Effect of uniaxial compressive stress on polarization switching and domain wall formation in tetragonal phase BaTiO3 via machine learning potential

This study utilizes a machine learning potential to demonstrate that uniaxial compressive stress significantly influences polarization switching and domain wall evolution in tetragonal BaTiO3, revealing a critical threshold of approximately 120 MPa for 90-degree switching, stress-induced reductions in remnant polarization and coercive field, and the emergence of double hysteresis loops at 80 MPa.

The Gist

The Big Picture: Squeezing a Memory Chip

Imagine a tiny, magical crystal called Barium Titanate (BaTiO₃). This crystal is the heart of many modern devices, like memory sticks, sensors, and actuators. Its superpower is polarization: think of it as a tiny internal compass needle that points in a specific direction. When you apply an electric field, you can flip this needle to point the other way. This flipping is how the device stores data (0s and 1s).

However, this crystal isn't just sitting in a vacuum; it's often squeezed or stretched by its surroundings. The researchers in this paper wanted to answer a simple question: What happens to this internal compass if we squeeze the crystal from the top?

To find out, they didn't just use a physical press; they used a "digital microscope" powered by Machine Learning. This allowed them to watch every single atom dance and react in real-time, something too expensive and slow to do with traditional computer models.

---

The Key Discoveries

1. The "Tipping Point" (120 MPa)

Imagine you are pushing down on a stack of dominoes. At first, nothing happens. You push a little harder, and they just bend slightly. But then, you hit a specific amount of force, and suddenly, the whole stack tips over sideways.

The researchers found a similar critical tipping point for this crystal at about 120 MPa (a measure of pressure).

• Below this pressure: The crystal stays upright, just bending a little.
• Above this pressure: The internal "compass needles" (polarization) suddenly flip 90 degrees, pointing sideways instead of up. It's like the crystal decided, "I can't stand up anymore under this weight, so I'm going to lie down."

2. The "Crowded Room" vs. The "Big Hall" (Supercells)

To see what happens inside the crystal, the researchers simulated it in different-sized virtual rooms.

• Small Room: When they used a small simulation box, the atoms felt "crowded" by the walls. The walls (periodic boundary conditions) forced the atoms to stay in a single, uniform line. It was hard for them to break apart.
• Big Hall: When they used a larger box, the atoms had more room to breathe. The "walls" were far away, so they didn't feel as constrained.

The Analogy: Imagine trying to form a line of people in a tiny closet versus a large gymnasium. In the closet, everyone is forced to stand shoulder-to-shoulder. In the gym, people can naturally drift apart and form smaller, distinct groups.
The study found that in the larger "gymnasium" simulations, it was much easier for the atoms to break their uniform line and form Domain Walls.

3. What are "Domain Walls"?

Think of the crystal as a field of wheat.

• Single Domain: All the wheat stalks are leaning in the exact same direction (North).
• Domain Wall: If you squeeze the field, some stalks might lean North, while others next to them lean South. The invisible line where the North-leaning stalks meet the South-leaning stalks is the Domain Wall.

The study showed that when you squeeze the crystal hard enough, these walls form naturally. Interestingly, the width of these walls (how many atoms it takes to transition from North to South) stays pretty much the same, no matter how hard you squeeze. It's like the "transition zone" has a fixed personality.

4. The "Double-Loop" Mystery

Usually, when you flip a switch, it clicks once. But under a specific amount of pressure (80 MPa), the researchers saw something weird: a Double Hysteresis Loop.

The Analogy: Imagine a door with a heavy spring.

• Normal: You push the door, it swings open, and when you let go, it swings back.
• Double Loop: You push the door, and it gets stuck halfway. You have to push harder to get it to the other side. Then, when you let go, it doesn't just swing back; it gets stuck in the middle again, and you have to pull it back.

This "double loop" means the material is confused. It's fighting between staying upright and lying down. The pressure makes the "up" state unstable, but not quite unstable enough to fully collapse, creating a weird, two-step switching process. If you squeeze even harder (160 MPa), the crystal gives up entirely and acts like a non-magnetic material (paraelectric), refusing to hold a memory at all.

---

Why Does This Matter?

This research is like a manual for engineers building the next generation of tiny electronic devices.

1. Designing Better Memory: If you are building a memory chip, you need to know exactly how much pressure the material can take before it flips its data or loses its memory. This paper tells us that 120 MPa is the danger zone.
2. Controlling the Switch: By understanding how "room size" (the size of the material grain) affects the formation of domain walls, engineers can design materials that switch faster or more reliably.
3. The "Double-Loop" Effect: This weird double-switching behavior might actually be useful for new types of sensors or switches that need to operate in a specific, narrow range of pressure.

The Bottom Line

The researchers used a super-smart AI to simulate squeezing a crystal. They found that:

• Squeezing too hard makes the crystal flip its internal direction.
• Giving the atoms more space makes it easier for them to form complex patterns (domain walls).
• Moderate squeezing creates a confusing "double-switch" behavior.

This helps scientists build stronger, smarter, and more reliable electronic devices that can handle the physical stress of the real world.

Technical Summary

1. Problem Statement

Ferroelectric materials like Barium Titanate (BaTiO₃) are critical for memory, actuator, and sensor applications due to their ability to reorient polarization under external fields. However, their performance is heavily influenced by mechanical boundary conditions (stress) due to intrinsic electromechanical coupling.

• The Gap: Existing methods to study stress-polarization coupling have limitations:
  • Phase-field and Landau-Ginzburg-Devonshire (LGD) models: Efficient for macroscopic scales ($\mu$m) but lack atomic-scale resolution and cannot capture specific microscopic mechanisms.
  • Core-shell models: Offer atomic resolution but often lack quantitative accuracy in energy and force predictions.
  • Ab initio Molecular Dynamics (AIMD): Highly accurate but computationally prohibitive for the large supercells and long timescales required to observe domain wall (DW) formation and switching dynamics.
• Objective: To bridge the gap between macroscopic stress effects and atomic-scale behavior by investigating how uniaxial compressive stress influences polarization switching, hysteresis loops, and the spontaneous formation of domain walls in tetragonal BaTiO₃.

2. Methodology

The study employs a hybrid computational framework combining Machine Learning Potentials (MLP) with Molecular Dynamics (MD) simulations.

• Machine Learning Potential (MLP):
  • Utilized a fine-tuned MACE-MP-0 model trained on a database of 4,045 structural configurations generated via Density Functional Theory (DFT) using the PBEsol functional.
  • Validated against DFT and experimental data for elastic constants, showing high agreement (particularly with PBEsol results).
  • Capable of reproducing temperature-dependent phase transitions and accurate atomic forces/energies comparable to AIMD but at a fraction of the cost.
• Born Effective Charge (BEC) Prediction:
  • Since MLPs do not inherently calculate electric fields, a Graph Neural Network (GNN) model (Equivar_eval) was fine-tuned on a specific BEC database to predict Born effective charges directly from structural information.
  • This reduced the Mean Absolute Error (MAE) from 0.1691 to 0.0697, enabling efficient calculation of polarization during MD.
• Simulation Setup:
  • Ensemble: NPT ensemble (constant pressure and temperature) at 250 K.
  • System Sizes:
    • $8\times8\times8$ supercell (2,560 atoms) for stress-strain analysis and P-E hysteresis loops.
    • $8\times8\times32$ supercell (10,240 atoms) and varying lengths (64–128 Å) to study domain wall nucleation and formation probability.
  • Stress Application: Uniaxial compressive stress applied along the c-axis (polarization direction) ranging from 0 to 800 MPa.
  • Electric Field: Triangular-wave electric field (max 100 kV/cm, 2.5 GHz) applied to simulate P-E hysteresis loops.

3. Key Contributions

1. Development of a Robust MLP-MD Framework: Successfully integrated a fine-tuned MACE potential with a GNN-based BEC predictor to simulate ferroelectric switching under mechanical stress with DFT-level accuracy.
2. Identification of Critical Stress Thresholds: Determined specific stress levels that trigger structural phase transitions and polarization switching mechanisms.
3. Mechanistic Insight into Domain Wall Formation: Provided atomistic evidence of how uniaxial stress induces the spontaneous formation of 180° domain walls, a process difficult to capture with previous methods.
4. Size-Dependent Switching Dynamics: Revealed the relationship between supercell size, activation energy, and the probability of domain wall nucleation.

4. Key Results

A. Stress-Induced Polarization Switching

• Critical Stress (~120 MPa): A distinct transition occurs at approximately 120 MPa. Below this, the lattice constants change linearly with stress. Above this threshold, a 90° polarization switching event occurs where the polarization vector rotates from the c-axis to the a/b-plane.
• Lattice Response: Beyond 120 MPa, the $b$-axis lattice constant becomes the largest, while the $c$-axis contracts, confirming the reorientation of the ferroelectric axis.

B. Domain Wall (DW) Formation

• Spontaneous Nucleation: Under high compressive stress (e.g., 800 MPa), the system transitions from a single-domain state to a multi-domain state featuring 180° domain walls.
• DW Characteristics:
  • The formed walls are Ising-type, where polarization rotates strictly along the $b$-axis without in-plane rotation.
  • DW Width: Decreases with increasing stress, stabilizing at ~5.5 Å (consistent with theoretical predictions of <1 nm).
  • Supercell Length Effect: DW formation probability is zero for short supercells (64 Å) due to periodic boundary condition (PBC) constraints. As length increases (up to 128 Å), the probability rises to ~0.9.
  • Activation Energy: Larger supercells exhibit lower activation energies for DW-mediated switching compared to single-domain switching, explaining the higher nucleation probability in larger systems.

C. P-E Hysteresis Loop Evolution

• Stress-Free State: Reproduces a standard hysteresis loop with a remnant polarization ($P_r$) of ~20 $\mu$C/cm² and coercive field ($E_c$) of ~50 kV/cm.
• Low Stress (80 MPa): A double hysteresis loop emerges. This indicates a two-step switching process ($c \to \pm a/b \to -c$) where the competition between phases creates distinct switching events.
• High Stress (>160 MPa): The hysteresis loop collapses into a nearly linear, paraelectric-like response. The coercive field drops significantly, and remnant polarization vanishes, indicating the suppression of ferroelectricity along the original axis.

5. Significance

• Atomic-Scale Validation: The study validates that machine learning potentials can faithfully reproduce complex electromechanical phenomena (switching, DW formation) that were previously inaccessible to large-scale simulations.
• Device Design Implications: The findings highlight that mechanical loading is a critical design parameter. Uniaxial stress can be used to:
  • Tune the coercive field and remnant polarization.
  • Induce or suppress domain wall formation, which is crucial for memory retention and switching speed.
  • Trigger double-hysteresis behaviors useful for specific sensor or actuator applications.
• Fundamental Physics: The research clarifies the mechanism of stress-induced 90° switching and the role of system size (PBC constraints) in lowering energy barriers for domain nucleation, providing a deeper understanding of ferroelectricity in perovskite oxides under mechanical load.