Cite

This course has no released version yet: the text at this address may change. Cite a released version when one exists.

/courses/degree-structure/exact-pairs
@misc{degree-structure-exact-pairs,
  title = {{The partial order of the Turing degrees}},
  howpublished = {Lesson 1 of \emph{Structure of the Turing degrees}, Computability Theory course notes},
  url = {/courses/degree-structure/exact-pairs},
  note = {Unreleased working version}
}

To cite a single statement, use the # link that appears next to its heading.

Lesson 1 of 3

The partial order of the Turing degrees

The Turing degrees form an upper semilattice with countable lower cones and no maximal element; every countable ideal has an exact pair; consequently the degrees do not form a lattice, and strictly increasing sequences have no least upper bound.

35 min2 questions
Before this lesson:The finite extension method (Oracles and Turing degrees) · The Turing jump (Oracles and Turing degrees)
Structure of the Turing degrees · lesson 1 of 3
  1. The partial order of the Turing degrees
  2. Embedding partial orders
  3. Minimal degrees

The Turing degrees D\mathcal D, ordered by Turing reducibility, form a partial order whose study began with Kleene and Post (1954). This course examines its global structure: which bounds exist, which partial orders appear inside it, and whether it has minimal elements above 0\obf. This first lesson gathers the elementary properties of D\mathcal D and proves Spector’s exact pair theorem, which shows that the degrees are far from forming a lattice.

For a set GG and k∈Nk \in \nat, the column G[k]G^{[k]} is the set {y:⟨k,y⟩∈G}\{ y : \langle k, y \rangle \in G \}. For sets A0,A1,…A_0, A_1, \dots, the set ⨁nAn\bigoplus_n A_n has columns AnA_n.

1.1 Elementary properties

Proposition 1.1.#
  1. 0\obf is the least degree, and every degree has at most countably many degrees below it.
  2. Every degree has continuum many degrees above it, and no degree is maximal.
  3. Every two degrees have a least upper bound, the degree of the join: D\mathcal D is an upper semilattice.
  4. Every countable set of degrees has an upper bound.
Proof of Proposition 1.1.
  1. Computable sets are computable from every set; lower cones are countable by Proposition 2.4 of “Turing reducibility and degrees”.
  2. Let AA be a set. The sets A⊕XA \oplus X, X∈2NX \in \cs, compute AA, and they are continuum many. Each degree contains only countably many sets, so they fall into continuum many degrees. Since only one of them is deg⁡T(A)\deg_T(A) itself, deg⁡T(A)\deg_T(A) is not maximal; an explicit degree strictly above it is deg⁡T(A′)\deg_T(A') (Theorem 3.2 of “The Turing jump”).
  3. This is Proposition 2.6 of “Turing reducibility and degrees”.
  4. Given representatives XnX_n, the set ⨁nXn\bigoplus_n X_n computes each of them.

An upper bound of a countable set of degrees given by (4) is not a least upper bound in general. The rest of the lesson shows that a strictly increasing sequence never has one.

1.2 Exact pairs

A non-empty set I\mathcal I of degrees is an ideal if it is closed downward and under ∨\vee. The countable ideals are exactly the sets of the form

I(A0,A1,… )={deg⁡T(X):X≤TA0⊕⋯⊕An for some n},\mathcal I(A_0, A_1, \dots) = \{ \deg_T(X) : X \leq_T A_0 \oplus \cdots \oplus A_n \text{ for some } n \},

the ideal generated by a sequence of sets: such a set is an ideal, and a countable ideal is generated by any sequence of representatives of its elements.

Definition 1.2.#

Two degrees a\abf and b\bbf form an exact pair for an ideal I\mathcal I if I={c:c≤a and c≤b}\mathcal I = \{ \cbf : \cbf \leq \abf \text{ and } \cbf \leq \bbf \}.

So a\abf and b\bbf are upper bounds of I\mathcal I, and every common lower bound of a\abf and b\bbf belongs to I\mathcal I.

Theorem 1.3 (Spector).#

Every countable ideal of Turing degrees has an exact pair.

Proof of Theorem 1.3.

Let I\mathcal I be generated by A0,A1,…A_0, A_1, \dots, and let Bs=A0⊕⋯⊕As−1B_s = A_0 \oplus \cdots \oplus A_{s-1}, with B0=∅B_0 = \emptyset. We build G=⋃sgsG = \bigcup_s g_s and H=⋃shsH = \bigcup_s h_s as limits of strings, so that the column kk of GG and of HH agrees with AkA_k from some point on.

For strings σ⪯τ\sigma \preceq \tau, say that τ\tau is an ss-admissible extension of σ\sigma if τ(⟨k,y⟩)=Ak(y)\tau(\langle k, y \rangle) = A_k(y) for every position ⟨k,y⟩\langle k, y \rangle with k<sk < s and ∣σ∣≤⟨k,y⟩<∣τ∣|\sigma| \leq \langle k, y \rangle < |\tau|. Columns k<sk < s are locked beyond σ\sigma: an admissible extension must copy AkA_k there. Admissibility of τ\tau over σ\sigma is decidable with oracle BsB_s, uniformly in ss.

Fix an enumeration s↦(is,js)s \mapsto (i_s, j_s) of all pairs of indices. Start with g0=h0=εg_0 = h_0 = \varepsilon. At stage ss:

  1. If there are xx and ss-admissible extensions g′g' of gsg_s and h′h' of hsh_s with Φisg′(x) ⁣ ⁣↓≠Φjsh′(x) ⁣ ⁣↓\Phi^{g'}_{i_s}(x)\converge \neq \Phi^{h'}_{j_s}(x)\converge, take such g′,h′g', h'. Otherwise let g′=gsg' = g_s and h′=hsh' = h_s.
  2. Extend g′g' and h′h' by one bit to obtain gs+1g_{s+1} and hs+1h_{s+1}: at a position p=⟨k,y⟩p = \langle k, y \rangle, the new bit is Ak(y)A_k(y) if k≤sk \leq s, and 00 otherwise.

So gs+1g_{s+1} is an ss-admissible extension of gsg_s, and hs+1h_{s+1} of hsh_s. Since tt-admissibility implies ss-admissibility for t≥st \geq s, every G ⁣↾mG \uh m with m≥∣gs∣m \geq |g_s| is an ss-admissible extension of gsg_s, and similarly for HH.

a=deg⁡T(G)\abf = \deg_T(G) and b=deg⁡T(H)\bbf = \deg_T(H) are upper bounds of I\mathcal I. Fix kk. At every stage s≥k+1s \geq k + 1, the new positions of GG in column kk receive the values of AkA_k, since all extensions made at stage ss are ss-admissible and the new bit follows AkA_k. So G(⟨k,y⟩)=Ak(y)G(\langle k, y \rangle) = A_k(y) for all positions ⟨k,y⟩≥∣gk+1∣\langle k, y \rangle \geq |g_{k+1}|. Hence AkA_k is a finite variant of G[k]G^{[k]}, and Ak≤TGA_k \leq_T G. The same holds for HH.

Common lower bounds belong to I\mathcal I. Let Y≤TGY \leq_T G and Y≤THY \leq_T H, say Y=ΦiG=ΦjHY = \Phi^G_i = \Phi^H_j, and let ss be the stage with (is,js)=(i,j)(i_s, j_s) = (i, j). At stage ss, no pair of admissible extensions with distinct convergent values was found: otherwise G⪰g′G \succeq g' and H⪰h′H \succeq h' would give ΦiG(x)≠ΦjH(x)\Phi^G_i(x) \neq \Phi^H_j(x). We compute YY with oracle BsB_s. Given xx, search for an ss-admissible extension g′g' of gsg_s with Φig′(x) ⁣ ⁣↓\Phi^{g'}_i(x)\converge, and output the value vv.

  • The search terminates: G ⁣↾mG \uh m is an ss-admissible extension of gsg_s for every m≥∣gs∣m \geq |g_s|, and ΦiG ⁣↾m(x) ⁣ ⁣↓\Phi^{G \uh m}_i(x)\converge for large mm.
  • The output is Y(x)Y(x): similarly, H ⁣↾mH \uh m is an ss-admissible extension of hsh_s with ΦjH ⁣↾m(x) ⁣ ⁣↓=Y(x)\Phi^{H \uh m}_j(x)\converge = Y(x) for large mm. Since no admissible pair gives distinct values, v=Y(x)v = Y(x).

So Y≤TBsY \leq_T B_s, and deg⁡T(Y)∈I\deg_T(Y) \in \mathcal I.

Each requirement is handled once, at a single stage: either it is satisfied by a disagreement, or the absence of any disagreement among admissible extensions gives a procedure computing the common value from the locked columns alone. The locking of columns makes GG and HH upper bounds; it also limits the freedom of later extensions, which is why the procedure needs the oracle BsB_s.

1.3 Consequences

Corollary 1.4 (Kleene, Post; Spector).#
  1. A strictly increasing sequence d0<d1<⋯\dbf_0 < \dbf_1 < \cdots of degrees has no least upper bound.
  2. D\mathcal D is not a lattice: there are two degrees without a greatest lower bound.
Proof of Corollary 1.4.

Let Dn∈dnD_n \in \dbf_n, and let a,b\abf, \bbf be an exact pair for the ideal generated by (Dn)(D_n). Since the sequence increases, D0⊕⋯⊕Dn≡TDnD_0 \oplus \cdots \oplus D_n \equiv_T D_n, so this ideal is the set of degrees c\cbf such that c≤dn\cbf \leq \dbf_n for some nn.

  1. Suppose l\mathbf l is a least upper bound of the sequence. The degrees a\abf and b\bbf are upper bounds of the sequence, so l≤a\mathbf l \leq \abf and l≤b\mathbf l \leq \bbf. Hence l\mathbf l is in the ideal: l≤dn\mathbf l \leq \dbf_n for some nn. But dn+1≤l\dbf_{n+1} \leq \mathbf l, so dn+1≤dn\dbf_{n+1} \leq \dbf_n, a contradiction.
  2. Every common lower bound c\cbf of a\abf and b\bbf is below some dn\dbf_n, hence strictly below dn+1\dbf_{n+1}, which is another common lower bound. So no common lower bound is the greatest.

For instance, the arithmetic degrees, those below some 0(n)\obf^{(n)}, are exactly the common lower bounds of some pair of degrees, although they have no least upper bound. Exact pairs are a basic tool for coding countable structures into D\mathcal D, used in the proof that the first-order theory of D\mathcal D is as complex as true second-order arithmetic (Simpson, 1977).

1.4 Exercises

Exercise 1.5 ★.

Show that the arithmetic degrees, the degrees below 0(n)\obf^{(n)} for some nn, have no least upper bound, and that there are two degrees whose common lower bounds are exactly the arithmetic degrees.

Solution

The sequence 0(n)\obf^{(n)} is strictly increasing (Theorem 3.2 of “The Turing jump”), so Corollary 1.4 (1) applies. The arithmetic degrees form the ideal generated by (∅(n))n(\emptyset^{(n)})_n, since ∅(0)⊕⋯⊕∅(n)≡T∅(n)\emptyset^{(0)} \oplus \cdots \oplus \emptyset^{(n)} \equiv_T \emptyset^{(n)}. Apply Theorem 1.3.

Exercise 1.6 ★★.

Show that in Theorem 1.3, the sets GG and HH can be chosen computable in (⨁nAn)′\big(\bigoplus_n A_n\big)'.

Solution

Let B=⨁nAnB = \bigoplus_n A_n; it computes each BsB_s uniformly in ss. The question asked at stage ss, whether there exist xx and ss-admissible extensions with distinct convergent values, is Σ10\Sigma^0_1 relative to BsB_s, uniformly in ss, gsg_s and hsh_s, hence decided by B′B'. When the answer is positive, the witnesses are found by a BB-computable search. The extension by one bit uses Ak(y)A_k(y) for k≤sk \leq s, computable in BB. So the sequences (gs)(g_s) and (hs)(h_s) are computable in B′B'.

1.5 Notes and references

Kleene and Post (1954) proved that the degrees form an upper semilattice that is not a lattice, and that increasing sequences have no least upper bound. Spector (1956) proved the exact pair theorem in the same paper as the existence of minimal degrees. Lerman’s monograph is the standard reference on the global structure of D\mathcal D.

  • S. C. Kleene and E. L. Post, The upper semi-lattice of degrees of recursive unsolvability, Annals of Mathematics 59 (1954), 379–407.
  • C. Spector, On degrees of recursive unsolvability, Annals of Mathematics 64 (1956), 581–592.
  • M. Lerman, Degrees of Unsolvability, Perspectives in Mathematical Logic, Springer, 1983.
  • S. G. Simpson, First-order theory of the degrees of recursive unsolvability, Annals of Mathematics 105 (1977), 121–139.