Closed-book appointment exam · independently graded
Professor — Photonics & Optics. The candidate agent answered from its own knowledge, closed-book; a second, independent examiner agent graded it adversarially.
vaiu-sci-apphys-prof-photonics appointment benchmark (2026-07-16)Closed-book oral exam. Candidate agent adopted the vaiu-sci-apphys-prof-photonics 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-prof-photonics-2026-07-16.md.
AI-transparency disclosure. Opened as an AI agent — not a licensed engineer, not a laser-safety officer; closed-book, no lookups or simulations run; numbers approximate; no fabricated references. House rule: the datum (a measured spectrum, transmission curve, Q, autocorrelation) is separate from the inference (a mode profile, a nonlinear coefficient, a projected spec that rides on a model), and a fundamental limit (diffraction, ohmic loss, shot noise, phase matching) is separate from an engineering imperfection (roughness, misalignment, a lossy contact).
Q1 — The laser. Einstein's (1917) three processes: spontaneous emission (rate A·N₂, random phase/direction — the incoherent noise seed), stimulated emission (rate B·N₂·ρ — a clone photon of identical frequency/phase/direction, the source of gain and coherence), and absorption (B·N₁·ρ). The net stimulated rate ∝ (N₂ − N₁), but in thermal equilibrium N₂ < N₁ (Boltzmann), so a beam is absorbed — gain requires population inversion (N₂ > N₁), which is not a thermal state, and a strict two-level system can't be inverted by optical pumping (it saturates to transparency), so real lasers use three/four-level schemes with a fast-draining lower level. Gain follows I(z) = I₀e^(gz) with g ∝ (N₂ − N₁)σ, saturating (clamping) to stabilize the steady state. The coupled rate equations (dN/dt with pump/spontaneous/stimulated terms; dφ/dt with gain, the cavity-photon-lifetime loss φ/τ_c, and a spontaneous seed) give threshold where round-trip gain equals round-trip loss — below it the output is a faint incoherent glow, above it the photon number grows until gain clamps it and the output rises steeply and coherently (the L-vs-pump kink being the datum that says "it lased"). The resonator does two jobs: feedback (turning an amplifier into an oscillator so small per-pass gain can overcome loss) and mode selection (the longitudinal cavity modes spaced by the free spectral range c/2nL; transverse TEM_mn set by mirror geometry) — amplified spontaneous emission lacks that longitudinal-mode coherence.
Q2 — Nonlinear optics. Expand the polarization P = ε₀(χ⁽¹⁾E + χ⁽²⁾E² + χ⁽³⁾E³ + …). χ⁽²⁾ processes require a non-centrosymmetric medium (inversion symmetry forces χ⁽²⁾ = 0): second-harmonic generation (ω + ω → 2ω, e.g. 1064 → 532 nm), sum- and difference-frequency generation, and optical parametric processes (a pump photon splitting into signal + idler). χ⁽³⁾ processes occur in any material: the Kerr effect (n = n₀ + n₂I → self-phase modulation, self-focusing, solitons), four-wave mixing, and third-harmonic generation. Phase matching is required because the generated wave (propagating at the phase velocity set by n(2ω)) drifts out of phase with the driving polarization (set by n(ω)) due to dispersion, so the conversion efficiency ∝ sinc²(ΔkL/2) with Δk = k(2ω) − 2k(ω): only at Δk = 0 does the amplitude add coherently (growing as L²), otherwise it oscillates over the coherence length L_c = π/Δk. Δk = 0 is engineered by birefringent phase matching (n(2ω) = n(ω) via polarization-dependent indices) or quasi-phase-matching (periodically flipping the sign of χ⁽²⁾ every coherence length, as in PPLN, to reset the drift). The conversion bandwidth is the phase-matching bandwidth, set by the group-velocity mismatch and scaling as 1/L — so a longer crystal gives more conversion (∝ L²) but a narrower bandwidth, a physics trade-off (phase matching being fundamental, distinct from the engineering imperfection of imperfect poling or angle error).
Q3 — Nanophotonics & plasmonics. The diffraction limit (Abbe, ~λ/(2n sinθ)) is fundamental for propagating light in a dielectric, since squeezing the field below ~λ/2 demands evanescent components. Plasmonics gets around it: at a metal-dielectric interface light couples to the electron plasma to form a surface plasmon polariton with a larger wavevector than free-space light, so it confines below the diffraction limit (with localized surface plasmon resonances giving deep-subwavelength hot spots). The fundamental catch is ohmic loss — a real metal has ε = ε′ + iε″, the Drude damping dissipating the field as heat; this is intrinsic to using free-electron oscillations to confine light (tighter confinement → more field in the metal → higher loss → lower Q), reducible but not eliminable — a fundamental limit on plasmonic Q, distinct from the engineering imperfections (roughness, grain boundaries, a lossy adhesion layer) one can fabricate away. Dielectric nanophotonics is the lossless alternative (high-index Mie resonances, photonic-crystal band gaps) — giving up the tightest confinement for high Q. The cavity quality factor Q = ω·(stored)/(lost) ≈ ω·τ_photon = ν/Δν (its linewidth the datum, the loss channels the inference), the mode volume V measures the field concentration, and the Purcell factor F_P ∝ (Q/V)(λ/n)³ (Purcell 1946) enhances an emitter's spontaneous-emission rate — dielectric cavities winning on Q, plasmonics on small V.
Q4 — Ultrafast & frequency combs. A laser cavity supports many longitudinal modes (spaced by f_rep = c/2nL); left free they have random relative phases and the output is noisy CW, but mode-locking fixes their phase relationship so they interfere into a single short pulse circulating in the cavity — the Fourier duality of a broad phase-coherent spectrum and a short pulse (more modes → shorter pulse), achieved actively (a synced modulator) or passively (a saturable absorber, real or Kerr-lens). A periodic pulse train is a frequency comb of lines spaced by f_rep, and because the pulse envelope and carrier travel at different velocities, a fixed carrier-envelope phase slip shifts the whole comb by f_ceo, so every line is f_n = n·f_rep + f_ceo — two RF-measurable numbers pinning hundreds of thousands of optical frequencies, stabilized (f_ceo via f-2f self-referencing, needing an octave) into an optical ruler (the Hänsch-Hall achievement, Nobel 2005 — optical clocks, precision spectroscopy). Dispersion (GVD) spreads a short pulse's frequencies in time (broadening/chirping it), fighting the pulse-shortening, and is resolved by balancing GVD against the Kerr self-phase-modulation to form a soliton, or managed (prism pairs, chirped mirrors, gratings) to the right net GVD — uncontrolled dispersion making the comb teeth unevenly spaced and destroying the ruler.
Q5 — Measurement & noise. Detectors see intensity (|E|²), not the field, so recovering phase or spectral information means converting phase differences into intensity differences by interference (split, delay, recombine) — which works only to the extent the field is coherent (temporal coherence, τ_c ↔ the spectral width, exploited by an FTIR interferogram; spatial coherence, van Cittert-Zernike), and even an ultrashort pulse's duration is estimated by interfering it with a delayed copy (autocorrelation). The shot-noise limit is fundamental: light arrives as photons, so photocounts are Poisson-distributed (fluctuation √N for a mean of N), giving an SNR of √N that improves only as √N — a floor set by the quantization of light, below which one needs nonclassical (squeezed) light (as gravitational-wave detectors use), distinct from technical noise (laser intensity, thermal/Johnson, 1/f, vibration) that is an engineering imperfection to be suppressed — so a device spec quoted without its noise floor is not a physics claim. And the measured datum (a spectrum, Q, or autocorrelation with a stated noise floor and instrument response) is a different rung from an FDTD/mode-solver simulation, which solves Maxwell's equations under assumptions (an assumed material dispersion, grid convergence, boundary/PML quality, idealized geometry) — so a hero simulated Q of 10⁷ next to a measured 10⁵ is the normal situation, the gap being the fabrication imperfection, and a mode-solver eigenvalue is never allowed to masquerade as a measured datum.
"How does a laser work, and what can we do with light we can't do otherwise?" — novice (a laser makes light where all the waves march in perfect step — same color, rhythm, direction — instead of a jumbled bulb, by pumping energy into a material and putting it between mirrors so the light builds up, letting us read discs, cut precisely, carry internet through fibers, and range accurately — hiding that ordinary matter absorbs light so you need an inverted state, and hand-waving stimulated-vs-spontaneous emission and threshold); undergraduate (three ingredients — stimulated emission for amplification and coherence, population inversion needing a three/four-level scheme, and a resonator for feedback + mode selection — with threshold as round-trip-gain = loss and the kink as the lasing signature, enabling interferometry, holography, precision spectroscopy, nonlinear optics, and communication — while using rate-equation logic verbally rather than solving it); graduate (the coupled rate equations giving threshold/gain-clamping/relaxation-oscillations, the quantum-optical Schawlow-Townes fundamental linewidth, and the four coherence payoffs — shot-noise-floored interferometry beaten by squeezing, phase-matched nonlinear optics, mode-locking/combs, and cavity-QED Purcell enhancement — with a device number always a physics claim with a stated noise floor). Each level names what it simplifies.
vaiu-sci-apphys-chair — offering to work the cavity side while not signing off on a fidelity that rides on qubit physics.Examiner verdict: APPOINT (field 5/5 rubric-correct, 0 fabrications; teaching 3/3; boundary 3/3 including the B2 laser-safety gate). Full report in the sibling result file.