The star’s right half has not slowed down. It has become less visible. A weighted average turns that distinction into an apparent velocity, and the same weights also move the star’s light center.
Use projected coordinates across a unit stellar disk. The observer lies on negative z, the visible surface has z<0 and μ=−z is the local line-of-sight cosine. The quadratic quiet intensity describes the darker limb. The chosen coefficients are illustrative, fixed for this solar-radius teaching star, not a fit to CoRoT-7.
Iq(μ)W=1−0.45(1−μ)−0.20(1−μ)2,=Iq(μ)[1−csg]
Quiet intensity Iq and total weight W share an arbitrary intensity unit. The contrast cs is positive for a dark region and negative for a bright one, and g gives its dimensionless spatial profile. Projected area belongs in every integral. On a surface-area grid that same projection would require an extra factor of the line-of-sight cosine.
Let n be a surface unit normal, ns the unit direction to the region centre, and σs its width parameter. The region is a smooth spherical Gaussian, g=exp[−(1−n⋅ns)/σs2], with angular width in radians. Its value is one at the region centre and fades smoothly with angular distance. This avoids a hard cell-membership edge while keeping a localized region that rotates. It is not a uniform circular spot with a sharp radius. Surface latitude and longitude specify its center; direct dragging changes those physical coordinates.
Assume every patch has the same symmetric Gaussian absorption profile, width and equivalent width. Its centre combines the whole-star reflex velocity v⋆, local rotation and an active convective redshift Δvc. The angular speed Ω uses radians per second, R⋆ is the stellar radius in metres and i⋆ is the spin-axis inclination to the line of sight. The horizontal coordinate x is dimensionless, in stellar radii, increasing toward the receding rotational limb. A patch’s continuum weight also multiplies its absorption deficit. Summing first, then integrating the line, gives a particularly useful simplification: each symmetric local profile has first moment equal to its center.
vcellv=v⋆+ΩR⋆sini⋆x+Δvcg,=∫WdA∫WvcelldA
The projected disk-area element is dA, and v is the resulting line centroid in metres per second. Positive velocity recedes. The displayed line is synthesized from these same weighted centers. A real cross-correlation velocity or bisector can respond differently when local line shapes and equivalent widths vary.
With no active perturbation, symmetry cancels the odd rotational moment exactly. The centroid becomes the Kepler reflex velocity. In the implementation the quiet odd moment is removed analytically before summation, which avoids cancellation between large positive and negative contributions. The plotted difference from a quiet spectrum reveals a small distortion that would otherwise be difficult to see beside the full line.
Imagine an otherwise symmetric unit-flux disk. Remove a fraction 0.01 of its original light from a small patch moving away at 1000 metres per second, at horizontal position half a stellar radius. Ignore convection and companions. The remaining flux is 0.99. The original first moment was zero; after removal it is minus 10 in flux-weighted velocity units. Divide by the remaining flux to obtain about minus 10.1 metres per second.
The denominator matters because the star is now fainter. The horizontal photocenter is about minus 0.00505 stellar radii. At 20 parsecs, a nominal solar radius subtends about 233 microarcseconds, giving about minus 1.18 microarcseconds.
In a rigidly rotating, flux-only model, the rotational velocity is linear in horizontal position. The integrals therefore share an exact proportionality after subtracting the stellar orbit. This is a useful independent check, not a universal activity correction.
Δvrot=Ωsini⋆Xphot,surface
The position here is a physical length along the rotational velocity gradient. Adding an active convective shift breaks this relation even if the continuum brightness is unchanged.
The browser integrates a smooth visible-hemisphere coordinate chart using Gauss–Legendre nodes: weighted sampling points whose sum approximates the area integral. Increasing the order adds points to resolve smaller regions. Ordinary presets use order 48 in each coordinate; a two-degree region raises that order to 129. The independent reference uses polar area quadrature and an independently solved eccentric anomaly. Across 35 preset and extreme samples, the adaptive grid differs by at most 0.0000161 metres per second in centroid and about 0.000000501 microarcseconds in photocenter. These are finite reference-comparison errors, not physical accuracy claims.
A fixed order-48 grid misses the narrow-region extreme by 0.175 metres per second. That failed resolution test is retained. The sampled spectrum uses 80-metre-per-second velocity bins with linear deposition, which preserves the first moment; this bin width is not an instrument’s achievable precision. The continuous centroid and the sampled profile agree to the stored numerical tolerance.
The model omits differential rotation, evolving multiple regions, granulation, limb-dependent convection, spectral-type changes and occultations. Its active-area proxy has no calcium calibration; its Gaussian-equivalent line width is not the published HARPS fitted FWHM. It predicts a controlled mechanism, not the true surface of the observed star.
Actual synthetic records generated by check-surface.ts. Nominal solar radius and GM, distance 20 pc; sample phase 0.9 radian of rotation. Full settings and 121 acquired times per case are in default-results.json. These are not HARPS observations.
Double the removed-light fraction in the worked example. Does the centroid exactly double?
Read the explanation
It becomes −20/0.98≃−20.4 metres per second, slightly more than twice the original magnitude. The missing first moment doubles, but the remaining-flux denominator also decreases.
A photocenter stays fixed while the line shifts. Which assumption in the proportionality could have failed?
Read the explanation
A changed local convective shift is one possibility: it changes the line center without needing a matching brightness displacement. Unequal local equivalent widths or line-shape changes also break the shared weighting assumption. It would be premature to identify the residual as a planet.
11 September 2014 version. Sections 2.1–2.3, Equations 1–7 and Figure 1: surface-profile integration, flux contrast and convective blueshift. Our identical Gaussian local profiles are a simpler independent mechanism model.
24 June 2015 version. Sections 3.1–3.4 and 4.1–4.2, Figures 3–4, Appendix A.2. Derivative covariance signs follow a lag equal to first time minus second time; kernel-width normalizations must be kept with their equations.
Official adopted table: nominal solar and terrestrial gravitational parameters and nominal solar radius are exact conversion constants, not uncertainty-free measurements of the real Sun and Earth.