Generation of Phase Transitions Boundaries via Convolutional Neural Networks

Christopher Alexis Ibarra, University of Texas at El Paso


Accurate mapping of phase transitions boundaries is crucial in accurately modeling theequation of state of materials. The phase transitions can be structural (solid-solid) drivenby temperature or pressure or a phase change like melting which defines the solid-liquidmelt line. There exist many computational methods for evaluating the phase diagramat a particular point in temperature (T) and pressure (P). Most of these methods involveevaluation of a single (P,T) point at a time. The present work partially automates the searchfor phase boundaries lines utilizing a machine learning method based on convolutionalneural networks and an efficient search algorithm and a shrinking enclosure. This neuralnetwork (NN) approach is applied to the prediction of the melt line of metals as a functionof pressure. The proposed NN method is implemented using the so-called Z-method,a molecular-dynamics-based computational approach for determining upper bounds in thesolid-liquid melt line of a material. In this method, the system is subjected to ”jumps” intemperature until melting is achieved. The usefulness of our proposed NN search methodis that it can be easily applied to a wide range of inter-atomic potentials and hence helptest their accuracy and agreement with experiments. Future machine learning applicationscan be similarly applied for determining more subtle and complex phase diagrams.

Subject Area

Physics|Thermodynamics|Particle physics

Recommended Citation

Ibarra, Christopher Alexis, "Generation of Phase Transitions Boundaries via Convolutional Neural Networks" (2022). ETD Collection for University of Texas, El Paso. AAI30241940.