This note proves that smoothness and Gorensteinness are Fourier–Mukai invariants for proper flat families of Noetherian algebraic spaces. Moreover, a kernel realizing such a partnership is perfect if, and only if, the families are smooth.
Introduction
Fourier–Mukai partnerships provide rich comparisons of geometric objects through their derived categories. It first appeared in the work of [10], and has been extended to relative settings [6, 4, 5]. A fundamental question is which geometric properties are preserved under Fourier–Mukai partnerships.
We first fix notation. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces for i=1,2. Denote by pi:Y1×SY2→Yi the projections. For E∈Dcohb(Y1×SY2), the associated integral transform is ΦE:Dqc(Y1)→Dqc(Y2) where A↦R(p2)∗(L(p1)∗A⊗LE). We say that Y1 and Y2 are Fourier–Mukai S-partners if ΦE restricts to an equivalence Dcohb(Y1)→Dcohb(Y2) for some such kernel E∈Dcohb(Y1×SY2).
Results
For smooth projective schemes over a field, kernels of such transforms are perfect complexes. For nonsmooth families, perfectness need not hold.
Theorem 1.1. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by a kernel E∈Dcohb(Y1×SY2). Then the following are equivalent:
In particular, if either family has a nonsmooth fiber, then kernels giving a Fourier–Mukai partnership cannot be perfect. Consequently, bounded pseudocoherent kernels, rather than perfect ones, are crucial in relative Fourier–Mukai theory.
It is not possible to replace smoothness with regularity in Theorem 1.1. Over a field, regularity can be detected by the equality Perf(X)=Dcohb(X). However, over an imperfect field, a regular proper scheme need not be smooth. For such a scheme, no Fourier–Mukai autoequivalence has a perfect kernel. We give a concrete case.
Example 1.2. Consider any regular scheme Y proper over a field k that is not smooth (i.e. not geometrically regular). For instance, let k:=F3(t) be the rational function field in one variable, and let Y be the projective plane curve over k given by the equation
y2z+x3−tz3=0
[7 Remark 16]. By Proposition 5.10, any autoequivalence of the form ΦE on Dcohb(Y) with E∈Dcohb(Y×kY) satisfies E∈/Perf(Y×kY). For example, the pushforward of OY along the diagonal, i.e. the kernel of the identity functor, is not perfect.
Next, we show that Gorensteinness is preserved under Fourier–Mukai partnerships.
Theorem 1.3. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by a kernel E∈Dcohb(Y1×SY2). Then f1 is Gorenstein if, and only if, f2 is Gorenstein.
For projective schemes over an algebraically closed field, [6 Theorem 4.4] shows that Gorensteinness passes to Fourier–Mukai partners under Cohen–Macaulay and equidimensionality constraints. Over a field, a proper variety is Gorenstein if, and only if, its category of perfect complexes admits a Serre functor. See [1 Lemma 6.6] for projective varieties and [13 Proposition 2.1] for proper varieties (i.e. integrality included).
Theorem 1.3 strengthens these statements for Gorensteinness to proper flat families of Noetherian algebraic spaces. We do not require the fi to be projective, nor do we impose Cohen–Macaulay, equidimensionality, or integrality hypotheses. Moreover, the result applies over any Noetherian base, including in mixed characteristic.
Methodology
For the proof of Theorem 1.1, we base change the partnership along morphisms from fields to S. This uses [4 Theorem 10.13 and Corollary 6.18]. This reduces the question to regularity, which is characterized by the equality Perf=Dcohb. See Lemma 5.6. For the perfectness of kernels, we use that perfectness is detected on fibers. See Lemma 5.9. Moreover, over a regular base, we show that a perfect kernel forces the partners to be regular. See Lemma 5.8.
Our proof of Theorem 1.3 does not pass to fibers, and it is independent of [6] (e.g. l.c.i. cycles are not used). Instead, we work over the base with relative Serre functors. This is motivated by [14 Definition 2.5]. Specifically, we prove that a proper flat morphism is Gorenstein if, and only if, its category of perfect complexes admits a relative Serre functor. See Lemma 4.4. This is a relative version of [1 Lemma 6.6].
The relationship with diagonal morphisms with smoothness is classical: for a finite type morphism of locally Noetherian schemes, smoothness is equivalent to flatness and perfectness of the diagonal [15 Tag 0fDP]. Lunts–Schnürer have given a dg-categorical viewpoint which connects geometric smoothness to homological smoothness of enhancements of Perf[9]. Calabrese showed a reconstruction result for algebraic spaces with affine diagonals and proper flat morphisms with Gorenstein fibers with suitable ampleness constraints [3]. Our results complement these perspectives by addressing invariance of smoothness and Gorensteinness without projectivity- or ampleness-type constraints.
Acknowledgments. The authors thank Kabeer Manali Rahul and Nebojsa Pavic for discussions. Lank was supported by the ERC Advanced Grant 101095900-TriCatApp. Guisado Villalgordo was supported under the grant PRE2022-104796 by the Spanish Ministry of Science and Innovation.
AI disclosure. The authors used large language models (LLMs) for assistance with proofreading (e.g. grammar, notation, and exposition) and checking arguments (e.g. identifying possible gaps or errors). The main results and methods are due to the authors.
Preliminaries
Generation
We discuss a form of generation for triangulated categories. See [BV03] for details. Let T be a triangulated category with shift functor [1]:T→T. Consider a subcategory S⊆T. A triangulated subcategory of T is called thick if it is closed under direct summands. Denote by ⟨S⟩ the smallest thick subcategory of T containing S. If S consists of a single object G, we write ⟨G⟩:=⟨S⟩. Set add(S) to be the smallest strictly full subcategory of T containing S that is closed under shifts, finite coproducts, and direct summands. Inductively, let ⟨S⟩0 consist of all zero objects in T, ⟨S⟩1:=add(S), and
⟨S⟩n:=add{cone(ϕ):ϕ∈HomT(⟨S⟩n−1,⟨S⟩1)}.
It can be checked that ⟨S⟩=⋃n=0∞⟨S⟩n. We say E∈T is finitely built byS if E∈⟨S⟩.
Example 2.1. Let X=Spec(R) where R is a regular ring. Then ⟨OX⟩=Dcohb(X). See [15 Tag 066Z] and [11 Lemma 1.2].
Algebraic spaces
We follow [15] for conventions. We call a morphism natural if it is functorial, and canonical if it is given by an explicit construction that need not be functorial. Let Y be a quasi-compact algebraic space. An étale presentation of Y is a finitely presented, étale, surjective morphism U→Y from a scheme.
Notation
We work with the small étale ringed site (Yet,OY). Write D(Y) for the unbounded derived category of sheaves of OY-modules. Define Dqc(Y) to be the strictly full subcategory of D(Y) consisting of complexes with quasi-coherent cohomology. For ?∈{b,+,−}, let Dqc?(Y) be the strictly full subcategory of Dqc(Y) with prescribed cohomological vanishing. Denote by Perf(Y) the strictly full subcategory of Dqc(Y) consisting of perfect complexes. If Y is Noetherian, Dcohb(Y) is the strictly full subcategory of Dqc(Y) consisting of bounded pseudocoherent complexes. Recall that the derived internal Hom RHomOY is formed in the ambient category D(Y)[15 Tag 08J7]. In particular, it need not agree with the right adjoint to tensor product in Dqc(Y), which is denoted by RHom. See [4 Lemma 2.11] and [15 Tag 0A8A].
Duality
We use the duality conventions of [4 §2.4] and [12]. Let f:Y→X be a proper morphism of Noetherian algebraic spaces. As f is proper, f! is naturally isomorphic to the right adjoint f× of Rf∗ on quasi-coherent derived categories. See [4 Lemma 2.22]. Recall that f! is the upper shriek functor. It is constructed from Nagata compactifications, and defined on Dqc. See [4 §2.4] for details. In particular, loc. cit. discusses the results of [12 Theorem 1.8] for algebraic spaces, where Neeman’s result constructs the so-called upper shriek functor (e.g. f!) on Dqc and not just on Dqc+.
Integral transforms
We refer to [4]. Let S be a Noetherian algebraic space. Suppose Y1 and Y2 are proper and flat algebraic S-spaces. Denote by pi:Y1×SY2→Yi the canonical projections. For E∈Dqc(Y1×SY2), the integral transform with kernel E is the functor
ΦE:=R(p2)∗(Lp1∗(−)⊗LE):Dqc(Y1)→Dqc(Y2).
We say that E∈Dcohb(Y1×SY2) induces a Fourier–Mukai equivalence (over S) if ΦE restricts to an equivalence Dcohb(Y1)→Dcohb(Y2). In this case, Y1 and Y2 are called Fourier–Mukai S-partners, and we say it is given by E. If clear from context, we omit the ‘S-’ and say ‘Fourier–Mukai partners’. We say that A∈Dqc(Y1×SY2) is relatively perfect over Yi if A⊗LLpi∗Dqc(Yi)⊆Dqc(Y1×SY2). By [4 Corollary 1.4], a pseudocoherent complex E on Y1×SY2 is relatively perfect over Y1 (resp. Y2) if, and only if, ΦE(Dcohb(Y1))⊆Dcohb(Y2) (resp. ΦE(Perf(Y1))⊆Perf(Y2)). In both cases, any such E is bounded.
Auxiliary lemmas
Lemma 3.1. Let X be a quasi-compact quasi-separated algebraic space. Let E∈D(X). Then E is invertible if, and only if, there exists an étale presentation s:U→X such that Ls∗E is invertible.
Proof. If E is invertible, then so is Ls∗E for any étale presentation s. Indeed, the derived pullback is monoidal [15 Tag 07A4]. Conversely, suppose s:U→X is an étale presentation such that Ls∗E is invertible. Fix any étale morphism t:V→X from a scheme. Consider the fibered square
V×XUs′↓⏐Vt′tU↓⏐sX
Suppose h:W→V×XU is an étale surjective morphism from a scheme. Note that s′∘h is an étale covering of V in Xet. As the derived pullback is monoidal, it follows that L(s∘t′∘h)∗E is invertible. Recall that an algebraic space endowed with the structure on its small étale site is a locally ringed site [15 Tag 04KH]. Furthermore, the derived pullback along an étale morphism from a scheme to X can be identified with restriction on the small étale site [15 Tag 04M4]. By [15 Tag 0FPY], there exists an étale covering {gi:Wi→W} by schemes such that each E∣Wi can be represented by a shift of an invertible OWi-module. Consequently, we have a covering {s′∘h∘gi:Wi→V} for which [15 Tag 0FPY] holds. Since V was arbitrary, this completes the proof. □
Lemma 3.2. Let f:Y→X be a separated étale morphism of Noetherian algebraic spaces. Then f! is naturally isomorphic to Lf∗ on Dqc(X).
Proof. As f is flat, [4 Lemma 2.24] gives a natural isomorphism Lf∗E⊗Lf!′OX≅f!′E for all E∈Dqc(X). Hence, it suffices to prove that f!′OX≅OY. Consider the fibered square
Y×XYr1↓⏐Yr2fY↓⏐fX
Denote by Δf:Y→Y×XY the diagonal. Note that r1∘Δf and r2∘Δf are the identity of Y. By [15 Tag 05W1], Δf is an open immersion. It also is a closed immersion because f is separated. The open immersion Δf admits the Nagata compactification given by
YΔfY×XY1Y×XYY×XY,
it follows that LΔf∗≅Δf!. There exists a string of isomorphisms
Lemma 3.3. Let f:Y→X be a separated morphism of finite type between Noetherian algebraic spaces. If K is a dualizing complex on X, then f!K is a dualizing complex on Y.
Proof. Choose étale presentations t:V→Y and s:U→X from affine schemes. There exists a commutative diagram
V×XUs′′↓⏐Vt′tY×XUs′↓⏐Yf′fU↓⏐sX
obtained by base change. By [4 Lemma 2.19], t, s, and f∘t are quasi-affine. Base change implies s′, s′′, and t′ are quasi-affine, finitely presented, and étale. In particular, V×XU is a Noetherian scheme. There is a direct computation,
Note that Ls∗K is a dualizing complex. As f′∘t′ is a separated finite type morphism of Noetherian schemes, [15 Tag 0AA3] says that (f′∘t′)!Ls∗K is a dualizing complex. Since t∘s′′ is an étale presentation of Y, f!K is a dualizing complex [15 Tag 0E4Y]. □
Gorensteinness
Definition 4.1. Let f:Y→X be a proper flat morphism of Noetherian algebraic spaces. An f-Serre functor is an autoequivalence Sf:Perf(Y)→Perf(Y) equipped with a bifunctorial isomorphism
RHom(Rf∗RHom(P,Q),OX)≅Rf∗RHom(Q,Sf(P))
for all P,Q∈Perf(Y).
Remark 4.2. The definition above is an analog of [14 Definition 2.5].
Lemma 4.3. Let f:Y→X be a proper flat morphism of Noetherian algebraic spaces. Then f is Gorenstein if, and only if, f!OX is invertible.
Proof. Choose an étale presentation s:U→X from an affine scheme. Note that U is Noetherian. Consider the fibered square
Y×XUs′↓⏐Yf′fU↓⏐sX
Choose an étale presentation w:V→Y×XU from an affine scheme. Note that V is Noetherian, and that w is separated by [4 Lemma 2.19]. The composition f′∘w:V→U is a flat separated finite type morphism of Noetherian schemes. There is a string of isomorphisms,
Since s′∘w is an étale presentation, Lemma 3.1 implies that f!OX is invertible if, and only if, (f′∘w)!OU is invertible. By [15 Tag 0C08], f′∘w is Gorenstein if, and only if, (f′∘w)!OU is isomorphic Zariski locally on V to an invertible object. By [15 Tag 0FPY], the latter holds if, and only if, (f′∘w)!OU is invertible. Now, s and s′∘w are étale, and s′∘w is surjective. Hence, by [15 Tags 0E14 and 04NC], f is Gorenstein if, and only if, f′∘w is Gorenstein. This completes the proof. □
Lemma 4.4. Let f:Y→X be a proper flat morphism of Noetherian algebraic spaces. Then Perf(Y) admits an f-Serre functor if, and only if, f is Gorenstein.
Proof. There is a string of bifunctorial isomorphisms for all P,Q∈Perf(Y):
If f is Gorenstein, then Lemma 4.3 implies f!OX is invertible. In such a case, (−)⊗Lf!OX:Perf(Y)→Perf(Y) is an autoequivalence, and the bifunctorial isomorphisms above show that this defines an f-Serre functor.
Conversely, if Sf is an f-Serre functor on Perf(Y), then for all P,Q∈Perf(Y) there is a string of natural isomorphisms:
Hom(Q,Sf(P))≅Hom(OY,RHom(Q,Sf(P)))≅Hom(Lf∗OX,RHom(Q,Sf(P)))≅Hom(OX,Rf∗RHom(Q,Sf(P)))≅Hom(OX,RHom(Rf∗RHom(P,Q),OX))≅Hom(Rf∗RHom(P,Q),OX)≅Hom(RHom(P,Q),f×OX)≅Hom(RHom(P,Q),f!OX)≅Hom(Q,P⊗Lf!OX).(citep‘[Sta26,Tag08J7and0D0S]‘)(pull/push)(Serre functor)(right adjoint of push)(citep‘[GL26,Lemma2.22]‘)
Fix P∈Perf(Y). As Sf(P) is perfect, [15 Tag 001P], applied to these isomorphisms as functors of Q∈Perf(Y), yields
ϕP:Sf(P)⟶P⊗Lf!OX
such that Hom(Q,ϕP) is an isomorphism for all Q∈Perf(Y). Pick a compact generator G for Dqc(Y), see [15 Tag 09IY]. Apply Hom(G,−) to the distinguished triangle
Sf(P)ϕPP⊗Lf!OX⟶cone(ϕP)⟶Sf(P)[1].
There exists a long exact sequence (see [15 Tag 0149]),
As Hom(G[n],ϕP) is an isomorphism for all n∈Z, it follows that Hom(G,cone(ϕP)[n])≅0 for all n∈Z. Since G compactly generates Dqc(Y), we obtain that cone(ϕP)≅0. In other words, ϕP is an isomorphism. Taking P:=OY implies f!OX is perfect. Since Sf is essentially surjective, there exists P0∈Perf(Y) such that Sf(P0)≅OY. Then
OY≅Sf(P0)≅P0⊗Lf!OX.
Since P0 and f!OX are perfect, the preceding isomorphism shows that f!OX is invertible, with inverse P0. Therefore, by Lemma 4.3, f is Gorenstein.
Lemma 4.5. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by B∈Dcohb(Y1×SY2). There are natural isomorphisms for all P,Q∈Perf(Y1):
Proof. We prove the first claim. By [4 Remark 2.7 and Lemma 10.9], ΦB induces an equivalence on Dqc. Denote by pi:Y1×SY2→Yi the projections. For P,Q∈Perf(Y1) and E∈Dqc(S), there exist a string of natural isomorphisms:
Hom(E,R(f1)∗RHom(P,Q))≅Hom(Lf1∗E,RHom(P,Q))≅Hom(Lf1∗E⊗LP,Q)≅Hom(ΦB(Lf1∗E⊗LP),ΦB(Q))≅Hom(R(p2)∗(Lp1∗(Lf1∗E⊗LP)⊗LB),ΦB(Q))≅Hom(R(p2)∗(Lp1∗Lf1∗E⊗LLp1∗P⊗LB),ΦB(Q))≅Hom(R(p2)∗(Lp2∗Lf2∗E⊗LLp1∗P⊗LB),ΦB(Q))≅Hom(R(p2)∗(Lp1∗P⊗LB)⊗LLf2∗E,ΦB(Q))≅Hom(ΦB(P)⊗LLf2∗E,ΦB(Q))≅Hom(Lf2∗E,RHom(ΦB(P),ΦB(Q)))≅Hom(E,R(f2)∗RHom(ΦB(P),ΦB(Q)))(pull/push)([Sta26, Tag 08J7 and 0D0S])(ΦB is an equivalence)(def. of ΦB)([Sta26, Tag 07A4])(f1∘p1=f2∘p2)(projection formula)(def. of ΦB)([Sta26, Tag 08J7 and 0D0S])(pull/push).
These isomorphisms are natural in E∈Dqc(S) and in P,Q∈Perf(Y1). Hence, [15 Tag 001P] gives an isomorphism
R(f1)∗RHom(P,Q)⟶R(f2)∗RHom(ΦB(P),ΦB(Q))
which is functorial in P and Q. This proves the first claim. Applying RHom(−,OS) gives the second claim, and completes the proof. □
Lemma 4.6. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by B∈Dcohb(Y1×SY2). Denote by pi:Y1×SY2→Yi the projections. If S1 is an f1-Serre functor, then there are natural isomorphisms for all P,Q∈Perf(Y1):
where S2:=ΦB∘S1∘ΦRHom(B,p2∗OY2). Here, ΦRHom(B,p2∗OY2) denotes the integral transform from Dqc(Y2) to Dqc(Y1), i.e. the roles of p1 and p2 are interchanged.
Proof. Fix P∈Perf(Y1). By [4 Remark 2.7 and Lemma 10.9], ΦB is an equivalence on Dqc which restricts to an equivalence on Perf. By [4 Proposition 9.3], applied with t the identity, ΦRHom(B,p2∗OY2) is right adjoint to ΦB on Dqc. Hence, it is a quasi-inverse of ΦB. We claim that S2(ΦB(P))≅ΦB(S1(P)). Indeed,
S2(ΦB(P))≅(ΦB∘S1∘ΦRHom(B,p2∗OY2)∘ΦB)(P)(def. of S2)
Proof of Theorem 1.3. By symmetry, it suffices to show that if f1 is Gorenstein, then so is f2. By [4 Remark 2.7], E is relatively perfect over each Yi. As f1 is Gorenstein, Lemma 4.4 gives an f1-Serre functor S1. Define S2:=ΦE∘S1∘ΦRHom(E,p2∗OY2) as in Lemma 4.6. By [4 Lemma 10.9], ΦE restricts to an equivalence Perf(Y1)→Perf(Y2), and so S2 is an autoequivalence of Perf(Y2). For all P,Q∈Perf(Y1), there is a string of natural isomorphisms
RHom(R(f2)∗RHom(ΦE(P),ΦE(Q)),OS)≅RHom(R(f1)∗RHom(P,Q),OS)≅R(f1)∗RHom(Q,S1(P))≅R(f2)∗RHom(ΦE(Q),S2(ΦE(P)))(xref‘Lemma 4.5‘)(S1 is an f1-Serre functor)(xref‘Lemma 4.6‘).
Note that every object of Perf(Y2) is isomorphic to one of the form ΦE(P) with P∈Perf(Y1). Hence, from the equivalence ΦE, it follows that S2 is an f2-Serre functor. Consequently, Lemma 4.4 implies f2 is Gorenstein.
Smoothness
Lemma 5.1. Let f:Y→X be a finitely presented flat morphism of Noetherian algebraic spaces. If the base change of f along every Spec(k)→X is smooth where k is a field, then f is smooth.
Proof. Choose an étale presentation s:U→X. Consider the fibered square
Y×XUs′↓⏐Yf′fU↓⏐sX
Let t:V→Y×XU be an étale presentation. Fix u∈U. Denote by iu:Spec(κ(u))→U the canonical morphism. There exists a fibered square
Base change says tu is an étale presentation. Hence, the composition fu′∘tu is a smooth finitely presented morphism of Noetherian schemes. By [15 Tag 01V8], f′∘t is smooth. Appealing to [15 Tag 03ZF], we have that f is smooth. □
Lemma 5.2. Let f:Y→X be a finitely presented flat morphism of Noetherian algebraic spaces. If for every p∈∣X∣ there exists a representative Spec(k)→X of p such that the base change of f along Spec(k)→X is smooth, then the base change of f along every Spec(k)→X is smooth where k is a field.
Proof. Fix p∈∣X∣. Choose a representative t:Spec(k)→X of p such that the base change of f along t is smooth. Suppose t′:Spec(k′)→X is any representative of p. By the definition of the points of an algebraic space [15 Tag 03BT], there exist a field ℓ and a commutative diagram
Base change implies sk, and hence sℓ, are étale presentations. Hence, fℓ∘sℓ is smooth and finitely presented. Denote by T# the set of points of ∣U×XSpec(#)∣ where f#∘s# is smooth where #∈{k′,ℓ}. Applying [15 Tag 02V4], (q1′′)−1(Tk′)=Tℓ. Since q1 is surjective, so is q1′′. Hence, q1′′((q1′′)−1(Tk′))=Tk′. However, Tℓ=∣U×XSpec(ℓ)∣, and so fk′∘sk′ is smooth. As sk′ is an étale presentation from a scheme, [15 Tag 03ZF] says fk′ is smooth, which completes the proof. □
Lemma 5.3. Let f:Y→X be a finitely presented flat morphism of Noetherian algebraic spaces. Then the following are equivalent:
f is smooth
the base change of f along every Spec(k)→X is smooth where k is a field
for every p∈∣X∣ there exists a representative Spec(k)→X of p such that the base change of f along Spec(k)→X is smooth.
Proof. (1) ⟹ (3) follows from [15 Tag 03ZF], (3) ⟹ (2) is Lemma 5.2, and (2) ⟹ (1) is Lemma 5.1. □
Lemma 5.4. Let k be a field. A finitely presented morphism f:X→Spec(k) of algebraic spaces is smooth if, and only if, X×kSpec(k′) is regular for all field extensions k′/k.
Proof. If f is smooth, then the base change of f along all field extensions k′/k is smooth, and hence, X×kSpec(k′) is regular. Conversely, let s:U→X be an étale presentation. Fix a field extension k′/k. If X×kSpec(k′) is regular, then U×kSpec(k′) is regular because the base change of s along k′/k yields an étale presentation. By [15 Tag 038V], U is geometrically regular. Then [15 Tag 038X] implies U is smooth over k, i.e. f∘s is smooth. Thus, [15 Tag 03ZF] says f is smooth. □
Remark 5.5. Let X be a Noetherian algebraic space. Recall that the locus of x∈∣X∣ for which the étale local ring OX,x is regular is stable under generalization. Indeed, choose an étale presentation s:U→X. Let x⇝x′ be a specialization in ∣X∣ such that OX,x′ is regular. By [15 Tags 03JX, 03JV, and 03K2], there exists a specialization u⇝u′ in ∣U∣ such that s(u)=x and s(u′)=x′. By [15 Tag 04KF and 06LN], OU,u′ is regular. Hence, OU,u is regular because it is a localization of OU,u′[15 Tag 0AFS]. Once more, from [15 Tag 04KF and 06LN], OX,x is regular.
Lemma 5.6. A Noetherian algebraic space X is regular if, and only if, Perf(X)=Dcohb(X).
Proof. Fix an étale presentation s:U→X with U affine. If X is regular, then U is regular. Hence, for any E∈Dcohb(X), Ls∗E is perfect [15 Tag 0FDC], and so E is perfect [4 Lemma 3.5]. It follows that Perf(X)=Dcohb(X). We prove the converse. Note that every point of ∣X∣ specializes to a closed point [15 Tags 03K6 and 005E]. Hence, by Remark 5.5, it suffices to prove that every closed point of ∣X∣ belongs to the locus of points whose étale local ring is regular. Let p∈∣X∣ be closed. Denote by tp:Spec(κ(p))→X the associated residual space of p[15 Tag 06R0, 0H1R, and 0318]. By [15 Tag 0H1U], tp is a closed immersion. There exists a fibered square
U×XSpec(κ(p))tp′↓⏐UspsSpec(κ(p))↓⏐tpX
By [4 Lemma 2.19], s is quasi-affine, and hence representable. Base change implies sp is an étale presentation. By [15 Tag 02GL], U×XSpec(κ(p)) is a disjoint union of spectra of finite separable field extensions of κ(p). As sp is finitely presented, this disjoint union is finite, and hence its source is affine and regular. By Example 2.1, OU×XSpec(κ(p)) finitely builds every object of Dcohb(U×XSpec(κ(p))). As tp is a closed immersion, R(tp)∗OSpec(κ(p)) belongs to Dcohb(X), which equals Perf(X) by hypothesis. By flat base change, R(tp′)∗OU×XSpec(κ(p)) is perfect. Hence, for all E∈Dcohb(U×XSpec(κ(p))), R(tp′)∗E is perfect. Choose a closed point in ∣U×XSpec(κ(p))∣. Set u∈∣U∣ to be its image under ∣tp′∣. Denote by hu:Spec(OU,u)→U the canonical morphism. Consider the fibered square
Base change implies tp′′ is a closed immersion. By [8 Lemma 4.8], L(hu′)∗ is essentially dense on Dcohb. Coupled with flat base change, it follows that
Denote by gu:Spec(κ(u))→Spec(OU,u) the canonical morphism. By [15 Tag 0356], gu factors through tp′′. Since gu and tp′′ are closed immersions, the induced morphism from Spec(κ(u)) to the source of tp′′ is a closed immersion. Its structure sheaf pushforward is coherent, and pushforwards along closed immersions are exact [15 Tag 04CJ]. Thus, R(gu)∗OSpec(κ(u)) belongs to the left-hand side of the above inclusion, and hence is perfect. This implies OU,u is a regular local ring [15 Tag 00OC]. Hence, u is a regular point of U. As s(u)=p, it follows that X is regular at p[15 Tag 0AH9]. □
Proposition 5.7. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by a kernel E∈Dcohb(Y1×SY2). Then f1 is smooth if, and only if, f2 is smooth.
Proof. By [4 Remark 2.7], E is relatively perfect over each Yi. By symmetry, it suffices to prove the case where f1 is smooth. Fix a morphism Spec(k)→S from a field. Let L/k be a field extension. The composition Spec(L)→S is affine [15 Tag 09TF]. By [4 Theorem 10.13], Y1×SSpec(L) and Y2×SSpec(L) are Fourier–Mukai Spec(L)-partners. The kernel is the derived pullback of E to Y1×SY2×SSpec(L), which is relatively perfect over each Yi×SSpec(L)[4 Corollary 6.18]. Consequently, [4 Lemma 10.9] implies the associated integral transform restricts to equivalences on both Dcohb and Perf. As Y1×SSpec(L) is smooth over L, it is regular. By Lemma 5.6, it follows that
Perf(Y1×SSpec(L))=Dcohb(Y1×SSpec(L)).
The integral transform takes the left-hand side to Perf(Y2×SSpec(L)) and the right-hand side to Dcohb(Y2×SSpec(L)). Hence, every object of Dcohb(Y2×SSpec(L)) is perfect, and so Lemma 5.6 implies Y2×SSpec(L) is regular. Applying Lemma 5.4, Y2×SSpec(k) is smooth over k because L was arbitrary. Therefore, by Lemma 5.3, f2 is smooth. □
Lemma 5.8. Let f1:Y1→S and f2:Y2→S be proper flat morphisms to a regular Noetherian algebraic space. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by a kernel B∈Dcohb(Y1×SY2). If Y1 is not regular, then B∈/Perf(Y1×SY2).
Proof. By [4 Remark 2.7], B is relatively perfect over each Yi. We prove the claim by contradiction. Assume that B∈Perf(Y1×SY2). By [4 Lemma 10.9], ΦB restricts to triangulated equivalences on Dcohb and Perf. We first show that ΦB(Dcohb(Y1))⊆Perf(Y2). Choose a classical generator Gi for Perf(Yi), see [15 Tags 09IY, 09M8, and 09SR]. By [4 Lemma 5.4] and [15 Tag 09SR], Lp1∗G1⊗LLp2∗G2 is a classical generator for Perf(Y1×SY2). In particular, B is finitely built by Lp1∗G1⊗LLp2∗G2. Fix E∈Dcohb(Y1). As R(p2)∗(Lp1∗E⊗L−) is an exact functor, it follows that
ΦB(E)∈⟨R(p2)∗(Lp1∗E⊗LLp1∗G1⊗LLp2∗G2)⟩.
By the projection formula and flat base change, there are isomorphisms
As f1 is proper, R(f1)∗(E⊗LG1)∈Dcohb(S). Since S is regular, Lemma 5.6 implies Dcohb(S)=Perf(S). Hence, ΦB(E) is perfect, as desired. Now, ΦB takes Dcohb(Y1) to Dcohb(Y2), so Dcohb(Y2)=Perf(Y2). As ΦB restricts to equivalences on Dcohb and Perf, it follows that Dcohb(Y1)=Perf(Y1). Thus, Lemma 5.6 implies Y1 is regular, which is a contradiction. □
Lemma 5.9. Let f:Y→X be a finitely presented morphism of Noetherian algebraic spaces. For each p∈∣X∣, denote by tp:Spec(κ(p))→X the canonical morphism from the residue field of p, and by tp′:Y×XSpec(κ(p))→Y the morphism obtained by base change along tp. If E∈Dcohb(Y) satisfies L(tp′)∗E∈Perf(Y×XSpec(κ(p))) for all p∈∣X∣, then E∈Perf(Y).
Proof. As f is finitely presented, Y×XSpec(κ(p)) is Noetherian for all p∈∣X∣. Choose q∈∣Y∣, and set p:=f(q). By [15 Tag 03H4], there exists q′∈∣Y×XSpec(κ(p))∣ which maps to q. Denote by b:Spec(κ(q′))→Y×XSpec(κ(p)) the canonical morphism from the residue field of q′, see [15 Tag 0EMV]. The image of tp′∘b is q, and so tp′∘b represents q[4 Lemma 2.8]. As L(tp′)∗E is perfect, so is L(tp′∘b)∗E≃Lb∗L(tp′)∗E. In particular, L(tp′∘b)∗E has bounded cohomology. Since q was arbitrary, [4 Proposition 3.12] implies E is perfect. □
Proposition 5.10. Let fi:Yi→S be proper flat morphisms of Noetherian algebraic spaces. Suppose Y1 and Y2 are Fourier–Mukai S-partners given by a kernel E∈Dcohb(Y1×SY2). Then f1 is smooth if, and only if, E∈Perf(Y1×SY2).
Proof. By [4 Remark 2.7], E is relatively perfect over each Yi. Assume that f1 is smooth. Choose p∈∣S∣. Denote by tp:Spec(κ(p))→S the canonical morphism associated to the residue field of p. Denote by tp′:Y1×SY2×SSpec(κ(p))→Y1×SY2 the canonical morphism obtained by base change along tp. As f1 is smooth, Proposition 5.7 implies f2 is smooth. Hence, by base change, Y1×SY2×SSpec(κ(p)) is smooth over κ(p). Then Lemma 5.6 implies
::: By [4 Corollary 6.18], L(tp′∗E) is relatively perfect over each Yi×SSpec(κ(p)). In particular, L(tp′∗E) has bounded cohomology, and so it is perfect. Since this holds for all p∈∣S∣, Lemma 5.9 asserts that E is perfect.
Assume that E is perfect. Choose p∈∣S∣. Denote by tp:Spec(κ(p))→S a morphism from an algebraic closure of the residue field of p. Denote by tp′:Y1×SY2×SSpec(κ(p))→Y1×SY2 the canonical morphism obtained by base change along tp. By [4 Corollary 6.18], L(tp′∗E) is relatively perfect over each Yi×SSpec(κ(p)). Moreover, tp is affine [15 Tag 09TF], so [4 Theorem 10.13] implies L(tp′∗E) induces an equivalence on Dcohb.
Furthermore, L(tp′∗E) is perfect, and so Lemma 5.8 implies Y1×SSpec(κ(p)) is regular. As κ(p) is perfect, it has no nontrivial finite purely inseparable extensions. Applying [15 Tags 038V and 038X] to an étale presentation of Y1×SSpec(κ(p)), it follows from [15 Tag 03ZF] that Y1×SSpec(κ(p)) is smooth over κ(p). Since tp represents p and p∈∣S∣ was arbitrary, Lemma 5.3 asserts that f1 is smooth, which completes the proof. □
References
E. Guisado Villalgordo, BCAM – Basque Center for Applied Mathematics, Mazarredo 14, 48009 Bilbao, Basque Country – Spain
[1][Bal11] Matthew Robert Ballard. Derived categories of sheaves on singular schemes with an application to reconstruction. Adv. Math., 227(2):895–919, 2011.arxiv.org/abs/0801.2599
[4][GL26] Elías Guisado Villalgordo and Pat Lank. Relative Fourier–Mukai theory for algebraic spaces. arXiv:2609.01298, 2026.arxiv.org/abs/2609.01298
[5][GLMP25] Elías Guisado Villalgordo, Pat Lank, Kabeer Manali Rahul, and Nebojsa Pavic. Fiberwise criteria for Fourier–Mukai equivalences. arXiv:2512.16503, 2025.arxiv.org/abs/2512.16503
[6][HLS09] Daniel Hernández Ruipérez, Ana Cristina López Martín, and Fernando Sancho de Salas. Relative integral functors for singular fibrations and singular partners. J. Eur. Math. Soc. (JEMS), 11(3):597–625, 2009.arxiv.org/abs/math/0610319
[7][Kol11] János Kollár. Simultaneous normalization and algebra husks. Asian J. Math., 15(3):437–450, 2011.arxiv.org/abs/0910.1076
[8][Lan26] Pat Lank. Germwise singular Fourier–Mukai partnerships. arXiv:2610.29944, 2026.arxiv.org/abs/2610.29944
[9][LS16] Valery A. Lunts and Olaf M. Schnürer. New enhancements of derived categories of coherent sheaves and applications. J. Algebra, 446:203–274, 2016.arxiv.org/abs/1406.7559
[10][Muk81] Shigeru Mukai. Duality between D(X) and D(X̂) with its application to Picard sheaves. Nagoya Math. J., 81:153–175, 1981.
[11][Nee92] Amnon Neeman. The chromatic tower for D(R). With an appendix by Marcel Böckstedt. Topology, 31(3):519–532, 1992.
[12][Nee23] Amnon Neeman. An improvement on the base-change theorem and the functor f!. Bull. Iran. Math. Soc., 49(3):163, 2023. Id/No 25.DOI
[13][Spe23] Dylan Spence. Reconstruction of projective curves from the derived category. Mich. Math. J., 73(1):171–194, 2023.
[14][SS12] Carlos Sancho de Salas and Fernando Sancho de Salas. Reconstructing schemes from the derived category. Proc. Edinb. Math. Soc., II Ser., 55(3):781–796, 2012.DOI
[15][Sta26] The Stacks Project Authors. Stacks Project. https://stacks.math.columbia.edu, 2026.