Paper 3, Section II, 40E

Numerical Analysis
Part II, 2020

(a) Give the definition of a normal matrix. Prove that if AA is normal, then the (Euclidean) matrix 2\ell_{2}-norm of AA is equal to its spectral radius, i.e., A2=ρ(A)\|A\|_{2}=\rho(A).

(b) The advection equation

ut=ux,0x1,0t<u_{t}=u_{x}, \quad 0 \leqslant x \leqslant 1, \quad 0 \leqslant t<\infty

is discretized by the Crank-Nicolson scheme

umn+1umn=14μ(um+1n+1um1n+1)+14μ(um+1num1n),m=1,2,,M,nZ+u_{m}^{n+1}-u_{m}^{n}=\frac{1}{4} \mu\left(u_{m+1}^{n+1}-u_{m-1}^{n+1}\right)+\frac{1}{4} \mu\left(u_{m+1}^{n}-u_{m-1}^{n}\right), \quad m=1,2, \ldots, M, \quad n \in \mathbb{Z}_{+}

Here, μ=kh\mu=\frac{k}{h} is the Courant number, with k=Δt,h=Δx=1M+1k=\Delta t, h=\Delta x=\frac{1}{M+1}, and umnu_{m}^{n} is an approximation to u(mh,nk)u(m h, n k).

Using the eigenvalue analysis and carefully justifying each step, determine conditions on μ>0\mu>0 for which the method is stable. [Hint: All M ×M\times M Toeplitz anti-symmetric tridiagonal (TAT) matrices have the same set of orthogonal eigenvectors, and a TAT matrix with the elements aj,j=aa_{j, j}=a and aj,j+1=aj,j1=ba_{j, j+1}=-a_{j, j-1}=b has the eigenvalues λk=a+2ibcosπkM+1\lambda_{k}=a+2 \mathrm{i} b \cos \frac{\pi k}{M+1} where i=1\mathrm{i}=\sqrt{-1}. ]

(c) Consider the same advection equation for the Cauchy problem (xR,0t(x \in \mathbb{R}, 0 \leqslant t \leqslant T)T). Now it is discretized by the two-step leapfrog scheme

umn+1=μ(um+1num1n)+umn1.u_{m}^{n+1}=\mu\left(u_{m+1}^{n}-u_{m-1}^{n}\right)+u_{m}^{n-1} .

Applying the Fourier technique, find the range of μ>0\mu>0 for which the method is stable.