The problem of the title is one which seems a reasonable one to ask, but I can find no reference to it even in Dickson's History. This does not mean that the problem has never been considered - I just can't find any mention.
Consider first x2 + N2y 2 = b2 , and define w = b/x and z = y/x , so that w2 = N2z2 + 1, which has the obvious rational solution z = 0 , w = 1.
The line w = 1 + m z meets w2 = N2 z2 + 1 at z = 0 and z = 2 m / ( N2 - m2 ).
Define m = p / q, then z = y / x = 2 p q /
( N2 q2 - p2 ),
giving a parameterisation of
x2 + N2 y2 = b2
as x = N2 q2 - p2
and y = 2 p q.
Substituting these into x2 + N y 2 = a2, we find
|
|
Thus, the quartic is birationally equivalent to the elliptic curve
|
|
The three rational roots imply at least 4 torsion points, namely the pt. at infinity, ( 0 , 0 ) , ( N , 0 ), and ( -N ( N - 1 ) , 0 ).
Numerical evidence suggests these are the only ones, but I am unable to prove this. If this is so, then to get a non-trivial solution we need to have EN to have rank at least 1.
For example, the Birch and Swinnerton-Dyer conjecture shows that EN has rank at least 1 for N=74.
We find the following point on the curve h = 13686697 / 11664 , g = 116021624725 / 1259712.
This gives p = 3670525 and q = 46764 , and hence x = 1497444368329 , y = 343296862200, a = 3311108888329 and b = 25448063182921.
We used the BSD conjecture to estimate the rank for
2 £ N £
999 . The distribution of ranks is given in the following table.
| N range | Rank = 0 | Rank = 1 | Rank > 1 |
| [ 2 , 99 ] | 44 | 49 | 5 |
| [ 100 , 199 ] | 39 | 51 | 10 |
| [ 200 , 299 ] | 36 | 56 | 8 |
| [ 300 , 399 ] | 37 | 48 | 15 |
| [ 400 , 499 ] | 39 | 49 | 12 |
| [ 500 , 599 ] | 43 | 47 | 10 |
| [ 600 , 699 ] | 40 | 52 | 8 |
| [ 700 , 799 ] | 39 | 50 | 11 |
| [ 800 , 899 ] | 36 | 57 | 7 |
| [ 900 , 999 ] | 45 | 45 | 10 |
The solutions found for the values of N with rank greater than 0 can be found in the following file .
There are several values of N where the Birch and Swinnerton-Dyer
conjecture predicts a rank greater than 0 but where we cannot find a solution.
Currently, there are 73 values of N unsolved. Many of these curves have
predicted heights of several hundred which means we are unlikely to be
able to find a solution, so we only give the smallest 10 heights (updated
26 August 2004).
| N | Height | Information |
| 759 | 30 | |
| 938 | 37 | |
| 795 | 44 | |
| 602 | 48 | |
| 706 | 48 | |
| 844 | 48 | |
| 327 | 50 | |
| 397 | 52 | |
| 509 | 52 | |
| 515 | 53 |