Funkcialaj Ekvacioj, 19 (1976) 175-189 On a Linear System of Pfaffian Equations with Regular Singular Points By Masaaki YOSHIDA and Kyoichi TAKANO* (Kobe University, Japan) Introduction. The purpose of this paper is to extend some well known theorems on ordinary differential equations to Pfaffian systems. We consider a completely integrable system (1) $dX=(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{i})X$ where $X$ is an unknown $m¥times m$ matrix and $P_{i}(x)$ is an at $x=0$ , i.e. $P_{i}(x)$ permits a power series expansion $P_{i}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ $m¥times m$ matrix holomorphic , convergent in a neighborhood of $x=0$ . Here denotes a multi-index , nonnegative integer, $0=(0, ¥cdots, 0)$ and . For two multi-indices and , $k¥geq l’$ means for all and $k>l’$ means $k¥geq l$ and for some . We propose to construct the solutions of (1) concretely and to find out the dominant coefficients of which determine the local behavior of the solutions. As in the local theory of ordinary differential equations, our study is divided into two parts: (i) to find out a formal transformation which reduces (1) to a solvable equa$k$ $(k_{1^{ }},¥cdots, k_{n})$ $k$ $x^{k}=x_{l^{l}}^{k}¥cdots x_{n^{n}}^{k}$ $l$ “ ’ “ $k_{i}¥geq l_{i}$ $i’$ ’ “ ’ $k_{i}$ “ $k_{i}>l_{i}$ $i’’$ $¥{P_{i,k}¥}$ tion, to prove the convergence of the formal transformation. We call the former the formal part and the latter the analytic part. The analytic part for system (1) is easily achieved; in fact, the technique is the same as in that for ordinary differential equations. However, the formal part for system (1) contains an essential difficulty, which does not appear in that for ordinary differential equations, because the recursion formulas to determine the coefficients of a formal tronsforma- (ii) * This paper contains the complete proofs of the results published in the Proceedings of the Japan Academy, 51 (1975), 219-223. 176 M. YOSHIDA and K. TAKANO tion and the integrability condition are extremely complicated for our case. Thus our effort is concentrated on making clear the effect of a transformation on the consequent equation and the role of integrability condition. We show that the complication can be resolved by decomposing a transformation into a sequence of simple transformations of special form. As far as the authors know, the idea of decomposing a transformation in the formal theory is due to Professor M. Hukuhara. In Section 1, we prepare, in advance, a convergence theorem guaranteeing that the formal transformations considered in this paper are actually convergent. , In Section 2, we state the integrability condition in terms of the coefficients since the expression of the integrability condition in this form is required to achieve the calculations below. In Section 3, first we give the definition of the notion that a system of the form (1) has ’reduced’ form. Next we prove that by a suitable change of variables $X=U(x)Y$ with $U(x)$ invertible and holomorphic at $x=0$ , system (1) is transformed to a system $¥{P_{t,k}¥}$ (2) $d¥mathrm{Y}=(¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i})¥mathrm{Y}$ which has ’reduced’ form. We determine $U(x)$ as a product of infinitely many simple matrices of special form. This section is the main part of our paper. Indeed, once the reduction to ’reduced’ form is completed, there remains no difficulty. In Section 4, we show that a substitution $¥mathrm{Y}=x_{1}^{L_{1}}¥cdots x_{n^{n}}^{L}¥cdot Z$ , being a nonnegative integer, changes system (2), which with $L_{i}=$ diag has ’reduced’ form, to a system $(l_{i}^{1_{ }},¥cdots, l_{i}^{m})$ $l_{i}^{a}$ (3) $dZ=(¥sum_{i=1}^{n}¥frac{B_{i}}{x_{i}}dx_{i})Z$ , . Combining the results in Sections 3 and 4, we with constant matrices , can establish one of our two main theorems, which is stated in the last section. In the arguments above, it was not the question if some $P_{i}(x)/x_{i}$ is holomorphic at $x=0$ or not. However, in the first alternative, we had better eliminate the variable , before making the change of variables studied in Sections 3 and 4. Thus Section 5 is devoted to this elimination problem; in more detail, it shows that if $P_{i}(x)/x_{i}$ is holomorphic at $x=0$ for $¥nu<i¥leq n$ , system (1) can be changed to the system $B_{1}$ $¥cdots$ $B_{n}$ $x_{i}$ (4) $d¥mathrm{Y}=(¥sum_{i=1}^{¥nu}¥frac{P_{i}(x_{1},,x_{¥nu},0,,0)}{x_{i}}¥cdots¥cdots dx_{i})¥mathrm{Y}$ On a Linear System of Pfaffian Equations 177 $U(x)Y$ with $U(x)$ invertible and holomorphic at $x=0$ . by a suitable substitution Combining the results of Sections 3, 4 and 5, we have the other main theorem, which is also stated in the last section. Convergence theorem. §1. In this section, as was announced in the introduction, we shall prove the following convergence theorem. It is an extension of a well known theorem for one variable (cf. [4]) to several variables. Theorem 1. Let (5) $du=(¥sum_{i=1}^{n}¥frac{F_{i}(x)}{x_{i}}dx_{i})u$ be a completely integrable system, where $F_{i}(x)=¥sum_{k¥geq 0}F_{i,k}x^{k}$ $u$ , is a matrix convergent and holomorphic for solution of (5) written as (6) is a vector and $i=1$ , $|x|<_{¥epsilon}$ $¥cdots$ , $n$ Then any formal power series . $u(x)=¥sum_{k¥geq 0}u_{k}x^{k}$ converges for $|x|<_{¥epsilon}$ and represents a holomorphic solution of (5). Proof. Since (6) formally satisfies (5), we have $F_{i,0}u_{0}=0$ and $(F_{i,0}-k_{i}I)u_{k}=-¥sum_{0<l¥leq k}F_{i,l}u_{k-l}$ , . There exists a positive integer for all $i=1$ , such that $¥det(F_{i,0}-k_{i}I)¥neq 0$ . $||(F_{i,0}-k_{i}I)^{-1}|| ¥ leq 1$ for all . Here and denotes the usual Euclidian norm. Hence $¥cdots$ $n$ $k_{i}^{0}$ $k_{i}>k_{i}^{0}$ (7) for $k¥not¥leq k^{0}$ $||$ $||$ $||u_{k}||¥leq¥sum_{0<l¥leq k}f_{l}||u_{k-l}||$ , where $f_{l}=¥max_{1¥leq i¥leq n}||F_{i,l}||$ and so is holomorphic for $v(x)$ defined by . $|x|<_{¥epsilon}$ We note that the series $f(x)=¥sum_{¥iota>0}f_{l}x^{¥iota}$ converges . Now we introduce the scalar majorizing function M. YOSHmA and K. TAKANO 178 (8) $v(x)=f(x)v(x)+||u_{0}||+¥sum_{0<k¥leq k^{0}}¥{||u_{¥mathrm{k}}||-¥sum_{0<¥iota¥leq k}f_{l}||u_{k-l}||¥}x^{k}$ Since for there exists a positive number Let $f(0)=0$ , $|x|<_{¥epsilon^{¥prime}}$ . (9) $v(x)=¥sum_{k¥underline{>}0}v_{k}x^{k}$ , $6^{¥prime<_{¥epsilon}}$ such that $|x|<_{¥epsilon^{¥prime}}$ $v(x)$ . is holomorphic . Substitution of (9) in (8) leads to the recursion formulas $v_{k}=||u_{k}||$ , $k¥leq k^{0}$ and $v_{k}=¥sum_{0<¥iota¥leq k}f_{l}||u_{k-l}||$ , $k¥not¥leq k^{0}$ for all Comparison with (7) shows series (9) implies the convergence of (6) in is now immediate. $||u_{k}||¥leq v_{k}$ $k$ . $|x|<_{¥epsilon^{¥prime}}$ §2. . Hence the convergence of the . The completion of the proof Integrability condition. The integrability condition for a system (10) $dX=¥Omega(x)X$ $d¥Omega(x)-¥Omega(x)¥wedge¥Omega(x)=0$ In case $¥Omega(x)=¥sum_{i=1}^{n}¥frac{P_{i}(x_{i})}{x_{i}}dx_{i}$ is . where $P_{i}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ is convergent and holomorphic in a neighborhood of equivalent to (10) , and $k=(k_{1^{ }},¥cdots, k_{n})$ . for all , $j=1$ , matrices: $[A, B]=AB-BA$ . §3. the condition (10) is $k_{j}P_{i,k}-k_{i}P_{j,k}+¥sum_{k^{¥prime}+k^{¥prime¥prime}=k}[P_{i,k^{¥prime}}, P_{j,k^{¥prime¥prime}}]=0$ $i,j;k$ $i$ $x=0$ , $¥cdots$ $n$ Here $[, ]$ denotes the usual bracket of Reduction to a ‘reduced’ form. The purpose of this section is to reduce system (1) to a system which has a form as simple as possible by a change of variables $X=U(x)Y$ with $U(x)$ invertible and holomorphic at $x=0$ . We shall begin with giving the following definition. Definition. We say the equation On a Linear System of Pfaffian 179 Equations $dX=(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{t})X$ is ‘reduced’ with respect to , , where for all $j=1,2$ , $(k, (¥alpha, ¥beta))$ $¥cdots$ if $a_{i}^{aa}-a_{i}^{¥beta¥beta}-k_{i}¥neq ¥mathit{0}$ $k=(k_{1^{ }},¥cdots, k_{n})$ $n$ , for some $P_{t}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ , $i$ implies $p_{jk}^{a¥beta}=0$ $P_{i,k}=(p_{ik}^{**})$ and Furthermore we say the equation has a‘reduced’ form if it is reduced . with respect to all First we shall show that by a suitable formal transformation $X=U(x)¥mathrm{Y}$ , $U(x)$ being a formal power series of , (1) can be changed to a system which $P_{i,0}=(a_{i}^{**})$ . $(k, (¥alpha, ¥beta))$ $x$ $=¥sum_{k¥geq 0}U_{k}x^{k}$ has ‘reduced’ form. Next we shall prove the convergence of the formal power . We remark that the formal reduction will be carried out by deseries $a$ $¥sum_{k¥geq 0}U_{k}x^{k}$ composing a transformation $X=(¥sum_{h¥underline{>}0}U_{k}x^{k})¥mathrm{Y}$ into a product of infinitely many transformations. 3.1. Before exposing the process to decompose $X=$ , we explain the reason why such a decomposition is necessary. Let Formal reduction. $(¥sum_{k¥geq 0}U_{k}x^{k})¥mathrm{Y}$ (2) be the system changed from (1) by $X=U(x)¥mathrm{Y}$ , then we have $Q_{i}(x)=U(x)^{-1}P_{t}(x)U(x)-U(x)^{-1}x_{i}¥frac{¥partial U(x)}{¥partial x_{i}}$ In the usual way, we substitute $U(x)=¥sum_{k¥geq 0}U_{k}x^{k}$ finitely many relations between the coefficients in $Q_{i}(x)=¥sum_{k¥geq 0}Q_{i,k}x^{k})$ $i=1$ , , ⃝, n. in the above equations and get in$¥{U_{k}¥}$ , and determine the values of , $¥{P_{i,k}¥}$ $¥{U_{k}¥}$ , $¥{Q_{i,k}¥}(¥mathrm{t}¥mathrm{h}¥mathrm{e}$ coefficients such that almost all of vanish. Considering these relations as recursion formulas to determine successively, we can easily verify that each must satisfy n relations at the same time. The integrability condition must assure that can be determined compatibly. However, the recursion formulas and the integrability condition are so comph.cated that it is very difficult to see to what extent system (1) can be simplified and that even if a canonical form is incidentally found out, it is hard to show that can be determined compatibly. Thus, in order to make clear the effect of the coefficients on the consequent equation and the role of integrability condition, we decompose the trans$¥{Q_{i,k}¥}$ $¥{U_{k}¥}$ $U_{k}$ $U_{k}$ $U_{k}$ $¥{U_{k}¥}$ formation $X=(¥sum_{k¥geq 0}U_{k}x^{k})¥mathrm{Y}$ into a product of simple transformations. Now we return to our subject. It is easy to see that every formal power series $U(x)=¥sum_{k¥geq 0}U_{k}x^{k}(¥det U_{0}¥neq 0)$ can be uniquely decomposed as follows 180 M. YOSHiDA and K. TAKANO $ U(x)=U_{0}¥cdot U_{1}(x)¥cdots U_{N}(x)¥cdots$ where $U_{0}$ is the constant matrix of the given $¥sum_{k¥geq 0}U_{k}x^{k}$ and $U_{N}(x)=U_{l¥backslash ¥Gamma}^{(1,m)}(x)¥cdot U_{N}^{(2,m)}(x)¥cdots U_{N}^{(m,m)}(x)$ $¥times U_{N}^{(1,m-1)}(x)¥cdot U_{N}^{(2,m-1)}(x)¥cdots U_{N}^{(m,m-1)}(x)$ $¥times U_{N}^{(1,1)}(x)¥cdot U_{N}^{(2,1)}(x)¥cdots U_{N}^{(m,1)}(x)$ $U_{N}^{(a,¥beta)}(x)=I+¥sum_{|k|=N}U_{N,k}^{(a,¥beta)}x^{k}$ Here $|k|=¥sum_{i=1}^{n}k_{i}$ zero except for , and $U_{N}^{(a,¥beta)}$ , . is a constant matrix of which the component is $(¥gamma, ¥delta)$ will be denoted by component of . (The of the special form as above Conversely the infinite product of can be expanded into a formal power series of . In other words, the formal $(¥gamma, ¥delta)=(¥alpha, ¥beta)$ $(¥alpha, ¥beta)$ $U_{N,k}^{(a,¥beta)}$ $¥{U_{N}^{(a,¥beta)}(x)¥}$ $u_{Nk}^{a¥beta}.)$ $x$ transformation transformations $¥{X=U_{N}^{(a,¥beta)}(x)¥mathrm{Y}¥}$ $X=(¥sum_{k¥geq 0}U_{h}x^{k})¥mathrm{Y}$ can be uniquely decomposed into a sequence of $¥{X=U_{N}^{(a,¥beta)}(x)¥mathrm{Y}¥}$ and conversely a sequence of transformations determines a formal transformation $X=(¥sum_{k¥geq 0}U_{k}x^{k})¥mathrm{Y}$ . If we decompose $X=U(x)Y$ in this way, then we can easily see the effect of each on the consequent equation and the role of the integrability condition. , , , , , , We shall show how to determine successively. $U_{N}^{(a,¥beta)}(x)$ $U_{0}$ Determination of in matrices $P_{i,0}$ $U_{0}$ . $U_{1}^{(1,m)}(x)$ $¥cdots$ $U_{1}^{(m,1)}(x)$ $¥cdots$ $U_{N}^{(a,¥beta)}(x)$ ? , the constant By the integrability condition (11) in (1) are commutin and $i,j:0$ $P_{i}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ $P_{j,0}$ $P_{j}(x)=¥sum_{k¥geq 0}P_{j,k}x^{k}$ , . Therefore, by the elementary matrix ative: $[P_{i,0}, P_{j,0}]=0$ for all , $j=1$ , such that matrix constant nonsingular theory, we can choose a $¥cdots$ $i$ $n$ $U_{0}$ $¥mathrm{e}¥cdots i,’ ¥mathrm{t}n¥mathrm{i}¥mathrm{s}1¥mathrm{o}¥mathrm{w}¥mathrm{e}¥mathrm{r}¥mathrm{t}¥mathrm{r}¥mathrm{i}¥mathrm{a}¥mathrm{n}¥mathrm{g}¥mathrm{u}1¥mathrm{a}¥mathrm{r}¥mathrm{h}¥mathrm{e}¥mathrm{n}a_{j}^{a¥beta}=0¥mathrm{f}¥mathrm{o}¥mathrm{r}¥mathrm{a}11j’=1$ (12) $¥{((¥mathrm{i})¥mathrm{i}¥mathrm{i})$ , $¥mathrm{i}A_{i}=(a_{i}^{¥alpha¥beta})=U_{0}^{-1}.P_{i,0}U_{0},i=¥mathrm{l}¥mathrm{f}a_{i}^{aa}-a_{i}^{¥beta¥beta}¥neq 0¥mathrm{f}¥mathrm{o}¥mathrm{r}¥mathrm{s}¥mathrm{o}’ ¥mathrm{m}$ We note that the equation transformed by . for all $(0, (¥gamma, ¥delta))$ $X=U_{¥mathit{0}}Y$ $¥cdots$ , , where $n$ $A_{i}$ ’reduced’ with respect to $(¥gamma, ¥delta)$ namely the , , , ,?. Determination of described process is determined successively by the induction coefiBcient , regarding as below. We apply the following proposition to determine the equation in $X$ in the proposition the equation transformed from (1) by $X=$ $U_{1}^{(1,m)}(x)$ $U_{1}^{(2,m)}(x)$ $¥cdots$ $U_{N}^{(a,¥beta)}(x)$ $U_{N}^{(a,¥beta)}(x)$ $u_{Nk}^{a¥beta}(|k|=N)$ $U_{N}^{(a,¥beta)}(x)$ On a Linear System $U_{0}¥cdot U_{1}^{(1,m)}(x)¥cdots U_{N}^{(a-1,¥beta)}(x)¥mathrm{Y}$ ing the notation scribed as follows. of Pfaffian 181 Equations which is obviously completely integrable. By introducor for , , the induction process is de- $(¥gamma, ¥delta)<$ $(¥alpha, ¥beta)$ $¥delta>¥beta$ $¥delta=¥beta$ Proposition 1 (Induction process). $¥gamma¥backslash ^{/}¥alpha$ Assume that the completely integrable system $dX=(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{i})X$ with $P_{i}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ , $P_{i,k}=(p_{ik}^{**})$ , is ’reduced’ with respect to both for the cases when $|k|<N$ , arbitrary $P_{i,0}=A_{i}=(a_{i}^{**})$ and when $|k|=N$ , . Assume further satisfies the condition (12). Then one finds a transformation $(k, (¥gamma, ¥delta))$ $(¥gamma, ¥delta)$ $(¥gamma, ¥delta)<$ $(¥alpha, ¥beta)$ $X=U_{N}^{(a,¥beta)}(x)¥mathrm{Y}$ such that the consequent equation $d¥mathrm{Y}=(¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i})¥mathrm{Y}$ with $Q_{t}(x)=¥sum_{k¥geq 0}Q_{i,k}x^{k}$ is ’reduced’ with respect to and when $|k|=N$ , Furthermore $(k, (¥gamma, ¥delta))$ $(¥gamma, ¥delta)¥leq(¥alpha, ¥beta)$ (13) (14) If $Q_{i,k}=P_{i,k}$ $q_{ih}^{¥gamma¥delta}=p_{ik}^{¥gamma¥delta}$ $a_{i_{0}}^{aa}-a_{i_{0}}^{¥beta¥beta}-k_{i_{0}}¥neq 0$ for , $Q_{i,k}=(q_{ik}^{**})$ both for the cases when $|k|<N$ , $(¥gamma, ¥delta)$ arbitrary . for $k$ $(|k|<N)$ , $i=1$ , $k(|k|=N)$ , $(¥gamma, ¥delta)<(¥alpha, ¥beta)$ for some , then the value of $i_{0}$ , , $¥cdots$ , , $i=1$ , $u_{Nk}^{¥alpha¥beta}(|k|=N)$ $(a_{i_{0}}^{¥alpha a}-a_{i_{0}}^{¥beta¥beta}-k_{i_{0}})u_{Nk}^{a¥beta}+p_{i_{0}k}^{¥alpha¥beta}=0$ $n$ $¥cdots$ , . $n$ is determined by . Proof. It is easy to see $Q_{i}(x)=(I+¥sum_{|k|=N}U_{N,k}^{(a,¥beta)}x^{k})^{-1}P_{i}(x)(I+¥sum_{|k|=N}U_{N,k}^{(a,¥beta)}x^{k})$ $-(I+¥sum_{|k|=N}U_{N,k}^{(a,¥beta)}x^{k})^{-1}(¥sum_{|k|=N}k_{i}U_{N,k}^{(a,¥beta)}x^{k})$ . 182 M. YOSHIDA and K. TAKANO Comparing the coefficients of (15) with and $x^{k}$ , $|k|<N$ or $|k|=N$ , we obtain (13) and $Q_{i,k}=P_{i,k}+[A_{t}, U_{N,k}^{(a,¥beta)}]-k_{i}U_{N,k}^{(a,¥beta)}$ $|k|=N$ , $i=1$ , $¥cdots$ , . Since $n$ $A_{i}$ is lower triangular, we can easily derive (14) $q_{ik}^{a¥beta}=p_{ik}^{a¥beta}+(a_{i}^{aa}-a_{i}^{¥beta¥beta}-k_{i})u_{Nk}^{a¥beta}$ from (15). Now suppose $k(|k|=N)$ (16) , $|k|=N$ , $i=1$ , $¥cdots$ , , $n$ satisfies $a_{i_{0}}^{aa}-a_{i_{0}}^{¥beta¥beta}-k_{i_{0}}¥neq 0$ for some . In order to prove that can be chosen so that sufficient to show the relation $u_{Nk}^{a¥beta}$ (17) , $i_{0}$ $q_{ik}=0$ for all , $i=1$ , $(a_{i_{¥mathit{0}}}^{aa}-a_{i_{¥mathit{0}}}^{¥beta¥beta}-k_{i_{¥mathit{0}}})p_{ik}^{a¥beta}-(a_{i}^{aa}-a_{i}^{¥beta¥beta}-k_{i})p_{i_{¥mathit{0}}k}^{a¥beta}=¥mathit{0}$ $i=1$ , $¥cdots$ , $¥cdots$ , , it is $n$ $n$ holds, which will be called the compatibility condition. It will be shown that the condition (17) can be derived from the integrabih.ty condition, which will complete the proof of Proposition 1. Let us calculate the component of the integrability condition $(¥alpha, ¥beta)$ (11) $k_{i}P_{i_{0},k}-k_{i}{}_{0}P_{i,k}+¥sum_{k^{¥prime}+k^{¥prime¥prime}=k}[P_{t_{0},k^{¥prime}}, P_{t,k^{¥prime¥prime}}]=0$ $i_{0},t;k$ For the third term of (11) (18) $i_{0},i;k$ . , we have $¥sum_{k^{¥prime}+k^{¥prime¥prime}=k}[P_{i_{0},k^{¥prime}}, P_{i,k^{¥prime¥prime}}]=[A_{i_{0}}, P_{i,k}]-[A_{i}, P_{i_{0},k}]+h^{¥prime}k^{¥prime},+k^{¥prime¥prime}-kk^{¥prime¥prime}>0¥sum_{-}[P_{i_{0},k^{¥prime}}, P_{i,k^{¥prime¥prime}}]$ The $(¥alpha, ¥beta)$ . component of the first term of (18) is expressed as $[A_{i_{0}},, P_{i,k}]^{a¥beta}=(a_{i_{0}}^{aa}-a_{i_{0}}^{¥beta¥beta})p_{ik}^{a¥beta}+¥sum_{¥xi<a}a_{i_{0}}^{a¥xi}p_{ih}^{¥xi¥beta}-¥sum_{¥xi>¥beta}a_{i_{0}}^{¥xi¥beta}p_{ik}^{a¥xi}$ , is lower triangular. In this relation, if because for some , then for all and we have for all /, by the assumption of the proposition that the system considered is ’reduced’ with respect to both for the cases when $k=0$ , . Hence arbitrary and when $|k|=N$ , for all /, which is a contradiction to (16). Therefore, for all . For the same reason, for all . Thus we have $A_{i_{0}}$ $a_{j}^{aa}-a_{j}^{¥xi¥xi}=0$ $a_{i_{0}}^{a¥xi}p_{ik}^{¥xi¥beta}¥neq 0$ $j$ $¥xi<_{¥alpha}$ $a_{j}^{¥xi¥xi}-a_{j}^{¥beta¥beta}-k_{J}=0$ $(k, (¥gamma, ¥delta))$ $(¥gamma, ¥delta)<(¥alpha, ¥beta)$ $(¥gamma, ¥delta)$ $a_{j}^{aa}-a_{j}^{¥beta¥beta}-k_{f}=¥mathit{0}$ $¥xi<_{¥alpha}$ (19) $a_{i_{0}}^{a¥xi}p_{ik}^{¥xi¥beta}=0$ $a_{i_{0}}^{¥xi¥beta}p_{ik}^{a¥xi}=0$ $¥xi>¥beta$ $[A_{i_{0}}, P_{i,k}]^{a¥beta}=(a_{i_{0}}^{aa}-a_{i_{0}}^{¥beta¥beta})p_{ik}^{a¥beta}$ , $[A_{i}, P_{i_{0},k}]^{a¥beta}=(a_{i}^{aa}-a_{i}^{¥beta¥beta})p_{i_{0}k}^{a¥beta}$ . and analogously (20) On a Linear System of Finally, it will be shown that the Pfaffian 183 Equations component of the third term of (18), $(¥alpha, ¥beta)$ i.e. $[P_{i_{0},k^{¥prime}}, P_{i,k^{¥prime¥prime}}]^{a¥beta}=¥sum_{¥xi=1}^{m}p_{i_{0}k^{¥prime}}^{a¥xi}p_{ik}^{¥xi¥beta},,-¥sum_{¥xi=1}^{m}p_{i_{0}k^{¥prime}}^{¥xi¥beta}p_{ik^{¥prime¥prime}}^{a¥xi}$ vanishes. If $p_{i_{0}h^{¥prime}}^{a¥xi}p_{ik^{¥prime¥prime}}^{¥xi¥beta}¥neq 0$ , $k^{¥prime}+k^{¥prime¥prime}=k$ , $k^{¥prime}$ , $k^{¥prime¥prime}>0$ for some , then we have $¥xi$ $a_{j}^{aa}-a_{j}^{¥xi¥xi}-k_{j}^{¥prime}=¥mathit{0}$ for all $j$ and $a_{j}^{¥xi¥xi}-a_{j}^{¥beta¥beta}-k_{j}^{¥prime¥prime}=0$ for all , $j$ by the assumption of the proposition that our system is ’reduced’ with respect to for $|k|<N$ , arbitrary. Hence for all /, which is a contradiction to (16). Therefore for all . In the same way, we have for all . Thus $(k, (¥gamma, ¥delta))$ $a_{j}^{aa}-a_{j}^{¥beta¥beta}-k_{j}=¥mathit{0}$ $(¥gamma, ¥delta)$ $¥xi$ $p_{i_{0}k^{¥prime}}^{a¥xi}p_{ik^{¥prime¥prime}}^{¥xi¥beta}=0$ $¥xi$ $p_{i_{0}k^{¥prime}}^{¥xi¥beta}p_{ik^{¥prime¥prime}}^{a¥xi}=0$ (21) $[P_{i_{0},k^{¥prime}}, P_{i,k^{¥prime¥prime}}]^{a¥beta}=0$ . Therefore, from (11) $i_{0},i:k’(18)$ , (19), (20) , , for all , with and (21), we obtain the compatibility condition (17). Thus we have completed the proof of Proposition 1. thus determined, we have By expanding the infinite product of $k^{¥prime}$ $k^{¥prime}+k^{¥prime¥prime}=k$ $k^{¥prime¥prime}$ $k^{¥prime}$ $k^{¥prime¥prime}>0$ $¥{U_{N}^{(a,¥beta)}(x)¥}$ Theorem 2. One can find a formal power series that the formal substitution has $a$ $X=(¥sum_{k¥geq 0}U_{k}x^{k})¥mathrm{Y}$ with $¥sum_{k¥geq 0}U_{k}x^{k}$ $¥det U_{0}¥neq 0$ , such changes system (1) to a system which ’reduced’ form. Remark 1. In the induction process above, for sufficiently large $N$ , in Theorem 2 is not uniquely is uniquely determined. Therefore, although $U_{N}^{(a,¥beta)}(x)$ $¥sum_{k¥geq 0}U_{k}x^{k}$ determined, it contains only a finite number of undetermined parameters. Remark 2. In Theorem 2, if no two eigenvalues of for every , we can choose $i$ $¥sum_{k¥geq 0}U_{k}x^{k}$ with $U_{0}=I$ $P_{i,0}$ such that, by differ by an integer $X=(¥sum_{h¥geq 0}U_{k}x^{k})¥mathrm{Y}$ , (1) is changed to the equation $d¥mathrm{Y}=(¥sum_{i=1}^{n}¥frac{P_{i,0}}{x_{i}}dx_{i})¥mathrm{Y}$ In this case, 3.2, $¥sum_{k¥geq 0}U_{k}x^{k}$ $U_{k}(k>0)$ . is uniquely determined. Analytic reduction. As is easily seen, the formal power series in Theorem 2 satisfies the equation $U(x)=$ 184 M. YOSHIDA and K. TAKANO (22) $dU(x)=(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{i})U(x)-U(x)(¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i})$ in the formal sense. (23) If we regard du(x) $U(x)$ as an $¥mathrm{m}^{2}$ -vector , , (22) is equivalent to $u(x)$ $=(¥sum_{i=1}^{n}¥frac{1}{x_{i}}(P_{i}(x)¥otimes I-I¥otimes^{¥iota}Q_{i}(x))dx_{i})u(x)$ . Since $Q_{i}(x)$ is a polynomial of , we can apply Theorem 1 to equation (23) to prove . Thus, by the definition of a‘reduced’ form, we have the convergence of $x$ $¥sum_{k¥geq 0}U_{k}x^{k}$ Given a completely integrable system (1), we have a convergent with $¥det U_{0}¥neq 0$ , such that the transformation $X=U(x)Y$ takes Theorem 3. series ¥ ¥ $U(x)= sum_{h geq 0}U_{k}x^{k}$ (1) to (24) $d¥mathrm{Y}=(¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i})¥mathrm{Y}$ , where (i) $Q_{i}(x)=A_{i}+¥sum_{k>0}Q_{i,k}x^{k}$ $A_{i}=(a_{i}^{**})$ (ii) the (finite sum), being lower triangular, $(¥alpha, ¥beta)$ component $q_{i}^{a¥beta}(x)$ of $Q_{i}(x)$ is a monomial of , $x$ $q_{i}^{a¥beta}(x)=q_{i}^{a¥beta}x_{1}^{h_{1}}¥cdots x_{n^{n}}^{k}$ with $k_{¥mu}=a_{¥mu}^{aa}-a_{¥mu}^{¥beta¥beta}$ integer §4. $k_{¥mu}$ for all , $¥mu=1$ $¥mu=1$ , , $¥cdots$ $¥cdots$ , . $n$ $q_{i}^{a¥beta}(x)¥neq 0$ , implies that $a_{¥mu}^{aa}-a_{¥mu}^{¥beta¥beta}$ $n$ Singular transformation. Consider the ’reduced’ equation (24) in Theorem 3. . A singular transformation matrix diag $(l_{i^{ }}^{1},--, l_{i}^{m})$ $¥mathrm{Y}=x_{1}^{L_{1}}-x_{n}^{L_{n}}Z$ changes (24) to $dZ=(¥sum_{i=1}^{n}¥frac{B_{i}(x)}{x_{i}}dx_{i})Z$ where is a nonnegative , . Let $L_{i}$ be a diagonal On a Linear System of Pfaffian $B_{i}(x)=x_{l}^{-L_{l}}¥cdots x_{n}^{-L_{n}}Q_{i}(x)x_{l}^{L_{l}}$ 185 Equations ? $x_{n^{n}}^{L}-L_{i}$ , or equivalently . $b_{i}^{a¥beta}(x)=q_{i}^{a¥beta}(x)x_{1}^{¥iota_{1}^{¥beta}-¥iota_{1}^{a}}¥cdots x_{n^{n}}^{l^{¥beta}-¥iota_{n}^{a}}-¥delta_{¥rho}^{a}l_{i}^{a}$ ¥ ¥ , Here, , and denotes the Kronecker symbol. We shall show that becomes constant by choosing nonnegative integers suitably. We classify and belong to the same class , so that iff is an integer. We denote by the class of . For every , we denote by a member of which has the minimum real part. Then by tak$ing$ , becomes a constant , by virtue of the properties (i) and (ii) in Theorem 3. Thus we have proved $Q_{i}(x)=(q_{i}^{a beta}(x))$ $B_{t}(x)=(b_{i}^{a beta}(x))$ $¥delta_{¥beta}^{a}$ $b_{i}^{a¥beta}(x)$ $¥{a_{i}^{aa}¥}_{a=1,m}¥cdots$ $¥{l_{i}^{a}¥}$ $a_{i}^{aa}-a_{i}^{¥beta¥beta}$ $a_{i}^{¥beta¥beta}$ $a_{i}^{aa}$ $[a_{i}^{aa}]$ $a_{i}^{aa}$ $a_{i}^{aa}$ $[a_{i}^{aa}]$ $a_{i^{00}}^{a¥alpha}$ $b_{i}^{a¥beta}(x)$ $l_{i}^{a}=a_{i}^{aa}-a_{i^{¥mathit{0}¥mathit{0}}}^{aa}$ $b_{i}^{a¥beta}$ By a change Theorem 4. of variables $¥mathrm{Y}=x_{1}^{L_{1}}¥cdots x_{n^{n}}^{L}Z$ with , being nonnegative integer, the ’reduced: equation (24) in Theorem 3 is transformed to $¥mathrm{L}_{¥mathrm{i}}=¥mathrm{d}¥mathrm{i}¥mathrm{a}¥mathrm{g}$ $(l_{i^{ }}^{1},¥cdots, l_{i}^{m})$ $l_{i}^{a}$ $dZ=(¥sum_{i=1}^{n}¥frac{B_{i}}{x_{i}}dx_{i})Z$ where $B_{i}$ is a constant matrix given by (finite sum) $B_{i}=A_{i}-L_{i}+¥sum_{k>0}Q_{i,h}$ and $[B_{t}, B_{j}]=0$ , , $j=1$ , $i$ $¥cdots$ , . $n$ in Theorem 4 is not uniquely determined but is unique up to Remark 3. integers in the following sense: Let and be two diagonal matrices satisfying the conditions of Theorem 4, then for all , with . $L_{¥ell}$ $L_{i}$ $L_{i}^{¥prime}$ $l_{i}^{a}-l_{i}^{¥beta}=l_{i}^{¥prime_{a}}-l_{i}^{¥prime¥beta}$ §5. $¥alpha$ $¥beta$ $[a_{i}^{aa}]=[a_{i}^{¥beta¥beta}]$ Elimination of nonessential variables. In this section, we consider the special case when $P_{i}(x)/x_{i}$ is holomorphic at $x=0$ for $¥nu<i¥leq n$ . In order to eliminate the variables , , , we consider a sequence of transformations $¥{X=U_{N}(x)¥mathrm{Y}¥}_{N=1,2},¥ldots$ where $x_{¥nu+1}$ $U_{N}(x)=I+¥sum_{|h|=N,¥hat{k}>0}U_{N,k}x^{k}$ with $¥hat{k}=(k_{¥nu+1^{ }},¥cdots, k_{n})$ . $x_{n}$ , As an induction process we have Proposition 2 (Induction process). system $¥cdots$ Assume that the completely integrable 186 M. YOSHIDA and K. TAKANO $dX=(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{i})X$ , $P_{i}(x)=¥sum_{k¥geq 0}P_{i,k}x^{k}$ , satisfies the conditions: is holomorphic at $x=0$ for $¥nu<i¥leq n$ , (i) , $0¥leq|k|<N$ . , $0<|k|<N,¥hat{k}>0$ and for (u) $P_{i,k}=0$ for $X=U_{N}(x)Y$ with Then we can choose a suitable substitution $P_{i}(x)/x_{i}$ $i>_{¥nu}$ $ i¥leq¥nu$ $U_{N}(x)=I+¥sum_{|k|=N,¥hat{k}>0}U_{N,k}x^{h}$ , such that the consequent system $d¥mathrm{Y}=(¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i})¥mathrm{Y}$ , $Q_{i}(x)=¥sum_{k¥geq 0}Q_{i,k}x^{k}$ satisfies the conditions: (i)’ $(¥mathrm{i}¥mathrm{i})$ ’ (iii)’ $Q_{i}(x)/x_{i}$ $Q_{i,h}=0$ is holomorphic at $x=0$ for $¥nu<i¥leq n$ , , $0<|k|¥leq N,¥hat{k}>0$ and for for $i>_{¥nu}$ $ i¥leq¥nu$ , $Q_{i}(x_{1^{ }},¥cdots, x_{¥nu}, 0, ¥cdots, 0)=P_{i}(x_{1^{ }},¥cdots, x_{¥nu}, 0, ¥cdots, 0)$ $0¥leq|k|¥leq N$ for $ i¥leq¥nu$ and the condition (ii)’ for , Proof, The conditions , $0¥leq k<N$ are immediately verified by and for $(¥mathrm{i})’$ $¥hat{k}>0$ $(¥mathrm{i}¥mathrm{i}¥mathrm{i})^{¥prime}$ , . $ i¥geq¥nu$ , $0<|k|<N$ , $i>_{¥nu}$ (25) $¥sum_{i=1}^{n}¥frac{Q_{i}(x)}{x_{i}}dx_{i}=U_{N}(x)^{-1}(¥sum_{i=1}^{n}¥frac{P_{i}(x)}{x_{i}}dx_{i})U_{N}(x)-U_{N}(x)^{-1}dU_{N}(x)$ Therefore, we have only to show that by choosing the values of , $|k|=N$ . By (25), we have , $|k|=N,¥hat{k}>0$ and for for $Q_{i,k}=P_{i,k}+[P_{i,0}, U_{N,k}]-k_{i}U_{N,k}$ $k$ , $(¥mathrm{i}¥mathrm{i})^{¥prime}$ holds $i>_{¥nu}$ $ i¥leq¥nu$ For all by $¥{U_{N,k}¥}$ . $(|k|=N,¥hat{k}>0)$ let $i(k)$ , $i=1$ , be an index satisfying $U_{N,k}=-¥frac{1}{k_{i(k)}}P_{i(k),k}$ For thus determined, we can prove by the integrability condition (11). $U_{N,k}$ $¥cdots$ $Q_{i,k}=0$ , , $n$ $|k|=N,¥hat{k}>0$ $k_{i(k)}¥neq 0$ . , and determine $U_{N,k}$ . for $i=1$ , $¥cdots$ , , $n$ $|k|=N,¥hat{k}>0$ , Q.E.D. By this proposition, one determines a sequence of transformations which formally reduces (1) to $ U_{N}(x)¥mathrm{Y}¥}_{N=1,2},¥ldots$ . $d¥mathrm{Y}=(¥sum_{i=1}^{¥nu}¥frac{P_{i}(x_{1},,x_{¥nu},0,,0)}{x_{i}}¥cdots¥cdots dx_{i})¥mathrm{Y}$ $¥{X=$ On a Linear System of Pfaffian 187 Equations The convergence of the formal power series $U(x)=U_{1}(x)U_{2}(x)¥cdots U_{N}(x)¥cdots=I+¥sum_{k>0_{¥hat{k}}>0},U_{k}x^{k}$ can be proved by Theorem 1 in an analogous way as in Section 3. Thus we have Theorem 5. Let a completely integrable system (1) be given. If $P_{i}(x)/x_{i}$ is holomorphic at $x=0$ for $¥nu<i¥leq n$ , then (1) can be changed to the system $d¥mathrm{Y}=(¥sum_{i=1}^{¥nu}¥frac{P_{i}(x_{1},,x_{¥nu},0,,0)}{x_{i}}¥cdots¥cdots dx_{i})¥mathrm{Y}$ by a substitution $X=U(x)Y$ where the form $U(x)$ is invertible and holomorphic at $U(x)=I+k_{¥nu+1}+.¥cdot¥sum_{k>0}.+k_{n}>0U_{k}x^{k}$ $x=0$ of . Remark 4. It must be noted that, in the above substitution $U(x)$ , we have for $k_{v+1}=¥cdots=k_{n}=0$ . This fact plays an important role in applications. $U_{k}=0$ §6. Main theorems. Combining Theorems 3 and 4, we have Theorem . Let a completely integrable system (1) be given, where , $=1$ , , , is holomorphic at $x=0$ . Then we have a matrix $U(x)$ invertible and holomorphic at $x=0$ and a diagonal matrix , $i=1$ , , , whose components are nonnegative integers, such that the transformation $P_{i}(x)$ $¥epsilon$ $¥cdots$ $i$ $n$ $L_{i}$ $¥cdots$ $n$ $X=U(x)x_{l}^{L_{l}}¥cdots x_{n^{n}}^{L}Z$ changes (I) to $dZ=(¥sum_{i=1}^{n}¥frac{B_{i}}{x_{i}}dx_{i})Z$ , , is a constant matrix satisfying $[B_{i},B_{j}]=0$ for all $i,i=1$ , $¥cdots,n$ . where , $i=1$ , and the coefficients in the power series for , The calculation of the matrices $U(x)$ can be concretely performed by algebraic operations. $¥cdots$ $B_{i}$ $n$ $B_{i}$ $L_{i}$ Combining Theorems 3, 4 and 5, we have Theorem 7. Let a completely integrable system (1) be given, where , $=1$ ,?, , is holomorphic at $x=0$ . $P_{i}(x)/x_{i}$ , $i=¥nu+1$ , , , is holoIf further morphic at $x=0$ , then we have two matrices $V(x)$ , invertible and $P_{i}(x)$ $¥cdots$ $n$ $W(x_{1^{ }},¥cdots, x_{¥nu})$ $n$ $i$ 188 M. YOSHIDA and K. TAKANO holomorphic at $x=0$ having the form $V(x)=I+¥sum_{h>0,k_{p+1}++h_{n}>0}¥cdots V_{h}x^{k}$ , $W(x_{1^{ }},¥cdots, x_{¥nu})=W_{0}+¥ldots¥sum_{k_{1}++k_{y}>0}W_{k_{1},k_{p}}¥cdots,x_{1^{1}}^{h}¥cdots x_{¥nu}^{k_{y}}$ and a diagonal matrix , $i=1$ , such that the transformation $L_{i}$ $¥cdots$ , , whose components are nonnegative integers, $¥nu$ $X=U(x)W(x_{1^{ }},¥cdots, x_{¥nu})x_{l}^{L_{l}}¥cdots x_{¥nu}^{L_{¥nu}}Z$ changes (1) to $dZ=(¥sum_{i=1}^{¥nu}¥frac{B_{i}}{x_{i}}dx_{i})Z$ . Here, , $i=1$ , , , is a constant matrix satisfying $[B_{i}, B_{j}]=0$ for all , $i=1$ , , . The calculation of , and the coefficients , can be concretely carried out by algebraic operations. $¥cdots$ $B_{i}$ $i$ $¥nu$ $B_{i}$ $¥{V_{h}¥}$ $L_{i}$ $¥cdots$ $¥nu$ $¥{W_{k_{1},k_{¥nu}}¥cdots,¥}$ Remark 5. As an immediate consequence of Theorem 1, we have the following theorem which is equivalent to a result obtained by R. Gerard in an abstract way ([3]). Theorem. A completely integrable system $dX=(¥sum_{i=1}^{¥nu}¥frac{P_{i}(x)}{x_{i}}dx_{i}+¥sum_{i=¥nu+1}^{n}¥hat{P}_{i}(x)dx_{i})X$ with , $i=1$ , , , $i=¥nu+1$ , mental matrix solution of the form $P_{t}(x)$ $¥cdots$ $¥nu,¥hat{P}_{i}(x)$ $¥cdots$ , , , holomorphic at $n$ $U(x)x_{l}^{L_{l}}¥cdots x_{¥nu}^{L_{y}}x_{l}^{B_{l}}$ ? $x_{¥nu}^{B_{p}}$ $x=0$ , has a funda- . Here $U(x)$ is a matrix invertible and holomorphic at $x=0$ , is a diagonal matrix whose components are nonnegative integers and is a constant matrix with $=0$ for all , $j=1$ , , . $L_{i}$ $B_{i}$ $i$ $¥cdots$ $[B_{i}, B_{j}]$ $¥nu$ References a Deligne, P., Equations differentielles points reguliers singuliers, Lee. Notes in Math., 163. Springer (1970). [2] Gantmacher, F. R., The Theory of Matrices, Vol. I. Chelsea Publishing Company, New York (1964). [3] Gerard, R., Theorie de Fuchs sur une variete analytique complexe, J. Math. Pures Appl., 47 (1968), 321-404. [1] On a Linear System [4] [5] of Pfaffian Equations 189 Wasow, W., Asymptotic Expansions for Ordinary Differential Equations, Interscience Publishers (1965). Yoshida, M. and Takano, K., Local theory of Fuchsian systems I, Proc. Japan Acad., 51 (1975), 219-223. nuna adreso: Department of Mathematics Kobe University Rokko, Kobe Japan (Ricevita la 13-an de junio, 1975)