1

In the literature

see Proposition 4.3 in page 1405 in: Letac, Gérard; Massam, Hélène, The noncentral Wishart as an exponential family, and its moments, J. Multivariate Anal. 99, No. 7, 1393-1417 (2008). ZBL1140.62043.

there is a result on the expected value of an inverse of a non-central Wishart matrix. Unfortunately the result is very complicated, i.e. it is an infinite series over partitions of integers with coefficients that are expressed as certain integrals over the space of positive definite square matrices with the integrand involving derivatives of zonal polynomials. As such it is unclear, at least for me, how to compute that.

Now, by using integration by parts I managed to simplify the expressions and reduce them to the following integral below.

Let $y $ and ${\mathfrak a}$ be positive definite square matrices of dimension $N \ge 1$ and let$T \ge (N+1)/2$. Then let $\kappa:= (k_1,k_2,\cdots,k_N)$ be a partition of an integer $k$. This means that we have $k= k_1+k_2+\cdots+k_N$ and $k_1 \ge k_2 \ge \cdots \ge k_N \ge 0 $ . Finally let $C_\kappa(y)$ denote the zonal polynomial of the matrix $y$. Then the question is to compute the following integral:

\begin{equation} {\mathcal I}_N({\mathfrak a}) := \int\limits_{y>0} C_\kappa(y) \cdot \det(y)^{T-\frac{N+1}{2}} \cdot (y^{-1}) \cdot e^{-Tr[y\cdot {\mathfrak a}^{-1}]} dy \tag{1} \end{equation}

Now, by using the canonical parametrization of positive definite matrices $y = L^T \cdot L$ with $L$ being positive and upper triangular we have computed (with a little help of Mathematica) the integral above for $N=2$. Let $(a_1,a_2)$ denote the eigenvalues of the matrix ${\mathfrak a}^{-1}$. Then let ${\mathfrak P}_{n_1,n_2,n_3}$ denote the coefficient in the expansion of the zonal polynomial $C_\kappa(L^T L )$ at the powers $l_1^{n_1} l_2^{n_2} l_3^{n_3}$ with $L = [l_1, l_2]; [0, l_3]$.

Then the result reads:

\begin{eqnarray} {\mathcal I}_2({\mathfrak a}) &=& \sum\limits_{\begin{array}{l} n_1+n_2+n_3 = 2 k \\ n_1,n_2,n_3 \ge 0 \end{array}} {\mathfrak P}_{n_1,n_2,n_3} \frac{1}{2} \frac{ a_1^{-\frac{n_1}{2}} a_2^{\frac{1}{2} \left(-n_2-n_3\right)} }{ \left(a_1 a_2\right){}^{T} } \Gamma \left(T+\frac{n_1}{2}-1\right) \Gamma \left(T+\frac{n_3}{2}-\frac{3}{2}\right) \cdot \Gamma(\frac{n_2+1}{2}) \cdot \left( \begin{array}{cc} a_1 \left(\frac{1}{2} \left(n_2+n_3\right)+T-1\right) & -\frac{\sqrt{a_1 a_2} \Gamma \left(\frac{n_2}{2}+1\right) \left(T+\frac{n_1}{2}-1\right)_{\frac{1}{2}}}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right)} \\ -\frac{\sqrt{a_1 a_2} \Gamma \left(\frac{n_2}{2}+1\right) \left(T+\frac{n_1}{2}-1\right)_{\frac{1}{2}}}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right)} & a_2 \left(\frac{n_1}{2}+T-1\right) \\ \end{array} \right) \tag{2} \end{eqnarray}

As always, the code snippet below validates the result above. Here we go:

In[1227]:= 
Clear[l];
NN = 2; T = 10;
NN2 = (NN + 1) NN/2;
k = 5;
myParts = IntegerPartitions[k, {NN}];
\[Kappa] = myParts[[RandomInteger[{1, Length[myParts]}]]];
A = DiagonalMatrix[RandomReal[{1, 2}, {NN}]];

(The parametrization of positively definite matrices) myRoutine[l_List, NN_] := Module[{L, S, count, i, j}, L = ConstantArray[0, {NN, NN}]; count = 1; Do[L[[i, j]] = l[[count]]; count++;, {i, 1, NN}, {j, i, NN}]; S = Transpose[L] . L; {S, L} ];

(The actual integral in the canonical parametrization.) myFunction[n1_, n2_, n3_, a1_, a2_] := (a1 a2)^-T a1^(-(n1/2)) a2^( 1/2 (-n2 - n3)) 1/ 2 Gamma[-(3/2) + n3/2 + T] Gamma[-1 + n1/2 + T] Gamma[(1 + n2)/ 2] {{a1 (-1 + (n2 + n3)/2 + T) , - Sqrt[a1 a2] Gamma[ 1 + n2/2]/ Gamma[(1 + n2)/2] Pochhammer[-1 + n1/2 + T, 1/2] }, {-Sqrt[a1 a2] Gamma[1 + n2/2]/ Gamma[(1 + n2)/2] Pochhammer[-1 + n1/2 + T, 1/2] , a2 (-1 + n1/2 + T)}};

Clear[l, ll]; (Zonal polynomial in canonical parametrization.) ll = Table[l[i], {i, 1, NN2}]; tmp = myRoutine[ll, NN]; S = tmp[[1]]; L = tmp[[2]]; myList = {# /. l[_] :> 1, Exponent[#, ll]} & /@ List @@ Expand[Simplify[ZonalPolynomial[[Kappa], Eigenvalues[S]]]];

(Compute integral numericaly.) (Include the Jacobian as in page 250 in book by Muirhead) NIntegrate[ With[{ll = Table[l[i], {i, 1, NN2}]}, With[{tmp = myRoutine[ll, NN]}, With[{S = tmp[[1]], L = tmp[[2]]}, ZonalPolynomial[[Kappa], Eigenvalues[S]] Det[S]^(T - (NN + 1)/2) Inverse[ S] Exp[-Tr[S . A]] (2^NN Product[ L[[i, i]]^(NN + 1 - i), {i, 1, NN}]) ]]], Evaluate[Sequence @@ Table[{l[i], 0, Infinity}, {i, 1, NN2}]] ]

(Compute integral analytically.) Sum[myList[[i, 1]] myFunction @@ Join[myList[[i, 2]], Diagonal[A]], {i, 1, Length[myList]}]

Output of the code. Numerically and analytically top and bottom respectively.

In the same way we managed to obtain the result for $N=3$. It reads:

\begin{eqnarray} &&{\mathcal I}_3({\mathfrak a}) = \sum\limits_{\begin{array}{l} n_1+n_2+n_3+n_4+n_5+n_6 = 2 k \\ n_1,n_2,n_3,n_4,n_5,n_6 \ge 0 \end{array}} {\mathfrak P}_{n_1,n_2,n_3,n_4,n_5,n_6} \cdot \frac{1}{4} \frac{a_1^{-\frac{n_1}{2}} a_2^{-\frac{n_2+n_4}{2}} a_3^{-\frac{n_3+n_5+n_6}{2}} }{(a_1 a_2 a_3)^T} \cdot \Gamma(T + \frac{n_1}{2}-1) \Gamma(T + \frac{n_4}{2}-\frac{3}{2}) \Gamma(T + \frac{n_6}{2}-2) \cdot \Gamma(\frac{1+n_2}{2}) \Gamma(\frac{1+n_3}{2}) \Gamma(\frac{1+n_5}{2}) \cdot \left( \begin{array}{ccc} \frac{1}{8} a_1 \left(\left(n_4+2 T-3\right) \left(n_3+n_6+2 T-3\right)+\left(n_2+1\right) \left(n_5+n_6+2 T-3\right)-\frac{8 \Gamma \left(\frac{n_2}{2}+1\right) \Gamma \left(\frac{n_3}{2}+1\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_3+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)}\right) & \frac{\sqrt{a_1} \sqrt{a_2} \Gamma \left(T+\frac{1}{2} \left(n_1-1\right)\right) \left(\frac{2 \Gamma \left(\frac{n_3}{2}+1\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{\Gamma \left(\frac{1}{2} \left(n_3+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)}-\frac{\left(n_5+n_6+2 T-3\right) \Gamma \left(\frac{n_2}{2}+1\right)}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right)}\right)}{4 \Gamma \left(T+\frac{n_1}{2}-1\right)} & \frac{\sqrt{a_1} \sqrt{a_3} \Gamma \left(T+\frac{1}{2} \left(n_1-1\right)\right) \left(\frac{2 \Gamma \left(\frac{n_2}{2}+1\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)}-\frac{\left(n_4+2 T-3\right) \Gamma \left(\frac{n_3}{2}+1\right)}{\Gamma \left(\frac{1}{2} \left(n_3+1\right)\right)}\right)}{4 \Gamma \left(T+\frac{n_1}{2}-1\right)} \\ \frac{\sqrt{a_1} \sqrt{a_2} \Gamma \left(T+\frac{1}{2} \left(n_1-1\right)\right) \left(\frac{2 \Gamma \left(\frac{n_3}{2}+1\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{\Gamma \left(\frac{1}{2} \left(n_3+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)}-\frac{\left(n_5+n_6+2 T-3\right) \Gamma \left(\frac{n_2}{2}+1\right)}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right)}\right)}{4 \Gamma \left(T+\frac{n_1}{2}-1\right)} & \frac{1}{8} a_2 \left(n_1+2 T-2\right) \left(n_5+n_6+2 T-3\right) & -\frac{\sqrt{a_2} \sqrt{a_3} \left(n_1+2 T-2\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{4 \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)} \\ \frac{\sqrt{a_1} \sqrt{a_3} \Gamma \left(T+\frac{1}{2} \left(n_1-1\right)\right) \left(\frac{2 \Gamma \left(\frac{n_2}{2}+1\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{\Gamma \left(\frac{1}{2} \left(n_2+1\right)\right) \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)}-\frac{\left(n_4+2 T-3\right) \Gamma \left(\frac{n_3}{2}+1\right)}{\Gamma \left(\frac{1}{2} \left(n_3+1\right)\right)}\right)}{4 \Gamma \left(T+\frac{n_1}{2}-1\right)} & -\frac{\sqrt{a_2} \sqrt{a_3} \left(n_1+2 T-2\right) \Gamma \left(\frac{n_5}{2}+1\right) \Gamma \left(T+\frac{n_4}{2}-1\right)}{4 \Gamma \left(\frac{1}{2} \left(n_5+1\right)\right) \Gamma \left(T+\frac{1}{2} \left(n_4-3\right)\right)} & \frac{1}{8} a_3 \left(n_1+2 T-2\right) \left(n_4+2 T-3\right) \\ \end{array} \right) \end{eqnarray}

Likewise ${\mathfrak P}_{n_1,n_2,n_3,n_4,n_5,n_6}$ denotes the coefficient in the expansion of the zonal polynomial $C_\kappa(L^T L )$ at the powers $l_1^{n_1} l_2^{n_2} l_3^{n_3} l_4^{n_4} l_5^{n_5} l_6^{n_6}$ with $L = [l_1, l_2, l_3]; [0, l_4, l_5], [0, 0, l_6]$.

Again the relevant Mathematica code is below:

NN = 3; T =.;
n1 =.; n2 =.; n3 =.; n4 =.; n5 =.; n6 =.; a1 =.; a2 =.; a3 =.; l1 =.; \
l2 =.; l3 =.; l4 =.; l5 =.; l6 =.;
A = DiagonalMatrix[{a1, a2, a3}];
myRoutine[l_List, NN_] := Module[{L, S, count, i, j},
   L = ConstantArray[0, {NN, NN}];
   count = 1;
   Do[L[[i, j]] = l[[count]]; count++;, {i, 1, NN}, {j, i, NN}];
   S = Transpose[L] . L;
   {S, L}
   ];
tmp = myRoutine[{l1, l2, l3, l4, l5, l6}, 3];
S = tmp[[1]]; L = tmp[[2]];
(*Parametrize the integrand. Don't forget about the Jacobian.*)
mat = Simplify[
     Det[S]^(T - (NN + 1)/2)
        Inverse[
       S] Exp[-Tr[S . A]] (2^NN  Product[
         L[[i, i]]^(NN + 1 - i), {i, 1, NN}]), Assumptions -> T > 3] //
     PowerExpand // Simplify;
(*Include the zonal polynomial do do the integration in \
thel-variables.*)
(*Here n1+n2+n3+n4+n5+n6==2 k*)
t0 = TimeUsed[];
myoutput = 
  Integrate[
    l1^n1  l2^n2  l3^n3  l4^n4  l5^n5  l6^n6  mat, {l1, 0, 
     Infinity}, {l2, 0, Infinity}, {l3, 0, Infinity}, {l4, 0, 
     Infinity}, {l5, 0, Infinity}, {l6, 0, Infinity}, 
    Assumptions -> 
     n1 >= 0 && n2 >= 0 && n3 >= 0 && n4 >= 0 && n5 >= 0 && n6 >= 0 &&
       T > 4 && a1 > 0 && a2 > 0 && a3 > 0] // Simplify;
Print["Symbolic integration done in time=", TimeUsed[] - t0];
myoutput = ((myoutput/(1/4 (a1  a2  a3)^(-T) a1^(-(n1/2))  a2^(
          1/2 (-n2 - n4))  a3^(1/2 (-n3 - n5 - n6))
           Gamma[-1 + n1/2 + T]  Gamma[-3/2 + 1/2  (n4) + 
            T]  Gamma[-2 + n6/2 + T]   Gamma[(1 + n2)/2]  Gamma[(
           1 + n3)/2]  Gamma[(1 + n5)/2])) // Simplify // 
     PowerExpand // FullSimplify);
myoutput // MatrixForm

Having said all this my question is obvious.

How do we compute the integral in question for generic $N \in {\mathbb N} $.

Przemo
  • 11,971
  • 1
  • 25
  • 56

1 Answers1

1

This will by no means be a full answer. Yet I think that the approach which I intend to sketch here is fruitful and even though it might not be complete for the time being it can be eventually made more rigorous.

It is clear that all what we need to do is to parametrize the space of positive definite symmetric matrices. We choose the obvious parametrization $y = L^T \cdot L$ with $L_{i,j} = 1_{j \ge i} \cdot l_{-\frac{i^2}{2} + j -N + i(\frac{1}{2} + N)}$ for $i,j=1,\cdots,N$ being upper triangular composed of non-negative entries (here the index of the $l$-terms increases row-wise from top-left to bottom -right).

Let $M:= \binom{N+1}{2}$ and ${\mathcal a}:= T-(N+1)/2$. Now let us now analyze the terms in the integrand in $(1)$ from the left to the right.

The zonal polynomial:

The quantity in question is a homogeneous polynomial of order $2 k$ in the variables $\vec{l}:= (l_\xi)_{\xi=1}^M$ viz:

\begin{equation} C_\kappa(y) = \sum\limits_{\begin{array}{lll} n_1+n_2+\cdots+n_M=2 k \\ n_1\ge 0, \cdots, n_M \ge 0 \end{array}} {\mathfrak P}_{n_1,\cdots, n_M} \cdot l_1^{n_1} l_2^{n_2} \cdots l_M^{n_M} \tag{1} \end{equation}

The equation above follows from the following fact. Note that, for every $p\in {\mathbb N}$ we have that $Tr[y^p]$ is a homogeneous polynomial of order $2 p$ in the variable $\vec{l}$ . Now, since the zonal polynomial $C_\kappa(y)$ is a linear combination of power-sum symmetric function each one of order $k$, due to the observation above, the later quantities are each or order $2 k$ in the variable $\vec{l}$ .As such the whole thing, i.e. the zonal polynomial, is or order $2 k$ as well.

The power of the determinant:

\begin{equation} \det(y)^{\mathcal a} = l_1^{2 {\mathcal a}} l_{N+1}^{2 {\mathcal a}} l_{2 N}^{2 {\mathcal a}} l_{3 N-2}^{2 {\mathcal a}} l_{4 N-5}^{2 {\mathcal a}} \cdots l_M^{2 {\mathcal a}} \tag{2} \end{equation}

This is follows in an obvious way from the fact that $\det(y) = \det(L)^2$ and the fact that the eigenvalues of $L$ are its diagonal elements.

The inverse matrix:

Let us define auxiliary (upper-triangular) matrices $\Lambda^{(p)}$ viz:

\begin{eqnarray} (\Lambda^{(p)})_{i,j} &=& 1_{j-i \ge p} \sum\limits_{i+1 \le \xi_{p-2} < \cdots < \xi_0 \le j-1} \left( \prod\limits_{q=-1}^{p-1} \frac{l_{(\left(N+\frac{3}{2}\right) \xi _q-N-\frac{\xi _q^2}{2})+(\xi _{q-1}-\xi_q)}}{ l_{(\left(N+\frac{3}{2}\right) \xi _q-N-\frac{\xi _q^2}{2})}} \right) \end{eqnarray}

subject to $\xi_{-2},\xi_{-1},\xi_{p-1}=\infty,j,i$ and the convention that $l_\infty =1$.

Note, that for each sequence $(\xi_q)_{q=0}^{p-2}$ above each term in the product is a ratio of two $l$'s being located in the $\xi_q$th row of the matrix $L$.Here the denominator and the numerator are located at the first non-zero position and at the subsequent positions respectively.

Then by using my answer to the question about an inverse of an upper triangular matrix the following identity holds true:

\begin{equation} (y^{-1})_{i,j} = \sum\limits_{p_1=0}^{N-1} \sum\limits_{p_2=0}^{N-1} (-1)^{p_1+p_2} \sum\limits_{k=(i+p_1) \vee (j+p_2)}^{N} \Lambda^{(p_1)}_{i,k} \Lambda^{(p_2)}_{j,k} \tag{3} \end{equation}

The exponent from the trace:

It is not hard to see that the expression in the exponent is a linear combination of squares of $l$'s only, viz:

\begin{equation} e^{-Tr[y \cdot {\mathfrak a}^{-1}]} = e^{-\sum\limits_{p=1}^M A_p l_p^2} \tag{4} \end{equation}

where $(A_p)_{p=1}^M := \left((a_1,\cdots a_N);(a_2,\cdots,a_N);(a_3,\cdots,a_N); \cdots (a_N)\right)$ with $(a_p)_{p=1}^N$ being the eigenvalues of the matrix ${\mathfrak a}^{-1}$.

The Jacobian:

We have:

\begin{equation} dy = 2^N \cdot l_1^N l_{N+1}^{N-1} l_{2 N}^{N-2} l_{3N-2}^{N-3} l_{4N-5}^{N-4} \cdots l_M^{1} \cdot \prod\limits_{\xi=1}^M d l_\xi \tag{5} \end{equation}

See page 250 in

Muirhead, Robb J., Aspects of multivariate statistical theory, Wiley Series in Probability and Mathematical Statistics. New York: John Wiley & Sons, Inc. XIX, 673 p. (1982). ZBL0556.62028.


We define two row vectors with non zero entries at positions related to diagonal entries of the matrix $L$. We have:

\begin{eqnarray} \vec{{\mathcal D}_1} &:=& \left( \sum\limits_{\zeta=1}^N \delta_{\lambda, (\zeta-1) N- \frac{\zeta}{2} (\zeta-3)}\right)_{\lambda=1}^M \\ \vec{{\mathcal D}_2} &:=& \left( \sum\limits_{\zeta=1}^N (N-\zeta+1) \delta_{\lambda, (\zeta-1) N- \frac{\zeta}{2} (\zeta-3)}\right)_{\lambda=1}^M \end{eqnarray}

In addition to this we define a row vector that now depends on summation indices $\vec{\xi} := \left( \xi_q \right)_{q=-1}^{p-1}$ . We have:

\begin{eqnarray} \vec{{\mathcal D}_3}(\vec{\xi}) &:=& \left( % \sum\limits_{q=-1}^{p-1} \delta_{\lambda, (\left(N+\frac{3}{2}\right) \xi _q-N-\frac{\xi _q^2}{2})+(\xi _{q-1}-\xi_q)} - \delta_{\lambda, (\left(N+\frac{3}{2}\right) \xi _q-N-\frac{\xi _q^2}{2})} \right)_{\lambda=1}^M \end{eqnarray}

Finally, we define $\vec{l}^\vec{n} := \prod\limits_{\lambda=1}^M l_\lambda^{n_\lambda} $ and $d\vec{l}:= \prod\limits_{\lambda=1}^M d l_\lambda$.


Now with this notation in place we bring everything together as follows:

\begin{eqnarray} ({\mathcal I}_N({\mathfrak a}))_{i,j} &=& 2^N \cdot \sum\limits_{\begin{array}{lll} n_1+n_2+\cdots+n_M=2 k \\ n_1\ge 0, \cdots, n_M \ge 0 \end{array}} {\mathfrak P}_{n_1,\cdots, n_M} \sum\limits_{p_1=0}^{N-1} \sum\limits_{p_2=0}^{N-1} (-1)^{p_1+p_2} \sum\limits_{k=(i+p_1) \vee (j+p_2)} \\ && \int\limits_{{\mathbb R}_+^M} \vec{l}^{\vec{n} + 2 {\mathcal a} \vec{{\mathcal D}_1} + \vec{{\mathcal D}_2}} \cdot \Lambda^{(p_1)}_{i,k} \Lambda^{(p_2)}_{j,k} \cdot e^{-\sum\limits_{\lambda=1}^M A_\lambda l_\lambda^2} d\vec{l} \\ &=& 2^N \cdot \sum\limits_{\begin{array}{lll} n_1+n_2+\cdots+n_M=2 k \\ n_1\ge 0, \cdots, n_M \ge 0 \end{array}} {\mathfrak P}_{n_1,\cdots, n_M} \sum\limits_{p_1=0}^{N-1} \sum\limits_{p_2=0}^{N-1} (-1)^{p_1+p_2} \sum\limits_{k=(i+p_1) \vee (j+p_2)} \\ && % \sum\limits_{\begin{array}{lll} i+1 \le \xi_{p_1-2} < \cdots < \xi_0 \le k-1 \\ \xi_{-2},\xi_{-1},\xi_{p_1-1} = \infty,k,i \end{array}} % \sum\limits_{\begin{array}{lll} j+1 \le \eta_{p_2-2} < \cdots < \eta_0 \le k-1 \\ \eta_{-2},\eta_{-1},\eta_{p_2-1} = \infty,k,j \end{array}} % \underbrace{ \int\limits_{{\mathbb R}_+^M} \vec{l}^{\vec{n} + 2 {\mathcal a} \vec{{\mathcal D}_1} + \vec{{\mathcal D}_2}} \cdot \vec{l}^{\vec{{\mathcal D}_3}(\vec{\xi})} \cdot \vec{l}^{\vec{{\mathcal D}_3}(\vec{\eta})} \cdot e^{-\sum\limits_{\lambda=1}^M A_\lambda l_\lambda^2} d\vec{l} }_{ \left. \frac{1}{2^M} \prod\limits_{\lambda=1}^M \frac{ \Gamma(\frac{1+n_\lambda}{2}) } { (\sqrt{A_\lambda})^{1+n_\lambda} } \right|_{ n_\lambda \rightarrow n_\lambda + 2 {\mathcal a} ({\mathcal D}_1)_\lambda + ({\mathcal D}_2)_\lambda + ({\mathcal D}_3)(\vec{\xi})_\lambda+ ({\mathcal D}_3)(\vec{\eta})_\lambda } } \tag{6} \end{eqnarray}

Now the result of integration can be simplified by taking out a common factor, i.e. one that does not depend on the summation indices $\vec{\xi}$ and $\vec{\eta}$, and a multiplicative residual term that does depend on those indices. Here we go:

\begin{eqnarray} &&\left. \frac{1}{2^M} \prod\limits_{\lambda=1}^M \frac{ \Gamma(\frac{1+n_\lambda}{2}) } { (\sqrt{A_\lambda})^{1+n_\lambda} } \right|_{ n_\lambda \rightarrow n_\lambda + 2 {\mathcal a} ({\mathcal D}_1)_\lambda + ({\mathcal D}_2)_\lambda + ({\mathcal D}_3)(\vec{\xi})_\lambda+ ({\mathcal D}_3)(\vec{\eta})_\lambda } = \\ && \frac{ \prod _{\lambda =1}^N a_{\lambda }^{-\frac{1}{2} \sum _{\zeta =1}^{\lambda } n_{-\frac{\zeta ^2}{2}+\lambda +\zeta \left(N+\frac{1}{2}\right)-N}} } { \left(\prod _{\lambda =1}^N a_{\lambda }\right){}^{T} } \cdot \\ && % \prod _{\zeta =1}^N \Gamma \left[\frac{1}{2} \left(-\zeta +n_{-\frac{\zeta ^2}{2}+\zeta +\zeta \left(N+\frac{1}{2}\right)-N}+1\right)+T\right] \cdot \\ && % \prod _{\lambda =1}^N \prod _{\zeta =\lambda +1}^N \Gamma \left[\frac{1}{2} \left(n_{\zeta -\frac{\lambda ^2}{2}+\lambda \left(N+\frac{1}{2}\right)-N}+1\right)\right] \cdot \\ && % \left. \prod\limits_{\lambda=1}^M \frac{\Gamma(\frac{1+{\bar n}_\lambda + {\mathcal D}_3)(\vec{\xi})_\lambda+ ({\mathcal D}_3)(\vec{\eta})_\lambda }{2})}{\Gamma(\frac{1+{\bar n}_\lambda}{2})} \frac{1}{(\sqrt{A_\lambda})^{({\mathcal D}_3)(\vec{\xi})_\lambda+ ({\mathcal D}_3)(\vec{\eta})_\lambda}} \right|_{{\bar n}_\lambda \rightarrow n_\lambda + 2 {\mathcal a} ({\mathcal D}_1)_\lambda + ({\mathcal D}_2)_\lambda} \end{eqnarray}

We can see that the common factor matches perfectly well with the one given in the body of the question , i.e. for $N=2,3$.

Przemo
  • 11,971
  • 1
  • 25
  • 56