---
title: Relativized Diagonal Non-Computation
url: https://www.emergentmind.com/topics/relativized-diagonal-non-computation
type: topic
---

# Relativized Diagonal Non-Computation

Searching arXiv for recent and foundational papers on relativized diagonal non-computation and related oracle/axiomatic frameworks.
Relativized diagonal non-computation denotes a family of constructions in which diagonal avoidance, undecidability, or non-definability is studied relative to an added parameter: an oracle set, a growth bound, a subrecursive machine, or a relativized semantics. In computability theory, the standard instance is a function \(f\) that is diagonally non-computable relative to \(X\), meaning \(f(e)\neq \Phi_e^X(e)\) for every index \(e\) whenever \(\Phi_e^X(e)\downarrow\); in reverse mathematics this yields principles such as \(\mathrm{DNR}_h\); and in algebraic logic related phenomena appear through diagonal-free relativized semantics. The papers considered here also include axiomatic accounts of why diagonal arguments relativize, subrecursive analogues of Busy Beaver and halting-probability constructions, and several nonstandard reinterpretations of the scope of diagonalization [1412.0917], [1811.06259], [1807.00690], [1612.05221], [2603.06012].

## 1. Oracle-relative diagonal non-computability

A function is diagonally non-computable relative to a set \(X\) if it avoids the diagonal values of the universal partial computable functions with oracle \(X\). Formally, if \(\Phi_e^X\) is an effective enumeration of the partial computable functionals, then \(f\) is d.n.c. relative to \(X\) if for every index \(e\),
\[
f(e)\neq \Phi_e^X(e).
\]
In the unrelativized notation based on a universal partial computable function \(\varphi_e\), one writes \(f(e)\neq \varphi_e(e)\) for all \(e\) whenever \(\varphi_e(e)\downarrow\). This is the canonical computability-theoretic form of diagonal non-computation: the function \(f\) “diagonalizes” against the \(e\)-th program on its own index [1412.0917].

Several bounded versions refine this notion. In one formulation, a function \(f\) is \(h\)-bounded if \(f(x)\leq h(x)\) for all \(x\), and \(\mathrm{DNR}_h\) asserts that for every set \(X\) there exists an \(h\)-bounded function d.n.c. relative to \(X\), with \(h\) a computable order, namely nondecreasing and unbounded [1412.0917]. In another formulation, the class
\[
DNC_h=\{f:\omega\to\omega \mid f \text{ is DNC and }(\forall n)\, f(n)<h(n)\}
\]
and its relativized analogue
\[
DNC_h^X=\{f:\omega\to\omega \mid f \text{ is DNC}^X \text{ and }(\forall n)\, f(n)<h(n)\}
\]
are used to study measure-theoretic and degree-theoretic behavior [1411.6846]. For computable \(h\), the negligibility threshold is explicit:
\[
DNC_h \text{ is negligible } \iff \sum_n \frac1{h(n)}=\infty.
\]
This gives a quantitative refinement of diagonal non-computation in terms of growth rates rather than mere existence [1411.6846].

The same notion appears in reverse mathematics with \(k\)-boundedness. If \(g\) is \(k\)-bounded when \(\operatorname{ran}(g)\subseteq \{0,1,\dots,k-1\}\), then \(DNR(k,f)\) means that \(g\) is both \(k\)-bounded and diagonally non-recursive relative to \(f\), that is,
\[
\forall e\;\bigl(g(e)\neq \Phi_e^f(e)\bigr).
\]
This formulation is central when the strength of diagonal principles is compared to weak König’s lemma and to induction schemes [1401.3823].

## 2. Reverse mathematics, forcing, and degree-theoretic structure

The bounded hierarchy of relativized diagonal non-computation has been used to separate principles in reverse mathematics. For every computable order \(h\), there exists an \(\omega\)-model of \(\mathrm{DNR}_h\) that is not a model of \(\mathrm{RCOLOR}_2\), and therefore not a model of \(\mathrm{RWKL}\). The proof combines bushy tree forcing with the Lerman–Solomon–Towsner framework for transforming a computable non-reducibility into a separation over \(\omega\)-models. Its forcing conditions are pairs \((\sigma,B)\), with \(B\) a “bad” set controlled through \(k\)-bushy trees, \(k\)-big and \(k\)-small sets, smallness closure, concatenation, and smallness additivity; the iterated construction then builds a Turing ideal closed under \(\mathrm{DNR}_h\) while preserving a computable locally \(2\)-colorable graph \(G\) with no solution in the ideal [1412.0917].

A distinct line of work calibrates the dependence of these reductions on induction. Over \(RCA_0\), for each fixed \(k\ge 2\), \(WKL\) is equivalent to \(\forall f\, DNR(k,f)\), and also to \(\forall f\, DNR(2,f)\). By contrast, over \(RCA_0 + B\Sigma^0_2\),
\[
RCA_0 + B\Sigma^0_2 + \exists k\,\forall f\, DNR(k,f)\ \nvdash\ WKL,
\]
and the finer strict hierarchy
\[
\forall f\,\exists k\,DNR(k,f) \;<\; \exists k\,\forall f\,DNR(k,f) \;<\; WKL
\]
holds over the same base. The paper further states a recursion-theoretic corollary: if a model satisfies \(B\Sigma^0_2\) but not \(I\Sigma^0_2\), then there exists some nonstandard \(k\) and a \(k\)-bounded DNR function that computes no \(2\)-bounded DNR function. In this account, \(\Sigma^0_2\)-induction is the exact missing ingredient for the classical reduction from \(k\)-bounded to \(2\)-bounded DNR to work uniformly [1401.3823].

Relativized DNC also organizes degree-theoretic constructions. There is an increasing \(\omega\)-sequence \((a_n)\) of Turing degrees forming an initial segment of the Turing degrees such that each \(a_{n+1}\) is diagonally noncomputable relative to \(a_n\). The proof iterates bushy-tree forcing through a hierarchy of forcing notions \(\mathbb P_n\) and coherent restriction maps, ensuring that each extension is a strong minimal cover of the preceding degree. The resulting corollary is that the reverse-mathematical principle \(\mathsf{DNR}\) does not imply the existence of Turing incomparable degrees [1504.02805].

Measure-theoretic refinements show that relativized diagonal non-computation is not confined to rare degree phenomena. For every sufficiently fast-growing computable \(h\), every \(2\)-random real computes some \(f\in DNC_h\) which does not compute any Martin-Löf random real; more strongly, for any real \(X\) and sufficiently fast-growing computable \(h\), every real \(Z\) that is both \(X\)-random and \(2\)-random computes a function \(f\in DNC_h^X\) which itself computes no Martin-Löf random real. The proofs combine Kautz-style “fireworks” arguments with bushy tree forcing inside the \(h\)-bushy tree
\[
T=\{\sigma\in\omega^{<\omega} : (\forall i)\,\sigma(i)<h(i)\},
\]
using the DNC-bad set
\[
B_{DNC}=\{\sigma\in\omega^{<\omega} : (\exists e<|\sigma|)\,\varphi_e(e)\downarrow=\sigma(e)\}
\]
and a measure estimate that turns bushy-tree smallness into positive probability of avoiding bad strings [1411.6846].

## 3. Axiomatic explanations of relativization

An axiomatic account of relativized computability isolates the exact structural ingredients under which diagonal and non-computability arguments relativize. Let \(K\) be a class of partial functions with natural-number arguments and values. The first axiom requires that \(K\) contain all partial recursive functions and be closed under substitution, primitive recursion, and the \(u\)-operator. The second axiom, the computation-record axiom, says that for every unary \(f\in K\) there exist a set \(M\subseteq \mathbb N\) and functions \(a,\beta\in K\) whose domains contain \(M\), such that the characteristic function of \(M\) belongs to \(K\) and \(f(x)=y\) iff there exists \(m\in M\) with \(a(m)=x\) and \(\beta(m)=y\). The third axiom is a programs axiom: there exists a binary universal function \(F\in K\) such that for every unary \(f\in K\) there is some code \(n\) with \(F(n,\cdot)=f\) [1811.06259].

The main theorem states that if \(K\) satisfies conditions 1–3, then there exists a set \(A\subseteq \mathbb N\) such that \(K\) is exactly the class of all partial functions partial recursive relative to \(A\). Equivalently,
\[
K = \{ f : f \text{ is partial recursive relative to } A \}.
\]
The proof develops \(K\)-enumerable and \(K\)-decidable sets, shows that domains, ranges, and graphs of \(K\)-functions are \(K\)-enumerable, introduces \(K\)-enumeration reducibility, and proves that if a set is \(K\)-enumeration reducible to a \(K\)-enumerable set, then it is \(K\)-enumerable. It then shows, first, that a partial function belongs to \(K\) iff it is partial recursive relative to the universal function \(F\), and, second, that the relevant oracle can be coded by a set \(S\oplus S^c\) so that enumeration reducibility to \(S\oplus S^c\) is equivalent to recursive enumerability relative to \(S\) [1811.06259].

Within this framework, relativized diagonal non-computation is not a separate phenomenon but a consequence of the same closure, universality, and enumeration machinery used in the unrelativized setting. The paper explicitly identifies standard results such as the non-existence of a total decider for the halting problem relative to \(A\), relativized versions of the recursion theorem, and relativized diagonal non-computation statements as results whose proofs go through because they depend only on these axioms. This suggests that the persistence of oracle-relative diagonal arguments is explained structurally rather than by case-by-case simulation [1811.06259].

## 4. Subrecursive relativization and relative incompressibility

A different relativization program internalizes diagonal non-computation inside total, time-bounded subsystems of a universal machine. A Turing submachine is defined from a total program \(P_f\) by
\[
(U/f)(w)=U(P_f\circ w),
\]
and a time-bounded submachine \(U_{P_T}\) is obtained when \(P_T\) computes a total time bound. A concrete construction uses a program \(P_{SM}\) that returns a default value if \(U(w)\) does not halt within time \(\le U(P_T\circ w)\), and otherwise shifts the output label by one; this yields
\[
U_{P_T}(w)=U(P_{SM}\circ P_T\circ w).
\]
Because \(P_T\) is total, \(U_{P_T}\) is total. The paper then defines the Busy Beaver Plus function relative to a submachine \(U_{P'_T}\) by
\[
BB^+_{P'_T}(N)=\max\{\,U_{P'_T}(w): |w|\le N\,\}+1,
\]
so the output is strictly above the value of every program of size at most \(N\) running on that same submachine [1612.05221].

The corresponding relative halting probability is a time-bounded analogue of Chaitin’s \(\Omega\):
\[
\Omega_{P'_T}=\sum_{\substack{p\ \text{program in }L\\ U(P'_T\circ p)\ \text{halts}}}2^{-|p|}.
\]
The paper introduces a self-referential submachine \(U_{P^{**}_T\circ P_T}\) and proves that if \(P_T\) is total, then \(U(P^{**}_T\circ P_T\circ w)\) is well-defined for every \(w\in L\); hence \(U_{P^{**}_T\circ P_T}\) is itself a Turing submachine. It then proves that for every natural number \(N\) there exists a program of the form
\[
\pi'_{\Omega\circ P^{**}_T\circ P_T\circ 0^{|\rho|}1\circ \rho}
\]
whose output on \(U_{P^{**}_T\circ P_T}\) is at least \(BB^+_{P^{**}_T\circ P_T}(N)\), with program length bounded by \(2N+C\) for some constant \(C\). This is the submachine-relative analogue of the theorem that sufficiently informative approximations to \(\Omega\) determine Busy Beaver values [1612.05221].

The principal incomputability statement is that for any total \(P'_T\), the function \(BB^+_{P'_T}(N)\) is uncomputable by any program running on \(U_{P'_T}\). The paper states this as a strict dominance condition: for every program \(P\), there exists \(N_0\) such that for all \(N\ge N_0\),
\[
U_{P'_T}(P\circ N)<BB^+_{P'_T}(N),
\]
and similarly with \(P*N\). Since no program of size \(\le N\) can output a value \(\ge BB^+_{P'_T}(N)\), the function is also relatively incompressible on that submachine. Here relativized diagonal non-computation is the claim that a function may be computable in the ambient universal machine while remaining uncomputable and incompressible relative to the submachine that serves as the local computational universe [1612.05221].

## 5. Diagonal-free relativized semantics and non-definability

In algebraic logic, related phenomena arise when diagonal elements are removed and semantics are relativized. For any ordinal \(\alpha\ge 2\), the class \(\mathsf{Drs}_\alpha\) consists of representable relativized diagonal-free set algebras of dimension \(\alpha\), built from structures
\[
\big(P(V), \cap, \cup, \complement, \emptyset, V, c_i\big)_{i\in \alpha},
\]
where \(V\subseteq U^\alpha\) is nonempty and
\[
c_i X=\{\,f\in V:\exists g\in X\text{ with }f={}_i g\,\}.
\]
Because there are no diagonal elements \(d_{ij}\), equality is absent from the logic. The paper proves that the representable relativized class coincides with its equational axiomatization \(\mathsf{Dra}_\alpha\), that if \(X\neq \emptyset\) then the free algebra \(\mathfrak{Fr}_X \mathsf{Drs}_\alpha\) is atomless, and that in any free algebra the only zero-dimensional elements are \(0\) and \(1\). Its logical conclusion is that in first-order logic without equality on relativized semantics there is no finitely axiomatizable, complete and consistent theory [1807.00690].

The same paper explicitly presents this as a kind of relativized diagonal non-computation or non-definability. The absence of diagonals means that equality-like information cannot be recovered by the algebraic operations, and the triviality of zero-dimensional elements means that there are no nontrivial variable-independent propositions. The logical translation uses general assignment models \((M,V)\), with \(V\subseteq \mathrm{VAR}^M\) nonempty and quantifiers interpreted relative to \(V\), and yields finite schema axiomatizability, the finite model property, decidable validities, Craig interpolation, and Beth definability, while still excluding finitely axiomatizable complete consistent theories [1807.00690].

A broader blow-up-and-blur program connects these diagonal-free and relativized constructions to failures of omitting types. For finite \(n>2\), any class between \(S\Nr_n\CA_{n+3}\) and \(RCA_n\) is not atom canonical, and any class containing the class of completely representable algebras and contained in \(S_c\Nr_n\CA_{n+3}\) is not elementary. The paper proves that there is no finite variable universal axiomatization of several representable diagonal-free reducts of cylindric algebras of dimension \(n\), and that the omitting types theorem fails for finite variable fragments of first-order logic with and without equality even when severely relativized models are allowed. A central logical formulation is that there exists a countable consistent \(L_m\)-theory with a non-principal type that cannot be omitted in any \(m+3\)-flat model [1401.1103]. In this part of the literature, relativized diagonal non-computation becomes a problem of what cannot be defined, represented, or omitted once equality and full-square semantics are removed.

## 6. Reinterpretations, resource-bounded variants, and barrier claims

Some papers reinterpret the force of diagonal arguments rather than accepting the standard non-computability reading. One paper argues that reflexive and diagonal proofs of limitative theorems are being over-interpreted. It proposes an axiom schema
\[
\forall z \in Z,\; W(z)=W(z\wedge ap),
\]
where \(ap\) is the sentence “the sentence on the left of this conjunction is true,” and presents this as a “reflexivity blocker.” It also proposes a “large tester” \(T'\) intended to avoid the standard halting contradiction by running in parallel with a tester \(T\), and advances the formal thesis that “there exists a finite algorithm capable of computing all real numbers (and or binary strings) with an arbitrary precision (exactly in limit).” The core algorithm \(AS\) generates all finite binary strings stage by stage on a binary tree, and the paper argues that the diagonal argument “does not apply because obtained matrices are rectangular and complete.” The same work does not develop relativization in the standard complexity-theoretic sense, but it explicitly gestures toward a framework in which diagonal non-computation is blocked by additional structure internal to the system [1111.6954].

A resource-bounded reformulation treats diagonal obstruction as a failure of bounded self-certification. In this framework, if a machine \(D\) correctly decides whether \(A\) halts within \(T\) steps, then \(D\) requires at least \(T+1\) steps. The paper turns this \(+1\) overhead into an operator \(F\) on a domain \(D\) of partial halting observations, proves that \(F\) is monotone and Scott-continuous, and defines an ascending chain
\[
p_0 \sqsubseteq p_1 \sqsubseteq p_2 \sqsubseteq \cdots,\qquad p_{i+1}=F(p_i),
\]
whose Scott supremum is
\[
p_\omega=\operatorname{lfp}(F)=\bigsqcup_{n<\omega}F^n(p_0).
\]
No bounded observation is a fixed point, but the least fixed point exists at the \(\omega\)-limit. The paper explicitly states that it does not develop a formal oracle-relative theory, yet it suggests an analogy: relative to a finite bound \(T\), the machine cannot decide its own halting, whereas relative to the \(\omega\)-limit of iterated self-observation, the fixed point is obtained [2603.06012].

In oracle complexity theory, another line of work relocates the obstacle to diagonalization from self-reference to enumerability. In the Baker–Gill–Solovay setting, the papers claim that diagonalization requires three prerequisites: effective representation of oracle machines as strings, existence of a universal nondeterministic oracle machine that can simulate deterministic oracle machines and flip their answers, and bounded simulation overhead of order \(O(T(n)\log T(n))\). On this basis they argue that if
\[
\mathcal P^O=\mathcal{NP}^O,
\]
then the set \(P^O\) of polynomial-time deterministic oracle Turing machines with oracle \(O\) is not enumerable; therefore a Cantor- or Turing-style diagonalization over \(P^O\) cannot be carried out [2112.03677], [2110.06211]. This yields a diagnosis of the relativization barrier that differs sharply from the axiomatic account of relativized computability: rather than explaining why diagonal arguments survive oracle extension, it treats non-enumerability of the target class as the reason a universal-machine diagonal construction does not go through in that setting [1811.06259].

Across these lines of work, “relativized diagonal non-computation” names not one theorem but a cluster of precise technical programs. In the oracle-DNC tradition it concerns functions avoiding \(\Phi_e^X(e)\); in reverse mathematics it calibrates the strength of existence principles such as \(\mathrm{DNR}_h\); in axiomatic recursion theory it is explained by closure, computation records, and universality; in subrecursive settings it becomes relative incomputability and incompressibility against a local machine universe; and in diagonal-free algebraic logic it appears as a failure of definability, atomicity, and omitting types under relativized semantics. The more revisionary papers preserve the vocabulary of diagonalization while revising either its interpretation or its prerequisites, especially through limit computation, self-certification at the \(\omega\)-stage, or oracle-dependent claims about enumerability [1111.6954], [2603.06012], [2112.03677].

Source: https://www.emergentmind.com/topics/relativized-diagonal-non-computation