---
title: Higher Rational Singularities
url: https://www.emergentmind.com/topics/higher-rational-singularities
type: topic
---

# Higher Rational Singularities

Higher rational singularities are refinements of classical rational singularities in which the vanishing and comparison properties ordinarily imposed on functions are extended to differential forms, Hodge-theoretic graded pieces, or their positive-characteristic analogues up to a fixed level \(k\) or \(m\). In current literature, the expression appears in several related frameworks rather than as a single universal definition. The central characteristic-zero theory defines \(k\)-rational singularities using the Deligne–Du Bois complex and Grothendieck duality, recovering ordinary rational singularities at \(k=0\); related approaches recast the same phenomenon through intersection Du Bois complexes, irrationality complexes, derived splittings, mixed Hodge modules, Witt vector cohomology, and Frobenius-theoretic conditions after reduction modulo \(p\) [2205.04729].

## 1. Classical origin and the characteristic-zero \(k\)-rational definition

In the classical setting over the complex numbers, a normal variety \(X\) has rational singularities if for a resolution of singularities \(\pi:\widehat{X}\to X\) one has
\[
\mathcal{O}_X \simeq R\pi_*\mathcal{O}_{\widehat{X}},
\]
equivalently \(R^i\pi_*\mathcal{O}_{\widehat{X}}=0\) for \(i>0\). The higher theory begins from the filtered Deligne–Du Bois complex \(\underline{\Omega}_X^\bullet\), whose graded pieces \(\underline{\Omega}_X^p[p]\) play the role of \(\Omega_X^p\), and from the functorial comparison maps
\[
\phi^p:\Omega_X^p\longrightarrow \underline{\Omega}_X^p.
\]
A variety is \(k\)-Du Bois if \(\phi^p\) is a quasi-isomorphism for all \(0\le p\le k\). The higher rational condition is then formulated by combining \(\phi^p\) with Grothendieck duality: if \(n=\dim X\) and
\[
\mathbb{D}_X(-):=\operatorname{RHom}(-,\omega_X^\bullet)[-n],
\]
there is a canonical map
\[
\underline{\Omega}_X^p \xrightarrow{\ \psi^p\ } \mathbb{D}_X(\underline{\Omega}_X^{n-p}),
\]
and \(X\) is \(k\)-rational if for all \(0\le p\le k\) the composite
\[
\Omega_X^p \xrightarrow{\ \phi^p\ } \underline{\Omega}_X^p \xrightarrow{\ \psi^p\ } \mathbb{D}_X(\underline{\Omega}_X^{n-p})
\]
is a quasi-isomorphism. At \(k=0\), this recovers ordinary rational singularities [2205.04729].

This definition admits a log-resolution description under codimension bounds on the singular locus \(\Sigma=X_{\mathrm{sing}}\). If \(\dim \Sigma\le d\) and \(p<n-d\), then
\[
\mathbb{D}_X(\underline{\Omega}_X^{n-p}) \cong R\pi_*\Omega_{\widehat{X}}^p(\log E),
\]
where \(\pi:\widehat{X}\to X\) is a log resolution with reduced exceptional divisor \(E\). In particular, if \(\operatorname{codim}\Sigma\ge 2k+1\), then for \(p\le k\),
\[
X \text{ is } k\text{-rational} \iff \Omega_X^p \to R\pi_*\Omega_{\widehat{X}}^p(\log E)
\text{ is an isomorphism for all } p\le k.
\]
For local complete intersections this simplifies further:
\[
X \text{ is } k\text{-rational} \iff R^q\pi_*\Omega_{\widehat{X}}^p(\log E)=0
\quad \text{for all } p\le k,\ q>0.
\]
This formulation makes explicit that higher rationality is a vanishing theory for logarithmic differentials, not only for functions [2205.04729].

A later generalization isolates the part of the theory depending only on higher cohomology sheaves of the Du Bois complex. For a normal variety \(X\), one says that \(X\) is pre-\(k\)-rational if
\[
\mathcal{H}^i\bigl(\mathbb{D}_X(\underline{\Omega}_X^{n-p})\bigr)=0
\quad\text{for all }0\le p\le k,\ i>0,
\]
and then defines a new \(k\)-rational notion by adjoining normality, a codimension bound \(\operatorname{codim}_X(X_{\mathrm{sing}})>2k+1\), and reflexivity of the degree-zero pieces. In the local complete intersection case these definitions agree with the strict ones above [2306.03977].

## 2. Relation to higher Du Bois singularities

The higher rational and higher Du Bois conditions are linked by a direct higher-level analogue of the classical implication rational \(\Rightarrow\) Du Bois. If \(X\) has either local complete intersection singularities or isolated singularities, then
\[
X \text{ \(k\)-rational } \Longrightarrow X \text{ \(k\)-Du Bois}.
\]
The proof generalizes Kovács’ method for \(k=0\), using compatible left inverses to the maps \(\Omega_X^p\to \underline{\Omega}_X^p\) and hyperplane-section arguments based on Navarro Aznar’s triangle. In the local complete intersection case Mustaţă–Popa also gave an independent proof [2205.04729].

Outside the local complete intersection setting, the same implication survives at the level of the “pre-” notions:
\[
\text{pre-}k\text{-rational} \Longrightarrow \text{pre-}k\text{-Du Bois}.
\]
This is proved by working with higher cohomology of Du Bois graded pieces rather than with Kähler differentials in degree zero. The resulting non-lci framework emphasizes that some parts of the theory depend only on vanishing in the Du Bois complex and its duals, whereas reflexivity and codimension constraints are required to recover a robust geometric notion of \(k\)-rationality [2306.03977].

A converse does not generally hold at the same level. In the hypersurface case, and more generally for local complete intersection singularities, one has a one-step implication
\[
X \text{ \(k\)-Du Bois } \Longrightarrow X \text{ \((k-1)\)-rational}.
\]
For hypersurfaces this follows from Morihiko Saito’s appendix identifying the duality-based definition of \(k\)-rationality with a numerical criterion in terms of the minimal exponent; Chen–Dirks–Mustaţă established the corresponding statement for all local complete intersections [2205.04729].

The isolated local complete intersection case is particularly rigid. There,
\[
X \text{ is } k\text{-rational} \iff X \text{ is } k\text{-Du Bois and } \ell^{k,n-k-1}=0,
\]
while
\[
X \text{ is } k\text{-Du Bois} \iff X \text{ is } (k-1)\text{-rational and } b^{k,n-k-1}=0.
\]
These formulas express the difference between the two notions in terms of mixed-Hodge-theoretic invariants of the link and of logarithmic cohomology on a resolution [2207.07566].

## 3. Hodge theory, families, and deformation-theoretic consequences

One of the main motivations for higher rational singularities is the control of Hodge-theoretic behavior in singular families. If \(f:Y\to S\) is a flat proper family of complex algebraic varieties and one fiber \(Y_s\) has \(k\)-Du Bois local complete intersection singularities, then after shrinking \(S\) around \(s\),
\[
R^q f_*\Omega_{Y/S}^p
\]
is locally free and compatible with arbitrary base change for all \(0\le p\le k\) and all \(q\ge 0\). This extends Du Bois’ \(k=0\) theorem and yields constancy of Hodge numbers in the front range:
\[
h^{p,q}(Y_s)=h^{p,q}(Y_t)
\quad \text{for all }0\le p\le k
\]
and every smooth fiber \(Y_t\) over the same irreducible base [2205.04729].

For proper local complete intersection varieties with \(k\)-rational singularities, the Hodge-theoretic consequences are stronger. In dimension \(n\),
\[
h^{p,q}(Y)=h^{q,p}(Y)=h^{n-p,n-q}
\quad \text{for } p\le k,
\]
and one has the identification
\[
h^{p,q}(Y_s)=
\dim \operatorname{Gr}_F^p \operatorname{Im}\bigl(H^{p+q}(Y_s)\xrightarrow{\ \pi_s^*\ }H^{p+q}(\widehat{Y}_s)\bigr)
\]
for any projective resolution \(\pi_s:\widehat{Y}_s\to Y_s\). This realizes the frontier of the Hodge diamond up to coniveau \(k\) as the image of resolved cohomology [2205.04729].

The same techniques also produce deformation-theoretic applications. If \(Y\) is a canonical Calabi–Yau variety, meaning Gorenstein, canonical, and \(\omega_Y\simeq \mathcal{O}_Y\), with \(1\)-Du Bois local complete intersection singularities, then the deformation functor \(\operatorname{Def}(Y)\) is unobstructed. The proof uses the \(p=1\) case of the base-change theorem above, relative duality, and \(T^1\)-lifting [2205.04729].

This suggests that higher rationality is not merely a local birational condition. It also controls how much of the smooth Hodge package survives in singular settings, especially in low differential degree.

## 4. Numerical criteria, isolated singularities, and explicit examples

For isolated local complete intersection singularities, higher rationality admits a complete description in terms of mixed Hodge invariants of the Milnor fiber. If \(s_p=\dim \operatorname{Gr}_F^p H^n(M)\), where \(M\) is the Milnor fiber, then
\[
X \text{ is } k\text{-Du Bois} \iff s_p=0 \text{ for all } p\le k,
\]
and for \(k\le (n-1)/2\),
\[
X \text{ is } k\text{-rational} \iff s_{n-p}=0 \text{ for all } p\le k.
\]
Equivalently, \(k\)-rationality is characterized by the vanishing
\[
H^q(\widetilde{X};\Omega_{\widetilde{X}}^p(\log E))=0
\quad \text{for } 0\le p\le k,\ q>0
\]
on a log resolution \(\widetilde{X}\to X\) with reduced exceptional divisor \(E\) [2207.07566].

For hypersurface singularities, the minimal exponent \(\widetilde{\alpha}_X\) gives a sharp numerical criterion. One has
\[
X \text{ is } k\text{-Du Bois } \iff \widetilde{\alpha}_X \ge k+1,
\qquad
X \text{ is } k\text{-rational } \iff \widetilde{\alpha}_X > k+1.
\]
Saito’s appendix proves that the duality-based definition of \(k\)-rationality agrees with this numerical condition for arbitrary hypersurface singularities, not necessarily isolated [2205.04729].

Weighted homogeneous hypersurfaces provide concrete examples. For
\[
f=\sum z_i^{d_i}
\]
in \(\mathbb{A}^{n+1}\),
\[
X \text{ is } k\text{-Du Bois at }0
\iff \sum_{i=1}^{n+1}\frac{1}{d_i}\ge k+1,
\]
and
\[
X \text{ is } k\text{-rational at }0
\iff \sum_{i=1}^{n+1}\frac{1}{d_i}>k+1.
\]
For an ordinary double point in dimension \(n\), \(k\)-Du Bois holds for \(k\le \lfloor (n-1)/2\rfloor\), while \(k\)-rational holds iff \(k<(n-1)/2\). In particular, a \(3\)-fold node is \(1\)-Du Bois but not \(1\)-rational [2205.04729].

The hypersurface theory can also be read through vanishing cycles and Hirzebruch–Milnor classes. For globally defined reduced hypersurfaces \(X=\{f=0\}\subset Y\) with \(Y\) smooth, higher Du Bois and higher rational singularities can be characterized via vanishing of Hodge-graded pieces of the vanishing-cycle mixed Hodge module, and the spectral Hirzebruch–Milnor class detects the relevant minimal exponent provided the singular locus is projective [2302.00970].

## 5. Alternative cohomological and Hodge-module formulations

Recent work has supplied simpler criteria for higher rationality using the intersection Du Bois complex and the irrationality complex. If \(X\) is normal, let
\[
\mathcal{I}\underline{\Omega}_X^p
:=\operatorname{gr}_{-p}^F\operatorname{DR}_X(\mathcal{IC}_X^H[p-\dim X]),
\]
and define the graded irrationality piece
\[
{}^p_X := \bigl(\underline{\Omega}_X^{\dim X-p},\omega_X^\bullet\bigr)[-\dim X].
\]
Then \(X\) is pre-\(m\)-rational iff, for each \(0\le p\le m\), the natural map
\[
\mathcal{H}^0\bigl(\mathcal{I}\underline{\Omega}_X^p\bigr)\longrightarrow \mathcal{I}\underline{\Omega}_X^p
\]
is an isomorphism in \(D^b_{\mathrm{coh}}(X)\). Equivalently, it is enough that this map admit a left inverse, and equivalently that
\[
\widetilde{{}^p_X}\longrightarrow {}^p_X
\]
admit a left inverse. This directly generalizes the classical splitting criterion for rational singularities [2507.07351].

A different but related generalization shifts the focus to the derived pushforward of torsion-free sheaves under modifications. If \(f:Y\to X\) is a modification of integral Noetherian schemes and \(E\) is torsion-free on \(Y\), then all higher direct images vanish precisely when, stalkwise,
\[
(\mathbb{R}f_*E)_p \in \langle (f_*E)_p\rangle_1
\quad \text{for all }p\in X.
\]
Applied to \(E=\mathcal{O}_{\widetilde{X}}\) on a regular modification \(\widetilde{X}\to X\), this yields a criterion for rational singularities based on “one-cone” generation in the derived category. The same paper introduces mild rational singularities, birational derived splinters, and recipes for producing singularities with rational or Du Bois behavior under projective bundles, base change, and fiber products [2504.19907].

There is also a Hodge-module refinement in which a pure Hodge module \(N\) with strict support \(X\) has rational singularities when
\[
Q_X(N)\simeq \mathcal{H}^{-n}(Q_X(N))[n].
\]
Equivalently, \(S_X(N)\) is maximal Cohen–Macaulay of pure dimension \(n\). For the intersection cohomology Hodge module \(\mathrm{IC}_X^H\), this recovers ordinary rational singularities, and it yields a generalization of Boutot’s theorem for GIT quotients with stable points [2212.01947].

This suggests that higher rationality has become a family of cohomological formalisms centered on the same theme: concentration of Hodge-theoretic or derived data in degree zero, together with vanishing of higher direct images in controlled differential degree.

## 6. Positive-characteristic analogues and reduction modulo \(p\)

In positive characteristic, the absence of resolutions in full generality leads to alternative cohomological analogues. One such notion is Witt-rational singularities. For an integral \(k\)-scheme \(S\) over a perfect field of characteristic \(p>0\), \(S\) has Witt-rational singularities if for every quasi-resolution \(f:X\to S\) with \(X\) smooth,
\[
f^*:W\mathcal{O}_{S,\mathbb{Q}}\xrightarrow{\sim}f_*W\mathcal{O}_{X,\mathbb{Q}},
\]
\[
R^if_*W\mathcal{O}_{X,\mathbb{Q}}=0 \quad (i>0),
\qquad
R^if_*W\omega_{X,\mathbb{Q}}=0 \quad (i>0).
\]
Finite quotients and topological finite quotients are Witt-rational, and the property is stable under universal homeomorphisms. Here “higher” refers to the vanishing of higher direct images of Witt vector and Witt canonical sheaves, together with the action of correspondences on Hodge–Witt cohomology [1104.2145].

A more direct Frobenius-theoretic extension is \(m\)-\(F\)-rationality. For a normal affine variety \(X\) over a perfect field of characteristic \(p>0\), the definition requires surjectivity of the Cartier operator on reflexive \(i\)-forms for \(i\le m\) and injectivity of certain maps on local cohomology built from iterated inverse Cartier operators. The main comparison theorem states that a normal variety over a field of characteristic zero is \(m\)-rational if and only if it is \(m\)-\(F\)-rational after reduction modulo a sufficiently large prime \(p\). The same work proves logarithmic extension theorems for differential forms under \(m\)-\(F\)-rationality [2604.12362].

These positive-characteristic theories are not formally identical to the characteristic-zero \(k\)-rational definition. A plausible implication is that “higher rational singularities” has become an umbrella expression for several parallel attempts to transport rational-singularity phenomena—vanishing of higher direct images, extension of forms, Hodge-theoretic concentration, and birational invariance—into settings where ordinary rational singularities are too rigid or too narrow [1104.2145].

## 7. Scope, non-uniformity, and current directions

The literature makes clear that higher rational singularities are not a single invariant with one canonical definition. In characteristic zero alone, there are strict \(k\)-rational, pre-\(k\)-rational, and new \(k\)-rational notions; on hypersurfaces there are vanishing-cycle and minimal-exponent formulations; in mixed-Hodge-theoretic language there are rationality conditions for Hodge modules; in positive characteristic there are Witt-rational and \(m\)-\(F\)-rational analogues [2306.03977].

Several structural themes recur across these definitions. First, the relevant degree-zero sheaf is typically reflexive or identified with a pushforward of logarithmic differentials from a resolution. Second, the essential condition is vanishing of higher cohomology or higher direct images in a prescribed range \(p\le k\) or \(i\le m\). Third, the resulting theory interacts strongly with hyperplane sections, finite maps, and quotient constructions. For example, pre-\(k\)-rationality is preserved by general hyperplane sections, and if \(f:Y\to X\) is finite surjective with \(Y\) pre-\(m\)-rational, then \(X\) is pre-\(m\)-rational as well [2306.03977].

Open directions are explicit in the current work. Extending the theory beyond local complete intersections remains central because many arguments still rely on control of \(\Omega_X^p\). Shen–Venkatesh–Vo propose alternative non-lci definitions agreeing in the local complete intersection case, and later work on pairs develops higher Du Bois and higher rational pairs with Bertini theorems, finite-map stability, and an \(m\)-rational \(\Rightarrow m\)-Du Bois implication via a generalized Kovács–Schwede injectivity theorem [2205.04729].

The modern theory therefore presents higher rational singularities as a stratified refinement of rationality: classical rational singularities occupy level \(0\), while higher levels record how far singular spaces continue to behave like smooth ones with respect to differential forms, duality, Hodge filtration, and cohomological descent.

Source: https://www.emergentmind.com/topics/higher-rational-singularities