Open sangallidavide opened 4 months ago
The first comment is that in version 3.20 we made a change that is probably relevant for this: we changed the default BV
type from BVSVEC
to BVMAT
. The latter avoids the reported problem in some cases. Maybe the installed version of SLEPc is older than 3.20. In that case I would suggest to upgrade.
Another comment is that when computing such a large number of eigenvalues, one must be careful how to choose de dimension parameters, in particular it is highly recommended to specify an appropriate value of mpd
in the call to EPSSetDimensions()
, e.g., 500 in this case would be ok. This is explained in section 2.6.5 of the SLEPc users manual "Computing a Large Portion of the Spectrum". However, if I am now wrong, yambo's interface does not have support for this parameter.
Ok, we can easily add control for the mpd
variable.
@imarri do you know which version of Slepc is presently linked ?
Dear all, Currently I’m using a version of Yambo (5.2.0) linked with the Slepc 3.18.1
Sincerely Ivan
On 16 May 2024, at 13:42, Davide Sangalli @.***> wrote:
Ok, we can easily add control for the mpd variable.
@imarri https://www.google.com/url?q=https://github.com/imarri&source=gmail-imap&ust=1716464576000000&usg=AOvVaw0WHFTLgC47NCSKivqjd9N6 do you know which version of Slepc is presently linked ?
— Reply to this email directly, view it on GitHub https://www.google.com/url?q=https://github.com/yambo-code/yambo/issues/88%23issuecomment-2115004836&source=gmail-imap&ust=1716464576000000&usg=AOvVaw1VJGok3EaZhSz7dC6sY5PO, or unsubscribe https://www.google.com/url?q=https://github.com/notifications/unsubscribe-auth/AGJQZKISQ4AGHJVYFCPSEMTZCSLT5AVCNFSM6AAAAABHZ2BZ7CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMJVGAYDIOBTGY&source=gmail-imap&ust=1716464576000000&usg=AOvVaw2dvcTGgHaOutgyCPAmvWC4. You are receiving this because you were mentioned.
In the software available on Leonardo there is also the v5.2.3 of Yambo that links PETSc v3.20.1 and SLEPc v 3.20.0.
[nspallan@login07 ~]$ module load profile/chem-phys
[nspallan@login07 ~]$ module show yambo/5.2.3--openmpi--4.1.6--nvhpc--23.11
-------------------------------------------------------------------
/leonardo/prod/opt/modulefiles/chem-phys/applications/yambo/5.2.3--openmpi--4.1.6--nvhpc--23.11:
module-whatis {YAMBO is an open-source code released within the GPL licence.}
module load fftw/3.3.10--openmpi--4.1.6--nvhpc--23.11
module load hdf5/1.14.3--openmpi--4.1.6--nvhpc--23.11
module load libxc/5.2.3--nvhpc--23.11-bwqd5we
module load netcdf-c/4.9.2--openmpi--4.1.6--nvhpc--23.11
module load netcdf-fortran/4.6.1--openmpi--4.1.6--nvhpc--23.11
module load netlib-scalapack/2.2.0--openmpi--4.1.6--nvhpc--23.11
module load openblas/0.3.24--nvhpc--23.11
module load openmpi/4.1.6--nvhpc--23.11
module load petsc/3.20.1--openmpi--4.1.6--nvhpc--23.11-complex-6q7sfss
module load slepc/3.20.0--openmpi--4.1.6--nvhpc--23.11-lqtaegm
conflict yambo
prepend-path PATH /leonardo/prod/spack/5.2/install/0.21/linux-rhel8-icelake/nvhpc-23.11/yambo-5.2.3-7sf76hoc4w2vpegvb57n2a5wjh2iaqng/bin
prepend-path CMAKE_PREFIX_PATH /leonardo/prod/spack/5.2/install/0.21/linux-rhel8-icelake/nvhpc-23.11/yambo-5.2.3-7sf76hoc4w2vpegvb57n2a5wjh2iaqng/.
setenv YAMBO_HOME /leonardo/prod/spack/5.2/install/0.21/linux-rhel8-icelake/nvhpc-23.11/yambo-5.2.3-7sf76hoc4w2vpegvb57n2a5wjh2iaqng
However, if I remember well, @imarri you already tried it with the same problem, it's right?
The control on the MPD variable was added with commit 2581d6b in the bug-fixes branch (gpl repo).
You can compile it and try to run setting BSSSlepcMPD=500
in the input file as suggested by Jose.
Branch bug-fixes installed on Leonardo
Increasing the system size I get:
12500 is the number of requested eigenvalues (BSSNEig=12500), and 180000 is the BSK matrix size. @nicspalla tried to recompile with
--with-64-bit-indices
without succes.By @imarri