THE 1729 K3 SURFACE

KEN ONO AND SARAH TREBAT-LEDER

Abstract. We revisit the mathematics that Ramanujan developed in connectionwith the famous “taxi-cab” number 1729. A study of his writings reveals that hehad been studying Euler’s diophantine equation

a3 + b3 = c3 + d3.

It turns out that Ramanujan’s work anticipated deep structures and phenomenawhich have become fundamental objects in arithmetic geometry and number theory.We find that he discovered a K3 surface with Picard number 18, one which can beused to obtain infinitely many cubic twists over Q with rank ≥ 2.

1. Introduction and Statement of Results

1.1. 1729 and Ramanujan. Srinivasa Ramanujan is said to have possessed an un-canny memory for idiosyncratic properties of numbers. J. E. Littlewood famouslyremarked that “every number is a personal friend of Ramanujan.” This opinion issupported by the famous story of 1729, the so-called “Hardy-Ramanujan” numberafter the famous anecdote of G. H. Hardy concerning a visit to the hospital to see hiscollaborator Srinivasa Ramanujan: (page 12 of [6]):

I remember once going to see him when he was ill at Putney. I had ridden in taxi-cabnumber 1729 and remarked that the number seemed to me rather a dull one, and thatI hoped it was not an unfavorable omen. “No,” he replied, “it is a very interestingnumber; it is the smallest number expressible as the sum of two cubes in two differentways.”

Indeed, 1729 is the smallest natural number which is the sum of two positive cubesin two different ways. We have that

(1.1) 1729 = 93 + 103 = 123 + 13.

At first glance, it is very surprising that Ramanujan knew this strange fact. How-ever, it turns out that Ramanujan had been studying Euler’s diophantine equation(see Section 8.5 of [1])

x3 + y3 = z3 + w3.

2000 Mathematics Subject Classification. 14G05, 14H52.The authors are supported by the NSF. The first author also thanks the Asa Griggs Candler Fund

for their generous support. The authors would also like to thank Bartosz Naskrecki for correctingan error in an earlier version of the paper.

1

arX

iv:1

510.

0073

5v3

[m

ath.

NT

] 1

4 O

ct 2

015

2 KEN ONO AND SARAH TREBAT-LEDER

In his Lost Notebook, he offered a remarkable method for finding an infinite familyof solutions, which can be found in the appendix and involves expanding rationalfunctions at zero and infinity. The integer 1729 was one example he produced thisway. Hirschhorn later devoted four papers [5, 7, 8, 9] to examining this. Hirschhornproposed that Ramanujan might have used a parametrization of solutions he hadpreviously discovered along with some recurrence relations to arrive at his rationalfunctions. The parametrization [10, 11] is

(3a2 + 5ab− 5b2)3 + (4a2 − 4ab+ 6b2)3 = −(5a2 − 5ab− 3b2)3 + (6a2 − 4ab+ 4b2)3.

We will work with this parametrization, as opposed to the rational functions, thesolutions it gives grow polynomially. Dehomogenizing, these give us two solutions to

x3 + y3 = m(t)

for each value of t, where m(t) := (7 + t+ t2)(1 + t+ 7t2)(13− 23t+ 13t2). We have

P1 = (x1, y1) = (3t2 + 5t− 5, 4t2 − 4t+ 6) and(1.2)

P2 = (x2, y2) = (−5t2 + 5t+ 3, 6t2 − 4t+ 4).

1.2. A modern interpretation. For cube-free integers k, we can view

x3 + y3 = k

as an elliptic curve over Q, and its Weierstrass equation is given by y2 = x3 − 432k2.This is a cubic twist of E : y2 = x3−432 by k, and over Q is torsion-free for all k > 2.So we can view Ramanujan’s family of solutions as giving us two points on infinitelymany cubic twists of E. However, we will first consider them all together as a singleelliptic curve over Q(t).

Theorem 1.1. The elliptic curve

Em(t)/Q(t) : y2 = x3 − 432m(t)2

has rank 2.

This allows us to give a lower bound on the number of cubic twists of E withMordell-Weil rank ≥ 2 over Q.

Theorem 1.2. There exist positive numbers C1, C2 such that if T > C1, then thenumber of cube-free integers d for which the curve Ed(Q) has rank at least 2 is atleast C2T

1/3.

Assuming the Parity Conjecture, which says that the rank of Ed(Q) is even if andonly if the sign of the functional equation of L(Ed, s) is +1, we can also say somethingabout cubic twists of E with rank at least 3.

Theorem 1.3. Assume the Parity Conjecture. Then there exist positive numbersC3, C4, such that if T > C3, then the number of cube-free integers d for which thecurve Ed has odd rank ≥ 3 is at least C4T

1/3.

THE 1729 K3 SURFACE 3

Remark. Tim and Vladimir Dokchitser showed (see [4]) that finiteness of the Tate-Shafarevich group implies the parity conjecture for elliptic curves over number fields.

Remark. In earlier work, Stewart and Top [13] also obtained estimates for the numberof cubic twists with rank ≥ 2 and rank ≥ 3. They previously obtained the estimatesin Theorem 1.2 and Theorems 1.3 (see Theorem 7 of [13]).

We can also view Em(t) as an elliptic surface E over Q. We will show that it isactually a K3 surface.

Theorem 1.4. The smooth minimal surface E/Q associated with the equation

y2 = x3 − 432m(t)2

is an elliptic K3-surface with Picard number 18.

We will give the definition of a K3 surface and its Picard number in Section 4.1.In Section 2, we will prove Theorem 1.1 by studying a map to the space of holomor-

phic differentials on an auxiliary curve C. In Section 3, we will use this along withresults on power-free values of binary quadratic forms to prove Theorems 1.2 and 1.3.Lastly, in Section 4, we will use Theorem 1.1 and results on elliptic K3 surfaces toprove Theorem 1.4.

2. Proof of Theorem 1.1

2.1. Background. In this section, we will show that Em(t)/Q(t) has rank two. Tostudy this rank, we use a map, described in [13], from the Q(t) points of Em(t) to thevector space of holomorphic differentials on the auxiliary curve C/Q : s3 = m(t). Foreach point P = (x(t), y(t)) in Em(t)(Q(t)), we define an element φP of MorQ(C,E),where E/Q is given by y2 = x3 − 432 as in the introduction, by

φP (t, s) =

(x(t)

s2,y(t)

s3

).

Then the mapλ : Em(t)(Q(t))→ H0(C,Ω1

C/Q)

is given by λ(P ) = φ∗PωE, where φ∗PωE denotes the pullback via φP of the invariantdifferential ωE. Proposition 1 of [13] states that λ is a homomorphism with finitekernel. Therefore, we just need to understand its image.

2.2. Rank at least two. To prove Theorem 1.1, we first need to show that the rankis at least 2. As Ramanujan has given us two points, we just need to check theirimages and see that the differentials are linearly independent. We compute that

φ∗P1ωE =

1

3s2(−5− 38t+ 16t2)dt and φ∗P2

ωE =−1

3s2(−16 + 38t+ 5t2)dt,

where P1 and P2 are as in 1.2.

4 KEN ONO AND SARAH TREBAT-LEDER

2.3. Rank at most two. Lastly, we need to exclude two independent holomor-phic differentials from the image to show that the rank is at most two. The spaceH0(C,Ω1

C/Q) has dimension 4, and a basis is given by

ω1 =dt

s,

ω2 =t2 dt

s2,

ω3 =t dt

s2, and

ω4 =dt

s2.

As is explained in [13], if ζ3 is a primitive cube root of unity, and ζ is an automor-phism of C defined by ζ(t, s) = (t, ζ3s), then ζ acts on the differentials of C via thepullback ζ∗f(t, s)dt = f(t, ζ3s) dt for any function f on C. This gives a linear actionon the space H0(C,Ω1

C/Q) of holomorphic differentials on C, and a decomposition of

H0(C,Ω1C/Q) into eigenspaces on which ζ acts by multiplication by ζ i3 for i = 0, 1, 2.

The image of λ always lands in the ζ13 eigenspace, and therefore ω1, which is in theζ23 eigenspace, cannot be in the image. This shows that the dimension of the imageis at most 3.

Next, we need to get the image down to two dimensions. Consider the involution

σ(t, s) :=

(304t2 − 281t− 95

256t2 − 608t+ 361,

441s

256t2 − 608t+ 2361

).

The space of σ∗-invariant holomorphic differentials is generated by

ω := ω2 −19ω3

8− 5ω4

16.

Hence the quotient of C by ω is elliptic; in fact it is given by

y2 − 132003308704176245102247936y

= x3 − 7550778520501689214254602155146485718479954847046041.

This is not isogenous to E over Q - it is isogenous to E4. We show that this impliesthat ω is not in the image.

Assume that ω is in the image. Say P = (x(t), y(t)) ∈ Em(t) has λ(P ) = ω. Then

φP ∈ MorQ(C,E) gives a morphism φP ∈ MorQ(C/σ,E). Since ω is non-constant,φP must be a finite morphism, and hence be an isogeny between C/σ and E. This isa contradiction. Therefore, we have shown that the rank is exactly two.

THE 1729 K3 SURFACE 5

Remark. We did these calculations using Magma. If G is the automorphism group ofC, then σ = G.1 ∗ G.2. To recognize the quotient as an elliptic curve, a point on itmust be specified. We used [0, 1, 72402101023/2666490096, 0, 0, 7957649/26496].

3. Cubic Twists

3.1. Studying the cubic twists. If E/Q(t) is an elliptic curve which is not isomor-phic over Q(t) to an elliptic curve defined over Q, we can specialize t to a rationalnumber t0. A result of Silverman [12] gives that for all but finitely many rationalnumbers t0, this specialization map φt0 : E(Q(t))→ Et0(Q) is an injective homomor-phism. Therefore, all but finitely many values of t yield elliptic curves Em(t)(Q) withrank at least 2. Now, we need to count them.

3.2. Proof of Theorem 1.2. We use a result from [13]. Let F be a binary formwith integer coefficients, non-zero discriminant, and degree r ≥ 3. Let A,B,M, k beintegers with M ≥ 1 and k ≥ 2. Let m be the largest degree of an irreducible factorof F over Q and suppose that m ≥ 2k + 1. Let w be the largest positive integersuch that wk divides F (a, b) for all integers a and b with a ≡ A (mod M) and b ≡ B(mod M). Let Rk denote the number of k-free integers t with |t| ≤ x such that thereexists integers a and b with a ≡ A (mod M), b ≡ B (mod M), and F (a, b) = twk.Then there exist positive numbers C1, C2 which depend on M,k, F such that for allx > C1, we have that Rk(x) > C2x

2/r.We apply this result with k = 2 and M = 1 to the binary quadratic form

F (X, Y ) = Y 6m(X/Y ) = (7Y 2 +XY +Y 2)(Y 2 +XY + 7Y 2)(13Y 2 + 23XY + 13Y 2).

Note that in this case w = 1. We get that R3(x) > C2x1/3 for all x > C1, which

proves Theorem 1.2.

3.3. Proof of Theorem 1.3. Let d be a cube free integer and r(d) denote the rankof Ed(Q). In this situation, the Partiy Conjecture implies (see ??) that

(−1)r(d) = −w3 ·∏p 6=3

wp,

where

w3 =

−1 if d ≡ ±1,±3 (mod 9)1 otherwise,

and for p 6= 3,

wp =

−1 if p | d and p ≡ 2 (mod 3)1 otherwise.

We will find cubic twists with rank at least 2 and root number −1, and hence rankat least 3 assuming the Parity Conjecture. Taking a ≡ 0 (mod 9) and b ≡ 1 (mod 9)gives us that F (a, b) ≡ 8 (mod 9), and as long as F (a, b) is square-free, the rootnumber will be −1.

6 KEN ONO AND SARAH TREBAT-LEDER

So we have that if k = 2, A = 0, B = 1,M = 9, then there exists C3, C4 such thatif x > C3, then R2 > C4x

1/3. This proves Theorem 1.3.

4. A K3 surface

4.1. Background on K3 surfaces. A K3 surface is a smooth minimal completesurface that is regular and has trivial canonical bundle. Some examples of K3 surfacesinclude intersections of three quadratics in P5, intersections of a quadric and a cubicin P4, and a non-singular degree 4 surface in P3. Andre Weil named them in honorKummer, Kahler, Kodaira, and the mountain K2.

For a K3 surface X, we define the Picard group to be

Pic(X) := Div(X)/Princ(X).

Restricting to divisors of degree zero, we get

Pic0(X) := Div0(X)/Princ(X).

The Neron-Severi group is then defined as

NS(X) := Pic(X)/Pic0(X).

It is a finitely generated abelian group, and its rank is the Picard number of X. ForK3 surfaces, this Picard Number is always ≤ 20.

4.2. Proof of Theorem 1.4. Consider a Weierstrass equation

y2 + a1(t)xy + a3(t)y = x3 + a2(t)x2 + a4(t)x+ a6(t)

with ai(t) ∈ Q[t] and non-constant discriminant ∆(t). The criterion given in [2] onpages 276-277 show that it’s a K3 surface if the conditions below hold for N = 2 butnot for N = 1.

• The degree of ai(t) ≤ Ni.• No 12th power of a non-constant poly in Q[t] divides

gcd(g2(t)3, g3(t)

2)

or

gcd(t12Ng2(t−1)3, t12Ng3(t

−1)2).

These hold for our E/Q, which has g2(t) = 0 and g3(t) = 1728m(t)2. Therefore, wehave a K3 surface.

To complete the proof, it suffices to compute the Picard number. Using Tate’salgorithm, this surface has six bad fibers, each of type IV. The Shioda-Tata formulafor the Picard number ρ then says that

ρ = r + 2 + 6 · 2 = r + 14,

THE 1729 K3 SURFACE 7

with r the C(t)-rank of the elliptic curve defined by our equation. We know that theQ(t)-rank is 2. Also, note that y2 = x3 − 432k2 has CM over Q(

√−3), and

φ(x, y) =

(−13x3 + 576k2

x2,

√−39x3y + 384

√−3k2y

x3

)is an endomorphism of degree 3. The action of the endomorphism ring on our twoindependent points gives a Z-module of rank 4. So ρ ≥ 4 + 14 = 18.

By considering the prime of good reduction p = 17 and computing the L-function ofthe surface after reduction, we see that the corresponding Picard number over F17 isat most 18. Using the specialization theorem for the Neron-Severi groups, we obtainthat the rank over C(t) is 4 and the Picard number is exactly 18.

8 KEN ONO AND SARAH TREBAT-LEDER

Appendix

This is the famous page from Ramanujan‘s Lost Notebook on which one finds hisrepresentations of 1729 as the sum of two cubes in two different ways.

THE 1729 K3 SURFACE 9

References

[1] G. E. Andrews and B. C. Berndt, Ramanujan’s lost notebook, Part IV, Springer, New York,2013.

[2] F. Beukers and J. Stienstra, On the Picard-Fuchs equation and the formal Brauer group ofcertain elliptic K3-surfaces, Math. Ann. 271, no. 2 (1985), pages 269–304.

[3] B. J. Birch and N. Stephens, The parity of the rank of the Mordell-Weil group, Topology 5(1966), pages 295–299.

[4] T. Dokchitser, Notes on the parity conjecture. In Elliptic curves, Hilbert modular forms andGalois deformations, Adv. Courses Math. CRM Barcelona, pages 201–249. Birkhauser/Springer,Basel, 2013.

[5] J. H. Han and M. D. Hirschhorn, Another look at an amazing identity of Ramanujan, Math.Magazine 79 (2006), pages 302–304.

[6] G. H. Hardy, Ramanujan, Cambridge Univ. Press. Cambridge, 1940.[7] M. D. Hischhorn, An amazing identity of Ramanujan, Math. Magazine 68 (1995), pages 199–

201.[8] M. D. Hischhorn, A proof in the spirit of Zeilberger of an amazing identity of Ramanujan, Math.

Magazine 69 (1996), pages 267-269.[9] M. D. Hischhorn, Ramanujan and Fermat’s last theorem, Austral. Math. Soc. Gaz. 31 (2004),

pages 256–257.[10] S. Ramanujan, Question 441, J. Indian Math. Soc. 6 (1914), pages 226-227.[11] S. Ramanujan, Question 571, J. Indian Math. Soc. 7 (1915), page 32.[12] J. H. Silverman, Taxicabs and sums of two cubes, Amer. Math. Monthly 100, no. 4 (1993),

pages 331–340.[13] C. L. Stewart and J. Top, On ranks of twists of elliptic curves and power-free values of binary

forms, J. Amer. Math. Soc. 8, no. 4 (1995), pages 943–973.

Department of Mathematics and Computer Science, Emory University, Atlanta,Georgia 30322

E-mail address: ono@mathcs.emory.eduE-mail address: strebat@emory.edu