try to make the minimum autocorrellation estimation lower

Currently, the act estimate seems to be almost always the same as walks.

This is due to an artefact of the translation from CPNest.

This MR attempts to fix this.