Preempting Fermion Sign Problem: Unveiling Quantum Criticality through

Nonequilibrium Dynamics
Yin-Kai Yu,1, 2, 3 Zhi-Xuan Li,1 Shuai Yin,1, ∗ and Zi-Xiang Li2, 3, †
1
Guangdong Provincial Key Laboratory of Magnetoelectric Physics and Devices,
School of Physics, Sun Yat-sen University, Guangzhou 510275, China
2
Beijing National Laboratory for Condensed Matter Physics & Institute of Physics,
Chinese Academy of Sciences, Beijing 100190, China
3
University of Chinese Academy of Sciences, Beijing 100049, China
(Dated: October 25, 2024)
The notorious fermion sign problem, arising from fermion statistics, constitutes one of the main
obstacles of deciphering quantum many-body systems by numerical approach. The progress in
arXiv:2410.18854v1 [cond-mat.str-el] 24 Oct 2024

overcoming sign problem will definitely lead to a great leap in various areas of modern physics.
Here, by deviating from the conventional cognition that nonequilibrium studies should be more
complicated than equilibrium cases, we propose an innovative framework based on nonequilibrium
critical dynamics to preempt sign problem and investigate quantum critical point in fermionic model
through numerically exact quantum Monte Carlo (QMC) simulation. By virtue of universal scaling
theory of imaginary-time relaxation dynamics, we demonstrate that accurate critical point and
critical exponents can be obtained in the short-time stage, in which the sign problem is not severe
such that the QMC is accessible. After confirming the effectiveness of the method in two typical
interacting fermionic models featuring Dirac quantum critical point (QCP), we for the first time
reveal the quantum phase diagram in the Hubbard model hosting SU(3)-symmetric Dirac fermions,
and find that the QCP between Dirac semi-metal and a λ8 -antiferromagnetic phase belongs to a
new universality class different from the previously known Gross-Neveu transitions.

Introduction — The sign structure of the quantum 44] and strange metallicity [45, 46]. The understanding
many-body wave functions works like a double-edged of QCP in a non-perturbative way is severely hindered
sword. On the one hand, it endows fertile exotic phe- by the sign problem likewise. Developing a generic unbi-
nomena ranging from condensed matter physics [1, 2] to ased theoretical approach to decipher quantum criticality
high energy physics [3]. Quantum Monte-Carlo (QMC) not only has overarching meaning in fundamental theory,
is among the most important theoretical approaches to but also immensely promotes the research on the exotic
study the exotic properties of strongly correlated quan- phenomena emerging in quantum materials.
tum many-body systems [4–6], because it is intrinsically Here, we propose a general framework to preempt sign
unbiased and numerically exact. However, on the other problem in QMC and unveil the quantum criticality.
hand, the sign structure in the wave function results in Generally, fathoming the nonequilibrium properties in
the notorious sign problem [7–11], which considerably quantum many-body systems is substantially challeng-
plagues the application of QMC to investigating interact- ing, compared with the equilibrium ones. However, we
ing models potentially featuring intriguing physics, for leverage the nonequilibrium behavior and demonstrate
instance, the Hubbard model at generic filling [12, 13] that the short-time critical dynamics, which was first
and the lattice QCD at finite baryon density [14]. Since proposed in classical systems [47] and used to detect the
the general solution of sign problem is lacking and nonde- classical critical behaviors [48] and then generalized to
terministic polynomial (NP)-hard [9], establishing novel imaginary-time dynamics of QCP [49], provides an inge-
generic strategies to attack the sign problem in QMC nious strategy to reliably probe quantum criticality in the
will definitely provoke significant advances in the realm presence of sign problem, as illustrated in Fig. 1a. The
of quantum many-body physics [15–36]. underling mechanism is that in the short-time stage of
Among various exotic phenomena in quantum many- imaginary-time relaxation for some typical initial states,
body systems, quantum criticality emerges as a particu- (1) universal scaling behaviors manifesting the quantum
larly crucial and fascinating one. The underlying physics criticality appear [49, 50]; (2) the sign problem remains
of the quantum critical point (QCP) lays the foundation mild in this stage such that the reliable results are acces-
for achieving a unified theoretical framework to char- sible by QMC simulation. Hence, bestowing the scaling
acterize different phases and phase transitions [37–41]. of short-time dynamics [47–50], we can accurately de-
Moreover, quantum criticality is intimately associated termine critical properties of the QCP, while the sign
with fundamentally important phenomena in condensed problem therein, whose severity generally exponentially
matter physics such as high-Tc superconductivity [42– increases with the imaginary time [9, 51], is largely alle-
viated compared with the equilibrium one involving long
imaginary-time evolution.
∗ yinsh6@mail.sysu.edu.cn In the following, we demonstrate the framework by
† zixiangli@iphy.ac.cn studying two typical interacting fermionic models fea-
 2

a b c

Ū -1

Nonequilibrium
critical region
d e

sign problem
weak strong

Uc U

FIG. 1. Scheme of preempting sign problem to probe quantum criticality via short-time critical dynamics and
the application in single-Dirac-fermion Hubbard model. a, With some typical initial states, scaling behaviors governed
by the QCP are reflected in the short-time stage, in which the sign problem is still weak as shown in the inset. The red dashed
line therein marks the average sign at τ = 0.3Lz . b, Determination of QCP as Uc = 7.220(37) via the intersection points of
curves of the correlation length ratio RFM versus U for different sizes at τ = 0.3Lz with DSM initial state. c, Determination
of 1/ν = 1.18(3) by scaling collapse of RFM versus rescaled (U − Uc ). d, Determination of ηϕ = 0.33(2) via scaling collapse of
curves of the structure factor SFM versus rescaled τ at Uc . e, Determination of ηψ = 0.135(2) via scaling collapse of curves of
the fermion correlation Gf versus rescaled τ at Uc .

turing Dirac QCP. We unambiguously show that ac- which is generally one in Dirac QCP. Note that all criti-
curate critical properties are accessed through short- cal exponents in Eq. (1) are controlled by the QCP in the
time dynamics, despite the presence of sign problem. ground state. Accordingly, from Eq. (1), one can achieve
More remarkably, we adopt the approach to investi- critical properties from the expectation value of O in the
gate a paradigmatic strongly interacting model hosting relaxation process. This nonequilibrium approach is no-
Dirac fermions with SU(3) symmetry, namely the SU(3) tably different from the conventional one which requires
staggered-flux Hubbard model, which is sign problem- that the ground state should be determined through the
atic in any known algorithm. For the first time we un- evolution of long imaginary time. Because in general the
ravel the ground-state phase diagram of the model and severity of sign problem exponentially increases with τ in
reveal the critical properties of the quantum phase tran- QMC [9], the sign problem in short-time stage of relax-
sition between the Dirac semi-metal (DSM) and a λ8 - ation is significantly mitigated compared with the sim-
antiferromagnetic (AFM) phase. Intriguingly, the QCP ulation on ground-state properties, thereby enabling the
belongs to a novel universality class distinct from previ- large-scale QMC simulation with high accuracy despite
ous Gross-Neveu transitions. the presence of sign problem. In the following, we eluci-
Theoretical framework — We consider the imaginary- date the theoretical framework by systematically study-
time relaxation dynamics for which the wave func- ing several representative models with sign problem.
tion |ψ(τ )⟩ evolves according to the imaginary-time Single-Dirac-fermion Hubbard model — We first con-
∂
Schrödinger equation − ∂τ |ψ(τ )⟩ = H|ψ(τ )⟩ imposed by sider the SLAC fermion Hubbard model with the Hamil-
the normalization condition. As its long-time solution is tonian [52–54]:
the ground state, this equation provides a routine method
to access ground-state properties in numerical computa- X †
X 1

1

H= tR ci↑ ci+R↓ + h.c. + U ni↑ − ni↓ − ,
tions [51]. Moreover, near the QCP, it was shown that 2 2
i,R i
universal scaling behaviors manifest themselves not only
at the ground state, but also in the imaginary-time re- (2)
laxation process. Particularly, with initial states corre-
i(−1)Rx (−1)Ry
sponding to the fixed points of scale transformation, the where tR = L
sin π RLx
δRy ,0 + L Ry δRx ,0 is the am-
π π sin π L
short-time dynamics of observable O satisfies the follow- plitude of long-range hopping and U is the strength of
ing scaling form [49, 50]: repulsive interaction. The model is sign problematic in
1
O(τ, g, L) = L−κ fO (gL ν , τ L−z ), (1) any known QMC algorithm. Nonetheless, since the sign
problem is relatively benign, previous QMC study reveals
where L is linear system size and g is the distance to the a chiral Ising QCP separating Dirac semimetal (DSM)
critical point, κ is the scaling dimension of O, and ν is phase and ferromagnetic (FM) phase [52].
the correlation-length exponent, z is dynamical exponent Here, we present the procedure for unraveling the crit-
 3

a b evaluated here with previous results, we find that all

are approximately consistent with previous study of
Gutzwiller QMC [52] and functional renormalization
group [56]. Notice that since sign problem is largely
alleviated in our work owing to the short evolution time,
we can achieve larger system size L = 23 [55], compared
with previous studies, and thus reach the reliable results
c d of critical exponents. Consequently, we unambiguously
demonstrate that the short-time dynamics via QMC sim-
ulation provides a powerful approach to fathom quantum
critical properties in sign-problematic strongly interact-
ing models.
Spinless t-V model — To further demonstrate the
framework of our approach, we study another typical
interacting Dirac-fermion model, termed as Honeycomb
FIG. 2. Probing quantum criticality via short-time spinless t-V model, with the Hamiltonian: [17, 57–59]:
critical dynamics in spinless t-V model with CDW
X 1

1

initial state. a, Determination of QCP as Vc = 1.35(1) via
X †
H = −t ci cj + V ni − nj − , (3)
the intersection points of RCDW versus V for different L at 2 2
⟨ij⟩ ⟨ij⟩
τ = 0.3Lz . Shown in Inset is the evolution of average sign
with red dashed curve marks τ = 0.3Lz . b, Determination of
1/ν = 1.30(18) via scaling collapse of RCDW versus rescaled where t is nearest-neighbor (NN) hopping amplitude and
(V − Vc ). c-d, Determination of ηϕ = 0.49(5) and ηψ = V denotes the strength of NN density interaction. The
0.073(4) via scaling collapse of short-time dynamics of SCDW appearance of sign problem in Eq. (3) depends on the
and Gf versus rescaled τ , respectively. channel of Hubbard-Strotonovich (H-S) transformation.
Hence, the model provides a genuine platform to con-
firm the accuracy and feasibility of our approach to un-
ravel QCP in the presence of sign problem. Previous
ical properties via the short-time scaling of the observ- sign-free QMC studies, with the H-S transformation in
able as dictated in Eq. (1). In the conventional projector the hopping channel, reveal the critical properties of the
QMC approach, a sufficiently long evolution time should QCP separating DSM and charge-density-wave (CDW)
be implemented to ensure that the ground-state prop- phases [58]. Here, we decouple the interaction in the
erties are accessed, resulting in the severe sign problem sign-problematic density channel and study the QCP via
exponentially increasing with τ , as shown in Fig. 1. We short-time relaxation dynamics [60].
overcome this critical difficulty owing to the short-time The procedure is the same as the previous section, ex-
critical relaxation dynamics. To determine the critical cept that the initial state is chosen as the CDW fully
point via Eq. (1), we consider observable O as the di- ordered state. We will demonstrate that the initial state
mensionless correlation length ratio RFM for the FM or- can be flexibly selected in our approach, thereby offer-
der (See SM Sec. II for the detailed definition [55]) and ing various routes to extract critical properties. We fix
fix τ at short-time stage as τ = 0.3Lz , where z = 1 τ = 0.3Lz , and determine QCP from the crossing point of
due to the nature of Dirac QCP. Accordingly, RFM sat- curves of correlation-length ratio RCDW versus V (RCDW
isfies RFM = fR (gL1/ν ) where g = U − Uc . We show is defined in SM Sec. III [55]), as displayed in Fig. 2a,
the numerical results in Fig. 1b, in which the crossing giving rise to Vc = 1.35(1). The scaling collapse anal-
point of RFM versus U for different L reveals a quan- ysis determines the value of ν as 1/ν = 1.30(18). Ad-
tum phase transition from DSM to FM phase occurring ditionally, the anomalous dimensions ηϕ = 0.49(5) and
at Uc = 7.220(37). Furthermore, we perform data col- ηψ = 0.073(4) are obtained from the evolution of the
lapse analysis of RFM for different L versus (U − Uc )L1/ν CDW structure factor SCDW and the fermion correla-
and achieve accurate critical exponent 1/ν = 1.18(3), tion Gf (defined in SM Sec. III [55]), respectively, at Vc .
as shown in Fig. 1c. Furthermore, fixing U = Uc , we Both Vc and ηϕ are consistent with previous numerical
calculate the imaginary-time evolution of FM structure results, while ηψ , to the best of our knowledge, is nu-
order SFM and fermion correlation Gf (defined in SM merically determined for the first time, and is consistent
Sec. II [55]), whose scaling dimensions are (1 + ηϕ ) and with the previous results of the functional renormaliza-
ηψ , respectively [50, 52]. By making the rescaled curves tion group [56]. The similar computation with DSM ini-
of SFM L(1+ηϕ ) and Gf Lηψ versus τ L−z for different L col- tial state yields the consistent results, with the details in-
lapse according to Eq. (1) as shown in Figs. 1d and 1e, we cluded in the SM [55]. Hence, these results for model (3)
achieve the anomalous dimensions of the order parameter further establish the feasibility of the approach based on
boson ηϕ = 0.33(2) and the Dirac fermion ηψ = 0.135(2), short-time dynamics in accessing the critical properties
respectively. in the presence of sign problem. Moreover, the flexible
Comparing the critical point and critical exponents choice of initial states provide benchmarks for the results,
 4

a b c

DSM d e

ф -ф

λ 8-AFM

FIG. 3. Phase diagram and quantum criticality in SU(3) Hubbard model detected by short-time dynamics with
λ8 -AFM initial state. a, Phase diagram determined via short-time dynamics. Insets show the energy spectra of DSM state
(upper left) and Sketch of λ8 -AFM order in which a fermion with one flavor (blue) is situated at one sublattice and double
fermions with the other two flavors (pink and violet) are situated at the other sublattice (lower right). b-c, Critical point
Uc = 1.10(5) for ϕ = 0.075π and 1/ν = 0.68(5) determined via curves of Rλ8 -AFM versus U for different L at τ = 0.25Lz .
Shown in Inset of b is the evolution of average sign with red dashed curve marks τ = 0.25Lz . d-e, ηϕ = 0.55(5) and ηψ = 0.15(3)
determined via the scaling collapse of evolution curves of Sλ8 -AFM and Gf , respectively.

further improving the accuracy of the approach. QMC simulation, are devoted to understanding the ex-
SU(3) Hubbard model — The successful application otic phenomena arising from the interplay between strong
of the method in previous two models encourages us to interaction, Dirac fermions and multi-flavor physics [63–
study the critical properties in unexplored models. In the 67]. However, despite its fundamental importance, the
following, we implement the approach to systematically unbiased numerical study on SU(N)-symmetric model
investigate a sign-problematic interacting model featur- with repulsive Hubbard interaction for odd N is scarce,
ing exotic Dirac QCP belonging to a novel universality largely due to the presence of notorious sign problem in
class. We consider SU(3) Hubbard model with staggered QMC simulation. Hence, we implement our approach to
magnetic flux on square lattice, described by the follow- preempt sign problem and unravel the quantum phases
ing Hamiltonian: and exotic QCP in Eq. (4), which is a minimal interacting
!2 model featuring SU(N) symmetry with odd N .
X † UX X 3
H=− tij ciα cjα + niα − , (4) Before embarking on QMC simulation, we perform a
2 i α
2 mean-field calculation to detect the feature of the ground
⟨ij⟩α
state schematically. The mean-field calculation shows
where α = 1, 2, 3 is the flavor index of fermion, U that an antiferromagnetic (AFM) order is favored by the
is repulsive Hubbard interaction strength, and tij = strong Hubbard interaction [55]. The order parameter is
teiθij , in which t = 1 is set as energy unit. As illus- characterized in terms of the generators of SU(3) sym-
trated in Fig. 3a, the magnetic flux in each plaquette is metry group, which is expressed by the eight Gell-Mann
= (−1)ix +iy ϕ. For non-zero ϕ, the energy disper-
P
□ θ ij matrices in flavor space for convenience [68] (The de-
sion of Eq. (4) in the non-interacting limit features two tails of the SU(3) algebra are introduced in SM [55]). In
Dirac points located at momenta (± π2 , ± π2 ), as shown in stark contrast to the case of SU(2), only one Gell-Mann
SM [55]. We fix our simulation at half filling such that matrix is full-rank without zero eigenvalue, dubbed as
the Fermi level is located at Dirac points. The model of λ8 = diag( √13 , √13 , − √23 ). Consequently, for the AFM
Eq. (2) respects SU(3) × Z2 symmetry, where SU(3) is order in SU(3) fermions, only the order parameter asso-
the rotation symmetry in flavor space and Z2 is the sub- ciated with λ8 is the mass term fully opening spectral
lattice symmetry which exchanges the two sublattices on gap in Dirac fermions, which is expressed as mλ8 -AFM =
the staggered-flux square lattice with each unit cell con- 1
P † αβ ix +iy
taining two sites. L2 i,α,β ciα λ8 ciβ (−1) . The mean-field calcula-
The SU(N) symmetry plays overarching roles in mod- tion also confirms that such AFM order is energetically
ern physics. For instance, the SU(3) symmetry lays the favorable in model Eq. (4) [55].
foundation for strong interaction between quarks [3]. Re- In the λ8 -AFM ordered phase, the Z2 sublattice sym-
cently, fermionic interacting models with SU(N) symme- metry is broken and the SU(3) symmetry is broken into
try have been realized in optical lattice [61, 62]. Exten- SU(2) × U(1) (See SM Sec. IV for details [55]), as illus-
sive theoretical and numerical efforts, including sign-free trated in Fig. 3a. Accordingly, the order parameter man-
 5

SU(3)×Z2
ifold is SU(2)×U(1) , which is remarkably different from the ple, for the SU(3) Hubbard model with L = 10, the es-
universality classes of the Gross-Neveu QCP studied pre- timated computational time in equilibrium QMC is mil-
viously. Thus, the QCP separating DSM and λ8 -AFM lions of times that in our approach [55]. Consequently,
phase is a novel Dirac QCP. our work paves a novel avenue to studying exotic QCP in
To reveal the critical properties, we perform short-time quantum many-body systems, which are not amendable
QMC simulation with fully ordered initial state by fix- to the unbiased numerical approach previously.
ing ϕ = 0.075π and τ L−z = 0.25 where z = 1. The More intriguingly, employing the newly developed ap-
results of correlation length ratio Rλ8 -AFM for λ8 -AFM proach, for the first time we reveal the ground-state
order (defined in SM Sec. IV [55]) are shown in Fig. 3b properties of the staggered-flux SU(3) repulsive Hub-
and the crossing points dictate the phase transition point bard model, and reveal a novel Dirac QCP different from
from DSM to λ8 -AFM phase occurring at Uc = 1.10(5). previously known Gross-Neveu transitions. Our numer-
We also compute the correlation-length ratios for other ical studies motivate systematic theoretical analysis of
types of AFM order instead of λ8 , further verifying that the QCP in SU(3) Dirac fermions by field-theory ap-
only λ8 -AFM long-range order is present in the ground proach. Moreover, the critical properties are potentially
state [55]. With varying magnetic flux ϕ, we implement detectable in the synthetic quantum simulators such as
a similar procedure and arrive at the ground-state phase optical lattice of cold atoms [61, 62]. Our results on the
diagram of the Eq. (4), as shown in Fig. 3a, in which Uc SU(3) Hubbard model provide theoretical guidance for
increases with ϕ. the future experimental exploration of exotic physics in
strongly correlated SU(N) Dirac fermions.
After accessing the phase diagram, we investigate the The basic procedure in our approach can be straight-
critical properties. The data collapse analysis of Rλ8 -AFM forwardly generalized to other fermionic QCP, including
versus (U − Uc )L1/ν in the regime close to QCP with metallic QCP involving Fermi surface, for which the criti-
fixed ϕ = 0.075 gives 1/ν = 0.68(5), as shown in Fig. 3c. cal properties attract considerable attentions, but remain
Moreover, at Uc , the short-time dynamics of the struc- largely elusive. Additionally, we believe that short-time
ture factor Sλ8 -AFM and the fermion correlation function nonequilibrium QMC is also applicable in sign problem-
Gf (defined in SM Sec. IV [55]) yield ηϕ = 0.55(5) and atic bosonic models. We should mention that in some
ηψ = 0.15(3), as presented in Fig. 3d and 3e, respectively. specific cases, although sign problem is alleviated, it re-
To confirm the universality of this phase transition, we mains difficult to access the critical properties with large
also evaluate the critical exponents for other values of ϕ, system size in our current simulation. For example, when
as presented in SM [55], which are consistent with each staggered magnetic flux is large in SU(3) Hubbard model,
other within the error bar. Remarkably, the critical ex- even at short-time stage the sign problem is too severe to
ponents found here are different from those of the Gross- allow us reach the results with high accuracy close to the
Neveu universality classes [69, 70] (for comparison, see QCP. Nonetheless, bestowing the flexibility of the frame-
SM [55]), confirming that this phase transition belongs work, it is feasible to further ameliorate the efficiency of
to a new universality class. our approach, for instance through improvement of ini-
Concluding remarks — In summary, we establish an in- tial state and optimization of H-S channel [31, 32], which
novative theoretical framework to unravel critical prop- is an interesting direction in future. We believe that our
erties in quantum many-body systems. The approach work promises a pathway towards a general strategy of
proposed in our work enables preempting the notori- studying the QCP in strongly correlated systems by un-
ous sign problem in QMC by leveraging the short-stage biased numerical approach.
imaginary-time relaxation dynamics. We demonstrate Acknowledgments — We acknowledge helpful discus-
the accuracy and efficiency of the approach by systemat- sions with Yiwen Pan. Yin-Kai Yu, Zhi-Xuan Li and
ically investigating several strongly interacting models by Shuai Yin are supported by the National Natural Sci-
sign-problematic QMC. The comparison of our approach ence Foundation of China (Grants No. 12222515 and
with conventional QMC shows that the sign problem is No. 12075324). Zi-Xiang Li is supported by the NSFC
enormously reduced in our approach, thus largely reduc- under Grant No. 12347107. Shuai Yin is also supported
ing the computational time and enabling the accessing by the Science and Technology Projects in Guangdong
the numerically accurate critical exponents. For exam- Province (Grants No. 2021QN02X561).

[1] S. Sachdev, Quantum Phase Transitions, 2nd ed. (Cam- Rev. D 24, 2278 (1981).
bridge Univ. Press, 2011). [5] J. E. Hirsch, D. J. Scalapino, R. L. Sugar, and
[2] J. Zaanen, Science 319, 1205 (2008), R. Blankenbecler, Phys. Rev. Lett. 47, 1628 (1981).
https://www.science.org/doi/pdf/10.1126/science.1152443. [6] J. E. Hirsch, Phys. Rev. B 31, 4403 (1985).
[3] S. Weinberg, The quantum theory of fields, Vol. 1 (Cam- [7] E. Y. Loh, J. E. Gubernatis, R. T. Scalettar, S. R. White,
bridge university press, 1995). D. J. Scalapino, and R. L. Sugar, Phys. Rev. B 41, 9301
[4] R. Blankenbecler, D. J. Scalapino, and R. L. Sugar, Phys. (1990).
 6

[8] P. Henelius and A. W. Sandvik, Phys. Rev. B 62, 1102 [36] O. Grossman and E. Berg, Phys. Rev. Lett. 131, 056501
(2000). (2023).
[9] M. Troyer and U.-J. Wiese, Phys. Rev. Lett. 94, 170201 [37] S. L. Sondhi, S. M. Girvin, J. P. Carini, and D. Shahar,
(2005). Rev. Mod. Phys. 69, 315 (1997).
[10] M. B. Hastings, Journal of Mathematical Physics 57, [38] M. Vojta, Reports on Progress in Physics 66, 2069
015210 (2015), https://pubs.aip.org/aip/jmp/article- (2003).
pdf/doi/10.1063/1.4936216/11140394/015210 1 online.pdf. [39] J. A. Hertz, Phys. Rev. B 14, 1165 (1976).
[11] Z. Ringel and D. L. Kovrizhin, Sci- [40] A. J. Millis, Phys. Rev. B 48, 7183 (1993).
ence Advances 3, e1701758 (2017), [41] E. Berg, S. Lederer, Y. Schattner, and S. Trebst, Annual
https://www.science.org/doi/pdf/10.1126/sciadv.1701758. Review of Condensed Matter Physics 10, 63 (2019),
[12] D. P. Arovas, E. Berg, S. A. Kivelson, and S. Raghu, An- https://doi.org/10.1146/annurev-conmatphys-031218-
nual Review of Condensed Matter Physics 13, 239 (2022). 013339.
[13] M. Qin, T. Schäfer, S. Andergassen, P. Corboz, and [42] D. J. Scalapino, Rev. Mod. Phys. 84, 1383 (2012).
E. Gull, Annual Review of Condensed Matter Physics [43] P. A. Lee, N. Nagaosa, and X.-G. Wen, Rev. Mod. Phys.
13, 275 (2022). 78, 17 (2006).
[14] T. DeGrand and C. DeTar, Lattice Methods for Quan- [44] S. Lederer, Y. Schattner, E. Berg, and S. A. Kivelson,
tum Chromodynamics (WORLD SCIENTIFIC, 2006) Phys. Rev. Lett. 114, 097001 (2015).
https://www.worldscientific.com/doi/pdf/10.1142/6065. [45] C. M. Varma, Rev. Mod. Phys. 92, 031001 (2020).
[15] Z.-X. Li and H. Yao, Annual Review of [46] P. W. Phillips, N. E. Hussey, and P. Ab-
Condensed Matter Physics 10, 337 (2019), bamonte, Science 377, eabh4273 (2022),
https://doi.org/10.1146/annurev-conmatphys-033117- https://www.science.org/doi/pdf/10.1126/science.abh4273.
054307. [47] H. K. Janssen, B. Schaub, and B. Schmittmann,
[16] S. Chandrasekharan and U.-J. Wiese, Phys. Rev. Lett. Zeitschrift für Physik B Condensed Matter 73, 539
83, 3116 (1999). (1989).
[17] E. F. Huffman and S. Chandrasekharan, Phys. Rev. B [48] Z. B. Li, L. Schülke, and B. Zheng, Phys. Rev. Lett. 74,
89, 111101 (2014). 3396 (1995).
[18] C. Wu and S.-C. Zhang, Phys. Rev. B 71, 155115 (2005). [49] S. Yin, P. Mai, and F. Zhong, Phys. Rev. B 89, 144115
[19] E. Berg, M. A. Metlitski, and (2014).
S. Sachdev, Science 338, 1606 (2012), [50] Y.-K. Yu, Z. Zeng, Y.-R. Shu, Z.-X. Li, and S. Yin,
https://www.science.org/doi/pdf/10.1126/science.1227769. Nonequilibrium dynamics in dirac quantum criticality
[20] Z.-X. Li, Y.-F. Jiang, and H. Yao, Physical Review B 91, (2023), arXiv:2310.10601 [cond-mat.str-el].
241117 (2015). [51] F. Assaad and H. Evertz, World-line and determinan-
[21] Z.-X. Li, Y.-F. Jiang, and H. Yao, Physical Review Let- tal quantum monte carlo methods for spins, phonons
ters 117, 267002 (2016). and electrons, in Computational Many-Particle Physics
[22] S. Zhang and H. Krakauer, Phys. Rev. Lett. 90, 136401 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2008)
(2003). pp. 277–356.
[23] L. Wang, Y.-H. Liu, M. Iazzi, M. Troyer, and G. Harcos, [52] S. M. Tabatabaei, A.-R. Negari, J. Maciejko, and
Phys. Rev. Lett. 115, 250601 (2015). A. Vaezi, Phys. Rev. Lett. 128, 225701 (2022).
[24] Z. C. Wei, C. Wu, Y. Li, S. Zhang, and T. Xiang, Phys. [53] Z.-X. Li, A. Vaezi, C. B. Mendl, and
Rev. Lett. 116, 250601 (2016). H. Yao, Science Advances 4, eaau1463 (2018),
[25] R. Mondaini, S. Tarat, and R. T. Scalettar, Science 375, https://www.science.org/doi/pdf/10.1126/sciadv.aau1463.
418 (2022). [54] T. C. Lang and A. M. Läuchli, Physical Review Letters
[26] D. Hangleiter, I. Roth, D. Nagaj, and J. Eis- 123, 137602 (2019).
ert, Science Advances 6, eabb8341 (2020), [55] See Supplementary Materials at [URL will be inserted
https://www.science.org/doi/pdf/10.1126/sciadv.abb8341. by publisher] for more details of nonequilibrium PQMC
[27] R. Levy and B. K. Clark, Phys. Rev. Lett. 126, 216401 simulation and more details of the three fermion models
(2021). in this study.
[28] Z.-Y. Han, Z.-Q. Wan, and H. Yao, Pfaffian quantum [56] G. P. Vacca and L. Zambelli, Phys. Rev. D 91, 125003
monte carlo: solution to majorana sign ambiguity and (2015).
applications (2024), arXiv:2408.10311 [cond-mat.str-el]. [57] L. Wang, P. Corboz, and M. Troyer, New Journal of
[29] A. Alexandru, G. m. c. Başar, P. F. Bedaque, and N. C. Physics 16, 103008 (2014).
Warrington, Rev. Mod. Phys. 94, 015006 (2022). [58] Z.-X. Li, Y.-F. Jiang, and H. Yao, New Journal of Physics
[30] O. Golan, A. Smith, and Z. Ringel, Phys. Rev. Res. 2, 17, 085003 (2015).
043032 (2020). [59] S. Hesselmann and S. Wessel, Phys. Rev. B 93, 155157
[31] W.-X. Chang and Z.-X. Li, Phys. Rev. B 110, 085152 (2016).
(2024). [60] Z.-X. Li, Z.-Q. Wan, and H. Yao, Asymptotic sign free
[32] Z.-Q. Wan, S.-X. Zhang, and H. Yao, Phys. Rev. B 106, in interacting fermion models (2022), arXiv:2211.00663
L241109 (2022). [cond-mat.str-el].
[33] X. Zhang, G. Pan, X. Y. Xu, and Z. Y. Meng, Phys. Rev. [61] H. Ozawa, S. Taie, Y. Takasu, and Y. Takahashi, Phys.
B 106, 035121 (2022). Rev. Lett. 121, 225303 (2018).
[34] T. Sato and F. F. Assaad, Phys. Rev. B 104, L081106 [62] S. Taie, E. Ibarra-Garcı́a-Padilla, N. Nishizawa,
(2021). Y. Takasu, Y. Kuno, H.-T. Wei, R. T. Scalettar, K. R. A.
[35] M.-S. Vaezi, A.-R. Negari, A. Moharramipour, and Hazzard, and Y. Takahashi, Nature Physics 18, 1356
A. Vaezi, Phys. Rev. Lett. 127, 217003 (2021). (2022).
 7

[63] T. C. Lang, Z. Y. Meng, A. Muramatsu, S. Wessel, [68] M. Gell-Mann, Phys. Rev. 125, 1067 (1962).
and F. F. Assaad, Physical Review Letters 111, 066401 [69] B. Rosenstein, Hoi-Lai Yu, and A. Kovner, Phys. Lett.
(2013), arXiv:1306.3258 [cond-mat]. B 314, 381 (1993).
[64] Z. Zhou, D. Wang, Z. Y. Meng, Y. Wang, and C. Wu, [70] L. Janssen and I. F. Herbut, Phys. Rev. B 89, 205403
Physical Review B 93, 245157 (2016), arXiv:1512.03994 (2014).
[cond-mat]. [71] Y. Otsuka, K. Seki, S. Sorella, and S. Yunoki, Physical
[65] Y.-Y. He, H.-Q. Wu, Y.-Z. You, C. Xu, Z. Y. Meng, and Review B 102, 235105 (2020).
Z.-Y. Lu, Phys. Rev. B 94, 241111 (2016). [72] G. Cassella, P. d’Ornellas, T. Hodson, W. M. H. Natori,
[66] Z.-X. Li, Y.-F. Jiang, S.-K. Jian, and H. Yao, Nature and J. Knolle, Nature Communications 14, 6663 (2023).
Communications 8, 314 (2017). [73] D. E. Burlankov, Theoretical and Mathematical Physics
[67] H. Xu, X. Li, Z. Zhou, X. Wang, L. Wang, C. Wu, and 138, 78 (2004).
Y. Wang, Phys. Rev. Res. 5, 023180 (2023). [74] R. Gilmore, Journal of Geometry and Symmetry in
Physics 28, 1 (2012).
 Supplementary Materials for
Preempting Fermion Sign Problem: Unveiling Quantum Criticality through
Nonequilibrium Dynamics

CONTENTS

I. Details of nonequilibrium imaginary-time dynamics in PQMC simulation 9

A. Imaginary-time relaxation dynamics simulated by PQMC 9
B. Hubbard-Stratonovich transformation 9
C. Sign problem in PQMC 10

II. More details for the single-Dirac-fermion Hubbard model 10

A. Sign problem behaviors 10
B. Verification of the QCP and critical exponents 11

III. More details for the spinless t-V model 12

A. Sign problem behaviors 12
B. Verification of the QCP and critical exponents 13

IV. More details for the SU(3) Hubbard model 14

A. SU(3) algebra 14
B. Mean-field analyses 15
C. Phase boundary and critical exponents 17
D. The new universality class 19
E. Sign problem behaviors and computational efficiency 20
 9

I. DETAILS OF NONEQUILIBRIUM IMAGINARY-TIME DYNAMICS IN PQMC SIMULATION

A. Imaginary-time relaxation dynamics simulated by PQMC

We focus on the imaginary-time relaxation dynamics from a fully ordered or Dirac semimetal initial state |ψ0 ⟩. The
initial state is prepared by solving H0 |ψ0 ⟩ = E0 |ψ0 ⟩, where H0 is the initial Hamiltonian and E0 is the ground state
energy. With these initial states, the evolution of the observable O(τ ) is given by
τ τ
⟨ψ0 | e− 2 H O e− 2 H |ψ0 ⟩
⟨O(τ )⟩ = . (S1)
⟨ψ0 | e−τ H |ψ0 ⟩
τ
As τ → ∞, e− 2 H projects the system onto the ground state of H.
The imaginary-time relaxation dynamics can be simulated via the projector quantum Monte Carlo (PQMC) [51].
In conventional PQMC studies, a sufficiently large τ (usually τ should be several times as large as L) is needed to
ensure that the ground state is obtained. Then the physical qualities are calculated in the ground state. In contrast,
in our work, we focus on the short-time stage of the imaginary-time relaxation process, and τ does not need to be
very large compared with L.

B. Hubbard-Stratonovich transformation

In the PQMC simulations, the interaction terms in the form of four-fermion operators should be decoupled via
Hubbard-Stratonovich (HS) transformation. We at first use the Trotter decomposition to discretize the imaginary-
time evolution operator into M = τ /∆τ (where M is an integer) time slices, i.e.,

M
Y
e−τ H =
 −∆τ Ht −∆τ HU
+ O ∆τ 2 ,


e e (S2)
m=1

where Ht and HU represent the free fermion hopping term and the interaction term in the Hamiltonian, respectively.
Then, we use the HS transformation on HU to decouple the fermion-fermion interaction into interactions between
non-interacting fermions and auxiliary fields.
For the single-Dirac-fermion Hubbard model, we use the following HS transformation:

1 − ∆τ U X λsi (c†i↑ ci↓ +c†i↓ ci↑ )

e−∆τ U (ni↑ − 2 )(ni↓ − 2 ) =
1 1
e 4 e . (S3)
2 s =±1
i

∆τ U
where cosh λ = e 2 . For U > 0 (Hubbard repulsive interaction), the sign problem arises for all channels. We here
choose the σx channel to mitigate the sign problem [52].
For the spinless t-V model, previous studies have shown that at half-filling, the model can be decoupled into the
hopping channel without the sign problem, which has been demonstrated from various perspectives [20, 21, 23, 24].
In our study, however, we use the HS transformation in a sign-problematic channel, namely the density channel:

1 − ∆τ V
e−∆τ V (ni − 2 )(nj − 2 ) =
1 1 X
e 4 eλsij (ni −nj ) . (S4)
2 sij =±1

∆τ V
where cosh λ = e 2 . Despite the presence of sign problem, the numerical results we obtained are consistent with the
previously established results, demonstrating the reliability of our new method.
For the SU(3) repulsive Hubbard model, we use the following HS transformation:

U 2 1 X λsi (niα −niβ )

e∆τ 2 (niα −niβ ) = e . (S5)
2 s =±1
i

∆τ U
where cosh λ = e 2 . For all the known algorithms, the SU(3) repulsive Hubbard model is sign-problematic for any
decoupling channel in HS transformation.
 10

C. Sign problem in PQMC

Through the HS transformation, the Hamiltonian can be converted into a quadratic effective form of fermionic
operators that depends on the spacetime configuration of P the auxiliary fields. The partition function can then be
expressed as a sum of configuration weight w(c), i.e., Z = c w(c). These weights are given by the determinant of
the effective Hamiltonian of fermions [51].
In PQMC simulations, we sample the space-time dependent configuration of the auxiliary field. For sign-free models,
the sampling probability is proportional to the configuration weight w(c). However, for the sign-problematic models,
the configuration weight w(c) is not positive definite, so it cannot be used directly as the sampling probability. Instead,
the absolute value |w(c)| is used as the sampling probability, and the observables are computed as follows [9, 15]:
⟨O⟩|w|
P P P
c w(c)O(c) c |w(c)|sign(c)O(c)/ c |w(c)|
⟨O⟩ = P = P P = . (S6)
c w(c) c |w(c)|sign(c)/ c |w(c)| ⟨sign⟩|w|
Here, we have used
P
c □ |w(c)|
⟨ □ ⟩|w| = P , (S7)
c |w(c)|

to denote the expectation value obtained using |w(c)| as the sampling probability. The sign problem introduces a
cost that the average sign ⟨sign⟩|w| tends to zero due to the frequent cancellation of positive and negative weights
across different configurations, leading to the consequence that (S6) becomes a ratio between two tiny numbers. This
is numerically unstable and introduces significant statistical errors. Specifically, it is proven that the error generally
follows [9]:
1
∆ ⟨O⟩ ∝ ∝ eτ N ∆f , (S8)
⟨sign⟩|w|
where ∆f denotes the difference in the free energy density between the actual fermionic system and its corresponding
bosonic system. The exponential dependence of the error amplification factor on the imaginary time τ and the number
of particles N means that QMC requires exponentially long computational times to achieve controllable statistical
errors when solving ground-state problems of quantum systems in the thermodynamic limit.
In subsequent analyses, we use the average sign ⟨sign⟩|w| to measure the severity of the sign problem. The lower
value of the average sign ⟨sign⟩|w| indicates a more severe sign problem.

II. MORE DETAILS FOR THE SINGLE-DIRAC-FERMION HUBBARD MODEL

A. Sign problem behaviors

a b

FIG. S1. Evolution of the average sign at the critical point. a, Evolution from the DSM initial state. b, Evolution from
the FM initial state.

Fig. S1 shows that near the critical point of the single-Dirac-fermion Hubbard model, the average sign ⟨sign⟩ decays
in the imaginary-time relaxation with the Dirac semimetal (DSM) and the ferromagnetic (FM) initial states. We find
that different initial states can have different decay rates. In the relaxation with the DSM initial state, the average
sign decays more slowly. The data presented in Fig. 1 of the main text show the relaxation results starting from the
DSM initial state.
 11

B. Verification of the QCP and critical exponents

Several physical quantities are needed to describe the universal scaling behaviors near the QCP. The ferromagnetic
structure factor SFM is defined as
1 X ik·(ri −rj ) z z
SFM (k) ≡ e ⟨Si Sj ⟩, (S9)
L2d i,j

where the local spin operator is Siz ≡ c†i σ z ci with c† ≡ (c†↑ , c†↓ ). In addition, the correlation length ratio RFM is
defined as
SFM (k = ∆k)
RFM ≡ 1 − , (S10)
SFM (k = 0)

where ∆k = L1 b1 + L1 b2 is the minimum momentum of electrons in a lattice with periodic boundary conditions, and
b1 , b2 are the reciprocal lattice vectors. The fermion correlation function Gf (k) is defined as
1 X ik(ri −rj ) †
Gf (k) ≡ e ⟨ci↑ cj↓ ⟩ . (S11)
L2 ij

In the main text and figures of data, we abbreviate SFM ≡ SFM (k = 0) and Gf ≡ Gf (k = ∆k), unless otherwise
specified.
The correlation length ratio RFM is a dimensionless quantity. In the nonequilibrium critical region, the correlation
length ratio RFM satisfies the following scaling form:

RFM (g, τ, L) = fR (gL1/ν , τ L−z ), (S12)

where g = U − Uc . To determine the quantum critical point Uc , we fix τ L−z to be constant (e.g., we take τ L−z = 0.3
in the main text), so the scaling form of RFM reduces to R(g, τ, L) = fR1 (gL1/ν ), which is similar to the traditional
finite-size scaling. Accordingly, the critical point can be determined by the intersection of curves of RFM versus U for
different L. We fit Uc and ν based on the expansion of the scaling form:
nX
max

RFM (g, L) = fR1 (gL1/ν ) = an g n Ln/ν . (S13)

n=0

where we appropriately truncate the scaling functions with polynomials [54].

In the main text, based on the results from the DSM initial state, we fit according to Eq. (S13) and obtain
Uc = 7.220(37) and 1/ν = 1.18(3). Here we supplement with results from the FM initial state, as shown in Fig. S2.
Using the same procedure (here we take τ L−z = 0.5), we fit according to Eq. (S13) and obtain Uc = 7.214(44) and
1/ν = 1.05(10), which are close to the results from the DSM initial state. These results not only confirm the values
of the critical point Uc and the critical exponent 1/ν, but also show that the initial states can be chosen flexibly in
our method, which provides a route to achieve reliable values of critical exponents by benchmarking the results with
different initial states.
a b

FIG. S2. Verification of the quantum criticality with FM initial state at τ = 0.5Lz . a, Curves of RFM versus U
intersect for different L. b, Data collapse after rescaling with 1/ν = 1.18(3).
 12

a b c d

FIG. S3. Relaxation dynamics at QCP with FM initial state in single-Dirac-fermion Hubbard model. a-b,
Curves of SFM ≡ SFM (k = 0) versus τ for different sizes before and after rescaling. c-d, Curves of Gf ≡ Gf (k = ∆k) versus τ
before and after rescaling. Data collapse in the relaxation dynamics shows ηϕ = 0.34(5) and ηψ = 0.131(20).

The anomalous dimensions of the bosonic field ηϕ and the fermionic field ηψ can be determined by the following
scaling relations:
 
SFM (k = 0) = L−(1+ηϕ ) fS gL1/ν , τ L−z , (S14)

 
Gf (k = ∆k) = L−ηψ fG gL1/ν , τ L−z . (S15)

We obtain ηϕ = 0.34(5) and ηψ = 0.131(20) from the data collapse in the relaxation dynamics at g = 0, as shown in
Fig. S3.
Table S1 compares the results in the present work with those of other methods. The consistent results shown in this
table demonstrate that the method based on the short-time dynamics can accurately determine the critical properties
of the ground-state QCP with much fewer computational costs.

TABLE S1. Comparison of critical properties for the single-Dirac-fermion Hubbard model calculated by different
methods. The method used in this work is the nonequilibrium short-time PQMC.

Methods Uc ν −1 ηϕ ηψ
z
This work (from DSM, τ = 0.3L ) 7.220(37) 1.18(3) 0.33(2) 0.135(2)
This work (from FM, τ = 0.5Lz ) 7.214(44) 1.05(10) 0.34(5) 0.131(20)
Gutzwiller-PQMC (equilibrium) [52] 7.275(25) 1.19(3) 0.31(1) 0.136(5)
FRG [56] - 1.229 0.372 0.131

III. MORE DETAILS FOR THE SPINLESS t-V MODEL

A. Sign problem behaviors

As shown in Fig. S4, for both DSM initial state and CDW initial state, the average sign in the short-time stage is
close to one, indicating the sign problem is very weak. In particular, for τ = 0.3Lz , as shown in the insets of Fig. S4,
the system almost remains sign-free near the critical point Vc = 1.35(1) determined in the main text.
Incidentally, Fig. S4 shows that the decay rate of ⟨sign⟩ does not monotonically change with the system size L. In
some cases, a larger system size L results in a weaker sign problem. Such phenomena have also been reported in the
equilibrium QMC study [60].
 13

a b

FIG. S4. Evolution of average sign at the QCP (inset shows ⟨sign⟩ versus V near the QCP for τ = 0.3Lz ). a,
Evolution from the DSM initial state. b, Evolution from the CDW initial state.

B. Verification of the QCP and critical exponents

For this model, the ordered phase is the charge density wave (CDW) state, for which the structure factor SCDW
can be defined as:
1 X ik·(ri −rj )
SCDW (k) ≡ 2d e ⟨mi mj ⟩, (S16)
L i,j

where the local order operator is mi ≡ 12 (ni,A − ni,B ), which represents the difference in particle number density
between the two sublattices in a unit cell. The associated correlation length ratio RCDW is defined as:

SCDW (k = ∆k)
RCDW ≡ 1 − . (S17)
SCDW (k = 0)

The fermion correlation Gf (k) is defined similarly to Eq. (S11):

1 X ik(ri −rj ) †
Gf (k) ≡ e ⟨ci,A cj,B ⟩ . (S18)
L2 ij

Here, the sublattice indices A and B function as Dirac spinor indices. In the
main text and figures of data, we
abbreviate SFM ≡ SFM (k = 0) and Gf ≡ Gf (k = K + ∆k), where K = ± 4π 3 , 0 represents the momentum at the
Dirac points.

a b

FIG. S5. Determination of the critical point and 1/ν with DSM initial state. a, At τ = 0.3Lz , curves of RCDW versus
V for different sizes intersect at Vc = 1.37(2). b, Data collapse of curves of RCDW versus rescaled (V − Vc ) with ν = 0.79(5).

In Fig. 2 of the main text, we showed the results of determination of critical properties via the short-time dynamics
from the ordered CDW initial state. In contrast, here we show consistent results can also be obtained from the DSM
initial state.
Fig. S5 illustrates the correlation length ratio RCDW as a function of interaction strength V at τ = 0.3Lz . We fit
the data in Fig. S5a using the scaling form in Eq. S13, determining the intersection point of the curves for different
sizes as Vc = 1.37(2) and the scaling collapse exponent as ν = 0.79(5). These results are close to those obtained
 14

a b c d

FIG. S6. Relaxation dynamics at QCP with DSM initial state in spinless t-V model. a-b, Curves of SCDW ≡
SCDW (k = 0) versus τ for different sizes before and after rescaling. c-d, Curves of Gf ≡ Gf (k = K + ∆k) versus τ before and
after rescaling. Data collapse in the relaxation dynamics shows ηϕ = 0.44(2) and ηψ = 0.072(4).

by equilibrium methods [57, 58] and the short-time dynamics with ordered CDW initial state discussed in the main
text. Fig. S6 shows the relaxation dynamics of SCDW = SCDW (k = 0) and Gf = Gf (k = K + ∆k) starting from
the DSM initial state. Their scaling behavior is described by Eq. (S14) and Eq. (S15). According to the scaling
collapse of the evolution of SCDW in Fig. S6a-S6b, one finds ηϕ = 0.44(2). In addition, from the scaling collapse of the
evolution of Gf in Fig. S6c-S6d, we obtain ηψ = 0.072(4). The value of ηϕ is close to that obtained by the equilibrium
methods [57, 58] and the short-time dynamics with ordered CDW initial state discussed in the main text; while the
value of ηψ is close to the FRG result [56], and also consistent with that obtained from the CDW initial state shown
in the main text. Consequently, the results of critical properties in the t-V model further demonstrate the efficiency
and accuracy of our method. More crucially, for the first time, through our method we achieve the reliable result of
fermionic anomalous dimension ηψ for the QCP in the spinless t-V model by unbiased QMC simulation, as shown in
Table S2.

TABLE S2. Comparison of critical properties for the spinless t-V model calculated by different methods. The
method used in this work is the nonequilibrium short-time PQMC.

Methods Uc ν ηϕ ηψ
z
This work (from CDW, τ = 0.3L ) 1.35(1) 0.77(12) 0.49(5) 0.073(4)
This work (from DSM, τ = 0.3Lz ) 1.37(2) 0.79(5) 0.44(2) 0.072(4)
Majorana QMC (equilibrium) [58] 1.355(1) 0.77(2) 0.45(2) -
Continuous-time QMC (equilibrium) [57] 1.356(1) 0.80(3) 0.302(7) -
FRG [56] - 0.929 0.602 0.069

IV. MORE DETAILS FOR THE SU(3) HUBBARD MODEL

A. SU(3) algebra

The SU(3) Hubbard model with staggered flux is invariant under SU(3) transformations in the flavor space of
fermions. Here we give a brief review on the SU(3) algebra.

The generators of the SU(3) group are represented by the well-known Gell-Mann matrices. Below are the eight
 15

Gell-Mann matrices, corresponding to the generators of SU(3) group:

     
0 1 0 0 −i 0 1 0 0
λ1 = 1 0 0 , λ2 =  i 0 0 , λ3 = 0 −1 0 ,
0 0 0 0 0 0 0 0 0
     
0 0 1 0 0 −i 0 0 0
λ4 = 0 0 0 , λ5 = 0 0 0  , λ6 = 0 0 1 , (S19)
1 0 0 i 0 0 0 1 0
   √ 
0 0 0 1/ 3 0√ 0
λ7 = 0 0 −i , λ8 =  0 1/ 3 0√  .
0 i 0 0 0 −2/ 3

The Lie algebra structure of the SU(3) group is determined by the commutation relations of its generators. The
commutation relations between the generators of the SU(3) group are listed as follows:
X
[λa , λb ] = 2i fabc λc , (S20)
c

where fabc are the structure constants, specifically:

f123 = 1, (S21)

1
f147 = f246 = f257 = f345 = , (S22)
2

1
f156 = f367 = − , (S23)
2
√
3
f458 = f678 = . (S24)
2
These commutation relations will determine the manifold shape of the ground-state degeneracy space of the ordered
phase, which we will further analyze with numerical evidence in the subsequent sections.

B. Mean-field analyses

To qualitatively understand the salient features of ground-state phase diagram and the dominant symmetry spon-
taneous breaking ordering in the SU(3) Hubbard model, we first perform a mean-field analysis. We rewrite the
interaction term in the following form and apply the mean-field approximation:
!2
UX X 3 U XX
niα − = niα niα′ + const.
2 i α
2 2 i ′
α,α
U XX 2
=− (niα − niα′ ) + const.
4 i
α,α′ (S25)
3U X X † 2
=− ci λn ci + const.
16 i n
3U X
(−1)i c†i λn ci + const. ,
X
≈− ⟨mλn -AFM ⟩
16 n i

where c†i ≡ c†i1 c†i2 c†i3 , and λn =(λ1 , λ2 , . . . , λ8 ) are the eight Gell-Mann matrices. The mean-field order parameter


is defined as
1 X
⟨mλn -AFM ⟩ ≡ 2 (−1)i ⟨c†i λn ci ⟩ (S26)
L i
 16

DSM

λ 8-AFM

FIG. S7. Mean-field results for the λ8 -AFM order. a, For ϕ = 0.1π, L = 30, a DSM-AFM transition in the mean-field
approximation is seen at U = 3.20(1). b, Mean-field phase diagram with L = 30.

After applying the mean-field approximation, the Hamiltonian is entirely expressed as a quadratic form of fermionic
operators and can be solved for the ground state using exact diagonalization. The problem is then simply reduced
to a self-consistent calculation of the order parameter {mn }. Fig. S7a shows the variation of the order parameter m8
with interaction strength U calculated using the mean-field method. When the interaction is strong, m8 starts to
increase with U , showing clear characteristics of a continuous phase transition. We control different magnetic flux ϕ
to find the corresponding phase boundary U , as shown in the mean-field phase diagram in Fig. S7b.
We can understand why the λ8 -AFM order is preferred by examining the ground state energy of the system.
Consider placing the eight different order parameter operators in the same external field h:

(−1)i c†i λn ci ,
X
Hnsat = −h (S27)
i

whose ground state corresponds to the saturated λn -AFM ordered state. At half-filling, the ground state energy of
H8sat (h) is − √23 hL2 . The energy gap at half-filling, between the ground state of H8sat (h) and the first excited state
of H8sat (h), is a finite value √23 h. For Hn̸sat 2
=8 (h), the ground state energy is −hL and the energy gap is 0. In other
words, even considering the saturated ordered state, only the λ8 -AFM order can open a gap in Dirac fermions, while
other types of λn -AFM cannot.
Ground state energy per site

2.5

2.6

2.7
8
n 8
2.8
0 1 2 3 4 5 6 7 8
U
FIG. S8. Mean-field ground state energy per site at half-filling of different types of λn -AFM. For the mean-field
calculation with ϕ = 0.1, L = 30, the ground state energy of the λ8 -AFM mean-field Hamiltonian is lower than that of other
types of Hamiltonians.

Next, we calculate the mean-field ground-state energy for different λn -AFM orders. From Fig. S8, we find that
 17

λ8 -AFM mean-field states have lower ground-state energy than other types of antiferromagnetic orders in the whole
interaction regime of the ordered phases. Thus, from a mean-field perspective, the system favors the λ8 -AFM order.

C. Phase boundary and critical exponents

Here, we perform QMC simulation through short-time relaxation to systematically study the phase boundary and
critical properties of the quantum phase transition. To determine the phase boundary of the SU(3) Hubbard model
with staggered flux, we compute the structure factor and correlation length ratio. For the λ8 -AFM order, the structure
factor is defined as:
1 X ik·(ri −rj )
Sλ8 -AFM (k) = e (−1)i+j ⟨c†i λ8 ci c†j λ8 cj ⟩ , (S28)
L2d i,j

The associated correlation length ratio is defined as

Sλ8 -AFM (k = ∆k)

Rλ8 -AFM ≡ 1 − . (S29)
Sλ8 -AFM (k = 0)

The fermion correlation Gf (k) is defined the same as Eq. (S18):

1 X ik(ri −rj ) †
Gf (k) ≡ e ⟨ci,A cj,B ⟩ , (S30)
L2 ij

where A, B represent two inequivalent sublattices in a square lattice with staggered flux. In the main text and data
figures, unless otherwise specified, we omit
the momentum variables and denote Sλ8 -AFM ≡ Sλ8 -AFM (k = 0) and
Gf ≡ Gf (k = K + ∆k), where K = ± π2 , ± π2 represents the momentum at the Dirac points.

a b

c d

FIG. S9. Curves of Rλ8 -AFM versus U for different sizes with λ8 -AFM initial state at τ = 0.25Lz . a-b, For
ϕ = 0.05π, the fitted critical point is Uc = 0.347(4), with the critical exponent 1/ν = 0.67(1). c-d, For ϕ = 0.1π, the fitted
critical point is Uc = 2.0(3), with 1/ν = 0.58(7).

To identify phase boundary between the DSM phase and the λ8 -AFM phase, we prepare a λ8 -AFM initial state and
calculate the critical point for fixed ϕ by the method of short-time dynamics. For fixed τ L−z = 0.25. We calculate
the critical point for ϕ = 0.05π, ϕ = 0.075π and 0.1π by the intersection points of curves of Rλ8 -AFM versus U for
different L. As shown in Fig. S9, critical points are Uc = 0.347(4) for ϕ = 0.05π, Uc = 2.0(3) for ϕ = 0.1π, and
Uc = 1.10(5) for ϕ = 0.075π as shown in the main text.
 18

a b c

d e f g

FIG. S10. Correlation-length ratios for λ1 -AFM to λ7 -AFM versus U for different sizes with λ8 -AFM initial
state at τ = 0.25Lz , ϕ = 0.075π. There is no crossing point in the curves of Rλn -AFM versus U .

Here, to further confirm the λ8 -AFM is the dominant ordering, we also compute the correlation length ratio for
other quantities. We find that there is no crossing in the curves of Rλn -AFM for the AFM orders defined by λ1 to λ7
versus U , as shown in Fig. S10. The results of the correlation length ratio decrease with L, demonstrating the absence
of long-range order. Hence, the λ8 -AFM is the dominant ordering, and other AFM orders are all short-range in the
half-filled SU(3) Hubbard model with staggered flux.
Moreover, scaling collapse for the curves of Rλ8 -AFM versus rescaled (U − Uc ) gives the value of 1/ν. Accordingly,
we obtain 1/ν = 0.67(1) and 1/ν = 0.58(7) for ϕ = 0.05π and ϕ = 0.1π, respectively. Combining 1/ν = 0.68(5) with
ϕ = 0.075π shown in the main text, we find that the values of 1/ν are consistent with each other within error bar,
showing the universality of the quantum phase transition between DSM and λ8 -AFM ordered phase.
To determine the anomalous dimensions ηϕ and ηψ of the bosonic field and the fermionic field we study the relaxation
dynamics of the structure factor Sλ8 -AFM = Sλ8 -AFM (k = 0) and the fermion correlation Gf = Gf (k = K + ∆k) at
the critical point. The curves before and after rescaling are shown in Figs. S11 and S12 and Fig. 3 in the main
text for different ϕ. The critical exponents determined above are summarized in Table S3, from which we find that
the anomalous dimensions for different ϕ are consistent with each other within error bar, further confirming the
universality of the phase transition.

a b c d

FIG. S11. Relaxation dynamics at the QCP ϕ = 0.05π, U = 0.347(4) with λ8 -AFM initial state in SU(3) Hubbard
model. a-b, Curves of S versus τ for different sizes before and after rescaling. c-d, Curves of G versus τ before and after
rescaling. Data collapse in the relaxation dynamics shows ηϕ = 0.51(3) and ηψ = 0.16(2).
 19

a b c d

FIG. S12. Relaxation dynamics at the QCP ϕ = 0.1π, U = 2.0(3) with λ8 -AFM initial state in SU(3) Hubbard
model. a-b, Curves of S versus τ for different sizes before and after rescaling. c-d, Curves of G versus τ before and after
rescaling. Data collapse in the relaxation dynamics shows ηϕ = 0.58(10) and ηψ = 0.16(3).

TABLE S3. Critical properties for the SU(3) Hubbard model probed under different staggered flux ϕ.

ϕ Uc ν −1 ηϕ ηψ
0.05π 0.347(4) 0.67(1) 0.51(3) 0.16(2)
0.075π 1.10(5) 0.68(5) 0.55(5) 0.15(3)
0.1π 2.0(3) 0.58(7) 0.58(10) 0.16(3)

D. The new universality class

Our numerical results demonstrate the presence of a continuous phase transition which does not belong to any of
known universality classes [71, 72]. Instead, it represents a new universality class.
In the saturated λ8 -AFM phase, two flavors of fermions (referred to as flavors 1 and 2) are localized in one
sublattice (designated as the A lattice), while the remaining flavor of fermions (referred to as flavor 3) is localized
in the other sublattice (designated as the B lattice), as illustrated in Fig. 3a of the main text. It is evident that
the Z2 symmetry between the AB sublattices is significantly broken. The diagonal λ8 generator induces a global
U(1) transformation, while the generators λ1 , λ2 , λ3 act solely on the subspace of flavors 1 and 2, generating a closed
SU(2) transformation that only mixes these two flavors. Since these four generators all commute with the order
parameter operator, the SU(2) × U(1) is the largest symmetry group that remains invariant under the λ8 -AFM order
parameter m after spontaneous symmetry breaking. The other four generators of the SU(3) group, λ4 , λ5 , λ6 , λ7 ,
produce transformations that alter the direction of the AFM order parameter {mn } in a compact manifold, leading to
other λ8 -AFM degenerate ground states, which are four independent Goldstone modes. All degenerate ground states
span a 4-dimensional ground state degeneracy space, and these states can be mapped one-to-one onto points on the
manifold [SU(3) × Z2 ] / [SU(2) × U(1)] [73, 74].
The critical exponents determined via the short-time dynamics in our work are summarized in Table S4 and
compared with those of other Gross-Neveu universality classes with the same fermion components. It is clear that
the new universality class we have discovered is distinct from the previously known classes.

TABLE S4. Comparison of critical exponents for different Gross-Neveu-Yukawa universality classes in d = 2 + 1
with Nf = 6. The first row, chiral [SU(3) × Z2 ]/[SU(2) × U(1)] denotes the universality class for SU (3) Dirac fermion Hubbard
model, with exponents determined from nonequilibrium PQMC.

Universality class ν −1 ηϕ ηψ
chiral [SU(3) × Z2 ]/ [SU(2) × U(1)] (this work) 0.68(5) 0.55(5) 0.15(3)
chiral Heisenberg (4 − ϵ, 2nd order)[69] 1.478 1.023 0.058
chiral XY (4 − ϵ, 2nd order)[69] 1.809 0.698 0.082
chiral Ising (4 − ϵ, 2nd order)[69] 0.750 0.865 0.011
chiral Ising (FRG)[70] 0.993 0.912 0.013
 20

E. Sign problem behaviors and computational efficiency

a b c

FIG. S13. The sign problem in SU(3) Hubbard model at the QCP ϕ = 0.1, U = 2.0(3). a, The average sign
with different L and τ in short-time stage. b, Comparison for the average sign between short-time stage (τ = 0.25Lz ) and
equilibrium stage (τ = 1.5Lz ). c, Efficiency gain of QMC with the short-time method (τ = 0.25Lz ) compared to equilibrium
QMC (τ = 1.5Lz ) for different sizes L.

For this model, we set the parameters at the critical point U = 2.0(3), ϕ = 0.1π. The average sign as a function of
imaginary time τ and size L is shown in Fig. S13a. The sign problem for this model is significantly more severe than
for the previous two models. For τ = 0.25Lz , the average sign is approximately 10−1 ∼ 10−2 , meaning that compared
to the sign-problem-free case, 10 to 100 times more computational resources are needed to obtain reliable results.
From Figs. S11 and S12 and Fig. 3 in the main text, it can be seen that τ = 0.25Lz is in the nonequilibrium scaling
region controlled by the critical point. Even though the ground state is not reached, its nonequilibrium scaling still
reflects the quantum criticality of the ground state. In equilibrium QMC studies, it is typically necessary to set the
imaginary time τ to more than 1.5 times the size Lz to reach the ground state. However, evolving for such a long time,
the average sign decays to approximately 10−5 ∼ 10−6 . In Fig. S13b, we compare the average sign for τ = 0.25Lz
and τ = 1.5Lz . Since the computational error (according to Eq. S8) is inversely proportional to the average sign
⟨sign⟩ (according to Eq. S8), we can measure the difference in computational efficiency by multiplying the ratio of the
average sign ⟨sign⟩ by the length of imaginary-time evolution, as follows:

1/ ⟨sign⟩eq. τeq.
Efficiency acceleration factor = × , (S31)
1/ ⟨sign⟩neq. τneq.

where the nonequilibrium evolution time is τneq. = 0.25Lz , and the equilibrium evolution time is τeq. = 1.5Lz .
⟨sign⟩neq. and ⟨sign⟩eq. are their corresponding average signs, as shown in Fig. S13b. We specifically compared the
differences in computational efficiency for several system sizes, as shown in Fig. S13c. The computational resources
required by the nonequilibrium method are only a few millionths of those required by the equilibrium method, and this
efficiency improvement roughly grows exponentially with the system size. Consequently, our nonequilibrium method
enables the QMC simulation on SU(3) Hubbard model with large system size, which is not accessible in previous
unbiased numerical approaches.