Free42 possible accuracy flaw

03222022, 12:33 PM
Post: #1




Free42 possible accuracy flaw
I may have identified an accuracy flaw in the power operation in Free42/Decimal.
This came from the last [VA] SRC #010  Pi Day 2022 Special thread where a calculation done by two methods gave significantly different results, with no clear explanation. I found a problem with this calculation: (1+1/X)^X [1] with X=1E8. A comparison with the expended and mathematically equivalent expression: exp( X * ln(1+1/X) ) [2] gives the following results [1] 2.718281814867636217652977244308419 [2] 2.718281814867636217652977243009177 Although we could expect expression [2] to be less accurate due to the intermediate calculations, it turns out that result [2] is correct (according to Wolfram reference) and result [1] is in error in its last 7 digits. Checked on latest Free42 3.0.10. JF 

« Next Oldest  Next Newest »

User(s) browsing this thread: 1 Guest(s)