T.R | Title | User | Personal Name | Date | Lines |
---|
1236.1 | 2 variables, 1 equation ? | AQUA::GRUNDMANN | | Wed May 09 1990 08:36 | 8 |
| Since this is too easy, I must be missing something about the question.
+ ___________
x = - \/ 1 + 92y^2
+ ________________
y = - \/ (x^2 - 1) / 92
|
1236.2 | More interesting to look for integral solutions? | VANDAL::STRANGEWAYS | Infested with sponge-cats | Wed May 09 1990 09:12 | 1 |
|
|
1236.3 | still too unconstrained? Oh... you want ALL the solutions!? | AQUA::GRUNDMANN | | Wed May 09 1990 10:37 | 1 |
| x=1 y=0
|
1236.4 | there's at least one interesting answer | AQUA::GRUNDMANN | | Wed May 09 1990 11:26 | 1 |
| a quick computer search yielded x=1151 y=120
|
1236.5 | | GUESS::DERAMO | Dan D'Eramo | Wed May 09 1990 14:22 | 4 |
| Yes, all integral (or just as easily, all non-negative
integral) solutions are desired.
Dan
|
1236.6 | | GUESS::DERAMO | Dan D'Eramo | Wed May 09 1990 15:29 | 5 |
| For what its worth ... sqrt(92) is 9 + something, where the
continued fraction representation of something is the
repeating block [1 1 2 4 2 1 1 18 ... ].
Dan
|
1236.7 | | GUESS::DERAMO | Dan D'Eramo | Fri May 11 1990 10:10 | 47 |
| If x^2 - 92y^2 = 1, then x/y is a pretty good approximation
to sqrt(92). This gives the idea of looking at the best
rational approximations of sqrt(92), and these come from
the continued fraction expansion of sqrt(92). If you learned
continued fractions as [a0 a1 a2 ...] = 1/(a0 + 1/(a1 + 1/(a2 + ...
________________
then sqrt(92) = 9 + [1 1 2 4 2 1 1 18 ...] where the overlined
block repeats. If you learned [a0 a1 a2 ...] = a0 + 1/(a1 + 1/(a2 + ...
________________
then sqrt(92) = [9 1 1 2 4 2 1 1 18 ...] where again it is the
overlined block that repeats. It is easier to use the latter
form here, so I will. Note they are reciprocals of each other,
so you just exchange the numerator and denominator to go from
one to the other.
Computing the successive approximations to the continued
fraction according to the formulas
x(n) = a(n) * x(n-1) + x(n-2)
y(n) = a(n) * y(n-1) + y(n-2)
yields
n a(n) x(n) y(n) x^2 - 92y^2
------------------------------------------------------
-2 0 1
-1 1 0
0 9 9 1 -11
1 1 10 1 8
2 1 19 2 -7
3 2 48 5 4
4 4 211 22 -7
5 2 470 49 8
6 1 681 71 -11
7 1 1151 120 1 (Eureka!)
8 18 21399 2231 -11
9 1 22550 2351 8
10 1 43949 4582 -7
11 2 110448 11515 4
12 4 485741 50642 -7
Notice how the successive rational approximations x/y alternate
being greater than then less than sqrt(92). Anyway, this shows
that (1151, 120) is a solution over the positive integers to
x^2 - 92y^2 = 1 (although we knew that already, from .4).
Dan
|
1236.8 | | GUESS::DERAMO | Dan D'Eramo | Fri May 11 1990 10:30 | 38 |
| Suppose now that (x,y) is a solution to x^2 - 92y^2 = 1.
This factors into (x + y sqrt(92))(x - y sqrt(92)) = 1.
Raising to the n-th power, you also have that
(x + y sqrt(92))^n (x - y sqrt(92))^n = 1. Now, if
you use the binomial theorem and separate the even and
odd powers of sqrt(92), you will see there is an A and B
(dependent on n) such that (x + y sqrt(92))^n = A + B sqrt(92)
and (x - y sqrt(92))^n = A - B sqrt(92). Then you will
also have (A + B sqrt(92)) (A - B sqrt(92)) = A^2 - 92B^2 = 1.
So given one nontrivial solution, we can generate an
infinite sequence of others. Let A(1) = x, B(1) = y.
Then (x + y sqrt(92))^(n+1) = (A(n) + B(n) sqrt(92))(x + y sqrt(92))
which multiplies out to
A(n+1) = x A(n) + 92 y B(n)
B(n+1) = y A(n) + x B(n)
Note: using n=0 for the exponent corresponds to the trivial
solution (1,0), and using n=1 for the exponent corresponds
to the solutiion (x,y). We need this first solution to start
generating the others. Here our first solution is (x,y) =
(1151,120). Thus our infinite sequence of solutions (A,B)
is given by
A(0)=1, B(0)=0
A(n+1) = 1151 A(n) + 11040 B(n)
B(n+1) = 120 A(n) + 1151 B(n)
For n = 0,1,2,3 this gives (1,0), (1151,120) and the new
solutions (2649601,276240) and (6099380351,635904360).
This gives infinitely many solutions. Are there any
others? :-) Would the A(n) for n > 1 have been found
if the continued fraction table in .-1 were continued?
Dan
|
1236.9 | | TRACE::GILBERT | Ownership Obligates | Fri May 11 1990 10:53 | 14 |
| > This gives infinitely many solutions. Are there any
> others? :-) Would the A(n) for n > 1 have been found
> if the continued fraction table in .-1 were continued?
You'd have seen those A values in the continued fraction
tables at lines n=7, 15, 23, 31, ... (note that the
values of x^2 - 92y^2 are cyclic with period 8).
The coefficients 1151, 11040, 120, and 1151 can be
directly derived by expanding the x(n), y(n) recurrences
in your previous note.
Does this yield all (positive) solutions? Yes, but I don't
remember why.
|
1236.10 | | GUESS::DERAMO | Dan D'Eramo | Fri May 11 1990 14:31 | 19 |
| >> Does this yield all (positive) solutions? Yes, but I don't
>> remember why.
See .0 ... if you're a mathematician you'll remember why
within the year. :-)
My vague recollection of this part of the proof is that
you bracket a solution (X,Y) between (A(n),B(n)) and
(A(n+1),B(n+1)), i.e., A(n) <= X < A(n+1) and B(n) <= Y < B(n+1),
then show it must be (A(n),B(n)). But the proof was only
for the special case where the "92" differed from a square
by one (I forget whether it was a square plus one or a square
minus one). Anyway that's the approach I would try first.
Looking at the structure of the units of Z[sqrt(92)] would
come next.
Dan
p.s. .7 and .8 were from memory, not from solving it :-)
|
1236.11 | | GUESS::DERAMO | that Colorado Rocky Mountain high | Wed May 30 1990 14:28 | 9 |
| The original usenet signature has now expanded to:
>> --
>> Chris Long, 272 Hamilton St. Apt. 1, New Brunswick NJ 08901 (201) 846-5569
>>
>> "A person who can within a year solve x^2 - 92y^2 = 1 is a mathematician."
>> Brahmagupta, 650 AD
Dan
|
1236.12 | | GUESS::DERAMO | Dan D'Eramo | Mon Nov 12 1990 23:04 | 7 |
| re .0
>> "A person who can within a year solve x^2 - 92y^2 = 1 is a mathematician."
Less than six months to go. :-)
Dan
|