Suppose that in the commutative diagram (4), the sequences (u,v) and (u′,v′) are exact. Then: (i) We have v1∘u1=0; if u′ is injective, the sequence (u1,v1) is exact. (ii) We have v2∘u2=0; if v is surjective, the sequence (u2,v2) is exact. (iii) Suppose that u′ is injective and v is surjective. There exists then one and only one homomorphism d:Ker (h)→Coker (f) having the following property: if z∈Ker (h),y∈N and x′∈M′ satisfy the relations v(y)=k(z) and u′(x′)=g(y), we have d(z)=p(x′). Moreover the sequence (*) Ker (f)u1Ker (g)v1Ker (h)dCoker (f)u2Coker (g)v2Coker (h) is exact.
Let us prove (i). Since u1 and v1 have the same graphs as the restrictions of u and v to Ker(f) and Ker(g) respectively, we have v1∘u1=0. We have
Ker(v1)=Ker(g)∩Ker(v)=Ker(g)∩Im(u)=Im(j)∩Im(u).
But by prop. 1 (i), we have Ker(v1)=Im(j∘u1)=Im(u1) if u′ is injective.
Let us prove (ii). Since u2 and v2 are obtained from u and v by passing to quotients, it is clear that v2∘u2=0. Suppose that v is surjective; since q and p are surjective, we have, by virtue of the hypotheses and prop. 1 (ii)
Let us finally prove (iii). For z∈Ker(h), there exists y∈N such that v(y)=k(z) since v is surjective; moreover, we have v′(g(y))=h(k(z))=0, and consequently there exists a unique x′∈M′ such that u′(x′)=g(y) since u′ is injective. Let us show that the element p(x′)∈Coker(f) is independent of the element y∈N such that v(y)=k(z). Indeed, if y1∈N is a second element such that v(y1)=k(z), we have y1=y+u(x) where x∈M; let us show that if x1′∈M′ is such that u′(x1′)=g(y1), we have x1′=x′+f(x); indeed, we have u′(x′+f(x))=u′(x′)+u′(f(x))=g(y)+g(u(x))=g(y+u(x))=g(y1). Finally, we conclude that p(x1′)=p(x′)+p(f(x))=p(x′). We can therefore set d(z)=p(x′) and we have thus defined a mapping d:Ker(h)→Coker(f).
If now z1,z2 are elements of Ker(h), if λ1,λ2∈A and z=λ1z1+λ2z2, we take elements y1 and y2 of N such that v(y1)=k(z1) and v(y2)=k(z2) and we choose for y∈N the element λ1y1+λ2y2; it is then immediate that
d(z)=λ1d(z1)+λ2d(z2),
so d is a homomorphism.
Suppose that z=v1(t) for some t∈Ker(g); we then take for y∈N the element j(t). Since g(j(t))=0, we conclude that d(z)=0, hence d∘v1=0. Conversely, suppose that d(z)=0. With the preceding notations, we thus have x′=f(x), where x ∈ M. In this case, we have g(y)=u′(f(x))=g(u(x)), or equivalently g(y−u(x))=0. The element y−u(x) is therefore of the form j(n) for n∈Ker(g), and we have
k(z)=v(y)=v(u(x)+j(n))=v(j(n))=k(v1(n));
as k is injective, z=v1(n), which proves that the sequence (*) is exact at Ker(h).
Conversely, suppose that an element w=p(x′) of Coker(f) is such that
u2(w)=u2(p(x′))=0(with x′∈M′).
We thus have q(u′(x′))=0, and consequently u′(x′)=g(y) for a y∈N; since v′(u′(x′))=0, we have v′(g(y))=0, hence h(v(y))=0, in other words v(y)=k(z) for a z∈Ker(h), and by definition w=d(z), which shows that the sequence (*) is exact at Coker(f). We have seen in (i) that it is exact at Ker(g) and in (ii) that it is exact at Coker(g), which completes the proving of (iii).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.