Sunday, October 23, 2016

The concordance model strikes back

Two weeks ago, I summarized a recent paper by McGaugh et al who reported a correlation in galactic structures. The researchers studied a data-set with the rotation curves of 153 galaxies and showed that the gravitational acceleration inferred from the rotational velocity (including dark matter), gobs, is strongly correlated to the gravitational acceleration from the normal matter (stars and gas), gbar.

Figure from arXiv:1609.05917 [astro-ph.GA] 

This isn’t actually new data or a new correlation, but a new way to look at correlations in previously available data.

The authors of the paper were very careful not to jump to conclusions from their results, but merely stated that this correlation requires some explanation. That galactic rotation curves have surprising regularities, however, has been evidence in favor of modified gravity for two decades, so the implication was clear: Here is something that the concordance model might have trouble explaining.

As I remarked in my previous blogpost, while the correlation does seem to be strong, it would be good to see the results of a simulation with the concordance model that describes dark matter, as usual, as a pressureless, cold fluid. In this case too one would expect there to be some relation. Normal matter forms galaxies in the gravitational potentials previously created by dark matter, so the two components should have some correlation with each other. The question is how much.

Just the other day, a new paper appeared on the arxiv, which looked at exactly this. The authors of the new paper analyzed the result of a specific numerical simulation within the concordance model. And they find that the correlation in this simulated sample is actually stronger than the observed one!

Figure from arXiv:1610.06183 [astro-ph.GA]


Moreover, they also demonstrate that in the concordance model, the slope of the best-fit curve should depend on the galaxies’ redshift (z), ie the age of the galaxy. This would be a way to test which explanation is correct.

Figure from arXiv:1610.06183 [astro-ph.GA]

I am not familiar with the specific numerical code that the authors use and hence I am not sure what to make of this. It’s been known for a long time that the concordance model has difficulties getting structures on galactic size right, especially galactic cores, and so it isn’t clear to me just how many parameters this model uses to work right. If the parameters were previously chosen so as to match observations already, then this result is hardly surprising.

McGaugh, one of the authors of the first paper, has already offered some comments (ht Yves). He notes that the sample size of the galaxies in the simulation is small, which might at least partly account for the small scatter. He also expresses himself skeptical of the results: “It is true that a single model does something like this as a result of dissipative collapse. It is not true that an ensemble of such models are guaranteed to fall on the same relation.”

I am somewhat puzzled by this result because, as I mentioned above, the correlation in the McGaugh paper is based on previously known correlations, such as the brightness-velocity relation which, to my knowledge, hadn’t been explained by the concordance model. So I would find it surprising should the results of the new paper hold up. I’m sure we’ll hear more about this in the soon future.

42 comments:

Henry Norman said...

How is galactic redshift related to galactic age?

Sabine Hossenfelder said...

Higher redshift (larger z) means larger distance means younger galaxy.

Carsten said...

Dr B.,

thanks for the nice pointers. I hope you continue to watch at and comment on this discussion closely, this is most interesting. The authors of the 2nd paper make a few bold statements based on only 18 simulated galaxies in a very limited mass range using a single algorithm. Interesting to see how far they go out on a limb ... I also liked your comment "If the parameters were previously chosen so as to match observations already, then this result is hardly surprising."
I think this highlights a problem in modern astronomy. Who from the outside can really judge what a specific simulation does, what errors is includes, what assumptions, what parameters were adjusted to get pleasing results?

Also interesting how fast two new papers were posted commenting on the observation, must have been a few quick reviewers. :)

I have a question:
I guess proper rotational curves can only be obtained for the nearest galaxies? At least the SPARC galaxies observed are all within a range of 128Mpc and I guess the redshift z even for the farest of them is way too low to analyze if there is a z-dependency on the g_obs/g_bar relation?

Cheers,

Carsten

Uncle Al said...

That fails versus z. Massive central black holes seed galaxies. Early massive metal-deficient stars rapidly evolve into black holes. Later metallized stars are smaller, giving abundant neutron stars. Dark matter is primordial and hot with no mechanism to radiate or absorb, with only gravitational interactions. Galaxies irreversibly scavenge dark matter in deep gravitational wells over time. Distributed dark matter monotonically declines locally and overall. It is worse if dark matter self-annihilates into photons.

Nick M. said...

Hi Sabine,

Just some clarification on terminology:

In the abstract of the "La Fin du MOND?..." paper, use is made of the phrase "dissipative collapse of baryons". Now I believe that the word 'baryons' is a reference to ordinary matter, but what to make of the phrase 'dissipative collapse'? Googling 'dissipative collapse of baryons' only provides links to various research papers where the use of these words is already understood. So what does "dissipative collapse of baryons" mean?

Also, in your definition of dark matter, you describe it as "a pressureless, cold fluid". Is the condition of 'pressureless' a consequence of the fact that dark matter, in addition to possessing little capacity to interact with ordinary matter, also has (apart from gravity) little, to no, self interaction?

Thanks In Advance For Your Reply,
Nick

Ben Keller said...

Hi Sabine,

Thanks for reading my paper :). The simulations this paper used, the MUGS2 sample (see https://arxiv.org/abs/1505.06268 and https://arxiv.org/abs/1604.08244) cover about half the dynamic range of SPARC (in dex, much less if you calculate overlap linearly). We went out of our way in these simulations to NOT tune any parameters from the top-down. Our model for supernova feedback is based on known, small-scale physics, rather than simply trying to tweak until the galaxies look correct. There was no guarantee when we started MUGS2 that things would come out looking good, and in fact, one of the big results we have is that for galaxies larger than our own, supernova cannot regulate star formation (more evidence that feedback from SMBHs is dominant for massive galaxies).

As for the standard "LCDM problems", like DM cores, missing satellites, etc, there have been a number of studies in the last ~5ish years that have started to show that baryonic physics (feedback from stars, collisional collapse of gas) need to be taken into account to actually understand what LCDM predicts. It's not yet conclusive, but simulations like APOSTLE (https://arxiv.org/abs/1511.01098) show that LCDM does predict DM cores, if you include feedback from massive stars in your simulations.

Ben

Sabine Hossenfelder said...

Ben,

Thanks for the explanation. I don't doubt that. It's just that, well, I know some things about model building, and I know that if you're looking to explain something 'naturally' you will eventually find a way to do it. So, I'm pretty sure that you are right that adding the right combination and right amounts of baryonic physics can explain all the data. But there's confirmation bias and post-selection in this model building. How many models have been tried before one worked? Theorists have much more freedom in their hypothesis tinkering than experimentalists. In the end, the only thing that helps sorting out models is making predictions. Hence, I appreciate you made a prediction and I hope it can be tested.

And while you're here: what happened with the Tully-Fisher relation? Any wiser on that? Best,

B.

Shantanu said...

Ben or anyone else: Do you know if the Millenium and Bolshoi simulations also agree
with your simulations on this point?

Sabine Hossenfelder said...

Millenium I think is dark matter only, doesn't tell you anything about the baryons.

Phillip Helbig said...

"Higher redshift (larger z) means larger distance means younger galaxy."

It means a longer lookback time to the galaxy, so a younger galaxy if compared to a galaxy at lower redshift if the two formed at the same time.

The question might have been "How does one calculate lookback time from redshift".

Phillip Helbig said...

"Thanks for the explanation. I don't doubt that. It's just that, well, I know some things about model building, and I know that if you're looking to explain something 'naturally' you will eventually find a way to do it. So, I'm pretty sure that you are right that adding the right combination and right amounts of baryonic physics can explain all the data. But there's confirmation bias and post-selection in this model building. How many models have been tried before one worked? Theorists have much more freedom in their hypothesis tinkering than experimentalists. In the end, the only thing that helps sorting out models is making predictions."

I agree. Not long ago, I read a paper with a similar theme which sounded quite promising. Then I read how many million hours of CPU time were involved. So, no, I haven't checked the calculations. :-|

This is not necessarily a show-stopper. Climate models are similarly complex.* But there is a consensus (leaving aside the crackpots, as one should) on anthropogenic global warming, for example. But it took a long time. Climate modellers were also in a better position to get funding, because they could cite "impact".

---------------------
*I worked in climate modelling for a while. (In fact, one of my more highly cited papers (which admittedly is not saying much) was in this field.) When I started out (this was back in the early 1990s, when RAM was DM 100 per MB and a 1-GB disk was huge), a colleague told me that one was classified into a more advanced group if one had more than 100 MB of disk space. I remember thinking that that wasn't too much, even back then, until I realized that they were talking about source code. Similarly, I thought that a day of Cray time wasn't that much, until I realized that they were talking about just the compile time.

Sabine Hossenfelder said...

Henry, Phillip,

Phillip is right. Sorry that I was sloppy there. Higher z doesn't necessarily mean the galaxy is younger, it means you look at a galaxy at a time when the universe was younger. This will on the average mean the galaxy is younger, but each single case depends on when the galaxy formed.

Phillip Helbig said...

And if you want to calculate the lookback time as a function of redshift, then Ned Wright's Cosmology Calculator is your friend. There is also an inverse version which can calculate redshift from light-travel time. (In general, the solution involves elliptic integrals, though of course one can use numerical integration is computing power is not an issue. For certain special cases, including the spatially flat case, there are special analytic solutions. (Interestingly, there is an analytic solution for light-travel time for the general flat universe, but not for distances.)

nicolas poupart said...

D.G. Russell (2005), Intrinsic Redshifts and the Tully–Fisher Distance Scale, Astrophysique Space Science, 299: 405. doi:10.1007/s10509-005-3426-2, Springer Link.

Shantanu said...

Just FYI, Milgrom has written a rebuttal to this paper
https://arxiv.org/abs/1610.07538

Ben Keller said...

RE: Model building.
I couldn't agree more. We are always desperate to get observers to tell us the answers, because parameter space is vast compared to the actual data. The hope with simulations at the scale I work in is that enough constraints from below (small-scale, individual stars/molecular clouds) and above (galaxy populations, etc) can work to really confine what is plausible. And unlike pure theoretical physics, all of the underlying physics is ostensibly "known": gravity, hydrodynamics, and a bit of radiative transport. Not that "known physics" makes our lives all that much easier :P

RE: Millenium simulations.
Best bet for looking at a large box like Millenium or Bolshoi would be either Eagle or Illustris. They don't have the resolution of a zoom-in though, so their might be some dynamical effects they miss (and definitely wouldn't be able to do some of the smaller dwarf galaxies from SPARC).

RE: TF relation.
I'm swamped with postdoc applications right now, so I probably won't be able to get much new science done for a few weeks. Hoping to get a followup to this brief letter before 2017 though!

Louis Wilbur said...

The preprint at https://arxiv.org/abs/1610.07663 is relevant to this issue. I hope that readers of this blog find it useful.

akidbelle said...

Hi Sabine,

despite your efforts I am not sure I understand (or maybe there is too much smoke from various papers). Could the correlation mean that baryonic matter is actually "also dark"? (the only source term in the equations).

If so (and if I understand..) it should be the only source of gravity terms except dark energy in the Friedman equations. Would that be a big blow to the big bang scenario?

Thanks,
J.

Ben Keller said...

Ah, there we go: similar analysis, done with EAGLE. General agreement with my paper, interesting that they don't see a redshift dependence though! It will be fun to tease apart the differences that lead to this.

Sabine Hossenfelder said...

Nick M,

Yes, "baryons" are basically normal matter - stars and gas. Dissipative just means they take into account friction/heat transfer. Best,

B.

Sabine Hossenfelder said...

akidbelle,

Yes, lots of papers on this in a rapid sequence. I was just reading Milgrom's one. In a nutshell, he says the same as I did in my post, but with more details and less politely ;)

No, this doesn't mean that baryonic matter is also dark.

It's roughly as follows: We know there's a discrepancy between the acceleration of stars in galaxies and the total baryonic ("normal") mass in these galaxies. Leaving aside otherworldly speculations, there's basically two ways to explain that a) there's some stuff in the galaxies which is non-baryonic and hard to see ("dark matter") or b) General Relativity must be amended at this scale.

The current consensus is a, the competitor is b. The benefit of a is that you can fit with this pretty much anything because it's a very flexible model. The benefit of b is basically the opposite, it's a very unflexible model but still it seems to get much right. Otoh, b doesn't seem to work well on scales of clusters, or at least it isn't understood exactly how. John Moffat claims everything works.

Now in the above mentioned paper the authors claim basically that in a numerical simulation they've done, a also reproduces some of the correlations of b. Problem is, they use a numerical model which has been produced to the end of generating galaxies that look pretty much like the ones we observe. And as Milgrom points out in his paper (I didn't know) not even that is the case. This leaves you to wonder how much went into the simulation. Ben Keller above says it's just baryonic physics. I don't actually doubt that. It's just that if you've ever had anything to do with astrophysics, you know that there isn't any such thing as 'just baryonic physics'. This isn't the standard model. There's lots and lots of assumptions about how and when stars form, or collapse, or supernovae blow up, and what their magnetic fields do, and so on. It's highly complex (partly even chaotic) physics, many-body, which you don't compute like, say, a Higgs cross-section, if you see what I mean.

Best,

B.

akidbelle said...

Many thanks Sabine for this long explanation.

Please could you tell me what to read for GR modifications at this scale?

Best,
J.

Sabine Hossenfelder said...

akidbelle,

This might be a good starting point.

Uncle Al said...

Gravitation? Two black hole mergers had immediate equilibrium, small binding energy, no quantum or information anomalies. No frou-frou.

Equivalence Principle (EP): The pendulum equation has no bob; Einstein's inertial elevator. Torsion pendulums have bobs. Their angular momenta are mirror-asymmetric pseudovectors. Einstein-Cartan-Kibble-Sciama gravitation: EP = true, achiral spacetime curvature. EP = false, chiral spacetime torsion, a trace vacuum left foot. Opposite shoes embed with different energies, vacuum free falling along non-identical minimum action trajectories. Divergence is baryogenesis and Milgrom acceleration, ~10^(-10) relative.

Test masses are chemical (atom-scale geometry) not physical (composition, field). Resolving gravitation is empirically trivial but politically impossible.

Shantanu said...

Ben: If you are reading this, do you know if your simulations or any other simulations can explain why the asymptotic centripetal
accelerations of Dark matter dominated systems is approximately constant (See Figure 3 of 1401.1146)?
I have asked this question to many people who do simulations and from what they told me, simulations cannot reproduce this
particular observation.

akidbelle said...

Dear Sabine,

thanks for the reference; after reading it, I understand that a) and b) are competing because no-one solves the main failure of quantum theory and GR, which is to explain the existence of a big bang. In facts, all I see (in both cases) is an attempt to minimal progress; essentially modelling without new physical understanding (- but maybe I see what I want to see). What do you think?

Best,
J.

Sabine Hossenfelder said...

akidbelle,

Yes, you see what you want to see. Compared to big bang/toe/gut scales these are low energy effective theories. What happens at the big bang is irrelevant for this discussion. What happens at the big bang might one day reveal a more fundamental explanation for either dark matter or modified gravity, but it doesn't remove the question which one is correct, right here, right now. Best,

B.

8c7793aa-15b2-11e5-898a-67ca934bd1df said...

I guess this means that WIMPs and MACHOs are back in play, since axions are much more amenable to the high density, strongly coupled, low temperature scenario implied by coherence in the Lambda CDM concordance model as suggested by the rotational curves.

That should make a whole lotta people happy.

akidbelle said...

Dear Sabine,

OK, maybe I see what I want to see.

With all respect, the question "which one is correct" does not imply that any is physically significant. (i.e. dark matter particles do not exist AND the current scents of modified gravity are only a trick in the equation without physical insight or enlightening principle to back it.)

It can also be that gravity theory needs a complete conceptual reset! What I mean is that you would not name such horror (or beauty?) "modified gravity".

Best,
J.

Nick M. said...

Thanks Sabine,

And sorry for the rather elementary question. I was ultimately able to surmise that the use of the term 'dissipative' was in reference to the action of non-conservative forces. What threw me off was the use of the phrase "dissipative collapse of baryons". It just came across as so incredibly apocalyptic sounding. ☺

Stuart said...

Are you folks aware that MOND fails dismally when it comes to rotation curves of early type galaxies dominated by population II stars? https://arxiv.org/abs/1606.05003

Phillip Helbig said...

From the paper: "Also, while the slopes of the mass density profiles inferred from galaxy dynamics show consistency with those expected from their stellar content assuming MOND, some profiles of individual galaxies show discrepancies."

"Some show discrepancies" is not the same as "fails dismally".

piein skee said...

Doc Hossenfelder says "Now in the above mentioned paper the authors claim basically that in a numerical simulation they've done, a also reproduces some of the correlations of b. Problem is, they use a numerical model which has been produced to the end of generating galaxies that look pretty much like the ones we observe."

This and the other statements you're making such as the part of prediction in science, combined as they are with a willingness to cite your own record for case-in-point instances (e.g. your first hand experience of simulation work), begin to stand you out. Is anyone else doing that? I don't think so (if anyone knows pls say).

I hope you commit to this. If you do, I will help you.

Sabine Hossenfelder said...

piein,

Yeah, I'm pretty good at standing out, though not always for the best reasons ;) In any case, I haven't spent a lot of time on this topic. I happen to know that there's a philosopher of science who wrote her thesis on MOND vs CDM (and it's really MOND, not modified gravity), which I think is very interesting, though maybe for a physicist not quantitative enough. However, as I've said numerous times elsewhere, I don't think you can understand the philosophy of science without the sociology of science. In summary: it's a mess. There's no easy answer. Time will tell.

Also, I'll have another post on the topic in the next days I think because there have been a few more papers. Best,

B.

Sabine Hossenfelder said...

Stuart,

There's a double-standard here. CDM fits galaxies only after a lot of parameter twiddling and tuning but then it somehow counts against a much simpler model if one of its approximations (please keep in mind everyone agrees that MOND needs a relativistic completion) doesn't immediately work out for all and everything. I find this very bizarre.

Look, modified gravity is a theory with additional fields which in some limit reproduces MOND. Galaxies with different symmetries and/or which haven't yet settled are very plausibly described by different solutions than spiral galaxies, and the MOND limit might not apply. To figure that out, you need to do the same thing as in GR, basically, solve the field equations for a given mass distribution. So really the referral to MOND is a red herring. Best,

B.

Phillip Helbig said...

"There's a double-standard here. CDM fits galaxies only after a lot of parameter twiddling and tuning but then it somehow counts against a much simpler model if one of its approximations (please keep in mind everyone agrees that MOND needs a relativistic completion) doesn't immediately work out for all and everything. I find this very bizarre."

Indeed. I am on the fence about MOND. I think both sides overdo the rhetoric. But I'm on the fence because I don't know enough. Certainly the MOND people are not cranks, and my impression is that not only do they know more about conventional astrophysics than conventional astrophysicists know about MOND, they know more about conventional astrophysics than conventional astrophysicists know about conventional astrophysics.

There is also the problem of moving the goalposts. First people said that it isn't relativistic. Then Bekenstein (hard to find a better relativist than Bekenstein) published a relativistic theory. Then people said "OK, but it's not elegant". (Whether his theory has yet been ruled out is another question. But at least it makes predictions.)

There is also the elephant in the room, namely bias on the part of the scientists. There is confirmation bias: write the paper if you can reproduce MOND phenomenology, otherwise tune your model more. Sabine has hinted at this. But there is also publication bias. "LambdaCDM fails to model galaxies realistically" is probably more difficult to publish than "MOND falls naturally out of LambdaCDM" (even if, in fact, it doesn't). Also, what do you show? Some sort of average? One of the strengths of MOND is not only some good fit, but also no scatter other than that explicable by observational uncertainties. The simulations need to show the scatter as well, not just the mean. And so on.

I've just finished a book on MOND, by the way. Not because I'm convinced that it is true, but because I am still learning.

What certainly is true is "MOND phenomenology". Whatever causes it, it has to be explained. It's not impossible, but saying that the simulations spit it out after millions of hours of CPU time is not elegant. Not that this means that it is wrong, only that it takes a lot of (not only CPU-) time to check.

Stuart said...

@Phillip kindly read this part of the paper "With this in mind, the following could be considered as a more
serious challenge for MOND than the mismatches for individual
galaxies. We showed our comparisons with the MOND predictions
when using the simple interpolating function, since it gives more
consistent results for our sample than the standard one (cf. also
Famaey & Binney 2005; Sanders & Noordermeer 2007; Weijmans
et al. 2008; Milgrom 2012; Chae & Gong 2015). While changing
the MOND constant a0 within the range of values used for spiral
galaxies only leads to minor changes, switching to the standard in-
terpolating function has more severe effects. The number of density
profiles that are still consistent with MOND given our assumptions
and estimates of uncertainties is roughly halved as compared to that
when using the simple interpolating function. In the MDA relation
this is even more evident. The calculation with the standard interpo-
lating function largely underpredicts the mass discrepancy for our
sample (Fig. 2) and marks essentially the lower edge of the trend
of observed MDA curves, while the MDA relation calculated with
the simple interpolating function runs through the middle of this
trend. However, the opposite is true for the comparison sample of
spirals from Famaey & McGaugh (2012), for which the standard
interpolating function provides a superior representation. This is at
odds with MOND, where there should be one universal interpolat-
ing function and MDA relation.
The above is consistent with G"
@Bee
MOG theories works much better than MOND and this poses a problem to LCDM advocates since they are compelled to explain MOG results using very few parameters. I find MOG successes more of a menace to quantum gravity theorists since it demands extra particles or properties of the graviton complicating QG. In my opinion MOG theories are not favored by most physicists partly because of this aspect.

Sabine Hossenfelder said...

Phillip,

I'm totally with you on that. There might be good reasons against it. But if so, I'd really like to see them.

Phillip Helbig said...

"I've just finished a book on MOND, by the way. Not because I'm convinced that it is true, but because I am still learning."

Just to be clear, I've just finished READING a book on MOND.* :-)

A review will appear in due course in my series of book reviews for The Observatory.

________________________
*Old joke: This summer, I plan to go to the south of France and finish my second book. I'm a slow reader.

Uncle Al said...

Elegant gravitation theory is untestable, parameterized untestable, or Yukawa potential always short an observed decimal place. Postulated mirror-symmetric universes suppress willful violation testing, Cox recanted vs. Yang and Lee (and Madame Wu!).

Matter universes cannot be exactly mirror-symmetric (Sakharov conditions). Milgrom acceleration over time and space is treacherously effective. Both arise from trace mirror-asymmetric spacetime. Postulated spacetime geometry is falsified with test mass geometry (chemistry) in a geometric Eötvös experiment. Composition and field are inert.

Pertinent test suites, qualified apparatus, and commercial test masses exist. Gravitation postulates "not the answer." Falsify the postulate, not derived theory. Observe the naked Emperor.

Shantanu said...

Sabine sorry for the OT question. But are the videos of the QG symposium in Frankfurt a month ago online?
I couldn't find it. Am very eager to watch the talks

Sabine Hossenfelder said...

Shantanu,

No, sorry, they aren't. I know, I too find it annoying. Last thing I heard was that the editing turned out to be more time-consuming than expected and they've outsourced it and now who knows when it'll be done.