Predicting Excited-State Energies from Ground-State Descriptors in Thermally Fluctuating π-Conjugated Molecules

Predicting excitation energies for thermally accessible ensembles of organic semiconductors remains computationally challenging because high-accuracy excited-state methods are prohibitively expensive for extensive thermal sampling. We report an interpretable regression framework that predicts singlet and triplet excitation energies directly from the ground-state electronic descriptors. The model reveals an approximately planar relationship of excitation energies (e.g., the lowest singlet (S1) and triplet (T1) excited states) with GapHL (HOMO-LUMO gap) and KHL (HOMO-LUMO exchange integral), together with a near-constant offset δ that captures the contributions beyond frontier molecular orbital effects. Validated on thermally sampled configurations of tetracene derivatives, the framework reproduces quantum-chemical benchmarks and yields E(S1) distribution maxima consistent with available experimental lowest-energy S1 absorption maxima, suggesting transferability to related π-conjugated systems. Importantly, we further demonstrate the feasibility of this descriptor-based mapping for derivatives built on nontetracene backbones, supporting backbone-level transferability beyond tetracene. By capturing excitation-energy distributions induced by structural fluctuations, the approach provides an efficient and physically transparent description of thermally modulated excited-state energetics relevant to excitonic materials and organic semiconductor design.