Inverse of Hessian matrix

Hello everyone,

Could anyone help me to solve the following problem please?

If the inverse of Hessian matrix (W) , which is estimated directly from the optimization package of Chris Sim, is not positive-definite, how can we calculate its square root to make a jump distribution N(mu, W) for random walk MCMC as suggested in the literature please?

Thank you very much,
Best regards,
A

See the discussion in the bottom part here [Usual Bayesian estimation problem). The bottom line: there are numerical ways to deal with this, but this is not recommended as the non-positive definite Hessian usually signals deeper problems.