A slice through the    Millennium XXL    simulation showing the cosmic web of structures. The bright yellow regions are the high density clusters that form at the intersection of filaments.

A slice through the Millennium XXL simulation showing the cosmic web of structures. The bright yellow regions are the high density clusters that form at the intersection of filaments.

Simulating the Universe

If you could zoom out on the Universe - so far that entire galaxies containing hundreds of billions of stars are just tiny points of light - you would find that the Universe is ordered in complex and interconnected structures. If you could then turn back the clock on the Universe - so far that the Big Bang was just tens of thousands of years previous, instead of billions - you would instead find a Universe that looks almost exactly the same everywhere, with hardly any structures. The way in which the smooth and homogeneous early Universe evolved, through gravity pulling mass together while the fabric of space itself expands, into the “cosmic web” of filaments, clusters, and voids we see today is the subject of my research.

Even in a good budget situation, it is not entirely feasible to test models of structure formation by creating a universe and measuring what happens as it evolves. Instead, cosmologists in my field rely on computer simulations that solve the complex nonlinear equations of gravity for different cosmological models. We then can compare the results of these simulations to the distribution of galaxies observed with our telescopes and determine which model produces the best match to the observations.

But there is a problem. It turns out that most of the “stuff” in the Universe can't be seen by telescopes at all, but only inferred through the effect it has on the movements of stars and gas via gravity. But because this stuff, called “dark matter”, makes up most of the mass in the Universe, we can understand structure formation on large scales by simulating the gravitational evolution of dark matter and ignoring the formation of stars and galaxies. Even though we don’t know what the dark matter actually is.

My research uses cosmological simulations to address these questions: How do large-scale structures form? Why is the Universe accelerating? How can we test cosmological models that tweak general relativity to explain this accelerated expansion?

<Publication list from ADS> <Publication list from arXiv>



Peta-scale simulations

Since we can’t simulate our observed Universe exactly, only its statistical properties, one large simulation is not enough to measure correlation statistics on very large scales. Indra is a suite of cosmological N-body simulations that will capture the large-scale modes of the matter power spectrum with an excellent handle on cosmic variance, required for precise covariance matrices and mock catalogs of future galaxy surveys.

The dark matter particle positions and velocities of all 64 time steps of each of the 512 gigaparsec simulations, as well as the halo catalogs, amount to over a peta-byte of data. We are developing the data storage framework and tools for server-side analysis of the full suite of simulations by the public, planned to be hosted on the SciServer.


The Cosmic Web

The large-scale structure of the Universe is ordered into a cosmic web of halos, filaments, walls, and voids. ORIGAMI is a C package that determines the cosmic web morphology of each dark matter particle in a simulation by associating structures with the dimensionality of their phase-space folds.

The image on the left shows the initial (Lagrangian) positions of particles colored by their final-positions ORIGAMI morphology: halos are big red blobs in Lagrangian-space, surrounded by yellow filaments, in a background of cyan walls and a sea of blue voids.


Testing Gravity

The expansion of the Universe is accelerating, and we don’t know why. Observations are consistent with a cosmological constant, but its measured value is 120 orders of magnitude off of our theoretical expectation. A possible way out is to tweak general relativity, modifying gravity on cosmological scales while suppressing new scalar interactions on Solar System scales.

My work has found that the Vainshtein screening mechanism depends on the cosmic web morphology of simulation particles, screening all halos equally, while the chameleon mechanism depends on halo mass and environment. Voids are a potentially fruitful test-bed for these theories. My students have studied the degeneracies between these models and hydrodynamical effects of galaxy formation and characterized their effect on cosmic filaments.