We extend the notion of the right relative Du Bois complex to a smooth base of arbitrary dimension. We further show that this complex satisfies base change in the sense of Kovács–Taji, for flat families with strict higher Du Bois fibers, extending the result by Ning over a smooth curve.
Introduction
For a smooth morphism f:X→S of smooth complex varieties, the relative de Rham complexΩX/S∙ is characterized by the exact sequence:
0⟶f∗ΩS1⟶ΩX1⟶ΩX/S1⟶0.
Exterior powers of this sequence give rise to a natural filtration of ΩXp with graded pieces f∗ΩSq⊗ΩX/Sp−q. A key property is that ΩX/Sp satisfies base change, that is, for f smooth,
ΩX/Sp⊗OXLOXs≃qisΩXsp.
For X singular, Du Bois [1] replaced the de Rham complex ΩXq with the filtered complex ΩXq in the derived category of X. It agrees with ΩXq when X is smooth, and, from a cohomological point of view, this complex behaves very much like the usual de Rham complex. It is therefore natural to ask whether there is a way to produce an object analogous to the relative de Rham complex of a smooth family, equipped with a similar base change property.
Kovács constructed a (left) relative Du Bois complex, which we will denote ΩX/Sp,−, over a smooth curve in [6] and over a smooth base of arbitrary dimension in [8, 7]. It shows many applications to families of singular varieties; see [6, 7, 10, 11]. One important question of Kovács–Taji [12] asks whether the relative Du Bois complex satisfies base change, that is, does
ΩX/Sp,−⊗OXLOXs≃qisΩXsp
hold? It turns out that, in general, base change for the left relative Du Bois complex fails in every degree [4, 13]. Ning proposed an alternative construction, dubbed the right relative Du Bois complex ΩX/Cp,+ for f:X→C, where C is a smooth curve. It satisfies base change [13 Theorem 1.1] for flat families with strict higher Du Bois singularities (cf. Definition 2.2). As shown in loc. cit., base change properties have geometric applications such as deformation invariance of higher Du Bois singularities.
The purpose of this paper is to extend the notion of the right relative Du Bois complex ΩX/Sp,+ to a smooth base of arbitrary dimension while preserving base change properties, see Definition 4.1 for the precise definition. We construct the following.
Theorem 1.1. (=Theorem 3.12) Let f:X→S be a morphism of complex algebraic varieties with dimX=n, dimS=k and S smooth. Then there exist objects ΩX/Sp,+,ΩX/Sp,− in Dcohb(X) for all p∈Z with the following properties:
(1) ΩX/Sp,+≃qis0 for p<0, ΩX/Sp,−≃qis0 for p>n−k.
:::`
(2) There exist hyperfiltrations Fq,+ΩXp, Fq,−ΩXp of ΩXp with 0≤q≤k+1, such that F0,+ΩXp≃qisF0,−ΩXp≃qisΩXp, Fk+1,+ΩXp≃qisFk+1,−ΩXp≃qis0, whose associated graded complexes, with 0≤q≤k are, Grq,+ΩXp≃qisΩX/Sp−q,+⊗f∗ΩSq, Grq,−ΩXp≃qisΩX/Sp−q,−⊗f∗ΩSq.
(3) Fq,+ΩXp and Fq,−ΩXp are functorial.
(4) If f is smooth, then ΩX/Sp,+≃qisΩX/Sp,−≃qisΩX/Sp.
(5) Let {νi:Ui↪X} be a finite open cover of X. Then ΩX/Sp,±≃qisR(ν∙)∗ΩU∙/Sp,±.
Note that the statement for the left relative Du Bois complex ΩX/Sp,− follows from [7 Theorem 3.4]. Our construction for the right relative complex agrees with the one constructed in [13] over a smooth curve, see Proposition 3.14.
We also show base change properties for this generalized right relative Du Bois complex for fibers that are strict higher Du Bois (cf. Definition 2.2).
Theorem 1.2. (= Theorem 4.7) Let f:X→S be a flat morphism of algebraic varieties with dimX=n, dimS=k and S smooth. Let s∈S be a closed point and suppose the fiber Xs has strict-m-Du Bois singularities. Then Xs satisfies right-m-base change (cf. Definition 4.1) for the relative Du Bois complex, that is, for all 0≤p≤m,
ΩX/Sp,+⊗LOXOXs≃qisΩXsp.
We remark that a recent work of Dirks in [2] constructs an alternative relative Du Bois complex using the theory of mixed Hodge modules, over smooth bases of arbitrary dimensions. It behaves similarly to the right relative Du Bois complex, and Dirks asks [2 §5, Questions 3, 4] whether the existing constructions agree. This result will be discussed in an upcoming work [14].
Paper Outline. Section 2 reviews preliminary material essential to the construction. In Section 3, we construct the right relative Du Bois complex ΩX/Sp,+ directly for a singular total space, refining the methods of [7]. We verify that both relative Du Bois complexes ΩX/Sp,± agree with the existing definitions in the literature [6, 13] over a smooth curve. Section 4 shows the base change property of ΩX/Sp,+ for flat families with strict higher Du Bois fibers.
Acknowledgments. The authors would like to thank Sándor Kovács and Brad Dirks for helpful discussions.
AI Disclosure. No generative AI is used for the mathematical contents of this paper.
Preliminaries
The Du Bois Complex
The Du Bois complex was originally constructed in [1] using simplicial hyperresolutions. In this paper, by hyperresolution we will mean cubical hyperresolutions in the sense of [3]. We refer the reader there for detail, see also [5 §10.6].
Theorem 2.1 ([1, 3]). Let X be a reduced scheme of finite type over C. Then there exists an object ΩX∙ in Dfilt,cohb(X), with associated graded ΩXp=GrfiltpΩX∙[p] satisfying the following properties.
(1) If π∙:X∙→X is any hyperresolution of X, then
ΩX∙≃qisRπ∙∗ΩX∙∙.
(2) The construction is functorial, i.e. if ϕ:Y→X is a morphism of reduced schemes of finite type over C, then there exists a natural map ϕ∗:ΩX∙→Rϕ∗ΩY∙.
Furthermore, if ϕ is proper, then ϕ∗ is a morphism in Dfilt,cohb(X).
(3) Let ΩX∙ be the usual De Rham complex of Kähler differentials considered with the ‘filtration bête’. Then there exists a natural map of filtered complexes
ΩX∙⟶ΩX∙
and if X is smooth, this map is a quasi-isomorphism.
(4) Let {νi:Ui↪X} be a finite open cover of X, with ν:U∙→X the associated map. Then
ΩX∙≃qisRν∗ΩU∙∙.
The definitions of higher Du Bois singularities have several distinct flavors outside of the local complete intersection setting. As observed in [13], the notion of strict-m-Du Bois singularities appears to be the most suitable for base change type questions, so we only include this variant here. We refer the reader to [SVV23, Kov26, NN25] for detailed discussions on other notions.
Definition 2.2. Let X be a reduced scheme of finite type over C and let m∈N. We say X has strict-m-Du Bois singularities if the natural map
ΩXp≃qisΩXp
is a quasi-isomorphism for each 0≤p≤m.
Hyperfiltrations
Let A be an abelian category and D(A) its derived category. Let Γ:A→Ab be a left exact additive functor from A to the category of abelian groups.
Definition 2.3.[7 1.2.1] Let K∈Ob(Db(A)) be a bounded complex. A hyperfiltration F of K consists of a set of objects FqK∈Ob(Db(A)) for q=l,…,k+1, where l,k∈Z and morphisms
ϕq∈HomDb(A)(Fq+1K,FqK)
for q=l,…,k, with Fl≃qisK and Fk+1≃qis0. The p-th associated graded complex of a hyperfiltration F is
Gp:=GrFpK=Cone(ϕp).
This is the class of the mapping cone of a morphism in C(A) whose class in D(A) is ϕp, so that Fp+1⟶Fp⟶Gp+1 forms a distinguished triangle.
The right relative Du Bois complex
We extend the construction of the right relative Du Bois complex ΩX/Bp,+ from [13], where the base is a smooth curve, to a smooth base S of arbitrary dimension. This is the ascending analogue of the construction of [7], in which the pieces ΩX/Sp are built upward from p=0 rather than downward from p=dimX−dimS.
Our approach further simplifies the constructions in loc. cit. by building ΩX/Sp directly without assuming X is smooth. The central observation is that the only data required from a hyperresolution is a wedge map on the Du Bois complex of X. The following lemma provides the naturality that we need.
Lemma 3.1. Let f:X→S be a morphism of algebraic varieties, with S smooth. Then, for each p,q≥0, there exists a functorial wedge map
Proof. This is a simple extension of the observations in [6 §1], see also [13 Construction 3.1].
Fix a cubical hyperresolution π∙:X∙→X, so that we have a specific representative of ΩXp, namely
R(π∙)∗ΩX∙p≃qisΩXp.
Using the projection formula to identify forms pulled back from S, we may define
∧p,q:R(π∙)∗ΩX∙p⊗f∗ΩSq→R(π∙)∗ΩX∙p+q.
determined by
∧p,q((πα)∗ηα⊗ξ)=(πα)∗(ηα∧(πα)∗ξ).
where πα:Xα→X is a component of the hyperresolution chosen, and ηα,ξ are the appropriate differential forms. It is then clear that the diagram in the statement of the lemma commutes with the chosen representative.
We now only need to show that ∧p,q is independent of the hyperresolution chosen. Suppose π∙:X∙→X is another cubical hyperresolution and α:X∙→X∙ is a morphism between them. Then we have a commutative diagram
where the vertical arrows are quasi-isomorphisms induced by α, see [1 Lemme 2.1.4]. Therefore, the equivalence class of ∧p,q in D(X) is independent of the chosen hyperresolution, as desired. □
The following construction is an analogue to the induction presented in [7], built upward to match the construction presented in [13].
Construction 3.2. Let f:X→S be a morphism of algebraic varieties, with S smooth of dimension k. With induction on i≥0, for each fixed i, our goal is to define auxiliary objects Fip,+ with maps
μip,q:Fip,+⊗f∗ΩSq→Fip+q,+,
for all p,q≥0. We further require at each step that μip,q is compatible with wedge maps from the base, meaning that the diagram
The two columns are distinguished triangles of the form 3.3 above, and the compatibility is given by Lemma 3.1. We see that it induces a map μ1p,q:F1p,+⊗f∗ΩSq→F1p+q,+ on the bottom. This gives the first step of the induction.
Suppose the inductive hypothesis holds for some i≥0. In particular, there are maps
μip,q:Fip,+⊗f∗ΩSq→Fip+q,+
for all p,q≥0 that commutes with the wedge maps on the base. Define Fi+1p,+:=Cone(μii,p−i), so that
Fip,+⊗f∗ΩSp−iμip−iFip,+⟶Fi+1p,++1(2)
is a distinguished triangle. Note that for p<i this gives Fip,+≃qisFi+1p,+ since by convention ΩSp−i=0 for p<i. Similar to before, this gives the diagram with exact columns of the form (2)
The top square commutes by the inductive hypothesis, therefore we obtain the induced bottom map to be μi+1p,q:Fi+1p,+⊗f∗ΩSq→Fi+1p+q,+. To check the required compatibility, consider the following diagram:
The vertical faces commute as they are all diagrams of the form (3). The top square commutes by associativity of the wedge product, and the middle square commutes by the inductive hypothesis. Therefore, the bottom square is commutative on any chosen representative. For clarity, the bottom square reads:
This is precisely the compatibility required for the inductive step. Therefore, we conclude the construction of Ωip,+ and the necessary accompanying maps μip,q for all i,p,q≥0. We summarize this information as:
Definition 3.6. Following Construction 3.2, for p,i≥0, we define
Ωi+1p,+:=Cone(Ωii,+⊗f∗ΩSp−iμii,p−iΩip,+).
By convention, we put Ωip,+=0 if p<0. Note that Ω0p,+=ΩXp. Furthermore,
0≃qisΩp+1p,+≃qisΩip,+
for all i>p.
This allows us to define the right relative Du Bois complex.
Definition 3.7. Let f:X→S be a morphism of varieties with S smooth. We define ΩX/Sp,+∈Ob(D(X)), called the right relative Du Bois complex of f:X→S, as follows. Put
We use the objects defined in Definition 3.6 to construct a hyperfiltration of ΩXp.
Construction 3.8. As in Construction 3.2, let f:X→S be a morphism of varieties with S smooth of dimension k. Note that by the iterative definition of Fip,+, there is a natural map Fip,+→Fi+1p,+ given by the second map of the triangle (2). These yield a chain
The third column is the shifted triangle (2) at i=p−q, and the top two rows are triangles by construction. By commutativity of the top right square, we may extend to obtain Fq+1,+ΩXp→Fq,+ΩXp, and thereafter define the associated graded complex Grq,+ΩXp to be its cone. The derived nine-lemma gives exactness of the third row, which implies that
Grq,+ΩXp≃qisΩX/Sp−q,+⊗f∗ΩSq(4)
We will take a quick detour to establish the functorial properties of Fq,+ΩXp.
Lemma 3.10. Let ϕ:Y→X be an S-morphism, and g=f∘ϕ. Then for each p,q≥0, there are natural maps of hyperfiltrations
FSq,+ΩXp→Rϕ∗FSq,+ΩYp.
That is, the maps are compatible with the hyperfiltration and descend to the associated graded. In particular, there are natural maps
ΩX/Sp,+→Rϕ∗ΩY/Sp,+.
Proof. We will temporarily use the notation Fi,Xp,+ and Fi,Yp,+ for the auxiliary object in Definition 3.6 associated to X and Y, respectively. Using the triangle (2), consider the following diagram
Note the use of the projection formula to identify the bottom left object. For i=0, the vertical arrows are given by the map of Du Bois complexes, as F0,Xp,+≃qisΩXp and similarly for Y, see Theorem 2.1(2). Furthermore, the left square commutes by the functoriality of the wedge map μp,q=∧p,q in Lemma 3.1. Therefore, the same inductive argument on i≥0 as in Construction 3.2 shows that we may extend iteratively to obtain a natural map
Fi+1,Xp,+→Rϕ∗Fi+1,Yp,+.
Moreover, it is compatible with the map on Du Bois complexes ΩXp→Rϕ∗ΩYp. This allows us to extend the map to the hyperfiltrations, as in the following diagram:
This gives the first statement of the lemma. Applying the triangle for the associated graded in Construction 3.8 gives the compatibility. Lastly, the map ΩX/Sp,+→Rϕ∗ΩY/Sp,+ is simply given by Fp,Xp,+→Rϕ∗Fp,Yp,+. This completes the proof. □
Lemma 3.11. Let π∙:X∙→X be a cubical hyperresolution of X. Then for each p≥0,
R(π∙)∗ΩX∙/Sp,+≃qisΩX/Sp,+.
Proof. By (4) and Lemma 3.10 we obtain a morphism of distinguished triangles:
Note the use of the projection formula for the last term. The proof follows from the same inductive argument as in Lemma 3.10 and the five-lemma, reducing down to the statement about the Du Bois complex, ΩXp≃qisR(π∙)∗ΩX∙p (cf. [1 2.3][3 V.3.3]). See also [7 Lemma 3.3] for an analogous statement. □
We collect the results in this section in the following theorem. For easy comparison, we also include the left relative Du Bois complex by [7].
Theorem 3.12. Let f:X→S be a morphism of algebraic varieties with dimX=n, dimS=k and S smooth. Then there exist objects ΩX/Sp,+,ΩX/Sp,− in Dcohb(X) for all p∈Z with the following properties:
(1) ΩX/Sp,+≃qis0 for p<0, ΩX/Sp,−≃qis0 for p>n−k.
(2) There exist hyperfiltrations Fq,+ΩXp,Fq,−ΩXp of ΩXp with 0≤q≤k+1, such that
(3) Fq,+ΩXp and Fq,−ΩXp are functorial, i.e. if ϕ:Y→X is an S-morphism, then there are natural maps Fq,±ΩXp→Rϕ∗Fq,±ΩYp in D(X) forming the commutative diagram of distinguished triangles:
In particular, there are natural maps in D(X) forming the commutative diagram:
ΩXp↓⏐Rϕ∗ΩYpΩX/Sp,±↓⏐Rϕ∗ΩY/Sp,±
(4) If f is smooth, then ΩX/Sp,+≃qisΩX/Sp,−≃qisΩX/Sp.
(5) Let {νi:Ui↪X} be a finite open cover of X. Then
ΩX/Sp,±≃qisR(ν∙)∗ΩU∙/Sp,±.
Proof. The statements for ΩX/Sp,− follow from [7 Theorem 3.4]. For the right complex ΩX/Sp,+, its existence and (1) follow from Construction 3.2 and Definition 3.7. (2) is Construction 3.8 and (3.9). (3) is Lemma 3.10. The rest are standard consequences of the analogous properties of the absolute Du Bois complex and the constructions of the hyperfiltrations, see [6 Theorem 1.3], [7 Theorem 3.4], [13 Theorem 3.2]. □
Corollary 3.13. Let f:X→S be a morphism of algebraic varieties such that S is smooth. Then for any p≥0 and any locally free sheaf E on X there exists a spectral sequence
Proof. The statement for ΩX/Sp,− is [7 Corollary 3.5]. The analogous statement for ΩX/Sp,+ follows immediately from Theorem 3.12(2) and [7 Theorem 1.2.2]. □
The next proposition shows that the relative Du Bois complexes ΩX/Sp,± agree with the existing definition in the literature when the base is a smooth curve (cf. [6, 13]), justifying the notation. We include the comparison for the left relative Du Bois complex of Kovács here for completeness, as it is not included in [7].
Proposition 3.14. Let f:X→S be a morphism of algebraic varieties, where S is a smooth curve. Then ΩX/Sp,+,ΩX/Sp,− agree with the objects defined in [13 Theorem 3.2] for all p∈N.
Proof. To avoid confusion, we will always use the notations ΩX/Sp,+,ΩX/Sp,− to mean the objects constructed in Theorem 3.12, and use ΩX/Sp,+,ΩX/Sp,− to temporarily denote those constructed in [13 Theorem 3.2]. The statement trivially holds if p<0 for ΩX/Sp,+ and if p>dimX−dimf(X) for ΩX/Sp,−. So let p be in the appropriate range where Fip,+,Fip,− are iteratively defined as cones.
Since dimS=1, ΩSq=0 for q>1. We see that for all 0≤i≤p−2,
Fi+1p,+=Cone(Fip,+⊗f∗ΩS1→Fip,+)≃qisFip,+.
In other words, for all 0≤i≤p−1
Fip,+≃qisF0p,+≃qisΩXp.
Therefore the triangle for ΩX/Sp,+=Fpp,+ reads
ΩX/Sp−1,+⊗f∗ΩS1→ΩXp→ΩX/Sp,++1.
Furthermore, for a fixed representative Kp of ΩXp, the first map is given by the natural factorization of the wedge map Kp−1⊗f∗ΩS1→Kp. Therefore starting with the base case ΩX/S−1,+≃qisΩX/S−1,+=0, we see that this agrees precisely with the defining triangle of ΩX/Sp,+ in [13 Theorem 3.2(2)], and consequently
ΩX/Sp,+≃qisΩX/Sp,+
for all p≥0. A similar argument works for ΩX/Sp,−, with the difference being the inductive cone construction. The distinguished triangle in [8 2.4] (see also [7 §2]) reads, after index shift
which gives the defining triangle for ΩX/Sp,−, cf. [6 Theorem 1.3.2]. Note that after twisting with f∗ωSp, the second map in the triangle factorizes the wedge map KXn−p⊗f∗ωS⟶KXn−p+1. Therefore, the same argument as the right complex concludes that
ΩX/Sp,+≃qisΩX/Sp,−,
once we identify the base case ΩX/Sk,−≃qisΩX/Sk,−=0 for k>dimX−dimf(X), as desired. □
Higher dimensional base change
The notions in [13 Definition 4.1] extend naturally to our setting over a higher dimensional base.
Definition 4.1. Let f:X→S be a morphism of algebraic varieties with dimX=n, dimS=k and S smooth. Let s∈S, m∈Z. We say the fiber Xs satisfies
left-m-base change for the relative Du Bois complex if there is a natural quasi-isomorphism
ΩX/Sp,+⊗LOXOXs≃qisΩXsp,
for all p≤m.
right-m-base change for the relative Du Bois complex if there is a natural quasi-isomorphism
ΩX/Sp,+⊗LOXOXs≃qisΩXsp,
for all p≤m.
m-base change for the relative Du Bois complex if it satisfies both left- and right-m-base change.
base change (resp. left-base change, right-base change) for the relative Du Bois complex if it satisfies m-base change (resp. left-m-base change, right-m-base change) for all m∈Z.
We will frequently omit “for the relative Du Bois complex” if it is clear from the context.
The following lemma identifies the relative complex over a point.
Lemma 4.2. Let f:X→S be a morphism of algebraic varieties, with S smooth of dimension 0. Then for each p∈N,
ΩX/Sp,+≃qisΩX/Sp,−≃qisΩXp.
Proof. Recall as in Construction 3.2 that we iteratively defined
Fi+1p,+=Cone(Fip,+⊗f∗ΩSp−i→Fip,+).
Since ΩSp−i=0 for all i=p, we have
ΩXp≃qisF0p,+≃qis⋯≃qisFpp,+≃qisΩX/Sp,+.
For the left complex ΩX/Sp,−, it can be defined similarly with a representative of ΩXp as in Theorem 3.12 and then the proof would follow. However, since [7] defined it first assuming X is smooth, we will check this case instead and use [7 Lemma 3.3] to conclude that it holds for singular X as well. When X is smooth with dimX=n, we have
commutes. In particular, they are also compatible with the maps from ΩXp.
Proof. Recall in Construction 3.2, we iteratively defined intermediate objects Fi,Sp,+,Fi,Tp,+ for p,i≥0 associated to the morphisms f:X→S, g∘f:X→T, with base case
F0,Sp,+≃qisF0,Tp,+≃qisΩXp.
Here the additional subscripts distinguish the bases S and T. The defining triangle for i=0 reads
where the first map is induced by the natural inclusion of base forms g∗ΩTp→ΩSp. At each step i>0, we may obtain comparison maps from the diagram of (3.5),
using instead g∗ΩTp−i→ΩSp−i. For i+1=p, this gives the desired ΩX/Tp,+→ΩX/Sp,+. Now, we populate the maps for the hyperfiltrations analogously using Construction 3.8. We have a diagram
identifying the associated graded as in Theorem 3.12(2). When q=0, the two hyperfiltration objects are naturally quasi-isomorphic
FT0,+⊗ΩXp≃qisFS0,+⊗ΩXp≃qisΩXp.
Hence we may iteratively extend to FTq+1,+→FSq+1,+ using the functorial map ΩX/Tp−q,+→ΩX/Sp−q,+ defined previously, combined with the inclusion of base forms. This shows all the claimed properties. □
The following is a relative version of the triangle for the relative Du Bois complex over a curve.
Proposition 4.4. Let f:X→S, g:S→T be morphisms of algebraic varieties, with S,T smooth and g smooth of relative dimension 1. Then for each p∈N, there is a triangle
ΩX/Sp−1,+⊗f∗ΩS/T1⟶ΩX/Tp,+⟶ΩX/Sp,++1.
Proof. This is an exercise of patience working with the hyperfiltrations associated with ΩX/Sp,+ and ΩX/Tp,+. We proceed first by induction on p, noting
We can simply read off the shifted third column to obtain the desired triangle
Cp,1≃qisΩX/Sp−1,+⊗f∗ΩS/T1→ΩX/Tp,+→ΩX/Sp,++1
This completes the proof of Proposition 4.4. □
Proof of Claim 4.6. We work with the same inductive assumptions as in the proof of Proposition 4.4, i.e. assuming the proposition holds up to some p≥0. Now, the claim is local on X by Theorem 3.12(5). Note that g is smooth of relative dimension 1, so after shrinking the base if necessary, we make use of the local splitting
ΩSq≅g∗ΩTq⊕(ΩS/T1⊗g∗ΩTq−1).
Thus for q≥1, we may apply the inductive hypothesis to the third column of (5), and obtain
Note that as in Theorem 3.12, FSq,+⊗ΩXp=0 for q>dimS and FTq,+⊗ΩXp=0 for q>dimT=dimS−1. Therefore when q=dimS, (5) consists of all zeroes except in the lower right square, and yields
with the last map induced by a split injection (gf)∗ΩTq↪f∗ΩSq. Using both inductive hypotheses on p and q for the source and the target, the composition realizes a quasi-isomorphism
Furthermore, the first map is split injective induced by the splitting (gf)∗ΩTq↪f∗ΩSq. More precisely, it comes from this diagram extending left inverses:
Combining the above, we see that the triangle coming from (5)
Cp,q⟶Dp,q⟶Cp,q+1[1]+1
is split. In particular, again applying the inductive hypothesis on q identifying Cp,q+1 shows that
Cp,q≃qisΩX/Sp−q,+⊗f∗ΩS/Tq⊗(gf)∗ΩTq−1,
thus proving the claim. □
Theorem 4.7. Let f:X→S be a flat morphism of algebraic varieties with dimX=n, dimS=k and S smooth. Let s∈S be a closed point and suppose the fiber Xs has strict-m-Du Bois singularities. Then Xs satisfies right-m-base change for the relative Du Bois complex, that is, for all 0≤p≤m,
ΩX/Sp,+⊗OXLOXs≃qisΩXsp.
Proof. The case dimS=0 is Lemma 4.2 and the case dimS=1 follows from Proposition 3.14 and [13 Theorem 1.1(2)]. We will prove this by induction on the dimension of the base, so suppose dimS=k≥2 and the theorem has been proved for all bases of dim≤k−1.
The statement is local on the base, so after possibly shrinking S, choose étale coordinates t1,…,tk and project from tk to obtain g:S→T=At1,…,tk−1k−1. We see that g is smooth of relative dimension 1 and g∘f is flat.
Fix s∈S in the theorem and let t=g(s), so that we have a family Xt→St over a smooth curve St. Over s∈St, we have an inclusion of an effective Cartier divisor Xs→Xt by flatness of f. By assumption, Xs is strict-m-Du Bois. In particular note that Xt is reduced in a neighborhood of Xs. Therefore, after further shrinking of X to a neighborhood containing Xs, [13 Theorem 1.3] implies that Xt is strict-m-Du Bois. We have thus arrived at the scenario of the diagram
[1]Philippe Du Bois. Complexe de de Rham filtré d’une variété singulière. Bull. Soc. Math. France, 109(1):41–81, 1981.DOI
[2]Bradley Dirks. A hodge module construction of relative du bois complexes, 2026. arXiv:2609.17148.arxiv.org/abs/2609.17148
[3]F. Guillén, V. Navarro Aznar, P. Pascual Gainza, and F. Puerta. Hyperrésolutions cubiques et descente co-homologique, volume 1335 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1988. Papers from the Seminar on Hodge-Deligne Theory held in Barcelona, 1982.
[4]Caleb Ji and Sándor Kovács. General base change for relative Du Bois complexes, 2025. arXiv:2508.02848.arxiv.org/abs/2508.02848
[5]János Kollár. Singularities of the minimal model program, volume 200 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2013. With a collaboration of Sándor Kovács.DOI
[6]Sándor J. Kovács. Smooth families over rational and elliptic curves. J. Algebraic Geom., 5(2):369–385, 1996.
[7]Sándor J. Kovács. Relative de Rham complex for non-smooth morphisms. In Birational algebraic geometry (Baltimore, MD, 1996), volume 207 of Contemp. Math., pages 89–100. Amer. Math. Soc., Providence, RI, 1997.DOI
[8]Sándor J. Kovács. Spectral sequences associated to morphisms of locally free sheaves. In Recent progress in arithmetic and algebraic geometry, volume 386 of Contemp. Math., pages 57–85. Amer. Math. Soc., Providence, RI, 2005.DOI
[10]Sándor J. Kovács and Behrouz Taji. Hodge sheaves underlying flat projective families. Math. Z., 303(3):Paper No. 75, 34, 2023.arxiv.org/abs/2103.03515
[11]Sándor J. Kovács and Behrouz Taji. Arakelov inequalities in higher dimensions. J. Reine Angew. Math., 806:115–145, 2024.arxiv.org/abs/2205.00761
[12]Sándor J. Kovács and Behrouz Taji. The relative Du Bois complex—on a question of S. Zucker. In Higher dimensional algebraic geometry—a volume in honor of V. V. Shokurov, volume 489 of London Math. Soc. Lecture Note Ser., pages 151–162. Cambridge Univ. Press, Cambridge, 2025.
[13]Haoming Ning. Deformations, local freeness, and base change for higher Du Bois singularities, 2026. arXiv:2608.07849.
[14]Haoming Ning. Duality and comparisons for relative Du Bois complexes, 2026. Preprint.