REVIEW 3 minor 27 references
Solving the problem of simultaneous diagonalization of complex symmetric matrices via congruence
T0 review · 0 major / 3 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The paper proves that a finite set of complex symmetric matrices is simultaneously diagonalizable by congruence exactly when its common kernel has the maximal possible dimension and the reduced matrices $L_j$ are simultaneously…
desk verdict A rigorous, self-contained solution to a long-standing problem in matrix analysis; worth a serious referee despite a few minor blemishes. 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 object is the linear matrix pencil $A(\lambda) = \sum_{j=1}^m \lambda_j A_j$ and its maximum rank $r = \max_\lambda \operatorname{rank} A(\lambda)$. A first lemma shows the common kernel of the $A_j$ sits inside the kernel of $A(\lambda_0)$ for a maximal point $\lambda_0$, and the two coincide exactly when $\dim(\bigcap_j \ker A_j) = n - r$. This equality is what allows Lemma 10 to compress the matrices by congruence to $\tilde{A}_j \oplus 0_{n-r}$, with $\tilde{A}_j$ of size $r$ and the reduced pencil $\tilde{A}(\lambda_0)$ invertible. Then Theorem 7 does the main work: for a nonsingular pencil, $P^T A(\lambda_0)^{-1} A_j P$ diagonalizes by similarity exactly when $P^T A_j P$ diagonalizes by congruence, using the identity $(P^T A(\lambda)P)(P^{-1}A(\lambda)^{-1}A_j P) = P^T A_j P$ and a blockwise diagonalization of the symmetric matrix $B(\lambda_0)$. The reduced matrices $L_j = \tilde{A}(\lambda_0)^{-1} \tilde{A}_j$ inherit the property $\sum_j (\lambda_0)_j L_j = I_r$, so only $m-1$ pairwise commutation checks are needed.
What would settle it
Search over 2-by-2 and 3-by-3 complex symmetric pairs $(A_1,A_2)$: if any pair satisfies $\dim(\ker A_1 \cap \ker A_2) = n - r$ and $L_2 = \tilde{A}(\lambda_0)^{-1} \tilde{A}_2$ is diagonalizable, but no nonsingular $P$ makes both $P^T A_1 P$ and $P^T A_2 P$ diagonal, then Theorem 14 is false.
Extended reading notes
Core claim
The central theorem states that complex symmetric matrices $A_1,\ldots,A_m$ with maximum pencil rank $r$ are SDC if and only if $\dim(\bigcap_j \ker A_j) = n - r$ and the reduced matrices $L_j = \tilde{A}(\lambda_0)^{-1} \tilde{A}_j$ are SDS (simultaneously diagonalizable via similarity), where $\lambda_0$ is any point where the pencil $A(\lambda) = \sum_j \lambda_j A_j$ attains its maximum rank. The kernel condition is necessary: under SDC the common kernel must be exactly the kernel of the pencil at a maximal point, of dimension $n - r$. When it holds, the matrices compress by congruence to $\tilde{A}_j \oplus 0_{n-r}$ with invertible reduced pencil, and the main transfer theorem converts diagonalization by congruence of the $\tilde{A}_j$ into diagonalization by similarity of the $L_j$. Combined with the classical criterion that a family is SDS iff its members pairwise commute and each is diagonalizable, this yields a three-step decision procedure. The proof of the converse direction leans on the standard factorization that every complex symmetric block can be diagonalized by a unitary congruence with real nonnegative diagonal entries.
Load-bearing premise
The converse direction of the main theorem assumes the classical factorization that every complex symmetric matrix can be diagonalized by a unitary congruence to a real nonnegative diagonal matrix; if that standard result were false, the constructed congruence in the proof would not exist.
Editorial extensions
If this is right
- Any finite set of complex symmetric matrices can be decided in finitely many steps: compute $r$, test the kernel dimension, then test the reduced matrices for pairwise commutation and individual diagonalizability.
- The complex SDC problem for arbitrarily many matrices is thereby reduced to the classical similarity problem, for which a simple pairwise test exists.
- The criterion extends earlier results that handled only pairs or required at least one nonsingular matrix; the kernel reduction removes the nonsingularity restriction.
- In the motivating application, an algebra is an evolution algebra exactly when its structure matrices pass this SDC test, giving a finite criterion for recognizing evolution algebras.
- For exact blind source separation, the result identifies precisely when a set of measured second-characteristic-function matrices can be jointly diagonalized to recover the sources.
Reading between the lines
- A natural next step is to turn the decision procedure into a numerical algorithm; the main computational bottleneck the paper leaves open is an efficient way to locate a point $\lambda_0$ where the pencil attains its maximum rank.
- For approximate joint diagonalization, the kernel condition $\dim(\bigcap_j \ker A_j) = n - r$ suggests that cost functions should penalize or exploit the common kernel explicitly, something the ad-hoc cost functions mentioned in the paper do not do.
- The same reduction may extend to other settings, such as Hermitian matrices under *-congruence, where an analogous maximal-rank point and kernel reduction would need a replacement for the unitary factorization step.
- In the real case, Theorem 14 would need the additional constraint that the eigenvectors and eigenvalues of the reduced matrices $L_j$ be real; the paper notes this but does not develop the real criterion.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper characterizes simultaneous diagonalization by congruence (SDC) of a finite set of complex symmetric matrices. After defining the linear pencil A(λ)=Σ λ_j A_j and its maximum rank r, the authors prove in Theorem 14 that A_1,...,A_m are SDC if and only if dim(∩ ker A_j)=n−r and the reduced r×r matrices L_j = \tilde A(λ0)^{-1}\tilde A_j, obtained after a kernel reduction, are simultaneously diagonalizable by similarity (SDS). Since SDS is equivalent by Theorem 3 to pairwise commutation plus individual diagonalizability, the criterion is checkable in finitely many steps. The proof proceeds through a nonsingular-pencil case (Theorem 7, using Takagi factorization), a diagonal-case kernel reduction (Lemma 8), and a general kernel reduction (Lemmas 9 and 10); two examples illustrate the procedure, including a case that is not SDC.
Significance. The result is a complete solution to a long-standing question and provides a clean, externally checkable criterion: the SDC problem is reduced to the classical SDS problem, whose own criterion is pairwise commutation and diagonalizability. The derivation is rigorous and self-contained modulo standard matrix-analysis theorems, uses no fitted parameters, and is not circular. The finite-step procedure and the worked examples make the criterion concrete, and the applications to evolution algebras and blind source separation are plausible. The paper does not provide a complexity analysis or a numerical algorithm for finding a maximizing λ0, but the central mathematical characterization is sound.
minor comments (3)
- [Theorem 7, proof after Eq. (3.3)] The block-decomposition argument constructs n1 from the first run of identical diagonal entries of D(j). If all D(j) are scalar multiples of the identity, then p_j=n for every j and the quantity α(j)_2 = α^j_{n1+1} is undefined. This endpoint case is trivial (take d=1 and diagonalize B(λ0) directly), but it should be stated explicitly so that the proof covers all cases.
- [§3.3, step (2), and Definition 5] The procedure requires λ0 ∈ C^m with rank A(λ0)=r but does not explain how such a point is to be obtained. Because the maximum rank is attained on a nonempty Zariski-open set, a generic choice works; adding one sentence to that effect, or an algebraic elimination method, would make the advertised 'finite number of steps' claim precise. Section 4's note that an efficient method is future work is acceptable, but the gap between procedure and algorithm should be acknowledged in §3.3.
- [Remarks 11] The remark is incorrect as stated: orthogonality of columns with respect to the bilinear form ⟨z,w⟩=z·w gives Q^T Q=I, i.e., Q is complex orthogonal, not unitary in the usual sense Q^*Q=I. Moreover, such an orthonormal basis for this form need not exist when the common kernel is totally isotropic. The remark is not used in any proof, so the central conclusions are unaffected, but it should be corrected or deleted.
Circularity Check
No significant circularity: the SDC-to-SDS reduction is proved from independent linear-algebra facts.
full rationale
The paper's central claim, Theorem 14, is a genuine reduction rather than a tautology. SDC is independently defined in Definition 1, and the proof establishes equivalence with two checkable conditions: the kernel-dimension condition dim(∩ ker A_j) = n − r and the SDS condition on the reduced matrices L_j = ~A(λ0)^−1 ~A_j. The forward direction uses Theorem 9 to produce P^T A_j P = ~D_j ⊕ 0_{n−r} and then shows explicitly that S^{−1} L_j S is diagonal for an invertible S derived from the congruence. The converse invokes Lemma 10 to reduce to the nonsingular pencil case and then Theorem 7, which converts SDS of the reduced pencil into SDC. Theorem 7's converse relies on Takagi's factorization, cited to Horn and Johnson [12, Cor. 2.6.6(a)], a standard external theorem, to diagonalize the complex symmetric blocks. No fitted parameter is introduced, no property is defined in terms of the desired conclusion, and the SDS criterion itself is checked by pairwise commutation and diagonalizability via the independent classical Theorem 3. The only self-citation, [4], appears in the introduction as an application motivation and is not load-bearing in the proof of Theorem 14. Remark 11 contains a questionable claim about choosing Q unitary, but that remark is not used in any proof and does not affect the derivation. Overall, the reduction chain is self-contained against standard matrix-analysis results, so no circularity is present.
Assumptions & free parameters
assumptions (3)
- standard math Every complex symmetric matrix can be diagonalized by a unitary congruence with real nonnegative diagonal entries (Takagi factorization).
- standard math A family of matrices is simultaneously diagonalizable by similarity if and only if all members are diagonalizable and they pairwise commute.
- standard math The rank of the pencil A(lambda) = sum lambda_j A_j attains its maximum over C^m, so a lambda0 with maximum rank r exists.
Cite this review
Pith. "Pith review of Solving the problem of simultaneous diagonalization of complex symmetric matrices via congruence." pith.science (2026). https://pith.science/paper/XVHYLXS6
@misc{pith2026190804228,
author = {Pith},
title = {Pith review of: Solving the problem of simultaneous diagonalization of complex symmetric matrices via congruence},
year = {2026},
howpublished = {\url{https://pith.science/paper/XVHYLXS6}},
note = {Machine review of arXiv:1908.04228}
}
abstract
We provide a solution to the problem of simultaneous $diagonalization$ $via$ $congruence$ of a given set of $m$ complex symmetric $n\times n$ matrices $\{A_{1},\ldots,A_{m}\}$, by showing that it can be reduced to a possibly lower-dimensional problem where the question is rephrased in terms of the classical problem of simultaneous $diagonalization$ $via$ $similarity$ of a new related set of matrices. We provide a procedure to determine in a finite number of steps whether or not a set of matrices is simultaneously diagonalizable by congruence. This solves a long standing problem in the complex case.
Reference graph
Works this paper leans on
-
[1]
Afsari B., Sensitivity Analysis for the Problem of Matri x Joint Diagonalisation, SIAM J. Matrix Anal. Appl., 30(3), (2008), 1148–1171
work page 2008
-
[2]
Becker, R.I. , Necessary and sufficient conditions for the simultaneous diagonability of two quadratic forms, Linear Algebra and its Applications, 30, ( 1980), 129–139
work page 1980
-
[3]
and Moul ines E., A blind source separation technique using second-order statistics
Belouchrani A., Abed-Meraim, K., Cardoso J.-F. and Moul ines E., A blind source separation technique using second-order statistics. IEEE Transactio ns on signal processing, 45(2) (1997), 434–444
work page 1997
-
[4]
Bustamante, M. D., Mellon P. and Velasco M. V., Determini ng when an algebra is an evolution algebra. Mathematics, 8 (2020), 1349
work page 2020
-
[5]
IEE Proc-F (Radar and Signal Process.), 140(6), (1993), 362–370
Cardoso J.-F., Souloumiac A., Blind beamforming for non -Gaussian signals. IEE Proc-F (Radar and Signal Process.), 140(6), (1993), 362–370
work page 1993
-
[6]
In: Adali T., Jutten C., Romano J.M.T., Ba rros A.K
Pham D.T., Congedo M., Least Square Joint Diagonalisati on of Matrices under an Intrinsic Scale Constraint. In: Adali T., Jutten C., Romano J.M.T., Ba rros A.K. (eds) Independent Component Analysis and Signal Separation. ICA (2009). Lect ure Notes in Computer Science, vol 5441. Springer, Berlin, Heidelberg
work page 2009
-
[7]
Hiriart-Urruty J. B., Potpourri of conjectures and open questions in nonlinear analysis and optimisation, SIAM Rev., 49, (2007), 255–273
work page 2007
-
[8]
Hiriart-Urruty, J.B., Malick, J., A Fresh Variational- Analysis Look at the Positive Semidefi- nite Matrices W orld, J Optim Theory Appl, 153 (3), (2012), 55 1-577
work page 2012
Show all 27 references
-
[9]
Quadrat- icW orld
J. B. Hiriart-Urruty and M. Torki, Permanently Going Bac k and Forth between the “Quadrat- icW orld” and the “ConvexityW orld” in Optimization, Appl Math Optim, 45, (2002), 169–184
2002
-
[10]
P., Horn R
Hong Y. P., Horn R. A., Johnson C. R., On the reduction of p airs of hermitian or symmetric matrices to diagonal form by congruence, Linear Algebra and its Applications, 73, (1986), 213–226
1986
-
[11]
Hong Y.P., Horn R.A., On simultaneous reduction of fami lies of matrices to triangular or diagonal form by unitary congruences, Linear and Multiline ar Algebra, 17:3-4, (1985), 271– 288
1985
-
[12]
A.; Johnson, C
Horn, R. A.; Johnson, C. R., Matrix Analysis, second edi tion. Cambridge University Press. (2013)
2013
-
[13]
Hsia Y., Lin G.X., Sheu R.L., A revisit to quadractic pro gramming with one inequality Quadratic Constraint via Matrix Pencil, Pacific Journal of O ptimization, 10, (2014), 461- 481
2014
-
[14]
Jiang R., Li D., Simultaneous Diagonalisation of Matri ces and Its Applications in Quadrati- cally Constrained Quadratic Programming. SIAM J. Optim., 2 6, (2016), 1649-1669
2016
-
[15]
Matrix Anal
De Lathauwer L., A link between the canonical decomposi tion in multilinear algebra and simultaneous matrix diagonalization, SIAM J. Matrix Anal. Appl., 28(3), (2006), 642–666
2006
-
[16]
T., Joint Approximate Diagonalisation of Posit ive Definite Matrices
Pham D. T., Joint Approximate Diagonalisation of Posit ive Definite Matrices. SIAM. J. Matrix Anal. Appl., 22 (4), (2001), 1136–1152. 14 MIGUEL D. BUSTAMANTE, PAULINE MELLON, AND M. VICTORIA VEL ASCO
2001
-
[17]
Equipe SIGNAL - Pˆ ole SIS - F´ evrier 2012, 29 pages
Sorensen M., Comon P., A Pair Sweeping Method for some Si multaneous Matrix Diagonali- sation. Equipe SIGNAL - Pˆ ole SIS - F´ evrier 2012, 29 pages
2012
-
[18]
Tian J. P. & Vojtechovsky P., Mathematical concepts of e volution algebras in non-mendelian genetics, Quasigroup and Related Systems, 24, (2006), 111- 122
2006
-
[19]
P., Evolution algebras and their applications
Tian J. P., Evolution algebras and their applications. Lecture Notes in Mathematics, vol. 1921, Springer-Verlag (2008)
2008
-
[20]
IEEE Trans
Tichavsky P., Yeredor A., Fast Approximate Joint Diago nalisation Incorporating W eight Matrices. IEEE Trans. Sig Process. 57(3), (2009). 878–891
2009
-
[21]
Uhlig, F., Simultaneous block diagonalization of two r eal symmetric matrices, Linear Algebra Appl., 7, (1973): 281-289
1973
-
[22]
Uhlig F., A recurring theorem about pairs of quadratic f orms and extensions: A survey, Linear Algebra Appl., 25, (1979), 219-237
1979
-
[23]
Z., and Sen hadji L., Nonnegative joint diago- nalisation by congruence based on LU matrix factorization
W ang, L., Albera, L., Kachenoura, A., Shu, H. Z., and Sen hadji L., Nonnegative joint diago- nalisation by congruence based on LU matrix factorization. IEEE Signal Processing Letters, 20 (8) (2013), 807–810
2013
-
[24]
W eierstrass K., Zur Theorie der quadratischen und bili nearen Formen, Monatsber. Akad. Wiss., Berlin, (1868), 310-338
-
[25]
Signal Processing, 80(5) (2000), 897–902
Yeredor A., Blind source separation via the second char acteristic function. Signal Processing, 80(5) (2000), 897–902
2000
-
[26]
IEEE Transactions on signal pro cessing, 50 (7) (2002),1545–1553
Yeredor A., Non-orthogonal joint diagonalization in t he least-squares sense with application in blind source separation. IEEE Transactions on signal pro cessing, 50 (7) (2002),1545–1553
2002
-
[27]
Y., A necessary and sufficient condition for si multaneously diagonalisation of two hermitian matrices and its applications, Glasgow Mathemat ical Journal 11 (1970), 81-83
Yik-Hoi A. Y., A necessary and sufficient condition for si multaneously diagonalisation of two hermitian matrices and its applications, Glasgow Mathemat ical Journal 11 (1970), 81-83. School of Mathematics and Statistics, University College Du blin, Dublin 4, Ireland Email addre...
1970
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.