Title: Spectral Densities from Dynamic Density-Matrix Renormalization

Abstract: Dynamic density-matrix renormalization provides valuable numerical information on dynamic correlations by computing convolutions of the corresponding spectral densities. Here we discuss and illustrate how and to which extent such data can be deconvolved to retrieve the wanted spectral densities. We advocate a nonlinear deconvolution scheme which minimizes the bias in the ansatz for the spectral density. The procedure is illustrated for the line shape and width of the Kondo peak (low energy feature) and for the line shape of the Hubbard satellites (high energy feature) of the single impurity Anderson model. It is found that the Hubbard satellites are strongly asymmetric.