Author 
Message 
Robert B. Israel science forum Guru
Joined: 24 Mar 2005
Posts: 2151

Posted: Wed May 31, 2006 4:00 pm Post subject:
Re: Maple's linear algebra routine and higher precision arithmetic



In article <pan.2006.05.31.15.21.21.254816@acadiau.ca>,
Ronald Haynes <ronald.haynes@acadiau.ca> wrote:
Quote:  On Mon, 29 May 2006 13:12:03 0700, Ronald Bruck wrote:
In article <pan.2006.05.29.18.47.40.640507@acadiau.ca>, Ronald Haynes
ronald.haynes@acadiau.ca> wrote:
Hi, many thanks... I just want to test the validity of some lower
precision results with results generated by increasing levels of precision.
Would defining the matrix entries as 1.0 4.0 1.0 do the trick to force
floats?
In Mapleyes. If you do this in Mathematica you're welladvised to
use a notation like N[1,64], N[4, 64], N[1,64]. The "4.0" will be
read as a standard doubleprecision floatingpoint number.
(Mathematica doesn't have a "Digits" commandeach number has its own
intrinsic precision.)
But you asked about Maple, not Mathematica, so this is just for your
general edification...
Hi, upon simple testing this does not seem to work. i.e. setting
Digits:=1 and computing the inverse does not give different results from
setting Digits:=100 and computing the inverse.
Sometime ago I recall hearing that Maple was changing that way it dealt
with linear algebra routines  maybe by using NAG routines (or similar)
under the hood? If Maple is actually using compiled routines how could
Digits:= whatever actually have an effect, I would presume the routines
would be double precision by default and not be affected by a change of
Digits (except possibly by rounding the input matrix). In my case the
matrix is integer valued. In fact only a slight modification of the
tridiag(1,4,1) matrix.

The environment variable UseHardwareFloats determines whether
floatingpoint operations on Arrays are done in hardware
(which is essentially doubleprecision) or software (with whatever
setting of Digits is current). The default value is "deduced",
meaning that hardware is used if Digits <= evalhf(Digits) (i.e.
the precision of hardware floatingpoint, which is 14 on my Windows
system), and software is used if Digits > evalhf(Digits). Compiled
NAG code is used for both hardware and software floatingpoint
computations in LinearAlgebra.
Thus:
Quote:  A:= <<7.0, 2.0><3.0, 5.0>>;
Digits:= 1;
A^(1); # this will use hardware floats 
[ 0.172413793103448260 0.103448275862068950]
[ ]
[0.0689655172413792956 0.241379310344827569 ]
Quote:  UseHardwareFloats:= false:
A^(1); # now it will use software floats with Digits = 1 
[ 0.1 0.06]
[ ]
[0.04 0.2 ]
Quote:  UseHardwareFloats:= deduced: Digits:= 40:
A^(1); # this will use software floats with Digits = 40 
[0.1724137931034482758620689655172413793104 ,
0.1034482758620689655172413793103448275862]
[0.06896551724137931034482758620689655172416 ,
0.2413793103448275862068965517241379310345]
Robert Israel israel@math.ubc.ca
Department of Mathematics http://www.math.ubc.ca/~israel
University of British Columbia Vancouver, BC, Canada 

Back to top 


Ronald Haynes science forum beginner
Joined: 10 Feb 2005
Posts: 5

Posted: Wed May 31, 2006 3:21 pm Post subject:
Re: Maple's linear algebra routine and higher precision arithmetic



On Mon, 29 May 2006 13:12:03 0700, Ronald Bruck wrote:
Quote:  In article <pan.2006.05.29.18.47.40.640507@acadiau.ca>, Ronald Haynes
ronald.haynes@acadiau.ca> wrote:
Hi, many thanks... I just want to test the validity of some lower
precision results with results generated by increasing levels of precision.
Would defining the matrix entries as 1.0 4.0 1.0 do the trick to force
floats?
In Mapleyes. If you do this in Mathematica you're welladvised to
use a notation like N[1,64], N[4, 64], N[1,64]. The "4.0" will be
read as a standard doubleprecision floatingpoint number.
(Mathematica doesn't have a "Digits" commandeach number has its own
intrinsic precision.)
But you asked about Maple, not Mathematica, so this is just for your
general edification...

Hi, upon simple testing this does not seem to work. i.e. setting
Digits:=1 and computing the inverse does not give different results from
setting Digits:=100 and computing the inverse.
Sometime ago I recall hearing that Maple was changing that way it dealt
with linear algebra routines  maybe by using NAG routines (or similar)
under the hood? If Maple is actually using compiled routines how could
Digits:= whatever actually have an effect, I would presume the routines
would be double precision by default and not be affected by a change of
Digits (except possibly by rounding the input matrix). In my case the
matrix is integer valued. In fact only a slight modification of the
tridiag(1,4,1) matrix.
Any further thoughts?
Thanks,
R Haynes 

Back to top 


Ronald Bruck science forum Guru
Joined: 05 Jun 2005
Posts: 356

Posted: Mon May 29, 2006 8:12 pm Post subject:
Re: Maple's linear algebra routine and higher precision arithmetic



In article <pan.2006.05.29.18.47.40.640507@acadiau.ca>, Ronald Haynes
<ronald.haynes@acadiau.ca> wrote:
Quote:  Hi, many thanks... I just want to test the validity of some lower
precision results with results generated by increasing levels of precision.
Would defining the matrix entries as 1.0 4.0 1.0 do the trick to force
floats?

In Mapleyes. If you do this in Mathematica you're welladvised to
use a notation like N[1,64], N[4, 64], N[1,64]. The "4.0" will be
read as a standard doubleprecision floatingpoint number.
(Mathematica doesn't have a "Digits" commandeach number has its own
intrinsic precision.)
But you asked about Maple, not Mathematica, so this is just for your
general edification...

Ron Bruck
Posted Via Usenet.com Premium Usenet Newsgroup Services

** SPEED ** RETENTION ** COMPLETION ** ANONYMITY **

http://www.usenet.com 

Back to top 


Ronald Haynes science forum beginner
Joined: 10 Feb 2005
Posts: 5

Posted: Mon May 29, 2006 6:47 pm Post subject:
Re: Maple's linear algebra routine and higher precision arithmetic



Hi, many thanks... I just want to test the validity of some lower
precision results with results generated by increasing levels of precision.
Would defining the matrix entries as 1.0 4.0 1.0 do the trick to force
floats?
Thanks,
R Haynes
On Mon, 29 May 2006 17:44:19 +0000, Robert Israel wrote:
Quote:  In article <pan.2006.05.29.15.19.31.155994@acadiau.ca>,
Ronald Haynes <ronald.haynes@acadiau.ca> wrote:
Hi, does Maple's linear algebra routine MatrixInverse play nicely with the
Digits command? If I set Digits:=50 and compute a Matrix Inverse will the
arithmetic all be done to 50 digits?
Yes, it will (if the original matrix involves floats). Note that
"to 50 digits" doesn't mean that the results are accurate to 50 digits,
but rather that the calculations are done with Digits set to 50.
Robert Israel israel@math.ubc.ca
Department of Mathematics http://www.math.ubc.ca/~israel
University of British Columbia Vancouver, BC, Canada 


Back to top 


Robert B. Israel science forum Guru
Joined: 24 Mar 2005
Posts: 2151

Posted: Mon May 29, 2006 5:44 pm Post subject:
Re: Maple's linear algebra routine and higher precision arithmetic



In article <pan.2006.05.29.15.19.31.155994@acadiau.ca>,
Ronald Haynes <ronald.haynes@acadiau.ca> wrote:
Quote:  Hi, does Maple's linear algebra routine MatrixInverse play nicely with the
Digits command? If I set Digits:=50 and compute a Matrix Inverse will the
arithmetic all be done to 50 digits?

Yes, it will (if the original matrix involves floats). Note that
"to 50 digits" doesn't mean that the results are accurate to 50 digits,
but rather that the calculations are done with Digits set to 50.
Robert Israel israel@math.ubc.ca
Department of Mathematics http://www.math.ubc.ca/~israel
University of British Columbia Vancouver, BC, Canada 

Back to top 


Ronald Haynes science forum beginner
Joined: 10 Feb 2005
Posts: 5

Posted: Mon May 29, 2006 3:19 pm Post subject:
Maple's linear algebra routine and higher precision arithmetic



Hi, does Maple's linear algebra routine MatrixInverse play nicely with the
Digits command? If I set Digits:=50 and compute a Matrix Inverse will the
arithmetic all be done to 50 digits?
Thanks,
R Haynes 

Back to top 


Google


Back to top 



The time now is Fri Sep 22, 2017 8:39 pm  All times are GMT

