Algorithm accuracy vs calculator precision

[dm319]

Ha ha, can I summarise this in one paragraph? Here goes:

Take a collection of 13 difficult TVM problems and use arbitrary precision maths in R to solve the problems giving a highly accurate reference value. Then test the calculators, extracting all the digits you can from the answer. Calculate an absolute error by taking the difference from the reference value. Sometimes take the relative error by dividing the absolute error by the reference value (to normalise for very big and small results). Then take the -log10 of these values to give you a measure of accuracy that is roughly comparable to number of digits. Finally, divide by the actual number of digits precision of the calculator.

There is more here and here if that didn't put you off.