Calif., California Inst. of Technology, Diss.--Pasadena, 2003. / Kopie, ersch. im Verl. UMI, Ann Arbor, Mich.

Thesis (M.C.E.)--University of Delaware, 2008. / Principal faculty advisor: Chin-Pao Huang, Dept. of Civil and Environmental Engineering. Includes bibliographical references.

Techn. University, Diss., 2006--München.

Techn. University, Diss., 2002--München.

Thesis (M.S.)--Brigham Young University. Dept. of Computer Science, 2006. / Includes bibliographical references (p. 65-72).

Universiẗat, Diss., 2002--Marburg.

Universiẗat, Diss., 2005--Düsseldorf.

Transthyretin-related amyloidosis (ATTR) is a disease involving the formation of a misfolded transthyretin (TTR) protein and resulting insoluble aggregates that deposit in extracellular regions of various tissues and organs. There are hereditary forms of the disease, referred to as ATTRm, and more than 100 TTR amyloid-forming mutants have been reported. The major goal of this work was to analyze the biochemical and biophysical properties of a unique and recently identified TTR mutant protein, TTR Glu51_Ser52dup, found in a patient with ATTRm. Unlike other single nucleotide replacements that have been described as amyloidogenic, the gene abnormality in the present case is the first identification of a TTR duplication mutation. The patient with TTR Glu51_Ser52dup exhibited an extremely aggressive form of ATTRm; clinical symptoms included peripheral neuropathy at baseline evaluation and rapid disease progression to early death from pneumonia and congestive heart failure. We hypothesized that the TTR Glu51_Ser52dup variant would be less stable than the wild-type protein and similar in stability to another highly amyloidogenic mutant, TTR L55P; moreover, the highly unstable nature of this TTR variant would provide a basis for understanding the extremely aggressive clinical phenotype observed in this case. Using Escherichia coli (E. coli) as an expression system and an appropriately modified expression vector, we produced histidine-tagged recombinant human TTR Glu51_Ser52dup protein in high yield and purified to homogeneity. Structural and stability studies were performed by circular dichroism (CD) spectroscopy and SDS-PAGE analysis. We demonstrated that TTR Glu51_Ser52dup was less stable than the wild-type or L55P proteins when measured under different types of denaturing conditions, including thermal and chemical stress. The presence of diflunisal, a drug that stabilizes tetrameric TTR and is currently approved for treatment of ATTRm, was also investigated; our results indicated that diflunisal stabilized the TTR Glu51_Ser52dup protein. Collectively, the data obtained from these studies suggest that Glu51_Ser52dup is one of the least stable and most amyloidogenic TTR variant described to date. Future investigations are necessary to determine which specific structural elements of the protein destabilize the TTR tetramer, and precisely characterize the binding of small molecules, including diflunisal, to the protein.

Tungsten is considered the prime candidate material for plasma facing components within fusion reactors. However, exposure of tungsten to neutron flux brings about transmutation of tungsten into by-products: Re, Os and Ta. Under increasing levels of radiation damage, irradiation induced clustering/precipitation takes place that embrittles and thus reduces lifetime of such tungsten components. This thesis was undertaken to explore this subject on a deeper level. There are three components to this study. The first part considers the effect of Re content on irradiation induced clustering. Lab-made plate W-xRe (x: 2, 5, 10 and 25at.%) alloys were exposed to 1.2, 3.9 and 33dpa, self-ion irradiated at 773K. Analysis of cluster number density and volume fraction found they increase with damage level and bulk Re content. Based on these trends and existing literature data, a hypothesis was proposed suggesting clusters originate from vacancy clusters. Also, at 33dpa, rod shaped clusters form in W-25Re alloys while spherical clusters are present in other alloys. The clusters show close correspondence with irradiation induced precipitates and appear to be precursor phase. In the second part of this thesis, the effect of Os and Ta on cluster formation and alloy mechanical properties is examined. Lab-made plate W-1Re-1Os and W-2Re-1Ta alloys were irradiated at 33dpa at 573 and 773K and compared against control W-2Re alloy. At 33dpa and 573K, the Os and Ta presence suppresses cluster formation. Both ternary alloys contain smaller cluster diameter, composition, number density and volume fraction than the W-2Re alloy. However, at 33dpa and 773K, Os and Ta have opposing effects on cluster behaviour. Os increases the cluster nucleation rate and raises irradiation hardening (compared to W-2Re). Meanwhile, Ta presence decreased cluster number density and reduced the irradiation hardening (compared to W-2Re alloy). As well, Ta showed no evidence of clustering, only Re clusters form in the W-2Re-1Ta alloy. The final aspect of the thesis analyzes the effect of material microstructure and external variables on cluster formation in W-Re alloys. Commercial wire form W-25Re alloy was irradiated at 1.2dpa at 573 and 773K as atom probe needles and bulk sample. The larger free surface on atom probe needle samples appears to act as a sink for self-interstitials and vacancies at both temperatures. The effect of grain size and dislocation density was examined by irradiating commercial W-5Re wire (0.5-1μm diameter) and plate (1-3mm diameter) samples (annealed and unannealed) to 33dpa and 573K. It was found grain boundaries and dislocations act as defect sinks at 573K. Additionally, radiation enhanced Re grain boundary enrichment was observed for first time. The effect of grain size on cluster behaviour at 773K was also analysed. Commercial wire and lab-made plate W-3Re, W-5Re and W-25Re alloys were irradiated to 33dpa at 773K. The larger grain boundary area in wire samples is suspected of acting as a sink for self-interstitials leaving more vacancies for promoting cluster formation compared to lab-made samples. The discoveries made in this thesis broaden our current understanding of irradiation induced phase formation in tungsten. Their implications on plasma facing component design are discussed as well as recommendations for improvements. Further, areas requiring further research in this field are also highlighted.

Dans cette thèse, nous donnons une introduction systématique à la condition dépendance faible, introduit par Doukhan and Louhichi (1999) , qui est plus générale que les cadres classiques de mélange ou de séquences associées. La notion est suffisamment large pour inclure des modèles standards tels que les modèles stables de Markov , les modèles bilinéaires , et plus généralement , les schémas de Bernoulli. Dans certains cas, aucunes des propriétés de mélangeant ne peut s'attendre sans hypothèse de régularité supplémentaire sur la distribution innovations pour lesquelles une condition de dépendance faible peut être facile- ment dérivée. Nous étudions la relation entre dépendance faible et mélangeant pour les processus de valeurs discrètes. Nous montrons que la dépendance faible implique des conditions de mélangeant sous des hypothèses naturelles. Les ré- sultats se spécialisent au cas des processus Markovian. Plusieurs exemples de processus à valeur entier sont examinés et leurs propriétés de dépendance faibles sont étudiés à l'aide d'une contraction principale.Dans la deuxième partie, nous établissons des vitesses de convergences en apprentissage statistique pour les prédictions d'une série chronologique. En util- isant l'approche PAC- bayésienne, les vitesses lentes de convergence d/n pour l'estimateur de Gibbs sous la perte absolue ont été donnés dans un travail précé- dent Alquier and Wintenberger (2012), où n est la taille de l'échantillon et d la dimension de l'ensemble des prédicteurs. Sous les mêmes conditions de dépendance faible, nous étendons ce résultat à une fonction de perte Lipschitz convexe. Nous identifions également une condition sur l'espace des paramètres qui assure des vitesses similaires pour la procédure classique de l'ERM pé- nalisé. Nous appliquons cette méthode pour la prédiction quantile du PIB français. Dans des conditions supplémentaires sur les fonctions de perte ( satis- faites par la fonction de perte quadratique ) et pour les processus uniformément mélangeant, nous montrons que l'estimateur de Gibbs atteint effectivement lesivvitesses rapides de convergence d/n. Nous discutons de l' optimalité de ces dif- férentes vitesses à abaisser les limites en soulignant des références quand elles sont disponibles. En particulier, ces résultats apportent une généralisation des résultats de Dalalyan and Tsybakov (2008) sur l'estimation en régression sparse à certains auto-régression. / This thesis aims at a systematic introduction to a weak dependence condition, provided by Doukhan and Louhichi (1999), which is more general than the clas- sical frameworks of mixing or associated sequences. The notion is broad enough to include some standard models such as stable Markov models, bilinear models, and more generally, Bernoulli shifts. In some cases no mixing properties can be expected without additional regularity assumption on the distribution of the innovations distribution for which a weak dependence condition can be easily de- rived. We investigate the relationship between weak dependence and mixing for discrete valued processes. We show that weak dependence implies mixing con- ditions under natural assumptions. The results specialize to the case of Markov processes. Several examples of integer valued processes are discussed and their weak dependence properties are investigated by means of a contraction principle.In the second part, we establish rates of convergences in statistical learning for time series forecasting. Using the PAC-Bayesian approach, slow rates of con- vergence d/n for the Gibbs estimator under the absolute loss were given in a previous work Alquier and Wintenberger (2012), where n is the sample size and d the dimension of the set of predictors. Under the same weak dependence conditions, we extend this result to any convex Lipschitz loss function. We also identify a condition on the parameter space that ensures similar rates for the clas- sical penalized ERM procedure. We apply this method for quantile forecasting of the French GDP. Under additional conditions on the loss functions (satisfied by the quadratic loss function) and for uniformly mixing processes, we prove that the Gibbs estimator actually achieves fast rates of convergence d/n. We discuss the optimality of these different rates pointing out references to lower bounds when they are available. In particular, these results bring a generalization of the results of Dalalyan and Tsybakov (2008) on sparse regression estimation to some autoregression.