Thesis (Ph.D.)--Lehigh University, 2008. / Adviser: Slava V. Rotkin.

Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2008. / Source: Dissertation Abstracts International, Volume: 69-11, Section: B, page: 6859. Adviser: Paul M. Ricker. Includes bibliographical references (leaves 141-156) Available on microfilm from Pro Quest Information and Learning.

This thesis presents five contributions to machine learning, with themes of differentiability and Bayesian inference. We present Firefly Monte Carlo, an auxiliary variable Markov chain Monte Carlo algorithm that only queries a potentially small subset of data at each iteration yet simulates from the exact posterior distribution. We describe the design and implementation of Autograd, a software package for efficiently computing derivatives of functions written in Python/Numpy using reverse accumulation mode differentiation. Using Autograd, we develop a convolutional neural network that takes arbitrary graphs, such as organic molecules, as input. This generalizes standard molecular feature representations and allows end-to-end adaptation of the feature extraction pipeline to particular tasks. We show how to compute gradients of cross-validation loss with respect to hyperparameters of learning algorithms, with both time and memory efficiency, by chaining gradients backwards through an exactly reversed optimization procedure. Finally, by accounting for the entropy destroyed by optimization, we show that early stopping and ensembling, popular tricks for avoiding overfitting, can be interpreted as variational Bayesian inference. / Physics

Seventy-five synthetic powder trioctahedral mica samples (between Mg, Co, Ni, and Fe end members, with different degrees of oxidation, vacancy and Al/Si contents, and including an OH/F substitution series) were studied by room-temperature powder X-ray diffraction. The iron-bearing samples were studied by 57Fe Mossbauer spectroscopy. Subsets of the samples were also characterized by scanning electron microscopy combined with energy dispersive spectroscopy, optical microscopy, X-ray fluorescence spectroscopy, and gas chromatography. Lattice parameters (refined under the 1M stacking polytype, space group C2/m) were determined for all powder samples and iron site populations ([4]Fe 3+, [6]Fe2+, and [6]Fe 2+) were obtained from Mossbauer spectroscopy. The relation (c/a)cosbeta* = 113 was found to hold exactly (within experimental error) for all synthetic powders whereas it does not hold in general for synthetic and natural 1M single-crystals. The above relation is predicted to hold for geometric home-octahedral sheets (having equal M1 and M2 site bond lengths) and not to hold for geometric meso-octahedral sheets (having unequal M1 and M2 site bond lengths). The counter-rotation of the M2 site of 1M single-crystals exactly (within experimental error) follows the geometric meso-octahedral sheet model, which, assuming a uniform octahedral sheet height and site-specific M1 and M2 bond lengths, predicts site-specific flattening angles and a counter-rotation angle for the M2 site which is uniquely determined by the bond length difference between the M1 and M2 sites. A geometric meso-octahedral 2:1 layer silicate was shown to require corrugated tetrahedral sheets composed of bond-distorted tetrahedra. Key geometric meso-octahedral distortions in 1M single-crystals were identified and elucidated: (i) intra-layer top-bottom displacements within a TOT layer; and (ii) a tetrahedral bending angle between the apical bond and the pyramidal base formed by the three basal bonds. Plots of lattice parameter b versus average-octahedral-bond-length allowed the following distinction to be made: Unoxidized divalent synthetic solid solution series tend to evolve along constant flattening-angle lines whereas trivalent octahedral cation and vacancy bearing natural single-crystals and synthetic powders follow trends with varying flattening angles. We found that the bond length of a given interlayer cationic species monotonously increases as the tetrahedral rotation angle alpha decreases in trioctahedral-1 M single-crystals. An upper limit of tetrahedral rotation of alpha = 9.5° was demonstrated to occur in trioctahedral-1M K-rich micas having an <AlSi3> sheet, for both synthetic powders and natural single-crystals. Other attempts at applying geometric crystal chemical models and at identifying structure-chemical relationships from structural refinement data will benefit from the perspective of our more complete and systematic approach based on pursuing simple geometrical models using 'regular' coordination polyhedrons and characteristic cation-specific bond lengths up to the limit beyond which such models are shown to necessarily breakdown because of unavoidable 'non-regular' polyhedral distortions. (Abstract shortened by UMI.)

No description available.

Master of Science / Department of Physics / Nobel S. Rebello / Over the past 20 years there have been limited efforts to improve students’ interest and knowledge of electronics and to offer students experiences to integrate and apply their knowledge of electronics with experimental physics. None of the reform efforts cited in the literature have performed a careful assessment of student learning and attitudes, and most of them report anecdotal success. These programs share several commonalities. They typically have a capstone project experience in which students apply their knowledge and skills in electronics and instrumentation to a particular context. The KSU Physics Department has embarked on an endeavor to improve the PMI (Physical Measurement and Instrumentation) class taken by physics majors. Capstone project experiences for students in PMI will provide them with an opportunity to revisit experiments they completed in previous courses. They then apply the knowledge and skills in electronics and instrumentation learned at the beginning of the PMI course to automate these experiments. The use of LabVIEW and NI ELVIS provides a range of opportunities to students due to their visual interface and easy learning curve. However, they do have some disadvantages such as speed and resolution when compared to more traditional measurements with oscilloscopes. Three specific capstone experiences have been developed in PMI. These include saturated absorption in Rubidium, the Franck-Hertz experiment, and the speed of light measurement. In each case, students first complete the traditional experiments and then use NI ELVIS and LabVIEW to automate these experiments. Students are provided minimal explicit guidance in completing the capstone projects. These include one-page handouts describing the goals, basic procedures and questions that students have to answer for themselves. Comparing data from traditional experiments and those from automated using LabVIEW and NI ELVIS provides a context in which to discuss the trade-offs between the traditional and automated experiments. Future efforts include the development of more experiments as well as careful assessment of student learning and attitudes as a result of the capstone experiences in the PMI class. This project can potentially inform similar efforts at other institutions in the future.

<p> The stellar halo that surrounds our Galaxy contains clues to understanding galaxy formation, cosmology, stellar evolution, and the nature of dark matter. Gravitationally disrupted dwarf galaxies form tidal streams, which roughly trace orbits through the Galactic halo. The Sagittarius (Sgr) dwarf tidal debris is the most dominant of these streams, and its properties place important constraints on the distribution of mass (including dark matter) in the Galaxy. Stars not associated with substructures form the "smooth" component of the stellar halo, the origin of which is still under investigation. Characterizing halo substructures such as the Sgr stream and the smooth halo provides valuable information on the formation history and evolution of our galaxy, and places constraints on cosmological models. This thesis is primarily concerned with characterizing the 3-dimensional stellar densities of the Sgr tidal debris system and the smooth stellar halo, using data from the Sloan Digital Sky Survey (SDSS). F turnoff stars are used to infer distances, as they are relatively bright, numerous, and distributed about a single intrinsic brightness (magnitude). The inherent spread in brightnesses of these stars is overcome through the use of the recently-developed technique of statistical photometric parallax, in which the bulk properties of a stellar population are used to create a probability distribution for a given star's distance. This was used to build a spatial density model for the smooth stellar halo and tidal streams. The free parameters in this model are then fit to SDSS data with a maximum likelihood technique, and the parameters are optimized by advanced computational methods. Several computing platforms are used in this study, including the RPI SUR Bluegene and the Milkyway@home volunteer computing project. Fits to the Sgr stream in 18 SDSS data stripes were performed, and a continuous density profile is found for the major Sgr stream. The stellar halo is found to be strongly oblate (flattening parameter q=0.53). A catalog of stars consistent with this density profile is produced as a template for matching future disruption models. The results of this analysis favor a description of the Sgr debris system that includes more than one dwarf galaxy progenitor, with the major streams above and below the Galactic disk being separate substructures. Preliminary results for the minor tidal stream characterizations are presented and discussed. Additionally, a more robust characterization of halo turnoff star brightnesses is performed, and it is found that increasing color errors with distance result in a previously unaccounted for incompleteness in star counts as the SDSS magnitude limit is approached. These corrections are currently in the process of being implemented on MilkyWay@home.</p>

<p> Terahertz (THz) heterodyne receiver systems are required by NASA to monitor gas concentrations related to the Earth's ozone depletion. To this end, NASA needs compact, solid state, tunable THz local oscillators. THz LOs have been developed using three means: 1) All-electronic LOs using mixers in combination with Gunn oscillators, 2) Hybrid Photo-electronic LOs using a cw analog of the Auston switch, and 3) All-photonic THz LOs using coherent sources, such as vapor lasers or solid-state Quantum Cascade Lasers, and down converting lasers using nonlinear crystals. In this dissertation, we began with two frequency stabilized Nd:YAG lasers, locked to a common reference cavity, as a starting point to having a stable input into a nonlinear optical frequency conversion system. Following this, we explored the nonlinear crystals useful for THz generation, and the phasematching schemes that could be employed by each. We concluded by settling on highly insulating III-V semiconductor crystals as the proper choice of nonlinear element, and put together a new phasematching method that is most useful for them.</p>

A fundamental quantum-mechanical problem on the phase of quantum harmonic oscillators, which has remained an enigma for more than sixty years since the first treatment by Dirac, is completely solved. Contrary to the common belief that no Hermitian phase operators can be found to describe the phase properties of a quantum harmonic oscillator, a well-defined Hermitian phase operator with an appropriate classical limit is constructed unambiguously. The approach is different in nature from those of many previous attempts which were more or less based on the idea of polar decomposition of the annihilation operator. The fundamental difference between the quantum phase and the classical phase in spite of their conceptual consistency is pointed out and explained. The eigenvalue spectrum and eigenstates of the phase operator are obtained. Some important properties of the phase operator and phase states are investigated. / The rest of this research is devoted to the studies of multimode Gaussian squeezed states of quantum harmonic oscillators. Multimode squeeze operators and rotation operators are defined such that they have extremely similar algebraic properties as those of their single-mode counterparts. It is shown that the introduction of N-mode squeeze operators provides a convenient set of parameters to describe squeezing in multimode Gaussian squeezed states. The disentangling, normal ordering, and some other properties of N-mode squeeze operators are investigated. It is also shown that the time-evolution operator for a general N-mode quadratic Hamiltonian can be conveniently expressed as an operator product containing an N-mode squeeze operator, an N-mode rotation operator, and an N-mode displacement operator. As an application of this result, the dynamics of N-mode harmonic oscillators with time-dependent normal coordinates and frequencies is investigated and formulated by the use of these N-mode unitary operators. A general expression for the time-dependent transition amplitudes between two arbitrary N-mode boson number states is obtained and explicit results are given for the case of sudden changes in normal coordinates and frequencies. / Source: Dissertation Abstracts International, Volume: 50-12, Section: B, page: 5703. / Major Professor: William Clifford Rhodes. / Thesis (Ph.D.)--The Florida State University, 1989.

The anatase form of titanium dioxide has been shown to be of technological importance. Little is known of the surface structure and the electronic structure of the anatase polymorph due to the unavailability of suitable samples. Recent studies within our group have shown that the mineral form of anatase may be amenable to surface studies. This dissertation focuses on the study of the surface and electronic properties of the anatase polymorph of titanium dioxide. Presented here are surface studies on the most stable faces of anatase, chemical surface interactions, and electronic studies for the mineral anatase. Surface studies were carried out on the (101), (100), (103) and (001) faces of mineral anatase. Geometrical models of these surfaces were developed through the use of surface science techniques including Scanning Tunneling Microscopy (STM), Low Energy Electron Diffraction (LEED), and Low-Energy Ion Scattering (LEIS). Chemical studies (Thermal Programmed Desorption (TPD) and X-ray Photoelectron Spectroscopy (XPS)) were undertaken to study the interactions of the anatase (101) surface with water and methanol. Transport measurements were conducted to determine the number of charge carriers in the material. Investigations were held to probe the electronic states of the material through the use of x-ray Absorption Spectroscopy (XAS) / acase@tulane.edu