Talk:Second-order cone programming

Expert needed? Too technical?

There is an expert tag on this page but I cannot see where it is really needed. I'd like to get some feedback before deleting the tag. Zfeinst (talk) 20:14, 15 October 2011 (UTC)

It says "too technical" and I didn't understand the notation so I've tried to wikilink some. - Rod57 (talk) 09:55, 21 January 2018 (UTC)

One decade on, does this tag still serve any useful purpose? This is a a topic that is mainly going to be read by reasonably technical people and should be evaluated mainly in terms of it's usefulness to them, so I suggest removing it. --Servalo (talk) 09:17, 29 November 2021 (UTC)

Second order?

Can we explain why this is second-order - rather than first or third order ? - Rod57 (talk) 09:55, 21 January 2018 (UTC)

Can someone add a picture?

I don't want to read all these boring maths. Please add a picture that explains the problem and solution. 128.196.56.50 (talk) 20:33, 28 January 2014 (UTC)

The Solvers section should not be deleted

I don't think the deletion of the "Solvers (...)" section is justified. See, for instance, that linear programming and quadratic programming both have big solvers tables. What do you think MrOllie? BernardoSulzbach (talk) 15:32, 23 August 2019 (UTC)

I went on and reverted the deletion after 48 hours without response. BernardoSulzbach (talk) 17:54, 25 August 2019 (UTC)

Add MATLAB's coneprog (and ECOS?)

The solvers section is missing MATLAB's coneprog function which has been available since version 2020b. I won't add it myself since I work at MathWorks but I think that this will improve the coverage of this section. I also note that there are only commercial solvers in this section. If you google 'ECOS: An SOCP Solver for Embedded Systems' you'll find an open source one that you may wish to add for balance. — Preceding unsigned comment added by WalkingRandomly2021 (talk • contribs) 2021-07-14T17:23:48 (UTC)

@WalkingRandomly2021: I support this. Please help expedite this by providing good references (just the URL is OK) for both. Please don't provide URLs for something you wrote yourself.

ECOS was removed in 912453495 by MrOllie. @MrOllie: do you think ECOS should not be listed for some reason or just not have an external link? BernardoSulzbach (talk) 18:13, 14 July 2021 (UTC)

I think that only stuff with established notability (usually a preexisting Wikipedia article) should appear on lists like this. - MrOllie (talk) 18:37, 14 July 2021 (UTC)

@MrOllie: Since ECOS is the standard solver on CVXPY for SOCP problems, shouldn't it be added by this reason? In my opinion, CVXPY has notability in academy. Though it do not have a preexisting Wikipedia article, its associated paper has been cited by more than 900 papers and its GitHub repository has more than 3k stars. Saung Tadashi (talk) 19:13, 14 July 2021 (UTC)

Notability on Wikipedia means that independent people have written in-depth about it. Citations are often trivial mentions, and social media metrics like github stars are even less than that. Even so, if CVXPY is notable that doesn't mean that every algorithm or bit of technology they used is also notable. Of course, the elephant in the room is that per WP:NOT we probably shouldn't have this list in the article at all, any more than we should have lists of suppliers on any other software related article. - MrOllie (talk) 19:31, 14 July 2021 (UTC)

@MrOllie: Got it, thanks for the explanation. Saung Tadashi (talk) 19:48, 14 July 2021 (UTC)

@MrOllie: you make sober points, thank you. I guess you're referring specifically to WP:NOTDIRECTORY. I'd say that this shouldn't be addressed on a per-article basis: as you may know, linear programming has a much bigger instance of this same table. I think it'd only be reasonable to remove all of these tables, or none of them. Lastly, I agree with the (doesn't have an article here yet $\rightarrow$ doesn't get mentioned in a table) rule and, therefore, I no longer support including ECOS in the table. I still stand for MATLAB, as it got to be mentioned in linear programming for LINPROG. BernardoSulzbach (talk) 20:29, 14 July 2021 (UTC)

I think Bernardo raised an important point regarding the consistency between articles. So I have one more question in this context: if CVXPY get sufficient notability (according to MrOllie's rule, which I also find reasonable), then CVXPY (rather than ECOS) could be added on the list, right? I am asking this because AMPL (which is a modelling language like CVXPY, not a solver) is also on the list. Saung Tadashi (talk) 20:41, 14 July 2021 (UTC)

@WalkingRandomly2021: I also support this edit. By the way, kudos for your blog - I am a longtime reader :) Saung Tadashi (talk) 19:13, 14 July 2021 (UTC)

I agree that if we're going to have the list (I accept that consensus is not with me on dropping it entirely), MATLAB should have a place on it. If ECOS (or CVXPY) gets an article and/or 2-3 independent reliable sources, that should be on the list as well. - MrOllie (talk) 21:27, 14 July 2021 (UTC)

Thank you for considering this and for a very illuminating discussion. I've made the edit. Thanks for your kind words Saung, blogging is fun..I wish I had more time for it. WalkingRandomly2021 (talk) 10:01, 15 July 2021 (UTC)

Why was a convenient formula removed?

Hi! I'm new to Wikipedia editing/commenting, but I thought I'd make an account to post here.

On March 5, 2021, an edit titled "correct errors" removed an extremely useful formula. In particular, there used to be a formula for converting $x^{T}A^{T}Ax+b^{T}x+c\leq 0$ into an SOCP constraint, but it was replaced by a different one for $x^{T}Ax+b^{T}x+c\leq 0$ with $A\succ 0$.

The formula in the newer version of the article requires $A$ to be invertible (notice the $A^{-1/2}$ term), whereas the older version worked even when $A$ isn't invertible. I actually was using this article as a reference specifically for the old version of the formula, which comes up a lot in my research.

Was the old formula incorrect? At least in the applications I've worked on, the old formula seemed to work properly. And it's much more useful in practice (no matrix square root, and often we know the factorization of the quadratic term a priori).

If the old formula was correct, maybe we can include both the old version and the new version in the article? Currently I have to dig through the Wikipedia edit history whenever I'm converting convex quadratic programs to SOCPs :-)

Thanks! - Justin Solomon Justinmsolomon (talk) 09:38, 1 September 2023 (UTC)

(Note the new formula is less general than the old one since it requires invertibility, but maybe I'm missing a boundary case where it's untrue! Thanks.) Justinmsolomon (talk) 09:39, 1 September 2023 (UTC)