23
$\begingroup$

[EDITED to make the question more suitable for MO. See meta.mathoverflow.net for discussion about re-opening.]

According to Wikipedia, Shing-Tung Yau expressed some doubts about Perelman's proof of the Poincaré conjecture in his 2019 book The Shape of a Life. Yau wrote

"I am not certain that the proof is totally nailed down. … there are very few experts in the area of Ricci flow, and I have not yet met anyone who claims to have a complete understanding of the last, most difficult part of Perelman's proof … As far as I'm aware, no one has taken some of the techniques Perelman introduced toward the end of his paper and successfully used them to solve any other significant problem. This suggests to me that other mathematicians don't yet have full command of this work and its methodologies either."

To what extent are the doubts that Yau expressed well-founded?


Here is a fuller quotation of the relevant passage from Yau's book The Shape of a Life (pages 258–260).

One problem I'm not actively working on is the Poincaré conjecture, as I'm happy to put the controversy surrounding it behind me. But I can't keep my mind from turning to that problem, upon occasion, and I still have some lingering doubts that—if expressed out loud—are likely to get me in trouble. Although it may be heresy for me to say this, I am not certain that the proof is totally nailed down. I am convinced, as I've said many times before, that Perelman did brilliant work regarding the formation and structure of singularities in three-dimensional spaces—work that was indeed worthy of the Fields Medal he was awarded (but chose not to accept). Perelman built upon a foundation painstakingly laid down by Hamilton and carried us further along the path laid out by Poincaré than we've ever ventured before. About this I have no doubts, and for that, Perelman deserves tremendous credit. Yet, I still wonder how far his work involving Ricci flow "technology" has taken us. And I also can't keep from wondering whether another approach—making use of some of the minimal surface techniques I developed many years ago with Bill Meeks, Rich Schoen, and Leon Simon—might lend some clarity to the situation.

In 2003, Perelman told Dana Mackenzie, a reporter for Science magazine, that it would be "premature" to make a public announcement regarding a proof of the geometrization and Poincaré conjectures until other experts in the field weighed in on the matter. Confirmation of this proof resided largely with outside "experts," given that Perelman receded almost completely from the mathematics scene, which is a great loss to the field. The thing is, there are very few experts in the area of Ricci flow, and I have not yet met anyone who claims to have a complete understanding of the last, most difficult part of Perelman's proof.

In 2006 or thereabouts, a visiting mathematician who was knowledgeable about this area stopped by my Harvard office to reproach me for raising questions about Perelman's work. Yet he admitted, when I asked him, that he did not entirely grasp the latter part of Perelman's argument. That's no knock on him, as that admission puts him in a rather sizable group. In fact, I don't know whether anyone else, including Hamilton, has fully gotten it, and I'd put myself in that category as well. As far as I'm aware, no one has taken some of the techniques Perelman introduced toward the end of his paper and successfully used them to solve any other significant problem. This suggests to me that other mathematicians don't yet have full command of this work and its methodologies either.

Hamilton, who's now in his seventies, has told me that it is still his dream to prove the Poincaré conjecture. That does not mean that he thinks Perelman did anything wrong. Hamilton, a truly independent spirit, is not one to follow in someone else's footsteps, nor would he be inclined to "connect the dots" of another's argument. He just may want to do it his own way and complete his life's work of the past three and a half decades.

Nevertheless, that still leaves me with the sense that this situation is not unequivocally resolved, perhaps leaving theorems of incredibly broad sweep hanging in the balance. Expressing my doubts on this subject, I know from experience, is a politically fraught proposition. But for the sake of my own questions—and for mathematics as a whole—I'd still like to be more certain of where we stand. If that makes me a pariah, so be it. In the end I care more about mathematics—the path I chose to follow more than a half century ago—than I do about what others think of me.

$\endgroup$
34
  • 36
    $\begingroup$ One should make a distinction between Perelman's proof of the Poincare conjecture and his proof of the geometrisation conjecture. For the former there are shortcuts that allow one to avoid the most difficult components of his arguments, which is presumably what Yau is alluding to here . In any event the subject has progressed since 2019, see e.g., Bamler's survey at arxiv.org/abs/2102.12615 $\endgroup$
    – Terry Tao
    Nov 6, 2021 at 22:02
  • 22
    $\begingroup$ I think this is a legitimate question (not trolling). Not clear why people are voting to close it. $\endgroup$
    – GH from MO
    Nov 6, 2021 at 22:23
  • 18
    $\begingroup$ I voted to close as my impression is this is not a serious question. Cherry picking a comment from a Wikipedia page does not even reflect a balanced reading of that same Wikipedia page. This isn't a serious question, it's Reddit trolling. $\endgroup$ Nov 6, 2021 at 22:38
  • 22
    $\begingroup$ I find Yau's quote quite bizarre (he surely knows the authors of 3 detailed accounts of the full proof). To make things worse, the last part is seriously outdated, given papers by Kleiner and Lott on the orbifold geometrization and by Bamler and Kleiner on the generalized Smale conjecture $\endgroup$ Nov 7, 2021 at 4:06
  • 23
    $\begingroup$ @Libli: This quote speaks volumes about Yau and why he is not on speaking terms with people advancing the theory of RF for the last 15 years. It does not make the current question on-topic for MO. $\endgroup$ Nov 7, 2021 at 16:01

2 Answers 2

32
$\begingroup$

First, let me make some preliminary remarks. We sometimes like to think that "being proved" is a black-and-white property, but in fact there are shades of gray. At one extreme are things like the infinitude of the primes, whose proof every mathematician understands. But then there are results that are widely accepted but no proof has appeared. Vladimir Voevodsky has pointed out that "a technical argument by a trusted author, which is hard to check and looks similar to arguments known to be correct, is hardly ever checked in detail," and that this practice can lead to false statements being erroneously accepted as proved; recognizing this point led Voevodsky to spend much of the later part of his career on computer verification of formal proofs. Although the mathematical literature is generally very reliable, it is far from perfect; this point has been addressed in more detail in another MO question about the extent of wrong papers in research mathematics. So in vast majority of cases, "being proved" isn't about being 100% confident that there is no error; it's about whether the proof has been sufficiently scrutinized that the chances of a serious mistake are negligible.

Returning to Yau, if you look carefully at what he is saying, you will see that, technically, he does not say that he thinks Perelman's proof is wrong, or that it has a serious gap, or even that there are parts of the proof that nobody understands. He says only that he is not certain that the proof is totally nailed down, and that he has not met anyone who understands the most difficult part of the proof. He also points out that if a powerful new idea is properly digested by the mathematical community then it usually leads to the solution of new problems, and that if this has not happened with the most difficult part of Perelman's proof then it probably means that this part of the proof deserves more study.

In principle, calling for the mathematical community to devote more time to studying an important and difficult proof in order to "nail it down" and acquire a "complete understanding" and a "full command of this work" is unobjectionable. In the past, I have heard colleagues say that the original work of various Fields Medalists—Hironaka and Freedman come to mind—was very difficult to understand and that there was a need for the community to study and assimilate those groundbreaking ideas more thoroughly. In both the cases of Hironaka and Freedman, the community has indeed put in effort to study their work, and rich dividends have resulted, so this type of activity is definitely worth encouraging. Note that this doesn't mean that the original proofs were wrong or had serious gaps; it just means that the proofs moved closer to the infinitely-many-primes ideal of universal understanding, and the chance of an unnoticed significant gap or error was driven down even closer to zero.

Unfortunately, Yau chose to phrase his remarks in a "politically fraught" manner that he knew would "get him in trouble." He says things in a way that (probably intentionally) gives many readers the impression that he is casting doubt on the correctness and completeness of Perelman's proof (even though, as I said, technically he doesn't explicitly say that the proof is wrong or incomplete). The book appeared in 2019 but the most recent conversation he cites was from 2006. He makes no mention of recent research in the area which does in fact apply Perelman's ideas to solve new problems.

It should therefore not be surprising that the consensus of the mathematical community is that Yau's remarks do not pose any serious challenge to the conclusion that Perelman's proofs—especially of the Poincaré Conjecture, which involves fewer technicalities than the Geometrization Conjecture—are correct. There were at least three separate efforts which came to this conclusion. Kleiner and Lott's detailed notes say, regarding Perelman's original papers [51] and [52]:

Regarding the proofs, the papers [51, 52] contain some incorrect statements and incomplete arguments, which we have attempted to point out to the reader. (Some of the mistakes in [51] were corrected in [52].) We did not find any serious problems, meaning problems that cannot be corrected using the methods introduced by Perelman.

Similarly, Morgan and Tian wrote:

In this book we present a complete and detailed proof of the Poincaré Conjecture. … The arguments we give here are a detailed version of those that appear in Perelman’s three preprints.

There is also the account of Huai-Dong Cao and Xi-Ping Zhu, which Yau himself refereed.

On top of those three detailed accounts of Perelman's proof, there have been more recent developments. Terry Tao mentions the recent survey by Richard Bamler. Moishe Kohan mentions Kleiner and Lott's Geometrization of Three-Dimensional Orbifolds via Ricci Flow and Bamler and Kleiner's proof of the Generalized Smale Conjecture. So contrary to the impression you might form from what Yau said, the community is indeed continuing to milk Perelman's ideas and apply them to solving new problems. If there are specific technical points which Yau thinks are obscure, I am sure that other researchers would be happy to address them if Yau were to spell them out explicitly. Until then, there is no credible reason to doubt the fundamental correctness of Perelman's arguments.

$\endgroup$
7
  • 2
    $\begingroup$ @QuartoBendir I am prepared to tone down my remarks; can you point out more specifically what you think I should change? Remember, Yau himself admits that he is being controversial, so saying that he is being controversial does not by itself cast aspersions on him. Regarding community wiki, note that the comments are almost uniformly against Yau, so that is why I tried to capture some of that in my answer. But I agree that it is better to err on the side of neutrality if possible. $\endgroup$ Nov 18, 2021 at 3:58
  • 27
    $\begingroup$ I found these remarks about Yao to be careful and measured. $\endgroup$
    – Nik Weaver
    Nov 18, 2021 at 4:15
  • 15
    $\begingroup$ @QuartoBendir It seems to be a fact, not disputed by either side, that Yau is not on good terms with much of the rest of the relevant community. To understand why the mathematical community at large rejects Yau's remarks, it seems that one needs to be aware of this tension. Otherwise, that rejection is mysterious. That's why I think that commenting on this tension is relevant to the answer. Why try to pretend there is no tension? Yau himself does not pretend. He admits his comments are "politically fraught" and will "get him in trouble." $\endgroup$ Nov 18, 2021 at 5:31
  • 1
    $\begingroup$ Clearly enough my concerns aren't shared by others, I've deleted my previous comments $\endgroup$ Nov 18, 2021 at 14:44
  • 1
    $\begingroup$ @MoisheKohan Please feel free to add it yourself. $\endgroup$ Nov 30, 2021 at 2:00
21
$\begingroup$

The question is, unavoidably, somewhat ambiguous. Here are seven points, hopefully rather objective:

  1. There are well-known expositions of Perelman's work by Cao & Zhu, Kleiner & Lott, and Morgan & Tian. Many parts also appear in Bennett Chow's series of six books on Ricci flow.

  2. As also said in the wikipedia article linked to in the question, Morgan and Tian's exposition of Perelman's shortcut to the Poincaré conjecture contained an incorrectly computed evolution equation, not identified as such until 2015. (This is part of the "shortcut" argument alluded to by Terry Tao in his comment.)

  3. As also said in the wikipedia article linked to in the question, Kleiner and Lott's exposition of Perelman's work used an incorrect statement of Hamilton's compactness theorem, in which completeness and local uniform curvature bounds are falsely said to imply completeness of the limit

  4. The errors in #1-2 are of the type which could in principle lead to fatal problems; compactness arguments in particular are used at nearly every stage of the work.

  5. The errors in #1-2 are strikingly elementary in relation to the complexity of the arguments.

  6. It is common for mathematical papers to have mistakes of a similar nature, as widely acknowledged. See e.g. Why doesn't mathematics collapse even though humans quite often make mistakes in their proofs?.

  7. In the last ten years, seemingly the full range of Perelman's results have been extended and developed by Bamler, Kleiner, and Lott, and so Yau's "no one" sentence is not accurate. Kleiner and Lott's orbifold geometrization paper claims to give a new proof of standard geometrization differing from Perelman's in certain respects.

  8. The part of Perelman's work to do with finite-time analysis (which I don't believe Yau is referring to) has been extended to higher dimensions by Brendle, and adapted to mean curvature flow by Brendle, Huisken, and Sinestrari, and others.

There is thus a timeline of Perelman publishing his work in 2002-2003, with careful expositions appearing 2003-2006, and potentially problematic errors identified and resolved in 2013 and 2015. To me, and especially given #4, this suggests that despite all of the attention given to Perelman's work (and claims for 10+ years from some people about how carefully it has been checked), a certain amount of humility is required in approaching the question. As such, Yau's first sentence (as stated) seems perfectly well-founded as a personal opinion, and his last two sentences become well-founded if changed to acknowledge the three mathematicians in #6. The question of whether three is curiously low, or reliable to external observers, is perhaps too subjective.

Ideally it would be possible for an expositor or researcher to reorganize the logic of some of Perelman's arguments and so to present them from a novel perspective. Of course this is a tall order, although one closely aligned with the accepted answer in the MO question linked to in #5. Something like this has been (for some time) very successfully accomplished, for instance, in Perelman's classification of shrinking Ricci solitons (although this is not one of the more demanding parts of the proof). Also, the very recent results on kappa solutions by Brendle, Daskalopoulos, and Sesum lends some conceptual/structural clarity to parts of Perelman's circle of results, although I haven't studied their proofs. However, Yau is explicitly not referring to these parts of the argument.

$\endgroup$
9
  • 2
    $\begingroup$ What are in your opinion the least understood part of the proof (of the geometrization)? Here "least" refers to the number of people who digested the part. Also isn't there also arxiv.org/abs/0706.2065 that treats an important special case? $\endgroup$ Nov 18, 2021 at 3:55
  • 11
    $\begingroup$ Errors found in expositions of the proof is very different from errors found in the proof. In particular, a big source of confidence in the proof is the existence of multiple independent expositions, each of which has been worked on or read by however many people. For each individual exposition, the source of confidence is not there, so it's not surprising if there's an error. $\endgroup$
    – Will Sawin
    Nov 18, 2021 at 4:03
  • 7
    $\begingroup$ In fact it's possible for finding and correcting an error in the exposition to increase our confidence in the proof, if we go back to the original and see there is some terse remark that the expositors were originally unable to interpret but upon reflection clearly refers to the error-fixing argument. If this happens (and I have no idea if it did in this case) it suggests the original author really was very careful and then didn't write all the details. $\endgroup$
    – Will Sawin
    Nov 18, 2021 at 4:05
  • 2
    $\begingroup$ @RW In this case, Wikipedia provides excellent pointers to the relevant primary literature. $\endgroup$ Nov 18, 2021 at 4:05
  • 14
    $\begingroup$ @WillSawin Indeed, for the 2015 Morgan-Tian errata in arxiv.org/abs/1512.00699 , the original paper of Perelman arxiv.org/abs/math/0307245 did indeed contain "some terse remark that the expositors were originally unable to interpret but upon reflection clearly refers to the error-fixing argument"; see the second paragraph of the Morgan-Tian correction and the sentence after (3) in Perelman's paper (Morgan-Tian initially bounded the ... terms incorrectly by $O(k^2)$ but Perelman correctly asserted the bound of $O(k^2+k)$, which the Morgan-Tian correction verifies). $\endgroup$
    – Terry Tao
    Nov 18, 2021 at 23:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.