Random matrix theory can be used to predict the frequency of rank 2 curves in a family of quadratic twists, but it does not seem to be able to tell us what the prediction is for quadratic twists of rank 3 or higher.
The reason that the RMT approach fails is that the height of a generating point is a factor in the formula for the central value of the derivative of the L-function of a rank 1 curve. We don't have a prediction for the distribution of the heights of the generating points, and so can't predict (within the collection of curves of odd rank) the distribution the critical derivatives of the -functions. Since the frequency of vanishing is determined by analyzing the tail of that distribution, we are unable to predict how often the derivative of the L-series is 0.
In [ arXiv:math/0010056] (to appear in Exp. Math.), Rubin and Silverberg show that for several infinite families of elliptic curves , the number of quadratic twists , , with rank at least is . (Assuming the Parity Conjecture, the same result holds with replaced by .) Some of these examples were previously obtained by Stewart and Top [ MR 95m:11055].
Rubin and Silverberg ask whether it is possible to find a hyperelliptic curve over whose jacobian contains copies of a fixed elliptic curve . This would give a lower bound for the frequency of quadratic twists of of rank at least .
Back to the main index for L-functions and Random Matrix Theory.