The Data that Threatened to Break Physics

[Andy44]

Good read about a neutrino experiment a few years ago that appeared to violate the cosmic speed limit, how they handled it, how they figured out what was causing the readings, and what the fallout was.

http://nautil.us/issue/24/error/the-data-that-threatened-to-break-physics

I have to say that when I read the initial part of the article my experience with electronics led me to suspect the equipment immediately. There is no such thing as a true square edge pulse; if you look at such a thing on an oscilloscope with high enough time resolution you will see that the leading and trailing edges of a square wave is never perfectly vertical. Even still, this was a tough problem to solve.

I also think it's pretty sad that even though the team and their leader handled everything correctly a scapegoat needed to be chosen after all was said and done.

 

[Ravenous]

It struck me as a really stupid story even at the time. I read the original paper (no I didn't understand 80% of it) and the conclusions were worded very carefully, and obviously not a claim but an appeal for opinions on what went wrong.

Just goes to show you science is just as political as everything else.

 

[Face]

Interesting write-up. Thanks for the link.

This makes me wonder what will happen at Eagleworks if the EmDrive bubble implodes. Bets on who will be the scapegoat then, anyone?

 

[Artlav]

Nevertheless, it seemed clear that someone, somewhere, had made a mistake.

:facepalm:
That's just sad.

 

[kamaz]

The guy did everything correctly, so the fact that he was fired was highly damaging to science in my opinion.

Invalidating the result required setting up another experiment, which had to be budgeted for. Every budget request must contain a justification and that alone required disclosing the anomaly (if only to the funding agency and the reviewers).

 

[Ravenous]

The guy did everything correctly, so the fact that he was fired was highly damaging to science in my opinion.

Exactly. From now on, being the good scientist by reporting an experimental problem is likely to get you fired, if your experiment is unlucky enough to attract press interest.

I heard an interview of someone outside that group, perhaps a month or two later, and he was very critical and stating they simply should not have published the original paper - but it seems they were compelled to.

 

[Face]

Exactly. From now on, being the good scientist by reporting an experimental problem is likely to get you fired, if your experiment is unlucky enough to attract press interest.

To be fair, the interview read to me like the group voted to distrust the leadership, so the leadership resigned "for the benefit of the team" . It was not like some external authority actively fired him.

To me, this is more of an example of the current global practice of "opinion fascism": get enough people loudly proclaiming one opinion, and it will become irrefutable truth for everybody (in the eyes of the proponents). If you dare to counter such an opinion with ratio, you'll be "stormed" into oblivion. If two parties emerge, they will fight so long till everybody is either in the one or the other camp, not giving neutral or - heaven forbid! - rational arguments any right to live at all.

In this case, the media hyped the observations so much that it didn't matter what he really did... only the opinion of the loud crowd mattered to the group, so that they had but no other choice than to distrust their leader. And even the guy who got dumped acknowledges that it was better that way.

I honestly don't like the direction we are going with this.

 

[statickid]

I work with alot of scientists these days, I'm not impressed by the current atmosphere created in the sciences used for federal and state policy.

Funding is provided to "interesting" new studies that are never replicated or verified, because replicating and verifying studies isn't that "interesting".

Many of the scientists I know are reluctant to come to any solid conclusions for fear of being wrong, sued, or fired. Unfortunately, IMO, this makes the whole process invalid.

 

[boogabooga]

I am very skeptical about ultra-precise measurements.

The experimental error was not so easy to discover.

Had the result happened to not exceed the speed of light, an erroneous value would have been obtained without much question.

 

[Artlav]

Had the result happened to not exceed the speed of light, an erroneous value would have been obtained without much question.

Well, the precision was not the goal of the experiment - determining whether neutrinos flip or not was.

 

[kamaz]

After reading this, as a telecom engineer turned scientist...

Whoever made the synchronization system has never taken an intro course in fiber optic technology. A correctly designed system would have detected and signaled a fault condition. (I guess another case of using grad students because you cannot afford the professionals.)

I'm tired at the moment, so if you want the details, please say thanks to this post, if there is interest, I can explain this in more detail during the weekend.

 

[RisingFury]

Experimental setups can't be compared to end user electronics or industry technology.

Industry is all about polish, science is all about finding new stuff. If you spend all your time working on polishing your experiments, you never get any experimenting done


The experiment did not intend to measure the speed of the neutrinos, as Artlav pointed out.

 

[kamaz]

Yet, apparently the experimental protocol required clock synchronization between source and detector for whatever reason. So, it's conceivable that if the timing error coincided with the expected neutrino oscillation period, the experiment could have ended with a false positive.

It isn't about polishing, it's about using industry expertise. You don't use grad students to weld pipes, and you shouldn't use grad students to design fiber optic systems either.
There are professionals who know how to do that correctly and efficiently. Telecom industry keeps their clocks synchronized to sub-ns levels across continents. Synchronizing 2 clocks across 500km is not exactly rocket science.

 

[Urwumpe]

Synchronizing 2 clocks across 500km is not exactly rocket science.

oh, you are pretty far away with that one... it is rocket science. Or at least spaceflight assisted metrology. The best source of synchronization is the GPS after all. If you need nanosecond accuracy, the process is just not done by setting the clock alone, that's a lot of engineering happening before you have confirmed sync.

 

[kamaz]

It does not matter what their primary clock source was, because they have killed the accuracy between the GPS receiver and the detector by using a self-made fiber optic system.

The connector was dislodged, so the rise time of the signal increased from ~100ns to ~200ns. But as anyone who has passed a metrology course will tell you, if your trigger signal has a ~100ns nominal rise time, then it can be maybe be used to measure something to 1us precision. This is the actual problem here, not the fact that the connector was dislodged. Their system is so bad, that if they switch a DAQ card, and the input transistor now switches at 0.60V while the previous one switched at 0.65V, the timing will be off by 10ns. (A CMOS transistor will switch somewhere between 0.6 and 0.8V, depending on the process, batch, or even location on the wafer.)

And ~100ns rise time in a fiber system is absurdly long. An off-the-shelf optical Gig-E PHY has a data clock rate of 1250MHz, which is 0.8 ns per bit. The rise time must be no more than 350ps per the spec. Otherwise the receiver PLL will fail to lock and the link will drop.

Someone here has tried to reinvent a wheel and failed miserably, getting a lower accuracy than available with off-the-shelf components. Simply using a Gig-E/FiberChannel PHY and flipping a bit in the input bitstream on event would provide both better accuracy and protection from dislodged connectors.

 

[RGClark]

...
There are professionals who know how to do that correctly and efficiently. Telecom industry keeps their clocks synchronized to sub-ns levels across continents. Synchronizing 2 clocks across 500km is not exactly rocket science.

Is it really trillionths of a second?

Bob Clark

 

[Ravenous]

Is it really trillionths of a second?

Bob Clark

Very good question. Light won't go that far in a nanosecond. I suppose you could synchronise clocks over that distance using regular sync pulses and locking into the phase, but timing a discrete event is probably more complicated...

 

[Notebook]

This doc shows the old analogue 625 PAL video format timings:
http://www.itelcast.com/_download/ITEL-video-course.pdf

Page 16 shows the tolerance for the leading-edge of line-sync pulse to colour-burst, its +- 50 nsec.

Its all digital these days, so its not a problem(haven't got the equipment to measure it, so it must be ok...)

N.

 

[kamaz]

Depends how you define precision exactly.

As I said above, Gigabit Ethernet (and this is not high-end gear at all) uses 0.8ns laser pulses, driven by clock at the transmitter. At the receiver, a PLL is used to reconstruct the transmitter clock signal (and reconstruct data). So the output of the receiver's PLL will match the master clock to 100ps plus the transmission delay.

The devil is in the details however. First, the speed of light in fiber is variable and depends on the temperature. Second problem is the stability of the master oscillator. The short-term stability (jitter) per the SDH/SONET spec should be +/-46ns, and equipment clocks should be within 3us of astronomical time. See this: http://www.gps.gov/cgsic/meetings/2012/weiss1.pdf

Also, it looks like I have undeservedly criticised their setup above After checking, it appears that GPS is accurate to 40ns, but is GUARANTEED to be accurate to 1000ns. If someone designed the system to the latter value, then having 100ns rise time is excusable. (Still no excuse for not using a proper synchronous transmission system, though!) HOWEVER, if your master clock source (GPS) is accurate to 40ns (or 1000ns), then you don't really have the accuracy to measure a 70ns anomaly in flight time...

The most logical way to run this experiment would be to run the fiber pair between the labs (not that expensive, a telecom company should be able to assemble the circuit together using the dark fiber they have already in the ground). At CERN you have master oscillator (running at say 1GHz) which controls both the laser shining into the fiber, and keys the neutrino source on/off (say every 1 second, or 1E9 cycles). At the detector site, you record the neutrino signals and send them back to CERN together with their clock signal. So you have time t0 when both the neutrinos and sync pulse was sent, time t1 the neutrinos arrived at the detector, and time t2 when the sync signal arrived at the detector, and time t3 when the sync signal returned to CERN -- all measured to the same clock reference. Now, the optical length of the fiber is different than the flight path of neutrinos, but that does not matter, because (t3-t0)/2 gives you the optical length of the fiber. Jitter of the master clock is a problem for each measurement, but it also has an average value of zero, so if you keep the experiment running for an entire day, it will average out.

---------- Post added at 05:04 PM ---------- Previous post was at 04:31 PM ----------

Very good question. Light won't go that far in a nanosecond. I suppose you could synchronise clocks over that distance using regular sync pulses and locking into the phase, but timing a discrete event is probably more complicated...

No, why? You use the carrier frequency to synchronize clocks, and modulation of the carrier to transmit data.

[ame="http://en.wikipedia.org/wiki/DCF77"]DCF77 - Wikipedia, the free encyclopedia[/ame]

 

[RGClark]

....
Also, it looks like I have undeservedly criticised their setup above After checking, it appears that GPS is accurate to 40ns, but is GUARANTEED to be accurate to 1000ns. If someone designed the system to the latter value, then having 100ns rise time is excusable. (Still no excuse for not using a proper synchronous transmission system, though!) HOWEVER, if your master clock source (GPS) is accurate to 40ns (or 1000ns), then you don't really have the accuracy to measure a 70ns anomaly in flight time...

Thanks for that. How is the proposed centimeter accuracy GPS supposed to work?

Bob Clark