machine precision -- smooth region

[diedro]

hi everyone,

I'ma writing a code about diffusion. It is veru sensible to fortran precision. I men negative value (taht I do not want) star from -1.e-60 and than rise ud to 1.e2.

I compile in double precision. Hos is possible that the program can handle so negative errors, Can I set them to zero when I compile.

Really really thanks

[Anonymous66]

The option, -fpe1 will turn on underflow to zero rather than allowing gradual underflow. Any value that is too small to be represented precisely in double precision will be rounded down to 0. Is something like this what you are looking for?

[diedro]

yes is that the things that I want to know. However I got the same error. Do you know a forum where to talk about numerical scheme problem? thanks a lot

[Anonymous66]

We do not have a forum specifically for numerical scheme problems. This white paper may be useful to you though.

[Ron_Green]

There are several articles on the -fp-model and related articles: I've attached a white paper to this post. Also see: https://secure-software.intel.com/sites/default/files/article/326703/fp-control-2012-08.pdf http://software.intel.com/en-us/articles/consistency-of-floating-point-results-using-the-intel-compiler/

[jimdempseyatthecove]

diedro wrote:

hi everyone,

I'ma writing a code about diffusion. It is veru sensible to fortran precision. I men negative value (taht I do not want) star from -1.e-60 and than rise ud to 1.e2.

I compile in double precision. Hos is possible that the program can handle so negative errors, Can I set them to zero when I compile.

Really really thanks

Try changing your parameters and literals from "e" to "d". Example, change -1.e-60 to -1.d-60 "e" defaults to REAL(4) Jim Dempsey

[TimP]

In case the implication of changing from a single to a double precision constant isn't evident, the smallest magnitude non-zero single precision numbers are TINY(1.)*EPSILON(1.) when gradual underflow is enabled, and TINY(1.) in the case of abrupt underflow (ifort option -ftz). In either case, 1.e-60 is the same as 0. (except possibly in some contexts of -mia32)