Let γ ∈ R \gamma\in\mathbb{R} γ ∈ R be positive and fix z ∈ A z\in A z ∈ A , which is possible because A A A is nonempty.
Preliminaries. The map x ↦ ∣ x ∣ x\mapsto|x| x ↦ ∣ x ∣ is continuous on H H H , since ∣ ∣ x ∣ − ∣ x ′ ∣ ∣ ≤ ∣ x − x ′ ∣ \bigl||x|-|x'|\bigr|\le|x-x'| ∣ x ∣ − ∣ x ′ ∣ ≤ ∣ x − x ′ ∣ by The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity (ii) ; hence, by Continuity Between Metric Spaces is Equivalent to Sequential Continuity and claims 2 and 3 of Arithmetic of Limits of Real Sequences , for every c ∈ R c\in\mathbb{R} c ∈ R the map x ↦ c ∣ x ∣ 2 x\mapsto c\,|x|^{2} x ↦ c ∣ x ∣ 2 is continuous on H H H .
Put B = C − Φ ( z ) + κ 2 B=C-\Phi(z)+\tfrac{\kappa}{2} B = C − Φ ( z ) + 2 κ . From the hypothesis, κ ∣ z ∣ 2 ≤ C − Φ ( z ) \kappa|z|^{2}\le C-\Phi(z) κ ∣ z ∣ 2 ≤ C − Φ ( z ) , and 0 ≤ κ ∣ z ∣ 2 0\le\kappa|z|^{2} 0 ≤ κ ∣ z ∣ 2 by claim 5 of Elementary Arithmetic in an Ordered Field , so 0 ≤ C − Φ ( z ) 0\le C-\Phi(z) 0 ≤ C − Φ ( z ) and hence κ 2 ≤ B \tfrac{\kappa}{2}\le B 2 κ ≤ B by claim 3 of Elementary Arithmetic in an Ordered Field ; since 0 < κ 2 0<\tfrac{\kappa}{2} 0 < 2 κ by claim 8 of Elementary Order Arithmetic in an Ordered Field , claim 2 of that lemma gives 0 < B 0<B 0 < B . Let β \beta β be the nonnegative real number with β 2 = 2 B κ \beta^{2}=\tfrac{2B}{\kappa} β 2 = κ 2 B , given by Existence and Uniqueness of the Nonnegative Square Root . Then β + 4 \beta+4 β + 4 is positive, and we may choose a positive μ ∈ R \mu\in\mathbb{R} μ ∈ R with
μ ≤ κ 2 and 2 μ ( β + 4 ) ≤ γ , \mu\le\tfrac{\kappa}{2}\qquad\text{and}\qquad 2\mu\,(\beta+4)\le\gamma, μ ≤ 2 κ and 2 μ ( β + 4 ) ≤ γ ,
for instance the least of κ 2 \tfrac{\kappa}{2} 2 κ and γ 2 ( β + 4 ) \tfrac{\gamma}{2(\beta+4)} 2 ( β + 4 ) γ , which is one of them and hence positive, by claim 9 of Elementary Order Arithmetic in an Ordered Field .
The auxiliary function. Let Φ μ : A → R \Phi_{\mu}:A\to\mathbb{R} Φ μ : A → R be given by Φ μ ( x ) = Φ ( x ) + μ ∣ x ∣ 2 \Phi_{\mu}(x)=\Phi(x)+\mu|x|^{2} Φ μ ( x ) = Φ ( x ) + μ ∣ x ∣ 2 . By the preliminaries the map g : H → R g:H\to\mathbb{R} g : H → R , g ( x ) = − μ ∣ x ∣ 2 g(x)=-\mu|x|^{2} g ( x ) = − μ ∣ x ∣ 2 , is continuous, and Φ μ ( x ) = Φ ( x ) − g ( x ) \Phi_{\mu}(x)=\Phi(x)-g(x) Φ μ ( x ) = Φ ( x ) − g ( x ) , so Φ μ \Phi_{\mu} Φ μ has closed superlevel sets in H H H by claim 3 of Functions with Closed Superlevel Sets: Sequential Characterisation, Semicontinuity, Perturbation and Limits . Moreover, for x ∈ A x\in A x ∈ A ,
Φ μ ( x ) ≤ C − κ ∣ x ∣ 2 + μ ∣ x ∣ 2 = C − ( κ − μ ) ∣ x ∣ 2 ≤ C , \Phi_{\mu}(x)\le C-\kappa|x|^{2}+\mu|x|^{2}=C-(\kappa-\mu)|x|^{2}\le C, Φ μ ( x ) ≤ C − κ ∣ x ∣ 2 + μ ∣ x ∣ 2 = C − ( κ − μ ) ∣ x ∣ 2 ≤ C ,
since 0 ≤ κ 2 ≤ κ − μ 0\le\tfrac{\kappa}{2}\le\kappa-\mu 0 ≤ 2 κ ≤ κ − μ and 0 ≤ ∣ x ∣ 2 0\le|x|^{2} 0 ≤ ∣ x ∣ 2 , using claim 5 of Elementary Arithmetic in an Ordered Field . So Φ μ \Phi_{\mu} Φ μ is bounded above and sup x ∈ A Φ μ ( x ) \sup_{x\in A}\Phi_{\mu}(x) sup x ∈ A Φ μ ( x ) exists.
A near-maximiser of bounded norm. By claim 3 of Approximation Property of the Supremum and the Infimum in R \mathbb{R} R there is x 0 ∈ A x_{0}\in A x 0 ∈ A with
sup x ∈ A Φ μ ( x ) ≤ Φ μ ( x 0 ) + μ . \sup_{x\in A}\Phi_{\mu}(x)\le\Phi_{\mu}(x_{0})+\mu . x ∈ A sup Φ μ ( x ) ≤ Φ μ ( x 0 ) + μ .
Since Φ μ ( z ) ≤ sup x ∈ A Φ μ ( x ) \Phi_{\mu}(z)\le\sup_{x\in A}\Phi_{\mu}(x) Φ μ ( z ) ≤ sup x ∈ A Φ μ ( x ) and Φ ( z ) ≤ Φ μ ( z ) \Phi(z)\le\Phi_{\mu}(z) Φ ( z ) ≤ Φ μ ( z ) , we get Φ ( z ) − μ ≤ Φ μ ( x 0 ) \Phi(z)-\mu\le\Phi_{\mu}(x_{0}) Φ ( z ) − μ ≤ Φ μ ( x 0 ) . Combining with the displayed bound Φ μ ( x 0 ) ≤ C − ( κ − μ ) ∣ x 0 ∣ 2 ≤ C − κ 2 ∣ x 0 ∣ 2 \Phi_{\mu}(x_{0})\le C-(\kappa-\mu)|x_{0}|^{2}\le C-\tfrac{\kappa}{2}|x_{0}|^{2} Φ μ ( x 0 ) ≤ C − ( κ − μ ) ∣ x 0 ∣ 2 ≤ C − 2 κ ∣ x 0 ∣ 2 gives
κ 2 ∣ x 0 ∣ 2 ≤ C − Φ ( z ) + μ ≤ C − Φ ( z ) + κ 2 = B , \tfrac{\kappa}{2}\,|x_{0}|^{2}\le C-\Phi(z)+\mu\le C-\Phi(z)+\tfrac{\kappa}{2}=B, 2 κ ∣ x 0 ∣ 2 ≤ C − Φ ( z ) + μ ≤ C − Φ ( z ) + 2 κ = B ,
hence ∣ x 0 ∣ 2 ≤ 2 B κ = β 2 |x_{0}|^{2}\le\tfrac{2B}{\kappa}=\beta^{2} ∣ x 0 ∣ 2 ≤ κ 2 B = β 2 and therefore ∣ x 0 ∣ ≤ β |x_{0}|\le\beta ∣ x 0 ∣ ≤ β , both numbers being nonnegative, by claim 1 of Monotonicity of Squaring on the Nonnegative Elements of an Ordered Field and trichotomy.
Applying the perturbed maximum principle. Apply A Perturbed Maximum Principle of Borwein-Preiss Type in a Real Hilbert Space to H H H , A A A , Φ μ \Phi_{\mu} Φ μ , the positive numbers μ \mu μ and λ = 1 \lambda=1 λ = 1 , and the point x 0 x_{0} x 0 , whose defining inequality is the displayed one above since μ λ 2 = μ \mu\lambda^{2}=\mu μ λ 2 = μ . It yields y ˉ ∈ H \bar{y}\in H y ˉ ∈ H and x ˉ ∈ A \bar{x}\in A x ˉ ∈ A such that ∣ y ˉ − x 0 ∣ ≤ 4 |\bar{y}-x_{0}|\le4 ∣ y ˉ − x 0 ∣ ≤ 4 and such that the function x ↦ Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 x\mapsto\Phi_{\mu}(x)-\mu|x-\bar{y}|^{2} x ↦ Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 attains a sequentially strict maximum on A A A at x ˉ \bar{x} x ˉ .
By The Norm Metric of a Real Inner Product Space: Triangle Inequalities, Limits and Continuity (i) , ∣ y ˉ ∣ ≤ ∣ y ˉ − x 0 ∣ + ∣ x 0 ∣ ≤ β + 4 |\bar{y}|\le|\bar{y}-x_{0}|+|x_{0}|\le\beta+4 ∣ y ˉ ∣ ≤ ∣ y ˉ − x 0 ∣ + ∣ x 0 ∣ ≤ β + 4 . Set p = − 2 μ y ˉ p=-2\mu\,\bar{y} p = − 2 μ y ˉ . Then, by Elementary Identities in a Real Inner Product Space (iv) ,
∣ p ∣ = ∣ − 2 μ ∣ ∣ y ˉ ∣ = 2 μ ∣ y ˉ ∣ ≤ 2 μ ( β + 4 ) ≤ γ . |p|=|{-2\mu}|\,|\bar{y}|=2\mu\,|\bar{y}|\le2\mu\,(\beta+4)\le\gamma . ∣ p ∣ = ∣ − 2 μ ∣ ∣ y ˉ ∣ = 2 μ ∣ y ˉ ∣ ≤ 2 μ ( β + 4 ) ≤ γ .
Identifying the perturbation. By Elementary Identities in a Real Inner Product Space (v) , ∣ x − y ˉ ∣ 2 = ∣ x ∣ 2 − 2 ⟨ x , y ˉ ⟩ + ∣ y ˉ ∣ 2 |x-\bar{y}|^{2}=|x|^{2}-2\langle x,\bar{y}\rangle+|\bar{y}|^{2} ∣ x − y ˉ ∣ 2 = ∣ x ∣ 2 − 2 ⟨ x , y ˉ ⟩ + ∣ y ˉ ∣ 2 , so for x ∈ A x\in A x ∈ A
Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 = Φ ( x ) + μ ∣ x ∣ 2 − μ ∣ x ∣ 2 + 2 μ ⟨ x , y ˉ ⟩ − μ ∣ y ˉ ∣ 2 = Φ ( x ) − ⟨ p , x ⟩ − μ ∣ y ˉ ∣ 2 , \Phi_{\mu}(x)-\mu|x-\bar{y}|^{2}=\Phi(x)+\mu|x|^{2}-\mu|x|^{2}+2\mu\langle x,\bar{y}\rangle-\mu|\bar{y}|^{2}=\Phi(x)-\langle p,x\rangle-\mu|\bar{y}|^{2}, Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 = Φ ( x ) + μ ∣ x ∣ 2 − μ ∣ x ∣ 2 + 2 μ ⟨ x , y ˉ ⟩ − μ ∣ y ˉ ∣ 2 = Φ ( x ) − ⟨ p , x ⟩ − μ ∣ y ˉ ∣ 2 ,
where the last step uses ⟨ p , x ⟩ = ⟨ − 2 μ y ˉ , x ⟩ = − 2 μ ⟨ y ˉ , x ⟩ = − 2 μ ⟨ x , y ˉ ⟩ \langle p,x\rangle=\langle-2\mu\bar{y},x\rangle=-2\mu\langle\bar{y},x\rangle=-2\mu\langle x,\bar{y}\rangle ⟨ p , x ⟩ = ⟨ − 2 μ y ˉ , x ⟩ = − 2 μ ⟨ y ˉ , x ⟩ = − 2 μ ⟨ x , y ˉ ⟩ , by conditions (a) and (c) of Real Inner Product Spaces (i) .
Thus the function x ↦ Φ ( x ) − ⟨ p , x ⟩ x\mapsto\Phi(x)-\langle p,x\rangle x ↦ Φ ( x ) − ⟨ p , x ⟩ differs from x ↦ Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 x\mapsto\Phi_{\mu}(x)-\mu|x-\bar{y}|^{2} x ↦ Φ μ ( x ) − μ ∣ x − y ˉ ∣ 2 by the constant μ ∣ y ˉ ∣ 2 \mu|\bar{y}|^{2} μ ∣ y ˉ ∣ 2 , and therefore attains a sequentially strict maximum on A A A at x ˉ \bar{x} x ˉ by claim 3 of Elementary Properties of Sequentially Strict Extrema . Since ∣ p ∣ ≤ γ |p|\le\gamma ∣ p ∣ ≤ γ , this proves the corollary.