Milnor-Witt Motivic Cohomology

• Milnor–Witt motivic cohomology is a refined cohomology theory that incorporates quadratic form data to recover Chow–Witt groups and Milnor–Witt K-theory.
• The framework employs MW-transfers based on Chow–Witt groups and Rost–Schmid complexes, offering a detailed account of intersection products and transfer functoriality.
• Its structure bridges classical motivic cohomology and hermitian K-theory through explicit computations, symplectic orientations, and a graded-commutative cup product with a quadratic refinement.

Searching arXiv for primary sources on Milnor–Witt motivic cohomology and closely related developments. First, I’ll locate the foundational paper and several comparison/structural papers needed to support a comprehensive article. Milnor–Witt motivic cohomology is the version of motivic cohomology obtained by enriching algebraic cycles with symmetric bilinear form data, so that transfers and intersection products keep track of quadratic refinements. Its construction is modeled on Voevodsky’s theory, but it starts from Chow–Witt theory and Milnor–Witt $K$-theory; the resulting transfers are weaker than Voevodsky transfers, and the derived category of motives is correspondingly closer to the stable homotopy theory of schemes (Bachmann et al., 2020). In its basic form, for a smooth scheme $X$ over a perfect field $k$, Milnor–Witt motivic cohomology is defined by

$$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$

and it recovers Chow–Witt groups in bidegree $(2n,n)$ and Milnor–Witt $K$-theory on the diagonal $(n,n)$ (Bachmann et al., 2020).

1. Finite Milnor–Witt correspondences

The foundational object is the category of finite Milnor–Witt correspondences. Let $k$ be a perfect field and $Sm$ the category of smooth separated $k$-schemes of finite type. For $X$0 equidimensional with $X$1, one replaces Voevodsky’s finite correspondences by correspondences whose coefficients lie in Chow–Witt groups. A closed subset $X$2 is admissible if $X$3 is finite and each component surjects onto a component of $X$4, and the group of finite MW-correspondences is

$X$5

Here the coefficients are Chow–Witt groups, equivalently cohomology groups computed via the Rost–Schmid complex with coefficients in $X$6 and twist by the pullback of $X$7 (Bachmann et al., 2020).

Composition is defined by the same pullback–product–pushforward pattern as for Voevodsky correspondences, but orientations are encoded by quadratic form data. This yields a canonical additive symmetric monoidal category $X$8 with objects smooth schemes and $X$9. The graph functor

$k$0

is faithful and symmetric monoidal. There is also a functor

$k$1

induced by the forgetful map from Chow–Witt to Chow; after inverting $k$2, $k$3 is surjective and split, so $k$4 is full when $k$5 is invertible in the coefficient ring (Bachmann et al., 2020).

A characteristic feature of the theory is that MW transfers are weaker than Voevodsky transfers. The representable presheaves $k$6 on $k$7 are generally not Nisnevich sheaves, and MW sheafification must be imposed separately. This is one of the first concrete ways in which the theory departs from ordinary motivic cohomology: the loss of some transfer functoriality is compensated by the retention of quadratic information (Bachmann et al., 2020).

2. Effective and stable MW motives

For a Grothendieck topology $k$8, a presheaf with MW-transfers is an additive contravariant functor $k$9, and an MW $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$0-sheaf is such a presheaf whose restriction to $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$1 is a $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$2-sheaf. If $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$3 is a coefficient ring, the abelian category of MW $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$4-sheaves of $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$5-modules is denoted $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$6. The representable MW presheaf $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$7 has sheafification $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$8, and these representables generate $$H^{p,q}_{MW}(X,R):=\mathrm{Hom}_{DM^{\mathrm{eff}}_{MW}(k,R)}\big(M(X),\,\tilde R(q)[p]\big),$$9 (Bachmann et al., 2020).

Passing to complexes and then $(2n,n)$0-localizing the derived category gives the effective category of MW motives

$(2n,n)$1

This is a triangulated symmetric monoidal category. For a smooth $(2n,n)$2, the effective MW motive $(2n,n)$3 is $(2n,n)$4 in degree $(2n,n)$5 (Bachmann et al., 2020).

The Tate object is

$(2n,n)$6

and the twist notation is organized by $(2n,n)$7, $(2n,n)$8, and $(2n,n)$9. With coefficients $K$0, one introduces

$K$1

For $K$2 perfect and infinite,

$K$3

and for $K$4 and $K$5,

$K$6

In particular,

$K$7

so the theory recovers the Chow–Witt group in codimension $K$8 (Bachmann et al., 2020).

The stable category $K$9 is obtained by inverting $(n,n)$0 via Tate MW spectra. Over an infinite perfect field of characteristic $(n,n)$1, tensoring with the Tate object is fully faithful: $(n,n)$2 equivalently $(n,n)$3 is fully faithful. This is the MW cancellation theorem, first proved at the correspondence level and then lifted to motives (Fasel et al., 2017).

3. Diagonal cohomology and Milnor–Witt $(n,n)$4-theory

The diagonal part of MW motivic cohomology is controlled by Milnor–Witt $(n,n)$5-theory. For a field $(n,n)$6, the graded ring $(n,n)$7 is generated by symbols $(n,n)$8 of degree $(n,n)$9 for $k$0 and $k$1 of degree $k$2, subject to the relations

$k$3

The degree-zero class $k$4 identifies $k$5 with $k$6, and negative degrees identify with Witt groups (Calmès et al., 2017).

For finitely generated field extensions $k$7 with $k$8 infinite perfect and $k$9, the comparison theorem states that for every $Sm$0,

$Sm$1

is a canonical isomorphism of graded rings, natural for field extensions and compatible with transfers (Calmès et al., 2017). In the systematic theory, this extends sheaf-theoretically: for any smooth $Sm$2 and $Sm$3,

$Sm$4

and the cohomology sheaf of the MW motivic complex satisfies

$Sm$5

as Nisnevich sheaves (Bachmann et al., 2020).

This diagonal identification is one of the central structural facts of the theory. It shows that MW motivic cohomology is not merely “motivic cohomology with extra signs”: it is built so that the basic coefficient objects already encode Grothendieck–Witt and Witt-theoretic data. The Rost–Schmid complex provides the computational mechanism for this identification, and the category of MW cycle modules gives the corresponding abstract formalism. In Feld’s equivalence, the heart of the homotopy $Sm$6-structure on $Sm$7 is equivalent to the category of MW cycle modules; for the MW Eilenberg–MacLane spectrum $Sm$8, MW motivic cohomology is computed by MW Rost–Schmid hypercohomology, recovering

$Sm$9

and its twisted variants (Feld, 2019).

For a finitely generated field extension $k$0, the field computation takes the form

$k$1

This isolates the genuinely quadratic contribution precisely on the diagonal (Bachmann et al., 2020).

4. Representability, operations, and non-oriented behavior

Milnor–Witt motivic cohomology is representable both in the $k$2-derived category and in the stable motivic homotopy category. Using the adjunction between $k$3 and $k$4, one defines the MW motivic ring spectrum

$k$5

and for smooth $k$6,

$k$7

A parallel construction in the $k$8-derived category also produces Borel–Moore homology, homology, and cohomology with compact support for singular schemes, with twists by Thom objects of virtual bundles (Déglise et al., 2017).

The formal properties follow the pattern of a six-functor theory, but with the twists and orientations made explicit. MW motivic cohomology enjoys homotopy invariance, localization exact sequences, base change for Tor-independent squares, projection formulas, and Gysin morphisms. For smooth $k$9 and virtual bundle $X$00,

$X$01

The cup product is graded-commutative with a quadratic sign: $X$02 Pushforwards for finite morphisms depend on twists by relative canonical bundles and orientations of $X$03; this is exactly where the quadratic refinement enters the transfer formalism (Bachmann et al., 2020).

A common misconception is that MW motivic cohomology is an oriented theory analogous to ordinary motivic cohomology. The theory is explicitly non-orientable globally: universal Thom classes and Chern classes do not exist in general, and the projective bundle theorem fails in the MW setting (Déglise et al., 2017). However, this does not mean that all orientation theory disappears. Yang shows that MW-motivic cohomology is symplectically oriented: quaternionic projective bundle theorems and Thom isomorphisms exist for symplectic bundles, and the opposite category of effective Chow–Witt motives embeds fully faithfully into the effective MW-motivic category (Yang, 2018). This contrast between non-orientability and symplectic orientation is one of the distinctive structural features of the theory.

5. Comparison with ordinary motives, Witt theory, and hermitian $X$04-theory

There are canonical comparison functors from MW motives to ordinary motives and to motivic stable homotopy: $X$05 When $X$06, the MW categories split into a $X$07-part and an ordinary motivic part: $X$08 After rationalization, MW motives agree with ordinary motives over a perfect field (Bachmann et al., 2020).

The comparison with ordinary motivic cohomology is mediated by the forgetful map $X$09 and the induced maps $X$10. Over fields, ordinary motivic cohomology is recovered from MW motivic cohomology by killing $X$11, while the remaining quadratic information is controlled by Grothendieck–Witt and Witt theory (Calmès et al., 2017). This suggests that MW motivic cohomology should be read as a quadratic refinement of classical motivic cohomology rather than as a separate replacement for it.

The connection with hermitian $X$12-theory is made explicit in the recent arithmetic extension over Dedekind bases. Over schemes essentially smooth over a Dedekind scheme $X$13, the Milnor–Witt motivic cohomology spectrum $X$14 is defined by a homotopy pullback

$X$15

and fits into a distinguished triangle

$X$16

In low degrees over a field $X$17, the comparison map to hermitian $X$18-theory gives an exact sequence

$X$19

and $X$20 is an isomorphism for $X$21. Rationally, the very effective slice spectral sequence yields a splitting

$X$22

which furnishes a Grothendieck–Riemann–Roch statement relating hermitian $X$23-theory to MW and ordinary motivic cohomology (Kolderup et al., 19 Sep 2025).

6. Computations and geometric examples

The theory admits explicit computations on geometric families where units, residues, and Gysin triangles can be controlled. For the complement $X$24 of a finite arrangement of affine hyperplanes in $X$25, with $X$26 perfect of characteristic $X$27, the total MW motivic cohomology ring is generated by degree-$X$28 unit classes $X$29 subject to the relations

$X$30

Equivalently, the algebra is a quadratic refinement of the Orlik–Solomon algebra with anti-commutativity governed by $X$31 and circuit relations refined by $X$32 and $X$33 (Peng, 2020). Under complex realization, this collapses to the usual Orlik–Solomon presentation; under real realization, the theory factors through $X$34-cohomology and the singular cohomology of the real spectrum (Peng, 2020).

A different computational regime appears for split MW-motives, namely finite direct sums of $X$35 and $X$36. In that situation, an MW-motivic cohomology class is determined by a motivic cohomology class and a Witt cohomology class, subject to a compatibility condition expressed through the motivic Bockstein associated to $X$37. This framework yields splitting formulas for Grassmannian bundles and complete flag bundles and implies that the integral cohomology of real complete flags has only $X$38-torsions (Yang, 2020).

For Stiefel varieties $X$39, the integral MW-motivic cohomology groups have been computed additively as $X$40-modules, and the MW-motive decomposes into tensor products of punctured affine-space motives and $X$41-cone summands. The basic building blocks are

$X$42

and the Euler classes of the frame-forgetting fibrations satisfy

$X$43

This gives a concrete illustration of how $X$44-sensitive summands refine classical motivic decompositions (Peng, 2024).

7. Hypotheses, limitations, and scope

The standard hypotheses are not cosmetic. Many structural statements require that the base field $X$45 be perfect and infinite, and characteristic $X$46 is used repeatedly in the theory of quadratic forms, Chow–Witt groups, MW transfers, and the cancellation theorem (Bachmann et al., 2020). Smoothness is built into the definition of MW $X$47-theory sheaves, the Rost–Schmid complex, and the purity and Gysin formalism. Singular schemes can be treated via Borel–Moore homology and the ring-spectrum formalism, but many core statements are still formulated first for smooth schemes (Déglise et al., 2017).

Several limitations are intrinsic rather than accidental. MW transfers are weaker than Voevodsky transfers, the representable presheaves are not automatically Nisnevich sheaves, and formulas require explicit twists by canonical bundles and virtual orientations (Bachmann et al., 2020). The projective bundle theorem fails in general, and symplectic orientation replaces full orientation only in special cases (Yang, 2018). In the MW cycle-module framework, general pullbacks for flat morphisms were noted as not yet constructed, even though localization, coniveau spectral sequences, and the bivariant formalism suffice for many applications (Feld, 2019).

Within those constraints, Milnor–Witt motivic cohomology provides a coherent quadratic refinement of motivic cohomology. Its diagonal recovers Milnor–Witt $X$48-theory, its codimension-$X$49 part recovers Chow–Witt groups, its ring spectrum sits naturally between ordinary motivic cohomology and hermitian $X$50-theory, and its computational apparatus—Rost–Schmid complexes, Gysin triangles, cancellation, and slice methods—makes the quadratic layer of motivic homotopy theory accessible in explicit terms (Bachmann et al., 2020).