We now combine the natural numbers, covers, and truncations to formulate finiteness conditions on animae. There are two constructively different notions already for sets. A finite set comes with the mere existence of a cardinal, whereas a finitely enumerable set comes only with the mere existence of a finite surjection onto it. The distinction propagates to higher animae and must be kept visible in a theory that does not assume excluded middle.
For a natural number \(n\), we write \(\{1,\ldots,n\}\) for the standard set with \(n\) elements, with \(\{1,\ldots,0\}=\emptyset\).
11.6.1 Finite and finitely enumerable sets
Definition 11.6.1. Let \(S\) be a set. A cardinal presentation of \(S\) is a natural number \(n\) together with a bijection \[ \{1,\ldots,n\}\iso S. \] We say that \(S\) is finite if the anima of cardinal presentations of \(S\) is merely inhabited. We denote the resulting proposition by \(\mathtt{isFinite}(S)\).
Thus finiteness is a property rather than a choice of an enumeration or of a cardinal. It is often called Bishop finiteness when it is necessary to distinguish it from weaker constructive notions.
Definition 11.6.2. Let \(S\) be a set. A finite enumeration of \(S\) is a natural number \(n\) together with a cover \[ \{1,\ldots,n\}\longrightarrow S. \] We say that \(S\) is finitely enumerable if the anima of its finite enumerations is merely inhabited.
Every finite set is finitely enumerable, but the converse is not constructively valid. The missing ingredient is precisely decidable equality.
Proposition 11.6.3 (Finite sets and finite enumerations). A set is finite if and only if it is finitely enumerable and has decidable equality.
Proof
The standard sets \(\{1,\ldots,n\}\) have decidable equality, by induction on \(n\). Hence a cardinal presentation of \(S\) gives both a finite enumeration and decidable equality on \(S\). These are properties of \(S\), so the same conclusion follows from the mere existence of a cardinal presentation.
Conversely, assume that \(S\) has decidable equality. Given a finite enumeration \(f\colon\{1,\ldots,n\}\to S\), we may remove repetitions from the list \(f(1),\ldots,f(n)\): at each step, decidable equality determines whether the next value has already occurred. This produces a natural number \(m\), an embedding \(\{1,\ldots,m\}\to S\), and the same image as \(f\). Since \(f\) is a cover, the resulting embedding is also a cover and hence an equivalence by Proposition 11.2.10. It is therefore a cardinal presentation of \(S\). Again, the desired conclusion is a proposition, so a merely given finite enumeration suffices.
Example 11.6.4. A proposition \(P\), regarded as a set, is finite if and only if it is decidable. Indeed, decidability identifies \(P\) either with the empty set or with the one-point set. Conversely, a cardinal presentation of \(P\) decides \(P\): cardinal zero gives \(\neg P\), and a positive cardinal gives a term of \(P\). Since decidability of a proposition is itself a proposition, the mere existence of a cardinal presentation suffices.
The following elementary closure properties will be used implicitly below.
Proposition 11.6.5 (Finite sums, products, and choice). Let \(S\) be a finite set and let \(T\colon S\to\Set\) be a family of finite sets. Then the dependent sum \(\sum_{s:S}T(s)\) and the dependent product \(\prod_{s:S}T(s)\) are finite. Moreover, for any family of animae \(X\colon S\to\An\), the canonical map \[ \truncation{-1}{\prod_{s:S}X(s)} \longrightarrow \prod_{s:S}\truncation{-1}{X(s)} \] is an equivalence.
Proof
All three assertions are propositions, so we may choose a cardinal presentation of \(S\) and argue by induction on its cardinal. We first prove the final assertion. For the empty set it is immediate. Adjoining one element reduces it to the canonical equivalence \[ \truncation{-1}{A\times B} \simeq \truncation{-1}{A}\times\truncation{-1}{B}. \] This equivalence follows by applying the induction principle for propositional truncation twice.
Apply the resulting finite-choice statement to the anima of cardinal presentations of each \(T(s)\). We may now, merely, choose cardinal presentations of all the \(T(s)\) simultaneously. The assertions about the dependent sum and dependent product then reduce to the corresponding elementary cardinal calculations for standard finite sets.
The last assertion is finite choice. It does not choose data from a merely inhabited arbitrary family. Rather, it says that finitely many merely inhabited animae have a merely inhabited product.
11.6.2 The recursive finiteness hierarchy
For an anima \(X\), its set truncation \(\truncation{0}{X}\) is its set of connected components. Finiteness of this set controls degree zero, while finiteness of the identity animae controls all higher degrees. This gives a definition that does not require homotopy groups or chosen basepoints.
Definition 11.6.6. Let \(X\) be an anima. We define recursively what it means for \(X\) to be \(n\)-finite, for \(n:\N\).
(1) The anima \(X\) is \(0\)-finite if it is a finite set.
(2) The anima \(X\) is \((n+1)\)-finite if its set of components \(\truncation{0}{X}\) is finite and, for every \(x,y:X\), the identity anima \(x=y\) is \(n\)-finite.
We denote the resulting proposition by \(\mathtt{isFinite}_n(X)\).
The word “proposition” in the last sentence is justified inductively. At height zero, finiteness was defined using propositional truncation. At the successor step, both conditions are products of propositions.
Proposition 11.6.7 (Basic properties of \(n\)-finite animae). Let \(n:\N\) and let \(X\) be an anima.
(1) If \(X\) is \(n\)-finite, then it is \(n\)-truncated.
(2) If \(X\) is \(n\)-finite, then it is \((n+1)\)-finite.
(3) Being \(n\)-finite is invariant under equivalence.
(4) Finite coproducts and finite products of \(n\)-finite animae are \(n\)-finite.
Proof
All four statements follow by induction on \(n\). At height zero, the first statement is part of the definition, the second uses decidable equality of finite sets, and the remaining statements follow from Proposition 11.6.5. At the successor step, apply the induction hypothesis to the identity animae. The set of components remains finite under equivalence and under finite sums and products. For products, use \[ (x_1,x_2)=(y_1,y_2) \simeq (x_1=y_1)\times(x_2=y_2), \] and for coproducts use that identity animae between distinct summands are empty.
Example 11.6.8. A finite set is \(n\)-finite for every \(n:\N\). If \(G\) is a finite group, its classifying anima \(\B G\) is \(1\)-finite: it is connected, and its identity anima at the distinguished object is the finite set \(G\). More generally, the recursive definition recognizes an \(n\)-truncated anima as \(n\)-finite precisely when its set of components and all its iterated identity animae are finite at the appropriate heights.
Definition 11.6.9. An anima \(X\) is \(\pi\)-finite if there merely exists a natural number \(n\) for which \(X\) is \(n\)-finite. In formulas, \[ \mathtt{isPiFinite}(X) \coloneqq \truncation{-1}{\sum_{n:\N}\mathtt{isFinite}_n(X)}. \]
The truncation in this formula matters. A proof of \(\pi\)-finiteness does not choose a truncation bound. By Proposition 11.6.7, every \(\pi\)-finite anima is merely \(n\)-truncated for some \(n\), has finitely many components, and has finite iterated identity animae. Conversely, these conditions, with a common finite truncation bound, recover the recursive definition. For a set, \(\pi\)-finiteness is equivalent to finiteness. In particular, Example 11.6.4 shows that a proposition is \(\pi\)-finite precisely when it is decidable.
11.6.3 Finite maps
Definition 11.6.10. A map \(f\colon X\to Y\) of animae is \(n\)-finite if each of its fibers is an \(n\)-finite anima. It is \(\pi\)-finite if each of its fibers is \(\pi\)-finite.
These are fiberwise properties, hence are stable under pullback. They are also stable under composition. We record the stronger dependent-sum formulation.
Proposition 11.6.11 (Dependent sums of finite animae). Let \(B\) be an \(n\)-finite anima and let \(E\colon B\to\An\) be a family of \(n\)-finite animae. Then its total anima \[ \sum_{b:B}E(b) \] is \(n\)-finite. Consequently, \(n\)-finite maps are closed under composition. The same assertions hold with “\(n\)-finite” replaced by “\(\pi\)-finite”.
Proof
For the statement about \(n\)-finite animae, the assertion follows by induction on \(n\). At height zero it is the dependent-sum assertion of Proposition 11.6.5. At the successor step, the identity anima between \((b,e)\) and \((b',e')\) is \[ \sum_{p:b=b'}\bigl(p_*e=e'\bigr). \] It is \(n\)-finite by the induction hypothesis. The same induction also shows that decidable subanimae of \(n\)-finite animae are \(n\)-finite. To verify the remaining component condition, reduce to the finitely many components of \(B\) and merely choose a representative \(b\) of each of them, using finite choice. The components of the total anima lying over the component of \(b\) are the quotient of the finite set \(\truncation{0}{E(b)}\) by transport along loops in \(B\) at \(b\). This orbit relation is decidable: for two components of \(E(b)\), the anima of loops carrying one to the other is a decidable subanima of the \(n\)-finite anima \(b=b\), so its mere inhabitation is decidable. A quotient of a finite set by a decidable equivalence relation is finite, by the same deduplication argument as in Proposition 11.6.3. Hence the total anima has finitely many components.
Now suppose that the base and fibers are \(\pi\)-finite. There merely exists a bound for the base. Over each of its finitely many components, there merely exists a bound for the corresponding fiber. Finite choice from Proposition 11.6.5 produces, merely, finitely many bounds; their maximum is a common bound. After increasing all bounds to this maximum by Proposition 11.6.7, the fixed-height case applies.
11.6.4 Finite covers and iterated diagonals
There is another natural way to express finiteness using only covers and diagonals. It is particularly useful because it makes sense for maps from the outset, but its constructive meaning differs from that of the recursive definition above.
Definition 11.6.12. An anima \(X\) is finitely covered if there is merely a finite set \(S\) and a cover \(S\to X\). A map of animae is finitely covered if all its fibers are finitely covered.
For a map \(f\colon X\to Y\), put \(\Delta^0(f)=f\) and define its iterated relative diagonals recursively by \[ \Delta^{k+1}(f)=\Delta\bigl(\Delta^k(f)\bigr). \] We say that \(f\) satisfies the Kuratowski finiteness condition if \(\Delta^k(f)\) is finitely covered for every \(k:\N\).
For a set \(S\), being finitely covered is exactly being finitely enumerable. It is therefore weaker than finiteness unless equality in \(S\) is decidable. The same issue occurs at every stage of the iterated-diagonal condition: the fibers detect finite enumerability of component sets, but do not supply their decidable equality.
In the ordinary category of animae, classical logic erases this distinction. Anel proves that an anima has finitely many components and finite homotopy groups in every degree if and only if its terminal map satisfies the Kuratowski finiteness condition [Ane25, Proposition 2.1.1]. He identifies these animae with the coherent animae, and identifies the \(\pi\)-finite animae with the truncated coherent animae. Thus, in the classical model, one may recognize \(\pi\)-finite animae as the truncated animae whose terminal maps and all iterated diagonals are finitely covered.
Constructively, the recursive definition in Definition 11.6.6 is the stronger one and is the convention we adopt. Every recursively \(n\)-finite anima is \(n\)-truncated and satisfies the Kuratowski finiteness condition. A converse requires decidable equality for the component sets detected at the successive stages, or a classicality principle that supplies it. Moreover, in an arbitrary model of synthetic category theory the internally defined Kuratowski condition should not be identified with categorical coherence without additional hypotheses on the ambient model.