A new, direct proof of the formulas for the conic intrinsic volumes of the Weyl chambers of types An−1, Bn and Dn is given. These formulas express the conic intrinsic volumes in terms of the Stirling numbers of the first kind and their B- and D-analogues. The proof involves an explicit determination of the internal and external angles of the faces of the Weyl chambers.
Stochastic geometrypolyhedral conesWeyl chambersconic intrinsic volumesexternal anglesinternal anglesStirling numbersrandom walks and bridges52A2260D0552A5511B7351F1552A39German Research Foundation390685587Both authors are supported by the German Research Foundation under Germany’s Excellence Strategy EXC 2044 – 390685587, Mathematics Münster: Dynamics – Geometry – Structure and by the DFG priority program SPP 2265 Random Geometric SystemsIntroduction
A polyhedral cone in the Euclidean space Rn is a set of solutions to a finite system of linear homogeneous inequalities. That is, a polyhedral cone C⊆Rn can be represented as
C={β∈Rn:⟨β,xi⟩≤0for alli=1,…,m}
for some finite collection of vectors x1,…,xm∈Rn, where ⟨·,·⟩ denotes the standard Euclidean scalar product. The fundamental Weyl chambers of types An−1, Bn and Dn are the polyhedral cones defined by
C(An−1):={β∈Rn:β1≥β2≥⋯≥βn},C(Bn):={β∈Rn:β1≥β2≥⋯≥βn≥0},C(Dn):={β∈Rn:β1≥β2≥⋯≥βn−1≥|βn|},
where β=(β1,…,βn) is the coordinate representation of β∈Rn.
In this paper, we shall be interested in the conic intrinsic volumes of the Weyl chambers. The conic intrinsic volumes of cones are analogues of the classical Euclidean intrinsic volumes of convex bodies in the setting of conic or spherical geometry. A recent increase of interest to conic intrinsic volumes is due to their relevance in convex optimization [6, 4, 5]. Let us briefly define the conic intrinsic volumes, referring to Section 2.1 for more details and to [25, Section 6.5] and [6, 7] for an extensive account of the theory. Given some point x∈Rn, the Euclidean projection ΠC(x) of x to a polyhedral cone C⊆Rn is the unique vector y∈C minimizing the Euclidean distance ‖x−y‖. For k∈{0,…,n}, the k-th conic intrinsic volume υk(C) of C is defined as the probability that the Euclidean projection ΠC(g) onto C of an n-dimensional standard Gaussian random vector g lies in the relative interior of a k-dimensional face of C (meaning that the projection is contained in this face but not in any face of smaller dimension). The conic intrinsic volumes of the Weyl chambers are given by the following theorem.
For alln∈{1,2,…}and allk∈{0,…,n}we haveυk(C(An−1))=nk1n!,υk(C(Bn))=B[n,k]2nn!,υk(C(Dn))=D[n,k]2n−1n!,where thenk’s denote the Stirling numbers of the first kind, theB[n,k]’s their B-analogues and theD[n,k]’s their D-analogues defined as the coefficients of the following polynomials:t(t+1)·…·(t+n−1)=∑k=0nnktk,(t+1)(t+3)·…·(t+2n−1)=∑k=0nB[n,k]tk,(t+1)(t+3)·…·(t+2n−3)(t+n−1)=∑k=0nD[n,k]tk.
The theorem is known and appeared in various, seemingly unrelated forms. One of the approaches to prove it is to relate the conic intrinsic volumes to the coefficients of the characteristic polynomial of the hyperplane arrangement generating the Weyl chambers; see [13, Theorems 3,4], [21, Theorem 5], [7, Section 6], [24], [20, Theorems 4.1 and 4.2]. The characteristic polynomials of Weyl arrangements can be computed by the finite field method; see Section 5.1 in [31] or Section 1.7.4 in [11].
On the other hand, the An−1-case of the formula (1.1) is closely related to a result, due to Sparre Andersen [30, Theorem 5], on the least concave majorant of a random walk. Consider a random walk given by Sk:=g1+⋯+gk, 1≤k≤n, and S0:=0, where the increments g1,…,gn are independent standard Gaussian random variables. (In fact, the result stated below holds, essentially, under the assumption of exchangeability of the increments). The graph of the random walk is the set in R2 consisting of the points (k,Sk), for k∈{0,1,…,n}. The least concave majorant of the random walk is the smallest concave function defined on [0,n] that stays above this graph. This function is piecewise linear. If Nn denotes the number of linearity segments of the least concave majorant, then the formula due to Sparre Andersen [30, Theorem 5] states that
P[Nn=k]=nk1n!,k∈{1,…,n}.
For further development of this probabilistic result we refer to [1–3]. Now, the projection of the point g=(g1,…,gn) on the Weyl chamber C(An−1) can explicitly be described in terms of the least concave majorant of the corresponding random walk, and, in particular, it is possible to check that the number of linearity segments of the majorant coincides with the dimension of the unique face F such that ΠC(An−1)(g) lies in the relative interior of F. With this knowledge, the An−1-case of the formula (1.1) can be viewed as a consequence of Sparre Andersen’s formula (1.2).
In the present paper, we shall give yet another proof of Theorem 1.1. The starting point of this proof is a formula, see (2.2) below, expressing the conic intrinsic volumes of a polyhedral cone in terms of the internal and external angles of its faces. Probably the main advantage of our proof is an explicit identification of all internal and external angles involved in this formula. The determination of the external angles is closely related to a result of Gao and Vitale [15] who computed the classical intrinsic volumes of the Schläfli orthoscheme which is defined as the simplex in Rn with the vertices 0 and e1+⋯+ei, 1≤i≤n, where e1,…,en denotes the standard orthonormal basis in Rn. Equivalently, this simplex is given by
{β∈Rn:1≥β1≥β2≥⋯≥βn≥0}.
Later, Gao [14] computed the intrinsic volumes of the simplex with the vertices e1+⋯+ei−in(e1+⋯+en), 1≤i≤n, which can be also given by
{β∈Rn:β1≥β2≥⋯≥βn,β1+⋯+βn=0,β1−βn≤1}.
These simplices are closely related to the Weyl chambers of types Bn and An−1, respectively. Both the computation of the external angles of the faces of the Weyl chambers (which follows the method of Gao and Vitale [15] and Gao [14]), and the computation of the internal angle sums proceed by re-arranging the solid angles under interest in such a way that they cover the whole space, from which we conclude that the sum of the angles is 1.
The rest of the paper is mostly devoted to the proof of Theorem 1.1.
PreliminariesConic intrinsic volumes and solid angles
In this section we collect some information on polyhedral cones (called just cones, for simplicity) that is needed in later proofs. A supporting hyperplane for a cone C⊆Rn is a linear hyperplane H with the property that C lies entirely in one of the closed half-spaces H+ and H− induced by H. A face of C is a set of the form F=C∩H, for a supporting hyperplane H, or the cone C itself. We denote by Fk(C) the set of all k-dimensional faces of C, for k∈{0,…,n}. The dimension of a face F is defined as the dimension of its linear hull, i.e. dimF=dimlin(F). Equivalently, the faces of C are obtained by replacing some of the half-spaces whose intersection defines the polyhedral cone by their boundaries and taking the intersection.
The dual cone (or the polar cone) of a cone C⊆Rn is defined as
C∘={x∈Rn:⟨x,y⟩≤0for ally∈C}.
For example, if C=L is a linear subspace, then C∘=L⊥ is its orthogonal complement. There is [7, Section 2.1] a one-to-one correspondence between the k-faces of C and the (n−k)-faces of C∘ via the bijective mapping
Fk(C)→Fn−k(C∘),F↦N(F,C),
where N(F,C):=(linF)⊥∩C∘ is called the normal face (or normal cone) of F with respect to C. A version of this one-to-one correspondence for polytopes can be found in [33, Corollary 2.14].
The positive hull of a finite set {x1,…,xm}⊆Rn is defined as the smallest cone containing this set, that is
pos{x1,…,xm}={λ1x1+⋯+λmxm:λ1≥0,…,λm≥0}.
We will repeatedly make use of the well-known duality relations
pos{x1,…,xm}∘=⋂i=1mxi−andpos{x1,…,xm}=(⋂i=1mxi−)∘
for x1,…,xm∈Rn and xi−:={v∈Rn:⟨v,xi⟩≤0}, i=1,…,m.
Now, we define the conic intrinsic volumes. The definition and further properties are taken from [7, Section 2.2] and [17, Section 2].
Let C⊆Rn be a polyhedral cone, and g be an n-dimensional standard Gaussian random vector. Then, for k∈{0,…,n}, the k-th conic intrinsic volume (or, for simplicity, just intrinsic volume) of C is defined by
υk(C):=∑F∈Fk(C)P(ΠC(g)∈relint(F)),
Here, ΠC denotes the Euclidean projection on C, that is ΠC(x) is the vector in C minimizing the Euclidean distance to x∈Rn. Also, relint(F) denotes the interior of F taken with respect to its linear hull lin(F) as an ambient space.
The solid angle (or just angle) of a cone C⊆Rn, denoted by α(C), is defined as
α(C)=P(Z∈C),
where Z is uniformly distributed on the unit sphere in the linear hull linC of C. In fact, for the distribution of Z we can choose any rotation invariant distribution living on the linear hull of C. For a k-dimensional cone C⊆Rn, k∈{1,…,n}, the k-th conic intrinsic volume coincides with the solid angle of C, that is
υk(C)=α(C).
The external angle of a cone C⊆Rn at a face F is defined as the solid angle of the normal face of F with respect to C, that is the external angle is α(N(F,C)). The internal angle of a face F at 0 is defined as α(F). For the purpose of this paper, we need the well-known expression of the intrinsic volumes in terms of internal and external angles:
υk(C)=∑F∈Fk(C)α(F)α(N(F,C)),k∈{0,…,n},
see Section 2.2 from [7]. We shall compute the intrinsic volumes of Weyl chambers by evaluating the internal and external angles at their faces.
Stirling numbers of the first kind and their generating functions
In this section we recall some facts on the Stirling numbers and their B- and D-analogues. These numbers, well known in combinatorics, appear in the formulas for the intrinsic volumes of the Weyl chambers. The (unsigned) Stirling numbers of the first kind are denoted by nk and defined as the coefficients of the polynomial
t(t+1)·…·(t+n−1)=∑k=1nnktk.
By convention, nk=0 for n∈N, k∉{1,…,n}. Equivalently, nk can be defined as the number of permutations of the set {1,…,n} having exactly k cycles. Other representations of the Stirling numbers of the first kind are known, e.g.,
nk=n!k!∑i1,…,ik∈Ni1+⋯+ik=n1i1i2·…·ik;
see [23, (1.9) and (1.13)]. We shall also need the generating functions of the Stirling numbers of the first kind:
∑n=0∞nktnn!=(−log(1−t))kk!and∑n=0∞∑k=0nnktnn!yk=(1−t)−y,
for all complex t such that |t|<1 and all y∈C. By convention, 00=1.
The B-analogues of the (signless) Stirling numbers of the first kind are denoted by B[n,k] and defined as the coefficients of the polynomial
(t+1)(t+3)·…·(t+2n−1)=∑k=0nB[n,k]tk.
Again, by convention, we put B[n,k]=0 for k∉{0,…,n}. These numbers appear as entry A028338 (or A039757 for the signed version) in the On-Line Encyclopedia of Integer Sequences [27] and also in [7, 9, 8, 10, 12, 13, 16, 19, 20, 22, 32]. As follows easily from their definition, the numbers B[n,k] satisfy the recurrence relation
B[n,k]=(2n−1)B[n−1,k]+B[n−1,k−1];
see [20, Section 2.2]. The following proposition collects some useful explicit formulas for B[n,k] including their exponential generating function.
The B-analoguesB[n,k]of the Stirling numbers of the first kind are explicitly given by the formulasB[n,k]=∑i=kn2n−iikni,B[n,k]=∑r=0n−k2n−k−2r2rrn−rkn!(n−r)!,fork∈{0,…,n}. The exponential generating function of(B[n,k])n,k≥0is given by∑n=0∞∑k=0nB[n,k]tnn!yk=(1−2t)−12(y+1)for all complex|t|<1/2andy∈C. By convention,B[0,0]=1.
In Entry A028338 of [27] the explicit formula (2.8) for the number B[n,k] in terms of the Stirling numbers of the first kind was stated by F. Woodhouse without a proof.
We start with the proof of (2.8). We want to check whether the numbers on the right-hand side of (2.8) coincide with the coefficients of the polynomial in (2.6). We have
∑k=0n∑i=kn2n−iiknitk=2n∑i=0n2−ini(∑k=0iiktk)=2n∑i=0nni(t+12)i=2n(t+12)(t+12+1)·…·(t+12+n−1)=(t+1)(t+3)·…·(t+2n−1)
using the Binomial Theorem and (2.3), which completes the proof of (2.8).
Next, we prove (2.10). By (2.6) and (2.3) we have
∑k=0nB[n,k]yk=2n(y+12)(y+12+1)·…·(y+12+n−1)=∑i=0n2nni(y+12)i.
Thus, the generating function is given by
∑n=0∞∑k=0nB[n,k]tnn!yk=∑n=0∞∑i=0nni(y+12)i(2t)nn!=(1−2t)−12(y+1),
where we used (2.5) in the last step.
At last, we use both formulas (2.8) and (2.10) to prove (2.9). To this end, we compare the generating functions of the sequences on both sides of (2.9). Consider the following sum:
ak(t):=∑n=k∞∑r=0n−k2n−k−2r2rrn−rkn!(n−r)!tnn!,k∈N0.
Then, we get
ak(t)=∑n=k∞∑r=0n−k2n−k−2r2rrn−rktn(n−r)!=∑r=0∞∑n=r+k∞2n−k−2r2rrn−rktn(n−r)!=∑r=0∞2−r−k2rrtr∑n=r+k∞n−rk(2t)n−r(n−r)!.
By shifting the index in the inner sum and using the generating function of the Stirling numbers of the first kind stated in (2.5), we can rewrite (2.11) to obtain
ak(t)=∑r=0∞2−r−k2rrtr∑n=k∞nk(2t)nn!=∑r=0∞2−r−k2rrtr(−log(1−2t))kk!=2−k(−log(1−2t))kk!∑r=0∞(t2)r2rr=2−k(−log(1−2t))kk!11−2t
for |t|<1/2. Thus, the generating function of the right-hand side of (2.9) is given by
∑k=0∞ak(t)yk=11−2texp(−y2log(1−2t))=(1−2t)−12(y+1),
which coincides with the generating function of (B[n,k])n,k≥0. This completes the proof. □
The D-analogues of the (signless) Stirling numbers of the first kind are denoted by D[n,k] and defined as the coefficients of the polynomial
(t+1)(t+3)·…·(t+2n−3)(t+n−1)=∑k=0nD[n,k]tk.
By convention, D[n,k]=0 for k∉{0,…,n}. The signed version of these numbers appears as entry A039762 in [27]. The D-analogues can be expressed through the B-analogues by
D[n,k]=(n−1)B[n−1,k]+B[n−1,k−1]=B[n,k]−nB[n−1,k];
see [20, Section 2.2] and use (2.7). In probabilistic terms, (2.3), (2.6) and (2.12) state that
P(Nn=k)=nk1n!,P(NnB=k)=B[n,k]2nn!,P(NnD=k)=D[n,k]2n−1n!,
for all k∈{0,…,n}, where Nn, NnB and NnD are random variables defined by
Nn=∑j=1nBern1j,NnB=∑j=1nBern12j,NnD=∑j=1n−1Bern12j+Bern1n,
and Bern(·) are independent Bernoulli variables with corresponding parameters. It is well known that Nn has the same distribution as the number of cycles in a uniform random permutation of n elements. This can be deduced from the Feller coupling or from the Chinese restaurant construction of the uniform random permutation, see [23, Section 3.1]. Similarly, it can be shown that NnB has the same distribution as the number of even cycles in a uniform random signed permutation of n elements, where a signed permutation is a pair (σ,ε) consisting of a permutation σ acting on {1,…,n} together with a vector of signs ε=(ε1,…,εn)∈{−1,+1}n, and a cycle i1↦⋯↦ik↦i1 of σ is called even if εi1·…·εik=+1. Observe that this gives a combinatorial interpretation of (2.8). More generally, it is known from [26, 5.3] or [18, p. 59, p. 63] that the numbers nk, B[n,k] and D[n,k] count the elements in the reflection group of the corresponding type whose space of invariant vectors is k-dimensional. For example, NnD has the same distribution as the number of even cycles in a random uniform D-permutation of n elements, where a D-permutation is a signed permutation (σ,ε) such that ε1·…·εn=1.
Conic intrinsic volumes of Weyl chambers
This section is dedicated to proving the formulas for the intrinsic volumes of the Weyl chambers stated in (1.1). To this end, we want to use the formula (2.2) and evaluate the internal and external angles of the faces of the Weyl chambers. The computation of the external angles relies on the ideas of Gao and Vitale [15] who derived the classical intrinsic volumes of the simplex (1.3).
We start with the simpler An−1 case. Recall the definition of the fundamental Weyl chamber of type An−1:
CA:=C(An−1)={β∈Rn:β1≥β2≥⋯≥βn}.
Then our first result is the following.
Fork=1,…,n, the k-th conic intrinsic volume of the Weyl chamber of typeAn−1is given byυk(C(An−1))=nk1n!,where the Stirling numbersnkare defined as in Section2.2.
Before we start with the proof of Theorem 3.1, we compute the solid angle of the cone whose points correspond to “bridges staying below zero”.
The solid angle of the coneDm:={x∈Rm:x1≤0,x1+x2≤0,…,x1+⋯+xm−1≤0,x1+⋯+xm=0}is given byα(Dm)=1m, form∈N.
The idea is to define m cones in Rm with the same solid angle as Dm, such that their union equals the linear subspace Hm:={x∈Rm:x1+⋯+xm=0} and their pairwise intersections are Lebesgue null sets (in the sense of the ambient linear subspace Hm). To this end, define
Dm(i)={(x1,…,xm)∈Rm:xi+1≤0,xi+1+xi+2≤0,…,xi+1+⋯+xm+x1+⋯+xi−1≤0,xi+1+⋯+xm+x1+⋯+xi=0},
for i∈{0,…,m−1}, where obviously Dm(0)=Dm. Observe that Dm(i)=ODm for a suitable permutation matrix O∈Rm×m, and thus, the rotation invariance of the spherical Lebesgue measure yields that the solid angle is the same for each Dm(i), 0≤i≤m. Furthermore, it is easy to see that Dm(i) can be written as follows:
Dm(i)={x∈Rm:x1+⋯+xm=0,x1+⋯+xi≥max{x1,x1+x2,…,x1+⋯+xm}}.
From this representation of Dm(i) we easily observe that for any point x∈Hm, there is a suitable index i∈{0,…,m−1}, namely where x1+⋯+xi is maximal, such that x∈Dm(i). On the other hand, if x is in the intersection of two cones Dm(i) and Dm(j), i≠j, then x belongs to {x∈Rm:x1+⋯+xm=0,x1+⋯+xi=x1+⋯+xj}, which is a subspace of dimension m−2, and thus, a Lebesgue null set (in the sense of the ambient subspace Hm). This yields that α(Dm)=1m. Note that a probabilistic version of this reasoning is due to Sparre Andersen [29] who proved that the probability that a random bridge (satisfying certain minor conditions) of length m stays below zero is 1/m. □
In order to prove Theorem 3.1, we want to evaluate the internal and external angles of the faces of CA. Observe that the faces of CA can be obtained by replacing some of the inequalities defining CA by the corresponding equalities. It follows that for 2≤k≤n, any k-face of CA is determined by some collection of indices 1≤l1<⋯<lk−1<n and given by
CA(l1,…,lk−1)={β∈Rn:β1=⋯=βl1≥βl1+1=⋯=βl2≥…≥βlk−1+1=⋯=βn}.
Then, (2.2) yields the formula
υk(CA)=∑F∈Fk(CA)α(F)α(N(F,CA))=∑1≤l1<⋯<lk−1≤n−1α(CA(l1,…,lk−1))α(N(CA(l1,…,lk−1),CA)).
Note that CA has no vertices and the only 1-face is {β∈Rn:β1=⋯=βn}, which corresponds to an empty collection of l’s.
External Angles. Let 1≤k≤n. Take some 1≤l1<⋯<lk−1<n. We start by computing the normal face N(CA(l1,…,lk−1),CA). Our method of proof partially relies on the approach of Gao and Vitale [15]. By definition, we have
N(CA(l1,…,lk−1),CA)=(linCA(l1,…,lk−1))⊥∩(CA)∘.
Using (2.1), we obtain
(CA)∘={x∈Rn:x1≥x2≥⋯≥xn}∘=pos{e1,e1+e2,…,e1+⋯+en,−e1−⋯−en}∘={x∈Rn:x1≤0,…,x1+⋯+xn−1≤0,x1+⋯+xn=0}.
Here, e1,…,en denotes the standard Euclidean orthonormal basis of Rn. Furthermore,
(linCA(l1,…,lk−1))⊥={x∈Rn:x1=⋯=xl1,xl1+1=⋯=xl2,…,xlk−1+1=⋯=xn}⊥={x∈Rn:x1+⋯+xl1=0,xl1+1+⋯+xl2=0,…,xlk−1+1+⋯+xn=0}.
The explicit representations of (CA)∘ and (linCA(l1,…,lk−1))⊥ imply the following representation of the normal face as the orthogonal product (or direct sum) of k cones:
N(CA(l1,…,lk−1),CA)=Dl1×Dl2−l1×⋯×Dlk−1−lk−2×Dn−lk−1,
where again
Dm:={x∈Rm:x1≤0,…,x1+⋯+xm−1≤0,x1+⋯+xm=0}.
In Lemma 3.2 we proved that α(Dm)=1/m holds for any m∈N. Since the normal cone N(CA(l1,…,lk−1),CA) is the orthogonal product of the k cones Dl1,Dl2−l1,…,Dn−lk−1, it follows that
α(N(CA(l1,…,lk−1),CA))=1l1(l2−l1)·…·(n−lk−1).
Internal Angles. Now, we need to compute the internal angles at the faces of CA. To this end, we consider the linear hull of CA(l1,…,lk−1) which forms a k-dimensional linear subspace of Rn. Recall that for 1≤l1<⋯<lk−1≤n−1 the linear hull of CA(l1,…,lk−1) is given by
linCA(l1,…,lk−1)={β∈Rn:β1=⋯=βl1,…,βlk−1+1=⋯=βn}.
Thus, the vectors yi given by
y1=1l1(1,…,1︷1,…,l1,0,…,0),y2=1l2−l1(0,…,0,1,…,1︷l1+1,…,l2,0,…,0),…,yk−1=1lk−1−lk−2(0,…,0,1,…,1︷lk−2+1,…,lk−1,0,…,0),yk=1n−lk−1(0,…,0,1,…,1)︷lk−1+1,…,n
form an orthonormal basis of linCA(l1,…,lk−1). For independent and standard Gaussian random variables ξ1,…,ξk, the random vector N:=ξ1y1+⋯+ξkyk is k-dimensional standard Gaussian distributed on the linear subspace linCA(l1,…,lk−1). Thus, we obtain for the angle of CA(l1,…,lk−1) the following formula:
α(CA(l1,…,lk−1))=P(N∈CA(l1,…,lk−1))=P(ξ1l1≥ξ2l2−l1≥⋯≥ξk−1lk−1−lk−2≥ξkn−lk−1).
Conic intrinsic volumes. Let 1≤k≤n. Using the formulas (2.2), (3.1) and (3.2) we obtain
υk(CA)=∑1≤l1<⋯<lk−1≤n−1α(CA(l1,…,lk−1))α(N(CA(l1,…,lk−1),CA))=∑1≤l1<⋯<lk−1≤n−1P(ξ1l1≥ξ2l2−l1≥⋯≥ξk−1lk−1−lk−2≥ξkn−lk−1)l1(l2−l1)·…·(lk−1−lk−2)(n−lk−1).
Defining i1=l1, i2=l2−l1,…, ik−1=lk−1−lk−2, ik=n−lk−1, we can change the summation indices in the above sum and obtain
υk(CA)=∑i1,…,ik∈Ni1+⋯+ik=n1i1i2·…·ikP(ξ1i1≥ξ2i2≥⋯≥ξk−1ik−1≥ξkik).
Now, fix a permutation π from the symmetric group Sk. For each tuple (i1,…,ik)∈Nk satisfying i1+⋯+ik=n, the tuple (iπ(1),…,iπ(k)) also satisfies iπ(1)+⋯+iπ(k)=n. Thus, the sum in (3.3) does not change if we replace the tuple (i1,…,ik) by (iπ(1),…,iπ(k)) inside the sum. It follows that
υk(CA)=1k!∑π∈Sk∑i1,…,ik∈Ni1+⋯+ik=nP(ξ1iπ(1)≥ξ2iπ(2)≥⋯≥ξkiπ(k))iπ(1)iπ(2)·…·iπ(k)=1k!∑i1,…,ik∈Ni1+⋯+ik=n1i1i2·…·ik∑π∈SkP(ξ1iπ(1)≥ξ2iπ(2)≥⋯≥ξkiπ(k))=1k!∑i1,…,ik∈Ni1+⋯+ik=n1i1i2·…·ik∑π∈SkP(ξπ(1)iπ(1)≥ξπ(2)iπ(2)≥⋯≥ξπ(k)iπ(k))=1k!∑i1,…,ik∈Ni1+⋯+ik=n1i1i2·…·ik,
where we used that ξ1,…,ξk are independent and identically distributed, hence exchangeable. Using the representation (2.4) of the Stirling numbers of the first kind, we obtain
υk(CA)=1k!nkk!n!=nk1n!,
which completes the proof. □
Now, we proceed with the Bn-case. Recall the definition of the fundamental Weyl chamber of type Bn:
CB:=C(Bn)={β∈Rn:β1≥β2≥⋯≥βn≥0}.
Then the analogue of Theorem 3.1 is the following.
Fork=0,1,…,n, the k-th conic intrinsic volume of the Weyl chamberC(Bn)is given byυk(C(Bn))=B[n,k]2nn!.
Similar to the An−1-case, we first need a formula for the solid angle of a cone whose points correspond to “walks staying below zero”.
The solid angle of the coneBm:={x∈Rm:x1≤0,x1+x2≤0,…,x1+⋯+xm≤0},is given byα(Bm)=2mm14m, form∈N.
Take an m-dimensional standard Gaussian random vector N=(N1,…,Nm). The solid angle of Bm coincides with the probability that N∈Bm. Using the well-known formula of Sparre Andersen [28] (which is valid in the special case when N1,…,Nm are independent and standard Gaussian variables) we obtain
α(Bm)=P[N∈Bm]=P[N1≤0,N1+N2≤0,…,N1+⋯+Nm≤0]=2mm122m,
which completes the proof. □
To obtain the faces of CB we have to replace some of the inequalities defining CB by the corresponding equalities. It follows that for 1≤k≤n any k-face of CB is determined by a collection of indices 1≤l1<⋯<lk≤n and given by
CB(l1,…,lk)={β∈Rn:β1=⋯=βl1≥⋯≥βlk−1+1=⋯=βlk≥βlk+1=⋯=βn=0}.
Then, (2.2) yields the formula
υk(CB)=∑1≤l1<⋯<lk≤nα(CB(l1,…,lk))α(N(CB(l1,…,lk),CB)).
External angles. The computation of the external angles for the Bn-case is similar to that for the An−1-case. Take some k∈{1,…,n} and 1≤l1<⋯<lk≤n. We start by computing the normal face N(CB(l1,…,lk),CB) of CB(l1,…,lk). First of all, we have
(CB)∘=pos{e1,e1+e2,…,e1+⋯+en}∘={x∈Rn:x1≤0,x1+x2≤0,…,x1+⋯+xn≤0}.
Since
(linCB(l1,…,lk))⊥={x∈Rn:x1=⋯=xl1,…,xlk−1+1=⋯=xlk,xlk+1=⋯=xn=0}⊥={x∈Rn:x1+⋯+xl1=0,…,xlk−1+1+⋯+xlk=0},
it follows that the normal cone N(CB(l1,…,lk),CB) can be written as the orthogonal product
N(CB(l1,…,lk),CB)=Dl1×Dl2−l1×⋯×Dlk−lk−1×Bn−lk,
where again
Dm={x∈Rm:x1≤0,…,x1+⋯+xm−1≤0,x1+⋯+xm=0},Bm={x∈Rm:x1≤0,x1+x2≤0,…,x1+⋯+xm≤0},B0:={0}.
Using Lemmas 3.2 and 3.4 and the orthogonal product structure of the normal face we arrive at
α(N(CB(l1,…,lk),CB))=2(n−lk)n−lkl1(l2−l1)·…·(lk−lk−1)22(n−lk).
Note that the above formula is also valid for the case k=0 where we have CB(l1,…,lk)={0}.
Internal angles. Now, we need to compute the internal angles for the faces of CB, that is, the solid angles α(CB(l1,…,lk)) for 1≤l1<⋯<lk≤n. Recall that the linear hull of CB(l1,…,lk) is the k-dimensional linear subspace given by
linCB(l1,…,lk)={x∈Rn:x1=⋯=xl1,xl1+1=⋯=xl2,…,xlk+1=⋯=xn=0}.
The following vectors y1,…,yk form an orthonormal basis of linCB(l1,…,lk):
y1=1l1(1,…,1︷1,…,l1,0,…,0),y2=1l2−l1(0,…,0,1,…,1︷l1+1,…,l2,0,…,0),…,yk=1lk−lk−1(0,…,0,1,…,1︷lk−1+1,…,lk,0,…,0).
For independent and standard Gaussian random variables ξ1,…,ξk, the random vector N:=ξ1y1+⋯+ξkyk is k-dimensional standard Gaussian on the linear subspace linCB(l1,…,lk). Thus, we obtain for the solid angle of CB(l1,…,lk) the following formula:
α(CB(l1,…,lk))=P(N∈CB(l1,…,lk))=P(ξ1l1≥ξ2l2−l1≥⋯≥ξklk−lk−1≥0).
Conic intrinsic volumes. Let 1≤k≤n. Using the formulas (3.5) and (3.4) for the internal and external angles, we have
υk(CB)=∑1≤l1<⋯<lk≤nα(CB(l1,…,lk))α(N(CB(l1,…,lk),CB))=∑1≤l1<⋯<lk≤n2(n−lk)n−lkP(ξ1l1≥ξ2l2−l1≥⋯≥ξklk−lk−1≥0)l1(l2−l1)·…·(lk−lk−1)22(n−lk).
Defining i1=l1, i2=l2−l1,…, ik=lk−lk−1, we can change the summation indices in the above sum and obtain
υk(CB)=∑i1,…,ik∈Ni1+⋯+ik≤n2(n−i1−⋯−ik)n−i1−⋯−ikP(ξ1i1≥ξ2i2≥⋯≥ξkik≥0)i1i2·…·ik22(n−i1−⋯−ik)=∑r=0n−k∑i1,…,ik∈Ni1+⋯+ik=n−r2rrP(ξ1i1≥ξ2i2≥⋯≥ξkik≥0)i1i2·…·ik22r.
Now, fix a number r∈{0,1,…,n−k}, a permutation π∈Sk and a vector of signs ε=(ε1,…,εk)∈{±1}k. For each tuple (i1,…,ik)∈Nk satisfying i1+⋯+ik=n−r, the tuple (iπ(1),…,iπ(k)) also satisfies iπ(1)+⋯+iπ(k)=n−r. Thus, the inner sum in (3.6) does not change if we replace the tuple (i1,…,ik) by (iπ(1),…,iπ(k)) inside the sum. Furthermore, the sum does not change if we additionally replace ξ1,…,ξk by ε1ξ1,…,εkξk, since ξ1,…,ξk are independent and standard normal. Thus, we obtain
υk(CB)=∑r=0n−k12kk!∑(ε,π)∈{±1}k×Sk∑i1,…,ik∈Ni1+⋯+ik=n−r2rr×P(ε1ξ1iπ(1)≥ε2ξ2iπ(2)≥⋯≥εkξkiπ(k)≥0)iπ(1)iπ(2)·…·iπ(k)22r=12kk!∑r=0n−k∑i1,…,ik∈Ni1+⋯+ik=n−r2rri1i2·…·ik22r×∑(ε,π)∈{±1}k×SkP(ε1ξ1iπ(1)≥⋯≥εkξkiπ(k)≥0)=12kk!∑r=0n−k∑i1,…,ik∈Ni1+⋯+ik=n−r2rri1i2·…·ik22r×∑(ε,π)∈{±1}k×SkP(ε1ξπ(1)iπ(1)≥⋯≥εkξπ(k)iπ(k)≥0)=12kk!∑r=0n−k∑i1,…,ik∈Ni1+⋯+ik=n−r2rri1i2·…·ik22r.
Using the representation (2.4) for the Stirling numbers of the first kind and the formula (2.9), we obtain
υk(CB)=∑r=0n−k12kk!2rr22rn−rkk!(n−r)!=∑r=0n−k2−k−2r2rrn−rk1(n−r)!=B[n,k]2nn!.
The case k=0 is easy since the only 0-face is the origin {0}, and thus
υ0(CB)=α({0})α(N({0},CB))=α((CB)∘)=α(Bn)=2nn122n=(2n−1)!!2nn!=B[n,0]2nn!,
where we applied the formula for α(Bn) stated in Lemma 3.4. □
At last, we can also compute the conic intrinsic volumes of the Weyl chambers of type Dn. These formulas follow from the intrinsic volumes of the type Bn chambers and additivity of the intrinsic volumes.
Fork=0,1,…,n, the k-th conic intrinsic volume of the Weyl chamber of typeDnis given byυk(C(Dn))=D[n,k]2n−1n!,where the numbersD[n,k]are given by (2.12).
Consider the Weyl chamber
C(Dn)={β∈Rn:β1≥⋯≥βn−1≥|βn|}=(C(Dn)∩{βn≥0})∪(C(Dn)∩{βn≤0})={β∈Rn:β1≥⋯≥βn−1≥βn≥0}∪{β∈Rn:β1≥⋯≥βn−1≥−βn≥0},
where the first set on the right-hand side is the Weyl chamber of type Bn, while the second set is isometric to it. Their intersection
{β∈Rn:β1≥⋯≥βn−1≥βn≥0}∩{β∈Rn:β1≥⋯≥βn−1≥−βn≥0}={β1≥⋯≥βn−1≥βn=0}
is a Weyl chamber of type Bn−1 if we identify Rn−1 and Rn−1×{0}. The conic intrinsic volumes υk are additive functionals, see [25, Theorem 6.5.2], and thus, we obtain
υk(C(Dn))=2υk({β∈Rn:β1≥⋯≥βn−1≥βn≥0})−υk({β1≥⋯≥βn−1≥βn=0})=2B[n,k]2nn!−B[n−1,k]2n−1(n−1)!=B[n,k]−nB[n−1,k]2n−1n!=D[n,k]2n−1n!.
Here, we used the formula for the intrinsic volumes of the Weyl chambers of type Bn from Theorem 3.3, and in the last step the relation (2.13). □
ReferencesAbramson, J., Pitman, J.: Concave majorants of random walks and related Poisson processes. Combin. Probab. Comput.20(5), 651–682 (2011). MR2825583. https://doi.org/10.1017/S0963548311000307Abramson, J., Pitman, J., Ross, N., Uribe Bravo, G.: Convex minorants of random walks and Lévy processes. Electron. Commun. Probab.16, 423–434 (2011). MR2831081. https://doi.org/10.1214/ECP.v16-1648Alsmeyer, G., Kabluchko, Z., Marynych, A., Vysotsky, V.: How long is the convex minorant of a one-dimensional random walk?Electron. J. Probab.25, 1–22 (2020). MR4147518. https://doi.org/10.1214/20-ejp497Amelunxen, D., Bürgisser, P.: Intrinsic volumes of symmetric cones and applications in convex programming. Math. Program.149(1-2, Ser. A), 105–130 (2015). MR3300458. https://doi.org/10.1007/s10107-013-0740-2Amelunxen, D., Bürgisser, P.: Probabilistic analysis of the Grassmann condition number. Found. Comput. Math.15(1), 3–51 (2015). MR3303690. https://doi.org/10.1007/s10208-013-9178-4Amelunxen, D., Lotz, M., McCoy, M.B., Tropp, J.A.: Living on the edge: phase transitions in convex programs with random data. Information and Inference3(3), 224–294 (2014). MR3311453. https://doi.org/10.1093/imaiai/iau005Amelunxen, D., Lotz, M.: Intrinsic volumes of polyhedral cones: A combinatorial perspective. Discrete Comput. Geom.58(2), 371–409 (2017). MR3679941. https://doi.org/10.1007/s00454-017-9904-9Bagno, E., Garber, D.: Signed partitions – A balls into urns approach. Preprint at arXiv:1903.02877 (2019). MR4408200Bagno, E., Biagioli, R., Garber, D.: Some identities involving second kind Stirling numbers of types B and D. Elect. J. Combin.26(3), 3–9 (2019). MR3982318. https://doi.org/10.37236/8703Bala, P.: A 3-parameter family of generalized Stirling numbers. Preprint at https://oeis.org/A143395/a143395.pdf (2015)Bóna, M. (ed.): Handbook of Enumerative Combinatorics. Discrete Mathematics and its Applications (Boca Raton). CRC Press, Boca Raton, FL (2015). MR3408702Dowling, T.A.: A class of geometric lattices based on finite groups. J. Combinatorial Theory Ser. B14, 61–86 (1973). MR0307951. https://doi.org/10.1016/s0095-8956(73)80007-3Drton, M., Klivans, C.J.: A geometric interpretation of the characteristic polynomial of reflection arrangements.Proc. Amer. Math. Soc.138(8), 2873–2887 (2010). MR2644900. https://doi.org/10.1090/S0002-9939-10-10369-4Gao, F.: The mean of a maximum likelihood estimator associated with the Brownian bridge. Electron. Comm. Probab.8, 1–5 (2003). MR1961284. https://doi.org/10.1214/ECP.v8-1064Gao, F., Vitale, R.A.: Intrinsic volumes of the Brownian motion body. Discrete Comput. Geom.26(1), 41–50 (2001). MR1832728. https://doi.org/10.1007/s00454-001-0023-1Henze, N.: Weitere Überraschungen im Zusammenhang mit dem Schnur-Orakel. Stochastik in der Schule33(3), 18–23 (2013)Hug, D., Schneider, R.: Random conical tessellations. Discrete Comput. Geom.56(2), 395–426 (2016). MR3530973. https://doi.org/10.1007/s00454-016-9788-0Humphreys, J.E.: Reflection Groups and Coxeter Groups. Cambridge Studies in Advanced Mathematics, vol. 29. Cambridge University Press, Cambridge (1990). https://doi.org/10.1017/CBO9780511623646Kabluchko, Z., Vysotsky, V., Zaporozhets, D.: Convex hulls of random walks: expected number of faces and face probabilities. Adv. Math.320, 595–629 (2017). MR3709116. https://doi.org/10.1016/j.aim.2017.09.002Kabluchko, Z., Vysotsky, V., Zaporozhets, D.: Convex hulls of random walks, hyperplane arrangements, and Weyl chambers. Geom. Funct. Anal.27(4), 880–918 (2017). MR3678504. https://doi.org/10.1007/s00039-017-0415-xKlivans, C.J., Swartz, E.: Projection volumes of hyperplane arrangements.Discrete Comput. Geom.46(3), 417–426 (2011). MR2826960. https://doi.org/10.1007/s00454-011-9363-7Lang, W.: On Sums of Powers of Arithmetic Progressions, and Generalized Stirling, Eulerian and Bernoulli numbers. Preprint at arXiv:1707.04451 (2017)Pitman, J.: Combinatorial Stochastic Processes. Lecture Notes in Mathematics. Springer (2006). MR2245368Schneider, R.: Combinatorial identities for polyhedral cones. St. Petersburg Math. J.29(1), 209–221 (2018). MR3660692. https://doi.org/10.1090/spmj/1489Schneider, R., Weil, W.: Stochastic and Integral Geometry. Springer (2008). MR2455326. https://doi.org/10.1007/978-3-540-78859-1Shephard, G.C., Todd, J.A.: Finite unitary reflection groups. Canad. J. Math.6, 274–304 (1954). MR0059914. https://doi.org/10.4153/cjm-1954-028-3Sloane (editor), N.J.A.: The On-Line Encyclopedia of Integer Sequences. https://oeis.orgSparre Andersen, E.: On the number of positive sums of random variables. Scandinavian Actuarial Journal1949(1), 27–36 (1949). MR0032115. https://doi.org/10.1080/03461238.1949.10419756Sparre Andersen, E.: On the fluctuations of sums of random variables.Math. Scand.1, 263–285 (1953). MR0058893. https://doi.org/10.7146/math.scand.a-10385Sparre Andersen, E.: On the fluctuations of sums of random variables II. Math. Scand.2, 195–223 (1954). MR0068154Stanley, R.P.: An introduction to hyperplane arrangements. In: Miller, E., Reiner, V., Sturmfels, B. (eds.) Geometric Combinatorics. IAS/Park City Mathematics Series, vol. 13, pp. 389–496. AMS (2007). MR2383131. https://doi.org/10.1090/pcms/013/08Suter, R.: Two analogues of a classical sequence. J. Integer Seq.3(1), 1–18 (2000). MR1750742Ziegler, G.M.: Lectures on Polytopes. Springer, New York (1995). MR1311028. https://doi.org/10.1007/978-1-4613-8431-1