Keller sequences of exact \infty-categories

We first show that if n < -\infty then \cal{C}_{[m,n]} is closed under retracts in \cal{C}_{[m,n+1]}. For this, let x \to y \to x be a retract diagram in \cal{C}_{[m,n+1]} with y \in \cal{C}_{[m,n]} and set z = \fib[y \to x] =\cofib[x \to y] \in \cal{C}. Now the exact sequence x \to y \to z implies that z \in \cal{C}_{\geq m} while the exact sequence \Omega x \to z \to y implies that z \in \cal{C}_{\leq n}. We conclude that z \in \cal{C}_{[m,n]}. The exact sequence x \to y \to z now implies that x \in \cal{C}_{[m,n]}, as desired.

Replacing (\cal{C},\cal{C}_{\geq 0},\cal{C}_{\leq 0}) with the opposite heart \infty-category (\cal{C}^{\op},(\cal{C}_{\leq 0})^{\op},(\cal{C}_{\geq 0})^{\op}) we conclude that if -\infty < m then \cal{C}_{[m,n]} is closed under retracts in \cal{C}_{[m-1,n]}. By induction we conclude that for any -\infty < m'\leq m \leq n \leq n' < \infty the inclusion \cal{C}_{[m,n]} \subseteq \cal{C}_{[m',n']} is closed under retracts. Since the heart structure is assumed bounded it now follows that \cal{C}_{[m,n]} is closed under retracts in \cal{C} for every -\infty \leq m \leq n \leq \infty.

Given a small \infty-category I and a diagram

let us write \cal{D} := \lim_IU(\cal{C}_\bullet) for its limit in \cat_\infty, where U \colon\exact \to\cat_\infty is the forgetful functor. It follows from [HTT, Proposition 5.1.2.2] that a diagram in \cal{D} admits a (co)limit as soon as its image under each of the structure maps p_i\colon\cal{D} \to U(\cal{C}_i) admits a (co)limit which is furthermore preserved under each of the transition functors in U(\cal{C}_i) \to U(\cal{C}_j) in the diagram, in which case the corresponding (co)limit in \cal{D} is computed pointwise. Since all the U(\cal{C}_i) are additive and all the transition functors U(\cal{C}_i) \to U(\cal{C}_j) are direct sum preserving it now follows that \cal{D} is additive and each of the structure maps p_i is direct sum preserving. By a similar argument, if we declare that a map in \cal{D} is an inclusion/projection if and only if its image under all the p_i is an inclusion/projection then the resulting classes of maps form an exact structure on \cal{D}. A functor f\colon\cal{E} \to \cal{D} from an exact category \cal{E} is then exact with respect to this structure precisely if each of the composites p_i\circ f\colon\cal{E} \to \cal{C}_i is so, and so this exact structure exhibits \cal{D} as the limit of this diagram in \exact.

From the universal property of \Psh_\lex^\mrm{st}(\C^\ex), we obtain a commutative square

where U is a colimit-preserving functor. Moreover, the exact inclusion functor j \colon \C^\ex \to \Psh_\lex^\mrm{st}(\C) induces a colimit-preserving functor

Since JUi^{\mrm{st}}_\C \simeq Ji^{\mrm{st}}_{\C^\ex}u \simeq ju \simeq i^{\mrm{st}}_\C, the composite JU is the identity functor.

We claim that UJ is also the identity. To see this, note that i^{\mrm{st}}_{\C^\ex} \colon \C^\ex \to \Psh_\lex^\mrm{st}(\C^\ex) exhibits \C^\ex as an extension-closed subcategory. Consequently, its preimage U^{-1}(\C^\ex) is an extension-closed subcategory of \Psh_\lex^\mrm{st}(\C). Since \C is contained in U^{-1}(\C^\ex), it follows that \C^\ex \subseteq U^{-1}(\C^\ex), and it follows that UJ restricts to an exact endofunctor f \colon \C^\ex \to \C^\ex. From UJi^{\mrm{st}}_{\C^\ex} \simeq i^{\mrm{st}}_{\C^\ex}f we conclude that

Since j is fully faithful, it follows that f, and thus also UJ, is the identity functor. In particular, J and U are inverse equivalences. Now from the commutative diagram

we conclude that that the middle vertical arrow is an equivalence. Since \E \to \Psh_\lex^\mrm{st}(\E) exhibits \E as an extension-closed subcategory, the left square is a pullback. Consequently, u^* is an equivalence.

Let I be a filtered \infty-category and \cal{C}_\bullet\colon I \to \exact a diagram. We wish to show that the colimit \cal{D} of \{\cal{C}_i\}_{i \in I} in \wald is exact. For this, note that filtered colimits in \wald are computed as follows: the underlying \infty-category of \cal{D} is the colimit \colim_i U(\cal{C}_i) \in \cat of the underlying \infty-categories U(\cal{C}_i). In particular, for each i the resulting functor U(\cal{C}_i) \to \cal{D} preserves all finite (co)limits which exist in U(\cal{C}_i) and which are preserved by the transition functors U(\cal{C}_i) \to U(\cal{C}_j) for all [i \to j] \in I_{i/} (where we note that I_{i/} is filtered and I_{i/} \to I is cofinal, so that we can assume if we wish that the colimit is indexed by I_{i/}). In addition, every finite diagram in \cal{D} factors through one of the U(\cal{C}_i) \to \cal{D} by compactness. Consequently, since each U(\cal{C}_i) is additive and each transition functor U(\cal{C}_i) \to U(\cal{C}_j) in the diagram is direct sum preserving we have that \cal{D} is additive. The colimit Waldhausen structure on \cal{D} is then such that a map is an inclusion if and only if it is the image of an inclusion in \cal{C}_i for some i \in I. Then every span x \leftarrow y \mono z in \cal{D} with right leg inclusion comes from a span with right leg inclusion in some \cal{C}_i, and so we see that this is indeed a Waldhausen structure and that the universal property of colimit (in \wald) holds.

Since we already know that \cal{D} is additive, to show that \cal{D} is exact it will suffice to show that there is some exact structure on \cal{D} with this collection of inclusions. Indeed, define a map to be a projection if it is the image of a projection in \cal{C}_i for some i \in I. Since any of the following types of diagrams

in \cal{D} comes a diagram of the same type in one of the \cal{C}_i we see that this is indeed an exact structure, as desired.

Let c' \epi a' be a projection in \cal{C}' with a'\in \cal{A}'. By assumption there exists a projection a \to a' in \cal{A}' with a \in \cal{A}. For the pullback square

in \cal{C'}. By assumption there exists a projection c \epi d with c \in \cal{C}. The composite c \epi d \epi a is then a projection in \cal{C}' among objects in \cal{C} and hence a projection in \cal{C} by assumption. Since \cal{A} \hookrightarrow \cal{C} is left-special there exists a projection b \epi a in \cal{A} whose image in \cal{C} factors through c. The composite b \epi a \epi a' is then a projection in \cal{A}' whose image in \cal{C'} factors through c', and so the bottom horizontal arrow is left-special.

? \Leftrightarrow ?: Since \cal{C} is closed under extensions in \st(\cal{C}) the data of a map \Sigma^{-1}i(a) \to x in \st(\cal{C}) with a \in \cal{A},c \in \cal{C} is equivalent to that of an exact sequence x \mono y \epi i(a) in \cal{C} with a \in \cal{A}, which in turn is equivalent to the data of a projection y \epi i(a) in \cal{C} with a \in \cal{A}. Then in the diagram

the existence of the dashed and dotted extensions with b,b'\in \cal{A} is equivalent, as one can pass between them by taking b = \fib[b'\epi a] and b'=\cofib[\Sigma^{-1}a \to b].

For right-special inclusions apply Corollary 3.18 and use the fact that fully faithful functors induces fully faithful functors on Spanier-Whitehead stabilizations. For left-special inclusions pass to opposites.

We prove the first claim, the second then follows by passing to opposite exact \infty-categories. Choose a heart decomposition

with A' \in \st(\cal{A})_{[0,n-1]} and a \in \cal{A}. Since the inclusion \cal{C} \subseteq \pst(\cal{C}) is left-special by Example 3.5, Lemma 3.8 inductively yields a factorization of the composite c \to \Sigma^ni(a) as

for some c_1,...,c_{n-1} \in \cal{C}. Now since \cal{A} \subseteq \cal{C} is right-special, the last of these maps moreover factors by (the dual of) Lemma 3.8? as \Sigma^{n-1}c' \to \Sigma^{n-1}i(b) \to \Sigma^ni(a) for some b\in\cal{A} and map \Sigma^{n-1}b \to \Sigma a in \pst(\cal{A}), so that \cofib(\Sigma^{n-1}b \to \Sigma^na) \simeq \Sigma^na' for some a' \in \cal{A}. The pullback B := \Sigma^{n-1}b \times_{\Sigma^na}A then participates in an exact sequence A' \to B \to \Sigma^{n-1}b and hence lies in \st(\cal{A})_{[0, n-1]}, and moreover there is an equivalence \cofib(B \to A) \simeq \cofib(\Sigma^{n-1}b \to \Sigma^na) \in \Sigma^n\cal{A} \subseteq \st(\cal{A})_{[1,n]}. By the universal property of pullbacks the map \phi\colon c \to i(A) factors as c \to i(B) \to i(A).

Choose n\ge 0 such that A lies in \st(\cal{A})_{[0,n]}\subseteq\st(\cal{C})_{[0,n]}, which exists since the heart structure on \st(\cal{A}) is bounded (and the functor \st(\cal{A})\subseteq\st(\cal{C}) is heart exact). For every m \in \{0,...,n\} consider the factorization \infty-category

We claim that the inclusion

is cofinal for every m=0,...,n-1, and in particular a weak homotopy equivalence. Since (\st(\cal{A})_{[0,n]})_{c//A} admits a terminal object this will finish the proof. Next we note that the comma \infty-categories associated to these inclusions are of the form (\st(\cal{A})_{[0,m]})_{c//A'} for some A'\in \cal{A}_{[0,m+1]}. Replacing m with n it will hence suffice to show that (\st(\cal{A})_{[0,n-1]})_{c//A} is weakly contractible.

Let (\st(\cal{A})_{[0,n-1]})^{\circ}_{c//A} \subseteq (\st(\cal{A})_{[0,n-1]})_{c//A} be the full subcategory spanned by those factorizations c \to i(B) \to i(A) such that \cofib[B \to A] \in \st(\cal{A})_{[1,n]}. Then (\st(\cal{A})_{[0,n-1]})^{\circ}_{c//A} is non-empty by Lemma 3.11. We now claim that (\st(\cal{A})^{\circ}_{[0,n-1]})_{c//A} admits binary cartesian products, and is hence co-sifted, and in particular weakly contractible. Indeed, since the functor \Map_{\pst(\cal{C})}(c,i(-)) is product preserving it will suffice by [HTT, Lemma 6.3.3.6] to show that the full subcategory (\pst(\cal{A})^{\circ}_{[0,n-1]})_{/A} \subseteq (\st(\cal{A})^{\circ}_{[0,n]})_{/A} spanned by those B \to A whose cofibre lies in \st(\cal{A})_{[1,n]} has binary products. Indeed, this is a full subcategory of the over category \st(\cal{A})_{/A} spanned by the maps B \to A such that B and \fib[B \to A] belong to \st(\cal{A})_{[0,n]}; this over category then has finite products and the full subcategory in question is closed under binary products since \st(\cal{A})_{[0,n-1]} is closed under extensions in \st(\cal{A}).

It is left to show that (\st(\cal{A})_{[0,n-1]})^{\circ}_{c//A} \subseteq (\st(\cal{A})_{[0,n-1]})_{c//A} is cofinal. Indeed, inspecting the associated comma \infty-categories we see that they are of the same form (\st(\cal{A})_{[0,n-1]})^{\circ}_{A'//A} (defined this time with respect to the identity \cal{A} \to \cal{A} in place of i) with A' \in \st(\cal{A})_{[0,n-1]}. These are then again weakly contractible by the above argument, and so the proof is complete.

Let A := \cofib(f). Then c = i(A) = \cofib(i(f)) \in \pst(\cal{C}) lies in \cal{C}. By Lemma 3.12 the identity on c factors through i(a) for some a \in \cal{A}. Since i is retract closed we get that \cofib(i(f)) \in \cal{A}, and hence f is an inclusion in \cal{A}. The dual statement follows by replacing i with i^{\op}.

Since \pst(\cal{A})_{/C} \to \pst(\cal{A}) is a right fibration and \cal{A}_{/C} \to \cal{A} is its base change along \cal{A} \to \pst(\cal{A}) the comma \infty-categories of the inclusion \cal{A}_{/C} \subseteq \pst(\cal{A})_{/C} are the same as those of \cal{A} \subseteq \pst(\cal{A}). By the pointwise formula for Kan extensions, we then need to show that for every map A \to C with A \in \pst(\cal{A}), the natural map

is an equivalence. By universality of colimits in spaces, this is equivalent to the statement that for each \phi\colon c \to i(A) in \pst(\cal{C})_{/C} the functor

has a contractible colimit. By [HTT, Corollary 3.3.4.6], it suffices to show that the unstraightening of this functor is weakly contractible. Indeed, this is exactly what is shown in Lemma 3.12.

For (?), we claim that for every A \in \pst(\cal{A}) the induced functor on comma \infty-categories \cal{A}_{/A} \to \C_{/A} (which appear in the pointwise formula for the relevant left Kan extensions) is cofinal, and for (?) we claim that for every c \in \C the induced functor on comma categories \cal{A}_{c/} \to \pst(\cal{A})_{c/} is coinitial; both statements are equivalent to the statement that for any map c \to A in \pst(\C) from an object of \cal{C} to an object of \pst(\cal{A}) the \infty-category of factorizations \cal{A}_{c//A} is weakly contractible, which we have established in Lemma 3.12.

Applying Corollary 3.14 with c = i(a) for some a \in \cal{A} one concludes that the induced map

is an equivalence for every A \in \pst(\cal{A}). Since every object in \pst(\cal{A}) is generated under finite colimits by objects in \cal{A} it follows that the map

is an equivalence for every B \in \pst(\cal{A}).

Consider some retract diagram X \to A \to X in \st(\cal{C}) with A \in \st(\cal{A}). By shifting, we may suppose A \in \st(\cal{A})_{[0, n]} for some n \ge 0. We prove that X \in \st(\cal{A})_{[0,n]} as well by induction on n. For n=0, this follows from the fact that \cal{A} is closed under retracts in \cal{C}, and \cal{C} under retracts in \st(\cal{C}) by Corollary 2.2. For n>0, write Y for the complementary summand to X in A, which exists by weak idempotent completeness of \st(\cal{C}), and consider a heart decomposition

for a \in \cal{A} and B \in \st(\cal{A})_{[1,n]}. The map h factors as a \to a \oplus a \to X \oplus Y = A, where the first map is the diagonal and the second the direct sum of the composites f\colon a \to X and g\colon a \to Y of h with the canonical projections. The associated exact sequence of cofibres is then

so that \cofib(f)\oplus\cofib(g) is an extension of B and \Sigma a, and hence lies in \st(\cal{A})_{[1,n]}. By the induction hypothesis \cofib(f) \in \st(\cal{A})_{[1,n]}. The exact sequence a \to X \to \cofib(f) then show that X belongs to \st(\cal{A})_{[0, n]}.

Suppose given some X \in \st(\cal{A}) \cap \st(\cal{C})_{\le 0}. Take a heart decomposition

in \st(\cal{A}) with X' \in \st(\cal{A})_{\le -1} and X'' \in \st(\cal{A})_{\ge 0}. Then the image of X'' in \st(\cal{C}) lies in \st(\cal{A})_{\ge 0} \cap \st(\cal{C})_{\le 0} = \cal{C} \cap \st(\cal{A})_{\ge 0}. In particular, the identity \id_{X''} is a map from an object of \cal{C} to an object of \st(\cal{A})_{\ge 0}, and so factors through an object of \cal{A} by Lemma 3.12. But \cal{A} is closed under retracts in \cal{C}, so X'' \in \cal{A}, and X is an extension of objects in \st(\cal{A})_{\le 0} and itself in \st(\cal{A})_{\le 0}.

Apply [Cis19, Theorem 7.2.16 and 7.2.8] (see also [nuiten16, Corollary 3.14]) using that the collection of inclusions with cofibre in \cal{A} is closed under composition (since \cal{A} is closed under extensions) and cobase change.

Suppose given a map f\colon x \to y in \cal{B}^l which is the image of an inclusion f'\colon x'\mono y' in \cal{C} and let \tilde x \in \cal{C} be a lift of x. Then the equivalence p(\tilde x) \simeq p(x') can be represented by a zig zag x' \rightarrow w \hookleftarrow \tilde x with the second arrow an inclusion with cofibre in \cal{A} and the first arrow mapping to an equivalence in \cal{B}^l. We may then consider the cobase change

of f'. The square on the right is then also cocartesian in \cal{B}^l by assumption, and so its right vertical arrow is also an equivalence in \cal{B}^l. The composite inclusion \tilde x \mono w \mono z then provides the desired lift of f.

We prove the right-special case, the left-special case can be deduced by passing to opposite exact \infty-categories. We view the pre-stable envelope \pst(\cal{C}) as an exact \infty-category in which all maps are inclusions. Given y \in \cal{C}, we claim that for every x \in \cal{C}, the functor

is left Kan extended from its restriction to y \rtimes \cal{A}. Indeed, viewing \Sigma y as an object of \st(\cal{C}), note that the functor [y \mono y'] \mapsto \cofib[y \mono y'] \to \Sigma y induces an equivalence

and similarly for y\rtimes \pst(\cal{A}). In addition, under these equivalences the functor [y \mono y'] \mapsto \Map_{\pst(\cal{C})}(x,y') corresponds to the functor [z \to \Sigma y] \mapsto \Map_{\cal{C}_{/\Sigma y}}(x,z), where x is viewed as the object of \cal{C}_{/\Sigma y} corresponds to the zero map x \xrightarrow{0} \Sigma y. The left Kan extension claim is then an instance of Corollary 3.14. Now since \pst(\cal{C}) admits finite colimits and the inclusion \pst(\cal{A}) \to \pst(\cal{C}) preserves finite colimits it follows that y\rtimes \pst(\cal{A}) admits finite colimits, and is hence filtered. We conclude that the presheaf

is a filtered colimit of presheaves of the form \Map_{\pst(\cal{C})}(-,y'). Now since the inclusion \cal{C} \to \pst(\cal{C}) is an exact functor every cocartesian square in \cal{C} with parallel legs inclusions is also cocartesian in \pst(\cal{C}), and hence each of the presheaves \Map_{\pst(\cal{C})}(-,y') send such a square to a pullback square. We hence conclude that \Map_{\cal{C}'}(-,p(y)) sends such squares to pullback squares. It follows that p^l sends cocartesian square in \cal{C} with parallel legs inclusions to cocartesian squares in \cal{B}^{l}, as desired.

The implication ? \Rightarrow ? follows from Proposition 3.29. We now show that ? \Rightarrow ?. For this we first show that if p is a left-special projection with kernel i then i is a left-special inclusion (and hence, passing to opposite exact \infty-categories, that if p is a right-special projection with kernel i then i is a right-special inclusion). Given a projection x \epi a in \cal{C} with a \in \cal{A}, write f:y \mono x for its fibre. Since p is a left-special projection, there exists a b \in \cal{A} and factorization

where q is a projection with fibre in \cal{A} and the left diagonal map is the corresponding summand inclusion. We claim that the composite b \to y\oplus b \epi x \epi a is a projection in \cal{A}. Indeed, there is a map of exact sequences

where the upper row is split. Since the right square induces an equivalence on horizontal fibres it is cartesian in \st(\cal{C}), and so also induces an equivalence on vertical fibres in \st(\cal{C}). Since q is a projection these vertical fibres are in \cal{C} and b \to a is a projection in \cal{C}. Furthermore, since the fibre of q is assumed in \cal{A} we have that the fibre of b \to a is in \cal{A}, so that b \to a is a also a projection in \cal{A}.

Now since we assume that \cal{A} \to \cal{C} is the inclusion of the kernel of p it is closed under retracts, and hence a Keller inclusion. It is left to show that the sequence exhibits \cal{B} as the cofibre of \cal{A} \to \cal{C}. Let W denote the collection of maps sent to equivalences by p. Since the underlying functor of p is a localisation by W and the inclusions/projections in \cal{B} are exactly the images of the inclusions/projections in \cal{C} we have that for any exact \infty-category \cal{E} the induced functor

is fully faithful with essential image those exact functors \cal{C} \to \cal{E} which send the arrows in W to equivalences. Now since the arrow from 0 to (or from) any object in \cal{A} is in W we have that such exact functors vanish on \cal{A}. On the other hand, either of the axioms 3.24, 3.24 implies that any exact functor \cal{C} \to \cal{E} which vanishes on \cal{A} inverts all the arrows in W. Comparing universal properties we conclude that p exhibits \cal{B} as the cofibre of i\colon \cal{A} \to \cal{C}, as desired.

Note first that since the subcategory W is closed under direct sum the localisation \cal{B} is additive and the map \cal{C} \to \cal{B} is direct sum preserving. Let now f\colon x \to y be a map in W, so that the cofibre A = \cofib[f] \in \st(\cal{C}) lies in \st(\cal{A}), and hence in \st(\cal{A})_{[0,1]} by Corollary 3.20. We claim that there exists a b \in \cal{A} such that f factors as a composite x \mono b \oplus y \epi y with the second map being the projection to y. To see this, note that by Lemma 3.11 the map y \to A factors as y \to a \to A with a \in \cal{A} and \cofib[a \to A] = \Sigma b for some b \in \cal{A}. Let z be the fibre product in the right square in

where the map x \to z is induced by the universal property pullback from the map f\colon x \to y and the zero map 0\colon x \to a. Then \fib[z \to y] \simeq b \in \cal{A} and so z belongs to \cal{C} and the map z \to y is a projection in \cal{C}. At the same time, the cofibre of x \to z in \st(\cal{C}) is a, and so x \to z is an inclusion in \cal{C}. Now since the map y \to A factors through a by construction the map z \to y admits a section and hence splits as the summand projection y \oplus b \to y. Applying this argument to the opposite exact \infty-categories we similarly get that any f\colon x \to y in W factors as a composite x \mono x \oplus a \epi y, where the first map is the associated summand inclusion.

We conclude that the projection \cal{C} \to \cal{B} also exhibits \cal{B} as the localisation of \cal{C} by \cal{C}^{\cal{A}}_{\inc}, and also as the localisation of \cal{C} by \cal{C}^{\cal{A}}_{\pr}. In particular, by Lemma 3.21 this localisation has both a left calculus of fractions with respect to \cal{C}^{\cal{A}}_{\inc} and a right calculus of fractions with respect to \cal{C}^{\cal{A}}_{\pr}. Furthermore, by Lemma 3.23 the map p\colon \cal{C} \to \cal{B} preserves cocartesian squares with parallel legs inclusions and cartesian squares with parallel legs projections. Using that the axioms of exact structures are satisfied in \cal{C}, to show that under the proposed collection of inclusions and projections the \infty-category \cal{B} and the functor p are exact it will hence suffice to show that any of the following types of diagrams

in \cal{B} lifts to a diagram of the same type in \cal{C}. All of these follow from Lemmas 3.21 and 3.22. Indeed, since we assume that \cal{A} \hookrightarrow \cal{C} is both left- and right-special these lemmas together tell us that we can lift any map in \cal{B} to a map in \cal{C} given a prescribed lift of either its domain or its target, that we can lift any inclusion in \cal{B} to an inclusion in \cal{C} given a prescribed lift of its domain, and that we can lift any projection in \cal{B} to a projection in \cal{C} given a prescribed lift of its target. To obtain the first two diagram we first lift z to \tilde z, then lift the map x \to z to map \tilde x \to \tilde z using the right calculus of fractions (which, in the second case we can choose to be a projection), and then lift the inclusion x \mono y to an inclusion \tilde x \to \tilde y. For the latter two diagrams, we first lift z to \tilde z, then lift the map z \to w to a map \tilde z \to \tilde w (which in the second case we can choose to be an inclusion), and then lift z \epi w to a projection \tilde z \epi \tilde w. We have thus established that \cal{B} is an exact \infty-category with respect to the proposed structure and that p\colon \cal{C} \to \cal{B} is exact.

We now claim that the resulting null-composite sequence

is both cartesian and cocartesian. For this, note first that by the argument above, the cocartesian squares with parallel legs inclusions in \cal{B} are the exactly the images of the same types of squares in \cal{C}, and similarly for cartesian squares with parallel legs projections. The universal property of localisations then implies that for any exact \infty-category \cal{E}, the induced functor

is fully faithful with essential image those exact functors \cal{C} \to \cal{E} which send the arrows in W to equivalences. Clearly any such exact functor vanishes on \cal{A} (since the maps from 0 to/from any object in \cal{A} is in W), but on the other hand since we have shown in the beginning of the proof that any map in W is a composite of an inclusion with cofibre in \cal{A} and a projection with fibre in \cal{A} we see that any exact functor that vanishes on \cal{A} also inverts all arrows in W. Comparing universal properties we hence conclude that the sequence (?) exhibits \cal{B} as the cofibre of \cal{A} \hookrightarrow \cal{C}.

Now since forming stable envelops is a colimit preserving operation we have \st(\cal{B}) = \st(\cal{C})/\st(\cal{A}). At the same time, by Corollary 3.19 the induced functor \st(\cal{A}) \to \st(\cal{C}) is fully faithful and closed under retracts, and so the map \st(\cal{A}) \to \fib[\st(\cal{C}) \to \st(\cal{B})] is an equivalence by [9-authors-II, Corollary A.1.10]. Since \cal{C} embeds in \st(\cal{C}) and \cal{A} = \cal{C} \cap \st(\cal{A}) by Corollary 3.20 we now conclude that the sequence (?) is a fibre sequence in \cat, and eventually in \exact by Corollary 3.13.

It is left to show that p is a Keller projection. Note that since the fibre of p was shown to coincide with \cal{A}, the factorization properties we proved for W in the beginning of the proof imply that p satisfies 3.24 and 3.24 as soon as we show that W is saturated, that is, that every map f\colon x \to y such that p(f) is an equivalence is in W. For this, note that if p(f) is an equivalence then \cofib[f] \in \st(\cal{C}) lies in the fibre of \st(\cal{C}) \to \st(\cal{B}). But we have already saw above that this fibre is \st(\cal{A}), and so the proof is complete.

Given a Keller sequence (i,p)\colon\cal{A} \to \cal{C} \to \cal{B}, the induced map \st{p}\colon\st(\cal{C}) \to \st(\cal{B}) is a Verdier projection by Proposition 3.27 since \st(-) is left adjoint, with kernel the retract-closure of \st(\cal{A}) in \st(\cal{C}), which by Corollary 3.19 coincides with \st(\cal{A}).

The map on the left is fully faithful by Theorem 3.10 and closed under retracts by Corollaries 3.20 and 3.19. Now if m=-\infty and n=\infty then this is Corollary 3.30 (see Examples 3.2 and 3.26). If n=\infty and m \neq \infty then we may as well assume that m=0. Then \st(\cal{A})_{[0,\infty]} = \pst(\cal{A}) \to \pst(\cal{C}) = \st(\cal{C})_{[0,\infty]} is left-special by Lemma 3.6 and is right-special since all maps are inclusions on both sides. By Proposition 3.27 we then have that the map \pst(\cal{C}) \to \pst(\cal{C})/\pst(\cal{A}) to the cofibre in \exact is a Keller projection. Since all maps in \pst(\cal{C}) are inclusions it then follows that all the maps in \pst(\cal{C})/\pst(\cal{A}) are inclusions, so that this cofibre is also the cofibre in the full subcategory \exact^{\pst} \subseteq \exact spanned by the exact \infty-categories in which all maps are inclusions, which is just the \infty-category of pre-stable \infty-categories and finite colimit preserving functors. Since \pst(-) constitutes a left adjoint to the inclusion \exact^{\pst} \subseteq \exact we conclude that the functor \pst(\cal{C})/\pst(\cal{A}) \to \pst(\cal{C}/\cal{A}) = \pst(\cal{B}) is an equivalence, and so

is a Keller sequence. The case where m=-\infty and n\neq \infty follows in a dual manner.

Finally, consider the case where m \neq -\infty and n \neq \infty. Without loss of generality we may assume that m=0. Applying both Lemma 3.6 and its dual we see that \st(\cal{A})_{[0,n]} \to \st(\cal{C})_{[0,n]} is a Keller inclusion. To finish the proof we need to show that the induced exact functor

is an equivalence. Now since the inclusion \cal{C} \to \st(\cal{C})_{[0,n]} induces an equivalence on stable envelopes by [Sau23, Theorem 2.9] (and its proof), the inclusion \st(\cal{C})_{[0,n]} \subseteq \st(\cal{C}) exhibits its target as the stable envelope of its source, and since taking stable envelopes preserves cofibres we conclude that (?) induces an equivalence on stable envelopes. The exact functor (?) is hence an extension-closed inclusion by [Kle22, Proposition 4.25]. It will hence suffice to show that \st(\cal{C})_{[0,n]} \to \st(\cal{B})_{[0,n]} is essentially surjective. We now argue by induction on n. For n=0 this is clear. Now suppose that n is such that the claim is true for n-1. Then \st(\cal{C})_{[0,n-1]} \to \st(\cal{B})_{[0,n-1]} is a Keller projection and in particular every map f\colon x \to y in \st(\cal{B})_{[0,n-1]} is the image of a map in \st(\cal{C})_{[0,n-1]} in the sense that it lifts to a zig-zag \tilde{x} \to w \leftarrowtail \tilde{y} with right-hand map an inclusion with cofibre in \st(\cal{A})_{[0,n-1]} (so the image of the left-hand map in \st(\cal{B})_{[0,n-1]} is equivalent to f). Since any object in \st(\cal{B})_{[0,n]} is a cofibre of a map in \st(\cal{B})_{[0,n-1]} it now follows that \st(\cal{C})_{[0,n]} \to \st(\cal{B})_{[0,n]} is essentially surjective, as desired.

We prove the claim for left-special projections. The case of right-special projections follows by taking opposite exact \infty-categories. Let f\colon x \to y be a map such that p(f) is a projection and let C = \cofib[f] \in \st(\cal{C}), so that C lies in \st(\cal{C})_{[0,1]}. Since p(f) is a projection we have p(C) \in \Sigma\cal{B} belongs to \cal{B}. It follows that the map 0 \to p(C) is a projection in \st(\cal{B})_{[0,1]}. Since \st(\cal{C})_{[0,1]} \to \st(\cal{B})_{[0,1]} is a Keller projection with kernel \st(\cal{A})_{[0,1]} by Corollary 3.31 it follows from Remark 3.25 that there exists a projection A \epi C in \st(\cal{C})_{[0,1]} with A \in \st(\cal{A})_{[0,1]}. Choose a heart decomposition b \to A \to \Sigma b' with b,b' \in \cal{A}. Then b \to A is also a projection in \st(\cal{A})_{[0,1]} and so the composite b \epi A \epi C is a projection in \st(\cal{C})_{[0,1]}. Let z be the fibre product in the right square in

Then the map z \epi y is a projection in \st(\cal{C})_{[0,1]} with fibre b' \in \cal{A} \subseteq \cal{C} and hence z belongs to \cal{C} and z \epi y is a projection in \cal{C}. The cofibre of x \to z is b by construction and so this map is in particular an inclusion in \cal{C} which is sent to an equivalence in \cal{B}. By Property (3.24) there exists an a\in \cal{A} and a factorization of x \to z as x \mono x \oplus a \epi z with the first map being the summand inclusion. The triangle

hence gives the desired factorization of f.

The statements (?) and (?) are equivalent, see Remark 3.16. We now prove (?).

Set \cal{A} = \ker(p). Recall from Remark 3.25 that p admits a right calculus of fractions with respect to \cal{C}_{\pr}^{\cal{A}}, so that for every x,y \in \cal{C} the induced map

is an equivalence of spaces, where \cal{A} \rtimes x \subseteq \cal{C}_{/x} is the full subcategory spanned by the projections x'\epi x with fibre in \cal{A}. This means that for every map f\colon p(x) \to p(y) we have \colim_{x' \epi x \in \cal{A} \rtimes x}\Map_f(x',y) \simeq \ast, where \Map_f(x',y) is the fibre of \Map(x',y) \to \Map(p(x),p(y)) over f. Viewing the last colimit as the comma category of the composed functor

over (x,f), we conclude that F is coinitial, and so for every \phi\colon \cal{C}^{\op} \to \E we have

Noting that \pst(\cal{C}) \to \pst(\cal{B}) is again a Keller projection with kernel \pst(\cal{A}) by Corollary 3.31, to prove the claim it will now suffice to show that for every x \in \cal{C} the induced functor

is coinitial. Shifting exact sequences inside the stable envelop of \cal{C} we may identify this functor with the inclusion

whose comma category over a given g\colon \Sigma^{-1} x \to A with A \in \pst(\cal{A}) is the category of factorizations \cal{A}_{\Sigma^{-1}x//A} of g through an object of \cal{A}. This category of factorizations is indeed weakly contractible, as can be obtained by applying Lemma 3.12 to the composed right-special inclusion \cal{A} \to \cal{C} \to \st(\cal{C})_{[-1,0]} (see Example 3.5).

We need to show that if

is a pointwise exact sequence in \fun(\cal{I},\cal{E}) with \vphi Reedy cofibrant then the following holds:

1. If \phi is Reedy cofibrant then \psi is Reedy cofibrant.

2. If \psi is Reedy cofibrant then \phi is Reedy cofibrant if and only if \psi \Rightarrow \vphi is a Reedy cofibration.

For i \in \cal{I} consider the commutative diagram

in \st(\cal{E}) where the last row is obtained by taking vertical cofibres. In particular, all three rows are exact, and since we assume that \vphi is Reedy cofibrant we have B \in \cal{E}. Now if \phi is also Reedy cofibrant then C is also in \cal{E}, in which case A \in \st(\cal{E})_{\leq 0}. But A also lies in \st(\cal{E})_{\geq 0} (since it is by definition the cofibre of a map in \st(\cal{E})_{\geq 0}) and so A \in \cal{E}. This shows the first claim. For the second claim, assume that \psi is Reedy cofibrant, so that A is in \cal{E}. The second claim can then be restated as saying that A \to B is an inclusion in \cal{E} if and only if C is in \cal{E}, which holds since \cal{E} \hookrightarrow \st(\cal{E}) preserves and detects exact sequences.

We prove by induction on the size of \cal{I}. Clearly the claim holds if \cal{I} is empty. Now suppose that \cal{I} is non-empty and that the claim has been proven for all posets of strictly smaller size. Since \cal{I} is finite and non-empty there exists an i \in \cal{I} which is maximal in the sense that there does not exist a j \in \cal{I} strictly bigger than i. Then the complement \cal{I} \setminus \{i\} is downward closed and \cal{I} is the union of its two downward closed full subposets \cal{I} \setminus \{i\} and \cal{I}_{\leq i} = \{j \in \cal{I}\;|\; j \leq i\}, whose intersection is \cal{I}_{< i} = \{j \in \cal{I}\;|\; j < i\}. Passing to nerves we then obtain the nerve of \cal{I} is the pushout of the nerve of \cal{I}_{j \neq i} and the nerve of \cal{I}_{\leq i} along the nerve of \cal{I}_{< i}, which implies that the colimit of \vphi over \cal{I} fits in a pushout square

Now \cal{I}\setminus \{i\} and \cal{I}_{< i} are finite posets of strictly smaller size than \cal{I}. By the induction hypothesis the terms in the left column are hence in \cal{E}. At the same time, since \vphi is a Reedy cofibration the cofibre of the top horizontal map is in \cal{E}, and so the same holds for the cofibre of the bottom horizontal map. We conclude that \colim_{j}\vphi(j) \in \cal{E}, as desired. For the last claim, note that \colim is induced by an exact functor among stable \infty-categories

and so once we know that it sends the extension-closed full subcategory \fun^{\reedy}(\cal{I},\cal{E}) on the left to the extension-closed full subcategory \cal{E} on the right the induced functor

is automatically exact with respect to the induced exact structures on both sides.

It is clear that \fun(\cal{I},\st(\cal{E}))_{\geq 0} is closed under finite colimits, \fun(\cal{I},\st(\cal{E}))_{\leq 0} under finite limits , and both are closed under extensions, simply because \st(\cal{E})_{\geq 0} and \st(\cal{E})_{\leq 0} have the analogous properties (where we note that finite limits commute with finite colimits in \st(\cal{E}) by stability). It is also clear from the definitions that \fun(\cal{I},\st(\cal{E}))_{\geq 0} \cap \fun(\cal{I},\st(\cal{E}))_{\leq 0} coincides with Reedy cofibrant diagram to \st(\cal{E})^{\heartsuit}, and hence to \cal{E} when \cal{E} is weakly idempotent complete.

It is left to show that there is a sufficient supply of heart decompositions. Let \vphi\colon \cal{I} \to \cal{E} be some diagram. We wish to construct an exact sequence \psi^- \Rightarrow \vphi \Rightarrow \psi^+ with \psi^- \in \fun(\cal{I},\st(\cal{E}))_{\leq 0} and \psi^+ \in \fun(\cal{I},\st(\cal{E}))_{\geq 1}. We construct \psi^- and \pst^+ by induction on the poset \cal{I}. Let i \in \cal{I}. Suppose that i \in \cal{I} is such that we have already constructed an exact

with \psi^- \in \fun(\cal{I}_{< i},\st(\cal{E}))_{\leq 0} and \psi^+ \in \fun(\cal{I}_{< i},\st(\cal{E}))_{\geq 1}, where \cal{I}_{<i} = \{j \in \cal{I} \;|\; j < i\} (note that there is always at least one element in \cal{I} for which \cal{I}_{<i} = \emptyset, so the induction can always start somewhere). Note that this implies that \colim_{j < i}\psi^-(j) \in \st(\cal{E})_{\leq 0} and \colim_{j < i}\psi^+(j) \in \st(\cal{E})_{\geq 1}. In particular we have maps

such that the horizontal row is a heart decomposition for \colim_{j < i}\vphi(j). Now consider the pushout X := \vphi(i) \displaystyle\mathop{\coprod}_{\colim_{j < i}\vphi(j)} \colim_{j < i}\psi^+(j) and let

be a heart decomposition with X^- \in \st(\cal{E})_{\leq 0} and X^+ \in \st(\cal{E})_{\geq 1}, and set Y^-:=\fib[\vphi(i) \to X^+]. Consider the resulting diagram

with exact rows. By construction, the total cofibre of the right square is \cofib[X \to X^+] = \Sigma X^-. It follows that the cofibre of the left vertical map is equivalent to X^-, and is hence in \st(\cal{E})_{\leq 0}. The right hand side then encodes an extension of \psi^+ to \fun(\cal{I}_{\leq i},\st(\cal{E}))_{\geq 1} with \psi^+(i) = X^+, and the left hand side encodes an extension of \psi^- to \fun(\cal{I}_{\leq i},\st(\cal{E}))_{\leq 0}. We have thus extended our desired heart decomposition from \cal{I}_{< i} to \cal{I}_{\leq i}. Since \cal{I} is a finite poset this process terminates after finitely many steps, yielding the desired heart decomposition on all of \cal{I}.

We first show that the inclusion \fun^{\reedy}(\cal{I},\cal{A}) \to \fun^{\reedy}(\cal{I},\cal{C)} is left-special. Let \vphi\colon \cal{I} \to \cal{C} be a Reedy cofibrant diagram and \vphi\Rightarrow \psi a pointwise projection with \psi a Reedy cofibrant diagram taking values in \cal{A}. Since \cal{A} \to \cal{C} is left-special there exists, for every i \in \cal{I}, an object b_i equipped with a map b_i \to \vphi(i) such that the composite b_i \to \vphi(i) \to \psi(i) is an inclusion in \cal{A}. Define \phi\colon \cal{I} \to \cal{A} by the formula

Note that this formula indeed defines a functor, since the association i \mapsto S_i := \{j \in \cal{I} \;|\; j < i\} is a functor from \cal{P} to sets. Let us now view this functor to \infty-groupoids which happens to take values in discrete ones. We then observe that for every j \in\cal{I}_{< i} the poset \cal{J}_j := \{k \in \cal{I}_{< i} \;|\; j \in S_k\} contains j as an initial object, and is hence weakly contractible. We conclude that for every i \in \cal{I} the maps S_j \to S_i \setminus \{i\} for j \in \cal{I}_{< i} exhibit S_i \setminus \{i\} as the colimit of the functor \cal{I}_{< i} \to \Sps sending j to S_j. Consequently, for every i \in \cal{I} the induced map \colim_{j < i}\phi(j) \to \phi(i) is equivalent to the map \oplus_{j < i} b_i \to \oplus_{j \leq i} b_i, which is an inclusion with cofibre b_i. We conclude that \phi is Reedy cofibrant. The composite \phi \Rightarrow \vphi \Rightarrow \psi is then a pointwise projection by construction and hence a projection by Lemma 3.35.

We now show that the inclusion \fun^{\reedy}(\cal{I},\cal{A}) \to \fun^{\reedy}(\cal{I},\cal{C)} is right-special. Let \vphi\colon \cal{I} \to \cal{C} be a Reedy cofibrant diagram and \psi\Rightarrow \vphi a Reedy cofibration with \psi a Reedy cofibrant diagram taking values in \cal{A}. We construct a diagram \phi\colon \cal{I} \to \cal{A} and a map \vphi\Rightarrow \phi such that the composite \psi \Rightarrow \vphi \Rightarrow \phi is a Reedy cofibration. As in the proof of Lemma 3.37 we construct \phi by induction on the poset \cal{I}. In particular, given an i \in \cal{I} such that we have already constructed \phi\colon \cal{I}_{<i} \to \cal{A} with the desired property with respect to \vphi|_{I_{<i}}, we show how to extend \phi from \cal{I}_{<i} to \cal{I}_{\leq i}. Consider the diagram

where the map a \to c is obtained by forming pushouts in the left and right faces. Here, all terms on the left face are in \cal{A}, all terms on the right face are in \cal{C}, and all indicated maps are inclusions, as can be deduced from Lemma 3.36 and the assumptions on \vphi,\psi and \phi. In addition, since the front-top horizontal map is the identity the \cofib[a \to c] \in \st(\cal{C}) coincides with the total cofibre of the back face, which lies in \cal{C} since \psi \Rightarrow \vphi is a Reedy cofibration. We conclude that a \to c is an inclusion. Now since \cal{A} \to \cal{C} is right-special there exists a map c \to b in \cal{C} such that the composite a \to c \to b is an inclusion. The rightmost vertical map in the resulting diagram

then encodes an extension \tilde\phi of \phi from \cal{I}_{< i} to \cal{I}_{\leq i} with \phi(i) = b, and \tilde\phi is Reedy cofibrant since the rightmost vertical map is an inclusion. In addition, the total cofibre of the external rectangle is equivalent to \cofib[a \hookrightarrow b], which lies in \cal{A}, and so the resulting composite \psi|_{\cal{I}_{\leq i}} \Rightarrow \vphi|_{\cal{I}_{\leq i}} \Rightarrow \tilde\phi is a Reedy cofibration.

We have thus established that \fun^{\reedy}(\cal{I},\cal{A}) \to \fun^{\reedy}(\cal{I},\cal{C)} is left- and right-special. It is also closed under retracts, since \cal{A} is closed under retracts in \cal{C}, and hence a Keller inclusion. By Proposition 3.27 it will now suffice to show that

is a cofibre sequence. Now by Corollary 3.38 the induced sequence on stable envelopes is given by

which is a Verdier sequence by Corollary 3.30 and [9-authors-I, Proposition 6.5.10], and in particular a cofibre sequence. Since forming stable envelopes is colimit preserving it follows that the induced exact functor from \cofib[\fun^{\reedy}(\cal{I},\cal{A}) \to \fun^{\reedy}(\cal{I},\cal{C)}] to \fun^{\reedy}(\cal{I},\cal{B}) is an equivalence on stable envelopes, and hence in particular fully faithful. To finish the proof it will hence suffice to show that

is essentially surjective. Given a Reedy cofibrant diagram in \vphi\colon \cal{I} \to \cal{B} we may construct a \phi\colon \cal{I} \to \cal{C} such that p(\phi) \simeq \vphi by induction on \cal{I} as above. In particular, given a \phi constructed on \cal{I}_{<i} for some i \in \cal{I}, extending \phi to \cal{I}_{\leq i} requires solving the following lifting problem: given an object c \in \cal{C} and an inclusion j\colon p(c) \mono b in \cal{B}, we need to find an inclusion \tilde{j}\colon c \mono \tilde{b} such that p(\tilde{j}) = j. Indeed, this lifting property holds for Keller projections, see Remark 3.25.

For (?) this is an instance of Example 3.3. As for (?) let us just give the argument for p, the one for q being completely analogous. Suppose given a map f\colon x \to y in \cal{C} such that p(f) is an equivalence in \cal{B}. Consider the commutative diagram

in which the rows are exact sequences. Then the left square is a pushout, so that the induced map jq(y) \oplus x \to y is a projection with kernel jq(x). Setting a := jq(y) we then obtain a factorization

as a summand inclusion with complement in \ker(p) followed by a projection, so that p is a left-special projection as desired.

We now prove (?). Since \ker(q)=\cal{B} and \ker(p)=\cal{A} we have that the sequences in question are fibre sequences. We now show that the sequence

is also cofibre sequences in \exact; the argument for the second sequence is completely analogous. Let \cal{E} \in \exact be a test object. Then the adjunction p\dashv i induces via pre-composition an adjunction

on the level of exact functor categories, whose counit is again an equivalence. In particular, p^* is fully-faithful, and an arbitrary exact functor \vphi\colon \cal{C} \to \cal{E} is in the image of p^* if and only if the unit is an equivalence on \vphi, that is, if and only if the map \vphi(x) \to \vphi(ip(x)) is an equivalence for every x. Now since the unit fits in an exact sequence

we have that such an exact \vphi inverts the map x \epi ip(x) if and only if it vanishes on jq(x). We conclude that an exact \vphi\colon \cal{C} \to \cal{E} is in the image of p^* if and only if it vanishes on jq(x) for every x, which is equivalent to saying that \vphi vanishes on the essential image of j. We conclude that p exhibits \cal{B} as the cofibre of j\colon \cal{A} \to \cal{C}, as desired.

The fact that arrows with these properties are (co)cartesian follows immediately from the mapping space criterion [HTT, 2.4.4.3]. It is also clear that there is a sufficient supply of lifts of this form since one can always form the pullback/pushout along the unit/counit since these are projections and inclusions, respectively, by assumption (where we have also used that p and q have fully-faithful adjoints).

Apply Proposition 4.12 to the induced functor

using the backward-forward unique factorization system on \Span^{\dagger}(\cal{B}) of [two-variable-fibrations, Proposition 4.9], where we note that the first assumption is satisfied by Lemma 4.10 and the second assumption is satisfied since the exact functor q\colon \cal{C} \to \cal{A} induces a retraction \Span^{\dagger}(q)\colon \Span^{\dagger}(\cal{C}) \to \Span^{\dagger}(\cal{A}) to the inclusion of the fibre of \Span^{\dagger}(p) over 0 \in \Span^{\dagger}(\cal{B}).

The dual variant for q can be obtained from the statement for p by replacing (\cal{C};\cal{A},\cal{B}) with (\cal{C}^{\op};\cal{B}^{\op},\cal{A}^{\op}) and using Remark 4.1. We now prove the statement for p.

We apply [two-variable-fibrations, Theorem 3.1] with f being an equivalence (a case for which the two assumptions of that theorem clearly hold), once to the map of adequate triples (\cal{C},\cal{C}^{\pr},\cal{C}^{\inc}) \to (\cal{B},\cal{B}^{\pr},\cal{B}^{\inc}) and once to the map of adequate triples (\cal{C},\cal{C}^{\inc},\cal{C}^{\pr}) \to (\cal{B},\cal{B}^{\inc},\cal{B}^{\pr}) (noting that switching the role of ingressive and egressive arrows replaces the associated span \infty-category with its opposite). To obtain the desired result it will now suffice to show that any p-cartesian projection in \C is also p^{\pr}-cartesian and any p-cartesian inclusion in \C is also p^{\inc}-cartesian, where p^{\pr} and p^{\inc} are the functors induces on the wide subcategories of projections and inclusions, respectively. Concretely, we need to show that in any commutative triangle

in \C such that g is p-cartesian, if g,h and p(f) are projections then so is f, and if g,h and p(f) are inclusion then so is f. Indeed, this follows from Remark 4.2 since q\colon \cal{C} \to \cal{A} sends all p-cartesian arrows to equivalences (as can be seen by their explicit description appearing in Lemma 4.6).

Set \cal{E}_{/\cal{D}} := \cal{E} \times_{\cal{D}^{\Delta^{\{0\}}}} \cal{D}^{\Delta^1}, so that the composite

is a cocartesian fibration whose fibre over x \in \cal{D} is the comma category \cal{E}_{/x}. Note that the inclusion \cal{E} \to \cal{E}_{/\cal{D}} sending a to (a,\id\colon p(a)\to p(a)) is co-initial and in particular a weak homotopy equivalence, and so it will suffice to show that the functor

is a weak homotopy equivalence, where r'\colon \cal{E}_{/\cal{D}} \to \cal{F} is given by r'(a,p(a) \to x) = r(a). Viewing (r',q) as a map among cocartesian fibrations over \cal{D} (where on the right hand side the map to \cal{D} is the projection on the second factor) it preserves cocartesian arrows over \cal{D} (since the q-cocartesian arrows (a,p(a) \to x) \to (a',p(a') \to x') are exactly those for which the map a \to a' is an equivalence). To show that (r',q) is a weak homotopy equivalence it will hence suffice to show that the induced map

on fibres over x \in \cal{D} is a weak homotopy equivalence for every x.

Let \cal{E}^r := \cal{E} \times_{\cal{D}} \cal{D}^r be the wide subcategory of \cal{E} spanned by those morphisms whose image in \cal{D} lies in \cal{D}^r, and let \cal{E}^r_{/x} be the comma category of \cal{E}^r \to \cal{D}^r over x. Then the induced functor

is fully-faithful: indeed, by uniqueness of factorizations arrows in \cal{D}^r have the right cancellation property in the sense that if f,g are two composable arrows in \cal{D} such that g and g\circ f are in \cal{D}^r, then f is in \cal{D}^r. Now, given (a, f\colon p(a) \to x) \in \cal{E}_{/x}, we may factor f as p(a) \stackrel{g}\epi y \stackrel{i}\hookrightarrow x, and choose a p-cocartesian lift a \to \tilde{y} of g.

We claim that the collection of maps

assemble into the unit of an adjunction \cal{E}_{/x} \adj \cal{E}^r_{/x}. Indeed, given (a', f'\colon p(a') \mono x) \in \cal{E}^r_{/x}, the map

is induced on pullbacks by the map of cospans

in which the bottom face is a composition of squares

We now observe that the left square is cartesian by virtue of the facts that a \to \tilde{y} is p-cocartesian and the right square is cartesian by applying Remark 4.11 to g\colon f(a) \epi y and f'\colon p(a')\mono x.

It then follows by pasting that the external rectangle is cartesian and hence the map (?) is an equivalence.

Let \cal{F} \to \cal{E}^r_{/x} be the fully-faithful functor sending b \in \cal{F} = p^{-1}(0) to the pair (b,0 \hookrightarrow x), where 0 \hookrightarrow x is the essentially unique arrow in \cal{D}^r from 0 to x. Given (a, f) \in \cal{E}^r_{/x}, write b \to a for a cartesian lift of the unique \cal{D}^r-map 0 \hookrightarrow p(a); then by an analogous argument, the collection of maps (b, 0 \hookrightarrow x) \to (a, p(a) \hookrightarrow x) assemble into the counit of an adjunction \cal{F} \adj \cal{E}^r_{/x}.

Since adjunctions induce inverse equivalences on geometric realizations the inclusions \cal{F} \subseteq \cal{E}^r_{/x} \subseteq \cal{E}_{/x} are both weak homotopy equivalences. Since r'_x\colon \cal{E}_{/x} \to \cal{F} restricts to the identity on \cal{F} by construction we conclude that r'_x is indeed a weak homotopy equivalence, as desired.

Given an extension-closed full inclusion \cal{D} \subseteq \cal{E} of exact \infty-categories, and [n] \in \bbDelta let us denote by P_n(\cal{E};\cal{D}) \subseteq \fun(\Delta^n,\cal{E}) the full subcategory spanned by those sequences e_0 \to ... \to e_n such that each e_i \to e_{i+1} is an inclusion with cofibre in \cal{D}. We note that P_n(\cal{E};\cal{D}) is closed under extensions in \fun(\Delta^n,\cal{E}) and hence inherits an exact structure. We also note that for every \rho\colon [n] \to [m] in \bbDelta the restriction functor \fun(\Delta^m,\cal{E}) \to \fun(\Delta^n,\cal{E}) sends P_m(\cal{E};\cal{D}) to P_n(\cal{E};\cal{D}), so that the various P_n(\cal{E};\cal{D}) assemble to give a simplicial object P_\bullet(\cal{E};\cal{D}). Given a Keller sequence

we have an induced fibre sequence

of simplicial objects in \exact. We note that at \bullet=0 this is just the original Keller sequence. Applying K-theory and taking geometric realizations we hence obtain a commutative diagram

in \Sp. We now prove the following three claims, which together imply that the top row is an exact sequence in \Sp:

1. The rightmost vertical map is an equivalence.

2. The left square is cocartesian.

3. The bottom row is a cofibre sequence.

The first statement is clear, since the simplicial object P_\bullet(\cal{B};0) is homotopy constant. For the second statement, note that the \infty-category \exact is semi-additive, and hence we can take the Bar construction B_\bullet(\cal{C};\cal{A}) of the exact functor i\colon \cal{A} \to \cal{C}, which by definition is the simplicial object obtained by left Kan extending the diagram 0 \leftarrow \cal{A} \xrightarrow{i} \cal{C} along the diagram [0] \xleftarrow{d_0} [1] \xrightarrow{d_0} [0] in \bbDelta^{\op}. Explicitly, B_n(\cal{C};\cal{A}) is given by \cal{C} \oplus \cal{A} \oplus ... \oplus \cal{A} with the summand \cal{A} appearing n times. We have a map of simplicial objects

which at the n’th level is the exact functor

We claim that the exact functor f_n induces an equivalence on K-theory for every n. Indeed, each f_n admits a retraction

though, to avoid confusion, let us point out that these retractions do not assemble to a map of simplicial objects. Nonetheless, g_n\circ f_n = \id, while the functor \K(P_n(\cal{C};\cal{A})) \to \K(P_n(\cal{C};\cal{A})) induced by f_n \circ g_n is homotopic to the identity by additivity of \K, as witnessed by the natural filtration

in P_n(\cal{C};\cal{A}) whose associated graded pieces are [c_0 = ... = c_0], [0 \to \cofib(i_1) = ... = \cofib(i_1)], ... [0 \to ... \to 0 \to \cofib(i_n)], the direct sum of which is f_ng_n(c_0 \to ... \to c_n), see Remark 4.8. It hence follows that f_n is indeed an equivalence on K-theory, as claimed. To prove (?) above it will hence suffice to show that the square

is cocartesian in \Sp_{\geq 0}. Now since \K-theory preserves direct sums it sends the Bar construction of \cal{A} \to \cal{C} to the Bar construction of \K(\cal{A}) \to \K(\cal{C}) in \Sp_{\geq 0} and the Bar construction of \cal{A} \to \cal{A} to the Bar construction of \K(\cal{A}) \to \K(\cal{A}). But Bar constructions are left Kan extensions, so their geometric realizations are just the cofibres of the arrows from which they are generated. In other words, the desired statement is equivalent to saying that the square

is cocartesian, which is clear. It is hence left to prove (?). Now the above argument also shows that |\K(P_\bullet(\cal{A};\cal{A}))| \simeq |\K(B_\bullet(\cal{A};\cal{A}))| \simeq 0, and so the desired claim is equivalent to the map

being an equivalence. For this, it will suffice to show that for each n the map

is an equivalence of spaces. We now observe that we can commute S_n and P_\bullet, that is, the above map can be rewritten as

This map, in turn, can be identified with the map induced on geometric realizations by the functor

It will hence suffice to show that this functor is a weak homotopy equivalence. Indeed, this follows from Proposition 4.15 applied to the induced Keller sequence on S_n established in Lemma 4.19.

We prove the claim for \cal{C}^{\cal{A}}_{\pr}, the claim for \cal{C}^{\cal{A}}_{\inc} can be inferred by replacing \cal{C} with its opposite (equipped with the opposite exact structure). Applying Proposition 4.16 it will suffice to show that \pst(\cal{C})^{\pst(\cal{A})}_{\pr} \to \pst(\cal{B})^{\simeq} is a weak homotopy equivalence, and applying Proposition 4.18 we may further replace the domain by \pst(\cal{C})^{\pst(p)}. Now the map

is a Keller projection (Corollary 3.31) and so in particular exhibits \pst(\cal{B}) as the localisation of \pst(\cal{C}) by the collection of maps belonging to \pst(\cal{C})^{\pst(p)}. At the same time, since \pst(\cal{C}) has finite colimits and these are preserved by \pst(p) we see that the collection of maps belonging to \pst(\cal{C})^{\pst(p)} is closed under cobase change. The \infty-category \pst(\cal{C}) is then an \infty-category with weak equivalences and cofibrations in the sense of [Cis19, Definition 7.4.12], where all maps are cofibrations and the weak equivalences are the maps belonging to \pst(\cal{C})^{\pst(p)}. Applying [Cis19, Corollary 7.6.9] we now conclude that the map

is a weak homotopy equivalence, as desired.

We show that the square

is cartesian for every n \geq 1, thus implying the result upon pasting the squares and passing to the colimit. Equivalently, since geometric realizations commute with base change along maps among \infty-groupoids, what we need to show is that the induced functor

is an equivalence on geometric realizations. We claim that this functor is coinitial, hence in particular an equivalence on geometric realizations. For this, let C be an object of \cal{P}, which we can view as an object of \st(\cal{C})_{[0,n]} whose image in \pst(\cal{B})_{[0,n]} lies in \pst(\cal{B})_{[0,n-1]}, and let \cal{I}_C be the comma category of the above functor over C. We need to show that \cal{I}_{C} is weakly contractible. We note that if D \epi C is a projection in \st(\cal{C})_{[0,n]} with domain in \st(\cal{C})_{[0,n-1]} then \fib(D \epi C) also lies in \st(\cal{C})_{[0,n-1]}. We hence describe \cal{I}_C as the category whose objects are projections D \epi C in \st(\cal{C})_{[0,n]} with domain in \st(\cal{C})_{[0,n-1]} and fibre in \st(\cal{A})_{[0,n-1]}, and whose morphisms are commutative triangles

of projections. We first show that \cal{I}_{C} is not empty. Since p(C) belong to \pst(\cal{B})_{[0,n-1]} we have that the map C \to 0 in \st(\cal{C})_{[0,n]} is sent to an inclusion in \pst(\cal{B})_{[0,n]}. Now \st(\cal{C})_{[0,n]} \to \pst(\cal{B})_{[0,n]} is a Keller projection by Corollary 3.31 and hence inclusions in \pst(\cal{B})_{[0,n]} can be lifted to inclusions in \st(\cal{C})_{[0,n]} with prescribed lift of their domain, see Remark 3.25. We conclude that there exists an inclusion C \mono A in \st(\cal{C})_{[0,n]} with A \in \st(\cal{A})_{[0,n]}. Choose a heart decomposition a \to A \to A' with A' \in \st(\cal{A})_{[1,n]} and a \in \cal{A}, so that A \to A' is in particular an inclusion in \st(\cal{A})_{[0,n]} (with cofibre \Sigma a \in \st(\cal{A})_{[1,1]} \subseteq \st(\cal{C})_{[0,n]}). Finally, consider the composite inclusion

Since A' \in \st(\cal{A})_{[1,n]} the cofibre of the inclusion C \mono A' must lie in \st(\cal{C})_{[1,n]} and so this inclusion is also a projection and its fibre D = \fib[C \to A'] lies in \st(\cal{C})_{[0,n-1]}. Since the cofibre of D \to C lies in \st(\cal{A})_{[1,n]} this map is also a projection with fibre in \st(\cal{A})_{[0,n-1]}), and hence determines an object of \cal{I}_C. We have thus shown that \cal{I}_C is non-empty. Now we note that \cal{I}_C almost has cartesian products. More precisely, we may embed \cal{I}_C as a wide subcategory of \cal{J}_C, where \cal{J}_C is the full subcategory of the comma category (\st(\cal{C})_{[0,n]})_{/C} spanned by the projections D \epi C with fibre in \st(\cal{A})_{[0,n-1]}, then \cal{J}_C has cartesian products, given by taking fibre product over \cal{C}. This cartesian product restricts to a functorial binary operation (-)\times_C(-)\colon\cal{I}_C \times \cal{I}_C \to \cal{I}_C, though this operation is not the cartesian product in \cal{I}_C. Nonetheless, the natural projections from (-)\times_C(-) to each of the two factors are pointwise in \cal{I}_C, and can hence be viewed as natural transformations of functors \cal{I}_C \times \cal{I}_C \to \cal{I}_C. Since \cal{I}_C is non empty, this is enough to deduce that \cal{I}_C is weakly contractible using Quillen’s classical trick: given an X \in \cal{I}_C the maps Y \leftarrow X \times Y \rightarrow X determine a zig-zag of natural transformations from the identity to a constant functor.

Let \tilde{\cal{E}} \subseteq \cal{E}^{\Delta^1} \times_{\cal{E}^{\Delta^{\{1\}}}} \cal{E}^{\circ} be the full subcategory spanned by the arrows x \to y in \cal{E}^{\circ}. In other words, the objects of \tilde{\cal{E}} are the arrows in \cal{E}^{\circ} and the morphisms from [x \to y] to [x' \to y'] are the commutative squares

such that y \to y' is in \cal{E}^{\circ}. We may then factor the map \cal{E}^{\circ} \to \cal{E} as a composite

where j(y) = [\id_y\colon y \to y] and q(x \to y) = x. We claim that both j and q are weak homotopy equivalences. For j, note that it admits a retraction

sending [x \to y] to y. Furthermore, r is a cocartesian fibration with fibre over y given by the full subcategory of \cal{E}_{/y} given by the arrows in \cal{E}^{\circ} and cocartesian transition functors given by post-composition. Since the fibres of r have terminal objects we see that r is a weak homotopy equivalence, and hence j is a weak homotopy equivalence as well.

We now claim that q is a localisation functor, and so in particular a weak homotopy equivalence. To see this, we use Rezk’s sufficient criterion for localisation, namely, we show that for every [m] \in \Delta the map

is a weak homotopy equivalence, where on the right we have the core \infty-groupoid of the \infty-category of functors \Delta^m \to \cal{E} and on the left the wide subcategory of \fun(\Delta^m,\tilde{\cal{E}}) whose morphisms are the natural transformations sent to equivalences in \fun(\Delta^m,\cal{E}). In particular, the right hand side is an \infty-groupoid and hence it suffices to check that the fibres of q_m are weakly contractible. Now given an \alpha = (x_0 \to ... \to x_m)\in \fun(\Delta^m,\cal{E}), the fibre of q_m over \alpha has objects diagrams

where all vertical morphisms and all horizontal morphisms in the bottom row belong to \cal{E}^{\circ}, and morphisms natural transformations of such diagrams which are the identity on the top row and pointwise in \cal{E}^{\circ} on the bottom row. To finish the proof we now show that \cal{I}_{\alpha} is non-empty and admits binary coproducts; this implies that \cal{I} is sifted, and in particular weakly contractible. Concerning coproducts, we note that by Assumption ? we may identify \cal{I}_C with a full subcategory of the slice \infty-category \fun(\Delta^m,\cal{E})_{[x_0 \to ... \to x_m]/} spanned by diagrams as above. Now \fun(\Delta^m,\cal{E}_{[x_0 \to ... \to x_m]/}) admits coproducts given by pushouts under [x_0 \to ... \to x_m] and by Assumptions ? and ? we see that \cal{I}_{\alpha} is closed under binary coproducts in \fun(\Delta^m,\cal{E})_{[x_0 \to ... \to x_m]/} and hence inherits itself binary coproducts. To finish the proof we now show that \cal{I}_{\alpha} is not empty. For this, we construct the y_i’s by induction on i. For i=0 we set y_0=x_0. Now suppose that for 0 \leq i < m we have constructed a partial diagram of the form

such that the vertical arrows and the horizontal arrows in the bottom row are in \cal{E}^{\circ}. Let z_{i+1} := y_{i} \coprod_{x_i} x_{i+1}, so that the map x_{i+1} \to z_{i+1} is in \cal{E}^{\circ} by Assumption ?. Then by Assumption ? there exists a map z_{i+1} \to y_{i+1} in \cal{E}^{\circ} such that the composite y_i \to z_{i+1} \to y_{i+1} is also in \cal{E}^{\circ}. We can then use the \cal{E}^{\circ} maps y_i \to y_{i+1} and x_{i+1} \to y_{i+1} to extend the above diagram from i to i+1. Arriving at i=m we have thus constructed an object of \cal{I}_{\alpha}, showing that the latter is non-empty, as desired.

We verify that \iota satisfies the assumptions of Lemma 4.17. First note that \pst(\cal{C}) admits pushouts which are preserved by \pst(\cal{C}) \to \pst(\cal{B}), and \pst(\cal{B})^{\simeq} has pushouts (since any \infty-groupoid admits colimits indexed by weakly contractible categories), so that \pst(\cal{C})^{\pst(p)} admits pushouts. Next we note that that projections in \pst(\cal{C}) are closed under cobase change. Indeed, all maps in \pst(\cal{C}) are inclusions and projections are closed under cobase change along inclusions in any exact \infty-category. It follows that \iota satisfies Assumption (?) holds. In addition, if C \xrightarrow{f} D \xrightarrow{g} E is a composable sequence in \pst(\cal{C}) such that f and g\circ f are projections then g is a projection: indeed, the cofibres of f,g and g\circ f fit into an exact sequence

and a map in \pst(\cal{C}) is a projection if and only if its cofibre is in the full subcategory \pst(\cal{C})_{[1,\infty]} \subseteq \pst(\cal{C}) which is closed under cofibres. We conclude that \iota satisfies Assumption (?). Finally, suppose that f\colon C \to D is a map in \pst(\cal{C}) lying over an equivalence in \pst(\cal{B}). The map D \to \cofib[C \to D] to the cofibre is a projection in \pst(\cal{C}) with target in \pst(\cal{A}). Since \pst(\cal{A}) \to \pst(\cal{C}) is a Keller inclusion (see Corollary 3.31) there exists an A \in \pst(\cal{A}) and a map A \to D such that the composite A \to D \to \cofib[C \to D] is a projection in \pst(\cal{A}). Set E := \cofib[A \to D]. Then the map D \to E is a projection with kernel in \pst(\cal{A}), hence a morphism in \pst(\cal{C})^{\pst(\cal{A})}_{\pr}, and the composite C \to D \to E is also a projection with kernel in \pst(\cal{A}), as can be seen from the exact sequence

and the fact that A \to \cofib[C \to D] is a projection in \pst(\cal{A}). We conclude that \iota satisfies Assumption (?), and so the proof is complete.

This is a special case of Proposition 3.39, as we can identify \rS_q(-) \simeq \fun^{\reedy}(\Delta^q,-) with the induced exact structure.

Combine Corollary 4.23 with [SW25, Proposition 5.4].

Since \exact is compactly generated (Corollary 2.7) the inclusion of finitary functors \exact \to \Sp inside all functors admits a right adjoint, sending \cal{G}\colon \exact \to \Sp to the left Kan extension of its restriction to the full subcategory of compact objects. It will hence suffice to show that \K is the initial finitary Keller localising functor under \Sigma^{\infty}(-)^{\simeq}. But by Theorem 4.20 all finitary Keller localising functors \exact \to \Sp are right Kan extended from their restriction to \catex \subseteq \exact. In addition, any Keller localising functor on \exact restricts to a Verdier localising functor on \catex. The desired result hence follows from the universal property of algebraic K-theory as a functor on \catex, see [BGT13] or [HLS24, Theorem 5.1].

The inclusion i_0 is Keller by Example 3.4. By Proposition 3.27, it suffices to show that the functor \ev_2:\cal{E}_n \to \cal{C}_{[0,n]} coincides with the Dwyer-Kan localisation at the projections in \cal{E}_n with fibre in \cal{C}_{[0,n-1]}. Since \ev_2 certainly inverts maps in w := (\cal{E}_n)_\mathrm{pr}^{\cal{C}_{[0,n-1]}} and is essentially surjective since \cal{C}_{[0,n-1]} \subseteq \cal{C}_{[0,n]} is resolving, it remains to show that the induced functor \cal{E}_n[w^{-1}] \to \cal{C}_{[0,n]} is fully faithful. For X an object of \cal{E}_n, write w_{/X} \subseteq (\cal{E}_n)_{/X} for the full subcategory of the slice on those projections W \epi X with fibre in \cal{C}_{[0,n-1]}. Given X_i := (x_i \mono y_i \epi z_i) \in \cal{E}_n for i=0,1, consider the natural map

By universality of colimits, it suffices to show for each map f:z_0 \to z_1 that the functor

has contractible colimit, and by [HTT, Corollary 3.3.4.6], we may show that the unstraightening \mrm{Un}(\phi_f) is weakly contractible. But \mrm{Un}(\phi_f) is equivalent to the \infty-category of spans of the form

with \fib(y' \epi y_0) \in \cal{C}_{[0,n-1]}. Consider the pullback Q := \lim(y_1 \epi z_1 \leftarrow y_0) where the map on the right is the composite y_0 \to z_0 \xrightarrow{f} z_1. We note that Q is an extension of y_0 by x_1, and hence lies in \cal{C}_{[0,n-1]}, and moreover \fib(Q \epi z_0) is an extension of x_0 by x_1, and hence lies in \cal{C}_{[0,n-1]} as well. The object

is then easily seen to be terminal in \mrm{Un}(\phi_f), so that the latter is weakly contractible, as claimed.

Write \ev_0, \ev_1:\cal{E}_n \to \cal{C}_{[0,n-1]} for the evaluations (x \mono y \epi z) \mapsto x, y. Then the composite

is the identity on \cal{C}_{[0,n-1]}^{\times 2}, and the composite

sends (x \mono y \epi z) to (x \xto{0} y \xto{(\id,0)} y \oplus \Sigma x). Its action on K-theory is then homotopic to the identity by virtue of additivity and the diagram

and so we conclude that (\ev_0,\ev_1) induces an equivalence on K-theory. At the same time, the composite

is given by (x,y) \mapsto (x, x \oplus y), and hence induces an equivalence on K-theory by direct sum preservation of the latter. We conclude that i_0 \oplus i_1 is an equivalence on K-theory. Finally, consider the commutative diagram

where the top row is clearly a Keller sequence and the bottom row is a Keller sequence by Proposition 4.22. Since the left-most and middle vertical maps are equivalences on K-theory we conclude from Theorem 4.14 that the rightmost vertical map is an equivalence, as desired.