The BOS-TMC Dataset: DFT Properties of 159k Experimentally Characterized Transition Metal Complexes Spanning Multiple Charge and Spin States

Aaron G. Garrison; Christopher J. Stein; Heather J. Kulik; Jacob W. Toney; Roland G. St. Michel; Tatiana Nikolaeva

arxiv: 2604.07623 · v1 · submitted 2026-04-08 · ⚛️ physics.chem-ph · cond-mat.mtrl-sci

The BOS-TMC Dataset: DFT Properties of 159k Experimentally Characterized Transition Metal Complexes Spanning Multiple Charge and Spin States

Aaron G. Garrison , Jacob W. Toney , Tatiana Nikolaeva , Roland G. St. Michel , Christopher J. Stein , Heather J. Kulik This is my paper

Pith reviewed 2026-05-10 16:53 UTC · model grok-4.3

classification ⚛️ physics.chem-ph cond-mat.mtrl-sci

keywords transition metal complexesdensity functional theorydatasetspin statescharge assignmentCambridge Structural Databaseatomization energiesmachine learning

0 comments

The pith

The BOS-TMC dataset supplies DFT-computed properties for 159k experimentally characterized transition metal complexes across multiple charges and spin states.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper presents a large collection of quantum chemical calculations on real-world transition metal complexes taken from the Cambridge Structural Database. The authors developed an iterative method to determine the overall charge on each complex and then computed properties in up to three different spin states for each one. They kept the experimental positions of the heavy atoms fixed during geometry optimization to stay close to observed structures. The resulting set includes over 2.9 million individual property values and is substantially larger and more varied in charge and spin than previous collections. Such a resource should support better machine-learning models for predicting properties of these chemically important molecules.

Core claim

The paper introduces the Boston Open-Shell Transition Metal Complex dataset containing density functional theory properties for 159,000 mononuclear transition metal complexes in multiple spin states and formal charges, derived from the Cambridge Structural Database with experimental heavy-atom coordinates preserved during optimization and single-point energies evaluated at the PBE0/def2-TZVP level, along with a scheme for metal-spin-dependent atomization energies.

What carries the argument

The iterative procedure for confidently assigning overall TMC charge, combined with preservation of experimental heavy-atom coordinates during optimization and PBE0 single-point energy calculations on structures in compatible spin states.

Load-bearing premise

The iterative procedure confidently assigns overall TMC charge and that preserving experimental heavy-atom coordinates during optimization yields reliable properties across the diverse set of complexes and spin states.

What would settle it

Finding a substantial number of complexes where the assigned charges conflict with independently determined experimental charges or where fully relaxed geometries produce properties that differ enough from the fixed-coordinate results to change chemical conclusions.

Figures

Figures reproduced from arXiv: 2604.07623 by Aaron G. Garrison, Christopher J. Stein, Heather J. Kulik, Jacob W. Toney, Roland G. St. Michel, Tatiana Nikolaeva.

**Figure 1.** Figure 1: (a) Stacked histogram (i.e., grouped by converged and failed) of calculations by net complex charge. (b) Stacked histogram (i.e., grouped by converged and failed) of calculations by molecular weight (in amu). (c) Principal component analysis in metal-centered depth-two revised autocorrelations104 of the full dataset grouped by period (i.e., 3d, 4d, or 5d metals). We next evaluated the properties of converg… view at source ↗

**Figure 2.** Figure 2: Molecular weight (amu) vs. HOMO-LUMO gap (eV, left) and center of mass dipole moment (Debye, right). Marginal 1D histograms are shown at top and right for each of the two quantities. The data is colored by KDE density in 100 bins for each dimension with a shared color bar ranging from 10-7 (dark purple) to 0.0009 (yellow) density. BOS-TMC is not the first dataset of experimentally characterized structures … view at source ↗

**Figure 3.** Figure 3: Effect of sequential addition of data on mean (top), max (middle), and min (bottom) of the energetic gap (eV, left), dipole moment magnitude (Debye, middle), and Löwdin metal charge (right) starting from a set of TMCs in common with tmQMg and adding ca. 30k |q| > 1 complexes last. The shaded regions correspond to the standard deviation (2x for the mean plot and 1x for max. and min.) observed when the proce… view at source ↗

**Figure 4.** Figure 4: Distribution of ground state spins for transition metal complexes (left), grouped by the identity of the metal center. All transition metals omitted have >98% low-spin ground states. Distribution of vertical intermediate–low (top right) and high–low (bottom right) spin-splitting energies for both all transition metal complexes and for mid-row 3d metals. Only transition metal complexes which have converged … view at source ↗

**Figure 5.** Figure 5: Parity between the ground state and low-spin state for the energetic HOMO-LUMO gap (left), magnitude of the dipole moment (middle), and metal Löwdin charge (right). Representative structures with large deviations from parity are shown. Only transition metal complexes which have converged and are not outliers in any property, for every spin state modeled for their corresponding metal and oxidation state, ar… view at source ↗

**Figure 6.** Figure 6: Atomization energy statistics for BOS-TMC. The top-left panel shows the joint distribution of atomization energy and molecular weight for low-spin (singlets or doublets) structures as a log-scaled 2D density map with marginal histograms and a linear best-fit trend (dashed line). The bottom-left panel compares the distribution of the relative atomization energies with respect to molecular weight (eV mol g-1… view at source ↗

**Figure 7.** Figure 7: Distributions of properties in the manyDFA set compared to the full distribution in BOSTMC. Principal component analysis of depth-2 metal-centered revised autocorrelations of the manyDFA set plotted over the full BOS-TMC set (left). Comparison how frequently different metal centers occur in the sets (top right). Distribution of molecular weights in the manyDFA set overlaid on the distribution in the full … view at source ↗

**Figure 8.** Figure 8: Principal component analysis of depth-2, metal-centered revised autocorrelations in the manyDFA set, colored and sized by the standard deviation in the intermediate-low vertical spinsplitting energies among 12 functionals (left). Only structures which converged and were not spin contaminated for all 12 functionals and were not outliers for any property for any functional are included. Selected structures … view at source ↗

read the original abstract

We present the Boston Open-Shell Transition Metal Complex (BOS-TMC) dataset, a set of density functional theory (DFT) properties for 159k experimentally characterized mononuclear transition metal complexes (TMCs) in multiple spin states with a range of formal charges derived from the Cambridge Structural Database (CSD). To curate this set, we carried out an iterative procedure to confidently assign overall TMC charge. From this information, we then obtained properties in up to three spin states, i.e., low-, intermediate-, and high-spin for 3d metals and low- and intermediate-spin for 4d and 5d metals, depending on compatibility with the metal electron configuration, for a total of 343.8k TMC/spin combinations. At odds with prior sets, we preserved experimental heavy-atom coordinates in these structures during optimization. We report all properties using PBE0/def2-TZVP single-point energies on these structures. We introduce a scheme for computing metal-spin-dependent atomization energies, which we report for each TMC. Alongside electronic energies, we report up to seven additional properties including: HOMO, LUMO, HOMO-LUMO gap, atomic partial charges, dipole moments, atomization energies, and spin-splitting energies for a total of over 2.9M TMC-associated properties. For a representative subset of over 10k complexes chosen based on size, we evaluate the sensitivity of computed properties to exchange-correlation (xc) functional choice from a set of twelve xcs spanning rungs of "Jacob's ladder", highlighting hotspots of TMC space that have the greatest uncertainty. In comparison to prior transition-metal datasets, BOS-TMC is both larger and more diverse in terms of charge and spin configurations and, as a result, more diverse in its range of properties. This dataset is expected to provide a high-fidelity foundation for machine-learning model development, DFT benchmarking, and exploration.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

BOS-TMC supplies a bigger, more charge- and spin-diverse TMC dataset than prior work but leaves the charge-assignment success rate and fixed-geometry effects unquantified.

read the letter

The main takeaway is that this paper releases DFT properties for 159k experimental transition metal complexes across multiple formal charges and spin states, for a total of 343k combinations. They start from CSD structures, apply an iterative charge assignment, compute PBE0/def2-TZVP single-point energies while holding heavy-atom coordinates fixed, and add a metal-spin-dependent atomization energy scheme plus several other properties. On a 10k subset they also scan twelve functionals to flag areas of high sensitivity. That scale and the explicit inclusion of charge and spin variety are the concrete advances over earlier TMC collections. Preserving experimental coordinates is a deliberate choice that avoids some optimization artifacts but also limits how far the structures sit from DFT minima. The functional sensitivity check is a useful addition for users who care about method dependence. The curation workflow is described at a high level and the data release itself would be valuable for ML training or DFT benchmarking in inorganic chemistry. The soft spots sit in the validation. The abstract gives no success statistics or cross-checks for the iterative charge assignment, so it is hard to judge error rates on particular metal-ligand combinations. Single-point energies on unrelaxed experimental geometries are reported without any comparison to fully optimized structures, which leaves open how much the atomization energies, gaps, and spin splittings shift under relaxation. No direct error bars versus experiment appear either. These gaps are real but not fatal for a data paper; they mainly mean downstream users will need to run their own spot checks. The work is aimed at people who build models on TMC properties or benchmark DFT methods for open-shell systems. A reader who needs volume and diversity in charge and spin will find usable material here even before the validation details are tightened. It is worth sending to peer review because the dataset scale is large enough to matter and the authors have done the heavy lifting on curation; referees can push for the missing quantitative checks on charge assignment and geometry effects without rejecting the core contribution.

Referee Report

2 major / 2 minor

Summary. The manuscript presents the BOS-TMC dataset of DFT properties for 159k experimentally characterized mononuclear transition metal complexes (TMCs) drawn from the CSD, spanning multiple formal charges and spin states (low-, intermediate-, and high-spin where compatible) for a total of 343.8k TMC/spin combinations. The workflow uses an iterative charge-assignment procedure, preserves experimental heavy-atom coordinates during optimization, performs PBE0/def2-TZVP single-point calculations, introduces metal-spin-dependent atomization energies, and reports additional properties (HOMO, LUMO, gaps, partial charges, dipoles, spin splittings) totaling over 2.9M entries. A 10k-subset sensitivity analysis to twelve xc functionals is included, with the dataset positioned as larger and more diverse than prior TMC collections for ML, benchmarking, and exploration.

Significance. If the curation steps prove reliable, the dataset would provide a substantial, experimentally anchored resource for open-shell transition-metal chemistry. Its scale, coverage of charge/spin diversity, metal-spin-dependent atomization energies, and explicit functional-sensitivity study on a representative subset address documented gaps in existing TMC data collections and would support improved ML model training and DFT benchmarking in this challenging domain.

major comments (2)

[Methods (iterative charge assignment)] The iterative charge-assignment procedure is load-bearing for selecting valid spin states and generating the 343.8k TMC/spin entries, yet the manuscript reports no quantitative validation metrics (success rate, agreement with literature charges for benchmark subsets, or error rates stratified by metal or ligand class).
[Computational details and results (geometry handling)] Preservation of experimental heavy-atom coordinates is presented as ensuring high fidelity, but no comparison of key properties (atomization energies, spin splittings, HOMO-LUMO gaps) between fixed-geometry single points and fully DFT-optimized structures is provided, leaving the reliability of the reported values for the full set unquantified.

minor comments (2)

[Abstract] Clarify whether atomization energies are included in the 'up to seven additional properties' count or reported separately, and ensure consistent terminology between abstract and main text.
[Introduction] Add explicit numerical comparisons (size, charge/spin coverage, property ranges) to the prior TMC datasets referenced in the introduction to substantiate the 'larger and more diverse' claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript on the BOS-TMC dataset. We address each major comment below, indicating planned revisions where appropriate.

read point-by-point responses

Referee: The iterative charge-assignment procedure is load-bearing for selecting valid spin states and generating the 343.8k TMC/spin entries, yet the manuscript reports no quantitative validation metrics (success rate, agreement with literature charges for benchmark subsets, or error rates stratified by metal or ligand class).

Authors: We agree that explicit quantitative validation of the iterative charge-assignment procedure would strengthen the manuscript. The procedure applies standard electron-counting rules with iterative refinement to ensure consistency, but we acknowledge the absence of reported success rates or stratified comparisons in the original text. In the revised version, we will add a new subsection to the Methods section reporting the overall success rate, agreement with a manually verified benchmark subset of 500 complexes drawn from the literature, and error rates stratified by metal and ligand class. This addition will directly quantify the reliability of the curation step. revision: yes
Referee: Preservation of experimental heavy-atom coordinates is presented as ensuring high fidelity, but no comparison of key properties (atomization energies, spin splittings, HOMO-LUMO gaps) between fixed-geometry single points and fully DFT-optimized structures is provided, leaving the reliability of the reported values for the full set unquantified.

Authors: We thank the referee for this point. Preserving experimental heavy-atom coordinates was a deliberate choice to maintain fidelity to measured structures rather than allowing unconstrained optimization, which can introduce significant deviations in open-shell TMCs. To address the request for quantification, we will add an analysis in the revised manuscript comparing the specified properties (atomization energies, spin splittings, and HOMO-LUMO gaps) on a diverse representative subset of 5,000 complexes between the fixed-geometry single-point calculations and fully optimized structures. This will be reported in the Computational Details section. A full-set comparison is not feasible due to computational cost, but the subset will be selected to reflect the diversity of the dataset. revision: partial

Circularity Check

0 steps flagged

No circularity: standard data curation from external structures

full rationale

This is a dataset-generation paper that applies standard DFT (PBE0/def2-TZVP single-points) to experimentally determined CSD structures after an iterative charge-assignment curation step. No derivations, predictions, or first-principles results are claimed that reduce to fitted parameters or self-citations by construction. The charge-assignment procedure and fixed-geometry choice are methodological choices whose validity is external to any internal equation; they do not create a self-definitional loop or rename a fitted input as a prediction. All reported quantities (energies, gaps, atomization energies, etc.) are direct outputs of the chosen electronic-structure method on the curated inputs. No load-bearing self-citation chains or uniqueness theorems are invoked. The paper is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is an empirical dataset curation paper; it introduces no mathematical free parameters, axioms, or invented physical entities. All computations rely on standard DFT protocols and experimental input structures from the CSD.

pith-pipeline@v0.9.0 · 5691 in / 1134 out tokens · 68184 ms · 2026-05-10T16:53:21.412222+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We present the Boston Open-Shell Transition Metal Complex (BOS-TMC) dataset, a set of density functional theory (DFT) properties for 159k experimentally characterized mononuclear transition metal complexes (TMCs) in multiple spin states with a range of formal charges derived from the Cambridge Structural Database (CSD).
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce a scheme for computing metal-spin-dependent atomization energies, which we report for each TMC.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

2 extracted references · 2 canonical work pages

[1]

Calcs which completed but the graph changed: 1,026 a

All attempted TeraChem calcs (total//unique): 162,109//128,357 2. Calcs which completed but the graph changed: 1,026 a. 22 of these were problematic (i.e., hydrogen atoms flew away from the complex, identified by any hydrogen atom being over 1.5 Å away from the nearest heavy atom) and marked as failures b. All others (1,004) were kept in (see explanation ...

work page
[2]

weakly charged

In Psi4, attempt to converge the DFA with def2-TZVP starting from a functional that is not PBE0. a. In the case of semilocal DFAs, take a converged PBE result and initialize the calculations for other semilocal DFAs from PBE. i. If the PBE calculation did not converge, successively converge PBE+x% calculations, where x is the amount of Hartree-Fock exchan...

work page 2059

[1] [1]

Calcs which completed but the graph changed: 1,026 a

All attempted TeraChem calcs (total//unique): 162,109//128,357 2. Calcs which completed but the graph changed: 1,026 a. 22 of these were problematic (i.e., hydrogen atoms flew away from the complex, identified by any hydrogen atom being over 1.5 Å away from the nearest heavy atom) and marked as failures b. All others (1,004) were kept in (see explanation ...

work page

[2] [2]

weakly charged

In Psi4, attempt to converge the DFA with def2-TZVP starting from a functional that is not PBE0. a. In the case of semilocal DFAs, take a converged PBE result and initialize the calculations for other semilocal DFAs from PBE. i. If the PBE calculation did not converge, successively converge PBE+x% calculations, where x is the amount of Hartree-Fock exchan...

work page 2059