• Buchner, Florian
• Madsen, Georg Kent Hellerup
• Carrete, Jesús
• Kovács, Péter
• Wanzenböck, Ralf

Abstract

<strong>General description</strong> This is the code associated to the manuscript <em>Clinamen2: Versatile functional-style Python CMA-ES</em> including example applications of the code presented therein. The core algorithm is an implementation of the covariance matrix adaptation evolution strategy (CMA-ES) utilizing Cholesky decomposition and adhering to the principles of functional-style programming. The code offers a framework of building blocks and utility functions to enable the application of the CMA-ES to different problems, with a particular focus on material science, e.g., atomistic structure search. In principle, any Python function with a compatible signature can be used for loss evaluation, i.e. to drive an evolution. <strong>Installation</strong> You can get the latest published Clinamen2 release from pypi with <pre><code class="language-bash">pip install clinamen2</code></pre> but the example applications are not included there. Alternatively, to get the full package, download the source or clone the repository from GitHub and execute <pre><code class="language-bash">pip install -U pip pip install -U setuptools pip install -e .</code></pre> If you would rather use the exact version employed for the manuscript, follow the same procedure, but with the files bundled with this release instead. In any case, it might be advisable to create a dedicated virtual environment for the installation. Also see the Getting started page that also includes a simple tutorial. <strong>Example applications</strong> Four examples are included with this release with more details, including additional installation requirements, available in the manuscript and documentation. Function trial: Optimization of various test functions. Silver cluster with density functional theory: Utilizing Dask and NWChem to optimize silver clusters. Si bulk with neural-network force field: With NeuralIL as a surrogate model Si bulk is optimized to investigate defect structures. Lennard-Jones cluster: These are investigated utilizing Google JAX and the BI-Population restart scheme.

Topics

• density
• impedance spectroscopy
• cluster
• silver
• theory
• defect
• density functional theory
• decomposition
• defect structure