6
$\begingroup$

Let $\mathbf{Sch}_k$ be the category of $k$-schemes of finite type, and let $K_0(\mathbf{Sch}_k)$ be the Grothendieck ring of $k$-schemes. Let $\mathbb{Z}[\mathbb{L}]$ the subring generated by the virtual Lefschetz motive $\mathbb{L} := [\mathbb{A}^1_k]$. Is every element of $\mathbb{Z}[\mathbb{L}]$ the class of a $k$-scheme (or at least a $k$-constructible set) ? It might be convenient to recall that $K_0(\mathbf{Sch}_k)$ is generated by isomorphism classes of $k$-schemes $[X]$, and that $$[X] = [X \setminus C] + [C]$$ whenever $C$ is a closed subscheme in $X$; the product is given by $$[X] \cdot [Y] = [X \times Y].$$

$\endgroup$
5
  • 3
    $\begingroup$ Not $-1$, surely. $\endgroup$ – Will Sawin May 4 '17 at 15:02
  • $\begingroup$ Do you know of a general property on the form of the elements of $\mathbb{Z}[\mathbb{L}]$ so that the answer would be "yes" ? (Say, with $k$ a finite prime field.) Thanks ! $\endgroup$ – THC May 4 '17 at 15:12
  • $\begingroup$ I think the leading term must be positive. $\endgroup$ – Will Sawin May 4 '17 at 15:13
  • $\begingroup$ Or maybe "slightly" more restrictive: if $P(\mathbb{L}) \in \mathbb{Z}[\mathbb{L}]$ and $k = \mathbb{F}_p$, then $P(p^n) > 0$ for all positive integers $n$. What do you think ? $\endgroup$ – THC May 4 '17 at 15:59
  • $\begingroup$ I was thinking over an infinite field. I bet you can extract a proof from arxiv.org/pdf/0907.0488.pdf $\endgroup$ – Will Sawin May 4 '17 at 20:36
6
$\begingroup$

Here is a proof of the fact that over an infinite field $f(\mathbb{L})$ is a class of a scheme if and only if the leading term of $f$ is positive, as suggested by Will Sawin.

Example. For $f(t)=a_nt^n+a_{n-1}t^{n-1}+\dots+a_0$ with $a_n>0$ throw away from affine space of dimension $n$ planes $\mathbb{A}^i$ in amount $-a_i$ for each negative $a_i$ such that these planes are pairwise disojoint(this is possible due to the infinitness of the base field). Then take the disjoint union of the resulting complement with $a_n-1$ copies $\mathbb{A}^n$ and with $a_j$ copies of $\mathbb{A}^j$ for each positive $a_j$. The variety which we get clearly equals to $f(\mathbb{L})$ in $K_0$(one can construct an irreducible example by doing blow-ups instead of disjoint unions).

All other polynomials do not give a class of a scheme. 1) First, consider the case $\mathrm{char}\, k=0$ and reduce by a Lefschetz principle type argument to the case $k=\mathbb{C}$. Assume that $[f(\mathbb{L})]=[X]$ and $a_n<0$. This identity is a linear combination of a finite number of standard identities. There exists a subfield $F$ of $k$, finitely generated over $\mathbb{Q}$ such that all schemes involved in these identities are defined over $F$. Thus, we have $[p(\mathbb{L})]=[X_F]$ already in $K_0(Sch_F)$. Then embedding $F$ into $\mathbb{C}$ and base changing to $\mathbb{C}$ we arrive at the case $k=\mathbb{C}$.

Over $\mathbb{C}$ we can use the following motivic measure(strictly speaking, this formula defines it only on $K_0(Var_{\mathbb{C}})$ but if I got the definition of $K_0(Sch)$ right, the relations imply $[X]=[X_{red}]$ -- there is still something to be said about not-separatedness, but let's omit that): $$\mu([X])=\sum_{k,p,q}(-1)^k\dim H^k_c(X,\mathbb{C})^{p,q}u^pv^q\in \mathbb{Z}[u,v]$$ where by $H^k_c(X,\mathbb{C})^{p,q}$ I mean the piece of weight $(p,q)$ in the MHS on $H^k_c(X)$. For an irreducible variety of dimension $n$ the leading term of $\mu([X])$ is $u^nv^n$ so it is at least positive for arbitrary varirty. But for $f(\mathbb{L})$ this motivic measure gives $f(uv)$, qed.

2)The case $\mathrm{char}\, k=p>0$. Exactly as in the first case, we may assume that we have a counterexample defined over a ring which is a finite extension of $\mathbb{F}_p[t_1,\dots,t_n]$(say, Noether normalization). Specializing at some values of $t_i$ we arrive at the case $k=\mathbb{F_q}$ where we can use the motivic measure suggested by THC: $$\mu_n([X])=\#X(\mathbb{F}_{q^n})$$ and for $f(\mathbb{L})$ with negative leading coefficient there exists a big enough $n$ such that $\mu_n(f(\mathbb{L}))<0$.

For finite fields everything seems more delicate -- a necessary condition is $f(q^{nm})\geq f(q^n)\geq 0$ for any $n,m$.

$\endgroup$
1
  • 1
    $\begingroup$ One can also prove the characteristic zero case by a spreading out method, as you do for the characteristic p case. $\endgroup$ – Will Sawin May 6 '17 at 5:09

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.