I understand that the Tannakian formalism (which I only "know" extremely superficially) is very important for the theory of motives. I guess "the" conjectural category of motives should be Tannakian, or a full subcategory of a Tannakian category (I hope what I write makes sense).

Now let $\mathcal{V}_k$ be the category of $k$-varieties with $k$ a field, and consider the Grothendieck ring $K_0(\mathcal{V}_k)$. It is the quotient of the free abelian group generated by the isomorphism classes $[X]$ of $k$-varieties by the relations $[A] = [A \setminus C] + [C]$, where $C$ is a closed $k$-subvariety in the $k$-variety $A$; the fiber product over $k$ then induces a ring structure.

As you may know, some aspects of motives over $k$ can be seen in $K_0(\mathcal{V}_k)$, but many others can not.

My question is: how much does $K_0(\mathcal{V}_k)$ "capture" from the (full subcategory of the) Tannakian category which the category of $k$-motives should be ?

  • $\begingroup$ I know you are asking about K_0(Var) and Mot, but just as in Will Sawin's answer I'd like to suggest looking at K_0(Mot). In here you can say very concretely what is lost: K_0 sees no difference between split and non-split extensions, and in particular, every mixed motive agrees with the sum of its weight-graded pieces in this K_0. I think you could be very concrete about this in the case of say, hodge structures, K_0 is somehow only seeing the reductive part of the motivic fundamental group. $\endgroup$
    – pupshaw
    Jul 1, 2021 at 12:32

1 Answer 1


The conjectural category of motives, being an abelian category, has its own notion of $K$-theory, $K_0 ( \operatorname{Mot}_k)$.

This would be the quotient of the free abelian group generated by isomorphism classes of motives by the relations $[B] = [A]+[C]$ when $0 \to A \to B \to C \to 0$ is an exact sequence. (A similar construction works for the non-conjectural triangulated category of motives.)

The ring $K_0 ( \operatorname{Mot}_k)$ contains a lot of the information of the category of motives $\operatorname{Mot}_k$, but not all.

There is a natural map $K_0 (\mathcal V_k) \to K_0 ( \operatorname{Mot}_k)$ that sends the class of a smooth projective variety to its motive.

The fact that this map is not an isomorphism could be interpreted to say that $K_0 (\mathcal V_k)$ loses some of the information in $K_0 ( \operatorname{Mot}_k)$, or alternately that it contains some information about varieties that is missing in $K_0 ( \operatorname{Mot}_k)$. This second perspective is justified by the observation that, despite the seemingly similar definition, it is often more difficult to work with in practice.

I don't know of any reason to believe that there is overlap between the information lost when passing from $\operatorname{Mot}_k$ to $K_0(\operatorname{Mot}_k)$ and the information lost when passing from $K_0(\mathcal V_k)$ to $K_0(\operatorname{Mot}_k)$, but there might be some overlap.

In summary, the Grothendieck group of varieties is removed by two steps from the category of motives, and one of the steps causes a clear loss of information while the other step is more ambiguous and arguably involves a gain of information.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.