Em teoria dos números, o teorema de Erdös-Kac, também conhecido como o teorema fundamental de teoria probabilística dos números, diz que se w(n) denota a quantidade de fatores primos distintos de n, então a sequência de funções de distribuições N definidas por FN(x) = (1/N) #{n <= N : (w(n) log log N)/(log log N)^(1/2)} <= x}, converge uniformemente sobre R para a distribuição normal padrão. Neste trabalho desenvolvemos todos os teoremas necessários para uma demonstração analítica, que nos permitirá encontrar a ordem de erro da convergência acima. / In number theory, the Erdös-Kac theorem, also known as the fundamental theorem of probabilistic number theory, states that if w(n) is the number of distinct prime factors of n, then the sequence of distribution functions N, defined by FN(x) = (1/N) #{n <= N : (w(n) log log N)/(log log N)^(1/2)} <= x}, converges uniformly on R to the standard normal distribution. In this work we developed all theorems needed to an analytic demonstration, which will allow us to find an order of error of the above convergence.

Em teoria dos números, o teorema de Erdös-Kac, também conhecido como o teorema fundamental de teoria probabilística dos números, diz que se w(n) denota a quantidade de fatores primos distintos de n, então a sequência de funções de distribuições N definidas por FN(x) = (1/N) #{n <= N : (w(n) log log N)/(log log N)^(1/2)} <= x}, converge uniformemente sobre R para a distribuição normal padrão. Neste trabalho desenvolvemos todos os teoremas necessários para uma demonstração analítica, que nos permitirá encontrar a ordem de erro da convergência acima. / In number theory, the Erdös-Kac theorem, also known as the fundamental theorem of probabilistic number theory, states that if w(n) is the number of distinct prime factors of n, then the sequence of distribution functions N, defined by FN(x) = (1/N) #{n <= N : (w(n) log log N)/(log log N)^(1/2)} <= x}, converges uniformly on R to the standard normal distribution. In this work we developed all theorems needed to an analytic demonstration, which will allow us to find an order of error of the above convergence.

Cette thèse se compose de deux projets. Le premier concerne la structure des fonctions multiplicatives dont les moyennes sont petites. En particulier, dans ce projet, nous établissons le comportement moyen des valeurs \(f(p)\) de \(f\) aux nombres premiers pour des fonctions \(f\) multiplicatives appropriées lorsque leurs sommes partielles \(\sum_{n\leqslant x}f(n)\) sont plus petites que leur borne supérieure triviale par un facteur d′une puissance de \(\log x\). Ce résultat poursuit un travail antérieur de Koukoulopoulos et Soundararajan et il est construit sur des idées provenant du traitement plus soigné de Koukoulopoulos sur le cas special des fonctions multiplicatives bornées. Le deuxième projet de la thèse est inspiré par un analogue d’une estimation que Linnik a déduit dans sa tentative de prouver son célèbre théorème concernant la taille du plus petit nombre premier d’une progression arithmétique. Cette estimation fournit une formule asymptotique fortement uniforme pour les sommes de la fonction de von Mangoldt \(\Lambda\) sur les progressions arithmétiques. Dans la littérature, ses preuves existantes utilisent des informations non triviales sur les zéros des fonctions \(L\) de Dirichlet \(L(\cdot,\chi)\) et le but du deuxième projet est de présenter une approche différente, plus élémentaire qui récupère cette estimation en évitant la “langue” de ces zéros. Pour le développement de cette méthode alternative, nous utilisons des idées qui apparaissent dans le grand crible prétentieux (pretentious large sieve) de Granville, Harper et Soundararajan. De plus, comme dans le cas du premier projet, nous empruntons également des idées du travail de Koukoulopoulos sur la structure des fonctions multiplicatives bornées à petites moyennes. / This thesis consists of two projects. The first one is concerned with the structure of multiplicative functions whose averages are small. In particular, in this project, we establish the average behaviour of the prime values \(f(p)\) for suitable multiplicative functions \(f\) when their partial sums \(\sum_{n\leqslant x}f(n)\) admit logarithmic cancellations over their trivial upper bound. This result extends previous related work of Koukoulopoulos and Soundararajan and it is built upon ideas coming from the more careful treatment of Koukoulopoulos on the special case of bounded multiplicative functions. The second project of the dissertation is inspired by an analogue of an estimate that Linnik deduced in his attempt to prove his celebrated theorem regarding the size of the smallest prime number of an arithmetic progression. This estimate provides a strongly uniform asymptotic formula for the sums of the von Mangoldt function \(\Lambda\) on arithmetic progressions. In the literature, its existing proofs involve non-trivial information about the zeroes of Dirichlet \(L\)-functions \(L(\cdot,\chi)\) and the purpose of the second project is to present a different, more elementary approach which recovers this estimate by avoiding the “language” of those zeroes. For the development of this alternative method, we make use of ideas that appear in the pretentious large sieve of Granville, Harper and Soundararajan. Moreover, as in the case of the first project, we also borrow insights from the work of Koukoulopoulos on the structure of bounded multiplicative functions with small averages.