A \(k\)-linear triangulated category without a model (Q2296296)

There are a few known examples of triangulated categories without model, given by \textit{F. Muro} et al. [Invent. Math. 170, No. 2, 231--241 (2007; Zbl 1125.18009)]. Those are however not linear over a field, and relies on very particular characteristic. This article considers triangulated categories over a field of characteristic \(0\) and prove the existence of such without a model. By a model, is meant a DG enhancement, or equivalently an \(A_\infty\)-enhancement. Then the main result of the article is an example of a \(k\)-linear triangulated category which does not carry an \(A_\infty\) enhancement. Let \(n\geq 3\) be a fixed number, \(k\) either a field of characteristic \(0\) or an infinite field of characteristic \(>n\), \(R=k[x_1,\dots,x_n]\), \(K=R_{(0)}\), \(R[\varepsilon]\) the \(R\)-linear DG-algebra with \(|\varepsilon|=-n+2,\;\varepsilon^2=0,\;d\varepsilon=0\), \(C(R,R)\) the Hochshild cochain complex of \(R\), \(\operatorname{HH}^n(R,R)=H^n(C(R,R))\) and \(T^n_{R/k}=\wedge^n_R\operatorname{Der}_k(R,R)\). The Hochschild-Kostant-Rosenberg theorem says that there is an inclusion \(T^n_{R/k}\subset Z^nC(R,R)\) inducing an isomorphism \(T^n_{R/k}\simeq\operatorname{HH}^n(R,R)\). Let \(\eta\in T^n_{R/k}\) and let \(R_\eta\) be the \(k[\varepsilon]\)-linear \(A_\infty\)-deformation of \(R[\varepsilon]\) whose only non-trivial higher multiplication is given by \(\varepsilon\eta\). The authors prove that for \(n\geq 14\) and \(\eta\neq 0\), there exists a triangulated category without \(A_\infty\)-enhancement with semi-orthogonal decomposition \(\langle D(K),D(R_\eta)\rangle\). Most of the introductory sections give the basics needed to give this example. An \(A_\infty\)-category is a DG-graph equipped with higher compositions \((m_i)_{i\geq 1}\) which satisfy certain natural quadratic relations. If only \(m_i\) with \(i\leq n\) are defined,this is called an \(A_n\)-category. The general principle throughout this text is that for any \(A_\infty\)-notion, there is a corresponding \(A_n\)-notion in which only operations with less than \(n\) arguments are considered. Notice that \(A_\infty\)-algebras are not categories, but if \(\mathfrak a\) is an \(A_n\)-category for \(n\geq 3\), the homotopy category \(H^0(\mathfrak a)\) is an honest category. A DG category is an \(A_\infty\) category with \(m_i=0\) for \(i>2\), \textit{A. I. Bondal} and \textit{M. M. Kapranov} [Math. USSR, Sb. 70, No. 1, 93--107 (1991; Zbl 0729.18008); translation from Mat. Sb. 181, No. 5, 669--683 (1990)] introduced pre-triangulated DG-categories by the property that their homotopy category is canonically triangulated. A pre-triangulated DG-category is thus a DG-category with additional properties. This idea is prolonged to define the analogous notion of a pre-triangulated \(A_\infty\)-category. It is pre-triangulated if the natural functor \(\mathfrak a\rightarrow\mathsf{Tw}\;\mathfrak a\) is a quasi-equivalence, where \(\mathsf{Tw}\;\mathfrak a\) is the category of twisted complexes over \(\mathfrak a\). This is equivalent to \(\mathfrak a\) being closed under suspensions, desuspensions and cones of closed maps, up to isomorphism in \(H^0(\mathsf{Tw}\;\mathfrak a)\). The statement of these operations requires a finite number of higher operations on \(\mathfrak a\) and make sense for \(A_n\)-categories, \(n\gg 0\). For an \(A_\infty\)-category \(\mathfrak a\), \(H^0(\mathsf{Tw}\;\mathfrak a)\) is canonically triangulated, so that if \(\mathfrak a\) is pre-triangulated, then \(H^0(\mathfrak a)\) is canonically triangulated. It is most reasonable that this can be proven by only finitely many of the higher operations on \(\mathfrak a\), so it should be possible to define a notion of a pre-triangulated \(A_n\)-category for \(n\gg 0\) which induces a canonical triangulation on its homotopy category. The authors choose to approach the explicit computations by techniques from \(A_\infty\)-context. The main difficulty is that the definition of \(\mathsf{Tw}\;\mathfrak a\) depends on higher compositions in \(\mathfrak a\) of unbounded arity, and so does not generalize to \(A_n\)-categories. This is solved by using twisted complexes of uniformly bounded length. Then only \(\mathsf{Tw}_{\leq 1}\;\mathfrak a\) is needed, which consists of twisted complexes of length two. The first main result of the article reads, verbatim: Theorem. If \(\mathfrak a\) is an \(A_n\)-category, then \(\mathsf{Tw}_{\leq 1}\;\mathfrak a\) is an \(A_{\lfloor(n-1)/2\rfloor}\)-category. If \(n\geq 7\), then we say that \(\mathfrak a\) is pre-triangulated if \(H^\ast(\mathfrak a)\rightarrow H^\ast(\mathsf{Tw}_{\leq 1}\;\mathfrak a)\) is a graded equivalence. If \(\mathfrak a\) is pre-triangulated and \(n\geq 13\), then \(H^0(\mathfrak a)\) is canonically triangulated. The authors study a glueing procedure: If \(\mathfrak a\) is an \(A_n\)-category, its pre-triangulated hull \(\mathsf{Tw}\;\mathfrak a\) is not well defined. However, there is a gluing procedure for which starting with pre-triangulated \(A_\infty\)-categories, one can produce pre-triangulated \(A_n\)-categories which are not themselves \(A_\infty\)-categories. For triangulated categories \(\mathcal A,\mathcal B\) and \(\mathcal M\) a \(\mathcal B-\mathcal A\)-bimodule, a gluing of \(\mathcal A,\mathcal B\) across \(\mathcal M\) is a triangulated category \(\mathcal C\) together with a semi-orthogonal decomposition \(\mathcal C=\langle\mathcal A,\mathcal B\rangle\) such that \(\mathcal C(A,B)=\mathcal M(A,B)\) of objects \(A,B\) in \(\mathcal A,\mathcal B\) respectively. The data \((\mathcal A,\mathcal B,\mathcal M)\) determines the objects of \(\mathcal C\) up to isomorphism, and there is a long exact sequence relating the \(\operatorname{Hom}\)-spaces in \(\mathcal C\) to those in \(\mathcal A,\mathcal B\) and the elements of \(\mathcal M\). However, the authors claim that triangulated categories are too flabby to allow the construction of \(\mathcal C\) from the triple \((\mathcal A,\mathcal B,\mathcal M)\). For \(A_\infty\)-categories \(\mathfrak a,\mathfrak b\) and \(M\) an \(A_\infty-\mathfrak b-\mathfrak a\)-module, one can define an \(A_\infty\)-gluing category \(\mathfrak c=\mathfrak a\coprod_M\mathfrak b\) such that \(\mathfrak a,\mathfrak b\) triangulated implies \(\mathfrak c\) pre-triangulated such that \(H^0(\mathfrak c)=\langle H^0(\mathfrak a),H^0(\mathfrak b)\rangle\) is a semi-orthogonal decomposition with associated bimodule \(H^0(M)\). Again, this only involves a finite number of higher operations, and so works for \(A_n\)-categories. Again verbatim: Theorem. Assume \(n\geq 13\), that \(\mathfrak a,\mathfrak b\) are pre-triangulated \(A_n\)-categories, and that \(M\) is an \(A_n-\mathfrak b-\mathfrak a\)-bimodule. Then \(\mathfrak a\coprod\mathfrak b\) is a pre-triangulated \(A_{n-1}\)-category. If \(n\geq 14\), so that \(H^0(\mathfrak a\coprod_M\mathfrak b)\) is triangulated by the previous theorem, then we have a semi-orthogonal decomposition \(H^0(\mathfrak a\coprod_M\mathfrak b)=\langle H^0(\mathfrak a),H^0(\mathfrak b)\rangle\) whose associated bimodule is \(H^0(M)\). The previously given counterexample is given on the form \(\mathcal D=H^0(\mathfrak a\coprod_M\mathfrak b)\) with \(\mathfrak a,\mathfrak b\) pre-triangulated \(A_\infty\)-categories and \(M\) an \(A_n-\mathfrak a-\mathfrak b\)-bimodule. For \(n\) sufficiently big, \(\mathcal D\) is canonically triangulated, and any \(A_\infty\)-enhancement on \(\mathcal D\) induces \(A_\infty\)-enhancements \(\mathfrak a', \mathfrak b'\) on \(H^0(\mathfrak a),H^0(\mathfrak b)\) as well as an \(A_\infty\)-functor \(F':\mathfrak a'\rightarrow \mathfrak b'\) such that \(H^0(F')=H^0(F)\). The authors prove that such \(F'\) doesn't exist, thus implying that an \(A_\infty\)-enhancement on \(\mathcal D\) doesn't exist. This is done with \(\mathfrak a,\mathfrak b\) the standard \(A_\infty\)-enhancements of \(D(K),D(R_\eta),\;\eta\neq 0\). The exact functor \(f:D(K)\rightarrow D(R_\eta):K\rightarrow K_\eta\) lifts to an \(A_{n-1}\)-functor \(F\). As long as \(\eta\neq 0\), \(f\) does not lift to an \(A_\infty\)-functor, even up to shift of enhancements. The proof this follows from among other tings, Toda brackets. The article is very well written, it contains (all?) necessary preliminaries to make it self-contained. In addition, it gives a very nice application of \(A_\infty\) categories, and very interesting results and computational techniques.