As an aside - just because the determinant of a matrix is "low" doesn't
mean inverting it or solving a linear system with the matrix as
coefficients will give numerical problems. For example, if I_n is an n x
n identity matrix, the matrix a*I_n where a = 10^-10 (say) has a
determinant of 10^(-10*n). Yet there is clearly no problem in inverting
such a matrix. One way of normalizing is to look at the ratio of the
largest to smallest eigenvalue. In this example, the ratio is one (no
problem) - but for Hilbert matrices, the ratio becomes < < 1 very
quickly as the size of the matrix increases.
Al Feiveson
-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of William
Gould, Stata
Sent: Friday, May 04, 2007 10:10 AM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: Hilbert matrices Mata testing
Marcos <M.Delprato@westminster.ac.uk> writes,
> I found in the Mata manual Hilbert matrices. They are defined as a
> matrix H with elements H[i,j]=1/(i+j-1). As it says in the help (type
> help
> mf_hilbert): Hilbert matrices are notoriously ill conditioned, with
> near zero determinants.
>
> The help also adds Hilbert(n) and invHilbert(n) are used in testing
> Mata. I think that is intuitive as since they are ill matrices with
> very low determinants and nearly singular you could use them to test
> the quality of, for instance, Mata's solver (e.g., qrsolve, lusolve,
> qrinv) and different inverters (e.g., cholinv, invsym).
>
> Am I right? Could anybody supply an example of how Hilbert matrices
> are used to test Mata?
We at StataCorp have do-files used for testing Stata. One of the
do-files simply serves to call all othe others. Currently, there are
3,103 do-files totalling 568,032 lines.
The part that tests Mata includes 325 files comprising 45,626 lines.
Here are a few lines from one of the files:
...
assert(reldif(1/det(Hilbert(2)), 12)<1e-16)
assert(reldif(1/det(Hilbert(3)), 2160)<1e-14)
assert(reldif(1/det(Hilbert(4)), 6048000)<1e-13)
assert(reldif(1/det(Hilbert(5)), 266716800000)<1e-12)
...
assert(mreldif(pinv(pinv(Hilbert(4))), Hilbert(4))<1e-12))
assert(mreldif(pinv(pinv(Hilbert(5))), Hilbert(5))<1e-11))
...
-- Bill
wgould@stata.com
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/