Unconventional superconductivity from lattice quantum disorder Preprint: APS/123-QED Yu-Cheng Zhu, Jia-Xi Zeng, Xin-Zheng Li Abstract Unconventional superconductivity presents a defining and enduring challenge in condensed matter physics. Prevailing theoretical frameworks have predominantly emphasized electronic degrees of freedom, largely neglecting the rich physics inherent in the lattice. Although conventional phonon theory offers an elegant description of structural phase diagrams and lattice dynamics, its omission of nuclear quantum many-body effects results in misleading phase diagram interpretations and, consequently, an unsound foundation for superconducting theory. Here, by incorporating nuclear quantum many-body effects within first-principles calculations, we discover a lattice quantum disordered phase in superconductors H 3 ​ S \textup{H}_{3}\textup{S} and La 3 ​ Ni 2 ​ O 7 \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} . This phase occupies a triangular region in the pressure–temperature phase diagram, whose left boundary aligns precisely with T c T_{c} of the left flank of the superconducting dome. The T c max T_{c}^{\textup{max}} of this quantum disordered phase coincides with the maximum of superconducting T c T_{c} , indicating this phase as both the origin of superconductivity on the dome’s left flank and a key ingredient of its pairing mechanism. Our findings advance the understanding of high-temperature superconductivity and establish the lattice quantum disordered phase as a unifying framework, both for predicting new superconductors and for elucidating phenomena in a broader context of condensed matter physics. The left flank of the superconducting dome—spanning the underdoped or low-pressure regime up to the peak T_{c} —is pivotal for deciphering the mechanism of high-temperature superconductivity 11 ; 41 ; 34 . This region is characterized by an intimate entanglement of magnetic, charge, and structural orders 42 ; 37 . While such complexity has inspired influential theoretical paradigms, including those based on magnetic fluctuations and strong electron correlations, it has also led to a lack of consensus, obscuring a unified physical picture 31 ; 36 . The growing family of unconventional superconductors 38 and advances in first-principles methods 32 now highlight structural phase transitions as a clarifying lens, given their universal link to competing orders and their essential role in ab initio modeling. Nevertheless, prevailing theoretical approaches remain predominantly electron-centric, largely overlooking the rich and decisive physics of the lattice. Recent studies of high-pressure superconductors have opened new avenues for understanding high-temperature superconductivity, bringing structural phase transitions into sharp focus 40 ; 17 ; 48 ; 45 ; 25 ; 47 ; 12 ; 14 ; 13 . In such systems, nuclear quantum effects are known to reshape superconducting phase diagrams by altering structural boundaries 16 ; 15 . A prominent example is \textup{H}_{3}\textup{S} , where accounting for these effects dramatically lowers the pressure for hydrogen-bond symmetrization relative to classical predictions, shifting the entire superconducting dome into the \textup{Im}\bar{3}\textup{m} phase and challenging the prior two-phase interpretation 16 ; 46 . While such studies incorporated essential quantum corrections, they did not fully address the many-body nature of the interacting nuclei. In our recent work, by rigorously including nuclear quantum many-body effects within a first-principles framework, we have achieved a more accurate determination of the structural phase boundaries and, significantly, revealed a lattice quantum disordered (LQD) phase 50 . In this regime, quantum fluctuations stabilize a higher-symmetry disordered state, giving rise to lattice dynamics beyond the conventional phonon picture. This finding not only redefines the structural phase diagram but also directly motivates the present investigation into the link between such quantum lattice disorder and the unresolved physics on the left flank of the T_{c} dome. The recent discovery of nickel-based superconductors, accompanied by enhanced structural characterization under pressure, provides a timely platform for validating this picture 40 ; 17 ; 44 ; 48 ; 45 ; 25 . Figure 1: Lattice quantum disordered phase in \textup{H}_{3}\textup{S} and \textup{D}_{3}\textup{S} from first-principles calculation. a , A schematic diagram of a 1-D double-well chain, which describes the LQD phase. The nucleus on each lattice site (the ball) lies on a double-well potential (red curve) connected to the neighboring sites by spring interactions (zigzag lines). The many-body nuclear quantum states are determined by the competition between the on-site tunneling effects and the inter-site interactions. In Ref. 50 , we have a detailed description of the LQD phase described by this model and the PIMD method. b , The dispersion relation of lattice dynamics of \textup{Im}\bar{3}\textup{m} \textup{H}_{3}\textup{S} ( T=200 K, P=141 GPa) by PIMD (solid lines) in comparison with the harmonic phonon spectra (dashed lines). The structural instability indicated by the soft phonon mode is suppressed. c , PIMD frequency at \Gamma point for the soft mode defined in ( b ), as a function of temperature and pressure. The open black and red circles correspond to \textup{H}_{3}\textup{S} and \textup{D}_{3}\textup{S} , respectively. The structural phase transition is defined by the point at which this frequency changes sign. d , The triangular region of the LQD phase on the P-T phase diagram is bounded by the quantum (PIMD) and classical (MD) phase boundaries. The solid symbols represent the experimental superconducting T_{c,\textup{SC}} from Refs. 12 ; 15 ; 29 . Notably, the left boundary of the LQD phase (the PIMD line) aligns with the left flank of the superconducting dome, and T_{c,\textup{LQD}}^{\textup{max}} coincides with T_{c,\textup{SC}}^{\textup{max}} for both \textup{H}_{3}\textup{S} and \textup{D}_{3}\textup{S} . Here, we find that the superconducting transition on the left flank of the dome stems from the structural transition of a low-symmetry phase into a LQD phase. We employ path-integral molecular dynamics (PIMD) 7 ; 28 to treat the nuclear quantum many-body effects from first principles. The structural phase boundary is determined by the free-energy surface, which is constructed by the centroid potential of mean force from PIMD. The boundary obtained via this approach coincides precisely with the left flank of the superconducting dome. Its intersection with the classical boundary from conventional molecular dynamics delineates the region of the LQD phase. The extent of this LQD phase increases linearly with decreasing temperature, with its origin, which we label as T_{c,\textup{LQD}}^{\textup{max}} , locating at \sim 220 K for \textup{H}_{3}\textup{S} , \sim 160 K for \textup{D}_{3}\textup{S} and \sim 77 K for \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} . These values align exactly with their respective highest observed superconducting transition temperatures T_{c,\textup{SC}}^{\textup{max}} . This indicates not only that superconductivity occurs entirely within the high-symmetry phase but also establishes a direct mechanistic link between the LQD phase and unconventional superconductivity. We expect this picture to be broadly applicable to other unconventional superconductors. LQD phase by PIMD Thermal and quantum fluctuations are key factors that can suppress a structural phase transition. A soft phonon mode in the potential energy surface (PES) does not necessarily imply a structural instability in the free energy surface (FES) once these fluctuations are included. Thus, the true phase transition point must be identified by examining the curvature of the FES that fully accounts for both thermal and quantum effects. This offers a fundamental and broadly applicable criterion for structural instability. While thermal fluctuations are well described by conventional molecular dynamics (MD), the key challenge lies in correctly including nuclear quantum effects 27 when calculating the FES. At low temperatures, when the potential barrier of a soft mode is modest, nuclear tunneling induces strong quantum fluctuations (sketched by the green wavefunctions of double-well potential in Fig. 1 a) that favor a quantum disordered state 50 . In competition, there is a global tendency for all lattice sites to slide toward the same side of the potential well (as the blue balls) to lower the energy, thereby establishing long-range structural order. The resulting transition from a low-symmetry to a high-symmetry phase is therefore a quantum order–disorder transition 35 , governed by the balance between quantum fluctuations and long-range ordering (effect of inter-site interaction as the zigzag springs) 50 . This physics necessitates a treatment that captures the many-body nature of the quantum nuclei. To this end, we employ path-integral molecular dynamics (PIMD), a powerful framework for including such nuclear quantum many-body effects and the thermal effects 7 ; 28 ; 27 ; 5 . Within PIMD, the centroid potential of mean force provides a rigorous route to construct the relevant FES 4 ; 19 . Leveraging lattice symmetry, the emergence of a soft mode—and hence the phase boundary—can be determined efficiently from the centroid effective forces of only a few finite-displaced structures. The region between this quantum phase boundary and the classical boundary from MD defines the LQD phase. In the following, we apply this approach to map the LQD phase in two representative systems: \textup{H}_{3}\textup{S} and \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} . To make the PIMD sampling feasible, we employ machine learning inter-atomic potentials at the density-functional theory (DFT) level 3 ; 22 as the PES, with the underlying DFT calculations performed using the PBE functional 33 . The computational details are provided in the Methods section. \textup{H}_{3}\textup{S} The lattice dispersion obtained via the centroid effective forces (Fig. 1 b) show that imaginary harmonic frequencies—indicating a saddle point in the PES—are suppressed in the PIMD spectrum when quantum fluctuations are strong. Tracking the soft-mode frequency at the \Gamma point across pressure and temperature (Fig. 1 c) allows us to locate the phase boundary where this frequency changes sign. The resulting boundary shifts to higher pressure with increasing temperature (Fig. 1 d), in contrast to the classical trend and to other quantum correction methods 30 (see Extended Data Fig. 1 ). This arises from two competing influences: thermal and quantum effects both suppress structural order, but thermal agitation also disrupts quantum effects, rapidly diminishing the LQD phase. Consequently, the structural phase boundary in the T-P plane forms a dome: it first rises with temperature before eventually crossing over to the classical regime. The classical boundary, identified from the pair distribution function via MD (Extended Data Fig. 2 ), together with the PIMD quantum boundary, delineates a triangular LQD region. Their intersection defines the T_{c,\textup{LQD}}^{\textup{max}} of the LQD phase—the temperature at which quantum effects are completely overwhelmed by thermal fluctuations. Due to the greater mass of deuterium compared to hydrogen, its quantum effects are weaker. As a result, the quantum phase boundary of \textup{D}_{3}\textup{S} shifts to higher pressures, and its T_{c,\textup{LQD}}^{\textup{max}} is about 160 K, which is lower than the 220 K for \textup{H}_{3}\textup{S} as in Fig. 1 d. In \textup{H}_{3}\textup{S} , the superconducting transition temperature T_{c} exhibits a characteristic dome in the P-T plane 15 . A distinct, continuous kink at the dome’s maximum—evident in the experimental data (the gray filled dots in Fig. 1 d) and also observed in nickel-based superconductors 40 —provides a natural marker that defines the lower-pressure region (below 150 GPa) as the dome’s left flank. The key finding is that the left boundary of the LQD phase aligns with the left flank. Moreover, the isotope effect in superconductivity is accurately captured by the LQD phase. Although the quantitative agreement with experiment is limited by DFT precision (results with other functionals are in the Extended Data Fig. 1 ), the correspondence between the phase boundaries strongly indicates that the superconducting transition on the left flank originates from the LQD phase—and, consequently, that superconductivity occurs entirely within the high-symmetry \textup{Im}\bar{3}\textup{m} phase, thereby invalidating earlier two-phase interpretations and the calculations based upon them 1 . A more important point is that the T_{c,\textup{LQD}}^{\textup{max}} coincides precisely with the peak superconducting temperature T_{c,\textup{SC}}^{\textup{max}} for both \textup{H}_{3}\textup{S} and \textup{D}_{3}\textup{S} . This agreement, robust across different exchange-correlation functionals (Extended Data Fig. 1 ), establishes a direct link between LQD and the superconducting mechanism. Having established this framework for \textup{H}_{3}\textup{S} , we now examine \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} to test its generality, before finally presenting a unified picture in the subsequent discussion. \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} Figure 2: Lattice quantum disordered phase in \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} from first-principles calculation. The solid symbols represent the experimental superconducting T_{c,\textup{SC}} from Refs. 40 ; 17 . On the left flank of the superconducting dome (below 14 GPa), T_{c,\textup{SC}} is marked by a sudden and rapid decrease. The quantum structural phase boundary by PIMD aligns with the left flank, and T_{c,\textup{LQD}}^{\textup{max}}\approx 77 K is in consistence with the experimental T_{c,\textup{SC}}^{\textup{max}}=80 K. Nickel-based superconductors offer a compelling platform to clarify the link between the LQD phase and unconventional superconductivity. In these systems, first-principles calculations have played a central role 40 , and the assumed crystal structure—as the starting point of such calculations—directly influences theoretical interpretations of the superconducting mechanism. It is therefore critical to correctly determine the structural phase diagram 48 ; 45 ; 25 . Current studies suggest a transition from I4/mmm to a lower-symmetry phase around a few tens of GPa, and—reminiscent of the earlier, erroneous two‑phase picture for \textup{H}_{3}\textup{S} —assign the left flank of the superconducting dome to that low‑symmetry phase 44 . Here, we resolve the structural transitions in nickel‑based superconductors using \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} as a representative example, and reveal the relationship between the LQD phase and the superconducting dome. We note that, unlike the earlier studies 40 ; 18 , we do not employ the PBE+U approach because it fails to reproduce the stable structures and yields phase‑transition pressures far from experimental values (see Extended Data Fig. 3 ). For generality and computational feasibility, we use the PBE functional throughout. As shown in Fig. 2 , the left boundary of the LQD phase shifts to higher pressure with increasing temperature, matching the steep left flank of the superconducting dome. Within the precision of DFT, this indicates that—as in \textup{H}_{3}\textup{S} —the left boundary of the LQD phase coincides with the left flank of the dome, and that superconducting transition on this flank originates from the quantum order‑disorder transition into the LQD phase. More importantly, the T_{c,\textup{LQD}}^{\textup{max}} aligns with the T_{c,\textup{SC}}^{\textup{max}} again. The intersection of the PIMD and MD boundaries gives a maximum T_{c} of about 77 K for the LQD phase of \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} at the PBE level, in consistence with the experimental maximum of superconducting T_{c} of 80 K. General phase diagram Fig. 3 summarizes the proposed picture for the phase diagram. At low temperatures, the structural transition is a quantum order-disorder transition, which shifts the phase boundary to lower pressures compared to the classical limit. Above T_{c,\textup{LQD}}^{\textup{max}} , the system recovers the classical first-order transition behavior. Here, we suggest that experiments aiming to locate this boundary should track the change in lattice parameters and focus on temperatures just above the maximum superconducting T_{c} rather than at room temperature. The intersection of the quantum and first-order transition lines—which is the T_{c,\textup{LQD}}^{\textup{max}} —constitutes a tricritical point. Given the observed coincidence between the left boundary of the LQD phase and the left flank of the superconducting dome, we conclude that superconductivity occurs entirely within the high-symmetry phase, thereby refuting earlier two-phase interpretations. Crucially, the LQD phase is distinct from a static high-symmetry phase; its lattice dynamics, which transcend the conventional phonon picture, may host a novel pairing mechanism. Superconductivity should be viewed not only as a macroscopic quantum state of electronic degrees of freedom, but equally as one of the lattice. The energy scale of lattice dynamics (tens to hundreds of kelvins) is comparable to the T_{c} of high-temperature superconductors; at the very least, the process by which thermal effects suppress nuclear quantum fluctuations cannot be ignored. If one attributes the right flank of the dome to a BCS‑like mechanism, where T_{c} rises slowly with decreasing pressure, the quantum order‑disorder transition could be seen as interrupting this rise. However, this cannot explain why the T_{c,\textup{LQD}}^{\textup{max}} coincides precisely with the T_{c,\textup{SC}}^{\textup{max}} —that is, why a scenario such as that sketched in Extended Data Fig. 4 does not occur. Based on consistent first‑principles results across \textup{H}_{3}\textup{S} and \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} , we argue that this agreement is not coincidental. We instead propose that the right flank of the dome extends from the tricritical point, indicating a profound connection between LQD and the unconventional superconducting mechanism, with the LQD phase playing a decisive role in determining the maximum superconducting transition temperature. Figure 3: Schematic phase diagram of lattice quantum disorder and unconventional superconductivity. The quantum order-disorder transition boundary (solid red line), dominant at low temperatures, is shifted to lower pressures compared to the classical transition boundary (dashed black line). Their intersection defines T_{c,\textup{LQD}}^{\textup{max}} ——a tricritical point (red point). The region enclosed by these two boundaries is the LQD phase (shaded orange area). Experimentally observed superconductivity forms a dome (orange area). Crucially, the left flank of the superconducting dome coincides with the left boundary of the LQD phase, and T_{c,\textup{SC}}^{\textup{max}} aligns precisely with the tricritical point ( T_{c,\textup{LQD}}^{\textup{max}} ). We contend that this unequivocally points to a superconducting mechanism inherent to the LQD phase, with the tricritical point determining the maximum superconducting transition temperature. This picture suggests a practical route for predicting and discovering superconductors with higher T_{c} : first identify materials that host a large LQD phase, and subsequently consider appropriate carrier introduction. Regarding doping, we expect the LQD framework to remain relevant for doping‑dependent high- T_{c} families such as the cuprates 10 . Notably, the antiferromagnetic phase (AFM) that competes with superconductivity in these systems often also exhibits a dome shape in the T-P plane 9 ; 2 , suggesting that the AFM phase is coupled to a structural phase transition and the left flank of their superconducting domes may likewise originate from a quantum order‑disorder structural transition. Doping may influence the soft mode on the potential energy surface in a manner analogous to pressure, though its combined effect on both the lattice and carrier concentration is more complex and places higher demands on first‑principles treatments 8 . Finally, we suggest that the LQD phase could be a widespread phenomenon, with its extent varying across different materials. We anticipate that this perspective will offer fresh insights into other unresolved puzzles in condensed matter, particularly concerning anomalous transport properties such as glass‑like thermal conductivity in crystals 39 . Methods PIMD simulations We performed fixed-centroid PIMD to obtain the centroid mean force. During the PIMD simulation, the centroid was fixed at a finite-displaced structure and then the mean force was averaged from centroid forces of all the steps in the trajectory. A set of finite-displaced structures (0.03 Å from equilibrium position) were generated according to lattice symmetry by Phonopy software 43 . Then the corresponding set of centroid mean forces were used to construct the force constant, from which we can obtain the dispersion relation. The PIMD simulations were performed using the i-pi software 26 (version 3.1). We used NVT ensemble and Path-integral Langevin equation thermostat. For each centroid mean force, we ran 600,000 steps to ensure a sufficient sampling, with the time step being 0.5 fs. The number of beads N_{\textup{b}} was set to ensure N_{\textup{b}}\times T\geq 5,120 K, specifically, N_{\textup{b}}=32 for T=200,160 K and N_{\textup{b}}=64 for T=120,80 K. For \textup{H}_{3}\textup{S} and \textup{D}_{3}\textup{S} , we simulated in a 3\times 3\times 3 supercell of unit cell containing totally 216 atoms. For \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} , we simulated in a 2\sqrt{2}\times 2\sqrt{2}\times 1 supercell containing totally 192 atoms. The convergence on number of beads and size of supercell is shown in Extended Data Fig. 5 . Machine-learning potentials at DFT level The Born-Oppenheimer PES was constructed using a combination of density-functional theory (DFT) and the machine learning inter-atomic potential method. The DFT calculations are performed by the Vienna Ab initio Simulation Package (VASP) 23 at GGA level using the Perdew-Berke-Ernzerhof (PBE) functional 33 and projector augmented wave (PAW) pseudopotentials 24 . The machine learning techniques are used to save computational cost so that our PIMD simulation is doable 49 . Here we employ the machine-learning force field (MLFF) built in VASP 21 ; 20 As a practical test on the training accuracy of the MLFF, the harmonic phonon spectra (by Phonopy software) in comparison with the DFT one are shown in Extended Data Fig. 6 . All the atomic simulations in this work including PIMD and MD were performed on the MLFF. For \textup{H}_{3}\textup{S} , the plane-wave cutoff was set as 550 eV. We used a 8\times 8\times 8 Gamma k -mesh for a 2\times 2\times 2 supercell of unit cell. This unit cell of \textup{Im}\bar{3}\textup{m} phase contains 8 atoms. We ran MD in the 2\times 2\times 2 supercell containing 64 atoms at 150 Gpa, 125 GPa, 100 GPa and 75 GPa, to generate data set for training MLFF. At each pressure, the system was gradually heated from 1 K to 400 K in 25 ps with a time step of 0.5 fs. 3926 configurations were selected from the MD trajectories, constructing the training dataset. To better describe the local environment around each atom, we used a Gaussian broadening of 0.3 Å for atomic distribution function. The standard derivation of the training dataset is 0.18 eV/atom for energy, 0.39 eV/Å for force and 19.0 GPa for stress. The standard error of the MLFF on the dataset is 0.15 meV/atom for energy, 0.02 eV/Å for force and 0.3 GPa for stress. For \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} , the plane-wave cutoff was set as 700 eV. We used a 6\times 6\times 1 Gamma k -mesh for a 2\times 2\times 1 supercell of unit cell. This unit cell of I4/mmm phase contains 24 atoms. We ran MD in the 2\times 2\times 1 supercell containing 96 atoms at 0 Gpa, 10 GPa and 20 GPa, to generate data set for training MLFF. At each pressure, the system was gradually heated from 1 K to 200 K in 100 ps with a time step of 2 fs. 1854 configurations were selected from the MD trajectories, constructing the training dataset. To better describe the local environment around each atom, we used a Gaussian broadening of 0.3 Å for atomic distribution function. The standard derivation of the training dataset is 29.5 meV/atom for energy, 0.30 eV/Å for force and 5.5 GPa for stress. The standard error of the MLFF on the dataset is 0.05 meV/atom for energy, 0.01 eV/Å for force and 0.2 GPa for stress. Equation of state The pressure at a specific lattice parameter was determined by conventional (unfixed-centroid) PIMD. The number of beads N_{\textup{b}} was set to ensure N_{\textup{b}}\times T\geq 5,120 K. We used NVT ensemble and Path-integral Langevin Equation thermostat with a local langevin thermostat to the centroid mode 6 . The friction factor was \tau= 100 fs. The time step was 0.5 fs. The P-V curves at different T are shown in Extended Data Fig. 7 . Classical boundaries by MD For \textup{H}_{3}\textup{S} , the classical boundary of the structural phase transition between \textup{Im}\bar{3}\textup{m} and R3m was determined by the pair distribution function (PDF) from MD simulation. For each temperature and volume, we performed a 50 ps MD simulation in the 3\times 3\times 3 supercell. We used the NVT ensemble and the Langevin thermostat, with the friction factor being \tau= 100 fs. The S-H PDFs at a range of temperatures and lattice parameters were summarized in the upper panel of Extended Data Fig. 2 . The transition points were determined by the splitting of peaks. For \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} , PDF analysis lacks sufficient sensitivity to pinpoint the phase boundary in tetragonal-orthorhombic transitions driven by octahedral rotations. Instead, here, the classical boundary was identified by the pressure where the derivative of the lattice parameter ratio c/a with respect to pressure becomes zero. This criterion works because at structural phase boundaries involving changes in crystal symmetry (like tetragonal to orthorhombic), the axial ratio c/a typically exhibits an extremum. The point where \textup{d}(\frac{c}{a})/\textup{d}P_{\textup{MD}}=0 thus marks the transition pressure where one structure becomes more stable than another. For each temperature and pressure, we performed a 6 ns MD simulation in the 4\times 4\times 1 supercell. We used the NPT ensemble and the Langevin thermostat, with the friction factor being \tau= 100 fs. The results are summarized in the lower panel of Extended Data Fig. 2 . Figure 1: LQD phase and the SSCHA phase boundary in \textup{H}_{3}\textup{S} by different functionals. The structural phase boundary between \textup{Im}\bar{3}\textup{m} (high-pressure) and R3m (low-pressure) by the stochastic self-consistent harmonic approximation (SSCHA) method 30 is shown by the cyan line. SSCHA predicts a much lower transition pressure compared to PIMD. In comparison with PBE, the structural phase boundary by R2SCAN functional shifts to higher pressure because, at the same pressure, R2SCAN predicts a higher potential barrier hence weaker quantum fluctuation. Figure 2: Classical structural phase boundaries by MD. a-f , S-H pair distribution function for \textup{H}_{3}\textup{S} at different temperatures and pressures. The PDFs obtained from MD are identical for \textup{D}_{3}\textup{S} and \textup{H}_{3}\textup{S} . g , pair distribution function for \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} . h , derivative of the lattice parameter ratio c/a with respect to pressure. i , PIMD predicts a higher pressure than MD at the same volume. j , the classical transition pressure was determined by \textup{d}(\frac{c}{a})/\textup{d}P_{\textup{MD}}=0 . Figure 3: Harmonic phonon spectra of \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} by PBE+U. a-c , Hubbard U=4.0 eV at 0, 10 and 20 GPa. d-f , Hubbard U=6.0 eV at 0, 10 and 20 GPa. Figure 4: Schematic diagram of another possible phase diagram. Figure 5: Convergence on number of beads N_{\textup{b}} and size of supercell in PIMD simulation. a , the PIMD phase boundaries of \textup{H}_{3}\textup{S} when N_{\textup{b}}\times T=2,400,\ 5,120 and 10,240 K in a 3\times 3\times 3 or 2\times 2\times 2 supercell of unit cell. b , the PIMD phase boundaries of \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} when N_{\textup{b}}\times T=2,560 and 5,120 K. Figure 6: Comparison of MLFF PES and ab\ initio PES. a , harmonic phonon spectra of \textup{H}_{3}\textup{S} at P_{\textup{PIMD}}=141 Gpa (lattice parameter a=3.10~\textup{\AA} ). b , frequency of soft mode at \Gamma point when varying lattice parameter. c , harmonic phonon spectra of \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} at V_{\textup{unitcell}}=384.1~\textup{\AA}^{3} . d , frequency of soft mode at ( 1/2,1/2,0 ) k -point when varying volume. Figure 7: P-V curves for \textup{H}_{3}\textup{S} (a) and \textup{La}_{3}\textup{Ni}_{2}\textup{O}_{7} (b).