Inve, Math, 25, 105-121 1960 Inventiones mathematicae Springer Verlag 1984 On K3 surfaces with large Picard number DR. Morrison* Department of Mathematics, Princeton University, Fine Hall Box 37, Princeton, NJ 08544, USA A K3 surface is a simply connected compact complex manifold of dimension two with a nowhere-vanishing holomorphic 2-form. K3 surfaces have received much attention in the last 25 years, both because of the important place they occupy in the classification of compact complex surfaces [4], and because the “period map” for K3 surfaces is quite well-behaved (there is a “global Torelli theorem” [15, 1, 17, 7], and a “surjectivity of the period map” theorem [5, 21, 6, 19, 9}) Among classical examples of K3 surfaces are the “Kummer sur- faces”, which play @ crucial role in the theory, and which are defined as follows. Let Z be a complex torus of dimension two, 1 be an involution on Z induced by multiplication by ~1 on the universal cover €?, and ¥ be the minimal resolution of singularities of Zji. Then Y is a Kummer surface; all Kummer surfaces are K3 surfaces The Picard number of a K3 surface is the rank of its group of line bundles; this rank ranges from 0 to 20. If X is a K3 surface with Picard number 20, then Shioda and Inose [18] have constructed an involution 1 on X such that the quotient X/1 is birational to. a Kummer surface, This gives rise to a diagram in which the dotted arrows are rational maps of degree 2, X and Y are K3 surfaces, and Z is a complex torus. Shioda and Inose further show that this diagram induces an isomorphism of integral Hodge structures on the transcen- dental lattices of X and Z. The main result of this paper is a generalization of the construction of Shioda and Inose, We show that there is a diagram analogous to theits for any algebraic K3 surface of Picard number 19 or 20, and give precise conditions ‘National Science Foundation Postdoctoral Fellow106 DR Morrison for the existence of such a diagram when the Picard number is 17 or 18, (Such diagram cannot exist for algebraic K3 surfaces of Picard number less than 17) This work was prompted by = remark of Takayuki Oda [13], who ‘conjectured that analogues of the Sbiods-Inose construction should exist when the Picard number is 17, 18, oF 19. ‘The plan of the paper is as follows: Seotion 1 reviews definitions and results about Hodge theory, complex tori, and K3 surfaces, Section 2 summarizes the ‘work of Nikulin [12] oa embeddings of quadratic forms, and draws some consequences for K} surfaces and complex tori. Sections 3, 4, and 5 discuss involutions on K3 surfaces and complex tori, relying heavily on two other papers of Nikulin [10, 11]. In Sect. 6, we generalize the Shioda-Inose con- struction, and in Sect. 7 we discuss Oda’s conjecture. L. Hodge structures, complex tori, and K3 surfaces Definition 1.1. A lattice is a free Z-module of finite rank equipped with a Z- valued symmetric bilinear form b(x,y}. If Ly and L, are two lattices, then L,@L, denotes the orthogonal direct surn of L, and Lp; 17 denotes L=LOL®...OL (factors), If Lis a lattice and m is an integer, then L(m) denotes the same Z-module with ‘a form which has been altered by multiplication by m, that is, Byam I= Mlb 8 YD) ‘An isomorphism of lattices preserving the bilinear form is called an isometry: hote that 1 is not isometrie to Lon) when |m|>1. The group of sel-isometries of a lattice Lis denoted by O(L) ‘A lattice is even ifthe associated quadratic form takes on only even values, and is odd if the quadratic form takes on some odd value. The discriminant of a lattice L, written discr(L), is the determinant of the matrix of its bilinear form. A lattice is non-degenerate if its discriminant is non-zero, and unimodular if its Giscriminant is +1. If Lis 2 non-degenerate lattice, the signature of Lis a pair Gin 5.) Where s,., denotes the multiplicity of the eigenvalue +1 for the ‘quadratic form on LOR. (Note that with this definition, the statement that a given lattice has a signature (5,,,.%)..) automatically implies that L is non- degenerate) A lattice is indefinive if the associated quadratic form takes on both positive and negative values; in the non-degenerate case, this is true if and only if mints, 5. -)>0- Examples 1.2. ()) U denotes the hyperbolic plane, that is, U is a free Z-module of rank 2 whose bilinear form has matrix o 1 (0): This is clearly an even lattice; note that U(—m)= U(m) for any m. (i) E, denotes the unique even unimodular positive definite lattice of rank 8; the bilinear form on F, is given by the matrix(On K3 surfaces with large Picard number 107 2-1 -102-1 -1 2-1-1 -1 2.0 -100 2-1 eee ed (ii) ¢K) denotes the lattice of rank 1 such that b(x,x)=k for any generator x of (hp. “Theorem 1.3 (Milnor (8D, Let L be an indefinite unimodular tatice, If Lis od, se Ledyre(-D* for some m and n. If Lis even, then LeU"@E( +1" for some m and n. In particular, the signature and parity of L determine L up 10 If X is a compact Kahler surface with the property that H?(X,Z) is torsion-free, then the intersection pairing gives H°(X,Z) the structure of a lattice. This lattice is unimodular by Poincaré duality, so that H7(X,Z) is determined by its signature and parity. The Hodge index theorem [2] says that the signature of the lattice H*(X,2) is (2h=°+1, h'—1), where hid =dim HX). Definition 14. Let Le a lattice. A Hodge structure of weight 2 on L consists of a “Hodge decomposition” L@C=L@L OL such that L>=1°?, and L} and such that =" (where ~~ denotes complex conjugation), bx,9>0 for 04 xeL**, Dx,y)=0 for x,yeL*, and bix,y)=0 if xeL*°@ 19? and yeLh! A Hodge isometry is an isometry $: LyL, between lattices with Hodge strvetutes which preserves the Hodge decompositions. A signed Hodge structure (of weight two) on a lattice L consists of a Hodge structure on L such that the quadratic form restricted to L''m(LOR) has signature (1, n—1), together with a choice of one of the (two) components of {x€L!! (LDR) :b(x,x)> 0}. 0108 DR Morrison AA signed Hodge isomerry is a Hodge isometry between two lattices with signed Hodge structures which preserves the choice of component of (4) A polarized Hodge structure (of weight two) is a Hodge structure with the property that the quadratic form is negative-definite when restricted to L*n(L@R), Let X be a compact Kilhler surface such that H*(X,Z) is torsion-free. The lattice H(X,2) has a natural signed Hodge structure of weight two: we take the usual Hodge decomposition HX, EH» (XOH(XYOHO*X) together with the component of {xeH™(X,R):6(4,3)> 0} which contains the cohomology class of any Kahler metric. The Hodge dex theorem [2] guarantees that the signature of the form on H™1(X,R) is (Ah), Let NS(X) be the Néron-Severi group of X, that is, the group of line bundles on X, modulo those algebraically equivalent to zero. NS(X) has a natural embedding in H°(X,2), and can be identified with H2(X,Z)0H™"(X), giving NS(X) the structure of a lattice. The Picard mumber of X, denoted by p(X), is the rank of NS(X). The transcendental lattice of X, denoted by Tj, is the orthogonal complement of NS(X) in H(X,Z). The lattice Ty inherits a Hodge structure from the one on H*(X, 2). Ifthe Hodge structure on Ty is polarized, then its signature must be (242, h')—p+1) so that the signature of the lattice NS(X) is (1, p—1). But then 'NS(X) contains an element of positive square-length; by a theorem of Kodaira [4; Theorem 8], this implies that X is algebraic. Conversely, if X is algebraic, then the signature of NS(X) is (1, p—1), which implies that the Hodge struc- ture on Ty is polarized. Let ruop(X) denote the topological Euler characteristic of X. ‘Theorem-Definition LS (Kodaira [4; Sect. 6]). Let X be a compact Kahler surface with trivial canonical bundle. Then h*°(X)=1, and either ( X=€7/L is a complex torus of (complex) dimension 2; in this case, W(X)e2 and YoqIX)=0, OF (i) X is a KF surface, that is, h*%(X)=0 and y49(X)=24. (in fact, a K3 surface can be defined 25 a compact complex surface with trivial canonical bundle such that h?°(X)=0, but Sin [20] has recently shown that every K3 surface is Kabler.) If X is a complex torus, then itis easy to see directly that H*(X,Z)=U. In particular, H°(X,Z) is torsion-free, so. that H*(X,Z) and T, carry natural Hodge structures. X is algebraic when Ty is polarized; in this case, we call X an abelian surface If X is a K3 surface, then HX, Z) has no torsion (16; Chap. IX, Sect. 3} Thus, H2(X,Z) and Ty carry natural Hodge structures. Moreover, a com- putation involving the Wu formula [22] shows that H?(X,2) is an even lattice‘Gn K3 surfaces with large Picard number 199 {ck [8] oF (16). The signature of this lattice is (3,19) by the Hodge index heorem, so Theorem 1.3 implies that H*(X,Z) is isometric to the K3 lattice A =U2@E,(-1). For complex tori and K3 surfaces, the following results go by the name “the surjectivity of the period mapping” ‘Theorem 1.6 (Sbioda (17). Given a signed Hodge structure on U?, there exists a complex torus X of dimension two and a signed Hodge isometry UX DAU (with respect to the given signed Hodge structure) ‘Theorem 1.7 ({5, 21, 6, 19, 9). Given a signed Hodge structure on the K3 lattice ‘A, there exists a K3 surface X and a signed Hodge isometry 6: HAUX,Z)A (with respect t0 the given signed Hodge structure). Definition 18. An embedding M<-L of latices is primitive if L/M is free. Two primitive embeddings Mo+L, M=+L: are isomorphic if there is an isometry LE which induces the identity map on M. Corollary 1.9. Let A be the K3 lattice i) Suppose S“+U® (resp. SA) is a primitive sublattice of signature (dp 1). Then there exists an abelian surface (resp. algebraic K 3 surface) X and an isometry NS(X}>S. iii) Suppose T=+U5 (resp. Te+A) is a primitive sublattice of ‘signature (2,4-~p) (resp. (2,20-—p)}. Then there exists an abelian surface (resp. algebraic K 3 surface) X and an isometry TT. Proof. Let Ldenote U? (resp. 4), and let b(x,y) denote the bilinear form on L. (i) Choose a subspace FL=S, and bly has signature 1, 3) (esp. signature (1, 19)}. Pick some non-zero @eL@C such that 1X and bfex.@)=0. Define P8-Co; Li? =2@C; L*=Co. ‘Then L@C=L2°OL' OL is a Hodge decomposition; choosing either com- ponent of {xeL'! o(LOR):b(,2)>9), ‘makes this into a signed Hodge structure. By Theorems 1,6 and 1.7, there is @ Complex torus (resp. K3 surface) X and a signed Hodge isometry @:K,DAL But now glysin Bives an isometry of NSLX) with LM ab=EoL=s.0 DR. Mortison Since NS(X) has signature (1, p—1), X is algebraic, (ii) Let $ be the orthogonal complement of T in L, and apply part (i): we get Ty2NSXPSS4= 7. QED. 2. Diseriminant-forms and embeddings of lattices, Definition 2.1. Let A be a finite abelian group, The length of A, denoted I(A), is the minimum number of generators of 4. A quadratic form on 4 is'a map, + Q/22, together with a symmetrie bilinear form BAK ASQIE such that 1) g(na)=n*q(a) for all neZ and aca. 2) gla-+a')~qla)—gla}=20(a,a’\mod2Z), Note that if q is a quadratic form on A, then so is ~g (with bilinear form —b), Mf Lis a non-degenerate even lattice, then there is a natural embedding jom(L, Z). The (Q-valued) quadratic form on L¥ induces a quadratic form ay, called the discrminant-form of L, on the finite abelian group A,=L*/1. ‘Notice that fu.» —4,, and that dy o4%9,S4y, Given a non-degenerate even lattice L, We attach as invariants (6,,».5,» duh Where (%.4),5)-.) i8 the signature of L ‘The usefulness of these invariants is shown by the following ‘Theorem 22 (Kneser [3], Nikulin [12; Corollary 1.133} Let L be an even lattice with imoariants (5.,), 5,-,4z)- Suppose that () OAusox(On K3 suriaes with large Picard number mn preserving the diseriminant-forms. Define = {le(M,OK)*:$(eLAM, OK). ‘Then there is an embedding M,coL. If meMoL, then d(me(M}OLV (M,@K) so that d(m)eM,, since M,<+L is primitive Thus, meM, so that the embedding M.-L is primitive. Since @ preserves the discriminant-form, q,%4,. Moreover, since M,@KeL and M,@KeL, L and L have the same signature. Thus, LEL. QED. Another easy argument yields the following. Lemma 24 (Nikulin [12; Proposition 1.6.1). Let M=+L be a primitive embedi- ing of non-degenerate even lattices, and suppose that Lis unimodular. Then ws ~ 0 Conversely, if M, and M, are non-degenerate even lattices which satisfy qu, ~ ay. then there is a primitive embedding of M, into an even unimodular lattice L such that M}=M,. Corollary 25 (cf. (7; Theorem 24]}. Let T be a non-degenerate even lastice of rank r, Then there is a primitive embedding TU" Proof. By Lemma 24, since gr,_,)%—dy there is an even unimodular lattice L and a primitive embedding ‘TeoL such that T!%T(—1). But then L has signature (r) 80 that LU" by Theorem 13. QED. Recall that by Corollary 1.9(ii), the possible transcendental lattices of abe- lian surfaces are all primitive sublattices TU? of signature (2, 4~ 9). Corollary 26. Let T be an even lattice of signature (2, 4 @ fk or 1, then there is a primitive embedding TU. (i) I k=2, then there is a primitive embedding TU> if and only if T=ver. (ii) If, k=3, then there is a primitive embedding TU? if and only if T=U"@T" Proof. If k=0 or 1, then T has rank $3, so that 7+U* by Corollary 2.5. If k=2 and TOU? let S=T+. Then U@S(—1) has the same signature as T; by Lemma 24, it also has the same discriminant-form. Since I(4,)=UA9)S 4-2, by Theorem 22, T=U @S(~ 1), Conversely, if T=U @ T-, then by Corol- lary 25, THU" thus, TU If k23 and T+U%, let S= T+, Then U?@)S(—1) has the same signature as T; by Lemma 24, it also has the same discriminant-form. Since [(47)=I(49)S 32, by Theorem 2.2, T= U?@S(—1), Conversely, if TU? ® 1’, then by Cor- ollary 2.5, TU; thus, TOU". QED. We will need one furtherue DR. Morrison Corollary 2.7. Ler Ty be a sub-Q-lattice of U>@Q of signature (2,2), Then the quadratic form of Ty represents zero. Proof. Let T=TynU?. By Corollary 26(i), T= U@T". But U represents zero, hence Tg also represents zer0. QED. ‘The main result on embeddings of even lattices is ‘Theorem 28 (Nikulin [12; Theorem 1.144). Let M be an even lattice with invariants (4). dygh and let L be an even unimodular lattice of signature (.j05,-9) Suppose that tr0, let 2: X'~» ¥ be the rational quotient map corresponding to the double cover branched on T.C,, and let ReX be the points corresponding to C;, Then Hol XI Lag X~ (PY) + b= Dpegh V— (CM) += 204-24) 4k Now Ky=n"(Ky)=0, so that by Theorem 1.5, X is a complex torus or a K3 surface. In the first c88e, Zo,(X)=0 and k=165 in the second, fgg(X)=24 and k=8 QED. 4, Kummer surfaces Definition 4.1. Let Z be a complex torus of dimension 2, and let 1 be an involution on Z induced by multiplication by —1 on the universal cover €2. If zr: Z-»¥ is the rational quotient by 1, then ¥Y, which is a K3 surface, is called a(On K3 surfaces with large Picard number us Kummer surface. + has sixteen fixed points on Z, so that Y has sixteen excep- tional curves. Note that Z is an abelian surface if and only if ¥ is an algebraic K3 surface. ‘Theorem 42 (Nikulin [IOD. There exists an even, negative-definite rank 16 lattice K, called the Kummer lastce, withthe following properties () diser(k)=2* (i) 7 ¥ is @ Kummer surface, then the minimal primitive sublattice of HYY,2) containing the classes of the exceptional curves on ¥ is isomorphic to K. (ii) K admits a unique primitive embedding into the K3 lattice A x) A K3 surface ¥ surface ¥ is a Kummer surface if and only if there is 0 primitive embedding K-+NS(Y). ‘The Kummer lattice was first described by D.B, Fuks (cf. (15; appendix to Sect. Sp. Proposition 43. Let Y be a Kummer surface, Z be the corresponding complex torus, x: Z->¥ be the rational quotient map, and Ty (resp. T,) be the transcen- deraal latice of ¥ (resp. Z). Then (i (ef. Nikulin [10; Remark 2}) , induces a Hodge isometry Te2)2 Ty. Gi) de = Guy Proof, The Kummer involution acts as the identity on H°(Z,2)2U%; more- over, by (42), xg(H(2,2)'=K which has discriminant 2° by (4.2)( Statement (?) now follows immediately from Proposition 32, which also tes us that Kian (HIZ,Z)=UQ? so that 46% 42> Gea)" (ince du dud) QED Corollary 4.4. Let ¥ be an algebraic K3 surface. (i) If p(Y)=19 or 20, then Y is a Kummer surface if and only if there is an even lattice T’ with Ty= TQ, (i) If pt¥)=18, then V is a Kummer surface if and only if there is an even lattice T’ with T, = UQ\@T'(2). (iii) If p(Y)=17, then Y is a Kummer surface if and only if there is an even lattice T’ with T,xUQVOTQ) iv) If p(Y)<17, then ¥ is not a Kummer surface. Proof. By (42Miv), Y is a Kummer surface if and only if there is a primitive embedding KNS(¥). Since K admits a unique primitive embedding into A bby (42){ii), such a primitive embedding Ke+NS(V) exists if and only if there is a primitive embedding T,>K*= U(2)*. But then there is some even lattice T” with T,=7"Q) and a primitive embedding T’<+U?. The corollary now follows from Corollary2.6. QED. Note that when 175 p20, every even lattice of the appropriate signature ‘occurs as the transcendental lattice of a K3 surface, by Corollary 2.10. Corol-16 DR. Moctsoo lary 44 thus shows that Kummer surfaces are rather rare among K3 surfaces with such Picard numbers. 5. Nikulin invoh jons Definition 5.1. An involution 1 on a K3 surface X is a Nikulin involution if 1c) =a for every oe H®°C0), Lemma 5.2 (Nikulin [11; Sect. 5) (i) Every Nikulin involution has eight isolated ‘fixed points fn X-r¥ is the rational quotient by a Nikulin involution, then Y is a K3 surface. Definition 5.3. The Nikulin lattice is an even lattice N of rank 8 generated by {e,]h.: and d-=£3¢,, with the bilinear form induced by (cp -28 Lemma 54, (i) The discriminant of N is 2° ii) If X is a K3 surface with a Nikulin involution 1, and X-+Y is the rational quotient mop, then the minimal primitive sublattice of H°(%,Z) containing the ‘classes of the exceptional curves on Y is isomorphic to N. Proof. (i) If N’ is the sublattice of N spanned by {c,}, then diser(N’) [N:N"] =2, so that and discr(N') 56 ) 296 ENN? Gi) Let Cyyin€ be the exceptional divisors of Xvr¥. Since XY is the rational quotient by an involution, D=}EGeNS(¥) as well, so that NesNS(Y}=HY,2), To show that the embedding is primitive, suppose that C=Em,CeNS(Y) with meQ. Then (C, C)= —2m,e2, so that C=aY4C, (mod for some 1<{1,...,8), But by Lemma 33, #(1)=0 or 8; if #(0)=0, then CeN’, while if #(/)=8 then CusDmod N’ so that CeN. QED. Definition 55, Let X be a K3 surface. The Wey! group of X is the subgroup WUX)c AutH(X,Z) generated by reflections in all elements of NS(X) of square-length —2. The following theorem of Nikulin is a consequence of the global Torelli theorem for K3 surfaces ({15, 1,17, 7) ‘Theorem 56 (Nikulin [11; Theorems 43, 47, 4.15). Let X be a K3 surface, let GE=Z2Z be a subgroup of OH*(X,Z), and let Sg—(H*4X,Z)9). Suppose tharOn K3 surfaces with large Pieasd number a the lattice Sq is negative definice, (i) no element of Sg has square-length —2, and Gi) SceNSX) ‘Then there is a Nikulin involution 1 on X and an element weW/(X) such that iawen! where g is the generator of G. Asa consequence, we get the following ‘Theorem 5.7. Let X be a K3 surface such that E, = 1)74.NS(X). Then there is 4@ Nikulin incolution 1 on X such that if x: XY i the rational quotient map, () there isa primitive embedding N@Eq(—1)-oNS(Y), (i) rq induces a Hodge isometry Ty) I, (ii) N@Eg(~1) has diseriminant.form (gus) Proof Let {) ('=12; 15/8) be a basis of Es(—1)?=Fy(—1)@E,(—1) fuck that cj e£s(— OO, 76(0}@E4(—1), and for each fixed |, (¢) forms basis of Ea(—1) whose matrix isthe negative of that in’ (12). Let @:EA-IPHX,Z) be the embeddig and define an action of G=2/2Z on H"(X,2) as follows: the generator geG acts as BOACP= HE: LHC = Gh): B)=e, for all ee p((Eg(—1)?))4 (This is well-defined since the embedding @ is primitive, and Ey(—1 is ymimodular) So=(H*X,2)")* is generated by. {¢v(c})—Gle})}, 80. that See H(Es(—1)J=NS(X) and $2 Eq(—2). Since Ey is an even, positive de- finite lattice, Sq is a negative-definite lattice which contains no element of square-length—2. Thus, by Theorem 5.6, there is a Nikulin involution ton X and an element we W(X) such that 1*=wew"!. For xeE«(—1)2 let WE)= w(x). Then y: Ey(—IoNS(X)c H*(X,Z) is another primitive embedding, since W(X) preserves NS(X). Moreover, MUD =WE ME P=We}); rr for all ecwi(E.(—1)°)¥ Let : XY be the rational quotient map. The minimal primitive lattice containing the exceptional divisors spans a copy of N--NS(P) Moreover the classes x,(H(cD).....g¥((ch)) ate orthogonal to N. Now by the formules te Se (WUD, THC) = Han WOH), 2° x Ale) = HIE) +H WUE LYE +H YEE)) Hef.eh)+ 4c}, c2)us DR. Morrison since {c}] and {c7) have identical bilinear form matrices, But this means that {r,W(c})} also spans a copy of E4(—1), so that NBE,(—1)e+NS(Y), Since Ex(—1) is unimodular and N is primitively embedded, the embedding N@Eq(~1)}NS(Y) is also primitive Let L be the orthogonal complement of Y(Eg(—1)2) in H2(X,Z) Then LEU? Since L-oH*(X,Z)° and discr(N@E,)=2% by Proposition 32. m, in. duces a Hodge isometry 7;(2)% Tj, and n,(L) is a primitive sublattice of B¥Y,Z), isomorphic to U2). ‘We thus see that Ings =o)”. QED. woe, 6. Shioda-Inose structures Definition 6.1. A K3 surface X admits a Shioda-Inose structure if there is a Nikulin involution r on X with rational quotient map x: X-»Y such that Vis Kummer surface, and n, induces a Hodge isometry Ty(2 Remark 6.2. If X admits a Shioda-Inose structure, let Z be the complex torus whose Kummer surface is ¥. This gives a diagram of rational maps of degree 2. 73(2)% 7, by definition, and Ty2)=7, by Pro- Position 4.3. Thus, this diagram induces a Hodge isometry T, =, Theorem 63, Ler X be an algebraic K3 surface. Then the following are equiva lent: (i) X admits a Shioda-tnose structure. (ii) There exists an abelian surface A and a Hodge isometry Ty=T, ii) There is a primitive embedding Ty=—+U? Gv) There is an embedding E_(—1)PoNS(X). Proof. (ij-»(i) follows from Remark 62; the complex torus Z is an abelian surface because the Hodge structure T; = Ty is polarized (i)->Gi): I Ty%T%, the natural primitive embedding T,+H?(4,2) induces a primitive embedding Ty=+U5 (iipetivy: We extend the given primitive embedding : TU to an embedding 0: TU BE(—1P 2A. Since X is algebraic and o(X)217, by Corollary 2.10, the lattice Ty admits « ‘unique primitive embedding into the K3 lattice A, Thus, the embedding 30 is isomorphic to the canonical embedding; in particular, vu Eg(-IPOT = NSN),(On 3 surfaces with large Piatd number us (iv): By Theorem 5.7, since E,(—1)*+NS(X), there is a Nikulin in- volution 1 on X such that, if x: XY is the rational quotient map, then z, induces a Hodge isometry T,(2)=T,, and there is a primitive embedding N@E,(—1)=+NS(Y). Note that ¥ is an algebraic K3 surface and p(¥)217. Hence, NS(Y) is uniquely determined by its signature and discriminant-form {Corollary 2.10). Furthermore, N@E,(~1) and the Kummer lattice K have ‘isomorphic discriminant-lorms (by (4:3)Gi) and (5.7)Gi)). Thus, by Lemma 23, the primitive embedding N@E,(—1)=+NS(¥) determines a primitive embed: ding KNS(Y). But now by Theorem 42(iv), ¥ is a Kammer surface. QED. We should point out that the “algebraic” hypothesis is used in an essential way (in guaranteeing the uniqueness of the lattice NS(Y), given its signature and discriminantform). In fect, the generic K3 surface with Ey(—1)? in its Néron-Severi group has a Nikulin involution of the right type, but the quotient 3s not Kummer; conversely, the generic Kummer surface has no double cover which has a Nikulin involution of the right type. This happens because the lattices K and N@E,(~ 1) are not isomorphic, even though they have the same signatures and diseriminant-forms, Corollary 6.4. Let X be an algebraic K3 surface. () If p(X)=19 or 20, then X admits a Shioda-Inose structure. Gi) If pXV=18, then X admits @ Shioda-Inose structure if and only if T2U@T Gif 900 T2U°@T Proof. This follows immediately from Theorem 63 and Corollary 26. QE.D. 7, then X admits a Shioda-Inose structure if and only if Corollary 6.4 in the case p(X)=20 was first proved by Shioda and Inose [18], using somewhat different methods, 7. Remarks on a conjecture of Takayuki Oda In [13], Takayuki Oda made the following Conjecture. Let X be an algebraic K3 surface, and suppose that either p(X)=18, 19, of 20, or that p(X)=17 and the discriminant of the intersection.form on NS(X) is a square, Then there exists an abelian surface A and a correspondance between X and A whick induces a Hodge isometry (Tx@Q)(T, BQ). Corollary 7.1. Oda’s conjecture holds whenever p=19 or 20. Proof. By Corollary 64, X admits a Shioda-Inose structure in this case, The Shioda-Inose structure induces such an isometry which is defined over Z. QED.120 DR Morrison Remark 7.2. The following hypothesis must be added to Oda's conjecture: “There exists an embedding of Q-lattices (FOQYH(U7SQ)” Proof. Note that since (TAYH(4,Q)=(V7DQ, this bypothesis must hold for any K3 surface satisfying the conjecture. How- cover, there exist K3 surfaces with p= 17 or 18 which do not satisly this hypothesis: if T is a lattice of signature (2, 2) which does not represent zero ‘over Q then T@Q has no such embedding, by Corollary 27. On the other hand, by Corollary 2.10, T is the transcendental lattice of some K3 surface with Picard number 18, (There is a similar construction for p=17.) QE.D. Remark 7.3. When p=17, the hypothesis in Oda’s conjecture that the discrim- inant of the intersection-form on NS(X) be a square is unnecessary. Proof. In case p=17, X admits a Shiods-Inose structure if and only if Ty U?@T-, where Tis a negative even lattice of rank 1; such an X will satisfy the conclusion of Oda’s conjecture. On the other hand, any positive even integer 2k defines a negative rank { even lattice T’= (2k), and the lattice T= U2@¢—-21) occurs as the transcendental lattice of some algebraic K3 surface X by Corol- lary 2.10, But now, diser(NS(X)) = —diser( Ty diser(T)=24, which need not be a square. QED We thus propose the following Modified conjecture. Let X be an algebraic K 3 surface, and suppose that there is an embedding $: (Ty®Q)(U? @Q) of Qulattices. Then there exists an abelian surface A and a correspondance between X and A which induces a Hodge isometry (F,@Q)>(T,2O). Remark 7.4. The Hodge conjecture implies this “modified conjecture” Proof, Let T=U*AG(Tx@Q). By Corollary 1.94), A such that TST. induces an isometry (T,@Q(T, BQ) there is an abelian surface which gives a class in HBX x A\HAX x A,Q) (ying in the Kinneth component H"(X,Q)@H7(A,Q); cf [14]}. But the Hodge conjecture asserts that such a class is given by @ Q-linear combination of irreducible algebraic cycles; one of these will be a correspondance inducing the given isometry. QED.(On 3 surfaces with large Picard number nt References 1. Burns, B. Rapoport. M: On the Toei problem for Kilesian K3 surfaces, Ann. Scien. Ee Norm. Sup 8, 235-274 1975) 2 Hodge, WV.D2 The topological invariants of algebraic varieties Proc. Intern. Cong. Math ‘Cambridge 1. 182-192 (1950), 3. Kneser, M.) Klssenzabln indenter quadraticcher Formen io drei oder mehr Verindertichen ‘Arsh. Math, (Basel) 7, 323-352 (1086) 4, Kodaita, Ki On the structore of compact complex analytic surface, 1. Ams. J. Math. 86, 751~ 798 1964) 5, Kullhov, Vs Epimorphicity of the period mapping for surfaces of ype K3. fin Russian) Uspehi Mat. Nauk. 32, (4) 257-288 (197) 6, Looijngiy Es A Toveli theorem for Kibleiastein K3 surfaces. Lecture Notes in Mathe- retics vol 494, 107-112. Belin-Heidlberp-New York: Springer 1981 7, Looiengs, £ Peters C.: Torelli theorems for Kabler K3 surtaces. Compasitio Math. 42, 145- 386 (1981) & Milnor. 1: On simply connected 4-manifolés, Symposium Internacional de Topologia Alge braica, La Universidad Nacional Autaoma de México y Ia UNESCO pp. 122-128, 1958 9, Namikawa, ¥.2 Surectviy of period map for K3 surface, Classification of algebraic and analstic manifolds Progress im Mathematics, vol. 38, 379-397, Boston-Basel-Stutgar: Birkhduser 1983 40. Nikulin, Vo On Kummer susaces. ly, Akad, Nauk SSSR 38, 278-293 (1975): Math, USSR aves 9, 261-275 (1975) 11, Nikulin, Vi Finite groupe of automorphisms of Kaberin surfaces of «ype K3. Trudy Mosk. Mat. Ob. 38, 18-137 (1979): Trans. Moscow Math, Soe, 38, 71-135 (980) 12, Nikulin, Wi: Integral symmetric bilinear forms and some of their applications v. Akad Nauk ‘SSSR 43, 111-177 (1979), Math, USSR l2vesta 14, 108-167 (1980) 13, Oda, TA note on the Tate conjecture for KS soriuces. Proc, Japan, Acad Ser. ASS, 296-300 (0980) 14, Dkamoto, M.: On a certain decomposition of 2-dimenionsl cycles on & product of two Algebraic surfaces. Proc, Japan Acad. Sor AST, 321-325 (981) 15, Ptecki Shapiro Shafarevieh, LR--A Torelh theorem for alebrai surfaces of «ype K3. lx ‘Akad Nauk SSSR 36, 530-572 (1972); Math. USSR Izvestia§,S47-587 (1971) 16, Shafarovich, LR. ed Algebraic surfaces. Proc. Steklov Institute of Math. 76, (1965) 17. Sbiods, T. The peried map of abelian surfaces J. Fae. Sc, Uns. Tokyo 28, 47-S9 (1978) 1. Shion, T, Inose, H.: On singular K3 surfaces. Complex analysis and algebraic geometry papers dedicated to” K, Kedaira. Iwanam) Shoteo and. Cambridge University Press 1977, Pp 136 19. Sia, ¥eT. A simple proof of the surjectivity of the period map of K3 suefaces, Manuscripts salt 36, 311-31 (1981), 20, Sh, ¥-T. Every K3 surfoce is KHMer. Invent math. 73, 139-150 (1983) 21, Todorov, A> Applications of the Kahler Finsten-Calabi-Yaw metic to moduli of K3 surfaces Invent math. 61. 281-268 (1980) 22, We, WT: Classes caractristigues et Fearés dune yard, CR. Acad, Sci, Paris 230, S08 0950) Oblatam $-V-1983 Added in prot Im a recent preprint entitled “On the moduli space of vector bundles on K3 surfaces and its application to the Hodge conjecture", §. Muhai has showra that if X and ¥ are algebraic K2 ‘sinfaces with Picard number a least Ui, aad f$: T,@Q > T,@Q isa Hodge isometry, then there is some integer n such that ng i induced by an algeblaic cycle on X'xY. The “modified ‘conecure™ 8 Sect. 7 follons (rom this combined With Thearem 63}, by ao argument similar to aa.