ScaLAPACK Archives

[Scalapack] complex pivoted qr factorisation

There was a relatively recently repaired bug of this kind in the LAPACK 
code;
the problem was that when some R(i,i) is very tiny (as in the examples 
below),
cancellation from updating the column norms can cause them to be inaccurate
and not reflect the true R(i,i) values, which then do not appear in 
sorted order.

But that should not cause R(2,2) and R(3,3) in the first example below to
come out in the wrong order, since they are of vastly different magnitudes.
Is this really the wrong order that appeared?

Jim



On 10/19/11 10:07 AM, James, Rodney wrote:
Hello David,

Thank you for the report.  It looks like this is probably a bug -- we will 
take a look at this and hopefully get a fix in place sometime in the near 
future.

--Rodney

On Oct 18, 2011, at 9:32 PM, David Wang wrote:

Hello Scalapack,

I have been using the pivoted QR factorisation routine (pzgeqpf) from
scalapack-1.8.0.  I have found that, in some cases, the diagonal
elements of the R matrix are not arranged in order of descending
magnitude as described in the user's guide.  I have attached a test
case for you to verify.  e.g. I get something like:

R(0,0) -0.707107 + 0i
R(1,1) -0.707107 + 0i
R(2,2) 8.16273e-17 + 0i
R(3,3) -0.243173 + 0i
R(4,4) 4.82605e-17 + 0i

This behaviour can be reproduced for the single precision complex case
(although not with the supplied test matrix).  The same test case
works for the real case (pdgeqpf).  e.g.

R(0,0) -0.707107
R(1,1) -0.707107
R(2,2) 0.5
R(3,3) 3.96991e-17
R(4,4) 2.2336e-17

I have not had any problems with the equivalent lapack routines (zgeqp3),
and have not see any mention of this in the errata.

David
<pzgeqpf_test.c><pdgeqpf_test.c>_______________________________________________
Scalapack mailing list
Scalapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/scalapack
_______________________________________________
Scalapack mailing list
Scalapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/scalapack


<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or