ترغب بنشر مسار تعليمي؟ اضغط هنا

Kernel Phase and Coronagraphy with Automatic Differentiation

345   0   0.0 ( 0 )
 نشر من قبل Benjamin Pope
 تاريخ النشر 2020
  مجال البحث فيزياء
والبحث باللغة English




اسأل ChatGPT حول البحث

The accumulation of aberrations along the optical path in a telescope produces distortions and speckles in the resulting images, limiting the performance of cameras at high angular resolution. It is important to achieve the highest possible sensitivity to faint sources such as planets, using both hardware and data analysis software. While analytic methods are efficient, real systems are better-modelled numerically, but such models with many parameters can be hard to understand, optimize and apply. Automatic differentiation software developed for machine learning now makes calculating derivatives with respect to aberrations straightforward for arbitrary optical systems. We apply this powerful new tool to enhance high-angular-resolution astronomical imaging. Self-calibrating observables such as the closure phase or bispectrum have been widely used in optical and radio astronomy to mitigate optical aberrations and achieve high-fidelity imagery. Kernel phases are a generalization of closure phases in the limit of small phase errors. Using automatic differentiation, we reproduce existing kernel phase theory within this framework and demonstrate an extension to the Lyot coronagraph, finding self-calibrating combinations of speckles which are resistant to phase noise, but only in the very high-wavefront-quality regime. As an illustrative example, we reanalyze Palomar adaptive optics observations of the binary alpha Ophiuchi, finding consistency between the new pipeline and the existing standard. We present a new Python package morphine that incorporates these ideas, with an interface similar to the popular package poppy, for optical simulation with automatic differentiation. These methods may be useful for designing improved astronomical optical systems by gradient descent.



قيم البحث

اقرأ أيضاً

The vector-Apodizing Phase Plate (vAPP) is a pupil-plane coronagraph that manipulates phase to create dark holes in the stellar PSF. The phase is induced on the circular polarization states through the inherently achromatic geometric phase by spatial ly varying the fast axis orientation of a half-wave liquid-crystal layer. The two polarized PSFs can be separated, either by a quarter-wave plate (QWP) followed by a polarizing beamsplitter (PBS) for broadband operation, or a polarization sensitive grating (PSG) for narrowband or IFS operation. Here we present new vAPP concepts that lift the restrictions of previous designs and report on their performance. We demonstrated that the QWP+PBS combination puts tight tolerances on the components to prevent leakage of non-coronagraphic light into the dark-hole. We present a new broadband design using an innovative two-stage patterned liquid-crystal element system based on multi-color holography, alleviating the leakage problem and relaxing manufacturing tolerances. Furthermore, we have shown that focal-plane wavefront sensing (FPWFS) can be integrated into the vAPP by an asymmetric pupil. However, such vAPPs suffer from a reduced throughput and have only been demonstrated with a PSG in narrowband operation. We present advanced designs that maintain throughput and enable phase and amplitude wavefront sensing. We also present broadband vAPP FPWFS designs and outline a broadband FPWFS algorithm. Finally, previous dual-beam vAPP designs for sensitive polarimetry with one-sided dark holes were very complex. We show new dual-beam designs that significantly reduce the complexity.
The successes of deep learning, variational inference, and many other fields have been aided by specialized implementations of reverse-mode automatic differentiation (AD) to compute gradients of mega-dimensional objectives. The AD techniques underlyi ng these tools were designed to compute exact gradients to numerical precision, but modern machine learning models are almost always trained with stochastic gradient descent. Why spend computation and memory on exact (minibatch) gradients only to use them for stochastic optimization? We develop a general framework and approach for randomized automatic differentiation (RAD), which can allow unbiased gradient estimates to be computed with reduced memory in return for variance. We examine limitations of the general approach, and argue that we must leverage problem specific structure to realize benefits. We develop RAD techniques for a variety of simple neural network architectures, and show that for a fixed memory budget, RAD converges in fewer iterations than using a small batch size for feedforward networks, and in a similar number for recurrent networks. We also show that RAD can be applied to scientific computing, and use it to develop a low-memory stochastic gradient method for optimizing the control parameters of a linear reaction-diffusion PDE representing a fission reactor.
The Keck Planet Imager and Characterizer (KPIC) is a purpose-built instrument for high-dispersion coronagraphy in the K and L bands on Keck. This instrument will provide the first high resolution (R$>$30,000) spectra of known directly imaged exoplane ts and low-mass brown dwarf companions visible in the northern hemisphere. KPIC is developed in phases. Phase I is currently at Keck in the early operations stage, and the phase II upgrade will deploy in late 2021. The goal of phase II is to maximize the throughput for planet light and minimize the stellar leakage, hence reducing the exposure time needed to acquire spectra with a given signal-to-noise ratio. To achieve this, KPIC phase II exploits several innovative technologies that have not been combined this way before. These include a 1000-element deformable mirror for wavefront correction and speckle control, a set of lossless beam shaping optics to maximize coupling into the fiber, a pupil apodizer to suppress unwanted starlight, a pupil plane vortex mask to enable the acquisition of spectra at and within the diffraction limit, and an atmospheric dispersion compensator. These modules, when combined with the active fiber injection unit present in phase I, will make for a highly efficient exoplanet characterization platform. In this paper, we will present the final design of the optics and opto-mechanics and highlight some innovative solutions we implemented to facilitate all the new capabilities. We will provide an overview of the assembly and laboratory testing of the sub-modules and some of the results. Finally, we will outline the deployment timeline.
Automatic Differentiation Variational Inference (ADVI) is a useful tool for efficiently learning probabilistic models in machine learning. Generally approximate posteriors learned by ADVI are forced to be unimodal in order to facilitate use of the re parameterization trick. In this paper, we show how stratified sampling may be used to enable mixture distributions as the approximate posterior, and derive a new lower bound on the evidence analogous to the importance weighted autoencoder (IWAE). We show that this SIWAE is a tighter bound than both IWAE and the traditional ELBO, both of which are special instances of this bound. We verify empirically that the traditional ELBO objective disfavors the presence of multimodal posterior distributions and may therefore not be able to fully capture structure in the latent space. Our experiments show that using the SIWAE objective allows the encoder to learn more complex distributions which regularly contain multimodality, resulting in higher accuracy and better calibration in the presence of incomplete, limited, or corrupted data.
We discuss the use of parametric phase-diverse phase retrieval as an in-situ high-fidelity wavefront measurement method to characterize and optimize the transmitted wavefront of a high-contrast coronagraphic instrument. We apply our method to correct the transmitted wavefront of the HiCAT (High contrast imager for Complex Aperture Telescopes) coronagraphic testbed. This correction requires a series of calibration steps, which we describe. The correction improves the system wavefront from 16 nm RMS to 3.0 nm RMS.
التعليقات
جاري جلب التعليقات جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
mircosoft-partner

هل ترغب بارسال اشعارات عن اخر التحديثات في شمرا-اكاديميا