Data-driven exploration of synthesizable strained hydrocarbons as high-energy-density candidates for sustainable aviation fuels

Abstract

This study explores the chemical space of strained polycyclic hydrocarbons to identify novel candidates for high-energy-density sustainable aviation fuels (SAFs) and for additional applications as 3D building blocks in natural product synthesis. Using the AIQM2 quantum machine-learning potential, which approaches CCSD(T)/CBS-level accuracy, we generated a high-fidelity thermochemical database of novel hydrocarbon architectures. The workflow integrates PyAR-based molecular generation, multilevel geometry optimization, AIQM2 thermochemical evaluation, and transfer-learned D-MPNN property prediction. High-throughput screening identified 20 novel hydrocarbon candidates (up to C₁₀) that satisfy first-pass criteria for net heat of combustion, liquid density, melting point, and synthetic plausibility, providing a prioritized set of high-energy-density molecules for further computational and experimental evaluation. In addition, a distinct topology-selected set of 16 novel cage-like hydrocarbons was identified, illustrating the broader strained-hydrocarbon structural space accessible through the workflow. Overall, this work demonstrates a data-driven discovery framework for prioritizing strained hydrocarbon candidates, while emphasizing that full aviation-fuel suitability will require further evaluation of mixture-level and operability properties.