Perfect, thanks a lot for your very constructive and accurate help.
With this non linear oscillator, tuning the numpoints(with numpoints=10*tmax) enables to keep the solution with the limit cycle.
However, I tried to used the same tuning (that is to say to use the same number of points numpoints) in a more complex coupled oscillator and this time it doesn't work.
My code is the following :
K:=Matrix([<0, -1, 1, -1>,<-1, 0, -1, 1>,<-1, 1, 0,-1>,<1, -1, -1,0>]);
for i to 4
ic:=[u(0)=0, v(0)=0,u(0)=0, v(0)=-0.1,u(0)=0, v(0)=0.1,u(0)=0, v(0)=0.1];
res:=dsolve([sys,ic],numeric, maxfun = 0, output = listprocedure):
plots:-odeplot(res,[u(t),v(t)],0..tmax, scaling = constrained, numpoints = 10*tmax);
However, it seems to work with numpoints = 100*tmax.
Do you have ideas so as to tune numpoints correctly ? I would like to build a rule of thumb so as to test directly a more appropriated number of point (numpoints).
Thank you in advance for your feedback.