Fraud in superconductor physics | Hacker News Reader

Fraud in superconductor physics

twitter.com

168 pointsgumby40 comments

The comment by Brian Skinner who discovered the correlated noise in a graph is available on Arxiv [1]. It's not claiming fraud but points out that such an effect of correlated noise below a threshold temperature between seemingly independent measurements has not been observed before.

Quit-your-bullshit played safe ;)

[1] https://arxiv.org/pdf/1808.02929.pdf

It is indeed a very well written comment drawing attention to the issue. One imagines that there could be some explanation that isn't a deliberate attempt at data fabrication. Perhaps there was some temperature dependent process in their measurement apparatus that they didn't understand and it was getting suppressed by high magnetic fields. In their rush to show their results, this could have been overlooked. I don't know how likely that is, but it seems plausible in some sense.

That is a really smart way to play it, because while it is astronomically unlikely, it's possible the answer is "woops, we charted the wrong data there, here's the actual results" and we enter a beautiful new world of room temperature superconductivity.

I get kind of exhausted of the typical response to Hot New Shit In Science, which hasn't changed in about four hundred years: outright, arrogant dismissal.

One would hope that the correlated noise was a problem in a plot command. Two observations. 1) The correlated noise in that plot seems obvious. If one was to fake data, why make it that obvious? 2) If it was a plotting error, that would be easy to check and to respond about.

This is giving me flashbacks to the whole STAP cells fiasco of a few years ago where a research group in Japan published a "simple" way to generate pluripotent stem cells. Everyone and their mother tried to replicate the result and eventually the paper was shown to be fraudulent. On the bright side, the larger scientific community acted as a backstop to the journal/peer-review system. On the other hand, the amount of resources expended to nullify this claim was probably enormous and is hard to quantify. Putting out bad data not only poisons the well, it also chews up lots of productivity cycles of everyone in the field.

It's giving me flashbacks to

https://en.wikipedia.org/wiki/Schön_scandal

You are so right. It feels like a dejavu:

"Lydia Sohn, then of Princeton University, noticed that two experiments carried out at very different temperatures had identical noise.[3] When the editors of Nature pointed this out to Schön, he claimed to have accidentally submitted the same graph twice."

https://en.wikipedia.org/wiki/Sch%C3%B6n_scandal#Allegations...

The title is a bit aggressive I think. While the correlated noise discussed in the tweets is suspicious, there doesn't seem to be any definitive evidence of fraud yet.

It's also possible that the plot of the noise was an honest error or something else.

Well, besides apparently spoofing one of their senior colleagues' email addresses to send another colleague an email asking him not to criticise the authors' works. Through Protonmail.

It seems to me that the forged emails are what is being called out as a bizarre fraud, since they definitely are fraudulent. The rest of the thread makes it clear that the experimental results are still to achieve a consensus one way or the other (although it doesn't look good, does it?)

Matching noise in plots isn't proof of fraud, but it's something that is so suspicious that it changes the game from "convince me that this paper is fraudulent" to "convince me that this paper isn't fraud." In that light, other worrisome elements of the situation are problematic:

* Claiming one of the holy grail properties of modern physics is achievable with rather pedestrian materials.

* The methodology section is apparently on the vague side. Sort of on the order of “we prepared silver nanoparticles using standard techniques, then incorporated them into into a gold matrix.”

* Response to the announcement of "your data presentation has serious errors" is "we won't share our raw data to confirm."

I think the bizarre thing was that someone (allegedly) posed as a researcher in an attempt to get a different researcher to back down.

Could be an honest mistake, except that the authors are defending it, which rings the alarm bells. Not even getting to the spoofed email...

The reason this looks more like fraud then accident is the forged email and the weird facebook request, both described down thread.

A FB request from a rando by itself isn't weird -- FB pretty much invites this to happen -- and a FB request from a brand new account would be both believable and, in the detail, suspicious (because of the name). But a random request from an account name from the forged mail but which was created a month before makes me wonder if somebody hired some sort of "reputation" service to defend the paper, which is, in my mind, just completely bonkers.

is there a service that turns these into readable correspondence to the editor? because I am not going to read a series of tweets alleging fraud. It's like there is an entire user-visible packetizing mechanism in the way of the content.

https://threadreaderapp.com/ will do it -- https://threadreaderapp.com/thread/1027717419400392705.html

and yes, it's ridiculous that this is necessary

Have you got something that will handle replies (in date order) too?

It's not quite what you're asking for, but Treeverse represents the conversation (including replies) as a tree https://github.com/paulgb/Treeverse (disclosure: I wrote it)

Very handy, thanks

If you do not like to read a thread of tweets on Twitter, ping @threadreaderapp with the word "unroll" on the last comment and a bot will convert the thread into a single post and add the link to it in a reply to your request.

Scrolling worked fine for me.

It's almost like Twitter isn't intended for this sort of communication.

it definitely isn't but unfortunately some of the smartest and most informative people have moved to Twitter and so we have to read what we can where we can

But what's the point of faking this? Wouldn't this be debunked very fast anyway if it was fake? So what's there to gain?

People can do irrational things under pressure to publish.

I'd like to point out 90 Kelvin isn't the highest temperature superconductor (likely) reported, not even close. There were Mercury-based HgBa2Ca2Cu3O8 "high temperature superconductors" (thought to operate similar to the common YBCO ones) that have achieved up to 134 Kelvin years ago.

Recently, there was also a conventional superconductor at up to 203 Kelvin (!) at extremely high pressure based on HS3 or somesuch and that seems to have been pretty widely accepted and replicated. https://www.nature.com/articles/nature14964

It's thought that modifying the sulfur hydride based superconductors could result in near-room-temperature superconductivity (above 0 Celsius).

Occasionally, people claim room temperature superconductivity in carbon allotropes such as graphene, but take those with a grain of salt as they haven't really been replicated.

Anyway, yeah, this silver and gold material is almost certainly bogus.

The plot has other problems also in the symbols choosen to print the values.

The gray cross into a square symbol looks particularly ugly. If the goal is to assure that nobody can understand the real structure of the data; is the perfect symbol. Many values printed in the red set are also hidden by the blue and green dots.

As a guy who does a lot of scientific data analysis and plotting, I'm surprised that no one brought up the obvious explanation; in the general scrum of final manuscript preparation, the data analysis guy mistakenly made a plot with the same data replicated several times, but shifted each time so that the plot looked plausible at first pass. It would require a lot of things to go wrong at once, sure... but with experience, you learn to count on a lot of things going wrong at once, especially just before something like a manuscript submission. :)

But that the authors, who presumably could check the raw data the plot is coming from, didn't immediately come back with this explanation, makes it sound much more likely that this is just fraud. Oh well.

your mixup does not make any sense: look at the 2 traces in question, on the left half both traces seem like a simple small vertical shift of the other, but in the right half of the same plot one plateaus earlier than the other, both plateauing at the same height.

the conclusion is unmistakable: either the data is real and unexplained (or at least no agreement on any explanation yet) or the data was doctored by summing vertically displaced noise with horizontally displaced sigmoid.

If this is fraud, why on earth would you claim something that (theoretically) that is: 1) a holy grail of any field of science. 2) likely to attract a lot of attention 3) likely to be disproved quickly. 4) likely to result in you being compared unfavorably with Fleischmann–Pons?

People are irrational.

https://threadreaderapp.com/thread/1027717419400392705.html

Maybe it’s been answered though I didn’t see it addressed, but couldn’t the noise issue simply be some sort of copy paste error in excel?

In principal it could, and this is why Brian Skinner wrote his note in the way he did, simply pointing out the facts. It is odd, however, that the two data series in question match at lower temperatures but not higher temperatures. It would be odd to be manipulating one portion of a series of data like that.

When you announce room temperature superconductors, you sure as heck better have found and fixed every single error, including dumb ones like excel copy/paste errors. You touch that topic like you touch cold fusion.

And besides, if it was something as simple as that then all they need to do is release their data. They haven't.

It's an arXiv preprint. If it were submitted to a journal I'd agree with you, but the fast sharing of new data is the whole point of arXiv.

Copy and paste would produce the same numbers, and therefore the same pattern. As mentioned, the figures aren't the same at higher temperatures, as well Excel wouldn't copy, paste, add 1 unit which would produce the offset.

Here somebody claims superconductivity above 200 Celsius.

http://www.superconductors.org/216C209C.htm

Superconductors.org is an old and well-known sham.

Reply on news.ycombinator.com