Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > comp.soft-sys.math.mathematica > #7589 > unrolled thread
| Started by | Leslaw Bieniasz <nbbienia@cyf-kr.edu.pl> |
|---|---|
| First post | 2012-01-05 11:01 +0000 |
| Last post | 2012-01-09 08:26 +0000 |
| Articles | 7 — 5 participants |
Back to article view | Back to comp.soft-sys.math.mathematica
JordanDecompose[] Leslaw Bieniasz <nbbienia@cyf-kr.edu.pl> - 2012-01-05 11:01 +0000
Re: JordanDecompose[] danl@wolfram.com - 2012-01-06 09:16 +0000
For a variety of plotting experiences, maybe bugs? (small intervals) Richard Fateman <fateman@cs.berkeley.edu> - 2012-01-08 09:23 +0000
Re: JordanDecompose[] Leslaw Bieniasz <nbbienia@cyf-kr.edu.pl> - 2012-01-09 08:16 +0000
Re: JordanDecompose[] danl@wolfram.com - 2012-01-10 10:59 +0000
RE: For a variety of plotting experiences, maybe bugs? (small intervals) "David Park" <djmpark@comcast.net> - 2012-01-09 08:25 +0000
Re: For a variety of plotting experiences, maybe bugs? Richard Fateman <fateman@eecs.berkeley.edu> - 2012-01-09 08:26 +0000
| From | Leslaw Bieniasz <nbbienia@cyf-kr.edu.pl> |
|---|---|
| Date | 2012-01-05 11:01 +0000 |
| Subject | JordanDecompose[] |
| Message-ID | <je3vti$eeh$1@smc.vnet.net> |
Hi, MATHEMATICA has a built-in function JordanDecompose[] that can computes a Jordan normal form of a given matrix, and the related similarity matrix. Most importantly, it can obtain symbolic representations of these matrices, for a matrix given exclusively in terms of symbols (no concrete numbers, but symbols representing matrix elements). My question is: what kind of an algorithm is used by this function, in the symbolic mode of operation? I would appreciate references to the mathematical literature describing the algorithm. Leslaw
[toc] | [next] | [standalone]
| From | danl@wolfram.com |
|---|---|
| Date | 2012-01-06 09:16 +0000 |
| Message-ID | <je6e5h$q7b$1@smc.vnet.net> |
| In reply to | #7589 |
Here is a mildly edited comment from the code. "This routine uses the method based on matrix presentations. It is a more general method than the one used in [...] (the latter is faster but does not always work, for derogatory matrices). Initially the matrix is decomposed into subblocks, where each subblock has eigenvalues of the same value. Eigenvalues in different subblocks are distinct. Then elementary row and column operations are done on the characteristic presentations of the subblocks." This was from 1993, when two of us worked on the implementation. It was based on course notes I took in 1985. (Coincidently, the class was taught by one of the original Mathematica developers, Dan Grayson.) It's been a while since bugs were spotted in that code, so I have to confess I do not remember the details. Daniel Lichtblau Wolfram Research
[toc] | [prev] | [next] | [standalone]
| From | Richard Fateman <fateman@cs.berkeley.edu> |
|---|---|
| Date | 2012-01-08 09:23 +0000 |
| Subject | For a variety of plotting experiences, maybe bugs? (small intervals) |
| Message-ID | <jebnb6$18m$1@smc.vnet.net> |
| In reply to | #7596 |
try this (I did it on version 7...)
>
Table[Plot[Cos[x], {x, 1 , 1 + 2.0^-i}], {i, 40, 45}]
[toc] | [prev] | [next] | [standalone]
| From | Leslaw Bieniasz <nbbienia@cyf-kr.edu.pl> |
|---|---|
| Date | 2012-01-09 08:16 +0000 |
| Message-ID | <jee7pc$cjq$1@smc.vnet.net> |
| In reply to | #7596 |
On Fri, 6 Jan 2012, danl@wolfram.com wrote: > > Here is a mildly edited comment from the code. > > "This routine uses the method based on matrix presentations. > It is a more general method than the one used in [...] > (the latter is faster but does not always work, for derogatory matrices). What is exactly the reference marked by [...], and is there any description of your algorithm available anywhere? I believe this kind of algorithms can't be a proprietary secret of Wolfram? Leslaw
[toc] | [prev] | [next] | [standalone]
| From | danl@wolfram.com |
|---|---|
| Date | 2012-01-10 10:59 +0000 |
| Message-ID | <jeh5ms$r55$1@smc.vnet.net> |
| In reply to | #7674 |
I should have made that clear. The [...] referred by name to a function in the code, not a reference from the literature. I did not want to put internal function names in my post. That function handles the case where the matrix is nonderogatory. More specifically, it handles all eigenvalues that have geometric multiplicity of 1 (I think it also handles the simple case where gemoetric multiplicity = algebraic multiplicity). The algorithm is not a proprietary secret. I simply cannot find a reference in the literature that seems to describe it. All I have is old lecture notes, if indeed I still have those. There is something slightly similar, I think, in the text Linear Algebra by Charles Curtis. Itworks out the rational canonical form, then notes that this gives the Jordan decomposition in the special case where we work over the algebraic closure of the matrix coefficients field. Also I think there may be something similar in the algebra text by Dummit and Foote. Daniel Lichtblau Wolfram Research
[toc] | [prev] | [next] | [standalone]
| From | "David Park" <djmpark@comcast.net> |
|---|---|
| Date | 2012-01-09 08:25 +0000 |
| Subject | RE: For a variety of plotting experiences, maybe bugs? (small intervals) |
| Message-ID | <jee89n$d24$1@smc.vnet.net> |
| In reply to | #7596 |
Small domains and small intervals are often treated masochistically. And why
the powers of 2 instead of 10?
This is the easy way to make the plots. Use a reasonable domain and range
and enough WorkingPrecision.
With[{n = 140},
Plot[(Cos[x 10^-n + 1] - Cos[1]) 10^n, {x, 0, 1},
WorkingPrecision -> n + 3,
Frame -> True,
FrameLabel -> {(y - 1) 10.^n, 10.^n (Cos[y] - Cos[1])}]
]
Or use a series expansion:
Series[10^n (Cos[1 + x 10^-n] - Cos[1]), {x, 0, 2}] // Normal
-2^(-1 - n) 5^-n x^2 Cos[1] - x Sin[1]
and then use MachinePrecision:
With[{n = 145},
Plot[-2^(-1 - n) 5^-n x^2 Cos[1] - x Sin[1], {x, 0, 1},
WorkingPrecision -> MachinePrecision,
Frame -> True,
FrameLabel -> {(y - 1) 10.^n, 10.^n (Cos[y] - Cos[1])}]
]
That is the way they do it in technical journals all the time.
David Park
djmpark@comcast.net
http://home.comcast.net/~djmpark/index.html
From: Richard Fateman [mailto:fateman@cs.berkeley.edu]
try this (I did it on version 7...)
>
Table[Plot[Cos[x], {x, 1 , 1 + 2.0^-i}], {i, 40, 45}]
[toc] | [prev] | [next] | [standalone]
| From | Richard Fateman <fateman@eecs.berkeley.edu> |
|---|---|
| Date | 2012-01-09 08:26 +0000 |
| Subject | Re: For a variety of plotting experiences, maybe bugs? |
| Message-ID | <jee8bk$d2m$1@smc.vnet.net> |
| In reply to | #7596 |
David:
Thanks. Your suggestions would be useful if I were interested in exploring
the cosine function. However, I was interested in exploring
Cos[MachinePrecisionNumbers].
(actually, some other numerical functions. I just used cos to simplify
the example).
My interest was whether the numerical subroutine reflected the
mathematical functions and
were appropriately monotonic in small ranges. Looking "microscopically"
one expects to see some kind of step function,
but if poorly implemented there may be some reversals-- i.e. local
extrema-- that are not
mathematically justified, but are effects of the implemented algorithm,
roundoff, etc.
I found that Mathematica refused to plot the most straightforward
versions of these
investigations.
As far as why the powers of 2 instead of 10, the fraction part of a
machine float x has, as one
unit in the last place, one part in 2^(52) of x. I find 2^-52 to
be much nicer to look at than
2.220446049250313*^-16
as well as more accurate.
On 1/8/2012 10:03 AM, David Park wrote:
> Small domains and small intervals are often treated masochistically. And why
> the powers of 2 instead of 10?
>
> This is the easy way to make the plots. Use a reasonable domain and range
> and enough WorkingPrecision.
>
> With[{n = 140},
> Plot[(Cos[x 10^-n + 1] - Cos[1]) 10^n, {x, 0, 1},
> WorkingPrecision -> n + 3,
> Frame -> True,
> FrameLabel -> {(y - 1) 10.^n, 10.^n (Cos[y] - Cos[1])}]
> ]
>
> Or use a series expansion:
>
> Series[10^n (Cos[1 + x 10^-n] - Cos[1]), {x, 0, 2}] // Normal
>
> -2^(-1 - n) 5^-n x^2 Cos[1] - x Sin[1]
>
> and then use MachinePrecision:
>
> With[{n = 145},
> Plot[-2^(-1 - n) 5^-n x^2 Cos[1] - x Sin[1], {x, 0, 1},
> WorkingPrecision -> MachinePrecision,
> Frame -> True,
> FrameLabel -> {(y - 1) 10.^n, 10.^n (Cos[y] - Cos[1])}]
> ]
>
> That is the way they do it in technical journals all the time.
>
>
> David Park
> djmpark@comcast.net
> http://home.comcast.net/~djmpark/index.html
>
>
>
>
>
>
> From: Richard Fateman [mailto:fateman@cs.berkeley.edu]
>
>
> try this (I did it on version 7...)
> Table[Plot[Cos[x], {x, 1 , 1 + 2.0^-i}], {i, 40, 45}]
>
[toc] | [prev] | [standalone]
Back to top | Article view | comp.soft-sys.math.mathematica
csiph-web