The engine’s results are ours. The measured absorption maxima come from a public collection of experimental UV-vis spectra. The engine computes orbital energies, not a spectrum, so the gap is converted to a wavelength (1240 divided by the gap in eV) and compared with the measured maximum. That conversion on its own is far from the measurement, so the page also shows the result after a straight-line calibration, scored so that no molecule is in both the fit and the test.
What the comparison shows
The gap tracks the colour a molecule absorbs: its raw correlation with the measured maximum is 0.78. The raw gap wavelength averages 187 nm while the measured maxima average 433 nm, so it is not a usable wavelength prediction by itself (mean absolute error 245 nm). After a straight-line calibration it predicts absorption maxima to within 45 nm on average, a clear improvement on guessing the average (77 nm), and about two predictions in three fall within 50 nm. That accuracy is strongest where most of the data lies. Between 300 and 500 nm the error is 32 to 39 nm and 72% to 81% of predictions are within 50 nm. It falls off for long wavelengths: above 600 nm the error rises to 96 to 155 nm. The overall correlation comes mostly from the gap separating molecules that absorb in the UV from those that absorb in the visible. Within a narrow wavelength band the gap orders molecules only weakly, so use it to sort molecules by colour, not to resolve small shifts.Predicted against measured
Pick a wavelength region. The calibrated prediction is in blue; the raw gap wavelength is in grey and sits well below the dashed line because it is not calibrated.How the errors are distributed
Where it works and where it does not
How the comparison was made
Measurements and molecules
Measurements and molecules
The benchmark has 19,713 measured absorption maxima. We dropped 3 whose maximum lies outside 150 to 1000 nm, leaving 19,710 measurements of 10,932 molecules. A molecule can appear several times, in different solvents.
Calibration and scoring
Calibration and scoring
The gap in eV is converted to a wavelength as 1240 divided by the gap. A straight line from that wavelength to the measured maximum is fitted by least squares. To keep the scoring honest, the data are split into five folds so that all measurements of a molecule fall in the same fold, and each fold is predicted by a line fitted only on the other four. Fitted on everything, the line is roughly measured = 2.83 × (gap wavelength) − 97 nm.
The baseline
The baseline
The comparison value of 77 nm is the mean absolute error of always predicting the average measured maximum. A useful predictor has to beat it.
Reading these results
A proxy, not a spectrum
A proxy, not a spectrum
The gap approximates the lowest electronic excitation, but real absorption maxima depend on the bright transitions, solvent effects and vibrational structure that a single orbital gap does not capture. This page measures how useful the gap is as a quick proxy, not how well the engine predicts a full spectrum.
Calibrated on the same kind of data
Calibrated on the same kind of data
The calibration line is fitted on molecules from the same measured collection. Cross-validation by molecule guards against fitting to the molecules being scored, but a molecule class far from the collection could behave differently.
Solvent is not modelled in the gap
Solvent is not modelled in the gap
The measurements come from 358 different solvents. The error barely changes with solvent polarity (43 to 45 nm), although R² is lower in more polar solvents (0.69 down to 0.51). Whether the engine accounts for the solvent in its orbital energies is not recorded in the files this page was built from.
No competing method is shown
No competing method is shown
This page reports the engine’s orbital gap on its own. It does not compare it with other ways of predicting UV-vis absorption.
Sources
- Measured absorption maxima: a public collection of experimental UV-vis absorption data for organic molecules in solution.
- Orbital energies: the HOMO and LUMO energies computed by our molecular orbital engine for each molecule.