Deep learning of dynamically responsive chemical Hamiltonians with semiempirical quantum mechanics.

Conventional machine-learning (ML) models in computational chemistry learn to directly predict molecular properties using quantum chemistry only for reference data. While these heuristic ML methods show quantum-level accuracy with speeds several orders of magnitude faster than traditional quantum chemistry methods, they suffer from poor extensibility and transferability; i.e., their accuracy degrades on large or new chemical systems. Incorporating quantum chemistry frameworks into the ML models directly solves this problem. Here we take the structure of semiempirical quantum mechanics (SEQM) methods to construct dynamically responsive Hamiltonians. SEQM methods use empirical parameters fitted to experimental properties to construct reduced-order Hamiltonians, facilitating much faster calculations than ab initio methods but with compromised accuracy. By replacing these static parameters with machine-learned dynamic values inferred from the local environment, we greatly improve the accuracy of the SEQM methods. Trained on molecular energies and atomic forces, these dynamically generated Hamiltonian parameters show a strong correlation with atomic hybridization and bonding. Trained with only about 60,000 small organic molecular conformers, the resulting model retains interpretability, extensibility, and transferability when testing on much larger chemical systems and predicting various molecular properties. Overall, this work demonstrates the virtues of incorporating physics-based descriptions with ML to develop models that are simultaneously accurate, transferable, and interpretable.

Dielectric Screening and Charge-Transfer in 2D Lead-Halide Perovskites for Reduced Exciton Binding Energies.

Layered lead-halide perovskites have shown tremendous success as an active material for optoelectronics. This is attributed to the electronic structure of the inorganic sublattice and large exciton binding energies due to quantum and dielectric confinement. Expanding functionalities for applications that depend on free-carrier generation requires new material design routes to decrease the binding energy. Here we use electronic structure methods with model Bethe-Salpeter equation (BSE) to examine the contributions of the dielectric screening and charge-transfer excited-states to the exciton binding energy of phenylethylammonium (PEA2PbBr4) and naphthlethylammonium (NEA2PbBr4) lead-bromide perovskites. Our model BSE calculations show that NEA introduces hole acceptor states which impose charge-transfer character on the exciton along with larger dielectric screening. This substantially decreases the exciton binding compared to PEA. This result suggests the use of organic cations with high dielectric screening and hole acceptor states as a viable strategy for reducing exciton binding energies in two-dimensional halide perovskites.

Light-Induced Structural Dynamics and Charge Transport in Layered Halide Perovskite Thin Films.

The dynamic nature of the metal halide perovskite lattice upon photoexcitation plays a vital role in their properties. Here we report an observation of light-induced structure dynamics in quasi-2D Ruddlesden-Popper phase perovskite thin films and its impact on the carrier transport properties. By a time-resolved X-ray scattering technique, we observe a rapid lattice expansion upon photoexcitation, followed by a slow relaxation over the course of 100 ns in the dark. Theoretical modeling suggests that the expansion originates from the lattice's thermal fluctuations caused by photon energy deposition. Power dependent optical spectroscopy and photoconductivity indicate that high laser powers triggered a strong local structural disorder, which increased the charge dissociation activation energy that results in localized transport. Our study investigates the impact of laser energy deposition on the lattices and the subsequent carrier transport properties, that are relevant to device operations.

X-ray and Optical Circular Dichroism as Local and Global Ultrafast Chiral Probes of [12]Helicene Racemization.

Chirality is a fundamental molecular property that plays a crucial role in biophysics and drug design. Optical circular dichroism (OCD) is a well-established chiral spectroscopic probe in the UV-visible regime. Chirality is most commonly associated with a localized chiral center. However, some compounds such as helicenes (Figure 1) are chiral due to their screwlike global structure. In these highly conjugated systems, some electric and magnetic allowed transitions are distributed across the entire molecule, and OCD thus probes the global molecular chirality. Recent advances in X-ray sources, in particular the control of their polarization and spatial profiles, have enabled X-ray circular dichroism (XCD), which, in contrast to OCD, can exploit the localized and element-specific nature of X-ray electronic transitions. XCD therefore is more sensitive to local structures, and the chirality probed with it can be referred to as local. During the racemization of helicene, between opposite helical structures, the screw handedness can flip locally, making the molecule globally achiral while retaining a local handedness. Here, we use the racemization mechanism of [12]helicene as a model to demonstrate the capabilities of OCD and XCD as time-dependent probes for global and local chiralities, respectively. Our simulations demonstrate that XCD provides an excellent spectroscopic probe for the time-dependent local chirality of molecules.

Neural network atomistic potentials for global energy minima search in carbon clusters.

The global energy optimization problem is an acute and important problem in chemistry. It is crucial to know the geometry of the lowest energy isomer (global minimum, GM) of a given compound for the evaluation of its chemical and physical properties. This problem is especially relevant for atomic clusters. Due to the exponential growth of the number of local minima geometries with the increase of the number of atoms in the cluster, it is important to find a computationally efficient and reliable method to navigate the energy landscape and locate a true global minima structure. Newly developed neural network (NN) atomistic potentials offer a numerically efficient and relatively accurate approach for molecular structure optimization. An important question that needs to be answered is "Can NN potentials, trained on a given set, represent the potential energy surface (PES) of a neighboring domain?". In this work, we tested the applicability of ANI-1ccx and ANI-nr NN atomistic potentials for the global minima optimization of carbon clusters Cn (n = 3-10). We showed that with the introduction of the cluster connectivity restriction and consequent DFT or ab initio calculations, ANI-1ccx and ANI-nr can be considered as robust PES pre-samplers that can capture the GM structure even for large clusters such as C20.

Machine Learning Models Capture Plasmon Dynamics in Ag Nanoparticles.

Highly energetic electron-hole pairs (hot carriers) formed from plasmon decay in metallic nanostructures promise sustainable pathways for energy-harvesting devices. However, efficient collection before thermalization remains an obstacle for realization of their full energy generating potential. Addressing this challenge requires detailed understanding of physical processes from plasmon excitation in the metal to their collection in a molecule or a semiconductor, where atomistic theoretical investigation may be particularly beneficial. Unfortunately, first-principles theoretical modeling of these processes is extremely costly, preventing a detailed analysis over a large number of potential nanostructures and limiting the analysis to systems with a few 100s of atoms. Recent advances in machine learned interatomic potentials suggest that dynamics can be accelerated with surrogate models which replace the full solution of the Schrödinger Equation. Here, we modify an existing neural network, Hierarchically Interacting Particle Neural Network (HIP-NN), to predict plasmon dynamics in Ag nanoparticles. The model takes as a minimum as three time steps of the reference real-time time-dependent density functional theory (rt-TDDFT) calculated charges as history and predicts trajectories for 5 fs in great agreement with the reference simulation. Further, we show that a multistep training approach in which the loss function includes errors from future time-step predictions can stabilize the model predictions for the entire simulated trajectory (∼25 fs). This extends the model's capability to accurately predict plasmon dynamics in large nanoparticles of up to 561 atoms, not present in the training data set. More importantly, with machine learning models on GPUs, we gain a speed-up factor of ∼103 as compared with the rt-TDDFT calculations when predicting important physical quantities such as dynamic dipole moments in Ag55 and a factor of ∼104 for extended nanoparticles that are 10 times larger. This underscores the promise of future machine learning accelerated electron/nuclear dynamics simulations for understanding fundamental properties of plasmon-driven hot carrier devices.

Synergy of semiempirical models and machine learning in computational chemistry.

Catalyzed by enormous success in the industrial sector, many research programs have been exploring data-driven, machine learning approaches. Performance can be poor when the model is extrapolated to new regions of chemical space, e.g., new bonding types, new many-body interactions. Another important limitation is the spatial locality assumption in model architecture, and this limitation cannot be overcome with larger or more diverse datasets. The outlined challenges are primarily associated with the lack of electronic structure information in surrogate models such as interatomic potentials. Given the fast development of machine learning and computational chemistry methods, we expect some limitations of surrogate models to be addressed in the near future; nevertheless spatial locality assumption will likely remain a limiting factor for their transferability. Here, we suggest focusing on an equally important effort-design of physics-informed models that leverage the domain knowledge and employ machine learning only as a corrective tool. In the context of material science, we will focus on semi-empirical quantum mechanics, using machine learning to predict corrections to the reduced-order Hamiltonian model parameters. The resulting models are broadly applicable, retain the speed of semiempirical chemistry, and frequently achieve accuracy on par with much more expensive ab initio calculations. These early results indicate that future work, in which machine learning and quantum chemistry methods are developed jointly, may provide the best of all worlds for chemistry applications that demand both high accuracy and high numerical efficiency.

Two Dimensional MOene: From Superconductors to Direct Semiconductors and Weyl Fermions.

The number of semiconducting MXenes with direct band gaps is extremely low; thus, it is highly desirable to broaden the MXene family beyond carbides and nitrides to expand the palette of desired chemical and physical properties. Here, we theoretically report the existence of the single-layer (SL) dititanium oxide 2H-Ti2O MOene (MXene-like 2D transition oxides), showing an Ising superconducting feature. Moreover, SL halogenated 2H- and 1T-Ti2O monolayers display tunable semiconducting features and strong light-harvesting ability. In addition, the external strains can induce Weyl fermions via quantum phase transition in 2H-Ti2OF2 and Ti2OCl2 monolayers. Specifically, 2H- and 1T-Ti2OF2 are direct semiconductors with band gaps of 0.82 and 1.18 eV, respectively. Furthermore, the carrier lifetimes of SL 2H- and 1T-Ti2OF2 are evaluated to be 0.39 and 2.8 ns, respectively. This study extends emerging phenomena in a rich family of 2D MXene-like MOene materials, which provides a novel platform for next-generation optoelectronic and photovoltaic fields.

Charge Delocalization and Vibronic Couplings in Quadrupolar Squaraine Dyes.

Squaraines are prototypical quadrupolar charge-transfer chromophores that have recently attracted much attention as building blocks for solution-processed photovoltaics, fluorescent probes with large two-photon absorption cross sections, and aggregates with large circular dichroism. Their optical properties are often rationalized in terms of phenomenological essential state models, considering the coupling of two zwitterionic excited states to a neutral ground state. As a result, optical transitions to the lowest S1 excited state are one-photon allowed, whereas the next higher S2 state can only be accessed by two-photon transitions. A further implication of these models is a substantial reduction of vibronic coupling to the ubiquitous high-frequency vinyl-stretching modes of organic materials. Here, we combine time-resolved vibrational spectroscopy, two-dimensional electronic spectroscopy, and quantum-chemical simulations to test and rationalize these predictions for nonaggregated molecules. We find small Huang-Rhys factors below 0.01 for the high-frequency, 1500 cm-1 modes in particular, as well as a noticeable reduction for those of lower frequency modes in general for the electronic S0 → S1 transition. The two-photon allowed state S2 is well separated energetically from S1 and has weak vibronic signatures as well. Thus, the resulting pronounced concentration of the oscillator strength in a narrow region relevant to the lowest electronic transition makes squaraines and their aggregates exceptionally interesting for strong and ultrastrong coupling of excitons to localized light modes in external resonators with chiral properties that can largely be controlled by the molecular architecture.

Non-adiabatic Excited-State Molecular Dynamics: Theory and Applications for Modeling Photophysics in Extended Molecular Materials.

Optically active molecular materials, such as organic conjugated polymers and biological systems, are characterized by strong coupling between electronic and vibrational degrees of freedom. Typically, simulations must go beyond the Born-Oppenheimer approximation to account for non-adiabatic coupling between excited states. Indeed, non-adiabatic dynamics is commonly associated with exciton dynamics and photophysics involving charge and energy transfer, as well as exciton dissociation and charge recombination. Understanding the photoinduced dynamics in such materials is vital to providing an accurate description of exciton formation, evolution, and decay. This interdisciplinary field has matured significantly over the past decades. Formulation of new theoretical frameworks, development of more efficient and accurate computational algorithms, and evolution of high-performance computer hardware has extended these simulations to very large molecular systems with hundreds of atoms, including numerous studies of organic semiconductors and biomolecules. In this Review, we will describe recent theoretical advances including treatment of electronic decoherence in surface-hopping methods, the role of solvent effects, trivial unavoided crossings, analysis of data based on transition densities, and efficient computational implementations of these numerical methods. We also emphasize newly developed semiclassical approaches, based on the Gaussian approximation, which retain phase and width information to account for significant decoherence and interference effects while maintaining the high efficiency of surface-hopping approaches. The above developments have been employed to successfully describe photophysics in a variety of molecular materials.

Exciton-vibrational dynamics induces efficient self-trapping in a substituted nanoring.

Cycloparaphenylenes, being the smallest segments of carbon nanotubes, have emerged as prototypes of the simplest carbon nanohoops. Their unique structure-dynamics-optical properties relationships have motivated a wide variety of synthesis of new related nanohoop species. Studies of how chemical changes, introduced in these new materials, lead to systems with new structural, dynamics and optical properties, expand their functionalities for optoelectronics applications. Herein, we study the effect that conjugation extension of a cycloparaphenylene through the introduction of a satellite tetraphenyl substitution has on its structural and dynamical properties. Our non-adiabatic excited state molecular dynamics simulations suggest that this substitution accelerates the electronic relaxation from the high-energy band to the lowest excited state. This is partially due to efficient conjugation achieved between specific phenyl units as introduced by the tetraphenyl substitution. We observe a particular exciton redistribution during relaxation, in which the tetraphenyl substitution plays a significant role. As a result, an efficient inter-band energy transfer takes place. Besides, the observed phonon-exciton interplay induces a significant exciton self-trapping. Our results encourage and guide the future studies of new phenyl substitutions in carbon nanorings with desired optoelectronic properties.

Vibronic Photoexcitation Dynamics of Perylene Diimide: Computational Insights.

Perylene diimide (PDI) represents a prototype material for organic optoelectronic devices because of its strong optical absorbance, chemical stability, efficient energy transfer, and optical and chemical tunability. Herein, we analyze in detail the vibronic relaxation of its photoexcitation using nonadiabatic excited-state molecular dynamics simulations. We find that after the absorption of a photon, which excites the electron to the second excited state, S2, induced vibronic dynamics features persistent modulations in the spatial localization of electronic and vibrational excitations. These energy exchanges are dictated by strong vibronic couplings that overcome structural disorders and thermal fluctuations. Specifically, the electronic wavefunction periodically swaps between localizations on the right and left sides of the molecule. Within 1 ps of such dynamics, a nonradiative transition to the lowest electronic state, S1, takes place, resulting in a complete delocalization of the wavefunction. The observed vibronic dynamics emerges following the electronic energy deposition in the direction that excites a combination of two dominant vibrational normal modes. This behavior is maintained even with a chemical substitution that breaks the symmetry of the molecule. We believe that our findings elucidate the nature of the complex dynamics of the optically excited states and, therefore, contribute to the development of tunable functionalities of PDIs and their derivatives.

High-efficiency two-dimensional Ruddlesden-Popper perovskite solar cells.

Three-dimensional organic-inorganic perovskites have emerged as one of the most promising thin-film solar cell materials owing to their remarkable photophysical properties, which have led to power conversion efficiencies exceeding 20 per cent, with the prospect of further improvements towards the Shockley-Queisser limit for a single­junction solar cell (33.5 per cent). Besides efficiency, another critical factor for photovoltaics and other optoelectronic applications is environmental stability and photostability under operating conditions. In contrast to their three-dimensional counterparts, Ruddlesden-Popper phases--layered two-dimensional perovskite films--have shown promising stability, but poor efficiency at only 4.73 per cent. This relatively poor efficiency is attributed to the inhibition of out-of-plane charge transport by the organic cations, which act like insulating spacing layers between the conducting inorganic slabs. Here we overcome this issue in layered perovskites by producing thin films of near-single-crystalline quality, in which the crystallographic planes of the inorganic perovskite component have a strongly preferential out-of-plane alignment with respect to the contacts in planar solar cells to facilitate efficient charge transport. We report a photovoltaic efficiency of 12.52 per cent with no hysteresis, and the devices exhibit greatly improved stability in comparison to their three-dimensional counterparts when subjected to light, humidity and heat stress tests. Unencapsulated two-dimensional perovskite devices retain over 60 per cent of their efficiency for over 2,250 hours under constant, standard (AM1.5G) illumination, and exhibit greater tolerance to 65 per cent relative humidity than do three-dimensional equivalents. When the devices are encapsulated, the layered devices do not show any degradation under constant AM1.5G illumination or humidity. We anticipate that these results will lead to the growth of single-crystalline, solution-processed, layered, hybrid, perovskite thin films, which are essential for high-performance opto-electronic devices with technologically relevant long-term stability.

Investigation of plasmon relaxation mechanisms using nonadiabatic molecular dynamics.

Hot carriers generated from the decay of plasmon excitation can be harvested to drive a wide range of physical or chemical processes. However, their generation efficiency is limited by the concomitant phonon-induced relaxation processes by which the energy in excited carriers is transformed into heat. However, simulations of dynamics of nanoscale clusters are challenging due to the computational complexity involved. Here, we adopt our newly developed Trajectory Surface Hopping (TSH) nonadiabatic molecular dynamics algorithm to simulate plasmon relaxation in Au20 clusters, taking the atomistic details into account. The electronic properties are treated within the Linear Response Time-Dependent Tight-binding Density Functional Theory (LR-TDDFTB) framework. The relaxation of plasmon due to coupling to phonon modes in Au20 beyond the Born-Oppenheimer approximation is described by the TSH algorithm. The numerically efficient LR-TDDFTB method allows us to address a dense manifold of excited states to ensure the inclusion of plasmon excitation. Starting from the photoexcited plasmon states in Au20 cluster, we find that the time constant for relaxation from plasmon excited states to the lowest excited states is about 2.7 ps, mainly resulting from a stepwise decay process caused by low-frequency phonons of the Au20 cluster. Furthermore, our simulations show that the lifetime of the phonon-induced plasmon dephasing process is ∼10.4 fs and that such a swift process can be attributed to the strong nonadiabatic effect in small clusters. Our simulations demonstrate a detailed description of the dynamic processes in nanoclusters, including plasmon excitation, hot carrier generation from plasmon excitation dephasing, and the subsequent phonon-induced relaxation process.

Nonadiabatic molecular dynamics simulations based on time-dependent density functional tight-binding method.

Nonadiabatic excited state molecular dynamics underpin many photophysical and photochemical phenomena, such as exciton dynamics, and charge separation and transport. In this work, we present an efficient nonadiabatic molecular dynamics (NAMD) simulation method based on time-dependent density functional tight-binding (TDDFTB) theory. Specifically, the adiabatic electronic structure, an essential NAMD input, is described at the TDDFTB level. The nonadiabatic effects originating from the coupled motions of electrons and nuclei are treated by the trajectory surface hopping algorithm. To improve the computational efficiency, nonadiabatic couplings between excited states within the TDDFTB method are derived and implemented using an analytical approach. Furthermore, the time-dependent nonadiabatic coupling scalars are calculated based on the overlap between molecular orbitals rather than the Slater determinants to speed up the simulations. In addition, the electronic decoherence scheme and a state reassigned unavoided crossings algorithm, which has been implemented in the NEXMD software, are used to improve the accuracy of the simulated dynamics and handle trivial unavoided crossings. Finally, the photoinduced nonadiabatic dynamics of a benzene molecule are simulated to demonstrate our implementation. The results for excited state NAMD simulations of benzene molecule based on TDDFTB method compare well to those obtained with numerically expensive time-dependent density functional theory. The proposed methodology provides an attractive theoretical simulation tool for predicting the photophysical and photochemical properties of complex materials.

Controlling Defect-State Photophysics in Covalently Functionalized Single-Walled Carbon Nanotubes.

ConspectusSingle-walled carbon nanotubes (SWCNTs) show promise as light sources for modern fiber optical communications due to their emission wavelengths tunable via chirality and diameter dependency. However, the emission quantum yields are relatively low owing to the existence of low-lying dark electronic states and fast excitonic diffusion leading to carrier quenching at defects. Covalent functionalization of SWCNTs addresses this problem by brightening their infrared emission. Namely, introduction of sp3-hybridized defects makes the lowest energy transitions optically active for some defect geometries and enables further control of their optical properties. Such functionalized SWCNTs are currently the only material exhibiting room-temperature single photon emission at telecom relevant infrared wavelengths. While this fluorescence is strong and has the right wavelength, functionalization introduces a variety of emission peaks resulting in spectrally broad inhomogeneous photoluminescence that prohibits the use of SWCNTs in practical applications. Consequently, there is a strong need to control the emission diversity in order to render these materials useful for applications. Recent experimental and computational work has attributed the emissive diversity to the presence of multiple localized defect geometries each resulting in distinct emission energy. This Account outlines methods by which the morphology of the defect in functionalized SWCNTs can be controlled to reduce emissive diversity and to tune the fluorescence wavelengths. The chirality-dependent trends of emission energies with respect to individual defect morphologies are explored. It is demonstrated that defect geometries originating from functionalization of SWCNT carbon atoms along bonds with strong π-orbital mismatch are favorable. Furthermore, the effect of controlling the defect itself through use of different chemical groups is also discussed. Such tunability is enabled due to the formation of specific defect geometries in close proximity to other existing defects. This takes advantage of the changes in π-orbital mismatch enforced by existing defects and the resulting changes in reactivities toward formation of specific defect morphologies. Furthermore, the trends in emissive energies are highly dependent on the value of mod(n-m,3) for an (n,m) tube chirality. These powerful concepts allow for a targeted formation of defects that emit at desired energies based on SWCNT single chirality enriched samples. Finally, the impact of functionalization with specific types of defects that enforce certain defect geometries due to steric constraints in bond lengths and angles to the SWCNT are discussed. We further relate to a similar effect that is present in systems where high density of surface defects is formed due to high reactant concentration. The outlined strategies suggested by simulations are instrumental in guiding experimental efforts toward the generation of functionalized SWCNTs with tunable emission energies.

Excitation Energy Transfer between bodipy Dyes in a Symmetric Molecular Excitonic Seesaw.

We examine the redistribution of energy between electronic and vibrational degrees of freedom that takes place between a π-conjugated oligomer, a phenylene-butadiynylene, and two identical boron-dipyrromethene (bodipy) end-caps using femtosecond transient absorption spectroscopy, single-molecule spectroscopy, and nonadiabatic excited-state molecular dynamics (NEXMD) modeling techniques. The molecular structure represents an excitonic seesaw in that the excitation energy on the oligomer backbone can migrate to either one end-cap or the other, but not to both. The NEXMD simulations closely reproduce the characteristic time scale for redistribution of electronic and vibrational energy of 2.2 ps and uncover the vibrational modes contributing to the intramolecular relaxation. The calculations indicate that the dihedral angle between the bodipy dye and the oligomer change upon excitation of the oligomer. Single-molecule experiments reveal a difference in photoluminescence lifetime of the bodipy dyes depending on whether they are excited by direct absorption or by redistribution of energy from the backbone. This difference in lifetime may be attributed to the difference in dihedral angle. The simulations also suggest that a strong coupling can occur between the two end-caps, giving rise to a reversible shuttling of excitation energy between them. Strong coupling should lead to a pronounced loss in polarization memory of the fluorescence since the oligomer backbone tends to be slightly distorted and the two bodipy transition dipoles have different orientations. A sensitive single-molecule technique is presented to test for such coupling. However, although redistribution of electronic and vibrational energy between the end-caps can occur, it appears to be unidirectional and irreversible, suggesting that an additional localization mechanism is at play which is, as yet, not fully accounted for in the simulations.

Machine learned Hückel theory: Interfacing physics and deep neural networks.

The Hückel Hamiltonian is an incredibly simple tight-binding model known for its ability to capture qualitative physics phenomena arising from electron interactions in molecules and materials. Part of its simplicity arises from using only two types of empirically fit physics-motivated parameters: the first describes the orbital energies on each atom and the second describes electronic interactions and bonding between atoms. By replacing these empirical parameters with machine-learned dynamic values, we vastly increase the accuracy of the extended Hückel model. The dynamic values are generated with a deep neural network, which is trained to reproduce orbital energies and densities derived from density functional theory. The resulting model retains interpretability, while the deep neural network parameterization is smooth and accurate and reproduces insightful features of the original empirical parameterization. Overall, this work shows the promise of utilizing machine learning to formulate simple, accurate, and dynamically parameterized physics models.

Passivating Nucleobases Bring Charge Transfer Character to Optically Active Transitions in Small Silver Nanoclusters.

DNA-wrapped silver nanoclusters (DNA-AgNCs) are known for their efficient luminescence. However, their emission is highly sensitive to the DNA sequence, the cluster size, and its charge state. To get better insights into photophysics of these hybrid systems, simulations based on density functional theory (DFT) are performed. Our calculations elucidate the effect of the structural conformations, charges, solvent polarity, and passivating bases on optical spectra of DNA-AgNCs containing five and six Ag atoms. It is found that inclusion of water in calculations as a polar solvent media results in stabilization of nonplanar conformations of base-passivated clusters, while their planar conformations are more stable in vacuum, similar to the bare Ag5 and Ag6 clusters. Cytosines and guanines interact with the cluster twice stronger than thymines, due to their larger dipole moments. In addition to the base-cluster interactions, hydrogen bonds between bases notably contribute to the structure stabilization. While the relative intensity, line width, and the energy of absorption peaks are slightly changing depending on the cluster charge, conformations, and base types, the overall spectral shape with five well-resolved bands at 2.5-5.5 eV is consistent for all structures. Independent of the passivating bases and the cluster size and charge, the low energy optical transitions at 2.5-3.5 eV exhibit a metal to ligand charge transfer (MLCT) character with the main contribution emerging from Ag-core to the bases. Cytosines facilitate the MLCT character to a larger degree comparing to the other bases. However, the doublet transitions in clusters with the open shell electronic structure (Ag5 and Ag6+) result in appearance of additional red-shifted (<2.5 eV) and optically weak band with negligible MLCT character. The passivated clusters with the closed shell electronic structure (Ag5+ and Ag6) exhibit higher optical intensity of their lowest transitions with much higher MLCT contribution, thus having better potential for emission, than their open shell counterparts.

Machine learning approaches for structural and thermodynamic properties of a Lennard-Jones fluid.

Predicting the functional properties of many molecular systems relies on understanding how atomistic interactions give rise to macroscale observables. However, current attempts to develop predictive models for the structural and thermodynamic properties of condensed-phase systems often rely on extensive parameter fitting to empirically selected functional forms whose effectiveness is limited to a narrow range of physical conditions. In this article, we illustrate how these traditional fitting paradigms can be superseded using machine learning. Specifically, we use the results of molecular dynamics simulations to train machine learning protocols that are able to produce the radial distribution function, pressure, and internal energy of a Lennard-Jones fluid with increased accuracy in comparison to previous theoretical methods. The radial distribution function is determined using a variant of the segmented linear regression with the multivariate function decomposition approach developed by Craven et al. [J. Phys. Chem. Lett. 11, 4372 (2020)]. The pressure and internal energy are determined using expressions containing the learned radial distribution function and also a kernel ridge regression process that is trained directly on thermodynamic properties measured in simulation. The presented results suggest that the structural and thermodynamic properties of fluids may be determined more accurately through machine learning than through human-guided functional forms.