Smooth relaxation preserving Turing machines

Abstract: Clift and Murfet (2019) introduced a naive Bayesian smooth relaxation of Turing machines motivated by work in differential linear logic; this was subsequently used to endow spaces of program codes of bounded length with a smooth manifold structure via the staged-pseudo universal Turing machine introduced by Clift, Murfet and Wallbridge (2021). In this paper, we give a general construction for simulating n-tape Turing machines on a single tape Turing machine such that the (naive Bayesian) uncertainty is propagated in an equivalent manner. This result suggests a stronger kind of equivalence between single tape and n-tape Turing machines than that established by classical results, however, the clarification of these implications is open to future work. We then construct a pseudo universal Turing machine which similarly preserves the propagation of uncertainty in its simulations, and observe that this gives rise to a particularly natural smooth relaxation of the space of programs.