The finite-dimensional Schatten -quantity, from singular values
DefinitionHlawkaSchatten_SchattenNormhlawka-schattenoperator-normsschatten-normsingular-values
Fix a scalar field (Lean: RCLike ๐) and finite-dimensional inner product spaces over . Let be -linear, with singular values (Mathlib's LinearMap.singularValues, a sequence that is nonzero for only finitely many ).
singularValuePowerSum p Tโ the sum of the -th powers of the nonzero singular values, for any real . Lean sums over the finite support of the singular-value sequence, so this is a finite sum by construction, not merely because it happens to converge; for it agrees with a sum over every , since , but the two differ at (where the all- reading would count every zero singular value too, via ).schattenPNorm p Tโ the Schatten -quantity. The definition is total in and ; it is a genuine norm only in the intended range , and the analytic hypotheses needed for norm-type facts are placed on the theorems that use them rather than on this definition.schattenPowerSphere pโ the subtype of operators whose singular-value -power sum equals exactly one: the normalization actually used by the Mazur-map constructions, before the outer root that appears in itself.
The bundle also carries five short supporting facts: singularValuePowerSum is nonnegative, and strictly positive when ; consequently schattenPNorm is strictly positive for ; for , exactly when singularValuePowerSum p T = 1 (so the two normalizations above agree); and at , singularValuePowerSum 2 T equals , the real trace of the domain Gram operator.
Definition code
import Mathlib.Analysis.InnerProductSpace.SingularValues
import Mathlib.Analysis.InnerProductSpace.Trace
/-
Copyright (c) 2026 Ezzeri Esa. All rights reserved.
Released under Apache 2.0 license as described in the file LICENSE.
Authors: Ezzeri Esa
-/
/-!
# Finite-dimensional Schatten quantities
This file gives the definition needed by the Hlawka boundary. Mathlib's
singular-value sequence is finitely supported, so the power sum is finite
without choosing bases or matrix dimensions.
-/
namespace HlawkaSchatten
variable {๐ E F : Type*} [RCLike ๐]
[NormedAddCommGroup E] [InnerProductSpace ๐ E] [FiniteDimensional ๐ E]
[NormedAddCommGroup F] [InnerProductSpace ๐ F] [FiniteDimensional ๐ F]
/-- The sum of the `p`-th powers of the singular values of `T`. -/
noncomputable def singularValuePowerSum (p : โ) (T : E โโ[๐] F) : โ :=
โ i โ T.singularValues.support, (T.singularValues i) ^ p
/--
The finite-dimensional Schatten `p` quantity: the `1 / p` power of the
sum of the `p`-th powers of the singular values.
For the norm laws, the intended range is `1 โค p`. Keeping the definition
total makes boundary statements straightforward while putting analytic
hypotheses on the theorems that need them.
-/
noncomputable def schattenPNorm (p : โ) (T : E โโ[๐] F) : โ :=
(singularValuePowerSum p T) ^ (1 / p)
/-- The unit sphere described by a singular-value power sum. This is the
normalization naturally used by the Mazur map before taking the outer
`p`-th root in the Schatten norm. -/
def schattenPowerSphere (p : โ) :=
{T : E โโ[๐] F // singularValuePowerSum p T = 1}
theorem singularValuePowerSum_nonneg (p : โ) (T : E โโ[๐] F) :
0 โค singularValuePowerSum p T := by
unfold singularValuePowerSum
exact Finset.sum_nonneg fun i _ โฆ Real.rpow_nonneg (T.singularValues_nonneg i) _
theorem singularValuePowerSum_pos (p : โ) {T : E โโ[๐] F}
(hT : T โ 0) : 0 < singularValuePowerSum p T := by
unfold singularValuePowerSum
have hsupport : T.singularValues.support.Nonempty := by
rw [Finsupp.support_nonempty_iff]
exact T.singularValues_eq_zero_iff.not.mpr hT
apply Finset.sum_pos'
ยท exact fun i _ โฆ Real.rpow_nonneg (T.singularValues_nonneg i) _
ยท obtain โจi, hiโฉ := hsupport
refine โจi, hi, Real.rpow_pos_of_pos ?_ pโฉ
exact (T.singularValues_pos_iff_ne_zero i).mpr (Finsupp.mem_support_iff.mp hi)
theorem schattenPNorm_pos (p : โ) {T : E โโ[๐] F} (hT : T โ 0) :
0 < schattenPNorm p T :=
Real.rpow_pos_of_pos (singularValuePowerSum_pos p hT) _
/-- For a positive exponent, unit Schatten norm is equivalent to unit
singular-value power sum. -/
theorem schattenPNorm_eq_one_iff {p : โ} (hp : 0 < p) (T : E โโ[๐] F) :
schattenPNorm p T = 1 โ singularValuePowerSum p T = 1 := by
let s := singularValuePowerSum p T
have hs : 0 โค s := singularValuePowerSum_nonneg p T
constructor
ยท intro h
have hspos : 0 < s := by
by_contra hn
have hs0 : s = 0 := le_antisymm (le_of_not_gt hn) hs
unfold schattenPNorm at h
rw [show singularValuePowerSum p T = s by rfl, hs0,
Real.zero_rpow (one_div_ne_zero hp.ne')] at h
norm_num at h
have hpow := congrArg (fun x : โ => x ^ p) h
simp only [Real.one_rpow] at hpow
change (s ^ (1 / p)) ^ p = 1 at hpow
rw [โ Real.rpow_mul hs] at hpow
rw [show 1 / p * p = 1 by field_simp [hp.ne'], Real.rpow_one] at hpow
exact hpow
ยท intro h
unfold schattenPNorm
rw [h, Real.one_rpow]
/-- At exponent two, the singular-value power sum is the real trace of the
domain Gram operator. -/
theorem singularValuePowerSum_two_eq_re_trace (T : E โโ[๐] F) :
singularValuePowerSum 2 T =
RCLike.re ((T.adjoint.comp T).trace ๐ E) := by
have hrank : Module.finrank ๐ T.range โค Module.finrank ๐ E :=
T.finrank_range_le
have hsum :
(โ i โ Finset.range (Module.finrank ๐ T.range), T.singularValues i ^ 2) =
โ i โ Finset.range (Module.finrank ๐ E), T.singularValues i ^ 2 := by
apply Finset.sum_subset (Finset.range_mono hrank)
intro i hi hi'
have hirank : Module.finrank ๐ T.range โค i := by
exact Nat.le_of_not_gt (fun hlt โฆ hi' (Finset.mem_range.mpr hlt))
rw [T.singularValues_eq_zero_iff_le_finrank_range.mpr hirank]
norm_num
unfold singularValuePowerSum
rw [T.support_singularValues]
simp_rw [Real.rpow_two]
rw [T.isSymmetric_adjoint_comp_self.re_trace_eq_sum_eigenvalues rfl]
simp_rw [โ T.sq_singularValues_fin rfl]
exact hsum.trans (Fin.sum_univ_eq_sum_range
(fun i : โ โฆ T.singularValues i ^ 2) (Module.finrank ๐ E)).symm
end HlawkaSchatten
Source
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.