1: \begin{abstract}

2: Many problems in physics, chemistry and other fields are perturbative in

3: nature,~\emph{i.e.} differ only slightly from related problems with

4: known solutions. Prominent among these is the eigenvalue perturbation

5: problem, wherein one seeks the eigenvectors and eigenvalues of a matrix

6: with small off-diagonal elements. Here we introduce a novel iterative

7: algorithm to compute these eigenpairs based on fixed-point iteration for an algebraic equation in complex projective space.

8: We show from explicit and numerical examples that our algorithm

9: outperforms the usual Rayleigh-Schr\"odinger expansion on three counts.

10: First, since it is not defined as a power series, its domain of

11: convergence is not~\emph{a priori} confined~to a disk in the complex

12: plane; we find that it indeed usually extends beyond the standard perturbative radius of

13: convergence. Second, it converges at a faster rate than the Rayleigh-Schr\"odinger expansion, \emph{i.e.} fewer iterations are required to reach a given

14: precision. Third, the (time- and space-) algorithmic complexity of each

15: iteration does not increase with the order of the approximation,

16: allowing for higher precision computations. Because this complexity is

17: merely that of matrix multiplication, our dynamical scheme also scales

18: better with the size of the matrix than general-purpose eigenvalue

19: routines such as the shifted QR or divide-and-conquer algorithms.

20: Whether they are dense, sparse, symmetric or~unsymmetric, we confirm

21: that dynamical diagonalization quickly outpaces LAPACK drivers as the

22: size of matrices grows; for the computation of just the dominant

23: eigenvector, our method converges order of magnitudes faster than the

24: Arnoldi algorithm implemented in ARPACK.~~%

25: \end{abstract}%