1: \begin{abstract}

2: We study the spectrum of a random multigraph with

3: a degree sequence $\mathbf{D}_n=(D_i)_{i=1}^n$ and average degree $1 \ll \omega_n \ll n$, generated by the configuration model.

4: We show that, when the empirical spectral distribution (\abbr{ESD}) of $\omega_n^{-1} \mathbf{D}_n $ converges weakly to a limit $\nu$, under mild moment assumptions (\emph{e.g.}, $D_i/\omega_n$ are i.i.d.\ with a finite second moment),

5: the \abbr{ESD} of the normalized adjacency matrix converges in probability to

6: $\nu\boxtimes \sigma_{\semic}$, the free multiplicative convolution of $\nu$ with the semicircle law. Relating this limit

7: with a variant of the Marchenko--Pastur law yields the continuity of its

8: density (away from zero), and an effective procedure for determining its support.

9: 

10: Our proof of  convergence is based on a coupling of the graph to an inhomogeneous Erd\H{o}s-R\'enyi graph with the target \abbr{ESD}, using three intermediate random graphs, with a negligible number of edges modified in each step.

11:   \end{abstract}

12: