-
Incomplete selected inversion for linear-scaling electronic structure calculations
Authors:
Simon Etter
Abstract:
Pole Expansion and Selected Inversion (PEXSI) is an efficient scheme for evaluating selected entries of functions of large sparse matrices as required e.g. in electronic structure algorithms. We show that the triangular factorizations computed by the PEXSI scheme exhibit a localization property similar to that of matrix functions, and we present a modified PEXSI algorithm which exploits this obser…
▽ More
Pole Expansion and Selected Inversion (PEXSI) is an efficient scheme for evaluating selected entries of functions of large sparse matrices as required e.g. in electronic structure algorithms. We show that the triangular factorizations computed by the PEXSI scheme exhibit a localization property similar to that of matrix functions, and we present a modified PEXSI algorithm which exploits this observation to achieve linear scaling. To the best of our knowledge, the resulting incomplete PEXSI (iPEXSI) algorithm is the first linear-scaling algorithm which scales provably better than cubically even in the absence of localization, and we hope that this will help to further lower the critical system size where linear-scaling algorithms begin to outperform the diagonalization algorithm.
△ Less
Submitted 17 January, 2020;
originally announced January 2020.
-
Atomic Cluster Expansion: Completeness, Efficiency and Stability
Authors:
Genevieve Dusson,
Markus Bachmayr,
Gabor Csanyi,
Ralf Drautz,
Simon Etter,
Cas van der Oord,
Christoph Ortner
Abstract:
The Atomic Cluster Expansion (Drautz, Phys. Rev. B 99, 2019) provides 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. Our presentation extends the derivation by proposing a precomputation algorithm that yields immediate guarantees that a complete ba…
▽ More
The Atomic Cluster Expansion (Drautz, Phys. Rev. B 99, 2019) provides 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. Our presentation extends the derivation by proposing a precomputation algorithm that yields immediate guarantees that a complete basis is obtained. We provide a fast recursive algorithm for efficient evaluation and illustrate its performance in numerical tests. Finally, we discuss generalisations and open challenges, particularly from a numerical stability perspective, around basis optimisation and parameter estimation, paving the way towards a comprehensive analysis of the convergence to a high-fidelity reference model.
△ Less
Submitted 12 May, 2021; v1 submitted 8 November, 2019;
originally announced November 2019.
-
Modeling and Computation of Kubo Conductivity for 2D Incommensurate Bilayers
Authors:
Simon Etter,
Daniel Massatt,
Mitchell Luskin,
Christoph Ortner
Abstract:
This paper presents a unified approach to the modeling and computation of the Kubo conductivity of incommensurate bilayer heterostructures at finite temperature. Firstly, we derive an expression for the large-body limit of Kubo-Greenwood conductivity in terms of an integral of the conductivity function with respect to a current-current correlation measure. We then observe that the incommensurate s…
▽ More
This paper presents a unified approach to the modeling and computation of the Kubo conductivity of incommensurate bilayer heterostructures at finite temperature. Firstly, we derive an expression for the large-body limit of Kubo-Greenwood conductivity in terms of an integral of the conductivity function with respect to a current-current correlation measure. We then observe that the incommensurate structure can be exploited to decompose the current-current correlation measure into local contribution and deduce an approximation scheme which is exponentially convergent in terms of domain size. Secondly, we analyze the cost of computing local conductivities via Chebyshev approximation. Our main finding is that if the inverse temperature $β$ is sufficiently small compared to the inverse relaxation time $η$, namely $β\lesssim η^{-1/2}$, then the dominant computational cost is $\mathcal{O}\bigl(η^{-3/2}\bigr)$ inner products for a suitably truncated Chebyshev series, which significantly improves on the $\mathcal{O}\bigl(η^{-2}\bigr)$ inner products required by a naive Chebyshev approximation. Thirdly, we propose a rational approximation scheme for the low temperature regime $η^{-1/2} \lesssim β$, where the cost of the polynomial method increases up to $\mathcal{O}\bigl(β^2\bigr),$ but the rational scheme scales much more mildly with respect to $β$.
△ Less
Submitted 23 September, 2020; v1 submitted 2 July, 2019;
originally announced July 2019.
