Transition from Homogeneous to Domain-Wall-Mediated Polarization Switching in BaTiO3: A Machine-Learning Molecular Dynamics Study

Using machine-learning molecular dynamics, this study reveals that polarization switching in BaTiO3 transitions from a homogeneous to a domain-wall-mediated mechanism as supercell size increases, driven by size-dependent fluctuations that significantly raise the coercive field and depend critically on system geometry and stress-field orientation.

The Gist

Imagine a block of special material called Barium Titanate (BaTiO₃). Inside this material, tiny atoms act like millions of tiny compass needles. Normally, they all point in the same direction, creating an electric "memory" (polarization). When you apply an electric field, you want these needles to flip over to point the other way. This flipping is called polarization switching, and it's the heart of how ferroelectric devices store data.

For a long time, scientists weren't sure exactly how these needles flip. They thought there were two main ways this could happen, but they didn't know what decided which way the material would choose.

This paper acts like a detective story, using a super-powerful computer simulation (powered by Machine Learning) to watch these atoms flip in real-time. Here is what they found, explained simply:

1. The Two Ways to Flip a Switch

Think of the material as a crowd of people in a room.

• Homogeneous Switching (The "Wave"): Imagine everyone in the room turning around at the exact same time, in perfect unison. It's smooth, fast, and requires less effort. This happens in small blocks of material.
• Domain-Wall Switching (The "Ripple"): Imagine a small group in the corner decides to turn around first. Then, the "turning" spreads out like a ripple or a wave moving through the crowd until everyone is facing the other way. This happens in large blocks of material.

2. The "Size" Surprise

The biggest discovery in this paper is that size matters more than anyone thought.

• When the researchers simulated a small block of material, the atoms flipped all together (the "Wave").
• When they simulated a larger block, the atoms didn't flip together. Instead, they started flipping in small pockets that grew and merged (the "Ripple").

The Analogy: Think of a small rubber band versus a giant rubber sheet. If you pull a small rubber band, it stretches evenly. If you pull a giant sheet, it might wrinkle or fold in specific spots before the whole thing moves. The paper shows that as the material gets bigger, it naturally prefers to "fold" (create domain walls) rather than stretch evenly.

3. The "Chaos" Meter (Shannon Entropy)

How did they know why this happened? They used a concept called Shannon Entropy, which is basically a "Chaos Meter."

• In the small blocks, the atoms were very orderly and predictable.
• In the large blocks, the atoms were much more "chaotic" or jiggly.
• The Finding: This extra jiggling (fluctuation) in the large blocks makes it easier for a small group of atoms to break away and start a new "domain" (a ripple). The paper proves that this local chaos is the trigger that forces the material to switch from the "Wave" method to the "Ripple" method.

4. The Cost of Flipping

Because the "Ripple" method involves creating these new boundaries (domain walls) and overcoming the chaos, it is harder to do.

• The Result: The larger blocks required a much stronger electric push (about 50% more force) to flip the switch compared to the small blocks.
• The Takeaway: If you simulate a small piece of material, you might think the material is easy to switch. But in the real world (where materials are large), it's actually much harder because it switches via the "Ripple" method.

5. Direction and Pressure Matter Too

The paper also found that the shape of the block and the direction you push it changes the story:

• Direction: Pushing the electric field along the long side of the block is harder than pushing it along the short side. It's like trying to push a long line of dominoes from the end versus the side; the physics changes based on the geometry.
• Pressure: If you squeeze the material (apply stress) in the same direction you are trying to flip the switch, it makes the "Ripple" method even more dominant and changes how the material behaves. If you squeeze it from the side, it barely matters.

Summary

This paper tells us that system size isn't just a number in a computer code; it's a physical law.

• Small systems = Smooth, easy flipping (Homogeneous).
• Large systems = Chaotic, ripple-based flipping (Domain-Wall), which requires much more energy.

The authors conclude that to understand how real-world devices work, scientists must simulate large enough blocks of material to see these "ripples." If they only look at tiny blocks, they are missing the true, more difficult way nature flips the switch.

Technical Summary

Technical Summary: Transition from Homogeneous to Domain-Wall-Mediated Polarization Switching in BaTiO3

Problem Statement
Ferroelectric polarization switching in tetragonal Barium Titanate (BaTiO3) is known to proceed via fundamentally different microscopic pathways: homogeneous collective reversal or domain-wall (DW)-mediated switching involving nucleation and propagation. While extensive experimental and computational studies have characterized these behaviors, the specific conditions determining which mechanism is operative remain poorly understood. Previous atomistic simulations have yielded conflicting results, with NVT ensemble simulations often favoring domain nucleation and NPT simulations favoring homogeneous switching. It is unclear whether these discrepancies arise solely from simulation ensemble choices or reflect an intrinsic physical crossover dependent on system size and geometric constraints. Furthermore, the relationship between supercell size, polarization fluctuations, and the resulting coercive fields has not been systematically clarified.

Methodology
The authors employed machine-learning potential-based molecular dynamics (MLP-MD) simulations to investigate polarization switching in tetragonal BaTiO3 under the NPT ensemble with external electric fields.

• Potential Models: The study utilized a finetuned MACE (Machine Learning Atomic Cluster Expansion) model for interatomic forces and a specialized MACEField model for predicting Born Effective Charges (BEC). The MACEField model was trained on a dataset of 2,038 solid-phase BaTiO3 structures with BEC data calculated via Density Functional Perturbation Theory (DFPT). This model demonstrated a mean absolute error (MAE) of ~0.01 for BEC predictions, significantly outperforming the previous Equivar model (MAE ~0.03) for solid-state structures, while maintaining high computational efficiency through cuEquivariance acceleration.
• Simulation Setup: Simulations were conducted at 250 K using triangular electric fields with a maximum amplitude of 150 kV/cm and a frequency of 1.67 GHz. The study systematically varied the supercell size along the crystallographic c-axis (from 8×8×8 to 8×8×32 unit cells) to probe finite-size effects.
• Analysis Metrics: Polarization switching mechanisms were analyzed by tracking Ti-atom displacements relative to oxygen coordination centers to determine local polarization vectors. Shannon entropy analysis was applied to the distribution of these local polarization directions to quantify configurational disorder and polarization fluctuations. Hysteresis loops (P-E curves) were generated to determine coercive fields and switching pathways under various conditions, including different electric field orientations (b-axis vs. c-axis) and uniaxial compressive stresses (parallel and perpendicular to the field).

Key Results

1. Size-Dependent Mechanism Transition: A clear transition from homogeneous switching to domain-wall-mediated switching was observed as supercell size increased.
   • Small Supercells (8×8×8): Switching occurred homogeneously with a coercive field of approximately 55 kV/cm.
   • Large Supercells (≥8×8×16): Switching proceeded via the nucleation of 180° domains followed by domain-wall propagation. This transition was accompanied by a >50% increase in the coercive field (rising to ~85 kV/cm for b-axis fields and ~110 kV/cm for c-axis fields).
2. Role of Polarization Fluctuations: Shannon entropy analysis revealed that larger supercells exhibit significantly higher polarization fluctuations. These enhanced fluctuations facilitate the local instability required for 180° domain-wall nucleation, establishing a quantitative link between local configurational disorder and the macroscopic switching mechanism.
3. Anisotropy and Geometry: The switching behavior is highly sensitive to the orientation of the applied electric field relative to the supercell geometry.
   • Under a c-axis electric field, the 8×8×32 supercell exhibited a higher coercive field (~110 kV/cm) and a switching pathway involving enhanced lateral fluctuations and transient 90° DW-like structures before forming 180° DWs.
   • Under a b-axis electric field, the coercive field was lower (~85 kV/cm), and domain growth was more gradual.
   • The authors attribute these differences to periodic boundary conditions: the elongated c-axis allows greater spatial variation, while the constrained a/b directions (8 unit cells) impose stronger lateral constraints on DW formation.
4. Electromechanical Coupling: The influence of uniaxial compressive stress on switching depends critically on its orientation relative to the electric field.
   • Parallel Stress: Compressive stress applied parallel to the electric field stabilizes intermediate polarization states, leading to double hysteresis loops. However, the critical stress required to induce this behavior increases with supercell size (e.g., requiring 160 MPa for the 8×8×32 c-axis case vs. 80 MPa for smaller systems).
   • Perpendicular Stress: Stress applied perpendicular to the switching direction has a minimal effect on the hysteresis behavior.

Significance and Claims
The paper asserts that homogeneous and domain-wall-mediated switching represent distinct physical regimes in BaTiO3, governed primarily by system size rather than just simulation ensemble artifacts. The study establishes that:

• System size is an intrinsic physical variable: It is not merely a computational parameter to be minimized but a determinant of the operative switching mechanism.
• Finite-size effects are physically meaningful: The transition to domain-wall-mediated switching in larger systems, characterized by higher coercive fields and specific fluctuation patterns, may be more representative of the behavior observed in real ferroelectric materials and devices than homogeneous switching observed in small supercells.
• Methodological Validation: The MACEField model provides a robust and computationally efficient framework for simulating electric-field-driven dynamics in large-scale ferroelectric systems, overcoming previous limitations in BEC calculation speed and accuracy.

The authors conclude that future atomistic simulations of ferroelectric switching must carefully account for system size to correctly capture the operative mechanism, and that the domain-wall-mediated regime is likely the dominant behavior in macroscopic systems.