Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Effort-based Versioning.

Unreleased

(no summary yet)

1.3.0 - 2026-10-12

This release introduces the Debye spectral model.

Added

  • The DebyeModel has been added to the list of built-in models. It complements the LorentzModel and is suitable for spectra with a minimum at zero frequency. Such spectra are typically the result of hopping-like diffusion superimposed on a higher-frequency oscillatory motion. The DebyeModel does not attempt to model these resonances in full, and only accounts for the onset towards such a resonance peak, to keep the number of parameters small.

Changed

  • The molten salt example has been refactored to use only the DebyeModel, since it is expected to be the right model for this system a priori.

  • The LorentzModel now also rejects fits with a non-positive autocorrelation integral (p0 <= 0) when deriving the exponential correlation time. Such fits are discarded from the average over cutoff frequencies, like fits with an ill-defined correlation time.

  • The package metadata now declares only the LGPL-3.0-or-later license, as an SPDX expression, because the dual-licensed documentation is not part of the package.

Fixed

  • The variances of exp_block_time and exp_simulation_time in the LorentzModel were propagated with a linear instead of a squared prefactor, which is now fixed.

1.2.2 - 2026-09-24

This is a preview of an upcoming version, featuring the designed quadrature module. It can benefit from some more testing and feedback, which this release will facilitate.

Added

  • stacie.dq module for constructing designed quadrature rules with equal (or user-specified) weights. These grids are suitable for integrating over a low-dimensional parameter space, where the integrand is noisy and expensive, typically the result of a stochastic simulation. In the context of STACIE, the parameter space can consist of one or a few thermodynamic parameters (e.g. temperature, pressure) that are varied across a set of simulations. The integral of interest is typically an ensemble average of a transport coefficient. One can provide the distribution of thermodynamic parameters a priori, and the quadrature rule will be optimized to integrate over this distribution, by taking a simple average of the function values at the quadrature points. The advantage of equal weights is that the variance of the average, due to (identically distributed and independent) noise in the function values, is minimal.

Changed

  • Refactored several aspects of the development environment.

Fixed

  • Many documentation improvements and clarifications, including updated literature references.

1.2.1 - 2025-12-28

Minor documentation and dependency improvements.

Changed

  • Several documentation improvements and clarifications.

  • Removed unused dependency cattrs and lowered minimum required of attrs to 23.1.0 to facilitate installation in more constrained environments.

1.2.0 - 2025-12-23

Reformulation of the Lorentz model parameters.

Changed

  • New parametrization of the linear coefficients in the Lorentz model: \(C_0\) and \(C_1\) instead of \(A\) and \(B\).

  • Documentation improvements and clarifications.

1.1.3 - 2025-11-19

Support for Python 3.10

Changed

  • Add support for Python 3.10

1.1.2 - 2025-11-19

Additional packaging cleanups.

Changed

  • Removed a few more files from the source package that could cause confusion.

  • Removed an unused testing dependency.

1.1.1 - 2025-11-18

Smaller source package (10x size reduction) and dependency simplification.

Changed

  • PyPI Package size reduction by a factor 10. Only essential files are now included in the package.

  • Remove path dependency to facilitate packaging and distribution.

1.1.0 - 2025-11-10

Improved Lorentz model, with a more robust estimate of the exponential correlation time and its uncertainty. Improved examples.

Added

  • Cloud cover example

  • Comparison between STACIE’s diffusion coefficient and a more conventional MSD analysis in the surface diffusion example.

Changed

  • The license of the documentation has been updated to a choice of license (CC-BY-SA-4.0 OR LGPL-3.0-or-later). STACIE’s code is still distributed under a single LGPL-3.0-or-later license.

  • The penalty in the LorentzModel has been improved to exclude and down-weight results from frequency cutoffs with a high relative error of the estimated exponential correlation time. In this regime, the maximum a posteriori estimate of uncertainty is not reliable.

Fixed

  • Several documentation improvements:

    • Clarify how the derivation of block-averaged velocities for diffusion and electrical conductivity, using atomic positions (or dipole vectors) as input.

    • Improve explanation on discarding the DC-component of the spectrum.

    • Add more helpful comments on how to deal with unit conversion.

    • Fixed a typo in the equation of the marginalization weights.

  • Repocard images were added.

  • Dataset metadata improvements.

  • Several other minor issues in documentation and tooling were fixed.

1.0.0 - 2025-06-26

This is the first stable release of STACIE!

Changed

  • Metadata and citation updates

1.0.0rc1 - 2025-06-25

This is the first release candidate of STACIE, with a final release expected very soon. The main remaining issues are related to (back)linking of external resources in the documentation and README files.