Initializing implicit linear multistep methods

Question

A sixth order backward differentiation formula (BDF) need six (five plus initial value) previous solutions to get started. How I can get these previous solutions? I need a method accurate to sixth order capable of handling stiff problems.

Answer 1

The standard approach is to use a self-starting time-marching algorithm with sufficiently small timestep (such that the order of accuracy is not spoiled) and compute the 5 non-initial value previous solutions. These are then used to "start" the BDF formula.

Answer 2

Just to add a bit to @Jesse Chan's answer: you can preserve 6th-order convergence if you use a 5th-order starting method; in general, the starting method can be one order lower than the multistep method without decreasing the global convergence rate.

The most common starting methods are, of course, Runge-Kutta methods.