An on-the-fly deep neural network for simulating time-resolved spectroscopy: predicting the ultrafast ring opening dynamics of 1,2-dithiane

Abstract

Revolutionary developments in ultrafast light source technology are enabling experimental spectroscopists to probe the structural dynamics of molecules and materials on the femtosecond timescale. The capacity to investigate ultrafast processes...

Artificial-Intelligence-Enhanced On-the-Fly Simulation of Nonlinear Time-Resolved Spectra

“…As is becoming increasingly common in such situations, machine learning (ML) tools developed for artificial intelligence (AI) research can be applied to overcome the cost limitations . Successful examples in the related fields include creating surrogate ML models for excited-state properties, , which can be used to increase the precision of linear absorption spectra , and uncertainty quantification, predict two-photon absorption cross sections, and perform non-adiabatic molecular dynamics of molecular systems. − Beyond linear spectra, ML methods were also successfully used in the interpretation of nonlinear spectroscopic signals, i.e., for the reconstruction of certain facets of system dynamics from experimental transient absorption pump–probe , and 2D electronic spectra − as well as “denoising” experimental signals . A few studies invoked ML for the evaluation of nonlinear signals, i.e., predicting 2D electronic spectra of proteins , and extracting relevant parameters, such as orientations of transition dipole moments, from these spectra .…”

Uncertainty quantification of spectral predictions using deep neural networks

“…Recently progress in X-ray science driven by the emergence of facilities capable of delivering high-brilliance ultrashort pulses of X-rays 11 has been the driving force for significant progress in ML models capable of predicting X-ray spectroscopic observables from an input property or structure. 12,13 Very recently, works extending these models to quantify uncertainty have also been demonstrated for X-ray emission 14 and absorption at the C, N and O K-edges. 15 Consequently, in this article, we apply the deep ensembles and bootstrap resampling methods to our DNN used to predict the lineshape of first-row transition metal K-edge XANES spectra 16,17 and demonstrate its ability to assess pointwise uncertainty using only information about the local coordination geometry of the transition metal complexes.…”

Machine-learning strategies for the accurate and efficient analysis of x-ray spectroscopy

“…An alternative approach, the so-called 'Type II' method, is to tailor one's model to a more specialised problem. These models are trained using data for a specific class of systems [162][163][164][165]. Indeed, this is the concept behind the originally developed FitIt [154] approach.…”

“…from a molecular dynamics (MD) simulation] to describe adequately the x-ray spectrum. This is a time-and resource-intensive task for first-principles simulations but can be addressed using ML algorithms [130,145,146,156,162,174,[254][255][256][257] that alleviate the bottleneck associated with computing the x-ray spectra for all of the sampled configurations.…”