Past exam of the mathematics course of the year 2 of the natural sciences course of the University of Cambridge / 2018

Apply the product rule in Einstein notation, using the Levi-Civita symbol to express the cross product:The third line uses a cyclic permutation of indices in the first term and an interchange of two indices in the second. Rearranging proves the divergence of a cross product identity:

Take any constant vector $\mathbf{a}$. Its curl is zero, so part (a) and the divergence theorem givewhere $\mathbf{n}$ is the outward unit normal vector. The scalar triple product identity rewrites the last integrand as $\mathbf{a}\cdot (\mathbf{F}\times \mathbf{n})$. Since $\mathbf{a}$ is arbitrary, equality of these scalar products implies the vector identity

First compute the curl directly:Reflection symmetry in $x$ makes the last component integrate to zero. The middle component gives the hemisphere's volume, soFor an independent surface integral, the closed boundary consists of the curved hemisphere and its flat base. The right-hand side in part (b) is ${\int }_S\mathbf{n}\times \mathbf{F}dS$.

On the curved part use spherical coordinates with $0\le \theta \le \pi /2$ and $0\le \varphi <2\pi$. There,The last component has zero angular integral. The first two components giveOn the base, $z=0$ and the outward normal vector is $-{\mathbf{e}}_z$. Thus $\mathbf{n}\times \mathbf{F}=(x^2+y^2,0,0)$. In polar coordinates its integral isThe first components cancel, and therefore both required sides equal

For the linearly damped string, substitute $u(x,t)=X(x)T(t)$ and divide by $XT$:Thus separation of variables givesThe fixed-end Dirichlet boundary conditions require $X(0)=X(L)=0$. Nonzero solutions are $X_n(x)=\sin (n\pi x/L)$, with ${\alpha }_n=n\pi /L$, $n\ge 1$. Neither $\alpha =0$ nor a negative spatial eigenvalue produces a nonzero function obeying both endpoint conditions.

For each sine normal mode, the temporal characteristic equation is $s^2+2\lambda s+{\alpha }_n^2=0$, with roots $s=-\lambda \pm \sqrt{{\lambda }^2-{\alpha }_n^2}$. Since $\lambda L/\pi =M+1/2$, the first $M$ modes are overdamped and all higher modes are underdamped. No integer $n$ gives critical damping.

The zero initial displacement excludes the hyperbolic cosine and cosine terms. HencewhereTo impose the initial velocity, let $f_n=\frac{2}{L}{\int }_0^{L}f(x)\sin ({\alpha }_{n}x)dx$ be its Fourier sine series coefficients. Differentiating the expansion at $t=0$ gives $A_n{\Omega }_n=f_n$ for the overdamped modes and $A_n{\omega }_n=f_n$ for the underdamped modes. These determine all coefficients. Despite the hyperbolic sine, the first modes decay: ${\Omega }_n<\lambda$, so both temporal exponential rates are negative.

Here $\lambda =7/2$ and ${\alpha }_n=n$. LetTwo applications of integration by parts giveSolving these equations,Consequently the required amplitudes areThe stated initial velocity is nonzero at the fixed endpoints, whereas a classical solution differentiable there would require zero endpoint velocity. Thus it is understood on the open interval, or as an L2 space initial datum in the usual finite-energy solution. Its Fourier sine series takes value zero at the endpoints; this changes no initial energy or interior initial data.

For $x\ne \xi$, the homogeneous solutions are $e^x,e^{-x}$. The left boundary condition selects $\sinh x$, while boundedness at infinity selects $e^{-x}$ on the right. Thus write $G=A(\xi )\sinh x$ for $x<\xi$ and $G=B(\xi )e^{-x}$ for $x>\xi$.

A unit Dirac delta distribution in the second derivative requires continuity of $G$ and a derivative jump of one:The second equation and the first give $-A{e}^{\xi }=1$, so the Dirichlet half-line Green function for d2 minus 1 isThe second expression is also a method of images construction. Applying the Green's function to the forcing givesFor bounded forcing this is bounded, vanishes at zero and solves the equation; a bounded homogeneous solution vanishing at zero is zero, proving uniqueness. More general forcing is allowed when the integrals converge and the resulting function has the required growth. Arbitrary unbounded forcing need not admit a bounded solution; this distinction is relevant in part (b).

For $x>0$, differentiation of $u=y/x$ givesThus the radial Laplacian equation reduces toBoundedness of $u$ at zero implies $y(0)=0$. Conversely, a regular reduced solution with $y(0)=0$ gives finite $u(0)=y^{\prime }(0)$.

There is an important qualification to the reduction in the printed question: bounded $u$ at infinity requires $y=O(x)$, not bounded $y$. The example in part (ii) actually has $y\sim -x$, so it cannot be the bounded solution described in part (a). The differential equation and origin condition reduce as asserted, and the same Green's function can still be used with the correct growth condition. This is the radial reduction and growth at infinity distinction.

The reduced forcing is $f(\xi )=\xi 1_{[1,\infty )}(\xi )$. Its exponential-weighted integrals converge, even though the resulting reduced function is not bounded. For $0\le x<1$, the Green's function formula givesFor $x\ge 1$ it givesUse $\int \xi \sinh \xi d\xi =\xi \cosh \xi -\sinh \xi$ and ${\int }_x^{\infty }\xi e^{-\xi }d\xi =(x+1)e^{-x}$ to simplify this to $y(x)=-x+e^{-x-1}$. Dividing by $x$ yieldsThus $a(x)=-(2/e)\sinh x/x$ with its continuous value at zero. At $x=1$, both pieces give $u=-1+e^{-2}$ and $u^{\prime }=-2e^{-2}$, so neither a Dirac delta distribution nor its derivative is introduced at the forcing jump. The solution is bounded at both ends, with $u\to -1$ at infinity. Its reduced function $y=xu\sim -x$ confirms the growth qualification in part (i).

Using Fourier inversion with the stated convention,Hence applying the transform twice gives the duality property of the Fourier transform:The formula holds directly for rapidly decreasing smooth functions, and extends through the usual L2 space or tempered distribution formulations when absolute convergence is unavailable.

For a real nonzero $b$, put $s=bt$ in the Fourier transform. Reversing the integration limits when $b<0$ produces the absolute value:This is the scaling property of the Fourier transform. If $b=0$ and $f(0)$ exists, the transformed function is constant in time; its transform is $2\pi f(0)\delta (\omega )$ in the sense of tempered distributions, rather than an ordinary convergent Fourier integral.

The Heaviside step function restricts the integral defining the Fourier transform to $t\ge 0$, soThe convergence condition is essential. For real $\alpha$ it means $\alpha >0$, which is the decay assumption used in the subsequent parts. At $\alpha =0$ a distributional Fourier transform is needed; for negative real $\alpha$ the defining improper integral diverges.

For $\mathrm{Re}\alpha >0$, the even exponential is $h(t)=g(t)+g(-t)$ except for an irrelevant single value at zero. Linearity and the scaling property of the Fourier transform at $b=-1$ giveIn particular, for positive real $\alpha$ the transform is real and even, as expected from the symmetry of $h$.

Set $\alpha =1$ in part (d), obtaining $\mathcal{F}[e^{-|t|}](\omega )=2/(1+{\omega }^2)$. Apply the duality property of the Fourier transform from part (a):Therefore

For $T>0$ the difference of Heaviside step functions is an interval indicator, and $v=1/2$ on $(-T,T)$ and zero elsewhere. Its Fourier transform isThe apparent singularity at $\omega =0$ is removable. The formula remains valid for real negative $T$, because the original step-function difference and the oriented integral both reverse sign; for $T=0$ both vanish. This is the Fourier transform of an interval indicator and produces a scaled sinc function.

By Fourier inversion, the sinc function $\sin (At)/(\pi t)$ has Fourier transform equal to one for $|\omega |<A$ and zero for $|\omega |>A$. Thus the two bandwidths add to giveThe requested squared-magnitude spectrum therefore has heights four and one:Values at the jump points depend on the chosen Fourier convergence convention and do not affect any integral. The graph is shown below.

The Parseval identity with this Fourier transform normalization gives

Induction shows $(PA{P}^{-1}{)}^k=P{A}^k{P}^{-1}$ for every nonnegative integer $k$: in a product, adjacent factors $P^{-1}P$ cancel. At $k=0$, this is $I=PI{P}^{-1}$. Hence each partial sum of the matrix exponential satisfiesThe series is absolutely convergent, since a submultiplicative matrix norm bounds it by ${\sum }_k\Vert A{\Vert }^k/k!=e^{\Vert A\Vert }$. Passing to the limit therefore proves that the matrix exponential respects similarity:No diagonalizability assumption is required.

If $A^{†}=-A$, then $H=-iA$ is a Hermitian matrix, because $H^{†}=i{A}^{†}=-iA=H$. By the finite-dimensional spectral theorem, there is a unitary matrix $Q$ withThus $A=Q\mathrm{diag}(i{\theta }_j)Q^{†}$. Since the matrix exponential respects similarity,Each diagonal entry has modulus one, so $C^{†}C=I$. Using $Q^{†}Q=Q{Q}^{†}=I$,Hence the exponential of a skew-Hermitian matrix is unitary.

The finite-dimensional spectral theorem gives a spectral decomposition of the Hermitian matrix:The matrix exponential respects similarity, soThe diagonal entries are real, so the conjugate transpose of this expression equals itself. Its eigenvalues are $e^{{\lambda }_j}>0$, provingIn particular, for any nonzero $v$, $v^{†}{e}^{A}v={\sum }_j{e}^{{\lambda }_j}|(Q^{†}v{)}_j{|}^2>0$. Thus the exponential of a Hermitian matrix is a Hermitian positive-definite matrix, even when $A$ itself has negative eigenvalues.

A square matrix $A$ is a diagonalizable matrix over a specified field if some invertible matrix $P$ makes $P^{-1}AP$ diagonal. Equivalently, the underlying vector space has a basis of eigenvectors of $A$.

For example,The first is already diagonal. The second has only the eigenvalue one, of algebraic multiplicity two, but its eigenspace consists of $(x,0{)}^T$ and has dimension one. It cannot supply a two-vector eigenbasis.

Every row sums to six, so $\mathbf{e}=(1,1,1{)}^T$ is an eigenvector with eigenvalue six. Since $M$ is a real symmetric matrix, its other two eigenvalues, say ${\lambda }_2,{\lambda }_3$, are real. The trace givesAlso, for a real symmetric matrix, $\mathrm{tr}(M^2)$ is the sum of the squares of all its entries. HereThus ${\lambda }_2^2+{\lambda }_3^2=42-36=6$. Together with their zero sum, this implies

By the spectral theorem for real symmetric matrices, choose an orthonormal eigenbasis ${\mathbf{u}}_1,{\mathbf{u}}_2,{\mathbf{u}}_3$ corresponding to $6,\sqrt{3},-\sqrt{3}$, with ${\mathbf{u}}_1=(1,1,1{)}^T/\sqrt{3}$. Write $P=[{\mathbf{u}}_1{\mathbf{u}}_2{\mathbf{u}}_3]$ and $\mathbf{x}=P\mathbf{\xi }$. This orthogonal change of coordinates givesThe plane constraint is therefore ${\xi }_1=0$. On that plane the quadratic form vanishes exactly whenConsequently the entire set isThe two direction vectors are nonzero and not parallel, since ${\mathbf{u}}_2,{\mathbf{u}}_3$ are linearly independent. An invertible orthogonal transformation maps each full line to a full line. Thus the set consists of precisely two infinite lines through the origin.

The derivative of an analytic function must be independent of the direction of approach. Taking a real increment gives $f^{\prime }(z)=u_x+i{v}_x$; taking an imaginary increment givesEquality of real and imaginary parts proves the Cauchy-Riemann equations:A real harmonic function is a twice continuously differentiable function whose Laplacian vanishes. For the given logarithmic function and $x^2+y^2>0$,It is therefore harmonic everywhere on the punctured complex plane.

On any domain admitting a branch of the complex logarithm, chooseIts derivatives are $v_x=-2y/(x^2+y^2)$ and $v_y=2x/(x^2+y^2)$, as required. For example, the principal complex logarithm gives such a harmonic conjugate on the plane cut along the negative real axis.

There is no single-valued harmonic conjugate on the whole punctured plane. Indeed its putative differential would satisfywhereas a globally defined function has zero integral of its differential around a closed curve. This is the log modulus has no global harmonic conjugate on the punctured plane obstruction. Thus the conjugate requested in the question must be interpreted locally or on a cut plane.

Put $w=z-4i$. Factor out the nonzero constant in the denominator and use the geometric series:This Taylor series converges when $|z-4i|<|3-4i|$. The simple pole at $z=3$ is the only finite obstruction to analyticity, whose distance from the expansion centre is five. Hence the radius of convergence is

Expand the exponential in inverse powers of $z$:This is a Laurent series, not a Taylor series, and converges throughout $0<|z|<\infty$. Every displayed coefficient of a negative power is nonzero, so its principal part has infinitely many terms. ThusThe residue is the coefficient of $z^{-1}$:

Divide by $2x^2$ to put the ordinary differential equation into normalized form $y^{\prime \prime }+P(x)y^{\prime }+Q(x)y=0$:The coefficients are singular at zero, butare both analytic there. This is precisely the regular singular point criterion for a second-order equation, so $x=0$ is a regular singular point.

Apply the Frobenius method. For $y=x^{\sigma }{\sum }_{n=0}^{\infty }{a}_n{x}^n$, the coefficient of the lowest power $x^{\sigma }$ comes only from $2x^2{y}^{\prime \prime }-x{y}^{\prime }+y$; the remaining term $xy$ starts one power higher. As $a_0\ne 0$, this lowest coefficient gives the indicial equationTherefore the two possible characteristic exponents at a regular singular point areTheir difference is not an integer, so the two corresponding Frobenius solutions do not require a logarithmic correction.

For $n\ge 1$, the coefficient of $x^{n+\sigma }$ isFactoring the bracket gives the recurrence relationFor $\sigma =1/2$ this is $a_n=-a_{n-1}/[n(2n-1)]$. Iteration and $(2n)!=2^{n}n!(2n-1)!!$ yieldFor $\sigma =1$, the recurrence is $a_n=-a_{n-1}/[n(2n+1)]$, givingHere the double factorial products are one at $n=0$, consistent with the chosen normalization. Recognizing the cosine and sine power series, for positive $x$ one may also write the two Frobenius solutions asThe constants multiplying the two independent solutions may be chosen separately.