Hi,
We have recently installed the ALPS code, downloaded from the links on the Wiki. We are using version 1.3.2 of the libraries and applications.
The install went smoothly, and the tests worked fine. We have both Cygwin and Ubuntu installs and the problems. We successfully ran some spin models with qmc, dmrg, and exact diagonalization. However, we attempted to run some DMRG calculations with a fermion Hubbard model, and this was unsuccesful. We used a periodic boundary condition chain lattice and a non-interacting system to start with, and use the default "models.xml" and "lattices.xml" that came with the packages. Our input file was
---- BEGIN INPUT FILE, parm-fermion.xml ----
LATTICE="chain lattice" SWEEPS=4 MAXSTATES=26 CONSERVED_QUANTUMNUMBERS="Nup,Ndown" { MODEL="fermion Hubbard", L=10, Nup_total=1, Ndown_total=1 mu=1.0 t=1.0 U=0 NUMBER_EIGENVALUES=1 MEASURE_LOCAL[local spin up density]=n_up MEASURE_CORRELATIONS[up up Green function]="cdag_up:c_up" MEASURE_CORRELATIONS[up down Green function]="cdag_up:c_down" MEASURE_CORRELATIONS[down down Green function]="cdag_down:c_down" MEASURE_CORRELATIONS[up density up density correlation function]=n_up MEASURE_CORRELATIONS[down density down density correlation function]=n_down MEASURE_CORRELATIONS[up density down density correlation function]="n_up:n_down" }
---- END INPUT FILE ----
When we run this ("parameter2xml parm-fermion" then "dmrg parm-fermion.in.xml"), though the energy appears to be correct, we (1) obtain correlation functions which are completely different than the expected result (often coming out to be zero) and (2) most of the site-to-site correlations (e.g., site i to site j, with i not equal to zero) are missing. A symptom may be provided by the (stderr?) output of the code: "Did not get the right number of measurements". The problem is independent of the SWEEPS and MAXSTATES used. If we run the exact diagonalization code "sparsediag", however, we obtain the expected results. So it seems we are setting up something incorrectly with the dmrg input.
To further examine the problem, we re-ran with open boundary conditions (just changing "chain lattice" into "open chain lattice" in the file above). Although now the right number of measurements appear to be taken and the right correlation function elements are non-zero, the results still disagree with our expectations and exact diagonalization.
Finally, we obtain similar problems with bosons. Spin models, however, work without problem and agree with exact diagonalization.
We have not included the output files since they are somewhat lengthy and should be reproducible, but if it would help we would be happy to send them.
Thanks for any help and the wonderful program! Kaden and Stefan
Hi,
Just a quick update. We fixed the periodic boundary issue and are looking into the other discrepancies you mentioned. You can expect a new release soon.
Matthias
On 1 Jul 2008, at 11:57, Kaden Hazzard wrote:
Hi,
We have recently installed the ALPS code, downloaded from the links on the Wiki. We are using version 1.3.2 of the libraries and applications.
The install went smoothly, and the tests worked fine. We have both Cygwin and Ubuntu installs and the problems. We successfully ran some spin models with qmc, dmrg, and exact diagonalization. However, we attempted to run some DMRG calculations with a fermion Hubbard model, and this was unsuccesful. We used a periodic boundary condition chain lattice and a non-interacting system to start with, and use the default "models.xml" and "lattices.xml" that came with the packages. Our input file was
---- BEGIN INPUT FILE, parm-fermion.xml ----
LATTICE="chain lattice" SWEEPS=4 MAXSTATES=26 CONSERVED_QUANTUMNUMBERS="Nup,Ndown" { MODEL="fermion Hubbard", L=10, Nup_total=1, Ndown_total=1 mu=1.0 t=1.0 U=0 NUMBER_EIGENVALUES=1 MEASURE_LOCAL[local spin up density]=n_up MEASURE_CORRELATIONS[up up Green function]="cdag_up:c_up" MEASURE_CORRELATIONS[up down Green function]="cdag_up:c_down" MEASURE_CORRELATIONS[down down Green function]="cdag_down:c_down" MEASURE_CORRELATIONS[up density up density correlation function]=n_up MEASURE_CORRELATIONS[down density down density correlation function]=n_down MEASURE_CORRELATIONS[up density down density correlation function]="n_up:n_down" }
---- END INPUT FILE ----
When we run this ("parameter2xml parm-fermion" then "dmrg parm- fermion.in.xml"), though the energy appears to be correct, we (1) obtain correlation functions which are completely different than the expected result (often coming out to be zero) and (2) most of the site-to-site correlations (e.g., site i to site j, with i not equal to zero) are missing. A symptom may be provided by the (stderr?) output of the code: "Did not get the right number of measurements". The problem is independent of the SWEEPS and MAXSTATES used. If we run the exact diagonalization code "sparsediag", however, we obtain the expected results. So it seems we are setting up something incorrectly with the dmrg input.
To further examine the problem, we re-ran with open boundary conditions (just changing "chain lattice" into "open chain lattice" in the file above). Although now the right number of measurements appear to be taken and the right correlation function elements are non-zero, the results still disagree with our expectations and exact diagonalization.
Finally, we obtain similar problems with bosons. Spin models, however, work without problem and agree with exact diagonalization.
We have not included the output files since they are somewhat lengthy and should be reproducible, but if it would help we would be happy to send them.
Thanks for any help and the wonderful program! Kaden and Stefan
comp-phys-alps-users@lists.phys.ethz.ch