⚛️ FLUXMATERIA — CHEMISTRY

Eight spectroscopy types,
one physics engine

Predict UV-Vis, IR, NMR, Raman, CD, EPR, Emission, and X-ray spectra from a SMILES — with peak assignments and confidence indicators. UV-Vis at 6.2% error, IR inside 1%, NMR at 0.3–0.5 ppm MAE. ~25 ms per prediction.

UV-Vis IR / Raman NMR CD / EPR / X-ray Peak assignments

Open spectroscopy benchmark Request Pilot Access

Spectroscopy types from one engine

6.2%

UV-Vis mean error on 50 molecules

< 1%

IR mean error on 32 NIST molecules

0.3–0.5

NMR MAE (ppm) on 10 SDBS molecules

~25 ms

Per prediction on a single CPU

The breakthrough

Predict the spectrum before the synthesis.

UV-Vis, IR, NMR, Raman, CD, EPR, Emission, and X-ray — eight spectroscopy types from one physics engine, with peak assignments and confidence indicators per feature. ~25 ms per prediction means you’ve checked “will this compound actually give the IR I expect?” before you’ve booked the bench time. Every peak traces to a molecular feature; nothing is opaque.

Eight spectroscopy types, one call

Same engine, same interpretable output surface, every spectrum.

🌈

UV-Vis

Electronic transitions, absorption maxima, extinction coefficients — 6.2% mean error on the 50-molecule validation set.

📊

IR

Vibrational mode analysis with peak assignments and functional-group ID. < 1% mean error on 32 NIST molecules.

🧲

NMR

¹H and ¹³C shift predictions. 0.3–0.5 ppm MAE on the 10-molecule SDBS validation set.

🔬

Raman

Raman-active modes, Stokes / anti-Stokes ratios, SERS enhancement factors — production-ready alongside IR.

🔄

Circular dichroism

Chiroptical spectra for stereochemistry analysis. Cotton effects and secondary-structure signatures.

⚛️

EPR

Electron paramagnetic resonance for radical and metal-centre characterisation. g-factors and hyperfine coupling.

💡

Emission

Fluorescence and phosphorescence with Stokes shifts and quantum-yield estimates.

☢️

X-ray (XAS / XES)

Element-specific electronic structure — edge energies, pre-edge features — for metal-centre and coordination chemistry.

How a spectrum is predicted

From SMILES to spectrum with assigned peaks in one call.

Input

SMILES or structure file. Single molecule or batch.

Select

Pick the spectroscopy types to compute — UV-Vis, IR, NMR, Raman, CD, EPR, Emission, X-ray, or any combination.

Compute

Physics engine evaluates transitions / modes / shifts for every selected technique. ~25 ms per spectrum.

Assign

Every peak labelled with its molecular feature + confidence tier — not a bare spectrum.

Compare & export

Overlay predicted vs experimental, export spectrum files (JCAMP-DX, JSON, CSV) and per-peak assignment tables.

Why you can trust it

Validated against the NIST, SDBS, and curated literature datasets — every production spectrum type has a benchmark.

6.2%

UV-Vis absorption-max error on 50 validated molecules — chromophores from aromatic dyes to carbonyls.

< 1%

IR vibrational-mode error on the 32-molecule NIST validation set. Every peak reproduces the correct functional-group region.

0.3–0.5 ppm

NMR chemical-shift MAE on 10 SDBS molecules — within typical experimental resolution.

Spectroscopy types exposed today. UV-Vis / IR / Raman / NMR production-validated; CD / EPR / Emission / X-ray shipping, actively benchmarked.

~25 ms

Per-prediction runtime on a single CPU. Batch several hundred molecules per second.

Fitted parameters. Every peak traces to a physical feature — re-runnable bit-for-bit.

How FluxMateria compares

Head-to-head against the standard ways to predict a spectrum.

Metric	FluxMateria	DFT (TDDFT / DFT-D)	ML surrogate	Experimental reference
UV-Vis abs-max error	6.2%	5–15%	10–30% (OOD)	Reference (exact)
IR mode error	< 1%	2–5%	Depends on set	Reference (exact)
NMR MAE (ppm)	0.3–0.5	0.3–1.0	Depends on set	Reference (exact)
Runtime per spectrum	~25 ms	Minutes to hours	ms–s	Measure time + cost
Techniques covered	8 from one engine	Per-method setup	One model per technique	Per-instrument
Peak assignments	Built-in + confidence	Manual	Usually not	Manual
Training data	None	None	Thousands per technique	Data is the tool
Out-of-distribution	Physics-grounded	Physics-grounded	Confidently wrong	N/A

The key insight: TDDFT / DFT predicts spectra but takes minutes to hours and needs per-technique setup. ML surrogates are fast within their training distribution and confidently wrong outside it. Spectroscopy gives you eight techniques from one engine with built-in peak assignments, at ~25 ms per prediction. See the full benchmark →

Where Spectroscopy wins

Characterisation workflows where the predicted spectrum changes the wet-lab plan.

Use case 1

Synthesis planning

Predict the spectrum you’ll want to see before you start the synthesis. Plan the characterisation strategy and set acceptance criteria.

Use case 2

Structure confirmation

Overlay predicted vs experimental IR / NMR to confirm product identity — with per-peak assignment + confidence for the review meeting.

Use case 3

Photophysical screening

Screen candidates for absorption ranges, emission colour, Stokes shift, and quantum-yield hints — in milliseconds, not days.

Use case 4

Impurity identification

Unknown peak in the experimental spectrum? Match it to predicted signatures of likely impurities or degradation products.

Use case 5

Metal-centre coordination

XAS pre-edge + EPR g-factors + emission colour all from one SMILES+ligand-set call — useful for bioinorganic and catalyst work.

Use case 6

Teaching & auditing

Every peak is a physical feature with a confidence tier. Graduate courses see the spectrum “open up” instead of being a black box.

Spectroscopy in the product

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

UV-Vis absorption spectrum with peak assignments — **UV-Vis**Predicted absorption spectrum with electronic-transition assignments and extinction coefficients.

IR spectrum with vibrational-mode assignments — IRVibrational-mode assignments with functional-group annotations — < 1% error on NIST validation.

NMR predicted chemical-shift table with per-carbon/per-proton confidence — **NMR**Predicted ¹H and ¹³C shifts with confidence tiers, 0.3–0.5 ppm MAE.

Overlay comparison of predicted vs experimental spectrum — **Predicted vs experimental**Overlay comparison with peak alignment and per-peak deviation — review-ready evidence for structure confirmation.

Scope & Limitations

Strengths

Eight spectroscopy types from one engine with peak assignments and confidence indicators per feature.
UV-Vis 6.2% / IR < 1% / NMR 0.3–0.5 ppm — within typical experimental resolution.
~25 ms per prediction — batches of several hundred molecules in a second.
Physics-based, so out-of-distribution chemistry is handled gracefully with confidence tiers, not silent failure.
Shares the same bond lengths / force constants the rest of the platform rides on — consistent numbers across the product.

Known limitations

Production-validated: UV-Vis, IR, Raman, NMR. Shipping but actively benchmarked: CD, EPR, Emission, X-ray.
Benchmarks target small molecules today; macromolecular and solid-state spectra live in the Materials module.
Explicit-solvent effects are handled via solvatochromic shifts; full explicit-solvent MD lives in the MD & Free Energy module.
Peak-assignment confidence is calibrated per-technique; treat low-confidence assignments as leads for manual review, not final labels.

Predict a spectrum

Pilot access includes Spectroscopy (all 8 types), the Chemistry bond engine, MechanismOS, Synthesis Planning, and a Workspace seat for audit-ready runs.

Request Pilot Access →

Eight spectroscopy types,one physics engine