arxiv: 2605.05560 · v1 · submitted 2026-05-07 · 📡 eess.SP

Recognition: unknown

Covariance Square Root Second-Order Mapping

Keith A. LeGrand , Braden Hastings , Jackson Kulik

Authors on Pith no claims yet

Pith reviewed 2026-05-08 07:36 UTC · model grok-4.3

classification 📡 eess.SP

keywords square root covariancesecond-order mappingstate estimationnumerical stabilitynonlinear filteringcovariance factorizationrecursive estimation

0 comments

The pith

A direct square root computation of the second-order covariance mapping improves accuracy and often cuts operations in nonlinear state estimation.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a square root version of the second-order covariance mapping for recursive state estimation. Roundoff errors in finite-precision arithmetic can break symmetry or positive-semidefiniteness in covariance matrices when higher-order mappings are used. Square root factorizations have long helped first-order methods like the extended Kalman filter avoid these problems. This work extends the same idea to second-order mappings, which are valuable for strongly nonlinear dynamics. Two numerical experiments show the new square-root form is more accurate and typically requires fewer floating-point operations than forming the full covariance matrix.

Core claim

The paper presents the first known square root computation of the second-order covariance mapping. This computation preserves the numerical advantages of square root factorizations while handling the second-order moment mapping that appears in highly nonlinear estimation problems.

What carries the argument

The square root factorization applied directly to the second-order covariance mapping, avoiding explicit formation of the full matrix.

If this is right

State estimates remain more reliable under finite-precision arithmetic when second-order mappings are employed.
Algorithms using higher-order moments can inherit the same numerical safeguards previously limited to first-order filters.
Computational cost decreases in many cases because the square-root form avoids some matrix multiplications required for the full covariance.
Positive semidefiniteness is maintained by construction rather than enforced after the fact.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The technique could be inserted into existing square-root Kalman filter code bases with minimal changes to the propagation step.
Similar square-root constructions might be derived for third- or higher-order moment mappings if the algebraic pattern generalizes.
Applications with strict numerical requirements, such as onboard spacecraft navigation, would be natural test cases for the method.

Load-bearing premise

A stable square-root factorization of the second-order moment mapping exists and can be computed without introducing new instabilities beyond those already present in first-order square-root filters.

What would settle it

Run the same two numerical experiments and observe that the square-root version produces a non-positive-semidefinite matrix or higher estimation error than the full-matrix version.

Figures

Figures reproduced from arXiv: 2605.05560 by Braden Hastings, Jackson Kulik, Keith A. LeGrand.

**Figure 1.** Figure 1: A uniform weighting factor is also associated with this geometry allowing for easy analytical solution of the view at source ↗

read the original abstract

In recursive state estimation, numerical error can play a major role in an algorithm's overall performance and reliability. Roundoff errors due to finite precision arithmetic can violate theoretical guarantees, leading to asymmetric and non-positive-semidefinite covariance matrices. In algorithms employing first-order covariance mappings, such as the extended Kalman filter, these issues have been mitigated by employing square root factorizations of the covariance matrix. However, existing techniques do not directly extend to higher-order moment mappings, which show great value in highly nonlinear settings. This paper presents the first known square root computation of the second-order covariance mapping. The square root computation is not only more accurate, as is shown in two distinct numerical experiments, but generally requires fewer floating point operations compared to the full covariance matrix computation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper gives the first square-root factorization for the second-order covariance update in nonlinear recursive estimation.

read the letter

The core advance is a direct algebraic square-root form for the second-order moment mapping that avoids forming the full covariance matrix at each step. This extends the well-known first-order square-root techniques without obvious circularity or re-packaging of old results. The two numerical experiments show measurable gains in accuracy and a reduction in floating-point operations, which lines up with the practical motivation around roundoff errors in finite-precision arithmetic. That part is useful and worth having on record for people who already use second-order filters in highly nonlinear settings. The derivation itself looks like straightforward matrix algebra once the second-order terms are written out, and the abstract does not hide any load-bearing assumptions beyond those already common in the first-order case. The main soft spot is that stability of the factorization is only demonstrated experimentally rather than bounded analytically, so edge cases with near-singular Jacobians or higher noise levels could still produce surprises. The efficiency claim is also qualified as “generally,” which will depend on the specific implementation of the square-root operations. Overall this is incremental but cleanly executed work aimed at a narrow but real pain point in control and robotics estimation. It is worth a serious referee who can check the algebra and ask for a bit more on the stability question; the experiments already give enough evidence that the idea is not obviously broken.

Referee Report

0 major / 1 minor

Summary. The paper claims to present the first square-root computation of the second-order covariance mapping for recursive state estimation. It asserts that this approach is more accurate than direct full-matrix computation (as shown in two numerical experiments) and generally requires fewer floating-point operations, addressing roundoff-induced violations of positive-semidefiniteness in higher-order nonlinear filters.

Significance. If the derivation and stability claims hold, the work would usefully extend square-root techniques beyond first-order mappings (e.g., EKF) to second-order moment mappings that are valuable in highly nonlinear regimes. The reported accuracy gains and FLOP reduction, if reproducible, would be a practical contribution to numerically robust filtering implementations.

minor comments (1)

[Abstract] The abstract refers to 'two distinct numerical experiments' without naming the systems, noise levels, or metrics used; adding one sentence of context would help readers gauge the scope of the validation.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for reviewing our manuscript and for the positive assessment of its potential contribution to numerically robust filtering. The referee summary correctly identifies the core claim: the first square-root formulation of the second-order covariance mapping. No specific major comments were enumerated in the report, so we provide no point-by-point rebuttals. We remain available to supply additional derivation details or numerical evidence if that would resolve the current 'uncertain' recommendation.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained algebraic extension

full rationale

The paper derives a square-root factorization for the second-order covariance mapping as a direct algebraic extension of first-order square-root techniques. The abstract and structure present this as a new computation, validated by two numerical experiments comparing accuracy and FLOPs against the full matrix form. No load-bearing steps reduce to self-citation chains, fitted parameters renamed as predictions, or self-definitional mappings. The central result is not equivalent to its inputs by construction; it introduces an independent factorization whose stability is asserted under the same assumptions as prior first-order work, without circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method relies on standard linear-algebra properties of covariance matrices and square-root factorizations already established in the Kalman-filter literature; no new free parameters, axioms, or invented entities are introduced in the abstract.

axioms (1)

standard math Covariance matrices admit stable square-root factorizations that preserve positive-semidefiniteness under finite-precision arithmetic.
Invoked implicitly when extending first-order square-root methods to second-order mappings.

pith-pipeline@v0.9.0 · 5419 in / 1075 out tokens · 41804 ms · 2026-05-08T07:36:32.358396+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

21 extracted references · 14 canonical work pages

[1]

H.,Astronautical Guidance, McGraw-Hill, New York, 1964, Chap

Battin, R. H.,Astronautical Guidance, McGraw-Hill, New York, 1964, Chap. 9

1964
[2]

S.,Stochastic Models, Estimation and Control, Academic Press, New York, 1979, Mathematics in Science and Engineering, Vol

Maybeck, P. S.,Stochastic Models, Estimation and Control, Academic Press, New York, 1979, Mathematics in Science and Engineering, Vol. 1, Chap. 7

1979
[3]

Computational Techniques in Kalman Filtering,

Schmidt, S. F., “Computational Techniques in Kalman Filtering,”Theory and Applications of Kalman Filtering, Vol. AGARD-AG-139, Technical Editing and Reproduction Ltd., 1970

1970
[4]

Fast Triangular Formulation of the Square Root Filter

Carlson, N. A., “Fast Triangular Formulation of the Square Root Filter.”AIAA Journal, Vol. 11, No. 9, 1973, pp. 1259–1265. https://doi.org/10.2514/3.6907

work page doi:10.2514/3.6907 1973
[5]

The Square-Root Unscented Kalman Filter for State and Parameter-Estimation,

Van der Merwe, R., and Wan, E., “The Square-Root Unscented Kalman Filter for State and Parameter-Estimation,”2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.01CH37221), Vol. 6, 2001, pp. 3461–3464. https://doi.org/10.1109/ICASSP.2001.940586

work page doi:10.1109/icassp.2001.940586 2001
[6]

Square-Root Consider Filters with Hyperbolic Householder Reflections,

McCabe, J. S., and DeMars, K. J., “Square-Root Consider Filters with Hyperbolic Householder Reflections,”Journal of Guidance, Control, and Dynamics, 2018, pp. 1–14. https://doi.org/10.2514/1.G003417

work page doi:10.2514/1.g003417 2018
[7]

Nonlinear mapping of Gaussian statistics: theory and applications to spacecraft trajectory design,

Park, R. S., and Scheeres, D. J., “Nonlinear mapping of Gaussian statistics: theory and applications to spacecraft trajectory design,”Journal of Guidance, Control, and Dynamics, Vol. 29, No. 6, 2006, pp. 1367–1375. https://doi.org/10.2514/1.2017

work page doi:10.2514/1.2017 2006
[8]

Higher-Order Tensor-Based Deferral of Gaussian Splitting for Orbit Uncertainty Propagation,

Siciliano, G. A., LeGrand, K. A., and Kulik, J., “Higher-Order Tensor-Based Deferral of Gaussian Splitting for Orbit Uncertainty Propagation,”IEEE Transactions on Aerospace and Electronic Systems, Vol. 62, 2026, pp. 5352–5368. https://doi.org/10.1109/TAES.2026.3662318

work page doi:10.1109/taes.2026.3662318 2026
[9]

Fast Second-Order Covariance Propagation Based on the Canonical Polyadic Decomposition,

Hastings, B., and Kulik, J., “Fast Second-Order Covariance Propagation Based on the Canonical Polyadic Decomposition,” Proceedings of the 2025 AIAA/AAS Astrodynamics Specialist, AIAA/AAS, Boston, MA, 2025

2025
[10]

The ASA’ s Statement on p-Values: Context, Process, and Purpose.The American Statisti- cian2016;70(2):129–133

Kessy, A., Lewin, A., and Strimmer, K., “Optimal whitening and decorrelation,”The American Statistician, Vol. 72, No. 4, 2018, pp. 309–314. https://doi.org/10.1080/00031305.2016.1277159. 13

work page doi:10.1080/00031305.2016.1277159 2018
[11]

The Effects of Rounding Error on an Algorithm for Downdating a Cholesky Factorization,

Stewart, G. W., “The Effects of Rounding Error on an Algorithm for Downdating a Cholesky Factorization,”IMA Journal of Applied Mathematics, Vol. 23, No. 2, 1979, pp. 203–213. https://doi.org/10.1093/imamat/23.2.203

work page doi:10.1093/imamat/23.2.203 1979
[12]

H., and Van Loan, C

Golub, G. H., and Van Loan, C. F.,Matrix computations, JHU press, Baltimore, MD, 2013, Chap. 5

2013
[13]

doi: 10.1016/j.jsc.2005.05.007

Qi, L., “Eigenvalues of a real supersymmetric tensor,”Journal of Symbolic Computation, Vol. 40, No. 6, 2005, pp. 1302–1324. https://doi.org/10.1016/j.jsc.2005.05.007

work page doi:10.1016/j.jsc.2005.05.007 2005
[14]

Shifted power method for computing tensor eigenpairs,

Kolda, T. G., and Mayo, J. R., “Shifted power method for computing tensor eigenpairs,”SIAM Journal on Matrix Analysis and Applications, Vol. 32, No. 4, 2011, pp. 1095–1124. https://doi.org/10.1137/100801482

work page doi:10.1137/100801482 2011
[15]

Higham.Functions of Matrices: Theory and Computation

Higham, N. J.,Functions of Matrices: Theory and Computation, Society for Industrial and Applied Mathematics, 2008, pp. 335–337. https://doi.org/10.1137/1.9780898717778

work page doi:10.1137/1.9780898717778 2008
[16]

Kolda and Brett W

Kolda, T. G., and Bader, B. W., “Tensor decompositions and applications,”SIAM review, Vol. 51, No. 3, 2009, pp. 455–500. https://doi.org/10.1137/07070111X

work page doi:10.1137/07070111x 2009
[17]

Analysis of individual differ- ences in multidimensional scaling via an n-way generalization of “Eckart-Young

Carroll, J. D., and Chang, J.-J., “Analysis of individual differences in multidimensional scaling via an N-way generalization of “Eckart-Young” decomposition,”Psychometrika, Vol. 35, No. 3, 1970, pp. 283–319. https://doi.org/10.1007/BF02310791

work page doi:10.1007/bf02310791 1970
[18]

Foracchia, A

Faber, N. K. M., Bro, R., and Hopke, P. K., “Recent developments in CANDECOMP/PARAFAC algorithms: a critical review,”Chemometrics and Intelligent Laboratory Systems, Vol. 65, No. 1, 2003, pp. 119–137. https://doi.org/10.1016/S0169- 7439(02)00089-8

work page doi:10.1016/s0169- 2003
[19]

Numerical optimization for symmetric tensor decomposition,

Kolda, T. G., “Numerical optimization for symmetric tensor decomposition,”Mathematical Programming, Vol. 151, No. 1, 2015, pp. 225–248. https://doi.org/10.1007/s10107-015-0895-0

work page doi:10.1007/s10107-015-0895-0 2015
[20]

Tensor Toolbox for MATLAB,

Kolda, T. G., and Bader, B. W., “Tensor Toolbox for MATLAB,” , 2025. URL www.tensortoolbox.org

2025
[21]

A Numerical Comparison of Discrete Kalman Filtering Algorithms: An Orbit Determination Case Study,

Thornton, C., and Bierman, G., “A Numerical Comparison of Discrete Kalman Filtering Algorithms: An Orbit Determination Case Study,” Tech. Rep. JPL-TM-33-771;, 1976. 14

1976