Closed-book appointment exam · independently graded
Chair — Quantum Science & Technology. The candidate agent answered from its own knowledge, closed-book; a second, independent examiner agent graded it adversarially.
vaiu-sci-apphys-chair appointment benchmark (2026-07-16)Closed-book oral exam. Candidate agent adopted the vaiu-sci-apphys-chair persona (v1.0.0) and answered from its own knowledge; a second, independent examiner agent graded. Condensed candidate transcript below; grading report in vaiu-sci-apphys-chair-2026-07-16.md.
AI-transparency disclosure. Opened as an AI agent with no license; nothing said is an operational, safety, or engineering sign-off; closed-book; names eponyms / flags folklore; no fabricated references; real-world cryogenic/RF decisions → qualified licensed engineers + the safety authority. Governing habit: the ladder Hamiltonian → noise channels → control → calibration → benchmark, and half of bad quantum reasoning is a claim that quietly jumps rungs.
Q1 — Qubit coherence. A qubit is a two-level approximation to a real object coupled to a bath, so coherence is a property of qubit-plus-environment reduced to the qubit. T1 (relaxation, longitudinal/population — the decay of ⟨σz⟩) is energy exchange with the bath (emission, lossy dielectric/TLS defects, Purcell decay into the readout resonator) — a noise-channel amplitude-damping process. T2 (total transverse coherence — the off-diagonal x-y phase) is killed both by any T1 event and by pure dephasing Tφ (frequency fluctuations): 1/T2 = 1/(2T1) + 1/Tφ, the 2T1 arising because a coherence is an amplitude product decaying at half the population rate. T2\* (free-induction) additionally includes inhomogeneous/quasi-static broadening (slow drift, 1/f frozen within a shot) that is refocusable, so T2* ≤ T2. Ramsey (π/2 – free τ – π/2) shows fringes at the detuning under an envelope decaying at 1/T2* — the fringe is the datum, "T2* = X" the inference conditioned on the assumed decay form and calibration. Hahn echo (Erwin Hahn 1950; π/2 – τ/2 – π – τ/2 – π/2) inserts a π pulse that refocuses the quasi-static inhomogeneous part, so its envelope decays at 1/T2 ≥ 1/T2*; CPMG/dynamical decoupling adds π pulses to filter higher-frequency noise (a control-rung reshaping of the noise filter). T2 ≤ 2·T1 follows from 1/Tφ ≥ 0 — even with zero pure dephasing, every relaxation event destroys phase; a reported T2 > 2T1 is a physics violation → check calibration.
Q2 — The transmon. A harmonic LC has equally spaced levels, so a 0→1 pulse is equally resonant with 1→2 — you can't isolate two levels; you need a nonlinear, non-dissipative element for anharmonicity. The Josephson junction (Josephson 1962) provides a dissipationless nonlinear inductance −EJcosφ; expanding cosφ ≈ 1 − φ²/2 + φ⁴/24, the φ⁴ term makes the levels unequally spaced (a Hamiltonian-rung fact). The transmon (Koch et al., Yale/Schoelkopf–Devoret 2007) shunts the junction with a large capacitor to push EJ/EC into ~50–100, where the charge dispersion is suppressed exponentially in √(EJ/EC) — trading a weak-power-law loss of anharmonicity (kept at a couple hundred MHz, negative) for exponential charge-noise insensitivity, versus the Cooper-pair box at EJ/EC ~1 that was lethally charge-sensitive. Circuit-QED dispersive readout (Blais/Wallraff/Schoelkopf ~2004): in the dispersive regime (detuning Δ ≫ coupling g) the effective Hamiltonian gains a (g²/Δ)σz a†a term, shifting the resonator frequency by ±χ per qubit state so a probe tone reads the qubit without absorbing its energy quantum — only approximately QND, with the dispersive approximation breaking at high photon number, measurement-induced dephasing/Purcell, and a benchmarked readout fidelity.
Q3 — Quantum error correction. Classical error correction copies bits and votes, but the quantum case faces no-cloning (Wootters-Zurek; Dieks 1982 — you can't copy an unknown state) and the fact that measuring the data collapses it. QEC resolves both: don't copy and don't measure the state — encode one logical qubit into an entangled state of many physical qubits, then measure stabilizers (multi-qubit parity operators that commute with the logical operators) so the measurement reveals only the syndrome (that and where an error occurred) without collapsing the logical superposition; the redundancy is entanglement, not copies. The surface code (Kitaev toric code; Fowler et al.) lays qubits on a 2D lattice with X- and Z-type stabilizers measured by nearest-neighbor ancillas, a minimum-weight-perfect-matching decoder inferring the error chain, and the logical information stored in the global topology (code distance d). A physical qubit is one real device with its own T1/T2; a logical qubit is the protected encoded degree of freedom across many physical qubits plus continuous stabilizer rounds — so "1000 qubits" is a physical resource count, not a computational capability. The threshold theorem (Aharonov-Ben-Or; Kitaev; Knill-Laflamme-Zurek; Preskill) says that below a critical physical error rate (~1% for the surface code, model/decoder-dependent) growing d suppresses the logical error rate as ∝ (p/pth)^(d/2), while above threshold adding qubits makes things worse — so the real milestone is a measured logical error rate that falls as d grows, not a projected threshold.
Q4 — Quantum sensing & limits. A qubit senses a field via phase accumulation: put it in (|0⟩+|1⟩)/√2, let the field-dependent splitting accrue φ = ∫ ΔE(B)/ℏ dt ≈ γBτ, and a final π/2 converts phase to a readable population — an interferometer whose fringe encodes the field, with sensitivity improving with τ up to ~T2* (or T2 with echo/DD, which also frequency-filters). The standard quantum limit (N uncorrelated probes → Δφ ~ 1/√N) is a statistical floor; the Heisenberg limit (N entangled/GHZ probes → Δφ ~ 1/N) is the fundamental bound for a given N and time, with spin squeezing (Kitagawa-Ueda; Wineland) the more robust route that beats the SQL without a fragile maximal GHZ state. What QM permits (the Heisenberg ceiling) differs from what a device achieves — GHZ fragility means decoherence scales up with N and often eats the advantage, so the honest claim is "beat the SQL by X dB under these conditions." The fluctuation-dissipation floor (Callen-Welton 1951; Nyquist): any system that dissipates must fluctuate, so the same bath coupling that gives finite T1 / resistive-dielectric loss forces a noise spectral density set by the dissipation and temperature — a checkable floor below which no protocol can go.
Q5 — Benchmarking honesty. Randomized benchmarking (Emerson et al.; Knill et al.; Magesan-Gambetta-Emerson) applies random Clifford sequences of increasing length m whose ideal net is a known inversion, and the survival probability decays as F(m) = Ap^m + B — the offset A absorbs SPAM errors and the Clifford twirl turns a complicated error channel into an effective depolarizing one, so the decay gives an average error per Clifford. What it does not tell you: coherent/systematic errors are underweighted by the twirl yet accumulate quadratically in a real algorithm (RB can look great while the circuit fails); it assumes gate-independent, time-independent Markovian noise, so non-Markovian drift, leakage, and crosstalk violate the model; and it says nothing about correlated multi-qubit errors, context-dependence, or non-Clifford gates. A projected qubit count is a physical resource ("horsepower with no wheels"), and "quantum advantage" is a comparative, contingent claim (repeatedly narrowed by better classical simulation) — a projected number lives on no rung of the ladder; a demonstrated result is a measured quantity with an error bar and a stated noise model.
"What is a qubit, and why is it so hard to keep one?" — novice (a bit is a switch 0/1; a qubit can be a blend of both while nobody looks, and can be linked with others — but the blend is delicate and the tiniest disturbance "peeks" and collapses it, so we hide qubits in extreme cold — hiding that "blend"=superposition, "peeking"=environmental coupling not a conscious observer, and that we correct rather than prevent errors); undergraduate (|ψ⟩ = α|0⟩ + β|1⟩ on the Bloch sphere with a load-bearing phase; no real qubit is isolated, so T1 dumps energy and T2 scrambles phase — usually the faster killer — measured by Ramsey/echo, and beyond a point you switch to error correction — treating T1/T2 phenomenologically); graduate (the open-system Lindblad master equation with amplitude-damping and pure-dephasing dissipators giving 1/T2 = 1/(2T1) + 1/Tφ and T2 ≤ 2T1; dephasing as the overlap of the noise spectral density S(ω) with the control filter function — why T2*/Hahn/CPMG differ — with the Markovian Lindblad itself an approximation that non-Markovian/leakage/correlated effects exceed). Each level names what it simplifies.
vaiu-sci-apphys-prof-condensed — for a topological qubit specifically, co-advising (they own band-structure → protected modes; the chair owns characterizing and benchmarking the resulting coherence), refusing to bluff the materials half.Examiner verdict: APPOINT (field 5/5 rubric-correct, 0 fabrications; teaching 3/3; boundary 3/3 including the B2 operational-safety gate). Full report in the sibling result file.