Assessing Crystal Stability via Thermodynamic Convex Hulls

46 concepts39 papers1 proposal

When searching for new materials, the intuitive first step is to calculate a crystal structure's formation energy. If this energy is negative, the material releases energy upon forming from its constituent elements, suggesting it might be stable.

However, this absolute energy metric is a dangerous illusion. A compound can be highly exothermic yet completely un-synthesizable. This happens because nature is inherently competitive: a newly proposed material does not just compete against its isolated source elements; it must compete against every possible combination of other phases within that chemical space. As Raymundo Arróyave (Phase Stability Through Machine Learning) notes, understanding this thermodynamic competition is the "beginning of wisdom" in materials science.

To accurately evaluate whether a material can exist, researchers construct a thermodynamic convex hull.

The Geometric Intuition of the Convex Hull

Imagine plotting the formation energy of every known and hypothetical compound in a chemical system against its compositional ratios. To find the stable states, imagine wrapping a rubber band around the bottom of this data plot. The segments of the rubber band that stretch tightly between the lowest-energy points form the convex hull.

Formation Energy (E)
▲
│      ● Hypothetical Phase A (Unstable)
│     /  \
│    /    ● Hypothetical Phase B (Metastable, ΔE_hull > 0)
│   ●───────_
│  /         ──────● Stable Phase C (On the Hull, ΔE_hull = 0)
│ /
└────────────────────────────────────────► Composition (x)

The points that touch this lower boundary represent the absolute lowest-energy states for those specific compositions. These are the thermodynamically stable phases. Any compound sitting above this line is unstable or metastable; it can lower its total energy by decomposing into a mixture of the stable phases that define the hull beneath it. As Prashant Singh and Daniel Sauceda (The effect of chemical disorder on defect formation and migration in disordered max phases) emphasize, evaluating a material's stability without this full convex hull analysis leaves its true viability unverified against competing decomposition pathways.

High-Throughput Screening and Metastability

In practice, computing these hulls requires mapping vast, multi-dimensional chemical spaces. Platforms like AFLOW-CHULL, developed by Corey Oses et al. (AFLOW-CHULL: Cloud-Oriented Platform for Autonomous Phase Stability Analysis), automate this process across massive databases of calculated material properties.

This high-throughput approach has driven numerous experimental breakthroughs. For instance, D. J. Fredeman et al. (Computationally driven experimental discovery of the CeIr4In compound) combined density functional theory (DFT) simulations with experimental arc-melting to discover and synthesize the brand-new compound $\text{CeIr}_4\text{In}$ by proving it lies on the ternary convex hull. Similarly, Soo Kim et al. (High-throughput computational design of cathode coatings for Li-ion batteries) screened over 130,000 materials to identify stable battery coatings that resist decomposition under harsh operating conditions.

However, being off the hull does not completely rule out synthesis. Many valuable materials are metastable. The distance to the convex hull ($\Delta E_{\text{hull}}$) quantifies exactly how unstable a material is. In complex oxide systems, for example, the authors of a study on the $\text{SrO–Al}_2\text{O}_3\text{–SiO}_2$ phase field (A computationally-guided non-equilibrium synthesis approach to materials discovery in the SrO–Al2O3–SiO2 phase field) demonstrated that targeting structures within a narrow energy window—typically within 50 meV/atom above the hull—provides a high-probability pathway for isolating novel metastable phases using non-equilibrium synthesis techniques.

Hull-Aware Machine Learning

As machine learning has integrated into materials science, researchers realized that standard regression models—which minimize mean squared error across all data points—are poorly suited for finding stable crystals. A model might have excellent overall accuracy but fail completely if its small energy errors occur right at the stability boundary, leading to false positives or missed discoveries.

To fix this, Janosh Riebesell et al. (A framework to evaluate machine learning crystal stability predictions) introduced the Matbench Discovery framework. This framework shifts the evaluation metric from raw energy regression to stability-boundary classification, prioritizing precision and recall right at the convex hull. This paradigm shift has validated models like Wren, developed by Abhijith S Parackal et al. (Rapid Discovery of Stable Materials by Coordinate-free Coarse Graining), which leverages coordinate-free crystal symmetry representations and exhibits its highest accuracy precisely near the thermodynamic stability boundary.

Traditional ML Focus:                 Hull-Aware ML Focus:
Minimize average error everywhere     Maximize precision/recall at the boundary
        ▲                                     ▲
        │   ●                                 │   ●
  ──────┼───_─────── Error                    │    \
        │    ──────●                          ──────●◄─── Focus here!

Modern workflows now integrate machine learning and thermodynamics directly. Joseph H. Montoya et al. (Autonomous intelligent agents for accelerated materials discovery) developed autonomous agents that feed ML energy predictions directly into convex hull calculators, allowing the system to evaluate a material's realizability based on its decomposition potential rather than raw energy.

To make this search even more efficient, Andrew C. Novick et al. (Probabilistic prediction of material stability: integrating convex hulls into active learning) proposed Convex Hull-Aware Active Learning (CAL). Instead of wasting computational resources mapping an entire energy landscape, CAL uses a Bayesian framework to intelligently select calculations that resolve the exact shape of the convex hull boundary, drastically accelerating the confident discovery of stable inorganic solids.

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_convex_hull_stability_analysis_d84d4b97,
  title = {convex hull stability analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/convex-hull-stability-analysis/rdr_d84d4b970d6040c78eb04495a316af20},
  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_convex_hull_stability_analysis_d84d4b97_v1,
  title = {convex hull stability analysis},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/convex-hull-stability-analysis/rdr_d84d4b970d6040c78eb04495a316af20/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