Skip to content

Supporting Information: Optimization of ReaxFF Reactive Force Field Parameters for Cu/Si/O Systems via Neural Network Inversion (with application to copper oxide interaction with silicon)

Evidence and attribution

Authority of statements

Prose sections below (Summary, Methods, Findings, etc.) are curated summaries of the publication identified by doi, title, and pdf_path in the front matter above. They are not new primary claims by this wiki.

For definitive numerical values, reaction schemes, and interpretations, use the peer-reviewed article (and optional records under normalized/papers/ when present)—not this page alone.

Summary

This Supporting Information PDF (papers/Roshan_JPCC_CuSiO_2023_SI.pdf) accompanies a J. Phys. Chem. C article on optimizing ReaxFF parameters for Cu/Si/O systems using neural-network inversion, with application emphasis on copper oxide interactions with silicon surfaces relevant to microelectronics and oxidation chemistry. The SI’s visible content (extract) is dominated by tabular disclosures: parameter indices, optimization bounds, and sensitivity metrics for bond, off-diagonal, and valence terms in the Cu–O and O–Cu–O blocks—material needed to reproduce the optimized force field alongside the main text loss functions and QM training sets. These tables are the machine-readable backbone for auditing how NN inversion redistributes error across interaction classes when fitting Cu/Si/O chemistry relevant to oxidized Cu on Si-bearing substrates. Until the parent article’s DOI is mirrored in front matter, treat bibliographic anchors as provisional and follow the JPCC PDF title in papers/ when it appears.

Methods

Document role (non-primary / SI catalog). The corpus ingests the JPCC Supporting Information at papers/Roshan_JPCC_CuSiO_2023_SI.pdf. All kinetic/oxide/interface science and any validation RMD belong in the parent JPCC VOR article; citable claims should be anchored there, not on this SI-only note alone.

1 — MD application (not owned by this SI-only PDF in this form)

N/A in this SI for a stand-alone, repro-ready LAMMPS story: the JPCC VOR+full SI for the same work would state how many atoms sit in the slab/supercell, which 3D PBC/periodic boundary conditions and NVT/NPT blocks apply, 0.1–0.5 fs class timestep choices typical for ReaxFF (fs-resolved RMD), ps/ns equilibration+production duration targets, and K-scale substrate temperature (plus Nose–Hoover-class thermostat when NVT is used). None of that is retyped on this page—use the parent JPCC PDF/SI bytes. N/A for E-field or rare-event (metadynamics) on this summary; E-field/metadynamics are out of scope for the table-centric excerpt.

2 — Force-field training (NN-*inversion* of ReaxFF for Cu/Si/O; this SI = tables )**

  • ReaxFF reoptimization: JPCC work uses a neural-network-inversion workflow to adjust selected ReaxFF classes in Cu–O / O–Cu–O-related manifolds; the local SI publishes parameter index mappings, bound constraints, and sensitivity to reproduce the fit; objectives and the DFT/QM training set that drove the loss landscape are in the JPCC main (not retyped here).

N/A (block 3) — static DFT in this file: the JPCC VOR+SI list Jaguar-class and other static QM reference data used to build the ReaxFF dataset—copy k-points, functionals, and basis from the primary PDF/SI, not this wiki summary.

Maintainer note (DOI/merge): add doi: 10.1021/acs.jpcc.3c03079 and a VOR paper: page when the parent article is fully registered in the corpus; the empty doi in front matter reflects a legacy stub, not a missing community DOI.

Findings

Operational content

The SI’s primary “result” is reproducibility metadata—which ReaxFF parameters moved, by how much, and with what sensitivity in the fit, versus a black-box release without indices and bounds. Compared to a bare field file, the tables help an operator reproduce the same inversion outcome when paired with the JPCC main-text objectives and QM data.

Scientific interpretation and corpus routing

DFT-referenced error metrics, reaction barriers, and oxide/interface morphology are stated in the JPCC main figures and text—not in this SI-only extract as a self-contained story. Limitation (inherent to SI-only routing): the laser-style or reactor-grade “experiment-level” agreement claims, if any, are in the VOR article, not the table list alone. Open question for MAS: whether future work will publish a unified JPCC+SI wiki page here—until then, use the VOR PDF and this papers/ path to resolve provenance (see ## Limitations for version-of-record caveats).

Corpus honesty: proof the JPCC PDF/SI if you need T(K), pressure (bar), or NVT timestep (fs)—not the summary line above; this page tracks ingested Roshan_JPCC_CuSiO_2023_SI.pdf for the BOM and repro of the ReaxFF tables only.

Limitations

SI-only ingest: no standalone narrative results; DOI must be taken from the parent article. Per docs/corpus/NON_PRIMARY_ARTICLE_PAPER_SLUGS.md, this row may stay pointer-heavy by policy. NN inversion workflows can be sensitive to training set coverage—treat tabulated bounds as necessary but not sufficient without the main text loss landscape and validation simulations.

Relevance to group

Documents group methodology connecting ReaxFF fitting to neural-network inversion for Cu/Si/O chemistry relevant to microelectronics and oxidation at Si interfaces. Operators should treat this page as SI provenance paired with a main JPCC article once the DOI is registered in front matter during corpus hygiene passes.

Citations and evidence anchors

  • papers/Roshan_JPCC_CuSiO_2023_SI.pdf; extract normalized/extracts/2023roshan-venue-paper_p1-2.txt. Locate the primary J. Phys. Chem. C article (filename stem Roshan_JPCC_CuSiO_2023) for DOI and pagination.