Thursday, February 22, 2018

Shut up and simulate. (In which I try to understand how dark matter forms galaxies, and end up very confused.)

Galactic structures.
Illustris Simulation.
[Image Source]
Most of the mass in the universe isn’t a type of matter we are familiar with. Instead, it’s a mysterious kind of “dark matter” that doesn’t emit or absorb light. It also interacts rarely both with itself and with normal matter, too rarely to have left any trace in our detectors.

We know dark matter is out there because we see its gravitational pull. Without dark matter, Einstein’s theory of general relativity does not predict a universe that looks like what we see; neither galaxies, nor galaxy clusters, nor galactic filaments come out right. At least that’s what I used to think.

But the large-scale structure we observe in the universe also don’t come out right with dark matter.

These are not calculations anyone can do with a pen on paper, so almost all of it is computer simulations. It’s terra-flopping, super-clustering, parallel computing that takes months even on the world’s best hardware. The outcome is achingly beautiful videos that show how initially homogeneous matter clumps under its own gravitational pull, slowly creating the sponge-like structures we see today.

Dark matter begins to clump first, then the normal matter follows the dark matter’s gravitational pull, forming dense clouds of gas, stars, and solar systems: The cradles of life.

structure formation, Magneticum simulation
Credit: Dolag et al. 2015
It is impressive work that simply wouldn’t have been possible two decades ago.

But the results of the computer simulations are problem-ridden, and have been since the very first ones. The clumping matter, it turns out, creates too many small “dwarf” galaxies. Also, the distribution of dark matter inside the galaxies is too peaked towards the middle, a trouble known as the “cusp problem.”

The simulations also leave some observations unexplained, such as an empirically well-established relation between the brightness of a galaxy and the velocity of its outermost stars, known as the Tully-Fisher-relation. And this is just to mention the problems that I understand well enough to even mention them.

It’s not something I used to worry about. Frankly I’ve been rather uninterested in the whole thing because for all I know dark matter is just another particle and really I don’t care much what it’s called.

Whenever I spoke to an astrophysicist about the shortcomings of the computer simulations they told me that mismatches with data are to be expected. That’s because the simulations don’t yet properly take into account the – often complicated – physics of normal matter, such as the pressure generated when stars go supernovae, the dynamics of interstellar gas, or the accretion and ejection of matter by supermassive black holes which are at the center of most galaxies.

Fair enough, I thought. Something with supernovae and so on that creates pressure and prevents the density peaks in the center of galaxies. Sounds plausible. These “feedback” processes, as they are called, must be highly efficient to fit the data, and make use of almost 100% of supernovae energy. This doesn’t seem realistic. But then again, astrophysicists aren’t known for high precision data. When the universe is your lab, error margins tend to be large. So maybe “almost 100%” in the end turns out to be more like 30%. I could live with that.

Eta Carinae. An almost supernova.
Image Source: NASA

Then I learned about the curious case of low surface brightness galaxies. I learned that from Stacy McGaugh who blogs next door. How I learned about that is a story by itself.

The first time someone sent me a link to Stacy’s blog, I read one sentence and closed the window right away. Some modified gravity guy, I thought. And modified gravity, you must know, is the crazy alternative to dark matter. The idea is that rather than adding dark matter to the universe, you fiddle with Einstein’s theory of gravity. And who in their right mind messes with Einstein.

The second time someone sent me a link to Stacy’s blog it came with the remark I might have something in common with the modified gravity dude. I wasn’t flattered. Also I didn’t bother clicking on the link.

The third time I heard of Stacy it was because I had a conversation with my husband about low surface brightness galaxies. Yes, I know, not the most romantic topic of a dinner conversation, but things happen when you marry a physicist. Turned out my dear husband clearly knew more about the subject than I. And when prompted for the source of his wisdom he referred to me to no other than Stacy-the-modified-gravity-dude.

So I had another look at that guy’s blog.

Upon closer inspection it became apparent Stacy isn’t a modified gravity dude. He isn’t even a theorist. He’s an observational astrophysicist somewhere in the US North-East who has become, rather unwillingly, a lone fighter for modified gravity. Not because he advocates a particular theory, but because he has his thumb on the pulse of incoming data.

I am not much of an astrophysicist and understand like 5% of what Stacy writes on his blog. There are so many words I can’t parse. Is it low-surface brightness galaxy or low surface-brightness galaxy? And what’s the surface of a galaxy anyway? If there are finite size galaxies, does that mean there are also infinite size galaxies? What the heck is an UFD? What means NFW, ISM, RAR, and EFE?* And why do astrophysicists use so many acronyms that you can’t tell a galaxy from an experiment? Questions over questions.

Though I barely understood what the man was saying, it was also clear why other people thought I may have something in common with him. Even if you don’t have a clue what he’s on about, frustration pours out of his writing. That’s a guy shouting at a scientific community to stop deluding themselves. A guy whose criticism is totally and utterly ignored while everybody goes on doing what they’ve been doing for decades, never mind that it doesn’t work. Oh yes, I know that feeling.

Still, I had no particular reason to look at the galactic literature and reassess which party is the crazier one, modified gravity or particle dark matter. I merely piped Stacy’s blog into my feed just for the occasional amusement. It took yet another guy to finally make me look at this.

I get a lot of requests from students. Not because I am such a famous physicists, I am afraid, but just because I am easy to find. So far I have deterred these students by pointing out that I have no money to pay them and that my own contract will likely run out before they have even graduated. But last year I was confronted with a student who was entirely unperturbed by my bleak future vision. He simply moved to Frankfurt and one day showed up in my office to announce he was here to work with me. On modified gravity, out of all things.

So now that I carry responsibility for somebody else’s career, I thought, I should at least get an opinion on the matter of dark matter.

That’s why I finally looked at a bunch of papers from different simulations for galaxy formation. I had the rather modest goal of trying to find out how many parameters they use, which of the simulations fare best in terms of explaining the most with the least input, and how those simulations compare to what you can do with modified gravity. I still don’t know. I don’t think anyone knows.

But after looking at a dozen or so papers the problem Stacy is going on about became apparent. These papers typically start with a brief survey of other, previous, simulations, none of which got the structures right, all of which have been adapted over and over and over again to produce results that fit better to observations. It screams “epicycles” directly into your face.

Now, there isn’t anything scientifically wrong with this procedure. It’s all well and fine to adapt a model so that it describes what you observe. But this way you’ll not get a model that has much predictive power. Instead, you will just extract fitting parameters from data. It is highly implausible that you can spend twenty or so years fiddling with the details of computer simulations to then find what’s supposedly a universal relation. It doesn’t add up. It doesn’t make sense. I get this cognitive dissonance.

And then there are the low surface-brightness galaxies. These are interesting because 30 years ago they were thought to be not existent. They do exist though, they are just difficult to see. And they spelled trouble for dark matter, just that no one wants to admit it.

Low surface brightness galaxies are basically dilute types of galaxies, so that there is less brightness per surface area, hence the name. If you believe that dark matter is a type of particle, then you’d naively expect these galaxies to not obey the Tully-Fisher relation. That’s because if you stretch out the matter in a galaxy, then the orbital velocity of the outermost stars should decrease while the total luminosity doesn’t, hence the relation between them should change.

But the data don’t comply. The low surface brightness things, they obey the very same Tully-Fisher relation than all the other galaxies. This came as a surprise to the dark matter community. It did not come as a surprise to Mordehai Milgrom, the inventor of modified Newtonian dynamics, who had predicted this in 1983, long before there was any data.

You’d think this would have counted as strong evidence for modified gravity. But it barely made a difference. What happened instead is that the dark matter models were adapted.

You can explain the observations of low surface brightness galaxies with dark matter, but it comes at a cost. To make it work, you have to readjust the amount of dark matter relative to normal matter. The lower the surface-brightness, the higher the fraction of dark matter in a galaxy.

And you must be good in your adjustment to match just the right ratio. Because that is fixed by the Tully-Fisher relation. And then you have to come up with a dynamical process for ridding your galaxies of normal matter to get the right ratio. And you have to get the same ratio pretty much regardless of how the galaxies formed, whether they formed directly, or whether they formed through mergers of smaller galaxies.

The stellar feedback is supposed to do it. Apparently it works. As someone who has nothing to do with the computer simulations for galaxy structures, the codes are black boxes to me. I have little doubt that it works. But how much fiddling and tuning is necessary to make it work, I have no telling.

My attempts to find out just how many parameters the computer simulations use were not very successful. It is not information that you readily find in the papers, which is odd enough. Isn’t this the major, most relevant information you’d want to have about the simulations? One person I contacted referred me to someone else who referred me to a paper which didn’t contain the list I was looking for. When I asked again, I got no response. On another attempt my question how many parameters there are in a simulations was answered with “in general, quite a few.”

But I did eventually get a straight reply from Volker Springel. In the Illustris Simulation, he told me, there are 10 physically relevant parameters, in addition to the 8 cosmological parameters. (That’s not counting the parameters necessary to initialize the simulation, like the resolution and so on.) I assume the other simulations have comparable numbers. That’s not so many. Indeed, that’s not bad at all, given how many different galaxy types there are!

Still, you have to compare this to Milgrom’s prediction from modified gravity. He needs one parameter. One. And out falls a relation that computer simulations haven’t been able to explain for twenty years.

And even if the simulations would get the right result, would that count as an explanation?

From the outside, it looks much like dark magic.

* ultra faint dwarfs, Navarro-Frenk-White, interstellar medium, radial acceleration relation, external field effect


  1. I wonder if anyone is producing a model with both dark matter and gravity modifications?
    Why just one solution?

  2. "With four parameters I can fit an elephant, and with five I can make him wiggle his trunk." John von Neumann

  3. You could look at papers by Justin Khour, he's not exactly doing that, but he has some ideas on how make it look as if gravity was modified in the 'right places'.

  4. From "Introducing the Illustris Project"

    "The 15 or so free parameters of the models, associated mostly with the various feedback processes, all have a physical meaning, and can be assigned numerical values based on underlying principles, but given our ignorance and uncertainties regarding the complicated physics of, e.g. star-formation and black-hole accretion, there is freedom in their exact values. In practice, a subset of them was tuned to their particular values based on test simulations [...] Full details on the physics models and the choice of parameters appear in ..."

  5. what's your take on Lee Smolin Verlinde et al, theory that there is no "modified" graviy in MOND, MOND is simply QG in the deep IR, possibly in association with cc?

    so in effect the modification of GR MOND requires is simply accounting for both dark energy and QG. this is distinct from other proposals such as additional fields in say Scalar–tensor–vector gravity or Tensor–vector–scalar gravity.

    here, MOND is simply explained as a QG phenomenology. GR is just a classical theory.

    since QG is your forte, have you thought about working with Smolin, Verlinde, Stacy McGaugh, Milgrom et al, on showing how MOND can arise from QG of GR in a rigorous way?


    any thoughts comments on the more recent paper on Satellite galaxies of Centaurus A not randomly aligned as a problem for dark matter, but in a plane but is also explained by MOND?
    i wish stacy mcgaugh would blog on it.

  6. "Is it low-surface brightness galaxy or low surface-brightness galaxy? "

    It is definitely "low--surface-brightness galaxy". Definitely.

  7. You didn't mention MOND (MOdified Newtonian Dynamics). While technically "modified gravity", the MOND and modified-gravity communities are quite distinct.

    Also, Stacy isn't completely alone in his fight, though he is arguably the most visible MOND proponent.

  8. So the bottom line seems to be about going back and looking again at everything.. It should not be surprising that phenomena on scales vastly larger than those directly impacting our biology are confusing and not easy to untangle

  9. Very interesting, thanks.
    So, because these low-surface-brightness galaxies behave the same as normal galaxies, it is suggested that they have a much larger proportion of dark matter than normal galaxies. That doesn't sound very convincing at all. How convenient. It really does scream epicycles.

    I would feel tempted to ignore all the simulation results (which are way too easy to fiddle to get a nice result) and only consider the velocity curves. Otherwise, as you say, it's all too confusing.

    (I see you're going to talk about this at the Hay-On-Wye festival in May. I'll be there for the four days, it's a fantastic festival. Hopefully get to say hello. I hope your talk isn't on the Sunday as I can't make the Sunday.)

  10. neo,

    My take is that it's too vague for my comfort.

  11. Phillip,

    Because I wish people would stop talking about MOND. We already know that it's only a non-relativistic limit and can't be the full story. What happens all too often if you talk about MOND is that they will - correctly - point out that MOND is wrong and - falsely - proclaim that this settles the situation.

  12. Andrew,

    Eh, what am I going to talk about in May? I thought I was supposed to talk about my book.

  13. We know dark matter is out there because we see its gravitational pull or ppb Milgrom acceleration, from baryogenesis’ ppb parity violation, from ppb spacetime chiral anisotropy. Noether then leaks ppb non-conservation of angular momentum everywhere and everywhen. they obey the very same Tully-Fisher relation than all the other galaxies because Milgrom acceleration is ugly spacetime torsion.

    shouting at a scientific community to stop deluding themselves Theory’s million+ published pages fail. Dark matter begins to clump first "Dark matter" has no cooling mechanism, cannot lose binding energy. Starting unbound it remains unbound.

    Some modified gravity guy Milgrom acceleration is one-day sourced in commercial microwave spectrometers (Bright Spec FT-MRR) fed milligrams of calculated organic chemistry. Look. moved to Frankfurt and one day showed up in my office to announce he was here to work with me. On modified gravity When you need research in the worst way it Uncle Al’s way. No other cheap, fast, viable observation exists.

  14. ok,

    have you heard of the recent reports on Satellite galaxies of Centaurus A not randomly aligned as a problem for dark matter, but in a plane but is also explained by MOND?

    computer simulation also suggests that Satellite galaxies should be *randomly* distributed based on dark matter halo, but actual observation on Satellite galaxies of Centaurus A, Milky way and Andromeda shows they are aligned on a plane, following baryon distribution. the authors says this is easily explained by MOND as well.

  15. You're listed for "The Dark Universe" with Erik Verlinde:

  16. neo,

    Yes, I have seen the paper and read it. It's literally lying on my desk.

  17. Andrew,

    You are right. I should read my emails better...

  18. It would be cool if at the end of such a post, you'd put a link to a contrary post/paper, where the opposite is argued (MOND doesn't work, Lambda-CDM it is), just so people can get both side of the story.

  19. Sabine

    can you comment on it? it seems relevant to this post and in mond vs dm debate

    thanks in advance

  20. Maro,

    On my blog you get my side of the story.

  21. neo,

    No, I can't. Same problem: It's a numerical simulation and it's basically a black box. The best I can do to is try to guess whether the authors seem to know what they are talking about but that's more psychology than science. Hence I'll keep my mouth shut.

  22. Sabine,

    Thanks for an enlightening and entertaining blog post.

    I have a question concerning the computer models/simulations: Are spinor fields coupled to the gravitational field via the spin connection (and vierbein), or are only standard GR couplings of the tensorial fields via the Christoffel connection considered? Or do the calculations not use such formalism at all because the gravitational field is simply assumed weak?

    I ask for the following reason (perhaps this is in violation of comment rule no. 4?, hope not): The Dirac field couples to the gravitational field via only the axial vector part of the spin connection, that is, via only 4 of the 24 degrees of freedom of the spin connection. As any other field, at least in some local Lorentz frame, can be constructed from spinor fields, its coupling to the gravitational field must also be describable in terms of this axial vector part of the spin connection. This makes me wonder if it is strictly necessary to have all 24 spin connection coefficients enter into the Einstein-Hilbert Lagrangian written in terms of the vierbein and the spin connection; or in other words: is it perhaps possible to 'truncate' the Einstein-Hilbert action?

  23. ok, but if the paper is valid, it seems DM has a lot of very strong constraints, it has to avoid core-cusp problem, explain baryon tuller fisher relation, explain missing satellite yet also explain why they are planar and follow baryon distrubution, explain low surface brightness galaxies, non detection in detectors etc.

    going back to an earlier comment

    "it's too vague for my comfort. "

    starting with GR and cc what additional requirements are necessary to derive MOND as a QG effect

  24. Two cents from an astrophysicist studying the ISM (interstellar medium) in our own Galaxy: these simulations often use few parameters to describe feedback, but only because they are forced to by practical concerns; for accuracy, we would prefer to be able to use many more parameters. The output of the simulations is strongly dependent on the way feedback is parameterized because all parameterizations are simplifications. This is a major field of study that has only been tackled by simulators in the last 5-10 years, again primarily because of computational expense. The current models of feedback are probably very wrong, but they're rapidly getting better.

  25. Sabine,
    regarding the parameters used, I would distinguish between parameters related to the physics of dark matter and parameters used to effectively describe higher concentrations of matter - faormation, evolution and fate of stars. I'm pretty sure, the latter is not simulated using basic particle physics parameters, but rather using some seperately fitted effective models to make a simulation on this arge scale feasible at all. As far as this is done 'honestly', this is ok from my point of view and would not necessarily interfere with the dark matter sector of the simulation. Of course all parameters or embedded models should be handled transparently.
    For the severely dark matter dependend part, I don't understand, why respective obsevable parameters are not extracted from the simulation - or are they?
    Points like the 'cuspiness', Tully-Fisher-Relation for various types of Galaxiies should be essential results of such a simulation. And such results should be analysed with their sensitivity to chosen parameters to gain some insight.

  26. Regarding MOND, Modified Gravity etc. it's somewhat difficult to distinguish obviously different approaches appearing under very similar titles.
    Also, some of these seem to be ruled out by the recent BH-NS merger observation.
    I understand, there remain at least one/some model, which gets the galaxy rotation right. May be, also the relation to only normal matter.
    But how far does it explain the satellite galaxy disks and other observational results?
    I have the impression, that here are opposing camps pointing to specific successes of their model/theory, but both sides not clearly checking the whole picture.

  27. We humans are not aware of the complexity of the things we are dealing with. I once used to simulate just a few Oort cloud comets embedded in a Galactic potential. Already the attempt to predict the dynamics based on simple Newton's law for few point masses is almost impossible, because of the high non-linearity of the system and because it depends in reality on dozens, if not hundreds of parameters. The longer I thought about how to make it more and more realistic and the more parameters to consider came to my mind. And changing only a little bit the value of one of these parameters, led to entirely different simulation results. But just for this reason it was also easy to adjust the parameters in order to obtain whatever desired results. Considering then that an Oort cloud is a relatively simple system compared to Galaxies, I find this belief in the predictive power of simulations very naive. Only if they can be described for sure by only few parameters it makes sense, otherwise not. But nowadays we are in the full AI swing and hype, and hope to simulate even an entire human brain. Frankly, I think these expectations rely on a lack of understanding of the immense complexity of the things we are trying to simulate. I'm not claiming that I understand it, but at least I know that I don't. Then, every ten years we will be told that other ten years of research are needed because "we didn't expects things to be so complicate". A vicious circle we honestly should begin to question.

  28. An absolutely great article, however I am pessimistic the sensibility of it will spread much in the scientific community. It was encouraging to learn about Stacy McGaugh, specifically others in that community are trying to make it more self-aware of issue's you've often written about.

  29. I don't understand all this focus on galaxies regarding dark matter.
    Galaxies are polluted by all the baryonic physics and they are just not the best target to address dark matter (although historically they did represent one of the first evidences).
    Robust evidence for DM is in the observations of the cosmic microwave background fluctuations and the power spectrum of large scale structures. No modified theory of gravity has up to now been able to explain these observations. DM does it. These are basic cosmology facts that modified gravity should explain before it can be taken seriously.

  30. curve fitting of parameters and simulations, with little or no predictive power ... not much different from climate scientists trying to predict what might happen in 50 or 100 years from. At least the dark matter people aren't demanding Trillions of $USD in new taxes.
    -- TomH

  31. Reminds me of a speaker we had on climate change: he showed us the history, over time, and how all the models agreed! They fit the data! Then he blushed and said "the extrapolations of each of these models into the future, differ wildly." He still believes in global warming, saying that increased CO2 is very dangerous.

  32. Excellent piece, and good to see that you're coming out more and more forcefully against the Standard Model, which has seemed to many observers (including me) to be very wanting for many years now. Interestingly, I just completed and published this draft version of a detailed interview with Stacy McGaugh. Feedback appreciated: Also any contacts you have in cosmology or astronomy-oriented online publications that may be interested in publishing this would be appreciated.

  33. The most successful recent efforts to reproduce the baryonic Tully-Fischer relation with CDM models is L.V. Sales, et al., "The low-mass end of the baryonic Tully-Fisher relation" (February 5, 2016). And it has way too many parameters. It explains:
    "[T]he literature is littered with failed attempts to reproduce the Tully-Fisher relation in a cold dark matter-dominated universe. Direct galaxy formation simulations, for example, have for many years consistently produced galaxies so massive and compact that their rotation curves were steeply declining and, generally, a poor match to observation. Even semi-analytic models, where galaxy masses and sizes can be adjusted to match observation, have had difficulty reproducing the Tully-Fisher relation, typically predicting velocities at given mass that are significantly higher than observed unless somewhat arbitrary adjustments are made to the response of the dark halo."

    The paper manages to simulate the Tully-Fisher relation only with a model that has sixteen parameters carefully "calibrated to match the observed galaxy stellar mass function and the sizes of galaxies at z = 0" and "chosen to resemble the surroundings of the Local Group of Galaxies", however, and still struggles to reproduce the one parameter fits of the MOND toy-model from three decades ago. Any data set can be described by almost any model so long as it has enough adjustable parameters.

    I've tracked both the modified gravity and dark matter particle literature for years, and you can get a compressed and select overview of it at my "dark matter" tag.

  34. So what are the ten parameters?

  35. A related rant out the very same day: Is It A Bird ? Is It A Plane ? NO, It's A Cloud !

    Rhys Taylor talks a little about the missing-satellite-galaxies problem and low surface-brightness galaxies, but mostly about the so-called satellite "planes," which don't seem to be statistically robust except around our own galaxy (and I have some doubts even there; not even Rhys has added radial error bars to his 3D models, and distances are notoriously tricky to get right for small, dim objects).

    More to the point, he did a four-part lecture series summarizing the state of the field with special attention to the available observations, the pitfalls of applying them to models, etc. It's well worth reading if you want a good, more-than-layman's understanding of, well, almost all the things you mentioned, Dr. H.

  36. It is definitely "low--surface-brightness galaxy". Definitely.

    The compound is open, or at least it was when the AAS journals had a consistent style. The en dash is a hypercorrection, and it's also a pain to set on narrow measure.

  37. Thomas,

    See table 1 in this paper. (Though, as someone pointed out above, they seem to have added a few parameters since 2013.)

  38. konoha4,

    I, in return, don't understand why you think you can ignore galaxies. I want to have an explanation for this. Saying "but dark matter fits the CMB and in galaxies magic happens" doesn't cut it.

  39. Sabine,
    From what I know most galaxy dynamics and cosmological simulations use Newtonian dynamics, especially such large simulations as those you refer to.
    So it should be OK to compare to nonrelativistic MOND, shouldn't it?

  40. opamanfred,

    I don't know and I think you don't know either. What we do know is that (a) MOND does a good job explaining some correlations and that (b) MOND is not relativistic, hence it can only be correct as a limiting case. Without knowing the full theory, however, it's not clear just in which situation the limit it is even valid.

    Take the (above mentioned) Khoury-model as example. It has a MOND-like non-relativistic limit. It also turns out, however, that this limit is only good in deep but not too deep gravitational potentials. Ie, it doesn't work on scales of galactic clusters and it doesn't work on the scale of solar systems (though for different reasons). Just from looking at the MOND you just can't tell what a completion would do.

  41. Sabine,

    I am not an expert, but I do follow a bit this literature, and I am fairly confident that most simulations of structure formation are indeed Newtonian. GR simulations exist but are very recent and rare, see the paper by Adamek ( and corresponding News and Views comment.
    From what I understand of the Illustris project (from their website) it is also newtonian, as is the well-known RAMSES code (astro-ph/0111367).

    My point was not to discuss the validity of MOND, but to say that it is consistent to compare current simulations (mostly nonrelativistic) with a nonrelativistic theory such as MOND.

    As to Khoury's model, there is no GR there. Look at the last paper you mention in your earlier post (arXiv:1711.05748), eq. 29 is Poisson's equation, and the authors say on the same page (right column) "the gravitational potential is of course determined by Poisson’s equation". So I don't understand what you mean when you say that Khoury's model "has a MOND-like non-relativistic limit". It is nonrelativistic to start with.


  42. opamanfred,

    You entirely missed my point.

    "it is consistent to compare current simulations (mostly nonrelativistic) with a nonrelativistic theory such as MOND."

    I am telling you that no one knows because MOND is not a complete theory. We know this exactly because it's not relativistic. All we know is that this non-relativistic limit works for some things in some circumstances. It does not follow from this that MOND must work in all non-relativistic GR limits. It's a non sequitur. It's logically wrong. Above I gave you an explicit example for just exactly how it can go wrong.

  43. This is a fascinating post. Despite having devoted quite a bit of time to trying to understanding the intersection between data and modified gravity/inertia theories, and visiting Stacy McGaugh's excellent website, I was still lost in the labyrinth of details. This post helps greatly by providing a birds eye view, (or I should say Bee's eye view), of the overall landscape in this area of cosmology.

  44. Sabine,
    I think there is a misunderstanding here. I have reread your post, especially the last part on MOND vs simulations, and I agree on every word of it. These simulations are technically awesome but conceptually awful. 15 parameters! You could fit my grandmother orbiting the moon with that. I also had the same experience as you that the very people who use the code are not sure how many free parameters there are.
    My only purpose was to point out that both the simulations and MOND are Newtonian physics, because I thought this was useful information. Yes, for this reason MOND is incomplete, but so are the simulations, for the same reason. I don't see any non sequitur here, but perhaps I have misunderstood you.
    Also, you did not respond to my comment that Khoury's model is nonrelativistic (non GR) from the very beginning.

  45. opamanfred,

    Yes, Khoury's model is non-relativistic! That's my point. It's non-relativistic but the MOND limit *still* doesn't apply on cluster scales. That's why I am telling you: The only thing we know is that MOND is a limit of something. We don't know in which cases it applies. Just because it's non-relativistic you can not conclude that it must apply in all non-relativistic limits of GR.

  46. Sabine,

    OK, I see better now. Just one more point about your last sentence: why do you say "all non-relativistic limitS of GR" (in the plural). The only nonrelativistic limit of GR is Newton's gravity, is it not? There are post-newtonian limits, but they are semi- not non-relativistic. And MOND is fully non-relativistic (both c-->infty and G-->0).

    For me, all this seems to point out to MOND being too crude, even as a nonrelativistiuc theory. Perhaps an unpgraded (though still nonrelativistic) version of it will manage to describe BOTH galaxies and clusters. Who knows. But of course, if we start putting 15 parameters in a modified-generalized-deformed-whatever MOND theory, then we're back to square one...

  47. Professor this is perhaps your best blog post, and the comments are great. You are so patient.
    BUT you do yourself a disservice. ..

    "I barely understood what the man was saying,[so] it was also clear why other people thought I may have something in common with him."

    You write very clearly especially for the layman ando provide a great service. For me iam all.. yes DM exists but we keep eliminating all the places it can hide so where is it?.. this leads to the same dissatisfaction that it is just a contrived explanation.

  48. Joey,

    It's a pretty dumb move to misquote me directly below the article where the quote is supposedly coming from.

  49. opamanfred,

    I think that's a semantic issue. I'd say there's a non-relativistic limit for each solution to the field equations. You seem to say instead there's one non-relativistic limit of the field equations. *shrugs* Not sure there is something to learn here.

  50. Thumb typing: as you are probably finding out, the arena of galactic structure is a gong show rife with corruption. We are trapped in the fallacy of the ecluded middle. MOND should at least be listened to and DM given the critique it demands of MOND. Both are insufficient as models agreed, but some sobriety and maturity in the discussion would help a great deal.

  51. Chemistry operates without deep knowledge. contains an extraordinary NMR spectrum simulator. I broke it with a small, rigid, rotationally symmetric molecule, C_12H_14. Chemaxion found it could not be simulated. Its furious simplicity disallowed approximations.

    Non-classical gravitation, SUSY, and dark matter theories suffer unending parameterizations of a kind, and therefore diagnostic. Feed them their own lethally simplified selves in a one day observation. An emergent symmetry’s divergence must be maximized.

    Pure math has an n-dimensional solution. Make that two experiments, one orthogonal to that solution lest a spider be caught in its own web. Never settle for a lesser evil. Interesting problem…

  52. So you are saying that the parameters written into the simulation are not complete, so we are getting a inaccurate demonstration of the simulation?

  53. Hi Sabine, Has anyone ever considered that the galaxies, or space-time engulfing the galaxies, could be rotating like a solid or semi-solid and the stars and matter are resisting the drag of the space-time rotation.

  54. Michael,

    Well, you explain galaxy formation based on that idea and we can talk again.

  55. Plato,

    Was that question addressed to me? No, that's not what I am saying. The simulations are fine. But they are designed to fit what we observe after we have observed it, when necessary by adding more parameters. They're not very predictive.

  56. This comment section is getting a lot of comments in which people propose their own theory for dark matter. Please note that I do not approve such comments. You are wasting your time submitting them.

  57. I must also comment that just as MOND phenomenology could emerge from a much more sophisticated theory, so could LCDM (as a cosmological model) be the "limit" of some underling theory.

    I think that a more precise way of stating the strength of LCDM is that in each and every regime it is a predictive paradigm, it did not fail. The predictions on a sub-cluster scale are still premature and should be taken less seriously in my opinion.
    The thing is that MOND, as a paradigm, is in a very similar situation, it is very successful on a galactic scale and though many tried, they all failed killing in that scale. On other scales however and, as you said, in the absence of a full theory it is hard to say much about the validity if MOND as a paradigm.

    The most frustrating part is that both paradigms are successful in a quite complementary region and are hard to compare in a model independent way. I wish I'll be able to address this point in the near future.

  58. >>But they are designed to fit what we observe after we have observed it, when necessary by adding more parameters.<<

    Is that really the case?
    The cosmological simulations seem to me like a big forging hammer but increasingly demanded to do ever finer chiselled work. So the hammer is always refined to produce even the most detailed things. And reality is probably even more complex than that.

    Perhaps one should better ask the question which statements can really be expected from numerical simulations. In particular I would be careful to basically condemn the model to be mapped in the simulations just because the simulations can not map observation details (for example "CDM is wrong because we don't see the satellites/cusps/alignments in the simulation"). The parameters that are included in the simulations do not appear from nowhere but are justified by astrophysics (and are included as soon as it is technically possible).

    >>They're not very predictive.<<

    Is that a feature you expect from numerical simulations? Predictive power I would expect from physical theories rather from numerical simulations that are fed with physical theories AND derived models AND empirical parameters. For me, numerical simulations already serve their purpose when they draw my attention to phenomena that I did not think of before. If the predict something testable - great!

  59. Michael,

    The question isn't what I would expect or not, the question is how do we make progress. We can spend the next 30 years fitting computer simulations of increasingly contrived interacting particle hidden sectors that somehow no one manages to detect. Or we can do what scientists have been doing for centuries, formulating laws based on observed patterns and try to develop consistent theories for what we observe.

  60. It screams “epicycles” directly into your face.

    No, it doesn't, not if you actually understand what's involved in trying to model complex, real-world phenomena, and not if you understand what "epicycle" means.

    If you want a good example of a genuine epicycle, then that's exactly what MOND is: a physically unmotivated complication of an existing model (in this case, Newtonian dynamics) added to solely to explain one set or class of observations.

    Of course, you could justifiably make the same observation about dark matter: it's an epicycle, too. (Although the history of astronomy has shown that adding an extra source of mass to explain mysterious dynamical phenomena -- Neptune; white dwarfs; hot X-ray emitting gas in clusters -- can at least sometimes turn out to be justified.)

    And sometimes that is a useful way to go. But that's different from attempts to move beyond the unrealistic simplicity of early simulations ("assume a spherical horse") by including effects, or more accurate treatments of effects, which we know actually exist.

    And of course the vast majority of complications added to simulations of galaxy formation in dark-matter-based studies will be necessary -- and indeed probably useful -- for any other attempts to explain galaxy formation using, e.g., alternate theories of gravity. Stellar winds, ionizing radiation, supernovae, active galactic nuclei, and the hydrodynamics of gas all exist, and will not magically go away just because your new theory of gravity does away with the need for dark matter.

  61. Peter,

    I use the word epicycle because what we are seeing here are increasingly more amendments to an underlying model which is kept unaltered, in this case the hypothesis of particle dark matter, in the attempt to make data fit that clearly doesn't want to fit.

    The explanation due to modified gravity on the other hand has remained virtually unchanged since the early 1983 and it still fits on the present data, while older dark matter simulations don't.

    I never questioned the presence of the astrophysical phenomena that the simulations try to capture. If you think so, you entirely misunderstood what I wrote and you should read it again. What I said was that it seems exceedingly implausible to me all these twiddles and thumbs will one day all magically add up to give rise to a MOND-like relation.

  62. Simulations have rules. Observe reality, then debate bellybuttons.

    Observed vacuum is exactly mirror symmetric toward massless boson photons over billion lightyear paths. Observed vacuum is ~10^(-9) mirror asymmetric toward hadronic matter: baryogenesis, proton-antiproton magnetic moments. Copious chiral and parity parameterizations include Chern-Simons repair of Einstein-Hilbert action. Electroweak consequences are left-handed toward matter.

    A high resolution microwave cryogenic rotational spectrum of large dipole moment enantiomer molecules will show ppb line broadening for being left-handed vacuum-embedment non-identical. Exceed line resolution by 3:1 enantiomer ratio giving 3:1 lopsided line broadening. Chiral bubble D_3-trishomonocubane-2-carbonitrile is one extreme example. Spiky bicyclo[2.2.1]heptane-endo-2,5-exo-3,6-tetracarbonitrile is another. Look.

  63. Is the WIMP miracle still viable or has Xenon ruled it out?

    has there been research in both MOND and dark matter as both valid but on different scales? MOND for galaxy scale, dark matter for large scale structure

  64. I think this reference is relevant:

    Andromeda Dwarfs in Light of MOND. II. Testing Prior Predictions
    Stacy McGaugh, Moti Milgrom

  65. Where does the evidence from BAOs in the CMB fit into this? I thought that these clearly pointed to dark matter. Can some form of MOND explain these?

  66. I was at a conference last decade where some talks on galaxy formation and /\-CDM were given. These guys talked stuff I had not a clue about.

    The MOND idea is simply ugly. I see it as similar to the attempts to write Newtonian gravity force as F = -GMm/r^x, for x something like x = 2.001. This was an attempt to account for the precession in the orbit of Mercury. The idea is ugly, but it was a sign that maybe something fundamental was wrong with classical physics. Maybe the same is the case here. This might mean Verlinde is in some ways on the right track.

    Of course we have to be careful. This stuff involves a lot of complexity. It would be as if we were looking at the properties of DNA in a cell to try to ferret out quantum mechanics. It might not be the best arena for looking at physical foundations.

  67. Somewhat OT: recently the long-running Galaxy Zoo citizen science project served up “galaxies” from Illustris (blog post about it:

    While I am far from the super-zooite many of my colleagues are, in terms the number of galaxies classified (Galaxy Zoo is principally a morphology classification study), I’ve done enough to instantly recognize that ~none of the Illustris sims are realistic. And this is in the optical; Illustris would surely be utterly hopeless if viewed in the x-ray, midIR, microwave, or radio.

    Part of why is due to Illustris ignoring dust; true, dust is a truly minor mass component of almost all galaxies, but is it also minor in its role in galaxy evolution? No one knows.

    Regarding sims in general: standard solar models are remarkably faithful to real stars. Has anyone tried to create a sim of a tropical rainforest, say, starting from the pre-Cambrian?

  68. Waterbergs,

    Well, John Moffatt says he can explain everything. You may or may not believe that. Personally I think it's pretty irrelevant because saying that particle dark matter explains BAO doesn't negate that MOND explains Tully-Fisher.

  69. "Where does the evidence from BAOs in the CMB fit into this? I thought that these clearly pointed to dark matter. Can some form of MOND explain these?"

    Even most MOND enthusiasts agree that dark matter is needed in a cosmological context. The fact that MOND can explain some things better than dark matter doesn't mean that it does away with the need for dark matter entirely. On the other hand, dark matter can explain MOND phenomenology, but only in a very contrived way.

    So, we are sure that most of the matter in the universe is some sort of unknown non-baryonic matter. It does not follow, however, that there is no need for a non--dark-matter explanation for MOND phenomenology.

  70. Well, there is a point I havent seen here in the discussion, so let me just go forward with it. The problem is with making MOND a field theory, both relativistic and non-relativistic.

    Already at the non-relativistic level, it is difficult to keep MOND-based theories stable. Consider for instance a star moving in a deep-MOND field at the outskirts of the galaxy as compared to a completely isolated star. The typical MOND-based field theory will make these two stars very different, because the gravitational field passes from the weak external (MOND-regime) field to moderate Newtonian (non-MOND) field at around the surface of the star, and to the weak (MOND) field in the center of the star again. As a result, at least one of the two stars will actually be ripped apart by the MOND non-linearity.

    In relativity, a few more issues arise. We know from the works of Karel Kuchař and others that if you have a metric gravity with a good Cauchy problem and the degrees of freedom match Einstein gravity in number, then you are necessarily talking *only* about Einstein gravity. If you modify gravity while maintaining a good Cauchy problem (some call it 'no-ghost' gravity), you are *necessarily* adding new degrees of freedom to your theory. I.e., if you are modifying gravity, you are adding something like dark matter anyway, you are just calling it with a different name and coupling it to the other fields in some pretty weird way.

    With metric gravity having so many direct and well-controlled confirmations, the arguments of the last paragraph are ironclad. If you do not want to modify gravity, you get simple and natural dark matter, if you want to modify gravity, you get weird and complicated dark matter. There is really no way out of that. And even though the weird and complicated sometimes proves to be correct, I do understand that the community first goes with the simple and natural and tries to explore whether enough sophistication and including known, well-explored effects can reproduce the observations.

  71. Sabine and Philip

    Thanks for the clarification there.

    So, the BAO data implies a certain amount of dark matter - everyone seems fairly certain of that. Can the effects of this dark matter then be calculated on cosmological scales in the most "vanila", non-contrived, way and then can one see what might be left to be explained by MOND? Would such a subtraction leave the required MOND theory itself looking reasonable, or would it then seem rather contrived to fit the left over elements not explained by the BAO required amount of dark matter?

  72. @Jean Tate "Has anyone tried to create a sim of a tropical rainforest, say, starting from the pre-Cambrian??

    European weather models trounce American models. European predictions are then run backward. If initial states do not appear, the model is refined.

    Amazon rains hugely wash out minerals. Exposed soil laterizes. Major winds blow Saharan lake sediment dust across the Atlantic into Amazonia, providing a trace phosphate lifeline. Local civilizations large scale created "terra preta" ("Indian black earth") that retains prodigious fertility. They hugely propagated “useful” trees.

    Modeling Amazon creation from scratch using cosmological cherry picking methods will fail. Begin with now and run it backwards. "Uff da/Usch då!"

  73. Just an ad-hoc idea:
    What would it mean for DM, when the MOND-like behavior of observations at galactic scales is correct? Is it possible to assume/describe this as some attribute or feature of DM, even if some 'strange' assumptions have to be made??
    If this is an impossible feature (not only strange) – bad for DM.
    If it is possible to describe a DM characteristic this way and derive a not impossible feature, perhaps it helps.

  74. @Phillip Helbig

    "So, we are sure that most of the matter in the universe is some sort of unknown non-baryonic matter. It does not follow, however, that there is no need for a non--dark-matter explanation for MOND phenomenology."

    Milgrom McGaugh et al, points out how much "dark matter" is needed is reduced by 5x, and it could be ordinary baryonic matter or neutrinos with mass on order of keV


    there is more recent literature from Verlinde Smolin et al, that suggests MOND is a QG effect, based on the observation that MOND scale of acceleration is very close, on the same order as cc

  75. "It's a pretty dumb move to misquote me directly below the article where the quote is supposedly coming from."

    It was just a joke. I put the fake part in brackets. I probably should have put in a smiley face since it was such a dumb joke.

    I am a big fan...

  76. Joey,

    Ok, then. Sorry for the misunderstanding.

  77. Simulation code that is used for scientific publications ought to be open-source, and available e.g., via

  78. Perhaps one should shed light on the question of what the purpose of numerical simulations is or can be. For me, it's a tool I can use to check if I understood a phenomenon correctly. By that I mean understanding the problem as well as understanding possible solutions to the problem. And I think that this view can certainly contribute to scientific progress.
    In numerical simulations of cosmic structure formation, I introduce fairly simple principles (LambdaCDM) and get results that are surprisingly close to what I can observe - at least on larger scales. In detail it is indeed dirty. For example, I get more substructure than is observed. The simulation that delivers this result is still very simple. So the idea of putting more physics into the simulations to solve this discrepancy is obvious.
    As far as I know there is still no full-fledged numerical simulation (analogous to LCDM) of cosmic structure that has implemented the MOND principle. Here is one approach:

    In the end it does not matter to me if something MOND-like or W-C-DM-like model explains the observations, I'm agnostic about it. However, I am convinced that numerical simulations can help us to figure this out.

  79. Re numerical simulations and block boxen: y'all physicists ever heard of open source?

  80. Have the LIGO detections appreciably changed the motivation for particle DM (or for MOND-like models)?

  81. Latest results show no signs of the elusive WIMP

  82. Thanks Sabine for this interesting post and blog overall. I think I found a small error:

    "The do exist though, they are just difficult to see. And they spelled trouble for dark matter, just that no one wants to admit it."

    I think it should be "They do exist though, ..."

  83. Felix,

    Thanks for pointing out, I've fixed that.

  84. I second Michael Bertschik's comments. The state of the art for these simulations seems to be in an early stage where everyone is hoping to see improvement. Until the improvements happen, researchers will want to see what is the result of their (over)simplified models; what happens when I rerun with somewhat different initial conditions or a tweaked model? Do the changes in results make sense and does it tell me anything useful about the models?

    Apology if this is off topic, but it reminds me of the early days of computational fluid dynamics compared to today. In the late 1980's CFD and the best computers could manage a model with the fluid domain divided into 100K cells or nodes. A user had to choose among a dozen empirically derived turbulence models (parameters). The solvers were inefficient and prone to diverging. Only an expert user could nurse a solution to completion. Much time was spent checking the results to see if they made sense. Still there was value. One could compare how design changes caused behavior to change -- even if neither was quite physically accurate. One could determine the most sensitive parameters; one could locate turbulent eddies and design them away. Then, CFD was seldom an integral part of product's design cycle, but it had some promise.

    Today one can use a billion or more cells to model a fluid domain. CFD is accessible to many engineers not just the experts. The results are amazingly accurate and CFD is an integral part of product design. Solvers are better and almost self correcting. Still have several turbulence models, though (Look up the Lamb/Heisenberg quotes.)

    With support these cosmological evolution codes should also see regular improvements.

  85. 1) Never condemn a million pages of empirically sterile nonclassical gravitation, SUSY, and dark matter. That is Popper shaming!
    2) Never condemn a million pages of, etc. That is business model insubordination!
    3) Never pursue falsifying observations. Validating observations are accepted theory!
    4) 10× detector size is 1000× detector volume. Another three decimal places will do it!

    Fifty years of non-physical physical theory is not failed ability or failed dedication. It is disjointed perspective. Administrative value is about administration not value.

  86. Can MOND handle collisions such as the Bullet Cluster? And do we have proof there was Dark Matter, ten billions years ago? If the answer is no to both, a sort of drastic remedy, similar to the remedy for the epicycle remedy may be in order, "MOND" tries to modify gravity (at cosmological distance). Why not modify Quantum mechanics, at cosmological distances, instead?

    That would imply a mechanism where “Dark Matter” is emergent, and not really matter.

    Epicycle theory was partly a consequence of a non material Zeitgeist caused by the obviously false Aristotelian physics. When Buridan, around 1350 CE, got the physics right with his “IMPETUS” theory (first and second laws of mechanics combined), the heliocentric theory came right out (Buridan was put at the Index 120 years later, but Copernic learned his science, including heliocentrism, at the university in Cracow, it was mandatory, then and there, as Eastern Europe hated the Vatican, because of the treacherous burning alive of professor, dean and rector Jan Hus!)

    We may have a similar situation now: it is NOT gravity which is wrong (wit its connection with inertia, thus Buridan. Disjointing gravity from inertia, as MOND does, is a bridge too far!) Instead of accusing gravity-inertia, as MOND does, one should rather guess that Quantum Mechanics is, slightly, wrong (at cosmological scales). Indeed Quantum Mechanics, Copenhagen Interpretation, assumes an extraneous element, collapse, just as Aristotelian physics assumed an extraneous element, namely a force to create inertia…

  87. There is something that bugs me about dark matter. And it is the question "how are galaxy mergers at all possible"? According to conventional wisdom galaxies are large, roughly spherical blobs of dark matter, that contain a disc of gas and stars. Assume two of those collide: the gas from both galaxies collides, so you get a blob of hot gas - but the stars mostly don't collide, because they are small, and the dark matter blobs don't collide at all, because dark matter does not interact with itself. So the DM blobs should just pass each other and happily continue their motion they had before the collision. And, as they contain most of the mass of the system, the majority of stars should stay bound to their DM blobs and continue together with them. So, the result of a collision should usually be two galaxies, largely devoid of gas and thus not creating any new stars, plus a blob of gas between them. Probably expanding gas, because it would be hot after the collision and the main sources of gravity that previously kept that gas bound in the galaxies are now speeding away from the collision point. So how comes we have large numbers of elliptical galaxies, that are results of galaxy mergers?

  88. Jerzy,

    The kind of high-speed head-on collisions that you have in mind are extremely rare, basically for the same reason that collisions of stars within the galaxies are rare (as you note). Much more common are mergers of early galaxies that just get too close to each other and gravity does the rest. I don't know much about the statistics in this case, maybe someone else has a good reference, but I don't see the concern that you see.

    What is a concern though is that it seems whatever the galaxies have done in their past, the fraction of dark matter that they end up with is the same and only dictated by the amount of baryons. At least that's what you have to believe if you don't like modified gravity.

  89. Bee

    perhaps this should be a new blog post, but since lamda cdm model predicts 5x as much dark matter as baryon, i understand our solar system and other solar systems may not be represenative sampling, but on average, shouldn't there be at least as much dark matter in our solar system if not 5x as much dark matter as visible matter in our solar system?

    and wouldn't all the dark matter in our solar system, and other solar systems, change the orbits of planets in disagreement from prediction?

    esp orbits of the most far out objects of our solar system like pluto and Kuiper belt comets?

    essentially our solar system is like a minutiae version of the galaxy, and the sum total of all the dark matter in our solar system should have an effect much like what was observed in galaxies.

    i understand that the dark matter is spread very thinly and may be difficult to detect, but the some total all dark matter plus the sun and visible planets over the entire solar system should have some effect on the orbits of the most outer objects of the solar system

    to put it another way, can the orbits of the farthest objects of our solar system of pluto and kuiper belt be explained assuming GR and Newton solely in terms of visible matter, sun and planets, or does it also require dark matter?

  90. "So the DM blobs should just pass each other and happily continue their motion they had before the collision. And, as they contain most of the mass of the system, the majority of stars should stay bound to their DM blobs and continue together with them."

    The problem I see with this statement is, how do the stars of the respective galaxies "know" which blob of DM to be bound to?

    That is, the stars should be attracted by gravity to both blobs, and the blobs should be attracted to each other. This attraction will change their momentum - they may continue on their way, but not as happily; or not at all, depending on their initial momentums or after several collisions.

    The Bullet Cluster seemed to be an example of DM + LM blobs colliding, with gravitational lensing indicating a separate DM blob - although as Dr. Hossenfelder has reported in previous posts, some modified gravity system has also been able to model that effect.

    I of course have no idea which hypothesis is closer to reality, but retreating back to (modified) Newtonian Dynamics seems almost crazy, despite the sheer magnitude of the coincidence that it seems to model so much of galactic dynamics so well.

    When things like this happen in one of my computer programs, it makes me work much harder and longer, feeling that there must be an answer that makes sense and I desperately want to know what it is. Of course learning by trial and error is much easier with artificial computer problems. (And yes, sometimes I dismiss problems as flukes rather than flaws until I get more information.)

  91. I'm still confused: does Dr. B believe in modified gravity or not?

  92. I'm confused: does Dr. B still believe in general relativity or not?

    We have nothing new for reason to throw away the main principles of GR.

  93. Palladium couplings revolutionized organic synthesis. Discovery of like iron catalysts was amazing, but inert in new equipment. Thus contemporary physics...right up to “say what?” Physics parameterizes.

    Chemistry looks. Iron “catalysts” mobilized ppm Pd adsorbed to used Pyrex and Teflon.
    ... The gorilla isn't the fun part. Look.

  94. We know something is wrong.

    Dark matter theories assume that the problem is with the Standard Model. We understand GR and gravitation, but there are more particles out there than we have detected so far.

    The alternative theories argue that the Standard Model may or may not be complete, but that we don't understand GR or gravitation as well as we think. The latter has been less fashionable since we've been so successful at finding new particles to add to the Standard Model for so long.

    Now that the LHC has further confirmed the Standard Model, it is time for us to move down the block and search under a different street lamp.


COMMENTS ON THIS BLOG ARE PERMANENTLY CLOSED. You can join the discussion on Patreon.

Note: Only a member of this blog may post a comment.