On Friday, July 15, 2016 at 10:31:20 PM UTC+1, Kurolong wrote: > > Turns out that it actually did fix the error, but i was too tired and > confused things. > BUT the components of the fractions were a lot too large. > A Gauß-algorithm should have returned very managable fractions, not > fractions of 300-decimal-place-numbers by 300-decimal-place-numbers. >
How do you know that? In general, the number of digits in the denominators for exact rational arithmetic increases linearly with the number of computational steps. Since each output in A \ b by Gaussian elimination on an nxn matrix A requires O(n^2) arithmetic operations (O(n^3) operations for n outputs), 300 decimal places is reasonable even for a 10x10 matrix. (This is why rational bignum arithmetic is usually even more of an impractical luxury than ordinary bignum arithmetic.)
