I had an MC chain that crashed due to the java heap space error. I now tried to continue the chain using mh_recover and feeding in the old mode-file from mode_compute=6 by specifying mode_compute=0 and mode_file=filename.
Whenever I do this, the acceptance rate of the continued MC starts at exactly 100% and only slowly moves down, while it was around 15% before the crash. Considering that the recovery should start at last draw values and the hessian should be the same as before, the acceptance rate when continuing should be also 15% (the chain ran long enough to reach a stable point). Is this a bug with the display of the acceptance rate or what is going on?