Translations on a Space of Square-Integrable Random Vectors: Constant Classes, Law Invariance, the Translation Derivative and the Translation Hessian
lemmaAnalysisProbabilitylem:translation-lift-2026aThe constant classes form a linear isometry of Euclidean space into the space of square-integrable random vectors, so translation by a constant is an isometry of the parameter into that space. For a law-invariant function the translated function depends only on the law; if the function is continuously differentiable, the translated function is of class with partial derivatives the pairings of the gradient with constant directions; and if it is twice continuously differentiable along translations, the translation Laplacian is the trace of the translation Hessian.
In the setting of Plans, Marginals, Vector Fields and Symmetric Matrices on the Wasserstein Space: Standing Notation, let satisfy , let and let . The real Hilbert space is open in itself by Plans, Marginals, Vector Fields and Symmetric Matrices on the Wasserstein Space: Standing Notation §dimensions, so that the differential calculus fixed there applies to on it: that is differentiable at a point , its gradient there, and the class is as defined there. Let be the constant class with value , and let
Partial derivatives, the classes and on and the Hessian matrix of a function of class are those of Differential Calculus and Convexity on Euclidean Open Sets: Standing Notation §derivatives, the Laplacian is that of The Laplacian of a Twice Continuously Differentiable Function §laplacian, the standard basis vectors of , the initial segment , the origin and the trace are as in Plans, Marginals, Vector Fields and Symmetric Matrices on the Wasserstein Space: Standing Notation §matrices, and the Euclidean norm and Euclidean distance are as fixed there. Then the following hold.
1. (Constant classes and translation)¶ For all and ,
and is the zero vector of . Consequently the map with satisfies and , and is continuous.
2. (Law invariance)¶ Suppose is law-invariant, that definition being read with in place of the dimension named there, and let satisfy . Then . In particular, if is of class on then so is , with the same Hessian matrix at every point.
3. (The translation derivative)¶ Suppose . Then is of class on , with
4. (The translation Hessian and its trace)¶ Suppose is twice continuously differentiable along translations at , that definition being read with in place of the dimension named there; that is, is of class on . Then , and the translation Laplacian of at is
Loading…
Prerequisites
No prerequisites tracked.
Dependents
No dependents yet.
Dependent proofs
No dependent proofs yet.
No relations recorded yet.