A long time ago (the late '80s or early '90s) I needed a square root routine for a floating point number and found an article by a Jack Crenshaw (a real rocket scientist), that always converged in less than 5 iterations. First you normalized your FP number (converted the mantissa to 1.xxxxx) while adjusting the exponent and then multiplied by 1.38 which gave a surprisingly close answer.
I'm guessing that (if I could find the article) the constant used in this floating point technique and the one used in this article are somehow inversely related too!
Gee, you've triggered a memory for me too. I remember doing something very similar (1.38 was the trigger) around the same time. It probably writing floating point routines for processors that didn't have them, perhaps the Z-80.
I'm guessing that (if I could find the article) the constant used in this floating point technique and the one used in this article are somehow inversely related too!
In any case, it's a very nice article ... thanks!