Critical analysis of two-dimensional classical XY model

Raghav G. Jha
Perimeter Institute for Theoretical Physics
31 Caroline Street North
Waterloo, ON, Canada

Abstract

We consider the two-dimensional classical XY model on a square lattice in the thermody-

arXiv:2004.06314v2 [hep-lat] 30 Jun 2020

namic limit using tensor renormalization group and precisely determine the critical tem-
perature corresponding to the Berezinskii-Kosterlitz-Thouless (BKT) phase transition to
be 0.89290(5) which is an improvement compared to earlier studies using tensor network
methods.

Contents

1 Introduction 1

2 Phase transition, Tensor construction, and Observables 3

3 Results & Conclusion 6

1. Introduction

The XY model, also known as planar rotor model is one of the most widely studied and
the simplest two-dimensional spin model1 with continuous symmetry because of the inter-
esting features it possesses and the possibility of analytical treatment using some reasonable
approximations. The model has U(1) or O(2) symmetry corresponding to the rotation on
a two-dimensional plane. The spins on each lattice site interact via a nearest neighbour
Hamiltonian. The model has numerous practical applications in studies related to superfluid
helium, thin-films, superconductivity, liquid crystals, melting of two-dimensional crystals,
and dielectric plasma transition from a dielectric phase (charges bound as neutral dipoles)
to the conducting phase (free charges).
The seminal work by Berezinskii and subsequently followed by Kosterlitz and Thouless
elaborated on the features of this model and provided an understanding of the infinite order
phase transition without any order parameter. This phase transition does not accompany

1
The history of this model is not well established but it appears that it was first discussed by Nambu in
1950 [1] and was named XY for the first time by Lieb et al. [2]

1
any symmetry breaking and from the point of view of the Ginzburg-Landau description
of phase transitions, this is a very special case. The change in the functional form of
the correlation functions across the critical temperature signals the phase transition. The
correlation length obeys an exponential dependence above the phase transition unlike the
familiar power-law and it is infinite below the critical temperature. Any determination of
the critical temperature by probing the system close to the critical temperature (i.e. TBKT )
is a challenging numerical problem. For temperatures below this critical value, there is a
line of fixed points down to zero temperature. This is a special property that it shares for
example with supersymmetric N = 4 Yang–Mills in four dimensions. The XY model at
zero temperature in the absence of an external field has all spins aligned in a particular
direction corresponding to a ferromagnetic phase and shows long-range order while at high
temperatures it is in a symmetric paramagnetic phase in which all the directions on the
plane are equivalent for all spins.
For many years, it was common folklore that two-dimensional nearest-neighbour spin
systems cannot have phase transitions but the possibility of a finite temperature phase
transition inferred by extrapolation from the high-temperature expansion was pointed out
by Stanley and Kaplan in [3]. This led to interesting developments eventually culminating
in a series of works [4–6] after which many groups continued studying this model in great
detail. Even though the two-dimensional XY model can be understood to a great extent
through analytical arguments based on mean-field theory, renormalization group methods,
and by maps to other models in the same universality class (such as Coulomb gas, SOS
(solid-on-solid) model), the exact determination of the critical temperature on some fixed
lattice remains much a numerical problem. The study of the critical behaviour of the two-
dimensional XY model has now been pursued for over four decades using high-temperature
expansions (HTE), Monte-Carlo (MC) numerical approach, Hamiltonian strong coupling
expansion and most recently using efficient tensor network (TN) algorithms. However, there
is no general agreement on the critical temperature corresponding to the BKT transition
and the available results differ from one other by several error bars.
In the majority of these explorations, MC approach (single-flip Metropolis) was applied.
However, this computational approach is prone to critical slowing down around the critical
temperature and makes it highly inefficient. Several improved algorithms such as cluster
algorithms2 where a group of spins is updated at once were then formulated and was then
used with parallel computations to locate the phase transition on large systems close to the
thermodynamic limit. Though until a few decades ago, the most preferred tool to study
lower-dimensional (d ≤ 2) statistical systems were large-scale MC simulations, this is now
giving way to tensor network techniques with two such studies already existing in the litera-
ture [7, 8] for the model we consider in this work. The reason of this paradigm shift is clear –

2
For example, these include the Swendsen-Wang, Wolff, and embedded cluster algorithms

2
in the last decade many efficient real-space tensor network algorithms have been developed
which have enabled studies of the critical and off-critical statistical systems in the thermo-
dynamic limit by carefully exploiting the interesting region of the otherwise huge Hilbert
Space. In the tensor network description, the physical properties of the system is encoded in
the local tensors and by performing the coarse-graining sufficiently many times, we can un-
derstand the global properties in a more intuitive way. Tensor networks have also been used
to study spin-foam models in quantum gravity in various dimensions. They are also starting
to play a crucial role in studying real-time path integrals which are otherwise impossible
to study using MC because of highly oscillatory form of integrals for which the importance
sampling method fails. The tensor networks have also been useful in understanding some
aspects of holography and understanding the connections between the many-body systems
and gravity. For a review on tensor networks and its various applications, see [9]. It is now
also becoming evident that these methods can also be used to study gauge theories and
some works along this direction are [10–14] 3 .
In this paper, we take a step in this direction by applying a real-space coarse-graining
algorithm and systematically determining the critical temperature for the topological tran-
sition in the XY model. By keeping a bigger fraction of the local space than achieved
before, we have considerably improved on the earlier works. This resulted in our numerical
estimate as being the most precise achieved to date using any tensor network approach. It
also agrees with the most precise available MC result [15]. We introduce a small external
magnetic field that breaks the O(2) symmetry to determine the critical temperatures at
small field strengths by looking at the response of the network to the field by computing
the magnetization and susceptibility. We now briefly outline the structure of the paper. In
Sec. 2, we discuss the phase transition and lack of ordering in two dimensions and explain
the tensor construction of the model before discussing the observables we study. In Sec. 3,
we present the results for the free energy, magnetization, and determine the critical tem-
perature from the peak of the susceptibility. We put down some numerical details in the
Appendix for the convenience of the reader.

2. Phase transition, Tensor construction, and Observables

In two dimensions, a no-go theorem due to Mermin-Wagner-Hohenberg-Coleman (MWHC)

states that it is not possible to have an ordered phase (long-range order) 4 which implies
that there is no possibility of a phase transition from a disordered to an ordered phase
accompanied by spontaneous symmetry breaking at any finite temperature. In high energy

3
It is impossible to list the extended literature along these lines and we apologize in advance for all
omissions.
4
We mean the absence of long-range order in the sense explained by Polyakov in [16]. If one considers
a large but finite system and fixes the value of S~ at the boundary to be 1, the average value of hSi~ inside
the system vanishes as system size goes to infinity. This is equivalent to the fact that the angular deviation
between spins increases with r i.e. h(θ(r) – θ(0))2 i ∼ log(r).

3
theory language, this is usually expressed as - there are no Goldstone bosons in two di-
mensions. The absence (or destruction) of long-range order is due to the strong low-energy
fluctuations. Since there is no long-range order, there is no true order parameter. However,
there does exist a well-known phase transition where in the absence of an order parameter,
the task of classification of the phases depends on the behaviour of the correlation function.
In the low-temperature phase of the XY model, there is a quasi-long-range order (QLRO)
phase and we have:
1
hS(r) · S(0)i ∼ e–η ln r =
, (1)
rη
while in the high-temperature disordered phase, we have:

hS(r) · S(0)i ∼ e–r/ξ , (2)

where S(r) is a planar spin of magnitude one on site r = (x, y) of a square lattice, ξ is the
correlation length and the exponent η = T /(2πJ)5 is proportional to the temperature and is
equal to its maximum value of 1/4 at TBKT . This power-law behaviour is not just at critical
point but for all temperatures less than TBKT . The QLRO is destroyed by unbinding of the
vortex and anti-vortex pair at some critical temperature TBKT , which is now well-known
as the Berezinskii-Kosterlitz-Thouless (BKT) transition. This transition has a topological
character associated with the proliferation of vortices. The BKT transition is a very special
case of bypassing the MWHC theorem which rests on the fact that for temperatures T <
TBKT , the algebraic (‘power-law’) decay of correlations functions means that there is no long
range order but only QLRO where topological excitations/defects (vortex and anti-vortex
with opposite topological charges) are bound together. At T > TBKT , defects unbind and
starts to grow rapidly and results in exponential decay (also referred to as ‘screening’) given
by:  h i
-C
∼ exp √
τ
, T > TBKT
ξ= , (3)
∞, T < TBKT
where τ = (T – TBKT )/TBKT and C is a positive constant. We now consider a square lattice
with a spin on each lattice site Si . The Hamiltonian of the XY model is given by:
X X
H = –J cos(θi – θj ) – h cos θi , (4)
hiji i

where hiji denotes the summation over the nearest neighbouring sites i and j, θi is the spin
angle at site i, and we have set the modulus |Si | = 1 with J being the exchange coupling
between the nearest neighbour spins, and h the applied external magnetic field. We will
set J = 1 in what follows while the Boltzmann’s constant, kB , has already been set to 1.
The Hamiltonian (4) with h = 0 has global U(1) symmetry: θi → θi + c which amounts

5
The was first noted down by Peierls and Landau in 1935 and 1937 respectively

4
to changing the angle by same amount at each lattice site and also has periodicity i.e.
θi → θi + 2πni . We can write the partition function of the the model as:
Z Y !
1 X X
Z= dθi exp β cos(θi – θj ) + βh cos θi , (5)
2π
i hiji i

where β = 1/T is the reciprocal of temperature and angles vary continuously in the interval
[0, 2π)6 . In order to construct the tensor representation, we decompose the Boltzmann factor
(for h = 0) in a basis (also known as Jacobi-Anger expansion) as follows:
∞
!
X
exp β cos(θi – θj ) = I0 (β) + Iν (β)eiν(θi –θj ) , (6)
ν=–∞,6=0

where Iν (β) is the modified Bessel function of the first kind. The partition function can
then be written as [17]:
Z Y Y
Z= dθi Iνij (β)Iµi (βh)eiνij (θi –θj )+iµi θi . (7)
i νij ,µi

By performing integration over the physical angular degrees of freedom θi , we proceed to

define a tensor (dual) as:
i+k
for h=0, this enforces δj+l
q z }| {
Tijkl = Ti,j,k,l = Ii (β)Ij (β)Ik (β)Il (β) Ii+k–j–l (βh), (8)

where indices (i, j, k, l) denote the four legs of the tensor7 and run from –∞, · · · , ∞. The
size of each leg, which is infinite in principle from the expansion formula in (6) is truncated
down to run only in range [d–χ/2e , · · · , bχ/2c], where we refer to χ as bond dimension. The
reason this truncation is reasonable is that the series expansion coefficient Iν (β) decreases
quickly with increasing ν for all β considered in this work. Thus, we can truncate the series
and approximate Tijkl by a tensor with finite bond dimension with relatively high precision.
However, the errors due to this truncation can depend strongly on the temperature and the
external field h as we will see later. This leads to a finite-dimensional tensor representation
for the partition function: Y 
Z = tTr Tijkl , (9)
where tTr denotes the tensor trace and product is over all sites. It is well-known that in
absence of magnetic field, the magnetization i.e. M = hS~x i = cos(θ(r)) is zero in d = 2
i
for all non-zero temperatures. The average magnetization in presence of external field is
evaluated as:
!
–∂F 1 ∂ ln Z
q Ii+k–j–l–1 (βh) + Ii+k–j–l+1 (βh)
M= = = tTr Ii (β)Ij (β)Ik (β)Il (β) .
∂h β ∂h 2
(10)

6
The model where this interval is split in Q-intervals is known as the Q-state Potts model which for
Q → ∞ is the XY model
7
We always use the order: left, right, up, down to denote a rank-four tensor in this work

5
 j j′

Ii+ j− k− l− 1(βh) + Ii+ j− k− l+ 1(βh)

( )
𝚃˜ 𝚒𝚔𝚓𝚕 = Ii(β)Ik(β)Ij(β)Il(β) i k k i
2

l l′
l l′

i′ k′ k′ i′

j j′
= M
j

i k
𝚃𝚒𝚔𝚓𝚕 = Ii(β)Ik(β)Ij(β)Il(β)Ii+ j− k− l(βh)

Figure 1: The ‘observable’ (sometimes also called ‘impure’) tensor shown in orange is inserted
in place of one of the usual (blue) tensors. We then evaluate magnetization by dividing the
contraction of the network in the numerator and the denominator. The contraction is over the
same index label and the ordering and label of indices in the denominator is the same as in the
numerator.

Since the system has translational invariance, the average magnetization is the same for all
spins. In principle, one can evaluate the magnetization by taking the numerical derivative
of the logarithm of the partition function with respect to external field as in (10), but,
this method is prone to errors and is not suited for determining the critical temperature.
We evaluate the magnetization by inserting an additional tensor in the tensor network as
described in detail in Figure 1. The response function of the system when the external field
is turned on is given by magnetic susceptibility defined as:

∂M 
S=  . (11)
∂h 
T
.

3. Results & Conclusion

We use higher-order tensor renormalization group (HOTRG) [18] using periodic bound-
ary conditions to study this model. This real-space coarse-graining scheme has been very
useful in studying quantum and classical statistical systems in various dimensions [19, 20].
Instead of coarse-graining along a constant time-slice and then constructing the transfer
matrix, we have used simultaneous coarse-graining along x and y direction. We then use
the singular value decomposition (SVD) to truncate the tensor back to the original size while

6
 0.4

0.6

0.8

1
f
1.2

1.4

1.6 h=0
h = 0.1
h = 0.5
1.8
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
T

Figure 2: Free energy against temperature for different values of magnetic field. The system size
is 28 × 28 and χ = 25.

introducing a small error in the description of the system. This truncation becomes severe
with decreasing h. We first calculate the free energy (using 12) and show the its variation
with temperature for different external magnetic field h in Figure 2. Unlike in the Monte
Carlo simulations, the free energy (or density) is the simplest quantity to measure in the
tensor network formulation. We checked that the free energy at T = 1.0 is the same up to an
error of 10–4 whether (Nx Ny , χ) is (260 , 47) or (216 , 25)8 . One important feature of the phase
transition in this system is that all derivatives of the free energy are zero as T → Tcritical and
there is no way of locating the transition by looking at any of its derivatives since there are
no discontinuities at the critical temperature. This makes this system an exception to the
usual Landau-Ginzburg paradigm. They belong to the class of topological phase transitions
that are not characterized by a Landau order parameter. Our determination of the critical
temperature was done by looking at S for different h and then taking the lim of the critical
h→0
temperature using an ansatz (see Appendix). In an external magnetic field h, the magneti-
1
zation behaves like M ∼ h δ at the critical temperature and the critical exponent describing
the response to the field i.e. δ is given by δ = (d + 2 – η)/(d – 2 + η) = 15. We evaluate
the magnetization for various h in the thermodynamic limit (lattice of size 250 × 250 ) and
show a small subset of those measurements in Figure 3. In order to determine the critical
temperature, we find out the temperature corresponding to the peak of susceptibility for
different field strengths as shown in Figure 4. We then extrapolate our numerical data and

8
Nx Ny denotes the number of sites along the x and y direction respectively. One of these is about 200
times faster than the other even without worrying about the memory

7
 0.6

0.5

0.4

M 0.3

0.2

0.1 h = 1e-8
h = 1e-6
h = 1e-4
0
0.95 1 1.05 1.1 1.15 1.2

Figure 3: The dependence of magnetization on inverse temperature for different h.

find Tcritical = 0.89290(5) in the limit of vanishing field shown in Figure 5. In order to
precisely determine the critical temperature, it was essential to go to very small values of
the external field which in turn was strongly dependent on the choice of χ. We initially
started with χ = 41 and found that it was not possible to explore less than h ≈ 10–7 while
accurately determining the peak of susceptibility. We believe it was fortuitous that this
bond dimension (χ) was sufficient to determine the critical temperature reasonably well and
critical exponent to a within few percent of accuracy in [7]. For instance, even with χ = 53,
we could not precisely determine the critical exponent, δ, but certainly been able to reduce
few orders of magnitude of error on the previous estimate of Tcritical . As compared to earlier
work, we kept almost double the number of states at each coarse-graining step because of
using a slightly improved version of HOTRG and memory available. These extra states were
crucial to explore much smaller magnetic fields down to h = 10–15 compared to approxi-
mately h = 10–7 in [7]. For instance, in [7], the critical temperature for h = 1.5 × 10–7 was
found to be T = 0.9172 while we found T = 0.9215(3). This is a clear sign that increasing
χ on basically similar lattice size still helps the precision as one probes lower temperature
and goes to zero field limit.
We now briefly mention some earlier works locating the critical temperature for this
model (see Table 1 for the list). Very recently, using uniform MPS and a well-defined
method of extracting correlation length for discrete system as given in [27], this model was
studied in [8] and Tcritical = 0.8930(1) was found. This was an improvement over the old
work using tensor networks which found Tcritical = 0.8921(19) [7]. This model was also
studied in [15] using 256 GPUs on a 65536 (= 216 ) × 65536 lattice using Monte Carlo.

8
 700000

650000

600000

550000

500000
𝕊
450000

400000
h = 1.5 × 10−7
350000 Tcrit. = 0.92150(25)
300000

250000
0.91 0.915 0.92 0.925 0.93 0.935
T

Figure 4: A representative example of the determination of the critical temperature for a given
magnetic field. The vertical shaded band denotes the error in the estimate of the critical tem-
perature.

Method Year System Size Tcritical

Monte Carlo [21] 1992 29 × 29 0.89400(500)

HTE [22] 1993 – 0.89440(250)
Monte Carlo [23] 1995 28 × 28 0.89213(10)
Monte Carlo [24] 2005 211 × 211 0.89294(8)
HTE [25] 2011 – 0.89286(8)
Monte Carlo [15] 2012 216 × 216 0.89289(5)
Monte Carlo [26] 2013 29 × 29 0.89350(10)
Higher-order TRG [7] 2013 240 × 240 0.89210(190)
Uniform MPS [8] 2019 – 0.89300(10) TN
Higher-order TRG [This work] 2020 250 × 250 0.89290(5)

Table 1: The estimate of the temperature for the BKT transition in classical XY model with
different methods. The three most recent estimates have all been done using real-space renor-
malization with tensor networks (TN).

9
 1
0.99
0.98
0.97
0.96
0.95
Tcrit. 0.94
0.93
0.92
0.91
0.9
0.89
0.88
19 16 13 10 7
1 ⇥ 10 1 ⇥ 10 1 ⇥ 10 1 ⇥ 10 1 ⇥ 10 0.0001
h

Figure 5: The dependence of critical temperature on external magnetic field h. We fit the data
to the ansatz and obtain lim Tcrit. = 0.89290 (5). The errors have been scaled by a factor of 10
h→0
for visibility on the plot and determined as in Figure (4). The data is for square lattice of size
250 × 250 with χmax. = 53.

Some earlier studies differ significantly from our estimate and we guess that it might be
because the lattice is too small. It was shown in [8] that already around T = 0.95, the
correlation length can be more than a thousand sites. We note that among all these, the
numerical study using Monte Carlo on the largest lattice is indeed also the most precise
estimate to date which we have been able to match using tensor networks in this work.
Though the procedure in this paper is similar in spirit to the one in [7], we have been able
to refine the results in that paper by going to a large bond dimension. For instance, for
h ∼ 1.5 × 10–4 , the peak appears at T = 0.98 in the earlier work while we found that the
peak is at T = 0.9810(5). A more prominent difference is observed for h ∼ 1.5 × 10–7 , where
earlier study found T = 0.9172 while we deduced T = 0.9215(3).
In this work, the estimate of the critical temperature is the most precise that has yet been
achieved using any tensor network approach and are level with precision from Monte Carlo
methods. In the preliminary stages, we also implemented the tensor network renormalization
(TNR) [28] algorithm which removes the short-range entanglement by using disentanglers
and makes the study at and near criticality more efficient compared to the other standard
numerical RG approaches. We found that while the results for free energy and magnetization
obtained using TNR are consistent with HOTRG, the complexity and the scaling of the
TNR code was less encouraging. However, we believe that a systematic large-scale study
of this model using efficient disentangling algorithms might be able to further improve on
the errors we have reported here and extract the critical exponents precisely but it would
be challenging. In the future, it will be interesting to explore phase transitions in frustated

10
XY model and other spin models such as J-Q model exhibiting continuous symmetry using
these methods.

Acknowledgements
We thank Simon Catterall, Judah Unmuth-Yockey, Yuzhi Liu, Haiyuan Zou, Suraj Shankar,
Guifré Vidal, Stefan Kühn, Glen Evenbly, Nikhil Kalyanapuram, and Utkarsh Giri for dis-
cussions. The numerical computations were done on - Symmetry (Perimeter’s HPC system).
Research at Perimeter Institute is supported in part by the Government of Canada through
the Department of Innovation, Science and Economic Development Canada and by the
Province of Ontario through the Ministry of Colleges and Universities.

Appendix A: Numerical details

In this appendix, we elaborate on some numerical details and steps to determine the
critical temperature. We normalize the tensor Tν at each coarse-graining step (denoted
by ν) by the maximum element of the tensor having a total of χ4 elements which can
be implemented in Python using NumPy library as normν = numpy.max(Tν ). We can then
calculate the free energy density (f ) from these normalization factors, see for instance [29]:
N N
! !
1 X log(normν ) log(ZN ) 1 X
f =– + =– N log(normν )4N –ν + log ZN , (12)
β 4ν 4N β4
ν=0 ν=0

where ZN = tTr(TN ) is calculated from the tensor contraction after the last step of CG and
4N is the lattice volume and N is the number of times we do coarse-graining along x + y
direction. To compute expectation values, one needs to insert the appropriate ‘observable’
tensor in the network. By inserting ~
T, which is just the derivative of T with respect to h,
we evaluated the magnetization for a range of magnetic fields at various temperatures. The
~ν , is normalized by normν as well and is calculated as M = tTr(~Tν ) at each step. The
tensor, T tTr(Tν )
diagram representing this contraction is shown in Figure 1. Though the free energy density
converges rather quickly, we really need to work in the thermodynamic limit with as large
χ as possible to precisely evaluate magnetization. In order to compute the magnetization
for 10–15 ≤ h < 10–9 , we used single-precision float (np.float32) rather than the default
double precision. This enabled us to go up to χ = 53 compared to χ = 47 for h ≥ 10–9 .
Once the magnetization M is computed for different h, we evaluate ∆M /∆h|T ,hmid and plot
it against temperature and determine the critical temperature corresponding to the peak of
the susceptibility. The data for the determination of critical temperature for various h can
be found at [30]. We used Python’s curve_fit which uses Levenberg-Marquardt algorithm
and fit our data to the ansatz: Tc = Tc,h=0 + ahb , with a and b as two fitting parameters.
The result of the fit is:

Tc,h=0 = 0.89290(5), a = 0.364(2), and b = 0.162(1),

11
and is shown by the solid line in Figure (5).

References

[1] Y. Nambu, “A Note on the Eigenvalue Problem in Crystal Statistics,” Progress of

Theoretical Physics 5 no. 1, (Jan., 1950) 1–13.

[2] E. H. Lieb, T. Schultz, and D. Mattis, “Two soluble models of an antiferromagnetic

chain,” Annals Phys. 16 (1961) 407–466.

[3] H. E. Stanley and T. A. Kaplan, “Possibility of a phase transition for the

two-dimensional heisenberg model,” Phys. Rev. Lett. 17 (Oct, 1966) 913–915.
https://link.aps.org/doi/10.1103/PhysRevLett.17.913.

[4] V. L. Berezinsky, “Destruction of long range order in one-dimensional and

two-dimensional systems having a continuous symmetry group. I. Classical systems,”
Sov. Phys. JETP . http://jetp.ac.ru/cgi-bin/dn/e_032_03_0493.pdf.

[5] J. M. Kosterlitz and D. J. Thouless, “Ordering, metastability and phase transitions in

two-dimensional systems,” J. Phys. C6 (1973) 1181–1203. [,349(1973)].

[6] J. M. Kosterlitz, “The Critical properties of the two-dimensional x y model,” J. Phys.

C7 (1974) 1046–1060. [,377(1974)].

[7] J. F. Yu, Z. Y. Xie, Y. Meurice, Y. Liu, A. Denbleyker, H. Zou, M. P. Qin, and

J. Chen, “Tensor Renormalization Group Study of Classical XY Model on the Square
Lattice,” Phys. Rev. E89 no. 1, (2014) 013308, arXiv:1309.4963
[cond-mat.stat-mech].

[8] L. Vanderstraeten, B. Vanhecke, A. M. Laeuchli, and F. Verstraete, “Approaching the

Kosterlitz-Thouless transition for the classical XY model with tensor networks,”
arXiv:1907.04576 [cond-mat.stat-mech].

[9] R. Orus, “Tensor networks for complex quantum systems,” arXiv e-prints (Dec., 2018)
, arXiv:1812.04011 [cond-mat.str-el].

[10] M. C. Banuls and K. Cichy, “Review on Novel Methods for Lattice Gauge Theories,”
Rept. Prog. Phys. 83 no. 2, (2020) 024401, arXiv:1910.00257 [hep-lat].

[11] S. Kuehn, J. I. Cirac, and M. C. Banuls, “Non-Abelian string breaking phenomena

with Matrix Product States,” JHEP 07 (2015) 130, arXiv:1505.04441 [hep-lat].

12
[12] D. Kadoh, Y. Kuramashi, Y. Nakamura, R. Sakai, S. Takeda, and Y. Yoshimura,
“Tensor network formulation for two-dimensional lattice N = 1 Wess-Zumino model,”
JHEP 03 (2018) 141, arXiv:1801.04183 [hep-lat].

[13] A. Bazavov, S. Catterall, R. G. Jha, and J. Unmuth-Yockey, “Tensor renormalization

group study of the non-Abelian Higgs model in two dimensions,” Phys. Rev. D99
no. 11, (2019) 114507, arXiv:1901.11443 [hep-lat].

[14] A. Franco-Rubio and G. Vidal, “Entanglement renormalization for gauge invariant

quantum fields,” arXiv:1910.11815 [quant-ph].

[15] Y. Komura and Y. Okabe, “Large-scale monte carlo simulation of two-dimensional

classical xy model using multiple gpus,” arXiv:1210.6116 [cond-mat.stat-mech].

[16] A. M. Polyakov, Gauge Fields and Strings, vol. 3. 1987.

[17] Y. Liu, Y. Meurice, M. Qin, J. Unmuth-Yockey, T. Xiang, Z. Xie, J. Yu, and H. Zou,
“Exact Blocking Formulas for Spin and Gauge Models,” Phys. Rev. D 88 (2013)
056005, arXiv:1307.6543 [hep-lat].

[18] Z. Y. Xie, J. Chen, M. P. Qin, J. W. Zhu, L. P. Yang, and T. Xiang, “Coarse-graining

renormalization by higher-order singular value decomposition,” arXiv:1201.1144
[cond-mat.stat-mech].

[19] J. F. Unmuth-Yockey, “Gauge-invariant rotor Hamiltonian from dual variables of 3D

U (1) gauge theory,” Phys. Rev. D99 no. 7, (2019) 074502, arXiv:1811.05884
[hep-lat].

[20] S. Akiyama, Y. Kuramashi, T. Yamashita, and Y. Yoshimura, “Phase transition of

four-dimensional Ising model with higher-order tensor renormalization group,” Phys.
Rev. D100 no. 5, (2019) 054510, arXiv:1906.06060 [hep-lat].

[21] R. Gupta and C. F. Baillie, “Critical behavior of the two-dimensional xy model,”

Phys. Rev. B 45 (Feb, 1992) 2883–2898.
https://link.aps.org/doi/10.1103/PhysRevB.45.2883.

[22] P. Butera and M. Comi, “Quantitative study of the kosterlitz-thouless phase

transition in an xy model of two-dimensional plane rotators: High-temperature
expansions to order β 20 ,” Phys. Rev. B 47 (May, 1993) 11969–11979.
https://link.aps.org/doi/10.1103/PhysRevB.47.11969.

[23] P. Olsson, “Monte carlo analysis of the two-dimensional xy model. ii. comparison with
the kosterlitz renormalization-group equations,” Phys. Rev. B 52 (Aug, 1995)
4526–4535. https://link.aps.org/doi/10.1103/PhysRevB.52.4526.

13
[24] M. Hasenbusch, “The Two dimensional XY model at the transition temperature: A
High precision Monte Carlo study,” J. Phys. A38 (2005) 5869–5884,
arXiv:cond-mat/0502556 [cond-mat].

[25] H. Arisue, “High-temperature expansion of the magnetic susceptibility and higher

moments of the correlation function for the two-dimensional xy model,” Phys. Rev. E
79 (Jan, 2009) 011107. https://link.aps.org/doi/10.1103/PhysRevE.79.011107.

[26] Y.-D. Hsieh, Y.-J. Kao, and A. W. Sandvik, “Finite-size scaling method for the
berezinskii-kosterlitz-thouless transition,” arXiv:1302.2900 [cond-mat.stat-mech].

[27] M. M. Rams, P. Czarnik, and L. Cincio, “Precise extrapolation of the correlation

function asymptotics in uniform tensor network states with application to the
bose-hubbard and xxz models,” Phys. Rev. X 8 (Nov, 2018) 041033.
https://link.aps.org/doi/10.1103/PhysRevX.8.041033.

[28] G. Evenbly and G. Vidal, “Tensor Network Renormalization,” Physical Review Letters
115 no. 18, (Oct., 2015) 180405, arXiv:1412.0732 [cond-mat.str-el].

[29] H.-H. Zhao, Z.-Y. Xie, T. Xiang, and M. Imada, “Tensor network algorithm by
coarse-graining tensor renormalization on finite periodic lattices,” arXiv:1510.03333
[cond-mat.str-el].

[30] https://github.com/rgjha/TensorCodes/blob/master/dataXY.pdf.

14