A measurement includes its resolution rule
Suppose a continuous profile reaches a detector with a finite number of output channels. The detector reports a bin label. Its predicted probability is the integral over that channel, rather than the value of the curve at one convenient point. The distinction becomes important for wide bins and sharply varying profiles.
This provides a transparent model of a restricted observation. Several fine descriptions can produce the same coarse record, but a finer detector or a shifted comparison can recover distinctions that the first device omitted. The choice of admitted tests therefore determines the appropriate operational description. [B05]
The laboratory uses a normalized nonnegative angular density on one cycle, f(x)=[1+v cos(x−φ)]/(2π), with 0≤v≤1. Here x and φ are angles in radians inside the equations. The visibility v and phase φ specify the supplied profile; N equal-width bins specify the detector. The drawing uses exact bin integrals, not sampled fake data.
Refine the detector without changing the source
Change the number of bins, the phase, or the profile visibility. The bars are channel probabilities. Increasing the bin count makes each individual probability smaller even when the underlying profile is unchanged. For comparing shapes across resolutions, divide a bin probability by its angular width to obtain its average density.
Interactive calculation
Play scans Phase φ (degrees). Step advances it. No animation starts automatically.
Enable JavaScript for adjustable values. The static figure and worked calculation remain available.
With v=1 and φ=0, two half-cycle bins each receive probability one half. Their coarse record conceals the cosine structure because the positive and negative portions cancel within each half. Four quarter-cycle bins instead have probabilities 1/4+1/(2π), 1/4−1/(2π), 1/4−1/(2π), and 1/4+1/(2π).
Integrate before interpreting the bars
Summing the sine differences telescopes around the full circle and leaves one. Nonnegativity follows from f(x)≥0. The coarse probabilities therefore satisfy the requirements of a measurement record for every allowed setting.
Here sinc(u)=sin(u)/u, with sinc(0)=1. A finite bin averages the cosine and reduces its contrast by sinc(Δ/2) when the bin-average density is expressed at its centre. Additional loss can occur when only a few centres sample the cycle. Neither effect means that the source itself became less coherent.
If a coarse bin is the union of fine bins, its probability is the sum of those fine probabilities. This is a stochastic map from a finer record to a coarser one. For any two distributions p and q, the triangle inequality gives Σ_coarse|Σ_fine(p−q)|≤Σ_fine|p−q|. Coarse graining cannot increase their total-variation separation.
Keep the two-bin detector but change φ from zero to π/2. The two probabilities become 1/2+v/π and 1/2−v/π, so this same coarse apparatus can now reveal the modulation. The earlier equal populations were caused by the alignment of the profile with the bin boundaries, not by a universal inability of two outcomes to carry phase information. This example makes calibration of the reference angle as important as the nominal number of detector channels.
Distinguish detector blur from dynamical loss
The preceding decoherence chapter changed a density matrix through a physical channel. This laboratory keeps its input profile fixed and changes how an apparatus reports it. Similar-looking flatter plots can therefore have different causes. A useful comparison must calibrate bin widths, channel efficiencies, background rates, and any reconstruction applied afterward.
The book's interference record makes coherence measurable through a specified comparison. Resolution adds an apparatus transfer rule to that comparison. Recovering the source profile then becomes an inference problem with its own assumptions, rather than a direct reading of every drawn point. [O01]
For SMM/DTT, the constructive target is a source and detector model that predicts the same retained distribution across calibrated resolutions. Agreement obtained only by changing free parameters for every binning would not establish that consistency. Continue to correlations to compare records from two separated instruments.

