Research work · 2025

Classical Machine Learning vs Quantum Machine Learning in Computational Chemistry: Hype, Hope, and Horizons

Samrat Chakraborty

American Chemical Society (ACS)

Abstract

Computational chemistry has entered a new era where machine learning (ML) models—particularly graph neural networks and machine learning force fields—routinely deliver quantum mechanical accuracy at classical speeds, scaling to millions of atoms and reshaping workflows in drug discovery, catalysis, and materials design. At the same time, quantum computing is promoted as chemistry’s “killer application,” with the promise of solving many body problems beyond classical reach. Between these poles lies quantum machine learning (QML), a rapidly developing field that integrates quantum algorithms with ML paradigms. But is QML today a revolution, or merely hype? This article surveys the current state of classical ML in computational chemistry, highlighting its extraordinary successes and real limitations of transferability, interpretability, and generalization. We then examine the current landscape of quantum computing hardware and algorithms, clarifying what has actually been achieved with NISQ devices and where fundamental roadblocks remain. Building from this foundation, we analyze the foundations of QML algorithms—quantum kernels, variational quantum circuits, and hybrid workflows—and their putative advantages in polynomial scaling and Hilbert space expressivity. To ground discussion in practice, we present a case study of H₂ molecular energies, contrasting classical ML’s near instantaneous accuracy with quantum algorithms like the Variational Quantum Eigensolver. The outcome is telling: performance parity, not advantage, achieved only for toy systems under heavy simplification. This provides proof of principle for QML’s viability while simultaneously underscoring the severe constraints of current hardware. From this synthesis, we conclude that QML is presently characterized more by hype than utility, given the industrial dominance of classical ML. Yet in the long term—likely the mid 2030s and beyond—fault tolerant quantum hardware could transform computational chemistry, especially for strongly correlated systems where classical surrogates fail. The future lies not in premature claims, but in hybrid strategies, honest benchmarks, interpretable models, and sustainable integration across classical ML and quantum approaches. QML is not today’s solution. It is tomorrow’s horizon. And whether it becomes a true game changer will depend not on marketing cycles, but on the community’s ability to bridge hype with the patient, long term engineering of methods and machines.

DOI: 10.26434/chemrxiv-2025-71814