What the study found
A machine-learned moment tensor potential, trained on density functional theory data for amorphous silicon, identified two-level systems that differed in detail from those found with a modified Stillinger-Weber potential. The samples reproduced experimental results and showed similar average structural and dissipative properties, but the atomistic details were significantly different.
Why the authors say this matters
The authors suggest that using this approach helps classify two-level systems in amorphous silicon while preserving agreement with experimental and average dissipative behavior. They also conclude that the model reveals atomistic differences that are not captured in the comparison potential.
What the researchers tested
The researchers combined a moment tensor potential, trained on density functional theory data, with the activation-relaxation technique nouveau (ARTn) to identify and classify two-level systems in amorphous silicon. They compared the resulting samples with those obtained using a modified Stillinger-Weber potential and examined radial distribution function, defect concentration, internal friction, and atomistic details.
What worked and what didn't
The MTP-based samples recovered experimental results and had average structural and dissipative properties similar to those from the modified Stillinger-Weber potential, including radial distribution function, defect concentration, and internal friction. However, the underlying atomistic details differed, including the density and type of two-level systems. Bond-hopping two-level systems had similar density in both potentials, while more complex two-level systems, including those involving a Wooten-Winer-Weaire bond exchange, were about twice as common in the MTP-based models. The analysis also showed that, for the MTP-based models, two-level systems were mostly isolated and oscillated independently from each other.
What to keep in mind
The abstract does not describe experimental limitations beyond the comparison between the two potentials. The findings are limited to amorphous silicon and to the specific methods and properties described in the summary.
Key points
- The study used a density functional theory-trained moment tensor potential with ARTn to identify two-level systems in amorphous silicon.
- The MTP-based samples reproduced experimental results and matched average structural and dissipative properties seen with a modified Stillinger-Weber potential.
- Atomistic details differed, including the density and type of two-level systems.
- Bond-hopping two-level systems had similar density in both potentials, but more complex systems were about twice as common in the MTP-based models.
- In the MTP-based models, two-level systems were mostly isolated and oscillated independently.
Disclosure
- Research title:
- Machine-learned model finds different two-level systems in amorphous silicon
- Authors:
- Renaude Girard, Carl Lévesque, Normand Mousseau, F. Schiettekatte
- Institutions:
- Université de Montréal
- Publication date:
- 2026-05-06
- DOI:
- 10.1103/9g5j-9q4n
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.