%0 Generic %A Titov, Ivan %C Heidelberg %D 2024 %F heidok:34250 %R 10.11588/heidok.00034250 %T Solovay Reducibility and Speedability Outside of left-c.e. Reals %U https://archiv.ub.uni-heidelberg.de/volltextserver/34250/ %X A real number is left-c.e.\ if it has a left-c.e.\ approximation, i.e., a computable nondecreasing sequence $a_0, a_1, ...$ of rationals that converges to the real number. Furthermore, a real number $\alpha$ is Solovay reducible to a real number $\beta$ if there exists a partially computable function~$g$ that maps every rational number~$q<\beta$ to some rational number~$g(q)<\alpha$ such that, for some real constant~$c$ and all~$q < \beta$, it holds that \[ \alpha - g(q) < c (\beta - q) . \] Solovay reducibility can be used to compare the speed at which left-c.e.\ numbers can be approximated: if a real number~$\alpha$ is Solovay reducible to a left-c.e.\ real number~$\beta$, then also~$\alpha$ is left-c.e.\ and, for every left-c.e.\ approximation of~$\beta$, there is a left-c.e.\ approximation of~$\alpha$ that converges at least as fast up to a constant factor. Among the left-c.e.\ reals, the Martin-Löf random ones have been intensively studied, and it is known that they have several natural equivalent characterizations. For example, by results of Solovay~\cite{1975} and of Calude, Hertling, Khoussainov and Wang~\cite{Calude-Hertling-Khoussainov-Wang}, the Martin-Löf random left-c.e.\ reals are exactly the halting probabilities of universal Turing machines. Furthermore, Kučera and Slaman~\cite{Kucera-Slaman-2001} demonstrated that, within the left-c.e.\ reals, the Martin-Löf random ones form a highest degree of Solovay reducibility, i.e., a left-c.e.\ real~$\beta$ is Martin-Löf random if and only if every left-c.e.\ real~$\alpha$ is reducible to~$\beta$. In fact, they showed that the latter holds via arbitrary left-c.e.\ approximations of~$\alpha$ and~$\beta$. As a consequence, given any Martin-Löf random left-c.e.\ reals~$\alpha$ and~$\beta$, they are mutually Solovay reducible to each other via arbitrary left-c.e.\ approximations~$a_0, a_1, \ldots$ and~$b_0, b_1, \ldots$ of~$\alpha$ and~$\beta$, respectively, hence, there are reals $c>0$ and $d$ such that, for all~$n$, it holds that \begin{equation}\label{eq:abstract-solovay-constant} c<\frac{\alpha - a_n}{\beta - b_n}0$ via any left-c.e.\ approximation of the real. Also speedability is a degree property, i.e., in a Solovay degree, either every or no real is speedable. Furthermore, Martin-Löf random left-c.e.\ reals are never speedable, while all nonhigh left-c.e.\ reals are speedable. For speedability defined in terms of monotone Solovay reducibility, some of these results extend to all reals, in particular, robustness with respect to the choice of nonzero~$\rho$ and the nonspeedability of Martin-Löf random reals. Being Martin-Löf random is not equivalent to being nonspeedable, neither for all reals nor when restricting attention to the left-c.e.\ reals. The former result is shown below by constructing a right-c.e.\ counterexample, i.e., a right-c.e.\ real that is neither Martin-Löf random nor speedable. The latter, more interesting and more difficult result is due to Hölzl and Janicki~\cite{Hoelzl-Janicki-2023}, who constructed a left-c.e.\ counterexample. Third, the theorem of Barmpalias and Lewis-Pye allows an equivalent reformulation in terms of monotone Solovay reducibility, which can be extended to all reals. This extension is one of the main results of this thesis. A corresponding reformulation in terms of Solovay reducibility is false in general and is actually false for all left-c.e.\ reals.