Temperature crossovers in the specific heat of amorphous magnets

Héctor Ochoa1
Department of Physics, Columbia University, New York, NY 10027, USA
(*Electronic mail: [email protected])
(Dated: 25 December 2024)
It is argued that the specific heat of amorphous solids at low temperatures can be understood to arise from a single
branch of collective modes. The idea is illustrated in a model of a correlated spin glass for which magnetic anisotropies
are present but they are completely frustrated by disorder. The low-energy spectrum is dominated by soft modes cor-
responding to propagating Halperin-Saslow spin waves at short wavelengths, evolving into a relaxation and a diffuson
arXiv:2412.18305v1 [cond-mat.mes-hall] 24 Dec 2024

mode at the long wavelengths. The latter gives rise to an anomalous temperature behavior of the specific heat at T ≪ T∗ :
C ∼ T in d = 3 solids, C ∼ T ln(1/T ) in d = 2. The temperature scale T∗ has a non-trivial dependence on the damping
coefficient describing magnetic friction, which can be related to fluctuations of the spin-torque operator via a Green-
Kubo formula. Halperin-Saslow modes can also become diffusive due to disclination motion (plastic flow). Extensions
of these ideas to other systems (including disordered phases of correlated electrons in cuprate superconductors and of
moiré superlattices) are discussed.

I. INTRODUCTION really there as T 3 . There are no theorems, but

I have a feeling (I think Brenig has made this
The central theme of this article is an old question rela- point also) that there are in some sense two en-
tive to the low-temperature properties of amorphous (or more tirely separate branches to the spectrum. There
generically, disordered or glass) materials and their appar- is a long wave length collective branch, but there
ent excess of boson modes at low energies. Specifically, the is at the same time something else at very short
question addressed here was directly posed by Brian Coles to wave lengths that is very local.
Philip Anderson (different accounts point to each of them for Anderson’s final point is certainly valid in many situations.
coining the term spin glass) about 50 years ago:1 The TTLS picture gained quickly acceptance thanks to vari-
If one were to start from a totally different ous experimental observations, in particular, the dependence
view and regard the spin glasses and real glasses of the specific heat on the experimental observation times and
as not too bad a deviation from a long range fer- the saturation of acoustic attenuation at high intensities.4
romagnet or antiferromagnet and a long range However, these observations do not imply that the TTLS
lattice, and say what we have are excitations or model is necessarily of universal application. An eloquent cri-
damped spin waves on the one hand and damped tique in that regard has been expressed by Leggett and Vural.5
phonons on the other which would distort our More broadly, the appearance of these short-wavelength exci-
spin wave specific heat, could you possibly get tations at low energies, specifically, the finite density of states
out the linear specific heat? at zero energy (constant in the original model) is the conse-
quence of the chosen distribution of model parameters of the
Coles’ different view expressed in this question was in con- two-level systems. This choice is ad hoc and not easily justi-
trast to the tunnelling two-level states (TTLS) of Anderson, fied from microscopics. For example, Anderson et al. origi-
Halperin, Varma2 and, independently, of Phillips,3 which An- nally wrote:2
derson had just presented in the same conference. Anderson’s
answer was, according to the same record (here abbreviated): It is not easy to speculate meaningfully on
the distribution of the barriers. Most of the sim-
That doesn’t work because the phonon spe- ple models we have tried tend to suggest a pre-
cific heat is visible and it is T 3 . On the other dominance of rather high barriers; but the ob-
hand, one of the most fascinating things about servations, as discussed above, suggest rather
the ordinary glasses is that the coefficient of T 3 that there is a very broad distribution of barrier
is wrong, relative to sound velocity. heights, with low ones as a probable as high ones.
Coles insisted: Here the barriers is one of the TTLS model parameters: the
energy barrier separating two minima in the complex energy
Doesn’t that mean that part of the quasi landscape of the amorphous material. Since the model is
phonon like excitations are no longer going as heuristic, it cannot provide us with an estimate of the tem-
T 3 but as T ? perature scale corresponding to a crossover from a TTLS- to a
To what Anderson finally replied: phonon-dominated specific heat. This is the main goal of this
manuscript starting from Coles’ point of view instead.
Well, the point is you see the phonons, in fact, This point of view has been recently put forward by Bag-
you see more than you expect from the phonons gioli and Zaccone6,7 borrowing intuition in part from in-
already and that is something else again. That is commensurate lattices,8–10 where a phason can play the role
 2

where ⟨·⟩ denotes a statistical averages (see concrete definition

below).
Alternatively, we may consider a different model to de-
scribe amorphous magnets (e.g. in the case of rare-earth
transition-metal compounds) given by the Hamiltonian17
 2
Ĥ = −J ∑ Ŝi · Ŝ j − K ∑ ξi · Ŝi , (3)
⟨i j⟩ i

where exchange interactions are restricted to nearest neigh-

bors in a lattice of period a, and K represents an easy-axis
magnetic anisotropy imposed by spin-orbit coupling and the
crystalline environment. In a polycrystalline material, the
FIG. 1. Possible spin textures in the ground state of the model of anisotropy axis ξi can be taken as a random variable with zero
Eq. (3). The second case corresponds to a correlated spin glass.12 mean (indicating no global preferred direction of anisotropy)
and correlated over a length Ldis .
These models differ in two crucial aspects: i) The Hamil-
of a(n) (over-)damped phonon to give a linear-in-T specific tonian in Eq. (1) is invariant under spin rotations; the Hamil-
heat.11 In this article, I elaborate on these ideas trying to es- tonian in Eq. (3) is not. ii) In the model described by Eq. (1)
tablish a connection with microscopic models. I present a dis- there is a single microscopic energy/length scale; in the model
cussion of the long-time dynamics of a spin model for amor- of Eq. (3), there are actually two, the length Ldis introduced by
phous magnets forming a so-called correlated spin glass.12 At disorder correlations in the ξi -orientations, and the intrinsic
short wavelengths, the hydrodynamic modes of this phase re- length scale resulting from the competition between the two
semble propagating Halperin-Saslow spin waves.13 At long
p
terms in the Hamiltonian, LDW = a J/K. In a crystalline ma-
wavelengths, one mode is fully relaxed and the other be- terial, this length can be interpreted as the characteristic width
comes diffusive. In this model the mechanism is ultimately of a domain wall connecting spin configurations parallel and
related to the lack of a local conservation law for spin an- anti-parallel to a fixed anisotropy axis. In a polycrystalline
gular momentum. Alternatively, in an Edwards-Anderson material, this length together with a finite Ldis (think of it as
spin glass,14 Halperin-Saslow modes become diffusive due a the crystalline grain size) defines a third mesoscopic scale
to the dissipative motion of vortex disclinations of the order corresponding to the magnetic correlation length in the sense
pararmeter.15,16 Associated with the change in character of the of Imry-Ma,18
soft modes there are crossovers in the temperature behavior of
various physical properties, in particular, in the spin wave-   4
LDW 4−d
dominated specific heat. The discussion is extended to other Lc ≈ Ldis , (4)
disordered systems. Ldis

where d = 2, 3 is the dimensionality of the system.

II. TWO MODELS OF A SPIN GLASS This last equation describes two different types of spin tex-
tures (see Fig. 1) below the freezing temperature arising from
the competition between Ldis and LDW , or equivalently, the
Textbook models of a spin glass are usually described by a
two energy scales in the Hamiltonian of Eq. (3). If LDW ≪
Hamiltonian of the Heisenberg type,
Ldis , anisotropy dominates over exchange and one should ex-
1 pect a texture in which the magnetization follows closely the
Ĥ = − Ji j Ŝi · Ŝ j , (1)
2∑
local anisotropy direction; the magnetic correlation length sat-
ij urates to Lc = Ldis . In the opposite limit, LDW ≫ Ldis , ex-
β γ
change dominates over anisotropy, producing a smooth spin
where Ŝi are spin operators satisfying [Ŝiα , Ŝ j ] = iεαβ γ Ŝi δi j texture varying on the scale Lc ≫ Ldis . This is known as a
defined on a lattice of N sites, not necessarily periodic, and correlated spin glass.12
Ji j are exchange coupling constants between sites i and j;
εαβ γ is the Levi-Civita symbol and repeated greek indices are
summed hereafter. In a spin glass, the couplings Ji j are treated III. HYDRODYNAMIC MODES
as random variables with zero mean. Collinear magnetic or-
der with a net magnetization is precluded, yet individual spins
A. Order parameter
can freeze below certain temperature to form a complex non-
collinear spin texture. This form of order is described by a
finite value q of an Edwards-Anderson order parameter,14 Consider the following 1-spin operator:

1 1
R̂αβ =
β β
qδα,β = ⟨Ŝiα ⟩⟨Ŝi ⟩, (2) ⟨Ŝi ⟩Ŝiα . (5)
N∑i N∑i
 3

In both the Edwards-Anderson and correlated spin glass Similarly, we can define the local operator
cases we have non-zero expectation values of this operator, 1
R̂αβ = qδαβ ̸= 0. For concreteness, the statistical averages
β
R̂αβ (r) = ⟨Ŝi ⟩Ŝiα δ (r − ri ) . (10)
qN ∑
are taking with respect to a density matrix operator ρ̂G repre- i
senting a macrostate G, ⟨·⟩ = Tr [ρ̂G ·]. From the spin commutator algebra we just have
Let us consider now a transformation g of the macrostate,
−i 
G → G′ = gG, consisting of a global spin rotation, g ∈ SO(3):

ŝα (r1 ) , R̂β γ (r2 ) = εαβ λ R̂λ γ (r1 ) δ (r1 − r2 ) . (11)
h̄
ρ̂G′ = Û (θ) ρ̂GÛ † (θ) , with Û (θ) = e−iθ·Ŝtot , (6) However, these are not good hydrodynamic operators since
(even after coarse-graining) ⟨R̂αβ (r)⟩ = ̸ 0 in a glass. Follow-
and Ŝtot is the total spin operator, ing Halperin and Saslow, we consider instead
1
Ŝtot = ∑ Ŝi . (7) θ̂α (r) = εαβ γ R̂γβ (r). (12)
2
i
This definition only makes full sense if the expectation value
Note that in a spin glass, ⟨Ŝtot ⟩ = 0 (no equilibrium magner- of operator R̂γβ (r) does not deviate much from the Edwards-
tization). Generically, ρ̂G′ ̸= ρ̂G ; the expectation value of the Anderson order parameter, ⟨R̂αβ (r)⟩ ≈ δαβ , and therefore,
operator in Eq. (5) in the new macrostate G′ is ⟨θ̂α (r)⟩ ≈ 0 upon appropriate coarse-graining.
Next, we consider spectral functions of two arbitrary oper-
′
R̂αβ = qRαβ (θ) , (8) ators Â, B̂ with time dependence described in the Heisenberg
picture,
where Rαβ (θ) are matrix elements of the SO(3) rotation R̂(θ) 1 ∞
Z
′′
dt eiωt Â(t), B̂† (0) ,
 
associated with Û(θ). χÂ,B̂
(ω) = (13)
2h̄ −∞
For both the Edwards-Anderson and the correlated spin
glass, G and G′ represent physically different macrostates but and complex response functions given by
with the same free energy, i.e., rigid spin rotations of the ′′
dω χÂ,B̂ (ω)
Z ∞
glass introduce no free-energy cost. This is evident for the χÂ,B̂ (z) = (Imz ̸= 0). (14)
−∞ π ω −z
Edwards-Anderson spin glass of the first model since Û(θ)
is a symmetry of the Hamiltonian in Eq. (1). The Edwards- As stated before, the low-frequency response of the system is
Anderson spin glass breaks spontaneously this symmetry, and dominated by fluctuations of the Fourier components of op-
θ = (θx , θy , θz ) parametrize three branches of soft (Goldstone) erators θ̂(r), ŝ(r); in particular, cross correlations satisfy the
modes.13 This is less evident in the case of a correlated spin sum rule
glass since Û(θ) is not a symmetry of the Hamiltonian. Yet dω ′′ 1 
Z ∞ 
magnetic anisotropy can be assumed to be completely frus- χθ̂ ,ŝ (ω) = θ̂α,k1 , ŝβ ,−k2 (15)
−∞ π α,k1 β ,k2 h̄
trated in the limit LDS ≫ Ldis . In that case we still expect i
Z
d d r e−i(k1 −k2 )·r δαβ R̂γγ (r) − R̂αβ (r)
 
θ to parametrize three branches of soft modes dominating =
2V
the response at low frequencies. However, the dynamics of
≈ iδαβ δk1 ,k2 .
these modes is not protected by a local conservation law and
they should be overdamped at long wavelengths. The low- The last result is true if ⟨R̂αβ (r)⟩ ≈ δαβ . It means that if
frequency spin response is diffusive, and that suffices to give coarse-grained (without worrying much about the exact pro-
a linear in T specific heat (T ln 1/T in d = 2) at the lowest cedure, only that it exists) density operators associated with
temperatures. rotation angles of the SO(3) order parameter and the gener-
ator of rotations can be defined locally, then these operators
must be related by canonical conjugation relations quickly de-
B. Hydrodynamic operators caying in space. There is at least one situation in which this
is questionable and will be discussed later. Note that the last
The argument goes as follows. The idea is that the long- result does not guarantee the existence of Goldstone modes,
time response of the spin glass is dominated by collective which require the more stringent sum rules19,20
modes associated with fluctuations of the three angles θ = Z ∞
dω
(θx , θy , θz ), which relax on long (hydrodymamic) times com- lim ω n χθ̂′′ (ω) = 0 ∀n > 0. (16)
k1 ,k2 →0 −∞ π α,k1 ,ŝβ ,k2
pared to the rest of modes of the spin dynamics due to the
free-energy invariance described above. A minimal hydro- To probe the latter, one needs to show 1) that there are no long-
dynamic theory should include these variables as well as a range forces between local operators and 2) that there is a local
coarse-grained spin density associated with the generator of conservation law for the coarse-grained spin density. So long
rotations, Ji j decays quickly as the separation of sites i and j grows,
condition 1) should be satisfied for local operators. Condition
ŝ (r) = h̄ ∑ Ŝi δ (r − ri ) . (9) 2) can only be the result of a microscopic symmetry, which is
i present in model 1, but not in model 2.
 4

C. Dispersion relation

Consider now specifically the complex response functions

χθ̂i ,θ̂ j (z) grouped in a matrix, [χ̂(z)]i j = χθ̂i ,θ̂ j (z), where the 2

subscript i span spin and spatial indices/coordinates; in par-

ticular, χθ̂i ,θ̂ j = χθ̂α (r1 ),θ̂ (r2 ) with i = (α, r1 ), j = (α, r2 )
β
describes the local rotation of the order parameter in po-
1
sition r1 due to a magnetic torque produced by e.g. spin
accumulation21 in position r2 . For its Fourier components,
i = (α, k1 ), j = (α, k2 ), the result in Eq. (15) allows us to
write a dispersion-relation representation of the form
−1 1
h̄2 χm

χ̂(z) = χ̂ −1 (0) − 2 2 z2 − iz Σ̂(z) . (17)
g µB
The first term is the static susceptibility of rotation vari-
3
ables
 −1 corresponding to a second derivative of the free energy,
χ̂ (0) i j = ∂ 2 F/∂ θi ∂ θ j . Given the free-energy invariance


with respect to spin rotations discussed before, χ̂ −1 (0) should

2
go to 0 in the limit k1 , k2 → 0. We parametrize this by
 −1 
χ̂ (0) (α,k ),(β ,k ) = Aα |k1 |2 δα,β δk1 ,k2 , (18)
1 2
1
where Aα is some fraction of Jqa2−d and represents the order-
parameter stiffness with respect to rotations along α-axis.
Here I am retaining only diagonal terms in momenta as-
suming that translational invariance is effectively restored at 1
long wavelengths (or upon coarse-graining of hydrodynami-
cal variables).
In the second term we have the magnetic susceptibility re-
lating the macroscopic magnetization density with an applied FIG. 2. Density of states of branch α in units of the inverse of
the cut-off frequency ωc as deduced from the numerical evaluation
magnetic field, m = M/V = χm B. In a spin glass the response
of Eq. (27) for different values of the damping coefficient γα in (a)
is similar to an isotropic paramagnet; in particular, in the cor- d = 2 and (b) d = 3 solids.
related spin glass from Imry-Ma arguments one can write
d  2d
g2 µB2 Lc 2 g2 µB2 LDW 4−d
   
χm ≈ d = d . (19) where τ̂ ≡ −i Ŝtot , Ĥ is the spin torque operator. We con-
qa K Ldis qa K Ldis clude that in the hydrodynamical limit (taking z = ω + i0+ →
Finally, the third term is the Laplace transform of a mem- 0) the response in Eq. (17) can be approximated by
ory matrix describing the effect of magnetic forces produced g2 µB2 δαβ
by the rest of microscopic degrees of freedom on the hydro- χ(α,k),(β ,k) (ω) ≈ 2
, (21)
dynamic variables. In the Mori formalism,20 the memory ma- h̄ χm cα |k| − ω 2 − iγα ω
2 2

trix can be written as a Kubo correlation function of the spin where cα is the spin-wave velocity of the correlated spin glass,
current density operator, ŝ˙ = −i[ŝ, Ĥ]/h̄, with hydrodynamic √ 
s  d
modes projected-out from the Hamiltonian evolution. The g2 µB2 Aα aq JK Ldis 4−d
important observation is that for a Hamiltonian of the type cα = ∼ , (22)
h̄2 χm h̄ LDW
of model 1, there must be a local conservation law for the
(coarse-grained) spin density as a consequence of the symme- and γα is a damping coefficient given by a Green-Kubo for-
try under spin rotations, implying that limk1 ,k2 →0 Σ(z) → 0. mula of the form
In this case, the restoring variables θ parametrize long-lived
g2 µB2 2
excitation with lifetimes satisfying τ1 ∝ |k|2 . These are the γα = lim χτ̂′′ ,τ̂ (ω) . (23)
Halperin-Saslow waves of spin glasses.13 h̄2V χm ω→0 ω α α
However, for microscopic Hamiltonian of the type of model The poles of the susceptibility in Eq. (21) lie at
2, in which anisotropies or other form relativistic corrections r
to the spin Hamiltonian are present, spin angular momentum iγ 2 γα
ωα (k) = ± c2α |k|2 − α − i (24a)
is not conserved. In the long-wavelength limit we have 4 2
( γα
1
Z ∞ ′′
dω χτ̂α ,τ̂β (ω) ±cα |k| − i 2 if cα |k| ≫ γ2α ,
 
lim Σ̂(z) (α,k = , (20) ≈ c2 |k|2 (24b)
k1 ,k2 →0 1 ),(β ,k2 ) V −∞ iπ ω(ω − z) −iγα , −i αγα if cα |k| ≪ γ2α .
 5

In the first limit we have propagating spin waves of the type of

Halpein-Saslow. In the second limit, one evolves into a fully 1
gapped relaxation mode, and the other to a diffusive mode
which grabs most of the spectral weight at low frequencies. 0.20

0.100

0.15

IV. SPECIFIC HEAT 0.010

0.10

The specific heat per site at constant field (taken 0) can be

0.001
written as 0.05

h̄ω 2
     
∂S h̄ω
Z ∞
−2
C=T = kB dω sinh D (ω) , 10-4
∂ T B=0 0 kB T kB T
0.005 0.010 0.050 0.100 0.500 1
(25)
where D(ω) is the density of modes per site at frequency ω.
At low frequencies, we have argued that the response is dom-
inated by the collective modes θα in Eq. (17), defining 1

2
Tr χ̂ −1 (0) · Imχ̂ ω + i0+ .


D (ω) = (26)
πNω 0.100 0.12

Here the trace is over indices i = (α, k). In the approximation 0.10

of Eq. (21), each branch labelled by α contributes as 0.010

0.08

  0.06

−1 ω γα
Dα (ω) = ωc fd , , with (27a) 0.04
ωc ωc 0.001
zd+1
Z 1 0.02

2dy
fd (x, y) = dz . (27b)
π 0 (z2 − x2 )2 + x2 y2
0.005 0.010 0.050 0.100 0.500 1
The remaining integral in the definition of the dimensionless
function fd (x, y) comes from the summation over momenta.
The subindex d = 2, 3 refers to dimensionality of the systems FIG. 3. Specific heat as a function of temperature for different
considered below. Note that in the formula for d = 2 the ef- values of γα (from bottom to top in units of ωc : 0, 0.01, 0.1, 1, 10).
fects of quantum confinement in thin films22 are not taken into The inset shows the specific heat as a function of temperature for the
lowest temperatures, T < Tγ = 0.1Tc . The dashed blue and red lines
account but they could be included following the same steps
correspond to the analytical estimate in Eq. 29 and Debye theory,
as in Ref. 23. The specific heat reads respectively.
 
T Tγ
C = kB gd , , with (28a)
Tc Tc
Z ∞ What is the crossover temperature T∗ ? Figure 2 shows that
gd (x, y) = 2x dz z2 sinh−2 z fd (2xz, y) , (28b) there are two regimes we must consider. If γα < 2ωc , only a
0
fraction of the spin waves are overdamped. This translates to
and where Tγ ≡ h̄γα /kB , Tc ≡ h̄ωc /kB . In all these expres- a non-monotonic behavior of Dα (ω): it first decreases, then
sions ωc is a cut-off for the linear dispersion of the collective it grows following the otherwise expected ω d−1 dependence,
modes, which is ultimately related to the inverse of the coarse- and then it decreases again to make up for the excess of den-
graining length of hydrodynamic variables proportional to Lc sity of states at low frequencies. In d = 2, the logarithmic de-
in Eq. (4). pendence in f2 (x, y) ceases to be appreciable when x ∼ y/2π.
If γα = 0, then fd (x, 0) = dxd−1 , and at low temperatures In d = 3, it is around x2 ∼ 2y/π. However, if γα > 2ωc , then
one recovers Debye’s law for the contribution of mode α to the whole spectrum is incoherent and the density of states is a
the specific heat, C ∼ T d . However, for finite damping, the monotonically decreasing function of ω, fully dominated by
evaluation of Eq. (27) in Fig. 2 shows that there is a transfer the diffusive poles. Moreover, in this regime the density of
of spectral weight from high to low energies required by the states scales as fd (x, y) = yhd (xy), i.e., the dependence on the
presence of a broad diffusive peak in the response function at first argument x is really through the combination xy, and thus
long wavelengths. In d = 3 the density of states is finite at ω = the separation between high and low temperatures is given by
0, and diverges logarithmically in the case of d = 2. This gives the condition xy ∼ 1. Putting all these arguments together, one
rise to a deviation of the specific heat at low temperatures T ≪ finds
T∗ of the form 
T T2
h i
min 4πγ , 2Tcγ if d = 2,

 2πT Tγ ln Tc2
  
if d = 2,
3Tc2 2Tγ T T∗ ≈ (30)
q 
C = kB × 2πT T (29) Tc Tγ Tc2
min 2π , 2Tγ if d = 3.
γ
if d = 3.
 2 
Tc
 6

Figure 3 shows C as a function of T in double logarithmic which is the Jacobi identity for the covariant derivative Dµ =
scale from the numerical evaluation of Eq. (28). The calcula- ∂µ − Ωµ ×.
tion agrees well with the analytical estimates in Eq. (29) for If the quaternion field is single valued, ∂i ∂ j q = ∂ j ∂i q,
the lowest temperatures T < T∗ estimated in Eq. (30). The then we have ρk = 0 in the definition above. The property
two regimes in the density of states translates to two differ- ∂i Ω j − ∂ j Ωi = Ωi × Ω j is akin to the Mermin-Ho relation25
ent scenarios for the temperature dependence of the specific in the A-phase of superfluid 3 He. It expresses that vorticity in
heat. When the density of states is non-monotonic, C displays one of the angles θα parametrizing the local spin frame can
three different behaviors with T : The anomalous dependence be unwinded (in multiples of 4π) by a texture of the Edwards-
in Eq.(29) (incoherent regime), followed by the otherwise ex- Anderson order parameter and does not introduce a singular-
pected T d dependence (Debye regime) at T∗ < Tc , and finally ity. On the contrary, if this relation is not satisfied, it im-
the classical result (equipartition regime) around Tc . The in- plies winding in multiples of 2π, which cannot be undone
sets in Fig. 3 illustrates the first crossover in linear scale. As smoothly: There is a line of vortex cores associated with an
the damping coefficient increases, T∗ approaches Tc and the essential branch cut where the order parameter is multivalued
Debye regime disappears. When the density of states is mono- (where antipodal points of the SO(3) hypersphere are identi-
tonically decreasing, i.e., when all the modes are overdamped, fied).
there is a direct crossover from the incoherent to the equipar- For a glass described by the Hamiltonian in Eq. (1), Volovik
tition regime at T∗ ≪ Tc . and Dzyaloshinskii proposed15,16 that in account for discli-
nations the macroscopic spin dynamics (coarse-grained over
many separated vortex cores) should be derived from a phase-
V. DISCLINATIONS space Lagrangian density of the form L = gµh̄B Ω0 · m −
H [m, Ωi ], with the Hamiltonian density accounting for the
One instance in which the definition of local operators θ̂α is free energy cost of non-equilibrium magnetization and ex-
questionable is if the non-collinear texture characterizing the change couplings,
glass state does not admit a description in terms of a smoothly
varying rotation of a local spin frame due to the presence of m2 A
Z2 vortex disclinations. The possibility of these defects can H [m, Ωi ] = + Ω2i , (35)
2χm 2
be understood from the topology of the SO(3) manifold intro-
duced before. Since the SU(2) group of spin rotations is iso- and supplemented by Poisson brackets derived from the spin
morphic to the unit hypersphere S3 , a generic operator Û can operator algebra,26
be represented by a 4-dimensional unit vector (or quaternion) n o gµ 
B
q = (w, v), with w2 + |v|2 = 1. The elements of the associated mα (r), Ωi (r′ ) =
β
−δαβ ∂i + εαβ γ Ωi (r) δ r − r′ .
γ 

SO(3) matrix are h̄
(36)
Rαβ = (1 − 2|v|2 )δαβ + 2vα vβ − 2εαβ γ . (31) The equations of motion derived from this theory are
Since q and −q correspond to the same rotation matrix, we h̄
can conclude that SO(3) ∼ = RP3 , i.e., the hypersphere S3 with ṁ = A ∂i Ωi , (37a)
gµB
antipodal points identified as the same. gµB
The macroscopic description of these defects is provided Ω̇i = (∂i m + m × Ωi ) . (37b)
by a non-abelian generalization of the disclination density in h̄χm
2D crystals. It is convenient to introduce the velocity field The first one describes the conservation of spin angular mo-
Ωi = 2∂i q ∧ q∗ , where q∗ = (w, −v) and ∧ represents the
mentum. Together with the constitutive relation m = h̄χ m
gµB Ω0 ,
Hamilton product of two quaternions (i.e., a matrix product);
by definition24 the first entry in Ωi is identically zero (i.e., Ωi the second one is equivalent to ji = 0. Thus in the ab-
is really a vector in spin space, not a matrix). This field plays a sence of dissipative terms there is no disclination current and
role analogous to the strain tensor. The density of disclination Dt ρi = ρ̇i − Ω0 × ρi = 0, which describes macroscopic spin
lines is given by precession in the presence of a non-equilibrium magnetiza-
tion density; these oscillations correspond to Halperin-Saslow
εi jk ρk = ∂i Ω j − ∂ j Ωi − Ωi × Ω j . (32) waves. However, in the presence of dissipative terms the first
equation is still true for the model in Eq. (1), but now there
Similarly, the disclination current can be defined as is also a dissipative disclination current ji = −γ Ωi . Halperin-
Saslow modes disperse as in Eq. (24) and become overdamped
ji = Ω̇i − ∂i Ω0 − Ω0 × Ωi , (33) at long wavelengths.15,16 This mechanism can be understood
in analogy with supercurrent relaxation due to vortex motion
where Ω0 = 2q̇ ∧ q∗ is the precession frequency. The conti- in superfluids: The disclination current modifies the constitu-
nuity equation for disclinations follows from the geometrical tive relation equivalent to the Josephson formula, introducing
relation damping in the analogue of the second sound.27 These argu-
ments can be extended to disclination motion in crystalline
Dt ρi + εi jk D j jk = 0, (34) and amorphous solid.28,29
 7

VI. DISCUSSION above, the new one (Bloch-Grüneisen scale) associated with
electron kinematics in the Fermi surface. In a theory involv-
The theory can be extended to other amorphous systems ing overdamped soft modes a disparity between Debye and
not involving magnetic order. The non-collinear magnetic Bloch-Grüneisen temperatures must be invoked in order to ex-
texture describing the correlated spin glass would be a com- plain the simultaneous observation of anomalous specific heat
plex ion mass density in structurally disordered materials. In and linear-in-T resistivity since the latter is just an extension
that case the phases θα would represent generalized position of classical equipartition to low temperatures due to the trans-
coordinates of a group of ions. If phase shifts of the mass fer of spectral weight to a diffusive peak. On the contrary, in a
density produce no free-energy cost, then we would expect a theory involving TTLS of electronic origin, the statistical dis-
soft mode for each θα . However, these phases correspond to tribution of these events in energy must be assumed to be very
complex atomic re-arrangements involving motion of a group broad.
of ions with respect to others, and consequently, these modes In conclusion, I have presented a discussion of two possible
should become diffusive at long wavelengths due to mechan- microscopic mechanisms for the transfer of spectral weight of
ical friction. The best analogy is with the sliding phason in a a collective mode from high to low energies in amorphous
bipartite incommensurate lattice.8–10 Although TTLS events magnets. In one case it is just the result of magnetic fric-
are typically ascribed to collective dynamics of group of ions tion with local anisotropy axes that are globally frustrated.
too, note the difference between these two scenarios: In the In the other case it is the result of the dissipative motion of
present theory (as elaborated in Sec. III), θα parametrize disclinations. In both cases one has overdamped Halperin-
small non-equilibrium deviations in which the system remains Saslow modes at long wavelengths, which gives rise to a spe-
within a given (local) minimum basin of the free energy asso- cific heat C ∼ T in bulk d = 3 solids, and C ∼ T ln(1/T ) in
ciated with macrostate G, while in the TTLS theory groups of d = 2 layered materials, below some scale T∗ identified in
ions tunnel between local minima. Eq. (30). Both mechanism relies on a finite (albeit probably
Back to the central question: Do we need to invoke two small) stiffness constant. In that regard, note that signatures
branches of excitations, one propagating (acoustic waves) and of long-range spin signals have been reported in amorphous
the other localized (TTLS) in order to explain a deviation from magnets,37 which might be an indication of collective spin
Debye theory? Or is it enough to consider a single branch of transport encoded in the rigid dynamics of a non-collinear or-
excitations that evolve from propagating to diffusive waves? der parameter.38
The honest conclusion is that there is no way to discriminate
between these two scenarios if we focus on the specific heat
alone. In particular, if we look at the insets of Fig. 3, the total ACKNOWLEDGMENTS
specific heat (black curve) at low temperatures is very well
described by the sum of a Debye (in blue) and a anomalous This work is supported by the Spanish MCI/AEI/FEDER,
(Eq. 29, in red) components, although the black curve was Grant No. PID2021-128760NB-I00. I would like to acknowl-
generated by a model including a single branch of excitations. edge Ricardo Zarzuela, Yaroslav Tserkovnyak and Rafael M.
It is also instructive to consider the present theory in the Fernandes for past collaborations that contributed to consoli-
context of possible glassy phases of strongly correlated mate- date some of these ideas. Special thanks should go to Matteo
rials such as in the cuprate superconductors or in moiré sys- Baggioli for electronic correspondence that clarified some of
tems. Bashan et al.30 have recently proposed that scattering these ideas before consolidation.
off TTLS events in a metallic glass phase can give rise to non-
Fermi liquid behavior. Note that the cuprates, which are lay-
ered materials (thus effectively d = 2) display a specific heat DATA AVAILABILITY STATEMENT
C ∼ T ln(1/T ), which fits well with the present scenario of
a single overdamped branch of excitations. In the context of All plots are generated by direct numerical evaluation of the
moiré systems, Fernandes and I31 have recently proposed that expressions provided in the body of the manuscript. Model
scattering with phason modes of the moiré pattern could ex- parameters are indicated in the figure and/or in the figures’
plain the strange metal behavior observed in twisted bilayer caption. If the reader has issues with the integrals I will be
graphene.32–34 Damping of the phason modes is the manifes- happy to share my Mathematica notebook.
tation of interlayer mechanical friction in this case.35
1 Amorphous
Nevertheless, there are important differences between these Magnetism (edited by H. O. Hooper and A. M. de Graaf,
two examples both at the theoretical and phenomenological Plenum Press, New York, 1973).
2 P. W. Anderson, B. I. Halperin, and C. M. Varma, Philos. Mag. 25, 1 (1972).
levels. In the theory of Bashan et al. all the degrees of freedom 3 W. A. Phillips, J. Low Temp. Phys. 7, 351 (1972).
are (in principle) electronic, while in our theory for moiré sys- 4 See W. A. Phillips Rep. Prog. Phys. 50, 1657 (1987), and references therein.
tems the phasons represent lattice degrees of freedom. Note 5 A. J. Leggett and D. C. Vural, J. Phys. Chem. B 117, 12966 (2013).

also that cuprates are both strange and bad metals,36 while 6 M. Baggioli and A. Zaccone, Phys. Rev. Research 1, 012010(R) (2019).
7 M. Baggioli and A. Zaccone, Int. J. Mod. Phys. B 35, 2130002 (2021).
in twisted bilayer graphene the resistivity saturates in appar- 8 R. Zeyher and W. Finger, Phys. Rev. Lett. 49, 1833 (1982).
ent compliance with the Mott-Ioffe-Regel limit.34 Finally, the 9 W. Finger and T. M. Rice, Phys. Rev. Lett. 49, 468 (1982).
inclusion of two different degrees of freedom (mechanic and 10 W. Finger and T. M. Rice, Phys. Rev. B 28, 340 (1983).

electronic) introduces also two scales playing the role of Tc 11 A. Cano and A. P. Levanyuk, Phys. Rev. Lett. 93, 245902 (2004).
 8

12 E. M. Chudnovsky, W. M. Saslow, and R. A. Serota, Phys. Rev. B 33, 251 27 R. A. Davison, L. V. Delacre’taz, B. Goute’raux, and S. A. Hartnoll, Phys.
(1986). Rev. B 94, 054502 (2016).
13 B. I. Halperin and W. M. Saslow, Phys. Rev. B 16, 2154 (1977). 28 M. Baggioli, M. Landry, and A. Zaccone, Phys. Rev. E 105, 024602 (2022).
14 S. F. Edwards and P. W. Anderson, J. Phys. F 5, 965 (1975). 29 M. Baggioli and B. Gouteraux, Rev. Mod. Phys., 95, 011001 (2023).
15 I. E. Dzyaloshinskii and G. E.Volovik, J. Phys. (Paris) 39, 693 (1978). 30 N. Bashan, E. Tulipman, J. Schmalian, and E. Berg, Phys. Rev. Lett. 132,
16 G. E.Volovik and I. E. Dzyaloshinskii, Zh. Eksp. Teor. Fiz. 75, 1110 [Sov. 236501 (2024).
Phys. JETP 48(3)] (1978). 31 H. Ochoa and R. M. Fernandes, Phys. Rev. B 108, 075168 (2023).
17 R. Harris, M. Plischke, and M. J. Zuckermann, Phys. Rev. Lett. 31, 160 32 H. Polshyn, M. Yankowitz, S. Chen, Y. Zhang, K. Watanabe, T. Taniguchi,

(1973). C. R. Dean, and A. F. Young, Nat. Phys. 15, 1011 (2019).

18 Y. Imry and S. K. Ma, Phys. Rev. Lett. 35, 1399 (1975). 33 Y. Cao, D. Chowdhury, D. Rodan-Legrain, O. Rubies-Bigorda, K. Watan-
19 R. V. Lange, Phys. Rev. 146, 301 (1966). abe, T. Taniguchi, T. Senthil, and P. Jarillo-Herrero, Phys. Rev. Lett. 124,
20 D. Forster, Hydrodynamic Fluctuations, Broken Symmetry, and Correlation 076801 (2020).
Functions (CRC Press, Boca Raton, FL, 2019). 34 A. Jaoui, I. Das, G. Di Battista, J. Díez-Mérida, X. Lu, K. Watanabe, T.
21 Y. Tserkovnyak and H. Ochoa, Phys. Rev. B 96, 100402(R) (2017). Taniguchi, H. Ishizuka, L. Levitov, and D. K. Efetov, Nat. Phys. 18, 633
22 M. Sidorova, A. D. Semenov, A. Zaccone, I. Charaev, M. Gonzalez, A. (2022).
Schilling, S. Gyger, and S. Steinhauer, Phys. Rev. B 110, 134513 (2024). 35 H. Ochoa and R. M. Fernandes, Phys. Rev. Lett. 128, 065901 (2022).
23 A. Zaccone, Phys. Rev. Mater. 8, 056001 (2024). 36 V. J. Emery and S. A. Kivelson, Phys. Rev. Lett. 74, 3253 (1995).
24 The Hamilton product of two quaternions is defined as q ∧ q = (w w − 37 D. Wesenberg, T. Liu, D. Balzar, M. Wu, and B. L. Zink, Nat. Phys. 13, 987
1 2 1 2
v1 · v2 , w1 v2 + w2 v1 + v1 × v2 ). (2017).
25 N. D. Mermin and T.-L. Ho, Phys. Rev. Lett. 36, 594 (1976). 38 H. Ochoa, R. Zarzuela, and Y. Tserkovnyak, Phys. Rev. B 98, 054424
26 I. E. Dzyaloshinskii and G. E.Volovik, Ann. Phys. 125, 67 (1980). (2018).