wu :: forums (http://www.ocf.berkeley.edu/~wwu/cgi-bin/yabb/YaBB.cgi)
riddles >> putnam exam (pure math) >> (a^2+b^2)/(ab+1)
(Message started by: BNC on Jun 10th, 2004, 6:29pm)

Title: (a^2+b^2)/(ab+1)
Post by BNC on Jun 10th, 2004, 6:29pm
After deciding not to attempt adaptation after 2AM, I post this interesting question in it's original form here (it's 4AM...). If I had to choose, I'd place it in "hard" (see background below).



Let a and b be positive integers , such that t = (a2+b2)/(ab+1) is an integer.
Show that t is also a perfect square.

For extra credit, characterise all triplets (a,b,t).



background

This problem is from IMO98 (so yes, it's solution is available on the web -- not sure about the "extra").

The problem selection committee (6 members, including G. Szekeres), who always try to solve the problems themselvs in order to get first-hand impression about the difficulty of the problem, couldn't solve it.

The committee then presented the problem to several number theory experts, asking them to solve it in 6 hours. They failed.

Finally, it was decided to use the problem in the 2nd day of the olympics (when the contestants have 4.5 hours to solve 3 questions). Out of ~200 participants that year, 11 solved the problem.


So, yes, I'd say it is "hard".
Or is it?

Title: Re: (a^2+b^2)/(ab+1)
Post by towr on Jun 11th, 2004, 12:50am
::[hide]
(x, x3, x2) or (x3, x, x2)
[/hide]::

Title: Re: (a^2+b^2)/(ab+1)
Post by BNC on Jun 11th, 2004, 12:56am
1. That's not a proof!

2. You get partial credit. It doesn't cover, e.g., the triplet (8,30,4).

Title: Re: (a^2+b^2)/(ab+1)
Post by towr on Jun 11th, 2004, 2:30am
I suppose it would have been too simple if that was the only sort of triple..
::[hide]I only found (8,2,4) to being with, and just guessed at the pattern (x3, x, x2) and after filling it in it worked..[/hide]::

Title: Re: (a^2+b^2)/(ab+1)
Post by Barukh on Jun 11th, 2004, 4:28am
Excellent puzzle, BNC!

2. Conjecture:
[smiley=square.gif][hide]
For every integer t, define a sequence {sn} as follows: s0 = 0; s1 = t; sn = t2sn-1 - sn-2. Then, triple (sn, sn+1, t2) satisfies the condition of the problem.

The conjecture is that these sequences cover all the cases.
[/hide][smiley=square.gif]

Of course, it’s not a proof…  ::)

Title: Re: (a^2+b^2)/(ab+1)
Post by BNC on Jun 11th, 2004, 5:30am
Briliant!

The solution I had requires using a couple of special cases, which your conjecture covers!

Now, all you have left is the proof (and, offcourse, the original problem's solution).

Title: Re: (a^2+b^2)/(ab+1)
Post by THUDandBLUNDER on Jun 18th, 2004, 6:46pm
Let t = (a2+b2) / (ab+1) be a positive integer.
And, as a = b => t = 1, WLOG let a > b

Then a2 - tba + b2 - t = 0.......[1]

If the roots of this quadratic in a are a1 and a2 then
a1 + a2 = tb........[2]
a1a2 = b2 - t.......[3]

Hence, if (a1, b) is a solution, then so is (tb - a1, b)

And from [3], if a1 > b then a2 < b

We thus get a strictly decreasing sequence {sn} of integers given by
s0 = a1
s1 = b
s2 = tb - a1
.................
..................
...................
sk = tsk-1 - sk-2


Title: Re: (a^2+b^2)/(ab+1)
Post by Barukh on Jun 19th, 2004, 2:13am
Yes, that's it! ... And the conjecture is correct.

Once you see the solution, it looks so simple, doesn't it?  ;)

What were all those number theorists and other experts thinking?  ;D

By the way, the problem wasn't at IMO98, it was at IMO88.

Title: Re: (a^2+b^2)/(ab+1)
Post by BNC on Jun 19th, 2004, 2:21am

on 06/19/04 at 02:13:25, Barukh wrote:
By the way, the problem wasn't at IMO98, it was at IMO88.


Sorry, typo...  :(

Title: Re: (a^2+b^2)/(ab+1)
Post by BNC on Jun 22nd, 2004, 4:34am

on 06/19/04 at 02:13:25, Barukh wrote:
What were all those number theorists and other experts thinking?  ;D


How does it show t to be a perfect square?

Title: Re: (a^2+b^2)/(ab+1)
Post by THUDandBLUNDER on Jun 22nd, 2004, 5:55am

on 06/22/04 at 04:34:34, BNC wrote:
How does it show t to be a perfect square?

I was afraid you would ask that.   ::)

Well, if we relax the condition that a,b > 0 (provided t > 0), we have
sk2 + sk+12 / 1 + sksk+1 = t for k = 0,1,2,3....

If sk+1 = 0 for some value of k, then t = sk2
Assume that there is NO k such that sk+1 = 0
Then, as the sequence is strictly decreasing, there must exist two consecutive terms which are of opposite sign.
Thus (a2+b2) / (ab+1) must be either infinite (if ab = -1) or negative (if ab < -1), contradicting the condition that t > 0




Powered by YaBB 1 Gold - SP 1.4!
Forum software copyright © 2000-2004 Yet another Bulletin Board