Abstracts

Activity indicators are a useful way of checking and/or modeling radial velocity signals arising from stellar activity processes. However, it is essential to bear in mind that these indicators are not directly correlated with the radial velocities, but rather shifted by several days. In this talk I will give an overview of this effect as studied on the Solar data from the HARPS-N spectrograph.

The EXPRES Stellar Signals Project (ESSP) is an international research network of scientists working on disentangling stellar signals from true center-of-mass shifts due to planets. With our first round of work, we established the current state of the field through a self-consistent comparison of 21 different methods implemented on the same extreme-precision spectroscopic data from the EXPRES instrument. I will give a brief overview of the project so far and highlight commonalities between the different methods. The bulk of the session will be a discussion of future directions for such collaboration and next step data challenges.

The intrinsic variability of young active stars is one of the main limitations in detection and characterisation of exoplanets with Doppler spectroscopy. Indeed, stellar activity affects the measurement of radial velocity (RV) often preventing the correct detection of the planet, in particular in presence of fast rotation.

We present a method to optimize the RVs extraction based on the cross-correlation function (CCF) technique that takes into account the activity of the star modelled by the presence of one or more spots on the stellar disc.

We applied our technique to 146 HARPS-N spectra of V830 Tau, a young fast rotating star. We find that the star is well modelled by applying a “Two-spots model” to the CCFs. We verified that the method removes efficiently part of the rotation signal, and that the new RVs show a significantly lower dispersion with respect to the TERRA pipeline dataset. We also tested the detection sensitivity of the method by direct injection of a planetary signal into the data and we demonstrated that our method improves significantly the detection sensitivity.

The “Two-spots model” provides us also the physical properties of the spots that results modulated by rotation and with filling factor consistent with data present in the literature.

Current theories predict that very young giant planets have large radii and very low densities before they slowly contract to reach their final size after several hundred million years. Open stellar clusters and young moving groups of stars offer a unique opportunity to study exoplanets at their infant stages because of the well-constrained ages.

The extreme stellar activity of young stars makes measuring the masses of their planets a very challenging task. Planet-induced radial velocity signals are typically smaller in amplitude and at lower frequencies than stellar-induced signals. These situations place very demanding requirements for RV observation campaigns and puts us in a worst-case scenario for the standard methods of stellar activity modelling.

With an estimated age of 20 million years, V1298 Tau is one of the youngest solar-type stars known to host transiting planets; it harbours a system composed of four planets, two Neptune-sized, one Saturn-sized and one Jupiter-sized. Using more than 200 radial velocity measurements, condensed into a 4-month period, we were able to measure the radial velocity amplitude induced by the planet V1298 Tau b, and to pick up a second signal that was suspected to be linked to V1298 Tau e. The analysis required the combination of radial velocities, ground-based photometry contemporary to the RV campaign, and K2 photometry. The result obtained indicated that V1298 Tau might have contracted much faster than predicted by theoretical models, highlighting the importance of attempting these difficult cases.

Stellar activity mitigation is one of the major challenges for the detection of earth-like exoplanets in radial velocity measurements. Several promising techniques are now investigating the use of spectral time-series, to differentiate between stellar and planetary perturbations. In this context, developing a software that can efficiently explore the parameter space of stellar activity at the spectral level is of great importance.

In this tutorial, we introduce a new version of the Spot Oscillation And Planet (SOAP) 2.0 code that can model stellar activity at the spectral level using graphical processing units (GPUs). Benchmarking calculations show that this new code improves the computational speed by a factor of 60 while having the same accuracy. On top of that, we implemented a realistic simulation of activity on solar-type stars (convective blueshift, number of active regions, evolution), therefore allowing to generate realistic spectral time-series affected by activity perturbations within seconds. The outcome of this code is essential to test any algorithm that tries improving planetary signal detection by mitigating stellar activity at the spectral level.

Exoplanet detections are claimed based on the value of a statistical significance metric: signal-to-noise-ratio, false alarm probability, Bayes factor, etc. If the statistical significance of a planetary signal is greater than a certain threshold, a detection is claimed. It is unclear which metric and which detection threshold should be chosen. In this talk, we address the question: what is the optimal choice? We pose the problem in the following terms. If the orbital elements of a planet whose detection is claimed are too far, in a precise sense, from those of a planet truly in the data, we count a false detection. On the other hand, if the number of planets is underestimated, we count missed detections. We exhibit a detection criterion called the false inclusion probability (FIP) which minimises the number of missed detection for a certain tolerance to false ones. The optimality is provable mathematically, and illustrated on simulated examples. We show on simulated radial velocity data that the new criterion leads up to 30% more true detections than all its alternatives. We discuss the choice of the detection threshold and the robustness to a model change.

Stellar activity is today the main limitation to the detection of Earth-like exoplanets by the RV method. When dealing with stellar activity, there exist several spaces where the corrections can be performed. From the spectra, to the CCFs up to the time-domain regressions. Each of this space present advantages and drawbacks related to different SNR limits, magnification of stellar activity signatures or collinearity with planetary signals. We proposed a new space between the classical spectrum and the CCF which is a compact representation of the spectrum that we call a shell. A shell can be understood as a generalization of the template matching method, except that the method also provides non-Dopplerian time-series, related to spectral lines’ distortions. We demonstrated that in that space, the stellar activity is clearly visible and can be used to efficiently correct for the rotational period of the star. The required SNR for the correction is estimated around 250 which makes the method currently applicable to stars intensively observed. We found that the first harmonic of the rotational period is more robust that the rotational period.

Radial velocity (RV) measurements are crucial to determine the masses of small exoplanets and investigate the architecture of planetary systems with non-transiting planets. To push the RV exoplanet detection threshold, it is vital to find more reliable RV extraction methods. In this talk I will present the Multi-Mask Least-Squares Deconvolution (MM-LSD) RV extraction pipeline, which I have recently submitted to the MNRAS journal. This pipeline extracts the RV from two-dimensional echelle-order spectra using LSD with multiple tailored spectral line masks after continuum normalisation and telluric absorption line correction. The MM-LSD pipeline was tested on HARPS-N data for the Sun and selected well-observed FGK stars with 5.7 < Vmag < 12.6. For stars with median signal-to-noise ratio above 100, the pipeline delivered RV time series with on average 12 per cent lower scatter as compared to the standard HARPS-N RV extraction pipeline based on the Cross-Correlation Function technique. I will show how using LSD with multiple masks removes the necessity to optimise any parameters and provides precise RVs while retaining the planetary signal. Furthermore, I will outline how the pipeline can be extended to extract a proxy for the hemispherically-averaged stellar magnetic flux. The latter is known to correlate with solar RV variations and is thus a promising way forward to disentangle stellar and planetary signals in RV data and uncover the population of Earth analogues.

Stellar activity limits the ability to detect and characterize earth analogs using state-of-the-art spectrographs. In particular, dynamical mass measurements of these planets require Doppler precision which is equivalent to the noise contribution of line-profile variability. Therefore, detecting periodicity in the estimated radial velocity is often insufficient, and one may also face the challenge of attributing it to the reflex motion of the star rather than other activity-related effects. This talk will present new periodograms that effectively distinguish Doppler shift from spectral shape variability in time-series of astronomical spectra. Based on the concept of partial distance correlation, these periodograms separate the periodic radial velocity modulation induced by orbital motion from that induced by stellar activity. These tools can be used to explore large spectroscopic databases in search of targets in which spectral shape variations obscure the orbital motion; such systems include active planet-hosting stars on the one hand or binary systems with an intrinsically variable component on the other.

Recently, several extreme precision radial velocity (EPRV) spectrographs have been providing high-resolution and high signal-to-noise spectra with the express purpose of exoplanet discovery and characterization. A significant barrier to this endeavor is the existence of time-variable features in the spectra from both telluric absorption and stellar variability. Traditional methods discard significant portions of data to minimize the effects of telluric contamination, but new data-driven methods may enable the use of a larger fraction of the available data. While there exist methods for modeling out the telluric features (e.g. Bedell et al. 2019) or the stellar variability (e.g. Gilbertson et al. 2020) individually, there is a need for new tools that are capable of modeling them simultaneously. Here we present StellarSpectraObservationFitting.jl (SSOF), a Julia package for creating data-driven linear models (with fast, physically-motivated Gaussian Process priors) for the time-variable spectral features in both the observer and observed frames. SSOF outputs estimates for the radial velocities, template spectra in both the observer and barycentric frames, and scores and basis vectors that quantify the shapes and amplitudes for the temporal variability of time-variable telluric and stellar features, while accounting for the wavelength-dependent instrumental line-spread function. We have demonstrated SSOF's state-of-the-art RV precision performance on EXPRES and NEID data and discuss how the resulting model can be used to aid in mitigating remaining sources of correlated noise in the radial velocity time series.

Previous efforts (Dumusque 2018, Cretignier et al. 2020) have shown that individual spectral lines are differently affected by stellar activity, with deep lines generally exhibiting a weaker sensitivity to activity signals. Here, we present a novel approach to investigate how RVs depend on the average formation temperature of spectral lines, retrieved from 1D LTE spectral synthesis. By measuring RV time series of observed line parts—segmented according to their theoretical formation temperature—we study how activity signatures vary within the photosphere. Our analysis is performed on ~3 years of HARPS-N high-resolution (R~115,000), high-SNR (>2800) daily-binned solar spectra taken during the end of Solar Cycle 24. We find a variation in amplitude and periodicity of the activity signal registered at separate temperature intervals, at both short- and long-term time scales. Combined with simulated RVs based on SDO/HMI images of the Sun, our method enables the disentanglement of various RV contributions by matching the effects of resolved surface features with signals seen on the spectral level. This could shed more light on the shifts and asymmetries induced by active regions and convective motion and ultimately enable new methods of mitigating stellar activity to extract planet signals on the sub-m/s level.

Active regions on stellar surfaces induce signals in our spectroscopic time-series that limit our ability to detect planetary induced signals. Such active regions affect observed spectroscopic parameters (e.g. RVs or activity indicators) in different ways. Some of them are only affected by the projected area that is covered by active regions, while others also by how these regions evolve in time. This generates discrepancies between RV and activity indicators time-series, such as different harmonic complexity and apparent phase shifts. In this contribution we discuss how a multidimensional Gaussian Processes framework can be used to constrain the stellar signal in multiple spectroscopic time-series simultaneously. This approach also deals intrinsically with harmonic complexity or phase shift differences that can be present in various spectroscopic time-series. We will show some practical examples of how this approach has been used to detect planet signals in active stars with different levels of activity.

The NEID spectrograph routinely collects ~250 high-resolution solar spectra per clear day to support research in mitigating the effects of stellar variability on RV surveys. NEID is complementary to other spectrographs taking Sun-as-a-star observations thanks to the combination of its instrumental precision and stability, large spectral grasp, and high cadence of solar observations. I propose to present early results to characterize the effects of granulation on RV measurements based on applying GP regression models to NEID solar observations. We employ physically-motivated GP kernels that account for pulsations and granulation ((Guo et al. submitted), short-lived active regions (Gilbertson et al. 2020) and finite exposure time effects (Luhn et al. submitted). Depending on progress, we may report additional characteristics (e.g., chromatic effects). The goals include stimulating ideas for further improvements in the GP model and discussing implications for the observing cadence of upcoming RV surveys.

Stellar activity poses one of the main obstacles for the detection and characterisation of exoplanets around cool stars, as it deforms the absorption line profiles of stellar spectra creating signals that can hide or mimic the presence of companions. A wide variety of techniques are being developed with the aim of disentangling signals of stellar and planetary origins. These range from the decorrelation or modelling of the activity signals in the radial velocity (RV) time series, to the RV extraction using carefully selected wavelength regions of the stellar spectrum which are less affected by activity. In this work, we evaluate the performance of a set of common spectroscopic activity indicators for a set of M dwarfs observed with CARMENES. We show that different indicators behave differently depending on the mass and activity level of the target star. We then use some of these indicators to study their correlation with RVs computed from single lines in the stellar spectra. We apply this to a small sample of stars observed with CARMENES and EXPRES. Based on this correlation, we classify lines according to their sensitivity to activity. This allows us to select differently affected lines and use them to recompute RVs for which we mitigate or enhance the activity signal to varying degrees. We also note that the same lines on similar stars show different sensitivities to activity. Our results highlight the wealth of information available in the stellar spectrum that can be used to better understand stellar activity effects.

Stellar activity can induce radial velocity variations that dilute or even mimic the signature of a planet. A widely recognized method for disentangling these signals is to model the radial velocity time series, jointly with stellar activity indicators, using Gaussian processes and their derivatives. However, such modeling is prohibitive in terms of computational resources for large data sets, as the cost typically scales as the total number of measurements cubed.

We present S+LEAF 2, a Gaussian process framework that can be used to jointly model several time series, with a computational cost that scales linearly with the data set size. This framework thus provides a state-of-the-art Gaussian process model, with tractable computations even for large data sets.

We illustrate the power of this framework by reanalyzing the 246 HARPS radial velocity measurements of the nearby K2 dwarf HD 138038, together with two activity indicators. We reproduce the results of a previous analysis of these data, but with a strongly decreased computational cost (more than two order of magnitude). The gain would be even greater for larger data sets.

Ref: https://ui.adsabs.harvard.edu/abs/2022arXiv220102440D/abstract

Radial-velocity (RV) measurements are impacted by stellar and instrumental systematics at the m/s or sometimes even lower level. These small amplitudes make it very challenging to probe the different signal contributions and correct for them. A lot of those effects, such as the ones induced by detector systematics or micro-telluric lines, can be much stronger at the spectrum level, as they can induce dozens of m/s effects locally on the spectrum, but then are averaged-out when compressing the information from the entire spectrum in a CCF or when measuring RV using template matching. Regarding stellar activity, the variation of the line shape due to for example the inhibition of convection induced by active regions is also averaged-out when combining spectral line with different weights in a CCF. I believe that to reach the extreme RV precision allowing the detection of Earth-twins, we need to focus on correcting for instrumental systematics at the spectrum level. In addition, mainly for stellar activity systematics, we need to investigate different ways of gathering the information of spectral line together based on stellar physics arguments rather than only photon-noise and RV content arguments, as it was done so far. The goal of this discussion is therefore to reflect on future techniques that we could use to reach an extreme precision in RV.

Exoplanet detection with precise radial velocity (RV) observations is currently limited by spurious RV signals introduced by stellar activity (i.e. faculae, starspots). Here we show that machine learning techniques such as linear regression and neural networks can significantly remove the activity signals (primarily starspots/faculae) from real center-of-mass RV shifts. Many EPRV efforts have focused on carefully filtering out activity signals in time using Gaussian process regression. Instead, we separate activity signals from true center-of-mass RV shifts using only changes to the average shape of spectral lines, and no information about when the observations were collected.

We have tested our machine learning methods on both solar observations from the HARPS-N Solar Telescope and extrasolar observations from EXPRES and HARPS. For the solar observations, we find that these techniques can successfully predict and remove stellar activity and reduce the RMS by a factor of ~1.7 or about 40%. For extrasolar observations from EXPRES, we found a similar reduction in RMS for the most active stars in our sample. Lastly, we were able to successfully apply our methods to reveal the mass of K2-167, a planet which was first found using the transit method in 2015 and where stellar jitter previously limited our ability to measure its mass. This promising result inspires us to apply these or similar techniques to solar-type (FGK) stars, help measure masses of planets, and eventually help us detect habitable-zone Earth-mass exoplanets.

Well studied young exoplanets have the potential to guide our understanding of planet evolution and formation while taking us from a snapshot of mainly mature worlds to a plethora of exoplanets at all evolutionary stages. Despite their scientific value, young exoplanets remain elusive, mainly because of the violently active nature of their host stars. Recently, several studies have targeted relatively young planet-hosting systems, but getting precise masses of young planets is still challenging and extremely expensive in terms of telescope time e.g. AU Mic (22 Myr, Zicher+2022, Wittrock+2022, Szabó+2022, Klein+2021, Plavchan+2020 to mention only a few), V1298 Tau (20 Myr, Suárez Mascareño+2021), TOI-1201 (600-800 Myr, Kossakowski+2021), and K2-100 (~750 Myr, Mann+2017, Barragán+2019).

I propose to facilitate and chair a 1hr discussion session with the aim of identifying the main obstacles and potential solutions towards unbiased RV-detections and robust masses of young exoplanets. This discussion would intersect perfectly between the main topics of the workshop: data analysis, pipelines and instrumentation, and I forsee we could get a great exchange of knowledge, experience and ideas going.

Using Sun-as-a-star observations from the NASA/NSF NEID precision radial velocity (RV) facility, we investigate a variety of magnetic activity metrics to isolate activity signatures and improve the quality of the measured solar RVs. I will present an overview of recent work to diagnose and model solar activity using both ground and space-based data from the Solar Dynamics Observatory (SDO) and the NEID instrument. I will discuss our recently released SDO analysis pipeline, which provides the community with an accessible tool for measuring a wealth of valuable solar activity data products, as well as our initial results in identifying specific activity signatures in the NEID solar spectra using a variety of analysis tools, including line-by-line analyses and more classical line index measurements.

Combining the large collecting area of the VLT, an exquisite spectral fidelity and high efficiency, the ESPRESSO spectrograph is opening a new parameter space in exoplanet detection with radial velocities. As part of the follow-up of transiting planets and the blind search GTO programs, several stars are routinely observed with an RV precision at the cm/s level. These measurements are also revealing clear stellar activity signals contaminating the spectra.

In this talk, I will present the methods we are using to derive precise radial velocities and to disentangle activity from planetary signals, using Gaussian processes to extract information from activity indicators and exploiting the large spectral coverage of ESPRESSO. These methods have already allowed for the detection of some of the lightest exoplanets around active M dwarfs like L98-59 and Proxima, and are paving the way towards finding Earth-mass planets in the habitable zones of other nearby stars.

The RVx project (RVxTESS.com) uses simultaneous photometry and radial velocity observations to investigate strategies on mitigating stellar jitter. This talk will present our latest results on modeling the stellar oscillation and granulation signals on two different subgiants observed by Kepler/K2 + Keck/HIRES and TESS + Magellan/PFS. We'll share our lessons learned so far and some current explorations to make progress in battling the stellar jitter terms caused by asteroseismic signals.

In this tutorial, I will present some of the technical and computational challenges faced when using Gaussian Process models to analyze astronomical data in general, and (E)PRV datasets more specifically. I will discuss modeling decisions (including kernel and mean function choices) and how these decisions can be traded off against the computational tractability of the analysis. The second half of the tutorial will be an interactive introduction to the available open source tools, and an exploration of how to implement the discussed models for real RV applications.

Stellar magnetic activity induces distortions in the absorption line profiles of solar-like stars. Those produce radial velocity signals which highly hamper the search for habitable zone Earth-like planets. With the advent of state-of-the-art velocimeters such as ESPRESSO, EXPRES or HARPS3, it becomes crucial to come up with efficient and robust methods to get rid of the activity contributions while preserving signatures of planetary origin. In the last few years, innovative methods to exploit the wealth of information present in line profiles/spectra rather than just radial velocity points have shown promising results in this venture.

I will present the application of some of these methods to the high-precision solar line profiles collected with the HARPS-N spectrograph over the last 6 years. On the one hand, data-driven methods based on the separation of shift- and shape-driven variations in the line profiles provide a fast and flexible framework along with a set ancillary coefficients that can be included in multi-dimensional Gaussian process analyses. On the other hand, more physically-based approaches like Doppler Imaging, applicable to slowly rotating stars provided that their rotational cycle is densely sampled, might provide a reliable complementary approach as well as information on the physical properties of stellar magnetic activity.

A joint NASA-NSF initiative focusing on the RV variability of FGK stars ran in 2019-2020 and outlined future requirements needed to enable the detection and characterisation of very low mass planets around Sun-like stars. Although originally a US initiative, many researchers outside the US were involved -- especially in the stellar variability working group. The final report of this initiative concluded that international efforts are needed to achieve small planet characterisation, such as a coordinated global coordination networks and the funding of stellar variability research as it was identified as the biggest challenge to detecting Earth analog exoplanets. This is in keeping with a recent STFC Astronomy Advisory Panel White Paper advocating for a UK Stellar Variability and EPRV roadmap that would liaise with other international bodies. The session herein will focus on potential EPRV road maps for the 2020s/30s, with an emphasis on building sustained collaborative networks between the various research groups around the world that are tackling challenges relevant to achieving 10 cm/s RV precision.

Gaussian Process (GP) regression has become an increasing popular data analysis tool in stellar and exoplanet astronomy, yet questions remain as to the extent to which GP hyperparamters relate to physical stellar properties. I will present the results of tests of GP regression with popular kernels on a simulated stellar lightcurves and radial velocity times series, with the aim of determining the correlation, if any, between physical properties of a star, and the recovered GP hyper-parameters. I also explore the differences in the recovered GP hyper-parameters between light curve data and ‘perfectly’ sampled radial velocity data.

Tremendous progress has been made over the last decades in the instrumental precision of radial velocity instruments. In order to detect and measure the mass of temperate terrestrial planets, radial velocity measurements must meet extremely demanding requirements for long-term precision in order to be successful. Much effort and progress has been devoted to the critical task of reducing stellar activity, but mitigation of instrumental systematics has received less attention and are often dealt with by pipelines that are generally less accessible to the broader community. Next-generation extreme precision instruments have recently been commissioned or are coming online in the near future, and to fully utilize the potential of these instruments and novel calibration sources like Laser Frequency Combs (LFCs), advanced mitigation of instrumental systematics and sophisticated reduction pipelines are vital. This topic of this session is to discuss techniques and ideas that could promote mitigation of instrumental systematics to ensure centimeter per second stability over year-long timescales.

In many astronomical applications, Gaussian process inference tends to be relatively insensitive to the exact functional form one chooses for the kernel, meaning one is free to choose from a set of well-studied, widely-implemented, few-parameter kernel functions. However, in the case of EPRV exoplanet searches, the coherence of the stellar signal over timescales comparable to planetary orbits means the functional form of the kernel can matter quite a bit. In this talk I will discuss ongoing work to construct a new GP kernel for stellar variability from first principles, which has the potential to greatly improve our ability to discriminate between stellar and planetary signals in timeseries spectroscopy.

In this tutorial I will introduce starry, a framework for modeling stellar and planetary timeseries data, and starry_process, its probabilistic cousin. While both codes were created with photometric timeseries data in mind, they have since evolved and can be used to model both radial velocity and spectral timeseries data. Both packages are open source, thoroughly documented, and implemented to maximize both efficiency and ease-of-use. Both starry and starry_process are also actively welcoming community contributions!