Session L01: Material Science and Computational Physics

Revealing Microdynamics that Underlie Tire Tread Performance

Filled elastomers are used in many applications, including tire tread. Functionalized chain additives and coupling agents have been used to enhance interactions between filler and polymer matrix in silica-filled styrene butadiene rubber (SBR). These interactions create bound polymer layers around the filler that alter the dynamic performance by changing aggregate size and connections among particles, forming strong “force chains”.

We have synthesized well-defined functionalized polymeric chain additives and nonfunctional analogs to probe how changing the filler/polymer interface chemistry changes the filler microscopic response and thus the macrodynamics. This study spans length scales from that of the networked agglomerates to that of filler/polymer interfaces. We probe the filler network’s microscopic response using X-ray Photon Correlation Spectroscopy (XPCS) while the sample is under oscillatory strain. The rate of change in the intensity-intensity time autocorrelation during the measurement provides information on the filler network microdynamics. The microscopic responses of samples containing functionalized chain additives differ from those of samples containing nonfunctional additives.   

Structural and Optoelectronic Characterization of AgSbI4 through Machine Learning and Density Functional Theory

Lead-free metal halides, as emerging materials for photovoltaic and optoelectronic applications, have garnered substantial attention, especially given the increasing global emphasis on environmental sustainability. AgSbI₄ is an example of such a material that has recently been synthesized and shows intriguing characteristics but has not yet been adequately explored with density functional theory due to its pronounced site-disorder in its cation sublattice. We harness the potential of a kernel ridge regression machine learning model of the total energy in AgSbI₄ to choose a few simulation cells out of ~10⁷ possibilities; a task out of reach for current first principles techniques. With these models, we calculate structural and optoelectronic properties ranging from X-ray diffraction patterns to absorption and reflection spectra. We compare these results with existing experimental data, for example, the average band gap of 1.96 eV, and lattice constants (a, c = 4.4, 21.0 Å). We calculate effective masses (〈m^*_e〉 = 0.4 m₀ and 〈m^*_h〉 = 4.1 m₀), bulk modulus (35 GPa), and formation energies with respect to AgI and SbI₃ (~50 meV per formula unit). We analyze the density of states, use LOBSTER to calculate Crystal Orbital Hamilton Populations, and perform Bader charge analysis to compute charge transfer. Significantly, our predictions concerning optoelectronic properties indicate AgSbI₄'s potential as an absorber layer in next-generation tandem photovoltaic cells.   

Trions in Monolayer MoS2 using Faddeev Scheme in Momentum Space

In this presentation, we study the binding energy and geometrical structure of negative trions, the bound state of two electrons and one hole interacting via the Rytova-Keldysh interaction in Monolayer MoS2. We introduce a general framework developed in momentum space for the bound state of three different charged particles, interacting with distinct interactions in two dimensions. We discuss the numerical challenges and their solutions for handling repulsive electron-electron interaction when solving the coupled Faddeev integral equations and calculating the trion's binding energy and wave function.   

Temperature Replica Exchange Gaussian Accelerated Molecular Dynamics: Improved Enhanced Sampling and Energetic Reweighting

Gaussian accelerated molecular dynamics (GaMD) provides enhanced sampling and energy reweighting of biomolecules. GaMD works through adding a harmonic boost potential, defined by a force constant and a threshold energy, to smooth the potential energy surface and accelerate sampling between different states of a biomolecular system separated by large energy barriers. Previously, GaMD has been combined with replica exchange algorithms to improve the acceleration power and energy reweighting of the conventional GaMD simulation. Replica exchange GaMD (Rex-GaMD) comes in two varieties: force constant Rex-GaMD and threshold energy Rex-GaMD, which exchange between replicas of different harmonic force constants and threshold energies, respectively. Recently, Rex-GaMD has been combined with the parallel tempering algorithm that exchanges replicas which vary over a range of temperatures. This new method of temperature Rex-GaMD (T-Rex-GaMD) is able to exchange any combination of replicas defined over a range of different values of the force constant, energy threshold, and temperature. Our hope is that the use of high temperature replicas, in addition to force constant and energy threshold replicas, will lead to accelerated sampling of interesting biomolecular systems with multiple different conformational states separated by large energy barriers. For this new method to be useful we must still be able to perform accurate energetic reweighting, allowing us to recover the true free energy profile of our biomolecular system. To this date, we have performed T-Rex-GaMD simulations on three test systems: alanine dipeptide, chignolin, and HIV protease.   

Invariance in Deep Network Learning: Mathematical Representation, Probabilistic Symmetry, Variable Exchangeability, and Sufficient Statistics

Algebraic Lie groups provide the foundation for describing many physical symmetries, which play a key role in studying the geometry of smooth manifolds and analyzing complex high-dimensional observations. In particular, Noether’s symmetries articulate the correspondence between physical symmetries and conservation quantities, which often correspond to physical laws that can be prescribed as differential equations. This interplay between mechanical dynamics, symmetries, information, and geometry characterizes the importance of Lie group actions on modeling classical physical systems, obtaining rigorous statistical inference, and training of deep artificial intelligence (AI) networks. 

Lie group characterize data symmetries in the underlying learning problems. Classical machine learning approaches for examining symmetries include data augmentation (data-driven) and architectural modifications to construct invariant models through weight constraint designs (model-based). A model-based G-invariant design often composed several equivariant functions followed by a final invariant function. A concurrent work suggests that building group invariance and partial invariance into string theory Kreuzer Skarke dataset improve model performance. Non group invariant models benefit from group invariant preprocessing.

From a statistical point of view, classical symmetry is related to probabilistic distributional symmetry, where exchangeability and stationarity are the primary examples. For instance, a sufficient sample statistic contains all the information needed for an inferential procedure and is directly related to probabilistic symmetry. That is, sufficiency describes the information that is relevant to the statistical inference. Probabilistic symmetry and invariance identify information that is irrelevant to the statistical inference. 

In this talk, we will review and investigate the relations between mathematical invariance, DNN invariants, (probabilistic) symmetries, physical modeling, PDE solutions, geometries and information.   

Influences of Blood Flow Waveform Uncertainties on Computational Hemodynamic Evaluation for Intracranial Aneurysms

Boundary condition (BC) is one of the most critical factors in the accuracy of hemodynamic evaluations of intracranial aneurysms (IAs) using computational fluid dynamics (CFD) modeling. Most previous investigations used a uniform rather than the patient-specific physiological blood flow waveform as BCs for numerical modeling of IAs, which could induce significant errors in risk evaluations and lead to wrong diagnoses for patients with IA symptoms. To secure the prime BC for hemodynamic modeling for IAs and quantify the hemodynamic differences under various BC strategies, this study conducted a comprehensive investigation based on Doppler ultrasound measurements and the discrete Fourier transform (DFT) simulation. First, the periodically pulsatile blood velocity at the internal carotid artery (ICA) was measured by the ultrasound flowmeter for the IA patient with a heart rate of 57 Hz, which was plotted and then phase-averaged as the baseline physiological BC for CFD modeling. Subsequently, the number of discrete points, i.e., N = 8, 16, 22, 36, and 106 on the phase-averaged waveform, were employed to generate five simulated waveforms as BCs for CFD modeling by comparing agreements in hemodynamics with the phase-averaged scenario. In addition, hemodynamic performances under the patient-specific physiological BC and a previously employed uniform BC were compared to check the errors induced by the uniform waveform assumption. The preliminary results showed that more discrete points are selected for a DFT waveform, better agreements in hemodynamics (i.e., maximum wall shear stress (WSS), surface-averaged (SA-WSS), oscillatory shear index (OSI)) on the IA sac wall can be obtained. It results in the correlation coefficients of ~ 0.94, ~0.98, ~ 0.993, ~0.996, and ~0.998 under scenarios of N = 8, 16, 22, 36, and 106, respectively. It suggests that N = 22 are acceptable prime DFT points to generate a BC for hemodynamic modeling by considering the balance of modeling accuracy and DFT complexity. Additionally, significant differences in hemodynamics resulting from uniform and patient-specific BCs suggest the latter is essential to ensure the accuracy of hemodynamic predictions for IAs.