Andrew Valentini

Andrew Valentini

Second-year Physics PhD
Generative models for materials discovery

Research

Diffusion-Based Discovery of Ordered Superconductors

Abstract: To address the disconnect between novel computational predictions and experimental realization in AI-driven superconductor discovery pipelines, we fine-tune a diffusion foundation model on 7,183 superconductors labeled with crystallographic disorder probability and superconducting critical temperature, and employ classifier-free guidance to steer sampling towards ordered and high-temperature superconductors. We demonstrate that this model exhibits a 2.05x improvement in the ability to generate ordered structures with respect to an unguided baseline, from 31.80% to 65.13%, in addition to a 4.16x improvement in the median superconducting critical temperature of generated materials, from a median Tc of 0.99K to 4.13K. Structures generated from this fine-tuned model are then screened through a multi-stage pipeline which employs machine learning and density functional theory calculations to assess disorder probability, thermodynamic stability, and superconducting properties, with selected structures advancing to experimental synthesis.

Density plots of critical temperature and disorder probability for the training data, Tc-guided samples, and Tc and disorder-guided samples
Distributions of superconducting critical temperature (left) and disorder probability (right) for the training data and for structures generated with Tc guidance alone and with joint Tc & disorder guidance.

Open Quantum Systems Theory and Simulation (NSF-REU at Penn State)

Abstract: Motivated by the possible information to be gained from studying the evolution of ensembles of interacting quantum systems in a cosmological context, we develop a simulation of an idealized quantum circuit. We treat the full circuit as an ensemble of open systems and track the state of these subsystems as correlations between them develop. We use violations of complete positivity on the mappings that evolve subsystems as a diagnosis for entanglement in the circuit and study the domain of positivity for these mappings to determine between which subsystems the entanglement exists at a given time. We additionally define and track thermodynamic properties on the circuit to study how subsystems are able to resist thermalization while the full circuit evolves. This process is carried out for circuits of varying interaction, strengthening our understanding of the impact conditions on these mappings has on the evolution of subsystems.

Gravitational Wave Detector Characterization (NSF-REU)

Abstract: The prompt discovery of gravitational wave events is performed by a variety of matched-filtering search pipelines. Due to the non-Gaussian nature of the Advanced LIGO detector, these pipelines often react to transient noise sources, commonly referred to as glitches, and mistake them for gravitational wave events. Analyzing the properties of these glitches and a pipeline’s reaction to them is thus crucial to improve pipeline efficiency and event validation accuracy. In this study, we perform such an analysis on one of these pipelines, GstLAL, with the intent of providing a battery of tests and models by which a given gravitational wave candidate may be evaluated against to provide a higher degree of confidence in the identification of the event origin in addition to offering a valuable perspective on the responsiveness of this pipeline to instrumental noise.

Correlation Functions in Materials Theory and Cosmology

Abstract: Dislocations are a type of defect or irregularity in the structure of a material that is otherwise ordered and crystalline. These dislocations either extend across the entirety of a material or self-terminate, forming a loop. Dislocation loops are an important subject of research in solid state physics as well as in materials science because they are the fundamental carriers of deformation in metals. Modeling the distribution of these dislocation loops is crucial to understanding material properties, such as predicting the plastic response of metals. Previous research has developed pair correlation functions for dislocation loops, which describe the relative arrangement of pairs of dislocation lines, using a stochastic geometric approach. In this work, we summarize this geometric approach and propose an alternative method based on Fourier analysis for computing these dislocation loop correlations. We discuss the benefits of this proposed method and consider applications to other linear objects such as cosmological defects.