The matched projection and geodesics of the Grassmann manifold
Esteban Andruchow · 2025
Read The matched projection and geodesics of the Grassmann manifold by Esteban Andruchow (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-2508-14870.txt.
Introduction
Esteban Andruchow settles a question about two ways of rounding a slanted projection into a flat one, in a 2025 math paper. The setting is a complex Hilbert space, the infinite-dimensional home of quantum states. An idempotent operator acts like a projection but may lean; orthogonal projections are its upright cousins.
One upright version comes from the polar decomposition, a standard factorization of the operator. The other, the matched projection, is defined as the closest orthogonal projection in norm. Earlier authors found each separately. Andruchow proves the two are the same operator. He also studies the space of all subspaces, the Grassmann manifold, and finds a unique shortest path between the ranges of the operator and its adjoint. The projection onto the path's midpoint also equals the matched projection. The triple agreement ties algebra, approximation, and geometry together.
Catalog record
| CREATOR | Esteban Andruchow |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-14870
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The matched projection and geodesics of the Grassmann manifold Esteban Andruchow Abstract Given an idempotent operator E E in a complex Hilbert space ℋ {\cal H} , one can associate to it two orthogonal projections: • The polar decomposition 2 E − 1 = ( 2 P − 1 ) | 2 E − 1 | 2E-1=(2P-1)|2E-1| provides an orthogonal projection P P . That the unitary part in the decomposition of 2 E − 1 2E-1 is of this form, i.e., a selfadjoint unitary operator, is a remarkable observation done by G. Corach, H. Porta and L. Recht (see references below). • The question of which, among all orthogonal projections, is the one closest in norm to E E , provides another projection, the so called matched projection m ( E ) m(E) , which answers this question. It was found by X. Tian, Q. Xu and C. Fu (see references below). In this paper we show that these projections coincide. Moreover, we show that there exists a unique minimal geodesic of the Grassmann manifold of ℋ {\cal H} (the manifold of closed subspaces of ℋ {\cal H} ) that joins R ( E ) R(E) and R ( E ∗ ) R(E^{*}) . The orthogonal projection onto the midpoint of this geodesic, also coincides with m ( E ) m(E) . 2020 MSC: 47B02, 58BXX, 58B20 . Keywords: Idempotents, Projections, Grassmann manifold . 1 Introduction Let E be an idempotent in {\cal B}({\cal H}) , which written as a 2\times 2 matrix in terms of the decomposition {\cal H}=R(E)\oplus R(E)^{\perp} is of the form E=\left(\begin{array}[]{cc}1&B\\ 0&0\end{array}\right), (1) where B:R(E)^{\perp}\to R(E) . If {\cal S}\subset{\cal H} is a closed subspace, denote by P_{\cal S} the orthogonal projection onto {\cal S} . In this note we study the position of the projections P_{R(E)} and P_{R(E^{*})}=P_{N(E)}^{\perp} as points in the Grassmann manifold {\cal P}({\cal H}) , of all orthogonal projections in {\cal H} . The points P_{R(E)} and P_{R(E^{*})} lie at distance strictly less than 1 , namely (Theorem 3.4 ) \|P_{R(E)}-P_{R(E^{*})}\|=\frac{\|B\|}{(1+\|B\|^{2})^{1/2}}. In [ 11 ] H. Porta and L. Recht, introduced a natural linear connection in {\cal P}({\cal H}) , and proved that two projections P,Q\in{\cal P}({\cal H}) such that \|P-Q\|<1 are joined by a unique geodesic of {\cal P}({\cal H}) (up to reparametrization), which is minimal for the Finsler metric given by the usual norm of {\cal B}({\cal H}) at very tangent space. It follows that P_{R(E)} and P_{R(E^{*})} are joined at times t=0 and t=1 by a (unique, minimal) geodesic \delta_{E} , which is of the form \delta_{E}(t)=e^{itX_{E}}P_{R(E)}e^{-itX_{E}} , for X_{E}^{*}=X_{E} , \|X_{E}\|<\pi/2 , co-diagonal in the above 2\times 2 matrix representation. In [ 12 ] X. Tian, Q. Xu and C. Fu introduced the matched projection m(E) of an idempotent operator E . Among remarkable properties, they prove that m(E) is the orthogonal projection which is closest in norm to E : \|m(E)-E\|\leq\|P-E\|,\ \hbox{ for all }P\in{\cal P}({\cal H}). In [ 7 ] , G. Corach, H. Porta and L. Recht studied the inmersion {\cal P}({\cal H})\hookrightarrow{\cal Q}({\cal H}) , of {\cal P}({\cal H}) into the larger space {\cal Q}({\cal H})=\{E\in{\cal B}({\cal H}):E^{2}=E\} of idempotent operators, or oblique projections. For instance, they prove that the C^{\infty} -map \pi:{\cal Q}({\cal H})\to{\cal P}({\cal H}), is a retraction, where 2E-1=(2\pi(E)-1)|2E-1| is the polar decomposition of 2E-1 . In this paper we show that m(E)=\delta_{E}(\frac{1}{2})=\pi(E). The contents of the paper are the following. In Section 2 we give basic properties of R(E) and R(E^{*}) . In Section 3 we compute the spectrum (and norm) of P_{R(E)}-P_{R(E^{*})} . In Section 4 we characterize the geodesic \delta_{E} , by giving an expression of the exponent X_{E} in terms of the operator B of ( 1 ). In Section 5 we consider the matched projection m(E) of E [ 12 ] , and show that m(E)=\delta_{E}(\frac{1}{2}) . In Section 6 we use Halmos’ theory for two subspaces [ 10 ] and Davis’ results on the difference of orthogonal projections [ 8 ] , to give further descriptions of m(E) . In Section 7 we study an example: E_{a}=\frac{1}{2}(\Gamma_{a}-1) , where \Gamma_{a} is the composition operator \Gamma_{a}f=f\circ\varphi_{a} acting in {\cal H}=L^{2}(\mathbb{T},dm) ( \mathbb{T} the unit circle, dm the normalized Lebesgue measure), and \varphi_{a}(z)=\frac{a-z}{1-\bar{a}z} , for |a|<1 . In Section 8 we consider the retraction \pi:{\cal Q}({\cal H})\to{\cal P}({\cal H}) and prove that \pi(E)=\delta_{E}(\frac{1}{2}) . We also examine certain metric properties of m(E) . 2 Preliminaries We say that an operator C is a reflection if C^{2}=1 ; a symmetry S is a selfadjoint reflection. We identify a closed subspace {\cal S}\subset{\cal H} with the orthogonal projection P_{\cal S} onto {\cal S} , and with the symmetry 2P_{\cal S}-1 which equals the identity on {\cal S} and minus the identity on {\cal S}^{\perp} . The set of subspaces, projections or symmetries will be denoted {\cal P}({\cal H}) . Similarly, we denote by {\cal Q}({\cal H}) the space of idempotents Q or reflections 2Q-1 . When considering a pair of subspaces {\cal S},{\cal T}\subset{\cal H} (or projections P_{\cal S}.P_{\cal T} ) as points in {\cal P}({\cal H}) , a key geometric fact are the intersections {\cal S}\cap{\cal T}^{\perp} and {\cal S}^{\perp}\cap{\cal T} : the necessary and sufficient condition for the existence geodesics of {\cal P}({\cal H}) joining P_{\cal S} and P_{\cal T} is that these intersections have the same dimension; the geodesic is unique (up to reparametrization) iff these intersections are trivial. The geodesic \delta which joins \delta(0)={\cal S} and \delta(1)={\cal T} ( {\cal S} and {\cal T} satisfying the above condiion) takes the form of a one-parameter group of unitaries acting on {\cal S} . The infinitesimal generator of this group is a bounded selfadjoint operator X , which can be chosen with norm \|X\|\leq\pi/2 , and its 2\times 2 matrix with respect to {\cal S}\oplus{\cal S}^{\perp}={\cal H} (and also with respecto to {\cal T}\oplus{\cal T}^{\perp}={\cal H} ) is codiagonal. The geodesic is then, as a curve of subspaces \delta(t)=e^{itX}{\cal S}, or \delta(t)=e^{itX}P_{\cal S}e^{-itX},\ \delta(t)=e^{itX}(2P_{\cal S}-1)e^{-itX} as projections or symmetries. Regarded as symmetries, it has the useful computational feature, that since X is codiagonal with respect to {\cal S}\oplus{\cal S}^{\perp}={\cal H} , X anti-commutes with 2P_{\cal S}-1 . Therefore \delta takes the alternative forms \delta(t)=e^{2itX}(2P_{\cal S}-1)=(2P_{\cal S}-1)e^{-2itX}. These facts were shown in [ 11 ] . In our case, R(E)\cap R(E^{*})^{\perp}=R(E)\cap N(E)=\{0\}\hbox{ and }R(E)^{\perp}\cap R(E^{*})=N(E^{*})\cap R(E^{*})=\{0\}, so that there exists a unique geodesic joining P_{R(E)} and P_{R(E^{*})} . Also note that R(E)\cap R(E^{*})=\{f\in R(E):B^{*}f=0\}=N(B^{*}), (2) and R(E)^{\perp}\cap R(E^{*})^{\perp}=\{g\in R(E)^{\perp}:Bg=0\}=N(B). (3) Two subspaces {\cal S},{\cal T} are in generic position [ 10 ] if {\cal S}\cap{\cal T}={\cal S}\cap{\cal T}^{\perp}={\cal S}^{\perp}\cap{\cal T}={\cal S}^{\perp}\cap{\cal T}^{\perp}=\{0\} . In general, we call generic part of {\cal S} and {\cal T} the subspace {\cal H}_{0}=\left({\cal S}\cap{\cal T}\oplus{\cal S}\cap{\cal T}^{\perp}\oplus{\cal S}^{\perp}\cap{\cal T}\oplus{\cal S}^{\perp}\cap{\cal T}^{\perp}\right)^{\perp}. In our case R(E) and R(E^{*}) are in generic position iff B has trivial nullspace and dense range. The generic part of R(E) and R(E^{*}) {\cal H}_{0}=\left(R(E)\cap R(E^{*})\oplus R(E)^{\perp}\cap R(E^{*})^{\perp}\right)^{\perp}=\left(N(B)\oplus N(B^{*})\right)^{\perp} is trivial iff B=0 , i.e. E=E^{*} is an orthogonal projection. Since E and E^{*} coincide in {\cal H}_{0}^{\perp} , in order to characterize the unique geodesic between P_{R(E)}