Selected Exercise Solutions¶
This appendix provides solutions to selected exercises marked with \((\star)\) throughout the book. Solutions to all programming exercises can be found in the companion GitHub repository.
Chapter 1: Mathematical Prerequisites¶
Exercise 1.1 (Fundamental): Show that the set of \(2\times2\) Hermitian matrices forms a real vector space. Find a basis.
Solution: A Hermitian matrix \(A\) satisfies \(A=A^\dagger\), i.e., \(A_{ij} = A_{ji}^*\). The diagonal entries are real; the off-diagonal entries are complex conjugates. A basis for \(2\times2\) Hermitian matrices: \(\{I, X, Y, Z\}\) (the Pauli matrices plus identity). Dimension: 4 (real). Linear combinations with real coefficients: \(aI + bX + cY + dZ\) for \(a,b,c,d\in\mathbb{R}\).
Chapter 2: Quantum Computing Essentials¶
Exercise 2.1 (Fundamental): Prove that the Hadamard gate is its own inverse.
Solution: \(H^2 = \frac{1}{\sqrt{2}}\begin{pmatrix}1&1\\1&-1\end{pmatrix} \cdot\frac{1}{\sqrt{2}}\begin{pmatrix}1&1\\1&-1\end{pmatrix} = \frac{1}{2}\begin{pmatrix}2&0\\0&2\end{pmatrix} = I\). ✓
Exercise 2.2 (Intermediate): Show that CNOT transforms \(|++\rangle\) to \(|++\rangle\) and \(|+-\rangle\) to \(|--\rangle\) (Hadamard basis).
Solution: \(|+\rangle = (|0\rangle+|1\rangle)/\sqrt{2}\), \(|-\rangle = (|0\rangle-|1\rangle)/\sqrt{2}\).
\(\text{CNOT}|++\rangle = \text{CNOT}\cdot\frac{1}{2}(|00\rangle+|01\rangle+|10\rangle+|11\rangle) = \frac{1}{2}(|00\rangle+|01\rangle+|11\rangle+|10\rangle) = |++\rangle\). ✓
\(\text{CNOT}|+-\rangle = \text{CNOT}\cdot\frac{1}{2}(|00\rangle-|01\rangle+|10\rangle-|11\rangle) = \frac{1}{2}(|00\rangle-|01\rangle+|11\rangle-|10\rangle) = |--\rangle\). ✓
Chapter 6: Variational Quantum Circuits¶
Exercise 6.1: For a QAOA circuit with \(p=1\) layer on a MaxCut instance with 4 nodes arranged in a cycle (edges \(\{01, 12, 23, 30\}\)), write out the cost and mixer Hamiltonians and derive the QAOA expectation value as a function of \((\gamma, \beta)\).
Solution: Cost Hamiltonian for MaxCut: \(H_C = \frac{1}{2}\sum_{\{i,j\}\in E}(1 - Z_iZ_j) = 2 - \frac{1}{2}(Z_0Z_1 + Z_1Z_2 + Z_2Z_3 + Z_3Z_0)\).
Mixer: \(H_B = \sum_{i=0}^3 X_i\).
\(p=1\) QAOA state: \(|\psi(\gamma,\beta)\rangle = e^{-i\beta H_B}e^{-i\gamma H_C}|+\rangle^4\).
After calculation, \(\langle H_C\rangle = 2 - \sin(4\beta)\sin(\gamma)\cos^3(\gamma) \cdot[1+\cos(\gamma)\text{ terms}]\). Maximum cut size for a 4-cycle is 4 (bipartite); \(p=1\) QAOA achieves \(\langle H_C\rangle_{\max} = 2.5\) (approximation ratio \(5/8\)).
Chapter 10: Quantum Neural Networks¶
Exercise 10.1: Derive the PSR for \(f(\theta) = \langle\psi|R_y^\dagger(\theta) Z R_y(\theta)|\psi\rangle\) where \(|\psi\rangle = |+\rangle\).
Solution: \(|\psi(\theta)\rangle = R_y(\theta)|+\rangle\). \(f(\theta) = \langle+|R_y^\dagger(\theta)Z R_y(\theta)|+\rangle\).
\(R_y(\theta)|+\rangle = \frac{1}{\sqrt{2}}(R_y(\theta)|0\rangle + R_y(\theta)|1\rangle) = \frac{1}{\sqrt{2}}\left[\begin{pmatrix}\cos(\theta/2)\\\sin(\theta/2)\end{pmatrix} + \begin{pmatrix}-\sin(\theta/2)\\\cos(\theta/2)\end{pmatrix}\right]\).
Computing: \(f(\theta) = \langle\psi(\theta)|Z|\psi(\theta)\rangle = -\sin\theta\).
PSR: \(f'(\theta) = \frac{1}{2}[f(\theta+\pi/2) - f(\theta-\pi/2)] = \frac{1}{2}[-\sin(\theta+\pi/2) - (-\sin(\theta-\pi/2))] = \frac{1}{2}[-\cos\theta + \cos\theta]\cdot 2 = -\cos\theta\).
Direct: \(\frac{d}{d\theta}(-\sin\theta) = -\cos\theta\). ✓
Chapter 15: Optimization¶
Exercise 15.1: Prove that for \(f(\theta) = a\cos\theta + b\sin\theta + c\), the PSR \(\frac{1}{2}[f(\theta+\pi/2) - f(\theta-\pi/2)] = \frac{df}{d\theta}\).
Solution: LHS \(= \frac{1}{2}[a\cos(\theta+\pi/2) + b\sin(\theta+\pi/2) + c - a\cos(\theta-\pi/2) - b\sin(\theta-\pi/2) - c]\)
\(= \frac{1}{2}[-a\sin\theta + b\cos\theta - a\sin\theta + b\cos\theta\cdot(-1)\cdot(-1) …]\)
Wait, let us compute directly: \(\cos(\theta\pm\pi/2) = \mp\sin\theta\), \(\sin(\theta\pm\pi/2) = \pm\cos\theta\).
LHS \(= \frac{1}{2}[a(-\sin\theta) + b\cos\theta - a\sin\theta - b(-\cos\theta)]\)
\(= \frac{1}{2}[-2a\sin\theta + 2b\cos\theta] = -a\sin\theta + b\cos\theta\).
RHS \(= \frac{d}{d\theta}(a\cos\theta + b\sin\theta + c) = -a\sin\theta + b\cos\theta\). ✓
Chapter 16: Barren Plateaus¶
Exercise 16.1: For \(n=2\) qubits and Haar-random unitaries, compute \(\mathrm{Var}_U[\partial_k\langle Z_0\rangle]\) (local cost).
Solution: By the 2-design integration formula, for local observable \(Z_0\): \(\mathbb{E}_U[\langle Z_0\rangle] = \mathrm{Tr}[Z_0]/2^n = 0\). \(\mathbb{E}_U[\langle Z_0\rangle^2] = \frac{\mathrm{Tr}[Z_0^2]}{2^n(2^n+1)} \cdot 2^n = \frac{2}{2^2+1} = \frac{2}{5}\) for \(n=2\).
The PSR gives \(\partial_k\langle Z_0\rangle = \frac{1}{2}[\langle Z_0\rangle_+ - \langle Z_0\rangle_-]\). \(\mathrm{Var}[\partial_k\langle Z_0\rangle] = \frac{1}{4}\cdot 2\cdot\mathbb{E}[\langle Z_0\rangle^2] = \frac{1}{4}\cdot 2\cdot\frac{2}{5} = \frac{1}{5}\).
This is \(O(1)\) (independent of \(n\)), confirming local costs avoid barren plateaus. ✓
Chapter 20: Finance¶
Exercise 20.1: Formulate the binary knapsack problem as a QUBO.
Solution: Knapsack: maximize \(\sum_i v_i x_i\) subject to \(\sum_i w_i x_i \leq W\), \(x_i\in\{0,1\}\).
QUBO (penalty method): \(\mathcal{Q}(\mathbf{x}) = -\sum_i v_i x_i + \lambda\left(\sum_i w_i x_i - W\right)^2 \cdot \mathbf{1}[\sum_i w_i x_i > W]\)
Approximate continuous penalty (always active): \(\mathcal{Q}(\mathbf{x}) = -\sum_i v_i x_i + \lambda\left(\sum_i w_i x_i - W\right)^2\)
Q matrix entries: \(Q_{ii} = -v_i + \lambda w_i^2 - 2\lambda W w_i\), \(Q_{ij} = 2\lambda w_i w_j\) for \(i\neq j\).
Compare portfolio QUBO: both are binary, quadratic, with linear and quadratic terms. Knapsack uses an inequality constraint (converted via slack variable or soft penalty); portfolio uses an equality constraint (cardinality \(\sum x_i = k\)).