Transformation-free generation of a quasi-diabatic representation from the state-average orbital-optimized variational quantum eigensolver

Silvie Illésová, Martin Beseda, Saad Yalouz, Benjamin Lasorne, Bruno Senjean · 2025

Read Transformation-free generation of a quasi-diabatic representation from the state-average orbital-optimized variational quantum eigensolver by Silvie Illésová, Martin Beseda, Saad Yalouz, 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-2502-18194.txt.

Introduction

Five authors show that a recent quantum chemistry algorithm hands over a prized representation almost for free. Chemists who model molecules changing shape need a diabatic picture, where states keep their character as atoms move, instead of the natural adiabatic one where they swap. The algorithm, a state-averaged, orbital-optimized variational eigensolver run on quantum hardware, behaves like a classic multi-configuration method. Viewed as a gentlest-possible reshuffling of its blocks, its raw output already sits close to diabatic.

The team defines gauges for how diabatic a representation is, perfect and leftover, and reworks the algebra of the ideal form. They measure how far one may stray before the coupling terms that drive shape change suffer. A test case, formaldimine, with its famous crossing of ground and excited states, carries the numbers.

Catalog record

CREATORSilvie Illésová, Martin Beseda, Saad Yalouz, Benjamin Lasorne, Bruno Senjean
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE132 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-18194

Browse the full library · About the Tower of Babel

Full text

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

Transformation-free generation of a quasi-diabatic representation from the state-average orbital-optimized variational quantum eigensolver Silvie Illésová, Martin Beseda, Saad Yalouz, Benjamin Lasorne, Bruno Senjean Abstract In the present work, we examine how the recent quantum-computing algorithm known as the state-average orbital-optimized variational quantum eigensolver (SA-OO-VQE), viewed within the context of quantum chemistry as a type of multiconfiguration self-consistent field (MCSCF) electronic-structure approach, exhibits a propensity to produce an ab initio quasi-diabatic representation “for free” if considered as a least-transformed block-diagonalization procedure, as alluded to in our previous work [S. Yalouz et al. , J. Chem. Theory Comput. 18 (2022) 776] and thoroughly assessed herein. To this end, we introduce intrinsic and residual descriptors of diabaticity and re-explore the definition and linear-algebra properties – as well as their consequences on the vibronic nonadiabatic couplings – of an optimal diabatic representation within this context, and how much one may deviate from it. Such considerations are illustrated numerically on the prototypical case of formaldimine, which presents a well-known conical intersection between its ground and first-excited singlet electronic states. Keywords: quantum computing; quantum algorithms; ensemble variational principle; diabatization; conical intersection; nonadiabatic couplings. I Introduction Ultrafast (subpicosecond) photochemical and photophysical ‘energy/charge/matter/information’-transfer processes induced upon light absorption by molecules within the UV-visible domain are ubiquitous. Subsequent radiationless processes, in particular, are governed by so-called vibronic nonadiabatic couplings (NACs) among several adiabatic electronic states. They are mediated by molecular vibrations and how these affect the electronic system when the nuclei move from their equilibrium positions. In other words, vibronic NACs – when they are large – imply a theoretical description that goes beyond the Born-Oppenheimer approximation (BOA) [ 1 , 2 , 3 , 4 ] . Their effects have been observed and rationalized, both for time-resolved experiments (femtochemistry; see, e.g. , Ref. [ 5 ] ) and energy-resolved ones (photoabsorption or photoelectron spectroscopy; see, e.g. , Refs. [ 6 , 7 ] ). Any faithful theoretical/computational description (modelling and simulation) of such processes beyond BOA requires – in principle – a quantum treatment of the nuclear motion (molecular quantum dynamics) [ 8 , 9 ] because the vibrational system experiences a dynamical regime that is “as quantum” as that of the electronic system as regards energy and time scales. This has been one of the main incentives for the rise of nuclear wavepacket methods within the field of theoretical chemistry over the past few decades, incarnated in particular with the now golden standard known as the ‘multiconfiguration time-dependent Hartree’ (MCTDH) method [ 10 , 11 ] . Meanwhile, alternative approximate treatments relying on statistical ensembles of nuclear trajectories attached to quantum superpositions of electronic states, rather than molecular wavepackets – known as mixed quantum-classical and sometimes semiclassical – have been devised over the years. The most popular one is perhaps the Tully ‘trajectory surface hopping’ (TSH) approach (see, e.g. , Ref. [ 12 ] for a recent review comparing various implementations of the NACs). In practice, both types of approaches exhibit very different interfacial connections toward electronic-structure computations. While grid-based wavepacket methods rely on the availability, hence the preconstruction, of a global functional representation of the electronic Hamiltonian matrix at any molecular geometry (also known as a ‘fit’, much as in a parameterized force field), trajectory-based methods – known as ‘nonadiabatic molecular dynamics’ (NAMD) [ 13 ] – can benefit from direct, ab initio , local evaluations of energy derivatives and NACs on-the-fly, within a spirit similar to ‘Born-Oppenheimer molecular dynamics’ (BOMD) or ‘ ab initio molecular dynamics’ (AIMD) [ 14 ] . A rich variety of intermediate approaches, aiming at combining the pros and cons of the two fully-quantum and mixed quantum-classical extremes, often Gaussian-based, have been proposed over the past decades. This has become a very active field of research, with the common aim of proposing a rational hierarchy of methods [ 4 , 15 , 16 , 17 ] . A full exposition of them is beyond the scope of the present work. Now, in practice, it must be understood that wavepacket methods – either grid-based or Gaussian-based – are quantum, hence not pointwise with respect to the nuclear coordinates. They thus involve eventually the evaluation of multidimensional Hamiltonian matrix elements. As such, any singular or ill-behaved operator is not to be welcome, being a potential source for numerical difficulties upon local differentiation and/or integration over a spatial distribution. This is the very reason why so-called (quasi)diabatic representations have become preferred in this context. The problem is well-known and concerns in particular the treatment of tamed vibronic NACs. This aspect is one of the main focuses of the present work. In a nutshell, the most evident representation of the electronic problem is the adiabatic one (eigensolutions of the electronic Hamiltonian at any molecular geometry). This is the very purpose of quantum chemistry. However, when two well-defined adiabatic eigenstates become degenerate at a certain geometry, known as a conical intersection, which is a common situation of physical interest for radiationless decay, they exhibit a divergent NAC between them and the problem becomes mathematically ill-defined. A solution to this consists of invoking a global geometry-dependent unitary change of basis (known as the ‘adiabatic-to-diabatic transformation’ – ADT), which is meant to generate a well-behaved electronic basis set with vanishing NACs at all molecular geometries. Such an ADT is not unique and rarely exists globally in a strict sense. This has been the subject of some abundant literature in the field during the 1980s (see, e.g. , Ref. [ 1 ] and references therein for a quite exhaustive compilation). A plethora of practical approximate solutions has been proposed over the years, typically known as ‘quasi-diabatizations’. A complete overview is beyond the scope of the present work. Let us mention that there are essentially three types of approaches: (i) NAC-based, with line-integration of the vector criterion of “diabaticity” (see, e.g. , Ref. [ 18 ] for a recent implementation in the context of Gaussian-based direct quantum dynamics); (ii) property-based, with smooth and minimal variations of expectation values of operators (observables), which somewhat includes the famous “diabatization by ansatz” where the electronic Hamiltonian matrix elements are sought from the onset as regular functions (low-degree polynomials) of the nuclear coordinates (see, e.g. , Refs. [ 6 , 7 ] for two milestone illustrations and [ 19 ] for a recent extension making use of neural-network techniques); (iii) wavefunction-based, according to a maximal overlap criterion of the electronic states at various molecular geometries (see, e.g. , Refs. [ 20 , 21 ] for recent considerations about the Householder compression with valence-bond wavefunctions and some related approaches for the automatization of the global phase-consistency issue). In the present work, we shall investigate an approach of the third kind. It strongly relies on the concept of a least-transformed block-diagonalization, which was first proposed as an avenue by Cederbaum et al. in Refs. [ 22 , 23 ] and further explored in practice by Werner et al. in Refs. [ 24 , 25 ] . The formal basis invokes a constrained singular-value decomposition (SVD) between the model and target subspaces. Its connection with effective Hamiltonian theory and quasidegenerate perturbation theory (QDPT) [ 26 , 27 , 28 ] is evident and gave rise to strong

Read the complete text