Paper: Enhancing SfePy with Isogeometric Analysis (Robert Cimrman)

[rc]

Hi! Here it is...

[pdebuyl]

Hi Robert, thanks for submitting. The paper builds fine.

Cheers, Pierre

[khinsen]

I will review this paper. Consider me a non-specialist: I have some notions of finite-element techniques but I have never heard of isogeometric analysis before.

[rc]

The paper should be comprehensible to non-specialists - if it is not, it's a bug. :)

[khinsen]

I left detailed comments in line notes. Overall, you did a very good job explaining FEM in a short space, but the IGA part is too concise, in particular in not defining most of the terms that are being used. What I retain from reading this paper is that IGA's main advantage over FEM is that it doesn't need meshes, but defines smooth basis functions on the whole domain. I just have no idea of how this is done and what the limitations and drawbacks are.

The description of the implementation in SfePy seems clear enough for someone who understands IGA, and the instructions for using IGA in SfePy look quite straightforward.

[rc]

Thank you for your review. The conciseness of the IGA part was deliberate, but now (also after discussing with people at euroscipy) I agree fully that the basic notions should be explained in the article, not only referenced. I will address the notes soon.

[rc]

I have tried to address @khinsen's comments. I still have to proof-read the additions and maybe also reorder the images a bit for a better text flow.

[rc]

I have re-read the paper and fixed a few other things. Awaiting more comments :)

[pdebuyl]

A second reviewer is coming up shortly :-)

[khinsen]

Is there a straightforward way to get a PDF of the current state? By "straightforward", I mean not requiring the installation of a dozen libraries.

[rc]

FYI: I did not need to install anything that I did not have, but I am using sphinx for my docs, and git for my projects.

[pdebuyl]

I updated the draft pdf with all papers at http://pdebuyl.be/tmp/esp2014_draft.pdf

[khinsen]

Thanks @pdebuyl for the PDF, and to @rc for the improvements. I cannot check the equations, but I can follow the overall description, which in fact strikes the right balance between clarity and conciseness. IGA does look like a pretty neat idea: reusing the bases used for describing the domain also for the subsequent computations.

[wd15]

Review of version 31a52c376408, "Enhancing SfePy with Isogeometric Analysis"

Overall this is an excellent paper and I enjoyed reviewing it. I have quite a few suggestions for changes though none of these are mandatory, implementing at least some of the suggestions would greatly benefit the paper.

I am not familiar with IGA/NURBs so I am not qualified to comment on some of the technical aspects of the paper although I am an Sfepy user and I am broadly familiar with finite element analysis. Many of my criticisms are aimed at the style of the paper rather than the technical parts.

• [ ] Scipy attendees are drawn from extremely varied scientific backgrounds and therefore it would be helpful if the paper was more approachable to a general audience especially the abstract and introduction.
• [ ] The abstract reads more like an introduction. I think the results from the paper need to be highlighted as well as the results/conclusions. The abstract needs to make clear why this work was undertaken and why it is important (motivation). Why did you go to the trouble of implementing this in Sfepy?
• [ ] The introduction needs to make an extended argument about why NURBS/IGA is important as well as the relative advantages and drawbacks over FEM (and need to have this implemented in Sfepy). None of that comes across from the abstract or introduction. Also a few sentences on the history of NURBs/IGA as this isn't as familiar to most readers as the FEM.
• [ ] This may or may not be possible, but could there be an image at the start of section 2 that demonstrates the various differences between Bezier curves, B-splines and NURBS. Maybe a cartoon-type image with explanation about why we need each of these geometric representations. Something so that the general reader can get a feel for these concepts.
• [ ] It would help the general reader to have some explanation of the notation used in section 2. This could be part of an image describing Bezier curves, B-splines and NURBS.
• [ ] For the analysis in sections 5 and 6, it would be nice to have the git SHA for the version of the code that produced each of the figures along with the URL to the example file gists. Make it reproducible!
• [ ] The comparison between IGA and FEM doesn't include CPU usage, memory high water mark and method accuracy (against more refined FEM and also order of accuracy) for each technique. It would be really nice to see the numbers side by side in a table for each example and for varying mesh sizes. One of the benefits of implementing both techniques in Sfepy is that practitioners can readily compare both techniques with little effort.
• [ ] The symbols in the explanation text for $D_{ijkl}$ after the first two equations in section 5.2 seem to be messed up.
• [ ] Might it be better to reduce the number of figures in the paper. For example, all the images in figures 4 and 6 could be replaced with just a few, but with a far more detailed caption.

[rc]

Thank you for your review. You raise a number of good points, I will try to address them soon(ish).

[pdebuyl]

Dear @khinsen and @wd15 , thank you for your reviews.

Dear @rc , please propose modifications addressing the comments made in @wd15 's review.

Your updates should arrive for mid-october so that the reviewers can assess the update before the end of october.

[rc]

Sure, @pdebuyl I will update it in time. I will get to it after the end of September (two other deadlines by the end...). Cheers, r.

[pdebuyl]

Thanks for the notice :-)

[wd15]

On Fri, Sep 26, 2014 at 4:45 AM, Pierre de Buyl notifications@github.com wrote:

Dear @rc https://github.com/rc , please propose modifications addressing the comments made in @wd15 https://github.com/wd15 's review.

Just to point out, my comments are only suggestions and not mandatory changes.

Daniel Wheeler

[rc]

I have updated one example and the Dirichlet boundary conditions description, as SfePy 2014.3 supports the conditions given by a function. I will mention the version in the text.

Now I am going to address (at least some of) the Daniel's points.

[rc]

I updated the text hoping it addresses some of @wd15 comments.

I would like to make the code that generated the results available as suggested - is there some "standardish" way of doing that (using gists)? Any pointer/advice will be appreciated.

The convergence analysis and comparison of IGA with FEM is IMHO out-of-scope of this paper, and moreover, it will take time and effort to make it properly. Surely, it has to be done - it will be in another, more focused, paper (maybe next year!).

[khinsen]

For your demonstration code, just create a repository on GitHub. It's better to keep them together in one place, rather than in unrelated gists.

[wd15]

On Thu, Oct 2, 2014 at 10:20 AM, Robert Cimrman notifications@github.com wrote:

I updated the text hoping it addresses some of @wd15 https://github.com/wd15 comments.

I would like to make the code that generated the results available as suggested - is there some "standardish" way of doing that (using gists)? Any pointer/advice will be appreciated.

Definitely a new repository. You'll need to include both the SHA from the example repository and the SHA from the Sfepy repository.

Off topic, but I've been using Sumatra recently to keep records of simulations. I find it very helpful.

The convergence analysis and comparison of IGA with FEM is IMHO out-of-scope of this paper, and moreover, it will take time and effort to make it properly. Surely, it has to be done - it will be in another, more focused, paper (maybe next year!).

Understood. Make clear that the paper is about the implementation and not the comparison, but will make the comparison easy in the future.

Daniel Wheeler

[rc]

I have just created [1] with the examples. I will update the text accordingly.

[1] https://github.com/rc/euroscipy2014-sfepy-iga-examples

[rc]

I have moved the repository with examples to https://github.com/sfepy/euroscipy2014-iga-examples and updated the text.

[pdebuyl]

Hi @rc it would be nice, as the arxiv publishing process does not allow to include source code, to link to your code via figshare or zenodo (for instance) so that this resource remains available with more longevity than in a github repo.

@khinsen and @wd15 if I understood well your position is both a final accept?

Pierre

[rc]

Hi @pdebuyl

why do you think that those services offer more longevity than github? (just curious :-), I am not against the idea). I can put an archive of the referenced version of the examples to sfepy.org - if that disappears, the examples would be useless anyway. What do you think?

[khinsen]

@pdebuyl Yes, I think this is fine for a final version.

As for the longevity issue, nobody knows which service will be around for how long, but DOI-issuing services like figshare and Zenodo have a higher moral obligation to maintain their data, because that is part of the DOI culture. Moreover, those services store a specific version rather than an ever evolving repository, which I think is a better match with the idea of archiving.

[pdebuyl]

The comment of @khinsen reflects my main motivation. There is no guarantee in either case but the a priori is in favor of archival service (that github is not) with respect to development hosting.

[rc]

I see, thanks for explaining. I have no experience with either of those - which one would you recommend?

[khinsen]

@rc Throw a coin... both are fine, and in fact very similar. Zenodo accepts bigger files (2 GB, compared to figshare's 150 MB), but that doesn't matter for our purposes.

[pdebuyl]

As far as these proceedings go, there is no preference :-)

[wd15]

On Tue, Oct 14, 2014 at 4:43 AM, Pierre de Buyl notifications@github.com wrote:

@khinsen https://github.com/khinsen and @wd15 https://github.com/wd15 if I understood well your position is both a final accept?

Yes.

Daniel Wheeler

[rc]

Thanks, @khinsen and @wd15, for your kind and constructive reviews. I hope that the article is more accessible now.

I have just added a zenodo doi link to archived examples.

[pdebuyl]

Thanks to all, this version will be included after routine checks (layout, etc).

Pierre

[pdebuyl]

Hi Robert,

The filesize of your paper is rather large, at 6.1 MB for the compiled pdf. Is it possible rework the figures for a lighter result? I say this mostly for the convenience of readers as large file are sometimes annoying (data transfer and printing issues).

[rc]

Hi Pierre, my pdf has only 4.5 MB (similar to the last year), but I will check whether the size could be reduced without excessive loss of quality.

[rc]

The above gives me a 3MB pdf, is it ok?

[pdebuyl]

It's allright, thanks for the quick update!

P