This week’s installment covers a new Bayesian analysis code (particularly designed for cosmological studies!) called Cobaya, a gravitational wave constraint on the number π (yes, you’ve read correctly) as a null test of General Relativity, and a new proposal for holographic dark energy based on John Barrow’s recent COVID-19-inspired proposal for fractal structure on a black hole event horizon. Enjoy!
#1 2005.05290: Cobaya: Code for Bayesian Analysis of hierarchical physical models
This piece constitutes somewhat of an exception to my earlier reviews, in that what I found most interesting is not the paper itself, but the code therein described. There is by now a wealth of cosmological observations of exquisite precision, including CMB, large-scale structure, weak lensing, BBN, Supernovae measurements just to mention a few. At the same time, new cosmological models are being proposed every day (just open the astro-ph.CO section on arXiv), and to check whether these theories makes sense entails constraining them against the aforementioned cosmological data. This is usually done within a Bayesian probabilistic framework, where the question of interest is: “What are the posterior probability distributions for the parameters of interest, given the data I have analyzed?”. In practice, to answer this question one usually makes use of Markov Chain Monte Carlo (MCMC) methods to sample the relevant parameter space.
There are many cosmological MCMC samplers out there, but it is fair to say that two dominate the market: CosmoMC (written in Fortran and interfaced with the Boltzmann solver CAMB, also written in Fortran) and MontePython (written in Python and interfaced with the Boltzmann solver CLASS, written in C). I’ve personally used both, although I lean considerably more towards CosmoMC for academic genealogy reasons. Each comes with its strengths and weaknesses, and there is so much to say on the matter that it could easily constitute a post on its own. However, having worked hands-on on both, I can say that adding new likelihoods (i.e. the bit of code necessary to analyze a new dataset), priors, theory computations, sampling methods, and so on, can be a huge pain in the a**. As the amount of data and theory complexity are both increasing considerably, it is important that these codes be as flexible as possible. Cobaya is a newly-developed Bayesian analysis code which goes towards the direction I flagged above. The name comes from the Spanish for “Guinea pig”, though I am not sure if there is a deeper connection between this animal and the code itself! Cobaya has already been around for some time (at least an year or more) and it is very likely it will, going forward in time, gradually replace both CosmoMC (whose main developer is Antony Lewis, also one of Cobaya’s developers) and MontePython.
Cobaya is more flexible than CosmoMC and MontePython, especially in terms of modularity, rapid prototyping, and supporting external components (see more details in Section III, and a simplified picture of Cobaya’s source and main classes in Figure 1). In plain English, this means that it is much easier to interface new theory or likelihood codes, and the amount of work required for external collaborations to release their code packages in a format which is directly usable for Cobaya is close to zero. Being written in Python, there is in principle a slight computational overhead, which one can in principle compensate for by writing more computationally expensive chunks of code in NUMBA/Cython or C/Fortran which then get interfaced to Python, but overall the runtime is comparable to CosmoMC. A case where Cobaya would outperform CosmoMC would be either if there are many primordial power spectrum parameters, or if one wants to reconstruct the latter (e.g. through piece-wise splines), as done in a huge number of works including the recent 1711.08441 1908.00906 (I literally quoted the first two papers that came to my mind, apologies for all those I have missed!). There is still more work to do to improve Cobaya (as discussed e.g. at the end of Section VI), but overall I applaud the amount of solid work which went into developing this code. It is also worth noting that Cobaya can be used as a stand-alone generic sampler with no connection to cosmology. I definitely plan on learning to use Cobaya sooner rather than later, and this pandemic might well be providing me the right occasion to do so!
#2 2005.05472: Pi from the sky -- A null test of general relativity from a population of gravitational wave observations by Carl-Johan Haster
Observations of gravitational waves (GWs) from compact binary coalescences (CBCs), e.g. binary black hole (BBH) or binary neutron star (BNS) mergers, have allowed us to test General Relativity (GR) in the strong-field regime to an unprecedented level. So far GR appears to be completely consistent with observations, although the tests of GR performed so far have generically been consistency tests, mostly introducing ad-hoc modifications to the GR waveforms, for instance in the form of model-agnostic deviations from the usual Post-Newtonian (PN) expansion coefficients. While possible deviations from GR can then be mapped to specific beyond-GR theories of interest, a flaw of this method is that it does not account for the fact that deviations from the GR-predicted PN expansion coefficients values should depend on the CBC source properties themselves. Moreover, this method has been successful in testing each PN order separately, but provides little information when more than one PN term is allowed to vary. These flaws are the starting point of this week’s paper, which seeks to provide a more robust null test of GR. In particular, Haster envisages treating the number π in the PN expansion as a variable, constraining it in a Bayesian sense (i.e. determining its posterior distribution), and checking whether the obtained posterior is consistent with the value of π=3.1415926535… we know and love. In practice, in the usual PN expansion of the GW waveform phase, π and multiples thereof appear in various expansion coefficients, see the discussion around Eq.(2) in this week’s paper, and Haster’s proposal is to treat π as a free parameter to be constrained (although it is the same parameter for all PN expansion orders). Haster then takes 13 CBC observations from the LIGO/Virgo O1, O2, and O3 observation runs, namely GW150914, GW151012, GW151226, GW170104, GW170608, GW170729, GW170809, GW170814, GW170817, GW170818, GW170823, GW190412, and GW190425.
Combining all these observations, Haster derives a constraint of 3.022<π<3.162 at 90% CL, which is perfectly consistent with the value we know and love, albeit slightly skewed towards lower values (but at a level which is not even close to being statistically significant). Extending the observations to include other 9 events leads to an even better constraint on π which provides an extremely strong preference for GR (with a Bayes factor of 321). Figure 2 is a very nice representation of how the posterior of π converges towards the final distribution as more and more GW events are included. This is a strong consistency test of GR, in particular of 4 out of the 8 included PN orders, and is particularly useful in testing so-called GW tail effects, non-linear effects due to the backscattering of linear waves onto space-time curvature generated by the source mass. Haster then tests the reliability of this method against false-positives, a test which the method passes, and then shows for a specific case related to the presence of a massive graviton how the test itself could reveal the presence of non-GR effects (although it cannot guide us as to what specific theory is behind these effects) should one recover a value of π other than the one we know and love. This was a very interesting paper which reminded me somewhat of the paper I discussed in my Week 17 post treating 8 different instances of the physical baryon density in the Boltzmann solver CAMB as being independent (although, I remark once more, the value of π at each relevant PN order here is treated as one and the same variable). The only tiny concern I might have is that, looking at Fig. 1, the individual posteriors on π from each GW event appear in some cases to be mildly inconsistent with each other, with some of them appearing to be multi-modal, and one of them appearing to be cut off by the prior on π. If these posteriors were truly inconsistent, it is unclear whether the overall combination of all GW events is sensible to begin with (this is an issue which is being hotly debated in cosmology, with people such as Will Handley or Marco Raveri - just to mention two - constantly thinking of ways to quantify concordance or tension between correlated and uncorrelated datasets, see e.g. 1806.04649 or 1902.04029). But GWs aren’t exactly my field, so it might be that this potential issue is not a concern at all!
#3 2005.04115: Barrow holographic dark energy by Manos Saridakis
The idea of holographic dark energy (HDE) is an interesting proposal to explain cosmic acceleration, and stems from applying the holographic principle to the entire Universe. The basic idea is that, in a given region, the largest length of a consistent QFT (which determines its infrared - IR - cutoff) should be related to the smallest length thereof (which determines its ultraviolatet - UV - cutoff). The relation requires that the maximum energy in a region whose size is determined by the IR cutoff should be less than or equal to the mass of a black hole (BH) with Schwarzschild radius of the same size. Saturating this inequality gives an effective dark energy-like component, whose energy density scales as L^-2, with L the length associated to the IR cutoff, and the standard choice associates L to the future event horizon of the Universe. HDE was first proposed in hep-th/9803132 as a possible way to reconcile the breakdown of QFT in large volumes as required by the Bekenstein bound with the success of local QFT in describing particle phenomenology, was studied in more detail later in hep-th/0403127, and a recent review is 1612.00345.
The usual HDE argument relies on the standard Bekenstein-Hawking entropy (BHE) wherein the entropy S of a BH is proportional to its area. However, many other extended entropy proposals have been made in the literature. One new proposal was recently made by John Barrow in 2004.09444, inspired by images of the COVID-19 virus, a son-of-a-b**** often pictured as a sphere with many attachments growing off its surface to increase its surface area and provide links to connect to our cells. Barrow wondered what would happen if quantum gravity effects (and possibly space-time foam, which I discussed in my Week 4 post) would introduce complicated fractal structure for the horizon geometry, in the form of a 3D analogue of the Koch snowflake. The result is a modification to the usual BHE from the standard S going as A to S going as A^(1+Delta), with the exponent Delta being between 0 and 1, with Delta=0 corresponding to smooth spacetime and Delta=1 to the largest deviation from the BHE, and correspondingly the maximal amount of fractalized horizon structure for the resulting “rough black hole”. We will call this form of the entropy “Barrow entropy”.
In this week’s paper, Saridakis asks a simple question: “What if one applies the usual HDE arguments using Barrow’s entropy instead?”, with the rest of this short paper devoted to a first exploration of the resulting “holographic Barrow dark energy” (HBDE). Although the generalization from HDE to HBDE is in principle rather straightforward, there are a few subtle differences between the two. For instance, unlike the case of HDE, in HBDE one cannot get a closed form expression for the dark energy density parameter as a function of time, or for the dark energy equation of state. Of course this is not a problem since both can be determined numerically, as Saridakis does. Saridakis finds that HBDE, unsurprisingly, possesses more freedom in terms of cosmological dynamics than HDE. The interesting thing is perhaps that for Delta=1, i.e. BHs with a maximal amount of fractalized horizon structure, HBDE returns a constant energy density for the dark energy component. This is the standard behavior of the cosmological constant in LCDM, only this time the effective cosmological constant emerges from a completely different picture compared to the standard case. Of course, the same shortcomings from which Barrow’s new entropy proposal suffer apply to Saridakis’ paper as well. The main issue, as Barrow himself points out, is that he does not “create the fractal substructures by any single quantum gravity model”. In other words there is (yet) no convincing quantum gravity model underlying the specific Koch snowflake-like BH horizon structure which gives rise to the corresponding form of Barrow entropy. Should this gap be filled, Saridakis’ paper might be shedding light on very intriguing connections between the horizon structure of BHs and dark energy. Or, to put it differently, connections between the structure of space-time on the smallest possible scales and the mechanism responsible for cosmic acceleration on the largest possible scales.