Ab initio perspectives on surface photocatalysis: advances, challenges, and opportunities

Abstract

Surface photocatalysis holds significant promise for converting solar energy into chemical fuels and addressing environmental challenges. While ab initio calculations provide critical insights into the thermodynamic and kinetic aspects of catalytic reactions, applying these methods to surface photocatalysis remains challenging. In this work, we discuss the key challenges that need to be addressed when using ab initio calculations to understand surface photocatalytic processes, the reasons behind these challenges, and the potential directions and opportunities for overcoming them in the future.

---

Introduction

The increasing reliance on fossil fuels has led to severe environmental challenges, including air pollution and global climate change driven by greenhouse gas emissions. Consequently, the development of clean and renewable energy sources, such as solar energy, is essential for achieving energy sustainability. A fundamental scientific challenge in this pursuit is the efficient conversion of solar energy into other usable forms. Among various solar energy conversion processes, photocatalysis, fundamentally, facilitates the conversion of light energy into chemical energy, enabling light-driven chemical reactions. This process plays a crucial role in addressing global energy and environmental challenges, as it underpins a wide range of applications, including solar fuel production, environmental remediation, and green chemical synthesis.

Thanks to extensive research efforts, photocatalytic reactions have been successfully observed in various systems, including metal oxide-based photocatalysts,^1–4 metal single-atom and cluster photocatalysts,^5,6 metal-free organic photocatalysts,^7,8 semiconductor quantum dot photocatalysts,^9,10 metal–organic framework (MOF) based photocatalysts,^11,12 and hybrid heterojunction photocatalytic systems.^13,14 These systems have demonstrated promising photocatalytic activity for applications such as water splitting, CO₂ reduction, and environmental remediation.^3,9,11 However, despite notable progress, most photocatalytic reactions remain confined to laboratory-scale studies. A major challenge hindering large-scale implementation of photocatalysis is the insufficient efficiency of most photocatalytic reactions. To enhance photocatalytic efficiency, it is crucial to gain a deeper understanding of the underlying physical mechanisms governing the light-driven processes. At this stage, ab initio calculations play a pivotal role in elucidating the fundamental physics of photocatalytic reactions, providing valuable insights into charge carrier dynamics, reaction pathways, and materials design strategies for optimizing photocatalytic performance.

Ab initio calculations are playing an increasingly critical role in various branches of physical chemistry, particularly in the field of catalysis, where the ab initio calculations provide valuable insights from both thermodynamic and kinetic perspectives. Thermodynamic analysis provides insights into the spontaneity of a reaction through changes in free energy,¹⁵ while kinetic studies enable the exploration of reaction mechanisms, estimation of rate constants, and identification of the rate-determining step.¹⁶ While ab initio calculations have been successful in studying different catalytic reactions, applying them to photocatalysis, especially photocatalysis on solid surfaces, remains challenging. In this work, we discuss the key challenges that need to be addressed when using ab initio calculations to understand surface photocatalytic processes, the reasons behind these challenges, and the potential directions and opportunities for overcoming them in the future.

Key steps of surface photocatalysis

Surface photocatalysis typically occurs on semiconductor surfaces, including traditional semiconductor materials such as TiO₂, CdS, ZnO, and BiVO₄,^2,17–19 as well as two-dimensional (2D) semiconductors like MoS₂ and C₃N₄.^20,21 On one hand, semiconductors have bandgaps that allow them to absorb light and generate e–h pairs. On the other hand, semiconductor surfaces often contain active sites that facilitate molecular adsorption and reactions. As illustrated in Fig. 1, which schematically depicts the three key steps in surface photocatalysis, a comprehensive understanding of these processes is essential for optimizing photocatalytic efficiency on surfaces. In the first step, photoabsorption leads to the generation of e–h pairs, making it crucial to enhance the efficiency of this excitation process. The second step involves the migration of photoexcited carriers to the surface, where they can be captured by the adsorbed molecules. The key factor in this process is the excited carrier dynamics, as photoexcited electrons and holes must reach the surface before recombination occurs. Finally, in the third step, surface chemical reactions take place when the adsorbed molecules capture electrons or holes, entering an excited state. Because the excited state potential energy surface usually differs from the ground state, reaction barriers may be reduced, facilitating photochemical reactions. Additionally, the hot carriers transfer their energy to the surface and molecules via electron–phonon (e–ph) coupling, inducing thermal effects, and possible influences from the electromagnetic field of light can further affect surface chemical reactions. Next, we will discuss the role of ab initio calculations in three key aspects: photoabsorption, excited carrier dynamics, and surface photochemical reactions.

Fig. 1 Schematic of key steps in surface photocatalysis: (I) generation of e–h pairs upon photoabsorption; (II) migration of photoexcited carriers to the surface; and (III) trapping of photoexcited carriers by the adsorbates (A and D represent electron acceptors and donors, respectively).

Photoabsorption

Light absorption is the first and most fundamental step in surface photocatalysis, where incident photons generate electron–hole (e–h) pairs within a solid material. The primary challenge in this process lies in maximizing the efficiency of solar light absorption. Since most surface photocatalytic reactions occur on semiconductor materials, the light absorption properties are inherently tied to the band gap of the semiconductor. When the photon energy is lower than the band gap of the semiconductor, photoabsorption is significantly suppressed, limiting the material's ability to harness solar energy for photocatalysis. Consequently, tuning the band gap to extend the absorption range into the visible and near-infrared regions has been a major focus in photocatalysis research.

Since the 1990s, many efforts have been devoted to modifying the band gap of semiconductors through defect engineering and impurity doping to enhance light absorption.^22–25 A notable early example is the work by Asahi et al., who demonstrated that N doping in TiO₂ effectively reduced the band gap, leading to enhanced absorption in the visible range.²⁶ This discovery spurred further research into various dopant elements and their effects on electronic structures.^27–29 Beyond chemical modifications, strain and stress engineering have also been explored as viable strategies to manipulate the band gap.^30,31 With the advent of 2D materials, novel strategies have emerged for optimizing light absorption. One effective approach is constructing heterostructures by stacking different 2D materials, such as BCN and C₂N, to engineer band alignments and enhance absorption across a broader spectrum.^32,33 In recent years, the increasing computational power of supercomputers has allowed researchers to move beyond individual material studies and instead perform high-throughput screening of entire material families.^34,35

It should be noted that light absorption in photocatalytic materials is directly related to their optical band gap. However, due to the well-known self-interaction error, DFT tends to underestimate band gaps and fails to accurately predict impurity states introduced by doping or defects.^36,37 Hybrid functionals can partially improve these issues. In addition, in 2D materials, reduced dielectric screening leads to strong e–h interactions, namely the exciton effects, which distinctly influence the optical band gap.^38,39 In such cases, GW–Bethe–Salpeter equation (GW + BSE) provides a more accurate description of the quasiparticle energies and excitonic spectra.^40,41 Linear response time-dependent DFT (LR-TDDFT) with hybrid functionals is also a viable alternative for capturing excitonic effects at a lower computational cost.^42,43 While these beyond-DFT methods offer higher accuracy, their computational expense often limits their application in high-throughput screening. In particular, GW + BSE remains impractical for large-scale material searches. This challenge presents an opportunity for machine learning, which may accelerate the prediction of optical properties and aid in the discovery of efficient photocatalysts. For example, Thygesen et al. developed a “state-based” representation (ENDOME/RAD-PDOS) that uses DFT-level information to predict G₀W₀ self-energy corrections for 2D materials, achieving mean absolute errors around 0.14 eV across different materials.⁴⁴ Galli et al. learned the mapping from bare to screened Coulomb interactions to speed up finite-temperature BSE spectral calculations for solids and solid–liquid interfaces, achieving near-BSE accuracy with orders-of-magnitude speedup.⁴⁵

Excited carrier dynamics

i. Electron–hole recombination

After photoexcitation generates e–h pairs, their participation in subsequent photocatalytic reactions requires sufficiently long lifetimes to avoid recombination. It is crucial to have ab initio approaches for determining the recombination timescales of the electrons and holes in semiconductor materials. E–h recombination can occur via two main pathways: radiative recombination and non-radiative recombination. In radiative recombination, the electron and hole recombine, resulting in the emission of energy in the form of a photon, while the non-radiative recombination entails the dissipation of the recombination energy into the lattice, leading to the generation of heat. The timescale for radiative recombination can be estimated by calculating the transition dipole moments and applying Fermi's golden rule.⁴⁶ On the other hand, the timescale for non-radiative recombination is more complex and can be obtained using ab initio nonadiabatic molecular dynamics (NAMD) simulations based on the classical path approximation (CPA).^47–49 In non-radiative recombination processes, e–ph coupling plays a crucial role. The NAMD method combines the time-dependent Kohn–Sham equation (TDKS) with surface hopping techniques, utilizing molecular dynamics (MD) simulations to model phonons in solid-state materials. By calculating non-adiabatic couplings, this method can effectively describe the e–ph coupling in materials and provide insights into the dynamics of electrons and holes. In recent years, the NAMD method has been implemented in programs such as Pyxaid and Hefei-NAMD.^48–50 These tools have been widely used to calculate the lifetimes of excited carriers in solid-state materials,^51–54 contributing to the understanding of excited carrier dynamics in photocatalytic systems.

The NAMD method has been employed to study the lifetimes of excited carriers in semiconductors that contain defects or dopants. The presence of defects and impurities in semiconductors is usually inherent, and element doping is often utilized to modify the electronic structure of materials, thereby optimizing their optical absorption properties.^52,55 Therefore, understanding the impact of doping and defects on the lifetime of excited carriers is of paramount importance. As early as the 1950s, Shockley, Reed, and Hall proposed the Shockley–Read–Hall (SRH) model, which asserts that when impurity or defect states are located near the middle of the band gap, they can act as e–h recombination centers and accelerate the e–h recombination process.^56,57 However, the SRH model does not account for the e–ph coupling in different materials in a quantitative manner. In this regard, ab initio NAMD simulations provide a more advanced framework for validating the applicability of the SRH model. Recent studies have shown that for semiconductors with high rigidity, such as TiO₂, the results obtained from NAMD simulations are qualitatively consistent with those predicted by the SRH model.⁵¹ However, for relatively soft materials, such as black phosphorus and perovskite-based materials, it was reported that regardless of whether the impurity level is located at the mid-gap, it will not form an e–h recombination center.^58,59 These findings suggest that materials with relatively low rigidity may be less susceptible to defect-induced e–h recombination, which in turn could allow for the stabilization of longer-lived e–h pairs. Such properties may offer advantages in photocatalytic and photovoltaic applications. However, these theoretical predictions require further experimental validation.

ii. Electron–hole separation

To prolong the lifetime of photoexcited carriers, one straightforward approach is to separate the electron and hole. Constructing heterojunctions using different semiconductors is one possibility to achieve e–h separation. The development of 2D materials has further facilitated the construction of heterojunctions. As shown in Fig. 2, both type-II and Z-scheme heterojunctions can effectively separate e–h pairs.^60,61 In type-II heterojunctions, photoexcited electrons and holes migrate to opposite sides of the junction, resulting in effective charge separation but reduced redox potential. In contrast, Z-scheme heterojunctions involve interfacial recombination of low-energy carriers, retaining high-energy electrons and holes in their respective semiconductors for enhanced redox activity and efficient charge separation. The distinction between type-II and Z-scheme heterojunctions is primarily governed by the behavior of excited carrier dynamics. If charge transfer at the interface occurs faster than e–h recombination, the system functions as a type-II heterojunction. Conversely, if e–h recombination at the interface is faster than charge transfer, it forms a Z-scheme heterojunction. In recent years, substantial theoretical work has been dedicated to designing type-II and Z-scheme heterojunctions.^62–64 The NAMD approach has been widely applied to study the e–h recombination and the interface charge transfer.^32,65,66 Experimental studies have also validated the high photocatalytic efficiency of Z-scheme materials and the related S-scheme photocatalysts.^67–69 These studies provide a deeper understanding of the mechanisms behind e–h pair separation and their potential applications in enhancing photocatalytic performance.

Fig. 2 Schematic of effective e–h pair separation in (a) type-II heterojunctions and (b) Z-scheme heterojunctions. Type-II heterostructures separate electrons and holes across two semiconductors, reducing recombination and also redox power. Z-scheme heterostructures retain strong redox abilities by recombining low-energy carriers internally, leaving high-energy electrons and holes in their respective materials for enhanced photocatalytic activity.

iii. Photoexcited carrier trapping by surface molecules

When photoexcited carriers have sufficiently long lifetimes, the next crucial step in photocatalysis is the process of surface-adsorbed molecules capturing electrons or holes. This process is primarily influenced by two factors. First, the energy alignment between the molecular levels and the semiconductor's band edges plays a critical role. Photoexcited carriers tend to relax towards lower energy states, so the energy level alignment determines the direction of carrier transfer at the interface. For instance, the famous hole scavenger CH₃OH on the TiO₂ surface is effective at capturing holes due to its highest occupied molecular orbital (HOMO) being close to the valence band maximum (VBM) of TiO₂. Further studies have shown that when CH₃OH dissociates to form CH₃O, its HOMO lies above the VBM of TiO₂, enhancing its ability to capture holes.^70–72 The second key factor is the strength of the orbital interaction between the molecule and the surface and the strength of e–ph coupling, which governs the rate of carrier capture. In general, molecules that are chemically adsorbed onto the surface capture carriers faster than those that are physically adsorbed. The dynamics of excited-state carrier capture can also be investigated using the ab initio NAMD method, which provides detailed insights into the interactions between the photoexcited carriers and surface adsorbates.^70,71

Furthermore, an additional consideration arises when examining solid–liquid interfaces under realistic conditions, as opposed to simplified systems with low-coverage adsorbed molecules on semiconductor surfaces. In such cases, the complexity increases significantly due to solvent effects, particularly the formation of electric double layers.^73,74 Addressing the energy level alignment at the interface from a first-principles perspective presents a formidable challenge, requiring explicit treatment of solvent-induced electrostatic interactions and their interplay with electronic states at the interface.^75,76

Table 1 presents a comparison between the results obtained from NAMD simulations and the corresponding experimental measurements. The simulated time scales agree with experimental values within one order of magnitude. It should be emphasized, however, that NAMD simulations are not intended to yield quantitatively precise time scales, as the results can be significantly affected by the choice of methods and parameters employed.

Table 1 Comparison of different carrier dynamics’ time scales obtained from NAMD simulations and experiments

Dynamics	Materials		Calculated results	Experimental results
Electron–hole recombination	Rutile TiO2		∼500 ps^51	∼20 ps^77
	Black phosphorus		375 ps^59	∼100 ps^78
Electron–hole separation	MoS2/WS2	Hole transfer	20 fs^65	<50 fs^79
	MoS2/MoSe2	Hole transfer	213 fs^80	<200 fs^81
		Electron transfer	105 fs^80	<1 ps^81
	MoSe2/WSe2	Hole transfer	∼250 fs^82	∼200 fs^82
	MoSe2/hBN/WSe2	Hole transfer	1197 ps^83	∼500 ps^84
Photoexcited carrier trapping by surface molecules	CH3OH/TiO2	Hole trapping	∼150 fs^70	<100 fs^85
	CH3OH/g-C3N4		(CH3OH/TiO2)	(CH3OH/g-C3N4)
	H2O/TiO2	Hole trapping	∼200 fs^86	<285 fs^86

---

iv. Challenges in excited carrier dynamics simulations

Recent advancements in ab initio simulations of excited carrier dynamics are noteworthy, especially with the widespread application of the CPA-based NAMD method in simulating carrier dynamics in different materials. Table 2 summarizes the representative NAMD simulation frameworks and typical codes for modeling the excited carrier dynamics. Nevertheless, several challenges and issues still remain. First, the accuracy of NAMD simulations is directly tied to the precision of electronic structure calculations. For many materials, especially those containing d- and f-orbitals, hybrid functionals are required to achieve accurate energy levels and wavefunctions, particularly for defect states.^87,88 However, for computationally demanding methods like NAMD, using hybrid functionals for large-scale simulations remains impractical. Similarly, the process of surface molecules trapping charge carriers is highly dependent on the energy level alignment at the semiconductor–molecule interface. Since molecules and semiconductors are fundamentally different systems, their dependence on different functionals varies. Thus, selecting an appropriate functional for simulations remains a key challenge. For instance, studies on the CH₃OH/TiO₂ interface have shown that the band alignment obtained using GW methods differs from that derived from DFT calculations.⁷² With the rapid development of machine learning, there is hope that machine learning techniques could assist in constructing Hamiltonians for large systems with accuracy close to hybrid functional and GW calculations, potentially providing a solution to this problem.

Table 2 Representative NAMD simulation frameworks and typical codes for modeling the excited carrier dynamics

Theoretical framework	Codes
Ehrenfest dynamics	Octopus (https://www.octopus-code.org/)
	TDAP (https://tdap.iphy.ac.cn/)
	GPAW (https://gpaw.readthedocs.io)
	QE (https://www.quantum-espresso.org/)
	PWmat (https://www.pwmat.com/)
Surface hopping	PYXAID (https://quantum-dynamics-hub.github.io/)
	Libra (https://quantum-dynamics-hub.github.io/)
	Hefei-NAMD (https://hefei-namd.org/)
	SPADE (https://www.linjun-wang-group.com/)
	PWmat (https://www.pwmat.com/)

---

Another critical issue lies in the fact that most widely used NAMD implementations rely on the CPA, which assumes that excited-state electrons do not influence lattice dynamics.^48,89 Consequently, NAMD simulations are typically performed using ground-state ab initio molecular dynamics (AIMD) trajectories.^48,50 While the CPA accounts for the effect of phonons on electron dynamics, it neglects the feedback of electron excitations on lattice dynamics. This limitation significantly impacts the study of e–h recombination processes in semiconductors, particularly in ionic crystals. When excited electrons and holes are present in an ionic semiconductor, the lattice structure often undergoes distortion due to e–ph coupling, sometimes leading to formation of polarons.⁹⁰ In 2D semiconductors, the coexistence of e–h and e–ph interactions may further result in exciton–polaron complexes.^91,92 In such cases, relying solely on ground-state AIMD to simulate lattice dynamics becomes inadequate. Moreover, CPA introduces inaccuracies in defective systems. For instance, when a defect state captures an electron or a hole, its charge state and local atomic structure may undergo distinct changes—a critical factor unaccounted for in ground-state AIMD simulations under the CPA framework.

Overcoming the limitations of the CPA requires reliable and efficient methods to compute forces on excited-state PES in solids. Achieving this would enable on-the-fly surface hopping simulations, akin to those used for small molecules in photochemistry.^93,94 While some approaches, such as LR-TDDFT or GW + BSE, have been proposed to be able to calculate excited-state forces in solids,^95,96 their computational complexity has hindered their widespread adoption in solid-state systems. A more practical alternative may involve the ΔSCF method, also known as constrained DFT, to perform AIMD simulations.^90,97 This approach could capture the dynamics of polaron formation induced by electrons or holes in semiconductors, thereby elucidating their impact on carrier recombination.^90,97 Such methods might also be extended to defective systems. Ultimately, we anticipate the development of more robust and efficient electronic structure methodologies to eliminate the computational bottleneck associated with excited-state force calculations in solids, thereby circumventing the limitations of the CPA.

Another noteworthy consideration is that nuclear quantum effects (NQEs) can influence electron dynamics. For instance, many photocatalytic systems involve solid–aqueous interfaces, where surface-bound molecules may form hydrogen-bonded networks.⁹⁸ In such cases, protons frequently transfer within these networks, and NQEs often become significant.^99–102 For solid-state systems, the computational cost of fully quantum dynamical methods is prohibitive, unlike in small-molecule studies. A practical alternative is to combine NAMD with path-integral molecular dynamics (PIMD). This approach employs ring-polymer molecular dynamics (RPMD) to generate nuclear trajectories, replacing conventional AIMD, and subsequently performs NAMD simulations on these trajectories to account for quantum nuclear motion.^71,103,104

Surface photochemical reactions

Next, we will discuss the process of photochemical reactions on the semiconducting surfaces. As shown in Fig. 3, the influence of light on surface chemical reactions can occur through three distinct channels. If the optical excitation generates hot carriers far from the band edges, these hot carriers will transfer their energy to the lattice through e–ph coupling during their relaxation to the band edge, resulting in localized heating, which is referred to as the ‘photothermal effect’. On the other hand, when photoexcited carriers are captured by molecules adsorbed on the surface and possess a sufficiently long lifetime, the chemical reaction PES will shift from the ground state to the excited state, thus affecting the reaction process and products. Finally, the light field may also directly interact with the molecules adsorbed on the surface, promoting the reaction. Below, we will discuss surface photochemical reactions from the three aspects mentioned above.

Fig. 3 Schematic of distinct channels of photochemical reactions on the semiconducting surfaces: (I) heat localization on the surface after hot carrier relaxation; (II) capture of photoexcited carriers by adsorbates; and (III) resonant excitation of the light field to interface vibration modes (optical phonons on the surface or chemical bond vibrations in molecules).

i. Photothermal effects

Recently, the role of the photothermal effects in photocatalysis has received increasing attention.^105,106 When hot carriers have not reached the band edges or been captured by surface molecules, the photothermal effect becomes the most important factor influencing the photocatalytic process. The photothermal effects can lead to localized heating of the material.^106,107 Through special designs, such as adsorbing metal nanoparticles onto the solid surface, spatially selective heating can be achieved.^108,109 The photothermal heating may affect photocatalysis in various ways, with both positive and negative potential impacts. Possible positive effects include helping surface-adsorbed molecules overcome reaction barriers or potentially accelerating the separation of excited charge carriers, and the potential negative effects include the possibility that localized heating accelerates e–h recombination. The exact outcome depends on the specific system, and ab initio investigations are crucial to understand the photothermal effects.

To study the photothermal effects from ab initio calculations, it is necessary to accurately describe the energy transfer process from the hot electrons to the lattice. For the NAMD method based on CPA, the influence of excited-state electrons on the lattice is not considered, and the electron system and lattice system are completely separated, resulting in a lack of overall energy conservation. However, the recently developed NAMD_k method introduces e–ph coupling elements into the Hamiltonian and incorporates energy conservation conditions.¹¹⁰ This approach not only provides time-resolved electron and hole dynamics but also simultaneously yields information on phonon population changes over time as hot carriers relax.^110–112 Therefore, it becomes possible to accurately track phonon excitation as hot carriers relax. If this method is applied to the molecule–semiconductor interface, it could provide a precise understanding of the photothermal effects, making it an effective approach for using ab initio calculations to study the photothermal effects.

ii. Effects of excited state potential energy surface

When excited-state e–h pairs are captured by surface-adsorbed molecules, the molecules transition to the excited state, necessitating the consideration of excited-state potential energy surfaces in chemical reaction studies. However, as shown in Fig. 4, when the excited and ground-state potential energy surfaces get close in energy to each other, non-adiabatic transitions may occur, returning the system to the ground state. Two common approaches are used to address such non-adiabatic effects. The first combines Ehrenfest dynamics with real-time TDDFT (rt-TDDFT), where electrons and the lattice evolve on a mean-field potential energy surface.¹¹³ This method has been applied to several different materials, including Meng et al., who studied the light-induced dissociation of H₂O on the TiO₂ surface.^114–117 The second is the on-the-fly surface hopping method,^93,94 where the system occupies either the ground or excited state, with transitions dictated by non-adiabatic coupling.⁹⁴ However, its reliance on excited-state force calculations limits its application to solids. Chu et al. and Wang et al. explored surface photocatalytic reactions using constrained DFT, though without fully incorporating non-adiabatic effects.^118,119

Fig. 4 Schematic of Ehrenfest dynamics and trajectory surface hopping dynamics. Ehrenfest dynamics uses a mean-field approach where nuclei evolve on an averaged potential energy surface, treating electronic states as a superposition. Surface hopping allows nuclei to evolve on individual adiabatic surfaces with stochastic hops between states to capture non-adiabatic effects.

iii. Light-field interaction

The interaction between light and matter manifests differently depending on the photon energy. For visible and ultraviolet light, a prominent effect is the generation of excited charge carriers in semiconductors, as discussed previously. When considering lower-energy light, such as terahertz (THz) radiation, whose photon energy aligns with the energy scales of optical phonons in ionic crystals or chemical bond vibration in molecules, resonant excitation of specific vibrational modes becomes feasible.¹²⁰ Such interaction can induce direct lattice molecular bond vibrations. On one hand, when the optical field is sufficiently strong, direct bond breaking may occur. On the other hand, THz technology can be combined with higher-energy optical excitation, where THz light can excite vibrational modes at the molecule–solid interface, affecting the energy alignment. This, in turn, can influence the relaxation of photoexcited charge carriers generated by higher-energy light and the capture of these carriers by molecular adsorbates. Furthermore, when metallic nanoparticles are adsorbed on the surface, THz light may also excite surface plasmon resonances, which can further modulate surface photochemical reactions.^121,122 While extensive studies have been conducted on the influence of surface plasmons on chemical reactions,^123,124 a detailed discussion of this topic is beyond the scope of this work due to thematic and space limitations.

Due to the low frequency of THz light, its interaction with materials can be approximated as a slowly varying electric field. This enables the simulation of THz-induced phonon and molecular vibrational excitations by calculating the Born effective charges of the material to determine the field-driven atomic forces, combined with MD simulations.¹²⁵ This approach has already been applied to topological materials and is expected to be broadly applicable to semiconductor–molecule interfaces.^125,126 Furthermore, this methodology can be seamlessly integrated with NAMD to investigate carrier dynamics under multifield synergy involving THz light and higher-energy photons (visible to ultraviolet range), where THz-driven lattice vibrations may modulate hot carrier relaxation pathways or interfacial charge transfer dynamics.

Recently, Chen et al. incorporated light–matter interactions into Hefei-NAMD, enabling a unified and continuous simulation of photoabsorption and carrier dynamics.¹²⁷ This advancement also allows researchers to investigate how the presence of phonons affects photoexcitation. Notably, it becomes possible to first excite phonons using THz light, and the resulting phonon modes may in turn induce selective optical excitations and influence the subsequent carrier dynamics. This represents a particularly intriguing direction for future research.

To clearly illustrate how theoretical calculations can be applied to investigate photochemical reaction mechanisms, we present an example using NAMD. Meng et al. applied rt-TDDFT combined with Ehrenfest dynamics, explicitly incorporating the light field, to investigate the photoinduced water dissociation on rutile TiO₂.¹¹⁴ Two distinct dissociation pathways were identified. In the first pathway, the light field initiates electron migration from a surface bridging oxygen on the TiO₂ surface to the oxygen atom of H₂O, weakening the O–H bond and inducing proton transfer to the bridging oxygen within ∼15 fs during irradiation. This field-initiated process lowers the proton-transfer barrier, enabling dissociation at temperatures above 100 K. In the second one, an adsorbed water molecule hydrogen bonded to a second-layer H₂O undergoes proton transfer to the latter, coupled with photoexcited hole transfer to the adsorbate within ∼35–42 fs. This hole-driven process occurs exclusively under 3.1 eV excitation and is promoted by surface polarons and the associated lattice distortions. The study found that the ratio of hole-driven to field-initiated dissociation is approximately 1.25 : 1 based on the statistical analysis of several non-adiabatic trajectories, indicating that both pathways make comparably important contributions to the reaction. This work demonstrates the capability of advanced first-principles excited-state dynamics to model photocatalytic reactions.

iv. Challenges in surface photochemical reaction simulations

To date, ab initio studies on surface photochemical reactions remain relatively scarce. For photothermal effects, NAMD_k may offer a promising approach.¹¹⁰ However, NAMD_k is a recently developed methodology that has yet to be extensively validated across diverse systems. Furthermore, the computation of e–ph coupling elements—critical for NAMD_k—becomes computationally prohibitive for molecular-adsorbed surfaces containing tens to hundreds of atoms. Advances in machine learning methods may provide opportunities to bypass this bottleneck by approximating coupling terms with reduced computational cost.¹²⁸

For modeling photochemical reactions involving surface-adsorbed molecules in excited states, combining Ehrenfest dynamics with rt-TDDFT presents a viable strategy. This approach simulates non-adiabatic effects on excited-state PES under a mean-field approximation while inherently incorporating light–matter interactions, thereby directly modeling light absorption characteristics. While this method yields reliable results when the excited-state and ground-state PES are relatively close, its accuracy diminishes when these PES diverge distinctly.¹²⁹ Thus, further theoretical investigations based on this approach and experimental validations are essential to assess its robustness in surface photocatalysis.

The primary limitation of on-the-fly surface hopping-based NAMD for simulating surface photochemical reactions lies in the computational demand for excited-state force calculations, as discussed earlier. A critical additional challenge is the statistical sampling requirement: surface hopping simulations typically necessitate hundreds to thousands of trajectories to converge reaction probabilities,^130,131 imposing formidable computational burdens. Developing efficient methods to compute excited-state forces in solids with both accuracy and affordability is imperative to enable a broader application of on-the-fly surface hopping in surface photochemistry.

Summary

In this article, we systematically examine the advances, challenges, and opportunities of ab initio calculations in surface photocatalysis. While ab initio methods have achieved notable success in modeling light absorption properties, notable challenges persist in simulating the excited carrier dynamics and surface photochemical reactions. These challenges, however, underscore the untapped potential for advancing ab initio methodologies in photocatalysis. From an applied standpoint, emerging techniques such as rt-TDDFT, NAMD, and NAMD_k require rigorous benchmarking against experimental data to validate their predictive power in photocatalytic systems. Methodologically, critical developments are needed to enhance computational efficiency and accuracy, including scalable hybrid density functionals, accelerated GW schemes and e–ph coupling calculations, and robust algorithms for excited-state force evaluations in solids. Addressing these challenges will deepen our mechanistic understanding of surface photocatalysis and provide a theoretical foundation for designing advanced photocatalytic materials with tailored light-harvesting, charge separation, and surface reactivity properties.

Conflicts of interest

There are no conflicts to declare.

Data availability

No primary research results, software or code have been included and no new data were generated or analysed as part of this review.

Acknowledgements

J. Z. acknowledges the support of the Innovation Program for Quantum Science and Technology (2021ZD0303306), the Strategic Priority Research Program of the Chinese Academy of Sciences (grant no. XDB0450101), and the National Natural Science Foundation of China (NSFC) (grant no. 12125408 and 12334004). Q. Z. acknowledges the support of the NSFC (grant no. 12174363). L. Z. acknowledges the support of the Natural Science Foundation of Henan Province (no. 242300421161) and the China Postdoctoral Science Foundation (no. 2020M682326).

References

1. Q. Guo, C. Y. Zhou, Z. B. Ma and X. M. Yang, Adv. Mater., 2019, 31, 1901997.
2. M. A. Henderson, Surf. Sci. Rep., 2011, 66, 185–297.
3. C. V. Reddy, K. R. Reddy, N. P. Shetti, J. Shim, T. M. Aminabhavi and D. D. Dionysiou, Int. J. Hydrogen Energy, 2020, 45, 18331–18347.
4. U. Diebold, Surf. Sci. Rep., 2003, 48, 53–229.
5. Y. Attia and M. Samer, Renewable Sustainable Energy Rev., 2017, 79, 878–892.
6. Z. P. Yu, Y. F. Li, A. Torres-Pinto, A. P. LaGrow, V. M. Diaconescu, L. Simonelli, M. J. Sampaio, O. Bondarchuk, I. Amorim, A. Araujo, A. M. T. Silva, C. G. Silva, J. L. Faria and L. F. Liu, Appl. Catal., B, 2022, 310, 121318.
7. Z. P. Li, Y. F. Zhi, P. P. Shao, H. Xia, G. S. Li, X. Feng, X. Chen, Z. Shi and X. M. Liu, Appl. Catal., B, 2019, 245, 334–342.
8. L. Yang, Y. T. Peng, X. D. Luo, Y. Dan, J. H. Ye, Y. Zhou and Z. G. Zou, Chem. Soc. Rev., 2021, 50, 2147–2172.
9. S. Sharma, V. Dutta, P. Singh, P. Raizada, A. Rahmani-Sani, A. Hosseini-Bandegharaei and V. K. Thakur, J. Cleaner Prod., 2019, 228, 755–769.
10. P. Sun, Z. P. Xing, Z. Z. Li and W. Zhou, Chem. Eng. J., 2023, 458, 141399.
11. R. Li, W. Zhang and K. Zhou, Adv. Mater., 2018, 30, 1705512.
12. D. D. Li, M. Kassymova, X. C. Cai, S. Q. Zang and H. L. Jiang, Coord. Chem. Rev., 2020, 412, 213262.
13. W. X. Yang, G. Z. Ma, Y. Fu, K. Peng, H. L. Yang, X. Q. Zhan, W. Y. Yang, L. Wang and H. L. Hou, Chem. Eng. J., 2022, 429, 132381.
14. R. Q. Gao, H. He, J. X. Bai, L. Hao, R. C. Shen, P. Zhang, Y. J. Li and X. Li, Chin. J. Struct. Chem., 2022, 41, 2206031.
15. Y. Wang and P. B. Balbuena, J. Phys. Chem. B, 2005, 109, 18902–18906.
16. A. J. Cowan, C. J. Barnett, S. R. Pendlebury, M. Barroso, K. Sivula, M. Grätzel, J. R. Durrant and D. R. Klug, J. Am. Chem. Soc., 2011, 133, 10134–10140.
17. L. Cheng, Q. Xiang, Y. Liao and H. Zhang, Energy Environ. Sci., 2018, 11, 1362–1391.
18. L. Jing, Z. Xu, X. Sun, J. Shang and W. Cai, Appl. Surf. Sci., 2001, 180, 308–314.
19. A. Malathi, J. Madhavan, M. Ashokkumar and P. Arunachalam, Appl. Catal., A, 2018, 555, 47–74.
20. N. Thomas, S. Mathew, K. M. Nair, K. O'Dowd, P. Forouzandeh, A. Goswami, G. McGranaghan and S. C. Pillai, Mater. Today Sustain., 2021, 13, 100073.
21. Y. Li, X. Li, H. Zhang, J. Fan and Q. Xiang, J. Mater. Sci. Technol., 2020, 56, 69–88.
22. J. Nowotny, M. A. Alim, T. Bak, M. A. Idris, M. Ionescu, K. Prince, M. Z. Sahdan, K. Sopian, M. A. Mat Teridi and W. Sigmund, Chem. Soc. Rev., 2015, 44, 8424–8442.
23. D. Maarisetty and S. S. Baral, J. Mater. Chem. A, 2020, 8, 18560–18604.
24. Y. Zhang, B. Fluegel, A. Mascarenhas, H. P. Xin and C. W. Tu, Phys. Rev. B: Condens. Matter Mater. Phys., 2000, 62, 4493–4500.
25. M. Arshad, M. Meenhaz Ansari, A. S. Ahmed, P. Tripathi, S. S. Z. Ashraf, A. H. Naqvi and A. Azam, J. Lumin., 2015, 161, 275–280.
26. R. Asahi, T. Morikawa, T. Ohwaki, K. Aoki and Y. Taga, Science, 2001, 293, 269–271.
27. T. Umebayashi, T. Yamaki, H. Itoh and K. Asai, J. Phys. Chem. Solids, 2002, 63, 1909–1920.
28. F. Tian and C. Liu, J. Phys. Chem. B, 2006, 110, 17866–17871.
29. M. Guo and J. Du, Phys. B, 2012, 407, 1003–1007.
30. H. J. Conley, B. Wang, J. I. Ziegler, R. F. Haglund, Jr., S. T. Pantelides and K. I. Bolotin, Nano Lett., 2013, 13, 3626–3630.
31. Q. Tu, I. Spanopoulos, S. Hao, C. Wolverton, M. G. Kanatzidis, G. S. Shekhawat and V. P. Dravid, ACS Energy Lett., 2019, 4, 796–802.
32. R. Zhang, L. Zhang, Q. Zheng, P. Gao, J. Zhao and J. Yang, J. Phys. Chem. Lett., 2018, 9, 5419–5424.
33. J. Meng, J. Wang, J. Wang, Q. Li and J. Yang, J. Phys. Chem. Lett., 2022, 13, 1473–1479.
34. R. Potyrailo, K. Rajan, K. Stoewe, I. Takeuchi, B. Chisholm and H. Lam, ACS Comb. Sci., 2011, 13, 579–633.
35. N. M. Nursam, X. Wang and R. A. Caruso, ACS Comb. Sci., 2015, 17, 548–569.
36. J. P. Perdew, Int. J. Quantum Chem., 1985, 28, 497–523.
37. J. L. Lyons and C. G. Van de Walle, npj Comput. Mater., 2017, 3, 12.
38. T. Mueller and E. Malic, npj 2D Mater. Appl., 2018, 2, 29.
39. T. Ketolainen, N. Macháčová and F. Karlický, J. Chem. Theory Comput., 2020, 16, 5876–5883.
40. W. Kang and M. S. Hybertsen, Phys. Rev. B: Condens. Matter Mater. Phys., 2010, 82, 085203.
41. M. O. Atambo, D. Varsano, A. Ferretti, S. S. Ataei, M. J. Caldas, E. Molinari and A. Selloni, Phys. Rev. Mater., 2019, 3, 045401.
42. N. Govind, K. Lopata, R. Rousseau, A. Andersen and K. Kowalski, J. Phys. Chem. Lett., 2011, 2, 2696–2701.
43. J. Liu, W. Hu and J. Yang, J. Chem. Theory Comput., 2022, 18, 6713–6721.
44. N. R. Knøsgaard and K. S. Thygesen, Nat. Commun., 2022, 13, 468.
45. S. S. Dong, M. Govoni and G. Galli, Chem. Sci., 2021, 12, 4970–4980.
46. H.-Y. Chen, V. A. Jhalani, M. Palummo and M. Bernardi, Phys. Rev. B, 2019, 100, 075135.
47. C. F. Craig, W. R. Duncan and O. V. Prezhdo, Phys. Rev. Lett., 2005, 95, 163001.
48. A. V. Akimov and O. V. Prezhdo, J. Chem. Theory Comput., 2013, 9, 4959–4972.
49. A. V. Akimov and O. V. Prezhdo, J. Chem. Theory Comput., 2014, 10, 789–804.
50. Q. Zheng, W. Chu, C. Zhao, L. Zhang, H. Guo, Y. Wang, X. Jiang and J. Zhao, Wiley Interdiscip. Rev.: Comput. Mol. Sci., 2019, 9, e1411.
51. L. Zhang, Q. Zheng, Y. Xie, Z. Lan, O. V. Prezhdo, W. A. Saidi and J. Zhao, Nano Lett., 2018, 18, 1592–1599.
52. R. Long, W. Fang and A. V. Akimov, J. Phys. Chem. Lett., 2016, 7, 653–659.
53. H. Jin, T. Wang, Z.-R. Gong, C. Long and Y. Dai, Nanoscale, 2018, 10, 19310–19315.
54. W. Li, Y. She, A. S. Vasenko and O. V. Prezhdo, Nanoscale, 2021, 13, 10239–10265.
55. H. Guo, W. Chu, Q. Zheng and J. Zhao, J. Phys. Chem. Lett., 2020, 11, 4662–4667.
56. R. Hall, Phys. Rev., 1951, 83, 228.
57. W. Shockley and W. T. Read, Phys. Rev., 1952, 87, 835–842.
58. W. Chu, Q. Zheng, O. V. Prezhdo, J. Zhao and W. A. Saidi, Sci. Adv., 2020, 6, eaaw7453.
59. L. Zhang, W. Chu, Q. Zheng, A. V. Benderskii, O. V. Prezhdo and J. Zhao, J. Phys. Chem. Lett., 2019, 10, 6151–6158.
60. S. S. Lo, T. Mirkovic, C.-H. Chuang, C. Burda and G. D. Scholes, Adv. Mater., 2011, 23, 180–197.
61. J. Low, C. Jiang, B. Cheng, S. Wageh, A. A. Al-Ghamdi and J. Yu, Small Methods, 2017, 1, 1700080.
62. G. Di Liberto, S. Tosoni and G. Pacchioni, Catal. Sci. Technol., 2021, 11, 3589–3598.
63. M. Su, H. Sun, Z. Tian, Z. Zhao and P. Li, Appl. Catal., A, 2022, 631, 118485.
64. R. Kumar, D. Das and A. K. Singh, J. Catal., 2018, 359, 143–150.
65. Q. Zheng, W. A. Saidi, Y. Xie, Z. Lan, O. V. Prezhdo, H. Petek and J. Zhao, Nano Lett., 2017, 17, 6435–6442.
66. Y. Liu, Z. Jiang, J. Jia, J. Robertson and Y. Guo, Appl. Surf. Sci., 2023, 611, 155674.
67. Q. Xu, L. Zhang, B. Cheng, J. Fan and J. Yu, Chem, 2020, 6, 1543–1559.
68. X.-h Wu, G.-q Chen, J. Wang, J. Li and G. Wang, Acta Phys. Chim. Sin., 2023, 39, 221201.
69. G. Wang, K. Dou, H. Cao, R. Du, J. Liu, N. Tsidaeva and W. Wang, Sep. Purif. Technol., 2022, 291, 120976.
70. W. Chu, W. A. Saidi, Q. Zheng, Y. Xie, Z. Lan, O. V. Prezhdo, H. Petek and J. Zhao, J. Am. Chem. Soc., 2016, 138, 13740–13749.
71. W. Chu, S. Tan, Q. Zheng, W. Fang, Y. Feng, O. V. Prezhdo, B. Wang, X.-Z. Li and J. Zhao, Sci. Adv., 2022, 8, eabo2675.
72. A. Migani, D. J. Mowbray, A. Iacomino, J. Zhao, H. Petek and A. Rubio, J. Am. Chem. Soc., 2013, 135, 11429–11432.
73. D. Henderson and L. Blum, J. Electroanal. Chem., 1982, 132, 1–13.
74. Y. Burak and D. Andelman, Phys. Rev. E: Stat. Phys., Plasmas, Fluids, Relat. Interdiscip. Top., 2000, 62, 5296–5312.
75. J. Cheng and M. Sprik, Phys. Chem. Chem. Phys., 2012, 14, 11245–11267.
76. J.-B. Le and J. Cheng, Curr. Opin. Electrochem., 2021, 27, 100693.
77. K. Ozawa, M. Emori, S. Yamamoto, R. Yukawa, S. Yamamoto, R. Hobara, K. Fujikawa, H. Sakama and I. Matsuda, J. Phys. Chem. Lett., 2014, 5, 1953–1957.
78. J. He, D. He, Y. Wang, Q. Cui, M. Z. Bellus, H.-Y. Chiu and H. Zhao, ACS Nano, 2015, 9, 6436–6442.
79. X. Hong, J. Kim, S.-F. Shi, Y. Zhang, C. Jin, Y. Sun, S. Tongay, J. Wu, Y. Zhang and F. Wang, Nat. Nanotechnol., 2014, 9, 682–686.
80. R. Long and O. V. Prezhdo, Nano Lett., 2016, 16, 1996–2003.
81. F. Ceballos, M. Z. Bellus, H.-Y. Chiu and H. Zhao, ACS Nano, 2014, 8, 12717–12724.
82. Z. Wang, P. Altmann, C. Gadermaier, Y. Yang, W. Li, L. Ghirardini, C. Trovatello, M. Finazzi, L. Duò, M. Celebrano, R. Long, D. Akinwande, O. V. Prezhdo, G. Cerullo and S. Dal Conte, Nano Lett., 2021, 21, 2165–2173.
83. Y. Lei, X. Xie, H. Ma and J. Ma, J. Phys. Chem. Lett., 2023, 14, 11190–11199.
84. Y. Yoon, Z. Zhang, R. Qi, A. Y. Joe, R. Sailus, K. Watanabe, T. Taniguchi, S. Tongay and F. Wang, Nano Lett., 2022, 22, 10140–10146.
85. Z. Chen, Q. Zhang and Y. Luo, Angew. Chem., Int. Ed., 2018, 57, 5320–5324.
86. M. Wagstaffe, A. Dominguez-Castro, L. Wenthaus, S. Palutke, D. Kutnyakhov, M. Heber, F. Pressacco, S. Dziarzhytski, H. Gleißner, V. K. Gupta, H. Redlin, A. Dominguez, T. Frauenheim, A. Rubio, A. Stierle and H. Noei, Phys. Rev. Lett., 2023, 130, 108001.
87. P. Deák, B. Aradi, T. Frauenheim, E. Janzén and A. Gali, Phys. Rev. B: Condens. Matter Mater. Phys., 2010, 81, 153203.
88. A. Alkauskas, P. Broqvist and A. Pasquarello, Phys. Status Solidi B, 2011, 248, 775–789.
89. L. Wang, A. Akimov and O. V. Prezhdo, J. Phys. Chem. Lett., 2016, 7, 2100–2112.
90. L. Zhang, W. Chu, C. Zhao, Q. Zheng, O. V. Prezhdo and J. Zhao, J. Phys. Chem. Lett., 2021, 12, 2191–2198.
91. X. Li, A. Wang, H. Chen, W. Tao, Z. Chen, C. Zhang, Y. Li, Y. Zhang, H. Shang, Y.-X. Weng, J. Zhao and H. Zhu, Nano Lett., 2022, 22, 8755–8762.
92. W. Tao, Y. Zhang and H. Zhu, Acc. Chem. Res., 2022, 55, 345–353.
93. M. Barbatti, G. Granucci, M. Persico, M. Ruckenbauer, M. Vazdar, M. Eckert-Maksić and H. Lischka, J. Photochem. Photobiol., A, 2007, 190, 228–240.
94. Z. Lan, E. Fabiano and W. Thiel, J. Phys. Chem. B, 2009, 113, 3548–3555.
95. S. Ismail-Beigi and S. G. Louie, Phys. Rev. Lett., 2005, 95, 156401.
96. Y. Jin, V. W.-z Yu, M. Govoni, A. C. Xu and G. Galli, J. Chem. Theory Comput., 2023, 19, 8689–8705.
97. C. Gao, L. Zhang, Q. Zheng and J. Zhao, J. Phys. Chem. C, 2021, 125, 27275–27282.
98. M. Ceriotti, W. Fang, P. G. Kusalik, R. H. McKenzie, A. Michaelides, M. A. Morales and T. E. Markland, Chem. Rev., 2016, 116, 7529–7550.
99. S. P. Webb, T. Iordanov and S. Hammes-Schiffer, J. Chem. Phys., 2002, 117, 4106–4118.
100. M. E. Tuckerman, D. Marx and M. Parrinello, Nature, 2002, 417, 925–929.
101. J. A. Morrone and R. Car, Phys. Rev. Lett., 2008, 101, 017801.
102. M. Ceriotti, G. Bussi and M. Parrinello, Phys. Rev. Lett., 2009, 103, 030603.
103. Y. Liu, R. Long, W.-H. Fang and O. V. Prezhdo, J. Am. Chem. Soc., 2023, 145, 14112–14123.
104. J. Wang and R. Long, Nano Lett., 2024, 24, 3476–3483.
105. M. Xu, X. Hu, S. Wang, J. Yu, D. Zhu and J. Wang, J. Catal., 2019, 377, 652–661.
106. Y. Chen, J. Fang, B. Dai, J. Kou, C. Lu and Z. Xu, Appl. Surf. Sci., 2020, 534, 147640.
107. Y. Lu, H. Zhang, D. Fan, Z. Chen and X. Yang, J. Hazard. Mater., 2022, 423, 127128.
108. T. K. Sau, A. L. Rogach, F. Jäckel, T. A. Klar and J. Feldmann, Adv. Mater., 2010, 22, 1805–1825.
109. S. Maity, J. R. Bochinski and L. I. Clarke, Adv. Funct. Mater., 2012, 22, 5259–5270.
110. Z. Zheng, Y. Shi, J.-J. Zhou, O. V. Prezhdo, Q. Zheng and J. Zhao, Nat. Comput. Sci., 2023, 3, 532–541.
111. Z. Wang, Z. Zheng, Q. Zheng and J. Zhao, J. Phys. Chem. Lett., 2024, 15, 3907–3913.
112. Z. Shi, Z. Zheng, Q. Zheng and J. Zhao, Sci. Sin.-Phys. Mech. Astron., 2025, 55, 214210.
113. S. Meng and E. Kaxiras, J. Chem. Phys., 2008, 129, 054110.
114. P. You, D. Chen, X. Liu, C. Zhang, A. Selloni and S. Meng, Nat. Mater., 2024, 23, 1100–1106.
115. L. Yan, J. Xu, F. Wang and S. Meng, J. Phys. Chem. Lett., 2018, 9, 63–69.
116. P. You, C. Lian, D. Chen, J. Xu, C. Zhang, S. Meng and E. Wang, Nano Lett., 2021, 21, 6449–6455.
117. Y. Zhang, L. Yan, M. Guan, D. Chen, Z. Xu, H. Guo, S. Hu, S. Zhang, X. Liu, Z. Guo, S. Li and S. Meng, Adv. Sci., 2022, 9, 2102978.
118. W. Chu, Q. Zheng, O. V. Prezhdo and J. Zhao, J. Am. Chem. Soc., 2020, 142, 3214–3221.
119. J. Zhang, C. Peng, H. Wang and P. Hu, ACS Catal., 2017, 7, 2374–2380.
120. M. Guan, D. Chen, Q. Chen, Y. Yao and S. Meng, Phys. Rev. Lett., 2023, 131, 256503.
121. D. Polyushkin, E. Hendry, E. Stone and W. Barnes, Nano Lett., 2011, 11, 4718–4724.
122. J. Jana, M. Ganguly and T. Pal, RSC Adv., 2016, 6, 86174–86211.
123. M. Xiao, R. Jiang, F. Wang, C. Fang, J. Wang and J. C. Yu, J. Mater. Chem. A, 2013, 1, 5790–5805.
124. C. Zhan, X.-J. Chen, Y.-F. Huang, D.-Y. Wu and Z.-Q. Tian, Acc. Chem. Res., 2019, 52, 2784–2792.
125. D. Shin, A. Rubio and P. Tang, Phys. Rev. Lett., 2024, 132, 016603.
126. C. Bao, P. Tang, D. Sun and S. Zhou, Nat. Rev. Phys., 2022, 4, 33–48.
127. L. Chen, Z. Li, Q. Li, Q. Zheng and J. Zhao, Adv. Opt. Mater., 2025, 13, 2403069.
128. H. Li, Z. Wang, N. Zou, M. Ye, R. Xu, X. Gong, W. Duan and Y. Xu, Nat. Comput. Sci., 2022, 2, 367–377.
129. R. Crespo-Otero and M. Barbatti, Chem. Rev., 2018, 118, 7026–7068.
130. J. C. Tully, J. Chem. Phys., 1990, 93, 1061–1071.
131. S. Hammes-Schiffer and J. C. Tully, J. Chem. Phys., 1994, 101, 4657–4667.

---