← Benchmarks | Synthesis Planning

Synthesis Planning BENCHMARK

29 reaction-type barriers validated against experimental ranges. 200 specific reactions with 0.119% MAE. All barriers from closed-form FLUX expressions. Published methodology.

3.1%
MAE
Reaction-Type Barriers
29/29 within exp. range
29/29
Pass Rate
All within experimental range
200
Specific Reactions
0.119% MAE
82
Reagents
26 categories
<50ms
Per Plan
Single-threaded CPU

Methodology

How FLUX barrier theory derives activation energies from first principles

FLUX Barrier Theory

All activation barriers derive from a single base energy computed from FLUX vacuum geometry. Each reaction type applies a closed-form geometric modification to the base energy — no empirical fitting, no database lookups.

  • 29 reaction types, each with a distinct geometric factor
  • Barrier ordering emerges from reaction topology
  • Evans-Polanyi corrections for exo/endothermic reactions
  • All factors traceable to FLUX vacuum geometry

Evaluation Criteria

Barrier Accuracy

FLUX prediction must fall within published experimental range for each of the 29 reaction types.

Reproducibility

All 29 cases can be run in the FluxMateria app. Pilot participants can validate independently.

Full Results: 29 Reaction Types

All reaction-type barriers, predictions, and experimental ranges

Reaction FLUX (kJ/mol) Exp (kJ/mol) Error% Range Status
SN2 76.4 76.0 0.5% (65–85) OK
SN1 91.7 100.0 8.3% (85–115) OK
E2 84.0 90.0 6.6% (80–100) OK
E1 91.7 95.0 3.5% (85–115) OK
Aldol 59.1 60.0 1.4% (50–70) OK
Grignard 53.5 55.0 2.8% (45–65) OK
Wittig 68.8 60.0 14.7% (50–70) OK
Addition 59.2 60.0 1.3% (50–70) OK
Michael 64.9 55.0 18.1% (45–65) OK
Claisen 91.7 92.0 0.4% (80–105) OK
Oxidation 114.6 120.0 4.5% (100–140) OK
Reduction 68.8 65.0 5.8% (55–80) OK
Ester hydrolysis 57.3 57.0 0.5% (50–65) OK
Amide hydrolysis 80.3 80.0 0.3% (70–90) OK
Halogenation 53.5 55.0 2.8% (40–70) OK
SNAr 91.7 95.0 3.5% (85–105) OK
EAS 94.7 92.0 3.0% (85–100) OK
Cycloaddition 114.6 115.0 0.3% (105–125) OK
Cycloaddition (1 EWG) 76.4 75.0 1.9% (65–85) OK
Cycloaddition (2 EWG) 64.2 65.0 1.2% (55–75) OK
Radical addition 31.8 32.0 0.5% (25–40) OK
Radical chain 31.8 32.0 0.5% (25–40) OK
Proton transfer 11.8 12.0 1.9% (5–20) OK
Epoxide opening (acid) 66.3 65.0 2.0% (55–75) OK
Epoxide opening (base) 76.4 78.0 2.0% (68–88) OK
Beckmann 114.6 115.0 0.3% (100–130) OK
Hofmann 99.4 100.0 0.6% (85–115) OK
Friedel-Crafts alkylation 140.1 140.0 0.1% (120–160) OK
Friedel-Crafts acylation 121.0 121.0 0.0% (105–140) OK

MAE: 3.1% | Pass rate: 29/29 (100%)

Specific Activation Energies

200 individual reactions validated against experimental barriers

Specific Activation Energies: 200 Reactions

  • Total reactions: 200
  • MAE: 0.119%
  • Exact match (0.0% error): 72 reactions
  • High error (>5%): 0 reactions
  • H-abstraction subset (11 rxns): 0.034% MAE
  • All 200 barriers derive from closed-form FLUX geometric expressions. Every result reproducible and auditable.

Comparison with Other Methods

How FluxMateria synthesis planning compares to established approaches

Metric FluxMateria DFT (B3LYP) Reaction Databases ML Retrosynthesis
Barrier Accuracy 3.1% MAE 4–8 kJ/mol MAE Reference (exact) N/A
Coverage 29 types + 200 specific Unlimited (per-molecule) Limited to known rxns Limited to training set
Speed <50ms per plan Hours per barrier Instant lookup ~1 sec
Parameters Fitted 0 Many (functional) N/A (empirical) Millions
Generalization Any SMILES Any (slow) Known reactions only Training domain

Physical Consistency Tests

Verification that FLUX barriers obey known chemical ordering rules

  • All barriers deterministic and traceable ✓ 100%
  • Barrier order preserved for EWG-activated cycloadditions ✓ 100%
  • Radical barriers always below polar barriers ✓ 100%
  • Proton transfer near-barrierless ✓ 100%
  • H-abstraction Evans-Polanyi corrections from FLUX ✓ 100%

Scope & Limitations

Honest assessment of strengths and known gaps

Strengths

  • 21/29 types within 5% error
  • Proton transfer and radical barriers very accurate
  • Friedel-Crafts within 0.1%
  • Every result reproducible — no retraining required

Known Limitations

  • Wittig (14.7%) and Michael (18.1%) near boundary
  • Substrate-specific effects not captured in generic barriers
  • FGI route depth limited to single-step
  • 200-reaction database may not cover exotic transformations

References

Primary data sources for experimental validation

  1. F.A. Carey, R.J. Sundberg, Advanced Organic Chemistry, 5th ed., Springer, 2007.
  2. M.B. Smith, J. March, March's Advanced Organic Chemistry, 6th ed., Wiley, 2007.
  3. NIST Chemical Kinetics Database, National Institute of Standards and Technology.
  4. E.V. Anslyn, D.A. Dougherty, Modern Physical Organic Chemistry, University Science Books, 2006.

Run the validation yourself

Pilot participants get full access to validation scripts and datasets.

← Back to Module Request Access