1: \documentclass[envcountsame,runningheads]{llncs}
2: \usepackage{xspace}
3: \usepackage{amsmath}
4: \usepackage{amssymb}
5: \usepackage{graphicx}
6: \DeclareFontFamily{U}{mathb}{\hyphenchar\font45}
7: \DeclareFontShape{U}{mathb}{m}{n}{
8: <5> <6> <7> <8> <9> <10> gen * mathb
9: <10.95> mathb10 <12> <14.4> <17.28> <20.74> <24.88> mathb12
10: }{}
11: \DeclareSymbolFont{mathb}{U}{mathb}{m}{n}
12: \DeclareMathSymbol{\precneq}{3}{mathb}{"AC}
13: \newcommand{\Nreduceq}{\reduceq_{\text{n}}}
14: \newcommand{\Nequiv}{\equiv_{\text{n}}}
15: \newcommand{\Tequiv}{\equiv_{\text{T}}}
16: \newcommand{\reduceq}{\preceq}
17: \newcommand{\reducneq}{\precneq}
18: \newcommand{\notreduceq}{\npreceq}
19: \newcommand{\reduceqt}{\reduceq_\text{t}}
20: \newcommand{\reducneqt}{\precneq_\text{t}}
21: \newcommand{\notreduceqt}{\npreceq_\text{t}}
22: \newcommand{\BCSS}{BCSS\xspace} %or BSS
23: \newcommand{\IR}{\mathbb{R}}
24: \newcommand{\IRc}{\mathbb{R}_c}
25: \newcommand{\IA}{\mathbb{A}}
26: \newcommand{\IT}{\mathbb{T}}
27: \newcommand{\IQ}{\mathbb{Q}}
28: \newcommand{\IZ}{\mathbb{Z}}
29: \newcommand{\IH}{\mathbb{H}}
30: \newcommand{\IN}{\mathbb{N}}
31: \newcommand{\dom}{\operatorname{dom}}
32: \newcommand{\bin}{\operatorname{bin}}
33: \newcommand{\id}{\operatorname{id}}
34: \newcommand{\myjoin}{\wedge} %alternative: \sqcap
35: \DeclareMathOperator*{\mysup}{\smash{\sup}}
36: \DeclareMathOperator*{\myinf}{\smash{\inf}}
37: \DeclareMathOperator*{\mytheta}{\Theta}
38: \DeclareMathOperator*{\ulim}{ulim}
39: \newcommand{\calO}{\mathcal{O}}
40: \newcommand{\calM}{\mathcal{M}}
41: \newcommand{\Heavi}{h}
42: \newcommand{\barH}{\overline{\Heavi}}
43: \newcommand{\RDelta}[1]{\Delta_{#1}\IR}
44: \newcommand{\RSigma}[1]{\Sigma_{#1}\IR}
45: \newcommand{\RPi}[1]{\Pi_{#1}\IR}
46: \newcommand{\person}[1]{\textsc{#1}}
47: \newcommand{\aname}[1]{\textsf{#1}}
48: \newcommand{\mycite}[2]{\cite[\textsc{#1}]{#2}}
49: \newcommand{\myto}{\!\to\!}
50: \newcommand{\myrho}{\rho}
51: \newcommand{\myl}{{\scriptscriptstyle<}}
52: \newcommand{\myg}{{\scriptscriptstyle>}}
53: \newcommand{\myrhol}{\myrho_{\raisebox{0.2ex}{$\myl$}}}
54: \newcommand{\myrhog}{\myrho_{\raisebox{0.2ex}{$\myg$}}}
55: \newcommand{\myrhon}{\myrho_{\text{Cn}}}
56: \newcommand{\myrhod}{\myrho_{\text{b},10}}
57: \newcommand{\myrhob}{\myrho_{\text{b},2}}
58: \newcommand{\myrhoh}{\myrho_{\text{H}}}
59: \newcommand{\tildemyrhol}{\tilde\myrho_{\raisebox{0.2ex}{$\myl$}}}
60: \newcommand{\hatmyrhol}{\hat\myrho_{\raisebox{0.2ex}{$\myl$}}}
61: \newcommand{\nuQ}{\nu_{\IQ}}
62: \newcommand{\Card}[1]{|{#1}|}
63: \newcommand{\Gdelta}{\text{G}_{\delta}}
64: \newcommand{\lcp}{\operatorname{lcp}}
65: \newcommand{\substr}{\sqsubseteq}
66: \newcommand{\supstr}{\sqsupseteq}
67: \newcommand{\LSC}{\operatorname{LSC}}
68: \newcommand{\MLSC}[1]{\operatorname{MLSC}(#1)}
69: \newcommand{\COMMENTED}[1]{}
70: \newcommand{\comment}[1]{\maginpar{\footnotesize\sf #1}}
71: \spnewtheorem{observation}[theorem]{Observation}{\bfseries}{\itshape}
72: \spnewtheorem{fact}[theorem]{Fact}{\bfseries}{\itshape}
73: \spnewtheorem{myclaim}[theorem]{Claim}{\bfseries}{\itshape}
74: \spnewtheorem{scholium}[theorem]{Scholium}{\bfseries}{\itshape}
75: \newtheorem{scholiumf}[theorem]{Scholium\footnotemark}
76: \begin{document}
77: \title{Real Hypercomputation and Continuity\thanks{An extended abstract
78: of this work, mostly lacking proofs, has appeared as \cite{CIE}.}}
79: \titlerunning{Real Hypercomputation and Continuity}
80: \author{Martin Ziegler\thanks{supported by the German Research Fundation
81: \textsf{DFG} with project \texttt{Zi1009/1-1}}}
82: \authorrunning{Martin Ziegler}
83: \institute{University of Paderborn, 33095 GERMANY}
84: \maketitle
85: \def\thefootnote{\fnsymbol{footnote}}
86: \addtocounter{footnote}{2}
87: \begin{abstract}
88: By the sometimes so-called \emph{Main Theorem} of
89: Recursive Analysis, every computable real function
90: is necessarily continuous.
91: We wonder whether and which kinds of \emph{hyper}computation
92: allow for the effective evaluation of also discontinuous $f:\IR\to\IR$.
93: More precisely the present work considers the following three
94: super-Turing notions of real function computability:
95: \begin{itemize}
96: \item relativized computation; \quad specifically
97: given oracle access to the Halting Problem
98: $\emptyset'$ or its jump $\emptyset''$;
99: \item encoding input $x\in\IR$ and/or output $y=f(x)$ in weaker
100: ways also related to the Arithmetic Hierarchy;
101: \item non-deterministic computation.
102: \end{itemize}
103: It turns out that any $f:\IR\to\IR$ computable in the
104: first or second sense is still necessarily continuous
105: whereas the third type of hypercomputation
106: does provide the required power to evaluate for instance
107: the discontinuous Heaviside function.
108: \end{abstract}
109: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
110: \section{Motivation}
111: What does it mean for a Turing Machine,
112: capable of operating only on discrete objects,
113: to compute a real number $x$:
114:
115: \smallskip
116: \noindent\begin{tabular}{rl}
117: $\pmb{\myrhob}$\textbf{:} & To determine its binary expansion, i.e.,
118: to decide $A\subseteq\IN$ with $x=\sum\limits_{n\in A} 2^{-n}$ ?\\
119: $\pmb{\myrhon}$\textbf{:} & To compute a sequence $(q_n)$ of rational numbers
120: eventually converging to $x$?\\[0.7ex]
121: $\pmb{\myrho}$\textbf{:} & To compute a \emph{fast} convergent sequence
122: $(q_n)\subseteq\IQ$ for $x$, ~i.e. with $|x-q_n|$ \\
123: & $\leq2^{-n}$ ~ (in other words:
124: % and $(\varepsilon_n)$ with ~$|x-q_n|\leq\varepsilon_n\to0$; i.e.,
125: to approximate $x$ with effective error bounds)?\\[0.5ex]
126: $\pmb{\myrhol}$\textbf{:} & To approximate $x$ from below, i.e., to
127: compute $(q_n)$ such that $x=\sup_n q_n$ ?
128: \end{tabular}
129:
130: \medskip
131: \noindent
132: All these notions make sense in being closed under arithmetic
133: operations like addition and multiplication. In fact they are
134: well (known equivalent to variants) studied in literature\footnote{%
135: Their above names by indexed Greek letters are taken from
136: \mycite{Section~4.1}{Weihrauch}.};
137: e.g. \cite{Turing}, \cite{Naive}, \cite{Turing2}, \cite{Weihrauch} in order.
138:
139: \medskip
140: Now what does it mean for a Turing Machine $\calM$
141: to compute a real function $f:\IR\to\IR$?
142: Most naturally it means that $\calM$ realizes effective
143: evaluation $x\mapsto f(x)$ in that, upon input of $x\in\IR$
144: given in one of the above ways, it outputs $y=f(x)$ also
145: in one (not necessarily the same) of the above ways.
146: And, again, many possible combinations have already been investigated.
147: For instance the standard notion of real function computation
148: in Recursive Analysis \cite{Grzegorczyk,PER,Ko,Weihrauch} refers
149: (or is equivalent) to input and output given according to $\myrho$.
150: Here, the \aname{Main Theorem} of Computable Analysis implies that
151: any computable $f$ will necessarily be continuous
152: \mycite{Theorem~4.3.1}{Weihrauch}.
153:
154: We are interested in ways of lifting this restriction,
155: that is, in the following
156: \begin{question}
157: Does hypercomputation in some sense permit
158: the computational evaluation of (at least certain)
159: discontinuous real functions?
160: \end{question}
161: That is related to the \aname{Church-Turing Hypothesis}:
162: A Turing Machine's ability to simulate \emph{every} physical
163: process would imply all such processes to behave continuously%
164: ---a property \person{G.~Leibniz} was convinced of
165: (``\emph{Natura non facit saltus}'')
166: but which we nowadays know to be violated for instance
167: by the \aname{Quantum Hall Effect} awarded a Nobel Prize in 1985.
168: Since this (nor any other) discontinuous physical process cannot
169: be simulated on a classical Turing Machine, it constitutes a putative
170: candidate for a system capable of realizing hypercomputation.
171:
172: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
173: \subsection{Summary}
174: The standard (and indeed the most general) way of turning a
175: Turing Machine into a hypercomputer is to grant it access
176: to an oracle like, say, the
177: Halting Problem $\emptyset'$ or its iterated jumps like $\emptyset''$
178: and $\emptyset^{(d)}$
179: in \person{Kleene}'s \aname{Arithmetic Hierarchy}.
180: However regarding computational evaluation of real functions,
181: closer inspection in Section~\ref{secOracles}
182: reveals that this Main Theorem
183: relies solely on information rather than recursion theoretic arguments
184: and therefore requires continuity also
185: for oracle-computable real functions
186: with respect to input and output of form $\myrho$.
187: (For the special case of an $\emptyset'$--oracle,
188: this had been observed in \mycite{Theorem~16}{Ho}.)
189:
190: \medskip
191: A second idea, applicable to real but not to discrete computability,
192: changes the input and output representation
193: for $x$ and $y=f(x)$ from $\myrho$ to a weaker
194: form like, say, $\myrhon$. This relates to
195: the Arithmetic Hierarchy, too, however in a %completely
196: different way: Computing $x$ in the sense
197: of $\myrhon$ is equivalent to computing
198: $x$ in the sense of $\myrho$ \mycite{Theorem~9}{Ho}
199: \emph{relative} (i.e., given access) to the
200: Halting Problem $\emptyset'$ and thus suggests
201: to write $\myrho':=\myrhon$.
202: Most promisingly, the Main Theorem
203: \mycite{Corollary~3.2.12}{Weihrauch}
204: which requires continuity of $(\myrho\to\myrho)$--computable
205: real functions applies to $\myrho$ but not to
206: $\myrho'$ because the latter lacks
207: the technical property of \emph{admissibility}.
208:
209: It therefore came to quite a surprise when
210: \person{Brattka} and \person{Hertling} established
211: that any $(\myrho'\to\myrho')$--computable $f$
212: (that is, with respect to input $x$
213: and output $f(x)$ encoded according to $\myrhon$) still
214: satisfies continuity;
215: see \cite[\textsc{Exercise~4.1.13}d]{Weihrauch}
216: or \mycite{Section~6}{Naive}.
217:
218: Section~\ref{secWeak} contains an extension of this and
219: a series of related results.
220: Specifically we manage to prove that
221: continuity is necessary for
222: $(\myrho''\to\myrho'')$--computability of $f$;
223: %with respect to $x\in\IR$ and $y=f(x)$ both given
224: %by \emph{any} representation $\myrho^{(d)}$;
225: here, ~$\myrho\:\reducneq\:
226: \myrho'\:\reducneq\:\myrho''\:\reducneq\ldots$~
227: denote the first levels of an entire hierarchy of real
228: number representations explained in Lemma~\ref{lemRepr}
229: which emerge naturally from
230: the Real Arithmetic Hierarchy of
231: \person{Weihrauch} and \person{Zheng} \cite{Zheng}.
232:
233: \medskip
234: In Section~\ref{secHierarchies}, we closer investigate
235: the two approaches to real function hypercomputation.
236: Specifically it is established (Section~\ref{secWeakly})
237: that the hierarchy of real number representation actually
238: does yield a hierarchy of weakly computable real functions.
239: Furthermore a comparison of both oracle-supported and weakly
240: computable (and each hence necessarily continuous)
241: real functions in Section~\ref{secWeier}
242: reveals a relativized version of the
243: Effective Weierstra\ss{} Theorem to fail.
244:
245: \medskip
246: Our third approach to real hypercomputation (Section~\ref{secNondet})
247: finally allows the Turing Machines
248: under consideration to behave \emph{non}deterministically.
249: Remarkably and in contrast to the classical (Type-1) theory,
250: this does significantly increase their principal capabilities.
251: For example, all quasi-strongly $\delta$--$\IQ$--analytic functions
252: in the sense of \person{Chadzelek} and \person{Hotz} \cite{Hotz}%
253: ---and in particular many discontinuous real functions---now become
254: computable as well as conversion among the
255: aforementioned representations $\myrhon$ and $\myrhob$.
256: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
257: \section{Arithmetic Hierarchy and Reals} \label{secRAH}
258: In \cite{Ho}, \person{Ho} observed an
259: interesting relation between computability of a real number $x$
260: in the respective senses of $\myrho$ and $\myrhon$ in terms of oracles:
261: $x=\lim_n q_n$ for an (eventually convergent)
262: computable rational sequence $(q_n)$ ~iff~ $x$
263: admits a \emph{fast} convergent rational sequence
264: computable with oracle $\emptyset'$,
265: that is, a sequence $(p_m)\subseteq\IQ$
266: recursive in $\emptyset'$ with $|x-p_m|\leq2^{-m}$.
267: This suggests to use $\myrho'$ synonymously for $\myrhon$;
268: and denoting by $\RDelta{1}=\IRc$ the set of
269: reals computable in the sense of Recursive Analysis
270: (that is with respect to $\myrho$), it is therefore natural
271: to write, in analogy to
272: \person{Kleene}'s classical \aname{Arithmetic Hierarchy},
273: $\RDelta{2}$ for the set of all $x\in\IR$ computable
274: with respect to $\myrho'$.
275: \person{Weihrauch} and \person{Zheng} extended
276: these considerations and obtained for instance
277: \mycite{Corollary~7.3}{Zheng} the following characterization
278: of $\RDelta{3}$: A real $x\in\IR$
279: admits a %$\emptyset''$--computable
280: fast convergent rational sequence
281: recursive in $\emptyset''$ iff $x$ is computable in the
282: sense of $\myrho''$ defined as follows:
283:
284: \smallskip
285: \noindent\begin{tabular}{rl}
286: \hspace*{1.5em}$\pmb{\myrho''}$\textbf{:} & ~ $\displaystyle
287: x=\lim\nolimits_i \lim\nolimits_j q_{\scriptscriptstyle\langle i,j\rangle}$
288: ~for some computable rational sequence $(q_n)$
289: \end{tabular}
290:
291: \smallskip\noindent
292: where $\langle\cdots\rangle:\IN^*\to\IN$
293: denotes some fixed computable pairing or, more generally,
294: tupling function.
295: Similarly, $\RSigma{1}$ contains of all $x\in\IR$
296: computable with respect to $\myrhol$
297: whereas $\RSigma{2}$ includes all $x$ computable
298: in the sense of $\myrhol'$ defined as follows:
299:
300: \smallskip
301: \noindent\begin{tabular}{rl}
302: \hspace*{1.5em}$\pmb{\myrhol'}$\textbf{:} & ~
303: $\displaystyle x=\mysup\nolimits_i \myinf\nolimits_j
304: q_{\scriptscriptstyle\langle i,j\rangle}$
305: ~for some computable rational sequence $(q_n)$.
306: \end{tabular}
307:
308: \smallskip\noindent
309: To $\RSigma{2}$ belongs for instance the radius of
310: convergence $r=1/\limsup_{n\to\infty}\sqrt[n]{a_n}$ of
311: a computable power series $\sum_{n=0}^\infty a_nx^n$
312: \mycite{Theorem~6.2}{Zheng}. More generally we
313: take from \cite[\textsc{Definition~7.1} and
314: \textsc{Corollary~7.3}]{Zheng} the following
315: \begin{definition}[Real Arithmetic Hierarchy]
316: \label{defRAH} Let $d=0,1,2,\ldots$
317: \\\begin{tabular}{rl}
318: $\pmb{\myrhol^{(d)}}$\textbf{:} &
319: $\RSigma{d+1}$ consists of all $x\in\IR$ of the form
320: ~ $\displaystyle x\;=\;
321: \mysup_{n_1} \; \myinf_{n_2} \;\ldots \; \mytheta_{n_{d+1}}
322: \;q_{\scriptscriptstyle\langle n_1,\ldots,n_{d+1}\rangle}$
323: \\[-1.1ex] & for a computable rational sequence $(q_n)$,
324: \\ & where $\mytheta\!\!=\!\mysup$ or $\mytheta\!\!=\!\myinf$
325: depending on $d$'s parity; \\
326: $\pmb{\myrhog^{(d)}}$\textbf{:} & $\RPi{d+1}$ similarly for
327: ~$\displaystyle x=\myinf_{n_1} \mysup_{n_2} \;\ldots$ \\
328: $\pmb{\myrho^{(d)}}$\textbf{:} & $\RDelta{d+1}$ contains all $x\in\IR$
329: of the form
330: \quad$\displaystyle x\;=\;
331: \lim_{n_1} \; \lim_{n_2} \;\ldots \; \lim_{n_d}
332: \;q_{\scriptscriptstyle\langle n_1,\ldots,n_d\rangle}$
333: \\[-1.1ex] & for a computable rational sequence $(q_n)$.
334: \end{tabular}
335: \end{definition}
336: (For an extension to levels beyond $\omega$ see \cite{Barmpalias}\ldots)
337:
338: %\pagebreak[2]
339: The close analogy between the discrete and this real variant of the
340: Arithmetic Hierarchy is expressed in \cite{Zheng}
341: by a variety of elegant results like, e.g.,
342: \begin{fact} \label{facRAH}
343: \begin{enumerate}
344: \item[a)] $x\in\RDelta{d}$ ~iff~ deciding its binary expansion is in
345: $\Delta_d$.
346: \item[b)] $x$ is computable with respect to $\myrho^{(d)}$ \\
347: ~iff~ there is a
348: $\emptyset^{(d)}$--computable fast convergent rational sequence
349: for $x$.% \\%\makebox[0pt][l]{$x$} iff~
350: \item[c)] $x$ is computable with respect to
351: $\myrhol^{(d)}$ \\
352: ~iff~ $x$ is the supremum of a
353: $\emptyset^{(d)}$--computable rational sequence.
354: \item[d)] $\RDelta{d}\;=\;\RSigma{d}\,\cap\,\RPi{d}$.
355: \item[e)] $\RSigma{d}\,\cup\,\RPi{d}\;\subsetneq\;\RDelta{d+1}$.
356: \end{enumerate}
357: \end{fact}
358: \begin{proof}
359: %See \cite{Zheng}.
360: a) \textsc{Theorem~7.8},
361: b+c) \textsc{Lemma~7.2},
362: d) \textsc{Definition~7.1},
363: and e) \textsc{Theorem~7.8}
364: in \cite{Zheng}, respectively. \qed
365: \end{proof}
366: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
367: \subsection{Type-2 Theory of Effectivity}
368: Specifying an encoding formalizes how to feed some general form of input
369: like graphs or integers into a Turing Machine with fixed alphabet $\Sigma$.
370: Already in the discrete case, the complexity of a problem usually depends
371: heavily on the chosen encoding; e.g., numbers in unary versus binary.
372: This dependence becomes even more important when dealing with objects
373: from a continuum like the set of reals and their computability.
374: While Recursive Analysis usually considers one particular encoding
375: for $\IR$, the Type-2 Theory of Effectivity (TTE) due to
376: \person{Weihrauch} provides
377: (a convenient formal framework for studying and comparing)
378: a variety of encodings for different universes.
379: Formally speaking, a \emph{representation} $\alpha$ for $\IR$
380: is a partial\footnote{indicated by the symbol ``$\subseteq$'',
381: whose absence here generally refers to total functions}
382: surjective mapping $\alpha:\subseteq\Sigma^\omega\to\IR$;
383: and an infinite string $\bar\sigma\in\dom(\alpha)$ is
384: regarded as an $\alpha$--name for the real number $x=\alpha(\bar\sigma)$.
385:
386: In this way, $(\alpha\to\beta)$--computing\footnote{We use this
387: notation instead of \cite{Weihrauch}'s $(\alpha,\beta)$--computability
388: to stress its connection (but not to be confused) with
389: $[\alpha\myto\beta]$--computability appearing in Section~\ref{secWeier}.}
390: a real function $f:\IR\to\IR$ means to compute a transformation
391: on infinite strings $F:\subseteq\Sigma^\omega\to\Sigma^\omega$
392: such that any $\alpha$--name $\bar\sigma$ for $x=\alpha(\bar\sigma)$
393: gets transformed to a $\beta$--name $\bar\tau=F(\bar\sigma)$
394: for $f(x)=y$, that is, satisfying $\beta(\bar\tau)=y$;
395: cf. \mycite{Section~3}{Weihrauch}.
396: Converting
397: $\alpha$--names to $\beta$--names thus amounts to
398: $(\alpha\to\beta)$--computability of $\id:\IR\to\IR$, $x\mapsto x$,
399: and is called \aname{reducibility} ``$\alpha\reduceq\beta$''
400: \mycite{Definition~2.3.2}{Weihrauch}.
401: Computational \aname{equivalence}, that is mutual
402: reducibility ~$\alpha\reduceq\beta$
403: and $\beta\reduceq\alpha$,
404: is denoted by ``$\alpha\equiv\beta$''
405: whereas ``$\alpha\reducneq\beta$'' means
406: $\alpha\reduceq\beta$ but $\beta\not\reduceq\alpha$.
407:
408: We borrow from TTE also two ways of constructing new representations
409: from giving ones: The \aname{conjunction} $\alpha\myjoin\tilde\alpha$
410: of $\alpha$ and $\tilde\alpha$ is the least upper bound
411: with respect to ``\,$\reduceq\!$'' \mycite{Lemma~3.3.8}{Weihrauch};
412: and for (finitely or countably many) representations
413: $\alpha_i:\subseteq\Sigma^\omega\to A_i$,
414: their \aname{product}
415: $\prod_i \alpha_i$ denotes a natural representation
416: for the set $\prod_i A_i$ \mycite{Definition~3.3.3.2}{Weihrauch}.
417: In particular, in order to encode $x\in\IR$ as a rational sequence
418: $(q_n)\in\IQ^\omega$, we (often implicitly) refer to the representation
419: $[\nuQ]^\omega:\subseteq\Sigma^\omega\to\IQ^\omega$ due to
420: \cite[\textsc{Definition~3.1.2.4} and \textsc{Lemma~3.3.16}]{Weihrauch}.
421:
422: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
423: \subsection{Arithmetic Hierarchy of Real Representations}
424: Observe that (the characterizations due to Fact~\ref{facRAH} of)
425: each level of the Real Arithmetic Hierarchy gives rise not only
426: to a notion of computability for real numbers
427: but also canonically to a representation for $\IR$;
428: for instance let
429:
430: \smallskip
431: \noindent\begin{tabular}{l@{\,:\;\:}l}
432: $\pmb{\myrho}$ & encode (arbitrary!)
433: $x\in\IR$ as a fast convergent rational sequence $(q_n)$; \\[0.2ex]
434: $\pmb{\myrhol}$ & encode
435: $x\in\IR$ as a rational sequence $(q_n)$ with supremum $x=\sup_n q_n$; \\[0.2ex]
436: $\pmb{\myrho'}$ & encode
437: $x\in\IR$ as a rational sequence $(q_n)$ with limit $x=\lim_n q_n$; \\[0.2ex]
438: $\pmb{\myrhol'}$ & encode
439: $x\in\IR$ as $(q_n)\subseteq\IQ$ with
440: $x=\mysup_i \myinf_j q_{\scriptscriptstyle\langle i,j\rangle}$; \\[0.2ex]
441: $\pmb{\myrho''}$ & encode
442: $x\in\IR$ as $(q_n)\subseteq\IQ$ with
443: $x=\lim_i \lim_j q_{\scriptscriptstyle\langle i,j\rangle}$.
444: \end{tabular}
445:
446: \medskip
447: \noindent
448: As already pointed out, the first three of them are already
449: known and used in TTE as $\myrho$, $\myrhol$, and $\myrhon$,
450: respectively \mycite{Section~4.1}{Weihrauch}.
451: In general one obtains, similar to Definition~\ref{defRAH},
452: a hierarchy of real representations as follows:
453: \begin{definition} \label{defRepr}
454: Let ~$\myrho^{(0)}:=\myrho$, ~$\myrhol^{(0)}:=\myrhol$,
455: ~$\myrhog^{(0)}:=\myrhog$. ~ Now fix $1\leq d\in\IN$: \\
456: A $\myrho^{(d)}$--name for $x\in\IR$ is
457: (a $[\nuQ]^\omega$--name for) a rational sequence
458: $(q_n)$ such that%
459: $$\displaystyle x\;=\;
460: \lim_{n_1} \; \lim_{n_2} \;\ldots \; \lim_{n_d}
461: \;q_{\scriptscriptstyle\langle n_1,\ldots,n_d\rangle}
462: \enspace . $$
463: A $\myrhol^{(d)}$--name for $x\in\IR$ is
464: a (name for a) sequence
465: $(q_n)\subseteq\IQ$ such that
466: $$\displaystyle x\;=\;
467: \mysup_{n_1} \; \myinf_{n_2} \;\ldots \; \mytheta_{n_{d+1}}
468: \;q_{\scriptscriptstyle\langle n_1,\ldots,n_{d+1}\rangle}
469: \enspace .$$
470: A $\myrhog^{(d)}$--name for $x\in\IR$ is a sequence
471: $(q_n)\subseteq\IQ$ such that %, similarly,
472: $\displaystyle x\;=\;
473: \myinf_{n_1} \; \mysup_{n_2} \;\ldots$
474: \end{definition}
475:
476: \noindent
477: Regarding Fact~\ref{facRAH}, one may see $\myrho'$ and $\myrho''$
478: as the first and second \emph{Jump} of $\myrho$,
479: respectively; same for $\myrhol'$ and $\myrhol$.
480:
481: Results from \cite{Zheng} about the Real Arithmetic Hierarchy
482: are easily re-phrased in terms of these representations.
483: Fact~\ref{facRAH}d) for example translates as follows:
484: \begin{quote}
485: $x$ is $\myrho^{(d)}$--computable ~ iff ~ it is both
486: $\myrhol^{(d)}$--computable and $\myrhog^{(d)}$--computable.
487: \end{quote}
488: Observe that this is a non-uniform claim whereas
489: closer inspection of the proofs in particular of
490: \textsc{Lemma~3.2} and \textsc{Lemma~3.3} in \cite{Zheng}
491: reveals them to hold fully uniformly so that we have
492: \begin{lemma} \label{lemRepr}
493: \;\; %It holds $
494: $\displaystyle\myrho\;\equiv\;\;
495: \myrhol\!\myjoin\myrhog\;\;\reducneq\;\;
496: \myrhol\;\;\reducneq\;\;\myrho'\;\,\equiv\;\;\,
497: \myrhol'\!\myjoin\myrhog'\;\;\reducneq\;\;
498: \myrhol'\;\;\reducneq\;\;\myrho''
499: \;\equiv\;\ldots\;$.
500: \end{lemma}
501:
502: \noindent
503: Moreover, the uniformity of \mycite{Lemma3.2}{Zheng} yields
504: the following interesting
505: \begin{scholiumf}\footnotetext{A \aname{scholium} is ``\emph{a
506: note amplifying a proof or course of reasoning,
507: as in mathematics}'' \cite{Dictionary}} \label{schLiminf}
508: Let $\tildemyrhol'$ denote the representation encoding
509: $x\in\IR$ as $(q_n)\subseteq\IQ$ with $x=\liminf_n q_n$;
510: and $\hatmyrhol'$ similarly with the additional
511: requirement that $q_n<x$ for infinitely many $n$. \\
512: Then it holds ~$\hatmyrhol'\equiv\tildemyrhol'\equiv\myrhol'$~
513: ($\hatmyrhol'\reduceq\tildemyrhol'\reduceq\,\myrhol'$~
514: being the trivial direction).
515: \end{scholiumf}
516: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
517: \section{Computability and Continuity} \label{secContinuous}
518: Recursive Analysis has established as folklore that any
519: computable real function is continuous.
520: More precisely, computability of a partial function
521: from/to infinite strings $f:\subseteq\Sigma^\omega\to\Sigma^\omega$
522: requires continuity with respect to the Cantor Topology $\tau_{\text{C}}$
523: \mycite{Theorem~2.2.3}{Weihrauch};
524: and this requirement carries over to functions
525: $f:\subseteq A\to B$ on other topological spaces
526: $(A,\tau_A)$ and $(B,\tau_B)$
527: where input $a\in A$ and output $b=f(a)$
528: are encoded by respective \emph{admissible} representations
529: $\alpha$ and $\beta$. Roughly speaking, this
530: property expresses that the mappings
531: $\alpha:\subseteq\Sigma^\omega\to A$ and
532: $\beta:\subseteq\Sigma^\omega\to B$
533: satisfy a certain compatibility condition
534: with respect to the topologies $\tau_A$/$\tau_B$
535: and $\tau_{\text{C}}$ involved.
536: For $A=B=\IR$, the (standard) representation
537: $\myrho$ for example is admissible \mycite{Lemma~4.1.4.1}{Weihrauch},
538: thus recovering the folklore claim.
539:
540: Now in order to treat and non-trivially investigate computability
541: also of discontinuous real functions $f:\IR\to\IR$,
542: there are basically two ways out:
543: Either enhance the underlying Type-2 Machine model
544: or resort to non-admissible representations.
545: It turns out that for either choice, at least the
546: straight-forward approaches fail: \\ \noindent
547: \textbullet ~ extending Turing Machines with oracles
548: \quad as well as \\ \noindent
549: \textbullet ~ considering weakened representations for $\IR$.
550: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
551: \subsection{Type-2 Oracle Computation} \label{secOracles}
552: Specifically concerning the first approach,
553: most results in Computable Analysis relativize.
554: Specifically we make
555: \begin{observation} \label{obsRelative}
556: Let $\calO\subseteq\Sigma^*$ be arbitrary.
557: Replace in {\rm\mycite{Definition~2.1.1}{Weihrauch}}
558: the Turing Machine $\calM$ by $\calM^\calO$, that is,
559: one with oracle access to $\calO$. This
560: Type-2 Computability \emph{in $\calO$} still satisfies%
561: \begin{enumerate}
562: \item[a)]
563: closure under composition
564: {\rm\mycite{Theorem~2.1.12}{Weihrauch}};
565: \item[b)]
566: computability of string functions requires continuity
567: {\rm\mycite{Theorem~2.2.3}{Weihrauch}};
568: \item[c)]
569: same for computable functions on represented spaces
570: with respect to \emph{admissible} representations
571: {\rm\mycite{Corollary~3.2.12}{Weihrauch}}.
572: \end{enumerate}
573: In particular, the Main Theorem of Recursive Analysis
574: {\rm\mycite{Theorem~4.3.1}{Weihrauch}}
575: relativizes.
576: \end{observation}
577: A strengthening and counterpart to
578: Observation~\ref{obsRelative}b), we have
579: \begin{lemma} \label{lemRelative}
580: For a partial function on infinite strings
581: ~$f:\subseteq\Sigma^\omega\to\Sigma^\omega$,
582: the following are equivalent:
583: \begin{itemize}
584: \item[\textbullet] There exists an oracle $\calO$
585: such that $f$ is computable relative to $\calO$;
586: \item[\textbullet] $f$ is Cantor-continuous
587: and $\dom(f)$ is a $\Gdelta$--set.
588: \end{itemize}
589: \end{lemma}
590: Compare this with Type-1 Theory (that is,
591: computability on finite strings) where every function
592: $f:\subseteq\Sigma^*\to\Sigma^*$ is recursive
593: in some appropriate $\calO\subseteq\Sigma^*$.
594: \begin{proof}[Lemma~\ref{lemRelative}]
595: If $f$ is recursive in $\calO$, then it is also continuous
596: by Observation~\ref{obsRelative}b), that is,
597: the relativized version of \mycite{Theorem~2.2.3}{Weihrauch}.
598: Furthermore the relativization of \mycite{Theorem~2.2.4}{Weihrauch}
599: reveals $\dom(f)$ to be a $\Gdelta$--set.
600:
601: Conversely suppose that continuous $f$ has $\Gdelta$ domain.
602: Then $f=h_\omega$ for some monotone total function
603: $h:\Sigma^*\to\Sigma^*$ according to
604: \mycite{Theorem~2.3.7.2}{Weihrauch} where, by
605: \mycite{Definition~2.1.10.2}{Weihrauch},
606: $h_\omega:\subseteq\Sigma^\omega\ni\bar\sigma\mapsto
607: \sup_{n} h(\sigma_1\ldots\sigma_n)$ denotes the (existing
608: and unique) extension of $h$ from $\Sigma^*$ to $\subseteq\Sigma^\omega$.
609: A classical Type-1 function on finite strings, this $h$
610: is recursive in a certain oracle $\calO\subseteq\Sigma^*$.
611: The relativization of \mycite{Lemma~2.1.11.2}{Weihrauch}
612: then asserts also $h_\omega=f$ to be computable in $\calO$.
613: \COMMENTED{
614: To this end let
615: $\lcp(S)\in\Sigma^*$ denote the \emph{longest common prefix}
616: of a set $S\subseteq\Sigma^\omega$ of at least
617: $2\leq\Card{S}$ infinite strings,
618: $\lcp(\{\bar\sigma\}):=\bar\sigma\in\Sigma^\omega$.
619: Now define $\tilde h:\subseteq\Sigma^*\to\Sigma^*\cup\Sigma^\omega$ by
620: $$ \tilde h(a) \;:=\; \left\{
621: \begin{array}{ll} \bot & \;\text{ if } a\Sigma^\omega\cap\dom(f)=\emptyset
622: \\[0.5ex]
623: \lcp\big(f[a\Sigma^\omega\cap\dom(f)]\big)_{\leq n(a)} & \;\text{ otherwise}
624: % \min\{n(a),|a|\}}
625: \end{array}\right. $$
626: with the abbreviation
627: $n(a):=\sup\{n\in\IN: a\Sigma^\omega\subseteq U_n\}\in\IN\cup\{\infty\}$
628: for $a\in\Sigma^*$. %It follows
629: \begin{enumerate}
630: \item[a)]
631: If $a\substr a'\in\Sigma^*$, then
632: $a\Sigma^\omega\supseteq a'\Sigma^\omega$.
633: \item[b)]
634: Consequently, $a\in\dom(\tilde h)$ whenever
635: $a\substr a'$ and $a'\in\dom(\tilde h)$.
636: In particular, $\lambda\in\dom(\tilde h)$ and $\tilde h(\lambda)=\lambda$.
637: \item[c)]
638: For $a\substr a'\in\dom(f)$,
639: $\tilde h(a)\substr\tilde h(a')$, i.e.,
640: $\tilde h$ is monotone on its domain.
641: \\
642: Indeed,
643: $A:=f[a\Sigma^\omega\cap\dom(f)]\supseteq A':=f[a'\Sigma^\omega\cap\dom(f)$
644: yields $\lcp(A)\substr\lcp(A')$, and
645: $b\substr b'$ with $n:=n(a)\leq n(a')=:n'$ implies
646: $b_{\leq n}\substr b'_{\leq n'}$.
647: \item[d)]
648: For every $\bar\sigma\in\dom(f)$,
649: $\sup_n \tilde h(\bar\sigma_{<n})=f(\bar\sigma)$.
650: \end{enumerate}
651: To see the latter, let $b\substr f(\bar\sigma)\in\Sigma^\omega$
652: denote a finite prefix of length $m$. By continuity of $f$,
653: $f^{-1}[b\Sigma^\omega]$ is an open
654: subset of $\dom(f)$ containing $\bar\sigma$ and thus
655: an entire ball in the relative topology, that is,
656: $\bar\sigma\in a\Sigma^\omega\cap\dom(f)\subseteq f^{-1}[b\Sigma^\omega]$
657: with $a:=\bar\sigma_{\leq n}$ for any sufficiently large $n$;
658: in particular $a\in\dom(\tilde h)$.
659: By possibly further increasing $n$, we may even suppose that
660: $a\Sigma^\omega\subseteq U_m$ because $\bar\sigma\in U_m$ open,
661: so $n(a)\geq m$.
662: Then $f[a\Sigma^\omega\cap\dom(f)]\subseteq b\Sigma^\omega$, hence
663: $$ \tilde h(a)\;=\;\lcp\big(f[a\Sigma^\omega\cap\dom(f)]\big)_{\leq n(a)}
664: \;\;\supstr\quad\lcp(b\Sigma^\omega)_{\leq n(a)}
665: \;\;=\; b$$
666: since $n(a)\geq m=|b|$.
667: \begin{enumerate}
668: \item[e)]
669: Claim~d) includes that $\tilde h$ be unbounded (or even infinite) on every
670: $\bar\sigma\in\dom(f)=\bigcup_n U_n$.
671: If on the other hand $\bar\sigma\not\in U_N$ for some $N$,
672: then $a\Sigma^\omega\not\subseteq U_n$ for all $n\geq N$
673: and $a\substr\bar\sigma$, so $n(a)<N$.
674: Hence, if $\tilde h$ is at all
675: defined on those $a$, then it is still bounded on $\bar\sigma$.
676: \end{enumerate}
677: The total function to finite strings $h:\Sigma^*\to\Sigma^*$
678: is finally obtained as $$h(a):=\tilde h
679: \big(\max\{\tilde a\substr a:\tilde a\in\dom(\tilde h)\}\big)_{\leq|a|}
680: \enspace . $$
681: Because of b), this preserves monotonicity c)
682: while maintaining properties d) and e),
683: that is, establishing $h_{\omega}=f$.
684: }
685: \qed
686: \end{proof}
687: The conclusion of this subsection is that
688: oracles do not increase the computational power
689: of a Type-2 Machine sufficiently in order to handle
690: also discontinuous functions. So let us proceed
691: to the second approach to real hypercomputation:
692: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
693: \subsection{Weaker Representations for Reals} \label{secWeak}
694: In the present section we are interested in
695: relaxations of the standard representation $\myrho$
696: for single reals
697: and their effect on the computability of function
698: evaluation $x\mapsto f(x)$.
699: Since, with exception of $\myrho$, none of the
700: ones introduced in Definition~\ref{defRepr} is admissible
701: with respect to the usual Euclidean\footnote{%
702: it might be admissible w.r.t. some other, typically artificial
703: topology, though} topology on $\IR$
704: \mycite{Lemma~4.1.4, Example~4.1.14.1}{Weihrauch},
705: the relativized Main Theorem (Observation~\ref{obsRelative}c)
706: is not applicable. Hence,
707: chances are good for evaluation $x\mapsto f(x)$ to become computable
708: even for discontinuous $f:\IR\to\IR$; and indeed we have the following
709: \\[1ex]\begin{minipage}[c]{60ex}
710: \begin{example} \label{exHeaviside}
711: \person{Heaviside}'s function \\[0.7ex]
712: \centerline{$\displaystyle\Heavi:\IR\to\IR, \qquad x\mapsto 0 \text{ ~for~ }x\leq0,
713: \quad x\mapsto 1 \text{ ~for~ }x>0$}
714: \\[0.5ex]
715: is both $(\myrhol\to\myrhol)$--computable
716: and $(\myrhol'\to\myrhol')$--computable.%
717: \end{example}
718: \end{minipage}\hfill\begin{minipage}[c]{22ex}
719: \includegraphics[width=\textwidth]{heaviside}%
720: \end{minipage}
721: \begin{proof}
722: Given $(q_n)\subseteq\IQ$ with $x=\sup_n q_n$,
723: exploit $(\nuQ\to\nuQ)$--computability
724: of the restriction $\Heavi|_{\IQ}:\IQ\to\{0,1\}$ to
725: obtain $p_n:=\Heavi(q_n)$. Then indeed, $(p_n)\subseteq\IQ$ has
726: $\sup_n p_n=\Heavi(x)$: In case $x\leq0$, $q_n\leq0$ and hence
727: $p_n=0$ for all $n$; whereas in case $x>0$, $q_n>0$
728: and hence $p_n=1$ for some $n$.
729:
730: Let $x\in\IR$ be given by a rational double sequence
731: $(q_{i,j})$ with $x=\mysup_i\myinf_j q_{i,j}$.
732: Proceeding from $q_{i,j}$ to
733: $\tilde q_{i,j}:=\max\{q_{0,j},\ldots,q_{i,j}\}$,
734: we assert $\myinf_j\tilde q_{i+1,j}\geq\myinf_j\tilde q_{i,j}$.
735: Now compute $p_{i,j}:=\Heavi(\tilde q_{i,j}-2^{-i})$. Then in case $x\leq0$,
736: it holds
737: $\forall i \exists j: \tilde q_{i,j}\leq 2^{-i}$,
738: i.e., $p_{i,j}=0$
739: and thus $\mysup_i\myinf_j p_{i,j}=0=\Heavi(x)$. Similarly in case
740: $x>0$, there is some $i_0$ such that
741: $\myinf_j \tilde q_{i_0,j}>x/2$ and thus
742: $\myinf_j \tilde q_{i,j}>x/2$ for all $i\geq i_0$.
743: For $i\geq i_0$ with $2^{-i}\leq x/2$,
744: it follows $p_{i,j}=1$ $\forall j$
745: and therefore $\mysup_i\myinf_j p_{i,j}=1=\Heavi(x)$.
746: \qed
747: \end{proof}
748: %
749: So real function hypercomputation based on weaker representations
750: indeed does allow for effective evaluation of some discontinuous
751: functions. On the other hand, they still impose well-known
752: topological restrictions:
753: \begin{fact} \label{facFolklore}
754: Consider $f:\IR\to\IR$.%
755: \begin{itemize}
756: \item[a)]
757: If $f$ is ~$(\myrho\to\myrho)$--computable,
758: then it is continuous.
759: \item[b)]
760: If $f$ is ~$(\myrho\to\myrhol)$--computable,
761: then it is lower semi-continuous.
762: \item[c)]
763: If $f$ is ~$(\myrhol\to\myrhol)$--computable,
764: then it is monotonically %(though not strictly)
765: increasing.
766: \item[d)]
767: If $f$ is ~$(\myrho'\to\myrho')$--computable,
768: then it is continuous.
769: \end{itemize}
770: The claims remain valid under oracle-supported computation.
771: \end{fact}
772: Claim~a) is the Main Theorem.
773: For b) see \cite{SemiTCS} and recall,
774: e.g. from \mycite{Chapter~6.7}{Analysis},
775: that $f:\IR\to\IR$ is lower semi-continuous iff
776: $f(\lim_n x_n)\leq\liminf_n f(x_n)$
777: for all convergent sequences $(x_n)$;
778: equivalently: $f^{-1}\big[(y,\infty)\big]$
779: is open for any $y\in\IR$.
780: The establishing of d) in \mycite{Section~6}{Naive}
781: caused some surprise.
782: We briefly sketch the according proofs as a preparation
783: for those of Theorem~\ref{thNecessary} below.
784: \begin{proof}
785: \begin{enumerate}
786: \item[a)] Suppose for a start that Heaviside's function,
787: in spite of its discontinuity at $x=0$, be $(\myrho\to\myrho)$--computable
788: by some Type-2 Machine $\calM$.
789: Feed the rational sequence $q_n:=2^{-n}$,
790: a valid $\myrho$--name for $x$, to this $\calM$.
791: By presumption it will then spit out a sequence
792: $(p_m)_{_m}\subseteq\IQ$ with $|p_m-y|\leq 2^{-m}$ for $y=\Heavi(x)=0$;
793: in particular, $|p_2-\tilde y|>2^{-2}$ for $\tilde y:=1$.
794: Up to output of $p_2$, $\calM$ has executed a finite number $N\in\IN$ of
795: operations and in particular read at most the initial part
796: $p_0,p_1,\ldots,p_N$ of the input.
797:
798: Now re-use $\calM$ in order to evaluate $\Heavi$ at $\tilde x:=p_N>0$
799: $\myrho$--encoded as the rational sequence
800: $(\tilde q_n):=(q_0,q_1,\ldots,q_N,q_N,\ldots)$
801: coinciding with $(p_n)$ for $n\leq N$.
802: Being a deterministic machine, $\calM$ will then proceed
803: exactly as before for its first $N$ steps;
804: in particular the output $(\tilde p_m)$ agrees with $(p_m)$ up to $m=2$.
805: Hence $|\tilde p_2-\tilde y|>2^{-2}$
806: contradicting that $\calM$ is supposed to output a
807: $\myrho$--name for $\tilde y=\Heavi(\tilde x)$.
808:
809: \smallskip
810: For the case of a general function $f:\IR\to\IR$ with
811: discontinuity at some $x\in\IR$, let
812: $y=f(x)\not=\lim_k f(x_k)=\tilde y$ with a
813: real sequence $x_k$ converging to $x$.
814: There exists $M\in\IN$ with $|y-\tilde y|>2^{-M+2}$;
815: by possibly proceeding to an appropriate subsequence of $(x_k)$,
816: we may suppose w.l.o.g. that $|x-x_k|\leq2^{-k-2}$
817: and $|f(x_k)-\tilde y|\leq2^{-M}$.
818: Then there is a rational double sequence $(q_{k,n})$ such that
819: $|x_k-q_{k,n}|\leq2^{-n-1}$; thus $|x-q_{n,n}|\leq2^{-n}$.
820: We may therefore feed $(q_{n,n})$ as a $\myrho$--name in order
821: to evaluate $f$ at $x$ and obtain in turn a $\myrho$--name
822: $(p_m)\subseteq\IQ$ for $y$.
823: As before, $p_M$ is output after having only read
824: some finite initial part $(q_{n,n})_{_{n\leq N}}$ of the
825: input. Then
826: $$|q_{n,n}-x_N|\leq|q_{n,n}-x_n|+|x_n-x|+|x-x_N|
827: \leq 2^{-n-1}+2^{-n-2}+2^{-N-2}\leq2^{-n}$$
828: for $n\leq N$
829: reveals this very initial part to also be the start
830: of a valid $\myrho$--name for $\tilde x:=x_N$ whereas
831: $$2^{-M+2}<|y-\tilde y|
832: \leq |y-p_M|+|p_M-f(\tilde x)|+|f(\tilde x)-\tilde y|
833: \leq 2^{-M} + |p_M-f(\tilde x)|+ 2^{-M}$$
834: shows %implies $|p_M-f(\tilde x)|>2^{-M}$, showing
835: that $(p_m)_{m\leq M}$ is not a valid
836: initial part of a $\myrho$--name for $f(\tilde x)$: contradiction.
837: \item[b)]
838: We prove $(\myrho\to\myrhol)$--uncomputability of the flipped
839: Heaviside Function%
840: \[\barH:0\geq x\mapsto 1, \qquad 0<x\mapsto 0 \qquad\qquad
841: \smash{\raisebox{-5ex}{\includegraphics[height=9ex]{rheaviside}}} \]
842: as a prototype lacking lower semi-continuity.
843: \\
844: Consider again the $\myrho$--name $q_n:=2^{-n}$ for
845: $x=0$ which the hypothetical Type-2 Machine transforms
846: into a $\myrhol$--name for $y=\barH(x)=1$, that is,
847: a sequence $(p_m)\subseteq\IQ$ with $\sup_m p_m=y$;
848: In particular $p_M\geq\tfrac{2}{3}$ for some $M\in\IN$
849: gets output
850: having read only $(q_n)_{_{n\leq N}}$ for some $N\in\IN$.
851: The latter finite segment is also the initial part of
852: a valid $\myrho$--name for $\tilde x=q_N>0$ whereas
853: $(p_m)_{_{m\leq M}}$ has $\sup\geq\tfrac{2}{3}$ and
854: thus is not the initial part of a valid $\myrhol$--name
855: for $\tilde y=\barH(\tilde x)=0$: contradiction.
856:
857: This proof for the case $\barH$ carries over to
858: an arbitrary $f:\IR\to\IR$
859: just like in a), that is, by replacing $q_n=2^{-n}$
860: with rational approximations to a general sequence
861: $x_n\in\IR$ witnessing violated lower semi-continuity
862: of $f$ in that $f(\lim_n x_n)>\liminf_n f(x_n)$.
863: \item[c)]
864: As in a) and b), we treat for notational simplicity
865: the case of $f:\IR\to\IR$ violating
866: monotonicity in that $f(0)=1$ and $f(1)=0$;
867: the general case can again be handled similarly.
868: Feed the $\myrhol$--name $(q_n)=(0,0,\ldots)$ for $x=0$
869: into a machine which be presumption produces a sequence
870: $(p_m)\subseteq\IQ$ with $\sup p_m=1$ and in particular
871: $p_M\geq\tfrac{2}{3}$ for some $M\in\IN$.
872: Up to output of $p_M$, only $(q_n)_{_{n\leq N}}$ has
873: been read for some $N\in\IN$. Now consider
874: the rational sequence $(\tilde q_n)$ consisting
875: of $N$ zeros followed by an infinity of 1s,
876: that is, a valid $\myrhol$--name for $\tilde x=1$.
877: This new input will cause the machine to output
878: a sequence $(\tilde p_m)\subseteq\IQ$ coinciding
879: with $(p_m)$ for $m\leq M$; in particular
880: $\tilde p_M\geq\tfrac{2}{3}$ contradicting that
881: $(\tilde p_m)$ is supposed to satisfy
882: $\sup_m\tilde p_m=f(\tilde x)=0$.
883: \item[d)]
884: Suppose that, in spite of its discontinuity at $x=0$,
885: $\barH$ be $(\myrho'\to\myrho')$--computable by some
886: Type-2 Machine $\calM$.
887:
888: Consider the sequence $q^{(1)}:=\big(q^{(1)}_n\big)\subseteq\IQ$,
889: $q^{(1)}_n:\equiv1$, which is by definition a valid $\myrho'$--name
890: for $1=:x^{(1)}=\lim_n q^{(1)}_n$.
891: So upon input of $q^{(1)}$, $\calM$ will generate
892: a corresponding sequence $p^{(1)}\subseteq\IQ$ as a
893: $\myrho'$--name for $y^{(1)}=\barH\big(x^{(1)}\big)=0$,
894: that is, satisfying $\lim_m p^{(1)}_m=0$; in particular,
895: $p^{(1)}_{m_1}\leq\tfrac{1}{3}$ for some $m_1\in\IN$.
896: Up to this output, $\calM$ has read only a finite initial part
897: of the input $q^{(1)}$, say, up to $n\leq n_1$.
898:
899: Next consider the sequence $q^{(2)}\subseteq\IQ$ defined by
900: $q^{(2)}_n:=1$ for $n\leq n_1$ and $q^{(2)}_n:=\tfrac{1}{2}$
901: for $n>n_1$: a valid $\myrho'$--name for $x^{(2)}=\tfrac{1}{2}$
902: which $\calM$ by presumption transforms into a sequence $p^{(2)}\subseteq\IQ$
903: with $\lim_m p^{(2)}_m=y^{(2)}=\barH\big(x^{(2)}\big)=0$;
904: in particular, $q^{(2)}_{m_2}\leq\tfrac{1}{3}$ for some
905: $m_2>m_1$. However, due to $\calM$'s deterministic behavior
906: and since $q^{(1)}$ and $q^{(2)}$ initially coincide,
907: it still holds $p^{(2)}_{m_1}\leq\tfrac{1}{3}$.
908:
909: Now by repeating the above argument we obtain a sequence
910: of sequences $q^{(k)}\subseteq\IQ$, each constant for $n\geq n_k$
911: of value (and thus a valid $\myrho'$--name for) $x^{(k)}=2^{-k+1}$
912: and transformed by $\calM$ into a sequence $p^{(k)}\subseteq\IQ$
913: satisfying $p^{(k)}_{m_i}\leq\tfrac{1}{3}$ for $i=1,\ldots k$
914: with strictly increasing $(n_k),(m_k)\subseteq\IN$.
915: The ultimate sequence $q^{(\omega)}\subseteq\IQ$, well-defined by
916: $q^{(\omega)}_n:=q^{(k)}_n$ for $n\leq n_k$
917: (and in fact the limit of the sequence of sequences
918: $\big(q^{(k)}\big)_{_k}$ with respect to Baire's Topology),
919: therefore converges
920: to (and is hence a valid $\myrho'$--name for) $x^{(\omega)}=0$;
921: and it gets mapped by $\calM$ to a sequence $q^{(\omega)}\subseteq\IQ$
922: satisfying $q^{(\omega)}_m\leq\tfrac{1}{3}$ for infinitely
923: many $m$ contradicting that a valid $\myrho'$--name
924: for $y^{(\omega)}=\barH\big(x^{(\omega)}\big)=1$
925: should have $\lim_m=1$.
926: \end{enumerate}
927: \noindent
928: Being only information-theoretic,
929: the above arguments obviously relativize.%
930: \qed\end{proof}
931: %
932: The main result of the present section is an extension
933: of Fact~\ref{facFolklore}
934: to one level up on the hierarchy of real representations
935: from Definition~\ref{defRepr}. This suggests similar
936: claims to hold for the entire hierarchy and might not
937: be as surprising any more as Fact~\ref{facFolklore}d) in \cite{Naive};
938: nevertheless, already this additional step makes
939: proofs significantly more involved.
940: \begin{theorem}[First Main Theorem of Real Hypercomputation]
941: \label{thNecessary} ~ \\ Consider $f:\IR\to\IR$.
942: \begin{itemize}
943: \item[a)]
944: If $f$ is ~$(\myrho'\to\myrhol')$--computable,
945: then it is lower semi-continuous.
946: \item[b)]
947: If $f$ is ~$(\myrhol'\to\myrhol')$--computable,
948: then it is monotonically increasing.
949: \item[c)]
950: If $f$ is ~$(\myrho''\to\myrho'')$--computable,
951: then it is continuous.
952: \end{itemize}
953: The claims remain valid under oracle-supported computation.
954: \end{theorem}
955: We point out that the proofs of Fact~\ref{facFolklore} proceed
956: by constructing an input for which a presumed machine $\calM$ fails to
957: produce the correct output. They differ however in the `length'
958: of these constructions: for Claims~a) to c), the counter-example
959: inputs are obtained by running $\calM$ for a finite number of steps
960: on a single, fixed argument; whereas in the proof of Claim~d),
961: $\calM$ is repeatedly started on an adaptively extended sequence of
962: arguments. The latter argument may thus be considered as of
963: length $\omega$, the first infinite ordinal.
964: Our proof of Theorem~\ref{thNecessary}c) will be even longer
965: and is therefore put into the following subsection.
966: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
967: \subsection{Proof of Theorem~\ref{thNecessary}} \label{secProof}
968: As in the proof of Fact~\ref{facFolklore},
969: we treat the special case of the flipped
970: Heaviside Function $\barH$ for reasons of
971: notational convenience and clarity of
972: presentation; the according arguments
973: can be immediately extended to the general case.
974: \begin{myclaim} \label{clOne}
975: $\barH:\IR\to\IR$ is not $(\myrho'\to\myrhol')$--computable.
976: \end{myclaim}
977: \begin{proof}
978: Suppose a Type-2 Machine $\calM$
979: $(\myrho'\to\myrhol')$--computes $\barH$.
980: In particular, upon input of $x^{(1)}=1$ in form of
981: the sequence $q^{(1)}=(q^{(1)}_n)$ with $q^{(1)}_n:\equiv1$, $\calM$ will output
982: a rational double sequence $p^{(1)}=(p^{(1)}_{k,\ell})$ with
983: $0=y^{(1)}:=\barH(x^{(1)})=
984: \mysup\limits_k\myinf\limits_\ell p^{(1)}_{k,\ell}$.
985: Observe that $p^{(1)}_{1,\ell_1}\leq\tfrac{1}{3}$ for some
986: $\ell_1$. When writing $p^{(1)}_{1,\ell_1}$, $\calM$ has only
987: read a finite part of $(q^{(1)}_n)$, say, up to $n_1$.
988:
989: Now consider $x^{(2)}:=\tfrac{1}{2}$, given by way of
990: the sequence $q^{(2)}$ with
991: \quad $q^{(2)}_n:=1$ ~ for ~ $n<n_1$ \quad and\quad
992: $q^{(2)}_n:=\tfrac{1}{2}$ ~ for ~ $n\geq n_1$.
993: Then, too, $\calM$ will output a double sequence $p^{(2)}$
994: with $0=y^{(2)}=\mysup\limits_k\myinf\limits_\ell p^{(2)}_{k,\ell}$.
995: Observe that, similarly, some $p^{(2)}_{2,\ell_2}\leq\tfrac{1}{3}$
996: is output having read only a finite part of $(q^{(2)}_n)$,
997: say, up to $n_2$. Moreover, as
998: $q^{(1)}$ and $q^{(2)}$ coincide up to $n_1$
999: and since $\calM$ operates deterministically,
1000: $p^{(2)}_{1,\ell_1}=p^{(1)}_{1,\ell_1}\leq\tfrac{1}{3}$.
1001: \begin{figure}[htb]
1002: \includegraphics[width=0.99\textwidth]{lowersemi}\vspace*{-4ex}%
1003: \caption{\label{figLowersemi}Illustration to the
1004: iterative construction employed in the proof of
1005: Claim~\ref{clOne}}%
1006: \end{figure}
1007:
1008: Continuing this process with $x^{(k)}:=2^{-k+1}$
1009: for $k=3,4,\ldots$ as
1010: indicated in Figure~\ref{figLowersemi} eventually yields a
1011: rational sequence $q^{(\omega)}$ with
1012: $\lim_n q^{(\omega)}_n=:x^{(\omega)}=0$, upon input of which $\calM$ outputs
1013: a double sequence $p^{(\omega)}$
1014: such that $p^{(\omega)}_{k,\ell_k}\leq\tfrac{1}{3}$
1015: for all $k=1,2,\ldots$. In particular,
1016: $y^{(\omega)}:=
1017: \mysup\limits_k\myinf\limits_\ell p^{(\omega)}_{k,\ell}\leq\tfrac{1}{3}$
1018: whereas $\barH(x^{(\omega)})=1$ : contradiction.
1019: %$\calM$ realizes a $(\myrho'\to\myrhol')$--computation of $\barH$.
1020: \qed\end{proof}
1021: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1022: Notice that the above proof involves one-dimensionally
1023: indexed sequences $(q_n)$ for input and two-dimensionally
1024: indexed ones $(p_{k,\ell})$ for output.
1025: We now proceed a step further in proof difficulty,
1026: namely involving two-dimensional indices for both
1027: input and output in order to establish Item~b).
1028: \begin{myclaim} \label{clTwo}
1029: Let $f:\IR\to\IR$ violate monotonicity
1030: in that $f(0)=1$ and $f(1)=0$.
1031: Then, $f$ is not $(\myrhol'\to\myrhol')$--computable.
1032: \end{myclaim}
1033: \begin{proof}
1034: We construct a $\myrhol'$--name for $x=0$
1035: from an iteratively defined sequence of initial segments of
1036: $\myrhol'$--names for $x=1$:
1037:
1038: Start with $q^{(1)}_{i,j}:=1$ for all $i,j$. Then,
1039: $q^{(1)}=(q^{(1)}_{i,j})$ is obviously a $\myrhol'$--name for $x=1$
1040: and thus yields by presumption, upon input to $\calM$, a
1041: $\myrhol'$--name $p^{(1)}_{k,\ell}$ for $f(1)=0$,
1042: that is, with $0=\mysup\limits_k\myinf\limits_{\ell} p^{(1)}_{k,\ell}$.
1043: In particular, $p^{(1)}_{1,\ell_1}\leq\tfrac{1}{3}$ for
1044: some $\ell_1$.
1045: \begin{figure}[htb]
1046: \includegraphics[width=0.99\textwidth]{monoton}\vspace*{-4ex}%
1047: \caption{\label{figMonoton}Illustration to the
1048: iterative construction employed in the proof of
1049: Claim~\ref{clTwo}}%
1050: \end{figure}
1051:
1052: Until output of $p^{(1)}_{1,\ell_1}$, $\calM$
1053: has read only finitely many entries of $q^{(1)}$;
1054: say, up to $i_1$ and $j_1$, that is, covered
1055: in Figure~\ref{figMonoton} by the light gray rectangle.
1056: Now consider $q^{(2)}$ defined as in this figure:
1057: Since $\myinf_j q^{(2)}_{i,j}=0$ for $i\leq i_1$ and
1058: $\myinf_j q^{(2)}_{i,j}=1$ for $i>i_1$,
1059: $\mysup\limits_i\myinf\limits_j q^{(2)}_{i,j}=1$, that is, this is again
1060: valid $\myrhol'$--name for $x=1$; and again, $\calM$ will
1061: by presumption convert $q^{(2)}$ into a $\myrhol'$--name
1062: $p^{(2)}$ for $f(1)=0$. In particular,
1063: $p^{(2)}_{2,\ell_2}\leq\tfrac{1}{3}$ for
1064: some $\ell_2$; and, being a deterministic machine, $\calM$'s
1065: operation on the initial part (dark gray) on which input $q^{(2)}$
1066: coincides with input $q^{(1)}$ will first have generated
1067: the same initial output, namely
1068: $p^{(2)}_{1,\ell_1}=p^{(1)}_{1,\ell_1}\leq\tfrac{1}{2}$.
1069: \\
1070: Again, until output of $p^{(2)}_{2,\ell_2}$, $\calM$
1071: has read only a finite part of $q^{(2)}$ of, say, up to $i_2>i_1$
1072: (light gray). By now considering input $q^{(3)}$ with
1073: $\myinf_j q^{(3)}_{i,j}=0$ for $i\leq i_2$ as in
1074: Figure~\ref{figMonoton}, we arrive at $p^{(3)}$ and $\ell_3$ with
1075: $p^{(3)}_{1,\ell_1},p^{(3)}_{2,\ell_2},p^{(3)}_{3,\ell_3}\leq\tfrac{1}{3}$;
1076: and so on with $i_3, q^{(4)}, p^{(4)}, \ell_4, i_4$, \ldots
1077: \\
1078: Finally observe that continuing these arguments eventually
1079: leads to a rational double sequence
1080: $q^{(\omega)}=(q^{(\omega)}_{i,j})$ which has
1081: $\myinf_j q^{(\infty)}_{i,j}=0$ for $i\leq i_{\infty}=\infty$%
1082: ---and is therefore a valid $\myrhol'$--name for $x=0$ (rather
1083: than $x=1$)---but gets mapped by $\calM$ to
1084: $p^{(\omega)}=(p^{(\omega)}_{k,\ell})$ with
1085: $\myinf_\ell p^{(\infty)}_{k,\ell}
1086: \leq p^{(\infty)}_{k,\ell_k}\leq\tfrac{1}{3}$ for all $k$.
1087: % and therefore $\mysup_k\myinf_\ell p^{(\infty)}_{k,\ell}\leq\tfrac{1}{3}$.
1088: Since $f(0)=1$, this contradicts our presumption that
1089: $\calM$ maps $\myrhol'$--names for $x$
1090: to $\myrhol'$--names for $f(x)$.
1091: \qed\end{proof}
1092: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1093: The above proofs involving
1094: $\myrho'$ and $\myrhol'$
1095: proceeded by constructing an infinite sequence of inputs
1096: $q^{(1)},q^{(2)},\ldots,q^{(\omega)}$~
1097: (each possibly a multi-indexed sequence of its own).
1098: For finally asserting Claim~c) involving $\myrho''$,
1099: we will extend this method from length $\omega$,
1100: the first infinite ordinal, to an even longer one.
1101: \begin{myclaim} \label{clThree}
1102: $\barH:\IR\to\IR$
1103: is not $(\myrho''\to\myrho'')$--computable.
1104: \end{myclaim}
1105: \begin{proof}
1106: Outwit a Type-2 Machine $\calM$, presumed to realize
1107: this computation, as follows:%
1108: \begin{enumerate}
1109: \item[i)]
1110: Take $q^{(1)}$ to be the constant double sequence 1,
1111: i.e., $q^{(1)}_{i,j}:=1$ for all $i,j$.
1112: Being a $\myrho''$--name for $1$, it is by
1113: presumption mapped to a $\myrho''$--name
1114: $p^{(1)}$ for $\barH(1)=0$, that is, satisfying
1115: $\lim_k\lim_\ell p^{(1)}_{k,\ell}=0$. In particular,
1116: almost every column $\#k$ contains an entry $\#\ell$
1117: with $p^{(1)}_{k,\ell}\leq\tfrac{1}{3}$. Until output of
1118: the first such $p^{(1)}_{k_1,\ell_1}$, $\calM$ has
1119: read only a finite part of $q^{(1)}$---say,
1120: up to $i_1,j_1$.
1121:
1122: \begin{figure}[htb]
1123: \includegraphics[width=0.99\textwidth]{omega1}\vspace*{-4ex}%
1124: \caption{\label{figOmega1}The first infinitely long
1125: iterative construction employed in the proof of
1126: Claim~\ref{clThree}}%
1127: \end{figure}
1128:
1129: \item[ii)]
1130: Observe that this Argument~i) equally applies to the
1131: scaled input sequence $2^{-m}\cdot q^{(1)}$ for any $m$.
1132: So define $q^{(2)}_{i,j}:=q^{(1)}_{i,j}$ for
1133: $j\leq j_1$ (i.e., inherit the initial part of $q^{(1)}$)
1134: and $q^{(2)}_{i,j}:\equiv\tfrac{1}{2}$ for $j>j_1$.
1135: Now upon input of this $q^{(2)}$,
1136: $\calM$ will output $p^{(2)}$ with, again,
1137: infinitely many $p^{(2)}_{k,\ell}\leq\tfrac{1}{3}$,
1138: the first one---$(k_2,\ell_2)$, say---after having
1139: read $q^{(2)}$ only up to some $(i_2,j_2)$. Furthermore
1140: $\calM$'s determinism implies
1141: $p^{(2)}_{k_1,\ell_1}=p^{(1)}_{k_1,\ell_1}\leq\tfrac{1}{3}$.
1142:
1143: By repeating for $m=2,3,\ldots$, we eventually obtain%
1144: ---similarly to the proof of Claim~\ref{clTwo}---an input
1145: sequence $q^{(\omega)}$ with $q^{(\omega)}_{i,j}$
1146: with $\lim_i\lim_j q^{(\omega)}_{i,j}=0$, that is,
1147: a valid $\myrho''$--name for $x=0$ (rather than 1).
1148: This is mapped by $\calM$ to $p^{(\omega)}$ with
1149: $p^{(\omega)}_{k_m,\ell_m}\leq\tfrac{1}{3}$ for all $m$.
1150: On the other hand, $p^{(\omega)}$ is by presumption a
1151: $\myrho''$--name for $\barH(0)=1$.
1152: Therefore, there are infinitely many $m$
1153: with $p^{(\omega)}_{m,\ell}\geq\tfrac{2}{3}$
1154: for some $\ell>\ell_m$ \underline{and}
1155: $p^{(\omega)}_{m,\ell_m}\leq\tfrac{1}{3}$;
1156: see the grey columns in the right part of
1157: Figure~\ref{figOmega1}.
1158: \item[iii)]
1159: Since this gives no contradiction yet, we proceed
1160: by considering the first such column $m$ containing
1161: an entry $\leq\tfrac{1}{3}$ as well as an entry
1162: $\geq\tfrac{2}{3}$. Take the initial
1163: part of the input $q^{(\omega)}$ ---
1164: up to $(i_{\omega},j_{\omega})$, say,
1165: depicted in grey in the left part of Figure~\ref{figOmega2} ---
1166: that $\calM$ has read until output of both of them;
1167: extend it with $\tfrac{1}{2}$s in top direction
1168: and with $1$s to the right. Feed this $\myrho''$--name
1169: for $x=1$ into $\calM$ until output of an entry
1170: $p_{k,\ell}\leq\tfrac{1}{3}$
1171: in some column $k$ beyond $m$. Then repeat extending
1172: to the right with $1$s replaced by $\tfrac{1}{2}$s
1173: for a second entry $p_{k,\ell}\leq\tfrac{1}{3}$.
1174:
1175: \begin{figure}[htb]
1176: \includegraphics[width=0.99\textwidth]{omega2}\vspace*{-4ex}%
1177: \caption{\label{figOmega2}Second infinitely long
1178: iterative construction employed in the proof of
1179: Claim~\ref{clThree}}%
1180: \end{figure}
1181:
1182: More generally, proceed similarly as in ii)
1183: and extend $q^{(\omega)}_{i_{\omega},i_{\omega}}$
1184: to the right in such a way with some $\myrho''$--name
1185: $q'^{(\omega)}$ for $x=0$
1186: as to obtain another column $m'$
1187: with both entries $\leq\tfrac{1}{3}$ and
1188: $\geq\tfrac{2}{3}$; see the middle part of Figure~\ref{figOmega2}.
1189: Again, $\calM$ outputs the latter two entries having
1190: read only a finite part; say, up to $(i'_\omega,j'_\omega)$.
1191:
1192: Now extend this part, too, with $\tfrac{1}{2}$ in
1193: top direction and with another $q''^{(\omega)}$
1194: obtained, again, as in ii) for a third column $m''$
1195: with both entries $\leq\tfrac{1}{3}$ and
1196: $\geq\tfrac{2}{3}$; and so on.
1197:
1198: This eventually
1199: leads to an input $q^{(2\omega)}$ which,
1200: due to the extensions to the top, represents
1201: a $\myrho''$--name for $x=\tfrac{1}{2}$ and is
1202: thus mapped by presumption to a $\myrho''$--name
1203: $p^{(2\omega)}$ for $\barH(\tfrac{1}{2})=0$.
1204: In particular, almost every column of $p^{(2\omega)}$
1205: has almost every entry $\leq\tfrac{1}{3}$ while maintaining
1206: infinitely many columns with preceding entries
1207: $\leq\tfrac{1}{3}$ and $\geq\tfrac{2}{3}$;
1208: see the right part of Figure~\ref{figOmega2}. This asserts the existence
1209: of infinitely many columns in $p^{(2\omega)}$
1210: containing $\leq\tfrac{1}{3}$, $\geq\tfrac{2}{3}$, and
1211: $\leq\tfrac{1}{3}$ in order.
1212: And again, already a finite initial part
1213: of $q^{(2\omega)}$ up to some $(i_{2\omega},j_{2\omega})$
1214: gives rise to the first such triple.
1215: \item[iv)]
1216: Notice that the arguments in iii)
1217: similarly yield the existence of an appropriate,
1218: scaled counter-part $\tfrac{1}{2}q'^{(2\omega)}$ of
1219: $q^{(2\omega)}$, of some $\tfrac{1}{4}q''^{(2\omega)}$,
1220: and so on, all leading to output containing infinitely
1221: many columns with alternating triples as above.
1222: We now construct input $q^{(3\omega)}$ leading to
1223: output $p^{(3\omega)}$ containing an infinity of
1224: columns, each with four entries
1225: $\leq\tfrac{1}{3}$, $\geq\tfrac{2}{3}$,
1226: $\leq\tfrac{1}{3}$, \underline{and} $\geq\tfrac{2}{3}$.
1227:
1228: To this end, take the initial part of $q^{(2\omega)}$
1229: leading to output of the first column with alternating
1230: triple in the above sense; then extend it with the
1231: initial part of the scaled version $\tfrac{1}{2}q'^{(2\omega)}$
1232: leading to another column with such a triple; and so on.
1233: Observing that, due to the scaling, the thus obtained
1234: $q^{(3\omega)}$ represents a $\myrho''$--name for
1235: $x=0$, almost every column of the output $p^{(3\omega)}$
1236: representing $\barH(0)=1$ contains entries $\geq\tfrac{2}{3}$
1237: in addition to the infinitely many columns with triples as
1238: above; see the left part of Figure~\ref{figOmega3}.
1239:
1240: \begin{figure}[htb]
1241: \includegraphics[width=0.99\textwidth]{omega3}\vspace*{-4ex}%
1242: \caption{\label{figOmega3}Third, fourth, and fifth
1243: infinitely long iterative construction employed in the proof of
1244: Claim~\ref{clThree}}%
1245: \end{figure}
1246: \item[v)]
1247: Our next step is a $\myrho''$--name $q^{(4\omega)}$ for
1248: $x=\tfrac{1}{4}$ giving rise to $p^{(4\omega)}$ with
1249: infinitely many columns containing alternating
1250: quintuples. This is obtained by repeating the arguments
1251: in iv) to obtain initial segments of (variants of) $q^{(3\omega)}$,
1252: stacking them horizontally---in order to obtain an infinity
1253: of columns with alternating quadruples---while extending in top direction
1254: with $\tfrac{1}{4}$; see the middle part of Figure~\ref{figOmega3}.
1255: This forces $\calM$ to output a $\myrho''$--name
1256: $q^{(4\omega)}$ for $\barH(\tfrac{1}{4})=0$ and thus
1257: with in almost every column almost every entry being $\leq\tfrac{1}{4}$,
1258: thus extending the alternating quadruples to quintuples.
1259: \item[vi)]
1260: Noticing that the vertical extension in v) was similar to step
1261: iii), we now take a step similar to iv) based on horizontally
1262: stacked initial parts
1263: of scaled counterparts of $q^{(4\omega)}$ in order to obtain
1264: a $\myrho''$--name $q^{(5\omega)}$ for $x=0$ which $\calM$ maps to
1265: some $p^{(5\omega)}$ containing infinitely many alternating six-tuples.
1266:
1267: Then again construct a $\myrho''$--name $q^{(6\omega)}$ for
1268: $x=\tfrac{1}{8}$ by horizontally stacking initial segments of
1269: (variants of) $q^{(5\omega)}$ while extending them vertically
1270: with $\tfrac{1}{8}$ and so on.
1271: \end{enumerate}
1272: Now for the bottom line:
1273: By proceeding the above construction, one eventually obtains
1274: a rational double sequence $q^{(\omega^2)}$ with
1275: $\lim_j q^{(\omega^2)}_{i,j}=0$ for all $i$ ---
1276: that is, a $\myrho''$--name for $x=0$ ---
1277: mapped by $\calM$
1278: to some $p^{(\omega^2)}$ containing (infinitely many) columns $\#k$
1279: with infinitely many alternating entries
1280: $\leq\tfrac{1}{3}$ and $\geq\tfrac{2}{3}$ ---
1281: contradicting that, for $\myrho''$-names $p=(p_{k,\ell})$,
1282: $\lim_\ell p_{k,\ell}$ is required to exist for every $k$.
1283: \qed
1284: \end{proof}
1285: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1286: \section{Hierarchies of Hypercomputable Real Functions}
1287: \label{secHierarchies}
1288: The present section investigates and compares the first levels
1289: of the two hierarchies of hypercomputable real functions induced
1290: by the two approaches to real function hypercomputation considered
1291: in Section~\ref{secContinuous}:
1292: based on oracle support and based on weakened encodings.
1293: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1294: \subsection{Weakly Computable Real Functions} \label{secWeakly}
1295: For every $(\alpha\to\beta)$--computable function $f:A\to B$,
1296: one may obviously replace representation $\alpha$ for $A$
1297: by a stronger one and $\beta$ for $B$ by a weaker one
1298: while maintaining computability of $f$:
1299: \\[0.5ex]
1300: $\displaystyle
1301: f \quad (\alpha\to\beta)\text{--computable}
1302: \;\wedge\;\;\alpha,\reduceq\alpha
1303: \;\;\wedge\;\beta\reduceq\beta' \quad\Rightarrow\quad
1304: f \quad (\alpha,\to\beta')\text{--computable}$.
1305: \\[0.5ex]
1306: However if both $\alpha$ and $\beta$ are made, say,
1307: weaker then $(\alpha'\to\beta')$--computability
1308: of $f$ may in general be violated.
1309: For $\alpha=\beta=\myrhol$, though, we have seen
1310: in Example~\ref{exHeaviside} that the implication
1311: ``$(\myrhol\to\myrhol)\Rightarrow(\myrhol'\to\myrhol')$''
1312: does hold at least for the case of $f$ being
1313: Heaviside's function. By the following result,
1314: it holds in fact for every $f$:
1315: \begin{theorem}[Second Main Theorem of Real Hypercomputation]
1316: \label{thSufficient}
1317: ~ \\ Consider $f:\IR\to\IR$.
1318: \begin{enumerate}
1319: \item[a)]
1320: If $f$ is $(\myrho\to\myrho)$--computable,
1321: then it is also $(\myrho'\to\myrho')$--computable.
1322: \item[b)]
1323: If $f$ is $(\myrho\to\myrhol)$--computable,
1324: then it is also $(\myrho'\to\myrhol')$--computable.
1325: \item[c)]
1326: If $f$ is $(\myrhol\to\myrhol)$--computable,
1327: then it is also $(\myrhol'\to\myrhol')$--computable.
1328: \item[d)]
1329: If $f$ is $(\myrho'\to\myrho')$--computable,
1330: then it is also $(\myrho''\to\myrho'')$--computable.%
1331: \item[e)]
1332: If $f$ is $(\myrho''\to\myrho'')$--computable,
1333: then it is also $(\myrho'''\to\myrho''')$--computable.
1334: \end{enumerate}
1335: The claims remain valid under oracle-supported computation.
1336: \end{theorem}
1337:
1338: \medskip
1339: \noindent
1340: As a consequence, we obtain
1341: the following partial strengthening of Lemma~\ref{lemRepr}:
1342: \begin{corollary} \label{corRepr}
1343: ~ It holds ~
1344: $\displaystyle\myrho\;\:\equiv\;\:\,
1345: \myrhol\!\myjoin\myrhog\;\:\reducneqt\;\:
1346: \myrhol\;\:\reducneqt\;\myrho'\;\:\equiv\;\:\,
1347: \myrhol'\!\myjoin\myrhog'\;\:\reducneqt\;\:
1348: \myrhol'\;\:\reducneqt\;\:\myrho''$ \\
1349: where ``\,$\reduceqt\!$'' denotes \emph{continuous} reducibility
1350: of representations {\rm\mycite{Def.~2.3.2}{Weihrauch}}.
1351: \end{corollary}
1352: \begin{proof}
1353: The positive claims follow from Lemmas~\ref{lemRepr}
1354: and \ref{lemRelative}.
1355: For a negative claim like ``$\myrhol'\notreduceqt\myrho'$''
1356: suppose the contrary. Then by Lemma~\ref{lemRelative},
1357: with the help of some appropriate oracle $\calO$,
1358: one can convert $\myrhol'$--names to $\myrho'$--names.
1359: As Heaviside's function $\Heavi$ is $(\myrho'\to\myrhol')$--computable
1360: by Example~\ref{exHeaviside} and Theorem~\ref{thSufficient},
1361: composition with the presumed conversion
1362: implies $(\myrho'\to\myrho')$--computability of
1363: $\Heavi$ relative to $\calO$---contradicting Theorem~\ref{thNecessary}c).
1364: \qed\end{proof}
1365:
1366: \begin{proof}[Theorem~\ref{thSufficient}d]
1367: Let $f$ be $(\myrho'\to\myrho')$--computable and
1368: $x$ given by a $\myrho''$--name, that is, a rational
1369: sequence $q=(q_n)$ with
1370: $x=\lim_i \lim_j q_{\scriptscriptstyle\langle i,j\rangle}$.
1371: For each $i$, compute by assumption from the $\myrho'$--name
1372: $q_{\scriptscriptstyle\langle i,\cdot\rangle}=
1373: (q_{\scriptscriptstyle\langle i,j\rangle})_{_j}$ of
1374: $x_i:=\lim_j q_{\scriptscriptstyle \langle i,j\rangle}$
1375: a $\myrho'$--name of $f(x_i)$, that is, a sequence
1376: $p=p_{\scriptscriptstyle\langle i,\cdot\rangle}=
1377: (p_{\scriptscriptstyle \langle i,j\rangle})_{_j}$ with
1378: $f(x_i)=\lim_j p_{\scriptscriptstyle\langle i,j\rangle}$.
1379: Continuity of $f$ due to Fact~\ref{facFolklore}c) asserts
1380: $$\lim_i \lim_j p_{\scriptscriptstyle\langle i,j\rangle}
1381: \;=\;\lim_i f(x_i)\;\overset{!}{=}\;f\big(\lim_i x_i\big)
1382: \;=\;f\big(\lim_i \lim_j
1383: q_{\scriptscriptstyle\langle i,j\rangle}\big)\;=\;f(x) $$
1384: this sequence $p$ to be a $\myrho''$--name
1385: for $y=f(x)$.
1386: \qed\end{proof}
1387:
1388: Where the last proof exploited Fact~\ref{facFolklore}c),
1389: the next one relies on Theorem~\ref{thNecessary}c):
1390:
1391: \begin{proof}[Theorem~\ref{thSufficient}e]
1392: A $\myrho'''$--name for $x\in\IR$ is a rational sequence $a=(q_n)$ with
1393: $x=\lim_i\lim_j\lim_k q_{\scriptscriptstyle\langle i,j,k\rangle}$.
1394: For each $i$, exploit $(\myrho''\to\myrho'')$--computability of $f$
1395: to obtain, from the $\myrho''$--name
1396: $q_{\scriptscriptstyle\langle i,\,\cdot\,,\,\cdot\,\rangle}$ of
1397: $x_i:=\lim_j\lim_k q_{\scriptscriptstyle \langle i,j,k\rangle}\in\IR$,
1398: a sequence
1399: $p_{\scriptscriptstyle\langle i,\,\cdot\,,\,\cdot\,\rangle}$
1400: with $\lim_j\lim_k p_{\scriptscriptstyle\langle i,j,k\rangle}$
1401: as $\myrho''$--name of $f(x_i)$.
1402: Similarly to case d), this sequence $p$ constitutes a
1403: $\myrho'''$--name for $y=f(x)$
1404: by continuity of $f$ due to Theorem~\ref{thNecessary}c).
1405: \qed\end{proof}
1406:
1407: \begin{proof}[Theorem~\ref{thSufficient}a]
1408: Let $f$ be $(\myrho\to\myrho)$--computable. Its
1409: $(\myrho'\to\myrho')$--computability is established as follows:
1410: Given $(q_n)\subseteq\IQ$ with $x=\lim_n q_n$,
1411: apply the assumption to evaluate
1412: $f(q_n)$ for each $n$ up to error $2^{-n}$; that is,
1413: obtain $p_n\in\IQ$ with $|p_n-f(q_n)|\leq2^{-n}$.
1414: Since $f$ is continuous by
1415: Fact~\ref{facFolklore}a), it follows
1416: $f(x)=\lim_n f(q_n)=\lim_n p_n$
1417: so that $(p_n)$ is a $\myrho'$--name for $y=f(x)$.
1418: \qed\end{proof}
1419:
1420: \noindent
1421: It is interesting
1422: that the latter proof works in fact uniformly in $f$,
1423: i.e., we have%
1424: \begin{scholium}
1425: The \emph{apply} operator
1426: ~$C(\IR)\times\IR\ni(f,x)\mapsto f(x)$~
1427: is $([\myrho\myto\myrho]\times\myrho'\to\myrho')$--computable.
1428: \end{scholium}
1429: Similarly, Theorem~\ref{thSufficient}b) follows from
1430: Lemma~\ref{lemLowersemi} below together with the observation
1431: that every $(\myrho\to\myrhol)$--computable $f$ has a
1432: computable $[\myrho\myto\myrhol]$--name
1433: \cite[\textsc{Corollary~5.1(2)} and \textsc{Theorem~3.7}]{SemiTCS};
1434: here, $[\myrho\myto\myrhol]$ denotes a natural representation
1435: for the space $\LSC(\IR)$ of lower semi-continuous
1436: functions $f:\IR\to\IR$ considered in \cite{SemiTCS}.
1437: Specifically, a $[\myrho\myto\myrhol]$--name for such a $f$
1438: is an enumeration of all rational triples $(a,b,c)$ such that
1439: $c<\min f\big[[a,b]\big]$---the latter making sense as a
1440: lower semi-continuous function attains its minimum (though
1441: not necessarily its maximum) on any compact set.
1442: $[\myrho\myto\myrhol]$ indeed \emph{is} a representation for
1443: $\LSC(\IR)$ because different lower semi-continuous functions
1444: give rise to different such collections $\{(a,b,c)\in\IQ^3:\ldots\}$;
1445: cf.~\mycite{Lemma~3.3}{SemiTCS}.
1446: \begin{lemma} \label{lemLowersemi}
1447: ~ $\LSC(\IR)\times\IR\ni(f,x)\mapsto f(x)$~
1448: is ~$([\myrho\myto\myrhol]\times\myrho'\to\myrhol')$--computable.
1449: \end{lemma}
1450: \begin{proof}
1451: Let $(a_k,b_k,c_k)_{_k}$ denote the given $[\myrho\myto\myrhol]$--name
1452: of $f\in\LSC(\IR)$ and $(q_n)_{_n}$ the given $\myrho'$--name for $x\in\IR$.
1453: Our goal is to $\myrhol'$--compute $y:=f(x)$.
1454: Define the sequence $p=(p_m)_{_m}\subseteq\IQ\cup\{+\infty\}$ by
1455: \begin{equation} \label{eqLowersemi}
1456: p_{_{\langle k,\ell,n\rangle}}
1457: \;:=\; \left\{ \begin{array}{ll}
1458: \max\big\{c_m \;: \;m\leq k \;\wedge& [a_m,b_m]\supseteq[a_k,b_k]\big\} \\[0.2ex]
1459: &\text{ if } q_n\in(a_k,b_k) \;\wedge\; |b_k-a_k|=2^{-\ell} \\[0.5ex]
1460: +\infty &\text{ otherwise} \end{array} \right.
1461: \end{equation}
1462: %with a computable pairing (or, in this case rather, tripling)
1463: %function $\langle\,\cdot\,,\,\cdot\,,\,\cdot\,\rangle:\IN^3\to\IN$.
1464: From the given information, one can
1465: obviously compute $p$. Moreover this sequence satisfies
1466: \begin{itemize}
1467: \item[\labelitemii] ~$\liminf p\geq y$: \\
1468: Let $\epsilon>0$ be arbitrary. Since $f$ is lower semi-continuous,
1469: its preimage $f^{-1}\big[(y-\epsilon,\infty)\big]\ni x$ is an open set
1470: and therefore contains an entire ball around $x$. In fact,
1471: the center of this ball may be chosen as rational
1472: and its diameter of the form $2^{-L}$ for some $L\in\IN$;
1473: formally (see Figure~\ref{figUnifival}): %there exist $K,L,K'\in\IN$ such that
1474: \begin{multline} \label{eqUnifival}
1475: \exists K,L,K'\in\IN: \quad
1476: x\in(a_{K'},b_{K'})\subseteq[a_K,b_K]\subseteq
1477: f^{-1}\big[(y-\epsilon,\infty)\big] \;\;\wedge\\
1478: |b_K-a_K|=2^{-L} \;
1479: \wedge\; a_{K'}=a_K+\tfrac{3}{2}\cdot 2^{-L-2}
1480: \;\wedge\; b_{K'}=b_K-\tfrac{3}{2}\cdot 2^{-L-2}
1481: \end{multline}
1482: where we have exploited that \emph{every} rational pair
1483: $(a,b)$ occurs in the list representing the $[\myrho\myto\myrhol]$--name.
1484: Moreover, as it consists of all rational triples $(a,b,c)$
1485: with $c<\min f\big[[a,b]\big]$,
1486: \begin{equation} \label{eqLowersemi2}
1487: \exists M\geq K: \;\; [a_K,b_K]=[a_M,b_M] \;\wedge\;
1488: c_M\geq\min f\big[[a_M,b_M]\big]-\epsilon
1489: \overset{\text{(*)}}{\;\geq\;}y-2\epsilon
1490: \end{equation}
1491: with (*) a consequence of
1492: $[a_K,b_K]\subseteq f^{-1}\big[(y-\epsilon,\infty)\big]$
1493: in Equation~(\ref{eqUnifival}). Finally,
1494: \begin{equation} \label{eqLowersemi3}
1495: \lim_n q_n=x\in(a_{K'},b_{K'})
1496: \qquad\Rightarrow\quad
1497: \exists N: \;\forall n\geq N: \;\; q_n\in(a_{K'},b_{K'})\enspace .
1498: \end{equation}
1499: So putting things together, for each $n\geq N$, $\ell\geq L'$, and $k\geq M$,
1500: we either have $p_{_{\langle k,\ell,n\rangle}}=+\infty\geq y-2\epsilon$;
1501: or we are in the first case of Equation~(\ref{eqLowersemi}), thus%
1502: \begin{itemize}
1503: \item[\labelitemi] $q_n\in(a_k,b_k)$ with $|b_k-a_k|\leq2^{-\ell}$
1504: \item[\labelitemi] $q_n\in(a_{K'},b_{K'})$ by Equation~(\ref{eqLowersemi3})
1505: \item[\labelitemi] hence $[a_k,b_k]\subseteq[a_K,b_K]$
1506: by Equation~(\ref{eqUnifival}) due to $\ell\geq L'$;
1507: cf. Figure~\ref{figUnifival}.
1508: \item[\labelitemi] So $[a_k,b_k]\subseteq[a_M,b_M]$ by Equation~(\ref{eqLowersemi2})
1509: \item[\labelitemi] implying $p_{_{\langle k,\ell,n\rangle}}\geq c_M\geq y-2\epsilon$
1510: by Equations~(\ref{eqLowersemi}) and (\ref{eqLowersemi2}) since $k\geq M$.
1511: \end{itemize}
1512: Summarizing, it holds $p_{_{\langle k,\ell,n\rangle}}\geq y-2\epsilon$
1513: for all $(k,\ell,n)\in\IN^3$ not belonging to the finite set
1514: $\{0,1,\ldots,N-1\}\times\{0,1,\ldots,L'-1\}\times\{0,1,\ldots,M-1\}$
1515: of exceptions. Consequently $\liminf p\geq y-2\epsilon$;
1516: even $\liminf p\geq y$ because $\epsilon>0$ was arbitrary.
1517: \begin{figure}[hbt]
1518: \centerline{\includegraphics[width=0.8\textwidth]{unifival}\vspace*{-3ex}}%
1519: \caption{\label{figUnifival}Nesting of some rational intervals of
1520: dyadic length contained in $f^{-1}\big[(y-\epsilon,\infty)\big]$. \newline
1521: The parameters are chosen in such a way that,
1522: whenever $(a_{K'},b_{K'})$ meets some other
1523: $(a_k,b_k)$ of length $|b_k-a_k|=2^{-\ell}$
1524: for $\ell\geq L':=L+2$, then $[a_k,b_k]$ is entirely contained
1525: within the larger $[a_K,b_K]$.}%
1526: \end{figure}
1527: \item[\labelitemii] ~$\liminf p\leq y$ \\
1528: Indeed: Since the $[\myrho\myto\myrhol]$--name contains in particular all
1529: rational pairs $(a_k,b_k)$ and these intervals are dense in $\IR$,
1530: there exists to every $\ell\in\IN$ some $k$ such that
1531: $|b_k-a_k|=2^{-\ell}$ and $x\in(a_k,b_k)$. Furthermore
1532: it holds $q_n\in(a_k,b_k)$ for some sufficiently large $n$
1533: because $\lim_n q_n=x$.
1534: We have thus infinitely many triples $(k,\ell,n)$ for which
1535: $p_{_{\langle k,\ell,n\rangle}}$ is defined by the first
1536: case in Equation~(\ref{eqLowersemi}) and thus agrees with
1537: some $c_m<\min f\big[[a_m,b_m]\big]\leq f(x)=y$ as
1538: $x\in(a_k,b_k)\subseteq[a_m,b_m]$.
1539: \end{itemize}
1540: Concluding, we have $\liminf_m p_m=y$.
1541: Although $p$ may attain the value $+\infty$, this can easily
1542: be overcome by proceeding to
1543: $\tilde p_m:=p_m$ for $p_m\not=\infty$
1544: and $\tilde p_m:=\max\{0,\tilde p_0,\ldots,\tilde p_{m-1}\}$ for
1545: $p_m=\infty$ because this transformation $p\mapsto\tilde p$
1546: on sequences obviously does not affect their $\liminf<\infty$.
1547: This yields a $\tildemyrhol'$--name for $y$
1548: which can finally be converted to
1549: the desired $\myrhol'$--name due to the easy part of
1550: Scholium~\ref{schLiminf}.
1551: \qed
1552: \end{proof}
1553:
1554: In order to obtain a similar uniform claim yielding
1555: Theorem~\ref{thSufficient}c), recall that every
1556: $(\myrhol\to\myrhol)$--computable function $f:\IR\to\IR$
1557: is necessarily both monotonically increasing and
1558: lower semi-continuous
1559: (Fact~\ref{facFolklore}b+c). This suggests% the following
1560: \begin{definition} \label{defMLSC}
1561: Let $\MLSC{\IR}$ denote the class of all
1562: monotonically increasing, lower semi-continuous functions
1563: $f:\IR\to\IR$. A $[\myrhol\myto\myrhol]$--name
1564: for $f\in\MLSC{\IR}$ is an enumeration of the set
1565: $\{(a,c)\in\IQ^2: c<f(a)\}$.
1566: \end{definition}
1567: \begin{lemma} \label{lemMLSC}
1568: \begin{enumerate}
1569: \item[a)] Distinct $f,g\in\MLSC{\IR}$ have
1570: different sets $\{(a,c):\ldots\}$ according to
1571: Definition~\ref{defMLSC}; that is,
1572: $[\myrhol\myto\myrhol]$ constitutes a well-defined
1573: representation.
1574: \item[b)] A function $f\in\MLSC{\IR}$ is
1575: $(\myrhol\to\myrhol)$--computable iff it
1576: has a computable $[\myrhol\myto\myrhol]$--name.
1577: \item[c)] Let $f\in\MLSC{\IR}$, $(a_k,c_k)_{_k}$ with
1578: $\{(a,c)\in\IQ^2:c<f(a)\}=\{(a_k,c_k):k\in\IN\}$,
1579: $x\in\IR$, and $q=(q_n)\subseteq\IQ$ with
1580: $x=\liminf_n q_n$. Then, the rational sequence $p$ defined by
1581: $$p_{_{\langle k,n,\ell\rangle}}
1582: \;:=\;\left\{ \begin{array}{ll}
1583: \max\big\{c_m \;: \;m\leq k \;\wedge\; a_m\geq a_k\big\}
1584: &\text{ ~if~ } a_k<q_n<a_k+2^{-\ell} \\[0.5ex]
1585: +\infty &\text{ ~otherwise} \end{array} \right. $$
1586: satisfies ~$\liminf p=f(x)=:y$.
1587: \item[d)] Therefore, the \emph{apply} operator
1588: ~$\MLSC{\IR}\times\IR \;\ni\; (f,x)
1589: \;\mapsto\; f(x)$~ is
1590: ~$([\myrhol\myto\myrhol]\times\myrhol'\to\myrhol')$--computable.
1591: \end{enumerate}
1592: \end{lemma}
1593: \begin{proof}%[Sketch for Claim~c]
1594: \begin{enumerate}
1595: \item[a)]
1596: Let $f,g\in\MLSC{\IR}$ with $f\not=g$,
1597: that is, w.l.o.g. $f(x_0)<g(x_0)$ for some $x_0\in\IR$.
1598: There exists some $c_0\in\IQ$ with $f(x_0)<c_0<g(x_0)$.
1599: Being monotonically increasing and lower semi-continuous,
1600: their pre-images $f^{-1}\big[(c_0,\infty)\big]\not\ni x_0$
1601: and $g^{-1}\big[(c_0,\infty)\big]\ni x_0$ on open half-interval
1602: $(c_0,\infty)$ are again open half-intervals
1603: $(x_f,\infty)$ and $(x_g,\infty)$, respectively.
1604: As $x_0$ belongs to the second but not to the first,
1605: we have $x_g<x_0<x_f$ and therefore
1606: $x_g<a_0<x_f$ for some $a_0\in\IQ$.
1607: Then $a_0\in (x_g,\infty)=g^{-1}\big[(c_0,\infty)\big]$
1608: yields $c_0<g(a_0)$
1609: whereas $a_0\not\in (x_f,\infty)=f^{-1}\big[(c_0,\infty)\big]$
1610: asserts $c_0\not< f(a_0)$.
1611: \item[b)]
1612: Let $\calM$ denote a Type-2 Machine $(\myrhol\to\myrhol)$--computing
1613: $f\in\MLSC{\IR}$. Evaluating $f$ at $a\in\IQ$ by simulating $\calM$
1614: on the $\myrhol$--name $(a,a,a,\ldots)$ for $a$ thus yields
1615: a $\myrhol$--name for $f(a)$ which is (equivalent to) a list
1616: of all $c\in\IQ$ with $c<f(a)$ \mycite{Lemma~4.1.8}{Weihrauch}. So
1617: dove-tailing this simulation for all $a\in\IQ$ yields the
1618: desired $[\myrhol\myto\myrhol]$--name for $f$.
1619:
1620: Conversely, knowing a $[\myrhol\myto\myrhol]$--name
1621: $(a_k,c_k)_{_k}$ for $f\in\MLSC{\IR}$ and given an
1622: increasing sequence $(q_n)\subseteq\IQ$ with $x=\sup_n q_n$,
1623: let
1624: $$p_n:=c_n \text{ ~if~ } a_n\leq q_n,
1625: \qquad p_n:=-\infty \text{ ~otherwise} \enspace . $$
1626: Then, in the first case, $p_n=c_n<f(a_n)\leq f(q_n)\leq f(x)=:y$
1627: by monotonicity, and $p_n=-\infty\leq y$ in the second;
1628: hence $\sup_n p_n\leq y$. To see $\sup_n p_n\geq y$,
1629: fix arbitrary $\epsilon>0$ and consider
1630: the open half-interval
1631: $f^{-1}\big[(y-\epsilon,\infty)\big]=(x_\epsilon,\infty)$
1632: containing $x$ and thus also some rational $a=a_K\in(x_\epsilon,x)$,
1633: $K\in\IN$. Furthermore $q_n\nearrow x$ yields some $N\in\IN$
1634: such that $q_n\in(a_K,x)$ for all $n\geq N$.
1635: And finally there exists $M\geq N$ with
1636: $a_M=a_K$ and $c_M\geq f(a_M)-\epsilon$.
1637: Together this asserts $q_M>a_K=a_M$ because $M\geq N$
1638: and thus $p_M=c_M\geq f(a_K)-\epsilon>y-2\epsilon$
1639: due to $a_k\in f^{-1}\big[(y-\epsilon,\infty)\big]$.
1640: \item[c)]
1641: Take arbitrary $\epsilon>0$. As $f$ is increasing and
1642: lower semi-continuous, the pre-image $f^{-1}\big[(y-\epsilon,\infty)\big]$
1643: is an open half-interval $(x_\epsilon,\infty)$ containing $x$.
1644: Therefore there exist $K,L\in\IN$ such that
1645: $x_\epsilon<a_K$ and $a_K+2^{-L}<x$;
1646: furthermore, the sequence $(a_k,c_k)_{_k}$ containing
1647: \emph{all} rational pairs $(a,c)$ with $c<f(a)$,
1648: there is $M\geq K$ such that $a_M=a_K$ and
1649: $c_M\geq f(a_M)-\epsilon$;
1650: and finally, since $\liminf_n q_n=x>a_M+2^{-L}$,
1651: it holds $q_n>a_M+2^{-L}$ for all $n\geq N$ with an appropriate $N\in\IN$.
1652: Observe that $q>a_M+2^{-L}$ and $a<q<a+2^{-L}$ implies
1653: $a\geq a_M$; so
1654: together we have for all $n\geq N$, $\ell\geq L$, and $k\geq M$
1655: that $p_{_{\langle k,n,\ell\rangle}}$ is either $+\infty$
1656: or $\geq c_M\geq f(a_K)-\epsilon%\overset{\text{(*)}}{\geq}
1657: \geq f(x_\epsilon)-\epsilon\geq y-2\epsilon$
1658: due to monotonicity of $f$ and by definition of $x_\epsilon<a_K$.
1659: This proves $\liminf p\geq y$ because $\epsilon$ was arbitrary.
1660:
1661: To see the reverse inequality
1662: ``$\liminf p\leq y$'', take arbitrary $\ell\in\IN$.
1663: There exists $k\in\IN$ with $a_k<x<a_k+2^{-\ell}$
1664: and, because of $\liminf_n q_n=x$,
1665: also $n\in\IN$ with $a_k<q_n<a_k+2^{-\ell}$.
1666: We therefore have infinitely many triples $(n,k,\ell)$
1667: for which $p_{_{\langle n,k,\ell\rangle}}$ agrees with
1668: a certain $c_m<f(a_m)\leq f(a_k)\leq f(x)=y$.
1669: \item[d)]
1670: Given a $\myrhol'$--name for $x$, one can obtain a
1671: sequence $(q_n)\subseteq\IQ$ with $x=\liminf_n q_n$
1672: by virtue of Scholium~\ref{schLiminf}.
1673: From this, the sequence $p\subseteq\IQ$ with
1674: $\liminf p=f(x)$ according to c) is obviously
1675: computable and yields, again by Scholium~\ref{schLiminf},
1676: a $\myrhol'$--name for $y=f(x)$.
1677: \qed\end{enumerate}\end{proof}
1678: Concluding this subsection, the classes of
1679: $(\myrho^{(d)}\to\myrho^{(d)})$--computable
1680: real functions $f:\IR\to\IR$ form,
1681: for $d=0,1,\ldots$ respectively,
1682: a hierarchy. By Fact~\ref{facRAH},
1683: this hierarchy is strict as can be seen from
1684: the constant functions $f(x)\equiv c$
1685: with $c\in\RDelta{d+1}$.
1686: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1687: \subsection{Arithmetic Weierstra\ss{} Hierarchy} \label{secWeier}
1688: Section~\ref{secWeakly} established the
1689: sequence ~$\myrho, \myrho', \myrho'', \ldots$~
1690: of increasingly weaker representations for $\IR$
1691: to yield the strict hierarchy of
1692: $(\myrho\to\myrho)$--computable,
1693: $(\myrho'\to\myrho')$--computable,
1694: and $(\myrho''\to\myrho'')$--computable
1695: functions $f:[0,1]\to\IR$.
1696: We now compare these classes with those induced
1697: by the other kind of real hypercomputation suggested
1698: in Section~\ref{secContinuous}: relative to
1699: the Halting Problem $H=\emptyset'$ and its
1700: iterated jumps $\emptyset''$, \ldots
1701:
1702: Such a comparison makes sense because both
1703: weakly and oracle-computable real functions
1704: are necessarily continuous according to
1705: Fact~\ref{facFolklore}d)/Theorem~\ref{thNecessary}c)
1706: and Lemma~\ref{lemRelative}, respectively.
1707:
1708: \medskip
1709: The classical \person{Weierstra\ss} Approximation Theorem
1710: establishes any continuous real function $f:[0,1]\to\IR$
1711: to be the uniform limit $f=\ulim_n P_n$
1712: of a sequence of rational polynomials $(P_n)\subseteq\IQ[X]$.
1713: Here, `$\ulim$' suggestively denotes uniform convergence of continuous
1714: functions on $[0,1]$, that is the requirement
1715: $$\sup\limits_{0\leq x\leq1}|f(x)-P_n(x)| \;\;=:\;\; \|f-P_n\|\;\to\;0
1716: \qquad\text{as~ } n\to\infty \enspace . $$
1717: The famous \aname{Effective Weierstra\ss{} Theorem}
1718: due to \person{Pour-El}, \person{Caldwell}, and \person{Hauck}
1719: relates effectively evaluable to effectively approximable
1720: real functions:
1721: \begin{fact} \label{facWeier}
1722: A function $f:[0,1]\to\IR$ is $(\myrho\to\myrho)$--computable
1723: if and only if it holds%
1724: \\[0.5ex]\begin{tabular}{rl}
1725: ~$\pmb{[\myrho\myto\myrho]}$\textbf{:} &
1726: There exists a computable sequence of
1727: ~(degrees and coefficients of) \\
1728: & rational polynomials $(P_n)\subseteq\IQ[X]$ such that%
1729: \hfill%
1730: \makebox[9.0em][l]{\rule[1ex]{0pt}{1.3ex}\smash{\begin{minipage}[b]{10.3em}
1731: \begin{equation} \label{eqWeier}
1732: \|f-P_n\| \;\leq\; 2^{-n}
1733: \end{equation}
1734: \end{minipage}}}%
1735: \end{tabular}%
1736: \end{fact}%
1737: \begin{proof}
1738: See \mycite{Section~0.7}{PER}, \cite{WeierA}, or \cite{WeierB}. \\
1739: The notion ``$[\myrho\myto\myrho]$'' is justified as the list $(P_n)_{_n}$
1740: constitutes (or is equivalent to) a $[\myrho\myto\myrho]$--name for
1741: $f=\ulim_n P_n$; cf. \cite[bottom of p.160]{Weihrauch}.%
1742: \qed\end{proof}
1743: %
1744: The aforementioned other approach to continuous real hypercomputation
1745: arises from allowing the fast convergent sequence $(P_n)_{_n}\subseteq\IQ[X]$
1746: to be computable in $\emptyset'$ or $\emptyset''$.
1747: The $\emptyset'$--computable $f:[0,1]\to\IR$ have in fact
1748: already been characterized by \person{Ho} as Claim~a) of the following
1749: \begin{lemma} \label{lemWeier}
1750: \begin{enumerate}
1751: \item[a)]
1752: To a real function $f:[0,1]\to\IR$, there exists a $\emptyset'$--computable
1753: sequence of polynomials $(P_n)$ satisfying Equation~(\ref{eqWeier})
1754: ~if and only if~ it holds
1755: \\[0.5ex]\begin{tabular}{rl}
1756: $\pmb{[\myrho\myto\myrho]'}$\textbf{:} &
1757: There is a computable sequence $(Q_m)\subseteq\IQ[X]$
1758: converging uniformly \\ & (although not necessarily `fast')
1759: to $f$, that is, with $f=\ulim\limits_{m\to\infty} Q_m$.
1760: \end{tabular}
1761: \item[b)]
1762: For an arbitrary oracle $A$, the sequence
1763: (of discrete degrees and numerators/denominators of the coefficients of)
1764: ~$(P_n)_{_n}\subseteq\IQ[X]$~
1765: is $A'$--computable ~iff~
1766: there exists an $A$--computable sequence ~$(Q_{n,m})_{_{n,m}}\subseteq\IQ[X]$~
1767: such that
1768: $$ \forall n \;\exists M\;\forall m\geq M: \quad P_n=Q_{n,m} \enspace . $$
1769: %\item[c)]
1770: %The set ~$\big\{p\in\IQ[X]\,\big|\,\forall x\in[0,1]: -1\leq p(x)\leq+1\big\}$~
1771: %is decidable (in the classical discrete sense).
1772: \item[c)]
1773: To a real function $f:[0,1]\to\IR$, there exists a $\emptyset''$--computable
1774: sequence of polynomials $(P_n)$ satisfying Equation~(\ref{eqWeier})
1775: ~if and only if~ it holds
1776: \\[0.5ex]\begin{tabular}{rl}
1777: $\pmb{[\myrho\myto\myrho]''}$\textbf{:} &
1778: There is a computable sequence $(Q_m)\subseteq\IQ[X]$
1779: s.t.
1780: $f=\ulim\limits_i \ulim\limits_j Q_{\scriptscriptstyle\langle i,j\rangle}$.%
1781: \end{tabular}
1782: \end{enumerate}
1783: \end{lemma}
1784: Notice the similarity of Claims~a+c) to Fact~\ref{facRAH}b).
1785: \begin{proof}
1786: \begin{enumerate}
1787: \item[a)] See \mycite{Theorem~16}{Ho}.
1788: \item[b)] is a straight-forward extension of
1789: \person{Shoenfield}'s Limit Lemma \mycite{Lemma III.3.3}{Soare}
1790: and its generalization to sequences of rational numbers
1791: \mycite{Lemma 4.1}{Zheng}.
1792: %\item[c)]
1793: % The solvability of ~``$p(x)=1$''~ on $[0,1]$ is decidable
1794: % based on \person{Sturm}'s Theorem\footnote{\person{Tarski}'s
1795: % \aname{Quantifier Elimination} yields a generalization to
1796: % the multi-variate case.} using exact arithmetic on $\IQ$;
1797: % similarly for ~``$p(x)=-1$''.
1798: % If both assertions fail then the maximum of $|p|$ on $[0,1]$
1799: % is either $>1$~ or ~$<1$, both cases being semi-decidable
1800: % by virtue of \mycite{Corollary~6.2.5}{Weihrauch}.
1801: \item[c)]
1802: If $\emptyset''$--computable ~$(P_n)_{_n}\subseteq\IQ[X]$~
1803: satisfies Equation~(\ref{eqWeier}), then by virtue of
1804: the relativization of \mycite{Theorem~16}{Ho} there exists some
1805: $\emptyset'$--computable ~$(\tilde P_n)_{_n}\subseteq\IQ[X]$~
1806: converging to the same $f$ uniformly on $[0,1]$.
1807: By Claim~a) in turn, $\tilde P_n=\ulim_m Q_{n,m}$ for some
1808: computable sequence $(Q_{n,m})\subseteq\IQ[X]$.
1809:
1810: Conversely if $f=\ulim_n\tilde P_n$ with $\tilde P_n:=\ulim_m Q_{n,m}$
1811: for a computable $(Q_{n,m})$, then let $P_n:=Q_{n,m_n}$ where
1812: \begin{equation} \label{eqFastsub}
1813: m_n\quad:=\quad \min\big\{m\in\IN \:\big|\,
1814: \forall k,\ell\geq m \,:\, \|Q_{n,k}-Q_{n,\ell}\|\leq2^{-n}\big\}
1815: \enspace .
1816: \end{equation}
1817: This sequence $(m_n)_{_n}$ is well-defined and yields
1818: ~$\|P_n-\tilde P_n\|\leq2^{-n}$, so $f=\ulim_n\tilde P_n=\ulim_n P_n$.
1819: Moreover, the minimum in Equation~(\ref{eqFastsub}) is taken
1820: over a co-r.e. set ---
1821: $r:=\|Q_{n,k}-Q_{n,\ell}\|\cdot2^n$~ being ~$\myrho$--computable
1822: by virtue of \mycite{Corollary~6.2.5}{Weihrauch}
1823: and the complementary condition ~``$r>1$''~
1824: $\myrho$--r.e. open
1825: and hence recursive in $\emptyset'$.
1826: Similar to Equation~(\ref{eqFastsub}), this $\emptyset'$--computable
1827: sequence $(P_n)_{_n}\subseteq\IQ[X]$
1828: converging uniformly though just ultimately to $f$
1829: can be turned into a $\emptyset''$--computable, fast convergent one.
1830: \qed
1831: \end{enumerate}
1832: \end{proof}
1833: We thus have two hierarchies of hypercomputable
1834: continuous real functions:
1835: \\ \noindent \textbullet\quad\,\!
1836: $[\myrho\myto\myrho]$, \;\,\quad $[\myrho\myto\myrho]'$,
1837: \;\:\,\quad $[\myrho\myto\myrho]''$,
1838: \quad\ldots
1839: \\ \noindent \textbullet\quad
1840: $(\myrho\to\myrho)$, ~ $(\myrho'\to\myrho')$, ~
1841: $(\myrho''\to\myrho'')$, ~ \ldots
1842: \\ \noindent
1843: By the Fact~\ref{facWeier},
1844: their respective ground-levels coincide.
1845: Our next result compares their respective higher levels.
1846: They turn out to lie skewly to each other (Claim~c).
1847: \begin{theorem} \label{thWeier}
1848: \begin{enumerate}
1849: \item[a)] Let $f:[0,1]\to\IR$ be $[\myrho\myto\myrho]'$--computable
1850: (in the sense of Lemma~\ref{lemWeier}a).
1851: Then, $f$ is $(\myrho'\to\myrho')$--computable.
1852: \item[b)] Let $f:[0,1]\to\IR$ be $(\myrho'\to\myrho')$--computable.
1853: Then, $f$ is $[\myrho\myto\myrho]''$--computable.
1854: \item[c)] There is a $(\myrho'\to\myrho')$--computable
1855: but not $[\myrho\myto\myrho]'$--computable $f:[0,1]\to\IR$.
1856: \end{enumerate}
1857: \end{theorem}
1858: The idea to c) is that
1859: every $[\myrho\myto\myrho]'$--computable $f:[0,1]\to\IR$
1860: has a modulus of uniform continuity recursive in $\emptyset'$;
1861: whereas a $(\myrho'\to\myrho')$--computable $f$,
1862: although uniformly continuous as well, in general does not.
1863:
1864: Before proceeding to the proof, % of Theorem~\ref{thWeier},
1865: we first provide a tool which turns out to be useful in the sequel.
1866: It is well-known in Recursive Analysis that, although equality
1867: of real numbers is $\myrho$--undecidable due to the Main Theorem,
1868: inequality is at least semi-decidable. The following lemma
1869: generalizes this to $\myrho'$ and to
1870: $(\myrho'\to\myrhol')$--computable functions:
1871: \begin{lemma} \label{lemBound}
1872: \begin{enumerate}
1873: \item[a)]
1874: Let $f:\IR\to\IR$ be $(\myrho\to\myrhol)$--computable. Then
1875: the property
1876: $$ \big\{ (a,b,c)\in\IQ^3 \,\big|\, \exists x\in[a,b]: f(x)>c \big\} $$
1877: whether $f$ on $[a,b]$ exceeds $c$ ~ is semi-decidable.
1878: \item[b)]
1879: Let $f:\IR\to\IR$ be $(\myrho'\to\myrhol')$--computable. Then
1880: the property
1881: $$ \big\{ (a,b,c)\in\IQ^3 \,\big|\, \exists x\in[a,b]: f(x)>c \big\} $$
1882: whether $f$ on $[a,b]$ exceeds $c$ ~ is semi-decidable
1883: relative to $\emptyset'$.
1884: \item[c)]
1885: Let $f:\IR\to\IR$ be $(\myrho'\to\myrho')$--computable.
1886: Then the property
1887: $$ \big\{ (a,b,c,m)\in\IQ^3\times\IN \,\big|\,
1888: \forall x\in[a,b]: c-2^{-m}\leq f(x)\leq c+2^{-m}\big\}$$
1889: is decidable relative to $\emptyset''$.
1890: \end{enumerate}
1891: \end{lemma}
1892: \begin{proof}
1893: a) is standard; c) follows from b) which is established as follows: \\
1894: By lower semi-continuity of $f$ due to Theorem~\ref{thNecessary}a),
1895: if $f$ exceeds $c$ on the compact interval $[a,b]$, then it does so
1896: on some \emph{rational} $x$. Feeding, for any such $x\in[a,b]\cap\IQ$,
1897: the $\myrho'$--name $(x,x,x,x,\ldots)$ for $x$
1898: into the Type-2 Machine computing
1899: $f$ reveals the mapping $\IQ\ni x\mapsto f(x)$ to be
1900: $(\nuQ\to\myrhol')$--computable. With $\emptyset'$--oracle,
1901: it thus becomes $(\nuQ\to\myrhol)$--computable by virtue
1902: of \mycite{Lemma~4.2}{Zheng}. Since
1903: $\{(y,c):y>c\}$ is $(\myrhol\times\nuQ)$--semi-decidable, the claim follows.
1904: \qed
1905: \end{proof}
1906:
1907: \begin{proof}[Theorem~\ref{thWeier}]
1908: \begin{enumerate}
1909: \item[a)] Let $(P_n)\subseteq\IQ[X]$ denote a computable sequence
1910: converging uniformly (yet not necessarily fast) to $f$.
1911: Let $x\in[0,1]$ be given as the limit of a sequence $(q_n)\subseteq\IQ$.
1912: Then, $p_n:=P_n(q_n)$ eventually converges to $f(x)$.
1913: \item[b)]
1914: Let $x\in[0,1]$ be given by (an equivalent to)
1915: its $\myrho$--name in form of two
1916: rational sequences $(a_n)$ and $(b_n)$ with
1917: $\{x\}=\bigcap_n [a_n,b_n]$. There exists a rational
1918: sequence $(c_m)$ forming a $\myrho$--name for $f(x)$,
1919: that is, satisfying
1920: $c-2^{-m}\leq f(x)\leq c+2^{-m}$ for all $m$;
1921: and by virtue of Lemma~\ref{lemBound}c), such a sequence can
1922: be found with the help of a $\emptyset''$--oracle.
1923: This reveals that $f$ is $\emptyset''$--recursive in
1924: the sense of \mycite{Section~4}{Ho} and thus,
1925: similarly to \mycite{Corollary~17}{Ho},
1926: $[\myrho\myto\myrho]''$--computable.
1927: \item[c)] Let $h:\IN\to\IN$ denote a $\emptyset'$--computable
1928: injective total enumeration of some subset
1929: $H=h[\IN]\in\Sigma_2\setminus\Delta_2$.
1930: Observe that $a_m:=2^{-h(m)}$ is a
1931: $\myrho'$--computable real sequence
1932: converging to 0 with modulus of convergence
1933: \mycite{Definition~4.2.2}{Weihrauch} lacking $\emptyset'$--recursivity;
1934: compare \cite[\textsc{Exercise~4.2.4}c)]{Weihrauch}.
1935: Let $\varphi:\IR\to\IR$ denote some $(\myrho\to\myrho)$--computable
1936: unit pulse, that is, vanishing outside $[0,1]$ and having
1937: height $\max_x\varphi(x)=\varphi(\tfrac{1}{2})=1$; a piecewise linear
1938: `\emph{hat}' function for instance will do
1939: fine but we can even choose $\varphi$ as in
1940: \mycite{Theorem~1.1.1}{PER} to obtain the counter-example
1941: \begin{equation} \label{eqPulses}
1942: f(x) \;:=\; \sum_{m\in\IN} a_m\cdot\varphi(2^m x-1) \enspace ,
1943: \end{equation}
1944: (that is, a non-overlapping superposition of scaled translates
1945: of such pulses) to be $C^\infty$.
1946: By Theorem~\ref{thSufficient}a),
1947: $x\mapsto a_m\cdot\varphi(2^m x-1)$ is $(\myrho'\to\myrho')$--computable;
1948: in fact even uniformly in $m$: Given $(q_n)_{_n}\subseteq\IQ$
1949: with $x=\lim_n q_n$, one can for each $M\in\IN$ obtain
1950: a sequence $(p_{k,M})_{_{k}}\subseteq\IQ$ with
1951: $\lim_k p_{k,M}=\sum_{m\leq\pmb{M}} a_m\cdot\varphi(2^mx-1)=:f_M$.
1952: The functions $f_M$ converge uniformly (though not effectively) to $f$
1953: because of the disjoint supports of the terms $\varphi(2^mx-1)$
1954: in Equation~(\ref{eqPulses}).
1955: Therefore $\lim\limits_M p_{M,M}=f(x)$, thus establishing
1956: $(\myrho'\to\myrho')$--computability of $f$.
1957: \\ Suppose $f$ was $[\myrho\myto\myrho]'$--computable.
1958: Then, by virtue of \mycite{Lemma~15}{Ho}, it has a
1959: $\emptyset'$--recursive modulus
1960: of uniform continuity; cf. \mycite{Definition~6.2.6.2}{Weihrauch}.
1961: In particular given $n\in\IN$, one can $\emptyset'$--compute
1962: $m\in\IN$ such that
1963: $x:=2^{-m}$ and $y:=\tfrac{3}{2}x$ satisfy
1964: \quad$2^{-n} \;{\geq}\;
1965: |f(x)-f(y)| \;=\; |0-a_m|$\quad
1966: contradicting that $(a_m)$ has no $\emptyset'$--recursive
1967: modulus of continuity.
1968: \qed\end{enumerate}
1969: \end{proof}
1970: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1971: \section{Type-2 Nondeterminism} \label{secNondet}
1972: Concerning the two kinds of
1973: real hypercomputation considered so far---based on oracle-support
1974: and weak real number encodings that is---recall that the
1975: according proofs of Fact~\ref{facFolklore} and Theorem~\ref{thNecessary}
1976: crucially rely on the underlying Turing Machines
1977: to behave deterministically.
1978: This raises the question whether nondeterminism
1979: might yield the additional power necessary for evaluating
1980: discontinuous real functions like Heaviside's.
1981:
1982: In the discrete (i.e., Type-1) setting where any computation
1983: is required to terminate, the finitely many possible choices
1984: of a nondeterministic machine can of course be simulated by a
1985: deterministic one---however already here subject to the important
1986: condition that all paths of the nondeterministic computation
1987: indeed terminate, cf. \cite{Boas}.
1988: In contrast, a Type-2 computation realizes a transformation
1989: from/to infinite strings and is therefore a generally non-terminating
1990: process. Therefore, nondeterminism here involves an infinite
1991: number of guesses which turns out cannot be simulated by a deterministic
1992: Type-2 machine.
1993:
1994: We also point out that nondeterminism has already before
1995: been revealed not only a useful but indeed the most natural concept
1996: of computation on $\Sigma^\omega$. More precisely,
1997: \person{B\"{u}chi} extended Finite Automata from
1998: finite to infinite strings
1999: and proved that here, as opposed deterministic,
2000: \emph{non}deterministic ones are closed under complement \cite{Buechi}
2001: and thus the appropriate model of computation.
2002: \begin{figure}[htb]
2003: \vspace*{-2.4ex}%
2004: \centerline{\sf\normalsize\begin{tabular}{l||c@{\;}|@{\;}c}
2005: Chomsky-Level & $\displaystyle\Sigma^*$ & $\displaystyle\Sigma^\omega$ \\[0.2ex]
2006: \hline
2007: 3: regular & Finite Automata & B\"{u}chi Automata (nondeterministic) \\[0.4ex]
2008: 2: context-free & \\[0.4ex]
2009: 1: context-sensitive & \\[0.4ex]
2010: 0: unrestricted & (Type-1) Turing Machines & \emph{non}deterministic Type-2 Machines
2011: \end{tabular}}\caption{Models of Computation
2012: in Chomsky's Hierarchies over
2013: finite/infinite strings\label{figChomsky}}%
2014: \vspace*{-2.2ex}%
2015: \end{figure}
2016: Since automata and Turing Machines
2017: constitute the bottom and top levels, respectively, of \person{Chomsky}'s
2018: Hierarchy of classical languages $L\subseteq\Sigma^*$ ~(Type-1 setting),
2019: we suggest that over infinite strings $\Sigma^\omega$ ~(Type-2 setting)
2020: both their respective counterparts, that is
2021: B\"{u}chi Automata \emph{and Type-2 Machines}
2022: be considered nondeterministically; compare Figure~\ref{figChomsky}.
2023:
2024: The concept of nondeterministic computation of a
2025: function ~$f:\subseteq\Sigma^*\to\Sigma^*$~
2026: (as opposed to a decision problem) is taken from the famous
2027: \person{Immerman}-\person{Szelepsc\'{e}nyi} Theorem in
2028: computational complexity;
2029: cf. for instance
2030: \cite[the paragraph preceding \textsc{Theorem~7.6}]{Papadimitriou}:
2031: For $\bar x\in\dom(f)$, some computing paths of
2032: the according machine $\calM$ may fail by leading to rejecting
2033: states, as long as%
2034: \begin{enumerate}
2035: \item[1)] there \emph{is} an accepting computation
2036: of $\calM$ on $\bar x$ \quad and
2037: \item[2)] \emph{every} accepting computation of
2038: $\calM$ on $\bar x$ yields the correct output ~$f(\bar x)$.
2039: \end{enumerate}
2040: This notion extends straight-forwardly from Type-1 to the Type-2 setting:
2041: \begin{definition} \label{defNondet}
2042: Let $A$ and $B$ be sets with respective representations
2043: $\alpha:\subseteq\Sigma^\omega\to A$ and
2044: $\beta:\subseteq\Sigma^\omega\to B$.
2045: A function $f:\subseteq A\to B$ is called
2046: \emph{nondeterministically} $(\alpha\to\beta)$--computable
2047: if some nondeterministic one-way Turing Machine $\calM$,%
2048: \begin{itemize}
2049: \item upon input of any $\alpha$--name $\bar\sigma\in\Sigma^\omega$
2050: for some $a\in\dom(f)$,
2051: \item has a computation which outputs a $\beta$--name
2052: %$\bar\tau\in\Sigma^\omega$
2053: for $b=f(a)$ \hfill\underline{and}
2054: \item every infinite computation\footnote{This condition is slightly
2055: stronger than the one required in \mycite{Definition~14}{CIE}.}
2056: of $\calM$ on $\bar\sigma$
2057: %$\bar\tau\in\Sigma^\omega$
2058: outputs a %(possibly different)
2059: $\beta$--name for $b=f(a)$.%
2060: \end{itemize}
2061: \end{definition}
2062: This definition is sensible insofar as it leads
2063: to closure under composition:
2064: \begin{observation} \label{obsComposition}
2065: Let $f:\subseteq A\to B$ be nondeterministically
2066: $(\alpha\to\beta)$--computable
2067: and $g:\subseteq B\to C$ be nondeterministically
2068: $(\beta\to\gamma)$--computable.
2069: Then, $g\circ f:\subseteq A\to C$ is nondeterministically
2070: $(\alpha\to\gamma)$--computable.
2071: \end{observation}
2072: A subtle point in Definition~\ref{defNondet},
2073: the nondeterministic machine may `withdraw' a
2074: guess as long as it does so within finite time.
2075: \begin{example}[`Deciding' the Arithmetic Hierarchy] \label{x:Kleene}
2076: Let $P\subseteq\IN$ be recursive,%
2077: \[ A \;=\; \big\{ x\in\IN\,|\, \forall y_1\in\IN \exists z_1\in\IN \;\;\forall y_2 \exists z_2
2078: \ldots \forall y_k \exists z_k : \langle x;y_1,z_1,\ldots,y_k,z_k\rangle\in P\big\} \]
2079: on (or below) level $\Pi_{2k}$ of Kleene's Arithmetic Hierarchy.
2080: Then the function $\tilde\chi_A:\IN\to\{0,1\}\times\{\text{\textvisiblespace}\,\}^{\omega}$
2081: is nondeterministically computable:
2082: \\
2083: Observe that $x\in A$ iff
2084: \[ %x\in A \quad\Leftrightarrow\quad
2085: \exists f_1,f_2,\ldots,f_k:\IN\to\IN \;
2086: \forall y_1,y_2,\ldots,y_k\in\IN\;:\;
2087: \langle x;y_1,f(y_1),\ldots,y_k,f(y_k)\rangle\in P \]
2088: So given $x\in\IN$, let $\calM_+$ output ``1''
2089: and then verify, while continuously spitting out blanks ``\textvisiblespace'',
2090: that $\chi_A(x)=1$ indeed holds.
2091: To this end, the machine starts `guessing' the values of
2092: $\bar f=(f_1,\ldots,f_k)$ restricted to $\{0,1,\ldots,n\}$
2093: for $n=1,2,\ldots$
2094: Simultaneously by means of dove-tailing,
2095: $\calM_+$ tries all $\bar y\in\{0,1,\ldots,n\}^k$
2096: and aborts in case that the assertion
2097: ``$\langle x;y_1,f(y_1),\ldots,y_k,f(y_k)\rangle\in P$''
2098: fails.
2099: \\
2100: Now if $x\in A$,
2101: then an appropriate $\bar f$ exists, is ultimately
2102: `found' by $\calM_+$, and leads to indefinite execution;
2103: whereas if $x\not\in A$, then $\calM_+$ will eventually
2104: terminate for any guessed $\bar f$.
2105: \\
2106: Since $\IN\setminus A\in\Pi_{2k+2}$, a machine
2107: $\calM_-$ can output ``0'' and then similarly
2108: verify $\chi_A(x)=0$. The final machine $\calM$,
2109: upon input of $x\in\IN$,
2110: nondeterministically chooses to proceed
2111: either like $\calM_+$
2112: or like $\calM_-$. Its computation %obviously
2113: satisfies the requirements
2114: of Definition~\ref{defNondet}.
2115: \qed\end{example}
2116: %
2117: The power of nondeterministic computation permits
2118: conversion forth and back among
2119: representations on the Real Arithmetic Hierarchy
2120: from Definition~\ref{defRAH}:
2121: \begin{theorem}[Third Main Theorem of Real Hypercomputation] \label{thNondet}
2122: For each $d=0,1,2,\ldots$, the identity $\IR\ni x\mapsto x$
2123: is nondeterministically $(\myrho^{(d+1)}\to\myrho^{(d)})$--computable.
2124: It is furthermore nondeterministically $(\myrho\to\myrhob)$--computable.
2125: \end{theorem}
2126: \begin{proof}%[Theorem~\ref{thNondet}]
2127: Consider first the case $d=0$.
2128: Let $x\in\IR$ be given by a sequence $(q_n)\subseteq\IQ$
2129: eventually converging to $x$. Then, there exists a
2130: fast convergent Cauchy sub-sequence $(q_{n_k})_{_k}$,
2131: that is, satisfying
2132: \begin{equation} \label{eqFast}
2133: \forall k\geq\ell:
2134: \quad |q_{n_k}-q_{n_{\ell}}|\;\leq\;2^{-\ell-1}
2135: \end{equation}
2136: and thus forming a $\myrho$--name for $x$.
2137: To find this subsequence, guess iteratively for each $k\in\IN$
2138: some $n_k>n_{k-1}$ and check whether it complies with
2139: Inequality~(\ref{eqFast}) for the (finitely many)
2140: $\ell\leq k$; if it does not, we may abort
2141: this computation in accordance with
2142: Definition~\ref{defNondet}.
2143:
2144: For $d=1$, let $x=\lim_n x_n$ with $x_n=\lim_m q_{n,m}$.
2145: Then apply the case $d=0$ to
2146: convert for each $n$ the $\myrho'$--name $(q_{n,m})_{_m}$ of $x_n\in\IR$
2147: into an according $\myrho$--name, that is,
2148: a sequence $p_{n,m}$ satisfying $|x_n-p_{n,m}|\leq2^{-m}$.
2149: Its diagonal $(p_{n,n})_{_n}$ then has
2150: $|x-p_{n,n}|\leq|x-x_n|+2^{-n}\to0$ and is thus a $\myrho'$--name
2151: for $x$. Higher levels $d$ can be treated similarly by induction.
2152:
2153: For $(\myrho\to\myrhob)$--computability,
2154: let $x\in(0,2)$ be given by a fast convergent sequence $(q_n)\subseteq\IQ$.
2155: We guess the leading digit $b\in\{0,1\}$ for $x$'s binary expansion
2156: $b.*$; in case $b=0$, check whether $x>1$%
2157: ---a $\myrho$--semi decidable property---and if so, abort;
2158: similarly in case $b=1$, abort if it turns out that $x<1$.
2159: Otherwise (that is, proceeding while simultaneously continuing the
2160: above semi-decision process via dove-tailing)
2161: replace $x$ by $2(x-b)$ and repeat guessing the next bit.
2162: \qed\end{proof}
2163: It is also instructive to observe how, in the case of non-unique binary expansion
2164: (i.e., for dyadic $x$), nondeterminism in the above
2165: $(\myrho\to\myrhob)$--computation generates,
2166: in accordance with the third requirement of Definition~\ref{defNondet},
2167: both possible expansions.
2168:
2169: Theorem~\ref{thNondet} implies that
2170: \emph{non}deterministic computability of real functions is
2171: largely independent of the representation under consideration
2172: --- in striking contrast to the classical case (Corollary~\ref{corRepr})
2173: where the effectivity subtleties arising from different
2174: encodings had confused already Turing himself \cite{Turing2}.
2175: \begin{corollary} \label{corIndependent}
2176: \begin{enumerate}
2177: \item[a)] With respect to nondeterministic reduction ~``$\Nreduceq$''~,
2178: it holds ~
2179: $\displaystyle\myrhob\;\:\Nequiv\;\:\myrho\;\:\Nequiv\;\:
2180: \myrhol\;\:\Nequiv\;\myrho'\;\:\Nequiv\;\:
2181: \myrhol'\;\:\Nequiv\;\:\myrho''\;\:\Nequiv\;\ldots$.
2182: \item[b)]
2183: The entire Real Arithmetic Hierarchy
2184: of \person{Weihrauch} and \person{Zheng}
2185: is nondeterministically computable.
2186: \end{enumerate}
2187: \end{corollary}
2188: \begin{proof}%[Corollary~\ref{corIndependent}]
2189: \begin{enumerate}
2190: \item[a)] follows from Lemma~\ref{lemRepr} and Theorem~\ref{thNondet}.
2191: \item[b)] Let $x\in\RDelta{d+1}$ for some $d\in\IN$.
2192: Then, $x\in\IR$~ is ~$\myrho^{(d)}$--computable
2193: by Definition~\ref{defRAH}; hence nondeterministically
2194: also $\myrhob$--computable by a). Alternatively
2195: combine Example~\ref{x:Kleene} with Fact~\ref{facRAH}a).
2196: \qed
2197: \end{enumerate}
2198: \end{proof}
2199: %
2200: In particular, this kind of hypercomputation allows for
2201: nondeterministic
2202: $(\myrho\to\myrho)$--evaluation of Heaviside's function
2203: by appending to the $(\myrho\to\myrhol)$--computation in
2204: Example~\ref{exHeaviside} a conversion
2205: from $\myrhol\reduceq\myrho'$ back to $\myrho$.
2206: Section~\ref{secHotz} establishes many more real functions,
2207: both continuous and discontinuous ones,
2208: to be nondeterministically computable, too.
2209: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2210: \subsection{Nondeterministic and Analytic Computation} \label{secHotz}
2211: We now show that Type-2 nondeterminism
2212: includes the algebraic so called \aname{BCSS-model} of
2213: real number computation due to \person{Blum}, \person{Cucker},
2214: \person{Shub}, and \person{Smale} \cite{BSS,BCSS}
2215: employed for instance in Computational Geometry \mycite{Section~1.4}{CompGeom}.
2216: As a matter of fact, nondeterministic real hypercomputation even covers all
2217: \emph{quasi-strongly $\delta$--$\IQ$--analytic}
2218: functions $f:\subseteq\IR^d\to\IR$ in the sense of
2219: \person{Chadzelek} and \person{Hotz} \mycite{Definition~5}{Hotz}.
2220: The latter can be considered a
2221: synthesis of the Type-2 (i.e., infinite approximate)
2222: and the \BCSS (i.e., finite exact) model of real number computation.
2223: Its computational power admits
2224: an elegant characterization (see Lemma~\ref{lemHotz}b+c)
2225: in terms of the following
2226: \begin{definition} \label{defHotz}
2227: A $\myrhoh$--name for $x\in\IR$ is some $(q_n)_{_n}\subseteq\IQ$
2228: such that
2229: \begin{equation} \label{e:Hotz}
2230: \exists N \;\forall n\geq N: \quad |q_n-x|\leq2^{-n} \enspace .
2231: \end{equation}
2232: \end{definition}
2233: The encoding sequence of rational approximations must thus converge
2234: fast with the exception of some initial segment of finite yet
2235: unknown length. It corresponds to $\myrho$--computation by
2236: an \emph{Inductive} Turing Machine in the sense of
2237: \cite{Burgin} which is roughly speaking a Type-2
2238: Machine but whose output tape(s) need not be one-way
2239: \mycite{Section~2.1}{Weihrauch} provided that
2240: the contents of every cell ultimately stabilizes.
2241: \begin{lemma} \label{lemHotz}
2242: \begin{enumerate}
2243: \item[a)] It holds
2244: ~$\myrho\;\reducneq\;\myrhoh\;\reducneq\;\myrho'$.
2245: \item[b)] A function
2246: ~$f:\subseteq\IR^N\to\IR$~ is
2247: ~$(\myrho^N\to\myrhoh)$--computable
2248: ~ iff ~
2249: it is computable by a
2250: quasi-strongly $\delta$--$\IQ$--analytic machine.
2251: \item[c)] $(\myrho\to\myrhoh)$--computability
2252: is equivalent to $(\myrhoh\to\myrhoh)$--computability.
2253: \item[d)] The class of $(\myrhoh\to\myrhoh)$--computable
2254: functions is closed under composition.
2255: \end{enumerate}
2256: \noindent
2257: The above claims relativize.
2258: \end{lemma}
2259: \begin{proof}
2260: \begin{enumerate}
2261: \item[a)] is immediate.
2262: \item[b)] Observe that
2263: the \emph{robustness} of the program $\pi$ required
2264: in \cite[top of p.157]{Hotz} amounts to
2265: the argument $x\in\IR$ of $f$ being accessible
2266: by rational approximations $q_n\in\IQ$ of
2267: error $|q_n-x|\leq2^{-n}$, that is,
2268: in terms of a $\myrho$--name.
2269: The output $y=f(x)$ on the other hand proceeds
2270: by way of two sequences $(p_m)_{_m},(\epsilon_m)_{_m}\subseteq\IQ$
2271: such that $\epsilon_m\to0$ and $|p_m-y|\leq\epsilon_m$
2272: holds for all sufficiently large $m$.
2273: By effectively proceeding to an appropriate subsequence,
2274: we can w.l.o.g. suppose $\epsilon_m=2^{-m}$,
2275: hence $(p_m)$ is $\myrhoh$--name of $y$.
2276: %A quasi-strongly $\delta$--$\IQ$--analytic computation
2277: %of $f$ therefore corresponds to $(\myrho^N\to\myrho)$--computing
2278: %$\bar x\mapsto f(x)$.
2279: %Now a $\IQ$--machine, using only rational constants
2280: %\cite[\textsc{Section~2.1} and Instruction 1(b) in \textsc{Table~1} on p.154]{Hotz},
2281: %admits immediate simulation by a Type-2 Machine and
2282: %vice versa.
2283: \item[c)]
2284: %Consider for simplicity the case $N=1$.
2285: By a), every $(\myrhoh\to\myrhoh)$--computable
2286: function is $(\myrho\to\myrhoh)$--computable, too.
2287: For the converse implication, take the Type-2 Machine
2288: $\calM$
2289: converting $\myrho$--names for $x\in\IR$ to
2290: $\myrhoh$--names for $y=f(x)$. Let $(q_n)$ satisfy
2291: Equation~(\ref{e:Hotz}) for some unknown $N\in\IN$.
2292:
2293: Now simulate $\calM$ on $(q_n)_{_{n\geq0}}$,
2294: implicitly supposing that it is a valid $\myrho$--name,
2295: i.e., that $N=0$. Simultaneously check consistency of
2296: Condition~(\ref{e:Hotz}), that is,
2297: verify $|q_n-q_k|\leq2^{-n+1}\forall k\geq n\geq N$.
2298: If (or, rather, when) the latter fails for
2299: some $(k_0,n_0)$, $\calM$ has output only finitely
2300: (say $M_0\in\IN$) many $p_m\in\IQ$. In that case,
2301: restart $\calM$ on $(q_n)_{_{n\geq1}}$
2302: presuming $N=1$ while, again, checking this
2303: presumption consistent with (\ref{e:Hotz});
2304: but this time throw away the first $M_0$
2305: elements of the sequence printed by $\calM$.
2306: Continue analogously for $N=2,3,\ldots$.
2307:
2308: We claim that this yields output of a
2309: $\myrhoh$--name for $y$.
2310: Since $(q_n)$ is a valid $\myrhoh$--name,
2311: a feasible $N$ will eventually be found.
2312: Before that happens, the several partial runs
2313: of $\calM$ have produced only finitely (say $M\in\IN$)
2314: many rational numbers $p_m$; and after that,
2315: the final simulation generates
2316: by presumption a valid $\myrhoh$--name for $y$.
2317: Out of this sequence $(p_m)_{_m}$, the first $M$
2318: entries may have been exchanged by outputs of
2319: previous simulation trials; however according to
2320: Definition~\ref{defHotz}, the representation
2321: $\myrhoh$ is immune against such finite
2322: modifications.
2323: \item[d)]
2324: Quasi-strongly $\delta$--$\IQ$--analytic functions
2325: are closed under composition according to \mycite{Lemma~2}{Hotz};
2326: now apply b+c).
2327: \qed\end{enumerate}\end{proof}
2328: %
2329: A \BCSS (or, equivalently, an $\IR$--) machine $\calM$
2330: is permitted to store a finite number of \emph{arbitrary}
2331: real constants $r_1,\ldots,r_k$
2332: \cite[Instruction 1(b) in \textsc{Table~1} on p.154]{Hotz}
2333: and use it for instance to solve the Halting
2334: or any other fixed discrete problem \mycite{Example~6}{BSS}.
2335: Slightly correcting \mycite{Theorem~3}{Hotz}, $\calM$'s simulation
2336: by a rational machine thus \emph{requires} knowledge of $\bar r:=(r_1,\ldots,r_k)\in\IR^k$;
2337: e.g. by virtue of oracle access to
2338: ($\calO:=\{\bin(n):\sigma_n=1\}\subseteq\{0,1\}^*$
2339: as natural encoding of a $\myrhob^k$--name $\bar\sigma\in\{0,1\}^\omega$ of)
2340: $\bar r$---compare \cite{Boldi} for the case of
2341: simulating $\myrho$--semi decidability.
2342: \begin{proposition} \label{proHotz}
2343: \begin{enumerate}
2344: \item[a)]
2345: A function $f:\subseteq\IR\to\IR$ computable
2346: by a \BCSS--machine with constants $\bar r\in\IR^k$
2347: is also $(\myrhoh\to\myrhoh)$--computable
2348: \emph{relative to} $\bar r$.%
2349: \item[b)]
2350: Every $(\myrho\to\myrho)$--computable function
2351: $f:\subseteq\IR\to\IR$
2352: is also $(\myrhoh\to\myrhoh)$--computable.
2353: \item[c)]
2354: Let ~$f:\subseteq\IR\to\IR$~ be $(\myrhoh\to\myrhoh)$--computable
2355: relative to some oracle $\calO\subseteq\Sigma^*$ in
2356: (Kleene's) Arithmetic Hierarchy.
2357: Then $f$ is nondeterministically Type-2 computable.
2358: \end{enumerate}
2359: \end{proposition}
2360: \begin{proof}%[Proposition~\ref{proHotz}]
2361: \begin{enumerate}
2362: \item[a)] See (the proof of) \mycite{Theorem~3}{Hotz}.
2363: \item[b)] Combine Lemma~\ref{lemHotz}a+c).
2364: \item[c)] The nondeterministic simulation can answer
2365: queries to $\calO$ due to Example~\ref{x:Kleene}.
2366: As $\myrho\Nequiv\myrhoh\Nequiv\myrho'$ by
2367: Corollary~\ref{corIndependent}a)
2368: and Lemma~\ref{lemHotz}a),
2369: the claim follows.%
2370: \qed\end{enumerate}\end{proof}
2371: %
2372: Let us illustrate Proposition~\ref{proHotz}a) with the following%
2373: \begin{example} \label{x:Heavi}
2374: Heaviside's Function $h:\IR\to\{0,1\}$ is
2375: trivially \BCSS--computable.
2376: It is also
2377: $(\myrhoh\to\myrhoh)$--computable by
2378: means of \emph{conservative branching}:
2379: Given $x\in\IR$ by virtue of $(q_n)\subseteq\IQ$ with (\ref{e:Hotz})
2380: and unknown $N\in\IN$, let $p_n:=0$ if $q_n\leq2^{-n}$
2381: and $p_n:=1$ otherwise.
2382: \\
2383: Indeed if $x\leq 0$ then, for all $n\geq N$,
2384: $q_n\leq2^{-n}$ and thus $p_n=0=f(x)$.
2385: If on the other hand $x>0$, $x>2^{-M}$ for some $M\in\IN$;
2386: then, for all $n\geq\max\{M+1,N\}$,
2387: $q_n>2^{-n}$ so $p_n=1=f(x)$.
2388: \qed\end{example}
2389: %
2390: %XXX Infinite-time TMs, GTMs, higher recursive computation,
2391: %Borel-continuous Hierarchy, Dag Normann, Peter G. Hinman
2392: Of course the class of nondeterministic Type-2 Machines
2393: (and thus also that of the nondeterministically computable real functions)
2394: is still only countably infinite: most (even constant) functions
2395: $f:\IR\to\IR$
2396: actually remain infeasible to this kind of real hypercomputation.
2397: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2398: \section{Conclusion}\label{secConclusion}
2399: Recursive Analysis is often criticized for being unable,
2400: due to its Main Theorem, to
2401: non-trivially treat discontinuous functions.
2402: Although one can in Type-2 Theory devise sensible computability
2403: notions for, say, generalized (and in particular discontinuous)
2404: functions as for instance in \cite{Ning}, evaluation
2405: $x\mapsto f(x)$ of an $L^2$ function or a distribution $f$
2406: at a point $x\in\IR$ does not make sense here
2407: already mathematically.
2408: Regarding the Main Theorem's connection to the Church-Turing Hypothesis
2409: indicated in the introduction, the present work has investigated
2410: whether and which models of hypercomputation allows for
2411: lifting that restriction.
2412:
2413: A first idea, relativized computation on oracle Turing Machines,
2414: was ruled out right away. A second idea, computation based on
2415: weakened encodings of real numbers, renders evaluation
2416: ~$x\mapsto\Heavi(x)$~ of
2417: Heaviside's function---although discontinuous---for
2418: instance $(\myrho\to\myrhol)$--computable.
2419: The drawback of this notion of real hypercomputation:
2420: it lacks closure under composition.
2421: \begin{example} \label{exComposition}
2422: Let $f:\IR\to\IR$, ~ $f(0):=0$ ~and~ $f(x):=1$ for $x\not=0$.
2423: Let $g(x):=-x$.
2424: Then both $f$ and $g$ are $(\myrho\to\myrhol)$--computable
2425: but their composition ~ $g\circ f:0\mapsto 0, \;0\not=x\mapsto -1$ ~
2426: lacks lower semi--continuity.
2427: \end{example}
2428: %Closure under composition holds for
2429: %$(\myrhol\to\myrhol)$--computability,
2430: %but here closure under negation $f\mapsto -f$ fails.
2431: %
2432: Requiring both argument $x$ and value $y=f(x)$ to be encoded
2433: in the same way---say, $\myrho$, $\myrho'$, or $\myrho''$---%
2434: asserts closure under both composition and negation
2435: $f\mapsto -f$; and the prerequisites
2436: of the Main Theorem applies only to the case $(\myrho\to\myrho)$. Surprisingly,
2437: $(\myrho'\to\myrho')$--computability and $(\myrho''\to\myrho'')$--computability
2438: still require continuity!
2439: %
2440: These results
2441: extend to $(\myrho^{(d)}\to\myrho^{(d)})$--computability for
2442: arbitrary $d$, although already the step from $d=1$ to $2$
2443: made proofs significantly more involved.
2444:
2445: These claims immediately relativize, that is, even a mixture of
2446: oracle support and weak real number encodings does not allow
2447: for hypercomputational evaluation of discontinuous functions.
2448: This is due to the purely information-theoretic nature of the
2449: arguments employed, specifically: the deterministic behavior
2450: of the Turing Machines under consideration.
2451:
2452: %\medskip
2453: So we have finally looked at nondeterminism as a further way of
2454: enhancing the underlying machine model beyond Turing's barrier.
2455: Over the Type-2 setting of infinite strings $\Sigma^\omega$,
2456: this parallels B\"{u}chi's well-established generalization
2457: of finite automata to so-called \emph{$\omega$--regular} languages.
2458: While the practical realizability of Type-2 nondeterminism
2459: is admittedly even more questionable than that of classical
2460: $\cal{NP}$-machines, it does yield an elegant notion of
2461: hypercomputation with nice closure properties and
2462: invariant under various encodings.
2463:
2464: A precise characterization of the class of nondeterministically
2465: computable real functions will be subject of future work.
2466: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2467: \begin{thebibliography}{BCSS98}
2468: \bibitem[Bar03]{Barmpalias}
2469: \textsc{G.~Barmpalias}: ``A Transfinite Hierarchy of Reals'',
2470: pp.163--172
2471: in \emph{Mathematical Logic Quarterly} vol.{\bf49(2)} (2003).
2472: \bibitem[BSS89]{BSS}
2473: \textsc{L.~Blum, M.~Shub, S.~Smale}:
2474: ``On a theory of computation and complexity over the real numbers'',
2475: pp.1--46 in \emph{Bull. Amer. Math. Soc.} vol.{\bf21} (1989).
2476: \bibitem[BCSS98]{BCSS}
2477: \textsc{L.~Blum, F.~Cucker, M.~Shub, S.~Smale}:
2478: ``\emph{Complexity and Real Computation}'', Springer (1998).
2479: \bibitem[BV99]{Boldi}
2480: \textsc{P.~Boldi, S.~Vigna}: ``Equality is a jump'',
2481: pp.49--64 in \emph{Theoretical Computer Science}
2482: vol.\textbf{219} (1999).
2483: \bibitem[BH02]{Naive}
2484: \textsc{V.~Brattka, P.~Hertling}:
2485: ``Topological Properties of Real Number Representations'',
2486: pp.241--257 in \emph{Theoretical Computer Science}
2487: vol.{\bf284} (2002).
2488: \bibitem[Bur04]{Burgin}
2489: \textsc{M.~Burgin}: ``Algorithmic Complexity of Recursive
2490: and Inductive Algorithms'', pp.31--60 in \emph{Theoretical
2491: Computer Science} vol.\textbf{317} (2004).
2492: \bibitem[CH99]{Hotz}
2493: \textsc{T.~Chadzelek, G.~Hotz}:
2494: ``Analytic Machines'', pp.151--167 in \emph{Theoretical
2495: Computer Science} vol.{\bf219} (1999).
2496: \bibitem[Grz57]{Grzegorczyk}
2497: \textsc{A.~Grzegorczyk}: ``On the Definitions of Computable Real
2498: Continuous Functions'', pp.61--77 in \emph{Fundamenta Mathematicae}
2499: {\bf 44} (1957).
2500: \bibitem[Hau76]{WeierB}
2501: \textsc{J.~Hauck}: ``Berechenbare reelle Funktionenfolgen'',
2502: pp.265--282 in \emph{Zeitschrift f\"{u}r Mathematische Logik und
2503: Grundlagen der Mathematik} vol.{\bf22} (1976).
2504: \bibitem[Ho99]{Ho}
2505: \textsc{C.-K.~Ho}: ``Relatively Recursive Real Numbers and Real
2506: Functions'', pp.99--120 in \emph{Theoretical Computer Science}
2507: vol.{\bf 210} (1999).
2508: \bibitem[Ko91]{Ko}
2509: \textsc{Ker-I~Ko}: ``\emph{Complexity Theory of Real Functions}'',
2510: Birkh\"{a}user (1991).
2511: \bibitem[Mor69]{Dictionary}
2512: \textsc{W.~Morris} (Editor): ``\emph{American Heritage Dictionary
2513: of the English Language}'', American Heritage Publishing (1969).
2514: \bibitem[Pap94]{Papadimitriou}
2515: \textsc{C.H.~Papadimitriou}: ``\emph{Computational Complexity}'',
2516: Addison-Wesley (1994).
2517: \bibitem[PEC75]{WeierA}
2518: \textsc{M.B.~Pour-El, J.~Caldwell}:
2519: ``On a simple definition of computable functions of a real variable'',
2520: pp.1--19 in \emph{Zeitschrift f\"{u}r Mathematische Logik und
2521: Grundlagen der Mathematik} vol.{\bf21} (1975).
2522: \bibitem[PER89]{PER}
2523: \textsc{M.B.~Pour-El, J.I.~Richards}: ``\emph{Computability in
2524: Analysis and Physics}'', Springer (1989).
2525: \bibitem[PS85]{CompGeom}
2526: \textsc{F.P.~Preparata, M.I.~Shamos}:
2527: ``\emph{Computational Geometry: An Introduction}'',
2528: Springer (1985).
2529: \bibitem[Ran68]{Analysis}
2530: \textsc{J.F.~Randolph}: ``\emph{Basic Real and Abstract Analysis}'',
2531: Academic Press (1968).
2532: %\bibitem[Sch02]{Schmidhuber}
2533: % \textsc{J.~Schmidhuber}: ``Hierarchies of Generalized Kolmogorov
2534: % Complexities and Nonenumerable Universal Measures Computable in the Limit'',
2535: % pp.587--612 in \emph{International Journal of Foundations of Computer
2536: % Science} vol.\textbf{13:4} (2002).
2537: \bibitem[Soa87]{Soare}
2538: \textsc{R.I.~Soare}: ``\emph{Recursively Enumerable Sets and Degrees}'',
2539: Springer (1987).
2540: \bibitem[STvE89]{Boas}
2541: \textsc{E.~Spaan, L.~Torenvliet, P.~van Emde Boas}:
2542: ``Nondeterminism, Fairness and a Fundamental Analogy'',
2543: pp.186--193 in \emph{The Bulletin of the European Association for
2544: Theoretical Computer Science} (EATCS Bulletin) vol.{\bf37} (1989).
2545: \bibitem[Tho90]{Buechi}
2546: \textsc{Thomas, W.}: ``Automata on Infinite Objects'',
2547: pp.133--191 in \emph{Handbook of Theoretical Computer Science},
2548: vol.{\bf B} (Formal Models and Semantics), Elsevier (1990).
2549: \bibitem[Tur36]{Turing}
2550: \textsc{Turing, A.M.}: ``On Computable Numbers, with an Application to the
2551: Entscheidungsproblem'', pp.230--265 in \emph{Proc. London Math. Soc.}
2552: vol.{\bf42(2)} (1936).
2553: \bibitem[Tur37]{Turing2}
2554: \textsc{Turing, A.M.}: ``On Computable Numbers, with an Application to the
2555: Entscheidungsproblem. A correction'', pp.544--546
2556: in \emph{Proc. London Math. Soc.} vol.{\bf43(2)} (1937).
2557: \bibitem[Wei01]{Weihrauch}
2558: \textsc{K.~Weihrauch}: ``\emph{Computable Analysis}'', Springer (2001).
2559: \bibitem[WZ00]{SemiTCS}
2560: \textsc{K.~Weihrauch, X.~Zheng}:
2561: ``Computability on continuous, lower semi-continuous and
2562: upper semi-continuous real functions'', pp.109--133 in
2563: \emph{Theoretical Computer Science} vol.{234} (2000).
2564: \bibitem[Zie05]{CIE}
2565: \textsc{M.~Ziegler}: ``Computability and Continuity on the
2566: Real Arithmetic Hierarchy and the Power of Type-2 Nondeterminism'',
2567: pp.562--571 in \emph{Proc. 1st Conference on Computability in
2568: Europe} (CiE'2005), Springer LNCS vol.{\bf3526}.
2569: \bibitem[ZW01]{Zheng}
2570: \textsc{X.~Zheng, K.~Weihrauch}: ``The Arithmetical Hierarchy of
2571: Real Numbers'', pp.51--65 in \emph{Mathematical Logic Quarterly}
2572: vol.{\bf47} (2001).
2573: \bibitem[ZW03]{Ning}
2574: \textsc{N.~Zhong, K.~Weihrauch}: ``Computability Theory of
2575: Generalized Functions'', pp.469--505 in
2576: \emph{J. ACM} vol.{\bf50} (2003).
2577: \end{thebibliography}
2578: \end{document}
2579: