Approximation theorems for operator exponential and cosine functions (Q844366)

From MaRDI portal
Revision as of 10:03, 2 July 2024 by ReferenceBot (talk | contribs) (‎Changed an Item)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
scientific article
Language Label Description Also known as
English
Approximation theorems for operator exponential and cosine functions
scientific article

    Statements

    Approximation theorems for operator exponential and cosine functions (English)
    0 references
    0 references
    0 references
    19 January 2010
    0 references
    The subject of this work are abstract Cauchy problems in a Hilbert space \({\mathbf H}\) of the following form: \[ {{d^2x(t)}\over{dt^2}}+Ax(t)=0,\quad t\in(0,T],\quad x(0)=x_0,\quad x'(0)=0,\tag{1} \] and \[ {{dx(t)}\over{dt}}+Ax(t)=0,\quad t>0,\quad x(0)=x_0.\tag{2} \] For problem (1), it is assumed that the operator \(A\) acting in \({\mathbf H}\) with domain \(D(A)\) satisfies the following conditions \(A=A^*>\lambda_0 I\), \(\lambda_0>0\), \(x_0\in D(A^\sigma)\), \(\sigma>5/4\). Then, it follows from cited works that the solution \(x(t)=\cos(t\sqrt{A})\) can be represented by a series involving the Laguerre polynomials \(L^0_n(t)\) multiplied by \(e^{-\delta t}\), \(\delta>1/2\). Let \(z^2_N=\int_0^\infty t^{-1}e^{-(1-2\delta)t}\|x(t)-x_N(t)\|^2\,dt\), where \(x_N(t)\) is the \(N\)-th partial sum of the above mentioned series. The following theorem is proven: Theorem 1: If \(A=A^*>0\), \(\sigma>0\), and if the non decreasing sequence of positive numbers \(c_k\) is such that \(\sum_{k=0}^\infty {1\over{c_{2^k}}}\) converges, then the condition \(z^2_N\leq{1\over{N^{2\sigma}c_N}}\) implies that \(x_0\in D(A^{\sigma/2})\). For problem (2) with \(A\) self-adjoint and positive definite with domain \(D(A)\), the solution \(x(t)=e^{-tA}x_0\) can also to be expressed in the form \(e^{-t\gamma}\) \((\gamma>0)\) times certain series involving Laguerre polynomials \(L^0_n(2\gamma t)\) and the Cayley transform \(T_{\gamma}\) of \(A\). For effective calculations, this requires to solve recursively an elliptic problem with the operator \(\gamma I+A\). From cited papers, it follows also that \(\|x(t)-x_N(t)\|\leq cN^{-\sigma}\|A^{\sigma}x_0\|\), \(\sigma>1\), with independent constant \(c\). Theorem 2, proven in this work, specifies the constant \(c\) as \(c={{\sigma^{2\sigma-1}}\over{e^\sigma2^{2\sigma+1}\gamma^{2\sigma}}}\) for all \(N\). Theorem 3 states that, for any unbounded sequence of positive \(c_N\), there is no positive constant \(c\) satisfying \(z^2_{t,N}\leq{c\over{N^{2\sigma}c_N}}\), where \(z^2_{t,N}=\int_0^\infty t^{-t}\|x(t)-x_N(t)\|^2\,dt\), while Theorem 4 says that under the assumptions of Theorem 1, the condition \(z^2_{t,N}\leq{1\over{N^{2\sigma}c_N}}\) implies that \(x_0\in D(A^\sigma)\).
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    Cauchy problem
    0 references
    Cayley transform
    0 references
    rate of convergence
    0 references
    unimprovable estimate
    0 references
    inverse approximation theorem
    0 references
    0 references