In semiconductor theory, applying the kp-method to the monodimensional Schrödinger equation leads to a symmetric perturbed eigenvalue problem, i.e., to the diagonalization of a matrix $A(\epsilon)$ depending on a small parameter $\epsilon$, symmetric for all $\epsilon\in {\mathbb R}$. The eigenelements of $A(\epsilon)$ are expansions in fractional powers of $\epsilon$ (Puiseux series). Usually, physicists solve this problem by using Schrödinger perturbation formulas under some restrictive conditions, which make perturbed eigenvector symbolic approximation impossible. This is illustrated by the modified Kane matrix. To solve this problem completely from a symbolic computing point of view, we consider the symmetric perturbed eigenvalue problem in the case of analytic perturbations. We first review the classical characteristic polynomial approach, showing why it may not be optimal. We also present a direct matricial algorithm: transforming the analytic matrix $A(\epsilon)$ into its so-called $q$-reduced form allows to recover the information we need for the eigenvalues. This alternative method, as well as the classical one, can be described in terms of the Newton polygon. However, our approach uses only a finite number of terms of $A(\epsilon)$ and is more suitable for large matrices and a low approximation order. Besides, we show that the $q$-reduction process can simultaneously provide symbolic approximations of both the perturbed eigenvalues and eigenvectors. The implementation of this algorithm in Maple is used to diagonalize the modified Kane matrix up to a given order.