10 · Poincaré–Lelong and corner formulas

11 NAG Tags

10.1 Cycles, logarithmic singularities, and the divisor formula

An integration current retains cycle multiplicities

For a closed analytic subspace ZZ of pure codimension rr in a pure nn-dimensional space, its integration current acts on a complementary test by integration over the associated cycle: ⟨δZ,η⟩=∫Zanη,η∈Acn−r,n−r(X).\langle\delta_Z,\eta\rangle=\int_{Z^{\mathrm{an}}}\eta, \qquad \eta\in A_c^{n-r,n-r}(X). Here the notation includes the generic multiplicities of ZZ. For a nonreduced hypersurface defined by fmf^m, the integration current is mδdiv⁡(f)m\delta_{\operatorname{div}(f)} when the usual Cartier-divisor hypotheses hold. This is the analytic counterpart of the doubled tropical hypersurface in Chapter 5.

The current is closed on the boundaryless locus by Stokes. This does not mean every current supported on the same analytic set equals a scalar multiple of δZ\delta_Z. Support and degree constrain a current, but the coefficient structure and differential conditions still matter. The stronger support theorems in CLD25 have additional hypotheses.

The sign and the content of Poincaré–Lelong

For a nonzero meromorphic function under the divisor hypotheses of CLD, the formula is d′d′′[log⁡∣f∣]=δdiv⁡(f).\mathrm d'\mathrm d''[\log|f|]=\delta_{\operatorname{div}(f)}. Equivalently, the logarithmic Green function g=−log⁡∣f∣g=-\log|f| satisfies d′d′′[g]+δdiv⁡(f)=0.\mathrm d'\mathrm d''[g]+\delta_{\operatorname{div}(f)}=0. There are two assertions here. First, log⁡∣f∣\log|f| is locally integrable in the sense needed to define a current. Second, its distributional Hessian is the cycle with the correct multiplicities. A computation away from the zero and pole set proves only that the Hessian vanishes there; it does not determine the singular term.

The proof reduces local analytic geometry to finite maps and one-dimensional slope calculations, while retaining the degree normalization. The zeros and poles contribute precisely their orders. Compatibility with tropical refinements then turns the local computation into an identity on every actual test form. This is why the divisor theorem requires more than the observation that log⁡∣f∣\log|f| is affine wherever ff is a unit.

A projective-line check

On P1,an\mathbb P^{1,\mathrm{an}} with coordinate zz, let ξG\xi_G be the Gauss point and give O(1)\mathcal O(1) its standard model metric. On the torus put t=−log⁡∣z∣t=-\log|z|. The Green functions of 00 and ∞\infty can be written g0=max⁡(0,t),g∞=max⁡(0,−t).g_0=\max(0,t),\qquad g_\infty=\max(0,-t). Their current identities are d′d′′[g0]=δξG−δ0,d′d′′[g∞]=δξG−δ∞.\mathrm d'\mathrm d''[g_0]=\delta_{\xi_G}-\delta_0, \qquad \mathrm d'\mathrm d''[g_\infty]=\delta_{\xi_G}-\delta_\infty. On the finite tropical line, one sees the positive corner at t=0t=0. The missing negative term lies at the appropriate analytic endpoint of the compactification. This is an effective sign check: on a compact boundaryless curve, a global distributional Hessian has total mass zero, whereas the Chern measure of O(1)\mathcal O(1) has total mass one.

Weak convergence: fixed tests and varying tests

For currents, Tj⇀TT_j\rightharpoonup T means ⟨Tj,η⟩→⟨T,η⟩\langle T_j,\eta\rangle\to\langle T,\eta\rangle for each fixed test η\eta. It does not by itself permit replacing η\eta by an arbitrary sequence ηj\eta_j. If TjT_j are positive, the coefficient estimates proved above often provide the missing uniform bound for continuous scalar multipliers.

For example, if uj→uu_j\to u uniformly on a compact neighbourhood of a fixed test support, then ⟨d′d′′[uj]−d′d′′[u],η⟩=∫(uj−u)d′d′′η⟶0.\langle\mathrm d'\mathrm d''[u_j]-\mathrm d'\mathrm d''[u],\eta\rangle =\int (u_j-u)\mathrm d'\mathrm d''\eta\longrightarrow0. The estimate uses the total variation of the signed measure represented by d′d′′η\mathrm d'\mathrm d''\eta. No derivatives of uju_j need converge. Conversely, weak convergence of positive measures does not imply convergence in total variation: on a real interval, δ1/j⇀δ0\delta_{1/j}\rightharpoonup\delta_0, while ∥δ1/j−δ0∥TV=2\|\delta_{1/j}-\delta_0\|_{\mathrm{TV}}=2 for every jj.

Curvature currents and top-degree measures

On a real two-dimensional chart, consider the positive (1,1)(1,1)-forms α=d′x1∧d′′x1,β=d′x2∧d′′x2.\alpha=d'x_1\wedge d''x_1, \qquad \beta=d'x_2\wedge d''x_2. They are different, but their squares both vanish. Thus even the self-top-degree measure need not determine a positive lower-degree current. More refined uniqueness statements, such as uniqueness of a normalized global solution of a Monge–Ampère equation in an ample class, use global geometry and a specified class of potentials. They cannot be inferred from the notation T↦TnT\mapsto T^n alone.

References. [CLD12] (§§4.3–4.6); [Gubler] (Proposition 6.10).

Logarithmic clipping on bounded ordinary tests

The finite localization estimate uses the signed top-form measure of Proposition NAG-002970, the native topology of NAG-002472/002473, and the top-degree vanishing of NAG-002806. The clipping argument below is an application under the native fixed-test Poincaré–Lelong and current hypotheses of NAG-003103 and NAG-003106. Its setting is a specified compact support and a finite ordinary Cartier cover.

Test topology.

Let kk be a complete non-Archimedean field with its given absolute value, XX a good Hausdorff pure nn-dimensional kk-analytic space, n≥1n\ge1, and UU an ordinary open in Int⁡(X/k)\operatorname{Int}(X/k). Use real ordinary smooth moment forms, the structure-sheaf integral, and ordinary compact supports. Write E=Acn−1,n−1(U)E=\mathcal A^{n-1,n-1}_c(U). For an affinoid domain V⊂UV\subset U, a finite tuple gg of units on VV, an open Ω⊃tg(V)\Omega\supset t_g(V) with tg=−log⁡∣g∣t_g=-\log|g|, and an ordinary smooth χ\chi compactly supported in the topological interior of VV in UU, put Qiα=ext⁡0(χ tg∗α),pL,m(α)=max⁡I,J,∣γ∣≤msup⁡v∈L∣∂γαIJ(v)∣(L⋐Ω).Q_i\alpha=\operatorname{ext}_0(\chi\,t_g^*\alpha),\qquad p_{L,m}(\alpha)=\max_{I,J,|\gamma|\le m}\sup_{v\in L} |\partial^\gamma\alpha_{IJ}(v)|\quad(L\Subset\Omega). An empty coefficient space has zero seminorms. NAG-002472 defines τcut\tau_{\rm cut} by all seminorms PP for which PQiP Q_i is continuous for every ii. The ordinary stages instead glue finitely many ambient representatives agreeing on pairwise intersections of the ordinary chart interiors and vanishing on each interior outside a fixed compact contained in their union. Each coefficient subspace has the induced finite product topology and is divided by the kernel of gluing with its quotient topology. Their final locally convex topology τord\tau_{\rm ord} equals τcut\tau_{\rm cut} by NAG-002473. A set B⊂EB\subset E is bounded if sup⁡η∈BP(η)<∞\sup_{\eta\in B}P(\eta)<\infty for every continuous seminorm PP. The proof uses this definition directly on EE.

Finite localization lemma.

Fix a compact S⊂US\subset U and a finite ordinary open cover O1,…,OrO_1,\ldots,O_r of SS in UU. On each OjO_j fix analytic aj,bja_j,b_j which are non-zero-divisors at every stalk. Write Zj=V(aj)∪V(bj)Z_j=V(a_j)\cup V(b_j). Choose ordinary smooth nonnegative ρj\rho_j on UU with compact support in OjO_j and ∑jρj=1\sum_j\rho_j=1 near SS (NAG-002428). Define four compact top forms on OjO_j, restricting all products there, by Tj0η=ρjd′d′′η,Tj1η=d′ρj∧d′′η,Tj2η=−d′′ρj∧d′η,Tj3η=d′d′′ρj∧η.\begin{aligned} T_{j0}\eta&=\rho_j d'd''\eta,& T_{j1}\eta&=d'\rho_j\wedge d''\eta,\\ T_{j2}\eta&=-d''\rho_j\wedge d'\eta,& T_{j3}\eta&=d'd''\rho_j\wedge\eta. \end{aligned} Let μjt(η)\mu_{jt}(\eta) be their respective signed native Radon measures (NAG-002970). For every bounded family B⊂EB\subset E whose tests are supported in SS, there are C,H<∞C,H<\infty such that for all η∈B\eta\in B, ∑j,t∣μjt(η)∣(Oj)≤C,∣μjt(η)∣({min⁡(∣aj∣,∣bj∣)<e−H})=0for every j,t.\sum_{j,t}|\mu_{jt}(\eta)|(O_j)\le C,\qquad |\mu_{jt}(\eta)|\bigl(\{\min(|a_j|,|b_j|)< e^{-H}\}\bigr)=0 \quad\text{for every }j,t . Hence clipping either logarithm by any Borel profile equal to the identity above −H-H, including at −H-H, leaves each weighted signed measure unchanged on the complement of ZjZ_j; extend it by zero on ZjZ_j. The same holds for the difference of the two weighted measures. These are assertions about variation of localized top-form measures. The original tests can meet the divisor.

Proof.

Every output and every derivative used in its definition is local. In particular the support of TjtηT_{jt}\eta is contained in supp⁡ρj∩supp⁡η\operatorname{supp}\rho_j\cap\operatorname{supp}\eta. By NAG-002806 a compact ordinary top form on OjO_j vanishes on an ordinary neighbourhood of the zeros of each regular factor. Thus its compact support is disjoint from ZjZ_j. This assertion so far is per form.

Fix an incoming cutoff stage ii and an index jj. The compact set D=supp⁡χ∩supp⁡ρjD=\operatorname{supp}\chi\cap\operatorname{supp}\rho_j is contained in Int⁡top(V,U)∩Oj\operatorname{Int}_{\rm top}(V,U)\cap O_j. Cover DD by finitely many ordinary interiors of affinoid neighbourhoods in that intersection on which χ,ρj\chi,\rho_j have fixed ordinary moment representatives. Adjoin the restricted tuple gg to their coordinates. Use NAG-002428 on OjO_j for a compactly supported partition subordinate to these interiors, summing to one near DD. Apply that partition after differentiating. NAG-002424 and the coefficient product rule then give, for each tt, a finite expansion with terms ((∂γλαIλJλ)∘tg) βijtλ,∣γλ∣≤2.((\partial^{\gamma_\lambda}\alpha_{I_\lambda J_\lambda}) \circ t_g)\,\beta_{ijt\lambda},\qquad |\gamma_\lambda|\le2 . Indeed the undifferentiated variable form is a sum of coefficients times wedges of coordinate differentials. Each differentiation of a variable coefficient inserts a coordinate differential and an ambient partial derivative. Differentiations of the fixed cutoffs only change fixed factors. All those factors, signs, coordinate wedges and the partition element belong to the fixed compact ordinary top form βijtλ\beta_{ijt\lambda}, extended by zero in OjO_j. If DD is empty take the empty sum. The expansion uses one fixed cover for the whole incoming coefficient space. The tangential relations are preserved by the ordinary operations on the forms.

For a smooth scalar hh defined near supp⁡β\operatorname{supp}\beta, μhβ=hμβ\mu_{h\beta}=h\mu_\beta. Indeed cut hh off by a fixed ordinary function equal to one near the compact support, using NAG-002428 there (include the complementary open away from the support if needed). The two measures give the same values on every smooth scalar test, by multiplication and uniqueness in NAG-002970. Multiplication by a bounded continuous scalar on a compact carrier preserves Radon regularity, since errors on Borel differences are bounded by its supremum times the old variation error. The finite-partition definition of variation gives ∣hμ∣≤∥h∥∞,supp⁡μ∣μ∣|h\mu|\le\|h\|_{\infty,\operatorname{supp}\mu}|\mu| by integrating on each piece and summing, and gives the triangle inequality for sums. Consequently, with Li=tg(V)L_i=t_g(V), ∣μjt(Qiα)∣≤pLi,2(α) νijt,νijt=∑λ∣μβijtλ∣.|\mu_{jt}(Q_i\alpha)|\le p_{L_i,2}(\alpha)\,\nu_{ijt},\qquad \nu_{ijt}=\sum_\lambda|\mu_{\beta_{ijt\lambda}}|. The carrier Cijt=⋃λsupp⁡βijtλC_{ijt}=\bigcup_\lambda\operatorname{supp}\beta_{ijt\lambda} is compact in Oj∖ZjO_j\setminus Z_j by NAG-002806 separately on these finitely many fixed forms. This fixed carrier is the extra input needed for the passage to bounded families.

Take nonnegative locally bounded Borel weights wjw_j on Oj∖ZjO_j\setminus Z_j. The expression Pw(η)=∑j,t∫Oj∖Zjwj d∣μjt(η)∣P_w(\eta)=\sum_{j,t}\int_{O_j\setminus Z_j}w_j\,d|\mu_{jt}(\eta)| is finite on all of EE by the per-form support assertion, and subadditive and absolutely homogeneous by variation. The stage bound is Pw(Qiα)≤pLi,2(α)∑j,t(sup⁡Cijtwj)νijt(Oj),P_w(Q_i\alpha)\le p_{L_i,2}(\alpha) \sum_{j,t}\big(\sup_{C_{ijt}}w_j\big)\nu_{ijt}(O_j), with zero for an empty carrier. The coefficient is finite for each ii. Hence PwP_w is continuous in the original topology and is bounded on BB, without a common presentation of BB. The choice wj=1w_j=1 proves the bound CC.

On Oj∖ZjO_j\setminus Z_j set sj=max⁡(0,−log⁡∣aj∣,−log⁡∣bj∣)s_j=\max(0,-\log|a_j|,-\log|b_j|). If no common HH worked, for every integer k≥1k\ge1 we could choose ηk∈B,jk,tk\eta_k\in B,j_k,t_k with mk=∣μjktk(ηk)∣({sjk>k})>0m_k=|\mu_{j_kt_k}(\eta_k)|(\{s_{j_k}>k\})>0. Then put wj=∑k: jk=j(k/mk)1{sj>k}.w_j=\sum_{k:\,j_k=j}(k/m_k)\mathbf1_{\{s_j>k\}} . These weights are locally bounded Borel functions since sjs_j is finite and continuous off ZjZ_j. But Pw(ηk)≥kP_w(\eta_k)\ge k, a contradiction. Together with the individual zero mass on ZjZ_j this gives HH; the upper logarithms are bounded on the fixed compact supports of the cutoffs. Clipping is therefore stationary in each measure separately. The auxiliary Borel weights occur only in integrals against the already constructed Radon variations.

Finally the graded Leibniz rule gives d′d′′(ρjη)=∑tTjtηd'd''(\rho_j\eta)=\sum_tT_{jt}\eta on OjO_j. After zero extension, summing in jj gives d′d′′ηd'd''\eta for supp⁡η⊂S\operatorname{supp}\eta\subset S, since the partition sums to one near SS. Thus all four summands in localizing a Cartier potential can be estimated before either the partition cancellation or numerator and denominator subtraction. This does not assert that local clipped potentials glue. The bound concerns the native structure-sheaf measure of these ordinary top tests only.

Exact consumer.

NAG-003103, proof lines 240–254 of F8NM, applies scalar Poincaré–Lelong to ρjη\rho_j\eta for each regular factor. Its left side before subtraction is Lcj(d′d′′(ρjη))L_{c_j}(\mathrm d'\mathrm d''(\rho_j\eta)), where cj=ajc_j=a_j or bjb_j. NAG-003106, lines 372–381, uses the positive transpose sign (d′d′′T)(η)=T(d′d′′η)(\mathrm d'\mathrm d''T)(\eta)=T(\mathrm d'\mathrm d''\eta) for a degree-zero current in the native topology. We apply the reviewed cutoff lemma precisely to these operations.

Let kk be complete non-Archimedean with its given absolute value, XX good Hausdorff pure of dimension n≥1n\ge1, and U⊆Int⁡(X/k)U\subseteq\operatorname{Int}(X/k) ordinary open. Use the native structure-cycle integral, ordinary real smooth moment tests, and τa=τcut=τord\tau_a=\tau_{\rm cut}=\tau_{\rm ord} of NAG-002472/002473. Assume the native fixed-test PL inputs of NAG-003103, with their full structure and Cartier multiplicities. For an ambient regular meromorphic ff, fix the finite localization of that proof near a specified compact S⊂US\subset U. Thus f=aj/bjf=a_j/b_j on ordinary opens OjO_j covering SS, in the true interiors of the chosen affinoid neighborhoods as in lines 240–245. Each factor is an ambient non-zero-divisor. The fixed smooth ρj\rho_j have compact support in OjO_j and sum to one near SS. Let ESE_S be the (n−1,n−1)(n-1,n-1)-tests on UU supported in SS, with the topology induced by τa\tau_a.

The actual operation.

Use the smooth profile of NAG-002480. For M≥0M\ge0 set λM(t)=−M+(t+M)θ(t+M+1)\lambda_M(t)=-M+(t+M)\theta(t+M+1), where θ\theta is smooth, between zero and one, zero on (−∞,0](-\infty,0] and one on [1,∞)[1,\infty). Write ℓM(c)=λM(log⁡∣c∣)\ell_M(c)=\lambda_M(\log|c|) off its zero set and −M-M on it. This is ordinary smooth, locally constant near the zeros, and equals log⁡∣c∣\log|c| wherever log⁡∣c∣≥−M\log|c|\ge-M. Put uj,M=ℓM(aj)−ℓM(bj)u_{j,M}=\ell_M(a_j)-\ell_M(b_j). Consider the two continuous linear functionals AM(η)=∑j∫Ojuj,M d′d′′(ρjη),PM(η)=∑j∫Ojuj,M ρjd′d′′η.A_M(\eta)=\sum_j\int_{O_j}u_{j,M}\,\mathrm d'\mathrm d''(\rho_j\eta), \qquad P_M(\eta)=\sum_j\int_{O_j}u_{j,M}\,\rho_j\mathrm d'\mathrm d''\eta . Here AMA_M regularizes the local proof terms. The other operation is PM=(d′d′′TvM)∣ESP_M=(\mathrm d'\mathrm d''T_{v_M})|_{E_S}, where vM=∑jext⁡0(ρjuj,M)v_M=\sum_j\operatorname{ext}_0(\rho_j u_{j,M}) and Tv(α)=∫UvαT_v(\alpha)=\int_U v\alpha. This is the derivative of a specified partition-dependent smooth potential. Raw local clips need not agree on overlaps. For every bounded B⊂ESB\subset E_S there is HBH_B such that for every real M≥HBM\ge H_B and every η∈B\eta\in B, AM(η)=PM(η)=Lf(d′d′′η)=δdiv⁡f(η).A_M(\eta)=P_M(\eta)=L_f(\mathrm d'\mathrm d''\eta)=\delta_{\operatorname{div}f}(\eta). In particular both sequences of functionals (or real-parameter families) converge on bounded sets of the indicated supported space.

Proof.

Let μjt(η)\mu_{jt}(\eta), t=0,1,2,3t=0,1,2,3, represent the top forms ρjd′d′′η, d′ρj∧d′′η, −d′′ρj∧d′η, d′d′′ρj∧η\rho_j\mathrm d'\mathrm d''\eta,\ d'\rho_j\wedge d''\eta,\ -d''\rho_j\wedge d'\eta,\ \mathrm d'\mathrm d''\rho_j\wedge\eta, respectively. Their sum represents d′d′′(ρjη)\mathrm d'\mathrm d''(\rho_j\eta) on OjO_j. Boundedness in ESE_S implies boundedness under its inclusion in the full test space. The independently reviewed finite cutoff lemma supplies CB,HB≥0C_B,H_B\ge0, with total variation summed over j,tj,t at most CBC_B, and zero variation for each measure where either factor has absolute value <e−HB<e^{-H_B}. The original lower-degree test can have support meeting the divisor.

Choose L≥0L\ge0 so both factors on every supp⁡ρj\operatorname{supp}\rho_j have absolute value at most eLe^L, using compactness and finiteness. For either factor separately the logarithm lies in [−HB,L][-H_B,L] almost everywhere for every indicated variation (values on the zero set are immaterial). Thus with D=max⁡(HB,L)D=\max(H_B,L), the sum of the absolute logarithmic integrals for one chosen factor per chart is at most DCBD C_B; the absolute-log tails at levels R>DR>D are zero, uniformly. In particular for every M≥HBM\ge H_B, ∑j,t∑c∈(aj,bj)∫Oj∖(V(aj)∪V(bj))∣ℓM(c)−log⁡∣c∣∣ d∣μjt(η)∣=0.\sum_{j,t}\sum_{c\in(a_j,b_j)} \int_{O_j\setminus(V(a_j)\cup V(b_j))} \bigl|\ell_M(c)-\log|c|\bigr|\,d|\mu_{jt}(\eta)|=0 . The inner sum counts the two factors with repetition. This proves the required uniform integrability in the varying test measures before any signed cancellation. For each native integral this is indeed its Radon realization: the fixed top test vanishes near both zero loci by NAG-002806, so the genuine logarithm on its compact support can be cut off smoothly within the common unit locus. NAG-002970 identifies its smooth-multiplier pairing with the Radon integral. No Borel extension of the native coefficient integral enters.

For fixed MM, the reviewed stage estimate bounds each clipped term by the order-two input coefficient seminorm times the finite mass of a fixed stage measure times the supremum of the fixed multiplier on its compact carrier. The same works for the true logarithm since that carrier avoids the zero loci. Finality gives continuity, hence continuity on ESE_S by restriction. For the limit of AMA_M, all four true-log terms sum to Lf(∑jd′d′′(ρjη))L_f(\sum_j\mathrm d'\mathrm d''(\rho_j\eta)), by linearity and the ordinary restriction of the meromorphic pairing. The inner sum equals d′d′′η\mathrm d'\mathrm d''\eta. For PMP_M only term zero occurs and ∑jρjd′d′′η=d′d′′η\sum_j\rho_j\mathrm d'\mathrm d''\eta=\mathrm d'\mathrm d''\eta. NAG-003103 now identifies the common value with the full Cartier functional. Unit transition factors, if different representatives of the same quotient are used, leave the true pairing unchanged. No equality of these two localized expressions at shallow clipping depth is required.

For the smooth-metric clause of the same consumer take ordinary frames, s=fjejs=f_j e_j, hj=−log⁡∥ej∥h_j=-\log\|e_j\|. Replace both multipliers uj,Mu_{j,M} by hj−uj,Mh_j-u_{j,M}. The hjh_j terms are unchanged throughout the limit. Both resulting functionals eventually equal Gs(d′d′′η)=(Tc1−δDs)(η)G_s(\mathrm d'\mathrm d''\eta)=(T_{c_1}-\delta_{D_s})(\eta) on BB, by the frame gluing and Stokes calculation in NAG-003103, lines 259–285. The sign is also fixed by the norm-one trivial frame.

Native fixed-test identities and the current transfer

The following native proof excerpts use the stated hypotheses on the actual regular factors, on the compact interior tests and on the ordinary smooth metric. The closed-immersion pullback uses the native cutoff topology. The current identities require the fixed-test inputs as stated. Their transitive prerequisite cone is still under review. References to the detached quotient comparison and its two formulae are printed by their permanent identifiers.

Let KK be a complete non-Archimedean field with its specified absolute value, including trivial valuation. Generalized affinoids are allowed. Let XX be good, Hausdorff and pure of analytic dimension n≥1n\geq1, and fix an ordinary open U⊆Int⁡(X/K)U\subseteq\operatorname{Int}(X/K). There is no ambient S1S_1 assumption. All test supports are ordinary compact subsets of UU. Use the frozen ordinary moment sheaf, native integration and positive interleaved convention of NAG-002424, 002425 and 002388. Thus the scalar logarithm has the positive sign La(d′d′′η)=δdiv⁡(a)(η)L_a(\mathrm d'\mathrm d''\eta)=\delta_{\operatorname{div}(a)}(\eta), without an extra constant.

A function is ambient regular when multiplication on OUG\mathcal O_{U_G} is injective; on actual affinoid restrictions this is regularity in their coordinate rings. A regular meromorphic function is a unit of the local total quotient sheaf in the ordinary good-space scope, represented locally by a quotient of ambient regular functions. A line bundle has ordinary local analytic frames, a smooth metric has ordinary smooth weights, and a regular meromorphic section is given data. No existence theorem for such a section is asserted.

The proof inputs are distinguished as follows.

  1. The exact component source has SHA-256

    .

    F8CR-S gives reduced-component S1S_1, compact analytic-interior pullback and differential compatibility. F8CR-N/I give weighted native integration and absolute logarithmic integration. F8CR-C gives full Cartier comparison under its written ZPloc_{\rm loc} premise; F8CR-P gives the analytic scalar identity under ZPloc_{\rm loc} and RPL. The separate nonauthor interface-fit receipt discharges those premises on the actual pure affinoids and reduced components used there. These are reviewed applications, still separate from canonical admission.

  2. Frozen NAG-002772/002778 retain their actual quotient and pure-zero-space scopes. Their effective length comparison and regular-quotient construction do not require ambient S1S_1. The reviewed ZP input supplies empty-or-pure dimension n−1n-1 on all actual common refinements, including regular cross-products.

  3. Frozen NAG-002806 gives neighbourhood vanishing of ordinary top-degree tests near zeros of an ambient regular function. Frozen NAG-002577 supplies restricted Stokes for genuine analytic-interior compact supports. Neither is replaced by a component boundary argument.

  4. The exact reviewed meromorphic source has SHA-256

    .

    Its F8MG-O/T continuity results will be used only on the actual reduced components, where their stated S1S_1 hypothesis holds. The new ambient extension is proved below rather than read into those bytes.

For an affinoid V=M(A)V=\mathcal M(A) pure of dimension nn, set Min⁡A={qi}i=1r,Vi=M(A/qi),ji:Vi↪V,mi=length⁡AqiAqi.\operatorname{Min}A=\{\mathfrak q_i\}_{i=1}^r,\qquad V_i=\mathcal M(A/\mathfrak q_i),\quad j_i:V_i\hookrightarrow V,\qquad m_i=\operatorname{length}_{A_{\mathfrak q_i}}A_{\mathfrak q_i}. All mim_i are positive integers. They are not replaced by one. A general XX is treated on finitely many such neighbourhoods for each compact localization, not by an assumed finite global component list.

NAG-003103 · Open at full scope

Theorem NAG-003103. NAG-003103 (Non-S1S_1 fixed-test identities; F8NM-G). Under the exact reviewed component/scalar and ZP inputs above, every ambient regular meromorphic ff on UU satisfies Lf(d′d′′η)=δdiv⁡(f)(η)(η∈Acn−1,n−1(U)).\begin{equation} L_f(\mathrm d'\mathrm d''\eta)=\delta_{\operatorname{div}(f)}(\eta) \qquad(\eta\in\mathcal A_c^{n-1,n-1}(U)). \end{equation} Let L‾\overline{\mathcal L} have ordinary smooth metric weights and a given ambient regular meromorphic section ss. Its potential pairing GsG_s, full Cartier divisor DsD_s, and ordinary curvature satisfy Gs(d′d′′η)+δDs(η)=∫Uc1(L‾)∧η.\begin{equation} G_s(\mathrm d'\mathrm d''\eta)+\delta_{D_s}(\eta) =\int_U c_1(\overline{\mathcal L})\wedge\eta. \end{equation} These are fixed-test statements before any topology is invoked.

Proof. Choose a finite ordinary quotient atlas near Supp⁡η\operatorname{Supp}\eta and a compact smooth partition ρj\rho_j subordinate to it. Each ηj=ρjη\eta_j=\rho_j\eta is supported in the true analytic interior of a chosen affinoid neighbourhood: use Oj⊂Int⁡top(Vj,X)∩Int⁡(X/K)O_j\subset\operatorname{Int}_{\rm top}(V_j,X)\cap\operatorname{Int}(X/K) and frozen NAG-001630/001704, exactly as in reviewed F8CR-P. Apply that reviewed scalar implication to aja_j and bjb_j, then subtract. Proposition  f8nmTag:prop:quotient f8nmTag:prop:quotient I.0.15 identifies the full Cartier right side. Since η=∑jρjη\eta=\sum_j\rho_j\eta, their mixed derivatives sum to d′d′′η\mathrm d'\mathrm d''\eta. In particular all four terms d′d′′(ρjη)=ρjd′d′′η+d′ρj∧d′′η−d′′ρj∧d′η+d′d′′ρj∧η\mathrm d'\mathrm d''(\rho_j\eta)=\rho_j\mathrm d'\mathrm d''\eta+d'\rho_j\wedge d''\eta -d''\rho_j\wedge d'\eta+\mathrm d'\mathrm d''\rho_j\wedge\eta are retained. The finite sum proves f8nmTag:eq:meromorphic (f8nmTag:eq:meromorphic) (10.1.1) . The component tests are admissible by reviewed F8CR-S, and its pullback commutes with the ordered d′d′′d'd''. This supplies exactly the use of the reduced scalar input; ambient S1S_1 is never inferred.

For the metric identity, choose frames eie_i with ej=uijeie_j=u_{ij}e_i, put hi=−log⁡∥ei∥h_i=-\log\|e_i\|, and write s=fieis=f_i e_i. Then hj=hi−log⁡∣uij∣,fj=fi/uij.h_j=h_i-\log|u_{ij}|,\qquad f_j=f_i/u_{ij}. The unit logarithm is a linear moment coordinate, so its d′d′′d'd'' is zero without using PL. Consequently c1∣Ui=d′d′′hic_1|_{U_i}=\mathrm d'\mathrm d''h_i glues, as do Ds∣Ui=div⁡(fi)D_s|_{U_i}=\operatorname{div}(f_i) and Gs,i(α)=∫hiα−Lfi(α).G_{s,i}(\alpha)=\int h_i\alpha-L_{f_i}(\alpha). The displayed transition equalities cancel on every top test; finite product partitions again prove independence.

For a compact interior test in one frame, frozen restricted Stokes applies to hid′′ηh_i d''\eta and d′hi∧ηd'h_i\wedge\eta. Writing A=∫d′hi∧d′′ηA=\int d'h_i\wedge d''\eta, its two equations are 0=A+∫hid′d′′η,0=−∫d′d′′hi∧η−A.0=A+\int h_i\mathrm d'\mathrm d''\eta,\qquad 0=-\int\mathrm d'\mathrm d''h_i\wedge\eta-A. Thus ∫hid′d′′η=∫d′d′′hi∧η\int h_i\mathrm d'\mathrm d''\eta=\int\mathrm d'\mathrm d''h_i\wedge\eta, with the positive sign. Subtract f8nmTag:eq:meromorphic (f8nmTag:eq:meromorphic) (10.1.1) for fif_i and use a finite partition to obtain f8nmTag:eq:metric (f8nmTag:eq:metric) (10.1.2) . The support and all boundary hypotheses of Stokes remain in force. The norm-one trivial frame gives Gs=−LfG_s=-L_f, independently checking the divisor sign. ◻

Write τa=τcut=τord\tau_a=\tau_{\rm cut}=\tau_{\rm ord} for exactly the frozen final topology of NAG-002472/002473. No intrinsic or external topology is identified here.

NAG-003106 · Open at full scope

Proposition NAG-003106. NAG-003106 (Closed pullback and non-S1S_1 current transfer; F8NM-T). For a closed immersion j:Y↪Xj:Y\hookrightarrow X of good Hausdorff spaces, pullback is continuous on native ordinary compact test spaces with τa\tau_a, including after restriction to an ordinary open U⊆Int⁡(X/K)U\subseteq\operatorname{Int}(X/K) and its inverse image. It preserves compact interior support and commutes with d′d' and d′′d''.

Under the reviewed inputs of Theorem  f8nmTag:thm:fixed f8nmTag:thm:fixed NAG-003103 , LfL_f, GsG_s, δdiv⁡(f)\delta_{\operatorname{div}(f)}, δDs\delta_{D_s} and the smooth curvature pairing are continuous for this topology on UU. The fixed-test identities are therefore native current identities d′d′′Lf=δdiv⁡(f),d′d′′Gs+δDs=Tc1(L‾).\mathrm d'\mathrm d''L_f=\delta_{\operatorname{div}(f)},\qquad \mathrm d'\mathrm d''G_s+\delta_{D_s}=T_{c_1(\overline{\mathcal L})}. This extends ambient scope without enlarging the written S1S_1 statements of the reviewed meromorphic source.

Proof. For a cutoff stage Q=(V,f,Ω,χ)Q=(V,f,\Omega,\chi) on XX, its inverse image W=Y×XVW=Y\times_XV is the actual affinoid closed subspace of VV. The pulled-back tuple is a unit tuple. Closedness of the immersion makes j−1Supp⁡χj^{-1}\operatorname{Supp}\chi compact, and continuity gives j−1Int⁡top(V,X)⊆Int⁡top(W,Y).j^{-1}\operatorname{Int}_{\rm top}(V,X)\subseteq\operatorname{Int}_{\rm top}(W,Y). Hence j∗χj^*\chi is an admissible cutoff on YY, and j∗Qη=Q(W,f∣W,Ω,j∗χ)η.j^*Q\eta=Q_{(W,f|_W,\Omega,j^*\chi)}\eta. The empty inverse image gives zero. This is a continuous single-stage factorization, so finality proves pullback continuity. Reviewed F8CR-S gives the analytic-interior support and differential statements. Applying this argument to the base-changed closed immersion j−1U↪Uj^{-1}U\hookrightarrow U gives the restricted assertion.

First suppose UU is an ordinary open in a pure affinoid VV. Its finitely many reduced components restrict to closed immersions ji−1U→Uj_i^{-1}U\to U, each in the exact S1S_1 scope of F8CR-S. The reviewed F8MG-O/T are therefore applicable to their logarithmic, Cartier and metric pairings. Formulae f8nmTag:eq:logsplit (f8nmTag:eq:logsplit) (eq:logsplit) and f8nmTag:eq:cycles (f8nmTag:eq:cycles) (eq:cycles) express the ambient logarithmic and Cartier pairings as finite sums ∑imi(ji)∗Ti\sum_i m_i(j_i)_*T_i. Each summand is continuous by the pullback factorization just proved. The same decomposition for GsG_s follows locally from F8CR-N for hiαh_i\alpha and f8nmTag:eq:logsplit (f8nmTag:eq:logsplit) (eq:logsplit) , then from a finite frame partition. Smooth scalar and curvature pairings also have their direct fixed-chart coefficient bounds from the frozen native integrator.

For a general UU, test continuity on a fixed cutoff stage. Its support has a fixed finite cover by ordinary opens contained in the interiors of pure affinoid neighbourhoods and by quotient/frame opens. Choose a fixed compact partition there. Multiplying this stage by each partition member and regarding it as a test in its smaller open is continuous: cover its compact support by affinoids inside that open, adjoin the fixed cutoff moments to the original unit tuple, and write the variable coefficient as a fixed smooth coefficient times its pullback by a fixed projection. The product and chain rules give the coefficient bounds. This is precisely the finite-stage factorization used in frozen NAG-002473, without an S1S_1 assumption. On each such open the preceding finite-component argument applies. Their finite sum proves continuity on the original stage, and finality proves continuity on UU. No global finite component decomposition or common presentation for all bounded sets is asserted.

For completeness, the native test differentials are continuous in this same non-S1S_1 category. On a cutoff image use d′(χtf∗η)=d′χ∧tf∗η+χtf∗d′ηd'(\chi t_f^*\eta)=d'\chi\wedge t_f^*\eta+\chi t_f^*d'\eta. The first term has the fixed finite joint-chart factorization just used; the second uses one extra ambient derivative. The identical argument applies to d′′d''. It uses only the frozen ordinary sheaf. With the usual dual sign (−1)p+q+1(-1)^{p+q+1}, their composition on a degree-zero current is d′d′′T(η)=T(d′d′′η)\mathrm d'\mathrm d''T(\eta)=T(\mathrm d'\mathrm d''\eta). The equalities on every test from Theorem  f8nmTag:thm:fixed f8nmTag:thm:fixed NAG-003103 now give exactly the displayed current identities. ◻

10.2 The distributional calculation on an interval

A corner formula is an integration-by-parts identity. Keeping its endpoint terms visible settles the signs and explains why convexity on individual faces is insufficient. We use the normalization ∫a(t)d′t∧d′′t=∫a(t)dt\int a(t)d't\wedge d''t=\int a(t)dt.

Theorem 10.2.1. Let I⊂RI\subset\mathbb R be open and let uu be continuous, with a locally finite subdivision into intervals on whose closures it is C2C^2 from each side. Then d′d′′[u]=u′′(t) dt+∑p(u+′(p)−u−′(p))δp.\mathrm d'\mathrm d''[u]=u''(t)\,dt+\sum_{p}\bigl(u'_+(p)-u'_-(p)\bigr)\delta_p. The density is taken off the subdivision vertices. The current is positive if and only if the density is nonnegative on every open interval and every displayed jump is nonnegative.

Proof. A compact test support meets finitely many pieces. On one piece [a,b][a,b] two integrations by parts give ∫abuϕ′′ dt=∫abu′′ϕ dt+[uϕ′−u′ϕ]ab.\int_a^b u\phi''\,dt =\int_a^b u''\phi\,dt+[u\phi'-u'\phi]_a^b. At an internal vertex pp, continuity of uu cancels the two terms containing ϕ′(p)\phi'(p). The remaining contribution is (u+′(p)−u−′(p))ϕ(p)(u'_+(p)-u'_-(p))\phi(p). The outer endpoints lie off the test support. This proves the formula. For positivity, tests supported inside a single interval detect its density. Shrinking nonnegative tests around one vertex detect its atomic coefficient, because the integral of the locally bounded density tends to zero. Conversely both kinds of coefficients nonnegative make the displayed Radon measure positive. ◻

Example 10.2.2. The function u=−∣t∣u=-|t| is affine on both half-lines, so every facewise Hessian is zero. Its derivative jump is −2-2, and d′d′′[u]=−2δ0\mathrm d'\mathrm d''[u]=-2\delta_0. The function v=−t2v=-t^2 has no corners and d′d′′[v]=−2dt\mathrm d'\mathrm d''[v]=-2dt. Thus a criterion using only facewise convexity misses the first obstruction, and one using only slope sums at vertices misses the second.

10.3 Balanced graphs and endpoint conventions

Let CC be a locally finite rational polyhedral curve in NRN_{\mathbb R}, with constant positive edge weights mem_e. An edge coordinate is measured in its primitive lattice direction, not Euclidean arclength. Assume balancing at all vertices where tests may meet the curve. Tests are restrictions of ambient smooth functions; their directional derivatives at a vertex therefore come from one ambient differential.

Theorem 10.3.1. For a continuous edgewise C2C^2 function uu with one-sided derivatives at vertices, one has d′d′′[u]=∑emeue′′ dte+∑v(∑e∋vme∂eu(v))δv.\mathrm d'\mathrm d''[u]=\sum_e m_eu_e''\,dt_e+ \sum_v\left(\sum_{e\ni v}m_e\partial_eu(v)\right)\delta_v. Here ∂e\partial_e is the outgoing primitive derivative. The identity is on tests supported away from any designated graph boundary.

Proof. Apply the interval calculation on each edge. At a vertex the endpoint contribution is −u(v)∑e∋vme∂eϕ(v)+ϕ(v)∑e∋vme∂eu(v).-u(v)\sum_{e\ni v}m_e\partial_e\phi(v) +\phi(v)\sum_{e\ni v}m_e\partial_eu(v). The first sum is dϕv(∑emene,v)=0d\phi_v(\sum_em_e n_{e,v})=0 by balancing. This leaves exactly the asserted mass. Local finiteness reduces the calculation to finitely many edges for each test. ◻

If balancing is absent at an endpoint and tests are allowed to reach it, the term involving the derivative of the test remains. One must then formulate a boundary-current identity or impose a boundary test convention. It is incorrect to omit this term merely because the underlying graph is compact. In the analytic theory the relevant support condition is avoidance of the analytic boundary.

Corollary 10.3.2. On a compact connected metric graph without boundary, equipped with the usual Kirchhoff test condition, an edgewise C2C^2 continuous function with nonnegative Laplacian is constant.

Proof. Pair the Laplacian with the constant function one. Its total mass is zero. Positivity implies that the measure is zero, hence u′′=0u''=0 on edges and every outgoing slope sum vanishes. Edgewise integration by parts now gives ∑eme∫e(ue′)2dte=−∑eme∫eueue′′dte−∑vu(v)∑e∋vme∂eu(v)=0.\sum_e m_e\int_e(u'_e)^2dt_e =-\sum_e m_e\int_e u_eu_e''dt_e -\sum_v u(v)\sum_{e\ni v}m_e\partial_eu(v)=0. All derivatives vanish, and connectedness identifies the constants. ◻

10.4 An annulus as a sign check

On an annulus use t=−log⁡∣T∣t=-\log|T|. For a real number aa in its skeletal interval and a function pulled back from that interval, d′d′′max⁡(0,t−a)=δa,d′d′′min⁡(0,t−a)=−δa\mathrm d'\mathrm d''\max(0,t-a)=\delta_a,\qquad \mathrm d'\mathrm d''\min(0,t-a)=-\delta_a in the interval calculation. The analytic interpretation uses the annular tropicalization and its calibration. At this stage the formulas describe the tropical measure; identifying it with an analytic current on every analytic test additionally uses the comparison theorem.

For a meromorphic function ff, the analytic Poincaré–Lelong normalization in this book is d′d′′[log⁡∣f∣]=δdiv⁡(f),d′d′′[−log⁡∣f∣]=−δdiv⁡(f).\mathrm d'\mathrm d''[\log|f|]=\delta_{\operatorname{div}(f)},\qquad \mathrm d'\mathrm d''[-\log|f|]=-\delta_{\operatorname{div}(f)}. Consequently, for a meromorphic section s=fes=fe and a metric weight φe=−log⁡∥e∥\varphi_e=-\log\|e\|, −log⁡∥s∥=φe−log⁡∣f∣,d′d′′[−log⁡∥s∥]+δdiv⁡(s)=c1(L,∥⋅∥).-\log\|s\|=\varphi_e-\log|f|,\qquad \mathrm d'\mathrm d''[-\log\|s\|]+\delta_{\operatorname{div}(s)}=c_1(L,\|\cdot\|). This calculation checks the sign of the Green identity. It uses the analytic divisor theorem; the graph calculation alone is not a proof of Poincaré–Lelong in higher dimension.

10.5 Subdivision and normal lifts

The higher-dimensional corner weight has two terms because a primitive normal is a class modulo tangent directions. Let CC be balanced with constant weights and let uu be continuous and affine on its cells. For a facet τ\tau, write ℓσ\ell_\sigma and ℓτ\ell_\tau for the linear parts. Then cτ(u)=∑σ⊃τmσℓσ(nσ,τ)−ℓτ(∑σ⊃τmσnσ,τ).c_\tau(u)=\sum_{\sigma\supset\tau}m_\sigma\ell_\sigma(n_{\sigma,\tau}) -\ell_\tau\left(\sum_{\sigma\supset\tau}m_\sigma n_{\sigma,\tau}\right).

Proposition 10.5.1. This expression is independent of all normal lifts, vanishes for an ambient affine function, and gives zero on a new facet introduced inside a cell where uu was already affine.

Proof. Balancing puts the argument of ℓτ\ell_\tau in its domain. Changing a lift by a∈Nτa\in N_\tau changes the first term by mσℓσ(a)m_\sigma\ell_\sigma(a) and the second by mσℓτ(a)m_\sigma\ell_\tau(a). Continuity gives equality of these restrictions, so the changes cancel. For an ambient affine function all linear parts are restrictions of one functional, and cancellation is immediate. At a subdivision facet inside one cell the two normal classes are opposite, the weights agree, and the linear parts are the same. Choose opposite lifts; both sums then vanish. ◻

This proves the elementary independence statements used in calculations. Balancing of the resulting codimension-one cycle does not follow from lift independence alone. Theorem  f8nmTag:new:corner_balanced f8nmTag:new:corner_balanced 15.5.2 supplies the missing argument in the constant-weight PL setting, using the current identity and the Stokes characterization of balancing.

10.6 The two positive parts of a piecewise smooth Hessian

A piecewise smooth function has an interior Hessian and a corner contribution. Positivity of their sum can be separated because these two measures live in different dimensions. This is stronger than merely testing the Hessian on the interiors of maximal cells. It is the first step in the maximum principle of Gubler–Rabinoff.

Proposition 10.6.1. Let CC be a pure dd-dimensional rational tropical cycle with positive constant weights, and let ff be continuous and smooth on the cells of a locally finite subdivision. Then DD(fδC)=[dP′dP′′f]∧δC+δf⋅C.DD(f\delta_C)=[d'_Pd''_Pf]\wedge\delta_C+\delta_{f\cdot C}. This current is positive if and only if the cellwise Hessian is positive on every maximal cell and every smooth weight of the corner cycle f⋅Cf\cdot C is nonnegative.

Proof. The identity is the piecewise smooth Poincaré–Lelong formula: twice integrating by parts leaves the interior second derivatives and the normal slope jumps, with the tangential correction at each facet. This is the same computation as the PL formula, with the cellwise second derivative retained.

The sufficiency of the two positivity conditions follows by summing positive pairings. For necessity, positivity against tests supported in the interior of one maximal cell gives its Hessian positivity. To isolate the corner, fix a positive compactly supported test η\eta and let ZZ be the union of the facets meeting its support. Choose smooth cutoffs 0≤χj≤10\leq\chi_j\leq 1, equal to one on those facets near the support and tending to zero away from them. Then 0≤DD(fδC)(χjη)=([dP′dP′′f]∧δC)(χjη)+δf⋅C(η).0\leq DD(f\delta_C)(\chi_j\eta) =([d'_Pd''_Pf]\wedge\delta_C)(\chi_j\eta) +\delta_{f\cdot C}(\eta). The first term tends to zero. Indeed, it is integration against a signed measure with smooth densities on the dd-cells, and its total variation gives zero mass to their lower-dimensional faces. Dominated convergence for that total variation applies. This argument does not require applying monotone convergence directly to a signed measure. Thus the corner current is positive. Tests near the relative interior of a facet detect its scalar weight, proving nonnegativity of every weight; continuity extends the conclusion to facet boundaries. ◻

10.7 The tropical strong maximum principle

Theorem 10.7.1. In the setting of Proposition  f8nmTag:new:pws_positivity_split f8nmTag:new:pws_positivity_split 10.6.1 , if DD(fδC)≥0DD(f\delta_C)\geq 0 and ff has a local maximum at an interior point xx of the cycle, then ff is constant in a neighbourhood of xx in CC.

Proof. Translate xx to zero and restrict to a small ball. The local cycle is the intersection of a rational fan with the ball, and we may assume f≤f(0)f\leq f(0) throughout it. We prove the stronger claim that ff is constant on this entire small fan neighbourhood, and argue by induction on dd. Dimension zero is immediate. In dimension one, write ses_e for the outgoing derivative on a ray and me>0m_e>0 for its weight. The maximum gives se≤0s_e\leq 0, while positivity of the corner gives ∑emese≥0\sum_em_es_e\geq 0. Hence every se=0s_e=0. The nonnegative second derivative on each ray makes its derivative nondecreasing from zero. Thus f≥f(0)f\geq f(0) on each ray, while the maximum gives the reverse inequality. The function is constant.

Suppose d≥2d\geq 2 and ff is not constant. By continuity and density of rational points in the rational fan, choose a rational point yy with f(y)<f(0)f(y)<f(0), in the relative interior of a maximal cone and away from the finitely many unwanted face spans. Choose a rational linear hyperplane HH through 00 and yy, otherwise generic for the fan. The stable intersection C′=C⋅HC'=C\cdot H is an effective cycle of pure dimension d−1d-1, with support equal to C∩HC\cap H near these points. This is the proper-intersection case of the tropical displacement theorem.

The restricted cellwise Hessian stays positive. The divisor-intersection identity gives (f∣C′)⋅C′=(f⋅C)⋅H.(f|_{C'})\cdot C'=(f\cdot C)\cdot H. The right side has nonnegative weights: the corner cycle is effective and the intersection coefficients with the linear hyperplane are nonnegative lattice indices. Proposition  f8nmTag:new:pws_positivity_split f8nmTag:new:pws_positivity_split 10.6.1 therefore shows that f∣C′f|_{C'} is weakly tropically psh. The induction hypothesis, applied on the same fan neighbourhood, makes it constant there. This contradicts f(y)<f(0)f(y)<f(0) and proves the result. ◻

The positive weights and balancing have different jobs. Positive weights force each outgoing derivative to vanish in dimension one. Balancing makes the corner formula and the restriction by a tropical hyperplane compatible. A collection of convex functions on separate cells, without the corner inequalities, need not satisfy this maximum principle.

10.8 Separating an analytic point from the boundary

To apply the tropical theorem to an analytic function, it is not enough that the chosen analytic point is interior. Its tropical value could also be attained at a boundary point. Additional coordinates remove this overlap.

Lemma 10.8.1. Let XX be a strict affinoid space and let x∈Int⁡(X)x\in\operatorname{Int}(X). There are a strict affinoid neighbourhood UU of xx and analytic units g1,…,gsg_1,\ldots,g_s on UU such that the tropicalization b=(−log⁡∣gi∣)ib=(-\log|g_i|)_i satisfies b(x)∉b(∂U).b(x)\notin b(\partial U).

Proof. The affinoid boundary lemma of CLD12 supplies finitely many analytic functions gig_i on XX with ∣gi∣≤1|g_i|\leq 1, ∣gi(x)∣<1|g_i(x)|<1, and with at least one ∣gi(y)∣=1|g_i(y)|=1 at every y∈∂Xy\in\partial X. If gi(x)=0g_i(x)=0, add a nonzero constant of norm less than one; this preserves all three properties and makes the value at xx nonzero. This boundary lemma is the geometric input.

Choose rr in the positive value group, larger than every −log⁡∣gi(x)∣-\log|g_i(x)|, and set U=⋂i{∣gi∣≥e−r}U=\bigcap_i\{|g_i|\geq e^{-r}\}. It is a strict Laurent domain and a neighbourhood of xx. The functions gig_i are units on UU, and b(U)⊂[0,r]sb(U)\subset[0,r]^s. Points inherited from ∂X\partial X have at least one coordinate zero. Any new boundary point of the Laurent domain has at least one coordinate rr, by the relative-boundary formula. Hence b(∂U)b(\partial U) lies in the boundary of the box, while b(x)∈(0,r)sb(x)\in(0,r)^s. ◻

Theorem 10.8.2. (The analytic maximum principle in the piecewise smooth class). Let XX be a good strict analytic space of pure dimension over a complete nontrivially valued field. If a piecewise smooth function hh is current-psh for the original CLD tests and has a local maximum at an interior point xx, then it is locally constant at xx.

Proof. Choose a strict affinoid neighbourhood on which h=F∘ah=F\circ a for a moment tropicalization aa and a piecewise smooth FF, and shrink so that h≤h(x)h\leq h(x). Adjoin the boundary-separating coordinates of Lemma  f8nmTag:new:boundary_separating_units f8nmTag:new:boundary_separating_units 10.8.1 ; the new chart is a′=(a,b)a'=(a,b), and FF is pulled back from the first factor. Write ω=a′(x)\omega=a'(x). Then ω∉a′(∂U)\omega\notin a'(\partial U).

By Theorem  f8nmTag:new:affinoid_purity_theorem f8nmTag:new:affinoid_purity_theorem 6.10.2 , the image has pure dimension dim⁡X\dim X near ω\omega. The analytic calibration and balancing theorem equips it there with positive tropical multiplicities. For a positive compactly supported superform η\eta on this image, with support avoiding the tropicalized boundary, its pullback is an admissible positive CLD test. The integration formula gives DD(Fδa′(U))(η)=DD[h]((a′)∗η)≥0.DD(F\delta_{a'(U)})(\eta)=DD[h]((a')^*\eta)\geq0. The tropical function has a maximum at ω\omega, because h≤h(x)h\leq h(x) on all of UU. Theorem  f8nmTag:new:tropical_strong_maximum f8nmTag:new:tropical_strong_maximum 10.7.1 makes FF constant near ω\omega in the tropical image. Pulling back proves the assertion. ◻

The proof uses neither a converse from current positivity to model semipositivity nor a general smooth regularization theorem. It applies directly to the piecewise smooth class. On a compact connected boundaryless space in this setting, every such psh function is constant: its maximum set is nonempty and closed, and the local theorem makes that set open. This conclusion concerns scalar functions; a semipositive metric on a nontrivial line bundle has transition functions and is not a single global psh scalar.

References. [GRPL] (Section 7, Proposition 7.4 and Theorem 7.7); the boundary-separating input is [CLD12] (Lemma 3.4.3). The pure-dimensional analytic case is stated here so that no restriction-to-components step is implicit.