Topology plays an increasingly important role in understanding different phases and phase transitions in correlated quantum matters and materials. One prominent example is the Berezinskii–Kosterlitz–Thouless (BKT) mechanism in two-dimensional (2D) systems1,2,3,4,5, which is associated with the binding and unbinding of topological defects. The BKT transition cannot be characterized by conventional order parameters and constitutes the earliest example of phase transition beyond the Landau–Ginzburg paradigm of spontaneous symmetry breaking. Historically, the BKT mechanism was introduced in the XY spin model and long predicted to occur in magnetic thin films1,4. In experiments, signatures of the BKT transition have been observed in superfluid helium films6, as well as in 2D superconducting films7,8 and arrays9. Regarding the original proposal in layered XY-type magnets, despite intensive efforts10,11,12,13,14,15, direct and unambiguous observation of the BKT transition is still lacking. One major obstacle is the three-dimensional (3D) couplings in the magnets, although weak, will inevitably enhance the confining potential of vortices15, leading to 3D ordering that masks the BKT transition. Therefore, it is of fundamental interest to find and identify BKT materials that could overcome the obstacle and study the topology-related low-energy dynamics.

Recently, a layered frustrated rare-earth antiferromagnet TmMgGaO416,17,18 was reported to ideally realize the triangular-lattice quantum Ising (TLI) model19. The relatively large interlayer distance of 8.3774 Å along the c axis gives rise to excellent two dimensionality17 and no sign of conventional 3D phase transition was observed in either specific heat or magnetic susceptibility measurements. Nevertheless, it was reported from neutron scattering that TmMgGaO4 ordered below  ~1 K into an antiferromagnetic phase with a sixfold symmetry breaking16,18, which closely resembles the ground state of the TLI model originated from an order-by-disorder mechanism20,21. At higher temperatures, the effective XY degrees of freedom emerge and the BKT mechanism is expected to come into play21.

In TmMgGaO4, each Tm3+, with a 4f12 electron configuration and a spin–orbit moment J = 6, forms a non-Kramers doublet due to the crystal-electric-field splitting. The doublet is well separated from the rest levels by about 400 K16 and can thus be regarded as an effective spin-1/2. There is further a fine energy splitting within the doublet, induced by the local trigonal crystal field17, acting as an intrinsic “transverse field” applied on the effective spin. From the magnetization measurements16,17,18, one observes that Tm3+ ions contribute highly anisotropic Ising-type moments with Jz = ±6 along the c axis, resulting in an effective out-of-plane g-factor ~ 13.216,19. On the contrary, the effective in-plane g-factor and dipolar moment in the ab plane are negligible.

Facilitated by this feature in TmMgGaO4, in this work we employed nuclear magnetic resonance (NMR), a sensitive low-energy probe, to detect the BKT phase. We applied a moderate in-plane field of 3 T, which is adequate to collect the 69Ga NMR signals and, at the same time, hardly disturbs the low-energy electronic states of the material. This is important, as in the TLI model that is believed to accurately model TmMgGaO419, the BKT phase can be fragile against out-of-plane fields22,23,24, thus posing a challenge to NMR measurements. Taking advantage of the fact that in-plane moment in TmMgGaO4 is mostly multipolar16,19, our NMR experiments with in-plane fields, which merely couple to the nuclear spins, can clearly identify the BKT phase in the material.

As shown in Fig. 1, from our NMR measurements of the spin-lattice relaxation rate 1/T1, we identify TU 1.9 K and TL 0.9 K, which represent the upper- and lower-BKT transitions, where a critical BKT phase resides at zero magnetic field in between the high-T paramagnetic and low-T antiferromagnetic phases. This finding is further substantiated by our scaling analysis of the measured susceptibility data near TL, as well as the simulated NMR and susceptibility data using large-scale quantum Monte Carlo (QMC) calculations.

Fig. 1: Phase diagram of TmMgGaO4 under out-of-plane magnetic fields.
figure 1

Under zero field, there are paramagnetic (PM), BKT, and antiferromagnetic (AFM) phases. The TU (TL) is the upper (lower) BKT transition temperature, determined from the plateau structure in the NMR spin-lattice relaxation rate 1/69T1 (see Fig. 2c for details). The BKT phase between TU and TL is illustrated by the solid vertical line, while the AFM regime is indicated by the arrow. The contour background depicts the magnetic specific heat Cm/T at various fields and temperatures, with data adapted from ref. 16 and plotted in logarithmic scale. T* corresponds to the maximum of Cm/T at each field, signifying the position of strong magnetic fluctuations. \(T^{\prime}\)(T″) denotes the temperature at a specific field where a peak is found in the differential susceptibility dM/dH, shown in Fig. 3b. The Curie–Weiss temperature θCW is obtained from the 1/69T1T (see Supplementary Fig. 1). Remarkably, T*, \(T^{\prime}\), T″, and θCW all collapse to the same phase boundary between the BKT-like regime and AFM phase. A magnetically ordered phase is supposed to lie below the dome-like boundary. Errors represent 1 SD throughout the paper.


NMR probe of the BKT phase

The obtained NMR spectra with an in-plane magnetic field μ0H = 3 T are shown in Fig. 2a at representative temperatures. To better resolve the magnetic transition, the hyperfine shifts 69Kn of the spectra were analyzed and plotted in Fig. 2b as a function of temperature. Upon cooling, 69Kn peaks at about 0.8–0.95 K and then starts to drop at lower temperatures. Therefore, the ordering temperature is determined to be TL 0.9 K, consistent with neutron scattering experiments16,18. In addition, both the second moments (width of the NMR spectra) and the third moments (asymmetry of the spectra) of the spectra change dramatically below  ~ 2 K, suggesting the onset of local hyperfine fields enhanced by the static or quasi-static magnetic ordering (Supplementary Fig. 4). These two characteristic temperatures signal the two-step melting of magnetic order through two BKT transitions, suggesting an intermediate floating BKT phase in the system. We suspect that there is some inhomogeneity of the local hyperfine fields, which is very likely caused by the quenched disorder from Mg/Ga site mixing16, although no significant influence on the electronic and more importantly the magnetic properties is seen (see more detailed discussions in Supplementary Note 4).

Fig. 2: NMR spectra and spin-lattice relaxation rates of TmMgGaO4.
figure 2

a 69Ga NMR spectra at different temperatures under a 3 T in-plane field, with zero frequency corresponding to γH = 30.692 MHz. Data are shifted vertically for clarity. At high temperatures, the spectra are roughly symmetric, whereas for T ≤ 2 K, a shoulder-like structure can be resolved on the left side of the main peak. b NMR hyperfine shift \({}^{69}{K}_{n}=(\bar{f}/\gamma H-1)\times 100 \%\) as a function of temperature, where \(\bar{f}\) is the average frequency of each spectrum. c NMR spin-lattice relaxation rate 1/69T1 vs. temperature measured under in-plane fields of 3 T and 1 T. A plateau-like feature, characterizing strong magnetic fluctuations, is observed between TL 0.9 K (lower-BKT transition) and TU 1.9 K (upper BKT transition). d 1/T1 data computed from the dynamical spin–spin correlation function with contributions from all momentum points [c.f., Eq. (3) and see the hyperfine form factor in Supplementary Note 2] in the Brillouin zone (left scale) and from K′ point in the vicinity of the K point (right scale), through large-scale QMC simulations (see “Methods”).

The spin-lattice relaxation rate 1/T1 provides a highly sensitive detection of low-energy spin fluctuations25,26,27,28,29, and thus the BKT transition. In Fig. 2c, we show the 1/69T1 obtained under in-plane fields μ0H = 3 T and 1 T, which reflects intrinsic spin fluctuations with zero out-of-plane field. At 3 T, 1/69T1 first decreases upon cooling from 10 K then suddenly increases below TU ≈ 1.9 K, indicating the onset of strong low-energy spin fluctuations. The data at 1 T show similar behaviors. Below TL 0.9 K, 1/69T1 drops sharply, consistent with the onset of the magnetic ordering as also inferred from the hyperfine shift. Here, 1/69T1 is dominated by the gapped long wavelength excitations in the ordered state. At the magnetic phase transition, a peaked feature in 1/T1 develops, caused by the gapless low-energy spin fluctuations with diverging correlation length. Remarkably, at temperatures between TU 1.9 K and TL 0.9 K, 1/69T1 exhibits a plateau-like structure, indicating the emergence of a highly fluctuating phase with diverging spin correlations yet no true long-range order, which is the hallmark of a BKT phase1,2,3,4,5. Therefore, it is for the first time that such a phase is unambiguously observed in a magnetic crystalline material.

Our unbiased QMC simulations on the TLI model of the material (see “Methods”), with accurate coupling parameters determined in ref. 19, quantitatively justifies the existence of the BKT phase between TL and TU. We computed 1/T1 and compared with the experiment below. Figure 2d shows the calculated 1/T1 by including fluctuations from all momentum points in the Brillouin zone (cf. Supplementary Fig. 2) and compare to that from K′ (around the K point at the corner of Brillouin zone). The former shows a decrease upon cooling below 4 K and then an upturn above \({T}_{{\rm{U}}}^{* }\simeq 2\) K, followed by a rapid decrease below \({T}_{{\rm{L}}}^{* }\simeq 1\) K. These behaviors are in excellent agreement with the measured 1/69T1. The latter reflects gapless excitations of the XY degrees of freedom emergent in the BKT phase, where the calculated 1/T1 from K′ exhibits an anomalous increase down to \({T}_{{\rm{U}}}^{* }\), below which the increment slows down. The contribution to 1/T1 near the K point reaches a maximum at \({T}_{{\rm{L}}}^{* }\) and drops rapidly below it. The absence of critical spin fluctuations at momentum away from the K point suggests that the measured 1/69T1 below 2 K is mainly contributed by excitations around the K point (see Supplementary Note 3).

Overall, the quasi-plateau behaviors in the QMC results and the two characteristic temperature scales are in full consistency with the NMR measurements. This constitutes both strong support for the accurate quantum many-body modeling of the material TmMgGaO4 and also solid proof of the BKT phase therein detected by NMR. Nevertheless, we note that there are still subtle differences between the experimental and numerical data. Needless to say, the real material is always more complicated than our theoretical minimal model. For example, influences from higher crystal-electric-field levels above the non-Kramers doublet, the interlayer couplings not included in our model calculations, and the lack of knowledge on the precise local hyperfine coupling constant, etc., may explain the difference remaining between Fig. 2c, d.

Universal magnetic susceptibility scaling

Magnetic susceptibility χ measurements were also performed to strengthen the finding of the BKT phase. In Fig. 3a, we show the overall temperature dependence of χ at different out-of-plane fields. For T 2 K, χ increases monotonically upon cooling and barely changes with fields. However, for T 2 K, approximately the upper BKT transition TU as obtained from the 1/69T1 measurements, χ increases as the field decreases, suggesting the onset of peculiar magnetic correlations. With further cooling, the susceptibility gets flattened with temperature. The magnetization M(H) was further measured at selected temperatures (data shown in Supplementary Fig. 6), and for the sake of clarity, the differential susceptibility dM/dH is plotted in Fig. 3b. At around 2.5 T, a pronounced peak can be observed at low temperature, indicating the existence of a quantum phase transition, and the phase at lower fields should be a magnetically ordered phase, although its precise nature remains to be uncovered. Besides the high-field feature, for temperatures at 0.8 K and above, a kinked feature is clearly resolved on each dM/dH curve at low fields, whereas at 0.4 K, the low-field kink disappears, which posts a question of whether there is a quantum transition or merely a crossover from the zero-field AFM phase to the finite-field ordered phase under the dome in Fig. 1. The temperature and field values indicated by the down arrows in Fig. 3 are denoted as \(T^{\prime}\) and T″ in the phase diagram (Fig. 1).

Fig. 3: Uniform magnetic susceptibility of TmMgGaO4 and scaling analysis.
figure 3

a dc susceptibility χ as functions of temperatures under small out-of-plane (H//c) and in-plane (H//ab) fields. The latter is multiplied by a factor of 20 for visualizing purpose. The deviation of data below 2 K indicates the entry to the BKT phase and the field-suppression of magnetic correlations. b The differential susceptibility dM/dH under out-of-plane fields at different temperatures. The kinks at low fields, as denoted by the down arrows, suggest the transition from the BKT-like phase to the ordered phase (under the dome in Fig. 1) with increasing fields. The peaked features at high fields suggests a quantum phase transition to the polarized phase. c Fits of dM/dH to the power-law scaling function dM/dH ~ Hα with α = 2/3 for the 0.4 and 0.8 K data, and α = 0.123 for 2.1 K data in the log-log scale. The 3 K data follow the α = 0 line in the paramagnetic phase. dM/dH by the QMC calculations in the same field and temperature range as in c, and fits to the power-law function with exponents α, which give consistent results as experiments.

Field-theoretical analysis of the TLI model23,24 has predicted that upon applying a small out-of-plane field, the differential susceptibility dM/dH shall exhibit a divergent power-law behavior as dM/dH ~ Hα in proximity to the BKT phase. At TL, α has the value of 2/3, which corresponds to a critical exponent η = 1/9 at the lower-BKT transition and is originated from the sixfold symmetry breaking23. The exponent α gradually decreases as temperature increases, and above an intermediate temperature between TL and TU, α = 0 due to non-universal contributions. This is exactly what we observe in Fig. 3c. We fit the dM/dH with the power-law function at different temperatures, with the fitting regime chosen between 0.6–0.9 T. At 0.8 K, α is very close to the expected value of 2/3 (and thus η = 1/9) at TL, which constitutes a remarkable fingerprint evidence for the BKT transition. At lower temperatures such as 0.4 K, the exponent is also close to 2/3, because the susceptibility saturates with temperature, as shown in Fig. 3a. At high temperatures, α drops rapidly to a small value 0.12 at 2.1 K and becomes effectively zero at 3.0 K.

Therefore, the susceptibility scaling gives the lower-BKT transition at about 0.8 K and upper transition probably between 2.1 and 3 K, in good agreement with the TL and TU estimated from NMR. These results are also fully consistent with our QMC calculations on the susceptibility shown in Fig. 3d. At TL or lower, α is 2/3, then decreases to a very small exponent 0.086 at 2.67 K, and above 3 K, becomes zero within numerical uncertainty. Such a power-law behavior in dM/dH again signifies the finite-temperature window of the BKT phase with diverging magnetic correlations, which gives rise to the universal power-law scaling of magnetic susceptibility.


We believe the findings in this work are of various fundamental values. Since the original proposal of a BKT phase in magnetic films3,4,5, which also triggered the currently thriving research field of topology in quantum materials, tremendous efforts have been devoted to finding the BKT phase in magnetic crystalline materials, yet hindered by the obstacle outlined in the Introduction. Here, benefiting from NMR as a sensitive low-energy probe, and the nearly zero planar gyromagnetic factor in a TLI magnet TmMgGaO4, we are able to reveal two BKT transitions and a critical BKT phase with an emergent XY symmetry. Together with the power-law behavior of the susceptibility and excellent agreement between our QMC simulation and experiment data, we unambiguously identify the long-sought BKT phase in a magnetic crystalline material.

Many intriguing questions are stimulated, based on the phase diagram in Fig. 1 obtained here. First, what is the nature of the ordered phase under finite fields, are there further exotic phases and transitions in the phase diagram, and will there be a field-induced quantum phase transition at the high-field side of the dome—these are all of great interests to be addressed in future studies. Second, it should be noted that the dynamical properties obtained by QMC calculations in Fig. 2 are computed on a large, while finite-size, 36 × 36 lattice, which already produces 1/T1 data in very good agreement with the experimental measurements. Such a great agreement is surprising, given the possible existence of randomness from Ma/Ga site mixing in the material TmMgGaO416, and also the lattice disorder revealed by the large high-temperature second moment of the NMR spectra (Supplementary Note 4). Although the random distribution in intrinsic transverse fields and spin couplings does not seem to alter the low-temperature spin-ordered phase and the sharp spin excitation line shapes18,19, its intriguing effects on the finite-temperature phase diagram of TLI and also the compound TmMgGaO4 call for further studies.

Third, in the study of BKT transition in superfluid systems, it has been observed experimentally and understood theoretically that additional dissipations also appear above the transition temperature due to fluctuations of vortices6. Hence, the plateau of 1/T1 we observe may also cover regions slightly above the upper BKT transition temperature. We leave this subtlety to future numerical and experimental efforts. Lastly, in general terms, whether there are other rare-earth magnetic materials in the same family of TmMgGaO4 that, acquire similar 2D competing magnetic interactions from highly anisotropic gyromagnetic factor and unique triangular-lattice structures, and also exhibit the BKT physics, is quite intriguing and calls for future investigations. All these directions are ready to be explored from here.


Crystal growth and susceptibility measurements

Single crystals were grown by the optical-floating-zone method with an image furnace (IR Image Furnace G2, Quantum Design). The natural cleavage surface of the crystals is the ab plane, which allows us to align the field orientation within an error of 2. The dc susceptibility was measured in a PPMS VSM (Quantum Design) for temperatures above 2 K and in a He-3 MPMS (Quantum Design) for temperatures ranging from 0.4 to 2 K.

NMR measurements

The 69Ga (I = 3/2, γ = 10.219 MHz/T) NMR spectra were collected with the standard spin-echo sequence, with frequency sweep by a 50 kHz step using a topping tuning circuit. The NMR hyperfine shift was obtained by calculating the change of the first moment of the spectra. The spin-lattice relaxation rate 1/69T1 was measured by the inversion-recovery method, where a π/2 pulse was used as the inversion pulse. The NMR data from 1.8 K and above were measured in a variable temperature insert, and the data from 1.8 K and below were measured in a dilution refrigerator. The weak NMR signal at low fields and the rapid decrease of 69T2 upon cooling (Supplementary Note 5) prevented us to measure the 1/69T1 for in-plane fields <3 T, with temperature below 1.8 K. Whereas for in-plane fields of 4 T and higher, the sample could not be held in position because of the large anisotropy in the g-factor and unavoidable sample misalignment (2). At T = 1.2 K, we did not find any change of 1/69T1 with two different radio frequency excitation levels (14 mT and 24 mT), and with different frequencies across the NMR line, within the error bar.

Triangular-lattice Ising model

At zero field, the intralayer couplings in TmMgGaO4 can be described by the TLI Hamiltonian,

$$H={J}_{1}\sum _{\langle i,j\rangle }{S}_{i}^{z}{S}_{j}^{z}+{J}_{2}\sum _{\langle \langle i,j\rangle \rangle }{S}_{i}^{z}{S}_{j}^{z}+\sum _{i}\Delta {S}_{i}^{x},$$

where J1 and J2 are the superexchange interactions among Tm3+, 〈ij〉 and 〈〈ij〉〉 refer to the nearest neighbors and the next-nearest neighbors, respectively, and Δ is the energy splitting within the non-Kramers doublet imposed by the crystal field. We have shown that the parameter set J1 =  0.99 meV, J2/J1 ≈ 0.05 and Δ/J1 ≈ 0.54 reproduces the experimental results of the transition temperatures and the inelastic neutron scattering spectra19.

In the TLI model [Eq. (1)], we can define a complex field ψ as a combination of the Ising (Z) components \({m}_{{\rm{A}},{\rm{B}},{\rm{C}}}^{z}\) on three sublattices, i.e.,

$$\psi ={m}_{{\rm{A}}}^{z}+{{\rm{e}}}^{i2\pi /3}\ {m}_{{\rm{B}}}^{z}+{{\rm{e}}}^{i4\pi /3}\ {m}_{{\rm{C}}}^{z}.$$

Notably, ψ = ψeiθ is a complex order parameter that represents the emergent XY degree of freedom relevant to the BKT physics in the TLI model.

QMC calculations

QMC simulations were performed in the path integral in the \({S}_{i,\tau }^{z}\) basis with discretization in space and time. The lattice of L × L × Lτ, where L = 36 and Lτ = βτ with Δτ = 0.05 and β = 1/T, were simulated with both local and Wolff-cluster update schemes30,31. The 1/T1 results were obtained by first computing the dynamical spin–spin correlation function \(\langle {S}_{i}^{z}(\tau ){S}_{j}^{z}(0)\rangle\) and then acquiring its real-frequency dependence S(qω) from the stochastic analytic continuation32. We then determined the 1/T1 either by summing the contributions close to momentum K or over the entire Brillouin zone, as discussed in the Fig. 2d of the main text,

$${T}_{1}^{-1}({\bf{q}})=\frac{1}{{L}^{2}}\sum _{{\bf{q}}}| {A}_{{\rm{hf}}}({\bf{q}}){| }^{2}S({\bf{q}},\omega \to 0),$$

where Ahf(q) is the hyperfine coupling form factor (see Supplementary Note 2). Similar analyses have been successfully applied to the QMC computation of NMR 1/T1 for the spin-1/2 and spin-1 chains33,34.