Report Gene

7/21/20 14:24

Topic

Webinar ID

2020 Sagan

940 3997 6754

Question Det

Question

Answer(s)

What exactly true anomaly meant in reality?

It's an angular parameter that defines the position of a planet (or other body) along its Keplerian orbit. Specifically it is the angle between the direction of periastron and the current position of the planet, as seen from the centre of mass

There are several quantities that are different from star to star: the radial component of the true space velocity, the gravitational redshift, and the convective blueshift (= average of "up" and "down" motions in convective cells. Besides that, some data reduction packages compare the stellar RV at a specific time to an average of all spectra of the same star. In that case, all measured RVs have only a relative meaning. So gamma has to be determined for each star separately in any case.

How do we practically measure the gamma term in the RV expression?

You measure the RV using a measured spectrum against a spectrum that is calibrated to the wavelength in the stellar rest frame.

Why is measuring gamma accurately problematic? After years of doing an RV survey with a specific instrument (presumably including many planet-less stars), won’t you have a very good measurement of gamma as well? At least between stars relatively?

Convective blueshift for one will affect it as that is not a motion of the full body, but of the surface convection cells. Then it still depends on how you extract the RV, your mask weights for example.

The accuracy (not precision) of the absolute RV hinges upon your accuracy of wavelength calibration on a stellar rest-frame specrtum. This depends on fundamental astrophysics (line formation, quantum mechanics, etc.). Even in the Sun, this is very hard to achieve to better than 100 m/s precision, due to astrophysical processes that change the line shapes like Annelies pointed out, plus uncertainties in atomic physics / line calibration.

What is periastron?

The periastron is the point on the orbit that is closest to the star

Is the only way to obtain the RV semi amplitude through regression?

Not sure what you mean. There are many model fitting techniques you can use to derive the (K,omega,P,T0,e) set. “Regression” often implies a linear fit, which you can only do if you know the values of P and e a priori (because those parameters are nonlinear).

Is the big M in the mass function the total mass of the system or only the mass of the star ? What about planets around binaries stars in circumstellar or close orbits ?

In Jason's slides, big M was the mass of the star. You see a M+m term in the denominator, where M is the star mass and m is the planet mass. In my slides yesterday I used slightly different terminology: M1+M2; but the meaning is identical.

For Jason: Can you comment more on what you said about most systems being near edge-on? I don't get why that is true.

My favorite analogy, from a paper by Andrew Howard and BJ Fulton, is that assuming the orientation of all planetary systems is randomly distributed on the sky -> uniform distribution of an 3-D angle -> throw a dart randomly on a globe with equal chance of landing anywhere -> having sini > 2 is like throwing that dart at Earth, and landing in the polar regions -> <15% chance. ;)

what are the typical methods to get the mass of the stars, for the exoplanet community standards? let’s say from dwarfs? derived from spectra, sed or mass luminosity relations?

Via isochrone analysis using spectroscopically derived photospheric parameters (temperature, metallicity, ...), magnitudes, and a trigonometric parallax is a widely used method.

Do you know a source that compile these masses?

Depends on the sample. Safest thing to do is query Vizier and search the output. For nearby Sun-like stars, some recent big compilations come from Brewer+2016, Luck+2017. For M dwarfs, see Mann’s recent papers (especially 2015). For M dwarfs, one can get good masses from absolute K magnitudes (see calibrations by Mann+2018, Benedict+2016). Agreement between authors for main sequence FGKM stars tends to be good (typically <5%) but as soon as the stars evolve off the MS, and especially for subgiants and giants, the mass estimates can vary much more.

Edge on systems are statistically more likely. Is this due to the shape of the galaxy? Have we ever tried a survey looking at a piece of the sky perpendicular to the disk of the galaxy?

No. Systems are randomly oriented in space. This means that the polar axis can point in any direction with equal probability. But that means it is more likely nearly orthogonal to your line of sight than close to aligned with your line of sight, just as there is more "real estate" near the equator than there is near the pole.

You mentioned that a lot of planets are close to “on edge” with the observer. But statistically, wouldn’t all i values have the same probability ?

Yes, but what we measure is sin i, so it is no longer uniform.

No. Each value of cos i is equally probable. Think about it in this way: The polar axis points to any direction in space with equal probability. But there is a lot more "real estate" near the equator than near the pole.

Jason Wright’s talk - I thought he said the majority of observed systems are nearly edge on. Did I catch that right, and if so, why is this true if systems are randomly oriented?

"Random orientation" means that the polar axis (angular momentum vector) points to any direction of the sky with equal probability. But there is more "real estate" close to directions perpendicular to your line of sight than close to the direction along the line of sight, just as there is more real estate close to the equator than there is close to the pole.

Is there a reason why most systems are edge on? Looking from Earth, shouldn't a more uniform distribution be expected?

The distribution or orientations is uniform in 3D, but in 1D this projects to being uniform in sin(i), not uniform in i. There are “more ways” for a random sytem to be edge on than face-on.

Can we solve for the orbital elements of multiple planets from radial velocity or astrometric data?

Yes! I recommend my paper with Andrew Howard for all the hairy details: ApJS..182..205W

Do periodic residuals point towards more planets or bodies?

They could. It can also be stellar activity, see tomorrow’s talks

How do you wisely choose your grid in the periodogram? Spacing, number of trail periods (frequencies), min P, max P? How do you evaluate it based on your data?

You should sample the periodogram as densely as you need to to capture all of the structure. You can tell empirically by examining the peaks to see if they are well sampled or not.

Is there a simple relation to assess if numerical simulations of planet-planet interactions must computed or if the 2-body approximation remains valid ?

Not really; it’s mostly heuristic. But in general, it’s only a big deal in systems in mean motion resonance (periods are commensurate in a ratio of small numbers). First order resonances are usually the strongest.

Can you elaborate on dynamic planet- planet modelling?

I didn’t have time to do this, but the essence is that eveny model evaluation you preform is actually driven by an n-body simulation of the planets and the star. It can be very computationally expensive!

Do you understand which signal in the periodogram is true and which one false for the plot?

This is a huge topic! You can’t tell just from the plot, you have to compare models with and without planets, makes sure you have properly accounted for your noise, and assign probabilities.

You mentioned that the periastron can precess in the case of two body resonance so it made me wonder if general relativity orbital precession could also fool us when analyzing a periodogram ? (hot Jupiters being very close to their star for instance)

Thanks for your question! You are right that GR precession may be detectable for short-period palnets, particularly those with eccentric orbits. But, the effect is very small, less than 10 degrees per century change in the argument of periastron for nearly all known planets

do the periodograms consider single measurement error when looking for the signals?

They can do, if generalised versions are used

Does nutation play any important role in modeling?

Nutation is a critical component in calculating the barycentric correction (See Wright and Eastman 2014). That correction is necessary to remove the motion of the earth from RV measurements that you obtained at different times, and is a precursor step to the orbital parameter fitting that Jason described.

No, I think nutation always has to do with the orientation of a body in space, not an orbit. The nutation of Earth has a small role in the barycentric corrections.

Can Gaia be used for absolute RV

Yes, Gaia measures absolute RV, but with an accuracy and precision that is much poorer than what we need for exoplanets.

How much more computationally expensive is the N-body dynamical modelling of RVs compared to Keplerians?

Lots! You can solve the 55 Cnc system kinematically (lots of model evaluations!) on your laptop in a second or so. n-body simulations are much more expensive (especially if there is a short-period planet) and MCMC work with them generally involves supercomputers. See Ben Nelson’s paper on 55 Cnc for an example.

Where do I get the CCF mask from?

One can either use a "synthetic" mask from stellar model atmospheres and laboratory data, or use an average of all observed spectra from the star under consideration.

Why stellar activities are there in star and how we dignose by line bisector?

There are a number of physical phenomena that produce different stellar activity signals in the RV measurements. We’ll be talking about these in much more detail (and addressing things like how line bisectors work) during the first set of talks tomorrow.

What causes iodine lines?

You place a cell of gaseous iodine in the light path of the telescope. So the light from the star passes through the iodine cell and this imprints the iodine lines on top of the stellar spectrum

How low (how precise an RV measurement) can we go with CCFs? I think ESPRESSO has CCFs so getting by orders RVs from these CCFs for RV timeseries or even checking on stuffs like Rossiter effect wouldn`t be a good idea, right?

For the brightest stars, CCF RVs can be measured with precision of 10cm/s for ESPRESSO. 0.5 to 1m/s for other state of the art instruments. If stars get fainter, the precision gets worse of course.

Do you obtain different LSF across the wavelengths in the forward modelling?

Ideally yes, although this can be a difficult quantity to measure. We are attempting to do so with laser comb spectra, which have lines with very narrow intrinsic line shape. LFC though rarely cover the entire spectral range of our spectrometer. Etalons are broader in wavelength coverage, but also have broader intrinsic line width so are less good for deriving fundamental LSFs. Once you have a wavelength/spatial dependent LSF, then you also have to figure out how to use it. The CCF algorithms get a lot more complicated once you throw away the assumption of constant LSF.

What is the physical reason for the instrumental broadening?

Thanks for your question. Optical effects, such as diffraction, mean that a real optical system (like a spectrograph) will generally map even an really tiny, monochromatic input spot of light into a larger spot on the detector. This mapping (input illumination shape to detected illumination shape) can be thought of as a convolution and the kernel is called the specrograph line spread function (LSF).

Finite pixel size, finite slit/fiber size, and camera aberrations usually contribute to similar magnitudes.

What can be the possible effects of stellar atmosphere on CCF bisectors?

See Heather Cegla’s talk on Wednesday.

Are contamination and noise the same or different?

I think contam is things like cosmic rays which leads to noise?

We measure the light coming from the star and the reflected and doppler-shifted light from the planet. Should the reflected light be considered to reach very high precision RVs?

With EPRV work rarely detect light from the planets (only very close giant planets to very bright stars), but when we do we are dealing then with a “double-lined” spectroscopic binary. The velocities of the planets are very large in this case, and we measure these to relatively high precision (10’s of m/s, I think). Chad Bender has some nice papers on how to do this.

Which model will be more efficient in determing RV of a close binary system with an exoplanet/a brown dwarf?

Do you mean forward modeling vs. CCF? This is a complex issue, and both methods can work well for double-lined spectroscopic binaries.

Ok so the exmaple mask in the slides is not realistic then right? The mask looks more like the actual stellar spectrum at its zero point?

It depends - you can either use a stellar template (so comparing your later observations to a single spectrum of the star that you set as the zero point), or you can create a binary mask (like what Sharon showed) that allows you to include only those lines you select as being high quality. This means you can eliminate lines that are blended, saturated, and/or prone to stellar variability

Are these spectra affected by the exoplanet's atmosphere (during transit)?

There can be small effects from the atmosphere, and there’s a whole field of exoplanet science focused on this (called transmission spectroscopy) that uses that fact to try and identify what’s in the planet’s atmosphere. The bigger impact from and RV point of view, however, is that during transit the planet blocks out parts of the star’s surface which changes the RV signal. This is called the Rossiter-Mclaughlin effect and it can be used to infer the spin-orbit alignment of the sytem. There are a bunch of paper’s on these kinds of measurements, but here’s a nice review: https://www.cambridge.org/core/services/aop-cambridge-core/content/view/S1743921311020230

I was aware of transmission spectroscopy but not the Rossiter-Mclaughlin effect.

We ideally would not need blaze correction to determine RVs. Still, what is the best way to incorporate this correction. Dividing the blaze doesnt work always for less S/N data.

One needs a blaze correction in the data reduction because the blaze function introduces a slight shift of the line center - the spectrograph is a little more sensitive on one side of the line than on the other. This by itself would not be a problem, but the annual variation of the barycentric correction (the space velocity of the Earth) makes the lines slide up and down the blaze function, and that leads to an annual change of the shift.

How does not removing the blaze function affect measurements?

The blaze function introduces a slight shift of lines, as the spectrograph is a little more sensitive in one line wing than in the other. This by itself would not be a problem, but the annual variation of the barycentric correction slides the lines up and down the blaze function. So not removing it causes a slight variation of the line center in sync with the variation of the barycentric correction.

There is also posibble contamination from other close stars (few arcsec separation and poor seeing)

Certainly! There is also possible contamination from faint, wide binary companions you don’t even know about. Such a situation bit is in the MARVELS-1 system: ApJ 770, 2

When getting the final RV for an epoch from multiple echelle spectral orders, is there any difference (advantages/disadvantages) between combining the individual RVs per order and combining the CCFs to extract the RV once?

Generally we determine the RV using the entire spectral range because this increases the SNR of the measurement (more lines = more RV information) and often times the RV precision that can be obtained with just a single order isn’t very high. But there’s been some interesting work done recently looking at order-by-order measurements as a way to look for chromatic effects in RVs (see, e.g. https://ui.adsabs.harvard.edu/abs/2018A%26A...609A..12Z/abstract) which can be used to differentiate between Keplerian signals (which should have the same RV amplitude at any wavelength) and stellar variability (the amplitude of which varies as a function of wavelength)

In priniciple, the doppler shift is the same in all orders, so you could do it either way. In practice, different orders (or even different *parts* of orders) will exhibit different zero points, depending on the method used. So most methods measure the *change* in the RV separately for each order, and then perform some sort of weighting based on which sections of the spectrum are most indicative of the true shift.

Thanks!

Is the solar contamination by the lunar reflection expected to produce power in periodogram at preferentially period as 28 days for instance ?

It can, but it will also depend on when you happen to get telescope time, whether your star is near the ecliptic, whether you observe near twilight, and how much Mie scattering there is in the atmosphere (dust, cirrus).

Excited to read your new paper! For the telluric correction test where you also considered a wrong telluric profile - how different were the line profiles in terms of residual error? Were there line strength cut offs after which a telluric line was masked?

Great question. The residual is roughly around a couple of percent in RMS, mimicing the typical precision we could model tellurics these days, as you know. The amplitudes of this residual can be as large as 5% or maybe a bit larger, in reality. The way we mimic this “wrong profile” to represent our lack of knowledge on the Earth’s atmosphere is to use the telluric line profiles for Mauna Kea to fit for tellurics generated for Kitt Peak. We did not mask any telluric lines or regions with large modeling residuals. That’s a good point - perhaps one should…

How does the sky fiber help, to remove solar contamination, the signal from the sky fiber may be poor?

Yes, the idea is that you use the sky fiber to obtain a simultaneous spectrum of the local sky (due to moon light, sky glow, etc). You are right that a pretty high S/N is required in the sky spectrum to do any useful correction of your science specrum.

Fortunately, we know the solar spectrum extremely well (the biggest uncertainty is the RV shift from the air) so we only need to know an integrated estimate of the brightness in broad bandpasses.

If you have sunlight reflected by the moon (and perhaps reflected again by a cloud), you get (nearly) the same amount of that light into the stellar fiber and in the sky fiber.

How important is it to use a mask of the exact stellar type of the target if using a synthetic mask? For example, different subclasses between the mask and target?

While obviously a mask that "perfectly" matches your target star is ideal, it's also not realistic. We tend to use a small number of masks (3-5) with temperature steps of 500-1000 deg between them. This is an approximation, but one that does not dominate at 1 m/s precisions.

modeling the telluric lines is hard on M-dwarfs, how we can compute reliable RV in we dont have a good model for the these lines??

Cullen Blake is going to give a whole talk on exactly this topic tomorrow, so he’ll provide details there. But it’s a work in progress, and one that is becoming more and more important as instruments push further into the Red and NIR wavelength regions where telluric lines are more prominent

What is a Reduced Julian Date (RJD)?

Astronomers use Julian date (JD), which is just a counter of the days from a certain starting date. Since this counter is large, one can subtract 2400000. That's RJD. It is often used to make plots less busy.

Object and sky are sampled by different fibers, what accuracy we need to subtract the sky

Depends on what you’re trying to observe. Solar contaminaiton in a clear sky and oxygen lines are pretty slowly varying so sky fibers are pretty reliable. Water vapor is another story, and can be very hard to handle. Arpita Roy’s paper has a lot of great detail on how well solar contamination needs to be done: https://ui.adsabs.harvard.edu/abs/2020AJ....159..161R/abstract

Would be it false to select, for the number planet ,the « knee» of the chi_2 curve as currently performed for the number of principals components with PCA ?

It would be ad hoc, rather than a method with a sound statistical method. That said, it wouldn't necessarily give the wrong answer. But one would need to perform lots of tests to understand under what conditions such a procedure gives reasonable "answers".

Is the prior similar to the initial guess of the fit ? Could the best fitted solutions converge outside the prior or would such solutions be penalized in some way ?

They are penalized. In Bayesian statistics, the probability distribution of the fitted parameters (the posterior) is constrained by a combination of the data (the evidence) and the prior. In regions where the prior is zero the posterior is also zero.

If you had a massive EPRV survey running, such as TESS or Kepler, with massive amounts of data coming in from one facility, how would you define an initial “detection alarm” for a planet signal? (considering computational efficiency, automatization and minimizing false positives)

The Kepler pipeline set its detection threshold so as to expect one _statistical_ false positive out of all the planet search targets, _if_ the noise (after a pre-whitening process) were Gaussian.

How we can know/select the best model for the GP?

Applying different GP kernels to many simulated datasets can allow you to evaluate which GP kernels perform well for your purposes. Christian Gilbertson (grad student in my group at Penn State) has been working on this and found interesting results. Some kernels do perform slightly better than the quaesi-periodic kernel for solar-like variability (at least based on SOAP 2.0 simulated spectra). A draft is coming soon (once Christian gets back from his summer internship).

What is effect of uncorrected atmospheric dispersion correction that is variable with time on the RV signal

This is complicated. What will happen if you don't have a good ADC (or any ADC) is that light at different wavelengths will not be injected into your spectrograph in the same way. However, this will depend on the altitude that each observation is happening at. Say you have a spectrum with a lot of strong lines in the blue. If you observe at different airmasses without an ADC, then the SNR from those lines will be varying. Additionally, if your fiber scrambling is not perfect, the chromatic 'spot' of the star on the fiber tip will also move as the altitude changes. That will result in chromatic RV shifts in your spectrum that depend on the altitude angle.

Is there a criteria to choose a kernel for the Gaussian process?

Are these results available already? Would be interesting to look in detail

They're comin

Regarding using GPs for planets with periods similar to stellar activity periods, would you recommend using alternative parametric methods to obtain basic parameters for each system in blind searches and then using GPs to refine these, or can you use a two-step GP process?

Could you clarify what you mean by "basic parameters for each system"? What steps are you proposing be split up?

I was considering rotation/potential planet periods, amplitude of stellar activity etc, to act as more informative prioers for a GP fit.

Jason, when applying the barycentric corrections, is it better to correct the spectra and then calculate doppler shift, or apply to the final doppler shift? in the first option, is it necessary to interpolate the flux when shifting the wavelength range (since the pixels were measured to be sensitive to a certain wavelength)

Definitely don't “correct” the spectrum, except perhaps by shifting the wavelength solution in a precisely known way. Any process that includes interpolation can introduce lots of trouble at the precisions we care about.

Does telluric absorption and emission (as in the ones corrected for in todays second live talk) strongly depend on atmospheric changes and seeing? Are they not corrected by usual correction techniques used in all observations?

Yes, tellurics are strongly depednent on atmospheric conditions which could vary on timescales as short as even minutes (e.g. recent work by Cullen Blake and co.), and certaint vary w.r.t. line of sight / airmass and so on. They are corrected to various degrees with traditional methods like taking telluric standard frames and sky frames etc., but the question is “to what RV precision”. That’s the motivation for more sophisticated methods for correction, as we aim for 10 cm/s precision and better.

What is your opinion about RV measurements in binaries? What are the real limitations of detecting planets around close binaries such those discoveried by Kepler (periods of binary around 10 days, with period of planets around 100 days) ?

Circumbinary planets are more challenging because you have *two* sets of stellar lines to account for, and you are looking for a variation in their common center of mass due to the planet. This is a more complex problem and usually results in lower overall RV precision. Binaries with 10 day periods will also have other nasty effects from tidal interactions, that may interfere with the measurements. In general, single-star RVs will always be “cleaner”.

I think that tidal interaction is not such important, as we can see in Kepler binaries. I think that maybe the main constrain is dopler-modelling of binaries as first stage.

(I`m not super familiar with GP) is it possible to use GP for extracting RVs from CCFs? Perhaps using the gaussian fit as prior. Would we get a more accurate and precise RV value compared to only fitting a gaussian?

If you already have a CCF, you just need to find the peak location/arg max to get out an RV - no need for a GP to do the peak finding bit. You could e.g. fit a Gaussian or parabola to the CCF instead.

What is need of spectrograph to be maintained in stable pressure and temperature condition?

Changing pressure of the atmosphere inside of the spectrograph changes the index of refraction. Changing temperature causes the glasses and metals used in the spectrograph construction to expand or contract, which changes how the light beams hit them and results in RV shifts.

What is your take for the future, will we have a standardized method or are new promising statistical tools on the way ?

I think that both on the extraction the RV from the spectra and the analysis of the RV time series, we’re far from having explored everything. In my opinion, it is likely we’ll make significant progress

GPs have the danger to overfit data. I find it sometimes really difficult to decide how much one should constrain the GP hyperparameters, e.g., by using photometric rotation periods for example, especially since the effect of each hyperparameter can be captured by an other one, e.g. a faster decay of the GP model can reduce the harmonic complexity and vice versa (degeneracy). Is there any kind of metric (except for the log-likelihood or evidence of the posterior samples), which could help to define how flexible or constrained a GP should be to model stellar activity? And what about instrumental systematics, there should be a kernel component for such red-noise, too, which would make the GP even more flexible. This would be especially important for finding planetary signals hidden behind stellar activity.

It's hard to overfit with GPs - GP likelihood actually includes a built-in complexity penalty term - but quite easy to fit the wrong signal (e.g. planets instead of activity). In the example you gave, it'd be better to model photometry simultaneously with RVs, using a GP, rather than just condensing all the info in the photometry to e.g. a constraint on your GP period. That aside, there are many sensible ways to constrain GP hyper-parameters. E.g. evolution time scale must be longer than a single period, else the GP model is not even quasi-periodic. Harmonic complexity can be constrained by geometrical considerations/simulations with rotating active regions; extremely complex signals are very unlikely. Etc. The most general/robust way I can think of to quantify whether the use of GP is justified/necessary at all would be comparing Bayesian evidence for a model with and without the GP

Also it's worth remembering that the problem cuts both ways: if you have activity present and you *don't* model it adequately, your Keplerian terms (i.e. planets) might well absorb some of the activity variability and inflate

If we do not find any statistically significant peak in the Generlaized Lomb-Sacrgle periodogram, Is there any possibility of the presence of additional companion in extra-solar planetary system?

Certainly! There are likely many orbiting bodies at an amplitude too low to see, or there maby be a planet at a very long period or high eccentricity.

Sometimes stellar activity indicators such as Ca II HK or H-alpha or BIS, etc. How does one make the choose?

Look at all of them. If one (or several) of them shows a periodicity identical to the RV, the RV signal is very likely due to stellar activity, not to an exoplanet.

Question for Vinesh: can you elaborate a bit more in what ways one can prevent the GP from absorbing too much of the planet signal?

Fitting a GP simultaneously to RVs + activity-sensitive time series (e.g. BIS, Calcium HK, CCF FWHM, photometry), and allowing the GP only to fit signals present both in the activity-sensitive time series and the RVs. NB this ins't the same as just setting a prior on (e.g.) GP period hyper-parameter. Also worth remembering that the problem cuts both ways: if you have activity present and you *don't* model it, your Keplerian terms (i.e. planets) might well absorb some of the activity variability and inflate. Reference: https://doi.org/10.1093/mnras/stv1428

what is 5150,5155 used for

That is just a wavelength range in Angstroms chosen for convenience for the zoomed in plot.

is there any benefit on oversampling the spectra when transforming to log-linear grid?

Yes, it can be helpful to oversample the spectrum a bit, maybe 2x-4x. However, the interpolation technique is really important. It can have a major impact and actually bias your RVs. Best to test different techniques on simulated spectrum.

While taking the FFT, how would you deal with the high frequency noise in real data? Should you use a taper or window function?

This is a good question that none of us really know the answer to. In a lot of cases the template/reference is noiseless but of course the data are not. Ultimately, I believe this manifests as noise in the CCF and the peak finding algorithm has to deal with that noise.

Why do we need a log wavelength axis for the CCFs?

The canonical answer for this is Tonry & Davis 1979 AJ 84 1511. However, for a binary CCF mask approach we often don't use this.

102

Are there other spectral window functions other than deltas which are used while sampling? If so how does one choose? Whats criteria

Yes, sometimes Gaussians are used, or a real observed spectrum (sort of what we do in the notebook). There are pros and cons for each. Its best to try several different masks and see how your choice changes the result.

105

In this notebook, how would I estimate the uncertainty ? do I want to fit a gaussian around the peak corr and get the spread ?

Yep, thats definitely a viable technique. There isn’t one right answer here.

You might find astropy.modeling handy here. Take a look at “models” and “fitting”.

106

Can I ask you a question a bit out of context? So I am a freshman undergraduate and I didn't actually understand everything you've taught. So, can you tell me like how should I start learning about exoplanets to reach the level of understanding of this webinar?

One reference I found very useful to get a broad overview of exoplanets is ‘The Exoplanet Handbook’ by Michael Perryman: https://ui.adsabs.harvard.edu/abs/2018exha.book.....P/abstract

There is also the “Handbook of Exoplanets,” which you can access for free on arxiv.org (chapters are posted separately), but may be more techincal.

Josh Winn’s Great Courses lectures on exoplanets is also a great overview of the field for beginners. As a relative new comer to the field myself, I would also recommend reading papers and emailing authors for clarification as needed. That’s what has helped me most. I’m guessing a lot of your difficulties with these talks is the terminology and methods which you can learn through papers. Mayor and Queloz 1995 (the 51 Peg discovery) is a nice place to start.

Also check out the astronomy reading seminar by Cooke et al. 2020 for help with reading papers.

107

Will the correct answers to the activity be given at some point?

Yes, I’ll make sure there is an answer key version of the notebook up in the repo before the end of the week.

108

How can we find the habitable zone of a star?

I suggest saving this until Thursday, and asking Jim Kasting. He created the concept of the HZ many years ago.

110

If the exoplanets do have moon(s), how would this moons and planets system affect the stars wobble?

The star will be tugged by the center of mass of the planet-moon system. So you would simply see this as a larger amplitude signal (to first order). There are subtle effects, but these would be very difficult to disentangle from orbital paramters like eccentricity and/or other planets in the system.

113

Concerning the gamma term. If we are observing the same star with different instrument, is the offset between these RV sets coming almost entirely from the instrument? The stellar contribution is quite unpredictable and hard to derive, right? how do we account for this so that in the end we could get an absolute RV measurement? Or we do not need to know gamma for absolute RVs?

These are the zero point correction?

116

are there any planets which are found around a star in binary system ?

Yes there are, you can check Amaury Triaud’s papers for instance

Thank you

120

can you explain why we need to build the spline object for resampling the model spectrum?

This is simply the way the Python package works, in fact there are other Python packages/modules that do this in a way that don’t reaquire you to initialize that object first.

122

As mentioned in one of the talks, we can only find the minimum mass of the planet and its was also mentioned for an object to be a exoplanet mass should be less than 15 jupiter mass. Is there any possibility that an object which was classified as a planet might be a brown dwarf ?

Yes, this does happen. I highlighted on example yesterday in my talk - Latham's planet from 1989 was recently determined by GAIA to have a very face on inclination (~5 deg). That moved it from planet to low-mass star mass regime.

125

How does the detection of exoplanets help to understand the large scale picture of universe ?

By helping understand the frequency of other planets like our own.

128

Assuming we're able to detect planets in another galaxy, do we still just need to do the regular corrections for instance baricentric or somehow should we need to take into account the sun's velocity through the galaxy. What about the galaxy intrinsic velocity and stellar velocity around its host galaxy... ?

probably general relativity as well (?)

130

How accurate the radius values that we calculate for the planets using RV?

RV gives you mass, not radius. It is usually accurate to about 20%, but that depends on the particular planet and star system.