REVIEW 4 minor 17 references
Characterization and Computation of Matrices of Maximal Trace over Rotations
T0 review · 0 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read A $d\times d$ matrix is of maximal trace over rotation matrices if and only if it is symmetric and has at most one negative eigenvalue, whose absolute value does not exceed the other eigenvalues.
desk verdict A clean, correct characterization of maximal-trace rotations; the alleged proof flaw doesn't survive contact with the paper. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing mechanism is a trace bound for diagonal matrices under determinant-minus-one orthogonal transformations: if $D$ is diagonal with nonnegative entries and $W$ is orthogonal with $\det(W)=-1$, then $\operatorname{tr}(WD)$ is at most the sum of all diagonal entries except the smallest, minus that smallest entry. This bound, proved as Proposition 3 and sharpened in Proposition 4, lets the sufficiency argument reduce rotation maximality to a spectrum sign count. The necessity direction uses a Givens rotation construction: whenever a matrix is not symmetric, a carefully chosen plane rotation raises the trace, so maximality forces symmetry. For $d=3$ the resulting practical test is the positive semidefiniteness of $S=\operatorname{tr}(A)I-A$.
What would settle it
Take $A=\operatorname{diag}(1,-2)$. The paper's condition says $A$ should not be maximal over rotations; indeed, rotating by $\pi$ gives $\operatorname{tr}(UA)=1>-1=\operatorname{tr}(A)$. To challenge the opposite direction, take $A=\operatorname{diag}(1,-1/2)$ and check numerically over a fine grid of rotation angles that no rotation exceeds $\operatorname{tr}(A)=1/2$. A decisive scan would randomly generate symmetric matrices, compute their eigenvalue conditions, and compare with a numerical maximization of $\operatorname{tr}(UA)$ over $\operatorname{SO}(d)$.
Extended reading notes
Core claim
The paper's central claim is a necessary-and-sufficient eigenvalue characterization of maximal trace over rotations: a real $d\times d$ matrix $A$ satisfies $\operatorname{tr}(UA)\leq \operatorname{tr}(A)$ for every rotation $U$ exactly when $A$ is symmetric and its eigenvalues contain at most one negative number, that number being no larger in absolute value than the others. The proof forces symmetry by exhibiting a Givens rotation that strictly increases the trace of any nonsymmetric matrix, then shows by an eigenvalue sign-flipping argument that a maximizing matrix cannot have two negative eigenvalues or one oversized negative eigenvalue. Conversely, a spectral bound on orthogonal matrices of determinant minus one shows that no rotation can improve the trace once the symmetric matrix satisfies the eigenvalue condition. The authors state the result as Corollary 1 and treat it as the main goal of the paper; the SVD-free computational recipes in later sections are a secondary contribution.
Load-bearing premise
The proof that the eigenvalue condition is sufficient leans on a technical bound: a reflection of a nonnegative diagonal matrix cannot raise the trace by more than twice the smallest diagonal entry, and the paper's printed derivation of the needed block structure contains a garbled determinant identity that a reader must repair. If that bound were false, the characterization would not follow.
Editorial extensions
If this is right
- A matrix passing the symmetry-plus-eigenvalue test certifies that the identity rotation solves the associated Procrustes or Wahba alignment, without any SVD being run.
- For $d=2$, the criterion is equivalent to the trace of the symmetric matrix being nonnegative, so the paper's closed-form rotation formula is optimal whenever the identity is not.
- For $d=3$, maximality is equivalent to positive semidefiniteness of $\operatorname{tr}(A)I-A$, a condition checkable by principal-minor tests before any eigenvector computation.
- When a symmetric matrix fails the test, a rotation making it maximal is obtained by flipping the signs of its negative eigenvalues in an orthogonal diagonalization, with a parity rule deciding which eigenvalue's sign is preserved (Proposition 12).
- For nonsymmetric $3\times3$ matrices, the paper's Cayley-Newton procedure finds a rotation symmetrizing $UM$; the paper notes this procedure can fail when the required rotation has angle $\pi$, leaving the SVD method as fallback.
Reading between the lines
- Because the $d=3$ test is a linear matrix inequality, membership in the set of rotation-maximal matrices is a semidefinite programming feasibility problem; the paper does not draw this connection, but it means existing SDP solvers could certify optimality in the same step that checks the condition.
- The two-dimensional closed form depends only on two scalar sums (a dot-product sum and a signed-area sum), so weighted 2D Procrustes reduces to computing those two numbers and normalizing a vector; this is a simpler pipeline than calling a general SVD routine.
- The eigenvalue characterization is a symmetry-plus-spectrum condition, which suggests that the same 'almost positive semidefinite' notion may appear in other optimization problems over compact Lie groups where the objective is a trace; exploring that analogy would go beyond this paper.
- For odd dimensions the paper proves the semidefinite test is necessary but not sufficient in general, so a complete higher-dimensional version of the $d=3$ shortcut remains open.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the problem of maximizing tr(UM) over rotation matrices U for a fixed real d x d matrix M. Its main result (Corollary 1) characterizes matrices of maximal trace over rotations: after symmetry is established, at most one eigenvalue may be negative, and if one is negative, its absolute value must not exceed that of any other eigenvalue. The proof is developed in Section 3, with sufficiency proved through Propositions 3-5 and necessity through Proposition 6. For d=2 and d=3, the characterization is turned into simple tests (nonnegative trace for d=2; positive semidefiniteness of tr(A)I-A for d=3), and the paper then proposes SVD-free computational procedures: closed-form two-dimensional formulas, a trigonometric eigensolver for symmetric 3x3 matrices, and a Cayley-transform/Newton method for nonsymmetric 3x3 matrices with an SVD fallback. The numerical procedures are implemented in Fortran and Matlab and compared in Table 1.
Significance. If correct, the main characterization is clean and practically useful: checking whether the identity rotation is optimal reduces to a symmetry test and an eigenvalue count. The sufficiency proof is self-contained and does not rely on the SVD, and the necessity argument exhaustively handles the determinant sign cases. The paper also provides reproducible code and honestly describes the failure modes of the Newton-based method, which strengthens the numerical portion. The main defect I found is a local typographical error in the determinant argument of Proposition 3; it does not affect the validity of the theorem.
minor comments (4)
- [Section 3, Proposition 3(3)] The displayed determinant chain contains a false equality: det(W^T)det(W+I) is not generally equal to det(I+W^T). The correct identity to invoke is det(I+W) = det(W)det(I+W^T) = det(W)det(I+W), which, together with det(W)=-1, gives det(I+W)=0. Once this line is corrected, the rest of the proof of Proposition 3(3) goes through unchanged.
- [Corollary 1, last sentence] The assertion that a matrix of maximal trace over rotation matrices has nonnegative trace needs the standing assumption d >= 2. For d=1, the matrix [-1] is maximal over the only rotation matrix [1] but has trace -1. Please state the dimension assumption explicitly; the proof of Proposition 2 already requires d >= 2.
- [Proposition 4] There is a typographical error in the chain tr(WD) = tr(W_hat D_hat) =<= ...; the symbol '=<' should be '=' or the words should be split into two inequalities.
- [Section 6, numerical experiments] The claim that Newton's method is successful 'close to one hundred percent' of the time, and the statement that rank-one matrices are not successfully handled, are not quantified. Please report the number of failures and the rank breakdown, since Table 1 reports a different experiment (one million random matrices, all of which are stated to have been solved).
Circularity Check
No circularity: the main characterization is derived from first principles, with Proposition 3 restated and proved in the paper.
full rationale
The paper's central claim, Corollary 1, is derived in Section 3 from first principles. Proposition 2 establishes symmetry of a maximal-trace matrix; Proposition 3, though credited to the authors' earlier paper [9], is restated with a full proof whose chain (statements 1 to 2 to 3) does not invoke the target characterization; Propositions 4 and 5 convert these trace inequalities into sufficiency; Proposition 6 handles necessity case by case. No parameter is fitted, no target-inclusive assumption is used, and no 'prediction' is constructed from the quantity it claims to predict. The only overlap with the authors' prior work is the citation [9] for Proposition 3, but because the proof is reproduced in the paper, the citation is not load-bearing. The determinant identity in the proof of Proposition 3 is algebraically valid as printed: det(W^T)det(W+I)=det(I+W^T) by multiplicativity, and det(W)=-1 makes the first equality an identity. The secondary computational sections are honestly presented as consequences of the already-proved characterization, with limitations (Newton's method may fail, fallback to the SVD method) explicitly stated. Thus there is no significant circularity.
Assumptions & free parameters
assumptions (6)
- standard math Existence of a maximizer on the compact group SO(d).
- standard math Spectral theorem for real symmetric matrices.
- standard math Every orthogonal matrix with determinant -1 has -1 as an eigenvalue.
- standard math Roots of a depressed cubic can be expressed using the trigonometric identity cos(3theta) = 4cos^3(theta) - 3cos(theta).
- standard math The Cayley transform maps skew-symmetric matrices to rotation matrices and is involutive when defined.
- standard math Newton's method converges quadratically near a nonsingular root.
Cite this review
Pith. "Pith review of Characterization and Computation of Matrices of Maximal Trace over Rotations." pith.science (2026). https://pith.science/paper/ZETTAYNZ
@misc{pith2026190808824,
author = {Pith},
title = {Pith review of: Characterization and Computation of Matrices of Maximal Trace over Rotations},
year = {2026},
howpublished = {\url{https://pith.science/paper/ZETTAYNZ}},
note = {Machine review of arXiv:1908.08824}
}
read the original abstract
The constrained orthogonal Procrustes problem is the least-squares problem that calls for a rotation matrix that optimally aligns two corresponding sets of points in d-dimensional Euclidean space. This problem generalizes to the so-called Wahba's problem which is the same problem with nonnegative weights. Given a dxd matrix M, solutions to these problems are intimately related to the problem of finding a dxd rotation matrix U that maximizes the trace of UM, i.e., that makes UM a matrix of maximal trace over rotations, and it is well known this can be achieved with a method based on the computation of the singular value decomposition (SVD) of M. As the main goal of this paper, we characterize dxd matrices of maximal trace over rotation matrices in terms of their eigenvalues, and for d = 2, 3, we show how this characterization can be used to determine whether a matrix is of maximal trace over rotation matrices. Finally, although depending only slightly on the characterization, as a secondary goal of the paper, for d = 2, 3, we identify alternative ways, other than the SVD, of obtaining solutions to the aforementioned problems.
Reference graph
Works this paper leans on
-
[9]
Lawrence J., Bernal J. and Witzgall C., A Purely Algebraic Justification of the Kabsch-Umeyama Algorithm, Journal of Research of the National Institute of Standards and Technology 124 (2019) 1-6
work page 2019
-
[1]
Bellman R., Introduction to Matrix Analysis, 2nd Edn, McGraw-Hill, New York 1970
work page 1970
-
[2]
and Hagwood C., On Computing Elastic Shape Distances Between d-Dimensional Curves, In Preparation
Bernal J., Dogan G. and Hagwood C., On Computing Elastic Shape Distances Between d-Dimensional Curves, In Preparation
-
[3]
Cayley A., Sur Quelques Propri´ et´ es des D´ eterminants Gauches,Journal f¨ ur die Reine und Angewandte Mathematik 32 (1846) 119-123. 34
-
[4]
Dogan G., Bernal J. and Hagwood C., FFT-Based Alignment of 2d Closed Curves with Application to Elastic Shape Analysis, Proceedings of the 1st DIFF-CV Workshop, British Machine Vision Conference, Swansea 2015
work page 2015
-
[5]
Eberly D., A Robust Eigensolver for 3 × 3 Symmetric Matrices, Geo- metric Tools, https://www.geometrictools.com/ 2014
work page 2014
-
[6]
and Johnson C., Matrix Analysis, Cambridge University Press, Cambridge 1985
Horn R. and Johnson C., Matrix Analysis, Cambridge University Press, Cambridge 1985
work page 1985
-
[7]
Kabsch W., A Solution for the Best Rotation to Relate Two Sets of Vectors, Acta Crystallographica Section A: Crystal Physics 32 (1976) 922-923
work page 1976
Show all 17 references
-
[8]
Kabsch W., A Discussion of the Solution for the Best Rotation to Relate Two Sets of Vectors, Acta Crystallographica Section A: Crystal Physics 34 (1978) 827-828
1978
-
[10]
and McDonald J., Linear Algebra and Its Applications, 5th Edn, Pearson Education, Boston 2016
Lay D., Lay S. and McDonald J., Linear Algebra and Its Applications, 5th Edn, Pearson Education, Boston 2016
2016
-
[11]
Markley F., Equivalence of Two Solutions of Wahba’s Problem, J. Astr. Sci. 60 (2013) 303-312
2013
-
[12]
and Mladenova C., An Approach to Reduction of the Real Symmetric 3 × 3 Secular Problem and Applications, Phys
Mladenov I. and Mladenova C., An Approach to Reduction of the Real Symmetric 3 × 3 Secular Problem and Applications, Phys. Atomic Nucl. 68 (2005) 1916-1921
2005
-
[13]
Smith O., Eigenvalues of a Symmetric 3 × 3 Matrix, Communications ACM 4 (1961) p 168
1961
-
[14]
and Klassen E., Functional and Shape Data Analysis, Springer, New York 2016
Srivastava A. and Klassen E., Functional and Shape Data Analysis, Springer, New York 2016
2016
-
[15]
Pattern Analysis and Ma- chine Intelligence 13 (1991) 376-380
Umeyama S., Least-Squares Estimation of Transformation Parameters Between Two Point Patterns, IEEE Trans. Pattern Analysis and Ma- chine Intelligence 13 (1991) 376-380
1991
-
[16]
Wahba G., A Least-Squares Estimate of Satellite Attitude, SIAM Review 7 (1965) p 409
1965
-
[17]
Their Invariants and Representation s, 2nd Edn, Princeton University Press, Princeton 1946
Weyl H., The Classical Groups. Their Invariants and Representation s, 2nd Edn, Princeton University Press, Princeton 1946. 35
1946
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.