This research demonstrates how deep learning can model molecular properties, suggesting internal representations are equivalent to the density matrix.
Quantum state tomography has been widely used to reconstruct the quantum state of a system from a set of informationally-complete measurements. Obtaining enough information about, e.g., the wavefunction of a molecule allows its complete characterization. On the other hand, deep learning models for molecular property prediction have demonstrated the capability to predict properties of unseen molecules, thereby internally characterizing them in a latent space spanning potential virtual molecules. In this work, we argue that deep learning can achieve its characterization power by learning an internal representation that is informationally-equivalent to the molecule's density matrix. We call such a representation a deep tomography, akin to quantum state tomography. We discuss how it is possible for property prediction foundation models, trained on data of informationally-complete observables, to have representations that contain the same information as that of a molecule's reduced quantum density matrix. Thus, such learned representation should be sufficient to predict arbitrary properties of arbitrary unseen molecules.
No takes yet. Share an insight, caveat, or question.
Ortega-Ochoa et al. (2025) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: