Thermodynamic Collective Variables for Phase Transitions

15 concepts4 papers1 proposal

When simulating how a liquid freezes into a solid, computational chemists face a fundamental dilemma: how do you guide a simulation toward a crystal structure if you do not know what that structure looks like beforehand?

Traditionally, researchers have relied on geometric collective variables (CVs)—such as bond-orientational order parameters—to track and drive phase transitions in molecular dynamics. These variables count specific spatial symmetries or bond angles. While effective if the target crystal structure is known, geometric CVs act as a conceptual straightjacket. If you bias a simulation using a metric tailored for a face-centered cubic (FCC) lattice, the system is highly unlikely to explore or discover a body-centered cubic (BCC) or an entirely novel polymorph.

To break free from this geometric bias, a powerful alternative paradigm has emerged: using fundamental thermodynamic properties—specifically entropy and enthalpy—as the coordinates to identify and drive phase transitions. Because nature itself drives crystallization through the minimization of free energy ($G = H - TS$), biasing a system along the axes of enthalpy ($H$) and entropy ($S$) allows atoms to spontaneously organize into their preferred configurations without any predefined structural templates.

Geometric Biasing:    Liquid ──(Force specific angles)──> Specific Crystal (e.g., FCC)
Thermodynamic Biasing: Liquid ──(Reduce Enthalpy/Entropy)──> Most Stable Polymorph (BCC, FCC, etc.)

The Thermodynamic Fingerprint

To use thermodynamics to classify or drive a simulation, one must first solve a difficult problem: entropy and enthalpy are inherently macroscopic, global properties of an ensemble. How can we calculate them on the fly for local atomic configurations?

In Entropy based fingerprint for local crystalline order, Pablo M. Piaggi demonstrated that local structural order is naturally encoded within a system's local thermodynamic state. Instead of measuring bond angles, the author constructed a continuous and differentiable "fingerprint" by projecting an approximation of two-body excess entropy ($s^{(2)}$) and local enthalpy onto individual atoms. The local enthalpy is obtained by partitioning the system-wide potential energy and volume among individual atoms, while the local entropy is calculated using a localized radial distribution function.

This approach works because different crystal polymorphs possess distinct thermodynamic signatures. For example, FCC and hexagonal close-packed (HCP) structures share nearly identical local coordination numbers, making them notoriously difficult to distinguish with simple geometric cutoffs. However, they differ subtly in their potential energy and volume partitions. By mapping local entropy against local enthalpy, the author created a high-resolution classifier capable of distinguishing between these subtle stacking differences purely based on thermodynamic signatures, completely independent of predefined geometric symmetries.

Driving Structure-Agnostic Crystallization

Once we can compute local thermodynamic variables, we can use them as collective variables in enhanced sampling methods like metadynamics to actively drive phase transitions.

In Enhancing Entropy and Enthalpy Fluctuations to Drive Crystallization in Atomistic Simulations, Pablo M. Piaggi et al. applied this concept by biasing a system's global enthalpy and entropy fluctuations rather than its geometry. By penalizing the high-entropy, high-enthalpy liquid state, the simulation forces the system to explore regions of lower entropy and enthalpy. Because the bias does not favor any specific spatial symmetry, the atoms are free to find their own optimal packing. The system spontaneously crystallizes into whatever lattice arrangement is thermodynamically favorable under those conditions, such as BCC or FCC, without any prior structural knowledge embedded in the simulation parameters.

High S, High H (Liquid) ──[Metadynamics Bias]──> Fluctuations Enhanced ──> Spontaneous Nucleation of Lowest Free-Energy Lattice

This structure-agnostic approach becomes particularly valuable when dealing with complex molecular crystals, where molecules can arrange themselves in countless packing configurations (polymorphism). In Predicting polymorphism in molecular crystals using orientational entropy, Pablo M. Piaggi extended this framework by introducing a collective variable based on orientational entropy. By combining well-tempered metadynamics with this entropy surrogate, the author accelerated the crystallization process from a disordered liquid phase directly within finite-temperature simulations. This method successfully discovered both known and entirely new molecular polymorphs, demonstrating that accounting for entropic effects at finite temperatures can reveal stable crystal structures that traditional zero-temperature crystal structure prediction methods completely miss.

The Mathematical Foundation of Locality

The success of partitioning thermodynamic properties into per-atom descriptors raises a deeper theoretical question: is entropy truly a local property? While enthalpy can be intuitively carved up into local bond energies and atomic volumes, entropy is a measure of the available phase space of the entire system, suggesting a global dependency.

This theoretical gap was addressed by Tina Torabi et al. in Surrogate Models for Vibrational Entropy Based on a Spatial Decomposition. The authors provided a rigorous mathematical proof demonstrating that vibrational entropy exhibits strong spatial locality when decomposed into individual "site entropies."

Specifically, the authors proved that the sensitivity of an individual atom's site entropy decays algebraically with distance. This means that an atom's contribution to the overall vibrational entropy of a material is overwhelmingly determined by its immediate atomic neighborhood.

[Central Atom] ──(Strong Influence)──> Immediate Neighbors ──(Decays Algebraically)──> Distant Atoms

By establishing this rigorous foundation of locality, the authors justified the creation of highly efficient machine learning surrogate models. Using the Atomic Cluster Expansion (ACE) framework, the authors trained localized models to predict total vibrational entropy from local site environments. This dramatically reduces the computational cost of evaluating entropic contributions, paving the way for large-scale, thermodynamically accurate materials simulations.

Why It Works: A Summary

Thermodynamic collective variables succeed because they align the computational coordinates of a simulation with the physical drivers of nature.

  1. Symmetry Independence: By mapping states to entropy and enthalpy rather than bond angles, simulations can discover unexpected crystal structures without being guided toward a preconceived geometry.
  2. High Resolution: Thermodynamic signatures inherently capture subtle differences in potential energy, volume partitioning, and vibrational states, allowing them to distinguish between polymorphs that look geometrically identical.
  3. Physical Locality: Because an atom's entropic contribution decays rapidly with distance, global thermodynamic properties can be accurately decomposed into local site descriptors, enabling both precise classification and fast machine learning acceleration.

Key Foundational Work

Entropy based fingerprint for local crystalline order (2017) — contributed 5 concept elements to this direction.
Enhancing Entropy and Enthalpy Fluctuations to Drive Crystallization in Atomistic Simulations (2017) — contributed 2 concept elements to this direction.
Predicting polymorphism in molecular crystals using orientational entropy (2018) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_thermodynamic_collective_variables_a78a6132,
  title = {thermodynamic collective variables},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/thermodynamic-collective-variables/rdr_a78a6132f5c14c6487f2013cc0589fd9},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_thermodynamic_collective_variables_a78a6132_v1,
  title = {thermodynamic collective variables},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/thermodynamic-collective-variables/rdr_a78a6132f5c14c6487f2013cc0589fd9/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake