Dear All, I wish to report a suboptimal behaviour in the mps_optim:
setting the initial state with wrong L (L=10 and there are 12 spins defined)
L=10;
init_state="local_quantumnumbers" initial_local_S="0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5" initial_local_Sz="0.5,-0.5,0.5,0.5,-0.5,0.5,-0.5,-0.5,0.5,-0.5,0.5,0.5”
produces a nasty segfault with no warning (in almost all other cases of setting something wrong a message explaining a nature of the problem is printed). This is not a bug, but a nuissance.
Best, Mateusz Łącki
Dear Mateusz,
Thanks for reporting that. I increased the checks, now the error message should be more consistent.
Best regards, Michele
-- ETH Zurich Michele Dolfi Institute for Theoretical Physics HIT G 32.4 Wolfgang-Pauli-Str. 27 8093 Zurich Switzerland
dolfim@phys.ethz.ch www.itp.phys.ethz.ch
+41 44 633 78 56 phone +41 44 633 11 15 fax
On Aug 22, 2015, at 6:07 PM, Mateusz Łącki mateusz.lacki@gmail.com wrote:
Dear All, I wish to report a suboptimal behaviour in the mps_optim:
setting the initial state with wrong L (L=10 and there are 12 spins defined)
L=10;
init_state="local_quantumnumbers" initial_local_S="0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5" initial_local_Sz="0.5,-0.5,0.5,0.5,-0.5,0.5,-0.5,-0.5,0.5,-0.5,0.5,0.5”
produces a nasty segfault with no warning (in almost all other cases of setting something wrong a message explaining a nature of the problem is printed). This is not a bug, but a nuissance.
Best, Mateusz Łącki
Dear Michele, I would also like to give user’s feedback.
mps_optim writes states as a series of mpsXX.h5 files in s *chkp subdirectory. While this is convenient, I have encountered many situations when I wanted to do lots of computations, but the quantity number_of_vectors*chain_length was large to the point of crashing the filesystem or exceeding file number quotas.
While in all cases the problem is actually with the system (hard to persuade administrator, old file system), still my experience tells that the "de facto” usability could be icreased by limiting the number of files written per run (currently it is O(L), while O(1) would be more convenient).
Best, Mateusz Łącki
On 24 Aug 2015, at 09:28, Michele Dolfi dolfim@phys.ethz.ch wrote:
Dear Mateusz,
Thanks for reporting that. I increased the checks, now the error message should be more consistent.
Best regards, Michele
-- ETH Zurich Michele Dolfi Institute for Theoretical Physics HIT G 32.4 Wolfgang-Pauli-Str. 27 8093 Zurich Switzerland
dolfim@phys.ethz.ch mailto:dolfim@phys.ethz.ch www.itp.phys.ethz.ch http://www.itp.phys.ethz.ch/
+41 44 633 78 56 phone +41 44 633 11 15 fax
On Aug 22, 2015, at 6:07 PM, Mateusz Łącki <mateusz.lacki@gmail.com mailto:mateusz.lacki@gmail.com> wrote:
Dear All, I wish to report a suboptimal behaviour in the mps_optim:
setting the initial state with wrong L (L=10 and there are 12 spins defined)
L=10;
init_state="local_quantumnumbers" initial_local_S="0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5" initial_local_Sz="0.5,-0.5,0.5,0.5,-0.5,0.5,-0.5,-0.5,0.5,-0.5,0.5,0.5”
produces a nasty segfault with no warning (in almost all other cases of setting something wrong a message explaining a nature of the problem is printed). This is not a bug, but a nuissance.
Best, Mateusz Łącki
Dear Mateusz,
It used to be a single file, but at some point we decided to change format for higher I/O (on parallel filesystems) and parallelism. If you need to keep the MPS after a simulation we usually compress it in a tarball. Otherwise you could specify the `chkpfile` parameter to provide a location on a different (maybe temporary) filesystem, which you can easily prune at the end of the simulation.
Michele
-- ETH Zurich Michele Dolfi Institute for Theoretical Physics HIT G 32.4 Wolfgang-Pauli-Str. 27 8093 Zurich Switzerland
dolfim@phys.ethz.ch www.itp.phys.ethz.ch
+41 44 633 78 56 phone +41 44 633 11 15 fax
On Aug 24, 2015, at 9:46 AM, Mateusz Łącki mateusz.lacki@gmail.com wrote:
Dear Michele, I would also like to give user’s feedback.
mps_optim writes states as a series of mpsXX.h5 files in s *chkp subdirectory. While this is convenient, I have encountered many situations when I wanted to do lots of computations, but the quantity number_of_vectors*chain_length was large to the point of crashing the filesystem or exceeding file number quotas.
While in all cases the problem is actually with the system (hard to persuade administrator, old file system), still my experience tells that the "de facto” usability could be icreased by limiting the number of files written per run (currently it is O(L), while O(1) would be more convenient).
Best, Mateusz Łącki
On 24 Aug 2015, at 09:28, Michele Dolfi <dolfim@phys.ethz.ch mailto:dolfim@phys.ethz.ch> wrote:
Dear Mateusz,
Thanks for reporting that. I increased the checks, now the error message should be more consistent.
Best regards, Michele
-- ETH Zurich Michele Dolfi Institute for Theoretical Physics HIT G 32.4 Wolfgang-Pauli-Str. 27 8093 Zurich Switzerland
dolfim@phys.ethz.ch mailto:dolfim@phys.ethz.ch www.itp.phys.ethz.ch http://www.itp.phys.ethz.ch/
+41 44 633 78 56 phone +41 44 633 11 15 fax
On Aug 22, 2015, at 6:07 PM, Mateusz Łącki <mateusz.lacki@gmail.com mailto:mateusz.lacki@gmail.com> wrote:
Dear All, I wish to report a suboptimal behaviour in the mps_optim:
setting the initial state with wrong L (L=10 and there are 12 spins defined)
L=10;
init_state="local_quantumnumbers" initial_local_S="0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5" initial_local_Sz="0.5,-0.5,0.5,0.5,-0.5,0.5,-0.5,-0.5,0.5,-0.5,0.5,0.5”
produces a nasty segfault with no warning (in almost all other cases of setting something wrong a message explaining a nature of the problem is printed). This is not a bug, but a nuissance.
Best, Mateusz Łącki
comp-phys-alps-users@lists.phys.ethz.ch