Suppose 'the worst' happens. Stack Exchange folds (commerical entities have been known to), and at the same time the moderators lose all interest and/or decide the despise mathematics. Where would this leave us?

People are citing MathOverflow in many published papers, so it seems we have some obligation to attempt to ensure that the content remains accessible forever.

Public dumps exist (although they are up to 3 months), and it's likely that someone would have retained a copy, even if the site they are currently served from became inaccessible. From this dump it would be relatively easy to create a read-only website containing the contents of MO, and if one could obtain ownership of the domain name (which is owned by the MathOverflow corporation, not StackExchange) it would be relatively easy to have all the links from the literature resolve to the correct (but read-only) places.

Is there something else we can do to ensure the long-term archiving of MathOverflow?

Are there libraries that would be interested? The Internet archive? The arXiv itself?

It would be relatively easy to produce a giant PDF snapshot of MO (e.g. I know how to script creating PDFs from webpages, including rendered MathJax). The question becomes whether anyone really reliable would want to take them.

  • $\begingroup$ Copies could be sent periodically to math departments (there is usually at least a library liaison). This could probably be done automatically, and universities that express interest could be added to the list. $\endgroup$ Commented Sep 12, 2013 at 1:25
  • 4
    $\begingroup$ I don't think the arXiv would make sense. MO is way outside of the arXiv's mission, I don't think a periodically updated 100+ megabyte document would be a good fit, and there are serious format issues. (A PDF snapshot couldn't hurt as supplementary information, but it shouldn't be the primary format for backups.) On the other hand, the Internet Archive and academic libraries would both be great solutions. $\endgroup$
    – Henry Cohn
    Commented Sep 12, 2013 at 2:12
  • 10
    $\begingroup$ What I'd recommend is getting in touch with the Internet Archive, and also looking into a few universities with repositories based on DSpace or the like (e.g., dspace.mit.edu). These repositories are often focused on work with some connection to the university, but they might make an exception for something of broad interest like MO. The home universities of moderators or MO board members might have a particularly compelling reason to view archiving MO as part of their mission, but my guess is that it wouldn't be hard to get universities to do this. $\endgroup$
    – Henry Cohn
    Commented Sep 12, 2013 at 2:27
  • 5
    $\begingroup$ Might the AMS take an interest? Even if they don't want to be directly involved in archiving, perhaps they have people with the right kind of expertise to advise. $\endgroup$ Commented Sep 12, 2013 at 2:52
  • $\begingroup$ Does anyone know who the relevant people at the AMS would be? DSpace seems fantastic, and explicitly set up for creating new "collections". $\endgroup$ Commented Sep 12, 2013 at 3:27
  • 2
    $\begingroup$ Librarians are such nice people. Mine says we should talk. $\endgroup$ Commented Sep 12, 2013 at 3:59
  • $\begingroup$ Related question: meta.mathoverflow.net/questions/129 Public dumps for mathoverflow? $\endgroup$ Commented Sep 12, 2013 at 9:04
  • 3
    $\begingroup$ Why PDF? It's so bloated compared to markdown (the format that MO posts are currently in) and I don't see any additional convenience. $\endgroup$ Commented Sep 13, 2013 at 10:52
  • 21
    $\begingroup$ Can't we trust the NSA to keep archives? $\endgroup$ Commented Sep 13, 2013 at 20:26
  • $\begingroup$ Yes, I'm convinced that PDF isn't a great idea. It's perhaps more convenient right now in some ways (some people are going to be more comfortable reading or printing PDFs which already have sensible margins, for example) but should be at most secondary to plain text, which is more likely to be useful in 10 years. $\endgroup$ Commented Sep 15, 2013 at 0:50
  • $\begingroup$ @ScottMorrison: Sensible margins can be specified using CSS media types: w3schools.com/css/css_mediatypes.asp $\endgroup$ Commented Sep 15, 2013 at 18:26
  • $\begingroup$ @GeorgesElencwajg I am not sure if your comment about the NSA was meant as a serious suggestion, but assuming it was, the answer is no. Without going into too much detail, the short version is that this sort of archival project is highly vulnerable to budget cuts. $\endgroup$ Commented Jul 31, 2023 at 11:59

5 Answers 5


Contrary to some other answers, I think the MathOverflow archive should be as textually literal as possible, though there is no need to preserve fancy formatting, logos or any bells and whistles that are meant to make MathOverflow more pleasant to use rather than to read. I would be very happy with something very plain (such as what you see below the separation line when looking at the revision history of a post).

I don't think there is any need to convert to PDF or any other format. The reasons for this are multiple. Markdown (the format MO posts are currently composed and stored in) is very compact and simple. It's perfectly readable even without markdown processing, so little is lost even if markdown goes out of fashion. Storing the math as LaTeX code is also generally better than compiling it and preserving the output. Alternately, LaTeX code could be converted to something which is standardized (such as MathML, for example) but LaTeX is pretty much a standard at this point. The only thing that might require maintenance is upgrading the character encoding from UTF to whatever comes next but that won't be a problem at all if and when it happens.

I think the archive should include comments since many do contain much substance and some posts don't make sense without them. It's true that a lot of comments are just chatter but I don't see any good reason not to preserve those too and I can't think of a good way to sift through comments. Besides, knowing that their comments will be preserved forever might make users think a lot more before posting something inappropriate.

One contentious issue is how to handle deleted material. Archives typically work only one way: you can put stuff in but you can never remove anything. I can think of two ways of handling this: (a) make deleted material unavailable but still preserved, (b) clearly mark deleted material as redacted. Maybe there is another way? (Note that the current site uses (a) for <10k users and (b) for 10k+ users.)

  • 2
    $\begingroup$ I agree that for long-term storage, it is best to use a format (e.g., ASCII plain text) that requires minimal transformation to become human readable. $\endgroup$
    – S. Carnahan Mod
    Commented Sep 13, 2013 at 21:25
  • 6
    $\begingroup$ I'd be inclined to preserve deleted material in the archives (marked as deleted). There may be occasional cases where it would be useful, such as references in the literature to material that was later deleted by whoever posted it. Deleted material is already available to nearly a hundred high-reputation users, who are allowed to distribute it under the terms of the CC license. In principle putting it in the archives would make it a little more accessible, but frankly if someone cares enough to dig it up from the archives, I don't think there's any harm in letting them see deleted content. $\endgroup$
    – Henry Cohn
    Commented Sep 14, 2013 at 0:09
  • 1
    $\begingroup$ I think the design of the archive is probably important when considering deleted material. Option (b) is fine if the deleted content is cannot be discovered with too much ease. However, if each question comes along with all its answers (similarly to the actual site) then the redacted answers are perhaps too easy to find. $\endgroup$ Commented Sep 14, 2013 at 0:32
  • 3
    $\begingroup$ @FrançoisG.Dorais, thanks for this very helpful answer. Perhaps at some point I'll see if I can generate a "plain text" version of MathOverflow from the dumps, that is simultaneously human readable with no software, but also clearly computer digestible, in case anyone had a future need to do that. Presumably I'll just generate one text file per question. $\endgroup$ Commented Sep 14, 2013 at 7:45

The Keepers Registry currently has 8 participating archiving agencies:

A quick look at their descriptions suggests that some may be interested in archiving MathOverflow.

The Center for Research Libraries appears to be the main international entity for assessing and certifying the trustworthiness and reliability of major archiving agencies. Some reports can be found here.

I looked in more detail at Portico since it has a very nice description of its services. I think they would classify MathOverflow as an e-journal. Their aim is to provide digital preservation, not a backup system nor a redundant access point. So this service would be suitable for a separate read-only "e-journal" archive of MathOverflow content, not for a site mirror (even if read-only) nor for permanent storage of MathOverflow data dumps. Their publication agreement can be seen here and the terms look very reasonable to me. Their pricing model is also very reasonable: so long as MathOverflow does not make \$250k in revenues, our annual contribution would be \$250.

It appears that HathiTrust and e-Depot currently only accept "monolithic" documents such as books and cannot accept periodicals or similar objects. Archaeology Data Service only accepts documents relating to archaeology and history.

The Global LOCKSS Network has an interesting framework.

This section describes how a LOCKSS Box works. Specifically, a LOCKSS Box performs five main functions:

  • It ingests content from target websites using a web crawler similar to those used by search engines.

  • It preserves content by continually comparing the content it has collected with the same content collected by other LOCKSS Boxes, and repairing any differences.

  • It delivers authoritative content to readers by acting as a web proxy, cache or via Metadata resolvers when the publisher’s website is not available.

  • It provides management through a web interface that allows librarians to select new content for preservation, monitor the content being preserved and control access to the preserved content.

  • It dynamically migrates content to new formats as needed for display.

The system seems rather flexible but I couldn't determine what content formats they could support in their archival units. To use that option, we would need to maintain our own archive (distributing content in a format that LOCKSS can support) and the participating libraries would each have a permanent and authoritative cache of our archive in their LOCKSS Box.

The CLOCKSS Archive has a basic mechanism similar to the Global LOCKSS Network but the goal is not to provide continued access to the resource like some of the above. Instead, CLOCKSS is designed to publicly release an archive only after a trigger event. The Title No Longer Offered event corresponds to the type of failure that Scott mentioned in the question.

I haven't yet found the relevant information for the British Library and the National Science Library of the Chinese Academy of Sciences.

From what I can tell so far, Portico, the Global LOCKSS Network and the CLOCKSS Archive all offer potentially useful services for the purpose of archiving MathOverflow. The Global LOCKSS Network appears to be the most flexible but it does not offer as much security as Portico and CLOCKSS. CLOCKSS offers fewer services than Portico, but combining CLOCKSS with the Global LOCKSS Network is a very competitive alternative to Portico.


Preamble. When thinking about situations like this, it's tempting to fall back on traditional ideas. I realized earlier today that everything suggested so far would also work if MathOverflow was a giant stack of loose leaf paper instead. Though this is a characteristic of excellent ideas that are both simple and versatile, it does suggest that we might not be considering the full extent of our options. I want to encourage users to think outside the box while considering this issue. The suggestion below is a small step in this direction and I my hope is that our collective creativity will lead to a solution which is as innovative and practical as MathOverflow itself.

The idea that I will explain is motivated by several issues. One issue is the problem of deleted content that I raised in another answer. Similar issues are brought up by edits, revisions, later answers, etc. Another is that, although I expressed doubt about this, the fact that some users would prefer that only some of MathOverflow's content should be preserved suggests that this is a desirable feature. Finally, I realized during in my research on archiving agencies that they could find the total volume and regular flow of MathOverflow problematic.

This suggests that MathOverflow content could be archived only on demand. That is, there would be a MathOverflow Archive which stores a static copy of a question or answer in its current state when, and only when, a user requests it. Thus, if you want to cite something on MathOverflow, you make such a request for the content you want to cite, that content gets archived and your citation will always point to a static copy of that content.

This basically solves the deletion and editing issue since I doubt anyone would make such a request for deleted content and static copies remain as they are even if the original gets edited or modified in other ways. This also solves the selectivity problem since (presumably) only good content would be requested in this way. Finally, this also solves the volume problem since much less content would be archived by request.

This kind of archiving mechanism might also serve as support for other possibly desirable features such as an option to have the content peer reviewed (preferably in an open way using the Selected Papers Network, for example) and, if the content is approved, it would get stored in a designated "peer-reviewed" part of the MathOverflow Archive (along with the referee report). Such enhancements would definitely give more credibility to MathOverflow as a micropublishing platform.

Scott pointed out that a defect of this approach is that few would bother making such requests. A potential solution is to have posts automatically archived as soon as someone presses the cite button to get a bibtex/amsref entry and have the urls there point to the proper place on archive.mathoverflow.net instead of the original.

  • 2
    $\begingroup$ I'd suspect that this simply wouldn't be used. Until it's in widespread use, it won't occur to anyone to use it. $\endgroup$ Commented Sep 16, 2013 at 0:33
  • $\begingroup$ Your last paragraph is very interesting, however! $\endgroup$ Commented Sep 16, 2013 at 0:33
  • 1
    $\begingroup$ @Scott: Yeah, it's not a complete idea but I think it's a good lead. As I tried to communicate in the preamble, I'm hoping that creative minds will cooperate toward something much better. The last paragraph does reflect some of my long term thoughts for MO, independently of the ideas that led to it in this post. $\endgroup$ Commented Sep 16, 2013 at 0:47
  • $\begingroup$ @Scott: Also, regarding your first comment (and some unrelated discussions we had recently), wouldn't it be awesome if we had a decent place to publicize this kind of features? $\endgroup$ Commented Sep 16, 2013 at 0:57
  • $\begingroup$ Perhaps we should ask to turn on mathoverflow.blogoverflow.com $\endgroup$ Commented Sep 16, 2013 at 1:40
  • $\begingroup$ @Scott: I'm iffy about having that also being hosted by SE. Also, we need people to take care of it (whatever it ends up being). Unless one of us suddenly volunteers, it would have to be someone new. $\endgroup$ Commented Sep 16, 2013 at 3:41
  • 1
    $\begingroup$ I think the discussion has reached the point where it is clear that something should be preserved, and that it is unclear what the something is. Perhaps it should be subsumed as part of a larger project (e.g. mathoverflow.net/questions/122125/math-annotate-platform ), where it will become clear what is to be preserved. For example, I always hoped MathOverflow would morph from a series of reference requests to a collection of well written pointers to the literature, especially to free collections. Gerhard "Sometimes Nowhere Near The Box" Paseman, 2013.09.19 $\endgroup$ Commented Sep 19, 2013 at 19:26
  • 2
    $\begingroup$ In particular, if MathOverflow influenced the redesign of the storage, production, and accessibility of mathematical knowledge, and the new object became so useful that it left MathOverflow in the dust, I would attend MathOverflow's wake and sing its praises, and then go use the new tool. Gerhard "And Maybe Attend Anniversary Parties" Paseman, 2013.09.19 $\endgroup$ Commented Sep 19, 2013 at 19:29

One interest of long term archiving is that MathOverflow could then be issued an ISSN. One could make the case that MathOverflow is already eligible for an ISSN, but a long-term archive would make an even stronger case. Here is the relevant list of eligibility criteria from the ISSN Manual.

0.3.2 Ongoing integrating resources

Ongoing integrating resources are resources that are updated over time and with no predetermined conclusion, for which the updates are integrated into the resources and do not remain discrete.

ISSN are assigned to ongoing integrating resources which fulfill all the inclusion and exclusion criteria listed hereafter. National Centres can decide to exclude ephemeral ongoing integrating resources or ongoing integrating resources of purely local interest from systematic ISSN assignment (see section 0.7)

Meeting only one of the criteria is not sufficient for ISSN assignment. These criteria apply to all the categories of ongoing integrating resources, whether print or electronic: databases, websites, wikis, print loose-leaf services, etc. Although blogs are considered to be serials, these same criteria should be applied to blogs.

Note: If a part of an ongoing integrating resource (part of website, for instance) is eligible for ISSN assignment, e.g., a newsletter that is part of the website is assigned an ISSN, this does not mean necessarily that the whole website is eligible for ISSN assignment. Inclusion criteria

  1. There is editorial content (i.e., the resource mostly consists of written, textual content, and there is evidence of editorial or journalistic treatment);

  2. There is identified editorial responsibility (i.e., a statement indicating the name of the publisher / producer, and at least the country of publication). Generally, editorial responsibility will consist of more than one individual;

  3. There is a consistent title (i.e., a title which remains consistent when the resource is updated) and the title is prominently visible on the resource;

  4. There is a valid URL (i.e., a URL leading to the actual resource);

  5. The resource has subject-related content or has identified subject-related audience. Exclusion criteria

  1. Personal resources (personal web sites and web pages, online diaries);

  2. Resources focusing on a company, a product, an institution or organization (advertising and promotional web sites, commercial web sites, product information web sites, company and institutional web sites, web sites of organizations);

  3. Web sites consisting only of links;

  4. Ephemeral resources (i.e., resources known to be ephemeral).

  • $\begingroup$ Would having an ISSN have beneficial consequences? They're particularly useful for serials with titles that are vague, ambiguous, or difficult to search for. However, I doubt it would ever be easier to look up MO via ISSN than via the name "mathoverflow". (On the other hand, maybe there would be other benefits. Do librarians take serials with ISSNs more seriously?) $\endgroup$
    – Henry Cohn
    Commented Sep 13, 2013 at 13:04
  • 2
    $\begingroup$ @Henry: My understanding (which is limited) is that librarians do classify serials with ISSNs differently. Though, I don't think it matters that much for practical purposes. My personal view is that an ISSN would "certify" that MathOverflow is not ephemeral and has lasting value. $\endgroup$ Commented Sep 13, 2013 at 13:10
  • $\begingroup$ Also, an ISSN seems to be a requirement for inclusion in some listings, such as the Keepers Registry from my other answer. $\endgroup$ Commented Sep 13, 2013 at 13:37
  • $\begingroup$ @HenryCohn this might not (but perhaps eventually it might) be relevant for MO, but I do know that some 'activity reports' for researchers explicitly distinguish also between non-refereed contributions according to whether the thing where one contributed has an ISSN (or ISBN) or does not have. $\endgroup$
    – user9072
    Commented Sep 13, 2013 at 14:29
  • 7
    $\begingroup$ @quid, in that case, I'm looking forward to adding some 900 new "publications" to my CV! And doesn't the voting mechanism count as referring? ;-) $\endgroup$ Commented Sep 13, 2013 at 16:03
  • 2
    $\begingroup$ @Joel: This may not be as far fetched as you think if you believe those who say that academic micropublishing is inevitable. $\endgroup$ Commented Sep 13, 2013 at 19:12
  • 3
    $\begingroup$ @JoelDavidHamkins You are looking forward to this? Adding 900 items to a CV seems like an awful lot of busy-work to me! Good I am pseudonymous, also my 200 would be too much work. ;-) More seriosuly and generally, what I meant to convey is that also in some scientific evaluations having an ISSN can play a role (in general). What I could actually imagine is that overtime it becomes more common to see MO contrib as part of 'usual' (optional) scholarly activity. Say, like writing referee reports or perhaps closer reviewing for MathSciNet, which gets mentioned in a geenral/bulk form in some CVs. $\endgroup$
    – user9072
    Commented Sep 13, 2013 at 20:30

The decision to archive must be made with care. In addition to the technical issues of preserving rendering of typeset math, static and dynamic images, and various links, there are the social issues of how to keep and arrange the comments if any, and how to include chat room transcripts and even whether to post annotate to explain cultural and historical references. ("What's +1 Daddy? Did mathematicians use only one finger instead of five?")

I think a partial solution may be appropriate, which is to have a static rendering of some parts of the content, with references to the dynamic database. That way a curated and edited archivable version can be created and preserve something until all the other bugs are ironed out of making a full archive. The curated versioncan also be made useful independent of a full archive with some care.

  • 4
    $\begingroup$ I think you're letting perfect be the enemy of good. If someone want to do some serious "history" on MathOverflow, that's great, but no one here has the interest or ability to do it. I'm really only concerned with ensuring that a link in a paper to a MathOverflow question can be resolved (possibly 'by hand') to some version of the content, even assuming non-cooperation (or non-existence) of SE. It seems that archive.mathoverflow.net/question/1234, hosted by a university library, containing the 'last known' version of the page, rendered as PDF, would suffice. $\endgroup$ Commented Sep 12, 2013 at 23:46
  • 4
    $\begingroup$ Why should such sociological commentary be part of the archive? There are already academics who study MO (in the sociological sense) and they publish their work in scholarly journals and books that are preserved on their own. There is no need to integrate that into MO. $\endgroup$ Commented Sep 13, 2013 at 11:12
  • 1
    $\begingroup$ MathOverflow is not a database. Nor is it a collection of PDFs. We could argue what MathOverflow is (or, to stick to this post, what an archive of it should be), but it might be better just to make a thing that serves many of the purposes of an archive instead. Scott Morrison's suggestion of a collection of PDFs seems good, but makes the assumption that PDF will be an available format long term. I do not make that assumption in the post above; I argue that more than one thing be created until we know what would make a really good archive, letting good be a guide towards perfect. $\endgroup$ Commented Sep 13, 2013 at 15:02

You must log in to answer this question.

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