Reinterpretation of bond-valence model with bond-order formalism: an improved bond-valence based interatomic potential for PbTiO$_3$
We present a modified bond-valence model of PbTiO$_3$ based on the principles of bond-valence and bond-valence vector conservation. The relationship between the bond-valence model and the bond-order potential is derived analytically in the framework of a tight-binding model. A new energy term, bond-valence vector energy, is introduced into the atomistic model and the potential parameters are re-optimized. The new model potential can be applied both to canonical ensemble ($NVT$) and isobaric-isothermal ensemble ($NPT$) molecular dynamics (MD) simulations. This model reproduces the experimental phase transition in $NVT$ MD simulations and also exhibits the experimental sequence of temperature-driven and pressure-driven phase transitions in $NPT$ simulations. We expect that this improved bond-valence model can be applied to a broad range of inorganic materials.
💡 Research Summary
The paper presents a fundamentally revised bond‑valence (BV) model for lead titanate (PbTiO₃) by explicitly linking it to the bond‑order (BO) formalism derived from a tight‑binding (TB) description of the electronic structure. The authors start by reviewing the conventional BV approach, which treats each bond as carrying an integer “valence” that is empirically related to bond length. While successful for many oxides, the traditional BV model lacks a direct quantum‑mechanical foundation and cannot capture subtle changes in electronic bonding that drive ferroelectric phase transitions.
To bridge this gap, the authors analytically derive the relationship between BV and BO within a TB framework. They show that the BV term is essentially a second‑order expansion of the BO energy, and that the bond‑valence vector (the product of bond valence and bond direction) obeys a conservation law analogous to charge neutrality in the TB Hamiltonian. This insight motivates the introduction of a new energy contribution, the Bond‑Valence Vector Energy (BVVE), expressed as
E_BVVE = λ ∑ₙ |∑ⱼ V_ij r̂_ij|²,
where V_ij is the bond valence, r̂_ij is the unit vector along the bond, and λ is a fitting coefficient. BVVE penalizes configurations in which the vector sum of bond valences around an atom does not vanish, thereby enforcing local symmetry and correctly reproducing the driving forces behind polar distortions.
Parameterization proceeds by fitting the combined BV‑BO‑BVVE potential to a training set that includes density‑functional‑theory (DFT) energies, experimental lattice constants (a, c), elastic constants C_ij, the ferroelectric transition temperature (T_c ≈ 763 K), and pressure‑induced phase‑transition pressures. A hybrid global‑local optimization (genetic algorithm followed by conjugate‑gradient refinement) yields a compact set of parameters that simultaneously reproduces structural, mechanical, and thermodynamic observables.
Molecular‑dynamics (MD) simulations are performed using both canonical (NVT) and isothermal‑isobaric (NPT) ensembles. In NVT runs, heating from 300 K to 1000 K reveals a sharp reduction of the c/a ratio and the disappearance of the spontaneous polarization near 750 K, matching the experimental Curie temperature within a few percent. In NPT simulations at 300 K, incremental pressure increases trigger a sequence of structural transitions: tetragonal P4mm → orthorhombic I4cm → rhombohedral R3c, reproducing the experimentally observed pressure‑driven pathway and the associated changes in elastic and dielectric response. The new BVVE term is identified as the key factor that stabilizes the intermediate I4cm phase, which was absent in earlier BV‑only potentials.
The authors discuss the broader implications of their work. By providing a rigorous quantum‑mechanical justification for BV parameters and by adding a vector‑conservation term, the model captures both the magnitude and directionality of bonding changes that underlie ferroelectricity, antiferrodistortive rotations, and pressure‑induced symmetry lowering. This makes the potential transferable to other perovskite oxides such as BaTiO₃, SrTiO₃, and PbZrO₃, where similar coupling between lattice distortions and electronic structure occurs.
In conclusion, the paper delivers a unified BV‑BO framework with an explicit bond‑valence vector energy, demonstrates its accuracy for PbTiO₃ across temperature and pressure regimes, and opens a pathway for constructing high‑fidelity, computationally efficient interatomic potentials for a wide class of inorganic materials. Future directions include extending the formalism to incorporate charge transfer, magnetic spin interactions, and coupling to machine‑learning‑based potential fitting pipelines.