How an Equi-ensemble Description Systematically Outperforms the Weighted-ensemble Variational Quantum Eigensolver

Akilan Rajamani, Martin Beseda, Benjamin Lasorne, Bruno Senjean · 2025

Read How an Equi-ensemble Description Systematically Outperforms the Weighted-ensemble Variational Quantum Eigensolver by Akilan Rajamani, Martin Beseda, Benjamin Lasorne, Bruno Senjean (2025) 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-2509-17982.txt.

Introduction

Akilan Rajamani and 3 co-authors show that an even mix beats a weighted mix for quantum chemistry sums. Their 2025 letter deals with excited states of molecules, which drive how light changes matter. Quantum solvers for such states rest on a group form of the classic lowest-energy rule.

Two group styles are compared head to head. In the weighted style, states count by set weights. In the equi style, states count the same. Tests lead the team to back the equi style as the way to go. The claim rests on cost weighed against how close the sums land.

Catalog record

CREATORAkilan Rajamani, Martin Beseda, Benjamin Lasorne, Bruno Senjean
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE51 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-17982

Browse the full library · About the Tower of Babel

Full text

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

How an Equi-ensemble Description Systematically Outperforms the Weighted-ensemble Variational Quantum Eigensolver Akilan Rajamani, Martin Beseda, Benjamin Lasorne, Bruno Senjean Abstract Calculating excited states in chemistry is crucial to provide insight into photoinduced molecular behavior beyond the ground state, enabling innovations in spectroscopy, material sciences, and drug design. While several approaches have been developed to compute excited-state properties, finding the best ratio between computational cost and accuracy remains challenging. The advent of quantum computers brings new perspectives, with the development of quantum algorithms that promise an advantage over classical ones. Most of these new algorithms are inspired from previous classical ones, but with different pros and cons. In this Letter, we focus on the generalization of the variational principle for many-body excited-states that led to the ensemble variational quantum eigensolver (VQE). We compare the performance of two ensemble VQE approaches, the equi-ensemble and weighted-ensemble ones, and conclude that the equi-ensemble is the way to go. Let us consider an ensemble of K initial orthonormal states \{\ket{\Phi_j}\}_{j=0,\ldots,K-1} transformed by the same variationally parametrized unitary operator \hat{U}({\bm{\theta}}) . We define the ( {\bf w} -weighted) ensemble energy as Theophilou 1979 ; Gross et al. 1988a \displaystyle E_{C}^{{\bf w}}({\bm{\theta}}) \displaystyle= \displaystyle\sum_{j=0}^{K-1}w_{j}\bra{\Phi_j}\hat{U}^{\dagger}({\bm{\theta}})\hat{H}\hat{U}({\bm{\theta}})\ket{\Phi_j} (1) \displaystyle\geq\sum_{j=0}^{K-1}w_{j}E_{j}, where w_{j}\geq w_{j+1} , \hat{H} is the electronic structure Hamiltonian, and E_{j}\leq E_{j+1} its exact eigenvalues. The inequality in Eq. ( 1 ) reflects the variational principle generalized to an ensemble of states, such that minimizing the ensemble energy with respect to the parameters {\bm{\theta}} will lead to \hat{U}({\bm{\theta}})\ket{\Phi_j}=\ket{\Psi_j(\bmtheta)}\sim\ket{\Psi_j} , \ket{\Psi_j} being the j th eigenvector of \hat{H} with eigenvalue E_{j} , assuming that the unitary \hat{U}({\bm{\theta}}) can achieve this goal. This generalized variational principle has been used to extract excited-state properties from ensemble density functional theory Gross et al. 1988b ; Filatov and Shaik 1999 ; Kazaryan et al. 2008 ; Filatov 2015 ; Filatov et al. 2015 ; Filatov et al. 2017 ; Liu et al. 2021 ; Filatov et al. 2021 ; Gould and Pittalis 2017 ; Gould et al. 2018 ; Gould and Pittalis 2019 ; Gould and Pittalis 2020 ; Gould 2020 ; Gould et al. 2020 ; Gould and Kronik 2021 ; Gould et al. 2021 ; Sagredo and Burke 2018 ; Senjean et al. 2015 ; Senjean et al. 2016 ; Deur et al. 2018 ; Deur and Fromager 2019 ; Senjean and Fromager 2018 ; Marut et al. 2020 ; Cernatic et al. 2022 ; Cernatic and Fromager 2024 ; Scott et al. 2024 ; Cernatic et al. 2024a ; Dupuy and Fromager 2025 ; Fromager 2025 ; Gould 2025 ; Gould et al. 2025 ; Daas et al. 2025 and more recently in the context of the one-particle reduced density-matrix functional theory Schilling and Pittalis 2021 ; Liebert et al. 2021 ; Liebert and Schilling 2023a ; Liebert and Schilling 2023b , variational Monte–Carlo Schautz and Filippi 2004 ; Wheeler et al. 2024 and density-matrix embedding methods Cernatic et al. 2024b . Turning to multiconfigurational approaches, the generalized variational principle is used in conjunction with configuration interaction (CI) and multiconfigurational self-consistent field (MCSCF) approaches Helgaker et al. 2014 . For the latter, the variational principle is used not only for the linear many-body variational problem (CI/MC) but also for the nonlinear one-body variational orbital-optimization problem (SCF). In general, the equi-ensemble ( i.e. , w_{j}=w_{j+1} , \forall j ) version of Eq. ( 1 ) is considered for the SCF orbital optimization to provide a democratic description of all the MC states of the ensemble, which is especially important when vibronic couplings are large, i.e. , in the vicinity of avoided crossings and conical intersections Domcke et al. 2004 , thus leading to the so-called state-average MCSCF (SA-MCSCF). Recently, the quantum computing analog of the SA-MCSCF method has been derived and called state-average orbital-optimized variational quantum eigensolver (SA-OO-VQE) Yalouz et al. 2021 ; Yalouz et al. 2022 ; Beseda et al. 2024 ; Illésová et al. 2025 , inspired by the previous work of Nakanishi and coworkers on the subspsace-search VQE Nakanishi et al. 2019 , which is equivalent but without orbital optimization. SA-OO-VQE is analogous to SA-MCSCF as regards the state-average orbital optimization problem, which is performed classically. However, while SA-MCSCF always uses an actual – full or partial, but supposedly exact (numerically) for the targeted ensemble – diagonalization algorithm for the many-body part ( i.e. , for \hat{U}({\bm{\theta}}) ), SA-OO-VQE uses an approximate parameterized-circuit ansatz \hat{U}({\bm{\theta}}) to minimize the ensemble energy. How this ansatz – for which several different expressions have been derived Peruzzo et al. 2014 ; Yung et al. 2014 ; Barkoutsos et al. 2018 ; Romero et al. 2018a ; Lee et al. 2019 ; Evangelista et al. 2019 ; Sokolov et al. 2020 ; Khamoshi et al. 2022 ; Smart and Mazziotti 2021 ; Marécat et al. 2023 ; Materia et al. 2024 ; Kandala et al. 2017 ; Ganzhorn et al. 2019 ; O’Gorman et al. 2019 ; Burton 2024 ; Gard et al. 2020 ; Seki et al. 2020 ; Anselmetti et al. 2021 ; Lacroix et al. 2023 ; Ruiz Guzman and Lacroix 2024 – can actually achieve diagonalization without any convergence issues is one of the biggest challenges of VQE McClean et al. 2018 ; Bittel and Kliesch 2021 ; D’Cunha et al. 2023 ; Larocca et al. 2025 ; Leone et al. 2024 . Such a difference is also not without consequences as regards the choice of the initial orthonormal states \{\ket{\Phi_j}\}_{j=0,\ldots,K-1} , and has led to a revival of the use of ensemble theory to extract many-body excited states in the context of quantum computing Nakanishi et al. 2019 ; Parrish et al. 2019 ; Yalouz et al. 2021 ; Yalouz et al. 2022 ; Omiya et al. 2022 ; Bierman et al. 2022 ; Benavides-Riveros et al. 2022 ; Xu et al. 2023 ; Heya et al. 2023 ; Beseda et al. 2024 ; Chen et al. 2024 ; Bierman et al. 2024 ; Benavides-Riveros et al. 2024 ; Hong et al. 2024 ; Ding et al. 2024 ; Illésová et al. 2025 . Let us now discuss in more details the role of the weights {\bf w} in Eq. ( 1 ). To converge towards the eigenvectors of the problem, while the weights {\bf w} do not need to take particular values in principle apart from being sorted in descending order ( w_{j}>w_{j+1} ), they can have a huge impact in practice. Indeed, the choice of the weights can strongly influence the parameter landscape and impact the minimization of the cost function E_{C}^{\bf w}({\bm{\theta}}) , as shown recently in Refs. 83 and 84 . Interestingly, they also have an impact in ensemble DFT when approximate Hartree-exchange-correlation functionals of the density are used Senjean et al. 2015 , such that they are sometimes taken as Boltzmann weights (with a fictitious temperature arbitrarily fixed by the user), thus leading to thermal DFT Pittalis et al. 2011 ; Pribram-Jones and Burke 2016 ; Pastorczak et al. 2013 . Note that Boltzmann weights may also be used in SA-MCSCF to avoid active-space discontinuities with respect to a finite adiabatic target subspace at various geometries Olsen 2015 . Along the lines of the aforementioned

Read the complete text