Physics-Informed Machine Learning: Theory, Algorithms and Applications

Wang, Sifan University of Pennsylvania ProQuest Dissertations 2023 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The remarkable potential of deep learning in areas from computer vision to natural language processing has now found profound implications in modeling and simulating physical systems. Central to these advancements is the emerging field of physics-informed machine learning, a fusion of physical principles with machine learning techniques. There are three predominant strategies to integrate physics: inductive bias, learning bias, and observational bias.Our study delves into the inherent challenges and limitations of physics-informed machine learning, particularly in the physics-informed neural networks (PINNs) and deep operator networks (DeepONet). Our research is driven by overcoming fundamental challenges and enhancing the performance of these frameworks. Firstly, we investigate the gradient flow of PINNs, identifying a training failure stemming from unbalanced back-propagated gradients. This insight motivates us to generalize the neural tangent kernel (NTK) theory to PINNs. With this tool, we theoretically reveal that the training of PINNs suffer from spectral bias, causality violation and discrepancy in convergence rate of loss term. To address these critical issues, we propose several simple yet effective loss re-weighting algorithms and network architecture and validate them across a wide range range of representative benchmarks in computational physics. Besides, we present an extension of PINNs framework for solving free boundary problems.Moreover, we highlight the data-intensive demands of training neural operators and the potential inconsistency of their predictions with the underlying physics. To resolve these challenges, we propose physics-informed DeepONet, introducing a simple and effective regularization mechanism for biasing the outputs of DeepONet models towards ensuring physical consistency. Based on that, we propose a autoreressive training algorithm for performing long-time integration of evolution equations. We also analyze the training dynamics of DeepONets through the lens of NTK theory, uncovering a bias that favors the approximation of functions with larger magnitudes. Therefore, we propose a point-wise loss re-weighting algorithm to correct this bias and a novel network architecture that is more resilient to vanishing gradient pathologies. We leverage the proposed physics-informed DeepONet to build fast and differentiable surrogates for rapidly solving PDE-constrained optimization problems, even in the absence of any paired input-output training data. In summary, this thesis provides in-depth exploration into training, improving and applications aspects of physics-informed machine learning, paving a new way to developing scientific machine learning algorithms with better robustness and accuracy guarantees, as needed for many critical applications in computational science and engineering.

Studies of Higgs Boson Self-Interaction via Higgs Boson Pair Production in the ATLAS Experiment and the Development of Prior-Assisted Anomaly Detection Methods for New Physics Searches

Cheng, Chi Lung The University of Wisconsin - Madison ProQuest Dis 2025 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Particle physics stands at the forefront of humanity's endeavor to understand our physical reality, studying the elementary particles and the forces that govern their interactions. This thesis presents contributions to this scientific effort, detailing experimental research conducted as part of the ATLAS Collaboration at the CERN Large Hadron Collider (LHC). It also introduces novel machine learning methods designed for model-agnostic searches for new physics. The work presented herein addresses fundamental questions at the heart of the Standard Model (SM) and explores avenues for discovering physics beyond it.The first part focuses on an investigation into the Higgs boson's self-interaction via Higgs boson pair production (HH) processes. Measuring the Higgs self-coupling is essential for experimentally reconstructing the Higgs potential, providing important insights into the mechanism of electroweak symmetry breaking and offering a window into physics beyond the Standard Model. This work presents a dedicated search in the HH to bb final state using the full 140fb-1 Run 2 dataset collected by the ATLAS experiment at √s=13 TeV. The analysis strategy was optimized for sensitivity to both the dominant gluon-gluon fusion and sub-dominant vector-boson fusion production modes. Furthermore, this thesis details the statistical combination of this search with four other ATLAS Run 2 HH analyses. At the time of publication, this combination yielded the most stringent constraints derived from the LHC Run 2 data: an observed (expected) 95% CL upper limit is set on the SM HH production signal strength modifier of µHH < 2.9 (2.4). The combination constrains the Higgs boson trilinear self-coupling modifier, κλ = λHHH/λHHH SM, to the interval [-1.2, 7.2] (expected [-1.6, 7.2]) as well as the quartic HHVV coupling modifier, κ2V, to [0.6, 1.5] (expected [0.4, 1.6]) at 95% CL. These results are consistent with the Standard Model predictions.The second part shifts focus to model-agnostic search strategies, addressing the challenge of discovering unexpected new physics phenomena. Novel machine learning techniques for anomaly detection are introduced. This work presents a method called Prior Assisted Weak Supervision (PAWS), developed to significantly enhance search sensitivity by incorporating physics knowledge to guide the learning process. This physics knowledge, serving as a prior, defines a restricted, physically motivated space of potential signal functions, which focuses the search when looking for potential anomalies in the data. On benchmark simulated dijet resonance datasets, PAWS demonstrates over an order-of-magnitude improvement in sensitivity compared to conventional weakly supervised approaches. This method is further extended into a complete framework for statistical inference, Generator Based Inference (GBI), enabling not only the detection of anomalies but also the direct, quantitative measurement of their physical properties (e.g., particle mass, signal fraction) with well-defined confidence intervals. Studies on simulated datasets show the GBI-PAWS framework can detect signals with significances as low as 0.1σ and provide accurate, unbiased parameter estimates. While these methods show significant promise, their application to experimental data remains future work.Together, these efforts contribute to the ongoing exploration of fundamental physics, providing experimental constraints on the Higgs boson's self-interaction while introducing effective anomaly detection methodologies for future discoveries.

Applications of Complex Network Dynamics in Ultrafast Electronics

Charlot, Noeloikeau ProQuest Dissertations & Theses The Ohio State Uni 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The success of modern digital electronics relies on compartmentalizing logical functions into individual gates, and controlling their order of operations via a global clock. In the absence of such a timekeeping mechanism, systems of connected logic gates can quickly become chaotic and unpredictable -- exhibiting analog, asynchronous, autonomous dynamics. Such recurrent circuitry behaves in a manner more consistent with neural networks than digital computers, exchanging and conducting electricity as quickly as its hardware allows. These physics enable new forms of information processing that are faster and more complex than clocked digital circuitry. However, modern electronic design tools often fail to measure or predict the properties of large recurrent networks, and their presence can disrupt other clocked architectures.In this thesis, I study and apply the physics of complex networks of self-interacting logic gates at sub-ns timescales. At a high level, my unique contributions are: 1. I derive a general theory of network dynamics and develop open-source simulation libraries and experimental circuit designs to re-create this work; 2. I invent a best-in-class digital measurement system to experimentally analyze signals at the trillionth-of-a-second (ps) timescale; 3. I introduce a network computing architecture based on chaotic fractal dynamics, creating the first `physically unclonable function' with near-infinite entropy.In practice, I use a digital computer to reconfigure a tabletop electronic device containing millions of logic gates (a field-programmable gate array; FPGA) into a network of Boolean functions (a hybrid Boolean network; HBN). From within the FPGA, I release the HBN from initial conditions and measure the resulting state of the network over time. These data are transferred to an external computer and used to study the system experimentally and via a mathematical model.Existing mathematical theories and FPGA simulation tools produce incorrect results when predicting HBNs, and current FPGA-based measurement tools cannot reliably capture the ultrafast HBN dynamics. Thus I begin by generalizing prior mathematical models of Boolean networks in a way that reproduces extant models as limiting cases. Next I design a ps-scale digital measurement system (Waveform Capture Device; WCD). The WCD is an improvement to the state-of-the-art in FPGA measurement systems, having external application in e.g. medical imaging and particle physics. I validate the model and WCD independently, showing that they reproduce each-other in a self-consistent manner. I use the WCD to fit the model parameters and predict the behavior of simple HBNs on FPGAs.I go on to study chaotic HBN. I find that infinitesimal changes to the model parameters -- as well as uncontrollable manufacturing variations inherent to the FPGAs - cause near-identical HBNs to differ exponentially. The simulations predict that fractal patterns separate infinitesimally distinct networks over time, motivating the use of HBN dynamics as 'digital fingerprints' (Physically Unclonable Functions; PUFs) for hardware security. I conclude by rigorously analyzing the experimental properties of HBN-PUFs on FPGAs across a variety of statistical metrics, ultimately discovering super-exponential entropy scaling -- a significant improvement to the state-of-the-art.

Exploring the Thermal Transport and Phase Nucleation Physics of Nanowires via Suspended Microdevices

Jin, Lei University of California, Berkeley ProQuest Disser 2021 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The advance in nanofabrication and characterization methods enables researchers to study nanomaterials at small scales. Rich physics can emerge in nanomaterials due to their low-dimensional nature, such as boundary effect, quantum confinement effect etc. Also, nanomaterials usually have the most single-crystalline pristine qualities. This opens opportunities to study the intrinsic properties of many materials, which is hard to probe in bulk counterparts as the gain boundaries and defects in bulk samples can degrade their qualities. Combining the nanofabricated suspended microdevice and nanomaterials such as nanowires or nanoribbon, we can probe the intrinsic properties of materials which have interesting physics. Nanomaterials also have their unique structure and thermal physics properties which are very different from those of their bulk counterparts. In this dissertation, the author presents his work, in his pursuits of PhD degree at UC Berkeley, focused on exploring the thermal and transport physics of nanowire via transport measurement realized by using nanofabricated suspended microdevices.Chapter 2 and 3 are focused on vanadium dioxide (VO2), a strongly correlated materials with metal-to-insulator transition (MIT) occurring at temperature of 341 K. Although the first observation of the MIT of VO2 was more than 60 years ago, the transition mechanism and the correlation physics of VO2 is still under debate among theorists. Thanks to the advances in nanofabrication, measuring and modulating the MIT of single-crystalline VO2 nanowires are made possible and accessible to experimentalist. This provides us an efficient tool to study the intrinsic properties of VO2 and could help unveil its correlated nature. In chapter 2, a discovery of the recovery of Wiedemann-Franz (W-F) law in metallic VO2 is reported. This recovery is realized by introducing point defects using energetic ion irradiation. The experimental measurements and theoretical explanations are demonstrated to explain this full restoration of otherwise violated W-F law in strongly correlated system. In chapter 3, a deeply supercooled intrinsic VO2 metal is first realized by technique called irradiation shielding supported on suspended microdevices. Nucleation seeds created by ion surface milling are then introduced on nanowires' surface and the critical nucleation sizes of supercooled metallic VO2 are measured for the first time. A model derived from classical nucleation theory is also presented.In chapter 4, experimental observations of abnormal thermal transport behaviors of twisted layered GeS nanowires with a screw dislocation core in the center is reported. The thermal conductivity of twisted GeS nanowires is measured by using suspended microdevices, which increases by decreasing diameters of nanowires. This trend is against to the prediction from boundary scattering, which is commonly observed in nanowires such as silicon, silver etc. In contrast, the thermal conductivities of non-twisted layered GeS nanowires and GeS nanoribbon are also measured. Their thermal transport behaves the same as common nanowires, that is thinner wires have lower thermal conductivity. The abnormal increasement of thermal conductivity at smaller diameters of twisted nanowires must origin from either their dislocation cores or the twisting topology between layers. New thermal physics may arise in this type of materials as classical explanation of the measured results fails to give full explanations.

Light transport on path-space manifolds

Jakob, Wenzel Alban Cornell University 2013 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The pervasive use of computer-generated graphics in our society has led to strict demands on their visual realism. Generally, users of rendering software want their images to look, in various ways, "real", which has been a key driving force towards methods that are based on the physics of light transport. Until recently, industrial practice has relied on a different set of methods that had comparatively little rigorous grounding in physics---but within the last decade, advances in rendering methods and computing power have come together to create a sudden and dramatic shift, in which physics-based methods that were formerly thought impractical have become the standard tool. As a consequence, considerable attention is now devoted towards making these methods as robust as possible. In this context, robustness refers to an algorithm's ability to process arbitrary input without large increases of the rendering time or degradation of the output image. One particularly challenging aspect of robustness entails simulating the precise interaction of light with all the materials that comprise the input scene. This dissertation focuses on one specific group of materials that has fundamentally been the most important source of difficulties in this process. Specular materials, such as glass windows, mirrors or smooth coatings (e.g. on finished wood), account for a significant percentage of the objects that surround us every day. It is perhaps surprising, then, that it is not well-understood how they can be accommodated within the theoretical framework that underlies some of the most sophisticated rendering methods available today. Many of these methods operate using a theoretical framework known as path space integration. But this framework makes no provisions for specular materials: to date, it is not clear how to write down a path space integral involving something as simple as a piece of glass. Although implementations can in practice still render these materials by side-stepping limitations of the theory, they often suffer from unusably slow convergence; improvements to this situation have been hampered by the lack of a thorough theoretical understanding. We address these problems by developing a new theory of path-space light transport which, for the first time, cleanly incorporates specular scattering into the standard framework. Most of the results obtained in the analysis of the ideally smooth case can also be generalized to rendering of glossy materials and volumetric scattering so that this dissertation also provides a powerful new set of tools for dealing with them. The basis of our approach is that each specular material interaction locally collapses the dimension of the space of light paths so that all relevant paths lie on a submanifold of path space. We analyze the high-dimensional differential geometry of this submanifold and use the resulting information to construct an algorithm that is able to "walk" around on it using a simple and efficient equation-solving iteration. This manifold walking algorithm then constitutes the key operation of a new type of Markov Chain Monte Carlo (MCMC) rendering method that computes lighting through very general families of paths that can involve arbitrary combinations of specular, near-specular, glossy, and diffuse surface interactions as well as isotropic or highly anisotropic volume scattering. We demonstrate our implementation on a range of challenging scenes and evaluate it against previous methods.

Physics-Based Damage Modeling of Ceramic Matrix Composites for Extreme Heat Environments

Alabdullah, Mohammad A B F ProQuest Dissertations & Theses University of Cali 2019 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

High-temperature composite materials have broad applications in the energy and transportation sectors because of their unique capabilities to sustain strength at elevated temperatures, inert interaction with some coolants, and their high strength-to-weight ratio. These encompass C/C and SiC/SiC composites, developed as cladding in fission reactors, and as turbine blade materials for advanced jet engines and combustion liners. Cyclic thermal and mechanical loading associated with neutron irradiation effects of these composites leads to wide-spread and progressive micro-cracking that leads to loss of thermal conductivity and further enhancement of thermo-mechanical damage. A physics-based model of wide-spread micro-crcaking is developed within the thermodynamic framework of continuum damage mechanics. Evolution equations for damage parameters that describe the growth of continuum damage are developed, where the material variables are obtained from experiments. The model novelty is in coupling mechanical, thermal, and irradiation damage through a consistent thermodynamic framework, including loss of thermal conductivity due to the evolution of mechanically induced micro-cracks. A number of thermo-mechanical experiments were conducted to confirm model assumptions. The model is shown to be validated with out-of-pile experiments, and then implemented using commercial finite element code COMSOL to the fuel cladding problem and FNSF blanket where coupling between different physics was achieved along with an implementation of global local approach. Finally, a set of performance diagrams for a thin-walled SiC tube structure subjected to cyclic thermo-mechanical loading associated with neutron irradiation were introduced to aid the community with first stage design tool for a thin walled CMC cylindrical structures.

Thermodynamically Consistent Physics-Informed Data-Driven Computing and Reduced-Order Modeling of Nonlinear Materials

He, Xiaolong ProQuest Dissertations & Theses University of Cali 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Physical simulations have influenced the advancements in engineering, technology, and science more rapidly than ever before. However, it remains challenging for effective and efficient modeling of complex linear and nonlinear material systems based on phenomenological approaches which require predefined functional forms. The goal of this dissertation is to enhance the predictivity and efficiency of physical simulations by developing thermodynamically consistent data-driven computing and reduced-order materials modeling methods based on emerging machine learning techniques for manifold learning, dimensionality reduction, sequence learning, and system identification.For reversible mechanical systems, we first develop a new data-driven material solver built upon local convexity-preserving reconstruction to capture anisotropic material behaviors and enable data-driven modeling of nonlinear anisotropic elastic solids. A material anisotropic state characterizing the underlying material orientation is introduced for the manifold learning projection in the material solver. To counteract the curse of dimensionality and enhance the generalization ability of data-driven computing, we employ deep autoencoders to discover the underlying low-dimensional manifold of material database and integrate a convexity-preserving interpolation scheme into the novel autoencoder-based data-driven solver to further enhance efficiency and robustness of data searching and reconstruction during online data-driven computation. The proposed approach is shown to achieve enhanced efficiency and generalization ability over a few commonly used data-driven schemes.For irreversible mechanical systems, we develop a thermodynamically consistent machine learned data-driven constitutive modeling approach for path-dependent materials based on measurable material states, where the internal state variables essential to the material path-dependency are inferred automatically from the hidden state of recurrent neural networks. The proposed method is shown to successfully model soil behaviors under cyclic shear loading using experimental stress-strain data.Lastly, we develop a non-intrusive accurate and efficient reduced-order model based on physics-informed adaptive greedy latent space dynamics identification (gLaSDI) for general high-dimensional nonlinear dynamical systems. An autoencoder and dynamics identification models are trained simultaneously to discover intrinsic latent space and learn expressive governing equations of simple latent-space dynamics. To maximize and accelerate the exploration of the parameter space for optimal model performance, an adaptive greedy sampling algorithm integrated with a physics-informed residual-based error indicator and random-subset evaluation is introduced to search for optimal training samples on the fly, which outperforms the conventional predefined uniform sampling. Compared with the high-fidelity models of various nonlinear dynamical problems, gLaSDI achieves 66 to 4,417x speed-up with 1 to 5% relative errors.

Beyond the Standard Model Phenomena: From Model Building to Searches

Chiu, Wen Han The University of Chicago ProQuest Dissertations & 2023 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The Standard Model (SM) of particle physics is an effective theory for describing the interactions of fundamental particles and their properties. In spite of countless experimental agreements with the predictions of the SM, we have good reason to believe that the SM is not a complete description of particle physics. This unknown description of particle physics which is beyond that of the SM is the focus of this dissertation.We start by studying model-independent flavor non-universal UV effects at proton-proton colliders, low-energy flavor physics, and exotic Higgs decays. We then discuss the details of defining an arrival time for a jet to better search for long-lived particles which decay hadronically. Next, we study how an extended Higgs sector with an approximate ℤ2 symmetry can affect electroweak precision observables, collider experiments, and cosmology. Lastly, we study a framework for a dark matter model where the dark sector was described by an approximate conformal field theory at early times.

BICEP3 and CMB-S4: Current and Future CMB Polarization Experiments to Probe Fundamental Physics

Wu, Wai Ling Kimmy ProQuest Dissertations & Theses Stanford Universit 2015 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Cosmic Microwave Background (CMB) polarization is a powerful probe of fundamental physics – from the physics of inflation to the sum of neutrino masses. BICEP3 is a new 95GHz receiver in the BICEP/Keck Array series of inflationary probes at the South Pole sensitive to the CMB polarization at degree-angular scales. The goal of the BICEP/Keck program is to test one of inflation’s prediction: generation of stochastic gravitational wave background. This gravitational wave background imprints B-mode polarization patterns on the CMB which peaks at 2 degrees in the power spectrum. BICEP3 advances per-receiver sensitivity, while maintaining the advantages of a compact refractor with degree-angular resolution. BICEP3 doubles the aperture of BICEP2/Keck receivers, has faster optics, and can house 1280 dualpolarization pixels on its focal plane. This is made possible by a redesign of the receiver – we implemented large area metal mesh filters to reduce infrared loading, used alumina as absorptive filters and lenses to reduce optical loading, and repackaged the detectors and readout electronics into modules for scalability. This thesis details the instrument design of BICEP3, with discussions on initial performance from instrument characterization measurements and preliminary maps made from the first 3 months of observations in 2015. Together with multi-frequency observation data from Planck, BICEP2, and the Keck Array, BICEP3 is projected to be able to set upper limits on the tensor-to-scalar ratio to r ≤ 0.03 at 95% C.L..CMB-S4 is a future ground-based CMB polarization experiment, planned to observe large fractions of the sky (> 50%), have high resolution (≤ 3' ) focusing on the arc-minute scale B-mode spectrum generated by lensing, and orders of magnitude more detectors than current generation of experiments (200, 000+). The high signal-to-noise measurement of E-modes and B-modes of the CMB enables delensing through the E-to-B channel and reconstructing the lensing potential. We can therefore study 1) the gravitational wave generated B-modes with lensing removed, 2) physics that changes the shape of the lensing potential, and 3) further constraint and verify cosmological parameters through the E-modes. This thesis focuses on investigating how a wider range of input experiment configurations (sky fraction, detector count, beam sizes) for CMB-S4 changes the constraints on physics parameters of interests. With unprecedented sensitivity, CMB-S4 is projected to place competitive constraints on these areas of physics: cosmic neutrino background, dark matter, dark energy, and inflation. Specifically, combing a range of CMB-S4 experiment configurations with low redshift survey data can 1 sigma constraint on the total neutrino mass to σ(Mν) < 20meV, sufficient for a cosmological detection of sum of neutrino mass in the inverse mass hierarchy.

Search for New Physics Through a Long-Lived Di-Muon Resonance in the NuMI Beam With the Icarus Detector

Putnam, Gray Louis Campbell The University of Chicago ProQuest Dissertations & 2024 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Many of the unexplained phenomena in particle physics and cosmology today, such as the microphysical nature of dark matter, the strong CP problem, and the origin of the neutrino masses, can be resolved by the existence of a light (~GeV), weakly-coupled hidden sector of new physics. Such hidden sectors often predict the existence of "long-lived" particles (LLPs) that travel a far distance from production before decaying into Standard Model particles. Neutrino oscillation experiments, which combine intense particle beams with precise imaging detectors, are well equipped to probe LLP models with new sensitivity. This thesis details a search for a long-lived particle decaying to two muons with the ICARUS liquid argon time projection chamber (LArTPC) neutrino detector in the Short-Baseline Neutrino program at Fermilab. The calibration of the ICARUS time projection chamber (TPC) which enables the search is also presented. Notably, the calibration measures an angular dependence in electron-ion recombination in argon, a novel effect in the detector physics of LArTPCs. The search is performed using data taken with the Neutrinos at the Main Injector (NuMI) beam, with an exposure of 2.41 x 1020 protons on target. No significant excess over background is observed, and we set world-leading limits on two new physics models that predict this process: the Higgs portal scalar and a heavy axion model. We also present the sensitivity in a model-independent way applicable to any new physics model predicting the process K → π + S(→μμ), for a long-lived particle S.