The first effective level of genericity asks a set to meet every dense set of strings that can be enumerated by an algorithm. These weakly 1-generic sets are not computable, and Kurtz showed that the degrees that compute one are exactly the degrees computing a function that escapes domination by every computable function. Genericity, a topological notion, thus coincides at this level with a notion of growth. This lesson proves Kurtz’s theorem and introduces the hierarchy of weak -genericity.
2.1 Definition
We fix a computable enumeration of the c.e. sets of strings, through codes of strings, with finite approximations computable from and increasing in .
A set is weakly 1-generic if it meets every dense c.e. set of strings. More generally, is weakly -generic if it meets every dense set of strings.
There are countably many c.e. sets of strings, so weakly 1-generic sets exist, and the class of weakly -generic sets is co-meager for every (Proposition 1.6). A weakly 1-generic set is not computable: for a computable , the set is dense and computable. Since the family of dense sets contains the dense sets, weak -genericity implies weak -genericity.
2.2 Hyperimmune functions
A function is hyperimmune if no computable function dominates it, that is, for every computable there is with . A set has hyperimmune degree if it computes a hyperimmune function.
Equivalently, a set has hyperimmune degree if and only if it is not computably dominated, that is, if some total function it computes is dominated by no computable function (Definition 3.3 of “Basis theorems”). A hyperimmune function exceeds every computable function at infinitely many points: if for all , a finite modification of dominates everywhere. These functions and the corresponding sets are studied in the course on immunity and growth of functions (Immune and hyperimmune sets). Every non-computable set has hyperimmune degree, since it computes its computation function, which no computable function dominates (Martin–Miller).
2.3 Kurtz’s theorem
For every set , the following are equivalent:
- has hyperimmune degree;
- computes a function that is equal to every computable function at infinitely many points;
- computes a weakly 1-generic set.
(3 ⇒ 2) Let be weakly 1-generic. For each , the set of strings with a at some position at least is dense and computable, so is infinite. Let be the number of zeros of that follow position before the next : the largest such that . It is -computable. Let be computable and . The set is dense and computable, so has a prefix with , and then . So equals at infinitely many points.
(2 ⇒ 1) If equals every computable function infinitely often, then exceeds every computable function at some point.
(1 ⇒ 3) Let be hyperimmune. We build with oracle , one bit per stage: at stage , the prefix is known. Say that is met at stage if some prefix of belongs to for some ; once met, stays met, since . The construction may have a target: a pair with and .
Stage . Look for the least such that is not met at stage and some belongs to . If there is one, and there is no current target or the current target has an index larger than , make the current target, for the first such . Then let be the bit of the current target at position if the target is longer than , and otherwise. When the target is a prefix of , it is achieved, is met, and there is no current target.
The construction is computable in , so . We show that meets every dense c.e. set .
Each index sets a target finitely often. An index sets a target only when it is not met; its target is then copied bit by bit, and is achieved unless it is replaced by a target of smaller index. By induction on : if the indices smaller than set targets finitely often, then after they stop and their last targets are achieved or replaced, any target of is achieved, after which is met and never sets a target again.
Dense sets are met. Suppose is dense and is never met. Let be a stage after which no index smaller than sets a target and none of their targets is current. If at some stage some belonged to , then would be the least index with a candidate (a smaller index with a candidate would set a target at stage , since no target of smaller index is current), would set a target unless its own target is current, and this target would be achieved, so would be met. Hence, for every , no extension of belongs to . Now let be the least such that every string of length has an extension in ; is computable and total, since is dense and there are finitely many strings of length . Since has length , the previous observation gives for all , so a finite modification of dominates , contradicting the hyperimmunity of .
The construction of (1 ⇒ 3) is a finite injury argument without restraints: an index may lose its target to an index of smaller code, but only finitely often. The hyperimmune function serves as a bound on the search for extensions: a dense c.e. set that is never met would provide a computable bound on .
- Every non-computable set, in particular every non-computable c.e. set, computes a weakly 1-generic set.
- A computably dominated set computes no weakly 1-generic set; in particular, no weakly 1-generic set is computably dominated.
- No weakly 2-generic set is computable in .
- A non-computable set has hyperimmune degree, as recalled above.
- By the theorem, a set computing a weakly 1-generic set has hyperimmune degree.
- If , the set is dense and computable in , hence ; a weakly 2-generic set meets it, so it is not .
2.4 Exercises
Show that if is weakly 1-generic, then so is its complement .
Solution
For a set of strings , let , where flips all bits of . If is c.e. and dense, so is , since is a computable bijection preserving the prefix relation. meets , so meets .
Show that the principal function of a weakly 1-generic set is hyperimmune.
Solution
Recall from the proof of Theorem 2.3 that is infinite, so , the -th element of in increasing order (counting from ), is defined for every . Let be computable. The set is dense and computable, so has a prefix , with . At most elements of lie below , so the elements of index in increasing order cannot all lie below , and . No element of lies in the interval , so . Hence no computable function dominates .
2.5 Notes and references
Weakly 1-generic sets and the equivalence with hyperimmune degrees are due to Kurtz (1981), in his thesis on randomness and genericity; the presentation of (1 ⇒ 3) by one bit per stage follows the standard modern proofs. At the next level, Andrews, Gerdes and Miller showed that a set computes a weakly 2-generic set if and only if it computes a function dominated by no function. Downey and Hirschfeldt’s book treats these notions in detail.
- S. A. Kurtz, Randomness and genericity in the degrees of unsolvability, PhD thesis, University of Illinois at Urbana-Champaign, 1981.
- U. Andrews, P. Gerdes and J. S. Miller, The degrees of bi-hyperhyperimmune sets, Annals of Pure and Applied Logic 165 (2014), 803–811.
- R. G. Downey and D. R. Hirschfeldt, Algorithmic Randomness and Complexity, Springer, 2010.