Lemma 11.20.1. (Controlled convex extension).
Let be as in Lemma 11.19.12. There is with the following property. Let be real affine, let , and let be continuous on , a finite maximum of rational-slope, -constant affine functions on each facet. Suppose Then has a finite-max extension with the same coefficient conditions and
Argument in the manuscript
Proof. Fix a facet and a compatible affine barrier vanishing exactly on . In a finite-max representation of , discard branches never active on a relatively open subset. This does not change the function, including at its boundary, by continuity. For a remaining branch , the tangential part of on has norm at most . The tangent space of is rational. Its orthogonal complement therefore has dense rational points. Add to a function , where is a vertex and is rational, orthogonal to the tangent space, chosen so that the new branch satisfies The added constant is in . For a zero-dimensional facet the same construction uses its single boundary value and no tangential constraint. At a point where the original branch is active, . If , it follows that on .
We next bound the penalty needed to dominate every other facet. If , then for every one can choose such that with a fixed finite constant. Indeed, write as a convex combination of the vertices of , keep the vertices in , and replace all other vertices by one fixed vertex . One may take the largest ratio over vertices outside . Since and the function is -Lipschitz on , If the two facets are disjoint, has a positive minimum on , and the uniform bounds above give the same inequality with another fixed constant. Thus, independently of the branch and the boundary subdivision, there is such that every modified branch is dominated on the boundary after subtracting , where is any rational number in .
Take the maximum of these penalized branches over all facets. It extends exactly and retains the coefficient conditions. Every branch has gradient within of , where depends only on the chosen facet barriers. Hence , a maximum of these affine differences, is -Lipschitz on . Each branch is at most at the vertices, so throughout . Comparing with any boundary point gives . Taking proves the lemma. ◻