Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2508.11822 · The Journal of Chemical Physics

Efficient optimization of low-rank antisymmetric product of geminals wavefunction using the direct Givens rotation method

Airi Kawasaki, Rei Oshima, Naoki Nakatani, Hiromi Nakai

Published Nov 19, 2025• 1 citation
arXivHTMLPDF

Abstract

In our previous study [Kawasaki and Nakatani, Mol. Phys. 123(16), e2449186 (2025)], we proposed the low-rank antisymmetric product of geminals (APG) method, which reconstructs the wavefunction by extracting only the important eigenvalues from the APG wavefunction. However, its practical application was limited by the high computational cost from an orbital optimization process, making higher-rank calculations difficult. In this work, we reformulate the orbital part of the wavefunction using Givens rotation matrices, enabling an analytical treatment of the variational optimization. By combining the low-rank APG with the direct Givens rotation method, we achieved a significant improvement in optimization efficiency. We applied the developed method to small molecular systems and confirmed that it provides high accuracy, while also significantly reducing the computational time compared to conventional methods.

DOI
10.1063/5.0296913

Related papers

Low-rank decomposition on the antisymmetric product of geminals for strongly correlated electrons

Airi Kawasaki, Naoki Nakatani

We investigated some variational methods to compute a wavefunction based on antisymmetric product of geminals (APG). The Waring decomposition on the APG wavefunction leads a finite sum of antisymmetrized geminal power (AGP) wavefunctions, each for which the variational principle can be applied. We call this as AGP-CI method which provides a variational solution of the APG wavefunction efficiently. However, number of AGP wavefunctions in the exact AGP-CI formalism become exponentially large in case of many-electron systems. Therefore, we also investigate the low-rank APG wavefunction, in which the coefficient matrices of geminals are factorized by the Schur decomposition. Interestingly, only a few non-zero eigenvalues (up to half number of electrons) were found from the Schur decomposition on the APG wavefunction. We developed some methods to approximate the APG wavefunction by lowering the ranks of coefficient matrices of geminals, and demonstrate their performance. Our new geminal method based on the low-rank decomposition can drastically reduce the number of variational parameters, although no efficient formalism has been elaborated so far, due to the difficulty of optimization caused by the inability to compute analytical derivatives.

arXiv:2410.06666•Molecular Physics•Jan 21, 2025•5 citations

Geometrical constraints to reduce complexity in quantum molecular systems

Pérez, Thomas, Patrick Cassam‐Chenaï

The energy of a molecule is not the sum of the energies of its atomic components. Our aim is to compute an accurate approximation of this quantity based on an electron pair model, that is to say by using an antisymmetric product of two-electron wave functions, called “geminals”. In this model, the total wave function can also be described in a more practical way, by a set of matrices, one for each geminal. However, without further restrictions, such a model has a factorial computational complexity with the number of electrons and its applicability is therefore limited to small systems.We have introduced generalized orthogonality constraints between geminals to reduce the computational effort to a polynomial cost, without sacrificing the indistinguishability of the electrons. In other words, the electron pairs corresponding to the geminals are not fully distinguishable, so that our model remains complex, in the sense that geminal products have still to be antisymmetrized according to the Pauli principle to form bona fide electronic wave functions.Our geometrical constraints translate, in terms of geminal matrices, into simple equations involving the traces of products of these matrices. In the simplest non-trivial model, a set of solutions is given by block-diagonal matrices where each block is of size 2x2 and consists of a Pauli matrix or the Identity, multiplied by a complex parameter to be optimized. With this simplified ansatz for geminals, the number of terms in the calculation of the matrix elements of quantum observables is considerably reduced.

•Mar 25, 2019•1 citation

2D-block geminals: Guidelines to choose effective excitations

Patrick Cassam‐Chenaï, Louis Jourdan

A new family of geminal Ansätze called antisymmetrized product of 2D-block geminals (AP2D-BG) has been introduced recently to solve the electronic Schrödinger equation of molecules. They build on an antisymmetrized product of strongly orthogonal geminals (APSG) while lifting the "strong orthogonality" and "seniority zero" restrictions. The combinatorics to select the excitations which can be added to each geminal of the product grows exponentially with the number of geminals. However, in this study on diatomic molecules, we present simple guidelines to drastically reduce the number of excitations, which are useful to lower the electronic energy in variational calculations. Although different regimes to partition the orbital space into 1D- and 2D-blocks in an optimal way are observed along the potential energy curves (PEC), a simple strategy for choosing the block types associated to the geminals in each 2D-block is found successful in all cases. This makes the method quite practical to use. Moreover, the optimal geminals can be easily related to classical Lewis structures of chemistry. The quality of the PECs is assessed by calculating equilibrium geometries and fundamental frequencies; the improvement with respect to APSG is discussed.

The Journal of Chemical Physics•Nov 4, 2025•0 citations

SSpG: A strongly orthogonal geminal method with relaxed strong orthogonality

Brett A. Cagg, Vitaly A. Rassolov

Strong orthogonality is an important constraint placed on geminal wavefunctions in order to make variational minimization tractable. However, strong orthogonality prevents certain, possibly important, excited configurations from contributing to the ground state description of chemical systems. The presented method lifts strong orthogonality constraint from geminal wavefunction by computing a perturbative-like correction to each geminal independently from the corrections to all other geminals. The method is applied to the Singlet-type Strongly orthogonal Geminals variant of the geminal wavefunction. Comparisons of this new SS(p)G method are made to the non-orthogonal AP1roG and the unconstrained Geminal Mean-Field Configuration Interaction method using small atomic and molecular systems. The correction is also compared to Density Matrix Renormalization Group calculations performed on long polyene chains in order to assess its scalability and applicability to large strongly correlated systems. The results of these comparisons demonstrate that although the perturbative correction is small, it may be a necessary first step in the systematic improvement of any strongly orthogonal geminal method.

The Journal of Chemical Physics•Oct 28, 2014•12 citations

A new wavefunction hierarchy for interacting geminals

Peter A. Limacher

A new truncation scheme for non-orthogonal antisymmetrized products of interacting geminals (APIG) is introduced based on antisymmetrized products of strongly orthogonal geminals (APSG). This wavefunction hierarchy of interacting geminals (IG) allows us to gradually increase the accuracy at which the ground state of a seniority-zero Hamiltonian can be estimated, ranging from APSG up to approximation-free APIG. Mathematical expressions for the lowest four orders (IG0, IG1, IG2, and IG3) are given explicitly and the computational cost to evaluate their transition density matrix is verified to scale only cubically with system size. Exemplary numerical calculations indicate that already a very early truncation level leads to results virtually identical to APIG.

The Journal of Chemical Physics•Nov 15, 2016•32 citations

Efficient evaluation of AGP reduced density matrices

Armin Khamoshi, Thomas M. Henderson, Gustavo E. Scuseria

We propose and implement an algorithm to calculate the norm and reduced density matrices (RDMs) of the antisymmetrized geminal power of any rank with polynomial cost. Our method scales quadratically per element of the RDMs. Numerical tests indicate that our method is very fast and capable of treating systems with a few thousand orbitals and hundreds of electrons reliably in double-precision. In addition, we present reconstruction formulas that allow one to decompose higher order RDMs in terms of linear combinations of lower order ones and geminal coefficients, thereby reducing the computational cost significantly.

arXiv:1909.06345•The Journal of Chemical Physics•Nov 11, 2019•41 citations
In our previous study [Kawasaki and Nakatani, Mol. Phys. 123(16), e2449186 (2025)], we proposed the low-rank antisymmetric product of geminals (APG) method, which reconstructs the wavefunction by extracting only the important eigenvalues from the APG wavefunction. However, its practical application was limited by the high computational cost from an orbital optimization process, making higher-rank calculations difficult. In this work, we reformulate the orbital part of the wavefunction using Givens rotation matrices, enabling an analytical treatment of the variational optimization. By combining the low-rank APG with the direct Givens rotation method, we achieved a significant improvement in optimization efficiency. We applied the developed method to small molecular systems and confirmed that it provides high accuracy, while also significantly reducing the computational time compared to conventional methods.
We investigated some variational methods to compute a wavefunction based on antisymmetric product of geminals (APG). The Waring decomposition on the APG wavefunction leads a finite sum of antisymmetrized geminal power (AGP) wavefunctions, each for which the variational principle can be applied. We call this as AGP-CI method which provides a variational solution of the APG wavefunction efficiently. However, number of AGP wavefunctions in the exact AGP-CI formalism become exponentially large in case of many-electron systems. Therefore, we also investigate the low-rank APG wavefunction, in which the coefficient matrices of geminals are factorized by the Schur decomposition. Interestingly, only a few non-zero eigenvalues (up to half number of electrons) were found from the Schur decomposition on the APG wavefunction. We developed some methods to approximate the APG wavefunction by lowering the ranks of coefficient matrices of geminals, and demonstrate their performance. Our new geminal method based on the low-rank decomposition can drastically reduce the number of variational parameters, although no efficient formalism has been elaborated so far, due to the difficulty of optimization caused by the inability to compute analytical derivatives.
The energy of a molecule is not the sum of the energies of its atomic components. Our aim is to compute an accurate approximation of this quantity based on an electron pair model, that is to say by using an antisymmetric product of two-electron wave functions, called “geminals”. In this model, the total wave function can also be described in a more practical way, by a set of matrices, one for each geminal. However, without further restrictions, such a model has a factorial computational complexity with the number of electrons and its applicability is therefore limited to small systems.We have introduced generalized orthogonality constraints between geminals to reduce the computational effort to a polynomial cost, without sacrificing the indistinguishability of the electrons. In other words, the electron pairs corresponding to the geminals are not fully distinguishable, so that our model remains complex, in the sense that geminal products have still to be antisymmetrized according to the Pauli principle to form bona fide electronic wave functions.Our geometrical constraints translate, in terms of geminal matrices, into simple equations involving the traces of products of these matrices. In the simplest non-trivial model, a set of solutions is given by block-diagonal matrices where each block is of size 2x2 and consists of a Pauli matrix or the Identity, multiplied by a complex parameter to be optimized. With this simplified ansatz for geminals, the number of terms in the calculation of the matrix elements of quantum observables is considerably reduced.
A new family of geminal Ansätze called antisymmetrized product of 2D-block geminals (AP2D-BG) has been introduced recently to solve the electronic Schrödinger equation of molecules. They build on an antisymmetrized product of strongly orthogonal geminals (APSG) while lifting the "strong orthogonality" and "seniority zero" restrictions. The combinatorics to select the excitations which can be added to each geminal of the product grows exponentially with the number of geminals. However, in this study on diatomic molecules, we present simple guidelines to drastically reduce the number of excitations, which are useful to lower the electronic energy in variational calculations. Although different regimes to partition the orbital space into 1D- and 2D-blocks in an optimal way are observed along the potential energy curves (PEC), a simple strategy for choosing the block types associated to the geminals in each 2D-block is found successful in all cases. This makes the method quite practical to use. Moreover, the optimal geminals can be easily related to classical Lewis structures of chemistry. The quality of the PECs is assessed by calculating equilibrium geometries and fundamental frequencies; the improvement with respect to APSG is discussed.
Strong orthogonality is an important constraint placed on geminal wavefunctions in order to make variational minimization tractable. However, strong orthogonality prevents certain, possibly important, excited configurations from contributing to the ground state description of chemical systems. The presented method lifts strong orthogonality constraint from geminal wavefunction by computing a perturbative-like correction to each geminal independently from the corrections to all other geminals. The method is applied to the Singlet-type Strongly orthogonal Geminals variant of the geminal wavefunction. Comparisons of this new SS(p)G method are made to the non-orthogonal AP1roG and the unconstrained Geminal Mean-Field Configuration Interaction method using small atomic and molecular systems. The correction is also compared to Density Matrix Renormalization Group calculations performed on long polyene chains in order to assess its scalability and applicability to large strongly correlated systems. The results of these comparisons demonstrate that although the perturbative correction is small, it may be a necessary first step in the systematic improvement of any strongly orthogonal geminal method.
A new truncation scheme for non-orthogonal antisymmetrized products of interacting geminals (APIG) is introduced based on antisymmetrized products of strongly orthogonal geminals (APSG). This wavefunction hierarchy of interacting geminals (IG) allows us to gradually increase the accuracy at which the ground state of a seniority-zero Hamiltonian can be estimated, ranging from APSG up to approximation-free APIG. Mathematical expressions for the lowest four orders (IG0, IG1, IG2, and IG3) are given explicitly and the computational cost to evaluate their transition density matrix is verified to scale only cubically with system size. Exemplary numerical calculations indicate that already a very early truncation level leads to results virtually identical to APIG.
We propose and implement an algorithm to calculate the norm and reduced density matrices (RDMs) of the antisymmetrized geminal power of any rank with polynomial cost. Our method scales quadratically per element of the RDMs. Numerical tests indicate that our method is very fast and capable of treating systems with a few thousand orbitals and hundreds of electrons reliably in double-precision. In addition, we present reconstruction formulas that allow one to decompose higher order RDMs in terms of linear combinations of lower order ones and geminal coefficients, thereby reducing the computational cost significantly.