We work in the settings of Real Hilbert Spaces: Standing Notation and Background , Real Hilbert Spaces: Series, Products, Orthonormal Bases and Differential Calculus and Differential Calculus and Convexity on Euclidean Open Sets: Standing Notation , the last together with Euclidean Space and Real Matrices: Standing Notation on which it rests, used in the dimension m m m for a natural number m m m with 1 ≤ m 1\le m 1 ≤ m .
Let H H H be a real Hilbert space , with its inner product ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ and norm ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ as fixed there, let R m \mathbb{R}^{m} R m carry its Euclidean norm ∥ ⋅ ∥ \lVert\cdot\rVert ∥ ⋅ ∥ and Euclidean distance d E d_{E} d E , and regard R m \mathbb{R}^{m} R m as the metric space ( R m , d E ) (\mathbb{R}^{m},d_{E}) ( R m , d E ) . We write ∥ ζ ∥ 2 = ∥ ζ ∥ ∥ ζ ∥ \lVert\zeta\rVert^{2}=\lVert\zeta\rVert\lVert\zeta\rVert ∥ ζ ∥ 2 = ∥ ζ ∥ ∥ ζ ∥ , and α 2 \tfrac{\alpha}{2} 2 α for the quotient of α ∈ R \alpha\in\mathbb{R} α ∈ R by 2 = 1 + 1 2=1+1 2 = 1 + 1 .
Let A A A be a linear subspace of H H H , let e ∈ H m e\in H^{m} e ∈ H m be an m m m -tuple in H H H that is orthonormal and all of whose components lie in A A A , and let Λ : H → R m \Lambda:H\to\mathbb{R}^{m} Λ : H → R m be the coordinate map determined by e e e .
Let u ^ , v ^ : A → R \hat{u},\hat{v}:A\to\mathbb{R} u ^ , v ^ : A → R and let α ∈ R \alpha\in\mathbb{R} α ∈ R be positive. Assume that the set of values of u ^ \hat{u} u ^ is bounded above and that the set of values of v ^ \hat{v} v ^ is bounded below . Let x ˉ , y ˉ ∈ A \bar{x},\bar{y}\in A x ˉ , y ˉ ∈ A and assume that
u ^ ( x ) − v ^ ( y ) − α 2 ∥ Λ x − Λ y ∥ 2 ≤ u ^ ( x ˉ ) − v ^ ( y ˉ ) − α 2 ∥ Λ x ˉ − Λ y ˉ ∥ 2 for all x , y ∈ A . \hat{u}(x)-\hat{v}(y)-\tfrac{\alpha}{2}\lVert\Lambda x-\Lambda y\rVert^{2}
\ \le\
\hat{u}(\bar{x})-\hat{v}(\bar{y})-\tfrac{\alpha}{2}\lVert\Lambda\bar{x}-\Lambda\bar{y}\rVert^{2}
\qquad\text{for all }x,y\in A . u ^ ( x ) − v ^ ( y ) − 2 α ∥ Λ x − Λ y ∥ 2 ≤ u ^ ( x ˉ ) − v ^ ( y ˉ ) − 2 α ∥ Λ x ˉ − Λ y ˉ ∥ 2 for all x , y ∈ A .
Put ζ ˉ = Λ x ˉ \bar{\zeta}=\Lambda\bar{x} ζ ˉ = Λ x ˉ and ω ˉ = Λ y ˉ \bar{\omega}=\Lambda\bar{y} ω ˉ = Λ y ˉ , and for ζ ∈ R m \zeta\in\mathbb{R}^{m} ζ ∈ R m put A ζ = { x ∈ A : Λ x = ζ } A_{\zeta}=\{\,x\in A\ :\ \Lambda x=\zeta\,\} A ζ = { x ∈ A : Λ x = ζ } . Then the following hold.
1. (The fibre suprema) (i) For every ζ ∈ R m \zeta\in\mathbb{R}^{m} ζ ∈ R m the set A ζ A_{\zeta} A ζ is nonempty, the set { u ^ ( x ) : x ∈ A ζ } \{\hat{u}(x):x\in A_{\zeta}\} { u ^ ( x ) : x ∈ A ζ } has a least upper bound and the set { v ^ ( y ) : y ∈ A ζ } \{\hat{v}(y):y\in A_{\zeta}\} { v ^ ( y ) : y ∈ A ζ } has a greatest lower bound, unique by Existence of the Infimum of a Nonempty Subset of R \mathbb{R} R Bounded Below and Uniqueness of the Supremum and of the Infimum . The functions U , V : R m → R U,\mathcal{V}:\mathbb{R}^{m}\to\mathbb{R} U , V : R m → R given by
U ( ζ ) = sup { u ^ ( x ) : x ∈ A ζ } , V ( ζ ) = inf { v ^ ( y ) : y ∈ A ζ } U(\zeta)=\sup\{\hat{u}(x):x\in A_{\zeta}\},
\qquad
\mathcal{V}(\zeta)=\inf\{\hat{v}(y):y\in A_{\zeta}\} U ( ζ ) = sup { u ^ ( x ) : x ∈ A ζ } , V ( ζ ) = inf { v ^ ( y ) : y ∈ A ζ }
are therefore defined; the set of values of U U U is bounded above and that of V \mathcal{V} V is bounded below. Consequently U U U is bounded above near each point of R m \mathbb{R}^{m} R m and V \mathcal{V} V is bounded below near each point of R m \mathbb{R}^{m} R m .
2. (Envelopes and the maximum) (ii) Let U ∗ U^{*} U ∗ be the upper semicontinuous envelope of U U U and let V ∗ \mathcal{V}_{*} V ∗ be the lower semicontinuous envelope of V \mathcal{V} V , both taken in the metric space ( R m , d E ) (\mathbb{R}^{m},d_{E}) ( R m , d E ) . Then U ∗ U^{*} U ∗ is upper semicontinuous on R m \mathbb{R}^{m} R m , V ∗ \mathcal{V}_{*} V ∗ is lower semicontinuous on R m \mathbb{R}^{m} R m , and
U ∗ ( ζ ) − V ∗ ( ω ) − α 2 ∥ ζ − ω ∥ 2 ≤ U ∗ ( ζ ˉ ) − V ∗ ( ω ˉ ) − α 2 ∥ ζ ˉ − ω ˉ ∥ 2 for all ζ , ω ∈ R m . U^{*}(\zeta)-\mathcal{V}_{*}(\omega)-\tfrac{\alpha}{2}\lVert\zeta-\omega\rVert^{2}
\ \le\
U^{*}(\bar{\zeta})-\mathcal{V}_{*}(\bar{\omega})-\tfrac{\alpha}{2}\lVert\bar{\zeta}-\bar{\omega}\rVert^{2}
\qquad\text{for all }\zeta,\omega\in\mathbb{R}^{m}. U ∗ ( ζ ) − V ∗ ( ω ) − 2 α ∥ ζ − ω ∥ 2 ≤ U ∗ ( ζ ˉ ) − V ∗ ( ω ˉ ) − 2 α ∥ ζ ˉ − ω ˉ ∥ 2 for all ζ , ω ∈ R m .
3. (Values at the base point) (iii) U ∗ ( ζ ˉ ) = U ( ζ ˉ ) = u ^ ( x ˉ ) U^{*}(\bar{\zeta})=U(\bar{\zeta})=\hat{u}(\bar{x}) U ∗ ( ζ ˉ ) = U ( ζ ˉ ) = u ^ ( x ˉ ) and V ∗ ( ω ˉ ) = V ( ω ˉ ) = v ^ ( y ˉ ) \mathcal{V}_{*}(\bar{\omega})=\mathcal{V}(\bar{\omega})=\hat{v}(\bar{y}) V ∗ ( ω ˉ ) = V ( ω ˉ ) = v ^ ( y ˉ ) .
4. (Approximation over a fibre neighbourhood) (iv) Let ζ 1 ∈ R m \zeta_{1}\in\mathbb{R}^{m} ζ 1 ∈ R m and let r , ε ∈ R r,\varepsilon\in\mathbb{R} r , ε ∈ R be positive. Then there are x ∈ A x\in A x ∈ A and y ∈ A y\in A y ∈ A with
∥ Λ x − ζ 1 ∥ ≤ r , U ∗ ( ζ 1 ) − ε < u ^ ( x ) , ∥ Λ y − ζ 1 ∥ ≤ r , v ^ ( y ) < V ∗ ( ζ 1 ) + ε . \lVert\Lambda x-\zeta_{1}\rVert\le r,
\qquad
U^{*}(\zeta_{1})-\varepsilon<\hat{u}(x),
\qquad
\lVert\Lambda y-\zeta_{1}\rVert\le r,
\qquad
\hat{v}(y)<\mathcal{V}_{*}(\zeta_{1})+\varepsilon . ∥ Λ x − ζ 1 ∥ ≤ r , U ∗ ( ζ 1 ) − ε < u ^ ( x ) , ∥ Λ y − ζ 1 ∥ ≤ r , v ^ ( y ) < V ∗ ( ζ 1 ) + ε .