Closed-book appointment exam · independently graded
Professor — Biophysics. The candidate agent answered from its own knowledge, closed-book; a second, independent examiner agent graded it adversarially.
vaiu-sci-apphys-prof-biophysics appointment benchmark (2026-07-16)Closed-book oral exam. Candidate agent adopted the vaiu-sci-apphys-prof-biophysics 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-biophysics-2026-07-16.md.
AI-transparency disclosure. Opened as an AI agent (an LLM); closed-book, no lookups; eponyms flagged where confident, folklore marked; no fabricated references; stated numbers are order-of-magnitude teaching values, not lab-calibrated measurements. Governing habit: energy/length/time scales first, then biology; the curve/trajectory is the datum and the landscape/rate/mechanism is an interpretation that lives or dies by its calibration, assumed potential, and labeling; thermal noise (kBT) is the physics; a living system is driven and dissipative; a spherical cow carries its regime of validity, never mistaken for the cell.
Q1 — Life at low Reynolds number. The Reynolds number Re ~ ρvL/η is inertia over viscosity; for a bacterium (L ~ 1 μm, v ~ 30 μm/s in water) Re ~ 10⁻⁴–10⁻⁵ — the Stokes regime, viscosity-dominated and overdamped. There is no coasting (the stopping distance is a fraction of an ångström — turn the flagellum off and it halts on any biological timescale), so the world is pure force-balance. Because Stokes flow is linear and time-reversible, a reciprocal stroke (one degree of freedom retracing its path, like a scallop opening and closing) produces zero net displacement — Purcell's scallop theorem — so swimming requires a non-reciprocal stroke (a rotating helical flagellum, or a cilium bending differently on power and recovery). Mixing is by diffusion, not stirring (low Péclet number, no turbulence). And kBT is the ruler because at the mesoscale thermal energy is comparable to the structures' energies: kBT ≈ 4.1 pN·nm ≈ 25 meV ≈ 0.6 kcal/mol at 300 K, so kBT/(1 nm) ≈ 4 pN is exactly the scale of molecular-motor forces and single-molecule tweezers — structures held together by tens of kBT are marginally stable and fluctuate, exploring a Boltzmann-weighted landscape rather than running as a deterministic machine.
Q2 — Polymer physics of biopolymers. The datum is a force-extension curve (force f vs end-to-end extension x); the interpretation is a chain model. The freely-jointed chain (N Kuhn segments of length b, uncorrelated, ⟨R²⟩ = Nb², following the Langevin function under force) is a purely entropic random walk — good for very flexible chains. The worm-like chain (Kratky–Porod) is a continuous elastic rod with bending energy, whose tangent correlations decay as ⟨t(0)·t(s)⟩ = e^(−s/ℓp) with the persistence length ℓp = κ/kBT (the length over which the chain forgets its direction) — dsDNA ℓp ≈ 50 nm (~150 bp), ssDNA ~1 nm, actin ~10–17 μm, with the Marko–Siggia interpolation f ℓp/kBT ≈ x/L − 1/4 + 1/[4(1−x/L)²] (an approximation) whose divergence as x → L is the semiflexible signature. Entropic elasticity is the core: a stretched polymer resists mainly because stretching reduces the number of accessible configurations (lowers entropy, so F = U − TS rises through −TS), the telltale being that the spring gets stiffer as T rises. A dsDNA pull shows three regimes — the entropic regime (sub-pN to a few pN, where the WLC fit returns ℓp and the contour length), enthalpic stretching (~tens of pN), and the ~65 pN overstretching plateau (a B-to-S/melting transition, mechanism model-dependent) — with the caveats that trap/bead calibration sets the force axis, the WLC knows nothing of base sequence, and salt changes ℓp.
Q3 — Single-molecule methods. The recurring discipline: the trajectory is the datum, the landscape/rate/mechanism an inference conditioned on calibration + assumed potential + labeling. Optical tweezers trap a bead as a harmonic spring (stiffness k calibrated by equipartition ½k⟨Δx²⟩ = ½kBT or the power-spectrum corner frequency), reading force-extension curves and folding/motor-stepping trajectories — miscalibrate k and every reported energy is wrong. Magnetic tweezers are a natural constant-force instrument uniquely able to apply torque/twist (DNA supercoiling), calibrated from transverse Brownian fluctuations. AFM (cantilever deflection → force) reaches higher forces for mechanical protein unfolding (the sawtooth), where each peak's WLC fit gives an unfolded domain's added contour length and the apparent unfolding force is loading-rate-dependent (Bell/Evans dynamic force spectroscopy → the distance to the transition state and the intrinsic off-rate — so "unfolding force" is mechanism, not a fixed property). Single-molecule FRET (E = 1/[1 + (r/R₀)⁶], Förster radius R₀ ~ 3–8 nm) is a molecular ruler in the ~2–8 nm window, but converting E to an absolute distance hides the orientation factor κ² (assumed 2/3, valid only if the dyes rotate freely), the quantum yields, and the assumption the labels don't perturb — and dye blinking/bleaching can masquerade as conformational dynamics, so smFRET robustly reports relative changes and state connectivity. Across all of them: a single trajectory is a hint (not a law) until reproduced, the free-energy landscape is reconstructed (never measured) under Kramers/Bell assumptions and a presumed reaction coordinate, and pulling is a nonequilibrium perturbation — motivating the fluctuation theorems.
Q4 — Molecular motors & nonequilibrium. Kinesin walks a microtubule toward the plus end in 8 nm steps (the tubulin-dimer spacing set by the track — a clean datum), tightly coupled at ~1 ATP/step, hand-over-hand (a 16 nm periodicity when one head is labeled), stalling at ~5–7 pN — with the thermodynamic ceiling ~(20 kBT ≈ 80 pN·nm)/8 nm ~ 10 pN again on the pN mesoscale set by kBT. Myosin II is the non-processive contrast (small duty ratio, ~5 nm working stroke), myosin V processive (~36 nm). The power-stroke (a deterministic conformational lever swing) and thermal-ratchet (rectifying unbiased diffusion via a cyclically-fueled asymmetric potential) pictures are a false dichotomy at the extremes — real motors sit on a continuum with a chemically-gated bias plus thermal barrier crossing (every step a Kramers crossing) — and neither works without an energy input breaking detailed balance. Life is driven and dissipative: a motor at equilibrium would step forward and backward equally and do zero net work (no directed work from a single heat bath — no free Maxwell demon), so directed motion requires continuous dissipation, with metabolism holding ATP far from equilibrium (ΔμATP ~ 20 kBT) to bias the cycle. The fluctuation theorems turn the noise into a tool: Jarzynski (⟨e^(−W/kBT)⟩ = e^(−ΔF/kBT)) recovers the equilibrium free-energy difference from fast, irreversible pulls (with ⟨W⟩ ≥ ΔF just Jensen's inequality), and Crooks (P_F(+W)/P_R(−W) = e^[(W−ΔF)/kBT], the forward and reverse work distributions crossing at W = ΔF) is the finer statement — used to extract folding/binding free energies from single-molecule pulling, with the caveat that the exponential average is dominated by rare low-work trajectories and converges slowly.
Q5 — Physical limits to sensing. The diffusion-limited association rate to an absorbing sphere is Smoluchowski's k = 4πDa — scaling with the radius a, not the surface area (a 10× smaller target loses only a factor of 10), so a cell needs only a sparse patchwork of small absorbers to capture nearly as well (diffusion explores and re-explores) — setting the ~10⁸–10⁹ M⁻¹s⁻¹ ceiling on cellular kinetics and the physics of molecular search. The Berg–Purcell limit bounds how precisely a cell can measure a concentration c by counting diffusive arrivals at a detector of size a: (δc/c)² ~ 1/(Dacτ), so precision improves only as √τ (slow) and with larger a, faster D, higher c — the deep subtlety being that the number of independent measurements is set by the a²/D refresh time, since re-counting the same molecule buys nothing. This is a fundamental bound no downstream signaling can beat at equilibrium — but cells spend energy to do better (kinetic proofreading, a driven ATP/GTP-consuming scheme, sharpens specificity beyond the equilibrium limit; adaptation via dissipative feedback), so "is the process at equilibrium or driven?" changes the bound itself. And every element ("absorbing sphere," "Poisson arrivals," "concentration c") is a spherical cow — magnificent as a bound and scaling law, but the prefactors and independence assumption are where the model meets the crowded, hydrodynamically-coupled, receptor-clustering reality it only approximates.
"How does physics constrain what a living cell can do?" — novice (imagine being a swimmer so small the water feels as thick as honey, so you stop dead the moment you stop paddling, and you're constantly jostled into trembling by the water molecules — so a cell can't coast, can't stir its soup, and must work with the jiggling — hiding the numbers behind the "honey"/"trembling" metaphors); undergraduate (two dimensionless numbers — Re ~ 10⁻⁴ making motion overdamped and reciprocal strokes useless, and the energy-to-kBT ratio making structures fluctuate and setting motor forces at ~pN — give the constraints: no inertial swimming, diffusion-limited kinetics k = 4πDa, the Berg–Purcell sensing floor, and motors that must burn ATP because there's no directed work from one heat bath — using scaling laws without deriving them); graduate (reframing the cell as a driven, dissipative, mesoscopic system where the constraints become theorems: overdamped Langevin dynamics, detailed balance broken by clamping ΔμATP ~ 20 kBT, the thermodynamic uncertainty relation pricing precision in entropy production, kinetic proofreading buying specificity at a dissipation cost, and Crooks/Jarzynski recovering the equilibrium landscapes — while flagging that the TUR/fluctuation theorems are stated not derived and that non-Markovian crowding and active cytoskeletal noise are ignored). Each level names what it simplifies.
vaiu-sci-stat-*), proposing co-supervision (he defines the physical model and calibration; they make the inference rigorous).Examiner verdict: APPOINT (field 5/5 rubric-correct, 0 fabrications; teaching 3/3; boundary 3/3 including the B2 clinical-safety gate). Full report in the sibling result file.