Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

By Fermat's last theorem, the equation $u^3+v^3=w^3$ has no solutions in positive integers $u,v,w$. Now consider the following variant : call $\rho(x)$ the distance between $x$ and the nearest integer, for any real number $x$ (thus $\rho(3)=0,\rho(3.2)=0.2,\rho(3.5)=0.5, \rho(3.9)=0.1$ etc).

An "approximate" version of the Fermat equation is to ask for $\rho({( u^3+v^3 )}^{\frac{1}{3}})$ to be arbitrarily small. A trivial way to achieve this is to make one of the variables very small compared to the other, say $v$ very small compared to $u$, so that ${( u^3+v^3 )}^{\frac{1}{3}}$ is very near to $u$.

It is therefore natural to ask if there is an absolute constant $C>0$ such that $\rho({( u^3+v^3 )}^{\frac{1}{3}})$ can be made arbitrarily small with $u,v$ positive and $u \leq C v, v \leq C u$ (so that neither of $u$ or $v$ dominates).

Can a (reasonably) explicit $(u_n,v_n)$ sequence be found, such that $\rho(u_n^3+v_n^3)$ tends to $0$ as $n$ goes to infinity and $u_n \leq C v_n, v_n \leq C u_n$ ? "Closed-form" formula would be the best, of course, but even a simple recursion would be nice.

share|improve this question
    
Why $\rho(3)=3$? I think that $\rho(3)=0$. And do you actually mean $\rho(\sqrt[3]{u^3+v^3})$? Then just consider $u_n=n, v_n=1$. –  Nurdin Takenov Jan 23 '10 at 12:12
    
Yes, indeed. I'll correct that. –  Ewan Delanoy Jan 23 '10 at 12:14
1  
Or if you want u_n and v_n tending to infinity, u_n=n^10 and v_n=n will do. It's hard for me to see whether this can be turned into an interesting question. –  Kevin Buzzard Jan 23 '10 at 12:57
    
I'll make one last try, and delete this question if it turns out to be trivial again. –  Ewan Delanoy Jan 23 '10 at 13:27
1  
Let p/q be a close approximation to 2^(1/3), say p/q = 2^(1/3) + e with |e| < 1/(sqrt(5)*q^2); there are infinitely many of these by Hurwitz's theorem. Letting u=v=q, we have (u^3+v^3)^(1/3) = 2^(1/3) * q = p - qe, so the error qe is at most 1/(sqrt(5)*q) in magnitude, and as q goes to infinity the error goes to 0. –  Steven Sivek Jan 23 '10 at 16:12
show 2 more comments

2 Answers

up vote 5 down vote accepted

The condition you want is very weak, and there are clearly many accidental solutions.

You can add severe restrictions and still find many solutions. For example, (as Steven Sivek pointed out) you can force $u_n = v_n$ and then by the theory of simple continued fractions, there are infinitely many $p_n/q_n$ so that

$|\frac{p_n}{q_n}-\sqrt[3]2|\lt q_n^{-2}.$

Then $|p_n - q_n \sqrt[3]2| < 1/q_n$ so $\rho(\sqrt[3]{(q_n^3+q_n^3)} )$ decreases to 0 rapidly.

This might not be viewed as explicit since the coefficients of the simple continued fraction for $2^{1/3}$ don't have a clear pattern, although you can define them recursively if you allow functions like $\lfloor1/\rho\rfloor$.

If you can find $x^3 + y^3 = z^3$ in numbers with known simple continued faction expansions, then you may be able to use this to construct closed form families of examples. For example,

$(5-\sqrt{6})^3 + 3\sqrt{6}^3 = (5+\sqrt{6})^3$.

Convergents of $\sqrt6$, $p_n/q_n$, satisfy $|\frac{p_n}{q_n} - \sqrt{6}| < 1/q_n^2$.

Then $(5q_n-p_n)^3+(3p_n)^3 = (5q_n+p_n)^3 + O(q_n) = (5q_n+p_n)^3 + o((5q_n+p_n)^2)$,

so $\rho(\sqrt[3]{(5q_n-p_n)^3+(3p_n)^3}) = o(1).$

For example, $\sqrt[3]{(5\times 881 - 2158)^3 + (3 \times 2158)^3} = 6552.99916...$

Since $\sqrt{6} = [2;2,4,2,4,2,4...]$ which is periodic, you can construct a closed form expression for the convergents $p_n/q_n$.

share|improve this answer
add comment

Elkies' paper "Rational points near curves and small nonzero |x^3-y^2| via lattice reduction" is relevant to your question: he discusses the efficient computation of solutions to |x^3 + y^3 - z^3| < M, and remarks that there are conjectured to be infinitely many of these "near-Fermat" triples. Indeed, his argument is very general and is not restricted to algebraic curves: he finds, for instance, that

2063^pi + 8093^pi = 8128^pi + .019....!

share|improve this answer
6  
I saw this question and thought, "Oh, this is a good chance to mention this cool paper of Elkies," and then I saw this answer and thought, "Oh well, someone already mentioned it," and then I realized it was me. –  JSE Dec 22 '10 at 1:54
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.