Applications Of SFFT Image Subtraction Algorithm And AI Spectroscopy Prediction In Time Domain Astronomy

The upcoming massive time domain surveys including LSST can empower a huge number of transient discoveries from deep universe.Image subtraction is essential for the transient detections.The point spread function（PSF）of images from ground-based optical telescopes generally varies across the image field due to instrumental optical distortion and atmospheric seeing.Image subtraction algorithm is to match the PSF and sensitivity variations across the entire imaging field,as a result,any brightness variation becomes detectable on the difference.An image subtraction algorithm that can be fully parallelized and enables accurate transient detections is highly desirable for the future transient surveys.However,the current popular methods are not sufficent enough to meet this demand.Another major challenge of the future time domain surveys is the limited spectroscopy resource for transient follow-up observations.It is not realistic nor costeffective to trigger spectroscopy follow-up for every transient candidate discoveried by the powerful imaging surveys.Nevertheless,the current arichival spectral dataset accumulated in the last decades is likely representative enough to develop data driven models for some common transients events,such as Type Ⅰa supernovae（SNeIa）.This promising approach is likely to alleivate the deficiency issue of spectroscopic resources and helps to optimize the planning of spectroscopic follow-up observations.Our paper is comprised of the following parts:In chapter 1,we give a brief introduction to SNela from an observational view point.we also present a short review of the processing modules in transient detection pipeline,and explain the fundamental ideas underlying the poplular image subtraction methods.In chapter 2,we present a data driven method based on Long Short-Term Memory（LSTM）neural networks,which allows predicting spectrum at given phase from arbitrarily time-sampled spectral sequence.The LSTM model makes it possible to reconstruct complete spectral evolution from very limited spectral observations.We show that our method can perform accurate prediction even from a single observed spectrum around maximum light.Although our spectral dataset is dominated by samples of Ⅰanorm SNe,the neural networks can still capture the prominent spectral properties for Ia subtypes.In chapter 3,we introduce the newly developed image subtraction algorithm SFFT.The algorithm allows for image subtraction in Fourier Space.Unlike other popular methods,the method we developed does not rely on any isolated stars for PSF homogenization,instead it can make use of all statistically valid pixels in the image for the construction of the image matching.The algorithm has been tested extensively in real astronomical data taken by a variety of telescopes.We have incoperated the method into the transient detection pipeline of several time-domain surveys,that are based on the Tsinghua university-Ma huateng Telescopes for Survey（TMTS）,Antarctic Survey Telescopes（AST3）and Dark Engery Camera（DECam）supernova program DECamERON.In chapter 4,we introduce the observations of GW 170817 by AST3-Ⅱ telescope located in Antarctica Dome A.We report the optical observations of the GW counterpart in the nearby galaxy NGC 4993 using AST3-Ⅱ.SFFT algorithm is used to detect the optical signal and the photometric data shows a rapidly fading transient at around 1 day after the GW trigger.The brightness and time evolution of the optical transient associated with GW 170817 are broadly consistent with the predictions of models involving merging binary neutron stars.We infer from our data that the merging process ejected about 10^-2 solar mass of radioactive material at a speed of up to 30%the speed of light.In chapter 5,we make a brief summary and discuss the potential applications of our methods.