
2
tions are gapped. In the discussion section, we establish
the connection and consistency between our simulations
and recent experimental studies on TMI phases in h-BN
nonaligned TBG devices [41].
BM model and projected interaction.— We utilize the
BM model and project interactions between fermions to
the moir´e flat bands. The BM model Hamiltonian [3]
for the τvalley takes the following form Hτ
k,k0=
−~vF(k−K1)·σδk,k0V
V†−~vF(k−K2)·σδk,k0
where K1and K2mark the two Dirac points
in the τvalley from layers 1 and 2 respectively.
V=U0δk,k0+U1δk,k0−G1+U2δk,k0−G1−G2and
V†=U†
0δk,k0+U†
1δk,k0+G1+U†
2δk,k0+G1+G2
are the inter-layer tunnelings with matrixes
U0=u0u1
u1u0,U1=u0u1e−i2π
3
u1ei2π
3u0and
U2=u0u1ei2π
3
u1e−i2π
3u0where u0and u1are the
intra- and inter-sublattice inter-layer tunneling am-
plitudes. G1= (−1/2,−√3/2),G2= (1,0) are the
reciprocal vectors of the moir´e Brillouin zone (mBZ),
and K1= (0,1/2√3)|G1,2|,K2= (0,−1/2√3)|G1,2|.
For control parameters, we set (θ, ~vF/a0, u0, u1) =
(1.08◦,2.37745 eV,0 eV,0.11 eV) for the chiral limit, and
for non-chiral model, we set u0= 0.06 eV following
Refs. [50, 55, 57, 67]. Here, θis the twisting angle and
a0is the lattice constant of monolayer graphene.
For interactions, in addition to the Coulomb repulsion,
here we have the option to include one more interaction
term and the sign problem will still remain polynomial.
HI=1
2Ω X
q
(V1(q)δρ1,qδρ1,−q+V2(q)δρ2,qδρ2,−q)
δρ1,q=X
k,α,τ,s
(c†
k,α,τ,sck+q,α,τ,s −ν+ 4
8δq,0)
δρ2,q=X
k,α,s
(c†
k,α,τ,sck+q,α,τ,s −c†
k,α,−τ,sck+q,α,−τ,s)
(1)
The first term in HI(V1>0) is the Coulomb interac-
tions, and the second term V2≥0 introduces repulsive
interactions for fermions in the same valley and attrac-
tions between the two valleys, which can be introduced
as a phenomenology term describing inter-valley attrac-
tions [68–72]. At V2= 0, this model recovers the stan-
dard TBG model with Coulomb repulsions. When V2is
turned on, the inter-valley attraction favors inter-valley
pairing and could stabilize a superconducting ground
state. The normalization factor in HIis Ω = L2√3
2a2
M
with Lbeing the linear system size of the system. kand
qcover the whole momentum space, νis the filling factor
and α, τ, s represent layer/sublattice, valley, spin indices,
respectively. The mometum dependence for non-negative
V1and V2is unimportant to polynomial sign problem.
Here, for simplicity, we set V2=γV1with γbeing a
non-negative constant and for V1, we use Coulomb inter-
action screened by single gate in our simulation V1(q) =
e2
4πε Rd2r1
r−1
√r2+d2eiq·r=e2
2ε
1
|q|1−e−|q|dwhere
d
2= 20 nm is the distance between graphene layer and
single gate, = 70is the dielectric constant. We then
project the interactions HIto the moir´e flat bands (See
SM [73]) and use the projected Hamiltonian to carry out
sign bounds analysis and QMC simulations.
Polynomial sign bounds.— In QMC simulations, the ex-
pectation value of a physical observable Ois measured as
hˆ
Oi=PlWlhˆ
Oil, where Wland hˆ
Oilare the weight and
the expectation value for the configuration l. Instead of
summing over all configurations, a QMC simulation sam-
ples the configuration space using the probability Wl. In
sign-problem-free QMC simulations, Wl≥0 for all land
an accurate expectation value can be obtained by only
sampling a small number of configurations – the impor-
tance sampling, and this number scales as a power-law
function of the system size. However, for many quantum
systems, Wlcan be negative or even complex, and thus
to obtain an accurate expectation value, it requires to
sample a large number of configurations, which usually
scales as an exponential function of the system size [74].
It is worthwhile to emphasize that the sign prob-
lem doesn’t always lead to an exponentially high com-
putational cost. To measure the severity of the sign
problem, here we use the average sign hsigni=
PlWl/Pl|Re(Wl)|, where |Re(Wl)|is the absolute
value of the real part of Wl. For physical partition func-
tion Z=PlWl=PlRe(Wl), this average sign is be-
tween 0 and 1, and hsigni= 1 means that the system
is sign problem free, while smaller hsignimeans sev-
erer sign problem. In a d-dimension quantum systems
that suffers from the sign problem, hsigni ∼ exp(−βLd)
where β= 1/T the inverse temperature, indicating that
the number of configurations needed in QMC simulations
scales as an exponential function of the space-time vol-
ume. For polynomial sign problem, although hsigni<1
(i.e., the system does suffer from the sign problem),
1/hsigniis a polynomial function of the system size, and
thus the number of configurations needed only scales as
a power-law function of the system size.
Although the average sign can be easily measured in
QMC simulations, it usually doesn’t have a simple an-
alytic formula. To estimate the numerical cost to over-
come the sign problem, we utilize the sign bound hsignib
defined in Ref. [60]. As proved in Ref. [60], hsignibis
the lower bound of hsigni(i.e., hsignib6hsigni). Thus,
if the sign bound scale is a power-law function of the
system size, the sign problem is (at most) polynomial.
Remarkably, the low temperature sign bound in moir´e
flat bands can be easily calculated by counting ground
state degeneracy, which can be obtained using SU(4) and
SU(2) Young diagram as employed in Refs. [50, 51, 56]