must include the multipliers that affect the absolute
values of the amplitudes:
     
ik1 l1
A re 0 1
 F 1 = U  1 U  . (13)
AF 2 0 r2 eik2 l2 0
and for a non-coherent transformation, we obtain
     
I R 0 1
 F 1 = T  1 T  . (14)
IF 2 0 R2 0
√ √
Here,r1 = R1 , r2 = R2 , and the parameters R1 and R2 must be determined experimen-
tally.
The conditions for maximal visibility is analogous to that of Eq. (11):
     
2 2
a b r 0 a b a r1 − b r2 ab(r1 + r2 )
  1  = . (15)
2 2
b −a 0 −r2 b −a 2ab(r1 + r2 ) −a r1 + b r2
We find that the condition of maximum visibility , which is a2 r1 −b2 r2 in this case, has the
same physical meaning as that of the ideal case: at the recombination of beams, intensities
on both sides should be equal.
When we take into account the absorption of light by mirrors, we can not obtain the
same intensity at the output as at the input. Therefore, we do not aim for the complete
reversibility of the whole experiment. What we want to demonstrate is the reversibility of
the process of transmission/reflection of light at the interface between the air and the glass.
We achieve this, by showing that all the light that is left in the system after the reflections
at the mirrors, is aimed in one direction.

9
III. THE EXPERIMENT

A. Experimental Setup

The light source that we define as a “coherent” wave is a commercial Helium-Neon laser
(Research Electro Optics model 30989, λ = 632.8 nm). The “incoherent” light source is a red
laser diode driven below the threshold current, with a peak emission wavelength at 650 nm.
In practice, there is no absolute coherent or incoherent sources. Rather, all light waves show
a certain degree of temporal coherence, given by the average correlation between the wave
and a delayed copy of itself. The parameter τc (correlation time) gives a characteristic delay
for which this correlation is maintained. Correspondingly, the coherence length ℓc = cτc
is the distance the wave travels during the time τc . At the single photon level, this effect
can be related to the length or duration of the photon wavepacket. The He-Ne laser has
a coherence length between 10 and 30 cm, when used in the operational regime., whileOn
the other hand, the spontaneous emission from the laser diode when it is operated below
the laser threshold, due to its large bandwidth has a reduced coherence length of less than
100 µm, due to its large band width.

Each light source was sent to a multi-axis, high-resolution fiber coupler (Thorlabs PAF-
X-11-B) that allow light to be launched into a Single Mode Fiber (SMF). Therefore, the
input light beam for the experiment was generated by coupling either the coherent or the
incoherent source to a 2 meter SMF patchcord and sending it to the experiment using an
adjustable focus fiber collimator (Thorlabs CFC-11X-B). The experiment end of the fiber
patchcord was left untouched throughout the whole experiment, while the opposite end could
be switched between the two fiber couplers, to select either the coherent or the incoherent
light source. The purpose of the coupling and decoupling into a SMF is twofold: to obtain
good beam quality by spatial filtering with the single transverse mode propagation of the
fiber, and also to ensure a similar alignment for both light sources (Figure 2).

Using the adjustable lens of the collimator, we loosely focus the output beam down to a
400 µm waist at one meter from the lens, where we placed the interference experiment. A
polarizing beamsplitter cube (Thorlabs PBS202) allows us to define a specific polarization
for the input beam.

We use a 25 mm square, 6 mm thick N-BK7 borosilicate crown-glass optical window as

10
Figure 2. Experimental setup. Using a single mode fiber (SMF), we select either a coherent or an
incoherent light source as the beam probe. Light is decoupled from the fiber using an adjustable
focus fiber collimator (AFFC). The input polarization is selected with a polarizing beamsplitter
cube (PBS). The two possible outputs are monitored with a pair of Si- biased photodetectors
(PD); the intensity of the output, that propagates inside the glass is partially collected after an
additional reflection in M2 , and a transmission through the interface S. The coherent light source
is a He-Ne laser, while the incoherent light source is laser diode (LD) current-pumped below the
lasing threshold. Each beam is sent to a fixed-lens fiber coupler (FC). This allows us to select any
of the light sources by plugging the fiber patchcord into the corresponding fiber coupler. The size
of the gap g controls the spatial overlap between the two beam paths.

one of the propagating mediums (n2 = 1.515). Both surfaces are optically polished and
one of them has an aluminum-coating mirror deposited on it (M2 ). The uncoated surface
is the partially reflecting interface S between the BK7 glass and air. For the reflection in
air (M2 ), we use the first surface of a circular, 25 mm diameter, silver coated mirror. In
this configuration, the optimum incidence angle under ideal reflectivity conditions can be
calculated using the expression (12), which numerically givesresults in the value ϕ = 78.57◦ ,
for n1 = 1 and n2 = 1.5151.

Both mirrors are mounted on precision angular mounts, to ensure parallelism between
surfaces. Additionally, the silver mirror mount M1 , is placed on a manual, micrometer-

11
resolution translation stage, to allow for fine adjustment of the air gap g between the optical
surfaces., whichThis controls the overlap between both beams. By displacing laterally the
two mirrors as depicted in figure 2, we gain access to the two possible outputs of the light,
after the interaction between the reflected beam and the transmitted beams.

B. Preliminary characterization of surface reflectivities

For a coherent wave, complete reversibility with a single interaction can be expected,
if the splitting ratio at the interface is close to 50% and there are similar losses on both
paths. According to the previous discussions, a reflectivity of RS = 0.5 can be obtained at
an interface between air (n1 ≈ 1) and BK7 glass (n2 = 1.515) for an s-polarized field and
an incidence angle as large as 78.6◦ . We are assuming here that wavelengths fall within the
range between 633 nm and 650 nm. This is a fairly large incidence angle, and although the
transverse profile of the beam is small, it is difficult to avoid partial blocking and distortions
from the edges of the mirrors. Fortunately, a mismatch between the reflectivities of mirrors
M1 and M2 helps to avoid this problem. We measured the mirror reflectivites for the actual
experimental conditions: The silver mirror M1 has a reflectivity of R1 = 0.91 for an incidence
angle θi ≃ 75◦ , while the reflectivity of the aluminum mirror, behind the BK7 slab M2 , was
measured to be R2 = 0.61 for an internal incidence angle of 50◦ ., which This is, roughly, the
complementary angle of the refracted ray inside the glass, for incidence angles between 74◦
and 78◦ . This means that the beam propagating inside the glass suffers an increased loss,
and therefore the incidence angle has to be adjusted in the experiment to obtain a larger
transmissivity, which compensates for this loss imbalance, as discussed in Sec. II C.

C. Balanced loss configuration

Experimentally, we set the incidence angle at a value of θi = 76.1◦ that maximizes the
fringe visibility. Such angle leads to a ratio between transmissivity and reflectivity given by
the Fresnel coefficients of TS /RS = 1.31 and a refraction angle θr = 39.8◦ . Substituting the
measured reflectivities R1 and R2 into the expression (15), we find that the theoretical ideal
incidence angle would be 76.8◦ ., whichRemarkably, this is very close to the experimental
value of 76.1◦ (See Fig. 3).

12
 1 0.75

)
0.5
0.9

Intensity (
0.25

visibility 0.8 0
Phase difference

0.7

0.6

60 65 70 75 80 85
incidence angle [deg]

Figure 3. Measured visibility as a function of incidence angle atin radians. The maximum is located
at 1.34 radians, or 76.8◦ . TheIn orange circular symbol is depicted point is the experimentally
obtained the visibility value obtained experimentally,. It was measured for an incidence angle of
(76.1±0.6)◦ . The actual visibility is affected by several factors, other than the balance of intensities,
such as imperfections of the beam wavefront and scattering at the surfaces. The inset shows the
Inset: calculated complimentary modulation of the intensity at both outputs, as a function of the
phase difference between the two paths., This calculation considers considering a perfectly coherent
input beam. The normalized intensity departs from unity due to reflection losses.

For such configuration, an air gap of a nominal thickness of 1.24 mm equalizes the distance
traveled by both beams in the direction parallel to the interface. In this situation, depending
on the coherence of the source, total recombination of the beams into a single beam may be
obtained at the output., whichWe point out that this must can be understood as a signature
of reversibility. Furthermore, the light can be arranged to exit the deviceexperimental setup
as a single beam through either of the outputs.: dDepending on the phase difference between
the two paths, a coherent combination of the beams will lead to constructive interference
alternatively on the two outputs [C: previous sentence may need more rephrasing],. This is
shown as depicted in the inset of figure 3.

IV. RESULTS

13
 When the coherent source is used, aAfter a careful adjustment of theis separation g and
a furthecarefulr mirror alignment, the coherent source gives rise to a complementary single-
fringe interference pattern, can be observed in the two outputs. We use two large area
biased Si photodiodes (Thorlabs DET-36A) to register the light intensity at both outputs.
The output that propagates in air is fully collected by the detector, while the one that
propagates inside the glass is attenuated due to an internal reflection in M2 , and a partial
transmission from glass to air at the interface., which in overallThis reduces the intensity
by an overall factor of R2 TS = 0.346.
Figure 4 shows the normalized intensities for the input and output beams 1 and 2, as a
function of the optical path length variation (OPD), obtained by changing the size of the air
gap g between the two mirrors. When the coherent source is used, the relative phase between
the two beams produces a complete recombination of the light in a single beam, which exits
the experiment through either of the outputs. The coherent light signals are shown with
full circles in Fig. 4. For OPD differences that are integers of the light wavelength (N λ,
N ∈ N), all the available light exits through the air side of the interface (Out2 in Fig. 2),
while for path differences that are half integers of the wavelength (N + 1/2)λ, the light
is almost completely recombined on a single beam at the glass side (Out1 ). Due to the
extended length of the glass slab, we can sample both outputs: Out2 is directly coupled to a
photodiode, while a portion of Out1 exits the glass slab and is also collected by a photodiode
after a reflection in M2 and a successive transmission at the interface. The fringe visibility,
calculated using Eq. (9) is above 97.5% for both outputs.
In contrast, when the experiment is carried out using the incoherent light source , no
intensity modulation is obtained, and there is a single distribution of light at the outputs,
regardless of the phase condition given by the variation of the OPD. These are shown with
"+" symbols in Fig. 4.
Furthermore, taking into account the additional reflections and refractions suffered by
both outputs, the ideal mean throughput at each output can be calculated as:

Out1 = R1 (RS TS R1 + RS TS R2 ) R2 TS = 0.118,

(16)
Out2 = R1 RS 2 R1 + TS 2 R2 = 0.334.

These values exhibit good agreement with the normalized intensities obtained using the
incoherent source and are locatedsituated at the midpoint of the intensity excursions ob-

14
tained with the coherent light source. For comparison, the theoretical predictions for the
intensities at both outputs are also shown in full lines, and these also align well with the
experimental data.

1
Coherent Incoherent
0.9 Input
Out1
Out2
0.8
Normalized intensity

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Optical path length variation [μm]

Figure 4. Measured intensities at the input and at the two complimentary outputs for both,
coherent and incoherent light sources, as a function of the variation of the optical path length.
The modulation was generated by changing the air gap width between the two mirrors. The
coherent source shows almost ideal time reversibility when losses at the mirrors are taken into
account: all the light left in the system is combined into onea single beam (full circles). In
contrast, when the incoherent source is used no modulation is present (+ symbols). This shows
that the recombination of the two non-coherent beams always gives one result, which is half the
maximum possible intensity for the coherent beams. The In thin continuous lines are the theoretical
predictions are shown. These were obtained by using the experimentally determined incidence
angles and mirrors’ reflectivities measured experimentally.

Figure 4 also shows a temporal intensity modulation in our coherent source (He-Ne laser),
which is present both at the input and at the outputs [C: It is mentioned temporal but the
Figure invites to think in lengths and not in times(?)]. This effect is produced by the

15
longitudinal modes of the laser cavity sweeping through the gain profile due to temperature-
induced variations of the laser cavity length [37]. When lasing, these modes which emit with
orthogonal polarization from each other and. as a result, there is a polarization modulation
that is observed when the laser polarization is fixed outside the cavity.
It is worth to mention that, although this experiment has been carried out with intense
light sources, similar results can be obtained at the single photon level. The temporal
coherence, namely the length of the photon wavepacket, is inversely related to its spectral
dispersion. The reversibility of the splitting at the interface within the first interaction
is related to the interference at the second interaction: the fringe visibility will survive
at full strength (reversible process) if the optical path difference is much smaller than the
coherence length of the photon [38]. If the path difference exceeds the duration of the photon
wavepacket, both contributions of the photon at the transmitted and reflected superposition
state will arrive separately at the interface and the interference vanishes.
In this single-photon picture the matrices T , U have a different interpretation. The
elements of T may be interpreted as the probabilities of photon transmitting through the
interface and the elements of U as probability amplitudes of transmission through the in-
terface. The difference between the coherent and incoherent regimes is understood in this
case, as the regime where the transition of a photon is described by probability amplitudes
with specific phase values and the regime where only the absolute value of the probability
amplitude matters.[Miguel. What do you think of this? Is it clear?]

V. FINAL DISCUSSION AND CONCLUSIONS

In this work we have carried out the followingan experiment that shows a strong con-
nection between irreversibility and coherence and provide a bridge between the frameworks
statistical mechanics and optical physics. In the experiment, : two parts of a light beam,
split by the incidence on the interface, were incident again on the interface, and met at the
same spot. In the condition such that the difference between the lengths of their optical
paths is much greater than the coherence length of light (non-coherent regime), the in-
tensities measured on the outputs agree with those calculated by the formula (14). In the
opposite case, when the difference between the lengths of optical paths was much shorter
than the coherence length of light (coherent regime), the measured intensities that agree

16
with those calculated by the formula (13). In the latter case, by varying the difference be-
tween the lengths of optical paths, we were able to combine the split parts of the light back
into a single beam, which confirms the reversibility of the process in the coherent regime.
The non-coherent regime, on the contrary, does not allow for such a recombination, thus
producing an irreversible process.
This seemingly simple optical experiment, is interpreted as an analogy to the process of
heat transport through an interface. This analogy allows for studying and understanding
the emergence of irreversibility in statistical-mechanics systems, in the powerful and clean
framework of optical physics. If we examine different types of quantum particles such as
phonons, electrons, and photons, transitioning through the interface, the equations that
connect amplitudes of incident waves and departing waves take, universally, the form[13]
   
A A
 D1  = U  I1  , (17)
AD2 AI2

where U is a unitary matrix. The expressions for the elements of U are different for the
different types of particles and the different models, but the unitarity is universal.
From this equation, we derive another expression, that relates the distribution functions
of particles departing from the interface with the distribution functions of particles incident
on the interface
   
fD1 fI1
 =T  . (18)
fD2 fI2

Distribution functions f -s are defined as the mean number of particles in the given state, T
is a matrix in which the elements are probabilities of transmission or reflection of particles at
the interface. The elements of T are calculated as squares of absolute values of the elements
of U from equation (17). We call this equation a matching equation for the distribution
functions.
To calculate the Kapitza resistance, we accompany the matching equations with ther-
malization conditions on each interface side. These conditions are needed to connect the
distribution functions of particles with the temperatures at both sides of the interface. These
differ by the value of the Kapitza temperature jump. In the setting of heat transport, the
distribution functions of particles are not given directly, they should be expressed as func-
tions of temperatures. In an optical setting, however, we do not have physically meaningful

17
analogous conditions. Instead, we point out that the intensity of light is the mean number
of photons multiplied by the flux associated with each photon, I = f ch̄ω, which means the
distribution of particles can be measured directly. Therefore, we only replicate the matching
equations in the optical experiment.
We obtain the matching equations by finding an expression for the matrix T . To this end,
we need to find an explicit form for matrix U. The natural way to find U is by solving the
following problem: let the particle be incident on the interface from one side with probability
one, and find the probability amplitudes of its transmission and reflection. The solution for
the case of incidence from a side 1, for example, gives the values for the first column of
U, while the solution for the case of incidence from side 2, gives the values of the second
column (see Fig. 1).[C and A: may be label 1 and 2 are needed in fig. 1] This physical
situation can be also stated in a different way. Let us assume that some particle is departing
from the interface on the given side with certainty, and find the probability amplitudes of
its incidence from each side. By solving this problem for both sides of departure, we get the
equation
   
AI1 AD1
  = U†  , (19)
AI2 AD2

which is mathematically equivalent to the equation (17). Indeed, it can be obtained from
it, by multiplying both sides of Eq. (17) by U † .
Intuitively, the equation (19) is a time-reversed version of equation (17), since the latter
describes the splitting of waves at the interface, and the former the merging of two waves
at the interface. Since wave equations are time-symmetric, both are equivalent. If we apply
to Eq. (19)the same procedure that we used for (17), we obtain the following matching
equation for the distribution functions
   
f f
 D1  = T  I1  . (20)
fD2 fI2

By solving these equations with the same thermalization conditions, we find that for the
same value of temperature difference between sides, we get the opposite values of heat flux.
This means that heat flows from the colder body to the hotter body. We indeed made the
time reverse.

18
 However, equations (18) are not equivalent to (20), since T ̸= T −1 . We started with two
equivalent equations,namely Eqs. 17 and 19, applied the procedure to them, and obtained
two non-equivalent equations, i.e. Eqs. 18 and 20. Thus we have lost reversibility during
the derivation of the matching equation.
In Ref. [13] was conjectured that the juncture in the derivation where the time symmetry
is disrupted, is the assumption of the absence of phase correlation between particles that
are incident on the interface from opposite sides. That allows for averaging over amplitude
phases which, in turn, allows for the description of processes in terms of probabilities and
their bistochastic transformations (5) instead of amplitudes and their unitary transforma-
tions (4). As we have pointed out before, bistochastic matrices (ingnoring exceptional cases)
do not have inverses, which means that they represent irreversible transformations.
The presented experimental results entirely confirm this suggestion. In the coherent
regime the phase information is present, complete, and can be measured exactly, and we
have found perfect reversibility. In constrast, for the non-coherent regime all the informa-
tion about phases is lost and, indeed, we found that this state is irreversible. This is proved
because the transformations of intensities at the interface are successfully described by bis-
tochastic matrices (5), obtained from unitary matrices by squaring absolute values of their
elements. Since intensities I and the mean number of particles f are connected by formula
I = f ch̄ω, they transform identically. This is exactly what we observe by comparing (8)
with (18) and bearing in mind the former is a two-step transformation.
For this experiment, we have specifically employed light sources with close wavelengths
and intensities (see Section III), and the only feature that accounts for the qualitative
difference in the results is the coherence length of the sources.
The main difference between the experimental setting studied in this work and the case
of heat transport is that the former allows to explore both, the coherent and the incoherent
regimes. In a heat transport setup, the indistinguishability condition that eventually leads
to interference (i.e. coherence) and reversibility is very hard to achieve. This condition
involves the interaction of particles that are incident on the interface from opposite sides
and hence, in a classical scenario, cannot have phase correlation. As a consequence, heat
transport occurs permanently in a non-coherent regime and it is always irreversible.
We hope that this powerful analogy between coherence and reversibility, and our suc-
cessful mapping from the physics of heat transfer to that of optics of beams incident in

19
an interface, can provide a bridge between these two research areas and motivate further
experimental and theoretical work. Possibly, this analogy can be stretched far forward from
the interfacial phenomena and explain the irreversibility of the Boltzmann equations.

[1] L. D. Landau and E. M. Lifšic, Landau, vol. 1 of Course of theoretical physics. Oxford:
Pergamon Press, 3. ed., rev. and enl. ed., 1994.
[2] D. Kondepudi, Modern Thermodynamics. Chichester: Wiley, second edition ed., 2015. De-
scription based upon print version of record.
[3] L. D. Landau and E. M. Lifshitz, Course of Theoretical Physics. [s.l.]: Elsevier Reference
Monographs, 3. aufl. ed., 2013. Description based upon print version of record.
[4] S. Harris, An introduction to the theory of the Boltzmann equation. Dover books on physics,
Mineola, N.Y. [u.a.]: Dover Publications, 2004. Unabridged republ. of the work originally
publ. by Holt, Rinehart and Winston, New York, 1971.
[5] S. M. CARROLL and J. CHEN, “Does inflation provide natural initial conditions for the
universe?,” International Journal of Modern Physics D, vol. 14, pp. 2335–2339, Dec. 2005.
[6] R. M. Wald, “The arrow of time and the initial conditions of the universe,” Studies in History
and Philosophy of Science Part B: Studies in History and Philosophy of Modern Physics,
vol. 37, pp. 394–398, Sept. 2006.
[7] E. Cohen and T. Berlin, “Note on the derivation of the boltzmann equation from the liouville
equation,” Physica, vol. 26, pp. 717–729, Sept. 1960.
[8] L. Maccone, “Quantum solution to the arrow-of-time dilemma,” Physical Review Letters,
vol. 103, p. 080401, Aug. 2009.
[9] G. C. Ghirardi, A. Rimini, and T. Weber, “Unified dynamics for microscopic and macroscopic
systems,” Physical Review D, vol. 34, pp. 470–491, July 1986.
[10] L. S. Schulman, Time’s arrows and quantum measurement. Cambridge: Cambridge University
Press, 1997. Title from publisher’s bibliographic system (viewed on 05 Oct 2015).
[11] E. Joos, ed., Decoherence and the appearance of a classical world in quantum theory. Physics
and astronomy online library, Berlin: Springer, 2. ed. ed., 2003. Hier auch später erschienene,
unveränd. Nachdr.
[12] S. L. Adler, “Why decoherence has not solved the measurement problem: a response to

20
 p.w. anderson,” Studies in History and Philosophy of Science Part B: Studies in History and
Philosophy of Modern Physics, vol. 34, pp. 135–142, Mar. 2003.
[13] A. Meilakhs, “Transmission of waves and particles through the interface: reversibility and
coherence,” Annals of Physics, vol. 466, p. 169686, 2024.
[14] K. Sääkilahti, J. Oksanen, J. Tulkki, and S. Volz, “Role of anharmonic phonon scattering
in the spectrally decomposed thermal conductance at planar interfaces,” Physical Review B,
vol. 90, p. 134312, Oct. 2014.
[15] N. Yang, T. Luo, K. Esfarjani, A. Henry, Z. Tian, J. Shiomi, Y. Chalopin, B. Li, and G. Chen,
“Thermal interface conductance between aluminum and silicon by molecular dynamics sim-
ulations,” Journal of Computational and Theoretical Nanoscience, vol. 12, pp. 168–174, Feb.
2015.
[16] K. Bi, Y. Liu, C. Zhang, J. Li, M. Chen, and Y. Chen, “Thermal transport across symmetric
and asymmetric solid–solid interfaces,” Applied Physics A, vol. 122, Sept. 2016.
[17] A. Alkurdi, S. Pailhès, and S. Merabia, “Critical angle for interfacial phonon scattering:
Results from ab initio lattice dynamics calculations,” Applied Physics Letters, vol. 111, Aug.
2017.
[18] R. R. Kakodkar and J. P. Feser, “Probing the validity of the diffuse mismatch model for
phonons using atomistic simulations,” Physical Review B, vol. 95, p. 125434, Mar. 2017.
[19] Z. Huang, C. Huang, D. Wu, and Z. Rao, “Influence of chemical bonding on thermal contact
resistance at silica interface: A molecular dynamics simulation,” Computational Materials
Science, vol. 149, pp. 316–323, June 2018.
[20] D. A. Young and H. J. Maris, “Lattice-dynamical calculation of the kapitza resistance between
fcc lattices,” Physical Review B, vol. 40, pp. 3685–3693, Aug. 1989.
[21] L. Zhang, P. Keblinski, J.-S. Wang, and B. Li, “Interfacial thermal transport in atomic junc-
tions,” Physical Review B, vol. 83, p. 064303, Feb. 2011.
[22] A. Meilakhs, “Phonon transmission across an interface between two crystals,” Nanosystems:
Physics, Chemistry, Mathematics, pp. 971–982, Dec. 2016.
[23] A. Majumdar and P. Reddy, “Role of electron–phonon coupling in thermal conductance of
metal–nonmetal interfaces,” Applied Physics Letters, vol. 84, pp. 4768–4770, June 2004.
[24] S. Merabia and K. Termentzidis, “Thermal conductance at the interface between crystals using
equilibrium and nonequilibrium molecular dynamics,” Physical Review B, vol. 86, p. 094303,

21
 Sept. 2012.
[25] K. Alaili, J. Ordonez-Miranda, and Y. Ezzahri, “Simultaneous determination of thermal diffu-
sivity and thermal conductivity of a thin layer using double modulated thermal excitations,”
Journal of Applied Physics, vol. 126, Oct. 2019.
[26] G. Varnavides, A. S. Jermyn, P. Anikeeva, and P. Narang, “Nonequilibrium phonon transport
across nanoscale interfaces,” Physical Review B, vol. 100, p. 115402, Sept. 2019.
[27] Z. Tian, K. Esfarjani, and G. Chen, “Enhancing phonon transmission across a si/ge inter-
face by atomic roughness: First-principles study with the green’s function method,” Physical
Review B, vol. 86, p. 235304, Dec. 2012.
[28] J.-S. Wang, N. Zeng, J. Wang, and C. K. Gan, “Nonequilibrium green’s function method for
thermal transport in junctions,” Physical Review E, vol. 75, p. 061128, June 2007.
[29] J.-S. Wang, J. Wang, and J. T. Lü, “Quantum thermal transport in nanostructures,” The
European Physical Journal B, vol. 62, pp. 381–404, Apr. 2008.
[30] R. M. Costescu, M. A. Wall, and D. G. Cahill, “Thermal conductance of epitaxial interfaces,”
Physical Review B, vol. 67, p. 054302, Feb. 2003.
[31] N. Ye, J. P. Feser, S. Sadasivam, T. S. Fisher, T. Wang, C. Ni, and A. Janotti, “Thermal
transport across metal silicide-silicon interfaces: An experimental comparison between epitax-
ial and nonepitaxial interfaces,” Physical Review B, vol. 95, p. 085430, Feb. 2017.
[32] A. P. Meilakhs, “Electronic kapitza conductance and related kinetic coefficients at an interface
between n-type semiconductors,” Journal of Physics: Condensed Matter, vol. 36, p. 045302,
Oct. 2023.
[33] C. Pastorino, I. Urrutia, M. Fiora, and F. Condado, “Heat flow through a liquid–vapor in-
terface in a nano-channel: the effect of end-grafting polymers on a wall,” Journal of Physics:
Condensed Matter, vol. 34, p. 344004, June 2022.
[34] I. M. Khalatnikov JETP, vol. 22, p. 687, 1952.
[35] A. P. Meilakhs, “Nonequilibrium distribution function in the presence of a heat flux at the
interface between two crystals,” Physics of the Solid State, vol. 57, pp. 148–152, Jan. 2015.
[36] A. Meilakhs and B. Semak, “Calculation of kapitza resistance with kinetic equation,” physica
status solidi (b), vol. 258, June 2021.
[37] T. Niebauer, J. E. Faller, H. Godwin, J. L. Hall, and R. Barger, “Frequency stability measure-
ments on polarization-stabilized he–ne lasers,” Applied optics, vol. 27, no. 7, pp. 1285–1289,

22
 1988.
[38] R. Loudon, The quantum theory of light, ch. 6. OUP Oxford, 2000.

23