Interatomic potentials for platinum

Imagine Platinum (Pt) as a very sophisticated, high-performance athlete. It's strong, doesn't rust easily even in extreme heat, and is used in everything from catalytic converters to medical devices. To understand how this "athlete" behaves under stress, heat, or pressure, scientists use computer simulations. But to run these simulations, they need a rulebook—a set of instructions that tells the computer how every single atom of platinum interacts with its neighbors. This rulebook is called an interatomic potential.

For a long time, the rulebooks available for platinum were a bit like old, worn-out maps. They had some errors: they predicted the metal would melt at the wrong temperature, or that it would be too easy to break certain internal bonds.

In this paper, the authors (Koju, Li, and Mishin) decided to write two brand-new, highly accurate rulebooks for platinum. Here is the breakdown of their work in simple terms:

1. The "Training" (No Human Guessing)

Usually, when scientists make these rulebooks, they look at real-world experiments to see if they are right. However, this team decided to be purely digital. They used a super-accurate quantum physics method (called DFT) to generate a massive "training database."

The Analogy: Imagine teaching a robot to play chess. Instead of showing it real games played by humans, you have the robot play millions of games against a perfect, math-based opponent. The robot learns the rules purely from the math, not from watching people.
The Result: They trained two new models on this pure math data. They didn't use any experimental measurements during the training phase.

2. The Two New Rulebooks

The authors created two different types of rulebooks, each with a different style:

The ADP Model (The "Flexible" Rulebook): This is an upgrade to an older, standard method. Think of the old method as a rule that says, "Atoms only care about how close their neighbors are." The new ADP version adds a twist: "Atoms also care about the angles their neighbors make." It's like saying a person doesn't just care about who is standing next to them, but also who is standing to their left or right. This makes the model very good at predicting how the metal bends and vibrates.
The MT Model (The "Adapted" Rulebook): This model was originally designed for things like diamonds or silicon (materials with very rigid, directional bonds). The authors took this rigid model and "stretched" it to fit a metal like platinum.
- The Analogy: Imagine a rulebook designed for a rigid wooden chair. The authors modified it so it could describe a soft, squishy metal pillow. Surprisingly, this "stretched" rulebook turned out to be incredibly accurate, sometimes even better than the ADP one.

3. The Results: Who Wins?

The team tested both new rulebooks against the old ones (the "worn-out maps") and the super-accurate quantum math.

Melting Point: The old rulebooks said platinum melts at a temperature that is hundreds of degrees too low. The new ADP rulebook got the melting point almost exactly right (within a tiny fraction of a degree). The MT rulebook was also very close, just slightly too high.
Breaking and Bending: The old rulebooks failed to predict how much energy it takes to create a "defect" (a missing atom) or to slide layers of atoms past each other (like shuffling a deck of cards). The new models fixed these errors, predicting the energy needed to break or slide the metal much more accurately.
Vibrations: When the metal vibrates (like a guitar string), the new models predicted the "notes" (frequencies) much better than the old ones.

4. The Trade-off: Speed vs. Accuracy

There is a catch.

The ADP model is like a fast sports car. It is very accurate and runs simulations quickly.
The MT model is like a high-tech, heavy tank. It is extremely accurate (sometimes even better than ADP), but it is very slow to run. It takes over 100 times longer to run a simulation with the MT model than with the ADP model because it has to calculate complex angles between atoms constantly.

5. Why This Matters (According to the Paper)

The authors suggest that while the MT model is slow for pure platinum, it might be the "missing link" for future materials.

The Analogy: Imagine you have a rulebook for water (liquid) and a rulebook for concrete (solid). But what if you need to simulate a material that is half-water and half-concrete, like wet cement? Neither rulebook works well alone.
The MT model is special because it can handle both metals (like platinum) and covalent materials (like carbon or silicon) using the same mathematical language.
Specific Applications Mentioned: The paper explicitly notes that this new model could be used to simulate platinum silicides (used in microchips) and platinum-based cancer drugs (where platinum bonds with nitrogen). It allows scientists to simulate how these mixed materials behave at the atomic level, which was very difficult to do before.

Summary

The authors built two new, highly accurate digital rulebooks for platinum atoms. They trained them using pure math, not experiments. Both are much better than the old versions, especially at predicting melting points and how the metal breaks. One is fast (ADP), and one is slow but incredibly versatile (MT). The versatile one might be the key to simulating complex materials that mix metals with other elements, like the chips in your phone or specific medicines.

Technical Summary: Interatomic Potentials for Platinum

Problem Statement
Classical atomistic simulations are essential for understanding the mechanical behavior and microstructural development of platinum (Pt), particularly under extreme conditions. However, the reliability of these simulations is contingent upon the accuracy of the interatomic potentials used. Existing potentials for Pt, primarily developed in the Embedded-Atom Method (EAM) format, exhibit significant limitations. These include the underestimation of stacking-fault energies and melting temperatures that deviate by several hundred degrees from experimental values. Such deficiencies can lead to reduced quantitative accuracy and, in some cases, nonphysical simulation behavior. Furthermore, existing potentials struggle to capture subtle electronic effects governing defect configurations and surface reconstructions.

Methodology
To address these limitations, the authors developed two new interatomic potentials for Pt: one in the Angular-Dependent Potential (ADP) format and another in the Modified Tersoff (MT) format.

Training Database: Both potentials were trained exclusively on a reference database generated from first-principles Density Functional Theory (DFT) calculations using the Vienna Ab initio Simulation Package (VASP). No experimental data were used during the training process. The database included:
- Equations of state (EOS) for various crystal structures (FCC, BCC, HCP, SC, A15, Diamond Cubic) under isotropic deformations.
- Uniaxial tension/compression and shear deformations for the FCC structure.
- Minimum energy paths (MEP) for vacancy migration via the Nudged Elastic Band (NEB) method.
- Surface energies, twin boundary energies, and $\gamma$ -surfaces.
- Point-defect formation energies (vacancies and interstitials) and dimer energies.
- Phonon dispersion relations and density of states.
Potential Formats:
- ADP: An extension of EAM that includes angular-dependent interactions (dipole vectors and quadrupole tensors) to account for non-central bonding. It utilizes 36 fitting parameters.
- MT: Originally designed for covalent elements (e.g., Si, C), the MT model was adapted for the metal Pt by expanding the cutoff radius to include multiple coordination shells (up to 6.57 Å). This allows the model to capture many-body interactions similar to MEAM and ADP. It utilizes 16 fitting parameters.
Optimization: Parameters for both potentials were optimized using a simulated annealing algorithm based on the Nelder-Mead simplex method to minimize the weighted mean-square deviation between potential predictions and the DFT reference database.
Validation: The new potentials were rigorously tested against the DFT database and experimental data. For consistency, two recently published EAM potentials (EAM1 by Zhou et al. and EAM2 by O'Brien et al.) were also re-evaluated using the same computational protocols.

Key Contributions and Results
The study presents a comprehensive comparison of the new ADP and MT potentials against existing EAM potentials, DFT calculations, and experimental data.

Lattice and Thermodynamic Properties:
- Both new potentials predict equilibrium lattice constants and cohesive energies with high accuracy.
- Melting Temperature ( $T_m$ ): The ADP potential reproduces the experimental melting temperature within 0.2% (2041 K vs. experimental ~2045 K). The MT potential overestimates $T_m$ by 7.3% (2195 K). In contrast, the EAM1 and EAM2 potentials significantly underestimate $T_m$ by several hundred Kelvin.
- Thermal Expansion: The MT potential shows excellent agreement with experimental thermal expansion data up to 1200 K, outperforming the ADP and EAM potentials.
- Phonons: The new potentials reproduce phonon dispersion relations significantly better than the EAM potentials, although all potentials slightly underestimate frequencies at high-symmetry points compared to experiment.
Defect Properties:
- Vacancies: The new potentials correctly predict vacancy formation energies within the experimental range, whereas the DFT calculations (PBE functional) grossly underestimate this value. Both ADP and MT potentials accurately reproduce the vacancy migration energy barrier, while EAM potentials significantly underestimate it.
- Interstitials: DFT calculations identify the octahedral interstitial as the ground state in Pt, a deviation from the typical $\langle 100 \rangle$ dumbbell preference in other FCC metals. While the new potentials correctly predict the $\langle 100 \rangle$ dumbbell as the ground state (failing to capture the subtle electronic effects causing the octahedral preference), they provide more accurate energy values than the EAM potentials, which systematically underestimate interstitial energies.
- Surfaces: All potentials correctly rank surface energies ( $\gamma_{111} < \gamma_{100} < \gamma_{110}$ ). However, none of the four potentials (ADP, MT, EAM1, EAM2) correctly predict the energy lowering associated with surface reconstructions (e.g., the $(1 \times 2)$ reconstruction of the (110) surface), a known deficiency.
Stacking Faults and Twin Boundaries:
- The new potentials demonstrate a marked improvement in predicting stacking fault energies (SFE) and twin boundary energies compared to EAM potentials.
- The ADP potential reproduces the DFT twin boundary energy (142 mJ m $^{-2}$ ) with excellent agreement. The MT potential overestimates this by ~50 mJ m $^{-2}$ , while EAM potentials drastically underestimate it (e.g., EAM1 yields 46 mJ m $^{-2}$ ).
- Similarly, the ADP potential provides the closest agreement to DFT for SFE, while EAM potentials grossly underestimate these values.
Structural Stability:
- All potentials correctly identify the FCC structure as the ground state. Parity plots of relative energies for non-equilibrium structures (BCC, HCP, etc.) show that the ADP and MT potentials exhibit unbiased scatter relative to DFT, whereas EAM potentials show a trend of underestimating energies.

Significance and Claims
The paper claims that the newly developed ADP and MT potentials offer superior accuracy for platinum compared to existing EAM potentials, particularly regarding melting temperature, stacking fault energies, and defect migration barriers.

Efficiency vs. Accuracy: The authors note a trade-off in computational cost. The MT potential, while highly accurate and requiring fewer parameters (16 vs. 36 for ADP), is computationally expensive (over $10^2$ times slower than ADP in MD simulations) due to the explicit calculation of bond angles in nested loops. The ADP potential offers a balance of high accuracy and computational efficiency.
Future Applications: The authors highlight the potential of the MT model to bridge the gap between metallic and covalent potentials. Because the MT formalism can describe both metallic and covalent bonding within a single analytical framework, it is proposed as a candidate for modeling mixed-bonding systems, such as platinum silicides (PtSi, Pt $_2$ Si) and platinum-nitrogen coordination compounds used in anticancer drugs. The authors suggest that the MT model's ability to handle hybrid bonding chemistry could enable large-scale simulations of these technologically relevant materials, despite its current computational overhead.

In conclusion, the work provides a robust, first-principles-trained set of potentials that corrects specific deficiencies in the literature, offering researchers improved tools for simulating the mechanical and physical behavior of platinum and potentially its compounds.