Dear Dr. Chern,
Cc: comp-phys-alps-users
From: "CHERN, Chyh-Hong" <chern(a)issp.u-tokyo.ac.jp>
Date: Wed, 30 May 2007 13:58:20 +0900
> Dear Professor Todo,
>
> Sorry for occupying your valuable time. We encounter some problems when
> using your ALPS. We will appreciate highly if you can kindly help.
>
> I am a postdoctoral researcher in Oshikawa's group in ISSP. Recently, we
> are interested in the transverse Ising model in the kagome system. We are
> doing the Monte Carlo simulation and would like use your exact
> diagonalization in ALPS to calibrate. I attached the lattice.xml and parm1
> that I used. We doubt on the results and would like to listen to your
> comments. Let me list what happened to us.
>
> 1. We obtained the exactly the same results for L=1 and L=2 on energy,
> entropy, specific heat, and the free energy. When running L=3, we failed to
> get any results even though the memory is only used ~ 25%.
Your kagome lattice with L=3 has 27 vertices, which obviously requires
too much amount of memory (8*(2^27)^2 bytes = 10^8 GB!) to store the
full Hamiltonian matrix...
Instead of the fulldiag code, you can use my qmc code (qmc/looper), as
long as the diagonal (Sz) interaction is ferromagnetic (i.e. no
frustration).
> 2. If we do not add CONSERVED_QUANTNUMBERS="Sz" in the parm1, we do not have
> the result on the susceptibility but do have others. If we do include that
> line, we have susceptibility, but lose the results on other thermodynamics
> quantities. As we know, Sz is not the conserved quantum number for the
> Ising model in the transverse magnetic field.
Calculating canonical correlations of the operators which do not
commute with the Hamiltonian requires substantial CPU time (N^3?).
This is the reason why the fulldiag skips measuring the susceptibility
unless total Sz is a conserved quantity.
Again you can use qmc/loop, which calculates the susceptibility even
in this case, instead for calibrating your simulation code.
Best regards,
Synge