1. Rounding error

Many textbooks say that when performing calculations where significant figures matter, carry extra digits and ONLY ROUND AT THE END.

I'm sure it was determined that rounding at the end minimizes error in calculation/ error propagation. But, does anyone know how to prove that?

I have little statistics background, so I've only assumed this is a statistics related question. I'm much stronger in Calculus and Diff. eq.

2. Re: Rounding error

Hey Tclack.

You should try using a form of the triangle inequality. As an example, lets say we have three points of rounding call them x, y, z. We also have a form of rounding w which does it at the end.

You have to show that |x| + |y| + |z| >= |w| where |.| denotes the magnitude of the error introduced by doing a round-off.