Difference between revisions of "Comments: which code to choose for your calculation"

From ALPS
Jump to: navigation, search
(Based on Prof. Troyer's short presentation)
 
Line 1: Line 1:
 
{{Languages|Comments:_which_code_to_choose_for_your_calculation}}
 
{{Languages|Comments:_which_code_to_choose_for_your_calculation}}
  
 +
There are currently four QMC representations / algorithms: looper, dirloop_sse, worm, and quantum Wang-Landau.
  
* looper: models with inversion symmetry in spin space
+
All four methods (except sometimes Looper) may be used to study (unfrustrated) spin models. Only worm and sometimes dirloop_sse may be used for boson models.
* dirloop_sse: models with anisotropy that breaks  inversion symmetry in spin space e.g., Heisenberg models in a magnetic field
 
* worm: Bose-Hubbard models or spin models in very strong fields
 
* quantum Wang-Landau: calculations of free energy and entropy
 
  
Simple Heisenberg models without a magnetic field can be simulated with either of these.
+
1. Looper: 
Looper has the smaller range of applicability, but if applicable, it shows the best performance (shortest autocorrelation time).
+
Only usable for models with inversion symmetry in spin space (for Heisenberg models, no magnetic field). Looper has the smaller range of applicability, but if applicable, it shows the best performance (shortest autocorrelation time).
 +
 
 +
2. dirloop_sse:
 +
Stochastic Series Expansion representation, using directed loops (essentially worms). Good for spin models with anisotropy that breaks  inversion symmetry in spin space e.g., Heisenberg models in a magnetic field. Also good for hard core bosons, with at most one boson per site. Extremely inefficient for soft core boson models where a few bosons on a site give a very large U term in the Hamiltonian. Can measure the Green function.
 +
 
 +
3. Worm:
 +
Path integral representation, using worms. Good for Bose-Hubbard models and for spin models in very strong fields. Can simulate Bose-Hubbard models also with non-small filling (set the parameter N_max).
 +
[If you have an action which is non-local in time, the path integral representation in the worm algorithm is a good starting point to write your own code.]
 +
 
 +
4. quantum Wang-Landau:
 +
Good for calculations of free energy and entropy.

Latest revision as of 01:08, 6 June 2012

There are currently four QMC representations / algorithms: looper, dirloop_sse, worm, and quantum Wang-Landau.

All four methods (except sometimes Looper) may be used to study (unfrustrated) spin models. Only worm and sometimes dirloop_sse may be used for boson models.

1. Looper: Only usable for models with inversion symmetry in spin space (for Heisenberg models, no magnetic field). Looper has the smaller range of applicability, but if applicable, it shows the best performance (shortest autocorrelation time).

2. dirloop_sse: Stochastic Series Expansion representation, using directed loops (essentially worms). Good for spin models with anisotropy that breaks inversion symmetry in spin space e.g., Heisenberg models in a magnetic field. Also good for hard core bosons, with at most one boson per site. Extremely inefficient for soft core boson models where a few bosons on a site give a very large U term in the Hamiltonian. Can measure the Green function.

3. Worm: Path integral representation, using worms. Good for Bose-Hubbard models and for spin models in very strong fields. Can simulate Bose-Hubbard models also with non-small filling (set the parameter N_max). [If you have an action which is non-local in time, the path integral representation in the worm algorithm is a good starting point to write your own code.]

4. quantum Wang-Landau: Good for calculations of free energy and entropy.