Implicit Time Integration
Implicit methods define the next state through an equation that includes future values, allowing large stable steps at the cost of a solve each step.
Solving for the future state
An implicit time integrator writes the update in terms of quantities at the new time level, so the unknown appears on both sides. Backward Euler, u_{n+1} = u_n + dt f(u_{n+1}), is the prototype. Because f is evaluated at the unknown state, each step requires solving an algebraic system, linear if f is linear and nonlinear otherwise, typically with Newton-Krylov iteration.
The reward for this extra work is stability. Many implicit methods are unconditionally stable for diffusive and stiff problems, meaning the time step is limited by accuracy requirements rather than by a stability bound. This lets a simulation take steps far larger than an explicit method could survive.
Stiffness and A-stability
Stiff systems contain very fast decaying modes alongside the slow dynamics of interest. Explicit methods must resolve the fast modes for stability even when they are physically irrelevant, forcing tiny steps. A-stable and L-stable implicit methods damp those fast modes at any step size, so the step can be chosen to resolve only the slow physics.
Cost per step
The linear or nonlinear solve dominates implicit cost. Efficiency depends heavily on good preconditioning and on matrix-free Jacobian-free techniques that avoid forming and storing the full Jacobian. When steps are large enough, the reduced number of steps outweighs the higher per-step cost.
- Update depends on unknown future state, requiring a solve
- Unconditional stability for stiff and diffusive problems
- Time step limited by accuracy, not stability
- Preconditioning is decisive for solver efficiency
Implicit and semi-implicit integration is essential for transport and extended-MHD simulations that must step over fast waves to follow slow evolution on device-relevant time scales.