Search NASA⌕ Search

SEARCH · Search NASA

Results for “eigenvector localization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

65 records · Page 4

Exploring the energy landscape of RBMs: reciprocal space insights into bosons, hierarchical learning and symmetry breaking

Deep generative models have become ubiquitous due to their ability to learn and sample from complex distributions. Despite the proliferation of various frameworks, the relationships among these models remain largely unexplored, a gap that hinders the development of a unified theory of AI learning. In this work, we address two central challenges: clarifying the connections between different deep generative models and deepening our understanding of their learning mechanisms. We focus on Restricted Boltzmann Machines (RBMs), a class of generative models known for their universal approximation capabilities for discrete distributions. By introducing a reciprocal space formulation for RBMs, we reveal a connection between these models, diffusion processes, and systems of coupled bosons. Our analysis shows that at initialization, the RBM operates at a saddle point, where the local curvature is determined by the singular values of the weight matrix, whose distribution follows the Marc̆enko-Pastur law and exhibits rotational symmetry. During training, this rotational symmetry is broken due to hierarchical learning, where different degrees of freedom progressively capture features at multiple levels of abstraction. This leads to a symmetry breaking in the energy landscape, reminiscent of Landau’s theory. This symmetry breaking in the energy landscape is characterized by the singular values and the weight matrix eigenvector matrix. We derive the corresponding free energy in a mean-field approximation. We show that in the limit of infinite size RBM, the reciprocal variables are Gaussian distributed. Our findings indicate that in this regime, there will be some modes for which the diffusion process will not converge to the Boltzmann distribution. To illustrate our results, we trained replicas of RBMs with different hidden layer sizes using the MNIST dataset. Our findings not only bridge the gap between disparate generative frameworks but also shed light on the fundamental processes underpinning learning in deep generative models.

97 MATHEMATICS AND COMPUTING↗

A Shifted Block Lanczos Algorithm 1: The Block Recurrence

In this paper we describe a block Lanczos algorithm that is used as the key building block of a software package for the extraction of eigenvalues and eigenvectors of large sparse symmetric generalized eigenproblems. The software package comprises: a version of the block Lanczos algorithm specialized for spectrally transformed eigenproblems; an adaptive strategy for choosing shifts, and efficient codes for factoring large sparse symmetric indefinite matrices. This paper describes the algorithmic details of our block Lanczos recurrence. This uses a novel combination of block generalizations of several features that have only been investigated independently in the past. In particular new forms of partial reorthogonalization, selective reorthogonalization and local reorthogonalization are used, as is a new algorithm for obtaining the M-orthogonal factorization of a matrix. The heuristic shifting strategy, the integration with sparse linear equation solvers and numerical experience with the code are described in a companion paper.

Grimes, Roger G.↗

SU-C-209-02: 3D Fluoroscopic Image Generation From Patient-Specific 4DCBCT-Based Motion Models Derived From Clinical Patient Images

Purpose: We develop a method to generate time varying volumetric images (3D fluoroscopic images) using patient-specific motion models derived from four-dimensional cone-beam CT (4DCBCT). Methods: Motion models are derived by selecting one 4DCBCT phase as a reference image, and registering the remaining images to it. Principal component analysis (PCA) is performed on the resultant displacement vector fields (DVFs) to create a reduced set of PCA eigenvectors that capture the majority of respiratory motion. 3D fluoroscopic images are generated by optimizing the weights of the PCA eigenvectors iteratively through comparison of measured cone-beam projections and simulated projections generated from the motion model. This method was applied to images from five lung-cancer patients. The spatial accuracy of this method is evaluated by comparing landmark positions in the 3D fluoroscopic images to manually defined ground truth positions in the patient cone-beam projections. Results: 4DCBCT motion models were shown to accurately generate 3D fluoroscopic images when the patient cone-beam projections contained clearly visible structures moving with respiration (e.g., the diaphragm). When no moving anatomical structure was clearly visible in the projections, the 3D fluoroscopic images generated did not capture breathing deformations, and reverted to the reference image. For the subset of 3D fluoroscopic images generated from projections with visibly moving anatomy, the average tumor localization error and the 95th percentile were 1.6 mm and 3.1 mm respectively. Conclusion: This study showed that 4DCBCT-based 3D fluoroscopic images can accurately capture respiratory deformations in a patient dataset, so long as the cone-beam projections used contain visible structures that move with respiration. For clinical implementation of 3D fluoroscopic imaging for treatment verification, an imaging field of view (FOV) that contains visible structures moving with respiration should be selected. If no other appropriate structures are visible, the images should include the diaphragm. This project was supported, in part, through a Master Research Agreement with Varian Medical Systems, Inc, Palo Alto, CA.

60 APPLIED LIFE SCIENCES↗

Impact of Bone Marrow Radiation Dose on Acute Hematologic Toxicity in Cervical Cancer: Principal Component Analysis on High Dimensional Data

Purpose: To study the effects of increasing pelvic bone marrow (BM) radiation dose on acute hematologic toxicity in patients undergoing chemoradiotherapy, using a novel modeling approach to preserve the local spatial dose information. Methods and Materials: The study included 37 cervical cancer patients treated with concurrent weekly cisplatin and pelvic radiation therapy. The white blood cell count nadir during treatment was used as the indicator for acute hematologic toxicity. Pelvic BM radiation dose distributions were standardized across patients by registering the pelvic BM volumes to a common template, followed by dose remapping using deformable image registration, resulting in a dose array. Principal component (PC) analysis was applied to the dose array, and the significant eigenvectors were identified by linear regression on the PCs. The coefficients for PC regression and significant eigenvectors were represented in three dimensions to identify critical BM subregions where dose accumulation is associated with hematologic toxicity. Results: We identified five PCs associated with acute hematologic toxicity. PC analysis regression modeling explained a high proportion of the variation in acute hematologicity (adjusted R{sup 2}, 0.49). Three-dimensional rendering of a linear combination of the significant eigenvectors revealed patterns consistent with anatomical distributions of hematopoietically active BM. Conclusions: We have developed a novel approach that preserves spatial dose information to model effects of radiation dose on toxicity, which may be useful in optimizing radiation techniques to avoid critical subregions of normal tissues. Further validation of this approach in a large cohort is ongoing.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Distributed Link Removal Using Local Estimation of Network Topology

In this paper, we consider the problem of network structure manipulation in the absence of information on the global network topology. In particular, the problem of removing some of links is investigated in order to slow or stop the spread of disease in a network while preserving its connectivity. Existing methods solve this combinatorial problem in a centralized manner and they require the global information of network structure. In this paper, we propose a distributed design algorithm to compute a suboptimal solution to this problem efficiently by mimicking gradient based method, namely by iteratively removing one or multiple links at a time from the network. Specifically, using matrix perturbation analysis we formulate an optimization problem involving the eigenvector associated with the largest eigenvalue of the adjacency matrix and whose solution is equal to a suboptimal solution to the original problem. This strategy also enables us to overcome the combinatorial issue of the problem. Distributed algorithms to estimate the eigenvector and to verify network’s connectivity are then proposed which facilitate us to solve the new optimization problem. In addition, topological insights into the proposed algorithm and optimality of its solution are also discussed. Lastly, the proposed distributed design method is demonstrated and evaluated via several numerical examples.

24 POWER TRANSMISSION AND DISTRIBUTION↗

Hamiltonian Truncation Effective Theory

Hamiltonian truncation is a non-perturbative numerical method for calculating observables of a quantum field theory. The starting point for this method is to truncate the interacting Hamiltonian to a finite-dimensional space of states spanned by the eigenvectors of the free Hamiltonian H_0 H 0 with eigenvalues below some energy cutoff E_\text{max} E max . In this work, we show how to treat Hamiltonian truncation systematically using effective field theory methodology. We define the finite-dimensional effective Hamiltonian by integrating out the states above E_\text{max} E max . The effective Hamiltonian can be computed by matching a transition amplitude to the full theory, and gives corrections order by order as an expansion in powers of 1/E_\text{max} 1 / E max . The effective Hamiltonian is non-local, with the non-locality controlled in an expansion in powers of H_0/E_\text{max} H 0 / E max . The effective Hamiltonian is also non-Hermitian, and we discuss whether this is a necessary feature or an artifact of our definition. We apply our formalism to 2D \lambda\phi^4 λ ϕ 4 theory, and compute the the leading 1/E_\text{max}^2 1 / E max 2 corrections to the effective Hamiltonian. We show that these corrections nontrivially satisfy the crucial property of separation of scales. Numerical diagonalization of the effective Hamiltonian gives residual errors of order 1/E_\text{max}^3 1 / E max 3 , as expected by our power counting. We also present the power counting for 3D \lambda \phi^4 λ ϕ 4 theory and perform calculations that demonstrate the separation of scales in this theory.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Projecting the Thermal Response in a HTGR-Type System during Conduction Cooldown Using Graph-Laplacian Based Machine Learning

Accurate prediction of an off-normal event in a nuclear reactor is dependent upon the availability of sensory data, reactor core physical condition, and understanding of the underlying phenomenon. This work presents a method to project the data from some discrete sensory locations to the overall reactor domain during conduction cooldown scenarios similar to High Temperature Gas-cooled Reactors (HTGRs). The existing models for conductive cooldown in a heterogeneous multi-body system, such as an assembly of prismatic blocks or pebble beds relies on knowledge of the thermal contact conductance, requiring significant knowledge of local thermal contacts and heat transport possibilities across those contacts. With a priori knowledge of bulk geometry features and some discrete sensors, a machine learning approach was devised. The presented work uses an experimental facility to mimic conduction cooldown with an assembly of 68 cylindrical rods initially heated to 1200 K. High-fidelity temperature data were collected using an infrared (IR) camera to provide training data to the model and validate the predicted temperature data. The machine learning approach used here first converts the macroscopic bulk geometry information into Graph-Laplacian, and then uses the eigenvectors of the Graph-Laplacian to develop Kernel functions. Support vector regression (SVR) was implemented on the obtained Kernels and used to predict the thermal response in a packed rod assembly during a conduction cooldown experiment. The usage of SVR modeling differs from most models today because of its representation of thermal coupling between rods in the core. When trained with thermographic data, the average normalized error is less than 2% over 400 s, during which temperatures of the assembly have dropped by more than 500 K. The rod temperature prediction performance was significantly better for rods in the interior of the assembly compared to those near the exterior, likely due to the model simplification of the surroundings.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

GPU-acceleration of the ELPA2 distributed eigensolver for dense symmetric and hermitian eigenproblems

The solution of eigenproblems is often a key computational bottleneck that limits the tractable system size of numerical algorithms, among them electronic structure theory in chemistry and in condensed matter physics. Large eigenproblems can easily exceed the capacity of a single compute node, thus must be solved on distributed-memory parallel computers. We here present GPU-oriented optimizations of the ELPA two-stage tridiagonalization eigensolver (ELPA2). On top of cuBLAS-based GPU offloading, we add a CUDA kernel to speed up the back-transformation of eigenvectors, which can be the computationally most expensive part of the two-stage tridiagonalization algorithm. Furthermore, we benchmark the performance of this GPU-accelerated eigensolver on two hybrid CPU–GPU architectures, namely a compute cluster based on Intel Xeon Gold CPUs and NVIDIA Volta GPUs, and the Summit supercomputer based on IBM POWER9 CPUs and NVIDIA Volta GPUs. Consistent with previous benchmarks on CPU-only architectures, the GPU-accelerated two-stage solver exhibits a parallel performance superior to the one-stage counterpart. Finally, we demonstrate the performance of the GPU-accelerated eigensolver developed in this work for routine semi-local KS-DFT calculations comprising thousands of atoms.

97 MATHEMATICS AND COMPUTING↗

A method of surface marker location optimization for tumor motion estimation in lung stereotactic body radiation therapy

Purpose: Accurately localizing lung tumor localization is essential for high-precision radiation therapy techniques such as stereotactic body radiation therapy (SBRT). Since direct monitoring of tumor motion is not always achievable due to the limitation of imaging modalities for treatment guidance, placement of fiducial markers on the patient’s body surface to act as a surrogate for tumor position prediction is a practical alternative for tracking lung tumor motion during SBRT treatments. In this work, the authors propose an innovative and robust model to solve the multimarker position optimization problem. The model is able to overcome the major drawbacks of the sparse optimization approach (SOA) model. Methods: The principle-component-analysis (PCA) method was employed as the framework to build the authors’ statistical prediction model. The method can be divided into two stages. The first stage is to build the surrogate tumor matrix and calculate its eigenvalues and associated eigenvectors. The second stage is to determine the “best represented” columns of the eigenvector matrix obtained from stage one and subsequently acquire the optimal marker positions as well as numbers. Using 4-dimensional CT (4DCT) and breath hold CT imaging data, the PCA method was compared to the SOA method with respect to calculation time, average prediction accuracy, prediction stability, noise resistance, marker position consistency, and marker distribution. Results: The PCA and SOA methods which were both tested were on all 11 patients for a total of 130 cases including 4DCT and breath-hold CT scenarios. The maximum calculation time for the PCA method was less than 1 s with 64 752 surface points, whereas the average calculation time for the SOA method was over 12 min with 400 surface points. Overall, the tumor center position prediction errors were comparable between the two methods, and all were less than 1.5 mm. However, for the extreme scenarios (breath hold), the prediction errors for the PCA method were not only smaller, but were also more stable than for the SOA method. Results obtained by imposing a series of random noises to the surrogates indicated that the PCA method was much more noise resistant than the SOA method. The marker position consistency tests using various combinations of 4DCT phases to construct the surrogates suggested that the marker position predictions of the PCA method were more consistent than those of the SOA method, in spite of surrogate construction. Marker distribution tests indicated that greater than 80% of the calculated marker positions fell into the high cross correlation and high motion magnitude regions for both of the algorithms. Conclusions: The PCA model is an accurate, efficient, robust, and practical model for solving the multimarker position optimization problem to predict lung tumor motion during SBRT treatments. Due to its generality, PCA model can also be applied to other imaging guidance system whichever using surface motion as the surrogates.

60 APPLIED LIFE SCIENCES↗

Derivation of a Markov state model of the dynamics of a protein-like chain immersed in an implicit solvent

A Markov state model of the dynamics of a protein-like chain immersed in an implicit hard sphere solvent is derived from first principles for a system of monomers that interact via discontinuous potentials designed to account for local structure and bonding in a coarse-grained sense. The model is based on the assumption that the implicit solvent interacts on a fast time scale with the monomers of the chain compared to the time scale for structural rearrangements of the chain and provides sufficient friction so that the motion of monomers is governed by the Smoluchowski equation. A microscopic theory for the dynamics of the system is developed that reduces to a Markovian model of the kinetics under well-defined conditions. Microscopic expressions for the rate constants that appear in the Markov state model are analyzed and expressed in terms of a temperature-dependent linear combination of escape rates that themselves are independent of temperature. Excellent agreement is demonstrated between the theoretical predictions of the escape rates and those obtained through simulation of a stochastic model of the dynamics of bond formation. Finally, the Markov model is studied by analyzing the eigenvalues and eigenvectors of the matrix of transition rates, and the equilibration process for a simple helix-forming system from an ensemble of initially extended configurations to mainly folded configurations is investigated as a function of temperature for a number of different chain lengths. For short chains, the relaxation is primarily single-exponential and becomes independent of temperature in the low-temperature regime. The profile is more complicated for longer chains, where multi-exponential relaxation behavior is seen at intermediate temperatures followed by a low temperature regime in which the folding becomes rapid and single exponential. It is demonstrated that the behavior of the equilibration profile as the temperature is lowered can be understood in terms of the number of relaxation modes or “folding pathways” that contribute to the evolution of the state populations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Mean Flow Augmented Acoustics in Rocket Systems

Oscillatory motion in solid rocket motors and liquid engines has long been a subject of concern. Many rockets display violent fluctuations in pressure, velocity, and temperature originating from the complex interactions between the combustion process and gas dynamics. The customary approach to modeling acoustic waves inside a rocket chamber is to apply the classical inhomogeneous wave equation to the combustion gas. The assumption of a linear, non-dissipative wave in a quiescent fluid remains valid while the acoustic amplitudes are small and local gas velocities stay below Mach 0.2. The converging section of a rocket nozzle, where gradients in pressure, density, and velocity become large, is a notable region where this approach is not applicable. The expulsion of unsteady energy through the nozzle of a rocket is identified as the predominate source of acoustic damping for most rocket systems. An accurate model of the acoustic behavior within this region where acoustic modes are influenced by the presence of a steady mean flow is required for reliable stability predictions. Recently, an approach to address nozzle damping with mean flow effects was implemented by French [1]. This new approach extends the work originated by Sigman and Zinn [2] by solving the acoustic velocity potential equation (AVPE) formulated by perturbing the Euler equations [3]. The acoustic velocity potential (psi) describing the acoustic wave motion in the presence of an inhomogeneous steady high-speed flow is defined by, (del squared)(psi) − (lambda/c)(exp 2)(psi) − M(dot)[M(dot)(del)(del(psi))] − 2(lambda(M/c) + (M(dot)del(M))(dot)del(psi)−2(lambda)(psi)[M(dot)del(1/c)]=0 (1) with M as the Mach vector, c as the speed of sound, and lambda as the complex eigenvalue. French apply the finite volume method to solve the steady flow field within the combustion chamber and nozzle with inviscid walls. The complex eigenvalues and eigenvector are determined with the use of the ARPACK eigensolver. The present study employs the COMSOL Multphysics framework to solve the coupled eigenvalue problem using the finite element approach. The study requires one way coupling of the CFD High Mach Number Flow (HMNF) and mathematics module. The HMNF module evaluated the gas flow inside of a solid rocket motor using St. Robert's law modeling solid propellant burn rate, slip boundary conditions, and the supersonic outflow condition. Results from the HMNF model are used by the coefficient form of the mathematics module to determine the eigenvalues of the AVPE. The mathematics model is truncated at the nozzle sonic line, where a zero flux boundary condition is self-satisfying. The remaining boundaries are modeled with a zero flux boundary condition, assuming zero acoustic absorption on all surfaces. Pertinent results from these analyses are the complex valued eigenvalue and eigenvectors. Comparisons are made to the French results to evaluate the modeling approach. A comparison of the French results with that of the present analysis is displayed in figures 1 and 2, respectively. The graphic shows the first tangential eigenvector's real (a) and imaginary (b) values.

Fischbach, Sean R.↗