4.8. Limits and colimits

Axiom of dependent choice (DC)

The well-foundedness of a relation RE2 excludes the existence of a sequence uE such that ∀n∈ℕ, un+1 R un.
Indeed for any such sequence, Im uRIm u but Im u≠ ∅.

The axiom of dependent choice is a weak version of the axiom of choice, still stronger than the axiom of countable choice. Its multiple (equivalent) formulations, include the following.

  1. SetE, ∀RE2, ∀aE, Dom R = E ⇒ ∃uE, u0=a ∧ ∀n∈ℕ, un R un+1 (to any serial {0,S}-system there exists a {0,S}-morphism from ℕ).
  2. SetE, ∀RE2, Dom R = E ≠ ∅ ⇒ ∃uE, ∀n∈ℕ, un R un+1 (to any nonempty serial {S}-system there exists a {S}-morphism from ℕ).
  3. SetE, ∀RE2, (R not well-founded) ⇒ ∃uE, ∀n∈ℕ, un+1 R un.
  4. SetE, ∀R ∈ ℘(E2), (∀n∈ℕ, Im Rn⊂ Dom Rn+1) ⇒ ∀a∈Dom R0, ∃uE, u0=a ∧ ∀n∈ℕ, un Rn un+1
  5. SetE, ∀f ∈ ∏n∈ℕ EnEn+1, (∀n∈ℕ, Im fn = En) ⇒ ∀aE0, ∃u∈ ∏E, u0=a ∧ ∀n∈ℕ, un = fn(un+1) (to any serial {0,S}-draft there exists a {0,S}-morphism from ℕ).
Obviously (1. ⇒ 2.), (1. ⇔ 4.) and (4. ⇒ 5.).
2. ⇔ 3. is easy: to be not well founded, means ∃AE, A≠∅ ∧ Im(RA2)=A.
2. ⇒ 1. ∃u ∈ 〈{a}〉R, ∀n∈ℕ, un R un+1 to be composed with a sub-term of E with root u0.
5. ⇒ 1. using the sequence of sets (Mor{0,S}(Vn+1,E))n∈ℕ with the restriction functions.∎

4. ⇒ AC : from any sequence E of nonempty sets, let R = (En×En+1)n∈ℕ.
ACE ⇒ 1. : if fEE satisfies ∀xE, xRf(x) then u = (fn(a))n∈ℕ fits.

A generalized formulation of DC uses an R⊂(∐n∈ℕEnE, to modify 1. by the more sophisticated condition ∀n∈ℕ, (uk)k<n R un. It can be deduced from 5. like with the above proof of 5. ⇒ 1.

Countable downward Löwenheim–Skolem theorem

DC is equivalent to the following statement: for any countable algebraic language L and any L-system E, there exists a countable FE such that LF ∩ Dom EEF.
This condition on F can also be written Dom F = LF ∩ Dom E where F is seen as a subsystem by F = E ∩ (LF × F).

Proof. Assuming DC, let K ≠ ∅ the set of countable subsets of E, and

R = {(A,B)∈K2 | ABLA ∩ Dom EEB}

Dom R = K because, applying countable choice to the countable set LA ∩ Dom E,

f : LA ∩ Dom EE, Gr fE ∴ (A, A∪ Im f)∈R

Hence by DC 2.

AK, ∀n∈ℕ, AnAn+1LAn ∩ Dom EEAn+1

Taking then F = ⋃n∈ℕ An,

∀(s,u)∈LF, ∃n∈ℕ, Im uAn
LF ∩ Dom E = ⋃n∈ℕ LAn ∩ Dom EEn∈ℕ An+1 = EF

Conversely, if a serial {0,S}-system (E,a,R) has a countable serial subsystem F then there is a {0,S}-morphism from ℕ to F and thus to E because a bijection between F and ℕ allows to define ∀n∈ℕ, un+1 = min{xF | un R x}.∎

In particular if E is an L-algebra then F is a countable subalgebra of E (namely the minimal subalgebra of E if following the construction starting with ∅).

The downward Löwenheim–Skolem Theorem, says that under the axiom of choice, any infinite system E with countable language, has elementary subsystems (3.4.) with any infinite cardinalities from |ℕ| to |E|. We did not work enough with infinite cardialities to reach that generality, but from the above, we get the existence of a countable elementary subsystem F, as an equivalent formulation of DC.
Indeed a formula in prenex form with parameters in F and true in E, will stay true in F just if the truth of all its existential quantifiers is preserved; and this can be written LF ∩ Dom EEF after converting existential quantifiers into operation symbols added to L.

Reference : paper by Asaf Karagila - message and paper by Christian Espindola.

Limits in categories

The concept of limit generalizes those of product, equalizer and wide fiber product. Let us formalize it based on a slightly different conception of a diagram than usually done elsewhere.

Let us define a diagram D over a given category C, as the data of

Then a cone from an object N of C to D, is a family of arrows ψ ∈ ∏iI Mor(N, Xi) such that

i,jI, ∀fKij, f∘ψi = ψj

The cones to D form a co-action of C, namely the sub-co-action of the product of the C(Xi), given by the intersection of equalizers Eq(f∘πi, πj) for all i,jI and fKij. Then a co-egg of this co-action is called a limit of D. It is in the same way an intersection of equalizers inside a product in C, if these concepts are well-defined there.

From the results of 3.9 and 3.10, for any morphism b in C, any limit of a diagram made of b-modules is also a b-module.

The condition remains unchanged when replacing D by the small category it generates, i.e. with I as set of objects, and the sets Kij, once completed with the composites of their elements and the identity elements, become its sets of arrows. For this reason, without loss of generality, a diagram can be assumed to be a small category. At least, let us qualify a diagram as stable if it is stable by composition.

The condition for ψ to be a cone from N to a stable diagram D, can be rephrased saying the extended diagram D' = D ∪ ψ, namely with I' = I∪{N} ≠ I and K'Nj = {ψj}, remains stable; then N is an initial object of the resulting small category (after adding identity elements).

If D is already a small category with an initial object 0∈I then X0 naturally serves as a limit of D.

Projective limits

A category is called thin if each Mor set is either empty or a singleton.

A projective limit is the limit of a special kind of diagram, called a projective system. It is a nonempty diagram such that:

If I had a greatest element, this element would be an initial object of the diagram, thus giving the projective limit as an already given object of C. So for a projective limit to differ from the given objects, I must have no greatest element, and thus be infinite.

If JI is such that ∀iI, ∃jJ, ij then the natural projection from cones over I to cones over J, is bijective. It thus forms an isomorphism in C between projective limits. Examples :

Projective limits of sets

The construction of limits in the category of sets, usually forms the bulk of their construction in other concrete categories.

In the category of sets, over any fixed directed set (I,≤), there is equivalence between

  1. The limit of any projective system of finite sets where all fij are surjective, is surjective over all Xi
  2. The limit of any projective system of nonempty finite sets is nonempty.
Proof.
1. ⇒ 2. : if (Xi)iI is a family of nonempty finite sets then we get a projective system with surjective functions by restriction to

Yi = ⋂ij Im fijXi

which is non-empty, otherwise Xk would be empty for an upper bound k of a tuple jIXi such that

xXi, ijxx∉ Im fijx

To check that ∀i,jI, ijfij[Yj] = Yi, the proof of Yifij[Yj] is easy, while that of fij[Yj] ⊂ Yi uses directedness.

2. ⇒ 1. : assuming all fij surjective, for any iI and xXi, let

jI, Yj = ⋃{fjk[fik(x)] |kIikjk}

It is easy to see that

jI, Yj ≠ ∅
j,lI, jlfjl[Yl] ⊂ Yj
Yi = {x}.

The conclusion follows.∎

Another, less obvious option for 2. ⇒ 1. is

jI, Yj = {yXj|∀kI, (kikj) ⇒ fki(x) = fkj(y)}

Still, Yj ≠ ∅ because

mI, ∃zXm, imjmfim(z) = x
kI, (kikj) ⇒ fki(x) = fkm(z) = fkj(fjm(z))

j,lI, jlfjl[Yl] ⊂ Yj because

yYl, ∀kI, (kikj) ⇒ fki(x) = fkl(y) = fkj(fjl(y))

Still another way involves first taking the projective limit of the fij(x) over ≤(i).

The generalization of 1. to infinite sets is a version of the axiom of choice, hard to deduce except when assuming that I is countable, in which case it is easily equivalent to dependent choice (DC 5.).

2. can be seen as a generalized form of the completeness theorem of propositional logic, also known as the compactness theorem, which we quickly deduced and used in the countable case without the axiom of choice, near the end of our proof of the completeness theorem. To extend the completeness theorem to theories with uncountable language, involves at that step the (roughly) full version of 2. (using the axiom of choice except in special cases). The other needed change is, we need term algebras for uncountable languages. These may be constructed as synonymity classes of terms; another construction of these will be done below.

The generalization of 2. to infinite sets is false. A counter-example is given by I=ℕ, Xi=ℕ\Vi and fij = IdXj.

Inductive limits

A colimit in a category, is a limit in the opposite category.
In particular, an inductive limit in a category C, is a projective limit in the opposite category : given an inductive system made of a directed set (I,≤), a family of objects (Xi)iI, and a family of morphisms fij ∈ Mor(Xi, Xj) for all ijI such that ijkfjkfij = fik, its inductive limit is an egg of the action of C whose elements over any object N are the ϕ ∈ ∏iI Mor(Xi, N) such that

i,jI, ij ⇒ ϕjfij = ϕi

The construction of colimits and inductive limits in the category of sets, usually forms the bulk of their construction in other concrete categories.
Directly applying the definition of a colimit to an inductive system of sets, would express their inductive limit as the quotient of the disjoint union U = ∐iI Xi by the equivalence relation generated by ⋃ij Gr fij where fij is read as a function from {iXi to {jXj.
This generation problem is resolved in the case of inductive systems, by more directly expressing this equivalence relation as

{((i,x),(j,y))∈U2 | ∃k∈I, ikijfik(x) = fjk(y)}

Let us qualify an inductive system as injective if all fij are injective ; then all ϕi of the inductive limit are also injective.

As a first example, the ground term algebra over any infinite language L, is the inductive limit in the category of L-systems, with I = ℘fin(L) with the inclusion order, Xi a ground i-term algebra, and for any ij, fij the unique i-morphism from Xi to Xj (which is injective, and also an L-morphism).

A second example is the construction of objects with infinite basis in any concrete category, out of those with bases of all finite cardinalities. For any two objects with equinumerous bases, every bijection between bases is uniquely extensible as an isomorphism. Thus, all objects with a finite basis are essentially described, up to isomorphism, by a sequence of objects with basis Vn for every n∈ℕ. Now let B an infinite set, I = ℘fin(B) with the inclusion order, Xi an object with basis i for every iI, and fij the unique morphism extending Idi. This forms an injective inductive system, since, as mentioned earlier, all fij are sections.
It is easy to see that any object with basis B is its inductive limit in the given category, and that, in the general case, its inductive limit in the category of sets is naturally a subset of it.
In practice, both will usually coincide. In particular for categories of algebras, for any inductive system of algebras, its inductive limit as defined in the category of sets is also naturally an algebra. So the above constructed inductive limit of algebras with finite basis, is an algebra, and thus is also an object (and the inductive limit in that category) just if that category admits an object with basis B and if any subalgebra of an object is also an object.

Birkhoff's variety theorem

We saw that, for any given equational system over a fixed algebraic language, the class of its modules is stable by quotienting (H), picking stable subsets (S), and products (P). Birkhoff's variety theorem, also called HSP theorem, states a converse : any class C of algebras stable by H, S and P, is the class of all algebras which satisfy some fixed set of equations. Such a class is called an equational class or a variety.

(the proof makes crucial use of the axiom of infinity in the form of infinite products; if only assuming stability of a class of finite algebras by finite products, the result no more holds, although a counter-example may fail to be expressible in finite set theory, as its definition uses a quantifier in ℕ ; the theorem may then be replaced by the more complicated Reiterman's theorem involving another concept of equations with infinite sizes)

Let us split it in two easier statements:

  1. If (S) and (P) then C has free objects with basis of every cardinality, generated by their basis
  2. If (H) and C has free objects FB with basis B of any cardinality where FB is generated by B, then C is an equational class.
Proof 1. For any set B, let TB a B-ary term algebra, and Q the set of congruences R of T such that M/R is in C.

Q is stable by intersections because any intersection ⋂iI Ri of congruences from Q, is the congruence of

iI πi ∈ Mor(TB, ∏iI M/Ri)

whose image is a subset of a product, thus in C according to (S) and (P).

B is a basis of FB = TB/⋂Q in C because

The congruence ∼B (which was ⋂Q) of the natural morphism from a B-ary term algebra TB to the corresponding free object FB, is the set of all equations with variables from B which are satisfied by all algebras in C.
Remember that the class of "all equations satisfied by all algebras in C" can be seen as a set independent of B, because these equations are finite and the copies of those from objects with finite basis, by the previously mentioned injectivity between free objects.

Proof 2. Let N be any algebra which satisfies them.
Let B be any generating subset of N (possibly B = N).
The unique f∈ Mor(TB, N) satisfies ∼B ⊂ ∼f. So f/∼B ∈ Mor(FB, N) is a quotient.
By (H), since FB is in C we conclude N is in C.∎

To be completed later


Set theory and foundations of mathematics
1. First foundations of mathematics
2. Set theory
3. Algebra 1
4. Arithmetic and first-order foundations
4.1. Algebraic terms
4.2. Quotient systems
4.3. Term algebras
4.4. Integers and recursion
4.5. Presburger Arithmetic
4.6. Finiteness
4.7. Countability and Completeness
4.8. Limits and colimits
4.9. More recursion tools
4.10. Non-standard models of Arithmetic
4.11. Developing theories : definitions and constructions
4.A. The Berry paradox
5. Second-order foundations
6. Foundations of Geometry