1: \begin{abstract}

2: Given a sequence of undirected connected graphs $G_n$ of $m_n$ vertices with bounded degrees and weights and an additional sequence of killing parameters $c_n$, one can define a sequence $\mu_n$ of natural non-normalized measures of non-trivial loops on $G_n$. Denote by $Q_n$ the transition matrix associated with $G_n$. Assuming that the empirical distributions $\nu_n$ of the eigenvalues of $Q_n$ converge, the main result is the determination of the limit of the $\mu_n-$proportion of loops which cover every vertex as $n\rightarrow\infty$. Let $\mathcal{L}_n$ be the Poissonian ``loop soup'' with intensity $\frac{\mu_n}{m_n}$. A corollary of the main result is the limit law of the number of the loops covering the whole graph. This paper also gives a criterion for the convergence of the empirical distributions of the eigenvalues of the transition matrices and states that modifications on the weights on a small number of edges of $G_n$ do not change the limit mentioned above. Several examples of graphs are analysed at the end of the paper.

3: \end{abstract}

4: