Introduction and main result
A triangulated category is topological if it admits a fully faithful triangle functor into the homotopy category of a stable model category. The first nontopological example was found by Muro, Schwede and Strickland, who constructed a non-topological triangulation on finitely generated free Z / 4 \mathbb{Z}/4 Z /4 -modules, with identity suspension [16 Theorem 1] . Their Hopf-map obstruction shows that every triangle functor from this category to a topological triangulated category is zero [16 Theorem 2] . Rizzardo and Van den Bergh constructed nonenhanceable triangulated categories by gluing algebraic ones along a triangle functor that does not lift to an A ∞ A_{\infty} A ∞ -functor [18 Theorem 1.1] . Recently Díaz Cabrera and Muro constructed a non-topological pretriangulation on finitely generated projective modules over the preprojective algebra P ( L 2 ) P(L_{2}) P ( L 2 ) by twisting the inverse suspension of an algebraic triangulation [6 ] . He and Le proved that this pretriangulation satisfies the octahedral axiom over every algebraically closed field [9 Theorem 5.6] .
For an infinite cardinal θ \theta θ , we call a triangulated category θ \theta θ -dg enhanced if every thick subcategory generated by a subset S S S with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ is algebraic. We ask whether triangulated categories with θ \theta θ -dg enhancement are algebraic or topological.
The following theorem gives a negative answer.
Main Theorem. Let R R R be a nonzero commutative unital ring and let θ \theta θ be an infinite cardinal. There is an triangulated category T \mathcal{T} T such that:
(1) T \mathcal{T} T is not topological;
(2) T \mathcal{T} T is θ \theta θ -dg enhanced, i.e., for each subset S S S with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ , thick T ( S ) \operatorname{thick}_{\mathcal{T}}(S) thick T ( S ) is algebraic;
(3) The scalar action R → Z ( T ) R \to Z(\mathcal{T}) R → Z ( T ) is faithful.
Corollary 1.1. For any field k k k , there is a nontopological triangulated category whose finitely generated thick subcategories are algebraic.
Remark 1.2. For every cardinal κ \kappa κ , neither the class of algebraic triangulated categories nor the class of topological triangulated categories can be characterised by diagram-extension axioms, each involving at most κ \kappa κ objects and morphisms.
By contrast, a pretriangulated category is triangulated if every finitely generated thick subcategory is triangulated with respect to the inherited structure. Roughly speaking, there are no easy definition for algebraic to topological triangulated categories.
We briefly describe the construction, and the obstruction to topological realisation.
First, we define C \mathcal{C} C to be the R R R -linearisation of ω 3 \omega_3 ω 3 , the first ordinal ω 3 \omega_3 ω 3 with cardinality ℵ 3 \aleph_3 ℵ 3 . Then we take a graded R R R -category B \mathcal{B} B by adding degree ( − 2 ) (-2) ( − 2 ) -terms on H o m \mathcal{H}om H o m , and define two minimal A ∞ A_\infty A ∞ -categories Λ ± \Lambda_{\pm} Λ ± on B \mathcal{B} B , with distinct m 4 m_4 m 4 . Then one has Perf ( Λ ± ) ⊆ V ± ⊆ D ∞ ( Λ ± ) \operatorname{Perf}(\Lambda_{\pm}) \subseteq\mathcal{V}_{\pm} \subseteq\mathcal{D}_{\infty}(\Lambda_{\pm}) Perf ( Λ ± ) ⊆ V ± ⊆ D ∞ ( Λ ± ) , where V ± \mathcal{V}_{\pm} V ± are generated by (< θ <\theta < θ ) copies of coproducts of representable functors.
Fix I = θ × { + , − } I = \theta\times\{+,-\} I = θ × { + , − } , Sign ( ξ , b ) = b \operatorname{Sign}(\xi,b)=b Sign ( ξ , b ) = b . The final category T \mathcal{T} T is a colimit of T J = ( ∏ i ∈ J V Sign ( i ) ) × Perf ( Λ − ( I ∖ J ) ) \mathcal{T}_J = \left(\prod_{i\in J}\mathcal{V}_{\operatorname{Sign}(i)}\right) \times\operatorname{Perf}(\Lambda_{-}(I\setminus J)) T J = ( ∏ i ∈ J V Sign ( i ) ) × Perf ( Λ − ( I ∖ J )) with ∣ J ∣ < ℵ 0 |J|<\aleph_0 ∣ J ∣ < ℵ 0 , i.e., the union of algebraic triangulated categories with “only finite large branches”. Some set-theoretic analysis shows that T \mathcal{T} T is θ \theta θ -dg enhanced.
The topological obstruction is based several “bridges” which connects fourfold Toda brackets, Yoneda extensions, and categorical cohomology of ω 3 \omega_3 ω 3 . Hence the difference of m 4 ± m^{\pm}_4 m 4 ± admits several representations, yielding a map ( χ ( t i ) ) i ∈ I (\chi(t_i))_{i\in I} ( χ ( t i ) ) i ∈ I from I I I . An analysis on infinite combinatorics shows that ( χ ( t i ) ) i ∈ I (\chi(t_i))_{i\in I} ( χ ( t i ) ) i ∈ I is constant except on a subset S ⊆ I S\subseteq I S ⊆ I with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ . Meanwhile standard computation on triangulated categories shows that ( χ ( t i ) ) i ∈ I (\chi(t_i))_{i\in I} ( χ ( t i ) ) i ∈ I vanishes on θ × { − } \theta\times\{-\} θ × { − } , while is nonvanishing on θ × { + } \theta\times\{+\} θ × { + } . This yields the contradiction.
Section 2 2 2 provides preliminaries; Section 3 3 3 begins with an outline followed by the proofs.
Preliminaries
Fix a commutative ring R R R . Unless otherwise specified, ⊗ \otimes ⊗ denotes ⊗ R \otimes_R ⊗ R . We use the Koszul sign rule ( f ⊗ g ) ( x ⊗ y ) = ( − 1 ) ∣ g ∣ ∣ x ∣ f ( x ) ⊗ g ( y ) (f\otimes g)(x\otimes y)=(-1)^{|g||x|}f(x)\otimes g(y) ( f ⊗ g ) ( x ⊗ y ) = ( − 1 ) ∣ g ∣∣ x ∣ f ( x ) ⊗ g ( y ) for homogeneous maps and elements. All A ∞ A_\infty A ∞ -categories, functors, modules and module morphisms are strictly unital. The triangulated categories constructed in the main theorem are idempotent complete.
An R R R -category is a category whose Hom \operatorname{Hom} Hom sets are R R R -modules, and the composition of morphisms is R R R -bilinear. A functor between R R R -linear categories is R R R -linear, if it induces R R R -homomorphisms on Hom \operatorname{Hom} Hom groups.
Triangulated categories
We use Verdier’s axioms and the usual suspension-compatible triangle functors [22 ] . Our cone convention is Cone ( f ) i = N i ⊕ M i + 1 \operatorname{Cone}(f)^i=N^i\oplus M^{i+1} Cone ( f ) i = N i ⊕ M i + 1 with d ( y , x ) = ( d N y + f ( x ) , − d M x ) d(y,x)=(d_Ny+f(x),-d_Mx) d ( y , x ) = ( d N y + f ( x ) , − d M x ) for a chain map f : M → N f:M\to N f : M → N . Write Z ( T ) Z(\mathcal{T}) Z ( T ) for the natural endomorphisms of id T \operatorname{id}_{\mathcal{T}} id T commuting with suspension; an R R R -linear scalar action is faithful when r ↦ ( r 1 X ) X r\mapsto(r1_X)_X r ↦ ( r 1 X ) X defines an injection R → Z ( T ) R\to Z(\mathcal{T}) R → Z ( T ) . A full triangulated subcategory is thick if it is closed under direct summands, and thick ( X ) \operatorname{thick}(\mathcal{X}) thick ( X ) denotes the thick subcategory generated by X \mathcal{X} X . A thick subcategory T 0 \mathcal{T}_0 T 0 is finitely generated if T 0 = thick ( G ) \mathcal{T}_0=\operatorname{thick}(G) T 0 = thick ( G ) for some object G G G , called a classical generator [4 §2.1] .
We use exact categories in Quillen’s sense [5 Definition 2.1] . A Frobenius exact category has enough projectives and injectives, with the two classes coinciding; its stable category , obtained by quotienting out morphisms factoring through these objects, is triangulated [8 Chapter I, Theorem 2.6] .
Definition 2.1. A triangulated category is algebraic (over R R R ) if it is R R R -linearly triangle equivalent to the stable category of an R R R -linear Frobenius exact category.
A model category is understood in the usual complete and cocomplete sense [10 Definitions 1.1.3–1.1.4] , originating in [17 ] . Its homotopy category is Ho ( M ) = M [ Weq − 1 ] \operatorname{Ho}(\mathcal{M})=\mathcal{M}[\operatorname{Weq}^{-1}] Ho ( M ) = M [ Weq − 1 ] , where Weq \operatorname{Weq} Weq is the class of weak equivalences. A pointed model category is stable if its derived suspension is an equivalence; its homotopy category is then canonically triangulated [10 Definition 7.1.1 and Proposition 7.1.6] .
Definition 2.2. A triangulated category is topological if it admits a fully faithful triangle functor into Ho ( M ) \operatorname{Ho}(\mathcal{M}) Ho ( M ) for some stable model category M \mathcal{M} M [20 ] .
A-infinity-categories and dg categories
Definition 2.3. Following [11 §3.1] , an A ∞ A_\infty A ∞ -category A \mathcal{A} A has Z \mathbb{Z} Z -graded R R R -modules Hom A ( X , Y ) \operatorname{Hom}_{\mathcal{A}}(X,Y) Hom A ( X , Y ) as its Hom sets, with R R R -multilinear operations
m n : Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n , X n − 1 ) ⟶ Hom A ( X n , X 0 ) ( n ≥ 1 ) m_n:\operatorname{Hom}_{\mathcal{A}}(X_1,X_0)\otimes\cdots\otimes\operatorname{Hom}_{\mathcal{A}}(X_n,X_{n-1})\longrightarrow\operatorname{Hom}_{\mathcal{A}}(X_n,X_0)\qquad(n\geq1) m n : Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n , X n − 1 ) ⟶ Hom A ( X n , X 0 ) ( n ≥ 1 ) of degree 2 − n 2-n 2 − n , satisfying Stasheff identities:
∑ r + s + t = n , r , t ≥ 0 , s ≥ 1 ( − 1 ) r + s t m r + 1 + t ( id ⊗ r ⊗ m s ⊗ id ⊗ t ) = 0. ( n ≥ 1 ) \sum_{\substack{r+s+t=n,\\ r,t\geq0,\ s\geq1}}(-1)^{r+st}m_{r+1+t}(\operatorname{id}^{\otimes r}\otimes m_s\otimes\operatorname{id}^{\otimes t})=0.\qquad(n\geq1) r + s + t = n , r , t ≥ 0 , s ≥ 1 ∑ ( − 1 ) r + s t m r + 1 + t ( id ⊗ r ⊗ m s ⊗ id ⊗ t ) = 0. ( n ≥ 1 ) Each object X X X is equipped with 1 X ∈ Hom A 0 ( X , X ) 1_X\in\operatorname{Hom}_{\mathcal{A}}^0(X,X) 1 X ∈ Hom A 0 ( X , X ) , satisfying
m 2 ( 1 Y , a ) = a = m 2 ( a , 1 X ) ( a ∈ Hom A ( X , Y ) ) . m_2(1_Y,a)=a=m_2(a,1_X)\qquad(a\in\operatorname{Hom}_{\mathcal{A}}(X,Y)). m 2 ( 1 Y , a ) = a = m 2 ( a , 1 X ) ( a ∈ Hom A ( X , Y )) . For every composable sequence ( a 1 , … , a n ) (a_1,\ldots,a_n) ( a 1 , … , a n ) , one also has
m n ( a 1 , … , a n ) = 0 if n ≠ 2 and some a j is an identity . m_n(a_1,\ldots,a_n)=0\quad\text{if }n\neq2\text{ and some }a_j\text{ is an identity}. m n ( a 1 , … , a n ) = 0 if n = 2 and some a j is an identity . In particular, m 1 ( 1 X ) = 0 m_1(1_X)=0 m 1 ( 1 X ) = 0 .
Definition 2.4. An A ∞ A_\infty A ∞ -functor F : A → B F:\mathcal{A}\to\mathcal{B} F : A → B between A ∞ A_\infty A ∞ -categories assigns an object F X ∈ B FX\in\mathcal{B} F X ∈ B to each X ∈ A X\in\mathcal{A} X ∈ A and, for each n ≥ 1 n\geq1 n ≥ 1 and X 0 , … , X n ∈ A X_0,\ldots,X_n\in\mathcal{A} X 0 , … , X n ∈ A , an R R R -linear map
F n : Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n , X n − 1 ) ⟶ Hom B ( F X n , F X 0 ) F_n:\operatorname{Hom}_{\mathcal{A}}(X_1,X_0)\otimes\cdots\otimes\operatorname{Hom}_{\mathcal{A}}(X_n,X_{n-1})\longrightarrow\operatorname{Hom}_{\mathcal{B}}(FX_n,FX_0) F n : Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n , X n − 1 ) ⟶ Hom B ( F X n , F X 0 ) of degree 1 − n 1-n 1 − n , satisfying, for every n ≥ 1 n\geq1 n ≥ 1 ,
∑ r + s + t = n , r , t ≥ 0 , s ≥ 1 ( − 1 ) r + s t F r + 1 + t ( id ⊗ r ⊗ m s A ⊗ id ⊗ t ) = ∑ k ≥ 1 , i 1 + ⋯ + i k = n i 1 , … , i k ≥ 1 ( − 1 ) ∑ j = 1 k ( k − j ) ( i j − 1 ) m k B ( F i 1 ⊗ ⋯ ⊗ F i k ) . \sum_{\substack{r+s+t=n,\\ r,t\geq0,\ s\geq1}}(-1)^{r+st}F_{r+1+t}(\operatorname{id}^{\otimes r}\otimes m_s^{\mathcal{A}}\otimes\operatorname{id}^{\otimes t})
=
\sum_{\substack{k\geq1,\ i_1+\cdots+i_k=n\\ i_1,\ldots,i_k\geq1}}(-1)^{\sum_{j=1}^{k}(k-j)(i_j-1)}m_k^{\mathcal{B}}(F_{i_1}\otimes\cdots\otimes F_{i_k}). r + s + t = n , r , t ≥ 0 , s ≥ 1 ∑ ( − 1 ) r + s t F r + 1 + t ( id ⊗ r ⊗ m s A ⊗ id ⊗ t ) = k ≥ 1 , i 1 + ⋯ + i k = n i 1 , … , i k ≥ 1 ∑ ( − 1 ) ∑ j = 1 k ( k − j ) ( i j − 1 ) m k B ( F i 1 ⊗ ⋯ ⊗ F i k ) . Graded tensor products of maps use the Koszul sign rule; see [11 §§3.4,7.3] . We also require
F 1 ( 1 X ) = 1 F X , F n ( a 1 , … , a n ) = 0 if n > 1 and some a j is an identity . F_1(1_X)=1_{FX},\qquad F_n(a_1,\ldots,a_n)=0\quad\text{if }n>1\text{ and some }a_j\text{ is an identity}. F 1 ( 1 X ) = 1 F X , F n ( a 1 , … , a n ) = 0 if n > 1 and some a j is an identity . An A ∞ A_\infty A ∞ -algebra is an A ∞ A_\infty A ∞ -category with one object. A morphism of A ∞ A_\infty A ∞ -algebras is an A ∞ A_\infty A ∞ -functor between the corresponding one-object categories. Such a morphism f : A → B f:\mathcal{A}\to\mathcal{B} f : A → B is a quasi-isomorphism if f 1 : ( A , m 1 ) → ( B , m 1 ) f_1:(\mathcal{A},m_1)\to(\mathcal{B},m_1) f 1 : ( A , m 1 ) → ( B , m 1 ) induces an isomorphism on cohomology.
An A ∞ A_\infty A ∞ -category is minimal if m 1 = 0 m_1=0 m 1 = 0 on every morphism space.
Lemma 2.5. For a minimal A ∞ A_\infty A ∞ -category A \mathcal{A} A , objects X , Y ∈ A X,Y\in\mathcal{A} X , Y ∈ A and d ∈ Z d\in\mathbb{Z} d ∈ Z ,
H d ( Hom A ( X , Y ) , m 1 ) = Hom A ( X , Y ) d . H^d(\operatorname{Hom}_{\mathcal{A}}(X,Y),m_1)=\operatorname{Hom}_{\mathcal{A}}(X,Y)^d. H d ( Hom A ( X , Y ) , m 1 ) = Hom A ( X , Y ) d .
Proof. This follows from m 1 = 0 m_1=0 m 1 = 0 . □
Definition 2.6. A right A ∞ A_\infty A ∞ -module over an A ∞ A_\infty A ∞ -category A \mathcal{A} A assigns a graded R R R -module M ( X ) M(X) M ( X ) to every object X X X , with R R R -linear maps
m n M : M ( X 0 ) ⊗ Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n − 1 , X n − 2 ) ⟶ M ( X n − 1 ) m_n^M:M(X_0)\otimes\operatorname{Hom}_{\mathcal{A}}(X_1,X_0)\otimes\cdots\otimes\operatorname{Hom}_{\mathcal{A}}(X_{n-1},X_{n-2})\longrightarrow M(X_{n-1}) m n M : M ( X 0 ) ⊗ Hom A ( X 1 , X 0 ) ⊗ ⋯ ⊗ Hom A ( X n − 1 , X n − 2 ) ⟶ M ( X n − 1 ) of degree 2 − n 2-n 2 − n for n ≥ 1 n\geq1 n ≥ 1 , satisfying the module identities of [11 §4.2] on composable strings. We require m 2 M ( x , 1 X ) = x m_2^M(x,1_X)=x m 2 M ( x , 1 X ) = x , and m n M m_n^M m n M vanishes on inputs containing an identity
when n ≠ 2 n \ne2 n = 2 . The representable module at Y Y Y is M ( X ) = H o m A ( X , Y ) M(X)=\mathcal{H}om_{\mathcal{A}}(X,Y) M ( X ) = H o m A ( X , Y ) with the category operations.
Definition 2.7. A morphism of right A ∞ A_{\infty} A ∞ -modules f : L → M f:L\to M f : L → M has R R R -linear components
f n : L ( X 0 ) ⊗ H o m A ( X 1 , X 0 ) ⊗ ⋯ ⊗ H o m A ( X n − 1 , X n − 2 ) ⟶ M ( X n − 1 ) f_n:L(X_0)\otimes\mathcal{H}om_{\mathcal{A}}(X_1,X_0)\otimes\cdots\otimes\mathcal{H}om_{\mathcal{A}}(X_{n-1},X_{n-2})\longrightarrow M(X_{n-1}) f n : L ( X 0 ) ⊗ H o m A ( X 1 , X 0 ) ⊗ ⋯ ⊗ H o m A ( X n − 1 , X n − 2 ) ⟶ M ( X n − 1 )
of degree 1 − n 1-n 1 − n for n ≥ 1 n\ge1 n ≥ 1 , with the identities and composition of [11 §4.2] on composable strings. For n > 1 n>1 n > 1 , f n f_n f n vanishes on inputs containing an identity. It is a quasi-isomorphism if f 1 f_1 f 1 induces an isomorphism on cohomology at every object.
Definition 2.8. A differential graded (dg) category is an A ∞ A_{\infty} A ∞ -category with m n = 0 m_n=0 m n = 0 for n ≥ 3 n\ge3 n ≥ 3 . See [12 §2.2] . A dg functor between dg categories is an A ∞ A_{\infty} A ∞ -functor with F n = 0 F_n=0 F n = 0 for n ≥ 2 n\ge2 n ≥ 2 . A dg algebra is a dg category with one object, and a morphism of dg algebras is a dg functor between the corresponding one-object categories.
Write C d g ( R ) \mathcal{C}_{\mathrm{dg}}(R) C dg ( R ) for the usual dg category of cochain complexes [12 §2.2] .
Definition 2.9. A right dg module over A \mathcal{A} A is a dg functor A o p → C d g ( R ) \mathcal{A}^{\mathrm{op}}\to\mathcal{C}_{\mathrm{dg}}(R) A op → C dg ( R ) . The representable module associated to X X X is denoted by H o m A ( − , X ) \mathcal{H}om_{\mathcal{A}}(-,X) H o m A ( − , X ) .
Homotopy categories and enhancements
For an A ∞ A_{\infty} A ∞ -algebra A \mathcal{A} A , its graded cohomology module is
H ∗ ( A , m 1 ) = ∏ p ∈ Z H p ( A , m 1 ) . H^*(\mathcal{A},m_1)=\prod_{p\in\mathbb{Z}}H^p(\mathcal{A},m_1). H ∗ ( A , m 1 ) = p ∈ Z ∏ H p ( A , m 1 ) .
The product [ a ] [ b ] = [ m 2 ( a , b ) ] [a][b]=[m_2(a,b)] [ a ] [ b ] = [ m 2 ( a , b )] makes H ∗ ( A , m 1 ) H^*(\mathcal{A},m_1) H ∗ ( A , m 1 ) a graded associative R R R -algebra with unit [ 1 A ] [1_{\mathcal{A}}] [ 1 A ] ; we also denote the induced product by m 2 m_2 m 2 .
The homotopy category H 0 ( A ) H^0(\mathcal{A}) H 0 ( A ) has the objects of A \mathcal{A} A , morphism groups
Hom H 0 ( A ) ( X , Y ) = H 0 ( H o m A ( X , Y ) , m 1 ) , \operatorname{Hom}_{H^0(\mathcal{A})}(X,Y)=H^0(\mathcal{H}om_{\mathcal{A}}(X,Y),m_1), Hom H 0 ( A ) ( X , Y ) = H 0 ( H o m A ( X , Y ) , m 1 ) ,
and composition induced by m 2 m_2 m 2 .
Definition 2.10. A dg category is pretriangulated if the zero module, all integer shifts of representable modules, and cones of closed degree-zero maps between representable modules are homotopy equivalent to representable modules [7 §2.4] .
Definition 2.11. An R R R -linear dg enhancement of a triangulated category T \mathcal{T} T is a pretriangulated dg category A \mathcal{A} A with an R R R -linear triangle equivalence H 0 ( A ) ≃ T H^0(\mathcal{A})\simeq\mathcal{T} H 0 ( A ) ≃ T .
Every small algebraic triangulated category admits a dg enhancement [12 Theorem 3.8] ; conversely, H 0 ( A ) H^0(\mathcal{A}) H 0 ( A ) is algebraic for every pretriangulated R R R -linear dg category A \mathcal{A} A [12 §4.5] .
Hochschild cohomology
Let P P P be a Z \mathbb{Z} Z -graded R R R -algebra and N N N a graded P P P -bimodule whose left and right R R R -actions agree. For graded R R R -modules M , N M,N M , N , write
H o m R ( M , N ) q = ∏ d ∈ Z Hom R ( M d , N d + q ) . \mathcal{H}om_R(M,N)^q=\prod_{d\in\mathbb{Z}}\operatorname{Hom}_R(M^d,N^{d+q}). H o m R ( M , N ) q = d ∈ Z ∏ Hom R ( M d , N d + q ) .
For each p ≥ 0 p\ge0 p ≥ 0 and q ∈ Z q\in\mathbb{Z} q ∈ Z , write C R p , q ( P , N ) = H o m R ( P ⊗ p , N ) q C_R^{p,q}(P,N)=\mathcal{H}om_R(P^{\otimes p},N)^q C R p , q ( P , N ) = H o m R ( P ⊗ p , N ) q , where P ⊗ 0 = R P^{\otimes0}=R P ⊗ 0 = R . Define the Hochschild differential by
δ : C R p , q ( P , N ) ⟶ C R p + 1 , q ( P , N ) , f ⟼ δ f , \delta:C_R^{p,q}(P,N)\longrightarrow C_R^{p+1,q}(P,N),\qquad f\longmapsto\delta f, δ : C R p , q ( P , N ) ⟶ C R p + 1 , q ( P , N ) , f ⟼ δ f ,
where, for homogeneous a 1 , … , a p + 1 ∈ P a_1,\ldots,a_{p+1}\in P a 1 , … , a p + 1 ∈ P ,
( δ f ) ( a 1 , … , a p + 1 ) = ( − 1 ) q ∣ a 1 ∣ a 1 f ( a 2 , … , a p + 1 ) + ∑ i = 1 p ( − 1 ) i f ( a 1 , … , a i − 1 , a i a i + 1 , a i + 2 , … , a p + 1 ) + ( − 1 ) p + 1 f ( a 1 , … , a p ) a p + 1 . (\delta f)(a_1,\ldots,a_{p+1})=(-1)^{q|a_1|}a_1f(a_2,\ldots,a_{p+1})+\sum_{i=1}^{p}(-1)^if(a_1,\ldots,a_{i-1},a_ia_{i+1},a_{i+2},\ldots,a_{p+1})+(-1)^{p+1}f(a_1,\ldots,a_p)a_{p+1}. ( δ f ) ( a 1 , … , a p + 1 ) = ( − 1 ) q ∣ a 1 ∣ a 1 f ( a 2 , … , a p + 1 ) + i = 1 ∑ p ( − 1 ) i f ( a 1 , … , a i − 1 , a i a i + 1 , a i + 2 , … , a p + 1 ) + ( − 1 ) p + 1 f ( a 1 , … , a p ) a p + 1 .
For p = 0 p=0 p = 0 , we identify f ∈ C R 0 , q ( P , N ) f\in C_R^{0,q}(P,N) f ∈ C R 0 , q ( P , N ) with f ( 1 ) ∈ N q f(1)\in N^q f ( 1 ) ∈ N q , so that ( δ f ) ( a ) = ( − 1 ) q ∣ a ∣ a f − f a (\delta f)(a)=(-1)^{q|a|}af-fa ( δ f ) ( a ) = ( − 1 ) q ∣ a ∣ a f − f a . Then δ 2 = 0 \delta^2=0 δ 2 = 0 , and the Hochschild cohomology is H H R p , q ( P , N ) = H p ( C R ∙ , q ( P , N ) , δ ) HH_R^{p,q}(P,N)=H^p(C_R^{\bullet,q}(P,N),\delta) H H R p , q ( P , N ) = H p ( C R ∙ , q ( P , N ) , δ ) ; [15 §2] .
For algebras and bimodules concentrated in degree zero, write H H R p ( P , N ) = H H R p , 0 ( P , N ) HH_R^p(P,N)=HH_R^{p,0}(P,N) H H R p ( P , N ) = H H R p , 0 ( P , N ) .
For an R R R -linear category C \mathcal{C} C and a bimodule E E E , the same definitions use products over composable object strings, with R R R -multilinear cochains valued in E E E and the same end actions and signs. We write C R p ( C , E ) C_R^p(\mathcal{C},E) C R p ( C , E ) and H H R p ( C , E ) HH_R^p(\mathcal{C},E) H H R p ( C , E ) when C \mathcal{C} C and E E E are concentrated in degree zero. Normalized cochains vanish on identity inputs.
Twisted complexes
2.5. 2.5. 2.5. Twisted complexes. Finite twisted complexes to compute homotopy categories in Proposition 3.11 . Let A \mathcal{A} A be an A ∞ A_\infty A ∞ -category, with the conventions of [11 §§3.6,7.6] .
The objects of Σ A \Sigma\mathcal{A} Σ A are formal finite coproducts E = ⨁ i = 1 N X i [ a i ] E=\bigoplus_{i=1}^{N}X_i[a_i] E = ⨁ i = 1 N X i [ a i ] , with X i ∈ A X_i\in\mathcal{A} X i ∈ A and a i ∈ Z a_i\in\mathbb{Z} a i ∈ Z . Write R [ a ] R[a] R [ a ] for the rank-one graded free R R R -module concentrated in degree − a -a − a . Set H o m Σ A ( X [ a ] , Y [ b ] ) = H o m R ( R [ a ] , R [ b ] ) ⊗ H o m A ( X , Y ) ≅ H o m A ( X , Y ) [ b − a ] \mathcal{H}om_{\Sigma\mathcal{A}}(X[a],Y[b])=\mathcal{H}om_R(R[a],R[b])\otimes\mathcal{H}om_{\mathcal{A}}(X,Y)\cong\mathcal{H}om_{\mathcal{A}}(X,Y)[b-a] H o m Σ A ( X [ a ] , Y [ b ]) = H o m R ( R [ a ] , R [ b ]) ⊗ H o m A ( X , Y ) ≅ H o m A ( X , Y ) [ b − a ] .
For homogeneous u i : R [ a i ] → R [ a i − 1 ] u_i:R[a_i]\to R[a_{i-1}] u i : R [ a i ] → R [ a i − 1 ] and f i ∈ H o m A ( X i , X i − 1 ) f_i\in\mathcal{H}om_{\mathcal{A}}(X_i,X_{i-1}) f i ∈ H o m A ( X i , X i − 1 ) , define
m n Σ A ( u 1 ⊗ f 1 , … , u n ⊗ f n ) = ( − 1 ) ∑ i < j ∣ f i ∣ ∣ u j ∣ + ( 2 − n ) ∑ i ∣ u i ∣ ⋅ ( u 1 ⋯ u n ) ⊗ m n A ( f 1 , … , f n ) , m_n^{\Sigma\mathcal{A}}(u_1\otimes f_1,\ldots,u_n\otimes f_n)=(-1)^{\sum_{i<j}|f_i||u_j|+(2-n)\sum_i|u_i|}\cdot(u_1\cdots u_n)\otimes m_n^{\mathcal{A}}(f_1,\ldots,f_n), m n Σ A ( u 1 ⊗ f 1 , … , u n ⊗ f n ) = ( − 1 ) ∑ i < j ∣ f i ∣∣ u j ∣ + ( 2 − n ) ∑ i ∣ u i ∣ ⋅ ( u 1 ⋯ u n ) ⊗ m n A ( f 1 , … , f n ) ,
where ∣ u i ∣ = a i − a i − 1 |u_i|=a_i-a_{i-1} ∣ u i ∣ = a i − a i − 1 . Morphisms between finite coproducts are matrices, and the operations extend by summing over composable matrix entries.
For a Z \mathbb{Z} Z -graded R R R -module V V V , write s : V → V [ 1 ] s:V\to V[1] s : V → V [ 1 ] for the canonical map of degree − 1 -1 − 1 , where V [ 1 ] d = V d + 1 V[1]^d=V^{d+1} V [ 1 ] d = V d + 1 . For n ≥ 1 n\ge1 n ≥ 1 and E 0 , … , E n ∈ Σ A E_0,\ldots,E_n\in\Sigma\mathcal{A} E 0 , … , E n ∈ Σ A , define the degree-one map b n b_n b n by the commutative diagram
⨂ i = 1 n H o m Σ A ( E i , E i − 1 ) → m n Σ A H o m Σ A ( E n , E 0 ) s ⊗ n ↓ ↓ s ⨂ i = 1 n ( H o m Σ A ( E i , E i − 1 ) [ 1 ] ) → b n H o m Σ A ( E n , E 0 ) [ 1 ] . \begin{CD}
\bigotimes_{i=1}^{n}\mathcal{H}om_{\Sigma\mathcal{A}}(E_i,E_{i-1}) @>{m_n^{\Sigma\mathcal{A}}}>> \mathcal{H}om_{\Sigma\mathcal{A}}(E_n,E_0)\\
@V{s^{\otimes n}}VV @VV{s}V\\
\bigotimes_{i=1}^{n}\bigl(\mathcal{H}om_{\Sigma\mathcal{A}}(E_i,E_{i-1})[1]\bigr) @>{b_n}>> \mathcal{H}om_{\Sigma\mathcal{A}}(E_n,E_0)[1].
\end{CD} i = 1 ⨂ n H o m Σ A ( E i , E i − 1 ) s ⊗ n ↓ ⏐ i = 1 ⨂ n ( H o m Σ A ( E i , E i − 1 ) [ 1 ] ) m n Σ A b n H o m Σ A ( E n , E 0 ) ↓ ⏐ s H o m Σ A ( E n , E 0 ) [ 1 ] .
Definition 2.12. A finite twisted complex over A \mathcal{A} A is a pair ( E , q ) (E,q) ( E , q ) , where E = ⨁ i = 1 N X i [ a i ] ∈ Σ A E=\bigoplus_{i=1}^{N}X_i[a_i]\in\Sigma\mathcal{A} E = ⨁ i = 1 N X i [ a i ] ∈ Σ A and q = ( q i j ) 1 ≤ i , j ≤ N ∈ ( H o m Σ A ( E , E ) [ 1 ] ) 0 q=(q_{ij})_{1\le i,j\le N}\in(\mathcal{H}om_{\Sigma\mathcal{A}}(E,E)[1])^0 q = ( q ij ) 1 ≤ i , j ≤ N ∈ ( H o m Σ A ( E , E ) [ 1 ] ) 0 is a matrix with
q i j ∈ ( H o m Σ A ( X j [ a j ] , X i [ a i ] ) [ 1 ] ) 0 , q i j = 0 ( i ≥ j ) . q_{ij}\in\bigl(\mathcal{H}om_{\Sigma\mathcal{A}}(X_j[a_j],X_i[a_i])[1]\bigr)^0,\qquad q_{ij}=0\quad(i\ge j). q ij ∈ ( H o m Σ A ( X j [ a j ] , X i [ a i ]) [ 1 ] ) 0 , q ij = 0 ( i ≥ j ) .
We require q q q to satisfy the Maurer–Cartan equation ∑ n ≥ 1 b n ( q ⊗ n ) = 0 \sum_{n\ge1}b_n(q^{\otimes n})=0 ∑ n ≥ 1 b n ( q ⊗ n ) = 0 [11 §7.6 and §8.6] . Since q i j = 0 q_{ij}=0 q ij = 0 for i ≥ j i\ge j i ≥ j , one has b n ( q ⊗ n ) = 0 b_n(q^{\otimes n})=0 b n ( q ⊗ n ) = 0 for n ≥ N n\ge N n ≥ N , so the sum is finite.
The A ∞ A_\infty A ∞ -category Tw ( A ) \operatorname{Tw}(\mathcal{A}) Tw ( A ) has finite twisted complexes as objects and graded morphism modules H o m Tw ( A ) ( ( E , q ) , ( F , r ) ) = H o m Σ A ( E , F ) \mathcal{H}om_{\operatorname{Tw}(\mathcal{A})}((E,q),(F,r))=\mathcal{H}om_{\Sigma\mathcal{A}}(E,F) H o m Tw ( A ) (( E , q ) , ( F , r )) = H o m Σ A ( E , F ) . For n ≥ 1 n\ge1 n ≥ 1 , objects ( E i , q i ) (E_i,q_i) ( E i , q i ) and x i ∈ H o m Σ A ( E i , E i − 1 ) [ 1 ] x_i\in\mathcal{H}om_{\Sigma\mathcal{A}}(E_i,E_{i-1})[1] x i ∈ H o m Σ A ( E i , E i − 1 ) [ 1 ] . Suspended operations are obtained by inserting twists in all positions:
b n Tw ( x 1 , … , x n ) = ∑ j 0 , … , j n ≥ 0 b n + j 0 + ⋯ + j n ( q 0 ⊗ j 0 , x 1 , q 1 ⊗ j 1 , … , x n , q n ⊗ j n ) . (1) b_n^{\operatorname{Tw}}(x_1,\ldots,x_n)=\sum_{j_0,\ldots,j_n\ge0}b_{n+j_0+\cdots+j_n}(q_0^{\otimes j_0},x_1,q_1^{\otimes j_1},\ldots,x_n,q_n^{\otimes j_n}). \tag*{(1)} b n Tw ( x 1 , … , x n ) = j 0 , … , j n ≥ 0 ∑ b n + j 0 + ⋯ + j n ( q 0 ⊗ j 0 , x 1 , q 1 ⊗ j 1 , … , x n , q n ⊗ j n ) . (1)
A zeroth tensor power inserts no twist, and the triangularity condition makes each sum finite. The higher compositions are determined by the commutative diagram
⨂ i = 1 n Hom Σ A ( E i , E i − 1 ) → m n T w Hom Σ A ( E n , E 0 ) s ⊗ n ↓ ↓ s ⨂ i = 1 n ( Hom Σ A ( E i , E i − 1 ) [ 1 ] ) → b n T w Hom Σ A ( E n , E 0 ) [ 1 ] . \begin{CD}
\bigotimes_{i=1}^{n}\operatorname{Hom}_{\Sigma\mathcal{A}}(E_i,E_{i-1}) @>{m_n^{\mathrm{Tw}}}>> \operatorname{Hom}_{\Sigma\mathcal{A}}(E_n,E_0) \\
@V{s^{\otimes n}}VV @VV{s}V \\
\bigotimes_{i=1}^{n}\left(\operatorname{Hom}_{\Sigma\mathcal{A}}(E_i,E_{i-1})[1]\right) @>{b_n^{\mathrm{Tw}}}>> \operatorname{Hom}_{\Sigma\mathcal{A}}(E_n,E_0)[1].
\end{CD} i = 1 ⨂ n Hom Σ A ( E i , E i − 1 ) s ⊗ n ↓ ⏐ i = 1 ⨂ n ( Hom Σ A ( E i , E i − 1 ) [ 1 ] ) m n Tw b n Tw Hom Σ A ( E n , E 0 ) ↓ ⏐ s Hom Σ A ( E n , E 0 ) [ 1 ] .
The Maurer–Cartan equations ensure the A ∞ A_\infty A ∞ -identities [11 §8.6] . The units are diagonal identity matrices, and X ↦ ( X , 0 ) X \mapsto(X,0) X ↦ ( X , 0 ) gives a fully faithful A ∞ A_\infty A ∞ -embedding A ↪ Tw ( A ) \mathcal{A} \hookrightarrow\operatorname{Tw}(\mathcal{A}) A ↪ Tw ( A ) whose components of order n ≥ 2 n \ge2 n ≥ 2 vanish.
If R R R is a field, then H 0 ( Tw ( A ) ) H^0(\operatorname{Tw}(\mathcal{A})) H 0 ( Tw ( A )) is triangulated, with suspension and distinguished triangles induced by shifts and cones of twisted complexes [11 §§7.5–7.6] .
For the categories used below, the same construction embeds in the dg category of strictly unital modules over R R R ; shifts and cones give the required triangulation. The arbitrary-ring homotopy Yoneda comparison is used explicitly in Lemma 3.12. We write Tw ( A ) = tw ( A ) \operatorname{Tw}(\mathcal{A})=\operatorname{tw}(\mathcal{A}) Tw ( A ) = tw ( A ) and Perf ( A ) \operatorname{Perf}(\mathcal{A}) Perf ( A ) for the idempotent completion of H 0 Tw ( A ) H^0\operatorname{Tw}(\mathcal{A}) H 0 Tw ( A ) .
Proposition 2.13 ([1 §11.20, §12.1 and Proposition 13.19] ). Every strictly unital A ∞ A_\infty A ∞ -functor f : A → B f:\mathcal{A}\to\mathcal{B} f : A → B extends to an A ∞ A_\infty A ∞ -functor
Tw ( f ) : Tw ( A ) ⟶ Tw ( B ) , \operatorname{Tw}(f):\operatorname{Tw}(\mathcal{A})\longrightarrow\operatorname{Tw}(\mathcal{B}), Tw ( f ) : Tw ( A ) ⟶ Tw ( B ) , which induces a triangle functor on H 0 H^0 H 0 .
Fourfold Toda brackets
Let T \mathcal{T} T be a triangulated category. We use the fourfold Toda bracket to distinguish the two branches in Proposition 3.14.
Definition 2.14 ([19 Definitions 4.2 and 4.4] ). Consider a five-term complex in T \mathcal{T} T ,
X 4 → d 4 X 3 → d 3 X 2 → d 2 X 1 → d 1 X 0 . X_4 \xrightarrow{d_4} X_3 \xrightarrow{d_3} X_2 \xrightarrow{d_2} X_1 \xrightarrow{d_1} X_0. X 4 d 4 X 3 d 3 X 2 d 2 X 1 d 1 X 0 . The fourfold Toda bracket ⟨ d 1 , d 2 , d 3 , d 4 ⟩ \langle d_1,d_2,d_3,d_4\rangle ⟨ d 1 , d 2 , d 3 , d 4 ⟩ consists of the morphisms γ : X 4 [ 2 ] → X 0 \gamma:X_4[2]\to X_0 γ : X 4 [ 2 ] → X 0 for which there exist objects C , Y C,Y C , Y and maps satisfying the following conditions:
The morphism γ \gamma γ factors as X 4 [ 2 ] → α Y → β X 0 X_4[2]\xrightarrow{\alpha}Y\xrightarrow{\beta}X_0 X 4 [ 2 ] α Y β X 0 for some γ = β ∘ α \gamma=\beta\circ\alpha γ = β ∘ α
The following triangles are distinguished:
X 1 → u C → q X 2 [ 1 ] → d 2 [ 1 ] X 1 [ 1 ] , C → v Y → p X 3 [ 2 ] → δ C [ 1 ] . X_1\xrightarrow{u}C\xrightarrow{q}X_2[1]\xrightarrow{d_2[1]}X_1[1],\qquad C\xrightarrow{v}Y\xrightarrow{p}X_3[2]\xrightarrow{\delta}C[1]. X 1 u C q X 2 [ 1 ] d 2 [ 1 ] X 1 [ 1 ] , C v Y p X 3 [ 2 ] δ C [ 1 ] . The following three diagrams commute:
X 3 [ 2 ] → δ C [ 1 ] d 3 [ 2 ] ↓ ↓ q [ 1 ] X 2 [ 2 ] = X 2 [ 2 ] X 4 [ 2 ] → α Y d 4 [ 2 ] ↓ ↓ p X 3 [ 2 ] = X 3 [ 2 ] X 1 → v ∘ u Y d 1 ↓ ↓ β X 0 = X 0 \begin{CD}
X_3[2] @>{\delta}>> C[1] \\
@V{d_3[2]}VV @VV{q[1]}V \\
X_2[2] @= X_2[2]
\end{CD}
\qquad
\begin{CD}
X_4[2] @>{\alpha}>> Y \\
@V{d_4[2]}VV @VV{p}V \\
X_3[2] @= X_3[2]
\end{CD}
\qquad
\begin{CD}
X_1 @>{v\circ u}>> Y \\
@V{d_1}VV @VV{\beta}V \\
X_0 @= X_0
\end{CD} X 3 [ 2 ] d 3 [ 2 ] ↓ ⏐ X 2 [ 2 ] δ C [ 1 ] ↓ ⏐ q [ 1 ] X 2 [ 2 ] X 4 [ 2 ] d 4 [ 2 ] ↓ ⏐ X 3 [ 2 ] α Y ↓ ⏐ p X 3 [ 2 ] X 1 d 1 ↓ ⏐ X 0 v ∘ u Y ↓ ⏐ β X 0
We regard ⟨ d 1 , d 2 , d 3 , d 4 ⟩ \langle d_1,d_2,d_3,d_4\rangle ⟨ d 1 , d 2 , d 3 , d 4 ⟩ as a subset of Hom T ( X 4 , X 0 [ − 2 ] ) \operatorname{Hom}_{\mathcal{T}}(X_4,X_0[-2]) Hom T ( X 4 , X 0 [ − 2 ]) .
A full additive subcategory U ⊆ T \mathcal{U}\subseteq\mathcal{T} U ⊆ T is 2 2 2 -split if Hom T ( P , Q [ r ] ) = 0 \operatorname{Hom}_{\mathcal{T}}(P,Q[r])=0 Hom T ( P , Q [ r ]) = 0 for all P , Q ∈ U P,Q\in\mathcal{U} P , Q ∈ U and every odd integer r r r . The subcategory need not be closed under suspension.
Proposition 2.15 ([19 Proposition 4.10] ). If X 0 , … , X 4 X_0,\ldots,X_4 X 0 , … , X 4 in Definition 2.14 belong to a 2-split subcategory, their fourfold Toda bracket is a nonempty coset of im Hom T ( X 4 , d 1 [ − 2 ] ) + im Hom T ( d 4 , X 0 [ − 2 ] ) ⊆ Hom T ( X 4 , X 0 [ − 2 ] ) \operatorname{im}\operatorname{Hom}_{\mathcal{T}}(X_4,d_1[-2])+\operatorname{im}\operatorname{Hom}_{\mathcal{T}}(d_4,X_0[-2])\subseteq\operatorname{Hom}_{\mathcal{T}}(X_4,X_0[-2]) im Hom T ( X 4 , d 1 [ − 2 ]) + im Hom T ( d 4 , X 0 [ − 2 ]) ⊆ Hom T ( X 4 , X 0 [ − 2 ]) .
Remark 2.16. Via its suspension isomorphism, a triangle functor F : T → T ′ F:\mathcal{T}\to\mathcal{T}' F : T → T ′ satisfies F ⟨ d 1 , d 2 , d 3 , d 4 ⟩ ⊆ ⟨ F d 1 , F d 2 , F d 3 , F d 4 ⟩ F\langle d_1,d_2,d_3,d_4\rangle\subseteq\langle Fd_1,Fd_2,Fd_3,Fd_4\rangle F ⟨ d 1 , d 2 , d 3 , d 4 ⟩ ⊆ ⟨ F d 1 , F d 2 , F d 3 , F d 4 ⟩ . Equality holds if F F F is fully faithful and X 0 , … , X 4 X_0,\ldots,X_4 X 0 , … , X 4 belong to a 2-split subcategory.
Category cohomology and characteristic classes
We introduce the category cohomology and characteristic maps used to prove show topological obstructions.
Definition 2.17 ([19 §2.1] ). Let P \mathcal{P} P be a small category and E : P op × P → A b \mathbb{E}:\mathcal{P}^{\operatorname{op}}\times\mathcal{P}\to\mathsf{Ab} E : P op × P → Ab a bifunctor. The cochain complex C ∙ ( P ; E ) C^\bullet(\mathcal{P};\mathbb{E}) C ∙ ( P ; E ) has terms
C 0 ( P ; E ) = ∏ P ∈ Ob ( P ) E ( P , P ) ; C n ( P ; E ) = ∏ P n → f n ⋯ → f 1 P 0 E ( P n , P 0 ) , ( n ≥ 1 ) . C^0(\mathcal{P};\mathbb{E})=\prod_{P\in\operatorname{Ob}(\mathcal{P})}\mathbb{E}(P,P);\qquad C^n(\mathcal{P};\mathbb{E})=\prod_{P_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}P_0}\mathbb{E}(P_n,P_0),\quad(n\geq1). C 0 ( P ; E ) = P ∈ Ob ( P ) ∏ E ( P , P ) ; C n ( P ; E ) = P n f n ⋯ f 1 P 0 ∏ E ( P n , P 0 ) , ( n ≥ 1 ) .
where the second product ranges over all strings of n n n composable morphisms in P \mathcal{P} P . For c ∈ C n ( P ; E ) c\in C^n(\mathcal{P};\mathbb{E}) c ∈ C n ( P ; E ) with n ≥ 1 n\geq1 n ≥ 1 , define δ c ∈ C n + 1 ( P ; E ) \delta c\in C^{n+1}(\mathcal{P};\mathbb{E}) δ c ∈ C n + 1 ( P ; E ) by
( δ c ) ( f 1 , … , f n + 1 ) = E ( 1 , f 1 ) c ( f 2 , … , f n + 1 ) + ∑ j = 1 n ( − 1 ) j c ( f 1 , … , f j ∘ f j + 1 , … , f n + 1 ) + ( − 1 ) n + 1 E ( f n + 1 , 1 ) c ( f 1 , … , f n ) . (\delta c)(f_1,\ldots,f_{n+1})=\mathbb{E}(1,f_1)c(f_2,\ldots,f_{n+1})+\sum_{j=1}^{n}(-1)^jc(f_1,\ldots,f_j\circ f_{j+1},\ldots,f_{n+1})+(-1)^{n+1}\mathbb{E}(f_{n+1},1)c(f_1,\ldots,f_n). ( δ c ) ( f 1 , … , f n + 1 ) = E ( 1 , f 1 ) c ( f 2 , … , f n + 1 ) + j = 1 ∑ n ( − 1 ) j c ( f 1 , … , f j ∘ f j + 1 , … , f n + 1 ) + ( − 1 ) n + 1 E ( f n + 1 , 1 ) c ( f 1 , … , f n ) .
In degree zero, ( δ c ) ( f : P 1 → P 0 ) = E ( 1 , f ) c ( P 1 ) − E ( f , 1 ) c ( P 0 ) (\delta c)(f:P_1\to P_0)=\mathbb{E}(1,f)c(P_1)-\mathbb{E}(f,1)c(P_0) ( δ c ) ( f : P 1 → P 0 ) = E ( 1 , f ) c ( P 1 ) − E ( f , 1 ) c ( P 0 ) .
The cohomology groups of P \mathcal{P} P with coefficients in E \mathbb{E} E are H n ( P ; E ) = H n ( C ∙ ( P ; E ) , δ ) H^n(\mathcal{P};\mathbb{E})=H^n(C^\bullet(\mathcal{P};\mathbb{E}),\delta) H n ( P ; E ) = H n ( C ∙ ( P ; E ) , δ ) .
No additivity condition is imposed on the cochains. A cochain on a category with a zero object is zero-normalised, if it vanishes whenever an input morphism is zero.
We also write H c a t n ( P , E ) = H n ( P ; E ) H^n_{\mathrm{cat}}(\mathcal{P},\mathbb{E})=H^n(\mathcal{P};\mathbb{E}) H cat n ( P , E ) = H n ( P ; E ) . The characteristic map for the ringoid used below is constructed in Definition 3.26.
All size statements are relative to fixed universes; essentially small categories are replaced by small equivalent categories when necessary. We use ⊕ \oplus ⊕ for finite coproducts and ∐ \coprod ∐ for arbitrary coproducts. Modules over ordinary ringoids are right modules.
The main construction
Outline
Let R R R be a nonzero commutative ring. Fix an infinite cardinal θ \theta θ with cf ( θ ) > ℵ 3 \operatorname{cf}(\theta)>\aleph_3 cf ( θ ) > ℵ 3 . We write ω 3 \omega_3 ω 3 for the initial ordinal of cardinality ℵ 3 \aleph_3 ℵ 3 , so θ > ℵ 3 = ∣ ω 3 ∣ \theta>\aleph_3=|\omega_3| θ > ℵ 3 = ∣ ω 3 ∣ .
The construction of T J T_J T J 's. Let C \mathcal{C} C be the R R R -linearisation of the poset category ω 3 \omega_3 ω 3 , i.e., the objects are unchanged but Hom sets are R R R -enriched. Let R : C op → Mod R R:\mathcal{C}^{\operatorname{op}}\to\operatorname{Mod}_R R : C op → Mod R be the constant presheaf taking the value R R R . Subsection 3.2 takes the 4-th syzygy A A A of R R R
⋯ ⟶ P 4 ⟶ P 3 ⟶ P 2 ⟶ P 1 ⟶ P 0 ⟶ R . (2) \cdots\longrightarrow P_4\longrightarrow P_3\longrightarrow P_2\longrightarrow P_1\longrightarrow P_0\longrightarrow R. \tag*{(2)} ⋯ ⟶ P 4 ⟶ P 3 ⟶ P 2 ⟶ P 1 ⟶ P 0 ⟶ R . (2)
Define the bimodule E : C op ⊗ R C → Mod R E:\mathcal{C}^{\operatorname{op}}\otimes_R\mathcal{C}\to\operatorname{Mod}_R E : C op ⊗ R C → Mod R via E ( α , β ) = A ( α ) E(\alpha,\beta)=A(\alpha) E ( α , β ) = A ( α ) . Then c : P 4 ↠ A c:P_4\twoheadrightarrow A c : P 4 ↠ A induces 0 ≠ [ c ] ∈ Ext C 4 ( R , A ) 0\ne[c]\in\operatorname{Ext}^4_{\mathcal{C}}(R,A) 0 = [ c ] ∈ Ext C 4 ( R , A ) by Lemma 3.3 . Consider the graded category B \mathcal{B} B with
Ob ( B ) = Ob ( C ) = ω 3 , H o m B ( α , β ) d = { Hom C ( α , β ) , d = 0 , E ( α , β ) = A ( α ) , d = − 2 , 0 , otherwise . \operatorname{Ob}(\mathcal{B})=\operatorname{Ob}(\mathcal{C})=\omega_3,\qquad
\mathcal{H}om_{\mathcal{B}}(\alpha,\beta)^d=
\begin{cases}
\operatorname{Hom}_{\mathcal{C}}(\alpha,\beta), & d=0,\\
E(\alpha,\beta)=A(\alpha), & d=-2,\\
0, & \text{otherwise}.
\end{cases} Ob ( B ) = Ob ( C ) = ω 3 , H o m B ( α , β ) d = ⎩ ⎨ ⎧ Hom C ( α , β ) , E ( α , β ) = A ( α ) , 0 , d = 0 , d = − 2 , otherwise .
We identify H o m B ( α , β ) ≅ Hom C ( α , β ) ⊕ E ( α , β ) \mathcal{H}om_{\mathcal{B}}(\alpha,\beta)\cong\operatorname{Hom}_{\mathcal{C}}(\alpha,\beta)\oplus E(\alpha,\beta) H o m B ( α , β ) ≅ Hom C ( α , β ) ⊕ E ( α , β ) , and the composition becomes
( g , y ) ∘ ( f , x ) = ( g ∘ f , g ⋅ x + y ⋅ f ) . (g,y)\circ(f,x)=(g\circ f,g\cdot x+y\cdot f). ( g , y ) ∘ ( f , x ) = ( g ∘ f , g ⋅ x + y ⋅ f ) .
Here, for f : α → β f:\alpha\to\beta f : α → β and g : β → γ g:\beta\to\gamma g : β → γ , by definition one has
g ⋅ x = { s x , β ≤ γ , ( g = s u β , γ ) , 0 , β > γ , y ⋅ f = { r A ( u α , β ) ( y ) , α ≤ β , ( f = r u α , β ) , 0 , α > β . g\cdot x=
\begin{cases}
sx,&\beta\leq\gamma,\quad(g=su_{\beta,\gamma}),\\
0,&\beta>\gamma,
\end{cases}
\qquad
y\cdot f=
\begin{cases}
rA(u_{\alpha,\beta})(y),&\alpha\leq\beta,\quad(f=ru_{\alpha,\beta}),\\
0,&\alpha>\beta.
\end{cases} g ⋅ x = { s x , 0 , β ≤ γ , ( g = s u β , γ ) , β > γ , y ⋅ f = { r A ( u α , β ) ( y ) , 0 , α ≤ β , ( f = r u α , β ) , α > β .
Note that B \mathcal{B} B has usual A ∞ A_\infty A ∞ -structure, with m 2 m_2 m 2 the usual multiplication and m ≠ 2 = 0 m_{\ne2}=0 m = 2 = 0 , to obtain Λ − ; \Lambda_-; Λ − ; meanwhile Lemma 3.5 gives another A ∞ A_\infty A ∞ -structure on B \mathcal{B} B with m 2 m_2 m 2 the usual multiplication, m 4 = m 4 + m_4=m_4^+ m 4 = m 4 + , and all other operations zero, to obtain Λ + \Lambda_+ Λ + . For either b ∈ { + , − } b\in\{+,-\} b ∈ { + , − } , let D ∞ ( Λ b ) D_\infty(\Lambda_b) D ∞ ( Λ b ) denote the derived category by localising its right A ∞ A_\infty A ∞ -modules at quasi-isomorphisms. Put V b = thick D ∞ ( Λ b ) ( G b ) \mathcal{V}_b=\operatorname{thick}_{D_\infty(\Lambda_b)}(\mathcal{G}_b) V b = thick D ∞ ( Λ b ) ( G b ) , where G b \mathcal{G}_b G b is the coproduct of ∣ ω 3 ∣ |\omega_3| ∣ ω 3 ∣ copies of each representable functors Hom Λ b ( − , α ) \operatorname{Hom}_{\Lambda_b}(-,\alpha) Hom Λ b ( − , α ) ’s.
By Proposition 3.11 , Perf ( Λ − ) ≃ Perf ( Λ + ) \operatorname{Perf}(\Lambda_-)\simeq\operatorname{Perf}(\Lambda_+) Perf ( Λ − ) ≃ Perf ( Λ + ) ; however, the larger categories V ± \mathcal{V}_{\pm} V ± detect the difference between the two m 4 m_4 m 4 -operations: the fourfold Toda brackets associated with the same projective resolution determine the distinct classes 0 0 0 and − [ c ] -[c] − [ c ] in Ext C 4 ( R , A ) \operatorname{Ext}^4_{\mathcal{C}}(R,A) Ext C 4 ( R , A ) .
Fix I = θ × { + , − } I=\theta\times\{+,-\} I = θ × { + , − } with projection Sign : I → { + , − } \operatorname{Sign}:I\to\{+,-\} Sign : I → { + , − } . For finite subset J ⊆ I J\subseteq I J ⊆ I , take
T J = ( ∏ i ∈ J V Sign ( i ) ) × Perf ( Λ − ( I ∖ J ) ) , ∣ J ∣ < ℵ 0 \mathcal{T}_J=\left(\prod_{i\in J}\mathcal{V}_{\operatorname{Sign}(i)}\right)\times\operatorname{Perf}(\Lambda_-(I\setminus J)),\qquad|J|<\aleph_0 T J = ( i ∈ J ∏ V Sign ( i ) ) × Perf ( Λ − ( I ∖ J )) , ∣ J ∣ < ℵ 0
(see Equation (1)).
We shall specify a filtered system in Subsection 3.5. Take its colimit T = l i m → J ⊆ I , ∣ J ∣ < ℵ 0 T J \mathcal{T}=\varinjlim_{J\subseteq I,\ |J|<\aleph_0}\mathcal{T}_J T = lim J ⊆ I , ∣ J ∣ < ℵ 0 T J .
Each < θ <\theta < θ -generated category is algebraic. Let ( X s ) s ∈ S (X_s)_{s\in S} ( X s ) s ∈ S be a family of objects of T \mathcal{T} T with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ . Fix X s ∈ T J s X_s\in\mathcal{T}_{J_s} X s ∈ T J s for a finite subset J s ⊆ I J_s\subseteq I J s ⊆ I . Then J = ⋃ s ∈ S J s J=\bigcup_{s\in S}J_s J = ⋃ s ∈ S J s . Since ∣ J ∣ < θ |J|<\theta ∣ J ∣ < θ , Proposition 3.22 gives an R R R -linear dg enhancement.
However T \mathcal{T} T is not topological. Suppose that T \mathcal{T} T admits a topological realisation. Proposition 3.30 gives a Toda class t = [ u ] t=[u] t = [ u ] . For each i ∈ I i\in I i ∈ I , let t i t_i t i be the branch class obtained from t t t as in Lemma 3.33 , and χ \chi χ the characteristic map of Definition 3.26.
On one hand, Lemma 3.27 and Proposition 3.15 give
χ ( t i ) = { 0 , Sign ( i ) = − , − [ c ] , Sign ( i ) = + . \chi(t_i)=
\begin{cases}
0,&\operatorname{Sign}(i)=-,\\
-[c],&\operatorname{Sign}(i)=+.
\end{cases} χ ( t i ) = { 0 , − [ c ] , Sign ( i ) = − , Sign ( i ) = + .
On the other hand, ( χ ( t i ) ) i ∈ I ∈ ∏ i ∈ I Ext C 4 ( R , A ) (\chi(t_i))_{i\in I}\in\prod_{i\in I}\operatorname{Ext}^4_{\mathcal{C}}(R,A) ( χ ( t i ) ) i ∈ I ∈ ∏ i ∈ I Ext C 4 ( R , A ) is constant except for a subset S ⊆ I S\subseteq I S ⊆ I with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ (Proposition 3.28).
These lead a contradiction.
Faithful central action by R R R . The constant scalar functions act faithfully on every Perf ( B ) \operatorname{Perf}(\mathcal{B}) Perf ( B ) . Therefore R R R acts on Z ( T ) Z(\mathcal{T}) Z ( T ) faithfully.
Two distinct A ∞ A_{\infty} A ∞ structures over B \mathcal{B} B
Lemma 3.1. For each β < ω 3 \beta< \omega_{3} β < ω 3 , the following sequence is contractible
⋯ ⟶ ∏ β ≤ α 0 < α 1 < α 2 < ω 3 R → d 2 ∏ β ≤ α 0 < α 1 < ω 3 R → d 1 ∏ β ≤ α 0 < ω 3 R → d 0 R ⟶ 0. \cdots\longrightarrow\prod_{\beta\leq\alpha_{0}<\alpha_{1}<\alpha_{2}<\omega_{3}} R \xrightarrow{d_{2}} \prod_{\beta\leq\alpha_{0}<\alpha_{1}<\omega_{3}} R \xrightarrow{d_{1}} \prod_{\beta\leq\alpha_{0}<\omega_{3}} R \xrightarrow{d_{0}} R \longrightarrow0. ⋯ ⟶ β ≤ α 0 < α 1 < α 2 < ω 3 ∏ R d 2 β ≤ α 0 < α 1 < ω 3 ∏ R d 1 β ≤ α 0 < ω 3 ∏ R d 0 R ⟶ 0. Writing α 0 , … , α p \alpha_{0},\ldots,\alpha_{p} α 0 , … , α p as coordinate basis of the corresponding summand, the differentials are
d p [ α 0 , … , α p ] = ∑ j = 0 p ( − 1 ) j [ α 0 , … , α j ^ , … , α p ] ( p ≥ 1 ) , d 0 [ α 0 ] = 1. d_{p}[\alpha_{0},\ldots,\alpha_{p}]=\sum_{j=0}^{p}(-1)^{j}[\alpha_{0},\ldots,\widehat{\alpha_{j}},\ldots,\alpha_{p}]\quad(p\geq1),\qquad d_{0}[\alpha_{0}]=1. d p [ α 0 , … , α p ] = j = 0 ∑ p ( − 1 ) j [ α 0 , … , α j , … , α p ] ( p ≥ 1 ) , d 0 [ α 0 ] = 1. Proof. Define R R R -linear maps by h − 1 ( 1 ) = [ β ] h_{-1}(1)=[\beta] h − 1 ( 1 ) = [ β ] and
h p [ α 0 , … , α p ] = { [ β , α 0 , … , α p ] , β < α 0 , 0 , β = α 0 . h_{p}[\alpha_{0},\ldots,\alpha_{p}]=
\begin{cases}
[\beta,\alpha_{0},\ldots,\alpha_{p}], & \beta<\alpha_{0},\\
0, & \beta=\alpha_{0}.
\end{cases} h p [ α 0 , … , α p ] = { [ β , α 0 , … , α p ] , 0 , β < α 0 , β = α 0 . Then d p d p + 1 = 0 d_{p}d_{p+1}=0 d p d p + 1 = 0 , d 0 h − 1 = id R d_{0}h_{-1}=\operatorname{id}_{R} d 0 h − 1 = id R and d p + 1 h p + h p − 1 d p = id d_{p+1}h_{p}+h_{p-1}d_{p}=\operatorname{id} d p + 1 h p + h p − 1 d p = id for p ≥ 0 p\geq0 p ≥ 0 , proving the lemma. □
Then gives a resolution of R R R , whose pointwise exactness is checked in Lemma 3.1
⋯ ⟶ P 4 → d 4 P 3 → d 3 P 2 → d 2 P 1 → d 1 P 0 → d 0 R ⟶ 0 , P p = ∏ α 0 < ⋯ < α p < ω 3 y α 0 . \cdots\longrightarrow P_{4}\xrightarrow{d_{4}}P_{3}\xrightarrow{d_{3}}P_{2}\xrightarrow{d_{2}}P_{1}\xrightarrow{d_{1}}P_{0}\xrightarrow{d_{0}}R\longrightarrow0,\qquad P_{p}=\prod_{\alpha_{0}<\cdots<\alpha_{p}<\omega_{3}}y_{\alpha_{0}}. ⋯ ⟶ P 4 d 4 P 3 d 3 P 2 d 2 P 1 d 1 P 0 d 0 R ⟶ 0 , P p = α 0 < ⋯ < α p < ω 3 ∏ y α 0 .
Put A = im ( d 4 : P 4 → P 3 ) A=\operatorname{im}(d_{4}:P_{4}\to P_{3}) A = im ( d 4 : P 4 → P 3 ) and let c : P 4 → A c:P_{4}\to A c : P 4 → A be the corestriction of d 4 d_{4} d 4 .
Lemma 3.2. The module R R R is flat and pd C R = 4 \operatorname{pd}_{\mathcal{C}}R=4 pd C R = 4 .
Proof. Since R = l i m → β < ω 3 y β R=\varinjlim_{\beta<\omega_{3}}y_{\beta} R = lim β < ω 3 y β is a filtered colimit of projective modules, it is flat.
Since ℵ 3 \aleph_{3} ℵ 3 is regular, ω 3 \omega_{3} ω 3 has cofinality ℵ 3 \aleph_{3} ℵ 3 . By Mitchell’s theorem [14 Theorem A] , ω 3 \omega_{3} ω 3 has R R R -cohomological dimension 4, i.e., l i m ← p N \varprojlim^{p}N lim p N vanishes for any inverse system N : ( ω 3 ) op → Mod R N:(\omega_{3})^{\operatorname{op}}\to\operatorname{Mod}_{R} N : ( ω 3 ) op → Mod R precisely when p > 4 p>4 p > 4 . Since limits are right adjoint to diagonal functor, Hom C ( R , − ) ≅ l i m ← α < ω 3 ( − ) ( α ) \operatorname{Hom}_{\mathcal{C}}(R,-)\cong\varprojlim_{\alpha<\omega_{3}}(-)(\alpha) Hom C ( R , − ) ≅ lim α < ω 3 ( − ) ( α ) . This yields exactly pd C R = 4 \operatorname{pd}_{\mathcal{C}}R=4 pd C R = 4 . □
Lemma 3.3. The cocycle c : P 4 → A c:P_{4}\to A c : P 4 → A represents a nonzero class in Ext C 4 ( R , A ) \operatorname{Ext}^{4}_{\mathcal{C}}(R,A) Ext C 4 ( R , A ) .
Proof. If [ c ] = 0 [c]=0 [ c ] = 0 , dimension shifting implies that 0 → A → P 3 → im d 3 → 0 0\to A\to P_{3}\to\operatorname{im}d_{3}\to0 0 → A → P 3 → im d 3 → 0 splits. Thus im d 3 \operatorname{im}d_{3} im d 3 is projective, contradicting pd C R = 4 \operatorname{pd}_{\mathcal{C}}R=4 pd C R = 4 .
We construct two distinct A ∞ A_{\infty} A ∞ -structures over the graded category B \mathcal{B} B as defined in the outline. The first A ∞ A_{\infty} A ∞ category Λ − \Lambda_{-} Λ − has underlying graded category B \mathcal{B} B , m 2 − m_{2}^{-} m 2 − the usual composition and m i ≠ 2 − = 0 m_{i\ne2}^{-}=0 m i = 2 − = 0 .
Definition 3.4. Let Λ + \Lambda_{+} Λ + be an A ∞ A_{\infty} A ∞ category with underlying graded category B \mathcal{B} B , m 2 + m_{2}^{+} m 2 + the usual multiplication, m i + = 0 m_{i}^{+}=0 m i + = 0 for i ∉ { 2 , 4 } i\notin\{2,4\} i ∈ / { 2 , 4 } , and m 4 + m_{4}^{+} m 4 + is defined as follows:
For α 0 < α 1 < α 2 < α 3 < α 4 < ω 3 \alpha_{0}<\alpha_{1}<\alpha_{2}<\alpha_{3}<\alpha_{4}<\omega_{3} α 0 < α 1 < α 2 < α 3 < α 4 < ω 3 , define
m 4 + ( u α 3 , α 4 , u α 2 , α 3 , u α 1 , α 2 , u α 0 , α 1 ) = c α 0 [ α 0 , α 1 , α 2 , α 3 , α 4 ] , m_{4}^{+}(u_{\alpha_{3},\alpha_{4}},u_{\alpha_{2},\alpha_{3}},u_{\alpha_{1},\alpha_{2}},u_{\alpha_{0},\alpha_{1}})=c_{\alpha_{0}}[\alpha_{0},\alpha_{1},\alpha_{2},\alpha_{3},\alpha_{4}], m 4 + ( u α 3 , α 4 , u α 2 , α 3 , u α 1 , α 2 , u α 0 , α 1 ) = c α 0 [ α 0 , α 1 , α 2 , α 3 , α 4 ] , where the result belongs to A ( α 0 ) = E ( α 0 , α 4 ) = H o m B ( α 0 , α 4 ) − 2 A(\alpha_{0})=E(\alpha_{0},\alpha_{4})=\mathcal{H}om_{\mathcal{B}}(\alpha_{0},\alpha_{4})^{-2} A ( α 0 ) = E ( α 0 , α 4 ) = H o m B ( α 0 , α 4 ) − 2 .
Set m 4 + ( f 1 , f 2 , f 3 , f 4 ) = 0 m_{4}^{+}(f_{1},f_{2},f_{3},f_{4})=0 m 4 + ( f 1 , f 2 , f 3 , f 4 ) = 0 if one of the morphisms is 1 1 1 , or has degree − 2 -2 − 2 .
Extend m 4 + m_{4}^{+} m 4 + to make it linear in 4 arguments.
Lemma 3.5. The categories Λ − \Lambda_{-} Λ − and Λ + \Lambda_{+} Λ + are strictly unital minimal R R R -linear A ∞ A_{\infty} A ∞ categories.
Proof. For Λ + \Lambda_{+} Λ + , we check Stasheff identities for Λ + \Lambda_{+} Λ + , where the identity for arity N N N is
∑ r + s + t = N ( − 1 ) r + s t m r + 1 + t + ( id ⊗ r ⊗ m s + ⊗ id ⊗ t ) = 0. \sum_{r+s+t=N}(-1)^{r+st}m^{+}_{r+1+t}\left(\operatorname{id}^{\otimes r}\otimes m^{+}_{s}\otimes\operatorname{id}^{\otimes t}\right)=0. r + s + t = N ∑ ( − 1 ) r + s t m r + 1 + t + ( id ⊗ r ⊗ m s + ⊗ id ⊗ t ) = 0.
It suffices to check the case r + 1 + t , s ∈ { 2 , 4 } r+1+t,s\in\{2,4\} r + 1 + t , s ∈ { 2 , 4 } , where N ∈ { 3 , 5 , 7 } N\in\{3,5,7\} N ∈ { 3 , 5 , 7 } .
( N = 3 ) (N=3) ( N = 3 ) Associativity yields m 2 + ( m 2 + ( f 3 , f 2 ) , f 1 ) − m 2 + ( f 3 , m 2 + ( f 2 , f 1 ) ) = 0 m^{+}_{2}(m^{+}_{2}(f_{3},f_{2}),f_{1})-m^{+}_{2}(f_{3},m^{+}_{2}(f_{2},f_{1}))=0 m 2 + ( m 2 + ( f 3 , f 2 ) , f 1 ) − m 2 + ( f 3 , m 2 + ( f 2 , f 1 )) = 0 .
( N = 5 ) (N=5) ( N = 5 ) Let L \mathcal{L} L denote the left hand side of the identity, i.e.,
L = m 2 + ( m 4 + ( f 5 , f 4 , f 3 , f 2 ) , f 1 ) − m 2 + ( f 5 , m 4 + ( f 4 , f 3 , f 2 , f 1 ) ) + m 4 + ( m 2 + ( f 5 , f 4 ) , f 3 , f 2 , f 1 ) − m 4 + ( f 5 , m 2 + ( f 4 , f 3 ) , f 2 , f 1 ) + m 4 + ( f 5 , f 4 , m 2 + ( f 3 , f 2 ) , f 1 ) − m 4 + ( f 5 , f 4 , f 3 , m 2 + ( f 2 , f 1 ) ) . \begin{aligned}
\mathcal{L}={}&m^{+}_{2}(m^{+}_{4}(f_{5},f_{4},f_{3},f_{2}),f_{1})-m^{+}_{2}(f_{5},m^{+}_{4}(f_{4},f_{3},f_{2},f_{1}))\\
&+m^{+}_{4}(m^{+}_{2}(f_{5},f_{4}),f_{3},f_{2},f_{1})-m^{+}_{4}(f_{5},m^{+}_{2}(f_{4},f_{3}),f_{2},f_{1})\\
&+m^{+}_{4}(f_{5},f_{4},m^{+}_{2}(f_{3},f_{2}),f_{1})-m^{+}_{4}(f_{5},f_{4},f_{3},m^{+}_{2}(f_{2},f_{1})).
\end{aligned} L = m 2 + ( m 4 + ( f 5 , f 4 , f 3 , f 2 ) , f 1 ) − m 2 + ( f 5 , m 4 + ( f 4 , f 3 , f 2 , f 1 )) + m 4 + ( m 2 + ( f 5 , f 4 ) , f 3 , f 2 , f 1 ) − m 4 + ( f 5 , m 2 + ( f 4 , f 3 ) , f 2 , f 1 ) + m 4 + ( f 5 , f 4 , m 2 + ( f 3 , f 2 ) , f 1 ) − m 4 + ( f 5 , f 4 , f 3 , m 2 + ( f 2 , f 1 )) .
We first show the case f j = u α j − 1 α j f_j=u_{\alpha_{j-1}\alpha_j} f j = u α j − 1 α j with α 0 < ⋯ < α 5 < ω 3 \alpha_0<\cdots<\alpha_5<\omega_3 α 0 < ⋯ < α 5 < ω 3 :
L = A ( u α 0 , α 1 ) ( c α 1 [ α 1 , … , α 5 ] ) − c α 0 [ α 0 , … , α 4 ] + c α 0 [ α 0 , α 1 , α 2 , α 3 , α 5 ] − c α 0 [ α 0 , α 1 , α 2 , α 4 , α 5 ] + c α 0 [ α 0 , α 1 , α 3 , α 4 , α 5 ] − c α 0 [ α 0 , α 2 , α 3 , α 4 , α 5 ] = c α 0 ( ∑ j = 0 5 ( − 1 ) j [ α 0 , … , α j ^ , … , α 5 ] ) = ( c ∘ d 5 ) α 0 [ α 0 , … , α 5 ] = 0. \begin{aligned}
\mathcal{L}={}&A(u_{\alpha_0,\alpha_1})\left(c_{\alpha_1}[\alpha_1,\ldots,\alpha_5]\right)-c_{\alpha_0}[\alpha_0,\ldots,\alpha_4]+c_{\alpha_0}[\alpha_0,\alpha_1,\alpha_2,\alpha_3,\alpha_5]\\
&-c_{\alpha_0}[\alpha_0,\alpha_1,\alpha_2,\alpha_4,\alpha_5]+c_{\alpha_0}[\alpha_0,\alpha_1,\alpha_3,\alpha_4,\alpha_5]-c_{\alpha_0}[\alpha_0,\alpha_2,\alpha_3,\alpha_4,\alpha_5]\\
={}&c_{\alpha_0}\left(\sum_{j=0}^{5}(-1)^j[\alpha_0,\ldots,\widehat{\alpha_j},\ldots,\alpha_5]\right)=(c\circ d_5)_{\alpha_0}[\alpha_0,\ldots,\alpha_5]=0.
\end{aligned} L = = A ( u α 0 , α 1 ) ( c α 1 [ α 1 , … , α 5 ] ) − c α 0 [ α 0 , … , α 4 ] + c α 0 [ α 0 , α 1 , α 2 , α 3 , α 5 ] − c α 0 [ α 0 , α 1 , α 2 , α 4 , α 5 ] + c α 0 [ α 0 , α 1 , α 3 , α 4 , α 5 ] − c α 0 [ α 0 , α 2 , α 3 , α 4 , α 5 ] c α 0 ( j = 0 ∑ 5 ( − 1 ) j [ α 0 , … , α j , … , α 5 ] ) = ( c ∘ d 5 ) α 0 [ α 0 , … , α 5 ] = 0.
If f i = 1 f_i=1 f i = 1 or f i f_i f i has degree − 2 -2 − 2 for some f i f_i f i , then is easy to check that L = 0 \mathcal{L}=0 L = 0 .
( N = 7 ) (N=7) ( N = 7 ) The identity becomes
0 = m 4 + ( m 4 + ( f 7 , f 6 , f 5 , f 4 ) , f 3 , f 2 , f 1 ) − m 4 + ( f 7 , m 4 + ( f 6 , f 5 , f 4 , f 3 ) , f 2 , f 1 ) + m 4 + ( f 7 , f 6 , m 4 + ( f 5 , f 4 , f 3 , f 2 ) , f 1 ) − m 4 + ( f 7 , f 6 , f 5 , m 4 + ( f 4 , f 3 , f 2 , f 1 ) ) . \begin{aligned}
0={}&m^{+}_{4}(m^{+}_{4}(f_{7},f_{6},f_{5},f_{4}),f_{3},f_{2},f_{1})-m^{+}_{4}(f_{7},m^{+}_{4}(f_{6},f_{5},f_{4},f_{3}),f_{2},f_{1})\\
&+m^{+}_{4}(f_{7},f_{6},m^{+}_{4}(f_{5},f_{4},f_{3},f_{2}),f_{1})-m^{+}_{4}(f_{7},f_{6},f_{5},m^{+}_{4}(f_{4},f_{3},f_{2},f_{1})).
\end{aligned} 0 = m 4 + ( m 4 + ( f 7 , f 6 , f 5 , f 4 ) , f 3 , f 2 , f 1 ) − m 4 + ( f 7 , m 4 + ( f 6 , f 5 , f 4 , f 3 ) , f 2 , f 1 ) + m 4 + ( f 7 , f 6 , m 4 + ( f 5 , f 4 , f 3 , f 2 ) , f 1 ) − m 4 + ( f 7 , f 6 , f 5 , m 4 + ( f 4 , f 3 , f 2 , f 1 )) .
Since the output of m 4 + m^{+}_{4} m 4 + has degree − 2 -2 − 2 , the right hand side vanishes.
Note that m 1 = 0 m_1=0 m 1 = 0 gives minimality. To see the strict units, it is clear to verify
m 1 + ( 1 ) = 0 , m 2 + ( 1 , f ) = f = m 2 + ( f , 1 ) , m ≥ 3 + ( … , 1 , … ) = 0. m^{+}_{1}(1)=0,\qquad m^{+}_{2}(1,f)=f=m^{+}_{2}(f,1),\qquad m^{+}_{\geq3}(\ldots,1,\ldots)=0. m 1 + ( 1 ) = 0 , m 2 + ( 1 , f ) = f = m 2 + ( f , 1 ) , m ≥ 3 + ( … , 1 , … ) = 0.
For Λ − \Lambda_{-} Λ − the verification is much easier. □
Equivalence of the perfect categories
For a subset Γ ⊂ ω 3 \Gamma\subset\omega_3 Γ ⊂ ω 3 , let C Γ C_{\Gamma} C Γ be the full subcategory of C C C on Γ \Gamma Γ , and let E Γ E_{\Gamma} E Γ be the restriction of E E E to C Γ op ⊗ R C Γ C_{\Gamma}^{\operatorname{op}}\otimes_R C_{\Gamma} C Γ op ⊗ R C Γ .
Based on the distinct A ∞ A_{\infty} A ∞ -structures Λ ± \Lambda_{\pm} Λ ± , we use twisted complexes show the equivalence ( H 0 Tw ( Λ + ∣ Γ ) ) Γ ≅ ( H 0 Tw ( Λ − ∣ Γ ) ) Γ (H^0\operatorname{Tw}(\Lambda_{+}|_{\Gamma}))_{\Gamma}\cong(H^0\operatorname{Tw}(\Lambda_{-}|_{\Gamma}))_{\Gamma} ( H 0 Tw ( Λ + ∣ Γ ) ) Γ ≅ ( H 0 Tw ( Λ − ∣ Γ ) ) Γ for each finite subset Γ ⊆ ω 3 \Gamma\subseteq\omega_3 Γ ⊆ ω 3 . The we show these equivalences are compatible with inclusions of finite subsets, and “glue” these equivalences to H 0 Tw ( Λ + ) ≃ H 0 Tw ( Λ − ) H^0\operatorname{Tw}(\Lambda_{+})\simeq H^0\operatorname{Tw}(\Lambda_{-}) H 0 Tw ( Λ + ) ≃ H 0 Tw ( Λ − ) . Finally idempotent completion gives Perf ( Λ + ) ≃ Perf ( Λ − ) \operatorname{Perf}(\Lambda_{+})\simeq\operatorname{Perf}(\Lambda_{-}) Perf ( Λ + ) ≃ Perf ( Λ − ) .
Lemma 3.6. For every nonempty finite set Γ ⊂ ω 3 \Gamma\subset\omega_3 Γ ⊂ ω 3 , H H R p ( C Γ , E Γ ) = 0 HH^p_R(C_{\Gamma},E_{\Gamma})=0 H H R p ( C Γ , E Γ ) = 0 for p > 0 p>0 p > 0 .
Proof. By definition, H H R p ( C Γ , E Γ ) = H p ( C R p ( C Γ , E Γ ) , δ ) HH^p_R(C_{\Gamma},E_{\Gamma})=H^p(C^p_R(C_{\Gamma},E_{\Gamma}),\delta) H H R p ( C Γ , E Γ ) = H p ( C R p ( C Γ , E Γ ) , δ ) , and
C R p ( C Γ , E Γ ) ≅ ∏ α 0 ≤ ⋯ ≤ α p α i ∈ Γ Hom R ( R ⊗ R p , A ( α 0 ) ) ≅ ∏ α 0 ≤ ⋯ ≤ α p α i ∈ Γ A ( α 0 ) . C^p_R(C_{\Gamma},E_{\Gamma})\cong\prod_{\substack{\alpha_0\leq\cdots\leq\alpha_p\\\alpha_i\in\Gamma}}\operatorname{Hom}_R(R^{\otimes Rp},A(\alpha_0))\cong\prod_{\substack{\alpha_0\leq\cdots\leq\alpha_p\\\alpha_i\in\Gamma}}A(\alpha_0). C R p ( C Γ , E Γ ) ≅ α 0 ≤ ⋯ ≤ α p α i ∈ Γ ∏ Hom R ( R ⊗ R p , A ( α 0 )) ≅ α 0 ≤ ⋯ ≤ α p α i ∈ Γ ∏ A ( α 0 ) . Take β = sup Γ \beta=\sup\Gamma β = sup Γ . Then β < ω 3 \beta<\omega_3 β < ω 3 . For each p ≥ 1 p\geq1 p ≥ 1 , put
h : C R p ( C Γ , E Γ ) ⟶ C R p − 1 ( C Γ , E Γ ) , ( h f ) α 0 , … , α p − 1 = f α 0 , … , α p − 1 , β . h:C^p_R(C_{\Gamma},E_{\Gamma})\longrightarrow C^{p-1}_R(C_{\Gamma},E_{\Gamma}),\qquad(hf)_{\alpha_0,\ldots,\alpha_{p-1}}=f_{\alpha_0,\ldots,\alpha_{p-1},\beta}. h : C R p ( C Γ , E Γ ) ⟶ C R p − 1 ( C Γ , E Γ ) , ( h f ) α 0 , … , α p − 1 = f α 0 , … , α p − 1 , β .
Then by construction one has δ h + h δ = 1 \delta h + h\delta= 1 δ h + h δ = 1 , since
( δ h f ) α 0 , … , α p = f α 0 , … , α p − 1 , β + ∑ j = 1 p − 1 ( − 1 ) p − j f α 0 , … , α j ^ , … , α p , β + ( − 1 ) p A ( u α 0 , α 1 ) f α 1 , … , α p , β , (\delta h f)_{\alpha_0,\ldots,\alpha_p}=f_{\alpha_0,\ldots,\alpha_{p-1},\beta}+\sum_{j=1}^{p-1}(-1)^{p-j}f_{\alpha_0,\ldots,\widehat{\alpha_j},\ldots,\alpha_p,\beta}+(-1)^p A(u_{\alpha_0,\alpha_1})f_{\alpha_1,\ldots,\alpha_p,\beta}, ( δ h f ) α 0 , … , α p = f α 0 , … , α p − 1 , β + j = 1 ∑ p − 1 ( − 1 ) p − j f α 0 , … , α j , … , α p , β + ( − 1 ) p A ( u α 0 , α 1 ) f α 1 , … , α p , β ,
( h δ f ) α 0 , … , α p = f α 0 , … , α p − f α 0 , … , α p − 1 , β + ∑ j = 1 p − 1 ( − 1 ) p + 1 − j f α 0 , … , α j ^ , … , α p , β + ( − 1 ) p + 1 A ( u α 0 , α 1 ) f α 1 , … , α p , β . (h\delta f)_{\alpha_0,\ldots,\alpha_p}=f_{\alpha_0,\ldots,\alpha_p}-f_{\alpha_0,\ldots,\alpha_{p-1},\beta}+\sum_{j=1}^{p-1}(-1)^{p+1-j}f_{\alpha_0,\ldots,\widehat{\alpha_j},\ldots,\alpha_p,\beta}+(-1)^{p+1}A(u_{\alpha_0,\alpha_1})f_{\alpha_1,\ldots,\alpha_p,\beta}. ( h δ f ) α 0 , … , α p = f α 0 , … , α p − f α 0 , … , α p − 1 , β + j = 1 ∑ p − 1 ( − 1 ) p + 1 − j f α 0 , … , α j , … , α p , β + ( − 1 ) p + 1 A ( u α 0 , α 1 ) f α 1 , … , α p , β .
By setwise homotopy [2 ] , C R p ( C Γ , E Γ ) C_{\mathrm{R}}^p(\mathcal{C}_{\Gamma},E_{\Gamma}) C R p ( C Γ , E Γ ) is acyclic, thus HH R p ( C Γ , E Γ ) = 0 \operatorname{HH}_{\mathrm{R}}^p(\mathcal{C}_{\Gamma},E_{\Gamma})=0 HH R p ( C Γ , E Γ ) = 0 . □
Hence m 4 + ∣ Γ = δ h Γ m_4^+|_{\Gamma}=\delta h_{\Gamma} m 4 + ∣ Γ = δ h Γ for some h Γ ∈ C R 3 ( C Γ , E Γ ) h_{\Gamma}\in C_{\mathrm{R}}^3(\mathcal{C}_{\Gamma},E_{\Gamma}) h Γ ∈ C R 3 ( C Γ , E Γ ) . Fix such h Γ h_{\Gamma} h Γ , and denote Λ ± ∣ Γ \Lambda_{\pm}|_{\Gamma} Λ ± ∣ Γ as the full subcategories on Γ \Gamma Γ . Then h Γ h_{\Gamma} h Γ gives an A ∞ A_{\infty} A ∞ -functor f Γ : Λ + ∣ Γ → Λ − ∣ Γ f_{\Gamma}:\Lambda_+|_{\Gamma}\to\Lambda_-|_{\Gamma} f Γ : Λ + ∣ Γ → Λ − ∣ Γ which satisfies
f Γ f_{\Gamma} f Γ is identical on objects, and ( f Γ ) 1 = id (f_{\Gamma})_1=\operatorname{id} ( f Γ ) 1 = id , i.e., f Γ f_{\Gamma} f Γ is identical on morphisms;
( f Γ ) 3 = h Γ (f_{\Gamma})_3=h_{\Gamma} ( f Γ ) 3 = h Γ , and ( f Γ ) i = 0 (f_{\Gamma})_i=0 ( f Γ ) i = 0 for i ∉ { 1 , 3 } i\notin\{1,3\} i ∈ / { 1 , 3 } .
Here we extend h Γ h_{\Gamma} h Γ to Λ + ∣ Γ \Lambda_+|_{\Gamma} Λ + ∣ Γ by sending the input in E E E to $0.
Lemma 3.7. For any nonempty finite subset Γ ⊂ ω 3 \Gamma\subset\omega_3 Γ ⊂ ω 3 , f Γ : Λ + ∣ Γ → Λ − ∣ Γ f_{\Gamma}:\Lambda_+|_{\Gamma}\to\Lambda_-|_{\Gamma} f Γ : Λ + ∣ Γ → Λ − ∣ Γ is an A ∞ A_{\infty} A ∞ functor.
Proof. Write f Γ = f f_{\Gamma}=f f Γ = f , h = h Γ h=h_{\Gamma} h = h Γ and m i ± ∣ Γ = m i ± m_i^{\pm}|_{\Gamma}=m_i^{\pm} m i ± ∣ Γ = m i ± for convenience in this proof. Since m i − = 0 m_i^-=0 m i − = 0 for i ≠ 2 i\ne2 i = 2 , and ( f Γ ) i = 0 (f_{\Gamma})_i=0 ( f Γ ) i = 0 for i ∉ { 1 , 3 } i\notin\{1,3\} i ∈ / { 1 , 3 } , it suffices to check the defining equalities with N N N arguments, where N ∈ { 2 , 4 , 6 } N\in\{2,4,6\} N ∈ { 2 , 4 , 6 } .
( N = 2 ) (N=2) ( N = 2 ) The identity is f 1 m 2 + = m 2 − ( f 1 ⊗ f 1 ) f_1m_2^+=m_2^-(f_1\otimes f_1) f 1 m 2 + = m 2 − ( f 1 ⊗ f 1 ) , which follows from m 2 + = m 2 − m_2^+=m_2^- m 2 + = m 2 − .
( N = 4 ) (N=4) ( N = 4 ) The identity is
f 1 m 4 + + f 3 ( m 2 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 2 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 2 + ) = m 2 − ( f 1 ⊗ f 3 ) + m 2 − ( f 3 ⊗ f 1 ) . f_1m_4^+ + f_3(m_2^+\otimes1\otimes1)-f_3(1\otimes m_2^+\otimes1)+f_3(1\otimes1\otimes m_2^+) = m_2^-(f_1\otimes f_3)+m_2^-(f_3\otimes f_1). f 1 m 4 + + f 3 ( m 2 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 2 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 2 + ) = m 2 − ( f 1 ⊗ f 3 ) + m 2 − ( f 3 ⊗ f 1 ) . Use f 1 = id f_1=\operatorname{id} f 1 = id , f 3 = h Γ f_3=h_{\Gamma} f 3 = h Γ , m 4 + = δ h Γ m_4^+=\delta h_{\Gamma} m 4 + = δ h Γ and m 2 − = m 2 + m_2^-=m_2^+ m 2 − = m 2 + . Then
f 1 m 4 + + f 3 ( m 2 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 2 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 2 + ) = δ h Γ + h Γ ( m 2 ⊗ 1 ⊗ 1 ) − h Γ ( 1 ⊗ m 2 ⊗ 1 ) + h Γ ( 1 ⊗ 1 ⊗ m 2 ) = m 2 ( 1 ⊗ h Γ ) − h Γ ( m 2 ⊗ 1 ⊗ 1 ) + h Γ ( 1 ⊗ m 2 ⊗ 1 ) − h Γ ( 1 ⊗ 1 ⊗ m 2 ) + m 2 ( h Γ ⊗ 1 ) + h Γ ( m 2 ⊗ 1 ⊗ 1 ) − h Γ ( 1 ⊗ m 2 ⊗ 1 ) + h Γ ( 1 ⊗ 1 ⊗ m 2 ) = m 2 ( 1 ⊗ h Γ ) + m 2 ( h Γ ⊗ 1 ) = m 2 − ( f 1 ⊗ f 3 ) + m 2 − ( f 3 ⊗ f 1 ) . \begin{aligned}
f_1m_4^+ + f_3(m_2^+\otimes1\otimes1)-f_3(1\otimes m_2^+\otimes1)+f_3(1\otimes1\otimes m_2^+)
&=\delta h_{\Gamma}+h_{\Gamma}(m_2\otimes1\otimes1)-h_{\Gamma}(1\otimes m_2\otimes1)+h_{\Gamma}(1\otimes1\otimes m_2)\\
&=m_2(1\otimes h_{\Gamma})-h_{\Gamma}(m_2\otimes1\otimes1)+h_{\Gamma}(1\otimes m_2\otimes1)-h_{\Gamma}(1\otimes1\otimes m_2)\\
&\quad+m_2(h_{\Gamma}\otimes1)+h_{\Gamma}(m_2\otimes1\otimes1)-h_{\Gamma}(1\otimes m_2\otimes1)+h_{\Gamma}(1\otimes1\otimes m_2)\\
&=m_2(1\otimes h_{\Gamma})+m_2(h_{\Gamma}\otimes1)=m_2^-(f_1\otimes f_3)+m_2^-(f_3\otimes f_1).
\end{aligned} f 1 m 4 + + f 3 ( m 2 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 2 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 2 + ) = δ h Γ + h Γ ( m 2 ⊗ 1 ⊗ 1 ) − h Γ ( 1 ⊗ m 2 ⊗ 1 ) + h Γ ( 1 ⊗ 1 ⊗ m 2 ) = m 2 ( 1 ⊗ h Γ ) − h Γ ( m 2 ⊗ 1 ⊗ 1 ) + h Γ ( 1 ⊗ m 2 ⊗ 1 ) − h Γ ( 1 ⊗ 1 ⊗ m 2 ) + m 2 ( h Γ ⊗ 1 ) + h Γ ( m 2 ⊗ 1 ⊗ 1 ) − h Γ ( 1 ⊗ m 2 ⊗ 1 ) + h Γ ( 1 ⊗ 1 ⊗ m 2 ) = m 2 ( 1 ⊗ h Γ ) + m 2 ( h Γ ⊗ 1 ) = m 2 − ( f 1 ⊗ f 3 ) + m 2 − ( f 3 ⊗ f 1 ) . ( N = 6 ) (N=6) ( N = 6 ) The identity is
f 3 ( m 4 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 4 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 4 + ) = m 2 − ( f 3 ⊗ f 3 ) . f_3(m_4^+\otimes1\otimes1)-f_3(1\otimes m_4^+\otimes1)+f_3(1\otimes1\otimes m_4^+)=m_2^-(f_3\otimes f_3). f 3 ( m 4 + ⊗ 1 ⊗ 1 ) − f 3 ( 1 ⊗ m 4 + ⊗ 1 ) + f 3 ( 1 ⊗ 1 ⊗ m 4 + ) = m 2 − ( f 3 ⊗ f 3 ) . Since both m 4 + m_4^+ m 4 + and f 3 f_3 f 3 take values in E E E and vanish for the inputs from E E E , the left hand side vanishes; by E 2 = 0 E^2=0 E 2 = 0 the right hand side also vanishes. □
Lemma 3.8. For every nonempty finite Γ ⊂ ω 3 \Gamma\subset\omega_3 Γ ⊂ ω 3 , the A ∞ A_{\infty} A ∞ -functor f Γ : Λ + ∣ Γ → Λ − ∣ Γ f_{\Gamma}:\Lambda_+|_{\Gamma}\to\Lambda_-|_{\Gamma} f Γ : Λ + ∣ Γ → Λ − ∣ Γ constructed above induces a triangle equivalence
F Γ : = H 0 Tw ( f Γ ) : H 0 Tw ( Λ + ∣ Γ ) → ∼ H 0 Tw ( Λ − ∣ Γ ) , F_{\Gamma}:=H^0\operatorname{Tw}(f_{\Gamma}):H^0\operatorname{Tw}(\Lambda_+|_{\Gamma})\xrightarrow{\sim}H^0\operatorname{Tw}(\Lambda_-|_{\Gamma}), F Γ := H 0 Tw ( f Γ ) : H 0 Tw ( Λ + ∣ Γ ) ∼ H 0 Tw ( Λ − ∣ Γ ) , which fixes the one-term objects and their graded morphism groups, i.e.,
F Γ ( α , 0 ) = ( α , 0 ) , Hom H 0 Tw ( Λ + ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) → ∼ Hom H 0 Tw ( Λ − ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) . F_{\Gamma}(\alpha,0)=(\alpha,0),\qquad\operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_+|_{\Gamma})}((\alpha,0),(\beta,0)[d])\xrightarrow{\sim}\operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_-|_{\Gamma})}((\alpha,0),(\beta,0)[d]). F Γ ( α , 0 ) = ( α , 0 ) , Hom H 0 Tw ( Λ + ∣ Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) ∼ Hom H 0 Tw ( Λ − ∣ Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) . Proof. Proposition 2.13 gives the triangle functor F Γ = H 0 Tw ( f Γ ) F_{\Gamma}=H^0\operatorname{Tw}(f_{\Gamma}) F Γ = H 0 Tw ( f Γ ) . Since f Γ f_{\Gamma} f Γ is the identity on objects, F Γ ( α , 0 ) = ( f Γ ( α ) , 0 ) = ( α , 0 ) F_{\Gamma}(\alpha,0)=(f_{\Gamma}(\alpha),0)=(\alpha,0) F Γ ( α , 0 ) = ( f Γ ( α ) , 0 ) = ( α , 0 ) . For α , β ∈ Γ \alpha,\beta\in\Gamma α , β ∈ Γ and d ∈ Z d\in\mathbb{Z} d ∈ Z ,
Hom H 0 Tw ( Λ ± ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) ≅ H d ( Hom Λ ± ( α , β ) , m 1 ± ) = Hom B ( α , β ) d . \operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_{\pm}|_{\Gamma})}((\alpha,0),(\beta,0)[d])\cong H^d(\operatorname{Hom}_{\Lambda_{\pm}}(\alpha,\beta),m_1^{\pm})=\operatorname{Hom}_{\mathcal{B}}(\alpha,\beta)^d. Hom H 0 Tw ( Λ ± ∣ Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) ≅ H d ( Hom Λ ± ( α , β ) , m 1 ± ) = Hom B ( α , β ) d . The first isomorphism follows from the definition of twisted complexes, and the equality follows from Lemma 2.5. Under these identifications, for the objects ( α , 0 ) (\alpha,0) ( α , 0 ) , ( β , 0 ) (\beta,0) ( β , 0 ) and a ∈ Hom B ( α , β ) d a\in\operatorname{Hom}_{\mathcal{B}}(\alpha,\beta)^d a ∈ Hom B ( α , β ) d , the formula defining Tw ( f Γ ) \operatorname{Tw}(f_{\Gamma}) Tw ( f Γ ) gives ( f Γ ) 1 ( a ) = ( f Γ ) 1 ( a ) = a (f_{\Gamma})_1(a)=(f_{\Gamma})_1(a)=a ( f Γ ) 1 ( a ) = ( f Γ ) 1 ( a ) = a . Thus F Γ F_{\Gamma} F Γ induces the identity after identifying both sides with Hom B ( α , β ) d \operatorname{Hom}_{\mathcal{B}}(\alpha,\beta)^d Hom B ( α , β ) d
Hom H 0 Tw ( Λ + ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) ⟶ Hom H 0 Tw ( Λ − ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) , [ a ] ⟼ [ a ] . \operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_+|_{\Gamma})}((\alpha,0),(\beta,0)[d])\longrightarrow\operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_-|_{\Gamma})}((\alpha,0),(\beta,0)[d]),\qquad[a]\longmapsto[a]. Hom H 0 Tw ( Λ + ∣ Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) ⟶ Hom H 0 Tw ( Λ − ∣ Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) , [ a ] ⟼ [ a ] .
For arbitrary T , U ∈ H 0 Tw ( Λ + ∣ Γ ) T,U \in H^0\operatorname{Tw}(\Lambda_+\vert_\Gamma) T , U ∈ H 0 Tw ( Λ + ∣ Γ ) , consider the natural comparison maps
Hom H 0 Tw ( Λ + ∣ Γ ) ( T , U [ d ] ) ⟶ Hom H 0 Tw ( Λ − ∣ Γ ) ( F Γ ( T ) , F Γ ( U ) [ d ] ) . \operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_+\vert_\Gamma)}(T,U[d]) \longrightarrow\operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_-\vert_\Gamma)}(F_\Gamma(T),F_\Gamma(U)[d]). Hom H 0 Tw ( Λ + ∣ Γ ) ( T , U [ d ]) ⟶ Hom H 0 Tw ( Λ − ∣ Γ ) ( F Γ ( T ) , F Γ ( U ) [ d ]) .
The isomorphism holds for T = ( α , 0 ) T=(\alpha,0) T = ( α , 0 ) and U = ( β , 0 ) U=(\beta,0) U = ( β , 0 ) , and thus extends to finite twisted complexes by taking (de)suspensions and mapping cones (using five lemma).
Thus F Γ F_\Gamma F Γ is fully faithful. It is essentially surjective, thus a triangle equivalence. □
For every nonempty finite Γ ⊂ ω 3 \Gamma\subset\omega_3 Γ ⊂ ω 3 , fix the normalized cochains h Γ h_\Gamma h Γ used to define f Γ f_\Gamma f Γ .
Lemma 3.9. For any inclusion of finite subsets Γ ⊂ Γ ′ ( ⊂ ω 3 ) \Gamma\subset\Gamma'(\subset\omega_3) Γ ⊂ Γ ′ ( ⊂ ω 3 ) , let ι Γ , Γ ′ ± \iota^\pm_{\Gamma,\Gamma'} ι Γ , Γ ′ ± denote H 0 Tw ( Λ ± ∣ Γ ) → H 0 Tw ( Λ ± ∣ Γ ′ ) H^0\operatorname{Tw}(\Lambda_\pm\vert_\Gamma)\to H^0\operatorname{Tw}(\Lambda_\pm\vert_{\Gamma'}) H 0 Tw ( Λ ± ∣ Γ ) → H 0 Tw ( Λ ± ∣ Γ ′ ) . Then there is a natural isomorphism ϑ Γ , Γ ′ : ι Γ , Γ ′ − F Γ ⇒ F Γ ′ ι Γ , Γ ′ + \vartheta_{\Gamma,\Gamma'}:\iota^-_{\Gamma,\Gamma'}F_\Gamma\Rightarrow F_{\Gamma'}\iota^+_{\Gamma,\Gamma'} ϑ Γ , Γ ′ : ι Γ , Γ ′ − F Γ ⇒ F Γ ′ ι Γ , Γ ′ +
H 0 Tw ( Λ + ∣ Γ ) → F Γ H 0 Tw ( Λ − ∣ Γ ) ι Γ , Γ ′ + ↓ ↓ ι Γ , Γ ′ − H 0 Tw ( Λ + ∣ Γ ′ ) → F Γ ′ H 0 Tw ( Λ − ∣ Γ ′ ) \begin{CD}
H^0\operatorname{Tw}(\Lambda_+\vert_\Gamma) @>F_\Gamma>> H^0\operatorname{Tw}(\Lambda_-\vert_\Gamma) \\
@V\iota^+_{\Gamma,\Gamma'}VV @VV\iota^-_{\Gamma,\Gamma'}V \\
H^0\operatorname{Tw}(\Lambda_+\vert_{\Gamma'}) @>F_{\Gamma'}>> H^0\operatorname{Tw}(\Lambda_-\vert_{\Gamma'})
\end{CD} H 0 Tw ( Λ + ∣ Γ ) ι Γ , Γ ′ + ↓ ⏐ H 0 Tw ( Λ + ∣ Γ ′ ) F Γ F Γ ′ H 0 Tw ( Λ − ∣ Γ ) ↓ ⏐ ι Γ , Γ ′ − H 0 Tw ( Λ − ∣ Γ ′ ) such that ϑ Γ , Γ = 1 \vartheta_{\Gamma,\Gamma}=1 ϑ Γ , Γ = 1 and ϑ Γ , Γ ′ ′ = ( ϑ Γ ′ , Γ ′ ′ ι Γ , Γ ′ − ) ∘ ( ι Γ ′ , Γ ′ ′ + ϑ Γ , Γ ′ ) \vartheta_{\Gamma,\Gamma''}=(\vartheta_{\Gamma',\Gamma''}\iota^-_{\Gamma,\Gamma'})\circ(\iota^+_{\Gamma',\Gamma''}\vartheta_{\Gamma,\Gamma'}) ϑ Γ , Γ ′′ = ( ϑ Γ ′ , Γ ′′ ι Γ , Γ ′ − ) ∘ ( ι Γ ′ , Γ ′′ + ϑ Γ , Γ ′ ) for finite subsets Γ ⊂ Γ ′ ⊂ Γ ′ ′ ( ⊂ ω 3 ) \Gamma\subset\Gamma'\subset\Gamma''(\subset\omega_3) Γ ⊂ Γ ′ ⊂ Γ ′′ ( ⊂ ω 3 ) . These isomorphisms commute with the suspensions.
Proof. Since δ h Γ = m 4 ∣ Γ + \delta h_\Gamma=m^+_{4\vert_\Gamma} δ h Γ = m 4 ∣ Γ + , the difference h Γ ′ ∣ Γ − h Γ h_{\Gamma'\vert_\Gamma}-h_\Gamma h Γ ′ ∣ Γ − h Γ is a normalized Hochschild cocycle. Lemma 3.6 gives a normalized k Γ , Γ ′ ∈ C R 2 ( C Γ , E Γ ) k_{\Gamma,\Gamma'}\in C^2_R(C_\Gamma,E_\Gamma) k Γ , Γ ′ ∈ C R 2 ( C Γ , E Γ ) with δ k Γ , Γ ′ = h Γ ′ ∣ Γ − h Γ \delta k_{\Gamma,\Gamma'}=h_{\Gamma'\vert_\Gamma}-h_\Gamma δ k Γ , Γ ′ = h Γ ′ ∣ Γ − h Γ . Choose k Γ , Γ = 0 k_{\Gamma,\Gamma}=0 k Γ , Γ = 0 . Extend k Γ , Γ ′ k_{\Gamma,\Gamma'} k Γ , Γ ′ by zero on E E E inputs, and to matrices:
( k Γ , Γ ′ ( B , A ) ) γ α = ∑ β k Γ , Γ ′ ( B γ β , A β α ) , ( A : X → Y , B : Y → Z are matrices ) . \left(k_{\Gamma,\Gamma'}(B,A)\right)_{\gamma\alpha}=\sum_\beta k_{\Gamma,\Gamma'}(B_{\gamma\beta},A_{\beta\alpha}),\qquad(A:X\to Y,\ B:Y\to Z\text{ are matrices}). ( k Γ , Γ ′ ( B , A ) ) γ α = β ∑ k Γ , Γ ′ ( B γ β , A β α ) , ( A : X → Y , B : Y → Z are matrices ) .
For canonical map s : V → V [ 1 ] s:V\to V[1] s : V → V [ 1 ] , write ( k Γ , Γ ′ s ) ⊗ 2 = s k Γ , Γ ′ (k^s_{\Gamma,\Gamma'})^{\otimes2}=sk_{\Gamma,\Gamma'} ( k Γ , Γ ′ s ) ⊗ 2 = s k Γ , Γ ′ . For X = ( V , q ) X=(V,q) X = ( V , q ) , define
t Γ , Γ ′ , X = 1 V − s − 1 k Γ , Γ ′ s ( q , q ) , ( ϑ Γ , Γ ′ ) X = [ t Γ , Γ ′ , X ] ∈ Hom H 0 Tw ( Λ − ∣ Γ ′ ) ( F Γ X , F Γ ′ X ) . t_{\Gamma,\Gamma',X}=1_V-s^{-1}k^s_{\Gamma,\Gamma'}(q,q),\qquad(\vartheta_{\Gamma,\Gamma'})_X=[t_{\Gamma,\Gamma',X}]\in\operatorname{Hom}_{H^0\operatorname{Tw}(\Lambda_-\vert_{\Gamma'})}(F_\Gamma X,F_{\Gamma'}X). t Γ , Γ ′ , X = 1 V − s − 1 k Γ , Γ ′ s ( q , q ) , ( ϑ Γ , Γ ′ ) X = [ t Γ , Γ ′ , X ] ∈ Hom H 0 Tw ( Λ − ∣ Γ ′ ) ( F Γ X , F Γ ′ X ) .
Here ∣ k Γ , Γ ′ s ∣ = − 1 \lvert k^s_{\Gamma,\Gamma'}\rvert=-1 ∣ k Γ , Γ ′ s ∣ = − 1 , thus ∣ t Γ , Γ ′ , X ∣ = 0 \lvert t_{\Gamma,\Gamma',X}\rvert=0 ∣ t Γ , Γ ′ , X ∣ = 0 . This defines ϑ \vartheta ϑ .
We then show t Γ , Γ ′ , X t_{\Gamma,\Gamma',X} t Γ , Γ ′ , X is closed, so ( ϑ Γ , Γ ′ ) X = [ t Γ , Γ ′ , X ] (\vartheta_{\Gamma,\Gamma'})_X=[t_{\Gamma,\Gamma',X}] ( ϑ Γ , Γ ′ ) X = [ t Γ , Γ ′ , X ] is a morphism in H 0 Tw ( Λ − ∣ Γ ′ ) H^0\operatorname{Tw}(\Lambda_-\vert_{\Gamma'}) H 0 Tw ( Λ − ∣ Γ ′ ) . Put q Γ = q + h Γ s ( q , q , q ) q_\Gamma=q+h^s_\Gamma(q,q,q) q Γ = q + h Γ s ( q , q , q ) and q Γ ′ = q + ( h Γ ′ ∣ Γ ) s ( q , q , q ) q_{\Gamma'}=q+(h_{\Gamma'\vert_\Gamma})^s(q,q,q) q Γ ′ = q + ( h Γ ′ ∣ Γ ) s ( q , q , q ) . We show
( 1 V − s − 1 k Γ , Γ ′ s ( q , q ) ) ∈ Z 0 Hom Tw ( Λ − ∣ Γ ′ ) ( ( V , q Γ ) , ( V , q Γ ′ ) ) . \left(1_V-s^{-1}k^s_{\Gamma,\Gamma'}(q,q)\right)\in Z^0\operatorname{Hom}_{\operatorname{Tw}(\Lambda_-\vert_{\Gamma'})}((V,q_\Gamma),(V,q_{\Gamma'})). ( 1 V − s − 1 k Γ , Γ ′ s ( q , q ) ) ∈ Z 0 Hom Tw ( Λ − ∣ Γ ′ ) (( V , q Γ ) , ( V , q Γ ′ )) .
Let d d d denote the differential on H o m Tw ( Λ − ∣ Γ ′ ) ( ( V , q Γ ) , ( V , q Γ ′ ) ) . \mathcal{H}om_{\operatorname{Tw}(\Lambda_-\vert_{\Gamma'})}((V,q_\Gamma),(V,q_{\Gamma'})). H o m Tw ( Λ − ∣ Γ ′ ) (( V , q Γ ) , ( V , q Γ ′ )) . Write b 2 = s ∘ m 2 ∘ ( s ⊗ s ) − 1 b_2=s\circ m_2\circ(s\otimes s)^{-1} b 2 = s ∘ m 2 ∘ ( s ⊗ s ) − 1 . Using ( δ k Γ , Γ ′ ) s ⊗ 3 = s ( δ k Γ , Γ ′ ) (\delta k_{\Gamma,\Gamma'})^{s\otimes3}=s(\delta k_{\Gamma,\Gamma'}) ( δ k Γ , Γ ′ ) s ⊗ 3 = s ( δ k Γ , Γ ′ ) , and evaluating at ( q , q , q ) (q,q,q) ( q , q , q ) gives
( δ k Γ , Γ ′ ) s ( q , q , q ) = − b 2 ( q , k Γ , Γ ′ s ( q , q ) ) − b 2 ( k Γ , Γ ′ s ( q , q ) , q ) − k Γ , Γ ′ s ( b 2 ( q , q ) , q ) − k Γ , Γ ′ s ( q , b 2 ( q , q ) ) . (\delta k_{\Gamma,\Gamma'})^s(q,q,q)=-b_2(q,k^s_{\Gamma,\Gamma'}(q,q))-b_2(k^s_{\Gamma,\Gamma'}(q,q),q)-k^s_{\Gamma,\Gamma'}(b_2(q,q),q)-k^s_{\Gamma,\Gamma'}(q,b_2(q,q)). ( δ k Γ , Γ ′ ) s ( q , q , q ) = − b 2 ( q , k Γ , Γ ′ s ( q , q )) − b 2 ( k Γ , Γ ′ s ( q , q ) , q ) − k Γ , Γ ′ s ( b 2 ( q , q ) , q ) − k Γ , Γ ′ s ( q , b 2 ( q , q )) .
The Maurer–Cartan equation for q q q is b 2 ( q , q ) + b 4 ( q , q , q , q ) = 0 b_2(q,q)+b^4(q,q,q,q)=0 b 2 ( q , q ) + b 4 ( q , q , q , q ) = 0 . Using E 2 = 0 E^2=0 E 2 = 0 , the vanishing of k Γ , Γ ′ k_{\Gamma,\Gamma'} k Γ , Γ ′ on E E E inputs, and δ k Γ , Γ ′ = h Γ ′ ∣ Γ − h Γ \delta k_{\Gamma,\Gamma'}=h_{\Gamma'\vert_\Gamma}-h_\Gamma δ k Γ , Γ ′ = h Γ ′ ∣ Γ − h Γ , we obtain
s d ( t Γ , Γ ′ , X ) = b 2 ( q Γ ′ , s t Γ , Γ ′ , X ) + b 2 ( s t Γ , Γ ′ , X , q Γ ) = q Γ − q Γ ′ − b 2 ( q Γ ′ , k Γ , Γ ′ s ( q , q ) ) − b 2 ( k Γ , Γ ′ s ( q , q ) , q Γ ) = q Γ − q Γ ′ − b 2 ( q , k Γ , Γ ′ s ( q , q ) ) − b 2 ( k Γ , Γ ′ s ( q , q ) , q ) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) + k Γ , Γ ′ s ( b 2 ( q , q ) , q ) + k Γ , Γ ′ s ( q , b 2 ( q , q ) ) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) − k Γ , Γ ′ s ( b 4 ( q , q , q , q ) , q ) − k Γ , Γ ′ s ( q , b 4 ( q , q , q ) ) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) = ( h Γ − h Γ ′ ∣ Γ + δ k Γ , Γ ′ ) s ( q , q , q ) = 0. \begin{aligned}
s\,d(t_{\Gamma,\Gamma',X})&=b_2(q_{\Gamma'},st_{\Gamma,\Gamma',X})+b_2(st_{\Gamma,\Gamma',X},q_\Gamma)\\
&=q_\Gamma-q_{\Gamma'}-b_2(q_{\Gamma'},k^s_{\Gamma,\Gamma'}(q,q))-b_2(k^s_{\Gamma,\Gamma'}(q,q),q_\Gamma)\\
&=q_\Gamma-q_{\Gamma'}-b_2(q,k^s_{\Gamma,\Gamma'}(q,q))-b_2(k^s_{\Gamma,\Gamma'}(q,q),q)\\
&=q_\Gamma-q_{\Gamma'}+(\delta k_{\Gamma,\Gamma'})^s(q,q,q)+k^s_{\Gamma,\Gamma'}(b_2(q,q),q)+k^s_{\Gamma,\Gamma'}(q,b_2(q,q))\\
&=q_\Gamma-q_{\Gamma'}+(\delta k_{\Gamma,\Gamma'})^s(q,q,q)-k^s_{\Gamma,\Gamma'}(b^4(q,q,q,q),q)-k^s_{\Gamma,\Gamma'}(q,b^4(q,q,q))\\
&=q_\Gamma-q_{\Gamma'}+(\delta k_{\Gamma,\Gamma'})^s(q,q,q)=(h_\Gamma-h_{\Gamma'\vert_\Gamma}+\delta k_{\Gamma,\Gamma'})^s(q,q,q)=0.
\end{aligned} s d ( t Γ , Γ ′ , X ) = b 2 ( q Γ ′ , s t Γ , Γ ′ , X ) + b 2 ( s t Γ , Γ ′ , X , q Γ ) = q Γ − q Γ ′ − b 2 ( q Γ ′ , k Γ , Γ ′ s ( q , q )) − b 2 ( k Γ , Γ ′ s ( q , q ) , q Γ ) = q Γ − q Γ ′ − b 2 ( q , k Γ , Γ ′ s ( q , q )) − b 2 ( k Γ , Γ ′ s ( q , q ) , q ) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) + k Γ , Γ ′ s ( b 2 ( q , q ) , q ) + k Γ , Γ ′ s ( q , b 2 ( q , q )) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) − k Γ , Γ ′ s ( b 4 ( q , q , q , q ) , q ) − k Γ , Γ ′ s ( q , b 4 ( q , q , q )) = q Γ − q Γ ′ + ( δ k Γ , Γ ′ ) s ( q , q , q ) = ( h Γ − h Γ ′ ∣ Γ + δ k Γ , Γ ′ ) s ( q , q , q ) = 0.
Since s s s is invertible, d ( t Γ , Γ ′ , X ) = 0 d(t_{\Gamma,\Gamma',X})=0 d ( t Γ , Γ ′ , X ) = 0 , as required.
For naturality, take a closed degree-zero matrix A : X = ( V , q X ) → Y = ( W , q Y ) A:X=(V,q_X)\to Y=(W,q_Y) A : X = ( V , q X ) → Y = ( W , q Y ) . Let d d d denote the differential on Hom Tw ( Λ − ∣ Γ ′ ) ( F Γ X , F Γ ′ Y ) \operatorname{Hom}_{\operatorname{Tw}(\Lambda_{-|\Gamma'})}(F_\Gamma X,F_{\Gamma'}Y) Hom Tw ( Λ − ∣ Γ ′ ) ( F Γ X , F Γ ′ Y ) . Then
t Γ , Γ ′ , Y ∘ Tw ( f Γ ) 1 ( A ) − Tw ( f Γ ′ ) 1 ( A ) ∘ t Γ , Γ ′ , X = ( 1 ) ( 1 V − s − 1 k Γ , Γ ′ s ( q Y , q Y ) ) ∘ s − 1 ( s A + ∑ r + t = 2 h Γ s ( q Y ⊗ r , s A , q X ⊗ t ) ) − s − 1 ( s A + ∑ r + t = 2 ( h Γ ′ ∣ Γ ) s ( q Y ⊗ r , s A , q X ⊗ t ) ) ∘ ( 1 V − s − 1 k Γ , Γ ′ s ( q X , q X ) ) (3) \begin{aligned}
&t_{\Gamma,\Gamma',Y}\circ\operatorname{Tw}(f_\Gamma)_1(A)-\operatorname{Tw}(f_{\Gamma'})_1(A)\circ t_{\Gamma,\Gamma',X}\\
&\underset{(1)}{=}(1_V-s^{-1}k^s_{\Gamma,\Gamma'}(q_Y,q_Y))\circ s^{-1}\left(sA+\sum_{r+t=2}h^s_\Gamma(q_Y^{\otimes r},sA,q_X^{\otimes t})\right)\\
&\quad-s^{-1}\left(sA+\sum_{r+t=2}(h_{\Gamma'|\Gamma})^s(q_Y^{\otimes r},sA,q_X^{\otimes t})\right)\circ(1_V-s^{-1}k^s_{\Gamma,\Gamma'}(q_X,q_X)) \tag*{(3)}
\end{aligned} t Γ , Γ ′ , Y ∘ Tw ( f Γ ) 1 ( A ) − Tw ( f Γ ′ ) 1 ( A ) ∘ t Γ , Γ ′ , X ( 1 ) = ( 1 V − s − 1 k Γ , Γ ′ s ( q Y , q Y )) ∘ s − 1 ( s A + r + t = 2 ∑ h Γ s ( q Y ⊗ r , s A , q X ⊗ t ) ) − s − 1 ( s A + r + t = 2 ∑ ( h Γ ′ ∣Γ ) s ( q Y ⊗ r , s A , q X ⊗ t ) ) ∘ ( 1 V − s − 1 k Γ , Γ ′ s ( q X , q X )) (3)
s − 1 ( ∑ r + t = 2 ( h Γ − h Γ ′ ∣ Γ ) s ( q Y ⊗ r , s A , q X ⊗ t ) − b 2 ( k Γ , Γ ′ s ( q Y , q Y ) , s A ) + b 2 ( s A , k Γ , Γ ′ s ( q X , q X ) ) ) (4) s^{-1}\left(\sum_{r+t=2}(h_\Gamma-h_{\Gamma'|\Gamma})^s(q_Y^{\otimes r},sA,q_X^{\otimes t})-b_2(k^s_{\Gamma,\Gamma'}(q_Y,q_Y),sA)+b_2(sA,k^s_{\Gamma,\Gamma'}(q_X,q_X))\right) \tag*{(4)} s − 1 ( r + t = 2 ∑ ( h Γ − h Γ ′ ∣Γ ) s ( q Y ⊗ r , s A , q X ⊗ t ) − b 2 ( k Γ , Γ ′ s ( q Y , q Y ) , s A ) + b 2 ( s A , k Γ , Γ ′ s ( q X , q X )) ) (4)
s − 1 ( − ∑ r + t = 2 ( δ k Γ , Γ ′ ) s ( q Y ⊗ r , s A , q X ⊗ t ) − b 2 ( k Γ , Γ ′ s ( q Y , q Y ) , s A ) + b 2 ( s A , k Γ , Γ ′ s ( q X , q X ) ) ) (5) s^{-1}\left(-\sum_{r+t=2}(\delta k_{\Gamma,\Gamma'})^s(q_Y^{\otimes r},sA,q_X^{\otimes t})-b_2(k^s_{\Gamma,\Gamma'}(q_Y,q_Y),sA)+b_2(sA,k^s_{\Gamma,\Gamma'}(q_X,q_X))\right) \tag*{(5)} s − 1 ( − r + t = 2 ∑ ( δ k Γ , Γ ′ ) s ( q Y ⊗ r , s A , q X ⊗ t ) − b 2 ( k Γ , Γ ′ s ( q Y , q Y ) , s A ) + b 2 ( s A , k Γ , Γ ′ s ( q X , q X )) ) (5)
s − 1 ( b 2 ( q Y , k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X ) ) + b 2 ( k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X ) , q X ) ) (6) s^{-1}\left(b_2(q_Y,k^s_{\Gamma,\Gamma'}(q_Y,sA)+k^s_{\Gamma,\Gamma'}(sA,q_X))+b_2(k^s_{\Gamma,\Gamma'}(q_Y,sA)+k^s_{\Gamma,\Gamma'}(sA,q_X),q_X)\right) \tag*{(6)} s − 1 ( b 2 ( q Y , k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X )) + b 2 ( k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X ) , q X ) ) (6)
d ( s − 1 ( k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X ) ) ) . (7) d\left(s^{-1}(k^s_{\Gamma,\Gamma'}(q_Y,sA)+k^s_{\Gamma,\Gamma'}(sA,q_X))\right). \tag*{(7)} d ( s − 1 ( k Γ , Γ ′ s ( q Y , s A ) + k Γ , Γ ′ s ( s A , q X )) ) . (7)
Here (1) is from the definition, (2) uses E 2 = 0 E^2=0 E 2 = 0 , (3) uses h Γ − h Γ ′ ∣ Γ = − δ k Γ , Γ ′ h_\Gamma-h_{\Gamma'|\Gamma}=-\delta k_{\Gamma,\Gamma'} h Γ − h Γ ′ ∣Γ = − δ k Γ , Γ ′ , (4) uses
( δ k Γ , Γ ′ ) s ( u , v , w ) = − b 2 ( k Γ , Γ ′ s ( u , v ) , w ) − ( − 1 ) ∣ u ∣ b 2 ( u , k Γ , Γ ′ s ( v , w ) ) − k Γ , Γ ′ s ( b 2 ( u , v ) , w ) − ( − 1 ) ∣ u ∣ k Γ , Γ ′ s ( u , b 2 ( v , w ) ) , (\delta k_{\Gamma,\Gamma'})^s(u,v,w)=-b_2(k^s_{\Gamma,\Gamma'}(u,v),w)-(-1)^{|u|}b_2(u,k^s_{\Gamma,\Gamma'}(v,w))-k^s_{\Gamma,\Gamma'}(b_2(u,v),w)-(-1)^{|u|}k^s_{\Gamma,\Gamma'}(u,b_2(v,w)), ( δ k Γ , Γ ′ ) s ( u , v , w ) = − b 2 ( k Γ , Γ ′ s ( u , v ) , w ) − ( − 1 ) ∣ u ∣ b 2 ( u , k Γ , Γ ′ s ( v , w )) − k Γ , Γ ′ s ( b 2 ( u , v ) , w ) − ( − 1 ) ∣ u ∣ k Γ , Γ ′ s ( u , b 2 ( v , w )) ,
(5) follows from the twisted differential formula and E 2 = 0 E^2=0 E 2 = 0 . Thus, in H 0 Tw ( Λ − ∣ Γ ′ ) H^0\operatorname{Tw}(\Lambda_{-|\Gamma'}) H 0 Tw ( Λ − ∣ Γ ′ ) ,
( ϑ Γ , Γ ′ ) Y ∘ F Γ ( [ A ] ) = F Γ ′ ( [ A ] ) ∘ ( ϑ Γ , Γ ′ ) X . (\vartheta_{\Gamma,\Gamma'})_Y\circ F_\Gamma([A])=F_{\Gamma'}([A])\circ(\vartheta_{\Gamma,\Gamma'})_X. ( ϑ Γ , Γ ′ ) Y ∘ F Γ ([ A ]) = F Γ ′ ([ A ]) ∘ ( ϑ Γ , Γ ′ ) X .
ϑ Γ , Γ ′ \vartheta_{\Gamma,\Gamma'} ϑ Γ , Γ ′ is a natural isomorphism, where 1 V + s − 1 k Γ , Γ ′ s ( q , q ) 1_V+s^{-1}k^s_{\Gamma,\Gamma'}(q,q) 1 V + s − 1 k Γ , Γ ′ s ( q , q ) represents the inverse.
Finally, ϑ Γ , Γ = 1 \vartheta_{\Gamma,\Gamma}=1 ϑ Γ , Γ = 1 follows from the construction. It remains to show for finite subsets Γ ⊆ Γ ′ ⊆ Γ ′ ′ ( ⊆ ω 3 ) \Gamma\subseteq\Gamma'\subseteq\Gamma''(\subseteq\omega_3) Γ ⊆ Γ ′ ⊆ Γ ′′ ( ⊆ ω 3 ) , one has ϑ Γ , Γ ′ ′ = ( ϑ Γ ′ , Γ ′ ′ ι Γ , Γ ′ ) ∘ ( ι Γ ′ , Γ ′ ′ ϑ Γ , Γ ′ ) \vartheta_{\Gamma,\Gamma''}=(\vartheta_{\Gamma',\Gamma''}\iota_{\Gamma,\Gamma'})\circ(\iota_{\Gamma',\Gamma''}\vartheta_{\Gamma,\Gamma'}) ϑ Γ , Γ ′′ = ( ϑ Γ ′ , Γ ′′ ι Γ , Γ ′ ) ∘ ( ι Γ ′ , Γ ′′ ϑ Γ , Γ ′ ) . Note that
δ ( k Γ , Γ ′ ′ − k Γ , Γ ′ − k Γ ′ , Γ ′ ′ ∣ Γ ) = ( h Γ ′ ′ ∣ Γ − h Γ ) − ( h Γ ′ ∣ Γ − h Γ ) − ( h Γ ′ ′ ∣ Γ − h Γ ′ ∣ Γ ) = 0. \delta(k_{\Gamma,\Gamma''}-k_{\Gamma,\Gamma'}-k_{\Gamma',\Gamma''|\Gamma})=(h_{\Gamma''|\Gamma}-h_\Gamma)-(h_{\Gamma'|\Gamma}-h_\Gamma)-(h_{\Gamma''|\Gamma}-h_{\Gamma'|\Gamma})=0. δ ( k Γ , Γ ′′ − k Γ , Γ ′ − k Γ ′ , Γ ′′ ∣Γ ) = ( h Γ ′′ ∣Γ − h Γ ) − ( h Γ ′ ∣Γ − h Γ ) − ( h Γ ′′ ∣Γ − h Γ ′ ∣Γ ) = 0.
By Lemma 3.6 , take l ∈ C R 1 ( C Γ , E Γ ) l\in C^1_R(C_\Gamma,E_\Gamma) l ∈ C R 1 ( C Γ , E Γ ) with δ l = k Γ , Γ ′ ′ − k Γ , Γ ′ − k Γ ′ , Γ ′ ′ ∣ Γ \delta l=k_{\Gamma,\Gamma''}-k_{\Gamma,\Gamma'}-k_{\Gamma',\Gamma''|\Gamma} δ l = k Γ , Γ ′′ − k Γ , Γ ′ − k Γ ′ , Γ ′′ ∣Γ . Fix X = ( V , q ) X=(V,q) X = ( V , q ) . Again we put q ⋆ = q + h s ( q , q , q ) q_\star=q+h^s(q,q,q) q ⋆ = q + h s ( q , q , q ) . Let d d d denote the differential on Hom Tw ( Λ − ∣ Γ ′ ′ ) ( ( V , q Γ ) , ( V , q Γ ′ ′ ) ) \operatorname{Hom}_{\operatorname{Tw}(\Lambda_{-|\Gamma''})}((V,q_\Gamma),(V,q_{\Gamma''})) Hom Tw ( Λ − ∣ Γ ′′ ) (( V , q Γ ) , ( V , q Γ ′′ )) . Write l s = s l l^s=sl l s = s l . Suppressing the natural inclusions, we have
t Γ , Γ ′ ′ , X − t Γ ′ , Γ ′ ′ , X ∘ t Γ , Γ ′ , X t_{\Gamma,\Gamma'',X}-t_{\Gamma',\Gamma'',X}\circ t_{\Gamma,\Gamma',X} t Γ , Γ ′′ , X − t Γ ′ , Γ ′′ , X ∘ t Γ , Γ ′ , X
= [ 1 ] 1 V − s − 1 k Γ , Γ ′ ′ s ( q , q ) − ( 1 V − s − 1 ( k Γ ′ , Γ ′ ′ ∣ Γ ) s ( q , q ) ) ∘ ( 1 V − s − 1 k Γ , Γ ′ s ( q , q ) ) \underset{[1]}{=}1_V-s^{-1}k^s_{\Gamma,\Gamma''}(q,q)-(1_V-s^{-1}(k_{\Gamma',\Gamma''|\Gamma})^s(q,q))\circ(1_V-s^{-1}k^s_{\Gamma,\Gamma'}(q,q)) [ 1 ] = 1 V − s − 1 k Γ , Γ ′′ s ( q , q ) − ( 1 V − s − 1 ( k Γ ′ , Γ ′′ ∣Γ ) s ( q , q )) ∘ ( 1 V − s − 1 k Γ , Γ ′ s ( q , q ))
= [ 2 ] − s − 1 ( k Γ , Γ ′ ′ − k Γ , Γ ′ − k Γ ′ , Γ ′ ′ ∣ Γ ) s ( q , q ) \underset{[2]}{=}-s^{-1}(k_{\Gamma,\Gamma''}-k_{\Gamma,\Gamma'}-k_{\Gamma',\Gamma''|\Gamma})^s(q,q) [ 2 ] = − s − 1 ( k Γ , Γ ′′ − k Γ , Γ ′ − k Γ ′ , Γ ′′ ∣Γ ) s ( q , q )
= [ 3 ] − s − 1 ( δ l ) s ( q , q ) \underset{[3]}{=}-s^{-1}(\delta l)^s(q,q) [ 3 ] = − s − 1 ( δ l ) s ( q , q )
= [ 4 ] − s − 1 ( b 2 ( l s ( q ) , q ) + b 2 ( q , l s ( q ) ) − l s ( b 2 ( q , q ) ) ) \underset{[4]}{=}-s^{-1}\left(b_2(l^s(q),q)+b_2(q,l^s(q))-l^s(b_2(q,q))\right) [ 4 ] = − s − 1 ( b 2 ( l s ( q ) , q ) + b 2 ( q , l s ( q )) − l s ( b 2 ( q , q )) )
= [ 5 ] − s − 1 ( b 2 ( l s ( q ) , q ) + b 2 ( q , l s ( q ) ) ) \underset{[5]}{=}-s^{-1}\left(b_2(l^s(q),q)+b_2(q,l^s(q))\right) [ 5 ] = − s − 1 ( b 2 ( l s ( q ) , q ) + b 2 ( q , l s ( q )) )
= [ 6 ] s − 1 ( b 2 ( q Γ ′ ′ , − l s ( q ) ) + b 2 ( − l s ( q ) , q Γ ) ) \underset{[6]}{=}s^{-1}\left(b_2(q_{\Gamma''},-l^s(q))+b_2(-l^s(q),q_\Gamma)\right) [ 6 ] = s − 1 ( b 2 ( q Γ ′′ , − l s ( q )) + b 2 ( − l s ( q ) , q Γ ) )
= [ 7 ] d ( − s − 1 l s ( q ) ) . \underset{[7]}{=}d(-s^{-1}l^s(q)). [ 7 ] = d ( − s − 1 l s ( q )) .
Here [2 ] and [6 ] use E 2 = 0 E^2=0 E 2 = 0 , [5 ] uses l s ( b 2 ( q , q ) ) = − l s ( b 4 + ( q , q , q , q ) ) = 0 l^s(b_2(q,q))=-l^s(b^+_4(q,q,q,q))=0 l s ( b 2 ( q , q )) = − l s ( b 4 + ( q , q , q , q )) = 0 , and the remaining follows from definitions or constructions. Taking H 0 H^0 H 0 gives the result. □
Remark 3.10. Some verifications in Lemma 3.9 are not necessary by [1, Corollary 11.22.
Proposition 3.11. There is a triangle equivalence Perf ( Λ + ) ≃ Perf ( Λ − ) \operatorname{Perf}(\Lambda_{+}) \simeq\operatorname{Perf}(\Lambda_{-}) Perf ( Λ + ) ≃ Perf ( Λ − ) fixing the one-term objects and their graded morphism groups, i.e.,
F Γ ( α , 0 ) = ( α , 0 ) , Hom H 0 Tw ( Λ + ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) → ∼ Hom H 0 Tw ( Λ − ∣ Γ ) ( ( α , 0 ) , ( β , 0 ) [ d ] ) . F_{\Gamma}(\alpha,0)=(\alpha,0),\qquad\operatorname{Hom}_{H^{0}\operatorname{Tw}(\Lambda_{+|\Gamma})}((\alpha,0),(\beta,0)[d]) \xrightarrow{\sim} \operatorname{Hom}_{H^{0}\operatorname{Tw}(\Lambda_{-|\Gamma})}((\alpha,0),(\beta,0)[d]). F Γ ( α , 0 ) = ( α , 0 ) , Hom H 0 Tw ( Λ + ∣Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) ∼ Hom H 0 Tw ( Λ − ∣Γ ) (( α , 0 ) , ( β , 0 ) [ d ]) .
Proof. We first construct a triangle equivalence F : H 0 Tw ( Λ + ) ⟶ H 0 Tw ( Λ − ) F:H^{0}\operatorname{Tw}(\Lambda_{+})\longrightarrow H^{0}\operatorname{Tw}(\Lambda_{-}) F : H 0 Tw ( Λ + ) ⟶ H 0 Tw ( Λ − ) .
For each twisted complex X X X , let Γ X \Gamma_X Γ X be the finite set of ordinals occurring in its underlying formal sum, taking Γ X = { 0 } \Gamma_X=\{0\} Γ X = { 0 } for the empty sum, and put F ( X ) = F Γ X ( X ) F(X)=F_{\Gamma_X}(X) F ( X ) = F Γ X ( X ) . For a morphism a : X → Y a:X\to Y a : X → Y , choose a finite set Γ ′ ⊇ Γ X ∪ Γ Y \Gamma'\supseteq\Gamma_X\cup\Gamma_Y Γ ′ ⊇ Γ X ∪ Γ Y and define F ( a ) = ( ϑ Γ Y , Γ ′ ) − 1 ∘ F Γ ′ ( a ) ∘ ( ϑ Γ X , Γ ′ ) X F(a)=(\vartheta_{\Gamma_Y,\Gamma'})^{-1}\circ F_{\Gamma'}(a)\circ(\vartheta_{\Gamma_X,\Gamma'})_X F ( a ) = ( ϑ Γ Y , Γ ′ ) − 1 ∘ F Γ ′ ( a ) ∘ ( ϑ Γ X , Γ ′ ) X . If Γ ′ ⊆ Γ ′ ′ \Gamma'\subseteq\Gamma'' Γ ′ ⊆ Γ ′′ , then Lemma 3.9 gives (suppressing natural inclusions)
( ϑ Γ Y , Γ ′ ′ ) − 1 ∘ F Γ ′ ′ ( a ) ∘ ( ϑ Γ X , Γ ′ ′ ) X = ( ϑ Γ Y , Γ ′ ) − 1 ∘ ( ϑ Γ ′ , Γ ′ ′ ) − 1 ∘ F Γ ′ ′ ( a ) ∘ ( ϑ Γ ′ , Γ ′ ′ ) X ∘ ( ϑ Γ X , Γ ′ ) X = ( ϑ Γ Y , Γ ′ ) − 1 ∘ (\vartheta_{\Gamma_Y,\Gamma''})^{-1}\circ F_{\Gamma''}(a)\circ(\vartheta_{\Gamma_X,\Gamma''})_X=(\vartheta_{\Gamma_Y,\Gamma'})^{-1}\circ(\vartheta_{\Gamma',\Gamma''})^{-1}\circ F_{\Gamma''}(a)\circ(\vartheta_{\Gamma',\Gamma''})_X\circ(\vartheta_{\Gamma_X,\Gamma'})_X=(\vartheta_{\Gamma_Y,\Gamma'})^{-1}\circ ( ϑ Γ Y , Γ ′′ ) − 1 ∘ F Γ ′′ ( a ) ∘ ( ϑ Γ X , Γ ′′ ) X = ( ϑ Γ Y , Γ ′ ) − 1 ∘ ( ϑ Γ ′ , Γ ′′ ) − 1 ∘ F Γ ′′ ( a ) ∘ ( ϑ Γ ′ , Γ ′′ ) X ∘ ( ϑ Γ X , Γ ′ ) X = ( ϑ Γ Y , Γ ′ ) − 1 ∘
Hence, F ( a ) F(a) F ( a ) is well-defined. Similarly F F F preserves composition, identities and R R R -linearity.
For fixed X , Y X,Y X , Y , take Γ X \Gamma_X Γ X , Γ Y \Gamma_Y Γ Y and Γ ′ \Gamma' Γ ′ as above. Then F F F acts on hom sets as the composite
Hom H 0 Tw ( Λ + ) ( X , Y ) → ∼ F Γ ′ Hom H 0 Tw ( Λ − ) ( F Γ ′ X , F Γ ′ Y ) → ∼ Hom H 0 Tw ( Λ − ) ( F X , F Y ) . \operatorname{Hom}_{H^{0}\operatorname{Tw}(\Lambda_{+})}(X,Y)\xrightarrow[\sim]{F_{\Gamma'}}\operatorname{Hom}_{H^{0}\operatorname{Tw}(\Lambda_{-})}(F_{\Gamma'}X,F_{\Gamma'}Y)\xrightarrow{\sim}\operatorname{Hom}_{H^{0}\operatorname{Tw}(\Lambda_{-})}(FX,FY). Hom H 0 Tw ( Λ + ) ( X , Y ) F Γ ′ ∼ Hom H 0 Tw ( Λ − ) ( F Γ ′ X , F Γ ′ Y ) ∼ Hom H 0 Tw ( Λ − ) ( F X , F Y ) .
Hence F F F is fully faithful by Lemma 3.8. For any Y ∈ H 0 Tw ( Λ − ) Y\in H^{0}\operatorname{Tw}(\Lambda_{-}) Y ∈ H 0 Tw ( Λ − ) , choose finite Γ ′ ⊇ Γ Y ∪ { 0 } \Gamma'\supseteq\Gamma_Y\cup\{0\} Γ ′ ⊇ Γ Y ∪ { 0 } . The same lemma gives X ∈ H 0 Tw ( Λ + ∣ Γ ′ ) X\in H^{0}\operatorname{Tw}(\Lambda_{+|\Gamma'}) X ∈ H 0 Tw ( Λ + ∣ Γ ′ ) with F Γ ′ ( X ) ≅ Y F_{\Gamma'}(X)\cong Y F Γ ′ ( X ) ≅ Y . Since ( ϑ Γ X , Γ ′ ) X : F ( X ) → ∼ F Γ ′ ( X ) (\vartheta_{\Gamma_X,\Gamma'})_X:F(X)\xrightarrow{\sim}F_{\Gamma'}(X) ( ϑ Γ X , Γ ′ ) X : F ( X ) ∼ F Γ ′ ( X ) , the functor F F F is essentially surjective.
Since the axioms of a triangulated category are checked in finite stages, F F F is an equivalence. Idempotent completion passes the isomorphism to perfect categories. □
Detecting differences in m 4 m_{4} m 4 via Toda brackets
The category Tw ( Λ ± ) \operatorname{Tw}(\Lambda_{\pm}) Tw ( Λ ± ) uses finite formal coproducts. In this subsection, we allow formal coproducts of ≤ ℵ 3 \leq\aleph_{3} ≤ ℵ 3 representable right modules, obtaining an inclusion Tw ( Λ ± ) ↪ Tw ( Λ ± ( ≤ ℵ 3 ) ) \operatorname{Tw}(\Lambda_{\pm})\hookrightarrow\operatorname{Tw}(\Lambda_{\pm}^{(\leq\aleph_{3})}) Tw ( Λ ± ) ↪ Tw ( Λ ± ( ≤ ℵ 3 ) ) . The resulting larger thick subcategories of the derived categories detect the difference between the two m 4 m_{4} m 4 operations.
Let Λ b ( ≤ ℵ 3 ) \Lambda_{b}^{(\leq\aleph_{3})} Λ b ( ≤ ℵ 3 ) have objects families ( α i ) i ∈ L (\alpha_i)_{i\in L} ( α i ) i ∈ L of ordinals below ω 3 \omega_{3} ω 3 , with ∣ L ∣ ≤ ∣ ω 3 ∣ |L|\leq|\omega_{3}| ∣ L ∣ ≤ ∣ ω 3 ∣ . The family represents the free right C C C -module ∐ i ∈ L y α i \coprod_{i\in L}y_{\alpha_i} ∐ i ∈ L y α i . For any such two families, set
H o m Λ b ( ≤ ℵ 3 ) ( ( α i ) i ∈ L , ( β j ) j ∈ K ) r = ∏ i ∈ L ∏ j ∈ K H o m Λ b ( α i , β j ) r . \mathcal{H}om_{\Lambda_{b}^{(\leq\aleph_{3})}}\bigl((\alpha_i)_{i\in L},(\beta_j)_{j\in K}\bigr)^{r}=\prod_{i\in L}\prod_{j\in K}\mathcal{H}om_{\Lambda_b}(\alpha_i,\beta_j)^{r}. H o m Λ b ( ≤ ℵ 3 ) ( ( α i ) i ∈ L , ( β j ) j ∈ K ) r = i ∈ L ∏ j ∈ K ∏ H o m Λ b ( α i , β j ) r .
Thus a morphism is a column-finite matrix. Hence one can extend m i m_i m i by summing over composable matrix entries. From a fixed input column, only finitely many paths of a fixed length occur, and these operations remain column-finite. The Stasheff identities are follow entrywise from Lemma 3.5. For the corresponding free modules P , Q P,Q P , Q , one has
H o m Λ b ( ≤ ℵ 3 ) ( P , Q ) 0 = Hom C ( P , Q ) , H o m Λ b ( ≤ ℵ 3 ) ( P , Q ) − 2 = Hom C ( P , Q ⊗ C E ) . \mathcal{H}om_{\Lambda_{b}^{(\leq\aleph_{3})}}(P,Q)^{0}=\operatorname{Hom}_{C}(P,Q),\qquad\mathcal{H}om_{\Lambda_{b}^{(\leq\aleph_{3})}}(P,Q)^{-2}=\operatorname{Hom}_{C}(P,Q\otimes_{C}E). H o m Λ b ( ≤ ℵ 3 ) ( P , Q ) 0 = Hom C ( P , Q ) , H o m Λ b ( ≤ ℵ 3 ) ( P , Q ) − 2 = Hom C ( P , Q ⊗ C E ) .
For b ∈ { + , − } b\in\{+,-\} b ∈ { + , − } , this gives an A ∞ A_{\infty} A ∞ -functor Y b : Λ b ( ≤ ℵ 3 ) ⟶ Mod ∞ ( Λ b ) Y_b:\Lambda_{b}^{(\leq\aleph_{3})}\longrightarrow\operatorname{Mod}_{\infty}(\Lambda_b) Y b : Λ b ( ≤ ℵ 3 ) ⟶ Mod ∞ ( Λ b ) via
( α i ) i ∈ L ⟼ H o m Λ b ( ≤ ℵ 3 ) ( ι b ( − ) , ( α i ) i ∈ L ) = ∏ i ∈ L H o m Λ b ( ≤ ℵ 3 ) ( ι b ( − ) , α i ) . (\alpha_i)_{i\in L}\longmapsto\mathcal{H}om_{\Lambda_{b}^{(\leq\aleph_{3})}}(\iota_b(-),(\alpha_i)_{i\in L})=\prod_{i\in L}\mathcal{H}om_{\Lambda_{b}^{(\leq\aleph_{3})}}(\iota_b(-),\alpha_i). ( α i ) i ∈ L ⟼ H o m Λ b ( ≤ ℵ 3 ) ( ι b ( − ) , ( α i ) i ∈ L ) = i ∈ L ∏ H o m Λ b ( ≤ ℵ 3 ) ( ι b ( − ) , α i ) .
Lemma 3.12. Y b Y_b Y b induces a fully faithful triangle functor H 0 Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ D ∞ ( Λ b ) H^{0}\operatorname{Tw}(\Lambda_{b}^{(\leq\aleph_{3})})\longrightarrow D_{\infty}(\Lambda_b) H 0 Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ D ∞ ( Λ b ) .
Proof. Let ι b : Λ b ↪ Λ b ( ≤ ℵ 3 ) \iota_b:\Lambda_b\hookrightarrow\Lambda_{b}^{(\leq\aleph_{3})} ι b : Λ b ↪ Λ b ( ≤ ℵ 3 ) be the full inclusion of singleton families. As in [11 §7.5, Theorem] , extend the Yoneda functor to finite twisted complexes and restrict along ι b \iota_b ι b :
Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ Mod ∞ ( Λ b ( ≤ ℵ 3 ) ) → ι b ∗ Mod ∞ ( Λ b ) . \operatorname{Tw}(\Lambda_{b}^{(\leq\aleph_{3})})\longrightarrow\operatorname{Mod}_{\infty}(\Lambda_{b}^{(\leq\aleph_{3})})\xrightarrow{\iota_b^{*}}\operatorname{Mod}_{\infty}(\Lambda_b). Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ Mod ∞ ( Λ b ( ≤ ℵ 3 ) ) ι b ∗ Mod ∞ ( Λ b ) .
Note that [11 §7.5] works over fields; over the commutative ring R R R , the extension is given by the same finite twisting formulas; see [1 §11.20 and Proposition 13.19] . Taking H 0 H^0 H 0 and then localising at objectwise quasi-isomorphisms gives a triangle functor
Y ~ b : H 0 Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ D ∞ ( Λ b ) \widetilde{Y}_{b}: H^0\operatorname{Tw}\left(\Lambda_b^{(\leq\aleph_3)}\right) \longrightarrow D_{\infty}(\Lambda_b) Y b : H 0 Tw ( Λ b ( ≤ ℵ 3 ) ) ⟶ D ∞ ( Λ b )
whose restriction to matrix objects is induced by Y b Y_b Y b .
By [13 Appendix A, Theorem A.1] one has a homotopy equivalence
Hom Mod ∞ ( Λ b ) ( Hom Λ b ( − , α ) , N ) ≃ N ( α ) . \operatorname{Hom}_{\operatorname{Mod}_{\infty}(\Lambda_b)}\left(\operatorname{Hom}_{\Lambda_b}(-,\alpha),N\right) \simeq N(\alpha). Hom Mod ∞ ( Λ b ) ( Hom Λ b ( − , α ) , N ) ≃ N ( α ) .
For P = ( α i ) i ∈ L P=(\alpha_i)_{i\in L} P = ( α i ) i ∈ L and a module N N N , it gives Hom Mod ∞ ( Λ b ) ( Y ~ b ( P ) , N ) ≃ ∏ i ∈ L N ( α i ) \operatorname{Hom}_{\operatorname{Mod}_{\infty}(\Lambda_b)}(\widetilde{Y}_b(P),N)\simeq\prod_{i\in L}N(\alpha_i) Hom Mod ∞ ( Λ b ) ( Y b ( P ) , N ) ≃ ∏ i ∈ L N ( α i ) . Hence
Hom D ∞ ( Λ b ) ( Y ~ b ( P ) , Y ~ b ( Q ) [ d ] ) ≅ ∏ i ∈ L H d ( Y ~ b ( Q ) ( α i ) ) = ∏ i ∈ L ∏ j ∈ K Hom Λ b ( α i , β j ) d = Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) d . \operatorname{Hom}_{D_{\infty}(\Lambda_b)}(\widetilde{Y}_b(P),\widetilde{Y}_b(Q)[d]) \cong\prod_{i\in L}H^d(\widetilde{Y}_b(Q)(\alpha_i))=\prod_{i\in L}\prod_{j\in K}\operatorname{Hom}_{\Lambda_b}(\alpha_i,\beta_j)^d=\operatorname{Hom}_{\Lambda_b^{(\leq\aleph_3)}}(P,Q)^d. Hom D ∞ ( Λ b ) ( Y b ( P ) , Y b ( Q ) [ d ]) ≅ i ∈ L ∏ H d ( Y b ( Q ) ( α i )) = i ∈ L ∏ j ∈ K ∏ Hom Λ b ( α i , β j ) d = Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) d .
The inverse isomorphism is induced by ( Y ~ b ) 1 (\widetilde{Y}_b)_1 ( Y b ) 1 and minimality:
Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) d → m 1 = 0 m 1 = 0 H d ( Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) ) \operatorname{Hom}_{\Lambda_b^{(\leq\aleph_3)}}(P,Q)^d \xrightarrow[\ m_1=0\ ]{\phantom{m_1=0}} H^d\left(\operatorname{Hom}_{\Lambda_b^{(\leq\aleph_3)}}(P,Q)\right) Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) d m 1 = 0 m 1 = 0 H d ( Hom Λ b ( ≤ ℵ 3 ) ( P , Q ) )
→ H d ( ( Y ~ b ) 1 ) H d ( Hom Mod ∞ ( Λ b ) ( Y ~ b ( P ) , Y ~ b ( Q ) ) ) → ∼ Hom D ∞ ( Λ b ) ( Y ~ b ( P ) , Y ~ b ( Q ) [ d ] ) . \xrightarrow{H^d((\widetilde{Y}_b)_1)} H^d\left(\operatorname{Hom}_{\operatorname{Mod}_{\infty}(\Lambda_b)}(\widetilde{Y}_b(P),\widetilde{Y}_b(Q))\right)\xrightarrow{\sim}\operatorname{Hom}_{D_{\infty}(\Lambda_b)}(\widetilde{Y}_b(P),\widetilde{Y}_b(Q)[d]). H d (( Y b ) 1 ) H d ( Hom Mod ∞ ( Λ b ) ( Y b ( P ) , Y b ( Q )) ) ∼ Hom D ∞ ( Λ b ) ( Y b ( P ) , Y b ( Q ) [ d ]) .
Thus Y ~ b \widetilde{Y}_b Y b is fully faithful on matrix objects and their shifts, i.e., P [ r ] P[r] P [ r ] ’s. Taking shifts and cones (by five lemma) proves full faithfulness on all finite twisted complexes. □
Let V b \mathcal{V}_b V b denote the image of Y ~ b \widetilde{Y}_b Y b . By Lemma 3.12, one has fully faithful inclusion of thick subcategories Perf ( Λ b ) ⊆ V b ⊆ D ∞ ( Λ b ) \operatorname{Perf}(\Lambda_b)\subseteq\mathcal{V}_b\subseteq D_{\infty}(\Lambda_b) Perf ( Λ b ) ⊆ V b ⊆ D ∞ ( Λ b ) , and all three categories are algebraic over R R R .
Note that each term ( P d ′ ) (P'_d) ( P d ′ ) in the projective resolution of R R R (Lemma 3.1 ) are in V b \mathcal{V}_b V b , by Lemma 3.12. The graded Hom groups are computed as
Hom V b ( P p , P q [ d ] ) ≅ { Hom C ( P p , P q ) , d = 0 , Hom C ( P p , P q ⊗ C E ) , d = − 2 , 0 , else . \operatorname{Hom}_{\mathcal{V}_b}(P_p,P_q[d])\cong
\begin{cases}
\operatorname{Hom}_{\mathcal{C}}(P_p,P_q), & d=0,\\
\operatorname{Hom}_{\mathcal{C}}(P_p,P_q\otimes_{\mathcal{C}}E), & d=-2,\\
0, & \text{else}.
\end{cases} Hom V b ( P p , P q [ d ]) ≅ ⎩ ⎨ ⎧ Hom C ( P p , P q ) , Hom C ( P p , P q ⊗ C E ) , 0 , d = 0 , d = − 2 , else .
Hence the full subcategory on P p P_p P p ’s is 2-split.
Remark 3.13. The computation for d = − 2 d=-2 d = − 2 is given by y β ⊗ C E ≅ E ( − , β ) y_\beta\otimes_{\mathcal{C}}E\cong E(-,\beta) y β ⊗ C E ≅ E ( − , β ) and coproducts.
Put I = ( d 1 ⊗ C E ) ∗ Hom C ( P 4 , P 1 ⊗ C E ) + d 4 ∗ Hom C ( P 3 , P 0 ⊗ C E ) ⊆ Hom C ( P 4 , P 0 ⊗ C E ) \mathcal{I}=(d_1\otimes_{\mathcal{C}}E)_*\operatorname{Hom}_{\mathcal{C}}(P_4,P_1\otimes_{\mathcal{C}}E)+d_4^*\operatorname{Hom}_{\mathcal{C}}(P_3,P_0\otimes_{\mathcal{C}}E)\subseteq\operatorname{Hom}_{\mathcal{C}}(P_4,P_0\otimes_{\mathcal{C}}E) I = ( d 1 ⊗ C E ) ∗ Hom C ( P 4 , P 1 ⊗ C E ) + d 4 ∗ Hom C ( P 3 , P 0 ⊗ C E ) ⊆ Hom C ( P 4 , P 0 ⊗ C E ) .
Proposition 3.14. With conventions of Definitions 2.12 and 2.14 , the Toda brackets are
⟨ d 1 , d 2 , d 3 , d 4 ⟩ V − = I , ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V + = − m 4 + ( d 1 , d 2 , d 3 , d 4 ) + I . \langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_{-}}=\mathcal{I},\qquad
\langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_{+}}=-m_4^{+}(d_1,d_2,d_3,d_4)+\mathcal{I}. ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V − = I , ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V + = − m 4 + ( d 1 , d 2 , d 3 , d 4 ) + I .
Proof. Consider P 4 → d 4 P 3 → d 3 P 2 → d 2 P 1 → d 1 P 0 P_4\xrightarrow{d_4}P_3\xrightarrow{d_3}P_2\xrightarrow{d_2}P_1\xrightarrow{d_1}P_0 P 4 d 4 P 3 d 3 P 2 d 2 P 1 d 1 P 0 . Since the terms belong to a 2-split subcategory, Proposition 2.15 gives a nonempty Toda coset with indeterminacy I \mathcal{I} I .
Fix b ∈ { + , − } b\in\{+,-\} b ∈ { + , − } . We work in Tw ( Λ b ( ≤ ℵ 3 ) ) \operatorname{Tw}(\Lambda_b^{(\leq\aleph_3)}) Tw ( Λ b ( ≤ ℵ 3 ) ) . By definition, ⟨ d 1 , d 2 , d 3 , d 4 ⟩ \langle d_1,d_2,d_3,d_4\rangle ⟨ d 1 , d 2 , d 3 , d 4 ⟩ consists of the morphism γ : P 4 [ 2 ] → P 0 \gamma:P_4[2]\to P_0 γ : P 4 [ 2 ] → P 0 for which there exist objects C , Y C,Y C , Y and morphisms satisfying the certain conditions. Take
C = ( P 1 ⊕ P 2 [ 1 ] , q C = ( 0 s d 2 0 0 ) ) , Y = ( P 1 ⊕ P 2 [ 1 ] ⊕ P 3 [ 2 ] , q Y = ( 0 s d 2 0 0 0 s d 3 0 0 0 ) ) . C=\left(P_1\oplus P_2[1],q_C=\begin{pmatrix}0&s d_2\\0&0\end{pmatrix}\right),\qquad
Y=\left(P_1\oplus P_2[1]\oplus P_3[2],q_Y=\begin{pmatrix}0&s d_2&0\\0&0&s d_3\\0&0&0\end{pmatrix}\right). C = ( P 1 ⊕ P 2 [ 1 ] , q C = ( 0 0 s d 2 0 ) ) , Y = P 1 ⊕ P 2 [ 1 ] ⊕ P 3 [ 2 ] , q Y = 0 0 0 s d 2 0 0 0 s d 3 0 .
The morphism γ \gamma γ is defined as the composite P 4 [ 2 ] → α Y → β P 0 P_4[2]\xrightarrow{\alpha}Y\xrightarrow{\beta}P_0 P 4 [ 2 ] α Y β P 0 , where α = ( 0 0 d 4 [ 2 ] ) \alpha=\begin{pmatrix}0\\0\\d_4[2]\end{pmatrix} α = 0 0 d 4 [ 2 ] and β = ( d 1 0 0 ) \beta=(d_1\ 0\ 0) β = ( d 1 0 0 ) .
(2) The following are distinguished triangles:
P 1 → u = ( 1 0 ) C → q = ( 0 1 ) P 2 [ 1 ] → d 2 [ 1 ] P 1 [ 1 ] , C → v = ( 1 0 0 1 0 0 ) Y → p = ( 0 0 1 ) P 3 [ 2 ] → δ = ( 0 d 3 [ 2 ] ) C [ 1 ] . P_1 \xrightarrow{u=\binom{1}{0}} C \xrightarrow{q=(0\ 1)} P_2[1] \xrightarrow{d_2[1]} P_1[1], \qquad C \xrightarrow{v=\left(\begin{smallmatrix}1&0\\0&1\\0&0\end{smallmatrix}\right)} Y \xrightarrow{p=(0\ 0\ 1)} P_3[2] \xrightarrow{\delta=\binom{0}{d_3[2]}} C[1]. P 1 u = ( 0 1 ) C q = ( 0 1 ) P 2 [ 1 ] d 2 [ 1 ] P 1 [ 1 ] , C v = ( 1 0 0 0 1 0 ) Y p = ( 0 0 1 ) P 3 [ 2 ] δ = ( d 3 [ 2 ] 0 ) C [ 1 ] .
(3) The following three diagrams commute:
P 3 [ 2 ] → δ C [ 1 ] d 3 [ 2 ] ↓ ↓ q [ 1 ] P 2 [ 2 ] = P 2 [ 2 ] P 4 [ 2 ] → α Y d 4 [ 2 ] ↓ ↓ p P 3 [ 2 ] = P 3 [ 2 ] P 1 → v ∘ u Y d 1 ↓ ↓ β P 0 = P 0 \begin{CD}
P_3[2] @>{\delta}>> C[1]\\
@V{d_3[2]}VV @VV{q[1]}V\\
P_2[2] @= P_2[2]
\end{CD}
\qquad
\begin{CD}
P_4[2] @>{\alpha}>> Y\\
@V{d_4[2]}VV @VV{p}V\\
P_3[2] @= P_3[2]
\end{CD}
\qquad
\begin{CD}
P_1 @>{v\circ u}>> Y\\
@V{d_1}VV @VV{\beta}V\\
P_0 @= P_0
\end{CD} P 3 [ 2 ] d 3 [ 2 ] ↓ ⏐ P 2 [ 2 ] δ C [ 1 ] ↓ ⏐ q [ 1 ] P 2 [ 2 ] P 4 [ 2 ] d 4 [ 2 ] ↓ ⏐ P 3 [ 2 ] α Y ↓ ⏐ p P 3 [ 2 ] P 1 d 1 ↓ ⏐ P 0 v ∘ u Y ↓ ⏐ β P 0
The above are easily checked. Hence we regard ⟨ d 1 , d 2 , d 3 , d 4 ⟩ \langle d_1,d_2,d_3,d_4\rangle ⟨ d 1 , d 2 , d 3 , d 4 ⟩ as a subset of Hom V b ( P 4 , P 0 [ − 2 ] ) \operatorname{Hom}_{\mathcal{V}_b}(P_4,P_0[-2]) Hom V b ( P 4 , P 0 [ − 2 ]) . It remains to show that − m 4 b ( d 1 , d 2 , d 3 , d 4 ) ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b -m_4^b(d_1,d_2,d_3,d_4)\in\langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_b} − m 4 b ( d 1 , d 2 , d 3 , d 4 ) ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b . Note that
[ γ ] = [ β ] ∘ [ α ] = [ m 2 T w ( β , α ) ] = s − 1 b 2 T w ( s β , s α ) = s − 1 ∑ j ≥ 0 b j + 2 ( s β , q Y ⊗ j , s α ) = s − 1 ( b 2 ( s β , s α ) + b 4 ( s β , q Y , q Y , s α ) ) = ⟨ 1 ⟩ s − 1 b 4 ( s d 1 , s d 2 , s d 3 , s ( d 4 [ 2 ] ) ) = − ( m 4 b ) Σ ( d 1 , d 2 , d 3 , d 4 [ 2 ] ) = ⟨ 2 ⟩ − m 4 b ( d 1 , d 2 , d 3 , d 4 [ 2 ] ) . \begin{aligned}
[\gamma]=[\beta]\circ[\alpha]=[m_2^{\mathrm{Tw}}(\beta,\alpha)]=s^{-1}b_2^{\mathrm{Tw}}(s\beta,s\alpha)=s^{-1}\sum_{j\geq0}b_{j+2}(s\beta,q_Y^{\otimes j},s\alpha) \\
&=s^{-1}\left(b_2(s\beta,s\alpha)+b_4(s\beta,q_Y,q_Y,s\alpha)\right)\mathrel{\overset{\langle1\rangle}{=} }s^{-1}b_4(sd_1,sd_2,sd_3,s(d_4[2])) \\
&=-\left(m_4^b\right)^\Sigma(d_1,d_2,d_3,d_4[2])\mathrel{\overset{\langle2\rangle}{=}}-m_4^b(d_1,d_2,d_3,d_4[2]).
\end{aligned} [ γ ] = [ β ] ∘ [ α ] = [ m 2 Tw ( β , α )] = s − 1 b 2 Tw ( s β , s α ) = s − 1 j ≥ 0 ∑ b j + 2 ( s β , q Y ⊗ j , s α ) = s − 1 ( b 2 ( s β , s α ) + b 4 ( s β , q Y , q Y , s α ) ) = ⟨ 1 ⟩ s − 1 b 4 ( s d 1 , s d 2 , s d 3 , s ( d 4 [ 2 ])) = − ( m 4 b ) Σ ( d 1 , d 2 , d 3 , d 4 [ 2 ]) = ⟨ 2 ⟩ − m 4 b ( d 1 , d 2 , d 3 , d 4 [ 2 ]) .
For ⟨ 1 ⟩ \langle1\rangle ⟨ 1 ⟩ , expanding the matrix gives b 2 ( s β , s α ) = b 2 ( s d 1 , 0 ) + b 2 ( 0 , 0 ) + b 2 ( 0 , s ( d 4 [ 2 ] ) ) = 0 b_2(s\beta,s\alpha)=b_2(sd_1,0)+b_2(0,0)+b_2(0,s(d_4[2]))=0 b 2 ( s β , s α ) = b 2 ( s d 1 , 0 ) + b 2 ( 0 , 0 ) + b 2 ( 0 , s ( d 4 [ 2 ])) = 0 and and the unique path contributing to the matrix expansion of b 4 ( s β , q Y , q Y , s α ) b_4(s\beta,q_Y,q_Y,s\alpha) b 4 ( s β , q Y , q Y , s α ) is [ s d 1 , s d 2 , s d 3 , s ( d 4 [ 2 ] ) ] [sd_1,sd_2,sd_3,s(d_4[2])] [ s d 1 , s d 2 , s d 3 , s ( d 4 [ 2 ])] . The Koszul rule for m Σ m^\Sigma m Σ in the Preliminaries gives ⟨ 2 ⟩ \langle2\rangle ⟨ 2 ⟩ .
Finally Lemma 3.12 and Remark 2.16 transport these Toda cosets to V b \mathcal{V}_b V b . Since m 4 − = 0 m_4^-=0 m 4 − = 0 , this proves the two asserted coset formulas.
Proposition 3.15. For every f ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b f\in\langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_b} f ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b , the composite ( d 0 ⊗ C E ) ∘ f : P 4 → P 0 ⊗ C E → R ⊗ C E = A (d_0\otimes_C E)\circ f:P_4\to P_0\otimes_C E\to R\otimes_C E=A ( d 0 ⊗ C E ) ∘ f : P 4 → P 0 ⊗ C E → R ⊗ C E = A is a cocycle and
[ ( d 0 ⊗ C E ) ∘ f ] = { 0 , b = − , − [ c ] ( ≠ 0 ) , b = + , in Ext C 4 ( R , A ) . [(d_0\otimes_C E)\circ f]=
\begin{cases}
0, & b=-,\\
-[c](\ne0), & b=+,
\end{cases}
\quad\text{in }\operatorname{Ext}^4_C(R,A). [( d 0 ⊗ C E ) ∘ f ] = { 0 , − [ c ] ( = 0 ) , b = − , b = + , in Ext C 4 ( R , A ) . Proof. Note that for ordinals α 0 < α 1 < α 2 < α 3 < α 4 < ω 3 \alpha_0<\alpha_1<\alpha_2<\alpha_3<\alpha_4<\omega_3 α 0 < α 1 < α 2 < α 3 < α 4 < ω 3 , one has
m 4 + ( u α 3 , α 4 , u α 2 , α 3 , u α 1 , α 2 , u α 0 , α 1 ) = c α 0 [ α 0 , … , α 4 ] . m_4^+(u_{\alpha_3,\alpha_4},u_{\alpha_2,\alpha_3},u_{\alpha_1,\alpha_2},u_{\alpha_0,\alpha_1})=c_{\alpha_0}[\alpha_0,\ldots,\alpha_4]. m 4 + ( u α 3 , α 4 , u α 2 , α 3 , u α 1 , α 2 , u α 0 , α 1 ) = c α 0 [ α 0 , … , α 4 ] . Indeed, since m 4 + ( … , 1 , … ) = 0 m_4^+(\ldots,1,\ldots)=0 m 4 + ( … , 1 , … ) = 0 , the only surviving path from σ = [ α 0 , … , α 4 ] \sigma=[\alpha_0,\ldots,\alpha_4] σ = [ α 0 , … , α 4 ] to ∅ \varnothing ∅ is
[ α 0 , … , α 4 ] → [ α 1 , … , α 4 ] → [ α 2 , α 3 , α 4 ] → [ α 3 , α 4 ] → [ α 4 ] → ∅ . [\alpha_0,\ldots,\alpha_4]\to[\alpha_1,\ldots,\alpha_4]\to[\alpha_2,\alpha_3,\alpha_4]\to[\alpha_3,\alpha_4]\to[\alpha_4]\to\varnothing. [ α 0 , … , α 4 ] → [ α 1 , … , α 4 ] → [ α 2 , α 3 , α 4 ] → [ α 3 , α 4 ] → [ α 4 ] → ∅ . Take the canonical inclusion ι α 4 : A ( α 0 ) → ( P 0 ⊗ C E ) ( α 0 ) ≅ ∐ β < ω 3 A ( α 0 ) \iota_{\alpha_4}:A(\alpha_0)\to(P_0\otimes_C E)(\alpha_0)\cong\coprod_{\beta<\omega_3}A(\alpha_0) ι α 4 : A ( α 0 ) → ( P 0 ⊗ C E ) ( α 0 ) ≅ ∐ β < ω 3 A ( α 0 ) . Then the following composite is identical
( d 0 ⊗ C E ) ∘ ι α 4 : A ( α 0 ) ⟶ ( P 0 ⊗ C E ) ( α 0 ) ⟶ ( R ⊗ C E ) ( α 0 ) = A ( α 0 ) . (d_0\otimes_C E)\circ\iota_{\alpha_4}:A(\alpha_0)\longrightarrow(P_0\otimes_C E)(\alpha_0)\longrightarrow(R\otimes_C E)(\alpha_0)=A(\alpha_0). ( d 0 ⊗ C E ) ∘ ι α 4 : A ( α 0 ) ⟶ ( P 0 ⊗ C E ) ( α 0 ) ⟶ ( R ⊗ C E ) ( α 0 ) = A ( α 0 ) . It follows that
( ( d 0 ⊗ C E ) ∘ ( m 4 + ( d 1 , d 2 , d 3 , d 4 ) ) ) α 0 ( σ ) = ( d 0 ⊗ C E ) α 0 ( ι α 4 ( c α 0 ( σ ) ) ) = c α 0 ( σ ) \left((d_0\otimes_C E)\circ\left(m_4^+(d_1,d_2,d_3,d_4)\right)\right)_{\alpha_0}(\sigma)=(d_0\otimes_C E)_{\alpha_0}\left(\iota_{\alpha_4}\left(c_{\alpha_0}(\sigma)\right)\right)=c_{\alpha_0}(\sigma) ( ( d 0 ⊗ C E ) ∘ ( m 4 + ( d 1 , d 2 , d 3 , d 4 ) ) ) α 0 ( σ ) = ( d 0 ⊗ C E ) α 0 ( ι α 4 ( c α 0 ( σ ) ) ) = c α 0 ( σ ) Since the above simplex summands generate P 4 P_4 P 4 , we obtain
( d 0 ⊗ C E ) ∘ m 4 + ( d 1 , d 2 , d 3 , d 4 ) = c , c ∘ d 5 = 0. (d_0\otimes_C E)\circ m_4^+(d_1,d_2,d_3,d_4)=c,\qquad c\circ d_5=0. ( d 0 ⊗ C E ) ∘ m 4 + ( d 1 , d 2 , d 3 , d 4 ) = c , c ∘ d 5 = 0. Thus the composite give the classes 0 0 0 and − [ c ] -[c] − [ c ] , respectively.
We then show the cohomology class is independent of the representative in Toda bracket. Take another f ′ ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b f'\in\langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_b} f ′ ∈ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V b . Then by the construction of I \mathcal{I} I in Proposition 3.14, take f − f ′ = ( d 1 ⊗ C E ) a + v d 4 f-f'=(d_1\otimes_C E)a+vd_4 f − f ′ = ( d 1 ⊗ C E ) a + v d 4 with a : P 4 → P 1 ⊗ C E a:P_4\to P_1\otimes_C E a : P 4 → P 1 ⊗ C E and v : P 3 → P 0 ⊗ C E v:P_3\to P_0\otimes_C E v : P 3 → P 0 ⊗ C E . Thus
( d 0 ⊗ C E ) ∘ ( ( d 1 ⊗ C E ) ∘ a + v ∘ d 4 ) = ( ( d 0 ⊗ C E ) ∘ v ) ∘ d 4 . (d_0\otimes_C E)\circ\left((d_1\otimes_C E)\circ a+v\circ d_4\right)=\left((d_0\otimes_C E)\circ v\right)\circ d_4. ( d 0 ⊗ C E ) ∘ ( ( d 1 ⊗ C E ) ∘ a + v ∘ d 4 ) = ( ( d 0 ⊗ C E ) ∘ v ) ∘ d 4 .
The right-hand side is a degree-four boundary in Hom C ( P ∙ , A ) \operatorname{Hom}_{\mathcal C}(\mathbf P_\bullet,A) Hom C ( P ∙ , A ) , thus is killed by [ − ] [-] [ − ] .
Finally, [ c ] ≠ 0 [c]\ne0 [ c ] = 0 follows from Lemma 3.3 . □
Construct T \mathcal T T from V ± V_\pm V ±
We retain the cardinal θ \theta θ in the outline and put
I = θ × { + , − } , Sign ( ξ , + ) = + , Sign ( ξ , − ) = − . I=\theta\times\{+,-\},\qquad\operatorname{Sign}(\xi,+)=+,\qquad\operatorname{Sign}(\xi,-)=-. I = θ × { + , − } , Sign ( ξ , + ) = + , Sign ( ξ , − ) = − .
We construct T \mathcal T T by adjoining triangulated categories T J \mathcal T_J T J ’s with a common perfect core:
T J = ( ∏ i ∈ J V Sign ( i ) ) × Perf ( Λ − ( I ∖ J ) ) , ∣ J ∣ < ℵ 0 (8) \mathcal T_J=\left(\prod_{i\in J}\mathcal V_{\operatorname{Sign}(i)}\right)\times\operatorname{Perf}(\Lambda_-(I\setminus J)),\qquad|J|<\aleph_0 \tag*{(8)} T J = ( i ∈ J ∏ V Sign ( i ) ) × Perf ( Λ − ( I ∖ J )) , ∣ J ∣ < ℵ 0 (8)
For V ∈ Mod R V\in\operatorname{Mod}_{R} V ∈ Mod R , define the set of “almost constant” maps as follows
L ( V ) = { f : I → V ∣ ∣ f ( I ) ∣ < ℵ 0 , ∃ S ⊆ I , ∣ S ∣ < θ , f ∣ I ∖ S is constant } . \mathrm L(V)=\{f:I\to V\mid|f(I)|<\aleph_0,\ \exists S\subseteq I,\ |S|<\theta,\ f|_{I\setminus S}\text{ is constant}\}. L ( V ) = { f : I → V ∣ ∣ f ( I ) ∣ < ℵ 0 , ∃ S ⊆ I , ∣ S ∣ < θ , f ∣ I ∖ S is constant } .
A subset U ⊆ I U\subseteq I U ⊆ I is admissible if ∣ U ∣ < θ |U|<\theta ∣ U ∣ < θ or ∣ I ∖ U ∣ < θ |I\setminus U|<\theta ∣ I ∖ U ∣ < θ . For admissible U ⊆ I U\subseteq I U ⊆ I , define
L U ( V ) = { f ∣ U ∣ f ∈ L ( V ) } . \mathrm L_U(V)=\{f|_U\mid f\in\mathrm L(V)\}. L U ( V ) = { f ∣ U ∣ f ∈ L ( V )} .
We write ev i \operatorname{ev}_i ev i for evaluation at i ∈ I i\in I i ∈ I . For functions in L ( V ) \mathrm L(V) L ( V ) , we write ev ∞ \operatorname{ev}_\infty ev ∞ for their unique common value outside a subset of cardinality less than θ \theta θ .
Lemma 3.16. Let U ⊆ I U\subseteq I U ⊆ I be admissible and let V V V be an R R R -module.
(1) Pointwise operations make L U ( V ) \mathrm L_U(V) L U ( V ) an R R R -module.
(2) The following canonical map is an isomorphism
L U ( R ) ⊗ R V ⟶ L U ( V ) , a ⊗ v ⟼ ( i ↦ a ( i ) v ) . \mathrm L_U(R)\otimes_R V\longrightarrow\mathrm L_U(V),\qquad a\otimes v\longmapsto(i\mapsto a(i)v). L U ( R ) ⊗ R V ⟶ L U ( V ) , a ⊗ v ⟼ ( i ↦ a ( i ) v ) . (3) A finite partition U = ∐ r = 1 n U r U=\coprod_{r=1}^{n}U_r U = ∐ r = 1 n U r of admissible subsets induces an isomorphism
L U ( V ) → ∼ ⨁ r = 1 n L U r ( V ) , f ⟼ ( f ∣ U 1 , f ∣ U 2 , … , f ∣ U n ) . \mathrm L_U(V)\xrightarrow{\sim}\bigoplus_{r=1}^{n}\mathrm L_{U_r}(V),\qquad f\longmapsto(f|_{U_1},f|_{U_2},\ldots,f|_{U_n}). L U ( V ) ∼ r = 1 ⨁ n L U r ( V ) , f ⟼ ( f ∣ U 1 , f ∣ U 2 , … , f ∣ U n ) .
Proof. Straightforward. □
Definition 3.17. For admissible U ⊆ I U\subseteq I U ⊆ I , let Λ − ( U ) \Lambda_-(U) Λ − ( U ) be the dg category with Ob = ω 3 \operatorname{Ob}=\omega_3 Ob = ω 3 and
Hom Λ − ( U ) ( α , β ) d = L U ( Hom B ( α , β ) d ) . \operatorname{Hom}_{\Lambda_-(U)}(\alpha,\beta)^d=\mathrm L_U\bigl(\operatorname{Hom}_{\mathcal B}(\alpha,\beta)^d\bigr). Hom Λ − ( U ) ( α , β ) d = L U ( Hom B ( α , β ) d ) . Its differential is zero and its composition is pointwise. Evaluation and restriction act on coefficients and leave the object labels unchanged.
Lemma 3.18. A finite partition U = ∐ r = 1 n U r U=\coprod_{r=1}^{n}U_r U = ∐ r = 1 n U r of admissible subsets induces a triangle equivalence by restriction
Perf ( Λ − ( U ) ) → ∼ ∏ r = 1 n Perf ( Λ − ( U r ) ) . \operatorname{Perf}(\Lambda_-(U))\xrightarrow{\sim}\prod_{r=1}^{n}\operatorname{Perf}(\Lambda_-(U_r)). Perf ( Λ − ( U )) ∼ r = 1 ∏ n Perf ( Λ − ( U r )) . Proof. By Lemma 3.16 , the morphism modules split according to the central idempotents 1 U r 1_{U_r} 1 U r . After passing to perfect categories, these idempotents split objects into their components on the parts U r U_r U r . This gives the asserted equivalence. □
We identify Perf ( Λ − ) → ∼ Perf ( Λ + ) \operatorname{Perf}(\Lambda_-)\xrightarrow{\sim}\operatorname{Perf}(\Lambda_+) Perf ( Λ − ) ∼ Perf ( Λ + ) and take the canonical fully faithful triangle functor ι ± : Perf ( Λ ± ) ↪ V ± \iota_\pm:\operatorname{Perf}(\Lambda_\pm)\hookrightarrow\mathcal V_\pm ι ± : Perf ( Λ ± ) ↪ V ± . For each finite subset J ⊆ I J\subseteq I J ⊆ I , define
T J = ( ∏ i ∈ J V Sign ( i ) ) × Perf ( Λ − ( I ∖ J ) ) . \mathcal T_J=\left(\prod_{i\in J}\mathcal V_{\operatorname{Sign}(i)}\right)\times\operatorname{Perf}(\Lambda_-(I\setminus J)). T J = ( i ∈ J ∏ V Sign ( i ) ) × Perf ( Λ − ( I ∖ J )) .
Proposition 3.19. The categories T J \mathcal{T}_J T J , indexed by finite subsets of I I I , form a directed system of fully faithful triangle functors with thick images.
Proof. For finite subsets J J J and K K K with J ⊆ K ⊆ I J \subseteq K \subseteq I J ⊆ K ⊆ I , applying Lemma 3.18 to the partition
I ∖ J = ∐ k ∈ K ∖ J { k } ⨿ ( I ∖ K ) I \setminus J = \coprod_{k\in K\setminus J}\{k\} \amalg(I \setminus K) I ∖ J = k ∈ K ∖ J ∐ { k } ⨿ ( I ∖ K ) gives an equivalence
Perf ( Λ − ( I ∖ J ) ) ≃ ( ∏ k ∈ K ∖ J Perf ( Λ − ) ) × Perf ( Λ − ( I ∖ K ) ) . \operatorname{Perf}(\Lambda_-(I \setminus J)) \simeq\left(\prod_{k\in K\setminus J} \operatorname{Perf}(\Lambda_-)\right) \times\operatorname{Perf}(\Lambda_-(I \setminus K)). Perf ( Λ − ( I ∖ J )) ≃ k ∈ K ∖ J ∏ Perf ( Λ − ) × Perf ( Λ − ( I ∖ K )) . Then use ι Sign ( k ) \iota_{\operatorname{Sign}(k)} ι Sign ( k ) to get the transition functor Φ J , K : T J → T K \Phi_{J,K}:\mathcal{T}_J\to\mathcal{T}_K Φ J , K : T J → T K . Explicitly,
Φ J , K ( ( X i ) i ∈ J , P ) = ( ( X i ) i ∈ J , ( ι Sign ( k ) ( ev k P ) ) k ∈ K ∖ J , P ∣ I ∖ K ) . \Phi_{J,K}((X_i)_{i\in J},P)=\left((X_i)_{i\in J},\left(\iota_{\operatorname{Sign}(k)}(\operatorname{ev}_k P)\right)_{k\in K\setminus J},P|_{I\setminus K}\right). Φ J , K (( X i ) i ∈ J , P ) = ( ( X i ) i ∈ J , ( ι Sign ( k ) ( ev k P ) ) k ∈ K ∖ J , P ∣ I ∖ K ) . Φ J , K \Phi_{J,K} Φ J , K is fully faithful and has thick image, since these properties hold for ι ± \iota_\pm ι ± .
For finite subsets J J J , K K K and L L L with J ⊆ K ⊆ L J\subseteq K\subseteq L J ⊆ K ⊆ L , one has Φ K , L Φ J , K = Φ J , L \Phi_{K,L}\Phi_{J,K}=\Phi_{J,L} Φ K , L Φ J , K = Φ J , L . This equality should be an isomorphism; for convenience we identify the finite stages with their images in a common union model, so that the transition functors are inclusions.
The system is clearly directed. □ \square □
For infinite subset J ⊆ I J\subseteq I J ⊆ I , put T J = l i m → K ⊆ J , ∣ K ∣ < ℵ 0 T K \mathcal{T}_J=\varinjlim_{K\subseteq J,\ |K|<\aleph_0}\mathcal{T}_K T J = lim K ⊆ J , ∣ K ∣ < ℵ 0 T K . Write T = T I \mathcal{T}=\mathcal{T}_I T = T I .
Subcategories generated by fewer than θ \theta θ objects We show that every thick subcategory of T \mathcal{T} T generated by fewer than θ \theta θ objects is algebraic over R R R ; equivalently, T J \mathcal{T}_J T J is algebraic for ∣ J ∣ < θ |J|<\theta ∣ J ∣ < θ . We fix such J J J , and let e J : I → R e_J:I\to R e J : I → R be map in L ( R ) \mathrm{L}(R) L ( R ) by setting
e J ( i ) = { 1 , i ∈ J , Sign ( i ) = + , 0 , otherwise . e_J(i)=
\begin{cases}
1, & i\in J,\ \operatorname{Sign}(i)=+,\\
0, & \text{otherwise}.
\end{cases} e J ( i ) = { 1 , 0 , i ∈ J , Sign ( i ) = + , otherwise . Lemma 3.20. There is an A ∞ A_\infty A ∞ category Λ J \Lambda_J Λ J whose underlying graded category and m ≠ 4 m_{\ne4} m = 4 coincide with the dg category Λ − ( I ) \Lambda_-(I) Λ − ( I ) , and
α 0 → f 1 α 1 → f 2 α 2 → f 3 α 3 → f 4 α 4 \alpha_0 \xrightarrow{f_1} \alpha_1 \xrightarrow{f_2} \alpha_2 \xrightarrow{f_3} \alpha_3 \xrightarrow{f_4} \alpha_4 α 0 f 1 α 1 f 2 α 2 f 3 α 3 f 4 α 4 gives that
m 4 ( f 4 , f 3 , f 2 , f 1 ) ( i ) = e J ( i ) ⋅ m 4 + ( f 4 ( i ) , f 3 ( i ) , f 2 ( i ) , f 1 ( i ) ) . m_4(f_4,f_3,f_2,f_1)(i)=e_J(i)\cdot m_4^+(f_4(i),f_3(i),f_2(i),f_1(i)). m 4 ( f 4 , f 3 , f 2 , f 1 ) ( i ) = e J ( i ) ⋅ m 4 + ( f 4 ( i ) , f 3 ( i ) , f 2 ( i ) , f 1 ( i )) . Proof. By construction, m 4 ( f 4 , f 3 , f 2 , f 1 ) ∈ L ( E ( α 0 , α 4 ) ) = L ( A ( α 0 ) ) m_4(f_4,f_3,f_2,f_1)\in\mathrm{L}(\mathrm{E}(\alpha_0,\alpha_4))=\mathrm{L}(\mathrm{A}(\alpha_0)) m 4 ( f 4 , f 3 , f 2 , f 1 ) ∈ L ( E ( α 0 , α 4 )) = L ( A ( α 0 )) .
The Stasheff identities are checked pointwise on i ∈ I i\in I i ∈ I , so they hold by Lemma 3.5 . □ \square □
Lemma 3.21. There are triangle equivalences which are natural in admissible subset U ⊆ I U\subseteq I U ⊆ I and ∣ J ∣ < θ |J|<\theta ∣ J ∣ < θ , denoted by Ψ J , U : Perf ( Λ − ( U ) ) → ∼ Perf ( Λ J ( U ) ) \Psi_{J,U}:\operatorname{Perf}(\Lambda_-(U))\xrightarrow{\sim}\operatorname{Perf}(\Lambda_J(U)) Ψ J , U : Perf ( Λ − ( U )) ∼ Perf ( Λ J ( U )) .
Proof. Take U + = U ∩ J ∩ Sign − 1 ( + ) U_+=U\cap J\cap\operatorname{Sign}^{-1}(+) U + = U ∩ J ∩ Sign − 1 ( + ) and U − = U ∖ U + U_-=U\setminus U_+ U − = U ∖ U + . Both parts are admissible. Clearly
Perf ( Λ − ( U ) ) → ∼ Perf ( Λ − ( U + ) ) × Perf ( Λ − ( U − ) ) → ∼ Perf ( Λ J ( U + ) ) × Perf ( Λ J ( U − ) ) → ∼ Perf ( Λ J ( U ) ) \operatorname{Perf}(\Lambda_-(U))\xrightarrow{\sim}\operatorname{Perf}(\Lambda_-(U_+))\times\operatorname{Perf}(\Lambda_-(U_-))\xrightarrow{\sim}\operatorname{Perf}(\Lambda_J(U_+))\times\operatorname{Perf}(\Lambda_J(U_-))\xrightarrow{\sim}\operatorname{Perf}(\Lambda_J(U)) Perf ( Λ − ( U )) ∼ Perf ( Λ − ( U + )) × Perf ( Λ − ( U − )) ∼ Perf ( Λ J ( U + )) × Perf ( Λ J ( U − )) ∼ Perf ( Λ J ( U )) are composite of isomorphisms natural in both arguments. □ \square □
The derived category D ∞ ( Λ J ) D_\infty(\Lambda_J) D ∞ ( Λ J ) admits an R R R -linear dg enhancement [1 §13.35] and [7 Theorem 1.6.2] . We then construct a fully faithful triangle inclusion T J ↪ D ∞ ( Λ J ) \mathcal{T}_J\hookrightarrow D_\infty(\Lambda_J) T J ↪ D ∞ ( Λ J ) .
Proposition 3.22. For J ⊆ I J\subseteq I J ⊆ I (∣ J ∣ < θ |J|<\theta ∣ J ∣ < θ ), there is an R R R -linear fully faithful triangle functor
Φ J : T J ⟶ D ∞ ( Λ J ) . \Phi_J:\mathcal{T}_J\longrightarrow D_\infty(\Lambda_J). Φ J : T J ⟶ D ∞ ( Λ J ) .
Proof. For admissible U ⊆ I U \subseteq I U ⊆ I , take the restriction A ∞ A_{\infty} A ∞ functor r U : Λ J → Λ J ( U ) r_U \colon\Lambda_J \to\Lambda_J(U) r U : Λ J → Λ J ( U ) which is identical on objects, and whose components are
( r U ) 1 : ⨁ Hom B ( α , β ) d ⟶ ⨁ Hom B ( α , β ) d , f ⟼ f ∣ U ; ( r U ) ≥ 2 = 0. (r_U)_1 \colon\bigoplus\operatorname{Hom}_{\mathcal{B}}(\alpha,\beta)^{d} \longrightarrow\bigoplus\operatorname{Hom}_{\mathcal{B}}(\alpha,\beta)^{d}, \qquad f \longmapsto f|_U; \qquad(r_U)_{\geq2}=0. ( r U ) 1 : ⨁ Hom B ( α , β ) d ⟶ ⨁ Hom B ( α , β ) d , f ⟼ f ∣ U ; ( r U ) ≥ 2 = 0.
Note that the Stasheff identities are pointwise checked, in particular
( m 4 Λ J ( f 4 , … , f 1 ) ∣ U ) ( u ) = e J ( u ) m 4 + ( f 4 ( u ) , … , f 1 ( u ) ) = m 4 Λ J ( U ) ( f 4 ∣ U , … , f 1 ∣ U ) ( u ) . \left(m^{\Lambda_J}_4(f_4,\ldots,f_1)|_U\right)(u)=e_J(u)m^+_4(f_4(u),\ldots,f_1(u))=m^{\Lambda_J(U)}_4(f_4|_U,\ldots,f_1|_U)(u). ( m 4 Λ J ( f 4 , … , f 1 ) ∣ U ) ( u ) = e J ( u ) m 4 + ( f 4 ( u ) , … , f 1 ( u )) = m 4 Λ J ( U ) ( f 4 ∣ U , … , f 1 ∣ U ) ( u ) .
We define Φ F : T F → D ∞ ( Λ J ) \Phi_F \colon\mathcal{T}_F \to D_{\infty}(\Lambda_J) Φ F : T F → D ∞ ( Λ J ) for finite subset F ⊆ J F \subseteq J F ⊆ J first. For U = I ∖ F U=I\setminus F U = I ∖ F , one has
T F = ( ∏ i ∈ F V Sign ( i ) ) × Perf ( Λ − ( U ) ) . \mathcal{T}_F=\left(\prod_{i\in F}\mathcal{V}_{\operatorname{Sign}(i)}\right)\times\operatorname{Perf}(\Lambda_-(U)). T F = ( i ∈ F ∏ V Sign ( i ) ) × Perf ( Λ − ( U )) .
For each i ∈ F ⊆ J i\in F\subseteq J i ∈ F ⊆ J , take V Sign ( i ) ↪ D ∞ ( Λ Sign ( i ) ) = D ∞ ( Λ J ( { i } ) ) → ( r { i } ) ∗ D ∞ ( Λ J ) \mathcal{V}_{\operatorname{Sign}(i)}\hookrightarrow D_{\infty}(\Lambda_{\operatorname{Sign}(i)})=D_{\infty}(\Lambda_J(\{i\}))\xrightarrow{(r_{\{i\}})_*}D_{\infty}(\Lambda_J) V Sign ( i ) ↪ D ∞ ( Λ Sign ( i ) ) = D ∞ ( Λ J ({ i })) ( r { i } ) ∗ D ∞ ( Λ J ) .
For Perf ( Λ − ( U ) ) \operatorname{Perf}(\Lambda_-(U)) Perf ( Λ − ( U )) , take Perf ( Λ − ( U ) ) → Ψ J , U ∼ Perf ( Λ J ( U ) ) ↪ D ∞ ( Λ J ( U ) ) → ( r U ) ∗ D ∞ ( Λ J ) \operatorname{Perf}(\Lambda_-(U))\xrightarrow{\Psi_{J,U}\sim}\operatorname{Perf}(\Lambda_J(U))\hookrightarrow D_{\infty}(\Lambda_J(U))\xrightarrow{(r_U)_*}D_{\infty}(\Lambda_J) Perf ( Λ − ( U )) Ψ J , U ∼ Perf ( Λ J ( U )) ↪ D ∞ ( Λ J ( U )) ( r U ) ∗ D ∞ ( Λ J ) .
The fully faithful functors above yield another full faithful triangle functor
Φ F : T F ⟶ D ∞ ( Λ J ) , ( ( X i ) i ∈ F , P ) ⟼ ( ⨁ i ∈ F r { i } ∗ X i ) ⊕ ( r U ) ∗ Ψ J , U ( P ) . \Phi_F\colon\mathcal{T}_F\longrightarrow D_{\infty}(\Lambda_J),\qquad((X_i)_{i\in F},P)\longmapsto\left(\bigoplus_{i\in F}r_{\{i\}}^*X_i\right)\oplus(r_U)^*\Psi_{J,U}(P). Φ F : T F ⟶ D ∞ ( Λ J ) , (( X i ) i ∈ F , P ) ⟼ ( i ∈ F ⨁ r { i } ∗ X i ) ⊕ ( r U ) ∗ Ψ J , U ( P ) .
Let F ⊆ G ⊆ J F\subseteq G\subseteq J F ⊆ G ⊆ J be finite, then I ∖ F = ( G ∖ F ) ⊔ ( I ∖ G ) I\setminus F=(G\setminus F)\sqcup(I\setminus G) I ∖ F = ( G ∖ F ) ⊔ ( I ∖ G ) and thus yields a natural isomorphism Perf ( Λ − ( I ∖ F ) ) ≃ Perf ( Λ − ( G ∖ F ) ) × Perf ( Λ − ( I ∖ G ) ) \operatorname{Perf}(\Lambda_-(I\setminus F))\simeq\operatorname{Perf}(\Lambda_-(G\setminus F))\times\operatorname{Perf}(\Lambda_-(I\setminus G)) Perf ( Λ − ( I ∖ F )) ≃ Perf ( Λ − ( G ∖ F )) × Perf ( Λ − ( I ∖ G )) . It is straightforward to check Φ G Φ F , G ≃ Φ F \Phi_G\Phi_{F,G}\simeq\Phi_F Φ G Φ F , G ≃ Φ F . These Φ F \Phi_F Φ F 's therefore induce a triangle functor
Φ J : T J = lim ⟶ F ⊆ J , ∣ F ∣ < ℵ 0 T F ⟶ D ∞ ( Λ J ) . \Phi_J\colon\mathcal{T}_J=\underset{\substack{F\subseteq J,\ |F|<\aleph_0}}{\lim\limits_{\longrightarrow}}\mathcal{T}_F\longrightarrow D_{\infty}(\Lambda_J). Φ J : T J = F ⊆ J , ∣ F ∣ < ℵ 0 ⟶ lim T F ⟶ D ∞ ( Λ J ) .
Any two objects lie in a common finite stage, where the functor is fully faithful. Hence Φ J \Phi_J Φ J is fully faithful. □ \square □
Almost constancy of characteristic values
We construct a map χ : H c a t 4 ( F , E ∣ F ) → Ext C 4 ( R , A ) \chi\colon H^4_{\mathrm{cat}}(\mathcal{F},\mathcal{E}|_{\mathcal{F}})\to\operatorname{Ext}^4_C(R,A) χ : H cat 4 ( F , E ∣ F ) → Ext C 4 ( R , A ) from the category cohomology to Ext C 4 ( R , A ) \operatorname{Ext}^4_C(R,A) Ext C 4 ( R , A ) , and show the map χ ( ( ev ( ⋅ ) ) ∗ z ) : I → Ext C 4 ( R , A ) \chi((\operatorname{ev}(\cdot))^*z)\colon I\to\operatorname{Ext}^4_C(R,A) χ (( ev ( ⋅ ) ) ∗ z ) : I → Ext C 4 ( R , A ) is constant except for a subset S ⊆ I S\subseteq I S ⊆ I with cardinality ∣ S ∣ < ∣ I ∣ |S|<|I| ∣ S ∣ < ∣ I ∣ .
Let F \mathcal{F} F be a skeleton consisting of finite direct sums of representable functors (y β y_\beta y β 's).
Let P \mathcal{P} P be an arbitrary fixed small full additive category of projective right C \mathcal{C} C -modules containing F \mathcal{F} F and every P p P_p P p in the resolution of R R R . Define the bifunctor
E : Mod C o p × Mod C ⟶ Mod R , ( P , Q ) ⟼ Hom C ( P , Q ⊗ C E ) , E ( y α , y β ) ≅ E ( α , β ) . \mathcal{E}\colon\operatorname{Mod}_{\mathcal{C}}^{\mathrm{op}}\times\operatorname{Mod}_{\mathcal{C}}\longrightarrow\operatorname{Mod}_R,\qquad(P,Q)\longmapsto\operatorname{Hom}_{\mathcal{C}}(P,Q\otimes_{\mathcal{C}}E),\qquad\mathcal{E}(y_\alpha,y_\beta)\cong E(\alpha,\beta). E : Mod C op × Mod C ⟶ Mod R , ( P , Q ) ⟼ Hom C ( P , Q ⊗ C E ) , E ( y α , y β ) ≅ E ( α , β ) .
Lemma 3.23. Let S \mathcal{S} S be an arbitrary small category of projective right C \mathcal{C} C -modules (e.g. F \mathcal{F} F or P \mathcal{P} P ). Take the inclusion I S : S → Mod C I_{\mathcal{S}}\colon\mathcal{S}\to\operatorname{Mod}_{\mathcal{C}} I S : S → Mod C . Then
H c a t ( S , E ∣ S ) ≅ Ext Fun ( S , Mod C ) ∙ ( I S , I S ⊗ C E ) . H^{}_{\mathrm{cat}}(\mathcal{S},\mathcal{E}|_{\mathcal{S}})\cong\operatorname{Ext}^{\bullet}_{\operatorname{Fun}(\mathcal{S},\operatorname{Mod}_{\mathcal{C}})}(I_{\mathcal{S}},I_{\mathcal{S}}\otimes_{\mathcal{C}}E). H cat ( S , E ∣ S ) ≅ Ext Fun ( S , Mod C ) ∙ ( I S , I S ⊗ C E ) .
Proof. The projective resolution of I S I_{\mathcal{S}} I S in the functor category A = Fun ( S , Mod C ) \mathcal{A}=\operatorname{Fun}(\mathcal{S},\operatorname{Mod}_{\mathcal{C}}) A = Fun ( S , Mod C ) is
⋯ ⟶ ∏ X 0 → X 1 → X 2 X 0 Hom S ( X 2 , − ) ⟶ ∏ X 0 → X 1 X 0 Hom S ( X 1 , − ) ⟶ ∏ X 0 ∈ S X 0 Hom S ( X 0 , − ) ⟶ I S ⟶ 0. \cdots\longrightarrow\prod_{X_0\to X_1\to X_2}X_0^{\operatorname{Hom}_{\mathcal{S}}(X_2,-)}
\longrightarrow\prod_{X_0\to X_1}X_0^{\operatorname{Hom}_{\mathcal{S}}(X_1,-)}
\longrightarrow\prod_{X_0\in\mathcal{S}}X_0^{\operatorname{Hom}_{\mathcal{S}}(X_0,-)}
\longrightarrow I_{\mathcal{S}}\longrightarrow0. ⋯ ⟶ X 0 → X 1 → X 2 ∏ X 0 Hom S ( X 2 , − ) ⟶ X 0 → X 1 ∏ X 0 Hom S ( X 1 , − ) ⟶ X 0 ∈ S ∏ X 0 Hom S ( X 0 , − ) ⟶ I S ⟶ 0.
Applying Hom A ( − , I S ⊗ C E ) \operatorname{Hom}_{\mathcal A}(-, I_S \otimes_{\mathcal C} E) Hom A ( − , I S ⊗ C E ) gives a cochain. The adjunction of universal coproduct and evaluation, ( P ↦ P Hom S ( Q , − ) ) ⊣ ( H ↦ H ( Q ) ) (P \mapsto P^{\operatorname{Hom}_{\mathcal S}(Q,-)}) \dashv(H \mapsto H(Q)) ( P ↦ P Hom S ( Q , − ) ) ⊣ ( H ↦ H ( Q )) , gives the natural isomorphisms
Hom A ( ∏ X 0 → ⋯ → X n X 0 Hom S ( X n , − ) , I S ⊗ C E ) ≅ ∏ X 0 → ⋯ → X n Hom A ( X 0 Hom S ( X n , − ) , I S ⊗ C E ) \operatorname{Hom}_{\mathcal A}\left(\prod_{X_0\to\cdots\to X_n} X_0^{\operatorname{Hom}_{\mathcal S}(X_n,-)}, I_S \otimes_{\mathcal C} E\right) \cong\prod_{X_0\to\cdots\to X_n} \operatorname{Hom}_{\mathcal A}\left(X_0^{\operatorname{Hom}_{\mathcal S}(X_n,-)}, I_S \otimes_{\mathcal C} E\right) Hom A ( X 0 → ⋯ → X n ∏ X 0 Hom S ( X n , − ) , I S ⊗ C E ) ≅ X 0 → ⋯ → X n ∏ Hom A ( X 0 Hom S ( X n , − ) , I S ⊗ C E )
≅ ∏ X 0 → ⋯ → X n Hom C ( X 0 , X n ⊗ C E ) = C c a t n ( S , E ∣ S ) ( n ≥ 0 ) . \cong\prod_{X_0\to\cdots\to X_n} \operatorname{Hom}_{\mathcal C}(X_0, X_n \otimes_{\mathcal C} E) = C_{\mathrm{cat}}^n(\mathcal S, E|_{\mathcal S}) \qquad(n \ge0). ≅ X 0 → ⋯ → X n ∏ Hom C ( X 0 , X n ⊗ C E ) = C cat n ( S , E ∣ S ) ( n ≥ 0 ) .
Taking fourth cohomology yields the desired isomorphism. □
Lemma 3.24. One has natural isomorphism H c a t 4 ( P , E ∣ P ) ≅ H c a t 4 ( F , E ∣ F ) H_{\mathrm{cat}}^4(\mathcal P, E|_{\mathcal P}) \cong H_{\mathrm{cat}}^4(\mathcal F, E|_{\mathcal F}) H cat 4 ( P , E ∣ P ) ≅ H cat 4 ( F , E ∣ F ) .
Proof. Let Q ∙ → I F Q_\bullet\to I_{\mathcal F} Q ∙ → I F be the projective resolution of I F I_{\mathcal F} I F in Fun ( F , Mod C ) \operatorname{Fun}(\mathcal F,\operatorname{Mod}_{\mathcal C}) Fun ( F , Mod C ) . Put k : F ⊆ P k:\mathcal F \subseteq\mathcal P k : F ⊆ P . Then for each X ∈ P X \in\mathcal P X ∈ P , the left Kan extension gives
( Lan k I F ) ( X ) = lim ( F → X ) ∈ F / X F ≅ X , (\operatorname{Lan}_k I_{\mathcal F})(X)=\lim_{\substack{(F\to X)\in\mathcal F/X}} F \cong X, ( Lan k I F ) ( X ) = ( F → X ) ∈ F / X lim F ≅ X , i.e., Lan k I F ≅ I P \operatorname{Lan}_k I_{\mathcal F}\cong I_{\mathcal P} Lan k I F ≅ I P .
Here for X ∈ F X \in\mathcal F X ∈ F , the colimit is filtered, thus is exact. Since Lan k \operatorname{Lan}_k Lan k is exact and has an exact right adjoint, it preserves projective resolutions. Hence Lan k Q ∙ \operatorname{Lan}_k Q_\bullet Lan k Q ∙ is a projective resolution of I P I_{\mathcal P} I P . Along with Lemma 3.23, one has natural isomorphisms
H c a t 4 ( P , E ∣ P ) ≅ Ext Fun ( P , Mod C ) 4 ( I P , I P ⊗ C E ) = H 4 Hom Fun ( P , Mod C ) ( Lan k Q ∙ , I P ⊗ C E ) H_{\mathrm{cat}}^4(\mathcal P,E|_{\mathcal P})\cong\operatorname{Ext}_{\operatorname{Fun}(\mathcal P,\operatorname{Mod}_{\mathcal C})}^4(I_{\mathcal P},I_{\mathcal P}\otimes_{\mathcal C}E)=H^4\operatorname{Hom}_{\operatorname{Fun}(\mathcal P,\operatorname{Mod}_{\mathcal C})}(\operatorname{Lan}_k Q_\bullet,I_{\mathcal P}\otimes_{\mathcal C}E) H cat 4 ( P , E ∣ P ) ≅ Ext Fun ( P , Mod C ) 4 ( I P , I P ⊗ C E ) = H 4 Hom Fun ( P , Mod C ) ( Lan k Q ∙ , I P ⊗ C E ) ≅ H 4 Hom Fun ( F , Mod C ) ( Q ∙ , I F ⊗ C E ) ≅ Ext Fun ( F , Mod C ) 4 ( I F , I F ⊗ C E ) ≅ H c a t 4 ( F , E ∣ F ) . \cong H^4\operatorname{Hom}_{\operatorname{Fun}(\mathcal F,\operatorname{Mod}_{\mathcal C})}(Q_\bullet,I_{\mathcal F}\otimes_{\mathcal C}E)\cong\operatorname{Ext}_{\operatorname{Fun}(\mathcal F,\operatorname{Mod}_{\mathcal C})}^4(I_{\mathcal F},I_{\mathcal F}\otimes_{\mathcal C}E)\cong H_{\mathrm{cat}}^4(\mathcal F,E|_{\mathcal F}). ≅ H 4 Hom Fun ( F , Mod C ) ( Q ∙ , I F ⊗ C E ) ≅ Ext Fun ( F , Mod C ) 4 ( I F , I F ⊗ C E ) ≅ H cat 4 ( F , E ∣ F ) . □
Let ω 3 \omega_3 ω 3 be the ordinary ordinal category, viewed as a subcategory of F \mathcal F F via β ↦ y β \beta\mapsto y_\beta β ↦ y β .
Lemma 3.25. There are natural isomorphisms H c a t n ( ω 3 , E ∣ ω 3 ) ≅ Ext C n ( R , A ) H_{\mathrm{cat}}^n(\omega_3,E|_{\omega_3})\cong\operatorname{Ext}_{\mathcal C}^n(R,A) H cat n ( ω 3 , E ∣ ω 3 ) ≅ Ext C n ( R , A ) for n ≥ 0 n\ge0 n ≥ 0 .
Proof. Induced by restriction ω 3 → F \omega_3\to\mathcal F ω 3 → F , via β ↦ y β \beta\mapsto y_\beta β ↦ y β . A cochain u ∈ C c a t n ( F , E ∣ F ) u\in C_{\mathrm{cat}}^n(\mathcal F,E|_{\mathcal F}) u ∈ C cat n ( F , E ∣ F ) gives
res ( u ) : [ β 0 → β 1 → ⋯ → β n ] ↦ u [ y β 0 → y β 1 → ⋯ → y β n ] . \operatorname{res}(u):[\beta_0\to\beta_1\to\cdots\to\beta_n]\mapsto u[y_{\beta_0}\to y_{\beta_1}\to\cdots\to y_{\beta_n}]. res ( u ) : [ β 0 → β 1 → ⋯ → β n ] ↦ u [ y β 0 → y β 1 → ⋯ → y β n ] . By construction, ( C c a t ∙ ( ω 3 , E ∣ ω 3 ) , δ ) (C_{\mathrm{cat}}^\bullet(\omega_3,E|_{\omega_3}),\delta) ( C cat ∙ ( ω 3 , E ∣ ω 3 ) , δ ) and Hom C ( P ∙ , A ) \operatorname{Hom}_{\mathcal C}(P_\bullet,A) Hom C ( P ∙ , A ) has isomorphic terms:
C c a t n ( ω 3 , E ∣ ω 3 ) = ∏ β 0 ≤ ⋯ ≤ β n < ω 3 E ( β 0 , β n ) = ∏ β 0 ≤ ⋯ ≤ β n < ω 3 A ( β 0 ) ≅ Hom C ( ∏ β 0 ≤ ⋯ ≤ β n y β 0 , A ) . C_{\mathrm{cat}}^n(\omega_3,E|_{\omega_3})=\prod_{\beta_0\le\cdots\le\beta_n<\omega_3}E(\beta_0,\beta_n)=\prod_{\beta_0\le\cdots\le\beta_n<\omega_3}A(\beta_0)\cong\operatorname{Hom}_{\mathcal C}\left(\prod_{\beta_0\le\cdots\le\beta_n}y_{\beta_0},A\right). C cat n ( ω 3 , E ∣ ω 3 ) = β 0 ≤ ⋯ ≤ β n < ω 3 ∏ E ( β 0 , β n ) = β 0 ≤ ⋯ ≤ β n < ω 3 ∏ A ( β 0 ) ≅ Hom C β 0 ≤ ⋯ ≤ β n ∏ y β 0 , A . The differential δ \delta δ is constructed by bar resolution, so ( C c a t ∙ ( ω 3 , E ∣ ω 3 ) , δ ) ≅ Hom C ( P ∙ , A ) (C_{\mathrm{cat}}^\bullet(\omega_3,E|_{\omega_3}),\delta)\cong\operatorname{Hom}_{\mathcal C}(P_\bullet,A) ( C cat ∙ ( ω 3 , E ∣ ω 3 ) , δ ) ≅ Hom C ( P ∙ , A ) . Taking cohomology gives H c a t n ( ω 3 , E ∣ ω 3 ) ≅ Ext C n ( R , A ) H_{\mathrm{cat}}^n(\omega_3,E|_{\omega_3})\cong\operatorname{Ext}_{\mathcal C}^n(R,A) H cat n ( ω 3 , E ∣ ω 3 ) ≅ Ext C n ( R , A ) . □
Definition 3.26. Define χ \chi χ as the composite H c a t 4 ( F , E ∣ F ) → res H c a t 4 ( ω 3 , E ∣ ω 3 ) → ∼ Ext C 4 ( R , A ) H_{\mathrm{cat}}^4(\mathcal F,E|_{\mathcal F})\xrightarrow{\operatorname{res}}H_{\mathrm{cat}}^4(\omega_3,E|_{\omega_3})\xrightarrow{\sim}\operatorname{Ext}_{\mathcal C}^4(R,A) H cat 4 ( F , E ∣ F ) res H cat 4 ( ω 3 , E ∣ ω 3 ) ∼ Ext C 4 ( R , A ) .
Lemma 3.27. For every zero-normalized cocycle u ∈ Z c a t 4 ( P , E ∣ P ) u\in Z_{\mathrm{cat}}^4(\mathcal P,E|_{\mathcal P}) u ∈ Z cat 4 ( P , E ∣ P ) , the composite
P 4 → u ( d 1 , d 2 , d 3 , d 4 ) P 0 ⊗ C E → d 0 ⊗ C E A \mathcal P_4 \xrightarrow{u(d_1,d_2,d_3,d_4)} \mathcal P_0\otimes_{\mathcal C}E \xrightarrow{d_0\otimes_{\mathcal C}E} A P 4 u ( d 1 , d 2 , d 3 , d 4 ) P 0 ⊗ C E d 0 ⊗ C E A represents χ ( [ u ∣ F ] ) \chi([u|_{\mathcal F}]) χ ([ u ∣ F ]) in Ext C 4 ( R , A ) \operatorname{Ext}_{\mathcal C}^4(R,A) Ext C 4 ( R , A ) .
Proof. Let P + \mathcal P^+ P + be the additive subcategory generated by P \mathcal P P and R R R . Write the inclusions F → k P → h P + \mathcal F\xrightarrow{k}\mathcal P\xrightarrow{h}\mathcal P^+ F k P h P + and j = h k j=hk j = hk . For a projective resolution Q ∙ → I F Q_\bullet\to I_{\mathcal F} Q ∙ → I F , Lemma 3.24 shows that Lan k Q ∙ \operatorname{Lan}_k Q_\bullet Lan k Q ∙ is a projective resolution of I P I_{\mathcal P} I P . Similarly, since
( Lan j I F ) ( R ) = lim ( F → a R ) ∈ F / R F ≅ lim β < ω 3 y β ≅ R , (\operatorname{Lan}_j I_{\mathcal F})(R)=\lim_{\substack{(F\xrightarrow{a}R)\in\mathcal F/R}}F\cong\lim_{\beta<\omega_3}y_\beta\cong R, ( Lan j I F ) ( R ) = ( F a R ) ∈ F / R lim F ≅ β < ω 3 lim y β ≅ R , Lan j \operatorname{Lan}_j Lan j is exact and Lan j Q ∙ \operatorname{Lan}_j Q_\bullet Lan j Q ∙ is a projective resolution of I P + I_{\mathcal P^+} I P + . Since Lan h Lan k Q ∙ ≅ Lan j Q ∙ \operatorname{Lan}_h\operatorname{Lan}_k Q_\bullet\cong\operatorname{Lan}_j Q_\bullet Lan h Lan k Q ∙ ≅ Lan j Q ∙ , we have L n Lan h ( I P ) = 0 L_n\operatorname{Lan}_h(I_{\mathcal P})=0 L n Lan h ( I P ) = 0 for n > 0 n>0 n > 0 . Thus Lan h \operatorname{Lan}_h Lan h sends any projective resolution of I P I_{\mathcal P} I P to one of I P + I_{\mathcal P^+} I P + ; projectivity is preserved because restriction is its exact right adjoint.
Take the zero-reduced bar resolution B ∙ r e d → I P B^{\mathrm{red}}_\bullet\to I_{\mathcal{P}} B ∙ red → I P in Fun ( P , Mod C ) \operatorname{Fun}(\mathcal{P},\operatorname{Mod}_{\mathcal{C}}) Fun ( P , Mod C ) , i.e., the quotient of usual bar complex B ∙ B_\bullet B ∙ by its subcomplex N ∙ N_\bullet N ∙ , with terms
N n ( Y ) = ∐ X n → ⋯ → X 0 → Y at least one arrow is zero X n , B n r e d ( Y ) = ∐ X n → f n ⋯ → f 1 X 0 → a Y X j ∈ P , f j ≠ 0 , a ≠ 0 X n . N_n(Y)=\coprod_{\substack{X_n\to\cdots\to X_0\to Y\\\text{at least one}\\\text{arrow is zero}}}X_n,\qquad B^{\mathrm{red}}_n(Y)=\coprod_{\substack{X_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}X_0\xrightarrow{a}Y\\X_j\in\mathcal{P},\ f_j\ne0,\ a\ne0}}X_n. N n ( Y ) = X n → ⋯ → X 0 → Y at least one arrow is zero ∐ X n , B n red ( Y ) = X n f n ⋯ f 1 X 0 a Y X j ∈ P , f j = 0 , a = 0 ∐ X n .
Since H ≥ n N ∙ H_{\ge n}N_\bullet H ≥ n N ∙ is immediate, H ≥ 1 B ∙ r e d = 0 H_{\ge1}B^{\mathrm{red}}_\bullet=0 H ≥ 1 B ∙ red = 0 and H 0 B ∙ r e d ≅ I P H_0B^{\mathrm{red}}_\bullet\cong I_{\mathcal{P}} H 0 B ∙ red ≅ I P . Each N → B N\to B N → B is degree wise split: at degree n n n , construct the left inverse of inclusion by sending the coordinates
( X n → f n ⋯ → f 1 X 0 → a Y ) ⟼ { ( X n → f n ⋯ → f 1 X 0 → a Y ) , ∃ j , f j = 0 , ( X n → f n ⋯ → f 1 X 0 → Y ) , ∀ j , f j ≠ 0. \left(X_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}X_0\xrightarrow{a}Y\right)\longmapsto
\begin{cases}
\left(X_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}X_0\xrightarrow{a}Y\right),&\exists j,\ f_j=0,\\
\left(X_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}X_0\to Y\right),&\forall j,\ f_j\ne0.
\end{cases} ( X n f n ⋯ f 1 X 0 a Y ) ⟼ ⎩ ⎨ ⎧ ( X n f n ⋯ f 1 X 0 a Y ) , ( X n f n ⋯ f 1 X 0 → Y ) , ∃ j , f j = 0 , ∀ j , f j = 0.
Hence B r e d B^{\mathrm{red}} B red is a well-defined projective resolution.
Note that Lan h B ∙ r e d \operatorname{Lan}_h B^{\mathrm{red}}_\bullet Lan h B ∙ red is a projective resolution of I P + I_{\mathcal{P}+} I P + . Evaluation at R R R gives
T ∙ = ( Lan h B ∙ r e d ) ( R ) ⟶ R , T n = ∐ X n → f n ⋯ → f 1 X 0 → a R X j ∈ P , f j ≠ 0 , a ≠ 0 X n . T_\bullet=(\operatorname{Lan}_h B^{\mathrm{red}}_\bullet)(R)\longrightarrow R,\qquad
T_n=\coprod_{\substack{X_n\xrightarrow{f_n}\cdots\xrightarrow{f_1}X_0\xrightarrow{a}R\\X_j\in\mathcal{P},\ f_j\ne0,\ a\ne0}}X_n. T ∙ = ( Lan h B ∙ red ) ( R ) ⟶ R , T n = X n f n ⋯ f 1 X 0 a R X j ∈ P , f j = 0 , a = 0 ∐ X n .
The zero-normalized cocycle u u u defines u ~ : T 4 → A \widetilde{u}:T_4\to A u : T 4 → A . By construction u ~ ∣ ( f 1 , … , f 4 , 1 ) = ( 1 ⊗ C E ) ∘ u ( f 1 , f 2 , f 3 , f 4 ) \widetilde{u}|_{(f_1,\ldots,f_4,1)}=(1\otimes_{\mathcal{C}}E)\circ u(f_1,f_2,f_3,f_4) u ∣ ( f 1 , … , f 4 , 1 ) = ( 1 ⊗ C E ) ∘ u ( f 1 , f 2 , f 3 , f 4 ) . Naturality gives u ~ ∣ ( f 1 , … , f 4 , a ) = ( a ⊗ C E ) ∘ u ( f 1 , f 2 , f 3 , f 4 ) \widetilde{u}|_{(f_1,\ldots,f_4,a)}=(a\otimes_{\mathcal{C}}E)\circ u(f_1,f_2,f_3,f_4) u ∣ ( f 1 , … , f 4 , a ) = ( a ⊗ C E ) ∘ u ( f 1 , f 2 , f 3 , f 4 ) . Let Q n = ∐ β 0 ≤ ⋯ ≤ β n y β 0 Q_n=\coprod_{\beta_0\le\cdots\le\beta_n}y_{\beta_0} Q n = ∐ β 0 ≤ ⋯ ≤ β n y β 0 be the ordinal bar resolution used to compute H c a t ∗ ( ω 3 , E ∣ ω 3 ) H^*_{\mathrm{cat}}(\omega_3,E|_{\omega_3}) H cat ∗ ( ω 3 , E ∣ ω 3 ) . Then one has a comparison of projective resolution
\begin{CD}
\cdots@>>> Q_1 @>>> Q_0 @>>> R\\
@. @V{\psi_1}VV @V{\psi_0}VV @|\\
\cdots@>>> T_1 @>>> T_0 @>>> R
\end{CD}
where ψ ∙ \psi_\bullet ψ ∙ send index ( β 0 ≤ ⋯ ≤ β n ) (\beta_0\le\cdots\le\beta_n) ( β 0 ≤ ⋯ ≤ β n ) to ( y β 0 → ⋯ → y β n → R ) (y_{\beta_0}\to\cdots\to y_{\beta_n}\to R) ( y β 0 → ⋯ → y β n → R ) . Fix σ = ( β 0 ≤ ⋯ ≤ β 4 ) \sigma=(\beta_0\le\cdots\le\beta_4) σ = ( β 0 ≤ ⋯ ≤ β 4 ) , and let c j : y β j − 1 → y β j c_j:y_{\beta_{j-1}}\to y_{\beta_j} c j : y β j − 1 → y β j and a β 4 : y β 4 → R a_{\beta_4}:y_{\beta_4}\to R a β 4 : y β 4 → R be the canonical maps. Write ι σ : y β 0 → Q 4 \iota_\sigma:y_{\beta_0}\to Q_4 ι σ : y β 0 → Q 4 for the corresponding coproduct inclusion. Then
u ~ ∘ ψ 4 ∘ ι σ = ( a β 4 ⊗ C E ) ∘ u ( c 4 , c 3 , c 2 , c 1 ) . \widetilde{u}\circ\psi_4\circ\iota_\sigma=(a_{\beta_4}\otimes_{\mathcal{C}}E)\circ u(c_4,c_3,c_2,c_1). u ∘ ψ 4 ∘ ι σ = ( a β 4 ⊗ C E ) ∘ u ( c 4 , c 3 , c 2 , c 1 ) .
Under the canonical identifications y β 4 ⊗ C E ≅ A y_{\beta_4}\otimes_{\mathcal{C}}E\cong A y β 4 ⊗ C E ≅ A and R ⊗ C E ≅ A R\otimes_{\mathcal{C}}E\cong A R ⊗ C E ≅ A , the map a β 4 ⊗ C E a_{\beta_4}\otimes_{\mathcal{C}}E a β 4 ⊗ C E is the identity. Consequently, under the Yoneda identification Hom C ( y β 0 , A ) ≅ A ( β 0 ) \operatorname{Hom}_{\mathcal{C}}(y_{\beta_0},A)\cong A(\beta_0) Hom C ( y β 0 , A ) ≅ A ( β 0 ) ,
( u ~ ∘ ψ 4 ∘ ι σ ) β 0 ( 1 β 0 ) = ( u ∣ ω 3 ) ( β 0 ≤ ⋯ ≤ β 4 ) . (\widetilde{u}\circ\psi_4\circ\iota_\sigma)_{\beta_0}(1_{\beta_0})=(u|_{\omega_3})(\beta_0\le\cdots\le\beta_4). ( u ∘ ψ 4 ∘ ι σ ) β 0 ( 1 β 0 ) = ( u ∣ ω 3 ) ( β 0 ≤ ⋯ ≤ β 4 ) .
Thus u ~ ∘ ψ 4 \widetilde{u}\circ\psi_4 u ∘ ψ 4 corresponds to the cochain u ∣ ω 3 u|_{\omega_3} u ∣ ω 3 .
Finally we consider u ~ φ ~ 4 \widetilde{u}\widetilde{\varphi}_4 u φ 4 in another projective resolution P ∙ P_\bullet P ∙ of R R R , which also computes the image in Ext C 4 ( R , A ) \operatorname{Ext}^4_{\mathcal{C}}(R,A) Ext C 4 ( R , A ) . For each n ≥ 0 n\ge0 n ≥ 0 , let ι n : P n → T n \iota_n:P_n\to T_n ι n : P n → T n be the canonical inclusion of the summand indexed by P n → d n ⋯ → d 2 P 1 → d 1 P 0 → d 0 R P_n\xrightarrow{d_n}\cdots\xrightarrow{d_2}P_1\xrightarrow{d_1}P_0\xrightarrow{d_0}R P n d n ⋯ d 2 P 1 d 1 P 0 d 0 R . Note that ι ∙ \iota_\bullet ι ∙ is not a chain map: consider the summand indexed by ⋅ → f n ⋯ ⋅ → f 0 ⋅ \cdot\xrightarrow{f_n}\cdots\cdot\xrightarrow{f_0}\cdot ⋅ f n ⋯ ⋅ f 0 ⋅ , taking bar resolution gives face ⋅ → f n − 1 ⋯ ⋅ → f 0 ⋅ \cdot\xrightarrow{f_{n-1}}\cdots\cdot\xrightarrow{f_0}\cdot ⋅ f n − 1 ⋯ ⋅ f 0 ⋅ with sign ( − 1 ) n (-1)^n ( − 1 ) n . Hence φ ∙ : P ∙ → T ∙ \varphi_\bullet:P_\bullet\to T_\bullet φ ∙ : P ∙ → T ∙ is a comparison of projective resolution, where φ n = ( − 1 ) n ( n + 1 ) / 2 ι n \varphi_n=(-1)^{n(n+1)/2}\iota_n φ n = ( − 1 ) n ( n + 1 ) /2 ι n . Since ( − 1 ) 4 ⋅ 5 / 2 = 1 (-1)^{4\cdot5/2}=1 ( − 1 ) 4 ⋅ 5/2 = 1 , u ~ φ ~ 4 \widetilde{u}\widetilde{\varphi}_4 u φ 4 and u ~ ψ ~ 4 \widetilde{u}\widetilde{\psi}_4 u ψ 4 represent the same Ext \operatorname{Ext} Ext class. Thus χ ( [ u ∣ F ] ) = [ u ~ φ ~ 4 ] = [ ( d 0 ⊗ C E ) ∘ u ( d 1 , d 2 , d 3 , d 4 ) ] \chi([u|_{\mathcal{F}}])=[\widetilde{u}\widetilde{\varphi}_4]=[(d_0\otimes_{\mathcal{C}}E)\circ u(d_1,d_2,d_3,d_4)] χ ([ u ∣ F ]) = [ u φ 4 ] = [( d 0 ⊗ C E ) ∘ u ( d 1 , d 2 , d 3 , d 4 )] . □
Proposition 3.28. For every z ∈ H c a t 4 ( F , L ( E ∣ F ) ) z\in H^4_{\mathrm{cat}}(\mathcal{F},\mathop{\mathrm{L}}(E|_{\mathcal{F}})) z ∈ H cat 4 ( F , L ( E ∣ F )) , there is S ⊆ I S\subseteq I S ⊆ I with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ such that
( I ∖ S ) ⟶ Ext C 4 ( R , A ) , i ⟼ χ ( ( ev i ) ∗ z ) is a constant map. (I\setminus S)\longrightarrow\operatorname{Ext}^4_{\mathcal{C}}(R,A),\qquad i\longmapsto\chi((\operatorname{ev}_i)_*z)\quad\text{is a constant map.} ( I ∖ S ) ⟶ Ext C 4 ( R , A ) , i ⟼ χ (( ev i ) ∗ z ) is a constant map.
Proof. Choose a cocycle v v v representing res ( z ) ∈ H c a t 4 ( ω 3 , L ( E ∣ ω 3 ) ) \operatorname{res}(z) \in H^4_{\mathrm{cat}}(\omega_3,\mathcal{L}(E|_{\omega_3})) res ( z ) ∈ H cat 4 ( ω 3 , L ( E ∣ ω 3 )) . Namely,
v = ( v ( β 0 ≤ ⋯ ≤ β 4 ) ) β 0 ≤ ⋯ ≤ β 4 ∈ ∏ β 0 ≤ ⋯ ≤ β 4 L ( A ( β 0 ) ) , δ v = 0. v = \bigl(v(\beta_0 \leq\cdots\leq\beta_4)\bigr)_{\beta_0\leq\cdots\leq\beta_4} \in\prod_{\beta_0\leq\cdots\leq\beta_4}\mathcal{L}(A(\beta_0)), \qquad\delta v=0. v = ( v ( β 0 ≤ ⋯ ≤ β 4 ) ) β 0 ≤ ⋯ ≤ β 4 ∈ β 0 ≤ ⋯ ≤ β 4 ∏ L ( A ( β 0 )) , δ v = 0.
For each σ = ( β 0 ≤ ⋯ ≤ β 4 ) \sigma=(\beta_0\leq\cdots\leq\beta_4) σ = ( β 0 ≤ ⋯ ≤ β 4 ) , the coefficient v ( σ ) v(\sigma) v ( σ ) belongs to L ( A ( β 0 ) ) \mathcal{L}(A(\beta_0)) L ( A ( β 0 )) . Hence
( I ∖ S σ ) ⟶ A ( β 0 ) , i ⟼ e v i ( v ( σ ) ) (I\setminus S_\sigma)\longrightarrow A(\beta_0), \qquad i\longmapsto e v_i(v(\sigma)) ( I ∖ S σ ) ⟶ A ( β 0 ) , i ⟼ e v i ( v ( σ ))
is constant for some S σ ⊆ I S_\sigma\subseteq I S σ ⊆ I , with ∣ S σ ∣ < θ |S_\sigma|<\theta ∣ S σ ∣ < θ . Put S = ⋃ σ S σ S=\bigcup_\sigma S_\sigma S = ⋃ σ S σ . There are only ℵ 3 \aleph_3 ℵ 3 such strings. Since cf ( θ ) > ℵ 3 \operatorname{cf}(\theta)>\aleph_3 cf ( θ ) > ℵ 3 , one has ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ . Now the result follows.
The topological obstruction
Lemma 3.29. Every small topological triangulated category admits a fully faithful triangle functor into the homotopy category of a stable topological model category.
Proof. Let T \mathcal{T} T be a small topological triangulated category. Choose a fully faithful triangle functor T → Ho ( M 0 ) \mathcal{T}\to\operatorname{Ho}(\mathcal{M}_0) T → Ho ( M 0 ) with M 0 \mathcal{M}_0 M 0 stable. By [3 Remark 2.13] , the underlying ∞ \infty ∞ -category of M 0 \mathcal{M}_0 M 0 is stable. By [3 Propositions 4.8–4.9] , the resulting small stable ∞ \infty ∞ -category admits a spectral model A \mathcal{A} A compatible with the stable Yoneda embedding. The spectral Yoneda embedding into right A \mathcal{A} A -modules then induces a fully faithful triangle functor T → D ( A ) \mathcal{T}\to D(\mathcal{A}) T → D ( A ) , where D ( A ) D(\mathcal{A}) D ( A ) denotes the derived category of A \mathcal{A} A -modules; see [3 §2.1] .
Choose a cofibrant replacement A c → A \mathcal{A}^{c}\to\mathcal{A} A c → A in the model category of spectral categories with fixed object set Ob ( A ) \operatorname{Ob}(\mathcal{A}) Ob ( A ) . Let ∣ A c ∣ |\mathcal{A}^{c}| ∣ A c ∣ denote its levelwise geometric realisation. By the module comparisons in [21 proof of Corollary 1.2, p. 331] , there are Quillen equivalences
Mod - A ≃ Q Mod - A c ≃ Q Mod - ∣ A c ∣ . \operatorname{Mod}\text{-}\mathcal{A}\simeq_Q\operatorname{Mod}\text{-}\mathcal{A}^{c}\simeq_Q\operatorname{Mod}\text{-}|\mathcal{A}^{c}|. Mod - A ≃ Q Mod - A c ≃ Q Mod - ∣ A c ∣. The last category is a stable topological model category M \mathcal{M} M . Passing to homotopy categories, one has a triangle equivalence D ( A ) ≃ Ho ( M ) D(\mathcal{A})\simeq\operatorname{Ho}(\mathcal{M}) D ( A ) ≃ Ho ( M ) , and hence the required fully faithful triangle functor T → Ho ( M ) \mathcal{T}\to\operatorname{Ho}(\mathcal{M}) T → Ho ( M ) .
Proposition 3.30. Let T \mathcal{T} T be a small topological triangulated category, with a 2-split subset Q ⊂ T Q\subset\mathcal{T} Q ⊂ T . There exists a zero-normalised cocycle u ∈ Z c a t 4 ( Q , Hom T ( − , − [ − 2 ] ) ∣ Q o p × Q ) u\in Z^4_{\mathrm{cat}}(Q,\operatorname{Hom}_{\mathcal{T}}(-,-[-2])|_{Q^{\mathrm{op}}\times Q}) u ∈ Z cat 4 ( Q , Hom T ( − , − [ − 2 ]) ∣ Q op × Q ) such that for every composable string
X 4 → f 4 X 3 → f 3 X 2 → f 2 X 1 → f 1 X 0 , f j f j + 1 = 0 , X j ∈ Q X_4\xrightarrow{f_4}X_3\xrightarrow{f_3}X_2\xrightarrow{f_2}X_1\xrightarrow{f_1}X_0,\qquad f_jf_{j+1}=0,\qquad X_j\in Q X 4 f 4 X 3 f 3 X 2 f 2 X 1 f 1 X 0 , f j f j + 1 = 0 , X j ∈ Q one has $u(f_1,f_2,f_3,f_4)\in\langle f_1,f_2,f_3,f_4\rangle\subseteq\operatorname{Hom}_{\mathcal{T}}(X_4,X_0[-2]).
Proof. Let F : T → Ho ( M ) F:\mathcal{T}\to\operatorname{Ho}(\mathcal{M}) F : T → Ho ( M ) be such a fully faithful triangle functor, where M \mathcal{M} M is a stable topological model category. Since Q Q Q is 2-split,
Hom Ho ( M ) ( F P , F Q [ 2 r + 1 ] ) ≅ Hom T ( P , Q [ 2 r + 1 ] ) , \operatorname{Hom}_{\operatorname{Ho}(\mathcal{M})}(FP,FQ[2r+1])\cong\operatorname{Hom}_{\mathcal{T}}(P,Q[2r+1]), Hom Ho ( M ) ( F P , F Q [ 2 r + 1 ]) ≅ Hom T ( P , Q [ 2 r + 1 ]) , Thus F ( Q ) F(Q) F ( Q ) is 2-split. Then [19 Construction 5.7] gives a zero-normalised cochain u ~ ∈ C c a t 4 ( F ( Q ) , Hom Ho ( M ) ( − , − [ − 2 ] ) ) \widetilde{u}\in C^4_{\mathrm{cat}}(F(Q),\operatorname{Hom}_{\operatorname{Ho}(\mathcal{M})}(-,-[-2])) u ∈ C cat 4 ( F ( Q ) , Hom Ho ( M ) ( − , − [ − 2 ])) , which assigns each chain of 4 composable morphisms
X 4 → g 4 X 3 → g 3 X 2 → g 2 X 1 → g 1 X 0 X_4\xrightarrow{g_4}X_3\xrightarrow{g_3}X_2\xrightarrow{g_2}X_1\xrightarrow{g_1}X_0 X 4 g 4 X 3 g 3 X 2 g 2 X 1 g 1 X 0 to a morphism u ~ ( g 1 , g 2 , g 3 , g 4 ) ∈ Hom Ho ( M ) ( X 4 , X 0 [ − 2 ] ) \widetilde{u}(g_1,g_2,g_3,g_4)\in\operatorname{Hom}_{\operatorname{Ho}(\mathcal{M})}(X_4,X_0[-2]) u ( g 1 , g 2 , g 3 , g 4 ) ∈ Hom Ho ( M ) ( X 4 , X 0 [ − 2 ]) . By [19 Lemma 5.8] , δ u ~ = 0 \delta\widetilde{u}=0 δ u = 0 . Take Θ P , Q : Hom T ( P , Q [ − 2 ] ) → ∼ Hom Ho ( M ) ( F P , ( F Q ) [ − 2 ] ) \Theta_{P,Q}:\operatorname{Hom}_{\mathcal{T}}(P,Q[-2])\xrightarrow{\sim}\operatorname{Hom}_{\operatorname{Ho}(\mathcal{M})}(FP,(FQ)[-2]) Θ P , Q : Hom T ( P , Q [ − 2 ]) ∼ Hom Ho ( M ) ( F P , ( F Q ) [ − 2 ]) .
For each composable string of morphisms P 4 → f 4 P 3 → f 3 P 2 → f 2 P 1 → f 1 P 0 P_4\xrightarrow{f_4}P_3\xrightarrow{f_3}P_2\xrightarrow{f_2}P_1\xrightarrow{f_1}P_0 P 4 f 4 P 3 f 3 P 2 f 2 P 1 f 1 P 0 , set
u ( f 1 , f 2 , f 3 , f 4 ) : = Θ P 4 , P 0 − 1 ( u ~ ( F f 1 , F f 2 , F f 3 , F f 4 ) ) . u(f_1,f_2,f_3,f_4):=\Theta^{-1}_{P_4,P_0}\bigl(\widetilde{u}(Ff_1,Ff_2,Ff_3,Ff_4)\bigr). u ( f 1 , f 2 , f 3 , f 4 ) := Θ P 4 , P 0 − 1 ( u ( F f 1 , F f 2 , F f 3 , F f 4 ) ) . By naturality, for a : P ′ → P a:P'\to P a : P ′ → P , b : Q → Q ′ b:Q\to Q' b : Q → Q ′ , and x : P → Q [ − 2 ] x:P\to Q[-2] x : P → Q [ − 2 ] , one has
Θ P ′ , Q ′ ( b [ − 2 ] ∘ x ∘ a ) = ( F b ) [ − 2 ] ∘ Θ P , Q ( x ) ∘ F a . \Theta_{P',Q'}\bigl(b[-2]\circ x\circ a\bigr)=(Fb)[-2]\circ\Theta_{P,Q}(x)\circ Fa. Θ P ′ , Q ′ ( b [ − 2 ] ∘ x ∘ a ) = ( F b ) [ − 2 ] ∘ Θ P , Q ( x ) ∘ F a . Put u = Θ − 1 ( F ∗ u ~ ) u=\Theta^{-1}(F^*\widetilde{u}) u = Θ − 1 ( F ∗ u ) , then one has ( δ u ) = 0 (\delta u)=0 ( δ u ) = 0 , and u u u is zero-normalized.
Since f j f j + 1 = 0 f_jf_{j+1}=0 f j f j + 1 = 0 , [19 Proposition 6.1] gives u ~ ( F f 1 , … , F f 4 ) ∈ ⟨ F f 1 , … , F f 4 ⟩ \widetilde{u}(\mathbb{F}f_1,\ldots,\mathbb{F}f_4)\in\langle\mathbb{F}f_1,\ldots,\mathbb{F}f_4\rangle u ( F f 1 , … , F f 4 ) ∈ ⟨ F f 1 , … , F f 4 ⟩ . Then Remark 2.16 gives u ( f 1 , f 2 , f 3 , f 4 ) ∈ ⟨ f 1 , f 2 , f 3 , f 4 ⟩ u(f_1,f_2,f_3,f_4)\in\langle f_1,f_2,f_3,f_4\rangle u ( f 1 , f 2 , f 3 , f 4 ) ∈ ⟨ f 1 , f 2 , f 3 , f 4 ⟩ .
Let P \mathcal{P} P be the full additive category generated by representable C \mathcal{C} C -modules and the P p \mathcal{P}_p P p . Define the diagonal functor
Δ : F ⟶ T ∅ = Perf ( Λ − ( I ) ) ⟶ T . \Delta:\mathcal{F}\longrightarrow\mathcal{T}_{\varnothing}=\operatorname{Perf}(\Lambda_-(I))\longrightarrow\mathcal{T}. Δ : F ⟶ T ∅ = Perf ( Λ − ( I )) ⟶ T .
It sends y α y_\alpha y α to Hom Λ − ( I ) ( − , α ) \operatorname{Hom}_{\Lambda_-(I)}(-,\alpha) Hom Λ − ( I ) ( − , α ) , f : y α → y β f:y_\alpha\to y_\beta f : y α → y β to a constant map I → Hom C ( α , β ) I\to\operatorname{Hom}_{\mathcal{C}}(\alpha,\beta) I → Hom C ( α , β ) , i ↦ f i\mapsto f i ↦ f .
For i ∈ I i\in I i ∈ I , let κ i : P → T \kappa_i:\mathcal{P}\to\mathcal{T} κ i : P → T be the composite
κ i : P ⟶ V Sign ( i ) → X ↦ ( X , 0 ) T { i } = V Sign ( i ) × Perf ( Λ − ( I ∖ { i } ) ) ⟶ T . \kappa_i:\mathcal{P}\longrightarrow\mathcal{V}_{\operatorname{Sign}(i)}
\xrightarrow{X\mapsto(X,0)}
\mathcal{T}_{\{i\}}=\mathcal{V}_{\operatorname{Sign}(i)}\times\operatorname{Perf}(\Lambda_-(I\setminus\{i\}))\longrightarrow\mathcal{T}. κ i : P ⟶ V Sign ( i ) X ↦ ( X , 0 ) T { i } = V Sign ( i ) × Perf ( Λ − ( I ∖ { i })) ⟶ T .
Let Q \mathcal{Q} Q be the small full additive closure in T \mathcal{T} T of the objects of im Δ \operatorname{im}\Delta im Δ and all im κ i \operatorname{im}\kappa_i im κ i .
Lemma 3.31. Q \mathcal{Q} Q is 2-split. For P , Q ∈ F P,Q\in\mathcal{F} P , Q ∈ F , one has Hom T ( Δ P , Δ Q [ − 2 ] ) ≅ L ( E ∣ F ( P , Q ) ) \operatorname{Hom}_{\mathcal{T}}(\Delta P,\Delta Q[-2])\cong\mathsf{L}(\mathbb{E}|_{\mathcal{F}}(P,Q)) Hom T ( Δ P , Δ Q [ − 2 ]) ≅ L ( E ∣ F ( P , Q )) .
Proof. For representable P = y α P=y_\alpha P = y α and Q = y β Q=y_\beta Q = y β , the definition of Λ − ( I ) \Lambda_-(I) Λ − ( I ) gives
Hom T ( Δ y α , Δ y β [ − 2 ] ) ≅ L ( E ( α , β ) ) . \operatorname{Hom}_{\mathcal{T}}(\Delta y_\alpha,\Delta y_\beta[-2])\cong\mathsf{L}(E(\alpha,\beta)). Hom T ( Δ y α , Δ y β [ − 2 ]) ≅ L ( E ( α , β )) . The general case follows by taking finite coproducts in both variables.
Lemma 3.32. For each i ∈ I i\in I i ∈ I , there are natural transformations
κ i ∣ F → in i Δ → pr i κ i ∣ F , pr i in i = id κ i ∣ F . \kappa_i|_{\mathcal{F}}\xrightarrow{\operatorname{in}_i}\Delta\xrightarrow{\operatorname{pr}_i}\kappa_i|_{\mathcal{F}},\qquad\operatorname{pr}_i\operatorname{in}_i=\operatorname{id}_{\kappa_i|_{\mathcal{F}}}. κ i ∣ F in i Δ pr i κ i ∣ F , pr i in i = id κ i ∣ F . For P , Q ∈ F P,Q\in\mathcal{F} P , Q ∈ F , evaluation at i i i corresponds to the map
Hom T ( Δ P , Δ Q [ − 2 ] ) ⟶ Hom T ( κ i P , κ i Q [ − 2 ] ) , x ⟼ pr i , Q [ − 2 ] ∘ x ∘ in i , P . \operatorname{Hom}_{\mathcal{T}}(\Delta P,\Delta Q[-2])\longrightarrow\operatorname{Hom}_{\mathcal{T}}(\kappa_iP,\kappa_iQ[-2]),\qquad x\longmapsto\operatorname{pr}_{i,Q[-2]}\circ x\circ\operatorname{in}_{i,P}. Hom T ( Δ P , Δ Q [ − 2 ]) ⟶ Hom T ( κ i P , κ i Q [ − 2 ]) , x ⟼ pr i , Q [ − 2 ] ∘ x ∘ in i , P . Proof. For P = ∐ a = 1 m y α a P=\coprod_{a=1}^{m}y_{\alpha_a} P = ∐ a = 1 m y α a and Q = ∐ b = 1 n y β b Q=\coprod_{b=1}^{n}y_{\beta_b} Q = ∐ b = 1 n y β b , one has
Hom T ( Δ P , Δ Q [ − 2 ] ) ≅ ∏ a , b L ( E ( α a , β b ) ) ≅ L ( ∏ a , b E ( α a , β b ) ) . \operatorname{Hom}_{\mathcal{T}}(\Delta P,\Delta Q[-2])\cong\prod_{a,b}\mathsf{L}(E(\alpha_a,\beta_b))\cong\mathsf{L}\left(\prod_{a,b}E(\alpha_a,\beta_b)\right). Hom T ( Δ P , Δ Q [ − 2 ]) ≅ a , b ∏ L ( E ( α a , β b )) ≅ L a , b ∏ E ( α a , β b ) . Under the displayed identification, write a morphism x : Δ P → Δ Q [ − 2 ] x:\Delta P\to\Delta Q[-2] x : Δ P → Δ Q [ − 2 ] as a matrix ( x b a ) (x_{ba}) ( x ba ) , where x b a ∈ L ( E ( α a , β b ) ) x_{ba}\in\mathsf{L}(E(\alpha_a,\beta_b)) x ba ∈ L ( E ( α a , β b )) . Evaluation at i i i sends this matrix to ev i ( x ) = ( x b a ( i ) ) b , a \operatorname{ev}_i(x)=(x_{ba}(i))_{b,a} ev i ( x ) = ( x ba ( i ) ) b , a .
On the other hand, under the decomposition T { i } = V Sign ( i ) × Perf ( Λ − ( I ∖ { i } ) ) \mathcal{T}_{\{i\}}=\mathcal{V}_{\operatorname{Sign}(i)}\times\operatorname{Perf}(\Lambda_-(I\setminus\{i\})) T { i } = V Sign ( i ) × Perf ( Λ − ( I ∖ { i })) , the first component of x x x is precisely ( x b a ( i ) ) b , a (x_{ba}(i))_{b,a} ( x ba ( i ) ) b , a . Hence
( pr i ) Q [ − 2 ] ∘ x ∘ ( in i ) P = ( x b a ( i ) ) b , a = ev i ( x ) . (\operatorname{pr}_i)_{Q[-2]}\circ x\circ(\operatorname{in}_i)_P=(x_{ba}(i))_{b,a}=\operatorname{ev}_i(x). ( pr i ) Q [ − 2 ] ∘ x ∘ ( in i ) P = ( x ba ( i ) ) b , a = ev i ( x ) .
Lemma 3.33. For t ∈ H c a t 4 ( Q , Hom T ( − , − [ − 2 ] ) ∣ Q o p × Q ) t\in H^4_{\mathrm{cat}}(\mathcal{Q},\operatorname{Hom}_{\mathcal{T}}(-,-[-2])|_{\mathcal{Q}^{\mathrm{op}\times\mathcal{Q}}}) t ∈ H cat 4 ( Q , Hom T ( − , − [ − 2 ]) ∣ Q op × Q ) , put t i = ( ev i ) ∗ Δ ∗ t t_i=(\operatorname{ev}_i)_*\Delta^*t t i = ( ev i ) ∗ Δ ∗ t . Then
res ( t i ) = res ( ( κ i ∗ t ) ∣ F ) in H c a t 4 ( ω 3 , E ∣ ω 3 ) . \operatorname{res}(t_i)=\operatorname{res}((\kappa_i^*t)|_{\mathcal{F}})\quad\text{in }H^4_{\mathrm{cat}}(\omega_3,E|_{\omega_3}). res ( t i ) = res (( κ i ∗ t ) ∣ F ) in H cat 4 ( ω 3 , E ∣ ω 3 ) . Consequently, χ ( t i ) = χ ( ( κ i ∗ t ) ∣ F ) \chi(t_i)=\chi((\kappa_i^*t)|_{\mathcal{F}}) χ ( t i ) = χ (( κ i ∗ t ) ∣ F ) .
Proof. Fix u ∈ Z c a t 4 ( Q , Hom T ( − , − [ − 2 ] ) ∣ Q o p × Q ) u\in Z^4_{\mathrm{cat}}(\mathcal{Q},\operatorname{Hom}_{\mathcal{T}}(-,-[-2])|_{\mathcal{Q}^{\mathrm{op}\times\mathcal{Q}}}) u ∈ Z cat 4 ( Q , Hom T ( − , − [ − 2 ]) ∣ Q op × Q ) with [ u ] = t [u]=t [ u ] = t . Consider a mapping sequence P 4 → f 4 P 3 → f 3 P 2 → f 2 P 1 → f 1 P 0 P_4\xrightarrow{f_4}P_3\xrightarrow{f_3}P_2\xrightarrow{f_2}P_1\xrightarrow{f_1}P_0 P 4 f 4 P 3 f 3 P 2 f 2 P 1 f 1 P 0 in F \mathcal{F} F . Then
( Δ ∗ u ) ( f 1 , … , f 4 ) = u ( Δ f 1 , … , Δ f 4 ) ∈ Hom T ( Δ P 4 , Δ P 0 [ − 2 ] ) . (\Delta^*u)(f_1,\ldots,f_4)=u(\Delta f_1,\ldots,\Delta f_4)\in\operatorname{Hom}_{\mathcal{T}}(\Delta P_4,\Delta P_0[-2]). ( Δ ∗ u ) ( f 1 , … , f 4 ) = u ( Δ f 1 , … , Δ f 4 ) ∈ Hom T ( Δ P 4 , Δ P 0 [ − 2 ]) . Hence Lemma 3.32 gives
( ( ev i ) ∗ ( Δ ∗ u ) ) ( f 1 , … , f 4 ) = ( pr i ) P 0 [ − 2 ] ∘ u ( Δ f 1 , … , Δ f 4 ) ∘ ( in i ) P 4 . ((\operatorname{ev}_i)_*(\Delta^*u))(f_1,\ldots,f_4)=(\operatorname{pr}_i)_{P_0[-2]}\circ u(\Delta f_1,\ldots,\Delta f_4)\circ(\operatorname{in}_i)_{P_4}. (( ev i ) ∗ ( Δ ∗ u )) ( f 1 , … , f 4 ) = ( pr i ) P 0 [ − 2 ] ∘ u ( Δ f 1 , … , Δ f 4 ) ∘ ( in i ) P 4 . Meanwhile, a cocycle w i w_i w i representing ( κ i ∗ t ) ∣ F (\kappa_i^*t)|_{\mathcal{F}} ( κ i ∗ t ) ∣ F gives w i ( f 1 , … , f 4 ) = u ( κ i f 1 , … , κ i f 4 ) w_i(f_1,\ldots,f_4)=u(\kappa_if_1,\ldots,\kappa_if_4) w i ( f 1 , … , f 4 ) = u ( κ i f 1 , … , κ i f 4 ) .
Both ( ev i ) ∗ ( Δ ∗ u ) (\operatorname{ev}_i)_*(\Delta^*u) ( ev i ) ∗ ( Δ ∗ u ) and w i w_i w i takes values in Hom T ( κ i P 4 , κ i P 0 [ − 2 ] ) \operatorname{Hom}_{\mathcal{T}}(\kappa_iP_4,\kappa_iP_0[-2]) Hom T ( κ i P 4 , κ i P 0 [ − 2 ]) , we then show these two cocycles differ by a coboundary. For P 3 → f 3 P 2 → f 2 P 1 → f 1 P 0 P_3 \xrightarrow{f_3} P_2 \xrightarrow{f_2} P_1 \xrightarrow{f_1} P_0 P 3 f 3 P 2 f 2 P 1 f 1 P 0 in F \mathcal{F} F , put
h ( f 1 , f 2 , f 3 ) = ( pr i ) P 0 [ − 2 ] ∘ ( u ( ( in i ) P 0 , κ i f 1 , κ i f 2 , κ i f 3 ) − u ( Δ f 1 , ( in i ) P 1 , κ i f 2 , κ i f 3 ) h(f_1,f_2,f_3)=(\operatorname{pr}_i)_{P_0}[-2]\circ\left(u((\operatorname{in}_i)_{P_0},\kappa_i f_1,\kappa_i f_2,\kappa_i f_3)-u(\Delta f_1,(\operatorname{in}_i)_{P_1},\kappa_i f_2,\kappa_i f_3)\right. h ( f 1 , f 2 , f 3 ) = ( pr i ) P 0 [ − 2 ] ∘ ( u (( in i ) P 0 , κ i f 1 , κ i f 2 , κ i f 3 ) − u ( Δ f 1 , ( in i ) P 1 , κ i f 2 , κ i f 3 )
+ u ( Δ f 1 , Δ f 2 , ( in i ) P 2 , κ i f 3 ) − u ( Δ f 1 , Δ f 2 , Δ f 3 , ( in i ) P 3 ) ) . \left.+u(\Delta f_1,\Delta f_2,(\operatorname{in}_i)_{P_2},\kappa_i f_3)-u(\Delta f_1,\Delta f_2,\Delta f_3,(\operatorname{in}_i)_{P_3})\right). + u ( Δ f 1 , Δ f 2 , ( in i ) P 2 , κ i f 3 ) − u ( Δ f 1 , Δ f 2 , Δ f 3 , ( in i ) P 3 ) ) .
Taking the above equations and using δ u = 0 \delta u=0 δ u = 0 , one has
( δ h ) ( f 1 , f 2 , f 3 , f 4 ) = ( κ i f 1 ) [ − 2 ] ∘ h ( f 2 , f 3 , f 4 ) − h ( f 1 f 2 , f 3 , f 4 ) + h ( f 1 , f 2 f 3 , f 4 ) − h ( f 1 , f 2 , f 3 f 4 ) + h ( f 1 , f 2 , f 3 ) ∘ κ i f 4 = u ( κ i f 1 , … , κ i f 4 ) − ( pr i ) P 0 [ − 2 ] ∘ u ( Δ f 1 , … , Δ f 4 ) ∘ ( in i ) P 4 \begin{aligned}
(\delta h)(f_1,f_2,f_3,f_4)={}&(\kappa_i f_1)[-2]\circ h(f_2,f_3,f_4)-h(f_1f_2,f_3,f_4)+h(f_1,f_2f_3,f_4)\\
&-h(f_1,f_2,f_3f_4)+h(f_1,f_2,f_3)\circ\kappa_i f_4\\
={}&u(\kappa_i f_1,\ldots,\kappa_i f_4)-(\operatorname{pr}_i)_{P_0}[-2]\circ u(\Delta f_1,\ldots,\Delta f_4)\circ(\operatorname{in}_i)_{P_4}
\end{aligned} ( δ h ) ( f 1 , f 2 , f 3 , f 4 ) = = ( κ i f 1 ) [ − 2 ] ∘ h ( f 2 , f 3 , f 4 ) − h ( f 1 f 2 , f 3 , f 4 ) + h ( f 1 , f 2 f 3 , f 4 ) − h ( f 1 , f 2 , f 3 f 4 ) + h ( f 1 , f 2 , f 3 ) ∘ κ i f 4 u ( κ i f 1 , … , κ i f 4 ) − ( pr i ) P 0 [ − 2 ] ∘ u ( Δ f 1 , … , Δ f 4 ) ∘ ( in i ) P 4
This proves [ w i ] = ( ev i ) ∗ Δ ∗ [ u ] = ( ev i ) ∗ Δ ∗ t ( = t i ) [w_i]=(\operatorname{ev}_i)_*\Delta^*[u]=(\operatorname{ev}_i)_*\Delta^*t(=t_i) [ w i ] = ( ev i ) ∗ Δ ∗ [ u ] = ( ev i ) ∗ Δ ∗ t ( = t i ) .
Hence res ( ( ev i ) ∗ Δ ∗ t ) = res ( ( κ i ∗ t ) ∣ F ) \operatorname{res}((\operatorname{ev}_i)_*\Delta^*t)=\operatorname{res}((\kappa_i^*t)|_{\mathcal{F}}) res (( ev i ) ∗ Δ ∗ t ) = res (( κ i ∗ t ) ∣ F ) in H c a t 4 ( ω 3 , E ∣ ω 3 ) H^4_{\mathrm{cat}}(\omega_3,E|_{\omega_3}) H cat 4 ( ω 3 , E ∣ ω 3 ) . □
Lemma 3.34. Suppose that T \mathcal{T} T is topological, and choose the zero-normalised cocycle u u u of Proposition 3.30 on Q \mathcal{Q} Q . Put t = [ u ] t=[u] t = [ u ] and t i = ( ev i ) ∗ Δ ∗ t t_i=(\operatorname{ev}_i)_*\Delta^*t t i = ( ev i ) ∗ Δ ∗ t . Then
χ ( t i ) = { 0 , Sign ( i ) = − , − [ c ] , Sign ( i ) = + . \chi(t_i)=
\begin{cases}
0, & \operatorname{Sign}(i)=-,\\
-[c], & \operatorname{Sign}(i)=+.
\end{cases} χ ( t i ) = { 0 , − [ c ] , Sign ( i ) = − , Sign ( i ) = + .
Proof. Since κ i \kappa_i κ i is additive, κ i ∗ u \kappa_i^*u κ i ∗ u is again zero-normalised. This cocycle is defined on P \mathcal{P} P , thus taking P 4 → d 4 ⋯ → d 1 P 0 P_4\xrightarrow{d_4}\cdots\xrightarrow{d_1}P_0 P 4 d 4 ⋯ d 1 P 0 and applying Lemmas 3.33 and 3.27 give
χ ( t i ) = χ ( [ ( κ i ∗ u ) ∣ F ] ) = [ ( d 0 ⊗ C E ) ∘ ( κ i ∗ u ) ( d 1 , d 2 , d 3 , d 4 ) ] . \chi(t_i)=\chi\left([(\kappa_i^*u)|_{\mathcal{F}}]\right)=\left[(d_0\otimes_{\mathbb{C}}E)\circ(\kappa_i^*u)(d_1,d_2,d_3,d_4)\right]. χ ( t i ) = χ ( [( κ i ∗ u ) ∣ F ] ) = [ ( d 0 ⊗ C E ) ∘ ( κ i ∗ u ) ( d 1 , d 2 , d 3 , d 4 ) ] .
By Proposition 3.30,
u ( κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ) ∈ ⟨ κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ⟩ T ⊆ Hom T ( κ i P 4 , κ i P 0 [ − 2 ] ) . u(\kappa_i d_1,\kappa_i d_2,\kappa_i d_3,\kappa_i d_4)\in\langle\kappa_i d_1,\kappa_i d_2,\kappa_i d_3,\kappa_i d_4\rangle_{\mathcal{T}}\subseteq\operatorname{Hom}_{\mathcal{T}}(\kappa_iP_4,\kappa_iP_0[-2]). u ( κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ) ∈ ⟨ κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ⟩ T ⊆ Hom T ( κ i P 4 , κ i P 0 [ − 2 ]) .
Then, under the identification Hom T ( κ i P 4 , κ i P 0 [ − 2 ] ) ≅ Hom V Sign ( i ) ( P 4 , P 0 [ − 2 ] ) \operatorname{Hom}_{\mathcal{T}}(\kappa_iP_4,\kappa_iP_0[-2])\cong\operatorname{Hom}_{\mathcal{V}_{\operatorname{Sign}(i)}}(P_4,P_0[-2]) Hom T ( κ i P 4 , κ i P 0 [ − 2 ]) ≅ Hom V Sign ( i ) ( P 4 , P 0 [ − 2 ]) , Remark 2.16 gives a canonical bijection ⟨ κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ⟩ T ≅ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V Sign ( i ) \langle\kappa_i d_1,\kappa_i d_2,\kappa_i d_3,\kappa_i d_4\rangle_{\mathcal{T}}\cong\langle d_1,d_2,d_3,d_4\rangle_{\mathcal{V}_{\operatorname{Sign}(i)}} ⟨ κ i d 1 , κ i d 2 , κ i d 3 , κ i d 4 ⟩ T ≅ ⟨ d 1 , d 2 , d 3 , d 4 ⟩ V Sign ( i ) . Finally, Proposition 3.15 now gives the asserted values. □
Proposition 3.35. The category T \mathcal{T} T is not topological.
Proof. Suppose otherwise, and choose u u u and t i t_i t i as in Lemma 3.34. By Lemma 3.32 ,
Δ ∗ [ u ] ∈ H c a t 4 ( F , L ( E ∣ F ) ) . \Delta^*[u]\in H^4_{\mathrm{cat}}(\mathcal{F},\mathcal{L}(E|_{\mathcal{F}})). Δ ∗ [ u ] ∈ H cat 4 ( F , L ( E ∣ F )) .
Proposition 3.28 gives S ⊆ I S\subseteq I S ⊆ I with ∣ S ∣ < θ |S|<\theta ∣ S ∣ < θ such that χ ( t i ) \chi(t_i) χ ( t i ) is constant for i ∈ I ∖ S i\in I\setminus S i ∈ I ∖ S .
However, by Lemma 3.34, χ ( t i ) = − [ c ] \chi(t_i)=-[c] χ ( t i ) = − [ c ] for Sign ( i ) = + \operatorname{Sign}(i)=+ Sign ( i ) = + , and χ ( t i ) = 0 \chi(t_i)=0 χ ( t i ) = 0 for Sign ( i ) = − \operatorname{Sign}(i)=- Sign ( i ) = − . Here [ c ] ≠ 0 [c]\ne0 [ c ] = 0 by Lemma 3.3 . Both ∣ Sign − 1 ( + ) ∣ = ∣ Sign − 1 ( − ) ∣ = θ |\operatorname{Sign}^{-1}(+)|=|\operatorname{Sign}^{-1}(-)|=\theta ∣ Sign − 1 ( + ) ∣ = ∣ Sign − 1 ( − ) ∣ = θ , a contradiction. □
Statement on the use of AI
In the first arXiv version (v1), the author proposed a construction using two perfect categories arising from two A ∞ A_\infty A ∞ -categories with the same underlying graded category. By taking alternating products, the construction produces a triangulated category which is algebraic at every finite stage but is not topological as a whole. To find a graded algebra supporting this construction, the author used GPT6.1 Sol and computer searches.
The author soon realised that the tensor construction in v1 was not essential: the obstruction could be realised directly over the ordinal category ω 3 \omega_3 ω 3 , with a much simpler graded category and A ∞ A_\infty A ∞ -structures, leading to this simpler version (v2). Sections 1 (Introduction) and 3 (Main construction) of v2 are written by the author. Section 2 (Preliminaries) is written with assistance from GPT6.1 Sol; it provides background material and does not contain the main results.
We thank Yicong Zhao for his formal verification of the original version of this article in Lean. Formal verification of the present version (in Lean) is ongoing.
References
School of Mathematical Sciences, Shanghai Jiao Tong University
Email address: zhangchencheng@sjtu.edu.cn