# Is it ok to ask about the correctness of preprints on crank-friendly topics?

See this question in which someone asked for an update on the status of an arxiv preprint on graph isomorphism. It was downvoted, with a comment "Are this kind of discussions on-topic here? I think there are too many claims of P = NP, P ≠ NP, etc. in ArXiv and elsewhere.", but my own feeling is that it's useful to have a place to publicly discuss whether it's worthwhile to pay any serious attention to certain controversial preprints.

I guess my question for here is, should this exchange be that place?

Update (suresh): Please review and edit the proposed FAQ entry below, so we can add it to the FAQ and resolve this question.

• FWIW, I added the comment but didn't downvote. Sep 7 '10 at 22:32
• Nov 16 '11 at 23:34

### A proposed policy for questions on preprints.

(in the form of an FAQ.)

There are good ways to ask about new seemingly exciting results in TCS, and there are bad ways. Below, we give a few examples of each:

• BAD: Have you read this new paper on the arXiv on (seminal problem in complexity)?

• Answer: Yes, we have. Please do not post this question on this site.

• BAD: Is the new claimed proof of (seminal problem in complexity) correct?

• Answer: If it's at all credible, be assured that people are reading it and trying to determine that. In fact, why not read it yourself ! But don't post this question on this site.

• GOOD: In this new preprint on the arXiv that claims to solve (seminal problem in TCS), the author refers to quantum range searching on squashed hypercubes. What is that?

• GOOD: This new paper claiming a big result is beyond my ken to read. Before I invest the time to learn all this stuff and try to read it, I am curious: have there been any technical rebuttals or discussions of it?

• note: Such cases are acceptable only if they are really reference-requests (see Josh's answer). This item cannot be used if the real motivation of the OP is not reading the paper. Using this item to circumvent the general policy will not be permitted.
Also this item doesn't apply if it is unlikely that there are such references, e.g.

1. the time is too short for existence of such references,
2. reasonable time has passed but the author's claims is not considered credible, e.g. the author is not considered an expert in the area and the author's claims has not been taken seriously by experts in the area.

Note that if a claim about progress on a famous long standing open problem (like "P vs. NP") is credible at all, there will be discussions about it online in a short time after its announcement (e.g. on theory blogs like Computational Complexity, Gödel's Lost Letter and P=NP, etc.) and if there is not then it is a sign that the claim is not taken seriously by experts.

• BAD: This new unpublished papers on arXiv proposes a new algorithm and claims it can solve problem X way faster than the best known algorithms. I don't understand the paper and don't want to spend time reading it, can someone verify that the author's claim is correct?

• Answer: If it's at all credible, be assured that people are reading it and trying to determine that. If you can't check the correctness of the algorithm yourself, you should wait until it is peer-reviewed and published. The peer-review process does not take a very long time, if the paper hasn't been published after several years it is a strong sign that the paper has failed to pass the peer-review process and the author's claim is at least doubtful. Discussing the general correctness of papers which have not passed the peer-review process are off-topic for this site.

• Good: What is the fastest known algorithm for problem X?

In summary, focus your question on specific technical aspects of the work that you'd like clarification on. You can also try to rephrase your question in a way that avoids mentioning a specific unpublished work but will ask essentially what you want to know.

Remember: if your question is closed as off-topic, you can still edit your question to address the issues stated for closing it, in which case it can be reopened.

• "why not read it yourself": reading the whole paper might not be practical. In fact, one may be interested in the credibility of the new results, but not in the technical details. Moreover, there might be people here who have read the paper before, and can comment on (or even judge) the credibility. Sep 8 '10 at 17:45
• but that's the kind of discussion this policy is trying to avoid. It's not the purpose of a Q&A site to foster discussion on soft matters like "credibility". Better to focus on technical questions. Sep 8 '10 at 20:13

... it's useful to have a place to publicly discuss ...

AFAIK, SE is designed intentionally to discourage discussions, so although I completely agree that it is useful to have a place to publicly discuss these issues, IMO TCS SE is not an appropriate place for such discussions.

Also IMHO having this kind of open ended endless discussion will severely decrease the usefulness of the site. Tolerating this kind of questions can attract lots of cranks.

If a question is not specific enough to have a definite answer and may lead to open-ended endless back and forth arguments, then IMHO it is not appropriate for TCS SE.

• We may end up having one (or even more) post per $P$ vs $NP$ claim that is submitted to arXiv. Sep 8 '10 at 4:15
• "Correctness of preprints" is going to mean questions of the form: Does so-and-so's purported proof work? These do fit the Q&A format. Sep 8 '10 at 17:16
• but (as the recent case demonstrates) they don't have clear definite answers and will lead to back and forth discussions. Mar 29 '11 at 1:39

As this issue recently came up again in regards to this question on matrix multiplication, I thought I'd throw in my two cents.

It seems to me entirely reasonable and in line with all the discussion in this thread that the following type of question be allowed:

This new paper claiming a big result is beyond my ken to read. Before I invest the time to learn all this stuff and try to read it, I am curious: have there been any technical rebuttals or discussions of it?

Note that this avoids issues of credibility, author responses, etc. It is simply a reference request.

(This seems to me essentially what has happened with the matrix multiplication question, and seems like it would satisfy the OP for that question, even though his question wasn't originally phrased this way.)

• Nice. I added your mod to the FAQ entry being compiled below. Nov 29 '10 at 17:45

Here's the scenario that could cause a problem. Let's say that on the thread cited above, the author in question decides to post what they consider to be a rebuttal of the claim that the paper is not well specified. This would after all be their right, to "defend" their work. But then a discussion might break out, and this platform is not ideal for that.

Maybe at that point the thread could be closed, with an edit stating that the participants are discussing the matter in private (so as not to give the impression that the discussion has resolved itself) ?

• Perhaps we could try to have the following policy in order to "play safe" and to avoid using this site for something for which it's not suitable: it is ok to give pointers to discussion elsewhere, but let's avoid giving new information (new opinions) here? With this policy it'd also make sense to CW this kind of questions. Sep 7 '10 at 22:41
• We could also try to have something in our FAQ that covers the most common case. For example, the answer to "is the community aware of paper X in ArXiv" is virtually always "yes". Sep 7 '10 at 22:43
• If the preprint's proof has a mistake or hole or ambiguity in it, the proof doesn't work. If the author says it can be fixed, then they can go off and revise their paper. If the author denies that the hole is there, well ... "discussion" can be closed down. The nice thing about talking about proofs is that the criterion of acceptability is pretty firmly established. Sep 8 '10 at 17:24

I think the same thing happened here: Is the recent proof that P != NP correct?.

When a new result appears, many people try to email "giants" in the community, asking them what is their idea on the paper. Many of the researchers are tired of such situations.

I recently witnessed such event due to Deolalikar's paper (See Scott Aaronson's, Dick Lipton's, and others' blogs).

In a blog post, Aaronson considers this forum as "the best thing ever to happen to his inbox," and offers the motto "Exponentially better than emailing Scott Aaronson" for TCS Stack Exchange.

Among other things, I think TCS SE must become the "de facto" place for discussing new results on theoretical computer science. Conferences and journals have their own review process, but why should this prevent us from discussing them here?

I believe Suresh's point is right: We must not embrace authors here, and we should give them enough right to defend their work. Yet, is it right to do so by not discussing any new result here??

I agree that this is a good site to have such discussions.

But again, there might be the concern that the author would get embarrassed publicly and that the right course of action should be to contact the author directly. I personally don't subscribe to this viewpoint.

EDIT: I suppose one could make a distinction between reviewed publications and online preprints. In the first case, it's better to contact the author directly if mistakes are found because it's very likely that the paper is not totally bogus. In the second case, discussions on this site are fine.

1) If an author posts something to the web, whether CoRR or their own web site, s/he is making that paper available for scientific discussion. There's already an example in-scope question about "Does Theorem X follow from Lemma Y in Paper Z." Contacting authors before posting a question here might be courteous, but isn't an enforceable policy. We have no way of knowing if there was such contact, or what happened between the parties if contact occurred. Best to avoid the need for telepathy.

2) Beating the drum about my great new awesome result is already a no-no, according to the Area 51 FAQ about pushing one's own product. My theory of Tautological Postmodernism might prove that Goldbach's Conjecture implies that P=EXPTIME, but someone else, not me or a sock puppet, can raise it for discussion on this site, and I don't get to try to sell my product all day long if someone does post a question. Exactly what "all day long" means is ultimately a moderator decision.

• But the problem, as I said above, is this: suppose there's a reasonable paper on the arxiv, and someone posts an answer trashing it. One would want the author to be able to come and defend themself - it seems only fair. However, this situation cannot be distinguished from the reverse without discussion. Sep 7 '10 at 22:59
• Close it if there is back-and-forth. If the question demonstrates it is "subjective, argumentative or requires extensive discussion" then existing rules already indicate it should be closed. Sep 7 '10 at 23:06
• Stack Overflow allows promotion of one's own products, within limits: "Be careful, because the community frowns on overt self-promotion and tends to vote it down and flag it as spam. Post good, relevant answers, and if they happen to be about your product, so be it." (stackoverflow.com/faq ) Perhaps we can have a similar policy?
– Emil
Sep 8 '10 at 14:27