Natural Orbital Non-Orthogonal Configuration Interaction

Daniel Graf, Alex J. W. Thom · 2024

Read Natural Orbital Non-Orthogonal Configuration Interaction by Daniel Graf, Alex J. W. Thom (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-17602.txt.

Introduction

Daniel Graf and Alex Thom introduce a new variant of a flexible method for computing the electronic states of molecules. The base method, non-orthogonal configuration interaction, allows more freedom than the standard approach in how states are built. That freedom creates a problem: no accepted recipe exists for choosing a compact yet accurate set of building blocks. The authors aim to supply one.

Their approach pursues three goals at once. It should be systematic, it should keep the expansion small, and it should treat ground and excited states even-handedly. They develop the method step by step, testing each building block on simple model systems as they go. The tests show the design works as intended. The paper offers the scheme as a step toward making this flexible family of methods routine in quantum chemistry.

Catalog record

CREATORDaniel Graf, Alex J. W. Thom
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE25 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-17602

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Natural Orbital Non-Orthogonal Configuration Interaction Daniel Graf, Alex J. W. Thom Abstract Non-orthogonal configuration interaction (NOCI) is a generalization of the standard orthogonal configuration interaction (CI) method and offers a highly flexible framework for describing ground and excited electronic states. However, this flexibility also comes with challenges, as there is still no clear or generally accepted approach for constructing a compact and accurate state basis for NOCI. In this work, we take a step toward addressing this challenge by introducing a novel NOCI approach designed with three primary objectives: (1) ensuring the method is systematic, (2) achieving a compact NOCI expansion, and (3) treating all electronic states of interest on equal footing. The development of our approach is presented step by step, with each building block evaluated and validated through applications to simple model systems, demonstrating its effectiveness and potential. keywords non-orthogonal configuration interaction, NOCI, natural orbitals † † affiliation: Yusuf Hamied Department of Chemistry, University of Cambridge, Cambridge † † affiliation: Department of Chemistry, Ludwig-Maximilians-Universität München, Munich † † email: [email protected] † † affiliation: Yusuf Hamied Department of Chemistry, University of Cambridge, Cambridge 1 Introduction Excited states lie at the core of many fundamental processes in nature, such as photosynthesis Cerullo et al. (2002) ; Cheng and Fleming (2009) , human vision Herbst et al. (2002) ; Tapavicza et al. (2007) , photoinduced damage of essential biomolecules Schultz et al. (2004) ; Schreier et al. (2007) , and even theories concerning the origin of life Roberts et al. (2018) ; Xu et al. (2020) — a detailed understanding of these processes is therefore highly desirable. While spectroscopic methods provide valuable insights, they face limitations in both spatial and temporal resolution, highlighting the indispensable role of theoretical investigations. However, accurately describing such processes remains a formidable challenge. The primary difficulty arises from the need to treat multiple electronic states on the same footing, requiring a balanced treatment of both dynamical and static electron correlation, while at the same time ensuring a feasible computational cost. Currently, time-dependent density functional theory Runge and Gross (1984) (TDDFT) is the most widely used framework for excited-state calculations due to its excellent cost-to-performance ratio. While TDDFT is, in principle, an exact theory, Hirata and Head-Gordon (1999) it depends on an unknown modification of the Hamiltonian — specifically, the exchange-correlation potential. Consequently, the accuracy of TDDFT is critically determined by the choice of exchange-correlation functional. However, given the vast array of available approximations, Mardirossian and Head-Gordon (2017) selecting an appropriate functional is often challenging. Additionally, in TDDFT, excitation energies are typically computed in the linear response regime, Casida (1995) which results in an incapability to describe double excitations. The most significant limitation of TDDFT, however, is its inability to accurately describe static correlation effects in the electronic ground state. Matsika (2021) This deficiency leads to an imbalance in the representation of the system’s electronic states, undermining its reliability for systems where static correlation plays a major role. The most widely used method for addressing the static correlation problem in the ground state, thereby enabling a balanced description of different electronic states, is the complete active space self-consistent field (CASSCF) method Roos et al. (1980) — the multi-reference analogue of Hartree–Fock theory. While CASSCF is highly effective at capturing static correlation effects, its computational cost increases combinatorially with the active space size and, therefore, one is frequently restricted to small active spaces including only the very most important orbitals. The selection of these active orbitals is far from trivial and necessitates expert knowledge, preventing its use as a black-box method. Often it is not even possible to include all orbitals which should be included due to the enormous cost of the method, resulting in a (possibly) unsatisfactory description of the system at hand. When considering the calculation of excited states, additional challenges arise regarding the orbital optimization: One approach is to target the excited state explicitly using a state-specific formalism. However, this method inherently biases the calculation towards the selected state and is prone to variational collapse onto the ground state. Tran et al. (2019) An alternative is the state-averaging formalism, which optimizes a weighted average of the energies of multiple states, allowing for simultaneous computation of several states. Werner and Meyer (1981) Unfortunately, the resulting wave functions are not true stationary points on the energy surface, which limits their accuracy compared to states obtained from state-specific procedures. Motivated by the aforementioned challenges, we here propose a non-orthogonal configuration interaction (NOCI) approach based on natural orbitals for multiple electronic states of interest. The approach was designed with three key objectives in mind: (1) being systematic, (2) achieving a compact CI expansion, and (3) ensuring a balanced treatment of the system’s electronic states. We will demonstrate, using simple test cases, that the presented approach forms a promising starting point to achieve these objectives. 2 Non-orthogonal Configuration Interaction Non-orthogonal configuration interaction is a generalized extension of the well-established orthogonal configuration interaction (CI) method. In the NOCI framework, the generalized eigenvalue problem to determine the electronic states is expressed as \sum_{J}\left(H_{IJ}-ES_{IJ}\right)C_{J}=0. (1) where H_{IJ} and S_{IJ} represent Hamiltonian and overlap matrix elements in the basis of Slater determinants; C_{J} and E denote the expansion coefficients and the eigenvalue — the energy — of a specific state, respectively. The matrix elements can either be computed using Löwdin’s pairing approach Löwdin (1962) ; Amos et al. (1961) in conjunction with the generalized Slater-Condon rules, or by employing the recently introduced generalized Wick’s theorem. Burton (2021) NOCI offers significant flexibility in constructing the basis of Slater determinants, but this flexibility also poses a key challenge: defining an accurate and compact basis for the states of interest remains an open problem. In this work, we propose a novel approach utilizing principal natural orbital determinants to address this challenge. 3 Principal Natural Orbital Determinants as NOCI Basis The natural orbitals (NOs) of a given electronic state are the eigenstates of the respective one-particle reduced density matrix (1-RDM): \gamma[\Psi]\ket{\phi_{p}}=n_{p}\ket{\phi_{p}} (2) with the 1-RDM defined in a real-space basis according to \gamma[\Psi]=N\int\Psi^{*}(\mathbf{x}_{1},\mathbf{x}_{2},\dots)\Psi(\mathbf{x}^{{}^{\prime}}_{1},\mathbf{x}_{2},\dots)\mathrm{d}\mathbf{x}_{2}\mathrm{d}\mathbf{x}_{3}\dots (3) The corresponding eigenvalues are the natural occupation numbers \{n_{p}\} for which: n_{p}\in[0,1];\qquad\sum_{p}n_{p}=N_{\text{el}} (4) where N_{\text{el}} denotes the number of electrons in the system. The NOs form an orthonormal basis and the determinant constructed from the N_{\text{el}} NOs with the highest occupation numbers is referred to as the principal natural orbital determinant. Yu et al. (2024) In the following, we will denote the principal NO determinant for state I (whose wavefunction is \ket{\Psi_{I}} ) as \ket{\Phi_{I}} . Since \gamma[\Phi_{I}] is the best possible idempotent approximation of \gamma[\Psi_{I}] , CI expansions based on these determinants are highly compact. Coe and Paterson (2015) ; Yu et al. (2024) The central idea of this work is to form, for each state of interest, a compact orthogonal CI basis based on its principal NO determinant. Since the CI expansions of the different states of interest are non-orthogonal to each other, this approach naturally leads to a NOCI problem. To evaluate the benefits of including principal NO determinants from multiple states in the CI expansion — thereby transforming the approach into a NOCI one — we conducted tests on three simple systems taken from the QUEST database. Loos et al.

Read the complete text