Pure-state extensions across a split ideal
Statement
QUESTION 3: Let a C*-algebra be the direct sum of an ideal and a MASA, . Under what circumstances do the pure states of have unique extension to states of ?
- Cynthia Vinzant
- there are existence proofs for KS, how to make them constructive? Likes stable polynomials and interlacers. How to make constructive proofs?
- Find an algorithm for Weaver's conjecture/equivalence/theorem (or call it the vector partitioning problem), similarly for Paving Conjecture
- Chuck started to describe a cooperative two-player game with partitioning the sets --> could we get an impossibility result via communication complexity?
- Greg Knese
- understanding intrinsic proofs of Feichtinger's conjecture
- extend the theory of interlacing families to several variables
- use a theorem to define interlacing in several variables
- how do operators that preserve stability affect locations of roots (i.e., quantitative versions of MSS)
- Petter Branden
- combinatorics generally
- better bounds, more quantitative results in Weaver's result
- for the random MSS, there's a witness for the best bound, is there a conjecture for Weaver's conjecture? Weaver has a construction that \sqrt\eps is necessary but MSS has no constructive lower bound. Pete C. does have a constructive lower bound.
- has a witness to a hyperbolic version of the problem
- more generally, matching lower bounds for the different equivalences (the paving lower bound in the \R case is 1/\eps^2 and authors showed 6/\eps^2)
- Pete's example: can we make it better? It will affect all of the equivalences! It's almost a counter-example.
- consequences to matroid theory: can we prove a continuous version of matroid theory using MSS theorem and what consequences to matroid theory does MSS have?
- serious about hyperbolic frame theory!
- extensions to hyperbolic polynomials --> just posted a new paper on the ArXiv
- Marcin Bownik
- interested in Feichtinger conjecture: suppose you have a Parseval frame and norms of vectors are bounded away from zero, what is the minimum number of partitions can we form that make Riesz basic sequences? Count the number of partitions? How does this number depend on \epsilon, the lower bound (Pete C. also will show some results tomorrow on this question)? How to exhibit/construct such partitions?
- Frames of complex of exponentials on a subset of the torus = exponential frames, derive better/best bounds for Feichtinger's conjecture for these frames.
- exponential frame L^2([0,1]), consider measurable subset E \subset [0,1]
- { \econst^{2 \pi \iunit n \ell \chi_E }_{n \in \Z} are an exponential frame
- Darrin Speegle
- interested in Feichtinger conjecture, how large the partition has to be to get Riesz basic sequences, especially for complex exponentials, this suggests that there's something about Fourier series that we don't understand.
- whether the number of partitions depends on |E| for exponential frames or any other properties of E? Are there any examples of two sets with same measure such that the number of partitions is different (or the constants are different)?
- More concretely, can we find E, |E| = \frac12 such that { \econst^{2 \pi \iunit n \ell \chi_E }_{n \in \Z} cannot be partitioned into two Riesz sequences?
- there's a connection to synthetic sets and arithmetic progressions (Bourgain's work with Tzafriri): we now know we can partition exponential frames into synthetic sets (MSS says we can do so). How to do this (constructively?)?
- Bill Johnson -Is paving true for ? Specifically:
Suppose .
: Does there exist for every a constant s.t. for every and every by zero diagonal matrix for which , there is a partition {1,\dots,n}$ s.t.
Here \|\dot \|_ is the operator norm for operators on , NOT the Schatten -class norm.
, due IIRC to Bourgain has been known for a long time and is relatively easy. It is more-or-less equivalent to for matrices that have non-negative entries, a result that is contained in
Berman, Kenneth; Halpern, Herbert; Kaftal, Victor; Weiss, Gary; Matrix norm inequalities and the relative Dixmier property. Integral Equations Operator Theory 11 (1988), no. 1, 28-48.
In their original paper, Bourgain and Tzafriri proved restricted invertibility for , but they got optimal estimates only for .
- Take A = n \times n matrix that's a commutator and write it as a commutator of B,C with B diagonal of norm 1 and C with bounded norm, with some minimization. Bill's recent PNAS paper with Ozawa and Schechtmann showed something that needed KS with optimal estimates to improve bound from n^\alpha (which uses only restricted invertibility) to poly(n), and that the existence of a bound independent of n implies KS. Can you improve the result using KS to get a bound independent of n?
- Deanna Needell
- design algorithms to get pavings, especially for stochastic optimization
- interested in algorithms that are efficient!
- algorithmic developments that use pavings
- sampling and RIP constructions for sparse approximation, do these lead to constructions or insight for KS
- Anna Gilbert
- Fourier sampling questions, RIP for the DFT, Rudelson selection lemma, how to remove extra factors on the bounds for RIP for DFT
- Olaf Mordhorst
- applications to convex geometry and Banach spaces
- Restricted invertibility: provide a concrete example of optimality which is proven via random matrices
- Pierre Youssef
- consequences in convex geometry
- if you have an identity decamp in \R^n, BSS approximated the identity with many fewer, albeit weighted, vectors. If you use MSS, you get an identity approximation without weights. Let's add a condition that the vectors are balanced (say they come from a convex body).
- \sum_j c_j x_j x_j^T with |x_j| = 1 and \sum_j c_j x_j = 0
- Find S \subset [m] , |S| \approx d so that C \sum_{j \in S} x_j x_j^T \approx I with a balancing condition | \sum x_j | \leq B
- this was done by Srivastava but got 4 + \eps approximation of the identity
- Given a convex body K in \R^n, Restricted Invertibility says there exists an E \subset \R^n with dim(E) = (1-\eps)n so that distance( K \cap E, B_1) \approx \sqrt{dim(E)}
- KS says that we can partition K into subspaces like this. Can we improve the Banach-Mazur distance of a convex body to the cube using this partition? The best bound currently is n^{5/6}.
- Algorithmic version of KS: try the Batson-S-S approach to KS? BUT, they had convinced themselves that their approach wouldn't work. Perhaps it is worth to revisit this?
- Leonid Gurvits
- there's an algorithm: start with k vectors in n-dimnl space. Can you partition them into 5 sets so that set spans the space. Take each vector, repeat it 5 times and augment those matrices by identity to get dimension k. You'll get k semi-def-pos. matrices. Now you check each matrix for some property xxxx. This is an algorithm to carry out Rado-Horn?
- Mirko Visontai
- understand limitations of the proof, we don't have any idea of how many good partitions there are? come up with more quantitative versions of the proof.
- can we generate new interlacing families?
- Mihai Putinar
- perturbation determinant: in interlacing families, we have quotients of such polynomials, they form rational functions in the Nevalinna class of functions that preserve the upper half plane
- understand KS in terms of the phase shift theory, perhaps getting a proof of the KS in terms that analysts can understand
- Dan Edidin
- wants to geometrize this work! Paving projections is especially interesting. Fix r. Find the conditions on the diagonal of a projection so it can be r-paved? Conditions should be independent of dimension.
- For KS2, the problem is static: given v, find a good partition. Wants a more dynamic version: look at all vectors together (in a manifold structure), so that you replace the discrete structure with a continuous one. Look over some parametrized space with geometric structure and this may be more helpful to characterize when there are lots of partitions (i.e., count) that satisfies KS2.
- Pablo Parrilo
- algorithmic questions!
- To try to understand why the MSS proof works: many existence proofs use probabilistic methods or fixed point methods and they are non-constructive. But MSS proof combines different techniques. Understand how these pieces fit together more generally.
- Rachel Ward
- see the above on algorithmic!
- Leonid Gurvits
- Is there an analog of KS for zonotopes? Let z_1,\ldots,z_k \in \R^n, k > n and let \bar z_i denote the interval { \alpha z_i | 0 \leq \alpha \leq 1}. Consider the Minkowski sum Zon = \alpha_1 \bar z_1 + \cdots \alpha_k \bar z_k.
- Conjecture: vol(Zon) = 1.
- KS analog: symmetrize Zon and then seek a partition S \cup T so that
- \sum_{i \in S} \bar z_i \subset (1/2 \pm \eps) Zon.
- Bernhard Bodmann
- how to eliminate those polynomials?! Borgain-Tzafriri's proof did not have such polynomials, they used rank-1 perturbation theory. Can we reproduce such a proof at the level of multi-variate polynomials?
- Adam Marcus
- one of the biggest comments we get is that KS was so popular because it was ubiquitous so it would require some new tool that other people could use in lots of other areas, unfortunately, these polynomial tools are not portable to other areas. If someone did prove KS in your area, what would it look like? Let's construct such a tool so that you have a hammer to bang away with in your area! So, please think about what this tool would like!
- There's a conjecture about repeated copies of Identity with zero padding that this gives you the worse bounds on roots?
- Using interlacing polynomials but if you only want the maximum root, you don't need all these other roots to exist, can we trim our conditions so that the machinery is more usable? Is there some weaker form of interlacing if you don't care about all the roots.
- There was something about adding rank-2 perturbations rather than rank-1 perturbations.
Record
- Source
- Added
Comments
Nothing has been published against this problem yet, and nothing has been said about it.
Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.
Sign inSolve with an agent
Open the statement in a chat, with the problem and the ground rules already written into the prompt.