Competing states in the S = 1 / 2 S=1/2 triangular-lattice J 1 J_{1} - J 2 J_{2} Heisenberg model: a dynamical density-matrix renormalization group study Shengtao Jiang (gbsn蒋晟韬), Steven R. White, Steven A. Kivelson, Hong-Chen Jiang Abstract Previous studies of the S = 1 / 2 S=1/2 triangular-lattice J 1 J_{1} – J 2 J_{2} Heisenberg antiferromagnet have inferred the existence of a non-magnetic ground-state phase for an intermediate range of J 2 J_{2} , but disagree concerning whether it is a gapped ℤ 2 \mathbb{Z}_{2} quantum spin liquid (QSL), a gapless (Dirac) QSL, or a weakly symmetry-broken phase. Using an improved dynamical density-matrix renormalization group method, we investigate the relevant intermediate J 2 J_{2} regime for cylinders with circumferences from 6 to 9. Depending on the initial state and boundary conditions, we find two distinct variational states. The higher energy state is consistent with a Dirac QSL. In the lower-energy state, both the static and dynamical properties are qualitatively similar to the magnetically ordered state at J 2 = 0 J_{2}=0 , suggestive of either a weakly magnetically ordered non-QSL or a gapped QSL proximate to a continuous transition to such an ordered state. Introduction. – Half a century ago, Anderson initiated the study of quantum spin liquids (QSL) with his seminal proposal of the resonating valence bond state in the triangular-lattice antiferromagnet (AFM) Anderson 1973 ; Kivelson and Sondhi 2023 , prototypical of what is now known as a gapped \mathbb{Z}_{2} QSL Broholm et al. 2020 . In this exotic state of matter, strong quantum fluctuations, facilitated by geometrical frustration in the triangular lattice, preclude conventional magnetic order even at zero temperature. Despite the later discovery that the ground state of the nearest-neighbor (NN) Heisenberg AFM has the classical 120 ∘ Néel order Zheng et al. 2006 ; J. Miyake 1992 ; Chubukov et al. 1994 ; Bernu et al. 1994 ; Capriotti et al. 1999 ; White and Chernyshev 2007 , the triangular lattice AFM remains one of the central focuses in the search for QSL and other exotic quantum states, both theoretically Kalmeyer and Laughlin 1987 ; Moessner et al. 2000 ; Peng et al. 2021 ; Gong et al. 2019 ; Szasz et al. 2020 ; Zhu et al. 2017 ; Zhu et al. 2018 ; Maksimov et al. 2019 ; Chi et al. 2024 ; Gallegos et al. 2025 ; Huang et al. 2024 ; Ulaga et al. 2025 ; Gao et al. 2024 and experimentally Scheie et al. 2024a ; Facheris et al. 2023 ; Chen et al. 2025 ; Li et al. 2020 ; Gao et al. 2022 ; Sheng et al. 2022 ; Xiang et al. 2024 ; Bag et al. 2024 ; Li et al. 2015a ; Scheie et al. 2024b ; Scheie et al. 2024c ; Shen et al. 2016 ; Ma et al. 2016 ; Zhong et al. 2019 ; Park et al. 2024 . While the simplest NN Heisenberg model fails to achieve a QSL state, quantum fluctuations are apparently enhanced by the inclusion of an AFM second NN exchange J_{2} . Numerous studies Zhu and White 2015 ; Iqbal et al. 2016 ; Mishmash et al. 2013 ; Saadatmand and McCulloch 2016 ; Hu et al. 2015 ; Li et al. 2015b ; Hu et al. 2016 ; Tang et al. 2022 ; Wietek and Läuchli 2017 ; Kaneko et al. 2014 ; Jiang and Jiang 2023 report a nonmagnetic phase, frequently interpreted as a QSL, in an intermediate J_{2} range 0.07\lesssim J_{2}\lesssim 0.16 , although the precise phase boundaries and whether the putative QSL is gapped or gapless remain unsettled. Several density-matrix renormalization group (DMRG) studies Zhu and White 2015 ; Hu et al. 2015 ; Saadatmand and McCulloch 2016 ; Jiang and Jiang 2023 produced suggestive evidence of a gapped QSL, leading to the proposed phase diagram sketched in Fig. 1 (a). Specifically, that the QSL has an emergent \mathbb{Z}_{2} gauge symmetry was inferred from the existence of two low-energy states on even-width cylinders, depending on boundary conditions – a feature that was interpreted as representing the two-fold topological degeneracy Thouless 1987 expected of a “ \mathbb{Z}_{2} QSL” on a cylinder. Additionally, a substantial spin triplet gap on finite-width cylinders was observed. However, on the basis of DMRG simulations with flux insertion Hu et al. 2019 and variational Monte Carlo Iqbal et al. 2016 ; Kaneko et al. 2014 , some subsequent studies argued that the relevant QSL is a gapless “Dirac QSL.” This view was further supported by studies that found evidence of low-energy magnetic excitations at the M points in the Brillouin zone (illustrated in Fig. 1 (a)) - consistent with the existence of nodal spinons at the Y points Ferrari and Becca 2019 ; Sherman et al. 2023 ; Drescher et al. 2023 ; Drescher et al. 2025 - and by other more complicated evidence of emergent U(1) gauge fields Wietek et al. 2024 . On the other hand, the latest DMRG simulations find that the spin triplet gap remains substantial even for cylinders with widths up to 10 Jiang and Jiang 2023 , behavior that is also seen in the spectral function on a six-leg cylinder Sherman et al. 2023 ; Drescher et al. 2023 . Thus, several fundamental questions remain concerning the nature of the intermediate J_{2} region. 1) Are the two nearly degenerate states seen in previous studies a reflection of the expected topological degeneracies of a \mathbb{Z}_{2} QSL or do they reflect the existence of two nearly degenerate, thermodynamically distinct “competing” ground-state phases? 2) Correspondingly, how do the two nearly degenerate states and their associated spectral properties evolve as the system size increases? 3) Most importantly, is the intermediate regime a QSL in the two-dimensional limit, or does it ultimately develop weak magnetic or valence-bond order? Figure 1: Top panel (a): Conjectured ground state phase diagram of the S=1/2 J_{1} - J_{2} Heisenberg model on the triangular lattice from previous studies Zhu and White 2015 ; Iqbal et al. 2016 ; Mishmash et al. 2013 ; Saadatmand and McCulloch 2016 ; Hu et al. 2015 ; Li et al. 2015b ; Hu et al. 2016 ; Tang et al. 2022 ; Wietek and Läuchli 2017 ; Kaneko et al. 2014 ; Jiang and Jiang 2023 , where an intermediate 0.07\lesssim J_{2}\lesssim 0.16 region was found with vanishing magnetic order, albeit with somewhat different phase boundaries and different putative QSLs. The middle point of this region J_{2}=0.12 , and specifically what QSL if any occurs there, is the focus of this paper. Insets of (a) show the XC6 cylinder with \vec{e}_{1} and \vec{e}_{2} being the unit vectors along its length and circumference, respectively, and the first Brillouin Zone with the high symmetry points marked. The Y point (midpoint between \Gamma and M ) is where the Dirac spinon node is located for a gapless Dirac QSL Song et al. 2019 ; Song et al. 2020 . Middle panel : Strength of the spin-spin correlation \langle\vec{S}_{i}\cdot\vec{S}_{i+\delta i}\rangle on three types of NN bonds (defined in (a)), averaged over the circumferential direction and plotted versus the length of the cylinder l_{x} , for four different states: (b) the 120 ∘ state at J_{2}=0 , (c) the LE state at J_{2}=0.12 , (d) the HE state at J_{2}=0.12 and (e) the striped state at J_{2}=0.2 . The strengths of bonds 1 and 2 are identical except in (e). Bottom panel : The equal-time spin structural factor S(\vec{q}) for (f) the 120 ∘ state at J_{2}=0 that has dominant peaks at K , (g) the LE state at J_{2}=0.12 with weaker peaks at K , (h) the HE state at J_{2}=0.12 with still weaker peaks at both K and M (marked by red circles), and (i) the striped state at J_{2}=0.2 with peaks at M . The white dashed lines mark the boundaries of the first Brillouin zone. In this Letter, we address these questions by systematically examining both the ground-state and dynamical properties of the S=1/2 triangular-lattice J_{1} - J_{2} Heisenberg model in the intermediate J_{2} regime using large-scale DMRG White 1992 ; White 1993 and dynamical DMRG Jeckelmann 2002 simulations, focusing on the representative value J_{2}=0.12 within the intermediate regime. Among other things, we establish the existence of two low-energy competing states that can be obtained from different initial conditions. This behavior is observed on cylinders with circumferences ranging from 6 to 9, i.e., it is not restricted to even-width cylinders as would be expected for a \mathbb{Z}_{2} QSL. To obtain dynamical information about the various quantum phases, we have improved and then utilized the dynamical DMRG Jeckelmann 2002 to probe their dynamical spin structure factors (DSSF). In this way, we have uncovered qualitatively distinct low-energy excitation spectra for the two states. Thus, in answer to question 1, the existence of two nearly degenerate states does not reflect a topological signature of a gapped \mathbb{Z}_{2} QSL. Instead, we will show that the higher energy state (HE state) most likely represents a Dirac QSL, characterized by low-frequency peaks of comparable intensity at both the K points and M points in the DSSF. In contrast, the lower energy state (LE state) exhibits both static and dynamical structure factors characterized by a single dominant peak at the K points, indicating it is probably a gapped \mathbb{Z}_{2} spin liquid, but could be a weakly ordered version of the J_{2}=0 phase close to a quantum critical point. Thus, while overall our findings lend support to the existence of a gapped \mathbb{Z}_{2} QSL as sketched in Fig. 1 (a), the possibility that the character of the ground-state may change for still larger L_{y} is non-negligible, i.e., questions 2 and 3 can only be answered very tentatively. Specifically, while the Dirac QSL is the higher energy state (and hence at best metastable), this could reverse with larger L_{y} , as previous studies show that the energy density difference is only of order 0.1% and shrinks as L_{y} increases Hu et al. 2015 ; Zhu and White 2015 . Moreover, even accepting that the low energy state is best characterized as a \mathbb{Z}_{2} QSL, because it appears close to a magnetically ordered state, its character could easily change with larger L_{y} . Model and ground state properties. – We study the S=1/2 , J_{1} - J_{2} Heisenberg model on triangular cylinders: H=J_{1}\sum_{\langle ij\rangle_{1}}\vec{S}_{i}\cdot\vec{S}_{j}+J_{2}\sum_{\langle ij\rangle_{2}}\vec{S}_{i}\cdot\vec{S}_{j}, (1) where \langle ij\rangle_{1/2} denotes the first/second NN pairs of sites, and \vec{S}_{j} are spin-1/2 operators. J_{1}=1.0 is set as the energy unit throughout the paper. We primarily use the so-called XC cylinders, depicted in Fig. 1 (a), where one of the NN bonds is along the length of the cylinder with unit vector \vec{e}_{1}=(1,0) . The circumference of the cylinder is along \vec{e}_{2}=(\frac{1}{2},\frac{\sqrt{3}}{2}) , meaning that sites are connected periodically along this direction. We also consider YC cylinders for comparison and find similar results sm . One of the major findings of previous DMRG studies was the existence, in the intermediate J_{2} region, of two distinct states on even-width cylinders Zhu and White 2015 ; Hu et al. 2015 . These were often referred to as the “odd” and “even” states, since it was conjectured that they represent the ground states in the two topological sectors expected in a \mathbb{Z}_{2} QSL on a cylinder Moessner and Sondhi 2001 ; Sachdev 1992 ; Wang and Vishwanath 2006 ; Thouless 1987 . Since our findings do not support this picture, we will refer to them as the LE and HE states instead. For a given system, the two states are found by starting DMRG with different initial states. For example, the LE state can always be reached starting with the J_{2}=0 ground state, while the HE state is accessed starting from a striped product state. The LE state can also be stabilized by isolating one site on each boundary of the cylinder, which might be thought of as corresponding to localizing a spinon at each end, thus favoring \mathbb{Z}_{2} QSL relative to any confined phase. The energy density difference between the two states is of order 10^{-3} , and shrinks as L_{y} increases Zhu and White 2015 ; Hu et al. 2015 . The HE state is metastable in the sense that further sweeps do not result in its conversion into the LE state. The existence of two topologically distinct ground states with energies that approach each other exponentially is an expected feature of a gapped \mathbb{Z}_{2} QSL. However, the big differences in the locally observable properties of the HE and LE states are inconsistent with this characterization. As shown in Figs. 1 (b) - (e), the spin-spin correlations on the NN bonds with the three possibly orientations are similar in the bulk for both the LE state and the 120 ∘ state, while for the HE state and the striped state, one orientation of the bonds encircling the cylinder is noticeably different from the others, i.e. there is an indication of vestigial Z_{3} nematic order that might reflect proximity to a striped state. Moreover, comparing the equal-time spin structure factors S(\vec{q}) in Fig. 1 (g) and (h), we see that the HE state has extra peaks at two of the M points \pm(\pi,\pi/\sqrt{3}) , which are not seen in the LE state. If we ignore the amplitude and focus on the sign pattern of the longer-ranged correlations, we find a more drastic contrast, suggesting that their low-energy excitations are essentially different, see the End Matter. Additionally, in contrast with previous studies Zhu and White 2015 and the analysis Yao and Kivelson 2012 ; Zhu and White 2015 of the quantum dimer model of a \mathbb{Z}_{2} QSL, we observe the existence of a metastable distinct HE state also on odd-width XC9 (see the End Matter) and XC7 sm cylinders. While our results clearly show that the LE and HE states are different in qualitative, locally observable ways that preclude a topological interpretation, it is notable that the LE state shares many similarities with the 120 ∘ state at J_{2}=0 : they both have pronounced peaks in their structure factors at the K points in the Brillouin zone, they both support edge spinons on even-width cylinders, weak columnar dimer order on odd-width cylinders (albeit significantly weaker in the J_{2}=0 state), as well as a similar L_{y} dependence of the spin triplet gap sm . Both the differences between the LE and HE states, and the similarities between the LE state and the 120 ∘ state at J_{2}=0 are more clearly reflected in the dynamical correlations to which we now turn. Improved Dynamical DMRG method. – Having investigated the ground state properties of the LE and the HE states, we aim to probe their spectral properties through the DSSF S(\vec{q},\omega) , defined as: \begin{split}S(\vec{q},\omega)=&\frac{-1}{\pi N}\sum_{\begin{subarray}{c}a,b=1\\ \alpha=x,y,z\end{subarray}}^{N}e^{i\vec{q}\cdot(\vec{r}_{a}-\vec{r}_{b})}\times\\ &\text{Im}\langle 0|S^{\alpha}_{\vec{r}_{a}}\frac{1}{E_{0}+\omega-H+i\eta}S^{\alpha}_{\vec{r}_{b}}|0\rangle,\end{split} (2) where |0\rangle is the ground state obtained by DMRG with energy E_{0} , and \eta is the chosen broadening factor. Dynamical-DMRG (DDMRG) Jeckelmann 2002 computes this quantity by finding the complex correction vector |c^{\alpha}_{\vec{r}_{b}}(\omega)\rangle=|X^{\alpha}_{\vec{r}_{b}}(\omega)\rangle+i|Y^{\alpha}_{\vec{r}_{b}}(\omega)\rangle that satisfies: (E_{0}+\omega-H+i\eta)|c^{\alpha}_{\vec{r}_{b}}(\omega)\rangle=S^{\alpha}_{\vec{r}_{b}}|0\rangle. (3) In this way, DDMRG computes the DSSF in the frequency domain directly, with each calculation targeting one specific frequency. Typically, the low-frequency spectra are easier to obtain since the corresponding correction vector is closer to the ground state and has lower entanglement. In fact, it has been shown that DDMRG can be more efficient than time-dependent DMRG in obtaining spectra at a specific frequency Ronca et al. 2017 . Unlike DMRG, DDMRG, when implementing H^{2} as two applications of H , is non-variational and solves a projected correction-vector equation, whose accuracy depends on the completeness of the truncated matrix product state (MPS) basis. Specifically, the standard state-averaged implementation Jeckelmann 2002 must accommodate four distinct targets— |0\rangle , S^{\alpha}_{\vec{r}_{b}}|0\rangle , |X^{\alpha}_{\vec{r}_{b}}(\omega)\rangle , and |Y^{\alpha}_{\vec{r}_{b}}(\omega)\rangle —within a single truncated MPS basis; with limited bond dimension m , the resulting truncation can significantly distort the linear problem, leading to slow or spurious convergence. See the End Matter for a more detailed discussion. After our own experimentation with different implementations, we found the following reliable and efficient approach to mitigate this issue, where we (i) decrease the number of simultaneously targeted states in state-averaging and (ii) re-balance their contributions during state averaging. Following Ref. Ronca et al. 2017 , we keep |0\rangle and the source state S^{\alpha}_{\vec{r}_{b}}|0\rangle as a separate (fixed) MPS, and remove |0\rangle from the state-averaging used to build the correction-vector basis. We then construct the state-averaged density matrix from three targets, \{\sqrt{\eta}\,S^{\alpha}_{\vec{r}_{b}}|0\rangle,\ \eta|X^{\alpha}_{\vec{r}_{b}}(\omega)\rangle,\ |Y^{\alpha}_{\vec{r}_{b}}(\omega)\rangle\} , where the \eta -dependent prefactors set their relative weights in the truncation. Empirically, we found this balancing stabilizes convergence and enables correct qualitative features of the DSSF to be captured at a moderate bond dimension, typically within twice of that for the ground state 1 1 1 For ground state DMRG, we typically keep a maximum bond dimension such that the truncation error is ~\sim O(10^{-5}) . , even though the truncation error is still \sim O(10^{-4}) We typically choose a tolerance of \eta and a maximum iteration of 30 in the conjugate gradient optimization at each step, and perform DDMRG sweeps until convergence. More detailed information regarding the original and improved DDMRG method, including its benchmark and convergence, is provided in the SM sm . We implemented the improved DDMRG algorithm using the ITensor library Fishman et al. 2022 and Krylovkit Haegeman 2024 . Figure 2: The low-energy DSSF S(\vec{q},\omega) at a specific \omega shown in the center of each figure, for the 120 ∘ state at J_{2}=0 (top row, (a)(b)(c)), the LE state at J_{2}=0.12 (middle row, (d)(e)(f)), and the HE state at J_{2}=0.12 (bottom row, (g)(h)(i)). Simulations were carried out on XC6-24 (left column, (a)(d)(g)), XC8-24 (middle column, (b)(e)(h)), and XC9-24 (right column, (c)(f)(i)), respectively. The color scale has an upper cutoff of 2. Dynamical spin structure factor. – Using the improved DDMRG algorithm, we computed the DSSF (Eq. 2 ) for various states. A broadening factor \eta=0.1 was used, and we have checked that the qualitative features of the states are insensitive to the choice of \eta sm . Calculations were carried out on XC6-24, XC8-24, and XC9-24 cylinders, keeping a maximum bond dimension of m =1800, 3000, and 5000, respectively, using U(1) symmetry. To better approximate the 2D system with C_{6} rotational symmetry, we choose data from a central L_{y}\times L_{y} region to perform the Fourier transform in Eq. 2 , such that the accessible momenta are approximately C_{6} symmetric. Additionally, we also performed a Fourier transform using data from the central half of the system to obtain the dispersive DSSF along several momentum cuts, as shown in Fig. 6 in the End Matter. In Fig. 2 we show the low-energy DSSF S(\vec{q},\omega) at fixed \omega , for the 120 ∘ state at J_{2}=0 , the LE and HE states at J_{2}=0.12 . In each panel, the value of \omega was taken to be that which maximizes either S(\vec{q}=K,\omega) or S(\vec{q}=M,\omega) , whichever lies lower in energy, see Fig. 3 below. The J_{2}=0 state has clear peaks at the ordering vector \vec{q}=K , corresponding to the 120 ∘ magnetic order, as shown in Fig. 2 (a)(b)(c) for XC6-24, XC8-24, and XC9-24 cylinders, respectively. The LE state at J_{2}=0.12 also has dominant peaks at \vec{q}=K with a similar amplitude to the J_{2}=0 case, but with a somewhat more diffuse structure around it. The intensity at the M point increases only slightly and does not exhibit a clear peak. In contrast, the low-energy DSSF for the HE state at J_{2}=0.12 exhibits conspicuous differences from the previous two states, as shown in Figs. 2 (g)(h)(i). The intensity at the K point is weaker, and there is a clear peak at the M point with comparable intensity, when it is accessible in XC6 and XC8 cylinders. In all cases, the DSSF is much more diffuse along \vec{q}=K\rightarrow M . The features in Fig. 2 can be better highlighted after D_{6} -symmetrization and interpolation, provided in the SM sm . We further compare the DSSF at or near the high symmetry points \vec{q}=K and \vec{q}=M in Fig. 3 . Due to the discrete transverse quantization of momenta, some of the exact high symmetry points are inaccessible on the XC8 or XC9 cylinder 2 2 2 On XC8 cylinder, the \sim K point is chosen at a nearby momentum (1.25\pi,~0.1443\pi) where the intensity reaches its maximum. The \sim M point is taken to be (1.111\pi,~0.385\pi) on XC9 cylinder, which is closest to the original M point along the zone boundary. For the HE state on XC9 cylinder, the intensity at (1.25\pi,~0.257\pi) is chosen, since the peak has slightly shifted away from the original K point (Fig. 2 (i)). . Another consequence of the cylindrical boundary conditions is that the spectra are not C_{6} symmetric, especially as is apparent in Fig. 2 , at the M points. On the other hand, due to translation and inversion symmetries, the intensities at or around the K points are the same. To compare intensities at these two momenta, the M points data in Fig. 3 have been averaged over the symmetry-related points. For J_{2}=0 , the strongest peak appears at \vec{q}=K . The associated gap, defined by the peak position of S(K,\omega) in Fig. 3 , decreases quickly as L_{y} increases, which is consistent with the expected appearance of a Goldstone mode in the 2D limit. In contrast, the low-energy excitation at \vec{q}=M is strongly suppressed. For the LE state at J_{2}=0.12 , the dominant peak remains at \vec{q}=K , although with a larger gap than the J_{2}=0 case. The spectral weight at \vec{q}=M is somewhat larger than for J_{2}=0 and moves to lower frequency with increasing L_{y} , but remains much smaller at low energy than the K point and never exhibits a well-defined peak frequency. For the HE state, the low-energy excitations at the K and M points are comparable in strength, and the gaps at both momenta appear to decrease as L_{y} increases. We have also verified that these characteristics of the HE and LE states behave similarly on YC6-24 cylinders sm . Figure 3: The DSSF S(\vec{q},\omega) at or around high symmetry points \vec{q}=K=(4\pi/3,0) and \vec{q}=M=(\pi,\pi/\sqrt{3}) after symmetrization (see text), for the 120 ∘ state at J_{2}=0 (top row, (a)(b)(c)), the LE state at J_{2}=0.12 (middle row, (d)(e)(f)), and the HE state at J_{2}=0.12 (bottom row, (g)(h)(i)). Simulations were carried out on XC6-24 (left column, (a)(d)(g)), XC8-24 (middle column, (b)(e)(h)), and XC9-24 (right column, (c)(f)(i)), respectively. Having established the main characteristics of the DSSF in different cases, we now compare them with the theoretical expectations for various candidate phases. According to theoretical predictions for the gapped \mathbb{Z}_{2} QSL Wang and Vishwanath 2006 ; Sachdev 1992 , the spinon dispersion should have a minimum at the K point, which leads to a magnon minimum at 2K (equivalent to K ). Within this framework, the system can undergo a continuous phase transition from the 120 ∘ state to the gapped \mathbb{Z}_{2} QSL accompanied by a gap opening at K . Although the LE state identified on finite cylinders is compatible with this scenario, the interpretation is not unique. For finite cylinder circumference, the 120 ∘ state at J_{2}=0 also exhibits a finite gap, as shown in Fig. 3 (a-c) (the gap in XC9-24 is around 0.08). As J_{2} increases and magnetic order is weakened, the gap on a finite cylinder correspondingly grows; this behavior does not preclude the presence of magnetic order in the 2D limit. Indeed, finite-size scaling of the magnetic order extrapolates suggestively to a tiny but non-zero value in the 2D limit sm . This is accompanied by a decrease in the gap as L_{y} widens, although it is unclear if it will close in 2D. Based on our results on cylinders up to width XC9, we cannot decisively determine whether the LE state evolves into a gapped \mathbb{Z}_{2} QSL or instead reflects a weakly magnetically ordered state in the 2D limit. On the other hand, the low-energy effective field theory of a U(1) -Dirac QSL on the triangular lattice is quantum electrodynamics in 2+1 dimensions (QED3) Song et al. 2019 ; Song et al. 2020 ; Wietek et al. 2024 , which predicts gapless spin-triplet excitations at both the M and K points of the Brillouin zone. The HE state observed in our calculations appears to be consistent with this expectation. It displays low-energy spectral weight at both M and K with comparable intensities, and its excitation gap decreases as L_{y} increases. Similar signatures of a Dirac QSL have also been reported in previous numerical studies Ferrari and Becca 2019 ; Sherman et al. 2023 ; Kaneko et al. 2014 ; Drescher et al. 2023 ; Drescher et al. 2025 ; Wietek et al. 2024 . In contrast, the LE state deviates from the Dirac QSL predictions, in that the low-energy triplet spectral weight is strongly suppressed at the M point compared to the K point. This appears to be inconsistent with its being a U(1) Dirac QSL in the 2D limit. Discussion. – While the HE state is most probably a gapless Dirac QSL in 2D, we are currently unable to convincingly determine the precise nature of the LE state, nor are we able to conclude which state will evolve into the equilibrium ground state in the 2D limit. Addressing these questions will require even wider systems, particularly those designed so that the allowed crystal momenta include both the K and M points. It will also require clever finite-size scaling analysis of the magnetic order parameter of the 120 ∘ state and the spin-triplet excitation gap, to distinguish between a weakly magnetically ordered state and a gapped \mathbb{Z}_{2} QSL. Note that the existence of two distinct, nearly degenerate QSL states has also been inferred in studies of the kagome-lattice AFM Changlani et al. 2018 ; Sachdev 1992 . Data availability : The data used to generate the figures are deposited in Zenodo 3 3 3 https://doi.org/10.5281/zenodo.18535976 . Acknowledgements. Acknowledgments. — We acknowledge Leon Balents, Cenke Xu, and Stephen Shenker for helpful discussions. S.J., S.A.K., and H.C.J. are supported by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division, under contract DE-AC02-76SF00515. S.J. and the DDMRG calculations were supported by the Department of Energy, Laboratory Directed Research and Development program at SLAC National Accelerator Laboratory, under contract DE-AC02-76SF00515, with partial support for H.C.J.. SRW is supported by the NSF under DMR-2412638. Calculations were performed on resources of the National Energy Research Scientific Computing Center, supported by the U.S. Department of Energy under contract DEAC02-05CH11231. End Matter Differences between the LE and HE state on the XC9 cylinder. – Here we summarize some further properties of the LE state and the HE state on XC9-24 cylinders, with a focus on the evidence against assigning a topological interpretation to the existence of a metastable HE state. On the basis of an analysis Yao and Kivelson 2012 ; Zhu and White 2015 of the quantum dimer model of a \mathbb{Z}_{2} QSL, one expects that on a cylinder of finite width with L_{y} odd, one should find a state with broken translational symmetry (i.e. columnar dimer order with strength) that decreases exponentially with L_{y} . In fact, as shown in Fig. 4 (a), we do see a columnar order, i.e. pronounced period 2 spatial oscillations in the strength of the spin correlations on bond-types 1 and 2 in the LE state, consistent with the above \mathbb{Z}_{2} QSL description. However, to the extent that there are any such oscillations in the HE state (Fig. 4 (b)), they are substantially weaker - again inconsistent with the two states being only topologically distinct. Figure 4: Correlation functions in the LE (top panels) and HE state (lower panels) in a XC9-24 cylinder, with J_{2}=0.12 . Panels (a) and (d) plot the strength of the spin-spin correlation \langle\vec{S}_{i}\cdot\vec{S}_{i+\delta i}\rangle on three types of NN bonds (defined in Fig. 1 (a) in the main text), averaged over the circumferential direction and plotted versus the position along the cylinder l_{x} . Panels (b) and (e) show the equal-time spin structural factor S(\vec{q}) . (c) and (f) show the the low-energy DSSF S(\vec{q},\omega=0.2) . For both the equal-time structural factor S(\vec{q}) and the low-energy dynamical structural factor S(\vec{q},\omega=0.2) , the LE state is qualitatively similar to the ordered state at J_{2}=0 with a dominate peak only at \vec{q}=K . The HE state has a more diffuse structure factor from K to M , similar to that on even-width cylinders and more consistent with a gapless Dirac QSL. We note that that there is an interesting difference with the situation in the even leg cylinders we have studied in that the HE state appears to be less robustly metastable: it remains metastable up to m=2400 , but evolves towards the LE state at larger m . The DMRG results shown in the figure for the LE states and the DDMRG results for both states are at m=4000 . Structure of the sign of the equal-time spin-spin correlation function — Here we further analyze the equal-time spin-spin correlation function, focusing on its sign pattern and ignoring its amplitude, plotted in Fig. 5 (a) and (b) for the LE and HE states, respectively. While the short-ranged correlations marked by the shaded region are identical for the two states, the longer-ranged correlations are significantly different. The LE state mostly follows a three-sublattice pattern, similar to the 120 ∘ state, while the HE state exhibits a striped pattern. Figure 5: Sign of the spin-spin correlation function \langle\vec{S}_{\vec{r}_{0}}\cdot\vec{S}_{\vec{r}}\rangle/|\langle\vec{S}_{\vec{r}_{0}}\cdot\vec{S}_{\vec{r}}\rangle| for (a) the LE state and (b) the HE state at J_{2}=0.12 on a XC6-24 cylinder at m=2400 , with three columns subtracted on each edge. The reference site \vec{r}_{0} is marked by the red circle. Green dots and yellow dots denote plus and minus signs at site \vec{r} , respectively. Those sites with a correlation amplitude smaller than 5\times 10^{-5} are excluded and marked by a cross, since their signs are susceptible to errors and can not be definitely determined. Outside the shaded vicinity of \vec{r}_{0} , the patterns of the correlation are different for the two states. The Fourier transform of the sign of the spin-spin correlation functions: \tilde{S}(\vec{q})=\sum_{\vec{r}}e^{-i\vec{q}\cdot\vec{r}}\langle\vec{S}_{\vec{r}_{0}}\cdot\vec{S}_{\vec{r}}\rangle/|\langle\vec{S}_{\vec{r}_{0}}\cdot\vec{S}_{\vec{r}}\rangle| , for (c) the LE state and (d) the HE state. The color scale is renormalized and has a lower cutoff of 0.2 to filter out the fuzzy background due to contributions from short-ranged correlations. We also look at the structural factor of the sign of the correlation functions, as shown in Fig. 5 (c) and (d). They also exhibit a significant distinction: the LE state resembles the 120 ∘ state with peaks at \vec{q}=K , while the HE state resembles the striped state with peaks at \vec{q}=M . This contrasts with the ordinary S(\vec{q}) in Fig. 1 (f) and (g), where the difference is less significant, since it is dominated by short-range correlations. Similar distinction in the sign pattern between the two states has also been observed in XC8 and XC9 cylinders, suggesting their different natures. Approximation in DDMRG and related issues. — While ground-state DMRG is variational, DDMRG is usually implemented in a way that is not. When solving the correction-vector equation (Eq. 3 ), one uses real arithmetic to solve the imaginary part |Y\rangle first via [(E_{0}+\omega-H)^{2}+\eta^{2}]|Y\rangle=-\eta S|0\rangle , which can then be used to generate the real part |X\rangle=(H-E_{0}-\omega)|Y\rangle/\eta Jeckelmann 2002 . Since the matrix product operator of H^{2} has a large bond dimension, the standard approach is to apply H twice in the same basis Jeckelmann 2002 , i.e., approximating H^{2} by (H^{\rm proj})^{2} with H^{\rm proj} being the reduced superblock Hamiltonian after projecting on the two-site basis. The accuracy of this approximation relies on the basis representing the intermediate state H^{\rm proj}|Y\rangle accurately, as well as the initial |Y\rangle and final states S|0\rangle Jeckelmann 2002 . In practice, a truncation on the intermediate state always occurs during DDMRG, which makes the calculation non-variational. This means that the iterative linear solver can have slow convergence or “fake” convergence for the projected problem when the true (full-space) residual remains large, leading to inaccurate spectra at limited bond dimension. This issue is exacerbated by the fact that the standard state-averaged implementation in DDMRG must accommodate four distinct targets— |0\rangle , S|0\rangle , |X\rangle , and |Y\rangle —within a single truncated MPS basis; with limited m , the resulting truncation can be substantial and significantly distort the reduced linear problem. Dynamical spin structure factor along momentum cuts. — To supplement the DSSF at specific low-energy (Fig. 2 ) and at specific momenta (Fig. 3 ), we plot the DSSF in XC6-24 cylinders along several momentum cuts. To better see the dispersive features, we calculate the DSSF in Eq. 2 with \vec{r}_{j} being the reference site in the center of the cylinder and \vec{r}_{i} running over the central half of the system. This allows us to access more momentum points along the length of the cylinder (compared with using data from the central L_{y}\times L_{y} region), such as cut 1 and cut 2 in Fig. 6 (j). After that, we perform a D_{6} symmetrization of the DSSF and interpolate the data between the accessible momenta. Note that the DSSF obtained this way will have a minor quantitative difference from running both \vec{r}_{j} and \vec{r}_{i} over the L_{y}\times L_{y} region as in the previous figures, but the qualitative features remain the same. Figure 6: On an XC6-24 cylinder, the DSSF S(\vec{q},\omega) along momentum cut 1 (top row, (a)(b)(c)), momentum cut 2 (middle row, (d)(e)(f)), and momentum cut 3 J_{2}=0.12 (bottom row, (g)(h)(i)); for the J_{2}=0 state (left column, (a)(d)(g)), the LE state at J_{2}=0.12 (middle column, (b)(e)(h)), and the HE state at J_{2}=0.12 (right column, (c)(f)(i)), respectively. (j) Illustrates the three momentum cuts. The color scale has an upper cutoff of 1.3. (k) Along cut 3, the low-energy S(\vec{q},\omega=0.2) for the three states. For both the J_{2}=0 and the LE state at J_{2}=0.12 , one can see a clear suppression of low-energy excitations at the M point. The LE state has a somewhat more diffuse structure at higher energies, but overall it appears qualitatively similar to the J_{2}=0 state. On the other hand, for the HE state at J_{2}=0.12 , low-energy modes emerge at M with intensities comparable to those of K . In Fig. 6 (k), we plot the intensity of DSSF along the momentum cut 3 at an energy cut \omega=0.2 , where one can see a clear peak at M that is absent in the other two states, signifying their different natures.