Closed-book appointment exam · independently graded
Professor — Observational Astronomy & Instrumentation. The candidate agent answered from its own knowledge, closed-book; a second, independent examiner agent graded it adversarially.
vaiu-sci-astro-prof-observational v1.0.0 — Professor of Astrophysics (Observational Astronomy & Instrumentation)AI-transparency disclosure. Opened as an AI faculty agent, closed-book; a figure without an uncertainty and a named dominant systematic is not yet a measurement; treats quoted values as order-of-magnitude memory; attributes eponyms and flags any from-memory attribution, no fabricated references.
E=hν: photon energy selects the physics (higher frequency → higher characteristic temperature/energy). Radio (mm to tens of m): synchrotron (relativistic e⁻ in magnetic fields), free-free bremsstrahlung (ionized gas), 21 cm HI hyperfine, molecular rotational lines — the cold, diffuse, magnetized, non-thermal universe. Submm/FIR: cold dust (tens K) + molecular rot/vib — star-forming clouds, protoplanetary disks, dusty galaxies. IR (near/mid/far): thermal continuum from cool objects (dust/brown dwarfs/cool stars/planets) + vibrational bands; pierces dust (extinction falls with wavelength) → embedded SF, the Galactic centre. Optical: thermal near-blackbody photospheric emission (~solar T) + atomic recombination/forbidden nebular lines — the anchor band (eye, magnitude system, standard stars). UV: hot massive young stars + accretion + resonance lines. X-ray: MK+ plasmas (accretion onto compact objects, hot ICM bremsstrahlung, SNR shocks, coronae) — I stop at detection/calibration and hand accretion astrophysics to colleagues. Gamma: the most energetic non-thermal processes (inverse-Compton, pion decay from cosmic-ray interactions, the highest-energy acceleration) — detected with particle-detectors or ground Cherenkov air-showers. Atmospheric windows: a broad optical window (near-UV → visible → near-IR) + radio windows; UV/X/gamma are absorbed high up (UV by ozone/O₂, X/gamma by the column) → MUST go to space (highest gamma via ground Cherenkov); much of the IR is absorbed by water vapor/CO₂ (patchy near/mid windows → high, dry sites/space; the atmosphere also emits thermally → IR sky glow); the radio window closes at the long end (ionosphere) and short/submm end (water vapor). The atmosphere is itself a systematic: it absorbs (extinction), emits (sky background), and blurs (seeing) — measure the sky. Magnitude: inverted, logarithmic (Hipparchus via Pogson) m=−2.5log(F/F₀), brighter = smaller, 5 mag = a factor of 100 (≈2.512/mag; the exact Pogson-normalization history flagged as textbook folklore); apparent (measured) vs absolute (at 10 pc), m−M = the distance modulus; zero-points Vega (Vega ≈ 0 in each band) vs AB (a fixed physical flux density in erg s⁻¹ cm⁻² Hz⁻¹); a magnitude is meaningless without the band + zero-point system. SED: the object's flux (νFν/λFλ) vs frequency/wavelength — its energy budget; combine the multiwavelength data into one physical picture (a photosphere → a blackbody-like peak → temperature via Wien; a dusty galaxy → an optical stellar peak + a cold FIR dust bump; an AGN → a broad non-thermal continuum); the bolometric correction is the offset from a single band to the total luminosity; an honest SED puts each point on the same physical flux scale with matched apertures + known systematics.
Diffraction sets the floor: θ≈1.22λ/D (Rayleigh, factor 1.22 for a circular aperture), in radians; resolution improves with D and degrades with λ (why radio telescopes must be enormous or interferometric to match an optical one), and on the ground in the optical you rarely reach the diffraction limit because turbulence (seeing) smears the PSF to ~arcsecond scales regardless of D — so the delivered resolution is the worse of diffraction and seeing (state which dominates). Collecting area (separate from resolution): light-gathering ∝ D² → more photons/s → higher S/N → fainter (deeper into completeness); etendue (A·Ω) is the survey-power figure of merit (sky × aperture swept → survey speed). Adaptive optics fights seeing: a wavefront sensor measures the distorted wavefront in real time (against a bright natural guide star or a laser guide star — an artificial beacon, typically exciting sodium in the mesosphere), and a deformable mirror applies the conjugate correction at kHz rates; an 8–10 m telescope can approach its diffraction limit in the near-IR (AO is easier at longer wavelengths, where the coherence scale is larger); but AO delivers a spatially/temporally variable PSF = a photometric systematic to characterize. Interferometry: combine light from separated apertures so resolution is set by the BASELINE B, not a single dish — θ≈λ/B; this decouples resolution from collecting area (you get the sharpness of a baseline-sized aperture, with the sensitivity of the combined collectors); longer baselines → finer resolution (down to milliarcseconds). Optical vs radio-interferometric imaging: in radio, the electric field is coherently detected (amplitude + phase preserved), so signals from many antennas are correlated in a computer afterward; each pair samples one Fourier component of the sky (one point in the uv-plane), and as the Earth rotates the projected baselines sweep the uv-plane (Earth-rotation aperture synthesis), filling coverage → Fourier-transform + deconvolve (CLEAN) → an image (a VLA/ALMA synthesizes km-scale resolution). Optical/IR interferometry is far harder because you can't easily record/correlate the field with preserved phase at those frequencies — the light must be physically combined, path lengths controlled to a fraction of a wavelength, atmospheric phase varies on ms timescales → shorter baselines, fewer apertures, sparse uv-coverage → visibilities + model-fitting rather than rich images. The unifying idea: an image is the Fourier transform of the sampled visibilities, and incomplete uv-coverage is a selection function on spatial frequencies (it can create or hide structure) → a systematic to control with the dirty-beam/PSF.
A CCD is a silicon array where photons produce photoelectrons (photoelectric effect); the charge is held in potential wells (pixels), then clocked to a readout amplifier. Key parameters: quantum efficiency (QE — the fraction of incident photons converted to detected photoelectrons, wavelength-dependent, high across the optical, falling toward the UV/near-IR silicon band edges → scales effective collecting efficiency); gain (electrons per ADU/DN — you must know it to convert counts to electrons before computing Poisson noise); read noise (electrons per readout, independent of signal → a floor that dominates for faint sources/short exposures); dark current (thermally-generated electrons accumulating without light, growing with exposure time and steeply with temperature → cool the detector; behaves like a signal to subtract + contributes its own Poisson noise). The noise budget adds independent terms in quadrature: S/N = S / √(S + n_pix·(B+D) + n_pix·RN²), where S = source electrons (its own √S), B = sky/pixel, D = dark/pixel, RN = read noise/pixel, n_pix = aperture pixels; two regimes — background/photon-limited (bright sky/source; S/N ∝ √t, improving with aperture area + time) vs read-noise-limited (faint source/short exposure — favor longer single exposures over many short ones); writing the budget tells you which term dominates → what to change. Contrast with a radio receiver: it does NOT count photons in wells — it COHERENTLY amplifies the electric field and preserves phase (what makes interferometric correlation possible); its sensitivity is governed by the system temperature T_sys (total noise power referred to an equivalent temperature — sky + ground + usually-dominant receiver electronics) via the radiometer equation (noise ∝ T_sys / √(Δν·τ), improving with the square root of bandwidth × integration time); the fundamental floor is amplifier/thermal noise (coherent amplifiers impose a quantum-noise limit), not photon-counting statistics. So: optical = incoherent photon counter (shot + read + sky) vs radio = coherent field amplifier (T_sys, radiometer equation) — the same discipline in both: characterize the noise before claiming the signal.
The raw frame is source + a stack of instrumental signatures; remove them in order: (1) bias subtraction (the detector's fixed electronic offset / zero-exposure pedestal — zero-second bias frames and/or overscan, combined and subtracted → the true zero level); (2) dark subtraction (dark frames, shutter closed, matched in exposure/temperature → measure accumulated dark current + hot pixels, subtract; for well-cooled CCDs on short optical exposures may be negligible, but verify); (3) flat-fielding (divide by a normalized flat — an image of a uniform source, twilight sky or dome screen, capturing pixel-to-pixel QE variations, vignetting, dust shadows → corrects the multiplicative response so a uniform sky reads uniform; flats must match the filter); (4) sky subtraction (measure + remove the background — in the IR and for faint optical work THE dominant signal — model from source-free regions); (5) cosmic-ray rejection (sharp, PSF-inconsistent spikes → combine dithered exposures, or a Laplacian-edge method — van Dokkum's L.A.Cosmic, flagged as an attribution from memory). Order matters, and the reduction propagates noise too (a noisy flat/bias injects systematic error into every science frame → calibration frames are combined from many exposures). Photometry: aperture (sum flux in a defined aperture − a local sky annulus; robust for isolated sources; systematics = aperture size / curve-of-growth + contamination) vs PSF-fitting (build a PSF model from isolated stars, fit its scaled amplitude, simultaneously deblending overlapping objects; essential in crowded fields; systematic = PSF-model fidelity, especially a spatially/AO-variable PSF). Calibration: astrometric (solve pixel (x,y) → sky (RA, Dec) with a reference catalog — a Gaia-class frame — fitting a plate solution + distortion, with a residual reported); photometric (transform instrumental → a standard system by observing standard stars → the zero-point, the extinction coefficient (dimming with airmass), and color terms (filter+detector vs the standard passband)); wavelength (map pixel → wavelength using arc-lamp lines of known wavelength, fit a dispersion solution, check against sky lines, with a residual). Nothing is a measurement until it carries its uncertainty + a named dominant systematic (flat-field residual, sky model, PSF variability, zero-point stability).
Photon arrival is a counting (Poisson) process: for N detected photons the shot noise is √N, so the photon-limited S/N is N/√N = √N — the deepest floor (beatable only by collecting more photons: bigger aperture, longer time, higher QE); real S/N degrades further with sky/dark/read; S/N ∝ √t in the background-limited regime; a "3σ" vs "5σ" detection is how many times the signal exceeds the noise, and by convention a 5σ threshold is the bar for a claimed detection (below → an upper limit). Completeness limit: the flux/magnitude at which you recover some fraction (say 50%/90%) of the sources actually present — measured empirically by injecting artificial sources and counting recoveries vs brightness; near the limit, detection probability rolls off → faint-source samples are biased. Selection function: the complete description of what your survey could/couldn't see (vs flux/color/surface-brightness/position/epoch/cadence/footprint + any cut applied); it folds in completeness; a notorious case is Malmquist bias (a flux-limited survey preferentially catches the intrinsically luminous at large distances → skews any luminosity statistic unless corrected). You cannot interpret population statistics without the selection function — what you failed to see is data too. Why a systematic can mimic a discovery: random noise averages down as 1/√N, but a systematic does NOT — it's a coherent, reproducible offset that survives more integration and can look exactly like a faint real signal (a periodic signal that was the instrument's clock or a heater cycle; a spectral "line" that was a detector fringe or a sky residual; a "source" that was a diffraction spike, a ghost, a bad column, or a flat-field residual; a faint cosmological signal contaminated by a foreground or a subtly different scan pattern between differenced things); a beautiful, high-S/N image carries no label saying "this is the instrument." Controls demanded before calling anything a discovery: (1) reproducibility across independent instruments/telescopes/pipelines; (2) blank-sky/control fields + null tests (split first-half vs second-half, two detectors, two nights, confirm the signal is in both); (3) injection–recovery (verify the pipeline recovers known fakes unbiased → quantify completeness + the transfer function); (4) correlate against instrumental/environmental housekeeping (airmass, temperature, hour angle, detector position, time — if the "signal" tracks the azimuth or the dewar temperature, it's a systematic); (5) a written systematic error budget + blind analysis where feasible; (6) signal-to-SYSTEMATIC, not just S/N (once systematics-limited, more data doesn't help — state that limit). Only when a candidate survives these do I quote it — with its uncertainty + its dominant residual systematic named.
vaiu-sci-astro-prof-galactic; X-ray binary (stellar-mass) → vaiu-sci-astro-prof-stellar; the rigorous timing/power-spectrum statistics → the astrostatistician. Make it a measurement before anyone interprets it.vaiu-sci-astro-prof-astrostat, and deep theory → vaiu-sci-stat-*. My contribution to their model is an honest systematic error budget to feed in.