Higher traces of linear maps on finite-dimensional normed spaces

Tomasz Kania · 2025

Read Higher traces of linear maps on finite-dimensional normed spaces by Tomasz Kania (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-2510-16501.txt.

Introduction

Tomasz Kania, the sole author, studies trace formulas in a 2025 maths paper. A trace adds up key values of a map on a space. His spaces have fixed size but any shape of distance rule. He looks at blown-up maps on wedge forms, which blend directions in a signed way. The question asks which chance weights on unit points make a mean-based trace formula hold.

He shows the formula holds just when a linked mean equals the identity map. One cone-based weight works for all distance rules. A skin-based weight works when the space has enough symmetry. He also gives sparse, point-based forms of the formula. One blocked form shows where the skin weight must fail.

Catalog record

CREATORTomasz Kania
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-16501

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Higher traces of linear maps on finite-dimensional normed spaces Thanks: IM CAS (RVO 67985840). Tomasz Kania Abstract . We prove a unified trace–average formula for the k k -th higher trace λ k ​ ( A ) = tr ⁡ ( Λ k ​ A ) \lambda_{k}(A)=\tr(\Lambda^{k}A) of a linear operator A A on a finite-dimensional normed space. The formula averages the matrix coefficient w ↦ ⟨ ( Λ k ​ A ) ​ w , w ∗ ⟩ w\mapsto\left\langle(\Lambda^{k}A)w,w^{*}\right\rangle over the unit sphere of Λ k ​ X \Lambda^{k}X against a probability measure η \eta ; it holds for all A A if and only if the operator-valued average T η = ( N k ) ​ ∫ w ⊗ w ∗ ​ 𝑑 η T_{\eta}=\binom{N}{k}\int w\otimes w^{*}\,\mathrm{d}\eta equals the identity. Two natural choices of η \eta satisfy this isotropy: (i) the hypersurface measure when a finite isometry group acts as an orthogonal 2 2 -design on Λ k ​ ℝ N \Lambda^{k}\mathbb{R}^{N} ; and (ii) the cone probability measure (no symmetry needed). We also identify a first-order obstruction for hypersurface averages at k = 1 k=1 : only degree– 2 2 spherical harmonics of the support function contribute. 2020 Mathematics Subject Classification Primary 15A75, 52A40; Secondary 15A15, 46B20, 42C10, 20C15, 52A20 1. Introduction and main results Let X be an N -dimensional real normed space and A\in\End(X) . For 1\leqslant k\leqslant N , the higher traces \lambda_{k}(A)\ :=\ \tr(\Lambda^{k}A), are the coefficients of the characteristic polynomial \det(I-tA)=\sum_{k=0}^{N}(-1)^{k}\lambda_{k}(A)\,t^{k},\qquad\lambda_{0}=1. There are several complementary ways to understand \lambda_{k} : • Exterior-power/ k -volume distortion: \Lambda^{k}A acts on oriented k -volumes; \lambda_{k}(A) is the trace of this action and equals the sum of all principal k\times k minors of A . • Averaging over Grassmannians: If X=\mathbb{R}^{N} is Euclidean, then \lambda_{k}(A)=\binom{N}{k}\int_{G_{k,N}}\det(P_{E}A|_{E})\,\,\mathrm{d}\sigma(E), the O(N) -invariant average of the determinant of the compression P_{E}A|_{E} to a k -plane E [ 2 , pp. 231–234] . K. Morrison independently observed this viewpoint (unpublished note). • Representation theory: \lambda_{k}(A) is the character of the \Lambda^{k} -representation evaluated at A . Averaging the matrix coefficient \left\langle(\Lambda^{k}A)w,w\right\rangle over the O(N) -orbit of unit simple k -vectors reproduces \tr(\Lambda^{k}A) after normalisation. The main theme of the paper is the question of whether we can realise these averages for arbitrary norms by integrating over the unit sphere of the normed exterior power? For k=1 , Morrison and the author [ 6 ] gave such a formula under the hypothesis of the existence of a 1-symmetric basis, that is, the action of a big enough hypooctahedral group on self-isometries of the space; the present work gives a unified framework for all k and all norms by isolating a single isotropy condition on a measure \eta on the unit sphere of \Lambda^{k}X . The main result of the paper is that the trace average \lambda_{k}(A)=n_{k}\!\int_{S_{V}}\!\langle(\Lambda^{k}A)w,w^{*}\rangle\,\,\mathrm{d}\eta(w),\qquad V:=\Lambda^{k}X,\ \ n_{k}=\binom{N}{k}, holds for all A if and only if T_{\eta}:=n_{k}\!\int_{S_{V}}w\otimes w^{*}\,\,\mathrm{d}\eta(w)=I_{V}. This extends the familiar trace-average formula (case k=1 ; see [ 5 ] ) and higher-trace formulae by Eberlein [ 2 ] . This isotropy is automatic for the cone probability measure , by the Gauss–Green theorem on B_{V} , and it holds for the hypersurface measure whenever a finite isometry group acts on X and induces an orthogonal 2 -design on \Lambda^{k}\mathbb{R}^{N} . We also quantify the obstruction to isotropy for \mu_{k} at k=1 : only degree– 2 spherical harmonics of the support function enter at first order. 2. Preliminaries We identify X with \mathbb{R}^{N} via a fixed auxiliary Euclidean inner product \left\langle\cdot,\cdot\right\rangle_{2} , used only to define Euclidean volumes, surface measures, and normals. Denote by B_{X} the unit ball and by S_{X}=\partial B_{X} the unit sphere. Let \mu be the normalised Euclidean hypersurface measure on S_{X} . At \mu -a.e. x\in S_{X} there is a unique norming functional x^{*} with \|x^{*}\|_{*}=1 and \left\langle x,x^{*}\right\rangle=1 (see [ 6 , Prop. 2.4] ). Orthogonal 2 -designs and symmetry Let W be a Euclidean space and let G\leqslant O(W) be a finite subgroup of the orthogonal group. We say that G is an orthogonal 2 -design on W if (2.1) \frac{1}{|G|}\sum_{Q\in G}Q^{\top}B\,Q\;=\;\frac{\tr B}{\dim W}\,I_{W}\qquad\text{for all }B\in\End(W). Equivalently, if w is any unit vector and \nu_{G} is the uniform measure on the orbit G\cdot w , then \int(u\otimes u)\,\,\mathrm{d}\nu_{G}(u)\;=\;\frac{1}{\dim W}\,I_{W}. In our applications W=\Lambda^{k}\mathbb{R}^{N} with the induced Euclidean structure, and the representation is Q\mapsto\Lambda^{k}Q . Example 2.1 (Hyperoctahedral group) . Let B_{N} be the group of signed permutation matrices. Then for every k the induced action of B_{N} on W=\Lambda^{k}\mathbb{R}^{N} is an orthogonal 2 -design. Indeed, the commutant of \Lambda^{k}(B_{N}) is scalar: commuting with all sign-flips forces diagonality in the basis \{e_{i_{1}}\wedge\cdots\wedge e_{i_{k}}\} , while commuting with permutations forces equality of all diagonal entries; hence \End(W)^{B_{N}}=\mathbb{R}I_{W} . Averaging as in ( 2.1 ) then yields \frac{1}{|B_{N}|}\sum Q^{\top}BQ=(\tr B/\dim W)\,I_{W} . Example 2.2 (Low-dimensional geometric groups) . In \mathbb{R}^{2} , the dihedral group D_{m} ( m\geqslant 3 ) is a 2 -design on \Lambda^{1}\mathbb{R}^{2}\cong\mathbb{R}^{2} . In \mathbb{R}^{3} , the rotation groups of the Platonic solids are 2 -designs on \Lambda^{1}\mathbb{R}^{3}\cong\mathbb{R}^{3} and, via the Hodge isometry \star:\Lambda^{2}\mathbb{R}^{3}\overset{\cong}{\to}\Lambda^{1}\mathbb{R}^{3} , also on \Lambda^{2}\mathbb{R}^{3} . Exterior powers For 1\leqslant k\leqslant N , write V=\Lambda^{k}X and n_{k}=\dim V=\binom{N}{k} . We equip V with the exterior projective norm as introduced in [ 8 ] : \|w\|_{\wedge,\pi}=\inf\Big\{\sum_{r}\|x^{(r)}_{1}\|\cdots\|x^{(r)}_{k}\|\ \colon\ w=\sum_{r}x^{(r)}_{1}\wedge\cdots\wedge x^{(r)}_{k}\Big\}. The canonical Euclidean inner product on V is induced by \left\langle x_{1}\wedge\cdots\wedge x_{k},y_{1}\wedge\cdots\wedge y_{k}\right\rangle_{2}=\det[\left\langle x_{i},y_{j}\right\rangle_{2}]_{i,j} . Let \sigma be the Euclidean surface measure on S_{V}=\partial B_{V} . Denote by \mu_{k} the normalised Euclidean hypersurface measure on S_{V} . Lemma 2.3 . Let X be a finite-dimensional normed space and equip \Lambda^{k}X with the exterior projective norm. (a) If Q:X\to X is a linear isometry, then \Lambda^{k}Q:(\Lambda^{k}X,\|\cdot\|_{\wedge,\pi})\to(\Lambda^{k}X,\|\cdot\|_{\wedge,\pi}) is an isometry. (b) More generally, for any T\in\End(X) , one has \|\Lambda^{k}T\|_{\Lambda^{k}X\to\Lambda^{k}X}\leqslant\|T\|^{k} . Proof. (a) Fix w\in\Lambda^{k}X and \varepsilon>0 . Choose a representation w=\sum_{r=1}^{m}x^{(r)}_{1}\wedge\cdots\wedge x^{(r)}_{k} such that \|w\|_{\wedge,\pi}\geqslant\sum_{r=1}^{m}\|x^{(r)}_{1}\|\cdots\|x^{(r)}_{k}\|-\varepsilon. Applying \Lambda^{k}Q and using that Q is an isometry on X , \Lambda^{k}Q(w)=\sum_{r=1}^{m}(Qx^{(r)}_{1})\wedge\cdots\wedge(Qx^{(r)}_{k}),\qquad\prod_{i=1}^{k}\|Qx^{(r)}_{i}\|=\prod_{i=1}^{k}\|x^{(r)}_{i}\|. Taking the infimum over all such decompositions on the right yields \|\Lambda^{k}Q(w)\|_{\wedge,\pi}\leqslant\|w\|_{\wedge,\pi}+\varepsilon . Since \varepsilon>0 is arbitrary, \|\Lambda^{k}Q(w)\|_{\wedge,\pi}\leqslant\|w\|_{\wedge,\pi} . Applying the same argument to Q^{-1} gives the reverse inequality, hence \|\Lambda^{k}Q(w)\|_{\wedge,\pi}=\|w\|_{\wedge,\pi} for all w , i.e. , \Lambda^{k}Q is an isometry. (b) For any representation of w as above, \|\Lambda^{k}T(w)\|_{\wedge,\pi}\;\leqslant\;\sum_{r=1}^{m}\prod_{i=1}^{k}\|Tx^{(r)}_{i}\|\;\leqslant\;\sum_{r=1}^{m}\prod_{i=1}^{k}\|T\|\,\|x^{(r)}_{i}\|\;=\;\|T\|^{k}\sum_{r=1}^{m}\prod_{i=1}^{k}\|x^{(r)}_{i}\|. Taking the infimum over all decompositions of w gives \|\Lambda^{k}T(w)\|_{\wedge,\pi}\leqslant\|T\|^{k}\,\|w\|_{\wedge,\pi} , hence \|\Lambda^{k}T\|\leqslant\|T\|^{k} . ∎ Cone measures Let K\subset\mathbb{R}^{m} be a convex body with 0\in\mathrm{int}\,K . The cone probability measure \nu_{K} on \partial K is \nu_{K}(A)=\frac{\vol(\{tx:x\in A,\ 0\leqslant t\leqslant 1\})}{\vol(K)}. If n_{e} exists a.e., then \,\mathrm{d}\nu_{K}(x)=\frac{\langle n_{e}(x),x\rangle}{m\,\vol(K)}\,\,\mathrm{d}\sigma(x) (Schneider [ 7 , §§2.2, 4.2, 8.2] ; Gardner [ 4 , §§8.4, B.2] ). Grassmannian viewpoint and the \Lambda^{k} -sphere in the Euclidean case Let us consider X=\mathbb{R}^{N} with its Euclidean inner product. Fix w_{0}=e_{1}\wedge\cdots\wedge e_{k} and \mathcal{O}_{k}:=\{(\Lambda^{k}g)w_{0}:g\in O(N)\}\subset S_{\Lambda^{k}\mathbb{R}^{N}} . Then \mathcal{O}_{k}\ \cong\ O(N)\big/\big(O(k)\times O(N-k)\big), and the map E\in G_{k,N}\mapsto w_{E}:=v_{1}\wedge\cdots\wedge v_{k} (for any orthonormal basis (v_{i}) of E ) identifies the unoriented Grassmannian with \mathcal{O}_{k}/\{\pm 1\} . For such w_{E} , \big\langle(\Lambda^{k}A)w_{E},\,w_{E}\big\rangle=\det\!\big([\langle Av_{i},v_{j}\rangle]_{i,j\leqslant k}\big)=\det\!\big(P_{E}A|_{E}\big), so averaging \langle(\Lambda^{k}A)w,w\rangle over \mathcal{O}_{k} (or G_{k,N} ) yields Eberlein’s formula for \lambda_{k}(A) with factor \binom{N}{k} [ 2 , pp. 231–234] . The Gauss–Green theorem and sets of finite perimeter Lemma 2.4

Read the complete text