Critical point search and linear response theory for computing electronic excitation energies of molecular systems. Part II: CASSCF

Laura Grazioli, Yukuan Hu, Tommaso Nottoli, Filippo Lipparini, Eric Cancès · 2026

Read Critical point search and linear response theory for computing electronic excitation energies of molecular systems. Part II: CASSCF by Laura Grazioli, Yukuan Hu, Tommaso Nottoli, Filippo Lipparini, Eric Cancès (2026) 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-2604-13753.txt.

Introduction

Grazioli and four colleagues extend a geometric method for electronic excitation levels to CASSCF theory. Excited states are the higher energy levels of molecules. They carry the Kahler manifold framework from Part I into CASSCF, linking time-dependent equations to excited-state methods. CI and orbital coupling complicates the geometry. They derive linear response equations directly and build a state-specific method needing only first derivatives. Tests on water, formaldehyde, and ethylene show it works. The method is built on shapes, not on brute force.

Catalog record

CREATORLaura Grazioli, Yukuan Hu, Tommaso Nottoli, Filippo Lipparini, Eric Cancès
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74437
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.13753

Browse the full library · About the Tower of Babel

Full text

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

Critical point search and linear response theory for computing electronic excitation energies of molecular systems. Part II: CASSCF Laura Grazioli, Yukuan Hu, Tommaso Nottoli, Filippo Lipparini, Eric Cancès Abstract The computation of excited states within the Complete Active Space Self-Consistent Field (CASSCF) framework remains a significant challenge in quantum chemistry, both theoretically and algorithmically. In this work, we extend the Kähler manifold formalism introduced in Part I of this series to the CASSCF theory, and draw a geometrical connection from the time-dependent CASSCF equations to state-specific and linear response methodologies for excited states. This is achieved by first investigating the underlying CASSCF manifold and identifying its Kähler structure, which is complicated by the nontrivial coupling of CI and orbital degrees of freedom. Building on these theoretical findings, we derive the CASSCF linear response equations in a straightforward manner, and develop a robust state-specific method that relies solely on first-order derivatives of the CASSCF energy functional. Numerical results on representative molecular systems—water, formaldehyde, and ethylene—demonstrate the effectiveness of the proposed state-specific method, while revealing the difficulty of reliable identification of excited states due to nonlinearity induced by the CASSCF theory. I Introduction The Complete Active Space Self-Consistent Field (CASSCF) is a powerful method to qualitatively describe strongly correlated molecular systems. 41 ; 40 ; 39 ; 58 ; 46 ; 42 It offers access to both ground and excited states, and is therefore a popular choice for investigating photophysical and photochemical processes. However, applying CASSCF suffers from many limitations. The two most apparent problems of CASSCF are its computational cost, given the exponential scaling of the underlying CASCI problem, and the arbitrariness introduced by the choice of the active space. In other words, CASSCF calculations are sometimes more an art than a science, even though a lot of progress has been made to formulate robust algorithms 57 ; 56 ; 24 ; 23 ; 50 ; 28 and automated active space selection strategies. 38 ; 51 ; 49 ; 45 There are, however, several other aspects that make CASSCF a method that is far from being straightforward or black-box. From a numerical point of view, the optimization of the CASSCF wavefunction is a challenging problem even for ground-state calculations. This is mainly due to the fact that CASSCF requires one to optimize at the same time both the molecular orbitals and the CI coefficients. These are two very different tasks, the former being a highly nonlinear optimization problem of moderate size, and the latter being a very high-dimensional linear eigenvalue problem. What really makes CASSCF optimization hard is that the two tasks are not independent, and the subtle interplay between the two sets of degrees of freedom can lead to poor convergence or instabilities. 57 ; 56 Computing excited states is an even more challenging task. As CASSCF is a nonlinear method, the definition of excited state itself is not straightforward, i.e., there is no corresponding linear and self-adjoint Hamiltonian operator whose eigenfunctions are the excited states. The most popular approach, state-averaged CASSCF, 55 optimizes the orbitals for an average of CASCI electronic states that are obtained by seeking the few lowest eigenvalues of the CASCI Hamiltonian. State-averaged CASSCF has the advantage of being relatively straightforward to implement starting from an available ground-state CASSCF implementation, but limits the flexibility of the method by relegating the treatment of the electronic excitation to the active space, and produces orbitals that are not optimal for any particular state. 18 Nevertheless, it represents a good compromise between cost-effectiveness, simplicity, and accuracy. Two more approaches can be used to compute excited states with CASSCF. State-specific calculations 17 ; 20 ; 23 ; 24 ; 52 ; 53 can be performed, where one optimizes a CASSCF wavefunction that corresponds to a saddle point of the CASSCF energy as a function of the variational parameters. Although this sounds like a straightforward definition, it comes with some problems. First, it is not obvious that all the saddle points of the CASSCF energy correspond to physical states, due to the nonlinearity of the model. Second, saddle point optimizations are, in general, more difficult than minimizations. Another popular approach is given by linear response theory, 13 ; 15 ; 19 ; 25 ; 37 ; 63 where excitation energies are obtained from poles of the molecular linear response function. This is a rigorous and well-defined approach, but the so-obtained excitation energies may strongly depend on the CASSCF ground state around which the dynamics is linearized. Furthermore, it can be computationally demanding, and the numerical methods for solving linear response equations can encounter convergence difficulties, especially if many states are sought. 19 ; 3 ; 36 Recently, real-time time-dependent CASSCF has been proposed as a method to investigate electronic dynamics. Electronic excitation energies, and even whole spectra, can also be obtained from real-time simulations by analyzing appropriate correlation functions. 32 ; 44 ; 43 In summary, not only is CASSCF a complex method per se , but its extension to excited states introduces significant complications, even speaking of the strategy used to define them. In this paper, we aim to shed some light on the method by exploring and clarifying the geometrical structure of CASSCF, and by using such a structure to connect, in a clear, unambiguous way, the CASSCF dynamics (i.e., time-dependent formulation) to the solution of the CASSCF equations for ground and excited states in a state-specific context, and to CASSCF linear response theory. We achieve these by recognizing and exploring the Kähler structure of the CASSCF manifold, which provides a clear path to deriving time-dependent CASSCF equations. The Kähler formalism was introduced by Erich Kähler in the 1930s to unify holomorphic geometry, Riemannian geometry and Hamiltonian dynamics. 26 It also plays a key role in theoretical physics, for example in Calabi-Yau manifolds, 61 supersymmetry and string theory. 16 Building on the Kähler structure, we provide a new, more geometrical way of deriving CASSCF linear response equations, corresponding to the linearization of the CASSCF dynamics around the CASSCF ground state. We also show that the stationary points of the CASSCF dynamics correspond to (state-specific) ground and excited states, and develop a constrained saddle-search method, which robustly locates the saddle points on the CASSCF energy landscape with any given Morse index and requires only first-order derivatives of the energy functional. Understanding the geometrical structure of CASSCF is not straightforward. If one decouples the CI problem from the orbital optimization, one obtains two problems for which the geometrical structure is already well defined. The CI problem corresponds in fact to the minimization of a quadratic function on the unit sphere, which leads to the CASCI linear eigenvalue problem. The orbital optimization problem, on the other hand, corresponds to a highly nonlinear optimization problem on a flag manifold, which is more complex, but already well understood. 62 ; 54 The real challenge is to understand the full picture, where both the orbitals and CI coefficients are variable, as the resulting manifold is not just the simple union of the two aforementioned manifolds, but a more complex entity that allows for the subtle interplay of the two sets of degrees of freedom, which is what makes CASSCF such a complicated method. We note that

Read the complete text