Ground-state energy: VQE in practice, with a worked small molecule
The variational quantum eigensolver (Chapter 122) has computed the ground-state energy of H₂ on almost every quantum processor ever built. Reaching chemical accuracy for H₂ is easy. What is hard is the next step up: every added electron multiplies the number of Hamiltonian terms to measure, and the shots needed per term. This chapter runs a complete VQE on H₂ in the book’s simulator, with realistic shot noise, and shows where the cost goes.
On the plate: the H₂ dissociation curve, exact and estimated by VQE.
Ansätze
The unitary coupled-cluster ansatz, UCCSD, starts from the Hartree–Fock state and applies with the single and double excitations, each with an adjustable amplitude. For H₂ in a minimal basis, symmetry leaves one double excitation, , so the ansatz is , which contains the exact answer. For larger molecules UCCSD has many parameters and deep circuits; hardware-efficient ansätze are shallow but lack chemical structure and risk barren plateaus (Chapter 123); adaptive schemes such as ADAPT-VQE, proposed by Harper Grimsley and colleagues in 2019, grow the ansatz one operator at a time. Established
Run VQE with shot noise
Bond length 1.40 bohr; shots per energy estimate 10^4.00; . The plate evaluates the ansatz energy from the exact integrals (Book.H2), adds the statistical error of a finite number of shots, with standard deviation about 1 hartree per shot, and minimizes over θ on a grid. Notice two effects. Random error: the estimate scatters around the true minimum. Bias: picking the lowest of many noisy estimates systematically undershoots, a “winner’s curse” that can even put the VQE estimate below the exact energy, seemingly violating the variational principle. Chemical accuracy needs about shots per energy evaluation. Established
Grouping, mitigation, and the scaling problem
Pauli terms that commute qubit by qubit can be measured together, reducing the number of distinct circuits several-fold; factorizing the Hamiltonian into low-rank pieces reduces it further. Error mitigation (Chapter 152) removes some hardware bias at extra sampling cost. Still, Jérôme Gonthier and colleagues estimated in 2022 that measuring the energy of modest molecules such as methane or ethanol to chemical accuracy would take on the order of days per energy evaluation on current hardware, and an optimization needs thousands of evaluations. VQE on small molecules is a solved demonstration; at scale, measurement cost, noise and trainability together make advantage unlikely without error correction. Established for small molecules; Debated at scale.
What you now have
UCCSD, hardware-efficient and adaptive ansätze, a complete H₂ VQE with exact integrals and shot noise, statistical error and winner’s-curse bias, the shot budget for chemical accuracy, measurement grouping, and the measurement-cost critique.
· · ·
Three questions to carry with you.
And one to work out. With hartree and target error 1.6 mHa, how many shots does give?
Sources: A. Peruzzo et al., Nat. Commun. 5, 4213 (2014). P. J. J. O’Malley et al., Phys. Rev. X 6, 031007 (2016). H. R. Grimsley et al., Nat. Commun. 10, 3007 (2019). J. F. Gonthier et al., Phys. Rev. Research 4, 033154 (2022). Checked by the plate: the noise-free minimum equals the exact FCI energy from Book.H2 at every bond length.