[IFT-UAM/CSIC-19-116]

Scale Invariant SolidsMatteo Baggioli∗

Instituto de Fisica Teorica UAM/CSIC, c/ Nicolas Cabrera 13-15, Cantoblanco, 28049 Madrid, Spain

Vıctor Cancer Castillo† and Oriol Pujolas‡

Institut de Fısica d’Altes Energies (IFAE),The Barcelona Institute of Science and Technology (BIST)

Campus UAB, 08193 Bellaterra, Barcelona.

Scale invariance (SI) can in principle be realized in the elastic response of solid materials. There aretwo basic options: that SI is a manifest symmetry or that it is spontaneously broken. The manifestcase corresponds physically to the existence of a non-trivial infrared fixed point with phonons amongits degrees of freedom. We use simple bottom-up AdS/CFT constructions to model this case. Wecharacterize the types of possible elastic response and discuss how the sound speeds can be realistic,that is, sufficiently small compared to the speed of light. We also study the spontaneously brokencase using Effective Field Theory (EFT) methods. We present a new one-parameter family ofnontrivial EFTs that includes the previously known ‘conformal solid’ as a particular case as well asothers which display small sound speeds. We also point out that an emergent Lorentz invariance atlow energies could affect by order-one factors the relation between sound speeds and elastic moduli.

I. INTRODUCTION

The mechanical response of matter under small ap-plied stresses is a well-known subject [1, 2]. At suffi-ciently low energies, it can be described in a continuumlimit by the so-called elasticity theory. Just like in hy-drodynamics, the main assumption is that the displace-ments in the solid are described by an effective set of fieldsφi(t, xj) that represent the deformations of the materialfrom its equilibrium position at each point. The effec-tive Lagrangian for φi(t, xj) is then automatically fixedby symmetries. It was shown in [3] (see also [4]) thatthe form of the nonlinearities in the effective Lagrangianis greatly constrained by the fact that the phonon fieldφi(t, xj) can be viewed as the Goldstone boson arisingfrom the spontaneous breaking of translation invariance.More recently, it has been shown how to derive the effec-tive Lagrangian applying the Coset construction to thespontaneous breaking of Poincare symmetry [5, 6].

These developments taught us how to promote elas-ticity theory into a fully nonlinear Effective Field The-ory (EFT). We shall refer to this EFT simply as SolidEFT and give more details on it below. As it happenswith other known EFTs, one expects that this providesfor an efficient way to re-sum certain low-energy observ-ables that are difficult to compute directly from the mi-croscopic theory. It is natural, then, to ask what arethe phenomenological consequences that can be extractedand how the procedure works. Given that the EFT meth-

∗ matteo.baggioli@uam.es; https://members.ift.uam-csic.es/

matteo.baggioli/† vcancer@ifae.es‡ pujolas@ifae.es

ods mainly provide nontrivial information concerning thenonlinear part of the theory, one expects that the SolidEFT provides interesting constraints/information aboutthe phonon interactions (e.g., phonon 2→ 2 scattering),but more generally also regarding the nonlinear elasticresponse. Ref. [7], initiated a study in this direction,showing that nontrivial relations among several nonlin-ear observables can indeed be identified. This motivatesus to continue the analysis to more sophisticated cases.

The purpose of this work is to focus on the special casewhere the solid exhibits scale invariance (SI), in additionto the broken symmetries of a regular solid. Aside frombeing interesting per se, this case seems to be quite closeto real world of materials that exhibit criticality in theform of a quantum critical point. In order to possiblymake contact with these especially interesting materials,it is desirable to understand well how SI is compatiblewith solid EFT or similar techniques.1

It is worth spending some words on what are the pos-sible ways how SI can be realized generically. Concep-tually, the main division arises from whether or not thelow energy dynamics is governed by a nontrivial infraredfixed point (IRFP). To some extent, in the presence ofan IRFP one can say that SI is an unbroken symmetry.By the same logic, in the absence of an IRFP, then, SIcan only be spontaneously broken – a nonlinearly real-ized symmetry. This criterion allows to separate possiblerealizations of SI in solids in two basic groups:

• solids with spontaneously broken SI. In this

1 In this work, SI is meant to be realized in the mechanical sector –by the phonons. It isn’t our goal to identify what kind of physicalsystem accomplishes this, but the idea is very well posed so wejust take it as an assumption.

arX

iv:1

910.

0528

1v2

[he

p-th

] 2

0 M

ar 2

020

2

FIG. 1. Sketch of the two basic options on how scale in-variance (SI) can be realized, in terms of the beta functionβ = µ dλ

dµof a certain coupling λ. The arrows indicate the

flow towards low energies. The left plot represents the spon-taneous breaking case, which we discuss in Section 3. Theright cartoon represents the case with manifest SI, which wediscuss in Section 4.

case one expects a gapped spectrum and that thephonons are isolated degrees of freedom at low en-ergies. In this case EFT methods are applicablein order to describe the lightest excitations (thephonons) in the mechanical sector as the Goldstonebosons of the spontaneously broken spacetime sym-metries. We discuss this case in Section 3.

• solids with manifest SI, where by assumptionthere is a dynamical IRFP. In this case the phononsare not isolated degrees of freedom and one ex-pects that the dispersion relation develops an imag-inary part. Will use bottom-up AdS/CFT meth-ods, which are well suited to construct simple mod-els with these properties, in Section 4.

These options can be better visualized using the renor-malization group (RG) language, that is, in terms of thebeta function β = µ dλdµ of a certain coupling λ. The main

two options are depicted in Fig. 1. The spontaneouslybroken SI case can be viewed as a departure (an RG-flow) from a UVFP induced by the vacuum expectationvalue of some operator. The manifest SI case correspondsto the presence of an infrared or emergent fixed point.

The pictures in Fig. 1 also immediately suggest thatone can construct more options by ‘combining’ the twopossibilities, that is by having both an IR and a UVfixed points. For instance, one can break spontaneouslythe UV SI but then ‘land’ on an IRFP which realizesan emergent SI. This case would combine both sponta-neously broken and manifest realizations.

These possibilities seem to apply both to Lorentz in-variant and non-invarant situations, and one can eas-ily construct examples in bottom-up holographic models.For instance, Lorentz-invariant examples of the SB casecan be found in [8] (see also [9]) and of the emergent +SB case in [10, 11].

An important qualitative distinction in the Lorentz in-variant case is that SB of scale (and conformal) invariance

is accompanied with the appearance of a massless dila-ton. It is well known that this requires fine-tuning of thetheory, however assuming the tuning, the dilaton polemust appear and has indeed been found in both SB [8, 9]and emergent+SB cases [10, 11]. In condensed mattersetups, however, Lorentz boosts are broken and the dila-ton does not appear even if SI is broken spontaneously[12, 13] – in a sense it is replaced by other Goldstonebosons, the phonons. This motivates a deeper study ofthe possible realizations of SI in solid materials from thelow energies effective point of view, with the main focusin whether SI is a spontaneously broken or a manifestsymmetry.

The second main motivation for our work is, perhaps,more down-to-earth: a sine qua non condition for thetheories that aim to describe realistic solids (SI or not)is that the sound speeds (both transverse and longitudi-nal) are tiny compared with the speed of light c – indeed,the fastest sound speeds in known materials are around10−4 in units of c. To the best of our knowledge, theonly know previous example of a SI solid effective the-ory is the so-called conformal solid [13], and it displaysinevitably relativistic longitudinal sound waves. Any ef-fective description of realistic SI solids must overcomethis difficulty. Below, we will show how SI is compati-ble with ‘slow sound’ both for spontaneously broken andmanifest SI.

The mechanical response in critical materials is also thesubject of recent research from a more condensed matterperspective [14–16]. They are also of interest since it isin this class of material that deviations from the KSSviscosity bound [17] have been identified [18–21].

The rest of this work is organized as follows. In Sec-tion 2.1 we review the text-book elasticity theory, andin Section 2.2 we review how it is reformulated in theEFT of solids. In Section 3 we discuss the special caseof solid EFTs that incorporate SI (which corresponds tothe spontaneous breaking case). In Section 4 we discusscase with manifest SI, and we conclude in Section 5.

II. LINEAR ELASTICITY

In this Section we review a few basic notions on theelasticity theory for general (isotropic and homogeneous)solids. The main focus will be to introduce the shear andbulk (or compressibility) moduli, and their relation withthe sound/phonon properties.

A. Elastic response

The elastic response describes the produced stress ina material with respect to an external mechanical de-formation, i.e. the strain [2]. The state of mechanicaldeformation of the solid can be described by the map-ping

Φi = xi + φi(t, x) (1)

3

which gives the position of every given solid element. Thedeviations from equilibrium are directly encoded in φi,which act in all respects like a set of dynamical scalarfields. Their (small) wave excitations are the phonons.And the time-constant configurations of the form φi ∝xi encode small shear or bulk strain deformations. Theuseful way to parametrize them is the tensorial objectknown as strain tensor :

εij =1

2(∂iφj + ∂jφi) (2)

where φi ≡ r′i−ri is the displacement vector, i.e. the de-formation from equilibrium (see Fig. 2). The bulk strainis defined as the trace of the strain tensor,

εii = ~∂ · ~φ (3)

and it can be either positive or negative. It physically cor-responds to an external compression/traction on the sys-tem which changes the volume of the sample. The shearstrain ε is on the contrary the traceless part, which canbe reduced to the off-diagonal component of the straintensor,

ε ≡ 2 εxy (4)

and it encodes the angular deflection of a point from itsoriginal position (see Fig. 2). Obviously a generic me-chanical deformation contains a superposition of a bulkand shear strain.

For homogeneous and isotropic solids, the response tosmall external strain is described at linear level by twoelastic moduli. In the linear response approximation, i.e.for small deformations, the deviation of the stress tensorfrom equilibrium

σij ≡ Tij − p δij

for an isotropic solid (see textbooks [1, 2]), in d spacedimensions, takes the simple form:

σij = K δij εkk + 2G(εij −

1

d− 1δij εkk

)(5)

where G,K are referred to as the linear shear and bulkelastic moduli respectively.

The shear elastic modulus can be obtained in linearresponse theory from the shear component of the stresstensor as:

Txy = G ε+O(ε2) , G = <[GRTxy,Txy

(ω = k = 0)]

(6)with GRTxy,Txy

the retarded Green function of the stress

tensor operator. Let us notice that the shear modulus Gpertains just the static or zero frequency response to ashear strain and it is purely non dissipative.

The same can be done for the bulk modulus K fromthe correlator of the trace part of the Tij , that is, as σii =K ∂ ·φ+O((∂ ·φ)2). This is compatible with the another

y

x

ΦI

2ϵxy

Φ’I

y

x

FIG. 2. The description of the mechanical deformations interms of the displacement vector Φi = xi + φi. Top: Theequilibrium configuration is simply ΦIeq = xI . Bottom: Anexample of a shear deformation and the geometrical inter-pretation of the strain tensor εij . The configuration changesfrom Φieq to Φi such that Φi−Φieq = (εxx dx+ εxy dy, εyy dy +εxy dx).

notion of the bulk modulus, as simply the inverse of thecompressibility, which applies to more general systems,

K = −V dp

dV(7)

where V is the volume of the system and p ≡ Txx themechanical pressure.

Another simple and important parameter in order tocharacterize different kind of materials is the so-calledPoisson’s ratio. It parametrizes how much a materialcompresses (or dilates) in the transverse direction whenunder an applied axial tensile strain,

R = − εtransεaxial

. (8)

It is possible to express this ratio in terms of the elasticmoduli [22], by re-writing the elastic response as εij =1E [(1 +R)σij −Rσkkδij ], with E the Young modulus. It

4

follows that

R =(d− 1)K − 2G

(d− 1)(d− 2)K + 2G. (9)

This allows us to classify models accordingly to their lin-ear elastic properties. By construction, in two spatialdimension the Poisson ratio is bounded −1 < R < 1

d−2 ,and the most auxetic behaviour corresponds to K � G.

Notice that R can be negative, giving a rather exotictype of response where the material actually dilates inthe transverse directions. Materials of this type are calledauxetic and have a number of applications. Let us ad-vance one of the results of Section IV is that we willconstruct planar black hole solutions that are auxetic inexactly the same sense as this.

As we review in Section II B, the elastic moduli deter-mine completely the speed of propagation of transverseand longitudinal phonons in homogeneous and isotropicsolids.

B. Phonons and solid EFTs

Let us now review how the two elastic moduli deter-mine the speed of propagation of transverse and longi-tudinal phonons, cT,L, with no additional microscopicinformation on the solid characteristics required. Therelations are long known but we find illustrative to de-rive them using Effective Field Theory (EFT) methods,by treating the phonons as the Goldstone bosons asso-ciated to the spontaneous symmetry breaking patternwhich takes place in solids. The resulting Solid EFT arediscussed at lowest order in derivatives for the phononfields in [3–6], see also [7]. Let us emphasize that theEFT description below corresponds to the spontaneousbreaking of spacetime symmetries.

We want to work with dynamical degrees of freedomthat are in a sense responsible for the spontaneous break-ing of the translations that take place in solids. In factin the language of the previous subsection, the sponta-neous breaking of the translations can be ascribed to thescalar fields ΦI , taking the vacuum expectation value(vev) Φi = xi in equilibrium. Looking at (1), then oneidentifies the phonons as φi, the perturbations aroundthis vev (or ‘background’).

The EFT can be formalized more sharply by labelingthe set of scalar fields with an ‘internal’ index, so fromnow on we swictch to the notation ΦI(x). In d spacetimedimensions, we need d − 1 scalars, so the internal indexruns over I = 1, . . . , d−1. The theory can be then viewedas having an internal symmetry group given by the two-dimensional Euclidean group, ISO(d − 1), acting on ΦI

like standard translations and rotations in the internalspace. The equilibrium configuration of an homogeneousand isotropic material is identified with the vev

ΦIeq = δIj xj . (10)

which breaks the symmetry group ISO(d−1)×ISO(d−1, 1) down to the diagonal subgroup preserved by (10).

At lowest order in derivatives, the effective Lagrangianis a free function of the scalar field strength

IIJ = gµν∂µΦI∂νΦJ (11)

traced with the internal ISO(d−1) symmetry group met-ric δIJ . We denote by gµν the spacetime metric, whichit is assumed to be Minkowski. Since IIJ is a rank d− 1matrix, there are d − 1 invariants that can be split intoZ = det

(IIJ

), and d−2 traces Xn = tr

{(IIJ)n

}. Then,

the most general effective action at lowest order in deriva-tives can be written as [3–6]

S = −∫ddx√−g V

(Z, {Xn}

). (12)

The function V (Z, {Xn}) is arbitrary (up to stability con-straints) and its form depends on the solid material inquestion.

The phonons fields are identified as the small excita-tions around the equilibrium configuration, φI = ΦI −ΦIeq.

In order to see that the effective action (12) encodesa mechanical response, it is illustrative to show how theelasic moduli are determined by the form of V . One canfind these moduli by introducing a small strain as theconfiguration

ΦI =(δIj + εIj

)xj

with a small matrix εIj and working how the stress tensor

depends at linear order on εIj . The details of the compu-tation are deferred to Appendix A. Let us note, however,that the expressions simplify remarkably once one usesthe variable xn ≡ Xn/Z

n/(d−1) instead of Xn – that iswe view V to be a generic function of Z and xn. For thebulk modulus we obtain

K = 4Z2VZZ + 2ZVZ (13)

and for the shear modulus

G = 2

d−2∑n=1

n2 ∂V

∂xn. (14)

The energy density and pressure for the backgroundconfiguration read

ρ = V and p = 2Z VZ − V ,

and as a result the standard link between the soundspeeds and the moduli follows,

c2T =G

ρ+ pc2, c2L =

K + 2d−2d−1G

ρ+ pc2 . (15)

Note that in the non-relativistic limit where the massdensity ρm dominates so that ρ+ p ' ρmc2 +O(c0), andone recovers the classic expressions [1, 2]

c2T 'Gρm

, c2L 'K + 2 d−2

d−1 Gρm

(16)

5

which do not involve the speed of light.

Equation (15), together with the definition of the bulkmodulus (7), implies a general relation between the lon-gitudinal and transverse sound speeds,

c2L =dp

dρ

∣∣∣�

c2 + 2d− 2

d− 1c2T (17)

where the |�

subscript here is to remind that the deriva-tive is taken while keeping zero shear strain, that is, withno shape deformation. In particular in the fluid limit(G → 0), this expression recovers the usual relation be-tween the sound speed and the equation of state in fluids.

The relation between the sound speeds (15) supersedesthe one that was found for the particular case of a con-formal solid in [13],

c2L =1

d− 1c2 + 2

d− 2

d− 1c2T . (18)

In Sec. III we will see how the formula (15) (valid fora general Solid EFT) simplifies when one imposes ScaleInvariance, and how it reduces to the one found in [13]for the conformal sub-case.

Lastly, we remark that the Poisson ratio (9) dependsonly on G/K and therefore it can be expressed entirelyin function of the ratio of the two speeds, cT /cL. Using(15), one finds

R =1− 2

c2Tc2L

d− 2− (d− 3)c2Tc2L

. (19)

The auxetic limit (R = −1) corresponds transversesound as close as possible to longitudinal sound, specifi-cally c2L = 2 d−2

d−1 c2T (which follows also by setting K = 0

in (15)).

III. SOLIDS WITH SPONTANEOUSLYBROKEN SCALE INVARIANCE

Spontaneously broken Scale Invariance (SI) is easy tobe implemented in the solid EFT framework. We simplyrequire the Lagrangian to be invariant under scale trans-formations, which however are not a symmetry of theground state. Physically, scale transformations are justa rescaling of the coordinates. However, in this theorythere are two objects that play the role of coordinates:the ‘external’ coordinates xµ and the internal space co-ordinates (solid element positions) φI . Therefore, it isconceivable that a rescaling of coordinates acts (perhapsdifferently) on each of them. This leads us to considerthe scale transformation as

xµ → λ−1 xµ

ΦI → λ∆ ΦI (20)

with some ‘weight’ ∆ for the fields φI .To proceed, we construct SI combinations of IIJ . Out

of the invariants Z, Xn, the combinations of the formxn ≡ Xn/Z

n/(d−1) are manifestly SI (for arbitrary val-ues of ∆). The the most general action invariant under(20) must have V which transforms homogeneously. Thisleads to V (Z, {xm}) of the form

Vw(X,Z) = Z1+w

2 F ({xn}) (21)

for some constant w and with an arbitrary functionF ({xn}) of only d− 2 arguments.

The important restriction is that there is a power of,say, Z which factors out, with some exponent. Invarianceunder (20) uniquely fixes w in terms of ∆, by (1+w)(d−1)(1 + ∆) = d, giving

∆ =1− (d− 1)w

(d− 1)(w + 1), or w =

1− (d− 1) ∆

(d− 1)(∆ + 1), (22)

which is shown in Fig. 3 for d = 4.Several comments are in order:

1. The physical meaning of the parameter w in (21)is readily found by computing the energy densityρ and pressure p for the background configuration(10) (that is, of the solid in its state of mechanicalequilibrium). One finds that w is none other thanthe equation of state parameter,

w =p

ρ, (23)

which in turn gives Tµµ = (1 − (d − 1)w) ρ. Notethat SI fixes the equation of state to be constantfor arbitrary values of the energy density/pressure,that is, that the equation of state is linear,

p = p(ρ) = w ρ, (24)

to all orders. As usual, w must comply with theusual The Null Energy Condition (NEC), w > −1,as a necessary condition order to ensure the absenceof ghosts. See below for further constraints fromother consistency conditions.

2. The weight ∆ introduced in (20) plays the samerole as the scaling dimension for ΦI , formally inthe same way as for scalar operators in conformalfield theories CFTs.

3. The previous point immediately raises the ques-tion: can one apply standard logic and results fromCFTs? In fact, the fields ΦI play the role of somescalar operators, and (21) is formally relativistic(it is built out of ΦI and the Minkowski metric)so one might even wonder: does (21) actually de-fine a relativistic CFT? From our perspective, theanswer to both questions is no. Despite appear-ances, the theories (21) are not really relativistic

6

-1 1 2 3Δ

-1.0

-0.5

0.5

1.0

1.5

2.0

ω

FIG. 3. Relation between the equation of state parameter wand the scaling dimension of the scalars ΦI for SI solids in3 + 1 dimensions. The gray shaded area is where the naiveunitarity bound ∆ > 1 holds. The red shaded area (w ≥ 1)should be excluded, as the longitudinal sound speed cL issuperluminal. The red disk corresponds to the Weyl-invariantcase, with Tµµ = 0 and ∆ = 0. The blue circle includes thefree theory, which is known to admit an improved the stresstensor that is traceless too. The blue shaded region suffersfrom longitudinal gradient instability (c2L < 0) in the limitcT � c.

for an essential reason: they lack a well definedPoincare invariant ground state. Indeed, the onlyPoincare invariant configuration would be ΦI = 0(or ΦI = const, which is equivalent by the shiftsymmetry). However, around this ‘vacuum’ the ki-netic terms for ΦI are not even analytic, so thetheory doesn’t admit a well defined vacuum. Onlyaround configurations with nonzero gradients ∂µΦI

the theory can be quantized perturbatively.

The scale invariance of the theory implies that once∂µΦI = αδIµ then all the values of α are equivalentup to rescalings. Hence, there are only two distinctconfigurations in principle: α = 1, which is breakspart of the Poincare group but admits a perturba-tive quantization; or 0, which would be relativisticbut isn’t really well defined. In other words, thereis no continuous controlable way to approach theunbroken symmetry case – the theory is bound todescribe only nonrelativistic states. Therefore thetheory isn’t really relativistic, even if we used arelativistic-looking language in (20) and (21).

A more heuristic reason to see that the EFTs (21)are essentially non-relativistic is to realize that thefields ΦI themselves play the role of spatial coordi-nates. Therefore, in some sense, the scale transfor-mation (20) acts on time and space differently (for∆ 6= 0). This has important consequences at thelevel of reconciling the features that the SI theories(21) with standard CFT results.

4. The ∆ = 0 case is special: it is invariant under localWeyl (or conformal) transformations of the metric

gµν → Λ2(xρ) gµν (25)

with an arbitrary function Λ(xρ) and no action onthe scalars. The symmetry group is bigger than justglobal SI because Λ(x) is a free function, and thisensures vanishing trace of the stress tensor Tµµ =0. This is the ‘conformal solid’ EFT consideredin [13], which leads to the relativistic sound speeds(18) and therefore doesn’t seem appropriate for realworld (i.e., with nonrelativistic sound speeds) SIsolids. This point is marked in Fig. 3 with a reddot.

Let us also remark that in the examples of Sec. IVof solids with manifest SI the scaling dimension ofthe operators that are naturally identified as thephonons also have ∆ = 0.

5. The case ∆ = d−22 (corresponding to w = −d−3

d−1 )

with F ({xn}) = x1 is also special: it is the freetheory, which is known to admit an improved stresstensor that is also traceless, see e.g. [23–25]. Thiscase is marked with a blue circle in Figs. 3, 4.

6. In the generic case (∆ 6= 0 or d−22 ) the stress ten-

sor isn’t traceless. It isn’t obvious whether one canconstruct an ‘improved’ traceless stress tensor [24–28], but it seems highly unlikely in the full nonlineartheory2. See [24, 29] for discussions on this point insimilar theories. Therefore these cases are scale butnot conformal invariant (i.e. with Tµµ 6= 0). Onemight find this surprising, however let us remindthat the theories (21) are intrinsically nonrelativis-tic (see point 3) because they can be quantized onlyaround Lorentz-breaking backgrounds.

The scale but non-conformal invariant elasticitytheories (21) might remind the reader the wellknown previous example given in [30]. Despitenaive similarities, however, the constructions arevery different. For instance, the example in [30]is a theory in Euclidean signature, and it managesto be scale- but not conformal-invariant because itlacks reflection-positivity [30] (see [23–25] for re-views on this point). The SI solid EFTs insteadescape conformal invariance by breaking Lorentzinvariance.

2 Even ignoring interactions, the theory consists two types of ‘free’scalars fields, the longitudinal and transverse modes, with gener-ically different sound speeds. In this case, one can see that thestandard improvement method [23] eliminates only the contri-bution from one of the scalars to the trace. This already hintsthat there is no possible improvement and the trace is genericallynonzero.

7

7. An intriguing feature of the relation (22) is that therange compatible with the NEC and with gradient(in)stability, corresponds to a surprising range in ∆that includes even negative values, as seen in Fig. 3.It is inevitable to compare with true CFTs, wherethe scaling dimension of scalar operator must obeythe so-called unitarity bound ∆ > d−2

2 . Even theconformal solid case (∆ = 0, w = 1/(d− 1)) evadesthis bound. Again, in our view this is not a signalof violating unitarity, but simply the consequencethat without full Lorentz invariance the unitaritybound is expected to be more permissive.

8. The natural question, then, is what are the boundson ∆ that apply for the theories (21). Stabilityand consistency arguments give rise to 3 types ofbounds: i) absence of ghosts (which amounts tothe the NEC, w > −1) and which translates into∆ > −1; ii) absence of gradient instabilities andiii) subluminality cL, T < c. We postpone this dis-cussion to Sec. III A, once the values for the soundspeeds are presented.

A. Sound speeds

Let us now return to the main point – how SI con-strains the phonon speeds. As mentioned in point 1),SI demands that the equation of state is linear p = w ρto all orders in ρ. In particular, this implies that thelinear bulk modulus is fixed in terms of the backgroundpressure as3

K = (1 + w)p = w (ρ+ p) . (26)

The general formulas for the sound speeds in any (SI ornot) solid EFT are

c2T =G

ρ+ pc2, c2L =

K + 2d−2d−1G

ρ+ pc2 . (27)

Plugging (26) into those, one obtains

c2L = w c2 + 2d− 2

d− 1c2T . (28)

The first evident remark is that, once w � 1, this equa-tion allows that the two sound speeds are small, whilethe solid being SI.

Next, we discuss the stability/constitency constraints.Absence of gradient instability and subluminality in thetransverse sector only places a constraint on the shearmodulus

0 < G ≤ ρ+ p (29)

3 This also restricts the nonlinear response for bulk strain defor-mations. We postpone this discussion to a forthcoming work[31].

-1.5 -1.0 -0.5 0.5 1.0 1.5 2.0Δ

-1.0

-0.5

0.5

1.0

1.5

2.0

ω

FIG. 4. The same as in Fig. 3 for a SI superfluid.

Once this is ensured, the analogous bounds on cL thenconstrain w. The absence of gradient instability, c2L > 0,places a stronger constraint on w than the NEC. From(28), we find4

w > − 2d− 2

d− 1

c2Tc2

. (30)

Since in most solids cT /c ∼ 10−4 at most, we representthis constraint in Figs. 3 and 4 as basically excludingthe region w < 0. From (22), this translates into anupper bound on the scaling dimension ∆ for SI solids∆ < 1/(d− 1).

On the other hand, the subluminality condition on (28)requires that w < 1, which translates into ∆ > − d−2

2(d−1) .

All in all, then, we are left with allowed scaling dimen-sions in the window

− d− 2

2(d− 1)< ∆ <

1

d− 1. (31)

Finally, using (28) into (9), one can relate the Poissonratio in a SI solid in terms of one sound speed and w,

c2Lc2

=(d− 1)(1− (d− 3)R)

R+ 1w (32)

or R =(d− 1)w − c2L

c2

(d− 3)(d− 1)w +c2Lc2

. (33)

Notice that small w and cL (i.e. more non-relativistic thesolid) allows for more auxetic behaviour – more negativeR is allowed. Conversely, in the conformal solid limit(w = 1/(d− 1)), one finds at most R > 0 (from cL < c).

4 In the exceptional case of the free theory (with V = X and

cL ,T = c), leads to w = − d−3d−1

so (30) is automatically satisfied

for any d.

8

B. Scale invariant superfluids

It is easy to extend the the above analysis of solidsthat realize SI to a simpler case, namely a SI relativisticsuperfluid, which consists in a single scalar field that hasa temporal vev for the gradient, Ψ = t + ψ. This casehas also been studied in [29]. The most general action atleading order in derivatives is S =

∫ddxP (X

(Ψ)) where

X(Ψ)≡ ∂µΨgµν∂νΨ. Scale invariance also allows for a

nontrivial scaling dimension for Ψ, ∆(Ψ)

, defined similarly

to (20) and it restricts the Lagrangian to be power lawin X

(Ψ). The only difference with respect to the SI solid

case is how the exponent relates to w. One can easily

find that Pw(X(Ψ)

) = Xw+1

2(Ψ) , and thus the ∆−w relation

is now

∆ =(d− 1)w − 1

w + 1, or w =

1− (d− 1)∆

(d− 1)(∆ + 1), (34)

which is shown in Fig. 4 for d = 4. Intrerestingly, theWeyl/conformal-invariant case also requires ∆ = 0. Inthis case, there is only one speed of sound c2s = w c2 sothe constraints from (no) gradient instability and sublu-minality are 0 < w < 1, which gives

− 1 < ∆ <d− 2

2− for SI superfluids (35)

IV. SOLIDS WITH MANIFEST SCALEINVARIANCE

Let us focus now on the other possibility mentionedin the introduction: that scale invariance (SI) is in factmanifest in the low energy theory, namely as a nontriv-ial infrared fixed point of the renormalization group. Themain physical effect of having the phonons as part of non-trivial IRFP is that the phonons are not isolated degreesof freedom, so one expects that they inevitably have dif-fusive behaviour. This translates in their dispersion rela-tion as acquiring an imaginary part, w = cT,L k − iΓ(k).Given that the diffusive part Γ(k) scales as k2, at lowenough energy the dispersion relation is still basicallyreal, linear and propagating5. Under this condition, itis justified to focus mainly on the real part of the dis-persion relation (i.e. on their the speeds), as we shallassume henceforth6.

Our goal is to study this case using holography, thatis, modelling the CFT as an effective theory in AdS

5 More precisely, the phonon propagates until the so-called Ioffe-Regel crossover, whose momentum depends on how strong isdissipation (the diffusion constant) compared to speed of prop-agation – the elastic modulus. For more details see [32] and inparticular fig.4 therein.

6 The complete study of the low energy dynamics (including thedissipative terms) can be obtained using Hydrodynamic tech-niques. See [33–35].

space and using the AdS/CFT dictionary and keepingin mind that we need an elastic sector which allowsfor a well defined elastic response. The simplest holo-graphic model is to consider a non-dynamical AdSd+1

space with d− 1 scalar fields ΦI propagating in it (withno backreaction on the metric). In Poincare coordinates7

ds2 = gabdxadxb = (`2/z2)(dz2 + dxµdx

µ), one can takethe scalar Lagrangian as in the previous section, V(Z, xn)with some function V and the invariants constructed fromIIJAdS = ∂aΦI∂bΦ

Jgab. These theories admit solutionswith spatial gradients of the form ΦI = δIµx

µ, whichbreak both Lorentz and SI. For a certain type of thepotential V , the holographic interpretation [37] of thesolutions is that the breaking of both Lorentz and SI isspontaneous.

Given that the stress tensor plays a prominent role inelasticity, and that a model with non-dynamical metricis interpreted holographically as a theory with no stresstensor operator (see e.g. [25]), we shall not discuss thelimit of no backreaction below.

Nevertheless, this simple model allows us to highlightan important point. The holographic CFT constructedas an AdS dual is meant to represent the infrared fixedpoint (IRFP) which controls the quantum critical ma-terial at low energies. Therefore it is clear that it onlystands for the effective field theory description that in asense emerges at low energies – SI itself is an emergentsymmetry for IRFPs. It is conceivable, then, that othersymmetries might be emergent as well.

In the present context, it is particularly relevant to in-clude the possibility that the IRFP exhibits an emergentLorentz invariance. Besides the fact that this allows usto treat the IRFP as a true CFT (invariant under emer-gent boost invariance, SI and special conformal transfor-mations), what this means in practice is that the fieldtheory is characterized by a well defined light cone speedce, generically different from the speed of light c. In or-der to comply with the fundamental principles of specialrelativity one needs to have ce < c and (in fact the limitof interest will be ce � c). The possibility that Lorentzinvariance arises as an emergent symmetry has been stud-ied e.g. in [38–46]. For the present work, we shall onlytake this as an assumption in order to construct a simplemodel.

The emergence of LI can be formulated a bit moreprecisely by saying that in addition to the standard (fun-damental) Minkowski metric ηµν the theory contains (orproduces dynamically) a spin-2 object in addition to thestandard Minkowski metric, and that the all the CFT op-erators couple to this emergent metric. To distinguish itfrom the fundamental Minkowski metric, we will denoteit as ηeµν . By definition, it allows to define an emergent

7 To make the distinction with the previous section clearer, lower-case latin indices a, b, . . . will refer to d+ 1 dimensional coordi-nates, so that schematically xa = {z , xµ}.

9

line-cone structure

ds2e = −c2e dt2 + dxidxi .

In terms of the usual spacetime coordinates with homo-geneous dimensionality xµ = {c t, xi }, the fundamen-tal metric reads simply ηµν = diag(−1, 1, ..., 1), and theemergent one

ηeµν = diag

(−c

2e

c2, 1, ..., 1

). (36)

The statement that the CFT exhibits emergent Lorentzinvariance then amounts to saying that all the correlatorsare constructed for instance from the emergent covariantdistance ∆xµ ∆xνηeµν .

Another clear consequence is that two different notionsof trace, which is especially relevant for the stress tensoroperator Tµν . Particularizing to the perfect fluid formTµν = diag(ρ, p, ...p) that describes the solid in the ho-mogeneous background, the two traces yield

Tµνηµν = −ρ+(d−1) p , and Tµνηeµν = −c2e

c2ρ+(d−1) p .

(37)This shows that the notion of an emergent conformaltheory

Tµνηeµν = 0 (38)

is perfectly compatible with the usual non-relativisticlimit (in the sense that p/ρ � 1) which is required forreal-world materials, so long as ce � c.

Another important lesson from (37) is that the emer-gent light cone speed is related to the equation of stateparameter (w = p/ρ) of the background,

ce =√

(d− 1)w c . (39)

Snooping for a moment at Eq. (40) this will match withEq. (28) found in Section 3 for the spontaneous breakingcase. Let us remark, however, that the symmetries arevery different in the two cases.

In retrospect, assuming that the IRFP has emer-gent Lorentz invariance with ce � c use the standardAdS/CFT dictionary in a setup that is non-relativisticin the sense that the speeds are small compared to thespeed of light. The only conceptually important pointis that the light-cone structure of the AdS space is alsocharacterized by the emergent speed ce. Since it is con-venient to use natural units where the speed parameterdoesn’t appear explicitly, the only point to keep in mindis that natural units are those where ce = 1. Alterna-tively, one can keep track of the factors of ce/c by simpledimensional analysis as above.

In the rest of this Section, we will consider this precisesetup (we model the IRFP as an emergent CFT withsmall ce light-cone speed parameter) and study the elasticresponse. In order to do this, we introduces an elasticsector as a set of scalars with nonzero spatial gradients

∂aφI as before, but we allow it to couple to the stress

tensor (in order to extract the elastic response in theusual fashion). In the AdS picture this means that thescalars backreact on the metric, which will be AdS4 onlyasymptotically.

This analysis follows very closely the steps of [19, 37,47–51] (see also [13, 52–61] for other treatments of theelastic response), with the only main difference that nowwe keep in mind that the light-cone speed is emergentand therefore we treat it basically as a new parameter.

As we shall see, in this case what happens is that trans-verse and longitudinal sound speeds satisfy8

c2L =1

d− 1c2e + 2

d− 2

d− 1c2T . (40)

This is structurally the same as derived for the conformalsolid EFT – it is the same as (18) but replacing c → ce.However, the quantitative difference is huge in the ce � climit. The longitudinal sound speeds is now bounded by

1d−1c

2e, and so taking ce of order 10−4c brings us to

realistic sound speed values.Another interesting aspect of the holographic models

such as the one presented here is that they allow to char-acterize rather systematically the elastic properties of thedifferent models. In particular we will see that thesemodels allow for one can have a significant auxetic be-haviour. Lastly, these models also incorporate finite tem-perature effects in a straightforward way. These effectsare important because it is possible to capture the melt-ing transition that happens at sufficiently large tempera-ture9. Needless to say, the mechanical response changesdrastically above or below the melting crossover.

A. Holographic models

As mentioned above, we shall model the low energy’critical’ behaviour using the standard holographic dictio-nary that maps CFT to gravitational physics in asymp-totically AdS spacetime. For simplicity, from now on weassume that the CFT lives in 2+1 spacetime dimensions,so that the gravitational dual is AdS4.

By assumption, the CFT contains operators that canbe identified with the displacement vectors. Their dualincarnation in the AdS4, are an identical a set of fields,ΦI , which propagate into the holographic dimension too.The equilibrium configuration for the scalars is ΦI = xI

and it defines the equilibrium configuration which breaks

8 Strictly speaking, we show this relation below for d = 3 and thed-depence is only an ‘educated guess’ here. However the mainpoint in the present discussion how the emergent speed ce enters.

9 To be precise, this transition is totally continuous and differentfrom the first order typical melting phase transition – ice to wa-ter. There are some models [50] where this transition can bediscontinuous, but second order.

10

the 2 + 1 Poincare group and more precisely translationsand rotations. The perturbations of those scalars aroundequilibrium encode the mechanical deformations of thesystem and directly the strain tensor as:

ΦI = xI + φI , εij =1

2(∂iφj + ∂j φi) (41)

The above identification permits to rewrite the full elasticresponse in terms of the dynamics of the scalar fields ΦI .For example an external shear deformation would simplycorrespond to a perturbation for which ∂xφy 6= 0.

We consider the generic holographic massive gravitymodels introduced in [47, 48], and studied in severaldirections in [19, 37, 49, 62–64]. The models are de-fined as a gravitational theory with negative cosmolog-ical constant Λ. The metric is locally Minkowskian,gµν ∼ diag(−c2e/c2, 1, 1, 1) + ..., with an input speed pa-rameter ce that is unrelated to the speed of light be-cause the 4D space is only holographic. We can thustake ce � c without affecting at all the consistency of thetheory, and which gives the important benefit of realizingconformal material with slow sound speeds as describedin hte introduction. In the following, we will work in theunits ce = 1 unless otherwise stated.

The model is then defined by the following action inthe 4D bulk space,

S =

∫d4x√−g

[R

2− Λ− m2 V (X,Z)

](42)

with IIJAdS = ∂aΦI∂bΦJgab and X ≡ 1

2Tr(IIJAdS) and Z =

det(IIJAdS). For simplicity, we focus on d = 3 but we willcomment on generic and universal features.

For specific choices of the potential V (X,Z), the model(42) represents the gravity dual of a CFT at finite tem-perature and zero charge density where translational in-variance is broken spontaneously. Using the standardAdS/CFT dictionary, this defines for us a CFT that willhave non-zero elastic moduli and so it can be interpretedas a model for a solid in a quantum critical regime. Moreprecisely, a well-defined elastic response can be definedfor potentials which decay at the boundary as V ∼ u3 orfaster [49]. Moreover, for potentials whose fall-off at theboundary is V ∼ u5 or faster, this elastic response is as-sociated to the presence of massless propagating phonons[37].

This field configuration admits an AdS black branegeometry

ds2 =1

u2

(−f(u) dt2 +

du2

f(u)+ dxidxj

),

f(u) = u3

∫ uh

u

(3

v4− m2

v4V (v2, v4)

)dv (43)

We fix the cosmological constant to Λ = −3. We assumethe presence of an event horizon at u = uh defined byf(uh) = 0. The associated entropy density is s = 2π/u2

h

and the corresponding temperature reads T = − f′(uh)4π .

The shear elastic modulus for these models can be ob-tained solving numerically the equation:

h′′ +

(f ′

f− 2

u

)h′ − 2m2 VX(u2, u4)

fh = 0 (44)

for the metric perturbation h ≡ δgxy. The perturbationis assumed to be static, ω = 0, and the subscript Xindicates the derivative with respect to X. In order toextract the retarded correlator we have to impose ingoingboundary conditions at the horizon, see [19] for moredetails. The UV expansion of the shear perturbationreads

h(u) = h0 (1 + . . . ) + h3 u3 + . . . (45)

where h0 represents a source for the Txy operator andh3 encodes the VEV of the stress tensor 〈Txy〉 [65]. Fol-lowing equation (6), which defines the shear modulus (asused in previous works [19, 37, 48–50]), we simply find

G =3

2

h3

h0(46)

At this stage, it is important to make a stop to discussthis result. To be more precise, here we are consideringthe response of the stress tensor to a geometrical sheardeformation h0. A more physical approach is to considerthe response of the stress tensor h3 in terms of a mechan-ical shear deformation ∂xφy 6= φs. At linear level, it doesnot make much difference. It is straightforward to checkthat:

h3

h0= − h3

φs(47)

The previous result can be understood noticing that thegauge invariant perturbation encoding a “gauge invariantshear strain” is indeed a combination of φs and h0 [47].Therefore it is clear the two will produce the same result.Once the framework will be extended at non-linear level,the interpretation in terms of the scalars perturbations ismuch more direct and simpler and it will be convenientto fix h0 = 0 once and forever.The solution for the shear modulus can be found analyt-ically in the limit m� T [19, 37]

G = m2

∫ uh

0

VX(ζ2 , ζ4

)ζ2

dζ + O(m4) (48)

On the other hand, the bulk modulus is defined inequation (7). Due to conformal symmetry, the stress en-ergy tensor is traceless, thus

p = Tii = Ttt/2 ≡ ρ/2 (49)

We can guess the volume dependence of the energy den-sity ρ quite easily. Consider a homogeneous system withequation of state γ = p/ρ in a box of volume V. In an

11

adiabatic process that changes volume, the energy den-sity scales with volume as ρ ∝ V−1−γ . Therefore in ourcase

ρ ∝ V−3/2 (50)

and the total bulk modulus is just

K =3

4ρ . (51)

B. Sound speeds

In the current setup, the phonons can be found asthe poles in the TijTkl correlator at finite wavenumberk. In the gravitational dual this is done by findingthe spectrum of quasi-normal modes. We shall not re-peat this exercise here, since it has already been done in[35, 37, 49, 66]. The conclusion of these works is thatfor b < 5/2 the spectrum of QNMs in our benchmark ofmodels contains gapless modes with a gapless dispersionrelation of the form

w = csk − iDk2 + . . . (52)

both for transverse [37] and longitudinal [35, 66] waves.Since the diffusive part scales as a higher power of k itis still possible to preserve a clear notion of propagatingsound modes and sound speeds – at low enough k.

Moreover, the sound speeds of the QNMs can be foundnumerically by following the motion of the pole as kchanges [35, 37, 66, 67]. The numerical result thus ob-tained for the sound speed agree formally with what oneexpects from elasticity theory, that is Eq. (15),

c2T =G

ρ+ p, c2L =

G + Kρ+ p

(53)

with ρ, P , G and K the energy density, pressure, and elas-tic the moduli respectively, for d = 3 space-time dimen-sions. This agreement justifies the physical identificationof these modes as physical phonons.

The speeds in Eq.(53) are expressed in the units of thelight-cone speed present in the AdS theory. In a trulyrelativistic CFT (with light cone speed identical to thespeed of light), the units are restored in Eq.(53) trivially:by a multiplicative c2 factor.

However, it is interesting to take the AdS theorymerely as a model for a low energy CFT with emergentlight-cone speed ce � c (in order that sound can be slowcompared to light). With this in mind, let us now restorethe factors of ce-dependence in (53). At the technicallevel, this can be done by performing a rescaling of thetime coordinate,

∂t →c

ce∂t , ω → c

ceω . (54)

From this we can immediately derive10 that:

cT,L →ceccT,L (56)

This has two immediate consequences. First, it followsthat the speeds satisfy

c2L =1

2c2e + c2T , (57)

which is the same as (40) for d = 3.Second, since the rescaling (54) affects the energy den-

sity ρ but not the pressures Tij , the full expression forthe sound speeds is

c2T =G

(ce/c)2ρ+ pc2e , c2L =

G + K(ce/c)2ρ+ p

c2e .

(58)It is interesting to rewrite these in terms of the physicalmass density, which relates to the energy density in theusual form

ρm ≡ ρ/c2 .

One then finds that Eq. (58) reduces to

c2T =G

ρm + p/c2e, c2L =

G + Kρm + p/c2e

. (59)

Recall that the spontaneous breaking case (Sec.III) leadsto expressions of the form

c2T =G

ρm + p/c2(60)

where the pressure contribution in the denominator ismuch more suppressed. Therefore, one can say that theeffect of having manifest SI with a ‘slow’ emergent cone(ce � c) ends up enhancing the pressure contribution inthe demonimator of the sound speed formulas, therebyreducing the sound speeds.

Upgrading the discussion to d dimensions, and usingthe tracelessness condition (p = (ce/c)

2ρ/(d − 1) in ddimensions) Eq. (59) further simplifies to

c2T =d− 1

d

Gρm

(manifest SI) . (61)

This is to be contrasted with the conventional expression(16)

c2T =Gρm

(spontaneous breaking) (62)

10 A similar conclusion can be reached by noticing that underrestoring ce:

G → G , χPP →(c

ce

)2

χPP (55)

12

(up to tiny O( pρmc2

) corrections), which holds in the EFT

picture (spontaneous breaking) but is not granted to ap-ply in the presence of an emergent CFT-like fixed point.

Note that the difference between (60) and (61) is in-dependent of ce. This deserves two comments. First, thediscrepancy looks surprising but actually it is due to thefact that the low energy theories are very different – SIis realized in a completely different way in the two cases.Second, this implies that the discrepancy persists even inthe limit ce � c. In particular, taking ce/c in the range10−5 − 10−4 brings the sound speeds into the range ofreal-world materials so from this point of view this has achance to correspond to a realistic material at a criticalpoint. It is tempting to say that the relation betwee cT ,G and ρm can provide a signature of whether a materialis controlled by such peculiar IR dynamics. (A similardiscrepancy arises also in the longitudinal sector.)

Let us remark that the case of spontaneous breaking ofSI with emergent Lorentz symmetry at low energy repre-sents another well defined option. A proper discussion ofit is beyond the scope of this paper, but let us offer onecomment. In this case, one wonders whether the EFTshould be obtained from coset construction referred tothe breaking of the ‘fundamental’ Poincare group (withspeed c) or from the emergent one (with speed ce), whichis also spontaneously broken in the ground state. If thelatter option is the relevant one, then we would expectEq. (59) to apply. In this case, however, the pressure isnot constrained by the emergent-tracelessness conditionso one wouldn’t obtain (61). Still, one would also ex-pect order-one deviations from (62), basically due to theenhancement of the pressure term in the denominator.

C. Elastic response in a benchmark model

After defining the linear response in abstract terms, werestrict ourselves to a specific and quite generic form ofthe potential V to make more quantitative statements.In particular, for the rest of the paper we consider thebenchmark potential:

V (X,Z) = Xa Zb−a

2 (63)

In order to ensure the consistency of this choice (63), andof the model (42) in general, one must impose a numberof requirements. First, absence of ghosts, absence of gra-dient instabilities locally in the 4D theory leads to limitthe parameters a, b in the range [7]:

a ≥ 0 ∧ b ≥ 1 (64)

Demanding the positivity of the linear elastic moduli, andof the energy density at low temperatures (see Eqs. (67)and (68) below) restricts b further as

b ≥ 3

2. (65)

FIG. 5. Poisson Ratio −1 < R < 1 for the benchmarkmodel (63) within the consistency region at low temperaturesT/m � 1. From orange to blue the material becomes moreand more auxetic, i.e. with a negative Poisson ratio. In theregion above the purple dashed line, the phonons are gap-less. The red dashed line marks the simple choice of potentialV (X,Z) = Xa. The black dashed line shows the region whereall local speeds are sub-luminal in the bulk (see Fig. 8 for theboundary values of the phonon speeds).

Finally, in order to restrict to the theories where thephonons are gapless we need to further impose [37, 49]:

b ≥ 5

2. (66)

Below 5/2 the phonons acquire a mass gap. This canbe translated as having additional (‘explicit’) sourcesthat break translational invariance, suggesting that thephonon speeds might depart from the expressions (53).

Notice that the constraints considered are purelybulk requirements and they represent just necessary butnot sufficient conditions for the full consistency of ourboundary field theory. In order to have a final verdict, adetailed QNMs computation would be needed.

Restricting ourselves to our benchmark model (63), wecan write the expressions for the energy density and pres-sure of the background as

ρ =1

u3h

+ m2 u2 b− 3h

2 b − 3, p =

1

2ρ (67)

as well as for the elastic moduli

G =a

2b− 3m2 u2b−3

h + O(m4) , K =3

4ρ (68)

and noticing immediately that the parameter a is whatdistinguish in the static response a solid with respect to

13

0 10 20 30 40/-1.0

-0.5

0.0

0.5

1.0ℛ

0 2 4 6 8

1

2

3

4

5

FIG. 6. Poisson ratio as a function of K/G for the model(63). The colors represent specific points in the parameterspace, which are shown in the inset, within the region of con-sistency. The various colored lines are produced changing thedimensionless parameter T/m at fixed a, b.

a fluid. For a = 0, the static shear modulus is zero andthe system does behave like a fluid. From (68) we canimmediately obtain the Poisson ratio R (9) for our holo-graphic models. We show how R depends on a, b at lowtemperature in Fig. 5.

At large T/m � 1, the Poisson ratio always goes tothe fluid value R = 1. This indicates that within ourmodel, no matter the choice of the potential V (X,Z),the limit of large temperature correspond to a fluid phasewith a maximum Poisson ratio. We can easily under-stand this phenomenon by noticing that at T/m� 1 thegraviton mass and therefore the additional structure in-duced by the scalars ΦI are completely negligible andthe phenomenology is simply the one of a relativisticstrongly coupled fluid. On the contrary, at small tem-perature, the effects of the scalars are dominant and thephenomenology depends crucially on the choice of the po-tential V (X,Z) and the Poisson ratio differs consistentlyfrom the fluid value. More precisely we can provide arough classification of benchmark model (63) as follows(see related Fig. 5):

• For small a and large b the Poisson ratio is largeand close to its upper limit R = 1. This class ofmodels refers therefore to incompressible and elas-tic materials such as rubber.

• For a ∼ b the Poisson ratio is in the range −0.5 <R < 0.5, similarly to the typical values for moststeels and rigid polymers.

• For large a and small b the Poisson ratio is negative(i.e. it exhibits auxetic behaviour) and close to itslower limit R = −1. As we will see later this iscorrelated with the presence of superluminal speeds

0.2 0.4 0.6 0.8 1.0T/m

0.2

0.4

0.6

0.8

1.0

/

0.2 0.4 0.6 0.8 1.0T/m

-0.2

0.2

0.4

0.6

0.8

1.0

ℛ

FIG. 7. Top: The elastic moduli for the potential V (X,Z) =X3 in function of T/m. Bottom: Poisson Ratio R in func-tion of the dimensionless temperature T/m for various choicesof potential. The specific (a, b) are indicated with the samecolors of Fig. 6. Notice that at large temperature the Poissonratio always goes towards the fluid limit R = 1.

of sound hinting towards a possible instability.

An even better way of classifying our theories consistsin plotting their Poisson artio R in function of thedimensionless quantity K/G as for example presentedin [68]. The similarities with the realistic results arepresented in Fig. 6. As already hinted in Fig. 5, smalla corresponds to foam-like material whether large a torubber-like materials.

As another important feature, we can analyze the be-haviour of the linear elastic moduli which are shown inFig. 7. It is evident from the Figure that both the mod-uli goes to zero in the limit of T/m� 1 in a continuousfashion which typical of viscoelastic and glassy materi-als [69]. Additionally, their ratio G/K goes to zero atlarge temperatures indicating again that at T/m � 1we are always in a fluid phase. Moreover, we can com-pute the behaviour of the Poisson ratio in function ofthe temperature (Fig. 7). Our results suggest that as

14

0.0 0.2 0.4 0.6 0.8 1.0T/m0.0

0.2

0.4

0.6

0.8

1.0c2

0 1 2 3 4 51.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

0.50.6

0.7

0.8

0.9

1

FIG. 8. Top: The longitudinal (thick) and transverse(dashed) speeds in function of T/m. The specific (a, b) are(3, 3), (3, 4), (4, 4) for red, blue and green. Bottom: Valuesof c2L at T = 0 as extracted from the formula (53). Thedashed purple line indicates the region where the phononsare gapless, which happens for b > 5/2 [37]. Below this line,the formulae in (53) (and consequently this plot) need notapply. The dashed red line is the potential V (X,Z) = Xa.More importantly the white region signals the appearance ofsuperluminal longitudinal phonons.

a generic property such a ratio decreases with increas-ing the dimensionless ratio T/m. We can think of theprevious properties as the “melting” in our holographicsystem, which is very similar indeed to the phenomenol-ogy of amorphous solids and glasses as already hinted inprevious literature [37, 70, 71].

As already introduced in the previous sections, the lin-ear elastic response directly defines the speed of propa-gation of transverse and longitudinal sounds in terms ofthe elastic moduli via eq.(15). Notice that formulae (15)are strictly speaking valid only in presence of masslessphonons ωT,L = cT,L k and therefore only for b > 5/2.The latter disagreement for b < 5/2 has been explic-itly checked for in [49]. The validity of formula (15) hasbeen ascertained directly by a direct comparison with

FIG. 9. The longitudinal speed c2L in function of the Pois-son ratio R, for various values of ce/c. The orange line is force/c = 1 and it coincides with the plot obtained for the con-formal solid EFT (shown as green bullets). The other linescorrespond to CFTs the representative values ce/c = 1/

√2

and 1/√

10. Notice that the auxetic behaviour (correspond-ing to R < 0) can be achieved in the CFT case with ce/c < 1,thus the conformal solid EFTs must have R > 0 and can’tbe much auxetic. Similar plots are obtained for the family ofsolids with spontaneously broken SI introduced in Section 3,by decreasing the equation of state parameter w, which playsa role analogous to c2e/c

2.

the QNMs spectrum for transverse [37] and longitudinal[35, 66] waves. Two important results follow:

• For a = 0 the speed of transverse sound is zero. Inaddition the speed of longitudinal sound is constantc2L = 1/2 and independent of the power b. Thephase dual to a = 0 is a fluid.

• At any value of T/m the relation c2L = 12 + c2T

holds. This is ensured by conformal symmetry [13]and it is proven by direct computation.

The results for the speeds are shown in Fig. 8. The leftpanel of Fig. 8 shows a typical behaviour of the speedsin function of the dimensionless temperature T/m. Thespeeds exhibit a continuous transition towards the largeT values cT = 0, c2L = 1/2. The right panel shows thevalue of the longitudinal speed at zero T , the maximalspeed in the system, inside the parameter space. Thewhite region evidentiates the region where the longitu-dinal speed is superluminal cL > 1. Curiously, the speedbecomes superluminal in the direction where the dualCFT becomes more and more exotic (auxetic).

Restoring the units in the previous equation is simplydone by noticing that cL, T are expressed in the units ofthe universal light-cone speed present in AdS gravity side,ce. Therefore, restoring the units we recover readily (40).As emphasized in the introduction, this has an importantphysical consequence: taking ce � c, we obtain a model

15

for a scale invariant material whose sound speeds can beas slow as necessary. For real world applications, noticethat ce/c should be at most of the order 10−4.

This also makes manifest another important point: inthe present construction, the AdS gravity model alsoshares this small universal speed ce. Therefore, it is clearthat our framework has no knowledge of the underly-ing ultraviolet completion (where eventually the speed oflight c plays a key role), including any relation to stringcompactifications. This is in tune with the view thatholography can be used as an effective method as arguedbefore e.g. in [72, 73].

Let us finish by showing the relation between the Pois-son ratio R and the longitudinal sound speed in the holo-graphic CFT model,

c2L =c2eR+ 1

. (69)

This makes manifest that in this theory R can be neg-ative (auxetic) if ce < c, and the how much auxetic thesolid can be depends on how small the ratio ce/c ie. Weshow this in Fig. 9. A similar plot can be obtained forthe generic EFTs with spontaneously broken scale invari-ance presented in Section 3, with the equation of stateparameter w playing the role of c2e/c

2.

V. CONCLUSIONS

We have studied the possible realization of scale invari-ance (SI) in the mechanical response of solid materials,considering both the cases for which SI is a spontaneouslybroken or a manifest symmetry. The latter case takes theform of a nontrivial infrared fixed point (IRFP), and wehave studied it using AdS/CFT methods.

For the spontaneously broken case, it is consistent toassume that the mass-spectrum is gapped and then useEffective Field Theory (EFT) methods to study the gap-less phonons as Goldstone bosons of the spontaneouslybroken spacetime symmetries. We have presented anew one-parameter family of EFTs of sponteneously bro-ken SI, which is basically parametrized by the equationof state parameter of the material in the background,w = p/ρ. This family includes the previously studiedcase of the so-called conformal solid EFT [13] as a spe-cial point, however for the generic values it nonlinearlyrealizes SI, but not the full conformal group.

We have shown that for both types of realization of SIit is possible to have slow speeds of sound compared tothe speed of light, once w is small. More specifically, wefound that the longitudinal and transverse sound speedsare related as

c2L = w c2 +2 (d− 2)

d− 1c2T (70)

with w the equation of state parameter of the materialand c the speed of light. For small w the two speeds are

small compared to c, as needed for potentially realisticapplications.

We have constructed manifest SI theories as holo-graphic duals of AdS models. In these case, we also foundthat (70) applies. We have argued that our models canbe interpreted as IRFP where in addition to having man-ifest SI, the theory is also Lorentz invariant with respectto a small ‘emergent’ light-cone speed ce (so that thetheory is actually a full Conformal Field Theory). Thesymmetries then relate

w =1

d− 1

c2ec2

.

Importantly, note that assuming ce/c� 1 is a consistentpossibility, and it immediately leads to slow phonons alsoin this CFT-like case. Another interesting consequence ofhaving a full emergent CFT with is that the relation be-tween the speeds and the elastic moduli change by O(1)factor. For instance the transverse sound speed becomesc2T = d−1

dGρm

instead of the usual form Gρm

, with G the

shear modulus and ρm the mass density. Thus, a de-viation from the usual formula G

ρmcould be used as a

smoking gun of the emergent SI and Lorentz in the ma-terial. The measurement of elastic properties in somecuprates has been recently done in Ref. [16]. Our workmotivates further investigation in this direction.

In our models with manifest SI, we have used stan-dard holographic methods to compute several elastic re-sponse parameters: the elastic moduli, the Poisson ratio,the propagation speed of the phonons in function of thevarious parameters of the model. We find that the max-imally auxetic solids (most negative Poisson ratio) canarise only for w � 1, that is, for slow sound speeds.Also, the temperature dependence of the different fea-tures suggest once more that these holographic modelsseem to interpolate between a fluid phase to a solid phaseby decreasing temperature. The crossover is continuousand very analogous to what happens to certain extentin glasses and amorphous materials. The behaviour ofthe vibrational modes in these holographic systems hasalready produced important developments in the studyof the latter [70, 71].

An obvious extension of this work, which is currentlyunder investigation, amounts to generalize these resultsto the non-linear regime up to arbitrarily large deforma-tions. An EFT description has been recently introducedin [7]. We plan to report in a forthcoming work [31] theanalysis of nonlinear elasticity in scale invariant system,comparing also the spontaneously broken vs. manifestcases.

Finally we hope that this and related works stimulatefurther experimental investigation towards the mechan-ical properties and the phonons dynamics in quantumcritical situations and scale invariant systems. Prelimi-nary interesting studies have been presented in [14, 15];more has definitely to come.

16

ACKNOWLEDGEMENTS

We thank L.Alberte, M.Ammon, T.Andrade, A. Es-posito, C, Hoyos, A. Jimenez Alba, A. Nicolis, R. Penco,A.Zaccone, K.Trachenko, Jan Zaanen, Weijia Li and Mi-lan Allan for useful discussions. OP thanks the orga-nizers of the workshop “Effective theories of quantumphases of matter” in Nordita, which stimulated the com-pletion of this work. We acknowledge the support ofthe Spanish Agencia Estatal de Investigacion throughthe grant FPA2017-88915-P and the Severo Ochoa Excel-lence grants SEV-2016-0597 and SEV-2016-0588, as wellas from the DURSI through grant 2017-SGR-1069.MBthanks Jiao Tong Shangai University, TDLee Institute,ITP Beijing and DUT Dalian for the kind hospitalityduring the completion of this work.

Appendix A: More on solid EFT in d dimensions

In order to compute the speeds of the phonon modes weneed to calculate the quadratic action on the perturba-tions around the background of the fields ΦI = α (xI+πI)which give us

IIJ = α2(δIJ + ∂IπJ + ∂JπI + ∂ µπI ∂µπJ). (A1)

The most relevant expressions up to quadratic order aresummarized below

Tr(I n) =α2n((d− 1)− n πi2 + 2n∂i π

Li

+ n (2n− 1) (∂i πLi )2 + n2 (∂i π

Tj )2), (A2)

Z = α2(d−1)(1 + 2 ∂i π

Li − πi2 + (∂i π

Li )2), (A3)

xn ≡Tr(I n)

Zn

d−1= d−1+n2 (∂i π

Tj )2 +

2(d− 2)

d− 1n2 (∂i π

Li )2,

(A4)where we have split the perturbation into longitudinaland transverse modes

∂i πTi = 0 , ∂[i π

Lj] = 0. (A5)

The action at second order is then

δS(2) = −∫ddx

(−N πi

2 + c2L (∂i πLi )2 + c2T (∂i π

Tj )2)

(A6)where

N = Z VZ , (A7)

c2L = 1 +2VZZ Z

VZ+

2(d− 2)

d− 1c2T , (A8)

c2T =

d−2∑n=1

n2 Vxn

Z VZ. (A9)

We would like to relate this to the bulk and shear moduli.The stress-energy tensor is

Tµν = − 2√−g

δS

δgµν= −ηµν V + 2

∂V

∂IIJ∂µΦI ∂νΦJ .

(A10)Our potential is a function of Z and xn, so

∂V

∂IIJ=

∂Z

∂IIJ∂V

∂Z

+

d−2∑n=1

(∂Tr(I n)

∂IIJ1

Zn/d−1− ∂Z

∂IIJn

d− 1

xnZ

) ∂V∂xn

.

(A11)

Let’s start computing the shear modulus. The shearstrain changes our background to

ΦI = xI + εIkxk (A12)

where we can take εIJ = εJI with no loss of generality. We

assume that εji 6= 0 for i 6= j and look at the component

Tij at first order in εji , and exptract the shear modu-

lus comparing with Eq.(5). Notice that the term ∂Z∂IIJ

cancels with ∂iΦI∂jΦ

J . To check this, first we make thederivative of Z with respect to IIJ using Jacobi’s formula

∂Z

∂IIJ= adjT (I)IJ . (A13)

Contracting this with

∂iΦI∂jΦ

J = (δIi δJj + δIi ε

Jj + δJj ε

Ii )α

2 + O(ε2). (A14)

At linear order we find that adjT (I)ij = −2 εij α2d−4

(with i 6= j) and adjT (I)ii = α2d−4(1 +O(ε)). Therefore

∂Z

∂IIJ(δIi δ

Jj + δIi ε

Jj + δJj ε

Ii )α

2

= adjT (I)ij α2 + 2α2(d−1) εij = 0. (A15)

The only non-zero term is then

Tij = 2

d−2∑n=1

∂V

∂xn

∂Tr((IKL)n)

∂IIJ1

Zn/d−1∂iΦ

I ∂jΦJ .

(A16)For the derivative of the traces,Tr(In) = II1 I2 II2 I3 . . . IIn−1 In IIn I1 , one finds

∂Tr(I)

∂IIJ= δIJ ,

∂Tr(I2)

∂IIJ= 2 IIJ ,

and

∂Tr(In)

∂IIJ= n II I3 II3 I4 . . . IIn−1 In IIn J (A17)

for n > 2, where we have used the cyclic property of thetrace. Since IIJ = α2(δIJ + εIJ + εJI ), finally we can findthat at linear order

Tij = 4 εij

d−2∑n=1

n2 ∂V

∂xn= 2 εij G. (A18)

17

For the bulk modulus we consider a purely volume defor-mation (zero shear), which can be parametrized as

α = 1 +κ

d− 1(A19)

where κ is the bulk strain, and we look at Tii.Notice that Vxn

doesn’t appear here as we can easilycheck using ∂iΦ

I ∂iΦJ = α2δIi δ

Ji

(∂Tr(I n)

∂Iii1

Zn/d−1− ∂Z

∂Iiin

d− 1

xnZ

)α2 = 0. (A20)

Therefore we arrive to

Tii = −V + 2Z VZ (A21)

and from the equation above we can already find theimportant result

ρ+ p = 2Z VZ . (A22)

Finally, using the definition of the bulk modulus (7)together with V ∝ α1−d and (A19), one arrives at

K ≡ −V dp

dV=

dp

dκ=

dTiidκ

(A23)

From (A21), then, one finds

K = 2Z VZ + 4Z2 VZZ = 4Z3/2 ∂Z

(√Z VZ

). (A24)

It is also possible to rewrite, the bulk modulus in termsof the equation of state of the solid, understood as thefunctional dependence of the pressure on the energy den-sity, that is P (ρ)|

�, by changing only the density – that

is at zero shear strain. Note that

K =dp

dρ

dρ

dZ

dZ

dκ=

dp

dρ2Z VZ , (A25)

where we use ρ = V . Therefore we find

Kρ+ p

=dp

dρ

∣∣∣�

, (A26)

which leads to (17). The subscript |�

stands to recallthat the derivative is at vanishing shear deformation.

[1] L. D. Landau and E. M. Lifshitz,Course of Theoretical Physics, Vol. 7,Theory of Elasticity(Pergamon Press, 1970).

[2] P. M. Chaikin and T. C. Lubensky,Principles of Condensed Matter Physics (CambridgeUniversity Press, 1995).

[3] H. Leutwyler, Helv. Phys. Acta 70, 275 (1997),arXiv:hep-ph/9609466 [hep-ph].

[4] S. Dubovsky, T. Gregoire, A. Nicolis, and R. Rattazzi,JHEP 03, 025 (2006), arXiv:hep-th/0512260 [hep-th].

[5] A. Nicolis, R. Penco, and R. A. Rosen, Phys. Rev. D89,045002 (2014), arXiv:1307.0517 [hep-th].

[6] A. Nicolis, R. Penco, F. Piazza, and R. Rattazzi, JHEP06, 155 (2015), arXiv:1501.03845 [hep-th].

[7] L. Alberte, M. Baggioli, V. C. Castillo, and O. Pujolas,Phys. Rev. D100, 065015 (2019), arXiv:1807.07474 [hep-th].

[8] M. Bianchi, D. Z. Freedman, and K. Skenderis, JHEP08, 041 (2001), arXiv:hep-th/0105276 [hep-th].

[9] E. Megias and O. Pujolas, JHEP 08, 081 (2014),arXiv:1401.4998 [hep-th].

[10] C. Hoyos, U. Kol, J. Sonnenschein, and S. Yankielowicz,JHEP 10, 181 (2013), arXiv:1307.2572 [hep-th].

[11] B. Bajc and A. R. Lugo, JHEP 07, 056 (2013),arXiv:1304.3051 [hep-th].

[12] I. Low and A. V. Manohar, Phys. Rev. Lett. 88, 101602(2002), arXiv:hep-th/0110285 [hep-th].

[13] A. Esposito, S. Garcia-Saenz, A. Nicolis, and R. Penco,JHEP 12, 113 (2017), arXiv:1708.09391 [hep-th].

[14] Y. Ishii, Y. Ouchi, S. Kawaguchi, H. Ishibashi, Y. Kub-ota, and S. Mori, arXiv preprint arXiv:1901.09502(2019).

[15] C. Setty, arXiv preprint arXiv:1902.00516 (2019).[16] S. K. Sahu, D. Jangade, A. Thamizhavel, M. M. Desh-

mukh, and V. Singh, (2019), arXiv:1910.02221 [cond-mat].

[17] P. Kovtun, D. T. Son, and A. O. Starinets, Phys. Rev.Lett. 94, 111601 (2005), arXiv:hep-th/0405231 [hep-th].

[18] S. A. Hartnoll, D. M. Ramirez, and J. E. Santos, JHEP03, 170 (2016), arXiv:1601.02757 [hep-th].

[19] L. Alberte, M. Baggioli, and O. Pujolas, JHEP 07, 074(2016), arXiv:1601.03384 [hep-th].

[20] P. Burikham and N. Poovuttikul, Phys. Rev. D94,106001 (2016), arXiv:1601.04624 [hep-th].

[21] M. P. Gochan, H. Li, and K. S. Bedell, ArXiv e-prints(2018), arXiv:1801.08627 [cond-mat.str-el].

[22] M. F. Thorpe and I. Jasiuk, Proceedings: Mathematicaland Physical Sciences 438, 531 (1992).

[23] R. Jackiw and S. Y. Pi, J. Phys. A44, 223001 (2011),arXiv:1101.4886 [math-ph].

[24] Y. Nakayama, Phys. Rept. 569, 1 (2015),arXiv:1302.0884 [hep-th].

[25] S. Rychkov, Conformal Field Theory in D >= 3 Dimensions,SpringerBriefs in Physics (2016) arXiv:1601.05000 [hep-th].

[26] S. R. Coleman and R. Jackiw, Annals Phys. 67, 552(1971).

[27] C. G. Callan, Jr., S. R. Coleman, and R. Jackiw, AnnalsPhys. 59, 42 (1970).

[28] J. Polchinski, Nucl. Phys. B303, 226 (1988).[29] E. Pajer and D. Stefanyszyn, JHEP 06, 008 (2019),

arXiv:1812.05133 [hep-th].[30] V. Riva and J. L. Cardy, Phys. Lett. B622, 339 (2005),

arXiv:hep-th/0504197 [hep-th].

18

[31] M. Baggioli, V. Cancer-Castillo, O. Pujolas, S. Renaux-Petel, and K. Yang, .

[32] M. Ammon, M. Baggioli, and A. Jimenez-Alba, (2019),arXiv:1904.05785 [hep-th].

[33] P. C. Martin, O. Parodi, and P. S. Pershan, Phys. Rev.A 6, 2401 (1972).

[34] L. V. Delacretaz, B. Gouteraux, S. A. Hartnoll,and A. Karlsson, Phys. Rev. B96, 195128 (2017),arXiv:1702.05104 [cond-mat.str-el].

[35] M. Ammon, M. Baggioli, S. Gray, and S. Grieninger,(2019), arXiv:1905.09164 [hep-th].

[36] J. Armas and A. Jain, (2019), arXiv:1908.01175 [hep-th].[37] L. Alberte, M. Ammon, M. Baggioli, A. Jimenez-Alba,

and O. Pujolas, (2017), arXiv:1711.03100 [hep-th].[38] S. Chadha and H. B. Nielsen, Nucl. Phys. B217, 125

(1983).[39] H. B. Nielsen and M. Ninomiya, Nucl. Phys. B141, 153

(1978).[40] O. Vafek, Z. Tesanovic, and M. Franz, Phys. Rev. Lett.

89, 157003 (2002), arXiv:cond-mat/0203047 [cond-mat].[41] S. Groot Nibbelink and M. Pospelov, Phys. Rev. Lett.

94, 081601 (2005), arXiv:hep-ph/0404271 [hep-ph].[42] R. Iengo, J. G. Russo, and M. Serone, JHEP 11, 020

(2009), arXiv:0906.3477 [hep-th].[43] G. F. Giudice, M. Raidal, and A. Strumia, Phys. Lett.

B690, 272 (2010), arXiv:1003.2364 [hep-ph].[44] M. M. Anber and J. F. Donoghue, Phys. Rev. D83,

105027 (2011), arXiv:1102.0789 [hep-th].[45] O. Pujolas and S. Sibiryakov, JHEP 01, 062 (2012),

arXiv:1109.4495 [hep-th].[46] G. Bednik, O. Pujolas, and S. Sibiryakov, JHEP 11, 064

(2013), arXiv:1305.0011 [hep-th].[47] M. Baggioli and O. Pujolas, Phys. Rev. Lett. 114, 251602

(2015), arXiv:1411.1003 [hep-th].[48] L. Alberte, M. Baggioli, A. Khmelnitsky, and O. Pujolas,

JHEP 02, 114 (2016), arXiv:1510.09089 [hep-th].[49] L. Alberte, M. Ammon, M. Baggioli, A. Jimenez, and

O. Pujolas, JHEP 01, 129 (2018), arXiv:1708.08477 [hep-th].

[50] M. Baggioli and A. Buchel, JHEP 03, 146 (2019),arXiv:1805.06756 [hep-th].

[51] T. Andrade, M. Baggioli, and O. Pujolas, (2019),arXiv:1903.02859 [hep-th].

[52] D. Vegh, (2013), arXiv:1301.0537 [hep-th].

[53] M. Blake, D. Tong, and D. Vegh, Phys. Rev. Lett. 112,071602 (2014), arXiv:1310.3832 [hep-th].

[54] T. Andrade and B. Withers, JHEP 05, 101 (2014),arXiv:1311.5157 [hep-th].

[55] T. Andrade, M. Baggioli, A. Krikun, and N. Poovuttikul,JHEP 02, 085 (2018), arXiv:1708.08306 [hep-th].

[56] M. Baggioli and K. Trachenko, JHEP 03, 093 (2019),arXiv:1807.10530 [hep-th].

[57] M. Baggioli and K. Trachenko, (2018), arXiv:1808.05391[hep-th].

[58] S. Grozdanov and N. Poovuttikul, (2018),arXiv:1801.03199 [hep-th].

[59] S. Grozdanov, A. Lucas, and N. Poovuttikul, (2018),arXiv:1810.10016 [hep-th].

[60] A. Amoretti, D. Arean, B. Gouteraux, and D. Musso,(2017), arXiv:1711.06610 [hep-th].

[61] A. Amoretti, D. Arean, B. Gouteraux, and D. Musso,(2018), arXiv:1812.08118 [hep-th].

[62] M. Baggioli and M. Goykhman, JHEP 07, 035 (2015),arXiv:1504.05561 [hep-th].

[63] M. Baggioli and D. K. Brattan, Class. Quant. Grav. 34,015008 (2017), arXiv:1504.07635 [hep-th].

[64] M. Baggioli and M. Goykhman, JHEP 01, 011 (2016),arXiv:1510.06363 [hep-th].

[65] K. Skenderis, , Class. Quant. Grav. 19, 5849 (2002),arXiv:hep-th/0209067 [hep-th].

[66] M. Baggioli and S. Grieninger, (2019), arXiv:1905.09488[hep-th].

[67] M. Baggioli, U. Gran, A. J. Alba, M. Tornso, andT. Zingg, JHEP 09, 013 (2019), arXiv:1905.00804 [hep-th].

[68] G. N. Greaves, A. L. Greer, R. S. Lakes, and T. Rouxel,Nature Materials 10, 823 EP (2011), review Article.

[69] A. Cavagna, 476, 51 (2009), arXiv:0903.4264 [cond-mat.stat-mech].

[70] M. Baggioli and A. Zaccone, (2018), arXiv:1810.09516[cond-mat.soft].

[71] M. Baggioli and A. Zaccone, arXiv preprintarXiv:1812.07245 (2018).

[72] M. Baggioli and O. Pujolas, JHEP 12, 107 (2016),arXiv:1604.08915 [hep-th].

[73] M. Baggioli and O. Pujolas, JHEP 01, 040 (2017),arXiv:1601.07897 [hep-th].