The possibility of a controlled manipulation with molecules at the nanoscale allows us to gain net work from thermal energy, although this seems to be in contradiction to the Second Law of thermodynamics. Any manipulation, however, causes some memory records somewhere in the system's surroundings. To complete the thermodynamic cycle, these records must be reset, which costs energy that cancels the previous gain. The question is, what happens when this memory (information) is recorded only in the work reservoir? Then it cannot be reset because the record means nothing but the work gain itself (e.g., the result position of a weight in the gravity field). Is this a violation of the Second Law? To answer the question, we study in this theoretical work an exchange of energy between a physical (possibly microscopic) system that is thermalized at the beginning and another (possibly microscopic) system -- the work reservoir -- during a deterministic process in an autonomous arrangement, including also an auxiliary device controlling the process. This arrangement is suitable for deriving some equalities which express the Second Law in a form incorporating explicitly relevant memory records (and related information). We use these equalities in studying a hypothetical process including many cycles in which the only non-reset memory record is that in the work reservoir during each cycle. The results show that either the work gain is canceled in following cycles (and the work reservoir fluctuates and cannot accumulate energy), or there exists an information flow from the system (an information engine), or the system cannot work in an expected way for a purely dynamic reason (this reveals a deeper connection of the studied questions with the concept of adiabatic accessibility).

Submitted to Phys. Rev. E