Structure

Cocongruences in extensive categories

We start with a general fact which is easier to visualize for the dual case of congruences.

Lemma 1.

Let q1,q2:E⇉Xq_1, q_2 : E \rightrightarrows X be a congruence such that the pullback E×XEE \times_X E of q2q_2 and q1q_1 exists, and let t:E×XE→Et : E \times_X E \to E be the transitivity morphism. Then we have pushout diagrams

E×XE→p1E E×XE→p2Et↓↓q1 t↓↓q2E→q1X E→q2X. \begin{CD} E \times_X E @> p_1 >> E @. \quad @. E \times_X E @> p_2 >> E \\ @V t VV @VV q_1 V @. @V t VV @VV q_2 V \\ E @> q_1 >> X @. \quad @. E @> q_2 >> X. \end{CD}

Proof. Suppose we have a commutative square

E×XE→p1Et↓↓fE→gT. \begin{CD} E \times_X E @> p_1 >> E \\ @V t VV @VV f V \\ E @> g >> T. \end{CD}

Then for any generalized element (x,y)(x, y) of EE, consider the generalized elements ((x,x),(x,y))((x, x), (x, y)) and ((x,y),(y,y))((x, y), (y, y)) of E×XEE \times_X E. From the commutativity of the square acting on these generalized elements, we have f(x,x)=g(x,y)f(x, x) = g(x, y) and f(x,y)=g(x,y).f(x, y) = g(x, y). Therefore, if we define h:X→Th : X \to T on generalized elements by h(x)≔f(x,x)h(x) \coloneqq f(x, x), we must have f=h∘q1f = h \circ q_1 and g=h∘q1g = h \circ q_1. Conversely, if hh is a morphism with f=h∘q1f = h \circ q_1, we have h(x)=f(x,y)h(x)=f(x,y) for all (x,y)∈E(x,y) \in E, which for y=xy=x yields h(x)=f(x,x)h(x)=f(x,x).

This concludes the proof that the first diagram is a pushout diagram; the proof for the second is similar. □\square

Lemma 2.

Suppose we have two pullback diagrams

A1→f1B A2→f2Bg1↓↓h g2↓↓hC1→k1D C2→k2D \begin{CD} A_1 @> f_1 >> B @. \quad @. A_2 @> f_2 >> B \\ @V g_1 VV @VV h V @. @V g_2 VV @VV h V \\ C_1 @> k_1 >> D @. \quad @. C_2 @> k_2 >> D \end{CD}

in an extensive category C\C. Then we can combine these into a pullback diagram

A1+A2→(f1;f2)Bg1+g2↓↓hC1+C2→(k1;k2)D. \begin{CD} A_1 + A_2 @> (f_1; f_2) >> B \\ @V g_1 + g_2 VV @VV h V \\ C_1 + C_2 @> (k_1; k_2) >> D. \end{CD}

Proof. Suppose we have compatible maps T→BT \to B and T→C1+C2T \to C_1 + C_2. Then using extensivity, we can find a decomposition of the latter into morphisms T1→C1T_1 \to C_1 and T2→C2T_2 \to C_2 where T≅T1+T2T \cong T_1 + T_2. Using the individual pullback diagrams, we can then construct morphisms T1→A1T_1 \to A_1 and T2→A2T_2 \to A_2 with the corresponding projections to B,C1,C2B, C_1, C_2. The sum T≅T1+T2→A1+A2T \cong T_1 + T_2 \to A_1 + A_2 then gives the required morphism for the desired pullback diagram.

For uniqueness, suppose we have two generalized elements s,s′∈(A1+A2)(T)s, s' \in (A_1 + A_2)(T) such that (f1;f2)(s)=(f1;f2)(s′)(f_1; f_2)(s) = (f_1; f_2)(s') and (g1+g2)(s)=(g1+g2)(s′).(g_1 + g_2)(s) = (g_1 + g_2)(s'). Then we can find a decomposition of ss into generalized elements a1∈A1(T1)a_1 \in A_1(T_1) and a2∈A2(T2)a_2 \in A_2(T_2) where T≅T1+T2T \cong T_1 + T_2, and similarly a decomposition of s′s' into generalized elements a1′∈A1(T1′)a_1' \in A_1(T_1') and a2′∈A2(T2′)a_2' \in A_2(T_2') where T≅T1′+T2′T \cong T_1' + T_2'. From the condition (g1+g2)(s)=(g1+g2)(s′)(g_1 + g_2)(s) = (g_1 + g_2)(s'), we get that T≅T1+T2→g1(a1)+g2(a2)C1+C2T \cong T_1 + T_2 \xrightarrow{g_1(a_1) + g_2(a_2)} C_1 + C_2 and T≅T1′+T2′→g1(a1′)+g2(a2′)C1+C2T \cong T_1' + T_2' \xrightarrow{g_1(a_1') + g_2(a_2')} C_1 + C_2 are equal. From the equivalence of categories C/C1×C/C2≅C/(C1+C2)\C / C_1 \times \C / C_2 \cong \C / (C_1 + C_2), we can conclude that the slices T1→g1(a1)C1T_1 \xrightarrow{g_1(a_1)} C_1 and T1′→g1(a1′)C1T_1' \xrightarrow{g_1(a_1')} C_1 are isomorphic, and similarly for the slices T2→g2(a2)C2T_2 \xrightarrow{g_2(a_2)} C_2 and T2′→g2(a2′)C2T_2' \xrightarrow{g_2(a_2')} C_2. Without loss of generality, then, we may assume that T1=T1′T_1 = T_1' and T2=T2′T_2 = T_2'; the isomorphism T1+T2≅T≅T1′+T2′T_1 + T_2 \cong T \cong T_1' + T_2' is the identity; and g1(a1)=g1(a1′)g_1(a_1) = g_1(a_1') and g2(a2)=g2(a2′)g_2(a_2) = g_2(a_2').

On the other hand, from (f1;f2)(s)=(f1;f2)(s′)(f_1; f_2)(s) = (f_1; f_2)(s'), we conclude f1(a1)=f1(a1′)f_1(a_1) = f_1(a_1') and f2(a2)=f2(a2′)f_2(a_2) = f_2(a_2'). Therefore, from the uniqueness conditions on the given pullback diagrams, we get a1=a1′a_1 = a_1' and a2=a2′a_2 = a_2'. Hence, s=(a1;a2)=(a1′;a2′)=s′s = (a_1; a_2) = (a_1'; a_2') = s'. □\square

Corollary 3.

Let j1,j2:X⇉Ej_1, j_2 : X \rightrightarrows E be a cocongruence in an extensive category such that the pushout E+XEE +_X E of j2j_2 and j1j_1 exists, and let t:E→E+XEt : E \to E +_X E be the corresponding cotransitivity morphism. Then we have a pullback diagram

X+X→(j1;j2)Ej1+j2↓↓tE+E→E+XE. \begin{CD} X + X @> (j_1; j_2) >> E \\ @V j_1 + j_2 VV @VV t V \\ E + E @>>> E +_X E. \end{CD}

In particular, if the category also has pullback-stable regular epimorphisms, then for any cocongruence j1,j2:X⇉Ej_1, j_2 : X \rightrightarrows E such that the pushout of j2j_2 and j1j_1 exists, the corresponding morphism (j1;j2):X+X→E(j_1; j_2) : X + X \to E is a regular epimorphism.

Proof. The pullback diagram comes from using Lemma 2 on the two pullback diagrams from the dual of Lemma 1. For the other conclusion, all we need to establish is that E+E→E+XEE + E \to E +_X E is a regular epimorphism. However, this is immediate from the fact that the pushout E+XEE +_X E is also a coequalizer of i1∘j2,i2∘j1:X⇉E+Ei_1 \circ j_2, i_2 \circ j_1 : X \rightrightarrows E + E (cf. this implication). □\square

In the following results, define a regular corelation to be a corelation j1,j2:X⇉Ej_1, j_2 : X \rightrightarrows E such that the binary copower of XX exists, and (j1;j2):X+X→E(j_1; j_2) : X + X \to E is a regular epimorphism.

Lemma 4.

In an extensive category, every regular coreflexive corelation j1,j2:X⇉Ej_1, j_2 : X \rightrightarrows E such that the equalizer of j1j_1 and j2j_2 exists is an effective cocongruence.

Proof. Let e:Y↪Xe : Y \hookrightarrow X be the equalizer of j1j_1 and j2j_2. We will prove that the parallel pair

(i1;i1∘e;i2∘e;i2),(i1;i2∘e;i1∘e;i2):X+Y+Y+X⇉X+X \begin{align*} (i_1; i_1 \circ e; i_2 \circ e; i_2) & ,\\ (i_1; i_2 \circ e; i_1 \circ e; i_2) & : X + Y + Y + X \rightrightarrows X + X \end{align*}

is a kernel pair of (j1;j2):X+X→E(j_1; j_2) : X + X \to E. Informally, we can think of this as the subobject of (X+X)2≅∑i,j=12X2(X + X)^2 \cong \sum_{i,j=1}^2 X^2 which is the diagonal of XX in the first and fourth quadrants, and the diagonal of YY in the second and third quadrants.

To see this, note that since j1j_1 is a (split) monomorphism, we have a pullback diagram

X→idXid↓↓j1X→j1E. \begin{CD} X @> \id >> X \\ @V \id VV @VV j_1 V \\ X @>> j_1 > E. \end{CD}

Similarly, since ee is formed as a coreflexive equalizer, we have a pullback diagram

Y→eXe↓↓j1X→j2E. \begin{CD} Y @> e >> X \\ @V e VV @VV j_1 V \\ X @>> j_2 > E. \end{CD}

To see this, suppose we have two generalized elements x1,x2∈X(T)x_1, x_2 \in X(T) such that j1(x1)=j2(x2)j_1(x_1) = j_2(x_2). Then letting r:E→Xr : E \to X be the coreflexivity morphism, we see x1=r(j1(x1))=r(j2(x2))=x2x_1 = r(j_1(x_1)) = r(j_2(x_2)) = x_2 and j1(x1)=j2(x2)=j2(x1),j_1(x_1) = j_2(x_2) = j_2(x_1), so x1=x2x_1 = x_2 factors uniquely through ee. (See also here; in particular, that result, along with the uniqueness of equalizers and pullbacks, would give an alternate proof with an additional hypothesis that the pullback exists.)

Combining these using Lemma 2, we get a pullback diagram

X+Y→(id;e)Xid+e↓↓j1X+X→(j1;j2)E. \begin{CD} X+Y @> (\id; e) >> X \\ @V \id + e VV @VV j_1 V \\ X+X @>> (j_1; j_2) > E. \end{CD}

Similarly, we get a pullback diagram

Y+X→(e;id)Xe+id↓↓j2X+X→(j1;j2)E. \begin{CD} Y+X @> (e; \id) >> X \\ @V e + \id VV @VV j_2 V \\ X+X @>> (j_1; j_2) > E. \end{CD}

Now, combining those two using Lemma 2 (transposed), we get a pullback diagram

(X+Y)+(Y+X)→(id;e)+(e;id)X+X(id+e;e+id)↓↓(j1;j2)X+X→(j1;j2)E. \begin{CD} (X+Y)+(Y+X) @> (\id; e) + (e; \id) >> X+X \\ @V (\id+e; e+\id) VV @VV (j_1; j_2) V \\ X+X @>> (j_1; j_2) > E. \end{CD}

However, under the canonical isomorphism (X+Y)+(Y+X)≅X+Y+Y+X(X+Y)+(Y+X) \cong X+Y+Y+X, we see that (id;e)+(e;id)(\id; e) + (e; \id) corresponds to (i1;i1∘e;i2∘e;i2):X+Y+Y+X→X+X,(i_1; i_1\circ e; i_2\circ e; i_2) : X+Y+Y+X \to X+X, whereas (id+e;e+id)(\id+e; e+\id) corresponds to (i1;i2∘e;i1∘e;i2):X+Y+Y+X→X+X.(i_1; i_2\circ e; i_1\circ e; i_2) : X+Y+Y+X \to X+X. This completes the proof that these two morphisms form a kernel pair of (j1;j2):X+X→E(j_1; j_2) : X + X \to E.

Therefore, since (j1;j2):X+X→E(j_1; j_2) : X + X \to E is a regular epimorphism, by this result, it is a coequalizer of this kernel pair. However, this coequalizer is exactly the congruence quotient in the construction of X+YXX +_Y X here. Thus, we see that the original corelation is equivalent to the cokernel pair of ee. □\square

Corollary 5.

In an extensive and regular category with quotients of congruences, every cocongruence is effective.

Proof. This follows by combining Corollary 3 and Lemma 4, using the existence of quotients of congruences so that E+XEE +_X E always exists by this result. □\square

Context

This page is referenced by the following category implications.