Systematic comparison of approximations and functionals in first-principle calculations of aluminum-based III-V ferroelectric nitrides

This study systematically evaluates the impact of chemical disorder modeling (VCA vs. SQS) and exchange-correlation functionals (PBE, PBESol, SCAN, SCAN+rVV10) on the structural and ferroelectric properties of Al-based III-V nitrides, revealing that the SQS approach combined with the SCAN functional provides the most reliable framework for predicting phase stability and identifying metastable states in these materials.

The Gist

Imagine you are an architect trying to design the perfect building material for a new type of super-fast, memory-storing computer chip. You have two main ingredients: Aluminum Nitride (a sturdy, reliable brick) and a second ingredient that you can mix in to change its properties. You can mix in either Scandium (a heavy, metallic element) or Boron (a tiny, lightweight element).

The goal is to create a material that acts like a "ferroelectric" switch—a material that can remember if it's "on" or "off" by flipping its internal electrical direction. However, predicting exactly how these mixed materials behave is like trying to guess the weather in a chaotic storm. You need a computer model to simulate the atoms, but the model itself has flaws depending on how you set it up.

This paper is essentially a massive "stress test" of different computer models to see which one tells the truth about these aluminum-based nitride materials.

The Two Main Problems the Authors Investigated

The authors found that getting the right answer depends on solving two specific puzzles:

1. The "Crowded Room" vs. "Average Person" Problem (Disorder)
When you mix Aluminum with Scandium or Boron, the atoms don't sit in a perfect, repeating pattern like soldiers in a line. They are messy and random, like a crowded party where everyone is jostling for space.

• The Old Way (Virtual Crystal Approximation): Imagine trying to describe this party by saying, "The average person is 5'9" and wearing a blue shirt." This is the Virtual Crystal Approximation (VCA). It smooths out the chaos. The paper shows this method is a bad liar; it makes the material look stable when it's actually unstable, or vice versa. It's like saying a house made of sand and water is solid because the "average" of sand and water is "mud."
• The New Way (Special Quasirandom Structures): This is like taking a photo of the actual messy party, with specific people standing in specific spots. This is the Special Quasirandom Structure (SQS). The authors found that to get the right answer, you must look at the specific, messy arrangement of atoms, not just the average.

2. The "Lens" Problem (Functionals)
Even if you have the right messy arrangement, you still need to look at it through a specific mathematical "lens" (called an exchange-correlation functional) to calculate the energy. The authors tested four different lenses: PBE, PBESol, SCAN, and SCAN+rVV10.

• The Result: Some lenses (like PBESol) were blurry and distorted the picture, making the material seem unstable too early. Others (like SCAN) were like high-definition glasses, showing the true stability of the material.

What They Discovered About the Two Mixtures

The paper reveals that mixing in Scandium and mixing in Boron are like two completely different stories, even though they start with the same base material.

Story A: Mixing with Scandium (The Heavy Metal)

• The Behavior: When you add Scandium, the atoms want to huddle closer together. They start preferring a "crowded" arrangement (called the Rocksalt phase) over the "spacious" arrangement (the Wurtzite phase) that holds the memory switch.
• The Surprise: The "blurry" models (VCA) predicted this switch would happen very quickly, at low Scandium levels. But the "high-definition" models (SQS + SCAN) showed that the material stays stable and useful for a much longer time—up to nearly 50% Scandium. This matches what real-world experiments have seen.
• The Twist: There is a weird, in-between state (a 5-sided hexagonal phase) that acts like a stepping stone. It's a metastable "rest stop" the atoms visit before settling into the final crowded state.

Story B: Mixing with Boron (The Tiny Element)

• The Behavior: Boron is tiny and prefers to sit in a flat, 3-sided triangle shape rather than a 3D pyramid. When you add Boron, it forces the structure to break apart and reconfigure.
• The Breakage: At moderate amounts of Boron, the bonds between atoms actually snap and rearrange. The material gets distorted, and the "memory switch" (polarization) actually gets stronger initially, which is a good thing.
• The End Game: If you add too much Boron, the material gives up on the 3D pyramid shape entirely and turns into a flat, layered sheet (like graphite or a stack of paper). This is a total change in personality.

The Final Verdict: The "Golden Standard"

After testing every combination of "messy room" models and "lenses," the authors concluded that the best way to predict how these materials will behave is to use:

1. SQS: To capture the real, messy randomness of the atoms.
2. SCAN: To use the most accurate mathematical lens available.

Why does this matter?
The paper doesn't claim to build a new computer chip today. Instead, it provides the blueprint for the blueprint. It tells scientists, "If you want to design a new ferroelectric material, don't use the old, easy math tools. Use this specific, more complex combination of tools, or your predictions will be wrong."

By using the right tools, they confirmed that Scandium mixtures are very stable and promising for memory devices, while Boron mixtures are tricky—they can boost performance but only if you stop adding them before the structure collapses into flat sheets.

In short: Don't trust the average; look at the chaos. And don't use a blurry lens; use the high-definition one.

Technical Summary

Technical Summary: Systematic Comparison of Approximations and Functionals in First-Principle Calculations of Aluminum-Based III-V Ferroelectric Nitrides

Problem Statement
The discovery of ferroelectricity in III-V nitride semiconductors, particularly aluminum scandium nitride (Al$_{1-x}$Sc$_x$N) and aluminum boron nitride (Al$_{1-x}$B$_x$N), has shifted focus from traditional perovskite oxides. While first-principles calculations based on Density Functional Theory (DFT) have been central to understanding these materials, significant methodological uncertainties remain. Previous studies have predominantly relied on the Virtual Crystal Approximation (VCA) to model chemical disorder and have often utilized specific exchange-correlation functionals like PBE or PBESol without systematic benchmarking. There is a lack of comprehensive studies examining how the choice of disorder treatment (VCA vs. Special Quasirandom Structures, SQS) and the selection of exchange-correlation functionals (PBE, PBESol, SCAN, SCAN+rVV10) quantitatively and qualitatively affect the predicted structural, energetic, and ferroelectric properties of these disordered ternary alloys. Specifically, it is unclear if meta-GGAs like SCAN offer improved predictive accuracy for nitride ferroelectrics and whether computationally cheaper methods like VCA are sufficient for capturing the complex phase stability landscapes.

Methodology
The authors employ a unified computational framework based on a 48-atom supercell to compare five competing polymorphs: polar wurtzite (WZ), zincblende (ZB), hexagonal (HX), hexagonal-layered (HL), and rocksalt (RS). This supercell approach allows for the description of all phases on an equal footing by transforming lattice vectors.

The study systematically varies two critical methodological parameters:

1. Disorder Treatment: The authors compare the Virtual Crystal Approximation (VCA), implemented in the Abinit code, against Special Quasirandom Structures (SQS), generated using the Alloy Theoretic Automated Toolkit (ATAT) and calculated within the Vienna Ab initio Simulation Package (VASP).
2. Exchange-Correlation Functionals: The study benchmarks four functionals: PBE, PBESol, SCAN (a meta-GGA), and SCAN+rVV10 (incorporating van der Waals interactions).

Calculations utilize Projector Augmented Wave (PAW) pseudopotentials with a 620 eV plane-wave cutoff. The authors analyze total Kohn-Sham energies, lattice constants, internal parameters ($u$), bond distributions, spontaneous polarization (via Berry phase), Born effective charges, and electronic bandgaps across varying Sc and B concentrations ($x$).

Key Contributions and Results

• Impact of Disorder Treatment (VCA vs. SQS):

  • Al$_{1-x}$Sc$_x$N: The VCA severely underestimates the stability domain of the ferroelectric WZ phase. VCA predicts a transition from WZ to the non-polar rocksalt (RS) phase at approximately 19% Sc content, whereas SQS calculations predict this transition occurs near 47.8% Sc. The SQS results align much better with experimental reports of ferroelectricity up to 43% Sc.
  • Al$_{1-x}$B$_x$N: VCA fails to capture strong structural distortions caused by boron's tendency toward three-fold coordination. SQS reveals that at intermediate boron concentrations (20–50%), bond breaking occurs, leading to distorted WZ and ZB structures that are dynamically stable but distinct from regular phases. VCA cannot reproduce these local disorder effects.

• Impact of Exchange-Correlation Functionals:

  • PBE vs. PBESol: Despite their similarity, these functionals yield divergent results. PBESol significantly underestimates the critical Sc concentration for the WZ-to-RS transition (predicting ~29.1% vs. ~47.8% for PBE) and incorrectly predicts the ZB phase as the ground state for high boron concentrations, whereas PBE predicts the hexagonal-layered (HL) phase.
  • PBE vs. SCAN: The SCAN meta-GGA functional shows excellent qualitative and quantitative agreement with PBE for most properties but provides a more consistent description of phase energetics. SCAN predicts the WZ-to-RS transition in Al$_{1-x}$Sc$_x$N at $x \approx 0.434$, which is closer to experimental observations than PBESol.
  • Van der Waals Effects: Including rVV10 corrections in SCAN (SCAN+rVV10) shifts the RS phase stability to lower Sc concentrations ($x \approx 0.359$) and reduces the stability window of the HL phase in Al$_{1-x}$B$_x$N. However, given the computational overhead and the fact that SCAN alone already provides robust agreement with experiments, the authors suggest SCAN is the preferred functional for these systems.

• Structural and Ferroelectric Properties:

  • Al$_{1-x}$Sc$_x$N: Increasing Sc content leads to a decrease in the $c/a$ ratio and a reduction in spontaneous polarization. The system exhibits a complex energetic landscape where WZ, HX, and RS phases are nearly degenerate near the transition point. The HX phase (5-fold coordinated) is identified as a low-energy metastable state predicted by Farrer and Bellaiche (2002), appearing between the WZ and RS phases.
  • Al$_{1-x}$B$_x$N: In contrast to Sc doping, B doping leads to a rapid destabilization of the WZ phase. At intermediate concentrations, B atoms prefer three-fold coordination, causing bond breaking and structural distortion. This results in an enhancement of spontaneous polarization at low boron concentrations before the structure transitions to ZB or HL phases.
  • Electronic Properties: Sc doping in AlN significantly reduces the bandgap, particularly in the RS phase, due to the hybridization of Sc $d$ states with N $p$ states, introducing a partial Mott character. Conversely, B doping in AlN generally increases the bandgap, though the evolution is non-monotonic due to bond reconfigurations.

Significance and Claims
The paper claims that the combination of Special Quasirandom Structures (SQS) with the SCAN meta-GGA functional provides the most consistent and reliable theoretical framework for predicting the properties of emerging nitride-based ferroelectric materials. The authors emphasize that:

1. Local disorder is critical: The VCA is insufficient for accurately modeling the stability windows and structural distortions in these alloys, particularly for Al$_{1-x}$B$_x$N where local bond breaking is significant.
2. Functional choice matters: Even among similar functionals (PBE vs. PBESol), quantitative differences are substantial. SCAN offers a superior balance of accuracy and computational cost compared to hybrid functionals, outperforming GGA functionals in describing the structural and energetic properties of these ferroelectrics.
3. Distinct physical mechanisms: The study establishes clear structure-property relationships driven by coordination tendencies: Sc promotes higher coordination (favoring RS/HX phases), while B promotes lower coordination (favoring ZB/HL phases).

The work does not propose new experimental applications but rather provides a methodological roadmap and fundamental physical insights necessary for the accurate design and optimization of next-generation nitride ferroelectrics. It clarifies that previous discrepancies between theory and experiment may stem from the use of inappropriate approximations (VCA) or functionals (PBESol) rather than a failure of the underlying physics.