Global ETD Search

11	Enhanced Conformational Sampling of Proteins Using TEE-REX / Verbessertes Sampling von Proteinkonformationen durch TEE-REX Kubitzki, Marcus 11 December 2007 (has links) No description available. 530 Physik WC 000 Mathematics and Natural Science Molekulardynamik Proteindynamik Computersimulation Simulationsmethodik Hauptkomponentenanalyse molecular dynamics protein dynamics simulation methodology replica exchange essential dynamics principal component analysis conformational sampling 42.12 33.10 33.06 33.25
12	Study of organic matter decomposition under geological conditions from replica exchange molecular dynamics simulations / Etude de la décomposition de matière organique dans des conditions géologiques par simulations numériques de replica exchange molecular dynamics Atmani, Léa 15 May 2017 (has links) Pétrole et gaz proviennent de la décomposition de la matière organique dans la croûte terrestre. En s’enfouissant, les résidus organiques se décomposent en un solide poreux et carboné, appelé kérogène et en un fluide composé d’hydrocarbures et de petites molécules telles que de l’eau. Le processus de formation du kérogène n’est pas totalement élucidé et une modélisation aiderait à une meilleure compréhension à la fois de sa structure et de sa composition et serait utile à l’industrie pétrolière.Dans le présent travail, nous adoptons une approche thermodynamique ayant pour but, à l’aide de simulations numériques, de d’étudier la décomposition de précurseurs de kérogène d’un type donné –ici le type III- dans les conditions d’un réservoir géologique. La méthode dite de Replica Exchange Molecular Dynamics (REMD) est appliquée pour étudier la décomposition de cristaux de cellulose et de lignine. Le potentiel d’interaction ReaxFF et le code LAMMPS sont utilisés. La REMD est une façon de surmonter de larges barrières d’énergie libre, en améliorant l’échantillonnage de configurations d’une dynamique moléculaire conventionnelle à température constante, en utilisant des états générés à températures supérieures.En fin de simulation, les systèmes ont atteint un état d’équilibre entre deux phases : une phase riche en carbone, composée d’amas de macromolécules, que nous appelons « solide » et d’une phase riche en oxygène et en hydrogène, composée de petites molécules, que nous dénommons « fluide ». L’évolution des parties solides de nos systèmes coïncide avec celle d’échantillons naturels de kérogènes de type III. / In deep underground, organic residues decompose into a carbonaceous porous solid, called kerogen and a fluid usually composed of hydrocarbons and other small molecules such as water, carbon monoxide. The formation process of the kerogen remains poorly understood. Modeling its geological maturation could widen the understanding of both structure and composition of kerogen, and could be useful to oil and gas industry.In this work we adopt a purely thermodynamic approach in which we aim, through molecular simulations, at determining the thermodynamic equilibrium corresponding to the decomposition of given organic precursors of a specific type of kerogen –namely type III- under reservoir conditions. Starting from cellulose and lignin crystal structures we use replica exchange molecular dynamics (REMD) simulations, using the reactive force field ReaxFF and the open-source code LAMMPS. The REMD method is a way ofovercoming large free energy barriers, by enhancing the configurational sampling of a conventional constant temperature MD using states from higher temperatures.At the end of the simulations, we have reached for both systems, a stage where they can clearly be cast into two phases: a carbon-rich phase made of large molecular clusters that we call here the "solid" phase, and a oxygen and hydrogen rich phase made of small molecules that we call "fluid" phase.The evolution of solid parts for both systems and the natural evolution of a type III kerogen clearly match. Evolution of our systems follows the one of natural samples, as well as the one of a type III kerogen submitted to an experimental confined pyrolysis. Kérogène Remd Simulations Gaz Pétrole Gaz de schistes Matière organique Lammps ReaxFF Cellulose Lignine Van Krevelen Hydrocarbures Simulations Numériques Carbone Potentiel réactif Carbone amorphe Thermodynamique Géologie Roche mère Maturation Dynamique Moléculaire Pyrolyse Kerogen Remd Replica Exchange Molecular Dynamics Gas Shale Gas Oil Organic Matter Lammps ReaxFF Cellulose Lignin Van Krevelen Pyrolysis Hydrocarbons Numerical simulations Carbon Reactive Potential Amorphous Carbon Thermodynamics Petroleum Geology Source rocks Maturation Molecular Dynamics 530
13	Non-convex Bayesian Learning via Stochastic Gradient Markov Chain Monte Carlo Wei Deng (11804435) 18 December 2021 (has links) <div>The rise of artificial intelligence (AI) hinges on the efficient training of modern deep neural networks (DNNs) for non-convex optimization and uncertainty quantification, which boils down to a non-convex Bayesian learning problem. A standard tool to handle the problem is Langevin Monte Carlo, which proposes to approximate the posterior distribution with theoretical guarantees. However, non-convex Bayesian learning in real big data applications can be arbitrarily slow and often fails to capture the uncertainty or informative modes given a limited time. As a result, advanced techniques are still required.</div><div><br></div><div>In this thesis, we start with the replica exchange Langevin Monte Carlo (also known as parallel tempering), which is a Markov jump process that proposes appropriate swaps between exploration and exploitation to achieve accelerations. However, the na\"ive extension of swaps to big data problems leads to a large bias, and the bias-corrected swaps are required. Such a mechanism leads to few effective swaps and insignificant accelerations. To alleviate this issue, we first propose a control variates method to reduce the variance of noisy energy estimators and show a potential to accelerate the exponential convergence. We also present the population-chain replica exchange and propose a generalized deterministic even-odd scheme to track the non-reversibility and obtain an optimal round trip rate. Further approximations are conducted based on stochastic gradient descents, which yield a user-friendly nature for large-scale uncertainty approximation tasks without much tuning costs. </div><div><br></div><div>In the second part of the thesis, we study scalable dynamic importance sampling algorithms based on stochastic approximation. Traditional dynamic importance sampling algorithms have achieved successes in bioinformatics and statistical physics, however, the lack of scalability has greatly limited their extensions to big data applications. To handle this scalability issue, we resolve the vanishing gradient problem and propose two dynamic importance sampling algorithms based on stochastic gradient Langevin dynamics. Theoretically, we establish the stability condition for the underlying ordinary differential equation (ODE) system and guarantee the asymptotic convergence of the latent variable to the desired fixed point. Interestingly, such a result still holds given non-convex energy landscapes. In addition, we also propose a pleasingly parallel version of such algorithms with interacting latent variables. We show that the interacting algorithm can be theoretically more efficient than the single-chain alternative with an equivalent computational budget.</div> Statistics Stochastic Analysis and Modelling Monte Carlo Algorithm Artificial intelligence Importance sampling Computer vision Langevin Dynamics Variance reduction techniques Wang-Landau algorithm Interacting particles Hamiltonian Monte Carlo Log-Sobolev inequality Metropolis Hasting Deep neural network Stochastic variance-reduced gradient Wasserstein distance Convolutional neural network Deterministic even odd scheme Non-reversibility Stochastic approximation Monte Carlo Stochastic differential equation Stochastic gradient descent Parallel tempering Stochastic approximation Replica exchange Stochastic gradient Langevin dynamics Markov Chain Monte Carlo

Page generated in 0.0429 seconds