Optimizing the performance of phase-change azobenzene: from trial and error to machine learning

Abstract

Molecular solar thermal (MOST) systems employ molecular photoswitches to store or release solar energy as heat under specific conditions. Among these systems, azobenzene (Azo) derivatives are particularly noteworthy owing to their superior photoresponsive properties and efficient reversible isomerization. Recent pioneering studies have focused on the molecular design of Azo derivatives to enhance the energy storage performance. Phase-change Azo (PC-Azo) derivatives—fabricated by integrating molecular photoswitches with phase-change materials—capture both the enthalpy of phase change and that of isomerization, thereby considerably increasing the energy storage density of Azo. This research provides a review of the crucial performance parameters of PC-Azo-based MOST systems and delves into the factors influencing the phase change and isomerization of PC-Azo derivatives to provide guidelines for the application of machine learning in MOST systems. A key focus of the research is the application of theoretical computing and machine learning in recent molecular design advancements. The study emphasizes the importance of employing machine learning techniques in the molecular design of PC-Azo derivatives, underlining their potential in facilitating the targeted design of photothermal energy storage materials. This approach marks a significant stride in the field, offering innovative avenues for the development of advanced energy storage solutions.

---

Kai Wang

Kai Wang is an MS student under the guidance of Prof. Wei Feng at the School of Materials Science and Engineering, Tianjin University. His research interests include the synthesis, mechanism and applications of photo-responsive phase change azobenzene materials.

Wei Feng

Wei Feng is a professor at the School of Materials Science and Engineering in Tianjin University. He obtained his PhD degree from Xi’an Jiaotong University (China) in 2000. Then, he worked at Osaka University and Tsinghua University as a JSPS fellow and a postdoctoral researcher, respectively. In 2004, he became a full professor at Tianjin University. He has obtained the support of the National Science Fund for Distinguished Young Scholars in China. His research interests include photo-responsive organic molecules and their derivatives, thermal-conductive and high-strength carbon-based composites, and two-dimensional fluorinated carbon materials and polymers.

---

1. Introduction

With rapid economic growth and escalating energy demands, the inadequacy of conventional fossil fuel reserves has become increasingly apparent.^1,2 Moreover, excessive combustion of traditional fossil fuels contributes to environmental pollution and the greenhouse effect, thus intensifying the imperative for sustainable and innovative energy storage solutions.^3,4 The limitless availability and renewable nature of solar energy have rendered it a highly promising clean energy source.⁵ Currently, solar energy is harnessed primarily through photovoltaic and photothermal conversion.^6,7 Solar–thermal conversion systems have garnered considerable research attention owing to their advantages in terms of efficient energy utilization and versatile thermal management.^8–11 However, the utilization of solar energy is hindered by temporal and meteorological constraints, leading to its discontinuity and instability.^12,13 Consequently, developing effective methods for the conversion, storage, and utilization of solar energy has emerged as a key research concern.

As molecular photoswitches serve as potent photothermal conversion and storage materials, they are posited for application in molecular solar–thermal (MOST) systems.^14–18Fig. 1 illustrates the energy storage mechanisms of various common molecular photoswitches. Norbornadiene (NBD)/quadricyclane (QC),^19–22 anthracene,^23–25 FvRu₂(CO)₄,^26,27 and dihydroazulene (DHA)/vinylheptafulvene (VHF)^28–31 store energy via molecular structure alterations, differing from the energy storage mechanism of azobenzene (Azo).^32,33 Azo comprises two benzene rings linked by an Azo bond and exists in both cis and trans configurations.³⁴ At room temperature, the Azo molecule predominantly adopts a stable trans structure (S₀). Under ultraviolet (UV) light excitation, the molecule transitions to a metastable cis isomer (S₁), concurrently storing solar energy.^35–37 The unstable S₁ state, when stimulated by visible light or heat, surmounts the activation free energy of thermal restitution, reverting to the ground state and releasing the stored energy as thermal energy.^38,39

Fig. 1 The energy storage and release mechanisms of common photoswitches and PC-Azo in MOST systems, where hv: photon energy, Δ: heating, E_a: activation energy barrier, E_am: ambient heat energy, ΔH_iso: enthalpy of isomerization, and ΔH_c: enthalpy of phase transition.

The synthesis of new Azo derivatives, tailored to various application requirements, is achievable through the molecular design and functionalization of the Azo structure. This approach enables not only the adjustment of the absorption spectral range but also the enhancement of other photothermal properties, such as photothermal conversion efficiency.^40–42 Owing to its excellent photostability, good cyclic stability, and highly tunable structure, Azo is widely recognized as one of the most promising components of MOST systems.^43–48 However, the advancement of Azo–MOST systems confronts some challenges, including low energy density (ED), brief half-life (t_1/2), and complications in the solid-state charging process.⁴⁹ Another issue limiting the development of Azo–MOST systems is their poor thermal stability, as most Azo compounds tend to decompose at temperatures below 200 °C.⁵⁰ Additionally, while UV light catalyzes the trans-to-cis isomerization of Azo, it also induces a partial reversion of cis-Azo to its ground state, resulting in suboptimal solar energy utilization efficiency.⁵¹ Addressing these issues through molecular design remains an essential task for advancing clean energy technologies.^52–55

To enhance the performance and application scope of Azo–MOST systems, researchers are adopting three main strategies:^56–59 (i) template-based methods: these involve anchoring Azo onto carbon-based or polymer-based templates, extending the t_1/2 of cis-isomers. (ii) Molecular design methods: adjusting intermolecular forces by modifying molecular polarity or introducing hydrogen bonding, among other techniques, can effectively induce a red shift in the UV absorption spectrum and increase ED. (iii) Combination with phase-change materials: combining Azo with phase-change materials captures both phase-change and isomerization enthalpies. Moreover, composite material fabrication methods are being explored to harness the full solar spectrum, thereby optimizing solar energy absorption and utilization.

In recent years, researchers have innovatively designed and developed several phase-change azobenzene (PC-Azo) derivatives. These derivatives are adept at capturing both ambient heat and solar thermal energy under specific conditions.⁶⁰ Compared to ordinary molecular photoswitches, PC-Azo has a higher energy storage density. Fig. 1 illustrates the difference between the two energy storage mechanisms. Unlike conventional phase change materials (PCMs), PC-Azo derivatives, as photo controlled PCMs, facilitate nonthermogenic phase changes at ambient temperatures through the photoisomerization of photoswitches.⁶¹ They can also store energy stably over a wider temperature range.⁶² PC-Azo derivatives are versatile in their phase transition capabilities, encompassing not only solid-to-isotropic liquid phase transitions,^63,64 but also transition from crystal I to crystal II⁶⁵ and from gel to sol.^66,67 Among these, the solid-to-liquid phase change is particularly noteworthy for their pronounced shift in physicochemical properties and is the primary focus of this paper.

Molecular structure engineering of PC-Azo is an effective way to obtain photothermal conversion materials with high energy density and excellent thermal stability. Current research is focused on developing novel PC-Azo derivatives with excellent performance by varying substituents, as well as the types and lengths of alkyl chains.⁶⁸ Controlled heat release in these materials is achieved by manipulating the free volume size and intermolecular forces. Adjustments in the molecular structure enable control over light- and heat-induced isomerization and phase change processes.⁶⁹ However, traditional molecular design methods, reliant on extensive trial and error and experimental processes, are time-consuming and resource-intensive.⁷⁰ Furthermore, given the vast array of molecular structures, discovering potential compounds with excellent properties can be challenging. A major limitation is the difficulty in fully elucidating the relationship between the structure and performance, often resulting in low energy storage density within a limited temperature range.⁷¹

During the last few years, researchers have conducted numerous studies on the factors that affect isomerization and phase change processes, and some notable results have been obtained. With advancements in information technology, machine learning (ML) and artificial intelligence (AI) have begun to revolutionize molecular design.⁷² If the properties of unknown materials can be predicted using theoretical calculations and ML techniques, it would expedite research and synthesis of photothermal conversion materials with excellent energy-storage properties and be instrumental in creating materials capable of on-demand heat release.⁷³ This approach promises to transform the field of molecular photoswitching, reducing research timelines and opening new possibilities for high-performance material synthesis.

One of the important steps to realize the application of ML in the field of PC-Azo derivatives is to determine the required feature inputs and outputs (Fig. 2). This paper comprehensively reviews the research advancements related to PC-Azo derivatives, aiming to guide future development of high-performing MOST systems and the integration of ML in this domain. Initially, the fundamental mechanisms of PC-Azo are summarized, followed by a presentation of the essential performance parameters necessary for an effective MOST system. The study then focuses on the factors influencing isomerization and phase transitions in PC-Azo derivatives. And these are critical for feature selection. Finally, it investigates the role of theoretical calculations and ML in performance prediction and molecular design, proposing a strategy for merging ML with molecular design for the systematic development of PC-Azo derivatives.

Fig. 2 Schematic representation of the possible future of ML in the field of PC-Azo, and the importance of feature selection.

2. Phase-change azobenzene

Thermal energy storage stands as a pivotal technology for ensuring a stable and uninterrupted energy supply. It is broadly categorized into three main types: sensible heat storage, thermal chemical energy storage, and latent heat storage (phase-change energy storage). Phase-change energy storage, utilizing PCMs to absorb and release heat during phase transitions, boasts high ED, cost-effectiveness, and minimal environmental impact.^74–76 PCMs are classified into inorganic and organic types. Although inorganic PCMs encounter challenges such as subcooling and phase separation during phase transitions,⁷⁷ organic PCMs have gained considerable interest owing to their low supercooling, absence of precipitation, and stable performance.^78–80 In recent years, some pioneering studies have combined photoisomerization with phase change to create a series of PC-Azo derivatives. These derivatives are distinguished by their high ED and broad operational temperature range, marking an important stride in the field of photothermal energy storage.

2.1 Energy storage mechanism

Azo typically undergoes changes in polarity and volume following isomerization, enabling the trans-isomer to exhibit a higher crystallization temperature (T_c) and melting point (T_m) than the cis-isomer. Most trans-Azo derivatives tend to crystallize at room temperature due to the strong π–π interactions between aromatic groups. The T_c and T_m of the cis- and trans-formers can be adjusted to suitable ranges through molecular modifications or linkages to phase-change alkane chains. Fig. 3 illustrates the concurrent light-induced solid–liquid phase change and trans–cis isomerization of PC-Azo. In the ambient temperature range between the T_c of both isomers, the trans isomer is solid, while the cis-isomer is liquid. The trans-isomer absorbs photon energy to transform into the cis-isomer, simultaneously absorbing ambient heat to effectuate a solid–liquid phase change. This process enables the co-capture of photon energy and ambient heat energy (E_am). The cis-liquid can revert to the trans-solid state upon stimulation with external light or heat, releasing thermal energy in the form of phase-change enthalpy (ΔH_c) and isomerization enthalpy (ΔH_iso). This strategy can considerably increase the energy storage density of Azo derivatives.

Fig. 3 Schematic representation of the simultaneous photo-induced solid–liquid and trans–cis isomerization of PC-Azo, where UV is ultraviolet and Vis is visible light.

PC-Azo derivatives store and release heat energy based on isomerization.⁸¹ The ED and the rate of energy release are affected by the phase change and isomerization of Azo and are also dependent on the molecular structure and interactions. Thus, molecular structure optimization via molecular design is a key strategy for controlling energy release at different temperatures.^82,83

Research on PC-Azo derivatives has primarily focused on three aspects: (i) optimizing the molecular structure to achieve appropriate T_c and T_m, (ii) enhancing the stability of the cis-isomer to extend its t_1/2, and (iii) optimizing ΔH_c and ΔH_iso to increase the ED.

Most reported phase-change molecular photoswitches share a common molecular structure, wherein the photoswitch is connected to an aliphatic tail, facilitating the introduction of various substituents and alkyl chains through molecular modifications. Although some current molecular photoswitches efficiently store energy from visible light, shifting the absorption band to longer wavelengths often leads to rapid heat release or a reduction in ED. This necessitates rational molecular design. One promising approach is to explore new molecular photoswitches.^84,85 A series of heterocyclic Azo derivatives have been synthesized by researchers, and these photoswitchable molecules not only have long t_1/2, but can also achieve a high degree of isomerization in the visible light range.^86,87 Their photothermal properties can be further improved by modifying the substituents on heterocyclic azobenzenes.⁸⁸ Gonzalez et al.⁸⁹ prepared azodipyrazoles with small terminal substituents (Me, Et, etc.), showing excellent effective light penetration depth. Under UV irradiation, these photoswitches can undergo solid–liquid phase transitions with an ED of more than 300 J g⁻¹.

Furthermore, the ED of molecular photoswitches can further be augmented by amalgamating them with PCMs. The photoisomerization of Azo induces a conformational shift in the molecule, often altering the intermolecular interactions surrounding it, thereby influencing the physical properties of the molecule. Incorporating molecular photoswitches into conventional PCMs can modify the phase transition process, addressing the issue of heat loss susceptibility in standard PCMs at ambient temperatures. Azo dopants have the capability to adjust the phase transition temperature by disrupting the PCM stacking through spatial repulsion and dipole interactions, facilitating a reversible solid–liquid phase change.⁹⁰ This approach also offers improved control over PCM heat release.

2.2 Performance parameters

2.2.1 Isomerization properties and stability. The isomerization properties of Azo derivatives are the basis for energy storage and release, including the isomerization rate and the degree of isomerization. The conversion efficiency of solar energy depends on the degree of isomerization of Azo derivatives from trans to cis, while the energy release rate depends on the degree of isomerization from cis to trans. A great degree of isomerization can achieve efficient energy conversion. By rational molecular structure design, the degree of isomerization can be increased and the rate of isomerization can be adjusted. Cyclic stability means that a molecule undergoes multiple reversible changes without decaying. MOST systems with excellent cyclic stability can store and release energy over multiple cycles. Thermal stability refers to the stability of a compound at high temperatures and its resistance to thermal decomposition. Azo derivatives are often applied at higher temperatures, so their thermal stability is a very important performance indicator.

2.2.2 Storage energy density and half-life. The ED and t_1/2 of Azo derivatives are two important performance metrics, which also serve as the evaluation criteria for MOST systems.^91,92 The ED shows the energy storage capacity and is determined by the difference in energy levels between the steady state and the metastable state. The ED is usually obtained by differential scanning calorimetry (DSC) in units of mole-ED (kJ mol⁻¹), mass-ED (kJ kg⁻¹) and volume-ED (kJ L⁻¹). The magnitude of t_1/2 is strongly dependent on the cis-to-trans energy barrier. Under ambient conditions, the t_1/2 of Azo derivatives varies widely, ranging from seconds to minutes and even to days.^93,94 The Azo derivatives with a short t_1/2 are not suitable for energy storage, but are suitable for applications that require rapid exothermic release. The Azo derivatives with a long t_1/2 have the ability to store energy for long periods of time and can be used in Azo–MOST systems.

2.2.3 Solar spectrum match and melting point. Molecular photoswitches store and release energy by absorbing light at specific wavelengths. Thus, the solar spectrum matching ratio is a key parameter, which is directly related to the solar energy utilization. Nevertheless, the absorption spectra of most reported Azo derivatives overlap with only a small portion of the solar spectrum. And there is a competition between the cis- and trans-formers for the absorption of photons. Recently, researchers have prepared a number of molecular photoswitches for MOST systems through molecular design and material composites, which have improved the solar spectrum matching ratio.

Melting point (T_m) refers to the temperature at which a substance transitions from a solid to a liquid state and is a crucial performance parameter for PC-Azo derivatives. The magnitude and difference between the T_m of trans- and cis-Azo directly determines whether energy storage can be achieved by photo-induced solid–liquid phase change, and also determines the temperature at which the material can be used. Unfortunately, a large number of PC-Azo derivatives reported so far need to be heated above 60 °C to release energy, which is impractical for practical applications. Meanwhile, unlike the trans-isomers, most of the cis-Azo derivatives are thermally unstable, and therefore the T_m of the cis-formers determined by heating methods may be inaccurate, which limits the further development of PC-Azo derivatives.

For the Azo derivatives used in MOST systems, apart from requiring a high degree of isomerization, they should also possess a high ED, a long t_1/2, and a good match with the solar spectrum. For PC-Azo derivatives, appropriate T_m of trans- and cis-isomers are also necessary. Simultaneously combining and optimizing these attributes may pose challenges. Understanding the factors that influence these properties and how they can be predicted by machine learning forms the core of this paper.

2.3 Factors affecting performance

2.3.1 Substituents and alkyl chains. The size, position, number and polarity of the substituent group can have a great impact on the properties of PC-Azo derivatives. The electronic and site-resistive effects of substituents not only change the electron density on the aromatic ring but also affect the packing density of the molecule.⁹⁵ In 2004, Woolley et al.⁹⁶ investigated the effect of substituents with different electron-donating capacities on the properties of para-substituted Azo derivatives. They found that by changing the substituents, they not only modulated the absorption spectrum of Azo, but also greatly affected the t_1/2 of the cis-isomer. This suggests that the adjustment of the molecular structure can control the redshift in the π–π* band of Azo, thus regulate its thermal relaxation process.⁹⁷

Han et al.⁹⁸ prepared a series of para-functionalized Azo derivatives by controlling the size and polarity of the functional groups (Fig. 4a). They further revealed a close correlation between the properties of Azo derivatives and their para-substituents. As shown in Fig. 4b–e, highly electron-donating or electron-withdrawing groups lead to a significant difference between the phases of the trans (crystalline) and cis (liquid) isomers, while also decreasing the t_1/2 of the cis-isomer. All t_1/2 data were obtained from testing the cis-liquid at room temperature. They analyzed the storage properties of these compounds using computer simulations, which demonstrated that the shape and size of the substituent groups have a strong influence on latent heat (ΔH_c), as well as affecting the molecular stacking and the weight of the compound. For example, Azo derivatives with electron-donating groups MeO and EtO increased ΔH_iso and decreased ΔH_c. Most importantly, the photo-switching properties of the trans isomer in the condensed phase and the subsequent formation of a stable cis liquid determined the feasibility of energy storage in MOST. Fig. 4f shows the screening process of PC-Azo derivatives. By optimizing the molecular structure and chemical groups, they prepared molecular compounds with higher phase transition storage coefficients and energy densities, achieving a maximum ED of 300 J g⁻¹ (Fig. 4d). A comparison of mass-ED revealed the need to reduce the molecular weight of the molecular photoswitches. A negative correlation is observed between latent heat and molecular weight, whereas a positive correlation is detected between dipole moment and T_m.

Fig. 4 (a) Studied Azo derivatives and their σ. (b) T_m of all Azo derivatives with different σ. (c) T_m difference between the trans- and cis-isomers. (d) Total ED of 1–9, including ΔH_iso and ΔH_c. (e) t_1/2 of cis-isomers at 25 °C. (f) Selection process of PC-Azo derivatives.⁹⁸ Copyright 2021, The Royal Society of Chemistry.

Saalfrank et al.⁹⁹ systematically investigated the effect of different types, numbers and positions of substituents on Azo derivatives, using computational methods of quantum chemistry.¹⁰⁰ The research revealed that Azo derivatives with electron-donating substituents (–OMe, –NH₂) at the ortho-position are more likely to exhibit higher energy barriers (E_a) and therefore possess longer t_1/2 compared to pure Azo molecules. This increased E_a is associated with the spatial instability of the compact cis-isomer. Han et al.¹⁰¹ reported compounds 1–5 with different substituents in the ortho-position, which undergo spontaneous isomerization and phase transitions in sunlight (Fig. 5). Among them, compounds 1, 2, and 4 exhibit concurrent phase transition and isomerization, while compounds 3 and 5 undergo a same-phase isomerization process. The melt phase of compound 3 is extensively supercooled due to a significant decrease in the planarity of Azo as a result of o-methoxy substitution, thus enabling photo-induced trans-to-cis isomerization in a supercooled liquid phase at room temperature. Compound 5 remains liquid at room temperature. Compound 1 exhibits the highest ED (70 kJ mol⁻¹) among all compounds, but ortho-substitution generally leads to a reduction in ΔH_iso compared to pristine Azo (41 kJ mol⁻¹) (Table 1, entries 3, 4, and 5). In the case of o-fluorinated compound 1, this is due to the lowered energy of its Z-isomer, which results from reduced electronic repulsion in its HOMO orbital. The substitution of F with Cl has minimal effect on the ΔH_iso of compounds 2, 4, and 5. For compound 3, the HOMO centered on the Azo group is in close proximity to the electron-rich methoxy groups for the E isomer geometry, raising its energy level. Isomerization to the Z isomer alleviates this interaction. The E-isomer crystallization enthalpy (ΔH_c) of compounds 1–4 is slightly reduced from that of the pristine Azobenzene equivalent (48 kJ mol⁻¹), as a result of reduced van der Waals interactions caused by ortho-substitution on azobenzene. The thermal stability of the cis isomer of ortho-functionalized Azo was significantly enhanced, with t_1/2 for 1–4 ranging from 258 days to 8.7 years, measured at 25 °C, and t_1/2 for 1–4 ranging from 9 to 40 days, measured at 50 °C.

Table 1 Summary of the thermal properties of phase change Azo derivatives

PC-Azo derivatives	T m-trans (°C)	T m-cis (°C)	ΔHiso (MJ kg^−1)	ΔHc (MJ kg^−1)	ΔHtotal (MJ kg^−1)	ΔHtotal (kJ mol^−1)	Ref.
1	55	Liq	0.24	0.06	0.30	62	98
2	78	Liq	0.25	0.1	0.35	79	98
3	45	−36	0.05	0.1	0.15	70	101
4	56	−42	0.05	0.09	0.14	66	101
5	78	Liq	0.05	0.07	0.12	62	101
6	61	—	0.15	0.1	0.25	67	102
7	44	—	0.17	0.07	0.24	64	102
8	36	—	0.14	0.06	0.20	56	102
9	80	31	0.13	0.04	0.17	64	103
10	84	40	0.12	0.08	0.20	80	103
11	56	—	0.04	0.06	0.10	36	104
12	45	—	0.08	0.07	0.15	67	104
13	43	—	0.06	0.03	0.09	45	104

---

Fig. 5 (a) Structural diagrams of compounds 1–5. The wavelength that maximizes E-to-Z isomerization of each compound is listed. E_storage: total energy storage.¹⁰¹ Copyright 2021, The Royal Society of Chemistry.

Different substituents affect the molecular weight, which in turn affects the properties of the molecule. In order to attenuate this effect, Feng et al.¹⁰² prepared three Azo derivatives with different substituents while maintaining similar molecular weights (Fig. 6a). As shown in Fig. 6b–d, the crystallization and melting processes of all E-s-Azo derivatives were observed very clearly, with differences in the rate and extent of crystallization. E-s-Azo-A, with an alkoxyl chain substituent, exhibited good flexibility and orderly molecular arrangement. It rapidly crystallized at higher temperatures (T_E-c-A = 34.76 °C), demonstrating a high degree of crystallinity and the highest melting point (T_E-m-A = 61.46 °C). However, when the alkoxyl chain contained a branching group, the regular arrangement of molecules was affected. Therefore, E-s-Azo-B required cooling to lower temperatures for crystallization (T_E-c-B = 19.72 °C), resulting in decreased crystallinity and melting point (T_E-m-B = 43.63 °C). The introduction of methyl substituents leads to a further reduction in the crystallization rate of E-s-Azo-C, with its cold crystallization temperature also decreasing to 18.14 °C. This also indicated the potential application of such materials in low-temperature environments. As shown in Fig. 6e, under UV irradiation, the s-Azo undergoes a phase transition, manifested by the melting of the crystal. Subsequently, upon blue light triggering, the Z-isomer undergoes conversion to the E-isomer, resulting in the reformation of the crystal structure. Alkoxyl-substituted Azo exhibited photothermal properties with a high heat release of up to 343.3 J g⁻¹ and 413 W kg⁻¹ within a wide temperature range of −60.49 to 34.76 °C, achieving a temperature increase of 6.3 at −5 °C.

Fig. 6 (a) Structural diagrams of the three s-Azo. (b) DSC curves of E-isomers of s-Azo-A, (c) s-Azo-B and (d) s-Azo-C. (e) Optical images of crystal structure changes of s-Azo-A under alternating UV and blue light irradiation.¹⁰² Copyright 2023, Science China Press.

One way to control the intermolecular interactions of PC-Azo derivatives is by controlling the type and length of the para-connected alkyl chains of the photoswitch. Not only can the energy storage density be changed, but also the thermal storage temperature as well as the phase state can be greatly affected. Feng et al.¹⁰³ reported two PC-Azo derivatives with different alkyl chain lengths, namely, T-Azo and F-Azo (Fig. 7a). Through optimization of the alkyl chain length, the T_c and T_m of the Azo derivatives were adjusted to suitable temperature ranges (Table 1, entries 9,10), and both photoswitches exhibited reversible solid–liquid phase change. Photoswitches with longer alkyl chains are more favorable for energy storage and photo-induced exotherm at room temperature because of stronger molecular interactions and steric hindrance. The ED (including ΔH_c and ΔH_iso) of F-Azo with longer alkyl chains was 0.20 MJ kg⁻¹, surpassing that of T-Azo (0.17 MJ kg⁻¹). When exposed to blue light, F-Azo was able to generate a temperature change of 2.1 °C above the environmental temperature, significantly higher than the 0.8 °C difference observed for T-Azo.

Fig. 7 (a) Schematic structure of T-Azo and F-Azo. Copyright 2020, Elsevier. (b) Schematic structure of Azo-C_n-Br.¹⁰⁴ Copyright 2022, American Chemical Society. (c) Schematic structure of Azo-coumarin derivatives. Optical micrographs from isotropic phase cooling using a polarizing microscope, (d) nematic phase of Azo-C₈-Br at 120 °C, (e) smectic A phase of Azo-C₁₄-Br at 170 °C, and (f) crystalline phase of Azo-C₁₀-Br at 150 °C. Magnification used is 10× and resolution is 200 μm.¹⁰⁵ Copyright 2019, Elsevier.

Yang et al.¹⁰⁴ adjusted the T_m and photo-induced solid–liquid phase change of Azo derivatives by modifying ortho-diethyl and altering the length of the alkyl chains. They designed and synthesized a series of photo-liquefiable Azo-C_n-Br compounds (n = 2, 4, 6, 8, 10, and 12) (Fig. 7b). At ambient temperature, Azo-C_n-Br compounds (n = 4, 6 and 10) remain stable in liquid form, whereas Azo-C_n-Br (n = 2, 8, and 12) exist in a solid state that can be transformed into a liquid phase upon exposure to UV irradiation. Detailed melting point data in shown in Table 1, entries 11, 12, 13. However, after inducing cis–trans conversion through visible light irradiation, the photo-liquefied Azo-C_n-Br (n = 2, 8, and 12) did not undergo recrystallization into a solid phase, even after an extended period of precipitation. It is possible that ortho-diethyl and flexible bromoalkyl chains affect the molecular rearrangement and inhibit the restoration of the arrangement of the trans-Azo-C_n-Br (n = 2, 8, and 12) molecule. In conclusion, the reversible trans–cis isomerization under alternating UV and visible light allows their application in solvent-free MOST systems. Eswaran et al.¹⁰⁵ synthesized a range of new Azo-coumarin derivatives with varying alkyl chain lengths (C_nH_2n+1, n = 6, 8, 10, 12, and 14) and characterized their liquid-crystalline properties and photo-switching behavior (Fig. 7c). The photophysical and spectral characteristics of Azo-coumarin derivatives depend on the specific substituents on the coumarin ring. The variation in phase change temperature and mesomorphic phases can be attributed to the increase in carbon atoms within the terminal alkoxy chains. The Azo coumarin derivatives with C₁₄ alkyl chains exhibited Smectic-A while the rest of the longer alkyl chains with Azo showed a nematic phase (Fig. 7d–f). The introduction of coumarin and alkoxyl chains into the Azo moiety played a significant role in enhancing the liquid-crystalline properties.

In addition, there are some reports exploring the effect of substituents on Azo polymers on the properties. Wu et al.¹⁰⁶ investigated the effect of electron-donating groups on the photo-induced solid-to-liquid change of Azo-polymers containing azobenzene in the side chain. It was found that Azo-polymers with para-strong electron-donating groups exhibit lower phase transition temperatures and faster phase transition rates, while those with weak electron-donating groups display higher phase transition temperatures and slower phase transition rates. Introducing electron-donating groups significantly alters the phase change behavior of Azo-polymers. This effect is mainly attributed to the modulation of the light absorption properties and photo-isomerization reaction rates of the Azo groups by electron-donating groups.

2.3.2 Asymmetric aromatic heterocycles. Many studies have focused on developing heterocyclic Azo molecules to increase molecular diversity and further modulate the performance of photoswitches. Asymmetric aromatic photoswitches can be obtained by replacing one or two phenyl rings with heteroaromatic rings.

In recent years, research has been conducted on some dual heteroaromatic Azo photoswitches (Het-N=N-Het), while the asymmetric dual heteroaromatic Azo switches (Het1-N=N-Het2) that can combine the advantages of different heterocycles have received little attention. Lin et al.¹⁰⁷ have synthesized some new five-membered “heteroaryl Azo” photoswitches that undergo reversible isomerization in visible light (Fig. 8). The spectral properties of these photoswitches are quite different from those of conventional Azo and other heteroaryl Azo. For example, the λ_max (π–π*) of 1 is 363 nm, which is red-shifted by 47 nm compared to conventional Azo. The spectral bands of both trans- and cis-isomers of 1 are red-shifted, enabling reversible isomerization of the photo-switching under visible (405 and 525 nm) irradiation, which provides a good ratio of the trans- to cis-isomers and a long t_1/2 for the cis-isomers at the photo-stationary state (PSS). The photo-switching properties (λ_max, t_1/2 and PSS ratio) of heterocyclic Azo can be enhanced by modifying adjacent- and para-substituents on the phenyl ring. Among them, 6 with a neighboring NH₂ substituent group can undergo reversible photo-isomerization under visible light irradiation at longer wavelengths (525 nm and 625 nm), with less compensating effects on thermal stability and a longer t_1/2 compared to common Azo.

Fig. 8 Schematic structure of differently substituted heterocyclic Azo switches.¹⁰⁷ Copyright 2023, American Chemical Society.

Furthermore, the molecular design of asymmetric Azo has focused mainly on designing new heterocycles, while the exploration of substitution effects has been neglected. Rational structural substitutions are expected to enhance the performance of photoswitches by altering the geometry and electronic structure. The rational combination of heterocycles and appropriate substitutions is crucial for the development of high-performance photoswitches. Dong et al.¹⁰⁸ synthesized thiazole-based asymmetric two-heteroaromatic Azo-switches, which integrate the visible light-responsive characteristics of the thiazole ring with the susceptibility to substitution of the pyrazole ring (Fig. 9a). For the cis-isomer, the C–H bond in the adjacent position of the pyrazole ring and the weak C–H⋯N hydrogen bonding induced by the nitrogen atom on the thiazole ring are conducive to the construction of a unique near-T-configuration and planar conformation. The adjacent carbonylation can induce intramolecular lone-pair⋯π interactions because of the strong electron-absorbing effect, which can lead to the construction of a perfect T-configuration. Thiazolylpyrazoles achieve nearly quantitative visible-light isomerization in both directions, and the thermal t_1/2 of the cis-isomer is several days long at 25 °C, as shown in Fig. 9b and c. In contrast to the adverse impact of o-methylation, the o-carbonylation of the pyrazole ring can effectively enhance the stability of the cis-isomer through the induction of favorable intramolecular interactions, including dispersion interactions, C–H⋯N hydrogen bonding, and lone-pair⋯π interactions. This research combines theoretical calculations with an in-depth study of the structure–property relationships, revealing the unique structural and performance advantages of asymmetric bisheterocycles. Furthermore, it emphasizes the importance of rationally combining two heterocycles and implementing appropriate structural substitutions in the development of two-heteroaromatic Azo compounds.

Fig. 9 (a) Schematic structure thiazolylazopyrazoles 1–9. (b) t_1/2 of cis-1–8 at 25 °C in acetonitrile. (c) Molecular and geometrical structures of 4 and 9–11. The illustration displays the distance between the S atom of thiazole and the ring centroid of pyrazole.¹⁰⁸ Copyright 2023, Wiley.

In conclusion, heteroaromatic azoles offer excellent stability, superior spectral separation properties and greater functional design flexibility. Several studies have used them as core structures for PC-Azo derivatives. Table 2 shows a list of heterocyclic Azo switches and their key properties.

Table 2 A list of heterocyclic Azo switches and their key properties, where λ represents the excitation wavelength

Azo switches	π–π* (E) λmax (nm)	Photoconversion		t 1/2 (25 °C)	Solvent	Ref.
		E→Z (λ)	Z→E (λ)
	364	85 (405 nm)	81 (525 nm)	2.8 h	Acetonitrile	107
	351	94 (405 nm)	63 (525 nm)	7.2 h	Acetonitrile	107
	384	>92 (405 nm)	86 (525 nm)	0.25 h	Acetonitrile	107
	366	89 (400 nm)	96 (550 nm)	2.1 d	Acetonitrile	108
	377	98 (400 nm)	97 (550 nm)	13.1 h	Acetonitrile	108
	369	97 (400 nm)	96 (550 nm)	3.9 d	Acetonitrile	108
	385	88 (408 nm)	82 (532 nm)	2.95 h	Dimethyl sulfoxide	109
	335	>98 (355 nm)	>98 (532 nm)	10 d	Dimethyl sulfoxide	109
	356	85 (400 nm)	>99 (532 nm)	7 h	Acetonitrile	110
	325	97 (350 nm)	99 (523 nm)	72 d	Dimethyl sulfoxide	110
	327	98 (350 nm)	99 (523 nm)	681 d	Dimethyl sulfoxide	110
	421	84 (448 nm)	—	520 s	Acetonitrile	111

---

Han et al.¹¹² synthesized new PC-Azo derivatives through dodecanecarboxylic acid group functionalization of arylazoles (Fig. 10a). The crystallinity of the trans-isomers was improved by the substituents, which enhanced the intermolecular van der Waals forces. The bulk and polarity of the dodecanoate moiety improve the stability of the liquid phase of the cis-isomer. Because of the long t_1/2 of aryl azoles, the total thermal storage time is longer and the temperature range of thermal storage is wider. As a result, the material not only stores latent heat for long periods of time, but also undergoes light-triggered crystallization at low temperatures down to −30 °C. However, the modification of this substituent also reduces the energy level difference between the cis- and trans-isomers and limits the high rate exotherm at low temperatures. Furthermore, the presence of asymmetric aromatic rings in azo compounds enables them to undergo cis-to-trans isomerization upon interaction with protons and metal ions, thereby providing the potential to create dual-responsive materials capable of releasing heat both under chemical stimuli and light irradiation. Three heat storage and release schemes involving different activation methods have been devised, such as optical, thermal, or combined approaches, all of which can produce liquid cis-isomers that store thermal energy (Fig. 10b–e).

Fig. 10 (a) Schematic representation of Azo derivatives containing different heterocycles. Coloured circles represent structural modifications. (b) Schematic representation of methods I and II for the activation of aryl pyrazole derivatives with latent heat storage in a stable liquid phase. (c) DSC plots of the trans- and cis-isomers of compound 4' as a representative of group I and II compounds. (d) Schematic representation of method III for the activation of arylazo-pyrazole derivatives, where the latent heat can be stored in a stable liquid phase. (e) DSC plots of the trans- and cis- isomers of compound 3, the only group III compound that does not require the absorption of photons for activation.¹¹² Copyright 2020, American Chemical Society.

Wu et al.¹¹³ presented two completely reversible solid–liquid transitions in a specific type of PC-Azo derivatives (Fig. 11b and c). They synthesized PC-Azo derivatives by modifying long-chain alkane molecules with a phenyl-azopyrazole group, and investigated the photo-controllable tunability of the solid–liquid transition of the material. The phenyl-azopyrazole group was chosen as a photo-switch because it has higher cis-thermal stability than pristine Azo. Based on the results of the thermal decomposition test, it was found that all four PC-Azo derivatives exhibited negligible decomposition below 200 °C, suggesting their excellent thermal stability. With UV-vis spectra and NMR spectra, the good reversibility of the photo-isomerization of PC-Azo derivatives in solution and condensed states was demonstrated as shown in Fig. 11d. The synthesized PC-Azo derivatives can adapt to variable thermal charging temperatures through the switching behavior induced by light. Because of the combined ΔH_c and ΔH_iso, PC-Azo derivatives exhibit a remarkable mass-ED of up to 0.36 MJ kg⁻¹.

Fig. 11 (a) Schematic of the molecular structure of the PC-Azo derivative. (b) Photographs of PC-Azo derivatives in four states. (c) Two reversible phase transition behaviors. (d) Partial 1 H NMR spectra of the PC-Azo derivative 14 sample. (e) Possible applications for variable temperature thermal energy storage and heat enhancement on the basis of PC-Azo derivatives.¹¹³ Copyright 2023, American Chemical Society.

2.3.3 Molecular weight. Molecular weight is a crucial factor for the solid–liquid phase distinction of photo-controlled PCMs and is one of the most important factors affecting the phase change temperature. Besides, the mass-ED refers to the overall energy stored in a material per unit mass. Therefore, the molecular weight of the MOST system plays a crucial role in determining the mass-ED (MJ kg⁻¹) of the system. A larger molecular weight of a molecule usually means that it possesses more atoms and bonds, which can enhance intermolecular interactions and thus correspondingly increase the phase transition temperature. However, this can also make the phase change difficult. Longer molecular chains can affect the arrangement and conformation of the molecules during isomerization, which further affects the rate and efficiency of the isomerization. The relationship between molecular weight and the phase transition of PC-Azo is intricate and characterized by multiple factors. It is essential to take the effect of molecular weight into account when designing and preparing photo-controlled PCMs.

Wu et al.¹¹⁴ synthesized linear liquid crystalline Azo polymers with different molecular weights using non-cross-linked Azo polymers. The M_n of the Azo polymers ranged from 5–100 kg mol⁻¹. With the increase in molecular weight, the T_g of trans-Azo polymers increases from 48 to 80 °C. In addition, all Azo polymers showed liquid crystal phases above their T_g. Azo polymers (5–53 kg mol⁻¹) lack entanglement between their polymer chains, which makes them stiff and fragile. However, Azo polymers (80–100 kg mol⁻¹) have characteristics of flexibility, stretchability and easy processing because of the entanglement between the polymer chains. Pessoni et al.¹¹⁵ synthesized four different Azo polymers using nitrogen oxide mediated polymerization (NMP). The study reveals that both the molecular weight and chemical structure of the Azo polymer largely influence the properties of the reversible phase change. Specifically, the longer the length of the substituent or linker group, the faster the transformation rate. These findings contribute to a better understanding of the mechanism of the photo-reversible phase change and provide valuable insights for the development of innovative Azo-polymers with diverse applications.

Homma et al.¹¹⁶ synthesized some polymers with a similar Azo content but different molecular weights, poly(AzoAA-r-DMA) (Fig. 12a). They discovered a non-classical phase separation phenomenon in Azo-containing polymers during non-equilibrium photo-isomerization. It was found that polarity is not important for the phase separation phenomenon of polymers induced by photo-isomerization, but molecular weight plays a critical role. Phase separation occurred for cis–trans isomerization of high molecular weights induced by visible light, whereas phase separation was achieved for low molecular weights induced by UV light for trans–cis isomerization. This contradicts the commonly held view that phase separation is determined by changes in polarity during the isomerization of Azo. The phase separation temperature of the polymer in the trans-state increases with decreasing molecular weight, while the phase separation temperature of the cis-state remains approximately constant. This confirms the difference in molecular weight dependence of the phase separation behavior of the trans- and cis-states. The concentration of the polymer also affects the phase separation temperature. When the polymer concentration increases from 0.1 wt% to 1.0 wt%, the temperature decreases significantly. As shown in Fig. 12b and c, within the concentration range of 1.0–3.0 wt%, the phase separation temperature remains constant. This finding provides guidance for further research and modulation of the phase behavior of photo-isomeric polymers.

Fig. 12 (a) Schematic structure of the poly(AzoAA-r-DMA). (b) and (c) Variation of phase separation temperature with concentration/molecular weight. The plots in (b) and (c) depict the phase separation temperatures before and after UV irradiation, respectively. (d) Micrographs for 0.5 wt% poly(AzoAA_4.0-r-DMA_96.0)_36kDa in phosphate buffered saline at different temperatures. Scale bar: 10 μm.¹¹⁶ Copyright 2023, Wiley.

2.3.4 Others. During the cis-to-trans isomerization of Azo derivatives, significant volume changes occur because of conformational changes such as internal bond rotation and twisting.¹¹⁷ For most Azo derivatives, in the cis-isomer, the adjacent benzene rings are essentially arranged in a perpendicular orientation, occupying more space. In the trans-isomer, however, the benzene rings are arranged at a certain angle, resulting in a planar molecule with a smaller overall size. In 2011, Harms et al.¹¹⁸ employed a pulsed low-energy positron beam current to study the free volume change that occurs during the photo-isomerization of Azo-poly(methyl methacrylate) blends (40% Azo), and showed that the free volume change during photo-isomerization is approximately 10%. In fact, isomerization of Azo derivatives requires large free volume changes.¹¹⁹ Since conformational transitions cause volume changes, the molecule needs to have enough free volume to accommodate such conformational changes.¹²⁰

One scheme for changing the free volume of a molecule is to introduce bulky groups on the molecule while weakening the intermolecular interaction forces.¹²¹ Grossman et al.⁵⁴ synthesized three Azo derivatives with phenyl (1), biphenyl (2) and tert-butylphenyl (3), respectively (Fig. 13a). Because of the large volume of aromatic rings attenuating the π–π stacking, the prepared amorphous films achieved a high ED of 153 J g⁻¹. The incorporation of tert-butylphenyl effectively increased the thermal stability from 75 to 180 °C. This validates the successful inhibition of π–π stacking in the Azo derivative through the incorporation of tert-butyl groups, particularly when subjected to elevated temperatures during thermal annealing. Meanwhile, this research confirms the hypothesis that the intramolecular π–π stacking effect can be adjusted by varying the quantity of aromatic rings or the distance between the aromatic rings, thus achieving the modulation of ΔH_iso.

Fig. 13 (a) Schematic structure of compounds 1–3.⁵⁴ Copyright 2017, American Chemical Society. (b) Visual depiction of the Azo transition occurring in the presence of the UVA component of sunlight, as the trans-isomer (‘discharged’ state) transforms into the cis-isomer (‘charged’ state). (c) The first order rate constants using data from solution and 37% styrene SEBS. (d) Ratio of the normalized peak maxima of the cis- and trans-Azo absorption features in SEBS films with different styrene contents. These ratios were taken from the UV-vis results after 8 hours of UV charging and subsequently after 5 minutes of yellow light photoinitiation.¹²⁴ Copyright 2022, American Chemical Society.

Apart from altering the volume size of the molecule itself, it is also possible to change the substrate's free volume.¹²² Embedding Azo derivatives in a substrate is an important method for solving the problem of controlled heat release. The size of the free volume must be considered in the design and synthesis process. Li et al.¹²³ designed and synthesized ZIF-90-AAT by grafting 2-amino-Azo-toluene (AAT) onto a metal–organic framework, zeolite imidazole framework (ZIF). They further determined the average pore sizes of the parent metal–organic framework, ZIF-90, and ZIF-90-AAT as 12.66 nm and 1.38 nm, respectively, using a non-local density functional theory model. This result provided evidence for the successful insertion of AAT within the pores of ZIF-90. Furthermore, density functional theory calculations indicated that the minimum and maximum molecular sizes of the diazo-phenyl AAT were 5.5 Å and 11.66 Å, respectively, both smaller than the pore size. This ensured that the grafted Azo could undergo isomerization reactions inside the pores without hinderance. ZIF-90-AAT exhibits good thermal stability at temperatures below 300 °C. However, beyond this threshold, a gradual decrease in the mass of ZIF-90-AAT is observed, probably due to the decomposition of the MOF skeleton. Notably, ZIF-90-AAT demonstrates comparable thermal stability to that of ZIF-90, indicating that the grafting of AAT does not compromise the inherent thermal stability of ZIF-90. Feng et al.⁴⁷ developed a flexible and stretchable solar thermal fuel film using polynorbornene-templated Azo (PNB-Azo). They increased the free volume of the film by stretching the film, which improved the ED. At a 20% strain rate, the film displayed exceptional optical charging efficiency (85%), remarkable ED (49.0 W h kg⁻¹), and a substantial heat release rate stimulated by blue light (475 nm) at ambient temperature. Nevertheless, the quantification of free volume generation proved challenging due to the limitations imposed by macroscopic stretching. Cooper et al.¹²⁴ prepared a stretchable flexible solar thermal battery by integrating multi-walled carbon nanotubes and Azo derivatives into a stretchable matrix of a styrene–ethylene–butylene–styrene (SEBS) triblock copolymer (Fig. 13b). The free volume can be selectively adjusted by varying the rigid and flexible blocks of SEBS, which can be achieved by controlling the styrene content. Comparison of the calculations of the first-order rate constants indicates that the photo-isomerization process of the films with high styrene content is significantly suppressed (Fig. 13c). As illustrated in Fig. 13d, the degree of photo-isomerization of Azo derivatives after UV charging decreased significantly with increasing styrene content, which suggests that the free volume of the polymer matrix is important for the storage capacity of the MOSTs.

Numerous experiments and studies have shown that the size of the free volume of Azo derivatives affects the isomerization kinetics. The relevant values of molecular free volume can be determined through theoretical calculations and experimental tests. However, the effect of intermolecular forces cannot be neglected while studying the free volume.^125,126 This is because the introduction of functional groups to change the free volume also changes the magnitude of intermolecular interactions.

The trans-to-cis isomerization of Azo derivatives must overcome intermolecular interactions in order to achieve sufficient free volume. Additionally, the phase transition is a macroscopic representation of intermolecular forces that exerts a substantial impact on the energy storage capacity of MOST systems.^127,128 The interaction forces governed by structural factors primarily encompass π–π stacking between aromatic groups, van der Waals forces between substituents and hydrogen bonding.¹²⁹ Enhancing intermolecular interactions offers a promising avenue for augmenting the photon energy storage capacity. However, it must be noted that the greater the intermolecular forces, the greater the thermal barrier to isomerization and the more difficult it is to charge in the solid state.^130,131 Certain intermolecular forces, such as π–π stacking, affect the molar extinction coefficient, and being too large or too small is detrimental to energy storage.

In recent years, Azo derivatives have been grafted onto rigid templates to fabricate Azo-MOST systems with ordered structures and dense arrangements.^132–136 The close-packed structure significantly enhances intermolecular interactions. The Azo energy storage materials prepared using carbon nanotemplates can not only improve the ED and t_1/2 of the materials, but also improve the photothermal performance, thermal stability, and chemical stability. Feng et al.¹³⁷ prepared highly efficient carbon nanomaterials for energy storage by covalently linking Azo derivatives on graphene surfaces (Fig. 14). The covalent linkage of RGO (prepared graphene nanosheets) with Azo raises the thermal decomposition temperature of Azo, providing it with good thermal stability in the temperature range of 30 to 240 °C. Because of the high degree of functionalization and strong intermolecular hydrogen bonding and adjacent-induced intermolecular forces, the ED of the template reaches 112 W h kg⁻¹ with a long t_1/2 of 33 days, maintaining exceptional stability after 50 charge/discharge cycles.

Fig. 14 Schematic representation of the inter-planar bundling mechanism in high functionalization density Azo–RGO hybrids. In the solution phase (left), there is no packing interaction due to a large interlayer spacing, while in the solid-state forms (right), packing interaction occurs as a result of bundling with a reduced interlayer space.¹³⁷ Copyright 2015, The Royal Society of Chemistry.

Ren et al.¹³⁸ combined an intrinsically microporous polymer with 4-bromomethyl Azo to prepare MOST systems, Azo-PDAT and Azo-MTLE (Fig. 15a). The ED of 180.2 J g⁻¹ for Azo-PDAT was achieved through the cooperative influence of cation–π interactions, intrinsic microporosity, and template-enhanced steric strain. In contrast, the relatively low ED of 129.8 J g⁻¹ for Azo-MTLE was attributed to the relatively flexible polymer backbone and reduced micropores, which hindered photo-isomerization. Azo-PDAT demonstrated excellent durability and a remarkable capacity to release heat on a larger scale, exhibiting a temperature difference of 6.6 °C. This makes it a promising candidate for applications in intelligent temperature-controlled fabrics. Yu et al.¹³⁹ designed and fabricated a solid-state STF device by combining photo-liquefiable Azo (PLAzo) derivatives with a flexible fabric template, which enabled a solvent-free charging process and improved energy storage capacity. Intermolecular interactions between the fabric and the PLAzo derivative in the STF device inhibit isomerization and extend the t_1/2. Meanwhile, the template can increase the grafting density of PLAzo derivatives, which significantly improves the energy storage capacity, achieving an ED of 201 J g⁻¹.

Fig. 15 (a) Schematic structure of Azo-PDAT and Azo-MTLE.¹³⁸ Copyright 2023, American Chemical Society. (b) Azo-functionalized diacetylenes exhibiting diverse alkyl chain lengths, highlighted H-bonding units (in red), terminal functional groups, and an extended π-system. (c) First DSC traces of monomers 1–4 obtained during a temperature increase at 5 °C min⁻¹ (1–3) and at 2 °C min⁻¹.¹⁴² Copyright 2021, The Royal Society of Chemistry.

Even without the template, the intermolecular interactions among Azo derivatives significantly affect their energy storage performance. After trans-to-cis isomerization, the cis-isomer of PC-Azo destroys the original crystal structure and weakens the intermolecular forces, which causes a significant decrease in the T_m of the cis-isomer.^140,141 Intermolecular forces, including π–π conjugation, van der Waals forces, and hydrogen bonding, affect the isomerization and phase change processes.¹²⁹ Increasing intermolecular forces can enhance the ED, but excessive molecular stacking may hinder isomerization and phase transition. Anyway, it is theoretically feasible to control the phase transition temperature and energy storage density by adjusting the magnitude of intermolecular interactions.

Han et al.¹⁴² synthesized a range of symmetric diacetylenes and polydiacetylenes functionalized with Azo, capable of storing energy up to 176.2 kJ mol⁻¹ (Fig. 15b and c). They explored the opportunity of achieving high-energy storage through different levels of intermolecular interactions in trans- and cis-isomers. The effects of alkyl spacer length between the diacetylene core and end Azo groups, hydrogen bonding between molecules, and different functional groups on the properties were analyzed. The compound 1 with the shortest spacer exhibited the highest T_m of 266 °C, while compounds 2 and 3 with longer spacers had a T_m of 233 and 213 °C, respectively. The results indicated that the weakening of van der Waals forces and π–π stacking affected the ordered arrangement of molecules, leading to a decrease in the crystallization ability of Azo while favoring the reduction of T_c and T_m. The photo-isomerization of Azo groups and the resulting steric repulsion between side chains provided a reversible means to regulate the conjugation length of the polymer backbone. The ED was notably influenced by intermolecular interactions, including hydrogen bonding, van der Waals forces, and π–π interactions.

Yu et al.¹⁴³ proposed an effective method to increase the ED of PC-Azo derivatives by introducing “cation–π” interactions. By adding a small number of cations to the PC-Azo derivatives, the ED was increased by 25%. The competitive mechanisms of “cation–π” interactions and “π–π stacking” not only weaken the energy barrier but also reduce the high molar extinction efficiency. Zhang et al.¹⁴⁴ prepared two PC-Azo derivatives incorporating distinct R-rings onto the ammonium head of surfactants, along with linear tail chains and spacer chains, as depicted in Fig. 16a. The π–π interactions between the phenyl rings and the van der Waals forces between the linear chains and the phenyl rings maintain strong intermolecular interactions. As illustrated in Fig. 16b–d, the interfragment interactions between molecules were calculated using the independent gradient model based on Hirshfeld partition of molecular density (IGMH) method, which confirmed that the intermolecular interactions of Azo-1 and Azo-2 were significantly higher than those of Azo. At the same time, the presence of the bulky head, flexible linear tail chains and spacer chains weakens molecular stacking, ultimately achieving a balance between molecular free volume and interaction forces. An isothermal phase transition can occur at room temperature. The ED was 131.18 J g⁻¹ for charging the solid sample and 160.50 J g⁻¹ for charging the solution. It is worth noting that the molar isomerization enthalpy is 2.4 times higher than that of Azo. Moreover, all stored energy can be released as heat when stimulated by visible light.

Fig. 16 (a) Storage strategies for photons and thermal energy. The distance (d) between molecular chains is increased due to the bulky head, flexible tail chains and spacer chains that increase the free volume. Also due to stronger intermolecular interactions, the molecules have higher energy storage density. 3D isosurfaces illustrating the (b) Azo, (c) Azo-1, and (d) Azo-2 dimers generated using the IGMH method.¹⁴⁴ Copyright 2023, Wiley.

3. Machine learning-based performance prediction

3.1 Performance prediction methods

The structure of PC-Azo derivatives is highly customizable, given the wide variety of substituents, alkyl chains, and photoswitching molecules available. Photothermal performance can be substantially enhanced through rational selection and design. In conclusion, the discovery, design, and performance prediction of materials are essential. Traditional methods for material design and synthesis rely on complex calculations and extensive experimental data, leading to highly complex, time-consuming, and resource-intensive design and screening processes.

However, recent advances in information technology have revolutionized the ability to predict material properties and structures using simulation computation and machine learning (ML). Simulation computation employs digital simulation based on electronic structure theory and molecular simulation to simulate material structures and properties. Within this field, density functional theory (DFT) can calculate the electronic structure of materials, providing insights into the band structure, energy gap, charge distribution, etc.^145,146 Molecular dynamics (MD) simulation simulates the motion trajectory of atoms or molecules, elucidating the thermodynamic properties and dynamic behavior of materials.^147,148 These techniques assist researchers in predicting material performance, stability, and interactions, thereby guiding material design and optimization. However, this approach incurs substantial computational costs and time. Furthermore, its predictive accuracy depends on the mathematical model and computational methods employed, and its validity depends on the reliability and interpretability of the calculation results.

Classical molecular dynamics (CMD) simulations offer the possibility to study larger systems than DFT, but they require accurate interatomic potentials. Machine learning interatomic potentials (MLIPs) represent an alternative approach for predicting molecular properties and guiding molecular design; they describe and predict physicochemical properties such as interactions between atoms and binding energies by learning from extensive datasets containing atomic structures and corresponding potential energy data to construct a potential energy function.¹⁴⁹ This approach not only has high computational accuracy and efficiency, but also does not require an unaffordable computational cost.^150,151 This approach views molecules as a series of interactions between atoms, and predicts the properties of molecules by learning these interaction patterns and rules.¹⁵² For example, MLIPs can be utilized to forecast the geometric configuration, vibrational frequencies, and thermodynamic properties of molecules. Zhuang et al.¹⁵³ developed a first-principles multiscale modeling approach that combines ab initio calculations with macroscopic continuum scaling using MLIPs. They successfully modelled the mechanical properties of complex nanostructures and demonstrated the superiority of MLIPs for performance prediction. In addition to this, they demonstrated the effective application of ML electronic potentials in evaluating the piezoelectric and bending electric response of 2D systems.¹⁵⁴ These works open up new possibilities for designing and optimizing materials while maintaining computational accuracy.

ML, a subset of artificial intelligence, encompasses techniques such as supervised, unsupervised, and reinforcement learning. Table 3 summarizes the common types of algorithms for machine learning as well as their advantages and disadvantages. These techniques can uncover hidden patterns and structural information in material data through extensive data processing and pattern recognition. ML is instrumental in establishing correlations between material properties and structures and in predicting optical, electrical, thermal, and other material properties.^155–158 The synergy of ML with computational chemistry is increasingly being utilized for molecular design and property prediction.^159–161 This integrated approach utilizes ML algorithms to analyze large datasets and extract valuable patterns and trends. Training models and algorithms within this framework enables highly accurate and efficient prediction of molecular properties, reactions, and behaviors. For example, Yan et al.¹⁶² successfully predicted the chemical stability of anion exchange membranes (AEM) in fuel cells using ML methods. Wang et al.¹⁶³ used density functional theory calculations and ML methods to design materials with efficient small molecule receptor properties.

Table 3 Classification of commonly used ML algorithms and their advantages and disadvantages

Machine learning	Database dependency	Categorization	Advantage	Disadvantage
Supervised learning	Large labeled datasets	Decision tree	Accurate classification and prediction	Requires a large amount of labeled data and prone to overfitting
		Neural networks
		Naive Bayes
		Random forest
Unsupervised learning	Only unlabeled datasets	K-means	Automatically discovers patterns and structures within data	Results can be subjective and evaluating performance can be difficult
		Principal component analysis
Semi-supervised learning	A small number of small datasets	—	Trains on limited labeled data and abundant unlabeled data	High quality unlabeled data is required and selecting appropriate labeled samples is challenging

---

The application of ML has two main goals. The first is to predict the properties of molecules based on known molecular structures through forward analysis by ML. This means that ML algorithms can be utilized to train on a large amount of molecular structure and corresponding property data to build a model to predict the properties of a new molecular structure.¹⁶⁴ This approach can speed up the material screening and optimization process and improve the efficiency of molecular design. Second, ML can also be used to generate molecular structures with targeted properties by means of reverse design. Given the desired material performance parameters, possible molecular structures or parameter combinations can be inferred using ML algorithms. This capability allows us to search for the most suitable molecular structure or parameters based on the desired performance metrics when designing new materials. In addition, ML methods can also solve both forward problems and inverse problems in the partial differential equation.¹⁶⁵ Forward problem usually refers to predicting the output or outcome based on a given input. Whereas inverse problem is to infer possible inputs or parameters based on a given output or result.¹⁶⁶ ML methods can be used to predict outputs by training models, and also to estimate possible inputs or parameters by inverse reasoning.¹⁶⁷ In the field of materials chemistry, the accuracy and reliability of the model can be further improved by incorporating experimental data, which can be better applied to molecular design and materials research. Despite certain challenges in model building and training for ML, such as data availability, model interpretability, and overfitting, the amalgamation of ML with computational chemistry holds immense potential as a catalyst for advancements in material molecular design, drug discovery, and other domains.^168–171

3.2 Application in MOST systems

Dabrowa et al.¹⁷² reported that selective and predictable control of thermal trans-to-cis isomerization of Azo urea derivatives can be achieved through anion binding. They found that different anions can alter the electron distribution and bond lengths of the molecule, thus affecting the rate and degree of the thermal isomerization reaction. The results of quantum chemical calculations demonstrate that the kinetic rate constants for the trans-to-cis isomerization depend on the binding constants of the trans body to the anion and the first order constants of several reactions. The rate of the whole isomerization process is controlled by the concentration of anions and their binding affinity with the Azo host (cis-isomer). As the concentration and binding affinity of the anion increase, the rate constant also increases. This finding demonstrated the utility of kinetic data as a straightforward and precise means of determining both the association constant between the trans-isomer and an ionic guest, as well as the maximum rate constant alteration for thermal trans–cis isomerization.

Wei et al.¹⁷³ investigated the effect of side chain π–π stacking on the properties of Azo polymers through molecular dynamics simulations. The results show that the thermal conductivity of the polymers can be effectively changed by adjusting the π–π stacking structure of the side chains. The trend and amount of thermal conductivity enhancement of the π–π stacking structure in the results are consistent with the experimental results. When strong π–π stacking interactions exist between the side chains, the polymers have high thermal conductivities; while when the π–π stacking interactions between the side chains are destroyed, the thermal conductivities of the polymers decrease significantly. This switching effect of thermal conductivity can be realized through changes in light or temperature. In addition, it was found that the chain length and side chain density of the polymer also have an effect on the thermal conductivity. Wang et al.¹⁷⁴ investigated the physical properties and temperature differences during the isomerization process of phase change materials doped with Azo. As shown in Fig. 17a–c, they established a microscopic model of the optically controlled composite system (capric acid/4-(phenyldiazenyl)phenyl decanoate) and used molecular dynamics simulations to investigate the isomerization of Azo. Phase transition is a thermodynamic phenomenon involving the transformation of a substance from one phase to another. During a phase transition, the structure and properties of the substance change significantly, such as density, heat capacity and thermal conductivity. The self-diffusion coefficient is one of the important parameters describing the diffusive motion of the molecules of a substance in a medium and can therefore be determined by observing a sharp change in the self-diffusion coefficient (D). Also, D can be obtained by calculating the long-time limit of the mean square displacement (MSD). They used the self-diffusion coefficient D calculated from the MSD of hydrogen atoms to determine the phase transition temperature of decanoic acid (Fig. 17d–i). Moreover, the study analyzed the thermal conductivity of the composite system using non-equilibrium molecular dynamics. The investigation revealed that the aggregation and nucleation of trans-Azo groups exhibited enhanced thermal conductivity in comparison to the cis-isomers. Overall, this research contributes to a better understanding of optically controlled phase change processes.

Fig. 17 Initial model representations for the (a) pure PCM system, (b) amorphous trans-PC-Azo, and (c) crystalline trans-PC-Azo. MSD and self-diffusion coefficient for (d–e) pure capric acid, (f–g) trans-PC-Azo, and (h–i) cis-PC-Azo.¹⁷⁴ Copyright 2022, American Chemical Society.

In addition to simulation calculations, some machine learning based phase prediction methods have been proposed. Armeli et al.¹⁷⁵ built a dataset by collecting experimental data on a large number of organic compounds, including information on molecular structure, physical properties and glass transition temperature (T_g). Two distinct input modes were devised, wherein the molecular information is encoded through either the type and quantity of functional groups or derived from a SMILES string. They then trained these data using the extremely randomized trees procedure to find potential correlations between the molecular structure and T_g of organic compounds. The results of the study show that through ML methods, it is possible to build a predictive model that can predict the T_g of organic compounds with a certain degree of accuracy. Patra et al.¹⁷⁶ developed a deep learning method called dPOLY for predicting polymer phases and phase transitions. Fig. 18 shows the workflow of dPOLY. By applying convolutional neural networks to structural data of polymers, automatic classification and identification of different polymer phases were achieved. The method can also predict polymer phase change temperatures and types, including solid-state transitions, liquid crystal transitions, and glass transitions. The results demonstrate that dPOLY exhibits good accuracy and robustness in predicting polymer phases and phase transition properties. More importantly, the application of dPOLY helps accelerate material research and development processes while providing guidance and optimization strategies for designing new polymer materials.

Fig. 18 The dPOLY workflow involves five steps: generating molecular dynamics (MD) trajectories, building an autoencoder, labeling frames away from intersections, developing a deep neural network (DNN) prediction model, and predicting and analyzing frame labels in feature space to identify intersections.¹⁷⁶ Copyright 2021, American Chemical Society.

Yu et al.¹⁷⁷ investigated the photochemical reactions of Azo at different wavelengths by means of nonadiabatic kinetic simulations, in particular the excitation to the nπ* and ππ* states. The simulations utilize the ab initio multiple spawning (AIMS) method in combination with the floating occupation molecular orbital hole–hole Tamm–Dancoff approximated density functional theory (FOMO-hh-TDA) electronic structure method. The FOMO-hh-TDA method is capable of capturing the electron correlation effects necessary for describing electronic surface crossings and excited-state reaction paths in Azo, while maintaining computational efficiency. The simulations provide insights into the photo-dynamics of Azo, allowing for a quantitative analysis of the excited-state decay process. Two common S₁/S₀ deactivation pathways are identified, irrespective of the excitation wavelength: one leading to the formation of the isomerized photoproduct (reactive pathway) and the other returning to the photo-reactants (unreactive pathway). Although there is a high-energy deactivation pathway associated with excitation to the ππ* state, it is a minor pathway that mainly favors the formation of the photoproduct cis-Azo. Thus, this pathway does not significantly influence the wavelength-dependent photochemistry of Azo. Instead, changes in the excitation wavelength can modulate the frequency of accessing these two common S₁/S₀ deactivation pathways, thereby impacting the quantum yield of photoisomerization in Azo. By quantitatively analyzing the frequencies at which these pathways are accessed, the simulations can predict the quantum yield of photoisomerization.

By simulating reactions, it becomes possible to discriminate photoswitches with high isomerization quantum yield from those with other photophysical properties. However, due to the need for a large number of trajectories and expensive quantum chemical methods to explain non-adiabatic excited-state effects, these simulations are rarely used. Liu et al.¹²¹ utilized an ab initio high-throughput simulation method to accelerate the research on molecular photo-switch design. In this way, possible metastable structures of specific materials can be analyzed and determined from a large number of molecules composed of elements abundant on Earth. In the screening step, possible metastable states are explored by forcing the molecular structure to rotate around specific bonds, and relaxed structures with significant energetic differences from the initial structure are considered as candidate metastable structures. This approach enables the identification of molecules with potential applications in solar thermal storage devices. Axelrod et al.¹⁷⁸ improved the simulation speed of Azo derivatives using a diabatic artificial neural network (DANN) and successfully screened 3100 hypothetical molecules to obtain molecules with high quantum yields (Fig. 19a and b). They also provided a tool that automatically and quickly predicts the isomerization barriers of Azo derivatives.¹⁷⁹ The potential energy barriers and absorption wavelengths of Azo derivatives can be calculated by entering the descriptive characters of the molecule. The descriptive characters can be generated programmatically or using programs such as Chemdraw. This method employs ML to compute potentials and utilizes cross-system virtual screening for molecular selection. It not only effectively screens a large number of Azo derivatives but also accurately predicts their t_1/2. This approach presents a valuable tool for designing and optimizing Azo derivatives with desired thermal properties. The findings highlight the potential of ML techniques to accelerate the discovery and development of functional materials with enhanced thermal stability.

Fig. 19 (a) Four different mechanisms of thermal isomerization. a or a′ ≈ 180°, ω ≈ 90°. (b) Active learning loop for training the neural network. NAMD: nonadiabatic molecular dynamics. TS: transition state.¹⁷⁹ Copyright 2023, American Chemical Society.

Koerstz et al.¹⁸⁰ used a high-throughput computational and virtual screening approach to identify potential candidate materials for solar thermal cells. First, they constructed a database of 230 billion molecular candidate materials (DHA/VHF) based on 42 different substituents (including hydrogen) and seven possible substituent positions. Then, using computational methods such as DFT and MD, these candidate materials were calculated and analyzed in terms of electronic structure, energy band structure, energy gap and optical properties. During the calculations, they used ML algorithms to process the large-scale data and identified potential solar thermal cell candidate materials by screening and sorting. Eventually, a number of candidate materials with good light absorption and thermal stability were successfully screened. Christensen et al.¹⁸¹ established a database consisting of 32,623 DHA/VHF derivatives and trained a neural network model capable of predicting dipole moments and energy. By inputting molecular features, the model can predict the solar cell performance of molecules, such as light absorption, charge transfer, and band structure. It results demonstrate that the neural network model exhibits high accuracy in performance prediction and effectively screens out candidate materials with potential for solar energy conversion. This approach provides an effective tool for the rapid screening and optimization of molecular solar cell materials, accelerating the process of molecular design and synthesis, and promoting advancements in the field of solar energy.

3.3 Future prospects

In the field of photothermal, simulation and ML have successfully screened many molecules that can be applied to the MOST system and accurately predicted properties such as t_1/2 and isomerization rate. In other fields, they have not only achieved accurate prediction of properties, but also successfully guided molecular design. However, simulation and ML also have obvious limitations. Virtual screening needs to be benchmarked to ensure that its predictions correlate with experimental results and the scale needs to be affordable. The biggest advantage of ML methods is that they can represent a variety of complex mappings that can be used for performance prediction. However, their accuracy can be exaggerated by the overlap of training sets and they may make meaningless predictions without incorporating physical principles and constraints. Combining semi-empirical and ML methods can reduce the prediction cost to some extent.¹⁸² In addition to this, ML requires a large amount of data to build a database. The database created is expected to grow continuously, highlighting the importance of automating quantum chemistry work.

PC-Azo derivatives, as very promising MOST systems, have highly controllable thermodynamic properties and excellent optical properties. Predicting the properties of PC-Azo derivatives and guiding the molecular design is not only beneficial for the development of photo-induced phase change materials with high energy storage, but also important for realizing on-demand design of molecules and on-demand exotherm. On the one hand, the reported PC-Azo derivatives all have similar molecular structures. On the other hand, the factors affecting the properties of PC-Azo derivatives have been extensively studied in recent years. So theoretically, ML can realize accurate prediction of the properties of PC-Azo derivatives and provide guidance for molecular design (Fig. 20). This paper explores the general process of performance prediction through ML, using PC-Azo derivatives as an example, which can also be applied to other materials. The whole process can be roughly divided into six parts:

Fig. 20 Simple flowchart for ML prediction of PC-Azo derivative properties.

(I) Data acquisition and feature extraction: the key to ML is data acquisition and feature extraction. This process requires the collection of a large amount of performance data, molecular structures, physical properties, thermodynamics and other parameters of PC-Azo derivatives. Feature extraction is then performed to transform this data into a form that can be understood by ML models through material genomics and other quantitative methods. A dataset is created from the collected data and the created dataset is divided into two parts: training set and test set. Specific feature extraction methods are important for different tasks as they can directly affect the performance of the algorithms. This paper describes the performance parameters that PC-Azo derivatives should have and devotes an extensive section to exploring the effect of the structure and properties of PC-Azo derivatives on performance. This will facilitate the selection of features being used to train the ML model.

(II) Model selection: among the many models of ML, neural network models can map various complex relationships and are more suitable for performance prediction. In addition to this, there are models such as random forest and support vector machine (SVM). The performance parameters to be predicted need to be determined before selecting the model. Choosing an appropriate ML model requires consideration of factors such as data types, data volume, feature importance, model complexity, interpretability, prediction accuracy and generalization ability, which need to be weighed and selected on a case-by-case basis. It is also possible to try multiple models and compare them to get the best performance prediction results.

(III) Model training: the model was trained using data from the training set, and the model parameters were continuously optimized in the process to achieve accurate prediction of PC-Azo derivative properties. The goal of training the model is to ensure that it fits the existing data as effectively as possible, and show good generalization ability on unknown data.

(IV) Performance prediction: Use a trained ML model to predict the performance of new PC-Azo derivatives, such as T_c, T_m, t_1/2, ΔH_c and other key performance parameters. By inputting the characteristic data of the new material, the important performance parameters of the material are predicted using the association laws learned by the model.

(V) Model evaluation and optimization: predictions need to be validated with test set data or experimental measurements to evaluate ML models. The model is optimized according to the actual situation to improve the prediction accuracy and generalization ability. In this process, it is worth noting to avoid the overfitting problem.

(VI) Molecular design guidance: based on the prediction results of the ML model, it can guide the design of new PC-Azo derivatives. That is, based on the performance predicted by the model, the molecular structure is adjusted to find more potential design solutions for PC-Azo derivatives. Or by visualizing complex mappings, the relationship between performance parameters and structural features and molecular properties can be found to provide guidance for the design of better PC-Azo derivatives.

The combination of simulation and ML has considerably expedited molecular design and demonstrated a wide range of applications in photothermal energy storage. Beyond the molecular design of PC-Azo derivatives, this approach can be extended to other photoswitchable molecules, unlocking new avenues in materials science. However, practical applications may encounter various challenges, and ML is not the sole approach for predicting material properties. Nevertheless, the integration of theoretical calculations and ML can deepen the understanding of the relation between the properties and structure of PC-Azo derivatives. This method may provide more precise strategies for the rational design of molecules with enhanced solar energy storage capacities. Moreover, it may advance material design and discovery processes and equip researchers with a powerful tool that is expected to play an important role in the development of solar energy storage materials.

4. Conclusions

PC-Azo derivatives can undergo phase changes at ambient temperatures under photocontrolled conditions. By harnessing both solar and ambient thermal energy, they can achieve a higher ED than non-PC-Azo systems without solvent usage. The properties of PC-Azo derivatives, such as the spectral absorption range, phase transition temperature, and energy storage density, can be altered by modifying the substituents and alkyl chains. The influence of free volume and intermolecular forces on these properties must be considered when altering these molecular structures. The employment of asymmetric heterocyclic Azo derivatives as photoswitches considerably prolongs their t_1/2 as well as diversifies the response and phase transition behaviors of PC-Azo derivatives. Extensive research has validated the high promise and effectiveness of PC-Azo derivatives as MOST systems.

However, the current molecular design of PC-Azo derivatives faces several challenges. First, the vast array of substituents and alkyl chains leads to lengthy preparation and performance testing cycles. Second, experimental research struggles to comprehensively consider multiple performance-affecting factors. To date, there is a lack of a conclusive theory that thoroughly explains the relationship between the molecular structure and phase change processes, limiting the discovery of new PC-Azo derivatives with optimal energy storage density and stability. Future molecular design efforts should concentrate on addressing these challenges, with specific emphasis on: (i) enhancing energy storage density and cycling stability, (ii) adjusting the spectral absorption range for improved solar energy utilization, and (iii) developing varied phase transition and response modes to accommodate different application scenarios. This focus will be pivotal in advancing the field of PC-Azo derivatives and optimizing their utility in MOST systems.

The advancement of information technology has significantly enhanced the application of ML in guiding molecular design. While the use of theoretical computing and ML is prevalent in other energy technology domains such as batteries and catalysts, its application in photothermal conversion research remains relatively unexplored. ML has the capability to process vast amounts of experimental data, identify relationships between the molecular structure and properties, and thereby enhance the efficiency and precision of molecular design. This approach can be equally beneficial in the development and investigation of PC-Azo derivatives, playing a crucial role in advancing the performance and application of photothermal conversion materials.

This article delves into and consolidates the factors that impact the performance of PC-Azo derivatives. These insights are instrumental in selecting feature values and determining molecular descriptors in ML models. Notably, the structure of currently known PC-Azo derivatives, typically featuring a molecular photoswitch linked to an alkyl chain, offers a solid basis for predicting the performance of new PC-Azo derivatives through ML or theoretical calculations. This structure-oriented approach provides valuable theoretical guidance for molecular design. Future research should merge experimental and theoretical methods to expedite the molecular design of energy storage materials characterized by high energy density and a broad spectral absorption range. Additionally, the integration of ML in photothermal conversion is imperative to stimulate molecular design and material innovation.

Further exploration into alternative triggering mechanisms, such as electrocatalysis, could facilitate the complete release of stored energy. Moreover, by incorporating PC-Azo derivatives into various devices, their applicability can be expanded to include solar energy storage, controllable heat release, and low-temperature thermal management. PC-Azo derivatives, which combine isomerization and phase change processes, are poised to become the most effective Azo systems, making important contributions to the field of new energy.

Author contributions

Kai Wang: conceptualization, methodology, investigation, and writing – original draft preparation. Huitao Yu: visualization and investigation. Jingli Gao: visualization and investigation. Yiyu Feng: visualization and investigation. Wei Feng: supervision, reviewing and editing.

Conflicts of interest

The authors declare no competing interests.

Acknowledgements

This work was financially supported by the National Key R&D Program of China (no. 2022YFB3805702), the National Natural Science Foundation of China (grant no. 52303101, 52327802, 52173078, 52130303, 51973158, and 51803151), the China Postdoctoral Science Foundation (2023M732579), the Young Elite Scientists Sponsorship Program by CAST (no. 2022QNRC001).

Notes and references

1. Z. Wang, P. Erhart, T. Li, Z.-Y. Zhang, D. Sampedro, Z. Hu, H. A. Wegner, O. Brummel, J. Libuda, M. B. Nielsen and K. Moth-Poulsen, Joule, 2021, 5, 3116–3136.
2. S. Chu and A. Majumdar, Nature, 2012, 488, 294–303.
3. L. C. Ferreira, M. Castro-Alférez, S. Nahim-Granados, M. I. Polo-López, M. S. Lucas, G. Li Puma and P. Fernández-Ibáñez, Chem. Eng. J., 2020, 381, 122275.
4. Y. Liu and C. Feng, Energy Econ., 2023, 118, 106504.
5. M. Perez and R. Perez, Sol. Energy Adv., 2022, 2, 100014.
6. M. Wang, X. Huang and H. Yang, Macromol. Mater. Eng., 2023, 308, 2300061.
7. Y. Huang, Y. Xu, H. K. Bisoyi, Z. Liu, J. Wang, Y. Tao, T. Yang, S. Huang, H. Yang and Q. Li, Adv. Mater., 2023, 35, 2304378.
8. X. Z. Zhao, Y. Y. Feng, C. Q. Qin, W. X. Yang, Q. Y. Si and W. Feng, ChemSusChem, 2017, 10, 1395–1404.
9. W. Luo, Y. Y. Feng, C. Cao, M. Li, E. Z. Liu, S. P. Li, C. Q. Qin, W. P. Hu and W. Feng, J. Mater. Chem. A, 2015, 3, 11787–11795.
10. H. Wang, Y. F. Yu, X. Y. Yang, S. Wang, J. Ge, Q. B. Yang, X. L. Zhou, G. Q. Zheng, K. Dai, X. Y. Dai, Y. Y. Feng, L. B. Huang and W. Feng, EcoMat, 2023, 5, e12337.
11. Y. Jiang, J. X. Liu, W. Luo, X. Q. Quan, H. J. Li, J. Huang and W. Feng, Surf. Interfaces, 2021, 24, 101071–101078.
12. D. G. Nocera, Acc. Chem. Res., 2017, 50, 616–619.
13. Q. Wang, C. Pornrungroj, S. Linley and E. Reisner, Nat. Energy, 2022, 7, 13–24.
14. S. P. Jeong, L. A. Renna, C. J. Boyle, H. S. Kwak, E. Harder, W. Damm and D. Venkataraman, Sci. Rep., 2017, 7, 17773.
15. X. Xu, B. Wu, P. Zhang, H. Yu and G. Wang, Sol. RRL, 2020, 4, 1900422.
16. Z. Wang, Z. Wu, Z. Hu, J. Orrego-Hernández, E. Mu, Z.-Y. Zhang, M. Jevric, Y. Liu, X. Fu, F. Wang, T. Li and K. Moth-Poulsen, Cell Rep. Phys. Sci., 2022, 3, 100789.
17. L. Dong, Y. Feng, L. Wang and W. Feng, Chem. Soc. Rev., 2018, 47, 7339–7368.
18. M. Mansø, A. U. Petersen, Z. Wang, P. Erhart, M. B. Nielsen and K. Moth-Poulsen, Nat. Commun., 2018, 9, 1945.
19. V. A. Bren, D. D. Alexander, I. M. Vladimir and V. A. Chernoivanov, Russ. Chem. Rev., 1991, 60, 451.
20. P. Lorenz, F. Wullschläger, A. Rüter, B. Meyer and A. Hirsch, Chem. – Eur. J., 2021, 27, 14501–14507.
21. J. Mony, C. Climent, A. U. Petersen, K. Moth-Poulsen, J. Feist and K. Börjesson, Adv. Funct. Mater., 2021, 31, 2010737.
22. Z. Wang, H. Moïse, M. Cacciarini, M. B. Nielsen, M.-A. Morikawa, N. Kimizuka and K. Moth-Poulsen, Adv. Sci., 2021, 8, 2103060.
23. A. Lennartson, A. Roffey and K. Moth-Poulsen, Tetrahedron Lett., 2015, 56, 1457–1465.
24. G. Ganguly, M. Sultana and A. Paul, J. Phys. Chem. Lett., 2018, 9, 328–334.
25. M. Y. Vorona, N. J. Yutronkie, O. A. Melville, A. J. Daszczynski, K. T. Agyei, J. S. Ovens, J. L. Brusso and B. H. Lessard, Materials, 2019, 12, 2726.
26. R. Boese, J. K. Cammack, A. J. Matzger, K. Pflug, W. B. Tolman, K. P. C. Vollhardt and T. W. Weidman, J. Am. Chem. Soc., 1997, 119, 6757–6773.
27. K. Moth-Poulsen, D. Ćoso, K. Börjesson, N. Vinokurov, S. K. Meier, A. Majumdar, K. P. C. Vollhardt and R. A. Segalman, Energy Environ. Sci., 2012, 5, 8534–8537.
28. I. L. H. Kjeldsen, J. F. Høvring, T. J. von Buchwald, A. E. Hillers-Bendtsen and K. V. Mikkelsen, Phys. Chem. Chem. Phys., 2022, 24, 5564–5577.
29. J. Mogensen, O. Christensen, M. D. Kilde, M. Abildgaard, L. Metz, A. Kadziola, M. Jevric, K. V. Mikkelsen and M. B. Nielsen, Eur. J. Org. Chem., 2019, 1986–1993.
30. Y. Kanai, V. Srinivasan, S. K. Meier, K. P. C. Vollhardt and J. C. Grossman, Angew. Chem., Int. Ed., 2010, 49, 8926–8929.
31. C. Schøttler, S. K. Vegge, M. Cacciarini and M. B. Nielsen, ChemPhotoChem, 2022, 6, e202200037.
32. A. Nisar, S. Tabassum, K. Ayub, T. Mahmood, H. AlMohamadi, A. L. Khan, M. Yasin, R. Nawaz and M. A. Gilani, Phys. Chem. Chem. Phys., 2023, 25, 20430–20450.
33. X. Xu, J. Feng, W.-Y. Li, G. Wang, W. Feng and H. Yu, Prog. Polym. Sci., 2023, 101782,  DOI:10.1016/j.progpolymsci.2023.101782.
34. A. K. Schnack-Petersen, M. Pápai and K. B. Møller, J. Photochem., 2022, 428, 113869.
35. E. Wei-Guang Diau, J. Phys. Chem., 2004, 108, 950–956.
36. B. Zhang, Y. Feng and W. Feng, Nano-Micro Lett., 2022, 14, 138.
37. F. Nainggolan, J. Mol. Model., 2023, 29, 177.
38. V. Quintano, V. Diez-Cabanes, S. Dell’Elce, L. Di Mario, S. Pelli Cresi, A. Paladini, D. Beljonne, A. Liscio and V. Palermo, Phys. Chem. Chem. Phys., 2021, 23, 11698–11708.
39. Z.-B. Wen, R.-F. Snap, J.-M. Raquez, N. A. Clark, K.-K. Yang and Y.-Z. Wang, Sci. China Mater., 2020, 63, 2590–2598.
40. H. Liu, J. Tang, L. Dong, H. Wang, T. Xu, W. Gao, F. Zhai, Y. Feng and W. Feng, Adv. Funct. Mater., 2021, 31, 2008496.
41. A.-L. Leistner and Z. L. Pianowski, Eur. J. Org. Chem., 2022, e202101271.
42. F. He, X. Ren, J. Jiang, G. Zhang and L. He, J. Phys. Chem. Lett., 2022, 13, 427–432.
43. Y. Feng, H. Liu, W. Luo, E. Liu, N. Zhao, K. Yoshino and W. Feng, Sci. Rep., 2013, 3, 3260.
44. E. Durgun and J. C. Grossman, J. Phys. Chem. Lett., 2013, 4, 854–860.
45. S.-T. Wang, P. Tan, W.-Q. Weng, L. Zheng, X.-Q. Liu and L.-B. Sun, Sci. China Mater., 2023, 66, 2726–2732.
46. X. Xu and G. Wang, Small, 2022, 18, 2107473.
47. L. Fu, J. Yang, L. Dong, H. Yu, Q. Yan, F. Zhao, F. Zhai, Y. Xu, Y. Dang, W. Hu, Y. Feng and W. Feng, Macromolecules, 2019, 52, 4222–4231.
48. W. He, P. Song, B. Yu, Z. Fang and H. Wang, Prog. Mater. Sci., 2020, 114, 100687.
49. A. K. Saydjari, P. Weis and S. Wu, Adv. Energy Mater., 2017, 7, 1601622.
50. X. Yang, S. Li, J. Zhao, X. Wang, H. Huang and Y. Wang, Sol. Energy Mater. Sol. Cells, 2021, 231, 111330.
51. Z. Shangguan, W. Sun, Z.-Y. Zhang, D. Fang, Z. Wang, S. Wu, C. Deng, X. Huang, Y. He, R. Wang, T. Li, K. Moth-Poulsen and T. Li, Chem. Sci., 2022, 13, 6950–6958.
52. Y. Chen, H. Yu, M. Quan, L. Zhang, H. Yang and Y. Lu, RSC Adv., 2015, 5, 4675–4680.
53. Z. Wang, R. Losantos, D. Sampedro, M.-A. Morikawa, K. Börjesson, N. Kimizuka and K. Moth-Poulsen, J. Mater. Chem. A, 2019, 7, 15042–15047.
54. E. N. Cho, D. Zhitomirsky, G. G. D. Han, Y. Liu and J. C. Grossman, ACS Appl. Mater. Interfaces, 2017, 9, 8679–8687.
55. F. A. Jerca, V. V. Jerca and R. Hoogenboom, Nat. Rev. Chem., 2022, 6, 51–69.
56. H. Wang, Y. Feng, J. Gao, W. Fang, J. Ge, X. Yang, F. Zhai, Y. Yu and W. Feng, Adv. Sci., 2022, 9, 2201657.
57. H.-T. Yu, J.-W. Tang, Y.-Y. Feng and W. Feng, Chin. J. Polym. Sci., 2019, 37, 1183–1199.
58. Y. Jiang, J. Liu, W. Luo, X. Quan, H. Li, J. Huang and W. Feng, Surf. Interfaces, 2021, 24, 101071.
59. X. Zhao, Y. Feng, C. Qin, W. Yang, Q. Si and W. Feng, ChemSusChem, 2017, 10, 1302.
60. R. Ehid and A. S. Fleischer, Energy Convers. Manage., 2012, 53, 84–91.
61. J. Hu, T. Song, M.-M. Yu and H. Yu, Chem. Mater., 2023, 35, 4621–4648.
62. M. Le and G. G. D. Han, Acc. Chem. Res., 2022, 3, 634–643.
63. U. Bhattacharjee, D. Freppon, L. Men, J. Vela, E. A. Smith and J. W. Petrich, ChemPhysChem, 2017, 18, 2526–2532.
64. Y. Kikkawa, M. Nagasaki and Y. Norikane, Phys. Chem. Chem. Phys., 2022, 24, 29757–29764.
65. P. Gupta, D. P. Karothu, E. Ahmed, P. Naumov and N. K. Nath, Angew. Chem., Int. Ed., 2018, 57, 8498–8502.
66. Y. Ni, X. Li, J. Hu, S. Huang and H. Yu, Chem. Mater., 2019, 31, 3388–3394.
67. M. Wang, C. Nie, J. Liu and S. Wu, Nat. Commun., 2023, 14, 1000.
68. H. Liu, T. Kobayashi and H. Yu, Macromol. Rapid Commun., 2011, 32, 378–383.
69. J. Gao, Y. Feng, W. Fang, H. Wang, J. Ge, X. Yang, H. Yu, M. Qin and W. Feng, Energy Environ. Sci., 2023, 0, e12607.
70. X. L. Pang Tan, T. Chen, Z. Qin, T. Yang, X. Liu and X. Liu, Chin. J. Org. Chem, 2021, 41, 2666–2675.
71. G. F. Martins, B. d P. Cardoso, N. Galamba and B. J. C. Cabral, Energy Fuels, 2023, 37, 1731–1756.
72. S. Lu, Q. Zhou, Y. Ouyang, Y. Guo, Q. Li and J. Wang, Nat. Commun., 2018, 9, 3405.
73. S. Gond, N. Krishnan and K. Ravi Kumar, Int. J. Energy Res., 2024, 25, 63–79.
74. J. W. Tang, Y. Y. Feng and W. Feng, Compos. Commun., 2021, 23, 100575–100581.
75. M. Felderhoff, R. Urbanczyk and S. Peil, Green, 2013, 3, 113–123.
76. K. Pielichowska and K. Pielichowski, Prog. Mater. Sci., 2014, 65, 67–123.
77. I. Shamseddine, F. Pennec, P. Biwole and F. Fardoun, Renewable Sustainable Energy Rev., 2022, 158, 112172.
78. A. S. Caldwell, G. Itskos and K. H. Sandhage, Mater. Today, 2021, 46, 9–17.
79. T. Yang, W. P. King and N. Miljkovic, Cell Rep. Phys. Sci., 2021, 2, 100540.
80. P. Khomein, A. Nallapaneni, J. Lau, D. Lilley, C. Zhu, S. Kaur, R. Prasher and G. Liu, Sol. Energy Mater. Sol. Cells, 2021, 225, 111030.
81. Z.-Y. Zhang, Y. He, Z. Wang, J. Xu, M. Xie, P. Tao, D. Ji, K. Moth-Poulsen and T. Li, J. Am. Chem. Soc., 2020, 142, 12256–12264.
82. K. Griffiths, N. R. Halcovitch and J. M. Griffin, New J. Chem., 2022, 46, 4057–4061.
83. E. R. Soulé, Chem. Phys. Lett., 2022, 794, 139503.
84. J. Calbo, C. E. Weston, A. J. P. White, H. S. Rzepa, J. Contreras-García and M. J. Fuchter, J. Am. Chem. Soc., 2017, 139, 1261–1274.
85. S. Crespi, N. A. Simeth and B. König, Nat. Rev. Chem., 2019, 3, 133–146.
86. A. H. Heindl and H. A. Wegner, Chem. – Eur. J., 2020, 26, 13730–13737.
87. A. Dolai, S. M. Box, S. Bhunia, S. Bera, A. Das and S. Samanta, J. Org. Chem., 2023, 88, 8236–8247.
88. V. Jelínková, A. Dellai, M. Vachtlová, M. Fecková, J. Podlesný, M. Klikar, F. Castet, A. Růžička, P. Pařík, O. Pytela and F. Bureš, J. Photochem., 2024, 449, 115390.
89. A. Gonzalez, M. Odaybat, M. Le, J. L. Greenfield, A. J. P. White, X. Li, M. J. Fuchter and G. G. D. Han, J. Am. Chem. Soc., 2022, 144, 19430–19436.
90. H. Liu, L. Zhang, G. Zhang, Q. Du, K. Wang, X. Luo and Z. Wu, Dyes Pigm., 2023, 215, 111300.
91. T. J. Kucharski, Y. Tian, S. Akbulatov and R. Boulatov, Energy Environ. Sci., 2011, 4, 4449–4472.
92. Q. Qiu, Y. Shi and G. G. D. Han, J. Mater. Chem. C, 2021, 9, 11444–11463.
93. T. Kobayashi, E. O. Degenkolb and P. M. Rentzepis, J. Phys. Chem. C, 1979, 83, 2431–2434.
94. O. Sadovski, A. A. Beharry, F. Zhang and G. A. Woolley, Angew. Chem., Int. Ed., 2009, 48, 1484–1486.
95. J. Yao, L. Sun, Y. Xiao, J. Liu, J. Jie, X. Liu, F. Yang, Y. Sun, R. Li and W. Hu, Sci. China Mater., 2023, 66, 1511–1517.
96. N. Pozhidaeva, M.-E. Cormier, A. Chaudhari and G. A. Woolley, Bioconjug. Chem., 2004, 15, 1297–1303.
97. G. Gabor and E. Fischer, J. Phys. Chem. C, 1971, 75, 581–583.
98. Q. Qiu, M. A. Gerkman, Y. Shi and G. G. D. Han, Chem. Commun., 2021, 57, 9458–9461.
99. H. M. D. Bandara and S. C. Burdette, Chem. Soc. Rev., 2012, 41, 1809–1825.
100. J. Dokić, M. Gothe, J. Wirth, M. V. Peters, J. Schwarz, S. Hecht and P. Saalfrank, J. Phys. Chem. A, 2009, 113, 6763–6773.
101. Y. Shi, M. A. Gerkman, Q. Qiu, S. Zhang and G. G. D. Han, J. Mater. Chem. A, 2021, 9, 9798–9808.
102. Q. Yang, J. Ge, M. Qin, H. Wang, X. Yang, X. Zhou, B. Zhang, Y. Feng and W. Feng, Sci. China Mater., 2023, 66, 3609–3620.
103. H. Liu, Y. Feng and W. Feng, Compos. Commun., 2020, 21, 100402.
104. Y. Yang, S. Huang, Y. Ma, J. Yi, Y. Jiang, X. Chang and Q. Li, ACS Appl. Mater. Interfaces, 2022, 14, 35623–35634.
105. E. Madiahlagan, B. N. Sunil, Z. Ngaini and G. Hegde, J. Mol. Liq., 2019, 292, 111328.
106. J. Xu, B. Niu, S. Guo, X. Zhao, X. Li, J. Peng, W. Deng, S. Wu and Y. Liu, Polymers, 2020, 12, 901.
107. R. Lin, P. K. Hashim, S. Sahu, A. S. Amrutha, N. M. Cheruthu, S. Thazhathethil, K. Takahashi, T. Nakamura, T. Kikukawa and N. Tamaoki, J. Am. Chem. Soc., 2023, 145, 9072–9080.
108. T. Dang, D. Dong, J. Zhang, Y. He, Z.-Y. Zhang and T. Li, Angew. Chem., Int. Ed., 2023, 62, e202301992.
109. C. E. Weston, R. D. Richardson, P. R. Haycock, A. J. P. White and M. J. Fuchter, J. Am. Chem. Soc., 2014, 136, 11878–11881.
110. Y. He, Z. Shangguan, Z.-Y. Zhang, M. Xie, C. Yu and T. Li, Angew. Chem., Int. Ed., 2021, 60, 16539–16546.
111. H. Wang, H. K. Bisoyi, X. Zhang, F. Hassan and Q. Li, Chem. – Eur. J., 2022, 28, e202103906.
112. M. A. Gerkman, R. S. L. Gibson, J. Calbo, Y. Shi, M. J. Fuchter and G. G. D. Han, J. Am. Chem. Soc., 2020, 142, 8688–8695.
113. S. Wu, Z.-Y. Zhang, T. Li, R. Wang and T. Li, ACS Mater. Lett., 2023, 5, 2019–2027.
114. M. Chen, B. Yao, M. Kappl, S. Liu, J. Yuan, R. Berger, F. Zhang, H.-J. Butt, Y. Liu and S. Wu, Adv. Funct. Mater., 2020, 30, 1906752.
115. L. Pessoni, D. Siniscalco, A. Boussonnière, A.-S. Castanet, L. Billon and N. Delorme, Eur. Polym. J., 2022, 174, 111297.
116. K. Homma, A. C. Chang, S. Yamamoto, T. Ueki and J. Nakanishi, Macromol. Rapid Commun., 2023, 44, 2300118.
117. A. A. Beharry and G. A. Woolley, Chem. Soc. Rev., 2011, 40, 4422–4437.
118. S. Harms, K. Rätzke, C. Pakula, V. Zaporojtchenko, T. Strunskus, W. Egger, P. Sperr and F. Faupel, J. Polym. Sci., Part B: Polym. Phys., 2011, 49, 404–408.
119. A. Natansohn, S. Xie and P. Rochon, Macromolecules, 1992, 25, 5531–5532.
120. H. Akiyama and M. Yoshida, Adv. Mater., 2012, 24, 2353–2356.
121. Y. Liu and J. C. Grossman, Nano Lett., 2014, 14, 7046–7050.
122. F. Cai, T. Song, B. Yang, X. Lv, L. Zhang and H. Yu, Chem. Mater., 2021, 33, 9750–9759.
123. C. Li, Z. Zhang, T. Zhong and G. Wang, J. Energy Storage, 2022, 52, 104971.
124. T. W. Colburn, A. C. Delmastro, M. Figueroa, F. Lopez and C. B. Cooper, J. Phys. Chem. C, 2022, 126, 15565–15572.
125. X. Zhang, L. Hou and P. Samorì, Nat. Commun., 2016, 7, 11118.
126. A. Kunz, A. H. Heindl, A. Dreos, Z. Wang, K. Moth-Poulsen, J. Becker and H. A. Wegner, ChemPlusChem, 2019, 84, 1145–1148.
127. S. Sheng, Z. Zhu, Z. Wang, T. Hao, Z. He and J. Wang, Sci. China Mater., 2022, 65, 1369–1376.
128. A. M. Kolpak and J. C. Grossman, Nano Lett., 2011, 11, 3156–3162.
129. H.-Y. Zhong, L. Chen, X.-M. Ding, R. Yang and Y.-Z. Wang, Sci. China Mater., 2018, 61, 1225–1236.
130. Y. Liu, H. Wang, P. Liu, H. Zhu, B. Shi, X. Hong and F. Huang, Angew. Chem., Int. Ed., 2021, 60, 5766–5770.
131. J. Hu, X. Li, Y. Ni, S. Ma and H. Yu, J. Mater. Chem. C, 2018, 6, 10815–10821.
132. Y. Feng, W. Feng, H. Noda, A. Fujii, M. Ozaki, K. Yoshino and K. Yoshino, J. Appl. Phys., 2007, 102, 053102.
133. Y. Feng, W. Feng, H. Noda, T. Sekino, A. Fujii, M. Ozaki and K. Yoshino, Carbon, 2007, 45, 2445–2448.
134. W. Feng, W. Luo and Y. Feng, Nanoscale, 2012, 4, 6118–6134.
135. T. J. Kucharski, N. Ferralis, A. M. Kolpak, J. O. Zheng, D. G. Nocera and J. C. Grossman, Nat. Chem., 2014, 6, 441–447.
136. W. Luo, Y. Feng, C. Qin, M. Li, S. Li, C. Cao, P. Long, E. Liu, W. Hu, K. Yoshino and W. Feng, Nanoscale, 2015, 7, 16214–16221.
137. W. Luo, Y. Feng, C. Cao, M. Li, E. Liu, S. Li, C. Qin, W. Hu and W. Feng, J. Mater. Chem. A, 2015, 3, 11787–11795.
138. H. Ren, H. Chen, M. Lu, X. Cui, H. Wang, L. Wang and Q. Tang, ACS Appl. Polym. Mater., 2023, 5, 5592–5599.
139. J. Hu, S. Huang, M. Yu and H. Yu, Adv. Energy Mater., 2019, 9, 1970145.
140. J.-a Lv, Y. Liu, J. Wei, E. Chen, L. Qin and Y. Yu, Nature, 2016, 537, 179–184.
141. W.-C. Xu, S. Sun and S. Wu, Angew. Chem., Int. Ed., 2019, 58, 9712–9740.
142. G. D. Han, S. S. Park, Y. Liu, D. Zhitomirsky, E. Cho, M. Dincă and J. C. Grossman, J. Mater. Chem. A, 2016, 4, 16157–16165.
143. T. Song, H. Lei, F. Cai, Y. Kang, H. Yu and L. Zhang, ACS Appl. Mater. Interfaces, 2022, 14, 1940–1949.
144. L. Zhang, H. Liu, Q. Du, G. Zhang, S. Zhu, Z. Wu and X. Luo, Small, 2023, 19, 2206623.
145. J. Yeston, Science, 2017, 356, 1041–1042.
146. B. Huang, G. F. von Rudorff and O. A. von Lilienfeld, Science, 2023, 381, 170–175.
147. H. Tafrishi, S. Sadeghzadeh and R. Ahmadi, RSC Adv., 2022, 12, 14776–14807.
148. Y.-X. Xu, H.-Z. Fan and Y.-G. Zhou, Rare Met., 2023, 42, 3914–3944.
149. V. L. Deringer, M. A. Caro and G. Csányi, Adv. Mater., 2019, 31, 1902765.
150. M. H. Müser, S. V. Sukhomlinov and L. Pastewka, Adv. Phys.: X, 2023, 8, 2093129.
151. Y. Liu, X. He and Y. Mo, npj Comput. Mater., 2023, 9, 174.
152. T. W. Ko and S. P. Ong, Nat. Comput. Sci., 2023, 3, 998–1000.
153. B. Mortazavi, M. Silani, E. V. Podryabinkin, T. Rabczuk, X. Zhuang and A. V. Shapeev, Adv. Mater., 2021, 33, 2102807.
154. B. Mortazavi, B. Javvaji, F. Shojaei, T. Rabczuk, A. V. Shapeev and X. Zhuang, Nano Energy, 2021, 82, 105716.
155. M. Ishfaq, M. Aamir, F. Ahmad, A. M. Mebed and S. Elshahat, ACS Omega, 2022, 7, 48139–48149.
156. Z. Zhu, Z. Rahman, M. Aamir, S. Z. A. Shah, S. Hamid, A. Bilawal, S. Li and M. Ishfaq, RSC Adv., 2023, 13, 2057–2069.
157. T. Mubashir, M. Hussain Tahir, Y. Altaf, F. Ahmad, M. Arshad, A. Hakamy and M. Sulaman, J. Photochem., 2023, 437, 114501.
158. C. Zhang, Nat. Energy, 2018, 3, 535.
159. A. Mahmood, A. Irfan and J. L. Wang, Chin. J. Polym. Sci., 2022, 40, 870–876.
160. M. R. S. A. Janjua, A. Irfan, M. Hussien, M. Ali, M. Saqib and M. Sulaman, Energy Technol., 2022, 10, 6934–6941.
161. A. Mahmood, Y. Sandali and J. L. Wang, Phys. Chem. Chem. Phys., 2023, 25, 10417–10426.
162. X. Zou, J. Pan, Z. Sun, B. Wang, Z. Jin, G. Xu and F. Yan, Energy Environ. Sci., 2021, 14, 3965–3975.
163. A. Mahmood, A. Irfan and J.-L. Wang, Chem. – Eur. J., 2022, 28, e202103712.
164. Z. Liu, D. Zhu, L. Raju and W. Cai, Adv. Sci., 2021, 8, 2002923.
165. E. Samaniego, C. Anitescu, S. Goswami, V. M. Nguyen-Thanh, H. Guo, K. Hamdia, X. Zhuang and T. Rabczuk, Comput. Methods Appl. Mech. Eng., 2020, 362, 112790.
166. W.-F. Hu, Y.-J. Shih, T.-S. Lin and M.-C. Lai, Comput. Methods Appl. Mech. Eng., 2024, 418, 116486.
167. D. Wu and A. Lisser, Math. Comput. Simul., 2024, 219, 424–434.
168. J.-P. Correa-Baena, K. Hippalgaonkar, J. van Duren, S. Jaffer, V. R. Chandrasekhar, V. Stevanovic, C. Wadia, S. Guha and T. Buonassisi, Joule, 2018, 2, 1410–1420.
169. S. Hussain, S. Ali, S. Shahzadi, M. N. Tahir and M. Shahid, J. Coord. Chem., 2015, 68, 2369–2387.
170. M. U. Khan, R. Hussain, M. Y. Mehboob, M. Khalid, M. A. Ehsan, A. Rehman and M. R. S. A. Janjua, Spectrochim. Acta, Part A, 2021, 245, 1386–1425.
171. S. Zhu, B. H. Nguyen, Y. Xia, K. Frost, S. Xie, V. Viswanathan and J. A. Smith, Green Chem., 2023, 25, 6612–6617.
172. K. Dąbrowa, P. Niedbała and J. Jurczak, Chem. Commun., 2014, 50, 15748–15751.
173. X. Wei and T. Luo, Phys. Chem. Chem. Phys., 2022, 24, 10272–10279.
174. Y. Wang, S. Wang, J. Shi, Z. Chen, L. Sheng and T. Zhang, J. Phys. Chem. C, 2022, 126, 5443–5456.
175. G. Armeli, J.-H. Peters and T. Koop, ACS Omega, 2023, 8, 12298–12309.
176. D. Bhattacharya and T. K. Patra, Macromolecules, 2021, 54, 3065–3074.
177. J. K. Yu, C. Bannwarth, R. Liang, E. G. Hohenstein and T. J. Martínez, J. Am. Chem. Soc., 2020, 142, 20680–20690.
178. S. Axelrod, E. Shakhnovich and R. Gómez-Bombarelli, Nat. Commun., 2022, 13, 3440.
179. S. Axelrod, E. Shakhnovich and R. Gómez-Bombarelli, ACS Cent. Sci., 2023, 9, 166–176.
180. M. Koerstz, A. Christensen, K. Mikkelsen, M. Nielsen and J. Jensen, PeerJ Phys. Chem., 2021, 3, e16.
181. O. Christensen, R. D. Schlosser, R. B. Nielsen, J. Johansen, M. Koerstz, J. H. Jensen and K. V. Mikkelsen, J. Phys. Chem., 2022, 126, 1681–1688.
182. S. Axelrod, D. Schwalbe-Koda, S. Mohapatra, J. Damewood, K. P. Greenman and R. Gómez-Bombarelli, Acc. Mater. Res., 2022, 3, 343–357.

---