We detail a framework to systematically derive polynomial basis functions for approximating isometry and permutation invariant functions, particularly with an eye to modelling properties of atomistic systems and analyzing the convergence to a high-fidelity reference model. Our presentation builds on (Drautz, Phys. Rev. B 99, 2019), extending the derivation in a way that yields immediate guarantees that a complete basis is indeed obtained. In addition, we discuss generalisations, a variety of related open challenges, particularly from a numerical analysis perspective, around basis optimisation and parameter estimation.
翻译:我们详细制定了一个框架,以便系统地得出近似异度测量和变异功能的多元基函数,特别是着眼于原子系统建模特性和分析与高不忠参考模型的趋同性。我们的发言以(Drautz,Phys.Rev.B.99,2019)为基础,扩展推论,以便立即保证确实获得完整的基础。此外,我们讨论了一般化、各种相关的公开挑战,特别是从数字分析的角度,围绕基础优化和参数估计来讨论。