Dear All, I have been running the following input: <pasted at the bottom>
I have observed that for Bose-Hubbard model on chain of length L=256, unit filled, under open boundary conditions the DMRG code uses over 32G memory (this is the hardware limit, it clearly wants more, for L=128 it was around 20G). is that normal? I would estimate that an MPS state of 256 sites, with bond dimension 60, local Hilbert space 7 should be much smaller (even if we allow for a small constant factor for temporary data)
Kind Regards, Mateusz Łącki
LATTICE="open chain lattice" MODEL="boson Hubbard" SWEEPS=8 Nmax=6 NUMBER_EIGENVALUES=1 L=256 CONSERVED_QUANTUMNUMBERS="N" N_total=256 VERBOSE=0
t=0.25 U=1 mu=0
{MAXSTATES=60}
Dear Mateusz,
I guess you are using the dmrg application in ALPS, right?
If you have a recent ALPS version you may try the MPS applications, i.e. running exactly the same parameters with the application 'mps_optim'. More details here: http://arxiv.org/abs/1407.0872
If the MPS application exceeds memory you can dump some temporary objects to disk by setting the parameter "storagedir".
Best regards, Michele
On 01.09.2014 19:10, Mateusz Łącki wrote:
Dear All, I have been running the following input:
<pasted at the bottom>
I have observed that for Bose-Hubbard model on chain of length L=256, unit filled, under open boundary conditions the DMRG code uses over 32G memory (this is the hardware limit, it clearly wants more, for L=128 it was around 20G). is that normal? I would estimate that an MPS state of 256 sites, with bond dimension 60, local Hilbert space 7 should be much smaller (even if we allow for a small constant factor for temporary data)
Kind Regards, Mateusz Łącki
LATTICE="open chain lattice" MODEL="boson Hubbard" SWEEPS=8 Nmax=6 NUMBER_EIGENVALUES=1 L=256 CONSERVED_QUANTUMNUMBERS="N" N_total=256 VERBOSE=0
t=0.25 U=1 mu=0
{MAXSTATES=60}
Dear Michele, thanks for the answer! Indeed the code looks superb.
Regards, Mateusz
2014-09-02 13:27 GMT+02:00, Michele Dolfi dolfim@phys.ethz.ch:
Dear Mateusz,
I guess you are using the dmrg application in ALPS, right?
If you have a recent ALPS version you may try the MPS applications, i.e. running exactly the same parameters with the application 'mps_optim'. More details here: http://arxiv.org/abs/1407.0872
If the MPS application exceeds memory you can dump some temporary objects to disk by setting the parameter "storagedir".
Best regards, Michele
On 01.09.2014 19:10, Mateusz Łącki wrote:
Dear All, I have been running the following input:
<pasted at the bottom>
I have observed that for Bose-Hubbard model on chain of length L=256, unit filled, under open boundary conditions the DMRG code uses over 32G memory (this is the hardware limit, it clearly wants more, for L=128 it was around 20G). is that normal? I would estimate that an MPS state of 256 sites, with bond dimension 60, local Hilbert space 7 should be much smaller (even if we allow for a small constant factor for temporary data)
Kind Regards, Mateusz Łącki
LATTICE="open chain lattice" MODEL="boson Hubbard" SWEEPS=8 Nmax=6 NUMBER_EIGENVALUES=1 L=256 CONSERVED_QUANTUMNUMBERS="N" N_total=256 VERBOSE=0
t=0.25 U=1 mu=0
{MAXSTATES=60}
-- 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
comp-phys-alps-users@lists.phys.ethz.ch