OctoInform Beta

OpenAI preprint, September 2026 · §2. Pure states, excision, and peaking projections · §2

Lemma 2.5 [Projection excision]: For projections (pn)(p_n) peaking to a state hh, ∥pn(a−h(a)1)pn∥→0\|p_n(a - h(a)1)p_n\| \to 0 for every aa, and ∥pnapt∥→0\|p_n a p_t\| \to 0 as n,t→∞n, t \to \infty when h(a)=0h(a) = 0

Peaking projections excise the state they peak to.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let BB be a C*-algebra with unit 11, let hh be a state of BB, and let p0≥p1≥p2≥⋯p_0 \ge p_1 \ge p_2 \ge \cdots be projections in BB with h(pn)=1h(p_n) = 1 for every nn. Cutting an element aa of BB down to the corner pnBpnp_n B p_n gives its compression pnapnp_n a p_n. Decreasing projections (pn)(p_n) in BB excise a state hh of BB when, for every a∈Ba \in B, the compression pnapnp_n a p_n approaches the multiple h(a)pnh(a)p_n of pnp_n in norm as n→∞n \to \infty. The question is how close this compression comes to the multiple h(a)pnh(a)p_n of pnp_n as nn grows, and, when h(a)=0h(a) = 0, how small the off-diagonal pieces pnaptp_n a p_t become for large nn and tt.

The library answers both questions when hh is the only state of BB that takes the value 11 at every pnp_n.

States and Peaking Sequences

A state hh of a C*-algebra BB is a continuous linear functional on BB of norm one such that h(a*a)h(a^*a) is real and nonnegative for every aa in BB: ∥h∥=1,h(a*a)≥0for every a∈B.\|h\| = 1, \qquad h(a^*a) \ge 0 \quad \text{for every } a \in B.(1)✦

A decreasing sequence (pn)n≥0(p_n)_{n \ge 0} of projections in a C*-algebra BB, so that pn+1≤pnp_{n+1} \le p_n for every nn, peaks to a state hh of BB when hh takes the value 11 at every pnp_n and is the only state of BB that does: {g a state of B:g(pn)=1 for every n}={h}.\{\, g \text{ a state of } B : g(p_n) = 1 \text{ for every } n \,\} = \{h\}.(2)✦

The Claim

Theorem 1. Let BB be a C*-algebra and let (pn)(p_n) be a decreasing sequence of projections in BB that peaks to a state hh of BB. Then, for every element a∈Ba \in B, limn→∞∥pn(a−h(a)1)pn∥=0.\lim_{n \to \infty} \bigl\| p_n \bigl(a - h(a)1\bigr) p_n \bigr\| = 0.(3)✦ Furthermore, for every self-adjoint element a∈Ba \in B, limn→∞∥pn(a−h(a)1)pn∥=0,\lim_{n \to \infty} \bigl\| p_n \bigl(a - h(a)1\bigr) p_n \bigr\| = 0,(4)✦ If moreover h(a)=0h(a) = 0, the double sequence (pnapt)n,t≥0(p_n a p_t)_{n, t \ge 0} tends to zero in norm as nn and tt tend to infinity independently: limn,t→∞∥pnapt∥=0.\lim_{n, t \to \infty} \| p_n a p_t \| = 0.(5)✦

In the paper, the result reads: if decreasing projections (pn)(p_n) in BB peak to hh, then pn↓rhp_n \downarrow r_h strongly, where rhr_h is the support projection of hh, and ∥pn(a−h(a)1)pn∥→0\|p_n(a - h(a)1)p_n\| \to 0 for every a∈Ba \in B, so that ∥pnapt∥→0\|p_n a p_t\| \to 0 on the product index set when h(a)=0h(a) = 0, the paper writing PnP_n for pnp_n.

What Is Checked

The statements (3), (4) and (5) are stated over an arbitrary C*-algebra BB, a sequence of projections in BB that peaks to a state hh of BB in the sense of (1) and (2), and an element aa of BB, taken self-adjoint in (4) and with h(a)=0h(a) = 0 in (5).