logo IMB
Back

Séminaire de Calcul Scientifique et Modélisation

Explicit, IMEX or implicit? Order- and stiffness-adaptive Deferred Correction methods

Davide Torlo

( Sapienza University of Rome )

Salle 1

November 19, 2026 at 11:00 AM

In many applications, stiffness comes and goes. Reactive flows are stiff at ignition and in reaction zones, chemical kinetics turns slow after fast transients, and in phase-field or radiative cooling models the stiff terms switch on and off in time. Treating the stiff terms implicitly everywhere pays for Newton solves where an explicit step would do. Explicit methods, on the other hand, fail, or need tiny time steps, where the problem is stiff.


In this talk, I present time integrators that choose, step by step, the cheapest treatment that works. They are based on Deferred Correction (DeC) methods, which gain one order of accuracy per iteration. In their efficient versions, the number of subtimenodes grows with the iterations, so the methods are naturally p-adaptive: they iterate until the correction falls below a tolerance. Each time step starts with the explicit DeC. If the corrections grow or non-finite values appear, it switches to IMEX DeC, then to IMEX Euler, and only as a last resort halves the time step. The same correction that selects the order therefore also detects stiffness, without eigenvalue estimates or stiffness indicators.


I will show results on the Van der Pol oscillator, Robertson chemical kinetics, advection–diffusion, a 2D Cahn–Hilliard equation with intermittent stiffness, a radiative cooling model and the 3D reactive Euler equations. The adaptive methods stay robust where fixed-order IMEX DeC methods fail. On the PDE tests, the adaptive sDeCdu reduces the computational time by a factor of 5 to 9 compared with its non-adaptive IMEX version.


This is joint work with Lorenzo Micalizzi (North Carolina State University).