I was reading the post on the self-evaluation of the site and just wondered if it would be a good idea for the popularity of this site to build "big standard" community questions?

For example:

  • "Which linear solver for my problem?"
  • "Which method should i use to approximate my PDE/ODE?"

These questions should be normally closed (too broad) but that's the kind of answers that many people would need.


I think both points of view expressed here are correct:

  1. It is impossible to give general recommendations for which method to use; rather, the choice of method depends on the characteristics of the problem(s) to be solved.
  2. It would be very nice to have a set of basic suggestions of the form "if your problem has properties x and y, consider the class of methods z".

The question in my mind is whether the suggestions referenced in #2 should all appear in answers to a single huge question, or whether each question should be about a specific class of problems so that specific classes of methods can be recommended.

One alternative would be to have a broad question (or just a wiki page?) with pointers to many specific questions.

  • 2
    Another alternative would be to have a "Frequently Asked Questions" question on Meta, with answers containing links to relevant already existing questions on the main site. (For example, I remember there being a question on how to choose a linear solver already.) An added advantage is that if a similar new question is being closed as too broad, we can link to the corresponding answer with a comment such as "Look at these previous questions, and if the answers fail to address a specific point you're interested in, edit your question accordingly." – Christian Clason Oct 22 '14 at 7:40

I respectfully disagree. The words "best" and "should" are too ambiguous. The manner in which a person defines them within the context of their question is the only way to give a meaningful, cannonical answer. It's impossible to say what is best even for a single toy problem like the laplace equation. The quality expectations of the asker matter in terms of how to respond appropriately.

  • I agree with Paul: these questions are too broad to give useful answers, and attempting to give answers to these questions in a general way is likely to result in giving someone bad advice. – Geoff Oxberry Oct 20 '14 at 17:31
  • Ok, maybe my question was not clear enough (or maybe I'm just wrong, that's possible, too). I did not mean that we can summarize the "optimal" resolution of linear system (let's take that as example) easily, but if I encounter a new system that I need to solve, I will probably look at 3-4 properties (sparsity, dimension, symmetry,...) and decide for a first solver to try (that I could find in the community question as a decision tree). If that does not work, I will go for more custom solutions, and maybe ask for it on scicomp. – Dr_Sam Oct 21 '14 at 5:23

Okay, here's an idea that I like:

Establish a tag called "choosing-a-method" (or "method-selection") or something similar. Then, if I'm wondering how to choose a solver for $Ax=b$, I just search for questions tagged with BOTH "choosing-a-method" and "linear-solver".

This solution is lightweight and easy to implement. The drawbacks are that users might not know about it (so we should put it in the FAQ) and that we need to retag lots of questions to make it work.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .