Generalization of Fourier's Law Into Viscous Heat Equations

PHYSICAL REVIEW X 10, 011019 (2020)
Generalization of Fourier’s Law into Viscous Heat Equations

Michele Simoncelli ,1,* Nicola Marzari ,1 and Andrea Cepellotti 2
1
Theory and Simulation of Materials (THEOS) and National Centre for Computational Design
and Discovery of Novel Materials (MARVEL),
École Polytechnique Fédérale de Lausanne, Lausanne 1015, Switzerland
2
Department of Physics, University of California at Berkeley and Materials Sciences Division,
Lawrence Berkeley National Laboratory, Berkeley, California 94720, USA
(Received 26 June 2019; revised manuscript received 16 October 2019; published 28 January 2020)
Heat conduction in dielectric crystals originates from the dynamics of atomic vibrations, whose
evolution is well described by the linearized Boltzmann transport equation for the phonon populations.
Recently, it was shown that thermal conductivity can be resolved exactly and in a closed form as a sum over
relaxons, i.e., collective phonon excitations that are the eigenvectors of Boltzmann equation’s scattering
matrix [A. Cepellotti and N. Marzari, Phys. Rev. X 6, 041013 (2016)]. Relaxons have a well-defined parity,
and only odd relaxons contribute to the thermal conductivity. Here, we show that the complementary set of
even relaxons determines another quantity—the thermal viscosity—that enters into the description of heat
transport, and is especially relevant in the hydrodynamic regime, where dissipation of crystal momentum
by umklapp scattering phases out. We also show how the thermal conductivity and viscosity parametrize
two novel viscous heat equations—two coupled equations for the temperature and drift-velocity fields—
which represent the thermal counterpart of the Navier-Stokes equations of hydrodynamics in the linear,
laminar regime. These viscous heat equations are derived from a coarse graining of the linearized
Boltzmann transport equation for phonons, and encompass both the limit of Fourier’s law and that of
second sound, taking place, respectively, in the regime of strong or weak momentum dissipation. Last, we
introduce the Fourier deviation number as a descriptor that captures the deviations from Fourier’s law due
to hydrodynamic effects. We showcase these findings in a test case of a complex-shaped device made of
graphite, obtaining a remarkable agreement with the recent experimental demonstration of hydrodynamic
transport in this material, and also suggesting that hydrodynamic behavior can appear at room temperature
in micrometer-sized diamond crystals. The present formulation rigorously generalizes Fourier’s heat
equation, extending the reach of physical and computational models for heat conduction also to the
hydrodynamic regime.
DOI: 10.1103/PhysRevX.10.011019 Subject Areas: Computational Physics,
Condensed Matter Physics,
Materials Science
I. INTRODUCTION progress took place in this field, propelled by newly

discovered hydrodynamic phenomena in crystals, with
Thermal transport in insulating crystals takes place
striking signatures such as Poiseuille-like heat flow [4]
through the evolution and dynamics of the vibrations of
and second sound [5]. The former manifests itself with a
atoms around their equilibrium positions. The first pre-
heat flux that is akin to the flow of a fluid in a pipe (i.e.,
dictive theoretical framework to describe thermal transport
showing a paraboliclike profile with a maximum in the
was developed by Peierls in 1929 [1–3], who envisioned a
center and minimum at the boundaries, due to viscous
microscopic theory in terms of a Boltzmann transport
effects); the latter instead results in heat propagation in the
equation (BTE) for the propagation of vibrational excita- form of a coherent temperature wave, rather than a
tions (phonon wave packets). In the 1960s significant diffusing heat front. Second sound has been observed
experimentally in a handful of solids: first, in solid helium
*
[email protected] [5], followed by sodium fluoride [6,7], bismuth [8],
sapphire [9], and strontium titanate [10,11]—all at cryo-
Published by the American Physical Society under the terms of genic conditions. Importantly, neither Poiseuille flow nor
the Creative Commons Attribution 4.0 International license.
Further distribution of this work must maintain attribution to second sound can be described by the macroscopic
the author(s) and the published article’s title, journal citation, Fourier’s equation, which is limited to a diffusive descrip-
and DOI. tion of heat propagation.
2160-3308=20=10(1)=011019(35) 011019-1 Published by the American Physical Society

SIMONCELLI, MARZARI, and CEPELLOTTI PHYS. REV. X 10, 011019 (2020)
These experimental observations have been accompanied but its complexity prevents a straightforward application to
by several pioneering efforts aimed at providing a quanti- materials with complex geometries (used in experiments
tative description of heat hydrodynamics at the mesoscopic and relevant for applications), thus posing limitations to the
level, i.e., in terms of partial differential equations (PDEs) study of how a material’s shape alters transport [35]. Recent
that are simpler than the microscopic integro-differential research efforts have been directed at developing meso-
BTE (we use “mesoscopic model” to denote any description scopic models that correct the shortcomings of Fourier’s
requiring more fields or PDEs than the Fourier’s PDE for the law and extend it at a lower computational cost than
temperature field [12]). Sussmann and Thellung [13], the solution of the full LBTE. Different strategies have
starting from the linearized BTE (LBTE) in the absence been suggested: some approaches reduce the complexity
of momentum-dissipating (umklapp) phonon-phonon scat- of the LBTE by neglecting the effects of the phonon
tering events, derived mesoscopic equations in terms of the modes’ repopulation due to scattering events [the so-called
temperature and of the phonon drift velocity, i.e., the single-mode relaxation-time approximation (SMA)], thus
thermal counterparts of pressure and fluid velocity in allowing for analytical [47–50] or asymptotic [51] solu-
liquids. Further advances came from Gurzhi [14,15] and tions. From the LBTE in the SMA, mesoscopic models that
Guyer and Krumhansl [16,17] who, including the effect generalize Fourier’s law accounting for ultrafast thermal
of weak crystal-momentum dissipation, obtained equations processes or ballistic effects have been derived [52–56].
for damped second sound and for Poiseuille heat flow. Other works have derived mesoscopic models without
Among early works, we also mention the discussions of relying on the LBTE [57], or have generalized the
phonon hydrodynamics in the framework of many-body Guyer-Krumhansl equation to account for the effect of
theory, as in Refs. [18,19]. While correctly capturing the the boundaries on the heat flow [58–62]. A hydrodynamic
qualitative features of phonon hydrodynamics, all the transport model has been derived from the LBTE in the
theoretical investigations mentioned above assume simpli- Callaway approximation, defining a phonon viscosity
fied phonon dispersion relations (either power law [14,15] which can be computed from atomistic data [63].
or linear isotropic [13,16,17]), or neglect momentum dis- Here, we provide a general and universal solution to the
sipation [13]. A more rigorous and general formulation— challenge of extending Fourier’s law all the way to the
albeit valid only in the hydrodynamic regime of weak hydrodynamic regime, deriving from the LBTE two novel
umklapp scattering—was introduced by Hardy, who coupled mesoscopic heat transport equations that cover
extended the study of second sound [20] and, together with exactly and on equal footing Fourier diffusion, hydro-
Albers, of Poiseuille flow in terms of mesoscopic transport dynamic propagation, and all regimes in between.
equations [21]. To this aim, we first show that one can define the thermal
The turn of the century brought renewed interest in the viscosity of a crystal starting from an exact solution of the
theory of heat conduction; computational and algorithmic LBTE in terms of the eigenvectors of the scattering matrix
advances now allow us to solve exactly the LBTE— (i.e., the relaxons introduced in Ref. [27] to determine
employing iterative [22–24], variational [25], or exact thermal conductivity), and evaluate from it the crystal-
diagonalization [26,27] methods—and thus investigate momentum flux generated in response to a drift-velocity
the accuracy of the LBTE and the models derived from gradient. The relaxons’ parity [27] highlights the comple-
it. In addition, nowadays it is possible to solve the LBTE mentary character of thermal conductivity and viscosity,
without any fitting parameter, deriving all quantities from with the former being determined by odd relaxons, and the
first principles; this has been shown to describe accurately latter by even relaxons.
the thermal properties of bulk crystals [23,28–37], provided Next, we use a coarse-graining procedure to derive two
phonon branches remain well separated [38]. novel “viscous” heat equations: these are two coupled
Further applications of the LBTE, combined with state- equations for the local temperature and drift-velocity fields,
of-the-art first-principles simulations, have also recently and are parametrized in terms of the thermal conductivity
predicted the existence of hydrodynamic phenomena at and viscosity. The viscous heat equations represent the
noncryogenic temperatures in low-dimensional or layered thermal counterpart of the Navier-Stokes equations for
materials such as graphene [31,39,40], other 2D materials fluids in the laminar regime, and, as mentioned, include
[31], carbon nanotubes [41], and graphite [42]. Fittingly, Fourier’s law and second sound in the limits of strong and
and remarkably, these theoretical suggestions have now weak crystal-momentum dissipation, respectively.
been confirmed by the experimental finding of second Lastly, we introduce the Fourier deviation number
sound in graphite [43] at ∼100 K. (FDN), a dimensionless parameter that quantifies the
In the hydrodynamic regime, where Poiseuille flow or deviation from Fourier’s law due to hydrodynamic effects.
second sound occurs, Fourier’s law fails [43–46], depriving We test this formalism on graphite, diamond, and silicon,
us of the most common tool used to predict the temperature showing that the FDN predicts a temperature and size for
profile in a device. The LBTE, in principle, allows us to the window of hydrodynamic thermal transport that rep-
predict accurately thermal transport under these conditions, licates the explicit solution of the viscous heat equations,
011019-2
GENERALIZATION OF FOURIER’S LAW INTO VISCOUS HEAT … PHYS. REV. X 10, 011019 (2020)
but at a negligible computational cost. Most importantly, as the temperature Tðr; tÞ does for the local energy;
the prediction of the present formulation for the hydro- formally, Tðr; tÞ and uðr; tÞ can be related to the
dynamic window of graphite is found to be in excellent Lagrange multipliers that enforce the constraints of fixed
agreement with recent, pioneering experiments [43]. In local energy and momentum, respectively. The drifting
passing, we also predict that hydrodynamic behavior can distribution Eq. (3) depends implicitly on r, t through
appear in diamond at room temperature for micrometer- Tðr; tÞ and uðr; tÞ. Next, we study the effect of small
sized crystals. perturbations in the temperature and drift velocity. To this
aim, we expand the deviation from equilibrium Eq. (2) in
II. THERMAL VISCOSITY proximity of the local equilibrium Eq. (3) [20], finding

A microscopic description of thermal transport is given ∂N D
ν
∂N D
ν

nν ðr; tÞ ¼ ½Tðr; tÞ − T̄ þ · uðr; tÞ þ nδν ðr; tÞ
∂u eq
by the LBTE:
∂T eq
∂nν ðr; tÞ 1X ¼ nTν ðr; tÞ þ nD δ
ν ðr; tÞ þ nν ðr; tÞ; ð4Þ
þ vν · ∇nν ðr; tÞ ¼ − Ω 0 n 0 ðr; tÞ; ð1Þ
∂t V ν0 νν ν
where nTν arises from the local temperature [68], nD ν from
δ
where ν labels a phonon state (i.e., an index running on all the local drift velocity, and nν accounts for all the
the phonon wave vectors q and phonon branches s), vν is information that cannot be mapped to a local equilibrium
the phonon group velocity, V is the crystal volume [64], state; the subscript “eq” means that the derivatives are
and Ωνν0 is the phonon scattering matrix [27]. Equation (1) computed at equilibrium where Tðr; tÞ ¼ T̄ and uðr; tÞ ¼ 0,
governs the evolution of the deviation nν ðr; tÞ of the and to ease the notation this will be omitted in the
phonon populations from equilibrium: following. In analogy with previous work [25,27], we
consider the steady-state case and linearize the LBTE
nν ðr; tÞ ¼ N ν ðr; tÞ − N̄ ν ; ð2Þ around the constant temperature and drift-velocity gra-
dients (i.e., nδν , ∇T, and ∇u are constant). Then, we
where N ν ðr; tÞ are the out-of-equilibrium phonon popula- substitute Eq. (4) in Eq. (1) and, keeping only terms linear
tions at position r and time t, N̄ ν ¼ ðeℏων =kB T̄ − 1Þ−1 is the in the temperature and drift-velocity gradients, we obtain
equilibrium Bose-Einstein distribution at temperature T̄, D
∂ N̄ ν ∂N ν
and ων are the phonon frequencies. From the solution of the v · ∇T þ vν · · ∇u
∂T ν ∂u
LBTE P one can derive the local lattice energy Eðr; tÞ ¼
ð1=VÞ ν ℏων NP ν ðr; tÞ and the total crystal momentum
1X δ
¼− Ω 0 ½nT0 ðr; tÞ þ nDν0 ðr; tÞ þ nν0 : ð5Þ
Pðr; tÞ ¼ ð1=VÞ ν ℏqN ν ðr; tÞ [20]. The former is often V ν0 νν ν
studied in connection with the thermal conductivity [27],
while the latter becomes relevant in the hydrodynamic We recast Eq. (5) in the symmetric (thus diagonalizable) form,ffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
regime of thermal transport [13,15,65]. We note in passing i.e., in terms of Ω̃νν0 ¼Ωνν0 ½N̄ ν0 ðN̄ ν0 þ1Þ=½N̄ ν ðN̄ ν þ1Þ
that this latter emerges only in “simple” crystals, i.e., those and ñν ðr; tÞ ¼ nν ðr; tÞ½N̄ ν ðN̄ ν þ 1Þ−1=2 with the goal of
where phonon interbranch spacings are much larger than using the relaxon picture [27] to gain insight into the physics
their linewidths [38]. underlying transport. We then simplify the symmetrized
The energy flux generated in response to a temperature Eq. (5) exploiting parity: we recall that a function f ν is even
gradient determines the thermal conductivity; correspond- if f ν ¼ f −ν (this is, e.g., the case of the phonon energy
ingly, the crystal-momentum flux generated in response to ℏων ¼ ℏω−ν ) and odd if f ν ¼ −f −ν (e.g., the phonon group
a perturbation of the drift velocity determines the thermal velocity vν ¼ −v−ν ), and use the notation −ν ¼ ð−q; sÞ.
viscosity (for the analogous quantity in electronic transport, Therefore, ∂ N̄ ν =∂T and thus ñTν ðr; tÞ are even, whereas
see Ref. [66]). Therefore, we start by considering a crystal ∂ N̄ D
ν =∂u and thus ñν ðr; tÞ are odd.
D
in the hydrodynamic regime of thermal transport (i.e., Since the eigenvectors of the scattering matrix have a
carrying a finite amount of crystal momentum); the local well-defined parity [20,69], we can split ñδν into ñδE δO
ν þ ñν ,
equilibrium phonon distribution, obtained maximizing the δE δO
separating the even (ñν ) and odd (ñν ) components. At
local entropy under the constraints of fixed local energy and steady state, Eq. (5) decouples into two equations, one for
momentum [67], is the phonon drifting distribution [15], each parity. The equation for the odd part is
1
ν ½Tðr; tÞ; uðr; tÞ ¼
ND : ð3Þ vν ∂ N̄ ν 1X
e½1=kB Tðr;tÞ½ℏων −ℏq·uðr;tÞ − 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · ∇T ¼ − Ω̃ 0 ñδO ð6Þ
V ν0 νν ν
0
N̄ ν ðN̄ ν þ 1Þ ∂T
The drifting distribution differs from the Bose-Einstein
distribution due to the presence of the drift velocity uðr; tÞ, and describes the response to a thermal gradient [25,27],
a parameter expressing the amount of local momentum, just where ñδO
ν is the odd out-of-equilibrium phonon
011019-3
population generated in response to a temperature normal part of the scattering matrix conserves crystal
gradient. In writing Eq. (6) we made the assumption that
P P momentum, there exists a set of 3 eigenvectors ϕiν
δO δO (i ¼ 1; …; 3, where 3 is the dimensionality of the system)
ð1=VÞ ν0 Ω̃νν0 ½ñD ν0 ðr; tÞ þ ñν0 ≃ ð1=VÞ ν0 Ω̃νν0 ñν0 , as
explained in Appendix A. The solution of Eq. (6) can then with zero eigenvalue for Ω̃Nνν0 , which are associated to the
be used to determine the heat flux and the thermal conduc- conservation of crystal momentum in the 3 Cartesian
tivity (see, e.g., Refs. [25,27]). The equation for the even directions. Because the viscosity describes the response
part is of the crystal-momentum flux to a change of drift velocity, it
D is not surprising that the eigenvectors ϕiν appear in its
vν ∂N ν 1X
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · · ∇u ¼ − Ω̃ 0 ñδE0 ; ð7Þ definition. In fact, the deviation-from-equilibrium distribu-
N̄ ν ðN̄ ν þ 1Þ ∂u V ν0 νν ν tion [Eq. (4)] is linear in the drift velocity, with the
proportionality coefficients being these special eigenvectors
and describes the response to a drift-velocity gradient, where (see Appendix A for a proof), and therefore they appear in
ñδE
ν is the even out-of-equilibrium phonon population gen- the viscosity as well, to describe a perturbation to the local
erated in response to it. In writing Eq. (7) we used the property equilibrium. We note in passing that the thermal viscosity
that ñTν ðr; tÞ is an eigenvector of the scattering matrix with
P defined in Eq. (9) has the dimensions of a dynamic viscosity
zero eigenvalue: ð1=VÞ ν0 Ω̃νν0 ñTν0 ðr; tÞ ¼ 0 (also detailed in (Pa s) and satisfies the property μijkl ¼ μilkj ¼ μkjil.
Appendix A). We report in Fig. 1 the first-principles estimates of the
The phonon deviation ñδE ν gives rise to a crystal-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi thermal viscosity for graphite, diamond, and silicon. We
P
momentum flux, ΠδE ¼ð1=VÞ ν ℏq vν N̄ ν ðN̄ ν þ1ÞñδE
ij i j
ν choose these three materials as prototypes for two different
[15,20]. Analogously to the electronic case [66], the flux behaviors, the former two displaying hydrodynamic ther-
of crystal momentum allows us to define the thermal mal transport [25,42,43,45,70,71], as opposed to the more
viscosity as the fourth-rank tensor relating the local conventional case of silicon [25,27]. We account for finite-
response Πij δE ðr; tÞ to the local perturbation ∇uðr; tÞ:
size effects by combining the bulk viscosity in Eq. (9) with
its ballistic limit via Matthiessen’s rule for a sample having
X ∂uk ðr; tÞ size (or grain size) of 10 μm; the same renormalization is
Πij
δE ðr; tÞ ¼ − ηijkl ; ð8Þ
kl
∂rl applied also to the thermal conductivity (see Appendix C
for details). This is an approximate treatment of surface
the uniform drift-velocity gradient is a special space- scattering effects, which provides an estimate of size effects
independent case of this local relationship. Equation (7) at a much lower computational complexity compared to
has the same mathematical form of the steady-state LBTE more refined models [70,72]. In general, when samples’
linearized in the temperature gradient and used to compute sizes become comparable or smaller than the carriers’
the thermal conductivity. Therefore, we can readily solve diffusion lengths [61,73–77], transport coefficients cannot
it applying the methodology introduced in Ref. [27], be rigorously defined and a more accurate treatment is
and based on the eigenvectors of the scattering matrix required, solving the much more complex and computa-
(relaxons), to find a closed expression for ñδE
ν . Combining tionally expensive space-dependent LBTE [24,35,47,78].
such a solution with Eq. (8) we find, after symmetrizing, Matthiessen’s approach is a good approximation when the
the following expression for the symmetrized thermal values of thermal conductivity and viscosity do not differ
viscosity: significantly from their bulk counterparts, as is the case for
the materials, sizes, and temperatures considered in this
ηijkl þ ηilkj pffiffiffiffiffiffiffiffiffi
ffiX wj wl þ wl wj work (e.g., in Fig. 1 the size renormalization is negligibly
μijkl ¼ ¼ Ai Ak iα kα iα kα
τα ; ð9Þ small in the temperature range considered for graphite and
2 α>0
2
silicon, and becomes so above 300 K for diamond).
P For graphite in the low-temperature (≲100 K) regime,
where Ai ¼ ð∂Pi =∂ui Þjeq ¼ ð1=kB T̄VÞ ν N̄ ν ðN̄ ν þ1Þðℏqi Þ2
the bulk in-plane viscosity components are constant or
is referred to as the specific momentum, τα is the relaxation
slowly increasing with temperature; at higher temperatures,
time of relaxon α (i.e., the inverse eigenvalue associated to
the viscosity components increase with temperature up to
the eigenvector θαν of the symmetrized scattering matrix
reaching their saturation values. For a graphite polycrystal-
Ω̃νν0 [27]), and wjiα is the velocity tensor wjiα ¼ line sample with a grain size of 10 μm, finite-size effects on
P
ð1=VÞ ν ϕiν vjν θαν for relaxon θαν and eigenvector ϕiν viscosity (and on conductivity) are small. For diamond in
(see Appendix B for a full demonstration). ϕiν are three the temperature regime below 90 K, the bulk values of the
special eigenvectors linked to the crystal momentum of thermal viscosity increase with increasing temperature;
the system; to see this, we first decompose the scattering then in the intermediate temperature regime (100 ≲ T≲
matrix as Ω̃νν0 ¼ Ω̃Nνν0 þ Ω̃Uνν0 , where Ω̃νν0 and Ω̃νν0 contain
N U
300 K), they decrease with temperature; and finally
only momentum-conserving (normal) and momentum- for increasingly higher temperatures, they gently increase
dissipating (umklapp) processes, respectively. Since the up to saturating to the high-temperature limiting values.
011019-4
(a) temperature yields a decrease, up to a high-temperature

saturation value, of the bulk thermal viscosity; size effects
renormalize the viscosity below ∼100 K, yielding viscosity
components that increase with temperature (i.e., a behavior
analogous to that of diamond). We note that in the high-
temperature limit, i.e., when the thermal conductivity
decays as T −1 [3,79], the viscosity components tend to
constant values for all these three materials. The total
bulk thermal conductivities are shown in the insets of
Fig. 1(a)–1(c) for comparison (we only show one compo-
nent; for diamond and silicon the conductivity tensor is
isotropic, for graphite we show the in-plane component).
× ×
The off-plane transport coefficients for graphite are dis-
cussed in Fig. 9(a). In graphite, the bulk thermal conduc-
(b)
tivity below 100 K does not increase monotonically as the
temperature is decreased due to the presence of natural-
abundance isotopic scattering. Natural-abundance isotopic
scattering has been considered also for diamond and
silicon, and the thermal conductivity at low temperatures
is much lower than for isotopically-pure diamond [25,
80–85] or silicon [86]. We have also verified that the effects
of phonon coherences are negligible in these crystals [38].
We note in passing that, even if the thermal conductivities
of diamond and silicon differ by more than one order of
magnitude, their largest thermal-viscosity components
× × differ only by a factor of 3. These results may be compared
with the case of water, whose dynamic (shear) viscosity is
(c) 8.9 × 10−4 Pa s at room temperature, indicating that the
thermal viscosity found here is comparable or larger. To a
good approximation, water is an incompressible fluid,
and thus its largest viscosity component is ijij, also called
“first viscosity” or “shear viscosity.” For compressible fluids,
the iiii components of the viscosity tensor—also called
“second viscosity” or “volume viscosity” [87]—are non-
negligible. Here, in contrast with water, the iiii component of
the thermal viscosity tensor is the largest, which elicits an
analogy to a compressible fluid. It is important to mention
that the present formulation may need further extension to
deal with unstrained 2D materials, since their flexural
×
phonon modes have a quadratic dispersion ων ∝ q2 at
T ¼ 0 K [88]. Under these conditions, long-wavelength
FIG. 1. Largest components of the fourth-order thermal viscosity phonons lead to q · u being larger than ων , causing negative
tensor for graphite (a), diamond (b), and silicon (c), as a function of values of the phonon drifting distribution which are not
temperature (the off-plane tensor components for graphite are compatible with a semiclassical description of transport.
reported in Fig. 9). Insets: Total thermal conductivities (in-plane Further work is needed to address this issue, for example,
components for graphite and diagonal components of the isotropic considering the phonon renormalization due to coupling
tensor for diamond and silicon) as a function of temperature. The between bending and stretching degrees of freedom of the
dashed lines refer to the bulk materials [Eq. (9) for viscosity and
monolayer [89] or the presence of a substrate [90], or
Eq. (B12) for conductivity], the solid lines show finite-size
predictions computed using Eqs. (C1) and (C2) for a sample
introducing the Wigner-function formalism [38,91].
having size (or grain size) 10 μm (see main text and Appendix C; Finally, it is worth checking the complementary char-
for diamond and silicon the component μijji is negligible). acter of the thermal conductivity and viscosity that arises
from their decoupled relaxons’ contributions. As com-
However, size effects renormalize this behavior for mented above, decomposing the thermal conductivity [27]
temperatures below 300 K, leaving viscosity components and viscosity Eq. (9) in terms of single relaxons makes it
that increase with temperature. In silicon, increasing possible to show that the thermal viscosity is uniquely
011019-5
(a) determined by the even part of the relaxon spectrum, while

the thermal conductivity is determined uniquely by the odd
part of the relaxon spectrum [27]. In Fig. 2 we highlight the
contributions of each relaxon to the total thermal conduc-
tivity and viscosity, confirming numerically this picture.
III. VISCOUS HEAT EQUATIONS
We show here that heat conduction can be described by
two novel viscous heat equations that cover on the same
footing both the Fourier and hydrodynamic limits, and all
intermediate regimes. These are two coupled equations in
the temperature Tðr; tÞ and drift-velocity uðr; tÞ fields,
which are parametrized by the thermal conductivity and
viscosity. These equations represent the thermal counter-
(b) part of the Stokes equations of fluid dynamics—i.e., the
Navier-Stokes equations in the linear regime, whose
solution yields the laminar flow—where temperature takes
the role of pressure and the phonon drift velocity that of
the fluid velocity. In the kinetic regime, when momentum-
dissipating (umklapp) scattering processes dominate [31],
these viscous heat equations become equivalent to Fourier’s
heat equation.
As underscored before, hydrodynamic thermal transport
is characterized by energy conservation and crystal-
momentum quasiconservation (the latter being exactly
conserved only in the absence of umklapp processes
[13]). Conserved quantities in the LBTE dynamics can
be related to the eigenvectors of the full or normal
scattering matrix with zero eigenvalues [20] (see also
(c) Appendix A). Four of these eigenvectors (i.e., phonon
distribution functions) can be identified. The first one is the
Bose-Einstein eigenvector, ϕ0ν ∝ ℏων ∝ ∂ N̄ ν =∂T ∝ ñTν ,
which is the eigenvector of zero eigenvalue for the
symmetrized full scattering matrix Ω̃νν0 ; its zero eigenvalue
is associated to energy conservation in scattering events
(both normal and umklapp). The other three eigenvectors
are the drift eigenvectors, ϕiν ∝ ℏqi ∝ ∂N D ν =∂u ∝ ñν ,
i D
i ¼ 1, 2, 3 (where 3 is the dimensionality of the system)

already introduced for the evaluation of the viscosity; these
ϕiν are eigenvectors with zero eigenvalue for the normal
part of the scattering matrix Ω̃Nνν0 and are associated to the
conservation of crystal momentum by normal scattering
events [13,16,20]. We note that these four eigenvectors
FIG. 2. Relaxons’ contributions to the bulk thermal conduc- constitute the first two terms of the phonon distribution
tivity and to the thermal viscosity at 300 K for graphite [(a), in- expansion in Eq. (4). We can thus derive the mesoscopic
plane components], diamond (b), and silicon (c). Each dot equations that govern the evolution of the temperature
represents a relaxon, with its color labeling its relaxation time, [Tðr; tÞ] and drift-velocity [uðr; tÞ] fields projecting the
and its area being proportional to the sum of its percentage microscopic LBTE in the subspaces spanned by ϕ0ν and by
contributions to the thermal conductivity and viscosity. The
ϕiν (i ¼ 1; …; 3). In order to derive a closed-form equation
dashed lines are plotted as a guide to the eye, to underscore
how even and odd relaxons are fully decoupled. Odd relaxons,
for the drift velocity, when projecting in the subspace
which determine the thermal conductivity, yield negligible (zero) spanned by ϕiν we consider the effects of momentum
contributions to the thermal viscosity; conversely, even relaxons dissipation only within that subspace. The result is the
determine the thermal viscosity and yield negligible (zero) following set of equations (see Appendix D for a detailed
contributions to thermal conductivity. derivation):
011019-6
∂Tðr; tÞ X 3 pffiffiffiffiffiffiffiffiffiffiffiffi ∂uj ðr; tÞ X 3

∂ 2 Tðr; tÞ
C þ i
W 0j T̄A C j
− κ ij ¼ 0; ð10Þ
∂t i;j¼1
∂r i
i;j¼1
∂ri ∂rj
rffiffiffiffiffiffiffiffi 3 3 pffiffiffiffiffiffiffiffiffiffi
∂ui ðr; tÞ CAi X j ∂Tðr; tÞ X 3 2 k
ijkl ∂ u ðr; tÞ
X
A i
þ W i0 − μ ¼ − Ai Aj Dij U u ðr; tÞ;
j
ð11Þ
∂t T̄ j¼1 ∂rj j;k;l¼1
∂r j
∂r l
j¼1
P
where C ¼ ð1=kB T̄ 2 VÞ ν N̄ ν ðN̄ ν þ 1Þðℏων Þ2 is the where, on the basis of the phonon population expansion in
P
specific heat, W i0j ¼ ð1=VÞ ν ϕ0ν viν ϕjν is a velocity tensor Eq. (4), we distinguish the drifting heat flux into the
that arises from the nondiagonal form of the diffusion contributions from the local temperature gradient,
P
operator in the basis of the eigenvectors of the normal part of Qδ;i ðr; tÞ ¼ − j κij ∇j Tðr; tÞ, and from the drift-velocity
the scattering matrix (see Appendix D), T̄ is the reference P pffiffiffiffiffiffiffiffiffiffiffiffi
field, QD;i ðr; tÞ ¼ j W i0j T̄Aj Cuj ðr; tÞ. Similarly, the
(equilibrium) temperature on which a perturbation is ap-
plied (see Sec. V and Appendix A 1), Ai is the specific nonequilibrium momentum flux receives separate con-
momentum in direction i (defined in Sec. II), μijkl is the tributions from the local temperature, Πij T ðr; tÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi j
thermal viscosity tensor, κij is the thermal conductivity CA =T̄ W i0 ½Tðr; tÞ − T̄, and from variations of the drift
i
tensor [27], and Dij U is the momentum dissipation rate; the velocity through Eq. (8). Finally, ð∂Pi =∂tÞjumkl accounts for
latter is caused both by the presence of umklapp processes the dissipation of crystal momentum by umklapp processes
as well as boundary scattering (i.e., Dij U is sensitive to size or scattering with boundaries; further details can be found in
effects like the thermal conductivity and viscosity). Treating Appendix D. The distinction between temperature-driven
boundary scattering as in Ref. [70], the momentum- and drifting components of the heat flux [Qδ ðr; tÞ and
dissipating phonon collisions described by the matrix QD ðr; tÞ, respectively] is essential for hydrodynamic trans-
Ω̃U
νν0 can be rewritten as Ω̃νν0 ¼ Ω̃νν0
U U;bulk
þ Ω̃U;boundary
νν0 ðLS Þ, port. We show in Sec. VI that Fourier’s law is recovered in
where Ω̃νν0U;bulk
accounts for umklapp scattering in the bulk the limiting case where crystal-momentum dissipation
U;boundary dominates over viscous effects and is the fastest timescale
and Ω̃νν0 ðLS Þ approximates the momentum-
of the phonon dynamics.
dissipating scattering of phonons with boundaries at a
It is worth mentioning that the viscous heat equations
distance LS . Therefore, we compute Dij ij ij
U as DU ¼ DU;bulk þ introduced here differ from the Stokes equations for fluids
−2
P
Dij ij
U;boundary ðLS Þ, where DU;bulk ¼ V
i U;bulk ϕj
νν0 ϕν Ω̃νν0 ν0
in two major ways. First, there is no analogy with to the
ij −2
P U;boundary j mass conservation satisfied by Stokes equations, since the
and DU;boundary ðLS Þ¼V νν0 ϕν Ω̃νν0
i ðLS Þϕν0 ¼
−1
P i j total phonon number is not a constant of motion (e.g., a
V ν ϕν ðjvν j=LS Þϕν . The scalar equation (10) and the phonon coalescence event decreases the number of pho-
vectorial (3-components) equations (11) rule the coupled nons in the system). Second, while collisions between
evolution of the scalar temperature field and of the vector molecules in the fluid conserve momentum, scattering
drift-velocity field; they constitute the main result of this among phonons does not conserve crystal momentum in
work and we name these “viscous heat equations.” These the presence of umklapp processes.
transport equations are reminiscent of the linearized Stokes The most relevant feature of the viscous heat equations is
equations for fluids: to see this more clearly, we note that their capability to describe hydrodynamic thermal transport
local energy Eðr; tÞ and crystal momentum Pðr; tÞ are in terms of mesoscopic quantities, i.e., temperature and
proportional to temperature and drift velocity, respectively drift velocity, resulting in a much simpler and computa-
[Eðr; tÞ ¼ CTðr; tÞ and Pi ðr; tÞ ¼ Ai ui ðr; tÞ, where C is the tionally less expensive approach than the microscopic
specific heat and A is the specific momentum]. Exploiting LBTE. The parameters entering Eqs. (10) and (11) can
these relationships, it is possible to rewrite the viscous heat be determined from first-principles calculations or, possibly
equations in a more familiar form, namely, as energy E and less accurately, from classical potentials, and are tabulated
momentum Pi balance equations: in Appendix E for graphite, diamond, and silicon.
In order to be solved, the viscous heat equations require
∂Eðr; tÞ
þ ∇ · ½Qδ ðr; tÞ þ QD ðr; tÞ ¼ 0; ð12Þ appropriate boundary conditions on the temperature and
∂t drift velocity. Boundary conditions on temperature have

∂Pi ðr;tÞ X ∂Πij ij
T ðr;tÞ ∂ΠδE ðr;tÞ ∂Pi ðr;tÞ been widely studied in conjunction with Fourier’s heat
þ þ ¼− ;
∂t j
∂rj ∂rj ∂t umkl equation [92]: typically, one makes assumptions on the
system’s capability to exchange heat at the boundaries, and
ð13Þ on the temperature at those boundaries (Neumann and
011019-7
Dirichlet boundary conditions, respectively [92]). In Sec. V where τss and vss are the second-sound relaxation time
we consider a system in which the temperature is fixed at and propagation velocity, yet to be determined. In contrast
some boundaries, while the others are assumed to be to Fourier’s law, which has the form of a classical (non-
adiabatic (that is, the heat flux across these boundaries relativistic) diffusion equation and states that a temper-
is zero). In contrast, boundary conditions on the drift ature-gradient variation causes an instantaneous variation
velocity, i.e., on crystal momentum at the sample’s borders, of the heat flux, Eq. (14) has the physical property that
have not been studied as extensively. Since crystal momen- a sudden localized change of temperature propagates in
tum is not conserved at boundaries [13], we impose a no-slip space with a finite speed (i.e., it is not felt instantly
condition of zero drift velocity uðr; tÞ on all boundaries, everywhere in space) [12,97–99]. The temperature profile
ensuring thus zero drifting heat, QD ðr; tÞ ∝ uðr; tÞ. As that solves Eq. (14) has the form of a damped wave:
discussed in past work [3,35], more comprehensive boun-
dary conditions require quantifying phonon reflection at Tðx; tÞ ¼ T̄ þ δTei½kx−ω̄ðkÞt e−t=2τss ; ð15Þ
surfaces, and are beyond the scope of this work. where the second-sound frequency ω̄ðkÞ depends on the
The viscous heat equations (10) and (11) improve second-sound wave vector k. In Appendix F we derive the
upon past work on different levels. First, they are valid second-sound equation from the viscous heat equations (10)
for general dispersion relations for phonons, and thus and (11) following two different approaches (bottom up
overcome the limitations—pointed out by Hardy and and top down). In the first bottom-up approach we find the
Albers [21] already in 1974—of the pioneering mesoscopic conditions for which the damped-wave equation (14)
models developed in the 1960s by Guyer and Krumhansl emerges from the viscous heat equations (10) and (11).
[16,17] or Gurzhi [15], which assumed linear-isotropic or When this happens, the solution of Eq. (14) is the damped-
power-law phonon dispersion relations, respectively wave equation for temperature Eq. (15) shown above, with
(assumptions which arose from the hypothesis that hydro- the second-sound dispersion relation given by ω̄ðkÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
dynamic phenomena would occur at cryogenic temper- v2ss k2 − ð2τss Þ−2 [this can be easily verified substituting
atures only). The limitations of the Guyer-Krumhansl Eq. (15) into Eq. (14)]. This allows us to express τss and
[16,17] and Gurzhi [15] models were overcome in 1974 vss in terms of the parameters appearing in the viscous
by Hardy and Albers [21], who derived from the LBTE a heat equations; in particular, τss ¼ ½CðW xx0 Þ2 =½κxx ðDxx 2
UÞ þ
set of mesoscopic equations for energy and crystal momen- x 2 x 2
DU CðW x0 Þ and vss ¼ ½κ DU þCðW x0 Þ =CW x0 . The
xx xx xx x
tum valid for a general phonon dispersion relation (i.e., not propagation velocity of second sound is affected by
necessarily linear isotropic). The formulation of Hardy and damping and depends on the wave vector k: it
Albers relies on the hydrodynamic approximation—i.e., is given by the group velocity, vg ðkÞ ¼ ∂ ω̄ðkÞ=∂k ¼
that the fastest timescale of phonon dynamics is that of kvss ½k2 − ð2τss vss Þ−2 −1=2 , and we note that it reduces to
normal processes—that is valid only within the hydro- the undamped propagation velocity vss in the undamped
dynamic regime, where umklapp collisions are rare events. limit τss → ∞.
In fact, the equations of Hardy and Albers are limited just to These results are consistent with empirical expectations
the hydrodynamic regime and do not incorporate Fourier’s on second sound: in the limit of weak crystal-momentum
law as a limit (and thus any intermediate regimes). The dissipation, the second-sound relaxation time increases,
viscous heat equations address exactly this issue, and while the velocity becomes smaller, making second sound
Fourier’s law emerges when crystal-momentum dissipation more likely to be observed in the hydrodynamic regime
(due to umklapp processes or scattering with boundaries) xx −1
dominates over viscous effects and is the fastest timescale
[17,20]. In fact, when Dxx U → 0, we find that τss → ðDU Þ
of the phonon dynamics (see Sec. VI). Finally, the viscous and vss → W x0 . We note that the viscous heat equations
x
heat equations take into account the entire collision matrix, describe not only the propagation of the temperature field,
at variance with recent mesoscopic models derived from but also that of the drift velocity. In Appendix F we show that
the LBTE either in the SMA [60] or in the Callaway when second sound emerges, the drift-velocity field prop-
approximation [63]. agates as a damped wave as well [i.e., similar to Eq. (15)],
with the same relaxation time and velocity of temperature,
IV. SECOND SOUND but with a phase shift. Phase shifts between hydrodynamic
and resistive components of the frequency-dependent heat
Second sound is the coherent propagation of a temper- flux have also been discussed recently in the context of an
ature wave [13,17,19,20,40,43,93–96], and it is an effect improved Callaway approximation for the LBTE [67,100].
properly described by the viscous heat equations. From a As an alternative, we took inspiration from Ref. [40],
phenomenological point of view, second sound appears which derived the second-sound dispersion relations by
when the temperature field satisfies the following damped- taking advantage of the Laplace transform of the LBTE, to
wave equation [20] (we define x as the second sound identify solutions in the form of a damped wave following a
propagation direction): top-down approach. In particular, we take a damped-wave
ansatz for temperature, Eq. (15) with ω̄ðkÞ replaced by ω̂ðkÞ
∂ 2 Tðx; tÞ 1 ∂Tðx; tÞ 2
2 ∂ Tðx; tÞ ¼ 0; and τss by τ̂ss (we use the hat to distinguish the results for
þ − v ð14Þ
∂t2 ∂x2
ss
τss ∂t the top-down approach from these of the bottom-up
011019-8
approach), and a similar one for the drift-velocity field The former emerges when normal processes dominate and
(with the same frequency and decay time of temperature); is described in terms of balance equations for energy and
we substitute these into the viscous heat equations (10) and momentum; the latter is determined by a uniform energy
(11) and determine the conditions under which these flux that decays exponentially. The second sound discussed
are acceptable solutions. As detailed in Appendix F, we here is of the drifting kind, as it emerges from a set of
find that the dispersion relations of second sound in the balance equations for energy and crystal momentum
long-wavelength limit reduce to ωðkÞ ^ − i=2^τss ≈
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi derived from the LBTE.
2 x 2 2
k ½ðW x0 Þ þ DU κ =2C − μ DU =2A − ðDxx
xx xx xxxx xx x
U =2Þ
−iDU =2. In this long-wavelength limit and in the hydro-
xx
V. CASE STUDY
dynamic regime Dxx U → 0, we have τ ^ ss ≈ 1=Dxx U and
v^ g ðkÞ ≈ W 0x , which is consistent with the first bottom-
x We showcase the solution for the viscous heat equations
up approach presented to derive the equations for second for graphite around the equilibrium temperature T̄ ¼ 70 K
sound. We further stress that the LBTE can only rigorously in the geometry shown in Fig. 3, often used as an illustrative
describe second sound in the long-wavelength limit k → 0 example in textbooks on fluid dynamics. The equations are
in order for the temperature to be slowly varying (for which solved numerically using a finite-element solver imple-
the two approaches shown in this section provide the same mented in Mathematica [102], imposing a temperature
result); for smaller wavelengths, the definition of temper- of 80 K on the left side (x ¼ 0 μm) and 60 K on the right
ature itself becomes questionable [101]. side (x ¼ 15 μm), assuming all boundaries at x ≠ 0 and
We also recall that Enz [19] and Hardy [20] distin- x ≠ 15 μm to be adiabatic, and imposing a no-slip con-
guished between “drifting” and “driftless” second sound. dition on u at all boundaries.
(a) (c)
(b)
FIG. 3. In-plane (x-y) heat fluxes in graphite around T̄ ¼ 70 K, for a sample infinite in the z direction (the z direction of the sample
coincides with the off-plane direction of graphite). Panel (a) shows the total heat fluxP(QD þ Qδ ) obtained from the viscous heat
equations (10) and (11). Panel (b) shows instead the Fourier heat flux [QiFourier ðr; tÞ ¼ − j κij ∇j Tðr; tÞ] obtained solving the steady-
P
state Fourier equation, ij κij ½∂ 2 Tðr; tÞ=∂ri ∂rj ¼ 0. Boundary conditions are as follows: the temperature is set at 80 K (60 K) on the
left (right) boundaries, a zero total heat flux is imposed at the other boundaries, and zero drift velocity is imposed on all boundaries (no-
slip boundary condition). With this choice of boundary conditions, the phonon distribution reduces to the Bose-Einstein distribution on
the left and right boundaries, where the temperature is fixed. Despite having local equilibrium at the left and right boundaries, a nonzero
drift velocity is present in the material as a consequence of the coupling in Eq. (10) between drift velocity and temperature.
Panel (c) shows the x component of the heat flux along sections taken at x ¼ 1.5 μm (orange, gray, red) and x ¼ 9.0 μm (black, blue,
green); solid lines are the results obtained from the viscous heat equations, dashed lines are the results from Fourier’s law, and dotted
lines are the heat fluxes due to the temperature gradient within the viscous heat equations (Qδ ).
011019-9
We show in Fig. 3 the total viscous heat flux Qδ þ QD and (a)

the flux predicted by Fourier’s law QFourier [Figs. 3(a)
and 3(b), respectively] and contrast the Fourier and viscous
heat flux components across two different sections
[Fig. 3(c)].
We stress that Fourier’s law lacks a description of the
contribution to the heat flux derived from the local drift
velocity [13,103,104]. As a result, Fourier’s law misses
qualitative and quantitative properties of the heat flux
profile. The largest differences are observed in proximity
of spatial inhomogeneities, such as boundaries or corners.
For example, QD quickly increases (decreases) in proximity
of the thermal reservoir on the hot (left) [cold (right)] side (b)
of the sample; these changes in QD determine opposite
changes in Qδ . Microscopically, these variations are caused
by the rapid transition of the phonon distribution from the
Bose-Einstein equilibrium distribution imposed at the
boundaries to an out-of-equilibrium distribution carrying
nonzero total crystal momentum (i.e., a drift velocity)
inside the sample.
We report in Fig. 3(c) the total heat flux profiles along
two transversal sections of the sample, contrasting the
prediction from the viscous heat equations (solid lines) with
that of Fourier’s law (dashed lines). Along these sections,
Fourier’s law predicts a flat heat flux profile, while the
viscous heat equations yield a Poiseuille-like profile. The FIG. 4. Difference between the temperature profiles predicted by
results from the viscous heat equations are thus substan- Fourier’s law and by the viscous heat equations, plotted in (a) along
y for various values of x and in (b) along x for y ¼ 4.5 μm. The
tially different from Fourier’s predictions, and the behavior
insets of (b) show the temperature profile along x for y ¼ 4.5 μm;
of the heat flux can be understood from a simple analytical the result of the viscous heat equations (10) and (11) is the solid
1D solution of the viscous heat equations in the absence of blue line, Fourier’s law prediction is the dashed red line.
umklapp processes [13]: as discussed in Appendix G, the
flux is described by hyperbolic functions with a character-
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi variations of the sample’s shape (green line corresponding
istic length scale λ ¼ μκ=ACW 2 (an estimate of the to x ¼ 3.3 μm), while they are merely shifted by a positive
friction lengths, see Ref. [35]). At distances from or negative offset away from these; the precise amount
the surface larger than λ we recover the flat behavior depends on the distance from the fixed-temperature boun-
typical of the bulk. We also note that these results mimic daries. These differences become more clear by inspecting
those from the (computationally very expensive) space- the longitudinal direction [Fig. 4(b)], where the discrepancy
dependent solution of the LBTE (either in the frequency- between the temperature predicted by Fourier’s law and
dependent SMA approximation [78] or considering the full Eqs. (10) and (11) is largest at x ≃ 3 μm, i.e., where the
scattering operator [35]), which generates a minimum flux sample of Fig. 3 changes geometry. We show in the inset of
on surfaces and maximum at the sample’s center. We Fig. 4(b) that also the longitudinal temperature profile (for
further note that, at variance with classical fluid dynamics y ¼ 4.5 μm) changes when going from Fourier’s law
and as pointed out in Ref. [35], the total heat flux does not (dashed red line) to the viscous heat equations (solid blue
drop to zero at the boundaries: the no-slip condition sets line). The difference is maximized close to variations of the
the drifting heat flux QD to zero, but the temperature-driven sample’s shape at x ¼ 3 μm and at the boundaries x ¼ 0
component Qδ is still allowed to be nonzero. and x ¼ 15 μm; in this latter case, the viscous heat
In Fig. 4 we plot the difference between the temperature equations predict steeper-than-Fourier’s law or nonlinear
profile predicted by Fourier’s law and the viscous heat temperature gradients that are reminiscent of those obtained
equations along transversal [Fig. 4(a)] and longitudinal in molecular dynamics simulations [105–114] and in
[Fig. 4(b)] directions. The insets in Fig. 4(b) show the explicit solutions of the LBTE [35].
results of the viscous equations (solid blue line) and
Fourier’s law (dashed red line) along the section
VI. FOURIER DEVIATION NUMBER
y ¼ 4.5 μm. Along the transversal direction [Fig. 4(a)],
Fourier’s law and Eqs. (10) and (11) predict temperature In this section, we introduce a descriptor that para-
profiles which are substantially different in the presence of metrizes the conditions under which hydrodynamic heat
011019-10
(a) 102 Graphite (b) Diamond (c) Silicon
0.5
Tviscous ) (K)
Total length ( m)
101 0.4
0.3
(TFourier
100 0.2
2
0.1
1
10 0.0
(d) 102 (e) (f)

0.12
0.10
Total length ( m)
101
FDN (adim)
0.08
0.06
100 0.04
0.02
1
10 0.00
102 103 102 103 102 103
T (K) T (K) T (K)
FIG. 5. Top row: L2 distance (see text) between the temperature profile predicted by Fourier’s law and that of the viscous heat
equations (10) and (11), shown as a function of temperature and total length of a sample similar (in the geometric sense) to the shape of
Fig. 3 for the cases of graphite (a), diamond (b), and silicon (c). The color quantifies the L2 distance between the temperature profiles
predicted by Fourier’s law [T Fourier ðx; yÞ] and the viscous heat equations (10) and (11) [T viscous ðx; yÞ], computed over the region G
corresponding to points with x > 15 lTOT , where lTOT is the total length of the sample geometrically similar to that in Fig. 3 (see main text
for details). Bottom row: Fourier deviation number [(FDN), as defined in Eq. (22)] for the same materials as a function of temperature
and total length of the sample (for the geometry of Fig. 3, the characteristic size L appearing in FDN is the shortest length scale, i.e., 1=5
of the total length lTOT ). It is apparent that the FDN correctly identifies the deviations between the solutions of Fourier’s equation and the
viscous heat equations.
conduction is observed; we refer to this as the FDN. In normalized L2 distance between the temperature
particular, we aim at distinguishing the diffusive regime profile predicted by Fourier’s law and the viscous heat
from the hydrodynamic regime: in the former case the equations for a given sample length lTOT and reference
viscous heat equations become equivalent to Fourier’s law, temperature T̄:
while in the latter case Fourier’s law no longer holds and
the viscous heat equations are required. L2 ½T Fourier − T viscous ðlTOT ; T̄Þ
In order to investigate how the hydrodynamic deviations sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
R ffi
from Fourier’s law depend on the sample’s sizes and ½T ðx; yÞ − T ðx; yÞ 2 dxdy
reference temperature T̄, we solve the viscous heat equa- ¼ G Fourier R viscous
; ð16Þ
G dxdy
tions and Fourier’s equation for several samples having
different dimensions, and at various reference temperatures
T̄. We perform this analysis for graphite [Fig. 5(a)], and in order to ease the qualitative interpretation of these
diamond [Fig. 5(b)], and silicon [Fig. 5(c)]. In particular, results later, we evaluate L2 in the spatially homogeneous
we consider samples that are similar—in the geometric region G defined by x > 15 lTOT. We also inspected the
sense—to the reference sample shown in Fig. 3; i.e., we effects of shape on the deviation L2 , finding that the
generate samples of different dimensions starting from magnitude of L2 is larger for geometries with spatial
the reference sample of Fig. 3 and varying the sizes nonhomogeneities that imply larger values of the drift
via a uniform scaling. Following this protocol, we velocity’s second derivative; nevertheless, the dependence
vary the sample’s length lTOT , reported on the y axis of of L2 on lTOT and T̄ is qualitatively unchanged for different
Figs. 5(a)–5(c), from 0.1 to 100 μm (e.g., the reference shapes.
sample of Fig. 3 corresponds to y ¼ 15 μm; y ¼ 100 μm Results in Figs. 5(a)–5(c) have been computed account-
corresponds to a sample obtained magnifying uniformly by ing for finite-size effects as discussed in Secs. II and III.
a factor of 100=15 ≃ 6.66 the sizes of the reference Additionally, to better compare with the experimental
sample). The color in Figs. 5(a)–5(c) represents the results of Ref. [43], we make the hypothesis of working
011019-11
with polycrystalline graphite with an average crystal grain (with DU;m ¼ maxi;j ðjDij ij ij
U jÞ and d ¼ DU =DU;m ); μ
ijkl ¼
size of 10 μm as in Ref. [43]: as a result, the transport μm m (with μm ¼ maxi;j;k;l ðjμ jÞ and m ¼ μ =μm ).
ijkl ijkl ijkl ijkl
coefficients of graphite are renormalized by the system size Then, we define a set of dimensionless variables, r ¼ r=L,
for lTOT < 10 μm, and by the grain size for lTOT > 10 μm u ¼ u=u0 , and T ¼ T=δT, where L, u0 , and δT are a
(therefore, for lTOT > 10 μm, size effects are given characteristic size, drift velocity, and temperature pertur-
only by the boundary conditions’ effects on the temperature bation (more on these later). Substituting these variables in
and drift-velocity fields). In graphite, the difference Eqs. (10) and (11), and limiting ourselves to the steady-
between Fourier’s law and the viscous heat equations is state regime, we obtain
largest in the temperature range 60–80 K and for sizes
pffiffiffiffiffi
10 ≲ lTOT ≲ 20 μm. X 3
wij aj ∂uj ðr Þ X 3
kij ∂ 2 T ðr Þ
Turning our attention to diamond, we first recall that π1 pffiffiffi i − ¼ 0;
it is characterized by a very large thermal conductivity, i;j¼1 gðw · aÞ2
∂r i;j¼1
gðkÞ2 ∂ri ∂rj
which originates from having large group velocities and ð17Þ
weak umklapp scattering [80–85]; the latter condition
being favorable to the emergence of hydrodynamic effects. pffiffiffiffi X 3
wji ∂T ðr Þ X 3
mijkl ∂ 2 uk ðr Þ
This is confirmed by the results shown in Fig. 5(b), with the π2 ai − π 3
j¼1
gi ðwÞ1 ∂rj j;k;l¼1
gi ðmÞ3 ∂rj ∂rl
hydrodynamic deviation being largest around room temper-
pffiffiffiffiffiffiffiffiffi
ature and dimension lTOT ≳ 1 μm (we note in passing that, X 3
ai aj dij
at this temperature and size, the finite-size renormalization ¼− pffiffiffiffiffiffiffiffiffiffiffiffi uj ðr Þ; ð18Þ
gi
ð a ⊗ a · dÞ1
of transport coefficients is small and well compatible with j¼1
the mesoscopic approach described in this work). These
results suggest that a hydrodynamic window exists also where we have introduced the quantities
for diamond, and that therefore hydrodynamic behavior pffiffiffi pffiffiffiffiffi X pffiffiffiffiffi
(e.g., second sound) might be measurable in this material at gðw · aÞ2 ¼ ðmaxij ½wij aj Þ−1 wij aj ;
ij
temperatures even larger than graphite. X
−1
For silicon, instead, the deviation from Fourier’s law is gðkÞ2 ¼ ðmaxij ½k Þ ij
kij ;
smaller and takes place at lower temperatures. Silicon is an ij
example of a material for which the thermal conductivity X
computed within the SMA approximation is very close to gi ðwÞ1 ¼ ðmaxj ½wij Þ−1 wij ;
the conductivity computed from the exact solution of the j
LBTE [23,25,27,115], and it is known that the SMA works pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi X pffiffiffiffiffiffiffiffiffi
reasonably well in systems where momentum-dissipating gi ð a ⊗ a · dÞ1 ¼ ðmaxj ½ ai aj dij Þ−1 ai aj dij ;
(umklapp) scattering events dominate over those conserv- j
ing crystal momentum (normal) [34,115]. Therefore, the X
negligible magnitude of hydrodynamic effects predicted by gi ðmÞ3 ¼ ðmaxjkl ½mijkl Þ−1 mijkl :
the viscous heat equations for silicon is consistent with the jkl
predominance of umklapp over normal scattering events
known to occur in this material. We lastly remark that, in The role of these parameters is to account for the correct
contrast to graphite, results for diamond and silicon have order of magnitude of the five summations appearing in
been computed for single crystals; i.e., transport coeffi- Eqs. (17) and (18); they carry a Cartesian superscript i
cients are not limited by the grains’ size. whenever they may depend on direction, and their
In order to capture intuitively and inexpensively all these subscripts indicate the number of indexes summed in
trends we introduce an approach inspired by the definition their definition. For example, for a 3 × 3 isotropic tensor,
of the Reynolds number, and we rewrite the viscous heat χ ij ¼ χ m δij (where χ m is a constant
P ij and δ is the Kronecker
ij
−1
equations (10) and (11) in adimensional form (we follow tensor), gðχÞ2 ¼ ðχ m Þ ij χ ¼ 3 and gi ðχÞ1 ¼
−1
P ij
the standard procedure used, e.g., in fluid dynamics, which ðχ m Þ j χ ¼ 1 ∀ i. Because we are computing the
is also called “Buckingham Pi theorem” [116,117]). First, FDN for the planar geometry discussed in Fig. 3, we can
we extract the magnitudes of the tensors appearing in the discard the z component (i.e., the Cartesian direction
viscous heat equations, factorizing the largest component: indexed by 3) from Eqs. (17) and (18). Then, for diamond,
Ai ¼ Am ai , where Am ¼ maxi ðjAi jÞ and ai ¼ Ai =Am is silicon, and in-plane graphite, transport properties are
an adimensional tensor with the largest component isotropic [i.e., all the second-rank tensors appearing in
having modulus equal to 1. Similarly, we factorize the Eqs. (17)
largest component of all other tensors: κij ¼ κ m kij (with pffiffiffi and (18) are diagonal]. From this it follows that
aÞ2 ¼ 2; gðkÞ2 ¼ 2; gi ðwÞ1 ¼ 1 ∀ i ¼ 1, 2; and
gðwp·ffiffiffiffiffiffiffiffiffiffiffiffi
κm ¼ maxi;j ðjκij jÞ and kij ¼ κij =κ m ); W ji0 ¼ W m wji (with gi ð a ⊗ a · dÞ1 ¼ 1 ∀ i ¼ 1, 2. Finally, the tensor mijkl
W m ¼ maxi;j ðjW ji0 jÞ and wji ¼ W ji0 =W m ); DijU ¼ DU;m d
ij
is not isotropic (because the thermal viscosity tensor from
011019-12
which it is derived is not isotropic), and from the values in discussed in Appendix G: the heat flux is described by
Fig. 1 it follows that gi ðmÞ1 ¼ 1 þ μxyxy =μxxxx ∀ i ¼ 1, 2, hyperbolic
p functions ffi with a characteristic length scale λ ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 for diamond and silicon, and gi ðmÞ1 ¼ 1 þ μxyxy = 2
μm κ m =Am CW m , and for a characteristic size L ≫ λ, a flat
μxxxx þ 2μxyyx =μxxxx ∀ i ¼ 1, 2 for graphite. Fourier-like heat flux profile is recovered. It is worth
The final expressions for the dimensionless parameters mentioning that π 3 ≪ 1—and thus Fourier-like behavior
π 1 , π 2 , π 3 are thus —can also emerge for very small (submicrometer) length
pffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffi scales, where ballistic effects are relevant and strongly
T̄Am CW m u0 L gðw · aÞ2 renormalize the transport coefficients, as discussed in
π1 ¼ ; ð19Þ Appendix C. The emergence of Fourier-like behavior in
κ m δT gðkÞ2
the ballistic regime (with a size-dependent thermal con-
sffiffiffiffiffiffiffiffiffi ductivity alike to that employed here) has been explained
C W m δT gi ðwÞ
π2 ¼ pffiffiffiffiffiffiffiffiffiffiffiffi1 ; ð20Þ in a recent work [118], confirming that the results for
Am T̄ LDU;m u0 g ð a ⊗ a · dÞ1
i
submicrometer sizes obtained with the aforementioned
approximated treatment of boundary scattering are quali-
μm gi ðmÞ tatively correct. If we insert the condition π 3 ≪ 1 in
π3 ¼ 2
pffiffiffiffiffiffiffiffiffiffiffiffi3 : ð21Þ
DU;m L Am g ð a ⊗ a · dÞ1
i Eq. (17), P we obtain a Fourier-like pffiffiffiffiffiffiffiffiffiffi equation for
pffiffiffithe temper-
ature: i;j f(π 1 π 2 ½gðkÞ2 gi ð a⊗a ·dÞ1 =gðw· aÞ2 gi ðwÞ1 ×
From the derivation of these parameters, it is clear that they P −1 kl j 2 i j
k;l wk ðd Þ wl )þk g½∂ T ðr Þ=∂r ∂r ¼0. So, the
i ij
can be interpreted in terms of average values (we indicate thermal conductivity is corrected by a term proportional
with h i the average over space) of physical quantities: to π 1 π 2 p(since
π 1 ∼ hQD i=hQδ i, π 2 ∼ h∂ΠT =∂ri=hð∂P=∂tÞjumkl i, and ffiffiffiffiffiffiffiffiffiffiffiffi in all thepffiffiffisetups considered here,
½gðkÞ2 gi ð a ⊗ a · dÞ1 =gðw · aÞ2 gi ðwÞ1 ¼ 1), but the
π 3 ∼ h∂ΠδE =∂ri=hð∂P=∂tÞjumkl i. Still, to evaluate these qualitative behavior is that of Fourier’s law. Numerically,
parameters we need to know the characteristic size L we find that π 1 π 2 ≪ 1 for graphite in the high-temperature
and temperature perturbation δT, and estimate the charac- limit [see Fig. 6(a)], so that the viscous heat flux is well
teristic values of the drift velocity u0. Focusing on the setup approximated by Fourier’s flux; the same limiting behavior
discussed in the previous section (Fig. 3), we have L ¼ is observed also in diamond and silicon. On the other
1
5 lTOT (corresponding to the shortest length scale of the hand, deviations from the temperature profile predicted by
geometry considered, which has total length lTOT along x) Fourier’s law appear when both π 1 and π 3 are large. In fact,
and δT ¼ 10 K. As shown in Appendix H, the character- π 3 ≳ 1 implies that the temperature gradient in Eq. (18) is
istic drift velocity u0 is found by interpolating the asymp- coupled to the second derivative of the drift velocity,
totic behavior at low (uL ) and high (uH ) temperatures, and one cannot obtain a Fourier-like equation as before;
u−1 −1 −1
0 ¼ uL þ uH . In the low-temperature limit, where
large values of π 1 imply that the drift velocity affects the
umklapp scattering is frozen, viscous effectspdetermine evolution of temperature in Eq. (17). It follows that
ffiffiffiffiffiffiffiffiffiffiffiffiffiffi Fourier’s law is mathematically recovered for crystal-
the drift velocity, and one can show that uL ¼ 37 CAx =T̄ ×
W xx0 δTL momentum dissipation dominating over viscous effects
μ þμxyxy
xxxx . In the high-temperature limit, the drift velocity is and/or large characteristic size (π 3 ≪ 1), but deviations
mainly determined by the crystal-momentum dissipation
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi between the viscous and Fourier temperature profile can be
2
rate and uH ¼ C=T̄Ax ðW xx0 =Dxx U Þ 7 ðδT=LÞ. We can
small even when viscous effects dominate over crystal-
therefore estimate the values of all the π 1 , π 2 , and π 3 momentum dissipation (π 3 ≫ 1) if the coupling between
factors. drift velocity and temperature in Eq. (17) is small (π 1 ≪ 1).
Looking at the definition of π 1 , π 2 , and π 3 , it is We note in passing that while this reasoning has been made
straightforward to identify the conditions for which here at steady state, it can be straightforwardly extended to
the viscous heat equations reduce to Fourier’s law. When the time-dependent case, considering time variations of the
π 3 ≪ 1, i.e., for crystal-momentum dissipation dominating drifting velocity slow compared to the crystal-momentum
over viscous effects and/or very large characteristic dissipation timescale.
size L, the temperature gradient in Eq. (18) is proportional On the basis of this reasoning, it is convenient to
P introduce a Fourier deviation number as
to the drift velocity: π 2 j ½wji =gi ðwÞ1 ½∂T ðr Þ=∂rj ≃
P pffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffi
− j ½ aj dij =gi ð a ⊗ a · dÞ1 uj ðr Þ. While it is intuitive 1 1 −1
FDN ¼ þ ; ð22Þ
to understand the emergence of Fourier’s law when crystal- π1 π3
momentum dissipation (e.g., due to umklapp scattering)
dominates over viscous effects, the Fourier-like behavior which provides a simple estimate of the deviations from
of large-size samples can be rationalized qualitatively Fourier’s law: the larger the FDN, the larger the deviations
recalling the simple analytical 1D solution of the viscous from Fourier’s law. In Figs. 5(d)–5(f), we plot the FDN for
heat equations in the absence of umklapp processes graphite, diamond, and silicon as a function of the sample’s
011019-13
(a) G experiments for second sound in graphite [43]. Then, we

rationalize the trend of the FDN as a function of temper-
ature in terms of the π terms entering in Eq. (22): π 1
increases with temperature for T ≲ 100 K and saturates to a
constant value at high temperature and π 3 decreases
asymptotically like T −2 . Figure 6(b) shows the analysis
of the FDN as a function of the sample’s size lTOT (varied
according to the aforementioned protocol) at fixed temper-
ature 85 K, i.e., the temperature fixed in the experiments by
Huberman et al. [43] to investigate the magnitude of
hydrodynamic effects as a function of size. Also in this
case, the FDN predicts hydrodynamic effects to be largest
in a size range that matches very well the sizes at which
(b) G hydrodynamic effects have been measured. We thus con-
clude that the present model is capable of describing
quantitatively the hydrodynamic window that has been
recently measured in graphite. In Appendix I, we also show
that for diamond extending the sizes in Fig. 5(e) up to
10 mm yields an increase of the FDN for temperatures
around 50–60 K and dimensions ∼1 mm. This result is in
qualitative agreement with the predictions for diamond’s
second-sound window performed using the reduced iso-
tropic crystal model or the Callaway model [45]. Finally,
the parameters entering in the π i (i ¼ 1, 2, 3) are exactly the
same used for the numerical solutions of the viscous heat
equations in Figs. 5(a)–5(c).
FIG. 6. Fourier deviation number (solid line) and its different VII. CONCLUSIONS
contributions π 1 and π 3 [Eq. (22)] for a graphite sample of fixed
length lTOT ¼ 10 μm as a function of temperature (a), and at fixed We have introduced a framework to describe heat con-
T̄ ¼ 85 K as a function of sample’s length lTOT (b). The shaded duction beyond Fourier’s law and to capture in the process
areas are the hydrodynamic (second-sound) windows measured the hydrodynamic transport regime where the phonon gas
at 10 μm (a) or 85 K (b) by Huberman et al. [43]. The terms π 2 assumes a drift velocity and heat propagation resembles fluid
and π 1 π 2 are also reported in (a) because in the high-temperature dynamics. Just as a perturbation of the temperature generates
limit Fourier’s law is recovered when π 1 π 2 ≪ 1 (see main text for an energy flux that is proportional to thermal conductivity,
details). Notably, the FDN predicts hydrodynamic effects to be
a perturbation of the drift velocity generates a crystal-
largest at temperatures and sizes that are in excellent agreement
with these observed in the experiments.
momentum flux, with the proportionality tensor coefficient
between the two being a thermal viscosity. We have shown
that thermal conductivity and viscosity can be determined
total length lTOT and reference temperature T̄. Remarkably, exactly and in a closed form as a sum over relaxons (i.e., the
the FDN captures the trends of the exact solution of the eigenvectors of the phonon scattering matrix). Relaxons have
viscous heat equations [Figs. 5(a)–5(c)], thus identifying a well-defined parity and even relaxons contribute exclu-
accurately the regime where hydrodynamic behavior sively to the thermal viscosity (odd relaxons contribute
emerges. The magnitude of the FDN is directly propor- exclusively to the thermal conductivity [27]).
tional to the magnitude of hydrodynamic effects: a larger Most importantly, the microscopic LBTE has been
FDN implies a larger difference between the Fourier and coarse grained into two novel mesoscopic viscous heat
viscous temperature profiles. A detailed analysis of the π equations, which are coupled equations parametrized by the
terms as a function of temperature for a graphite sample thermal conductivity and viscosity that govern the evolu-
geometrically similar to that in Fig. 3 and with a fixed tion of the temperature and drift-velocity fields. The
total length lTOT ¼ 10 μm is shown in Fig. 6(a), where viscous heat equations provide a general description of
lTOT ¼ 10 μm has been chosen to match the length fixed in heat transfer encompassing both the Fourier and the
the experiments by Huberman et al. [43] to investigate the hydrodynamic regimes, and all intermediate cases,
magnitude of hydrodynamic effects as a function of allowing for the emergence of second sound, and of
temperature. Importantly, the FDN at fixed lTOT ¼ Poiseuille-like heat flow associated to a temperature profile
10 μm predicts hydrodynamic effects to be largest at deviating from Fourier’s law. In addition, these viscous heat
temperatures that match very well those measured in recent equations show that the thermal conductivity is not
011019-14
sufficient to describe thermal transport in general terms, but where also the distribution nν ðr; tÞ appearing in Eqs. (6)
also its complementary quantity, the thermal viscosity, must and (7) has to be transformed for consistency. The
be taken into account. symmetrized scattering operator Ω̃νν0 is real and symmetric,
We have characterized the hydrodynamic behavior in and can thus be diagonalized [20,26,27]:
terms of the Fourier deviation number, a dimensionless
1X 1
parameter that quantifies hydrodynamic deviations from Ω̃ 0 θα0 ¼ θαν ; ðA3Þ
Fourier’s law. At a negligible computational cost, the FDN V ν0 νν ν τα
enables us to study the temperature-and-size window at where θαν denotes a relaxon (i.e., an eigenvector), α is the
which hydrodynamic phenomena such as second sound relaxon index, and the inverse eigenvalue τα is the relaxon
emerge. lifetime. We also define a scalar product [27],
The full solution of the viscous heat equations allows us to
predict measurable temperature and heat flux profiles in 1 X α α0
hαjα0 i ¼ θ θ ¼ δα;α0 ; ðA4Þ
complex-shaped devices in the mesoscopic heat transport V ν ν ν
regime at a much reduced cost, and more transparently than
solving the full LBTE, therefore paving the way toward used to normalize eigenvectors.
understanding shape and size effects in complex and micro- In order to show that eigenvectors with zero eigenvalues
scopic devices, and especially when novel physics arises, are related to conserved quantities in the LBTE dynamics,
such as is the case in phononic devices [119–131]. These we rewrite the scattering operator distinguishing scatte-
results are particularly relevant for the emerging field of ring events that conserve crystal momentum—normal
materials that display hydrodynamic thermal transport, often (N)—from those that do not—ukmlapp (U):
characterized by large thermal conductivities [31,39–46]. Ω̃νν0 ¼ Ω̃Nνν0 þ Ω̃U ðA5Þ
νν0 :
Such behavior has been investigated in devices made of
either graphite, diamond, or silicon. For graphite, the present As stated in the main text, there are four eigenvectors with
formulation predicts the hydrodynamic windows (as a zero eigenvalues, which we discuss in the following
function of temperature for a fixed size, or as a function of sections.
size for a fixed temperature) in excellent agreement with
recent experimental findings [43], while suggesting that 1. Bose-Einstein eigenvector: Local temperature
hydrodynamic behavior can also appear in diamond at room Applying the transformation Eq. (A2) to Eq. (5) and
temperature for micrometer-sized crystals. considering the steady-state conditions, one obtains the
Finally, we remark that the present methodology can be following equation for the even part (see below for the odd
adapted to describe viscous phenomena for electronic part):
conduction [132–135], or any other transport phenomena D
described by a linearized Boltzmann equation. vν ∂N ν
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · · ∇u
N̄ ν ðN̄ ν þ 1Þ ∂u
ACKNOWLEDGMENTS 1X
¼− Ω̃ 0 ½ñT0 ðr; tÞ þ ñδE
ν0 : ðA6Þ
We acknowledge support from the Swiss National V ν0 νν ν
Science Foundation under Project No. P2ELP2_168546
and the MARVEL NCCR. Computational resources have The distribution ñTν ðr; tÞ is obtained applying the symmet-
been provided by the Swiss National Supercomputing rization (A2) to the distribution nTν ðr; tÞ appearing in Eq. (4)
Center (CSCS) and PRACE. and it is thus evident that ñTν ðr; tÞ ∝ ων ½Tðr; tÞ − T̄. From
the energy conservation in scattering events (both normal
and umklapp), it follows that [20,136]
APPENDIX A: EIGENVECTORS OF THE
SCATTERING MATRIX 1X
Ω̃ 0 ñT0 ðr; tÞ
The scattering matrix Ωνν0 appearing in Eq. (1) is not V ν0 νν ν
symmetric, but it can be recast in a symmetric (and thus
1X 1 ∂ N̄ 0
diagonalizable) form by means of the following trans- ¼ Ω̃ 0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ν ½Tðr; tÞ − T̄
formation [20,26,27]: V ν0 νν N̄ ν0 ðN̄ ν0 þ 1Þ ∂T
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1X ℏων0
¼ Ω̃ 0 N̄ ν0 ðN̄ ν0 þ 1Þ ½Tðr; tÞ − T̄ ¼ 0:
N̄ ν0 ðN̄ ν0 þ 1Þ V ν0 νν kB T̄ 2
Ω̃νν0 ¼ Ωνν0 ; ðA1Þ
N̄ ν ðN̄ ν þ 1Þ
ðA7Þ
nν ðr; tÞ
ñν ðr; tÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ðA2Þ As a result, we identify ∂ N̄ ν =∂T as an eigenvector θ0ν with
N̄ ν ðN̄ ν þ 1Þ zero eigenvalue, that, after normalization, is
011019-15
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
0 N̄ ν ðN̄ ν þ 1Þ kB T̄ 2 ∂ N̄ ν 1 kB T̄ ∂N D N̄ ν ðN̄ ν þ 1Þ i
θν ¼ 2
ℏων ¼ ; ðA8Þ ϕiν ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ν
¼ ℏq ;
kB T̄ C CN̄ ν ðN̄ ν þ 1Þ ∂T N̄ ν ðN̄ ν þ 1Þ A ∂u
i i
kB T̄Ai
where the specific heat C is ðA12Þ

where i ¼ 1, 2, 3 and Ai is a normalization constant. The
∂E 1 X
C¼ ¼ N̄ ðN̄ þ 1Þðℏων Þ2 : ðA9Þ drift eigenvectors Eq. (A12) are, in general, not orthogonal
∂T eq kB T̄ 2 V ν ν ν
[21]. Nevertheless, we work in a Cartesian coordinate
From Eq. (A7), it follows that ñTν ðr; tÞ disappears from system for q0 and uðr; tÞ, so that these 3 eigenvectors are
Eq. (A6). Therefore, Eq. (A6) gives Eq. (7) in the main text. orthogonal and can also be normalized, choosing Ai from
In the context of the decomposition [Eq. (A5)], the Bose- the condition hϕi jϕi i ¼ 1. In computing the normalization
Einstein eigenvector Eq. (A8) is an eigenvector to both the constants Ai , we note that they can be expressed in terms of
normal and umklapp scattering operator, and will be physically meaningful quantities. In particular, we note that
denoted as ϕ0ν when we later consider the basis of the crystal momentum-density associated to the drifting
eigenvectors of the normal scattering operator in distribution is
Appendix D. We conclude by noting that these consid- 1X D
P¼ N ℏq; ðA13Þ
erations approximately hold also when the approximate V ν ν
treatment of scattering with boundaries discussed in Sec. III
(Ω̃U;boundary ) is included as Ω̃U þ Ω̃U;boundary and its derivative with respect to the drift velocity is
νν0 ¼ Ω̃νν0
U;bulk
νν0 νν0 ,

provided the phenomenological boundary-scattering term ∂Pi 1 X ∂N D ν

¼ ℏqi
—which does not strictly conserve energy—is small ∂u eq V ν ∂u eq
j j
compared to the bulk umklapp scattering matrix; this
happens e.g., in the mesoscopic regime where the finite- 1 X
¼ N̄ ðN̄ þ 1Þℏqi ℏqj : ðA14Þ
size renormalization of transport coefficients is small and kB T̄V ν ν ν
the viscous heat equations are valid.
Comparing Eq. (A12) with Eq. (A14), we note that Ai ¼
ð∂Pi =∂ui Þjeq and ð∂Pi =∂uj Þjeq ¼ 0 ⇔ i ≠ j. Therefore, we
2. Drift eigenvectors: Local drift velocity
refer to Ai as the specific momentum, due to its formal
Starting from Eq. (5) at the steady state, one obtains the similarity with specific heat. It can be shown that in the high-
following equation for the odd part: temperature limit, Ai ðT → ∞Þ ∝ T (see also Appendix E).
Finally, it is worth mentioning that in going from Eq. (5)
vν ∂ N̄ ν
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · ∇T to Eq. (6), the term nD ν ðr; tÞ has disappeared due to the
N̄ ν ðN̄ ν þ 1Þ ∂T following approximation:
1X 1X
¼− Ω̃ 0 ½ñD0 ðr; tÞ þ ñδO
ν0 : ðA10Þ Ω̃ 0 ½ñD0 ðr; tÞ þ ñδO
V ν0 νν ν ν0
V ν0 νν ν
The distribution ñD ν ðr; tÞ is obtained applying the symmet- 1X N δO
¼ ðΩ̃νν0 þ Ω̃U νν0 Þ½ñν0 ðr; tÞ þ ñν0
D
rization Eq. (A2) to the distribution nD ν ðr; tÞ appearing
P3 V ν0
in Eq. (4). We note in particular that ñD ν ðr; tÞ ∝
i
i¼1 q · 1X U D 1X
u ðr; tÞ.
i
¼ Ω̃νν0 ñν0 ðr; tÞ þ Ω̃ 0 ñδO
V ν0 νν ν
0
V ν0
The drifting distribution ñD ν ðr; tÞ is the stationary
distribution for a system conserving crystal momentum. 1X
≃ Ω̃ 0 ñδO0 : ðA15Þ
Therefore, recalling the decomposition Eq. (A5), we have V ν0 νν ν
that
This is correct at both high and low temperatures, because at
1X N D
Ω̃ 0 ñ 0 ðr;tÞ high temperatures the strong crystal-momentum dissipation
V ν0 νν ν
ν0 ðr; tÞ ∝ uðr; tÞ ≈ 0 (see also Fig. 10), and at low
ensures ñD
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1X N ℏ 0 temperatures umklapp processes are much less frequent than
¼ Ω̃νν0 N̄ ν0 ðN̄ ν0 þ 1Þ q · uðr;tÞ ¼ 0; ðA11Þ normal processes (Ω̃U νν0 ≪ Ω̃νν0 ).
N
V ν0 kB T̄
since Ω̃Nνν0 accounts only for normal scattering events that 3. Local equilibrium
conserve crystal momentum [13,16,20]. From Eq. (A11) it From Eq. (A7) and Eq. (A15) it follows that, in the
is possible to identify three eigenvectors of Ω̃Nνν0 with zero hydrodynamic regime, the distributions ñTν ðr; tÞ and
eigenvalues [20]: ν ðr; tÞ are left unchanged by the dynamics described
ñD
011019-16
by the LBTE; therefore, these are local equilibrium dis- Eqs. (A7) and (B2), the Bose eigenvector is related to
tributions. It follows that ñTν ðr; tÞ and ñD
ν ðr; tÞ do not appear the local-equilibrium temperature and does not contribute
in Eqs. (6) and (7) and thus do not contribute to the thermal to the out-of-equilibrium response to a drift-velocity
conductivity and viscosity, which, respectively, describe perturbation. Substituting this relation in the LBTE, and
the response to a perturbation of the local temperature and noting that the left-hand term is related to the eigenvector
drift velocity. It is worth mentioning that, in the kinetic ϕiν of the normal scattering matrix [Eq. (A12)], we obtain
regime, ñDν ðr; tÞ vanishes and ñν ðr; tÞ is still a stationary sffiffiffiffiffiffiffiffi
T
distribution for the LBTE. j Ai i X 1 ij β

vν ϕν ¼ − f θν : ðB5Þ
kB T̄ τ β
β>0 β
APPENDIX B: THERMAL VISCOSITY
Taking the scalar product with a generic eigenvector θαν
The total crystal-momentum flux tensor Πij tot [20] is (with α > 0), we find
defined as sffiffiffiffiffiffiffiffi
1X i j Ai j f ij
α
Πij
tot ðr; tÞ ¼ ℏq vν N ν ðr; tÞ: ðB1Þ wiα ¼ − ; ðB6Þ
V ν kB T̄ τα
Because of the odd parity of qi and vjν , only the even part where wjiα is a velocity tensor given by
of the phonon distribution contributes to the crystal-
1X i j α
momentum flux. Using the decomposition Eq. (4) intro- wjiα ¼ ϕ vν θ : ðB7Þ
duced in the main text, we identify three contributions to V ν ν ν
the crystal-momentum flux:
Thanks to the odd parity of ϕiν and vjν , the velocity wjiα is
1X i j different from zero only for even eigenvectors α; Eq. (B6)
Πij
tot ðr;tÞ ¼ ℏq vν ½N̄ ν þ nTν ðr;tÞ þ nδE
ν ðr;tÞ
V ν can thus be trivially solved for f ij α.
¼ Π̄ij þ Πij ij With the knowledge of the LBTE solution f ij α at hand,

T ðr;tÞ þ ΠδE ; ðB2Þ
the crystal-momentum flux tensor is readily computed. We
where Π̄ij is the equilibrium (constant) crystal-momentum thus express Πij δE in the relaxon basis, finding
flux, which is not affected by the LBTE’s dynamics,
Πij 1 X δE i j
T ðr; tÞ is the momentum flux related to the local-equi- Πij
δE ¼ n ℏq vν
librium temperature, and Πij δE ðr; tÞ is the out-of-equilibrium
V ν ν
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
momentum flux generated in response to deviations from 1 X δE
local-equilibrium conditions and is further discussed below. ¼ ñν N̄ ν ðN̄ ν þ 1Þℏqi vjν
V ν
As stated by Eq. (8), the unsymmetrized thermal
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
viscosity tensor ηijkl relates a drift-velocity perturbation 1 X kl α ∂uk
to the momentum flux generated as a response to that ¼ f α θν l N̄ ν ðN̄ ν þ 1Þℏqi vjν
V kl;ν;α>0 ∂r
perturbation. Equation (8) is valid in the mesoscopic regime
mentioned in Secs. II and V, where the thermal viscosity is pffiffiffiffiffiffiffiffiffiffiffiffiffi X j ∂u
k
space-independent and nonhomogeneities arise mainly ¼ kB T̄Ai f kl w
α iα : ðB8Þ
kl;α>0
∂rl
from variations of the local temperature and local drift
velocity. In particular, ηijkl is determined by deviations Substituting Eq. (B6) in Eq. (B8), we obtain the expression
from local equilibrium and thus depends only on ñδE ν . To for the asymmetric thermal viscosity tensor:
determine the distribution ñδE ν , we must solve the LBTE pffiffiffiffiffiffiffiffiffiffiX
linearized in the drift-velocity gradient Eq. (7). To this aim, ηijkl ¼ Ai Ak wjiα wlkα τα : ðB9Þ
we first symmetrize the LBTE using the transformations α>0
(A1) and (A2), finding
D The tensor ηijkl is, in general, not symmetric under
vν ∂N ν 1X exchange of indexes j ↔ l. We show later that, in a way
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · · ∇u ¼ − Ω̃ 0 ñδE0 : ðB3Þ
N̄ ν ðN̄ ν þ 1Þ ∂u V ν0 νν ν analogous to fluids, only the sum over the spatial deriv-
Next, using the relaxon approach discussed in Ref. [27] for atives of the momentum flux tensor is relevant for the
thermal conductivity, we write the response to the pertur- mesoscopic description of hydrodynamic thermal transport:
bation ∇u as a linear combination of even eigenvectors:
X ∂Πij ðr; tÞ X ∂ 2 uk ðr; tÞ
X ij β ∂ui δE
¼− ηijkl
ñδE
ν ¼ f β θν j ; ðB4Þ j
∂r j
∂rj ∂rl
ij; β>0
∂r jkl
X ∂ 2 uk ðr; tÞ
where the summation does not involve the Bose eigenvec- ¼− μijkl ; ðB10Þ
∂rj ∂rl
tor ðβ ¼ 0Þ since, as discussed in the Appendix for jkl
011019-17
where in the last term of Eq. (B10) we have rewritten the where we highlight the fact that in the SMA approximation
divergence of the momentum flux tensor using the sym- ηijkl ijkl ijkl
SMA ¼ μSMA , since in the SMA the viscosity tensor ηSMA
metrized viscosity tensor μijkl : has the j ↔ l symmetry. A comparison between the exact
bulk thermal viscosity Eq. (9) and the SMA bulk thermal
ηijkl þ ηilkj viscosity Eq. (B16) is shown in Fig. 7. We highlight how
μijkl ¼ : ðB11Þ
2 the SMA approximation—which neglects the off-diagonal
elements of the scattering operator and works well when
It is worth drawing a parallel between thermal viscosity umklapp processes dominate over normal processes
and conductivity, where the latter can be written as [27]
X
κ ij ¼ C wi0α wj0α τα : ðB12Þ (a)
α
Notably, wi0α is different from zero only for odd eigenvec-

tors. As a result, thermal conductivity and viscosity are two
quantities describing the transport due to the odd and even
part of the spectrum, respectively, i.e., energy and crystal
momentum.
1. Single-mode relaxation-time approximation

In this section, we derive the expression for thermal
viscosity within the single-mode relaxation-time approxi- × ×
mation. Using the SMA, the LBTE at Eq. (B3) becomes
D (b)
vν ∂N ν ñδE
ν
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · · ∇u ¼ − SMA : ðB13Þ
N̄ ν ðN̄ ν þ 1Þ ∂u τν
The deviation from equilibrium ñδE

ν is readily found as
D
vν ∂N ν
ñδE
ν ¼ − pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi · · ∇u τSMA
ν : ðB14Þ
N̄ ν ðN̄ ν þ 1Þ ∂u
We insert this result in the definition of momentum flux,

obtaining
× ×
1 X
Πij
δE;SMA ¼ nδE i j
ν ℏq vν
V ν (c)
1 X 2 i j k l N̄ ν ðN̄ ν þ1Þ SMA ∂uk
¼− ℏ q vν q vν τν ; ðB15Þ
V ν;kl kB T̄ ∂rl
where we stress that, due to the lack of energy conservation

in phonon collisions within the SMA treatment [68], in
Eq. (B15) it has not been possible to separate, as in
Eq. (B2), the momentum flux generated from a local
temperature from that arising from a drift-velocity pertur-
bation. From the definition of the thermal viscosity Eq. (8),
the SMA thermal viscosity is therefore
×
ηijkl þ ηilkj 1 X 2 i j k l N̄ ν ðN̄ ν þ 1Þ SMA

μijkl
SMA ¼
SMA SMA
¼ ℏ q vν q vν τν ;
2 V ν kB T̄ FIG. 7. Comparison between the exact bulk thermal viscosity
(solid lines) and the SMA bulk thermal viscosity (dashed line) for
ðB16Þ in-plane graphite (a), diamond (b), and silicon (c).
011019-18
[16,25,34]—overestimates the largest component of the when the mean-free path for normal collisions ΛN is much
thermal viscosity, especially at low temperatures. This smaller than the boundary scattering’s length LS (for a
overestimation is more pronounced in diamond compared single crystal LS is the sample’s size, in the case of a
to silicon, in agreement with results from previous works polycrystalline sample LS is the size of the grains) or the
where the SMA approximation was reported to yield quite mean-free path for umklapp collisions ΛU : ΛN ≪ LS ; ΛU
accurate results for the thermal conductivity of silicon [27] [15,31]. Under these conditions, the local equilibrium is
but not for diamond [25]. expressed in terms of the four special eigenvectors ϕ0ν (also
denoted θ0ν , since this is a common eigenvector for the full
APPENDIX C: SIZE EFFECTS ON THE and normal scattering operator, see Appendix A), ϕiν
THERMAL CONDUCTIVITY AND VISCOSITY (i ¼ 1, 2, 3) described in Appendixes A 1 and A 2, and
In order to obtain a simple estimate of size effects, we of the local temperature Tðr; tÞ and drift-velocity uðr; tÞ
compute the total effective thermal conductivity and fields. As explained in Appendixes A 1 and A 2, these four
viscosity using a Matthiessen sum of the bulk and ballistic special eigenvectors do not contribute to thermal conduc-
limit: tivity and viscosity, since these latter coefficients only
describe the response to a perturbation of the local
1 1 1 equilibrium.
¼ ij þ ij ; ðC1Þ In order to exploit the relationship between the drift
κ ij
κbulk κballistic
velocity uðr; tÞ and the drift eigenvectors ϕiν , we choose to
1 1 1 work with the basis of eigenvectors of the normal scattering
¼ þ : ðC2Þ matrix Ω̃Nνν0 . To this aim, we diagonalize Ω̃Nνν0 as
μijkl μijkl
bulk μijkl
ballistic
1X N β 1
Ω̃ 0 ϕ 0 ¼ N ϕβν ; ðD1Þ
The ballistic conductivity and viscosity are computed for a V ν0 νν ν τβ
sample of size LS as κij ij ijkl
ballistic ¼ K S · LS and μballistic ¼
Mijkl · LS , where the prefactors K S and M ijkl are where β ≥ 0 is an integer that labels the eigenvectors ϕβν
and eigenvalues 1=τNβ . Among these eigenvectors, we know
1X N̄ ðN̄ þ 1Þ i j 1
K ij
S ¼ ðℏων Þ2 ν ν 2 vν vν ; ðC3Þ the analytic expression for the 4 of them associated with
V ν kB T̄ jvν j energy and momentum conservation, which we label with
β ¼ 0 for the energy eigenvector [in this section, we use ϕ0ν
1 X 2 i j k l N̄ ν ðN̄ ν þ 1Þ 1 to label the Bose-Einstein eigenvector Eq. (A8)], and
Mijkl ¼ ℏ q vν q vν : ðC4Þ β ¼ 1; …; 3 for the momentum eigenvectors, Eq. (A12).
V ν kB T̄ jvν j
Noting that the set of “normal” eigenvectors fϕβν g
These prefactors are obtained after setting τν ¼ LS =jvν j in (“normal” in the sense that they diagonalize Ω̃Nνν0 , the part
the SMA expressions of κ and μ. LS is the length that of the scattering matrix that accounts for normal processes
determines boundary effects, i.e., the sample’s size for a only) is a complete basis set [16], we write the deviation
single crystal or the grain size for a polycrystalline sample. from equilibrium ñν ðr; tÞ as a linear combination of these
The numerical values of K ij S and M
ijkl
can be computed eigenvectors:
from first-principles and are tabulated in Tables I, II, III. X
ñν ðr; tÞ ¼ zβ ðr; tÞϕβν : ðD2Þ
β
APPENDIX D: VISCOUS HEAT EQUATIONS
In this Appendix, we derive an extension to Fourier’s law After symmetrising the LBTE [Eq. (1)] according to
from the LBTE, which describes mesoscopic hydrody- Eqs. (A1) and (A2), we use Eq. (D2) to write the LBTE
namic thermal transport in terms of the temperature Tðr; tÞ in the basis of the eigenvectors of the normal scattering
and drift-velocity uðr; tÞ fields. We focus on the meso- operator:
scopic regime where transport coefficients are well-defined X ∂zβ ðr; tÞ X
(i.e., the finite-size renormalization of transport coefficients ϕβν þ vν · β
∇zβ ðr; tÞϕν
is small) and nonhomogeneities arise from variations of the β
∂t β
temperature and drift velocity. In contrast to Sec. II, here X zβ ðr; tÞ β 1 X β
the temperature and drift-velocity gradients are space ¼− ϕν − zβ ðr; tÞΩ̃U
νν0 ϕν0 : ðD3Þ
dependent, and consequently the LBTE is not linearized β>3
τ N
β V 0
ν ;β>0
in the temperature and drift-velocity gradients. We start by
recalling that hydrodynamic thermal transport emerges This equation is formally equivalent to the LBTE, but
when most collisions between phonon wave packets con- allows us to take advantage of the knowledge of the first 4
serve the crystal momentum. This can happen, for example, eigenvectors to derive mesoscopic equations.
011019-19
1. Projection of the LBTE on the first 1X

Qðr; tÞ ¼ v ℏω N ðr; tÞ
(Bose-Einstein) eigenvector: Energy moment V ν ν ν ν
Here we show how to obtain an energy balance equation. 1X
¼ v ℏω ½nD ðr; tÞ þ nδO
ν ðr; tÞ; ðD10Þ
First, we note from Eq. (4) that the phonon population V ν ν ν ν
expansion of Eq. (D2) can be recast as
ñν ðr; tÞ ¼ ñTν ðr; tÞ þ ñD δ where we used the fact that only odd components of the
ν ðr; tÞ þ ñν ðr; tÞ
sffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffi phonon distribution contribute to the heat flux. Therefore,
C 0 X 3
Ai i i the heat flux receives contributions from both the drift
¼ ϕν ½Tðr; tÞ − T̄ þ ϕ u ðr; tÞ
kB T̄ 2 i¼1
kB T̄ ν velocity [104] and the temperature gradient [27]. In the
X β basis of normal eigenvectors, the drifting contribution can
þ ϕν zβ ðr; tÞ: ðD4Þ be written as
β>3
1X i
The coefficients zβ in front of the 4 special eigenvectors of QD;i ðr; tÞ ¼ v ℏω nD ðr; tÞ
V ν ν ν ν
Ω̃Nνν0 are associated to temperature and drift velocity, which
fully determine local equilibrium; in detail, 1X i ∂N D
ν
¼ vν ℏων · uðr; tÞ
sffiffiffiffiffiffiffiffiffiffi V ν ∂u
C 1 X i 0 pffiffiffiffiffiffiffiffiffiffiffi ffi
z0 ðr; tÞ ¼ ½Tðr; tÞ − T̄; ðD5Þ ¼ vν θν T̄Aj Cϕjν uj ðr; tÞ
kB T̄ 2 V ν;j
sffiffiffiffiffiffiffiffi X pffiffiffiffiffiffiffiffiffiffiffiffi
¼ T̄Aj C W i0j uj ðr; tÞ; ðD11Þ
Ai i
zi ðr; tÞ ¼ u ðr; tÞ; i ¼ 1; 2; 3: ðD6Þ j
kB T̄
while for the contribution from the deviation from local
We now project the LBTE Eq. (D3) in the subspace equilibrium nδO
ν ðr; tÞ, we find
spanned by the Bose-Einstein eigenvector ϕ0ν ; i.e., we take
the scalar product of Eq. (D3) with ϕ0ν , finding 1X i
QδO;i ðr; tÞ ¼ v ℏω nδO ðr; tÞ
V ν ν ν ν
∂z0 ðr; tÞ X qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
þ W 0β · ∇zβ ðr; tÞ ¼ 0; ðD7Þ 1X i
∂t ¼ v ℏω N̄ ν ðN̄ ν þ 1ÞñδO ν ðr; tÞ
β>0
V ν ν ν
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where we used the fact that ϕ0ν is an eigenvector of zero CkB T̄ 2 X X
eigenvalue to Ω̃U ¼ zβ ðr; tÞϕ0ν viν ϕβν
νν0 (see Appendix A 1) and we defined the V β>3 ν
velocity tensor: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiX
1X α j β ¼ CkB T̄ 2 W i0β zβ ðr; tÞ; ðD12Þ
W jαβ ¼ ϕ vν ϕν : ðD8Þ β>3
V ν ν
where only odd eigenvectors contribute (W i0β ¼ 0 for even
Note that the velocity W jαβ
differs from the velocity wjαβ β eigenvectors). As explained in Appendix A 3, the thermal
introduced in Sec. II for thermal viscosity; the difference conductivity is determined only from odd eigenvectors that
arises from the use in Eq. (D8) of the normal eigenvectors are not related to local equilibrium (that is, all the odd
(of Ω̃Nνν0 ) rather than the general eigenvectors of Ω̃νν0 eigenvectors minus the three drift eigenvectors). At this
(W jαβ ≠ wjαβ in the presence of umklapp processes). point it is crucial to recall that the relaxons have a well-
Substituting Eqs. (D5) and (D6) in Eq. (D7), we obtain defined parity (even or odd), deriving from the symmetries
of the full scattering operator Ω̃νν0 ¼ Ω̃−ν;−ν0 [20,138].
sffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffi
C ∂Tðr; tÞ X 3
Ai j ∂ui ðr; tÞ Because the same symmetries apply to Ω̃Nνν0 , also the
þ W eigenvectors of the normal scattering operator have a
kB T̄ 2 ∂t i;j¼1
kB T̄ 0i ∂rj
well-defined parity [20,21,138]. The odd component of
X
þ W 0β · ∇zβ ðr; tÞ ¼ 0: ðD9Þ the heat flux (and hence the thermal conductivity) is
β>3 determined only from the odd part of the LBTE’s solution
ñδO
ν ðr; tÞ [27,138], and this can be written either as a linear
To elucidate the meaning of this equation, we note that the combination of odd relaxons (as done in Ref. [27]) or,
harmonic heat flux can be written as [137] equivalently, as a linear combination of odd eigenvectors
011019-20
of the normal scattering matrix, as done here. From the Using the expression of the four special eigenvectors
completeness of these basis sets it follows that the heat flux discussed in Appendixes A 1 and A 2, we rewrite these
Eq. (D12) arising from the odd out-of-equilibrium phonon two momentum fluxes in the basis of eigenvectors of the
distribution determined from Eq. (6), and described as a normal scattering matrix, finding
linear combination of odd eigenvectors of the normal
scattering matrix, is equivalent to the heat flux of 1X i j T
Πij
T ðr; tÞ ¼ ℏq vν nν ðr; tÞ
Ref. [27] that is written as a linear combination of odd V ν
relaxons (eigenvectors of the full scattering matrix). In an 1 X ∂ N̄ ν i j
equivalent way, the equation for the odd part of the LBTE ¼ ℏq vν ½Tðr; tÞ − T̄
[Eq. (6)] is equivalent to the equation used to determine the V ν ∂T
rffiffiffiffiffiffiffiffi
thermal conductivity in Ref. [27], the only difference being 1 CAi X 0 j i
the basis chosen for the decomposition of the solution (the ¼ ϕ vν ϕν ½Tðr; tÞ − T̄
V T̄ ν ν
relaxons in Ref. [27] and in the computation of the thermal rffiffiffiffiffiffiffiffi
viscosity reported in Fig. 1, and the eigenvectors of the CAi j
normal scattering matrix here). The choice of the basis set ¼ W ½Tðr; tÞ − T̄ ðD15Þ
T̄ i0
does not affect the resulting local odd heat flux, which can
therefore be related to the local temperature gradient via the and
thermal conductivity [27]:
1 X i j δE
X
3 Πij
δE ðr; tÞ ¼ ℏq vν nν ðr; tÞ
V ν
QδO;i ðr; tÞ ¼ − κ ij ∇j Tðr; tÞ: ðD13Þ pffiffiffiffiffiffiffiffiffiffiffiffiffi
j¼1 kB T̄Ai X β j i
¼ ϕν vν ϕν zβ ðr; tÞ
In Eq. (D13) the thermal conductivity is space-independent; V ν;β>3
i.e., Eq. (D13) is valid in the mesoscopic regime where the pffiffiffiffiffiffiffiffiffiffiffiffiffiX
¼ kB T̄Ai W jiβ zβ ðr; tÞ; ðD16Þ
thermal conductivity and viscosity are space-independent
β>3
and nonhomogeneities arise solely from variations of the
local temperature and local drift velocity. Equations (D12) where we used the velocity tensor defined in Eq. (D8).
and (D13) can be used to rewrite Eq. (D9) in terms of the local Next, as in the previous section, we take the scalar
temperature and drift-velocity fields, obtaining Eq. (10) of product of Eq. (D3) with ϕiν (i ¼ 1, 2, 3), obtaining the
the main text: following three equations indexed by i ¼ 1, 2, 3,
∂Tðr; tÞ X 3 pffiffiffiffiffiffiffiffiffiffiffiffi ∂uj ðr; tÞ X
C þ W i0j T̄Aj C ∂zi ðr; tÞ
∂t ∂ri þ W i0 · ∇z0 ðr; tÞ þ W iβ · ∇zβ ðr; tÞ
i;j¼1 ∂t β>3
X
3
∂ 2 Tðr; tÞ X
3 X
− κ ij ¼ 0: ðD14Þ ¼− zj ðr; tÞDij zβ ðr; tÞDiβ
∂ri ∂rj U − U; ðD17Þ
i;j¼1 j¼1 β>3
Equation (D14) is clearly not sufficient to fully describe

where we used the fact that ϕiν are eigenvectors of Ω̃Nνν0 with
the hydrodynamic heat conduction problem in which
both Tðr; tÞ and uðr; tÞ are nonzero. In the next section, zero eigenvalues and we defined
1 X i U β
we derive a complementary set of equations that completes
the formulation. Diβ
U ¼ ϕν Ω̃νν0 ϕν0 : ðD18Þ
V 2 νν0
2. Projection of the LBTE on the second, third, iβ
and fourth eigenvectors: The momentum eigenspace From the property Ω̃U νν0 ¼ Ω̃−ν;−ν0 it can be shown that DU
U
vanishes when β indexes an even eigenvector. Since the

In this section, we derive a set of balance equations for
coefficients zβ ðr; tÞ for the first four eigenvectors (β ¼ 0, 1,
crystal momentum. We start by recalling from Eq. (B2) that
the momentum flux receives contributions from three 2, 3) are known, it is convenient to rewrite Eq. (D17) as
different terms. Of these three, the first term is a constant X
∂zi ðr; tÞ
related to the equilibrium temperature, and thus is not þ W i0 · ∇z0 ðr; tÞ þ ðW iβ · ∇ þ Diβ
U Þzβ ðr; tÞ
changed by the LBTE. Therefore, we focus only on the ∂t β>3
momentum flux related to the local equilibrium temper- X
3
ature Πij
T ðr; tÞ and the out-of-equilibrium momentum flux ¼− zj ðr; tÞDij
U: ðD19Þ
generated in response to a drift-velocity gradient ΠijδE ðr; tÞ.
j¼1
011019-21
X
In the hydrodynamic regime, umklapp momentum dissi- ∂zi ðr;tÞ X 3
iβ ij
pation is weak and thus Diβ U → 0 (ϕν is approximately
i ∂t þ ðW iβ · ∇ þ DU Þzβ ðr;tÞ ≪ − zj ðr;tÞDU ;
β>3 j¼1
an eigenvector with a vanishing eigenvalue for Ω̃U
νν0 ). There-
fore, we simplify Eq. (D19), noting that ðD23Þ

X X implying that the viscous heat equations reduce to Fourier’s
W iβ · ∇zβ ðr; tÞ ≫ Diβ zβ ðr; tÞ: ðD20Þ
U law, as discussed in Sec. VI.
β>3 β>3
From a mathematical point of view, the projection of the
LBTE in the Bose-Einstein subspace, performed comput-
Then, we use the expression of the coefficients zβ ðr; tÞ ing the scalar product between the LBTE in the normal
(β ¼ 0, 1, 2, 3), and substitute Eqs. (D5) and (D6) in the eigenvectors basis Eq. (D3) and the Bose-Einstein eigen-
simplified Eq. (D19), obtaining vector Eq. (A8) ∝ ων , is equivalent to calculating the
sffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffi energy moment of the LBTE. Analogously, the projection
Ai ∂ui ðr; tÞ C X 3
∂Tðr; tÞ in the momentum subspace, performed calculating the
þ 2
W ji0 scalar product between the LBTE Eq. (D3) and the
kB T̄ ∂t kB T̄ j¼1 ∂rj
momentum eigenvectors Eq. (A12) ∝ qi , is equivalent to
1 X ∂Πij ðr; tÞ computing the momentum moment of the LBTE.
δE
þ pffiffiffiffiffiffiffiffiffiffiffiffiffi
kB T̄A j i ∂rj
sffiffiffiffiffiffiffiffi
X3
Aj ij j APPENDIX E: PARAMETERS ENTERING
¼− D u ðr; tÞ: ðD21Þ
j¼1
kB T̄ U THE VISCOUS HEAT EQUATIONS
We report in this section the numerical values of the
Next, we note that only even eigenvectors different from the parameters needed to solve the viscous heat equations.
Bose-Einstein eigenvector determine the even distribution Because of the crystal symmetries of the materials con-
nδE
ν ðr; tÞ appearing in the expression for the out-of-
sidered, several components of vectors and tensors are
equilibrium momentum flux tensor Eq. (D16). As shown equivalent. Second-rank tensors such as thermal conduc-
in Eq. (B10), in the mesoscopic regime the sum over the tivity (κ ij ), momentum dissipation (Dij j
U ), and velocity (W 0i )
spatial derivatives of Πij
δE ðr; tÞ can be expressed in terms of
are isotropic and diagonal for diamond and silicon; for
the viscosity and second derivative of the drift velocity. We graphite they are isotropic and diagonal for in-plane (x, y)
thus find directions (as used in the geometry of Fig. 3). Similarly, the
specific momentum Ai is isotropic for diamond and silicon,
rffiffiffiffiffiffiffiffi 3
CAi X j ∂Tðr; tÞ
and independent from the in-plane direction in graphite
∂ui ðr; tÞ
A i
þ W (Ax ¼ Ay ). The values of Dij i j
U; bulk , A , C, and W 0i as a
∂t T̄ j¼1 i0 ∂rj
function of temperature are plotted in Fig. 8 (we only report
X
3
∂ 2 uk ðr; tÞ the in-plane or isotropic components, indexes are omitted).
− μijkl The numerical values of these parameters can be found in
j;k;l¼1
∂rj ∂rl
Tables I, II, III; in addition, these tables contain also the
3 pffiffiffiffiffiffiffiffiffiffi
X prefactors discussed in Secs. II and III and Appendix C
¼− Ai Aj Dij U u ðr; tÞ:
j
ðD22Þ needed to compute the thermal conductivity and viscosity
j¼1
in the ballistic limit (κ ij ijkl
ballistic and μballistic , respectively). In
particular, the parameters K ij S and M
ijkl
are defined as
Combining this with Eq. (D14) we obtain 4 equations to ij ij ijkl
be solved in terms of temperature Tðr; tÞ and drift velocity K S ¼ κ ballistic =LS and M ¼ μballistic =LS , where LS is the
ijkl
uðr; tÞ; these are the viscous heat equations at the core of length that determines surface effects, i.e., the sample’s size
this work and that are further discussed in the main text. for a single crystal or the grain size for a polycrystalline
As a final remark it is worth mentioning that in the sample. Similarly, the parameter Fij U is defined as
ij ij
kinetic regime, characterized by strong crystal-momentum FU ¼ DU;boundary ðLS Þ · LS . In the tables, we report the
dissipation, the inequality (D20) may not be valid. Never- populations’ ðκij ij
P Þ and coherences’ ðκ C Þ contributions to
theless, in such a regime umklapp processes are the
most frequent and crystal-momentum dissipation prevails the total bulk conductivity (κij ij ij
bulk ¼ κ P þ κ C [38]) and the
nonzero irreducible components of the bulk thermal vis-
over crystal-momentum redistribution ðmax3i; j¼1 ½Dij U ≫ cosity tensor, limiting graphite to in-plane components.
maxi∈½1; 2; 3; β>3 ½Diβ
U Þ, thus yielding the following (stronger) Additionally, Fig. 9 shows the off-plane transport coef-
inequality ficients [Fig. 9(a)] and the off-plane non-negligible values
011019-22
TABLE I. Parameters entering the viscous heat equations for graphite. We report here only the in-plane components of the tensors needed to perform the calculation of Fig. 3:
ij ij ij ij ij ij ij ij ij i j j ij
κ ij
P ¼ κ P δ , κ C ¼ κ C δ , K S ¼ K S δ , DU;bulk ¼ DU;bulk δ , F U ¼ F U δ , A ¼ A ∀ i, W 0i ¼ W i0 ¼ Wδ , where the indexes i, j represent the in-plane directions x, y only
(i, j ¼ 1, 2 and i ≠ j). Numbers in square brackets denote powers of 10.
iiii (Pa s/m) M ijij (Pa s/m) M ijji (Pa s/m) D −1
T (K) κ P (W/m K) κ C (W/m K) K S ðW=m2 KÞ μiiii
bulk (Pa s) μijij
bulk (Pa s) μijji
bulk (Pa s) M U;bulk ðns Þ FU (m/s) A ðpg=μm3 Þ C ðpg=μm ns2 KÞ W ðμm=nsÞ
50 4.059 37[3] 1.686 03½−4 2.570 75[8] 9.641 73½−4 5.026 73½−4 2.285 03½−4 4.391 76[2] 1.466 38[2] 1.462 69[2] 2.356 02½−1 4.268 77[3] 1.534 33½−4 1.009 00½−4 2.727 61
60 4.716 60[3] 2.579 01½−4 3.951 97[8] 9.719 61½−4 5.286 85½−4 2.191 76½−4 7.155 78[2] 2.392 44[2] 2.381 66[2] 3.438 27½−1 4.598 01[3] 2.294 98½−4 1.411 01½−4 3.006 47
70 5.302 27[3] 3.867 37½−4 5.590 73[8] 9.915 50½−4 5.608 89½−4 2.127 11½−4 1.072 64[3] 3.593 08[2] 3.566 60[2] 4.908 88½−1 4.886 42[3] 3.202 56½−4 1.845 24½−4 3.252 49
80 5.735 08[3] 5.994 94½−4 7.457 26[8] 1.021 83½−3 5.952 33½−4 2.105 80½−4 1.510 11[3] 5.071 90[2] 5.014 40[2] 6.990 24½−1 5.136 08[3] 4.251 02½−4 2.309 97½−4 3.459 24
85 5.883 16[3] 7.643 89½−4 8.467 09[8] 1.041 40½−3 6.126 52½−4 2.116 23½−4 1.758 24[3] 5.915 02[2] 5.833 34[2] 8.334 61½−1 5.246 94[3] 4.826 50½−4 2.553 80½−4 3.546 81
90 5.985 09[3] 9.898 03½−4 9.523 60[8] 1.064 50½−3 6.300 69½−4 2.144 43½−4 2.025 32[3] 6.826 40[2] 6.712 86[2] 9.922 75½−1 5.348 94[3] 5.435 32½−4 2.805 34½−4 3.623 71
100 6.059 68[3] 1.705 92½−3 1.176 48[9] 1.123 68½−3 6.645 94½−4 2.267 60½−4 2.613 92[3] 8.850 20[2] 8.643 31[2] 1.395 51 5.528 12[3] 6.750 30½−4 3.331 60½−4 3.746 25
125 5.688 08[3] 6.194 70½−3 1.799 32[9] 1.369 01½−3 7.471 69½−4 3.082 84½−4 4.370 38[3] 1.502 75[3] 1.433 36[3] 3.039 75 5.856 14[3] 1.056 85½−3 4.777 79½−4 3.894 27
150 4.950 01[3] 1.685 25½−2 2.488 30[9] 1.743 38½−3 8.224 66½−4 4.581 98½−4 6.462 59[3] 2.266 30[3] 2.097 01[3] 5.829 76 6.064 77[3] 1.505 04½−3 6.383 59½−4 3.886 37
175 4.190 48[3] 3.545 34½−2 3.216 91[9] 2.163 05½−3 8.89217½−4 6.352 06½−4 8.810 58[3] 3.156 09[3] 2.825 09[3] 9.952 30 6.203 14[3] 2.006 76½−3 8.098 22½−4 3.801 67
200 3.537 15[3] 6.220 28½−2 3.962 23[9] 2.544 85½−3 9.466 84½−4 7.980 06½−4 1.135 05[4] 4.151 36[3] 3.596 04[3] 1.547 40[1] 6.300 63[3] 2.549 90½−3 9.869 74½−4 3.689 20
225 3.009 49[3] 9.609 03½−2 4.706 45[9] 2.851 77½−3 9.947 91½−4 9.280 42½−4 1.403 36[4] 5.232 16[3] 4.395 58[3] 2.236 22[1] 6.373 37[3] 3.124 32½−3 1.165 81½−3 3.572 59
250 2.591 10[3] 1.353 78½−1 5.436 87[9] 3.082 41½−3 1.034 13½−3 1.024 27½−3 1.682 37[4] 6.380 73[3] 5.214 48[3] 3.051 05[1] 6.430 12[3] 3.722 11½−3 1.343 66½−3 3.461 22
275 2.259 23[3] 1.780 35½−1 6.144 89[9] 3.250 18½−3 1.065 72½−3 1.092 85½−3 1.969 38[4] 7.582 09[3] 6.046 80[3] 3.976 58[1] 6.475 77[3] 4.337 32½−3 1.518 81½−3 3.357 92
300 1.993 78[3] 2.220 75½−1 6.824 87[9] 3.370 78½−3 1.090 79½−3 1.141 03½−3 2.262 38[4] 8.824 02[3] 6.888 60[3] 4.995 51[1] 6.513 24[3] 4.965 54½−3 1.690 10½−3 3.263 00
350 1.603 18[3] 3.078 55½−1 8.087 84[9] 3.520 11½−3 1.125 96½−3 1.198 69½−3 2.860 57[4] 1.139 24[4] 8.590 58[3] 7.245 49[1] 6.570 48[3] 6.248 65½−3 2.017 75½−3 3.096 12
400 1.335 09[3] 3.837 50½−1 9.212 71[9] 3.599 39½−3 1.147 51½−3 1.227 90½−3 3.468 48[4] 1.402 97[4] 1.030 64[4] 9.681 55[1] 6.611 14[3] 7.553 50½−3 2.321 35½−3 2.955 75
500 9.987 02[2] 4.959 75½−1 1.105 94[10] 3.667 70½−3 1.169 05½−3 1.251 59½−3 4.694 90[4] 1.938 52[4] 1.375 02[4] 1.479 64[2] 6.662 15[3] 1.018 96½−2 2.845 81½−3 2.738 66
600 7.994 31[2] 5.634 51½−1 1.244 44[10] 3.690 35½−3 1.177 75½−3 1.258 63½−3 5.920 22[4] 2.474 75[4] 1.718 50[4] 1.996 54[2] 6.690 43[3] 1.282 83½−2 3.260 54½−3 2.585 52
800 5.754 23[2] 6.249 87½−1 1.425 02[10] 3.700 56½−3 1.183 56½−3 1.260 80½−3 8.343 95[4] 3.533 75[4] 2.398 75[4] 3.006 24[2] 6.717 47[3] 1.805 81½−2 3.827 84½−3 2.397 57
1000 4.520 04[2] 6.470 48½−1 1.529 10[10] 3.701 06½−3 1.185 35½−3 1.260 10½−3 1.072 99[5] 4.572 99[4] 3.070 07[4] 3.979 14[2] 6.728 94[3] 2.321 38½−2 4.167 66½−3 2.295 14
011019-23
2000 2.215 03[2] 6.746 60½−1 1.695 81[10] 3.700 84½−3 1.188 29½−3 1.258 38½−3 2.236 10[5] 9.612 71[4] 6.355 92[4] 8.573 30[2] 6.742 28[3] 4.837 15½−2 4.730 19½−3 2.139 73
3000 1.473 97[2] 6.929 56½−1 1.730 97[10] 3.702 78½−3 1.189 73½−3 1.258 60½−3 3.381 43[5] 1.456 00[5] 9.599 38[4] 1.301 89[3] 6.744 40[3] 7.314 98½−2 4.851 55½−3 2.108 27
4000 1.105 42[2] 7.117 33½−1 1.743 64[10] 3.704 41½−3 1.190 62½−3 1.258 96½−3 4.521 56[5] 1.948 03[5] 1.283 04[5] 1.742 34[3] 6.745 11[3] 9.781 54½−2 4.895 50½−3 2.097 05
GENERALIZATION OF FOURIER’S LAW INTO VISCOUS HEAT …
PHYS. REV. X 10, 011019 (2020)
ij ij ij ij ij
TABLE II. Parameters entering the viscous heat equations for diamond. As discussed above, due to the symmetries of diamond’s crystal, κij
P ¼ κP δ , κC ¼ κC δ , K S ¼ K S δ ,
ij ij ij ij i j j ij
DU;bulk ¼ DU;bulk δ , FU ¼ FU δ , A ¼ A ∀ i, W 0i ¼ W i0 ¼ Wδ , where i; j ¼ 1; …; 3 and i ≠ j. Numbers in square brackets denote powers of 10.
T [K] κP (W/m K) κC (W/m K) K S ðW=m2 KÞ μiiii

bulk (Pa s) μijij
bulk (Pa s) M iiii (Pa s/m) M ijij (Pa s/m) DU;bulk ðns−1 Þ FU (m/s) A ðpg=μm3 Þ C ðpg=μm ns2 KÞ W ðμm=nsÞ
50 3.431 11[4] 2.075 28½−5 2.739 09[7] 2.955 94½−3 2.391 02½−3 5.066 87 1.686 63 8.378 22½−2 1.208 87[4] 6.981 85½−7 6.533 93½−6 6.989 08
60 3.282 94[4] 4.824 83½−5 4.877 21[7] 3.422 58½−3 2.650 63½−3 1.109 24[1] 3.613 32 2.364 70½−1 1.176 03[4] 1.557 71½−6 1.188 25½−5 6.791 96
70 3.053 15[4] 9.235 59½−5 7.958 28[7] 3.772 82½−3 2.800 07½−3 2.167 28[1] 7.014 04 5.511 76½−1 1.130 49[4] 3.158 01½−6 1.995 10½−5 6.513 75
80 2.779 12[4] 1.572 29½−4 1.221 26[8] 3.974 91½−3 2.857 72½−3 3.909 00[1] 1.271 10[1] 1.041 79 1.077 56[4] 5.986 89½−6 3.167 87½−5 6.177 05
90 2.483 08[4] 2.526 25½−4 1.787 72[8] 4.034 66½−3 2.836 59½−3 6.622 41[1] 2.182 27[1] 1.663 05 1.024 76[4] 1.072 15½−5 4.809 97½−5 5.822 19
100 2.187 75[4] 3.967 22½−4 2.516 59[8] 3.984 59½−3 2.754 99½−3 1.063 64[2] 3.572 40[1] 2.355 06 9.775 00[3] 1.819 05½−5 7.020 07½−5 5.484 21
125 1.544 75[4] 1.176 00½−3 5.139 87[8] 3.604 62½−3 2.414 25½−3 2.853 97[2] 1.015 22[2] 4.211 25 8.920 29[3] 5.475 61½−5 1.550 97½−4 4.802 53
150 1.081 62[4] 3.126 15½−3 8.933 41[8] 3.142 94½−3 2.051 93½−3 6.099 64[2] 2.294 08[2] 6.316 94 8.416 44[3] 1.269 87½−4 2.865 10½−4 4.329 60
175 7.744 43[3] 7.089 19½−3 1.375 11[9] 2.763 84½−3 1.770 50½−3 1.101 84[3] 4.346 26[2] 8.870 28 8.095 18[3] 2.434 89½−4 4.638 17½−4 3.978 34
200 5.785 23[3] 1.380 27½−2 1.933 67[9] 2.498 42½−3 1.582 04½−3 1.761 24[3] 7.226 24[2] 1.191 99[1] 7.863 36[3] 4.077 76½−4 6.823 43½−4 3.692 65
225 4.538 06[3] 2.366 43½−2 2.541 39[9] 2.325 32½−3 1.465 02½−3 2.574 37[3] 1.090 85[3] 1.537 86[1] 7.678 79[3] 6.193 78½−4 9.352 00½−4 3.447 45
250 3.723 28[3] 3.662 15½−2 3.173 42[9] 2.215 60½−3 1.395 70½−3 3.520 71[3] 1.531 78[3] 1.910 87[1] 7.523 04[3] 8.752 13½−4 1.214 27½−3 3.232 70
275 3.169 07[3] 5.218 40½−2 3.809 54[9] 2.147 10½−3 1.356 66½−3 4.578 03[3] 2.035 52[3] 2.297 96[1] 7.387 78[3] 1.170 73½−3 1.511 04½−3 3.043 95
300 2.774 99[3] 6.953 43½−2 4.434 49[9] 2.105 28½−3 1.336 61½−3 5.725 18[3] 2.591 70[3] 2.688 86[1] 7.268 85[3] 1.500 73½−3 1.817 40½−3 2.878 43
350 2.257 27[3] 1.058 19½−1 5.611 42[9] 2.068 29½−3 1.327 79½−3 8.216 63[3] 3.822 99[3] 3.456 70[1] 7.070 83[3] 2.243 39½−3 2.431 60½−3 2.607 38
SIMONCELLI, MARZARI, and CEPELLOTTI
400 1.931 01[3] 1.394 34½−1 6.657 93[9] 2.064 10½−3 1.338 76½−3 1.087 79[4] 5.160 74[3] 4.185 82[1] 6.915 54[3] 3.065 47½−3 3.015 41½−3 2.400 95
500 1.532 18[3] 1.928 15½−1 8.336 67[9] 2.092 61½−3 1.377 70½−3 1.642 73[4] 7.989 02[3] 5.513 36[1] 6.696 37[3] 4.839 23½−3 4.018 07½−3 2.120 89
600 1.287 60[3] 2.337 80½−1 9.546 94[9] 2.133 44½−3 1.417 60½−3 2.204 29[4] 1.087 52[4] 6.699 46[1] 6.556 39[3] 6.679 50½−3 4.786 18½−3 1.949 40
800 9.891 66[2] 2.975 35½−1 1.105 17[10] 2.206 90½−3 1.481 08½−3 3.309 93[4] 1.657 61[4] 8.815 36[1] 6.399 25[3] 1.035 30½−2 5.788 11½−3 1.763 84
1000 8.078 34[2] 3.455 53½−1 1.188 01[10] 2.261 42½−3 1.525 41½−3 4.384 62[4] 2.211 56[4] 1.075 17[2] 6.319 58[3] 1.393 72½−2 6.359 96½−3 1.672 54
2000 4.260 34[2] 4.745 96½−1 1.314 90[10] 2.389 76½−3 1.625 63½−3 9.507 13[4] 4.842 27[4] 1.967 31[2] 6.205 73[3] 3.092 57½−2 7.261 49½−3 1.545 25
3000 2.901 12[2] 5.432 41½−1 1.340 78[10] 2.437 88½−3 1.662 17½−3 1.448 24[5] 7.389 84[4] 2.830 04[2] 6.183 62[3] 4.732 19½−2 7.448 92½−3 1.520 96
011019-24
4000 2.200 60[2] 5.952 81½−1 1.350 03[10] 2.463 06½−3 1.681 05½−3 1.941 51[5] 9.913 19[4] 3.686 65[2] 6.175 80[3] 6.354 07½−2 7.516 21½−3 1.512 41
PHYS. REV. X 10, 011019 (2020)
ij ij ij ij ij
TABLE III. Parameters entering the viscous heat equations for silicon. As discussed above, due to the symmetries of silicon’s crystal, κij
P ¼ κP δ , κC ¼ κC δ , K S ¼ K S δ ,
ij ij ij ij i j j ij
DU;bulk ¼ DU;bulk δ , FU ¼ FU δ , A ¼ A ∀ i, W 0i ¼ W i0 ¼ Wδ , where i; j ¼ 1; …; 3 and i ≠ j. Numbers in square brackets denote powers of 10.
T (K) κP (W/m K) κ C (W/m K) K S ðW=m2 KÞ μiiii

bulk (Pa s) μijij
bulk (Pa s) M iiii (Pa s/m) Mijij (Pa s/m) DU;bulk ðns−1 Þ FU (m/s) A ðpg=μm3 Þ C ðpg=μm ns2 KÞ W ðμm=nsÞ
50 2.237 95[3] 4.593 29½−2 2.202 01[8] 1.683 79½−3 8.582 40½−4 3.333 15[2] 2.762 88[2] 5.115 70 2.703 64[3] 3.276 65½−4 2.114 41½−4 9.027 35½−1
60 1.813 98[3] 7.095 41½−2 3.151 43[8] 1.588 68½−3 8.966 18½−4 5.524 69[2] 4.764 59[2] 6.623 66 2.715 05[3] 5.544 60½−4 3.024 89½−4 8.852 84½−1
70 1.472 66[3] 9.571 85½−2 4.081 43[8] 1.495 77½−3 9.080 48½−4 8.041 23[2] 7.075 99[2] 8.783 07 2.726 06[3] 8.141 13½−4 3.913 95½−4 8.738 74½−1
80 1.194 53[3] 1.183 29½−1 4.973 27[8] 1.400 19½−3 8.931 42½−4 1.078 64[3] 9.579 83[2] 1.165 82[1] 2.734 91[3] 1.094 95½−3 4.779 16½−4 8.610 85½−1
90 9.740 26[2] 1.372 74½−1 5.821 83[8] 1.308 33½−3 8.614 66½−4 1.369 70[3] 1.220 15[3] 1.523 08[1] 2.741 22[3] 1.389 89½−3 5.628 56½−4 8.447 38½−1
100 8.037 19[2] 1.517 20½−1 6.625 19[8] 1.225 79½−3 8.225 64½−4 1.672 98[3] 1.489 47[3] 1.942 47[1] 2.745 09[3] 1.694 64½−3 6.466 90½−4 8.250 91½−1
125 5.327 96[2] 1.692 76½−1 8.426 48[8] 1.068 29½−3 7.277 20½−4 2.465 64[3] 2.177 60[3] 3.191 46[1] 2.746 49[3] 2.483 48½−3 8.498 29½−4 7.693 40½−1
150 3.875 35[2] 1.705 07½−1 9.926 62[8] 9.668 38½−4 6.554 21½−4 3.284 05[3] 2.871 36[3] 4.600 65[1] 2.740 93[3] 3.293 33½−3 1.036 33½−3 7.161 83½−1
175 3.021 97[2] 1.668 98½−1 1.114 55[9] 9.006 19½−4 6.046 37½−4 4.110 81[3] 3.562 15[3] 6.062 65[1] 2.732 83[3] 4.111 17½−3 1.199 18½−3 6.719 55½−1
200 2.475 34[2] 1.633 44½−1 1.212 36[9] 8.558 50½−4 5.690 80½−4 4.937 07[3] 4.246 98[3] 7.523 49[1] 2.724 47[3] 4.929 77½−3 1.336 75½−3 6.369 72½−1
225 2.099 66[2] 1.609 00½−1 1.290 62[9] 8.244 99½−4 5.437 28½−4 5.758 79[3] 4.925 14[3] 8.960 94[1] 2.716 81[3] 5.745 36½−3 1.450 94½−3 6.096 81½−1
250 1.826 75[2] 1.594 48½−1 1.353 43[9] 8.018 53½−4 5.252 34½−4 6.574 43[3] 5.596 87[3] 1.036 82[2] 2.710 13[3] 6.556 21½−3 1.545 10½−3 5.883 50½−1
275 1.619 70[2] 1.586 83½−1 1.404 15[9] 7.850 55½−4 5.114 34½−4 7.383 70[3] 6.262 80[3] 1.174 54[2] 2.704 43[3] 7.361 73½−3 1.622 69½−3 5.715 33½−1
300 1.457 10[2] 1.583 55½−1 1.445 45[9] 7.723 10½−4 5.009 22½−4 8.186 90[3] 6.923 62[3] 1.309 50[2] 2.699 61[3] 8.161 96½−3 1.686 84½−3 5.581 28½−1
350 1.217 55[2] 1.584 09½−1 1.507 44[9] 7.546 99½−4 4.863 38½−4 9.777 22[3] 8.232 61[3] 1.572 51[2] 2.692 11[3] 9.747 89½−3 1.784 75½−3 5.384 97½−1
400 1.048 80[2] 1.589 14½−1 1.550 67[9] 7.435 11½−4 4.770 32½−4 1.134 99[4] 9.528 38[3] 1.828 36[2] 2.686 71[3] 1.131 74½−2 1.854 14½−3 5.251 64½−1
500 8.251 13[1] 1.604 39½−1 1.604 86[9] 7.308 95½−4 4.664 72½−4 1.445 77[4] 1.209 28[4] 2.325 71[2] 2.679 77[3] 1.442 04½−2 1.942 32½−3 5.088 69½−1
600 6.823 16[1] 1.623 05½−1 1.635 94[9] 7.245 32½−4 4.610 92½−4 1.753 21[4] 1.463 39[4] 2.811 13[2] 2.675 71[3] 1.749 06½−2 1.993 48½−3 4.997 32½−1
800 5.088 41[1] 1.666 61½−1 1.668 06[9] 7.189 68½−4 4.562 98½−4 2.362 51[4] 1.967 79[4] 3.763 31[2] 2.671 45[3] 2.357 55½−2 2.046 82½−3 4.904 46½−1
1000 4.065 52[1] 1.715 02½−1 1.683 37[9] 7.169 39½−4 4.544 79½−4 2.967 84[4] 2.469 51[4] 4.702 85[2] 2.669 41[3] 2.962 02½−2 2.072 39½−3 4.860 79½−1
1200 3.388 05[1] 1.765 61½−1 1.691 80[9] 7.161 46½−4 4.537 16½−4 3.571 09[4] 2.969 82[4] 5.636 06[2] 2.668 28[3] 3.564 37½−2 2.086 51½−3 4.836 88½−1
1400 2.905 38[1] 1.817 10½−1 1.696 92[9] 7.158 58½−4 4.533 94½−4 4.173 11[4] 3.469 33[4] 6.565 67[2] 2.667 59[3] 4.165 48½−2 2.095 12½−3 4.822 40½−1
011019-25
GENERALIZATION OF FOURIER’S LAW INTO VISCOUS HEAT …
PHYS. REV. X 10, 011019 (2020)
(a) (b) (c)
FIG. 8. Trends as a function of the temperature of various quantities reported in Tables I–III. (a) Values of the bulk momentum
dissipation inverse timescale DU;bulk ðTÞ, (b) specific heat CðTÞ and specific momentum AðTÞ, (c) values of the velocity WðTÞ.
of Ai , Dij j
U;bulk , and W 0i [Fig. 9(b)] for graphite; these
APPENDIX F: SECOND SOUND
parameters are not relevant for the geometry studied in In this Appendix, we show that drifting second sound
Sec. V and are reported for completeness. [17,20,40,43], i.e., thermal transport in terms of a temper-
ature damped wave, is described by the viscous heat
equations. We first show the emergence of second sound
(a) following a bottom-up approach, i.e., analyzing the con-
ditions under which the two viscous heat equations
decouple with the first equation (10) reducing to a damped
wave equation for the temperature field. Afterward, we
analyze second sound within the viscous heat equations
following a top-down approach, i.e., requiring the temper-
ature field to have the mathematical form of a damped wave
and analyzing if and under which conditions this can
emerge from the viscous heat equations.
(b) 1. Second sound from the viscous heat

equations (bottom-up approach)
For simplicity, we consider an isotropic system whose
tensors appearing in the viscous heat equations (10) and
(11) are proportional to the identity; the general derivation
can be obtained straightforwardly generalizing the pro-
cedure reported here. Without loss of generality, we
consider x̂ as the direction of second-sound propagation.
In an isotropic system, the drifting heat flux QD ðr; tÞ is
collinear with the drift velocity and the heat flux due to
local temperature changes Qδ ðr; tÞ is collinear with the
temperature gradient. Thus, it follows that the only nonzero
component of the drift velocity must be along the second-
sound propagation direction ux ¼ u (for simplicity we omit
FIG. 9. (a) Off-plane thermal viscosity and off-plane SMA all tensor indexes in the rest of this section, since the only
thermal conductivity (inset) of graphite for a sample having a the component having all the indexes equal to x is needed
grain size of 10 μm (solid line) or for the bulk case (dashed line). for this discussion). With these conditions, the viscous heat
The other components of the viscosity tensor involving off-plane equations (10) and (11) become
directions are negligible (at least one order of magnitude smaller
than the largest component reported here). (b) Specific momen- pffiffiffiffiffiffiffiffiffiffi ∂uðx; tÞ
tum Az , component W z0z of the velocity tensor, and component ∂Tðx; tÞ ∂ 2 Tðx; tÞ
C þ W T̄AC −κ ¼ 0; ðF1Þ
Dzz
U; bulk of the momentum dissipation tensor of graphite. ∂t ∂x ∂x2
011019-26
rffiffiffiffiffiffiffi
∂uðx; tÞ CA ∂Tðx; tÞ ∂ 2 uðx; tÞ ∂ 2 uðx; tÞ DU f ∂uðx; tÞ
A þ W −μ 2
þ
∂t T̄ ∂x ∂x2 ∂t f − ð1 − fÞτss DU ∂t
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
¼ −ADU uðx; tÞ: ðF2Þ c1
2
W ∂ 2 uðx; tÞ
− ¼ 0: ðF7Þ
In order to observe second sound, temperature changes f − ð1 − fÞτss DU ∂x2
|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
need to propagate following a damped wave equation. To c2
this aim, we require that drift velocity and temperature are
related as Therefore, if
pffiffiffiffiffiffiffiffiffiffi ∂uðx; tÞ τss DU

∂ 2 Tðx; tÞ ∂Tðx; tÞ f> ; ðF8Þ
W T̄AC ¼ Cτss ð1 − fÞ 2
− Cf ; 1 þ τss DU
∂x ∂t ∂t
ðF3Þ then both constants c1 and c2 are positive and the evolution
of the drift velocity is that of a damped-wave equation like
where τss is the second-sound relaxation time and 0 < the one for temperature.
jfj < 1 is a constant, both to be determined. To better The coefficients τss and f are determined solving
understand this requirement, we insert Eq. (F3) in Eq. (F1), Eqs. (F4) and (F7) and imposing the second-sound con-
finding the desired temperature damped-wave equation: dition Eq. (F3). The solutions are of the form
Z
1
Tðx; tÞ ¼ CT ðkÞe−t=2τss ei½kx−ω̄ðkÞt dk ðF9Þ
∂ 2 Tðx; tÞ 1 ∂Tðx; tÞ κ ∂ 2 Tðx; tÞ 2π
þ − ¼ 0: ðF4Þ
∂t2 τss ∂t Cτss ð1 − fÞ ∂x2
and
Z
Next, we show that condition (F3) implies that also the 1
drift-velocity field follows a damped-wave equation. To uðx; tÞ ¼ Cu ðkÞe−t=2τss ei½kx−ω̄ðkÞt dk; ðF10Þ
2π
this aim, we take the derivative with respect to x of Eq. (F3),
finding with
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffi ∂ 2 uðx; tÞ ∂2

∂ ∂Tðx; tÞ 1
W T̄AC ¼ Cτss ð1 − fÞ 2 − Cf ; ω̄ðkÞ ¼ v2ss k2 − 2 ; ðF11Þ
∂2x ∂t ∂t ∂x 4τss
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Ô
∂ ω̄ðkÞ kvss
ðF5Þ vg ðkÞ ¼ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi; ðF12Þ
∂k k − ð2τss vss Þ−2
2
where Ô is a differential operator. Next, by applying the f ¼ DU τss ; ðF13Þ

operator Ô to both sides of Eq. (F2) and using condition
(F5), we obtain CW 2
τss ¼ ; ðF14Þ
κD2U þ DU CW 2
∂ 3 uðx; tÞ f 1 ∂ 2 uðx; tÞ W2 ∂ 2 uðx; tÞ
− þ κDU þ CW 2
∂t3 1 − f τss ∂t2 τss ð1 − fÞ ∂x2 vss ¼ : ðF15Þ
4 3
CW
μ ∂ uðx; tÞ f 1 μ ∂ uðx; tÞ
− þ
A ∂t2 ∂x2 1 − f τss A ∂t∂x2 The condition (F13) is derived from the requirement
2 c1 ¼ 1=τss , and is consistent with the damped-wave
∂ uðx; tÞ f 1 ∂uðx; tÞ
¼ −DU − : ðF6Þ requirement (F8); condition (F14) is derived from the
∂t2 1 − f τss ∂t requirement that c2 ¼ κ=Cτss ð1 − fÞ and the second-sound
velocity (F15) has been obtained substituting Eq. (F13)
If we consider only the lowest-order derivatives in Eq. (F6), and Eq. (F14) into Eq. (F4). We note that for DU → 0
we obtain a simplified equation that holds in the close- (that is, negligible crystal-momentum dissipation),
to-equilibrium regime where variations in space and time vg ðkÞ → vss → W; i.e., the second-sound propagation
are small. In particular, neglecting higher-than-second- velocity approaches the drifting second-sound velocity
order derivatives gives defined by Hardy [20]) and τss → ðDU Þ−1 . Finally, the
011019-27
qffiffiffiffiffi
second-sound condition (F3) imposes the following rela- ik CA W
tion between the coefficients: T̄
u0 ¼ −δT ; ðF21Þ
μk2 þ ADU − iAω̃ðkÞ
rffiffiffiffiffiffiffi 2
−i τss C 1
Cu ðkÞ ¼ ð1 − τss DU Þ þ iω̄ðkÞ and substitute this expression into Eq. (F19), finding
k W T̄A 2τss

1 CAW 2 k2
þ DU þ iω̄ðkÞ CT ðkÞ: ðF16Þ −iCω̃ðkÞ þ 2
þ κk2 ¼ 0; ðF22Þ
2τss μk þ ADU − iAω̃ðkÞ
CT ðkÞ must be set according to initial conditions and the which gives
form of Cu ðkÞ follows from Eq. (F16). We note from
Eq. (F16) that, when second sound occurs, temperature and ½−iCω̃ðkÞ þ κk2 ½ADU þ μk2 − iAω̃ðkÞ þ CAW 2 k2 ¼ 0:
drift velocity are both damped waves that are characterized ðF23Þ
by a phase shift.
This is a quadratic equation that determines the dispersion
2. Second sound from the viscous heat equations relations for ω̃ðkÞ, given by
(top-down approach)
2 μ κ 2
In the previous section, we obtained second-sound ω̃ ðkÞ þ iω̃ðkÞ þ k þ DU þ
A C
properties by finding the conditions under which the
viscous equation for temperature [Eq. (10)] becomes a 2 κDU κμk2 2
− W þ þ k ¼ 0: ðF24Þ
damped-wave equation. However, we can also obtain a C CA
second-sound equation taking inspiration from the
approach outlined in Ref. [40], i.e., by looking for the This equation can be solved to obtain the complex
conditions upon which the microscopic degrees of freedom frequency ω̃ðkÞ and thus the oscillation frequency and
of the transport equation evolve as a damped wave. In decay time of second sound as a function of the wave vector
particular, we want to find the conditions such that the k. Solving for this quadratic equation, we obtain
solution of Eqs. (10) and (11) are
i μ 2 κ 2
ω̃ðkÞ ¼ − k þ DU þ k
Tðx; tÞ ¼ T̄ þ ðδTÞei½kx−ω̂ðkÞt e−t=2^τss ; ðF17Þ 2 A C
2
1 μ 2 κ
− k þ DU þ k2
uðx; tÞ ¼ u0 ei½kx−ω̂ðkÞt e−t=2^τss ; ðF18Þ 2 A C
2

2 2 DU κk κμk4 1=2
þ4 W k þ þ : ðF25Þ
where δT and u0 are, in general, complex numbers to allow C AC
for a phase difference between the two waves. We note in
particular that this guess for solution requires that both In order to compare this result with the expression for
temperature and drift velocity oscillate at the same fre- second sound derived in the previous section, it is worth
quency and decay with the same rate, which is consistent recalling that the semiclassical description of thermal
with the conditions (F9) and (F10) obtained in the previous transport used throughout this work holds for long-wave-
section. length perturbations. Therefore, we simplify the previous
Using this guess for the solution, the derivation of the expression retaining terms to smallest order in k, finding
dispersion relation and the decay time easily follows. To sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi
this aim, we substitute Eqs. (F17) and (F18) in the viscous iD DU D κ μD
heat equations (10) and (11) and introducing the complex ω̃ðkÞ ≈ − U − þ k2 W 2 þ U
− U
:
2 2 2C 2A
^
frequency ω̃ðkÞ ¼ ωðkÞ − i=2^τss to simplify the calcula-
tion, we find ðF26Þ
pffiffiffiffiffiffiffiffiffiffi Choosing for convenience the solution with posi-
−iCω̃ðkÞδT þ W T̄ACiku0 þ κk2 δT ¼ 0; ðF19Þ
tive real part, we can readily see that ω̂ðkÞ ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffi k2 ðW 2 þ DU κ=2C − μDU =2AÞ − ðDU =2Þ2 and τ̂ss ¼
−iAω̃ðkÞu0 þ
CA
WikδT þ μk2 u0 ¼ −ADU u0 . ðF20Þ 1=DU , that is, the second sound’s decay time is set by
T̄ the crystal momentum dissipation rate. In the limit of
decreasingly smaller crystal-momentum dissipation
Next, we rewrite Eq. (F20) as (DU → 0), we find ω̂ðkÞ ≈ Wk, that is, second sound
011019-28
disperses linearly with k, and has a velocity v̂g ðkÞ ¼ thermal transport, which is in turn dependent on both
∂ ω̂ðkÞ=∂k ≈ W. These limiting approximations (^vg ðkÞ ≈ conductivity and viscosity. Moreover, we note that the
W and τ^ ss ≈ 1=DU ) are consistent with the results found in mathematical form of the solution has the same qualitative
the previous section. behavior of the problem studied by Sussmann and Thellung
[13], which serves as a verification of the present model.
APPENDIX G: ANALYTICAL 1D EXAMPLE At variance with their work, however, the prefactors
introduced here allow us to go beyond the Debye approxi-
The viscous heat equations (10) and (11) can be solved mation for the phonon dispersion.
analytically in a handful of toy models. Here, for simplicity,
we neglect dissipation of momentum by umklapp processes APPENDIX H: ESTIMATE OF THE
(Dij
U ¼ 0) and consider steady-state heat diffusion along CHARACTERISTIC DRIFT VELOCITY
the transversal direction of a thin film, so that the problem
becomes effectively a 1D problem, with x labeling the In this Appendix, we estimate the characteristic value of
orthogonal direction position. In addition, since we are the drift velocity (u0 ) in the high- (uH ) and low-temperature
considering a 1D geometry, we label A ¼ Ax , W ¼ W x0x , (uL ) regimes. These characteristic values are determined
μ ¼ μxxxx and assume that temperature and velocity fields substituting in the viscous heat equations (10) and (11) the
depend only on the position x. Under these considerations characteristic values of the temperature (and related deriv-
the viscous heat equations (10) and (11) reduce to atives) and solving them approximatively for the velocity.
With this aim, we start estimating the characteristic temper-
pffiffiffiffiffiffiffiffiffiffi ∂uðxÞ ∂ 2 TðxÞ ature gradient in the setup of Fig. 3 when a temperature
T̄ACW −κ ¼ 0; ðG1Þ difference T̄ δT is imposed on the two opposite sides (at
∂x ∂x2
x ¼ 0 and x ¼ 15 μm). For a sample geometrically similar
rffiffiffiffiffiffiffi to that in Fig. 3 with total length lTOT (e.g., lTOT ¼ 15 μm
CA ∂TðxÞ ∂ 2 uðxÞ
W −μ ¼ 0: ðG2Þ for the sample in Fig. 3), we choose its shortest lengthscale,
T̄ ∂x ∂x2 L ¼ lTOT = 5, as a length unit.
To solve the problem, we specify the following no-slip In this setup we clearly distinguish two regions. The left
boundary conditions on a 1D geometry having length 2l: region has length lL ¼ 1 L and width wL ¼ 1 L, while the
right region has length lR ¼ 4 L and width wR ¼ 3 L.
uðx ¼ lÞ ¼ 0 ðG3Þ Energy conservation requires that the current in the left
side must be equal to that in the right side. Since the
and temperature gradient is mainly along the x direction, the x
component of the heat flux on the left side QxL must be
Tðx ¼ lÞ ¼ T̄ δT; ðG4Þ 3 times the heat flux on the right side, QxL ¼ 3QxR . Using
Fourier’s law, and supposing that the thermal conductivity
that is, we assume boundaries at thermal equilibrium. tensor is isotropic and constant throughout the sample, it
We look for solutions of the form follows that Qx ¼ −κxx ∇x T and the temperature gradients
in the two regions are related as ∇x T L ¼ 3∇x T R . Requiring
uðxÞ ¼ d coshðblÞ þ a coshðbxÞ; ðG5Þ the total temperature drop to be equal to the temperature
difference imposed by the boundary conditions, we can
TðxÞ ¼ T̄ þ c sinhðbxÞ. ðG6Þ write
After some algebra, one finds the solution −2δT ¼ ΔT xL þ ΔT xR ¼ lL ∇x T L þ lR ∇x T R ¼ 7L∇x T R ;
rffiffiffiffiffiffi ðH1Þ
κ coshðbxÞ
uðxÞ ¼ δT − cothðblÞ ; ðG7Þ
μT̄ sinhðblÞ where ΔT xL and ΔT xR are the temperature drops over the left
and right regions, respectively.
sinhðbxÞ
TðxÞ ¼ T̄ þ δT ; ðG8Þ Focusing from now on the larger region on the right, it
sinhðblÞ follows that the temperature drop taking place over this
sffiffiffiffiffiffiffiffiffiffiffiffiffi region is approximately given by ΔT R ¼ − 87 δT.
ACW 2 To determine a characteristic value of drift velocity u0,
b¼ ¼ λ−1 : ðG9Þ
μκ we substitute ∇T xR ≃ΔT R =lR in Eq. (11) and for simplicity
consider isotropic symmetry, valid, e.g., for silicon, dia-
This analytical solution shares several qualitative similar- mond, and for graphite in the in-plane directions. u0 can be
ities with the numerical example discussed in the main text, determined focusing on the steady-state limit of Eq. (11).
and more clearly highlights how the factor λ ¼ 1=b We simplify its estimate considering separately the limits of
represents a length scale at which surface scattering affects low and high temperatures.
011019-29
In the high-temperature limit, the term related to

momentum dissipation (∝ DijU ) is much larger than the
viscous term (∝ μ ) (see Fig. 8); therefore, Eq. (11) can
ijkl
be approximated as
rffiffiffiffiffiffiffiffiffi
CAx x x x
W x0 ∇ T ≃ −Ax Dxx x
Uu : ðH2Þ
T̄
Using the estimated temperature gradient ∇T xR ≃

ΔT R =lR , the high-temperature characteristic value of drift
velocity uH is found to be
rffiffiffiffiffiffiffiffi
C W xx0 2 δT
uH ¼ : ðH3Þ
T̄Ax Dxx U 7 L
FIG. 10. Estimates of the characteristic value of the drift
At low temperatures, viscous effects dominate over velocity u0 (dashed lines) as combination of the asymptotic
momentum dissipation (see Figs. 1 and 8), so that Eq. (11) trends given by uH [Eq. (H3), dotted orange line] and uL
[Eq. (H6), dotted red line] for graphite samples having total
is approximated as
length lTOT ¼ lL þ lR ¼ 15 μm (main plot), 1 and 100 μm
rffiffiffiffiffiffiffiffiffi (inset). hux i (solid lines) is the average value of the drift velocity
CAx x ∂Tðx; yÞ ∂ 2 ux ðx; yÞ 2 x
xyxy ∂ u ðx; yÞ computed from the numerical solution of the viscous heat
W x0 ≃ μxxxx þ μ ;
T̄ ∂x ∂x2 ∂y2 equations in the region lL < x < ðlL þ lR Þ.
ðH4Þ
where we considered only the two largest components of where we used wR ¼ 3 L and lR ¼ 4 L. We thus recover the
the viscosity tensor. To estimate the average value of these expression for uL given in Sec. VI in the main text. The
second derivatives, we note that, as shown in Fig. 3, uðx; yÞ characteristic value of the drift velocity u0 is obtained
has a bell-like profile in the sample interior, which vanishes interpolating the high- and low-temperature limiting values
at the boundaries. We thus proceed with a few assumptions as u−1 −1 −1
0 ¼ uH þ uL . The estimates of uH , uL , and u0 for
that allow us to make an estimate of uL . First, we suppose graphite are reported in Fig. 10. We also compare this rough
that the average values of the second derivatives of the drift estimate with the average value of ux ðx; yÞ computed from
velocity in Eq. (H4) are of the same order of magnitude, the numerical solution of the viscous heat equations on the
that is, h∂ 2 ux =∂x2 i ∼ h∂ 2 ux =∂y2 i, which can be checked region x > 15 lTOT of the geometry discussed in the main text
numerically. It follows that Eq. (H4) can be simplified as and denoted with hux i. Despite the qualitative arguments
used to derive u0 , the estimate is able to capture qualitative
rffiffiffiffiffiffiffiffiffi trends and approximately reproduce average results from
∂ 2 ux CAx W xx0 ∇x T R
≃ ¼ −a: ðH5Þ the numerical solution of the viscous heat equations.
∂y2 T̄ μxxxx þ μxyxy
Next, we note that the variation of u is stronger along the y APPENDIX I: HYDRODYNAMIC BEHAVIOR
coordinate. To mimic the Poiseuille-like shape, we assume IN MACROSCOPIC DIAMOND SAMPLES
the velocity profile to be constant along the x direction, In this Appendix, we show how extending the sizes in
and parabolic along the y direction with vanishing velocity Fig. 5(e) up to 10 mm yields another peak for the FDN
at the boundaries (y ¼ 0 μm and y ¼ wR ), so that for dimensions around 1 mm and temperatures around
uðx; yÞ ≃ ½−a · yðy − wR Þ; 0; 0. With these approxima- 50–60 K. This result has been confirmed by full solutions
tions, we can estimate the average value of the parabolic of the viscous heat equations similar to these reported in
velocity profile, i.e., the characteristic value of the drift Fig. 5(b). The increase of the FDN at low temperature and
velocity at low temperatures uL , as large size reported in Fig. 11 is in qualitative agreement
Z w with the predictions for the second-sound window per-
1 R aw2 formed using the reduced isotropic crystal model or the
uL ¼ ux ðyÞdy ¼ R
wR 0 6 Callaway model [45]. However, in contrast with this work,
rffiffiffiffiffiffiffiffiffi Ref. [45] reports that an isotopic concentration much lower
3 CA W x0 δT L
x x
¼ ; ðH6Þ than the natural isotope abundance is necessary for the
7 T̄ μxxxx þ μxyxy observation of second sound in diamond.
011019-30
grid and a Gaussian smearing of 8 cm−1 . The use of

an odd, Γ-centered q-points mesh is crucial to
correctly account for the parity symmetries of the scattering
operator.
[1] R. Peierls, Zur Kinetischen Theorie der Wärmeleitung in

Kristallen, Ann. Phys. (Berlin) 395, 1055 (1929).
[2] R. E. Peierls, Quantum Theory of Solids (Oxford
University Press, New York, 1955).
[3] J. M. Ziman, Electrons and Phonons: The Theory of
Transport Phenomena in Solids ( Oxford University Press,
FIG. 11. Hydrodynamic behavior in crystalline diamond at New York, 1960).
natural isotopic abundance: increasing the dimension of the [4] L. Mezhov-Deglin, Measurement of the Thermal Conduc-
diamond sample can yield an increase of the hydrodynamic tivity of Crystalline He4 , Sov. Phys. JETP 22, 66 (1965).
effects at low temperatures. [5] C. C. Ackerman, B. Bertman, H. A. Fairbank, and R. A.
Guyer, Second Sound in Solid Helium, Phys. Rev. Lett. 16,
789 (1966).
The analyses here and in Fig. 5 are limited to a minimum [6] H. E. Jackson, C. T. Walker, and T. F. McNelly, Second
reference temperature of T̄ ¼ 50 K because, for lower Sound in NaF, Phys. Rev. Lett. 25, 26 (1970).
temperatures, anharmonicity becomes decreasingly smaller [7] D. W. Pohl and V. Irniger, Observation of Second Sound in
NaF by Means of Light Scattering, Phys. Rev. Lett. 36, 480
while size effects become increasingly larger, resulting in
(1976).
convergence issues for the numerical calculations (which [8] V. Narayanamurti and R. C. Dynes, Observation of Second
would also require a refined, computationally expensive, Sound in Bismuth, Phys. Rev. Lett. 28, 1461 (1972).
treatment of surface effects [35]). Most importantly, [9] B. Danil’Chenko, V. Poroshin, and O. Sarbei, An Obser-
the temperature and size ranges reported in Fig. 5 highlight vation of Second Sound in Sapphire, JETP Lett. 30, 196
the most accessible experimental conditions under which (1979).
hydrodynamic behavior can appear in diamond, namely [10] B. Hehlen, A.-L. Pérou, E. Courtens, and R. Vacher,
Observation of a Doublet in the Quasielastic Central
micrometer-sized diamond samples (thus much less expen-
Peak of Quantum-Paraelectric SrTiO3 , Phys. Rev. Lett.
sive than millimeter-sized samples) and noncryogenic 75, 2416 (1995).
temperatures. [11] Y. K. Koh and D. G. Cahill, Frequency Dependence of the
Thermal Conductivity of Semiconductor Alloys, Phys. Rev.
B 76, 075207 (2007).
APPENDIX J: COMPUTATIONAL DETAILS [12] L. Bergamasco, M. Alberghini, M. Fasano, A. Cardellini,
First-principles calculations are performed with the E. Chiavazzo, and P. Asinari, Mesoscopic Moment Equa-
QUANTUM ESPRESSO distribution [139,140]. For all the tions for Heat Conduction: Characteristic Features
and Slow–Fast Mode Decomposition, Entropy 20, 126
materials analyzed, the local-density approximation (LDA)
(2018).
for the exchange-correlation energy functional is employed [13] J. A. Sussmann and A. Thellung, Thermal Conductivity of
due to its capability to accurately describe the structural and Perfect Dielectric Crystals in the Absence of Umklapp
vibrational [141] properties of graphite [70], diamond [25], Processes, Proc. Phys. Soc. 81, 1122 (1963).
and silicon [27], and its compatibility with the D3Q code [14] R. Gurzhi, Thermal Conductivity of Dielectrics and Fer-
[32,33] for first-principles calculations of anharmonic rodielectrics at Low Temperatures, J. Exp. Theor. Phys.
(third-order) interatomic force constants. Details on the 46, 490 (1964).
[15] R. N. Gurzhi, Hydrodynamic Effects in Solids at Low
computation of the second- and third-order interatomic
Temperature, Sov. Phys. Usp. 11, 255 (1968).
force constants are reported in Ref. [27] for silicon, in [16] R. A. Guyer and J. A. Krumhansl, Solution of the Linear-
Ref. [25] for diamond, and in Ref. [70] for graphite. The ized Phonon Boltzmann Equation, Phys. Rev. 148, 766
LBTE’s scattering matrix Ωνν0 is computed as in Ref. [25] (1966).
and accounts for third-order anharmonicity [32] and iso- [17] R. A. Guyer and J. A. Krumhansl, Thermal Conductivity,
topic disorder [142,143] at natural abundance. Thermal Second Sound, and Phonon Hydrodynamic Phenomena in
conductivity and viscosity calculations for silicon and Nonmetallic Crystals, Phys. Rev. 148, 778 (1966).
[18] A. Griffin, On the Detection of Second Sound in Crystals
diamond are performed using 27 × 27 × 27 q-point grids
by Light Scattering, Phys. Lett. 17, 208 (1965).
and a Gaussian smearing of 2 and 8 cm−1 , respectively. [19] C. P. Enz, One-Particle Densities, Thermal Propagation,
Thermal conductivity and viscosity calculations for and Second Sound in Dielectric Crystals, Ann. Phys.
graphite are performed using a 49 × 49 × 3 q-point (N.Y.) 46, 114 (1968).
011019-31
[20] R. J. Hardy, Phonon Boltzmann Equation and Second [37] A. J. McGaughey, A. Jain, H.-Y. Kim, and B. Fu,
Sound in Solids, Phys. Rev. B 2, 1193 (1970). Phonon Properties and Thermal Conductivity from First
[21] R. J. Hardy and D. L. Albers, Hydrodynamic Approxima- Principles, Lattice Dynamics, and the Boltzmann Trans-
tion to the Phonon Boltzmann Equation, Phys. Rev. B 10, port Equation, J. Appl. Phys. 125, 011101 (2019).
3546 (1974). [38] M. Simoncelli, N. Marzari, and F. Mauri, Unified Theory of
[22] M. Omini and A. Sparavigna, An Iterative Approach to the Thermal Transport in Crystals and Glasses, Nat. Phys. 15,
Phonon Boltzmann Equation in the Theory of Thermal 809 (2019).
Conductivity, Physica (Amsterdam) 212B, 101 (1995). [39] S. Lee, D. Broido, K. Esfarjani, and G. Chen, Hydro-
[23] D. A. Broido, M. Malorny, G. Birner, N. Mingo, and D. A. dynamic Phonon Transport in Suspended Graphene, Nat.
Stewart, Intrinsic Lattice Thermal Conductivity of Semi- Commun. 6, 6290 (2015).
conductors from First Principles, Appl. Phys. Lett. 91, [40] A. Cepellotti and N. Marzari, Transport Waves as Crystal
231922 (2007). Excitations, Phys. Rev. Mater. 1, 045406 (2017).
[24] J. Carrete, B. Vermeersch, A. Katre, A. van Roekeghem, T. [41] S. Lee and L. Lindsay, Hydrodynamic Phonon Drift and
Wang, G. K. Madsen, and N. Mingo, almaBTE: A Solver of Second Sound in a (20,20) Single-Wall Carbon Nanotube,
the Space–Time Dependent Boltzmann Transport Equa- Phys. Rev. B 95, 184304 (2017).
tion for Phonons in Structured Materials, Comput. Phys. [42] Z. Ding, J. Zhou, B. Song, V. Chiloyan, M. Li, T.-H. Liu,
Commun. 220, 351 (2017). and G. Chen, Phonon Hydrodynamic Heat Conduction
[25] G. Fugallo, M. Lazzeri, L. Paulatto, and F. Mauri, Ab Initio and Knudsen Minimum in Graphite, Nano Lett. 18, 638
Variational Approach for Evaluating Lattice Thermal (2018).
Conductivity, Phys. Rev. B 88, 045430 (2013). [43] S. Huberman, R. A. Duncan, K. Chen, B. Song, V.
[26] L. Chaput, Direct Solution to the Linearized Phonon Chiloyan, Z. Ding, A. A. Maznev, G. Chen, and K. A.
Boltzmann Equation, Phys. Rev. Lett. 110, 265506 (2013). Nelson, Observation of Second Sound in Graphite at
[27] A. Cepellotti and N. Marzari, Thermal Transport in Temperatures above 100 K, Science 364, 375 (2019).
Crystals as a Kinetic Theory of Relaxons, Phys. Rev. X [44] V. Martelli, J. L. Jiménez, M. Continentino, E. Baggio-
6, 041013 (2016). Saitovitch, and K. Behnia, Thermal Transport and Phonon
[28] K. Esfarjani, G. Chen, and H. T. Stokes, Heat Transport in Hydrodynamics in Strontium Titanate, Phys. Rev. Lett.
Silicon from First-Principles Calculations, Phys. Rev. B 120, 125901 (2018).
84, 085204 (2011). [45] V. Khodusov and A. Naumovets, Second Sound Waves in
[29] M. N. Luckyanova, J. Garg, K. Esfarjani, A. Jandl, M. T. Diamond, Diam. Relat. Mater. 21, 92 (2012).
Bulsara, A. J. Schmidt, A. J. Minnich, S. Chen, M. S. [46] Y. Machida, A. Subedi, K. Akiba, A. Miyake, M.
Dresselhaus, Z. Ren, E. A. Fitzgerald, and G. Chen, Tokunaga, Y. Akahama, K. Izawa, and K. Behnia, Ob-
Coherent Phonon Heat Conduction in Superlattices, Sci- servation of Poiseuille Flow of Phonons in Black Phos-
ence 338, 936 (2012). phorus, Sci. Adv. 4, eaat3374 (2018).
[30] D. G. Cahill, P. V. Braun, G. Chen, D. R. Clarke, S. Fan, [47] C. Hua and A. J. Minnich, Analytical Green’s Function of
K. E. Goodson, P. Keblinski, W. P. King, G. D. Mahan, A. the Multidimensional Frequency-Dependent Phonon
Majumdar, H. J. Maris, S. R. Phillpot, E. Pop, and L. Shi, Boltzmann Equation, Phys. Rev. B 90, 214306 (2014).
Nanoscale Thermal Transport. II. 2003–2012, Appl. Phys. [48] B. Vermeersch, J. Carrete, N. Mingo, and A. Shakouri,
Rev. 1, 011305 (2014). Superdiffusive Heat Conduction in Semiconductor Alloys.
[31] A. Cepellotti, G. Fugallo, L. Paulatto, M. Lazzeri, I. Theoretical Foundations, Phys. Rev. B 91, 085202
F. Mauri, and N. Marzari, Phonon Hydrodynamics in (2015).
Two-Dimensional Materials, Nat. Commun. 6, 6400 [49] A. A. Maznev, J. A. Johnson, and K. A. Nelson, Onset of
(2015). Nondiffusive Phonon Transport in Transient Thermal
[32] L. Paulatto, F. Mauri, and M. Lazzeri, Anharmonic Proper- Grating Decay, Phys. Rev. B 84, 195206 (2011).
ties from a Generalized Third-Order Ab Initio Approach: [50] F. Yang and C. Dames, Heating-Frequency-Dependent
Theory and Applications to Graphite and Graphene, Phys. Thermal Conductivity: An Analytical Solution from Dif-
Rev. B 87, 214303 (2013). fusive to Ballistic Regime and Its Relevance to Phonon
[33] L. Paulatto, I. Errea, M. Calandra, and F. Mauri, First- Scattering Measurements, Phys. Rev. B 91, 165311
Principles Calculations of Phonon Frequencies, Lifetimes, (2015).
and Spectral Functions from Weak to Strong Anharmo- [51] J.-P. M. Péraud and N. G. Hadjiconstantinou, Extending
nicity: The Example of Palladium Hydrides, Phys. Rev. B the Range of Validity of Fourier’s Law into the Kinetic
91, 054304 (2015). Transport Regime via Asymptotic Solution of the Phonon
[34] L. Lindsay, First Principles Peierls-Boltzmann Phonon Boltzmann Transport Equation, Phys. Rev. B 93, 045424
Thermal Transport: A Topical Review, Nanoscale Micro. (2016).
Thermophys. Eng. 20, 67 (2016). [52] G. Chen, Ballistic-Diffusive Heat-Conduction Equations,
[35] A. Cepellotti and N. Marzari, Boltzmann Transport in Phys. Rev. Lett. 86, 2297 (2001).
Nanostructures as a Friction Effect, Nano Lett. 17, 4675 [53] C. V. D. R. Anderson and K. K. Tamma, Novel Heat
(2017). Conduction Model for Bridging Different Space and Time
[36] M. Omini and A. Sparavigna, Beyond the Isotropic-Model Scales, Phys. Rev. Lett. 96, 184301 (2006).
Approximation in the Theory of Thermal Conductivity, [54] J. Ordonez-Miranda, R. Yang, and J. Alvarado-Gil, A
Phys. Rev. B 53, 9064 (1996). Constitutive Equation for Nano-to-Macro-Scale Heat
011019-32
Conduction Based on the Boltzmann Transport Equation, Thermal Conductivity of Semiconductors, Phys. Rev. B
J. Appl. Phys. 109, 084319 (2011). 95, 165407 (2017).
[55] A. T. Ramu and Y. Ma, An Enhanced Fourier Law [73] J. Heron, T. Fournier, N. Mingo, and O. Bourgeois,
Derivable from the Boltzmann Transport Equation and Mesoscopic Size Effects on the Thermal Conductance of
a Sample Application in Determining the Mean-Free Path Silicon Nanowire, Nano Lett. 9, 1861 (2009).
of Nondiffusive Phonon Modes, J. Appl. Phys. 116, [74] M. E. Siemens, Q. Li, R. Yang, K. A. Nelson, E. H.
093501 (2014). Anderson, M. M. Murnane, and H. C. Kapteyn, Quasi-
[56] C. Hua, L. Lindsay, X. Chen, and A. Minnich, Exper- Ballistic Thermal Transport from Nanoscale Interfaces
imental Demonstration of a Generalized Fourier’s Law for Observed Using Ultrafast Coherent Soft X-Ray Beams,
Non-Diffusive Thermal Transport, Phys. Rev. B 100, Nat. Mater. 9, 26 (2010).
085203 (2019). [75] J. A. Johnson, A. A. Maznev, J. Cuffe, J. K. Eliason, A. J.
[57] B.-Y. Cao and Z.-Y. Guo, Equation of Motion of a Phonon Minnich, T. Kehoe, C. M. S. Torres, G. Chen, and K. A.
Gas and Non-Fourier Heat Conduction, J. Appl. Phys. Nelson, Direct Measurement of Room-Temperature Non-
102, 053503 (2007). diffusive Thermal Transport Over Micron Distances in a
[58] F. Alvarez, D. Jou, and A. Sellitto, Phonon Hydrodynamics Silicon Membrane, Phys. Rev. Lett. 110, 025901 (2013).
and Phonon-Boundary Scattering in Nanosystems, J. [76] K. T. Regner, D. P. Sellan, Z. Su, C. H. Amon, A. J.
Appl. Phys. 105, 014317 (2009). McGaughey, and J. A. Malen, Broadband Phonon Mean
[59] Y. Guo and M. Wang, Phonon Hydrodynamics and Its Free Path Contributions to Thermal Conductivity Mea-
Applications in Nanoscale Heat Transport, Phys. Rep. sured Using Frequency Domain Thermoreflectance, Nat.
595, 1 (2015). Commun. 4, 1640 (2013).
[60] Y. Guo and M. Wang, Phonon Hydrodynamics for Nano- [77] R. B. Wilson and D. G. Cahill, Anisotropic Failure
scale Heat Transport at Ordinary Temperatures, Phys. of Fourier Theory in Time-Domain Thermoreflectance
Rev. B 97, 035421 (2018). Experiments, Nat. Commun. 5, 5075 (2014).
[61] A. Ziabari, P. Torres, B. Vermeersch, Y. Xuan, X. Cartoixà, [78] J.-P. M. Péraud and N. G. Hadjiconstantinou, Efficient
A. Torelló, J.-H. Bahk, Y. R. Koh, M. Parsa, P. D. Ye, F. X. Simulation of Multidimensional Phonon Transport Using
Alvarez, and A. Shakouri, Full-Field Thermal Imaging of Energy-Based Variance-Reduced Monte Carlo Formula-
Quasiballistic Crosstalk Reduction in Nanoscale Devices, tions, Phys. Rev. B 84, 205331 (2011).
Nat. Commun. 9, 255 (2018). [79] T. Sun and P. B. Allen, Lattice Thermal Conductivity:
[62] P. Torres, A. Ziabari, A. Torelló, J. Bafaluy, J. Camacho, X. Computations and Theory of the High-Temperature
Cartoixà, A. Shakouri, and F. X. Alvarez, Emergence of Breakdown of the Phonon-Gas Model, Phys. Rev. B 82,
Hydrodynamic Heat Transport in Semiconductors at the 224305 (2010).
Nanoscale, Phys. Rev. Mater. 2, 076001 (2018). [80] R. Berman, Thermal Conductivity of Isotopically Enriched
[63] X. Li and S. Lee, Role of Hydrodynamic Viscosity on Diamonds, Phys. Rev. B 45, 5726 (1992).
Phonon Transport in Suspended Graphene, Phys. Rev. B [81] D. G. Onn, A. Witek, Y. Z. Qiu, T. R. Anthony, and W. F.
97, 094309 (2018). Banholzer, Some Aspects of the Thermal Conductivity of
[64] V ¼ VN c , i.e., the unit cell volume V times the number of Isotopically Enriched Diamond Single Crystals, Phys.
unit cells that constitute the crystal N c . N c is also the Rev. Lett. 68, 2806 (1992).
number of wave vectors q used to sample the Brillouin [82] J. R. Olson, R. O. Pohl, J. W. Vandersande, A. Zoltan, T. R.
zone. Anthony, and W. F. Banholzer, Thermal Conductivity of
[65] P. G. Klemens and F. E. Simon, The Thermal Conductivity Diamond between 170 and 1200 K and the Isotope Effect,
of Dielectric Solids at Low Temperatures (Theoretical), Phys. Rev. B 47, 14850 (1993).
Proc. R. Soc. A 208, 108 (1951). [83] L. Wei, P. K. Kuo, R. L. Thomas, T. R. Anthony, and W. F.
[66] M. J. Rice, Theory of Viscosity in Nearly Ferromagnetic Banholzer, Thermal Conductivity of Isotopically Modified
Fermi Liquids, Phys. Rev. 162, 189 (1967). Single Crystal Diamond, Phys. Rev. Lett. 70, 3764 (1993).
[67] P. B. Allen, Improved Callaway Model for Lattice Thermal [84] A. Ward, D. A. Broido, D. A. Stewart, and G. Deinzer,
Conductivity, Phys. Rev. B 88, 144302 (2013). Ab Initio Theory of the Lattice Thermal Conductivity in
[68] P. B. Allen and V. Perebeinos, Temperature in a Peierls- Diamond, Phys. Rev. B 80, 125203 (2009).
Boltzmann Treatment of Nonlocal Phonon Heat Trans- [85] D. A. Broido, L. Lindsay, and A. Ward, Thermal Conduc-
port, Phys. Rev. B 98, 085427 (2018). tivity of Diamond under Extreme Pressure: A First-
[69] This follows from the property Ω̃νν0 ¼ Ω̃−ν;−ν0 . Principles Study, Phys. Rev. B 86, 115203 (2012).
[70] G. Fugallo, A. Cepellotti, L. Paulatto, M. Lazzeri, N. [86] A. V. Inyushkin, A. N. Taldenkov, J. W. Ager III, E. E.
Marzari, and F. Mauri, Thermal Conductivity of Graphene Haller, H. Riemann, N. V. Abrosimov, H.-J. Pohl, and P.
and Graphite: Collective Excitations and Mean Free Becker, Ultrahigh Thermal Conductivity of Isotopically
Paths, Nano Lett. 14, 6109 (2014). Enriched Silicon, J. Appl. Phys. 123, 095112 (2018).
[71] M. Gandolfi, G. Benetti, C. Glorieux, C. Giannetti, and F. [87] E. M. Lifshitz and L. P. Pitaevskii, Physical Kinetics.
Banfi, Accessing Temperature Waves: A Dispersion Rela- Course of Theoretical Physics (Pergamon Press, Oxford,
tion Perspective, Int. J. Heat Mass Transfer 143, 118553 1981), Vol. 10.
(2019). [88] N. Bonini, J. Garg, and N. Marzari, Acoustic Phonon
[72] P. Torres, A. Torelló, J. Bafaluy, J. Camacho, X. Cartoixà, Lifetimes and Thermal Transport in Free-Standing and
and F. X. Alvarez, First Principles Kinetic-Collective Strained Graphene, Nano Lett. 12, 2673 (2012).
011019-33
[89] E. Mariani and F. von Oppen, Flexural Phonons in Free- [112] Z. Fan, L. F. C. Pereira, P. Hirvonen, M. M. Ervasti, K. R.
Standing Graphene, Phys. Rev. Lett. 100, 076801 (2008). Elder, D. Donadio, T. Ala-Nissila, and A. Harju, Thermal
[90] Z.-Y. Ong and E. Pop, Effect of Substrate Modes on Conductivity Decomposition in Two-Dimensional Materi-
Thermal Transport in Supported Graphene, Phys. Rev. B als: Application to Graphene, Phys. Rev. B 95, 144309
84, 075471 (2011). (2017).
[91] J. Weinbub and D. K. Ferry, Recent Advances in Wigner [113] P. B. Allen, Analysis of Nonlocal Phonon Thermal Con-
Function Approaches, Appl. Phys. Rev. 5, 041104 (2018). ductivity Simulations Showing the Ballistic to Diffusive
[92] J. D. Logan, Applied Partial Differential Equations Crossover, Phys. Rev. B 97, 134307 (2018).
(Springer, New York, 2014). [114] Z. Li, S. Xiong, C. Sievers, Y. Hu, Z. Fan, N. Wei, H. Bao,
[93] P. C. Kwok, Dispersion and Damping of Second Sound in S. Chen, D. Donadio, and T. Ala-Nissila, Influence of
Non-Isotropic Solids, Phys. Phys. Fiz. 3, 221 (1967). Boundaries and Thermostatting on Nonequilibrium
[94] C. Ackerman and R. Guyer, Temperature Pulses in Molecular Dynamics Simulations of Heat Conduction in
Dielectric Solids, Ann. Phys. (N.Y.) 50, 128 (1968). Solids, arXiv:1905.11024.
[95] H. J. Maris, Second Sound and Heat Flow in Anisotropic [115] W. Li, J. Carrete, N. A. Katcho, and N. Mingo, ShengBTE:
Solids, Phys. Rev. B 24, 1205 (1981). A Solver of the Boltzmann Transport Equation for
[96] M. Chester, Second Sound in Solids, Phys. Rev. 131, 2013 Phonons, Comput. Phys. Commun. 185, 1747 (2014).
(1963). [116] E. Buckingham, On Physically Similar Systems; Illustra-
[97] C. Cattaneo, Sulla Conduzione del Calore, Atti Sem. Mat. tions of the Use of Dimensional Equations, Phys. Rev. 4,
Fis. Univ. Modena 3, 83 (1948). 345 (1914).
[98] D. D. Joseph and L. Preziosi, Heat Waves, Rev. Mod. [117] F. M. White, Fluid Mechanics, 5th ed. (McGraw-Hill Book
Phys. 61, 41 (1989). Company, Boston, 2003).
[99] D. Y. Tzou, The Generalized Lagging Response in Small- [118] J. Maassen and M. Lundstrom, Steady-State Heat Trans-
Scale and High-Rate Heating, Int. J. Heat Mass Transfer port: Ballistic-to-Diffusive with Fourier’s Law, J. Appl.
38, 3231 (1995). Phys. 117, 035104 (2015).
[100] A. K. Majee and Z. Aksamija, Dynamical Thermal [119] M. Terraneo, M. Peyrard, and G. Casati, Controlling the
Conductivity of Suspended Graphene Ribbons in the Energy Flow in Nonlinear Lattices: A Model for a Thermal
Hydrodynamic Regime, Phys. Rev. B 98, 024303 (2018). Rectifier, Phys. Rev. Lett. 88, 094302 (2002).
[101] S. Lepri, R. Livi, and A. Politi, Thermal Conduction in [120] B. Li, L. Wang, and G. Casati, Thermal Diode: Rectifi-
Classical Low-Dimensional Lattices, Phys. Rep. 377, 1 cation of Heat Flux, Phys. Rev. Lett. 93, 184301 (2004).
(2003). [121] C. W. Chang, D. Okawa, A. Majumdar, and A. Zettl, Solid-
[102] Wolfram Research, Inc., Mathematica, Version 12.0, State Thermal Rectifier, Science 314, 1121 (2006).
Champaign, IL (2019). [122] J. Hu, X. Ruan, and Y. P. Chen, Thermal Conductivity and
[103] H. J. Maris, Second Sound and Heat Flow in Anisotropic Thermal Rectification in Graphene Nanoribbons: A
Solids, Phys. Rev. B 24, 1205 (1981). Molecular Dynamics Study, Nano Lett. 9, 2730 (2009).
[104] S. Simons, The Relation between the Heat Current and the [123] Y. Wang, A. Vallabhaneni, J. Hu, B. Qiu, Y. P. Chen, and
Phonon Drift Velocity, Am. J. Phys. 51, 457 (1983). X. Ruan, Phonon Lateral Confinement Enables Thermal
[105] P. K. Schelling, S. R. Phillpot, and P. Keblinski, Compari- Rectification in Asymmetric Single-Material Nanostruc-
son of Atomic-Level Simulation Methods for Computing tures, Nano Lett. 14, 592 (2014).
Thermal Conductivity, Phys. Rev. B 65, 144306 (2002). [124] R. Anufriev and M. Nomura, Thermal Conductance Boost
[106] Y.-Y. Zhang, Q.-X. Pei, H.-Y. Liu, and N. Wei, Thermal in Phononic Crystal Nanostructures, Phys. Rev. B 91,
Conductivity of a h-BCN Monolayer, Phys. Chem. Chem. 245417 (2015).
Phys. 19, 27326 (2017). [125] S. Volz et al., Nanophononics: State of the Art and
[107] M. H. Khadem and A. P. Wemhoff, Comparison of Green– Perspectives, Eur. Phys. J. B 89, 15 (2016).
Kubo and NEMD Heat Flux Formulations for Thermal [126] H. Wang, S. Hu, K. Takahashi, X. Zhang, H. Takamatsu,
Conductivity Prediction Using the Tersoff Potential, and J. Chen, Experimental Study of Thermal Rectification
Comput. Mater. Sci. 69, 428 (2013). in Suspended Monolayer Graphene, Nat. Commun. 8,
[108] M. Puligheddu, F. Gygi, and G. Galli, First-Principles 15843 (2017).
Simulations of Heat Transport, Phys. Rev. Mater. 1, [127] Y. Liu, Y. Xu, S.-C. Zhang, and W. Duan, Model for
060802 (2017). Topological Phononics and Phonon Diode, Phys. Rev. B
[109] S. Iubini, P. Di Cintio, S. Lepri, R. Livi, and L. Casetti, 96, 064106 (2017).
Heat Transport in Oscillator Chains with Long-Range [128] S. Chen, D. Donadio, G. Benenti, and G. Casati, Efficient
Interactions Coupled to Thermal Reservoirs, Phys. Rev. E Thermal Diode with Ballistic Spacer, Phys. Rev. E 97,
97, 032102 (2018). 030101(R) (2018).
[110] Y. He, I. Savić, D. Donadio, and G. Galli, Lattice Thermal [129] I. M. Felix and L. F. C. Pereira, Thermal Conductivity of
Conductivity of Semiconducting Bulk Materials: Atomistic Graphene-hBN Superlattice Ribbons, Sci. Rep. 8, 2737
Simulations, Phys. Chem. Chem. Phys. 14, 16209 (2012). (2018).
[111] X. W. Zhou, S. Aubry, R. E. Jones, A. Greenstein, and P. K. [130] A. Sood, F. Xiong, S. Chen, H. Wang, D. Selli, J. Zhang,
Schelling, Towards More Accurate Molecular Dynamics C. J. McClellan, J. Sun, D. Donadio, Y. Cui et al., An
Calculation of Thermal Conductivity: Case Study of GaN Electrochemical Thermal Transistor, Nat. Commun. 9,
Bulk Crystals, Phys. Rev. B 79, 115201 (2009). 4510 (2018).
011019-34
[131] M. De Luca, C. Fasolato, M. A. Verheijen, Y. Ren, [137] R. J. Hardy, Energy-Flux Operator for a Lattice, Phys.
M. Y. Swinkels, S. Koelling, E. P. Bakkers, R. Rurali, Rev. 132, 168 (1963).
X. Cartoixa, and I. Zardo, Phonon Engineering in Twin- [138] R. J. Hardy, Lowest-Order Contribution to the Lattice
ning Superlattice Nanowires, Nano Lett. 19, 4702 (2019). Thermal Conductivity, J. Math. Phys. (N.Y.) 6, 1749
[132] L. Levitov and G. Falkovich, Electron Viscosity, Current (1965).
Vortices and Negative Nonlocal Resistance in Graphene, [139] P. Giannozzi et al., QUANTUM ESPRESSO: A Modular and
Nat. Phys. 12, 672 (2016). Open-Source Software Project for Quantum Simulations
[133] D. A. Bandurin, I. Torre, R. K. Kumar, M. B. Shalom, A. of Materials, J. Phys. Condens. Matter 21, 395502
Tomadin, A. Principi, G. H. Auton, E. Khestanova, K. S. (2009).
Novoselov, I. V. Grigorieva, L. A. Ponomarenko, A. K. [140] P. Giannozzi et al., Advanced Capabilities for Materials
Geim, and M. Polini, Negative Local Resistance Caused by Modelling with QUANTUM ESPRESSO, J. Phys. Condens.
Viscous Electron Backflow in Graphene, Science 351, Matter 29, 465901 (2017).
1055 (2016). [141] S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi,
[134] P. J. Moll, P. Kushwaha, N. Nandi, B. Schmidt, and A. P. Phonons and Related Crystal Properties from Density-
Mackenzie, Evidence for Hydrodynamic Electron Flow in Functional Perturbation Theory, Rev. Mod. Phys. 73, 515
PdCoO2 , Science 351, 1061 (2016). (2001).
[135] T. Scaffidi, N. Nandi, B. Schmidt, A. P. Mackenzie, and [142] J. Garg, N. Bonini, B. Kozinsky, and N. Marzari, Role of
J. E. Moore, Hydrodynamic Electron Flow and Hall Disorder and Anharmonicity in the Thermal Conductivity
Viscosity, Phys. Rev. Lett. 118, 226601 (2017). of Silicon-Germanium Alloys: A First-Principles Study,
[136] H. Spohn, The Phonon Boltzmann Equation, Properties Phys. Rev. Lett. 106, 045901 (2011).
and Link to Weakly Anharmonic Lattice Dynamics, J. Stat. [143] S.-i. Tamura, Isotope Scattering of Dispersive Phonons in
Phys. 124, 1041 (2006). Ge, Phys. Rev. B 27, 858 (1983).
011019-35

Generalization of Fourier's Law Into Viscous Heat Equations

Uploaded by

Copyright:

Available Formats

Generalization of Fourier's Law Into Viscous Heat Equations

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Generalization of Fourier's Law Into Viscous Heat Equations

Uploaded by

Copyright:

Available Formats

PHYSICAL REVIEW X 10, 011019 (2020)

Generalization of Fourier’s Law into Viscous Heat Equations

I. INTRODUCTION progress took place in this field, propelled by newly

2160-3308=20=10(1)=011019(35) 011019-1 Published by the American Physical Society

(a) temperature yields a decrease, up to a high-temperature

(a) determined by the even part of the relaxon spectrum, while

i ¼ 1, 2, 3 (where 3 is the dimensionality of the system)

∂Tðr; tÞ X 3 pffiffiffiffiffiffiffiffiffiffiffiffi ∂uj ðr; tÞ X 3

We show in Fig. 3 the total viscous heat flux Qδ þ QD and (a)

(a) 102 Graphite (b) Diamond (c) Silicon

(d) 102 (e) (f)

(a) G experiments for second sound in graphite [43]. Then, we

where the specific heat C is ðA12Þ

distribution for the LBTE. j Ai i X 1 ij β

¼ Π̄ij þ Πij ij With the knowledge of the LBTE solution f ij α at hand,

Notably, wi0α is different from zero only for odd eigenvec-

1. Single-mode relaxation-time approximation

The deviation from equilibrium ñδE

We insert this result in the definition of momentum flux,

where we stress that, due to the lack of energy conservation

ηijkl þ ηilkj 1 X 2 i j k l N̄ ν ðN̄ ν þ 1Þ SMA

1. Projection of the LBTE on the first 1X

Equation (D14) is clearly not sufficient to fully describe

vanishes when β indexes an even eigenvector. Since the

T [K] κP (W/m K) κC (W/m K) K S ðW=m2 KÞ μiiii

T (K) κP (W/m K) κ C (W/m K) K S ðW=m2 KÞ μiiii

(a) (b) (c)

(b) 1. Second sound from the viscous heat

pffiffiffiffiffiffiffiffiffiffi ∂uðx; tÞ τss DU

where Ô is a differential operator. Next, by applying the f ¼ DU τss ; ðF13Þ

In the high-temperature limit, the term related to

Using the estimated temperature gradient ∇T xR ≃

grid and a Gaussian smearing of 8 cm−1 . The use of

[1] R. Peierls, Zur Kinetischen Theorie der Wärmeleitung in

You might also like