partition negative weights - Research Institute for ...kenkyubu/bessatsu/open/B39/pdf/...184 Tatsuru Takakura partition function (or volume function) by its generating function. Such
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
RIMS Kôkyûroku BessatsuB39 (2013), 183−195
On vector partition functions with negative weights
By
Tatsuru Takakura *
Abstract
We introduce the notion of weights into the vector partition function and the volume
function associated to a sequence of vectors. We prove an explicit formula for the volume
function with possibly negative weights. It is a generalization of a formula given by Brion and
Vergne in [2].
§1. Introduction
The vector partition function and the volume function associated to a sequence
of vectors have been studied from various point of view. For instance, the Kostant
partition function for a root system, which is a typical example of vector partition
functions, plays an important role in representation theory of Lie groups. We note that
a vector partition function (resp. a volume function) counts the number of the lattice
points in (resp. measures the volume of) a certain polytope. Hence they are closelyrelated to combinatorics of convex polytopes. We refer to [2], [7], and [1] for known
results. For example, in [2] Brion and Vergne gave explicit closed formulas for vector
partition functions and volume functions.
In this paper, we introduce the notion of a vector partition function (or volume
function) with weights, where each weight is an integer. A positive weight merely
corresponds to the multiplicity of a vector in the given set of vectors used to define
the vector partition function (or volume function). The known results cited above are
also available in this positive weight case. On the other hand, the notion of negative
weights seems to be new, although it is quite natural when we characterize a vector
Received September 30, 2011. Revised December 16, 2011.
partition function (or volume function) by its generating function. Such negative weights
appear, for example, in the computation of the dimension of the invariant subspace
(V_{$\lambda$_{1}}\otimes\cdots\otimes V_{$\lambda$_{n}})^{G} in a tensor product representation, where V_{$\lambda$_{i}} is the irreducible
representation of a compact Lie group G with highest weight $\lambda$_{i} ,if some of $\lambda$_{i}(i=
1,
. . .
, n) lie on the boundary of the Weyl chamber (see, e.g., [5]).Our main result Theorem 4.1 gives explicit formulas for the volume functions with
possibly negative weights, which generalize some of the Brion‐Vergne formulas men‐
tioned above. It is stated as follows.
Theorem 4.1. Let \triangle=($\alpha$_{1}, \ldots, $\alpha$_{N}) , m=(m\mathrm{l}, . . . , m_{N}) be as in Denition3.1 and let M=m_{1}+\cdots+m_{N} . Suppose h is in a chamber $\gamma$ and y\in \mathbb{R}^{N} is generic.Then we have
where e^{v}(v\in $\Lambda$) are elements in the group ring \mathbb{Z}[ $\Lambda$] , obeying e^{v_{1}}e^{v_{2}}=e^{v_{1}+v_{2}}.
More generally, for y=(y_{1}, \ldots, y_{N})\in \mathbb{R}^{N} ,we define
\displaystyle \mathcal{P}_{\triangle}(y, $\lambda$):=\sum_{x_{1}$\alpha$_{1}+\cdots+x_{N}$\alpha$_{N}= $\lambda$}e^{-(x_{1}y_{1}+\cdots+x_{N}y_{N})},whose generating function is given by
\displaystyle \sum_{ $\lambda$}\mathcal{P}_{\triangle}(y, $\lambda$)e^{ $\lambda$}=\prod_{i=1}^{N}\frac{1}{1-e^{-y_{i}}e^{$\alpha$_{i}}}.By definition, we have \mathcal{P}_{\triangle}(0, $\lambda$)=\mathcal{P}_{\triangle}( $\lambda$) .
We call X(h) the partition polytope associated to \triangle . Note that it is compact and
\mathcal{P}_{\triangle}( $\lambda$) is the number of lattice points in it.
Next, we introduce continuous analogues of \mathcal{P}_{\triangle}( $\lambda$) and \mathcal{P}_{\triangle}(y, $\lambda$) . We normalize the
Lebesgue measure on E so that the volume of E/ $\Lambda$ is 1, and we consider the standard
Lebesgue measure on \mathbb{R}^{N} . They determine the Lebesgue measure ds on H_{\triangle}(h) . For
h\in E and y=(y_{1}, \ldots, y_{N})\in \mathbb{R}^{N} ,let us define
\mathcal{V}_{\triangle}(h) := volume of X_{\triangle}(h)=\displaystyle \int_{X_{\triangle}(h)}ds,\displaystyle \mathcal{V}_{\triangle}(y, h):=\int_{X_{\triangle}(h)}e^{-(x_{1}y_{1}+\cdots+x_{N}y_{N})}ds.
The function \mathcal{V}_{\triangle} is called the volume function (or asymptotic partition function) asso‐
ciated to \triangle.
186 Tatsuru Takakura
Remark 1. Under some conditions on \triangle and $\lambda$,
the partition polytope X_{\triangle}( $\lambda$)becomes an integral polytope and \mathcal{P}_{\triangle}(k\cdot $\lambda$) turns out to be a polynomial of k\in \mathbb{Z}_{\geq 0},which is called the Ehrhart polynomial. Then the coefficient of the top term of \mathcal{P}_{\triangle}(k\cdot $\lambda$)is equal to \mathcal{V}_{\triangle}( $\lambda$) . See, e.g., [7].
Moreover, \mathcal{P}_{\triangle}( $\lambda$) and \mathcal{V}_{\triangle}( $\lambda$) are related to certain invariants, the Riemann‐Roch
number and the volume, of the toric variety associated to the partition polytope X_{\triangle}( $\lambda$) .
§2.2. Brion‐Vergne formula
In [2], Brion and Vergne gave explicit closed formulas for \mathcal{P}_{\triangle}(y, $\lambda$) , \mathcal{V}_{\triangle}(y, h) ,and
\mathcal{V}_{\triangle}(h) . Let us recall those for \mathcal{V}_{\triangle}(y, h) and \mathcal{V}_{\triangle}(h) . In order to state them, we need
some notation.
A subset $\sigma$ of \{ 1, . . .
, N\} is called a basis of \triangle if the sequence ($\alpha$_{j})_{j\in $\sigma$} is a basis of
E . The set of all bases of \triangle is denoted by \mathcal{B}(\triangle) .
Let C(\triangle) :=\displaystyle \sum_{i=1}^{N}\mathbb{R}_{\geq 0}$\alpha$_{i} and let C( $\sigma$) :=\displaystyle \sum_{j\in $\sigma$}\mathbb{R}_{\geq 0}$\alpha$_{j} for $\sigma$\in \mathcal{B}(\triangle) . Consider the
subdivision of C() given by the intersections of the cones C( $\sigma$)( $\sigma$\in \mathcal{B}(\triangle)) . The
interior of a maximal cone of this subdivision is called a chamber. For a chamber $\gamma$ ,we
denote by \mathcal{B}(\triangle, $\gamma$) the set of all bases $\sigma$\in \mathcal{B}(\triangle) such that $\gamma$\subset C( $\sigma$) .
Let $\sigma$\in \mathcal{B}(\triangle) . We define a linear map v_{ $\sigma$} : E\rightarrow \mathbb{R}^{N} by v_{ $\sigma$}($\alpha$_{j})=w_{j} for j\in $\sigma$,where (wl, . . .
, w_{N} ) is the standard basis of \mathbb{R}^{N} . We denote by $\mu$( $\sigma$) the volume of the
parallelepiped
\displaystyle \{\sum_{j\in $\sigma$}t_{j}$\alpha$_{j}|0\leq t_{j}\leq 1 (j\in $\sigma$)\}Finally, for $\sigma$\in \mathcal{B}(\triangle) , j\in $\sigma$ and k\not\in $\sigma$ ,
we define a real number c_{jk}=c_{jk}^{ $\sigma$} by
$\alpha$_{k}=\displaystyle \sum_{j\in $\sigma$}c_{jk}$\alpha$_{j}.Remark 2. For h\in $\gamma$ ,
the set of vertices of the partition polytope X(h) coincides
with \{v_{ $\sigma$}(h)| $\sigma$\in \mathcal{B}(\triangle, $\gamma$ See [2, 3.1].
Now, their formulas are stated as follows.
Theorem 2.1 ([2]). Let $\gamma$ be a chamber and let h\in $\gamma$ . Suppose y\in \mathbb{R}^{N} is
for $\lambda$\in $\Lambda$ . Here M=m_{1}+\cdots+m_{N} . Then the generating function of \mathcal{P}_{\triangle}( $\lambda$;m) is given
by
(3.1) \displaystyle \sum_{ $\lambda$}\mathcal{P}_{\triangle}( $\lambda$;m)e^{ $\lambda$}=\frac{1}{\prod_{i=1}^{N}(1-e^{$\alpha$_{i}})^{m_{i}}}.If we set
for a fixed x\in \mathbb{R}_{\geq 0} ,then we have
\displaystyle \int_{ $\Sigma$(x)}dx_{1}\cdots dx_{m}=\frac{x^{m-1}}{(m-1)!}.Hence the volume function associated to \triangle and m should be defined by
the |m| ‐th derivative of the delta function with support at x=0 . (See, e.g., [3, 4].)We note that, in the situation above, a product of derivatives of delta functions, its
restriction to H_{\triangle}(h) ,and the integrals (3.2) and (3.4) over X_{\triangle} are well‐defined by
virtue of the theory of Sato hyperfunction ([4]).In conclusion, we have introduced the following definition.
Denition 3.1. Let \triangle=($\alpha$_{1}, \ldots, $\alpha$_{N}) be a sequence of vectors in $\Lambda$, lying in
an open half space of E and spanning E as vector space, and let m=(m_{1}, \ldots, m_{N})\in\mathbb{Z}^{N} . For $\lambda$\in $\Lambda$, h\in E ,
and \mathcal{V}_{\triangle}(y, h;m) respectively by (3.1), (3.2), (3.3), and (3.4). We call \mathcal{P}_{\triangle}( $\lambda$;m) (resp.\mathcal{V}_{\triangle}(h;m)) the vector partition function with weights (resp. the volume function with
weights).
§4. Result
Let y=(y_{1}, \ldots, y_{N})\in \mathbb{R}^{N} . For i=1,
. . .
,N
,we denote \displaystyle \partial_{i}=\frac{\partial}{\partial y_{i}} . Our main result
is the following, which generalizes the Brion‐Vergne formulas (2.1) and (2.2).
Theorem 4.1. Let \triangle=($\alpha$_{1}, \ldots, $\alpha$_{N}) , m=(m\mathrm{l}, . . . , m_{N}) be as in Denition3.1 and let M=m_{1}+\cdots+m_{N} . Suppose h is in a chamber $\gamma$ and y\in \mathbb{R}^{N} is generic.
On vector partition functions with negative weights 189
where if m_{i}\leq 0 ,we set \displaystyle \frac{\partial_{j}^{m_{j}-1}}{(m_{j}-1)!}=0.
A proof is given in the next section. Let us consider a simple example.
Example 4.2 (Volume function for A_{2} ). Let \triangle=($\alpha$_{1}, $\alpha$_{2}, $\alpha$_{3}) be the positiveroot system of type A_{2} ,
where $\alpha$_{3}=$\alpha$_{1}+$\alpha$_{2} . There are two chambers $\gamma$_{1}=\{p_{1}$\alpha$_{1}+p_{2}$\alpha$_{2}|p_{1}>p_{2}>0\} and $\gamma$_{2}=\{p_{1}$\alpha$_{1}+p_{2}$\alpha$_{2}|p_{2}>p_{1}>0\} in C(\triangle) . For simplicity,
suppose h=p_{1}$\alpha$_{1}+p_{2}$\alpha$_{2} is in $\gamma$_{1} . Then for m=(m_{1}, m_{2}, m_{3})\in \mathbb{Z}
\displaystyle \frac{\partial_{1}^{m_{1}-1}}{(m_{1}-1)!}\frac{\partial_{2}^{m_{2}-1}}{(m_{2}-1)!}\frac{(p_{1}y_{1}+p_{2}y_{2})^{M-2}}{(-y_{3}+y_{1}+y_{2})^{m_{3}}}+\frac{\partial_{1}^{m_{1}-1}}{(m_{1}-1)!}\frac{\partial_{3}^{m_{3}-1}}{(m_{3}-1)!}\frac{((p_{1}-p_{2})y_{1}+p_{2}y_{3})^{M-2}}{(-y_{2}-y_{1}+y_{3})^{m_{2}}}with M=m_{1}+m_{2}+m_{3} . It might be natural to ask if there is any relation between
our formula and the GKZ hypergeometric theory.
§5. Proof of Theorem 4.1
The formula (4.2) follows from (4.1) in the same way with the proof of (2.2) in [2,3.3]. Hence let us prove (4.1).
190 Tatsuru Takakura
Step 1. If all the m_{j}(j=1, \ldots, N) are positive, the proof is quite easy. In fact, by
applying \displaystyle \prod_{j=1}^{N}\frac{(-\partial_{j})^{m_{j}-1}}{(m_{j}-1)!} to the Brion‐Vergne formula ((2.1) in Theorem 2.1)
as required. Here note that if k\not\in $\sigma$ ,then \langle y, v_{ $\sigma$}(h)\rangle does not contain the variable y_{k}
and hence
\displaystyle \frac{(-\partial_{k})^{m_{k}-1}}{(m_{k}-1)!}(\frac{e^{-\langle y,v_{ $\sigma$}(h)\rangle}}{y_{k}-\sum_{j\in $\sigma$}c_{jk}y_{j}})=\frac{e^{-\langle y,v_{ $\sigma$}(h)\rangle}}{(y_{k}-\sum_{j\in $\sigma$}c_{jk}y_{j})^{m_{k}}}.Step 2. In order to treat general case, we proceed by induction on N=\#\triangle.
First, let N=1 . If m_{1}>0 , (4.1) holds by Step 1. If m_{1}\leq 0 ,then \mathcal{V}(y, h;m)=0
since X(h) and \{x_{1}=0\} ,the support of $\delta$^{(|m_{1}|)}(x_{1}) ,
are disjoint. On the other hand,(-\partial_{1})^{m_{1}-1}
the right hand side of (4.1) is also 0 by \overline{(m_{1}-1)!}=0.Next, let us suppose that our formula for \mathcal{V}_{\triangle}(y, h;m) holds for all \triangle and m with
\#\triangle\leq N-1 ,whether or not all of the m_{j} are positive.
Taking Step 1 into account, we consider the case where some of m_{j} are nonpositive.Without loss of generality, we may assume that m_{N}\leq 0 . Then since
(5.1) \displaystyle \mathcal{V}_{\triangle}(y, h;m)=\int_{X_{\triangle}(h)}(-\frac{\partial}{\partial x_{N}})^{|m_{N}|}(\prod_{j=1}^{N-1}\frac{x_{j}^{m_{j}-1}}{(m_{j}-1)!}e^{-\langle y,x\rangle})|_{x_{N}=0}ds.Let us fix a $\sigma$_{0}\in \mathcal{B}(\triangle, $\gamma$) such that N\not\in$\sigma$_{0} . (If such a $\sigma$_{0} does not exist, (4.1)
obviously holds. In fact, it is easy to see that X_{\triangle}(h)\cap\{x_{N}=0\}=\emptyset in this case. Hence
we have \mathcal{V}_{\triangle}(y, h;m)=0 . On the other hand, the right hand side of (4.1) is also 0 by
On vector partition functions with negative weights 191
\displaystyle \frac{(-\partial_{N})^{m_{N}-1}}{(m_{N}-1)!}=0.) Let us set h=\displaystyle \sum_{j\in$\sigma$_{0}}p_{j}$\alpha$_{j} . We take (x_{j})_{j\not\in$\sigma$_{0}} as a coordinate of the
hyperplane H_{\triangle}(h) . Recall that $\alpha$_{k}=\displaystyle \sum_{j\in$\sigma$_{0}}c_{jk}^{$\sigma$_{0}}$\alpha$_{j} for k\not\in$\sigma$_{0} . Hence we see that
\displaystyle \langle y, x\rangle=\sum_{j\in$\sigma$_{0}}y_{j}p_{j}+\sum_{k\not\in$\sigma$_{0}}(y_{k}-\sum_{j\in$\sigma$_{0}}c_{jk}^{$\sigma$_{0}}y_{j})x_{k}.It follows that
\displaystyle \frac{\partial}{\partial x_{N}}e^{-\langle y,x\rangle}=-(y_{N}-\sum_{j\in$\sigma$_{0}}c_{jN}^{$\sigma$_{0}}y_{j})e^{-\langle y,x\rangle}.Hence the integrand in (5.1) is computed as follows.
=\displaystyle \sum_{u_{j},u_{\mathrm{e}}}\left(\begin{array}{l}|m_{N}|\\u_{j},u_{e}\end{array}\right)\prod_{j\in$\sigma$_{0}^{c}}\frac{x_{j}^{m_{j}-1}}{(m_{j}-1)!}\prod_{j\in$\sigma$_{0}}\frac{(c_{jN}^{$\sigma$_{0}})^{u_{j}}x_{j}^{m_{j}-u_{j}-1}}{(m_{j}-u_{j}-1)!}(y_{N}-\sum_{j\in$\sigma$_{0}}c_{jN}^{$\sigma$_{0}}y_{j})^{u_{\mathrm{e}}}e^{-\langle y',x'\rangle},where the sum is taken over all u_{j}\in \mathbb{Z}_{\geq 0}(j\in$\sigma$_{0}) and u_{e}\in \mathbb{Z}_{\geq 0} such that \displaystyle \sum_{j\in$\sigma$_{0}}u_{j}+u_{e}=|m_{N}| . and we set
Step 3. Let $\gamma$' be the chamber in C(\triangle') that contains h . By the assumption of our
induction, the integral
\displaystyle \int_{X_{\triangle'}(h)}\prod_{j\in$\sigma$_{0}^{c}}\frac{x_{j}^{m_{j}-1}}{(m_{j}-1)!}\prod_{j\in$\sigma$_{0}}\frac{x_{j}^{m_{j}-u_{j}-1}}{(m_{j}-u_{j}-1)!}e^{-\langle y',x'\rangle}ds'in (5.2) is computed as follows.
where $\sigma$^{c} is the complement of $\sigma$ in \{1, . . . , N-1\}.
Let $\sigma$\in \mathcal{B}(\triangle', $\gamma$') . Then the factor \displaystyle \prod_{j\in$\sigma$_{0}}(c_{jN}^{$\sigma$_{0}})^{u_{j}}(y_{N}-\sum_{j\in$\sigma$_{0}}c_{jN}^{$\sigma$_{0}}y_{j})^{u_{\mathrm{e}}} in (5.2) is
On vector partition functions with negative weights 193
y_{N}-\displaystyle \sum_{j\in$\sigma$_{0}}c_{jN}^{$\sigma$_{0}}y_{j}+\sum_{k\in$\sigma$^{c}\cap$\sigma$_{0}}(c_{kN}^{$\sigma$_{0}}(y_{k}-\sum_{j\in $\sigma$}c_{jk}^{ $\sigma$}y_{j}))=y_{N}-\sum_{j\in $\sigma$}c_{jN}^{ $\sigma$}y_{j}.Proof. We see that
=\displaystyle \sum_{j\in $\sigma$\cap$\sigma$_{0}^{c}}\sum_{k\in$\sigma$^{c}\cap$\sigma$_{0}}c_{kN}^{$\sigma$_{0}}c_{jk}^{ $\sigma$}$\alpha$_{j}+\sum_{j\in $\sigma$\cap$\sigma$_{0}}(\sum_{k\in$\sigma$^{c}\cap$\sigma$_{0}}c_{kN}^{$\sigma$_{0}}c_{jk}^{ $\sigma$}+c_{jN}^{$\sigma$_{0}})$\alpha$_{j}.Hence we have
On vector partition functions with negative weights 195
Since \displaystyle \frac{(-\partial_{j})^{m_{j}-1}}{(m_{j}-1)!}=0 if m_{j}\leq 0 ,the sum above is actually taken over $\sigma$\in \mathcal{B}(\triangle', $\gamma$')
such that m_{j}>0 for \forall j\in $\sigma$ . Moreover, since h belongs to both $\gamma$' and $\gamma$ ,we see that
where this time $\sigma$^{c} is the complement of $\sigma$ in \{ 1, . . .
, N\} . Since \displaystyle \frac{(-\partial_{j})^{m_{N}-1}}{(m_{N}-1)!}=0 ,the
formula above is nothing but (4.1). This completes the proof of Theorem 4.1.
References
[1] Beck, M. and Robins, S., Computing the Continuous Discretely: Integer‐Point Enumera‐
tion in Polyhedra, Undergraduate Texts in Mathematics, Springer, 2007.
[2] Brion, M. and Vergne, M., Residue formulae, vector partition functions and lattice pointsin rational polytopes, J. Amer. Math. Soc., 10 (1997), 797‐833.
[3] Gel�fand, I. M. and Shilov, G. E., Generalized functions. Vol. 1. Properties and operations,Academic Press, 1964.
[4] Kashiwara, M., Kawai, T., and Kimura, T., Foundations of Algebraic Analysis, Princeton
Mathematical Series 37, Princeton Univ. Press, 1986.
[5] Suzuki, T. and Takakura, T., Symplectic volumes of certain symplectic quotients associ‐
ated with the special unitary group of degree three, Tokyo J. Math., 31 (2008), 1‐26.
[6] Takakura, T., On asymptotic partition functions for root systems, Toric Topology(eds. M. Harada, Y. Karshon, M. Masuda, and T. Panov), Contemp. Math., 460,Amer. Math. Soc., 2008, pp. 339‐348.
[7] Vergne, M., Residue formulae for Verlinde sums, and for number of integral points in
convex rational polytopes, Proceedings of the Tenth General Meeting of the EuropeanWomen in Mathematics (Malta, 2001), World Sci. Publ., 2003, pp. 223‐285.