top of page

- Carl Sagan -

“Somewhere something incredible is waiting to be known.”

EoR

Epoch of Reionization

All the gas in the Universe went through two major events: recombination and reionization. While recombination led to the formation of neutral Hydrogen (and Helium!), most of the Universe's Hydrogen is now ionised. Hence it is theorised that there was a period in the Universe's history when Hydrogen was reionized. 

However, the exact sources of reionization and their individual rates and contribution, as well as their effects on structure formation are unknown. The 21-cm line in Hydrogen is a good way to study this, as it is a tracer for neutral Hydrogen. It is temperature dependent, and thus when exposed to Lyman-alpha photons, the 21-cm line can strengthen due to spin flipping while absorbing and re-emitting the Lyman-alpha photon. 

This 21-cm signal, if detected, can thus give us an idea about structure formation and energy sources of reionisation. The LOFAR radio telescope is one of the telescopes used to detect this signal. For doing this, it is necessary to accurately identify the signal while removing all other foreground sources and noise. One way to have a strong identifier is to use a Machine Learning based algorithm to train the identifier. This can then be used along with Gaussian Process Regression to extract the signal. However, we do not have any observed training datasets, as the signal has not been observed yet. What we do have are simulations: ranging from semi-numerical code like 21cmFAST to full simulations (N-body + 3D radiative transfer). We also have GRIZZLY, which uses N-body + 1D radiative transfer, a compromise between the fastest and most "physical" codes. 

Right now, I am testing the improvement, if any, of using GRIZZLY trained identifiers over previously used fitting functions (decided by setting functions for covariance kernels). For this, we are generating mock datasets using various simulations, adding foreground and noise to simulate real data, and then testing out our ML based function on it. It is crucial to have a robust identifier, as the exact properties of the 21-cm signal are unknown. With this project, we hope to achieve that and highlight our best choice's limitations.

We are also exploring methods to minimise errors in theoretical modelling and improving cosmological simulations for the reionization era. For this, I am running a setup similar to the THESAN simulations to test the improvement in computational expense possible with the Fixed & Paired method. I am also testing improvements in resolution by using significantly larger N-body simulations (GADGET-4), post-processed with L-Galaxies for semi-analytical modelling of galaxies and GRIZZLY for radiative transfer.

The Circumgalactic Medium

The benefit of studying the Circumgalactic Medium (CGM) is twofold: 

  1. The CGM is the place where outflowing gases expelled from the galactic disk interact with the inflowing gases from the intergalactic medium and these interactions govern how star-formation proceeds in the galaxy. Thus, knowing more about the CGM (density, size, metallicity, etc.) allows us to know more about the overall structure and evolution of the galaxy. However, we cannot directly observe the CGM. Rather, we depend on absorption spectra, when the CGM of a galaxy obscures a UV background source, like a quasar. But if we only have the data on the absorption spectra, we are limited in our knowledge of the actual unabsorbed spectra emitted by the quasar in question. What we can do in this situation is to use a range of quasar models to predict the properties of the CGM, in order to know the maximum variability one can expect in the physical parameters of the CGM.

  2. Using a variety of models to predict the parameters of the CGM can be used as a tool to find which of the models is "closest" to the actual quasar in question. Thus, if we can successfully quantify how "good" or "bad" a model is in predicting the physical parameters of the CGM clouds, we can use it to predict the properties of more and more quasars!

We explore this in a project to quantify the extent of error possible on using different UV background models. Our work also indicates the possibility of obtaining measures of how "off" our chosen UVB model is, as compared to the true quasar spectra.

cgm

Other Research Interests

Having always been fascinated by the various aspects of astronomy, I have had my fair share of exploring a myriad of topics. Some of which are listed below:

  • Supermassive Black Holes, Galaxy Evolution: Numerical N-body simulations to study the behaviour and interactions of a Supermassive Black Hole binary formed during a galaxy merger event, with the stellar matter of the 2 galaxies can be a fascinating approach to understand and analyse the Final Parsec Problem. Working on this can also give us tighter constraints on the properties of the expected gravitational waves from the eventual merger of such an SMBH binary. I am interested in the development of such simulations, as well as in the analysis of data generated by them in order to study the individual interactions between stars and the black holes.

  • Observational X-ray Astronomy: Interactions can be of many forms, and to have an idea about the breadth of the same, I have explored observational data on the "small" scale of stellar X-ray astronomy for my Master's thesis. We focussed on the star system HD 179949, which has a Sun-like star with a Jupiter mass planet revolving around it, at an orbit of just ~0.04 AU (a tenth of a distance between Mercury and our Sun!). Such close distances suggest the strong possibility of Star-Planet Interactions (SPI). In particular, we found the possibility of interactions between their magnetic fields. We used archival data from Chandra, and built our phase coverage using data from XMM-Newton and Swift. We find multiple evidences pointing towards the existence of SPI, with variability tied to the planet's period of revolution, as well as its beat period with the polar rotation of the star. Even the abundance measurements of the stellar corona agree with results from similar stars with close-in Jupiter like planets, but not with similar stars with no detected planets. We also managed to model the stellar corona, by developing new methodologies to discriminate between fitting models and maximise our understanding of its structure, temperature and metallicity. Our goodness of fit estimator package can be found here: csresid.py

other
bottom of page