Why there is a large "Global Force Balance Relative Error" for Ritz vectors?

The global force balance for a given mode is a measure of how "true" an Eigenvector is. Eigenvectors should always have small errors. The lower Ritz modes should also have small errors, since they converge to the Eigenvectors. The higher Ritz modes will have larger errors, since they are actually linear combinations of higher-modes Eigenvectors, and do not exactly satisfy K*v - w²*M*v = 0. However, provided that you have enough lower-frequency modes, the upper modes should be responding statically and this dynamic equilibrium violation may not be important.