NormDispIncr vs. RelativeNormDispIncr

[murray710]

Hi,

When using NormDispIncr, is the comparison normalized in any way to the initial target solution vector, or is that only done with RelativeNormDispIncr? Basically, is the NormDispIncr tolerance 'unit dependent' and 'step-size dependent'?

For instance, if I reduce the size of a displacement step, for instance, would I need to proportionally tighten the convergence criteria in order to keep the tolerance 'the same', or is it normalized to the target value?

Thanks,
Justin

[fmk]

NormDispIncr is indeed unit dependent: norm(dU) < tol.
RelativeNorm is not: norm(dU(step))/norm(dU(initial step in iteration))< tol.

if however initial step size is very small, i.e. 1e-12 RelativeNorm may not converge!