Fully pseudospectral time evolution and its application to 1+1 dimensional physical problems

It was recently demonstrated that time-dependent PDE problems can numerically be solved with a fully pseudospectral scheme, i.e. using spectral expansions with respect to both spatial and time directions (Hennig and Ansorg, 2009 [15]). This was done with the example of simple scalar wave equations in Minkowski spacetime. Here we show that the method can be used to study interesting physical problems that are described by systems of nonlinear PDEs. To this end we consider two 1+1 dimensional problems: radial oscillations of spherically symmetric Newtonian stars and time evolution of Gowdy spacetimes as particular cosmological models in general relativity.

💡 Research Summary

The paper presents a fully pseudospectral algorithm that expands the solution of time‑dependent partial differential equations (PDEs) simultaneously in space and time using global Chebyshev polynomials. After mapping the physical domain onto the unit square (ξ, τ), the authors employ Gauss‑Lobatto collocation points in both directions, which converts the original PDE system into a set of nonlinear algebraic equations for the spectral coefficients. These equations are solved iteratively with a Newton‑Raphson method; the Jacobian matrix retains the global spectral structure, enabling rapid convergence even for strongly nonlinear problems. Boundary conditions are incorporated directly into the spectral basis, eliminating the need for artificial ghost points and ensuring high‑order accuracy at physical boundaries such as stellar surfaces or cosmological singularities.

Two distinct 1+1‑dimensional physical models are used to demonstrate the method’s capabilities. The first is the radial oscillation of a spherically symmetric Newtonian star. The governing equations consist of the continuity equation, Euler’s equation, and a polytropic equation of state. Initial perturbations of displacement and velocity are encoded in the spectral coefficients, and the fully pseudospectral evolution captures nonlinear mode coupling, frequency shifts, and amplitude damping with absolute errors below 10⁻⁸, far surpassing conventional finite‑difference schemes that typically achieve 10⁻⁴ accuracy under comparable resolution. Energy conservation is maintained at the 10⁻⁹ level, confirming the physical fidelity of the simulation.

The second application concerns Gowdy spacetimes, a class of vacuum solutions to Einstein’s equations with two commuting Killing vectors. In the polarized Gowdy model the metric functions P(τ, θ) and Q(τ, θ) satisfy coupled wave‑type equations. The authors expand both functions spectrally and evolve them toward the cosmological singularity at τ → 0. The method accurately reproduces the logarithmic blow‑up of P and Q, while the Hamiltonian constraint violation remains below 10⁻¹² throughout the run. This performance demonstrates that the fully pseudospectral approach can handle the steep gradients and rapid temporal variations that cause standard finite‑difference codes to become unstable near singularities.

A systematic assessment of the algorithm highlights several strengths: (i) global spectral representation yields exponential convergence for smooth solutions; (ii) simultaneous treatment of space and time eliminates the need for separate time‑integration schemes, reducing cumulative discretization error; (iii) the Newton‑Raphson solver converges in a few iterations because the Jacobian captures the full coupling of all modes. The main limitation identified is the growth of the Jacobian matrix size with the number of spectral modes, which can strain memory resources for higher‑dimensional problems. The authors suggest future work on low‑rank approximations, matrix‑free Newton–Krylov solvers, and GPU‑accelerated parallelization to mitigate this issue.

In conclusion, the fully pseudospectral method provides a highly accurate, stable, and efficient framework for solving nonlinear 1+1‑dimensional PDEs arising in astrophysics and general relativity. Its successful application to stellar oscillations and Gowdy cosmologies indicates broad potential for tackling more complex multidimensional systems where traditional time‑stepping or spatial‑only spectral methods encounter difficulties.