Re: svd_lapack vs eofunc/eofunc_ts

[1] What did you not like about the messages 6 & 7
     in response to your question? Di you read them?

[2] PLEASE ... do not include the whole digest email.
     That just clutters up every one on ncl-talk's email box.

On 11/30/2011 10:27 AM, Kerrie Geil wrote:
> Thanks to everyone for the input, but is it known why the eofunc
> eigenvalues are reduced by a factor of (n-1)? Solving for the
> eigenvalues by hand using the covariance matrix yields the eigenvalues
> produced by the SVD function, not the eofunc. From what I understand,
> the eigenvalues produced by each function should be the same. So I
> guess I'm still not understanding why there's a factor of (n-1) involved.
>
> Thanks for the help,
> Kerrie

>
> Today's Topics:
>
> 1. Re: Error: Undefined identifier (Mateus Teixeira)
> 2. Re: xMon Avg in netcdf (David Brown)
> 3. small bug in linint2_wrap?? (Cathy Smith)
> 4. mall bug in linint2_wrap?? p.s. (Cathy Smith)
> 5. svd_lapack vs eofunc/eofunc_ts (Kerrie Geil)
> 6. Re: svd_lapack vs eofunc/eofunc_ts (Gil Compo)
> 7. Re: svd_lapack vs eofunc/eofunc_ts (Gustavo Correa)
[SNIP]

>
> ------------------------------
>
> Message: 5
> Date: Tue, 29 Nov 2011 14:14:41 -0700
> From: Kerrie Geil <kgeil@email.arizona.edu
> <mailto:kgeil@email.arizona.edu>>
> Subject: svd_lapack vs eofunc/eofunc_ts
> To: ncl-talk@ucar.edu <mailto:ncl-talk@ucar.edu>
> Message-ID:
> <CAKBe+-ARF8sVMzHmmQM=R_wTWhgGAwgZj57H98FW_RN+3qsSfw@mail.gmail.com
> <mailto:R_wTWhgGAwgZj57H98FW_RN%2B3qsSfw@mail.gmail.com>>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Hi All,
>
> Can anyone tell me why the singular value decomposition (svd_lapack)
> below
> yields different eigenvalues than the eigenanalysis (eofunc, eofunc_ts),
> while both output the exact same EOFs and PCs? Can't see what I'm doing
> wrong. Code and output are below.
>
> Thanks!
> Kerrie
[SNIP]
> ------------------------------
>
> Message: 6
> Date: Tue, 29 Nov 2011 14:30:07 -0700
> From: Gil Compo <Gilbert.P.Compo@noaa.gov
> <mailto:Gilbert.P.Compo@noaa.gov>>
> Subject: Re: svd_lapack vs eofunc/eofunc_ts
> To: ncl-talk@ucar.edu <mailto:ncl-talk@ucar.edu>
> Message-ID: <4ED54EDF.6090707@noaa.gov
> <mailto:4ED54EDF.6090707@noaa.gov>>
> Content-Type: text/plain; charset="iso-8859-1"
>
> Kerrie,
>
> They should be different by a factor of 1/(n-1), since eofunc is
> computing the covariance matrix itself. In your case, that factor is
> exactly 1/3.
>
> best wishes,
>
> gil

> ------------------------------
>
> Message: 7
> Date: Tue, 29 Nov 2011 16:36:45 -0500
> From: Gustavo Correa <gus@ldeo.columbia.edu
> <mailto:gus@ldeo.columbia.edu>>
> Subject: Re: svd_lapack vs eofunc/eofunc_ts
> To: ncl-talk <ncl-talk@ucar.edu <mailto:ncl-talk@ucar.edu>>
> Message-ID: <43DF3514-1E15-4147-86B8-6BDDB2BBA96D@ldeo.columbia.edu
> <mailto:43DF3514-1E15-4147-86B8-6BDDB2BBA96D@ldeo.columbia.edu>>
> Content-Type: text/plain; charset=us-ascii
>
> Hi Kerrie
>
> For what it is worth, the second group of eigenvalues is equal to
> the first group
> divided by 3, which is the largest dimension of your matrix.
> It may depend a bit on how the SVD problem is posed, I suppose.
>
> Gus Correa
_______________________________________________
ncl-talk mailing list
List instructions, subscriber options, unsubscribe:
http://mailman.ucar.edu/mailman/listinfo/ncl-talk
Received on Wed Nov 30 11:46:24 2011