Conditional Gaussian score
DefinitionZOGlobalLB_Supplement_adaptiveScoregaussian-processeslikelihoodp2o-gran-per-chapterp2o-plan-paperp2o-v1probability
At a Gaussian round of an adaptive transcript, with conditional response mean and conditional variance , the conditional score is
At a parameter-independent deterministic round, . The realized response history is held fixed when the gradient with respect to the candidate parameter is taken.
Formalization Note The parameter is represented in a Euclidean coordinate chart, matching the tangent-coordinate expression in the theorem.
Definition code
import Mathlib
import Definitions.Def_ZOGlobalLB_Admissible
import Definitions.Def_ZOGlobalLB_Supplement_transcriptPrefix
namespace ZOGlobalLB
namespace Supplement
/-- The round-`i` conditional score in Xia--Yuan, arXiv:2610.19652,
Theorem A.4, p.34: `(Y_i - m_i(v)) / s_i²` times the gradient of the conditional mean,
and zero on a parameter-independent deterministic branch.
**Formalization Note.** The parameter is represented in a Euclidean coordinate chart `E p`;
the theorem is local in the candidate parameter and its score, so this is the tangent-coordinate
form of the paper's smooth-manifold notation. Histories are held fixed by applying `gradient` only
to the candidate-parameter argument. -/
noncomputable def adaptiveScore {n p : ℕ} {Ω : Type*} (v : E p)
(branch : (i : Fin n) → (Fin i.val → ℝ) → Bool)
(variance : (i : Fin n) → (Fin i.val → ℝ) → ℝ)
(mean : (i : Fin n) → (Fin i.val → ℝ) → E p → ℝ)
(Y : Ω → Fin n → ℝ) (i : Fin n) (ω : Ω) : E p :=
if branch i (transcriptPrefix Y i ω) then
((Y ω i - mean i (transcriptPrefix Y i ω) v) /
variance i (transcriptPrefix Y i ω)) •
gradient (mean i (transcriptPrefix Y i ω)) v
else 0
end Supplement
end ZOGlobalLB
Source
Xia, Yuan, On Global Joint Lower Bounds in Zeroth-Order Nonconvex Optimization, arXiv:2610.19652, Theorem A.4, p.34