Sum estimate A:
ProvedRHLinalg.sum_sq_diff_loweranalysislinear-algebrazeta23
Let be a finite index type, with for all , and suppose the number of indices where is at most . Let be real.
Statement.
This elementary inequality is the first of the two scalar estimates in the proof of the rank–trace inequality (paper reference lem:ranktrace); pointwise it amounts to for , applied only at the at most indices where and summed. In the module Zeta23.LinAlg.RankTrace it is consumed directly by RHLinalg.rank_trace_ineq, where plays the role of the eigenvalues of the positive semidefinite matrix (nonzero at most times) and the negative parts of the eigenvalues of .
Preamble
import Mathlib.Algebra.Order.Rearrangement
import Mathlib.Analysis.Convex.Birkhoff
import Mathlib.Analysis.Matrix.PosDef
open Matrix Finset
open scoped ComplexOrder
variable {𝕜 : Type*} [RCLike 𝕜]
variable {n : Type*} [Fintype n] [DecidableEq n]
variable {ι : Type*} [Fintype ι] [DecidableEq ι]
Formal statement
theorem RHLinalg.sum_sq_diff_lower {p m : ι → ℝ} (hm : ∀ i, 0 ≤ m i)
{r : ℕ} (hr : #{i | p i ≠ 0} ≤ r) {c : ℝ} (hc : 0 ≤ c) :
c * (∑ i, p i) - c ^ 2 / 4 * r - 2 * c * (∑ i, m i) ≤ ∑ i, (p i - m i) ^ 2 := by sorry
Source