Validating and Guiding Configurational Entropy Utilizing an Information-Theoretic Collective Variable
Year: 2026
Type: dissertation
Abstract: Entropy governs molecular self-assembly, phase transitions, and material stability, and yet, it remains challenging to quantify and directly engineer in molecular systems. Here we demonstrate that the computable information density (CID), a data compression-based information-theoretic metric, provides a per-configuration proxy for local and long-range structural order without the need for system-specific parameters or a priori knowledge. We validate CID across four systems of increasing complexity: (1) a single-component Lennard-Jones melt, (2) a binary Lennard-Jones phase separation, (3) a condensation and dispersion of a coarse-grained homopolymer, and (4) the assembly of carbon networks at various densities. With this, we establish CID as a practical collective variable (CV) with potential applications for enhanced sampling. Additionally, we lay the foundation for guiding a system’s configuration through CID. By utilizing a Monte Carlo tree search (MCTS), we set a target CID to explore from a metastable state to a stable state. We then compare the tree’s performance to an unguided control. The CID-guided tree search not only reaches the target CID, but also produces more ordered and energetically favorable configurations than its unguided counterpart.
DOI: 10.7282/t3-3rtb-wm50
Author: Chang, Kaelyn
Language: English
Subfield: Physical and Theoretical Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7221728825