betaFeedback

WorksMathematics

Semi-exact functors and the conjecture of Coulembier and Flake

Abstract

We introduce the notion of semi-exact functors between abelian categories, which are characterised by either preserving or breaking exactness of sequences. We prove semi-exact functors are always exact with respect to a canonically determined exact structure. For a symmetric tensor category \Cs, we define a support theory by considering all k\Bbbk-linear symmetric monoidal, semi-exact functors C→Ver⁡p∞\mathscr{C}\to\operatorname{Ver}_{p^\infty} up to equivalence, and conjecture that it agrees with the Balmer spectrum of the stable category when C\mathscr{C} is finite. We prove a weak form of this conjecture and other results for Tannakian categories and in characteristic 0. Critical to the positive characteristic case is a functor constructed by Coulembier and Flake who were interested in higher Frobenius functors. We show that their construction, which relies on the theory of abelian envelopes, gives a natural source for semi-exact functors, and at least in the Tannakian setting gives all k\Bbbk-linear, symmetric monoidal, semi-exact functors.

Introduction: semi-exact functors

Throughout, kk denotes an algebraically closed field. We begin with a definition.

Definition 1.1. Let A,B\mathcal{A}, \mathcal{B} be abelian categories. We say an additive functor φ:A→B\varphi: \mathcal{A} \to\mathcal{B} is semi-exact if for any short exact sequence

0→A→B→C→00 \to A \to B \to C \to0

the following are equivalent:

  1. the sequence

0→φ(A)→φ(B)→φ(C)→00 \to\varphi(A) \to\varphi(B) \to\varphi(C) \to0

is exact;

  1. the sequence

0→φ(A)→φ(B)0 \to\varphi(A) \to\varphi(B)

is exact;

  1. the sequence

φ(B)→φ(C)→0\varphi(B) \to\varphi(C) \to0

is exact.

It is clear that any exact functor is semi-exact; however, there are examples of semi-exact functors which are not exact. For example, if A=A-mod\mathcal{A} = A\text{-mod}, where A=k[x]/x2A = k[x]/x^2, and B=Vec\mathcal{B} = \mathrm{Vec}, then the functor ker⁡(x)/im⁡(x)\ker(x)/\operatorname{im}(x) is semi-exact.

The notion of semi-exact functor does not appear in the literature to the best of this author’s knowledge. The purpose of this paper is to make a study of such functors. We find that in the setting of symmetric tensor categories, semi-exact functors have remarkable properties, and a rich connection to support theory.

Conjectures and theorems

By a symmetric tensor category C\mathcal{C} over k\mathbb{k} we mean a symmetric monoidal, rigid, k\mathbb{k}-linear, locally finite abelian category such that End⁡(1)=k\operatorname{End}(1)=\mathbb{k}. If C\mathcal{C} has finitely many simples and enough projectives, we additionally say that C\mathcal{C} is finite.

If C\mathcal{C} is a finite symmetric tensor category, then we may form its stable module category C\mathcal{C}, which is a tensor triangulated category, and therefore has a Balmer spectrum Sp⁡C\operatorname{Sp}\mathcal{C} consisting of all thick prime tensor ideals in C\mathcal{C}. We write max⁡Sp⁡C\max\operatorname{Sp}\mathcal{C} for the closed points of the topological space Sp⁡C\operatorname{Sp}\mathcal{C}.

In the following conjecture, let pp denote the characteristic of our field k\mathbb{k}, which is either a prime number or 00. If p=0p=0, we set Ver⁡p∞=sVec\operatorname{Ver}_{p^\infty}=\mathrm{sVec}, and for a prime p>0p>0 we let Ver⁡p∞\operatorname{Ver}_{p^\infty} denote the union ⋃nVer⁡pn\bigcup_n \operatorname{Ver}_{p^n} ([3]). We use the abbreviation LSM functor for a k\mathbb{k}-linear, symmetric monoidal functor (not necessarily exact).

Conjecture 1.2. Let C\mathcal{C} be a finite symmetric tensor category. Then there is a bijection:

{non-exact, semi-exact LSM functorsφ:C→Ver⁡p∞}∼⟶max⁡Sp⁡C,\frac{ \left\{ \begin{array}{c} \text{non-exact, semi-exact LSM functors}\\ \varphi:\mathcal{C}\to\operatorname{Ver}_{p^\infty} \end{array} \right\} }{\sim} \longrightarrow\max\operatorname{Sp}\mathcal{C},

where on the left hand side we have φ∼ψ\varphi\sim\psi if the two functors φ,ψ:C→Ver⁡p∞\varphi,\psi:\mathcal{C}\to\operatorname{Ver}_{p^\infty} are naturally isomorphic as monoidal functors.

Let us explain where the map in Conjecture 1.2 comes from. Given a non-exact, semi-exact LSM functor φ:C→Ver⁡p∞\varphi:\mathcal{C}\to\operatorname{Ver}_{p^\infty}, its object kernel will define a point x∈Sp⁡Cx\in\operatorname{Sp}\mathcal{C} such that φ(M)≠0\varphi(M)\ne0 if and only if xx lies in the support of MM. If we are able to identify Sp⁡C\operatorname{Sp}\mathcal{C} with Proj⁡Ext⁡∙(1,1)\operatorname{Proj}\operatorname{Ext}^{\bullet}(1,1) (which happens for Tannakian categories [4] and conjecturally always holds [23]), then we show that xx must be a closed point. This gives a map in one direction. However it may be surprising that this correspondence could be a bijection—we return to this shortly. In the meanwhile, here is another (perhaps more outlandish) conjecture. We will say that an LSM functor φ:C→D\varphi:\mathcal{C}\to\mathcal{D} is surjective if every object in D\mathcal{D} is a subquotient of an object in the image of φ\varphi.

Conjecture 1.3. Let C,D\mathcal{C},\mathcal{D} be symmetric tensor categories, and additionally assume that C\mathcal{C} is finite. Let φ:C→D\varphi:\mathcal{C}\to\mathcal{D} be a surjective, semi-exact, LSM functor. Then D\mathcal{D} is also finite.

Note that Conjecture 1.3 holds for exact functors [15 Prop. 1.8.19], but is false if we drop the semi-exactness property; indeed, semisimplification functors are a large source of counterexamples. Here is some immediate evidence for Conjectures 1.2 and 1.3.

Theorem 1.4. Conjectures 1.2 and 1.3 hold if k\mathbb{k} is characteristic 00 or C\mathcal{C} is Tannakian, i.e. C≃Rep⁡G\mathcal{C}\simeq\operatorname{Rep}G for a finite group scheme GG.

Connection with abelian envelopes

What is perhaps interesting about the proof of Theorem 1.4 is it can be done with a uniform argument that suggests how the proofs of Conjectures 1.2 and 1.3 would go in general. The key observation is that in each case, given a closed point M∈max⁡Sp⁡C\mathcal{M}\in\max\operatorname{Sp}\mathcal{C}, we can construct a nonzero additive tensor ideal IM\mathcal{I}_{\mathcal{M}} with finitely many indecomposables that are all supported at the point M\mathcal{M}. From this, one can turn the crank of the theory of abelian envelopes (see [8 Thm. 2.4.1], [24]) to obtain a functor

φM:C→(C/Mmax⁡)ab,\varphi_{\mathcal{M}}:\mathcal{C}\to\left(\mathcal{C}/\mathcal{M}^{\max}\right)^{ab},

where Mmax⁡\mathcal{M}^{\max} is a certain tensor ideal in C\mathcal{C}, (C/Mmax⁡)ab\left(\mathcal{C}/\mathcal{M}^{\max}\right)^{ab} is the abelian envelope of C/Mmax⁡\mathcal{C}/\mathcal{M}^{\max}, and is a finite symmetric tensor category. We prove that the functor φM\varphi_{\mathcal{M}} is semi-exact, and moreover in good situations any semi-exact functor on C\mathcal{C} must factor through one of the form φM\varphi_{\mathcal{M}} for some M∈max⁡Sp⁡C\mathcal{M} \in\operatorname{max}\operatorname{Sp}\mathcal{C}. Using the recent work [11] which proves the Benson-Etingof-Ostrik conjecture for finite tensor categories, along with uniqueness of fibre functors to Ver⁡p∞\operatorname{Ver}_{p^\infty}, Theorem 1.4 follows.

The conjecture of Coulembier-Flake

While § 1.2 gives a path to proof, the very difficult part is to construct an additive tensor ideal IM\mathcal{I}_{\mathcal{M}} with the prescribed properties. In characteristic 0 this is not so difficult, but in characteristic p>0p > 0 we require (a generalisation of) the ‘Saturation Conjecture’ of Coulembier-Flake [9 Conj. 5.4.2(1)]. This conjecture provides such additive tensor ideals in the case of quasi-elementary abelian groups, meaning ones isomorphic to Cpr×Ga(s)C_p^r \times\mathbb{G}_{a(s)}, where Ga(s)\mathbb{G}_{a(s)} denotes the ssth Frobenius kernel of Ga\mathbb{G}_a, and CpC_p is the cyclic group of order p>0p > 0. Given such a quasi-elementary abelian group EE of rank n=r+sn = r + s, along with a generic point M∈max⁡Sp⁡Rep⁡E\mathcal{M} \in\operatorname{max}\operatorname{Sp}\operatorname{Rep} E, [9] provides an additive tensor ideal IM\mathcal{I}_{\mathcal{M}} with the desired properties, and a surjective, semi-exact LSM functor

φM:Rep⁡E→Ver⁡pn.\varphi_{\mathcal{M}} : \operatorname{Rep} E \to\operatorname{Ver}_{p^n}.

The Saturation Conjecture of Coulembier-Flake was recently proven in [11]. We will provide a different more geometric proof which was discovered independently.

Upshot in characteristic 0 setting

The upshot of Theorem 1.4 in characteristic 0 is as follows. In this case, Deligne’s theorem [12] implies that C≃Rep⁡(G,ϵ)\mathcal{C} \simeq\operatorname{Rep}(G,\epsilon) for a supergroup GG of the form H⋉ΠVH \ltimes\Pi V, where HH is a finite group and VV is an HH-module. From an element v∈Vv \in V we obtain a Duflo-Serganova functor ([13], [19])

DSv:C→sVec,DS_v : \mathcal{C} \to\mathsf{sVec},

which is a semi-exact (by Hinich’s lemma, for the experts in superalgebras) LSM functor. Theorem 1.4 implies that any semi-exact functor on C\mathcal{C} must be isomorphic to DSvDS_v for some v∈Vv \in V.

Upshot in characteristic p>0p > 0

Let GG be a finite group scheme in characteristic p>0p > 0, and write C=Rep⁡G\mathcal{C} = \operatorname{Rep} G. Then every point x∈max⁡Sp⁡Cx \in\operatorname{max}\operatorname{Sp}\mathcal{C} arises as a generic closed point in max⁡Sp⁡Rep⁡E\operatorname{max}\operatorname{Sp}\operatorname{Rep} E, where E≤GE \leq G is a quasi-elementary abelian subgroup of GG [18]. Using the construction of Coulembier-Flake, we obtain, by restriction to EE, a semi-exact LSM functor

φx:Rep⁡G→Ver⁡pn⊆Ver⁡p∞.\varphi_x : \operatorname{Rep} G \to\operatorname{Ver}_{p^n} \subseteq\operatorname{Ver}_{p^\infty}.

The upshot of Theorem 1.4 is that every (non-exact) semi-exact LSM functor on Rep⁡G\operatorname{Rep} G arises in this way.

Connection to support theory

The bijection in Conjecture 1.2 point toward a purely abelian approach to support theory which may be useful even beyond the finite setting. For a field extension L/k\mathbb{L}/\mathbb{k}, define (Ver⁡p∞)L(\operatorname{Ver}_{p^\infty})_{\mathbb{L}} to be the extension of scalars of Ver⁡p∞\operatorname{Ver}_{p^\infty}, which is once again a symmetric tensor category [16 Lem. 2.2]. If L,L′\mathbb{L}, \mathbb{L}' are field extensions of k\mathbb{k}, and

φ1:C→(Ver⁡p∞)Landφ2:C→(Ver⁡p∞)L′\varphi_1 : \mathcal{C} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}} \quad\text{and}\quad\varphi_2 : \mathcal{C} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}'}

are LSM functors, then we write φ1∼φ2\varphi_1 \sim\varphi_2 if there exist a field extension L′′\mathbb{L}'' of both L\mathbb{L} and L′\mathbb{L}' such that φ1\varphi_1 and φ2\varphi_2 are coequalised by the natural functors (Ver⁡p∞)L→(Ver⁡p∞)L′′(\operatorname{Ver}_{p^\infty})_{\mathbb{L}} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}''} and (Ver⁡p∞)L′→(Ver⁡p∞)L′′(\operatorname{Ver}_{p^\infty})_{\mathbb{L}'} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}''}. This defines an equivalence relation on the set of LSM functors C→(Ver⁡p∞)L\mathcal{C} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}} ranging over all field extensions of L/k\mathbb{L}/\mathbb{k}. Now define

SE(C)={φ:C→(Ver⁡p∞)L:φ is a non-exact, semi-exact LSM functor, L/k a field ext}/∼.SE(\mathcal{C}) = \{\varphi: \mathcal{C} \to(\operatorname{Ver}_{p^\infty})_{\mathbb{L}} : \varphi\text{ is a non-exact, semi-exact LSM functor, } \mathbb{L}/\mathbb{k} \text{ a field ext}\}/\sim.

We equip SE(C)SE(\mathcal{C}) with a topology by declaring a basis for the closed sets to be those of the form I(X)={φ:φ(X)≠0}I(X)=\{\varphi:\varphi(X)\ne0\}, where XX runs over all objects of C\mathcal{C}. Then for X∈CX\in\mathcal{C}, we define its semi-exact support to be

supp⁡SE(X):=I(X).\operatorname{supp}_{SE}(X):=I(X).

By definition supp⁡SE(X)\operatorname{supp}_{SE}(X) is closed, and in this way (SE(C),supp⁡SE)(SE(\mathcal{C}),\operatorname{supp}_{SE}) defines a support theory on C\mathcal{C} which can easily be shown to satisfy the tensor product property:

supp⁡SE(X⊗Y)=supp⁡SE(X)∩supp⁡SE(Y).\operatorname{supp}_{SE}(X\otimes Y)=\operatorname{supp}_{SE}(X)\cap\operatorname{supp}_{SE}(Y).

One advantage of SE(C)SE(\mathcal{C}) is that it does not require C\mathcal{C} to be quasi-Frobenius. However, if C\mathcal{C} is quasi-Frobenius, then (SE(C),supp⁡SE)(SE(\mathcal{C}),\operatorname{supp}_{SE}) defines a support theory on C\mathcal{C}.

Conjecture 1.5. If C\mathcal{C} is a finite symmetric tensor category, then SE(C)SE(\mathcal{C}) is homeomorphic to Sp⁡C\operatorname{Sp}\mathcal{C}, and (SE(C),supp⁡SE)(SE(\mathcal{C}),\operatorname{supp}_{SE}) is the universal support theory for C\mathcal{C}.

It is possible that Conjecture 1.5 holds for any quasi-Frobenius category, but we have not computed any examples yet.

Statement on AI usage and story

This project began when the author discovered a proof of the Coulembier-Flake Saturation Conjecture using ChatGPT Astra 6.1. While writing up this result, the author became aware that Coulembier and Flake had recently also found a proof, which can now be found in [11] (see also the proof given by ChatGPT in [6]). However the author was originally motivated by the semi-exactness property in connection with the paper [20], which led to this paper.

The author used ChatGPT Astra 6.1 to help with the proof of Theorem 1.4, whose statement evolved over the last two months of work. The author also used AI to find references, understand related works in the literature, and to check the final version. Anyone interested in a more detailed AI usage statement is welcome to write the author.

This paper has been written entirely by the author, and the author takes full responsibility for its accuracy.

Acknowledgements

The author would like to foremost thank Kevin Coulembier and Johannes Flake for their comments on an initial draft of this preprint, and for helpful discussions thereafter. The author gives a hearty thanks to Bregje Pauwels for many insightful discussions and for coining the term semi-exact. The author further thanks Jonathan Gruber for interesting conversations about this project. Finally, the author thanks OpenAI for providing a temporary ChatGPT for Academic Researchers account.

Semi-exact functors, continued

We return to the study of semi-exact functors φ:A→B\varphi:\mathcal{A}\to\mathcal{B}, where A,B\mathcal{A},\mathcal{B} are now arbitrary abelian categories. Let us begin with an example.

Example 2.1. Consider A=k[x]/(xn)\mathcal{A}=\mathbb{k}[x]/(x^n), and let A=A-mod\mathcal{A}=A\text{-}\mathrm{mod}, B=Vec\mathcal{B}=\mathrm{Vec}. For 1≤i≤n1\le i\le n, define φi:A→B\varphi_i:\mathcal{A}\to\mathcal{B} by

φi=ker⁡(x)∩im⁡(xi−1)ker⁡(x)∩im⁡(xi).\varphi_i=\frac{\ker(x)\cap\operatorname{im}(x^{i-1})}{\ker(x)\cap\operatorname{im}(x^i)}.

Then ⨁i=1n−1φi\bigoplus_{i=1}^{n-1}\varphi_i is a semi-exact functor [20 Lem. 3.3.3].

In particular, in characteristic p>0p > 0 the semisimplification functor Rep⁡Cp→Ver⁡p\operatorname{Rep} C_p \to\operatorname{Ver}_p, which is given by

⨁i=1p−1φi⊠Li,\bigoplus_{i=1}^{p-1} \varphi_i \boxtimes L_i,

is a semi-exact functor. (For the definition and properties of Ver⁡p\operatorname{Ver}_p, we refer to [16].)

Remark 2.2. Example 2.1 shows that a semi-exact functor need not be middle-exact.

Here is a simple but useful lemma, whose proof we omit.

Lemma 2.3. Suppose that φ:A→B\varphi: \mathcal{A} \to\mathcal{B} is semi-exact, and $0 \to X \to Y \to Z \to 0$ is a short exact sequence in A\mathcal{A}.

  1. If φ(X)=0\varphi(X) = 0, then φ(Y)≅φ(Z)\varphi(Y) \cong\varphi(Z).

  2. If φ(Z)=0\varphi(Z) = 0, then φ(X)≅φ(Y)\varphi(X) \cong\varphi(Y).

Exact structure from a semi-exact functor

Let φ:A→B\varphi: \mathcal{A} \to\mathcal{B} be a semi-exact functor between abelian categories. Consider the class Eφ\mathcal{E}_{\varphi} of short exact sequences $0 \to A \to B \to C \to 0inin\mathcal{A}suchthat$0→φ(A)→φ(B)→φ(C)→0such that \$0 \to\varphi(A) \to\varphi(B) \to\varphi(C) \to0 is exact.

Proposition 2.4. The set (A,Eφ)(\mathcal{A}, \mathcal{E}_{\varphi}) is an exact category, and φ:A→B\varphi: \mathcal{A} \to\mathcal{B} is exact with respect to this exact structure.

Proof. We verify the axioms of an exact category. The only nontrivial axioms to check are that the pushout of admissible monomorphisms and the pullback of admissible epimorphisms remain admissible. We will prove it for admissible monomorphisms, with the case of admissible epimorphisms being similar. Let 0→A→B0 \to A \to B be a monomorphism in A\mathcal{A} such that 0→φ(A)→φ(B)0 \to\varphi(A) \to\varphi(B) is exact. Consider a morphism A→CA \to C giving the pushforward DD which fits into a short exact sequence

0→A→B⊕C→D→0.0 \to A \to B \oplus C \to D \to0.

Because A→BA \to B is admissible, φ(A)→φ(B⊕C)=φ(B)⊕φ(C)\varphi(A) \to\varphi(B \oplus C) = \varphi(B) \oplus\varphi(C) will be a monomorphism, and thus by definition of a semi-exact functor, the following sequence is exact:

0→φ(A)→φ(B)⊕φ(C)→φ(D)→0.0 \to\varphi(A) \to\varphi(B) \oplus\varphi(C) \to\varphi(D) \to0.

It follows that φ(C)→φ(D)\varphi(C) \to\varphi(D) is the pushout of φ(A)→φ(B)\varphi(A) \to\varphi(B) along φ(A)→φ(C)\varphi(A) \to\varphi(C). Since the pushout of a monomorphism is a monomorphism in an abelian category, φ(C)→φ(D)\varphi(C) \to\varphi(D) is a monomorphism, implying that C→DC \to D is an admissible monomorphism, as desired.

It is now obvious that φ\varphi is exact on (A,Eφ)(\mathcal{A}, \mathcal{E}_{\varphi}), so this completes the proof. □

2.1.1. A further exactness property. We say that an abelian category A\mathcal{A} is quasi-Frobenius if it has enough injectives, enough projectives, and injectives and projectives agree. If A\mathcal{A} is quasi-Frobenius, we can consider the stable category A‾\underline{\mathcal{A}}, which is the quotient of A\mathcal{A} by the ideal of morphisms factoring through a projective (equivalently, an injective) object. The category A‾\underline{\mathcal{A}} has the structure of a triangulated category, where the triangulated shift is given by

Σ(X)=X[1]≅I/X,\Sigma(X) = X[1] \cong I/X,

where II is the injective hull of XX. Moreover, every exact triangle in A‾\underline{\mathcal{A}} is of the form

X→αY→βZ→δΣ(X),X \xrightarrow{\alpha} Y \xrightarrow{\beta} Z \xrightarrow{\delta} \Sigma(X),

where $0 \to X \xrightarrow{\alpha} Y \xrightarrow{\beta} Z \to 0$ is a short exact sequence in A\mathcal{A}. We do not describe the boundary map δ\delta here.

Lemma 2.5. Suppose that A\mathcal{A} is quasi-Frobenius, φ:A→B\varphi: \mathcal{A} \to\mathcal{B} is semi-exact, and φ(P)=0\varphi(P) = 0 for all projectives P∈AP \in\mathcal{A}. Let

0→X→Y→Z→00 \to X \to Y \to Z \to0

be a short exact sequence such that φ(Y)=0\varphi(Y) = 0. Then we have a natural isomorphism

φ(Z)≅φ(Σ(X)).\varphi(Z) \cong\varphi(\Sigma(X)).

Proof. We have an exact triangle Y→Z→Σ(X)→Σ(Y)Y \to Z \to\Sigma(X) \to\Sigma(Y) in A\mathcal{A}. Thus there exists a short exact sequence $0 \to Y \to Z \oplus I \to \Sigma(X) \to 0$ in A\mathcal{A}, for some injective II in A\mathcal{A}. Since φ(Y)=0\varphi(Y) = 0, the semi-exactness property implies that φ(Z)≅φ(Z⊕I)≅φ(Σ(X))\varphi(Z) \cong\varphi(Z \oplus I) \cong\varphi(\Sigma(X)) as desired. □

Preliminaries on symmetric monoidal categories

A symmetric pseudo-tensor category C\mathcal{C} will mean a Karoubian, k\mathbb{k}-linear, rigid symmetric monoidal category. We recall that a symmetric tensor category is a symmetric pseudo-tensor category which is additionally abelian and locally finite.

Given a symmetric pseudo-tensor C\mathcal{C}, a tensor ideal K\mathcal{K} of C\mathcal{C} is an ideal of C\mathcal{C} (meaning a choice of subgroups K(X,Y)≤C(X,Y)\mathcal{K}(X,Y) \le\mathcal{C}(X,Y) satisfying the usual closure properties) such that id⁡X⊗f∈K\operatorname{id}_{X} \otimes f \in\mathcal{K} for all f∈Kf \in\mathcal{K}.

An additive tensor ideal I\mathcal{I} of C\mathcal{C} (also known as a thick tensor ideal in some places) is a collection of objects of C\mathcal{C}, closed under isomorphisms and direct sum(mands), such that if X∈IX \in\mathcal{I} and Y∈CY \in\mathcal{C}, then X⊗Y∈IX \otimes Y \in\mathcal{I}.

Given a tensor ideal K\mathcal{K} of C\mathcal{C}, we write Ob⁡(K)\operatorname{Ob}(\mathcal{K}) for the collection of objects X∈CX \in\mathcal{C} such that id⁡X∈K\operatorname{id}_{X} \in\mathcal{K}, which will be an additive tensor ideal.

Given an LSM functor φ:C→D\varphi: \mathcal{C} \to\mathcal{D} of symmetric pseudo-tensor categories, we will write ker⁡φ\operatorname{ker}\varphi for tensor ideal of morphisms f∈Cf \in\mathcal{C} such that φ(f)=0\varphi(f) = 0, and set Kφ=Ob⁡(ker⁡φ)\mathcal{K}_{\varphi} = \operatorname{Ob}(\operatorname{ker}\varphi).

Semi-exact LSM functors

If C\mathcal{C} is a symmetric tensor category over k\mathbb{k} which has enough projectives, then it will be quasi-Frobenius and we may define its stable module category C‾\underline{\mathcal{C}}. In this case C‾\underline{\mathcal{C}} will be a symmetric monoidal, rigid, k\mathbb{k}-linear triangulated category (otherwise known as a tensor triangulated category).

The following lemma is well-known [9 Lem. 2.2.2].

Lemma 2.6. If φ:C→D\varphi: \mathcal{C} \to\mathcal{D} is an LSM functor and C\mathcal{C} is quasi-Frobenius, then the following are equivalent:

  1. φ\varphi is exact;

  2. φ(P)≠0\varphi(P) \ne0 for all non-zero projectives PP in C\mathcal{C};

  3. φ(P)≠0\varphi(P) \ne0 for some non-zero projective PP in C\mathcal{C}.

Let C\mathcal{C} be quasi-Frobenius, and let φ:C→D\varphi: \mathcal{C} \to\mathcal{D} be a non-exact, semi-exact LSM functor. By Lemma 2.6, φ(P)=0\varphi(P) = 0 for all projectives PP in C\mathcal{C}, and thus φ\varphi factors through a functor C‾→D\underline{\mathcal{C}} \to\mathcal{D}, which by abuse of notation we also denote φ\varphi. Write

L:=φ(Σ(1)).L := \varphi(\Sigma(\mathbb{1})).

Since Σ(1)\Sigma(\mathbb{1}) is invertible in C‾\underline{\mathcal{C}}, LL is invertible in D\mathcal{D}, and thus is simple. Moreover, we have

φ(Σi(1))≅φ(Σ(1))⊗i≅L⊗i.\varphi(\Sigma^{i}(\mathbb{1})) \cong\varphi(\Sigma(\mathbb{1}))^{\otimes i} \cong L^{\otimes i}.

Lemma 2.7. If $0 \to X \to Y \to Z \to 0$ is a short exact sequence in C\mathcal{C} such that φ(Y)=0\varphi(Y)=0, then we have a natural isomorphism

φ(Z)≅φ(X)⊗L.\varphi(Z) \cong\varphi(X) \otimes L.

Proof. This follows from Lemma 2.5 and the fact that φ\varphi is monoidal. □

The Balmer spectrum

We refer to [1] for more details of the following construction. We continue to assume that C\mathcal{C} is a quasi-Frobenius symmetric tensor category. Recall that a prime thick tensor ideal in C\mathcal{C} is a proper triangulated full subcategory P⊆C\mathcal{P} \subseteq\mathcal{C} such that

  1. X⊗Y∈PX \otimes Y \in\mathcal{P} if X∈PX \in\mathcal{P} and Y∈CY \in\mathcal{C}, and

  2. if X⊗Y∈PX \otimes Y \in\mathcal{P}, then either X∈PX \in\mathcal{P} or Y∈PY \in\mathcal{P}.

We define the Balmer spectrum Sp⁡C\operatorname{Sp}\mathcal{C} to be the topological space whose points are thick prime tensor ideals in C\mathcal{C}, with a basis of closed sets given by I(X)={P∈Sp⁡C:X∉P}I(X)=\{\mathcal{P}\in\operatorname{Sp}\mathcal{C}:X\notin\mathcal{P}\} where XX runs over objects of C\mathcal{C}. We write max⁡Sp⁡C\max\operatorname{Sp}\mathcal{C} for the closed points of Sp⁡C\operatorname{Sp}\mathcal{C}. Recall that given an object X∈CX\in\mathcal{C}, its support is defined to be I(X)I(X).

Lemma 2.8. Let φ:C→D\varphi:\mathcal{C}\to\mathcal{D} be a non-exact, semi-exact, LSM functor. If C\mathcal{C} is quasi-Frobenius, then Kφ=Ob⁡(ker⁡φ)\mathcal{K}_{\varphi}=\operatorname{Ob}(\ker\varphi) descends to a prime thick tensor ideal of C\mathcal{C}.

Proof. If φ\varphi is not exact, then Kφ\mathcal{K}_{\varphi} must contain all projective objects, so Kφ\mathcal{K}_{\varphi} descends to an additive tensor ideal of C\mathcal{C}. To show Kφ\mathcal{K}_{\varphi} is a prime thick tensor ideal, we first observe that it clearly satisfies properties (1) and (2), so it remains to show that it is a triangulated full subcategory of C\mathcal{C}. For this, it suffices to prove that if X→Y→Z→X[1]X\to Y\to Z\to X[1] is a triangle and X,Y∈KφX,Y\in\mathcal{K}_{\varphi}, then Z∈KφZ\in\mathcal{K}_{\varphi}. However all triangles in C\mathcal{C} arise from short exact sequences in C\mathcal{C}, thus we may apply e.g. Lemma 2.3 for the result. □

Corollary 2.9. Let φ:C→D\varphi:\mathcal{C}\to\mathcal{D} be a non-exact, semi-exact LSM functor. Then there exists P∈Sp⁡C\mathcal{P}\in\operatorname{Sp}\mathcal{C} such that for all M∈CM\in\mathcal{C}, φ(M)≠0\varphi(M)\ne0 if and only if P∈supp⁡(M)\mathcal{P}\in\operatorname{supp}(M).

Lemma 2.8 gives a mapping:

κ:{non-exact, semi-exact LSM functorsφ:C→Ver⁡p∞}/∼⟶Sp⁡C.\kappa:\left\{\begin{array}{c} \text{non-exact, semi-exact LSM functors}\\ \varphi:\mathcal{C}\to\operatorname{Ver}_{p^\infty} \end{array}\right\}\big/\mathord{\sim}\longrightarrow\operatorname{Sp}\mathcal{C}.

Question 2.10. Is κ\kappa a bijection onto max⁡Sp⁡C\max\operatorname{Sp}\mathcal{C}?

Remark 2.11. Note that Conjecture 1.2 is the case of Question 2.10 when C\mathcal{C} is additionally assumed to be finite. A natural first case to consider beyond the finite setting is representations of a quasi-reductive supergroup in characteristic 00.

The Ext-algebra

Let C,D\mathcal{C},\mathcal{D} be symmetric tensor categories, and suppose that C\mathcal{C} is quasi-Frobenius. Let φ:C→D\varphi:\mathcal{C}\to\mathcal{D} denote a non-exact, semi-exact LSM functor, and write L:=φ(Σ(1))L:=\varphi(\Sigma(1)). Consider the N\mathbb{N}-graded k\mathbb{k}-algebra

RL=⨁i≥0Hom⁡(1,L⊗i).R_L=\bigoplus_{i\ge0}\operatorname{Hom}(1,L^{\otimes i}).

where multiplication is defined as follows. Given f:1→L⊗if:1\to L^{\otimes i}, g:1→L⊗jg:1\to L^{\otimes j}, define gfgf to be

1→fL⊗i→g⊗L⊗iL⊗(i+j).1\xrightarrow{f}L^{\otimes i}\xrightarrow{g\otimes L^{\otimes i}}L^{\otimes(i+j)}.

Lemma 2.12. RLR_L is a domain.

Proof. Indeed, since L⊗iL^{\otimes i} is simple for all ii, any non-zero f:1→L⊗if: 1 \to L^{\otimes i} must be an isomorphism, so any product of non-zero homogeneous elements must be non-zero. □

Now write RC=⨁n≥0Ext⁡n(1,1)R_{\mathcal{C}} = \bigoplus_{n \ge0} \operatorname{Ext}^{n}(1,1), which is a graded-commutative kk-algebra [1 Prop. 3.3]. We obtain an algebra morphism

rφ:RC→RL,r_{\varphi}: R_{\mathcal{C}} \to R_{L},

via the maps

Ext⁡i(1,1)≅Hom⁡C(1,Σi(1))→φHom⁡D(1,φ(Σi(1)))≅Hom⁡D(1,L⊗i).\operatorname{Ext}^{i}(1,1) \cong\operatorname{Hom}_{\mathcal{C}}(1,\Sigma^{i}(1)) \xrightarrow{\varphi} \operatorname{Hom}_{\mathcal{D}}(1,\varphi(\Sigma^{i}(1))) \cong\operatorname{Hom}_{\mathcal{D}}(1,L^{\otimes i}).

Let mφ={ψ∈RC:rφ(ψ)=0}\mathfrak{m}_{\varphi} = \{\psi\in R_{\mathcal{C}} : r_{\varphi}(\psi) = 0\} denote the kernel of rφr_{\varphi}. Then since RLR_{L} is a domain, mφ\mathfrak{m}_{\varphi} is a homogeneous prime ideal of RCR_{\mathcal{C}}.

Let us write Spec⁡h(RC)\operatorname{Spec}^{h}(R_{\mathcal{C}}) for the homogeneous prime spectrum of RCR_{\mathcal{C}}, meaning the set of homogeneous prime ideals of RCR_{\mathcal{C}}. For P∈Sp⁡C\mathcal{P} \in\operatorname{Sp}\mathcal{C}, define

ρ(P)=∑i{f:1→Σi(1):cone⁡(f)∉P}.\rho(\mathcal{P}) = \sum_{i}\{f:1 \to\Sigma^{i}(1):\operatorname{cone}(f) \notin\mathcal{P}\}.

Then ρ(P)\rho(\mathcal{P}) defines a prime ideal in RCR_{\mathcal{C}} [1 Thm. 5.3]. Hence we obtain a map

ρ:Sp⁡C→Spec⁡h(RC).\rho:\operatorname{Sp}\mathcal{C} \to\operatorname{Spec}^{h}(R_{\mathcal{C}}).

Observe that Proj⁡(RC)\operatorname{Proj}(R_{\mathcal{C}}) is a subspace of Spec⁡h(RC)\operatorname{Spec}^{h}(R_{\mathcal{C}}). We now introduce the following assumption:

Assumption (⋆): ρ\rho is a homeomorphism of Sp⁡C\operatorname{Sp}\mathcal{C} onto Proj⁡RC\operatorname{Proj} R_{\mathcal{C}}.

Note that assumption (⋆) is known to hold for finite symmetric tensor categories in characteristic 0 ([2], [21] and [14]), for representations of finite group schemes [4], and for representations of unipotent finite supergroup schemes [5]. Conjecturally it is expected to hold for all finite symmetric tensor categories [23].

Theorem 2.13. Continuing with our assumptions above, we have the following:

(1) Either mφ\mathfrak{m}_{\varphi} is the irrelevant ideal, or mφ\mathfrak{m}_{\varphi} is a closed point in Proj⁡RC\operatorname{Proj} R_{\mathcal{C}}.

(2) If assumption (⋆) holds, then under this identification Kφ=mφ\mathcal{K}_{\varphi} = \mathfrak{m}_{\varphi}, and thus Kφ\mathcal{K}_{\varphi} is a closed point.

Proof. For part (1), suppose that mφ\mathfrak{m}_{\varphi} is not the irrelevant ideal. Then dim⁡Hom⁡D(1,L⊗i)≤1\dim\operatorname{Hom}_{\mathcal{D}}(1,L^{\otimes i}) \le1. It follows that Im⁡rφ\operatorname{Im} r_{\varphi} has Krull dimension 1, and so Proj⁡(Im⁡rφ)\operatorname{Proj}(\operatorname{Im} r_{\varphi}) is a point, and mφ\mathfrak{m}_{\varphi} will be the pullback of this point under rφr_{\varphi}, and thus is closed.

For part (2), we need to prove that for f:1→Σi(1)f:1 \to\Sigma^{i}(1), cone⁡(f)∉Kφ\operatorname{cone}(f) \notin\mathcal{K}_{\varphi} if and only if f∈mφf \in\mathfrak{m}_{\varphi}. And indeed, we see that rφ(f)≠0r_{\varphi}(f) \ne0 if and only if

0→1→φ(f)L⊗i0 \to1 \xrightarrow{\varphi(f)} L^{\otimes i}

is exact, which by semi-exactness is equivalent to

0→1→φ(f)L⊗i→φ(cone⁡(f))→00 \to1 \xrightarrow{\varphi(f)} L^{\otimes i} \to\varphi(\operatorname{cone}(f)) \to0

being exact. However L⊗iL^{\otimes i} is simple, meaning φ(f)\varphi(f) must be an isomorphism, so this is equivalent to φ(cone⁡(f))=0\varphi(\operatorname{cone}(f)) = 0. Now Corollary 2.9 implies this is equivalent to cone⁡(f)∉Kφ\operatorname{cone}(f) \notin\mathcal{K}_{\varphi}, as desired. □

Remark 2.14. It is possible that mφ\mathfrak{m}_{\varphi} is the irrelevant ideal if C\mathcal{C} is not finite. For example, consider the supergroup G=Gm⋉ΠkG = G_m \ltimes\Pi k, where GmG_m acts nontrivially on Πk\Pi k, and set C=Rep⁡(G)\mathcal{C} = \operatorname{Rep}(G). Then RC=kR_{\mathcal{C}} = k, meaning every prime ideal is irrelevant.

Saturation, abelian envelopes, and semi-exact functors

Let C\mathcal{C} be a quasi-Frobenius symmetric tensor category, and let M∈max⁡Sp⁡C\mathcal{M} \in\operatorname{max}\operatorname{Sp}\mathcal{C}. We say that a non-zero additive tensor ideal I\mathcal{I} of C\mathcal{C} realises M\mathcal{M} if

  1. every nonzero object X∈IX \in\mathcal{I} has support precisely M\mathcal{M}, i.e. X∉MX \notin\mathcal{M} but X∈PX \in\mathcal{P} for all P∈Sp⁡C∖{M}\mathcal{P} \in\operatorname{Sp}\mathcal{C} \setminus\{\mathcal{M}\}, and

  2. I\mathcal{I} has finitely many indecomposable objects.

Suppose that I\mathcal{I} is an additive tensor ideal in C\mathcal{C} which realises M\mathcal{M}. Let Mmax⁡\mathcal{M}^{\max} denote the largest tensor ideal of C\mathcal{C} such that Ob⁡(Mmax⁡)=M\operatorname{Ob}(\mathcal{M}^{\max}) = \mathcal{M}. By an abelian envelope of C/Mmax⁡\mathcal{C}/\mathcal{M}^{\max} we mean a symmetric tensor category D\mathcal{D} and a faithful LSM functor C/Mmax⁡→D\mathcal{C}/\mathcal{M}^{\max} \to\mathcal{D} such that if D′\mathcal{D}' is another symmetric tensor category then any faithful LSM functor C/Mmax⁡→D′\mathcal{C}/\mathcal{M}^{\max} \to\mathcal{D}' factors uniquely through an exact LSM functor D→D′\mathcal{D} \to\mathcal{D}'.

Theorem 2.15. There exists an abelian envelope C/Mmax⁡→D\mathcal{C}/\mathcal{M}^{\max} \to\mathcal{D}, where D\mathcal{D} is finite, such that the induced functor φ:C→D\varphi: \mathcal{C} \to\mathcal{D} given by the composition

C→C‾→C/Mmax⁡→D,\mathcal{C} \to\underline{\mathcal{C}} \to\mathcal{C}/\mathcal{M}^{\max} \to\mathcal{D},

is semi-exact.

For the proof, we begin with a lemma.

Lemma 2.16. There exists an additive tensor ideal J≤I\mathcal{J} \leq\mathcal{I} such that J+M\mathcal{J}+\mathcal{M} is the unique minimal additive tensor ideal in C\mathcal{C} which contains M\mathcal{M}. In other words if J′\mathcal{J}' is another additive tensor ideal in C\mathcal{C} strictly containing M\mathcal{M}, then it must contain J+M\mathcal{J}+\mathcal{M}.

Proof. We take for J\mathcal{J} a minimal non-zero additive thick tensor ideal in I\mathcal{I}, which exists because I\mathcal{I} has finitely many indecomposables. Then suppose that J′≤C\mathcal{J}' \leq\mathcal{C} is an additive tensor ideal which strictly contains M\mathcal{M}. Then since J′≠M\mathcal{J}' \neq\mathcal{M}, there must exist X∈J′X \in\mathcal{J}' which is supported at M\mathcal{M}. Therefore for non-zero Y∈JY \in\mathcal{J} we must have X⊗Y∈J∩J′X \otimes Y \in\mathcal{J} \cap\mathcal{J}', meaning that J′∩J\mathcal{J}' \cap\mathcal{J} is a non-zero additive tensor ideal contained in J\mathcal{J}. Since J\mathcal{J} is minimal, it must be equal to J\mathcal{J}, and therefore J′\mathcal{J}' must contain J+M\mathcal{J}+\mathcal{M}. □

Proof of Theorem 2.15. Let J\mathcal{J} be the additive tensor ideal from Lemma 2.16. Then M\mathcal{M} satisfies the assumptions of [8 Thm. 2.4.1 (2)], where the minimal cover is J+M\mathcal{J}+\mathcal{M}. Thus C/Mmax⁡\mathcal{C}/\mathcal{M}^{\max} admits an envelope ψ:C/Mmax⁡→D\psi: \mathcal{C}/\mathcal{M}^{\max} \to\mathcal{D}, and moreover (1) ψ\psi restricted to J\mathcal{J} is full and (2) ψ(J)\psi(\mathcal{J}) will be the category of projectives in D\mathcal{D}.

It remains to prove that φ:C→D\varphi: \mathcal{C} \to\mathcal{D} is semi-exact. Let $0 \to A \to B \to C \to 0beashortexactsequenceinbe a short exact sequence in\mathcal{C},andsupposethat$0→φ(A)→φ(B), and suppose that \$0 \to\varphi(A) \to\varphi(B) is exact. Let P∈JP \in\mathcal{J} be non-zero. Tensoring with φ(P)\varphi(P), we obtain an exact sequence

0→φ(A)⊗φ(P)→φ(B)⊗φ(P).0 \to\varphi(A) \otimes\varphi(P) \to\varphi(B) \otimes\varphi(P).

However φ(P)\varphi(P) is projective, and thus injective, meaning each term in the exact sequence is injective. Hence we obtain a retraction r:φ(B⊗P)→φ(A⊗P)r : \varphi(B \otimes P) \to\varphi(A \otimes P). Because φ\varphi is fully faithful on J\mathcal{J}, there exists a map r′:A⊗P→B⊗Pr' : A \otimes P \to B \otimes P such that r′r=1+nr'r = 1+n, where n∈Mmax⁡n \in\mathcal{M}^{\max}. However Mmax⁡\mathcal{M}^{\max} does not contain the identity on any object in J\mathcal{J}, and thus nn must lie in the Jacobson radical of J\mathcal{J}. Hence 1+n1+n is invertible, implying that the triangle

A⊗P→B⊗P→C⊗P→Σ(A⊗P)A \otimes P \to B \otimes P \to C \otimes P \to\Sigma(A \otimes P)

is split in C\mathcal{C}. Thus 0→φ(A⊗P)→φ(B⊗P)→φ(C⊗P)→00 \to\varphi(A \otimes P) \to\varphi(B \otimes P) \to\varphi(C \otimes P) \to0 is exact, and since φ(P)⊗(−)\varphi(P) \otimes(-) is faithful and exact, this forces 0→φ(A)→φ(B)→φ(C)→00 \to\varphi(A) \to\varphi(B) \to\varphi(C) \to0 to also be exact, as desired. □

Saturation conjecture of Coulembier-Flake

For this section, k\Bbbk denotes an algebraically closed field of characteristic p>0p>0. In this section, we establish a generalisation of [9 Conj. 5.4.2(1)] in order to construct semi-exact functors in positive characteristic.

Let G=SL2G=SL_2, and write

U={(1∗01)}≤GU=\left\{\begin{pmatrix}1&*\\0&1\end{pmatrix}\right\}\leq G

for a root subgroup with normalising Borel subgroup BB. For λ=(λ1,…,λr)∈Ar(k)\lambda=(\lambda_1,\ldots,\lambda_r)\in\mathbb{A}^r(\Bbbk), we obtain a morphism ψλ:Cpr→U\psi_\lambda:\mathbb{C}_p^r\to U, where Cp\mathbb{C}_p denotes the cyclic group of order pp, given by:

ψλ(a1,…,ar)=(1∑i=1raiλi01).\psi_\lambda(a_1,\ldots,a_r)=\begin{pmatrix}1&\sum_{i=1}^{r}a_i\lambda_i\\0&1\end{pmatrix}.

Let Ga(s)\mathbb{G}_{a(s)} denote the ssth Frobenius kernel of Ga\mathbb{G}_a. For μ=(μ0,…,μs−1)∈As(k)\mu=(\mu_0,\ldots,\mu_{s-1})\in\mathbb{A}^s(\Bbbk), we have a morphism

σμ:Ga(s)→U,t↦∑i=0s−1μitpi.\sigma_\mu:\mathbb{G}_{a(s)}\to U,\qquad t\mapsto\sum_{i=0}^{s-1}\mu_i t^{p^i}.

Write Uμ=im⁡(σμ)U_\mu=\operatorname{im}(\sigma_\mu) for the image subgroup scheme. Then σμ\sigma_\mu is an embedding if and only if μ0≠0\mu_0\ne0.

Putting them together, for a choice of (λ,μ)(\lambda,\mu), we obtain a map

ψλ×σμ:Cpr×Ga(s)→U,\psi_\lambda\times\sigma_\mu:\mathbb{C}_p^r\times\mathbb{G}_{a(s)}\to U,

and this map is an embedding if and only if the λi\lambda_i are linearly independent over Fp\mathbb{F}_p and μ0\mu_0, and the image depends only on [λ:μ]∈Pr+s−1[\lambda:\mu]\in\mathbb{P}^{r+s-1}.

For λ,μ\lambda,\mu which define an embedding, we set

Hλ,μ=Eλ×Uμ≅Cpr×Ga(s),H_{\lambda,\mu}=E_\lambda\times U_\mu\cong\mathbb{C}_p^r\times\mathbb{G}_{a(s)},

which is a subgroup of UU.

Let UnU_n denote the nnth Frobenius kernel of UU. We have natural restriction functors

RU:Rep⁡G→Rep⁡U,Rλ,μ:Rep⁡G→Rep⁡Hλ,μ,RUn:Rep⁡G→Rep⁡Un.R_U:\operatorname{Rep}G\to\operatorname{Rep}U,\qquad R_{\lambda,\mu}:\operatorname{Rep}G\to\operatorname{Rep}H_{\lambda,\mu},\qquad R_{U_n}:\operatorname{Rep}G\to\operatorname{Rep}U_n.

Write Tilt for the category of tilting modules for SL2SL_2. This is a pseudo-tensor category, and the tensor ideals are exactly I0=Tilt⁡⊃I1⊃I2⊃⋯I_0=\operatorname{Tilt}\supset I_1\supset I_2\supset\cdots, where In=⟨St⁡n⟩I_n=\langle\operatorname{St}_n\rangle (see [7]). Here St⁡n\operatorname{St}_n denotes the nnth Steinberg module of SL2SL_2. The main conjecture of [9] can be stated as follows.

Conjecture 3.1. Let V∈Rep⁡EλV\in\operatorname{Rep}E_\lambda. Then V⊗Rλ,0(St⁡r−1)∈Rλ,0(Ir−1)V\otimes R_{\lambda,0}(\operatorname{St}_{r-1})\in R_{\lambda,0}(I_{r-1}).

We aim to prove Conjecture 3.1 by proving the following stronger theorem.

Theorem 3.2. Let V∈Rep⁡Hλ,μV\in\operatorname{Rep}H_{\lambda,\mu}. Then V⊗Rλ,μ(St⁡r+s−1)∈Rλ,μ(Ir+s−1)V\otimes R_{\lambda,\mu}(\operatorname{St}_{r+s-1})\in R_{\lambda,\mu}(I_{r+s-1}).

In the case s=0s=0, we obtain the following as a corollary:

Corollary 3.3. Let V∈Rep⁡(Us)V\in\operatorname{Rep}(U_s). Then V⊗RUs(St⁡s−1)∈RUn(Is−1)V\otimes R_{U_s}(\operatorname{St}_{s-1})\in R_{U_n}(I_{s-1}).

We now turn to the proof of Theorem 3.2. Write Dist⁡(U)=k[u0,u1,…]/(u0p,u1p,…)\operatorname{Dist}(U)=k[u_0,u_1,\ldots]/(u_0^p,u_1^p,\ldots) for the distribution algebra of UU, where uiu_i denotes the (pi)(p^i)-th divided power of a generator uu of Lie⁡(U)\operatorname{Lie}(U). Recall that Rep⁡(U)\operatorname{Rep}(U) is equivalent to the category of Dist⁡(U)\operatorname{Dist}(U)-modules VV such that uiV=0u_iV=0 for i≫0i\gg0 [17]. Thus Rep⁡(Un)\operatorname{Rep}(U_n) is the category of modules over the subalgebra Γn=k[u0,…,un−1]/(u0p,…,un−1p)\Gamma_n=k[u_0,\ldots,u_{n-1}]/(u_0^p,\ldots,u_{n-1}^p). We have a natural, exact (although not monoidal) functor α:Rep⁡(Un)→Rep⁡(U)\alpha:\operatorname{Rep}(U_n)\to\operatorname{Rep}(U) such that α∘RU=id⁡Rep⁡(Un)\alpha\circ R_U=\operatorname{id}_{\operatorname{Rep}(U_n)}, and uiu_i acts by zero on α(V)\alpha(V) for i≥ni\ge n.

First step: formulating a sufficient condition

Let n=r+sn=r+s. We have a commutative diagram

Rep⁡(Un)→αRep⁡(U)β↓↓ρ\begin{CD} \operatorname{Rep}(U_n) @>{\alpha}>> \operatorname{Rep}(U) \\ @V{\beta}VV @VV{\rho}V \\ &\operatorname{Rep} H_{\lambda,\mu} \end{CD}

Here ρ\rho is the natural restriction functor.

Lemma 3.4. The functor β=ρ∘α\beta=\rho\circ\alpha is an equivalence of abelian categories.

Proof. Indeed, first observe that St⁡n\operatorname{St}_n restricts to the free module of rank one over both UnU_n and Hλ,μH_{\lambda,\mu} (this can be shown using the polynomial model of the Steinberg representation, see e.g. [9 §5.3] and [11 Lem. 3.2]). Moreover, α(RUn(St⁡n))=RU(St⁡n)\alpha(R_{U_n}(\operatorname{St}_n))=R_U(\operatorname{St}_n) by the explicit polynomial realisation of St⁡n\operatorname{St}_n. Thus β\beta takes a free module of rank one to a free module of rank one. Since β\beta is exact and faithfully, and fully faithful on projectives, it follows it is a Morita equivalence. □

Now, let Vr,s∈Rep⁡Hλ,μV_{r,s}\in\operatorname{Rep} H_{\lambda,\mu}, and let VUn∈Rep⁡UnV_{U_n}\in\operatorname{Rep}U_n be such that β(VUn)≅Vr,s\beta(V_{U_n})\cong V_{r,s}. Finally, let V=VU:=α(VUn)V=V_U:=\alpha(V_{U_n}), so that ρ(V)=Vr,s\rho(V)=V_{r,s} and V∣Un=VUnV|_{U_n}=V_{U_n}. Our goal is to prove the following.

Theorem 3.5. We have V⊗RU(St⁡n−1)∈R(In−1)V\otimes R_U(\operatorname{St}_{n-1})\in R(I_{n-1}).

It is easy to see that Theorem 3.5 implies both Theorem 3.2, and thus also Conjecture 3.1 and Corollary 3.3.

Second step: studying the fibre of an induced module

Consider M=Ind⁡UGVM=\operatorname{Ind}_U^G V. This is naturally a GG-equivariant module over Ind⁡UG(k)=k[A2∖{(0,0)}]=k[x,y]\operatorname{Ind}_U^G(k)=k[\mathbb{A}^2\setminus\{(0,0)\}]=k[x,y].

Lemma 3.6. MM is a free k[x,y]k[x,y]-module of rank dim⁡V\dim V.

Proof. Indeed, because Dist⁡(Un)\operatorname{Dist}(U_n) is self-injective, there exists an exact sequence of Dist⁡(Un)\operatorname{Dist}(U_n)-modules 0→V→F1→F20\to V\to F_1\to F_2, where F1F_1 and F2F_2 are free. However Dist⁡(Un)≅RUn(St⁡n)\operatorname{Dist}(U_n)\cong R_{U_n}(\operatorname{St}_n) as a module over itself, so Fi≅RUn(St⁡n)⊕aiF_i\cong R_{U_n}(\operatorname{St}_n)^{\oplus a_i}. Applying left exactness of induction and the isomorphisms

Ind⁡UG(St⁡n)⊕ai≅St⁡n⊕ai⊗k[x,y],\operatorname{Ind}_U^G(\operatorname{St}_n)^{\oplus a_i}\cong\operatorname{St}_n^{\oplus a_i}\otimes k[x,y],

yields short exact sequences of k[x,y]k[x,y]-modules

0→M→St⁡n⊕a1⊗k[x,y]→St⁡n⊕a2⊗k[x,y].0\to M\to\operatorname{St}_n^{\oplus a_1}\otimes k[x,y]\to\operatorname{St}_n^{\oplus a_2}\otimes k[x,y].

Because k[x,y]k[x,y] has global dimension 22, this implies MM is projective by a standard syzygy argument. □

Write m=(x,y)\mathfrak{m}=(x,y) for the maximal ideal of (0,0)∈A2(0,0)\in A^2, which is GG-invariant. Then

W:=M/mMW:=M/\mathfrak{m}M

will be a GG-module of dimension dim⁡V\dim V.

For the following, recall that uiu_i are the divided powers of a generator uu of Lie⁡(U)\operatorname{Lie}(U).

Lemma 3.7. We have ui(W)=0u_i(W)=0 for i≥ni\ge n.

Proof. Indeed, let A1=Spec⁡k[x]⊆A2A^1=\operatorname{Spec} k[x]\subseteq A^2 be the line fixed pointwise by UU, and let M‾=M⊗k[x,y]k[x]\overline{M}=M\otimes_{k[x,y]}k[x]. Then M‾\overline{M} is a BB-equivariant vector bundle on A1A^1 with fibre WW at 00 and fibre VV at 11. For a point a∈A1∖{0}a\in A^1\setminus\{0\}, we have

(a00a−1)⋅1=a\begin{pmatrix} a & 0 \\ 0 & a^{-1} \end{pmatrix}\cdot1=a

Thus the action of UU on the fibre of M‾\overline{M} at aa is obtained by conjugating the action at 11 by

(a00a−1).\begin{pmatrix} a & 0 \\ 0 & a^{-1} \end{pmatrix}.

The operators uiu_i are eigenvectors for this conjugation. Since uiu_i acts by 00 on the fibre at 11 for i≥ni\ge n, these operators must act by zero on the fibre of M‾\overline{M} at every nonzero aa. But UU acts trivially on the underlying space A1A^1, so the operators uiu_i will act OA1\mathcal{O}_{A^1}-linearly on M‾\overline{M}. Since M‾\overline{M} is a vector bundle and these operators act trivially away from 00, they must act trivially everywhere, including the fibre WW at 00, completing the proof. □

In the following, write LaL_a for the irreducible SL2SL_2 module of highest weight a∈Na\in\mathbb{N}.

Corollary 3.8. If LaL_a is a composition factor of WW, then a<pna<p^n.

Proof. Indeed, if a≥pna\ge p^n then un+1u_{n+1} must act nontrivially by the Steinberg tensor product theorem. □

Corollary 3.8 and [3 Lem. 3.3] now imply:

Corollary 3.9. W⊗St⁡n−1W\otimes\operatorname{St}_{n-1} is a tilting module.

Now consider Z:=M⊗St⁡n−1Z:=M\otimes\operatorname{St}_{n-1}. It is easy to see that T:=Z/mZ≅W⊗St⁡n−1T:=Z/\mathfrak{m}Z\cong W\otimes\operatorname{St}_{n-1}, which is tilting by Corollary 3.9.

Third step: producing a GG-equivariant splitting of Z→TZ\to T

Observe that for all ii, we have a natural morphism of GG-modules

mi/mi+1⊗T→miZ/mi+1Z.\mathfrak{m}^i/\mathfrak{m}^{i+1}\otimes T\to\mathfrak{m}^iZ/\mathfrak{m}^{i+1}Z.

This must be an isomorphism since M⊗St⁡n−1M\otimes\operatorname{St}_{n-1} is a free k[x,y]k[x,y]-module. On the other hand, we have mi/mi+1≅∇(i)\mathfrak{m}^i/\mathfrak{m}^{i+1}\cong\nabla(i), where ∇(i)\nabla(i) denotes the costandard module of highest weight ii. Thus mi/mi+1⊗T\mathfrak{m}^i/\mathfrak{m}^{i+1}\otimes T has a filtration by costandard modules.

Now consider the short exact sequences:

0→miZ/mi+1Z→Z/mi+1Z→Z/miZ→00\to\mathfrak{m}^iZ/\mathfrak{m}^{i+1}Z\to Z/\mathfrak{m}^{i+1}Z\to Z/\mathfrak{m}^iZ\to0

The object miZ/mi+1Z≅mi/mi+1⊗T\mathfrak{m}^iZ/\mathfrak{m}^{i+1}Z\cong\mathfrak{m}^i/\mathfrak{m}^{i+1}\otimes T has a filtration by costandard modules, and thus

Ext⁡1(T,miZ/mi+1Z)=0\operatorname{Ext}^1(T,\mathfrak{m}^iZ/\mathfrak{m}^{i+1}Z)=0

for all ii, since TT is tilting. For i=1i=1 this provides a splitting of TT off Z/m2ZZ/\mathfrak{m}^2Z, and then at i=2i=2 a splitting off Z/m3ZZ/\mathfrak{m}^3Z, and so on.

Working inductively, for each ii we obtain a splitting of TT off Z/mi+1ZZ/\mathfrak{m}^{i+1}Z, which is compatible with the splitting off Z/miZZ/\mathfrak{m}^{i}Z. Therefore we obtain a map of vector spaces

T⟶lim←⁡Z/miZT \longrightarrow\varprojlim Z/\mathfrak{m}^{i}Z

Moreover, this map is equivariant for the natural action of G(k)G(k).

Recall we have a short exact sequence of GG-equivariant k[x,y]\mathbb{k}[x,y]-modules

0⟶Z⟶St⁡n−1⊗St⁡n⊕a1⊗k[x,y]⟶St⁡n−1⊗St⁡n⊕a1⊗k[x,y].0 \longrightarrow Z \longrightarrow\operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\mathbb{k}[x,y]\longrightarrow\operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\mathbb{k}[x,y].

If we complete this sequence with respect to m\mathfrak{m} it will remain exact ([25 Lem. 10.97.1]), and thus we have embeddings of GG-modules

Z→St⁡n−1⊗St⁡n⊕a1⊗k[x,y]↓↓Z^→∏a=0∞St⁡n−1⊗St⁡n⊕a1⊗∇(a).\begin{CD} Z @>>> \operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\mathbb{k}[x,y]\\ @VVV @VVV\\ \widehat{Z} @>>> \prod_{a=0}^{\infty}\operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\nabla(a). \end{CD}

This diagram is moreover a pullback. We now observe that TT has a nonzero map to St⁡n−1⊗St⁡n⊕a1⊗∇(a)\operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\nabla(a) for only finitely many aa, since the socle of ∇(a)\nabla(a) is L(a)L(a). Hence TT lands in St⁡n−1⊗St⁡n⊕a1⊗k[x,y]\operatorname{St}_{n-1}\otimes\operatorname{St}_{n}^{\oplus a_{1}}\otimes\mathbb{k}[x,y] under the composition of the bottom horizontal embeddings, and hence must land in ZZ. We may conclude:

Lemma 3.10. The map Z⟶Z/mZ=TZ\longrightarrow Z/\mathfrak{m}Z=T splits as a GG-module.

Fourth step: obtaining a global GG-equivariant splitting

Consider the GG-equivariant map induced by the k[x,y]\mathbb{k}[x,y]-module action:

φ:k[x,y]⊗T⟶Z.\varphi:\mathbb{k}[x,y]\otimes T\longrightarrow Z.

We claim this is an isomorphism. Indeed, it is an isomorphism in a neighborhood of 00 since TT maps isomorphically onto the fibre there. Since the complement of 00 is a homogeneous GG-space, and φ\varphi is GG-equivariant, it must be a global isomorphism.

Taking the fibre at the base point eU∈G/U(k)eU\in G/U(k), we obtain an isomorphism of UU-modules

V⊗R(St⁡n−1)≅T,V\otimes R(\operatorname{St}_{n-1})\cong T,

which finishes the proof of Theorem 4.2.

Recalling that T=W⊗St⁡n−1T=W\otimes\operatorname{St}_{n-1}, we additionally obtain the following corollary.

Corollary 3.11. Let Vr,s∈Rep⁡(Hλ,μ)V_{r,s}\in\operatorname{Rep}(H_{\lambda,\mu}). Then there exists W∈Rep⁡(SL2)W\in\operatorname{Rep}(\mathrm{SL}_{2}), all of whose weights are pnp^{n}-restricted, such that

Vr,s⊗Rλ,μ(St⁡n−1)≅Rλ,μ(W⊗St⁡n−1).V_{r,s}\otimes R_{\lambda,\mu}(\operatorname{St}_{n-1})\cong R_{\lambda,\mu}(W\otimes\operatorname{St}_{n-1}).

Proof of Theorem 1.4

The Coulembier-Flake functor

In [9], an LSM functor Rep⁡Cpr⟶Ver⁡pr\operatorname{Rep} C_{p}^{r}\longrightarrow\operatorname{Ver}_{p}^{r} is constructed (subject to the validity of Conjecture 3.1), which depends on the choice of a sufficiently generic point [λ]∈Pr−1[\lambda]\in\mathbb{P}^{r-1}. We now generalise this construction to a quasi-elementary abelian group

E=Cpr×Ga(s),E=C_{p}^{r}\times G_{a(s)},

to produce a functor

φλ,μ:Rep⁡E⟶Ver⁡pn,\varphi_{\lambda,\mu}:\operatorname{Rep} E\longrightarrow\operatorname{Ver}_{p}^{n},

where n=r+sn=r+s, which will depend on the choice of [λ:μ]∈Pr+s−1[\lambda:\mu]\in\mathbb{P}^{r+s-1}. The construction we give follows the ideas from § 2.5. Let S:=Rλ,μ(Stn−1)S := R_{\lambda,\mu}(\mathrm{St}_{n-1}), and let I\mathcal{I} denote the additive tensor ideal in Rep⁡E\operatorname{Rep} E generated by SS. Then by Theorem 3.2, I\mathcal{I} realises a point M∈max⁡Sp⁡Rep⁡E\mathcal{M} \in\operatorname{max} \operatorname{Sp} \operatorname{Rep} E which is described in [10 Cor. 4.2.7]. Thus we may apply Theorem 2.15 to obtain an abelian envelope

Rep⁡E/Mmax⁡⟶D.\operatorname{Rep} E/\mathcal{M}^{\max} \longrightarrow\mathcal{D}.

Thus we obtain a functor

Tilt⁡SL2⟶Rep⁡E⟶Rep⁡E/Mmax⁡⟶D,\operatorname{Tilt} SL_2 \longrightarrow\operatorname{Rep} E \longrightarrow\operatorname{Rep} E/\mathcal{M}^{\max} \longrightarrow\mathcal{D},

and we know that Stn−1\mathrm{St}_{n-1} is not killed while Stn\mathrm{St}_n goes to 00. Thus, writing In\mathcal{I}_n for the tensor ideal of Tilt⁡SL2\operatorname{Tilt} SL_2 generated by Stn\mathrm{St}_n, and using the classification of tensor ideals in Tilt⁡SL2\operatorname{Tilt} SL_2 [7], we obtain a faithful LSM functor

Tilt⁡SL2/In⟶D,\operatorname{Tilt} SL_2/\mathcal{I}_n \longrightarrow\mathcal{D},

which therefore factors through a tensor functor from the abelian envelope of Tilt⁡SL2/In\operatorname{Tilt} SL_2/\mathcal{I}_n, which is Ver⁡pn\operatorname{Ver}_{p^n}:

F:Ver⁡pn⟶D.F : \operatorname{Ver}_{p^n} \longrightarrow\mathcal{D}.

Lemma 4.1. The functor FF is an equivalence of categories.

Proof. Because Ver⁡pn\operatorname{Ver}_{p^n} is incompressible [3], it suffices to prove that FF is a surjective tensor functor. By the theory of abelian envelopes explained in § 2.5, the indecomposables of I\mathcal{I} will make up all of the projective indecomposables of D\mathcal{D}. However I\mathcal{I} is in the image of the functor Tilt⁡SL2⟶Rep⁡E\operatorname{Tilt} SL_2 \longrightarrow\operatorname{Rep} E, and so the image of FF must contain all projectives in D\mathcal{D}. From here it is straightforward to show that FF is surjective. □

We have thus constructed the desired LSM functor

φλ,μ:Rep⁡E⟶Ver⁡pn.\varphi_{\lambda,\mu} : \operatorname{Rep} E \longrightarrow\operatorname{Ver}_{p^n}.

Theorem 4.2. The functor φλ,μ\varphi_{\lambda,\mu} is semi-exact.

Proof. Indeed, this follows from Theorem 2.15. □

Finite group schemes and quasi-elementary subgroups

Let GG be a finite group scheme, and let C=Rep⁡G\mathcal{C} = \operatorname{Rep} G. Recall that a quasielementary group scheme EE is one isomorphic to Cpr×Ga(s)\mathbb{C}_p^r \times\mathbb{G}_a(s) for some r,s∈Nr, s \in\mathbb{N}. For every point x∈max⁡Sp⁡Cx \in\operatorname{max} \operatorname{Sp} \mathcal{C}, there exists a minimal quasi-elementary subgroup E≤GE \leq G such that xx lies in the image of Sp⁡Rep⁡E→Sp⁡C\operatorname{Sp} \operatorname{Rep} E \to\operatorname{Sp} \mathcal{C} [18].

Finite supergroup schemes in characteristic 0

If kk is characteristic 00 and C\mathcal{C} is a finite symmetric tensor category, then by [12], C≃Rep⁡(H⋉ΠV,ε)\mathcal{C} \simeq\operatorname{Rep}(H \ltimes\Pi V,\varepsilon) for a finite group HH, an HH-module VV, and ε∈Z(H)\varepsilon\in Z(H) such that ε2=1\varepsilon^2 = 1, and ε\varepsilon acts by −1-1 on VV. We write G:=H⋉ΠVG := H \ltimes\Pi V. In this case, the works of [2], [21], and [14] give the Balmer spectrum as

Sp⁡C=P(V)/H.\operatorname{Sp} \mathcal{C} = \mathbb{P}(V)/H.

Thus to a point x∈max⁡Sp⁡Cx \in\operatorname{max} \operatorname{Sp} \mathcal{C} we will associate an element x∈Vx \in V, which is well-defined up to non-zero scalar and HH-orbit.

The proof of Theorem 1.4

We now give a proof of Theorem 1.4. Therefore we either assume that k\mathbb{k} is characteristic 00 and C=Rep⁡(G,ε)\mathcal{C} = \operatorname{Rep}(G,\varepsilon) with G=H⋉ΠVG = H \ltimes\Pi V as in the setup of § 4.3, or that k\mathbb{k} is characteristic p>0p > 0 and C=Rep⁡G\mathcal{C} = \operatorname{Rep} G for a finite group scheme GG.

Proof of Theorem 1.4. Let φ:C→D\varphi: \mathcal{C} \to\mathcal{D} be a surjective semi-exact functor. If φ\varphi is exact, then D\mathcal{D} must be finite by [15 Prop. 1.8.19], and thus we now assume that φ\varphi is not exact. Then by Theorem 2.13, φ\varphi determines a point Kφ∈max⁡Sp⁡C\mathcal{K}_{\varphi} \in\operatorname{max} \operatorname{Sp} \mathcal{C}.

If k\mathbb{k} is characteristic 00, then by § 4.3, Kφ\mathcal{K}_{\varphi} lifts to a non-zero element x∈Vx \in V, and in this case we let E=Π⟨x⟩E = \Pi\langle x\rangle denote the purely odd, one-dimensional supergroup generated by xx. Then we set S:=Ind⁡EGS‾S := \operatorname{Ind}_{E}^{G} \overline{S}, where S‾\overline{S} denotes the trivial EE-module.

If k\mathbb{k} is positive characteristic, then Kφ\mathcal{K}_{\varphi} arises from a quasi-elementary abelian subgroup EE of rank nn such that it restricts to a generic point ([λ],[μ])([\lambda],[\mu]) in Sp⁡Rep⁡E\operatorname{Sp} \operatorname{Rep} E. In this case, let S‾=Rλ,μ(St⁡n−1)∈Rep⁡E\overline{S} = R_{\lambda,\mu}(\operatorname{St}_{n-1}) \in\operatorname{Rep} E, and set S:=Ind⁡EGS‾S := \operatorname{Ind}_{E}^{G} \overline{S}.

Lemma 4.3. The support of SS is exactly the point Kφ\mathcal{K}_{\varphi}.

Proof. In characteristic p>0p > 0 we may apply [4 Prop. 9.5(2)]. In characteristic 00, the induced module is given by ⨁g∈GΛ(V/⟨gx⟩)\bigoplus_{g\in G} \Lambda(V/\langle gx\rangle), which has support precisely the GG-orbit of xx, as desired. □

Recall that for M∈CM \in\mathcal{C}, we have the identity

M⊗Ind⁡EGS‾≅Ind⁡EG(Res⁡E(M)⊗S‾).M \otimes\operatorname{Ind}_{E}^{G} \overline{S} \cong\operatorname{Ind}_{E}^{G}(\operatorname{Res}_{E}(M) \otimes\overline{S}).

In all cases we have that S‾\overline{S} generates an additive tensor ideal in Rep⁡E\operatorname{Rep} E with finitely many indecomposables, and thus S‾\overline{S} must do the same in C\mathcal{C}. Write I\mathcal{I} for the additive tensor ideal generated by SS. Then since SS is supported exactly at mφ\mathfrak{m}_{\varphi} by Lemma 4.3, I\mathcal{I} realises Kφ\mathcal{K}_{\varphi} in the sense of § 2.5. By Theorem 2.15 there exists an abelian envelope

C/Kφmax⁡⟶D′.\mathcal{C}/\mathcal{K}_{\varphi}^{\max} \longrightarrow\mathcal{D}'.

Lemma 4.4. ker⁡(φ)=Kφmax⁡\ker(\varphi) = \mathcal{K}_{\varphi}^{\max}.

Proof. Because φ(M)=0\varphi(M) = 0 if and only if Kφ\mathcal{K}_{\varphi} does not lie in the support of MM, it is clear that ker⁡(φ)⊆Kφmax⁡\ker(\varphi) \subseteq\mathcal{K}_{\varphi}^{\max}. We must prove the opposite inclusion.

Let f:X→Yf : X \to Y be a morphism in Kφmax⁡\mathcal{K}_{\varphi}^{\max}. Then f⊗S:X⊗S→Y⊗Sf \otimes S : X \otimes S \to Y \otimes S must lie in the categorical radical Rad⁡(I)\operatorname{Rad}(\mathcal{I}) of the finite additive category I\mathcal{I}. Since I\mathcal{I} has finitely many indecomposables, there exists N>0N > 0 such that Rad⁡(I)N=0\operatorname{Rad}(\mathcal{I})^{N} = 0. We claim that f⊗N⊗S=0f^{\otimes N} \otimes S = 0. Indeed, it may be factored as

X⊗N⊗S→f⊗SY⊗X⊗(N−1)⊗S→⋯Y⊗N⊗S,X^{\otimes N} \otimes S \xrightarrow{f\otimes S} Y \otimes X^{\otimes(N-1)} \otimes S \to\cdots Y^{\otimes N} \otimes S,

and thus by assumption is 00. It follows that S⊗F(f)⊗N=0S \otimes F(f)^{\otimes N} = 0 in D\mathcal{D}. But since D\mathcal{D} is a tensor category, this forces F(f)=0F(f) = 0, completing the proof. □

It now follows that φ\varphi defines a faithful symmetric monoidal functor C/Kφmax⁡→D\mathcal{C}/\mathcal{K}_{\varphi}^{\max} \to\mathcal{D}, and thus by the universal property we obtain a surjective, exact tensor functor F:D′→DF : \mathcal{D}' \to\mathcal{D} fitting into the following diagram:

C→C/Kφmax⁡→D′φ↓↓ω′↓D→ωVer⁡p∞\begin{CD} \mathcal{C} @>>> \mathcal{C}/\mathcal{K}_{\varphi}^{\max} @>>> \mathcal{D}' \\ @V{\varphi}VV @VVV @V{\omega'}VV \\ \mathcal{D} @>{\omega}>> \operatorname{Ver}_{p^\infty} \end{CD}

Since FF is surjective and exact, and D′\mathcal{D}' is finite, D\mathcal{D} must also be finite. Since D,D′\mathcal{D},\mathcal{D}' are finite, there exists exact tensor functors ω,ω′\omega,\omega' as above, by the main theorem of [11]. By [11 Thm. 4.7], ω∘F\omega\circ F must be isomorphic to ω′\omega', and thus

ω′∘φ~≃ω∘φ.\omega' \circ\widetilde{\varphi} \simeq\omega\circ\varphi.

It follows that if D′=Ver⁡p∞\mathcal{D}'=\operatorname{Ver}_{p^\infty}, then φ≃φ~\varphi\simeq\widetilde{\varphi}, which proves the injectivity of the map in Conjecture 1.2. Surjectivity follows from the existence of φ~\widetilde{\varphi}. □

References

A.S.: School of Mathematics and Statistics, UNSW Sydney, NSW 2052, Australia

Email address: [email protected]

References

  1. [1]P. Balmer, Spectra, spectra, spectra—tensor triangular spectra versus Zariski spectra of endomorphism rings, Algebr. Geom. Topol. 10 (2010), no. 3, 1521–1563; MR2661535.DOI
  2. [2]B. D. Boe, J. R. Kujawa and D. K. Nakano, Tensor triangular geometry for classical Lie superalgebras, Adv. Math. 314 (2017), 228–277; MR3658717.arxiv.org/abs/1402.3732
  3. [3]D. J. Benson, P. I. Etingof and V. Ostrik, New incompressible symmetric tensor categories in positive characteristic, Duke Math. J. 172 (2023), no. 1, 105–200; MR4533718.DOI
  4. [4]D. J. Benson, S. Iyengar, H. Krause, and J. Pevtsova, Stratification for module categories of finite group schemes, J. Amer. Math. Soc. 31 (2018), no. 1, 265–302; MR3718455.arxiv.org/abs/1510.06773
  5. [5]D. J. Benson S. Iyengar, H. Krause, and J. Pevtsova, Stratification and duality for unipotent finite supergroup schemes, in Equivariant topology and derived algebra, 241–275, London Math. Soc. Lecture Note Ser., 474, Cambridge Univ. Press, Cambridge, ; MR4327102.
  6. [6]OpenAI, Fiber functors for finite symmetric tensor categories in positive characteristic.DOI
  7. [7]K. Coulembier, Tensor ideals, Deligne categories and invariant theory, Selecta Math. (N.S.) 24 (2018), no. 5, 4659–4710; MR3874701arxiv.org/abs/1712.06248
  8. [8]K. Coulembier, P. I. Etingof, and V. Ostrik, Tensor ideals of abelian type and quantum groups, arXiv preprint arXiv:2511.08859 (2025).arxiv.org/abs/2511.08859
  9. [9]K. Coulembier and J. Flake, Towards higher Frobenius functors for symmetric tensor categories, Selecta Math. (N.S.) 32 (2026), no. 5, Paper No. 98, 54 pp.; MR5134613.arxiv.org/abs/2405.19506
  10. [10]K. Coulembier and J. Flake, A conjecture on the tensor ideal for an elementary p-group generated by the restriction of a Steinberg module. arXiv preprint arXiv:2507.02722.
  11. [11]K. Coulembier and J. Flake, The Benson–Etingof–Ostrik conjecture for finite tensor categories, arXiv preprint arXiv:2610.04096 (2026).arxiv.org/abs/2610.04096
  12. [12]P. Deligne, Catégories tensorielles, Mosc. Math. J. 2 (2002), no. 2, 227–248; MR1944506.DOI
  13. [13]M. Duflo and V. Serganova, On associated variety for Lie superalgebras. arXiv preprint math/0507198.
  14. [14]C. M. Duprieski and J. R. Kujawa, On support varieties for Lie superalgebras and finite supergroup schemes, J. Algebra 525 (2019), 64–110; MR3909818.
  15. [15]P. I. Etingof et al., Tensor categories, Mathematical Surveys and Monographs, 205, Amer. Math. Soc., Providence, RI, 2015; MR3242743.
  16. [16]P. I. Etingof and V. Ostrik, On the Frobenius functor for symmetric tensor categories in positive characteristic, J. Reine Angew. Math. 773 (2021), 165–198; MR4237968.
  17. [17]E. M. Friedlander, Support varieties for rational representations, Compos. Math. 151 (2015), no. 4, 765–792; MR3334895.DOI
  18. [18]E. M. Friedlander and J. Pevtsova, Representation-theoretic support spaces for finite group schemes, Amer. J. Math. 127 (2005), no. 2, 379–420; MR2130619.DOI
  19. [19]M. Gorelik, C. Hoyt, V. Serganova, and A. Sherman, The Duflo–Serganova Functor, Vingt Ans Après, Journal of the Indian Institute of Science, 102(3), pp.961-1000.
  20. [20]C. Hone, F. Klein, B. Pauwels, A. Sherman, O. Yacobi, and V. L. Zhang, Semisimplifying categorical Heisenberg actions and periodic equivalences, arXiv preprint arXiv:2509.07377 (2025).arxiv.org/abs/2509.07377
  21. [21]H. Huang and K. B. Vashaw, Group actions on monoidal triangulated categories and Balmer spectra, Doc. Math. 30 (2025), no. 5, 1055–1083; MR4934522.arxiv.org/abs/2311.18638
  22. [23]D. K. Nakano, K. B. Vashaw and M. T. Yakimov, On the spectrum and support theory of a finite tensor category, Math. Ann. 390 (2024), no. 1, 205–254; MR4800913.DOI
  23. [24]M. Stroiński, Identity in the presence of adjunction, Int. Math. Res. Not. IMRN 2024, no. 18, 12711–12745; MR4798647.arxiv.org/abs/2401.09892
  24. [25]The Stacks Project Authors, Stacks Project, https://stacks.math.columbia.edu (2018).

Paper details

vlnc:W_6pdqpdht
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 3

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents