Documentation

TauCeti.CategoryTheory.Exact.Projective

Relative projectives in an exact category, and the horseshoe lemma #

Let E be a Quillen exact structure on an additive category C. An object Q is E-projective when every morphism out of Q lifts along every deflation of E. This is the relative notion Bühler uses: it depends on the exact structure, not just on the category. For the split exact structure every object is projective, and for the canonical exact structure of an abelian category it is Mathlib's CategoryTheory.Projective.

Two theorems make relative projectives usable for resolutions, and both are proved by the same method — pull a conflation back along a deflation, then split the resulting conflation because its third term is projective — although each builds its own pullback square. Schanuel's lemma says that two conflations K ↪ Q ↠ X and K' ↪ Q' ↠ X with projective middle terms have stably isomorphic kernels, K ⊞ Q' ≅ K' ⊞ Q.

The main theorem is the horseshoe lemma. Given a conflation X ↪ Y ↠ Z and first steps K_X ↪ Q_X ↠ X and K_Z ↪ Q_Z ↠ Z of resolutions of the two outer terms, with Q_Z projective, there is a conflation

K ↪ Q_X ⊞ Q_Z ↠ Y

whose middle term is the direct sum of the two resolving terms and whose kernel K is itself an extension K_X ↪ K ↠ K_Z of the two syzygies, compatibly with the two given conflations. Iterating it resolves the middle term of a conflation from resolutions of the outer terms, and that is the form in which the last result of this file states it: for an object property P consisting of projectives, containing a zero object and closed under binary biproducts, admitting a finite P-resolution is closed under extensions, with no increase in length.

That closure property is what makes the objects of finite projective dimension an extension-closed subcategory, so that TauCeti.ExactStructure.fullSubcategory equips them with an induced exact structure — the setting in which the resolution theorem compares their Grothendieck group with the one generated by the projectives themselves.

Main definitions #

Main results #

Implementation notes #

The horseshoe is proved without ever choosing a lift of Q_Z into Y by hand. Pulling the conflation X ↪ Y ↠ Z back along the deflation Q_Z ↠ Z gives a conflation X ↪ W ↠ Q_Z, which splits because Q_Z is projective, while pulling the conflation K_Z ↪ Q_Z ↠ Z back along Y ↠ Z exhibits the same object W as an extension of Y by K_Z. Composing the resulting deflation W ↠ Y with the direct sum Q_X ⊞ Q_Z ↠ X ⊞ Q_Z ≅ W and applying the dual Noether package TauCeti.ExactStructure.exists_conflation_comp' produces the conflation and identifies its kernel in one step.

References #

The objects that are projective relative to the exact structure E: those Q for which every morphism out of Q factors through every deflation of E.

This depends on E and not only on C. For the split exact structure every object is projective, by TauCeti.ExactStructure.split_isProjective; for the canonical exact structure of an abelian category the notion is Mathlib's CategoryTheory.Projective, by TauCeti.ExactStructure.abelian_isProjective_iff.

Equations
Instances For

    The defining lifting property, as the characterisation of TauCeti.ExactStructure.isProjective available to code that cannot unfold its body.

    The chosen lift of f : Q ⟶ Y along a deflation p : X ⟶ Y, for Q projective.

    Equations
    Instances For

      A binary direct sum of projectives is projective: lift the two components separately.

      A conflation with projective third term splits. The section of its deflation is the lift of the identity of that term.

      This is the exact-structure analogue of Mathlib's CategoryTheory.ShortComplex.ShortExact.splittingOfProjective, with the balancedness hypothesis on the ambient category replaced by the kernel–cokernel pair carried by a conflation.

      Equations
      Instances For
        @[simp]

        Every object is projective for the split exact structure. Its deflations are split epimorphisms, along which every morphism visibly lifts.

        @[simp]

        For the canonical exact structure of an abelian category, relative projectivity is Mathlib's CategoryTheory.Projective. The deflations there are exactly the epimorphisms.

        A left adjoint carries relative projectives to relative projectives when its right adjoint carries deflations to deflations.

        A projective presentation of X relative to E is a conflation K → P → X whose middle term is E-projective.

        • K : C

          The kernel term of the presentation.

        • P : C

          The relatively projective middle term.

        • i : self.K ⟶ self.P

          The inflation into the projective term.

        • p : self.P ⟶ X

          The deflation onto the presented object.

        • The two presentation maps form a short complex.

        • conflation : E.Conflation { X₁ := self.K, X₂ := self.P, X₃ := X, f := self.i, g := self.p, zero := ⋯ }

          The presentation is a conflation of E.

        • isProjective : E.isProjective self.P

          The middle term is projective relative to E.

        Instances For

          An exact structure has enough projectives if every object admits a relative projective presentation.

          Instances For

            The image of a relative projective presentation under a conflation-exact functor that carries relative projectives to relative projectives.

            Equations
            • P.map hF hPP' = { K := F.obj P.K, P := F.obj P.P, i := F.map P.i, p := F.map P.p, zero := ⋯, conflation := ⋯, isProjective := ⋯ }
            Instances For

              The tautological projective presentation in the split exact structure.

              Equations
              Instances For

                Mathlib's projective presentation gives a relative projective presentation for the canonical exact structure of an abelian category.

                Equations
                • One or more equations did not get rendered due to their size.
                Instances For

                  The lift of f : X ⟶ Y to the projective middle terms of relative projective presentations P of X and Q of Y, chosen by relative projectivity of P.P.

                  Equations
                  Instances For

                    The morphism induced by f : X ⟶ Y on the kernel terms of relative projective presentations, through the chosen lift TauCeti.ExactStructure.ProjectivePresentation.middleMap of the projective middle terms.

                    Equations
                    Instances For

                      Enough ordinary projectives give enough relative projectives for the canonical exact structure on an abelian category.

                      theorem TauCeti.ExactStructure.nonempty_iso_biprod_of_projective {C : Type u} [CategoryTheory.Category.{v, u} C] [CategoryTheory.Preadditive C] [CategoryTheory.Limits.HasZeroObject C] [CategoryTheory.Limits.HasBinaryBiproducts C] {E : ExactStructure C} {X K Q : C} {m : K ⟶ Q} {a : Q ⟶ X} {hm : CategoryTheory.CategoryStruct.comp m a = 0} (hc : E.Conflation { X₁ := K, X₂ := Q, X₃ := X, f := m, g := a, zero := hm }) (hQ : E.isProjective Q) {K' Q' : C} {m' : K' ⟶ Q'} {a' : Q' ⟶ X} {hm' : CategoryTheory.CategoryStruct.comp m' a' = 0} (hc' : E.Conflation { X₁ := K', X₂ := Q', X₃ := X, f := m', g := a', zero := hm' }) (hQ' : E.isProjective Q') :
                      Nonempty (K ⊞ Q' ≅ K' ⊞ Q)

                      Schanuel's lemma. Two conflations K ↪ Q ↠ X and K' ↪ Q' ↠ X with projective middle terms have stably isomorphic kernels: K ⊞ Q' ≅ K' ⊞ Q.

                      Both isomorphisms come from the pullback Q ×_X Q', which base change exhibits at once as an extension of Q' by K and as an extension of Q by K'; projectivity splits both.

                      theorem TauCeti.ExactStructure.exists_conflation_biprod_of_conflation_of_projective {C : Type u} [CategoryTheory.Category.{v, u} C] [CategoryTheory.Preadditive C] [CategoryTheory.Limits.HasZeroObject C] [CategoryTheory.Limits.HasBinaryBiproducts C] {E : ExactStructure C} {S : CategoryTheory.ShortComplex C} (hS : E.Conflation S) {KX QX : C} {mX : KX ⟶ QX} {aX : QX ⟶ S.X₁} {hmX : CategoryTheory.CategoryStruct.comp mX aX = 0} (hcX : E.Conflation { X₁ := KX, X₂ := QX, X₃ := S.X₁, f := mX, g := aX, zero := hmX }) {KZ QZ : C} {mZ : KZ ⟶ QZ} {aZ : QZ ⟶ S.X₃} {hmZ : CategoryTheory.CategoryStruct.comp mZ aZ = 0} (hcZ : E.Conflation { X₁ := KZ, X₂ := QZ, X₃ := S.X₃, f := mZ, g := aZ, zero := hmZ }) (hQZ : E.isProjective QZ) :

                      The horseshoe lemma. Let S : X ↪ Y ↠ Z be a conflation and let K_X ↪ Q_X ↠ X and K_Z ↪ Q_Z ↠ Z be conflations resolving its two outer terms, with Q_Z projective. Then Y is the third term of a conflation whose middle term is Q_X ⊞ Q_Z and whose first term is an extension of K_Z by K_X.

                      The two conflations form a horseshoe over the given one: the new deflation a restricts on the first summand to Q_X ↠ X ↪ Y and covers Q_Z ↠ Z on the second, while the new inflation u restricts on K_X to K_X ↪ Q_X ↪ Q_X ⊞ Q_Z and covers K_Z ↪ Q_Z on the second summand.

                      Iterating this is what lifts resolutions of X and of Z to a resolution of Y; see TauCeti.ExactStructure.exists_finiteResolution_length_le_of_conflation.

                      A property consisting of projectives is extension closed, as soon as it is replete and closed under binary biproducts: a conflation whose third term is projective splits, so its middle term is the biproduct of the two outer ones.

                      Together with TauCeti.ExactStructure.fullSubcategory this equips the objects satisfying P with an induced exact structure — necessarily the split one, by TauCeti.ExactStructure.fullSubcategory_eq_split.

                      A property containing a zero object and closed under binary products is automatically replete, by CategoryTheory.ObjectProperty.isClosedUnderIsomorphisms_of_containsZero; so the results below need no separate repleteness hypothesis, even though the resolution API they call does.

                      Finite resolutions by projectives lift along a conflation. If P consists of E-projectives, contains a zero object and is closed under binary biproducts, and both outer terms of a conflation admit a finite P-resolution of length at most n, then so does its middle term.

                      This is the horseshoe lemma, iterated: at each stage the two resolving terms are added, and the new syzygy is an extension of the two old ones.

                      The objects of finite P-dimension are extension closed, for an object property P consisting of E-projectives which contains a zero object and is closed under binary biproducts. The middle term of a conflation is covered by TauCeti.ExactStructure.IsExtensionClosed.prop_X₂.

                      Together with TauCeti.ExactStructure.fullSubcategory this equips them with an induced exact structure, in which the resolution theorem compares their Grothendieck group with the one generated by the objects satisfying P.