Abstract
The Alesker–Bernig–Schuster theorem asserts that each irreducible representation of the special orthogonal group appears with multiplicity at most one as a subrepresentation of the space of continuous translation-invariant valuations with fixed degree of homogeneity. Moreover, the theorem describes in terms of highest weights which irreducible representations appear with multiplicity one. In this paper, we present a refinement of this result, namely the explicit construction of a highest weight vector in each irreducible subrepresentation. We then describe how important natural operations on valuations (pullback, pushforward, Fourier transform, Lefschetz operator, Alesker–Poincaré pairing) act on these highest weight vectors. We use this information to prove the Hodge–Riemann relations for valuations in the case of Euclidean balls as reference bodies. Since special cases of the Hodge–Riemann relations have recently been used to prove new geometric inequalities for convex bodies, our work immediately extends the scope of these inequalities.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
A valuation is a finitely additive function defined on a fixed class \({{\mathcal {S}}}\) of geometric sets, i.e., a function \(\phi :{\mathcal {S}}\rightarrow \mathbb {C}\) satisfying
whenever \(A,B,A\cap B, A \cup B\in {\mathcal {S}}\). With its origins in the solution of Hilbert’s third problem by Dehn and in Pick’s theorem from plane geometry, the concept has—as a generalization of measure—long played an important role in convex geometry, see, e.g., the monographs by Gruber [Gru07], Hadwiger [Had57], and Schneider [Sch14]. The (mixed) volume, the lattice point enumerator, and the Euler characteristic are fundamental examples of valuations, defined, respectively, on the classes of convex bodies, lattice polytopes, and finite unions of convex bodies.
With very few exceptions, the majority of the classical results on valuations are closely tied to the dissection theory of polytopes. Some of the most striking results in this direction are the proof of the g-theorem, which characterizes the possible numbers of faces of simple polytopes, and the proof of the Alexandrov–Fenchel inequality by McMullen [Mcm89, Mcm93]. In contrast, results on continuous valuations on convex bodies that are not directly implied by the corresponding discrete statements were scarce.
This situation changed completely in the early 2000s with the groundbreaking work of Alesker [Ale01, Ale03, Ale04, Ale11]. The most important technical innovation introduced by Alesker was the discovery of a natural dense subspace \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\subset {{\,\textrm{Val}\,}}(\mathbb {R}^n)\) of smooth valuations, see Sect. 3 for precise definitions. Apart from much better analytical properties, the salient feature of the subspace of smooth valuations is the existence of a natural multiplicative structure, namely the Alesker product, which turns this space into a graded commutative algebra. The Alesker product and related algebraic structures had a profound impact on integral geometry, see, e.g., [AB21, AF14, BF11, BFS14, Fu06, Fai17, Fai19].
The starting point for this rapid development of valuation theory was Alesker’s proof of a much stronger and more useful form of McMullen’s conjecture [Ale01] known today as the irreducibility theorem. To describe this result, let us first point out that by a classical result of McMullen [Mcm77] the space of continuous and translation-invariant valuations is naturally graded by the degree
and the parity of a valuation
Alesker’s irreducibility theorem is the statement that the natural action of the general linear group on each graded component is irreducible. This is a very strong property with far-reaching consequences. For example, it implies that smooth valuations can be represented by smooth differential forms.
A related question of great importance is how the space \({{\,\textrm{Val}\,}}(\mathbb {R}^n)\) decomposes under the action of the special orthogonal group \({{\,\textrm{SO}\,}}(n)\). By the Peter–Weyl theorem from abstract harmonic analysis, each representation E of the compact group \({{\,\textrm{SO}\,}}(n)\) decomposes into isotypic components \(\bigoplus _{\lambda } E_\lambda \), where the sum is over all equivalence classes of irreducible representations of \({{\,\textrm{SO}\,}}(n)\). The question thus becomes which \(\lambda \) actually appear in the isotypic decomposition of \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) and with what multiplicities. A complete answer to this question is given by the Alesker–Bernig–Schuster decomposition theorem [ABS11].
Here and in the following, the equivalence classes of irreducible representations of \({{\,\textrm{SO}\,}}(n)\) are identified with their highest weights, see §2.2 below.
Theorem 1.1
(Alesker–Bernig–Schuster). Let \( l=\lfloor \frac{n}{2}\rfloor \) and \( 0\le r \le l \). The non-trivial \({{\,\textrm{SO}\,}}(n)\)-types in \({{\,\textrm{Val}\,}}_{r}(\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_{n-r}(\mathbb {R}^n)\) are the same and given by the following set of highest weights:
Moreover, each appears with multiplicity one.
Theorem 1.1 is a useful tool for many problems in valuation theory and integral geometry. Its typical usage is to reduce a question concerning the existence of valuations with certain properties to a problem in representation theory which can be solved. See [ABS11, BDS21, Wan20] for a small sample of such applications.
Drawing a comparison with the decomposition of functions on the unit sphere into spherical harmonics, it is obvious that the information provided by Theorem 1.1, however useful, is only very limited. For many applications of the theory of spherical harmonics it is of crucial importance to have a manageable description of at least one spherical harmonic in each degree, namely the zonal harmonic, see, e.g., [Gro96]. One of the principal contributions of this paper is to provide something just as useful, namely a simple explicit description of a highest weight vector in each \({{\,\textrm{SO}\,}}(n)\)-type of \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\). A different approach to making the Alesker–Bernig–Schuster decomposition theorem more explicit was recently explored by Saienko [Sai19].
A problem of particular interest where the information provided by Theorem 1.1 is insufficient—and our main motivation to undertake this work—is to prove the Hodge–Riemann relations for valuations with Euclidean balls as reference bodies. Denote by \(*:{{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\times {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) the convolution of valuations, which is in a precise sense dual to the Alesker product (see Sects. 3.3.2 and 3.4.1), and let us first formulate the hard Lefschetz theorem and the Hodge–Riemann relations for general reference bodies.
Let \({\mathcal {K}}(\mathbb {R}^n)\) denote the space of convex bodies in \(\mathbb {R}^n\) and let \(V(K_1,\ldots , K_n)\) be the mixed volume of \(K_1,\dots ,K_n\in {\mathcal {K}}(\mathbb {R}^n)\).
Conjecture 1.2
Let \(0\le r\le \lfloor \frac{n}{2}\rfloor \). Consider \(C_0,\dots ,C_{n-2r}\in {\mathcal {K}}(\mathbb {R}^n)\) with smooth and strictly positively curved boundary and denote
for \(i=0,\dots ,n-2r\). Then the following properties hold:
-
(a)
Hard Lefschetz theorem. The map \({{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\) given by
$$\begin{aligned} \phi \mapsto \phi *{{{\mu _{C}}_1}}*\cdots *{{{\mu _{C}}_{n-2r}}} \end{aligned}$$is an isomorphism of topological vector spaces.
-
(b)
Hodge–Riemann relations. The sesquilinear form
$$\begin{aligned} (\phi ,\psi )\mapsto (-1)^r\,\phi *\overline{\psi }*{{{\mu _{C}}_1}}*\cdots *{{{\mu _{C}}_{n-2r}}} \end{aligned}$$is positive definite on
$$\begin{aligned} \left\{ \phi \in {{\,\textrm{Val}\,}}^\infty _{n-r}(\mathbb {R}^n)\mid \phi *{{{\mu _{C}}_0}}*\cdots *{{{\mu _{C}}_{n-2r}}}=0\right\} . \end{aligned}$$
Conjecture 1.2 has first been published only last year by the first-named author [Kot21], but it was certainly considered before as both statements are formally analogous to theorems in Kähler geometry. Moreover, they can be viewed as the continuous analogues of two theorems of McMullen [Mcm93] on the polytope algebra. However, in contrast to these cases, the algebra \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) is infinite-dimensional which makes numerous aspects of the problem somewhat more delicate.
Still, several instances of the conjecture are known to be true. The hard Lefschetz theorem was proved for Euclidean balls as reference bodies first by Alesker [Ale03] for even valuations and subsequently by Bernig and Bröcker [BB07] without this restriction. The authors [KW22] have recently established the hard Lefschetz theorem for \(r=1\). The Hodge–Riemann relations were proved for Euclidean balls as reference bodies by the first-named author [Kot21] for even valuations and for \(r=1\). The authors [KW22] have also recently established the Hodge–Riemann relations for \(r=1\).
The Hodge–Riemann relations imply both classical and new geometric inequalities for convex bodies. For first steps in this direction see [Ale22, KW22, Kot21] where in particular the Alexandrov–Fenchel inequality together with a novel improvement for lower-dimensional convex bodies have been deduced from the Hodge–Riemann relations.
1.1 Main results.
Our contributions in this work are threefold. First, we will refine the Alesker–Bernig–Schuster theorem by explicitly constructing a non-trivial highest weight vector for each \({{\,\textrm{SO}\,}}(n)\)-type in \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\). Second, we will describe the action of a number of important natural operations on valuations, namely pullback, pushforward, Fourier transform, Lefschetz operator, and Alesker–Poincaré pairing, on these highest weight vectors. Third, we will accomplish our original goal, proving the Hodge–Riemann relations (Conjecture 1.2(b)) for Euclidean balls as reference bodies.
Let us begin by outlining the construction of the highest weight vectors. Recall that in each irreducible representation V of \({{\,\textrm{SO}\,}}(n)\) there exists a (unique up to the choice of certain Lie-theoretic data) one-dimensional subspace \(V_\lambda \subset V\) corresponding to the highest weight of V. Since according to the Alesker–Bernig–Schuster theorem the decomposition into \({{\,\textrm{SO}\,}}(n)\)-types is multiplicity-free, we have to construct precisely one highest weight vector for each \({{\,\textrm{SO}\,}}(n)\)-type appearing in the isotypic decomposition of \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\).
Throughout the paper, we will work in Euclidean space of dimension \(n\ge 2\) and we will denote \(l=\left\lfloor \frac{n}{2}\right\rfloor \). For the sake of clarity, let us define
With this notation, the Alesker–Bernig–Schuster theorem asserts that the set of highest weights appearing in \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) is
Let \(v_n\) be the volume of the n-dimensional Euclidean unit ball \(D^n\) and \(s_n\) the volume of the n-dimensional Euclidean unit sphere \(S^n\). We write \(S\mathbb {R}^n=\mathbb {R}^n\times S^{n-1}\) for the sphere bundle of \(\mathbb {R}^n\) and denote by N(K) the normal cycle of a convex body \(K\in {\mathcal {K}}(\mathbb {R}^n)\). As a set, the normal cycle is the disjoint union of the outward unit normals to K. By \(\mathbb {N}\) we denote the set of positive integers.
Theorem 1.3
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\), let \(\omega _{r,k,m}\in \Omega ^{n-1}(S\mathbb {R}^n)\) be given by formula (9) below. The smooth valuation
is a non-trivial highest weight vector of weight \(\lambda _{k,m}\) of the \({{\,\textrm{SO}\,}}(n)\)-representation \({{\,\textrm{Val}\,}}_{r}(\mathbb {R}^n)\).
Observe that there are highest weights appearing in \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) that are formally not covered by Theorem 1.3, namely the trivial highest weight and \(\lambda _{-l,m}\). However, applying to \(\phi _{l,l,m}\) the reflection in the coordinate hyperplane \(e_n^\perp \) yields the highest weight vectors for the latter. As for the former, the valuations corresponding to the trivial highest weight are precisely the intrinsic volumes, which have been intensively studied in convex geometry. Since they do not fit naturally into the family of differential forms \(\omega _{r,k,m}\) we introduce and since many much simpler and geometrically more appealing descriptions of the intrinsic volumes are known (see, e.g., [Sch14]), we have preferred to omit them from Theorem 1.3.
A brilliant insight behind much of the recent progress in integral geometry is the observation of Fu [Fu06] that problems in the spirit of the kinematic formula of Blaschke–Chern–Federer [Bla55, Che52, Fed59] can be transformed into the problem of evaluating the product, the convolution, and the Fourier transform of invariant valuations. More recently, it was discovered that also problems concerning geometric inequalities for convex bodies can be reformulated in the algebraic language of valuation [Ale22, KW22, Kot21]. It should therefore come as no surprise that computing the algebraic structures explicitly is typically a challenge. In this connection, the main point of Theorem 1.3 is that it provides an expression for the highest weight vectors that is simple enough to allow the evaluation of most of the natural operations on valuations.
Since in the following theorem we consider valuations both on \({\mathcal {K}}(\mathbb {R}^n)\) and on \({\mathcal {K}}(\mathbb {R}^{n-1})\), we write \(\phi ^{(n)}_{r,k,m}\) instead of \(\phi _{r,k,m}\) to reduce the risk of confusion.
Theorem 1.4
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\), the highest weight vectors constructed in Theorem 1.3 satisfy the following properties:
-
(a)
Alesker–Poincaré pairing.
$$\begin{aligned} \overline{\phi _{r,k,m}^{(n)}}*\phi _{n-r,k,m}^{(n)}= (-1)^k \left( {\begin{array}{l} {n - 2k} \\ {r - k} \\ \end{array} } \right) \frac{(m+k-1)(n+m-k)v_{n+2m-2}}{v_{n+m-r-2}v_{r+m-2}s_{2m-3}}. \end{aligned}$$ -
(b)
Pullback along the inclusion \(\iota :\mathbb {R}^{n-1}\rightarrow \mathbb {R}^{n-1}\oplus \mathbb {R}\).
$$\begin{aligned} \iota ^*\phi _{r,k,m}^{(n)}= {\left\{ \begin{array}{ll} \phi _{r,k,m}^{(n-1)}&{}{}\text{ if } r<n-1 \text{ and } k<n-r, \\ \frac{1}{2}\phi _{k,k-1,m}^{(n-1)} &{}{}\text{ if } k=r=\frac{n}{2}, \\ 0&{}{}\text{ otherwise }. \end{array}\right. } \end{aligned}$$ -
(c)
Pushforward along the projection \(\pi :\mathbb {R}^{n-1}\oplus \mathbb {R}\rightarrow \mathbb {R}^{n-1}\).
$$\begin{aligned} \pi _* \phi _{r,k,m}^{(n)}= {\left\{ \begin{array}{ll} \phi _{r-1,k,m}^{(n-1)}&{}\text {if }k<r,\\ -\frac{1}{2}\phi _{k-1,k-1,m}^{(n-1)}&{}\text {if }k=r=\frac{n}{2},\\ 0&{}\text {otherwise}. \end{array}\right. } \end{aligned}$$ -
(d)
Fourier transform.
$$\begin{aligned} \mathbb {F}\phi _{r,k,m}^{(n)}=(-1)^{k-1}(\sqrt{-1})^m\phi _{n-r,k,m}^{(n)}. \end{aligned}$$ -
(e)
Lefschetz operator.
$$\begin{aligned} \Lambda \phi _{r,k,m}^{(n)}= {\left\{ \begin{array}{ll} (n-r-k+1)\frac{v_{n+m-r-1}}{v_{n+m-r-2}}\phi _{r-1,k,m}^{(n)} &{} \text {if }k<r,\\ 0 &{} \text {if }k=r. \end{array}\right. } \end{aligned}$$
Several special cases and consequences of Theorem 1.4 have appeared—sometimes in different guises—in the literature. Regarding the Fourier transform, the special cases \(k=1\) and \(k=l\) for even m have been treated by Bernig and Hug [BH18, Theorem 1] and Bernig and Solanes [BS17, Corollary 2.5], respectively. Moreover, in [BS17, Proposition 2.8] the action of \(\Lambda ^{n-2r}\circ \mathbb {F}\) on highest weight vectors was determined for even m and n. For \(k=1\), our result on the Alesker–Poincaré pairing is equivalent to [BH18, Proposition 4.11]. [GHW17, Theorem 3.1] by Goodey, Hug, and Weil is equivalent to the action of \(\mathbb {F}\circ \Lambda \circ \mathbb {F}\) on highest weight vectors with \(k=1\).
We expect that Theorems 1.3 and 1.4 will open up new avenues in convex and integral geometry. In this paper we present two applications. First, we give a new proof of the hard Lefschetz theorem for Euclidean balls as reference bodies. As discussed above, in this case the hard Lefschetz theorem is already known to be true; the point here is that Theorem 1.4 yields a short and transparent proof. Second, and here our results are indeed indispensable, we deduce from Theorem 1.4 the Hodge–Riemann relations for Euclidean balls as reference bodies. Altogether,
Theorem 1.5
Conjecture 1.2 is true if \(C_0=\cdots =C_{n-2r}=D^n\).
As pointed out to us by R. van Handel, combining Theorem 1.5 with the case \(r=1\) of the Hodge–Riemann relations recently proved by the authors [KW22] implies the validity of Conjecture 1.2 in yet another special case.
Corollary 1.6
For arbitrary \(C_0\), Conjecture 1.2 is true if \(r=2\) and \(C_1=\cdots = C_{n-4}=D^n\). In particular, Conjecture 1.2 holds for \(n\le 4\).
Let us point out that Theorem 1.5 and Corollary 1.6 immediately extend the scope of the geometric inequalities that Alesker [Ale22] has recently deduced from the Hodge–Riemann relations thus providing in particular a continuous analogue to previously established combinatorial inequalities [Mcm93, Tim99, Han21].
1.2 Further discussion.
The differential forms \(\omega _{r,k,m}\) of Theorem 1.3 are defined as certain determinantal expressions in the differentials of complexified coordinate functions on \(\mathbb {R}^n\times \mathbb {R}^{n}\supset S\mathbb {R}^n\). Apart from finding them, a major challenge we faced at the beginning of this work was to manipulate these bulky objects without too much pain. The notion of double form—a concept developed by de Rham [Rha84] and a key tool in the work of Gray [Gra84, Gra04] in integral geometry—turned out to be perfectly suited for our purposes. Double forms allow us to split up the proofs of the main identities cleanly into a series of simpler auxiliary statements.
Another point that we believe deserves mention is that the proof of the special case of Theorem 1.4(d) corresponding to \(k=1\), previously established by Bernig and Hug [BH18], rested on the idea of using the action of general linear group to move between different \({{\,\textrm{SO}\,}}(n)\)-types. Much to our surprise, our proof of the general case is shorter, using only the fact that the Fourier transform is compatible with the action of the maximal compact subgroup \({{\,\textrm{O}\,}}(n)\subset {{\,\textrm{GL}\,}}(n)\) and intertwines the operations of pullback and pushforward.
2 Representation theory
2.1 Abstract harmonic analysis.
Let us begin by collecting general information about smooth vectors, the decomposition into K-types, and the convergence of Fourier series. Our reference for this material are books by Warner [War72] and Knapp [Kna01] and an article by Sugiura [Sug71].
Let G be a Lie group and \(\rho \) be a continuous representation of G on a Fréchet space E. An element \(v\in E\) is called a smooth vector if the map \(G\rightarrow E\), \(g\mapsto \rho (g)v\) is smooth. The invariant subspace of smooth vectors is denoted \( E^\infty \) and is equipped with a natural Fréchet space topology. Moreover, \((E^\infty )^\infty = E^\infty \).
Let \(K\subset G\) be a compact subgroup with Haar probability measure dk. Denote by \({\widehat{K}}\) the set of all K-types, i.e., equivalence classes of irreducible representations of K. Let further \(\chi _\lambda \) denote the character and \(d(\lambda )\) the (finite) dimension of \(\lambda \in {\widehat{K}}\). Then
is a continuous projection of E onto the \(\lambda \)-isotypic component \(E_\lambda \subset E\), i.e., the closure of the span of all irreducible subrepresentations of E of type \(\lambda \). Smooth vectors satisfy the following important property.
Theorem 2.1
(Harish-Chandra). Let \(v\in E^\infty \). Then the Fourier series
convergences absolutely to v.
We will need more precise information on the decay of the Fourier coefficients. To this end, let us assume K is connected in which case \({\widehat{K}}\) is in one-to-one correspondence with the set of highest weights. Given a maximal torus \(T\subset K\) with Lie algebra \(\mathfrak {t}\) and a system of positive roots \(\Delta ^+\subset \mathfrak {t}^*_\mathbb {C}\), the complexification of the Lie algebra \(\mathfrak {k}\) of K decomposes as
where \(\mathfrak {n}^\pm =\bigoplus _{\alpha \in \Delta ^+}\mathfrak {k}_{\pm \alpha }\) is the direct sum of positive (negative) root spaces. Let \((V,\sigma )\in \lambda \) be an irreducible representation of K. There exist \(\lambda \in \mathfrak {t}_\mathbb {C}^*\), called the highest weight, and a one-dimensional subspace \(V_\lambda \subset V\), both unique, such that
holds for any \(H\in \mathfrak {t}_\mathbb {C}\), \(X\in \mathfrak {n}^+\), and \(v\in V_\lambda \). In fact, the highest weight characterizes the corresponding K-type uniquely, which justifies the abuse of notation. From now on these two meanings of \(\lambda \) will be used interchangeably. The elements of \(V_\lambda \) are referred to as highest weight vectors.
Fix an \({{\,\textrm{Ad}\,}}(K)\)-invariant complex bilinear form on \(\mathfrak {k}_\mathbb {C}\) such that its restriction to \(\sqrt{-1}\mathfrak {t}\) is real and positive definite. For example, if \(K\subset U(n)\), then the trace form \(B_0(X,Y)={{\,\textrm{tr}\,}}(XY)\) has this property. For every \(\lambda \in \mathfrak {t}_\mathbb {C}^*\) which is imaginary on \(\mathfrak {t}\) (such as a weight) define \(H_\lambda \in \sqrt{-1}\mathfrak {t}\) by \( \lambda (H)= \langle H, H_\lambda \rangle \) for all \(H\in \sqrt{-1}\mathfrak {t}\). Observe that \(\langle \lambda , \mu \rangle = \langle H_\lambda , H_\mu \rangle \) defines a positive definite inner product on \(\sqrt{-1}\mathfrak {t}^*\subset \mathfrak {t}_\mathbb {C}^*\) and denote the induced norm by . The following theorem is then essentially due to Sugiura [Sug71]; although the results of [Sug71] concern functions on K, the argument generalizes to an arbitrary continuous representation. For the sake of completeness, we will prove this more general statement here.
Theorem 2.2
(Sugiura). Let \(v\in E^\infty \). Then for every integer \(q\ge 0\) and every continuous seminorm \(|\cdot |\) on \(E^\infty \),
Proof
First, consider the extension of the representation \(d\rho \) of \(\mathfrak {k}\) to the universal enveloping algebra \(U(\mathfrak {k})\) and the Casimir operator
where \(X_1,\ldots , X_r\) is a basis of \(\mathfrak {k}_\mathbb {C}\) and \((g^{ij})=(g_{ij})^{-1}\) with \(g_{ij}=\langle X_i,X_j\rangle \). It is well known that \(d\rho (\Delta )\) acts on each isotypic component \(E_\lambda \) by multiplication by \(\langle \lambda ,\lambda +2\delta \rangle \), where \(\delta =\frac{1}{2} \sum _{\alpha \in \Delta ^+}\alpha \), and that
see, e.g., Lemma 1.1 and (1.10), respectively, in [Sug71]. Second, because K is compact, the family of operators \(\{\rho (k)\mid k\in K\}\) is equicontinuous. Consequently, for every continuous seminorm on \(E^\infty \), there exists another continuous seminorm such that
holds for all \(k\in K\) and \(v\in E^\infty \). Third, recall that since the absolute value of the trace of a unitary matrix cannot exceed its rank, we have for any \(\lambda \in {\widehat{K}}\) and \(k\in K\)
Finally, according to Weyl’s dimension formula, there exist \(c,j\in \mathbb {N}\) with
see (1.17) in [Sug71]. Combining these facts together, we obtain for every \(\lambda \in {\widehat{K}}\), \(i\in \mathbb {N}\), and \(v\in E^\infty \)
and choosing \(i\in \mathbb {N}\) such that \(i>j+q\) thus yields the result. \(\square \)
Finally, we will need the following consequence of Schur’s lemma.
Proposition 2.3
Let V and W be irreducible representations of K and \(q:V\times W \rightarrow \mathbb {C}\) a K-invariant sesquilinear form.
-
(a)
If \(V\ncong W\) then \(q=0\) identically.
-
(b)
Assume \(V=W\). If there is \(x\in V\) such that \(q(x,x)>0\) then q is a Hermitian inner product on V.
Proof
Observe first that there exists a K-invariant Hermitian inner product on W: Take any Hermitian inner product h on W and set
In particular, \(h_K\) induces a K-equivariant isomorphism \(\overline{W}^*\cong W\). Consequently, we have \(q\in (V^*\otimes \overline{W}^*)^K\cong {{\,\textrm{Hom}\,}}_K(V, W)\). By Schur’s lemma,
This directly implies item (a). As for item (b), because \(h_K\in (V^*\otimes \overline{V}^*)^K\), there is \(c\in \mathbb {C}\) with \(q=ch_K\). If \(q(x,x)>0\) for some \(x\in V\) then \(c>0\) and hence q is a Hermitian inner product itself. \(\square \)
2.2 Special orthogonal group.
Later we will apply the above general theory to the Fréchet spaces of continuous and smooth valuations and the special orthogonal group \(K={{\,\textrm{SO}\,}}(n)\). To this end, let us now take a closer look at the latter, following Knapp’s monograph [Kna01]. We distinguish two cases according to the parity of n.
Assume first \(n=2l\) and let us fix the maximal torus \(T\subset {{\,\textrm{SO}\,}}(2l)\) consisting of matrices of the form
We further denote by \(E_{i,j}\) the (i, j)-th element of the standard basis of \(\mathbb {C}^{n\times n}\) and choose a basis \(H_1,\dots , H_l\) of \(\mathfrak {t}_\mathbb {C}\) and the dual basis \(\epsilon _1,\dots \epsilon _l\) of \(\mathfrak {t}^*_\mathbb {C}\) as follows:
We introduce an ordering on the roots by declaring a linear function \(\lambda \in \mathfrak {t}_\mathbb {C}^*\) that is real on \(\sqrt{-1}\mathfrak {t}\) to be positive if \( \lambda (H_1)>0\), or \(\lambda (H_1)=0\) and \(\lambda (H_2)>0\), ..., or \(\lambda (H_1)=\cdots = \lambda (H_{n-1})=0\) and \(\lambda (H_n)>0\). The set of positive roots is then
Finally, set
For \(i<j\) the root space \(\mathfrak {k}_{\epsilon _i\pm \epsilon _j}\subset \mathfrak {k}_\mathbb {C}\) is spanned by the \(l\times l\) block matrix \(X_{\epsilon _i \pm \epsilon _j }\) given (block-wise) as
Second, if \(n=2l+1\), we choose the maximal torus \(T\subset {{\,\textrm{SO}\,}}(2l+1)\) to consist of
The bases for \(\mathfrak {t}_\mathbb {C}\) and \(\mathfrak {t}^*_\mathbb {C}\) are still (formally) given as above. The root system, on the contrary, has a different structure; namely,
Consequently, each positive root space \(\mathfrak {k}_\alpha \subset \mathfrak {k}_\mathbb {C}\) is spanned by a matrix \(X_\alpha \), where \(X_{\epsilon _i\pm \epsilon _j}\) is the image of (1) under the embedding \(\mathbb {C}^{2l\times 2l}\rightarrow \mathbb {C}^{(2l+1)\times (2l+1)}\) into the left upper corner, while
To conclude, recall that in the case \(K={{\,\textrm{SO}\,}}(n)\), \({\widehat{K}}\) is parametrized by the set
Explicitly, \((\lambda _1,\dots ,\lambda _l)\) corresponds to the highest weight \(\sum _{i=1}^l\lambda _i\epsilon _i\).
3 Valuations on convex bodies
To fix notation and for quick later reference we review the theory of translation-invariant valuations on convex bodies. For more information on valuations see the books by Alesker [Ale18], Klain and Rota [KR97], and Schneider [Sch14].
We denote by
the volume of the n-dimensional Euclidean unit ball \(D^n\subset \mathbb {R}^{n}\) and by \(s_n\) the volume of the Euclidean unit sphere \(S^n=\partial D^{n+1}\). We will use the well-known relations
Moreover, we will need the definite integral
3.1 Continuous translation-invariant valuations.
Let \({\mathcal {K}}(\mathbb {R}^n)\) denote the family of convex bodies in \(\mathbb {R}^n\). Equipped with the Hausdorff metric, \({\mathcal {K}}(\mathbb {R}^n)\) becomes a locally compact topological space. A function \(\phi :{\mathcal {K}}(\mathbb {R}^n)\rightarrow \mathbb {C}\) is called a valuation if
holds for \(K,L\in {\mathcal {K}}(\mathbb {R}^n)\), whenever \(K\cup L\) is convex. The space of continuous and translation-invariant valuations is denoted \({{\,\textrm{Val}\,}}(\mathbb {R}^n)\).
It should be mentioned that valuations defined on various other classes of sets have been investigated in the literature and that, more recently, valuations on function spaces have also attracted considerable attention. We refer the reader to the books cited at the beginning of this section and to the articles [CLM20, CLM20a, Kno20, Kno21] for more information on these subjects.
A fundamental classical result is McMullen’s decomposition theorem asserting that
where \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) denotes the subspace of r-homogeneous valuations, i.e., satisfying \(\phi (tK)=t^r \phi (K)\) for all \(t>0\) and \(K\in {\mathcal {K}}(\mathbb {R}^n)\). The above can be further refined by decomposing
into even and odd valuations with respect to the reflection in the origin. More precisely, \(\phi \in {{\,\textrm{Val}\,}}^s_r(\mathbb {R}^n)\) if and only if \(\phi (-K)=(-1)^s \phi (K)\) for all \(K\in {\mathcal {K}}(\mathbb {R}^n)\).
It is easy to see that \({{\,\textrm{Val}\,}}_0(\mathbb {R}^n)\) is spanned by the constant valuation \(\chi (K)=1\), \(K\in {\mathcal {K}}(\mathbb {R}^n)\). By a theorem of Hadwiger, \({{\,\textrm{Val}\,}}_n(\mathbb {R}^n)\) is also 1-dimensional and spanned by the volume \({{\,\textrm{vol}\,}}_n\). The r-th intrinsic volume \(\mu _r\) defined by Steiner’s formula
is an important example of valuation of degree r. More generally, the mixed volume \(V(K_1,\dots ,K_n)\), i.e., the symmetric function of \(K_1,\dots ,K_n\in {\mathcal {K}}(\mathbb {R}^n)\) given by
is further a 1-homogeneous valuation in every argument. Observe that
The space \({{\,\textrm{Val}\,}}(\mathbb {R}^n)\) carries a natural topology, namely, the topology of uniform convergence on compact subsets. McMullen’s decomposition theorem implies that it is in fact induced by the Banach space norm
The natural action of the general linear group \({{\,\textrm{GL}\,}}(n)\) on \({{\,\textrm{Val}\,}}(\mathbb {R}^n)\) is
\({{\,\textrm{Val}\,}}(\mathbb {R}^n)\) thus becomes a continuous Banach space representation. Alesker’s celebrated irreducibility theorem [Ale01] asserts that \({{\,\textrm{Val}\,}}_r^0(\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_r^1(\mathbb {R}^n)\) are irreducible, i.e., contain no non-trivial closed \({{\,\textrm{GL}\,}}(n)\)-invariant subspaces.
General theory dictates that under the action of the compact group \({{\,\textrm{SO}\,}}(n)\), the representations \({{\,\textrm{Val}\,}}_r^0(\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_r^1(\mathbb {R}^n)\) decompose into isotypic components. The following theorem characterizes the \({{\,\textrm{SO}\,}}(n)\)-types that appear.
Theorem 3.1
(Alesker–Bernig–Schuster [ABS11, Theorem 1]). Let \( l=\lfloor \frac{n}{2}\rfloor \) and \( 0\le r \le l \). The non-trivial \({{\,\textrm{SO}\,}}(n)\)-types in \({{\,\textrm{Val}\,}}_{r}(\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_{n-r}(\mathbb {R}^n)\) are the same and given by the following set of highest weights:
Moreover, each appears with multiplicity one.
Remark 3.2
Theorem 3.1 was first proved by Alesker [Ale01, Proposition 6.3] for even valuations: Replacing \({{\,\textrm{Val}\,}}_j(\mathbb {R}^n)\) by \({{\,\textrm{Val}\,}}_j^0(\mathbb {R}^n)\), \(j\in \{r,n-r\}\), \(\Lambda _r\) has to be replaced by
This shows that the parity of valuations in the irreducible subspace corresponding to \(\lambda _{k,m}\) is given by the parity of m. Below we will give a new proof of this statement by constructing for each \({{\,\textrm{SO}\,}}(n)\)-type a non-trivial highest weight vector.
3.2 Integration over the normal cycle.
3.2.1 Normal cycles.
Integration over the normal cycle is an important means of constructing valuations. Let \(S\mathbb {R}^n=\mathbb {R}^n\times S^{n-1}\) denote the sphere bundle of \(\mathbb {R}^n\). As a set, the normal cycle of a convex body \(K\in {\mathcal {K}}(\mathbb {R}^n)\) is
Since N(K) is also a naturally oriented \((n-1)\)-dimensional Lipschitz submanifold, smooth differential \((n-1)\)-forms on \(S\mathbb {R}^n\) may be integrated over the normal cycle. Let \(\Omega ^*(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\subset \Omega ^*(S\mathbb {R}^n)\) be the subspace of translation-invariant forms. The key fact is that each form \(\omega \in \Omega ^{n-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\) defines via
a continuous, translation-invariant valuation (see, e.g., [AF08] or [Fu17]). Observe that the space of differential forms on \(S\mathbb {R}^n\) is naturally bi-graded and that \(\phi \in {{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) provided \(\omega \in \Omega ^{r,n-1-r}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\).
3.2.2 The kernel theorem.
The differential forms \(\omega \) for which (4) yields the trivial valuation can be characterized in terms of the Rumin differential [Rum94]. This natural second order differential operator is defined for \((n-1)\)-forms on a contact manifold M of dimension \(2n-1\). Assume for simplicity that the contact structure on M is defined by a global contact form \(\alpha \). Then, restricted to each contact hyperplane
\(d\alpha \) is non-degenerate. By the Lefschetz decomposition theorem applied to the exterior algebra \(\wedge H_x^* \), multiplication by \(d\alpha \) defines an isomorphism \(\wedge ^{n-2} H_x^* \rightarrow \wedge ^{n} H_x^*\). Consequently, there exists a unique form \(\alpha \wedge \xi \in \Omega ^{n-1}(S\mathbb {R}^n)\) such that
is divisible by \(\alpha \). The latter n-form is denoted \(D\omega \) and called the Rumin differential of \(\omega \). The following is a special case of the general Bernig–Bröcker kernel theorem.
Theorem 3.3
(Bernig–Bröcker [BB07]). Let \(1\le r\le n-1\). The valuation defined by \(\omega \in \Omega ^{r,n-r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\) is identically zero if and only if \(D\omega =0\).
3.2.3 Smooth valuations.
A particularly important technical innovation introduced by Alesker is the notion of smooth valuation. A valuation \(\phi \in {{\,\textrm{Val}\,}}(\mathbb {R}^n)\) is called smooth if it is a smooth vector under the action of the general linear group. The dense subspace of smooth valuations is denoted by \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\). Similarly, we define \({{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_r^{s,\infty }(\mathbb {R}^n)\).
An important consequence of the irreducibility theorem and a big advantage of working with smooth valuations is that they can be represented by differential forms. More precisely, given \(\phi \in {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\), \(r<n\), there exists \(\omega \in \Omega ^{r,n-r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\) such that (4) holds, see [AF08]. Conversely, it is not hard to verify that every valuation of this form is smooth.
3.3 Product and convolution of smooth valuations.
3.3.1 Alesker product.
One of the most remarkable properties of the space of smooth valuations is that it carries the structure of a graded commutative algebra. Given \(A\in {\mathcal {K}}(\mathbb {R}^n)\) define . It is readily verified that \(\phi _A\in {{\,\textrm{Val}\,}}(\mathbb {R}^n)\). In fact, \(\phi _A\) is smooth provided the convex body A has a smooth and strictly positively curved boundary.
Theorem 3.4
(Alesker [Ale04]). There exists a unique continuous and bilinear product \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\times {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) such that
where \(\Delta :\mathbb {R}^n \rightarrow \mathbb {R}^n\times \mathbb {R}^n\) is the diagonal embedding, holds for all \(A,B\in {\mathcal {K}}(\mathbb {R}^n)\) with smooth and strictly positively curved boundary. Moreover,
-
(a)
\(\cdot \) is commutative and associative;
-
(b)
\(\chi \) is the identity element;
-
(c)
\({{\,\textrm{Val}\,}}^\infty _r (\mathbb {R}^n)\cdot {{\,\textrm{Val}\,}}^\infty _q(\mathbb {R}^n) \subset {{\,\textrm{Val}\,}}_{r+q}^\infty (\mathbb {R}^n)\);
-
(d)
the product pairing \( {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\times {{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}_n^\infty (\mathbb {R}^n)\cong \mathbb {C}\) is perfect.
Item (d) of the preceding theorem is a version of Poincaré duality for valuations. The pairing is usually referred to as the Alesker–Poincaré pairing.
A general formula for the product of valuations in terms of differential forms exists (see [AB12]) but it seems too unwieldy for explicit computations. However, in an important special case it boils down to a more manageable expression.
Theorem 3.5
(Bernig [Ber09, Theorem 4.1]). Let \(1\le r\le n-1\) and \(s\in \{0,1\}\). Assume that \(\phi _1\in {{\,\textrm{Val}\,}}_r^{\infty ,s}(\mathbb {R}^n)\) and \(\phi _2\in {{\,\textrm{Val}\,}}^{\infty ,s}_{n-r}(\mathbb {R}^n)\) are represented by \(\omega _1\in \Omega ^{r,n-r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\) and \(\omega _2\in \Omega ^{n-r,r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\), respectively. If the standard orientation on \(\mathbb {R}^n\) is used to identify \(\wedge ^n(\mathbb {R}^n)^*\) and \({{\,\textrm{Val}\,}}_n(\mathbb {R}^n)\), then
Let us spell out explicitly how to interpret the integral in the previous theorem. Observe that \(\omega _1\wedge D\omega _2\in \Omega ^{n,n-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\) and that this space is naturally identified with \(\wedge ^n(\mathbb {R}^n)^*\otimes \Omega ^{n-1}(S^{n-1})\). For \(\alpha \in \wedge ^n(\mathbb {R}^n)^*\) and \(\beta \in \Omega ^{n-1}(S^{n-1})\) one defines
3.3.2 Bernig–Fu convolution.
Remarkably, there is yet another continuous product on the space of smooth valuations. We have the following theorem.
Theorem 3.6
(Bernig–Fu [BF06]). There exists a unique continuous and bilinear product \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\times {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) such that
for all \(A,B\in {\mathcal {K}}(\mathbb {R}^n)\) with smooth and strictly positively curved boundary. Moreover,
-
(a)
\(*\) is commutative and associative;
-
(b)
\({{\,\textrm{vol}\,}}_n\) is the identity element;
-
(c)
\({{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n)* {{\,\textrm{Val}\,}}_{n-q}^\infty (\mathbb {R}^n) \subset {{\,\textrm{Val}\,}}_{n-r-q}^\infty (\mathbb {R}^n)\).
Let us point out that, twisting the space \({{\,\textrm{Val}\,}}^\infty \) by the one-dimensional space of densities, the definition of convolution can be reformulated without reference to any Euclidean structure so that it commutes with the natural action of the general linear group. This will be briefly touched upon in §3.4 below. In the above form, however, both convolution and product define a multiplicative structure on the same space. These algebras are different, but isomorphic (see §3.4.1). Moreover, the induced pairings coincide on \({{\,\textrm{Val}\,}}_r^{s,\infty }(\mathbb {R}^n)\) up to sign, after suitable identifications are made.
Proposition 3.7
([BF06, Lemma 2.3] and [Wan14, Proposition 4.2]). Let \(0\le r\le n\) and \(s \in \{0,1\}\). If \({{\,\textrm{Val}\,}}_0(\mathbb {R}^n)\) and \({{\,\textrm{Val}\,}}_n(\mathbb {R}^n)\) are identified with \(\mathbb {C}\) via \(\chi \) and \({{\,\textrm{vol}\,}}_n\), then for any \(\phi _1\in {{\,\textrm{Val}\,}}_r^{s,\infty }(\mathbb {R}^n)\) and \(\phi _2\in {{\,\textrm{Val}\,}}^{s,\infty }_{n-r}(\mathbb {R}^n)\) one has
The convolution is typically easier to evaluate than the product. One example that will be relevant for us is the following. Recall that the intrinsic volumes are smooth and consider the degree \(-1\) linear operator on \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) given by
It is an easy consequence of Steiner’s formula that for any \(\phi \in {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) one has
see [BF06, Corollary 1.8]. In particular,
In the language of differential forms, \(\Lambda \) is nothing else than the Lie derivative with respect to the Reeb vector field \(T_{(x,\xi )}= \sum _{i=1}^n \xi _i \frac{\partial }{\partial x_i}\) on \(S\mathbb {R}^n\):
Proposition 3.8
(Bernig–Bröcker [BB07, Lemma 3.4]). Assume \(\phi \in {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) is represented by \(\omega \in \Omega ^{n-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\). Then
3.3.3 Hard Lefschetz theorem.
Using Proposition 3.8 together with Kähler identities from Kähler geometry, Bernig and Bröcker proved that the operator \(\Lambda \) satisfies the hard Lefschetz property.
Theorem 3.9
(Bernig–Bröcker [BB07, Theorem 2]). Let \(0\le r\le \lfloor \frac{n}{2}\rfloor \). The map \(\Lambda ^{n-2r}:{{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n) \rightarrow {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\) is an isomorphism of topological vector spaces.
In §10 below we will give an alternative prove of this result and supplement it with the Hodge–Riemann relations. Our proof relies on Proposition 3.8 as well as on the Alesker–Bernig–Schuster decomposition theorem.
Remark 3.10
Strictly speaking, the proof of Theorem 3.1 given in [ABS11] depends on Theorem 3.9; namely, the Lefschetz isomorphism is used to extend the validity of the statement from \({{\,\textrm{Val}\,}}_{n-r}(\mathbb {R}^n)\) to \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\). Observe, however, that the use of the hard Lefschetz theorem can easily be avoided, using only Alesker–Poincaré duality instead: For any \(\lambda \in \Lambda \) let \(\widetilde{E}_\lambda \subset {{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) and \(E_\lambda \subset {{\,\textrm{Val}\,}}_{n-r}(\mathbb {R}^n)\) be the \(\lambda \)-isotypic components. By Alesker–Poincaré duality and Proposition 2.3 (a), there is a non-degenerate \({{\,\textrm{SO}\,}}(n)\)-invariant sesquilinear pairing \(\widetilde{E}_\lambda \times E_\lambda \rightarrow \mathbb {C}\). If Theorem 3.1 holds for \({{\,\textrm{Val}\,}}_{n-r}(\mathbb {R}^n)\), we have \(\dim E_\lambda <\infty \) and consequently \(\widetilde{E}_\lambda \cong \overline{E}^*_\lambda \cong E_\lambda \).
3.4 Pullback, pushforward, and the Fourier transform.
3.4.1 Fourier transform.
The Fourier transform for valuations discovered by Alesker is a linear isomorphism that respects the action of the general linear group, intertwines the product and the convolution, and satisfies an analogue of the Fourier inversion theorem. To trace the correct behaviour under the action of \({{\,\textrm{GL}\,}}(n)\), we first do not fix a Euclidean inner product and work with an abstract n-dimensional real vector space V instead. As a consequence, there exists no canonical choice of volume on V, but a one-dimensional space of Haar measures. The complexification of this space is denoted by \({{\,\textrm{Dens}\,}}(V)\) and its elements are called densities on V. Note that \({{\,\textrm{Val}\,}}_n(V)\cong {{\,\textrm{Dens}\,}}(V)\) by Hadwiger’s theorem.
Theorem 3.11
(Alesker [Ale11]). There exists an isomorphism of topological vector spaces
which satisfies the following properties:
-
(a)
\(\mathbb {F}_V\) commutes with the natural action of \({{\,\textrm{GL}\,}}(V)\) on both spaces;
-
(b)
\(\mathbb {F}_V\) is an isomorphism of algebras when the source is equipped with the product and the target with the convolution;
-
(c)
after the canonical identification \( {{\,\textrm{Dens}\,}}(V^*)\otimes {{\,\textrm{Dens}\,}}(V)\cong \mathbb {C}\), the composition
$$\begin{aligned} {\mathcal {E}}_V=\left( \mathbb {F}_{V^*}\otimes Id_{{{\,\textrm{Dens}\,}}(V)}\right) \circ \mathbb {F}_V:{{\,\textrm{Val}\,}}^\infty (V) \rightarrow {{\,\textrm{Val}\,}}^\infty (V) \end{aligned}$$satisfies
$$\begin{aligned} ({\mathcal {E}}_V\phi )(K)= \phi (-K). \end{aligned}$$
We will reduce our computation of the Fourier transform recursively to the case \(V=\mathbb {R}^2\) where an explicit description is available as follows. First, by an enhancement of a classical theorem of Hadwiger, every valuation \(\phi \in {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^2)\) can be (uniquely) expressed as
where \(c_0,c_2\in \mathbb {C}\) are constants and \(f:S^1\rightarrow \mathbb {C}\) is a smooth function orthogonal to the two-dimensional space of linear functionals restricted to \(S^1\). denotes here the area measure of \(K\in {\mathcal {K}}(\mathbb {R}^2)\). Decompose \(f=f_0+f_1\) into the even and odd part and the latter \(f_1=f_{1,h}+f_{1,a}\) into the holomorphic and anti-holomorphic part, where \(\mathbb {R}^2\cong \mathbb {C}\) as usual. We further use the inner product on \(\mathbb {R}^2\) to identify \({{\,\textrm{Val}\,}}^\infty \big ((\mathbb {R}^2)^*\big )\otimes {{\,\textrm{Dens}\,}}(\mathbb {R}^2) \cong {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^2)\). Then one has
where \(J:\mathbb {R}^2\rightarrow \mathbb {R}^2\) denotes the counter-clockwise rotation by \(\pi /2\).
3.4.2 Pullback and pushforward.
To evaluate the Fourier transform on highest weight vectors we will make use of its compatibility properties with respect to the pullback and pushforward of valuations. Since we will need the pullback under monomorphisms and the pushforward along epimorphisms exclusively, we only define them in these special cases here. For the general case see [Ale11].
The pullback is easy to define. If \(f:V\rightarrow W\) is an injective linear map of vector spaces, then \(f^*:{{\,\textrm{Val}\,}}^\infty (W)\rightarrow {{\,\textrm{Val}\,}}^\infty (V)\) defined by
is a morphism of algebras with respect to the product.
The definition of the pushforward, on the contrary, is slightly more involved. Let \(f:V\rightarrow W\) be a surjective linear map and choose \(U\subset V\) such that \(V=\ker f \oplus U \). \(g=f|_U:U \rightarrow W\) is an isomorphism which yields \({{\,\textrm{Dens}\,}}(U)\cong {{\,\textrm{Dens}\,}}(W)\). Observe further that there is a natural isomorphism \({{\,\textrm{Dens}\,}}(V^*)\cong {{\,\textrm{Dens}\,}}(V)^*\) and that
Then \(f_* :{{\,\textrm{Val}\,}}^\infty (V)\otimes {{\,\textrm{Dens}\,}}(V^*)\rightarrow {{\,\textrm{Val}\,}}^\infty (W)\otimes {{\,\textrm{Dens}\,}}(W^*)\) is defined as follows. For
and any \(K\in {\mathcal {K}}(W)\) we set
where \(k=\dim \ker f\) and \(S\in {\mathcal {K}}(\ker f)\) is chosen so that \({{\,\textrm{vol}\,}}_{\ker f} (S)=1\). One can check that the definition does not depend on the choices of U and S and that \(f_*\) is a morphism of algebras with respect to the convolution. Moreover,
Theorem 3.12
(Alesker [Ale11]). Let \(f:V\rightarrow W\) be an injective linear map and let \(f^\vee :W^*\rightarrow V^*\) denote the dual map. Then
4 Construction of highest weight vectors
The following two sections constitute the technical heart of this paper. Namely, we explicitly construct in each \({{\,\textrm{SO}\,}}(n)\)-type appearing in \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\) a non-zero highest weight vector. To this end, and this is the goal of the current section, we first construct a highest weight vector in \( \Omega ^{r,n-r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\). Integration over the normal cycle then yields a highest weight vector in \({{\,\textrm{Val}\,}}_r(\mathbb {R}^n)\).
To prove that a differential form defines a non-trivial valuation, it suffices by the Bernig–Bröcker kernel theorem to compute the Rumin differential and show it is non-zero. These two steps of our argument, to be completed in Sects. 5 and 6, respectively, are combinatorially challenging but the formalism of double forms allows us to conveniently handle the rather complicated determinantal expressions appearing below.
4.1 Double forms.
Let M be a smooth manifold and let V be a finite-dimensional, real vector space. We denote by \(\Omega ^k(M,V)\) the space of smooth V-valued differential k-forms, i.e., the smooth sections of the bundle \(\mathbb {C}\otimes \wedge ^k T^*M\otimes V \), where the tensor products are taken over \(\mathbb {R}\). The complex conjugation, pullback, exterior derivative, and Lie derivative of a V-valued form are defined in the usual way. Recall also that there exists a natural product
of V- and W-valued forms. In the following we will make use of differential forms with values in \(\wedge V\), the exterior algebra of a vector space V. In this setting, the composition of the above general product with the wedge product on \(\wedge V\) yields a wedge product on \( \Omega ^*(M,\wedge V)= \bigoplus _k \Omega ^k(M,\wedge V)\). We will refer to elements of this algebra as double forms. A double form of type (p, q) is a smooth section of the bundle \(\mathbb {C}\otimes \wedge ^pT^*M\otimes \wedge ^q V\). Note that
if \(\omega \) is of type (p, q) and \(\theta \) of (r, s), where here and in the following we suppress the wedge symbol. For any (1, 1)-double form \(\beta \) and \(j\in \mathbb {N}_0\) we denote
Then for any such double forms \(\beta _1,\beta _2\) and \(m\in \mathbb {N}_0\) one has
Moreover, it will be convenient and natural to set \(\beta ^{[-1]}=0\) and \( \left( {\begin{array}{c} {i} \\ {j} \\ \end{array} } \right) =0\) if \(i<j\).
For our purposes, M will usually be either \(\mathbb {R}^n\times \mathbb {R}^n\) or \(S\mathbb {R}^n\) and the double forms will take values in \(V= \mathbb {C}\otimes \wedge ( \mathbb {R}^n\times \mathbb {R}^n)^*\). In this connection, let us fix once and for all the following notation. First, the canonical coordinates in \(\mathbb {R}^n \times \mathbb {R}^n\) are denoted by \(x_1,\dots ,x_n,\xi _1,\dots ,\xi _n\). Second, put \(l=\lfloor \frac{n}{2}\rfloor \) and define for \(j\in \{1,\ldots , l\}\)
Let us point out that the deviation by the factor \(\frac{1}{\sqrt{2}}\) from standard conventions [Huy05] visually simplifies multiple formulas in the sequel. If \(n=2l+1\), we set in addition
Further, consider the set of indices
and equip it with the total order \(1\prec \overline{1}\prec \cdots \prec l\prec \overline{l} \prec l+1\). For any subset \(I=\{i_1,\dots ,i_j\}\subset {\mathcal {I}}\) with \(i_1\prec \cdots \prec i_j\) we define the following (double) forms:
where \(\bar{{\bar{i}}}=i\) and \(\overline{l+1}=l+1\). Notice that \(\alpha =\alpha _{\mathcal {I}}\) is the contact form on \(S\mathbb {R}^n\). Similarly, we set
Assume \(k\in \mathbb {N}\), \(k\le l\). Of particular interest will be the following subsets of \({\mathcal {I}}\):
where \(\overline{\{i_1,\dots ,i_j\}}=\{\overline{i_1},\dots ,\overline{i_j}\}\); for \(n=2l\) we will use in addition
Observe that \({\mathcal {I}}=K\cup \overline{K}\cup L\), \( J=\overline{K}\cup L\), and \({\mathcal {I}}=M\cup \overline{M}\) (if \(n=2l\)) are disjoint unions. Finally, for any \(I\subset {\mathcal {I}}\) and \(a_1,\dots ,a_p\in I\) we will denote
In general, we will not notationally distinguish (double) forms on \(\mathbb {R}^{n}\times \mathbb {R}^n\) from their restrictions to \(S\mathbb {R}^n\); however, an explicit distinction will be made whenever this difference is relevant. In this connection, let us state for later use the following obvious fact.
Lemma 4.1
Let \(\iota :S\mathbb {R}^n \hookrightarrow \mathbb {R}^n\times \mathbb {R}^n\) denote the inclusion map and let \(\omega \in \Omega ^*(\mathbb {R}^n\times \mathbb {R}^n)\). If \(\omega \wedge \gamma =0\), then \(\iota ^*\omega =0\).
4.2 Action of the orthogonal group.
Recall that if \(G\times M\rightarrow M\) is a smooth left action of a Lie group on a smooth manifold and if X is an element of the Lie algebra \(\mathfrak {g}\) of G, then
is called the fundamental vector field induced by X. On smooth differential forms \(\omega \in \Omega ^k(M,V)\) the group acts by \( g\cdot \omega = (g^{-1})^*\omega \) . Thus the infinitesimal action of \(\mathfrak {g}\) on differential forms is given by
where \({\mathcal {L}}\) denotes the Lie derivative. By linearity, this can be extended to an action of \(\mathfrak {g}_\mathbb {C}\), the complexification of \(\mathfrak {g}\).
The natural left action of \({{\,\textrm{SO}\,}}(n)\) on \(\mathbb {R}^n\times \mathbb {R}^n\) and \(S\mathbb {R}^n\) is \(g\cdot (x,\xi )= (gx, g\xi )\). A direct computation shows that the corresponding fundamental vector fields (on either \(\mathbb {R}^n\times \mathbb {R}^n\) or \(S\mathbb {R}^n\)) induced by the basis vectors of \({\mathfrak {n}^{+}}\oplus {\mathfrak {t}_{\mathbb {C}}}\subset {\mathfrak {so}} {(n)_{\mathbb {C}}}\) introduced in §2.2 are given by
where
for \(j\in \{ 1,\ldots , l\}\) and, provided \(n=2l+1\),
4.3 Highest weight vectors.
We will now describe a family of highest weight vectors in \(\Omega ^{r,n-r-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\). Already here double forms are a convenient tool for organizing our computation and stating our results.
Theorem 4.2
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
is a highest weight vector of weight \(\lambda _{k,m}\), where
If \(n=2l\), then in addition
is a highest weight vector of weight \(\lambda _{-l,m}\), where
Before we will proceed to the proof of this theorem, let us first show two simple auxiliary statements.
Proposition 4.3
For any \(r,k\in \mathbb {N}\) with \(r\le n-1\) and \(k \le \min \{ r,n-r\}\),
Proof
We have
Remark 4.4
-
(a)
Since some version of the same argument will appear numerous times throughout the following pages, the previous proof is perhaps worth a brief comment: Observe that three elementary facts are used here, namely the binomial formula (7), the disjointness of the union \(J=\overline{K}\cup L\), and that one has \(\bigwedge ^aV=0\) for any vector space V and any integer \(a>\dim V\).
-
(b)
Observe that one can define \(\omega _{r,k}\) also for \( k>n-r\). However, the same argument as above shows that in this case \(\omega _{r,k}\otimes \Theta _1 =0\).
Lemma 4.5
Let \(I\subset {\mathcal {I}}\) be non-empty; choose \(a\in I\) and \(b\in {\mathcal {I}}\) with \(a\ne b\); and denote \(q=|I|\). Let \(Z_{a,b}=z_a\frac{\partial }{\partial z_b}+ \zeta _a \frac{\partial }{\partial \zeta _b}\). Then
Proof
We may assume \(b\in I\); otherwise the claim is obvious. In this case, (11) follows easily from skew-symmetry. As for (12), we have
Proof of Theorem 4.2
First of all, observe that \(\zeta _{\overline{1}}^{m-2}\) is obviously a highest weight vector of weight \(\lambda _{1,m-2}\) and there is, thus, no loss of generality in assuming \(m=2\). There are several cases to be considered separately.
First, it follows at once from Proposition 4.3 and Lemma 4.5 that for any \(1\le i<j\le k\), \(X_{\epsilon _i\pm \epsilon _j}\omega _{r,k}=0\). Furthermore, from Lemma 4.5 we also easily infer that \(X_{\epsilon _i\pm \epsilon _j}\omega _{l,-l}=0\) for any \(1\le i<j\le l\) provided \(n=2l\).
Second, let \(\omega _{r,k}\otimes \Theta _1=\Omega _1+\Omega _2\) be the decomposition (10). If \(a\in K\) and \(b\in L\), for the vector field \(Y_{a,b}=z_{\overline{a}}\frac{\partial }{\partial z_{\overline{b}}}-z_b\frac{\partial }{\partial z_a}+\zeta _{\overline{a}}\frac{\partial }{\partial \zeta _{\overline{b}}}-\zeta _b\frac{\partial }{\partial \zeta _a}\) one has
This shows that \(X_{\epsilon _i\pm \epsilon _j}\omega _{r,k}=X_{\epsilon _i}\omega _{r,k}=0\) for \(1\le i \le k < j\le l\).
Finally, let \(a,b\in L\) with \(a\ne b\) and consider the vector field \(Z_{a,b}\) from Lemma 4.5. If \( n-r-k\ge 1\), then
If \(n-r-k=0\), we get the same result directly from Lemma 4.5. If \(n-r-k\ge 2\), we similarly have
If \(n-r-k=1\), we get the same directly from Lemma 4.5. Hence \(X_{\epsilon _i\pm \epsilon _j}\omega _{r,k}=0\) for \(k<i<j\le l\) as well as \(X_{\epsilon _i}\omega _{r,k}=0\) for \(k<i\le l\).
Altogether, we have shown that both \(\omega _{r,k}\) and \(\omega _{l,-l}\) are highest weight vectors in fact. Using Proposition 4.3, one immediately verifies that their weights are \(\lambda _{k,2}\) and \(\lambda _{-l,2}\), respectively. \(\square \)
Definition 4.6
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\), let us define a valuation by
If \(n=2l\), then we define in addition
Clearly, the map \(\Omega ^{n-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\rightarrow {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) given by (4) commutes with the action of \({{\,\textrm{SO}\,}}(n)\). Hence, by Theorem 4.2, \(\phi _{r,k,m}\in {{\,\textrm{Val}\,}}^\infty _r(\mathbb {R}^n)\), is a highest weight vector of weight \(\lambda _{k,m}\) and similarly \(\phi _{l,-l,m}\in {{\,\textrm{Val}\,}}^{\infty }_l(\mathbb {R}^{2l})\) is a highest weight vector of weight \(\lambda _{-l,m}\). Observe that \(\phi _{r,k,m}\in {{\,\textrm{Val}\,}}_r^s(\mathbb {R}^n)\) as well as \(\phi _{l,-l,m}\in {{\,\textrm{Val}\,}}_r^s(\mathbb {R}^{2l})\) for \(m\equiv s\mod 2\), cf. Remark 3.2. However, it is a priori not clear that these valuations are not identically zero; this will become evident only later, see Remark 6.8 below.
5 The Rumin differential
We will now compute the action of the Rumin differential D on the highest weight vectors \(\omega _{r,k,m}\) constructed in the previous section. We will assume
throughout the entire section. Let us also define more (double) forms as follows:
Theorem 5.1
Denote \(c_{r,m}=(-1)^{n+1}(n+m-r-2)\). Then
The key ingredient in the proof of this result will be the relation (22) among \(\delta _{r,k}\), \(\theta _{r,k}\), and \(\sigma _{r,k}\) below whose proof, in turn, decays into a series of auxiliary identities. Let us begin with stating and verifying those.
Proposition 5.2
One has
Proof
To show (14), observe that for each \(i\in K\) we have
and sum over i. The proof of (15) is completely analogous. As for (16), one has
from which the claim follows since \((d\zeta _{\overline{K}})^{[k]}(\overline{dz_K})^{[k]}=(-1)^k(\overline{d\zeta _K})^{[k]}(dz_{\overline{K}})^{[k]}\). (17) is proven in precisely the same way as (10) above. Further, for any \(i\in L\) we have
and
which proves (18) and (19), respectively. Similarly, for any \(i\in L\) we have
Taking the sum over \(i\in L\), the second term in the resulting expression clearly does not contribute while the first yields (20). Finally, let us prove (21). First, for any \(i\in K\) we have
Concerning the first term,
As for the second, by (14) we have
and summing over i thus finishes the proof. \(\square \)
Lemma 5.3
The following forms are trivial on \(\mathbb {R}^{n}\times \mathbb {R}^n\):
-
(a)
\(-\theta _{r,k}d\alpha _K+(-1)^n\sigma _{r,k}\alpha _K\),
-
(b)
\(\delta _{r,k}\nu _K+(-1)^n\sigma _{r,k}\alpha _{\overline{K}}\),
-
(c)
\(\theta _{r,k}d\alpha _{\overline{K}}\gamma _K\),
-
(d)
\(\sigma _{r,k}\gamma _{\overline{K}}\),
-
(e)
\(\delta _{r,k}\gamma _{\overline{K}}\),
-
(f)
\(\delta _{r,k}\nu _K\gamma _K-\theta _{r,k}d\alpha _{\overline{K}}\gamma _{\overline{K}}\),
-
(g)
\(\delta _{r,k}\nu _L\gamma _L+(-1)^n\sigma _{r,k}\alpha _L\gamma _L\),
-
(h)
\(\delta _{r,k}\nu _K\gamma _L-\theta _{r,k}d\alpha _L\gamma _{\overline{K}}\),
-
(i)
\(\theta _{r,k}d\alpha _L\gamma _L\),
-
(j)
\(\left( \delta _{r,k}\nu _L-\theta _{r,k}d\alpha _L+(-1)^n\sigma _{r,k}\alpha _L\right) \gamma _K-\theta _{r,k}d\alpha _{\overline{K}}\gamma _L\).
Proof
Obviously, each \(\beta \in \Omega ^*(\mathbb {R}^n\times \mathbb {R}^n)\) is trivial if and only if \(\beta \otimes \Theta _1=0\). This rule will be used throughout the proof.
-
(a)
For each \(i\in K\) one has
$$\begin{aligned}&\overline{\zeta _K}(\overline{d\zeta _K})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}d\zeta _{\overline{i}}dz_i\\&\quad =\overline{\zeta _K}(\overline{d\zeta _{K_i}})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}d\zeta _{\overline{i}}dz_i\\&\quad =(\zeta _{\overline{i}}\otimes dz_i)(\overline{d\zeta _{K_i}})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}d\zeta _{\overline{i}}dz_i\\&\quad =(-1)^n(d\zeta _{\overline{i}}\otimes dz_i)(\overline{d\zeta _{K_i}})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}\zeta _{\overline{i}}dz_i\\&\quad =(-1)^n(\overline{d\zeta _K})^{[k]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}\zeta _{\overline{i}}dz_i \end{aligned}$$and the claim follows by summing over i.
-
(b)
According to (17), for any \(i\in K\) we have
$$\begin{aligned}&(\overline{d\zeta _K})^{[k]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}\zeta _{i}dz_{\overline{i}}\\&\quad =(\overline{d\zeta _K})^{[k]}(\zeta _{\overline{i}}\otimes dz_{\overline{i}})(d\zeta _L)^{[n-r-k]}(dz_L)^{[r-k]}(dz_{\overline{K}_{\overline{i}}})^{[k-1]}\zeta _{i}dz_{\overline{i}}\\&\quad =(-1)^{n+k-1}(\overline{d\zeta _K})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_L)^{[r-k]}(dz_{\overline{K}_{\overline{i}}})^{[k-1]}(dz_{\overline{i}}\otimes dz_{\overline{i}})\zeta _{i}\zeta _{\overline{i}}\\&\quad =(-1)^{n+k-1}(\overline{d\zeta _K})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_L)^{[r-k]}(dz_{\overline{K}})^{[k]}\zeta _{i}\zeta _{\overline{i}}\\&\quad =(-1)^{n-1}(d\zeta _{\overline{K}})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_J)^{[r-k]}(\overline{dz_K})^{[k]}\zeta _{i}\zeta _{\overline{i}}\\&\quad =(-1)^{n-1}(d\zeta _J)^{[n-r]}(dz_J)^{[r-k]}(\overline{dz_K})^{[k]}\zeta _{i}\zeta _{\overline{i}} \end{aligned}$$and the claim follows by summing over i.
-
(c)
An immediate consequence of (21).
-
(d)
Obvious.
-
(e)
This follows at once when one employs the relation (16).
-
(f)
Using (15) for the first equality and (16) for the last, for any \(i\in K\) we have
$$\begin{aligned}&\overline{\zeta _K}(\overline{d\zeta _K})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}d\zeta _{i}dz_{\overline{i}}\gamma _{\overline{K}} \\&\quad =(-1)^n(\overline{d\zeta _K})^{[k]}\zeta _J(d\zeta _L)^{[n-r-k]}(dz_{J_{\overline{i}}})^{[r-1]}d\zeta _{i}dz_{\overline{i}}\nu _K\\&\quad =(-1)^n(\overline{d\zeta _K})^{[k]}(\zeta _{\overline{i}}\otimes dz_{\overline{i}})(d\zeta _L)^{[n-r-k]}(dz_{J_{\overline{i}}})^{[r-1]}d\zeta _{i}dz_{\overline{i}}\nu _K\\&\quad =(-1)^k(\overline{d\zeta _K})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_J)^{[r]}\zeta _{\overline{i}}d\zeta _{i}\nu _K\\&\quad =(-1)^k(\overline{d\zeta _K})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_L)^{[r-k]}(dz_{\overline{K}})^{[k]}\zeta _{\overline{i}}d\zeta _{i}\nu _K\\&\quad =(d\zeta _J)^{[n-r]}(dz_J)^{[r-k]}(\overline{dz_K})^{[k]}\zeta _{\overline{i}}d\zeta _{i}\nu _K. \end{aligned}$$ -
(g)
From (17), (18), (19), and (16) we infer
$$\begin{aligned}&(\overline{d\zeta _K})^{[k]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}\alpha _L\\&\quad \equiv (-1)^{n+1}(d\zeta _J)^{[n-r]}(dz_J)^{[r-k]}(\overline{dz_K})^{[k]}\nu _L\mod \gamma _L \end{aligned}$$which is clearly equivalent to the claim.
-
(h)
Using (15) and (20) for the first equality, we obtain
$$\begin{aligned}&\overline{\zeta _K}(\overline{d\zeta _K})^{[k-1]}\zeta _J(d\zeta _J)^{[n-r-k]}(dz_J)^{[r-1]}d\alpha _L\gamma _{\overline{K}}\\&\quad =(-1)^k(\overline{d\zeta _K})^{[k]}(d\zeta _J)^{[n-r-k]}(dz_J)^{[r]}\nu _K\gamma _L\\&\quad =(-1)^k(\overline{d\zeta _K})^{[k]}(d\zeta _L)^{[n-r-k]}(dz_L)^{[r-k]}(dz_{\overline{K}})^k\nu _K\gamma _L, \end{aligned}$$which by (16) implies the claim.
-
(i)
An immediate consequence of (20).
-
(j)
First, according to (20) and (14),
$$\begin{aligned} \theta _{r,k}d\alpha _L\otimes \Theta _1&=(-1)^{n-k}\overline{\zeta _K}(\overline{d\zeta _K})^{[k-1]}(\overline{d\zeta _L})^{[n-r-k]}(dz_J)^{[r]}\gamma _L\\&\quad +(-1)^{n+1}(\overline{d\zeta _K})^{[k]}\zeta _{\overline{K}}(\overline{d\zeta _L})^{[n-r-k-1]}(dz_J)^{[r]}\gamma _L. \end{aligned}$$Second, using (16)–(19) we infer
$$\begin{aligned}&\big [(-1)^n\sigma _{r,k}\alpha _L+\delta _{r,k}\nu _L\big ]\otimes \Theta _1\\&\quad =(-1)^{n+1}(\overline{d\zeta _K})^{[k]}\zeta _{\overline{K}}(\overline{d\zeta _L})^{[n-r-k-1]}(dz_L)^{[r-k+1]}(dz_{\overline{K}})^{[k-1]}\gamma _L\\&\qquad +(-1)^{n+1}(\overline{d\zeta _K})^{[k]}\zeta _L(\overline{d\zeta _L})^{[n-r-k-1]}(dz_L)^{[r-k]}(dz_{\overline{K}})^{[k]}\gamma _L. \end{aligned}$$Finally, by (21),
$$\begin{aligned} \theta _{r,k}d\alpha _{\overline{K}}\otimes \Theta _1&=(-1)^{n-k}\overline{\zeta _K}(\overline{d\zeta _K})^{[k-1]}(d\zeta _L)^{[n-r-k]}(dz_J)^{[r]}\gamma _K\\&\quad +(-1)^n(\overline{d\zeta _K})^{[k]}\zeta _L(d\zeta _L)^{[n-r-k-1]}(dz_J)^{[r]}\gamma _K. \end{aligned}$$Multiplying by \(\gamma _K\) and \(\gamma _L\), respectively, and summing everything up, the claim easily follows. \(\square \)
Corollary 5.4
The following relation holds in \( \Omega ^*(S\mathbb {R}^n)\):
Proof
Notice that \(\alpha =\alpha _K+\alpha _{\overline{K}}+\alpha _L\), analogously for \(\gamma \), and \(\nu =2\nu _K+\nu _L\). Consequently, one immediately verifies that the form
belongs to the ideal of \( \Omega ^*(\mathbb {R}^n\times \mathbb {R}^n)\) generated by items (a)–(j) of Lemma 5.3. According to the statement of the lemma, it must be trivial, which by Lemma 4.1 implies the claim if we take into account that \(\nu =1\) on \(S\mathbb {R}^n\). \(\square \)
Let us show four more auxiliary identities before we will finally proceed to the proof of the theorem.
Proposition 5.5
One has
Proof
(23) and (26) follow by direct computation. To prove (24), observe that
Finally, as for (25), one has
\(\square \)
Proof of Theorem 5.1
Using the relations (22)–(25), we compute
The first term in the resulting expression clearly vanishes and the claim follows at once from (26). \(\square \)
It remains to determine the action of D on the highest weight vectors \(\omega _{l,-l,m}\). However, this turns out to be an easy consequence of a special case of Theorem 5.1. In analogy with the previous definitions, let us denote
Corollary 5.6
Assume \(n=2l\). For any \(m\ge 2\) one has
Proof
Let \(e_1,\dots ,e_n\) be the canonical basis of \(\mathbb {R}^n\) and let \(R\in {{\,\textrm{O}\,}}(n)\) be the reflection in the hyperplane \(e_n^\perp \). Observe that \(R^*\omega _{l,l,m}=\omega _{l,-l,m}\), \(R^*\theta _{l,l}=\omega _{l,-l}\), \(R^*\sigma _{l,l}=\sigma _{l,-l}\), \(R^*\tau _{l,l}=\tau _{l,-l}\), and \(R^*\alpha =\alpha \). Thus, since both d and D commute with \(R^*\), the claim follows at once from Theorem 5.1. \(\square \)
6 The Alesker–Poincaré pairing
The goal of this section is to evaluate the Alesker–Poincaré pairing of the highest weight vectors constructed above. It will again suffice to consider only the valuations \(\phi _{r,k,m}\) for \(k\ge 1\) inasmuch as the remaining cases (\(n=2l\) and \(k=-l\)) can be then easily deduced by applying a hyperplane reflection. In this connection, let us keep the assumption (13) throughout the entire section again.
Theorem 6.1
One has
where
Proceeding similarly as in the previous section, we will first take the advantage of the double-form formalism in order to prove the important auxiliary relation (48) below. Its proof, again, splits further into a series of simpler statements with which we will start.
Proposition 6.2
One has
Proof
See the proof of (14). \(\square \)
Proposition 6.3
One has
Proof
(31) and (32) follow from (10) and (17), respectively. (33) follows from
using (27). \(\square \)
Proposition 6.4
One has
Remark 6.5
Observe that if \(k=n-r\), both sides of (38) as well as of (39) are zero by convention.
Proof
(34) and (35) are analogous to (18) and (19), respectively. (36) is straightforward. To prove (37), observe first
Second, since \(\overline{L}=L\), for any \(I\subset L\) with \(|I|=r-k\) there are \(\varepsilon _1,\varepsilon _2\in \{\pm 1\}\) with
and therefore,
Since there are \({ \left( {\begin{array}{c} {n-2k} \\ {r-k} \\ \end{array} } \right) }\) subsets of L of cardinality \(r-k\), the claim follows. As for (38), repeating the preceding argument shows that for any \(i\in L\) with \(i\ne l+1\) one has
while, if \(n=2l+1\),
Altogether, for any \(i\in L\),
The proof of (39) is completely similar. \(\square \)
Proposition 6.6
One has
Proof
(40) and (41) are obvious. (42) follows at once from (33). (43) further follows from (31) and (29); (44) similarly from (32) and (30); and (45) from (33) and (30). For (46), one uses (32), (34), and (35). Finally, (47) follows easily from (33) and (34). \(\square \)
Let us define
where \(\widehat{d\xi _i}\) means that \(d\xi _i \) is omitted. Restricted to \(S\mathbb {R}^n\), this form is the product of the standard Riemannian volume forms on \(\mathbb {R}^n\) and \(S^{n-1}\) (cf. [Lee13, Example 15.22]).
Lemma 6.7
One has
where
Proof
In view of Theorem 5.1, consider on \(\mathbb {R}^n\times \mathbb {R}^n\) the form
where
We will treat the two summands in (49) separately. Let us begin with the one proportional to \(\sigma _{n-r,k}\). Note that according to (40) and (41),
Using (43), (44), (36), and (37), we get
According to (31), (44), (36), and (38),
Similarly, according to (32), (43), (36) and (39),
Finally, from (31), (46), (36), and (38) we infer
Altogether, since \(\nu =2\nu _K+\nu _L\), one has
Proceeding to the second term of (48), by (40), (42), and (47), we have
According to (43), (45), (36), and (37),
By (31), (45), (36), and (38),
Further, using (33) and (43), we infer that for degree reasons
Finally, according to (47), (28), (31), (36), and (38),
Together, therefore,
Plugging (50) and (51) back in (49), we get (on \(\mathbb {R}^n\times \mathbb {R}^n\))
Further, using
Lemma 4.1, and the fact that \(\nu =1\) on \(S\mathbb {R}^n\), we conclude that (on \(S\mathbb {R}^n\))
which exactly was to be shown. \(\square \)
Remark 6.8
Observe that in particular \(D\omega _{n-r,k,m}\ne 0\).
Proof of Theorem 6.1
By Theorem 3.5, Proposition 3.7, and Lemma 6.7,
After change of variables and using \( s_a= \frac{2\pi }{a-1} s_{a-2}\) we have
and hence the claim follows at once from \(s_{a-1}=av_a\). \(\square \)
7 Pullback and pushforward
The goal of this section is to show that our choice of highest weight vectors in \({{\,\textrm{Val}\,}}(\mathbb {R}^n)\) is compatible with the natural transition between different dimensions, more precisely, with pullback and pushforward under inclusions \(\mathbb {R}^{n}\rightarrow \mathbb {R}^N\) and projections \(\mathbb {R}^N\rightarrow \mathbb {R}^n\), respectively.
Throughout the section we will assume \(n\ge 3\) and identify \(\mathbb {R}^n=\mathbb {R}^{n-1}\oplus \langle e_n\rangle \). To avoid any confusion, \(N^{(d)}(K)\) denotes the normal cycle with respect to the d-dimensional ambient space and similarly the superscript (d) is used to highlight the dimension valuations and (double) forms are considered in.
7.1 Pullback.
Let \(\iota :\mathbb {R}^{n-1} \rightarrow \mathbb {R}^n\) be the inclusion.
Theorem 7.1
For any \(r,k,m\in \mathbb {N}\) with \(r \le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
For the proof, we will consider the manifold \(M=\mathbb {R}^{n-1}\times \mathbb {R}^{n-1}\times [-\pi /2, \pi /2]\) and the map \(F:M \rightarrow \mathbb {R}^n\times \mathbb {R}^n\) given by
Lemma 7.2
If \(k<\frac{n}{2}\), then
If \(k=\frac{n}{2}\), then
Proof
Straightforward. \(\square \)
Proof of Theorem 7.1
Let \(K\in {\mathcal {K}}(\mathbb {R}^{n-1})\) have smooth boundary and positive Gauss curvature. Denote
Then, as currents,
It is obvious if \(r<n-1\) and follows from Proposition 4.3 and from our assumption \(n>2\) if \(r=n-1\) that \( \int _{N_0^\pm }\omega _{r,k,m}^{(n)}=0\). Consequently,
In particular, \(\iota ^*\phi _{r,k,m}^{(n)}=0\) if \(r=n-1\). We will assume from now on that \(r<n-1\).
Assume moreover \(k<\frac{n}{2} \). If we write \(F^* d\zeta _J^{(n)}= \cos \vartheta d\zeta _J^{(n-1)} + \upsilon \), then
with
Observe also that
Using Lemma 7.2, we thus compute
and hence we get
For \(k=n-r\) we thus have \(\iota ^*\phi _{r,k,m}^{(n)}=0\), see Remark 4.4(b). For \(k<n-r\), however,
i.e., \(\phi _{r,k,m}^{(n)}(\iota K)=\phi _{r,k,m}^{(n-1)}(K)\). Since any convex body can be approximated by smooth, positively curved ones, the claim follows.
Let us consider the remaining case \(k=\frac{n}{2}\). Similarly as above, writing
we have
Using Proposition 4.3 and Lemma 7.2, we obtain
Hence
and therefore
\(\square \)
7.2 Pushforward.
Let \(\pi :\mathbb {R}^n \rightarrow \mathbb {R}^{n-1}\) be the orthogonal projection.
Theorem 7.3
For any \(r,k,m\in \mathbb {N}\) with \(r \le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
Before carrying out the explicit computation, let us first prove a general lemma. For its proof we will consider the family of maps \(f_t:\mathbb {R}^n\times \mathbb {R}^n \rightarrow \mathbb {R}^n\times \mathbb {R}^n\), \(t\in \mathbb {R}\),
Similarly as above we also define \(M=\mathbb {R}^{n-1}\times \mathbb {R}^{n-1}\times \mathbb {R}\) and \(F:M \rightarrow \mathbb {R}^n\times \mathbb {R}^n\) by
Lemma 7.4
Let \(\phi \in {{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\) be represented by a translation-invariant form \(\omega \in \Omega ^{n-1}(S\mathbb {R}^n)^{{{\,\textrm{tr}\,}}}\). Then \(\pi _* \phi \) is represented by \( (-1)^n \iota ^* \left( i_{\frac{\partial }{\partial x_n}} \omega \right) \).
Proof
Take any \(K\in {\mathcal {K}}(\mathbb {R}^{n-1})\) with smooth boundary and positive Gauss curvature and denote
First, the normal cycle of \(K_t=\iota K+t[0,e_n]\) splits into the following sum of currents:
Second, the translation invariance of \(\omega \) implies in particular \(f_t^*\omega =\omega \). Third, we clearly have \(F^*dx_i=\iota ^*dx_i\) for \(1\le i<n\), \(F^*dx_n=dt\), and \(F^*d\xi _i=\iota ^*d\xi _i\) for \(1\le i\le n\). Altogether, we conclude that
Since by definition \((\pi _* \phi )(K) =\left. \frac{d}{dt}\right| _{t=0}\phi (K_t)\), \(K\in {\mathcal {K}}(\mathbb {R}^{n-1})\), the claim follows by approximation. \(\square \)
Lemma 7.5
If \(k<\frac{n}{2}\), then
and
If \(k=\frac{n}{2}\), then
and
Proof
Straightforward. \(\square \)
Given a double form \(\omega \) and a smooth vector field X on the same manifold, the contraction \(i_X \omega \) acts by definition on the first factor. If \(\omega \) is a (p, q)-form, then \(i_X(\omega \theta )= (i_X\omega )\theta + (-1)^p \omega (i_X \theta )\). In particular, if \(\omega \) is a (1, 1)-form, then
Proof of Theorem 7.3
Let us first assume \(k<\frac{n}{2} \). For \(k<r\), by Lemma 7.5 and (53), we obtain
where \(\widetilde{c}_n\) is given by (52). By Lemma 7.4, for any \(K\in {\mathcal {K}}(\mathbb {R}^{n-1})\) we thus have
i.e., \(\pi _*\phi _{r,k,m}^{(n)}=\phi _{r-1,k,m}^{(n-1)}\) which exactly was to be shown. If \(k=r\), the same computation clearly gives zero.
Assume now that \(k=\frac{n}{2} \). Then, by Lemma 7.5,
Thus we have
as claimed. \(\square \)
8 The Fourier transform
In the section to follow, an explicit description of the Fourier transform in terms of its action on the highest weight vectors is presented. Let us fix an inner product to identify \(V\cong V^*\cong \mathbb {R}^n\) and \({{\,\textrm{Dens}\,}}(V)\cong {{\,\textrm{Dens}\,}}(V^*)\cong \mathbb {C}\). Then \(\mathbb {F}\) is a linear operator on \({{\,\textrm{Val}\,}}^\infty (\mathbb {R}^n)\), commuting with the group \({{\,\textrm{O}\,}}(n)\).
Theorem 8.1
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
If \(n=2l\), then further
Since equivariant maps take highest weight vectors to highest weight vectors of the same weight, according to Theorem 3.1 there are constants \(f_{r,k,m}^{(n)}\in \mathbb {C}\) such that
where the notation is kept from §7. We first prove Theorem 8.1 in a special case.
Proposition 8.2
For any \(m\ge 2\) one has
Proof
Put \(\rho _1=\xi _1dx_2-\xi _2dx_1\in \Omega ^1(S\mathbb {R}^2)\). It is easily verified (e.g., by approximation by polygons) that
holds for any \(K\in {\mathcal {K}}(\mathbb {R}^2)\) and \(h\in C^\infty (S^1)\), cf. [BH18]. Since
we have on \(S\mathbb {R}^2\)
and consequently, for some \(c_m\in \mathbb {C}\),
Let \(J:S^1 \rightarrow S^1\) be the counter-clockwise rotation by \(\pi /2\). If m is even, the function \(\zeta _{\overline{1}}^{m}:S^1 \rightarrow \mathbb {C}\) is even and
If m is odd, then \(\zeta _{\overline{1}}^{m}\) is odd and antiholomorphic and
Now the claim follows at once from the explicit description of the Fourier transform in \(\mathbb {R}^2\) given in §3.4.1 above. \(\square \)
Lemma 8.3
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-2\), \(k\le \min \{r,n-r-1\}\), and \(m\ge 2\),
Moreover, if n is even, then for any \(l\ge 2\) and \(m\ge 2\),
Proof
Identifying \(V\cong V^*\cong \mathbb {R}^n\), the inclusion \(\iota :\mathbb {R}^{n-1} \rightarrow \mathbb {R}^n\) and the orthogonal projection \(\pi :\mathbb {R}^n \rightarrow \mathbb {R}^{n-1}\) are dual to each other. Consequently, Theorems 7.3, 3.12, and 7.1 imply, respectively,
Similarly,
\(\square \)
Proof of Theorem 8.1
We will first show (54), using induction on n. For \(n=2\), the claim was proven in Proposition 8.2 above. Assume thus (54) holds for \(n-1\ge 2\). First, by the first part of Lemma 8.3 we have
for \( r\le n-2\) and \( k\le \min \{r,n-1-r\}\). Second, using the relation
which follows directly from Theorem 3.11 (c), we extend the validity of (56) to \(r=n-1\) and \(k\le \min \{r,n-r\}\); indeed, in this case one necessarily has \(k=1\) but \(f_{1,1,m}^{(n)}\) was already covered by (56). Similarly, if \(k=n-r<r\) for some \(r\le n-2\), one has \(n-r\le n-2\) and \(k\le \min \{n-r,n-(n-r)-1\}\) and so we can compute \(f^{(n)}_{r,k,m}=(-1)^m\big (f_{n-r,k,m}^{(n)}\big )^{-1}\). Finally, \(k=n-r=r\) implies \(n=2l\) and \(r=k=l\), and Lemma 8.3 therefore finishes the proof of (54).
If \(n=2l\), (55) follows from (54) by applying the reflection \(R\in {{\,\textrm{O}\,}}(n)\) from the proof of Corollary 5.6. \(\square \)
9 The Lefschetz operator \(\Lambda \)
In the section to follow, the action of the operator \(\Lambda \) defined by (5) will be determined explicitly on the highest weight vectors. This result will later be used to prove that \(\Lambda \) satisfies the hard Lefschetz theorem and Hodge–Riemann relations. Our starting point is Proposition 3.8.
Theorem 9.1
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
If \(n=2l\), then for \(m\ge 2\),
Proposition 9.2
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
Proof
(57) is obvious. (58) follows easily by differentiating (25) and using (57). To show (59), observe that \(k<r\) implies \(r\ge 2\). Then on the one hand we have \(d\sigma _{r,k}=(n-r-k+1)\delta _{r-1,k}\) by (16) and (17); on the other hand, by (23) and (24), \(d\omega _{r-1,k,m}=(n+m-r-1)\zeta _{\overline{1}}^{m-2}\delta _{r-1,k}\). Finally, (60) follows at once from (17). \(\square \)
Lemma 9.3
For any \(r,k,m\in \mathbb {N}\) with \(r\le n-1\), \(k\le \min \{r,n-r\}\), and \(m\ge 2\),
where \(T= \sum _{i\in {\mathcal {I}}} \zeta _i \frac{\partial }{\partial z_i}\) is the Reeb vector field.
Proof
First of all, from Cartan’s magic formula and the fact that \(d\circ D=0\) we infer \({\mathcal {L}}_T(D\omega _{r,k,m})=d(\iota _T D\omega _{r,k,m})\). Further, observe that
and
note that in view of (14) this identity continues to hold for \(k=r\) if we set \(\theta _{k-1,k}=0\). Using this together with (26) and the second part of Theorem 5.1, we obtain
where
Consequently, using Proposition 9.2, we have
and, for \(k<r\),
Now the claim follows from the first part of Theorem 5.1. \(\square \)
If \(n=2l\), then the analogous statement for the remaining highest weight vectors can be further deduced from Lemma 9.3, just like Corollary 5.6 was deduced from Theorem 5.1; namely,
Corollary 9.4
For any \(m\ge 2\) one has
Proof
The reflection R considered in the proof of Corollary 5.6 clearly commutes with \({\mathcal {L}}_T\) and hence the claim follows at once from the second part of Lemma 9.3. \(\square \)
Proof of Theorem 9.1
The operator D commutes with pullback along contactomorphisms, in particular with \({\mathcal {L}}_T\). Consequently, if \(k<r\), the first part of Lemma 9.3 together with (2) and Theorem 3.3 implies that the form
defines the zero valuation and hence the claim follows from Proposition 3.8. The remaining cases follow similarly from the second part of Lemma 9.3 and Corollary 9.4, respectively. \(\square \)
10 Hard Lefschetz theorem and Hodge–Riemann relations
Finally, using the description of the operator
obtained in the previous section, we will give here a new proof of the hard Lefschetz theorem and prove the Hodge–Riemann relations for valuations, for Euclidean balls as reference bodies. Since our proofs rely profusely on the Alesker–Bernig–Schuster decomposition theorem, let us recall that the seeming dependence of this result on the hard Lefschetz theorem can be easily removed, see Remark 3.10.
Theorem 10.1
Let \(0\le r\le \lfloor \frac{n}{2}\rfloor \). Then the following properties hold:
-
(a)
Hard Lefschetz theorem. The map \({{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n)\rightarrow {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\) given by
$$\begin{aligned} \phi \mapsto \phi *(\mu _{n-1})^{n-2r} \end{aligned}$$is an isomorphism of topological vector spaces.
-
(b)
Hodge–Riemann relations. The sesquilinear form
$$\begin{aligned} Q(\phi ,\psi )=(-1)^r\,\phi *\overline{\psi }*(\mu _{n-1})^{n-2r} \end{aligned}$$is positive definite on
$$\begin{aligned} P_{n-r}=\left\{ \phi \in {{\,\textrm{Val}\,}}^\infty _{n-r}(\mathbb {R}^n)\mid \phi *\left( \mu _{n-1}\right) ^{n-2r+1}=0\right\} . \end{aligned}$$
Proof
If \(r=0\), both items follow at once from (6) so we will assume \(r>0\).
-
(a)
Injectivity follows from Theorem 3.1 and Theorem 9.1. To prove surjectivity, it suffices to show that the map
$$\begin{aligned} S_r=\Lambda ^{n-2r}\circ \mathbb {F}:{{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n) \rightarrow {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n) \end{aligned}$$is surjective. To this end, observe that \(S_r\) commutes with \({{\,\textrm{SO}\,}}(n)\) and let \(e_{r,k,m}\in \mathbb {C}\), for \(k\le \min \{r,n-r\}\) and \(m\ge 2\), denote its eigenvalues, i.e.,
$$\begin{aligned} S_r\phi _{r,k,m}=e_{r,k,m}\phi _{r,k,m}. \end{aligned}$$Since \(v_m/v_{m-1} = O(m^{1/2})\) as \(m\rightarrow \infty \), Theorems 8.1 and 9.1 imply \(|e_{r,k,m}|= O(m^{n/2-r})\). Consequently, by Theorem 2.2, given \(\phi \in {{\,\textrm{Val}\,}}_r^\infty (\mathbb {R}^n)\) the sum
$$\begin{aligned} \sum _{\lambda \in \Lambda _r} e_{r,k,m}^{-1} \pi _{\lambda } (\phi ) \end{aligned}$$converges to some \(\phi _0\in {{\,\textrm{Val}\,}}^{\infty }_r(\mathbb {R}^n)\). By linearity and continuity of \(S_r\) and by Theorem 2.1, this valuation satisfies \(S_r\phi _0=\phi \) which proves \(S_r\) is surjective. Using the open mapping theorem, we conclude that \(\Lambda ^{n-2r}\) is an isomorphism of topological vector spaces.
-
(b)
Clearly the form Q and the subspace \(P_{n-r}\subset {{\,\textrm{Val}\,}}_{n-r}^\infty (\mathbb {R}^n)\) are both \({{\,\textrm{O}\,}}(n)\) invariant for so is the valuation \(\mu _{n-1}\) and the convolution commutes with the action of \({{\,\textrm{O}\,}}(n)\). By Theorems 3.1 and 9.1, \(P_{n-r}\) decomposes into \({{\,\textrm{SO}\,}}(n)\)-types given by the following highest weights:
$$\begin{aligned} \Pi _r=\{(\lambda _1,\dots ,\lambda _l)\in \Lambda _r\mid \lambda _r\ne 0\}, \end{aligned}$$each occurring with multiplicity one. Using Proposition 2.3, by linearity and continuity it thus suffices to show \( Q(\phi _\lambda ,\phi _\lambda )>0\) for any \(\lambda \in \Pi _r\) and some vector \(\phi _\lambda \in P_{n-r}\) from the irreducible subspace corresponding to \(\lambda \). Fix \(m\ge 2\). First,
$$\begin{aligned} Q(\phi _{n-r,r,m},\phi _{n-r,r,m})>0 \end{aligned}$$by Theorem 9.1 and Theorem 6.1. Second, if \(n=2l\) and \(r=l\), applying the reflection \(R\in {{\,\textrm{O}\,}}(n)\) from the proof of Corollary 5.6 and using invariance of Q, we obtain
$$\begin{aligned} Q(\phi _{l,-l,m},\phi _{l,-l,m})>0. \square \end{aligned}$$
As pointed out to the authors by R. van Handel, combined with the authors’ previous result [KW22, Theorem 1.1], the special case of the Hodge–Riemann relations proven above implies a slightly more general statement. This provides further evidence in favour of Conjecture 1.2.
Proof of Corollary 1.6
Note that \(\mu _{D^{n}}\) is positively proportional to \(\mu _{n-1}\). Let \(\phi \in {{\,\textrm{Val}\,}}_{n-2}^\infty (\mathbb {R}^n)\) and suppose that
It was shown by the authors in [KW22, Theorem 1.1] that Conjecture 1.2 is true if \(r=1\). The hard Lefschetz theorem for \(r=1\) thus implies that there exists (a unique) \(\eta \in {{\,\textrm{Val}\,}}_{n-1}^\infty (\mathbb {R}^n)\) with
Clearly, \((\mu _{C_0})^{2}*(\mu _{n-1})^{n-3}*\eta =0\) and hence the Hodge–Riemann relations for \(r=1\) yield
(61) further implies \(\phi -\mu _{C_0}*\eta \in P_{n-2}\). Theorem 10.1 thus yields
Combining the two inequalities readily gives \(\phi * \overline{\phi }*(\mu _{n-1})^{n-4}\ge 0\).
If equality holds in the latter inequality, then equality also holds in the previous two inequalities. The equality conditions there imply, respectively, \(\phi -\mu _{C_0}*\eta =0\) and \(\eta =0\) and hence \(\phi =0\). \(\square \)
References
J. Abardia-Evéquoz and A. Bernig, Additive kinematic formulas for flag area measures, Math. Ann. (3-4) (2021), 1615–1652
S. Alesker, Description of translation invariant valuations on convex sets with solution of P. McMullen’s conjecture, Geom. Funct. Anal. (2)11 (2001), 244–272
S. Alesker, Hard Lefschetz theorem for valuations, complex integral geometry, and unitarily invariant valuations, J. Differential Geom. (1)63 (2003), 63–95
S. Alesker, The multiplicative structure on continuous polynomial valuations, Geom. Funct. Anal. (1)14 (2004), 1–26
S. Alesker, A Fourier-type transform on translation-invariant valuations on convex sets, Israel J. Math. 181 (2011), 189–294
S. Alesker, Introduction to the theory of valuations, American Mathematical Society, Providence, RI (2018)
S. Alesker, Kotrbatý’s theorem on valuations and geometric inequalities for convex bodies, Israel J. Math. (1)247 (2022), 361–378
S. Alesker and A. Bernig, The product on smooth and generalized valuations, Amer. J. Math. (2)134 (2012), 507–560
S. Alesker, A. Bernig, and F. E. Schuster, Harmonic analysis of translation invariant valuations, Geom. Funct. Anal. (4)21 (2011), 751–773
S. Alesker and J. H. G. Fu, Theory of valuations on manifolds. III. Multiplicative structure in the general case, Trans. Amer. Math. Soc. (4)360 (2008), 1951–1981
S. Alesker and J. H. G. Fu, Integral geometry and valuations, Birkhäuser/ Springer, Basel (2014)
A. Bernig, A product formula for valuations on manifolds with applications to the integral geometry of the quaternionic line, Comment. Math. Helv. (1)84 (2009), 1–19
A. Bernig and L. Bröcker, Valuations on manifolds and Rumin cohomology, J. Differential Geom. (3)75 (2007), 433–457
A. Bernig and J. H. G. Fu, Convolution of convex valuations, Geom. Dedicata 123 (2006), 153–169
A. Bernig and J. H. G. Fu, Hermitian integral geometry, Ann. of Math. (2) (2)173 (2011), 907–945
A. Bernig, J. H. G. Fu, and G. Solanes, Integral geometry of complex space forms, Geom. Funct. Anal. (2)24 (2014), 403–492
A. Bernig and D. Hug, Kinematic formulas for tensor valuations, J. Reine Angew. Math. 736 (2018), 141–191
A. Bernig and G. Solanes, Kinematic formulas on the quaternionic plane, Proc. Lond. Math. Soc. (3) (4)115 (2017), 725–762
W. Blaschke, Vorlesungen über Integralgeometrie, Deutscher Verlag der Wissenschaften, Berlin (1955)
K. J. Böröczky, M. Domokos, and G. Solanes, Dimension of the space of unitary equivariant translation invariant tensor valuations, J. Funct. Anal. (4)280 (2021), Paper No. 108862, 18
S.-S. Chern, On the kinematic formula in the Euclidean space of n dimensions, Amer. J. Math. 74 (1952), 227–236
A. Colesanti, M. Ludwig, and F. Mussnig, The Hadwiger theorem on convex functions. I, preprint (2020), arXiv:2009.03702
A. Colesanti, M. Ludwig, and F. Mussnig, A homogeneous decomposition theorem for valuations on convex functions, J. Funct. Anal. (5)279 (2020), 108573, 25
G. de Rham, Differentiable manifolds, Springer-Verlag, Berlin (1984)
D. Faifman, Crofton formulas and indefinite signature, Geom. Funct. Anal. (3)27 (2017), 489–540
D. Faifman, Contact integral geometry and the Heisenberg algebra, Geom. Topol. (6)23 (2019), 3041–3110
H. Federer, Curvature measures, Trans. Amer. Math. Soc. 93 (1959), 418–491
J. H. G. Fu, Structure of the unitary valuation algebra, J. Differential Geom. (3)72 (2006), 509–533
J. H. G. Fu, Integral geometric regularity, Tensor valuations and their applications in stochastic geometry and imaging, pp. 261–299, Lecture Notes in Math. 2177, Springer, Cham (2017)
P. Goodey, D. Hug, and W. Weil, Kinematic formulas for area measures, Indiana Univ. Math. J. (3)66 (2017), 997–1018
A. Gray, Volumes of tubes about Kähler submanifolds expressed in terms of Chern classes, J. Math. Soc. Japan (1)36 (1984), 23–35
A. Gray, Tubes, Birkhäuser Verlag, Basel (2004)
H. Groemer, Geometric applications of Fourier series and spherical harmonics, Cambridge University Press, Cambridge, (1996).
P. M. Gruber, Convex and discrete geometry, Springer, Berlin (2007)
H. Hadwiger, Vorlesungen über Inhalt, Oberfläche und Isoperimetrie, Springer-Verlag, Berlin-Göttingen-Heidelberg (1957)
D. Huybrechts, Complex geometry, Springer-Verlag, Berlin (2005)
D. A. Klain and G.-C. Rota, Introduction to geometric probability, Cambridge University Press, Cambridge (1997)
A. W. Knapp, Representation theory of semisimple groups, Princeton University Press, Princeton, NJ (2001)
J. Knoerr, Smooth valuations on convex functions, preprint (2020), arXiv:2006.12933
J. Knoerr, The support of dually epi-translation invariant valuations on convex functions, J. Funct. Anal. (5)281 (2021), Paper No. 109059, 52
J. Kotrbatý, On Hodge–Riemann relations for translation-invariant valuations, Adv. Math. 390 (2021), Paper No. 107914, 28
J. Kotrbatý and T. Wannerer, On mixed Hodge-Riemann relations for translation-invariant valuations and Aleksandrov-Fenchel inequalities, Commun. Contemp. Math. (7)24 (2022), Paper No. 2150049, 24
J. M. Lee, Introduction to smooth manifolds, Springer, New York (2013)
P. McMullen, Valuations and Euler-type relations on certain classes of convex polytopes, Proc. London Math. Soc. (3) (1)35 (1977), 113–135
P. McMullen, The polytope algebra, Adv. Math. (1)78 (1989), 76–130
P. McMullen, On simple polytopes, Invent. Math. (2)113 (1993), 419–444
M. Rumin, Formes différentielles sur les variétés de contact, J. Differential Geom. (2)39 (1994), 281–330
M. Saienko, Characterisation of valuations and curvature measures in Euclidean spaces, Israel J. Math. (2019), to appear
R. Schneider, Convex bodies: the Brunn–Minkowski theory, Cambridge University Press, Cambridge (2014)
M. Sugiura, Fourier series of smooth functions on compact Lie groups, Osaka Math. J. 8 (1971), 33–47
V. A. Timorin, An analogue of the Hodge–Riemann relations for simple convex polyhedra, Uspekhi Mat. Nauk 54 (1999), no. 2(326), 113–162
R. van Handel, Shephard’s inequalities, Hodge-Riemann relations, and a conjecture of Fedotov, preprint (2021), arXiv:2109.05169
T. Wannerer, Integral geometry of unitary area measures, Adv. Math. 263 (2014), 1–44
T. Wannerer, On the extendability by continuity of angular valuations on polytopes, J. Funct. Anal. (8)279 (2020), 108665, 25
G. Warner, Harmonic analysis on semi-simple Lie groups. I, Springer-Verlag, New York-Heidelberg (1972)
Acknowledgements
The authors wish to thank Ramon van Handel for illuminating discussions on the Hodge–Riemann relations and the anonymous referee for useful comments.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
JK was supported by FWF grant P31448-N35 and by DFG grant BE 2484/5-2. TW was supported by DFG grant WA 3510/3-1.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Kotrbatý, J., Wannerer, T. From harmonic analysis of translation-invariant valuations to geometric inequalities for convex bodies. Geom. Funct. Anal. 33, 541–592 (2023). https://doi.org/10.1007/s00039-023-00630-1
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00039-023-00630-1