21694 >>> Katherine Stange: So thank you to the organizers. And also thank you to Microsoft. Hello to Kristin, wherever she is out there in videoland. So I'm going to talk about work which is joint work with Joseph Min [phonetic] actually when he was at Microsoft Research. He spent a semester at the Cambridge branch, piece, whatever it is. And this was from then. And so I'm going to start with a motivation which I'm then going to forget for the rest of the talk, because I think the question that comes out of it sort of takes on its own interest. So but let's start with this. So suppose we have an integer sequence. And I want to look at the indices and such that N divides the nth term. So I'll call it the index divisibility set. And obviously the interest of such a question depends greatly on what sequence you're looking at. So, for example, if your sequence is B to the N minus B, these already have a name, pseudo prime to the base B. And I want to look at the structure of the set by making it a directed graph. So you can take the elements, the indices which divide their terms as the vertices and draw an arrow from N to M if N divides M and N is smaller than M, and if it's a direct connection, so there's nothing in between, which is also in the set. So this was actually Chris Smyth took a look at this for Lucas sequences. So these are just linear recurrent sequences. Anything that satisfies the linear recurrence like that where A and B are whatever coefficients you want. If you call delta the discriminant A squared minus 4B, then you can describe this set in terms of all the arrows in the graph. So suppose you have -- suppose you have some N in the index divisibility set, and then you can classify all the arrows originating at that vertex. So you have an arrow heading from N to N times P where P is prime, whenever that prime is a devisor of that particular term L sub N or divides the discriminant delta. Then you have a few extra ones in there, which are classified down here. Sometimes you can go from N to 6N or 12N. So we saw this paper, and we thought of asking the same question for elliptic divisibility sequences, which are related to elliptic curves. So suppose we have an elliptic curve and a point on the curve, both defined over rational numbers. Then the point, the X coordinate of the nth multiple of the point. So I'm told that not everybody in this room is a number theorist. So I'll sort of toss out the basic facts about elliptic curves. Elliptic curve, its points form a group, which I'm sure you've all heard, even if you don't enjoy studying them all the time. This means the nth point looking at the X coordinate. So because of the form of the equation, the X and Y coordinates will have denominator something squared and cubed. We'll call that something DN. We'll just extract those numbers, those integers, take the positive integer and make that a sequence. And this is called an elliptic divisibility sequence. I won't talk about them for the rest of the talk, but I'll just mention that they carry a lot of the information about the elliptic curve. So you can see a lot of things going on in the elliptic curve just by looking at these sequences. Okay. So these were the sequences we were interested in looking at. And we got a theorem which is not dissimilar to what Chris Smyth got. So you can ignore some of the technical stuff here. But let's suppose we have an elliptical divisibility sequence. You do get an arrow from N. So looking at part 1. I guess I'll lose the pointer. Part 1 here. You get an arrow from N to N times the prime, when the prime divides the nth term, just like in the other theorem. >>: Can you go back one slide? One question. So the index M indicates the multiple. >>> Katherine Stange: Yes. >>: Then where is the sequence, it's referring also to one fist ->>> Katherine Stange: Yes, exactly. You pick a fixed curve and pick a fixed point, then you get a sequence of integers. So for every point and curve. Okay. Yes. So we're picking a particular elliptic curve in a point and looking at an associated sequence. If you have some vertex in your graph, then you get an arrow from that, from N to N times P, if P divides that element of the sequence. And you also get some extra arrows. You get N going to N times D where D is what I've highlighted in red there, an aliquot number. This is not aliquot in the classical sense. We decided to call these aliquots for elliptic curves. I'll explain what that is in a second. But I wanted to highlight the relationship to the previous theorem. Here we get when you divide the term and then some extra special stuff. The special stuff for these sequences is aliquot numbers. And then parts 3 and 4 here are sort of the converse. So if you have an arrow going from N to N times the prime, then it did usually come from either dividing the term or being aliquot number, the prime itself. So when I say good "reduction here", what I'll talk about in the rest of the talk, I'll be talking about taking an elliptic curve, defined over Q, and reducing it modulo P and getting something defined over FP. Then the question is whether you still get an elliptic curve. If you do, then you call it good reduction. You could get something singular, and that's called bad reduction. And the last one here says, if you have an arrow with a composite multiple, then in certain cases you can say that's an aliquot number. It's not a full converse to this. So when we were looking at this, we came up with these aliquot numbers naturally from working on that project. Now I'm going to forget that project completely and define what -- move from the definition of an aliquot number to something just talking about elliptic curves that's a question in and of itself. What do I mean by an aliquot number? So if you have some primes of good reduction, all distinct primes, and the first index which PI appears as a devisor is PI plus one. For each link in the entire loop. So including that the first index at which PL appears as a devisor is P1. So moving right around. Then the product of those things is called an aliquot number. And the important piece of the elliptic curves that you see in the sequence is that if your prime divides the nth term of the sequence, that means that the nth multiple of the point to which the sequence was associated is zero mod P on the reduced curve. So where the primes appear in the sequence is telling you the order of the point. So, in particular, if you just have the order of your curve being prime, then your point has to have, if it has nontrivial order, it has to have that order. So you can get aliquot numbers. You can build such things by finding elliptic curves that when you reduce mod the ith prime have a number of points equal to I plus the first prime. And the special case of this where there's just one prime, and it references itself, is where the number of points when you reduce mod P is actually P. And that's already got a name. That's an anomalous prime. That's in particular an aliquot number by this definition here. So you can forget all of the stuff about sequences and just ask the question for pairs to start with. So when can you have two primes so that when you reduce modulo one of them the number of the points on the reduced curve is equal to the other? And so this is in vague analogy to the classical number theoretic meaning of amicable numbers. But not in any meaningful analogy. Okay. So here's a couple examples. So these were just searches. We took a couple of curves, a small conductor, and looked for such pairs up to 10 to the 7. And in one example we found one pair. In the other example we found four. And so this is a little background slide for those who want a little bit more about elliptic curves. If we have an elliptic curve over a finite field, let's say FP, then we define the trace of Frobenius to be the difference between the number of points on the reduced curve and P plus one because the reduced curve should have about P plus one points, but there's some variation. The variation can be bounded. We call this the Hasse interval in which that thing called the trace of Frobenius can lie. And we have a theorem of during that says that everything in that Hasse interval can be obtained if you pick the right curve for a particular prime. And also we have some conjectures which govern the distribution within that house interval, as you vary over different curves. So we can ask about -- we can ask about the number of amicable pairs up to a given bound. So this is for a particular curve E. So I'm giving this counting function Q, which is indexed by our particular elliptic curve E, and the first question is how does this grow with X. How many amicable pairs do you expect there to be. And the question is sort of a strange one. You don't really expect amicable pairs to occur for any reason other than just kind of randomly when you ask the question. Right? But it turns out this kind of has an interesting answer. The second question here is maybe we can relate it to a counting function which has already been studied, which is, I'll call N, which is the number of primes so that the reduction is prime. If you're going to have an amicable pair at all, then in particular when you reduce modulo P you have to get some other prime Q. So we can count the number of primes for which there's prime reduction up to some bound and we could maybe compare the two counting functions. So here's a sort of -- sorry. I put the counting function up at the top just for reference there. And there's a conjecture due to Koblitz, which gives a conjectural growth for this with respect to the bounds. X over log X squared. And the constant there, he actually gave some conjectural values for that context in different cases and Zywina gave some corrections to that portion of the conjecture, so I stuck his name up there as well. And, in particular, the conjecture is that that constant is 0 when there's only finite many primes, obviously, but that's the only time when it's 0. You could have, for example, you could have, over your original curve over the rationals, you could have some torsion and that torsion is going to survive reduction mod P and divide the order of the reduced curve all the times, in which case this you wouldn't expect to be growing. So here's a sort of vague heuristic for how many amicable pairs you should expect. We'll do it in terms of probability. So the probability that P is part of an amicable pair, well, first of all, you have to reduce modulo P and get some prime order. Call that Q and then reduce mod Q and see if you get back P. Let's just assume those two things are sort of independent. If that's the case, then we can analyze each of those probabilities separately. And turning around the conjecture of Koblitz, we get that the probability that when you reduce mod P you get something prime should be somewhere around log P. And we also have that the probability when you reduce modulo in some particular prime Q you get the particular value you had in mind which is P. Should be about 1 over root Q, which is about the size of root P. P and Q here, if you're going to be an amicable pair, they have to be about the same size, because they have to be in each other's Hasse intervals. >>: Estimates are up to like order ->> Katherine Stange: I'd have to look. I'd have to go back. I don't remember. >>: [inaudible]. >> Katherine Stange: I think that, yeah, I think that's an unusual notation that we used in that. It means that there's a constant in both directions, I think. Yeah, I think the referee said that's not normal notation for this. >>: I think it's usually like the smile and the frown, that one. >> Katherine Stange: Yeah, that's a more usual notation. So putting these two things together, then, we should have the probability that P is part of an amicable pair should be around 1 over root P log P. So we turn that back into a growth function. And we get something like that. The number of amicable pairs should grow something like root X over log X squared. You remember the conjecture from Koblitz was X over log X squared. So this is smaller which you would expect. So the first conjecture then is, well, let's just put this together. There's the counting function for amicable pairs, assuming you get infinitely many primes so you have prime reduction, because otherwise you're not going to be able to count more than finite many pairs. And we expect it to grow like that, where these constants should somehow depend on the curve. But who knows how. So I think that if somebody asked you this question about amicable pairs, you might do this heuristic. Come up with this growth rate and be like, okay, well, probably can't prove this and probably that's about the story. It kind of happens randomly, the amount of time you expect it to happen. I guess we can check against a little bit of data. But we don't have very many pairs to count. So for this particular curve, and thanks to Andrew Sutherland, who extended our calculations beyond what we could do and none of this was done with Sage as far as I know, we can count the number of pairs, and we only get even 117 up to the 10 to the 12. So we can compare our counting function to root X over log X squared. And I'm not sure that the data is very convincing. We can compare it to log X there. This should go to like a half or something. Not very convincing. So we just don't have enough data to even know. But let's look at another example. So these are the two that I had on the first time I put up some examples. So up to about 10 to the 7 we get one in one case and four in the second case, but for this curve down here we get 5,578 in the same range, of which I have not listed them all. So that curve, you may, if you're familiar with elliptic curves you might notice that's a curve with complex multiplication. So something special might be happening in the case of complex multiplication. So I'll just tell you what complex multiplication is. Usually if you look at the endomorphism with your elliptic curve, it's usually isomorphic to Z. You have an endomorphism given by multiplication by M for each possible integer M. But sometimes your endomorphism ring is the order of a class number one in some quadratic imaginary number field. Sometimes you get curves that have more interesting endomorphism ring and those are called curves with complex multiplication. So we can actually prove something. We can say suppose we have a curve with complex multiplication. And suppose we have the complex multiplications even by this field joined by minus D, some order in that field. And let's throw out the J equals 0 case, that particular case. And let's suppose P and Q are primes of good reduction. So not too small. And suppose we already know that when we reduce mod P we get a prime order Q. The question is, does Q reciprocate, does it return the love? Right? When you reduce mod Q do you still get P? If so, then that would be an inamicable pair. In fact, when you reduce mod Q, two things can happen. Either you do in fact get P back and you have a pair. In this case we have a pair or you can get this other quantity. If you're familiar with this sort of thing, you'll probably recognize this formula. What this is saying actually is the nontrivial quadratic twist over the finite field has P points. So you always get E and E tilde and one of the two has P points. So I'll quickly give you the proof. It actually fits on a slide. First of all, we're interested in, going back to the statement here, we're interested in getting prime reduction. We know when you reduce mod P, you get Q. So we're going to throw out curves with two torsion. Because those will survive the reduction mod P and we won't get prime order for the curve. So that leaves that D is congruent to 3 mod 4. There's only finitely many different cases. So I forget how many this leaves. But anyway, when we look at P, it has to split. This is because the inert case corresponds to supersingular reduction, and in supersingular reduction the number of points would be in particular even. So P splits. And if we look at the number of points, so pick -- so P splits back P in its conjugate. If we look at the number of points over FP, then this is just the usual formula. The number of points in the curve is P plus 1 minus the trace of Frobenius where the Grossencharacter gives you the Frobenius as an element of the order number field that you're identifying with your endomorphism ring. So this expression up here in number three there, that's the norm of 1 minus psi of P. So any way, in particular, we know this reduction is Q. This reduction modulo P is Q. That says Q is the norm of something in this order. That means that Q itself splits. And then we can look at the norm of psi of Q, which is also Q. So that means those are two different things which have the same norm. Now, up to picking which of the Q or Q bar we want here, that means that those two things are the same up to a unit. And in all the cases that we're in, because we threw out the J equals 0 case, it turns out that the units available to us are just plus or minus 1. So then you can just calculate the trace of Frobenius for Q, and in each of the two cases you get two different values and they correspond to these two cases. It comes from the complex multiplication structure which is this rich structure we have in the case of CM curves. So let's look at the data. So this is -- these are the different possible orders for the endomorphism ring. D was Q adjoined root minus D. And F there is the index of the order. That's the bound and those are the number of pairs. This is just the counting function up here. And then let's actually look at the ratio of the amicable pair counting function over the counting function for prime reduction. So what the theorem said was if you already have prime reduction then there's two possible cases. And one of those cases corresponds to you being an amicable pair. So you'd expect this to go to a half, except that the pair counting function has two primes for each pair. You actually expect it to go to a quarter. And here it does look, I'd say, at least encouraging. Okay. So that looks good. It looks like basically what this table is saying is that in the statement of theorem, going back for a second, it looks like these two cases really do sort of happen equally often, which would be your first guess, I suppose. So here's the conjecture we get to put together, we have the not complex multiplication case, which is what we had before. And if it has complex multiplication, then we expect it to go to a quarter of the counting function for prime reduction. Okay. So you could ask about aliquot cycles. You could ask about more than just pairs. So suppose you have a whole sequence of primes such that when you reduce modulo one of them you get the next one, all the way around and coming back to the beginning. So here's a couple of examples. This one was found just by searching. And this one shows signs of being constructed. Take a bunch of primes that fit inside each other's Hasse intervals so they at least could plausibly be a cycle and then you can construct such a thing. Actually it's not too hard to do something like this. First, make sure they all satisfy their sizes are good enough so that the Hasse interval condition is verified. And then for each prime you can find by doing some elliptic curve that has the correct reduction over FPI. So you get one curve for each of them, but of course then you can use Chinese remainder theorem to get some curve that reduces to all of those, that's why you get these giant coefficients. That's how we made that example. Now, in the CM case, it's actually interesting that you don't get any -- in the CM case, we had lots of pairs but you don't get any longer cycles. That's again from that same CM theorem we had before. So supposing you were trying to make a long cycle. Long meaning three or more primes. Well, we know from that theorem that there's two choices in the reduction at each stage. So when you put those together, you get a recurrence relation telling you what the next prime could be based on the previous two. So if you put together the recurrence relation, you can get a general term. You can find out in fact that each term that P1 is P2 plus some positive amount. L is the length of the cycle. If L is at least 3, this is some positive number. And it's P1 is P2 plus some positive amount. P1 is bigger than P2 and similarly P2 is bigger than P3. P3 is bigger than P4. You can see where I'm going. It's like the extra staircase. You can't come around and be bigger all the way around your cycle. Come back to where you started. You can't make bigger cycles in the CM case. All right. So the last -- the case that I left out of the CM theorem was the case of the J equals 0 curve. So this curve has a little more interesting number theory. So I'm going to look a little more closely at what happens. In fact, it turns out that there's a really interesting purely elementary number theory question that comes out of this, which I'm still looking for an answer to. So here we're working with Q ajoins square root minus 3 which has a sixth root of unity. That's why the previous proof there required that the units were just plus or minus 1. So here we have more units, the unit group has six. In fact, the ring of integers can be made by just joining that [inaudible] there, and this has a nice property that the unit group was six elements. If you reduce modulo three, they each go one each into the invertible classes. So that's an isomorphism. We can also talk about the sextic residue symbol when we're working down here, defined down here. It's quadratic residue quadratic cubic symbol. In particular, for any element you're getting some sextic root unity as you see. So this is the number field we're working with for the J equals 0 case, for the endomorphism ring. And so we get a similar theorem to the CM theorem that we had for the other CM curves. Now, I've stuck some of the proof into the statement here because we can give explicitly what the unit is. So in the CM proof, I just said there's some units either plus or minus 1 in the middle of that proof. And here we can actually describe what that unit is. Going back over the setup here, we have some elliptic curve. For J equals 0 we can write Y squared equals XQ plus K. And suppose that P and Q are primes of good reduction. P is not too small. And when you reduce modulo P you get Q points. Then as in the proof before, P has to split and you can take Q to be the ideal generated by 1 minus the Frobenius. So then Q also splits. Then we have these P and Q to talk about, related by a unit. Just as in the previous proof. Now that unit is actually given in terms of the sextic residue symbols of K, just from the K as in the equation for the curve. So if this unit comes out to be one, that's the case where we get an amicable pair. So if we already had when you reduce mod P we get Q, we also get, when you reduce Q, you get P, when this thing comes out to one. So there's six -- the moral of the story, there's six possible options for this unit. And one of them is the case of an amicable pair. So you would expect that a good first guess, just like in the other CM case, is that these six units occur evenly often. So you would get the relationship between the payer counting function and prime reduction counting function would be 1/12th, counting for the fact there's two primes in a pair rather than one-sixth. Okay. So just a couple of extra remarks about this. Just as in the original CM case, those six different units correspond to twists. So that's six different possible values also just like in the statement of the other CM theorem that you could get for reduction mod Q. So you can state it that way if you'd like. The reason we can actually give this explicitly is from the theorem of CM we know Frobenius as a sextic residue. So here's the data. We've got different values of K across the top. We expect the ratio of Q over N to be going to about 1/12th. And it looks suspicious. For K equals 2 we seem to be getting a quarter again like before. For K equals 5, it looks reasonable. And the other one it's not clear what they're doing. So something more interesting is happening with this unit in this case. So then the question is just to take a look at this unit and see if we can analyze what we would expect it to be doing. If it's not going to be distributed evenly as -- distributed evenly for different P, than for a fixed K, then what do you expect it to be doing? So let's look a little more closely. These are the six possible values. Here's a giant amount of data. So this corresponds to the unit being one. And this corresponds to the unit being minus one. And these are the other four possibilities for that unit. And down here we have the different values of K and then these are just the frequencies of falling into those six different possible cases. So I think looking at the data here, well, what I noticed, and then Joe wasn't so convinced, was that it seemed to me that 1 and 2 were always a little bit more than the other cases in each of them. But it certainly seems as if these two columns are about the same as each other. But then they differ somehow from the other four cases. Look at the data. On some of them it's really clear. For example, for K equals 2 you don't even get those other cases and those two cases seem to divide about evenly. And for this one, K equals 3, for example, it also looks fairly clear. The other one not so sure. Certainly seems to depend on K a lot. Okay. All right. I don't know how I'm doing for time. Oh, there's a clock at the back. Okay. So there's just the pieces of the info for our number field up at the top there. And we have cubic reciprocity, which I'm going to use in a moment. So I thought I'd state it nicely. This is my favorite way of stating quadratic reciprocity, which is to think of the primes, instead of being just the positive primes, think of the primes as minus 3-5, minus 7, minus 11 and so. If you put the signs on the primes, so they end up being 1 mod 4, then you can state quadratic reciprocity particularly nicely. This is how I stated cubic reciprocity. I've chosen the representative for prime ideal so that it's 1 or 2 mod 3. So we have this unit that we'd like to analyze. Okay. So this is what it looked like from the statement of the theorem. So from cubic reciprocity we can flip numerator and denominator of course to check all the details and whatnot. So there's a little bit missing. But we can turn it into this. And so here I've got sextic residue. But I'm just going to ignore the quadratic part and just put a plus or minus in front. I'm really interested in the cubic part. The reason for that is when we looked at that table of data, looked like plus or minus 1 would be hitting pretty evenly. So it's really the cubic part which is the mystery. So changing this around, we get this symbol here. Okay. So we're interested in this symbol where M is our Frobenius for different primes P. So a first guess might be that the Frobenius ranges evenly through whatever elements it would be allowed to be as P varies. So you could take all of the M for which M times 1 minus M is invertible. And then you could look at how many of those actually give this cubic residue coming out to 1, and you could ask whether that's actually a third or that's some other ratio. And when this comes out to 1, that corresponds to the case where this thing becomes plus or minus 1, and we're in those first two columns. We seem to be in the first two columns more often. Is that simply because there are more elements where, more elements M in the residue here, the residue ring, that M times 1 minus M is actually from, has cubic residue 1. So then the conjecture would be that we expect this counting function ratio to go to that ratio where we have a 4 counting for the fact that now we're allowing plus or minus 1 and we're still counting two primes in a pair. So don't worry about the 4. Really we're interested in this ratio of MK star over MK. Where MK star is the one where this is cubic residue 1 and MK is just all the things where that numerator is invertible. So this becomes a purely number theory question. Now we've sort of forgotten about the elliptic curve completely. We're just asking for different Ks how can you count this. So here's an example of how we can count it when K is congruent to 2 mod 3. If we want to look at when M times 1 minus M is a cube, cubic residue 1, then -- one way to do it would be to look at it might be to at the associated curve, Y times 1 minus Y is X cubed, how many points are there on that. That just happens to be a J equals 0 elliptic curve. I don't think this is for a reason. But it just happens to be. So we have then that this curve E related to our problem for K -when K is a prime congruent to 2 mod 3, this is actually supersingular and counting the points is really easy. So if we want to count the points over K, that's actually FK squared is the residue we're interested in. So has K plus 1 squared points. It's real easy to count the points over supersingular curve. If this wasn't supersingular, if we had 1 mod 3, then we would have some trace of Frobenius in here instead of a fixed number in terms of K. Okay. So then we just want to count the points on that that correspond to solutions. So we have to throw away the points infinity 0 and 01 because they don't really correspond to solutions for what we're looking for. We want the numerator to be invertible we have to throw away Y as 0 or 1. So we take away those three points. Now we have K squared, K plus one all squared minus 3 and of course we're triple counting because X can be three different cube roots for each Y. We just want to count the Ys. So we divide by 3. That's the number of solutions to the problem then. Okay. So number of residues. Not equal to 0, 1, having that be a cube. And so that lets us count those things. Okay. So the question -- so this particular case fits on a slide. But all the other cases, as we went, just got worse and worse and horribler and horribler. So, of course, if, as I said, you can get traces of Frobenius in there that you have to be counting somehow. You have the strange thing that happens which is sometimes your Frobenius doesn't want itself to be a quadratic or cubic residue. Sometimes you want to count a slightly different thing than what we were just counting. When you do the point counting, when it's avoiding a cubic, quadratic or cubic residue, then you end up having to go to some other curves, not necessarily just that elliptic curve that we were looking at. And it ends up you can break up K for K prime up into different cases mod 36. And you can actually work it out but the calculations were so messy we actually had the computer do the computations in the end. All the simplifying and stuff. And if K splits, of course you have to do the two cases separately and combine. If it isn't prime you have even more complicated stuff to look at. So I won't ->>: The relationship between these curves has to get started with? >> Katherine Stange: So there isn't really. The only reason I'm bringing this curve into it is because this is the numerator of a cubic residue and I want it to be a cube. So when you want it to be a cube but you also don't want this to be a quadratic residue or a cubic residue, then you can count the number of points on that other curve and then take away the ones that would fit on this with N equals 2. And so, okay, but basically we turn it into a number theory problem and then ask that problem naively and forget the original curves to do with the pairs. So those computations are a challenge. And now I'm putting up the guts of the actual answer in case anybody wants to actually see it written down. So basically instead of the MK which was just -- there's four cases here. Here's the second slide. The last case here, where K is congruent to 3 mod 4, or K, the prime factors of K are not congruent to plus or minus 1. This means each prime factor. Then the MK you want is just the original one. So just the fact where those are invertible in the residue. And then for these other ones you have cases where it might not be a cubic residue, might not be allowed to be a cubic residue or allowed to be a quadratic residue. It just depends on the behavior of K mod 4 and mod 9. And then again MK star is just all the things in MK, but where that's a cube. So that's the like technical details to writing this out. And then once you have those definitions we expect the limit going back to the amicable pairs question we expect this limit to be a quarter of the limit of those counting functions. And that doesn't even deal with the case when K is not co-prime to 6. So here's what it comes out to when you actually count all these things. Okay. It took pages and pages in the paper. And then it simplifies to these simple polynomials in K. So when you do the computations you're counting points on all of these curves, on these Jacobeans, these crazy curves, you do all this crazy stuff, traces of Frobenius are showing up, residue are showing up. They cancel. Somehow you get polynomial K in the end. So when we count these things, we get one-sixth plus a half of whatever polynomials in the different cases. And so in the one case that I showed that worked out when K was 2 mod 3 N prime, it wasn't maybe surprising it came out to be a polynomial in K. But for these ones it really is because the computation, there's no reason to expect to have all that stuff simplify. I feel like there should be some real good reason for this that I'm just not, I don't know what it is. And it's a purely -- you can state this problem purely in the case of as a number theory problem, and if the answers are so simple I feel there should be a good reason. And obviously there's not a good reason. So finally here's some data for this conjecture. So we went and computed these, what we expect the conjectural limit to be. And for each K it's just some rational number. So here are those rational numbers for each K. And these two columns can bear conjecture and experiment. And I think it's pretty convincing that we seem to have discovered that the Frobenius is ranging through at least those allowable classes according to the law, according to the way described. So it seems plausible. And fortunately with these curves we can get lots of pairs. This is the number of pairs. That gives you an idea how robust the data is. So that's up to 10 to the 8. So seems reasonable. So these are the final suggestions, ideas, questions, I don't know, that might come out of this project. So, first of all, this first remark is just what I was saying a second ago. These things I'm computing, the number of classes where this thing is a cube, are coming out to these very simple answers, and I don't know why. So one could ask that question more generally. So we had M times 1 minus M, but what happens if you have some polynomial in terms of M up here. I don't know what's known about that question in general. And so I mean I'm new to Sage, but it seems like we could ask about what Sage can already do with respect to these computations we didn't use them in Sage we used the computer PARI to do the computations because they were so messy and could doing these sorts of computation provide any insight and give some nicer answer to why these things came out nicely, or is it a special case for M times 1 minus M somehow. Another question that comes out of this is to think about this as a dynamical system. So going back to the pairs, it's sort of like taking reduction modulo P is something you could iterate. So you take reduction mod P with Q and you take it with Q, get back to P and you have a two-cycle. We'd have to define it more generally, because what if you reduce mod P and you get some composite number. What's the next step? What's the function you want to iterate. That's reduction mod P for primes. So one idea is to define, instead of your trace of Frobenius only for P to define it for composite numbers in terms of the L series. And a student of Joe Silverman, Sahinoglu, is working on this. So we shouldn't steal her problem. But there might be other ideas for how to think about this as a dynamical system. This might not be the right dynamical system. >>: I think it's weird. N plus 1 minus N seems a little suspicious. Shouldn't it be like -- NF times Q, seems to be P minus 1 AP, Q plus 1 minus AQ, the cardinality of either ->> Katherine Stange: Yeah, I have. >>: Mod NC. >> Katherine Stange: I have the same thought that this is suspicious. Yeah. It's fishy. Certainly there are other alternatives. It seems more natural to me. I also thought somehow ->>: Like the cardinality of E -- the cardinality of EZ mod. >> Katherine Stange: Exactly. But then what do you do -- well, yeah, so that's definitely something to ponder. Also one reason that we found this interesting was because if a prime is anomalous, it's actually a very special case. If you have an elliptic curve over FP where P is anomalous, it's susceptible to an attack, elliptic curve discrete problem. If you have an amicable pair on your elliptic curve, is there anything you get from that? We couldn't get anything, but it's an interesting question. And of course, I don't know, we could figure out other ways to construct pairs, cycles or better ways to search for them. We really did exhaustive searching just by reducing and checking what the order was. So, yeah, so, thank you. [applause] >>: Does anybody have any questions? >>: Sometimes when we look at these distributions, compare them with the certain matrix groups, how often those matrices are finalized, the probabilities end up matching up with that space? >> Katherine Stange: You're talking about this question here? >>: You were saying about that your probabilities where you have one-sixth plus something in K. Now you're saying is there any explanation for that. I was wondering, one place to look for that might be to look in to see whether any matrix groups associated with your problem and then trying to analyze the probability that those matrices have a certain value and eigenvalue. And so you think that the eigenvalue 1, for example, should occur about 1 over P time. It's actually a little bit off from that. So sometimes those -- sometimes that can give an explanation for something that ->> Katherine Stange: Oh, that's fascinating. Yeah, I'd like to find out more about that. I haven't thought about that. >>: So [inaudible] density theorem didn't pop up anywhere in your talk. Somehow feels like it should if you're trying to prove one of these cases but it kind of doesn't matter if we're just trying to find the conjecture. Why is that? >> Katherine Stange: Well, I think we did think about it a little bit. But I'm not sure if I have anything coherent to say about it. >>: And so in every possible case you have kind of a conjecture now but you don't have a proof of any cases at all? >> Katherine Stange: Right. All we've got is the proof that it's sort of restricted in the CM case. >>: You've given a conjecture in every case which is really nice. >> Katherine Stange: I think except for K not co-prime with 6. >>: Few cases J0. >> Katherine Stange: Yeah. >>: Are you going to ask the same questions over number fields as well, they cost everybody for the one CM curves, for example? >> Katherine Stange: Yeah, we did sort of note that the similar proof will work for CM over -- but of course there could be way more complications of this type that you get for the J equals 0 case. So saying something works means the same sort of proof works but the actual working it out, what the many cases are depending on the units and stuff, you know ->>: What about the non-CM case over certain number fields? >> Katherine Stange: There we only have a conjecture. We don't really have anything ->>: Conjecture there, easier, because it was easier to make if conjecture in the non-CM case. >> Katherine Stange: Yes. So I guess we should -- I don't think. >>: Make it easy. >> Katherine Stange: Yeah, I haven't had time to write it down. The paper was already like -- with all this computing for the J equals 0 case, 50 pages, we're like who is going to want this. The first referee came back and said it's interesting they don't prove anything. It's 50 pages long. [laughter] Which was slightly unfair. It's a lot of figuring out how to write down the right conjectures. >>: Regarding small jobs, so I've been talking some with [inaudible] into Sage. So that could be one of the projects. I don't think it would be too difficult to include in Sage. And then at least you might be able to replicate the tables you may be using -- he may be using small junk [inaudible]. >> Katherine Stange: Yeah. He sent us more data than I have, that's up here, but he was just using it, he thought it was good for testing. >>: Which is fast. >> Katherine Stange: Yeah, he got farther than we managed to do. We only got to here. >>: 13th to the ninth and small jobs a little farther, works in parallel. So it's really nice. >> Katherine Stange: Yeah. So it was really nice. >>>: Any other questions? Okay. Let's thank our speaker. [applause]