• 検索結果がありません。

On Elliptic Product Formulas for Jackson Integrals Associated with Reduced Root Systems

N/A
N/A
Protected

Academic year: 2022

シェア "On Elliptic Product Formulas for Jackson Integrals Associated with Reduced Root Systems"

Copied!
12
0
0

読み込み中.... (全文を見る)

全文

(1)

°c 1998 Kluwer Academic Publishers. Manufactured in The Netherlands.

On Elliptic Product Formulas for Jackson Integrals Associated with Reduced Root Systems

KAZUHIKO AOMOTO

Graduate School of Mathematics, Nagoya University, Furo-cho 1, Chikusa-ku, Nagoya 464-8602, Japan Received December 13, 1995; Revised May 5, 1997

Abstract. In this note, we state certain product formulae for Jackson integrals associated with any root systems, involved in elliptic theta functions which appear as connection coefficients. The fomulae arise naturally in case of arbitrary root systems by extending the connection problem which has been investigated in [1, 4] in case of A type root system. This is also connected with the Macdonald-Morris constant term identity investigated by I. Cherednik [6], and K. Kadell [15] on the one hand, and of the Askey-Habsieger-Kadell’s q-Selberg integral formula and its extensions [4, 8, 12, 14, 15] on the other. This is also related with some of the results due to R.A. Gustafson [10, 11], although our integrands are different from his.

Keywords: elliptic theta function, Jackson integral, reduced root system, product formula, q-difference

1. The product formula

Let q be the elliptic modulus satisfying|q|<1. Let h and H be the n-dimensional Cartan subalgebra and its Cartan subgroup, associated with a simple Lie algebraGof rank n over C. Let hbe the dual of h.Let R+hbe the positive root system on h.Forα, βR+, we denote by(α, β)the inner product induced by the Killing form onG. We may identify the vector space hwith its dual h through the inner product as follows:

αhhαh,

such that(α, µ)=µ(hα)for anyµh.

Let X be the n-dimensional coweight lattice∼=Zn in h consisting of hh such that α(h)Z.We take a suitable basisχ1, . . . , χn,so that an arbitrary elementχof X can be described as a linear combination

χ= Xn

j=1

νjχj

for1, . . . , νn)Zn.h and H are isomorphic to the tensor products XC,and XC, respectively. X can also be embedded in the n-dimensional algebraic torus H (Cartan subgroup) isomorphic to(C)n.We denote this identification by

χXqχ =(qν1, . . . ,qνn)(C)n.

(2)

Letγ1, γ2, γ3, . . .be arbitrary complex numbers. Eachµhdefines a monomial tµ= t1µ(χ1)· · ·tnµ(χn)for t =(t1, . . . ,tn)H.We denote the two linear functionsλandλ0 as λ= 12P

α∈R+(1−2γ(α,α)andλ0=λ12P

α∈R+α, respectively.

We denote by Qχf(t)= f(qχt)the q-shift byχX for a function f(t)on H.We consider the q-multiplicative function8(t)and80(t)on H as

8(t)=tλ Y

α∈R+

(q1−γ(α,α)·tα)

(qγ(α,α)·tα) , (1.1)

80(t)=8(t) Y

α∈R+

(tα/2t−α/2). (1.2)

Here(x)=(x;q)denotes the infinite productQ

ν=0(1−xqν).It is known that there are at most 2 differentγ(α,α)which appear in the RHS of (1.1).

Then8(t), 80(t)have the quasi-symmetry with respect to the Weyl group W associated with the root system R+.

σ8(t)=8(σ1(t))=Uσ(t)8(t), (1.3)

σ 80(t)=801(t))=sgn(σ )Uσ(t)8(t), (1.4) where{Uσ(t)}σ∈Wdefines one cocycle on W with values in pseudo-constants i.e., q-periodic functions with respect to t .

Indeed Uσ(t)satisfies the properties

Uσ τ(t)=Uσ(t)·σUτ(t) forσ, τW,Ue(t)=1 (e=the identity) and QχUσ(t)=Uσ(t)for anyχX.

It is given in an explicit form as Uσ(t)= Y

α∈R+,σ (α)<0

½

t(2γ(α,α)1 θ(qγ(α,α)tα) θ(q1−γ(α,α)tα)

¾

, (1.5)

whereθ(x)denotes the Jacobi elliptic theta functionθ(x)=(x)·(q/x)·(q). We now consider the following Jackson integral

J(ξ)= Z

[0,ξ∞]q

80(t)$, µ

$ =dqt1

t1 ∧ · · · ∧dqtn tn

=(1−q)nX

χ∈X

80(qχξ), (1.6)

whereξ denotes an arbitrary point of H. Assume now thatλ(χ)+12P

α∈R+α(χ) >0 for allχ 6=0 of the fundamental domain 1+in h defined by the inequalitiesα(χ)≥0 forαR+.

(3)

Then (1.6) is summable on X1+becauseQ

α∈R+{(q1−γ(α,α)tα)/(qγ(α,α)tα)}is bound- ed on1+.Hence, (1.6) is also summable, in view of the quasi-symmetry (1.4) of80(t). It is important to note that this property does not depend on the choice ofξ, whenceξ−λJ(ξ) becomes a meromorphic function ofξ(C)n.

Then by definition J(ξ)is a q-periodic function ofξ on H : QχJ(ξ)=J(ξ) forχX.

The formula which we want to propose is the following.

Proposition 1 J(ξ)can be described as J(ξ)=C1ξλ0 Y

α∈R+

θ(qξα)

θ(q1(α,α)ξα), (1.7)

(we shall denote byψ˜(ξ)the term in the RHS divided by C1in the sequel) or equivalently J(ξ)=C2

X

σ∈W

sgnσ·Uσ(ξ)1·σ ψ(ξ), (1.8)

where C1and C2are constants with respect toξ.Letψ(ξ)be a pseudo-constant defined as ψ(ξ)=ξλ0

Yn j=1

θ(qcj(ωj,ωj)ξωj)

θ(qγ(ωj,ωj)ξωj) , (1.9)

where{cj}nj=1 are uniquely determined by the expressionω=Pn

j=1cjωj with respect to the positive simple rootsj}nj=1such that the corresponding positive roots are exactly R+. Remark 1 Since the symmetric n-dimensional cohomology associated with the Jackson integrals (1.6) is one dimensional, it may be conjectured that C1 and C2can be written in product form by using q-gamma functions ofγ(α,α).The explicit forms are presented in [13]

but not yet proved except in the two-dimensional cases. In [17], Macdonald has given a formal proof of Ito’s formula in an entirely different way. In [10, 11] Gustafson obtains various product formulas under a slightly different situation from ours. In [7], van Diejen obtains a similar product formula for BCntype, by using Gustafson’s result.

Proof 1: J(ξ)satisfies the same quasi-symmetry as80(t):

σJ(ξ)=sgnσ·Uσ(ξ)·J(ξ), σW (1.10)

Since J(ξ)ξ−λis a meromorphic function on H with poles lying in the set:Q

α∈R+θ(q1(α,α) ξα)=0,J(ξ)can be written as

J(ξ)=ξλQ f(ξ)

α∈R+θ(q1(α,α)ξα), (1.11)

(4)

where f(ξ) denotes a quasi-periodic function ofξ in the sense that Qχf(ξ)/f(ξ) is a monomial inξ for anyχX,and satisfies the skew-symmetry: σf(ξ) =sgnσf(ξ)for σW.This implies f(ξ)vanishes ifξα =1,q±1,q±2, . . .Hence, f(ξ)is divided out by the productQ

α∈R+θ(qξα)and also by the productξλ0Q

α∈R+θ(qξα): J(ξ)=ξλ0g(ξ) Y

α∈R+

θ(qξα)

θ(q1(α,α)ξα), (1.12)

where g(ξ) denotes a holomorphic function on H.Since g(ξ) is q-periodic, it must be constant. And the formula (1.7) follows.

To prove (1.8), we denote byψ(ξ)˜ the sumP

σ∈WsgnσUσ1(ξ)·σψ(ξ).Since Uσ(t) has the expression (1.5), the poles ofψ(ξ)˜ lie in the set{ξ ∈ H;Q

α∈R+θ(q1(α,α)ξα)=0}.

Moreover, it satisfies the same quasi-symmetry as in (1.10). Hence, it can be expressed as in the RHS of (1.8). It is sufficient to prove that it does not vanish identically. We evaluate the residues ofψ(ξ)˜ at the points of the equationsθ(qγjj)ξωj)=0, say, of the equations

qγjj)ξωj =1 for 1≤ jn. (1.13)

These points appear only for the residues of the summandψ(ξ)itself in the sumψ(ξ).˜ In fact,σ(ωj)=ωj,1 ≤ jn if and only ifσ=the identity.Since they do not vanish identically,ψ(ξ)˜ does not vanish identically. Letξ =ζ be a point satisfying this system of equations. Then

Resξψ(ξ)˜ =Resξψ(ξ)=ζλ0 Qn

j=1θ(qcj) θ0(1)n .

Equation (1.8) is thus proved. 2

2. A-type root system

In the sequelν1, ν2, ν3, . . .will denote integers.

At first we can take8(t)in a slightly more general way than (1.1), namely we take8(t) as

8(t)=t1α1· · ·tnαn Yn

j=1

(tj) (qβtj)

Y

1i<jn

(q1−γtj/ti)

(qγtj/ti) (2.1)

forαj, β, γC such thatαj=α1+(j−1)(1−2γ ), and 80(t)=8(t)D(t), for D(t)= Y

1i<jn

(titj). (2.2)

(5)

X consists of the lattice points x =1, . . . , νn)Zn.In this case, by the same argument as in the preceding section, the sum (1.6) is summable provided

−β > α1+n−1>0, −β > α1+(n−1)(2−2γ ) >0.

The Weyl group W is isomorphic to the symmetric group Sn of nth degree and the one-cocycle{Uσ(t)}is given as

Uσ(t)= Y

i<j−1(i)>σ−1(j)

µtj ti

2γ1 θ(qγtj/ti)

θ(q1−γtj/ti) (2.3)

for any permutationσ among the n letters{1,2, . . . ,n}.

As a special case of Jackson integrals, we takeξ =ξF for ξ1=q, ξ2 =q1, . . . , ξn =q1+(n1.

The integral J over [0, ξF∞]q is done only over the sethξFiconsisting of the points t such that t1 =q11,t2/t1 =qγ2, . . . ,tn/tn1 =qγn for eachνjZ0.Fiis clearly a subset of [0, ξF∞]q.We call it “α-stable cycle”, since the absolute value|tα|is maximal atξFand smaller elsewhere inhξFithan the value atξF.(Here the terminology “stable” is used as a discrete analog of the one in case of ordinary integrals.)

Hence, the substitution of the point ξF into 80(t)gives the asymptotic behavior of J =J(α1)forα1→ +∞as

J =qAn Yn

j=1

0q+1+(j−1)γ )0q(jγ )

0q(γ ) (1+O(qα1)) (2.4)

for An =Pn

j=1j+nj)[1+(j−1], where0q(u)denotes the q-gamma function (1−q)1u(q)/(qu).

On the other hand, we can takeξ =ηF for

ξ1=q−β, ξ2=q−β−γ, . . . , ξn =q−β−(n1.

The Jackson integral J over [0, ηF∞]q is meaningless, since80(t)has poles there. We denote byhηFithe set of points t such that

t1=q−β−ν1, t2/t1=q−γ−ν2, . . . ,tn/tn1=q−γ−νn

forνjZ0.Fi is a subset of [0, ηF∞]q.Then we can replace the sum (1.6) by the following regularization (see [1] or [4] for details):

reg Z

[0F]q

80(t)·$ µ

also denoted by Z

regFi80(t)·$

=X

νj0

Yn j=1

Restj/tj−1=q−γ−νj

·

80(t)dt1

t1

∧ · · · ∧dtn

tn

¸

, (2.5)

(6)

whereQn

j=1Res denotes the residue at each point ofhηFi. Here we have put t0=q−β+γ. Equation (2.5) is well defined because80(t)has simple poles at each point ofhηFi.

We define the quotient of theta functions ψn(ξ, ηF)=(1−q)n

Yn j=1

¡ξjqβ+(j1¢αj (q)3θ(qαj+···+αn+1ξjj1)

θ(qαj+···+αn+1)θ(qγ+1ξjj1) (2.6) forξ0=q−β+γ. It is a pseudo constant. The connection coefficient (or ratio) between the Jackson integrals over the cycles [0, ξ∞]qand reghηFiis denoted by([0, ξ∞]q: reghηFi)80.

Then we can give the following formula (see [4]):

([0, ξ∞]q : reghηFi)80=X

σ∈Sn

σψn(ξ, ηF)·sgnσ·Uσ(ξ)1, (2.7)

whereσϕ(ξ)denotesϕ(σ1(ξ)).

We evaluate the RHS of (2.7) in case whereβvanishes. We denote by Fn(ξ)=Fn(ξ, α1, γ ) the sum

X

σ∈Sn

sgnσ·Uσ(ξ)1·σ ( n

Y

j=1

ξαjjθ(qαj+···+αn+1ξjj1) θ(qγ+1ξjj1)

)

(2.8)

forξ0 =qγ.Then we have the product formula (see [2]):

Fn(ξ, α1, γ )= fn1, γ )·gn(ξ, α1, γ ), (2.9) where

fn1, γ )=(−1)n(n1)/2 Yn

j=1

q−(j1)2γ θ(qαj+···+αn+1)

θ(q11−(n+j2), (2.10) gn(ξ, α1, γ )=

Yn j=1

ξαj12(j1θ(qα1+1−(n1ξj) θ(qξj) · Y

1i<jn

ξiθ(ξji)

θ(qγ+1ξji). (2.11) Hence,ξj(j ≥1)being replaced byξjqβin (2.9),

([0, ξ∞]q : reghηFi)80

=(1−q)n Yn

j=1

(q)3qαjγ (j1)

θ(qαj+···+αn+1)· fn1, γ )·gn(qβξ, α1, γ ). (2.12) It is also possible to evaluate explicitlyR

Fi80(t)·$itself. In fact, the following formula is known (see [16]).

(7)

Proposition 2 Z

Fi80(t)·$

=qAn Yn j=1

0q+1+(j−1)γ )0q1+n−1−(n+j−2)γ )0q(jγ ) 0q(γ )0q1+β+n(nj)γ ) .

In particular, we have (hξFi: reghηFi)80

=(1−q)n(q)3n·qKn· Yn j=1

θ(qα1+β+2−(nj)θ(qγ)

θ(qα1+1−(n+j2)θ(qβ+2+(j1)θ(qjγ), (2.13) where Kndenotes

Kn = −2

3n(n−1)(2n−12n(n−1)

2 γ+1(n−1}

+1{1+(n−1}. (2.14)

Next, we take as8(t)the q-multiplicative function 8(t)=t1α1· · ·tnαn (t1· · ·tn)

(qβt1· · ·tn)

Y

1i<jn

(q1−γtj/ti)

(qγtj/ti) (2.15)

for αj = α1+(j −1)(1−2γ ).Let80(t)be as in (2.2). As in case of (2.1), J(ξ)is summable, provided

−β > α1+n−1>0, −β > α1+(n−1)(2−2γ ) >0.

X consists of the points x =1, . . . , νn)Zn.Then by the argument in Section 1, we have the formula (1.7) for

J(ξ)=C1

Yn j=1

ξαj12(j1

θ(qξ1· · ·ξn) θ(q1ξ1· · ·ξn)

Y

1i<jn

θ(qξji)

θ(q1ξji). (2.16)

3. Bn,Cn,Dn,G2,F4and E6 E8type root systems

For Bn-type, we take as8(t)the multiplicative function 8(t)=t1α1· · ·tnαn

Yn j=1

(tjq1−β) (tjqβ)

Y

1i<jn

(q1−γtj/ti)·(q1−γtitj)

(qγtj/ti)·(qγtitj) (3.1)

(8)

forαj =α1+(j −1)(1−2γ ), α1 = 12β.This is exactly the one obtained from (1.1) for the root system Bnwhen an explicit realization of positive roots is properly chosen for β = γ1, γ = γ2.The Weyl group W is isomorphic to the semi-direct product Zn2 ×Sn. 80(t)is given by

80(t) = 8(t)A¡

t1−(1/2)+nt2−(3/2)+n· · ·tn(1/2)¢

= (−1)n(n+1)/28(t)(t1· · ·tn)n+(1/2) Yn

j=1

(1−tj)

· Y

1i<jn

(titj)(1−titj), (3.2)

whereAdenotes the alternating sum with respect to the Weyl group W. X consists of the points x=1, . . . , νn)Zn.

Letψ˜(ξ)denote the productξλ0Q

α∈R+ θ(qξα)

θ(q1+γ(α,α)ξα) in the RHS of (1.7). Thenψ˜(ξ)is represented more concretely as

ψ˜(ξ)= Yn j=1

ξαjj−(j12)

θ(qξj) θ(qβ+1ξj)

Y

1i<jn

θ(qξji)θ(qξiξj)

θ(q1ξji)θ(q1ξiξj). (3.3) The explicit formula for C1has given by Ito [13] in case n=2.

Similarly for Cn-type, we take as8(t)the function 8(t)=t1α1· · ·tnαn

Yn j=1

¡q1−βt2j¢

¡

qβt2j¢

Y

1i<jn

(q1−γtj/ti)·(q1−γtitj)

(qγtj/ti)·(qγtitj) (3.4) forα1=1−2β, αj=α1+(j−1)(1−2γ ). This corresponds to (1.1) forγ=γ2, β=γ4. The Weyl group W is isomorphic to the one of Bn-type.80(t)is equal to8(t)A(t1nt2n1· · ·tn).

X consists of the points x=1, . . . , νn)Znor1+12, . . . , νn+12)Zn+(12, . . . ,12).

ψ˜(ξ)is then given as ψ˜(ξ)=

Yn j=1

ξαjjj

θ¡ 2j

¢ θ¡

qβ+1ξ2j¢ Y

1i<jn

θ(qξji)θ(qξiξj)

θ(q1ξji)θ(q1ξiξj). (3.5) For Dn-type we take as8(t)the function

8(t)= Yn j=1

tαjjj+1 Y

1i<jn

(q1−γtj/ti)·(q1−γtitj)

(qγtj/ti)·(qγtitj) (3.6) forα1 =0 andαj =α1+(j−1)(1−2γ ). This corresponds to (1.1) forγ =γ2. 80(t)is equal to8(t)A(t1nt2n1· · ·tn).The Weyl group W is isomorphic to the semi-direct product of Zn2andSn.X is the same as in case of Cntype. ψ˜(ξ)is then equal to

ψ˜(ξ)= Yn j=1

ξαjj−(j1)

Y

1i<jn

θ(qξji)θ(qξiξj)

θ(q1ξji)θ(q1ξiξj). (3.7)

(9)

For E8-type we can choose an orthonormal basis{²j}8j=1in h∼=R8such that the positive simple roots are

ω1=1

21+²8²2²3²4²5²6²7),

ω2=²1+²2, ω3=²2²1, ω4 =²3²2, ω5=²4²3, ω6=²5²4, ω7=²6²5, ω8=²7²6.

The positive roots are±²i+²j(1≤i < j ≤8)and128+P7

i=1ν(i)²i)where{ν(k)}7k=1 moves over the set±1 such that the number ofν(k)which are equal to−1 is even. Then 8(t)has the quasi-symmetry when and only when

αj =(j−1)(1−2γ )(1≤ j ≤7), α8=23(1−2γ ).

8(t)= Y8

j=1

tαjj Y

1i<j8

(q1−γtj/ti)(q1−γtitj)

(qγtj/ti)(qγtitj)

· Y

ν(k)=±1

¡q1−γt1ν(1)/2· · ·t7ν(7)/2t81/2¢

¡

qγt1ν(1)/2· · ·t7ν(7)/2t81/2¢

. (3.8)

The pairing betweenµ=P8

j=1µj²jhand x =(x1, . . . ,x8)h(∼=R8)is given by µ,x

X8 j=1

µjxj.

X is the eight-dimensional lattice in h consisiting of the points x =(x1, . . . ,x6,x7,x8)

h such thatα(x)≡0(Z)forαR+.In other words, xX consists of the points x such that xj =νj, (1≤ j ≤8)or xj =νj+12, (1≤ j ≤8)satisfying the equality

ν1+ · · · +ν8≡0(2).

ψ˜(ξ)is then equal to ψ˜(ξ)=

( 7 Y

j=1

ξj2(j1 )

ξ846γ Y

1i<j8

θ(qξji)θ(qξiξj) θ(q1ξji)θ(q1ξiξj)

· Y

ν(k)=±1

θ¡

1ν(1)/2· · ·ξ7ν(7)/2ξ81/2

¢ θ¡

qγ+1ξ1ν(1)/2· · ·ξ7ν(7)/2ξ81/2

¢. (3.9)

For E7-type, in terms of the above basis{²j}8j=1,the positive simple roots are ω1=1

21+²8²2²3²4²5²6²7),

ω2=²1+²2, ω3=²2²1, ω4 =²3²2, ω5=²4²3, ω6=²5²4, ω7=²6²5.

(10)

The positive roots are±²ij(1≤i < j ≤6),−²78and12(−²78+P6

i=1ν(i)²i) where{ν(k)}6k=1moves over the set±1 such that the number ofν(k)which are equal to−1 is odd. Then8(t)has the quasi-symmetry when and only when

αj =(j−1)(1−2γ )(1≤ j ≤6), −α7=α8= 17

2 (1−2γ ).

8(t)= (Y6

j=1

tαjj )

t8α8· ( Y

1i<j6

(q1−γtj/ti)(q1−γtitj) (qγtj/ti)(qγtitj)

)

·(q1−γt8) (qγt8)

· Y

ν(k)=±1

¡q1−γt1ν(1)/2· · ·t6ν(6)/2t81/2¢

¡

qγt1ν(1)/2· · ·t6ν(6)/2t81/2¢

. (3.10)

X consists of the points x such that either xj=νj(1≤ j≤8),x7=0 whereP6

j=1νj+ ν8≡0(2),or xj =νj+12(1≤ j ≤6),x7=0,x8=ν8whereP6

j=1νj+ν8 ≡1(2).

We normalizeξasξ7=1. ψ˜(ξ)is then equal to ψ˜(ξ)=

( 6 Y

j=1

ξj2(j1

)

·ξ817γ

Y

1i<j6

θ(qξji)θ(qξiξj)

θ(q1ξji)θ(q1ξiξj)· θ(qξ8) θ(qγ+1ξ8)

· Y

ν(k)=±1

θ¡

1ν(1)/2· · ·ξ6ν(6)/2ξ81/2¢ θ¡

qγ+1ξ1ν(1)/2· · ·ξ6ν(6)/2ξ81/2¢. (3.11) For E6-type, in R8the positive simple roots are

ω1=1

21+²8²2²3²4²5²6²7),

ω2=²1+²2, ω3=²2²1, ω4 =²3²2, ω5=²4²3, ω6=²5²4. The positive roots are±²i+²j(1 ≤ i < j ≤ 5)and 128²6²7+P5

i=1ν(i)²i) where{ν(k)}6k=1moves over the set±1 such that the number ofν(k)which are equal to−1 is even. Then8(t)has the quasi-symmetry when and only when

αj =(j−1)(1−2γ )(1≤ j ≤5), −α6= −α7=α8=4(1−2γ ).

8(t)= ( 5

Y

j=1

tαjj )

t8α8· ( Y

1i<j5

(q1−γtj/ti)(q1−γtitj)

(qγtj/ti)(qγtitj) )

· Y

ν(k)=±1

¡q1−γt1ν(1)/2· · ·t5ν(5)/2t81/2¢

¡

qγt1ν(1)/2· · ·t5ν(5)/2t81/2¢

. (3.12)

X consists of the points x such that either xj = νj(1 ≤ j ≤ 8),x6 = x7 =0 where P5

j=1νj+ν8 ≡0(2),or xj = νj + 12(1 ≤ j ≤ 5),x6 = x7 =0,x8 =ν812 where P5

j=1νj+ν8≡1(2).

参照

関連したドキュメント

Ion showed [2] that it can be used to give a quick and elegant proof of the classical fact that for a finite dimensional simple Lie algebra g, the partition formed by listing

[10] introduced the following two subclasses of the bi-univalent function class Σ and obtained non-sharp estimates on the first two Taylor-Maclaurin coefficients |a 2 | and |a 3 |

In this note we discuss the geometrical relationship between bi- Hamiltonian systems and bi-differential calculi, introduced by Dimakis and M¨

The main purpose of this paper is to establish explicit bounds on the general versions of (1.1) which can be used more effectively in the study of certain classes of

Abstract: The purpose of the present paper is to investigate some argument properties for certain analytic functions in the open unit disk associated with the convolution

• • Shifted staircase minus 2x2 square, plus outer Shifted staircase minus 2x2 square, plus outer

Srivastava, “Some inclusion properties of a certain family of integral operators,” Journal of Mathematical Analysis and Applications, vol.. Minda, “An internal

Finally remove a ball between consecutive integers, as in the proof of Proposition 4.6, but leave the number of balls in the far left and far right unchanged, to get a placement