Review pending

Proposition 19.18.6 · PL weights are admitted differences

Proposition 19.18.6. (PL weights are admitted differences). Let XX be Hausdorff and almost good, and let h1,…,hsh_1,\ldots,h_s be finitely many real-valued PL functions. Every point has one open neighbourhood UU on which hi=ai−bi,ai,bi∈UX(U).h_i=a_i-b_i,\qquad a_i,b_i\in\mathcal U_X(U). The functions ai,bia_i,b_i can be chosen as pullbacks of finite maxima of affine functions under one analytic moment. They have decreasing smooth psh approximants on this same UU, with an explicit uniform error bound.

Argument in the manuscript

Proof. We use the local chart and polyhedral decomposition results of CLD25 Proposition 12.4.7, Proposition 6.3.4 and Proposition 8.8.2(b). Almost goodness makes the associated affine tropical space decent. After taking a compact domain neighbourhood and concatenating the finitely many moments, these results write each hih_i as the difference of two convex PL functions on the convex hull of the tropical image. A convex PL function on a finite polytope is a finite maximum of affine functions there. Extend those affine functions to the ambient real vector space. Their maximum is defined on that whole vector space and has the required restriction. Shrinking to the interior of the chosen domain gives one open UU for all the functions.

For v=max⁡1≤j≤Nℓjv=\max_{1\leq j\leq N}\ell_j, put vϵ(t)=ϵlog⁡ ⁣(∑j=1Neℓj(t)/ϵ),ϵ>0.v_\epsilon(t)=\epsilon\log\!\left(\sum_{j=1}^N e^{\ell_j(t)/\epsilon}\right),\qquad \epsilon>0. Then v≤vϵ≤v+ϵlog⁡Nv\leq v_\epsilon\leq v+\epsilon\log N. If pj=eℓj/ϵ/∑keℓk/ϵp_j=e^{\ell_j/\epsilon}/\sum_k e^{\ell_k/\epsilon} and bjb_j is the gradient of ℓj\ell_j, then ∇2vϵ=ϵ−1(∑jpjbjbjt−(∑jpjbj)(∑jpjbj)t)≥0,\nabla^2v_\epsilon =\epsilon^{-1}\left(\sum_jp_j b_jb_j^t -\Bigl(\sum_jp_jb_j\Bigr)\Bigl(\sum_jp_jb_j\Bigr)^t\right) \geq0, ∂vϵ∂ϵ=−∑jpjlog⁡pj≥0.\frac{\partial v_\epsilon}{\partial\epsilon} =-\sum_jp_j\log p_j\geq0. Thus vϵ↓vv_\epsilon\downarrow v as ϵ↓0\epsilon\downarrow 0. Pullback by the fixed moment produces original smooth psh functions on UU, with the same uniform error bound. Apply this construction to both summands of every hih_i. No assertion about later ordinal stages is used. ◻

Citation

Non-Archimedean Geometry Foundations, Tag NAG-900367.

Proof in chapter