2000 Mathematics Subject Classification. 14J28, 14J50, 20E05.
<ph f="cmbx">Free subgroup of automorphisms of a singular K3 surface</ph>

### Keiji Oguiso

Dedicated to Professor Yukihiko Namikawa on the occasion of his sixtieth birthday
Graduate School of Mathematical Sciences, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8914, Japan, and Korea Institute for Advanced Study, 207-43 Cheonryangni-2dong, Dongdaemun-gu, Seoul 130-722, Korea E-mail address : oguiso@ms.u-tokyo.ac.jp
• Abstract. Using results in linear algebraic groups, we show that the automorphism group of a complex K3 surface of maximum Picard number $20$  contains a free group of rank two. This is a generalization of a famous result of Shioda and Inose in a non-commutative direction.

1 Introduction Background and main results

Our main results are Theorems (1.1), (2.5) and Corollaries (1.2), (2.6).
0. Throughout this note, we work over the complex number field $\mathbf{C}$  . By a K3 surface we mean a compact simply-connected $2$  -dimensional complex manifold $M$  having nowhere vanishing holomorphic $2$  -form ${\sigma }_{M}$  . We denote by $T\left(M\right)$  , the transcendental lattice of $M$  , which is the minimal primitive sublattice of ${H}^{2}\left(M,\mathbf{Z}\right)$  , whose $\mathbf{C}$  -linear extension contains the class $\left[{\sigma }_{M}\right]$  . By $\rho \left(M\right)$  , we denote the Picard number of $M$  , i.e. the rank of the Néron-Severi group $NS\left(M\right)$  . It is well-known that $0\le \rho \left(M\right)\le 20$  and $M$  is projective if $\rho \left(M\right)=20$  . Following [SI], we call a (smooth) K3 surface with maximum Picard number $20$  , a singular K3 surface. A hyperkähler manifold is a compact simply-connected Kähler manifold having unique, up to scalar multiple, everywhere nondegenerate holomorphic $2$  -form. By [Su], K3 surfaces are nothing but $2$  -dimensional hyperkähler manifolds. For more details about K3 surfaces, see for instance [BPV].
1. In his paper [Mc], McMullen constructed a (necessarily non-projective) K3 surface of $\rho \left(M\right)=0$  , having a very impressive automorphism of infinite order.
Inspired by his work, we have studied the bimeromorphic automorphism group $Bir\left(X\right)$  of a non-projective hyperkähler manifold $X$  ([Og2]). It turns out that it is always almost abelian of rank at most $\rho \left(X\right)-1$  (so that finitely generated as well).
As a special case, the same is true for $Aut\left(M\right)$  of a non-projective K3 surface $M$  .
So, $Aut\left(M\right)$  is quite modest as an abstract group when $M$  is a non-projective K3 surface.
2. In their famous paper [SI], Shioda and Inose have shown, among other things, that the automorphism group of a singular K3 surface is an infinite group. They actually found a Jacobian fibration of positive Mordell-Weil rank. (See [Sh] for basic properties of Mordell-Weil groups.) Later, Vinberg [Vi] calculated the full automorphism groups of the so-called most algebraic K3 surfaces ${X}_{3}$  and ${X}_{4}$  . Both groups are highly non-commutative.
3. By Sterk [St], $Aut\left(M\right)$  is also finitely generated when $M$  is a projective K3 surface. So, in principle, it is possible to have reasonable descriptions of $Aut\left(M\right)$  in terms of generators. But, in practice, it is, in general, very hard to find them. For instance, no explicit description of the automorphism group of the Fermat quartic K3 surface, a singular K3 surface with discriminant $64$  , seems known (cf. [KK]). In somewhat different view, [Og1] observes that, in a family of projective K3 surfaces ${M}_{t}$  ( $t\in \Delta$  ), the behaviour of $Aut\left({M}_{t}\right)$  is rather complicated at which $\rho \left({M}_{t}\right)$  jumps up. These known and unknown results and a series of important work of Nikulin (see [Ni] and references therein) suggest some complexity of the automorphism group of a projective K3 surface of large Picard number.
4. The aim of this note is to examplify the last statement by showing the following1 :
Theorem 1.1. (Crescent Theorem) Let $M$  be a K3 surface having at least two Jacobian fibrations of positive Mordell-Weil rank. Then $Aut\left(M\right)$  contains a free group of rank $2$  .
Here a free group of rank $2$  is a group isomorphic to the free product $\mathbf{Z}*\mathbf{Z}$  , i.e.
the smallest non-abelian free group.
By [Og3], a singular K3 surface always admits at least two Jacobian fibrations of positive Mordell-Weil rank. So, as a special case of Theorem (1.1), we obtain the following generalization of a result of Shioda and Inose [SI]:
Corollary 1.2. The automorphism group of a singular K3 surface contains a free group of rank $2$  .
We also note that there is a projective K3 surface $M$  s.t. $\rho \left(M\right)=\rho$  and $|Aut\left(M\right)|<\infty$  for each integer $\rho$  with $1\le \rho \le 19$  ([Kn], [Ni]).
5. Our proof is based on two famous, deep results in linear algebraic groups; Lie-Kolchin Theorem and Tits Theorem ([Hm], [Ti]; see also Section 2 for the statements). Unfortunately, our proof does not tell us much about geometrical reason why non-abelian free groups should be in automorphism groups. It would be interesting to find a more ”visible” proof. For this, an observation of Cantat [Ca] might give us some hint.
Acknowledgement. An idea of this note has been grown up during my stay at KIAS March 2005. I would like to express my thanks to Professor Y. Kawamata for his valuable discussion and to Professors J.H. Keum and D.-Q. Zhang for their interest in this work. I would like to express my thanks to Professors J. Hwang and B. Kim for invitation.

1 This Theorem is inspired by a discussion with Y. Kawamata about remarkable difference of the appearance of crescent in Japan and Korea; it looks vertical in Japan while horizontal in Korea. This reminded me of the situation in Theorem (1.1).

2 Proof of Crescent Theorem

In the first paragrah 1, we recall Lie-Kolchin Theorem and Tits Theorem in linear algebraic groups. Both are very important in our proof. We fix some notation about lattices in 2. Then, applying Lie-Kolchin Theorem and Tits Theorem, we show Theorem (2.5) and its Corollary (2.6) in 3. These are the technical heart of our proof. Using Corollary (2.6), we complete the proof of Theorem (1.1) in 4.
1. For simplicity, we shall work over $\mathbf{C}$  . Let $V\ne \left\{0\right\}$  be a finite dimensional $\mathbf{C}$  -vector space. We regard the general linear group $GL\left(V\right)$  as an algebraic group defined over $\mathbf{C}$  , with Zariski topology. We identify $GL\left(V\right)$  with the group $GL\left(V\right)\left(\mathbf{C}\right)$  of $\mathbf{C}$  -valued points in a usual way. A subgroup of $GL\left(V\right)=GL\left(V\right)\left(\mathbf{C}\right)$  means a subgroup as an abstract group. A linear group is a subgroup of $GL\left(V\right)$  for some $V$  . If $G$  is a subgroup of $GL\left(V\right)$  , then both its Zariski closure $\overline{G}$  and the identity component of $\overline{G}$  are algebraic subgroups of $GL\left(V\right)$  (see eg. [Hm]). The next Lemma should be very well-known:
Lemma 2.1. Let $G$  be a solvable subgroup of $GL\left(V\right)$  . Then:
• (1) Any subgroup of $G$  is solvable.
• (2) $\overline{G}$  is also solvable.
• Proof. We only show (2). It suffices to check that $\overline{\left[G,G\right]}=\left[\overline{G},\overline{G}\right]$  . Note that $\left[\overline{G},\overline{G}\right]$  is closed in $\overline{G}$  (See for instance [Hm, 17.2]). Thus $\overline{\left[G,G\right]}\subset \left[\overline{G},\overline{G}\right]$  .
Let us show the other inclusion. Take $g\in G$  . Let us define the map ${\alpha }_{g}$  by ${\alpha }_{g}:\overline{G}⟶\overline{G};f↦{f}^{-1}{g}^{-1}fg.$  Clearly, ${\alpha }_{g}$  is continuous and satisfies ${\alpha }_{g}\left(G\right)\subset \left[G,G\right]$  . Thus ${\alpha }_{g}\left(\overline{G}\right)\subset \overline{\left[G,G\right]}$  .
Hence $\left[\overline{G},G\right]\subset \overline{\left[G,G\right]}$  . Next, take $f\in \overline{G}$  . Let us define the map ${\beta }_{f}$  by ${\beta }_{f}:\overline{G}⟶\overline{G};g↦{f}^{-1}{g}^{-1}fg.$  Clearly, ${\beta }_{f}$  is continuous and satisfies ${\beta }_{f}\left(G\right)\subset \left[\overline{G},G\right]$  . Since $\left[\overline{G},G\right]\subset \overline{\left[G,G\right]}$  , we have ${\beta }_{f}\left(G\right)\subset \overline{\left[G,G\right]}$  as well. Thus ${\beta }_{f}\left(\overline{G}\right)\subset \overline{\left[G,G\right]}$  and hence $\left[\overline{G},\overline{G}\right]\subset \overline{\left[G,G\right]}$  .
Lie-Kolchin Theorem and Tits Theorem are the following:
Theorem 2.2. (Lie-Kolchin Theorem, see eg. [Hm, Chap. VII, 17.6]) Let $G$  be a connected solvable subgroup of $GL\left(V\right)$  . Then $G$  has a common eigenvector in $V$  .
Theorem 2.3. (Tits Theorem [Ti]) Let $G$  be a linear group, say a subgroup of $GL\left(V\right)$  . Then $G$  is either almost solvable or contains a non-abelian free group.
Here a group $G$  is called almost solvable if $G$  contains a solvable subgroup of finite index. We also notice that a free group $G$  is abelian iff it is isomorphic to $\mathbf{Z}$  .
So, any non-abelian free group contains $\mathbf{Z}*\mathbf{Z}$  , i.e. the free group of rank $2$  .
2. By a lattice $L=\left(L,\left(*,**\right)\right)$  , we mean a pair consisting of a free abelian group $L\simeq {\mathbf{Z}}^{r}$  and its non-degenerate integral-valued symmetric bilinear form $\left(*,**\right):L×L⟶\mathbf{Z}$  . By ${L}_{K}$  , we denote the scalar extension $L{\otimes }_{\mathbf{Z}}K$  of $L$  by a field $K$  . The signature of $L$  is the pair of the numbers of positiveand negative-eigenvalues of a symmetric matrix associated to $\left(*,**\right)$  . We call $L$  hyperbolic (resp. elliptic) if the signature is $\left(1,r-1\right)$  (resp. $\left(0,r\right)$  ).
We call an element $v\in L\\left\{0\right\}$  primitive if the quotient group $L/〈v〉$  is torsion-free.
Let $M$  be a subgroup of $L$  . By ${M}_{L}^{\perp }$  , we denote the subgroup $\left\{v\in L|\left(v,w\right)=0\forall w\in M\right\}$  . Note that the restriction of $\left(*,**\right)$  on ${M}_{L}^{\perp }$  is non-degenerate iff so is on $M$  .
By $O\left(L\right)$  , we denote the group of isometries of $L$  . Note that $|O\left(L\right)|<\infty$  if $L$  is elliptic.
Let $L$  be a hyperbolic lattice. Then the set $\left\{v\in {L}_{\mathbf{R}}|\left({v}^{2}\right)>0\right\}$  consists of two connected components (w.r.t. Euclidean topology of ${L}_{\mathbf{R}}$  ). We choose and fix one of them and denote it by $\mathcal{P}\left(L\right)$  . We call $\mathcal{P}\left(L\right)$  the positive cone of $L$  . In general, there is no canonical way to choose the positive cone. When $L$  is the Néron-Severi lattice of a projective surface, we will choose the positive cone so that it contains ample classes.
Let $\overline{\mathcal{P}}\left(L\right)$  (resp. $\partial \mathcal{P}\left(L\right)$  ) be the closure (resp. the boundary) of the positive cone in ${L}_{\mathbf{R}}$  (w.r.t. Euclidean topology). By the Schwartz inequality, we have $\left(x,y\right)\ge 0$  for $x,y\in \overline{\mathcal{P}}\left(L\right)\\left\{0\right\}$  and the equality holds iff ${\mathbf{R}}_{>0}x={\mathbf{R}}_{>0}y\in \partial \mathcal{P}\left(L\right)$  .
Lemma 2.4. Let $L$  be a hyperbolic lattice of rank $r$  . Let $f\in O\left(L\right)$  .
Then:
• (1) If $f\left(v\right)=v$  for some $v\in \mathcal{P}\left(L\right)\cap L$  , then $ordf<\infty$  .
• (2) If $f\left(e\right)=e$  for some $e\in \partial P\left(L\right)\cap L\\left\{0\right\}$  , then there is a positive integer $m$  s.t. ${f}^{m}$  is unipotent, i.e. all the eigenvalues of ${f}^{m}$  are $1$  .
• Proof. These are implicit in [Og3]. For the convenience of the reader, we give a proof. Let us show (1). By $f\left(v\right)=v$  , one has an embedding $〈f〉\subset O\left(〈v{〉}_{L}^{\perp }\right)$  . Here the lattice $〈v{〉}_{L}^{\perp }$  is elliptic by $\left({v}^{2}\right)>0$  . Thus $ordf<\infty$  .
Next we show (2). We may assume that $e$  is primitive. Put $E:=〈e{〉}_{L}^{\perp }/〈e〉$  . Then the bilinear form of $L$  naturally descends to $E$  and makes $E$  an elliptic lattice of rank $r-2$  . Choose ${u}_{i}\in L$  ( $1\le i\le r-2$  ) s.t. $〈{\overline{u}}_{i}{〉}_{i=1}^{r-2}$  forms an integral basis of $E$  .
Then there is $w\in L$  s.t. $〈e,{u}_{i},w〉$  forms a rational basis of ${L}_{\mathbf{Q}}$  .
Since $f\left(e\right)=e$  , the isometry $f$  also descends to an isometry of $E$  , say $\overline{f}$  . Since $E$  is elliptic, one can find a positive integer $n$  s.t. ${\overline{f}}^{n}$  is identity on $E$  . Since $det{f}^{n}=±1$  (for $L$  being non-degenerate), we have $det{f}^{2n}=1$  . Set $g:={f}^{2n}$  .
Then one can write $g\left(e\right)=e$  , $g\left({u}_{i}\right)={u}_{i}+{a}_{i}e$  , and $g\left(w\right)=bw+ae+{\sum }_{i=1}^{r-2}{c}_{i}{u}_{i}$  by using some rational numbers $b$  , $a$  , ${a}_{i}$  , ${c}_{i}$  . Since $detg=1$  , we have $b=1$  . Thus $g={f}^{2n}$  is unipotent.
3. Let us now formulate our key results:
Theorem 2.5. Let $L$  be a hyperbolic lattice of rank $r$  . Take ${f}_{1},{f}_{2}\in O\left(L\right)$  and put $G:=〈{f}_{1},{f}_{2}〉$  . Assume that:
• (1) ${f}_{1}\left({e}_{1}\right)={e}_{1}$  and ${f}_{2}\left({e}_{2}\right)={e}_{2}$  for some ${e}_{1},{e}_{2}\in \partial \mathcal{P}\left(L\right)\cap L\\left\{0\right\}$  s.t.
${\mathbf{R}}_{>0}{e}_{1}\ne {\mathbf{R}}_{>0}{e}_{2}$  , and
• (2) $G$  is almost solvable.
Then either $ord{f}_{1}<\infty$  or $ord{f}_{2}<\infty$  holds.
• Proof. We shall proceed the proof by dividing into five steps.
Reduction Step 1. By the assumption (ii), $G$  has a solvable subgroup $N$  s.t.
$\left[G:N\right]<\infty$  . Then there is a positive integer ${m}_{1}$  s.t. ${f}_{i}^{{m}_{1}}\in N$  for both $i=1$  and $2$  . The group $〈{f}_{1}^{{m}_{1}},{f}_{2}^{{m}_{1}}〉\left(\subset N\right)$  is solvable by Lemma (2.1)(1). So, replacing $G$  by $〈{f}_{1}^{{m}_{1}},{f}_{2}^{{m}_{1}}〉$  and ${f}_{i}$  by ${f}_{i}^{{m}_{1}}$  , we may assume that $G$  itself is solvable. We will do so from now on.
Reduction Step 2. By Lemma (2.4)(1) and by the assumption (i), there is a positive integer ${m}_{2}$  s.t. both ${f}_{i}^{{m}_{2}}$  are unipotent. By replacing ${f}_{i}$  by ${f}_{i}^{{m}_{2}}$  and $G$  by it subgroup $〈{f}_{1}^{{m}_{2}},{f}_{2}^{{m}_{2}}〉$  , which is again solvable by Lemma (2.1)(1), we may furthermore assume that both ${f}_{i}$  are unipotent. We will do so from now on.
Setting Step 3. Put $V:={L}_{\mathbf{C}}$  . We have a natural embedding:
$G\subset O\left(L\right)\subset GL\left(V\right).$  Let $\overline{G}$  be the Zariski closure of $G$  in $GL\left(V\right)$  and $S$  be the identity component of $\overline{G}$  , i.e. the irreducible component containing the identity $1$  . Since $G$  is solvable, so is $\overline{G}$  by Lemma (2.1)(2). Thus, by Lemma (2.1)(1), $S$  is a connected solvable subgroup of $GL\left(V\right)$  . Since $\overline{G}$  is an algebraic subset of the noetherian space $GL\left(V\right)$  , it has only finitely many irreducible components. Thus, $\left[\overline{G}:S\right]<\infty$  . So, there is a positive integer ${m}_{3}$  s.t. ${f}_{i}^{{m}_{3}}\in S$  for both $i=1$  and $2$  . We put ${h}_{i}:={f}_{i}^{{m}_{3}}$  and $H:=〈{h}_{1},{h}_{2}〉$  . Now, it suffices to check that either $ord{h}_{1}<\infty$  or $ord{h}_{2}<\infty$  holds. This will be done in the next two steps.
Setting Step 4. Since $S$  is a connected solvable subgroup of $GL\left(V\right)$  , by applying Lie-Kolchin Theorem (2.2) for $S$  , we find a common eigenvector of $S$  , say $v\in V={L}_{\mathbf{C}}$  . Since ${h}_{i}\in S$  and since both ${h}_{i}$  are unipotent, we have ${h}_{1}\left(v\right)={h}_{2}\left(v\right)=v$  .
Since both ${h}_{i}$  are defined over $L$  , they are represented by integral matrices w.r.t.
an integral basis of $L$  . Thus the existence of $v\in {L}_{\mathbf{C}}\\left\{0\right\}$  with ${h}_{1}\left(v\right)={h}_{2}\left(v\right)=v$  implies the existence $u\in L\\left\{0\right\}$  s.t. ${h}_{1}\left(u\right)={h}_{2}\left(u\right)=u$  . This element $u$  will play an important role in the rest of proof. In apriori, this $u$  falls into one of the following four cases:
• (I) $\left({u}^{2}\right)>0$  ;
• (II) $\left({u}^{2}\right)=0$  ;
• (III) $\left({u}^{2}\right)<0$  and either $\left(u,{e}_{1}\right)\ne 0$  or $\left(u,{e}_{2}\right)\ne 0$  ;
• (IV) $\left({u}^{2}\right)<0$  and $\left(u,{e}_{1}\right)=\left(u,{e}_{2}\right)=0$  .
Step 5. We shall proceed the proof by dividing into four cases in Step 4. Crucial case is Case (IV).
Case (I). Replacing $u$  by $-u$  if necessary, we may assume $u\in \mathcal{P}\left(L\right)\cap L$  . Then, by Lemma (2.4)(1), $ord{h}_{i}<\infty$  for both $i=1$  , $2$  and we are done. Case (II). Replacing $u$  by $-u$  if necessary, we may assume $u\in \partial \mathcal{P}\left(L\right)\cap L\\left\{0\right\}$  .
Then, $\left(u,{e}_{1}\right)\ge 0$  and $\left(u,{e}_{2}\right)\ge 0$  . Moreover, at least one of these two inequalities has to be strict. This is because ${e}_{1}$  and ${e}_{2}$  are not proportional. Without loss of generality, we may assume that $\left(u,{e}_{1}\right)>0$  . Then $\left(u+{e}_{1}{\right)}^{2}=2\left(u,{e}_{1}\right)>0$  and ${h}_{1}\left(u+{e}_{1}\right)=u+{e}_{1}$  . Thus $ord{h}_{1}<\infty$  by Lemma (2.4)(1), and we are done.
Case (III). Without loss of generality, we may assume that $\left({u}^{2}\right)<0$  and $\left(u,{e}_{1}\right)\ne 0$  .
Then one has $\left(\left(u+m{e}_{1}{\right)}^{2}\right)={u}^{2}+2m\left(u,{e}_{1}\right)>0$  for some integer $m$  . One has also ${h}_{1}\left(u+m{e}_{1}\right)=u+m{e}_{1}$  . Thus $ord{h}_{1}<\infty$  by Lemma (2.4)(1), and we are done.
Case (IV). We shall show the result in Case (IV) by induction on $r$  . If $r=1$  , then the assumption (i) is false. If $r=2$  , then the case assumption (IV) is false. So, by the axiom of logic, the result in Case (IV) is true if $r\le 2$  .
Let $r\ge 3$  . Assume that the result holds for $r-1$  . Put $M:=〈u{〉}_{L}^{\perp }$  . Then $M$  is a hyperbolic lattice of rank $r-1$  . We have ${e}_{1},{e}_{2}\in M$  by the case assumption of (IV). Since ${h}_{i}\left(u\right)=u$  for both $i=1$  , $2$  , one has a natural embedding:
$H=〈{h}_{1},{h}_{2}〉\subset O\left(M\right)\subset GL\left({M}_{\mathbf{C}}\right).$  Since $H$  is solvable and since both ${h}_{i}$  are unipotent with ${h}_{i}\left({e}_{i}\right)={e}_{i}$  ( $i=1$  , $2$  ), one can repeat the argument in Steps 3 and 4 for $H\subset GL\left({M}_{\mathbf{C}}\right)$  . As a result, one finds a positive integer $n$  and an element $w\in M\\left\{0\right\}$  s.t. ${h}_{1}^{n}\left(w\right)={h}_{2}^{n}\left(w\right)=w$  .
This $w$  again falls into one of the four cases in Step 4. If $w$  is in Cases (I) (III), then by repeating the argument in Cases (I) (III) above, we obtain $ord{h}_{1}^{n}<\infty$  or $ord{h}_{2}^{n}<\infty$  , and we are done. If $w$  is in Case (IV), then we obtain $ord{h}_{1}^{n}<\infty$  or $ord{h}_{2}^{n}<\infty$  from the induction hypothesis, and we are done.
Corollary 2.6. Let $L$  be a hyperbolic lattice of rank $r$  . Take ${f}_{1},{f}_{2}\in O\left(L\right)$  and put $G:=〈{f}_{1},{f}_{2}〉$  . Assume that:
• (1) ${f}_{1}\left({e}_{1}\right)={e}_{1}$  and ${f}_{2}\left({e}_{2}\right)={e}_{2}$  for some ${e}_{1},{e}_{2}\in \partial \mathcal{P}\left(L\right)\cap L\\left\{0\right\}$  s.t.
${\mathbf{R}}_{>0}{e}_{1}\ne {\mathbf{R}}_{>0}{e}_{2}$  , and
• (2) $ord{f}_{1}=\infty$  and $ord{f}_{2}=\infty$  .
Then $G$  contains a free group of rank $2$  .
• Proof. By the contraposition of Theorem (2.5), $G$  is not almost solvable. So, the result follows from Tits Theorem (2.3).
4. Let us retrun back to our main Theorem. Proof of Theorem (1.1). Let $M$  be a K3 surface admitting at least two Jacobian fibrations of positive Mordell-Weil rank. We denote two of them by ${\phi }_{i}:M⟶{\mathbf{P}}^{1}$  ( $i=1$  , $2$  ) and the Mordell-Weil group of ${\phi }_{i}$  by $MW\left({\phi }_{i}\right)$  . Choose ${f}_{i}\in MW\left({\phi }_{i}\right)$  s.t. $ord\left({f}_{i}\right)=\infty$  . We naturally regard both ${f}_{i}$  as elements of $Aut\left(M\right)$  . Consider the subgroup $G:=〈{f}_{1},{f}_{2}〉$  of $Aut\left(M\right)$  , generated by ${f}_{1}$  and ${f}_{2}$  . We have a natural homomorphism: ${r}_{NS}:G⟶O\left(NS\left(M\right)\right);{g}^{-1}↦{g}^{*}|NS\left(M\right).$  Here $NS\left(M\right)$  is hyperbolic. This is because $M$  is projective by Kodaira's criterion (see eg. [BPV]). Let ${e}_{i}\in NS\left(M\right)$  be the class of a general fiber of ${\phi }_{i}$  . Then one has ${e}_{1},{e}_{2}\in \partial \mathcal{P}\left(NS\left(M\right)\right)\cap NS\left(M\right)\\left\{0\right\}$  and ${\mathbf{R}}_{>0}{e}_{1}\ne {\mathbf{R}}_{>0}{e}_{2}$  . By definition of ${e}_{i}$  , one has also ${f}_{1}^{*}\left({e}_{1}\right)={e}_{1}$  and ${f}_{2}^{*}\left({e}_{2}\right)={e}_{2}$  .
Claim 2.7. The homomorphism ${r}_{NS}$  is injective.
• Proof. We have a natural homomorphism:
${r}_{NS}×{r}_{T}:G⟶O\left(NS\left(M\right)\right)×O\left(T\left(M\right)\right).$  By the global Torelli Theorem for projective K3 surfaces ([PSS], [BPV]), this homomorphism is injective. Since ${f}_{i}$  acts trivially on ${\sigma }_{M}$  , one has ${f}_{i}^{*}|T\left(M\right)=id$  as well. Thus ${r}_{T}\left(G\right)=id$  . Hence ${r}_{NS}$  is injective.
By Claim (2.7), we have $ord{f}_{1}^{*}|NS\left(M\right)=\infty$  and $ord{f}_{2}^{*}|NS\left(M\right)=\infty$  . Thus ${r}_{NS}\left(G\right)$  contains a free subgroup of rank $2$  by Corollary (2.6). Since $G\simeq {r}_{NS}\left(G\right)$  , this implies the result. Q.E.D.
Remark 2.8. Let $M$  be a singular K3 surface and $m$  be a positive integer. Then the Hilbert scheme $Hil{b}^{m}\left(M\right)$  of $m$  points on $M$  is a $2m$  -dimensional projective hyperkähler manifold ([Be]). Note that $Aut\left(M\right)$  acts faithfully on $Hil{b}^{m}\left(M\right)$  in a natural way. Thus by Corollary (1.2), $Aut\left(Hil{b}^{m}\left(M\right)\right)$  , and hence $Bir\left(Hil{b}^{m}\left(M\right)\right)$  , also contains a free group of rank $2$  .
References

1. W. Barth, C. Peters, A. Van de Ven, Compact complex surfaces, Springer-Verlag (1984).
2. A. Beauville, Variété Kähleriennes dont la premiére classe de chern est nulle, J. Diff. Deom. 18 (1983) 755–782.
3. S. Cantat, Sur la dynamique du groupe d'automorphismes des surfaces $K3$  , Transform. Groups 6 (2001) 201–214.
4. J. E. Humphreys, Linear algebraic groups, Graduate Texts in Mathematics, 21 Springer-Verlag, New York-Heidelberg, 1975.
5. S. Kondo, Algebraic K3 surfaces with finite automorphism groups, Nagoya Math. J. 116 (1989) 1-15.
6. J.-H. Keum, and S. Kondo, The automorphism groups of Kummer surfaces associated with the product of two elliptic curves, Trans. Amer. Math. Soc. 354 (2002) 67–91.
7. C. T. McMullen, Dynamics on K3 surfaces: Salem numbers and Siegel disks, J. Reine Angew. Math. 545 (2002) 201–233.
8. V. V. Nikulin, Discrete reflection groups in Lobachevsky spaces and algebraic surfaces: in Proceedings of the International Congress of Mathematics (Berkley 1986) Amer. Math. Soc. (1987) 654-671.
9. K. Oguiso, Local families of $K3$  surfaces and applications, J. Algebraic Geom. 12 (2003) 405–433.
10. K. Oguiso, Bimeromorphic automorphism groups of non-projective hyperkähler manifolds a note inspired by C. T. McMullen, math.AG/0312515 (version 2), submitted.
11. K. Oguiso, Automorphisms of hyperkähler manifolds in the view of topological entropy, math.AG/0407476 (version 4), submitted.
12. I. Piatetski-Shapiro and I. R. Shafarevich, A Torelli Theorem for algebraic surfaces of type K3, Math. USSR Izv. 5 (1971) 547-587.
13. T. Shioda, Theory of Mordell-Weil lattices: in Proceedings of the International Congress of Mathematicians (Kyoto 1990) Math. Soc. Japan (1991) 473 489.
14. T. Shioda and H. Inose, On singular K3 surfaces, In: Complex analysis and algebraic geometry, Iwanami Shoten (1977) 119–136.
15. Y. T. Siu, Every K3 surface is Kähler, Invent. Math. 73 (1983) 139–150.
16. H. Sterk, Finiteness results for algebraic K3 surfaces, Math. Z. 189 (1985) 507-513.
17. J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972) 250–270.
18. E.B. Vinberg, The two most algebraic K3 surfaces, Math. Ann. 265 (1983) 1–21.

Graduate School of Mathematical Sciences, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8914, Japan, and Korea Institute for Advanced Study, 207-43 Cheonryangni-2dong, Dongdaemun-gu, Seoul 130-722, Korea E-mail address : oguiso@ms.u-tokyo.ac.jp