arXiv subjects

Yi Mao

Publications and source records attributed to Yi Mao.

At least 91 records · Page 5Linked to original sources

arXiv2018

Action-depedent Control Variates for Policy Optimization via Stein's Identity

Policy gradient methods have achieved remarkable successes in solving challenging reinforcement learning problems. However, it still often suffers from the large variance issue on policy gradient estimation, which leads to poor sample efficiency during training. In this work, we propose a control variate method to effectively reduce variance for policy gradient methods. Motivated by the Stein's identity, our method extends the previous control variate methods used in REINFORCE and advantage actor-critic by introducing more general action-dependent baseline functions. Empirical studies show that our method significantly improves the sample efficiency of the state-of-the-art policy gradient approaches.

stat.ML↗

arXiv2017

The Giant Radio Array for Neutrino Detection (GRAND): Present and Perspectives

The Giant Radio Array for Neutrino Detection (GRAND) aims at detecting ultra-high energy extraterrestrial neutrinos via the extensive air showers induced by the decay of tau leptons created in the interaction of neutrinos under the Earth's surface. Consisting of an array of $\sim10^5$ radio antennas deployed over $\sim 2\times10^5\,\rm {km}^2$, GRAND plans to reach, for the first time, an all-flavor sensitivity of $\sim1.5\times10^{-10} \,\rm GeV\, cm^{-2} \,s^{-1}\, sr^{-1}$ above $5\times10^{17}$ eV and a sub-degree angular resolution, beyond the reach of other planned detectors. We describe here preliminary designs and simulation results, plans for the ongoing, staged approach to the construction of GRAND, and the rich research program made possible by GRAND's design sensitivity and angular resolution.

astro-ph.IM↗

arXiv2015

The Impact of Nonlinear Structure Formation on the Power Spectrum of Transverse Momentum Fluctuations and the Kinetic Sunyaev-Zel'dovich Effect

Cosmological transverse momentum fields, whose directions are perpendicular to Fourier wave vectors, induce temperature anisotropies in the cosmic microwave background via the kinetic Sunyaev-Zeldovich (kSZ) effect. The transverse momentum power spectrum contains the four-point function of density and velocity fields, $\langleδδv v\rangle$. In the post-reionization epoch, nonlinear effects dominate in the power spectrum. We use perturbation theory and cosmological $N$-body simulations to calculate this nonlinearity. We derive the next-to-leading order expression for the power spectrum with a particular emphasis on the connected term that has been ignored in the literature. While the contribution from the connected term on small scales ($k>0.1\,h\,\rm{Mpc}^{-1}$) is subdominant relative to the unconnected term, we find that its contribution to the kSZ power spectrum at $\ell = 3000$ at $z<6$ can be as large as ten percent of the unconnected term, which would reduce the allowed contribution from the reionization epoch ($z>6$) by twenty percent. The power spectrum of transverse momentum on large scales is expected to scale as $k^2$ as a consequence of momentum conservation. We show that both the leading and the next-to-leading order terms satisfy this scaling. In particular, we find that both of the unconnected and connected terms are necessary to reproduce $k^2$.

astro-ph.CO↗

arXiv2015

The Linear Perturbation Theory of Reionization in Position-Space: Cosmological Radiative Transfer Along the Light-Cone

The linear perturbation theory of inhomogeneous reionization (LPTR) has been developed as an analytical tool for predicting the global ionized fraction and large-scale power spectrum of ionized density fluctuations during reionization. In the original formulation of the LPTR, the ionization balance and radiative transfer equations are linearized and solved in Fourier space. However, the LPTR's approximation to the full solution of the radiative transfer equation is not straightforward to interpret, since the latter is most intuitively conceptualized in position space. To bridge the gap between the LPTR and the language of numerical radiative transfer, we present a new, equivalent, position-space formulation of the LPTR that clarifies the approximations it makes and facilitates its interpretation. We offer a comparison between the LPTR and the excursion-set model of reionization (ESMR), and demonstrate the built-in capability of the LPTR to explore a wide range of reionization scenarios, and to go beyond the ESMR in exploring scenarios involving X-rays.

astro-ph.CO↗

arXiv2014

Cosmologically probing ultra-light particle dark matter using 21 cm signals

There can arise ubiquitous ultra-light scalar fields in the Universe, such as the pseudo-Goldstone bosons from the spontaneous breaking of an approximate symmetry, which can make a partial contribution to the dark matter and affect the large scale structure of the Universe. While the properties of those ultra-light dark matter are heavily model dependent and can vary in a wide range, we develop a model-independent analysis to forecast the constraints on their mass and abundance using futuristic but realistic 21 cm observables as well as CMB fluctuations, including CMB lensing measurements. Avoiding the highly nonlinear regime, the 21 cm emission line spectra are most sensitive to the ultra-light dark matter with mass m ~10^{-26} eV for which the precision attainable on mass and abundance bounds can be of order of a few percent.

hep-ph↗

arXiv2014

Light cone effect on the reionization 21-cm signal II: Evolution, anisotropies and observational implications

Measurements of the HI 21-cm power spectra from the reionization epoch will be influenced by the evolution of the signal along the line-of-sight direction of any observed volume. We use numerical as well as semi-numerical simulations of reionization in a cubic volume of 607 Mpc across to study this so-called light cone effect on the HI 21-cm power spectrum. We find that the light cone effect has the largest impact at two different stages of reionization: one when reionization is $\sim 20\%$ and other when it is $\sim 80\%$ completed. We find a factor of $\sim 4$ amplification of the power spectrum at the largest scale available in our simulations. We do not find any significant anisotropy in the 21-cm power spectrum due to the light cone effect. We argue that for the power spectrum to become anisotropic, the light cone effect would have to make the ionized bubbles significantly elongated or compressed along the line-of-sight, which would require extreme reionization scenarios. We also calculate the two-point correlation functions parallel and perpendicular to the line-of-sight and find them to differ. Finally, we calculate an optimum frequency bandwidth below which the light cone effect can be neglected when extracting power spectra from observations. We find that if one is willing to accept a $10 \%$ error due to the light cone effect, the optimum frequency bandwidth for $k= 0.056 \, \rm{Mpc}^{-1}$ is $\sim 7.5$ MHz. For $k = 0.15$ and $0.41 \, \rm{Mpc}^{-1}$ the optimum bandwidth is $\sim 11$ and $\sim 16$ MHz respectively.

astro-ph.CO↗

arXiv2013

Simulating cosmic reionization: How large a volume is large enough?

We present the largest-volume (425 Mpc/h=607 Mpc on a side) full radiative transfer simulation of cosmic reionization to date. We show that there is significant additional power in density fluctuations at very large scales. We systematically investigate the effects this additional power has on the progress, duration and features of reionization, as well as on selected reionization observables. We find that comoving simulation volume of ~100 Mpc/h per side is sufficient for deriving a convergent mean reionization history, but that the reionization patchiness is significantly underestimated. We use jackknife splitting to quantify the convergence of reionization properties with simulation volume for both mean-density and variable-density sub-regions. We find that sub-volumes of ~100 Mpc/h per side or larger yield convergent reionization histories, except for the earliest times, but smaller volumes of ~50 Mpc/h or less are not well converged at any redshift. Reionization history milestones show significant scatter between the sub-volumes, of Delta z=0.6-1 for ~50 Mpc/h volumes, decreasing to Delta z=0.3-0.5 for ~100 Mpc/h volumes, and $Δz$~0.1 for ~200 Mpc/h volumes. If we only consider mean-density sub-regions the scatter decreases, but remains at Delta z~0.1-0.2 for the different size sub-volumes. Consequently, many potential reionization observables like 21-cm rms, 21-cm PDF skewness and kurtosis all show good convergence for volumes of ~200 Mpc/h, but retain considerable scatter for smaller volumes. In contrast, the three-dimensional 21-cm power spectra at large scales (k<0.25 h/Mpc) do not fully converge for any sub-volume size. These additional large-scale fluctuations significantly enhance the 21-cm fluctuations, which should improve the prospects of detection considerably, given the lower foregrounds and greater interferometer sensitivity at higher frequencies. (abridged)

astro-ph.CO↗

arXiv2013

Primordial Non-Gaussianity Estimation using 21 cm Tomography from the Epoch of Reionization

Measuring the small primordial nonGaussianity (PNG) predicted by cosmic inflation theories may help diagnose them. The detectability of PNG by its imprint on the 21cm power spectrum from the epoch of reionization is reassessed here in terms of $f_{NL}$, the local nonlinearity parameter. We find that an optimum, multi-frequency observation by SKA can achieve $Δf_{NL} \sim 3$ (comparable to recent Planck CMB limits), while a cosmic-variance-limited array of this size like Omniscope can even detect $Δf_{NL} \sim 0.2$. This substantially revises the methods and results of previous work.

astro-ph.CO↗

arXiv2013

Probing reionization with LOFAR using 21-cm redshift space distortions

One of the most promising ways to study the epoch of reionization (EoR) is through radio observations of the redshifted 21-cm line emission from neutral hydrogen. These observations are complicated by the fact that the mapping of redshifts to line-of-sight positions is distorted by the peculiar velocities of the gas. Such distortions can be a source of error if they are not properly understood, but they also encode information about cosmology and astrophysics. We study the effects of redshift space distortions on the power spectrum of 21-cm radiation from the EoR using large scale $N$-body and radiative transfer simulations. We quantify the anisotropy introduced in the 21-cm power spectrum by redshift space distortions and show how it evolves as reionization progresses and how it relates to the underlying physics. We go on to study the effects of redshift space distortions on LOFAR observations, taking instrument noise and foreground subtraction into account. We find that LOFAR should be able to directly observe the power spectrum anisotropy due to redshift space distortions at spatial scales around $k \sim 0.1$ Mpc$^{-1}$ after $\gtrsim$ 1000 hours of integration time. At larger scales, sample errors become a limiting factor, while at smaller scales detector noise and foregrounds make the extraction of the signal problematic. Finally, we show how the astrophysical information contained in the evolution of the anisotropy of the 21-cm power spectrum can be extracted from LOFAR observations, and how it can be used to distinguish between different reionization scenarios.

astro-ph.CO↗

arXiv2013

The scale-dependent signature of primordial non-Gaussianity in the large-scale structure of cosmic reionization

(ABRIDGED)The rise of cosmic structure depends upon the statistical distribution of initial density fluctuations generated by inflation. While the simplest models predict an almost perfectly Gaussian distribution, more-general models predict a level of primordial non-Gaussianity (PNG) that observations might yet be sensitive enough to detect. Recent Planck Collaboration measurements of the CMB temperature anisotropy bispectrum significantly tighten the observational limits, but they are still far from the PNG level predicted by the simplest models of inflation. Probing levels below CMB sensitivities will require other methods, such as searching for the statistical imprint of PNG on galactic halo clustering. During the epoch of reionization (EoR), the first stars and galaxies released radiation into the intergalactic medium (IGM) that created ionized patches whose large-scale geometry and evolution reflected the underlying abundance and large-scale clustering of the star-forming galaxies. This statistical connection between ionized patches in the IGM and galactic halos suggests that observing reionization may be another way to constrain PNG. We employ the linear perturbation theory of reionization and semi-analytic models based on the excursion-set formalism to model the effects of PNG on the EoR. We quantify the effects of PNG on the large-scale structure of reionization by deriving the ionized density bias, i.e. ratio of ionized atomic to total matter overdensities in Fourier space, at small wavenumber. Just as previous studies found that PNG creates a scale-dependent signature in the halo bias, so, too, we find a scale-dependent signature in the ionized density bias. Our results, which differ significantly from previous attempts in the literature to characterize this PNG signature, will be applied elsewhere to predict its observable consequences, e.g. in the cosmic 21cm background.

astro-ph.CO↗

arXiv2013

Will Nonlinear Peculiar Velocity and Inhomogeneous Reionization Spoil 21cm Cosmology from the Epoch of Reionization?

The 21cm background from the epoch of reionization is a promising cosmological probe: line-of-sight velocity fluctuations distort redshift, so brightness fluctuations in Fourier space depend upon angle, which linear theory shows can separate cosmological from astrophysical information. Nonlinear fluctuations in ionization, density and velocity change this, however. The validity and accuracy of the separation scheme are tested here for the first time, by detailed reionization simulations. The scheme works reasonably well early in reionization (< 40% ionized), but not late (> 80% ionized).

astro-ph.CO↗

arXiv2012

Simulating Cosmic Reionization and the Radiation Backgrounds from the Epoch of Reionization

Large-scale reionization simulations are described which combine the results of cosmological N-body simulations that model the evolving density and velocity fields and identify the galactic halo sources, with ray-tracing radiative transfer calculations which model the nonequilibrium ionization of the intergalactic medium. These simulations have been used to predict some of the signature effects of reionization on cosmic radiation backgrounds, including the CMB, near-IR, and redshifted 21cm backgrounds. We summarize some of our recent progress in this work, and address the question of whether observations of such signature effects can be used to distinguish the relative contributions of galaxies of different masses to reionization.

astro-ph.CO↗

arXiv2012

Detecting the Rise and Fall of the First Stars by Their Impact on Cosmic Reionization

The intergalactic medium was reionized before redshift z~6, most likely by starlight which escaped from early galaxies. The very first stars formed when hydrogen molecules (H2) cooled gas inside the smallest galaxies, minihalos of mass between 10^5 and 10^8 solar masses. Although the very first stars began forming inside these minihalos before redshift z~40, their contribution has, to date, been ignored in large-scale simulations of this cosmic reionization. Here we report results from the first reionization simulations to include these first stars and the radiative feedback that limited their formation, in a volume large enough to follow the crucial spatial variations that influenced the process and its observability. We show that, while minihalo stars stopped far short of fully ionizing the universe, reionization began much earlier with minihalo sources than without, and was greatly extended, which boosts the intergalactic electron-scattering optical depth and the large-angle polarization fluctuations of the cosmic microwave background significantly. Although within current WMAP uncertainties, this boost should be readily detectable by Planck. If reionization ended as late as z_ov<~7, as suggested by other observations, Planck will thereby see the signature of the first stars at high redshift, currently undetectable by other probes.

astro-ph.CO↗

arXiv2012

Light cone effect on the reionization 21-cm power spectrum

Observations of redshifted 21-cm radiation from neutral hydrogen during the epoch of reionization (EoR) are considered to constitute the most promising tool to probe that epoch. One of the major goals of the first generation of low frequency radio telescopes is to measure the 3D 21-cm power spectrum. However, the 21-cm signal could evolve substantially along the line of sight (LOS) direction of an observed 3D volume, since the received signal from different planes transverse to the LOS originated from different look-back times and could therefore be statistically different. Using numerical simulations we investigate this so-called light cone effect on the spherically averaged 3D 21-cm power spectrum. For this version of the power spectrum, we find that the effect mostly `averages out' and observe a smaller change in the power spectrum compared to the amount of evolution in the mean 21-cm signal and its rms variations along the LOS direction. Nevertheless, changes up to 50% at large scales are possible. In general the power is enhanced/suppressed at large/small scales when the effect is included. The cross-over mode below/above which the power is enhanced/suppressed moves toward larger scales as reionization proceeds. When considering the 3D power spectrum we find it to be anisotropic at the late stages of reionization and on large scales. The effect is dominated by the evolution of the ionized fraction of hydrogen during reionization and including peculiar velocities hardly changes these conclusions. We present simple analytical models which explain qualitatively all the features we see in the simulations.

astro-ph.CO↗

arXiv2012

Statistical Translation, Heat Kernels and Expected Distances

High dimensional structured data such as text and images is often poorly understood and misrepresented in statistical modeling. The standard histogram representation suffers from high variance and performs poorly in general. We explore novel connections between statistical translation, heat kernels on manifolds and graphs, and expected distances. These connections provide a new framework for unsupervised metric learning for text documents. Experiments indicate that the resulting distances are generally superior to their more standard counterparts.

cs.LG↗

arXiv2012

Domain Knowledge Uncertainty and Probabilistic Parameter Constraints

Incorporating domain knowledge into the modeling process is an effective way to improve learning accuracy. However, as it is provided by humans, domain knowledge can only be specified with some degree of uncertainty. We propose to explicitly model such uncertainty through probabilistic constraints over the parameter space. In contrast to hard parameter constraints, our approach is effective also when the domain knowledge is inaccurate and generally results in superior modeling accuracy. We focus on generative and conditional modeling where the parameters are assigned a Dirichlet or Gaussian prior and demonstrate the framework with experiments on both synthetic and real-world data.

cs.LG↗

arXiv2012

Redshift Space Distortion of the 21cm Background from the Epoch of Reionization I: Methodology Re-examined

The peculiar velocity of the intergalactic gas responsible for the cosmic 21cm background from the epoch of reionization and beyond introduces an anisotropy in the three-dimensional power spectrum of brightness temperature fluctuations. Measurement of this anisotropy by future 21cm surveys is a promising tool for separating cosmology from 21cm astrophysics. However, previous attempts to model the signal have often neglected peculiar velocity or only approximated it crudely. This paper re-examines the effects of peculiar velocity on the 21cm signal in detail, improving upon past treatment and addressing several issues for the first time. (1) We show that properly accounting for finite optical depth eliminates the unphysical divergence of 21cm brightness temperature in overdense regions of the IGM found by previous work that employed the usual optically-thin approximation. (2) The approximation made previously to circumvent the diverging brightness temperature problem by capping velocity gradient can misestimate the power spectrum on all scales. (3) The observed power spectrum in redshift-space remains finite even in the optically-thin approximation if one properly accounts for the redshift-space distortion. However, results that take full account of finite optical depth show that this approximation is only accurate in the limit of high spin temperature. (4) The linear theory for redshift-space distortion results in ~30% error in the observationally relevant wavenumber range, at the 50% ionized epoch. (5) We describe and test two numerical schemes to calculate the 21cm signal from reionization simulations to incorporate peculiar velocity effects in the optically-thin approximation accurately. One is particle-based, the other grid-based, and while the former is most accurate, we demonstrate that the latter is computationally more efficient and can achieve sufficient accuracy. [Abridged]

astro-ph.CO↗

arXiv2011

Can 21-cm observations discriminate between high-mass and low-mass galaxies as reionization sources?

The prospect of detecting the first galaxies by observing their impact on the intergalactic medium as they reionized it during the first billion years leads us to ask whether such indirect observations are capable of diagnosing which types of galaxies were most responsible for reionization. We attempt to answer this by considering a set of large-scale radiative transfer simulations of reionization in sufficiently large volumes to make statistically meaningful predictions of observable signatures, while also directly resolving all atomically-cooling halos down to 10^8 M_solar. We focus here on predictions of the 21-cm background, to see if upcoming observations are capable of distinguishing a universe ionized primarily by high-mass halos from one in which both high-mass and low-mass halos are responsible, and to see how these results depend upon the uncertain source efficiencies. We find that 21-cm fluctuation power spectra observed by the first generation EoR/21-cm radio interferometer arrays should be able to distinguish the case of reionization by high-mass halos alone from that by both high- and low-mass halos, together. Some reionization scenarios yield very similar power spectra and rms evolution and thus can only be discriminated by their different mean reionization history and 21-cm PDF distributions. We find that the skewness of the 21-cm PDF distribution smoothed over LOFAR-like window shows a clear feature correlated with the rise of the rms due to patchiness. Measurements of the mean photoionization rates are sensitive to the average density of the regions being studied and therefore could be strongly skewed in certain cases. (abridged)

astro-ph.CO↗