Numerical Methods · interactive worked example

Explicit vs. Implicit: When a Time Step Blows Up

The same circular fin from the steady-state lab, now heating up from room temperature over time. Step it forward explicitly and there's a hard speed limit on the time step — cross it, and the "solution" detonates into nonsense within a few dozen steps. Step it forward implicitly and there's no limit at all.

A transient extension of the Circular Fin worked example — same fin, now solved in time as well as space.

∂θ/∂t = α·∂²θ/∂x² − α·m²·θ  ·  Stable explicit step: Δt ≤ 1 / (2α/Δx² + α·m²)
Same fin equation as before, with time added back in. Implicit time-stepping has no such limit — the price is solving a linear system every step instead of just evaluating a formula.

θ(x,t) — heating up from ambient toward the steady profile

Steady-state target Current θ(x,t)
Δt / Δt_critical
Elapsed time
Status

What you're seeing

Method

Time step, as a multiple of Δtcritical

0.50

Where the speed limit comes from

Explicit time-stepping computes tomorrow's temperature using only today's neighbors — fast, but it implicitly assumes the disturbance can't travel faster than information moves through the grid in one step. Make the time step too large relative to the grid spacing, and that assumption breaks: errors stop decaying and start amplifying each step instead, doubling and redoubling until the numbers overflow into nonsense. That's not the physics failing — the real fin doesn't explode — it's the numerical scheme failing, which is exactly why the stability limit has to be respected, not just hoped past.

Why implicit doesn't have this problem

Implicit time-stepping computes tomorrow's temperature using tomorrow's neighbors too — which means solving a small linear system (the same tridiagonal-matrix idea as the Gauss-Seidel lab) at every step instead of just evaluating a formula. That extra cost buys unconditional stability: any time step, however large, gives a bounded answer. Large steps still lose accuracy — you can step right past interesting transient behavior — but they never blow up the way explicit does.

Try it

Push the time-step ratio past 1.0 with Explicit selected and press Play — watch the curve start oscillating, then diverge entirely within a handful of steps. Switch to Implicit at the exact same ratio and watch it calmly settle onto the same steady profile from the Circular Fin lab instead.

EngineeringCandy · Same fin, same matrix-building ideas, now marching forward in time · break the time step on purpose