Learning with noise and regularizers in multilayer neural networks

We study the effect of two types of noise, data noise and model noise, in an on-line gradient-descent learning scenario for general two-layer student network with an arbitrary number of hidden units. Training examples are randomly drawn input vectors labeled by a two-layer teacher network with an arbitrary number of hidden units. Data is then corrupted by Gaussian noise affecting either the output or the model itself. We examine the effect of both types of noise on the evolution of order parameters and the generalization error in various phases of the learning process.

Learning with regularizers in multilayer neural networks

We study the effect of regularization in an on-line gradient-descent learning scenario for a general two-layer student network with an arbitrary number of hidden units. Training examples are randomly drawn input vectors labelled by a two-layer teacher network with an arbitrary number of hidden units which may be corrupted by Gaussian output noise. We examine the effect of weight decay regularization on the dynamical evolution of the order parameters and generalization error in various phases of the learning process, in both noiseless and noisy scenarios.

Experimental investigation into amplitude-dependent modal properties of an eleven-span motorway bridge

The authors would like to thank their supporters. New Zealand Earthquake Commission (EQC) Research Foundation provided financial support for experimental work (Grant No. UNI/578). New Zealand Transport Agency (NZTA) provided access to the bridge. Piotr Omenzetter’s work within the LRF Centre for Safety and Reliability Engineering at the University of Aberdeen is supported by Lloyd’s Register Foundation. The Foundation helps to protect life and property by supporting engineering-related education, public engagement and the application of research. Ge-Wei Chen’s doctoral study is supported by China Scholarship Council (CSC) (Grant No. 2011637065).

Quantum Otto heat engine based on a multiferroic chain working substance

We study a quantum Otto engine operating on the basis of a helical spin-1/2 multiferroic chain with strongly coupled magnetic and ferroelectric order parameters. The presence of a finite spin chirality in the working substance enables steering of the cycle by an external electric field that couples to the electric polarization. We observe a direct connection between the chirality, the entanglement and the efficiency of the engine. An electric-field dependent threshold temperature is identified, above which the pair correlations in the system, as quantified by the thermal entanglement, diminish. In contrast to the pair correlations, the collective many-body thermal entanglement is less sensitive to the electric field, and in the high temperature limit converges to a constant value. We also discuss the correlations between the threshold temperature of the pair entanglement, the spin chirality and the minimum of the fidelities in relation to the electric and magnetic fields. The efficiency of the quantum Otto cycle shows a saturation plateau with increasing electric field amplitude.

Preenchimento de cavidades provenientes de remoção de trinca em turbinas hidráulicas por meio do processo GMAW-P

Dissertação (mestrado)—Universidade de Brasília, Faculdade de Tecnologia, Departamento de Engenharia Mecânica, 2016.

A theory of the upper critical field in antiferromagnetic superconductors

A generalized Ginzburg-Landau approach is used to study the nonmonotonic temperature dependence of the upper critical field H c 2(T) in antiferromagnetic superconductors RE(Mo)6S8; RE = Dy, Tb, Gd. It is found that electrodynamic effects incorporated through screening and indirect coupling between the staggered magnetization M Q (T) and superconducting order parameter psgr cannot explain the observed nonmonotonicity. This suggests that the direct coupling between the two order parameters should be considered to understand the experimental results, a finding which is consistent with recent microscopic calculations.

Turbulence-induced melting of a nonequilibrium vortex crystal in a forced thin fluid film

To gain a better understanding of recent experiments on the turbulence-induced melting of a periodic array of vortices in a thin fluid film, we perform a direct numerical simulation of the two-dimensional Navier-Stokes equations forced such that, at low Reynolds numbers, the steady state of the film is a square lattice of vortices. We find that as we increase the Reynolds number, this lattice undergoes a series of nonequilibrium phase transitions, first to a crystal with a different reciprocal lattice and then to a sequence of crystals that oscillate in time. Initially, the temporal oscillations are periodic; this periodic behaviour becoming more and more complicated with increasing Reynolds number until the film enters a spatially disordered nonequilibrium statistical steady state that is turbulent. We study this sequence of transitions using fluid-dynamics measures, such as the Okubo-Weiss parameter that distinguishes between vortical and extensional regions in the flow, ideas from nonlinear dynamics, e.g. Poincare maps, and theoretical methods that have been developed to study the melting of an equilibrium crystal or the freezing of a liquid and that lead to a natural set of order parameters for the crystalline phases and spatial autocorrelation functions that characterize short- and long-range order in the turbulent and crystalline phases, respectively.

The Path to a Porous Semiconductor Multilayer

Thin films are the basis of much of recent technological advance, ranging from coatings with mechanical or optical benefits to platforms for nanoscale electronics. In the latter, semiconductors have been the norm ever since silicon became the main construction material for a multitude of electronical components. The array of characteristics of silicon-based systems can be widened by manipulating the structure of the thin films at the nanoscale - for instance, by making them porous. The different characteristics of different films can then to some extent be combined by simple superposition. Thin films can be manufactured using many different methods. One emerging field is cluster beam deposition, where aggregates of hundreds or thousands of atoms are deposited one by one to form a layer, the characteristics of which depend on the parameters of deposition. One critical parameter is deposition energy, which dictates how porous, if at all, the layer becomes. Other parameters, such as sputtering rate and aggregation conditions, have an effect on the size and consistency of the individual clusters. Understanding nanoscale processes, which cannot be observed experimentally, is fundamental to optimizing experimental techniques and inventing new possibilities for advances at this scale. Atomistic computer simulations offer a window to the world of nanometers and nanoseconds in a way unparalleled by the most accurate of microscopes. Transmission electron microscope image simulations can then bridge this gap by providing a tangible link between the simulated and the experimental. In this thesis, the entire process of cluster beam deposition is explored using molecular dynamics and image simulations. The process begins with the formation of the clusters, which is investigated for Si/Ge in an Ar atmosphere. The structure of the clusters is optimized to bring it as close to the experimental ideal as possible. Then, clusters are deposited, one by one, onto a substrate, until a sufficiently thick layer has been produced. Finally, the concept is expanded by further deposition with different parameters, resulting in multiple superimposed layers of different porosities. This work demonstrates how the aggregation of clusters is not entirely understood within the scope of the approximations used in the simulations; yet, it is also shown how the continued deposition of clusters with a varying deposition energy can lead to a novel kind of nanostructured thin film: a multielemental porous multilayer. According to theory, these new structures have characteristics that can be tailored for a variety of applications, with precision heretofore unseen in conventional multilayer manufacture.

Crossover dynamics at large metastability in gas-liquid nucleation

We have developed an alternate description of dynamics of nucleation in terms of an extended set of order parameters. The order parameters consist of an ordered set of kth largest clusters, ordered such that k = 1 is the largest cluster in the system, k = 2 is the second largest cluster, and so on. We have derived an analytic expression for the free energy for the kth largest cluster, which is in excellent agreement with the simulated results. At large supersaturation, the free energy barrier for the growth of the kth largest cluster disappears and the nucleation becomes barrierless. The major success of this extended theoretical formalism is that it can clearly explain the observed change in mechanism at large metastability P. Bhimalapuram et al., Phys. Rev. Lett. 98, 206104 (2007)] and the associated dynamical crossover. The classical nucleation theory cannot explain this crossover. The crossover from activated to barrierless nucleation is found to occur at a supersaturation where multiple clusters cross the critical size. We attribute the crossover as the onset of the kinetic spinodal. We have derived an expression for the rate of nucleation in the barrierless regime by modeling growth as diffusion on the free energy surface of the largest cluster. The model reproduces the slower increase in the rate of growth as a function of supersaturation, as observed in experiments.

C-13 MAS NMR spectra of cis,cis-mucononitrile in liquid-crystalline media and in the solid state. Orientation of the carbon chemical-shift tensors

The spinning sidebands observed in the C-13 MAS NMR spectra of cis,cis-mucononitrile oriented in liquid-crystalline media and of the neat sample in the solid state are studied. There are differences in the sideband intensity patterns in the two cases. These differences arise because the order parameters which characterize the orientation of the solute in the liquid-crystalline media differ for different axes. It is shown that, in general, the relative intensities of the sidebands contain information on the sign and magnitude of an effective chemical-shift parameter which is a function of the sum of the products of the principal components of the chemical-shift tensor and the corresponding order parameters with respect to the director. A method for obtaining the orientation of the carbon chemical-shift tensor is proposed. The carbon chemical-shift tensors obtained from gauge-including atomic orbital calculations are also presented for comparison. (C) 1996 Academic Press, Inc.

Monte Carlo study of the phase boundary of YBa2Cu3O6 +x

A model hamiltonian previously introduced to study the oxygen ordering is considered. The phase boundary is isolated by studying the relaxation behaviour of the order parameters. Our results are consistent with the published Monte Carlo results except at low temperatures.

Quinazoline�iodine interaction as studied by NMR in nematic solutions

The results of an NMR study of the interaction of quinazoline with iodine in the nematic phase indicate the formation of at least two different types of charge-transfer complexes. Significant changes in the molecular geometry of the quinazoline moiety were observed as a result of complexation with iodine. Detailed information on the formation of the charge-transfer complexes was derived from the changes in the molecular structure, order parameters and chemical shifts as functions of iodine concentration. The observed changes in the order parameters are interpreted in terms of bond interaction tensors.

Weak molecular complexes as studied by NMR spectroscopy of oriented molecules

Weak molecular interactions such as those in pyridine-iodine, benzene-iodine and benzene-chloroform systems oriented in thermotropic liquid crystals have been studied from the changes of the order parameters as a result of complex formation. The results indicate the formation of at least two types of charge transfer complexes in pyridine-iodine solutions. The pi-complexes in benzene-chloroform and benzene-iodine mixtures have also been detected. No detectable changes in the inter-proton distances in these systems were observed.

Cross-polarisation applied to the study of liquid crsytalline ordering

Cross polarisation is extensively used in solid state NMR for enhancing signals of nuclei with low gyromagnetic ratio. However, the use of the method for providing quantitative structural and dynamics information is limited. This arises due to the fact that the mechanism which is responsible for cross polarisation namely, the dipolar interaction, has a long range and is also anisotropic. In nematic liquid crystals these limitations are easily overcome since molecules orient in a magnetic field. The uniaxial ordering of the molecules essentially removes problems associated with the angular dependence of the interactions encountered in powdered solids. The molecular motion averages out intermolecular dipolar interaction, while retaining partially averaged intramolecular interaction. In this article the use of cross polarisation for obtaining heteronuclear dipolar couplings and hence the order parameters of liquid crystals is presented. Several modifications to the basic experiment were considered and their utility illustrated. A method for obtaining proton-proton dipolar couplings, by utilizing cross polarisation from the dipolar reservoir, is also presented.

The inhomogeneous extended Bose-Hubbard model: A mean-field theory

We develop an inhomogeneous mean-field theory for the extended Bose-Hubbard model with a quadratic, confining potential. In the absence of this potential, our mean-field theory yields the phase diagram of the homogeneous extended Bose-Hubbard model. This phase diagram shows a superfluid (SF) phase and lobes of Mott-insulator (MI), density-wave (DW), and supersolid (SS) phases in the plane of the chemical potential mu and on-site repulsion U; we present phase diagrams for representative values of V, the repulsive energy for bosons on nearest-neighbor sites. We demonstrate that, when the confining potential is present, superfluid and density-wave order parameters are nonuniform; in particular, we obtain, for a few representative values of parameters, spherical shells of SF, MI, DW, and SS phases. We explore the implications of our study for experiments on cold-atom dipolar condensates in optical lattices in a confining potential.