ANALYTIC–NUMERICAL SOLUTIONS WITH <i>A PRIORI</i> BOUNDS FOR MATRIX–VECTOR PARABOLIC PARTIAL DIFFERENTIAL EQUATIONS (Q4784785)

The following problem for a system of parabolic equations: \[ u_t=A u_{xx}, \qquad 0<x<1, \;t>0 \] \[ B_1 u(0,t)+B_2 u_x(0,t)=0, \quad t>0, \] \[ C_1 u(1,t)+C_2 u_x(1,t)=0,\quad t>0, \] \[ u(x,0)=f(x), \qquad 0 \leq x \leq 1, \] is studied. Here, \(A\) is an \(n\times n\) Hermitian positive definite complex matrix, and may depend on \(t\). The matrices \(B_1, B_2, C_1\) and \(C_2\) are complex \(n\times n\) constant matrices and satisfy two conditions \[ B_1 B_2^{*}= B_2 B_1^{*}, \qquad C_1 C_2^{*}= C_2 C_1^{*} \] with hypotheses \(\operatorname {rank}(B_1^{*}+i B_2^{*})=n= \operatorname {rank}(C_1^{*}+i C_2^*)\). For the time-independent case the authors construct approximate solutions with guaranteed error bounds using spectral information from certain matrix vector Sturm-Liouville problems. For the time-dependent case an approximation procedure which reduces the problem, on each time-step, to the time-independent case, is employed. An algorithm that may be used a priori at each time-step in the time-dependent case to guarantee accuracy to a specified tolerance.