` 21351 >>: Okay. So next speaker of this morning is Bianca Viray, who is going to talk about the [inaudible] in genus-two. >> Bianca Viray: I'd like to first thank the organizers for giving me the opportunity to speak. As you see, I am going to be giving a board talk. So if it's too small, then just yell write larger, just whenever you feel like it. Alternatively, you can sit closer. I promise I don't bite or will not spray water guns or anything like that. Okay. So I'm speaking on Igusa class polynomials embeddings of quartic CM fields and arithmetic intersection theory. I'll give background on all of it so it hopefully won't be as intimidating as it sounds. This is all joint work with Helen Grudman, Jennifer Johnson Loung and Kristine Lauter Audriana Solerno and Erica Whittenborn. So the motivation for this problem is looking at the CM method for genus-two. So first let me review the genus-one case. Francois Moran mentioned this briefly on day one but I'm just going to review it. So step one is to compute the Hilbert class polynomial and as mentioned there are three methods for doing this. There's the complex method. This was first done by Atkin and Morah in '93. And then there's the Chinese remainder theorem method which was done by Agashia Lauter and Vancutessen in '04. And then I guess, out of order, but there's also the p-adic method, first studied by Covenia and Hinnuke [phonetic] in '02. Okay. So first we compute this Hilbert class polynomial using your favorite method from here. And then step two is to find a root of your Hilbert class polynomial mod P. And step 3 is then to construct with an elliptic curve E with J invariant J 0. So I'm not going to go into details but this is the outline. So what is the analog in the genus-two case. Okay. So the J invariant gets replaced with the Igusa invariant. I-1, I-2, I-3. This isn't exactly correct to completely determine every genus-two curve you actually need two Igusa invariants but for the vast majority of cases -- for instance, if the first Igusa invariant is non-0 and characteristic of the ground field is not 2 or 3, then these three values completely the isomorphism class. It's not completely true, but it's true in enough cases that that's all we're going to work with. Then we have the Hilbert class polynomial. So this gives you genus-one curves which have complex multiplication by the ring of integers in a quadratic imaginary field. So instead of looking at a quadratic imaginary field we look at the ring of integers in K where K is is totally imaginary, quadratic extension. That's probably small. Totally imaginary quadratic extension over a real quadratic field F. So we call K a quartic CM field. And throughout my talk K will refer to a quartic CM field and F will always be the quadratic sub field. Then the Hilbert class polynomial gets replaced with three polynomials and you define it as you take the product over all genus-two curves and you have an embedding of okay into the endomorphism ring of the Jacobean. And then you take X minus the Igusa invariant at C. Okay. So we have three of these polynomials. And then we want to do the same thing. Step one is compute HJ of OK for all J. And again we have the exact analog. So there's a complex method, studied first by Shpalic [phonetic] in '94 and Fonwonlin [phonetic] in '99 and Vang [phonetic] in '03. I should mention, this is not a complete list of the people who have done this. There are many, many more people but the list is much longer than on this side. So this one I only have the first people. Okay. And then there's the CRT method, which was done by Isentrager [phonetic] and Lauter in '05. And then the p-adic method, which is studied by Gaudry [phonetic], Hautman, Cole [inaudible] and Vang. This was in '06. Okay. So that's step one. And even though what I wrote on the board looks like an exact analog, there's actually a major difference between the two cases. So the Hilbert class polynomial is always -- has integer coefficients. So these methods, for instance, the complex method you take the J invariants, compute them as complex numbers out to a certain precision, multiply all the roots together and then recognize what the integer coefficient should be. And as long as you have sufficient precision, this is possible. But on the other hand the Igusa class polynomials, these have coefficients in Q. So if you drew the same method, you need to have a bound on the size of the denominators in order to know what precision you compute to. And in fact that's true for all of these methods. All of these methods require as input not only the quartic CM field but also a bound on the denominators. Okay. So for the CM method in genus-two what we really want, you want a bound on the denominators of these Igusa class polynomials. And ideally you would like the bound to be very sharp. I mean, if you think of how I described it, computing the precision, if you have a bound but it's much too large then you're going to be computing lots of extraneous precision which can slow down your algorithm quite a bit. So we really -- it's not just that we want A bound but we'd like our bound to be as sharp as possible. Okay. So what's been done so far -- so in 2003 Lauter conjectured that the primes appearing in the denominators are bounded by the discriminant of your quartic CM field and that they divide the discriminant of the field minus some integer squared. Okay. So it turns out that this divisibility condition follows from work of Goran in '97 plus a few additional observations. But the main bulk of the work was in this paper of Goran. Okay. And then in 2007 Goran and Lauter proved a bound on the primes and then earlier this year they proved a bound on the valuations. So the valuations of the primes, of the denominator. Okay. So certainly while a bound is good, I mean this now turns these methods into something which only require a quartic CM field as the input. You don't have to give anything else about the denominators, because you can use these results. But unfortunately these bounds are not very sharp. So we would still like to get something that's a little bit better. Okay. So something I'll mention just as a note which we'll return to later, so both of these papers heavily use what we call the embedding problem to prove the results. We'll return to this later. So we've seen now two things out of the title, so that's good. So to continue and see how we can maybe get a better handle on these denominators we need to look closer at what the definition of these Igusa invariance is. All I told you before is they determine the isomorphism class but that doesn't help you compute them. Okay. So let me first mention that there are multiple ways of defining the Igusa invariants, some of which may turn out for better for other methods. For what I'm going to say in this talk, the definition I'll describe is the best. So this is only one of many other possible approaches, and I'm not claiming this is the best in general. Just the most relevant for today. So given a genus-two curve C, you can send it to an abelian surface. Namely by just taking its Jacobean. And this map is injected. If you have two nonisomorphic genus-two curves, you send them to your Jacobean, you'll get two nonisomorphic Jacobeans. So instead of thinking of the Igusa invariants as functions on genus-two curves, we can think of these Igusa invariants as functions on the modulo space of abelian surfaces. That's what we're going to do. >>: [inaudible]. >> Bianca Viray: No. No particular field. Just a field. So think of Igusa invariants as functions on the modelized space of principally polarized abelian varieties, abelian surfaces. Okay. So I'm going to write out an expression for these Igusa invariants it's not going to make much sense in the beginning but hopefully in a few seconds it will. So we can write the first one in this form. The second one can be defined like this. Okay. So they have all these kis and psis floating around. What they are is not so important. The remarks that are important is that psi four psi six, ki ten and ki 12, so all of these functions that are appearing have no poles on the modelized space of smooth abelian surfaces. So that means if I take any smooth abelian surface and evaluate these functions on that surface, I will always get a finite number. I will never get something that's infinity. And the other remark is that ki 10 is the only function in the denominator. So let's tie this back into what we were thinking about what we want to study is the denominator, what primes are appearing and to what power they appear. So if you have a prime appearing in the denominator, then if you look over FP, then that will be an undefined function. You'll have a pole at that function. So the primes that appear in the denominator correspond exactly to the primes such that there's a genus-two curve with CM so that ki 10 has a 0 there. So none of these can be infinity. The only way this function can be infinity is if ki 10 is 0. That's the only way this happens. So that means -- so that means that the P valuation of the denominators is bounded above by the number of 0s of ki 10 at the CM points over FP bar. So you take a curve over FP bar with CM and if the function -- if ki 10 is 0 at that point of whatever its order of managing is, you'll get that many primes appearing in your denominator. It's a little bit annoying, but if you sort of unravel these definitions, this is exactly what it's saying. I lied slightly because it's really bounded above by a suitable multiple. Because we have either K 10 to the 6 or K 10 cube but that factor is not so important. The upshot of this is that if you can understand the 0s of ki 10 to what multiplicity they occur then you understand the denominators of the Igusa class polynomials. That was an annoying trip into modelized spaces and modulu interpretation and hopefully we'll back into that and come back to something more concrete. Okay. So after doing this, the upshot is you can get a statement like the following. So if you take one-sixth times the P valuation of the denominator -Let me write bigger. This is maybe the main point. So if you can't see this then really you should tell me. So 16 of the P valuation of the denominator of the constant term of H1 OK, this up to cancellation -- so the only reason why I took H one here is because I want to know what factor to multiply by up front. But you could choose H2 or H3 and then you would take ki one-fourth or also one-fourth. Okay. So using the modular interpretation, what I just said, so this value is equal to the number of embeddings of OK into an endomorphism ring of a product of two elliptic curves. This is counted up to isomorphism and with multiplicity. So from going from here to here, so this by what I said is just basically telling you the number of 0s of ki 10. You can use the moduli interpretation of ki 10, the points in ki 10 where ki 10 is 0corresponds exactly to the abelian surfaces that are products of two elliptic curves. So your abelian surface is just of the form E cross E prime. It has CM. So you want OK embedded into it. The condition that OK embedded into it basically implies that these two elliptic curves are supersingular. So this is just trying to back us out of thinking of a concrete interpretation of these 0s of ki 10. And then this is also equal to the P part of the arithmetic intersection number. So you take DIV ki 10 dotted with the CM cycle and you want just the P part of that. Okay. So now using this interpretation, this moduli interpretation, we get these three equivalent things. And so maybe you're not so convinced. I told you I was trying to get to more concrete things, but I have all these ugly words up here. I have this arithmetic section number. Looks like maybe I just lied to you. But I promise, I didn't. Okay. Okay. So the point is that this ugly thing, there's a conjectural formula for, which removes all of the arithmetic intersection theory and all of this. So in 2006, Grunya and Yang [phonetic] gave a conjectural formula for this DIV ki 10, dot the CM cycle under the assumption that the discriminant of the real quadratic field is 1 mod 4 and prime. Okay. And actually Yang has proved this conjecture, assuming that D tilde, which is the norm of the relative discriminant, it's prime, and two that OK is freely generated over OF and eta has a certain form. So I'm pretty sure this condition on the form of eta follows from this condition. But I haven't worked out all the details. But anyway so the point is that this condition is not really that strong, but this one is a very strong condition. Okay. So let me now write their conjectural formula. So they actually gave a formula for a more general case than this intersection number, but it gives you a formula for the intersection number we care about. Okay. So this P part -- this 2 ki 10 with the CM points at P, this is equal to you sum over an integer M such that M is of the form D minus X squared over 4 for some X in Z. You want M to be positive. And then you sum over prime ideals -- sorry. I need a little bit of notation. So K tilde is the reflex field. And F tilde is the real sub field, real quadratic sub field of K tilde. You don't really need to have a good understanding of what the reflex field is. It's just some other quartic CM field and there's a formula, an explicit way to get it from K. It's just some other field. So you take primes in the ring of integers of OF, where they lie over your P and then you take something of the form M plus M square root of D tilde over true D. You want this to be an inverse of the relative discriminant ideal, and you want the absolute value of N to be less than M times square root of D tilde. And we have this BT of P. Okay. So this is 0 if P is split in K tilde. And we've got -- so we get the P valuation of T plus 1 times this row value of T. Then we get the inertial degree these otherwise. Okay. And this row -- so row of an ideal A is equal to the number of integral ideals in the ring of integers of K tilde such that the norm equals the one you started with. Okay. So that was a long time to write. And it looks very ugly and involved, but if you look at it, this is very easy to compute on a computer. So if you -- as long as your D tilde and D are not so large, which we also have the case in genus-one, these factors are very easy to compute. You're just looking at the splitting behavior of an ideal in a quartic field and you're taking the P valuation. This is something that's easy to plug in and easy to do, even though it's ugly to write down. So if this formula -- so we have it proved in these two cases, if it was true in all cases, this would be great. Because this 1-sixth of this, up to cancellation -- so this bound is very sharp. The only time that it's not equal to the actual valuation of what's in the denominators is when you have 0s of psi 6 ki 12 or psi 4 at the same points where you have 0s of ki 12. I don't know if this is a theorem but certainly experimentally this rarely happens. So we could use this conjecture, then this would give us a great bound to use for the genus-two case. >>: Course of two ->> Bianca Viray: No, it works for all of them. It's just a little bit. You have to think of how you're combining the polynomial. You have to do something with the symmetric functions and see what the bound of the denominator should be. I think it works out to the same bound, but I haven't worked through all of the details. >>: Even though you have sums. >> Bianca Viray: You have sums but somehow you have functions like you really have poles over an extension field and when you multiply them all together, it still gives you a bound, but just the formula might not be as nice. You just have to unravel the sums and think about what it means. But didn't want to tackle that during my talk. So we would like experimental evidence to verify this conjecture in cases where it's not proved. So it's the one -- so if we only compare this formula with the valuation of the denominators, we have this cancellation part which even though I said is is rare, it might happen and it might be hiding some of the errors. So ideally we'd also like additional ways to verify it. So what we're going to do is we're going to compute all three of these things, except we don't know what the cancellation factor is. We also don't know how to count multiplicity of embeddings, but the multiplicity is also something that is not there so often. So we'll just count all three and see if that gives us enough evidence for this conjecture. Okay? So let me quickly explain how we go about counting these embeddings. And then I'll present the evidence that we found. Okay. So we're going to keep one of the assumptions, which I just erased. So we're going to assume that OK is generated by just one element over OF. But we won't put any assumptions on the form of OF. Just whatever you want. Okay. So if we let omega equal D plus square root D over 2. So then 1 and omega are generators for the ring of integers. And you have the trace of eta, let me choose alpha 0 and alpha 1 so this is true. And beta 0 and beta 1 so that this is true. And these are just the relative norms and traces down to the real quadratic field. Okay. Then an embedding OK into an endomorphism ring, this is equivalent to giving elements lambda 1, lambda 2, in the endomorphism ring, such that the following conditions hold. We want lambda 1 plus lambda 1 joule. The joule represents the Rosadi revolution that equal each other. We want lambda 1 squared minus D lambda 1 plus D squared minus D over 4 to equal 0. Lambda 2 plus lambda 2 dual plus alpha 0 plus alpha 1, lambda 1. Four that the product and five that lambda 1 and lambda 2 commute. Probably isn't so hard for you to see why this is true. Basically we're thinking of lambda 1 is the image of eta. If OK is of this form, then the image of omega and image of eta and image of eta completely determine the embedding and these are the necessary sufficient conditions that need to be satisfied. So this seems like maybe a silly thing to do, because it's basically by definition. But the upshot is that if you unravel these conditions, and it tells you that it's only a finite computation. There's only finitely many possibilities for embeddings, and you just kind of go through them all and count them. The endomorphism ring of the product we can think of as matrices where you stick different isogeny rings and endomorphism rings in the different places and then composition of functions just corresponds to multiplying the matrix. So if lambda 1 is equal to this ABCD and lambda 2 is XYZW, so then conditions 1 and 2 imply that A is an integer, D is equal to the discriminant minus A. C is equal to B dual, and the degree of B is equal to minus A squared plus DA minus D squared minus D over 4. And this is positive. Okay. So this tells you that there's only finitely many possibilities for A and B, because there's only so many integers where this will be a positive number and then you want an isogeny of a fixed degree there's only finitely many of those. So this restricts lambda 1 to just finitely many cases. And then three, four and five tell you that Z is equal to alpha 1 B dual minus Y dual and that the degree of X, the degree of W and the degree of YV dual are all bounded. It gives you a bit more information than this. But this is enough to tell you that already this also reduces lambda 2 to finitely many possibilities. So I'm not saying that this is the smartest way to do it or the fastest way to do it, it is just a way that works and it's enough that you can compute it. I think if you wanted to compute a lot of these or your discriminants were very large, then you probably want to choose a correct basis, different omega and a different eta so that these are short vectors and that sort of thing. But if you just want to do it, then you can just pick whatever you want and hope it works. Okay. So we take this. So then for a fixed E and E prime we have finitely many choices for what's in the endomorphism ring, since our curves are supersingular, then we change this into a problem of counting elements into orders and ideals in the quartarian algebra infinity. We have MAGMA code that just goes through and does all of this. And then you count the number of bottom morphisms [inaudible] out by the isomorphisms. It's kind of tedious but that's what a computer is for. That's how we count out the number of embeddings. And okay. So let me just explain a few more things before I introduce the data. So one thing is that even though we can't count the multiplicity of the embeddings, we can sort of check if we're getting -- we can have a little bit of extra assurance, because the Bruinian Yang formula also includes counting multiplicity. So the Bruinian Yang formula, this order of P to the T plus 1, this counts the multiplicity. So we can pull this out of the formula right as a separate factor and we can see at least if we have the number correct and that all we're missing is the multiplicity. And the other thing is if we're missing multiplicity and this up to cancellation, the formula is correct, it should always be larger than the other two values. If there's cancellation that we're not seeing, it will make the primes appearing in the denominator smaller. If we're missing multiplicity, they will also reduce our number. So the Bruinian Yang formula should only be overcounting things, never undercounting. So I think that's all I need before I present the data. Does anyone have any questions before the board disappears? Okay. Then could I get the slide, please. Okay. So the examples we used were the 13 examples that Vonromblin [phonetic] computed. These are all of the genius 2 curves with CM defined over Q. In this slide I have the CM field, what D tilde is. And then so this is one-sixth of the constant term, and there's a little bit of fudging at 2 and 3, because these are their extra powers. You have to normalize the little - if you want to know the exact details of how we get the 2 and 3 powers you can ask me afterwards. This is the case where the conjecture is proved so it better be all right. And you see we do get agreement everywhere. The only place we're missing agreement is our number of embeddings, we only get two embeddings at 3 where we're supposed to get three to the fourth in the denominator but this is the multiplicity 2. This is the factor I pulled out. So looks like everything is okay. We just missed that multiplicity 2. And you can see at least in these five examples multiplicity doesn't appear that often. It only appeared at 1 prime. Only once. Okay. So let's look at the next group. So these are ones where D tilde is no longer prime. We have a square factor, but they're still all 1 mod 4. If you notice, we still you have all the agreement there. So the reason why there's two values for the Igusa invariant is for this field there are two curves with CM by that field and since these are all defined over Q, the Igusa invariants are over Q. So this is actually giving you the Igusa invariant of each curve, not just the product in the constant term. But if you take the product then you see that we get, what you get by all the other cases. Okay. So it's still looking good. And now we get to the times where D tilde is highly divisible by 2 and you see kind of goes haywire. So the ones where I have double stars, this one the discriminant of the real quadratic field is 8. So this doesn't even fall under the conjecture. But it would be nice to have a formula in all cases so we figured we'd run it anyway, see if we can see what happens and what discrepancy occurs. So I'm going to divide these up into a few different cases. So the first one is when D is 8, you'll see that there are just lots of extra primes. Many extra powers. I mean, all the other ones the number of primes should at least, the primes that are appearing should at least be agreeing. Maybe the powers are off. But the primes that are appearing should always be right. If you can't -- I mean, we can't be missing a prime just because we can't count multiplicity. We can find every embedding. So in this case Bruin yea and Yang -- well, so in the proof Yang shows that you need to consider only need to consider integers M and N such that M squared D tilde minus N squared over 4 G is a positive integer divisible by P. But if you go through his proof of why you only need to consider these, this really highly relies on the fact that D is 1 mod 4 and prime. And so when we were looking through the data of what M and N come up, it looks like you need an additional condition. So for these fields, our guess is that you need the condition 8 M plus N is congruent to 0 mod 16. So this is for these fields what it looked like numerically. If you impose that additional condition in the formula, then those all disappear. So again it's not a proof, just something to think about. Okay. So that looks much better. There's not these huge extra numbers, these ugly exponents. Okay. So then the second condition, the second discrepancy I want to focus on is these 2s. So you see most of the other differences or all of these 2s. When we find an embedding, there is actually an embedding. So that means that there is a curve that that 2 should actually be appearing in the denominator. So the reason why it's missing in the denominators is probably because of cancellation. But we don't -- we haven't proved that. But I mean we can miss things in the embedding but we can't count things that are not there. So definitely all the 2s should be appearing. And then the fractional multiplicity part -- so I'm not sure what's going wrong. But if you again go through Yang's proof, what Yang proves is that under his assumptions all of the times you will get a non-zero row value this will happen for the multiplicity, that the order of P at the integral prime you get M squared D tilde minus N squared over 4 D plus 1, that this will be equal to the prime lying over it, the order of this element plus 1 and in particular that this order is always odd. Okay. So this statement just falls apart when you're in these cases. I mean, if you try to work through the formula and you see the M and N that are causing these three halves, you'll see that this is not true. This is something like 16. So you're getting an even number. So you'll always have a fractional multiplicity, and I don't understand the rest of it enough to say how to fix it. But at least this is what's going well. And then maybe you found the last mistake by looking at it. But you'll see in this last field from the Igusa invariants from the embeddings we get a 23 to the fourth. But Bernie Yang [phonetic] has a 23 squared. As I mentioned Bernie Yang should never undercount. If we're missing things because of multiplicity or the because of this cancellation Bernie Yang should be larger. So this is not -- this discrepancy is not arising because of multiplicity or cancellation. It's arising because of something else. And this I cannot find in the proof. I mean, this discrepancy is very mysterious to me and I don't see where in the proof the assumption that D tilde is prime is causing you to get lesser, a lesser power for 23. Okay. So I'll leave you with that. [applause] >>: Thank you very much. Are there any questions? >>: Yes. So these easy to compute and psis you had so I'm sure you know more about this than I do but my experience there sort of modular form, some complex upper hand space. >> Bianca Viray: Yes. >>: So what does it mean for them to be [inaudible] B bar? >> Bianca Viray: Well. >>: Evaluating that [inaudible]. >> Bianca Viray: Okay. Okay. So Dave's question was in his experience these ki Is are modular forms and they're evaluated and defined on some complex upper half space, and he wanted to know how you evaluate them at FP points and Christine said if you evaluate them at CM points then you get values which are in number fields so then you can think about them and reduce mod P. >>: [inaudible]. >>: Any questions? Thanks again. [applause]