Freeze-and-release direct optimization method for variational calculations of excited electronic states
Yorick L. A. Schmerwitz, Elli Selenius, Gianluca Levi · 2025
Read Freeze-and-release direct optimization method for variational calculations of excited electronic states by Yorick L. A. Schmerwitz, Elli Selenius, Gianluca Levi (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-2501-18568.txt.
Introduction
Excited states of molecules sit on saddle points of the energy landscape. Direct variational searches tend to slide off them. A 2025 arXiv paper by Yorick Schmerwitz, Elli Selenius, and Gianluca Levi offers a two step fix. Freeze the orbitals taking part in the excitation, settle the rest, then release everything into a saddle search.
Both steps cost no more than a ground state run. On charge transfer states in single molecules and pairs, the method avoided the slide into false, spread out solutions. That slide breaks the standard overlap approach. The method also kept the right energy trend as paired molecules parted, without costly corrections.
Catalog record
| CREATOR | Yorick L. A. Schmerwitz, Elli Selenius, Gianluca Levi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-18568
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Freeze-and-release direct optimization method for variational calculations of excited electronic states Yorick L. A. Schmerwitz, Elli Selenius, Gianluca Levi Abstract Variational optimization of orbitals in time-independent density functional calculations of excited electronic states presents a significant challenge, as excited states typically correspond to saddle points on the electronic energy landscape. The optimization can be particularly difficult if the excitation involves significant rearrangement of the electron density, as for charge transfer excitations. A simple strategy for variational orbital optimization of excited states is presented. The approach involves minimizing the energy while freezing the orbitals directly involved in the excitation, followed by a fully unconstrained saddle point optimization. Both steps of this freeze-and-release strategy are carried out using direct optimization algorithms with the same computational scaling as ground state calculations. The performance of the method is extensively assessed in calculations of intramolecular and intermolecular charge transfer excited states of organic molecules and molecular dimers using a generalized gradient approximation functional. It is found that the freeze-and-release direct optimization approach can avoid variational collapse to spurious, charge-delocalized solutions for cases where conventional algorithms based on the maximum overlap method fail. For intermolecular charge transfer, the orbital-optimized calculations are found to provide the correct dependency of the energy on the donor-acceptor separation without requiring long-range exact exchange, something common time-dependent density functional theory approaches fail to achieve. † † affiliation: Max-Planck-Institut für Kohlenforschung, 45470 Mülheim an der Ruhr, Germany † † email: [email protected] † † affiliation: Science Institute of the University of Iceland, 107 Reykjavík, Iceland † † affiliation: Department of Chemical and Pharmaceutical Sciences, University of Trieste, 34127 Trieste, Italy † † affiliation: Science Institute of the University of Iceland, 107 Reykjavík, Iceland † † email: [email protected] † † suppinfo: 1 Introduction Accurate modeling of electronic excitations in large molecular chromophores and photoactive materials is essential for understanding phenomena such as photosynthesis and vision, as well as for developing efficient and sustainable solar energy conversion devices. Photoinduced processes typically involve multiple states with different electronic character, with charge transfer excited states often playing a key role. Thus, simulating photoinduced processes relevant to both natural phenomena and solar energy devices requires electronic structure methods that are both computationally affordable and can reliably describe states of different electronic character across different molecular geometries. While numerous wave-function-based and density functional theory (DFT) approaches for excited state calculations exist, achieving the required accuracy, and efficiency for routine application remains a fundamental challenge. This is particularly the case for charge transfer excitations Knepp et al. (2025) , which are characterized by a large rearrangement of the electron density with respect to that of the ground state. As a result, conventional methods typically hold limited predictive power in simulations of the photoinduced dynamics of electrons and nuclei Mukherjee et al. (2024) ; Janoš and Slavíček (2023) , calling for continued development of efficient excited state methodologies with sufficient accuracy. The most favorable balance between efficiency and accuracy is typically achieved in density-functional-based calculations. These calculations are most efficient when employing local or semilocal Kohn-Sham (KS) Kohn and Sham (1965) ; Hohenberg and Kohn (1964) functionals, which approximate the electron-electron exchange and correlation through density-based exchange-correlation (XC) functionals. However, recent algorithmic advances have reduced the computational cost of more elaborate hybrid functionals as well, which incorporate a portion of non-local exact exchange explicitly Helmich-Paris et al. (2021) ; D’Antoni et al. (2023) ; Rebolini et al. (2016) . Within the density functional framework, excited states are most frequently modelled using time-dependent DFT (TDDFT) Runge and Gross (1984) ; Hohenberg and Kohn (1964) . Practical implementations of TDDFT are usually based on linear-response perturbation theory applied to the time-dependent KS equations Casida (1995) , using ground state functionals within the adiabatic approximation Maitra (2016) . While comparable in efficiency to the time-independent ground state DFT formalism, the approximations used in TDDFT significantly limit its applicability in calculations of excited states involving large rearrangement of the electron density, including charge transfer Mester and Kállay (2022) ; Dev et al. (2012) ; Ghosh et al. (2015) ; Dreuw and Head-Gordon (2004) and Rydberg Cheng et al. (2008) excitations. In particular, TDDFT tends to underestimate the excitation energy of charge transfer excitation and fails to provide the correct, 1/R , variation of the energy as a function of the separation R between donor and acceptor fragments Mester and Kállay (2022) ; Dev et al. (2012) ; Ghosh et al. (2015) ; Dreuw and Head-Gordon (2004) . These failures are rooted in the lack of derivative discontinuity in standard KS functionals as well as the lack of divergence of the adiabatic XC kernel as R\rightarrow\infty Maitra (2022) ; Hellgren and Gross (2012) . Nonadiabatic approximations offer a promising route to address these flaws of TDDFT, but are still in the early stages of development Lacombe and Maitra (2023) . Range-separated hybrid functionals with optimally tuned range-separation parameters Stein et al. (2009) , which recover full exact exchange at long range, can also improve the description of charge-transfer excitations in TDDFT Bogo and Stein (2024) . The challenge there, however, is that the optimal tuning is system-specific, may not generalize across different states, and can introduce discontinuities in potential energy surfaces. Bogo et al. (2025) ; Maitra (2022) Indeed, a generally applicable TDDFT approach for treating charge transfer excitations is not yet available. The limitations of practical TDDFT implementations have spurred the development of time-independent DFT formalisms for excited states within both ensemble Cernatic and Fromager (2024) ; Gould et al. (2021) ; Cernatic et al. (2022) ; Gould et al. (2018) ; Yang et al. (2017) ; Theophilou (1979) and state-specific Ayers et al. (2015) ; Ayers et al. (2012) ; Ayers and Levy (2009) ; Levy and Nagy (1999) ; Görling (1999) ; Perdew and Levy (1985) frameworks. Recently, the first state-specific Loos (2025) ; Gould et al. (2025) and ensemble Gould and Pittalis (2024) density functionals designed for variationally optimized excited states have been developed, representing an important step forward in excited state DFTs. Despite these theoretical advancements, time-independent excited state density functional calculations are typically based on a practical orbital-optimized (OO) approach Hait and Head-Gordon (2021) , sometimes referred to as \Delta Self-Consistent Field ( \Delta SCF), where excited states are found as nonaufbau solutions to the time-independent KS equations beyond the lowest energy (ground state) solution using ground state XC functionals. OO-KS calculations of excited states find partial justification in adiabatic TDDFT, where nonaufbau KS solutions correspond to stationary densities Kowalczyk et al. (2011) . More recently, Gould Gould (2025) and Fromager Fromager (2025) have shown that the approach can be derived from an exact stationarity condition for ground and excited states with respect to the ensemble density when the ensemble XC functional is constructed from the regular ground state functional, providing further theoretical underpinning. Yang and Ayers Yang and Ayers (2024) have also recently provided a rationale based on functionals that depend on the KS potential. While these recent developments provide a theoretical foundation, the growing interest in OO density functional methods for excited states has largely been driven by their practical success. Compared to TDDFT, these methods usually strike a more balanced description of ground and excited states as