You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Free, open-source, and extensible tool for multi-component Calphad assessments
Currently the only open source user tool for Calphad UQ
Using pycalphad as the thermodynamic engine supports fitting custom Gibbs energy or property models
Declarative and deterministic optimization enables reproducible fitting and allows users to explicitly encode their prior beliefs and intuition
And who should use it (setting expectations for prior knowledge/experience):
Calphad modelers who want an open-source tool that they can understand, extend, and use for reproducible research
Python experience is not required to use ESPEI, but it may be useful for doing post-processing and analysis
Critical evaluation of data is an essential skill. By making it easy to optimize many parameters simultaneously, ESPEI accentuates the need to carefully select and encode knowledge into data and modeling hyperparameters. The need to understand your system of interest cannot be underemphasized.
You should not use ESPEI if you need results by tomorrow.
It is very likely and completely normal that running ESPEI for the first time in a new system will not produce a publication-quality assessment. It helps to have some prior experience with Calphad modeling concepts: modeling of individual phases, the compound energy formalism and sublattice models, using Redlich-Kister-Muggianu polynomials to model composition dependence, and an understanding of basic principles of optimization. These concepts are crucial for identifying and fixing deficiencies in data and/or parameterization.
ESPEI's performance is tightly linked to pycalphad. While ESPEI has been carefully optimized to limit the overhead of calling pycalphad's equilibrium function, ESPEI's MCMC optimization and UQ strategy still requires on the order of 10^4 - 10^6 equilibrium calculations for assessing moderately complex binary systems with UQ. Systems with many parameters or that are data-rich may benefit from utilizing ESPEI's built-in parallelization with HPC resources.
The text was updated successfully, but these errors were encountered:
Some thoughts on why ESPEI exists:
And who should use it (setting expectations for prior knowledge/experience):
You should not use ESPEI if you need results by tomorrow.
equilibrium
function, ESPEI's MCMC optimization and UQ strategy still requires on the order of 10^4 - 10^6 equilibrium calculations for assessing moderately complex binary systems with UQ. Systems with many parameters or that are data-rich may benefit from utilizing ESPEI's built-in parallelization with HPC resources.The text was updated successfully, but these errors were encountered: