An Operator Theoretic Approach to Birkhoff’s Problem 111
Miles Gould · 2024
Read An Operator Theoretic Approach to Birkhoff’s Problem 111 by Miles Gould (2024) 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-2412-06707.txt.
Introduction
Miles Gould, writing alone in 2024, took an operator theory route to a matrix problem posed by Garrett Birkhoff in 1946. Birkhoff showed that a set of square matrices with set row and column sums is the closed hull of the swap matrices. He asked whether the same holds for endless matrices under some sense of closeness.
A 1955 result said no for one such sense. Gould shows the answer stays no across the many senses that operator theory offers. He also sharpens a positive result from 1960, showing how far it extends and where its limit lies. The paper closes by mapping the extreme points of the hull under these senses.
Catalog record
| CREATOR | Miles Gould |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-06707
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
An Operator Theoretic Approach to Birkhoff’s Problem 111 Miles Gould Abstract . In 1946, Garrett Birkhoff proved that the n × n n\times n doubly stochastic matrices comprise the convex hull of the n × n n\times n permutation matrices, which in turn make up the extreme points of this polytope. He proposed his problem 111, which asks whether there exists a topology on infinite matrices for which this applies to the closed convex hull of the ℕ × ℕ \mathbb{N}\times\mathbb{N} permutation matrices. As Isbell showed in 1955, this equality is not achieved in the line-sum norm. In this paper, we use the domain of operator theory, and its many topologies, to improve on his negative result by showing that Birkhoff’s problem is not solved in any of these topologies. In Kendall’s 1960 paper on this problem, he gave an answer to the affirmative, as well as a topology for which closed convex hull comprises the doubly substochastic matrices. We also show that Kendall’s secondary theorem also applies for all the locally convex Hausdorff topologies finer than than Kendall’s (namely that of entry-wise convergence) which make the continuous dual of the matrix space no larger than the predual of the von Neumann algebra containing them. We then show that this is a theoretical upper limit topologies with this closure property. We also discuss the exposed points of this hull for these several topologies. Moreover, we show that, in these topologies, the closed affine hull of these permutation matrices comprise all operators with real-entry matrix coefficients. 1. The Standard Representation of \mathrm{S}_{\mathbb{N}} Definition 1.1 . Let H be a separable infinite dimensional Hilbert space. For any orthonormal basis E=(e_{n})_{n\in\mathbb{N}} of H, we define the standard unitary representation \pi^{E}:\mathrm{S}_{\mathbb{N}}\rightarrow B(H) (or simply \pi when E is clear) of \mathrm{S}_{\mathbb{N}} by \pi(\rho)e_{n}=e_{\rho(n)}. For the reader strictly interested in our considerations on convex hulls, this entire first section need not be read, as it serves only to justify the use of the standard representation \pi. Our definition of a unitary representation \pi of a topological group G is that of [ 3 ] , which in addition to being a group homomorphism into the unitaries U(H), must make also the map G\times H\rightarrow H by (g,x)\mapsto\pi(g)x continuous for the product topology. For any of the mainstream operator topologies, namely weak, ultraweak, strong, strong ∗ , ultrastrong, ultrastrong ∗ , Arens-Mackey, weak Banach, and norm, this map is continuous. Per [ 3 ] , the continuity of this map is equivalent to the continuity of \pi:G\rightarrow U(H) where the codomain is equipped with the strong topology. Moreover, we know the weak and strong coincide on U(H), so for any topology finer than the weak, this continuity follows. Proposition 1.2 . The pointwise topology is the coarsest topology which makes S_{\mathbb{N}} a Hausdorff topological group. Moreover, it is the coarsest topology on S_{\mathbb{N}} which makes \pi continuous with respect to the weak topology on B(H). Proof. The first part of the theorem is given by [ 5 ] . Next, a net (\pi_{\lambda}) in B(H) converges to \pi strongly in B(H) if and only if, for all x\in H, \lim_{\lambda}\sum_{n=1}^{\infty}|\langle(\pi_{\lambda}-\pi)x,e_{n}\rangle|^{2}=0. Therefore, if S_{\mathbb{N}} is equipped with the topology we seek, a net (\rho_{\lambda}) converges to \rho if and only if \lim_{\lambda}\sum_{n=1}^{\infty}|\langle(\pi_{\lambda}(\rho)-\pi(\rho))x,e_{n}\rangle|^{2}=0, which simplifies to \lim_{\lambda}\sum_{n=1}^{\infty}|x_{\rho^{-1}_{\lambda}(n)}-x_{\rho^{-1}(n)}|^{2}=0. Notice the following bound \displaystyle\lim_{\lambda}\sum_{n=1}^{\infty}|x_{\rho^{-1}_{\lambda}(n)}-x_{\rho^{-1}(n)}|^{2} \displaystyle\leq\sum_{n=1}^{\infty}\left(|x_{\rho^{-1}_{\lambda}(n)}|^{2}+|x_{\rho^{-1}(n)}|^{2}\right) \displaystyle\leq\sum_{n=1}^{\infty}|x_{\rho^{-1}_{\lambda}(n)}|^{2}+\sum_{n=1}^{\infty}|x_{\rho^{-1}(n)}|^{2} \displaystyle\leq 2\sum_{n=1}^{\infty}|x_{n}|^{2} \displaystyle\leq 2\|x\| \displaystyle<\infty. Thus, the prior limit holds if and only if, for each n\in\mathbb{N}, x_{\rho^{-1}_{\lambda}(n)}\rightarrow x_{\rho^{-1}(n)}. Moreover, since this must hold for every x\in H, convergence holds if and only if \rho^{-1}_{\lambda}(n)\rightarrow\rho^{-1}(n). Since \mathbb{N} is discrete, this holds if and only if \rho^{-1}_{\lambda}(n)=\rho^{-1}(n) for eventual \lambda. Finally, because every n=\rho(m) for some unique m\in\mathbb{N}, convergence is equivalent to the statement: for all m\in\mathbb{N}, \rho_{\lambda}(m)=\rho(m) for eventual \lambda. ∎ An interesting consequence of equipping S_{\mathbb{N}} with the pointwise topology is that every irreducible unitary representation of S_{\mathbb{N}} must be over a separable Hilbert space. Therefore the degree of every faithful irreducible unitary representation is bounded above by \aleph_{0}. We will now show that the degree must also be bounded below by \aleph_{0}. Lemma 1.3 . Let n\in\mathbb{N}. Then every unitary representation \pi:S_{\mathbb{N}}\rightarrow M_{n}(\mathbb{C}) is unfaithful. Proof. Suppose, towards contradiction that \pi:S_{\mathbb{N}}\rightarrow M_{n}(\mathbb{C}) is faithful. Let \sim be the equivalence relation on S_{\mathbb{N}} by \rho\sim\nu iff \rho\nu=\nu\rho. By assumption, the quotient S_{\mathbb{N}}/\sim must be finite. Consider the sequence (\rho_{k}) in S_{\mathbb{N}} by \rho_{k}:(1\text{ }2\cdots k+1) in cycle notation. Whenever m\neq k, \rho_{k}\rho_{m}\neq\rho_{m}\rho_{k}, a contradiction, since this sequence could not fit into S_{\mathbb{N}}/\sim. ∎ Proposition 1.4 . For any topology \mathcal{T} on S_{\mathbb{N}} finer than pointwise, \pi is a faithful non-degenerate unitary representation of (S_{\mathbb{N}},\mathcal{T}) on H. Proof. Let \rho,\nu\in S_{\mathbb{N}}. Then \pi(\rho)\pi(\nu)(e_{n})=\pi(\rho)(e_{\nu(n)})=e_{\rho\nu(n)}, so \pi is a group homomorphism (and so a representation). Moreover, \langle\pi(\rho)(e_{n}),e_{m}\rangle=\delta_{\rho(n),m} where \delta is the Kronecker delta. Since it is real, \delta_{\rho(n),m}=\delta_{\rho^{-1}(m),n}=\langle\pi(\rho)^{*}(e_{m}),e_{n}\rangle, so \pi(\rho)^{*}=\pi(\rho^{-1})=\pi(\rho)^{-1}. Therefore \pi is unitary. Now suppose \pi(\rho)=\pi(\nu). Then e_{\rho(n)}=e_{\nu(n)} for all n\in\mathbb{N}, i.e. \rho=\nu, so \pi is faithful. Since \mathrm{Id}_{H}\in\pi(S_{\mathbb{N}}), non-degeneracy is trivial. ∎ Next, we will show a property which differs between the standard representation of the infinite symmetric group and the standard representations of the finite symmetric groups, namely that the former is irreducible. Lemma 1.5 . Let \pi_{m}:S_{m}\rightarrow M_{m}(\mathbb{C}) by \pi_{m}(\rho)e_{n}=e_{\rho(n)}. Then the commutant \pi(S_{m})^{\prime}=\mathrm{span}(\mathrm{Id}_{m},U_{m}), where [U_{m}]_{ij}=1. In particular, \pi_{m} is reducible. Proof. (Induction) Clearly, U_{1}=\mathrm{Id}_{1}, so \pi_{1}(S_{1})^{\prime}=\mathbb{C}\mathrm{Id}_{1} is trivial. Now, let k\in\mathbb{N} and suppose \pi(S_{k})^{\prime}=\mathrm{span}(\mathrm{Id}_{k},U_{k}). Let a\in\pi_{k+1}(S_{k+1}). We can express it in k\times k and 1\times 1 block form. Then for every \rho\in S_{k}, \displaystyle\begin{pmatrix}\pi_{k}(\rho)&0\\ 0&1\end{pmatrix}\begin{pmatrix}A&B\\ C&D\end{pmatrix} \displaystyle=\begin{pmatrix}A&B\\ C&D\end{pmatrix}\begin{pmatrix}\pi_{k}(\rho)&0\\ 0&1\end{pmatrix} \displaystyle\begin{pmatrix}\pi_{k}(\rho)A&\pi_{k}(\rho)B\\ C&D\end{pmatrix} \displaystyle=\begin{pmatrix}A\pi_{k}(\rho)&B\\ C\pi_{k}(\rho)&D\end{pmatrix} Then A\in\pi_{k}(S_{k})^{\prime}, B is a 1\times k constant row, C a k\times 1 constant column, and D is free. We can then do the same for the 1\times 1 and k\times k block form, yielding the general equation \displaystyle\begin{pmatrix}a_{1}&a_{2}&\cdots&a_{2}&b\\ a_{2}&a_{1}&\cdots&a_{2}&b\\ \vdots&\vdots&\ddots&\vdots&\vdots\\ a_{2}&a_{2}&\cdots&a_{1}&b\\ c&c&\cdots&c&d\end{pmatrix} \displaystyle=\begin{pmatrix}a^{\prime}&b^{\prime}&\cdots&b^{\prime}&b^{\prime}\\ c^{\prime}&d^{\prime}_{1}&\cdots&d^{\prime}_{2}&d^{\prime}_{2}\\ \vdots&\vdots&\ddots&\vdots&\vdots\\ c^{\prime}&d^{\prime}_{2}&\cdots&d^{\prime}_{1}&d^{\prime}_{2}\\ c^{\prime}&d^{\prime}_{2}&\cdots&d^{\prime}_{2}&d^{\prime}_{1}\end{pmatrix} \displaystyle a^{\prime}=a_{1} \displaystyle=d=d^{\prime}_{1} \displaystyle a_{2}=d^{\prime}_{2}=b^{\prime} \displaystyle=b=c^{\prime}=c and so we have the final form a\in\mathrm{span}(\mathrm{Id}_{m},U_{m}). It is simple to check that all such matrices commute with \pi_{k+1}(S_{k+1}) . By induction, the lemma holds. ∎ Theorem 1.6 . For any topology finer than pointwise, \pi is irreducible. Proof. As before, we can express a general element a\in\pi(S_{\mathbb{N}})^{\prime} in m\times m and \mathbb{N}\times\mathbb{N} block form, and we can see that, for all such block decompositions A\in\mathrm{span}(\mathrm{Id}_{m},U_{m}), so every matrix coefficient must fit this pattern. However, if a\not\in\mathbb{C}\mathrm{Id}_{H} then a must be unbounded, a contradiction. Thus, \pi(S_{\mathbb{N}})^{\prime}=\mathbb{C}\mathrm{id}_{H}. ∎ 2. The Convex Hulls of \mathrm{S}_{\mathbb{N}} Quickly, we will show that the group algebra of \mathrm{S}_{\mathbb{N}} on H is strongly dense in B(H), which we will analogize for real-entry matrices later. Corollary 2.1 . \mathbb{C}[\pi(S_{\mathbb{N}})] is strongly dense in B(H). Moreover, \mathbb{C}[\pi(S_{\mathbb{N}})]\cap U(H) is strongly dense in U(H). Proof. This follows from the double commutant lemma and Kaplansky’s density theorem [ 12 , 123,131] . ∎ Definition 2.2 . For subsets S\subseteq B(H), we will use the following notation for particular closures of S, \overline{S} the uniform closure, \overline{S}^{s} the strong closure, \overline{S}^{w} the weak, etc. We are now going to define several crucial collections of permutations, partial permutations, and their polytopes. Definition 2.3 . Firstly, we should note that, just as \pi depends on E, so will all the following definitions, via the matrix coefficients