Unscented and Higher-Order Linear Covariance Fidelity Checks and Measures of Non-Gaussianity
Jackson Kulik, Braden Hastings, Keith A. LeGrand · 2025
Read Unscented and Higher-Order Linear Covariance Fidelity Checks and Measures of Non-Gaussianity by Jackson Kulik, Braden Hastings, Keith A. LeGrand (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-23152.txt.
Introduction
Jackson Kulik, Braden Hastings, and Keith LeGrand develop ways to check whether fast uncertainty calculations for spacecraft navigation can be trusted. Linear covariance methods estimate navigation uncertainty far more cheaply than Monte Carlo simulation, which is why mission designers use them in early studies. Their weakness is the linear step itself: on strongly nonlinear trajectories with large uncertainties, the shortcut can quietly fail and mislead planning.
The authors present tools for grading linear covariance performance before it is used. Their fidelity measures draw on higher-order statistics, constrained search, and the unscented transform, each probing a different way the linear answer can depart from the true spread. The tools are aimed at preliminary spacecraft navigation studies, where catching an unreliable covariance early prevents expensive redesign later. By flagging deficient cases, the checks let engineers spend Monte Carlo effort only where the cheap method is out of its depth.
Catalog record
| CREATOR | Jackson Kulik, Braden Hastings, Keith A. LeGrand |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-23152
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Unscented and Higher-Order Linear Covariance Fidelity Checks and Measures of Non-Gaussianity Jackson Kulik, Braden Hastings, Keith A. LeGrand Abstract Linear covariance (LinCov) techniques have gained widespread traction in the modeling of uncertainty, including in the preliminary study of spacecraft navigation performance. While LinCov methods offer improved computational efficiency compared to Monte Carlo based uncertainty analysis, they inherently rely on linearization approximations. Understanding the fidelity of these approximations and identifying when they are deficient is critically important for spacecraft navigation and mission planning, especially when dealing with highly nonlinear systems and large state uncertainties. This work presents a number of computational techniques for assessing linear covariance performance. These new LinCov fidelity measures are formulated using higher-order statistics, constrained optimization, and the unscented transform. keywords Linear Covariance, Uncertainty Propagation, Non-Gaussianity 1 Introduction Uncertainty propagation in astrodynamics underlies satellite navigation, space situational awareness, and robust/uncertainty-aware trajectory optimization. Often in these applications, theoretical and practical constraints motivate the usage of linearization and/or Gaussian assumptions in the approximation of propagated uncertainty. These constraints include limited computational resources in autonomous systems and, in other cases, the need for semi-analytical or convex formulations of uncertainty propagation. lincov ( lincov ) analysis is based around the same principles as the prediction step in the extended Kalman filter and employs first-order Taylor approximations of the dynamics to obtain a closed-form covariance propagation carpenter2025navigation ; geller2006linear . The validity of this approach relies on the approximate linearity of the dynamics within the effective support of the distribution, which is commonly assumed to be Gaussian. Other uncertainty propagation methods that handle weak nonlinearity while assuming Gaussianity of the distribution include the unscented transform julier2004unscented . Still other nonlinear non-Gaussian uncertainty propagation methods include Monte Carlo, conjugate unscented transform adurthi2012conjugate , state transition tensor-based moment propagation park2006nonlinear , Gaussian mixture propagation alspach1972NonlinearBayesianEstimation ; demars2013entropy , differential algebra servadio2022maximum , and polynomial chaos expansion approaches jones2013nonlinear . Given lincov ’s simplicity, speed, and overall adoption by practitioners, it is likely to still play a large role in future spacecraft mission planning. Thus, it is important to establish fast fidelity checks to accompany this technique to indicate whether the outputs are trustworthy or that more sophisticated nonlinear uncertainty propagation methods are needed. Thus, two related questions arise in measuring the fidelity of a lincov approximation: how non-Gaussian is a propagated distribution, and how nonlinear are the dynamics on the effective support of the distribution? To address the question of quantifying non-Gaussianity, Mercurio et al. demonstrated how moment-based approaches, information theoretic approaches including kl ( kl )-divergence, and shape-based metrics such as the Wasserstein distance can be employed to assess the non-Gaussianity of a distribution mercurio2018HowNongaussian . On the other hand, Junkins presented a methodology for assessing the degree of nonlinearity of the flow of a dynamical system. junkins2004nonlinear Recent work has improved on methods to calculate that nonlinearity index using stt jenson2023semianalytical ; kulik2024applications . Other recent work studied measures of linearized uncertainty realism through the lens of the nonlinearity index gutierrez2024classifying . Work on splitting in adaptive Gaussian mixture uncertainty propagation has considered uncertainty weighted measures of nonlinearity kulik2025NonlinearityUncertaintySplitting ; zanetti2025uncertainty ; calkins2024efficient ; calkins2025dynamics ; siciliano2025higher ; iannamorelli2025AdaptiveGaussianMixture . Additionally, recent work on covariance steering for robust/uncertain trajectory optimization has enforced the fidelity of the approximations employed in computations by leveraging nonlinearity indices in a statistical context fife-steering ; qi2025optimal . Other approaches have quantified non-Gaussianity in terms of skewness in Gaussian mixture splitting dunik2018directional ; horwood2011adaptive and covariance steering settings qioptimal often employing a scalar summary metric of overall skewness using tensor methods from higher-order statistics mccullagh2018tensor . This work summarizes, extends, and formalizes some of these approaches to uncertainty, nonlinearity, and non-Gaussianity quantification. In particular, this work presents a number of novel moment-based and constrained optimization-based approaches for quantifying both non-Gaussianity and the fidelity of the linearization assumption in lincov analysis. Where possible, we present multiple methods for calculating each metric that are based either on state transition tensor or unscented transform methodologies. Using higher-order approximations of the mean and covariance, we present multiple methods for assessing the approximation quality of linear approximation-based moment propagation. These measures are based on expected Mahalanobis distances and maximal differences in the marginal variances between two fidelities of approximation. In the area of higher-order statistics we present a new connection between standard multivariate Gaussian kurtosis and the fourth-order identity tensor. Based on that observation, we offer a new compact definition of the excess kurtosis tensor that offers a novel method of standardization for the multivariate case. We then present and demonstrate how to compute novel scalar summary statistics for the maximal standardized directional skewness and kurtosis. The remainder of this article is arranged as follows. Background information on partial derivative tensors, whitening transformations, and statistical moments is given in Section 2 . Section 3 presents novel expectation-based measures of fidelity and their Monte Carlo, unscented, and second-order approximations. Fidelity measures based on constrained optimization problems are discussed in 4 . Section 5 demonstrates the application of these methods to a cislunar orbit lincov analysis, and conclusions are drawn in Section 6 . 2 Background Consider a nonlinear function \mathbf{g}:\mathbb{R}^{n}\rightarrow\mathbb{R}^{m} that may represent the flow of a dynamical system or a measurement function. Let \mathbf{x}\in\mathbb{R}^{n} be a random variable with mean \bm{\mu}_{x} and covariance \mathbf{P}_{x} . Linear covariance techniques rely on the following simple principle for uncertainty propagation: for sufficiently small input covariance \mathbf{P}_{x} , the resulting output mean and covariance are approximately \displaystyle\textrm{E}[\mathbf{z}]\approx\bm{\mu}_{z}^{(1)} \displaystyle=\mathbf{g}(\bm{\mu}_{x}) (1) \displaystyle\mathbf{P}_{z}\approx\mathbf{P}_{z}^{(1)} \displaystyle=\mathbf{G}\mathbf{P}_{x}\mathbf{G}^{T} (2) where the superscript “ (1) ” denotes that these quantities arise from linear covariance analysis and \mathbf{G}=\frac{\mathrm{d}\mathbf{g}}{\mathrm{d}\mathbf{x}}\Big|_{\bm{\mu}_{x}} (3) is the Jacobian of \mathbf{g} evaluated at \bm{\mu}_{x} . Furthermore, if \mathbf{x} is Gaussian-distributed, then by the same aforementioned assumptions, \mathbf{z} is also approximately Gaussian distributed. Often, linear covariance techniques are employed for stochastic systems that include process noise or measurement noise. In this work, we will examine only deterministic systems, where the dominating source of output uncertainty is due the initial uncertainty in \mathbf{x} . This paper develops new methods and metrics for analyzing the quality of LinCov-estimated quantities in nonlinear systems. The methods in this work are based primarily on calculating expectations related to linearization error or solving constrained optimization problems related to linearization error. In both cases, the proposed methods leverage higher-order statistics or the unscented transform. For multivariate random variables, central moments beyond order two require tensor representations. Thus, our constrained optimization approach leverages the theory of tensor eigenvalues and tensor operator norms jenson2023semianalytical ; kulik2024applications . One of the most important elements of our approach is output-whitening, which enables each output to be meaningfully interpretable as a Mahalanobis distance and is unit and scale-invariant. 2.1 Partial Derivative Tensors In this work, we will heavily rely on higher-order partial derivative tensors of the nonlinear function \mathbf{g} in order to describe the higher-order terms in the Taylor series expansion of \mathbf{g} about the mean of \bm{\mu}_{x} as these will approximately describe the error in approximating \mathbf{g}(\bm{\mu}_{x}+\delta\mathbf{x}) by its linearization \mathbf{g}(\bm{\mu}_{x})+\mathbf{G}\delta\mathbf{x} . The Taylor series expansion of \mathbf{g} about the mean \bm{\mu}_{x} is given by \mathbf{g}(\bm{\mu}_{x}+\delta\mathbf{x})=\mathbf{g}(\bm{\mu}_{x})+\mathbf{G}\delta\mathbf{x}+\frac{1}{2}\frac{\partial^{2}\mathbf{g}}{\partial\mathbf{x}^{2}}\Big|_{\bm{\mu}_{x}}\delta\mathbf{x}^{2}+\mathcal{O}(\delta\mathbf{x}^{3}) (4) where we define the shorthand notation above for double contraction of the tensor with two copies of a vector