This is Lite Plone Theme
You are here: Home Info Research Highlights 2012-13

Research Highlights 2012-13

Includes searches for non-Gaussianity, an upper limit of the radiation efficiency of black holes, gravitational waves from the early universe, regulating star formation in galactic building blocks, a report on Supercomputing 2013, and an ExoMol highlight

COSMOS Research Highlights

The scientific output of the COSMOS consortium has continued with about 100 publications and arXiv papers since DiRAC-2 installation in four key inter-related areas: (i) extreme universe, (ii) cosmic microwave sky, (iii) dark energy and (iv) galaxy formation and (v) black holes.

Planck Satellite Science – Searches for Non-Gaussianity: The ESA Planck satellite, launched in May 2009, provides an unprecedented high resolution survey of the temperature of the cosmic microwave background (CMB) radiation. Over twenty papers describing the analysis of the first two (out of four) surveys, including full sky maps, were published on 21 March 2013. COSMOS and HPCS@DiRAC resources were vital for the science exploitation efforts in several key Planck papers. This effort has led to new estimates of 
cosmological parameters. However, 
Planck has also crossed important qualitative thresholds, making gravitational lensing and the study of non-Gaussian statistics truly quantitative subject areas for the first time.

Using the unique capabilities of the COSMOS system, consortium members have used Planck data to undertake the most stringent tests of the inflationary paradigm to date by studying the prediction that primordial fluctuations should have Gaussian statistics (i.e. most results in the ground-breaking Planck Non-Gaussianity paper (XXIV)). The three-point correlator (“triangles on the sky”) or bispectrum was evaluated to high precision for the first time (Figure 1); it is a 3D tetrahedral object depending on three triangle multipoles l1, l2, l3.

These limits, up to four times stronger than WMAP, significantly constrain broad classes of alternative inflationary models. Despite the stringent constraints on scale-invariant models, the Planck bispectrum reconstructions exhibit large non-Gaussian signals (Figure 1), inspiring further investigation of other non-Gaussian models using separable modal methods developed on COSMOS. The two most promising classes of models appear to be those with excited inital states (non-Bunch Davies vacua) and those exhibiting periodic features with apparent high significance. Intensive non-Gaussian analysis of the Planck data is ongoing for a broad range of primordial models, as well as investigating second-order late-time effects such as ISW-lensing.

An upper limit of the radiation efficiency of black holes: Collisions of black holes are probably the most violent possible events in the universe. Numerical calculations show that astrophysical stellar-mass black holes, when colliding, convert about 5% of their rest mass into gravitational radiation within less than a second. This burst contains more energy than the sun radiates in its entire lifetime. This remarkable efficiency, however, is even eclipsed by that of black holes colliding near the speed of light where up to 35% of the total mass can be radiated in gravitational waves. Such high-energy collisions may be unlikely in astrophysical systems, but they are of great relevance in ongoing efforts to test so-called TeV gravity scenarios at the Large Hadron Colider (LHC). In these scenarios, the strength of gravity may increase much more rapidly than predicted by Newtonian or Einstein gravity at microscopically small distances, an effect due to the presence of extra spacetime dimensions. This opens up the exciting possibility that collision experiments at the LHC may produce black holes. One assumption underlying present analysis of the experimental data is that the black holes thus formed have a mass of the same order of magnitude as the centre-of-mass energy. The enormous efficiency of gravitational wave emission in the collision of black holes has cast doubts on this assumption and led to the conjecture that nearly all the energy may instead be radiated away. Recent work by Sperhake and collaborators on COSMOS has demonstrated that this is not the case and that the usual assumption is correct. About half the total energy can be lost in gravitational waves, but no more than that, as the remaining energy is always absorbed. The figure below illustrates this absorption.energy_absorption.jpg Coming in from the right, a black hole represented by a small circle moves to the left, where it interacts with the other binary member (not shown). Later snapshots show how the black hole increases in size and slows down. This outcome is the same whether a merger occurs or not.

Gravitational waves from the Early Universe: A great deal of information lies in the gravitational wave signal from processes in the very early universe, which can be explored using future ground-based and space-based gravitational wave detectors. COSMOS@DiRAC has been used to calculate GW production from preheating (the period in the cosmological evolution right after inflation) [Bethke et al, PRL 111 (2013) 011301], from topological defects (Figueroa et al, PRL 110 (2013) 10, 101302), and from first-order phase transitions (Hindmarsh, et al, arXiv:1304.2433). The last includes the electroweak transition, where the Higgs field “turned on”. Highlights include: the discovery that the gravitational wave background from preheating is highly anisotropic; that the gravitational wave background from topological defects is scale-invariant; and that there is a new source of gravitational waves from the sound waves generated by the colliding bubbles in a first order phase transition. This has exciting implications for the detectability of an electroweak transition, which are being explored.

Regulating star formation in galactic building blocks: Because of their low mass, dwarf galaxies are the perfect laboratory for studying the impact of stellar feedback upon star formation, and the structure and dynamics of the interstellar medium. Being physically smaller, the same number of resolution elements as employed in Milky Way-scale simulations means one can probe significantly smaller scales. In addition, their low mass means the energy associated with massive stars (both prior to and subsequent to SN) can drive strong outflows of material from the galaxy, which may interact and fall back to fuel more stars or escape the system altogether. What has remained contentious is the exact mode of star formation which dominates within dwarf galaxies - is it driven by non-axisymmetric structures like spiral arms or is it self-regulated and driven by expanding bubbles? With an entirely novel approach to handling hydrodynamics, diffusion, and simulation timesteps within their GCD+ smoothed particle hydrodynamics code, COSMOS users (Kawata, Gibson, Barnes, Grand et al) have shown that expanding bubbles from stellar feedback create dense filaments for future star formation. Such generations of stars can propagate outwards until the gas density becomes too low. Successive bubble generations can produce the stars in a much larger area compared to models without strong stellar energy feedback. The figure below shows one such bubble’s expansion over a 25 Myr period, with purple showing young stars.


Comparing with Local Group galaxy WLM observations, the team demonstrate that bubble-induced star formation is the most likely way to spread star formation, to keep the overall metallicity and rotation velocity low, while at the same time allowing high stellar velocity dispersion.

COSMOS Technical Highlight – Supercomputing 2013

The Cosmos team, in collaboration with Intel, has so far successfully ported two of its consortium codes to the Xeon Phi. Namely Walls and CAMB. The Walls code simulates the evolution of domain walls in the early Universe. It is a grid-based stencil code that evolves a scalar field through time and calculates the area of any domain walls that form. The code was already parallelized via pure OpenMP and has ran on several generations of Cosmos. The code was ran with a 480^3 problem and performance comparisons are between 2 x Intel® Xeon® E5-4650L (Sandybridge) processors and 1 x Intel® Xeon Phi™ 5110P coprocessor.

Out of the box, the Walls code was ~2x faster on two Xeon sockets than on Xeon Phi. Following some fairly straightforward modifications to the algorithm the code was improved via reducing the memory requirement, increasing the vectorisation, and making the code more cache friendly. All in all, after approximately 3-4 weeks of programmer time, the Walls code was sped up by a factor of 5.51x on 2 Xeon’s, and 17x on Xeon Phi. The Xeon Phi version 1.38x faster than the 2 Sandybridge Xeon’s (figure below). The important point is that this was achieved without resorting to low-level optimizations. The code is now being ported to use multiple MICs with a view to running the biggest ever domain wall simulation, accelerated by Xeon Phi. These results were reported in a 15 minute talk at Supercomputing 2013 by Jim Jeffers (Intel).


The figure above shows the speed-up of Walls on 2x Xeon and 1 Xeon Phi with various optimisations. The speed-up is measured against the original runtime of the unimproved code on 2x Xeons.

CAMB is the key cosmological parameter estimation code used by the Planck satellite consortium. Several instances of this code can now be run on a single Xeon Phi card. The multiple Xeon Phi COSMOS system has allowed Planck analysis to be farmed out to the Phi’s, increasing analysis throughput especially for high accuracy cases, while decreasing the load on the rest of the system. Extensive parameter estimation surveys for new models have been undertaken using Planck likelihood data.

ExoMol Highlight

The aim of the ExoMol project is to compute comprehensive spectroscopic line lists for hot molecules thought likely to occur in the atmospheres of exoplanets, brown dwarfs and cool stars. The list of such molecules is quite long but only those for four or more atoms require the use of DiRAC resources. Our calculations have been split between two computers located at Cambridge: Darwin, which was used to diagonalise small matrices and to compute transition intensities, and COSMOS which was essential for diagonalising larger matrices. We encountered some performance issues with COSMOS when performing large matrix diagonalisations. This problem was referred to SGI who have very recently supplied us with a new diagonaliser for COSMOS. Our initial tests with this new diagonaliser are very encouraging and suggest that our performance issues should largely be resolved. However we have yet to use diagonaliser in productions runs and the results discussed below did not employ it.

Work using DiRAC during 2013 focussed on four main molecules: methane (CH4), phosphine (PH3), formaldehyde (H2CO) and sulphur trioxide (SO3).Methane is a major absorber in hot Jupiter exoplanets, brown dwarfs and cool carbon stars. There is a huge demand for a reliable hot methane line list and there are several groups working towards this objective. This project was therefore given priority. Our main result is that we have generated a new methane line list, called 10to10, which contains just under 10 billion transitions. This line list is complete for wavelengths longer than 1 μm and temperatures up to 1500 K It is by some distance the most comprehensive line list available for methane (see Fig.1). It is currently being actively used by about a dozen groups worldwide to model methane in a variety of astronomical objects (and one group for studies of the earth’s atmosphere). Fig. 2 shows the spectrum of the brown dwarf spectrum of 2MASS J0559-1404 compared to our simulations.

Our tests on this line list show that for photometric studies of the K, H, J bands show that previously available line lists (a) agree well with 10to10 at 300 K for the K and H bands but significantly underestimate J-band absorption due to lack of experimental data in this region; and (b) seriously underestimate absorption by methane for all bands at temperatures above 1000 K. We have also completed initial, room temperature line lists for PH3 and SO3, and a full line list for hot H2CO.


Absorption of methane at T=1500K - new theoretical spectrum (10to10) compared to the experiment (HITRAN12)