MATLAB Answers

Zheng
0

How to increase the precision of Matlab (compared to Fortran) ???

Asked by Zheng
on 17 Nov 2012
Hi,
I am using Matlab to translate subroutines coded by FORTRAN. However I found that the truncation error of matlab is very important, it means the precision of matlab can not be comprared to Frotran.
I tried format command, it does not work. Anyone knows how to increase the precision of matlab?
Regards
Zheng Li

  1 Comment

The question would be more clear, if you explain "truncation error of Matlab is very important" with an explicit example. Currently it seems, like you confuse the internal representation of numbers (double is exactly the same as real*8) and the output to the command window (affected by the format command).

Sign in to comment.

3 Answers

Answer by Walter Roberson
on 18 Nov 2012

MATLAB's "double" has the same precision as Fortran's REAL*8.
MATLAB does not have a numeric class equivalent to REAL*16.

  0 Comments

Sign in to comment.


Answer by Ben Barrowes on 18 Nov 2012

What do you mean when you say "compare"? "format long" in matlab should give many digits of display. In any case, like Strider says, matlab default double is the same precision as fortran's real*8. If you need arbitrary precision in matlab (but slow calculation) try the multiple precision toolbox:
If you are converting fortran to matlab, try f2matlab:
it can get you 95% of the way to working m-code.

  0 Comments

Sign in to comment.


Answer by Pavel Holoborodko on 20 Jul 2013

This addon: Multiprecision Computing Toolbox for MATLAB allows programming with arbitrary precision.
It also provides fast quadruple precision mode (equivalent to REAL*16) in particular.

  8 Comments

To me it is about disclaimers and quality of the answer. Pavel do you "work for" advanpix? If so, I think you should mention it. Further, an example would go a long way to demonstrating what you mean (especially since one cannot just download the product and try it out). For example a simple Fortran program showing the result of a REAL*16 opperation, the standard MATLAB result, and the result with the MP toolbox..
Hi all, I’m the technical marketing product manager of MATLAB Answers. Thanks, Pavel, for sharing the pointer to an add-on toolset which might help address Zheng’s question. Thanks, Jan, as well, for keeping an eye on the integrity of the Answers community. While MathWorks certainly does not want Answers to become an advertising mechanism for software distributors, we do encourage the offering of assistance to our customers, even when in the form of pointing to outside tools. We appreciate both of your perspectives, and believe each adds value to the general Answers audience.
Pavel, MathWorks offers a Connections program, which works with third-party organizations that develop and distribute complementary, commercially-available products and services based on MATLAB and Simulink. I’d like to encourage you to look into this venue as a means of raising awareness of this software capability: http://www.mathworks.com/products/connections/join/index.html . If this software provides the capability noted above, we’d certainly like to help spread the word. I’d also encourage you to share some code or examples from the toolset which might further assist Zheng, to enhance the value of the answer provided.
@Wendy: Thanks, this clears my question completely and I've removed the my flag.

Sign in to comment.