Fractal non-Fermi liquids from moiré-Hofstadter phonons

Ajesh Kumar,1, ∗ Zihan Cheng,1, ∗ and Andrew C. Potter1

1
Department of Physics, University of Texas at Austin, Austin, TX 78712, USA
We theoretically explore 2d moiré heterostructures in lattice-commensurate magnetic fields as
platforms for quantum simulation of a paradigmatic model of non-Fermi liquid physics: a Fermi-
surface coupled to a fluctuating gauge field. In these moiré-Hofstadter (MH) systems, long-
wavelength acoustic phonons exhibit singular interactions with electrons analogous to those of elec-
trons with 2d gauge fields. This leads to a breakdown of Fermi-liquid theory at low temperatures.
We show that a combination of large moiré-unit cell size, tunable Fermi-surface topology, and en-
hanced coupling to interlayer sliding modes, enhance these effects by over many orders-of-magnitude
compared to bulk crystals, placing them within experimental reach. Though we find that the asymp-
arXiv:2004.04744v3 [cond-mat.str-el] 11 Dec 2020

totic low-temperature non-Fermi liquid regime remains at prohibitively low temperatures, striking
precursor non-Fermi liquid signatures can be observed, and we propose surface acoustic wave atten-
uation and quantum oscillation transport experiments. We also study the motion of MH acoustic
(MHA)-polarons, which we predict exhibit logarithmically diverging effective mass and unconven-
tional magnetic field scaling for scaling of cyclotron resonance frequency and quantum oscillation
amplitude.

I. INTRODUCTION rather surprising, example: phonons of a crystal in a

The problem of metallic electrons strongly coupled to
fluctuating gapless (bosonic) collective modes is believed
to underly some of the least-understood quantum phe-
nomena, from strange-metal phases of high-temperature
superconductors [1–3], to metallic quantum critical sys-
tems [4, 5], composite fermion liquids [6], and gap-
less spin-liquids [7–10]. These systems lack well-defined
quasi-particles and are not captured by conventional
Fermi liquid theory paradigm. The detailed behavior of
these non-Fermi liquid (NFL) systems remain poorly un-
derstood due to the absence of naturally controlled the-
oretical calculations [10–12] or efficient numerical meth-
ods, complex materials chemistry, relatively high impu-
rity concentrations, and limited ability to tune the elec-
tron density or interactions in the underlying host mate-
rials. To this end alternative platforms to explore NFL
behavior in simpler, cleaner, and more tunable material
platforms are highly desirable.
A common and reliable source of gapless bosonic collec-
tive modes are: Nambu-Goldstone modes (NGM), such
as acoustic phonons and magnons, arising due to spon-
taneously broken continuous symmetries. However, the
same mechanism that ensures their masslessness ordi-
narily causes NGM to decouple from electrons at low
temperatures, producing conventional Fermi-liquid be-
havior. Exceptions occur for rotational NGM, includ-
ing nematic NGM [13] and magnons in spin-orbit cou-
pled metals [14, 15], for which the electron-NGM cou-
pling does not freeze out at low temperatures. However,
these examples require continuous rotation symmetry, a
situation that can be at best approximately realized in
crystalline materials.
Watanabe and Vishwanath [16] derived a general cri-
terion for exceptional NGM, and pointed out another,
∗ These authors contributed equally to this work
magnetic field. The non-commutative nature of trans-
lations in a magnetic field result in electrons coupling
to phonon fluctuations in much the same way as they
would couple to dynamically fluctuating magnetic fields.
An important caveat is that magnetic fields tend to pro-
duce non-dispersing Landau levels, destroying the Fermi-
surface at the single-particle level. In a crystal, the band-
width of Landau levels, which sets an upper bound on the
energy scale for observing NFL behavior, scales as e−1/ν
where ν is the number of flux per unit cell. For atomic-
scale lattices, reaching ν ∼ 1, would require astronomical
B & 104 T. For this reason, the prediction of [16] has not
been experimentally tested.
In contrast, moiré-superlattice potentials of small-
angle twisted structures can have sufficiently large unit-
cells to reach ν ∼ O(1) with laboratory magnetic fields.
In this paper we explore these systems as platforms
for exploring NFL physics, focusing on twisted bilayer
graphene (TBG) as a particularly promising example due
to its large moiré-potential.
II. MOIRÉ-HOFSTADTER (MH) BANDS AND
PHONONS
A. Electronic band structure
Our analysis begins from Bistritzer and Macdonald’s
(BM) continuum model for twisted bilayer graphene
(TBG) [17–19], with single-electron Hamiltonian:
 
h(−θ/2) T (r)
H(r) = , (1)
T † (r) h(θ/2)
where h(θ) = v Π + sgn(θ) k2θ · σ θ is the Hamiltonian


for a single graphene layer twisted by angle θ, Π = p+eA
is the canonical momentum with in magnetic field B =
∇ × A = Bẑ, kθ = −kθ ŷ ≡ − 8π θ
3a sin 2 ŷ is the vector
 2

FIG. 1. Magnetic band structure – (a), (b), (c) Magnetic bands structures of spin up valence bands in the magnetic
Brillouin zone for θ = 1.5◦ , ν = 1/3, B = 18.0T , θ = 1.1◦ , ν = 1/2, B = 14.5T and θ = 0.8◦ , ν = 1, B = 15.4T . In (a) and (b)
there is a pair of Dirac points at K, K 0 and an electron pocket at Γ point, whereas in (c) Dirac points at K, K 0 smear out to
oblong hole pockets near M, M 0 , forming a three-fold Fermi surface. The magnetic Brillouin zone is shown as a blue hexagon.
(d), (e), (f) bands structure along the high symmetry path Γ − M − K − Γ − M0 − K for θ = 1.5◦ , ν = 1/3, θ = 1.1◦ , ν = 1/2
and θ = 0.8◦ , ν = 1.

connecting the Dirac points in the two layers, and a is {a1 , qa2 }, and exhibits dispersive Bloch-bands labeled by
the graphene lattice spacing. T (r) represent interlayer magnetic quasi-momenta.
tunneling with the (approximate) spatial periodicity of
the moiré lattice Fig. 1(a) and (d) show the lowest two bands near
X charge-neutrality for θ = 1.5◦ , and B = 18T (ν = 13 ).
T (r) = w e−igj ·r Tj . (2) Apart from having lower bandwidth and spin-splitting,
j the bands resemble those of single-layer graphene in zero-
field. In particular, the conduction (valence) bands ex-
√
3kθ
√
hibits a pair of Dirac points near charge-neutrality at
Here g 0 = 0, g 1,2 = ∓x̂ + 3ŷ are moiré-
2
reciprocal-lattice vectors, and Tj = η + σ2πj/3 are sub- the moiré K, K 0 points, which merge into a single Γ-
lattice matrices. The real-space lattice vectors for the centered hole (electron) pocket at the top of the conduc-
moiré√cell are
(for small twist angle, θ  1): a1,2 = tion band (bottom of the valence band), crossing through
a a van-Hove (VH) singularity with nearly-nested hexago-
2θ ∓ 3x̂ + ŷ . For numerical results presented below, nal Fermi-surface at intermediate filling along the way.
we take ~v ≈ 610 meV nm, w ≈ 110 meV, and η ≈ 0.82,
Bands at θ = 1.1◦ (shown in Fig. 1(b) and (e)) exhibit a
which accounts corrugation [20].
similar Fermi surface evolution with doping, and also in-
Varying B produces a fractal “Hofstadter butterfly” herit the zero-magnetic field property of having reduced
energy spectrum [18, 19, 21, 22]. Here, instead, we focus bandwidth near the magic angle [17]. In this case, be-
on fixed B such that number of magnetic flux per moiré cause of strong electron-electron interactions relative to
unit cell is a rational fraction [18] (see also Appendix A): the small bandwidth, correlated insulating and supercon-
√ ducting behaviors have been observed [23]. However, at
3a2 p the magnetic fields that we consider here, the supercon-
ν= = , with p, q ∈ Z, (3)
4πθ2 `2B q ductivity near the magic angle is suppressed [23], and the
√ correlated insulating behavior arises only near integer fill-
and magnetic length `B = 1/ eB. At these com- ings of the flat moiré bands [24, 25], which can be gener-
mensurate fields, the Hamiltonian possesses a peri- ically avoided [26]. Therefore, we expect the correlated
odic magnetic-lattice translation symmetry generated by insulating and superconducting behaviors do not neces-
 3

(a) = 1.5° = 1/3 (b) = 1.1° = 1/2 (c) = 0.8° = 1

Coupling strength (meV/nm)

Coupling strength (meV/nm)

Coupling strength (meV/nm)

2 1
0.2
1
0 0.0 0
1
0.2 1
2
M0 K0 K M M0 K0 K M M0 K0 K M

FIG. 2. Electron-phonon coupling – electron-phonon coupling Γm , for x-polarization phonons in the limit of zero momentum
transfer for (a) θ = 1.5◦ , ν = 1/3, (b) θ = 1.1◦ , ν = 1/2 and (c) θ = 0.8◦ , ν = 1.

sarily obscure the non-Fermi liquid physics discussed in
this work.
At special flat Fermi-surface (FFS) fillings, which oc-
cur near (but not precisely at) VH fillings, the Fermi-
surface curvature vanishes at certain points on the Fermi-
surface, and remains small over extended patches, which
will enhance electron-phonon interaction effects. Taylor
expanding the dispersion in small momentum displace-
ments from the center of these flat-patches, the vanishing
Fermi-surface curvature requires a vanishing quadratic
term for the tangential dispersion. Due to mirror symme-
try around the center of the flat-patch, the next leading
term is quartic. Near the flat-patches both the quadratic
and the quartic terms contribute to the dispersion:
1 2
 k = vF k ⊥ +
2m k
where k⊥ and kk are patch-momentum components per-
pendicular and parallel to the Fermi-surface. Away from
the FFS points, there is a crossover momentum scale
1 the lattice: r ← (r − u(r)) [28], in which the lattice
q∗ = √2mλ , below which the leading dependence on kk is
k ∼ kk2 and above which it is k ∼ kk4 .
In SLG, FFS fillings were predicted to enable corre-
lated insulators or superconductors [27], but occur at pro-
hibitively high electron density ∼ 1015 cm−2 above charge
neutrality. In moiré systems, the FFS occur at much
lower fillings and can be experimentally explored. More-
over, the degree of Ferm-surface nesting (alignment be-
tween antipodal flat-patches), which controls the compe-
tition between interaction-driven orders, can be adjusted
by twist-engineering. For example, for the parameters
shown in Fig. 1(c), the FFS fillings exhibit a non-nested
FFS with three-fold rotational symmetry.
B. Electron-phonon coupling
To describe phonons, we introduce displacement fields:
u` (r) for each layer ` = 1, 2, which are conveniently re-
expressed in terms of the mean-displacement of the bi-
layer: ū = 12 (u1 + u2 ) and relative displacements of the
layers, d = u1 − u2 . Both types of phonon displacements
on the electronic Hamiltonian can be accounted by dis-
placing the tunneling operators T (r) → T (r −u) by with
a single effective displacement field [17, 28, 29]:
ẑ × d
u = ū − , (5)
2 tan θ/2
since the relative sliding of the two graphene sheets
is equivalent to a translation of the moiré pattern per-
pendicular to the sliding direction. This has two im-
portant consequences: First, at small twist angles, the
coupling to d is enhanced by a factor of ≈ θ−1 [29]. Sec-
ond, for a commensurate crystal relative displacements
would be gapped optical modes. Instead, for incommen-
surate twist angles uniform d-displacements have no en-
ergy cost [30, 31] and result in gapless acoustic modes (in
k + λkk4 (4)
analogy to the sliding “phason” mode of incommensurate
charge density wave orders [32]).
The electron-phonon coupling is conveniently identi-
fied by changing coordinates to a co-moving frame of
is restored to its undistorted form. To properly account
for the non-commutivity of magnetic translations, while
manifestly preserving gauge invariance, we implement
this frame transformation with the unitary operator:
Wu = e−iu(R)·Π (6)
where R = r − `2B ẑ × Π is the guiding center coordi-
nate (whose components all commute with Π). Neglect-
ing subleading terms of order O(∇u, u2 ), this transfor-
mation effectively restores the moiré tunneling potential:
Wu† T (r − u(r)) Wu ≈ T (~r), while transforming the ki-
netic energy by: Wu† ΠWu ≈ Π − eB × u(R), yielding a
direct (gradient-free) electron-phonon interaction:
 
~ × σ̂ −θ/2 0
He-ph = ev~u · B ≡ ū · Γm + d · Γr (7)
0 σ θ/2
For future convenience, we define the electron phonon
vertices:
Γn,α,λ (k, q) = hn, k + q|ε̂λ (q) · Γα |n, ki (8)
 4

description is:

dωd2 k †
Z
S= ψ (iω − k )ψω,k
(2π)3 ω,k
X dΩd2 q
Z
ρ Ω2 + c2s q 2 |uλ (Ω, q)|2


+ 3
(2π)
λ
X Z dωd2 kdΩd2 q †
+ Γλ (k, q)uλ (Ω, q)ψω+Ω,k+q ψω,k
(2π)6
λ
(9)

where ψ, u are electron, phonon operators, and ρ is

the mass density for SLG. For the dispersion, we use
1 2
FIG. 3. Schematic showing crossover scales ELD , E∗ and k ≈ vF k⊥ + 2m kk away from the FFS patches and
ENFL . ∆µ indicates the deviation in the chemical potential 4
k ≈ vF k⊥ + λkk near the FFS patches. The dynamical
from FFS fillings.
critical exponents for the phonons (zb ) and the electrons
(zf ) that we find below are also patch-dependent. In
the following, we calculate these exponents individually
for the nth magneto-Bloch band, with wave-vector k state near and away from the FFS patches, and further pro-
|n, ki, where α ∈ {m, r} labels the phonon type, and vide energy scales for crossovers between these regimes.
λ labels longitudinal (L) or transverse (T) polarizations A self-consistent one-loop perturbative calculation gives
corresponding to polarization vector ε̂λ (q). In the follow- a singular Landau-damping (LD) form of the phonon self-
ing, we focus on the relative (interlayer-sliding) phonon energy:
mode since their coupling to electrons is stronger by a 
factor ≈ 1/θ compared to the layer-symmetric phonon γλ |Ω| , zb = 3
|qk |
mode. For notational simplicity we omit the band-index Π1-loop (Ω, q; zb ) = 2 |Ω| (10)
γλ q∗ |q3 | , zb = 5
n and the phonon type index α in subsequent expres- k

sions. Crucially, these vertices generically do not vanish

in the limit of zero momentum transfer [16] (q → 0), as where qk is the component of q parallel to the Fermi-
demonstrated numerically in Fig. 2. surface, and we define the LD-parameters:
|m| 2 
At asymptotically low-temperatures, this direct- γλ = Γ (k, q̂)q̂kFS (11)
coupling is expected to produce a complex non-Fermi πρvF λ
liquid state [13–16], The NFL properties cannot be re- The expressions are labeled by the resulting dynamical
liably calculated except in artificial limits [11, 12]. We critical exponent zb = 3, 5 describing the low wave-vector
will make predictions based only on leading order per- dependence of the phonon damping rate (∼ q zb ) at low-
turbative calculations, which reliably predict the onset of temperatures, with zb = 3 (Ω ∼ q 3 ) behavior arising far
the non-Fermi liquid behavior approached from higher- from the FFS fillings, and zb = 5 (Ω ∼ q 5 ) behavior
temperatures or energy-scales, but are merely suggestive occurring at the FFS. Close to, but not precisely at the
of the possible asymptotic behavior at low-temperatures. FFS fillings, there is an finite-temperature crossover (see
Fig. 3) between zb = 5 dynamics at intermediate scales
to zb = 3 at sufficiently low temperature scales where
the phonons “notice” the Fermi-surface curvature, with
crossover scale:
c2s q∗3
E∗ ≈ . (12)
γ
III. FRACTAL NON-FERMI LIQUID
Furthermore, at sufficiently high energies or tempera-
tures E, T  ELD , the phonon-damping becomes unim-
A. Landau damping portant, and the phonons become sharp (underdamped)
quasi-particles, where
(√
In the presence of a Fermi-surface, this direct coupling cs γ, zb = 3
ELD (zb ) ≈
1 (13)
causes phonons to decay into the electron-hole contin- 2 5 5 12
E∗ cs γ , zb = 5
uum, and become soft and overdamped. For simplicity
we model the phonon dispersion as isotropic with speed For temperatures above ELD , the phonons recover their
cs (∼ 104 m/s) [30, 33]. The effective action in a patch ordinary zb = 1 (Ω ∼ q) dynamics.

The various dynamical scaling crossovers are depicted
in Fig. 3, and summarized by:
 ENFL (zb ) = E∗
1, E > ELD
 (zb − 1) sin (π/zb ) mE∗
zb = 5, E∗ < E < ELD (14)

3, E < E , E
∗ LD
For our numerical estimates of ELD below, we give results
for the zb = 3 regime only, since the zb = 5 behavior
arises only in a verypnarrow window around the FFS
fillings where m > Γ1 πρv2λF cs .
The crossover scale, ELD , will play an important role in
our following discussion, and numerical results are shown
in Fig. 4 for representative parameters. ELD is generally
deeply sub-Kelvin except near FFS fillings where the van-
ishing Fermi-surface curvature results in divergent effec-
tive mass and Landau damping coefficient for select di-
rections along the Fermi-surface. This divergence will be
rounded in practice by disorder and higher-order terms in
the electron dispersion, as shown in Fig. 3. For θ = 0.8◦ ,
ELD /kB exceeds 1K over appreciable ranges of chemical
potential (±0.3meV) and density (±10−13 cm−2 ).
B. Anomalous Surface Acoustic Wave Attenuation
Signatures
This Landau-damped form can be probed by surface
acoustic wave (SAW) attenuation experiments. SAWs
can be injected and detected by piezoelectric contacts,
which we model as semi-infinite in the y-direction, and
separated by finite distance L in the x direction. We
extract the SAW attenuation length, ξ, by computing
the fixed-frequency phonon propagator between source
to detector: dyD(ω, x = L, y) ∼ e−L/ξ , and find (see
R
Appendix C 2):
(
2ω/γ ω  ELD
ξ = 2cs ELD
1/3 , (15)
ω  ELD
ELD ω
with numerical results shown in Fig. 4. The FFS points
near VH-fillings produce a singular suppression of SAW
propagation, providing a characteristic fingerprint of the
Landau damping in the doping-dependence of ξ.
C. Transport properties
Strong coupling to MH acoustic (MHA)-phonons is
expected to result in unconventional non-Fermi liquid
scaling for transport and thermodynamic properties with
temperature and frequency [1, 2, 6, 34–36]. A one-loop
analysis predicts that, below ELD , scattering by over-
damped phonons with dynamical critical exponent zb re-
sults in (retarded) electron self-energy
1 (zb −1)
ΣR
f (ω; zb ) = i (ENFL (zb ))
zb
|ω| zb
(16)
5
where
zb
q∗2

(17)
For asymptotically low energies (E  ENFL ≡
minzb =3,5 ENFL (zb )), the phonon-scattering self-energy
dominates over the bare electron energy, resulting in a
destruction of Fermi liquid quasi-particles. The electron
dynamical critical exponent gets modified from zf = 1
to zf = zbz−1b
. However, in TBG, this occurs only at
inaccessibly low temperatures (in Appendix C, we esti-
mate ENFL /kB ≤ 10−4 K over the range of parameters
explored in Fig. 1). Therefore, in the experimentally ac-
cessible regime, the system will be in a “precursor” NFL
state, where electronic quasi-particles are ailing but not
yet expired.
Non-Fermi liquid behavior is often characterized
through scaling of resistivity with temperature. The pre-
cursor NFL regime is divided into high-, intermediate-
, and low-temperature regimes by two important cross-
over scales: i) the Bloch-Gruneisen (BG) temperature,
TBG above which thermally activated phonons carry
average momentum that is larger than 2kF so that
phonon-contributions to resistivity are dominated by
large-momentum scattering by thermal excitations, and
ii) the Landau-damping temperature TLD = ELD /kB , be-
low which the phonons become overdamped and non-
Fermi liquid temperature dependence arises. Using
the Born-approximation to compute the transport time,
the asymptotic temperature dependence of the electron-
phonon contributions to resistivity in each regime are
summarized as follows:

T ;
 T > TBG
ρ(T ) ∼ T ; 2
TBG > T > TLD . (18)
T 4/3 ; T < T

LD
The phonon-mechanism for linear-T resistivity in the
high-temperature regime (T > TBG ) is a standard ef-
fect that does not rely on Moire-Hofstadter physics and
can be observed in the absence of a B-field [31, 37].
By contrast, the phonon-induced ρ(T ) ∼ T 2 behav-
ior in the intermediate regime arises from the singu-
lar coupling between electrons and MHA-phonons, and
crucially relies on the magnetic field producing non-
vanishing electron-phonon coupling vertex in the limit of
q → 0 (see Appendix C 3 for detailed calculation and nu-
merical results). This produces a strong enhancement of
the electron-phonon-scattering contribution to resistivity
compared to the conventional T 4 behavior [37] expected
in the absence of a magnetic field. We remark that a dif-
ferent T 2 -contribution to resistivity arises from electron-
electron interactions in a Fermi-liquid, and a quantita-
tive and perhaps model-dependent comparison will be
required to distinguish the origin of this ρ(T ) ∼ T 2 scal-
ing. For example electron-electron interaction contribu-
tions to resistivity may be neglected in very clean samples
when Umklapp processes required to relax momentum
 6

FIG. 4. Landau damping energy and SAW attenuation length – (a), (b) Landau damping scale of relative-displacement
phonon ELD,r along path K − Γ − K 0 for twist angle θ = 1.5◦ , θ = 1.1◦ and flux filling ν = 1/3, ν = 1/2. (c) ELD,r along path
M − Γ − M 0 for twist angle θ = 0.8◦ and flux filling ν = 1. (d),(e) SAW attenuation length of relative-displacement phonon
ξr along path K − Γ − K 0 , for twist angle θ = 1.5◦ , θ = 1.1◦ and flux filling ν = 1/3, ν = 1/2 and typical driving frequency
ω/2π = 200MHz ∼ 0.96 × 10−2 K. (f) ξr along path M − Γ − M 0 , for twist angle θ = 0.8◦ , filling ν = 1 and driving frequency
ω/2π = 200MHz

are strongly suppressed by Fermi-surface geometry and D. MHA Polarons

momentum conservation requirements. Furthermore, as
we elaborate in the next section, the enhanced electron-
MHA-phonon scattering can have a an impact on the
scaling of quantum oscillatory and cyclotron resonance
phenomena that is qualitatively different than standard
electron-electron or electron-phonon behavior.
Finally, in the low temperature regimes for T <
TLD , scattering from overdamped phonons results in
departures from Fermi liquid behavior, whose proper-
ties can only be computed in artificial limits [11, 12].
For example, a potentially oversimplified scattering-rate
calculation (neglecting hydrodynamic effects and umk-
lapp scattering) predicts temperature-dependent resis-
tivity ρ(T ) ∼ T 4/3 for T < ELD [1, 38]. We note
that while flat-patches experience a larger scattering rate
−1
τtr,FFS ∼ T 8/5 , their contribution to transport is likely
shorted-out by the coexisting curved patches that have
lower ∼ T 4/3 scattering. Other signatures include spa-
tial decay of quasi-particle interference patterns, which
expose the anomalous dimension of the electrons in the
NFL state [4, 11].
Even for energy and temperature scales above ELD ,
electrons and MHA-phonons are strongly coupled into
unusual polaronic degrees of freedom, which we dub
“MHA-polarons”. The dynamics of MHA polarons dif-
fers dramatically from their counter-parts in ordinary
metals, since accelerating MHA polaron motion results
in radiation of soft-collinear phonons in direct analogy
to cyclotron radiation. In the intermediate-temperature
regime, TBG > T > TLD , a tree-level scattering rate
calculation using the undamped phonon propagator (see
Appendix C 3) shows that the rate of phonon emission
is logarithmically divergent in the infrared (though the
energy emitted is finite), in analogy to soft collinear
divergences in quantum electrodynamics. We analyze
the MHA-polaron motion and phonon-radiation within a
semiclassical framework (see Appendix refapp:polarons),
which amounts to an infinite-order resummation of soft
collinear divergences in the eikonal approximation [39].
To physically probe the dynamics of the MHA-polaron,
we consider detuning the magnetic field slightly away
from commensurate filling B = Bν + ∆B, with e∆B 
1/|~a1,2 |2 . Without phonons, electrons would exhibit cy-
clotron motion with effective cyclotron frequency, ωc,0 =
e∆B
m set by the detuning field ∆B rather than the full
field B (as can be seen by semiclassical motion of a wave-
 7

packet made from Bloch states of the bands at commen- The cyclotron motion of MHA-polarons differs
surate filling Bν moving in effective field, ∆B). Oscilla- markedly from that of bare electrons as we now
tions with the reduced field ∆B, rather than the full field show by solving the polaron equation of motion in
B could be measured by standard Shubnikov-de-Haas, Eq. (21) for damped oscillatory cyclotron motion r(t) =
cyclotron resonance, or current-focusing [40] techniques, r0 e−Γc t (cos(ωc t)x̂ + sin(ωc t)ŷ), in an excess field ∆B.
but to our knowledge have not yet been explored. Defining an effective frequency dependent mass mp (ωc ) ≡
eB/ωc we find that the MHA-polaron exhibits a
scale-dependent  logarithmically diverging effective mass:
1. Cyclotron motion 2 Λ
mp (ωc ) = m 1 + g log ωc . The cyclotron frequency
scales with an unconventional power of ∆B:
Numerical computation reveals the electron-phonon
coupling is proportional to the band-velocity, so that the ω g 2
c,0 1+g 2
phonon fields couple to the electron velocity, ṙ, where r is ωc = ωc,0 ∼ (∆B) (22)
Λ
the electron coordinate. This velocity coupling is directly
analogous to the coupling between charged-particles and and a ∆B independent decay rate Γc = g 2 Λ. Here
gauge fields. Incorporating this feature into a Schwinger- ωc,0 = e∆Bm is the bare (non-interacting) electron cy-
Keldysh path integral description of a single-electron (or clotron frequency of the MH-bands. For TBG we find
hole) coupled to a zero-temperature
R bath of phonons, we that the g 2 . 10−2 (even close to the VH-fillings), so
obtain an effective action C dtL[r, u] on the closed-time these effects may be challenging to observe. We note,
Keldysh contour C with the Lagrangian: however, that these predictions equally apply to a vari-
ety of analog non-Fermi liquid systems [4, 6, 13–15] where
L[r, u] = Le [r] + Le−ph [r, u(r)] + Lph [u] the polaron coupling constant would not be suppressed
1 by the electron-ion mass ratio and could potentially give
Le [r] = mṙ 2 − eṙ · A(r), appreciable modifications to semiclassical electron mo-
2h
ρ 2 2
i tion.
Lph [u] = (∂t u) − c2s (∇u) ,
2
Le−ph [r, u(r)] = αṙ · u(r). (19)
2. Quantum oscillations
Here, the vector potential A(r) corresponds to the excess
magnetic field (∆B = ∇ × A(r)), and does not include In addition to energy dissipation by phonon radiation,
the constant field used to reach the MH-regime, and α ≡ MHA-phonons cause characteristic dephasing of phase-
m ∂Γ(k,q=0) . sensitive measurements such as periodic-in-1/∆B quan-
∂k
We now integrate out the phonon field by approximat- tum oscillations in density of states and resistivity. As
ing the change of the electron position to be slow com- electrons experience MHA-phonon fields u as an effec-
pared to the wavelengths of the phonons being integrated tive electromagnetic vector potential, zero-point fluctua-
out [41], which leads to a quadratic, non-local in time ef- tions of MHA-phonons give rise to quantum-fluctuating
R
fective action for the MHA polaron coordinate r: geometric (Aharonov-Bohm-like) phase, eiα dt ṙ·u(r(t)) .
Quantum fluctuations in the phonon fields results in
α2
Z ∞ Z ∞
r q (t) · ṙ cl (t0 ) random geometric-phase accumulation that suppress the
Seff = S0 − dt dt0 2 (20) quantum oscillation amplitude.
2πρc2s −∞ t (t − t0 ) We compute the quantum oscillatory contribution to
density of states,N (ε), at energy ε via a semiclassical sum
where rcl/q are the so-called “classical” and “quantum” of the return amplitude for multiple classical cyclotron
combinations of r(t) on the two-time Keldysh contour, orbits of the MHA-polaron, averaged over fluctuating ge-
and S0 is the bare action for an electron in the absence ometric phases due to MHA-phonons [44]:
of phonon degrees of freedom. Varying this effective ac-
tion [42], we find the following semiclassical equation of 1 t
Z
motion (EOM) for r (see Appendix D 1): N (ε) ≈ − dt Im GR (r = r0 ; t)e−iεt
π 0
2πε
X R
≈h e−i ωc n eiα dt ṙ·u(r(t)) iu (23)
t
ṙ(t0 )
Z
F ext 0 n
r̈(t) = − g2 dt (21)
m −∞ (t − t0 )2
where h. . .iu indicates an average over quantum fluctua-
q
m ∂Γ(k,q=0) tions of u. We restrict our attention to zero temperature
where g = 2πρc2s ∂k is a dimensionless measure of here, as finite temperature effects with T  ωc do not
electron-phonon coupling strength, and F ext represents effect the scaling form that we identify. These fluctu-
the external force due to the excess magnetic field [43]. ations are approximately quenched on the time-scale of
The second term represents the non-Markovian effects the cyclotron motion since the electrons are fast com-
due to the gapless phonons. pared to the phonons, i.e. we may approximately replace

R
dt ṙ · u(r(t)) by Φ(u), the flux of a static effective mag-
netic field b = ∇ × u through the cyclotron orbit. Within
this approximation, applying the Poisson summation for-
mula (see Appendix D 2) yields.
X 2πε 1 2
n2 hΦΦi
N (ε) = e−i ωc n e−i 2 α
n the NFL physics will compete (or intertwine) with en-
 
X 1 2 hanced tendency to order with a nested Fermi surface.
= Ac exp − 2 (ε − nωc,0 ) (24)
n
2Γ
The resulting oscillator contribution to the density of
states takes the form of a comb of Gaussian peaks cen-
tered at integer multiples of the cyclotron frequency ωc,0
2g 2
with peak-amplitude Ac ∼ |∆B| and
√ width Γ ∼ (∆B) .
This contrasts the usual Ac ∼ ∆B scaling expected
from impurity scattering, providing a means to distin-
guish these two mechanisms.
[1] Patrick A. Lee, “Gauge field, aharonov-bohm flux, and
high-Tc superconductivity,” Phys. Rev. Lett. 63, 680
(1989).
[2] P. A. Lee and N. Nagaosa, “Gauge theory of the normal
state of high-t c superconductors,” Physical Review B
46, 5621 (1992).
[3] Patrick A. Lee, Naoto Nagaosa, and Xiao-Gang Wen,
“Doping a mott insulator: Physics of high-temperature
superconductivity,” Rev. Mod. Phys. 78, 17 (2006).
[4] Max A. Metlitski and Subir Sachdev, “Quantum phase
transitions of metals in two spatial dimensions. i. ising-
nematic order,” Phys. Rev. B 82, 075127 (2010).
[5] Max A. Metlitski and Subir Sachdev, “Quantum phase
transitions of metals in two spatial dimensions. ii. spin
density wave order,” Phys. Rev. B 82, 075128 (2010).
[6] B. I. Halperin, Patrick A. Lee, and Nicholas Read, “The-
ory of the half-filled landau level,” Phys. Rev. B 47, 7312
(1993).
[7] S.-S. Lee and P. A. Lee, “U(1) gauge theory of the hub-
bard model: Spin liquid states and possible application
to κ−(BEDT−TTF)2 cu2 (CN)3 ,” Physical Review Let-
ters 95, 036403 (2005).
[8] T. Senthil, “Theory of a continuous mott transition in
two dimensions,” Phys. Rev. B 78, 045109 (2008).
[9] Sung-Sik Lee, “Stability of the u(1) spin liquid with a
spinon fermi surface in 2 + 1 dimensions,” Phys. Rev. B
78, 085129 (2008).
[10] Sung-Sik Lee, “Low-energy effective theory of fermi sur-
face coupled with u(1) gauge field in 2 + 1 dimensions,”
8
IV. OUTLOOK
We have explored moiré super-lattice structures in high
magnetic fields as potential platforms for simulating non-
Fermi liquid (NFL) physics, due to unusually singular
electron-phonon coupling. While the asymptotic low-
temperature NFL fixed point remains at inaccessibly low
temperature, intermediate scale non-Fermi liquid precur-
sor behavior and unconventional polaron dynamics can
be observed at by tuning carrier density near a van-Hove
singularity. At or very close to the van Hove filling,
This competition appears to be controllable by twist-
engineering [45–50], and a more detailed study of the
interplay between exotic interaction-driven orders, and
non-Fermi liquid physics in these systems will be a com-
pelling subject for future investigation.
Acknowledgements – We thank Allan Macdonald, Elaine
Li, Sid Parameswaran, Brad Ramshaw, Isabelle Phinney,
Igor Blinov and Naichao Hu for insightful discussions.
This work was supported by NSF DMR-1653007 (AP),
and by the NSF through the Center for Dynamics and
Control of Materials: an NSF MRSEC under Coopera-
tive Agreement No. DMR-1720595 (AK). Part of this
work was performed at the Aspen Center for Physics,
which is supported by NSF grant PHY-1607611.
Phys. Rev. B 80, 165102 (2009).
[11] David F. Mross, John McGreevy, Hong Liu, and
T. Senthil, “Controlled expansion for certain non-fermi-
liquid metals,” Phys. Rev. B 82, 045121 (2010).
[12] D. Dalidovich and S.-S. Lee, “Perturbative non-fermi liq-
uids from dimensional regularization,” Physical Review
B 88, 245106 (2013).
[13] Vadim Oganesyan, Steven A. Kivelson, and Eduardo
Fradkin, “Quantum theory of a nematic fermi fluid,”
Phys. Rev. B 64, 195109 (2001).
[14] Cenke Xu, “Quantum critical points of helical fermi liq-
uids,” Phys. Rev. B 81, 054403 (2010).
[15] Yasaman Bahri and Andrew C. Potter, “Stable non-
fermi-liquid phase of itinerant spin-orbit coupled ferro-
magnets,” Phys. Rev. B 92, 035131 (2015).
[16] H. Watanabe and A. Vishwanath, “Criterion for stability
of goldstone modes and fermi liquid behavior in a metal
with broken symmetry,” Proc Natl Acad Sci U S A 111,
16314 (2014).
[17] R. Bistritzer and A. H. MacDonald, “Moire bands in
twisted double-layer graphene,” Proc Natl Acad Sci U
S A 108, 12233 (2011).
[18] R. Bistritzer and A. H. MacDonald, “Moiré butterflies
in twisted bilayer graphene,” Phys. Rev. B 84, 035440
(2011).
[19] Kasra Hejazi, Chunxiao Liu, and Leon Balents, “Lan-
dau levels in twisted bilayer graphene and semiclassical
orbits,” Phys. Rev. B 100, 035115 (2019).
[20] Mikito Koshino, Noah F. Q. Yuan, Takashi Koretsune,

Masayuki Ochi, Kazuhiko Kuroki, and Liang Fu, “Max-
imally localized wannier orbitals and the extended hub-
bard model for twisted bilayer graphene,” Phys. Rev. X
8, 031087 (2018).
[21] C. R. Dean, L. Wang, P. Maher, C. Forsythe, F. Ghahari,
Y. Gao, J. Katoch, M. Ishigami, P. Moon, M. Koshino,
T. Taniguchi, K. Watanabe, K. L. Shepard, J. Hone,
and P. Kim, “Hofstadter’s butterfly and the fractal quan-
tum hall effect in moire superlattices,” Nature 497, 598
(2013).
[22] B. Hunt, J. D. Sanchez-Yamagishi, A. F. Young,
M. Yankowitz, B. J. LeRoy, K. Watanabe, T. Taniguchi,
P. Moon, M. Koshino, P. Jarillo-Herrero, and R. C.
Ashoori, “Massive dirac fermions and hofstadter butter-
fly in a van der waals heterostructure,” Science 340, 1427
(2013).
[23] Yuan Cao, Valla Fatemi, Shiang Fang, Kenji Watanabe,
Takashi Taniguchi, Efthimios Kaxiras, and Pablo Jarillo-
Herrero, “Unconventional superconductivity in magic-
angle graphene superlattices,” Nature 556, 43 (2018).
[24] Yuan Cao, Valla Fatemi, Ahmet Demir, Shiang
Fang, Spencer L Tomarken, Jason Y Luo, Javier D
Sanchez-Yamagishi, Kenji Watanabe, Takashi Taniguchi,
Efthimios Kaxiras, et al., “Correlated insulator be-
haviour at half-filling in magic-angle graphene superlat-
tices,” Nature 556, 80 (2018).
[25] Xiaobo Lu, Petr Stepanov, Wei Yang, Ming Xie, Mo-
hammed Ali Aamir, Ipsita Das, Carles Urgell, Kenji
Watanabe, Takashi Taniguchi, Guangyu Zhang, et al.,
“Superconductors, orbital magnets and correlated states
in magic-angle bilayer graphene,” Nature 574, 653
(2019).
[26] The non-Fermi liquid effects that we find are enhanced
near the VH fillings, which can be tuned to generic non-
integer fillings via an in-plane (Zeeman) field.
[27] Rahul Nandkishore and Andrey V. Chubukov, “Inter-
play of superconductivity and spin-density-wave order in
doped graphene,” Phys. Rev. B 86, 115426 (2012).
[28] Leon Balents, “General continuum model for twisted bi-
layer graphene and arbitrary smooth deformations,” Sci-
Post Phys. 7, 48 (2019).
[29] B. Lian, Z. Wang, and B. A. Bernevig, “Twisted bilayer
graphene: A phonon-driven superconductor,” Phys Rev
Lett 122, 257002 (2019).
[30] Mikito Koshino and Young-Woo Son, “Moiré phonons
in twisted bilayer graphene,” Phys. Rev. B 100, 075416
(2019).
[31] Fengcheng Wu, Euyheon Hwang, and Sankar Das Sarma,
“Phonon-induced giant linear-in-t resistivity in magic
angle twisted bilayer graphene: Ordinary strangeness
and exotic superconductivity,” Phys. Rev. B 99, 165112
(2019).
[32] A. W. Overhauser, “Observability of charge-density
waves by neutron diffraction,” Physical Review B 3, 3173
(1971).
[33] Alexandr I. Cocemasov, Denis L. Nika, and Alexander A.
9
Balandin, “Phonons in twisted bilayer graphene,” Phys.
Rev. B 88, 035428 (2013).
[34] Joseph Polchinski, “Low-energy dynamics of the spinon-
gauge system,” Nuclear Physics B 422, 617 (1994).
[35] C. Nayak and F. Wilczek, “Non-fermi liquid fixed point
in 2+ 1 dimensions,” Nuclear Physics B 417, 359 (1994).
[36] B. L. Altshuler, L. B. Ioffe, and A. J. Millis, “Low-energy
properties of fermions with singular interactions,” Phys-
ical Review B 50, 14048 (1994).
[37] E. H. Hwang and S. Das Sarma, “Acoustic phonon scat-
tering limited carrier mobility in two-dimensional extrin-
sic graphene,” Phys. Rev. B 77, 115449 (2008).
[38] L. B. Ioffe and P. B. Wiegmann, “Linear temperature de-
pendence of resistivity as evidence of gauge interaction,”
Phys Rev Lett 65, 653 (1990).
[39] E. Laenen, G. Stavenga, and C. D. White, “Path integral
approach to eikonal and next-to-eikonal exponentiation,”
Journal of High Energy Physics 2009, 054 (2009).
[40] T. Taychatanapat, K. Watanabe, T. Taniguchi, and
P. Jarillo-Herrero, “Electrically tunable transverse mag-
netic focusing in graphene,” Nature Physics 9, 225
(2013).
[41] M. P. A. Fisher and W. Zwerger, “Ground-state sym-
metry of a generalized polaron,” Phys. Rev. B 34, 5912
(1986).
[42] A. Kamenev, Field theory of non-equilibrium systems
(Cambridge University Press, 2011).
[43] At non-zero temperature an additional stochastic force
would arise from thermal phonon fluctuations.
[44] A. D. Mirlin, E. Altshuler, and P. Wölfle, “Quasiclassical
approach to impurity effect on magnetooscillations in 2d
metals,” Annalen der Physik 508, 281 (1996).
[45] Yi-Ting Hsu, Fengcheng Wu, and S. Das Sarma, “Topo-
logical superconductivity, ferromagnetism, and valley-
polarized phases in moiré systems: Renormalization
group analysis for twisted double bilayer graphene,”
Phys. Rev. B 102, 085103 (2020).
[46] Y.-P. Lin and R. M. Nandkishore, “Chiral twist on the
high-Tc phase diagram in moiré heterostructures,” Phys.
Rev. B 100, 085136 (2019).
[47] L. Classen, C. Honerkamp, and M. M. Scherer, “Com-
peting phases of interacting electrons on triangular lat-
tices in moiré heterostructures,” Phys. Rev. B 99, 195120
(2019).
[48] Dmitry V. Chichinadze, Laura Classen, and Andrey V.
Chubukov, “Nematic superconductivity in twisted bi-
layer graphene,” Phys. Rev. B 101, 224513 (2020).
[49] H. Isobe, N. F. Q. Yuan, and L. Fu, “Unconventional
superconductivity and density waves in twisted bilayer
graphene,” Phys. Rev. X 8, 041041 (2018).
[50] N. F. Q. Yuan, H. Isobe, and L. Fu, “Magic of high-
order van hove singularity,” Nature Communications 10,
1 (2019).
[51] Y. B. Kim, P. A. Lee, and X.-G. Wen, “Quantum boltz-
mann equation of composite fermions interacting with a
gauge field,” Physical Review B 52, 17275 (1995).
 10

Appendix A: Moire-Hofstadter bands

In this section, we review the derivation of the magnetic Bloch bands for twisted bilayer graphene. Define the
magnetic lattice translational operators:

Ti = eiK·ai , (A1)
 √  √ 
a 3 1 a 3 1
where i = 1, 2, which perform translation by Moire lattice-vectors: a1 = θ − 2 , 2 and a2 = θ 2 , 2 . Further,
define guiding-center momenta:
y x
Kx = Πx + 2 , Ky = Πy − 2 , (A2)
lB lB

which satisfy [Kx , Ky ] = l2i , [Kα , Πβ ] = 0, and [Ti , H] = 0.

B
Unlike ordinary lattice-translations, magnetic translations do not generically commute except at special commen-
surate magnetic fields. Generically,
√ 2!
3a
T1 T2 = T2 T1 exp i 2 2 , (A3)
2θ lB

which vanishes if and only if:

√
3a2 p 4π q
2 2 = 2π or kθ2 lB
2
= √ , (A4)
2θ lB q 3 3p
which gives the condition for commensuration of the Moire lattice and magnetic field.
At commensurate filling, [T1 , T2q ] = 0, so that we can construct the magnetic Bloch states |αki with band index α,
as the simultaneous eigenstates of H, T1 , and T2q :

H|αki = Eα (k)|αki, (A5)

ik·a1
T1 |αki = e |αki, (A6)
T2q |αki =e iqk·a2
|αki. (A7)

An extended magnetic Brillouin zone (MBZ) is spanned by magnetic-reciprocal lattice vectors: g 1 and g 2 /q (where
g i are the original Moire reciprocal lattice vectors). As we now show, the commensurability integers (p, q), determine
the number of sub-bands, and size of the reduced MBZ. The commutation relations for T1 and T2 give:

T1 T2j |αki = ei(k+g1 jp/q)·a1 T2j |αki (A8)

implying a set of degenerate states T2j |αki ∼ |α, k + jp q g 1 i (j = 1, 2, . . . , q − 1), which have identical energy since
[H, T (a2 )] = 0. Thus there is a q-fold degeneracy along the g 1 direction. Equivalently, we can consider a reduced
magnetic
 Brillouin
 zone spanned by g 1 /q and g 2 /q with q-fold degeneracy. In addition, Eq.(A8) implies: Eα (k) =
Eα k + pq g 1 , that is, one period in a magnetic band extends over p MBZ’s in the direction of g 1 . The reduction of
energy bands to one MBZ will yield p different subbands. For convenience, we extend Eα (k) to Eα (k + lg 1 ) ≡ Eαl (k)
where l = 0, 1, . . . , p − 1 represents subbands and k is restricted to the reduced MBZ.
Having identified the distinct magnetic sub-bands and reduced MBZ, we can label states of TBG by: |τ, σ, n, l, ki,
where τ = 1, 2 represents layers, σ = A, B represents sublattices, n represents LL index, and l represents sub-
bands. For completeness we write the wavefunction of the LL states in the Landau gauge: hr|n, ki ∼ eiky y Hn (x +
2 2
2
k y lB )e−(x+ky lB )/2lB . However, the expressions obtained below are general and do not rely on fixing a particular
gauge. Under this basis, the single-layer Hamiltonian reads
√ 
2v iθ/2 √ 
h(θ/2) = e n + 1|2, B, n + 1, l, kih2, A, n, l, k| + h.c. (A9)
lB

To calculate matrix elements of the interlayer hopping term under this basis, we first split e−igj ·r as

e−igj ·r = e−igj ·η e−igj ·R . (A10)

 11

where
   2
  
Rx x − Πy lB x − ηx
R= = 2 ≡ =r−η (A11)
Ry y + Π x lB y − ηy
2 2
Notice that R = lB (ẑ × K) only acts on |l, ki and η = −lB (ẑ × Π) only acts on |ni, so we have

hn, l, k|e−igj ·r |n0 , l0 , ki = hn|e−igj ·η |n0 ihl, k|e−igj ·R |l0 , ki. (A12)
√

The first term on the right hand side of Eq.(A12) is given by hn|e−igj ·η |n0 i = Fnn0 g j lB / 2 , where
r 0
 n ! (−zx + izy )n−n0 e− z2 Ln−n
2 0
(z 2 ) n > n0

n0
n!

Fnn0 (z) = r (A13)
 n! (z + iz )n0 −n e− z2 Ln0 −n (z 2 ) n < n0

 2

0 x y n
n!
with L being the associated Laguerre polynomial.
The second term on the right hand side of Eq.(A12) can be converted to matrix elements of magnetic translation
2
operators by using e−igj ·R = e−ilB K·(gj ×ẑ) . Specifically:
q q
−p
e−ig1 ·R = T2 and e−ig2 ·R = T1p . (A14)
q
−p
Considering T2 |αki ∼ |α, k − g 1 i and applying the commutation relation between T1 and T2 , we can obtain
q
−p q
T2 |l + 1, ki = e−i p k·a2 |l, ki. (A15)
q q
p ip (k·a1 +2πl)
T1 |l, ki = e |l, ki (A16)

Hence, the interlayer hopping term in the MBZ reads

T (k) = wT0 |2, σ, n, l, kih1, σ 0 , n0 , l, k|

 √  q
+ wT1 Fnn0 g 1 lB / 2 e−i p k·a2 |2, σ, n, l + 1, kih1, σ 0 , n0 , l, k|
 √  q
+ wT2 Fnn0 g 2 lB / 2 ei p (k·a1 +2πl) |2, σ, n, l, kih1, σ 0 , n0 , l, k|. (A17)

The numerical plots shown in the main text were obtained by diagonalizing this inter-layer hopping matrix truncated
to a sufficiently large range of Landau indices n to achieve well-converged cutoff-independent results. Notice that the
direct truncation in LL basis will introduce √
an artifical zero mode in Eq.(A9). To remove it, we add a penalty term
2Ncutoff v
EP |2, A, Ncutoff ih2, A, Ncutoff | with EP  lB to Eq.(A9) in the numerical calculation.

Appendix B: MHA electron-phonon vertex

In this section, we derive the MHA electron-phonon coupling. For the convenience, we can write Hamiltonian under
the lattice distortion Hu as Hu = H0 (Π) + Ht (r − u(r)), where
 
h(−θ/2) 0
H0 (Π) = (B1)
0 h(θ/2)

and
 
0 T (r − u(r))
Ht (r − u(r)) = . (B2)
T † (r − u(r)) 0

Then we implement the co-moving frame transformation with Wu = e−iu(R)·Π on H0 (Π) and Ht (r−u(r)) respectively.
Since [R, Π] = 0, for small u we have

Wu† Πi Wu = Πi + i[u(R) · Π, Πi ] = Πi − eBεij uj (R) + O(u2 ) (B3)

 12

and corresponding transformed H0 (Π):

Wu† H0 (Π)Wu = H0 (Π − eB × u(R)) (B4)
 
B × σ −θ/2 0
= H0 (Π)+veu · + O(u2 ) (B5)
0 B × σ θ/2
where σ θ/2 ≡ Rθ/2 σ are rotated Pauli matrices.
Similarly, for the tunneling term, we have
Wu† T (r − u(r))Wu = T (r) + O(∇u, u2 ) (B6)
and therefore the transformed total Hamiltonian is
 
B × σ −θ/2 0
Wu† Hu Wu = Hu=0 + veu · + O(∇u, u2 ). (B7)
0 B × σ θ/2
which gives electron-phonon coupling for both the mean displacement and the relative displacement in the transformed
coordinate:
He−ph = ū · Γm + d · Γr (B8)
with
 
ẑ × σ −θ/2 0
Γm = veB (B9)
0 ẑ × σ θ/2
and
 
veB σ −θ/2 0
Γr = − (B10)
2 tan θ2 0 σ θ/2

Appendix C: Computation of non-Fermi liquid scales

In this section, we estimate the relevant energy scales for the onset of Landau damping (zb = 3) for the phonons,
and non-Fermi liquid behavior of electrons, using self-consistent one-loop propagators (random phase approximation).
While uncontrolled at asympotitically low-temperatures in the non-Fermi liquid regime, this approximation produces
a reliable estimate of the onset for non-Fermi liquid behavior approaching from the high-temperature perturbative
regime. Unless otherwise specified, all propagators are specified in imaginary time (Matsubara frequency).

1. One-loop self-energies and propagators

The (imaginary time/Matsubara) bare phonon propagator and bare electron propagator are given respectively by
D0−1 (q, Ω) = −ρλ Ω2 + ωq2


(C1)

G−1
0 (k, ω) = iω − εk (C2)
where for the long wave limit the phonon dispersion is ωq ≈ cs |q| and εk is the electron dispersion and λ = m/r is the
label for mean and relative displacement phonon modes respectively.
With electron-phonon interaction, the full phonon propagator depends on the phonon self-energy as D−1 = D0−1 −Π.
At the lowest order of electron-phonon interaction, the one-loop phonon self-energy can be calculated by
Z 2
d kdω
Π(q, Ω) = G0 (k, ω)G0 (k + q, ω + Ω)Γ2α,λ (k, q) (C3)
(2π)3
where α = L/T is the label for longitudinal and transverse mode, ρm = 2ρ, ρr = ρ2 (ρ is the mass density of single
layer graphene). At the limit q → 0, only the nonvanished coupling Γ2α,λ (k, q̂) is left. After performing the integrals
2
kk
for the patch dispersion εk = vF k⊥ + 2m , we can obtain
|Ω|
Π(q, Ω) = γα,λ (C4)
|qk |
 13

where qk is the component parallel to the fermi surface and the Landau damping coefficient γα,λ is given by

|m|
Γ2α,λ (k, q̂)q̂kFS

γα,λ (q̂) = (C5)
2πρλ vF
where m is the effective mass of electron. Then, the full phonon propagator reads
1
D(q, Ω) = −  . (C6)
ρλ Ω2 + ωq2 + γα,λ |q|Ω|
k|

Using Ω ∼ cs q, the energy scale below which Landau damped modes dominate is
s
√ cs |m| 
ELD,α,λ ∼ γα,λ cs = Γ2α,λ (k, q̂) (C7)

2πρλ vF q̂kFS

On the other hand, using the Landau damped phonon propagator

1
D(q, Ω) ≈ −  , (C8)
ρλ ωq2 + γα,λ |q|Ω|
k|

we can obtain the one-loop electron self-energy at the Fermi level defined by G−1 = G−1 0 −Σ
Z 2
d qdΩ
Σ(ω) = D(q, Ω)G0 (q + kF , Ω + ω)Γ2α,λ (k, q̂)
(2π)3
! 13
Γ4α,λ 2
=i √ 2 2
|ω| 3 sgn(ω) (C9)
2 4
12 3π ρλ vF |m|cs

which gives the Non-Fermi liquid energy scale

Γ4α,λ
ENFL ∼ √ (C10)
12 3π 2 ρ2λ vF2 |m|c4s
2
below which the quasiparticle lifetime goes as τ −1 ∝ ω 3 .

2. SAW attenuation length

In this section, we give the propagation intensity of a surface acoustic wave on twisted bilayer graphene, which
is attenuated by the electron-phonon scattering and decaying as e−r/ξ . Here we consider the surface acoustic wave
generated by a line source and the phonon propagator in the direction perpendicular to the source is given by, the
retarded phonon propagator:
Z
R 1 1
D (r, q̂, ω) = dqeiqr 2 ω (C11)
ρλ ω − c2S q 2 − iγ(q̂) |q|

(where we have analytically continued from Matsubara to retarded frequency: iΩ → ω + i0+ ).

For further convenience, we first define dimensionless quantities: x ≡ qcS /ω, y ≡ rω/cS , and λ ≡ ELD /ω and then
the integral becomes
Z ∞
2i x sin(xy)
DR = dx
ρλ cs ω 0 x − x3 − iλ2
Z ∞ Z ∞
2i (x2 − x4 ) sin(xy) 2 λ2 x sin(xy)
= dx 3 2 4
− dx (C12)
ρλ cs ω 0 (x − x ) + λ ρ λ cs ω 0 (x − x3 )2 + λ4
We focus on the first term I1 which is the imaginary part of D and dictates the attenuation of SAWs
Z ∞
i (x2 − x4 )eixy
I1 = Im dx (C13)
ρ λ cs ω −∞ (x − x3 )2 + λ4
 14

Note that for larger imaginary part of poles, the decaying rate of the integral is larger, thus we want to find the pole
with the smallest positive imaginary part which dominates the attenuation. Solving (x − x3 )2 + λ4 = 0 gives 6 roots:
β ± iα, −β ± iα and ±i2α, where
1 2 1 1 1 2 5
23 η3 − 2 · 33 23 32 η 3 + 2 · 36
α= 2 1 , β= 2 1 (C14)
2·6 η3 3 2·6 η 3 3

√
with η = 12 + 81λ4 + 9λ2 . Hence the smallest positive imaginary part of poles is α and correspondingly the typical
decaying length ξ is given by
r cS λ
ξ= = (C15)
yα ELD α
λ2
When energy scale of acoustic wave is much larger than Landau damping scale, i.e., λ  1, α = 2 + O(λ6 ), thus

2cS 2ω
ξ≈ = (C16)
ELD λ γ
2
 2
On the other hand, for energy scale much smaller than Landau damping scale, i.e., λ  1, α = 21 λ 3 + O λ− 3 , thus

  13
2cS ELD
ξ≈ (C17)
ELD ω

3. Perturbative electron lifetime in intermediate-temperature regime

To compute the electronic life-time for temperatures or frequencies above the Landau-damping scale ELD , we
compute the tree-level electron-phonon quantum and transport scattering rates using the bare (undamped) phonon
propagator, and focus on the relative (interlayer-sliding) phonons which have two orders of magnitude stronger
interactions with electrons.
We work in a patch description of the Fermi-surface, which captures the universal aspects of the leading-order
singularities in the electron-phonon scattering rate, while neglecting smooth, non-singular “background” scattering
processes. We consider an electron at initial momentum k = kx̂ above the Fermi surface in a particular (x) direction
(i.e. k is measured relative to the Fermi-momentum kF x̂). We linearize the electron dispersion near the Fermi energy,
and neglect the curvature of the Fermi-surface, εp ≈ vF px . In this computation, the electron dispersions dominates
the energetics in the x direction parallel to the Fermi-velocity, and the phonon dispersion controls the energetics in
the y (perpendicular) direction, allowing us to neglect the dispersion of the phonons along x: ωq ≈ cs |qy | (formally,
this is justified for cs  vF ). For tree-level scattering rates, we may consider the Fermi-sea as Pauli-blocked, since
at this order in perturbation theory, additional virtual electron-hole pairs cannot be excited. These approximations
capture the leading singular behavior of scattering rates in the perturbative intermediate-temperature scale regime.
Writing the relative layer displacement field d in terms of annihilation operators aλ,q that destroy a phonon with
polarization λ and wave-vector q:
X 1
λ,q aλ,q eiq·r + h.c. ,


d= p (C18)
λ,q
2ρλ cs |q|

In the above-described setup, the electron couples only to a single polarization of the phonons, so we may drop the
polarization indices, and write the corresponding electron-phonon coupling interaction coefficient as Γr,λ,k := Γ
We first compute the total electron scattering-rate, which enters the so-called “quantum” scattering rate that effects
quantum-coherent processes such as quantum-oscillations:
Z k Z k
Γ2 Γ2
Z
−1 dqx dqy dqx
τQ = 2π δ (vF qx − cs |qy |) = (C19)
0 2π 2π 2ρr c s |q y | 4πρr cs v F 0 qx

where the δ-function enforces energy conservation, and the integration limits [0, k] on qx reflect momentum conserva-
tion and Pauli exclusion. The integral is logarithmically divergent in the infrared (IR, small q). This effect is directly
analogous to similar infrared singularities arising in soft-photon emission in quantum electrodynamics (QED), where
the logarithmic divergence is physically cutoff by an external IR scale such as the energy resolution of the detector.
 15

(a) = 1.5° = 1/3 (b) = 1.1° = 1/2 (c) = 0.8° = 1

105
MFP T2 ( m K 2) 105 105

MFP T2 ( m K 2)

MFP T2 ( m K 2)
104 104 104

103 103 103

0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
electron filling electron filling electron filling

FIG. 5. Prefactor of mean free path – l × T 2 versus electron filling for (a) θ = 1.5◦ , ν = 1/3 and (b) θ = 1.1◦ , ν = 1/2
and (c) θ = 0.8◦ , ν = 1.

In this regime we see that the quantum lifetime is dominated by very small angle, low-energy scattering processes.
To properly account for these infra-red divergence, in the following section, we implement a semiclassical treatment
which is equivalent to resumming the divergent soft-collinear radiation processes to infinite order [39].
For transport, small angle scatterings are ineffective at relaxing momentum. A standard cheap way of accounting for
this is to weight the scattering processes by the change in angle, which for small angle scattering can be approximated
 2
q
by inserting a factor of kFy into the above integrals. This angle-weighting results in a non-diverging transport
scattering rate:
2
Γ 2 vF

−1 k
τtr = ∼ T2 (C20)
8πρr c3s kF

At temperature T , electrons typically have momentum k ∼ T /vF relative to the Fermi surface, and we see that
the above expression gives a ρ(T ) ∼ T 2 contribution to resistivity. Fig. 5 shows the prefactor of the mean free
path l ∼ T −2 versus electron filling. We emphasize that, despite the Fermi-liquid like scaling of resistivity, the
logarithmically divergent quantum-lifetime signals a breakdown of Fermi-liquid quasi-particles.

Appendix D: MHA Polaron dynamics

In this section, we study the problem of an electron directly coupled to MHA-phonons for intermediate to high
energy regime where the phonons are not yet overdamped (the analogous problem in the asymptotic low-temperature
non-Fermi liquid regime was previously studied using a quantum Boltzmann equation approach [51]). This situation
is relevant not only for MH-polaron problems involving a single electron, but also allows a non-perturbative effective
resummation of divergent radiative corrections due to emission of soft collinear phonons at energy scales exceeding
ELD . By numerical computation, we find that hn, k|σ̂i |n, ki ≈ β kkθi for small ki /kθ with β being a constant. The
electron-phonon vertex is then,

βevF B βevF B
Γn,m,λ (k, q = 0) = ẑ × k, Γn,r,λ (k, q = 0) = k (D1)
kθ θkθ

Because of its enhanced coupling by a factor of 1/θ, we focus on the layer-antisymmetric phonon mode, in the following.
The electron-phonon coupling, using first-quantized notation for the electron coordinate r and momentum k, and
second quantized for the phonon field u, and keeping only the leading term at small phonon momentum q (dropping
gradient coupling terms) is,
α X iq·r α
He-ph (r, k) = e u(q) · k = u(r) · k (D2)
m q m

with α = βevθkF Bm
θ
. The coupling is analogous to the paramagnetic coupling of the particle to a U (1) gauge field. Our
problem is therefore, equivalent to that of a charged particle coupled minimally to a fluctuating gauge field.
For simplicity, we consider the phonon spectrum to be isotropic with a polarization independent velocity, as it will
not effect the universal physics. The effective Feynman Lagrangian for the particle in the presence of a uniform excess
 16

magnetic field discussed in the main text ∆B = ∇ × A(r) is then:

Z
L[r, u] = Le [r] + Le−ph [r, u(r)] + d2 rLph [u]
1
Le [r] = mṙ 2 − eṙ · A(r)
2h
ρ 2 2
i
Lph [u] = (∂t u) − c2s (∇u)
2
Le−ph [r, u(r)] = αṙ · u(r) (D3)

The dynamics of the electron is naturally described in the Schwinger-Keldysh path integral formulation [42]. The
corresponding action on the closed-time Keldysh contour C is:
Z
S[r, u] = dtL[r, u] (D4)
C

Decomposing the fields u(r, t) and r(t) in terms of fields residing on the forward time contour – u+ (r, t) and r + (t),
and the backward time contour – u− (r, t) and r − (t), we rewrite the action:
Z ∞
dt L[r + , u+ ] − L[r − , u− ]


S[r, u] = (D5)
−∞

We now perform standard rotation for the fields:

1 + 1 +
r cl (t) = r (t) + r − (t) , ucl (r, t) = u (r, t) + u− (r, t)
 
2 2
1 + 1 +
r q (t) = r (t) − r − (t) , uq (r, t) = u (r, t) − u− (r, t)
 
(D6)
2 2
and obtain the following action:

S = Se [r] + Sph [u] + Se−ph [r, u]

Z ∞ h  
 
i
Se [r] = dt −2mr q · r̈ cl − ṙ cl + ṙ cl · A r cl + r q + ṙ cl − ṙ cl · A r cl − r q
−∞
1 ∞
Z
Sph [u] = dtd2 r~uT D̂−1 u
~
2 −∞
Z ∞    cl 
cl + q + cl − q −

1 1 ṙ
Se−ph [r, u] = α dt u (r ) + u (r ) u (r ) − u (r ) (D7)
−∞
−1 1 ṙ q

where we express the field u by a vector in the Keldysh cl-q space:

   −1 A !
ucl (t) −1 0 D
~ (t) =
u , D̂ = (D8)
uq (t)
 −1 R  −1 K
D D
R(A)  
2
Here, 12 D−1

= ρ (∂t ± i0+ ) − c2s ∇2 . To study the motion of the electron coupled to the phonon bath, we
first integrate out the phonons, and further obtain the classical equation of motion for the electron. Integrating out
u, the effective action for the electron is,

α2 ∞
Z
Seff [r] = Se [r] + dtdt0 ∂t~r̃(t)
2 −∞
(D + DR + DA ) (r + (t) − r + (t0 ), t − t0 ) (DK − DR + DA ) (r + (t) − r − (t0 ), t − t0 )
 K 
∂ 0~r̃(t0 ) (D9)
(DK + DR − DA ) (r − (t) − r + (t0 ), t − t0 ) (DK − DR − DA ) (r − (t) − r − (t0 ), t − t0 ) t

where
   cl 
~r̃(t) ≡ 1 1 r (t)
, (D10)
−1 1 r q (t)
 17

We now make two simplifying approximations on the effective action. Following Ref. [41], we are interested in electron
motion where its position is changing slowly compared to the relevant wavelengths of the phonons being integrated out
(Lamb-Dicke approximation). This corresponds to replacing e−iq·∆r ≈ 1 in the following expression for the advanced
and the retarded Green’s function components:
dΩd2 q e−iq·∆r e−iΩ∆t
Z
R/A 1
D (∆r, ∆t) =
ρ (2π)3 (Ω ± i0+ )2 − c2s q 2
1 θ(±∆t)
≈∓ (D11)
2πρc2s ∆t
with ∆~r = ~r0 − ~r and ∆t = t0 − t. The effective action [D9] is then quadratic in r:
α2 ∞
 
0 DA
Z
Seff = Se [r] + dtdt0 ṙ(t) ṙ(t0 ) (D12)
2 −∞ DR DK
Furthermore, we restrict ourselves to semiclassical dynamics of the electron at zero temperature; thus ignoring the
role of thermally excited phonons. In this limit, we can ignore the Keldysh component of the Green’s function DK ,
in which case, the effective action takes the form:
Z ∞ Z ∞ cl q 0
α2 0 ṙ (t) · ṙ (t )
Seff = Se [r] + dt dt (D13)
2πρc2s −∞ t t − t0
Expanding the vector potential A to first order in r q , and integrating the non-local in time term by parts we get,
Z ∞ " Z ∞ cl 0
#
dA(r cl ) α2 q
 
q cl q cl cl 0 ṙ (t )
Seff = dt −2mr · r̈ − 2r · ṙi ∇Ai (r ) − − r (t) · dt 2 (D14)
−∞ dt 2πρc2s t (t − t0 )

1. Polaron cyclotron motion

dA(r cl )
We obtain the classical equation of motion of the polaron by integrating over r q , and expanding dt =
ṙi ∂i A(r cl ):
cl

eṙ cl × ∆B
Z t
ṙ cl (t0 )
∂t2 r cl = − − g2 dt0 2 (D15)
m −∞ (t − t0 )
where the dimensionless coupling constant
2
α2 β2 m

2 vF ν
g ≡ = √ (D16)
2πρmc2s 8 3π M cs
We now solve for the cyclotron motion of the polaron: r(t) = r0 (cos(ωt)x̂ + sin(ωt)ŷ) where we have suppressed the
cl label, for convenience. We take into account damping effects by allowing ω to be complex. To solve for ω, we first
express r(t) as a complex variable with the x and y components denoting the real and imaginary parts respectively.
Now, using the polaron equation of motion (Eq. D15), we get:
Z ∞ 0
ieiω(t−t )
ω = ωc,0 − ig 2 dt0 (D17)
t (t − t0 )2
where ωc,0 ≡ eB
m . Defining an effective frequency-dependent mass for the polaron: mp (ω) ≡
eB
ω and introducing a
UV-cutoff the small time differences, we obtain an expression for mp :
 
Λ Λ
mp (ω) = m 1 + g 2 log − ig 2 (D18)
ω ω
which is enhanced at small frequencies, and furthermore, diverges at ω = 0. A signature of this effect is seen in the
unconventional power law dependence of the cyclotron frequency of the polaron ωc on ∆B, discussed in the main
text, which we obtain by solving for ω using ω = meB
p (ω)
:
Λ
ω = ωc,0 − g 2 ω log
+ ig 2 Λ (D19)
ω
The real and imaginary solutions of ω give ωc and Γc respectively, and have been stated in the main text.
 18

2. Dephasing of quantum oscillations

The previous section focused on the energy-damping of oscillatory MHA-polaron motion due to radiation of soft-
collinear MHA phonons. The cyclotron line-width computed there is relevant for phase-insensitive transport and
optics cyclotron resonance measurements. For quantum oscillations measurements (e.g. Shubnikov-deHaas (SdH)
oscillations in resistivity or related oscillations in tunneling density of states), the phase-coherence of the electronic
orbit is also important, and these phenomena are affected more strongly by MHA phonons.
To analyze this effect, we employ a semiclassical computation of the density of states for an MHA polaron in an
excess magnetic field ∆B away from commensurate filling, by summing up the “return” amplitudes for a particle to
start at some position, and return to that position time t later, and then Fourier transforming with respect to t.
For cyclotron motion of an electron coupled to MHA-phonons, the electron picks up a Berry phase equal to:
R
eiθB = eiα dt ṙ·u(r(t))
(D20)
This phase has quantum fluctuations due to the quantum fluctuations of u, which give a suppression factor for
α2 2
R
the quantum-oscillation amplitude ∼ e− 2 h[ dtṙ·u(r(t))] iu , where the average is taken over the thermal ensemble of
phonons (we will restrict our attention to zero temperature).
Since the phonons are much slower than the electrons, we can approximate the phonon configuration during a
cyclotron orbit as being static (fluctuations
R of phonon field are approximately quenched on the time scale of cyclotron
motion). In that limit, we can replace dtṙ · u(r(t)) → Φ where (by Stoke’s theorem) Φ is the flux of the effective
“magnetic field”
b=∇×u (D21)
through the cyclotron orbit. This approximation is valid for up to n ∼ vcFs cyclotron orbits, on longer time scales than
2πvF
cs ωc , then we would have to account for the dynamics of the phonons.
Let us begin by computing the huui correlator in real-space. We have a Matsubara action: ρ2 (∂τ u)2 + c2s (∇u)2 ,


in 2 + 1d Euclidean time. At zero temperature, the Green’s function of the phonon will just be the 3d Coulomb
potential in 2 + 1d spacetime:
1 1 1
DM (τ, r) = p ≈ . (D22)
2 2
2πρcs cs τ + r 2 τ r 2πρcs r

The hbbi correlator can be obtained by taking spatial derivatives of this propagator, giving: hbbi = −∇2 huui =
1
2πρcs r 3 . Integrating over the cyclotron orbit, we get a flux-flux correlator:

Rc
Ac Ac
Z Z
1 1 dδr
hΦΦi = ≈ 2π ≈ (D23)
2πρcs 0
r,r ⊂cyc. orbit |r − r 0 |3 4πρcs a δr2 ρcs a

where a is a short-distance (lattice-scale) cutoff, and

Sk
Ac = (D24)
(e∆B)2
is the real-space area of the cyclotron orbit, and Sk is the momentum-space area of the Fermi-surface.
Inserting this expression into the semiclassical version of the density of states, computed as a sum over return
probabilities for various number of cyclotron oscillations [44], gives:

−1 t
Z X 2πε X 2πε 1 2 2
N (ε) = dtImGR (r = r 0 ; t)e−iεt ≈ h e−i ωc n e−iαnΦ(u) iu = e−i ωc n e−i 2 α n hΦΦi
π 0 n n
r   2  
X 2πρcs a 1 π ρcs a 2
= 2A
exp − 2 A α2
(ε − nωc ) (D25)
n
α c 2 ω c c

where in the last step, we have used the Poisson summation formula.
This result predicts a comb of Gaussian peaks centered at integer multiples of the cyclotron frequency with peak-
amplitude:
r
πρcs a
Ac ≈ ∝ ∆B (D26)
α2 Ac
 19

and width:
ωc2 Ac α2 2g 2
Γ∼ 2
∝ (∆B) . (D27)
π ρcs a
2
By contrast, for elastic impurity scattering, the return probability for n-orbits is e−n (ωc τ ) , and the quantum
q
1
√
oscillation amplitude would scale like 1/(ωc τ ) ∼ ∆B. (Note that the amplitude suppression of multiple scattering
mechanisms combine in parallel not in series, so that the linear-∆B dependence of the MHA-phonon dephasing would
dominate at small ∆B over impurity contributions). Hence, the above linear-∆B scaling provides a signature of the
unconventional MHA polaron dynamics that is qualitatively distinct from conventional contributions from impurities.
We also remark that essentially identical expressions apply for analog systems of electrons coupled to emergent
gauge-like fields, such as the composite fermion liquid of the half-filled Landau level, spinon Fermi surfaces in gapless
U (1) spin liquids, and zero-wavevector quantum critical points in metals.