CosmoCoffee

Is the covariance matrix(outputting file of CosmoMC, eg.
test.covmat) the invertion of Fisher matrix at the best fitting
point?

I'm confused, because the params.ini says one "can use covariance
matrix for proposal density", but I think the covariance matrix
derived from Fisher matrix is very different from the proposal
matrix in the MCMC technology.

It is the covariance matrix of the samples.

If supplied for an MCMC run it is used to construct a decent proposal matrix that knows about parameter scales and correlations.

I see, thank you very much!