
2
numerical study, we exploit fermionic truncated Wigner
approximation (fTWA) to deal with system of larger sizes
[37, 42–46]. Such an analysis is possible because fTWA
gives a reliable description in the parameter space in
which together with the tilt potential, a harmonic poten-
tial has been added to the lattice and a spin dependence
of the linear field has been taken into account. The im-
portance of the spin-dependent local potential has been
previously linked to the full MBL in the disordered Hub-
bard system because it is responsible for the localization
of the spin degrees of freedom [47]. Here we observe a
similar effect for spin dynamics on a tilted lattice and
demonstrate that the prediction of fTWA dynamics is
highly enhanced in this limit.
To discuss the dynamics of a Hubbard model on the
tilted lattice we focus our analysis on the imbalance and
QFI for charges and spins. Both observables are related
to the on-site density measurements and are experimen-
tally accessible [17, 19, 20, 24, 38, 40, 48]. Imbalance and
QFI were chosen because both are well-established indi-
cators of non-ergodicity. Moreover QFI can distinguish
the Wannier-Stark localization from SMBL through a
logarithmic-in-time type growth in the SMBL phase [25].
In this work, we show that in two dimensions QFI ex-
hibits a slow logarithmic-like growth which is similar to
the QFI behavior of disordered systems [17, 37, 49–51]
and recently studied tilted triangular ladder [25]. More-
over, we discuss the way in which harmonic potential
together with spin-dependent tilt causes a change in the
charge imbalance decay from diffusive to subdiffusive be-
havior for intermediate strength of linear potential. In-
terestingly for spins we show that the decay of imbalance
is even more pronounced and changes from superdiffusive
to subdiffusive behavior. It is worth stressing that due to
the approximation made in studying dynamical behavior,
we cannot conclude about a possibility of a transition to
SMBL phase in two dimensions. However, we can in-
dicate certain dynamical features which are difficult to
handle by other computational methods.
Finally, the fTWA method also enables us to discuss
the appearance of phase separation of ergodic and non-
ergodic long-lived phases in a two-dimensional lattice,
which is an extension of previous theoretical studies per-
formed for one-dimensional lattices [29–31].
The manuscript is constructed as follows. In Sec. II,
the fTWA method is shortly discussed. In Sec. III,
the benchmark of fTWA method against exact diago-
nalization (ED) in one-dimensional Hubbard system is
provided together with the mean square error analysis
(MSE) for imbalances and QFI. It is realized for the
charge and spin density wave initial conditions and the
roles of harmonic and spin-dependent linear potentials
are described. The two-dimensional analysis of the many-
body dynamics in tilted lattices is given in Sec. IV. The
paper ends with a summary of the obtained results (Sec.
V).
II. FTWA FOR THE HUBBARD MODEL IN
DISORDERED-FREE POTENTIALS
Before we define the semiclassical dynamics within
fTWA we begin with writing the Hubbard Hamiltonian
in terms of the creation ˆc†
iσ and annihilation ˆciσ operators
H=−X
ij, σ
Jij ˆc†
iσ ˆcjσ +UX
i
ˆni↑ˆni↓+X
i,σ
∆(i, σ)ˆniσ,(1)
where the operator ˆc†
iσ (ˆciσ) creates (annihilates)
fermionic particle at position iwith spin σ∈ {↑,↓},
ˆniσ = ˆc†
iσ ˆciσ is the density operator, Jij is the hop-
ping energy, ∆(i, σ)is the spin-dependent on-site po-
tential and Uis the on-site interaction energy between
two spin species. Throughout this work it is assumed
that Jij is non-zero for the nearest neighbour sites only
for which we set Jij =J. Then, instead of solving the
Schrödinger equation, approximated quantum dynamics
in fTWA is obtained by equating Hamilton equations of
motion with the addition of quantum fluctuation encoded
in the initial conditions through the Wigner function W
[42, 52]. Equations of motion for the Hubbard take the
form [37, 42]
idρmσnσ
dt =−X
k
(Jnkρmσ,kσ −Jkmρkσ,n σ)
+ρmσnσ [∆(n, σ)−∆(m, σ) + U(ρn−σn−σ−ρm−σm−σ)] ,
(2)
where ρmσnσ are phase space variables corresponding to
fermionic bilinears ˆ
Enσ
mσ =ˆc†
nσ ˆcmσ −ˆc†
mσ ˆcnσ/2(ρnσmσ
are obtained by the Wigner-Weyl quantization procedure
[42]). Here the so-called ρrepresentation of Hamiltonian
Hwas used [37, 42]. In order to obtain the expectation
value of a given observable, e.g. ˆ
O, trajectories are sam-
pled from the initial Wigner function W(ρ0)and summed
up according to the following procedure
Dˆ
O(t)EfTWA
≈ZOW(ρ(t))W(ρ0)dρ0=hOW(t)icl ,(3)
where OWis a Weyl symbol of ˆ
O,ρ(t) =
{ρiσjσ0(t) : i, j ∈ {1,2, ..., N}, σ, σ0∈ {↑,↓}},Nis the
number of sites, ρ0=ρ(t= 0). Initial conditions en-
coded in the Wigner function W(ρ0)are obtained by ap-
proximating W(ρ0)as multivariate Gaussians and read-
ing off its first and second moments from matching the
semiclassical and quantum expectation values [42].
Except for non-interacting systems, fTWA gives an ac-
curate description of general systems only in the early
times [52]. However, in the next section, we numerically
show that slight modification of the linear potential leads
to the improvement of the long-time fTWA predictions.
In one-dimensional systems, we consider the following
form of the onsite potential
∆(j, σ)=∆1(δσ↓+Aδσ↑)j+ ∆2(j−j0)2,(4)