Skip to content
PaperFren

Research method

Density Functional Theory (DFT)

Density functional theory approximates the electronic energy of a molecule or extended solid from its electron density. Chemists use it to rank isomers, estimate barriers, assign redox character, and generate descriptors that a statistical model can then learn. The number that comes out is a computed energy or geometry at a chosen functional and basis, not a measured rate, and 'chemical accuracy' (about 1 kcal mol⁻¹) is a target some hybrid workflows hit only after training on DFT itself.

Researchers reach for DFT when an experiment cannot easily see a transition state, a spin-state gap, or which ligand orbital is being reduced. It answers 'which structure or path is lower at this level of theory?' Its main limitation in this library is domain: octahedral first-row complexes, SNAr barriers, Au(100)–water, or a gas-phase heme model do not automatically transfer to a different metal, solvent, or applied potential.

Evidence

What the evidence shows

Drawn from 28 studies in this library. Each finding starts with a plain-language takeaway, then the denser detail. Supports means evidence for a finding; Challenges means evidence against a stated position; Qualifies marks scope with a short note on each study’s contribution. Challenged positions are labeled — they are not findings.

  • ML trained on DFT descriptors can beat raw DFT on a held-out SNAr set and still depend on being able to compute the transition state. A Gaussian-process model reached R² = 0.93 and MAE = 0.77 kcal mol⁻¹ versus raw DFT MAE 2.93; selectivity top-1 accuracy was 86%. Acid/base catalysis and other reaction classes were not in the model; TS geometry remains the bottleneck.

    1 study
    1. 1ML plus transition states predict SNAr barriers
  • 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.

    1 study
    1. 1Neural nets predict TM spin states and bonds
  • DFT is often a supporting assignment tool, not the primary result. Periodic-DFT on a Zn phenanthroline–maleate crystal gave a 3.45 eV gap; gas-phase heme-model pathways accompanied FT-ICR epoxidation rates with estimated ±30% rate error; ligand-centered versus metal-centered redox in cobalt PY4/PY3PZ complexes was parsed with DFT after electrocatalysis.

    3 studies
    1. 1A new Zn phenanthroline–maleate crystal
    2. 2Gas-phase rates of heme-model olefin epoxidation
    3. 3Pyrazine reservoirs that lower cobalt HER overpotential

    Study comparison

    StudyRoleDesignNPopulationOutcome
    A new Zn phenanthroline–maleate crystal2026SupportsOtherCrystal structure, periodic DFT, and secondary MIC assays of a Zn–phenanthroline–maleate complexCoordination-chemistry characterization — no cohort NZn(II) phenanthroline/maleate coordination compoundStructure, electronics, and weak antibacterial activity vs S. mutans
    Gas-phase rates of heme-model olefin epoxidation2015SupportsOtherGas-phase FT-ICR bimolecular rate constants for [FeIV(O)(TPFPP+·)]+ with olefins plus DFTPhysical-organic kinetics panel of olefins — not a sample-N studyGas-phase iron-oxo porphyrin cation and olefin substratesEpoxidation rate constants and efficiency versus olefin ionization potential
    Pyrazine reservoirs that lower cobalt HER overpotential2015SupportsOtherCobalt PY4/PY3PZ complexes for aqueous electrocatalytic and photoredox H2 evolution with DFTMolecular catalysis study — no sample NCobalt redox-active ligand complexes in aqueous catalytic assaysOverpotential and H2 evolution as a function of pendant pyrazine redox reservoirs
  • Learned potentials trained on DFT can extend the timescale of a mechanism study. A PaiNN ensemble on PBE-D3 Au(100)–water data plus metadynamics gave an associative ORR path with an O₂-to-hydroxyl barrier of about 0.3 eV — a landscape from the neural-network potential, not a measured Tafel slope, and without explicit cations or applied potential.

    1 study
    1. 1NN potentials map gold–water ORR paths

Open questions

Tensions and limits

Some items are genuine disagreements on the same question. Others mark different assays, populations, or outcomes — limits on how far one study travels — not a forced fight between papers.

  • Scope / different questions

    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.

    4 studies
    1. 1ML plus transition states predict SNAr barriers
    2. 2Neural nets predict TM spin states and bonds
    3. 3NN potentials map gold–water ORR paths
    4. 4Pyrazine reservoirs that lower cobalt HER overpotential

    Study comparison

    StudyRoleDesignNPopulationOutcome
    ML plus transition states predict SNAr barriers2021SupportsComputational / modellingDFT/physical-organic descriptors with GPR for SNAr barriers and selectivityML on reaction barriers/selectivity; learning curves emphasize 50–150 training samples — no single cohort NNucleophilic aromatic substitution reactions (including patent-selectivity cases)Predicted activation free energies and regio-/chemoselectivity accuracy
    Neural nets predict TM spin states and bonds2017SupportsComputational / modellingANN trained on DFT octahedral Cr–Ni complexes to predict spin states and bond lengthsN=2690 · 2,690 DFT geometry optimizations after filtering spin-contaminated casesHomoleptic and heteroleptic octahedral transition-metal complexesPredicted spin-state energetics and metal–ligand bond lengths
    NN potentials map gold–water ORR paths2023SupportsComputational / modellingPaiNN active-learning NNPs enabling path-CV metadynamics of ORR on Au(100)–water2.5 ns production metadynamics after AIMD active learning — no sample NAu(100)–water models with O2/OH adsorbatesAssociative ORR pathway and ~0.3 eV O2-to-hydroxyl barrier
    Pyrazine reservoirs that lower cobalt HER overpotential2015SupportsOtherCobalt PY4/PY3PZ complexes for aqueous electrocatalytic and photoredox H2 evolution with DFTMolecular catalysis study — no sample NCobalt redox-active ligand complexes in aqueous catalytic assaysOverpotential and H2 evolution as a function of pendant pyrazine redox reservoirs
  • Scope / different questions

    Gas-phase ions, aqueous electrocatalysis, and a crystal gap are different DFT worlds. Naked [Feᴵⱽ(O)(porphyrin)]⁺ epoxidation in FT-ICR is not P450 Compound I in water; the Zn crystal gap was not tested in a device; cobalt HER overpotential is an electrochemical measurement with DFT as interpretation.

    3 studies
    1. 1Gas-phase rates of heme-model olefin epoxidation
    2. 2A new Zn phenanthroline–maleate crystal
    3. 3Pyrazine reservoirs that lower cobalt HER overpotential

    Study comparison

    StudyRoleDesignNPopulationOutcome
    Gas-phase rates of heme-model olefin epoxidation2015SupportsOtherGas-phase FT-ICR bimolecular rate constants for [FeIV(O)(TPFPP+·)]+ with olefins plus DFTPhysical-organic kinetics panel of olefins — not a sample-N studyGas-phase iron-oxo porphyrin cation and olefin substratesEpoxidation rate constants and efficiency versus olefin ionization potential
    A new Zn phenanthroline–maleate crystal2026SupportsOtherCrystal structure, periodic DFT, and secondary MIC assays of a Zn–phenanthroline–maleate complexCoordination-chemistry characterization — no cohort NZn(II) phenanthroline/maleate coordination compoundStructure, electronics, and weak antibacterial activity vs S. mutans
    Pyrazine reservoirs that lower cobalt HER overpotential2015SupportsOtherCobalt PY4/PY3PZ complexes for aqueous electrocatalytic and photoredox H2 evolution with DFTMolecular catalysis study — no sample NCobalt redox-active ligand complexes in aqueous catalytic assaysOverpotential and H2 evolution as a function of pendant pyrazine redox reservoirs

Common misconceptions

  • If a model reaches 0.77 kcal mol⁻¹ MAE, DFT has been replaced and no longer needs to be computed.

    The SNAr GP still needs reliable transition-state geometries, which the authors call the bottleneck, and it does not cover acid/base catalysis or other reaction classes.

    1. 1ML plus transition states predict SNAr barriers
  • A neural net that is 98% correct on spin states will be 98% correct on any transition-metal complex.

    Training was octahedral first-row Cr–Ni with a specified ligand set. CSD transfer was poorer, and cyclams and early metals were ~30 kcal mol⁻¹ outliers. The target is the DFT functional used in training, not experiment.

    1. 1Neural nets predict TM spin states and bonds
  • A 0.3 eV ORR barrier from metadynamics is a measured activation energy on a gold electrode in a real cell.

    It comes from a neural-network potential trained on PBE-D3 Au(100)–water without explicit cations or applied potential; facet dependence is left for later work.

    1. 1NN potentials map gold–water ORR paths

Exam-style questions

Short-answer questions that ask you to explain or compare, not recall.

The SNAr paper reports raw DFT MAE 2.93 kcal mol⁻¹ and GP MAE 0.77. What is the student supposed to conclude about 'DFT accuracy'?

On that set, uncorrected DFT barrier errors are larger than the hybrid model's errors. The improvement is a statistical correction using physical-organic and TS descriptors, not a new functional, and it still requires computing the TS.

Why is 528/538 correct ground states on a test set compatible with 10 kcal mol⁻¹ MUE on CSD complexes?

The test set is drawn from the same octahedral Cr–Ni design as training. CSD structures include chemistries outside that box; a Euclidean reliability cutoff halves the CSD MUE, which is an admission that out-of-domain complexes should not be trusted at the test-set figure.

A colleague treats the gold ORR 0.3 eV barrier as proof of the experimental mechanism in alkaline fuel cells. What is missing from the model?

Explicit electrolyte cations, applied potential, and other facets. The path is associative *OOH then two *OH in pure water on Au(100) at 350 K in a learned PBE-D3 potential.

When should DFT be cited as the result versus as an interpretation of an experiment in this library?

As the result: SNAr barrier models, TM spin-state networks, NNP metadynamics. As interpretation: Zn crystal electronics, heme-model paths next to FT-ICR rates, ligand versus metal redox after measuring HER overpotential. Mixing those citations makes a computed gap look like a measured MIC or a Faradaic efficiency.

The studies

28 studies in this library bear on Density Functional Theory (DFT), ordered by citations. The first 8 are shown.

  • Single-atom Co–N–C catalyzes nitroarene azo coupling

    A self-supporting Co–N–C single-atom catalyst with a CoN4C8-1-2O2 site hydrogenatively couples nitroarenes to azo compounds.

    Chemical science · 2016 · 202 citations

  • Halogen bonds are not purely electrostatic

    Relativistic DFT shows halogen and hydrogen bonds share HOMO–LUMO covalency; halogen bonds have weaker electrostatics but stronger orbital mixing.

    ChemistryOpen · 2012 · 147 citations

  • DFT: graphene SACs break CO2 scaling relations

    Single metal atoms in graphene vacancies are predicted to reduce CO2 to methanol or methane at milder limiting potentials than metal terraces by weakening *CO more than *CHO.

    Chemical science · 2017 · 143 citations

  • Charged amino acids absorb past 250 nm

    A 67-residue protein with no aromatic side chains still absorbs from 250 to 800 nm because Lys/Glu charge-transfer transitions create Protein Charge Transfer Spectra.

    Chemical science · 2017 · 116 citations

  • FeMn single-atom nanozyme dual-readout HER2 test

    An etched FeMn–N–C nanozyme boosts peroxidase-like activity 2.03-fold and reports HER2 by electrochemistry and 808 nm photothermal readout.

    Analytical chemistry · 2024 · 115 citations

  • Neural nets predict TM spin states and bonds

    Graph-based ANNs trained on 2690 DFT geometries of Cr–Ni octahedral complexes predict spin splitting and metal–ligand distances without 3D coordinates.

    Chemical science · 2017 · 114 citations

  • Unsymmetrical ligands still make one Pd2L4 cage isomer

    Steric and geometric ligand design plus solvent polarity steer unsymmetrical dipyridyl ligands into a single cis or trans Pd2L4 cage rather than a statistical isomer mix.

    Chemical science · 2019 · 94 citations

  • Bismuth oxyhalide films as photoelectrodes

    AACVD BiOX films show halide-tuned bandgaps; untreated BiOBr gives about 0.38 mA cm−2 photoanodic current without a sacrificial donor.

    Chemical science · 2016 · 90 citations

Show 20 more studies

Learn alongside