How matrix inversion in LDA makes data unstable?

Question

I read that during inversion of the covariance matrix in LDA, it is numerically unstable if too few observations are available. I don\'t get what does numerically unstable m