$1$

## On toric varieties of high arithmetical rank

### November 27, 2006

Abstract
We describe a class of toric varieties in the $N$  -dimensional affine space which are minimally defined by $N-2$  binomial equations.
Introduction The arithmetical rank (ara) of an algebraic variety is the minimum number of equations that are needed to define it set-theoretically. For every affine variety $V\subset {K}^{N}$  we have that codim $V\le$  ara $V\le N$  . This general upper bound was found by Eisenbud and Evans [7. In particular cases a better upper bound can be obtained by direct computations based on Hilbert's Nullstellensatz: this was done for certain toric varieties in [1, [2, [3, [4, [5. In all the examples treated there the arithmetical rank was close to the trivial lower bound, i.e., ara $V\le$  codim $V+1$  . In this paper we present a class of toric varieties whose arithmetical rank is close to the general upper bound, namely ara $V=N-2$  . For proving this result, of course, we need a more efficient lower bound:
this is provided by étale cohomology. The same kind of tools was used in [3.
There they were applied for showing that the arithmetical rank of certain toric varieties of codimension 2 depends on the characteristic of the ground field, and that ara $V=$  codim $V$  in exactly one positive characteristic. In the present paper, however, we study toric varieties of any codimension, and obtain the same arithmetical rank in all characteristics.

1 The main theorem

Let $K$  be an algebraically closed field, and let $n\ge 2$  be an integer. Let ${\mathbf{e}}_{1},...,{\mathbf{e}}_{n}$  be the standard basis of $\mathit{Z}{\mathit{Z}}^{n}$  . Set $N=2n-1$  and consider the following subset of $\mathit{I}{\mathit{N}}^{N}$  :
$T=\left\{{\mathbf{e}}_{1},...,{\mathbf{e}}_{n-1},d{\mathbf{e}}_{n},{a}_{1}{\mathbf{e}}_{1}+{\mathbf{e}}_{n},...,{a}_{n-1}{\mathbf{e}}_{n-1}+{\mathbf{e}}_{n}\right\},$  where $d,{a}_{1},...,{a}_{n-1}\in \mathit{I}{\mathit{N}}^{*}$  . We also suppose that $d$  is divisible by two distinct primes $p$  and $q$  . With $T$  we can associate the variety $V$  admitting the following parametrization $V:\left\{\begin{array}{ccc}{x}_{1}& =& {u}_{1}\\ & ...& \\ {x}_{n-1}& =& {u}_{n-1}\\ {x}_{n}& =& {u}_{n}^{d}\\ {y}_{1}& =& {u}_{1}^{{a}_{1}}{u}_{n}\\ & ...& \\ {y}_{n-1}& =& {u}_{n-1}^{{a}_{n-1}}{u}_{n}\end{array},$  which is a toric variety of codimension $n-1$  in the affine space ${K}^{N}$  . Our aim is to show the following
Theorem 1 ara $V=N-2$  .
This will be accomplished in two steps.

2 The defining equations

We show that ara $V\le N-2$  by exhibiting $N-2$  binomials which define $V$  set-theoretically.
Proposition 1 The variety $V$  is set-theoretically defined by the following binomials:
 $\begin{array}{cccc}{F}_{i}& =& {y}_{i}^{d}-{x}_{i}^{{a}_{i}d}{x}_{n}& \text{for}i=1,...,n-1\text{}\end{array}$
 $\begin{array}{cccc}{G}_{i}& =& {y}_{i}^{d-1}{y}_{i+1}-{x}_{i}^{{a}_{i}\left(d-1\right)}{x}_{i+1}^{{a}_{i+1}}{x}_{n}& \text{for}i=1,...,n-2\text{}\end{array}$
Proof .It is straightforward to check that for all $\mathbf{w}\in V$  , ${F}_{i}\left(\mathbf{w}\right)=0$  and ${G}_{i}\left(\mathbf{w}\right)=0$  for all indices $i$  . Conversely, we have to prove that for every $\mathbf{w}\in {K}^{N}$  such that
 $\begin{array}{ccc}{F}_{i}\left(\mathbf{w}\right)=0& \text{for}i=1,...,n-1& \end{array}$ (1)
 $\begin{array}{ccc}{G}_{i}\left(\mathbf{w}\right)=0& \text{for}i=1,...,n-2& \end{array}$ (2)
we have that $\mathbf{w}\in W$  . Let $\mathbf{w}=\left({\overline{x}}_{1},...,{\overline{x}}_{n},{\overline{y}}_{1},...,{\overline{y}}_{n-1}\right)\in {K}^{N}$  be a point fulfilling ( 1 ) and ( 2 ). We show that, for a suitable choice of parameters ${u}_{1},...,{u}_{n}$  , $\mathbf{w}=\left({u}_{1},...,{u}_{n-1},{u}_{n}^{d},{u}_{1}^{{a}_{1}}{u}_{n},...,{u}_{n-1}^{{a}_{n-1}}{u}_{n}\right)$  . This is certainly true if ${\overline{x}}_{n}=0$  : in this case ( 1 ) implies that ${\overline{y}}_{i}=0$  for $i=1,...,n-1$  , and we can take ${u}_{i}={\overline{x}}_{i}$  for all $i=1,...,n-1$  . Suppose that ${\overline{x}}_{n}\ne 0$  . Then, by ( 1 ), for all $i=1,...,n-1$  , we have ${\overline{x}}_{i}=0$  if and only if ${\overline{y}}_{i}=0$  . If $n=2$  and ${\overline{x}}_{1}=0$  , then take ${u}_{1}=0$  and set ${u}_{2}$  equal to any $d$  -th root of ${\overline{x}}_{2}$  . Otherwise, up to replacing $n$  with a smaller number, we may assume that ${\overline{x}}_{i}\ne 0$  for all $i=1,...,n-1$  .
Set ${u}_{i}={\overline{x}}_{i}$  for all $i=1,...,n-1$  and let ${u}_{n}$  be a $d$  -th root of ${\overline{x}}_{n}$  . By ( 1 ) we have that, for all $i=1,...,n-1$  , ${\overline{y}}_{i}^{d}={u}_{i}^{{a}_{i}d}{u}_{n}^{d},$  which implies that
 $\begin{array}{c}{\overline{y}}_{i}={u}_{i}^{{a}_{i}}{u}_{n}{\omega }_{i}\end{array}$ (3)
for some $d$  -th root ${\omega }_{i}$  of 1. On the other hand, from ( 2 ) we deduce that
 $\begin{array}{c}{\overline{y}}_{i}^{d-1}{\overline{y}}_{i+1}={u}_{i}^{{a}_{i}\left(d-1\right)}{u}_{i+1}^{{a}_{i+1}}{u}_{n}^{d}\end{array}$ (4)
Replacing ( 3 ) on the left-hand side of ( 4 ) gives
 $\begin{array}{c}{u}_{i}^{{a}_{i}\left(d-1\right)}{u}_{n}^{d-1}{\omega }_{i}^{d-1}{u}_{i+1}^{{a}_{i+1}}{u}_{n}{\omega }_{i+1}={u}_{i}^{{a}_{i}\left(d-1\right)}{u}_{i+1}^{{a}_{i+1}}{u}_{n}^{d},\end{array}$ (5)
and finally, being ${u}_{i},{u}_{i+1},{u}_{n}$  non zero, we have that, for all $i=1,...,n-2$  ,
 $\begin{array}{c}{\omega }_{i}^{d-1}{\omega }_{i+1}=1.\end{array}$ (6)
This implies that ${\omega }_{i}={\omega }_{i+1}$  for all $i=1,...,n-2$  , i.e., ${\omega }_{1},...,{\omega }_{n-1}$  are all equal to the same $d$  -th root $\omega$  of 1. Set ${u}_{n}^{\prime }=\omega {u}_{n}$  . Then ${\overline{x}}_{n}={u}_{{n}^{\prime }}^{d}$  and, by ( 3 ), ${\overline{y}}_{i}={u}_{i}^{{a}_{i}}{u}_{n}^{\prime }$  for $i=1,...,n-1$  .
Hence ${u}_{1},...,{u}_{n-1},{u}_{n}^{\prime }$  is the required set of parameters.

3 The lower bound

We show that ara $V\ge N-2$  by means of the following criterion, which is based on étale cohomology and is cited from [6, Lemma 3 ${}^{\prime }$  .
Lemma 1 Let $W\subset \stackrel{~}{W}$  be affine varieties. Let $d=dim\stackrel{~}{W}\W$  . If there are $s$  equations ${F}_{1},...,{F}_{s}$  such that $W=\stackrel{~}{W}\cap V\left({F}_{1},...,{F}_{s}\right)$  , then ${H}^{d+i}et\left(\stackrel{~}{W}\W,\mathit{Z}\mathit{Z}/r\mathit{Z}\mathit{Z}\right)=0\text{for all}i\ge s$  and for all $r\in \mathit{Z}\mathit{Z}$  which are prime to char $K$  .
Hence our claim will follow once we have shown
Proposition 2 If char $K\ne p$  , then ${H}^{2N-3}et\left({K}^{N}\V,\mathit{Z}\mathit{Z}/p\mathit{Z}\mathit{Z}\right)\ne 0.$  If char $K\ne q$  , then ${H}^{2N-3}et\left({K}^{N}\V,\mathit{Z}\mathit{Z}/q\mathit{Z}\mathit{Z}\right)\ne 0.$
Proof .Let $r\in \left\{p,q\right\}$  , and suppose that char $K\ne r$  . In the sequel $Het$  and $Hc$  will denote étale cohomology and étale cohomology with compact support with respect to the coefficient group $\mathit{Z}\mathit{Z}/r\mathit{Z}\mathit{Z}$  : we shall omit the latter for the sake of simplicity. By Poincaré Duality (s. [8, Cor. 11.2, p. 276) we have
 $\begin{array}{c}{H}^{2N-3}et\left({K}^{N}\V\right)\simeq {H}^{3}c\left({K}^{N}\V\right)\end{array}$ (7)
Moreover, we have a long exact sequence of cohomology with compact support $\cdots \to {H}^{2}c\left({K}^{N}\right)\to {H}^{2}c\left(V\right)\to {H}^{3}c\left({K}^{N}\V\right)\to {H}^{3}c\left({K}^{N}\right)\to \cdots ,$  where ${H}^{2}c\left({K}^{N}\right)={H}^{3}c\left({K}^{N}\right)=0$  , since $N\ge 3$  . Hence
 $\begin{array}{c}{H}^{3}c\left({K}^{N}\V\right)\simeq {H}^{2}c\left(V\right)\end{array}$ (8)
By ( 7 ) and ( 8 ) it thus suffices to show that
 $\begin{array}{c}{H}^{2}c\left(V\right)\ne 0\end{array}$ (9)
On ${K}^{n}$  fix the coordinates ${u}_{1},...,{u}_{n}$  and let $X$  be the subvariety of ${K}^{n}$  defined by ${u}_{1}={u}_{2}=\cdots ={u}_{n-1}=0$  . Then $X$  is a 1-dimensional affine space over $K$  , on which we fix the coordinate ${u}_{n}$  . Consider the surjective map $\phi :{K}^{n}\to V$  $\left({u}_{1},...,{u}_{n}\right)↦\left({u}_{1},...,{u}_{n-1},{u}_{n}^{d},{u}_{1}^{{a}_{1}}{u}_{n},...,{u}_{n-1}^{{a}_{n-1}}{u}_{n}\right)$  and the restriction map $\stackrel{~}{\phi }:{K}^{n}\X\to V\\phi \left(X\right),$  which is a bijective morphism of affine schemes. For all $i=1,...,n-1$  let ${V}_{i}=\left\{\left({\overline{x}}_{1},...,{\overline{x}}_{n-1},{\overline{x}}_{n},{\overline{y}}_{1},...,{\overline{y}}_{n-1}\right)\in V|{\overline{x}}_{i}\ne 0\right\}$  . These sets form an open cover of $V\\phi \left(X\right)$  , and ${U}_{i}={\phi }^{-1}\left({V}_{i}\right)=\left\{\left({u}_{1},...,{u}_{n}\right)\in {K}^{n}|{u}_{i}\ne 0\right\}$  . Moreover, for all $i=1...,n-1$  , the morphism ${\psi }_{i}:{V}_{i}\to {U}_{i}$  $\left({\overline{x}}_{1},...,{\overline{x}}_{n-1},{\overline{x}}_{n},{\overline{y}}_{1},...,{\overline{y}}_{n-1}\right)↦\left({\overline{x}}_{1},...,{\overline{x}}_{n-1},\frac{{\overline{y}}_{i}}{{\overline{x}}_{i}^{{a}_{i}}}\right)$  is the inverse map of the restriction of $\stackrel{~}{\phi }$  to ${U}_{i}$  . Hence $\stackrel{~}{\phi }$  is an isomorphism of affine schemes, so that it induces an isomorphism of groups ${\stackrel{~}{\phi }}_{i}^{*}:{H}^{i}c\left(V\\phi \left(X\right)\right)\simeq {H}^{i}c\left({K}^{n}\X\right)$  for all indices $i$  . The restriction of $\phi$  to $X$  $\overline{\phi }:X\to \phi \left(X\right)$  maps ${u}_{n}$  to $\left(0,...,0,{u}_{n}^{d},...,0\right)$  . Hence $\phi \left(X\right)$  is a 1-dimensional affine space and $\overline{\phi }$  induces multiplication by $d$  in cohomology with compact support. Now, as it is well-known, ${H}^{i}c\left(\phi \left(X\right)\right)\simeq {H}^{i}c\left(X\right)\simeq \left\{\begin{array}{cc}\mathit{Z}\mathit{Z}/r\mathit{Z}\mathit{Z}& \text{for}i=2\\ 0& \text{else}\end{array}$  Since $r$  divides $d$  , it follows that the induced maps ${\overline{\phi }}_{i}^{*}:{H}^{i}c\left(\phi \left(X\right)\right)⟶{H}^{i}c\left(X\right)$  are all equal to the zero map. Thus $\phi$  gives rise to the following morphism of acyclic complexes:
 $\begin{array}{cccccc}& & \mathit{Z}\mathit{Z}/r\mathit{Z}\mathit{Z}& & & \end{array}$
 $\begin{array}{cccccc}& & \wr |& f& & \end{array}$
 $\begin{array}{ccccc}{H}^{2}c\left(V\right)& \to & {H}^{2}c\left(\phi \left(X\right)\right)& \to & {H}^{3}c\left(V\\phi \left(X\right)\right)\end{array}$
 $\begin{array}{cccccc}& & & & & \end{array}$
 $\begin{array}{ccccc}↓& & {\overline{\phi }}_{2}^{*}↓0& & \wr |↓{\stackrel{~}{\phi }}_{3}^{*}\end{array}$
 $\begin{array}{cccccc}& & & & & \end{array}$
 $\begin{array}{ccccc}{H}^{2}c\left({K}^{n}\right)& \to & {H}^{2}c\left(X\right)& \to & {H}^{3}c\left({K}^{n}\X\right)\end{array}$
 $\begin{array}{cccccc}& & \wr |& g& & \end{array}$
 $\begin{array}{cccccc}& & \mathit{Z}\mathit{Z}/r\mathit{Z}\mathit{Z}& & & \end{array}$
Note that ${\stackrel{~}{\phi }}_{3}^{*}f=g0=0$  is the zero map, so that ${\stackrel{~}{\phi }}_{3}^{*}f$  is not injective. Since ${\stackrel{~}{\phi }}_{3}^{*}$  is an isomorphism, it follows that $f$  is not injective. This implies ( 9 ) and completes the proof. References

1. M. Barile, Almost set-theoretic complete intersections in characteristic zero. Preprint (2005). arXiv:math.AG/0504052
2. M. Barile, A note on Veronese varieties. Preprint (2005). arXiv:math.AG/0504055
3. M. Barile, G. Lyubeznik, Set-theoretic complete intersections in characteristic $p$  . Preprint (2004). To appear in: Proceedings of the American Mathematical Society.
4. M. Barile, M. Morales, A. Thoma, On Simplicial Toric Varieties Which Are Set-Theoretic Complete Intersections, J. Algebra 226 (2000), 880–892.
5. M. Barile, M. Morales, A. Thoma, Set-Theoretic Complete Intersections on Binomials, Proc. AMS 130 (2002), 1893–1903.
6. W. Bruns, R. Schwänzl, The number of equations defining a determinantal variety, Bull. London Math. Soc. 22 (1990), 439–445.
7. D. Eisenbud, E.G. Evans, jr. Every Algebraic Set in $n$  -Space is the Intersection of $n$  Hypersurfaces, Inv. Math. 19 (1973), 107–112.
8. J.S. Milne, “Étale cohomology”, Princeton University Press, Princeton, 1980.