N. I. Shkil
MATHEMATICS
Submitted 1963-01-01 | RussiaRxiv: ru-196301.57094 | Translated from Russian

Abstract

Full Text

N. I. Shkil

ON THE ASYMPTOTIC SOLUTION OF A SYSTEM OF LINEAR DIFFERENTIAL EQUATIONS CONTAINING A PARAMETER

(Presented by Academician N. N. Bogolyubov on 10 I 1963)

MATHEMATICS

  1. Numerous works have been devoted to the construction of the asymptotic solution of systems of differential equations containing a parameter \(\left({}^{1-7}\right)\). Fundamental results were obtained by N. N. Bogolyubov \(\left({}^{1}\right)\).

Consider, in \(n\)-dimensional space, the system of differential equations\(^*\)

\[ \frac{dx}{dt}=A(\tau,\varepsilon)x+\varepsilon B(\tau,\varepsilon)e^{i\theta}, \tag{1} \]

where \(A(\tau,\varepsilon)\) is a real matrix of order \(n\); \(B(\tau,\varepsilon)\) is an \(n\)-dimensional vector, admitting formal expansions

\[ A(\tau,\varepsilon)=\sum_{s=0}^{\infty}\varepsilon^{s}A^{(s)}(\tau),\qquad B(\tau,\varepsilon)=\sum_{s=0}^{\infty}\varepsilon^{s}B^{(s)}(\tau),\qquad 0\leqslant \tau=\varepsilon t\leqslant L; \tag{2} \]

\(\varepsilon\) is a small real parameter.

The asymptotic solution of system (1) is determined by the behavior of the roots of the equation

\[ \det \left\| A^{(0)}(\tau)-\lambda E \right\|=0, \tag{3} \]

where \(E\) is the identity matrix.

In \(\left({}^{2}\right)\) a solution was obtained for the case of simple roots on the segment \([0,L]\) of equation (3). The case of multiple roots is partially considered in \(\left({}^{3,4,7}\right)\). The asymptotic splitting of system (1) is given in \(\left({}^{5,6}\right)\) for arbitrary behavior of the roots of equation (3).

In the present article we consider the case when several multiple elementary divisors correspond to a multiple root.

  1. Denote by \(\lambda_1(\tau), \lambda_2(\tau), \ldots, \lambda_p(\tau)\) the roots of equation (3), having respectively the constant multiplicities \(k_1,k_2,\ldots,k_p\) \((k_1+k_2+\cdots+k_p=n)\), and suppose that to the root \(\lambda_j(\tau)\) \((j=1,\ldots,p)\) there correspond \(r_j\) elementary divisors
    \[ (\lambda-\lambda_j(\tau))^{s_{j1}},\ldots,(\lambda-\lambda_j(\tau))^{s_{jr_j}} \]
    \((s_{j1}+\cdots+s_{jr_j}=k_j;\; j=1,\ldots,p)\). Then one can indicate a nonsingular matrix \(T(\tau)\) such that

\[ T^{-1}(\tau)A^{(0)}(\tau)T(\tau)=W(\tau)= \left\| \begin{array}{ccc} W_{k_1}(\tau) & & 0\\ & \ddots & \\ 0 & & W_{k_p}(\tau) \end{array} \right\|, \tag{4} \]

\(^*\) Many differential equations are reducible to a system of the form (1), in particular differential equations with a small parameter at the highest derivatives.

where

\[ W_{k_j}(\tau)= \left\| \begin{array}{cccc} W_{s_{j1}}(\tau) & & & \\ 0 & \ddots & & 0\\ & & W_{s_{jr_j}}(\tau) \end{array} \right\|, \quad W_{s_{jn_j}}(\tau)= \left\| \begin{array}{ccccc} \lambda_j(\tau) & 1 & 0 & \ldots & 0\\ 0 & \lambda_j(\tau) & 1 & 0\ldots 0\\ \cdot & \cdot & \cdot & \cdot & \cdot\\ 0 & 0 & \ldots & \lambda_j(\tau) \end{array} \right\| \Biggr\}_{s_{jn_j}}, \tag{5} \]

\[ n_j=1,\ldots,r_j;\quad j=1,\ldots,p; \]

\(T^{-1}(\tau)\) is the matrix inverse to the matrix \(T(\tau)\).

Suppose that the function \(i\omega(\tau)\) \((\omega(\tau)=d\theta/dt,\ i=\sqrt{-1})\) at isolated points of the segment \([0,L]\) becomes equal to one of the roots of equation (3), for example \(\lambda_1(\tau)\), but \(i\omega(\tau)\ne \lambda_k(\tau)\) \((k=2,\ldots,p)\) for any \(\tau\in[0,L]\). This case is customarily called “resonant.”

Theorem 1. If \(A(\tau,\varepsilon)\), \(B(\tau,\varepsilon)\), \(\omega(\tau)\) have derivatives of all orders with respect to \(\tau\), and for all \(\tau\in[0,L]\) the \(\nu_{jn_j}\)-component of the vector

\[ C(\tau)=T^{-1}(\tau)\left[ A^{(1)}(\tau)T_{\nu_{jn_j-1}+1}(\tau) -\frac{dT_{\nu_{jn_j-1}+1}(\tau)}{d\tau} \right], \tag{6} \]

where \(T_{\nu_{jn_j-1}+1}(\tau)\) is the column of the matrix \(T(\tau)\) numbered \(\nu_{jn_j-1}+1\) \((\nu_{jn_j}=k_1+\ldots+k_{j-1}+s_{j1}+\ldots+s_{jn_j};\ n_j=1,\ldots,r_j;\ j=1,\ldots,p)\), is not equal to zero, then the asymptotic solution of system (1) in the “resonant” case can be represented in the form

\[ x=\sum_{n_1=1}^{r_1}\left[U_{1n_1}(\tau,\mu_{1n_1})h_{1n_1} +P_{n_1}(\tau,\mu_{1n_1})\right]e^{i\theta} + \]

\[ +\sum_{k=2}^{p}\sum_{n_k=1}^{r_k} U_{kn_k}(\tau,\mu_{kn_k})h_{kn_k}; \tag{7} \]

\[ \frac{dh_{1n_1}}{dt} = [\lambda_{1n_1}(\tau,\mu_{1n_1})-i\omega(\tau)]h_{1n_1} +z_{n_1}(\tau,\mu_{1n_1}), \quad n_1=1,\ldots,r_1; \tag{8} \]

\[ \frac{dh_{kn_k}}{dt} = \lambda_{kn_k}(\tau,\mu_{kn_k})h_{kn_k}, \quad n_k=1,\ldots,r_k;\quad k=2,\ldots,p, \tag{9} \]

where \(U_{jn_j}(\tau,\mu_{jn_j})\), \(P_{n_1}(\tau,\mu_{1n_1})\) are \(n\)-dimensional vectors; \(h_{jn_j}\), \(z_{n_1}(\tau,\mu_{1n_1})\) are scalar functions \((n_j=1,\ldots,r_j;\ j=1,\ldots,p)\), admitting formal expansions

\[ U_{jn_j}(\tau,\mu_{jn_j}) = \sum_{s=0}^{\infty}\mu_{jn_j}^{s}U_{jn_j}^{(s)}(\tau), \quad \lambda_{jn_j}(\tau,\mu_{jn_j}) = \lambda_j(\tau)+ \sum_{s=1}^{\infty}\mu_{jn_j}^{s}\lambda_{jn_j}^{(s)}(\tau); \tag{10} \]

\[ P_{n_1}(\tau,\mu_{1n_1}) = \sum_{s=0}^{\infty}\mu_{1n_1}^{s}P_{n_1}^{(s)}(\tau), \quad z_{n_1}(\tau,\mu_{1n_1}) = \sum_{s=0}^{\infty}\mu_{1n_1}^{s}z_{n_1}^{(s)}(\tau); \tag{11} \]

\[ \mu_{jn_j}=\sqrt[s_{jn_j}]{\varepsilon}, \quad n_j=1,\ldots,r_j;\quad j=1,\ldots,p. \tag{12} \]

Substituting the vector \(x\), determined by relations (7)—(9), into system (1) and, in the identity obtained, equating the coefficients of the functions \(h_{jn_j}\) and the free terms, we have

\[ (A-\lambda_{jn_j}E)U_{jn_j} = \varepsilon U'_{jn_j}, \quad n_j=1,\ldots,r_j;\quad j=1,\ldots,p^*, \tag{13} \]

\[ (A-i\omega E)P_{n_1} = U_{1n_1}z_{n_1} +\varepsilon(P'_{n_1}-B), \quad n_1=1,\ldots,r_1. \tag{14} \]

\[ \text{* Here and below we omit the arguments of the quantities.} \]

To determine the coefficients of the series (10), we shall use relation (13). Equating in it the coefficient of \(\mu_{j n_j}^{0}\), we have

\[ (A^{(0)}-\lambda_j E)U_{i n_j}^{(0)}=0,\qquad n_j=1,\ldots,r_j;\ j=1,\ldots,p. \tag{15} \]

Introducing the vector

\[ Q_{j n_j}^{(s)}=T^{-1}U_{j n_j}^{(s)},\qquad s=0,1,\ldots, \tag{16} \]

equation (15) may be put in the form

\[ (W-\lambda_j E)Q_{j n_j}^{(0)}=0, \tag{17} \]

or, according to (4),

\[ (W_{k_r}-\lambda_j E)Q_{j n_j k_r}^{(0)}=0,\qquad n_j=1,\ldots,r_j;\ r,j=1,\ldots,p, \tag{18} \]

where \(Q_{j n_j k_r}^{(0)}\) is the vector with components

\[ Q_{j n_j k_r}^{(0)}= \bigl(\{q_{j n_j}^{(0)}\}_{l_{r-1}+1},\ \{q_{j n_j}^{(0)}\}_{l_{r-1}+2},\ldots,\{q_{j n_j}^{(0)}\}_{l_r}\bigr), \tag{19} \]

\[ l_r=k_1+\cdots+k_r;\quad r=1,\ldots,p. \]

From (18) we find

\[ Q_{j n_j k_r}^{(0)}\equiv 0,\qquad j\ne r;\ j,r=1,\ldots,p;\ n_j=1,\ldots,r_j. \tag{20} \]

For \(r=j\), equation (18) is representable in the form

\[ (W_{s_j i_j}-\lambda_j E)Q_{j n_j s_j i_j}^{(0)}=0, \tag{21} \]

where

\[ Q_{j n_j s_j i_j}^{(0)} = \bigl(\{q_{j n_j}^{(0)}\}_{v_{i_j j-1}+1},\ \{q_{j n_j}^{(0)}\}_{v_{i_j j-1}+2},\ldots, \{q_{j n_j}^{(0)}\}_{v_{i_j j}}\bigr), \tag{22} \]

\[ v_{i_j j}=k_1+\cdots+k_{j-1}+s_{j1}+\cdots+s_{j i_j}; \]

\[ n_j,i_j=1,\ldots,r_j;\quad j=1,\ldots,p. \]

Taking (5) into account, from (21) we find

\[ \{q_{j n_j}^{(0)}\}_{v_{i_j j-1}+2} =\cdots= \{q_{j n_j}^{(0)}\}_{v_{i_j j}}=0,\qquad n_j,i_j=1,\ldots,r_j;\ j=1,\ldots,p. \tag{23} \]

The component \(\{q_{j n_j}^{(0)}\}_{v_{i_j j-1}+1}\) is arbitrary. Put

\[ \{q_{j n_j}^{(0)}\}_{v_{i_j j-1}+1} = \begin{cases} 1, & i_j=n_j,\\ 0, & i_j\ne n_j;\ i_j,n_j=1,\ldots,r_j;\ j=1,\ldots,p. \end{cases} \tag{24} \]

To determine the vector \(Q_{j n_j}^{(1)}\) and the function \(\lambda_{j n_j}^{(1)}\), compare in (13) the coefficients of \(\mu_{j n_j}^{1}\). Then, repeating the preceding arguments, we find:

\[ Q_{j n_j k_r}^{(1)}\equiv 0,\qquad Q_{j n_j s_j i_j}^{(1)}\equiv 0,\qquad r\ne j;\ i_j\ne n_j;\ i_j,n_j=1,\ldots,r_j; \tag{25} \]

\[ r,j=1,\ldots,p; \]

\[ \{q_{j n_j}^{(1)}\}_{v_{j n_j-1}+2} =\lambda_{j n_j}^{(1)},\qquad \{q_{j n_j}^{(1)}\}_{v_{j n_j-1}+3} =\cdots= \{q_{j n_j}^{(1)}\}_{v_{j n_j}}=0. \tag{26} \]

The component \(\{q_{j n_j}^{(1)}\}_{v_{j n_j-1}+1}\) is arbitrary. We put it equal to zero. Equating now in relation (13) the coefficients successively of \(\mu_{j n_j}^{2},\ldots,\mu_{j n_j}^{s_{j n_j}-1}\) and applying the method of induction, we have

\[ \{q_{j n_j}^{(s_{j n_j}-1)}\}_{v_{j n_j}} = (\lambda_{j n_j}^{(1)})^{s_{j n_j}-1},\qquad \{q_{j n_j}^{(\eta_{j n_j})}\}_{v_{j n_j}}=0,\qquad 0\le \eta_{j n_j}\le s_{j n_j}-2. \tag{27} \]

Then, equating in (13) the coefficients of \(\mu_{j n_j}^{s_{j n_j}}\) and taking (27) into account, we find

\[ \lambda_{i n_j}^{(1)}(\tau) = \sqrt[s_{j n_j}]{ \left\{ T^{-1}(\tau) \left[ A^{(1)}(\tau)T_{\nu_{j n_j}-1+1}(\tau) - \frac{dT_{\nu_{j n_j}-1+1}(\tau)}{dt} \right] \right\}_{\nu_{j n_j}} }, \tag{28} \]

\[ n_j=1,\ldots,r_j;\quad j=1,\ldots,p. \]

All subsequent coefficients of the series (10) are determined in the same way.

The vector \(P_{n_1}(\tau,\mu_{1n_1})\) and the function \(z_{n_1}(\tau,\mu_{1n_1})\) \((n_1=1,\ldots,r_1)\) are determined from relation (14).

Applying the method set forth in \((^1)\), one can prove the following theorem.

Theorem 2. If the conditions of Theorem 1 are satisfied and

\[ x\big|_{t=0}=x_m\big|_{t=0};\qquad \operatorname{Re}\left( \sum_{s=0}^{s_{j n_j}-1} \mu_{j n_j}^{s}\lambda_{j n_j}^{(s)} \right)\leq 0, \quad n_j=1,\ldots,r_j;\quad j=1,\ldots,p, \tag{29} \]

where \(x_m\) is the vector defined by relations (7)—(12); if the series are cut off after the \(m\)-th terms, then for any \(L>0\) and \(0<\mu_{j n_j}\leq \bar{\mu}_{j n_j}\) one can indicate constants \(C_{j n_j}\), independent of the parameter \(\mu_{j n_j}\), such that

\[ \|x-x_m\| \leq \sum_{j=1}^{p}\sum_{n_j=1}^{r_j} \mu_{j n_j}^{\,m+2-2s_{j n_j}}C_{j n_j}. \tag{30} \]

Relation (30) proves the asymptotic character of the solution \(x_m\).

Kyiv State Pedagogical Institute
named after A. M. Gorky

Received
3 I 1963

References

  1. N. N. Bogolyubov, On certain statistical methods in mathematical physics, 1945.
  2. S. F. Feshchenko, Scientific Notes of the Kyiv Pedagogical Institute, 9, No. 4 (1949).
  3. Ya. D. Tamarkin, On certain general problems of the theory of ordinary linear differential equations and on the expansion of arbitrary functions in series, 1917.
  4. Yu. A. Mitropolsky, Scientific Notes of Kyiv University, 26, No. 2, 53 (1957).
  5. S. F. Feshchenko, Reports of the Academy of Sciences of the Ukrainian SSR, 1, 11 (1949).
  6. Yu. L. Daletskii, S. G. Krein, Ukrainian Mathematical Journal, 2, No. 4, 71 (1950).
  7. M. I. Shkil, Reports of the Academy of Sciences of the Ukrainian SSR, 9, 1138 (1962).

Submission history

N. I. Shkil