SciPost Submission Page
The ALF (Algorithms for Lattice Fermions) project release 1.0. Documentation for the auxiliary field quantum Monte Carlo code
by Martin Bercx, Florian Goth, Johannes S. Hofmann, Fakher F. Assaad
This is not the latest submitted version.
This Submission thread is now published as
Submission summary
| Authors (as registered SciPost users): | Fakher Assaad · Martin Bercx · Florian Goth · Johannes Stephan Hofmann |
| Submission information | |
|---|---|
| Preprint Link: | http://arxiv.org/abs/1704.00131v1 (pdf) |
| Date submitted: | 2017-04-06 02:00 |
| Submitted by: | Bercx, Martin |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Computational |
Abstract
The Algorithms for Lattice Fermions package provides a general code for the finite temperature auxiliary field quantum Monte Carlo algorithm. The code is engineered to be able to simulate any model that can be written in terms of sums of single-body operators, of squares of single-body operators and single-body operators coupled to an Ising field with given dynamics. We provide predefined types that allow the user to specify the model, the Bravais lattice as well as equal time and time displaced observables. The code supports an MPI implementation. Examples such as the Hubbard model on the honeycomb lattice and the Hubbard model on the square lattice coupled to a transverse Ising field are provided and discussed in the documentation. We furthermore discuss how to use the package to implement the Kondo lattice model and the $SU(N)$ Hubbard-Heisenberg model. One can download the code from our Git instance at \path{https://alf.physik.uni-wuerzburg.de} and sign in to file issues.
Current status:
Reports on this Submission
Report #2 by Anonymous (Referee 2) on 2017-5-22 (Invited Report)
- Cite as: Anonymous, Report on arXiv:1704.00131v1, delivered 2017-05-22, doi: 10.21468/SciPost.Report.144
Strengths
1) first open-source code for lattice fermion models
2) impressively versatile framework
3) broad applicability in terms of model Hamiltonians
4) state-of-the-art implementation and documentation
Weaknesses
1) no compatibility with other similar open-source packages likes ALPS, iTensor
2) no roadmap to include obvious extensions like projective zero-temperature QMC approach
3) no community building effort
Report
The manuscript at hand accompanies an open-source project devoted to a state-of-the-art, parallelized implementation of auxiliary field quantum Monte Carlo approaches for lattice fermion models. Its primary purpose is to provide a documentation of this Fortran 90 code package and illustrate via a number of examples its impressive versatility that allows for the simulation of a broad variety of fermionic lattice models.
This is an important contribution that complements related open-source projects such as the ALPS project or the iTensor package that have been widely utilized over the last decade and found broad appreciation beyond the numerical community. Nevertheless, there is no standardized approach to credit the original authors of such open-source codes in the scientific community and it is therefore of paramount importance to allow for the publication of an accompanying documentation such that scientific credit can be measured in terms of citations.
I therefore strongly endorse the publication of the manuscript at hand in its current form with some minor revisions (see below).
Requested changes
While the manuscript is basically a documentation accompanying an open-source code, it is generally very well written. But I do want to urge the authors to expand the introduction to (i) include a broader introduction into the topic of open-source codes and community codes, and (ii) make a connection to similar efforts such as the ALPS project or the iTensor project (and probably motivate why there is no cross-compatibility with these codes).
One minor point of revision regards the discussion of the sign problem. In particular, Equation (27) should not be credited to the Troyer&Wiese PRL (as done in the introduction), but should be credited to the Loh,Gubernatis, … PRB of 1990.
When referring to introductory readings on fermionic QMC flavors, I suggest to add the very nice book of Gubernatis, Kawashima and Werner (Cambridge University Press, 2016).
In the discussion at the end of the manuscript, I would have expected to find a roadmap for future releases and whether the ALF project is expected to turn into a community project (similar to the ALPS project) with contributors beyond the Würzburg group.
Report #1 by Anonymous (Referee 1) on 2017-5-8 (Invited Report)
- Cite as: Anonymous, Report on arXiv:1704.00131v1, delivered 2017-05-08, doi: 10.21468/SciPost.Report.130
Strengths
1) Highly useful tool, extremely valuable and helpful and welcome
2) Very versatile, large class of models
3) Very impressive complete package, documentation, and examples.
4) Flexible re-binning tool for error analysis.
Weaknesses
Minor.
1) Introduction could be more clear on which models can and cannot be simulated with the present software
2) Autocorrelations: more guidance needed in a black box package
3) Information sometimes a bit terse
Report
This software package for auxiliary field quantum Monte Carlo is an extremely valuable and helpful and welcome tool which had been sorely missing from the existing code packages. It is a very impressive complete package with software, documentation and examples, ready to simulate a very large class of Hamiltonians. This referee would like to thank the authors for providing such a big service to the community.
The documentation is clearly written and complete. Some small additions would be welcome in order to improve accessibility and ease of use of the package, as explained below:
The introduction first provides a quick but very wide review of existing techniques and some corresponding models. The switch to the present package is a bit fuzzy and may leave the reader wondering which aspects of the previous review are included. The subsequent definition of the Hamiltonian is technically complete, but it would be really helpful to have a brief discussion of some models which are or are not included.
In the example section, some repeated information would help, escpecially about which files contain which kind of information (input, output).
The present package can be used almost as a black box. Therefore, some guidance is necessary to prevent accidents. Guidance is nicely provided with respect to stabilization, but not enough with respect to autocorrelations (see below).
Overall, with a few additions in the documentation, I highly recommend publication of this paper.
Requested changes
1) Chapter 1.2: mention some examples of which models can or cannot be simulated. Mention that "flavor" includes spin. Might mention that H_I formally includes the Hamiltonians of H_T.
2) Example section: could repeat which files contain what kind of information.
3) Autocorrelations: Need more discussion. There is no unique autocorrelation time T_auto. The autocorrelation time for convergence depends on the observable and can be very different for e.g. energies vs. long distance correlations. Thermalization usually needs much more than a single T_auto. Like expressed in the paper for the case of stabilization, the user needs to know that convergence ,,has to be judged for every simulation'', and indeed every observable (see for example the ALPS package). The possibility of re-binning is very nice. Some hint would be helpful on how fast to increase bin-sizes and on how many independent bins are necessary for a reliable jackknife error.
4) Page 4: Re(sign) instead of |sign| ?
5) Jackknife: might mention how the sign is handled (eq. 24).
6) For stabilization: scaling with L and beta, timing ?
Author: Martin Bercx on 2017-06-26 [id 151]
(in reply to Report 1 on 2017-05-08)
We would like to thank the referee for the very positive evaluation and their constructive report.
In the following we respond to the points raised by the referee:
1)
"Chapter 1.2: mention some examples of which models can or cannot be simulated."
At the end of Section 1.2, we have added a paragraph mentioning a class of models that cannot be simulated with ALF. The set of shortcomings listed in this paragraph will actually define future directions for the development of our package.
"Mention that "flavor" includes spin."
We have included the following paragraph (Section 1.2):
"Both the color and the flavor index can describe the spin degree of freedom, the choice depending on the spin symmetry
of the simulated model and the HS transformation. This point is illustrated in the examples, see Sec. 4.1 and 4.2."
"Might mention that H_I formally includes the Hamiltonians of H_T."
Please note that the fermionic hopping term H_T and the term H_I describing fermionic hopping coupled to Ising spins are in general independent terms.
2)
"Example section: could repeat which files contain what kind of information."
We have added the following paragraph to the example section:
"The input files are 'parameters' and 'seeds' (see Tab. 8). The output files are 'info', 'confout', and files with suffixes '_scal', '_eq', and '_tau' that
contain the raw measurements (see Tab. 9)."
3)
"Autocorrelations: Need more discussion. There is no unique autocorrelation time T_auto. The autocorrelation time for convergence depends on the observable and can be very different for e.g. energies vs. long distance correlations. Thermalization usually needs much more than a single T_auto. Like expressed in the paper for the case of stabilization, the user needs to know that convergence ,,has to be judged for every simulation'', and indeed every observable (see for example the ALPS package). The possibility of re-binning is very nice. Some hint would be helpful on how fast to increase bin-sizes and on how many independent bins are necessary for a reliable jackknife error"
We agree with the referee that there is no unique autocorrelation time. In the revised version of the manuscript, we have included a much longer version of the discussion of how to determine error bars (see Sec. 2.4). We have also included a highly non-trivial example such that the user can appreciate the intricacies involved in determining autocorrelation times and reliable error bars.
4)
"Page 4: Re(sign) instead of |sign| ?"
Please note that we formulate the weight as |exp[-S(C)]| to account for the, in general, complex phase.
5)
"Jackknife: might mention how the sign is handled (eq. 24)."
The sign is properly handled with the Jackknife method. In the revised version of the manuscript, we have included a paragraph on the Jackknife resampling method (Sec. 2.4.1).
6)
"For stabilization: scaling with L and beta, timing ?"
We have added the following paragraph to the end of Sec. 2.3, which discusses the stabilization:
"While this might seem like quite an effort that has to be performed for every multiplication it has to be noted that even with this stabilization scheme the algorithm preserves the time complexity class of $\mathcal{O}(\beta N^3)$ expressed in the physical parameters inverse temperature $\beta$ and lattice size $N$.
While there is no analytical expression for the dependence of the stability on the physical parameters our experience has been that for a given number of stabilization steps along the imaginary time axis [in the notation of Eq. (36) this number is $L_{\text{Trotter}}/\texttt{NWrap}$], the precision will be largely invariant of the system size $N$, whereas with increasing inverse temperature $\beta$ the number of stabilization steps often has to be increased to maintain a given precision.
Author: Martin Bercx on 2017-06-26 [id 150]
(in reply to Report 2 on 2017-05-22)We would like to thank the referee for carefully reading our manuscript and for the very positive report.
In the following we respond to the points raised by the referee:
"i) include a broader introduction into the topic of open-source codes and community codes, and (ii) make a connection to similar efforts such as the ALPS project or the iTensor project (and probably motivate why there is no cross-compatibility with these codes)."
In the revised version of the manuscript, we have added in the introduction a paragraph placing ALF in the broader context of open source packages for correlated matter. In the conclusions we have touched upon issues of compatibility with other packages. This is certainly desirable, but will be left for future developments.
"One minor point of revision regards the discussion of the sign problem. In particular, Equation (27) should not be credited to the Troyer&Wiese PRL (as done in the introduction), but should be credited to the Loh,Gubernatis, … PRB of 1990."
We have included the above reference.
"When referring to introductory readings on fermionic QMC flavors, I suggest to add the very nice book of Gubernatis, Kawashima and Werner (Cambridge University Press, 2016)."
We have added the reference.
"In the discussion at the end of the manuscript, I would have expected to find a roadmap for future releases and whether the ALF project is expected to turn into a community project (similar to the ALPS project) with contributors beyond the Würzburg group."
At the end of the manuscript, we have added a road map for future releases and new features we plan to add to the ALF package. In the introduction, we have mentioned the efforts we are following to establish a community beyond the Würzburg group.