Hi there. My environment: Win7==>Oracle VB==>Ubuntu==>GFortran. I need to check if my computations which involve long recurrences lose precision. There is an intrinsic function PRECISION (X) in Fortran. My question is: Let's say I defined a function f(x). Inside of this function there is a variable res which is a part of a do loop and the loop takes let's say 50 recurrences thus changing the variable res 50 times. Every time it is a bunch of additions and multiplications. Can I apply the PRECISION (res) function after that to determine if I've lost any significant digits?

Thanks, A.

No. That function would tell you what the precision of the current variable is (somewhere around 7 significant digits if single precision, 15 if double), but it wouldn't be able to tell you if some mathematical operation you've performed in the past was truncated due to lack of precision.

The best option is to run through the math yourself, either that or re-run the same process with a double or a quad to see if there is any significant difference.

2 members found this post helpful.

suicidaleggroll, thanks again. I sort of did some research last night and came to the same conclusion. I do have a question, however, while I have you here :-) Will this declaration REAL (KIND=16) :: x guarantee the quadruple precision for the variable x? Thanks A.

I usually just use REAL*16

with FORTRAN

Keep your HP calculator NEXT TO THE KEYBOARD
run the math !!! as you test the code

John W., My calculations are very long, unfortunately. Still, is it OK to use KIND=16 ?