----------------------------------------------------------------------------------
@MSGID:
<1fc4a61c-b795-4de4-a50f-9db95f208aa6n@googlegroups.com> 91e5549a
@REPLY: 1@dont-email.me> 2dfd8802
@REPLYADDR Farzad Tatar
<tatarfarzad1992@gmail.com>
@REPLYTO 2:5075/128 Farzad Tatar
@CHRS: CP866 2
@RFC: 1 0
@RFC-References:
<aa2d38fc-1df2-4703-b1b3-67adf45e2793n@googlegroups.com> <8a0d90c2-1912-41f9-8cf1-76c8f12698adn@googlegroups.com>
<c8ac7613-a6bc-4804-a668-40dbc9de519en@googlegroups.com> <26454e02-d247-4d86-a721-6166cdf11d56n@googlegroups.com>
<232d1483-dbdd-4833-8697-6e15686b90c7n@googlegroups.com> <klmb03F5oasU1@mid.individual.net>
<65c791c9-0095-42ed-aa69-9c2ba04b3caen@googlegroups.com> 1@dont-email.me>
@RFC-Message-ID:
<1fc4a61c-b795-4de4-a50f-9db95f208aa6n@googlegroups.com>
@TZUTC: -0700
@PID: G2/1.0
@TID: FIDOGATE-5.12-ge4e8b94
On Tuesday, September 5, 2023 at 12:52:32 PM UTC+2, David Jones wrote:
> Farzad Tatar wrote:
>
> >
> > > Before making uncontrolled assumptions, you should check the
> > > condition number of your matrix (see the definition of condition
> > > numebr for matices in
> > >
https://en.wikipedia.org/wiki/Condition number ).
> >
> > Thank you for your insightful advice. I calculated the condition
> > number of my matrix using the Forbenus norm and here are the norms
> > and the condition number of my RBF matrix.
> >
> > Norm , Norm of the inverse matrix and Condition number are:
> > 142.049233380836 1.829619706901719E+016 2.598960767438591E+018
> >
> > Considering the condition number, I may lose the accuracy up to 18
> > decimals. Am I right? Please let me know if you have any textbook or
> > reference on how to improve the condition number. I have read the RBF
> > interpolation is well-conditioned, or maybe I am wrong and that has
> > been compactly supported RBF interpolation. I need to delve more into
> > this topic. Thank you anyway for your valuable advice.
> >
> > Cheers,
> >
> > F Tatar
> There are two direct solutions:
> (1) reduce the effective correlation that you are assuming, which will
> move the matrix in the direction of an identity matrix. Ensure that the
> largest off-diagonal element isn`t too close to one.
> (2) reduce the size of the field over which you are doing
> interpolation. That is, for any given target point, limit the number of
> neighbours being used for fitting: maybe only 20 rather than 1400.
>
> You might also want to reconsider you choice of basis-function. You
> could also look at the problem from the point of view of statistical
> optimal interpolation rather than basis-functions: this gives a very
> similar algorithm but with a different interpretation, which may be
> helpful.
Thank you. I did not totally understand your first point. Regarding
your second advice, I have used a similar job, though I cannot change
the size of RBF matrix, I have set far values to be zero. However,
this did not help. I was expecting a meaningful improvement in reducing
the condition number of my problem, but it is still in the order of
10E+13.
I tried a different basis function but I did not improve the condition number.
A solution that I discovered on the internet is to use a Singular
Value Decomposition algorithm for the matrix inversion. Do you think it
could help me in my case?
Cheers,
Farzad
--- G2/1.0
* Origin: usenet.network (2:5075/128)
SEEN-BY: 5001/100 5005/49 5015/255 5019/40 5020/715
848 1042 4441 12000
SEEN-BY: 5030/49 1081 5058/104 5075/128
@PATH: 5075/128 5020/1042 4441