June 28, 2024

Abstract

We study the nature of quantum phase transitions out of $\mathbb{Z}_{4}$ ordered phases in quantum loop models on a zig-zag ladder. We report very rich critical behaviour that includes a pair of Ising transitions, a multi-critical Ashkin-Teller point and a remarkably extended interval of a chiral transition. Although plaquette states turn out to be essential to realize chiral transitions, we demonstrate that critical regimes can be manipulated by deforming the model as to increase the presence of leg-dimerized states. This can be done to the point where the chiral transition turns into first order, we argue that this is associated with the emergence of a critical end point.

 

Contents

1. 1 Introduction
2. 2 Model & Methods
   1. 2.1 Model
   2. 2.2 DMRG
   3. 2.3 Extraction of $\xi$ and $q$
3. 3 $\mathbb{Z}_{4}$ transitions for $t^{\prime}=t$
   1. 3.1 Overview of the phase diagram
   2. 3.2 Ising transitions
   3. 3.3 Ashkin-Teller point and extended chiral transition
4. 4 Manipulating the nature of $\mathbb{Z}_{4}$ transitions
   1. 4.1 Transition to the plaquette phase at $t^{\prime}=0$
   2. 4.2 First order transition for $t^{\prime}=2$
   3. 4.3 Phase diagram as a function of $t^{\prime}$
5. 5 Conclusion
6. A DMRG with quantum loop model constraints
   1. A.1 Implementing quantum loop constraints
   2. A.2 Matrix product operator
7. B Locating the disorder line
8. C Mapping to Rydberg atoms

 

1 Introduction

Identifying different universality classes of quantum phase transitions appearing in low dimensional many-body systems is one of the central topics in condensed matter physics [1]. Often, one can guess the underlying critical theory by analyzing the symmetry spontaneously broken at the transition. Let us consider, for example, the transition between an ordered phase with a $\mathbb{Z}_{p}$ symmetry and a disordered phase. Naïvely, the corresponding transition is expected to belong to the universality class of conformal minimal model with the matching value of $p$, i.e. Ising for $p=2$, three-state Potts for $p=3$, Ashkin-Teller for $p=4$. However, if short-range correlations in the disordered phase are incommensurate (IC), with the dominant wave-vector $q$ different from $2\pi/p$, the transition is classified as a commensurate-incommensurate (C-IC) one. Understanding the nature of these C-IC transitions is one of the biggest challenges of modern quantum physics that roots back to the classical study of absorbed monolayers [2, 3, 4, 5]. In this case, different sequences of ground-state domains, such as $ABC$ and $ACB$, have different sets of domain walls, and if free energy contributions of various domain walls are not identical, i.e. $AB\neq AC$, one type of sequence becomes energetically favored over the other and induce chiral perturbations [6]. For $p=2$, if the transition is continuous, one can always expect the transition to be of the Ising type, irregardless of chiral perturbations. But for $p>2$ chiral perturbations are (almost always) relevant and drastically change the nature of the transition [7, 8, 3, 6]. For instance, for ordered phases with periodicities $p\geq 5$ the C-IC transition cannot be direct but is a two-step transition with an intermediate floating phase [3, 9], i.e. an incommensurate Luttinger liquid [10]. This floating phase is separated from the C ordered phase by a Pokrovsky-Talapov [11, 2] transition and from the IC disordered phase by the Kosterlitz-Thouless transition [12], with in the latter case exponentially decaying correlations.

The transitions in the cases of $p=3$ and $p=4$ are much more exotic in nature. When the disordered phase is commensurate, i.e. characterized by $q=2\pi/p$, the transition is conformal and belongs to the 3-state Potts (for $p=3$) or Ashkin-Teller (for $p=4$) universality class. In the presence of strong chiral perturbations the transition to the $p=3$ and $p=4$ phases is through the floating phase, similar to the $p\geq 5$ cases. In the presence of weak chiral perturbations the C-IC transition to the $p=3$ phase remains direct, though it is no longer conformal but belongs to the chiral Huse-Fisher universality class[6, 3, 13, 14, 15, 16, 17]. Weak chiral perturbations might also lead to a direct chiral transition at the boundary of the $p=4$ phase. In this case, however, the appearance of the chiral transition is a subtle issue [3] due to the fact that the Ashkin-Teller family of conformal critical theories forms a so-called weak universality class [18, 19]. This means that some critical exponents, such as $\nu$ and $\beta$ describing the divergence of the correlation length and the scaling of the order parameter in the vicinity of the critical point correspondingly, are not fixed but vary as a function of an external parameter (usually called $\lambda$ [18, 20, 21, 22]), while others, including the central charge and the scaling dimension $d=\beta/\nu$, are universal within the family. The Ashkin-Teller family of transitions ranges from a pair of decoupled Ising chains with $\nu=1$ to the symmetric 4-state Potts point with $\nu=2/3$ [18]. It turns out that the way chiral perturbations affect the nature of the transition drastically depend on the properties of the conformal point. When $\nu\gtrsim 0.8$ even weak chiral perturbations always open up a floating phase [3, 22, 20]. On the other hand, when the Ashkin-Teller point is characterized by $(1+\sqrt{3})/4<\nu\lesssim 0.8$ a direct chiral transition emerges under weak chiral perturbations and is followed by an opening of the floating phase when chiral perturbations become strong. The two regimes are separated by the Lifshitz point characterized by the dynamical critical exponent $z=3$ [3, 20, 16]. Finally, when the Ashkin-Teller point is close to the 4-state Potts point with $2/3\leq\nu\leq(1+\sqrt{3})/4\approx 0.683$ chiral perturbations are irrelevant [2], allowing for an interval of the conformal Ashkin-Teller transition, followed by the chiral transition and the floating phase when chiral perturbations become stronger [22].

In addition to the chiral Ashkin-Teller model [20, 22], the $\mathbb{Z}_{4}$ chiral transition has been recently reported in the context of Rydberg atoms [23, 24, 25, 26, 27, 28, 29, 30]. In experiments Rydberg atoms are trapped with optical tweezers in a one-dimensional array with a well controlled inter-atomic distances. Lasers with Rabi frequency $\Omega$ bring atoms from their ground-state to excited Rydberg states. Competition between the strong van der Waals interaction of excited atoms and the laser detuning $\Delta$ leads to a rich phase diagram with the lobes of density-wave phases with different integer periodicities $p$ [24, 9]. A narrow intervals of the chiral transition has been observed at the boundary of the $p=4$ lobe [26, 27, 29].

In this paper we explore a possibility to realize the $\mathbb{Z}_{4}$ chiral transition in quasi-one-dimensional quantum magnets. The transition we are looking for takes place between two gapped phases, one of which - the disorder phase - does not break translation symmetry and has no long-range order. According to the Mermin-Wagner theorem [31] this naturally exclude half-integer spin chains. Previous studies of Heisenberg spin-1 chain with nearest- and next-nearest-neighbor interactions report a next-nearest-neighbor Haldane (NNN-Haldane) phase [32, 33, 34]. This phase, being topologically trivial without long-range order and with incommensurate short-range correlations, is therefore ideally suitable for the role of an incommensurate disordered phase and provides an excellent starting point for our study. The $\mathbb{Z}_{4}$ chiral transition, just like the previously reported Ising transition to the dimerized phase [35, 36], is expected to be non-magnetic with low-lying excitations taking place entirely within the singlet sector while a singlet-triplet gap remains open. This allow us to focus on $S=1$ quantum loop models (QLMs) [37] that effectively disregard all magnetic degrees of freedom such that the Hilbert space is formed by all possible dimer coverings satisfying the quantum loop constraints.

These quantum loop models provide a number of computational advantages. Firstly, QLMs are by definition constrained: at each node, originally hosting a spin-1 site, 2 and only 2 quantum dimers originate. Dimer coverings that violate this constraint are excluded from the Hilbert space, significantly reducing its size. Secondly, QLMs formulated directly in terms of dimers simplifies the construction of the parent Hamiltonian that would lead to a desired long-range order with spontaneously broken $\mathbb{Z}_{4}$ symmetry. Furthermore, as we will show later, there are several realization of the period-4 phase in spin-1 chain and quantum loop model provides a natural way to interpolate and distinguish between them. Thirdly, in special cases QLMs can be mapped to the blockade models of Rydberg atoms[38]. This provides an excellent starting point for our study and opens a by-path for experimental validation of our results. Fourthly, non-magnetic Ising transition into the dimerized phase in a frustrated Haldane chain mentioned previously has been successfully reproduced with the related quantum loop model [38]. In other words, if a chiral transition exists in quantum spin-1 chains, likely it will also appear in the toy quantum loop models. In turn, realization of chiral transitions in QLMs will help to narrow down the conditions for its appearance in a more realistic models of quantum magnetism. We address the problem numerically with the state-of-the-art density matrix renormalization group (DMRG) algorithm[39, 40] with explicitly implemented quantum loop constraints[38] such that the algorithm fully profits from a restricted Hilbert space allowing us to reach the convergence for critical chains with up to $N=3000$ sites.

In this paper we report an extremely extended chiral transition separating the $\mathbb{Z}_{4}$ leg-dimerized and the disordered NNN-Haldane phase. For a certain parameter range we see that disordered and $\mathbb{Z}_{4}$ phases are connected via a pair of Ising transitions with $\mathbb{Z}_{2}$ rung-dimerized phase between the two. Eventually the rung-dimerized phase disappears and two Ising transitions merge into a multi-critical Ashkin-Teller point, beyond which we resolve the $\mathbb{Z}_{4}$ chiral transition. We show that the nature of the Ashkin-Teller point and the extent of the chiral transition depends on the relative weight of two realizations of the $\mathbb{Z}_{4}$ phase - trivial double-dimers on legs and four-site plaquette loops. Furthermore, and quite surprisingly, when the weight of double leg dimers is too high the chiral transition turns into a first order transition. We argue that this might be due to the presence of an additional relevant operator in the underlying critical theory.

The rest of the paper is organized as follows. In Section 2 we define the model, provide details on our DMRG algorithm and explain how we extract main observables. In Section 3 we discuss the basic phase diagram of the quantum loop ladder with three main phases - $\mathbb{Z}_{2}$, $\mathbb{Z}_{4}$ and disordered. We also show that for an extremely extended parameter range the transition between the latter two phases is chiral. In Section 4 we show that the nature of the multi-critical point and the extent of the chiral transition can be manipulated by the relative ratio of the two kinetic terms of the QLM. Here we also discuss how the first order transition between the $\mathbb{Z}_{4}$ and the disordered phase develops as a function of the kinetic term responsible for the formation of double-dimers. Finally we summarize our results and put them in perspective in Section 5.

2 Model & Methods

2.1 Model

We study $\mathbb{Z}_{4}$ transitions in the QLM with two dimers per node on a zig-zag ladder. This choice allows us to realized a quantum dimer analogue of the NNN-Haldane phase of the $J_{1}-J_{2}$ Heisenberg spin-1 chain. We also include a potential term that stabilizes the $\mathbb{Z}_{4}$ phase. The microscopic Hamiltonian is defined as follows:

$$\mathcal{H}_{\mathrm{QLM}}=-\sum_{\mathrm{plaquettes}}\left(t|\includegraphics[height=8.53581pt]{sld_plaquette.pdf}\rangle\langle\includegraphics[height=8.53581pt]{srd_plaquette.pdf}|+t^{\prime}|\includegraphics[height=8.53581pt]{double_leg.pdf}\rangle\langle\includegraphics[height=8.53581pt]{plaquette.pdf}|+\mathrm{h.c}\right)-\theta\sum_{\mathrm{plaquettes}}|\includegraphics[height=8.53581pt]{plaquette.pdf}\rangle\langle\includegraphics[height=8.53581pt]{plaquette.pdf}|+\delta\sum_{\mathrm{rungs}}|\includegraphics[height=8.53581pt]{double_rung.pdf}\rangle\langle\includegraphics[height=8.53581pt]{double_rung.pdf}|,$$ (1)

where the respective sums run over all plaquette and rungs with both orientations of the zig-zag chain. The first kinetic term, parametrized by $t$, acts on a plaquette that has its rungs occupied by a dimer (either single or double) and flips one pair of dimers at a time over a given plaquette. The second kinetic term controlled by $t^{\prime}$ acts on a plaquette that has both its rungs and legs occupied by a single dimer and flips them to a plaquette with two dimers on both legs. Furthermore, the first potential term, controlled by the parameter $\theta$, favors 4-site quantum loops - the plaquette phase. Lastly, the parameter $\delta$ controls the second kinetic term that acts on all rungs in the ladder. This term aims to suppress the otherwise very stable rung dimer phase. Without loss of generality we set $t=1$ and explore the model as a function of three independent parameters $t^{\prime}$, $\theta$ and $\delta$.

2.2 DMRG

We study the model defined in Eq.(1) with the constrained three-site - optimizing three tensors per iteration - DMRG algorithm with explicitly implemented quantum loop constraints using the Matrix Product State (MPS) ansatz [40, 39, 38, 41, 42]. With all eigenstates consisting fully of local dimer configurations, the QLM degrees of freedom can be taken as the occupation of dimers on the rungs and legs instead of spins. We make use of this by taking as the local Hilbert space the occupation of dimers on a single rung with the preceding leg, with its dimension being equal to six - one without any dimers, two with two dimers on either the rung and leg, and three with single dimer occupations of the rung and leg. For different dimer occupations, we associate quantum numbers to label their respective sectors in the Hilbert space in iterative processes of the algorithm, such as the construction of the left and right normalized blocks. These blocks are constructed iteratively and the fusion between these quantum numbers under the addition of new sites to these blocks are described by a set of fusion rules. Since there is a one-to-one correspondence assigned quantum labels, the tensor can be projected into a block-diagonal form, significantly reducing computational cost. In addition, the dimension of the Hilbert space scales $\approx 1.68^{N}$ instead of $3^{N}$ for spin-1 chains. Further details on the construction of the matrix product operator for the QLM Hamiltonian and technical aspects of the implementation of the constraints are provided in Appendix A.

We simulate systems with up to $N=3000$ spin-1 lattice sites, which effectively translates to $N=2999$ QLM local degrees of freedom, each half sweep we increase the bond dimension by 200, up to a maximum of 10000 (that according to our record was never reached) and discard all singular values smaller than $10^{-8}$. We perform up to seven full sweeps and . As a convergence criteria we use the absolute error of the energy over the sweep to be below $10^{-13}$. We start each simulation with a random guess for our initial wave-function. We always use $4k-1$, with $k\in\mathbb{N}$, QLM sites to ensure that the ordered phases, including the $\mathbb{Z}_{4}$ plaquette and leg-double-dimers, fully cover the ladder with symmetric boundary conditions.

2.3 Extraction of $\xi$ and $q$

Our numerical analysis of the nature of the transition between the $\mathbb{Z}_{4}$ symmetric and the NNN-Haldane phase primarily relies on the scaling of the correlation length $\xi$ and of the incommensurate wave-vector $q$. We extract these from the correlation function $C_{i,j}\propto\langle n_{i}n_{j}\rangle-\langle n_{i}\rangle\langle n_{j}\rangle$, unless states otherwise we use $n_{i}=|\includegraphics[height=8.53581pt]{basis_h_3.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_3.pdf}|+|\includegraphics[height=8.53581pt]{basis_h_6.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_6.pdf}|$, by fitting it to the Ornstein-Zernicke form [43]:

$$C_{i,j}^{\mathrm{OZ}}\propto\frac{e^{-|i-j|/\xi}}{\sqrt{|i-j|}}\cos(q|i-j|\pi+\phi_{0}),$$ (2)

where $\xi$, $q$ and the phase $\phi_{0}$ are considered as fitting parameters. The correlation function is always calculated on the interval $N/2\leq i\leq N$ with $j=N/2$. We extract the correlation length and the wavevector in a two-step process. First, we plot the logarithm of the correlation function, as shown in Fig. 1 (a), and fit its slope with the function $\ln C_{i,j}(x=|i-j|)\approx A-\newline -x/\xi-\ln(x)/2$ to extract $\xi$. In doing so we fit $\xi$ and the amplitude $A$ at the same time. To extract the slope we first identify local maxima of $\ln|C_{i,j}|$ by comparing each point with its $2$ neighbours and then fitting these maxima.

Figure 1: Example of fitting the correlation function $C_{i,j}$ to the Ornstein-Zernicke form in a two step process. (a) First the correlation length $\xi$ is extracted from fitting the main slope of $\ln|C_{i,j}|$. (b) Then the reduced correlation function $\tilde{C}_{i,j}$ is calculated and fitted to extract the wavevector $q$.

In the second step we construct the reduced correlation function

$$\tilde{C}_{i,j}=C_{i,j}\cdot Ae^{|i-j|/\xi}\sqrt{|i-j|}$$ (3)

that we fit with

$$\tilde{C}_{i,j}^{\mathrm{OZ}}\approx a\cos(q|i-j|\pi+\phi_{0}).$$ (4)

In Fig. 1 (b) we show an example of the fit where we treat $q$, $\phi$ and $a$ as fitting parameters. The agreement between our numerical data (blue circles) and the fit (orange dots) is almost perfect. We expect the wave-vector $q$ to be affected by finite-size effect due to fixed boundary conditions. We can estimate the associated error to be of the order of $2\xi/N^{2}$, where $2/N$ is a single step in $q$ and the extra factor $\xi/N$ results from finite size effects that could arise at the boundaries.

We repeat the procedure outlined above for multiple points in proximity to the critical point. From these data points we extract the critical exponents $\nu$ and $\bar{\beta}$ from the scaling of the correlation $\xi$ and the wave-vector $q$ respectively. This we also do in two steps. First we fit the inverse of the correlation length $1/\xi\propto|\theta-\theta_{c}|^{\nu}$ to extract $\nu$ and the critical parameter $\theta_{c}$, indicating the location of the transition. We do this with the function $1/\xi=A\Theta(\theta_{c}-\theta)(\theta_{c}-\theta)^{\nu}\newline +B\Theta(\theta-\theta_{c})(\theta-\theta_{c})^{\nu}$, where $\Theta(x)$ is the heaviside step function and we fitted $\nu$ and $\theta_{c}$ but also the constants $A$ and $B$ all at the same time. After extract $\theta_{c}$ we use it to fit the wavevector $q\propto|\theta-\theta_{c}|^{\bar{\beta}}$ with $q=C\Theta(\theta_{c}-\theta)(\theta_{c}-\theta)^{\bar{\beta}}$ to obtain $\bar{\beta}$. In our analysis we also compute the product $|q/\pi-1/2|\times\xi$. We would like to emphasize here that we do this point-by-point, independently from each fit for $\nu$ and $\bar{\beta}$.

3 $\mathbb{Z}_{4}$ transitions for $t^{\prime}=t$

3.1 Overview of the phase diagram

We start our study of the quantum loop model defined in Eq. (1) by looking at a special case of $t^{\prime}=t=1$. The ground-state phase diagram as a function of $\delta$ and $\theta$ is presented in Fig.2 and contains three main phases - NNN-Haldane phase, $\mathbb{Z}_{2}$ rung-dimerized phase and $\mathbb{Z}_{4}$ leg-dimerized phase. The NNN-Haldane phase is a disordered phase realized for positive $\delta$ and small $\theta$. Within the disordered phase we distinguish two regions: one with commensurate and the other with incommensurate short-range order, which are separated by the so-called disorder line. In Appendix B we explain in details the method we use to locate this line. By increasing $\theta$ while keeping $\delta$ positive, the system enters the $\mathbb{Z}_{4}$ leg-dimerized phase phase with spontaneously broken translation symmetry. There are two main families of states realized with a broken $\mathbb{Z}_{4}$ symmetry - plaquette and columnar leg states. In the former every four consecutive sites form a single dimer loop while in the latter every other leg is occupied by a double dimer - a trivial loop. Example of both states are sketched in Fig.2. The leg-dimerized phase realized with $t^{\prime}=t$ is a superposition of these two families of states. The columnar states have their most significant weight near the boundary of the leg-dimerized phase, nonetheless always being a fraction of that of plaquette states. Further away from the boundary the density of columnar dimer states decreases down to zero. Negative $\theta$ stabilizes the rung-dimerized phase with spontaneously broken translation symmetry. Although the potential term in the Hamiltonian of Eq.(1) favors only one type of rung-dimerized state - the fully dimerized ones with every other rung occupied by a trivial two-dimer loop - in the $\mathbb{Z}_{2}$ phase these states appears in a superposition with partially dimerized states where every other rung is occupied by a single dimer while the full quantum loop coverage is completed by a uniform density of dimers resonating on the legs of the ladder. In Fig.3 we sketch both pairs of states and provide typical density profiles in the rung-dimerized phase.

Figure 2: Phase diagram of the quantum loop model defined in Eq. (1) for $t^{\prime}=t=1$ as a function of $\delta$ and $\theta$. There are three main phases: disordered, NNN-Haldane, phase (yellow) with a commensurate and incommensurate parts separated by the disorder line (grey line), $\mathbb{Z}_{2}$ rung-dimerized phase (red), and $\mathbb{Z}_{4}$ phase (blue) as a superposition of plaquette and columnar states. The transition between the disordered phase and the rung-dimerized phase and that between the rung-dimerized and $\mathbb{Z}_{4}$ leg-dimerized phase are of the Ising type (green circles). Two Ising transitions merge into the multi-critical Ashkin-Teller point (purple dot) within some uncertainty of its location indicated by the light-pink region around it. Beyond the Ashkin-Teller point the transition is chiral (blue squares) The $\theta$-axis is broken for visual clarity.

After identifying the phases, we take a closer look at the transitions. One way to go from the $\mathbb{Z}_{4}$ leg-dimerized phase to the disordered NNN-Haldane phase is through a pair of Ising transitions with the intermediate rung-dimerized phase. At each of the two Ising transitions $\mathbb{Z}_{2}$ symmetry is spontaneously broken (hence the $\mathbb{Z}_{2}$ rung-dimerized phase appears). Upon increasing $\delta$ the two Ising transitions, characterized by the central charge $c=1/2$, come closer and merge into an Ashkin-Teller multi-critical point with a central charge $c=1$ [19]. There is some uncertainty in the location of the Ashkin-Teller point due to a limited numerical resolution and a crossover between various critical regimes. Instead, we identify a finite interval along the boundary of the $\mathbb{Z}_{4}$ leg-dimerized phase where the Ashkin-Teller point is located. Beyond the Ashkin-Teller point the transition becomes chiral. For $t^{\prime}=t$ we do not observe a floating phase opening up, at least up to $\delta=50$, indicating that the chiral transition persists for an extremely extended interval.

Figure 3: Sketches of two families of $\mathbb{Z}_{2}$ symmetry broken states realized as a superposition in the rung-dimerized phase alongside their typical density profiles. (a) Fully dimerized state with every other rung occupied by a double dimer and (b) partially dimerized state with single dimer every other rung. Density profile of two dimers (c) and a single dimer (d) on each rung, computed for the parameters $t^{\prime}=0$, $N=199$ sites, $\delta=-1$ and $\theta=2$.

3.2 Ising transitions

We start our analysis with the region where the transition between the $\mathbb{Z}_{4}$ ordered and the disordered phase takes place through two Ising transitions and a $\mathbb{Z}_{2}$ ordered phase in between them. We locate these Ising transitions by performing a finite-size scaling of the relevant order parameter. We first focus on the transition between the $\mathbb{Z}_{2}$ rung-dimerized phase and NNN-Haldane phase. At this transition the translation symmetry between rungs is spontaneously broken. An operator that reflects this is the dimerization on the rungs $D_{\mathrm{rung}}=|\langle n_{i}\rangle-\langle n_{i+1}\rangle|$, where $n_{i}=|\includegraphics[height=8.53581pt]{basis_h_4.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_4.pdf}|+2|\includegraphics[height=8.53581pt]{basis_h_5.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_5.pdf}|$ is an operator that computes the local density of dimers on a rung. To reduce edge effects we extract dimerization in the center of the ladder. In the thermodynamic limit we expect $D_{\mathrm{rung}}$ to approach a finite value inside the ordered $\mathbb{Z}_{2}$ phase and to vanish in the uniform disordered phase. In a log-log plot these will appear as convex and concave curves respectively and in between these is the separatrix that we associate with the quantum critical point. The slope of the separatrix yields the scaling dimension $d$ of the corresponding operator. In Fig.4 (a) we show the finite-size scaling of the dimerization for $\delta=0.55$ and various values of $\theta$. Our numerical results suggest that the critical point is located at $\theta_{1}^{c}\approx 1.6242$ and the scaling dimension $d\approx 0.125$ extracted numerically is in excellent agreement with the conformal field theory predictions $d=1/8$ for the Ising transition.

Figure 4: Numerical evidence for the Ising transition between the $\mathbb{Z}_{2}$ rung-dimerized phase and (a)-(b) the disordered and (c)-(d) the $\mathbb{Z}_{4}$ leg-dimerized phase for $\delta=0.55$. (a) Finite-size scaling of the dimerization on the rungs computed in the middle of the ladder with open boundary conditions. The quantum critical point is located at $\theta^{c}_{1}\approx 1.6242$ (red circles). (c) Finite size scaling of the dimerization on one of the legs, computed in the middle of the ladder with open boundary conditions. The critical point is located at $\theta^{c}_{2}\approx 1.807$ (red circles). In both cases the scaling dimension $d\approx 0.125$ extracted at the critical points is in excellent agreement with theory prediction for Ising transition $d=1/8$ (black dashed lines). (b),(d) Scaling of the reduced entanglement entropy with $d_{N}(n)=\frac{2N}{\pi}\sin\frac{\pi n}{N}$ at the two critical points $\theta_{1}^{c}$ and $\theta_{2}^{c}$ obtained as shown in (a) and (c). Extracted central charges $c\approx 0.501$ at $\theta_{1}^{c}$ and $c\approx 0.505$ at $\theta_{2}^{c}$ agree within $1\%$ with Ising $c=1/2$.

To further verify the nature of the transition between the rung-dimerized and disordered phase we extract the central charge $c$. From the reduced density matrix we compute the entanglement entropy $S_{N}(n)$ and then compute the reduced entanglement entropy, by removing Friedel oscillations of local density [44] - in the present case translation symmetry is broken - which we define as:

$$\tilde{S}_{N}(n)=S_{N}(n)-\alpha\langle n_{i}\rangle,$$ (5)

where $\alpha$ is a non-universal parameter tuned such that the oscillations are removed. For conformal transitions the reduced entanglement entropy in 1D systems with open boundary conditions scales linearly with $c$ according to the Calabrese-Cardy formula [45]:

$$\tilde{S}_{N}=\frac{c}{6}\ln d_{N}(n)+\ln g+s_{1},$$ (6)

where $d_{N}(n)=\frac{2N}{\pi}\sin\frac{\pi n}{N}$ is the conformal distance, $\ln g$ accounts for boundary contributions to the entropy and $s_{1}$ is a non-universal constant. We extract the central charge at the previously identified critical point $\theta^{c}_{1}$ by performing a scaling of the reduced entanglement entropy $\tilde{S}_{N}(n)$ with the conformal distance as shown in Fig.4 (b). To reduce boundary effects we discarded 30% of all sites on both edges of the chain. We observe that as we increase $N$ the numerically extracted value approaches the theory prediction $c=1/2$ for the Ising transition.

Based on the symmetry arguments we also expect an Ising transition between the rung-dimerized ($\mathbb{Z}_{2}$) and leg-dimerized ($\mathbb{Z}_{4}$) phases. To confirm this, we follow the same protocol. First, we compute the finite-size scaling of the dimerization in the middle of the ladder, but now for the leg density operators $\tilde{n}_{i}=|\includegraphics[height=8.53581pt]{basis_h_3.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_3.pdf}|+2|\includegraphics[height=8.53581pt]{basis_h_6.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_6.pdf}|$. Here we only consider the top chain, but, of course, the results for the bottom one would be identical. We denote $\tilde{D}_{\mathrm{leg}}=|\tilde{n}_{i}-\tilde{n}_{i+1}|$. In the rung-dimerized phase the dimer density on legs is small and uniform thus dimerization on legs scales to zero in the $\mathbb{Z}_{2}$ phase, while it naturally goes to a finite-value in the leg-dimerized phase, in the thermodynamic limit. Fig.4 (c) we provide an example of the scaling of $\tilde{D}_{\mathrm{leg}}$. We use the same value of $\delta=0.55$ as in the Fig.4 (a). The second critical point associated with the separatrix is located at $\theta^{c}_{2}\approx 1.807$ and it is clearly different from the first Ising transition ($\theta^{c}_{1}\neq\theta^{c}_{2}$). At this point we extract the scaling dimension $d\approx 0.125$ (Fig.4 (c)) and a central charge $c\approx 0.505$ (Fig.4 (d)). Both are in excellent agreement with Ising universality class.

3.3 Ashkin-Teller point and extended chiral transition

By accurately locating the two Ising transitions at the boundary of the $\mathbb{Z}_{2}$ phase we notice that upon increasing $\delta$ the two transitions come closer and eventually the rung-dimerized phase disappears. We expect the multi-critical point where two Ising transitions meet to be in the Ashkin-Teller universality class [18], characterized by the central charge $c=1$. We also locate the disorder line beyond which the disordered phase has incommensurate short-range correlations. Due to a finite resolution of the numerical approach there is an uncertainty in the location of the multi-critical point and the disorder line. We thus define an interval (indicated in Fig.2 with light pink symbols) that starts once we no longer are able to resolve two Ising transitions with sufficient accuracy and terminates at the point where we detect a chiral transition. The multi-critical Ashkin-Teller point is located within this interval, but there are three possible scenarios that can be realized as sketched in Fig.5. The most probable situation is that the disorder line hits the Ashkin-Teller point and if the critical exponent at this point lies within the interval $0.8\lesssim\nu\lesssim 0.68$ the chiral transition starts immediately as sketched in Fig.5(a). In case when the critical exponent is $0.68\lesssim\nu\leq 2/3$ the chiral transition starts after a short interval of the conformal Ashkin-Teller transition [22] (see Fig.5(b)). It is also possible that the disorder line hits the transition at a certain distance from the multi-critical point, then, in the interval between the two, the transition will be commensurate and in the Ashkin-Teller universality class.

Figure 5: Sketches of the three possible scenarios of two Ising transitions merging into an Ashkin-Teller point followed up by a chiral transition. (a) Disorder line hits an Ashkin-Teller point at which $\nu\geq(1+\sqrt{3})/4\approx 0.68$. The transition immediately becomes chiral beyond this point. (b) If $2/3\leq\nu\lesssim 0.68$ the chiral perturbation can be irrelevant, allowing for a short Ashkin-Teller interval, even if the disorder line terminates in the Ashkin-Teller point. (c) The disorder line does not terminate in the Ashkin-Teller points but at the direct transition line, allowing for an Ashkin-Teller interval, even if $\nu\gtrsim 0.68$. Beyond the disorder line, in the presence of chiral perturbations, the transition becomes chiral.

Beyond the multi-critical Ashkin-Teller point the transition between the $\mathbb{Z}_{4}$ and the NNN-Haldane phase can be in one of the three regimes $-$ Ashkin-Teller, chiral, and an intermediate floating phase. Following Huse and Fisher [6], we use the product $\Delta q\times\xi$, where $\Delta q=|q/\pi-1/2|$ is the distance between the wave-vector $q$ from its commensurate value $q=\pi/2$ and $\xi$ is the correlation length. Upon approaching the transition $\Delta q$ vanishes with the critical exponent $\bar{\beta}$, while $\xi$ diverges with critical exponent $\nu$. Despite $\bar{\beta}$ not being known for the Ashkin-Teller universality class, for conformal transition one can expect $\bar{\beta}>\nu$ [6, 3]. Subsequently, the product $|q/\pi-1/2|\times\xi$ is expected to vanish approaching the Ashkin-Teller transition. For the chiral transition $\bar{\beta}=\nu$ and thus the product $|q/\pi-1/2|\times\xi$ goes to a constant. By contract, the floating phase, being incommensurate in nature, is separated from the disordered phase by the Kosterlitz-Thouless transition that is characterized by an exponential divergence of the correlation length [12] while at the same time the wave-vector $q$ remains incommensurate through the transition. Therefore, we expect the product $|q/\pi-1/2|\times\xi$ to diverge in this case.

Our numerical results are summarized in Fig.6 where we show the inverse of the correlation length $1/\xi$, the distance between the wave-vector $q$ from its commensurate value $|q/\pi-1/2|$, and the product of the two $|q/\pi-1/2|\times\xi$. The results are presented for three vertical cuts across the transition for various values of $\delta$. For $\delta=1.7$, that we expect to go through the Ashkin-Teller point, we extract a critical exponent $\nu\approx 0.678$ and $\bar{\beta}>\nu$ consistent with the Ashkin-Teller universality class. Furthermore, the product $|q/\pi-1/2|\times\xi$ vanishes upon approaching the transition. In principle, this Ashkin-Teller point belongs to the interval in which chiral perturbations can be irrelevant, i.e. $2/3\leq\nu\leq(1+\sqrt{3})/4$ [2, 22], and thus there might be a small interval of the Ashkin-Teller transition, as sketched in Fig.5(b). Our numerical accuracy does not allow to resolve this possibility. Starting from $\delta\approx 2$ we see signatures of the chiral transition. A typical example for $\delta=5$ is presented in Fig.6(b)(e),(h). Numerically extracted critical exponents $\nu\approx\bar{\beta}\approx 0.64\pm 0.03$ are very similar and the product $|q/\pi-1/2|\times\xi$ stays constant upon approaching the transition signaling a chiral transition.

Figure 6: (a)-(c) Inverse of the correlation length $\xi$, (d)-(f) $|q/\pi-1/2|$ - the distance of wave-vector $q$ with respect to the commensurate value $q=\pi/2$, and (g)-(i) the product $|q/\pi-1/2|\times\xi$ along three cuts across the transition between the leg-dimerized ($\mathbb{Z}_{4}$) and the NNN-Haldane (disordered) phases. We fit $1/\xi$ and $|q/\pi-1/2|$ with power laws to extract $\nu$ and $\bar{\beta}$. Dashed vertical lines show the boundary of the $\mathbb{Z}_{4}$ phase extracted by fitting $1/\xi$. Error bars of $|q/\pi-1/2|$ and $|q/\pi-1/2|\times\xi$ are shown as $2\times\xi/N^{2}$ and $2\times\xi^{2}/N^{2}$ (we only show those that exceed the size of the symbols).

Further away from the Ashkin-Teller point we expect the floating phase to open. However, we find, rather surprising, that in this quantum loop model the chiral transition extends over a very long interval. Even at $\delta=50$ (see Fig.6 (c),(f),(i)), the transition still appears to be chiral. If the floating phase appear, the transition that separates it from the $\mathbb{Z}_{4}$ ordered phase is expected to be in the Pokrovsky-Talapov [46, 11] universality class characterized by the critical exponents $\nu=1/2$ (correlation length in the ordered phase), and $\bar{\beta}=1/2$ (incommensurability as an order parameter in the floating phase). The critical exponents that we extract numerically across the transition at $\delta=50$ are still significantly far from these values. Furthermore, $|q/\pi-1/2|\times\xi$ does not diverge as expected for the floating phase, but still clearly converges to a constant in agreement with a chiral transition, and the larger system size we consider the flatter is the function $|q/\pi-1/2|\times\xi$ in the vicinity of the transition. A remarkably extended interval of the chiral transition in the quantum loop model is interesting and inspiring, as the biggest challenge of chiral transitions in the context of Rydberg arrays is their extremely small interval [27, 28].

4 Manipulating the nature of $\mathbb{Z}_{4}$ transitions

In the previous section we presented the phase diagram of the QLM for a special case of $t^{\prime}=t$. Let us now investigate by changing the value of $t^{\prime}$ how the nature of the $\mathrm{Z}_{4}$ transition is altered. In this section we will focus on three cases: first we consider $t^{\prime}=0$ associated with an additional fragmentation of the Hilbert space and for which we observe the floating phase; then we present the results for $t^{\prime}=2$ where we see no signature of a chiral transition; finally, we fix $\delta=50$ and study the QLM as a function of $\theta$ and $t^{\prime}$ to track how the nature of the $\mathrm{Z}_{4}$ transition changes.

Figure 7: Sketches and density profiles of the $\mathbb{Z_{4}}$ leg-dimerized phase. (a) Four possible configurations of the plaquette states (b) The four possible states of the columnar leg phase. (c) Local density of the plaquette (blue) and columnar leg (red) states inside the $\mathbb{Z}_{4}$ phase for $t^{\prime}=0$ and $t^{\prime}=1$. We present only the central part of the ladder for $N=199$, $\delta=50$ and respectively $\theta=1.9$ and $\theta=1.4$.

4.1 Transition to the plaquette phase at $t^{\prime}=0$

We start our analysis with $t^{\prime}=0$ where $\mathbb{Z}_{4}$ columnar leg dimerized states are completely disconnected from other sectors in the Hilbert space. As a result, the $\mathbb{Z}_{4}$ ordered phase for $t^{\prime}=0$ corresponds solely to a plaquette phase as demonstrated in Fig.7. Apart from that the main features of the phase diagram presented in Fig.8(a) to a certain extent resemble the previous case with $t^{\prime}=1$. We also detect the $\mathbb{Z}_{2}$ rung-dimerized phase separated from both $\mathbb{Z}_{4}$ plaquette and disordered NNN-Haldane phases by a pair of Ising transitions. The rung-dimerized phase disappears and two Ising transitions meet at the multi-critical Ashkin-Teller point followed by the chiral transition. However, this time, the properties of the Ashkin-Teller point and, in turn, the length of the chiral transition is different.

Figure 8: Phase diagrams of the QLM defined by Eq.(1) for (a) $t^{\prime}=0$ and (b) $t^{\prime}=2$. In each case there are three gapped phases: disordered NNN-Haldane, $\mathbb{Z}_{2}$ rung-dimerized and $\mathbb{Z}_{4}$ leg-dimerized. For $t^{\prime}=0$ columnar leg states are fully disconnected from the rest of the Hilbert space, so the $\mathbb{Z}_{4}$ ordered phase consists of the plaquette phase only. In both cases the $\mathbb{Z}_{2}$ ordered phase is separated from the disordered and the $\mathbb{Z}_{4}$ ordered phases by continuous Ising transitions (green circles). (a) For $t^{\prime}=0$ two Ising transitions meet at the multi-critical Ashkin-Teller point (purple circle), beyond which the transition is chiral (blue squares) until eventually it turns into a floating phase (red pentagons). There is some uncertainty in the location of the Ashkin-Teller point indicated with light pink diamonds. (b) For $t^{\prime}=2$, as soon as the $\mathbb{Z}_{2}$ phase disappears the transition between the disordered and $\mathbb{Z}_{4}$ phases is first order (dark gray hexagon).

To study the details of the Ashkin-Teller multi-critical point and the transition beyond it we follow the same protocols as previously and compute the scaling of the inverse of the correlation length $1/\xi$ and the product $|q/\pi-1/2|\times\xi$ for various cuts along the transition. A few typical examples for $t^{\prime}=0$ are presented in Fig.9 (a)-(d). Along the cut at $\delta=1.8$ that goes across the Ashkin-Teller critical point we see the product $|q/\pi-1/2|\times\xi$ going to zero (see Fig.9 (b)) in agreement with the conformal transition, while the critical exponent $\nu\approx 0.71$ is noticeably larger than in the previous case, as demonstrated in Fig.9 (a).

Beyond the Ashkin-Teller point we find that the transition is chiral for an extended interval. In Fig.8(c) we provide an example of $|q/\pi-1/2|\times\xi$ scaling to a finite-value at the transition. Further away from the Ashkin-Teller point we see a signature of the floating phase with a clear divergence of $|q/\pi-1/2|\times\xi$ for $\delta=50$, presented in Fig.8(d). Note that in this case the finite-size effect is opposite to the one we observed for the chiral transition in Fig.6 (i).

According to our data the floating phase opens up between $\delta=10$ and $\delta=20$, implying that the chiral transition for $t^{\prime}=0$ is shorter than the one for $t^{\prime}=1$ (though it is still remarkably extended). This observation fully agrees with the theory prediction that if the Ashkin-Teller conformal point becomes closer to the point with $\nu\approx 0.8$ where it crosses the Lifshitz line, the interval of the chiral transition shortens [20, 22].

Figure 9: Numerical evidences of various types of quantum phase transition between the $\mathbb{Z}_{4}$ ordered phase and the disordered one for (a)-(d) $t^{\prime}=0$ and (e)-(g) $t^{\prime}=2$. (a) Inverse of the correlation length $1/\xi$ at the Ashkin-Teller point. (b)-(d) The product $|q/\pi-1/2|\times\xi$ across (a) conformal Ashkin-Teller transition, (b) direct chiral transition, and (c) the floating phase. Dashed vertical lines indicate the location of the critical point extracted by fitting $1/\xi$. Error bars of $|q/\pi-1/2|$ estimated as $2\times\xi^{2}/N^{2}$ are only shown if they exceed the size of the symbols. (e)-(g) Numerical evidences of the first order transition at $\delta=50$ that includes (e) finite correlation length $\xi$ across the transition; (f) a kink in the ground-state energy per site, and (g) jump in the order parameter. Dashed lines in (e) indicate the power-law fit. (f) In addition to the finite-size energy per site $E_{N}/N$ we also compute the bulk estimate as $(E_{N_{1}}-E_{N_{2}})/(N_{1}-N_{2})$ (purple pentagons) that significantly reduces the boundary effect. (g) Vertical dashed lines indicate the location of the transitions. (h) Scaling of the $1/\xi$ for $t^{\prime}=2$ through the multi-critical point. Although the transition appears continuous, the critical exponent $\nu$ are outside the range of the Ashkin-Teller model and might signal a weak first order transition.

To summarize, we find that the nature of the multi-critical Ashkin-Teller point, and in turn the length of the chiral transition, can be manipulated by the $t^{\prime}$ responsible for the appearance of double dimers on the legs of the ladder. More precisely, we observe that by switching from $t^{\prime}=0$ to $t^{\prime}=1$ the critical exponent $\nu$ decays from $\nu\approx 0.71$ to $\nu\approx 0.68$ accompanied by a significant increase of the interval of chiral transition. However, one can also notice that at $t^{\prime}=1$ the Ashkin-Teller point is already pretty close to the four-state Potts point with $\nu=2/3$[18] and it is therefore natural to question the fate of the chiral transition if we increase $t^{\prime}$ even further.

4.2 First order transition for $t^{\prime}=2$

By increasing $t^{\prime}$ the density of double dimers on the legs in the $\mathbb{Z}_{4}$ phase is increasing. As a result the $\mathbb{Z}_{4}$ phase appears earlier and the transition takes place at smaller $\theta$. The phase diagram for $t^{\prime}=2$ is presented in Fig.8(b). Like before we observe the rung-dimerized $\mathbb{Z}_{2}$ phase to be separated by a pair of Ising transitions from the disordered NNN-Haldane phase on one side and the $\mathbb{Z}_{4}$ leg-dimerized phase on the other. However, starting from the point where two Ising transitions meet the two phases are separated by a first order transition. In Fig.9(e)-(g) we present numerical evidences for $\delta=50$ supporting this claim. First, the correlation length $\xi$ does not diverge at the transition as shown in Fig.9(e). Second, the ground-state energy has a kink typical for a first order transition that becomes more pronounced upon approaching the thermodynamic limit, as demonstrated in Fig.9(f). In addition to the finite-size results for $E_{N}/N$ we also consider the bulk energy per site $(E_{N_{1}}-E_{N_{2}})/(N_{1}-N_{2})$ for two different system sizes, effectively eliminating boundary effects. Finally, in Fig.9(g) we show that the dimerization on the legs - the order parameter of the $\mathbb{Z}_{4}$ leg-dimerized phase - has a clear jump at the transition.

Upon approaching the multi-critical point signatures of the first order transition weaken. In particular, the scaling of the inverse of the correlation length presented in Fig.9(h) looks very similar to a continuous transition. However, the numerically extracted critical exponent $\nu\approx 0.62$ lies outside the range $2/3\leq\nu\leq 1$ defined by the Ashkin-Teller critical theory, with the lowest value $\nu=2/3$ corresponding to the 4-state Potts point. So, what is the nature of the transition beyond this point? Recent study of the Ashkin-Teller model has shown that the system along the self-dual line becomes gapped [21]. This is compatible with the first order transition that we observe with the finite correlation length and co-existing ground-states.

4.3 Phase diagram as a function of $t^{\prime}$

Figure 10: (a) Phase diagram of the QLM as a function of $t^{\prime}$ and $\theta$ for $\delta=50$. $t^{\prime}_{c}\approx 1.37$ indicates the end point of the chiral transition (black star). (b) and (c) Product $|q/\pi-1/2|\times\xi$ for a vertical across (b) a floating phase and (c) a chiral transition. (d) Minimum in $1/\xi$ as a function of $t^{\prime}$ along the first order transition, solid line states for the power-law for with critical exponent s; (e) same results in the log-log scale.

For the three examples we considered thus far, i.e. $t^{\prime}=0,1$ and $2$, we observe that for large $\delta$ the transition out of the $\mathbb{Z}_{4}$ ordered phase can be either through a floating phase ($t^{\prime}=0$), or direct and chiral, ($t^{\prime}=1$) or first order ($t^{\prime}=2$). In this section we study how one type of the transition goes into another as a function of $t^{\prime}$ and take a closer look on how the first order transition opens up.

Since the floating phase appears only for very large values of $\delta$ we focus on $\delta=50$, which is also sufficiently far to reduce a crossover effect from the multi-critical point where the rung-dimerized phase disappears. Fig.10(a) presents the ground-state phase diagram, as a function of $\theta$ and $t^{\prime}$, that summarizes our results. The floating phase that we observe for small $t^{\prime}$ (see Fig.10(b) ) turns into a chiral transition (see Fig.10(c) ) between $\theta\approx 1.60$ and $t^{\prime}\approx 0.7$, and $\theta\approx 1.53$ and $t^{\prime}\approx 0.8$. In between these points we expect the multi-critical Lifshitz point, with dynamical critical exponent $z=3$, to be located. Determining the precise location of this point is an extremely challenging computational task. Then, starting from $t^{\prime}_{c}\approx 1.37$ we detect a first order transition with finite correlation length at the transition.

By taking a closer look at the minimum in $1/\xi$ we study how the characteristic length scale develops along the first order line and away from the end point of the chiral transition. We identify the location of the transition and minimum of $1/\xi$ by fitting both sides of the transition with a power law function and calculating the intersection of these fits. In Fig.10(d) we show how the inverse of this characteristic length develops along the transition. We fit the data point with the power-law fit that allows us to estimate the location of the critical point $t^{\prime}_{c}$ and its scaling. In Fig.10(e) we present the same set of data but in a log-log scale.

As a final remark let us mention that we observe a clear coexistence of two domains in the disordered phase, another clear signature of a first order transition. In Fig.11 we show the density profiles and entanglement entropy at the first order transition for $t\prime=1.8$. Close to the edges of the chain we observe a plaquette domain while in the bulk it corresponds to the NNN-Haldane phase. Furthermore, the location of the domain walls clearly matches the peaks in the entanglement entropy.

Figure 11: Numerical evidence for the coexistence of two domains in the disordered phase at the first order transition for $t^{\prime}=1.8$. Density profiles of (a) an entire plaquette and (b) occupation of one dimer on the leg. (c) Entanglement entropy throughout the chain. Quantities are computed for $N=1999$ sites, $\delta=50$ and $\theta=0.184$.

5 Conclusion

In this paper we investigate the nature of the transition out of $\mathbb{Z}_{4}$ leg-dimerized phases in quantum loop models constrained to two dimers per node on a zig-zag ladder. We report very rich critical behavior. First of all, the $\mathbb{Z}_{4}$ order can be destroyed in two steps via Ising transitions and a rung-dimerized phase - another gapped phase with spontaneously broken $\mathbb{Z}_{2}$ symmetry. Realization of other scenarios depends on the parameter $t^{\prime}$, that controls the density of double-dimers on the legs of the ladder. When $t^{\prime}$ is not too large we observe a multi-critical Ashkin-Teller point followed by an extended interval of a $\mathbb{Z}_{4}$ exotic chiral transition. For small $t^{\prime}$ the transition can also be through the floating phase - incommensurate Luttinger liquid phase bounded by the Kosterlitz-Thouless and Pokrovsky-Talapov transitions. However, for large values of $t^{\prime}$ we see no Ashkin-Teller conformal point, nor the chiral transition or the floating phase. Instead, our results predict a first order transition.

The way how this first order transition opens up is actually very interesting. The characteristic length scale diverges algebraically upon approaching the end point of the chiral transition and not exponentially. The latter scenario would suggest the presence of a marginal operator, while the former one, what we actually observe in Fig.10(a), hints towards an additional relevant operator present in the system. As a consequence, we expect the chiral transition to terminate with critical end point that might have different underlying critical theory than the continuous chiral transition above it. Numerically this question is very challenging and goes beyond the scope of this work, but we hope that our observation of the chiral transition turning into the first order (to the best of our knowledge it is the first reported example) will advance the development of the field theory of chiral transitions and stimulate further numerical studies in this direction.

Our results obtained for the QLM - the toy model of quantum magnets that completely discards the magnetic degrees of freedom - provide a number of important messages to advance the study of $\mathbb{Z}_{4}$ transitions in more realistic models including, for instance, a Heisenberg spin-1 ladder. First of all, we have confirmed our original conjecture that the VBS transition between leg-dimerized phase and NNN-Haldane phase can be in the $\mathbb{Z}_{4}$ chiral universality class. The remarkable extent of the chiral transition that we observed significantly increases its chances to be detected in the Heisenberg-like models. Both plaquette, with a partial dimerization on legs, and fully-dimerized double-dimer states are present in the $\mathbb{Z}_{4}$ phase of QLM - this opens up a wider range of interactions in Heisenberg models to realize the $\mathbb{Z}_{4}$ phases. At the same time our results point out that the presence of plaquette states is crucial as strong fully-dimerized state might be responsible for a relevant perturbations that destroys chiral criticality.

Let us also highlight the floating phase, emergent at the boundary of the $\mathbb{Z}_{4}$ when the plaquette states are dominant. Recently, magnetic floating phases predicted in half-integer $J_{1}-J_{2}$ spin chains attracted a lot of attention among theorists and experimentalists [47, 48, 49, 50]. Our results in quantum loop models suggest a non-magnetic counterpart of this exotic phenomena, opening an opportunity to realize them in integer spin-chains.

Furthermore, our results predict that the transition between the $\mathbb{Z}_{2}$ rung-dimerized and $\mathbb{Z}_{4}$ leg-dimerized phases, if continuous, will be in the Ising universality class. To the best of our knowledge this possibility has never been reported yet in the context of frustrated Haldane chains. At the same time, the second Ising transition that we observed in QLMs - between the rung-dimerized phase and the NNN-Haldane phase has been successfully realized in a $J_{1}-J_{2}$ Haldane chain in the presence of biquadratic[36] or three-site interactions[35]. In addition, our result highlight a chance to realize a non-magnetic Ashkin-Teller transition at the critical point where the two Ising transitions meet, of course, under the condition that at this multi-critical point magnetic degrees of freedom would still have a finite gap, which is nevertheless totally feasible.

Finally, let us highlight the possibility to tune the nature of the multi-critical point and in turn the phase transition beyond it by controlling the density of the double leg-dimerized states with $t^{\prime}$ in QLM. Similar mechanisms that allow to tune the Ashkin-Teller point, and the extent of the chiral transition beyond it, has recently been proposed in the context of multi-component Rydberg atoms[28]. Remarkably, the two models are deeply connected. Firstly, the fusion rules in these two models are identical, implying that the two Hilbert spaces can be rigorously mapped onto each other. The mapping of the two Hamiltonians is not rigorous, however, because of an additional three-site interaction between the Rydberg atoms arising from the plaquette term parameterized by $\theta$ (see Appendix C for details). Furthermore, a tunable $\mathbb{Z}_{4}$ chiral transition can also be realized with a single-component Rydberg ladder. For this, one has to ensure nearest-neighbor blockade on one leg and next-nearest-neighbor blockade on the second one. There might be several ways of realizing this in modern experiments, but perhaps the simplest is to curve a ladder in a circle: this way the inner chain will have smaller inter-atomic distances and thus stronger intra-chain interaction than the outer one. The corresponding blockade model also has a rigorously equivalent Hilbert space to the one of the QLM, but the mapping between the interaction terms in the two models is not mathematically exact. Nevertheless, all three families of models - QLM, an array of multi-component Rydberg atoms, and a single-component Rydberg atoms arranged in a curved ladder - open up an exciting opportunity to manipulate quantum phase transitions.

Appendix A DMRG with quantum loop model constraints

A.1 Implementing quantum loop constraints

To fully profit from the fragmented Hilbert space of QLMs, we implement constraints to which these models are subjected directly into key components of the algorithm, following [38]. We briefly recap the procedure of how this can be accomplished for a spin-1 QLM. QLMs are constrained since each site of the original spin-1 zig-zag lattice forms two, and only two, dimers with other sites. We further constrain the model by limiting the dimer formation to nearest and next-nearest neighbouring sites only. There are several consequences of this. First is that instead of considering spins on the lattice as the relevant degrees of freedom we can model the system by considering the configurations of dimers only. One possible basis to this is by considering for each site all possible dimer configurations on a rung and the preceding leg - for $N$ spins in the lattice there are $N-1$ QLM degrees of freedom. In total there six combinations possible, which we depict in Fig. 12 (a). We use these as the local Hilbert space for the DMRG algorithm.

Figure 12: (a) Local Hilbert space consists of dimer occupations on a rung and the preceding leg and has a dimension of $d=6$. (a) Quantum numbers of the left environment that we use to label separate sectors of the Hilbert space. (c) Fusion graphs for left and right environments. There is a one-to-one correspondence between the quantum numbers of the left and those of the right.

Second, constraints cause the full Hilbert space to be fragmented into sectors corresponding to different dimer configurations. This fragmentation is not merely a global property but also holds locally during distinct parts of the DMRG algorithm. Take for example the bipartition of the total system into left- and right-normalized blocks $-$ we refer to these as the left and right environments (see [42] for terminology). In the course of sweeping through the chain, one environment growths and the other shrinks. We construct these environments iteratively, in which, during each iteration we add a new site to the environment. In tensor language this corresponds to adding a tensor, and its Hermitian conjugate, of the MPS, and the Matrix Product operator (MPO) to the environment. Each iteration, we label all allowed states of both environments by a set of quantum numbers, corresponding to the different Hilbert space sectors. In total, there are five of such, which we number from 0 to 4. For the left environment these are - for elaborative purposes we use a Valence-Bond-Singlet notation where one spin-1 site is split up into two spin-1/2 dots: 0 has no free dots on either side of the chain, 1 has one free dot on both sides, 2 labels the state in which there are two free dots on the site added, 4 has two free dots on both sides of the chain, and 3 labels the state in which there are two free dots, but such that they do not belong to the last but second last site instead. We depict these quantum numbers for the both environments in Fig. 12 (b). We note that the labels 0, 1 and 2 are the ones used for the quantum loop model studied in [38]. Quantum labels 3 and 4 are added to allow the formation of the $\mathbb{Z}_{4}$ columnar phase.

Under the addition of a site to the environment, which translates to adding dimers, a quantum number is mapped to a new one. Not any basis is allowed added to any quantum number as a result of the QLM constraints. Take for example the quantum number zero in which there are no free dots available to form dimers with. As a result we can only add $|\psi_{1}\rangle$ - the basis state without any dimers. Doing so causes the new environment to now have two free dots on one side of the chain and is thus labeled 2. By following the same procedure for the other quantum numbers, we can construct a so called fusion graph, which describes the fusion between quantum numbers under the addition of a bases. In the left panel of Fig. 12 (c) we list the fusion graph for the left environment. We also constructed the fusion graph for the right environment (see right panel) by reversing the direction of all arrows and renaming the quantum numbers to be the same as for the left environment. We note that there is a one-to-one correspondence between the sectors Hilbert space sectors of the environments. Note that the left and right environment are to be connected through the following rule: (L, R) = {(0,2), (1,1), (2,0), (3,4), (4,3)}, meaning that quantum number 0 on the left can only connect to quantum number 2 on the right, etc.

Third, local constraints are also accounted for in determining optimal tensor(s) and an estimate of the ground state energy each iteration of the DMRG algorithm. In this process the many-body Hilbert space is mapped to a reduced one in the form of a so-called effective Hamiltonian, which is then diagonalized. The basis in which this eigenvalue problem is solved is subject to constraints as well. Consequently, the reduced Hilbert space is greatly reduced, speeding up the diagonalization.

Lastly, splitting up the Hilbert space into distinct section allows to write the tensors in a block-diagonal form. There are two main advantages of this. The first one being that numerical operations central in the DMRG algorithm, such as tensor contraction and singular value decomposition, can be carried out on each block separately [51], making them computationally cheaper. Secondly, the blocks require fewer memory than the full tensor as well. In turn, this makes it possible to simulate more states.

A.2 Matrix product operator

The quantum loop Hamiltonian, defined by Eq.(1), in terms of the local Hilbert space can be rewritten as

$$\mathcal{H}_{\mathrm{QLM}}=\sum_{i}(-tp_{i}^{(1)}p_{i+1}^{(2)}p_{i+2}^{(3)}-t^{\prime}q_{i}^{(1)}q_{i+1}^{(2)}q_{i+2}^{(3)}+\mathrm{h.c.})+\delta r_{i}-\theta s_{i}^{(1)}s_{i+1}^{(2)}s_{i+2}^{(3)},$$ (7)

where $p^{(1)}=f(4,1)$, $p^{(2)}=f(3,6)$, $p^{(3)}=f(2,6)$, $q^{(1)}=f(5,4)+f(2,3)$, $q^{(2)}=f(1,3)$, $1^{(3)}=f(5,2)+f(4,3)$, $s^{(1)}=f(4,4)$, $s^{(2)}=f(3,3)$, $s^{(3)}=f(2,2)$ and $r=f(5,5)$ and $f(n,m)$ is a $6\times 6$ matrix for which the element in the $n$’th row and the $m$’th column is 1. To illustrate

$$f(4,1)=\begin{pmatrix}0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ 1&0&0&0&0&0\\ 0&0&0&0&0&0\\ 0&0&0&0&0&0\\ \end{pmatrix}.$$ (8)

We converted the above Hamiltonian to its MPO form, for which we formulated its tensors as:

$$W=\setcounter{MaxMatrixCols}{12}\begin{pmatrix}\mathcal{I}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ p^{(3)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ (p^{(3)})^{\dagger}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ q^{(3)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ (q^{(3)})^{\dagger}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ v^{(3)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \cdot&p^{(2)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \cdot&\cdot&(p^{(2)})^{\dagger}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \cdot&\cdot&\cdot&q^{(2)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \cdot&\cdot&\cdot&\cdot&(q^{(2)})^{\dagger}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \cdot&\cdot&\cdot&\cdot&\cdot&v^{(2)}&\cdot&\cdot&\cdot&\cdot&\cdot&\cdot\\ \delta r&\cdot&\cdot&\cdot&\cdot&\cdot&-tp^{(1)}&-t(p^{(1)})^{\dagger}&-t^{\prime}q^{(1)}&-t^{\prime}(q^{(1)})^{\dagger}&v^{(1)}&\mathcal{I}\\ \end{pmatrix},$$ (9)

with $\cdot$ denoting a $6\times 6$ matrix with zeros elements only and $\mathcal{I}$ being the $6\times 6$ identity matrix.

Appendix B Locating the disorder line

Figure 13: Locating the disorder line via a two step process. (a) First we define a top and bottom segment in the correlation function $C_{i,j}$ in the crossover region. Both of these have their slope fitted to extract their respective correlation lengths $\xi_{\mathrm{top}}$ and $\xi_{\mathrm{bottom}}$. (b) We associate disorder line to be the point where $\xi_{\mathrm{top}}=\xi_{\mathrm{bottom}}$. In the incommensurate domain of the disordered phase we observe only one correlation length.

The disorder line separates the commensurate from the incommensurate part of the disordered phase and it is characterized by a kink, frequently a minimum, in the correlation length [52, 53]. Unfortunately, such a kink is not always sharp but smeared out, making a good estimate of detecting such a kink near impossible. We instead extract an estimate of the disorder line in a two step process, which we show in Fig. 13. Let us briefly illustrate it. In vicinity of the disorder line in the commensurate domain, there are two short range orders present in the correlation function; the top part has a smaller correlation length and shows incommensurate correlations while the bottom part reflects the actual order, for which we denote $\xi_{\mathrm{top}}$ and $\xi_{\mathrm{bottom}}$ as their respective correlation lengths (see Fig.13(a)). We refer to this domain as the crossover region. In the cross-over region $\xi_{\mathrm{top}}$ and $\xi_{\mathrm{bottom}}$ most often differ while the in the incommensurate part there is no distinction between the top and bottom anymore and there is just a single correlation length. In other words, we associate the point where $\xi_{\mathrm{top}}$ terminates as the estimate of the disorder line, which we define as $\xi_{\mathrm{top}}=\xi_{\mathrm{bottom}}$. In Fig.13(b) we show an example of this. We note that this is merely an estimate and not the actual disorder line. However, this estimate is consistent with expectations since it approaches the multi-critical Ashkin-Teller point. We are not able to extract the disorder line near this point and leave it out of the data. For indicative purposes we do sketch a possible trajectory to the multi-critical Ashkin-Teller point as a dashed line (see Fig.2 and Fig.8(a))

We locate the disorder for $N=599$ sites and calculate the correlation function with $n_{i}=|\includegraphics[height=8.53581pt]{basis_h_3.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_3.pdf}|+|\includegraphics[height=8.53581pt]{basis_h_6.pdf}\rangle\langle\includegraphics[height=8.53581pt]{basis_h_6.pdf}|$. We observe that different operators can result in different estimates of the disorder line, although all of them terminate in the vicinity of the Ashkin-Teller point.

Appendix C Mapping to Rydberg atoms

Quantum loop models on zig-zag ladders without states containing two dimes on any legs, such as for $t\prime=0$, are rigorously mapped to a Rydberg chain [38]. Let us briefly repeat the argument. For each rung and the legs it can form a triangle with, i.e. those directly above and below it, we associate a Rydberg atom in the excited state if all three are not occupied by any dimers. All other possibilities correspond to Rydberg atoms in the ground state. When states wth two dimers on the legs are included on the other hand, this mapping does not suffice to correctly map the Hilbert space of QLMs to that of Rydberg atoms since it does not allow to make a distinction between the $\mathbb{Z}_{4}$ ordered plaquette and columnar-dimer phases. To overcome this problem we use map to a Rydberg ladder instead, where we associate every rung with two legs in the quantum loop model to a rung with two Rydberg atoms. The same mapping as described above is used to map all states in the Hilbert space without any legs occupied by two dimers but applied to Rydberg atoms the top atom of the rung only. This yields the possibility to describe the NNN-Haldane phase, the $\mathbb{Z}_{2}$ rung-dimerized phase and the $\mathbb{Z}_{4}$ plaquette phase. Added to this, for a rung that has both legs it can form a triangle with occupied by two dimers each we associate the bottom Rydberg atom in the rung to be excited only. In Fig.14(a) we summarize the above with three illustrative sketches and we show the NNN-Haldane phase, rung-dimerized phase and the plaquette and columnar-dimerized phases mapped to a circular Rydberg ladder in Fig.14(b)-(e) (below we explain the choice for a circular geometry).

With this mapping we derive the constraints to which the Rydberg atoms are subjected to. We describe these in terms of an effective $n$-site blockade model following hard boson statistics. Let us define the operator $c_{i}^{\dagger}$ ($d_{i}^{\dagger}$) as the one that excites a Rydberg atom from the ground to the excited state on the outer (inner) chain of the $i$’th rung of the lattice. By doing so we can formulate the constrains as:

$$n_{i}^{\mathrm{outer}}n_{i+1}^{\mathrm{outer}}=n_{i}^{\mathrm{inner}}n_{i+1}^{\mathrm{inner}}=n_{i}^{\mathrm{inner}}n_{i+2}^{\mathrm{inner}}=n_{i}^{\mathrm{outer}}n_{i}^{\mathrm{inner}}=n_{i}^{\mathrm{outer}}n_{i+1}^{\mathrm{inner}}=0,$$ (10)

where $n_{i}^{\mathrm{outer}}=c^{\dagger}_{i}c_{i}$ and $n_{i}^{\mathrm{inner}}=d^{\dagger}_{i}d_{i}$ measure the occupation of a particle in the outer and inner chain respectively. From this set of constraints it is clear that Rydberg atoms on the inner chain experience longer range blockade than those on the outer chain $-$ two Rydberg atoms in the outer chain have to be separated by at least two sites while three in the inner. With that in mind we propose a circular geometry for the Rydberg ladder as shown in Fig. 14(b)-(e) to naturally impose these constraints. Furthermore, all relevant states are realizable in such a lattice configuration.

Figure 14: (a) Map between quantum loop models and a Rydberg ladder. Shown dimer configurations in outer two are not unique for this mapping (b) Empty Rydberg ladder configuration corresponding. Corresponds to the NNN-Haldane phase. (c) $\mathbb{Z}_{2}$ rung-dimerized phase. $\mathbb{Z}_{4}$ plaquette (d) and columnar-dimer phase (e).

To construct a Hamiltonian formulated in terms of hard bosons operators from the quantum loop model we use the mapping and the set of constraints. We start with the first term that swaps a pair of dimers. In the Rydberg ladder, this corresponds to bringing a Rydberg atom from its excited state to the ground state in the outer chain only. Such that

$$|\includegraphics[height=8.53581pt]{sld_plaquette.pdf}\rangle\langle\includegraphics[height=8.53581pt]{srd_plaquette.pdf}|+\mathrm{h.c}=c_{i}+c_{i}^{\dagger}.$$ (11)

Following similar reasoning, but now for Rydberg atoms in the inner chain only, the second kinetic term is mapped as

$$|\includegraphics[height=8.53581pt]{double_leg.pdf}\rangle\langle\includegraphics[height=8.53581pt]{plaquette.pdf}|+\mathrm{h.c.}=n_{i-2}^{\mathrm{outer}}(d_{i}+d_{i}^{\dagger})n_{i+2}^{\mathrm{outer}},$$ (12)

where the occupation number operators ensure that the columnar-dimerized phase can only be mapped to from the plaquette one. Moving on to first the potential term, in quantum loop models a rung with two dimers on them is preceded an followed by a rung and two legs that are not occupied by any dimers at all. In a Rydberg chain this translates to the Rydberg atoms corresponding to rung with two dimers on it being in the ground state while those atoms in the outer chain adjacent to it in the excited state:

$$|\includegraphics[height=8.53581pt]{double_rung.pdf}\rangle\langle\includegraphics[height=8.53581pt]{double_rung.pdf}|=n_{i-1}^{\mathrm{outer}}(1-n_{i}^{\mathrm{outer}})n_{i+1}^{\mathrm{outer}}=n_{i-1}^{\mathrm{outer}}n_{i+1}^{\mathrm{outer}},$$ (13)

where the constraint $n_{i}^{\mathrm{outer}}n_{i-1}^{\mathrm{outer}}=0$ is used to simplify the expression. For the second potential term, the three rungs part of the plaquette correspond to Rydberg atoms in the ground state on both the outer and inner chain while the neighbouring atoms of the outer chain on both sides of the plaquette are excited:

$$|\includegraphics[height=8.53581pt]{plaquette.pdf}\rangle\langle\includegraphics[height=8.53581pt]{plaquette.pdf}| =n_{i-2}^{\mathrm{outer}}(1-n_{i-1}^{\mathrm{outer}})(1-n_{i}^{\mathrm{outer}})(1-n_{i+1}^{\mathrm{outer}})n_{i+2}^{\mathrm{outer}}(1-n_{i}^{\mathrm{inner}})$$ (14)

$$=n_{i-2}^{\mathrm{outer}}(1-n_{i}^{\mathrm{outer}})n_{i+2}^{\mathrm{outer}}-n_{i-2}^{\mathrm{outer}}n_{i}^{\mathrm{inner}}n_{i+2}^{\mathrm{outer}},$$

where again the blockade constraints were used to simplify the expression. The couping $n_{i-2}^{\mathrm{outer}}(1-n_{i}^{\mathrm{outer}})n_{i+2}^{\mathrm{outer}}$ is likely to be much smaller than the first potential term since the coupling is between particles more distant from each other in a system where the interaction is distant dependent. We note that the term $n_{i-2}^{\mathrm{outer}}n_{i}^{\mathrm{inner}}n_{i+2}^{\mathrm{outer}}\approx n_{i}^{\mathrm{inner}}$ can not be well approximated since it helps to stabilize the plaquette phase - crucial for the realization of the $\mathbb{Z}_{4}$ chiral transition. Summarizing the above, (1) is mapped to a circular Rydberg ladder in accordance to

$$\mathcal{H}_{\mathrm{Rydberg}}\approx\sum_{i}-t(c_{i}+c_{i}^{\dagger})-t^{\prime}n_{i-2}^{\mathrm{outer}}(d_{i}+d_{i}^{\dagger})n_{i+2}^{\mathrm{outer}}+(\delta-\theta)n_{i}^{\mathrm{outer}}n_{i+2}^{\mathrm{outer}}+\theta n_{i-2}^{\mathrm{outer}}n_{i}^{\mathrm{inner}}n_{i+2}^{\mathrm{outer}}.$$ (15)