Diffusion is a fundamental biological process with profound implications for human health, traversing diverse areas such as drug delivery, understanding protein-protein interactions in disease pathways, and ensuring the safety and quality of food products. Yet, accurate characterization of various modes of diffusion presents formidable challenges, especially when dealing with short and noisy trajectories observed in single particle tracking (SPT) experiments, which have direct relevance to the study of cellular processes and drug interactions in living organisms. To overcome this challenge and attain insights with potential health applications, this study harnesses the power of artificial intelligence (AI) techniques to enhance the characterization of single particle trajectories. In this study, random forest algorithm has been used to classify different types of motion for single particles. Feature selection algorithms have been employed to identify the most influential features for precise trajectory characterization. Consistently across all algorithms, the seven features were identified indicating their crucial roles in characterizing single particle motion. To assess the impact of these features, a comparative analysis of model accuracy is conducted between the base model utilizing all nine features and model incorporating the six features from each algorithm. The results reveal that both models achieve the equal accuracy of 70

Progress in the field of nanochemistry for nanocrystal fabrication entails a three-stage workflow involving chemical synthesis, optical characterization, and scanning electron microscopy (SEM) analysis. Optical characterization provides qualitative insights into particle properties, while SEM analysis yields quantitative information on particle dimensions such as shape and size. Nevertheless, the use of SEM analysis poses limitations in terms of hindering the speed of mechanistic studies on nanocrystal growth and potential sample degradation due to the electron beam's impact. To address these challenges, a deep learning algorithm, specifically a convoluted neural network (CNN), was developed and trained on experimental spectra of nanoparticles exhibiting diverse morphologies (spherical, rod, prism, hexagonal) and size dimensions ranging from 40 nm to 300 nm. This CNN model enables accurate assessment of nanoparticle shape and size without relying on SEM analysis. Moreover, it mitigates the concerns associated with sample degradation and alteration caused by electron beam exposure. The CNN model, specifically designed for predicting the shape and size of gold nanoparticles (AuNPs) based on their scattering spectrum, achieved an impressive overall accuracy of 92%. This surpasses the performance of the base CNN model (84%) and the transfer learning model (86%). The success of this deep learning approach signifies the potential for replacing electron microscopy with convoluted neural networks in nanoparticle characterization, offering a promising avenue for efficient and reliable analysis.

In this study, we employ equation-of-motion coupled cluster methods in conjunction with recently developed hybrid basis sets, built from off-the-shelf Dunning’s correlation consistent bases, to compute x-ray photoelectron spectra (XPS) for small gas phase molecules and water clusters (H₂O)_n (n = 1-3). Our focus is to assess the efficacy of hybrid (combining basis sets on one atom) and mixed (combining basis sets for different atoms) basis sets in CVS-EOMIP-CCSD and TP-CCSD computations for the core-level states. Our findings will provide insights into the impact of basis sets on CC calculations of core-level states. This research aims to establish a protocol for deriving reliable theoretical estimates and computational protocols for XPS spectra of larger water clusters. This extension aims to isolate cooperative effects and deviations from ideal additive behavior, providing a detailed understanding of spectral features concerning the hydrogen bond network topology.

The calculation of core-hole electronic states, which appear for example in soft x-ray photoelectron (XPS), absorption (NEXAFS/XANES), and inelastic scattering (RIXS) spectroscopies, is especially challenging for traditional excited state theoretical methods such as time-dependent density functional theory (TD-DFT) and equation-of-motion coupled cluster theory (EOM-CC). The main issue is the large orbital relaxation after creation of the core hole. In EOM-CC theory, accurately accounting for this effect normally requires the inclusion of at least triple excitations. Alternatively, we have considered two new methods: first, we developed the transition-potential coupled cluster (TP-CC) method, which draws inspiration from the transition-potential density functional theory (TP-DFT) method in that a partial core-hole state is used to determine the reference orbitals. In contrast, though, TP-CC provides a rigorous wavefunction and transition or excited state properties. Second, we developed the STEOM-CCSD+cT method which extends the similarity-transformed equation-of-motion coupled cluster theory (STEOM-CC) with the inclusion of triple excitations only in the determination of the core ionization potential. This limits the additional computational cost and overall scaling of the method. In this talk, we briefly recap the theoretical features of these methods and present benchmark data for small gas-phase organic molecules illustrating the increase in accuracy of transition energies and intensities over standard EOM-CCSD calculations.

There exist several techniques in the electronic structure community to address the degree of nondynamical correlation for a chosen molecule within a particular correlation method. Popular choices include the T₁ and T₂ diagnostics, whereby the Euclidean norm of the T₁ vector of the coupled-cluster wavefunction or the largest T₂ amplitude in a CCSD calculation are examined, respectively. Monitoring both will detect several problematic cases, however, a certain degree of ambiguity persists in these diagnostics. Here, we present the beginnings of a new diagnostic – specifically the degree of asymmetry in the reduced density matrix – which will vanish in the full CI limit but persists in standard coupled-cluster calculations (due to the fact that the coupled-cluster method is an intrinsically non-hermitian theory). Results are given for several small astrophysically relevant molecules and the potential of using this diagnostic to assess the quality of coupled-cluster calculations is discussed.

The lowest energy anion of BH₃ has previously been studied by photoelectron spectroscopy and matrix isolation infrared absorption spectroscopy. However, a theoretical description of the anion is deceptively challenging. Previous work has found the anion to be unstable to electron detachment in the vicinity of its D_3h minimum energy structure—becoming stable at stretched D_3h geometries. Only after correction for zero-point vibrational energy (ZPVE) does the anion fall energetically below the neutral. In this work, the adiabatic electron affinity and photoelectron spectrum of BH₃ are studied by combining convergent EOMEA-CC methods with Regularized Analytical Continuation (RAC). Particular attention is given to converging the critically-important ZPVE with respect to expansion of the basis set (both in diffuseness and angular momentum).

We have developed a generic method to evaluate vibrational spectra using inputs from semiclassical ab initio molecular dynamics simulations. Extending the work of Heller and others, this approach builds off of a flexible distributed Gaussian basis set that can adapt to the shape of the potential, and includes a tunable phase parameter derived from the momentum of the system to allow for improved locality. Through a series of refinements to the treatment of the potential it is possible to obtain accurate results from relatively small trajectories. Moreover, the results may be converged to any desired accuracy by running more AIMD trajectories. This method provides a natural complement to perturbative approaches, as it does not suffer from issues with low-frequency vibrations or degeneracy handling. By exploiting the locality of molecular vibrations, it is particularly well suited to the evaluation of spectra involving transitions to a subset of the total space of vibrations for systems of moderate size. Making use of machine-learning inspired techniques for extracting potential surfaces with good interpolatory behavior from AIMD trajectories, the method allows for an entirely black-box approach while retaining good accuracy.

Spectroscopic modeling is generally considered a high-cost task, especially compared to other tasks that have been accelerated in recent years with machine learning approaches. In this talk, we will focus on the development of new methods for increasing the efficiency of molecular spectroscopy modeling. Our first method is motivated from a functional materials design perspective, where we integrate reinforcement learning frameworks with generative molecular design methods to produce molecules with desired functional properties-in this case, materials that meet some necessary requirements for singlet fission. Here we will demonstrate the generalizability of this method to other spectroscopic tasks, and show how to balance exploration and optimization. In our second approach, we discuss the more difficult problem of spectroscopic assignment. In this part, we will demonstrate the utility of several competing approaches, including reinforcement learning and optimization, highlighting the importance of the representation that is passed to both of these frameworks. Our discussion will focus on model systems, including idealized vibrionic and vibrational transitions.

In recent years, exciton models have been developed to calculate the electronic spectroscopy of pigments within the LH2 complex and many organic optoelectronic materials. These models often assume a form of the Hamiltonian where the coupling between the chromophores is estimated from a dipole-dipole coupling model. While the models often predict spectroscopy and other observables near quantum chemical accuracy for systems where the chromophores are relatively far apart, there is no general solution to predicting the electronic spectroscopy for sub-units closer together. In this talk, we highlight our current work in the construction of a machine-learning model to calculate corrections to electrostatic dipole-dipole couplings and sub-unit site energies for our proposed model Hamiltonian framework. We then show the utility of this framework on a set of oligomeric systems, which cannot be directly simulated with standard TD-DFT approaches.

In line profile analyses of observed spectrum using an equally weighted least squares method, it is statistically correct to use transmittance spectrum (TS) / absorbance spectrum (AS), a negative logarithm of the TS, when the detector noise (DN) / the source noise (SN) is dominant. To evaluate quantitative differences between the correct analyses, TS with DN and AS with SN, and the incorrect analyses, TS with SN and AS with DN, we calculate TS of a Lorentz profile with three profile parameters of absorption strength, center frequency, and width and simulate the observed spectrum by adding a certain magnitude of DN or SN to the calculated TS. The simulated TS and AS with DN or SN are fitted to the Lorentz profile using the least squares methods. The correct analyses reproduce the noise magnitude and the profile parameters and properly give expected uncertainties of the parameters regardless of the absorption strength and the noise magnitude. The incorrect analyses reproduce the profile parameters but not the noise magnitude and do not give the correct expected uncertainties of the parameters for the large absorption strength. Properly weighted least squares fits of TS with SN and AS with DN correctly provide the noise magnitude and the expected uncertainties of the parameters but the determined absorption strengths and widths are considerably biased in some cases. We also show how to determine the noise magnitudes of DN and SN from the observed spectrum. They are required for correct selection of either TS or AS and thereby must be reported in a paper on experimental spectroscopy.

The electronic structures of Sn₂H₂ and its cation are investigated in this work. Unlike acetylene, the linear structure of Sn₂H₂ is not the global energy minima, so different configurations of the molecule (the butterfly, monobridged, vinylidene-like, trans, cis, linear, and planar dibridged structures) are deeply investigated. The interest in this molecule and its isomers is motivated by the importance of understanding tin-hydrogen interactions and the peculiar electronic structures of similar group 14 molecules, such as Si₂H₂ and Ge₂H₂. The geometries, harmonic vibrational frequencies, and energies of the neutral and cationic structures were determined using the CCSD(T) level of theory and cc-pwCVXZ-PP (X = D, T, Q) basis sets with a small-core pseudopotential. Our results show that butterfly, monobridged, vinylidene-like, and nonplanar trans isomers are energy minima, while the planar trans, cis, linear, and planar dibridged isomers are transition states. Among the minima, the butterfly isomer lies lowest in energy, followed by the monobridged, vinylidene-like, and nonplanar trans isomers respectively. NBO analysis was also performed to observe changes in bond order between the neutral and cation structures.