Paperlane

no. 16 · Tuesday 17 November 2026

High-order Crank-Nicolson method via diagonal Padé factorization

Lou Freschi, Camille Lévêque, Jérémie Caillat, Richard Taïeb and Jonathan Dubois

math.NA

Abstract

We present a numerical method to solve the time-dependent Schrödinger equation at arbitrary order of accuracy in the time step, particularly well suited for tridiagonal Hamilton operators. The method is based on a factorization of the diagonal Padé approximants of the matrix exponential into products of linear factors, which preserves tridiagonal matrix structure at every time step and yields a propagator of order 2q in the time step at a cost of O(qN) operations. The propagator is exactly unitary by construction, and at lowest order q=1 it reduces to the standard Crank-Nicolson algorithm. We further extend the method to any Hamiltonian admitting a decomposition into a sum of tridiagonal or diagonal operators via high-order split-operator techniques (McLachlan, 2022), retaining O(qN) scaling at each time step. We also analytically show and numerically verify that the precision is dictated by the split-operator method. This covers a broad range of physically relevant cases, including higher-dimensional systems, higher-order spatial discretizations, and non-separable potentials.

fields
Numerical Analysis
pdf ↗ page arXiv:2611.20800 ↗