Relaxation approach to quantum-mechanical modeling of ferroelectric and antiferroelectric phase transitions Authors: Nikhilesh Maity, Sergey Lisenkov, Arlies Valdespino, Milan Haddad, Lewys Jones, Amit Kumar, Nazanin Bassiri-Gharb, Inna Ponomareva Published: 2026 ABSTRACT Ferroelectrics and antiferroelectrics are the electric counterparts of ferromagnets and antiferromagnets. These materials undergo temperature- and electric-field-induced phase transitions that give rise to their characteristic hysteresis loops. Modeling such hysteresis loops and associated phase transitions enables both a deeper fundamental understanding and reliable property predictions for this important class of materials. To date, modeling has largely relied on classical approaches, often remaining qualitative and/or empirical. Traditional interpretation of these transitions rests on two assumptions: (i) they are activated Arrhenius-type processes and (ii) they occur well within the classical regime. Here, we demonstrate that a model can instead be built on two “orthogonal" assumptions: (i) the phase transitions are relaxational processes and (ii) they require a quantum mechanical treatment. Applying this model to both antiferroelectrics and ferroelectrics overcomes the limitations of traditional models and enables efficient first-principles simulations of phase transitions. The success of our unconventional approach highlights the significance of quantum mechanics in transitions long regarded as purely classical. We anticipate that this framework will be applicable to a broad range of phase transitions, including magnetic, elastic, multiferroic, and electronic, along with modeling of quantum tunneling, rates of chemical reactions, and others. Introduction Ferroelectrics and antiferroelectrics are the electric counterparts of ferromagnets and antiferromagnets. They are an important class of materials both from fundamental science and applied perspectives. For example, ferroelectrics are used in Memory (FeRAM), sensors, actuators, tunable capacitors, energy harvesters1–3, while antiferroelectrics are utilized in high-energy-density capacitors, pulse power, electrocaloric cooling, and advanced thermal switches4–7. Ferroelectric and antiferroelectric phase transitions originate from the multiple-well free energy profile. As shown in Fig. 1a, d, the zero-Kelvin free-energy, U, as a function of polarization, P, exhibits a double-well form for the ferroelectric phase and a triple-well form for the antiferroelectric phase. The energies are computed from Density Functional Theory (DFT) simulations. Applied electric field tilts the energy profile (yellow line in Fig. 1) so that one of the minima becomes energetically more favorable (stable state) and the system will eventually transition into it. If the system was initially located in the other minima (metastable state, indicated by the black point in Fig. 1), it has to overcome the energy barrier. It is typically assumed that this is an Arrhenius-type process, which describes transitions over barriers significantly higher than kBT, where kB is the Boltzmann constant, and T is the temperature. This is an activated process since transition rate is proportional to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{E}}}}}_{A}/{k}_{B} {T}}$$\end{document}e−EA/kBT, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{E}}}}_{A}$$\end{document}EA is the activation energy, which is typically very close to the height of the energy barrier. As a result, if the barrier is very high, the transition rates are very low, giving rise to extremely long lifetimes of the metastable states.Fig. 1Comparison of computational predictions with experimental data for hysteresis loops.Energy per formula unit as a function of polarization for ferroelectric PbTiO3 (a) and antiferroelectric PbZrO3 (d) as computed from DFT simulations. Equation of state for PbTiO3 (b) and PbZrO3 (e) obtained from their supercell energies. Black and green dots indicate metastable state and ground state, respectively. The experimental RT polarization as a function of electric field for PbTiO3 from Ref. 11. c and for PbZrO3 measured in this work (f). For PbZrO3, the electric field is applied along the [111] pseudo-cubic direction. Panel c is reprinted from Materials Science and Engineering: B, Vol. 276, H. W. Shin and J. Y. Son, “Imprinted hysteresis loops and size-reduced ferroelectric polarization nanodots in epitaxial PbTiO3 thin film after heat treatment'', 115533, Copyright (2022), with permission from Elsevier. This panel is not covered by the Creative Commons license of this article. Comparison of computational predictions with experimental data for hysteresis loops. Energy per formula unit as a function of polarization for ferroelectric PbTiO3 (a) and antiferroelectric PbZrO3 (d) as computed from DFT simulations. Equation of state for PbTiO3 (b) and PbZrO3 (e) obtained from their supercell energies. Black and green dots indicate metastable state and ground state, respectively. The experimental RT polarization as a function of electric field for PbTiO3 from Ref. 11. c and for PbZrO3 measured in this work (f). For PbZrO3, the electric field is applied along the [111] pseudo-cubic direction. Panel c is reprinted from Materials Science and Engineering: B, Vol. 276, H. W. Shin and J. Y. Son, “Imprinted hysteresis loops and size-reduced ferroelectric polarization nanodots in epitaxial PbTiO3 thin film after heat treatment'', 115533, Copyright (2022), with permission from Elsevier. This panel is not covered by the Creative Commons license of this article. Figure 1d shows an energy landscape computed from DFT simulations for prototypical antiferroelectric PbZrO3. The transition rate can be estimated by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu {e}^{-{{{\mathcal{E}}}}_{A}/{k}_{B}T}$$\end{document}νe−EA/kBT, where ν is the attempt frequency and can be approximated with the soft mode frequency8. For the typical soft mode frequency of 1 THz, the transition rate at room temperature and the experimental forward-switching field of 315 kV/cm is 2.0 × 10−40 s−1, which is unphysical. As the field increases, the barrier \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {{{\mathcal{E}}}}_{A}$$\end{document}ΔEA decreases, and the rates go up. One way to quantify the fields required to overcome the barrier is through converting the energy U(P) in Fig. 1a, d into the equation of state as follows9: the free energy in the presence of the electric field, E, is F(P) = U(P) − EPV, where V is the volume. Setting the derivative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{\partial F}{\partial P}$$\end{document}∂F∂P to zero produces the equation of state, P(E), shown by the dashed line in Fig. 1b, e. Avoiding the thermodynamically unstable branches associated with negative susceptibility \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi=\frac{{\partial }^{2}F}{\partial {P}^{2}}$$\end{document}χ=∂2F∂P2 results in the loops shown by solid lines in Fig. 1b, e. The model predicts that at zero Kelvin, PbZrO3 exhibits mixed behavior, characterized by antiferroelectric-like double hysteresis loops together with a nonzero spontaneous polarization. In contrast, the experimentally measured hysteresis loop, shown in Fig. 1f, does not exhibit this mixed behavior. The measurement was performed on a highly 042-oriented PbZrO3 thin film (approximately 290 ± 5 nm thick), chemical solution processed on platinized silicon, under previously reported conditions10. For ferroelectrics (see Fig. 1b), we find coercive fields of ~ 3018 kV/cm, which exceeds by order of magnitude the experimental ones11 (~ 175 kV/cm) (see Fig. 1c). We elaborate that within the equation of state model the system transitions from the metastable phase to the stable one once the local minimum aligns with the barrier top, which corresponds to infinitely long lifetime of the metastable phase. Although the model does not give access to time or frequency dependence, the infinitely long lifetime of the metastable phase effectively models an extremely high frequency of the AC electric field so that the system never has time to experience the fluctuation needed to overcome the barrier as per the Arrhenius process. This explains unphysically high coercive/switching fields. In Ref. 12, this model was augmented with the frequency- and temperature-dependent transition rates, but the application of the model here still predicted coercive fields that are too high, suggesting that traditional transition state-based models13 may not be fully adequate to describe such phase transitions. Traditionally, the model failures are attributed to the inability of present DFT calculations to adequately capture the transition states and associated barriers14,15. Indeed, in DFT simulations, we are limited to small supercell simulations, which critically limits the phase space available to the system. For example, configurations with domains are unlikely to appear, although they have been shown to play a critical role in phase transitions16. The other possible culprit is the pristine nature of the material in simulations, while real materials have defects and surfaces that are expected to reduce the barriers for transitions. So far, success in modeling of ferroelectric and antiferroelectric phase transitions has been achieved with Landau-type approaches17,18, related phase-field modeling19,20, first-principles-based effective Hamiltonians21–23, shell models24,25, and machine-learned potentials26,27. These include predictions of phase transition temperatures, hysteresis loops, dynamics, temperature evolutions of polarizations among others. The common theme between them is that they all utilize classical frameworks. Nevertheless, the counterintuitive possibility that the classical treatment may not be sufficient for ferroelectrics has been raised in a few pioneering studies. Using the path integral technique, Zhong and Vandebilt showed that zero-point energy is sufficient to destroy ferroelectric ordering in quantum paraelectric SrTiO328. Subsequently, Geneste et al. applied path-integral molecular dynamics to BaTiO3 to reveal that, in contrast to the classical picture, this ferroelectric exhibits strong anharmonicity down to the lowest temperatures, which results in enhanced dielectric and piezoelectric responses29. Through the same path integral approach, it was found that quantum effects lead to a significant reduction of transition temperatures (up to 50 K) in BaTiO3, which revealed that quantum effects play a significant role even at temperatures as high as room temperature30. In methodologically different work, Esswein and Spaldin demonstrated that the quantum effects are responsible for classifying ferroic as ferroelectric, paraelectric, and quantum paraelectric, which can be captured through the single-particle Schrodinger equation with the DFT-calculated potential14. However, it has not been established whether these techniques are able to resolve the hysteresis loops controversies outlined above. Here, we introduce a fundamentally different first-principles framework for (anti)ferroelectricity. Instead of treating the phase transition classically and as an Arrhenius-type activated process, we describe it quantum mechanically and as a relaxation-driven evolution toward equilibrium. This unconventional framework (semi-)quantitatively reproduces ferroelectric and antiferroelectric hysteresis loops across a broad range of materials, providing strong evidence that quantum mechanics and relaxation dynamics are essential ingredients of these phase transitions. Results Ground-State Relaxation (GSR) model Let us begin with describing the system being in a quantum state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \psi \right\rangle$$\end{document}ψ that propagates in time, that is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \psi (t)\right\rangle$$\end{document}ψ(t). For time-independent Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H=\frac{{{{\mathcal{P}}}}^{2}}{2M}+U(P)$$\end{document}H=P22M+U(P), where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal{P}}}$$\end{document}P is the momentum, this can be achieved as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \psi (t)\right\rangle={\sum }_{n}{c}_{n}(0){e}^{-i\frac{{E}_{n}}{\hslash }t}\left\vert {\phi }_{n}\right\rangle$$\end{document}ψ(t)=∑ncn(0)e−iEnℏtϕn, where cn(0) = 〈ϕn∣ψ(t = 0)〉, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{n}\right\rangle$$\end{document}ϕn and En are the eigenstates and eigenvalues of the Hamiltonian, respectively. For a given Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left|{\phi }_{n}\right\rangle$$\end{document}ϕn En can be evaluated numerically. Suppose \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \psi (t=0)\right\rangle$$\end{document}ψ(t=0) corresponds to the metastable state of the system (black dot in Fig. 1a, d). Then at time t, the probability of finding the system in the state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{n}\right\rangle$$\end{document}ϕn is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$| {\langle {\phi }_{n}| \psi (t)| | \rangle }^{2}=| {c}_{n}(0){| }^{2}$$\end{document}∣⟨ϕn∣ψ(t)∣∣⟩2=∣cn(0)∣2 and independent of time. This means that the system will never settle into the ground state, which is the stable state associated with the green dot in Fig. 1a, d. However, any real system will relax from the excited to the ground state (black and green dots in Fig. 1a, d, respectively), owing to the interaction with the environment. This can be incorporated into the model by replacing En in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \psi (t)\right\rangle$$\end{document}ψ(t) with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{n}^{{\prime} }={E}_{n}-i\hslash \frac{{\gamma }_{n}}{2}$$\end{document}En′=En−iℏγn2, where γn = 1/τn is the relaxation rate, while τn is the associated relaxation time from the excited state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{n}\right\rangle$$\end{document}ϕn into the other states31. We will later discuss how relaxation rates can be computed but for now we take them to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{n}={k}_{scale}{\sum }_{{E}_{m} < {E}_{n}}\frac{{E}_{n}-{E}_{m}}{h}$$\end{document}γn=kscale∑Em \ {E}_{n}}\frac{{E}_{m}-{E}_{n}}{h}{e}^{-\frac{{E}_{m}-{E}_{n}}{{k}_{B}T}}\right]$$\end{document}γn=kscale∑EmEnEm−Enhe−Em−EnkBT. There exists analytical solution for the relaxation term: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{nm}^{rlx}(t)={\rho }_{nm}^{eq}(1-{e}^{-{\gamma }_{nm}t})+{\rho }_{nm}(0){e}^{-{\gamma }_{nm}t}$$\end{document}ρnmrlx(t)=ρnmeq(1−e−γnmt)+ρnm(0)e−γnmt. Finally, we augment Eq. (2) with a term that allows to conserve trace of the density matrix (see Supplementary Materials (SM) for derivation) 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nm}}{dt}=-\frac{i}{\hslash }{[H,\rho ]}_{nm}-{\gamma }_{nm}({\rho }_{nm}-{\delta }_{nm}{\rho }_{nm}^{eq})+{\rho }_{nm}{\sum}_{j}{\gamma }_{j}({\rho }_{jj}-{\rho }_{jj}^{eq})$$\end{document}dρnmdt=−iℏ[H,ρ]nm−γnm(ρnm−δnmρnmeq)+ρnm∑jγj(ρjj−ρjjeq) Figure 3a predicts polarization evolution during relaxation from the metastable state of PbZrO3 into a stable one computed within the DMB model for different temperatures. At 0.1 K, it agrees well with the one from relaxation to the GSR model. At higher temperatures, relaxation occurs much faster owing to the temperature-dependent excitation rates in Eq. (2). The P(E) loops were computed in the temperature range of 10 K to 1000 K and are given in Fig. S2a. Representative data are shown in Fig. 3c and demonstrate that the loops slim down as temperature increases, capturing the temperature evolution from antiferroelectric to paraelectric behavior. Note that for all the data presented so far, we turn off intrinsic dynamics, as it is expected to average to zero. Later on, we comment on its effects.Fig. 3Predictions from DMB and LO models.a, b Time evolution of the polarization in PbZrO3 during the transition from the metastable state to the stable state under a DC electric field of 400 kV/cm at different temperatures, computed using the models specified in the panel titles. c, d Polarization as a function of electric field in PbZrO3 at different temperatures, computed using the models specified in the panel titles. The simulations were performed using an AC field frequency of ν = 0.17 THz and an integration time step of Δt = 0.1 fs. Predictions from DMB and LO models. a, b Time evolution of the polarization in PbZrO3 during the transition from the metastable state to the stable state under a DC electric field of 400 kV/cm at different temperatures, computed using the models specified in the panel titles. c, d Polarization as a function of electric field in PbZrO3 at different temperatures, computed using the models specified in the panel titles. The simulations were performed using an AC field frequency of ν = 0.17 THz and an integration time step of Δt = 0.1 fs. Lindblad Operators (LO) models To provide further grounds for the relaxation approach for quantum-mechanical modeling of antiferroelectrics, we implement it within Lindblad formalism. We begin with stating the master equations, that govern dynamics in the presence of relaxational processes, in terms of Lindblad operators Ljk33, which represent interaction of the system with the environment: 4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d\rho }{dt}=-\frac{i}{\hslash }[H,\rho ]+{\sum}_{kj}[2{L}_{kj}\rho {L}_{kj}^{{\dagger} }-\{{L}_{kj}^{{\dagger} }{L}_{kj},\rho \}]$$\end{document}dρdt=−iℏ[H,ρ]+∑kj[2LkjρLkj†−{Lkj†Lkj,ρ}] where curly brackets indicate anticommutator. The Lindblad operator, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{kj}=\sqrt{{\gamma }_{kj}}\vert {\phi }_{k}\rangle \langle {\phi }_{j}\vert$$\end{document}Lkj=γkj∣ϕk⟩⟨ϕj∣, describes transition from state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\vert {\phi }_{j}\rangle$$\end{document}∣ϕj⟩ into state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{k}\right\rangle$$\end{document}ϕk caused by the interaction with the environment, at a rate given by γkj. The summations go over all eigenstates. We use the same transition rates as for DMB approach, namely for the down-transitions (Ek < Ej) the rate is γkj = kscale∣Ej − Ek∣/h, while for the up-transitions the rate is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{jk}={k}_{scale}{e}^{-\frac{{E}_{j}-{E}_{k}}{{k}_{B}T}}| {E}_{j}-{E}_{k}| /h$$\end{document}γjk=kscalee−Ej−EkkBT∣Ej−Ek∣/h. The Lindblad approach preserves the trace of the density matrix, so we do not have to renormalize it. In the basis of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{n}\right\rangle$$\end{document}ϕn the master equation (4) becomes 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nm}}{dt}=-\frac{i}{\hslash }{[H,\rho ]}_{nm}+2{\delta }_{nm}{\sum}_{j}{\gamma }_{nj}{\rho }_{jj}-{\rho }_{nm}{\sum}_{k}({\gamma }_{kn}+{\gamma }_{km})$$\end{document}dρnmdt=−iℏ[H,ρ]nm+2δnm∑jγnjρjj−ρnm∑k(γkn+γkm) The second term on the right hand side describes the density in-flow into populations ρnn due to transitions from all the other states, while the third term describes the density out-flow (from both populations and coherences, ρnm) due to transitions to all the other states. Comparing this equation with Eq. (3) we note that the overall rate of population decay from a given state is twice the one in DMB model. We bring the two models into closer correspondence by rescaling the rates in the DMB model by a factor of 2. We note that the LO model preserves detailed balance. Indeed, let us consider the change in ρnn due to relaxation and excitations to state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{m}\right\rangle$$\end{document}ϕm. From Eq. (5) we obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nn}}{dt}=2{\gamma }_{nm}{\rho }_{mm}-2{\gamma }_{mn}{\rho }_{nn}$$\end{document}dρnndt=2γnmρmm−2γmnρnn. At equilibrium \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nn}}{dt}=$$\end{document}dρnndt= 0, so we have γnmρmm = γmnρnn, which is the condition of detailed balance. Furthermore, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{nm}/{\gamma }_{mn}={\rho }_{nn}/{\rho }_{mm}={e}^{-\frac{{E}_{n}-{E}_{m}}{{k}_{B}T}}$$\end{document}γnm/γmn=ρnn/ρmm=e−En−EmkBT, which justifies our choice for the relaxation rates in LO model. Let us contrast this to previous models. From Eq. (3) we obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nn}}{dt}=-{\gamma }_{nn}({\rho }_{nn}-{\rho }_{nn}^{eq})+{\rho }_{nn}{\sum }_{j}{\gamma }_{j}({\rho }_{jj}-{\rho }_{jj}^{eq})$$\end{document}dρnndt=−γnn(ρnn−ρnneq)+ρnn∑jγj(ρjj−ρjjeq). The sum in the last term is the same for all populations. Therefore, the rate of population change due to normalization is proportional to the population. The equation does not have an explicit dependence on density matrix elements for any other state, unlike the LO approach. So it is clear that the detailed balance is not satisfied. This equation is better interpreted as the flow of populations to and from the heat reservoir. If we now apply this to relaxation to the ground state models, we get \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\rho }_{nn}}{dt}=-{\gamma }_{nn}{\rho }_{nn}+{\rho }_{nn}{\sum }_{j}{\gamma }_{j}{\rho }_{jj}$$\end{document}dρnndt=−γnnρnn+ρnn∑jγjρjj. Since both γj and ρnn are nonnegative, we can easily see that the first term on the right-hand side describes the rate of population loss, while the second one describes the rate of population gain. Figure 3b, d gives the prediction from the LO Model. We find that the shape of the hysteresis loops changes significantly with respect to the previous approaches. In particular, the low temperature loops are slanted rather than square in a better agreement with experimental data in the case of PbZrO3 (see Fig. 1d). On the basis of our data, we conclude that all models developed here are capable of describing antiferroelectric behavior with switching fields comparable to experimental ones. This result provides strong evidence that (anti)ferroelectric phase transitions can be well modeled as relaxational processes within a quantum mechanical framework. Furthermore, DMB and LO models give access to finite-temperature predictions. The methodological difference between the models is in the way they model dissipative terms. The GSR model accomplishes that through decay of individual states followed by subsequent renormalization of the state vector. In the DMB approach, the decay and repopulation of the states is achieved through implicit exchange with the thermal bath. In the LO model, a decrease in the population of one state is accompanied by an equal increase in the population of another state, thereby preserving detailed balance. We believe that these differences result in the differences in predictions of P(t) and P(E) evolutions given in Fig. 3. The DMB approach predicts step-like P(t) evolution at low temperatures consistent with some experimental data34,35, while LO-based models predict exponential evolution. Consequently, the DMB model predicts square hysteresis loops in contrast to slanted ones from the LO model. Discussion of the models parameters, approximations, and limitations Our models require the energy as a function of polarization along the distortion path (see Figs. 1 and S7), which is computed here using DFT. The distortion path itself is generated by linear interpolation between two or more phases, as described in the Technical Details of the DFT Simulations section. Such linear interpolation corresponds to a homogeneous polarization reversal. We argue that this homogeneous polarization reversal represents the nucleation of a nanodomain of the stable phase within the matrix of the metastable phase. Within the well-established nucleation-limited-switching model36, the switching dynamics are dominated by the time required for domain nucleation, whereas subsequent domain propagation occurs on a much shorter timescale. We therefore believe that our model accurately captures the rate-limiting process governing phase switching, which explains its predictive success. In the case of PbZrO3, the supercell volume is 4.4 nm3, which is sufficiently large to represent a realistic nanoscale nucleus. The model also requires the mass associated with the polar mode, which enters the Hamiltonian as a parameter. This mass can be determined from DFT calculations. In particular, the harmonic frequency at the bottom of the potential well satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }^{2}=\frac{1}{M}\left\langle {\phi }_{0}\left| \frac{{d}^{2}U}{d{x}^{2}}\right| {\phi }_{0}\right\rangle,$$\end{document}ω2=1Mϕ0d2Udx2ϕ0, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\phi }_{0}\right\rangle$$\end{document}ϕ0 is the ground-state wavefunction. In this work, we determine the mass using the frequency of the polar mode computed from DFT. Equivalently, the mass can be chosen such that, within the harmonic approximation, the energy spacing between the ground and first excited states satisfies ΔEharm = ℏω. The mass scales with the size of the simulation supercell and controls both the energy-level spacing and the spatial localization of the eigenstates. Larger masses lead to more closely spaced energy levels and more localized eigenstates, whereas smaller masses produce the opposite behavior. Consequently, increasing the supercell size drives the system from a quantum-mechanical regime toward a classical one. The model additionally requires the Born effective charge, Z*, which is used to convert polarization into the corresponding supercell distortion according to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x=\frac{PV}{{Z}^{*}}$$\end{document}x=PVZ*. The Born effective charge can be readily obtained from DFT calculations (see SM). The remaining model parameter is the relaxation rates, γnm, which are taken here to be proportional to Bohr frequencies. Relaxation rates for an open quantum system connected to a heat bath could be computed using the “golden rule" for the transition rates326\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{kj}=\frac{2\pi }{\hslash Z}{\sum}_{N{N}^{{\prime} }}| \langle kN| {{\mathcal{V}}}| j{N}^{{\prime} }\rangle {| }^{2}{e}^{-{E}_{{N}^{{\prime} }}/{k}_{B}T}\delta ({E}_{{N}^{{\prime} }}-{E}_{N}-\hslash {\omega }_{kj})$$\end{document}γkj=2πℏZ∑NN′∣⟨kN∣V∣jN′⟩∣2e−EN′/kBTδ(EN′−EN−ℏωkj) where N and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}^{{\prime} }$$\end{document}N′ label the states of the heat bath and Z is the partition function, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathcal{V}}}={\sum }_{i}{Q}_{i}{F}_{i}$$\end{document}V=∑iQiFi couples system and bath, where Fi and Qi operate on the system and bath, respectively. The reverse rate is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\gamma }_{jk}={\gamma }_{kj}{e}^{-\frac{{E}_{j}-{E}_{k}}{{k}_{B}T}}$$\end{document}γjk=γkje−Ej−EkkBT. Although the analytical expression is available, in practice, the rates are notoriously difficult to calculate as they require the knowledge of the coupling term and the eigenspectrum of the bath. As a result, the rates are often found in an empirical way, either using some judicious arguments37 or fitting to experimental data. Some examples include relaxation times for the Bloch equations38, relaxation parameter for stimulated emission and absorption39, and damping rates for the alternamagnet-cavity system40. In SM, we derive the expressions for transition rates due to coupling between the polar phonon and acoustic phonons in the framework of the Debye model, which, under certain approximations, predicts linear dependence of transition rates on the Bohr frequencies used here. However, since some of the parameters of the model are not readily available, we follow the standard practice to fit rates to experimental data. In particular, the thermal average relaxation rate for our model is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \gamma \rangle=2{\sum }_{n}{p}_{n}^{{{\rm{eq}}}}{\gamma }_{n}$$\end{document}⟨γ⟩=2∑npneqγn, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{n}^{{{\rm{eq}}}}$$\end{document}pneq is the Boltzmann probability of the eigenstate n. Then this value is fit to experimental γexp for the corresponding mode through kscale = γexp/〈γ〉. The experimental values can be obtained from the literature on IR or Raman measurements, as they are related to the width of the spectral lines. The values of γexp for different materials used here are given in the SM. The role of intrinsic dynamics Including intrinsic dynamics increases the computational cost substantially and, in some cases, requires reducing the integration time step by a factor of 100. We attribute this to the presence of high Bohr frequencies, which reach up to 17 THz in PbZrO3. However, we expect the effects of intrinsic dynamics to average out on the timescale of relaxation. Indeed, kscale represents the ratio of the relaxation and intrinsic-dynamics timescales and, in our simulations, ranges from 4.8 × 10−3 to 6.6 × 10−3. To test this expectation, we repeated a hysteresis-loop simulation for PbZrO3 at 300 K with intrinsic dynamics included and found no discernible differences in the resulting loop. Application to ferroelectrics Figure 4a, b shows hysteresis loops computed for ferroelectric PbTiO3 using the DFT-calculated U(P) shown in Fig. 1a and the two finite-temperature models developed in this work. The full set of results is provided in Fig. S4. These results demonstrate that the developed approaches perform well for ferroelectrics, reproducing ferroelectric hysteresis loops. We note that, for the LO model, the loops do not exhibit a significant temperature dependence at the chosen electric-field frequency. This may be due to the use of a temperature-independent relaxation rate.Fig. 4Models predictions for various ferroelectrics and antiferroelectrics.a, b Polarization-electric field hysteresis loops of PbTiO3 computed at different temperatures using the models indicated in the panel titles. The simulations were performed using ν = 1.7 THz and an integration time step of Δt = 0.01 fs. c Polarization--electric field hysteresis loops computed for different ferroelectric materials using the DMB model. The simulations were performed using ν = 1.7 THz and Δt = 0.01 fs. d Polarization--electric field hysteresis loops computed for different antiferroelectric materials using the DMB model. The simulations were performed using ν = 0.17 THz and Δt = 0.1 fs. All other simulation parameters are provided in the SM. Models predictions for various ferroelectrics and antiferroelectrics. a, b Polarization-electric field hysteresis loops of PbTiO3 computed at different temperatures using the models indicated in the panel titles. The simulations were performed using ν = 1.7 THz and an integration time step of Δt = 0.01 fs. c Polarization--electric field hysteresis loops computed for different ferroelectric materials using the DMB model. The simulations were performed using ν = 1.7 THz and Δt = 0.01 fs. d Polarization--electric field hysteresis loops computed for different antiferroelectric materials using the DMB model. The simulations were performed using ν = 0.17 THz and Δt = 0.1 fs. All other simulation parameters are provided in the SM. To further assess the performance of the models, we apply them to additional ferroelectric and antiferroelectric materials, including nonperovskites. Figure 4c shows hysteresis loops computed at 300 K for PbTiO3, CsGeBr3, HfO2, and AlN. The model reproduces the well-known trend that the coercive fields of AlN and HfO2 are significantly larger than those of oxide ferroelectrics such as PbTiO3. We also note that the coercive field predicted for CsGeBr3 is larger than that for PbTiO3, which may be related to the use of a different exchange-correlation functional (see SM for details). For PbHfO3 (Fig. 4d), the model correctly predicts antiferroelectric loops. Experimental fields for forward switching in PbHfO3 are in the range of 215–615 kV/cm, and for backward switching, they are in the range of 175–360 kV/cm41–43. So our predictions are in the range of experimental values. Note that we used the same AC-field frequency for all ferroelectric materials and the same AC-field frequency for all antiferroelectric materials. The frequencies differ between the two groups because DFT underestimates the coexistence field in PbZrO3 and PbHfO3. The coexistence field is defined as the electric field at which the polar and antipolar phases have the same energy and corresponds experimentally to the field at the center of the hysteresis loop. Within DFT, it is given by Ecoex = ΔU/(VP), where ΔU is the zero-field energy difference between the polar and antipolar phases. We obtain coexistence fields of 181.2 and 124.9 kV/cm for PbZrO3 and PbHfO3, respectively, from DFT, compared to experimental values of 216.5 kV/cm10 and 190–470 kV/cm41–43, respectively. This underestimation effectively contracts the hysteresis loops along the electric-field axis. To account for this effect, we use a lower AC-field frequency for the antiferroelectric materials. Based on our results, we conclude that treating (anti)ferroelectric phase transitions as relaxational processes within a quantum mechanical framework offers an unconventional yet powerful approach to their modeling and understanding. This framework overcomes the limitations of the traditional description of such transitions as Arrhenius-type processes within a classical regime and enables efficient, first-principles-based simulations of phase transitions. We anticipate that this approach will be applicable to a broad spectrum of other phase transitions – including magnetic, ferroelastic, and multiferroic ones. Furthermore, the framework may provide an alternative perspective for modeling chemical reaction rates and be applicable to modeling quantum tunneling phenomena and electronic transitions. Equally important, the model is directly applicable to phase transitions driven by electronic rather than ionic degrees of freedom44–47, such as proton-ordering and hopping ferroelectrics48. Beyond these specific cases, the framework provides a direct means to probe quantum-mechanical features underlying a wide variety of phase transitions. Technical details of DFT simulations All DFT calculations used to compute the energy profiles were performed using the Vienna Ab initio Simulation Package (VASP)49,50. The projector augmented-wave (PAW) method51,52 was employed to describe the ion–electron interaction. Exchange and correlation effects were treated within the local density approximation (LDA), except for CsGeBr3, see SM for justification. A plane-wave kinetic-energy cutoff of 600 eV was used throughout. Structural relaxations, where required, were performed using the conjugate-gradient algorithm with simultaneous relaxation of both the ionic coordinates and lattice parameters until the Hellmann–Feynman forces on all atoms were below 0.005 eV,Å−1. A Γ-centered k-point mesh corresponding to a reciprocal-space spacing of 0.22 Å−1 was used in all calculations. Polarization was computed using the modern theory of polarization based on the Berry-phase formalism53–55. Material-specific computational details are provided in the SM. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file