Abstract:The design of astronomical hardware operating at the diffraction limit requires optimization of physical optical simulations of the instrument with respect to desired figures of merit, such as throughput or astrometric accuracy. These systems can be high dimensional, with highly nonlinear relationships between outputs and the adjustable parameters of the hardware. In this series of papers we present and apply dLux, an open-source end-to-end differentiable optical modelling framework. Automatic differentiation enables not just efficient high-dimensional optimization of astronomical hardware designs, but also Bayesian experimental design directly targeting the precision of experimental outcomes. Automatic second derivatives enable the exact and numerically stable calculation of parameter covariance forecasts, and higher derivatives of these enable direct optimization of these forecasts. We validate this method against analytic theory and illustrate its utility in evaluating the astrometric precision of a parametrized telescope model, and the design of a diffractive pupil to achieve optimal astrometric performance for exoplanet searches. The source code and tutorial software are open source and publicly available, targeting researchers who may wish to harness dLux for their own optical simulation problems.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of optimizing the design of instruments operating in high - precision astronomical observations, especially under the diffraction limit. Specifically, the authors strive to optimize the optical design by maximizing Fisher information to improve the performance of astronomical hardware (such as telescopes). The following are the main objectives and methods of the paper: 1. **Optimizing physical - optical simulations**: - The paper proposes an open - source end - to - end differentiable optical modeling framework named ∂Lux, which uses automatic differentiation techniques to efficiently optimize high - dimensional, nonlinear optical systems. - Automatic differentiation not only makes high - dimensional optimization possible but also supports Bayesian experimental design, directly optimizing for the precision of experimental results. 2. **Calculating the parameter covariance matrix**: - By automatically calculating second - order derivatives, the parameter covariance matrix can be calculated precisely and numerically stably. - Higher - order derivatives enable the direct optimization of the predicted values of these covariance matrices, thereby better estimating the uncertainty of parameters. 3. **Application examples**: - The paper validates this method and demonstrates its practicality in evaluating the astrometric precision of a parameterized telescope model. - In particular, the paper details how to design a diffractive pupil to achieve optimal astrometric performance for exoplanet searches. 4. **Specific application scenarios**: - The authors applied ∂Lux to optimize the diffractive pupil design in the Toliman mission, which is a small - scale space telescope aiming to detect low - mass planets around Alpha Centauri AB by measuring the micro - arc - second - level angular displacements of stars. - The design challenge is to construct a binary - phase mask to maximize the encoding of instrument meteorology and scientific signals while ensuring the binarity of the pupil (0 or π phase). ### Summary The core problem of this paper is to optimize the optical design by maximizing Fisher information, especially for high - precision astrometry and exoplanet detection tasks. By introducing automatic differentiation techniques and the differentiable optical modeling framework ∂Lux, the authors provide a new method to handle the complex problem of optimizing optical systems, thereby improving the precision and reliability of astronomical observations.

Differentiable Optics with dLux II: Optical Design Maximising Fisher Information

Differentiable Optics with dLux I: Deep calibration of Flat Field and Phase Retrieval with Automatic Differentiation

4-K-resolution Minimalist Optical System Design Based on Deep Learning.

Fast Method for Designing Broadband Achromatic Diffractive Optical Elements.

Elimination of Varying Chromatic Aberrations Based on Diffractive Optics

Differentiable Design of Double-Freeform Lens with Multi-Level Radial Basis Functions for Extended Source Irradiance Tailoring

Differentiable optimization of the Debye-Wolf integral for light shaping and adaptive optics in two-photon microscopy

A New Approach for Tolerance Sensitivity Reduction in Lens Optimization by Controlling the First Order Derivatives of Real Ray Height on Pupil

TorchOptics: An open-source Python library for differentiable Fourier optics simulations

End-to-End Hybrid Refractive-Diffractive Lens Design with Differentiable Ray-Wave Model

Open-Source Computational Photonics with Auto Differentiable Topology Optimization

Automatic Design of Machine Vision Lens Based on Improved Genetic Algorithm and Damped Least Squares

Designing Diffractive Optical Elements for Special Functions Using a Rigorous Method

Forecasting the power of Higher Order Weak Lensing Statistics with automatically differentiable simulations

Lens Factory: Automatic Lens Generation Using Off-the-shelf Components

Rigorous electromagnetic design of finite-aperture diffractive optical elements by use of an iterative optimization algorithm.

Automatic differentiation accelerated shape optimization approaches to photonic inverse design on rectilinear simulation grids

Toward the End-to-End Optimization of Particle Physics Instruments with Differentiable Programming: a White Paper

Joint optimization of wavefront sensing and reconstruction with automatic differentiation

Phase retrieval and design with automatic differentiation: tutorial

Optical Design and Characterization of an Advanced Computational Imaging System