Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 11
Encontrada(s) 1 página(s)

BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Enhancing Sampling in Computational Statistics Using Modified Hamiltonians

  • Radivojevic, T.
The Hamiltonian Monte Carlo (HMC) method has been recognized as a powerful sampling tool in computational statistics. In this thesis, we show that performance of HMC can be dramatically improved by replacing Hamiltonians in the Metropolis test with modified Hamiltonians, and a complete momentum update with a partial momentum refreshment. The resulting generalized HMC importance sampler, which we called Mix & Match Hamiltonian Monte Carlo (MMHMC), arose as an extension of the Generalized Shadow Hybrid Monte Carlo (GSHMC) method, previously proposed for molecular simulation. The MMHMC method adapts GSHMC specifically to computational statistics and enriches it with new essential features: (i) the efficient algorithms for computation of modified Hamiltonians; (ii) the implicit momentum update procedure and (iii) the two-stage splitting integration schemes specially derived for the methods sampling with modified Hamiltonians. In addition, different optional strategies for momentum update and flipping are introduced as well as algorithms for adaptive tuning of parameters and efficient sampling of multimodal distributions are developed. MMHMC has been implemented in the in-house software package HaiCS (Hamiltonians in Computational Statistics) written in C, tested on the popular statistical models and compared in sampling efficiency with HMC, Generalized Hybrid Monte Carlo, Riemann Manifold Hamiltonian Monte Carlo, Metropolis Adjusted Langevin Algorithm and Random Walk Metropolis-Hastings. The analysis of time-normalized effective sample size reveals the superiority of MMHMC over popular sampling techniques, especially in solving high-dimensional problems., FPU12/05209, MTM2013–46553–C3–1–P
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

On the controllability of Partial Differential Equations involving non-local terms and singular potentials

  • Biccari, U.
In this thesis, we investigate controllability and observability properties of Partial Differential Equations describing various phenomena appearing in several fields of the applied sciences such as elasticity theory, ecology, anomalous transport and diffusion, material science, porous media flow and quantum mechanics. In particular, we focus on evolution Partial Differential Equations with non-local and singular terms. Concerning non-local problems, we analyse the interior controllability of a Schr\"odinger and a wave-type equation in which the Laplace operator is replaced by the fractional Laplacian $(-\Delta)^s$. Under appropriate assumptions on the order $s$ of the fractional Laplace operator involved, we prove the exact null controllability of both equations, employing a $L^2$ control supported in a neighbourhood $\omega$ of the boundary of a bounded $C^{1,1}$ domain $\Omega\subset\mathbb{R}^N$. More precisely, we show that both the Schrodinger and the wave equation are null-controllable, for $s\geq 1/2$ and for $s\geq 1$ respectively. Furthermore, these exponents are sharp and controllability fails for $s<1/2$ (resp. $s<1$) for the Schrödinger (resp. wave) equation. Our proof is based on multiplier techniques and the very classical Hilbert Uniqueness Method. For models involving singular terms, we firstly address the boundary controllability problem for a one-dimensional heat equation with the singular inverse-square potential $V(x):=\mu/x^2$, whose singularity is localised at one extreme of the space interval $(0,1)$ in which the PDE is defined. For all $0<\mu<1/4$, we obtain the null controllability of the equation, acting with a $L^2$ control located at $x=0$, which is both a boundary point and the pole of the potential. This result follows from analogous ones presented in \cite{gueye2014exact} for parabolic equations with variable degenerate coefficients. Finally, we study the interior controllability of a heat equation with the singular inverse-square potential $\Lambda(x):=\mu/\delta^2$, involving the distance $\delta$ to the boundary of a bounded and $C^2$ domain $\Omega\subset\mathbb{R}^N$, $N\geq 3$. For all $\mu\leq 1/4$ (the critical Hardy constant associated to the potential $\Lambda$), we obtain the null controllability employing a $L^2$ control supported in an open subset $\omega\subset\Omega$. Moreover, we show that the upper bound $\mu=1/4$ is sharp. Our proof relies on a new Carleman estimate, obtained employing a weight properly designed for compensating the singularities of the potential.
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Goal-Oriented Adaptivity using Unconventional Error Representations

  • Darrigrand, V.
In Goal-Oriented Adaptivity (GOA), the error in a Quantity of Interest (QoI) is represented using global error functions of the direct and adjoint problems. This error representation is subsequently bounded above by element-wise error indicators that are used to drive optimal refinements. In this work, we propose to replace, in the error representation, the adjoint problem by an alternative operator. The main advantage of the proposed approach is that, when judiciously selecting such alternative operator, the corresponding upper bound of the error representation becomes sharper, leading to a more efficient GOA. These representations can be employed to design novel h, p, and hp energy-norm and goal-oriented adaptive algorithms. While the method can be applied to a variety of problems, in this Dissertation we first focus on one-dimensional (1D) problems, including Helmholtz and steady-state convection-dominated diffusion problems. Numerical results in 1D show that for the Helmholtz problem, it is advantageous to select the Laplace operator for the alternative error representation. Specifically, the upper bounds of the new error representation are sharper than the classical ones used in both energy-norm and goal-oriented adaptive methods, especially when the dispersion (pollution) error is significant. The 1D steady-state convection-dominated diffusion problem with homogeneous Dirichlet boundary conditions exhibits a boundary layer that produces a loss of numerical stability. The new error representation based on the Laplace operator delivers sharper error upper bounds. When applied to a p-GOA, the alternative error representation captures earlier the boundary layer, despite the existing spurious numerical oscillations. We then focus on the two- and three-dimensional (2D and 3D) Helmholtz equation. We show via extensive numerical experimentation that the upper bounds provided by the alternative error representations are sharper than the classical ones. When using the alternative error indicators, a naive p-adaptive process converges, whereas under the same conditions, the classical method fails and requires the use of the so-called Projection Based Interpolation (PBI) operator or some other technique to regain convergence. We also provide guidelines for finding operators delivering sharp error representation upper bounds., Basque Government Consolidated Research Group Grant IT649-13
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Hierarchical modelling of patient-reported outcomes data based on the beta-binomial distribution

  • Najera-Zuloaga, J.
The beta-binomial distribution does not belong to the exponential family and, hence classical regression techniques cannot be used when dealing with outcomes following the mentioned distribution. In this thesis we propose and develop regression models based on the beta-binomial distribution for the analysis of U, J or inverse J-shaped discrete and bounded outcomes. In fact, although this thesis is focused on the analysis of patient-reported outcomes (PROs), which usually follow the mentioned distributional shapes, proposed models can also be extended to several fields. First of all, we make a review and comparison of existing beta-binomial regression approaches in independent data context, concluding that the marginal approach is the most adequate. However, PRO studies are usually carried out in a longitudinal framework, where patients' responses are measured over time. This leads to a multilevel or correlated data structure and consequently, we extend the marginal beta-binomial regression approach to the inclusion of random effects to accommodate the hierarchical structure of the data. We develop the estimation and inference procedure for the model proposal. Furthermore, we compare the performance of our proposal with similar approaches in the literature, showing that it gets better results in terms of reducing the bias of the estimates. We apply the model to a longitudinal Chronic Obstructive Pulmonary Disease study carried out at Galdakao Hospital, reaching clinically and statistically relevant results about the evolution of the patients over time. PROs are usually obtained using rating scale questionnaires consisting of questions or items, grouped into one or more subscales, often called dimensions or domains. Therefore, we also propose a multivariate regression model based on the beta-binomial distribution for the joint analysis of all the longitudinal dimensions provided by different questionnaires. Finally, it is worth mentioning that we have implemented all the proposed regression models in the PROreg R- package which is freely available at CRAN., Department of Education, Language Policy and Culture of the Basque Government IT-620-13 program, MTM2013-40941-P and MTM2016-74931-P
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Singular Perturbation of the Dirac Hamiltonian

  • Pizzichillo, F.
This thesis is devoted to the study of the Dirac Hamiltonian perturbed by delta-type potentials and Coulomb-type potentials. We analysed the delta-shell interaction on bounded and smooth domains and its approximation by the coupling of the free Dirac operator with shrinking short-range potentials. Under certain hypothesis of smallness of a regular potential, in three dimensions, we prove that the Dirac operator coupled with a suitable rescaling of the potential converges to the Hamiltonian coupled with the delta-shell potential, in the strong resolvent sense. Nevertheless, the coupling constant depends non-linearly on the potential: Klein's Paradox comes into play. As a particular case, we pay to focus on the Dirac Operator with spherical delta-shell interactions. We characterise the eigenstates of the couplings by finding sharp constants and minimisers of some precise inequalities related to an uncertainty principle. We also explore the spectral relation between the shell interaction and its approximation by short-range potentials with shrinking support, improving previous results. Finally, we investigate the Dirac operator perturbed by a particular class of Coulomb-type spherically symmetric potentials. We describe the self-adjoint realisations of this operator in terms of the behaviour of the functions of the domain in the origin, and we provide Hardy-type estimates on them. At the end, we describe the distinguished extension.
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Adapting Hybrid Monte Carlo methods for solving complex problems in life and materials sciences

  • Fernández-Pendás, M.
Efficient sampling is the key to success of molecular simulation of complex physical systems. Still, a unique recipe for achieving this goal is unavailable. Hybrid Monte Carlo (HMC) is a promising sampling tool offering a smart, free of discretization errors, propagation in phase space, rigorous temperature control, and flexibility. However, its inability to provide dynamical information and its weakness in simulations of reasonably large systems do not allow HMC to become a sampler of choice in molecular simulation of complex systems. In this thesis, we show that performance of HMC can be dramatically improved by introducing in the method the splitting numerical integrators and importance sampling. We propose a novel splitting integration scheme called Adaptive Integration Approach or AIA, which leads to very promising improvements in accuracy and sampling in HMC simulations. Given a simulation problem and a time step, AIA automatically chooses the optimal scheme out of the family of two-stage splitting integrators. A system-specific integrator identified by our approach is optimal in the sense that it provides the best conservation of energy for harmonic forces. The role of importance sampling on the performance of HMC is studied through the modified Hamiltonian Monte Carlo (MHMC) methods, sampling with respect to a modified or shadow Hamiltonian. The particular attention is paid to Generalized Shadow Hybrid Monte Carlo (GSHMC), introduced by Akhmatskaya and Reich in 2008. To improve the performance of MHMC in general and GSHMC in particular, we develop and test the new multi-stage splitting integrators, specially formulated for sampling with respect to modified Hamiltonians. The novel adaptive two-stage integration approach or MAIA, specifically derived for MHMC is presented. We also discuss in detail the adaptation of GSHMC to the NPT ensemble and provide the thorough analysis of its performance. Moreover, for the first time, we formulate GSHMC in the grand canonical ensemble. A general framework, useful for an extension of other Hybrid Monte Carlo methods to the grand canonical ensemble, is also provided. The software development is another fundamental part of the present work. The algorithms presented in this thesis are implemented in MultiHMC-GROMACS, an in-house version of the popular software package GROMACS. We explain the details of such implementation and give useful recommendations and hints for implementation of the new algorithms in other software packages. In summary, in this thesis, we propose the new numerical algorithms that are capable of improving the accuracy and sampling efficiency of molecular simulations with Hybrid Monte Carlo methods. We show that equipping the Hybrid Monte Carlo algorithm with extra features makes it even a “smarter” sampler and, no doubts, a strong competitor to the well-established molecular simulation techniques such as molecular dynamics (MD) and Monte Carlo. The up to 60 times increase in sampling efficiency of GSHMC over MD, due to the new algorithms in simulations of selected systems, supports such a belief., MTM2013-46553-C3-1-P
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Probabilistic Modelling of Classical and Quantum Systems

  • Rusconi, S.
While probabilistic modelling has been widely used in the last decades, the quantitative prediction in stochastic modelling of real physical problems remains a great challenge and requires sophisticated mathematical models and advanced numerical algorithms. In this study, we developed the mathematical tools for solving three long-standing problems in Polymer Science and Quantum Measurement theory. The question, “Why kinetic models cannot reproduce experimental observations in Controlled Radical Polymerization (CRP)?” has been answered by introducing in the kinetic model a delay and treating CRP as a non-Markovian process. The efficient stochastic simulation (SS) approach allowing for an accurate description of CRP has been formulated, theoretically grounded and tested using experimental data and the less advanced SS algorithms. An accurate prediction of a morphology development in multi-phase polymers is vital for synthesis of new materials but still not feasible due to its complexity. We proposed a Population Balance Equations (PBE)-based model and derived a conceptually new and computationally tractable numerical approach for its solution in order to provide a systematic tool for a morphology prediction in composite polymers. Finally, we designed a stochastic simulation framework for continuous measurements performed on quantum systems of theoretical and experimental interest, which helped us to re-examine the “fuzzy continuous measurements” theory by Audretsch and Mensky (1997) and expose some of its deficiencies, while making amendments where necessary. All developed modelling approaches are general enough to be applied to the broad range of physical applications and thus ultimately to contribute to the understanding and prediction of complex chemical and physical processes., BES-2014-06864, MTM2013-46553-C3-1-P
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

REFINED ISOGEOMETRIC ANALYSIS: A SOLVER-BASED DISCRETIZATION METHOD

  • Garcia, D.
Isogeometric analysis (IGA) is a computational approach frequently employed nowadays to study problems governed by partial differential equations (PDEs). This approach defines the geometry using conventional computer-aided design (CAD) functions and, in particular, NURBS. These functions represent complex geometries commonly found in engineering design and are capable of preserving exactly the geometry description under refinement as required in the analysis. Moreover, the use of NURBS as basis functions is compatible with the isoparametric concept, allowing to build algebraic systems directly from the computational domain representation based on spline functions, which arise from CAD. Therefore, it avoids to define a second space for the numerical analysis resulting in huge reductions in the total analysis time. To perform the numerical analysis, we can use either direct or iterative solvers. Direct solvers are preferred to study stiff linear problems that are not solvable with iterative solvers, as well as another type of problems, e.g., problems with multiple right-hand side (RHS). Moreover, when the problem size and the floating point operations (FLOPs) required to solve the problem are enormous, and the direct solvers become excessively expensive, the iterative solver turns into a more suitable alternative. For the case of direct solvers, the performance strongly depends upon the employed discretization method. In particular, on IGA, the continuity of the solution spaces plays a significant role in their performance. High continuous spaces degrade the direct solver's performance, increasing the solution times by a factor up to $\mathcal{O}(p^3)$ with respect to traditional finite element analysis (FEA) per unknown, being $p$ the polynomial order. In this work, we propose a solver-based discretization that employs highly continuous finite element spaces interconnected with low continuity hyperplanes to maximize the performance of direct solvers. Starting from a highly continuous (IGA) discretization, we introduce $C^0$ hyperplanes, which act as separators for the direct solver, to reduce the interconnection between the degrees of freedom (DoF) in the mesh. By doing so, both the solution time and best approximation errors are simultaneously improved. We call the resulting method ``refined Isogeometric Analysis (rIGA)". While this method can be applied to a variety of problems, in this Dissertation we focus on analyzing the impact of the continuity reduction when solving a Laplace problem with structured meshes and uniform polynomial orders in both 2D and 3D. Numerical results indicate that rIGA delivers speed-up factors proportional to $p^2$. For instance, in a 2D mesh with four million elements and $p=5$, the linear system resulting from rIGA is solved $22$ times faster than the one from highly continuous IGA. In a 3D mesh with one million elements and $p=3$, the linear rIGA system is solved $15$ times faster than the IGA one. We then develop a version of the rIGA strategy that introduces hyperplanes of arbitrary continuity ($C^k$ hyperplanes with $0\leq k\leq p-1$). This strategy, called Optimally refined Isogeometric Analysis (OrIGA), leads to more efficient discretization than those obtained with the original version of rIGA. By using separators of arbitrary continuity degree, we achieve a performance boost of up to 25$\%$ in the direct solvers with respect to rIGA. Thus, the savings with respect to IGA are larger than in rIGA. We have also designed and implemented a similar rIGA strategy for iterative solvers. This strategy splits the mesh into subdomains using $C^0$-hyperplanes and constructs the Schur complements of the subdomains (macro-elements) using a direct solver. We then assemble those Schur complements into a global skeleton system, which is only composed of the DoF located along the boundaries of all the subdomains. Subsequently, we solve this system iteratively using conjugate gradient method (CG) with an incomplete LU (ILU) preconditioner. Lastly, a backward substitution is performed to recover the eliminated DoF and obtain the solution of the original system. Thus, rIGA for iterative solvers is a hybrid solver strategy that combines a direct solver (static condensation step) to eliminate the internal macro-elements DoF, with an iterative method to solve the skeleton system. The hybrid solver strategy achieves moderate savings with respect to IGA when solving a 2D Poisson problem with a structured mesh and a uniform polynomial degree of approximation. For instance, for a mesh with four million elements and polynomial degree $p=3$, the iterative solver is approximately $2.6$ times faster (in time) when applied to the rIGA system than to the IGA one. These savings occur because the skeleton rIGA system contains fewer non-zero entries than the IGA one. The opposite situation occurs for 3D problems, and as a result, 3D rIGA discretizations provide no gains with respect to their IGA counterparts. In this work, we also apply rIGA to solve incompressible fluid flow problems on an enclosed domain. To satisfy the inf-sup stability condition and guarantee divergence-free discrete solutions, the implemented rIGA employs a combination of $C^0$ and $C^1$ hyperplanes to reduce the continuity on the solution spaces. Numerical results show that the $L^2$ norm of the discretization error improves as we reduce the continuity. Therefore, rIGA delivers smaller errors than $C^{p-1}$ IGA discretizations. In terms of the computational savings, rIGA provides a reduction in the computational cost of the direct solvers by a factor of $\mathcal{O}(p^2)$ in both 2D and 3D., Texas Advanced Computing Center (TACC) at The University of Texas at Austin which provides the HPC resources.
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Some classes of homeomorphisms that preserve multiplicity and tangent cones

  • Sampaio, J.E.
In this paper it is presented some classes of homeomorphisms that preserve multiplicity and tangent cones of complex analytic sets. Moreover, we present a class of homeomorphisms that has the multiplicity as an invariant when we consider right equivalence and this class contains many known classes of homeomorphisms that preserve tangent cones. Finally, we show versions of these results looking at infinity.
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

Fast One-dimensional Finite Element Approximation of Geophysical Measurements

  • Shahriari, M.
There exist a wide variety of geophysical prospection methods. In this work, we focus on resistivity methods. We categorize these resistivity prospection methods according to their acquisition location as (a) on the surface, such as the ones obtained using Controlled Source Electromagnetics (CSEM) and magnetotelluric, and (b) in the borehole, such as the ones obtained using Logging-While-Drilling (LWD) devices. LWD devices are useful both for reservoir characterization and geosteering purposes, which is the act of adjusting the tool direction to travel within a specific zone. When inverting LWD resistivity measurements, it is a common practice to consider a one-dimensional (1D) layered media to reduce the problem dimensionality using a Hankel transform. Using orthogonality of Bessel functions, we arrive at a system of Ordinary Differential Equations (ODEs); one system of ODEs per Hankel mode. The dimensionality of the resulting problem is referred to as 1.5D since the computational cost to resolve it is in between that needed to solve a 1D problem and a 2D problem. When material properties (namely, resistivity, permittivity, and magnetic permeability) are piecewise-constant, we can solve the resulting ODEs either (a) analytically, which leads to a so-called semi-analytic method after performing a numerical inverse Hankel transform or (b) numerically. Semi-analytic methods are faster, but they also have important limitations, for example, (a) the analytical solution can only account for piecewise constant material properties, and other resistivity distributions cannot be solved analytically, which prevents to accurately model, for example, an Oil-Water Transition (OWT) zone when fluids are considered to be immiscible; (b) a specific set of cumbersome formulas has to be derived for each physical process (e.g., electromagnetism, elasticity, etc.), anisotropy type, etc.; (c) analytical derivatives of specific models (e.g., cross-bedded formations, or derivatives with respect to the bed boundary positions) are often difficult to obtain and have not been published to the best of our knowledge. In view of the above limitations, we propose to solve our forward problems using a numerical solver. A traditional Finite Element Method (FEM) is slow, which makes it unfeasible for our application. To achieve high performance, we developed a multiscale FEM that pre-computes a set of optimal local basis functions that are used at all logging positions. The resulting method is slow when compared to a semi-analytic approach for a single logging position, but it becomes highly competitive for a large number of logging positions, as needed for LWD geosteering applications. Moreover, we can compute the derivatives using an adjoint state method at almost zero additional cost in time. We describe an adjoint-based formulation for computing the derivatives of the electromagnetic fields with respect to the bed boundary positions. The key idea to obtain this adjoint-based formulation is to separate the tangential and normal components of the field, and treat them differently. We then apply this method to a 1.5D borehole resistivity problem. Moreover, we compute the adjoint-state formulation to compute the derivative of the magnetic field with respect to the resistivity value of each layer. We verify the accuracy of our formulations via synthetic examples. When simulating borehole resistivity measurements in a reservoir, it is common to consider an Oil-Water Contact (OWC) planar interface. However, this consideration can lead to an unrealistic model since, in the presence of capillary pressure, the mix of two immiscible fluids (oil and water) often appears as an OWT zone. These transition zones may be large in the vertical direction (20 meters or above), and in context of geosteering, an efficient method to simulate an OWT zone can maximize the production of an oil reservoir. In this work, we prove that by using our proposed 1.5D numerical method, we can easily consider arbitrary resistivity distributions in the vertical direction, as it occurs in an OWT zone. Numerical results on synthetic examples demonstrate significant differences between the results recorded by a geosteering device when considering a realistic OWT zone vs. an OWC sharp interface. As an additional piece of work of this Ph.D. Dissertation, we explore the possibility of using a Deep Neural Network (DNN) to perform a rapid inversion of borehole resistivity measurements. Herein, we build a DNN that approximates the following inverse problem: given a set of borehole resistivity measurements, the DNN is designed to deliver a physically meaningful and data-consistent piecewise one-dimensional layered model of the surrounding subsurface. Once the DNN is built, the actual inversion of the field measurements is efficiently performed in real time. We illustrate the performance of a DNN designed to invert LWD measurements acquired on high-angle wells via synthetic examples.
BIRD. BCAM's Institutional Repository Data
Publicación Tesis doctoral (DoctoralThesis)

On Zariski’s multiplicity problem at infinity

  • Sampaio, J.E.
We address a metric version of Zariski's multiplicity conjecture at infinity that says that two complex algebraic affine sets which are bi-Lipschitz homeomorphic at infinity must have the same degree. More specifically, we prove that the degree is a bi-Lipschitz invariant at infinity when the bi-Lipschitz homeomorphism has Lipschitz constants close to 1. In particular, we have that a family of complex algebraic sets bi-Lipschitz equisingular at infinity has constant degree. Moreover, we prove that if two polynomials are weakly rugose equivalent at infinity, then they have the same degree. In particular, we obtain that if two polynomials are rugose equivalent at infinity or bi-Lipschitz contact equivalent at infinity or bi-Lipschitz right-left equivalent at infinity, then they have the same degree.