A blurred star does not name the fault
You point a telescope at a star and turn the focuser, but see a moving patch instead of a point. Tighten a screw? Shorten the exposure? Wait for cooling? Each action sometimes helps, but addresses a different cause. Without a test that separates causes, you can change three things together and never discover what was wrong.
Telescope diagnosis follows “symptom → separating test → permitted action → repeat the same test”. A spot shape suggests a hypothesis; a predicted change in the result supports it. We will first separate focus and geometry from air, motion, the detector and processing. Radio telescopes need a different chain: antenna, cable, receiver, software stream and interference.
Examples draw on the course refractor, Newtonian on a Dobsonian mount and Radio JOVE. This is not a report of hardware we assembled: documented cases, our numerical models and proposed future tests are identified separately. Visible branches add mathematics; the main route is readable without them.
A precise instrument can still test the wrong thing
NASA’s 1990 report traced Hubble’s error to a 1.3 mm lens-spacing mistake in the test corrector. Conflicting auxiliary results were dismissed. The mirror followed a faulty reference. NASA: Hubble optical systems failure report.
The scale can also be domestic. In Stellafane’s documented build, the calculated cell position was 2⅜″, but 3¾″ was needed: 34.925 mm apart. The author’s best explanation was a focal-length measurement error. This is his experience, not our measurement. Stellafane: installing the primary mirror.
The shared lesson is to check both the part and the condition used to accept it. Repeating the same calculation with the same wrong parameter is insufficient. For a home-built tube, a focus trial before final drilling checks something different from a tidy drawing. A numerical diagram benefits from an independent calculation that does not repeat the same algorithm.
Record a baseline before turning anything
Save the instrument type, eyepiece or camera, additional lenses, focuser position, target, time and conditions. Images need exposure, gain, filter and raw files; radio needs frequencies, gains and timestamps. “It looked bad” cannot reproduce a check. “Field centre, same star, 5 s and 30 s, unchanged focuser position” is more useful.
First remove an obvious obstruction: a cap, an incorrectly seated component or a missing software stream. Then choose one branch below. Change one factor at a time and record your prediction before repeating. If a shorter frame should remove drift while retaining an optical asymmetry, compare precisely that. Do not reduce exposure, gain and display scale together.
Choose the separating test first
- No sharp image anywhere in the travel→ Change the distance to a non-solar target
- Centre and edge look different→ Compare centre, edges and both sides of focus
- Shape changes rapidly→ Repeat after cooling, away from warm surfaces
- The star shakes or leaves a trail→ Compare a short and a long exposure
- A patch or clipped peak appears in the image→ Compare a shorter raw exposure and calibration
- The radio spectrum is silent or full of bands→ Compare antenna connected/disconnected
A focus can exist beyond the mechanism’s reach
Sharpness improving right up to a travel stop, with no travel remaining, differs from a smallest spot located within the travel that is still too large. Check eyepiece seating, documented spacing and added adapters. Do not force the mechanism past its stop. On a home-built tube, measure the available optical-element positions before final cutting or drilling.
For our course refractor we specified an objective with f=500 mm, a positive +50 mm eyepiece and a hypothetical separation range of 540–590 mm. A very distant target needs 550 mm in the model; at 10 m, 576.316 mm; at 5 m, 605.556 mm. The last is unreachable in this example, although the same tube can focus a distant target. This is a calculated scenario, not the measured travel of an assembled instrument.
The separating test changes the distance to a non-solar target while keeping the optics unchanged. If a distant object focuses within the travel but a nearby one requires going beyond it, do not immediately blame the lens. Change geometry only within the design’s limits and repeat with the required eyepieces. Do not use the Sun for this test; ordinary eclipse glasses are not telescope filters. OpenStax: thin lenses, NASA: observing safely.
Branch · With equationsOptional: geometry explains unreachable focus without assuming a defective lens.
Why a nearby target moves the image farther back
For a thin converging lens, a real target at distance u > f and an image at distance v:
(1)/(f) = (1)/(u) + (1)/(v), v = (fu)/(u − f),; Δv = v − f = ((f)^(2))/(u − f).
As u→∞, we obtain v→f. For f = 500 mm and u = 10000 mm:
v = (10000)/(19) mm ≈ 526.316 mm,; d = v + 50 mm ≈ 576.316 mm.
Adding 50 mm here describes the course refractor’s particular positive eyepiece in the relaxed-eye model. It is not a universal camera or compound-telescope back focus. Real thick lenses require distances measured from principal planes; the cell’s mechanical length need not equal the optical separation.
Collimation aligns axes; it does not repair every glass surface
In a Newtonian, the secondary’s position, its aim and the primary’s aim are related but distinct adjustments. Distinguish a component’s edge from its reflection. After an allowed change to one component, check the others again. Requiring every visible shadow to be concentric can be wrong: fast Newtonians include secondary offset. Stellafane: collimation sequence, Sky-Watcher Quattro: design-specific documentation.
A star check needs thermally settled optics, sufficiently steady air and a star at the field centre. Re-centre the star after correcting alignment. If its shape changes only as it moves to the edge, you are testing field behaviour rather than the same on-axis state. Do not prescribe a universal cooling time: mass, construction and changing temperature all matter.
A closed compound system has its own adjustment limits. For example, Celestron describes user adjustment of secondary tilt for its SCT/EdgeHD systems, not rotation of the Schmidt corrector. This does not authorise dismantling another telescope because a picture looks similar. If the permitted procedure is unclear, document the symptom and consult the manual or service. Re-centring and the same reference star should establish whether the on-axis result improved. Celestron: SCT/EdgeHD collimation.
Compare the centre, edge and both sides of focus
Coma: stars near the edge resemble asymmetric fans, improving towards the axis. In an ordinary parabolic Newtonian, edge coma can remain an inherent property after collimation. The control uses the same star at the centre and several edge positions. If the centre is good, do not spoil it by trying to remove all edge coma with the primary screws. Check the design’s usable field and compatibility of any corrector already in use. Sky-Watcher: star testing.
Astigmatism: in a simple model an elongated spot rotates to a perpendicular direction across best focus. Test with small symmetric focus changes while retaining the target and field centre. This indicates different focuses in two sections, but does not identify the faulty component: the cell, optics, an accessory or field aberration can be involved. If the design permits, try another sound eyepiece in a separate repeat; do not rotate closed optical elements at random. Celestron: types of aberration.
The panels below are not a photographic catalogue for automatic diagnosis. They are our calculation: change only wave phase and examine the redistribution of light. Several causes can overlap in reality. The ring in the unobstructed model also explains why a “doughnut” alone does not prove that a secondary mirror exists or is misplaced.
Synthetic stars: phase redistributes light
Ideal pupil

Defocus · a=+0.4

Astigmatism · a=−0.3, b=+0.6

Astigmatism · a=+0.3, b=+0.6

Coma · c=+0.5

Trefoil · d=+0.5

Branch · Wave opticsOptional: explains the six synthetic panels and limits their diagnostic use.
From phase error to a spot: a Fraunhofer model
Normalise circular-pupil coordinates to radius 1, (r)^(2) = (x)^(2) + (y)^(2). Amplitude P is 1 inside and 0 outside. Specify optical path difference w in waves with these unnormalised polynomials:
\begin{gathered} w=a(2r^2-1)+b(x^2-y^2)\\ {}+c(3r^2-2)x+d(x^3-3xy^2),\\ I\propto\left|\mathcal F\{P\exp(2\pi i w)\}\right|^2. \end{gathered}
Here a sets defocus, b an astigmatic term, c coma and d trefoil. These are coefficients of precisely the written polynomials, not RMS-normalised Zernike coefficients. Our FFT uses a 256×256 grid with a pupil radius of 32 samples. Normalisation makes full-grid intensity sum to 1; the displayed crop is not renormalised. All six maps share a nonlinear brightness mapping, so loss of peak intensity remains visible. The physical basis is described in the POPPY authors’ documentation; POPPY itself was not run here.
We checked energy conservation, independent direct wave sums at selected points, the displacement direction from phase tilt, rotation of the astigmatic spot and the ideal pupil’s central profile. Direct Python calculations at 24 points differ from the FFT by no more than about 5.73 × (10)^( − 17) in the chosen normalisation. This checks implementation, not the accuracy of a real-star diagnosis.
For a circular unobstructed pupil the first minimum lies at about 1.22λ/D. With λ = 550 nm and D = 150 mm, that is 0.923″. Do not equate this distance with FWHM, Gaussian σ or guaranteed resolution under the atmosphere. OpenStax: circular apertures and Rayleigh’s criterion.
A cell should retain the mirror without reshaping it
A pinched mirror can produce a persistent non-circular pattern. Three lobes do not, however, prove that three clips are responsible: the previous figure introduced threefold phase symmetry, not glass elasticity or clip forces. The mechanical assembly needs an independent check.
For an accessible home-built Newtonian cell, compare retention with the course’s Dobsonian design and the cell’s own instructions. A retainer prevents the mirror falling out; it should not force the optical surface to bend. Side loads also matter. Change retention only with the mirror safely supported and without risking its release; there is no universal gap or screw-turn count here. Stellafane: installing a mirror in its cell.
If the documentation permits removing excessive pressure, repeat collimation and the star check afterwards under the same conditions. Disappearance of the corresponding persistent pattern supports the stress hypothesis. No change calls for another test, not greater tightening. Do not dismantle a closed compound system for this experiment.
Air and a cold surface produce different symptoms
If the shape flickers and changes rapidly, making a stable best focus hard to choose, suspect a moving optical path through air. Warm optics and tube currents add local turbulence; the atmosphere above the site adds another contribution. Repeat after the instrument approaches ambient temperature, away from heated roofs or asphalt and at a comparable star altitude. Improvement after cooling supports a local thermal cause but does not establish perfect optics. ESO: slow deformations and atmospheric turbulence.
Dew more often gradually reduces contrast, increases scattering and may be visible on an accessible outer surface. Radiative cooling can lower a surface below air temperature, so an air thermometer alone is insufficient. Inspect without touching optics, then compare before/after drying or using dew protection allowed by the manufacturer. Do not wipe coatings or heat them arbitrarily. Restored clarity after condensation disappears distinguishes this case from a collimation error. ESO: surface temperature and dew point.
If conditions do not separate atmospheric and tube effects, the honest result is “the test did not distinguish the causes”. Do not compensate a momentary atmospheric distortion with a permanent screw adjustment. A star test without stated conditions can worsen a sound instrument.
Vibration, drift and field rotation need different tests
Vibration often follows touching the focuser, a gust or a moving cable: the star shakes and then settles. Compare a frame after a hands-off pause with one immediately after adjustment. Check support fastening, balance and cable clearance according to the manual; after changing one thing, repeat the same sequence. Tightening every component at random does not measure stiffness.
Tracking error builds a trail during exposure. A shorter frame in the same configuration helps separate motion from a persistent optical spot. Similar parallel trails across the field suggest shared translation but do not immediately name a cause: rate, polar alignment, periodic error or flexure can contribute. Choose a relevant check from mounts and tracking, change one parameter and repeat the pair of exposure times.
Field rotation can leave a reference star nearly fixed while others turn around it. Check several stars across the field, not just the central one. Alt-az tracking alone does not fix the whole field’s orientation; the mounting or derotation configuration must address that. Correcting a constant rate on one axis does not remove rotation. MMTO: field and derotator geometry.
Longer exposures reveal more drift
| Exposure · s | Trail · arcsec | Trail · pixel | σx · arcsec |
|---|---|---|---|
| 5 | 0.5 | 0.483524 | 1.010363 |
| 30 | 3.0 | 2.901146 | 1.322876 |
Branch · Numbers and uncertaintyOptional: checks scale, units and an explicitly supplied synthetic result, not a real exposure.
Does a supplied trail agree with the tracking model?
For small angles, pixel size p and focal length f, the scale s in arcsec/pixel is:
s = 206264.806(p)/(f),; p = 3.76 μm, f = 0.75 m,; s ≈ 1.034074 arcsec/pixel.
Our model specifies uniform motion v = 0.1 arcsec/s and a base Gaussian with σ = 1 arcsec. At T = 30 s:
L = vT = 3 arcsec ≈ 2.901146 pixel,; (σ_(x))^(2) = (σ)^(2) + ((vT)^(2))/(12),; σ_(x) ≈ 1.322876 arcsec.
The term containing 12 is the variance of a uniform position along a trail centred on mid-exposure. The profile is a convolution of the Gaussian with that motion. Transverse width remains unchanged; total light is conserved. The long profile is no longer Gaussian, so its second moment cannot automatically be converted to Gaussian FWHM.
Now supply a hypothetical estimated length: ℓ = 3.0 ± 0.2 pixel over the same 30 s. This is not our measurement. Treating s and T as exact only in this example:
\begin{gathered} \widehat v=\frac{\ell s}{T}\approx0.1034\,\mathrm{arcsec/s},\\ \delta v=\frac{s\,\delta\ell}{T}\approx0.0069\,\mathrm{arcsec/s}. \end{gathered}
The model’s 0.1 arcsec/s falls within 0.0965–0.1103 arcsec/s: the supplied result agrees within this uncertainty. That does not establish a correct drive. Real trail estimation must account for PSF, noise, the algorithm and scale/time uncertainties; do not arbitrarily subtract the short frame’s FWHM from the long frame’s.
Check the raw frame before repairing optics
Saturation makes a star’s peak unsuitable for ordinary linear comparison; a clipped profile or charge spilling can appear, depending on the detector. Test with a shorter exposure of the same field, unchanged optics, raw-value inspection and documented linearity limits/quality flags. If an unsaturated peak returns, acquisition settings caused the problem; the mirror was not necessarily involved. Choose a usable mode and repeat. A maximum code of 65535 is not a universal threshold: charge capacity, analogue electronics and the ADC can impose different limits. Hamamatsu: CCD technical note, §2-5 and §2-10.
An unsuitable flat can create patches or an incorrect gradient after calibration. Compare the same raw frame with and without this particular flat; keep the stretch fixed. Check filter, camera and dust positions, illumination geometry, linearity and masks. A defect appearing only after division supports a correction problem but does not prove the raw data were flawless. Correct the flat selection/preparation and reprocess the same file. Astropy: what a flat corrects.
A defective pixel, dust, sky gradient or processing artifact can also look like “bad optics”. Check coordinates and the stage of appearance: does the feature stay at the same pixel, move with the sky or appear only after an operation? From measurement to image provides a real calibration example. A finished JPEG’s colour and sharpness do not replace raw data for this conclusion.
Radio JOVE first needs a working signal path
An optical tube tests focus through an image; Radio JOVE tests signal passage through specific links. Our documented Radio JOVE route uses an RSP1B, 16–24 MHz, antenna and coaxial path, USB and a software bridge to Radio-Sky Spectrograph. If the computer detects the receiver but the spectrogram does not update, checking antenna alignment cannot restore an interrupted software stream. NASA Radio JOVE: receiver manual v1.0, March 2025.
Compare the antenna disconnected and connected with identical frequencies, gains and scale. A clear extra background supports the conclusion that external signal reaches the chain. No change calls for separate checks of connections, cable, input and mode, not an immediate verdict of a broken antenna. This test neither calibrates antenna temperature nor identifies an astronomical source. NASA Radio JOVE: basic testing.
Extra bands may be RFI, radio-frequency interference, or overload products. Record settings; separately test lower gain and safely switching off a suspected local device. Radio JOVE recommends disabling AGC, noise blanker and noise reduction for astronomy because they alter the incoming signal. A gain change needs renewed calibration control. A change matching a local source’s switching supports RFI; resemblance to an example alone does not establish a solar or Jovian burst. NASA Radio JOVE: settings and limits.
Radio JOVE: check the signal chain
| Separating test | What the result may mean | Action and repeat |
|---|---|---|
| Computer sees receiver, but RSS draws nothing | The software bridge may be interrupted | Check selected device, stream and SDRc2RSS; save and reopen a short file |
| Antenna off/on, unchanged gain | A background change shows an external contribution | If unchanged, separately check cable, connectors and input; repeat |
| Lower gain; AGC/NB/NR disabled | Disappearing extra bands suggest overload | Retain the new settings; repeat the control and calibration |
| Safely switch a local device off/on | A synchronous change suggests RFI | Repeat the switch; compare an independent station and timestamps |
Acceptance ends with a repeatable result
After correcting a problem, repeat the original test rather than a more convenient one. If you changed focus geometry, return to the required target and eyepiece; if camera settings, retain unsaturated raw frames; if the software stream, record a short radio file and reopen it. A beautiful image alone does not accept every property of the instrument.
Name the property you are accepting beforehand: focus reach, retention and axial geometry, or signal recording. Record configuration, prediction, obtained value with unit/uncertainty, conditions and remaining doubts. The table proposes future physical checks. No physical tests were performed for this article.
Acceptance checks for three course builds
| Route | Check | Retain |
|---|---|---|
| Course refractor | Distant non-solar target; focus away from a travel stop; tubes not tilted | Target distance, optical-element separation and available travel |
| Course Newtonian / Dobsonian | Secure mirror retention without clamping; collimation; field centre after cooling | Component positions, star-test conditions, repeat with the same eyepiece |
| Course Radio JOVE | Cable/receiver/software stream; antenna off/on; RFI check | 16–24 MHz, gain settings, time, file and a successful reopening |
Branch · Experimental reasoningOptional: helps distinguish repeating an error from testing its cause.
What makes a check independent?
Two matching results are useful, but their agreement may arise from a shared wrong parameter. Calculating focus on two calculators with the same wrong f checks arithmetic, not f itself. In the diffraction model, direct wave summation checks the FFT with another algorithm; energy conservation adds an independent physical constraint. Even together they do not test your mirror’s actual shape.
For a diagnosis, formulate competing predictions. If drift is responsible, a shorter exposure should reduce the trail; a persistent phase error will not vanish merely because exposure is shortened. If an unsuitable flat is responsible, the feature changes when that flat is applied to the same raw data. When predictions do not differ or conditions are not retained, another test is needed. “Not distinguished yet” is more accurate than confident, arbitrary repair.
Try explaining three results
1. A refractor focuses on a distant tower but not a target 5 m away, with sharpness improving right to the travel limit. What should you check? Answer: the geometry available for the nearer target; our example needs 605.556 mm instead of the available 540–590 mm. This does not prove an objective defect.
2. The central star stays fixed in a long exposure while edge stars move along arcs. Is correcting a constant rate on one axis enough? Answer: no; check field rotation with several stars and the mount/derotation mode. One reference point does not set the whole field’s orientation.
3. Radio JOVE’s background is higher with the antenna than without it. Has Jupiter been found? Answer: no; only an external-path contribution is supported. Interference checks, time, observing conditions and independent comparison are needed. Each useful conclusion is narrower than the initial temptation.
Next, choose one control test
Choose the refractor, Newtonian or Radio JOVE route and write a future test: symptom, two possible causes, the factor you will change and the different expected results. Before handling hardware, check the actual design and permitted adjustments. For a camera, add a control from the detector lesson and compare raw/calibrated data using the lesson on turning measurements into images.
The course began with how people learned to collect and direct light. That structure now has a way to check its result: what was measured, which model explains it and which alternative remains. A retained file, a value with units and a reproducible test let the next session start from evidence instead of another random screw turn.