Dear ALPS developers and users,
I am using the DMRG algorithm to check the entropic properties of the
ground state of an Heisenberg chain. More specifically in the region
Jx=Jy=1 and Jz>1 , which is a gapped region. I sweep until I see that
the entanglement entropy repeats on a certain pattern through next
sweeps. At that stage the simulation should have converged on a good
approximation of the ground state, and I take my data.
For systems close to the critical point (e.g. Jz=1.001) the results I
get are comparable to exact results, and everything works perfectly.
When I increase the gap though, I manage to get sensible results only on
very short chains.
To explain you what happens on longer chains I am attaching to this mail
a plot of the entropy against the iterations. This picture refers to
Jx=Jy=1
Jz=2.5
L=42
Nstates=400
Nsweeps=40
periodic boundary conditions
As you can see it struggles to converge to a sensible value. This
picture is a good representative of any simulation I run. There are two
features I see:
1- jumps from a lower value to an higher one: I think this could be explained
with the fact that as the system sweeps, it gets short range correlations
and it converges immediately to another ground state.
2- the slow convergence in the final stage to a sensible value. This I really don't know how to explain.
Do you know what is going on, and how could I enhance the simulation?
Just increasing the number of sweeps or there is a smarter way in your
opinion? (It looks like increasing the states kept doesn't change too
much the situation, and I really need periodic boundaries for the kind of plot I want to produce).
Many thanks again for your precious help.
Kindest regards.
--
Emanuele Levi
emanuele.levi@gmail.com