Computational chemistry
Neural nets predict TM spin states and bonds
Open access · cc by · source: Europe PMC
Graph-based ANNs trained on 2690 DFT geometries of Cr–Ni octahedral complexes predict spin splitting and metal–ligand distances without 3D coordinates.
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Neural nets can learn DFT spin-state gaps for a defined ligand set. On octahedral Cr–Ni complexes, test RMSE was 3.1 kcal mol⁻¹ and ground state was right in 528/538 test cases; CSD complexes were worse (MUE 10 kcal mol⁻¹ overall, 5 inside a reliability cutoff), and cyclams and early metals were outliers around 30 kcal mol⁻¹. Predictions target the B3LYP-family DFT the network was trained on.
Evidence for the claim as stated.
These papers do not agree on what DFT is for. One trains a statistical model to correct DFT barriers; one trains a net to emulate DFT spin states; others use a single-point or periodic calculation to rationalise a crystal, a gas-phase ion, or a ligand redox event. Quoting a kcal mol⁻¹ figure without saying whether it is raw DFT, a GP correction, or an NNP barrier mixes those jobs.
Evidence for the claim as stated.
Open questions
Tensions this paper is part of
From concept pages' “where studies disagree.” Disagreement means the same question; scope means different assays, populations, or outcomes.
These papers do not agree on what DFT is for. One trains a statistical model to correct DFT barriers; one trains a net to emulate DFT spin states; others use a single-point or periodic calculation to rationalise a crystal, a gas-phase ion, or a ligand redox event. Quoting a kcal mol⁻¹ figure without saying whether it is raw DFT, a GP correction, or an NNP barrier mixes those jobs.
Discoveries this paper informs or conflicts with
- A neural network predicted spin states well — and reported where it should not be trusted
This paper informs this development.
Study at a glance
- Design
- Computational / modelling — ANN trained on DFT octahedral Cr–Ni complexes to predict spin states and bond lengths
- N
- N=2690 · 2,690 DFT geometry optimizations after filtering spin-contaminated cases
- Population
- Homoleptic and heteroleptic octahedral transition-metal complexes
- Outcome
- Predicted spin-state energetics and metal–ligand bond lengths
Structured fields used in claim comparison tables when every cited study has a complete layer.
Inspect source passages
Neural nets predict TM spin states and bonds
These passages occur verbatim in the stored paper text. They support the points listed below; they do not verify every statement in the explanation.
Final DFT data set has 2690 geometry optimizations
Eliminating these cases produced a final data set of 2690 geometry optimizations (structures and energetics provided in ESI, as outlined in ESI Text S1).
Spin-state splitting RMSE is 3.0 and 3.1 kcal mol–1 for train and test
The comparable RMSE of 3.0 and 3.1 kcal mol –1 for the training and test data, respectively, indicate an appropriate degree of regularization.
Key findings
Spin-state RMSE is 3.0 (train) and 3.1 kcal mol–1 (test); ground state is right in 98% of test cases (528/538). Bond-length RMSE is 0.02 Å (LS) and 0.03 Å (HS test). On 35 CSD complexes MUE is 10 kcal mol–1 overall and 5 kcal mol–1 inside a Euclidean reliability cutoff; GGA-to-hybrid slope correction brings CSD MUE to 5 kcal mol–1.
Methodology
They DFT-optimized homoleptic and heteroleptic octahedral complexes of Cr–Ni at 0–30% HF exchange, kept 2690 jobs after dropping spin-contaminated cases, and trained dropout ANNs on connectivity descriptors (connecting atom, ligand charge, denticity, metal/oxidation state). Test split was 40%; a CSD set checked transfer.
Limitations
The model is for octahedral first-row complexes with the ligand set used in training; cyclams and early metals are outliers (~30 kcal mol–1). Descriptors ignore 3D interligand clashes. Predictions target DFT (B3LYP-family) numbers, not experiment directly.
Related papers in this topic
Same topic cluster — not a recommendation engine.