9 · Currents, order zero and continuous multipliers

29 NAG Tags

9.1 Test spaces, currents, and their operations

Definition 9.1.1. (Test forms). Write Asm,cp,q(X)A^{p,q}_{\mathrm{sm},c}(X) for the original smooth forms whose support is compact and disjoint from the analytic boundary ∂X\partial X. The topology is determined by finite tropical presentations: on a fixed finite family of tropical charts, with support in fixed compact subsets, the coefficients of smooth representatives and every derivative of their coefficients converge uniformly. Equivalent representatives are given the quotient topology, and the test space is given the final locally convex topology over these presentation spaces.

In the algebraic setting, this is the stagewise continuity formulation in [Gubler] (§6.1–§6.2); compare [CLD12] (§4.1.1). The description includes support control, fixed presentations, and all derivatives. Uniform convergence of functions alone is a different topology. No claim that this space is a strict countable LF limit is needed here.

The support convention is also substantive. In [CLD12] (§4.2.1), “support propre” means compact support avoiding ∂X\partial X. If XX has no analytic boundary, this reduces to ordinary compact support. On a space with boundary, a measure supported on ∂X\partial X vanishes on all these test functions and is not detected by the corresponding currents.

Definition 9.1.2. (Bidimension and bidegree). A current of bidimension (p,q)(p,q) is a continuous linear functional on Asm,cp,q(X)A^{p,q}_{\mathrm{sm},c}(X). Put Dp,qsm(X)=(Asm,cp,q(X))′,Dsmp,q(X)=Dn−p,n−qsm(X).D^{\mathrm{sm}}_{p,q}(X) =\bigl(A^{p,q}_{\mathrm{sm},c}(X)\bigr)', \qquad D^{p,q}_{\mathrm{sm}}(X)=D^{\mathrm{sm}}_{n-p,n-q}(X). The latter notation records bidegree. Thus a current of bidegree (p,q)(p,q) acts on test forms of bidegree (n−p,n−q)(n-p,n-q).

For example, [ω](α)=∫Xω∧α,ω∈Asmp,q(X),[\omega](\alpha)=\int_X\omega\wedge\alpha, \qquad \omega\in A^{p,q}_{\mathrm{sm}}(X), has bidegree (p,q)(p,q). Integration over a dd-dimensional algebraic cycle has bidimension (d,d)(d,d) and bidegree (n−d,n−d)(n-d,n-d). A Radon measure acts on functions, so its bidimension is (0,0)(0,0) and its bidegree is (n,n)(n,n). These examples prevent an ambiguous use of the phrase “a current of degree zero.” See [CLD12] (§4.2.2) and [Gubler] (Examples 6.3, 6.5, and 6.7).

Definition 9.1.3. (Differentiation and symmetry of currents). For TT of bidegree (p,q)(p,q), define ⟨d′T,α⟩=(−1)p+q+1⟨T,d′α⟩,⟨d′′T,β⟩=(−1)p+q+1⟨T,d′′β⟩,\begin{align*} \langle d'T,\alpha\rangle &=(-1)^{p+q+1}\langle T,d'\alpha\rangle,\\ \langle d''T,\beta\rangle &=(-1)^{p+q+1}\langle T,d''\beta\rangle, \end{align*} where α\alpha and β\beta have the complementary test degrees. Set ⟨JT,γ⟩=(−1)n⟨T,Jγ⟩.\langle JT,\gamma\rangle=(-1)^n\langle T,J\gamma\rangle. A current of bidegree (p,p)(p,p) is symmetric if JT=(−1)pTJT=(-1)^pT.

Proposition 9.1.4. (Compatibility with smooth forms). For every original smooth form ω\omega, d′[ω]=[d′ω],d′′[ω]=[d′′ω],J[ω]=[Jω].d'[\omega]=[d'\omega],\qquad d''[\omega]=[d''\omega],\qquad J[\omega]=[J\omega]. Moreover, for any current and a test form of the appropriate degree, ⟨d′d′′T,α⟩=⟨T,d′d′′α⟩.\begin{equation} \langle d'd''T,\alpha\rangle=\langle T,d'd''\alpha\rangle. \end{equation}

Proof. Stokes’ formula applies because test supports avoid the analytic boundary. If ω\omega has total degree kk, integrating d′(ω∧α)=d′ω∧α+(−1)kω∧d′αd'(\omega\wedge\alpha)=d'\omega\wedge\alpha+ (-1)^k\omega\wedge d'\alpha gives the first identity. The second is identical. On top-degree forms JJ acts by (−1)n(-1)^n, which gives the third identity with the normalization in Definition 9.1.3. The two transpose signs in d′d′′d'd'' multiply to −1-1; reversing d′′d′d''d' to d′d′′d'd'' supplies the second minus sign and proves (9.1.1). ◻

The operations supplied by the definitions have precise domains:

  1. A morphism pulls back smooth forms.

  2. A proper morphism pushes currents forward by ⟨f∗T,α⟩=⟨T,f∗α⟩.\langle f_*T,\alpha\rangle=\langle T,f^*\alpha\rangle. Properness ensures that the pulled-back test support is admissible.

  3. A current can be multiplied by a fixed smooth form, using ⟨T∧ω,α⟩=⟨T,ω∧α⟩.\langle T\wedge\omega,\alpha\rangle =\langle T,\omega\wedge\alpha\rangle.

These formulas do not define a pullback of an arbitrary current or a product of two arbitrary currents. They are the operations in [CLD12] (§4.2–§4.4) and [Gubler] (§6.4–§6.6).

Definition 9.1.5. (Positive currents). A symmetric current of bidegree (p,p)(p,p) is weakly positive, positive, or strongly positive if it is nonnegative, respectively, on strongly positive, positive, or weakly positive test forms of bidegree (n−p,n−p)(n-p,n-p).

Symmetry is explicit in this definition. Testing only against symmetric forms must not be used as a substitute for specifying symmetry of the current itself. In bidegree (1,1)(1,1) the three positivity conditions agree, because the complementary test degree is (n−1,n−1)(n-1,n-1). In middle degree, Example 7.4.3 prevents importing arguments that require strong positive tests to span all symmetric tests.

Positivity, order zero, and multiplication by a function

The test topology controls derivatives of all orders. Ordinary positivity supplies estimates involving coefficient values alone. We first establish those estimates, and then distinguish them from the stronger scalar-multiplier construction needed for arbitrary continuous functions.

Definition 9.1.6. (Order zero on a presentation). A current TT has order zero if, on each fixed finite tropical presentation and fixed compact support BB, its value satisfies an estimate ∣⟨T,α⟩∣≤CB∥α∥B,0,\begin{equation} |\langle T,\alpha\rangle|\leq C_B\|\alpha\|_{B,0}, \end{equation} where the seminorm uses the suprema of coefficient values, with quotient seminorms for representatives, and does not use their derivatives. On a real coordinate domain, this is equivalent to the distributional coefficients of TT being signed Radon measures.

Proposition 9.1.7. (The usable positive-current estimate). On a real coordinate domain, an ordinary positive current is of order zero. The same argument on finite tropical presentations gives order zero for ordinary positive currents on the original analytic test sheaf. It also applies to strongly positive currents. Weak positivity alone does not suffice in middle degree.

Open proof obligation (RC2). Supply the analytic positive-form decomposition and quotient estimate used after passing from the coordinate calculation to the original test sheaf; GJR Lemma 7.12. The real coordinate calculation is retained in historical evidence.

NAG-900067 · Proof status

Definition 9.1.8. (Continuous multiplication of an ordinary positive current). Let TT be an ordinary positive current and β\beta an ordinary positive smooth test of complementary degree. The positive functional ϕ↦⟨T,ϕβ⟩\phi\mapsto\langle T,\phi\beta\rangle on smooth scalar functions satisfies ∣⟨T,ϕβ⟩∣≤∥ϕ∥supp⁡β⟨T,β⟩.|\langle T,\phi\beta\rangle| \leq\|\phi\|_{\operatorname{supp}\beta}\langle T,\beta\rangle. Uniform density of original smooth scalar functions on compact supports extends this functional to a finite Radon measure μT,β\mu_{T,\beta} supported in supp⁡β\operatorname{supp}\beta. If a symmetric test is written α=β1−β2\alpha=\beta_1-\beta_2 with ordinary positive smooth tests, define (uT)(α)=∫u dμT,β1−∫u dμT,β2(uT)(\alpha)=\int u\,d\mu_{T,\beta_1}-\int u\,d\mu_{T,\beta_2} for a continuous function uu, and set its value on antisymmetric tests to zero.

The construction is independent of the decomposition: a linear relation among positive smooth tests gives the same relation among the measures, first on smooth scalar functions and then by uniform density. Finite-presentation domination proves continuity in the current topology. The density statement is [CLD12] (Corollaries 5.4.7–5.4.8); the ordinary-positive decomposition is [CLD25] (Proposition 6.5.2), also used in [GJR] (Lemma 7.12). The scalar-multiplier argument is [CLD25] (Lemma 8.5.1). Thus the definition applies even when uu is not a function of the coordinates of one fixed tropical chart. A chartwise order-zero estimate alone would not justify that extension for an arbitrary current; [CLD25] (Example 7.2.14) gives an order-zero scalar current on an infinite tropical star which is not a Radon measure.

If uj→uu_j\to u uniformly near a test support, then ujT→uTu_jT\to uT there. For u≥0u\geq0, the product is positive. It need not be closed, even when TT is closed; this distinction is essential in Section 13.1.

9.2 Tests, degrees and symmetry

Fix an open Ω⊂Rn\Omega\subset\mathbb R^n. For a compact K⋐ΩK\Subset\Omega, let AKp,qA_K^{p,q} consist of smooth forms supported in KK, with seminorms ∥η∥K,r=max⁡I,J, ∣ν∣≤rsup⁡x∈K∣∂νηIJ(x)∣(r=0,1,…).\left\lVert\eta\right\rVert_{K,r}=\max_{I,J,\ |\nu|\leq r} \sup_{x\in K}|\partial^\nu\eta_{IJ}(x)|\quad(r=0,1,\ldots). The space Acp,qA_c^{p,q} has the usual test-function inductive-limit topology. A linear functional is a current if its restriction to every AKp,qA_K^{p,q} is continuous. Equivalently, for each KK there are CKC_K and an integer rKr_K giving a bound by CK∥η∥K,rKC_K\left\lVert\eta\right\rVert_{K,r_K}.

Definition. A current of bidimension (q,q)(q,q) acts on Acq,qA_c^{q,q}; its bidegree is (n−q,n−q)(n-q,n-q). It has order zero if rKr_K can be chosen to be zero for every KK. It is symmetric if T(Jη)=(−1)qT(η).T(J\eta)=(-1)^qT(\eta).

Put ηs=(η+(−1)qJη)/2\eta_s=(\eta+(-1)^qJ\eta)/2. A symmetric current satisfies T(η)=T(ηs)T(\eta)=T(\eta_s) and vanishes on the complementary antisymmetric part. In matrix notation symmetrization replaces HH by (H+Ht)/2(H+H^t)/2.

Definition 9.2.2. A symmetric current is weakly positive if it is nonnegative on strongly positive tests; positive if it is nonnegative on positive tests; strongly positive if it is nonnegative on weakly positive tests.

Thus the test cones are reversed: strong positivity of a current is the most restrictive condition. We use positive without an adjective for the middle condition. Weak convergence Tj⇀TT_j\rightharpoonup T means convergence on every fixed smooth compactly supported test. It initially says nothing about continuous or varying tests.

Example 9.2.3. A locally integrable form defines [α](η)=∫α∧η[\alpha](\eta)=\int\alpha\wedge\eta. In particular [1][1] has bidegree (0,0)(0,0) and tests top-degree forms, while a scalar Radon measure has bidegree (n,n)(n,n) and tests functions. By wedge duality a smooth positive form gives a positive current.

9.3 Order zero and measure coefficients

The scalar argument below needs no continuity assumption. Positivity itself supplies the required bound.

Lemma 9.3.1. A linear functional L:Cc∞(Ω)→RL:C_c^\infty(\Omega)\to\mathbb R nonnegative on nonnegative functions extends uniquely to a positive Radon measure.

Proof. For tests supported in K⋐ΩK\Subset\Omega, choose 0≤χ∈Cc∞0\leq\chi\in C_c^\infty equal to one near KK. Then −∥f∥∞χ≤f≤∥f∥∞χ,∣L(f)∣≤L(χ)∥f∥∞.-\left\lVert f\right\rVert_\infty\chi\leq f\leq\left\lVert f\right\rVert_\infty\chi,\qquad |L(f)|\leq L(\chi)\left\lVert f\right\rVert_\infty. Every continuous compactly supported function has uniformly converging smooth approximants supported in one larger compact subset: extend by zero and convolve with a smooth compactly supported approximate identity. The bound defines an extension independent of the approximants; compare any two choices on a common compact set. Nonnegative functions admit nonnegative approximants, so the extension is positive. The Riesz representation theorem on Cc(Ω)C_c(\Omega) gives the unique locally finite Radon measure. ◻

Theorem 9.3.2. (Positive functionals have order zero). Let T:Acq,q(Ω)→RT:A_c^{q,q}(\Omega)\to\mathbb R be symmetric and linear, nonnegative on ordinary positive tests. Then TT is a current of order zero. There is a positive Radon measure μT\mu_T characterized by μT(ϕ)=T(ϕβq),ϕ∈Cc∞(Ω).\begin{equation} \mu_T(\phi)=T(\phi\beta_q),\qquad \phi\in C_c^\infty(\Omega). \end{equation} For a symmetric test η=αH\eta=\alpha_H supported in KK and χ\chi as in Proposition 7.3.2, ∣T(η)∣≤∥H∥∞,op μT(χ).\begin{equation} |T(\eta)|\leq\left\lVert H\right\rVert_{\infty,\mathrm{op}}\,\mu_T(\chi). \end{equation}

Proof. Since βq\beta_q is positive, ϕ↦T(ϕβq)\phi\mapsto T(\phi\beta_q) is a positive scalar functional. Lemma 9.3.1 gives μT\mu_T. Apply TT to the two positive forms in (7.3.1); this proves the estimate. Replace an arbitrary test by its symmetric part and use the coefficient norm comparison in Remark 7.3.3. A C0C^0 bound holds on every compact set, proving both current continuity and order zero. ◻

The proof uses ordinary-positive domination, not strong-cone spanning. The trace measure depends on coordinates: it is a control measure, not an additional intrinsic volume form.

Theorem 9.3.3. (Matrix of coefficient measures). Put m=(nq)m=\binom nq. There are unique signed Radon measures Mij=MjiM_{ij}=M_{ji} with T(αH)=∑i,j=1m∫Hij dMij,μT=∑iMii.\begin{equation} T(\alpha_H)=\sum_{i,j=1}^m\int H_{ij}\,dM_{ij},\qquad \mu_T=\sum_iM_{ii}. \end{equation} There is a measurable symmetric matrix AA such that dMij=Aij dμT,A≥0,tr⁡A=1μT-almost everywhere.\begin{equation} dM_{ij}=A_{ij}\,d\mu_T,\qquad A\geq0,\quad\operatorname{tr}A=1 \quad\text{\(\mu_T\)-almost everywhere}. \end{equation} Consequently ∣T(αH)∣≤∫∥H(x)∥op dμT(x).\begin{equation} |T(\alpha_H)|\leq\int\left\lVert H(x)\right\rVert_{\mathrm{op}}\,d\mu_T(x). \end{equation}

Proof. Use the diagonal matrix units EiiE_{ii} and put Fij=(Eij+Eji)/2F_{ij}=(E_{ij}+E_{ji})/2 for i≠ji\ne j. Define Mii(ϕ)=T(αϕEii),Mij(ϕ)=T(αϕFij).M_{ii}(\phi)=T(\alpha_{\phi E_{ii}}),\qquad M_{ij}(\phi)=T(\alpha_{\phi F_{ij}}). The diagonal functionals are positive measures. The others are signed measures, since Fij=14((ei+ej)(ei+ej)t−(ei−ej)(ei−ej)t),F_{ij}=\tfrac14\big((e_i+e_j)(e_i+e_j)^t -(e_i-e_j)(e_i-e_j)^t\big), and each square gives a positive measure by the scalar lemma. Expanding HH proves (9.3.3). The factor 1/21/2 is necessary because the sum includes both ijij and jiji.

For any constant vector vv, the measure ∑vivjMij\sum v_iv_jM_{ij} is positive. Applying this to ei+eje_i+e_j and ei−eje_i-e_j gives −(Mii+Mjj)≤2Mij≤Mii+Mjj.-(M_{ii}+M_{jj})\leq2M_{ij}\leq M_{ii}+M_{jj}. Thus ∣Mij∣≤(Mii+Mjj)/2|M_{ij}|\leq(M_{ii}+M_{jj})/2 and every coefficient measure is absolutely continuous with respect to μT\mu_T. The Radon–Nikodym theorem gives AijA_{ij}. For each rational vector vv, vtAv≥0v^tAv\geq 0 outside a null set. Remove the union of these countably many null sets; continuity in vv then yields A≥0A\geq 0. The trace identity follows from μT=∑iMii\mu_T=\sum_iM_{ii}. Finally −∥H∥opId≤H≤∥H∥opId⟹∣tr⁡(AH)∣≤∥H∥op.-\left\lVert H\right\rVert_{\mathrm{op}}\mathrm{Id}\leq H\leq\left\lVert H\right\rVert_{\mathrm{op}}\mathrm{Id} \quad\Longrightarrow\quad |\operatorname{tr}(AH)|\leq\left\lVert H\right\rVert_{\mathrm{op}}. Integrating proves the estimate. Testing the matrix units proves uniqueness. ◻

Corollary 9.3.4. If both TT and −T-T are positive, then T=0T=0. More generally μT=0\mu_T=0 implies T=0T=0.

Proof. In the first case μT\mu_T and its negative are positive, so μT=0\mu_T=0. In either case (9.3.5) makes every test pairing zero. ◻

Corollary 9.3.5. (Coefficient Cauchy–Schwarz inequality). For every relatively compact Borel set E⊂ΩE\subset\Omega, ∣Mij∣(E)≤Mii(E)Mjj(E).|M_{ij}|(E)\leq\sqrt{M_{ii}(E)M_{jj}(E)}. For any a,b≥0a,b\geq 0 there is also the measure inequality ab∣Mij∣≤12(a2Mii+b2Mjj).ab|M_{ij}|\leq\tfrac12(a^2M_{ii}+b^2M_{jj}).

Proof. Positive semidefiniteness of AA gives ∣Aij∣2≤AiiAjj|A_{ij}|^2\leq A_{ii}A_{jj}, by taking the determinant of its two-by-two principal minor. Integrate ∣Aij∣≤AiiAjj|A_{ij}|\leq\sqrt{A_{ii}A_{jj}} over EE and apply scalar Cauchy–Schwarz with respect to μT\mu_T. The second assertion follows by integrating ab∣Aij∣≤(a2Aii+b2Ajj)/2ab|A_{ij}|\leq(a^2A_{ii}+b^2A_{jj})/2 on Borel sets. ◻

The weighted inequality is the real-chart form of [BGJK] (Proposition 5.2.6). Their setting also includes boundary strata, which are not treated here. Conversely, any locally finite positive measure μ\mu and measurable positive semidefinite AA with tr⁡A=1\operatorname{tr}A=1 define a positive current by T(αH)=∫tr⁡(AH) dμT(\alpha_H)=\int\operatorname{tr}(AH)\,d\mu. The estimate above proves continuity, and the trace measure is μ\mu. Thus the matrix description characterizes positive currents in the present Euclidean setting.

9.4 Why weak positivity is insufficient

Weak positivity can allow derivatives of point masses. Thus it changes more than the constant in an estimate.

Theorem 9.4.1. There is a nonzero symmetric weakly positive current on R4\mathbb R^4 of bidimension (2,2)(2,2) which is not of order zero. Its negative is also weakly positive.

Proof. With BB from Section 7.4, define on any coefficient matrix HH TB(αH)=−∂x1tr⁡(BH)(0).\begin{equation} T_B(\alpha_H)=-\partial_{x_1}\operatorname{tr}(BH)(0). \end{equation} This is a continuous distribution of order at most one. Since BB is symmetric, it kills antisymmetric matrices. If αH\alpha_H is strongly positive pointwise, then tr⁡(BH(x))=0\operatorname{tr}(BH(x))=0 everywhere by Theorem 7.4.2. Hence both TBT_B and −TB-T_B vanish on all strong tests and are weakly positive.

Choose χ∈Cc∞(R4)\chi\in C_c^\infty(\mathbb R^4) equal one near zero, and g∈Cc∞(R)g\in C_c^\infty(\mathbb R) with g′(0)=1g'(0)=1. Set ϕε(x)=εg(x1/ε)χ(x),Hε=ϕεB.\phi_\varepsilon(x)=\varepsilon g(x_1/\varepsilon)\chi(x), \qquad H_\varepsilon=\phi_\varepsilon B. All supports lie in supp⁡χ\operatorname{supp}\chi and ∥Hε∥∞→0\left\lVert H_\varepsilon\right\rVert_\infty\to 0. But ∂1ϕε(0)=1\partial_1\phi_\varepsilon(0)=1 and tr⁡(B2)=6\operatorname{tr}(B^2)=6, whence TB(αHε)=−6.T_B(\alpha_{H_\varepsilon})=-6. This contradicts any order-zero estimate on that fixed compact set. ◻

Also TB(ϕβ2)=0T_B(\phi\beta_2)=0 since tr⁡B=0\operatorname{tr}B=0. Its formal trace vanishes although the current is nonzero. The trace estimate for ordinary positive currents therefore cannot be applied to weakly positive currents. This is a distributional consequence of the nonpointed cone in [BGJK] (Examples 2.3.6 and 5.2.8).

Remark 9.4.2 (The CLD25 order-zero assertion). [CLD25] (Proposition 7.2.9) deduces order zero from positivity on strongly positive tests, using strong domination in its proof. On the standard real tropical space R4\mathbb R^4, (9.4.1) satisfies its symmetry and positivity assumptions and violates its conclusion. In bidegree (2,2)(2,2) the symmetric matrix condition agrees in the two conventions. Equip R4\mathbb R^4 with its usual affine functions and constant volume calibration in dimension four. Its standard tropical structure has level four; locally one may use the interior of a constantly calibrated four-dimensional cube, as in [CLD25] (§6.6.3). Coordinate functions separate points, hence it is affinely and differentially separated in the sense of [CLD25] (§6.2.3). The counterexample is in this real tropical setting, not claimed on an analytic Berkovich space. Theorem 9.3.2 provides the ordinary-positive replacement on real charts.

9.5 Multiplication by continuous functions

Proposition 9.5.1. A positive current has a unique extension to compactly supported continuous coefficient forms, continuous for uniform coefficient convergence on a common compact set. On symmetric tests it is (9.3.3); it vanishes on antisymmetric tests.

Proof. The coefficient measures are finite on compact sets, so the integrals exist and converge under uniform convergence with fixed compact support. Uniqueness follows by uniformly approximating the finitely many coefficients by smooth functions supported in one larger compact set. Changing coordinates gives the same extension: the extensions agree on smooth forms and the fixed smooth transition matrix preserves uniform convergence on compact sets. ◻

Definition. For f∈C0(Ω)f\in C^0(\Omega) define (fT)(η)=T(fη)(fT)(\eta)=T(f\eta) using this extension.

Proposition 9.5.3. The current fTfT has order zero, and ∣(fT)(αH)∣≤∫∣f∣∥H∥op dμT.|(fT)(\alpha_H)|\leq\int |f|\left\lVert H\right\rVert_{\mathrm{op}}\,d\mu_T. If f≥0f\geq 0, it is positive and μfT=fμT\mu_{fT}=f\mu_T. Multiplication is associative, commutes with restriction, and satisfies supp⁡(fT)⊂supp⁡T∩supp⁡f.\operatorname{supp}(fT)\subset\operatorname{supp}T\cap\operatorname{supp}f.

Proof. The inequality is (9.3.5) applied to fHfH. If f,H≥0f,H\geq 0, then tr⁡(AfH)≥0\operatorname{tr}(AfH)\geq 0 almost everywhere. Testing ϕβq\phi\beta_q gives the trace identity. Associativity and restriction follow by multiplying each coefficient measure. On open sets where T=0T=0 its coefficient measures vanish, and on open sets where f=0f=0 the multiplied coefficients vanish. This proves the support statement. ◻

Remark. Locally bounded Borel functions also multiply the coefficient measures; the relevant equivalence is then equality μT\mu_T-almost everywhere. Continuous multipliers are better suited to limits of varying currents because they admit uniform approximation. For an arbitrary distribution, coefficient distributions need not be measures, so this argument does not define continuous multiplication.

9.6 Weak convergence and varying multipliers

Lemma 9.6.1. If positive currents of fixed bidimension satisfy Tj⇀TT_j\rightharpoonup T, then TT is positive and for every compact K⋐ΩK\Subset\Omega, sup⁡jμTj(K)<∞.\sup_j\mu_{T_j}(K)<\infty.

Proof. Symmetry and nonnegativity on each positive test pass to the limit. Choose 0≤χ∈Cc∞0\leq\chi\in C_c^\infty equal one near KK. Then 0≤μTj(K)≤Tj(χβq)⟶T(χβq)<∞.0\leq\mu_{T_j}(K)\leq T_j(\chi\beta_q)\longrightarrow T(\chi\beta_q)<\infty. A convergent real sequence, including its initial terms, is bounded. ◻

Proposition 9.6.2. Under these hypotheses, Tj(η)→T(η)T_j(\eta)\to T(\eta) for every fixed continuous compactly supported coefficient form η\eta.

Proof. Approximate η\eta uniformly by smooth ην\eta_\nu supported in one compact K′K'. Theorem 9.3.3 and Lemma 9.6.1 bound ∣Tj(η−ην)∣|T_j(\eta-\eta_\nu)| uniformly in jj by a constant times the coefficient error. The same estimate holds for TT. For fixed ν\nu, weak convergence applies to ην\eta_\nu. First choose ν\nu to control both approximation errors, then let jj tend to infinity. ◻

Theorem 9.6.3. (Joint continuity). If Tj⇀TT_j\rightharpoonup T are positive currents of fixed bidimension and fj→ff_j\to f locally uniformly with fj,ff_j,f continuous, then fjTj⇀fT.f_jT_j\rightharpoonup fT.

Proof. For a symmetric smooth test η=αH\eta=\alpha_H supported in KK, write Tj(fjη)−T(fη)=Tj((fj−f)η)+(Tj−T)(fη).T_j(f_j\eta)-T(f\eta) =T_j((f_j-f)\eta)+(T_j-T)(f\eta). The first term is bounded in absolute value by ∥fj−f∥K,∞∥H∥∞,op μTj(K),\left\lVert f_j-f\right\rVert_{K,\infty}\left\lVert H\right\rVert_{\infty,\mathrm{op}}\,\mu_{T_j}(K), which tends to zero by the uniform mass bound. The second tends to zero by Proposition 9.6.2, since fηf\eta is a fixed continuous test. Antisymmetric components vanish. ◻

Pointwise convergence of multipliers is insufficient. On R\mathbb R, take Tj=δ1/jT_j=\delta_{1/j} and fj(x)=g(jx)f_j(x)=g(jx) for g∈Cc∞((1/2,3/2))g\in C_c^\infty((1/2,3/2)) with g(1)=1g(1)=1. At each fixed xx, fj(x)→0f_j(x)\to 0, and Tj⇀δ0T_j\rightharpoonup\delta_0, but fjTj=δ1/j⇀δ0f_jT_j=\delta_{1/j}\rightharpoonup\delta_0, not zero.

9.7 Differentiation and closedness

For a current SS of total degree kk, use the distributional convention (d′S)(η)=(−1)k+1S(d′η),(d′′S)(η)=(−1)k+1S(d′′η).(d'S)(\eta)=(-1)^{k+1}S(d'\eta),\qquad (d''S)(\eta)=(-1)^{k+1}S(d''\eta). Closed means both derivatives vanish. These signs agree with derivatives of integration currents. Indeed coefficients of a compactly supported top-degree derivative are sums of partial derivatives; Fubini and the one-variable fundamental theorem of calculus make each integral zero.

Lemma 9.7.1. For a current of diagonal bidegree, (d′d′′T)(η)=T(d′d′′η).\begin{equation} (\mathrm d'\mathrm d''T)(\eta)=T(\mathrm d'\mathrm d''\eta). \end{equation} Differentiation is continuous for weak convergence.

Proof. The total degree kk is even. Applying the definitions twice gives (−1)k+2(−1)k+1T(d′′d′η)=T(d′d′′η).(-1)^{k+2}(-1)^{k+1}T(d''d'\eta)=T(d'd''\eta). For continuity, the derivative of a fixed smooth compactly supported test is another fixed test. ◻

For a smooth form γ\gamma and a current SS of total degree kk, left multiplication is defined by (γ∧S)(η)=(−1)kdeg⁡γS(γ∧η).(\gamma\wedge S)(\eta)=(-1)^{k\deg\gamma}S(\gamma\wedge\eta). This convention agrees with left multiplication of locally integrable forms and fixes the signs in the next statement.

Proposition 9.7.2. If TT is closed and ff smooth, then d′(fT)=d′f∧T,d′′(fT)=d′′f∧T,d′d′′(fT)=d′d′′f∧T.d'(fT)=d'f\wedge T,\qquad d''(fT)=d''f\wedge T,\qquad \mathrm d'\mathrm d''(fT)=\mathrm d'\mathrm d''f\wedge T.

Proof. Insert d′(fη)=d′f∧η+f d′ηd'(f\eta)=d'f\wedge\eta+f\,d'\eta into d′T=0d'T=0 to obtain the first formula; exchange primes for the second. In general, the graded Leibniz rules give d′d′′(fT)=d′d′′f∧T−d′′f∧d′T+d′f∧d′′T+f d′d′′T.\begin{align*} \mathrm d'\mathrm d''(fT)={}&\mathrm d'\mathrm d''f\wedge T-d''f\wedge d'T\\ &+d'f\wedge d''T+f\,\mathrm d'\mathrm d''T. \end{align*} Closedness removes the last three terms. This argument needs smoothness of ff: derivatives of a continuous multiplier cannot be inserted as ordinary smooth forms. ◻

Example 9.7.3. (Multiplication can destroy closedness). On R\mathbb R, T=[1]T=[1] is positive and closed. For f(x)=ex>0f(x)=e^x>0, the current fTfT is positive, but d′[ex](ϕ d′′x)=−∫exϕ′(x) dx=∫exϕ(x) dx.d'[e^x](\phi\,d''x) =-\int e^x\phi'(x)\,dx=\int e^x\phi(x)\,dx. This is positive for nonzero nonnegative ϕ∈Cc∞\phi\in C_c^\infty, so fTfT is not closed.

Remark 9.7.4 (CLD25 Corollary 8.6.7). Definition 8.6.2 of [CLD25] requires a Bedford–Taylor current to be positive and closed and to satisfy a curvature-kernel condition. Corollary 8.6.7 asserts preservation under multiplication by an arbitrary positive continuous function. Example 9.7.3 contradicts closedness even for a smooth multiplier. The starting current [1][1] has the kernel property since d′d′′(u[1])=d′d′′[u]\mathrm d'\mathrm d''(u[1])=\mathrm d'\mathrm d''[u] by definition. The valid replacement here is Proposition 9.5.3: positivity and order zero are preserved. Closedness needs extra assumptions. This does not invalidate the distinct operation d′d′′(uT)\mathrm d'\mathrm d''(uT), whose closedness follows from the differential identities.

9.8 A first curvature product

For positive TT and continuous uu, d′d′′(uT)\mathrm d'\mathrm d''(uT) is always a distributional current. Positivity requires more. We prove it under a stated approximation hypothesis.

Theorem 9.8.1. Let TT be positive and closed of bidimension (q,q)(q,q), q≥1q\geq 1. Suppose uu is continuous and every point has a neighbourhood UU on which smooth uju_j satisfy d′d′′uj≥0\mathrm d'\mathrm d''u_j\geq 0 and converge to uu locally uniformly on the same UU. Then R=d′d′′(uT)R=\mathrm d'\mathrm d''(uT) is positive and closed of bidimension (q−1,q−1)(q-1,q-1), and locally d′d′′uj∧T⇀R.\mathrm d'\mathrm d''u_j\wedge T\rightharpoonup R. The limit is independent of the approximating sequence.

Proof. Theorem 9.6.3 with fixed TT gives ujT⇀uTu_jT\rightharpoonup uT. Lemma 9.7.1 and Proposition 9.7.2 give the displayed convergence. For a positive test η\eta of bidegree (q−1,q−1)(q-1,q-1), Proposition 7.5.2 makes d′d′′uj∧η\mathrm d'\mathrm d''u_j\wedge\eta positive. Thus (d′d′′uj∧T)(η)≥0(\mathrm d'\mathrm d''u_j\wedge T)(\eta)\geq 0, and positivity and symmetry pass to the limit. A nonnegative smooth partition of unity on a test’s compact support transfers local positivity to global positivity. The identities (d′)2=(d′′)2=0(d')^2=(d'')^2=0 and d′d′′=−d′′d′d'd''=-d''d' give d′R=d′′R=0d'R=d''R=0. The limit equals the already defined d′d′′(uT)\mathrm d'\mathrm d''(uT), so it is independent of approximants. ◻

Proposition 9.8.2. A continuous convex function on a convex open subset Ω⊂Rn\Omega\subset\mathbb R^n satisfies this approximation hypothesis locally.

Proof. Choose a ball UU with compact closure in Ω\Omega and a nonnegative smooth kernel ρ\rho supported in the unit ball, of integral one. For sufficiently small ε\varepsilon put uε(x)=∫u(x−εy)ρ(y) dy,x∈U.u_\varepsilon(x)=\int u(x-\varepsilon y)\rho(y)\,dy,\qquad x\in U. All functions are defined on the same UU. Convolution gives smoothness. Apply convexity inside the integral to see that uεu_\varepsilon is convex. Uniform continuity of uu on a slightly larger compact neighbourhood gives uniform convergence on U‾\overline U. Convexity implies a positive semidefinite Hessian by the criterion in Chapter 7. ◻

Example. (The corner mass of an absolute value). On R\mathbb R, one has d′d′′[∣x∣]=2δ0\mathrm d'\mathrm d''[|x|]=2\delta_0. For ϕ∈Cc∞(R)\phi\in C_c^\infty(\mathbb R), integration by parts on the two half-lines gives ∫−∞0(−x)ϕ′′(x) dx+∫0∞xϕ′′(x) dx=2ϕ(0).\int_{-\infty}^0(-x)\phi''(x)\,dx+ \int_0^\infty x\phi''(x)\,dx=2\phi(0). There is no smooth Hessian density off zero, but the full curvature is nonzero. The smooth convex uε=x2+ε2u_\varepsilon=\sqrt{x^2+\varepsilon^2} converge locally uniformly to ∣x∣|x|, and their second derivatives converge weakly to this mass by Theorem 9.8.1.

Remark 9.8.4. Sequence independence for a fixed uu is not curvature-class independence. If d′d′′[u]=d′d′′[v]\mathrm d'\mathrm d''[u]=\mathrm d'\mathrm d''[v], identifying the products still requires d′d′′((u−v)T)=0.\mathrm d'\mathrm d''((u-v)T)=0. For smooth h=u−vh=u-v with d′d′′h=0\mathrm d'\mathrm d''h=0 as a form, Proposition 9.7.2 proves this. It does not apply to nonsmooth hh. No general kernel assertion follows from Theorem 9.8.1. Symmetry of iterated products is likewise a further theorem, not a consequence of being able to repeat a definition.

9.9 Transfer through charts

The estimates are finite-dimensional in their coefficient directions. They give a transfer statement with explicit hypotheses, but relations among chart presentations must remain visible.

Proposition 9.9.1. (Transfer of order zero). Let Ac\mathcal A_c have a symmetry projection and local linear presentation maps PDP_D from smooth matrix-valued coefficients on Euclidean chart domains. Assume:

  1. Tests are finite sums of chart-presented tests after localization.

  2. For each fixed chart and compact coefficient support there are an actual positive test θD\theta_D and cDc_D such that, for symmetric HH, cD∥H∥∞θD±PD(H)are actual positive tests.c_D\left\lVert H\right\rVert_\infty\theta_D\pm P_D(H) \quad\text{are actual positive tests.}

  3. The topology is the final locally convex topology of these presentation spaces, and presentations respect symmetrization.

Then a symmetric linear functional SS nonnegative on actual positive tests is a current, of order zero on every fixed presentation.

Proof. Applying SS to the two tests gives ∣S(PD(H))∣≤cDS(θD)∥H∥∞.|S(P_D(H))|\leq c_DS(\theta_D)\left\lVert H\right\rVert_\infty. Use symmetrization for the remaining coefficient part. Each composite with a presentation is continuous even in the coefficient C0C^0 topology. The final topology gives continuity on Ac\mathcal A_c, and the bound is the asserted order-zero estimate. ◻

The abstract transfer proposition alone does not define fSfS for arbitrary continuous ff on an analytic space. Such an ff need not come from a fixed tropical chart. Nor does separate extension in each presentation show that new continuous relations vanish. The next elementary criterion isolates this issue.

Lemma 9.9.2. (Extension across a quotient). Let EE be a normed space of smooth presentations, E^\widehat E its completion, N⊂EN\subset E the relation subspace and N^⊂E^\widehat N\subset\widehat E a proposed closed relation space. A bounded functional ℓ\ell vanishing on NN extends to E^/N^\widehat E/\widehat N if N^=N‾\widehat N=\overline N. In general it extends if and only if its continuous extension vanishes on all of N^\widehat N.

Proof. The functional extends uniquely to E^\widehat E and vanishes on N‾\overline N by continuity. A functional factors through a quotient precisely when it vanishes on the relation subspace. ◻

A chart construction must prove this relation statement, or use an intrinsic extension theorem that already contains it. Positivity provides continuity, not automatically descent through an enlarged relation space. In the algebraic CLD setting, Theorem 8.14.1 supplies a different route: measures attached directly to actual smooth tests.

9.10 Source concordance and subsequent use

The following list records the exact replacements proved in this instalment.

  1. [CLD12] (Lemma 5.2.3): Theorem 7.4.2 disproves strong spanning on real charts; Proposition 7.3.2 supplies ordinary-positive domination.

  2. [GK19] (Remark 1.12(i)): the weak cone is not pointed. Corollary 9.3.4 is an ordinary-positive statement, not a restoration of its weak analogue.

  3. [CLD25] (Proposition 7.2.9): Theorem 9.4.1 gives a real tropical counterexample; Theorems 9.3.2 and 9.3.3 prove the ordinary-positive result.

  4. [CLD25] (Corollary 8.6.7): Example 9.7.3 gives the closedness obstruction; Proposition 9.5.3 is the valid multiplier result here.

Chapter 8 constructs intrinsic scalar multiplication, and Chapter 13 treats mixed products in the stated algebraic CLD setting. The next comparison concerns generated potential classes and the curvature kernel, beyond the admitted approximation class. No general continuous current-psh regularization theorem is claimed.

9.11 Compactness from a uniform trace bound

Order zero makes continuous multiplication possible. A uniform order-zero bound gives a second useful conclusion: compactness for weak convergence. These statements use different quantifiers. Every positive current has a finite trace on a fixed compact set, but a family of positive currents need not have a common bound there.

Theorem 9.11.1. (Local compactness of positive currents). Let Ω⊂Rn\Omega\subset\mathbb R^n be open. For each compact K⋐ΩK\Subset\Omega, fix a finite number CKC_K. Let T\mathscr T be a family of ordinary positive currents of a fixed bidimension, and assume their trace measures in the fixed coordinates satisfy sup⁡T∈TμT(K)≤CK.\sup_{T\in\mathscr T}\mu_T(K)\leq C_K. Every sequence in T\mathscr T has a subsequence converging weakly on smooth compactly supported tests to a positive current. If all members are d′d'-closed, or d′′d''-closed, or both, so is the limit. The same conclusions hold for the closure of T\mathscr T.

Proof. Choose a compact exhaustion Kj⋐Kj+1∘K_j\Subset K_{j+1}^{\circ} of Ω\Omega. By Theorem 9.3.3, the coefficient measures MabM_{ab} of every member are bounded in total variation on KjK_j by a constant controlled by CKjC_{K_j}. Indeed, their densities satisfy ∣Aab∣≤1|A_{ab}|\leq 1 relative to μT\mu_T. On the space of continuous functions supported in KjK_j, choose a countable set dense in the uniform norm. Such a set exists because KjK_j is compact metrizable; one can equivalently work with C0(Kj∘)C_0(K_j^{\circ}) and its separability.

Diagonal extraction over these countable sets, all coefficient pairs and all jj gives a subsequence for which every selected coefficient pairing converges. The uniform total-variation bounds extend this convergence to every compactly supported continuous coefficient. For example, if fℓ→ff_\ell\to f uniformly with common compact support, then ∣∫(f−fℓ) dMab,T∣≤CK∥f−fℓ∥∞\left|\int(f-f_\ell)\,dM_{ab,T}\right| \leq C_K\|f-f_\ell\|_\infty uniformly in TT. Taking first the subsequence limit and then ℓ→∞\ell\to\infty gives a well-defined bounded functional on each coefficient space. The resulting current has order zero and the expected compatibility on overlapping compact sets.

For every ordinary positive test η\eta, the limit of T(η)≥0T(\eta)\geq 0 is nonnegative. Symmetry passes to the limit as well. This proves positivity. Closedness also passes to the limit: for every test ξ\xi, the value on d′ξd'\xi or d′′ξd''\xi is the limit of zero. No continuity of a differential on a space of measures is needed; the differential is applied to the fixed test. The estimates and the same argument remain valid for the weak closure, with the local mass bounds interpreted using compactly supported cutoffs. ◻

Remark 9.11.2. A weak limit need not satisfy the literal bound μ(K)≤CK\mu(K)\leq C_K for each originally chosen compact KK: mass can move into its boundary. What is inherited without ambiguity is the bound on each fixed nonnegative continuous cutoff. For compactness it is enough to retain bounds on slightly larger compact neighbourhoods. This is why the theorem asserts relative compactness and local boundedness, rather than preserving an arbitrary list of sharp compact-set constants.

What changes for an analytic test space

An analytic current is not defined using one ambient Euclidean coordinate space. Its tests are presented in finitely many moment charts, modulo analytic relations. The following formulation isolates the exact functional-analytic input and avoids a countability assumption on the field.

Proposition 9.11.3. Let EE be a real test space with a locally convex final topology from presentation spaces EaE_a. Suppose a family T⊂E′\mathscr T\subset E' satisfies, for every aa, a common estimate ∣T(ιaη)∣≤Capa(η),T∈T,|T(\iota_a\eta)|\leq C_a p_a(\eta),\qquad T\in\mathscr T, where pap_a is a continuous seminorm and the images of the EaE_a span EE. Then T\mathscr T is relatively compact for pointwise convergence on EE, in the sense of convergent subnets. Every limit is a continuous functional on EE. Positivity on a fixed test cone, vanishing on specified relations, and closedness under fixed continuous test differentials are preserved.

Proof. Each e∈Ee\in E is a finite sum of presented elements, so the estimates give a finite bound MeM_e for ∣T(e)∣|T(e)|, uniform in TT. Embed T\mathscr T in the compact product ∏e∈E[−Me,Me]\prod_{e\in E}[-M_e,M_e]. Every net has a subnet converging in this product. The identities expressing linearity are closed conditions on finitely many coordinates, so the limit is linear. On EaE_a it still satisfies the displayed estimate, hence is continuous. The defining final topology makes the resulting linear functional continuous on EE. Each positivity inequality or differential identity is another closed condition on evaluations. If tests are first defined in a presentation space and then quotiented, the equations annihilating the relation subspace are preserved in exactly the same way. ◻

A sequential statement requires a countable determining family with uniform approximation control, as in the Euclidean proof. Without such a family, the product compactness theorem gives subnets, not automatically subsequences. Conversely, convergence on a countable family alone need not imply convergence on all analytic tests unless the density and uniform bounds have been established.

9.12 Testing moving coefficients against a convergent current

Proposition 9.12.1. Suppose Ti→TT_i\to T on all tests and the currents satisfy a common order-zero bound on each fixed compact presentation: ∣Ti(vη)∣≤Cη∥v∥∞|T_i(v\eta)|\leq C_\eta\|v\|_\infty for smooth scalar multipliers vv, with CηC_\eta independent of ii. Let ui,uu_i,u be continuous functions on a fixed compact neighbourhood of the support of a test η\eta, with ui→uu_i\to u uniformly. Assume the chartwise order-zero extension and continuous approximation for uηu\eta used in Chapter 9. Then Ti(uiη)⟶T(uη).T_i(u_i\eta)\longrightarrow T(u\eta).

Proof. Choose smooth multipliers vk→uv_k\to u uniformly on that compact neighbourhood, with a common cutoff where needed. The difference splits as (Ti−T)(vkη)+Ti((ui−u)η)+(Ti−T)((u−vk)η).(T_i-T)(v_k\eta) +T_i((u_i-u)\eta) +(T_i-T)((u-v_k)\eta). The first term tends to zero for fixed kk. The second is bounded by a fixed constant times ∥ui−u∥∞\|u_i-u\|_\infty. The last is bounded by a fixed constant times ∥u−vk∥∞\|u-v_k\|_\infty, using the uniform order-zero estimates for TiT_i and TT. First let ii tend to its limit and then k→∞k\to\infty. ◻

Example 9.12.2. (Weak convergence is weaker than variation convergence). On R\mathbb R, δ1/j→δ0\delta_{1/j}\to\delta_0 on every continuous compactly supported function, but ∥δ1/j−δ0∥TV=2\|\delta_{1/j}-\delta_0\|_{\rm TV}=2. Moreover δj→0\delta_j\to 0 vaguely on R\mathbb R while every δj\delta_j has total mass one. For a direct moving-test counterexample, take continuous functions uju_j supported in small intervals around 1/j1/j that avoid zero, with 0≤uj≤10\leq u_j\leq 1, uj(1/j)=1u_j(1/j)=1, and interval lengths tending to zero fast enough that uj→0u_j\to 0 pointwise. Then δ1/j(uj)=1\delta_{1/j}(u_j)=1 while δ0(0)=0\delta_0(0)=0. Thus pointwise convergence of moving multipliers does not suffice. The second shows why compactness of the underlying space, or a separate tightness argument, is needed to retain total mass.

The current compactness argument supplies limits, not uniqueness of nonlinear products. In Bedford–Taylor theory, uniqueness comes from the recursion and its estimates; compactness alone could leave several cluster currents. Likewise, positivity and closedness of a limit do not identify its cohomology class unless a class-detecting pairing is continuous and has been retained. These are separate uses of the same weak topology.

9.13 Radial mass monotonicity for closed supercurrents

Closedness imposes a dimension constraint on positive currents. A positive measure can be concentrated at any point, but a positive closed current of positive bidimension cannot place coefficient mass on a set of too small Hausdorff dimension. The real proof uses the convex radial function ∣x∣|x|, not the complex logarithmic radial weight. We begin with a regularized formula which avoids multiplying a current by a singular form at the centre.

Fix Euclidean coordinates on Ω⊂Rn\Omega\subset\mathbb R^n and put β=∑id′xi∧d′′xi=12DD∣x∣2\beta=\sum_i d'x_i\wedge d''x_i=\tfrac 12DD|x|^2. For a weakly positive current TT of bidimension (p,p)(p,p), its scalar trace measure is μT=T∧βpp!.\mu_T=T\wedge\frac{\beta^p}{p!}. It is positive because βp\beta^p is strongly positive. For weak positivity alone, this scalar measure need not dominate all coefficients of TT. Ordinary positivity supplies that additional estimate by Theorem 9.3.3.

Lemma 9.13.1. (The regularized radial identity). Suppose TT is smooth, closed under both differentials, and has bidimension (p,p)(p,p) with p≥1p\geq 1. If B‾(a,r)⊂Ω\overline B(a,r)\subset\Omega and ρε(x)=(∣x−a∣2+ε2)1/2\rho_\varepsilon(x)=(|x-a|^2+\varepsilon^2)^{1/2}, then μT(B(a,r))(r2+ε2)p/2=1p!∫B(a,r)T∧(DDρε)p.\frac{\mu_T(B(a,r))}{(r^2+\varepsilon^2)^{p/2}} =\frac1{p!}\int_{B(a,r)}T\wedge(DD\rho_\varepsilon)^p.

Proof. One has β=ρεDDρε+d′ρε∧d′′ρε\beta=\rho_\varepsilon DD\rho_\varepsilon+d'\rho_\varepsilon\wedge d''\rho_\varepsilon. Stokes and closedness give ∫B(a,r)T∧βp=∫∂B(a,r)T∧d′′(ρε2/2)∧βp−1=∫∂B(a,r)ρεpT∧d′′ρε∧(DDρε)p−1=(r2+ε2)p/2∫B(a,r)T∧(DDρε)p.\begin{align*} \int_{B(a,r)}T\wedge\beta^p &=\int_{\partial B(a,r)}T\wedge d''(\rho_\varepsilon^2/2)\wedge\beta^{p-1}\\ &=\int_{\partial B(a,r)}\rho_\varepsilon^p T\wedge d''\rho_\varepsilon\wedge(DD\rho_\varepsilon)^{p-1}\\ &=(r^2+\varepsilon^2)^{p/2} \int_{B(a,r)}T\wedge(DD\rho_\varepsilon)^p. \end{align*} In the second line, every other term contains d′′ρεd''\rho_\varepsilon twice and vanishes. In the third line, ρε\rho_\varepsilon is constant on the boundary sphere and Stokes is used again. Dividing by p!p! gives the formula. All forms used here are smooth at aa because ε>0\varepsilon>0. ◻

Theorem 9.13.2. (Real mass monotonicity). If TT is weakly positive and d′d'- and d′′d''-closed of bidimension (p,p)(p,p), then r⟼μT(B(a,r))rpr\longmapsto\frac{\mu_T(B(a,r))}{r^p} is nondecreasing for balls compactly contained in Ω\Omega. For p=0p=0 this is simply monotonicity of the mass of increasing balls.

Proof. First suppose TT smooth and p≥1p\geq 1. The function ρε\rho_\varepsilon is convex, so (DDρε)p(DD\rho_\varepsilon)^p is strongly positive. The right side of Lemma 9.13.1 is the mass of a positive measure on the ball and is nondecreasing in rr. For r<sr<s, let ε↓0\varepsilon\downarrow 0 in the corresponding inequality. This gives the asserted inequality between the two radius ratios. No singular product with DD∣x−a∣DD|x-a| is needed.

For a general current, convolve its coefficient distributions on a slightly smaller domain with a nonnegative smooth approximate identity. The currents TδT_\delta are smooth, weakly positive and closed. Their trace measures converge weakly to μT\mu_T. For radii r,sr,s whose boundary spheres have zero μT\mu_T mass, the ball masses converge, so the inequality passes to the limit. There are at most countably many exceptional radii on each bounded interval, since the spheres are disjoint and the trace measure is locally finite. For arbitrary r<sr<s, approach both radii from below by nonexceptional radii. Continuity from below of a measure on open balls finishes the proof. ◻

Example 9.13.3. (Closedness cannot be omitted). On R\mathbb R, let TT be a nonzero nonnegative smooth scalar function supported in [−1,1][-1,1], viewed as a current of bidimension (1,1)(1,1). Its trace is T(x) dxT(x)\,dx. For r>1r>1, the mass μT((−r,r))\mu_T((-r,r)) is constant, so its ratio to rr is strictly decreasing. The current is positive but not closed. Thus the literal statement of Proposition 3.4 in the consulted arXiv v1 of Lagerberg requires a closedness hypothesis; its Stokes proof uses exactly that hypothesis. Corollary 3.5 in that source does include closedness.

9.14 Hausdorff noncharging and support dimension

Corollary 9.14.1. For a compact K⋐ΩK\Subset\Omega, there are R>0R>0 and CK<∞C_K<\infty such that μT(B(a,r))≤CKrp(a∈K, 0<r<R).\mu_T(B(a,r))\leq C_Kr^p\qquad(a\in K,\ 0<r<R).

Proof. Choose RR so that all closed balls B‾(a,R)\overline B(a,R) with a∈Ka\in K lie in a fixed compact subset K′⋐ΩK'\Subset\Omega. Monotonicity gives μT(B(a,r))≤(r/R)pμT(K′)\mu_T(B(a,r))\leq(r/R)^p\mu_T(K'). Take CK=R−pμT(K′)C_K=R^{-p}\mu_T(K'). ◻

Theorem 9.14.2. Let TT be an ordinary positive closed current of bidimension (p,p)(p,p) on Ω\Omega. If a Borel set E⊂ΩE\subset\Omega has zero pp-dimensional Hausdorff measure, then every coefficient measure of TT has zero total variation on EE. In particular, if the support of TT has zero pp-dimensional Hausdorff measure, then T=0T=0.

Proof. Work first with EE in a fixed compact subset KK. For every ε>0\varepsilon>0, the Hausdorff assumption provides a countable ball cover E⊂⋃jB(aj,rj)E\subset\bigcup_jB(a_j,r_j) with centres in KK, radii below the RR of Corollary 9.14.1, and ∑jrjp<ε\sum_jr_j^p<\varepsilon after adjusting an inessential dimensional constant. Therefore μT(E)≤∑jμT(B(aj,rj))≤CK∑jrjp<CKε.\mu_T(E)\leq\sum_j\mu_T(B(a_j,r_j)) \leq C_K\sum_jr_j^p<C_K\varepsilon. Letting ε↓0\varepsilon\downarrow 0 gives zero trace mass. Ordinary positivity gives the coefficient total-variation bounds by the trace measure, so every coefficient vanishes on EE. Exhaustion by compact subsets proves the general assertion. If all coefficients are supported on such an EE, they are zero. ◻

The last step is exactly where ordinary positivity is used. The monotonicity of a scalar trace alone cannot recover coefficients invisible to that trace. Lagerberg’s Remark 3.3 already records this real-versus-complex distinction. It is the same distinction behind the strong-cone correction in Chapter 7.

Lelong numbers and the metric normalization

If ωp\omega_p is the volume of the Euclidean unit ball in Rp\mathbb R^p, the limit ν(T,a)=lim⁡r↓0μT(B(a,r))ωprp\nu(T,a)=\lim_{r\downarrow0}\frac{\mu_T(B(a,r))}{\omega_pr^p} exists and is finite by monotonicity and local finiteness. For a smooth current of positive codimension, its trace has locally bounded density in nn dimensions and p<np<n, so ν(T,a)=0\nu(T,a)=0. A coordinate pp-plane with multiplicity mm has ν=m\nu=m in the standard coordinates.

Example 9.14.3. In R2\mathbb R^2, let f(x,y)=max⁡(0,x+2y)f(x,y)=\max(0,x+2y). Its Hessian matrix current is aatδ(⟨a,(x,y)⟩)aa^{\mathsf t}\delta(\langle a,(x,y)\rangle) with a=(1,2)a=(1,2). The trace is 5δ(x+2y) dx dy=5 ds5\delta(x+2y)\,dx\,dy=\sqrt 5\,ds on the line, where dsds is Euclidean arclength. Hence ν(DDf,0)=5\nu(DDf,0)=\sqrt 5. The tropical divisor has lattice weight one. There is no contradiction: Euclidean radial density and lattice multiplicity use different normalizations.

To obtain a support theorem on an analytic space, one must relate these real trace estimates to the measures attached to every actual analytic test. A single tropical pushforward may lose directions and even vanish when the map has insufficient rank. The analytic calibration and chart-detection arguments in CLD25 are therefore substantive extra steps. Section 19.14 uses the resulting all-test noncharging property, while the theorem above supplies its real measure-theoretic model.

References. [Lag12] (Definition 3.1, Remark 3.3, Proposition 3.4 and Corollary 3.5); compare [CLD25] (Section 16.8). Closedness and the distinction between weak and ordinary positivity have been retained explicitly.

9.15 Local equality, gluing and the full test space

A current identity is local because tests can be localized by a smooth partition of unity. This is different from checking the identity on a smaller chosen family of tests. Localization keeps the test category fixed. The distinction is particularly useful when a delta-form identity is proved in different local presentations.

Proposition 9.15.1. Let XX have a test-form sheaf admitting smooth partitions of unity, with continuous multiplication by fixed smooth functions. Suppose TiT_i are currents of one degree on an open cover UiU_i, and their restrictions agree on overlaps. Then they glue uniquely to a current TT on XX. Closedness is local. Positivity is local for a test cone stable under restriction and multiplication by nonnegative smooth functions.

Proof. Choose a locally finite nonnegative smooth partition ρi\rho_i subordinate to the cover and set T(η)=∑iTi(ρiη).T(\eta)=\sum_iT_i(\rho_i\eta). Only finitely many summands meet the compact support of a test. Each product is supported compactly inside UiU_i. If σj\sigma_j is another partition, insert ∑jσj=1\sum_j\sigma_j=1 and use agreement on overlaps: ∑iTi(ρiη)=∑i,jTi(ρiσjη)=∑i,jTj(ρiσjη)=∑jTj(σjη).\sum_iT_i(\rho_i\eta) =\sum_{i,j}T_i(\rho_i\sigma_j\eta) =\sum_{i,j}T_j(\rho_i\sigma_j\eta) =\sum_jT_j(\sigma_j\eta). Thus the definition is independent of the partition. The same calculation for a test supported in one UiU_i shows T∣Ui=TiT|_{U_i}=T_i. On each fixed presentation space, the sum is finite and multiplication by each ρi\rho_i is continuous after a common refinement, so the resulting functional is continuous in the final test topology. Uniqueness follows by applying any proposed global current to the partitioned test.

Multiplication by a nonnegative scalar preserves the chosen positive cone, so local positivity implies global positivity. Differentiation commutes with restriction, hence d′Td'T and d′′Td''T restrict to the corresponding local derivatives. If these vanish on the cover, uniqueness of gluing makes them zero globally. The converse implications are immediate by restriction. ◻

The proposition applies to the original CLD test theory on its usual paracompact domains. It also applies to an enlarged test theory when that theory is a module over the smooth scalar functions and has the stated topology. To prove an identity on all actual delta-tests, one still needs the local identity on every localized actual delta-test. A partition of unity does not turn an arbitrary delta-test into an external product or an original smooth test.

9.16 Zero extension is not automatic closed extension

An order-zero current on X∖EX\setminus E may extend by zero if its coefficient measures have finite local mass near EE. Positivity survives this operation when the coefficient cone is preserved, but closedness can fail because differentiation detects a boundary term. A simple real example is the scalar current 1(0,∞)1_{(0,\infty)} on R∖{0}\mathbb R\setminus\{0\}. It is closed on that open set; its zero extension to R\mathbb R has derivative δ0d′x\delta_0d'x.

The following sufficient criterion exhibits the estimate needed for a removable set. It is an elementary consequence of coefficient domination and a Hausdorff covering argument. Its growth hypothesis is explicit; closedness only on the complement does not supply that hypothesis at centres lying in the removed set.

Theorem 9.16.1. (An elementary removal criterion). Let EE be relatively closed in Ω⊂Rn\Omega\subset\mathbb R^n, and let TT be an ordinary positive closed current of bidimension (p,p)(p,p) on Ω∖E\Omega\setminus E, where p≥2p\geq 2. Assume its trace has finite mass on compact subsets up to EE. Assume also that on each fixed compact neighbourhood there is a uniform estimate μT(B(a,r)∖E)≤Crp(a∈E, 0<r<r0).\mu_T(B(a,r)\setminus E)\leq Cr^p \quad(a\in E,\ 0<r<r_0). If Hp−1(E)=0\mathcal H^{p-1}(E)=0, the zero extension T~\widetilde T is positive and closed on Ω\Omega.

Proof. The coefficient total variations are bounded by a fixed multiple of the trace locally. Their finite-mass zero extensions therefore define an order-zero current, and integration against positive tests remains nonnegative. Fix a smooth test η\eta with compact support, and work in a slightly larger compact neighbourhood. For arbitrary small ε,δ>0\varepsilon,\delta>0, cover the portion of EE there by finitely many balls B(aj,rj)B(a_j,r_j) centred on EE, with rj<δr_j<\delta and ∑jrjp−1<ε\sum_jr_j^{p-1}<\varepsilon. Compactness allows a finite subcover of a Hausdorff cover.

Choose smooth bumps χj\chi_j, equal to one on B(aj,rj)B(a_j,r_j), supported in B(aj,2rj)B(a_j,2r_j), with ∣∇χj∣≤C0/rj|\nabla\chi_j|\leq C_0/r_j. Apply a fixed smooth truncation to ∑jχj\sum_j\chi_j to obtain 0≤χ≤10\leq\chi\leq 1, equal to one near the relevant part of EE, supported in the union of the doubled balls, and satisfying ∣∇χ∣≤C1∑jrj−11B(aj,2rj)|\nabla\chi|\leq C_1\sum_jr_j^{-1}1_{B(a_j,2r_j)}. Choose radii small enough that all doubled balls remain in the fixed neighbourhood.

The test (1−χ)η(1-\chi)\eta is supported away from EE, so closedness of TT applies. Up to the harmless transpose sign, the pairing of d′T~d'\widetilde T with η\eta is bounded by ∣T~(d′η)∣≤∣T~(χd′η)∣+∣T~(d′χ∧η)∣≤Cη∑j(rjp+rjp−1)≤Cη(1+δ)ε.\begin{align*} |\widetilde T(d'\eta)| &\leq|\widetilde T(\chi d'\eta)| +|\widetilde T(d'\chi\wedge\eta)|\\ &\leq C_\eta\sum_j\bigl(r_j^p+r_j^{p-1}\bigr) \leq C_\eta(1+\delta)\varepsilon. \end{align*} The constants absorb the doubled radii and the coefficient estimate. Letting ε↓0\varepsilon\downarrow 0 gives d′T~=0d'\widetilde T=0. The same argument with the double-primed differential gives d′′T~=0d''\widetilde T=0. ◻

The threshold is dictated by the derivative of the cutoff: a ball mass of order rpr^p loses one power of rr under differentiation. This explains why the noncharging theorem of Section 9.14 and the removal criterion have different Hausdorff exponents. Noncharging concerns the current itself; removable closedness concerns its first derivative. Stronger extension theorems, such as the toric Skoda–El Mir comparison, use additional geometry and should be cited with their own hypotheses.

References. The gluing construction is the standard duality argument for the CLD test sheaf, [CLD12] (Section 4). The removal estimate combines the coefficient control and ball-growth bounds developed above; it does not replace [BGJK] (Theorem 7.2.4).