82

811 1992 82-97

Computational algebraic analysis and connection formula

Nobuki Takayama ( )
Department of Mathematics, Kobe University
Rokko, Kobe, Japan
(January 31, 1992 )

Abstract. We introduce ‘fundamental problems” in ‘conlputational algebraic analysis” and explain that mechanical derivation and
proving of connection formulas, which are those of the most important special function identities, are reduced to solving the fundamental
problems. We give a new efficient algorithm of solving the fundamental problems The algorithm decomposes a problem in -variables $n$

to problems in $7l-1$ variables

\S 1. Ftindamental problems in computational algebraic analysis

Put
$A_{n}=C\{x_{1}, \cdots, x_{n}, \partial_{1}, \cdots, \partial_{71}\}$ ,

$\partial_{i}x_{j}-x_{j}\partial_{i}=\{\begin{array}{l}1_{)}i=j0,i\neq j\end{array}$

and
$B_{n}=C\langle x_{1}, \cdots, x_{n}, E_{1}, \cdots, E_{n}\rangle$ ,

$E_{i}x_{j}-x_{j}E_{?}\cdot=\{\begin{array}{l}E_{i},i=j0,i\neq j\end{array}$

The ring is the ring of differential operators and

$A_{n}$
is the ring of difference operators. Let and S43 be $B_{n}$ $\mathfrak{U}$

left ideals of the ring and respectively. We will call the following problems “fundamental problems
$A_{n}$ $B_{n}$

in computational algebraic analysis” ;

Find Grobner basis of the following modules;

(1.1) $A_{n-1}$ module $A_{n}/(\mathfrak{U}+\partial_{n}A_{n})$ ,

(1.2) $A_{n-1}$ module $A_{n}/(\mathfrak{U}+x_{n}A_{n})$ ,

(1.3) $B_{n-1}$ module $B_{n}/(q_{3}+(E_{n}-1)B_{n})$ .

Notice that, for example, $\mathfrak{U}+\partial_{n}A_{n}$

is the sum of the left ideal $\mathfrak{U}$

and the right ideal $\partial_{n}A_{7l}$

of $A_{n}$
.
We can solve the following problems by solving problems (1.1), (1.2) and (1.3) respectively.

(1.4) Find $A_{n-1}\cap(\mathfrak{U}+(J_{r\iota}A_{n})$ .

(1.5) Find $A_{r\iota-1}\cap(\mathfrak{U}+x_{n}A_{n})$ .

(1.6) Find $B_{n-1}\cap(\mathfrak{P}+(E_{n}-1)B_{n})$ .

We will also call these problems “fundamental probleIns in computational algebraic analysis“.
These fundamental problems are related to $prol$ $lerns$ in the classical analysis as follows; )
 83

$A_{n}/\mathfrak{U}\mapsto A_{r\iota}/(\mathfrak{U}+\partial_{n}A_{n})$ $f(x_{1}, \cdots, x_{n})rightarrow\int_{-}^{\infty_{\infty}}f(x_{1)}. . , x_{n})dx_{n}$

definite integral with parameters

$A_{n}/\mathfrak{U}\mapsto A_{n}/(\mathfrak{U}+x_{n}A_{n})$ $f(x_{1}, \cdots, x_{n})\mapsto f(x_{1} , )x_{n-1},0)$

restriction
$B_{n}/\mathfrak{P}\mapsto B_{n}/(\mathfrak{P}+(E_{n}-1)B_{n})$ , .,
$f(x_{1}, \cdots , x_{n})\mapsto\sum_{x=-\infty}^{\infty_{n}}f(x_{1}$ $\cdot$

$x_{n})$

summation with parameters

Let us see the correspondence of the above by using an example. Put $f(x_{1}, x_{2})=(_{x_{2}}^{x_{1}})$ (the binomial
coefficient). Here, we put when $x_{2}<0$ or $x_{2}>x_{1}$ . We define the action of
$(_{x_{2}}^{x_{1}})=0$ $E_{1}$
and to the $E_{2}$

function as follows;
$E_{1}f(x_{1}, x_{2})=f(x_{1}+1, x_{2})$ , $E_{2}f(x_{1)}x_{2})=f(x_{1}, x_{2}+1)$ .

Put
$\ell_{1}=(x_{1}+1-x_{2})E_{1}-(x_{1}+1)$ ,
$\ell_{2}=(x_{2}+1)E_{2}-(x_{1}-x_{2})$ .
Then we have $l_{1}f=l_{2}f=0$ . Let be the left ideal of generated by and . Notice that we have
$\mathfrak{P}$ $B_{2}$ $\ell_{1}$ $p_{2}$

. Let us find an element of $(\mathfrak{P}+(E_{2}-1)B_{2})\cap B_{1}$ . It is convinient to write the operators and as
$\mathfrak{P}f=0$ $\ell_{1}$ $\ell_{2}$

$P_{1}=E_{1}(x_{1}-x_{2})-(x_{1}+1)$ ,

$P_{2}=(E_{2}-1)x_{2}-x_{1}+2x_{2}$ .

Since we have
$2\ell_{1}+E_{1}l_{2}=2E_{1}(x_{1}-x_{2})-2(x_{1}+1)+(E_{2}-1)E_{1^{X\underline{\cdot)}}}-E_{1}x_{1}+2E_{1}x_{2}$

$=2E_{1}x_{1}-2(x_{1}+1)-E_{1}x_{1}+(E_{2}-1)E_{1}x_{2}$

$=E_{1}x_{1}-2(x_{1}+1)+(E_{2}-1)E_{1}x_{2}$ ,

we can see that

$E_{1}x_{1}-2(x_{1}+1)\in(\mathfrak{P}+(E_{2}-1)B_{2})\cap B_{1}$ .

For any element $P\in B_{2}$ , we have $\sum_{x=-\infty}^{\infty_{2}}(E_{2}-1)l(x_{1}x_{2})=0$ . Therefore we have

$[E_{1}x_{1}-2(x_{1}+1)]$ $\sum_{x_{2}=-\infty}^{\infty}$ $(\begin{array}{l}x_{1}x_{2}\end{array})=0$

.

We have the following proposition in a similar way.

PROPOSITION 1.1 Let $\mathfrak{P}$

be a leftt ideal of $B_{n}$
and $f$ be a function of $x_{1}\cdots x_{n}$ such that $f(x_{1}, \cdots, x_{n})=$

$0$
$(|x_{n}|>>0)$ . If $\mathfrak{P}f(x_{1}, --, x_{n})=0$ , then

$(\mathfrak{P}+(E_{n}-1)B_{n})\cap B_{n-1}$ $\sum_{x_{n}=-\infty}^{\infty}$ $f(x_{1)}\cdots, x_{n})=0$ .

Thus, if the $B_{n}$

module structure of the function $f$ is $B_{n}/\mathfrak{P}$
, then the $B_{n-1}$ lnodule structure of the
function is
$\sum_{x=-\infty}^{\infty_{n}}f(x_{1)}\cdots , x_{n})$

$B_{n-1}/(\mathfrak{P}+(E_{n}-1)B_{n})\cap B_{n-i}$ .

In the theory of holonomic systems, we consider the $B_{n-1}$ module $B_{n}/(\mathfrak{P}+(E_{n}-1)B_{n})$ rather than the above
module, because it is more natural concept and to treat. It is important to consider the fundamental $ea\backslash y$

problems for -module and q-Weyl algebra. These studies have just started. As for
$D_{n}$
-module, see [SO1]. $\prime D_{n}$
84

\S 2. An application to the mechanical $theore\ln$ proving of binolnial identities and special func-
tion identities
Studies on mechanical theorem proving of mathematical theorems have been done since the beginning of
computer science and computers have proved many theorems which are hard to prove. For example, [Wul]
and [Choul] found algorithms and implemented systems to prove theorems on elementary geometry by using
the Wu-Ritt characteristic sets and the Grobner basis. The quantifier elimination method and the decision
procedure over the real numbers have been widely studied. As for special function identities, [Takl] found
an algorithm to find contiguity relations of special function by using the Grobner basis.
[Link] [Zeil] found a general method to prove binomial coefficients identities and special functions
identities; he explained that proving these identities reduces to solving the fundamental problems (1.4) and
(1.6).
Let us explain his ideal by using an example,
Our goal is to prove the identity
$\sum_{x_{2}=-\infty}^{\infty}$
$(\begin{array}{l}x_{1}x_{2}\end{array})=2^{x_{1}}$
.

We have shown in \S 2 that

$[E_{1}x_{1}-2(x_{1}+1)]$ $\sum_{x_{2}=-\infty}^{\infty}$
$(\begin{array}{l}x_{1}x_{2}\end{array})=0$

by computing the ideal intersection

$(\mathfrak{P}+(E_{2}-1)B_{2})\cap B_{2}$ .

Since $E_{1}x_{1}-2(x_{1}+1)=(x_{1}+1)(E_{1}-2)$ , we have

$[E_{1}x_{1}-2(x_{1}+1)]2^{x_{1}}=0$ ,

which means that the both sides of the identity satisfy the same difference equation. Hence, it is enough for
the proof to show that the both sides satisfy a same initial condition. In fact, when $x_{1}=0$ , we have $1=1$ .
It completes the proof.
The most important part of the above proof is to solve the fundamental problems (1.3) or (1.6). D.
Zeilberger gave an efficient algorithm, which is a modification of Gosper’s decision procedure ([Gol]), to
solve the problems (1.4) and (1.6) in case of annihilating ideals of binomial coefficients and hyperexponential
functions ([Zei2], [AZ1]). He has also a similar algorithm for the q-case. On the other hand, the author found
an algorithm, which is a modification of the Buchberger algorithm ([Buchl]), to solve the problems (1.1) $\sim$

(1.6) ([Tak2], [Tak3]). By using these algorithms, we can easily give proofs, for example, to the following
binomial identities;
$\sum_{k=0}^{n}k(\begin{array}{l}nk\end{array})=n2^{n-1}$ , $\sum_{k=0}^{n}\frac{1}{k+1}(\begin{array}{l}nk\end{array})=\frac{2^{n+1}-1}{n+1}$

$\sum_{k=0}^{n}(\begin{array}{l}nk\end{array})=\frac{(2n)!}{(n!)^{2}})$ $\sum_{k=0}^{p}(\begin{array}{l}nk\end{array})(\begin{array}{l}mp-k\end{array})=(\begin{array}{l}n+mp\end{array})$
, (Chu-Vandermonde),

$\sum_{k}(-1)^{k}(\begin{array}{l}n+bn+k\end{array})(\begin{array}{l}n+cc+k\end{array})(\begin{array}{l}b+cb+k\end{array})=\frac{(n+b+c)!}{n!b!c!}$ (Dixson).

\S 3. A new “fundamental problem” and its application

In the preceding sections, we considered mechanical theorems proving of binomial and special functions
identities. In this section, we will consider algorithms of deriving and proving connection formulas of special
functions and introduce a new fundamental problem in computational algebraic analysis.
Put . Let be an left ideal of the Weyl algebra
$\theta_{i}=x_{i}\partial_{i}$
. We will call also the following problem a
$\mathfrak{U}$
$A_{n}$

fundamental problem,
 85

Find

(3.1) $(\mathfrak{U}+x_{n}A_{n-1}\langle x_{n}, \theta_{n}\})\cap C\langle x_{1,}x_{n-1}, \theta_{1}\rangle$ ,

in particular, find the set of roots

$\{\theta\in C|^{\forall}f(x_{1}, \cdots, x_{n-1}, \theta)=0, f\in(3.1)\}$ .

We call the above set of roots exponents of the ideal along the line $x_{n}=0$ . $\mathfrak{U}$

We will give an efficient algorithm which give a sufficient answer for applications to the problem (3.1).
Computations of exponents along regular singularities, derivations and making proofs of connection formulas
and finding a sufficient answer to the fundamental problem (1.5) are applications of the problem (3.1) and
Algorithm 4.1 given in Section 4.
Before going to giving a general algorithm, we will explain the method of solving (3.1) and (1.5) and its
applications by using an example. The general procedure will be given in the next section. Our method is
a kind of the pruning.

It is known that the following identity holds.

If $c-b-1\not\in Z$ , then

(3.2) $z_{1}= \frac{\Gamma(c)\Gamma(c-(\iota-b)}{\Gamma(c-a)\Gamma(c-b)}z_{7}+\frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)}\approx 9$

where
$z_{1}= \sum_{m,n=0}^{\infty}\frac{(a)_{m+n}(b)_{m}(b’)_{n}}{(c)_{m+n}(1)_{m}(1)_{n}}x^{n}y^{n}$ ,

$z_{7}= \sum_{m,n=0}^{\infty}\frac{(a)_{m+n}(b)_{m}(b’)_{n}}{(1+a+b-c)_{m}(c-b)_{n}(1)_{m}(1)_{n}}(1-x)^{m}y^{n}$ ,

$z_{9}=(1-x)^{c-a-b} \sum_{m,n=0}^{\infty}\frac{(c-b)_{m+n}(c-a)_{m}(b’)_{n}}{(1+c-a-b)_{m}(c-b)_{n}(1)_{m}(1)_{n}}(1-x)^{m}y^{n}$ ,

$(\alpha_{1})_{k}=\alpha(\alpha+1)\cdots(\alpha+k-1)=\Gamma(\alpha+k)/\Gamma(\alpha)$

and $\Gamma(\alpha)$
is the Gamma function.

EXAMPLE 3.1 Prove the identity (3.2) by using the connection formula of the Gauss hypergeometric function
and the fact that the functions and are solutions of the system of differential equations
$z_{1},$ $z_{7}$ $z_{9}$

$l_{0}v=\ell_{1}u=\ell_{2}u=0$

where
$\ell_{0}=\theta_{x}(\theta_{x}+\theta_{y}+c-1)-x(\theta_{x}+\theta_{y}+a)(\theta_{x}+b)$

$\ell_{1}=\theta_{y}(\theta_{x}+\theta_{y}+c-1)-y(\theta_{x}+\theta_{y}+a)(\theta_{y}+b’)$

$l_{2}=(x-y) \frac{\partial^{2}}{\partial x\partial y}-b’\frac{\partial}{\partial x}+b\frac{\partial}{\partial y}$

$\theta_{x}=x\partial_{x}=x\frac{\partial}{\partial x}$
, $\theta_{y}=y\partial_{y}=y\frac{\partial}{\partial y}$
.

Solution. In order to prove the identity (3.2), we need the following lemma.
86

LEMMA 3.1 Let $P$ be an operator of the form

$P= \sum_{k=0}^{s}y^{k}p_{k}(\theta_{y}, x, \theta_{x})$

and $f$be a holomorphic function at $(x, y)=(0,0)$ such that $f(x, 0)\equiv 0$ and $Pf=0$ . Suppose that
does not depend on
$p_{0}=p_{0}(\theta_{y}, x)$ and $p_{0}(k, x)\neq 0$ for all non-negative integers . Then we have
$\theta_{x}$ $k$

$f(x, y)\equiv 0$ .

Proof. Since the function $f$ can be written as

$f= \sum_{k=0}^{\infty}y^{k}f_{k}(x)$ ,

we have
$Pf= \sum_{k=0}^{\infty}\sum_{i=0}^{s}y^{i+k}p_{k}(k, x, \theta_{x})f_{k}(x)=0$ .

Hence, we have the following difference equation

$p_{0}(k, x)f_{k}=-p_{1}(k-1, x, \theta_{x})f_{k-1}-\cdots-p_{s}(k-s, x, \theta_{x})f_{k-s}$ .

Since $f_{0}\equiv 0$ , then we have $f_{k}\equiv 0$ . $[$

Putting
$f=z_{1}- \frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}z_{7}-\frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)}\approx 9$

we will use Lemma 3.1. In order to use it, we need to find an operator of the form

$p_{0}(\theta_{y)}x)+y[\cdots\cdot\cdot]$

in the left ideal

$R\ell_{0}+RP_{1}+R\ell_{2}$ where $R=C\{x, y, \theta_{x}, \theta_{y}\}$ . This problem is the problem (3.1).
In order to find an operator of the form $p_{0}(\theta_{y}, x)+y[\cdot\cdot \cdot\cdot]$ , we need not to consider the higher order
terms with respect to the variable , because we have $y$

$\theta_{y}y=y(\theta_{y}+1)$ in $R$ .

In fact, the operators $l_{0},$ $l_{2}$

can be written as
$xyl_{2}=x\theta_{y}(\theta_{x}+b)-y\theta_{x}(\theta_{x}+b’)$

$=x\theta_{y}(\theta_{x}+b)+O(y)$

$\ell_{0}=\theta_{y}(\theta_{x}+\theta_{y}+c-1)+O(y)$

and we need to consider only

$l_{3}=x\theta_{y}(\theta_{x}+b)$ ,
$l_{4}=\theta_{y}(\theta_{f}+\theta_{y}+c-1)$ .
We compute
idea1 $(\ell_{3}, \ell_{4})\cap C(b, c)\{x, \theta_{y}\}$ .
This elimination can be done by the Buchberger algorithm for the ring of differential operators. In this case,
we can easily eliminate by taking the S-pair $\theta_{x}$

$x\ell_{3}-\ell_{4}=x\theta_{y}(\theta_{y}+c-1-b)\in C(b, c)\langle x,$ $\theta_{y}$

}.
 87

Finally, we have
$xxyP_{2}-\ell_{0}$

$=x(l_{3}+O(y))-(\ell_{4}+O(y))$
$=x\theta_{y}(\theta_{y}+c-1-b)+y[-x(\theta_{x}+\theta_{y}+a)(\theta_{y}+b’)+\theta_{x}(\theta_{y}+b’)]$ .
Here, $O(y)$ is a term expressed as . Hence, it follows from Lemma 3.1 that if we have
$y[\cdots\cdots]$ $f(x, 0)\equiv 0$ ,
then we complete the proof of (3.1).
The restricted function $f(x, 0)$ can be written as

$f(x, 0)= \sum_{m=0}^{\infty}\frac{(a)_{m}(b)_{m}}{(c)_{m}(1)_{m}}x^{m}$

$- \frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}\sum_{m=0}^{\infty}\frac{(a)_{m}(b)_{m}}{(1+a+b-c)_{m}(1)_{m}}x^{m}$

$- \frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)}(1-x)^{c-a-b}\sum_{m=0}^{\infty}\frac{(c-b)_{m}(c-a)_{m}}{(1+c-a-b)_{m}(1)_{m}}x^{m})$

which is equal to zero by reason of the connection formula of the Gauss hypergeometric function $([WW$
14.51]). It completes the proof of (3.2).

EXAMPLE 3.2 Put $b_{i}(x)=z_{i|_{y}=0}$ . Show that the function $b_{i}$

satisfies the Gauss hypergeometric differential
equation
$[\theta_{x}(\theta_{x}+c-1)-x(\theta_{x}+a)(\theta_{x}+b)]v=0$

by using only the fact that the function $z_{i}$

is a solution of the system of differential equation
$\ell_{0u=\ell_{1}u=l_{2}u=0}$ .

Solution. Notice that if we find an operator of the form

$p_{0}(x, \partial_{x})-yp_{1}(x, \partial_{x}, y, \partial_{y})\in \mathfrak{U}=R\ell_{0}+R\ell_{1}+R\ell_{2}$ ,

then we have $p_{0}(x, \partial_{x})b_{i}=0$ . Since

$\mathfrak{U}\ni l_{0}=\theta_{x}(\theta_{x}+c-1)-x(\theta_{T}+a)(\theta_{x}+b)+y[\partial_{y}\theta_{x}-x\partial_{y}(\partial_{x}+b)]$ ,

we have
$[\theta_{x}(\theta_{x}+c-1)-x(\theta_{x}+a)(\theta_{x}+b)]b;=0$ .

REMARK 3.2 Here, we have proved (3.2) by solving the fundamental problem (3.1). The next question
is how to derive connection formulas like (3.2). In fact, it is possible to derive connection formulas, with a
limitation, by solving the fundamental problems (3.1) and (1.5). The limitation is that we cannot obtain
higher order terms of connection formulas by the method. For example, we can find the following formula

$z_{1}= \frac{\Gamma(c)\Gamma(c-a-b)}{\Gamma(c-a)\Gamma(c-b)}(1+O(x-1, y))+\frac{\Gamma(c)\Gamma(a+b-c)}{\Gamma(a)\Gamma(b)}(1-x)^{c-a-b}(1+O(x-1, y))$

by the derivation method. We need new notions and theorems, which are explained in [Tak3], for the
derivation. Here, we only notice that we can derive connection formulas by solving the fundamental problelns
(3.1) and (1.5) and by using the notions and theorems in [Tak3].
88

REMARK 3.3 We can obtain the operator of the form

$p_{0}(x, \theta_{y})+yp_{1}(x, \partial_{x}ty, \theta_{y})$

by computing the intersection

$\mathfrak{U}\cap C\{x, y_{2}\theta_{y}\}$
.
The computation of the intersection above is more expensive than the method explained in Example 3.1,
because the elimination is done in the ring of 4 variables in this case and in the method explained in Example
3.1 the elimination is done in the ring of 3 variables. Notice that we can see that the operator
$x\theta_{y}(\theta_{y}-1)(b-c-\theta_{y}+1)+y\theta_{y}(\theta_{y}+b’)(ax-bx+c+x\theta_{y}+\theta_{y}-1)-y^{2}(\theta_{y}+b’)(\theta_{y}+b’+1)(\theta_{y}+a)$ ,

is in the left ideal

$R\ell_{0}+R\ell_{1}+RP_{2}$ .

\S 4. New algorithms
We will explain new efficient algorithms to solve the fundamental problems (3.1) and (15) in sufficient
forms for the application to proof and derivation of connection formulas. We assume the variables are $x,$ , $y$

and and the number of generators of a given left ideal is 2 for the simplicity.
$\partial_{x}$ $\partial_{y}$
$\mathfrak{U}$

Put
, $A_{2}’=C\{x,$ , .
$A_{2}=C\{x, y, \partial_{x}, \partial_{y}\}$ $y,$
$\theta_{x},$ $\partial_{y}\rangle$ $\tilde{A}=C\{\theta_{x},$
$y,$ $\partial_{y}\rangle$

For given operators $P,$ $Q\in A_{2}’$ , we have unique expansions

$P= \sum_{k=0}^{K_{1}}x^{k}p_{k}$
,

$Q= \sum_{k=0}^{K_{2}}x^{k}q_{k}$
, $p_{k},$
$q_{k}\in\tilde{A}$
.

Put $\mathfrak{U}=A_{2}P+A_{2}Q$ . Then the set is the left ideal of the ring . We will show an algorithm of solving
$\mathfrak{U}$
$A_{2}$

the fundamental problem (3.1) for the left ideal . $\mathfrak{U}$

We inductively define operators and $(j=0,1,2, . . )$ satisfying the condition $R_{i}^{(i)}$ $S_{j}^{(i)}$

$R_{j}^{(i)}P+S_{i}^{(i)}Q=O(x^{l+1})$

where $O(x^{J+1})$ denotes a term expressed as $x^{j+1}w\rangle$ $w\in A_{2}’$ . The space

$\{(R, S)|Rp_{0}+Sq_{0}=0, (R, S)\in\tilde{A}^{2}\}$

is the left $\tilde{A}$

module. For $j=0$ , let $(R_{0}^{(i)}, S_{0}^{(i)})(i=1, \cdots, g_{0})$

be generators of the left $\overline{A}$

module. For , we
$j$

put
$R_{j}^{(i)}= \sum_{k=1}^{9j-1}d_{jk}^{(i)}R_{j-1}^{(k)}+x^{j}r_{j}^{(i)}$
, $i=1_{\rangle}$ . ., $g_{j}$ ,

$S_{j}^{(i)}= \sum_{k=1}^{9j-1}d_{jk}^{(i)}S_{j-1}^{(k)}+x^{j}s_{j}^{(i)}$
, $i=1,$ $\cdots,$ $g_{j}$ .

Here, $(d_{j1}^{(i)}, \cdots , d_{jg}^{(i)_{j-1}}, r_{j}^{(i)}, s_{j}^{(i)}),$ $(i=1, \cdots, g_{j})$ are generators of the left $\tilde{A}$

module

$\{(d_{1}, \cdots, d_{g_{g-1)}}r_{j}, s_{j})\in\tilde{A}^{g+2}j-1|\sum_{k=1}^{g_{g-1}}d_{k}T_{j-1}^{(k)}+(r_{j}p_{0}+s_{j}q_{0})=0\}$

 89

where we define the operator $T_{j-1}^{(k)}\in\overline{A}$

as

$R_{j-1}^{(k)}P+S_{j-1}^{(k)}Q=x^{j}T_{j-1}^{(k)}+O(x^{j+1})$ .

Notice that we have

$R_{j}^{(i)}P+S_{j}^{(i)}Q=O(x^{j+1})$

from the definition. Eliminating the operator from , we can solve the fundamental $\partial_{y}$
$\{T_{j^{k}}|k=1, \cdot\cdot, g_{j}\}$

problem (3.1). The correctness will be proved in Theorem 4.1. Summarizing the above procedure, we have
the following algorithm.

ALGORITHM 4.1 findBoundaryData(P, $Q,$ ) $J$

(1) exponents $:=\emptyset$

(2) boundaryValues $:=\emptyset$

(3) $T_{-1}^{(1)}$
$:=p_{0}$

(4) $T_{-1}^{(2)}$
$:=q_{0}$

(5) $:=2$
$g_{-1}$

(6) exponents $:=findExponents(\{T_{-1}^{(k)}|k=1,2\})$

(7) boundaryValues $:=findBoundaryValues(exponents,\{T_{-1}^{(k)}|k=1_{\rangle}2\})$
(8) Let $(R_{0}^{(i)}, S_{0}^{(i)})(i=1, \cdots, g_{0})$ be generators of the left module: $\check{A}$

$\{(R, S)|Rp_{0}+Sq_{0}=0, (R, S)\in\tilde{A}^{2}\}$

(9) for $j$ $:=1$ to $J$

do
(10) $T_{i-}^{(k)_{1}}$
$:=the$ coefficient of $x^{j}$
of $R_{j-1}^{(k)}P+S_{i-}^{(k)_{1}}Q,$ $(k=1, \cdots, g_{j-1})$

$(* R_{j-1}^{(k)}P+S_{i-1}^{(k)}=x^{j}T_{j-1}^{(k)}mod x^{j+1}A_{2}’*)$

(11) exponents $:=findExponents(\{T_{j}^{(k)}|j=-1, \cdots, j-1, k=1, \cdots, g_{j-1}\})$

(12) boundaryValues $:=findBoundaryValues(\{T_{j}^{(k)}|j=-1, \cdots, j-1, k=1, \cdots, g_{j-1}\})$
(13) Let $(i=1, \cdots, g_{j})$ be generators of the left module
$(d_{j1}^{(i)}, \cdots, d_{jg}^{(i)_{j-1}}, r_{j}^{(i)}, s_{j}^{(i)}),$
$\tilde{A}$

$\{(d_{1}, \cdots, d_{gj-1}, r_{j}, s_{j})\in\tilde{A}^{9j-1}+2|\sum_{k=1}^{gj-1}d_{k}T_{j-1}^{(k)}+(r_{j}p_{0}+s_{j}q_{0})=0\}$

(14) Return(exponents, boundaryValues)

ALGORITHM 4.2 findExponents $( \{T_{i}|i=1, \cdots\ell\})$

(1) $G:=the$ Gr\"obner basis of in by the order $\{T_{i}\}$
$\tilde{A}$

$\{y, \partial_{y}\}>\theta_{x}$

(2) $G$ $:=G\cap C\langle y,$ } $\theta_{x}$

(3) Return $(\{\theta\in C|f(y, \theta)=0, \forall f\in G\})$

ALGORITHM 4.3 findBoundaryValues(exponents, $\{T_{i}|i=1,$ $\cdot$

. , $l\}$ )
(1) $B$ $;=\emptyset$

(2) for exponents do

$e\in$

(3) $G$ $:=\{T_{i|_{\theta_{x}arrow e}}|i=1_{\rangle}\cdots, l\}$

(4) $:=B\cup$ {Gr\"obner basis of

$B$
} $G$

(5) Return(B)
90

Suppose that there exist $R_{I}’S’\in A_{2}$ such that

$R’P+S’Q=T(\theta_{x}, y)+xU(x, y)\theta_{\tau},$ $\partial_{y}$

).

In this case, there exists $N$ satisfying

$x^{N}R’x^{N}S’)\in A_{2}’=C\{x,$ $y,$ $\theta_{x)}\partial_{y}\rangle$

.

The following theorem shows that if the number is sufficiently large, then Algorithm4.1 (findBoundaryData(P, $Q,$ )
$J$ $J$

returns the exponents and solves the fundamental problem (3.1). Notice that the upperbound of the number
$)$

is not known.
$J$

THEOREM 4.1 If there exist a number $N$ , an $operatorT\in C\{\theta_{x},$ $y\rangle$

and operators $P,$ $Q,$ $R,$ $S\in C\langle x, \theta_{x}, y, \partial_{y}\rangle$

such that
$RP+SQ=x^{N+1}T(\theta_{x}, y)+O(x^{N+2}))$

then the operator $T$

is the member of the left ideal of $\tilde{A}=C\{\theta_{x}, y, \partial_{y}\}$
generated by

$T_{N}^{(k)},$
$k=1,$ $\cdots,$ $g_{N},$
$T_{-1}^{(1)}=p_{0},$ $T_{-1}^{(2)}=q_{0}$
.

We need a lemma to prove the theorem.

LEMMA 4.1 Let $(R, S)$ be a solution of

$RP+SQ=O(x^{N+1})$ .

Then the solution $(R, S)$ can be expressed as

$(R, S)= \sum_{k=1}^{g_{N}}c_{k}^{N}(R_{N}^{(k)}, S_{N}^{(k)})+O(x^{N+i})$ , $c_{k}^{N}\in\tilde{A}$

.

Proof. We can suppose that the degrees of and are $N$ with respect to the variable $R$ $S$ $x$ . We use the
induction on $N$ . We have the conclusion in case of $N=0$ from the definition of and $R_{0}^{(k)}$ $S_{0}^{(k)}$
. We express
the operators $R$
andS as
$R=R’+x^{N}r$ , $S=S’+x^{N}s$

where the degrees of $R’$ and $S’$

with respect to $x$ are less than $N-1$ . Since

$RP+SQ=R’P+S’Q+x^{N}(rp_{0}+sq_{0})+O(x^{N+1})$ ,

we have $R’P+S’Q=0(x^{N})$ . It follows from the assumption of the induction that

$(R’, S’)= \sum_{i=1}^{g_{N-1}}c_{i}^{N-1}(R_{N-1)}^{(i)}S_{N-1}^{(i)})$ , $c_{i}^{N-1}\in\tilde{A}$

.

Therefore $R’P+S’Q$ can be written as

$R’P+S’Q=x^{N} \sum_{i=1}^{g_{N-1}}c_{i}^{N-1}T_{N-1}^{(i)}+O(x^{N+1})$ .
 91

Since $R’P+S’Q+x^{N}(rp_{0}+sq_{0})=O(x^{N+1})$ , we have

$\sum_{i=1}^{g_{N-1}}c_{i}^{N-1}T_{N-1}^{(i)}+rp_{0}+sq_{0}=0$ .

Any solution
$(c_{1}^{N-1} , )c_{g_{N-}}^{N-1_{1}},$ $r,$ $s$
)
of the above linear indefinite equation can be expressed as

$\sum_{i=1}^{Ji}d_{i}(d_{N1}^{(i)}, \cdots, d_{N}^{(i)_{JN-1}}, r_{N}^{(i)}, s_{N}^{(i)}))$

$d_{i}\in\tilde{A}$
.

We have the conclusion from the above and the definition of $(R_{N}^{(k)}, S_{N}^{(k)})$
. $[$

Proof of Theorem 4.1. The operators $R$ and $S$

can be expressed as

$R=R’+x^{N+1}r+O(x^{N+2})$ ,
$S=S’+x^{N+1}s+O(x^{N+2})$ ,

where the degree of $R’$ and $S’$ with respect to $x$ is less than $N$ . It follows from Lemma 4.1 that there exist
such that
$c_{k}^{N}\in\overline{A}$

$R’P+S’Q= \sum_{k=1}^{g_{N}}c_{k}^{N}T_{N}^{(k)}+O(x^{N+2})$ .

Since we have
$RP+SQ=x^{N+1}T(\theta_{x}, y)+O(x^{N+2})$ ,

then
$\sum_{k=1}^{g_{N}}c_{k}^{N}T_{N}^{(k)}+rp_{0}+sq_{0}=T(\theta_{x}, y))$

which means that

$T( \theta_{x}, y)\in\sum\tilde{A}T_{N}^{(k)}+\tilde{A}p_{0}+\tilde{A}q_{0}.[$

Let be one of the exponents of the ideal . In order to derive connection formulas, we need to find
$\alpha$
$\mathfrak{U}$

the left ideal for the boundary values with respect to ; $\alpha$

$(x^{-\alpha}\mathfrak{U}x^{\alpha}+xA_{2})\cap C\langle y,$ $\partial_{y}$

}.
The above problem is the fundamental problem (1.5). We gave algorithms for solving (1.5) in [Tak2] and
[Tak3]. Algorithm 4.1 does not always give correct answer, but it is faster than the algorithms given in
[Tak2] and [Tak3]. Let $T’$ be the generator of

$(x^{-\alpha}\mathfrak{U}x^{\alpha}+xA_{2})\cap C\langle y,$ $\partial_{y}$

}.
The following theorem gives a sufficient condition for the correctness of Algorithm 4.1 –findBoundaryData $($

$P,$ $Q,$ $J$

).

THEOREM 4.2. If there exist operators $l$

and $U$
satisfying

$l=T’+xU,$ $l\in x^{-\alpha}\mathfrak{U}x^{\alpha},$ $U\in A_{2}’=C\{x, \theta_{x}, y, \partial_{y}\}$ ,

92

then for sufficiently large number , Algorithm $J$

4.1 returns a generator of the left ideal for the boundary value
with respect to the exponent . $\alpha$

Proof. Without loss of generality, we assume that $\alpha=0$ . It is enough to prove that if there exists a
number $N$ such that
$RP+SQ=x^{N+1}[T(y, \partial_{y})+\theta_{x}U(\theta_{x}, y)\partial_{y})]+O(x^{N+2})$,

$T,$ $U,$ $R,$ $S\in C\{x,$ $\theta_{x},$

$y,$
$\partial_{y}\rangle$
,

then the operator $T$

is the member of the left ideal of $C\{y, \partial_{y}\}$
generated by

$T_{N1_{\theta_{x}=0}}^{(k)},$ $(k=1, \cdots, g_{k}),$ $T_{-11_{\theta_{x}=\cap}}^{(k)},$

$(k=1,2)$ .

We can see, in a similar way to the proof of Theorem 4.1, that

$T(y, \partial_{y})+\theta_{x}U(\theta_{x}, y, \partial_{y})\in\sum\tilde{A}T_{N}^{(k)}+\sum\tilde{A}T_{-1}^{(k)}$

Putting $\theta_{x}=0$ in the above, we obtain the conclusion. fi

REMARK. 4.1. (Complexity) The complexity of Algorithm 4.1 is a research problem. We only notice the
following observation. Let , be elements of
$l_{1}$
. Put $d= \max_{i=1,\cdot\cdot r)}deg\ell_{j}$ . Suppose that
)
$l_{r}$ $C\langle\theta_{x}, y, \partial_{y}\rangle$

$T(d)$ and $D(d)$ are upper bounds of the time of the construction of the Gr\"obner basis of , by the $\ell_{1},$ $\cdot$ $\ell_{r}$

order
$\partial_{y}>\{\theta_{x}, y\}$

and the maximum of the degrees of elements in the Gr\"obner basi, . Put $d_{0}= \max\{\deg(p_{0}), \deg(q_{0})\}$ . Then $h$

an upper bound of the execution time of Algorithm 4.1 is $\sum_{k=0}^{J}T(D_{k})$ where

$D_{k}= \max\{Do\cdots oD(d_{0}), D_{k-1}\}\bigvee_{k}$

.

EXAMPLE 4.1. (The Appell function $F_{1}.$

) Put

$P=\theta_{x}(\theta_{x}+y\partial_{y}+c-1)-x(\theta_{x}+y\partial_{y}+a)(\theta_{x}+b)$ ,
$Q=y\partial_{y}(\theta_{x}+y\partial_{y}+c-1)-y(\theta_{x}+y\partial_{y}+a)(y\partial_{y}+b’)$ .

We call findBoundaryData(P,Q,-l). In this case, we have

$p_{0}=\theta_{x}(\theta_{x}+y\partial_{y}+c-1)$ ,
$q_{0}=y\partial_{y}(\theta_{x}+y\partial_{y}+c-1)-y(\theta_{x}+y\partial_{y}+a)(y\partial_{y}+b’)$ .

Let $G$ be the Gr\"obner basis of the left ideal generated by $p_{0}$ and $q_{0}$ by the order $\{y, \partial_{y}\}>\theta_{x}$
. We have

$G\cap C\langle y,$ $\theta_{x}$

} $\ni(a-c+1)y\theta_{x}(\theta_{x}+c-b’-1)$ .

In fact, we have
$t_{1}$ $:=y\cdot y\partial_{y}p_{0}+\theta_{x}q_{0}$ ,
$t_{2}$ $:=t_{1}-y\partial_{y}p_{0}$ ,

$t_{3}$
$:=t_{2}-y(c-1-b’-a)p_{0}$ ,
$=(a-c+1)y\theta_{x}(\theta_{x}+c-b’-1)$ .
 93

Hence we have exponents $=\{0,1+b’-c\}$ . Since we have

$p_{0|_{\theta_{x}=0}}=0$ ,
$q_{0|_{\theta_{x}=0}}=y\partial_{y}(y\partial_{y}+c-1)-y(y\partial_{y}+a)(y\partial_{y}+b’)$

and
$p_{0|_{\theta_{x}=1+b’-c}}=(-c+b’+1)(y\partial_{y}+b’)$ ,
$q_{0|_{\theta_{x}=1+b’-c}}=[y\partial_{y}-y(y\partial_{y}+a-c+b’+1)](y\partial_{y}+b’)$ ,
then
boundaryValues $=\{q_{0|_{\theta_{x}=0}}, p_{0|_{\theta_{f}=1+b-c}},\}$ .

We can prove that the above “boundaryValues” are the maximal annihilating left ideal of $t11\Leftrightarrow$

boundary
values by using [Tak3; Proposition 2.1].

EXAMPLE 4.2 Put

$P=\theta_{x}(\theta_{x}+y\partial_{y}+c-1)-x(\theta_{x}+y\partial_{y}+a)(\theta_{x}+b))$

$Q=-y\theta_{x}(y\partial_{y}+b’)+xy\partial_{y}(\theta_{r}+b)$ .

We call findBoundaryData(P, $Q,$ ). Notice that the operators

$0$
and above generates the same left ideal $P$ $Q$

with the ideal generated by and in Example 4.1 in the ring $C(a, b, b’, c, x, y)\{\partial_{x}, \partial_{y}\}$ .
$P$ $Q$

Put
$p_{0}=\theta_{x}(\theta_{x}+y\partial_{y}+c-1)$ ,
$q_{0}=-y\theta_{x}(y\partial_{y}+b’)$ .

The Gr\"obner basis $G$

of the ideal generated by $p_{0}$ and $q_{0}$ by the order $\{y, \partial_{y}\}>\theta_{x}$
is

$\{\theta_{x}(\theta_{x}+y\partial_{y}+c-1), y\theta_{x}(\theta_{x}+c-1-b’), -\theta_{x}(\theta_{x}-2+c)(-\theta_{x}+c+b’-c)\}$ .

Hence, the step (6) returns $\{0,1+b’-c\}$ and the step (7) returns $\{0, (1+b’-c)(y\partial_{y}+b’)\}$ . In this case, the
ideal for the boundary value with respect to the exponent is not correct. The step (8) computes generators $0$

of the module, which can be easily obtained from the Grobner basis of and . In fact, we have $p_{0}$ $q_{0}$

$(R_{0}^{(1)}, S_{0}^{(1)})=(y(y\partial_{y}+b’), y\partial_{y}+c-2+\theta_{x}))$

$(R_{0}^{(2)}, S_{0}^{(2)})=$ . .,

In the step (10), we have

$T_{-1}^{(1)}=-y(y\partial_{y}+b’)(\theta_{x}+y\partial_{y}+0)(\theta_{x}+b)+(y\partial_{y}+c-1+\theta_{x})y\partial_{y}(\theta_{x}+b)$
,
$T_{-1}^{(2)}=\cdot$ .

The step (12) returns the correct ideal for the boundary value with the exponent .
$0_{\backslash }$

$b[y\partial_{y}(y\partial_{y}+c-1)-y(y\partial_{y}+a)(y\partial_{y}+b’)]$ .

EXAMPLE 4.3. (The zonal spherical system on $SL(3,$ $R)/SO(3)$ )

Consider the system of differential equation ([Sekl; $14p]$ )

(4.1) $(\triangle_{2}-L_{2})u=(\Delta_{3}+L_{3})v=0$
94

where $L_{2},$ $L_{3}\in C$ and

$\Delta_{2}=-4(\theta_{1}^{2}-\theta_{1}\theta_{2}+\theta_{2}^{2})$

$+(2 \frac{1+x_{1}}{1-x_{1}}-\frac{1+x_{2}}{1-x_{2}}+\frac{1+x_{1}x_{2}}{1-x_{1}x_{2}})\theta_{1}$

$+(2 \frac{1+x_{2}}{1-x_{2}}-\frac{1+x_{1}}{1-x_{1}}+\frac{1+x_{1}x_{2}}{1-x_{1}x_{2}})\theta_{2}-1$ ,

$\triangle_{3}=-8\theta_{1}\theta_{2}(\theta_{1}-\theta_{2})$

(4.2)
$+2( \frac{1+x_{2}}{1-x_{2}}+\frac{1+x_{1^{X}2}}{1-x_{1}x_{2}})\theta_{1}^{2}+4(\frac{1+x_{1}}{1-x_{1}}-\frac{1+x_{2}}{1-x_{2}})\theta_{1}\theta_{2}$

$-2( \frac{1+x_{1}}{1-x_{1}}+\frac{1+x_{1^{X}2}}{1-x_{1}x_{2}})\theta_{2}^{2}$

$- \frac{2(1+x_{1})(1-x_{1}x_{2}^{2})}{(1-x_{1})(1-x_{2})(1-x_{1}x_{2})}\theta_{1}+\frac{2(1+x_{2})(1.-x_{1}^{2}x_{2})}{(1-x_{1})(1-x\underline{)})(1-x_{1}x_{2})}\theta_{2}$
.

Here, we use the notation and Changing the independent variables

$\theta_{i}=x_{i}\partial_{i}$
into
$\partial_{i}=\frac{\partial}{\partial x_{t}}$ $(x_{1}, x_{2})$

$(x”, y”)=T_{1}oT_{2}(x_{1}, x_{2})$ where $T_{1}(x’, y’)=(x’-1, y’-1)$ and $T_{2}(x_{1}, x_{2})=(1/x_{1}, x_{2})$ , we obtain a system

of differential equations of which locus of singularities is

$\mathcal{P}_{L}$

$\{(x’’, y’’)|x’’y’’(x’’-1)(y’’-1)(x" -y’’)=0\}$ .

Notice $(0,0)=T_{1}oT_{2}(1,1)$ . Blowing-up $(x”, y”)$ -plane at the origin, we naturally obtain the following system
of differential equations from the system : $\mathcal{P}_{L}$

(4.3) $Pu=Qu=0$
where
$P=p_{0}+xp_{1}+x^{2}p_{2}$

$Q=q_{0}+xq_{1}+x^{2}q_{2}+x^{3}q_{3}$

$p_{0}=2(2y^{3}\theta_{x}^{2}-4y^{3}\theta_{x}\theta_{y}+2y^{3}\theta_{y}^{9}\sim-2y^{2}\theta_{x}^{2}+6y^{2}\theta_{x}\theta_{y}-4y^{2}\theta_{y}^{2}+$

$y^{2}\theta_{y}-2y\theta_{x}\theta_{y}-y\theta_{x}+4y\theta_{y}^{2}+y\theta_{y}-2\theta_{y}^{2})$

$p_{1}=2y(4y^{2}\theta_{x}^{2}-6y^{2}\theta_{x}\theta_{y}+y^{2}\theta_{x}+2y^{2}\theta_{y}^{2}-4y\theta_{x}^{2}+8y\theta_{x}\theta_{y}-2y$

$\theta_{x}+4y\theta_{y}-2\theta_{x}\theta_{y}-\theta_{x}-2\theta_{y}^{2})$

$p_{2}=y^{2}(L_{2}y-L_{2}+4y\theta_{x}^{2}-4y\theta_{x}\theta_{y}+2y\theta_{x}+4y\theta_{y}^{2}+2y\theta_{y}+y-4\theta_{x}^{2}$

$+4\theta_{x}\theta_{y}-4\theta_{x}-4\theta_{y}^{2}+2\theta_{y}-1)$

$q_{0}=4(2y^{2}\theta_{x}^{2}\theta_{y}+2y^{2}\theta_{x}^{2}-4y^{2}\theta_{x}\theta_{y}^{2}-4y^{2}\theta_{x}\theta_{y}+2y^{2}\theta_{y}^{3}$

$+2y^{2}\theta_{y}^{2}-2y\theta_{x}^{2}\theta_{y}-y\theta_{x}^{2}+6y\theta_{x}\theta_{y}^{2}+2y\theta_{x}\theta_{y}-4y\theta_{y}^{3}-2\theta_{x}\theta_{y}^{2}+2\theta_{y}^{3}-2\theta_{y}^{2})$

$q_{1}=2(4y^{3}\theta_{x}^{2}\theta_{y}+2y^{3}\theta_{x}^{2}-8y^{3}\theta_{x}\theta_{y}^{2}-4y^{3}\theta_{x}\theta_{y}+4y^{3}\theta_{y}^{3}$

$+2y^{3}\theta_{y}^{2}+4y^{2}\theta_{x}^{2}\theta_{y}+8y^{2}\theta_{x}^{2}-12y^{2}\theta_{x}\theta_{y}+2y^{2}\theta_{x}-4y^{2}\theta_{y}^{3}$

$+8y^{2}\theta_{y}^{2}-y^{2}\theta_{y}-8y\theta_{x}^{2}\theta_{y}-4y\theta_{x}^{2}+12y\theta_{x}\theta_{y}^{2}+4y\theta_{x}\theta_{y}-y\theta_{x}$

$-4y\theta_{y}^{3}-8y\theta_{y}^{2}-y\theta_{y}-4\theta_{x}\theta_{y}^{2}+4\theta_{y}^{3}-2\theta_{y}^{2})$

$q_{2}=2y(8y^{2}\theta_{x}^{2}\theta_{y}+4y^{2}\theta_{x}^{2}-12y^{2}\theta_{x}\theta_{y}^{2}-4y^{2}\theta_{x}\theta_{y}+y^{2}\theta_{x}$

$+4y^{2}\theta_{y}^{3}+2y^{2}\theta_{y}^{2}-4y\theta_{x}^{2}\theta_{y}+4y\theta_{x}^{2}+12y\theta_{x}\theta_{y}^{2}-8y\theta_{x}\theta_{y}$

$+2y\theta_{x}-8y\theta_{y}^{3}-4y\theta_{y}-4\theta_{x}^{2}\theta_{y}-2\theta_{x}^{2}-\theta_{x}+4\theta_{y}^{3}-2\theta_{y}^{2})$

$q_{3}=-y^{2}(L_{3}y-L_{3}-8y\theta_{x}^{2}\theta_{y}-4y\theta_{x}^{2}+8y\theta_{x}\theta_{y}^{2}-2y\theta_{x}+4y\theta_{y}^{\gamma}\sim$

$+2y\theta_{y}+8\theta_{x}^{2}\theta_{y}-8\theta_{x}\theta_{y}^{2}+8\theta_{x}\theta_{y}-4\theta_{y}^{2}+2\theta_{y})$

where $x”=x$ and $y”=xy$ . Notice that $x=0$ is the exceptional curve of the blowing-up.
In this example, we obtain the exponents and boundary values of the system (4.3) along the line $x=0$
which were not known ([Sekl]).
 95

PROPOSITION 4.1
(1) If there exists a solution of (4.3) around the origin of the form

$x^{\mu} \sum_{k=0}^{\infty}a_{k}(y)x^{k}$ , $a_{0}(y)\not\equiv 0$ ,

then $\mu=0or\pm\frac{1}{2}$ .
(2) $If\mu=0_{f}$ then the function $a_{0}(y)$
satisfies the differential equation

$\partial_{y}a_{0}(y)=0$ .

If $\mu=\frac{1}{2}$
then the function $a_{0}(y)$
satisfies the differential equation

$p_{01}a_{0}(y)=0$ .

If $\mu=-\frac{1}{2}$ , then the function $a_{0}(y)$

satisfies the differential equation

$p_{02}a_{0}(y)=0$ .

Here,
$p_{01}=p_{01_{\theta_{\mathcal{I}}=1/2}}$ ,
$=4y^{3}\theta_{y}^{2}-4y^{3}\theta_{y}+y^{3}-8y^{2}\theta_{y}^{2}+8y^{2}\theta_{y}-y^{2}+8y\theta_{y}^{2}-y-4\theta_{y}^{2}$

$=y^{3}(2\theta_{y}-1)^{2}-y^{2}(8\theta_{y}^{2}-8\theta_{y}+1)+y(8\theta_{y}^{2}-1)-4\theta_{y}^{2}$

$p_{02}=p_{01_{\theta_{x}=-1/2}}$ ,
$=4y^{3}\theta_{y}^{2}+4y^{3}\theta_{y}+y^{3}-8y^{2}\theta_{y}^{2}-4y^{2}\theta_{y}-y^{2}+8y\theta_{y}^{2}+4y\theta_{y}+y-4\theta_{y}^{2}$

$=y^{3}(2\theta_{y}+1)^{2}-y^{2}(8\theta_{y}^{2}+4\theta_{y}+1)+y(8\theta_{y}^{2}+4\theta_{y}+1)-4\theta_{y}^{2}$

Proof of (1). Put

$R=-(2y^{4}\partial_{y}-4y^{3}\partial_{y}-4y^{3}+4y^{2}\partial_{y}+7y^{2}-2y\partial_{y}-4y\theta_{x}-5y+2\theta_{x}+2)P$

$+(y^{4}-2y^{3}+3y^{2}-2y+1)Q$

and
$S=(6y^{4}\partial_{y}-12y^{3}\partial_{y}-8y^{3}\theta_{x}-24y^{3}+12y^{2}\partial_{y}+12y^{2}\theta_{x}+39y^{2}-6y\partial_{y}-33y-2\theta_{x}+12)R$

$+(y^{2}(4y\theta_{x}^{2}-y-4\theta_{x}^{2}+1))9(y-1)P$ .

Then the operator $S$

has the following form:
$S=s_{0}+xs_{1}+x^{2}s_{2}+x^{3}s_{3}$

where
$s_{i}\in C\{\theta_{x}, y)\partial_{y}\}$ ,
$s_{0}(\theta_{x}, y)=4y^{2}(y^{2}-y+1)^{2}(y-1)\theta_{x}^{2}(2\theta_{\tau}-1)(2\theta_{x}+1)$

and
$s_{i}(k, y, \partial_{y})\not\equiv 0,$ $s_{i}(k+1/2, y, \partial_{y})\not\equiv 0_{7}s_{?}\cdot(k-1/2, y, \partial_{y})\not\equiv 0$

for $i=1,2,3$ and $k\in\{0,1,2, \cdots\}$ . Since

$s_{0}( \mu+k, y)a_{k}(y)=-\sum_{i=1}^{3}s_{k}(\mu+k-i, y, \theta_{y})a_{k-i}$ ,

96

the parameter must be $0or\pm

$\mu$ 1/2$ .

Proof of (2). Omitted.

REMARK
$p_{01}=4y(y-1)(y^{2}-y+1)\partial_{y}^{2}+4(2y-1)\partial_{y}+(y^{2}-y-1)$

and
$p_{02}=4y(y-1)_{C}) \frac{\tau)}{y}+4(2y-1)\partial_{y}+1$ .

The singularity $y^{2}-y+1$ must be apparent.

References
[Buchl] Buchberger, B., Gr\"obner bases: An Algorithmic method in Polynomial Ideal Theory, N.K. Bose (ed.),
Multidimensional Systems Theory, (1985) 184-232.
[Choul] Chou, S.C., Mechanical Geometry Theorem Proving, D Reidel Publishing Company (1988).
[Sekl] Sekiguchi, J., Global representations of solutions to zonal spherical systems on
$SL(3)/SO(3)$ . preprint, University of electro-communications, Tokyo.
[ ul] Wu, W., Basis Principles of Mechanical Theorem Proving in Geometries, J. of Sys. Sci. and Math. Sci.
$1(V$

4(3), (1984) 207-235.

See also “a bibliography of computational algebraic analysis” below.

A bibliography of computational algebraic analysis

[AZ1] Almkvist,G., Zeilberger,D., The method of differentiating under the integral sign, Journal of Symbolic
computation 10, (1990), 571-591. (Gosper-Zeilberger’s Algorithm, Algorithm of computing differential
operators for definite integrals, Mecanical theorem proving of special function identities)
[Ber] Bernstein, I.N., The analytic continuation of generalized functions with respect to a parameter. Funk.
Anal. 6(1972), 26-40. (theory of holonomic system)
[Bjo] Bj\"ork, J.E., Rings of Differential Operators. North-Holland, New York, 1979. (theory of holonomic
system)
[Cas] Castro,F., Calculs effectifs pour les id\’eaux ‘op\’erateurs diff\’erentiells, G\’eom\’etrie alg\’ebrique et application
$d$

III. J.M. Aroca, T. Sancheg-Giralda, J.L. Vicen $teeds.$ , Tra vaux en Cours 24 (1987), 1-19. (theoretical
study of Grobner basis for ideals of the ring ) $D_{n}$

[Ekl] Ekhad, S., A very short proof of Dixon’s theorem. to appear in J. $Com$ \’oinatorics theory (an application
of Gosper-Zeilberger’s algorithm, Mecanical theorem proving of binomial identities)
[Ek2] Ehhad, S., A one-line proof of the Habsieger-Zeilberger constant term identity. to appear in .
$G_{2}$ $J$

Comp. an Appl. Math (an application of Gosper-Zeilberger’s algorithm, Mecanical theorem proving
$d$

of binomial identities)
[Gal] Galligo, A., Some algorithmic questions on ideals of differential operators. Lect. No in Comp. Sci., $te$

204(1985), 413-421. (Grobner basis for the rings of differential operators)

[Gol] Gosper, R.W., Jr., Decision procedure of indefinite summation. Proc. Natl. Acad. Sci. USA 75, 40-42.
(Gosper’s decision procedure)
[Kal] Kashiwara, M., Systems of microdifferential equations, Birkh\"auser, (1983), (theory of holonomic $D$

module)
[Nou] Noumi, M., Wronskian determinants and the Grobner representation of a linear differential equation,
in Alge braic Analysis. Ed. by M. Kashiwara, T. Kawai. Academic Press, 1989. (An application of
Grobner basis of the ring of differential operators for the soliton theory)
[Olv] Ollivier, F., Le probl\’em de l’identifiabilit\’e structurelle globalle: approche th\’eorique, m\’ethodes effec-
tives et bornes de complexit\’e. Th\‘ese, \’Ecole polytechnique (1990). (Algorithm, implementation and
application on the ideal of differential fields)
[SO1] Simoyama and Oaku, in this volume. (Algorithm of computing Grobner basis of ideals of the ring ,
$D_{n}$

Approximation method of powerseries)

 97

[Takl] Takayama, N., Gr\"obner basis and the problem of contiguous relation. $Jap$ an J. Appl. Math. 6(1989),
147-160. (An application of Grobner basis of the ring of differential-difference operators to the derivation
of contiguity relation)
[Tak2] Takayama,N., An algorithm of constructing the integral of a module –an infinite dimensional analog
of Grobner basis. Proceedings of the international symposium on symbolic and algebraic computation,
ACM press and Addison-Wesley (1990), 206-211. (Algorithm of computing differential operators for
definite integrals, An extension of the Buchberger algorithm, Mecanical theorem proving of binomial
and speical function identities)
[Tak3] Takayama, N., Boundary values and connection formulas of holonomic systems with regular singulari-
ties, Special differential equations, Proceedings of the Taniguchi workshop 1991, Edited by [Link],
Distributed from Kyushu University (1991), 125-149. (Mecanical derivation and proving of connection
formula, Algorithm of computing restrictions of differential operators)
[Tak4] Takayama,N., ODE section algorithm by Grobner basis (in Japanese). RIMS Kokyurok . Kyoto Univ.
$u$

681 (1989), 1-26. (Algorithm of computing restrictions of differential operators)

[Tak5] Takayama, N., An approach to the zero recognition problem by Buchberger algorithm. to appe $ar$ .
(Mechanical theorem proving of special function identities, An extension of the Buchberger algorithm)
[Tak6] Takayama, N., Mechanical theorem proving of identities. (In Japanese) Algebraic Aspect of special
functions, Edited by [Link] , Mathematica (Tokyo), (1989), 143-190. (Tutorial, Mechanical theorem
proving of binomial and special function identities by an extension of the Buchberger algorithm)
[Tak7] Takayama, N., Gr\"obner basis, Integration and Transcendental Functions, Proceedings of the interna-
tional symposium on symbolic and algebraic computation, ACM press and Addison-Wesley (1990),
152-156. (Survey, Algorithm of computing differential operators for definite integrals (Implementation
$))$

[UT] Tajima,S., Uchida,M., Computation of an integral of deRham system corresponding to a resolution of

singularity (in Japanese). RIMS Kokyuroku 693 (1989), 41-68. (Algorithm of computing differential
operators for definite integrals, An example)
[WZ1] Wilf, H.S. and Zeilberger, D., Rational functions certify combinatorial identities, J. of $AMS$ . (1990)
$3$

147-158. (Abstract, Mechanical theorem proving of binomial identities)

[WZ2] Wilf, H.S. and Zeilberger, D., Towards computational proofs of identities, Bull. of $AMS$ (1990). (Ab-
stract, Mechanical theorem proving of binomial identities, Gosper-Zeilberger’s Algorithm)
[Zeil] Zeilberger,D., A holonomic systems approach to special functions identities. (to appear in J. Comp.
Appl. Math.) (Foundation on mechanical theorem proving of binomial and special function identities,
Sylvester’s dialitic elimination)
[Zei2] Zeilberger,D. (1989). The method of creative telescoping. Journal of symbolic computation 11, (1991)
195-204. (Gosper-Zeilberger algorithm, Algorithm of computing difference operators for sums)
[Zei3] Zeilberger,D. (1988). A unified approach to Macdonald’s root-system conjectures. SIAM J. Math Anal.
19, 987-1013. (Computer aided proof of q-identities)
[Zei4] Zeilberger,D. (1990). A fast algorithm for proving terminating hypergeometric identities. Discrete math.
80, 297-211. (Algorithm of computing difference operators for sums)