Tartar Ab

Luc TARTAR to John MACKEY, September 30, 2009
******
Thanks for pointing out after I wrote the following note the web page where one learns about the history of
the problem and some references:
http://www.math.niu.edu/˜rusin/known-math/00 incoming/quad eqn
******
I have found something interesting by looking at all the solutions in positive integers A and B with
A B + 1 dividing A2 + B 2 .
Of course, one has the solution B = A3 or A = B 3 , but there are others, and there is an interesting
phenomenon, which must be already known, and it would be useful to check your source of the problem in
order to see if it is mentioned.
I assume that A ≤ B, and since A B + 1 divides A2 (A2 + B 2 ) = A4 + 1 + (A2 B 2 − 1) and A B + 1 divides
A B 2 − 1, one deduces that A B + 1 divides A4 + 1, so that B ≤ A3 (and B = A3 is a solution).
2
The Euclidean division gives B = A Q + R with 0 ≤ R ≤ A − 1, and one has 1 ≤ Q ≤ A2 . One observes
that Q (A B + 1) = (B − R) B + Q, so that A B + 1 divides A2 + B 2 = A2 + Q (A B + 1) + R B − Q, hence
A B +1 must divide A2 +R B −Q. Hovever, 0 ≤ R B ≤ (A−1) B < A B +1, and 0 ≤ A2 −Q < A2 < A B +1,
so that A2 + R B − Q must either be equal to 0 or to A B + 1; the first case implies R = 0 and Q = A2 , so
that B = A3 .
The second case means A2 +R B −Q = A B +1, so that R = 0 is not possible, and B (A−R) = A2 −Q−1
may be written Q = A2 − k B − 1 if one writes k = A − R, which satisfies 1 ≤ k ≤ A − 1; of course, one
must have Q ≥ 1, but since B = A (A2 − k B − 1) + A − k = A3 − k A B − k, one has B (1 + k A) = A3 − k,
and 1 + k A must divide A3 − k; then it divides k (A3 − k) = A2 (1 + k A) − (k 2 + A2 ), so that 1 + k A must
divide k 2 + A2 . There is a supplementary constraint, that if k 2 + A2 = λ (1 + k A), one needs A2 − λ to be a
2
multiple of k, since B = A k−λ , but the definition of λ gives B = λ A − k. Then, 1 + A B = 1 + λ A2 − k A, and
A2 +B 2 = A2 +λ2 A2 −2λ A k+k 2 = λ2 A2 −2λ A k+λ (1+k A) = λ (1+A B), so that the ratio λ is unchanged.
2
Then, B 2 −λ = λ2 A2 −2λ A k +k 2 −λ = λ2 A2 −2λ A k +λ k A−A2 , so that B A−λ = λ2 A−λ k −A = λ B −A.
Starting from a trivial solution (0, k), which corresponds to the ratio k 2 , the next value is k 2 · k − 0 = k3 ,
giving the solutions (k, k 3 ); the next value is k 2 · k 3 − k = k (k 4 − 1), giving the solutions k 3 , k (k 4 − 1) ;
the next value is k 2 · k (k 4 − 1) − k 3 = k 3 (k 4 − 2), giving the solutions k (k 4 − 1), k 3 (k 4 − 2) . The following
values are k 2 · k 3 (k 4 − 2) − k (k 4 − 1) = k (k 8 − 3k 4 + 1), k 2 · k (k 8 − 3k 4 + 1) − k 3 (k 4 − 2) = k 3 (k 8 − 4k 4 + 3),
k 2 · k 3 (k 8 − 4k 4 + 3) − k (k 8 − 3k 4 + 1) = k (k 12 − 5k 8 + 6k 4 − 1), etc. Of course, one finds values of the form
k fn (k 4 ), k 3 gn (k 4 ), k fn+1 (k 4 ), k 3 gn+1 (k 4 ), . . ., with fn , gn being polynomials of degree n satisfying simple
relations, namely gn = fn − gn−1 , and fn+1 (x) = x gn (x) − fn (x) for all n ≥ 0, with f0 = g0 = 1 (and one
may take g−1 = 0).

0 1 x y
If one considers the matrix M = then M = with z = k 2 y − x, then M sends
−1 k 2 y z
the conic section of equation x2 + y 2 = k 2 (1 + x y) into itself, since y 2 + z 2 = y 2 + k 4 y 2 − 2k 2 x y + x2 =
k 2 (1+x

4 2 2 2 2 2 2
y)+k y −2k x y = k (1+k y −x y) = k (1+y z). Starting from the point with integer coordinates
0
, one generates then other points of the conic section with integer coordinates, at least all those with
k
positive coordinates.
Of course the conic sectionis an hyperbola
for k > 1, and
fork = 1 one
has an ellipse, and one repeats
0 1 1 0 −1 −1
the same points with period 6, , , , , , .
1 1 0 −1 −1 0
It reminds of the generation of solutions of the “Pell” equation y 2 − d x2 = ±1 when d is not a square,
but I do not know if there is a general theory about for which hyperbolas with integer coefficients may be
dealt in such a way for describing their points with integer coordinates.
Of course, the method of descent which I used reminds of the ideas of FERMAT, and questions about
Euclidean rings, but I do not know more about that.
Actually, there is some kind of a multiplication structure behind, which comes out naturally.
1

x y
First, one notices that if u = and v = M u = with z = k 2 y − x, then y 2 + z 2 − k 2 y z =
y z
y 2 + (k 2 y − x)2 − k 2 y (k 2 y − x) = x2 + y 2 − k 2 x y, so that each conic section of equation x2 + y 2 − k 2 x y = `
is mapped into itself, and our particular case ` = k 2 is not significant from the point of view of linear
algebra, and the fact that k is an integer is not relevant for this computation. One is led to define a
“norm” N (u) = x2 + y 2 − k 2 x y, and √
the roots ω and ω 0 of ξ 2 + k 2 ξ + 1 = 0 play a role; ω + ω 0 = −k 2
0 −k2 ± k4 −4
and ω ω = 1, and they are and from the algebraic point of view it does not matter which
2 √
one is ω. Everything is then related to Z(ω), which is a little bigger than Z( k 4 − 4) (but one could
play with Q(ω) for questions of rational points). Writing (a, b) for a + b ω, the multiplication rule is then
(a, b) ? (c, d) = (a c − b d, a d + b c − k 2 b d), since (a + b ω) (c + d ω) = a c + b d (−k 2 ω − 1) + (a d + b c)ω; one
has a conjugation where (a + b ω)∗ = a + bω 0 and (a + bω 0 )∗ = a + bω, and it is compatible with addition and
multiplication, and then N (u) = u u∗ has the property that N (u1 u2 ) = N (u1 ) N (u2 ) for all u1 , u2 ∈ Z(ω).
The operation M is then the multiplication by η = (0, −1), which satisfies N (η) = 1.

Tartar Ab

Uploaded by

Copyright:

Available Formats

Tartar Ab

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Tartar Ab

Uploaded by

Copyright:

Available Formats

Luc TARTAR to John MACKEY, September 30, 2009

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.