Fangzhou Zhu

Fangzhou Zhu's picture
Quantitative Researcher
Citadel LLC
Research Areas: 
Cosmology -- Large Scale Structure of the universe
Education: 

BA, Physics; MA, Mathematics, University of Pennsylvania, 2012; Ph.D. 2018, Yale University

Advisor: 
Nikhil Padmanabhan
Dissertation Title: 
Information Mining in the Large Scale Structure of the universe
Dissertation Abstract: 

The Baryon Acoustic Oscillations signal has been an important tool to study the properties of dark energy. Designing efficient and robust data analysis methods that optimize the extraction of information is crucial to realize the immense potential of current and future galaxy surveys. To achieve this goal, this thesis presents the development and implementation of the ‘redshift weighting’ (Zhu et al 2015, 2016) and ‘BAO emulation’, two techniques that promise to increase the science return of future BAO experiments.

In current and future BAO surveys, the samples cover a wide range of redshifts. Traditional analyses split the sample into multiple redshift bins and analyze the signals in these narrower slices to improve the redshift resolution of the distance-redshift relation. This approach results in lower signal-to-noise ratio in each slice, reducing the robustness of BAO detection. Signals at disjoint bin boundaries are also lost. “Redshift weighting” seeks to tackle this problem by compressing a sample’s BAO information in the redshift direction onto a small number of weighted correlation functions that preserve nearly all of the signals (Zhu et al 2015).

Building off the work in Zhu et al 2015, we implement the redshift weighting scheme and validate the method on the SDSS-III BOSS DR12 mock galaxy catalogs. We demonstrate the method gives unbiased distance and Hubble parameter estimates. The constraint is also in agreement with a Fisher forecast to within 10%, suggesting the method’s efficiency in producing optimized BAO measurements (Zhu et al 2016).

We apply the redshift weights technique to the clustering of quasars in the SDSS-IV eBOSS DR14 quasar sample. We produce and report 4.6% distance and Hubble parameter measurements using this sample. This work is presented in Zhu et al 2018 (subumitted to MNRAS and available on ArXiv).

BAO emulation is a method we developed to address the computational challenge of BAO modeling. MCMC is one of the most commonly used methods to sample the parameter likelihood surface. To reach a convergent MCMC chain, one typically requires tens of thousands of model evaluations. This high demand puts considerable stress on the speed of BAO model prediction. We built a fast emulator for BAO correlation function predictions. The emulator uses a low discrepancy Sobol sequence to efficiently sample the BAO parameter space, and utilizes Principal Component Analysis (PCA) and Gaussian Process regression (GP) for model prediction. We construct the emulator for two popular BAO models at very modest computational costs. Finally, we implement the emulator and produce BAO measurements from the BOSS DR12 galaxy sample.

I conclude by assessing avenues for future work that can build on the methods developed in this thesis. Future redshift surveys present exciting opportunities to enrich the science return from the BAO method. Both ‘redshift weighting’ and ‘BAO emulation’ can contribute to unlocking the full potential of upcoming surveys, getting us closer to unveiling the mystery of dark energy.