Given a very large integer $n$, I wish to find integer multipliers $m$ such that the product is nearly a perfect square. Ideally this should be easy to compute and output a series of improved approximates. We'll define a quality measure as "nearly a perfect square" as minimizing the quantity $\sqrt{nm} - \left \lfloor\sqrt{nm}\right \rfloor$.
As an example with a small $n=87654321$ a useful series of better and better $m$ values can be brute force evaluated by exhaustively testing $m$ values starting at 1. Notice how the fractional remainder of the square root decreases monotonically as it finds better multipliers:
87654321 * 1 = 9362.3886375219 ^2
87654321 * 3 = 16216.1328003936 ^2
87654321 * 6 = 22933.0749355598 ^2
87654321 * 26 = 47739.0023565638 ^2
87654321 * 249 = 147736.0007885688 ^2
87654321 * 2766 = 492394.0006600405 ^2
87654321 * 3469 = 551428.0003309589 ^2
87654321 * 5751 = 710000.0000500000 ^2
87654321 * 15967 = 1183037.0000160604 ^2
87654321 * 44374 = 1972200.0000136902 ^2
87654321 * 156685 = 3705957.0000048568 ^2
87654321 * 392125 = 5862717.0000030706 ^2
87654321 * 790738 = 8325359.0000010207 ^2
87654321 * 1555754 = 11677695.0000003850 ^2
87654321 * 1856354 = 12756075.0000003520 ^2
My question is how to generate this, or a similar, series of "good" $m$ values. I suspect the answer has to do with a continued fraction representation of $\sqrt n$ but I have had no luck turning that hunch into a working method.
An answer I'm not looking for is solving for the smallest $m$ that produces a perfect square. That method is to completely factor $n$, and forming $m$ as a product of all of the factors of $n$ with odd multiplicity. That's a great result but factoring $n$ is impractical for very large (hundreds of digits) values of $n$.
Thanks for help with this fun puzzle!