ProofComplete

Proof of the Additivity Theorem

We present Waldhausen's proof of the additivity theorem, which establishes that K-theory takes cofibration sequences to sums. The proof is a masterpiece of simplicial technique combined with the yoga of Waldhausen categories.


Reformulation

Theorem7.5Additivity (equivalent form)

Let C\mathcal{C} be a Waldhausen category. For every exact functor F:AE(C)F: \mathcal{A} \to E(\mathcal{C}) (to the category of cofibration sequences), the induced maps on K-theory satisfy:

K(sF)+K(qF)=K(tF):K(A)K(C)K(s \circ F) + K(q \circ F) = K(t \circ F): K(\mathcal{A}) \to K(\mathcal{C})

where ss: source, qq: quotient, tt: total (middle term) of the cofibration sequence. That is, [B]=[A]+[C][B] = [A] + [C] whenever ABCA \rightarrowtail B \twoheadrightarrow C.


Proof

Proof

Step 1: Reduction to a comparison of simplicial spaces.

It suffices to show that the map of simplicial categories

wSE(C)(s,q)wSC×wSCwS_\bullet E(\mathcal{C}) \xrightarrow{(s,q)} wS_\bullet \mathcal{C} \times wS_\bullet \mathcal{C}

induces a homotopy equivalence on geometric realizations. We use the simplicial category Ar(wSC)\operatorname{Ar}(wS_\bullet \mathcal{C}) (the arrow category, i.e., morphisms in wSCwS_\bullet \mathcal{C}) as an intermediary.

Step 2: The arrow category. Define FnCF_n \mathcal{C} as the category whose objects are sequences of cofibrations in SnCS_n \mathcal{C}:

Objects of SnE(C)S_n E(\mathcal{C}) are "filtrations of cofibration sequences": {Ai,jBi,jCi,j}0ijn\{A_{i,j} \rightarrowtail B_{i,j} \twoheadrightarrow C_{i,j}\}_{0 \leq i \leq j \leq n} where (A0,B0,C0,)(A_{0,\bullet} \rightarrowtail B_{0,\bullet} \twoheadrightarrow C_{0,\bullet}) is the primary data.

This factors through two intermediate categories:

wSnE(C)αwFnCβwSnC×wSnCwS_n E(\mathcal{C}) \xrightarrow{\alpha} wF_n \mathcal{C} \xrightarrow{\beta} wS_n \mathcal{C} \times wS_n \mathcal{C}

where FnCF_n \mathcal{C} has objects consisting of chains A0,B0,A_{0,\bullet} \rightarrowtail B_{0,\bullet} (just the cofibration, without the explicit quotients).

Step 3: The functor α\alpha is a homotopy equivalence. The functor α\alpha forgets the explicit quotient C0,=B0,/A0,C_{0,\bullet} = B_{0,\bullet}/A_{0,\bullet}. But in a Waldhausen category, pushout along a cofibration exists and is unique up to unique isomorphism. So the "quotient data" is determined by the cofibration ABA \rightarrowtail B, and α\alpha has a (homotopy) inverse that constructs the quotient. More precisely, the nerve of the fiber of α\alpha over any object of FnCF_n \mathcal{C} is contractible.

Step 4: Analyzing β\beta. The functor β:FnCSnC×SnC\beta: F_n \mathcal{C} \to S_n \mathcal{C} \times S_n \mathcal{C} sends (A0,B0,)(A0,,B0,/A0,)(A_{0,\bullet} \rightarrowtail B_{0,\bullet}) \mapsto (A_{0,\bullet}, B_{0,\bullet}/A_{0,\bullet}).

We need to show wFCwSC×wSC|wF_\bullet \mathcal{C}| \simeq |wS_\bullet \mathcal{C}| \times |wS_\bullet \mathcal{C}|.

Step 5: The "path space" trick. Define PnCP_n \mathcal{C} to be the category with objects being filtrations:

=X0X1Xn* = X_0 \rightarrowtail X_1 \rightarrowtail \cdots \rightarrowtail X_n

(same as SnCS_n \mathcal{C}, but we think of it as the "total" filtration). And define F~nC\tilde{F}_n \mathcal{C} by:

Objects: pairs of filtrations A,BSnCA_\bullet, B_\bullet \in S_n \mathcal{C} with compatible cofibrations AiBiA_i \rightarrowtail B_i for all ii.

The projection F~nCSnC\tilde{F}_n \mathcal{C} \to S_n \mathcal{C} via AA_\bullet has fibers that are equivalent to SnCS_n \mathcal{C}: given the "base" filtration AA_\bullet, the "total" BB_\bullet is determined by its quotient C=B/AC_\bullet = B_\bullet/A_\bullet, which is an arbitrary object of SnCS_n \mathcal{C}.

Step 6: Simplicial identities. By the realization lemma for bisimplicial sets, it suffices to show for each nn that:

wFnCβnwSnC×wSnC|wF_n \mathcal{C}| \xrightarrow{\beta_n} |wS_n \mathcal{C}| \times |wS_n \mathcal{C}|

is a homotopy equivalence. This follows from the fact that FnF_n is the "twisted product" of two copies of SnS_n, and the twist (the cofibration relation) is contractible.

Formally: the functor βn\beta_n is a Quillen fibration (satisfies Theorem B conditions) with contractible fibers. The fiber over (A,C)(A_\bullet, C_\bullet) is the category of cofibration sequences AiBiCiA_i \rightarrowtail B_i \twoheadrightarrow C_i compatible with the filtrations. By the extension axiom of Waldhausen categories, such extensions exist and form a contractible category (any two extensions are connected by a zig-zag of weak equivalences).

Step 7: Conclusion. Combining Steps 3--6:

wSE(C)αwFCβwSC×wSC|wS_\bullet E(\mathcal{C})| \xrightarrow{\alpha} |wF_\bullet \mathcal{C}| \xrightarrow{\beta} |wS_\bullet \mathcal{C}| \times |wS_\bullet \mathcal{C}|

Both α\alpha and β\beta are homotopy equivalences, so (s,q)=βα(s,q) = \beta \circ \alpha is a homotopy equivalence:

K(E(C))=ΩwSE(C)Ω(wSC×wSC)=K(C)×K(C).K(E(\mathcal{C})) = \Omega|wS_\bullet E(\mathcal{C})| \simeq \Omega(|wS_\bullet \mathcal{C}| \times |wS_\bullet \mathcal{C}|) = K(\mathcal{C}) \times K(\mathcal{C}).

\square


Remarks on the proof

RemarkThe role of the gluing axiom

The gluing axiom in the definition of a Waldhausen category is used crucially in Step 6: the fact that pushouts along cofibrations preserve weak equivalences ensures that the fiber categories in the fibration β\beta are well-behaved.

Without the gluing axiom, the proof fails: there exist "bad" Waldhausen categories where additivity does not hold. The gluing axiom is the minimal condition ensuring that K-theory is well-defined and additive.

ExampleNon-example: when additivity fails

Consider the category C\mathcal{C} of finite sets with injections as cofibrations and bijections as weak equivalences. This is a Waldhausen category (after verification of the gluing axiom). Here K0(C)=ZK_0(\mathcal{C}) = \mathbb{Z} (generated by the one-element set).

Now consider instead: weak equivalences = maps inducing bijections on connected components (a coarser notion). The gluing axiom may fail: a pushout of a "bijection on components" along a cofibration may not be a bijection on components. In this case, the S-construction does not produce a well-behaved K-theory, and "additivity" in the naive sense (A+C=B|A| + |C| = |B|) may fail.

This illustrates that the Waldhausen axioms are precisely calibrated to make the additivity theorem work.