Math Labs

Worked solution: Marcus–Spielman–Srivastava proof via interlacing families (2013)

Step 1 of 8: From an infinite-dimensional operator question to Weaver's KS2KS_2
In plain words

In 1959 Richard Kadison and Isadore Singer asked a question about listening posts on an infinite-dimensional space ℓ2\ell^2: if you only know how a physical quantity behaves along one fixed set of coordinate axes (the diagonal), is there only one honest way to extend that knowledge to the whole space? For forty-five years nobody could decide.

Nik Weaver found a way to shrink this abstract question down to something almost combinatorial: a purely finite-dimensional statement KS2KS_2 about splitting a pile of vectors into two well-balanced groups. Solving the small, concrete puzzle KS2KS_2 would settle the huge, abstract one.

state extends uniquely on B(ℓ2)  ⟺  KS2:∑ivivi∗=Id, ∥vi∥2≤δ  ⟹  ∃ S1,S2\text{state extends uniquely on } B(\ell^2) \iff KS_2: \sum_i v_iv_i^*=I_d,\ \|v_i\|^2\le\delta \implies \exists\, S_1,S_2
Detailed analysis

Kadison and Singer's 1959 question is about states: a pure state on the diagonal algebra D≅ℓ∞D \cong \ell^\infty inside B(ℓ2)B(\ell^2) is determined by an ultrafilter on N\mathbb{N}, and the question was whether its extension by Hahn–Banach to a pure state on all of B(ℓ2)B(\ell^2) is forced to be unique. The hypothesis ∑ivivi∗=Id\sum_i v_i v_i^* = I_d in Weaver's KS2KS_2 is the finite-dimensional shadow of a tight frame decomposition of the identity, and ∥vi∥2≤δ\|v_i\|^2 \le \delta is the smallness condition that replaces "the diagonal is diffuse": no single vector carries a macroscopic share of IdI_d.

Weaver's 2004 theorem (Discrete Mathematics 278) reduces the operator-algebra problem to a countable family of finite-dimensional paving statements, of which KS2KS_2 is the sharp form: partition the index set into S1,S2S_1, S_2 so that both operator norms ∥∑i∈Sjvivi∗∥\|\sum_{i \in S_j} v_i v_i^*\| stay strictly below 11. The constant matters quantitatively, not just qualitatively — a bound of (1+2δ)2/2(1+\sqrt{2\delta})^2/2 is what the eventual proof delivers, and it is <1< 1 precisely because ∥vi∥2≤δ\|v_i\|^2 \le \delta lets δ→0\delta \to 0 when one needs it (by first subdividing the frame).

The rest of the proof never touches ℓ2\ell^2 or B(ℓ2)B(\ell^2) again. Every step operates on finitely many vectors viv_i and rank-one matrices vivi∗v_i v_i^*: this is why a polynomial-method argument can settle a problem that began life inside C∗C^*-algebra theory.

Terms in this step
Pure state
In operator algebra, a state is a way of assigning an "expected value" to every operator, consistent with the rules of quantum mechanics; a pure state is an extremal, indivisible such assignment, roughly analogous to a single definite physical configuration rather than a mixture of several.
Maximal abelian subalgebra (masa)
A subset of operators that all commute with each other (so they can be measured simultaneously) and that cannot be enlarged while keeping this property; the diagonal operators on ℓ2\ell^2 form one natural example.
Knowledge used in this step