Low Rank ADI Solution of Sylvester Equation via Exact Shifts
Low Rank ADI Solution of Sylvester Equation via Exact Shifts
Low Rank ADI Solution of Sylvester Equation via Exact Shifts
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
∑where γ = n 0j=1|µ j − λ j |τ (j)max σ (j)max and σ (j)max, τ (j)max are defined as in (3.20).Pro<strong>of</strong>. Take the norm <strong>of</strong> δX ≈ δX 1 + δX 2 + δX 3 and use bounds (3.21),(3.22), (3.23) to get (3.24). ✷Remark 3.3 From (3.17), and similar equalities for δX 2 and δX 3 , we canobtain a sharper boundn‖δX‖ 0‖X‖ ≤ ∑j=1(×|µ j − λ j | ‖S∆ Φ (j)S −1 ‖ ‖T ∆ Ψ (j)T −1 ‖‖δA‖ + ‖δB‖ + ‖S−1 (GδF ∗ + δGF ∗ )T ‖‖X‖)+ O(ɛ 2 ),where diagonal matrices ∆ Φ (j) and ∆ Ψ (j) are defined as in (3.18) and (3.19).Remark 3.4 Using ‖AB‖ F ≤ ‖A‖‖B‖ F and ‖AB‖ F ≤ ‖A‖ F ‖B‖ (see, e.g.,[45, Theorem II. 3.9.]) similarly like in the previous calculations one can obtainfollowing bound for Frobenius norm‖δX‖ F‖X‖ F≤∑where γ = n 0j=1(κ(T )‖S‖‖S −1 δA‖ + κ(S)‖T −1 ‖‖δB T ‖+‖S‖‖T −1 ‖ ‖S−1 (GδF ∗ + δGF ∗ ))T ‖ Fγ + O(ɛ 2 ), (3.25)‖X‖ F|µ j − λ j |τ (j)max σ (j)max and σ (j)max, τ (j)max are defined as in (3.20).As an illustration <strong>of</strong> the quality <strong>of</strong> the perturbation bound (3.25) we willpresent a comparison between this bound and two bounds from [23]. The firstone is‖δX‖ F‖X‖ F≤ √ 3Ψɛ + O(ɛ 2 ), (3.26)whereΨ = ‖P −1 [α(X T ⊗I m ) −β(I n ⊗X) −δI mn ]‖/‖X‖ F ,{ ‖δA‖Fand ɛ = max α , ‖δB‖ Fβ, ‖δC‖ Fδas in [23]. The second bound is a weaker version <strong>of</strong> (3.26):P = I n ⊗A−B T ⊗I m}, while α, β and δ are scaling factors‖δX‖ F‖X‖ F≤ √ 3Φɛ + O(ɛ 2 ), (3.27)13