1
$\begingroup$

$T>0$ is a parameter.

Consider the linear Diophantine equation $ax+by=c$ where $a,b$ are coprime.

Suppose $a,b$ are of magnitude $T^{1+\epsilon}$ and $c$ is of magnitude $T^2$.

  1. For how many such equations we can expect $x,y$ to be of magnitude $T^{1+\epsilon}$ for a fixed $c$ and we vary $a,b$ coprime of magnitude $T^{1+\epsilon}$? Call such solutions Fundamental.

Such solutions are also mininum normed.

  1. The usual way of solving such equations is to solve $ax'+by'=1$ and choose $x=x'c$ and $y=y'c$. But this does not provide a polynomial time algorithm to find $x,y$ of magnitude $T^{1+\epsilon}$ if there exists one. How do we find such solutions when they exist without using integer programming and directly using number theory?
$\endgroup$
3
  • 1
    $\begingroup$ doi.org/10.1016/S0747-7171(89)80025-2 describes an efficient procedure for finding all minimal solutions, in a general case of $n$ variables. $\endgroup$ Commented Mar 4, 2022 at 15:47
  • $\begingroup$ @DimaPasechnik Is the problem of counting all minimal solutions complete for any complexity class? $\endgroup$
    – Turbo
    Commented Mar 4, 2022 at 22:44
  • $\begingroup$ do you mean counting complexity classes? I don't know. $\endgroup$ Commented Mar 5, 2022 at 7:56

1 Answer 1

4
$\begingroup$

Solve $ax'+by'=1$, then take $$x = x'c - b \left\lfloor \frac{x'c}{ b} \right\rfloor$$ $$y = y'c +a\left\lfloor \frac{x'c}{ b} \right\rfloor$$

then we have $ax+by= ax'c +by'c = c$ and (if $b>0$ for simplicity) $0 \leq x < b$ so $$|x| < b$$ and $$|y| = \left| \frac{c-ax}{b} \right| \leq \frac{|c|}{|b|} + \frac{|a||x|}{|b|}< \frac{|c|}{|b|}+a\leq T^{1-\epsilon} + T^{1+\epsilon}.$$ So there is always a solution, and the algorithm to find it is clearly polynomial time.

We can optimize the norm by subtract one more copy of $b$ from $x$ if it helps.

$\endgroup$
4
  • $\begingroup$ $x'=\frac1a\in\mathbb Q$ (computed upto length of repeating decimals)? So $x=\frac{c}{a}-b\lfloor\frac{c}{ab}\rfloor$? $\endgroup$
    – Turbo
    Commented Mar 4, 2022 at 15:34
  • $\begingroup$ @Turbo I don't think that works. If $a = 10$ and $b=11$, $x'=-1$, not $1/10$. $\endgroup$
    – Will Sawin
    Commented Mar 4, 2022 at 15:38
  • $\begingroup$ Sorry I am confused $ax'\neq1$ then ($ax'=10\times -1=-10$). Is it solve $ax'+by'=1$ and not $ax'=by'=1$? $\endgroup$
    – Turbo
    Commented Mar 4, 2022 at 15:39
  • 1
    $\begingroup$ @Turbo Correct, we solve $ax'+by'=1$. $\endgroup$
    – Will Sawin
    Commented Mar 4, 2022 at 15:44

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .