15 · Polyhedral intersection calculus

9 NAG Tags

15.1 Tropical preforms and their representations

Let NN be a lattice, let Ω~⊂NR\widetilde\Omega\subset N_{\mathbb R} be open, and let CiC_i be tropical cycles with smooth real weights as in Chapter 5. Their currents δCi\delta_{C_i} have bidegree equal to their codimension.

Definition 15.1.1. (Tropical delta-preform). A delta-preform of bidegree (p,q)(p,q) on Ω~\widetilde\Omega is a supercurrent admitting an expression α=∑i=1mαi∧δCi,αi∈Ap−ℓi,q−ℓi(Ω~),codim⁡Ci=ℓi.\alpha=\sum_{i=1}^m\alpha_i\wedge\delta_{C_i}, \qquad \alpha_i\in A^{p-\ell_i,q-\ell_i}(\widetilde\Omega), \qquad \operatorname{codim}C_i=\ell_i. The vector space of these currents is denoted Pp,q(Ω~)P^{p,q}(\widetilde\Omega). If every ℓi=ℓ\ell_i=\ell, the expression has pure codimension ℓ\ell; the coefficient bidegree is then (p−ℓ,q−ℓ)(p-\ell,q-\ell).

The defining data are representations of a current, rather than additional labels attached to it. The product on PP combines the wedge product of the coefficients with stable intersection of tropical cycles. Its independence of the representation is a theorem [GK17] (Definition 2.9 and Proposition 2.12). It is not a product defined on arbitrary supercurrents.

Remark 15.1.2 (Codimension zero does not mean smooth). With smooth weights allowed, a codimension-zero tropical cycle on NRN_{\mathbf R} is a piecewise smooth function on that space. Consequently, a codimension-zero delta-preform can have piecewise smooth coefficients [GK17] (Remark 1.7 and Example 2.10). For example, ∣t∣[R]|t|[\mathbf R] has codimension zero but is not represented by a smooth function near 00. Thus codimension and regularity are different attributes.

15.2 Recovering coefficients from a polyhedral current

The product of delta-preforms is a product of currents, not of chosen lists of generators. Its construction therefore needs a detection statement before a multiplication formula. We give the argument in fixed polyhedral codimension, as in GK17, Sections 2.11–2.12.

Lemma 15.2.1. Let C\mathscr C be a locally finite polyhedral complex in NRN_{\mathbb R}, and let S=∑σ∈Cdασ∧δσ,S=\sum_{\sigma\in\mathscr C_d}\alpha_\sigma\wedge\delta_\sigma, where each ασ\alpha_\sigma is a smooth tangential superform on σ\sigma, extending smoothly to a neighbourhood in its affine span. If S=0S=0, then each ασ=0\alpha_\sigma=0. Consequently two such currents on a common complex are equal exactly when their tangential coefficients agree.

Proof. Fix a point in the relative interior of one dd-cell. A sufficiently small neighbourhood avoids the other dd-cells, since their intersections with this cell are proper faces and the complex is locally finite. On that neighbourhood, testing SS is testing ασ\alpha_\sigma by integration on the affine span of σ\sigma.

Choose affine coordinates on this span. For any coefficient of ασ\alpha_\sigma, wedge with the complementary primed and double-primed basis elements, choosing their order to fix the sign, and multiply by an arbitrary compactly supported scalar test. Such a test extends to the ambient space by a linear projection and a normal cutoff. The vanishing of SS says that the integral of the chosen coefficient against every scalar test is zero. The fundamental lemma for distributions gives that coefficient zero. All coefficients vanish on the relative interior and, by continuity, on its closure. Repeat for each cell. ◻

A representative ∑iαi∧δCi\sum_i\alpha_i\wedge\delta_{C_i} has, after a common subdivision, the coefficient ∑imi,σαi∣σ\sum_i m_{i,\sigma}\alpha_i|_\sigma on a cell σ\sigma. This coefficient is intrinsic to the current on that complex. It is a tangential form: adding an ambient form which restricts to zero on σ\sigma changes nothing. This last point is essential when the support has positive codimension.

15.3 The geometric input for stable intersection

We use the stable-intersection theorem for tropical cycles with smooth weights in GK17, Section 1.4. After common subdivision, a generic displacement computes the coefficient of a codimension ℓ+ℓ′\ell+\ell' intersection face τ\tau by summing over contributing pairs of faces σ,σ′\sigma,\sigma', of codimensions ℓ,ℓ′\ell,\ell'. A pair contributes only if it survives the displacement test and Nσ+Nσ′N_\sigma+N_{\sigma'} has full rank. Its factor is [N:Nσ+Nσ′] mσmσ′.[N:N_\sigma+N_{\sigma'}]\,m_\sigma m_{\sigma'}. The cycle theorem asserts that the resulting cycle is independent of the permitted sufficiently small generic displacement and of subdivision, and that the cycle product is associative and commutative. We use those assertions as geometric inputs. They are not consequences of the coefficient-detection lemma.

NAG-900283 · Conditional on stated inputs

Theorem 15.3.1. (Representation independence). Assume the stated cycle-intersection theorem. For delta-preforms of fixed polyhedral codimensions, the formula (∑iαi∧δCi)∧(∑jβj∧δDj)=∑i,j(αi∧βj)∧δCi⋅Dj\left(\sum_i\alpha_i\wedge\delta_{C_i}\right) \wedge\left(\sum_j\beta_j\wedge\delta_{D_j}\right) =\sum_{i,j}(\alpha_i\wedge\beta_j)\wedge\delta_{C_i\cdot D_j} is independent of the chosen representations. It is bilinear, associative and graded commutative for total superform degree, and agrees with multiplication by a smooth superform.

Proof. Choose a common complex for all cycles in two proposed representations of each factor, refining further so that the displacement formula applies. Let aσa_\sigma and bσ′b_{\sigma'} be the intrinsic tangential coefficients of the two currents on this complex. On an intersection face τ\tau the proposed product has coefficient ∑σ,σ′[N:Nσ+Nσ′](aσ∣τ)∧(bσ′∣τ),\sum_{\sigma,\sigma'}[N:N_\sigma+N_{\sigma'}] (a_\sigma|_\tau)\wedge(b_{\sigma'}|_\tau), with the same contributing-pair rule as the cycle formula. This expression follows by expanding the finite sums and collecting the cycle weights. Lemma 15.2.1 makes aσ,bσ′a_\sigma,b_{\sigma'} independent of the initial lists of generators. Restriction to τ\tau is well defined for tangential forms, because TτT\tau lies in both tangent spaces. Thus the coefficient of every intersection face is independent of those lists. Detection proves equality of the resulting currents.

Bilinearity is immediate. Once representation independence is established, expand a triple product using the original smooth coefficients and cycles. Associativity of wedge and of the cycle product identifies the two expansions. Swapping the factors changes the smooth coefficients by the usual total-degree sign; twice the polyhedral codimension is even and adds no sign. Finally, a smooth form is represented using the unit cycle NRN_{\mathbb R}, so the formula reduces to its usual multiplication with the current. ◻

This proof separates two independence questions. The tropical cycle theorem controls displacement and subdivision. The coefficient argument controls representations of an already defined current. Conflating them hides the main point of GK17, Proposition 2.12. Neither independence statement defines a multiplication on arbitrary distributional currents.

15.4 Transverse lines and the determinant

Proposition 15.4.1. In R2\mathbb R^2 with lattice Z2\mathbb Z^2, let La,LbL_a,L_b be distinct lines through the origin with primitive integral direction vectors a,ba,b. With weight one on each line, δLa∧δLb=∣det⁡(a,b)∣δ0.\delta_{L_a}\wedge\delta_{L_b}=|\det(a,b)|\delta_0. If the two lines coincide, their stable intersection is zero.

Proof. Distinct lines meet transversely. After any sufficiently small generic translation they meet at one point tending to the origin. The cycle multiplicity is [Z2:Za+Zb]=∣det⁡(a,b)∣[\mathbb Z^2:\mathbb Za+\mathbb Zb]=|\det(a,b)|. For coincident lines choose a translation transverse to the line. The displaced lines are disjoint, so no face contributes. Displacement independence makes this the stable product. ◻

For u=max⁡(0,x)u=\max(0,x) and v=max⁡(0,x+2y)v=\max(0,x+2y), the two divisor directions may be chosen as (0,1)(0,1) and (2,−1)(2,-1). Their determinant has absolute value two, so d′d′′[u]∧d′d′′[v]=2δ0\mathrm d'\mathrm d''[u]\wedge\mathrm d'\mathrm d''[v]=2\delta_0 in this calculus. In contrast, the stable square of the straight divisor {x=0}\{x=0\} is zero. These identities express geometric intersection rules; they are not prescriptions for squaring Dirac distributions in ordinary distribution theory.

15.5 Balancing as cancellation of boundary currents

The corner calculation in Chapter 10 has a useful converse. For constant weights, balancing is precisely what removes the boundary term in Stokes’ formula. This gives a direct way to prove that a PL corner divisor is again balanced. Smoothly varying weights require an additional distinction, because their derivatives need not vanish.

Proposition 15.5.1. Let CC be a locally finite, pure dd-dimensional rational polyhedral complex in NRN_{\mathbb R} with constant real weights mσm_\sigma. Tests have compact support away from any designated boundary. The following conditions are equivalent: CC is balanced; d′δC=0d'\delta_C=0; d′′δC=0d''\delta_C=0.

Proof. Fix a codimension-one face τ\tau. Write nσ,τn_{\sigma,\tau} for an outgoing primitive normal lift and put bτ=∑σ⊃τmσnσ,τ(modNτ,R).b_\tau=\sum_{\sigma\supset\tau}m_\sigma n_{\sigma,\tau} \pmod{N_{\tau,\mathbb R}}. Apply ordinary Stokes on each maximal polyhedron to the coefficients of a superform. The interior derivatives of the weights vanish. Along τ\tau, all tangential coordinates and the induced lattice measure agree, so the sum of the boundary contractions depends only on bτb_\tau. A tangent change of normal has zero contraction in the normal quotient. Hence bτ=0b_\tau=0 makes the boundary contribution vanish. Summing over the finitely many cells meeting a test support proves both closedness assertions.

Conversely, take a point in the relative interior of τ\tau and a small neighbourhood meeting no other codimension-one face. Choose a compactly supported scalar bump on τ\tau whose lattice integral is one. Given any real linear functional ℓ\ell on NR/Nτ,RN_{\mathbb R}/N_{\tau,\mathbb R}, extend the bump to an ambient test and choose its remaining differential factors to be a tangential volume element and the normal covector ℓ\ell. The same Stokes calculation pairs the boundary current with this test as ℓ(bτ)\ell(b_\tau), up to one fixed nonzero sign. If d′δC=0d'\delta_C=0, every such pairing vanishes, so bτ=0b_\tau=0. The argument for d′′d'' is identical with the two copies of the cotangent space exchanged. ◻

Theorem 15.5.2. (A PL corner is balanced). Let CC have constant real weights and be balanced. If uu is continuous and affine on the cells of a subdivision of CC, the corner weights of Proposition 10.5.1 define a balanced cycle u⋅Cu\cdot C, and d′d′′(uδC)=δu⋅C.\mathrm d'\mathrm d''(u\delta_C)=\delta_{u\cdot C}.

Proof. On a maximal cell, the second derivatives of uu vanish. Integrate twice by parts in the pairing of uδCu\delta_C with d′d′′\mathrm d'\mathrm d'' of a test. The terms containing the value of uu times a normal derivative of the test cancel along each facet by continuity and balancing. The remaining coefficient on a facet τ\tau is ∑σ⊃τmσℓσ(nσ,τ)−ℓτ(∑σ⊃τmσnσ,τ),\sum_{\sigma\supset\tau}m_\sigma\ell_\sigma(n_{\sigma,\tau}) -\ell_\tau\left(\sum_{\sigma\supset\tau} m_\sigma n_{\sigma,\tau}\right), where ℓσ\ell_\sigma and ℓτ\ell_\tau are the linear parts of uu. This proves the current identity, including the tangent correction. It also proves that the result is a polyhedral current with constant facet weights. Now d′δu⋅C=d′d′d′′(uδC)=0,d′′δu⋅C=0.d'\delta_{u\cdot C}=d'\mathrm d'\mathrm d''(u\delta_C)=0, \qquad d''\delta_{u\cdot C}=0. Proposition 15.5.1 therefore proves balancing of the corner cycle. This last step uses only the differential identities and the preceding Stokes characterization. ◻

The proof also completes subdivision independence in this constant-weight PL setting. The distribution d′d′′(uδC)\mathrm d'\mathrm d''(u\delta_C) is unchanged by subdivision. On a common refinement, coefficient detection identifies its facet weights. Thus two descriptions give the same weighted cycle; the new facets inside an affine piece already have zero weight by Proposition 10.5.1.

Smooth weights and the cellwise differential

Balancing of smooth weights is a pointwise condition along a facet. It cancels boundary contributions but does not kill derivatives inside cells. Already on NR=RN_{\mathbb R}=\mathbb R, every smooth weight m(t)m(t) on the full line satisfies the codimension-zero balancing condition, whereas d′(mδR)=m′(t)d′t∧δR.d'(m\delta_{\mathbb R})=m'(t)d't\wedge\delta_{\mathbb R}. For m(t)=etm(t)=e^t, the right side is nonzero. Thus “balanced with smooth weights” and “closed integration current” are different assertions. The operators dP′,dP′′d'_P,d''_P record the cellwise terms. The difference between a distributional differential and its cellwise counterpart is the boundary operator. This distinction is used when selecting the universally closed generators in the GK sheaf.

15.6 Why the fan product is independent of displacement

For a finite complete rational fan Σ\Sigma in a rank-rr lattice NN, a codimension-pp Minkowski weight assigns a real number to each (r−p)(r-p)-dimensional cone and satisfies balancing at the next lower-dimensional cones. There is a geometric reason for exactly these linear equations.

Let XΣX_\Sigma be the associated complete toric variety. A cone σ\sigma of codimension pp has orbit closure V(σ)V(\sigma) of dimension pp. An operational Chow class c∈Ap(XΣ)c\in A^p(X_\Sigma) consequently defines mc(σ)=deg⁡(c∩[V(σ)]).m_c(\sigma)=\deg(c\cap[V(\sigma)]). The dimension convention is important: the codimension of the cone is the dimension of its orbit closure. For a cone τ\tau of codimension p+1p+1 and a character u∈M∩τ⊥u\in M\cap\tau^\perp, the divisor of that character on V(τ)V(\tau) gives the relation ∑σ⊃τdim⁡σ=dim⁡τ+1⟨u,nσ,τ⟩[V(σ)]=0in Ap(XΣ).\sum_{\substack{\sigma\supset\tau\\\dim\sigma=\dim\tau+1}} \langle u,n_{\sigma,\tau}\rangle[V(\sigma)]=0 \quad\text{in }A_p(X_\Sigma). Applying cc and taking degree says that the weighted sum of the normal vectors pairs to zero with every such uu. This is precisely balancing modulo NτN_\tau. Thus balancing is the numerical expression of principal-divisor relations.

NAG-900287 · Conditional on stated inputs

Theorem 15.6.1. (The toric input). For complete Σ\Sigma, the assignment c↦mcc\mapsto m_c identifies operational Chow cohomology, with real coefficients, with Minkowski weights. Its product is computed by the fan displacement formula: at a cone γ\gamma of codimension p+qp+q, sum [N:Nσ+Nτ] mc(σ)mc~(τ),[N:N_\sigma+N_\tau]\,m_c(\sigma)m_{\widetilde c}(\tau), over cones of codimensions p,qp,q containing γ\gamma whose generic displacement meets. This is the Fulton–Sturmfels theorem; the identification and the diagonal-degeneration formula are used here as inputs.

Displacement independence now has a precise proof from that input. Two generic vectors give two expressions for the same number deg⁡((cc~)∩[V(γ)]).\deg\bigl((c\widetilde c)\cap[V(\gamma)]\bigr). They must agree even though the contributing pairs may differ. Associativity and commutativity are inherited from the Chow product, not from an assertion that the displaced set-theoretic intersections are identical. A common toric subdivision compares different fans. Pullback of operational classes and the projection formula preserve the relevant degrees, which is the geometric source of refinement compatibility.

From constant fans to smooth polyhedral weights

NAG-900288 · Conditional on stated inputs

Proposition 15.6.2. Assume the toric fan product and its refinement compatibility. The displacement construction for rational polyhedral cycles with smooth weights is well defined, balanced pointwise, and compatible with subdivision. Its products are associative and commutative.

Proof. Fix a point xx in a common polyhedral complex. Replace every cell containing xx by its local tangent cone at xx, and freeze its weight at the value at xx. The pointwise balancing equations become exactly the balancing equations for this constant-weight fan. A compatible complete fan is obtained by adding zero-weight cones and refining. For each prospective intersection face, the small-displacement incidence condition on the original cells becomes the displacement condition on these tangent cones. The local coefficient is therefore the constant-fan product coefficient. It is independent of the generic vector and of the chosen complete refinement by Theorem 15.6.1.

On any fixed face, this coefficient is a finite sum of products of restrictions of the original smooth weights, multiplied by constant lattice indices. It is therefore smooth on that face. At a lower-dimensional face, freeze the weights again at a point. The balancing equation for the product is the constant-fan balancing equation there, so it holds. The same procedure shows subdivision compatibility: both subdivisions give the same local fan product at every point.

Finally, the local fan of a product at xx is the product of the local fans at xx, with their frozen weights. Thus both bracketings of a triple product have the same coefficient by associativity of the constant-fan product. Swapping two factors is treated in the same way. These are pointwise identities of smooth coefficient functions and hence identities of the weighted cycles. ◻

This argument uses values of weights, not derivatives of weights. It explains why stable intersection extends to smooth weights even though their integration currents need not be closed. Combining it with Theorem 15.3.1 proves representation independence of the resulting delta-preform algebra. The geometric and distributional independence statements are different steps, now linked by an explicit local construction.

Two displacement calculations

Example 15.6.3. (A tropical line intersects itself). Let L⊂R2L\subset\mathbb R^2 consist of the rays generated by e1,e2,−e1−e2e_1,e_2,-e_1-e_2, all with weight one. For v=(a,b)v=(a,b) with 0<a<b0<a<b, the ray R≥0e2\mathbb R_{\geq 0}e_2 meets the translated diagonal ray v+R≥0(−e1−e2)v+\mathbb R_{\geq 0}(-e_1-e_2) at (0,b−a)(0,b-a). Checking the three pairs of coordinate and diagonal directions shows that this is the only transverse intersection of LL and L+vL+v. Its lattice index is ∣det⁡(e2,−e1−e2)∣=1|\det(e_2,-e_1-e_2)|=1. Replacing vv by ϵv\epsilon v sends this point to the origin. Hence L⋅L=[0].L\cdot L=[0]. A straight tropical line has stable square zero, as in Proposition 15.4.1. The difference comes from the three-ray geometry, not from a formal rule about repeated supports.

Example 15.6.4. (Nonprimitive equations and weights). Let u=max⁡(0,ax)u=\max(0,ax) and v=max⁡(0,by)v=\max(0,by) for positive integers a,ba,b. The supports of their divisors are the coordinate lines, but their weights are a,ba,b. The determinant of the primitive line directions is one, so d′d′′[u]∧d′d′′[v]=abδ0.\mathrm d'\mathrm d''[u]\wedge\mathrm d'\mathrm d''[v]=ab\delta_0. Counting only the geometric intersection point loses both multiplicities. Conversely, using the nonprimitive vectors in the determinant and also retaining the weights would count them twice. Primitive directions and cycle weights must therefore be recorded separately.

References. The Stokes and corner-current framework is [GK17] (Sections 2–3). The toric input is [FS] (Theorems 2.1 and 3.2); the passage to smooth weights follows the local-fan construction of [GK17] (Remark 1.4), with its pointwise steps written above.

15.7 The delta-algebra and intersection theory

Polyhedral curvature terms

For a smooth real function, the Hessian is a smooth (1,1)(1,1)-form. For a PL function, the cellwise Hessian vanishes and the distributional Hessian is a corner current. A calculus limited to original smooth forms cannot treat that corner current as an ordinary smooth coefficient. GK enlarges the algebra so that certain polyhedral currents become legitimate differential objects with products.

The enlargement is controlled. A typical preform is a finite sum of smooth superforms multiplied by integration currents on weighted polyhedral cycles. Stable intersection supplies the product of the cycle factors. Differential operators have both cellwise and boundary parts. The functorial closedness conditions ensure that the construction survives refinements and pullbacks in the specified algebraic category.

This is why the detailed definition in the preceding sections is needed. Declaring every polyhedral current to be a form would not automatically make its derivatives remain in the class or make products independent of the presentation.

A local multiplication that ordinary distribution theory cannot supply

On R2\mathbb R^2, let u=max⁡(0,x)u=\max(0,x) and v=max⁡(0,y)v=\max(0,y). Their curvature currents are integration currents on the coordinate lines, with the corresponding normal differential degrees. The delta product is the point mass at the origin, with multiplicity one. Replacing yy by x+2yx+2y gives the determinant factor two after the correct lattice normalization.

If two divisors coincide, ordinary multiplication of their distributional densities is not defined by writing a square of a Dirac delta. Tropical stable intersection provides a geometric intersection operation in its admissible cycle class. The product belongs to that calculus because its construction includes a displacement or equivalent intersection-theoretic definition. This does not supply a product for every pair of distributions with the same supports.

The metric application

A model metric has PL local weights. The corner formula produces its curvature as a delta-form in the appropriate GK setting. Products of these Chern delta-forms can then be integrated and compared with model intersection numbers. Green currents and star products lead to local heights, with Poincaré–Lelong providing the divisor term.

The arithmetic importance is that the calculus retains lower-degree curvature, not only a final scalar intersection number. It gives local formulas for metric variation and intersections, while compatibility with the model theory verifies the normalization. The resulting local heights can then be assembled in global arithmetic settings with their own adelic hypotheses.

Mihatsch’s analytic construction and the comparison boundary

Mihatsch develops delta-forms intrinsically on analytic spaces using compatible skeletal descriptions and a polyhedral-current calculus. The resulting trigrading records smooth bidegrees and polyhedral codimension. In the notation of that paper, the codimension-zero component is the piecewise smooth calculus, and the realization in CLD currents has an injectivity theorem under the stated analytic hypotheses.

These are useful improvements for actual representatives and higher-codimension Green identities. They should not be retroactively inserted into the original GK definition without a comparison argument. In particular, a property proved for the later analytic delta-sheaf does not resolve an original GK representability question solely because both are denoted by BB in informal discussions.

The complete-intersection Green formula is a case where the distinction matters constructively. CLD25 states a current identity on original smooth tests. Mihatsch’s theorem supplies an actual delta-form on the complement in its own category and an identity against the appropriate delta-form tests. The stronger representation and testing statement is additional mathematical content.

References. [GK17] (Sections 2–6, 9–12); [Mih24] (Propositions 4.7, 4.11 and Theorem 5.5).

15.8 Products and tropical intersection in CLD25 Chapter 9

External products come before internal intersections

If SS is a current on XX and TT a current on YY, their external product belongs on X×YX\times Y. Formally it is obtained by integrating a test in one variable against one current and then against the other. To make this rigorous, CLD25 Chapter 9 develops integration with parameters and the continuity of the resulting test-valued operations. The construction is more general than merely writing a product of two scalar measures.

An internal intersection on XX can then be related to the diagonal in X×XX\times X. But pulling a current back to the diagonal is a singular restriction problem. A Gysin construction supplies such an operation in a controlled class; an arbitrary pair of distributions cannot be intersected just because their external product exists.

Why displacement appears

In a real vector space, translate one tropical cycle by a small generic vector. The transverse intersections have lattice-index multiplicities. Passing to the stable value gives the familiar tropical intersection. This can be interpreted through a family of currents and a diagonal or Gysin operation. The analytic formulation explains why the result is compatible with differential operations and why multiplicities are conserved under small displacements.

For two transverse lines in R2\mathbb R^2, the determinant computation in Chapter 5 is the entire local calculation. For nontransverse cycles, the displacement construction resolves an excess-dimensional intersection into weighted pieces of the expected dimension. The support of the result is a subset of the set-theoretic intersection, but the multiplicities require the displacement and lattice data.

Relation to the GK product

NAG-900289 · Comparison requirements remain

GK begins with polyhedral currents and stable intersections to construct a delta-algebra on analytic spaces. CLD25 develops product spaces, parameter-dependent integrals, and Gysin operations inside its abstract tropical framework. These approaches have overlapping local models and related formulas. An identification on a chosen class still requires matching the calibration, the allowable generators, and the current realization. The existence of both products is not a proof that the entire algebras coincide.

This point will matter for the research questions in Chapter 24. Extending a geometric intersection product to arbitrary positive closed currents requires new analytic continuity and representation results. The tropical product explains the finite polyhedral case and provides examples, but it does not automatically supply all bounded-potential or unbounded-current products.

References. [CLD25] (§§7.5–7.8, 8.10–8.11, Chapter 9); [GK17] (Sections 2–4).

15.9 Changing conventions without changing curvature

The six operators of Section 16.5 use Mihatsch’s boundary convention. Cai’s convention is the opposite. The current derivatives and the polyhedral derivatives agree, while ∂C′=d′−dP′=−∂M′,∂C′′=d′′−dP′′=−∂M′′.\partial'_{\mathrm C}=d'-d'_P=-\partial'_{\mathrm M}, \qquad \partial''_{\mathrm C}=d''-d''_P=-\partial''_{\mathrm M}. Here the subscripts label conventions, not different geometric operations. For a continuous piecewise smooth function ff, both conventions give d′d′′f=dP′dP′′f+∂C′dP′′f=dP′dP′′f−∂M′dP′′f.\begin{align*} d'd''f&=d'_Pd''_Pf+\partial'_{\mathrm C}d''_Pf\\ &=d'_Pd''_Pf-\partial'_{\mathrm M}d''_Pf. \end{align*} Thus the corner current is +∂C′dP′′f+\partial'_{\mathrm C}d''_Pf in Cai’s notation and −∂M′dP′′f-\partial'_{\mathrm M}d''_Pf in Mihatsch’s. For f(t)=max⁡(0,t)f(t)=\max(0,t) it is δ0\delta_0, with positive coefficient in both descriptions. This example is a useful normalization test: changing the sign of the boundary operator must not change the sign of the Monge–Ampère mass.

References. [Mih23] (Definition 3.4); [Cai23] (Theorem 4.25 and the conventions preceding Theorem 1.2).

15.10 Graph pullback and two pushforwards

Let F:NR′→NRF:N'_{\mathbb R}\to N_{\mathbb R} be an integral affine map. The graph construction defines a pullback of delta-forms even when FF is not surjective. If p1,p2p_1,p_2 are the projections from NR′×NRN'_{\mathbb R}\times N_{\mathbb R} and ΓF\Gamma_F is the graph, the defining formula is F∗β=(p1)∗([ΓF]∧([NR′]×β)).F^*\beta=(p_1)_*\bigl([\Gamma_F]\wedge([N'_{\mathbb R}]\times\beta)\bigr). The intersection on the right is the delta-product. It is not restriction of the coefficients of an arbitrary distribution. The projection has the required properness on the graph support, locally over the source. The construction preserves total bidegree and commutes with the wedge product and all six operators. These are theorems of the polyhedral calculus, not consequences of continuity of FF.

Example 15.10.1. (A non-surjective pullback). Let i:R→R2i:\mathbb R\to\mathbb R^2 be t↦(t,0)t\mapsto(t,0), and let Dx=d′d′′max⁡(0,x)D_x=d'd''\max(0,x) and Dy=d′d′′max⁡(0,y)D_y=d'd''\max(0,y). Then i∗Dx=δ0,i∗Dy=0.i^*D_x=\delta_0,\qquad i^*D_y=0. Indeed pullback commutes with d′d′′d'd'' and with piecewise linear functions. The pulled-back functions are max⁡(0,t)\max(0,t) and 00. Although i(R)i(\mathbb R) is contained in the support of DyD_y, its delta-pullback is zero. A set-theoretic inverse image of supports does not determine a pullback multiplicity.

Two operations are denoted by pushforward in the literature. The current pushforward F∗F_* is defined by duality with smooth tests; it requires properness on support, or compact support in the version used in Cai’s Theorem 5.13(3). It can integrate coefficient degrees along positive-dimensional fibres. The cycle-type pushforward F^∗\widehat F_* instead preserves the dimension of the supporting cells. A cell on which dimension drops contributes zero. For an ss-dimensional cell σ\sigma mapped injectively to an ss-dimensional cell it is F^∗(ασ[σ])=[NF(σ):dF(Nσ)]((F∣σ)−1)∗ασ [F(σ)].\widehat F_*(\alpha_\sigma[\sigma]) =[N_{F(\sigma)}:dF(N_\sigma)] ((F|_\sigma)^{-1})^*\alpha_\sigma\,[F(\sigma)]. Cai proves a projection formula for each operation with its own domain. The formula involving F^∗\widehat F_* on the full affine spaces does not remove the support condition from the current pushforward. When FF is injective on every supporting cell the two constructions agree.

Proposition 15.10.2. (The lattice factor in dimension one). For F:R→RF:\mathbb R\to\mathbb R, t↦mtt\mapsto mt with a nonzero integer mm, normalized using the standard lattices, F∗[R]=∣m∣[R],F∗δ0=δ0,F∗δ0=∣m∣δ0.F_*[\mathbb R]=|m|[\mathbb R],\qquad F_*\delta_0=\delta_0, \qquad F^*\delta_0=|m|\delta_0.

Proof. For a compactly supported top-degree test a(y)d′y∧d′′ya(y)d'y\wedge d''y, its pullback is m2a(mt)d′t∧d′′tm^2a(mt)d't\wedge d''t. Change of variables in the real integral gives the factor m2/∣m∣=∣m∣m^2/|m|=|m|. The second formula follows by evaluation on a scalar test. For the third, write δ0=d′d′′max⁡(0,y)\delta_0=d'd''\max(0,y). The slopes of max⁡(0,mt)\max(0,mt) have jump ∣m∣|m| at zero. ◻

These formulas satisfy the projection formula: pushing F∗δ0∧[R]F^*\delta_0\wedge[\mathbb R] forward gives ∣m∣δ0|m|\delta_0, as does δ0∧F∗[R]\delta_0\wedge F_*[\mathbb R]. They also show why the mass of a point and the weight of a one-dimensional cycle must not be normalized by the same informal “degree rule”. Passing from this affine construction to analytic spaces requires compatibility of charts and their test quotients. Cai’s algebraic construction, GK’s original sheaf and Mihatsch’s analytic sheaf must therefore still be distinguished. None of the affine formulas alone supplies a trace of arbitrary actual delta-forms through every proper analytic map.

References. [Mih23] (Theorem 4.1 and Proposition/Definition 4.2); [Cai23] (Proposition 5.4, Remark 5.5, Definition 5.6 and Theorem 5.13). The displayed examples are direct computations.