Research

"Traveler, there is no path; the path is made by walking"

Antonio Machado 

Dynamical evolution of rotating star cluster

Significant fraction of globular clusters is rotating. This rotation accelerates the core-collapse via gravo-gyro catastrophe, which is conceptually similar to gravo-thermal catastrophe. One of our collaborator has extended Fokker-Planck(FP) model of rotating stellar system to study "after core-collapse", using suitable diffusion term in the FP equation. I have carried out direct nbody simulation of rotating star clusters using Nbody6 code and compared to FP results. Two results shows reasonably good agreements with small difference in detail due to intrinsically different approaches. More details.

Self gravitating system in the external potential well

I investigated the problem of equilibrium of isothermal sphere bounded by solid wall. Instead of solid wall, I posed a situtation where the self gravitating system is in the center of deep external potential well and calculated an equilirium condition of the system. This study showed that a potential well can be a heat bath increasing the velocity dispersion of the embedded system and proposed a new equilibrium model of the system. I also simulated the dynamical evolution of star cluster embedded in the external potential using the modified direct Nbody simulation code, Nbody6. This direct simulation shows an equilibrium density profile which is similar to the analytic model. See more detail.

Marginal likelihood computation

In Bayesian statistics, marginal likelihood or often called 'Bayesian evidence' is used for model comparison, which has a broad application to astronomical modeling. However it is challenging to compute a reliable marginal likelihood if the model has large dimensional parameter space, owing to the 'curse of dimensionality'. I tested a practical approach to improve the reliability of marginal likelihood computed from the simulated posterior samples. Combining samples from tempered MCMC simulation increases the samples in the posterior region which is sparsely sampled and makes the marginal likelihood computation quickly converged with smaller number of samples than running a single MCMC simulation.   

Bayesian galaxy image decomposition software GALPHAT

We have developed a new image decomposition software GALPHAT (GALaxy PHotometric ATtributes) to model galaxy structure in a statistically robust way. GALPHAT is an application of Bayesian Inference Engine which has been primarily developed by Prof. Martin Weinberg and computer science people in UMASS.

Bayesian MCMC draws the random samples from the posterior distribution of model parameters and thus we can characterize the full probability distribution of model parameters. This is a big difference from previous approaches in galaxy image analysis. In many cases, people have estimated just best-fit parameters using simple chi^2 minimization type technique, but have not carefully taken care of all the correlations in the parameters, which is required to derive any robust conclusion about galaxy evolution and formation hypotheses. 

We have extensively tested the software and convinced ourselves that it is working as what we expect. We will use this software for scientific analysis of galaxy image data and put statistically robust confidence on the theory/hypothesis of galaxy formation and evolution. This will be a part of my Ph.D. thesis.  More detail...

AzTEC data reduction pipeline for extended sources

AzTEC is a bolometer array camera for detecting thermal continnum emission from dust in the universe and will be one of primary instruments for LMT, a 50m single dish in Mexico. As my second year project in UMASS, I was involved in the development of AzTEC data reduction pipeline focusing on diffuse emission from any extended astronomical sources. In contrast to the detection of point sources like high-z submm galaxy population, reducing signal(which is basically time series) from the extended source is very diffucult since the signal is dominated by large temporal and spatial variation of thermal emission from water vapor. Therefore standard technique (e.g. principal component analysis) used in reducing the data from point sources can not cleanly separate the signal from the background due to the degeneracy.

I have implemented a simple algorithm for doing it and written an IDL code in Object Oriented Programming style. Preliminary test result looked promising however, the significant improvement is required to apply to real data. I am still interested in it.