davidwhogg / avast Goto Github PK

• bullet point about sparseness on page 3: I would change the last sentence to "we expect only a few, isolated lines to participate in (say) a given component of the surface-velocity-related distortion" because we definitely wouldn't expect ALL basis functions for surface-velocity effects to be isolated to a few wavelengths - something like convective blueshift will distort/change all the line profiles the same way to first order.
• L and T are not defined for the summations in equations 5, 6, 9 (on purpose? how do we define those anyway - arbitrary choice?)
• b_l not defined in equation 6
• for "systematics" (equation 9), are those meant to be instrumental or continuum normalization-related in origin? If so, shouldn't it be a function of plain wavelength, not stellar rest-frame wavelength?
• I'd also like to propose that the logo of Avast be a cartoon star wearing an eyepatch. (get it? because it's like a pirate but also like a starspot??)

Let's define

\xi \equiv \ln \sqrt{ \frac{v+c}{v-c} }

and fit for \xi not \beta. With this definition (or maybe with the + and - reversed),

x' = x - \xi

Spline will probably go crazy near the edges of the fitted region - how do we prevent that?

These can be constructed from the derivatives you have.

Then run it into scipy.optimize BFGS and see if that crushes.

the code is currently brittle here.

• Get a working model of the mean spectrum, spectral scaling, and RV adjustments
• Look at the residuals away from this model in small spectral patches
• See if there are variations at Teff and logg indicating regions, and also activity regions
• Plot these variations against the HARPS RV outputs

This will give us some idea whether we have a chance of getting anything.

Scale parameters should absorb the varying flux levels so that a_ms just fit the shape of continuum & spectral features.

Right now lots of things are hard-coded.

@dfm says we should

@dfm says we should