arXiv subjects

Xi Meng

Publications and source records attributed to Xi Meng.

7 recordsLinked to original sources

arXiv2022

Tidal disruption of star clusters in galaxy formation simulations

We investigate the evolution of the tidal field experienced by massive star clusters using cosmological simulations of Milky Way-sized galaxies. Clusters in our simulations experience the strongest tidal force in the first few hundred Myr after formation, when the maximum eigenvalue of the tidal tensor reaches several times $10^4$ Gyr$^{-2}$. After about 1 Gyr the tidal field plateaus at a lower value, with the median $\lambda_{\rm m} \sim 3 \times 10^3$ Gyr$^{-2}$. The fraction of time clusters spend in high tidal strength ($\lambda_{\rm m} > 3 \times 10^4$ Gyr$^{-2}$) regions also decreases with their age from $\sim$20% immediately after formation to less than 1% after 1 Gyr. At early ages both the in situ and ex situ clusters experience similar tidal fields, while at older ages the in situ clusters in general experience stronger tidal field due to their lower orbits in host galaxy. This difference is reflected in the survival of clusters: we looked into cluster disruption calculated in simulation runtime and found that ex situ star clusters of the same initial mass typically end up with higher bound fraction at the last available simulation snapshot than the in situ ones.

astro-ph.GA↗

arXiv2021

Reconstruction of the Density Power Spectrum from Quasar Spectra using Machine Learning

We describe a novel end-to-end approach using Machine Learning to reconstruct the power spectrum of cosmological density perturbations at high redshift from observed quasar spectra. State-of-the-art cosmological simulations of structure formation are used to generate a large synthetic dataset of line-of-sight absorption spectra paired with 1-dimensional fluid quantities along the same line-of-sight, such as the total density of matter and the density of neutral atomic hydrogen. With this dataset, we build a series of data-driven models to predict the power spectrum of total matter density. We are able to produce models which yield reconstruction to accuracy of about 1% for wavelengths $k \leq 2 h Mpc^{-1}$, while the error increases at larger $k$. We show the size of data sample required to reach a particular error rate, giving a sense of how much data is necessary to reach a desired accuracy. This work provides a foundation for developing methods to analyse very large upcoming datasets with the next-generation observational facilities.

astro-ph.CO↗

arXiv2020

Evolution of Disc Thickness in Simulated High-Redshift Galaxies

We study the growth of stellar discs of Milky Way-sized galaxies using a suite of cosmological simulations. We calculate the half-mass axis lengths and axis ratios of stellar populations split by age in isolated galaxies with stellar mass $M_* = 10^7 - 10^{10} M_{\odot}$ at redshifts $z$ > 1.5. We find that in our simulations stars always form in relatively thin discs, and at ages below 100 Myr are contained within half-mass height $z_{1/2}$ ~ 0.1 kpc and short-to-long axis ratio $z_{1/2}/x_{1/2}$ ~ 0.15. Disc thickness increases with the age of stellar population, reaching median $z_{1/2}$ ~ 0.8 kpc and $z_{1/2}/x_{1/2}$ ~ 0.6 for stars older than 500 Myr. We trace the same group of stars over the simulation snapshots and show explicitly that their intrinsic shape grows more spheroidal over time. We identify a new mechanism that contributes to the observed disc thickness: rapid changes in the orientation of the galactic plane mix the configuration of young stars. The frequently mentioned "upside-down" formation scenario of galactic discs, which posits that young stars form in already thick discs at high redshift, may be missing this additional mechanism of quick disc inflation. The actual formation of stars within a fairly thin plane is consistent with the correspondingly flat configuration of dense molecular gas that fuels star formation.

astro-ph.GA↗

arXiv2019

Origin of Giant Stellar Clumps in High-Redshift Galaxies

We examine the nature of kpc-scale clumps seen in high-redshift galaxies using a suite of cosmological simulations of galaxy formation. We identify rest-frame UV clumps in mock HST images smoothed to 500 pc resolution, and compare them with the intrinsic 3D clumps of young stars identified in the simulations with 100 pc resolution. According to this comparison, we expect that the stellar masses of the observed clumps are overestimated by as much as an order of magnitude, and that the sizes of these clumps are also overestimated by factor of several, due to a combination of spatial resolution and projection. The masses of young stars contributing most of the UV emission can also be overestimated by factor of a few. We find that most clumps of young stars present in a simulation at one time dissolve on a timescale shorter than $\sim$150 Myr. Some clumps with dense cores can last longer but eventually disperse. Most of the clumps are not bound structures, with virial parameter $\alpha_{\rm vir}$ > 1. We find similar results for clumps identified in mock maps of H$\alpha$ emission measure. We examine the predictions for effective clump sizes from the linear theory of gravitational perturbations and conclude that they are inconsistent with being formed by global disc instabilities. Instead, the observed clumps represent random projections of multiple smaller star-forming regions.

astro-ph.GA↗

arXiv2019

Masked Non-Autoregressive Image Captioning

Existing captioning models often adopt the encoder-decoder architecture, where the decoder uses autoregressive decoding to generate captions, such that each token is generated sequentially given the preceding generated tokens. However, autoregressive decoding results in issues such as sequential error accumulation, slow generation, improper semantics and lack of diversity. Non-autoregressive decoding has been proposed to tackle slow generation for neural machine translation but suffers from multimodality problem due to the indirect modeling of the target distribution. In this paper, we propose masked non-autoregressive decoding to tackle the issues of both autoregressive decoding and non-autoregressive decoding. In masked non-autoregressive decoding, we mask several kinds of ratios of the input sequences during training, and generate captions parallelly in several stages from a totally masked sequence to a totally non-masked sequence in a compositional manner during inference. Experimentally our proposed model can preserve semantic content more effectively and can generate more diverse captions.

cs.CV↗

arXiv2018

Structure and stability of high-redshift galaxies in cosmological simulations

We investigate the structure of galaxies formed in a suite of high-resolution cosmological simulations. Consistent with observations of high-redshift galaxies, our simulated galaxies show irregular, prolate shapes, which are dominated by turbulent motions instead of rotation. Yet molecular gas and young stars are restricted to a relatively thin plane. We examine the accuracy of applying the Toomre linear stability analysis to predict the location and amount of gas available for star formation. We find that the Toomre criterion still works for these irregular galaxies, after correcting for multiple gas and stellar components: the $Q$ parameter in $\rm{H_2}$ rich regions is in the range $0.5-1$, remarkably close to unity. Due to the violent stellar feedback from supernovae and strong turbulent motions, young stars and molecular gas are not always spatially associated. Neither the $Q$ map nor the $\rm{H_2}$ surface density map coincide with recent star formation exactly. We argue that the Toomre criterion is a better indicator of future star formation than a single $\rm{H_2}$ surface density threshold because of the smaller dynamic range of $Q$. The depletion time of molecular gas is below 1~Gyr on kpc scale, but with large scatter. Centering the aperture on density peaks of gas/young stars systematically biases the depletion time to larger/smaller values and increases the scatter.

astro-ph.GA↗

arXiv2016

Star cluster formation in cosmological simulations. I. Properties of young clusters

We present a new implementation of star formation in cosmological simulations, by considering star clusters as a unit of star formation. Cluster particles grow in mass over several million years at the rate determined by local gas properties, with high time resolution. The particle growth is terminated by its own energy and momentum feedback on the interstellar medium. We test this implementation for Milky Way-sized galaxies at high redshift, by comparing the properties of model clusters with observations of young star clusters. We find that the cluster initial mass function is best described by a Schechter function rather than a single power law. In agreement with observations, at low masses the logarithmic slope is $\alpha\approx 1.8-2$, while the cutoff at high mass scales with the star formation rate. A related trend is a positive correlation between the surface density of star formation rate and fraction of stars contained in massive clusters. Both trends indicate that the formation of massive star clusters is preferred during bursts of star formation. These bursts are often associated with major merger events. We also find that the median timescale for cluster formation ranges from 0.5 to 4 Myr and decreases systematically with increasing star formation efficiency. Local variations in the gas density and cluster accretion rate naturally lead to the scatter of the overall formation efficiency by an order of magnitude, even when the instantaneous efficiency is kept constant. Comparison of the formation timescale with the observed age spread of young star clusters provides an additional important constraint on the modeling of star formation and feedback schemes.

astro-ph.GA↗