7 · Superforms and positivity cones

18 NAG Tags

The real differential algebra fixes the signs of integration, curvature and the divisor formula. We develop its positivity cones through their matrix representations, then prove the domination estimates used for currents. The distinction between ordinary and strong positivity is essential in middle degree.

7.1 The differential algebra

Definition 7.1.1. (Real superforms). Let V=RnV=\mathbb R^n with coordinates x1,…,xnx_1,\ldots,x_n, and let Ω⊂V\Omega\subset V be open. Write V′∗V'^* and V′′∗V''^* for two copies of V∗V^*. The algebra of superforms is Ap,q(Ω)=C∞(Ω)⊗⋀pV′∗⊗⋀qV′′∗.A^{p,q}(\Omega)=C^\infty(\Omega)\otimes \bigwedge^p V'^*\otimes\bigwedge^q V''^*. All one-form generators anticommute, including generators from different copies. Thus a form is uniquely written as α=∑∣I∣=p, ∣J∣=qαIJ d′xI∧d′′xJ,d′xI=d′xi1∧⋯∧d′xip,\alpha=\sum_{|I|=p,\ |J|=q}\alpha_{IJ}\,d'x_I\wedge d''x_J, \qquad d'x_I=d'x_{i_1}\wedge\cdots\wedge d'x_{i_p}, where indices in a multiindex are increasing. The total degree is p+qp+q. In particular, α∧β=(−1)(p+q)(r+s)β∧α(β∈Ar,s).\alpha\wedge\beta=(-1)^{(p+q)(r+s)}\beta\wedge\alpha \quad(\beta\in A^{r,s}).

Definition. The involution JJ is the algebra automorphism exchanging d′xid'x_i and d′′xid''x_i and fixing functions. A (p,p)(p,p)-form is symmetric if Jα=(−1)pαJ\alpha=(-1)^p\alpha.

Moving the primed generators to the left gives J(d′xI∧d′′xJ)=(−1)pqd′xJ∧d′′xI.\begin{equation} J(d'x_I\wedge d''x_J)=(-1)^{pq}d'x_J\wedge d''x_I. \end{equation} Since p2p^2 and pp have the same parity, symmetry in bidegree (p,p)(p,p) means precisely αIJ=αJI\alpha_{IJ}=\alpha_{JI}. This is the symmetry that will become ordinary matrix symmetry.

Definition. The differentials are the total-degree derivations determined by d′f=∑i∂if d′xi,d′′f=∑i∂if d′′xi,d′d′xi=d′d′′xi=d′′d′xi=d′′d′′xi=0.\begin{gather*} d'f=\sum_i\partial_i f\,d'x_i,\qquad d''f=\sum_i\partial_i f\,d''x_i,\\ d'd'x_i=d'd''x_i=d''d'x_i=d''d''x_i=0. \end{gather*}

For the displayed expansion of α\alpha, these definitions give d′α=∑I,J,i∂iαIJ d′xi∧d′xI∧d′′xJ,d′′α=(−1)p∑I,J,i∂iαIJ d′xI∧d′′xi∧d′′xJ.\begin{align*} d'\alpha&=\sum_{I,J,i}\partial_i\alpha_{IJ}\, d'x_i\wedge d'x_I\wedge d''x_J,\\ d''\alpha&=(-1)^p\sum_{I,J,i}\partial_i\alpha_{IJ}\, d'x_I\wedge d''x_i\wedge d''x_J. \end{align*}

Proposition 7.1.4. The operators satisfy (d′)2=(d′′)2=0,d′d′′+d′′d′=0,d′′=Jd′J.(d')^2=(d'')^2=0,\quad d'd''+d''d'=0,\quad d''=Jd'J. For a smooth function ff, one has d′d′′f=∑i,j∂i∂jf d′xi∧d′′xj.\mathrm d'\mathrm d''f=\sum_{i,j}\partial_i\partial_j f\,d'x_i\wedge d''x_j.

Proof. The terms in (d′)2α(d')^2\alpha cancel in pairs: the second derivatives are symmetric in the two new indices, whereas the primed generators are antisymmetric. The same argument applies to d′′d''. In d′d′′α+d′′d′αd'd''\alpha+d''d'\alpha, the derivatives again agree, and exchanging the two new generators reverses the sign. The identity with JJ can be checked on functions and the one-form generators, after which the derivation rule proves it on the algebra. Applying d′d' to d′′fd''f gives the last formula. ◻

Remark 7.1.5 (A sign that must be corrected in the 2012 text). The displayed commutativity exponent in [CLD12] (§1.2.6, p. 14) is pp′+qq′pp'+qq'. Read literally, it would make d′xid'x_i commute with d′′xjd''x_j, contrary to (7.1.1) and to the total-degree Leibniz rule used later in that text. The displayed exponent in [CLD12] (§3.1.4, p. 36) is also incompatible with those formulas. Throughout these notes the exponent is (p+q)(p′+q′)(p+q)(p'+q'). This is an explicit correction of the displayed formulas, not a second convention being used implicitly.

Proposition 7.1.6. (Conversion of the two operators). Let JJ be the involution in Definition 7.1.1. The algebra automorphism in [CLD25] (§1.1.2) satisfies J25(d′xi)=d′′xi,J25(d′′xi)=−d′xi.J_{25}(d'x_i)=d''x_i,\qquad J_{25}(d''x_i)=-d'x_i. On forms of bidegree (p,q)(p,q) one has J25=(−1)qJ,J252=(−1)p+qid⁡.\begin{equation} J_{25}=(-1)^qJ,\qquad J_{25}^{2}=(-1)^{p+q}\operatorname{id}. \end{equation} For a form of type (p,p)(p,p), the old symmetry condition Jα=(−1)pαJ\alpha=(-1)^p\alpha is equivalent to J25α=αJ_{25}\alpha=\alpha. The strong generators and the signed squares (−1)p(p−1)/2β∧Jβ(-1)^{p(p-1)/2}\beta\wedge J\beta, with β\beta of type (p,0)(p,0), are unchanged when JJ is replaced by J25J_{25}. Hence the three cones in Definition 7.2.2 agree under this conversion.

Proof. Every double-primed generator contributes one extra minus sign under J25J_{25}, giving the first equality. Applying it twice gives (−1)p+qJ2(-1)^{p+q}J^2. The symmetry assertion follows by setting q=pq=p. A form β\beta of type (p,0)(p,0) contains no double-primed factor, so J25β=JβJ_{25}\beta=J\beta. This proves the statement about the generators and, by the complementary wedge definitions, all three cones. ◻

Put sp=(−1)p(p−1)/2s_p=(-1)^{p(p-1)/2} and fix the positive top form voln=d′x1∧d′′x1∧⋯∧d′xn∧d′′xn=sn d′x1⋯n∧d′′x1⋯n.\mathrm{vol}_n=d'x_1\wedge d''x_1\wedge\cdots\wedge d'x_n\wedge d''x_n =s_n\,d'x_{1\cdots n}\wedge d''x_{1\cdots n}. Integration means ∫Ωfvoln=∫Ωf dx1⋯dxn\int_\Omega f\mathrm{vol}_n=\int_\Omega f\,dx_1\cdots dx_n. The choice is compatible with the lattice normalization when the coordinates form a lattice basis. Under an arbitrary linear change of coordinates the coefficient of this top form changes by the square of the determinant; the normalization of the integral must therefore be transformed as well.

7.2 Symmetric forms as matrices

Remark 7.2.1 (Matrix normalization). The factor sps_p removes the sign caused by collecting all primed generators. For a real m×mm\times m matrix HH, where m=(np)m=\binom np, define αH=sp∑I,JHIJ d′xI∧d′′xJ.\begin{equation} \alpha_H=s_p\sum_{I,J}H_{IJ}\,d'x_I\wedge d''x_J. \end{equation} Thus HH is symmetric exactly when αH\alpha_H is symmetric. If a=∑IaId′xIa=\sum_I a_I d'x_I, then sp a∧Ja=αaat.s_p\,a\wedge Ja=\alpha_{aa^t}.

Definition 7.2.2. A symmetric form αH\alpha_H is positive if HH is positive semidefinite. It is strongly positive if HH is a finite sum of matrices aataa^t with a∈⋀pV∗a\in\bigwedge^p V^* decomposable. It is weakly positive if ztHz≥0z^tHz\geq 0 for every decomposable pp-vector z∈⋀pVz\in\bigwedge^p V.

The zero element is permitted in all three cones. For a smooth form these conditions are imposed at every point. In the definition of strong positivity, the pointwise decomposition need not have smoothly varying summands. This chapter uses the pointwise cone condition; no smooth choice of eigenvectors is needed.

Proposition 7.2.3. A symmetric form is positive if and only if it is a finite sum sp∑νaν∧Jaνs_p\sum_\nu a_\nu\wedge Ja_\nu, without a decomposability restriction. There are inclusions Csp⊂C+p⊂Cwp.C_s^p\subset C_+^p\subset C_w^p. All three cones coincide for p=0,1,n−1,np=0,1,n-1,n.

Proof. The spectral theorem gives H=∑νλνvνvνtH=\sum_\nu\lambda_\nu v_\nu v_\nu^t with λν≥0\lambda_\nu\geq 0 precisely when HH is positive semidefinite. Take aν=λνvνa_\nu=\sqrt{\lambda_\nu}v_\nu. The converse follows from ztaatz=(atz)2z^t aa^tz=(a^tz)^2. The two cone inclusions follow from the same formula.

In the four indicated degrees every exterior vector and covector is decomposable. For degree n−1n-1, a nonzero exterior vector is contraction of a volume vector by a nonzero covector. A basis in which that covector is the last dual basis vector expresses the contraction as a single wedge of n−1n-1 basis vectors. Thus the decomposability restrictions disappear. ◻

Proposition 7.2.4. (Hessian criterion). Let u∈C∞(Ω)u\in C^\infty(\Omega). Then d′d′′u=∑i,j=1n∂2u∂xi∂xjd′xi∧d′′xj.d'd''u=\sum_{i,j=1}^n \frac{\partial^2u}{\partial x_i\partial x_j} d'x_i\wedge d''x_j. This form is positive if and only if the real Hessian of uu is positive semidefinite at every point. If Ω\Omega is convex, this is equivalent to convexity of uu.

Proof. The displayed formula follows from the definitions. A symmetric (1,1)(1,1)-form is represented by a real symmetric matrix. Evaluating along a real direction vv gives the quadratic expression vTHess⁡(u)vv^{\mathsf T}\operatorname{Hess}(u)v. The three positivity cones in this degree agree with the cone of positive semidefinite matrices. Finally, the restriction of uu to the segment x+tvx+tv has second derivative vTHess⁡(u)(x+tv)vv^{\mathsf T}\operatorname{Hess}(u)(x+tv)v; the one-dimensional criterion for convexity proves the last assertion. ◻

Proposition 7.2.5. (Wedge duality). The wedge pairing between symmetric forms in complementary bidegrees is nondegenerate. Under this pairing the positive cone is self-dual across complementary degrees, and the strongly positive and weakly positive cones are dual.

Proof. Let q=n−pq=n-p. For increasing multiindices II of size pp and AA of size qq, let CI,AC_{I,A} be the sign of dxI∧dxAdx_I\wedge dx_A relative to dx1∧⋯∧dxndx_1\wedge\cdots\wedge dx_n when A=IcA=I^c, and zero otherwise. The matrix CC is a signed permutation matrix. Expanding the wedge gives αH∧αK=tr⁡(HCKCt) voln.\begin{equation} \alpha_H\wedge\alpha_K=\operatorname{tr}(HCKC^t)\,\mathrm{vol}_n. \end{equation} Indeed the scalar sign is spsq(−1)pq/sn=1s_ps_q(-1)^{pq}/s_n=1. Since CC is invertible, nondegeneracy follows. Testing against all positive semidefinite KK is equivalent to HH being positive semidefinite: rank-one tests give all inequalities ztHz≥0z^tHz\geq 0.

The complement map CC takes decomposable qq-covectors to decomposable pp-vectors. To see this, choose an orthonormal basis adapted to the span of the covectors in their wedge; its complementary basis gives the image under CC, up to a scalar and sign. Thus testing against strongly positive KK gives exactly weak positivity of HH.

For the reverse duality, the strong cone is closed. Normalize its nonzero generators by tr⁡(aat)=1\operatorname{tr}(aa^t)=1. The resulting set is compact, being the image of the closed set of unit decomposable covectors. Its convex hull is compact in finite dimension. Every nonzero element of the cone is its trace times a point of that convex hull. A convergent sequence in the cone has bounded traces, so its limit remains in the cone. The finite-dimensional separation theorem now identifies the double dual with the cone itself. ◻

7.3 A positive control form

The current estimates will require a fixed form dominating every symmetric coefficient matrix. Define ω=∑id′xi∧d′′xi,βp=ωpp!.\omega=\sum_i d'x_i\wedge d''x_i,\qquad \beta_p=\frac{\omega^p}{p!}.

Lemma 7.3.1. One has βp=αIdm\beta_p=\alpha_{\mathrm{Id}_m}, where m=(np)m=\binom np. In particular βp\beta_p is strongly positive and lies in the interior of the ordinary positive cone.

Proof. The degree-two factors in ω\omega commute, and a repeated factor has zero wedge. Every set of pp distinct indices therefore occurs p!p! times. Moving the double-primed factors to the right contributes sps_p, giving (7.2.1) with the identity matrix. The result is a sum of decomposable squares. If a symmetric matrix EE has operator norm less than one, Idm+E\mathrm{Id}_m+E is positive definite; this proves the interior assertion. ◻

Proposition 7.3.2. (Coefficient domination). Let η=αH\eta=\alpha_H be a smooth symmetric form supported in a compact set K⋐ΩK\Subset\Omega. Choose 0≤χ∈Cc∞(Ω)0\leq\chi\in C_c^\infty(\Omega) with χ=1\chi=1 near KK. If M=sup⁡K∥H(x)∥opM=\sup_K\left\lVert H(x)\right\rVert_{\mathrm{op}}, then Mχβp+η≥0,Mχβp−η≥0.\begin{equation} M\chi\beta_p+\eta\geq0,\qquad M\chi\beta_p-\eta\geq0. \end{equation} Both inequalities mean ordinary positivity.

Proof. On KK, every eigenvalue of H(x)H(x) lies in [−M,M][-M,M], so MIdm±H(x)M\mathrm{Id}_m\pm H(x) is positive semidefinite. Off KK, H=0H=0 and MχIdmM\chi\mathrm{Id}_m is positive semidefinite. This proves the assertions at every point, without choosing eigenvectors smoothly. ◻

Consequently every compactly supported symmetric form is a difference of two compactly supported positive forms. This is the decomposition needed for positive currents. Requiring both summands to be strongly positive changes the statement, and the next section shows that the stronger requirement is impossible in general.

Remark 7.3.3. One may replace the operator norm by a coefficient norm. If ∣HIJ∣≤A|H_{IJ}|\leq A, then ∥H∥op≤mA\left\lVert H\right\rVert_{\mathrm{op}}\leq mA: for ∣v∣=1|v|=1, ∣(Hv)I∣≤A∑J∣vJ∣≤Am|(Hv)_I|\leq A\sum_J|v_J|\leq A\sqrt m, and summing over II gives the bound. Constants in chart estimates may therefore depend on the number of coefficient directions, but not on derivatives of the test form.

7.4 The Plücker obstruction

Take n=4n=4 and p=2p=2. Order the exterior coordinates by 12, 13, 14, 23, 24, 34.12,\ 13,\ 14,\ 23,\ 24,\ 34. Define the symmetric matrix B=(0000010000−100001000010000−10000100000).B=\begin{pmatrix} 0&0&0&0&0&1\\ 0&0&0&0&-1&0\\ 0&0&0&1&0&0\\ 0&0&1&0&0&0\\ 0&-1&0&0&0&0\\ 1&0&0&0&0&0 \end{pmatrix}. Its quadratic form is ztBz=2(z12z34−z13z24+z14z23).\begin{equation} z^tBz=2(z_{12}z_{34}-z_{13}z_{24}+z_{14}z_{23}). \end{equation}

Lemma 7.4.1. The right side vanishes for every decomposable z∈⋀2R4z\in\bigwedge^2\mathbb R^4. Moreover B2=Id6B^2=\mathrm{Id}_6 and tr⁡B=0\operatorname{tr}B=0.

Proof. If z=a∧bz=a\wedge b, then z∧z=a∧b∧a∧b=0z\wedge z=a\wedge b\wedge a\wedge b=0. Expanding in the volume basis gives z∧z=2(z12z34−z13z24+z14z23) e1∧e2∧e3∧e4.z\wedge z =2(z_{12}z_{34}-z_{13}z_{24}+z_{14}z_{23})\,e_1\wedge e_2\wedge e_3\wedge e_4. This proves the relation. The matrix BB exchanges the first and sixth coordinates, exchanges the second and fifth with a minus sign, and exchanges the third and fourth. Applying it twice is the identity, and all its diagonal entries vanish. ◻

Theorem 7.4.2. (Failure of strong-cone spanning). In bidegree (2,2)(2,2) on R4\mathbb R^4, the strongly positive cone does not span the space of symmetric forms. The weakly positive cone contains the nonzero line RαB\mathbb R\alpha_B. In particular it is not pointed.

Proof. The nonzero linear functional ℓ(H)=tr⁡(BH)\ell(H)=\operatorname{tr}(BH) vanishes on every strong generator aataa^t, because tr⁡(Baat)=atBa=0\operatorname{tr}(Baa^t)=a^tBa=0 for decomposable aa. Hence it vanishes on the linear span of the strong cone. But ℓ(B)=tr⁡(B2)=6\ell(B)=\operatorname{tr}(B^2)=6, so that span is proper. The same Plücker relation on decomposable vectors gives zt(tB)z=0z^t(tB)z=0 for every t∈Rt\in\mathbb R. Thus every tαBt\alpha_B is weakly positive. ◻

Example 7.4.3. (A positive form which is not strongly positive). Let a=e12+e34a=e_{12}+e_{34} in exterior coordinate notation and put H=aatH=aa^t. Then HH is positive semidefinite, while tr⁡(BH)=2\operatorname{tr}(BH)=2. Since this functional vanishes on the entire strong span, αH\alpha_H is not even a difference of strongly positive forms.

This example pinpoints the interior error. The form β2\beta_2 is strongly positive, but it cannot lie in the interior of the strong cone in the full symmetric space. Every matrix MId6+HM\mathrm{Id}_6+H in Example 7.4.3 has ℓ(MId6+H)=2\ell(M\mathrm{Id}_6+H)=2, regardless of MM, and therefore is not strongly positive. Enlarging the coefficient of a fixed strong form cannot repair this obstruction.

Remark 7.4.4 (Relation to the sources). This is the obstruction in [BGJK] (Example 2.3.6). It contradicts the real statement of [CLD12] (Lemma 5.2.3), even after symmetry is imposed. That lemma’s proof uses the interior assertion just disproved. The analytic Lemma 5.3.3 cites it through tropical charts; its printed proof therefore loses this input. An analytic counterexample would additionally require a faithful realization of the relevant forms and positivity through the analytic charts, and is not supplied by the real calculation alone.

Similarly, [GK19] (Remark 1.12(i)) asserts that weak positivity of both α\alpha and −α-\alpha implies α=0\alpha=0. The form αB\alpha_B disproves this assertion in its stated real setting. The analytic uses of this remark, including Lemma 2.16, require a separate analysis of their realization and quotient relations. We do not label those analytic statements disproved solely by this example.

7.5 Pullbacks and products

The corrected coefficient estimates use ordinary positivity. Before passing to currents, we record the operations that preserve this cone. These proofs also explain which cone conditions are available when multiplying a current by a smooth form.

Proposition 7.5.1. Pullback by an affine map preserves each of strong, ordinary and weak positivity.

Proof. Let F:W→VF:W\to V have linear part LL. On exterior covectors the pullback is ⋀pL∗\bigwedge^pL^*. It takes a square aataa^t to the square of the pulled-back covector; decomposable covectors stay decomposable or become zero. This proves the strong and ordinary assertions. For a decomposable pp-vector zz on WW, the exterior image (⋀pL)z(\bigwedge^pL)z is decomposable or zero. The quadratic form of F∗αHF^*\alpha_H evaluated at zz is the quadratic form of HH evaluated at that image, proving weak positivity. ◻

Proposition 7.5.2. Products of positive forms are positive. Products of strongly positive forms are strongly positive. The product of a weakly positive form and a strongly positive form is weakly positive. Products in degrees exceeding the dimension are zero.

Proof. For exterior covectors a,ba,b of degrees p,rp,r, respectively, the signs give (spa∧Ja)∧(srb∧Jb)=sp+r(a∧b)∧J(a∧b).(s_p a\wedge Ja)\wedge(s_r b\wedge Jb) =s_{p+r}(a\wedge b)\wedge J(a\wedge b). Indeed moving bb past JaJa gives (−1)pr(-1)^{pr} and spsr(−1)pr=sp+rs_ps_r(-1)^{pr}=s_{p+r}. Decompose the positive matrices into squares and use this identity term by term. If both factors are strong, their generating covectors are decomposable, so their wedge is decomposable or zero.

For the last assertion it suffices to consider one decomposable square srb∧Jbs_r b\wedge Jb. Evaluate the product on a decomposable (p+r)(p+r)-vector and restrict to the subspace it spans. If bb restricts to zero there is nothing to prove. Otherwise choose a basis of this subspace so that bb is a scalar multiple of the wedge of the last rr dual basis vectors. The value of the product is a nonnegative square factor times the value of the first form on the wedge of the first pp basis vectors. Weak positivity makes the latter value nonnegative. ◻

For a smooth function, Proposition 7.1.4 identifies the matrix of d′d′′f\mathrm d'\mathrm d''f with its Hessian. Thus positivity, strong positivity and weak positivity of d′d′′f\mathrm d'\mathrm d''f are equivalent. On a convex open set they are equivalent to convexity of ff: along every line the second derivative is nonnegative, which is the one-variable criterion for convexity; conversely, second differences of a smooth convex function give a nonnegative Hessian. This special agreement in degree (1,1)(1,1) must not be extrapolated to all degrees.

7.6 A homotopy formula for the real double complex

The two exterior factors of a superform have different roles when one studies only one differential. This gives an elementary Poincaré lemma on an ordinary real open set, but also explains why a statement about the double complex requires more than two separate exactness statements.

Let Ω⊂Rn\Omega\subset\mathbb R^n be open and star-shaped about 00. Write every superform uniquely as α=∑∣J∣=qαJ∧d′′xJ,αJ∈Ωp(Ω),\alpha=\sum_{|J|=q}\alpha_J\wedge d''x_J, \qquad \alpha_J\in\Omega^p(\Omega), where the first factor is an ordinary differential form in the d′xid'x_i. In this description, d′d' is the ordinary exterior derivative on each αJ\alpha_J, while the d′′xJd''x_J form a fixed coefficient space.

Definition 7.6.1. For p>0p>0, define H′:Ap,q(Ω)→Ap−1,q(Ω)H':A^{p,q}(\Omega)\to A^{p-1,q}(\Omega) coefficientwise by (H′α)x(v1,…,vp−1;w1,…,wq)=∫01tp−1αtx(x,v1,…,vp−1;w1,…,wq) dt.(H'\alpha)_x(v_1,\ldots,v_{p-1};w_1,\ldots,w_q) =\int_0^1 t^{p-1} \alpha_{tx}(x,v_1,\ldots,v_{p-1};w_1,\ldots,w_q)\,dt. Set H′=0H'=0 for p=0p=0. Define E′α=0E'\alpha=0 for p>0p>0; for p=0p=0, let E′E' replace each scalar coefficient of α\alpha by its value at 00, retaining the fixed d′′d''-basis.

The semicolon separates the primed and double-primed alternating slots. Only the primed slots are scaled in this formula. In particular, E′E' is not the full geometric pullback by the constant map: that pullback would also kill positive double-primed degree.

Proposition 7.6.2. (Radial homotopy). The operators satisfy d′H′+H′d′=id⁡−E′.d'H'+H'd'=\operatorname{id}-E'. Consequently the d′d'-cohomology on Ω\Omega vanishes in positive primed degree, while in primed degree zero it is the space of constant double-primed forms ⋀q(Rn)∗\bigwedge^q(\mathbb R^n)^*.

Proof. It suffices to prove the usual homotopy identity for each αJ\alpha_J. Let F:[0,1]×Ω→ΩF:[0,1]\times\Omega\to\Omega be F(t,x)=txF(t,x)=tx. Decompose the ordinary pullback of a primed form as F∗αJ=dt∧βt+γtF^*\alpha_J=dt\wedge\beta_t+\gamma_t, where γt\gamma_t contains no dtdt. The coefficient of dtdt in the identity d(F∗αJ)=F∗(dαJ)d(F^*\alpha_J)=F^*(d\alpha_J) is ∂tγt−dxβt=the dt-coefficient of F∗(dαJ).\partial_t\gamma_t-d_x\beta_t =\text{the }dt\text{-coefficient of }F^*(d\alpha_J). Integrate from 00 to 11. The boundary term is αJ−F0∗αJ\alpha_J-F_0^*\alpha_J, and ∫01βt dt\int_0^1\beta_t\,dt is precisely the coefficientwise operator in Definition 7.6.1. This proves the formula. Smoothness at t=0t=0 follows directly from the displayed pullback, or from the factor tp−1t^{p-1} when p>0p>0.

If p>0p>0 and d′α=0d'\alpha=0, the identity gives α=d′H′α\alpha=d'H'\alpha. For p=0p=0, closedness says that every coefficient has zero differential. The star-shaped open set is connected, so those coefficients are constant. There are no forms of primed degree −1-1, giving the asserted degree-zero cohomology. ◻

There is an analogous statement for d′′d'', obtained by exchanging the exterior factors with the fixed sign convention. The two statements concern separate complexes. The operator H′H' was not constructed to commute with d′′d'', and no such commutation is needed for Proposition 7.6.2. This point matters when one tries to obtain a d′d′′d'd''-potential by applying two homotopies successively.

7.7 The symmetric Hessian potential theorem

In bidegree (1,1)(1,1) the extra compatibility has a familiar form: the coefficient matrix must be symmetric. On a star-shaped open set this condition, together with closedness, suffices for a scalar potential.

Theorem 7.7.1. Let Ω⊂Rn\Omega\subset\mathbb R^n be open and star-shaped, and let α=∑i,j=1nAij(x)d′xi∧d′′xj\alpha=\sum_{i,j=1}^n A_{ij}(x)d'x_i\wedge d''x_j have smooth coefficients with Aij=AjiA_{ij}=A_{ji}. If d′α=0d'\alpha=0, then there is a smooth function uu with α=d′d′′u\alpha=d'd''u. The potential is unique up to addition of an affine function. If Ω\Omega is convex, α\alpha is positive if and only if uu is convex.

Proof. For each jj, the primed one-form αj=∑iAijd′xi\alpha_j=\sum_iA_{ij}d'x_i is closed. Proposition 7.6.2, after translating the star centre to zero, gives a smooth function bjb_j with ∂ibj=Aij\partial_i b_j=A_{ij}. Symmetry implies ∂ibj=∂jbi\partial_i b_j=\partial_j b_i. Hence the ordinary one-form ∑jbjdxj\sum_jb_jdx_j is closed. Apply the same homotopy formula once more to obtain uu with ∂ju=bj\partial_j u=b_j. Its Hessian is AA, proving the asserted identity.

If uu and vv are two potentials, every first derivative of u−vu-v has zero differential and is therefore constant on the connected set Ω\Omega. Thus u−vu-v is affine. Finally, positivity of a symmetric (1,1)(1,1)-superform is positive semidefiniteness of its matrix. On a convex domain, the second derivative of the restriction of uu to every line segment is ξTAξ\xi^{\mathsf T}A\xi. Nonnegativity of these second derivatives is equivalent to convexity. ◻

The same calculation shows that symmetry and d′d'-closedness imply d′′d''-closedness here: the constructed Hessian is closed under both differentials. It also identifies the ambiguity more precisely than saying only that it is harmonic. In ordinary real coordinates the smooth Hessian kernel consists of affine functions; on an analytic space, current-harmonic functions can form a much larger class.

Example 7.7.2. (Double closedness without a scalar potential). On any open ball in R2\mathbb R^2, the constant form α=d′x1∧d′′x2−d′x2∧d′′x1\alpha=d'x_1\wedge d''x_2-d'x_2\wedge d''x_1 is both d′d'-closed and d′′d''-closed. It cannot equal d′d′′ud'd''u for a smooth scalar function, because that would give ∂1∂2u=1\partial_1\partial_2u=1 and ∂2∂1u=−1\partial_2\partial_1u=-1. It cannot have a scalar distributional potential either, since distributional partial derivatives commute. Thus two separate Poincaré lemmas do not imply an unrestricted local d′d′′d'd''-lemma for this real double complex.

The passage to a polyhedral or analytic space

The preceding results are on an open subset of one real vector space. A form on a polyhedral support is a quotient of ambient forms by tangential relations, and the tangent directions can change across faces. Before using a homotopy on that quotient, one must prove that it preserves those relations and has the required support and domain properties. An ambient homotopy formula alone does not do this.

On a Berkovich space there is a further compatibility under refinements of moment charts. A potential constructed on one tropical presentation must be independent of the presentation or be shown to glue with the permitted transition functions. These are precisely the additional issues in local-potential theorems for currents and in comparisons of tropical and analytic cohomology. The explicit Euclidean theorem is a local building block, not a theorem covering every analytic chart quotient.

7.8 A radial homotopy on a polyhedral star

The ordinary de Rham homotopy contracts all covectors of a differential form. For the d′′d'' complex of superforms, the primed degree must remain fixed. The correct operator contracts only the double-primed variables. This produces a local Poincaré lemma and identifies its degree-zero kernel with the multi-tangent coefficient space.

Let C⊂RrC\subset\mathbb R^r be the support of a finite fan, and let BB be an open ball centred at its vertex 00. Superforms on C∩BC\cap B are restrictions of smooth ambient superforms, with two forms identified when their evaluations on tangent vectors of every cone agree. It is enough to work with germs at 00, so a common ambient representative is available. Write a representative as α=∑∣I∣=pd′xI∧ωI,ωI=∑∣J∣=qaIJ(x) d′′xJ.\alpha=\sum_{|I|=p}d'x_I\wedge\omega_I, \qquad \omega_I=\sum_{|J|=q}a_{IJ}(x)\,d''x_J. With the ordered-wedge convention of this book, d′′d'' acts on each ωI\omega_I as (−1)p(-1)^p times the ordinary exterior derivative. This sign is essential in the homotopy formula.

For an ordinary qq-form ω\omega with q>0q>0, set (Hω)x(v1,…,vq−1)=∫01tq−1ωtx(x,v1,…,vq−1) dt.(H\omega)_x(v_1,\ldots,v_{q-1}) =\int_0^1 t^{q-1}\omega_{tx}(x,v_1,\ldots,v_{q-1})\,dt. For q=0q=0 set H=0H=0, and define Kpα=(−1)p∑∣I∣=pd′xI∧HωI.K_p\alpha=(-1)^p\sum_{|I|=p}d'x_I\wedge H\omega_I. The operator KpK_p has bidegree (0,−1)(0,-1). Define EαE\alpha to be zero for q>0q>0, and for q=0q=0 define it to be the constant-coefficient primed form ∑IaI(0)d′xI\sum_Ia_I(0)d'x_I, restricted to the fan.

Proposition 7.8.1. The operators descend to superforms on the fan, and satisfy d′′Kp+Kpd′′=id⁡−E.d''K_p+K_pd''=\operatorname{id}-E. Consequently every d′′d''-closed germ of bidegree (p,q)(p,q) with q>0q>0 is d′′d''-exact.

Proof. Suppose an ambient representative vanishes on tangent vectors of every cone. For xx in a cone σ\sigma, all the points txtx with t>0t>0 remain in σ\sigma, and x∈Lσx\in L_\sigma. Evaluating KpαK_p\alpha on primed and double-primed vectors in LσL_\sigma therefore evaluates αtx\alpha_{tx} only on vectors in LσL_\sigma, including the inserted vector xx. The integrand is zero. Its value at t=0t=0 follows by continuity and does not affect the integral. Thus KpK_p preserves the relation defining forms on CC.

For q=0q=0, if an ambient primed form vanishes on each cone, evaluate along an interior ray of a cone and let its parameter tend to zero. Its constant value at zero annihilates ⋀pLσ\bigwedge^pL_\sigma for every σ\sigma, so EE also preserves the relation. For q>0q>0 there is nothing to check for EE.

On ambient ordinary forms, differentiation under the integral gives dH+Hd=id⁡−e0∗dH+Hd=\operatorname{id}-e_0^*, where e0∗e_0^* keeps only degree-zero values at the origin. One can check this on a coefficient aJ(x)a_J(x): the surviving integrand is the derivative in tt of tqaJ(tx)t^q a_J(tx), with the evaluation on the appropriate vectors. The endpoint at t=1t=1 gives ω\omega; the endpoint at t=0t=0 is zero for q>0q>0 and is ω(0)\omega(0) for q=0q=0. The two factors (−1)p(-1)^p from d′′d'' and KpK_p cancel. Applying this identity to every ωI\omega_I proves the displayed formula on ambient forms, hence on the quotient. If d′′α=0d''\alpha=0 and q>0q>0, it gives α=d′′Kpα\alpha=d''K_p\alpha. ◻

The kernel is a multi-tangent dual

Put Fp(0)=∑σ∈C⋀pLσ ⊂ ⋀pRr,Fp(0)=Hom⁡R(Fp(0),R).F_p(0)=\sum_{\sigma\in C}\bigwedge^p L_\sigma \ \subset\ \bigwedge^p\mathbb R^r, \qquad F^p(0)=\operatorname{Hom}_\mathbb R(F_p(0),\mathbb R).

Corollary 7.8.2. The space of d′′d''-closed (p,0)(p,0) germs at the vertex is naturally Fp(0)F^p(0). The local sequence 0⟶Fp⟶Ap,0→d′′Ap,1→d′′Ap,2⟶⋯0\longrightarrow F^p\longrightarrow A^{p,0} \xrightarrow{d''}A^{p,1}\xrightarrow{d''}A^{p,2}\longrightarrow\cdots is exact on finite polyhedral charts, with the natural face restrictions on FpF^p.

Proof. For a closed (p,0)(p,0) form, Proposition 7.8.1 gives α=Eα\alpha=E\alpha on a sufficiently small star. Thus it is represented by a constant ambient pp-covector. Such a covector represents zero exactly when it annihilates every ⋀pLσ\bigwedge^pL_\sigma, equivalently their sum Fp(0)F_p(0). The quotient of ⋀p(Rr)∗\bigwedge^p(\mathbb R^r)^* by this annihilator is Fp(0)∗F_p(0)^*, since every linear functional on a subspace of a finite-dimensional vector space extends to the whole space. Exactness in positive double-prime degree is the proposition. Around a point in any face, translate the point to the origin and use the local star. Restriction to a smaller face star induces the indicated restrictions on the coefficient spaces. ◻

The sum in FpF_p must be formed after taking exterior powers. Replacing it by ⋀p(∑σLσ)\bigwedge^p(\sum_\sigma L_\sigma) changes the coefficient sheaf in general. The two-plane example in Section 24.7 makes the difference explicit. The homotopy proof identifies the correct space directly from the vanishing relation for actual forms.

Remark 7.8.3. If an integral linear map sends one fan to another, the homotopy commutes with its ordinary superform pullback: linear maps commute with radial scaling, and contraction by xx corresponds to contraction by its image. Translation gives the same assertion for affine charts with specified centres. This compatibility is enough for the local polyhedral argument. A global homotopy on a Berkovich space would additionally require compatible analytic domains and centres; it is not supplied by this computation.

From a local resolution to global cohomology

On a paracompact polyhedral space with finite-part charts, the sheaves Ap,qA^{p,q} admit smooth partitions of unity. They are therefore fine. The exact sequence above is a fine resolution of FpF^p, so its global cohomology computes Hq(X,Fp)H^q(X,F^p). Here we use the standard fine-resolution theorem of sheaf cohomology; its mechanism can be seen in a Čech double complex.

For a locally finite cover (Ui)(U_i) and subordinate partition (ρi)(\rho_i), a Čech cochain cc with values in Ap,qA^{p,q} has a homotopy (hc)i0…ik−1=∑jρjcji0…ik−1,δh+hδ=id⁡(k>0).(hc)_{i_0\ldots i_{k-1}}=\sum_j\rho_j c_{j i_0\ldots i_{k-1}}, \qquad \delta h+h\delta=\operatorname{id}\quad(k>0). The sum is locally finite; extension by zero is valid because the multiplier is supported in UjU_j. Thus the rows have no positive Čech cohomology. In the other direction, local exactness of the d′′d'' complex leaves only FpF^p. Passing to sufficiently fine covers, the two computations of the total complex give the desired comparison.

This proves the finite-part local mechanism behind the tropical Dolbeault comparison. Extended tropical spaces also have sedentary boundary strata. Their form definition includes compatibility with projections to those strata, and their local Poincaré lemma requires that additional structure. The full comparison, including compact supports and extended charts, is the theorem of Jell–Shaw–Smacka stated in Chapter 24. The present radial proof does not remove the boundary hypotheses from that theorem.

References. [JSS] (Sections 2–3); the explicit homotopy above is the radial de Rham argument applied only to the double-primed variables, with the quotient and sign checks included.