Du Bois Complexes in Algebraic Geometry

Updated 6 July 2026

Du Bois Complexes are filtered objects that generalize the de Rham complex to singular varieties by encoding Hodge-theoretic data via hyperresolutions.
They define Du Bois singularities through the key morphism from the structure sheaf to the zeroth graded piece and support descent and base change phenomena.
Applications include computing relative complexes, extending forms, and proving generic vanishing results, which bridge singular Hodge theory and birational geometry.

The Deligne–Du Bois complex of a complex algebraic variety (X) is a filtered object ((\underline{\Omega}_X^\bullet,F)) in the derived category that replaces the usual de Rham complex when (X) is singular. Its graded pieces are written
[
\underline{\Omega}_X^{p:=\operatorname{Gr}_F^{p\underline{\Omega}_X^\bullet[p],}}
]
and they recover (\Omega_X^p) on smooth varieties, while for singular spaces they encode the Hodge-theoretic graded pieces of the constant sheaf, define Du Bois singularities through the morphism (\mathcal O_X\to \underline{\Omega}_X^0), and now appear in relative, higher, and functorial forms that connect singular Hodge theory, birational geometry, local cohomology, and generic vanishing [1109.5569, 2410.11109, 2508.02848].

1. Definition and basic structure

A standard construction starts from a hyperresolution (\varepsilon_\bullet:X_\bullet\to X). In that form one has
[
\underline{\Omega}X^p \simeq R{\varepsilon\bullet}*\Omega^p{X_\bullet},
]
and the full filtered complex (\underline{\Omega}_X^\bullet) is quasi-isomorphic to (\mathbb C_X). This is the sense in which the Du Bois complex extends the de Rham complex from smooth to singular spaces [1109.5569, 2406.03593].

The graded pieces behave as singular analogues of holomorphic forms. There is always a canonical morphism
[
\Omega_X^p \longrightarrow \underline{\Omega}X^p,
]
which is an isomorphism when (X) is smooth. Restriction to opens is compatible with the construction, and any morphism (f:X\to Y) induces a canonical morphism
[
\underline{\Omega}_Y^p \to \mathbf{R}f*\underline{\Omega}_X^p.
]
This makes the collection ({\underline{\Omega}_X^p}) functorial enough to support descent, base change, and comparison arguments [2406.03593, 2507.07350].

One particularly concrete piece is the top graded part. If (f:\widetilde X\to X) is a log resolution of a (d_X)-dimensional variety, then
[
\underline{\Omega}X^{d_X}\cong Rf\omega_{\widetilde X}\cong f_\omega_{\widetilde X},
]
the second isomorphism coming from Grauert–Riemenschneider vanishing. This identifies the top Du Bois complex with a canonical birational invariant, and explains why (\underline{\Omega}_X^{d_X}) often behaves more rigidly than the lower pieces [2410.11109].

2. Hodge-theoretic role and the Du Bois condition

For proper (X), the central Hodge-theoretic structure is the spectral sequence
[
E_1^{pq}=\mathbb H^{q(X,\underline{\Omega}_X^{p)\Rightarrow}} H^{{p+q}(X,\mathbb} C),
]
which degenerates at (E_1) and induces Deligne’s Hodge filtration. In particular,
[
\operatorname{Gr}_F^p H^{{p+q}(X,\mathbb} C)\simeq \mathbb H^{q(X,\underline{\Omega}_X^p),}
\qquad
\operatorname{Gr}_F⁰ H^i(X,\mathbb C)\simeq \mathbb H^{i(X,\underline{\Omega}_X^0).}
]
Thus the zeroth graded piece (\underline{\Omega}_X⁰⁾ is the object that replaces (\mathcal O_X) in singular Hodge theory [1109.5569].

This leads to the standard definition: (X) has Du Bois singularities if the natural morphism
[
\mathcal O_X \longrightarrow \underline{\Omega}_X⁰
]
is a quasi-isomorphism. For proper varieties, Kovács gave an “intuitive” reformulation: if (X) is proper with a fixed basepoint-free linear system (\mathfrak d), then (X) is Du Bois if and only if for every (i>0) and every (L\subseteq X) obtained as an intersection of general members of (\mathfrak d), the natural map
[
\nu_i(L):H^i(L,\mathcal O_L)\longrightarrow \operatorname{Gr}_F^{0H^i(L,\mathbb} C)
]
is an isomorphism [1109.5569].

The theory extends to pairs. For a reduced pair ((X,\Sigma)), one has a pair complex (\underline{\Omega}{X,\Sigma}⁰⁾ and a natural morphism
[
\mathscr I{\Sigma\subset X}\longrightarrow \underline{\Omega}_{X,\Sigma}^0.
]
The pair is Du Bois if this morphism is a quasi-isomorphism. This formulation makes it possible to compare the absolute and relative behavior of singularities through distinguished triangles relating (X), (\Sigma), and ((X,\Sigma)) [1401.4976].

A further enlargement is the notion of potentially Du Bois singularities: a variety (X) is potentially Du Bois at a closed point (x) if there exists a Zariski open neighborhood (U\subseteq X) of (x) and a subvariety (\Sigma_U\subseteq U), containing no irreducible component of (U), such that ((U,\Sigma_U)) is a Du Bois pair. Graf–Kovács showed that for normal surfaces the Du Bois and potentially Du Bois notions coincide, while in dimension at least three a normal potentially Du Bois singularity with (K_X) (\mathbb Q)-Cartier need not be Du Bois; they also proved that a normal potentially Du Bois variety with Cartier (K_X) is log canonical and hence Du Bois [1401.4976].

3. Constructions through resolutions, pairs, and Hodge modules

The pair formalism is organized by distinguished triangles. For a closed subscheme (Z\subseteq X),
[
\underline{\Omega}{X,Z}⁰ \to \underline{\Omega}_X⁰ \to \underline{\Omega}_Z⁰ \xrightarrow{+1},
]
and ((X,Z)) is a Du Bois pair precisely when (\mathcal I_Z\to \underline{\Omega}{X,Z}⁰⁾ is a quasi-isomorphism. This viewpoint is especially effective when nilpotents are present, because (\underline{\Omega}X^{0=\underline{\Omega}}{X_{\mathrm{red}}}⁰⁾ and similarly for pairs [1605.02755].

Saito’s theory of mixed Hodge modules gives a more intrinsic realization. If (\mathbf Q_X^H) is the trivial mixed Hodge module on (X), then
[
\underline{\Omega}X^p \cong \operatorname{Gr}^F{-p}\DR(\mathbf Q_X^{H[d_X])[p-d_X].}
]
This formula identifies the Du Bois complexes as graded de Rham pieces of the trivial Hodge module, and it is the starting point for modern applications such as generic vanishing and intersection-complex refinements [2410.11109].

For reduced pairs ((X,Z)), Park proved a precise characterization when (X\setminus Z) has rational singularities. If (j:X\setminus Z\hookrightarrow X) is the open immersion, then
[
\operatorname{gr}^{F_0\mathrm{DR}(j_!\mathbb{Q}^{H_{X\setminus}} Z}) \simeq \underline{\Omega}^{0}_{X,Z},
]
and under the same rationality assumption the pair complex can also be computed birationally as
[
\underline{\Omega}^{0}_{X,Z} \simeq \mathbf{R}p_*\mathcal{O}_{\widetilde X}(-E),
]
for a log resolution (p:\widetilde X\to X) with reduced exceptional divisor (E). This gives a direct bridge between mixed Hodge modules, Du Bois pairs, and extension of forms [2311.15159].

A different alternative to hyperresolutions was developed through smooth poset schemes. If (X) is a reduced complex projective scheme and (\sigma:\mathcal X\to X) is a smooth complex projective poset scheme satisfying
[
\mathbb C_{X^{an}} \xrightarrow{\sim} \mathbf R\sigma^{{an}_*\mathbb} C_{\mathcal X^{an}},
]
then
[
\underline{\Omega}X^\bullet \xrightarrow{\sim} \mathbf R\sigma\Omega_{\mathcal X}^\bullet
]
as filtered complexes, hence
[
\underline{\Omega}Xⁱ \xrightarrow{\sim} \mathbf R\sigma\Omega_{\mathcal X}^i.
]
In the same framework, a reduced finite-type scheme over characteristic (0) admits a categorical resolution by a smooth poset scheme if and only if it has Du Bois singularities [1011.6089].

4. Higher graded pieces, higher Du Bois singularities, and extension of forms

Beyond (p=0), one can require the comparison maps
[
\Omega_X^q \xrightarrow{\sim} \underline{\Omega}_X^q
\qquad (0\le q\le p).
]
For reduced hypersurfaces, Jung, Kim, Saito, and Yoon formulated this as the notion of higher (p)-Du Bois singularities and proved that for a reduced hypersurface (X\subset Y), the following are equivalent:
[
\widetilde\alpha_X\ge p+1,
\qquad
X \text{ is higher }p\text{-Du Bois},
\qquad
X \text{ is higher }p\text{-log canonical}.
]
Here (\widetilde\alpha_X) is the minimal exponent, i.e. the maximal root of the reduced Bernstein–Sato polynomial with sign changed [2107.06619].

For a reduced hypersurface (Z\subset X) in a smooth ambient variety, Mustață, Olano, Popa, and Witaszek established the comparison theorem
[
\widetilde{\alpha}(Z)\ge p+1 \implies \Omega_Z^{p\longrightarrow} \underline{\Omega}Z^p \text{ is an isomorphism},
]
and complemented it with higher-cohomology nonvanishing. If (Z) is singular and (\widetilde{\alpha}(Z)>p\ge 2), then for every singular point (x\in Z),
[
\mathcal H^{{p-1}(\underline{\Omega}_Z^{\,n-p})_x} \simeq \mathcal O{X,x}/(J_f+(f))\neq 0,
]
and, for isolated singularities with (p\ge 3),
[
\mathcal H^{{p-2}(\underline{\Omega}_Z^{\,n-p})_x} \simeq (J_f:f)/J_f \neq 0.
]
These results show that low-degree classical behavior and higher-degree singular corrections coexist in a controlled way [2105.01245].

Extension of forms provides another interpretation of higher Du Bois behavior. For a normal complex algebraic variety (X) with singular locus (\Sigma) and a resolution (\pi:\widetilde X\to X), Kebekus–Schnell-type methods were strengthened as follows: if (X) has at worst Du Bois singularities, then logarithmic extension holds in all degrees,
[
\pi_\Omega_{\widetilde X}^p(\log E)\xrightarrow{\sim}\Omega_X^{[p]},
]
and holomorphic extension holds in the optimal range
[
\pi_\Omega_{\widetilde X}^p \xrightarrow{\sim} \Omega_X^{[p]}
\qquad \text{for } 0\le p<\operatorname{codim}_X(\Sigma).
]
This improves Flenner’s classical criterion by one degree under the Du Bois assumption [2312.01245].

A recent isolated-singularity refinement concerns the internal cohomology sheaves of higher Du Bois complexes. If (X) has isolated pre-((k-1))-Du Bois singularities, then
[
R\mathcal Hom_X(\underline{\Omega}_X^{k,\omega_X^\bullet)}
\longrightarrow
R\mathcal Hom_X(\mathcal H^{0(\underline{\Omega}_X^{k),\omega_X^\bullet)}}
]
is injective on cohomology sheaves, and consequently
[
\mathcal H^{i(\underline{\Omega}_X^k)=0}
\qquad
\text{for }0<i<\operatorname{depth}\mathcal H^{0(\underline{\Omega}_X^k)-1.}
]
This identifies the depth of (\mathcal H^{0(\underline{\Omega}_X^k))} as the key invariant governing higher vanishing in the isolated case [2409.18019].

5. Relative complexes, base change, and descent under morphisms

The relative theory asks for singular analogues of relative Kähler differentials. For a morphism
[
f:X\to C
]
to a smooth complex curve (C), with fiber (X_c=f^{-1}(c)) and inclusion (\jmath_c:X_c\hookrightarrow X), the relative Du Bois complex (\underline{\Omega}{X/C}^\bullet) is defined in earlier work and the basic base-change question is whether
[
L\jmath_c^*\, \underline{\Omega}{X/C}^p \simeq \underline{\Omega}{X_c}^p.
]
A general positive answer is now known only generically: there exists a non-empty open subset (U\subseteq C) such that for every (c\in U),
[
L\jmath_c^*\, \underline{\Omega}{X/C}^p \simeq \underline{\Omega}_{X_c}^p,
]
and similarly for the full filtered complex. If (f) admits a simultaneous relative hyperresolution, then the same statement holds for every closed point (c\in C) [2508.02848].

The generic theorem is sharp. Even when (X) is smooth, (f) is smooth over (C\setminus{0}), and the special fiber (Y=X_0) is a divisor with simple normal crossings, one typically has
[
L\imath^*\, \underline{\Omega}{X/C}^{n} \not\simeq \underline{\Omega}{Y}^{n}.
]
The obstruction is already visible in the top piece: (\underline{\Omega}Y^n\simeq \nu\omega_{\widetilde Y}) reflects the non-normality of the SNC fiber, whereas (\imath^\Omega_X^{n+1}) is a line bundle [2508.02848].

Finite morphisms provide another form of descent. For a finite group quotient (\pi:X\to X/G), the quotient theorem states that
[
\underline{\Omega}{X/G}^\bullet \xrightarrow{\sim} R^G R\pi\underline{\Omega}X^\bullet
]
in the filtered derived category, and therefore
[
\underline{\Omega}{X/G}^p \xrightarrow{\sim} R^G R\pi_\underline{\Omega}X^p
]
for every (p). More generally, if (f:Y\to X) is either a finite group quotient or a finite surjective morphism between normal varieties, there exists
[
t:Rf\underline{\Omega}Y^p \to \underline{\Omega}_X^p
]
such that the composition
[
\underline{\Omega}_X^p \to Rf\underline{\Omega}Y^p \xrightarrow{t} \underline{\Omega}_X^p
]
is an isomorphism. Thus each (\underline{\Omega}_X^p) is a derived direct summand of (Rf*\underline{\Omega}_Y^p) [2507.07350].

At degree (0), descent can be formulated in local-cohomological terms. For a pair ((X,E)), ((X,E)) is Du Bois if and only if the natural map
[
\mathscr I_{E\subseteq X}\to \underline{\Omega}^0_{X,E}
]
is a quasi-isomorphism, and this is equivalent to injectivity of the induced maps on local cohomology at every point. Using this criterion, cyclically pure maps (R\to S) of rings essentially of finite type over (\mathbb C) preserve Du Bois singularities: if (S) is Du Bois, then (R) is Du Bois [2208.14429].

6. Applications, variants, and current directions

One major recent development is generic vanishing on singular varieties. In this setting the correct analogues of holomorphic forms are the Du Bois complexes
[
\underline{\Omega}X^{p=\operatorname{Gr}^F}{-p}\DR(\mathbf Q_X^{H[d_X])[p-d_X],}
]
and the duality-friendly variants are the intersection Du Bois complexes
[
I_X^{p=\operatorname{Gr}^{F_{-p}\DR(\mathrm{IC}_X^H)[p-d_X].}}
]
For a morphism (a:X\to A) to an abelian variety, one has
[
\underline{\Omega}X^p \text{ is } GV{\,p-d_X-\delta(a)},
\qquad
I_X^p \text{ is } GV_{\,p-d_X-\delta_s(a)},
]
while the top pieces satisfy
[
\underline{\Omega}X^{d_X}\cong I_X^{d_X}\cong f*\omega_{\widetilde X}.
]
This recovers the classical Green–Lazarsfeld and Popa–Schnell theorems in the smooth case and explains singular corrections such as the Hacon–Kovács counterexample [2410.11109].

Cones provide a setting where the complexes can be computed explicitly. For the abstract affine cone
[
Z=C(X,L)=\operatorname{Spec}\Big(\bigoplus_{m\ge 0} H^{0(X,L^m)\Big)}
]
over a projective variety (X) with ample line bundle (L), Popa and Shen expressed the cohomology sheaves of (\underline{\Omega}Z^k) in terms of the Du Bois complexes of (X). For example,
[
\Gamma(Z,\mathcal H^{i\underline{\Omega}_Z^0)\simeq} \bigoplus{m\ge 1} H^i(X,\mathcal H^{0\underline{\Omega}_X^0\otimes} L^m)
\quad (i\ge 1),
]
and for (k\ge 1),
[
\Gamma(Z,\mathcal H^{i\underline{\Omega}_Z^k)}
]
is built from the groups (H^{i(X,\underline{\Omega}_X^k\otimes} L^m)) and (H^{i(X,\underline{\Omega}_X^{k-1}\otimes} L^m)). These computations yield criteria for seminormality, formulas for the local cohomological defect of the cone, and descriptions of non-positive (K)-groups [2406.03593].

The degree-zero complex also has strong commutative-algebraic consequences. If ((R,\mathfrak m)) is local and (R_{\mathrm{red}}) is Du Bois, then
[
H^i_{\mathfrak m}(R)\to H^i_{\mathfrak m}(R_{\mathrm{red}})
]
is surjective for every (i). This surjectivity has applications to Cohen–Macaulayness in families, Ext-injectivity, set-theoretic Cohen–Macaulayness, and the relation between Koszul cohomology and local cohomology [1605.02755]. On the dual side, if (R) is a reduced local ring essentially of finite type over a field of characteristic (0), (X=\operatorname{Spec}R) is Du Bois, and (H^i_{\mathfrak m}(R)) has finite length for all (i \[
\tau_{>-j}\omega_R^\bullet
]
is quasi-isomorphic to a complex of (k)-vector spaces. In particular, Du Bois singularities with isolated non-Cohen–Macaulay locus are Buchsbaum [1512.05374].

These developments suggest a stable picture. The Du Bois complex is no longer used only as the formal definition of Du Bois singularities. It now functions as a computable Hodge-theoretic replacement for differential forms on singular spaces, a relative object in families, a descent-theoretic invariant under finite morphisms and pure maps, and a source of vanishing, extension, and local-cohomological statements. A plausible implication is that future work will continue to sharpen the distinction between generic and special fibers, between absolute and pair-theoretic Du Bois phenomena, and between ordinary and intersection-complex versions of singular Hodge theory.