Hi Mattias, Should i use Jz=1, Jxy=1 and Jz0=1, Jxy0=1, Jz1=1, Jxy1=1 setting for my experiment? Can use anisotropic Heisenberg model for QWL simulations with ALPS rc4? Thanks, Ruben ________________________________ From: Matthias Troyer <troyer@phys.ethz.ch> To: comp-phys-alps-users@lists.phys.ethz.ch Sent: Sat, December 18, 2010 12:51:18 PM Subject: Re: [ALPS-users] anisotropic Heisenberg model setup Indeed, J is just a short cut for J0. Thus if you only set J, then J1=0 Matthias On Dec 18, 2010, at 4:50 PM, Ruben Ghulghazaryan wrote:

Hi ALPS team,

I am trying to make simple simulations with anisotropic Heisenberg model on a ladder with ALPS2 rc2.

According to models.xml and full diagonalization web page if I specify J0 and J1 parameters then for different bond types "0" and "1" I will have J0 and J1 coefficients in Heisenberg ("spin") model. I did a simple experiment by performing fulldiag two simulations: one by using J=1 and the other J0=1,J1=1 and compared the results.

I expected that the results should be the same according to "spin" model definition: defaults for J0and J1 is J if they are not specified. But I found that the results of energy and other quantities are different. below are my two parameter files:

LATTICE="ladder" MODEL="spin" local_S = 1 J = 1 CONSERVED_QUANTUMNUMBERS="Sz" {L = 4}

++++++++++++++++++++++++

LATTICE="ladder" MODEL="spin" local_S = 1 J0 = 1 J1 = 1 CONSERVED_QUANTUMNUMBERS="Sz" {L = 4}

========================== I run fulldiag then fulldiag_evaluate

fulldiag_evaluate --T_MIN 0.1 --T_MAX 10 --DELTA_T 0.1 parm6a.task1.out.xml

as given in the web page http://alps.comp-phys.org/mediawiki/index.php/ALPS_2_Tutorials:ED-06_FullDia... n

Would you please help me to understand why the results are different?

Thanks, Ruben

Dear Mattias, Thank you very much for your letter. Please find my questions and comments below. Thank you in advance, Ruben ________________________________ From: Matthias Troyer <troyer@phys.ethz.ch> To: comp-phys-alps-users@lists.phys.ethz.ch Sent: Sat, December 18, 2010 1:06:54 PM Subject: Re: [ALPS-users] anisotropic Heisenberg model setup On Dec 18, 2010, at 5:02 PM, Ruben Ghulghazaryan wrote:

Hi Mattias,

Should i use Jz=1, Jxy=1 and Jz0=1, Jxy0=1, Jz1=1, Jxy1=1 setting for my experiment?

This would be isotropic and not anisotropic. ---> That's correct. This is why, we should expect that the results of simulations be absolutely the same for fulldiag program. But simulations show different results for Jz=1, Jxy=1 and Jz0=1, Jxy0=1, Jz1=1, Jxy1=1 settings. That is confusing. I want to make sure that anisotropic program works as expected that is why I want to make this experiment and check whether anisotropic model with the same settings for different bond types and isotropic model with the same interaction bond term give same results. My tests show that the results are not the same and this may be an indication of a problem with program or miss usage from my side. Would you please help me to understand the problem?

Can use anisotropic Heisenberg model for QWL simulations with ALPS rc4?

No, the qwl at the moment can only do isotropic Heisenberg models at the moment. --> Ok. Thanks. Matthias

Thanks, Ruben

From: Matthias Troyer <troyer@phys.ethz.ch> To: comp-phys-alps-users@lists.phys.ethz.ch Sent: Sat, December 18, 2010 12:51:18 PM Subject: Re: [ALPS-users] anisotropic Heisenberg model setup

Indeed, J is just a short cut for J0. Thus if you only set J, then J1=0

Matthias

On Dec 18, 2010, at 4:50 PM, Ruben Ghulghazaryan wrote:

...

Hi ALPS team,

I am trying to make simple simulations with anisotropic Heisenberg model on a ladder with ALPS2 rc2.

According to models.xml and full diagonalization web page if I specify J0 and J1 parameters then for different bond types "0" and "1" I will have J0 and J1 coefficients in Heisenberg ("spin") model. I did a simple experiment by performing fulldiag two simulations: one by using J=1 and the other J0=1,J1=1 and compared the results.

...

I expected that the results should be the same according to "spin" model

definition: defaults for J0and J1 is J if they are not specified. But I found that the results of energy and other quantities are different.

...

below are my two parameter files:

LATTICE="ladder" MODEL="spin" local_S = 1 J = 1 CONSERVED_QUANTUMNUMBERS="Sz" {L = 4}

++++++++++++++++++++++++

LATTICE="ladder" MODEL="spin" local_S = 1 J0 = 1 J1 = 1 CONSERVED_QUANTUMNUMBERS="Sz" {L = 4}

========================== I run fulldiag then fulldiag_evaluate

fulldiag_evaluate --T_MIN 0.1 --T_MAX 10 --DELTA_T 0.1

parm6a.task1.out.xml

...

as given in the web page

http://alps.comp-phys.org/mediawiki/index.php/ALPS_2_Tutorials:ED-06_FullDia...

...

Would you please help me to understand why the results are different?

Thanks, Ruben

Hi Mattias, By anisotropic model on a ladder I mean that we have two different interaction terms in Hamiltonian as shown in the file attached. Thus we have two bond types "0" and "1" in horizontal and vertical directions and we want to have different interaction terms in horizontal and vertical directions. As far as I understand, according to "spin" model definition J0 parameter should be for bond type="0" and J1 for bonds type="1". If we set J=1 only then (even if J1=0) all bonds irrespective of their type should get 1 as bond term, don't you? In the case we specify J0=1 and J1=1 again all bonds should get 1 as bond term and the model results should be the same as for setting J only. In this case both isotropic and anisotropic models should give similar results. Am I missed something? Regards, Ruben From: Matthias Troyer <troyer@phys.ethz.ch> To: comp-phys-alps-users@lists.phys.ethz.ch Sent: Mon, December 20, 2010 11:39:53 AM Subject: Re: [ALPS-users] anisotropic Heisenberg model setup On Dec 20, 2010, at 2:47 PM, Ruben Ghulghazaryan wrote: Dear Mattias,

Thank you very much for your letter. Please find my questions and comments below.

Thank you in advance, Ruben

________________________________ From: Matthias Troyer <troyer@phys.ethz.ch>

To: comp-phys-alps-users@lists.phys.ethz.ch Sent: Sat, December 18, 2010 1:06:54 PM Subject: Re: [ALPS-users] anisotropic Heisenberg model setup

On Dec 18, 2010, at 5:02 PM, Ruben Ghulghazaryan wrote:

...

Hi Mattias,

Should i use Jz=1, Jxy=1 and Jz0=1, Jxy0=1, Jz1=1, Jxy1=1 setting for my experiment?

This would be isotropic and not anisotropic.

---> That's correct. This is why, we should expect that the results of simulations be absolutely the same for fulldiag program. But simulations show different results for Jz=1, Jxy=1 and Jz0=1, Jxy0=1, Jz1=1, Jxy1=1 settings. That is confusing.

That is because as I said, J=J0 but not J1, thus if you set only J, the J1 terms will all be zero. I want to make sure that anisotropic program works as expected that is why I want to make this experiment and check whether anisotropic model

with the same settings for different bond types and isotropic model with the same interaction bond term give same results. My tests show that the results are not the same and this may be an indication of a problem with program or miss usage from my side.

What do you mean by "isotropic model" and "anisotropic model"? Matthias

Dear ALPS team, I am having trouble to install alps-2.0.0rc4 with MKL support (I tried several things). What is the proper cmake command to compile with LAPACK and BLAS support from MKL 10.x ? Best regards, H. Hafermann

Dear Ryo, thanks a lot for the reply. Setting -DBLA_VENDOR did not work for me, but looking at FindLAPACK.cmake gave some hints. In the end setting export MKL_HOME=$SOFTSRV/intel/mkl/10.0.011 and on the cmake command line cmake ... -DQMC_BITS=64 -DLAPACK_LINKER_FLAGS="-lmkl_sequential -lmkl_core -lmkl_intel_lp64" ... worked fine. Best regards, Hartmut Le 21.12.2010 à 04:33, Ryo IGARASHI a écrit :

Hi, Harmut,

On Tue, Dec 21, 2010 at 5:31 AM, Hartmut Hafermann <hartmut.hafermann@cpht.polytechnique.fr> wrote:

...

I am having trouble to install alps-2.0.0rc4 with MKL support (I tried several things). What is the proper cmake command to compile with LAPACK and BLAS support from MKL 10.x ?

Well, LAPACK/BLAS setting in ALPS is rather complicated, but if you have

If you have Fortran compiler installed, and if you install MKL in the usual directory, and if you compile on x86_64 machine (64bit), -DBLA_VENDOR=Intel10_64lp will be the easiest setting. This setting comes from CMake upstream, so look at the $CMAKE_ROOT/lib/cmake-2.8/Modules/FindLAPACK.cmake for further information (and naming convention for BLA_VENDOR).

Note that this setting is executed at the end of LAPACK detection code from ALPS, you may have to remove MKL related environment variable.

Best regards, -- Ryo IGARASHI, Ph.D. rigarash@hosi.phys.s.u-tokyo.ac.jp OpenPGP fingerprint: BAD9 71E3 28F3 8952 5640 6A53 EC79 A280 6A19 2319