⚛️ FLUXMATERIA — CHEMISTRY

SN1 / SN2 / E1 / E2 / E1cb,
at ~1 ms per prediction

Mechanism Discovery classifies substitution and elimination pathways with 100% accuracy on the 336-case published benchmark, returns the activation barrier at 7.4 kJ/mol MAE, and models temperature crossovers, rearrangements, and neighbouring-group participation — in about a millisecond per prediction.

5 mechanisms 16 solvents 7 leaving groups Rearrangements & NGP No ML
Open mechanism benchmark Read the textbook case study
100%
Mechanism accuracy on 336 published cases
7.4 kJ/mol
Activation barrier MAE vs experiment
~1 ms
Per-prediction runtime (vs hours for DFT)
106×
Speed-up vs conventional quantum chemistry
0
Fitted parameters
The breakthrough

DFT-grade accuracy at rule-based speed.

Rule-based tools sit around 85% mechanism accuracy, have no barrier estimate, and don’t handle E1cb or rearrangements. DFT hits the accuracy but takes hours per transition-state search and needs an expert to set up. Mechanism Discovery lands 100% mechanism classification and 7.4 kJ/mol barrier MAE on the 336-case benchmark, in ~1 ms per prediction — fast enough for interactive exploration, accurate enough to ship.

What Mechanism Discovery does

Every capability below is validated on the 336-case published benchmark.

🎯

5-way classification

SN1, SN2, E1, E2, E1cb — all five mechanisms from a single call, with Boltzmann-weighted competition between them.

⚡

DFT-grade barriers

7.4 kJ/mol MAE on activation energies — inside the DFT accuracy window — in ~1 ms instead of hours.

🌡️

Temperature crossovers

Predicts SN1→E1 and SN2→E2 crossovers as temperature changes, including the crossover temperature itself.

🔀

Rearrangement detection

Hydride shifts, methyl shifts, and ring expansions flagged automatically under SN1 / E1 conditions via graph traversal.

🔗

Neighbouring-group participation

Anchimeric assistance from S, N, and phenyl groups detected with rate-enhancement estimates (episulfonium, aziridinium, phenonium).

💧

16 solvents

Water, ethanol, DMSO, DMF, acetone, TFE, t-BuOH, methanol, THF — protic, aprotic, and non-polar solvents all parameterised.

🧬

Full interpretability

Every kJ/mol of the barrier traces back to a physical factor — nucleophilicity, steric strain, solvent stabilisation, leaving-group ability. Not a black box.

🔬

Wide substrate scope

7 substrate classes (methyl, primary, secondary, tertiary, allylic, benzylic, vinyl) × 7 leaving groups (I, Br, Cl, F, OTs, OMs, OTf).

How a prediction is built

From reactant + nucleophile + solvent + temperature to mechanism + barrier in one call.

1

Parse inputs

Substrate SMILES, nucleophile / base, leaving group, solvent, temperature. Continuous 1–10 nucleophile / base strength scale.

2

Score all 5 mechanisms

SN1 / SN2 / E1 / E2 / E1cb activation barriers computed in parallel — deterministic physics, no optimisation loop.

3

Apply special effects

Graph traversal flags rearrangements; NGP neighbours are detected and rate-enhanced; EWG activation is applied where relevant.

4

Boltzmann-weight

Dominant mechanism at the chosen temperature returned with per-pathway probabilities; crossover temperatures reported.

5

Return & interpret

Mechanism label + barrier + per-factor contribution breakdown. Every kJ/mol attributed to a physical driver.

Why you can trust it

Validated on the 336-case published benchmark — every case has an experimental mechanism and barrier assignment.

100%
Mechanism classification accuracy on all 336 published cases — every substrate, every leaving group, every solvent.
7.4 kJ/mol
Activation-barrier MAE — inside the DFT (B3LYP) accuracy window of 4–8 kJ/mol, at 106× the speed.
336
Published experimental cases in the benchmark — not a sampled test split.
16 · 7 · 7
Solvents · substrate classes · leaving groups parameterised across the model.
~1 ms
Per-prediction runtime on a single CPU. Batches of 10k reactions in seconds.
0
Fitted parameters. Every prediction re-runnable bit-for-bit.

How FluxMateria compares

Head-to-head against the standard ways of answering “is this SN1 or SN2?”.

MetricFluxMateriaDFT (B3LYP)Rule-based toolsML classifiers
Mechanism accuracy (336 cases)100%85–92%~85%70–90%
Barrier MAE7.4 kJ/mol4–8 kJ/molNot providedNot provided
Time per prediction~1 ms1–4 hours< 1 ms1–100 ms
E1cb, rearrangements, NGPBuilt-inManual setupNoDepends
Temperature crossoversBuilt-inRe-run per TNoNo
Training dataNoneNoneRule setThousands of labelled
InterpretabilitykJ/mol per factorMethod traceRule traceOpaque
DeterministicYesYesYesSeed-dependent

The key insight: Rule-based tools are fast but miss barriers and borderline cases. DFT is accurate but too slow for screening campaigns. ML classifiers need thousands of labelled reactions and can’t predict unseen chemistry confidently. Mechanism Discovery gets both accuracy and speed, and every kJ/mol is traceable to a physical factor. See the full 336-case benchmark →

Where Mechanism Discovery wins

Reaction-design workflows where DFT is too slow and rules are too coarse.

Use case 1

Reaction optimisation

Pick the solvent + temperature that tips a borderline substrate from E1 to SN2 — see the crossover temperature explicitly.

Use case 2

Process-chemistry triage

Before committing a synthesis route, predict the dominant mechanism and activation barrier for each step in seconds, not hours.

Use case 3

Rearrangement risk check

Proposed a tertiary-substrate SN1? The engine flags hydride / methyl / ring-expansion risks automatically before the wet lab sees the substrate.

Use case 4

Teaching & intuition

Per-factor kJ/mol attribution makes “why is this E2 not SN2?” a concrete answer, not a hand-wave. Perfect for graduate courses.

Use case 5

NGP-aided rate design

Confirm whether a proposed episulfonium / phenonium assistance is feasible, and quantify the rate enhancement before running kinetics assays.

Use case 6

Audit & reproducibility

Zero fitted parameters, deterministic output. Reviewers re-running the same substrate + conditions get the same mechanism + barrier every time.

Mechanism Discovery in the product

Real captures from the live application. Click any image to zoom.

Mechanism Discovery input panel with substrate, nucleophile, leaving group, solvent, temperature
InputSubstrate, nucleophile / base, leaving group, solvent, and temperature — everything the physics model needs in one form.
Mechanism result with dominant pathway, barrier, and Boltzmann-weighted competition
ResultDominant mechanism + activation barrier + Boltzmann-weighted pathway probabilities at the chosen temperature.
Temperature crossover chart showing SN1/E1 vs SN2/E2 dominance vs temperature
Temperature crossoverSN1 ⇄ E1 and SN2 ⇄ E2 crossover points visualised across the 0–180 °C validated range.
Per-factor kJ/mol attribution bar chart
Factor attributionEvery kJ/mol of the barrier attributed to nucleophilicity, steric strain, solvent stabilisation, or leaving-group ability — not a black box.

Scope & Limitations

Strengths

  • 100% mechanism classification on the 336-case published benchmark — exhaustive, not sampled.
  • 7.4 kJ/mol activation-barrier MAE, inside the DFT (B3LYP) accuracy window, at 106× the speed.
  • Temperature crossovers, rearrangements, NGP, and 16-solvent coverage built in.
  • Per-factor kJ/mol attribution — every prediction explains itself.
  • Deterministic: same inputs, same mechanism and barrier, every run.

Known limitations

  • Scope is substitution and elimination (5 mechanisms). Other reaction classes live in the MechanismOS module.
  • Solvent coverage is the 16 parameterised choices; unusual ionic-liquid or molten-salt media fall back on the closest analogue.
  • Temperature validated 0–180 °C; extreme regimes (cryogenic, flash-pyrolysis) sit outside the validation envelope.
  • For transition-metal-catalysed reactions, pair Mechanism Discovery with Catalyst Discovery for the metal-cycle context.

Run a mechanism prediction

Pilot access includes Mechanism Discovery, MechanismOS, Synthesis Planning, Spectroscopy, and a Workspace seat for audit-ready runs.

Request Pilot Access →