“The Answer to the Great Question... Of Life, the Universe and Everything... Is... Forty-two,' said Deep Thought, with infinite majesty and calm.”
― Douglas Adams, The Hitchhiker's Guide to the Galaxy
Large-scale structure of the Universe, computational, theoretical, and observational
Primordial non-Gaussianity, inflation
LSSxCMB, cross-survey analysis
Cosmological probes of neutrino mass
Strong lensing for Hubble tension
Statistical and mathematical methods and machine learning
Ongoing and recent projects
Constraining primordial non-Gaussianity using reconstructed density field with field-level template fits and near-optimal estimate of bisepctrum
I am developing a new approach to constraining primordial non-Gaussianity (PNG), using reconstructed density field with field-level template fits and near-optimal estimate of the bispectrum. Reconstruction reverses the nonlinearity due to gravity; by removing this source of confusion, we have a cleaner signal of PNG. Our approach also provides a remedy to the challenges with using bispectrum to constrain PNG -- enormous data vector and covariance matrix resulted from a large number of triangle configurations. This new approach shows strong constraining power for PNG.
Paper I will appear soon
Effective cosmic density field reconstruction with convolutional neural network
This work introduces a method that augments the traditional cosmic density field reconstruction algorithms with a convolutional neural network (CNN). The performance is significantly better than traditional algorithms in high number density matter density field.
Poster: APS April Meeting 2023 poster
Optimal reconstruction for DESI Year 1 BAO constraints
I am co-leading a project on optimal reconstruction for DESI Year 1 BAO analysis. The standard reconstruction method has been used for BAO constraints in LSS surveys for the past decade and has achieved an improvement in BAO distance measurement by a factor of 1.2-2.4 on average. DESI Year 1 will continue using the standard reconstruction framework, but the algorithm for cutsky, cell resolution, smoothing scale etc. all need to be carefully analyzed before applying reconstruction to data. We aim to find the optimal reconstruction scheme for the Y1 analysis, such that we maximize the improvement on BAO measurement over BOSS.
Paper I is ongoing
Paper II is ongoing
Comprehensive comparison of an iterative reconstruction algorithm with the standard reconstrucion algorithm
Over the years, there have been a number of new reconstruction algorithms proposed, but there has not been a detailed comparison among them. In this study, we compare an iterative reconstruction algorithm with the standard reconstruction algorithm, which is the only reconstruction method that has been used in observation. The analysis includes two-point statistics as well as BAO measurement in real and redshift space, and matter and halo fields.
Code: triangular-shaped cloud particle assignment code, two-point statistics calculation (several scripts here), BAO fitting code, fast Python implementation of Hada & Eisenstein (2018) reconstruction algorithm, and standard (Eisenstein et al. 2007) reconstruction
Strong lens search in DESI spectra for Hubble tension study
A strong lens can be used to estimate the time-delay distances, which offers an independent measurement of the Hubble constant. In DESI, we have a large number of spectra that are from more than one object. We discern spectra that are potentially from lens systems from blends (random two or more objects) to make a list of lens candidates, with machine learning techniques. These potential lens will then need high quality imaging observations later to do calculation.