In performing a Bayesian analysis, two difficult problems often emerge. First, in estimating the parameters of some model for the data, the resulting posterior distribution may be multi-modal or exhibit pronounced (curving) degeneracies. Secondly, in
selecting between a set of competing models, calculation of the Bayesian evidence for each model is computationally expensive using existing methods such as thermodynamic integration. Nested Sampling is a Monte Carlo method targeted at the efficient calculation of the evidence, but also produces posterior inferences as a by-product and therefore provides means to carry out parameter estimation as well as model selection. The main challenge in implementing Nested Sampling is to sample from a constrained probability distribution. One possible solution to this problem is provided by the Galilean Monte Carlo (GMC) algorithm. We show results of applying Nested Sampling with GMC to some problems which have proven very difficult for standard Markov Chain Monte Carlo (MCMC) and down-hill methods, due to the presence of large number of local minima and/or pronounced (curving) degeneracies between the parameters. We also discuss the use of Nested Sampling with GMC in Bayesian object detection problems, which are inherently multi-modal and require the evaluation of Bayesian evidence for distinguishing between true and spurious detections.
We present the first public release of our Bayesian inference tool, Bayes-X, for the analysis of X-ray observations of galaxy clusters. We illustrate the use of Bayes-X by analysing a set of four simulated clusters at z=0.2-0.9 as they would be obser
ved by a Chandra-like X-ray observatory. In both the simulations and the analysis pipeline we assume that the dark matter density follows a spherically-symmetric Navarro, Frenk and White (NFW) profile and that the gas pressure is described by a generalised NFW (GNFW) profile. We then perform four sets of analyses. By numerically exploring the joint probability distribution of the cluster parameters given simulated Chandra-like data, we show that the model and analysis technique can robustly return the simulated cluster input quantities, constrain the cluster physical parameters and reveal the degeneracies among the model parameters and cluster physical parameters. We then analyse Chandra data on the nearby cluster, A262, and derive the cluster physical profiles. To illustrate the performance of the Bayesian model selection, we also carried out analyses assuming an Einasto profile for the matter density and calculated the Bayes factor. The results of the model selection analyses for the simulated data favour the NFW model as expected. However, we find that the Einasto profile is preferred in the analysis of A262. The Bayes-X software, which is implemented in Fortran 90, is available at http://www.mrao.cam.ac.uk/facilities/software/bayesx/.
Bayesian inference involves two main computational challenges. First, in estimating the parameters of some model for the data, the posterior distribution may well be highly multi-modal: a regime in which the convergence to stationarity of traditional
Markov Chain Monte Carlo (MCMC) techniques becomes incredibly slow. Second, in selecting between a set of competing models the necessary estimation of the Bayesian evidence for each is, by definition, a (possibly high-dimensional) integration over the entire parameter space; again this can be a daunting computational task, although new Monte Carlo (MC) integration algorithms offer solutions of ever increasing efficiency. Nested sampling (NS) is one such contemporary MC strategy targeted at calculation of the Bayesian evidence, but which also enables posterior inference as a by-product, thereby allowing simultaneous parameter estimation and model selection. The widely-used MultiNest algorithm presents a particularly efficient implementation of the NS technique for multi-modal posteriors. In this paper we discuss importance nested sampling (INS), an alternative summation of the MultiNest draws, which can calculate the Bayesian evidence at up to an order of magnitude higher accuracy than `vanilla NS with no change in the way MultiNest explores the parameter space. This is accomplished by treating as a (pseudo-)importance sample the totality of points collected by MultiNest, including those previously discarded under the constrained likelihood sampling of the NS algorithm. We apply this technique to several challenging test problems and compare the accuracy of Bayesian evidences obtained with INS against those from vanilla NS.
Weak gravitational lensing studies of galaxy clusters often assume a spherical cluster model to simplify the analysis, but some recent studies have suggested this simplifying assumption may result in large biases in estimated cluster masses and conce
ntration values, since clusters are expected to exhibit triaxiality. Several such analyses have, however, quoted expressions for the spatial derivatives of the lensing potential in triaxial models, which are open to misinterpretation. In this paper, we give a clear description of weak lensing by triaxial NFW galaxy clusters and also present an efficient and robust method to model these clusters and obtain parameter estimates. By considering four highly triaxial NFW galaxy clusters, we re-examine the impact of simplifying spherical assumptions and found that while the concentration estimates are largely unbiased except in one of our traixial NFW simulated clusters, for which the concentration is only slightly biased, the masses are significantly biased, by up to 40%, for all the clusters we analysed. Moreover, we find that such assumptions can lead to the erroneous conclusion that some substructure is present in the galaxy clusters or, even worse, that multiple galaxy clusters are present in the field. Our cluster fitting method also allows one to answer the question of whether a given cluster exhibits triaxiality or a simple spherical model is good enough.
We present results of a Bayesian analysis of radial velocity (RV) data for the star HIP 5158, confirming the presence of two companions and also constraining their orbital parameters. Assuming Keplerian orbits, the two-companion model is found to be
e^{48} times more probable than the one-planet model, although the orbital parameters of the second companion are only weakly constrained. The derived orbital periods are 345.6 +/- 2.0 d and 9017.8 +/- 3180.7 d respectively, and the corresponding eccentricities are 0.54 +/- 0.04 and 0.14 +/- 0.10. The limits on planetary mass (m sin i) and semimajor axis are (1.44 +/- 0.14 M_{J}, 0.89 +/- 0.01 AU) and (15.04 +/- 10.55 M_{J}, 7.70 +/- 1.88 AU) respectively. Owing to large uncertainty on the mass of the second companion, we are unable to determine whether it is a planet or a brown dwarf. The remaining `noise (stellar jitter) unaccounted for by the model is 2.28 +/- 0.31 m/s. We also analysed a three-companion model, but found it to be e^{8} times less probable than the two-companion model.
Stellar radial velocity (RV) measurements have proven to be a very successful method for detecting extrasolar planets. Analysing RV data to determine the parameters of the extrasolar planets is a significant statistical challenge owing to the presenc
e of multiple planets and various degeneracies between orbital parameters. Determining the number of planets favoured by the observed data is an even more difficult task. Bayesian model selection provides a mathematically rigorous solution to this problem by calculating marginal posterior probabilities of models with different number of planets, but the use of this method in extrasolar planetary searches has been hampered by the computational cost of the evaluating Bayesian evidence. Nonetheless, Bayesian model selection has the potential to improve the interpretation of existing observational data and possibly detect yet undiscovered planets. We present a new and efficient Bayesian method for determining the number of extrasolar planets, as well as for inferring their orbital parameters, without having to calculate directly the Bayesian evidence for models containing a large number of planets. Instead, we work iteratively and at each iteration obtain a conservative lower limit on the odds ratio for the inclusion of an additional planet into the model. We apply this method to simulated data-sets containing one and two planets and successfully recover the correct number of planets and reliable constraints on the orbital parameters. We also apply our method to RV measurements of HD 37124, 47 Ursae Majoris and HD 10180. For HD 37124, we confirm that the current data strongly favour a three-planet system. We find strong evidence for the presence of a fourth planet in 47 Ursae Majoris, but its orbital period is suspiciously close to one year, casting doubt on its validity. For HD 10180 we find strong evidence for a six-planet system.
In arXiv:0911.2150, Rutger van Haasteren seeks to criticize the nested sampling algorithm for Bayesian data analysis in general and its MultiNest implementation in particular. He introduces a new method for evidence evaluation based on the idea of Vo
ronoi tessellation and requiring samples from the posterior distribution obtained through MCMC based methods. He compares its accuracy and efficiency with MultiNest, concluding that it outperforms MultiNest in several cases. This comparison is completely unfair since the proposed method can not perform the complete Bayesian data analysis including posterior exploration and evidence evaluation on its own while MultiNest allows one to perform Bayesian data analysis end to end. Furthermore, their criticism of nested sampling (and in turn MultiNest) is based on a few conceptual misunderstandings of the algorithm. Here we seek to set the record straight.
We present a Bayesian approach to modelling galaxy clusters using multi-frequency pointed observations from telescopes that exploit the Sunyaev--Zeldovich effect. We use the recently developed MultiNest technique (Feroz, Hobson & Bridges, 2008) to ex
plore the high-dimensional parameter spaces and also to calculate the Bayesian evidence. This permits robust parameter estimation as well as model comparison. Tests on simulated Arcminute Microkelvin Imager observations of a cluster, in the presence of primary CMB signal, radio point sources (detected as well as an unresolved background) and receiver noise, show that our algorithm is able to analyse jointly the data from six frequency channels, sample the posterior space of the model and calculate the Bayesian evidence very efficiently on a single processor. We also illustrate the robustness of our detection process by applying it to a field with radio sources and primordial CMB but no cluster, and show that indeed no cluster is identified. The extension of our methodology to the detection and modelling of multiple clusters in multi-frequency SZ survey data will be described in a future work.
We study the properties of the constrained minimal supersymmetric standard model (mSUGRA) by performing fits to updated indirect data, including the relic density of dark matter inferred from WMAP5. In order to find the extent to which mu < 0 is disf
avoured compared to mu > 0, we compare the Bayesian evidence values for these models, which we obtain straightforwardly and with good precision from the recently developed multi-modal nested sampling (MultiNest) technique. We find weak to moderate evidence for the mu > 0 branch of mSUGRA over mu < 0 and estimate the ratio of probabilities to be P(mu > 0)/P(mu < 0) = 6-61 depending on the prior measure and range used. There is thus positive (but not overwhelming) evidence that mu > 0 in mSUGRA. The MultiNest technique also delivers probability distributions of parameters and other relevant quantities such as superpartner masses. We explore the dependence of our results on the choice of the prior measure used. We also use the Bayesian evidence to quantify the consistency between the mSUGRA parameter inferences coming from the constraints that have the largest effects: (g-2)_mu, BR(b -> s gamma) and cold dark matter (DM) relic density Omega_{DM}h^2.
We present an efficient and robust approach for extracting clusters of galaxies from weak lensing survey data and measuring their properties. We use simple, physically-motivated cluster models appropriate for such sparse, noisy data, and incorporate
our knowledge of the cluster mass function to optimise the detection of low-mass objects. Despite the methods non-linear nature, we are able to search at a rate of approximately half a square degree per hour on a single processor, making this technique a viable candidate for future wide-field surveys. We quantify, for two simulated data-sets, the accuracy of recovered cluster parameters, and discuss the completeness and purity of our shear-selected cluster catalogues.
