Proof of Basic Properties of Differentiability on an Open Subset of a Real Inner Product Space
lemmalem:frechet-basic-hilbert-2026aThe increment bound and continuity follow from the first-order expansion and the Cauchy-Schwarz inequality; the first-order condition at a local extremum comes from testing the expansion against and ; and restriction is a matter of shrinking the radius to stay inside the smaller open set.
Throughout, denotes where relevant, and we use repeatedly that for : indeed in the vector space , so by Real Inner Product Space §distance and Elementary Identities in a Real Inner Product Space §homogeneity.
Claim 1. Let be the radius supplied by Fréchet Differentiability and the Gradient on an Open Subset of a Real Inner Product Space §differentiable for . Every with satisfies and, by claims 2 and 5 of Properties of the Absolute Value in an Ordered Field and The Cauchy-Schwarz Inequality in a Real Inner Product Space,
Claim 2. Let be the radius supplied by claim 1 for the positive number , and put . Since by Real Inner Product Space §norm and by claim 6 of Elementary Order Arithmetic in an Ordered Field, claim 3 of Elementary Order Arithmetic in an Ordered Field gives .
Let be positive and let be the lesser of and (claim 9 of Elementary Order Arithmetic in an Ordered Field), which is positive because it is one of them. Let satisfy and put , so that and . By claim 1,
the strict step by claim 10 of Elementary Order Arithmetic in an Ordered Field and the last by claim 5 of Elementary Arithmetic in an Ordered Field applied to with the nonnegative multiplier . Hence , and is continuous at relative to by Continuous Map Between Metric Spaces.
If is differentiable on it is differentiable at every point of , hence continuous at every point of relative to , that is continuous on . Members of and of are differentiable on by The Classes and on an Open Subset of a Real Inner Product Space §c1 and The Classes and on an Open Subset of a Real Inner Product Space §c2.
Claim 3. Suppose first that has a local maximum at relative to , and let be a positive real number such that every with satisfies (Local Maximum of a Function Relative to a Subset of a Metric Space). Let be positive, let be the radius supplied by Fréchet Differentiability and the Gradient on an Open Subset of a Real Inner Product Space §differentiable for , and let be the lesser of and (claim 9 of Elementary Order Arithmetic in an Ordered Field), which is positive because it is one of them.
Let satisfy and abbreviate and , so that . Since and , we have . Now , and by claims 3 and 2 of Properties of the Absolute Value in an Ordered Field; adding the inequalities and , which is permitted by the compatibility of the order with addition (an axiom of Ordered Field) together with transitivity, gives
If instead has a local minimum at relative to , then and the same computation, using and , gives .
In either case the resulting inequality holds for every with . Applying it to , which also satisfies by Elementary Identities in a Real Inner Product Space §homogeneity, and using from Elementary Identities in a Real Inner Product Space §bilinear, we obtain both and ; the second gives by claim 4 of Elementary Order Arithmetic in an Ordered Field, so claim 6 of Properties of the Absolute Value in an Ordered Field yields
Since was an arbitrary positive real number, Vanishing of Uniformly Small Linear, Quadratic and Bilinear Terms in a Real Inner Product Space §linear gives .
Claim 4. Since is open in and , Open Subset of a Metric Space provides a positive with ; as , every with satisfies .
Suppose is differentiable at with gradient , let be positive, let be the radius supplied by Fréchet Differentiability and the Gradient on an Open Subset of a Real Inner Product Space §differentiable, and let be the lesser of and , which is positive because it is one of them. Every with satisfies and, because agrees with at and at , the inequality of that clause with in place of . Hence is differentiable at with gradient , and by the uniqueness in Fréchet Differentiability and the Gradient on an Open Subset of a Real Inner Product Space §gradient. This argument used nothing about beyond its lying in the open set and being differentiable there, so it applies verbatim at any point of at which is differentiable: for every subset such that is differentiable at every point of , the restriction is differentiable at every point of with for . In particular, if is differentiable on then is differentiable on with for every .
Now let be a second derivative of at , and let be a radius as in The Second Derivative and the Hessian on an Open Subset of a Real Inner Product Space §expansion: thus and is differentiable at every point of . Let be the lesser of and , which is positive. Then , and by the previous paragraph applied with the restriction is differentiable at every point of with the same gradients as . Given a positive , let be a radius witnessing the condition of that clause for , and , and let be the lesser of and ; then is positive, , and the displayed inequality of that clause holds for all with , with in place of . Hence is a second derivative of at , and by the uniqueness in The Second Derivative and the Hessian on an Open Subset of a Real Inner Product Space §hessian.
Finally, suppose . Then is differentiable on and its gradient map is the restriction of to , which is continuous on by claim 1 of Restriction Stability of Continuity and of the Derivative; hence by The Classes and on an Open Subset of a Real Inner Product Space §c1. If moreover , then by the previous paragraph has at every the second derivative , so its Hessian map is the restriction of to and is continuous on by the same claim; hence by The Classes and on an Open Subset of a Real Inner Product Space §c2, with for every .
Loading…
Prerequisites
19aca83b-7db0-4de1-bdf1-5adc75321b63