LSLXD / L2LXD + L4LXD

[Matthaios_Antoninis]

I was using DLSLXD in a project that is running and been compiled in CVF. Now I am trying to migrate this project to IVF, but eventually despite the fact that this subroutine LSLXD its statement contains the same numbers, I take FATAL ERROR - that the coefficient matrix is not positive definite.

I tried also L2LXD call with the parameters for workspace giving some more workspace but I did not received anything different.

Please find attached a project where we are reading from input file ,[IRow],[ICol], B =0.0 there is all the calls inside, the input file IMSLTest.txt contains the information that is running in CVF (I know that, because I am running in parallel)

Thank you in advance.

[Steven_L_Intel1]

I will take a look at this.

[Matthaios_Antoninis]

Manys thanks,

I am sorry but I get a wrong impression from the usage of this function LSLXD, I realise that matrix must include (according the example) only non zero elements, and in my project includes also zeros (something like skyline), but this works in CVF and gives wright results, that's why I did not mention the formulation of I just read the data.

I will try to reform the formulation of and I will inform you if this was the problem, still it is strange that it runs in CVF.

Many thanks

[Matthaios_Antoninis]

I just tried you own example (in order to have smth easy to compare) inside IMSL Math.pdf page 394, the simple example. I tried to transform matrix in skyline format with all the coefficients of lower triangle, also zeros. Ana I realized that whatever coefficients you give to with or without zeros the subroutine solves the system deleting zeros ... using sparsity.

This means that the initial problem still exist !

I am waiting your proposals.

Ps it is really strange how it works, and nowwhere it is mentioned, if I am wrong please forgive me.

[Matthaios_Antoninis]

I have not received any suggestion but I tried to the call of L2LXD (+ L4LXD) in order to be more parameterized the call of the subroutine, I resolve the problem using default all the other parameters exept IPARAM(2) = 1, Another Factorization method (Markowitz). Also I changed the Workspace LIWK,IWK (but this did not solve the problem).

[Steven_L_Intel1]

While we provide IMSL as part of the product with that option, IMSL itself comes from another company (Rogue Wave Software). We can help with simple issues, but detailed use questions we have to defer to Rogue Wave. Do you still need help with this? If so, I will send Rogue Wave your questions and pass on any reply.

[Matthaios_Antoninis]

Many thanks for your message Mr Steve, I don't any more help for this topic, I hope so that with my last recomendation I will help If somebody face the same problem.

Matthaios