Hi all,

I've got a question related to relative errors in the special case where the x(exact) is nearly 0.

Example : x(exact) = 1 e-10 x(simul) = 0.001

Er = |x(exact)-x(simul)|/|x(exact)| = 9999999!!!

I would like to know if there is a way to avoid this situation.

Thanks in advance.

Can you make your system non dimensional ? In that way you would not need to go for such high precision requirements.

Well, the purpose is to compare some methods via the absolute and relative errors. The equation resolved is of type: m.d2(y)/dt2 + k.y = f with y=1 and dy/dt=0 and the exact solution is y=cos(wt). It's clear to see that the y is zero, or nearly, at Pi/2 for example, as I explained before.