Figure 1 begins with the missing measurements
In Paper IV Figure 1, read the spatial-frequency axes before looking at the image later in the paper. A plotted position records a projected baseline in wavelengths. Distance from the center corresponds to angular fineness, and direction selects which part of the source shape is tested. Identify gaps between tracks as carefully as the tracks themselves.
A filled Earth-sized aperture would sample a much more continuous range of separations. EHT uses a sparse set of stations, so its extreme longest-baseline resolution comes with substantial gaps. A narrow beam core describes one useful scale; it does not say that every structure on that scale is equally well constrained.
Now compare observing days and source models using the same coverage. A change in reconstructed shape could arise because the sky changed, because the array sampled different directions, or because a method responded differently to the gaps. The figure tells you which of those possibilities must be checked later.
Reading: EHT Collaboration, First M87 Results IV, arXiv:1906.11241v1 · NRAO Essential Radio Astronomy, Chapter 3, aperture synthesis
Figure 3 tests phases that stations cannot freely change
Figure 3 plots closure phase against Greenwich mean sidereal time for selected triangles. Follow one triangle’s label through the panels. A closure point combines three simultaneous directed baselines; it is not the phase of one station. Near a visibility null, phase becomes noise-sensitive, so a large angle alone needs its amplitude and uncertainty context.
Use equation 7 and section 2.1.2 to identify the station-gain model. Station phase factors cancel from the triangular product. Gain amplitudes remain in visibility amplitudes; closure amplitudes use a different multi-baseline combination to cancel station amplitude factors. Separate those two calibration problems when reading a data term.
The tutorial’s three-row CSV uses decimal UTC hours. Figure 3 uses GMST, a rotation-based time coordinate. A direct overlay requires the correct time conversion, observing day, triangle ordering and averaging. The tutorial computes one measured triangle in its native timestamp; matching the figure would additionally require that conversion and averaging.
Reading: EHT Collaboration, First M87 Results IV, arXiv:1906.11241v1 · EHT 2019-D01-01 M87 data release, pinned commit 27ae76a
Figure 5 gives the algorithms a known sky
Figure 5 uses synthetic source models and the visibility signals they would produce. Start with the known input image, then trace the sampling and reconstruction. Here the truth is available because the authors constructed it. This test asks whether an algorithm can recover selected shapes through the same kind of incomplete observation.
Compare features at matched angular resolution. A sharp reconstruction can look impressive simply because it contains unsupported fine detail; a blurred reference can reveal which larger structures were recovered reliably. Also compare the predicted measurements. A close image match on one training source is useful, but it does not test every source the real sky might contain.
The actual M87 measurements enter a separate stage of the analysis. Keeping the synthetic tests distinct lets the authors tune and evaluate methods without deciding in advance that one reconstructed M87 image is the truth.
Reading: EHT Collaboration, First M87 Results IV, arXiv:1906.11241v1
Equation 14 and Figure 7 expose the image preferences
Equation 14 combines data-agreement terms with weighted image regularizers. A weight, often called a hyperparameter, sets how strongly the program prefers a property such as smoothness. Appendix A gives the mathematical definitions. Read a regularizer’s sign and normalization before comparing weights: a number has no transferable meaning without those conventions.
In Figure 7, follow how the reconstructed image changes across the MEM and TV parameter survey. MEM denotes an entropy-based image preference; TV penalizes total variation, a measure of local image changes. Locate settings that fit the data acceptably, then ask which image features survive changing those settings.
These survey images are reconstructions under different choices, not random samples from a posterior distribution. Their visual spread is therefore not automatically a calibrated uncertainty band. The paper combines parameter surveys with synthetic tests and comparisons between independent pipelines. Section 6.2.2 is useful for following how those choices were assessed.
Reading: EHT Collaboration, First M87 Results IV, arXiv:1906.11241v1
Earth rotation can confuse time and space
Move to sections 2.3–2.4 of Dynamical Imaging. A baseline’s changing visibility can mean it reached a new position, or that the source changed while it moved. Static imaging assigns both effects to one fixed image. If source variation is rapid enough, that assumption can create misleading spatial structure.
The paper explains ways to identify variability through nearby spatial-frequency measurements and then reconstruct a sequence. Before studying an attractive movie, ask how much direct data each frame has. When a frame has few or no measurements, assumptions connecting it to neighboring frames become especially influential.
Reading: Johnson et al., Dynamical Imaging with Interferometry, arXiv:1711.01286v1
Equation 3 states how neighboring frames are connected
The dynamic objective adds a temporal regularizer to the data terms and ordinary image regularization. The schematic form below separates those roles. The paper’s own equation 3 fixes its signs and detailed normalization; sections 3.1–3.3 define different temporal choices.
One choice favors continuity between successive frames. Another favors small departures from a time-averaged image. A flow-based choice favors approximately steady transport of brightness. They make different predictions where data are absent. Sending the temporal weight toward zero leaves frames more independent; making a static-favoring weight extremely large suppresses genuine changes.
Reading: Johnson et al., Dynamical Imaging with Interferometry, arXiv:1711.01286v1
Figures 7–12 need a provenance check
Figures 7–10 in the dynamic paper use simulated black-hole-system observations, including reconstructions based on visibility amplitudes and closure information. The known simulated frames allow a direct recovery test. Their apparent motion is part of the synthetic experiment; it is not a measured 2017 EHT movie.
Figure 12 instead uses actual 7 mm VLBA observations of the M87 jet across epochs. Check the wavelength, array and time intervals in the caption. Those observations test the method on a different source scale and sampling pattern from the 1.3 mm EHT excerpt in our tutorial.
For the next imaging paper, write down five things before judging its image: the measured quantities, the sampled spatial frequencies, the image and time assumptions, the known-truth tests, and the feature that survives changing the assumptions. Those notes turn a striking picture into a scientific claim you can examine.
Reading: Johnson et al., Dynamical Imaging with Interferometry, arXiv:1711.01286v1 · EHT Collaboration, First M87 Results IV, arXiv:1906.11241v1