Wirthmüller context





Special and general types

Special notions


Extra structure






A Wirthmüller context is a pair of two symmetric closed monoidal categories (𝒳, X,1 X)(\mathcal{X}, \otimes_X, 1_{X}), (𝒴, Y,1 Y)(\mathcal{Y}, \otimes_Y, 1_Y) which are connected by an adjoint triple of functors such that the middle one is a closed monoidal functor.

This is the variant/special case of the yoga of six operations consisting of two adjoint pairs (f !f !)(f_! \dashv f^!) and (f *f *)(f^\ast \dashv f_\ast) and the tensor product/internal hom adjunctions (()B[B,])((-)\otimes B \dashv [B,-]), specialized to the case that f !f *f^! \simeq f^\ast:

f !(f !=f *)f *:𝒳f *f !=f *f !𝒴. f_! \dashv (f^! = f^\ast) \dashv f_\ast \;\colon\; \mathcal{X} \stackrel{\overset{f_!}{\longrightarrow}}{\stackrel{\overset{f^! = f^\ast }{\leftarrow}}{\underset{f_\ast}{\longrightarrow}}} \mathcal{Y} \,.

(The other specialization of the six operations to f *f !f_\ast \simeq f_! is called the Grothendieck context).

Often one is interested in the case that there is an object C𝒳C \in \mathcal{X} and an equivalence

f *1 Xf !C. f_\ast 1_{X} \simeq f_! C \,.

If this induces a natural equivalence

f *Af !(A XC) f_\ast A \simeq f_!(A \otimes_X C)

for A𝒳A \in \mathcal{X}, then one says this is a Wirthmüller isomorphism, following (Wirthmueller 74). In particular there is a canonical natural transformation

f *Af !(A XC) f_\ast A \longrightarrow f_!(A \otimes_X C)

and one can ask this to be an equivalence, hence a Wirthmüller isomorphism (May 05).



Let (𝒳, X,1 X)(\mathcal{X}, \otimes_X, 1_{X}), (𝒴, Y,1 Y)(\mathcal{Y}, \otimes_Y, 1_Y) be two symmetric closed monoidal categories and let

f !(f !=f *)f *:𝒳f *f !=f *f !𝒴 f_! \dashv (f^! = f^\ast) \dashv f_\ast \;\colon\; \mathcal{X} \stackrel{\overset{f_!}{\longrightarrow}}{\stackrel{\overset{f^! = f^\ast }{\leftarrow}}{\underset{f_\ast}{\longrightarrow}}} \mathcal{Y}

be an adjoint triple of functors between them. We call this setup

(May 05, def. 2.12)

Definition (Notation)

We write [,][-,-] for the internal hom functors. For A𝒳A \in \mathcal{X} we write

𝔻A[A,1 X] \mathbb{D}A \coloneqq [A,1_X]

for the internal hom from AA into the unit object, hence for dual of AA with respect to the closed category structure, its dual object in a closed category.

We say “AA is dualizable” to mean that it is a dualizable object with respect, insead, to the (symmetric) monoidal category structure X\otimes_X. If AA is dualizable we write A A^\vee for its monoidal dual object.

Similarly for B𝒴B \in \mathcal{Y}.


If all objects in 𝒳\mathcal{X} and 𝒴\mathcal{Y} are dualizable, hence if they are compact closed categories, then they are in particular also star-autonomous categories with dualizing object the tensor unit. As such their internal logic is linear logic and their type theory is linear type theory. In terms of this a Wirthmüller morphism, def. 1, is the linear analog of a context extension morphism in a hyperdoctrine: it defines a dependent linear type theory. See there for more.


The comparison maps


In a pre-Wirthmüller context, def. 1, there is a canonical natural transformation

f !(A XB)(f !A) Y(f !B), f_!(A \otimes_X B) \longrightarrow (f_! A) \otimes_Y (f_! B) \,,

(not necessarily an equivalence) being the adjunct of the composite

A XB(f *f !A) X(f *f !B)f *((f !A) Y(f !B)), A \otimes_X B \stackrel{}{\longrightarrow} (f^\ast f_! A) \otimes_X (f^\ast f_! B) \stackrel{\simeq}{\longrightarrow} f^\ast ( (f_! A) \otimes_Y (f_! B) ) \,,

where the first morphism is the tensor product of two copies of the adjunction unit and where the second is the equivalence that exhibits f *f^\ast as a strong monoidal functor.


In a pre-Wirthmüller context, def. 1, write π¯\overline {\pi} for the natural transformation

π¯:f !((f *B)A)Bf !A \overline{\pi} \;\colon\; f_! ((f^\ast B) \otimes A) \longrightarrow B \otimes f_! A

given as the composite

π¯:f !((f *B)A)(f !f *B) Y(f !A)Bf !A, \overline{\pi} \;\colon\; f_! ((f^\ast B) \otimes A) \stackrel{}{\longrightarrow} (f_! f^\ast B) \otimes_Y (f_! A) \longrightarrow B \otimes f_! A \,,

where the first morphism is that of remark 2 and where the second is the (f !f *)(f_! \dashv f^\ast) counit (tensored with an identity).

Also write

γ¯:[f !A,B]f *[A,f *B] \overline{\gamma} \;\colon\; [f_! A, B] \longrightarrow f_\ast [A, f^\ast B]

for the (f *f *)(f^\ast \dashv f_\ast) adjunct of the natural transformation given as the composite

f *[f !A,B][f *f !A,f *B][A,f *B], f^\ast [f_! A, B] \stackrel{\simeq}{\longrightarrow} [f^\ast f_! A, \, f^\ast B] \longrightarrow [A, f^\ast B] \,,

where the first map exhibits f *f^\ast as a closed monoidal functor and where the second is the (f !f *)(f_! \dashv f^\ast)-unit (under the internal hom).

see (May 05, prop. 2.11)


In a pre-Wirthmüller context, def. 1, the comparison maps of def. 3 are equivalences when the argument B(𝒴, Y,1 Y)B \in (\mathcal{Y}, \otimes_Y, 1_Y) is a dualizable object.

If either of the two happens to be a natural equivalence (hence an equivalence for all arguments), then so is the other.

(May 05, prop. 2.8 and prop. 2.11)


Precisely if the pre-Wirthmüller context is a Wirthmüller context, def. 1, are both comparison maps of def. 3 are natural equivalences.


For all A𝒳A \in \mathcal{X} and B,C𝒴B,C \in \mathcal{Y} we have by the (f !f *)(f_! \dashv f^\ast)-adjunction and the tensor\dashvhom-adjunction a commuting diagram of the form

𝒴(Bf !A,C) 𝒴(π¯(A,B),C) 𝒴(f !((f *B)A),C) 𝒳(A,f *[B,C]) 𝒳(A,[(f *B),(f *C)]). \array{ \mathcal{Y}(B \otimes f_! A, \, C ) & \stackrel{ \mathcal{Y}(\overline{\pi}(A,B), C) }{ \longrightarrow } & \mathcal{Y}(f_! ((f^\ast B) \otimes A),\, C) \\ \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} \\ \mathcal{X}(A, f^\ast [B,C]) &\stackrel{}{\longrightarrow}& \mathcal{X}(A, [(f^\ast B), (f^\ast C)]) } \,.

By naturality in AA and by the Yoneda lemma this shows that π¯\overline{\pi} is an equivalence precisey if f *f^\ast is strong closed.

For γ¯\overline{\gamma} the same statement follows from this with prop. 1.


The first natural equivalence of prop. 2

f !(f *AB)Af !(B) f_!(f^\ast A \otimes B) \simeq A \otimes f_!(B)

is often called the projection formula. In representation theory this is also sometimes called Frobenius reciprocity, though mostly that term is used for (just) the existence of the (f !f *)(f_! \dashv f^\ast)-adjunction, where in representation theory the left adjoint f !f_! forms induced representations.

Comparison of push-forwards and Wirthmüller isomorphism


In a pre-Wirthmüller context, def. 1, the functors f !f_! and f *f_\ast are intertwined by dualization, in that there is a natural equivalence

𝔻(f !A)f *(𝔻A). \mathbb{D}(f_! A) \simeq f_\ast(\mathbb{D} A) \,.

This is the special case of the property of γ¯\overline{\gamma} in prop. 1 for B=1 YB = 1_Y:

𝔻(f !A)=[f !A,1 Y]γ¯f *[A,f *1 Y]f *[A,1 Y]=f *𝔻A. \mathbb{D}(f_! A) = [f_! A, 1_Y] \underoverset{\simeq}{\overline{\gamma}}{\longrightarrow} f_\ast [A, f^\ast 1_Y] \simeq f_\ast [A, 1_Y] = f_\ast \mathbb{D} A \,.

With a Wirthmüller context regarded as categorical semantics for dependent linear type theory (see there), then the statement of cor. 1 is an instance of de Morgan duality where linear dualization intertwines linear dependent sum \sum and dependent product \prod

f𝔻𝔻 f. \prod_f \mathbb{D} \simeq \mathbb{D} \sum_f \,.

For more on this see also (Schreiber 14, section 3.3).


In a pre-Wirthmüller context

f *1 X𝔻(f !1 X). f_\ast 1_X \simeq \mathbb{D}(f_! 1_X) \,.

By cor. 1, since 𝔻1 X1 X\mathbb{D} 1_X \simeq 1_X.


(Wirthmüller isomorphism)

In a Wirthmüller context, def. 1 if f !1 Xf_! 1_X is a dualizable object with dual f !Cf_! C, then there is a natural equivalence

ω:f *f *Af !((f *A)C). \omega \;\colon\; f_\ast f^\ast A \stackrel{\simeq}{\longrightarrow} f_!((f^\ast A) \otimes C) \,.

(May 05, prop. 4.13).


In particular if there is D𝒴D \in \mathcal{Y} with

𝔻(f !f *1 Y)f !f *D \mathbb{D}(f_! f^\ast 1_Y) \simeq f_! f^\ast D

(hence if Cf *DC \simeq f^\ast D in the notation of prop. 3) and using that by cor. 1, f *f *𝔻Bf *𝔻f *B𝔻(f !f *B) f_\ast f^\ast \mathbb{D}B \simeq f_\ast \mathbb{D} f^\ast B \simeq \mathbb{D}(f_! f^\ast B) then prop. 3 gives a natural equivalence of the form

𝔻(f !f *B)f !f *((𝔻B)D), \mathbb{D}(f_! f^\ast B) \stackrel{\simeq}{\longrightarrow} f_! f^\ast ((\mathbb{D}B) \otimes D) \,,

saying that the comonad f !f *f_! f^\ast commutes with dualization up to a “twist” given by tensoring with DD.


Cartesian Wirthmüller contexts in toposes

For H\mathbf{H} a topos and f:XYf \colon X \longrightarrow Y any morphism, then in the induced base change etale geometric morphism

( ff * f):H /XH /Y (\sum_f \dashv f^\ast \dashv \prod_f) \;\colon\; \mathbf{H}_{/X} \longrightarrow \mathbf{H}_{/Y}

the inverse image/context extension is a cartesian closed functor (see there for the proof). Therefore any base change of toposes constitutes a cartesian Wirthmüller context.

Conversely, this means that systems of Wirthmüller contexts are generalizations of categorical logic (hyperdoctrines) to non-cartesian contexts (see at dependent linear type theory).

Notice that in a cartesian Wirthmüller context duality is trivial, in that 𝔻X1\mathbb{D}X \simeq 1 for all objects XX. Therefore to the extent that the six operations yoga involves duality, it is interesting only the more non-cartesian (non-classical) the ambient Wirthmüller context is.

For instance the projection formula γ¯\overline{\gamma} in def. 3 for base change along a pointed connected type BG*\mathbf{B}G \to \ast equivalently says that genuine (Bredon) equivariant cohomology reduces to Borel equivariant cohomology when the action on the coefficients is trivial. See at equivariant cohomology for more on this.

Pointed objects with smash product

A first step away from the Cartesian example above is the following.

Let H\mathbf{H} be a topos. For XHX \in \mathbf{H} any object, write

𝒞 XH /X X/ \mathcal{C}_X \coloneqq \mathbf{H}_{/X}^{X/}

for the category of pointed objects in the slice topos H /X\mathbf{H}_{/X}. Equipped with the smash product X\wedge_X this is a closed symmetric monoidal category (𝒞 X, X,XX)(\mathcal{C}_X, \wedge_X, X \coprod X).


For f:XYf \colon X \longrightarrow Y any morphism in H\mathbf{H}, the base change inverse image f *f^\ast restricts to a functor f *:𝒞 Y𝒞 Xf^\ast \colon \mathcal{C}_Y \longrightarrow \mathcal{C}_X which is a Wirthmüller context.

This appears as (Shulman 07, examples 12.13 and 13.7) and (Shulman 12, example 2.33).


For f:XYf \colon X \longrightarrow Y any morphism in H\mathbf{H} then the base change inverse image f *:H /YH /Xf^\ast \colon \mathbf{H}_{/Y} \longrightarrow \mathbf{H}_{/X} preserves pointedness, and the pushout functor f !:H X/H /Yf_! \colon \mathbf{H}^{X/} \longrightarrow \mathbf{H}^{/Y} preserves co-pointedness. These two functors hence form an adjoint pair (f 1f *):𝒞 X𝒞 Y(f_1 \dashv f^\ast) \colon \mathcal{C}_X \longrightarrow \mathcal{C}_Y. Moreover, since colimits in the under-over category H /X X/\mathbf{H}_{/X}^{X/} are computed as colimits in H\mathbf{H} of diagrams with an initial object adjoined, and since by the Giraud axioms in the topos H\mathbf{H} pullback preserves these colimits, it follows that f *:𝒞 Y𝒞 Xf^\ast \colon \mathcal{C}_Y \to \mathcal{C}_X preserves colimits. Finally by the discussion at category of pointed objects we have that 𝒞 X\mathcal{C}_X and 𝒞 Y\mathcal{C}_Y are locally presentable categories, so that by the adjoint functor theorem it follows that f *f^\ast has also a right adjoint f *:𝒞 X𝒞 Yf_\ast \colon \mathcal{C}_X \to \mathcal{C}_Y.

To see that f *f^\ast is a strong monoidal functor observe that the smash product is, by the discussion there, given by a pushout over coproducts and products in the slice topos. As above these are all preserved by pullback. Finally to see that f *f^\ast is also a strong closed functor observe that the internal hom on pointed objects is, by the discussion there, a fiber product of cartesian internal homs. These are preserved by the above case, and the fiber product is preserved since f *f^\ast preserves all limits. Hence f *f^\ast preserves also the internal homs of pointed objects.

Bundles of modules

For RR a ring, RModR Mod its category of modules, there is a functor

[,RMod]:Set opClMonCat [-, R Mod] \;\colon\; Set^{op} \longrightarrow ClMonCat

which sends a set XX to the closed monoidal category of RR-modules parameterized over XX.

This takes values in Wirthmüller morphisms. (Shulman 12, example 2.2, 2.17).

Bundles of \infty-modules

for (infinity,1)-module bundles: (Nuiten 13, Hopkins-Lurie, Schreiber 14)

Becker-Gottlieb transfer

A transfer context is a Wirthmüller context in which also f *f_\ast satisfies its projection formula. In this context there is an abstract concept of Becker-Gottlieb transfer. See there for more.

In equivariant stable homotopy theory

In equivariant stable homotopy theory, see (May 05b).

For quasicoherent sheaves (in E E_\infty-geometry)

Pull-push of quasicoherent sheaves is usually discussed as a Grothendieck context of six operations, but under some conditions it also becomes a Wirthmüller context.

Using results of Lurie this follows in the full generality of E-∞ geometry (spectral geometry).

Consider quasi-compact and quasi-separated E-∞ algebraic spaces? (spectral algebraic spaces?). (This includes precisely those spectral Deligne-Mumford stacks which have a scallop decomposition, see here.)

If f:XYf \;\colon\; X \longrightarrow Y is a map between these which is

  1. locally almost of finite presentation;

  2. strongly proper;

  3. has finite Tor-amplitude?

then the left adjoint to pullback of quasicoherent sheaves exists

(f !f *):QCoh(X)f *f !QCoh(Y). (f_! \dashv f^\ast) \;\colon\; QCoh(X) \stackrel{\overset{f_!}{\longrightarrow}}{\underset{f^\ast}{\longleftarrow}} QCoh(Y) \,.

(LurieProper, proposition 3.3.23)

If ff is

  • quasi-affine

then the right adjoint exists

(f *f *):QCoh(X)f *f *QCoh(Y). (f^\ast \dashv f_\ast) \;\colon\; QCoh(X) \stackrel{\overset{f^\ast}{\longleftarrow}}{\underset{f_\ast}{\longrightarrow}} QCoh(Y) \,.

(LurieQC, prop. 2.5.12, LurieProper, proposition 2.5.12)

The projection formula in the dual form

f *ABf *(Af *B) f_\ast A \otimes B \longrightarrow f_\ast (A\otimes f^\ast B)

for ff quasi-compact and quasi-separated appears as (LurieProper, remark 1.3.14).

Now if all the conditions on ff hold, so that (f !f *f *):QCoh(X)QCoh(Y)(f_! \dashv f^\ast \dashv f_\ast) \;\colon\; QCoh(X) \longrightarrow QCoh(Y), then passing to opposite categories QCoh(X) opQCoh(Y) opQCoh(X)^{op} \longrightarrow QCoh(Y)^{op} exchanges the roles of f !f_! and f *f_\ast, makes the projection formula be as in the above discussion and hence yields a Wirthmüller context.

The existence of dualizing modules KK

DX=[X,K] D X = [X,K]

is discussed in (Lurie, Representability theorems, section 4.2.)


The construction goes back to and is named after

  • Klaus Wirthmüller, Equivariant homology and duality. Manuscripta Math. 11(1974), 373-390

A clear discussion of axioms of six operations and their consequences, with emphasis on the Wirthmüller isomorphisms, is in

Discussion of the Wirthmüller isomorphism in equivariant stable homotopy theory, based on this, is in

More elaboration of the Wirthmüller context is in

  • Baptiste Calmès, Jens Hornbostel, section 4 of Tensor-triangulated categories and dualities, Theory and Applications of Categories, Vol. 22, 2009, No. 6, pp 136-198 (TAC, arXiv:0806.0569)

Discussion in the context of pure motives includes

Wirthmüller morphisms between pointed objects and between bundles of modules are discussed in

Discussion in E-∞ geometry is in

Revised on December 11, 2016 12:26:15 by Urs Schreiber (