WorksMathematics
Quasi-F-split heights and derived Morita invariance
Abstract
In this paper, we study quasi-F-splitting from the viewpoint of non-commutative algebraic geometry. Let k be a perfect field of characteristic p. For a proper dg-category over k with a Serre functor, we define quasi-F^e-split heights for e\geq 1 by using finite topological restriction homology with coefficients in a shift of the Serre functor. Using the functoriality of polygonic topological Hochschild homology for laced categories due to Harpaz–Nikolaus–Saunier, we show that these heights are invariant under derived Morita equivalence. For a proper Gorenstein scheme of pure dimension over k, the height for e=1 of its dg-category of perfect complexes coincides with the quasi-F-split height in the sense of Yobuko. For a complete Noetherian Gorenstein local -algebra with residue field k, the dg-category of perfect complexes supported at the closed point recovers the n-quasi-F^e-splitting of Tanaka–Witaszek–Yobuko for all n,e\geq 1. The proofs use the computation of the zeroth homotopy groups of TR with coefficients due to Dotto–Krause–Nikolaus–Patchkoria, Zariski descent, and Serre duality or local duality.
Introduction
The aim of this paper is to construct an analogue of the quasi--split height for non-commutative algebraic varieties, and to compare it with the classical invariants of proper Gorenstein schemes and complete Gorenstein local rings.
Let be a perfect field of characteristic . Quasi--splitting was introduced by Yobuko [21] as a generalization of -splitting, in which a splitting of the Frobenius map is replaced by a splitting involving Witt vectors of finite length. It has applications in birational geometry [11], and the variant with higher Frobenius exponents, quasi--splitting, was introduced by Tanaka–Witaszek–Yobuko [20]. Fedder-type criteria make these notions computable for hypersurfaces [12, 22].
The definition of quasi--splitting involves the structure sheaf, the module structure on Witt vectors and a specific map (see § 2), whereas a general non-commutative algebra has no ring endomorphism .
In this paper, we use finite topological restriction homology , which has Frobenius and restriction maps, in place of Witt vectors. In order to obtain the splitting condition for a proper scheme from its dg-category of perfect complexes, we take coefficients in the Serre functor. For a Gorenstein scheme, this coefficient corresponds to the canonical bundle. Then the lowest homotopy group of a certain pushout of spectra is identified with the top cohomology of the classical pushout tensored with the canonical bundle, and Serre duality translates the injectivity of the induced map into the existence of the required splitting.
The categorical construction
Throughout this paper, all dg-categories are assumed to be small, idempotent-complete and -linear, and to admit a classical generator. Equivalently, such a dg-category is of the form for a dg-algebra over . Recall that a dg-category is proper if the total cohomology of is finite-dimensional for all objects . Let be a proper dg-category. We say that admits a Serre functor if the dual bimodule
is invertible with respect to the derived tensor product. In this case, is a Serre functor of , that is, there are natural equivalences , and we always regard as a dg-functor in this way. These notions do not depend on the choice of .
Fix an integer , and put , which we regard as a coefficient bimodule. We use the polygonic spectra of Krause–McCandless–Nikolaus [9], which refine topological Hochschild homology with coefficients. For an exact endofunctor of , we denote by the finite with respect to the truncation set ; as usual, we have . There are restriction and Frobenius maps
where the Frobenius changes the coefficient from to .
For , we define
and we denote by the induced map. We define
where we put . The integer is part of the data.
Theorem A (Morita invariance; Subsection 4.2). *Fix an integer . Let and be proper dg-categories admitting Serre functors. Then a -linear Morita equivalence induces an equivalence between the diagrams defining and for all , including the Frobenius and restriction maps. In particular, we have
Moreover, if is injective, then so is .
Theorems A and B rely on the functoriality of polygonic THH with respect to laced functors due to Harpaz–Nikolaus–Saunier (Theorem 3.2). A construction is given in the thesis of Saunier [17 Theorem 6.30] and in his lecture notes [18 Corollary 7.3.14], and complete proofs are announced in [7, 6]; see Remark 3.3. Theorem C below does not depend on it.
Comparison with the classical invariants
For a scheme over , we denote by the quasi--split height of in the sense of Yobuko (Definition 2.3). For a Noetherian -finite Gorenstein local -algebra and , we denote by the smallest integer such that is -quasi--split in the sense of [20], in the form given in [22 Definition 2.1] (Definition 2.4).
Theorem B (Proper Gorenstein schemes; Subsection 7.1 and Corollary 7.4). Let be a nonempty proper Gorenstein scheme over of pure dimension , and put . Then, for every , the scheme is -quasi--split if and only if the map
is injective. In particular, we have
Moreover, if is another such scheme and as -linear dg-categories, then and .
Theorem B applies to singular Gorenstein schemes and to smooth proper varieties with nontrivial canonical bundle. The Serre functor cannot be replaced by the identity coefficient: for an ordinary smooth projective curve of genus at least two, the Frobenius acts bijectively on , so that the construction with the identity coefficient gives an injective map, whereas is not quasi--split (Remark 7.5).
Theorem C (Complete local Gorenstein rings; Subsection 8.4). Let be a nonzero complete Noetherian Gorenstein local -algebra with residue field , and put . Let be the dg-category of perfect -complexes supported at . Then is proper, and its Serre functor is ; in particular, . Moreover, for all , the ring is -quasi--split if and only if the map
is injective. In particular, we have
The ring in Theorem C is automatically -finite, since it is a quotient of by the Cohen structure theorem. Theorem C applies to complete hypersurfaces , where is a nonzero element of the maximal ideal. In this case, we have , and the Fedder-type criterion of Yoshikawa [22 Theorem A] computes this categorical invariant. For example, if , is algebraically closed and with , then by Theorem C and [13 Theorem 4.31]; see Example 8.8. The dg-category consists of perfect complexes with finite length cohomology; it is not the singularity category.
Relation to previous work
Yobuko [21 Theorem 4.5] shows that the quasi--split height of a Calabi–Yau variety coincides with the height of its Artin–Mazur formal group. Antieau–Bragg [1 Proposition 1.1 and Theorem 5.1] prove that this formal group, and hence its height, is a derived invariant. Therefore the derived invariance of the quasi--split height of Calabi–Yau varieties, including K3 surfaces, is already known.
For smooth projective varieties, the consequence of Theorem B is new only in a limited range. If is ample or anti-ample, then by Bondal–Orlov [BO01]. If the Kodaira dimension of is positive, then so is that of (see [8 Section 6.1]), and neither nor is quasi--split by [11 Proposition 3.14]. Thus Theorem B gives new information for smooth projective varieties of nonpositive Kodaira dimension which are neither Calabi–Yau nor determined by their derived categories, and for singular Gorenstein schemes. In a related direction, Krah–Vial [KV25] prove that global -regularity is a derived invariant for normal Gorenstein projective varieties.
Our computation relies on the theory of Witt vectors with coefficients due to Dotto–Krause–Nikolaus–Patchkoria [4] and on the polygonic spectra of [9]. The localization of THH with coefficients, which we need for schemes, is provided by the theory of laced categories due to Harpaz–Nikolaus–Saunier [7], and for THH and TR without coefficients we use the localization theorems of Blumberg–Mandell [2]. Liu–Liu [15 Theorems 3.9 and 3.11] prove that coefficient Morita equivalences induce equivalences of THH with coefficients and of the associated genuine restriction systems; Theorem 3.2, due to Harpaz–Nikolaus–Saunier, provides the corresponding functoriality for polygonic spectra. On the classical side, Lemma 7.2 and Proposition 8.5 are versions of the cohomological criteria in [11 Lemmas 2.10 and 3.13], and dualizing complexes on Witt rings are studied in [10 Section 9].
The global comparison (Theorem B) concerns only the case , whereas the local comparison (Theorem C) holds for all and uses Matlis duality over . If , then the classical pushout is only a -module, and a global comparison would require duality for -modules. This leads to the following question.
Question 1.1. Does Theorem B extend to all , with respect to the definition of quasi--splitting for schemes in [20]?
We do not discuss quasi--regularity in this paper. Its definition [20] involves a multiplier which runs over all non-zero-divisors, and Theorem A does not provide the corresponding structure on dg-categories.
Organization of the paper
In § 2, we fix conventions and recall the definitions of quasi--splitting and quasi--splitting. In § 3, we recall finite TR with coefficients and its functoriality for laced categories, and we deduce its invariance under derived Morita equivalence (Corollary 3.4). In § 4, we define the invariants and prove Theorem A. In § 5, we compute the zeroth homotopy groups of TR with coefficients in a line bundle. In § 6, we prove Zariski descent for TR with coefficients in a line bundle, and we compute the lowest homotopy group of derived global sections. Using these results, we prove Theorem B in § 7 and Theorem C in § 8, where we also give examples.
Use of AI. The proof strategy of this paper is due to the author. In working out the details of the arguments, the author benefited greatly from ChatGPT Astra 6. Claude Opus 5.5 was of great help in writing this paper, in particular with its English and grammar. The author takes full responsibility for all the contents of this paper.
Conventions and classical definitions
Conventions
Throughout this paper, we fix a prime and a perfect field of characteristic . All tensor products of bimodules are derived, and we often omit from the notation. We regard a dg-algebra as an associative ring spectrum via the Eilenberg–MacLane functor. We always consider THH relative to the sphere spectrum.
For a complex , we use the convention , so that the shift corresponds to the suspension. A spectrum is connective if its negative homotopy groups vanish. For a finite group , we denote by , and the homotopy fixed points, the homotopy orbits and the Tate construction, respectively.
For an exact endofunctor of a stable -category , we consider the associated bimodule
With this convention, the bimodule associated with for a line bundle is . If , we use the same symbol for an endofunctor and the corresponding -bimodule. Following [7], we call a pair of a stable -category and a -bimodule a laced category.
Witt vectors and quasi--splitting
Let be an -algebra. We denote by the ring of -typical Witt vectors of length , by the Teichmüller lift of , and by
the restriction and the Verschiebung. Let be the absolute Frobenius. The induced ring endomorphism of is given by and preserves the length. We distinguish it from the Frobenius , which decreases the length and corresponds to the Frobenius of finite TR (Proposition 3.5). For a -module and , we denote by the -module obtained by restriction of scalars along .
Definition 2.1. Let . We define a -module and a -linear map by the pushout diagram of -modules
where is regarded as a -module via . We put and .
If , then is not an -module in general; for , we have the following description.
Lemma 2.2. The -module structure on factors through , and there is a natural isomorphism
Under this isomorphism, the -module structure on and the map are given by
Proof. Since is surjective with kernel , the pushout is the quotient of by the image of under . Since has characteristic , we have
Hence the elements of act on the quotient by zero, and the module structure factors through . The formulas follow by evaluating on Teichmüller lifts.
This construction is compatible with localization, and hence it sheafifies on -schemes. For an -scheme , we denote by and the resulting -module and map.
Definition 2.3. Let be an -scheme and . We say that is -quasi--split if admits an -linear retraction. The quasi--split height of is defined by
This definition agrees with [21 Definition 4.1].
Definition 2.4. Let be a Noetherian -finite Gorenstein local -algebra, and let be the canonical (dualizing) module of (see [10 Section 9]). Following [20 Sections 3.1 and 3.5] in the form of [22 Definition 2.1], we say that is -quasi--split if the map
is surjective. We denote by the smallest such integer , where we put if no such exists.
Finite TR with coefficients
Polygonic spectra and finite TR
3.1. Polygonic spectra and finite TR. For a ring spectrum and an -bimodule , we put
The cyclic bar construction gives a -action on this spectrum, and the Tate diagonal gives natural maps
which are equivariant with respect to the residual -actions. By [9 Theorem 6.31], these data form a polygonic spectrum , functorially in the -bimodule . We only use the part indexed by powers of .
Definition 3.1. For , we put and define
where the right-hand side is the finite TR of polygonic spectra with respect to the truncation set , which is defined as a mapping spectrum in [9 Section 3].
Explicitly, writing , the spectrum is the equalizer of two maps
one of which is induced by the canonical maps and the other by the maps . In particular, finite TR preserves finite limits of polygonic spectra.
The inclusion induces the restriction map . The shift of by is identified with , and the Frobenius of finite TR gives a map
These maps commute with the restriction maps and can be iterated.
Functoriality and Morita invariance
3.2. Functoriality and Morita invariance. We use the following functoriality of polygonic THH for laced categories, due to Harpaz–Nikolaus–Saunier.
Theorem 3.2 (Harpaz–Nikolaus–Saunier). The following assertions hold.
(i) The construction , together with the cyclic actions and the Tate diagonals, extends to a functor from the -category of laced categories to the -category of polygonic spectra in the sense of [9]; see [17 Theorem 6.30] and [18 Corollary 7.3.14]. For , it recovers the cyclotomic structure on .
(ii) If is a connective ring spectrum and is a connective -bimodule, then is naturally equivalent to the polygonic spectrum of [9 Theorem 6.31]; see [17 Section 7.2].
(iii) The structure maps satisfy , naturally in ; see [17 Section 6.4] and [18 Section 7.2]. In particular, the shift of by is identified with , naturally in .
Remark 3.3. The construction in [17 Section 6.4] is presented as a sketch, statement (ii) is stated without proof in [17 Section 7.2] (see also [17 Proposition 7.16]), and [18] consists of lecture notes. A genuine refinement of (i) is stated in [17 Theorem 7.18], and complete proofs are announced in [7 Introduction] and [6]. Theorems A and B depend on Theorem 3.2. Theorem C does not: its coefficient is equivalent to the identity, and if is defined via the cyclotomic spectrum , then its proof only uses the functoriality of in for a fixed dg-algebra [9 Theorem 6.31], the comparison of TR of cyclotomic spectra with the classical TR [9], and the invariance and localization theorems for THH and TR [2].
For a laced category , we write for the first level of and . By Theorem 3.2, the restriction and Frobenius maps of finite TR are natural with respect to laced functors.
Corollary 3.4. Let and be dg-algebras, and let be an --bimodule and a --bimodule such that
as bimodules. For an -bimodule , we put . Then there is an equivalence of polygonic spectra
which induces equivalences between the finite TR spectra compatible with the Frobenius and restriction maps.
Proof. The functor is an equivalence, and we have a natural equivalence since . Hence it underlies an equivalence of laced categories . By Theorem 3.2(i), it induces an equivalence of polygonic spectra, which is compatible with the identifications of the shifts by by Theorem 3.2(iii). Since finite TR and its restriction and Frobenius maps are natural in polygonic spectra, we obtain the claim. □
In particular, for a dg-category and an exact endofunctor of , the spectra and , together with their restriction and Frobenius maps, are invariant under derived Morita equivalence. If with and connective, then they are computed by by Theorem 3.2(ii).
Connectivity and the zeroth homotopy groups.
Proposition 3.5. Let be a connective ring spectrum and a connective -bimodule. Then is connective, and there is a natural isomorphism
compatible with restriction, Frobenius and Verschiebung. Here denotes the Witt vectors of length with coefficients in in the sense of [4].
Proof. The fiber of is given by the homotopy orbits of the corresponding cyclic tensor product. It is connective, since all tensor factors are connective and taking homotopy orbits preserves connectivity. Hence the first assertion follows by induction on ; compare [9 Lemma 3.14 and the proof of Proposition 5.2].
The second assertion is [4 Theorem 2.5], and the compatibility with the operators follows from [4 Proposition 2.13] together with the compatibility with restriction in its construction. Our index is the number of elements of the truncation set; thus in this paper is denoted by in [4]. □
The invariants and their Morita invariance
4.1. Definition. Let be a dg-category and an exact endofunctor of . We define by (1), and we denote by
the composite of the inclusion of the second summand and the canonical map to the cofiber. Thus we have a pushout square of spectra
Definition 4.1. Let be an integer and . We define
If is proper and admits a Serre functor , we put
Lemma 4.2. There are maps
induced by the restriction maps, whose composite with is . In particular, if is injective, then is injective for every .
Proof. We apply the restriction maps to the two TR terms and the identity to the THH term. Since and the two composites of restriction maps to THH agree, we obtain a map of pushout diagrams, and hence the required map. If is injective, then any element killed by is also killed by the composite , and hence it is zero. □
Proposition 4.3. A derived Morita equivalence which carries to induces an equivalence of arrows
In particular, we have for all and $d.
Proof. By Corollary 3.4, both arrows in the diagram defining are identified. Since pushouts are functorial, the pushouts and the maps from the THH terms are also identified. Hence the two maps on are conjugate by isomorphisms, and one of them is injective if and only if the other is. □
4.2. Proof of Theorem A. Let be the given equivalence, and choose dg-algebras and with and . Then is given by an invertible bimodule, and it carries the -bimodule to the -bimodule , since both are the -linear duals of the corresponding mapping bimodules. Hence and , and the first assertion follows from Proposition 4.3. The last assertion is Lemma 4.2. □
Remark 4.4. The definition of does not use a distinguished unit element in or in TR. Such an element is not invariant under Morita equivalence: under the Morita equivalence , the class of the identity matrix corresponds to .
Witt vectors with coefficients in a line bundle
Let be a Noetherian -scheme and an invertible -module. We denote by
the sheaf of Witt vectors of length with coefficients in , regarded as a symmetric bimodule, which is obtained by gluing the local constructions of [4].
Lemma 5.1. The sheaf is an invertible -module. More precisely, let and be local generators of , where is a unit. Then the transition map from the trivialization given by to the trivialization given by is
that is, the multiplication by .
Proof. If is trivialized by , then the tensor powers appearing in the construction of Witt vectors with coefficients are identified with the corresponding powers of , and hence is identified with . The multiplication by on acts on the -th coordinate, which lies in , by ; this is the multiplication by the Teichmüller lift . Since , these transition maps satisfy the cocycle condition, and the local modules glue. □
Remark 5.2. By Lemma 5.1, is the Teichmüller lift of . In particular, if is normal and for a Cartier divisor , then is isomorphic to the Witt divisorial sheaf of [11 Subsection 2.6], which is locally generated by if locally. For , this sheaf should be distinguished from the dualizing sheaf of .
We have the maps
Proposition 5.3. The pushout of the above two maps in the category of sheaves of abelian groups is naturally isomorphic to
where is equipped with the -module structure of Lemma 2.2. Under this isomorphism, the induced map from is $\Phi_{X,n}\otimes L.
Proof. The map is surjective with kernel
For general coefficients, the composite is the norm map of the cyclic action on tensor powers. Since this action is trivial on tensor powers of an invertible module regarded as a symmetric bimodule, we have on . Hence the pushout is the sheaf of abelian groups
We equip this quotient with the -module structure given by . If is a local generator of , then this quotient is identified with by Lemma 2.2. If is another local generator, then the transition map is the multiplication by by Lemma 5.1, which is exactly the transition map of with respect to the -module structure of Lemma 2.2.
Moreover, with respect to the generator , an element of is sent to the class of , which corresponds to . Hence these identifications glue, and the induced map is . □
Corollary 5.4. Let be an affine open subscheme of , and regard as a symmetric -bimodule. Then the spectra and are connective, and there are natural isomorphisms
compatible with the restriction to smaller affine open subschemes. Under these isomorphisms, the map corresponds to .
Proof. By Proposition 3.5, all the terms in the diagram defining are connective, and their zeroth homotopy groups are given by Witt vectors with coefficients. Since a cofiber of a map of connective spectra is connective and its is the cokernel of the induced map on , the assertion follows from Proposition 5.3. □
Zariski descent and the lowest homotopy group
Zariski descent
Let be a quasi-compact separated scheme and a line bundle on . For an open subscheme , we denote by the endofunctor of .
Proposition 6.1. The polygonic spectra , the spectra and the spectra satisfy Zariski descent; that is, they send a cover of a quasi-compact open subscheme by two quasi-compact open subschemes to a Mayer–Vietoris square. Moreover, if is Noetherian of finite Krull dimension, then these spectra for are identified with the derived global sections of the associated Zariski sheaves of spectra.
Proof. For quasi-compact open subschemes , the restriction functor commutes with , and hence it underlies a laced functor . By Theorem 3.2, we obtain a functor from quasi-compact open subschemes of to polygonic spectra. Since the Tate construction is exact, finite limits of polygonic spectra are computed levelwise, and hence it suffices to show that satisfies Mayer–Vietoris for every .
Let be a quasi-compact open immersion, and put . For every , the endofunctor preserves , and the endofunctor induced on is ; on Ind-categories, this follows from
Thus the localization sequence gives, for each , a laced Verdier sequence up to idempotent completion. By [7 Proposition 5.18 and Corollary 5.19], with these coefficients sends it to a fiber sequence. If , then the dg-categories of perfect complexes supported on in and in are equivalent, compatibly with the coefficients, by excision. Comparing the localization sequences, we obtain a Mayer–Vietoris square for .
Since finite preserves finite limits of polygonic spectra, and since the pushout defining is exact and its Frobenius term is natural by Theorem 3.2(iii), the spectra and also satisfy Mayer–Vietoris. Finally, if is Noetherian of finite Krull dimension, a finite affine open cover and its intersections compute the associated sheaf by iterated Mayer–Vietoris squares, and the convergence of the descent spectral sequence (Lemma 6.2) identifies the constructions for with the derived global sections. □
The lowest homotopy group
Lemma 6.2. Let be a Noetherian scheme of dimension , and let be a sheaf of spectra on the Zariski site of whose homotopy sheaves vanish for . Then we have
and the isomorphism is natural in .
Proof. Since the underlying topological space of is Noetherian of Krull dimension , the -topos of sheaves on is locally of homotopy dimension at most by [16 Proposition 7.2.4.7 and Corollary 7.2.4.17], and hence it is hypercomplete by [16 Corollary 7.2.1.12]. Moreover, the Zariski cohomological dimension of is at most by [19 Tag 02UZ]. Hence the descent spectral sequence
converges. Since unless and , there are no nonzero terms of total degree , and the only term of total degree is . No nonzero differential enters this term, since its source would have , and no nonzero differential leaves it, since its target would have . Hence we obtain the claim. □
Lemma 6.3. Let be a Noetherian ring, let , and put and for a subset . Let be a cubical diagram of connective spectra indexed by the subsets of , and let be its total fiber. Then for . Moreover, if the diagram is isomorphic to the diagram of localizations of an -module , then there is a natural isomorphism
Proof. Filtering the total fiber by the cardinality of , we obtain a spectral sequence
where and . Hence for , and is the -th cohomology of the complex . Under the assumption, this complex is the Čech complex of with respect to , whose -th cohomology is . □
Proper Gorenstein schemes
Let be a nonempty proper Gorenstein scheme over of pure dimension . Then is invertible, and is a dualizing complex of . Since is proper, is proper. By [19 Tag 09IS], there is a perfect complex which generates ; since the compact objects of are the perfect complexes [19 Tag 09M1], it is a classical generator of . The Serre functor of is
see [19 Tag 0FY8]. Hence the coefficient in Definition 4.1 is .
Proof of Theorem B
Proposition 7.1. The map is naturally identified with the map
Proof. By Proposition 6.1, the source and the target of , together with the map between them, are identified with the derived global sections of the corresponding Zariski sheaves of spectra. By Theorem 3.2(ii) on affine open subschemes and Corollary 5.4, the homotopy sheaves of these sheaves vanish in negative degrees, and their zeroth homotopy sheaves and the map between them are identified with
Hence the assertion follows from Lemma 6.2 and its naturality. □
Lemma 7.2. Let be a coherent -module and an -linear map. Then the induced map
is injective if and only if admits an -linear retraction.
Proof. Since is an equidimensional proper Cohen–Macaulay scheme, Serre duality gives a natural isomorphism
for every coherent -module ; see [19 Tag 0F VZ]. Applying it to and , and using the invertibility of , we identify the dual of the above map with
Since all the groups are finite-dimensional over , the above map is injective if and only if is surjective. If is surjective, then a preimage of is a retraction of . Conversely, if is a retraction of , then for every endomorphism of . □
Proof of the first assertion of Theorem B. Since is of finite type over a perfect field, the Frobenius of is finite, and hence is a coherent -module. By Proposition 7.1 and Lemma 7.2, the map is injective if and only if admits an -linear retraction, that is, is -quasi--split. Since this holds for every , we obtain . □
Dimension and derived invariance
Let be a nonzero proper dg-category with a Serre functor and a classical generator . Following Elagin–Lunts [5 Definition 5.3], we put
and define the upper and lower Serre dimensions by
They do not depend on the choice of [5], and they are invariant under -linear equivalences, which commute with Serre functors.
Proposition 7.3. Let be as above. Then . In particular, is invariant under -linear equivalences of these dg-categories.
Proof. For smooth projective , this is [5 Lemma 5.6]; the same argument applies. Let be a classical generator of . Since is perfect, its cohomology sheaves are concentrated in a bounded range of degrees, and this range does not change after tensoring with . Since coherent cohomology on vanishes in degrees greater than , there are integers , independent of , such that
Since
this group vanishes unless . On the other hand, at least one of these groups is nonzero, since otherwise we would have . Hence , and dividing by , we obtain the claim. □
Corollary 7.4. Let and be nonempty proper Gorenstein schemes over of pure dimension. If there is a -linear derived Morita equivalence , then
In particular, quasi--splitting and -splitting are preserved by such equivalences.
Proof. The assertion follows from Proposition 7.3, Theorem A and the first assertion of Theorem B, since is -split if and only if . □
The role of the Serre functor
Remark 7.5. Assume that is algebraically closed, and let be an ordinary smooth projective curve over of genus at least two. Since the Kodaira dimension of is one, is not quasi--split by [11 Proposition 3.14], and hence by Theorem B. On the other hand, the Frobenius acts bijectively on . The argument of Proposition 7.1 with identifies with the Frobenius on , which is injective. Hence the identity coefficient does not recover the quasi--split height, and the shift of the Serre functor is necessary in Theorem B.
Complete local Gorenstein rings
Throughout this section, let be a nonzero complete Noetherian Gorenstein local -algebra with residue field , and put . As mentioned in the introduction, is -finite.
The dg-category of supported perfect complexes.
Proposition 8.1. The dg-category is proper, has a classical generator, and admits the Serre functor .
Proof. For , the complex is a perfect -complex with finite length cohomology, and hence is proper. A Koszul complex on a system of parameters of is a classical generator of ; see [19 Tag 0A9A, Tag 0A9B]. We put , so that .
Since is Gorenstein, is a normalized dualizing complex of . For an -module of finite length, we have for and by local duality [19 Tag 0A81]. Since is a coefficient field, we have , and the evaluation at identifies with ; see [19 Tag 0A7Q]. Hence we obtain a natural equivalence
of functors on -complexes with finite length cohomology. Since the -bimodule structure on is -linear, applying this to gives an equivalence of -bimodules . Since is perfect, the trace pairing , , induces an equivalence of -bimodules . Hence as -bimodules, that is, the Serre functor of is .
Consequently, we have .
The zeroth homotopy groups.
For a discrete commutative -algebra , we put .
Proposition 8.2. The spectrum is connective, and there is a natural isomorphism
under which the map corresponds to .
Proof. By Proposition 3.5, the three spectra , and in the diagram defining are connective, and their zeroth homotopy groups are , and , respectively.
Since has characteristic , the Frobenius is given by
Hence it factors as the surjection followed by . The other map factors as . In particular, both maps vanish on .
Since all the spectra are connective, of the pushout is the pushout of abelian groups of the zeroth homotopy groups. Dividing the source by the common kernel , we identify it with the underlying abelian group of the pushout of -modules in Definition 2.1, compatibly with the maps from . This also equips with the -module structure of .
Local cohomology.
Proposition 8.3. The map is naturally identified with the map
Proof. Since the coefficient is the identity, the polygonic spectra here come from cyclotomic spectra, and finite TR agrees with the classical TR [9]. Put . By the localization theorem [2 Theorems 1.1 and 1.3], is identified with the fiber of , compatibly with the cyclotomic structures. Since finite TR and the cofiber defining are exact, is also identified with the fiber of .
Choose a system of parameters of . Then the principal open subschemes cover . By Zariski descent [2], and are identified with the total fibers of the cubes and , which consist of connective spectra. For THH, the diagram of zeroth homotopy groups is the diagram of localizations of .
For , by Proposition 8.2, the diagram of zeroth homotopy groups is , which is the diagram of localizations of the -module at the Teichmüller lifts . Indeed, we have , and localizing at is the same as localizing it at . Since the kernel of is nilpotent, the ideal of has the maximal ideal as its radical.
Applying Lemma 6.3 to the Noetherian ring (Lemma 8.4) and the elements , we obtain
where both local cohomology groups are taken over with respect to its maximal ideal; the first one agrees with the usual local cohomology of . By naturality, we obtain the claim. □
Duality over and the proof of Theorem C
For the convenience of the reader, we include a proof of the following ring-theoretic facts.
Lemma 8.4. The ring is a complete Noetherian Cohen–Macaulay local ring of dimension . Moreover, is a finite -module.
Proof. Put . The Verschiebung filtration of has graded pieces
On the -th graded piece, acts as the multiplication by ; in particular, the action of on each graded piece factors through the surjection .
Firstly, we prove that is Noetherian by induction on , using the exact sequence
The kernel is isomorphic to , which is a finite -module since is -finite. Hence it is a Noetherian -module. Since is a Noetherian ring by the induction hypothesis, it is also a Noetherian -module, and hence is Noetherian.
Secondly, since the kernel of is nilpotent, is local, its spectrum is homeomorphic to , and its dimension is . On each graded piece , the topology defined by the maximal ideal of coincides with the -adic topology, since the ideal generated by the -th powers of the elements of is -primary. These finite -modules are complete. Since completion is exact for finite modules over a Noetherian ring, the above exact sequence and induction on show that is complete.
Thirdly, let be a system of parameters of , which is a regular sequence since is Cohen–Macaulay. On the -th graded piece, act as , which form a regular sequence on . Since a sequence which is regular on modules and is regular on every extension of by , we see by induction on the filtration that is a regular sequence on . Hence is Cohen–Macaulay.
Finally, has a finite filtration whose graded pieces are , which are finite -modules. Hence is a finite -module. Since is also a finite -module, so is the pushout . □
Proposition 8.5. The map (5) is injective if and only if is -quasi--split.
:::
Proof. Let be a canonical module of , and let
be the Matlis duality over . By Lemma 8.4 and local duality [19 Tag 0A81], we have a natural isomorphism
for every finite -module . The local cohomology modules are Artinian, and Matlis duality is exact and faithful on Artinian modules. Hence, applying to (5), we see that (5) is injective if and only if the map
is surjective. This is the condition in Definition 2.4. □
Proof of Theorem C. By Proposition 8.1, the dg-category is proper and its Serre functor is , and hence . By Proposition 8.3, the map is identified with the map (5) of local cohomology groups. By Proposition 8.5, this map is injective if and only if is -quasi--split. Since this holds for every , we obtain . □
For , Proposition 8.5 and [11 Lemma 3.13] show that coincides with the quasi--split height of in the sense of Definition 2.3.
Remark 8.6 (A dg-algebra model of a singularity). For a Koszul complex as in the proof of Proposition 8.1, the dg-algebra is proper, and we have . Hence Theorem C expresses the classical heights in terms of a proper dg-algebra model of , and the result does not depend on the choice of .
Examples
Example 8.7. Let . Then is a zero-dimensional complete Gorenstein local ring, and is a proper dg-category which is not smooth and whose Serre functor is the identity. For all , , we have by the definition of the pushout. Since , the map (5) is itself, and hence for all by Theorem C. The same argument applies to every zero-dimensional Gorenstein local -algebra which is not a field: a nonzero element of the socle satisfies , and hence . Thus finite heights greater than one only appear in positive dimension.
Example 8.8. Assume that is algebraically closed of characteristic 2. For , let
which is a rational double point. By [13 Theorem 4.31], the quasi--split height of is . Hence, by Theorem C and the comparison for above, we have
In particular, the invariant takes every positive integer as a value on proper dg-categories of the form .
References
- [1]Benjamin Antieau and Daniel Bragg, Derived invariants from topological Hochschild homology, Algebr. Geom. 9 (2022), 364–399, doi:10.14231/AG-2022-011; arXiv:1906.12267.DOI
- [2]Andrew J. Blumberg and Michael A. Mandell, Localization theorems in topological Hochschild homology and topological cyclic homology, Geom. Topol. 16 (2012), 1053–1120, doi:10.2140/gt.2012.16.1053; arXiv:0802.3938.DOI
- [4]Emanuele Dotto, Achim Krause, Thomas Nikolaus, and Irakli Patchkoria, Witt vectors with coefficients and TR, Proc. Lond. Math. Soc. (3) 130 (2025), Paper No. e70047, doi:10.1112/plms.70047; arXiv:2312.12971.DOI
- [5]Alexey Elagin and Valery A. Lunts, Three notions of dimension for triangulated categories, J. Algebra 569 (2021), 334–376, doi:10.1016/j.jalgebra.2020.10.027; arXiv:1901.09461.DOI
- [6]Yonatan Harpaz, Thomas Nikolaus, Maxime Ramzi, and Victor Saunier, Trace methods for stable categories II: The cyclotomic trace map, In preparation, 2026.
- [7]Yonatan Harpaz, Thomas Nikolaus, and Victor Saunier, Trace methods for stable categories I: The linear approximation of algebraic K-theory, arXiv:2411.04743; numbering as in arXiv:2411.04743v2, 2024.arxiv.org/abs/2411.04743
- [8]Daniel Huybrechts, Fourier–Mukai transforms in algebraic geometry, Oxford Mathematical Monographs, The Clarendon Press, Oxford University Press, Oxford, 2006.DOI
- [9]Achim Krause, Jonas McCandless, and Thomas Nikolaus, Polygonic spectra and TR with coefficients, arXiv:2302.07686, 2023.arxiv.org/abs/2302.07686
- [10]Tatsuro Kawakami, Teppei Takamatsu, Hiromu Tanaka, Jakub Witaszek, Fuetaro Yobuko, and Shou Yoshikawa, Quasi-F-splittings in birational geometry III, arXiv:2408.01921, 2024.arxiv.org/abs/2408.01921
- [11]———, Quasi-F-splittings in birational geometry, Ann. Sci. Éc. Norm. Supér. (4) 58 (2025), 665–748, doi:10.24033/asens.2614; arXiv:2208.08016.DOI
- [12]Tatsuro Kawakami, Teppei Takamatsu, and Shou Yoshikawa, Fedder type criteria for quasi-F-splitting I, arXiv:2204.10076, 2022.arxiv.org/abs/2204.10076
- [13]———, Fedder type criteria for quasi-F-splitting II, arXiv:2511.17270, 2025.arxiv.org/abs/2511.17270
- [15]Haoyang Liu and Tianle Liu, Intrinsic restriction traces and toric dynamics, arXiv:2609.08129, 2026.arxiv.org/abs/2609.08129
- [16]Jacob Lurie, Higher Topos Theory, Annals of Mathematics Studies, vol. 170, Princeton University Press, Princeton, NJ, 2009.
- [17]Victor Saunier, Trace methods for stable ∞-categories, Ph.D. thesis, Université Sorbonne Paris Nord, 2025.
- [18]———, Topological Hochschild homology, traces and higher categories, Lecture notes, Universität Bielefeld, version of October 2026, available at the author’s webpage, 2026.
- [20]Hiromu Tanaka, Jakub Witaszek, and Fuetaro Yobuko, Quasi-Fᵉ-splittings and quasi-F-regularity, arXiv:2404.06788, 2024.arxiv.org/abs/2404.06788
- [21]Fuetaro Yobuko, Quasi-Frobenius splitting and lifting of Calabi–Yau varieties in characteristic p, Math. Z. 292 (2019), 307–316, doi:10.1007/s00209-018-2198-7; arXiv:1704.05604.DOI
- [22]Shou Yoshikawa, Fedder-type criterion for quasi-Fᵉ-splitting and quasi-F-regularity, To appear in Nagoya Math. J.; arXiv:2505.09015, 2025.