W.L. Kimmy Wu

My research is focused on searching for new physics through precision measurements of the cosmic microwave background (CMB). In particular, I use cutting-edge CMB data to look for imprints of the stochastic background of gravitational waves from inflation, and new physics beyond the Standard Model.

You can find a list of my publications at Google Scholar or arXiv or inSPIRE or orcid.org/0000-0001-5411-6920.


Science - Delensing

delens_spt

I am particularly excited about constraining $r$ through CMB B-mode measurements. Inflation generically predicts a background of primordial gravitational waves, which generate a primordial B-mode component in the polarization of the CMB. The measurement of such a B-mode signature would lend significant support to the paradigm of inflation. Observed B modes also contain a component from the gravitational lensing of primordial E modes, which can obscure the measurement of the primordial B modes. We reduce the sample variance in the BB spectrum contributed from this lensing component by a process called 'delensing.' In this paper, we present delensing of a measurement of the CMB B-mode power spectrum from SPTpol using data from Herschel as a tracer of the lensing potential. The measured B-mode power is reduced by 28% on sub-degree scales, in agreement with predictions from simulations, and the null hypothesis of no delensing is ruled out at 6.9 $\sigma$. Delensing will be essential for future CMB experiment for constraining $r$ [1].

Science - testing $\Lambda$CDM and extensions / Cross-correlations

Now, in particular, is an interesting time to test $\Lambda$CDM: while there is no definitive evidence for tension of cosmological parameters amongst existing datasets, a level of $2-3 \sigma$ discordance in parameters like $H_0$, the Hubble constant, and $\sigma_8$, matter fluctuations at 8 $h^{-1}$ Mpc, are demonstrated both between different CMB experiments and between CMB and optical survey data. We can use data from SPT-3G to improve the measurements of the $\Lambda$CDM parameters and constrain $\Lambda$CDM extension models, like sum of neutrino mass, number of dark radiation species, etc. We can also cross-correlation CMB maps with lower-redshift survey maps (e.g. optical, IR, 21cm) to test the concordance of the $\Lambda$CDM model at these vastly different redshifts.

Reconstruction of the lensing potential, using neural networks

QU2EK

For next-generation CMB experiments, we need higher signal-to-noise extraction of the CMB lensing potential. My collaborators and I built a neural network to do that and showed that the recovered lensing field is of comparable signal-to-noise to maximum-likelihood methods [1]. Besides being a fun project, I am particularly excited about how ML can be used for astrophysics. Because the CMB is very well understood and modeled, we can leverage our understanding of it to get better intuition of neural networks. This is potentially a very powerful tool for the future of science. A blog post for the highlights.

Forecasts

I dabbled in forecasting the capabilities of future generation CMB polarization experiments to constrain fundamental physics: e.g. the sum of neutrion mass ($\Sigma m_{\nu}$), dark radiation content of the universe ($N_{\rm eff}$), dark matter thermal annihilation cross-section, dark energy equation of state, and $r$. What I found is that for many of these fundamental physics parameters, we are sample variance limited -- meaning that measuring a larger sky area improves the constraints on these parameters faster than decreasing the noise of the CMB map. However, for the elusive tensor-to-scalar ratio $r$, reducing the noise of the map (along side with foreground cleaning and delensing) will improve its constraint faster, until we see a signal [1].

Instrumentation

bicep3

As CMB experiments' noise go down in the coming years, instrumental systematics will become the dominant limitation on constraining $r$ - the tensor-to-scalar ratio, a parameter that is directly related to the energy scale of inflation in simple inflation models.
In order to devise optimal extraction of information from the data, and design systematics checks and quantify systematic contamination efficiently, I spent half of my graduate school effort in the design, testing, and deployment of the BICEP3 telescope [1].