1: \begin{abstract}

2: \noindent

3: Algorithmic entropy can be seen as a special case of entropy as studied

4: in statistical mechanics.  This viewpoint allows us to apply many

5: techniques developed for use in thermodynamics to the subject of

6: algorithmic information theory.  In particular, suppose we fix a

7: universal prefix-free Turing machine and let $X$ be the set of

8: programs that halt for this machine.  Then we can regard $X$ as a set

9: of `microstates', and treat any function on $X$ as an `observable'.  For

10: any collection of observables, we can study the Gibbs ensemble that

11: maximizes entropy subject to constraints on expected values of these

12: observables.  We illustrate this by taking the log runtime, length,

13: and output of a program as observables analogous to the energy $E$,

14: volume $V$ and number of molecules $N$ in a container of gas.  The

15: conjugate variables of these observables allow us to define quantities

16: which we call the `algorithmic temperature' $T$, `algorithmic

17: pressure' $P$ and `algorithmic potential' $\mu$, since they are

18: analogous to the temperature, pressure and chemical potential.  We

19: derive an analogue of the fundamental thermodynamic relation $d E = T

20: dS - P d V + \mu d N$, and use it to study thermodynamic cycles

21: analogous to those for heat engines.  We also investigate the values

22: of $T, P$ and $\mu$ for which the partition function converges.  

23: At some points on the boundary of this domain of convergence, the 

24: partition function becomes uncomputable.  Indeed, at these points 

25: the partition function itself has nontrivial algorithmic entropy.

26: \end{abstract}