Coarse Reducibility in Computability

• Coarse reducibility is a computability notion that compares sets or functions based on approximate presentations that differ only on a density-0 set.
• It distinguishes between uniform and nonuniform reducibility, connecting robust information coding with algorithmic randomness, genericity, and K-triviality.
• The framework embeds classical Turing degrees into a richer coarse degree structure and extends to functions and even state-space reductions in stochastic and quantum dynamics.

Searching arXiv for the cited papers and related coarse reducibility work. arXiv search query: "coarse reducibility algorithmic randomness (Hirschfeldt et al., 2015)" Coarse reducibility is a computability-theoretic notion of robust information coding in which sets, and more generally functions, are compared up to errors on a density-$0$ set. A coarse description of a set $A \subseteq \omega$ is a set $D \subseteq \omega$ such that $\rho(A \triangle D)=0$, and coarse reducibility asks whether every such approximate presentation of one object still suffices to recover an approximate presentation of another. In the form developed by Hirschfeldt, Jockusch, Kuyper, and Schupp, the subject studies both uniform and nonuniform reducibility, the induced degree structures, and their interaction with algorithmic randomness and genericity (Hirschfeldt et al., 2015). In a broader framework of robust reducibilities, coarse reducibility sits between generic reducibility and Turing reducibility (Dzhafarov et al., 2014), and later work extends the notion from subsets of $\omega$ to functions in $\omega^\omega$ while showing that the corresponding coarse degrees remain arithmetical and that every uniform coarse degree contains a set (Gerdes, 9 Aug 2025).

1. Definitions and basic framework

Let $\omega=\{0,1,2,\dots\}$ and $A\subseteq \omega$. For $n\in \omega$, write $[0,n)=\{0,1,\dots,n-1\}$ and

$A \subseteq \omega$0

The upper density and lower density are

$A \subseteq \omega$1

and when these agree one writes

$A \subseteq \omega$2

A set $A \subseteq \omega$3 is a coarse description of $A \subseteq \omega$4 if

$A \subseteq \omega$5

where $A \subseteq \omega$6 is the symmetric difference. Thus a coarse description is total but may disagree with the target on a sparse set (Hirschfeldt et al., 2015).

The later function-theoretic version replaces subsets of $A \subseteq \omega$7 by total functions $A \subseteq \omega$8. A total function $A \subseteq \omega$9 is a coarse description of $D \subseteq \omega$0 if

$D \subseteq \omega$1

For sets $D \subseteq \omega$2, nonuniform coarse reducibility is defined by

$D \subseteq \omega$3

iff every coarse description of $D \subseteq \omega$4 computes some coarse description of $D \subseteq \omega$5, while uniform coarse reducibility is defined by

$D \subseteq \omega$6

iff there is a single Turing functional $D \subseteq \omega$7 such that for every coarse description $D \subseteq \omega$8 of $D \subseteq \omega$9, $\rho(A \triangle D)=0$0 is a coarse description of $\rho(A \triangle D)=0$1. Uniform reducibility implies nonuniform reducibility, but not conversely (Hirschfeldt et al., 2015).

In the earlier robust-coding formulation, coarse reducibility is presented as a relativization of coarse computability. A set $\rho(A \triangle D)=0$2 is coarsely computable if there is a total computable function $\rho(A \triangle D)=0$3 such that $\rho(A \triangle D)=0$4 has asymptotic density $\rho(A \triangle D)=0$5. The relativized notion requires that whenever a total oracle $\rho(A \triangle D)=0$6 agrees with $\rho(A \triangle D)=0$7 on a set of density $\rho(A \triangle D)=0$8, the functional $\rho(A \triangle D)=0$9 is total and agrees with $\omega$0 on a set of density $\omega$1 (Dzhafarov et al., 2014). This suggests that coarse reducibility is best viewed as stability of computation under sparse oracle corruption rather than exact oracle access.

2. Uniformity, nonuniformity, and robust coding

The distinction between uniform and nonuniform reducibility is structural. In the uniform case, a single Turing functional must transform every coarse description of the source into a coarse description of the target. In the nonuniform case, different coarse descriptions of the source may compute coarse descriptions of the target by different methods. The basic implication

$\omega$2

holds immediately, but the two notions diverge in the coarse degrees (Hirschfeldt et al., 2015).

Within the larger family of robust reducibilities, coarse reducibility appears in the implication chain

$\omega$3

and separately

$\omega$4

No other implications hold among these notions. In particular, generic reducibility and coarse reducibility are incomparable, and neither is equivalent to ordinary Turing reducibility (Dzhafarov et al., 2014).

The same work records several natural variants. One may allow the reduction functional to depend on the particular oracle close to $\omega$5, yielding a non-uniform coarse reducibility. One may also consider a partial-oracle version, in which the output computation is allowed to diverge on a density-$\omega$6 set, or an equality-on-$\omega$7 requirement, in which one asks additionally that $\omega$8 exactly. These variants are explicitly noted but not developed there in detail (Dzhafarov et al., 2014). A plausible implication is that coarse reducibility was recognized from the outset as one point in a broader design space of density-robust reducibilities rather than as a single isolated definition.

3. Randomness, genericity, and K-triviality

A central theorem of Hirschfeldt, Jockusch, Kuyper, and Schupp states that if $\omega$9 is $\omega^\omega$0-random and $\omega^\omega$1, then $\omega^\omega$2 is $\omega^\omega$3-trivial. Consequently, if $\omega^\omega$4 is weakly $\omega^\omega$5-random, then any $\omega^\omega$6 is computable. The same two-step conclusion holds when “$\omega^\omega$7 is weakly $\omega^\omega$8-random” is replaced by “$\omega^\omega$9 is $\omega=\{0,1,2,\dots\}$0-generic”: if $\omega=\{0,1,2,\dots\}$1 is $\omega=\{0,1,2,\dots\}$2-generic and $\omega=\{0,1,2,\dots\}$3, then $\omega=\{0,1,2,\dots\}$4 is computable (Hirschfeldt et al., 2015).

These results place $\omega=\{0,1,2,\dots\}$5-triviality at the boundary of information recoverable from all coarse descriptions of a random source. The paper also proves that for some $\omega=\{0,1,2,\dots\}$6-random $\omega=\{0,1,2,\dots\}$7, there is a noncomputable c.e. set $\omega=\{0,1,2,\dots\}$8 with $\omega=\{0,1,2,\dots\}$9. In the opposite direction, not every $A\subseteq \omega$0-trivial set arises this way: there is a $A\subseteq \omega$1-trivial $A\subseteq \omega$2 such that no $A\subseteq \omega$3-random $A\subseteq \omega$4 satisfies $A\subseteq \omega$5 (Hirschfeldt et al., 2015).

Earlier work already recorded a negative interaction between coarse reducibility and randomness or genericity in the relative setting: if $A\subseteq \omega$6 is $A\subseteq \omega$7-random, or even weakly $A\subseteq \omega$8-generic, relative to $A\subseteq \omega$9, then $n\in \omega$0 (Dzhafarov et al., 2014). Taken together, these results show that coarse reducibility is sharply constrained by randomness and genericity, but not in a way that collapses entirely to computability. The appearance of $n\in \omega$1-triviality indicates that the residual information preserved under coarse descriptions of a random set is exactly the low-complexity portion singled out by algorithmic randomness theory.

4. Cone-avoiding compactness and proof methods

The main technical tool in “Coarse Reducibility and Algorithmic Randomness” is a compactness theorem for cone-avoiding descriptions. Its statement is as follows: let $n\in \omega$2 be arbitrary. Suppose that for every $n\in \omega$3 there is a set $n\in \omega$4 with

$n\in \omega$5

and

$n\in \omega$6

Then there exists $n\in \omega$7 with

$n\in \omega$8

and

$n\in \omega$9

In other words, one can amalgamate approximate descriptions of $[0,n)=\{0,1,\dots,n-1\}$0 into a single coarse description while still avoiding the Turing cone above $[0,n)=\{0,1,\dots,n-1\}$1 (Hirschfeldt et al., 2015).

The proof proceeds by a finite-extension, forcing-style construction. At stage $[0,n)=\{0,1,\dots,n-1\}$2 one has an initial segment $[0,n)=\{0,1,\dots,n-1\}$3 of $[0,n)=\{0,1,\dots,n-1\}$4 and chooses a longer $[0,n)=\{0,1,\dots,n-1\}$5 so that $[0,n)=\{0,1,\dots,n-1\}$6 remains $[0,n)=\{0,1,\dots,n-1\}$7-close to $[0,n)=\{0,1,\dots,n-1\}$8 on its length and forces failure of the $[0,n)=\{0,1,\dots,n-1\}$9-th attempted computation of $A \subseteq \omega$00. If neither of two good-extension options ever applies, then for sufficiently small $A \subseteq \omega$01 every string sufficiently close to $A \subseteq \omega$02 would compute $A \subseteq \omega$03, contradicting the hypothesis that $A \subseteq \omega$04. The triangle inequality for Hamming distance and a careful choice of densities yield $A \subseteq \omega$05 in the limit (Hirschfeldt et al., 2015).

Later work on functions in $A \subseteq \omega$06 introduces another forcing notion tailored to coarse descriptions. Conditions are triples $A \subseteq \omega$07, where $A \subseteq \omega$08 selects a finite error-set of density at most $A \subseteq \omega$09 and $A \subseteq \omega$10 is a finite partial function giving proposed values of $A \subseteq \omega$11 on the support of $A \subseteq \omega$12. The forcing is used to show that $A \subseteq \omega$13 iff some finite condition forces that every extension computes, via some oracle index $A \subseteq \omega$14, a total oracle whose symmetric-difference with $A \subseteq \omega$15 has density $A \subseteq \omega$16 (Gerdes, 9 Aug 2025). This suggests that the nonuniform theory can often be reduced to finite combinatorial control together with uniform density bounds.

5. Degree structure, embeddings, and minimal pairs

A map $A \subseteq \omega$17 is called a coarsening if for every $A \subseteq \omega$18 one has $A \subseteq \omega$19 and every coarse description $A \subseteq \omega$20 of $A \subseteq \omega$21 satisfies $A \subseteq \omega$22. It is uniform if there is a single functional $A \subseteq \omega$23 so that from any coarse description $A \subseteq \omega$24 of $A \subseteq \omega$25 the computation of $A \subseteq \omega$26 from $A \subseteq \omega$27 converges in a stage-wise uniform manner. A canonical example uses intervals $A \subseteq \omega$28 and defines

$A \subseteq \omega$29

To obtain a uniform coarsening, one interleaves via

$A \subseteq \omega$30

and then sets

$A \subseteq \omega$31

The map $A \subseteq \omega$32 is a uniform coarsening (Hirschfeldt et al., 2015).

For any coarsening $A \subseteq \omega$33 and sets $A \subseteq \omega$34,

$A \subseteq \omega$35

and if $A \subseteq \omega$36 is uniform then also

$A \subseteq \omega$37

Thus $A \subseteq \omega$38 embeds the Turing degrees into the nonuniform coarse degrees, and into the uniform coarse degrees when $A \subseteq \omega$39 is uniform. The embedding is not surjective: for a $A \subseteq \omega$40-generic $A \subseteq \omega$41, the two natural coarsenings $A \subseteq \omega$42 and $A \subseteq \omega$43 are distinct in the uniform coarse degrees, in effect because $A \subseteq \omega$44 is not jump-autoreducible (Hirschfeldt et al., 2015).

Minimal-pair phenomena are also present. If $A \subseteq \omega$45 is not coarsely computable and $A \subseteq \omega$46 is weakly $A \subseteq \omega$47-random relative to $A \subseteq \omega$48, then no $A \subseteq \omega$49-computable coarse description of $A \subseteq \omega$50 exists. Hence if $A \subseteq \omega$51 is not coarsely computable and $A \subseteq \omega$52 is weakly $A \subseteq \omega$53-random relative to $A \subseteq \omega$54, then the nonuniform coarse degrees of $A \subseteq \omega$55 and $A \subseteq \omega$56 form a minimal pair, and so do their uniform coarse degrees. This fails at the level of $A \subseteq \omega$57-randomness in the nonuniform degrees: there exist $A \subseteq \omega$58 both $A \subseteq \omega$59-random for which the nonuniform coarse degrees do not form a minimal pair, and indeed one can construct a noncoarsely computable c.e. $A \subseteq \omega$60 with $A \subseteq \omega$61 (Hirschfeldt et al., 2015).

Earlier work showed that the coarse degrees form an upper-semilattice and that the join of two degrees is their ordinary Turing join: for any $A \subseteq \omega$62, $A \subseteq \omega$63 is the least coarse-upper-bound of $A \subseteq \omega$64 and $A \subseteq \omega$65 (Dzhafarov et al., 2014). It also exhibited an embedding of the mod-finite degrees into the coarse degrees via

$A \subseteq \omega$66

These facts situate coarse reducibility within a degree theory that is richer than Turing reducibility but still retains familiar algebraic features (Dzhafarov et al., 2014).

6. Extensions to functions and comparison with effective dense reducibility

The function-theoretic extension in $A \subseteq \omega$67 shows that every uniform coarse degree contains a set. Specifically, there are computable maps

$A \subseteq \omega$68

such that for all $A \subseteq \omega$69: $A \subseteq \omega$70 is a set, $A \subseteq \omega$71 is a uniform coarse reduction of $A \subseteq \omega$72 to $A \subseteq \omega$73, $A \subseteq \omega$74 is the identity on $A \subseteq \omega$75, and $A \subseteq \omega$76 is a uniform coarse reduction of $A \subseteq \omega$77 to $A \subseteq \omega$78. Hence every uniform coarse degree, and thus every nonuniform coarse degree, contains at least one set (Gerdes, 9 Aug 2025).

The same paper gives an arithmetic characterization of nonuniform coarse reducibility. One can show that

$A \subseteq \omega$79

iff there exist numbers $A \subseteq \omega$80 such that for every total $A \subseteq \omega$81 with

$A \subseteq \omega$82

the function $A \subseteq \omega$83 is total and

$A \subseteq \omega$84

Theorem 4.1 shows that $A \subseteq \omega$85 is a $A \subseteq \omega$86-predicate, equivalently $A \subseteq \omega$87, and the relation can be written by a $A \subseteq \omega$88 formula in the real parameters $A \subseteq \omega$89 (Gerdes, 9 Aug 2025).

This arithmetic behavior contrasts sharply with effective dense reducibility. In the coarse setting, every function-degree even uniformly contains a set-degree. In the effective dense setting, a $A \subseteq \omega$90-generic function $A \subseteq \omega$91 has no set $A \subseteq \omega$92 with $A \subseteq \omega$93, and the nonuniform effective dense relation is $A \subseteq \omega$94-complete as a relation of reals (Gerdes, 9 Aug 2025). A plausible implication is that coarse reducibility behaves as a comparatively tame generalization of ordinary degree theory, whereas effective dense reducibility lies substantially closer to descriptive-set-theoretic complexity.

7. Terminological extension in stochastic and quantum dynamics

The phrase “coarse reducibility” also appears in a distinct literature on state-space reduction for stochastic and quantum dynamics. There a coarse-graining of a finite Markov chain with state-space $A \subseteq \omega$95 is given by a partition into macro-blocks and a lumping map

$A \subseteq \omega$96

with Moore–Penrose pseudo-inverse $A \subseteq \omega$97 and projector $A \subseteq \omega$98. The coarse-graining reduces the dynamics precisely when the projected dynamics is again Markovian, equivalently when

$A \subseteq \omega$99

in which case the reduced generator is $D \subseteq \omega$00 (Kabernik, 2018).

In the quantum setting, a coarse-graining is defined from an isometric embedding

$D \subseteq \omega$01

and the map

$D \subseteq \omega$02

A quantum coarse-graining reduces the dynamics generated by $D \subseteq \omega$03 iff

$D \subseteq \omega$04

where $D \subseteq \omega$05 is the induced orthogonal projection; for unitary dynamics $D \subseteq \omega$06, this is equivalent to the condition that for every $D \subseteq \omega$07 in the span of the bipartition operators $D \subseteq \omega$08,

$D \subseteq \omega$09

The same paper further shows that symmetry-based reduction extends beyond strict Noether-type commutation: coarse-graining by symmetrization over a group $D \subseteq \omega$10 is compatible with $D \subseteq \omega$11 iff for every $D \subseteq \omega$12,

$D \subseteq \omega$13

equivalently in the algebra generated by the group (Kabernik, 2018).

This use of the term is conceptually related through error-tolerant or information-discarding reduction, but it is mathematically separate from the computability-theoretic theory. The shared terminology reflects a common concern with identifying which information is safely ignorable, while the formal objects, reducibility criteria, and applications are entirely different.