Bridging the Gap Between Radial Velocity and Direct Imaging

UNIVERSIDAD DE CHILE FACULTAD DE CIENCIAS F´ISICAS Y MATEMATICAS´ DEPARTAMENTO DE ASTRONOM´IA

SAFARI: BRIDGING THE GAP BETWEEN RADIAL VELOCITY AND DIRECT IMAGING

TESIS PARA OPTAR AL GRADO DE DOCTOR EN CIENCIAS, MENCION´ ASTRONOM´IA

BLAKE M PANTOJA

PROFESOR GU´IA: JAMES JENKINS

MIEMBROS DE LA COMISION:´ JULIEN GIRARD PAULINA LIRA TEILLERY PATRICIO ROJO RUBKE

Este trabajo ha sido parcialmente ﬁnanciado por la beca de doctorado nacional de CONICYT No. 21161783 del a˜no2016

SANTIAGO DE CHILE 2019 RESUMEN DE LA MEMORIA PARA OPTAR AL TÍTULO DE DOCTOR EN CIENCIAS, MENCION´ ASTRONOMÍA POR: BLAKE M PANTOJA FECHA: 2019 PROF. GUÍA: JAMES JENKINS

SAFARI: BRIDGING THE GAP BETWEEN RADIAL VELOCITY AND DIRECT IMAGING

SAFARI: DISMINUYENDO LA BRECHA ENTRE VELOCIDADES RADIALES E IMAGENES´ DIRECTAS

El método de velocidades radiales ha sido muy productivo en la búsquedade exoplanetas, comen- zando con el descubrimiento de 51 Peg b en 1995 hasta los aproximadamente 800 planetas detec- tados por este método al d´ıade hoy. Aunque este método ha sido útil en muchos descubrimientos, tambiéntiene limitaciones, como la gran cantidad de tiempo necesario para realizar detecciones de compañeroslejanos a su estrella y la imposibilidad de medir la inclinaciónde las órbitas.As´ı,esta técnicasólopermite medir la masa m´ınimade un compañero.El método de imágenesdirectas, por otro lado, es mássensible a la detecciónde compañerosa distancias mas lejanas de sus estrellas primarias, debido al gran contraste de luminosidad entre compañerosde baja masa y sus estrellas. Aúncuando este método ha producido relativamente pocos descubrimientos, nos da la capacidad de mapear la orbita astrométricay medir la inclinación.De esta forma, podemos combinar ambos métodos complementarios e investigar el régimenmasa-separaciónentre sus detecciones.

Surveys que utilizan la técnicade velocidades radiales, a menudo presentan evidencia de periodos largos, en donde es posible que exista un compañerolejano que no se ha movido lo suficiente para poder determinar su órbitade manera precisa. En estos casos, podemos estudiarlos utilizando imágenesdirectas. Para investigar el concepto de combinar estos dos métodos, utilizamos dichos periodos largos provenientes de surveys que utilizan velocidades radiales para luego observarlos utilizando instrumentos con ópticaadaptativa (cámarasy espectrógrafos)como VLT-SPHERE, Magellan AO y VLT-SINFONI. Se comprobóel funcionamiento de este método en el descubrimiento de un compañeroestelar de tipo M3 a la estrella tipo Sol, rica en metales HD86006 y en el descubrimiento de dos enanas ultra fr´ıasde transiciónM/L orbitando la estrella de tipo M tem- prano que posee un planeta GJ 3634. De esta forma, podemos utilizar espectroscop´ıapara precisar sus caracter´ısticas, y con suficiente movimiento orbital, podemos utilizar astrometr´ıa para esti- mar masas dinámicas. Estos objetos pueden ser usados como referencia para estudiar la relación masa-luminosidad-metalicidad, ademásde servir como limite para los modelos evolutivos.

Para investigar como mejorar el contraste de nuestras imágenesdirectas, analizamos los efec- tos de dos técnicasde post-procesamiento: imágenesdiferenciales angulares (”angular differential imaging”; ADI), donde podemos aprovechar la rotacióndel campo durante una observaciónpara hacer una colecciónde datos referenciales con los que podemos sustraer la observaciónde ciencia; e imágenesdiferenciales de referencia (”reference differential imaging”; RDI), donde una similar (pero distinta) PSF puede ser usada como referencia en el objeto de interés.Encontramos que RDI produce el mejor resultado a separaciones pequeñasde la estrella. Podemos usar estas técnicasjunto con nuestro survey de imágenessin detecciónde compañeroscomo candidatos para instrumentos de futuros telescopios espaciales y terrestres para buscar pequeñosplanetas en órbitas cercanas en estrellas viejas similares al Sol.

i ii RESUMEN DE LA MEMORIA PARA OPTAR AL TÍTULO DE DOCTOR EN CIENCIAS, MENCION´ ASTRONOMÍA POR: BLAKE M PANTOJA FECHA: 2019 PROF. GUÍA: JAMES JENKINS

SAFARI: BRIDGING THE GAP BETWEEN RADIAL VELOCITY AND DIRECT IMAGING

Radial velocities is a very productive method in the search of exoplanets starting with the landmark discovery of 51 Peg b in 1995 to the about 800 planets detected by the method today. Even though it has been advantageous in making many detections, the method has a few limitations as well, namely that extended amounts of time is needed to make detections of companions far from the star and that it is inherently impossible to constrain a companion’s inclination, thus limiting the technique to measuring a companion’s minimum mass. Direct imaging, on the other hand, is more sensitive to the detection of companions distant from their primary stars, owing to the huge contrast in luminosity between close low-mass companions and their stars. While the method has been responsible for relatively few exoplanet discoveries, it provides the ability to map an astrometric orbit and ﬁnd an inclination. We can thus combine these complementary methods and study the mass-separation regime between their detections.

Radial velocity surveys turn up a number of long-period or linear trends where there is distant companion that has not moved enough for precise constraint of its orbit. As we know there to be a distant companion, we can follow up with imaging to search for it. To further investigate this concept of combining these methods, we used long-period trends as targets from planet-search radial velocity surveys to image with the adaptive optics imagers and spectrographs VLT-SPHERE, Magellan AO, and VLT-SINFONI. This concept is proved to work with the discoveries of an M3 companion to the metal-rich Sun-like star HD 86006 and of two M/L transition ultracool dwarfs orbiting the planet-hosting early M dwarf GJ 3634. Following these detections, we can use spectroscopy to constrain their characteristics, and with enough orbital motion, we can use astrometric measurements to constrain dynamical masses. These objects can be used as benchmarks to study the mass-luminosity-metallicity relation, as well as observational constraints on evolutionary models.

To investigate into improving the contrast performance of our direct images, we studied the effects of the two post-processing techniques angular differential imaging (ADI), where the field rotation during an observation can be used to make a reference data set with which to subtract the science observation, and reference differential imaging (RDI), where a similar but separate point spread function can be used as a reference to the science target. We found RDI to provide the best result at short separations from the star. We can use the techniques along with our sample of imaging non-detections as targets for future space-based and ground-based imagers to search for close low-mass planets around old Sun-like stars.

iii iv Agradecimientos

First of all, I would like to thank my two advisors from Calánand ESO, James and Julien, who provided me with their expertise, knowledge, advice, and support. James helped me orient in a totally new country when I started this program while also being patient with my learning a new field of study, and Julien trained me firsthand in a more technical area of astronomy involving instrumentation in an observatory setting. Without them, this thesis and my PhD as a whole would not have been possible.

I would also like to thank my committee including Paulina and Pato and I appreciate their taking the time to be a part of this work. I also want to thank my collaborators Arthur, Graeme, Mat´ıas,Ga¨el,Bartek, Mikko, Anne-Marie, Julien, and others. Thank all of you for all the support in my work. Thank you to Dr. Williger and Dr. Morrison for your support in my undergraduate studies. Y grac´ıasa Natalie, Marta, y Gissela por su paciencia en hacer las cosas administrativas. Y grac´ıasSra. Mari por el almuerzo diario.

I would like to thank my Mom and Dad. They have each provided me the patience, perseverance, and love of learning that led me to go this far in my studies. Thank you for your support in my whole education, without which I could not have studied astronomy at the doctoral level. I also thank my brothers Corey and Landon and sister Marissa for being there with me while growing up and now. I also thank my aunts, uncles, cousins, grandpa y familia en Per´u.Y muchas gracias a mi abuelita. Es por ella que quer´ıaestudiar en un pa´ıs hispanofono.

Being so far away from home, I have had the chance to meet new friends here in Chile as well as keep up with old friends. Thank you to my friends from other countries who also came to Chile including Sudeep, Valentin, Robert, Nina, Armin, Nathen, Carla, Jonathan, Alessandro, and Murat who really made a family away from home for me with unforgetable trips and times spent together. Muchas grac´ıasa mis amigos y compañerosde mi generación Mat´ıas (tambiéngracias por la ayuda con este resumen en español!),Jorge, Juan, Seba, Grecco y Juanpi por su apoyo en mi primer año y el resto de mi doctorado y paciencia con mi españolmalo. ¡Sin ustedes no ser´ıaposible! Tambiéna mis otros amigos acáen Chile incluyendo Ricardo, Maritza, Paula, Pablo, Tere, José,Pato, Daniela, Bica, Gabriela, Gus, Seba, Piera, Mia, Alexander y Cambrilo por su consejo y apoyo en un pa´ısextranjero para m´ı. I want to thank my great friends back home who have stuck with me and gave me so much support through this long program away including Tommy, Doug, Dylan, Sadaf, Thomas, and Meena. I lastly want to also thank everyone else who I have met along the way in this unforgettable experience. I wish everyone the best, and each one of you is welcome

v to visit me in Lewisburg or Louisville or wherever I am!

I acknowledge support from CONICYT National Doctoral Scholarship Grant No. 21161783. Observations in this thesis are based on programs of the European Southern Observatory and time allocated by the Chilean National Telescope Allocation Committee for use of the Mag- elland telescopes. I would like to acknowledge the use of the SPHERE Data Center for their help in the reduction of the SPHERE data. I would like to thank Pierre Kervella for providing his code for mapping the trajectory of a star based on proper motion and parallax.

vi Contents

Contents vii

List of Tables ix

List of Figures x

1 Introduction 1 1.1 Red Dwarfs - The Lowest Mass of Main-Sequence Stars ...... 1 1.1.1 Characteristics of Red Dwarfs ...... 1 1.1.2 Broader Perspective ...... 3 1.2 Brown Dwarfs - The Bridge between Stars and Planets ...... 4 1.3 Methods of Detecting Binary Stars and Exoplanets ...... 8 1.3.1 Visual Binaries ...... 9 1.3.2 Spectroscopic Binaries ...... 10 1.3.3 Perspective on Binary Stars ...... 11 1.4 Extrasolar Planets ...... 12 1.4.1 Radial Velocities ...... 13 1.4.2 Direct Imaging ...... 14 1.4.3 Combining Radial Velocities and Direct Imaging ...... 15 1.5 Technique of Direct Imaging with AO Systems or High Contrast Imaging of Exoplanets ...... 16 1.5.1 AO System ...... 16 1.5.2 Diﬀerential Imaging and Post Processing Techniques ...... 20 1.6 Thesis Outline ...... 22

2 SAFARI I: A SPHERE discovery of a super metal-rich M dwarf companion to the star HD 86006 24 2.1 Introduction ...... 25 2.2 Target Selection, Observations and Reduction ...... 27 2.2.1 Radial Velocity Observations ...... 28 2.2.2 Direct Imaging Observations ...... 32 2.2.3 Spectroscopic Reduction ...... 37 2.2.4 HST Astrometric Calibration ...... 38 2.3 Results ...... 38 2.3.1 HD86006 Companion Detection ...... 38 2.3.2 Spectral Indices ...... 40 2.3.3 Model Comparison ...... 41

vii 2.3.4 Model Fitting ...... 47 2.3.5 HD 90520 ...... 50 2.4 Summary and Conclusions ...... 50

3 SAFARI II: A Pair of Ultracool Dwarf Companions to a Planet-Hosting Red Dwarf 55 3.1 Introduction ...... 55 3.1.1 Sub-stellar Objects ...... 55 3.1.2 Radial-velocity Measurements ...... 56 3.2 Observations and Reduction ...... 57 3.2.1 HARPS Radial Velocity Observations ...... 59 3.2.2 Direct Imaging and Spectroscopic Observations ...... 61 3.2.3 SPHERE Observation ...... 62 3.2.4 SINFONI Observation ...... 62 3.3 Results ...... 64 3.3.1 Detection of Companions to GJ 3634 ...... 64 3.3.2 Spectral Template and Model Comparison ...... 66 3.3.3 Spectral Index ...... 66 3.3.4 Isochrone Fitting ...... 74 3.3.5 Orbital Fitting ...... 76 3.4 Summary and Conclusions ...... 78

4 Achieving High Contrast with rapid Reference Star Diﬀerential Imaging on SPHERE at 1.6 µm 81 4.1 Introduction ...... 81 4.2 Observation and Reduction ...... 83 4.2.1 ADI Reduction ...... 85 4.2.2 RDI Reduction ...... 86 4.2.3 Contrast Curve Generation ...... 86 4.3 Results ...... 86 4.3.1 Tests to Compare RDI and ADI ...... 87 4.3.2 Full Data Set ...... 87 4.3.3 Cut Frame Set ...... 88 4.3.4 Position Dependece on SNR ...... 96 4.3.5 Quality of Retrieved FCs ...... 99 4.4 Discussion ...... 103 4.5 Conclusions ...... 105

5 Conclusions and Future Work 106

Bibliography 110

viii List of Tables

1.1 Physical Characteristics of M dwarfs (Reid & Hawley, 2005; Table 1 from Kaltenegger & Traub, 2009) ...... 2 1.2 Stellar multiplicity fractions. Table 1 from Cort´es-Contreras et al. (2017). References found in article...... 9

2.1 Stellar characteristics of the stars HD 86006 and HD 90520...... 28 2.2 Radial velocity data for HD 86006...... 29 2.3 Radial velocity data for HD 90520...... 30 2.4 Observations Table ...... 36 2.5 Photometry and Astrometry of HD 86006B ...... 38 2.6 HD 86006 System Characteristics ...... 39 2.7 HD 86006B Model Physical Characteristics ...... 45 2.8 HD 86006B Keplerian Elements ...... 49

3.1 Characteristics of GJ 3634 and GJ 3634b...... 58 3.2 Radial velocity data for GJ 3634...... 60 3.3 Observations Table ...... 64 3.4 Photometry and Astrometry of GJ 3634 ...... 64 3.5 GJ 3634 Keplerian Elements ...... 78

4.1 Observations Table ...... 84

ix List of Figures

1.1 Plot of K5, M3.5, and M9 dwarf stars in the red part of the optical wavelength range with absorption bands marked. Fig. 5c of Kirkpatrick et al. (1991). .4 1.2 Plot of SEDs from optical to mid-infrared of M dwarf Gl101A (blue), L dwarf 2MASS 1507-16 (red), T dwarf 2MASSJ 0559-14 (dark purple), Y dwarf WISE 0350-5658 (light purple), and planetary mass objects in green. In black dots are shown the BT-Settl model spectra. Fig. 1 from (Allard, 2014). See article for spectra references...... 6 1.3 Plot of the color-magnitude diagram for late M, L, and T dwarfs. Fig. 6 from Gauza et al. (2019)...... 6

1.4 Plot of the evolution of Teff over time up to 10 Gyr. The blue curves represent low-mass stars, the green curves represent intermediate deuterium burning brown dwarfs, and the red curves represent low-mass non-deuterium burning brown dwarfs. The gold-colored dots are where 50 % of the deuterium supply has been burned. The magenta dots are where 50 % of lithium has been burned. The dashed lines are where the M, L, and T spectral types are approximately divided. Fig. 8 of Burrows et al. (2001)...... 7 1.5 Plot of log(luminosity) compared to mass. The mass-luminosity relation for main-sequence stars is clearly visible. Fig. 5 of Torres et al. (2010)...... 12 1.6 Plot of extrasolar planets discovered with points colored by detection method. http://exoplanets.org/ The red points are from transits, the blue points are from radial velocities, the green points are from microlensing, and the yellow points are from direct imaging...... 13 1.7 Images of directly imaged planetary mass or exoplanet companions to the stars as follows (right to left): DH Tau b, UScoCTIO 108 b, 51 Eri b, AB Pic b, β Pic b, Fomalhaut b, 2M0122-2439 b, HR 8799 bcde, RXJ1609 b, HD 106906 b, κ And b , HIP65426 b, CD-35 2722 b, GJ 504 b, HD 95086 b, GU Psc b, HIP 77900 b, HR 6037 b, 2M1207 b and 2M0143AB b. Figure extracted from Fig. 4 of Chauvin (2018a)...... 15 1.8 (A) Image image of a model airy disk and rings. An observation of a star with a perfect telescope system and no atmosphere would produce this image. (B) Surface plot of the airy disk showing the PSF. Note that the central disk holds the majority of light compared to the fainter rings. Credit: Fig. 3 of Schmolze et al. (2011)...... 17 1.9 Schematic of the AO loop. Fig. 1 of Rigaut (2015) ...... 18 1.10 Light path through a Lyot coronagraph. Credit: http://lyot.org/background/ coronagraphy.html adapted from Sivaramakrishnan et al. (2001) . . . . . 19

x 1.11 Figure showing Angular Diﬀerential Imaging, with subtraction of the science frames by the median reference, derotation, and stacking of the frames. Credit: Christian Thalmann ...... 21

2.1 Plot of radial velocity values for HD 86006. The closed data points represent data from HARPS, while the open points represent data from CORALIE. Note the linear trend without inflections making orbital characteristics difficult to determine...... 33 2.2 Plot of radial velocity values for HD 86006. The closed data points represent data from HARPS, while the open points represent data from CORALIE. Note the linear trend without inflections making orbital characteristics difficult to determine...... 34 2.3 The left plot shows an image of HD 86006 with SPHERE IRDIS and the right is from MagAO. The M dwarf companion can be seen to the upper right of the star at a separation of „26 AU from the primary...... 39 2.4 Plot of the astrometric positions of the companion to HD 86006 from our data with SPHERE and MagAO, along with a model trend of where the companion would be if it were a background source ending in November 2016 (red curve). The astrometric points are given for each epoch in different colors for the instrument and data, as shown in the legend...... 40 2.5 The best fit 3600 K model (black) from the BT-Settl to an SED from the IFS of HD 86006 (blue). This model also used a metallicity of +0.3. We also show a 3300 K, +0.5 metallicity model as it visually matches very well at the bluer wavelengths...... 42 2.6 The comparison BT-Settl spectra (3200 K, 3300 K, 3400 K) in black with the LSS spectrum in red. The best fit using the maximum likelihood between all of the spectra was the one at 3300 K which also appears the best visually. For clarity, we have shown the spectrum with the water absorption feature removed. 43 2.7 The LSS spectrum of HD86006B in red compared with the SpeX spectra in black. The Gl388 spectrum is visually a good fit to the LSS spectrum. An M2.5V (Gl381) and an M3.5V (Gl273) spectrum are also shown to compare. For clarity, we have shown the spectrum with the water absorption feature removed...... 44

2.8 The isochrones for BT-Settl models at masses varying from 0.03 Md to 0.6 Md and ages from 1 Myr to 12 Gyr as labeled. In black, we give the SPHERE H2 photometry of HD 86006B and the range of temperatures by taking the average of the temperatures found from SED ﬁtting and spectral indices. We also show various solar neighborhood M3 dwarfs with their photometries and their temperatures and metallicities (for which we divide into two groups, with green representing low metallicity and blue representing high metallicity) from Rojas-Ayala et al. (2012)...... 46 2.9 Corner plot for the MCMC astrometry and radial velocity ﬁts. We show the 2D posterior probability distributions for each parameter with each of the others to demonstrate how well they correlate along with the 1D distribution for each at the top of their set of plots. Vx, Vy, and Vv are paramters representing the systematic noise or jitter for the x and y astrometric dimensions and the radial velocities, respectively...... 51

xi 2.10 Plot of the fit from the MCMC code to the RV data. The red points shown with their error are the observed radial velocities and are the same as the points show in Table 3.2. The black curves represent one fit from the MCMC chain. As all fits after the burn-in were similar for the radial velocities, they nearly appear as a line in this plot...... 52 2.11 Fit to the astrometric points from SPHERE and MagAO from the MCMC code. The red points represent the position of observations, while each black line is the fit output from the MCMC run. The radial scale is in AU. . . . . 53 2.12 The 5σ contrast performance curve over the separation from the SPHERE IRDIS SV observations of HD 90520. We also show the instrument’s performance as a companion mass detectability limit (right axis) using the AMES- COND model to convert between magnitude and mass. We show where the directly imaged planets β Pic b (Lagrange et al., 2010; Bonnefoy et al., 2013), HR 7899 bcde (Marois et al., 2008; Zurlo et al., 2016), and 51 Eri b (Macin- tosh et al., 2015) would lie with respect to the contrast performance of these observations. We also highlight the position of our detection of HD 86006B and of the expected stellar to substellar transition for old objects...... 54

3.1 Plot of the radial velocity data over time for GJ 3634. In grey, the ﬁt model from EMPEROR is shown including the short period signal from the planet GJ 3634b and the linear trend acceleration...... 59 3.2 All images are oriented with north facing up and east to the left, and all three are shown in log scale. (a) MagAO image of GJ 3634 saturated in the Ks band with the companions seen lower in the image. (b) SPHERE IRDIS image of GJ 3634 in the H2 band with the companions visible in the lower part of the image. (c) SINFONI image of the GJ 3634 B companions at 2.2 µm in the 0.8” ﬁeld of view...... 65 3.3 Plot of the astrometric points of the companions to GJ 3634 with the track (in black ending on May 2017) that would be followed if they were distant background objects, for which they do not. The Gaia detection is that of the blended companions, whilst those of MagAO and SPHERE in 2017 allowed for separate detections of the two companions...... 65 3.4 Plot comparing the SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5...... 67 3.5 Plot comparing the SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2...... 68 3.6 Plot comparing the H band SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5...... 69 3.7 Plot comparing the K band SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5...... 70 3.8 Plot comparing the H band SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2...... 71 3.9 Plot comparing the K band SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2...... 72

3.10 Plot comparing the absolute H band magnitude to the Teff of the two components (blue and green) of GJ 3634 B, along with the BT-Settl isochrones. In black, we show the points on the isochrone plot from Baron et al. (2015). . 75

xii 3.11 Plot comparing the mass ratio and separation for low-mass stars and brown dwarfs in binary conﬁguration. We show companions to stars also hosting a planet (red, Mugrauer et al., 2006b; Lodieu et al., 2014), directly imaged planets (yellow, Bowler, 2016), wide red dwarf and brown dwarf companions (black, Baron et al., 2015), and imaged substellar objects to young stars (purple, Chauvin, 2018a)...... 77

3.12 Corner plots showing the posterior distributions for the orbital ﬁt of GJ 3634. log(Vx) is the jitter along the x direction...... 79

3.13 Plots demonstrating (left) the MCMC astrometric orbital ﬁts in position, (mid- dle) position angle vs. time, and (right) separation vs time with the red points representing the data...... 80

4.1 Images of 55 Eri A (left) and B (right) under the SPHERE coronagraph, where the axes are given in milliarcsec. 55 Eri A is used the science target and 55 Eri B is used as the reference target for RDI...... 84

4.2 Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast...... 89

4.3 Images of 55 Eri A with an injected 110 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast...... 90

4.4 Images of 55 Eri A with an injected 120, 298, 306, and 404 mas fake companions at 5 diﬀerent angles after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast...... 91

xiii 4.5 Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (first half of set). (f) Plot of 5σ contrast performance curves for the five methods and the fake companion position and contrast...... 93 4.6 Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (skipping every other frame of set). (f) Plot of 5σ contrast performance curves for the five methods and the fake companion position and contrast...... 95 4.7 Maps of SNRs from annular ADI (left) and annular RDI (right) reductions of fake companions spaced in resolution elements (1 λ/D) with the noise being measured by the standard deviation of apertures at the same radius as the fake companions with a Student-t correction for low sample statistics as in Mawet et al. (2014). Two contours in white are shown for each image with innermost one being the 5σ level and outermost one being 25σ...... 97 4.8 Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (partial set). (f) Plot of 5σ contrast performance curves for the five methods and the fake companion position and contrast...... 98 4.9 Map of FCs injected into 55 Eri A at separations of less than 0.3 arcsec (left) and greater than 0.3 arcsec (right). The size of the points are adjusted for the contrast magnitude given for the FC...... 100 4.10 Histogram plots of the difference fraction between recovered fluxes and given fluxes of the FCs using the vip.phot.detection algorithm with orange being from aADI+PCA and blue from aRDI+PCA (left) and of the difference of the x measurements representing the difference in astrometry for FCs within 0.3 arcsec...... 100 4.11 Histogram plots of the difference fraction between recovered fluxes and given fluxes of the FCs using the vip.phot.detection algorithm with orange being from aADI+PCA and blue from aRDI+PCA (left) and of the difference of the x measurements representing the difference in astrometry for FCs outside 0.3 arcsec...... 101

xiv 4.12 Plot of 5σ contrast and separation for recovered FCs by the RDI and ADI methods for FCs within 0.3 arcsec (left) and outside of 0.3 arcsec (right). Blue points represent RDI detections and red points represent ADI detections, with the red points hiding blue points discovered by both methods to indicate where RDI shows detections where ADI does not. RDI shows better results at close separation to the star...... 102 4.13 Plot of 5σ contrast and separation for recovered FCs by the RDI and ADI methods for FCs within 0.3 arcsec (left) and outside of 0.3 arcsec (right). Blue points represent RDI detections and red points represent ADI detections, with the blue points hiding red points discovered by both methods to indicate where ADI shows detections where RDI does not. ADI shows better results at the 0.7 to 0.9 arcsec range...... 103 4.14 Plot of 5σ contrast performance and separation the annular RDI and ADI methods with points plotted for previously discovered exoplanets by direct imaging...... 105

5.1 Mass-magnitude (V) plot for M dwarfs that have dynamical masses and metallicities measured. The curves correspond to the theoretical models from Allard et al. (2012) with an age of 3 Gyr and metallicities of [Fe/H] = -0.5, 0.0, and 0.3 (The curves with metallicities of 0.0 and 0.3 are close to each other). . . 108 5.2 Plot of known exoplanets with sensitivities overlaid from current instruments such as SPHERE and GPI, space instruments such as JWST, and future ELT instruments. Fig. 5 of Chauvin (2018b)...... 109

xv xvi Chapter 1

Introduction

1.1 Red Dwarfs - The Lowest Mass of Main-Sequence Stars

Red dwarfs or dwarf stars with spectral class M are the lowest-mass and most abundant of stars, which gives them special importance when trying to understand the Universe. We can use their position and abundance in the galaxy to study galactic formation and kinematics and the dark matter of spiral galaxies (Bochanski et al., 2007), discover exoplanets, and study stellar evolution.

1.1.1 Characteristics of Red Dwarfs

Originally from Pickering (1890), stars were classiﬁed alphabetically by the strength of hydrogen lines in their spectra. K stars had stronger hydrogen Balmer lines (from absorption of photons of energy equal to the levels between n = 2 and n ą 2) and O stars showed a lack of them. Cannon & Pickering (1901) reordered the the stars’ spectral types, using the same letters as before after dropping some, by temperature (which was related to Balmer lines nonlinearly). This is similar to the order used today where M stars come as the reddest of stars.

Among stars, M dwarfs are the least massive at 0.075 to 0.6 Md (Benedict et al., 2016), are the coolest with Teff s between 2300 and 3800 K, are the smallest with radii of 0.08 to 0.62 Rd (Reid & Hawley, 2005), and they have the reddest colors with V-H between 3.6 and 8.4 (Pecaut & Mamajek, 2013),. In Table 1.1.1, we show the physical characteristics of early to late M dwarfs. M dwarfs are a diverse class of stars as seen by the given parameters (especially when compared to FGK stars) and by their structure, which we describe in the following section. Their diversity is evident in the evolution and structure of M dwarfs. Like all other stars, mass primarily deﬁnes its properties as a main-sequence star does not change in luminosity or temperature.

1 Table 1.1: Physical Characteristics of M dwarfs (Reid & Hawley, 2005; Table 1 from Kalteneg- ger & Traub, 2009)

SpType Dwarf T (K) R (Rsun) Mass (Msun) L/100 (Lsun) MV (mag) a(HZ) (AU) P(HZ) (hr) ∆T(HZ) (hr) ∆I/I (%) M0 3800 0.62 0.60 7.2 9.34 0.268 1571 5.37 0.022 M1 3600 0.49 0.49 3.5 9.65 0.190 1039 3.96 0.035 M2 3400 0.44 0.44 2.3 10.12 0.152 786 3.36 0.043 M3 3250 0.39 0.36 1.5 11.15 0.123 633 2.96 0.055 M4 3100 0.26a 0.20 0.55 12.13 0.075 401 2.06 0.124 M5 2800 0.20 0.14 0.22 16.0 0.047 238 1.50 0.209 M6 2600 0.15 0.10 0.09 16.6 0.030 147 1.07 0.372 M7 2500 0.12 0.09 0.05 18.8 0.022 98 0.78 0.582 M8 2400 0.11 0.08 0.03 19.8 0.019 81 0.69 0.69 M9 2300 0.08 0.075 0.015 17.4 0.013 46 0.43 1.31 Note. a The original reference Reid & Hawley, 2005 states 0.36, but should be 0.26 as shown.

M dwarfs form from the gravitational contraction of unstable cores (with masses greater than their Jeans mass) within molecular clouds into a protoplanetary disk with material falling into the center and becoming a protostar. The pre-main sequence star then con- tracts and brightens due to gravitational potential energy. Once the core has reached the appropriate temperature for hydrogen fusion, it has become a main-sequence M dwarf. One important aspect to the diversity of this spectral type is that at 0.35 Md, the radiative core that is seen in more massive stars, disappears and the star is fully convective (Chabrier & Baraﬀe, 1997). Having cooler cores than more massive stars with a higher density, these low- mass red dwarfs have a high opacity which leads to convection in the core. As the lifetime timescale of a main-sequence star scales as τ9M{L, it stands to reason that 0.1 Md will live roughly 100 times the lifetime of the Sun or about 1 trillion years (early M stars live about 5 times the age of the Sun; Ryden & Peterson, 2010.) Actually, the lifespan is about 10 times that length as its total convection makes for eﬃcient burning, therefore no red dwarfs in the Universe have evolved past the main sequence. The stars will, after such long timescales, gain a radiative core and get hotter (but will not evolve as red giants unlike more massive stars) before becoming white dwarfs at the end of their lives (Laughlin et al., 1997).

Hot stars are the best approximators of the blackbody (a body that absorbs all radiation) as the radiation is only absorbed by their ionized or neutral atomic lines. While these lines can significantly affect the spectra, they are relatively well studied. Cooler redder stars, on the other hand, are more influenced by the absorption of many molecular lines (as they have lower dissociation energies) blended with atomic lines, which are still not fully understood. M dwarfs were distinguished from earlier stars, such as the neighboring K dwarfs, by the presence of molecular TiO lines, and subtyped by the lines’ strength in the blue part of the spectrum (Morgan, 1938; Kuiper, 1942), as was available at the time. In Fig. 1.1, we show from Kirkpatrick et al. (1991) a comparison between a mid K star and two M dwarfs (M3.5 and M9) in the red optical bands (6300-9000 A).˚ At these wavelengths, some weak TiO bands are visible in the K5 star, besides at 7589, 7628 A˚ where the line is strong, while the visibility and strength (similar to the bluer wavelengths) increases with M spectral subtypes. Among other molecular bands that become visible at the later spectral types are VO at later spectral types as can be seen with the bands at around 7400, 7900, 8600 A˚ in the M9 star of Kirkpatrick et al. (1991), where it affects the pseudocontinuum. CaH shows strongest in the early M dwarfs, like that of the M3.5 star. At bluer wavelengths, CaOH appears at 5530 A˚ for later M dwarfs In near infrared bands, H2O plays a greater roles, which we show in

2 Chapter 2.

The difficulty in understanding the complex molecular absorption features of the atmospheres of cool stars, highlights our imprecision in measuring and understanding the metallicity and chemical compositions of these objects. The pseudocontinuum created by the TiO bands blended together makes the measuring of atomic lines used for metallicity measurement of more massive stars challenging for M dwarfs without using high resolution spectra of bright red dwarfs as in Woolf & Wallerstein (2005). Even so, we are able to approach this problem in a few ways. Bean et al. (2006) made metallicity measurements based on the fits of model spectra to the Fe I line. Bonfils et al. (2005), Johnson & Apps (2009), and Schlaufman & Laughlin (2010) made photometric-metallicity calibrations with ”metallicity-benchmark” M dwarfs in a system with a host FGK star where a metallicity is assumed to be the same in molecular cloud in which they formed. Rojas-Ayala et al. (2010) used the metallicity- dependent Na I and Ca I lines in the K band, where the spectrum is less affected by lines. Maldonado et al. (2015) used ratios of spectral features in the optical regime. Kuznetsov et al. (2019) studied the parameter using multiple metallicity-sensitive wavelength ranges of the optical spectrum. There are a variety of ways of finding the metallicity of low mass stars, and this search of new ones will only continue as we aim to improve on its precision to understand planet formation history.

The mass-luminosity relation (MLR) for low mass stars is the least well constrained of all stellar objects. (We explain more of the MLR in general in Sec. 1.3.3.) Similar to how having ”metallicity benchmarks” are important for calibrating metallicites, ”mass benchmarks” where we can use orbiting M dwarfs to constrain their masses are important for calibrating the MLR. The MLR is only as precise as the masses that describe it. Owing to its fundamental use in observational astronomy, the MLR at the red end of stars has been studied a few times (e.g. Henry & McCarthy, 1993; Delfosse et al., 2000; Benedict et al., 2016). The V band shows more intrinsic scatter when compared to the near-infrared bands, showing that the MLR is actually a mass-luminosity-metallicity relation, with other properties such as age and magnetism giving an effect (Delfosse et al., 2000; Benedict et al., 2016). At higher metallicity, the TiO and VO bands blanket more of the optical spectrum while emitting more flux in the near infrared while also lowering the bolometric flux. These two effects, show a significantly smaller difference in infrared luminosity than in the V band, demonstrating that there is not only one universal MLR for a star. Bonfils et al. (2005), with metallicity measurements to the infrared MLR derived masses, shows how precise masses, luminosities, and metallicities are needed in the V band. At high precision, metallicity would be a limiting factor to all of the MLRs, thus showing the importance of metallicity measurements of low mass stars as well as mass measurements.

1.1.2 Broader Perspective

The lowest of stars have a special interest for us in the study of exoplanets. Their low-mass, small, red characteristics lend themselves well to the major methods of detecting exoplanets, which are radial velocities, transits and imaging. On the other hand, their diﬃculty in constraining of characteristics (such as metallicity, age, radius, mass) bleeds into diﬃculty in good constraints on their planets. Therefore, a deeper understanding of the nature of these

3 Figure 1.1: Plot of K5, M3.5, and M9 dwarf stars in the red part of the optical wavelength range with absorption bands marked. Fig. 5c of Kirkpatrick et al. (1991). stars is critical to better understand their population of planets.

1.2 Brown Dwarfs - The Bridge between Stars and Planets

Brown dwarfs are divided into four spectral classes: M (at young age and high substellar mass later than M7), L, T, and Y. Similarly ultracool dwarfs covers all of these objects later than M7. For a brown dwarf at a mass just below the hydrogen burning limit ,e. g. 0.06 MJup, over its life it will evolve from being an M dwarf with Teff ą 2200K and age of younger than 3 Myr, thus confusable with a star, through all the lower spectral classes over time. It is interesting to note that the terms ”early-type” and ”late-type” make logical sense for substellar objects and follow the original theory of star evolution! In Fig. 1.2, we show the spectra of these spectral classes, and Jupiter, to show how the molecular bands become more profound and complex as stellar and substellar objects become cooler. L dwarfs cover the range of about 2200 K to 1300 K, while T dwarfs cover roughly 1300 to 600 K (Golimowski et al., 2004; Leggett et al., 2009). Y dwarfs are the cooler extension to T dwarfs. In L dwarfs metal hydride lines appear in the near-IR spectra such as FeH, CaH, and CrH with MgH in the optical, as well as alkali metal lines. (Reid & Hawley, 2005) Dust forming in the atmospheres of these substellar objects, remove VO and TiO as they become condensates. We can distinguish stellar L dwarfs and substellar L dwarfs, by the appearance of an absorption line at 6707 A˚ due to lithium. Hot brown dwarfs take longer than stars to deplete their

4 lithium while most others (ă 60MJup) are never hot enough to burn it at all (hotter stars may destroy their lithium). We can then say if this line is observed in L dwarfs (in addition to M dwarfs later than M7), we know that it is substellar (Basri, 1998). All T dwarfs are substellar and this spectral class is deﬁned by the many methane bands appearing and deepening in the near-IR (1-2.5 µm) part of the spectrum, which is evident in Fig. 1.2, along with water absorption also strengthening (Burgasser et al., 2002). The condensate clouds of L dwarfs break up and sediment at the T type, making variable patchy atmospheres which seems upheld by their having a greater upper limit on variability amplitude than for L dwarfs, but both types show variability indeed (Metchev et al., 2015).

In Fig. 1.3, we show the color-magnitude diagram for brown dwarfs from Gauza et al. (2019) with measured parallaxes from Dupuy & Liu (2012) and K-M sequence from Pecaut & Mamajek (2013) with near-infrared bands. We can see that the M type transitions smoothly to L, but the transition between the L and T spectral types shows a drastic change. For the J and Ks bands, the T dwarfs become bluer and slightly more luminous before becoming fainter and even bluer, caused by the processes explained. There are models for these substellar objects such as AMES-Dusty, which includes a model with maximum dust content and where sedimentation is ineﬃcient, AMES-Cond, which has no dust opacity and where sedimentation is fully eﬃcient (Allard et al., 2001), and BT-Settl(Allard et al., 2012), where there is a model for dust formation by using a cloud model. All three of these models seem to agree well with stars and each other down to nearly brown dwarf temperatures at about 2500 K; at redder Teff s they start to diverge. As somewhat expected by the models, for infrared colors AMES- Dusty agrees better with L dwarfs and AMES-Cond agrees better with T dwarfs with neither agreeing for the other spectral type. BT-Settl is able to reproduce the L/T transition where, for near-IR J and Ks colors, T dwarfs become 2 mag bluer than L dwarfs and 1 mag bluer than M dwarfs. (Allard et al., 2012) While the latest models do show an improvement in the whole brown dwarf regime, further observational constraints are still needed.

Unlike many other surprisingly discovered objects in astronomy, brown dwarfs have a relatively short history starting with a theoretical basis. The idea of objects that are too cool to start nuclear fusion but eventually evolve to be completely degenerate was given by Kumar (1962). The theory was eventually proven true with the discovery and characterization of Teide 1 by (Rebolo et al., 1995), thus opening the study of substellar objects, which represent the intermediate type between stars and planets.

Brown dwarfs are substellar objects where the Teff is less than about 2000 K. The division between low-mass stars and brown dwarfs happens around 0.07 to 0.08 Md, the transition from where hydrogen burning to where the body is held by electron degeneracy pressure (Hayashi & Nakano, 1963; Burrows et al., 1997). A constrained mass function is not yet available for brown dwarfs owing to their diﬃculty in connecting mass with luminosity, but results suggest that for L dwarfs the mass distribution is rising at lower masses (Reid et al., 1999), (Allen et al., 2005), while for T dwarfs, it is decreasing (Burningham et al., 2010). These results do suggest that especially early brown dwarfs are in abundance, similar to low mass stars. With a mass function appearing to be continuous accross the hydrogen burning limit, it is thought that the same processes that govern low mass stars could apply to high mass brown dwarfs (Kroupa, 2002; Whitworth, 2018).

5 Figure 1.2: Plot of SEDs from optical to mid-infrared of M dwarf Gl101A (blue), L dwarf 2MASS 1507-16 (red), T dwarf 2MASSJ 0559-14 (dark purple), Y dwarf WISE 0350-5658 (light purple), and planetary mass objects in green. In black dots are shown the BT-Settl model spectra. Fig. 1 from (Allard, 2014). See article for spectra references.

4 M stars (>M5) L dwarfs T dwarfs 6 M0 NLTT 51469A M1 46.6 pc M2 33 pc M3 8 M4 NLTT 51469B M5 10 (mag)

J 12 M SDSS 2131-01 14

-2.0 -1.0 0.0 1.0 2.0 3.0 J − Ks (mag)

Figure 1.3: Plot of the color-magnitude diagram for late M, L, and T dwarfs. Fig. 6 from Gauza et al. (2019).

6 Figure 1.4: Plot of the evolution of Teff over time up to 10 Gyr. The blue curves represent low-mass stars, the green curves represent intermediate deuterium burning brown dwarfs, and the red curves represent low-mass non-deuterium burning brown dwarfs. The gold-colored dots are where 50 % of the deuterium supply has been burned. The magenta dots are where 50 % of lithium has been burned. The dashed lines are where the M, L, and T spectral types are approximately divided. Fig. 8 of Burrows et al. (2001).

After their formation, brown dwarfs live their whole lives similar to pre-main sequence stars. Similar to low-mass red dwarfs, brown dwarfs are fully convective with many molecular bands, but unlike evolved stars, brown dwarfs never reach a high enough core temperature to compensate the radiative energy losses at the substellar surface and stabilize before the core is limited by electron degeneracy (even though early ones do burn some hydrogen). Late red dwarfs, on the other hand, spend the longest time of any star in the pre-main sequence stage. Their luminosity and Teff eventually stabilize at about 3 Gyr when they burn hydrogen. Brown dwarfs never stabilize and will always become fainter and cooler (redder) throughout their entire long lives, as shown in Fig. 1.4. Substellar objects always thus have the unique observational problem among ”star-type” objects where the mass and age are degenerate, requiring an observer to obtain an extra parameter to understand their spectral types. To highlight this, a 2200 K object could be either a 0.08 Md red dwarf star after 3 Gyr, or a young sub-13 MJup ”planetary-mass” brown dwarf with an age of 10 Myr, as shown in Fig. 1.4. (Burrows et al., 2001) On the other hand, age is easier to constrain than for an evolved star if multiple observable quantities (i.e. color and dynamical mass from a benchmark system) are obtained.

Deuterium (or hydrogen with a neutron) is able to be fused at lower temperatures than hydrogen, allowing it to be used as an energy source for a limited time in brown dwarfs. The green ”intermediate brown dwarf” curves in Fig. 1.4 show the mass range where deuterium burning is possible in substellar objects. They cause the bumps in the curves where it starts to stabilize until it relatively quickly runs out of fuel, preventing a deuterium fusion

7 main-sequence of ”stars”. This deuterium limit is at about 75 MJup Unlike the high mass end of brown dwarfs where they are neighboring stars, the definition of the low mass end where they are neighboring planets is more controversial. 13 MJup is the limit where deuterium burning is possible for a solar composition, with lower mass substellar objects having smoother descending tracks (blue) in Fig. 1.4 (Burrows et al., 2001). The question arises if the limit between planets and brown dwarfs should be defined using the deuterium limit or by formation mechanism (which would cause a mass overlap), as brown dwarfs are considered to form like stars and planets form by other mechanisms in a circumstellar disk. The IAU defines a brown dwarf to be above 13 MJup, a sub-brown dwarf to be below the limit and free floating, and a planet to be below the limit and orbiting a star or stellar remnant. Au- thors such as Burrows et al. (2001) and Chabrier et al. (2014) argue that formation should be the guiding principle, as deuterium burning has no effect on star formation and little effect on evolution. Non-deuterium burning substellar objects can be free-floating and there is no physical reason disk-forming objects cannot burn deuterium, so they state that they should be called brown dwarfs and planets respectively, even if it causes an overlap between populations. Deuterium-burning and non-deuterium burning objects around the limit show no transitions in their mass-radius relationships, as is seen in other transitions, like star to brown dwarf and Jupiter-like planets to Neptune-like planets (Chen & Kipping, 2017).

In practice, though, understanding a substellar object’s formation mechanism is difficult and would leave many objects undefined, making a simple definition, such as the 25 MJup cut between planets and brown dwarfs (Schneider et al., 2011), attractive. This is argued to be justified by the observational existence of a brown dwarf desert. The brown dwarf desert is a sparc+sity of mass distribution in companions detected by radial velocity with masses covering the whole brown dwarf range with exceptional scarcity at roughly 25-50 MJup (Grether & Lineweaver, 2006; Sahlmann et al., 2011 Ma & Ge, 2014). This does lend some credence to there being a difference between ”brown dwarfs” formed in a disk and ”brown dwarfs” formed as stars, and to definitions like the IAU’s use of a mass cut (even if the deuterium burning limit cut is still hard to justify). The argument does get diluted, however, when results suggesting the desert disappearing for the mass range at wider separations ( 0.2AU; Troup et al., 2016) or with stars more massive than the Sun (Guillot et al., 2014). Nonetheless, this debate will surely continue with GAIA’s potential to astrometrically discover numerous companions in these mass ranges.

1.3 Methods of Detecting Binary Stars and Exoplanets

Binary stars, or two stars orbiting the same center of mass, are of great importance in astronomy as each star can be used to derive parameters from the other star, and further applied to stars in general. Since antiquity, visual double stars (or stars that are in similar positions in the sky regardless of their boundedness) such as Mizar and Alcor have been observed and even used as an ancient eyesight test. Along with the invention of the telescope came the discovery of the first visual binary star, coincidentally also happening to be of Mizar (A and B) by J. B. Riccoli in 1650. William Herschel further studied them by making catalogs of double stars and later confirmed in that some of these are systems bound by mutual attraction thus observationally confirming Newton’s law of universal gravitation Herschel

8 Table 1.2: Stellar multiplicity fractions. Table 1 from Cort´es-Contreras et al. (2017). Refer- ences found in article.

Reference Investigated dlim Multiplicity Projected physical Survey spectral type [pc] fraction [%] separation, s [au] methoda

Duquennoy & Mayor 1991 F7–G9 22 „ 65 „ 0.01–225 RV, WI Raghavan et al. 2010 „ F6–K3 25 44 ˘ 3 „ 0.005–100 000 RV, AO, S, WI Reid & Gizis 1997 K2–M6 8 32 „ 0.1–1800 RV, S, WI Leinert et al. 1997 M0–M6 5 26 ˘ 9 „ 1–100 S Fischer & Marcy 1992 M 20 42 ˘ 9 0–10 000 RV, WI `6.9 J´odaret al. 2013 K5-M4 25 20.3´5.2 „ 0–80 LI Ward-Duong et al. 2015 K7–M6 15 23.5 ˘ 3.2 „ 3–10 000 AO, WI Bergfors et al. 2010 M0.0–M6.0 52 32 ˘ 6 3–180 LI Janson et al. 2012 M0.0–M5.0 52 27 ˘ 3 3–227 LI `6.5 Law et al. 2008 M4.5–M6.0 ă15.4ą 13.6´4.0 „ 0–80 LI `4 Siegler et al. 2005 M6.0–M7.5 30 9´3 ě 3 AO Janson et al. 2014a M5.0–M8.0 36 21–27 „ 0.5–100 LI Close et al. 2003 M8.0–L0.5 33 15 ˘ 7 ă 15 AO Bouy et al. 2003 M7.0–L8.0 20 10–15 1–8 HST `5.3 Reid et al. 2008 L 20 12.5´3.0 ă 3 HST `15 Burgasser et al. 2003 T ă10ą 9´4 1–5 HST Note. a AO: Adaptive optics; HST: Hubble Space Telescope; LI: Lucky imaging; RV: Radial velocity; S: Speckle; WI: Wide-ﬁeld imaging.

(1803). It is now realized that binary stars are very common; in Table 1.2. These discoveries set oﬀ the important ﬁeld of binary star detection and characterization. In this section, we describe the three main types of binaries, as described by their methods of detections, visual binaries, eclipsing binaries, and spectroscopic binaries, with further detail given to the visual and spectroscopic binary methods as they pertain to the work of this thesis.

1.3.1 Visual Binaries

As observed by Herschel, visual binary stars are stars that have wide enough separations to be resolved, and whose motions and therefore orbits can be tracked over time. The usefulness of these binary stars stems from the fact that their masses can derived from the orbits of the two stars. We can obtain that by the following relations.

As two stars orbit a common center of mass, we can express the relationship between the semi-major axes and masses as follows:

m a A “ B , mB aA

where mA and mB are the masses for the A and B components of the binary respectively

9 and aA and aB are their semi-major axes. The semi-major axes can be deduced by the parallax, and therefore distance measured from Earth to the star, and the angular separation from the center of mass to periastron using θ “ a{d, where θ is the angular separation and d is the distance. Combining these two relations, we have

m θ A “ B mB θA

Only a constraint of two orbits of the systems allows us to ﬁnd the mass ratio of the two components, independent of distance. Assuming we do know the distance, we can ﬁnd each mass separately using Kepler’s third law:

4π2 P 2 “ a3, GpmA ` mBq

where P is the period and a “ a1 ` a2. In a real observation, the situation is a little more complicated as the orbit must ﬁrst be deprojected from the observed ellipse in the sky. We discuss this further in Section 2.4.4.

1.3.2 Spectroscopic Binaries

In the case of two stars being too close or one being much fainter, the visual binary technique may be impossible, but it still may be able to be detected indirectly using the spectra. When stars are orbiting each other, their light as received on Earth actually becomes shifted (red if moving away or blue if moving toward us) due to the Doppler eﬀect. This motion is most evident in the absorption lines of the spectra. If both stars are of a similar brightness, we can actually observe the associated spectra of the two stars shift depending on their orbital motion, calling it a double-lined spectroscopic binary. If one star is much fainter, we can only see the one spectrum moving, but can still infer the existence of a companion, calling it a single-lined spectroscopic binary. This is the case for the radial velocity method of ﬁnding exoplanets, as planets are very faint.

For a double-lined spectroscopic binary, we can rewrite a previous equation from visual binaries, considering that the orbital speed of the system is related to its period as:

m a v v sin i A “ B “ B “ B , mB aA vA vA sin i

where vA and vB are the orbital velocities of the system and i is its inclination. In this method, we only observe the motion in the radial direction (v sin i), making the inclination unobservable. Using the relation P “ 2πa{v, we can rewrite Kepler’s third law as:

P P pv sin i ` v sin3 iq m ` m “ pv ` v q3 “ A B A B 2πG A B 2πG sin i3 10 Hence, with this method we cannot obtain the masses for each component, only the minimum mass. We often can only observe lines from one star, or the single-lined spectroscopic binary. In this case, we can use the mass, velocity relation above to replace vB sin i as

P P pv sin i ` pv m {m qq sin iq3 m ` m “ pv ` v q3 “ A A A B A B 2πG A B 2πG sin3 i or

3 mB sin i P 3 “ pvA sin iq mA ` mB 2πG

Here, we can ﬁnd the minimum mass for the B component but the A mass would depend on either detection as a visual binary or eclipsing binary or (assuming it is a main-sequence star) by the mass-luminosity relation. This leads us to our discussion of the importance of binary stars.

1.3.3 Perspective on Binary Stars

While we do not detail further into the method of observing eclipsing binaries, they are also fundamentally important to the knowledge of binary stars and thus the nature of stellar astrophysics. They are simply a pair of stars where one of them crosses in front or behind of the line of sight of the other and causes the light to be dimmed. From them, we can extract the information on their radii, and eﬀective temperature ratio.

Main-sequence stars in binary conﬁgurations can also be used as ”benchmarks” in model- independent mass or metallicity by assuming they has formed together as a more well- constrained one. Mass measurements are especially important, as they provide details into the MLR. The MLR is one of the most fundamental tools to come from binary studies. Since a relation between the binary-derived mass and luminosity (through spectral type) was ﬁrst noted in Halm (1911), we have understood that for main-sequence stars, there exists a dependent relation between the luminosity of the star and mass / temperature. Then Eddington (1924) derived a theoretical basis for the MLR. This is crucial to astronomy, as we can know the mass of a star solely based on its easily-measured magnitude. Over time, thanks to improvements in observational methods (as those we showed above) and instrumentation, we now have many precisely determined masses down to the level of À 3% using double-line spectroscopic binaries (190 stars used in Torres et al., 2010). In Fig. 1.5, we show the derived mass-luminosity relation. The red dwarf regime is the hardest part to constrain and is still dependent on these studies, as is further described in our prior section on red dwarfs.

We can study the occurrence rates of stars in binary or multiple configurations to make inferences on star formation in general. In Table 1.2, we show from Cortés-Contreras et al. (2017) the stellar multiplicity rate for stars according to their spectral type of late F stars to brown dwarfs. This combined with the information of Table 1 from Duchêne & Kraus (2013), we can see that brown dwarfs have as few as 10-20 % in multiplicity up to about 80 %

11 Figure 1.5: Plot of log(luminosity) compared to mass. The mass-luminosity relation for main-sequence stars is clearly visible. Fig. 5 of Torres et al. (2010). for high-mass O stars with about 65 % for Sun-like stars. Low-mass star multiplicity ratios are still quite unconstrained. Overall, multiplicity has a smooth relation with the mass of the primary stars, pointing to a similar formation mechanism for all prestellar cores such as fragmentation (Duchˆene& Kraus, 2013).

1.4 Extrasolar Planets

Extrasolar planets, or exoplanets, are planets found outside of our Solar System. While the existence of exoplanets has been speculated since at least as long ago as Giordano Bruno’s idea that the heliocentric model of Nicolaus Copernicus applies to other stars, it is only since the end of the twentieth century when the first exoplanets were truly discovered around the pulsar PSR 1257 + 12 (Wolszczan & Frail, 1992). A few years later, a planetary system more comparable to the Solar System was discovered with a hot Jupiter discovered around the Sun-like G-dwarf 51 Pegasi using the radial velocity method (Mayor & Queloz, 1995), for which we further describe in the following section. Since then, there has been an explosion in the number of exoplanets discovered, mainly by radial velocities and transits. As of August 2019, there are over 4000 confirmed exoplanets with over 3000 more candidates known. The vast majority of these planets was detected by the transit method, which has experienced a large boost in detections with the Kepler space telescope providing nearly 5000 (+900 from K2) total confirmed and candidate planet detections. (https://exoplanetarchive. ipac.caltech.edu/docs/counts_detail.html) The total sample of confirmed exoplanets is shown Fig. 1.6. We are currently looking forward to being at the beginning of the age of the Transiting Exoplanet Survey Satellite (TESS). TESS has already found nearly 700 candidate exoplanets and some 30 confirmations since its launch in April 2018 and is expected to

12 Figure 1.6: Plot of extrasolar planets discovered with points colored by detection method. http://exoplanets.org/ The red points are from transits, the blue points are from radial velocities, the green points are from microlensing, and the yellow points are from direct imaging.

ﬁnd more than 14,000 exoplanets (Barclay et al., 2018) It is an exciting time for exoplanet discoveries!

1.4.1 Radial Velocities

Radial velocities (RV) or Doppler spectroscopy is the second most productive method in detecting exoplanets, with about 800 planets detected by this method at the present time. This method is technically identical to the one-line spectroscopic binary method that we describe above but applied to planets where a much ﬁner level of precision is generally needed. After the ﬁrst detection of a planet by this method (51 Pegasi b, 0.47 MJ , 4.2 day period), the method was responsible for discovering most of the early known exoplanets until the Kepler mission, many being hot Jupiter mass planets orbiting a Sun-like star which possess many thin absorption lines, able to be used for precise high precision shifts in wavelength. Considering the importance that metallicity has on planet formation, Fischer & Valenti (2005) show that there is a relationship between the host star’s metallicity and gas giant planet fraction. More iron in the star’s atmosphere shows a higher probability of hosting a short period gas giant planet. They show that the metallicity should persist throughout the stars’ interior meaning they were enriched by the molecular cloud rather than polluted by accretion. This gives credence to the metallicity being responsible for increased planet formation by being available for the grains needed to make planetesimals. Neptune-mass planets, though do not show this same correlation as they seem to form in lower-metallicity environments (Udry et al., 2006; Jenkins et al., 2017).

13 The targets for the two detected objects in this thesis come from the surveys of Calán- Hertfordshire Extrasolar Planet Search (CHEPS; Jenkins et al., 2009) and Tuomi et al. (2019), as well as the broader program from the survey of EXoPlanets aRound Evolved StarS (EXPRESS; Jones et al., 2011). We discuss the sample of CHEPS and EXPRESS further in Chapter 2. The high metallicity nature of the CHEPS sample has been useful in studying the mass-metallicity correlation of planets, such as the finding that low-mass planet desert may exist with no very low mass planets (ă 0.03MJup) found at high metallicity (Jenkins et al., 2013a). Tuomi et al. (2019) studied nearby M dwarfs with which there was `4.58 RV data. They find that M dwarfs possess 2.39´1.36 planets on average, showing them to be common around these stars. Thus far, the statistics of RV planets have allowed us to peer into the processes that govern planet formation and occurrence.

1.4.2 Direct Imaging

Direct imaging of exoplanets is the simplest method in concept (taking images of planets around stars), but among the most diﬃcult to implement in practice. We will describe the technique behind this method in Sec. 1.5, but in this section, we describe some of the results this method has brought about. Planet and brown dwarf searches by direct imaging are typically intentionally biased toward young (often in comoving assocaiations), low mass stars, and/or that are close in distance (if possible as this is balanced by youth). As shown previously, as these substellar objects age, they become fainter and redden, making younger objects more easily visible in the near-infrared bands and with our current adaptive optics (AO) instrumentation. A planet orbiting a low mass red dwarf star needs a much lower contrast performance than it would if it were orbiting a higher mass one, as it is fainter than the other star and will have less scattered light interfering with the light from the planet. For example, in K band, an aged mid T5 dwarf would need to be detected with a contrast of 9.7 mag to an M0 dwarf compared to a contrast of 13.9 to an A0 star (http://www. pas.rochester.edu/~emamajek/EEM_dwarf_UBVIJHK_colors_Teff.txt), with a detection right at the limit of current technology („ 14mag). Closer stars allow an improvement to our ability to ﬁnd planets at a shorter separation from their host stars. This is useful as we know already from RVs that giant planets are have a higher occurrence at short separations than at long periods from direct imaging (Cumming et al., 2008; Bowler, 2016).

The ﬁrst orbiting substellar companions directly detected were the brown dwarfs GD165B Becklin & Zuckerman (1988) and Gl229B (Nakajima et al., 1995) which made use of AO and coronagraphy, techniques which would prove powerful to ﬁnding these companions. While there have not been nearly as many planet found using the direct imaging method („ 50; Bowler (2016)) as RVs, a number of interesting diverse planetary-mass companions have been found including the four À 10MJup planets of HR 8799 (Marois et al., 2008; Marois et al., 2010),8 MJup planet of the debris disk-possessing β Pictoris (Lagrange et al., 2009), 2 MJup 51 Eri b (Macintosh et al., 2015), and 9 MJup HIP 65426 b (Chauvin et al., 2017). We show these planetary-mass companion detections and more in Fig. 1.7, and for a list of the parameters of these companions, we refer the reader to Table 2 of the review of Chauvin (2018a).

While the sample is still limited compared to the other planet detection methods, especially

14 Figure 1.7: Images of directly imaged planetary mass or exoplanet companions to the stars as follows (right to left): DH Tau b, UScoCTIO 108 b, 51 Eri b, AB Pic b, β Pic b, Fomalhaut b, 2M0122-2439 b, HR 8799 bcde, RXJ1609 b, HD 106906 b, κ And b , HIP65426 b, CD-35 2722 b, GJ 504 b, HD 95086 b, GU Psc b, HIP 77900 b, HR 6037 b, 2M1207 b and 2M0143AB b. Figure extracted from Fig. 4 of Chauvin (2018a). when considering Earth analogues of which none have been detected and old planets of which only few have, there is still enough to make some scientiﬁc deductions. Bowler (2016) looked into results from previous imaging surveys and found a sparsity of giant planets on wide orbits (5-13 MJup from 10-100 AU) from imaging programs with an occurrence rate for stars from `1.2 B to M of 0.8´0.5%, with similar occurrence rates for planets up to 1000 AU. This reﬂects the few companions that have been found with the latest high contrast imaging instruments such as SPHERE and GPI.

1.4.3 Combining Radial Velocities and Direct Imaging

While RVs are shown to be more productive in finding low mass planets in general and especially ones at low separation (as it can only find periods that are as long as the data baseline), direct imaging does have a niche in finding widely separated ones. Using these two techniques together can combine the advantage of each one and make up for the other’s disadvantage. As RVs can find the planet’s m sin i, direct imaging can provide the inclination, i, through orbital tracking, which allows us to measure the planet’s ”dynamical mass” with similar importance to that which was explained for binary stars. Long period trends or linear trends, in particular, can be useful because even though the RVs cannot be constrained with an orbital solution, the trend gives indication that there is an unseen companion with a period at least two times the data time baseline. As the companion is probably at a high separation, this makes these targets ideal ones for observation with imaging. This has been used with success through the TRENDS survey including detection of 3 M dwarfs (Crepp et al., 2012) and 2 brown dwarfs (Crepp et al., 2014, 2016) to solar type stars, discoveries of

15 low mass stellar companions to intermediate giants in the program of Ryu et al. (2016), and follow-up combination data for previously detected companions from (i.e. Dupuy et al., 2019; Grandjean et al., 2019). Further investigation of low-mass companions with this method, along with using barycenters, eclipsing binaries with RVs, and GAIA astrometries will allow us more low-mass stars with metallicities with which to constrain models and the mass- luminosity relation.

1.5 Technique of Direct Imaging with AO Systems or High Contrast Imaging of Exoplanets

As we described above the history and current status of the direct imaging of exoplanets, we will now describe the techniques behind these discoveries. From the above status, we note a paucity of exoplanets detected by direct imaging compared to the other methods showing that direct imaging is a difficult method to implement. As a real-life analogy, detecting a planet by imaging is comparable to distinguishing a firefly near a lighthouse in Brazil from Santiago, Chile. While this is easily understood to not be an easy task, we fortunately have some methods to use to make this method more possible and in the future, we hope, very productive. These methods include building optimized hardware, such as large telescopes and adaptive optics, and postprocessing differential imaging techniques, for which we describe in this section.

1.5.1 AO System

Firstly, the most valuable thing we can do to improve our ability to detect and characterize faint objects near to their host stars, along with improve nearly anything in astronomy, is build a bigger telescope. For our goals this allows to improve the two crucial parts: the resolution and the limiting magnitude. To understand how a larger telescope improves the image resolution, it is important to understand the Airy disk. Due to the wave nature of photons (from quantum physics we know they have a matter-like nature as well), they diffract into a wave on the detector, similar to the effect of a water wave going through a slit. As objects observed in astronomy are viewed far from the objects themselves, making the light arrive in parallel, through a circular aperture (or the telescope mirror), we consider it a case of Fraunhofer diffraction by a circular aperture or the Airy diffraction pattern as shown in Fig. 1.8. The resolution limit of the telescope is given by angle where the first zero occurs for the diffraction pattern.

1.22λ θ “ , min D

where θmin is angle where the ﬁrst minimum occurs or the angle of the resolution limit, λ is the wavelength of light and D is the diameter of the telescope primary mirror (Peatross & Ware, 2015). As we can see, as a larger wavelength gives a larger resolution limit, it would

16 Figure 1.8: (A) Image image of a model airy disk and rings. An observation of a star with a perfect telescope system and no atmosphere would produce this image. (B) Surface plot of the airy disk showing the PSF. Note that the central disk holds the majority of light compared to the fainter rings. Credit: Fig. 3 of Schmolze et al. (2011).

be more difficult to detect planets close to a star with infrared (IR) than optical imaging (but for self-luminous planets, they are easier to find in these wavelengths due to their peak fluxes at cool temperatures).Resolution limit decreases with increasing telescope diameter, showing that a larger mirror would, indeed, produce are higher resolution image. A larger telescope also provides a larger photon collecting area as A “ πpD{2q2 where A is the area of the mirror.

While the above information of an inverse dependence between resolution and diameter of a telescope is true in theory, in practice it is more complicated due to the Earth’s atmosphere. Turbulence in the atmosphere causes an astronomical object to be distorted and blurred and it aﬀects our ground-based astronomical observations everywhere on Earth (building in space is possible but expensive). We often measure this with astronomical seeing usually given as the full width at half maximum of the PSF after some integration (without which we see rapidly changing speckles) over time, as this gives us a larger disk. Problematically, this means we are limited in resolution by the atmospheric seeing, no matter how big we build a telescope. For example, if we observe a star at a near-IR wavelength of 1 µm with an exceptional seeing of 0.5 arcsec, we would have the resolution of that of a 40 cm telescope! We therefore do not gain any resolution building a mirror above that of a small telescope (but we do still gain greatly in the also crucial photon collecting area) without additional modiﬁcation.

One method to improve image quality is active optics as it corrects deformations in the mirror due to external conditions such as wind as well as telescope conditions such as mirror temperature or deformations due to telescope movements with low frequency ( 0.1 Hz) corrections (Noethe, 2002). It can be used alone, like at the New Technology Telescope at La Silla Observatory or together with AO like with some instruments, such as SPHERE and SINFONI on the Very Large Telescope at Paranal Observatory, both in Chile. Due to the variables in telescope conditions, active optics actually allows to achieve seeing-limited resolution to begin with.

17 Figure 1.9: Schematic of the AO loop. Fig. 1 of Rigaut (2015)

To overcome this atmospheric limit and get closer to the telescope resolution (or diffraction) limit, we need to correct for the atmospheric turbulence itself using adaptive optics (AO), as briefly alluded to earlier. AO is similar to active optics in technique, but differs in the corrections in practice as it needs to correct for the very rapid variations of the turbulence. In Fig. 1.9, we show the loop of the AO system. The wavefront from the astronomical object enters the deformable mirror from the telescope where light is split with part of it sent to the detector and part to a wavefront sensor. The most common type is the Shack- Hartmann wavefront sensor, as used by SPHERE, where the beam is separated into many apertures by microlenses onto a detector. The location of the position of one of the focal spots corresponds to the slope of the local wavefront through an aperture. Using these slopes, a corrected wavefront can be processed by a control computer (or realtime computer) with the corrections sent to the actuators of the deformable mirrors. The output light to the camera is then the corrected PSF, which should be closer to a diffraction limited PSF. This information is constantly sent between the mirror, wavefront sensor, and control computer throughout the observation (Rigaut, 2015). We often measure the quality of PSF correction, by comparing the peak of the measured PSF to that of a diffraction limited one as the Strehl ratio. As an example to describe the current state of the art in PSF correction, SPHERE achieves a high ě 90 % Strehl ratio in the H band for a star of R = 9-10 (Beuzit et al., 2019). While this is indeed an improve on past instruments in which „ 50 % was the norm, there is still room for improvement in the next generation of AO instruments.

Even with a great Strehl ratio and high resolution, the light from the star and its rings or stellar halo have a huge eﬀect, as their light is still far higher than that of a planet. To reduce this light, we make use of coronagraphy. As shown in Fig. 1.10, it works by using two stops, one occulting stop in the image plane followed by a Lyot stop in the pupil plane after

18 Figure 1.10: Light path through a Lyot coronagraph. Credit: http://lyot.org/ background/coronagraphy.html adapted from Sivaramakrishnan et al. (2001)

a lens, described by Lyot (1939) as a way to observe the corona of the Sun, independent of a solar eclipse. As the photosphere of the Sun is 300,000 times brighter than its corona, to find planets around a star we similarly dim the light of star (Eddy & Ise, 1979). The difference is that a star is a celestial object with a far smaller angular size, which is below the diffraction limit. The advantage is that, once we dim the star and reduce its photon noise, we are able to increase the exposure time greatly without saturating the detector, reducing overheads in observations. We can thus gather more photons of the companion itself without increasing the total time of observation.

While coronagraphy and AO allow us improve our ability to ﬁnd faint companions at high resolution, we still have the problem of quasi-static speckles. Unlike the rapidly changing (ms level) speckles due to atmospheric aberrations, quasi-static speckles are caused by imperfections in the telescope system itself, such as the telescope support causing so-called ”spiders” in the image or aberrations not detected in the wavefront sensor outside the loop in the beam split toward the detector. These speckles have lifetime timescales on the order of some minutes (Hinkley et al., 2007). These speckles can take a similar appearance to a faint companion, making it necessary to be cautious about taking a planetary signal by imaging at face value. Fortunately, in addition to the hardware improvements we can make to an imaging instrument, as previously discussed, we have some post-processing methods such as diﬀerential imaging, where we make some reference PSF that should represent a PSF including these speckles, but without any astrophysical signals with which we can subtract the observations from, as well as optimizing the use of data set with a technique such as principal component analysis. We discuss these in the next section.

19 1.5.2 Diﬀerential Imaging and Post Processing Techniques

Differential imaging methods are techniques used to distinguish a low-mass companion of a star from the aforementioned speckles which can take the appearance of one. There are a wide variety of them, including the following: Simultaneous differential imaging is where an object is observed in two different filters at the same time, with one of the filters being on an absorption line of an expected exoplanet SED such as the 1.6 µm methane line, and where after the two images are subtracted, a planet could be extracted (Racine et al., 1999). This technique is available to SPHERE. Spectral differential imaging, with the use of an integral field spectrograph like those on SPHERE or GPI, makes use of the fact that the speckles scale with „ λ{D like the star, allowing us to subtract speckles that scale and leave a signal that does not (Sparks & Ford, 2002). There are multiple other differential techniques as well, but we will focus on the two most prominent techniques: angular differential imaging (ADI) and reference (star) differential imaging (RDI).

ADI is the most popular method at the current time and has been most responsible for recent planetary discoveries, such as those shown in the review of Chauvin (2018a). ADI works by maintaining a stable pupil, allowing the field to rotate relative to the change in the parallactic angle. We show its classical variant in Fig. 1.11, where each frame of the observation is derotated and taking their median to make a reference which would ideally include only the speckles and not the astronomical source (as it rotates with field). This can then be subtracted from the original observation frames to provide a deeper contrast of the stellar halo. (Marois et al., 2006) While it has been a very effective method in boosting our ability to find low-mass companions to far brighter stars, it does have a few limitations including that it has more difficulty for companions that are very close to the star itself, as the reference frame includes the companion when the parallactic angle is not very large, leading to self-subtraction of the companion in the final reduced frame. It also can require quite large amounts of time to achieve a large parallactic angle variation, even when a reduction in actual photon noise is not required leading to inefficiency. It also has difficulty for resolved objects, such as disks which have issues with self-subtraction especially at low inclination (Milli et al., 2012).

RDI is another differential imaging technique that has been used for some time, due to its simplicity. It uses another star, taken in similar conditions at a close location on the sky with similar properties such as brightness and color, as the reference PSF with which to subtract the science PSF and gain in contrast (Xuan et al., 2018; Ruane et al., 2019). This method was notably used in the detection of β Pic b (Lagrange et al., 2009). This can be achieved by either observing the reference close in time to the science observation, thus reducing the effect of the changing quasi-static speckles, or making a reference frame library from many observations in the same mode and filter of the instrument with which we can make a high- correlated reference to the science PSF and subtracting it out. We discuss this further with the following discussion on post-processing techniques. As the reference library is completely separate from data including a companion, self-subtraction issues of close companions can be eliminated and it can be more efficient (as well as useful for targets where the field hardly rotates) as the parallactic angle is irrelevant (except that the parallactic angles should be similar between the science and reference so that the pupil configuration is similar). As space based instrumentation do not follow targets across the sky as ground based ones do, RDI

20 Figure 1.11: Figure showing Angular Differential Imaging, with subtraction of the science frames by the median reference, derotation, and stacking of the frames. Credit: Christian Thalmann has been used as well for the Hubble Space Telescope data, like in the redetection of the b, c, and d planets of HR 8799 from 1998 data (Lafrenièreet al., 2009; Soummer et al., 2011). Binary differential imaging is a promising method similar to RDI, where a visual binary is observed in the same frame with which their identical PSFs, unaffected by speckles changing in even quickly taken observations, can be used to subtract the other. Rodigas et al. (2015) showed this to be promising with an improvement in contrast over ADI from a Magellan AO data set at the close separations. It would be limited in the case of a coronagraph, though, where it could only be centered on one star.

In addition to these differential imaging methods, further post-processing techniques can be used to make the most use of the data sets. Principal component analysis (PCA) has had use in astronomy for some time as a way to reduce the dimensions needed to model a large dataset, e.g. in Connolly et al. (1995), where they used it to reduce galaxy spectral components. PCA works by building an orthogonal set of dimensions, or images for an imaging program, known as principal components. The first of the principal components most accounts for the most variability in the data, with the variance decreasing in order by index of principal components with the total number of principal components equal to the number of dimensions in the data set. We would then subtract the reference PSF made up by the reconstruction of the frame by eigenvectors, determined by the number of principal components we want to use, from the data frames themselves (Soummer et al., 2012). Locally Optimized Combination of Images (LOCI) is another algorithm that is used, where a linear combination of sections of images, where the residual is lowest from least-squares, and is used to make a reference PSF to subtract from the data (Lafrenièreet al., 2007). These algorithms can be used with ADI or RDI to improve the capabilities of either by building a PSF that best represents the data set. For example, we could use PCA on a large data set that could consider the most correlation between the library frames to build a reference PSF with which to perform RDI, even if the data was taken within a large range of time. This

21 would allow us to gain contrast without doing taking any extra observation time to observe a reference PSF, intentionally for our science target.

1.6 Thesis Outline

In Chapter 1, we have provided the introduction to the background that forms the basis for the work done in this thesis. We have discussed how M dwarfs are structured, the observational characteristics they have, and the broader use they have in studying areas from galaxies to exoplanets. Importantly, the mass-luminosity-metallicity relation for low-mass stars was described and shown that there is still work to be done to constrain this relation precisely and in different wavelengths. Brown dwarfs were discussed as well, in relation to their observable characteristics and how they evolve. There is still more to be investigated on their structure and models through observations. Binary stars are necessary for finding dynamical masses, which are still needed for understanding M dwarfs and brown dwarfs. Using the imaging and spectroscopic methods together allow us to find these dynamical masses. The methods for detecting binary stars are used as well for exoplanets. While there has been an explosion in the number of planets we know of, there is investigation needed to further understand their formation mechanisms. Lastly, to be able to do direct imaging for the detection of low mass objects, we need to use a few tools such as adaptive optics and post-processing algorithms to remove the tremendous brightness of a star.

In Chapter 2, we explore the technique to follow up linear radial velocity trends to find low-mass companions with imaging with our first detection from the SPHERE Ao Additional Radial velocity companIons (SAFARI) search. The RVs allowed us to narrow our sample to stars with which we know to possess a relatively distant companion. We detected directly an M3 dwarf to a G dwarf star (HD 86006) using the CHEPS RV planet search program during SPHERE Science Verification time. Using the proper motion of the star, we were able to conclude that the two binary stars are gravitationally bound. This is a powerful technique as we can use the G star itself, which has an easily studied metallicity (which is high) as a benchmark to infer the metallicity for the M dwarf itself. We investigated the M dwarf’s spectral type, temperature, mass, and age using spectral fitting with templates and models, spectral indices, and isochrones. These all agree with the early M characteristic we could assume from its spectral type. We also saw that known M dwarfs with high and low metallicities are scattered throughout the BT-Settl isochrone ages, agreeing with our age discrepancy from being an old star. We also did a preliminary MCMC joint RV+astrometry orbital fit, noting that more data will be needed before a better fit can be made. Using our other target HD 90520, which had no detection but did achieve the highest contrast performance, we give an upper limit to the mass of the unknown companion.

In Chapter 3, we show our second major detection from the SAFARI program, where from a linear trend for an M star hosting a super-Earth (GJ 3634) from the program of Tuomi et al. (2019), we detected two low-mass companions orbiting the M star and one another. Between MagAO, SPHERE, and GAIA data, we were able to conﬁrm that the three objects are gravitationally bound to each another. With Director’s Discretionary Time with SINFONI, we could take a medium-resolution integral ﬁeld spectrum of the system. The spectra agreed

22 with the two B components being M8 and L0.5 dwarfs by spectral ﬁtting. That makes these objects ultracool dwarfs, making for a dynamically interesting system, especially since they are at a short 30 AU from the primary where a planet is hosted. Spectral indices agrees with this assessment. With a fully constrained orbit in the future, we would be able to take a look into the formation scenario of rare types of systems as this, with a red dwarf, two close ultracool dwarfs, and a planet.

In Chapter 4, we show a technical test we did with SPHERE in comparing the ADI and RDI methods. We performed a starhopping technique where we rapidly (duty cycle of 2 minutes) repositioned between two stars of a visual binary under the coronagraph one at a time for over 1 hour of total time. The conditions were also exceptional being at 0.5 arcsec seeing for most of the observation. This allowed us to have the ideal data set for comparing ADI, where we need the long time to obtain high field rotation, and RDI, where the rapid changing between a science and reference target allow us to minimize the changes in their PSFs. We did various tests on this sample using the VIP package where we injected fake companions and measured their signal-to-noises, measured the contrast performances of the reduced images, made cuts on the data sets to simulate observing conditions more similar to real ones, and measured the photometry and astrometry of many injected fake companions into the data. Overall, we found that RDI consistently provides a better result for finding companions at the close 0.3 arcsec separation than ADI, whereas outside of this separation they are more balanced. This shows that RDI is a promising technique for searching for close planets, where they are expected to commonly be, as well as being efficient especially when considering the possibility of creating a data library.

In Chapter 5, we summarize the results in the thesis and discuss the broader perspective to our technique. We also discuss the future work that can come from the full sample of the SAFARI program, as well as with future ground-based and space-based instrumentation.

23 Chapter 2

SAFARI I: A SPHERE discovery of a super metal-rich M dwarf companion to the star HD 86006

Authors: Pantoja, B. M., Jenkins, J. S., Girard, J. H., Vigan, A., Salter, G., and Jones, M. I. Article Reference: Monthly Notices of the Royal Astronomical Society, Volume 479, Issue 4, p.4958-4970

We report the direct detection of a fully convective, early-to-mid M-dwarf companion orbiting the star HD 86006, using ESO-SPHERE during Science Verification as part of the SAFARI program. HARPS+CORALIE radial velocity measurements first indicated a possible companion. Such work highlights the synergies that are now possible between these two observing methods. We studied the companion by comparing our observed spectra with BT-Settl models and template spectra, measuring spectral indices to obtain a spectral type, and used a joint radial velocity and astrometric fit to simulate the companion’s orbit. The companion was found to be 4.14 mag fainter than the primary in the H2 band, residing at a physical separation of „ 25 AU, with a Teff and spectral type of 3321 ˘ 111 K and M 4.1 ˘ 1.1, respectively. We note that the age derived from BT-Settl models for such a star is too low by over two orders of magnitude, similar to other known field mid-M stars. We searched for the radial velocity companion to HD 90520 without any clear detection, however we reached a low contrast level of ∆H2 = 10.3 mag (or 1.3 ˚ 104) at 0.22 and 12.6 mag (or 105) at 0.52, allowing us to rule out any low-mass companions with masses of 0.07 and 0.05 Md at these separations. This discovery provides us with the exciting opportunity to better constrain the mass-luminosity relation for low-mass stars in the super metal-rich domain, expanding our understanding of the most-common types of stars and substellar objects.

24 2.1 Introduction

Binary stars are of great importance in stellar astrophysics. As the two stars in a system are gravitationally bound, we can derive important parameters such as mass, luminosity, period, and radius from these, as well as metallicity from studying the stars’ spectra, which can give information about the underlying stellar physics. Solar-type stars, themselves, are in a multiple star conﬁguration at a rate of about 45 % (Raghavan et al., 2010) for metallicities between -1.0 and +0.6 dex (Jenkins et al., 2015). As solar dwarf stars are common and very well characterized, they make for ideal laboratories to study less understood stars, such as low-mass M dwarfs.

There are methods to investigate binary systems with a low-mass component, such as direct imaging (DI). At the current time, young stars (ă 1 Gyr) are being targeted by DI searches as they are shown to make for feasible targets to detect planets, since planets this young are comparatively hot and bright (Marley et al., 2007). Successful DI detections of young planets have allowed us to study some nearby systems, like HR 8799 (Marois et al., 2008), β Pictoris (Lagrange et al., 2010), 51 Eridani (Macintosh et al., 2015) . As of yet, no planets have been discovered orbiting solar-age stars with DI, owing to the large contrast between the planet and star. Nevertheless, a number of old low-mass stars and brown dwarfs at that age have been imaged (e.g. Burningham et al., 2009; Crepp et al., 2012, Mace et al., 2013; Crepp et al., 2014; Crepp et al., 2016; Ryu et al., 2016).

Radial velocity (RV) surveys have been popular in recent years due to their productivity in discovering extrasolar planets in the Earth to Jupiter-mass regime at close separation to the star. One disadvantage to the RV method is that it is diﬃcult to constrain orbits where the orbital period of the Doppler source is greater than the time baseline of the observations. We call these RV detections long-period trends. Linear trends are RV trends where we have little or no indication of any inﬂection (therefore appear linear) and can extract little of the source’s orbital parameters. Another characteristic of the RV method is that the constrained orbits provide a minimum mass as M sin i, not the absolute companion mass unless another method is used in conjunction with the RVs, like transit observations.

By combining RVs and DI, we can use the advantages of both methods to gain much more information of a star’s orbital companion than that which can be measured by each method individually. Long-period RV trends make for good targets for DI searches as their probable large separation between the companion and primary bias the sample to where DI is most productive, at a distant separation from the star’s bright halo. By tracking the orbits over time, the inclination can be derived in the images, providing a true dynamical mass for the companion.

For M dwarfs orbiting G dwarfs, we can use the masses from G stars’ evolutionary models, along with the RV and DI parameters, in the derivation of the M dwarfs’ dynamical masses. As well, we can use derived metallicities and ages from the more massive stars’ models along with the dynamical mass and use them to constrain M dwarf evolutionary models, as the stars can be assumed to have been formed concurrently. We can make statistics for RV trends and discover if they relate to a companion of stellar-mass or substellar-mass, providing information about the formation mechanism of these low-mass companions. As

25 we would obtain information about the frequency of M dwarfs orbiting G dwarfs, we gain a further understanding of how often M dwarfs are contaminants in planet searches, such as their being false positive eclipsing binaries in planet transit searches (Almenara et al., 2009). Imaging non-detections are also very useful, as an upper mass limit can be given for the orbital companion which would rule out a massive stellar companion (Guenther et al., 2005). These targets would be useful for future instrumentation to directly detect the brown dwarf or planetary-mass companion.

Currently, the search for small planets around M dwarfs is becoming ever more popular, as highlighted by recent discoveries of planets orbiting GJ1214 (Charbonneau et al., 2009), Proxima Centauri (Anglada-Escud´eet al., 2016), TRAPPIST-1 (Gillon et al., 2017), LHS1140 (Dittmann et al., 2017), and NGTS-1 (Bayliss et al., 2017). M dwarfs make for excellent targets to search for important orbiting planets as their habitable zones (HZs) are located at short distances from their surfaces, with only a relatively small diﬀerence in brightness compared to more massive Sun-like stars. The mass, radius, and density of detected planets can be inferred from the stars they are orbiting. To disentangle these stars’ characteristics and obtain precise masses for planets, it is important to measure well their luminosities and masses to characterize the population mass-luminosity relation, as has been done previously by Henry & McCarthy (1993), Delfosse et al. (2000), and Benedict et al. (2016), for example.

The Calán-HertfordshireExtrasolar Planet Search (CHEPS; Jenkins et al., 2009) and EXoPlanets aRound Evolved StarS (EXPRESS; Jones et al., 2011) surveys have provided us with a wealth of candidates to follow-up with DI methods in the southern hemisphere. The CHEPS sample comprises a subset of metal-rich ([Fe/H] ě 0.1 dex) stars that have colors corresponding to late-F to mid-K spectral type (0.5 ď B ´ V ď 0.9), are located in the o 1 Southern Hemisphere (δ ď 0 ), and exhibit low activity levels (log RHK ď -4.5). Metal-rich stars were chosen as they have been shown to have an increased probability of possessing giant planets (e.g. Fischer & Valenti, 2005; Sousa et al., 2011). Thus far the CHEPS has uncovered 14 new companions with masses ranging from that of Uranus up to the brown dwarf regime (Jenkins et al., 2009; Jenkins et al., 2013b; Jenkins et al., 2017). The EXPRESS sample was selected to be bright and nearby G and K (0.8 ď B ´ V ď 1.2) giant stars (-0.5 ď MV ď 4.0) in the Southern Hemisphere. These stars are shown to exhibit correlations between stellar mass, up to 2.5 Md, and metallicity with the occurrence rate of giant planets (Jones et al., 2016). The EXPRESS sample has currently discovered 11 substellar companions (Jones et al., 2013; Jones et al., 2015a; Jones et al., 2015b; Jones et al., 2016; Jones et al., 2017) and 24 spectroscopic binaries (Bluhm et al., 2016). Any DI detections from these two samples will provide important evidence for the formation mechanism of low-mass companions to metal- rich solar type dwarfs and giants. The detection and constraint of orbits for gravitational companions will follow up the study by Bonfils et al. (2005) where the mass-luminosity relation for low-mass stars was considered along with the effect from metallicity.

To date, there have been a small number of programs following up long-period radial velocity trends with DI, such as those of Crepp et al. (2012) and Ryu et al. (2016), which show detections of ”benchmark” low-mass companions that will be able to constrain dynamical masses in the future.

26 In this article, we introduce our SPHERE Ao Follow-up of Additional Radial velocity companIons (SAFARI) with our discovery of a low-mass stellar companion to HD 86006, a star with a linear trend from the CHEPS program. We discuss our selection of the two targets for this program, our observations, and our reduction procedure in Section 2.2. We show our estimation of the companion’s fundamental characteristics, such as mass and temperature, as well as our constraints on the orbital parameters of the system in Section 4.3. We also describe our observations of the star HD 90520 that resulted in a non-detection, and describe the lower-mass limit that was achieved in the same section. Lastly, we show our conclusions in Section 4.5.

2.2 Target Selection, Observations and Reduction

The targets we selected for SPHERE Science Verification (SV) time were taken from the CHEPS (Jenkins et al., 2009) and EXPRESS projects (Jones et al., 2011). From the CHEPS program of about 150 metal-rich F to K stars stars and EXPRESS program of about 170 giant stars, the targets were chosen, at the time of this article, from a subset of 19 stars from CHEPS and 14 stars from EXPRESS that were shown to exhibit long period trends in their radial velocities, indicating they were hosts to very long period companions. These numbers were increased to 40 stars we currently are using as targets, with 26 coming from CHEPS, 9 from EXPRESS, and 5 from Tuomi et al. (2019). The large sample of M dwarf planet hosts of Tuomi et al. (2019) was our latest addition to the sample, with further target additions of linear trends are planned. Following the SV program, the plan is to use AO instrumentation with short exposures to run a snapshot program, where we follow-up imaging is used to map out the orbits of detected companions. Non-detections provide targets for long-exposure imaging where we perform post-processing techniques such as angular differential imaging (Marois et al., 2006) to find fainter companions. Further non-detections would be saved as targets for future space and ground-based imaging on large telescopes.

Although some of our selected sample exhibited signiﬁcant curvature in their velocity timeseries, we decided that stars with linear trends were preferable in this case since the companions would be widely separated from the star, and given that this was a SV program where the instrument was being tested, the largest possible star-companion separation would maximise the possibility that we would obtain a positive result. HD 86006 and HD 90520, both from CHEPS, made for ideal targets due to their large linear RV trends and favourable sky positions for SV time, and in Table 3.1, we list their stellar characteristics. We are interested in targets that display some curvature as well and are including them for the next observations in our survey.

The [Fe/H], [α/H], Teff , mass, log g, and age parameters were obtained using the SPECIES code (Soto & Jenkins, 2018). SPECIES measures the FeI and FeII lines from high-resolution echelle spectra. By applying the ATLAS9 model atmospheres (Castelli & Kurucz, 2004) and local thermodynamic equilibrium, it can obtain the surface gravity, metallicity, and temperature. The stellar photosphere’s microturbulence velocity and the stellar rotational velocity are also obtained by using synthetic spectra to ﬁt the absorption lines along with temperature relations. From the Isochrones (Morton, 2015) package that interpolates the

27 MIST Isochrones (Dotter, 2016), the mass, radius, and age are obtained. Finally for these 1 stars, log RHK and v sin i were obtained from Jenkins et al. (2011).

Table 2.1: Stellar characteristics of the stars HD 86006 and HD 90520. Property HD 86006 HD 90520 Reference RA (J2000) 09:54:31.169 10:26:10.648 1 Dec. (J2000) -45:43:51.53 -45:33:45.16 1 P.M. RA (mas/yr) 53.388 ˘ 0.056 -85.150 ˘ 0.046 2 P.M. Dec. (mas/yr) 38.012 ˘ 0.054 79.833 ˘ 0.045 2 V (mag) 8.20 7.50 3 J (mag) 6.940 ˘ 0.021 6.412 ˘ 0.021 4 H (mag) 6.705 ˘ 0.051 6.148 ˘ 0.038 4 K (mag) 6.582 ˘ 0.017 6.064 ˘ 0.021 4 `1.55 `0.96 Distance (pc) 78.13´1.49 64.02´0.93 2 Sp. Type G5IV/V G0IV/V 5 [Fe/H] 0.33 ˘ 0.06 0.21 ˘ 0.07 6 [α/Fe] 0.39 0.24 6 Teﬀ (K) 5821 5946 6 Mass (Md) 1.27 ˘ 0.04 1.30 ˘ 0.05 6 log g (cm/s2) 4.129 ˘ 0.213 4.108 ˘ 0.244 6 Age (Gyr) 3.67 ˘ 0.39 2.92 ˘ 0.46 6 1 log(RHK) -5.05 -5.00 7 vsin(i) (km/s) 3.4 4.6 7 Age (Gyr) 7.95 ˘ 0.90 3.43 ˘ 0.89 8

1: van Leeuwen (2007) 2: Gaia Collaboration et al. (2016) 3: Egret et al. (1992) 4: Cutri et al. (2003) 5: Houk (1978) 6: Soto & Jenkins (2018) 7: Jenkins et al. (2011) 8: Casagrande et al. (2011)

2.2.1 Radial Velocity Observations

The radial velocity data for these stars were observed by CORALIE at the Swiss Leonhart Euler Telescope (Queloz et al., 2000) and using the High Accuracy Radial Velocity Planet Searcher (HARPS) at the 3.6m Telescope (Mayor et al., 2003) at the La Silla Observatory in Chile. The observations of HD 86006 were made over a timespan of nine years, running from April 2008 to July 2017. For HD 90520 they were made over six year period, from May 2007 to June 2013.

CORALIE is an echelle ﬁbre-fed spectrograph with a spectral resolving power of „50,000 covering the wavelength range between 3900-6800 A.˚ It regularly maintains a radial veloc-

28 Table 2.2: Radial velocity data for HD 86006. JD RV (m/s) * σRV Instrument 2454579.68825 -125.730 0.908 HARPS 2454580.65361 -124.656 0.840 HARPS 2454581.64833 -127.953 0.661 HARPS 2455883.87993 -101.684 0.674 HARPS 2455885.79942 -3.739 0.857 HARPS 2455992.62698 6.547 0.475 HARPS 2455993.59547 8.259 0.519 HARPS 2455994.57316 8.381 0.590 HARPS 2456063.52248 20.093 0.730 HARPS 2456065.56623 14.132 0.507 HARPS 2456442.57086 58.062 0.902 HARPS 2456443.55437 64.089 0.763 HARPS 2456444.53733 63.996 0.935 HARPS 2456449.48962 60.322 0.709 HARPS 2456461.50032 59.048 0.591 HARPS 2456462.51857 58.943 0.731 HARPS 2456463.49291 61.398 1.052 HARPS 2455208.76688 -151.761 9.000 CORALIE 2455210.76373 -165.761 10.000 CORALIE 2455229.65227 -156.762 9.000 CORALIE 2456307.78252 -66.784 11.000 CORALIE 2456463.50251 -24.788 12.000 CORALIE 2456464.49144 -32.788 14.000 CORALIE 2456467.52434 -22.788 16.000 CORALIE 2456675.79582 8.2074 12.000 CORALIE 2456733.68959 18.206 11.000 CORALIE 2456736.69728 7.206 12.000 CORALIE 2456824.54322 6.204 12.000 CORALIE 2456968.86876 22.201 13.000 CORALIE 2457075.72074 42.199 13.000 CORALIE 2457078.52769 51.199 15.000 CORALIE 2457466.50750 80.190 11.000 CORALIE 2457467.63531 86.190 12.000 CORALIE 2457550.50844 140.189 26.000 CORALIE 2457944.46112 158.879 9.000 CORALIE

* Note that RV is a modiﬁed RV. This is the RV measured by the instrument subtracted by the mean RV and secular acceleration.

29 ity precision of better than 10 m/s. The CORALIE data was reduced using the normal steps for echelle spectra, including debiasing the images, locating and measuring orders with polynomial fitting, flatfielding, removing scattered light, making a 2D barycentric corrected wavelength solution, cross-correlating the spectra with a binary mask suitable for the spectral type, and fitting the cross-correlation function to calculate a radial velocity. Finally, the instrumental drift was measured with a simultaneous observation of a Thorium-Argon lamp, and this drift was then subtracted from the velocity measurement. For this process we used the CERES pipeline package (Brahm et al., 2017).

HARPS is an even more powerful radial velocity machine, since this echelle spectrograph has a resolving power of „115,000 and is heavily optimized for thermal, pressure, and me- chanical stability by being housed in a mechanically stable vacuum chamber to negate the eﬀects of air pressure and temperature variations on the instrument. The HARPS data was reduced using the HARPS-DRS following the procedure outlined in Baranne et al. (1996) and Pepe et al. (2002), which follows similar steps to those outlines above for CORALIE. Long-term stability for the instrument has been demonstrated to be „ 1 m/s or better (see Lo Curto et al., 2010). More details of the radial velocity observations, pipeline processing, and computations for the CHEPS datasets can be found in Jenkins et al. (2017). For our observations for both stars, we used the G2 binary mask to cross-correlate with the observed echelle spectra. For both CORALIE and HARPS observations, the majority of the observations were taken with a S/N of 50-100 at 5500 A.˚

From CHEPS observations we have obtained 35 radial velocity observations using HARPS and CORALIE over 9.2 years for HD 86006 as shown in Table 3.2. Using the Systemic Console (Meschiari et al., 2009), we fit a long-period trend to the data, as shown in Fig. 2.1, where the filled circles are HARPS data and the open circles are CORALIE observations. We repeated the same exercise for the star HD 90520, shown in Fig. 2.2, where we have obtained 61 observations with the same instruments for this star, with a full baseline of over 6 years. While there may be some evidence for curvature in this data, a linear fit is also a viable solution, even though only an extreme lower limit can be placed on the companion period and minimum mass. We do not report the physical parameters of these solutions as they were only made to provide us with the orbital separation lower limit with a fixed eccentricity of 0, which are 16.8 AU and 6.7 AU for HD 86006 and HD 90520, respectively, and the fits return reduced χ2 statistics of 3.6 and 3.4,

respectively after removing outliers in the residuals. Therefore, HD 86006 and HD 90520 became ideal targets for observation in SV with SPHERE due to these long-period trends, their physical brightnesses, and their sky coordinates. Finally, we show the Julian dates, radial velocity measurements, measurement uncertainties, and the instrument used in Table 3.2 for HD 86006 and Table 2.3 for HD 90520. The discrepancy between the values shown in the Figs. 2.1 and 2.2 and Tables 3.2 and 2.3 are explained by instrumental oﬀsets determined for the CORALIE and HARPS data sets, used to combine the data into one set and best ﬁt an RV trend. Table 2.3: Radial velocity data for HD 90520.

JD RV (m/s) ˚ σRV Instrument 2454248.45544 -32.414 1.092 HARPS

30 2454248.45899 -33.627 1.045 HARPS 2454248.46261 -31.563 1.046 HARPS 2454249.45329 -27.746 6.817 HARPS 2454250.45131 -26.852 1.834 HARPS 2454250.45502 -21.344 1.855 HARPS 2454250.459 -20.758 1.566 HARPS 2454253.47072 -19.565 1.363 HARPS 2454253.47441 -13.223 1.917 HARPS 2454253.47883 -16.860 1.94 HARPS 2454577.52982 -10.284 0.456 HARPS 2454578.49819 -6.498 0.523 HARPS 2454578.72273 -4.004 0.602 HARPS 2454579.49081 -2.754 0.548 HARPS 2454580.66601 -3.297 0.896 HARPS 2454581.66212 -4.799 0.659 HARPS 2455188.66246 -2.663 0.735 HARPS 2455189.66311 -3.137 0.673 HARPS 2455190.64828 -4.274 0.657 HARPS 2455649.63021 13.689 0.558 HARPS 2455650.62848 16.429 0.689 HARPS 2455651.64304 13.590 0.616 HARPS 2455885.81022 18.426 0.844 HARPS 2455992.71069 15.473 0.537 HARPS 2455993.64336 21.557 0.672 HARPS 2455994.59803 23.014 0.958 HARPS 2456063.53402 18.569 0.782 HARPS 2456442.58337 27.677 0.915 HARPS 2456443.5437 20.111 0.979 HARPS 2456450.47734 17.525 0.818 HARPS 2456461.5162 23.315 0.577 HARPS 2456462.54358 24.858 0.769 HARPS 2456463.51781 29.659 0.902 HARPS 2455161.86317 -13.929 9.000 CORALIE 2455196.87138 -14.931 9.000 CORALIE 2455198.86911 -20.931 9.000 CORALIE 2455209.84632 -10.932 9.000 CORALIE 2455229.61835 -21.933 9.000 CORALIE 2455268.59634 -17.935 9.000 CORALIE 2455270.61985 -19.935 9.000 CORALIE 2455349.59621 -13.940 15.000 CORALIE 2455350.60811 -9.940 9.000 CORALIE 2455351.52453 -7.940 9.000 CORALIE 2455352.58114 8.060 9.000 CORALIE 2455514.85248 -29.949 9.000 CORALIE 2455516.86803 -20.949 9.000 CORALIE 2455552.86086 0.049 9.000 CORALIE 2455608.75706 8.046 25.000 CORALIE

31 2455608.83666 26.046 23.000 CORALIE 2455609.81951 20.045 9.000 CORALIE 2455610.82172 -3.955 9.000 CORALIE 2455611.8245 4.045 9.000 CORALIE 2455698.57298 10.040 9.000 CORALIE 2455877.84483 12.030 20.000 CORALIE 2455878.85582 10.030 9.000 CORALIE 2455879.83989 23.030 9.000 CORALIE 2456307.79163 2.006 9.000 CORALIE 2456308.79141 3.006 9.000 CORALIE 2456463.51059 28.997 9.000 CORALIE 2456464.49952 26.997 9.000 CORALIE 2456467.5313 23.997 15.000 CORALIE ˚ Note that RV is a modiﬁed RV. by the mean RV and secular acceleration. This is the RV measured by the instrument subtracted by the mean RV and secular acceleration.

2.2.2 Direct Imaging Observations

SPHERE Observations

We obtained observational imaging data of our targets HD 86006 and HD 90520 on January 4, 2015, and February 22 and 23, 2015 (SV time), during ESO Period 97 on April 6, 2016, and during ESO Period 98 on November 14, 2017, with long-slit spectroscopic data taken on April 6, 2016 and May 2, 2016 on SPHERE (Beuzit et al., 2008) on the VLT UT3 at Cerro Paranal in Chile (ESO Run ID: 60.A-9385(A), 097.C-0775(A)). We used SPHERE in IRDIFS mode, a mode making simultaneous observations with the Infrared Dual-band Imager and Spectrograph (IRDIS, Dohlen et al., 2008) and the Integral Field Spectrograph (IFS, Claudi et al., 2008a). We used the H23 dual-band filters (1583 nm, 1667 nm) for IRDIS (Vigan et al., 2010) and IFS in the Y-J part of the spectrum (R „ 50, 950 nm - 1350 nm). We used the long-slit spectrograph (LSS) mode to follow up with a medium-resolution (R „ 350) of HD 86006 (Vigan et al., 2008). Our SV observations were made with the four-quadrant phase mask (4QPM), while the P97 and P98 ones were taken with the apodized Lyot coronagraph (ALC) as the 4QPM is no longer offered. As the January 4, 2015 data was not centered properly and the LSS data of April 6, 2016 did not have the companion in the image, we have not used them in the final analysis.

All SPHERE IRDIS observations were taken with ﬂux calibration images that were done by oﬀsetting the PSF core so that the coronagraph is not covering it, which allows one to accurately measure the star’s photometry, and also with star center calibrations that allow accurate measurements of the star’s position for all images, which is done by making four symmetric satellite spots on the coronagraphic image where the intersection is the location of

32 Figure 2.1: Plot of radial velocity values for HD 86006. The closed data points represent data from HARPS, while the open points represent data from CORALIE. Note the linear trend without inflections making orbital characteristics difficult to determine. the star under the coronagraph. The flux observations used neutral density filters to lessen the light and not immediately saturate the detector.

The SPHERE IRDIS data was reduced in the typical way using the SPHERE Pipeline Recipes (v. 0.15.0, Pavlov et al., 2008) doing a background subtraction, flatfield division, star center determination, and deletion of bad pixels. The IRDIS detector does have a small anamorphic distortion, but as the distortion map calibration in the SPHERE pipeline does not report the appropriate mask offset, we use the distortion correction of Maire et al. (2016) by multiplying the Y coordinate by 1.0062 ˘ 0.0002. The SV data for HD 90520 was processed with angular differential imaging (Marois et al., 2006) and principal component analysis (Soummer et al., 2012) to be able to provide the best subtraction of the surrounding speckle halo and obtain the best possible contrast between a companion and its host star.

We used the centering calibration to measure the position of the primary star, as described above, for the SPHERE observations on Feb. 12, 2015. To obtain the position of the secondary star, we measured the centroid of a 2D Gaussian ﬁt to the PSF with a full-width at half maximum (FWHM) described by that of the primary star in the ﬂux calibration frame. To obtain a measurement of the separation between the star and the companion we used the pixel scale measured by Maire et al. (2016) as well as the parallax for the distance to provide us an absolute separation in AU, and we used their true north that was obtained using observations at a measurement date very close to our own to obtain the proper position angle.

33 Figure 2.2: Plot of radial velocity values for HD 86006. The closed data points represent data from HARPS, while the open points represent data from CORALIE. Note the linear trend without inﬂections making orbital characteristics diﬃcult to determine.

To obtain an uncertainty on the centroid measurement, we rotated and derotated each frame by 5 degrees multiplied by a random number from a normal distribution 10 separate times and used the standard deviation on the mean measurement. We then propagated the errors for each centroid measurement for each frame observed, as the frames were stacked. We propagated this error with the error on the detector distortion and the centroid measurements of the four satellite spots. We also propagated in the pixel scale error from Maire et al. (2016) to obtain an error on the separation between the primary and the companion. To obtain an error on the position angle, we propagated the angle measured by the centroid as done for the separation and error on the true north as measured in Maire et al. (2016).

For the same Feb. 12, 2015 data, we measured the flux of the primary star in the flux- calibration image by summing the detector counts for pixels within 1 FWHM of the centroid, correcting for the exposure time for the frame. We also corrected for the neutral density filter in place (SPHERE filter ND2.0 in this case, Vigan et al., 2010). To measure the flux of the companion, we measured the counts within 1 FWHM of the companion centroid in a science frame. We corrected for the exposure time and the speckle contribution by subtracting the flux using 1 FWHM apertures at the same distance as the companion surrounding the primary star under the coronagraph and took the mean of these values. These measurements provided us with the photometric contrast between the primary and the companion.

For SPHERE observations other than the one taken on Feb. 12, 2015, we measured the astrometry and photometry using the ﬂux-calibration image, as the companion is visible

34 in that frame but the coronagraphic images had the companion saturated. The separation and position angle was measured by ﬁnding the distance and angle between the centroids, respectively, taking the parallactic angle of the observation into account. For SPHERE, the plate scale for non-coronagraphic images and true north for the proper period of time were used from Maire et al. (2016). We used the Vortex Image Processing (VIP) package (Gomez Gonzalez et al., 2016). to implement our post-processing of our images, as described above.

MagAO Observations

We also made an observation of HD 86006 using the Magellan Adaptive Optics (MagAO) instrument (Close et al., 2008) at the 6.5 m Clay Telescope at Las Campanas Observatory in Chile with the Clio2 infrared detector on November 26, 2015. We made observations in the H and K bands with the narrow camera (pixel scale of „ 15.85 mas pixels). It should be noted that the H ﬁlter is read noise limited, rather than sky limited, with 70 electrons of read noise. The K band is sky limited. The weather conditions were not ideal during that night.

Our MagAO data was taken with an ABBA nodding pattern. We reduced the data by first subtracting each A frame by a B frame taken closest in time in the same mode to subtract the background and dark current. We then stacked the detector integrations and aligned the nods. We stacked the images after centering each image from a fit of the PSF with a Gaussian function. We did not apply any flatfield division, since there is out-of-focus image that is unremoveable by flat-correction (Morzinski et al., 2015).

To calibrate MagAO’s astrometry, we used a MagAO observation of 47 Tucanae that we compared with one from HST, eﬀectively allowing us to obtain a pixel scale and true north for this instrument; we describe this in more detail in Section 2.2.4. We propagate the astrometric error by considering the standard deviation of the centroids from separate clean observed images, along with the pixel scale error, and the standard deviation of the position angles between the two centroids, along with the true north error.

To measure the photometry, we measured the ﬂux within 1 FWHM aperture of the primary star and the ﬂux within 1 FWHM of the companion star. As with the SPHERE coronagraphic image, we subtracted the speckle background within the mean of apertures of the same radius and distance as the companion surrounding the primary in the center. The photometry error was measured by considering? that photon counts follow a Poisson distribution, and therefore we estimate the error as e´.

We also used the VIP package (Gomez Gonzalez et al., 2016) to implement the MagAO image post-processing. We summarize the observation dates, exposure times and modes in Table 4.1.

35 Table 2.4: Observations Table

Instrument Mode Date (DD/MM/YYYY) Universal Time JD Object Filter No. of Exposures Exposure Time (sec) Calibrations HST-WFC3 IR 18/03/2014 10:48:03 2456734.95004214 47 Tuc F110W 1 99 SPHERE IRDIFS 12/02/2015 03:51:07.591 2457065.66050452 HD 86006 H23 16x20 8 Star Center, Flux Calib. 36 SPHERE IRDIFS 13/02/2015 03:16:42.295 2457066.63660064 HD 90520 H23 8x20 16 Star Center, Flux Calib. MagAO Clio2 26/11/2015 02:15:45 2457352.59427 47 Tuc (center) Ks 20x2 1500 AB pattern MagAO Clio2 26/11/2015 07:33:05 2457352.81464 HD 86006 H 20x4 1500 ABBA pattern, 500 sec. Flux Imgs. MagAO Clio2 26/11/2015 07:41:14 2457352.82030 HD 86006 Ks 10x8 3000 ABBA pattern, 500 sec. Flux Imgs. SPHERE IRDIFS 06/04/2016 00:49:46.578 2457484.5356687 HD 86006 H23 1 64 Star Center, Flux Calib. SPHERE IRDIS-LSS 02/05/2016 01:02:29.29.699 2457510.53287012 HD 86006 MR-YJH 14 64 Flux Calib. SPHERE IRDIFS 14/11/2016 08:37:10.900 57706.3591539 HD 86006 H23 4 64 Star Center, Flux Calib. 2.2.3 Spectroscopic Reduction

In Table 4.1, we summarize the spectroscopic observations taken with SPHERE. To reduce the SPHERE IFS data, we used the pipeline from Vigan et al. (2015) and steps described in Mesa et al. (2015). This pipeline works by first subtracting the background frames and four sets of detector flats for the YJ mode which include those taken in white light, in 1020 nm, 1230 nm, and 1300 nm filters. Then the spectra positions were defined by a calibration with light illuminating the detector evenly through the instrument. To calibrate the positions of the wavelengths on the detector, three monochromatic lasers illuminated the detector through the system. The integral field unit (IFU) flat is used to correct the lenslet contribution to science images. The calibrations were processed using the SPHERE DRH Pipeline.

To analyze the SPHERE IFS data, we ﬁnd the position coordinates of the companion source for each spectral channel by using the daoﬁnd module, an implementation from the DAOPHOT algorithm described in Stetson (1987), in Photutils v. 0.2.2 (Bradley et al., 2016). Using these coordinates, we measured the FWHMs in the x and y directions, and used the mean as the FWHM for the PSF in each channel. The photons from the source were counted within 2 FWHMs of the centroid of the secondary companion and any residual background left after the background subtraction was subtracted using counts within apertures of 8.5 and 11.5 FWHMs from the centroid.

We reduced the SPHERE LSS coronagraphic data by using a reduction pipeline of Vigan (2016). The pipeline uses a combination of the standard SPHERE DRH recipes with custom IDL routines to provide reduced and aligned LSS spectra. The pipeline then offers different algorithms for the subtraction of the speckle pattern from the data before the spectrum of the companion can be extracted. For this object, we used an improved version of the method based on spectral differential imaging described in Vigan et al. (2008) as well as a simple subtraction of the symmetric speckle halo with respect to the star. After the speckle subtraction, the spectrum of the companion is extracted using an aperture of size λ{D in each of the spectral channels. The noise is estimated from an identical aperture located on a symmetric position with respect to the star. The spectrum of the companion is then normalised to the flux of the primary in each channel, extracted using a similar aperture. Since the companion is very bright compared to the stellar halo and speckles, the spectra extracted with the two speckle subtraction methods yielded completely equivalent results. Because IRDIS provides two identical fields of view in LSS mode, the spectra obtained in each field were combined with a mean to increase the signal-to-noise ratio of the companion spectrum.

To account for the telluric bands in the spectra, for both the LSS and the IFS, we divided the spectra extracted for the secondary companion by the primary PSF spectrum. Finally, the spectrum, calibrated in contrast, was multiplied by a Planck function at the primary star’s temperature (5620 K for HD 86006) to compensate for its spectral slope that was divided out in the telluric calibration. We compare the resulting spectrum to models and empirical objects in Section 2.3.3.

37 Table 2.5: Photometry and Astrometry of HD 86006B

Instrument Mode JD Filter Separation (mas) Position Angle (deg) ∆mag SPHERE IRDIS 2457065.66050 H2 331.46 ˘ 2.51 301.20 ˘ 0.13 4.21 ˘ 0.04 SPHERE IRDIS 2457065.66050 H3 331.53 ˘ 2.51 301.35 ˘ 0.13 4.12 ˘ 0.04 MagAO Clio2 2457352.81464 H 319.69 ˘ 9.53 302.28 ˘ 2.23 3.78 ˘ 0.14 MagAO Clio2 2457352.82030 Ks 314.49 ˘ 6.74 300.83 ˘ 1.26 3.82 ˘ 0.22 SPHERE IRDIS 2457484.53567 H2 319.79 ˘ 2.35 301.57 ˘ 0.20 4.20 ˘ 0.05 SPHERE IRDIS 2457484.53567 H3 320.33 ˘ 2.36 301.47 ˘ 0.20 4.13 ˘ 0.05 SPHERE IRDIS 2457706.35915 H2 313.33 ˘ 2.31 301.24 ˘ 0.19 4.01 ˘ 0.05 SPHERE IRDIS 2457706.35915 H3 313.55 ˘ 2.31 301.32 ˘ 0.20 4.00 ˘ 0.04

2.2.4 HST Astrometric Calibration

47 Tuc at RA: 00:23:58.12, Dec: -72:05:30.19 (J2000) is one of the astronomical objects used by SPHERE to calibrate it for astrometric measurements (Maire et al., 2016). To compare the MagAO and SPHERE images, we made observations of the cluster to calibrate MagAO’s astrometric measurements to that of SPHERE. MagAO’s observation was taken at a diﬀerent point in the cluster (RA: 00:24:05.36, Dec: -72:04:53.20, J2000) making it necessary to compare it to other calibrated observations such as those from the Hubble Space Telescope Wide Field Camera 3 (WFC3). We used the AstroDrizzle-reduced .drz image to correct for the distortion in the WFC3 detector.

We used two stars in the image and measured their separation in sky coordinates given from the frame. From this we could calibrate the pixel lengths to distances in arcsecs from the HST observation. We also calculated the position angle of one star relative to the other to obtain a true north for the MagAO system. This allows us to compare the observations with SPHERE as the measurements are then on an absolute scale.

From this we obtained a pixel scale for MagAO of 15.764 ˘ 0.125 mas/pix and a true north of -2.403 ˘ 0.033 °.

2.3 Results

2.3.1 HD86006 Companion Detection

We made the detection of a companion to HD 86006 during the SV period as shown in Fig. 2.3. This companion has an average contrast of ∆H2 = 4.14 ˘ 0.03 mag, ∆H3 = 4.08 ˘ 0.03 mag and a separation of 0.331 arcsec on the SV observation. We show the astrometric and photometric parameters and uncertainties in Table 2.3. The follow-up observations of the star with MagAO and SPHERE show that there is common proper motion as there is little change in the position for each image. In Fig. 2.4, we show the astrometric positions of the detected companion along with the track of where the companion would be if it were a background source moving slowly across the sky. As the companion’s astrometric positions do

38 Table 2.6: HD 86006 System Characteristics Characteristic HD86006A HD86006B Spectral Type G5IV/V M3.7 ˘ 1.1 a M4.5 ˘ 1.8 b M3V c d Teff pKq 5700 3300 ˘ 100 3600 ˘ 100 e `201 f 3258´180 `320 g 3127´370 a H2OA index b H2OC index c Best fit SpeX spectrum to LSS d Best fit BT-Settl model to LSS e Best fit BT-Settl model to IFS f H2OA index converted to Teff g H2OC index converted to Teff

Figure 2.3: The left plot shows an image of HD 86006 with SPHERE IRDIS and the right is from MagAO. The M dwarf companion can be seen to the upper right of the star at a separation of „26 AU from the primary.

39 Figure 2.4: Plot of the astrometric positions of the companion to HD 86006 from our data with SPHERE and MagAO, along with a model trend of where the companion would be if it were a background source ending in November 2016 (red curve). The astrometric points are given for each epoch in diﬀerent colors for the instrument and data, as shown in the legend. not follow the track, the companion is gravitationally bound to its host star. The companion shows movement between the three SPHERE points and MagAO point.

2.3.2 Spectral Indices

We measured the temperature and spectral type of HD 86006B using empirically derived spectral indices. We considered the H2OA, H2OC indices from McLean et al. (2003), which were shown to well-represent early M stars in Cushing et al. (2005). These indices correspond to water absorption features and have a significant effect in low-mass stars. As these features are shown to relate to the spectral index of these low-mass stars, we can use empirical relations to derive a spectral index and from that a temperature of the companion. To calculate them, we used the LSS spectrum and took the median values for flux ranges in the reduced spectrum.

We use the following relations to obtain a spectral type from McLean et al. (2003):

SpT = - 26.18 (H2OA) + 28.09 SpT = - 39.37 (H2OC) + 38.94 The H2OA index corresponds to that of an M 3.7 ˘ 1.1 and the H2OC index shows that

40 of an M 4.5 ˘ 1.8. To convert these spectral types into Teﬀ , we interpolated between M dwarf spectral types and corresponding Teﬀ from Table 5 from Pecaut & Mamajek (2013). `201 `320 This gave 3258´180 K and 3127´370 K for the H2OA and H2OB indices respectively, with the precision based on the index precision, as shown in Table 2.6.

2.3.3 Model Comparison

To obtain model-derived characteristic values for the companion, we used the BT-Settl models (Allard et al., 2012) using the solar abundance from Asplund et al. (2009). [On BT-Settl webpage, this is under model group AGSS2009.]

To see which BT-Settl model fits the observed spectra we obtained for the companion HD 86006B from the IFS and LSS of SPHERE, we convolved the model spectra to match the lower resolutions of the observations. We used a Gaussian maximum likelihood estimation of all models for low-mass stars with temperatures between 2600 and 5600 K, [Fe/H] either 0, 0.3, or 0.5 dex, and log(g) between -4.0 and -4.5 dex. For the IFS spectrum (shown in Fig. 2.5), the best fit obtained was 3600 K temperature given at a log(g) of -4.5 dex and [Fe/H] of +0.3 dex. In this, we see that the slope matches relatively well from 1150 nm to 1350 nm. At bluer wavelengths there is some divergence where the model rises higher than the observed SED. We also show in Fig. 2.5 the model for 3300 K, log(g) of -4.0 dex and [Fe/H] of +0.5 dex. Here visibly the SED for a high-metallicity cooler model matches very well in the blue from 950 nm to 1150 nm, but the slope diverges past 1150 nm to redder wavelengths. For the LSS spectrum, the maximum likelihood fit gave a 3300 K model with a log(g) of -4.5 dex and [Fe/H] of 0.0 dex as shown in Fig 2.6. We use 100 K uncertainties for our best fit, given that the spectra are provided in 100 K intervals.

We also make use of stellar spectra from the NASA Infrared Telescope Facility (IRTF) Spectral Library (Cushing et al., 2005; Rayner et al., 2009) taken with the SpeX medium- resolution spectrograph at Mauna Kea. We made a visual comparison of the LSS spectrum with the spectra in the Library as shown in Fig. 2.7. The M3V spectrum of Gl 388 appears to provide the best fit which has an effective temperature of 3390 K and [Fe/H] of 0.28 (Rojas-Ayala et al., 2012). This agrees quite well with our measured effective temperature from H2O indices, along with the metallicity of the host star (+0.33 ˘ 0.06 dex; Soto & Jenkins, 2018).

To obtain physical parameters for the companion, we used the Isochrones code (Morton, 2015), written in Python, which makes use of MultiNest (Feroz et al., 2009), implemented as PyMultiNest (Buchner et al., 2014). To make estimates of the age and mass, we provided the code with isochrones from an evolutionary model and used the interpolation from the Isochrones code. We put many random points on the isochrone, cut the area that agreed with our magnitude and temperature observations, and used these points to estimate the mass and age.

Using the AMES-Cond (Allard et al., 2001; Baraﬀe et al., 2003) with [Fe/H] = 0.0 dex and BT-Settl (Allard et al., 2012) with [Fe/H] = 0.3 dex model isochrones, we used the measured the SPHERE H2 and H3 photometry for the three epochs added with the Cutri

41 º

Figure 2.5: The best ﬁt 3600 K model (black) from the BT-Settl to an SED from the IFS of HD 86006 (blue). This model also used a metallicity of +0.3. We also show a 3300 K, +0.5 metallicity model as it visually matches very well at the bluer wavelengths.

42 Figure 2.6: The comparison BT-Settl spectra (3200 K, 3300 K, 3400 K) in black with the LSS spectrum in red. The best ﬁt using the maximum likelihood between all of the spectra was the one at 3300 K which also appears the best visually. For clarity, we have shown the spectrum with the water absorption feature removed.

43 Figure 2.7: The LSS spectrum of HD86006B in red compared with the SpeX spectra in black. The Gl388 spectrum is visually a good ﬁt to the LSS spectrum. An M2.5V (Gl381) and an M3.5V (Gl273) spectrum are also shown to compare. For clarity, we have shown the spectrum with the water absorption feature removed.

44 Table 2.7: HD 86006B Model Physical Characteristics Characteristic AMES-Cond BT-Settl `0.09 `0.10 Mass (Md) 0.23 ´0.07 0.23 ´0.04 `0.34 `0.34 log(Age [yr]) 7.38 ˘ ´0.30 7.43 ´0.25

et al. (2003) photometry for the primary star and derived temperature to find a mass and age for the companion. We note that the AMES-Cond model is recommended for objects with with Teff ă 1700, which is less than that of our discovered companion, while the BT-Settl model is considered valid from the stellar to planetary regimes. We use Teff of 3321 ˘ 111 K, which was obtained by taking the mean of the Teff s for HD86006B found by the BT-Settl best fits for the IFS and LSS spectra (3600 K and 3300 K, respectively) and the two H2O `0.10 indeces McLean et al. (2003). From the BT-Settl isochrone, we obtain 0.23 ´0.04 Md for the mass and a range of ages between 15 and 59 Myr which we show in Fig. 2.8. For the `0.09 AMES-Cond isochrone, we derive a mass of 0.23 ´0.07 Md and age between 12 and 53 Myr which is also below the ages derived from the primary star. We show the obtained model isochrone parameters in Table 2.7.

We can use the age derived from the primary as a proxy age for the M dwarf as we can assume that they had formed together. This allows us to make a test on the isochrone- derived age of the companion, but it should be noted that ages tend to be poorly constrained parameters in general.

The companion’s isochrone age estimate is very low, compared to the ages derived from 1 the primary star (3.67 ˘ 0.39 Gyr and 7.95 ˘ 0.90 Gyr), which also has a low log RHK of -5.05 dex suggesting an old age of Gyrs (Mamajek & Hillenbrand, 2008). This clear inconsistency between the measured ages for the M dwarf companion and the primary gives an indication that the models do not well constrain this parameter. To investigate further this discrepancy, we also plot the absolute H band magnitudes of solar neighborhood M3 stars with spectroscopically measured Teff s from Rojas-Ayala et al. (2012) in Fig. 2.8. Even though nearby stars, especially M dwarfs, tend to be old, on the order of Gyrs, their positions on the HR-diagram are scattered from ą 10 Gyr down to ă 0.02 Gyr of age, with the majority found to be clustered around or below the 0.1 Gyr isochrone. This shows that currently the model isochrones for low-mass stars do not necessarily give precise age estimates, likely significantly underestimating the ages of mid-M stars. We also factor in the spectroscopic metallicities ([Fe/H]) from Rojas-Ayala et al. (2012), dividing the sample into two groups, with a dividing line set as 0.1 dex in metallicity to represent the metal-rich and metal-poor samples, and we see that there is no clear difference in the groups’ positions on the isochrones, indicating that the shift towards a lower age estimate for HD 86006B is not due to its super metal-rich nature, as known from the primary star’s [Fe/H] of 0.33 ˘ 0.06 dex. As the ages determined for HD 86006B by the isochrones is shown to be inconsistent, we are not assured that the measured masses are reliable.

45 Figure 2.8: The isochrones for BT-Settl models at masses varying from 0.03 Md to 0.6 Md and ages from 1 Myr to 12 Gyr as labeled. In black, we give the SPHERE H2 photometry of HD 86006B and the range of temperatures by taking the average of the temperatures found from SED ﬁtting and spectral indices. We also show various solar neighborhood M3 dwarfs with their photometries and their temperatures and metallicities (for which we divide into two groups, with green representing low metallicity and blue representing high metallicity) from Rojas-Ayala et al. (2012).

46 2.3.4 Model Fitting

We developed a code in Python to make a joint analysis using a Markov-chain Monte Carlo method (MCMC) that combines the RV data and DI data, in order to best constrain an orbital solution. To do this, we developed a code that uses an orbital model based on Kepler’s laws of planetary motion as given by the following Keplerian elements:

a, semi-major axis,

e, eccentricity,

i, inclination,

Ω, longitude of the node,

ω, argument of periastron,

ν, true anomaly at a given epoch

This Keplerian orbit is represented by a simple ellipse with the primary star being at a focus:

x “ b cos ν

y “ a cos ν

z “ 0

The distance from the center to a focus of the ellipse is given as ? c “ a2 ´ b2

The ellipse is multiplied by a rotation matrix to transform it into a new ellipse that represents a projection of the real ellipse on the sky.

x1 x y1 “ R y »z1fi »zfi – fl – fl where

cos ω cos Ω ´ sin Ω sin ω cos i sin ω cos Ω ´ cos ω cos i sin Ω sin Ω sin i R “ cos ω sin Ω ` cos Ω sin ω cos i ´ sin Ω sin ω ` cos Ω cos ω cos i ´ cos Ω sin i » sin i sin ω cos ω sin i cos i ﬁ – ﬂ

47 x1 and y1 give the new coordinates for our projected orbit on the sky.

The semi-major axis, b, is deﬁned as

? b “ a 1 ´ e2

As we want the orbit to be deﬁned by the time (in Julian Date) of the astrometric points rather than by angle, we want to convert the angle (or true anomaly) to the time of periastron passage, T0. We do this by converting the true anomaly, ν, to the eccentric anomaly, E, and then the mean anomaly, M.

? 1 ´ e2 sin ν tan E “ 1 ` e cos ν

M “ E ´ e sin E

We can use the mean anomaly, as it is the position angle of the companion in if it were on a circular orbit with the same period and speed, and the mean angular motion n given by deﬁned by a gravitational parameter, to ﬁnd the time of its position, with t0 being the time it passes periastron.

GpM ` M q n “ 1 2 a3 c M ` nt t “ 0 n

where G is the universal gravitational constant, M1 is the mass of the primary star, and M2 is the mass of the secondary companion.

The radial velocity equation, based on Kepler’s third law of planetary motion, that was used as part of our model to ﬁt the radial velocity data, giving us the mass parameter and further constraining the orbital elements, is as follows:

G vr “ v0 ` 2 M2 sin ipcos ω ` ν ` e cos ωq a2pM1`M2qp1´e q b where vr is the radial velocity, v0 is the barycentric velocity, G is the universal gravitational constant, a2 is the semi-major axis of the orbital companion, M1 and M2 are the primary and secondary masses respectively, i is the inclination, e is the eccentricity, ω is the argument of periapsis, and ν is the true anomaly. For the astrometric orbit, Ω, the longitude of the node is also relevant.

Also we considered that the eccentricity and argument of periapsis are parameters that

48 Table 2.8: HD 86006B Keplerian Elements Keplerian element Prior Value `1.54 a (AU) Up10, 40q 27.44´1.57 1 `0.13 e 0.52´0.10 2 `9.36 i (°) Up40, 89.74q ` Up90.26, 120q 56.13´6.21 1 `10.85 : e and ω are tied Ω(°) Up170, 250q 181.62 ´7.80 1 `9.91 ω (°) 11.80´7.67 `1524 t0(days) Up2432000, 2442000q 2434540´1590 `0.067 M 2 (Md) Up0.2, 0.7q 0.365´0.055, parameters as described in Section 3.5. The prior we set was a uniform distribution where e2 ` ω2 ă 1 2: The priors do not include a small area near 90 °as the code cannot calculate a completely edge on orbit.

show a high degree of degeneracy, especially as the eccentricy approches 0, so we combine these two parameters to make new ones that are dependent on e and ω, as in Marsh et al. (2014), with the following equations:

? ? x “ e cos ω and y “ e sin ω e “ x2 ` y2 We used a 2D Gaussian likelihood function with correlated parameters for the astrometric model and a 1D Gaussian likelihood function for the radial velocity model with consideration given to correlated noise. To give a good starting point for the MCMC code to run, we used trial and error inputting different values for the parameters to get a close fit to the model and gave flat priors that were near these values. These ”close” parameters with a small random value added were used as our starting values to run the code. After running the chain with emcee (Foreman-Mackey et al., 2013), we could obtain posterior probability distributions on the parameters. We used these distributions to obtain our estimations of the parameters at the 16th, 50th, and 84th percentiles which we used as their uncertainties.

When running the code using the uniform priors shown in Table 2.8, we obtain a Keplerian `1.75 fit for the companion to HD 86006 with a semi-major axis of 23.26´1.80 AU, an eccentricity `0.12 `0.064 of 0.68´0.12, mass of 0.321´0.058 Md, and other parameters shown in the same table. This mass is measured to 20%, which is higher than the ă 10% mass measurements used in the mass-luminosity relation of Delfosse et al. (2000), making more data necessary to further constrain the mass. Fig. 3.12 shows the correlations each of the different parameters. We see correlation between the semi-major axis and the time of periastron passage. We also see some possible correlation between a and M2, i and Ω, and t0 and M2. Fig. 2.10 shows the RV fit, and Fig. 2.11 shows the astrometric fit. We see that the fit is not very well constrained and will need to await more data before being able to make a confidently well-constrained dynamical mass.

49 2.3.5 HD 90520

The SPHERE SV observations of HD 90520 did not produce a detection of a companion, but did provide us the best contrast performance for the SV run. We were able to place a strong constraint on the maximum mass and separation of the hidden companion using the contrast limit produced by an ADI-PCA method, along with accounting for small sample statistics close to the star (Mawet et al., 2014) The contrast curve is shown in Fig. 2.12, which demonstrates SPHERE’s sensitivity for this data set. Considering that the stellar SED agrees with that of spectral type of a G0IV/V star, 11 mag contrast in the H2 band at 200 mas corresponds to a maximum mass of 0.07 Md and a minimum mass detection of 0.06 Md at 500 mas. This means that at these very close separations we can rule out any stellar companion giving rise to the trend we ﬁnd in the RV measurements. Therefore, this target must be a brown dwarf or planetary companion if it lies at a separation outside of 200 mas. From our RV simulation with the Systemic Console, we are able to ﬁt a 2.2 MJup planet with a 6.6 AU semi-major axis assuming an eccentricity of 0 and 90° inclination at a reduced χ2 of 3.4 as shown in Fig. 2.2. This will make for an ideal target to observe with future adaptive optics instrumentation.

2.4 Summary and Conclusions

We used SPHERE SV in the H23 dual-ﬁlter IRDIFS mode to image directly the stars HD 86006 and HD 90520, chosen for their long-period linear radial velocity trends from the CHEPS survey. HD 86006 provided an M dwarf companion at about 25 AU from its primary which we have conﬁrmed with following observations with MagAO and SPHERE. The IFS and LSS spectra were used to characterize the spectral type of the star with spectral indices and using the BT-Settl. The analysis of the LSS spectrum agreed with the expected spectral type from the contrast.

We used spectral fitting, spectral indices, and model isochrones to derive a mass, temperature, and age for the companion. The spectral fitting agreed with the star being M3.7 or M4.5, while our best fit template to the LSS spectrum agreed with that of an M3V star. The temperatures also agreed with that assessment, giving about 3300 K, or about M3-M4 in spectral type (Pecaut & Mamajek, 2013). The ages derived from isochrone fitting were significantly lower than the expected age by over two orders of magnitude, and when comparing to nearby field M dwarfs with spectroscopically derived temperatures and metallicities, this was found to be the case also, indicating that theoretical models of mid-M dwarf evolution still require some fine tuning to match the observations. Due to this, we do not assume the isochrone mass measurements to be reliable.

We were able to also make a preliminary orbital solution for the data points taken with imaging. We will, over time, be able to map out the orbit of HD 86006B with imaging instruments, which will be able to provide a dynamical mass for the M dwarf. This discovery represents the ﬁrst low-mass companion detection from our sample of long-period radial velocity trends from the CHEPS and EXPRESS samples. As the stars in the CHEPS sample are intentionally biased to be high metallicity, and the EXPRESS sample also seems to favour

50 Figure 2.9: Corner plot for the MCMC astrometry and radial velocity ﬁts. We show the 2D posterior probability distributions for each parameter with each of the others to demonstrate how well they correlate along with the 1D distribution for each at the top of their set of plots. Vx, Vy, and Vv are paramters representing the systematic noise or jitter for the x and y astrometric dimensions and the radial velocities, respectively.

51 Figure 2.10: Plot of the fit from the MCMC code to the RV data. The red points shown with their error are the observed radial velocities and are the same as the points show in Table 3.2. The black curves represent one fit from the MCMC chain. As all fits after the burn-in were similar for the radial velocities, they nearly appear as a line in this plot. metal-rich giants, our SAFARI project will prove to be fundamental to help constrain the mass-luminosity relation for metal-rich stars, of which there are few (Delfosse et al., 2000; Gaidos et al., 2014; Newton et al., 2014; Terrien et al., 2015). These new dynamical masses will then contribute heavily to gain a better understanding of the physics of low-mass star formation, structure, and evolution.

52 Figure 2.11: Fit to the astrometric points from SPHERE and MagAO from the MCMC code. The red points represent the position of observations, while each black line is the ﬁt output from the MCMC run. The radial scale is in AU.

53 Figure 2.12: The 5σ contrast performance curve over the separation from the SPHERE IRDIS SV observations of HD 90520. We also show the instrument’s performance as a companion mass detectability limit (right axis) using the AMES-COND model to convert between magnitude and mass. We show where the directly imaged planets β Pic b (Lagrange et al., 2010; Bonnefoy et al., 2013), HR 7899 bcde (Marois et al., 2008; Zurlo et al., 2016), and 51 Eri b (Macintosh et al., 2015) would lie with respect to the contrast performance of these observations. We also highlight the position of our detection of HD 86006B and of the expected stellar to substellar transition for old objects.

54 Chapter 3

SAFARI II: A Pair of Ultracool Dwarf Companions to a Planet-Hosting Red Dwarf

Authors: Blake M. Pantoja, James S. Jenkins, Julien H. Girard, Ga¨elChauvin, Bartosz Gauza, Ricardo Ram´ırez,Mikko Tuomi, and Mat´ıasI. Jones In preparation for submission to the Astrophysical Journal

Ultracool dwarfs represent the boundary between the least massive stellar objects and substellar objects. We announce the discovery of two ultracool dwarfs at the M/L boundary to the super-Earth hosting M dwarf star GJ 3634. The companions were discovered using MagAO, whereby observations were made due to the presence of a long-period radial velocity trend found by HARPS, after fitting for the known planetary companion. With SPHERE imaging, we were able to confirm the pair are gravitational bound to the planet host, at a physical separation of 38 AU from the primary. The pair themselves are separated by 0.1 arcsec on the sky, which relates to a separation of 2 AU. By fitting the spectral energy distribution to template spectra and calculating spectral indices, both using SINFONI obser- `60 vations, we derive spectral types of M9.5 and L1.5 with Teff s of 2273´54 K for the brighter `42 Ba component and Teff s of 2085´40 for the fainter companion, respectively. Therefore, the GJ 3634 triple system represents a unique opportunity to further study planet and brown dwarf formation and evolution mechanisms.

3.1 Introduction

3.1.1 Sub-stellar Objects

Ultracool dwarfs represent both the lowest-mass stars and the substellar brown dwarfs. For the most part, and unlike stars, brown dwarfs never have suﬃcient mass for nuclear fusion of hydrogen in their cores, but instead tend to burn deuterium for some period of their lives.

55 While this sub-stellar boundary is expected to present itself at a mass of around 0.07-0.08 Md (Hayashi & Nakano, 1963, Burrows et al., 1997), there is diﬃculty in determining if an object lies above or below this limit. For example, young brown dwarfs may display atmospheric spectral signatures similar to those of old M dwarfs, and then their spectra change as they cool and dim over time. These characteristics are unlike the relative consistency of the temperature and luminosity of low-mass stars over long („ Gyr) timescales (Burrows et al., 1997, Burrows et al., 2001).

L dwarfs were ﬁrst diﬀerentiated from M dwarfs using spectral features witnessed in observations of GD 165B (Becklin & Zuckerman, 1988). This star is missing the conspicuous TiO and VO optical bands present in late M dwarfs, but does possesses FeH (8692 A)˚ and CrH (8692 A)˚ bands. The K I doublet (7665, 7699 A)˚ also appears as a broadened single line compared to its late M counterpart.

The most definitive test to distinguish stars from substellar objects is the lithium test. As stars will burn their lithium within about 100 Myr, brown dwarfs less than 60 MJupiter will not, as their cores will not reach high enough temperatures. We can therefore say that any object with a spectral type later than M7 (where we typically define the ”ultracool” dwarf region), and with a detection of lithium (6708 A),˚ is a substellar brown dwarf. This was first found with PPl 15 (Basri et al., 1996).

Of particular interest are so-called ”benchmark” brown dwarfs, or brown dwarfs in a system with another companion where parameters such as age and metallicity can be well determined and assumed for the brown dwarfs themselves. Dynamical mass can be determined by Kepler’s equations by combining the astrometric tracking of the orbit from imaging, with radial velocity (RV) observations (or from ﬁnding the barycenter when each individual orbit can be tracked, hence allowing the knowledge of each dynamical mass). A handful of these benchmark brown dwarfs have been discovered, including those of Crepp et al. (2012, 2016); Dupuy et al. (2015). These objects are crucial for constraining evolutionary and atmospheric models for substellar object. As there are, as of yet, few benchmark brown dwarfs, any additional objects provide information on the formation mechanisms of substellar companions, as mass deﬁnes the luminosity and age.

3.1.2 Radial-velocity Measurements

Doppler spectroscopy has been very popular in recent years for its successful use in the discovery of exoplanets. To demonstrate this, let there be the case where RVs have been observed by a target for a long period of time and only a linear trend has been detected. While we cannot constrain the orbital parameters in this case, we can be sure that the period of the companion inducing the trend must be longer that twice the time baseline. This means that the companion must be at a far separation from the car. These targets are therefore useful to follow up with direct imaging, as this method has a bias towards the detection of companions at large separations from the host star. Also, these two methods can be combined as RVs can provide only a minimum mass (M sin i), while orbital astrometric tracking can provide the inclination, i, to obtain an absolute dynamical mass for the object.

56 Low mass M dwarf stars themselves make for some of the best objects to search for planets and substellar companions. As their masses and luminosities are the lowest of all stars, detecting companions by RVs, direct imaging, and transits is easier when compared to higher- mass stars. For example, the recent planet discoveries in the systems of Proxima Centauri (Anglada-Escud´eet al., 2016), TRAPPIST-1 (Gillon et al., 2017), and Barnard’s Star (Ribas et al., 2018) reﬂect the gains that can be made here. Any detected companions have a dependency on the star’s parameters, such as mass, radius, and luminosity, making a well characterized mass-luminosity relation with independently measured masses and luminosities crucial, as investigated by Henry & McCarthy (1993), Delfosse et al. (2000), and Benedict et al. (2016).

Currently there are only a few examples of exoplanet host stars that also host widely orbiting ultracool and brown dwarf companions. These include a T7-8 companion to HD3651B (Mugrauer et al., 2006a) at a 480 AU separation and a T4.5 companion to HIP70849 (Lodieu et al., 2014) at greater than 100,000 AU.

The SAFARI (SPHERE Ao Follow-up of Additional Radial velocity companIons) program started with our detection and confirmation of a M3 companion to a solar-type star (Pantoja et al., 2018), first discovered to possess a linear trend from the Calán-HertfordshireExtrasolar Planet Search (CHEPS; Jenkins et al., 2009). The goal of this program is to follow up long period RV trends with direct imaging, with the long term goal being to track their orbits and obtain dynamical masses. With the expansion of targets from the sample of M dwarfs that show RV linear trends from Tuomi et al. (2019), we report the detection and characterization of two close ultracool dwarfs to GJ 3634. GJ 3634 is a M3 dwarf star (Gaidos et al., 2014) that hosts a m sin i of 7.0 MC super-Earth planet on a short 2.6 day period (Bonfils et al., 2011). In S 2, we discuss the observations taken and reduction process for both the RV and direct imaging observations. In S 3, we show the results from SED fitting, spectral indices, parameter measurement from isochrones, and orbital fitting the astrometry. In S 4 we give a discussion of the results and conclusions.

3.2 Observations and Reduction

GJ 3634 was chosen as a target from the program of Tuomi et al. (2019). This program consists of M dwarf stars as targets for a Doppler spectroscopic search for planets. These stars make for interesting planetary hosts considering that habitable-zone super-Earths are easier to detect around these stars than earlier-type ones, as the lower mass ratio provides larger amplitude signals at the planet-star separations where water can exist in liquid form. GJ 3634 is an early-type M dwarf star with a mass of 0.45 Md, [Fe/H] of -0.04 and Teff of 3330 K (Santos et al., 2013). It was discovered to host a super-Earth with mass m sin i of 7.0 MC with a 2.6 day period (Bonﬁls et al., 2011). This star was chosen as part of our SAFARI sample as it exhibits a long period linear acceleration, besides the clear planetary signal, as shown in Fig. 3.1. These long period trends make for good targets for direct imaging programs as they are likely to have a relatively wide separation from their primary, maximizing the possibility of obtaining a positive result. In Table 3.1, we show the stellar characteristics for GJ 3634, and the planetary characteristics for GJ 3634b.

57 Table 3.1: Characteristics of GJ 3634 and GJ 3634b. Property Value Reference GJ 3634 A RA (J2000) 10:58:35.088 1 Dec. (J2000) -31 08 38.201 1 P.M. RA (mas/yr) -566.861 ˘ 0.105 2 P.M. Dec. (mas/yr) -91.371 ˘ 0.089 2 V (mag) 11.93 ˘ 0.02 3 G (mag) 10.923 2 J (mag) 8.361 ˘ 0.023 4 H (mag) 7.762 ˘ 0.049 4 K (mag) 7.470 ˘ 0.027 4 Parallax (mas) 49.0440 ˘ 0.0756 2 Sp. Type M2.5 5 [Fe/H] -0.04 6 Teff (K) 3330 ˘ 174 6 Mass (Md) 0.45 6 log g (cm/s2) 4.83 6 log(R’HK ) -5.173 ˘ 0.135 7 vsin(i) (km/s) 0.85 8 GJ 3634 B Semi-amplitude (m/s) 5.59 ˘ 0.55 9 Period (days) 2.64561 ˘ 0.0006 9 `0.9 m sin ipMC) 7.0´0.8 9

1: Gaia Collaboration et al. (2018) 2: Gaia Collaboration et al. (2018) 3: Zacharias et al. (2013) 4: Cutri et al. (2003) 5: Gaidos et al. (2014) 6: Santos et al. (2013) 7: Astudillo-Defru et al. (2017) 8: Houdebine et al. (2016) 9: Bonﬁls et al. (2011)

58 Figure 3.1: Plot of the radial velocity data over time for GJ 3634. In grey, the ﬁt model from EMPEROR is shown including the short period signal from the planet GJ 3634b and the linear trend acceleration.

3.2.1 HARPS Radial Velocity Observations

The RV observations for GJ 3634 were performed over 4 years (from March 2009 to March 2013) using the High Accuracy Radial velocity Planet Searcher (HARPS) at the 3.6 m Tele- scope (Mayor et al., 2003) at La Silla Obervatory.

HARPS is a ﬁber-fed echelle spectrograph capable of performing precise RV observations. With a resolving power of „ 115,000 and spectral range of 378 nm to 691 nm, HARPS is housed in a vacuum chamber to maintain long-term stability against varying air pressure and temperature. After the reduction of the data from the HARPS DRS (Baranne et al., 1996; Pepe et al., 2002), the RVs were extracted by using a high SNR template with HARPS- TERRA pipeline (Anglada-Escud´e& Butler, 2012). Over the long term, HARPS observations have been shown to provide a high precision of stability at the 1 m/s level (see Lo Curto et al., 2010). For more information on HARPS reduction and calibration, see (Jenkins et al., 2017).

We have taken the 35 HARPS RV observations of GJ 3634 and used the EMPEROR code (https://github.com/ReddTea/astroEMPEROR; P. PeñaRojas et al. 2019, in preparation) to confirm and constrain the long period trend, along with the planet previously detected planet. EMPEROR quickly confirms the 2.6 day signal, along with an acceleration of 8.84 ˘ 0.32 m/s/year, which we show along with the data in 3.1. This significant linear trend made the star a good target for searching for companions by direct imaging. All RVs we analysed in this work are shown in Table 3.2.

59 Table 3.2: Radial velocity data for GJ 3634.

JD RV (m/s) * σRV 2454915.70975032 -1.064 5.800 2454932.6160119 -9.227 1.519 2454933.59600648 -17.037 1.071 2454934.57996961 -4.775 1.862 2454935.60739073 -15.458 1.498 2454936.61886197 -13.427 1.066 2454937.58616678 -9.826 1.746 2454938.63577069 -18.629 1.229 2454939.62624349 -8.594 1.880 2454940.64010538 -9.320 1.427 2454941.59307724 -15.586 1.076 2455231.82528481 -3.866 2.720 2455234.69968604 -6.999 2.156 2455236.83937496 -2.877 1.789 2455237.80772492 -6.667 1.302 2455238.83697972 2.160 1.037 2455239.79165781 -6.835 1.265 2455240.75711082 -4.299 2.128 2455241.74653379 1.174 1.424 2455241.88258196 0.969 1.519 2455242.67200584 -4.884 1.062 2455242.872455 -5.567 1.179 2455243.70451876 2.278 1.527 2455243.86572979 2.558 1.372 2455244.74046803 2.644 0.987 2455244.86182756 0.394 1.048 2455245.72825418 -5.729 1.537 2455245.85334057 -2.949 1.549 2455248.58877715 -8.494 1.412 2455248.76025379 -5.984 1.351 2455248.90744704 -3.886 2.424 2455272.67493779 -3.126 1.487 2455275.69126999 2.285 1.827 2455276.60366713 -2.875 2.081 2455277.67914168 -4.710 1.270 2455278.67145504 6.017 1.658 2455279.65482028 -5.740 2.643 2455281.60269262 3.359 1.275 2455284.67845363 -1.663 1.527 2455295.60381119 -9.729 1.446 2455296.6027665 1.443 1.875 2455298.63578787 -5.446 1.254 2455299.63316516 5.711 1.834 2455300.57068197 3.179 2.161

60 2455301.55797609 -2.627 1.929 2455349.50313664 -4.259 2.265 2455351.52827048 -11.187 1.805 2455352.5083798 0.588 1.653 2455353.51889148 -2.770 1.674 2455354.49992594 -5.365 1.310 2455372.47121904 -8.777 1.630 2455373.48491063 0.234 1.895 2455374.47029441 -0.105 2.813 2455375.47541052 -7.735 1.522 2455397.48089669 4.500 2.254 2455401.46842892 -2.074 6.009 2455567.83672334 2.496 1.438 2455575.85913514 3.506 1.427 2455579.84450381 5.817 1.316 2455612.70478533 7.008 1.050 2455616.66924282 8.026 1.221 2455619.70421802 12.495 1.546 2455628.73037516 4.520 1.342 2455645.65952823 8.411 2.140 2455660.64956882 5.668 1.814 2455928.83278575 12.421 2.192 2455945.85792291 15.454 1.289 2455962.80628096 10.260 1.338 2455987.75952786 20.513 1.207 2456287.84086125 12.946 2.708 2456304.86724688 25.206 2.228 2456334.87274427 21.515 1.913 2456356.65373143 20.822 3.308 2456370.65478335 23.132 1.544

3.2.2 Direct Imaging and Spectroscopic Observations

MagAO Observation

Our first imaging observations of GJ 3634 was performed on February 15, 2017 with the Magellan Adaptive Optics (MagAO) (Close et al., 2008) with the Clio2 near-infrared detector at the Clay Telescope at Las Campanas Observatory in Chile. They were taken in the Ks band in the narrow configuration with a pixel scale of „ 16 mas. We used an ABBA nod pattern, which we processed by subtracting the A images by the B images, but cutting out the frames with a sub-par AO correction. This subtraction accounts for the sky background and dark current. We then shifted the frames to the center of the images. A flat field correction was not applied, as there is an uncorrectable out-of-focus ”glow” on the detector, for which

61 a 10% photometric error was added.

We calibrated the plate scale and the true north for these observations using the Trapezium cluster in the Orion Nebula. We measured the position angle and separation between stars B1 and E1 in the cluster by finding the centroids of 2D Gaussian fits of their point spread functions (PSFs) with the fiducial values measured in Close et al. (2013). We obtained 15.832 ˘ 0.043 mas/pix for the plate scale and 1.77 ˘ 0.30 degrees for the true north.

From these observations, we detected two companions (Bb and Bc) at 1.8 arcsec from the host star and with 0.1 arcsec separation between the detected pair.

3.2.3 SPHERE Observation

GJ 3634 was also observed with the SPHERE extreme adaptive optics instrument (Beuzit et al., 2008) on UT3 of the Very Large Telescope (VLT) on Cerro Paranal in Chile. It was observed on the night of May 20, 2017 as a part of ESO observing program 099.C-0878(A) (PI: B. M. Pantoja). The observations were taken in the IRDIFS mode, which combines the capability of the Infrared Dual-band Imager and Spectrograph (IRDIS, Dohlen et al., 2008) in the simultaneous H23 bands (1583 nm, 1667 nm, Vigan et al., 2010) and the Integral Field Spectrograph (IFS, Claudi et al., 2008a) in YJ mode which covers 950 nm - 1350 nm with a resolving power of „ 50. We used the apodized Lyot coronagraph (ALC) to lessen the star’s light in the image. To calibrate the star’s position, we took a center calibration which makes a waffle pattern of four satellite spots on a coronagraphic observation, where their intersection measures the star’s position. To calibrate the photometry, we took a flux calibration observation, whereby the PSF is offset so that it is not blocked by the coronagraph, and a neutral density filter is used to prevent saturation of the detector.

We reduced the data from SPHERE using the SPHERE Pipeline Recipes (v. 0.18.0 (CHECK CALAN), Pavlov et al., 2008). We processed the data in the typical fashion with imaging data including correction using the background and of bad pixels, division of the flatfield, flux calibration, and star center calibration. These reduced images are those we will discuss below.

3.2.4 SINFONI Observation

We made the observations of GJ 3634 with SINFONI (Eisenhauer et al., 2003; Bonnet et al., 2004) on UT4 at the VLT on Cerro Paranal in Chile. The observations were performed on July 15, 2018 as a part of Director’s Discretionary Time on Program 2101.C-5021(A). To observe our targets spectroscopically with SINFONI’s IFS we set up the observations in three steps. First, we used GJ 3634 as the AO natural guide star with 0.83 sec acquisition. Once acquired, we shifted the detector 0.06 ” in the X direction and 1.8 ” in the Y direction to observe the Ba and Bb components together in the same image, using the 0.8” x 0.8” ﬁeld of view mode with 100 sec exposures. We did small dithers to stack and subtract a nearby sky image. We then returned to the primary A component to observe in the same ﬁeld of

62 view for ten 5 sec exposures. Lastly, we shifted the detector by 0.9” and changed the plate scale to the 3”x3” ﬁeld of view mode to observe the whole system, with the A components and blended B components in the same image with ﬁve 1.5 sec exposures.

To reduce the data, we used the standard SINFONI Pipeline Recipes (v. 3.1.1) with EsoRex (Freudling et al., 2013) to run through the pipeline steps in a systematic manner. This included steps such as sky subtraction, ﬂatﬁeld division, nonlinearity correction, distortion correction, and wavelength calibration.

The standard star observation, observed at a similar airmass, is used to eliminate the telluric lines from the Earth’s atmosphere in the science data. Using the standard star observation, we did a Levenberg-Marquardt algorithm fit with a 2D Moffat model, using the fitting algorithm and models implemeted in Astropy v. 2.0.9 (Astropy Collaboration et al., 2018), to the PSF. Using the Photutils package (v. 0.4.1, Bradley et al., 2016), we extract the photometry of the star’s PSF for each channel using aperture photometry. We define the apertures using the fit centroids for the center and 1.5x the full width at half maximum (FWHM) of the fit models.

The PSFs of the two companions are partially blended, with the effect greater in redder wavelengths due to their resolution limit increasing proportional to λ/D. To extract the spectra of the two companions, we first used a Levenberg-Marquardt algorthm to fit the two PSFs with two Moffat functions, tying their core widths and power indexes (called γ and α respectively) together so as to give them the same shape, with the algorithm from Astropy Collaboration et al. (2018). We then use the residuals between the fit models and image to improve models for the companions as Dumas et al. (2001) did for spectra of the Pluto-Charon system observed with HST, using the following relation:

1 FBa “ FBa ` pFBa{FT qres

1 FBb “ FBb ` pFBb{FT qres

where, for each pixel, FBa is the flux value for the fit Ba component, FBb is the value for the fit Bb component, and FT for the combined fit, with res being the residuals, defined as res “ Forig ´ FT where Forig is the pixel flux in the original SINFONI image.

This process provides us two corrected model images of the Ba and Bb companions separately, which we can then use for aperture photometry, as we did for the standard star. We use the centroids given from the Moffat fits upon each separated image to find the positions of the Bb and Bc components and use the standard star 1.5 FWHM to use the same apertures for all components per wavelength channel.

We deﬁne a Planck function at 18700 K, representing the standard B3 star HIP59830 (Wright et al., 2003). To ﬂux calibrate the spectra of the two companions, we divide the standard star spectra from the companion spectra and multiply by the Planck function to return the removed standard star spectrum shape from the science target spectra.

63 Table 3.3: Observations Table

Instrument Mode Date (DD/MM/YYYY) Universal Time Filter No. of Exposures Exposure Time (sec) Calibrations MagAO Clio2 15/02/2017 24850 Ks 13x4 3 non-saturated calib. SPHERE IRDIFS 20/05/2017 84939 H23 1x13 64 Star Center, Flux Calib. SINFONI 25 mas, 100 mas 15/07/2018 84759 H+K 9x1 100 Standard star

Table 3.4: Photometry and Astrometry of GJ 3634 Component Instrument Mode Filter Separation (mas) Position Angle (deg) ∆mag B (blended) GAIA G 1761.36 175.584 5.687 ˘ 0.030 A-Ba MagAO Clio2 Ks 1841.16 177.633 4.269 ˘ 0.022 Ba-Bb MagAO Clio2 Ks 99.98 ˘ 2.27 9.724 ˘ 1.52 A-Ba SPHERE IRDIS H2 1855.38 178.595 4.595 ˘ 0.041 Ba-Bb SPHERE IRDIS H2 105.07 ˘ 2.63 16.53 ˘ 0.75 Ba-Bb SINFONI 25 mas H 75.09 ˘ 0.58 35.5 ˘ 0.48 4.455 Ba-Bb SINFONI 25 mas K 74.64 ˘ 0.68 36.84 ˘ 0.45 4.304 Ba-Bb SINFONI 25 mas H+K 74.86 ˘ 0.68 36.19 ˘ 0.81 4.455

To ﬁnd the H and K magnitudes of each component, we simply sum the ﬂux of all of the photometry for each individual channel in the H and K bands, and divide them by the similarly summed photometry of the A star and adjust it by the given magnitudes for the A component. To calculate the errors on the photometry of each channel, we took e´ ` preadoutnoise2 ˚ npixq considering Poissonian statistics. The errors for the standard star and each component of GJ3634B were propagated. The gain used for SINFONI to aconvert to electron counts from detector counts is 2.42 from the SINFONI manual.

3.3 Results

3.3.1 Detection of Companions to GJ 3634

With our observation of GJ 3634 from MagAO, we made a detection of two companions at 1.84 and 1.76 arcsec at ∆Ks of 4.27 ˘ 0.02 and 4.71 ˘ 0.04 magnitudes, respectively, as shown in Fig. 3.2 From our follow-up observations with SPHERE, we measured a contrast of ∆H2 of 4.96 ˘ 0.04 and 5.44 ˘ 0.04 magnitudes, respectively. We show the full astrometric and photometric parameters in Table 3.3.1. The two companions were also previously detected from Bryan et al. (2018) (named GJ 3634 cc1 and cc2) and the blended components also have a detection from Gaia in 2015 (Gaia Collaboration et al., 2016). Comparing these the three astrometric data points (from GAIA, MagAO, and SPHERE) to the track of GJ 3634’s high proper motion and parallax, we can conﬁrm that they have a common proper motion with the primary star as shown in Fig.3.3, showing that the objects are all gravitationally bound. We show the measured photometry and astrometry in Table 3.3.1, with the images of the A and B components for MagAO and SPHERE and the image of the B component in SINFONI shown in Fig. 3.2.

64 Figure 3.2: All images are oriented with north facing up and east to the left, and all three are shown in log scale. (a) MagAO image of GJ 3634 saturated in the Ks band with the companions seen lower in the image. (b) SPHERE IRDIS image of GJ 3634 in the H2 band with the companions visible in the lower part of the image. (c) SINFONI image of the GJ 3634 B companions at 2.2 µm in the 0.8” ﬁeld of view.

Figure 3.3: Plot of the astrometric points of the companions to GJ 3634 with the track (in black ending on May 2017) that would be followed if they were distant background objects, for which they do not. The Gaia detection is that of the blended companions, whilst those of MagAO and SPHERE in 2017 allowed for separate detections of the two companions.

65 3.3.2 Spectral Template and Model Comparison

Using the public NASA Infrared Telescope Facility (IRTF) Spectral Library (Cushing et al., 2005; Rayner et al., 2009), obtained with SpeX medium-resolution spectrograph, we compared the SEDs of the two companions to GJ 3634 with library template spectra. In Fig. 3.4 and Fig. 3.5, we show the spectra of GJ 3634 Ba and GJ 3634 Bb, respectively, with the best fitting spectral type, along with three other spectral types around the best fit, represented. The wavelengths where the telluric absorption is very high were removed from the spectra. After a visual inspection of Fig. 3.4, where the spectra were normalized at 1700 nm, we see that the Ba component best agrees with that of an M8 ultracool dwarf (LP412-31), especially in the H band (Fig. 3.6), where the shape is well traced. In the bluer section, it still traces the observed spectrum better than the other templates. When the K band is considered alone (Fig. 3.7), all four compared spectra fit quite well, showing a low dependence of the K band shape to spectral type. The Na doublet at 2.21 µm are fit best with the M7 and M8 SEDs. This line has a nonlinear dependence on spectral type with it depleting into the L dwarf regime (Cushing et al., 2005; Rojas-Ayala et al., 2010).

For the Bb component, we see the best agreement when considering the whole H+K SED with the L0.5 template (2MASS J07464256+2000321AB), especially in the K band, and the L1 template (2MASSJ02081833+2542533) in the H band. When each band is taken individually, the L0.5 and L1 templates match best with the observed SED for the H band (Fig. 3.8). For the K band considered separately, there is again not much dependence of the shape on the spectral type, but the Na lines at 2.21 µm agree best with the L0.5 template (Fig. 3.9).

In order to gain confidence in our results, we also compare our spectra to BT-Settl models (Allard et al., 2012) at solar abundance (Asplund et al., 2009). We used a Gaussian likelihood estimation for low-mass models from 400 to 3000 K with a [Fe/H] of 0.0 and log(g) from -4.0 to -5.5. While the maximum likelihood yielded a model of 2700 K for both models with log(g) of -4.0, which is near the spectral type found by template fitting, the visual fit is poor with the BT-Settl H band having a more triangular shape, indicative of youth (Lucas et al., 2001; Allers et al., 2007) We therefore do not use this fit further in our analysis. Even where we put the log(g) as a more realistic number for a field substellar dwarf, the model better fits the H band part of the spectrum (with a flatter peak), but it is significantly under the flux in the K band.

3.3.3 Spectral Index

Similar to the process described in Pantoja et al. (2018), we calculated the H2OC and H2OD spectral indices empirically derived in McLean et al. (2003). The H2OC and H2OD wavelength ranges fall within our spectral windows. The H2OB index uses the ﬂux ratio ă 1.456µm ą { ă 1.570µm ą with the wavelengths centered in 0.004 µm range. As the wavelength range of our spectrum is 1.456 to 2.410 µm, the bluest wavelength falls at the center of the ﬂux window of the numerator. We do not use this index as part of the range is outside the edge of our spectrum.These H2O indices are both applicable in the spectral type range of late

66 Figure 3.4: Plot comparing the SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5.

67 Figure 3.5: Plot comparing the SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2.

68 Figure 3.6: Plot comparing the H band SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5.

69 Figure 3.7: Plot comparing the K band SED of GJ 3634 Ba (red) from SINFONI to SPeX template spectra from spectral types M7 to M9.5.

70 Figure 3.8: Plot comparing the H band SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2.

71 Figure 3.9: Plot comparing the K band SED of GJ 3634 Bb (red) from SINFONI to SPeX template spectra from spectral types M9.5 to L2.

72 M to throughout the L sequence (Cushing et al., 2005). The CO index also shows a decreasing trend from mid-M to late L, but with more scatter in the L sequence, making it more difficult to use as an index for these redder objects. The spectral index for M5-L5 field dwarfs from (Allers et al., 2007) considers the flux ratio ă 1.550 to 1.560µm ą { ă 1.492 to 1.502µm ą in the H2O absorption region. We name this index H2OA07.

To calculate these spectral indices, we took the mean of the ﬂuxes for our SINFONI spectra in the wavelength ranges as deﬁned in McLean et al. (2003). The corresponding spectral types and given uncertainties correspond to the following equations from McLean et al. (2003) and Allers et al. (2007):

SpT = - 39.37 (H2OC) + 38.94 SpT = - 25.06 (H2OD) + 34.51 SpT = (H2OA07 - 0.77) / 0.040 For H2OC, we obtain L2.0 ˘ 1.8 for Ba and L4.6 ˘ 1.8 Bb, and for H2OD, we obtain L0.0 ˘ 0.8 for Ba and L1.2 ˘ 0.8 for Bb.From the H2OA07 spectral index of Allers et al. (2007), we received spectral indeces of M8.27 ˘ 0.65 and L1.42 ˘ 0.76 for the Ba and Bb components respectively. We also use the H2O-K2 index from Rojas-Ayala et al. (2012), based on that of Covey et al. (2010). It is given as follows:

ăF p2.070 to 2.909qą{ăF p2.235 to 2.255qą H2O ´ K2 “ ăF p2.235 to 2.255qą{ăF p2.360 to 2.380qą

M subtype number = 24.699 + -23.788 (H2O-K2)

This spectral index was constructed using spectral types from Kirkpatrick et al. (1991), and provides us with value of M7.82 ˘ 0.62 for the Aa component and M8.75 ˘ 0.62 for the Bb component, with the uncertainty given as the root-mean square of the residuals of the H2O–K2 index.

The equivalent width (EW) of the Na I absorption doublet at 2.206 and 2.209 µm are shown in Rojas-Ayala et al. (2010) to have a correlation with metallicity for M dwarfs, where a higher EW would indicate a higher metallicity, but it also depends on surface gravity and temperature. As our system can be assumed to have the same metallicity between each component, we should be able to consider the dependence on the lines in the two spectra on the Teff and log(g), which is related to age. As our objects are dwarfs and not expected to be very young (we do not see a triangular H band shape, Lucas et al., 2001; Allers et al., 2007), we assume gravity to only subtly be affecting the spectrum and neglect it here. Between both visual inspection and measuring the EW, we can see that component Bb has a significantly weaker Na feature than that in the Ba spectrum. Cushing et al. (2005) shows the Na I line to weaken and deplete with scatter between the M7 and L2 spectral types. The Ca I line at 2.264 µm cannot be significantly seen in either spectrum, which would agree that they are later than a spectral type of M7. From Rojas-Ayala et al. (2012), Fig. 4, we see that from their comparison with the BT-Settl spectra (Allard et al., 2012), the Na I line has a complex relationship with metallicity and Teff . All metallicities show a peak where lower and higher temperatures weaken the line, but each curve with a constant metallicity shows a different shape from the others, which is difficult to explain and could be computational rather than physical. Nevertheless, if we consider our [Fe/H] to be -0.04 dex, as given for GJ 3634 A (Santos et al., 2013), we can assume solar metallicity. We see that the Na I line decreases

73 as the ultracool dwarf temperature decreases from less than 2700 K (or M6.5), as do most metallicities, down to at least 2200 K (or earlier than L1). This gives evidence that the Bb is of a later spectral type than Ba, but we note that our EWs are quite diﬀerent from those given by the models. For the Ca I line, the EWs are low (less than 0.7 for all metallicities) at Teff s less than 2700 K (later than M6, Pecaut & Mamajek, 2013; http://www.pas.rochester.edu/ ~emamajek/EEM_dwarf_UBVIJHK_colors_Teff.txt), agreeing with our lack of a detection of that line.

3.3.4 Isochrone Fitting

After combining the spectral type measurements we obtain from SED fitting and spectral indices, we obtain a spectral type of M9.5 ˘ 0.5 for the Ba component and L1.5 ˘ 0.6 for the Bb component, putting both targets at the M/L transition.This would agree with 0.077 to 0.079 Md and 0.075 to 0.076 Md by comparing to masses from Pecaut & Mamajek (2013). From Table B.1 of (Testi, 2009), we compare the spectral types and Teff s from the AMES- Dusty models (Allard et al., 2001) and plot a fifth degree polynomial to convert our spectral `60 types to temperatures. From this, we calculate Teff s of 2273´54 K for the Ba component and `42 Teff s of 2085´40 K for the Bb component. We use the Isochrone v. 1.2.2 Python package (Morton, 2015), which uses MultiNest (Feroz et al., 2009), to measure the mass and age parameters by interpolating the input AMES-Dusty (Allard et al., 2001) evolutionary isochrones. To measure the physical parameters for the companions, we used the absolute magnitudes that were calculated using the SINFONI relative photometry, after comparing to the primary star flux, as explained above. We gave the code many random points within the photometric and temperature error bars to find the measurements in the mass-age space from the isochrones.

`0.29 `0.0151 From this, we receive for Ba, a log(age [yr]) of 8.36´0.36 and mass of 0.0547´0.0134 and for `1.14 `0.0089 Bb, a log(age [yr]) of 8.94´0.32 Gyr and mass of 0.0705´0.0145 Md for the H band magnitudes. The mass for the Bb companion is somewhat near the estimate from their spectral types, but Ba’s mass highlighting the diﬃculty in ultracool dwarfs’ parameter measurements.

In Fig. 3.10, we show where our companion pair fall in a temperature-absolute H magnitude plot along with AMES-Dusty isochrones. We also show other ultracool M/L dwarfs with Teff s and luminosities from Baron et al. (2015). This sample of binary stars, where at least one component is a mid-M to mid-L dwarf, contains systems where the companions have separations greater than 250 AU, making them wide binaries, unlike our comparatively compact system. They also possess a metallicity similar to solar, like GJ 3634. We can see that these scatter on the on the more aged and higher mass part of the plot. Within the errors, some of the M dwarfs could be substellar if they skew toward a younger age.

For substellar objects, there is mass-age degeneracy, meaning results from isochrone fitting should be carefully considered (Faherty, 2014). The age estimates for the two companions are in statistical agreement, even though age estimates for older field objects are known to be difficult to precisely measure. Assuming that the age distribution for field M dwarfs is relatively flat from tens of Myr to about 10 Gyr (shown for M6 dwarfs in Allen et al., 2005),

74 Figure 3.10: Plot comparing the absolute H band magnitude to the Teff of the two components (blue and green) of GJ 3634 B, along with the BT-Settl isochrones. In black, we show the points on the isochrone plot from Baron et al. (2015).

75 our derived ages would agree with the wide range but typically older ages of ﬁeld objects. Another indication of the old age of these ultracool dwarfs is that H-band SEDs of younger ultracool dwarfs have a triangular shape (e.g. Kirkpatrick et al., 2006; Allers et al., 2007; Bihain et al., 2010; Bonnefoy et al., 2014). This could possibly be explained by low gravity and dusty atmospheres causing less opacity from H2 collision induced absorption (Kirkpatrick et al., 2006). Our spectra have ﬂat peaks in the H band and the Na I doublet at 2.2 µm, unlike young dwarfs where this line is weak (Bihain et al., 2010; Bonnefoy et al., 2014).

In Fig. 3.11, we show a scatter plot between the mass ratio of the companion and primary star, and the physical separation of the companion from its host. We include the wide companion sample from Baron et al. (2015), red and brown dwarf companions to stars which also host a planet (Mugrauer et al., 2006b; Lodieu et al., 2014), a sample of directly imaged planets (Bowler, 2016), and directly imaged substellar objects to young stars (Chauvin, 2018a and references therein). For objects with no given mass, we converted their spectral types to temperatures from Pecaut & Mamajek (2013) and then to mass with AMES-Cond (Allard et al., 2001) for T dwarfs and BT-Settl (Allard et al., 2012) for more massive objects. We see the GJ 3634 objects being very close to a separation of zero, like the directly imaged planets but with a higher mass ratio. As for the wider components, we see a lot of scatter in mass ratio up to about 2000 AU, whereas further than that they tend to be at a smaller ratio. This may be due to smaller statistics at higher separation, selection bias, or it could suggest that distant binaries are preferentially unequal in mass. Compared to the other planet hosting stars with low-mass companions, of which there is a small sample, they all seem to have a relatively low mass ratio with all being below 0.4. More detections would help understand this further.

3.3.5 Orbital Fitting

Using a Python code similar to that in (Pantoja et al., 2018) based on Kepler’s laws of planetary motion, we fit the astrometry of the GJ 3634 Ba-Bb component, as this orbit can be tracked in a much quicker amount of time, owing to its shorter separation than that of the system GJ 3634 A-B. In this case, we only considered astrometry which has the potential to provide a total mass for the system. To do the Markov-chain Monte Carlo (MCMC) method, we made use of the emcee code of Foreman-Mackey et al. (2013), where we set up a run of 10,000 steps with 200 walkers at random separations for a reasonable set of parameters. In Table 3.5, where we give the uniform priors, we show the 16th, 50th, and 84th percentiles of the posterior distributions for the orbital elements a, semi-major axis; e, eccenticity; i, inclination; Ω, longitude of the ascending node; ω, argument of periastron; t0, time of periastron passage; and Mtot, total mass of the B component system. As e and ω show degeneracy, we create two parameters based on the e ω together (Marsh et al., 2014) for the code, before separating them to show the results. We show the posterior distributions in the form of a corner plot in Fig. 3.12 and the orbit fits themselves in Fig.3.13. The total `0.266 mass for the system we receive is 0.226´0.138 Md. These are large errors, where at the 84th percentile, we find a total mass of 0.492, a value that could be supplied by two M4 dwarfs. On the other hand, at the low 16th percentile end, we have a total mass of 0.088, which would put two equal mass objects below the substellar limit. This result does not disagree with

76 Figure 3.11: Plot comparing the mass ratio and separation for low-mass stars and brown dwarfs in binary conﬁguration. We show companions to stars also hosting a planet (red, Mu- grauer et al., 2006b; Lodieu et al., 2014), directly imaged planets (yellow, Bowler, 2016), wide red dwarf and brown dwarf companions (black, Baron et al., 2015), and imaged substellar objects to young stars (purple, Chauvin, 2018a).

77 Table 3.5: GJ 3634 Keplerian Elements Keplerian element Prior Value `1.55 a (AU) Up0.1, 10q 2.13´0.51 1 `0.16 e 0.63´0.19 `26.39 i (°) Up01, 180q 37.49´24.86 `32.37 Ω(°) Up0, 360q 247.38´27.85 1 `65.38 ω (°) 49.13´34.50 `695 1 t0(days) Up2457000, 246000q 2458802´1545 `0.266 Mtot (Md) Up0.01, 0.1q 0.226´0.138 1: e and ω are tied parameters as described in Section 3.5. The prior we set was a uniform distribution where e2 ` ω2 ă 1 2 : t0 has a bimodal distribution, due to the periodic time of periastron passage.

a 0.078 M9.5 object with an L2 0.075 Md companion (0.15 Md total; Pecaut & Mamajek, 2013), but more data is evidently needed to better constrain this system.

3.4 Summary and Conclusions

Using MagAO in the Ks band and follow up from SPHERE in the H band and previous data with Gaia, we were able to confirm that the GJ 3634 has two companions at 1.8 arcsec with a common proper motion, showing that they are bound gravitationally, forming a common system consisting of a low-mass M dwarf star, two ultracool companions, and a super-Neptune planet found from HARPS. To further characterize the low mass nature of the two companions, we performed integral field spectrscopy imaging to retrieve their SEDs in the H+K bands by fitting them to SPeX templates, as well as calculating spectral indeces. Using these methods, we found that they have spectral types of M8 and L1 making them ultracool dwarfs. Using AMES-Dusty models, we also found model-dependent mass estimates of 0.07 and 0.05 Md with ages of 0.2 and 0.9 Gyr. The degeneracy between age and mass make these parameters difficult to measure.

Since GJ 3634 is an exoplanet hosting M dwarf star, for which we show to possess two low mass ultracool dwarf companions at a short 38 AU separation from the host, this system becomes a benchmark for studying planet/brown dwarf formation and evolution. The presence of ultracool dwarfs widely orbiting an inner planetary system could eﬀect evolution of the planet, depending on how close they come to the host star. Furthermore, early in system’s history, the presence of the binary did not seem to quench planet formation, at least very close to the primary star. More discoveries of similar types of systems would be very useful to better understand how planet formation and brown dwarf formation proceed within the same proto-planetary disk. Finally, mapping out the full orbit of the binary could provide indications about the details of their formation, and a concerted eﬀort to better constrain this system should be made.

78 Figure 3.12: Corner plots showing the posterior distributions for the orbital ﬁt of GJ 3634. log(Vx) is the jitter along the x direction.

79 Figure 3.13: Plots demonstrating (left) the MCMC astrometric orbital ﬁts in position, (mid- dle) position angle vs. time, and (right) separation vs time with the red points representing the data.

80 Chapter 4

Achieving High Contrast with rapid Reference Star Diﬀerential Imaging on SPHERE at 1.6 µm

Authors: B. M. Pantoja, J. H. Girard, A.-M. Lagrange, J. S. Jenkins, J. Milli, and V. Christiaens In preparation for submission to Monthly Notices of the Royal Astronomical Society

Direct imaging of exoplanets is a difficult endeavor due to the high contrast between the star and companion planets, along with the small angular separations. Besides hardware such as adaptive optics and coronagraphy, we can use post-processing techniques such as differential imaging to subtract the star’s light, with angular differential imaging (ADI) and reference differential imaging (RDI) among the most popular. Using a ”starhopping” technique where we rapidly changed between the A and B components of 55 Eri with SPHERE, we made an ideal data set for both ADI and RDI with 80 minutes of observations and 41° of parallactic angle change. We tested the two methods along with principal component analysis (PCA)- including counterparts by injecting fake companions and measured their signal-to-noise ratios and contrast curves. We consistently found better results for RDI at short separations. We also tested the photometry and astrometry retrieved after injecting 5000 fake companions inside and outside 0.3 arcsec with PCA+ADI and PCA+RDI, for which we found PCA+RDI providing the best measurements, especially at the inner separations. ADI turned up fewer false positives than RDI did. We show RDI to be a promising method to complement the ADI method, particularly in cases where we prefer short snapshot observations, imaging with space instrumentation, and where we want to image as close to the star as possible.

4.1 Introduction

Direct imaging of exoplanets, while being a simple method in theory, is very diﬃcult in practice due to the noise near the star being limited by quasi-static speckles stemming from

81 aberrations in the telescope system. Speckles can change in shape and position over time, making them diﬃcult to subtract and can cause them to be confused for a faint companions. Adaptive Optics (AO) can help mitigate the issue by correcting for point spread function (PSF) for a turbulence component, some aberrations from the telescope system remain.

While speckles make detecting faint companions difficult, we have a variety of methods to use as tools to subtract them known as differential imaging. Angular Differential Imaging (ADI; Marois et al., 2006) is currently among the most popular for attaining a high contrast. It works by stabilizing the pupil in the telescope to allow the stellar field to rotate as the star moves across the sky. These images are then averaged to make a median frame which would discount a companion. By subtracting this median image from the observations, the frames should be left without the speckles. This method works best when there is much field rotation, i. e. when the star is transiting the meridian. Planets farther from the star are easier to detect as they will move more for the same rotation than those close to the star.

ADI has been the common method of choice as of late for the for the detection of exoplanets by imaging. It has been used in the detection of planets around HR 8799 (Marois et al., 2008), κ And b (Carson et al., 2013), HD 95086 (Rameau et al., 2013), GJ 504 b (Kuzuhara et al., 2013), and 51 Eri b (Macintosh et al., 2015). While it is an effective method in greatly improving the contrast performance of the area around a star, it does have some inherent limitation such as its proneness to self-subtraction, as data sets without much angular rotation or with a companion very close to the star with subtract itself as the reference frames include the companion in a simlar location. Due to this, it is often necessary to spend a large amount of time observing to gain the proper amount of field rotation, even if the integration time necessary to detect the companion is well below this observing time. The self-subtraction is especially problematic for extended features, such as disks, where it can produce artifacts that are difficult to discern from real features.

Reference Differential Imaging (RDI; Lagrange et al., 2009; Lafrenièreet al., 2009; Soum- mer et al., 2011) is another method for subtracting the speckle halo of the star. A reference PSF is built and subtractracted from the science target in one of two ways. It can be observed as a nearby star with a similar magnitude (to diminish a dependence of the PSF created by the detector and telescope system on the brightness) and position (considering that isopla- natism affects the PSF the further the reference target is from the science target) in the sky at a similar point of time (as atmospheric changes and changes in conditions in the telescope system can affect the PSF), all to minimize the effect of differences between the target and science PSF. The other way RDI can be performed is through a large library from a survey or archive of observations performed with the same mode of an instrument, as similar PSFs with a high correlation with the science PSF can be used as the reference, as in Choquet et al. (2016)

RDI while less used recently on ground-based observations, has been useful on space-based observations as well disk observations. This has been demonstrated with the redetection of the companions of HR 8799 with archival HST data (Lafreni`ereet al., 2009; Soummer et al., 2011).

There have been some recent studies that have compared the performances of the ADI and RDI methods as well, especially in mid-IR. Ruane et al. (2017) showed an improvement

82 of RDI over ADI at small separations À 0.2arcsec with the Keck/NIRC2 vortex coronagraph in the L’ band using reference PSFs taken before and after the science observations. Xuan et al. (2018) shows there is a critical parallactic angle where, once surpassed, ADI will gain over RDI. With their L’ and M band data sets with the Keck/NIRC2 vortex coronagraph they find this to get higher with increasing separation with the angle being about 14° at 0.2 arcsec but only 6° at 0.4 arcsec. They also show a dependence of the contrast obtained by both methods with the coherence time, with ADI showing the larger effect. Ruane et al. (2019) demonstrates the improvement of SNR for close detected companions with RDI and how for a rotationally symmetric disk, RDI would be the only of the two methods to even detect it. For near-IR wavelengths, the coherence time is naturally lower which would effect how quickly the PSF thus making high contrast imaging more difficult. Rodigas et al. (2015) also did a study directly comparing contrast between ADI and Binary Differential Imaging with MagAO (in the narrow 3.9 µm filter), which can be considered an ideal case of RDI where two PSFs are observed simultaneously on the same detector. This allows the reference star to not be affected by time varying properties intrinsic when it is taken at a different time from the science observation. They found that binary differential imaging results in increased contrast within 1 arcsec over ADI.

In this article, we show a comparison between the ADI and RDI techniques with SPHERE at 1.6 µm with observations of the 55 Eri visual binary. In S 4.2, we show our observational setup and reduction algorithms applied. In S 4.3, we show the various tests we applied to compare ADI and RDI. In S 4.4, we discuss the broader implications of this test on scientiﬁc observations and results. In S 4.5, we give our conclusions to this work.

4.2 Observation and Reduction

We observed the 55 Eridani A and B visual binary star on February 5, 2017 with VLT UT3 at Cerro Paranal in Chile. We imaged the system with SPHERE (Beuzit et al., 2008) in IRDIFS mode, which images with a simultaneous dual-band ﬁlter with the Infrared Dual-band Imager and Spectrograph (IRDIS, Dohlen et al., 2008) and the Integral Field Spectrograph (IFS, Claudi et al., 2008b). We set the ﬁlters in H23 mode (1583 nm, 1667 nm).For our analysis we use the H2 images. The images were taken in coronagraphic mode with the apodized Lyot coronagraph (Guerri et al., 2011).

The 55 Eri A and B observations were taken using a ”starhopping” method where we repeatedly observed each star repeatedly in an ABAB... pattern. Each star was observed with 4x32 sec. frames (62 in total for A for a first 2 frame sequence and 64 in total for B) before moving to the other companion, separated by 40 sec. of overheads given by the 20 sec. file write time, and 20 sec. to open the AO loop, manually give the telescope offset, move the telescope, close the loop, retrieve and run the next observation block, without reacquisition. As we spent 62 min. on target and 80 min. of total telescope time, we achieved an excellent efficiency of „ 80% over the observation. With length of observation time, we retrieved a 41° total parallactic angle variation. This allows us to properly compare an RDI and ADI reduction in the ideal cases for both methods. We summarize the observation in Table 4.1 and show images of the full frames in Fig. 4.1

83 Figure 4.1: Images of 55 Eri A (left) and B (right) under the SPHERE coronagraph, where the axes are given in milliarcsec. 55 Eri A is used the science target and 55 Eri B is used as the reference target for RDI.

Table 4.1: Observations Table

Date (DD/MM/YYYY) SPHERE Mode Filter Object MJD No. of Exposures Exposure Time (sec) Calibrations 06/02/2017 IRDIFS H23 55 Eri A 57790.02318362 62 32 Star Center, Flux Calib. 06/02/2017 IRDIFS H23 55 Eri A 57790.02741027 64 32 Star Center, Flux Calib.

84 The observations turned out to be an ideal test for our comparison between ADI and RDI as the data was taken in conditions with seeing ď 0.6” for most of the observations, with it reaching ă 0.3” for about a half hour. We removed the end of the observations where the seeing deteriorated. The H-band Strehl ratio stayed at about 90% for the whole observation as measured by the SPARTA realtime computer (Fedrigo et al., 2006).

The observations were taken with a star center calibration by making four satellite spots, used to accurate find the position of the star under the coronagraph by taking their intersection. Also, flux calibrations were taken by tilting the mirror so the coronagraph is not in the image, which allows for an accurate measure of the star’s photometry. The flux calibration was taken with a neutral density filter, so the detector is not quickly saturated by the star.

The SPHERE observation was reduced in a typical fashion using the SPHERE Pipeline Recipes (v. 0.15.0, Pavlov et al., 2008), by subtracting the background, dividing by a flatfield, and deleting bad pixels. We processed this data using angular differential imaging (ADI; Marois et al., 2006) and reference differential imaging (RDI) to test which methods provide the best subtraction of the speckle halo of the host star and therefore give the best contrast performance for finding companions. We also tested the methods with and without principal component analysis (PCA; Soummer et al., 2012). We used the Vortex Image Processing Package (VIP) to implement these methods (Gomez Gonzalez et al., 2017).

For further analysis, we used the ﬁrst 58 images of each in the following tests as it cut frames where conditions worsened.

4.2.1 ADI Reduction

We used the VIP code to reduce the 55 Eri A with ADI, considering that it is the brighter star of the two and would have less photon noise, providing a cleaner star subtraction. We first processed the data using a classical ADI (cADI) reduction, where we simply derotate each image and subtract by the median image. We also processed the data with the annular PCA algorithm from VIP, which performs PCA for an arbitrary number of annuli radially emanating from the center of the PSF. To process our image, we used annuli of the size of 1.5 full width at half maximum (FWHM), measured from a 2D Gaussian fit to the PSF of 55 Eri A outside of the coronagraph in the flux calibration image.

To decide on the number of principal components (NPC ) to use, we inserted a faint fake companion (∆H = 10) at a short separation (0.16 arcsec). After the ADI-PCA process, we measure the signal-to-noise ratio (SNR) of the companion using by measuring the ﬂux in one FWHM radius centered on the companion position divided by the standard deviation of 1 FWHM apertures at other angles but at the same separation from center of the stellar PSF, while considering small-sample statistics as described in (Mawet et al., 2014). After doing this for various NPC , we ﬁnd that 4 principal components (PCs) and a parallactic angle threshold given by 0.1 FWHM gives us the best result for this companion, so we use this in our analysis.

85 4.2.2 RDI Reduction

For our RDI analysis, we use VIP to reduce 55 Eri A with the data for 55 Eri B as the reference star. Like with our ADI reduction, we did a classical RDI (cRDI) reduction where we subtract each frame from the A star by each from the B star that has been scaled to the brightness of A (B/A = 3.46). We then derotate the frames and take the median of the frames to obtain our PSF-subtracted image. We also use the annular RDI PCA algorithm from VIP with 1.5 FWHM sized annuli, as used in for the annular ADI PCA reduction.

We decide on the NPC to use for our RDI-PCA reduction similarly to the method we used above in the ADI-PCA reduction. For this reduction we found use 4 PCs and a minimum frame correlation setting of 0.6 provides the best SNR for the retrieved fake companion (FC) and a good visual ﬁeld subtraction. It should be noted that there appears to be an optimal minimum correlation to use. This would be because if we use too small of a minimum correlation, low-correlated frames (possibly due to a rapid seeing change or other condition change) would be included, whereas if the minimum correlation is set too high, we will have a smaller reference library to use. We use the same number of principal components between RDI and ADI to minimize a bias in the number chosen.

4.2.3 Contrast Curve Generation

Using VIP, we generated contrast curves measuring the sensitivity of the observations to the detection of planets or faint objects in the vicinity of primary star (55 Eri in our case). As explained in Gomez Gonzalez et al. (2017), where the noise is measured by using the standard deviation of fluxes within FWHM apertures along an annulus of a given radius from the star. This standard deviation is corrected by student-t correction, as small separations are more affected by small-sample statistics with fewer apertures, as described in Mawet et al. (2014). FCs are inserted into the data frames and recovered after processing to measure the throughput of the procedure. By dividing the throughput multiplied by the flux of the star from the noise multiplied by a factor corresponding to the significance of detection (five for five-sigma, as we use in our tests), we receive the contrast for a given radius. By repeating this for each radius from the star, VIP can make a curve of detectable contrast from the the star to the edge of the data frame. The inserted FCs and contrast curve is corrected by the transmission curve of the coronagraph, which will diminish about 50 % of the light at 0.1 arcsec (SPHERE Manual; Guerri et al., 2011)

4.3 Results

To demonstrate the capabilities between RDI and ADI reductions on SPHERE, and their implications for real observations with the instrument and the techniques, we introduce some different tests showing this. We give the recovered signal-to-noise ratio (SNR), astrometry, and photometry of injected FCs at different separations and contrasts after running the different reductions, along with contrast performance curves for RDI and ADI. We also

86 investigate the eﬀect the angle of the companion and how sparcer and more realistic data sets would have on our possible performance and ability for detection.

4.3.1 Tests to Compare RDI and ADI

To compare the RDI and ADI methods of PSF subtraction, we repeated the steps shown above in Sections 4.2.1 and 4.2.2, but varying the number, position, and brightness of the tested FCs to show how they can be recovered after processing in a two-dimensional image. We also vary the cadence and number of observations to simulate real observing conditions. For the following each of the following contrast curve tests where we vary the data ”conditions” with FCs, we demonstrate our result using a ﬁgure including four images and two plots. The four images show the resulting data with FC(s) after post-processing with the cRDI (a), PCA+RDI (b), cADI (c), PCA+ADI (d) algorithms along with their SNRs considering small-sample statistics as used in Mawet et al. (2014). We show the plot of the time observed on the A and B components along with the parallactic angle at the time of the observation, for the data used in the given test (e). At the bottom, the contrast curves associated with the images using injected FCss from VIP in ﬁve branches are shown along with the contrast curve for a median of the derotated frames and the scatter points for the FCs at the separation and contrast shown (f).

4.3.2 Full Data Set

We ﬁrst start our tests by applying FCs on our full set of data of 55 Eri A and B. For the RDI test, we use the frames of the A star as the science frames and of the B star as reference frames. For the ADI test, we use only the A star frames. In Figs. 4.2 - 4.4, we see in the contrast curves that RDI is beating the performance of ADI at close separations, up to approximately 0.3 mas. This would agree with the advantages and disadvantages of each method, as ADI, unlike RDI, has a limitation of self-subtraction of FCs at short separations from the star. We explain each of the individual cases in the following subsections.

160 mas Fake Companion

We test the performance of the two methods by inputting an FC at 160 mas from the center of the star at 150°. We show in Fig. 4.2 the comparison of the two classical methods (on the right) and their PCA-including counterparts (on the left) in the upper four images as described above. For the cRDI reduction, the companion appears very clearly with SNR of 7.49 with the central region decreasing in counts inward. For the PCA+RDI reduction, we see the FC clearly, as well, with a higher SNR of 7.76 but with some noisy spots at the same radius. The cADI and PCA+ADI reductions are shown to have a worse performance. While cADI does show the companion with an SNR of 5.57, there are two other fairly round and bright spots at a similar radius as the FC that could cause confusion in making a detection. For PCA+ADI, the signal is below our 5 sigma limit with an SNR of 4.55, but still above a

87 candidate follow-up level of 3. There is more noise at the same radius but it appears relatively well distinguished in the image. In the contrast curves, we see that the FC is detectable above the two RDI curves but under the ADI curves. The RDI reduction provides the best results for an FC at the given separation.

110 mas Fake Companion

Similar to the test using the 160 mas companion, we test the processing using an FC at 110 mas separation from the star center (as a close planet could very well be), as shown in Fig. 4.3. This separation is close to the inner working angle (IWA) of the coronagraph where about 50 % of the light is transmitted through at 0.1 arcsec. In the RDI images, the companion is slightly visible to the lower-left side of the inner stellar halo. From the cRDI reduction, it is considered a non-detection, being below an SNR of 3 and for the PCA+RDI reduction, it is just above with an SNR of 3.42. For the ADI images, the FC may be slightly distinguishable in the PCA+ADI, with the previously known position. In a real world condition, there would be no way to distinguish it from the noise. It is not visible in the cADI image, and neither give an SNR above 2. As with the 160 mas FC, the RDI processes, especially with PCA, give the best chance of signal retrieval.

Multiple Fake Companions

To highlight the dependence of detection capability on the position angle as well as the separation by post-processing algorithms, we show the RDI and ADI comparison with FCs given at four separations and at five different position angles in Fig. 4.4. The 20 FCs are injected at separations of 0.11, 0.21, 0.31, and 0.40 arcsec from the center of the star with the 72° separating the branches. For all algorithms, the outer three FCs are easily retrieved for all angles. For the FCs at 0.11 arcsec in the cRDI result, only three of them (lower right three) give possible low-signal indications of the FCs. In the PCA+RDI frame, all of the innermost 0.11 arcsec FCs provide some visible signal. For the two ADI reductions, the cADI provides three clearer signals (upper three) for the 0.11 arcsec FCs than for the cRDI, but with more distortion in the position and PSF shape. The other two signals provide some slight possible flux at the noise level. The PCA+ADI processed frame is distorted in the inner image as well, with the four rightmost FCs showing the highest signals in the image. Neither of the FCs are as clear as those in the PCA+RDI frame. For this test, the two PCA reduced images show the best FC differentiations from the noise than those of the two classical methods.

4.3.3 Cut Frame Set

To closely simulate real observing conditions, we compare ADI and RDI in the following examples by manipulating the number of data frames (i.e. on source exposure time as each frame has the same detector integration time) and cadence of observations. In all of the following tests, we use use half of the number of frames to input into the RDI algorithms as

88 89

Figure 4.2: Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast. 90

Figure 4.3: Images of 55 Eri A with an injected 110 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast. 91

Figure 4.4: Images of 55 Eri A with an injected 120, 298, 306, and 404 mas fake companions at 5 different angles after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the A star and orange to the B star. (f) Plot of 5σ contrast performance curves for the five methods and the fake companion position and contrast. the ADI ones. This is due to the fact that an observing decision where the the same amount of telescope time is assumed would necessitate less time on the science target for the RDI observations than for the ADI observations, as RDI requires reference target observations separate from the science target, whereas ADI uses the science frames as a self-reference. We show three different tests in the following subsections and explain their effects on companion extraction and contrast performances.

Using First Half of Frames for RDI

We test how the performances of the algorithms are affected when we inject a 160 mas FC using the first half of the sets of science (A) and reference (B) frames for the RDI algorithm compared to using all of the science frames for the ADI process. The results are shown in Fig. 4.5. The FC is detected above 5σ for PCA+RDI and cADI. In the PCA+RDI frame, the FC is seen clearly with a few PSF-like noise spots well below the peak of the FC. In the ADI frames, we see the same result as above for the 160 mas, as the test is identical. The cRDI FC can be clearly seen as well, but with a lower signal appearing. The SNR is between 4 and 5 σ making this FC a comfortable candidate. Viewing the images, we see a mixed result as the best SNR is detected from the PCA+RDI algorithm, whereas the cADI and PCA+ADI algorithms produce signals below the 5σ detection threshold. As expected the two RDI tests show decreased SNRs and therefore worse results when compared to the first test above (S 4.3.2), but they still compare quite favorably to the ADI algorithms, especially when comparing the contrast curves. At the 160 mas separation on the curve, we see that a 10 mag contrast companion would be just barely detectable by the PCA+RDI algorithm and just under 5 σ detectability for cRDI and PCA+ADI. Unlike the image, the cADI algorithm performance is higher than the others, which differs from the suggestion of the image and SNR. For separations under 0.15 arcsec, we see a clear performance advantage for the RDI algorithms over the two ADI algorithms. The largest effect on contrast performance appears to be on the range of 0.15 to 0.3 arcsec for which there is little difference for the ADI and RDI performances. This test may less reflect real observing conditions because for the RDI test, science and reference observations are finished in the first „ 40 min of observations, while the ADI test covers the whole observation time, with gaps where the reference observations are taken. In real conditions, the whole observation block would generally equal each other. On the other hand, RDI in this case preserves the high cadence and low duty cycle time, as would be the case in a real, efficient observation. This shows that even in the case where we are considering only half of the science and reference frames for RDI compared to ADI where we keep the same parallactic angle difference, RDI shows a better result for inner companions.

Skipping Every other Frame for RDI

Typically when a decision to choose whether to observe a target with RDI or ADI intended in the post-processing phase of the data acquisition, we would compare the results where both procedures took equal amounts of time in an equal range of time for the observing block. To simulate that, we skipped every other frame for the RDI data set. To further explain, our

92 93

Figure 4.5: Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (first half of set). (f) Plot of 5σ contrast performance curves for the five methods and the fake companion position and contrast. original observations after the first frames followed the following sequence: AAAA, BBBB, AAAA... For this test RDI uses the sequence AA, BB, AA..., while ADI maintains the whole science frame observation (all As). Compared to the last test, we use the same total exposure time for the A and B targets, while making RDI less efficient than the previous test and possibly decreasing the advantage of a stable PSF but beneficially adding a wider range of possible PSFs for the reference library with larger time difference. In Fig. 4.6, we see the results of these data sets injected with a 160 mas FC through the algorithms. As in the prior test, the cADI and PCA+ADI show identical results as they are the same, for the purpose of comparison. The cRDI and PCA+RDI algorithms both show signals over the 5σ threshold with SNRs of 6.88 and 5.59, respectively. While this indicates an improvement in cRDI over the previous test for the FC extraction, we see a diminishing of performance in PCA+RDI. In the contrast curves, we also see that for the contrasts closest to the coronagraph (0.1 to 0.15 arcsec), the cRDI curve has the biggest change where the contrast limit dimishes but improves slightly at the distance of the FC to about 0.4 arcsec. The PCA+RDI curve shows less of a change but slightly before at the separation of the FC compared to the previous test and under 0.2 arcsec a contrast of about 11 mag can be achieved. The PCA+RDI at the shortest separations even shows a very slight improvement over the PCA+RDI of the full data examples, like in the previous test where only the first half was considered. PCA+RDI shows no improvement over cRDI at 0.3 arcsec, while PCA+ADI shows better performance over cRDI from 0.3 to 0.4. Interestingly, the PCA+RDI shows a greater contrast performance than PCA+ADI to about 0.4 arcsec, which is a better performance at the 0.2 to 0.4 arcsec range (where about 13 mag is achieved at 0.3 arcsec) than even when using the full data set. This shows that the selection of frames for the reference library can have an effect on contrast performance. Between this test and the previous one where we only use the first half of frames in the data set, the contrasts for PCA+RDI shows the biggest difference especially for large separations where an improvement is evident possibly suggesting that a reference taken in a larger variety of conditions can improve the library. An improvement at outer separations is even evident over the full frame data set that could be due to including frames with less correlation in reference library. As with the last test, the cRDI algorithm shows a decline in performance from the full set.

94 95

Figure 4.6: Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (skipping every other frame of set). (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast. Snapshot Simulation

One possible situation where RDI could have highly marked improvement over ADI is in the case of ”snapshot” imaging. Snapshot imaging is where an object is observed with a short total exposure time to maximize the number of targets that can be observed at a given time. This is simulated by for ADI using a 10 min observation of the science A target and for RDI using 5 min on the science A target and 5 min on the reference B target. In this test, we have also injected a 160 mas companion, but both processes are expected to be significantly deteriorated as RDI will have fewer reference frames to use and ADI will have less parallactic anglular difference. We see the result in Fig. 4.8. For the cRDI algorithm, the signal is not significantly detected with an SNR of 2.13, but in the correct location there is a round blob corresponding to the FC. There is much noise in the same separation thus lowering the SNR, and if the signal were with the same separation but at another angle, it would be more noticeably difficult to detect. For PCA+RDI, the companion is visible below the 5σ detection limit but above a 3σ candidate detection limit, allowing further investigation to be continued. Neither ADI algorithms give any significant SNR, and the images do not show any signals either, besides some blobs in the PCA+ADI that are not in the position of the FC. The contrast curves show that below 0.2 arcsec we see the greatest contrast improvement between the best RDI and ADI algorithms with a distance reaching about 2 mag at the separations closest to the coronagraph. The ADI algorithms either barely or do not at all improve the contrast above the median image until 0.2 arcsec, making RDI the only choice for searching for close companions in the case of snapshot imaging.

4.3.4 Position Dependece on SNR

To make use of all of the data in our sample, we again use the full data sets in the following tests. While the previous tests showed how an FC companion signal is affected by four post-processing algorithms and how an image would appear after running them, we are also interested in how the position on the image affects the SNR of an FC as well. While our contrast performance curves integrate the performances on five branches in the image, it is important to see how the same companion would be detected in all parts of the image. We show this in Fig. 4.7 where we have measured the SNRs of one injected 10 mag contrast FC companion everywhere in the image after processing with the two PCA algorithms, where one ”pixel” is equal to the measured FWHM of the A star in the non-coronographic flux image. As expected we can see that highest SNRs for both the ADI and RDI reduced images are at higher separations from the center of the frame where the light from the star under the coronagraph is the highest. We can also see that the RDI algorithm has a higher SNR throughout the image except for the very inner separations of a few resolution elements where neither have significant signals. Almost all of the RDI processed map outside of a few hundred arcsec has SNRs above 50, while for ADI no region has SNRs much above 40.

96 Figure 4.7: Maps of SNRs from annular ADI (left) and annular RDI (right) reductions of fake companions spaced in resolution elements (1 λ/D) with the noise being measured by the standard deviation of apertures at the same radius as the fake companions with a Student-t correction for low sample statistics as in Mawet et al. (2014). Two contours in white are shown for each image with innermost one being the 5σ level and outermost one being 25σ.

97 98

Figure 4.8: Images of 55 Eri A with an injected 160 mas fake companion after processing the observation with the classical RDI (a), annular RDI with PCA (b), classical ADI (c), and annular ADI (d) algorithms. (e) Plot of parallactic angle at time exposure for each observation used in the reduction with blue points corresponding to the ADI with the A star and orange to the RDI with the B star, where the A star is similarly used in the same range as B for RDI (partial set). (f) Plot of 5σ contrast performance curves for the ﬁve methods and the fake companion position and contrast. 4.3.5 Quality of Retrieved FCs

While the previous tests showed the overall contrast limits and SNRs of retrieved companions, we are also interested in our ability to actually detect individual companions and the precision of parameters retrieved. As shown in the contrast curves above, the inner part of the image where the star light has a large effect on the detectability of companions differs from outer part of the image to the correction radius. In this outer region, the contrast performance is flatter, which is why using a log scale better emphasizes where contrast differs most. Taking this into account, we injected 10000 companions companions thoroughout the image separating them into two categories (into which we injected 5000 each): inner where the companions are ă 0.3 arcsec in separation and outer where companions are ą 0.3 arcsec in separation. We do this for the PCA+ADI and PCA+RDI algorithms, which for this subsection (S 4.3.5) we simply refer to as ADI and RDI, respectively.

The inner companions are injected randomly on a uniform distribution in a separation range of 0.10 to 0.31 arcsec with a contrasts in the range 6.3x10´6 to 6.3x10´4 (or 8 to 13 mag but distributed linearly). The outer companions are injected similarly but with a separation range of 0.31 to 1.16 arcsec and, considering the better contrast performance at farther separations, contrasts of 1.0x10´6 to 4.0x10´5 (or 11 to 15 mag distributed linearly). We show the positions that were used in Fig. 4.9. We then attempt to recover the signals using the vip.phot.detection algorithm in VIP. This algorithm searches for blobs in the output two-dimensional array from our differential imaging algorithm. We chose the ”Laplacian of Gaussian” option, which applies a Gaussian filter and then Laplacian filter to the image thus removing out the high and low frequencies of the image. This is applied with increasing σs for the Gaussian and then stacked, where the maxima are considered ”blobs”. The blobs that are below our detection threshold (SNR of 5) are not considered. The rest are considered extracted FCs. We then measure the astrometries directly from the output centroids of the detection algorithm and measure the FCs’ photometries by measuring the flux in a 1 FWHM aperture subtracted by the average of background apertures at the same separation along an annulus as the FC. This flux is corrected by the coronagraphic transmission. To show the sampling distributions for these two parameters, we produce histograms (as in Figs. 4.10 and 4.11 which we describe in the next paragraph), where the bins are defined using the Freedman-Diaconis rule Freedman & Diaconis (1981), which we give as follows:

maxpzq ´ minpzq N , bin “ 1{3 2IQRpzqNz where z represents the parameter being investigated and IQR is the interquartile range of that parameter’s distribution.

We show the histogram plots of the inner-separation distributions of the retrieved FCs in Fig. 4.10. In the inner case, we find that out of the 5000 injected FCs, the 4672 (93%) were detected in the RDI case and 4170 (83%) were detected in the ADI case. This shows that the FCs were clearly easier to detect after RDI processing than ADI. The distributions of the ratio of the recovered FC flux to the given FC flux (where 1 means a perfect extraction) show a skewness to the right for ADI and to the left for RDI. To measure and compare the average and dispersion in these distributions, we use their medians and IQRs, respectively. For RDI

99 Figure 4.9: Map of FCs injected into 55 Eri A at separations of less than 0.3 arcsec (left) and greater than 0.3 arcsec (right). The size of the points are adjusted for the contrast magnitude given for the FC.

Figure 4.10: Histogram plots of the difference fraction between recovered fluxes and given fluxes of the FCs using the vip.phot.detection algorithm with orange being from aADI+PCA and blue from aRDI+PCA (left) and of the difference of the x measurements representing the difference in astrometry for FCs within 0.3 arcsec.

100 Figure 4.11: Histogram plots of the difference fraction between recovered fluxes and given fluxes of the FCs using the vip.phot.detection algorithm with orange being from aADI+PCA and blue from aRDI+PCA (left) and of the difference of the x measurements representing the difference in astrometry for FCs outside 0.3 arcsec. we find a median of 0.886 and IQR of 0.058, while ADI gives a median of 0.289 and IQR of 0.140. RDI results very significantly in photometries closer to their given values than ADI. This agrees with the problem of self-subtraction which is inherent in the ADI method at short separations as large parallactic angle variation is needed for the companion to be outside of the position of itself in the other images, as ADI uses the science frames as a reference. RDI reference frames do not possess the companion and so do not have this issue. A procedure for finding more accurate photometries is using forward modeling, where negative FCs are used at the beginning of the process, as in Pueyo (2016). The photometry and astrometry is found by finding the best signal deletion. In the figure, we show astrometry in the x direction, as y shows somewhat similar results. For ∆x (where 0 means that that centroid is perfectly on the given position) between the retrieved FC and given FC, we find that RDI provides a median of -0.050 arcsec and an IQR of 1.004 arcsec, whereas ADI provides a median of -0.119 arcsec and an IQR of 2.823 arcsec. We find that RDI provides a closer median to zero, along with an IQR of about 1/3 of that of ADI. This shows is a large advantage in studying companions from the RDI processed image over ADI processed.

The distributions for the outer separation companions provides a different result. Out of the 5000 given FCs, 4580 were extracted for RDI, while 4614 were extracted for ADI. Slightly more FCs were extracted for ADI than RDI, and both were comparable to the number of FCs extracted in the inner set. The photometry ratios extracted from this set of FCs also shows RDI providing a better result with a median of 0.919 and IQR of 0.126 than ADI with a median of 0.653 and IQR of 0.126. This result is not as significantly superior for RDI as in the last test, but it still demonstrates that ADI’s self-subtraction plays a role in its loss of flux in post-procsessing. While the both histograms show less skewness than the ones for the inner separation, the ADI histogram is representing a bimodal distribution. This appears to be a result of a dependence on the separation, where 300 to 600 mas FCs are recovered with a flux ratio of 0.5 to 0.6 and ą 800 mas FCs are recovered with a flux ratio stably at about 0.7. When investigating the photometry, we find that for the outer companions, RDI

101 Figure 4.12: Plot of 5σ contrast and separation for recovered FCs by the RDI and ADI methods for FCs within 0.3 arcsec (left) and outside of 0.3 arcsec (right). Blue points represent RDI detections and red points represent ADI detections, with the red points hiding blue points discovered by both methods to indicate where RDI shows detections where ADI does not. RDI shows better results at close separation to the star. slightly provides better results with a median of -0.060 with an IQR of 0.810, while ADI has a median of -0.067 and IQR of 1.01. Unlike the astrometry for the inner cases, the astrometry measurements for the outer cases are nearly the same.

Our most important test is to investigate where (by separation) companions for the two methods are actually detected and with what contrast performance, as this allows us to study which method works best for possible cases and to test the contrast performance curves we were obtaining above. In Fig. 4.12 and 4.13, we show the 5σ detected targets from ADI and RDI and their given contrasts and separations, separated by the inner targets and outer targets. Similar to the contrast curves calculated earlier, we find that RDI reveals more companions at very short separations at higher contrasts than ADI for this data sample. It is showing companions of 9 mag contrast at 100 mas and even 10 mag well at 150 mas, which is about 1 mag better contrast than ADI. Within 0.3 arcsec, we find RDI winning. At higher separations, we see that the situation changes where, for 0.6 to 0.9 arcsec, we find more companions for ADI at high contrast at around 13.5 mag. This result is similar that of Rodigas et al. (2015), where they found the similar-to-RDI binary differential imaging finding fainter companions than ADI at 3.9 µm with MagAO up to 0.6 arcsec, with the techniques being slightly more even at longer separations.

Another useful metric to the quality of the ADI and RDI FC retrievals in the number of false positives detected. For the inner injected companions, above the 5σ level, RDI detected 130 false positive signals (where in 5 cases it was the only detection in the image and in all others it was detected along with the true signal) while ADI detected 23 (where in 1 case it was the only detection in the image and in all others it was detected along with the true

102 Figure 4.13: Plot of 5σ contrast and separation for recovered FCs by the RDI and ADI methods for FCs within 0.3 arcsec (left) and outside of 0.3 arcsec (right). Blue points represent RDI detections and red points represent ADI detections, with the blue points hiding red points discovered by both methods to indicate where ADI shows detections where RDI does not. ADI shows better results at the 0.7 to 0.9 arcsec range. signal). This suggests that companions themselves can cause an eﬀect. Thus for RDI, 3 % of our injections yield a false positive, while for ADI, 0.5 % of the cases yield a false positive for close separation companions. Unlike the other metrics used for comparing RDI and ADI at short separations, we see ADI having having a large advantage in avoiding the detection of fake signals. For the outer companions, we see no advantage for either as both only detect 1 false positive, where both were detected along with the real signal in the image. It is noted that as speckles follow a non-Gaussian noise distribution (especially close to the star) and that we give a set (5 σ) detection threshold, we can expect more false positives than if the noise was Gaussian (Aime & Soummer, 2004; Ruane et al., 2019). This indicates that independent observations can still be important to conﬁrm candidates.

4.4 Discussion

Overall, we see that for FCs injected within 0.3 arcsec, RDI shows the provide the best results by contrast limits, number of recovered companions, measured photometry and measured astrometry over ADI. The one disadvantage we show for RDI is that shows many more false positives than ADI at these separations, therefore it is more important to maintain caution over these inner detections. For outer companions, the results are more equal, which is expected with ADI’s self-subtraction having less of an effect at these separations. As we have shown a technical demonstration of the contrast limits of this SPHERE data set, we will discuss scientific and observations applications showing how this can be used for finding

103 real signals in high-contrast imaging data.

As we showed especially in Sec. 4.3.3, one of the most useful cases to RDI is when observing within a narrow time range, as there is commonly not enough parallactic rotation to give ADI its full potential. In cases where many stars are required to be observed in a short time, RDI may be the best way to significantly improve the contrast while observing efficiently. Star forming regions of 120-150 pc or young moving groups, which are nearby (ă 100 pc) associations of young stars moving together with the same age would make for ideal cases for RDI, as they possess stars in a similar portion of the sky (reducing the effect from using a different part of the atmosphere thus changing the AO correction). For ADI this would be problematic as they will all have the highest parallactic angle change simultaneously making multiple nights of observations possibly necessary depending on the region. As RDI has no dependence on parallactic angle and therefore not requiring large amounts of time, but a large dependence on similarity to reference PSF, observing many stars in a region would be its forte. On non-coronagraphic instruments with wide fields of view, the technique would work especially well, since one could image multiple stars simultaneously greatly reducing PSF changes due to atmospheric changes, similar to the binar differential imaging technique (Rodigas et al., 2015). Xuan et al. (2018) showed that using many stars in a single night showed improved RDI performance over ADI, at short separations or with limited field rotation. Similarly, an advantage for this method would be when using a long term snapshot survey or using archival data, observed in the same mode, to build a reference library PSF. This method would allow for minimal use of time on an expensive instrument while potentially achieving a high contrast for many stars which could be used as references for the other stars to subtract and search for faint signals.

In the case of circumstellar disks with nearly face-on inclinations, ADI will self subtract the disk features. Even in cases where the inclination is higher, it is necessary to interpret features with care as artificial effects can be evident and confused for features with physical origins (Milli et al., 2012). For the study of debris and protoplanetary disks, RDI would be advantageous as the disks themselves would not be affected by self-subtraction as the reference frames are completely separate to the science frames. In fact, it has already been used commonly to observe disks on the Hubble Space Telescope (Choquet et al., 2016).

To show the possible results we can have from our method and data set, we show the RDI and ADI 5σ plots given above, with discovered exoplanets by direct imaging in Fig. 4.14. The planets given are HIP 656426 b (Chauvin et al., 2017), 51 Eri b (Macintosh et al., 2015; Samland et al., 2017), HR 8799 b, c, d, and e (Zurlo et al., 2016), and β Pic b (Bonnefoy et al., 2013). All of these planets have hot-star evolution model masses of under 13 MJup and ages under 50 Myr (Bowler, 2016; Chauvin et al., 2017; Cheetham et al., 2018). 51 Eri b is the only of these planets that could be be diﬃcult to redetect and would need a similar observatonal time to retrieve the signal. The rest would be detectable even with the highly cut data of Sec. 4.3.3. With our method given, we could detect a detect a young 6-12 MJup (Cheetham et al., 2018) planet like HIP 65426 b, even to a short separation of 0.2 arcsec.

104 Figure 4.14: Plot of 5σ contrast performance and separation the annular RDI and ADI methods with points plotted for previously discovered exoplanets by direct imaging.

4.5 Conclusions

In this article we have shown that RDI for short separations of less than 0.3 arcsec in our data set, shows better results by most metrics including number of FCs detected in total, SNR, photometry and astrometry measurements in post-processed image, with an important exception being number of false positives detected. At higher separations, we do not find the same advantage, where, while the photometry and astrometry are better detected, between 0.6 and 0.9 arcsec ADI detects more FCs. This method is promising for SPHERE and warrants more research into spectral differential imaging with integral field spectroscopy (Sparks & Ford, 2002) combined with ADI and RDI, forward modeling to find astrometry and photometry while minimizing subtraction from the algorithm. RDI will also be important for finding new planets on future space-based imagers on the JWST and WFIRST.

105 Chapter 5

Conclusions and Future Work

As we saw in the introduction, there is a gap in the detections of low-mass companions by radial velocities (RVs) and direct imaging. The main purpose of this thesis was to continue work in connecting the bridge between these two methods to be able to detect and characterize low mass and low separation red dwarfs, brown dwarfs, and exoplanets.

To do this, we used previously ran RV surveys from CHEPS (Jenkins et al., 2009) and Tuomi et al. (2019) to make a list of targets to follow up with direct imaging AO instruments such as SPHERE and MagAO. We made headway into this with the two detections of HD 86006 B (Chapter 2), an M3 dwarf orbiting a metal-rich Sun like star and GJ 3634 B (Chapter 3), a pair of ultracool dwarfs orbiting an M dwarf. For these detections, we used multiple imaging epochs to confirm their gravitational boundedness. We then took spectra of the companions with a long-slit spectrograph for HD 86006 B and the integral field spectrograph SINFONI for GJ 3634 B. Armed with these spectra, we could confirm their spectral types through continuum fits to template and model spectra and measuring their spectral indices. Using atmospheric models for low-mass stellar and substellar objects, we derived parameters from isochrone fitting. The age determined (which is known to be a difficult parameter to constrain) by isochrones for HD 86006 B is underestimated from that determined for the primary G star. Comparing HD 86006 B to solar neighborhood M dwarfs, we see a spread in ages as well, even though they would be expected to be of old age with little dependence on the parameter with metallicity. This suggests that the underestimated age is not due to the high metallicity nature of star. For GJ 3634 B, we were able to take a look into a rare system including a low-mass star, pair of ultracool dwarfs on the M/L transition and a short period super-Earth planet. This overall shows the kind of detections that can be made when RVs are used to follow up with direct imaging, as our chance of detection is boosted, knowing that there is a companion to the target.

In Chapter 4, we discuss the best differential imaging techniques we can use with SPHERE to allow us to achieve the highest possible contrast performance at the closest separation for the instrument. We compared the angular differential imaging (ADI) technique, which is the most popular one used at the current time, and the reference star differential imaging (RDI) technique. By ”starhopping” or rapidly switching observations between a science and reference star under good conditions while observing with enough time for significant change

106 in the parallactic angle, we are able to have an ideal data set for both RDI and ADI, thus allowing a test between the methods. Applying various tests, including fake companion injections at different separations and positions on the image, trimming the reference data set to compare RDI more evenly by total observation time with ADI, and measuring the quality of the recovered photometries and astrometries of fake companions. We did find that the contrast performance we obtain for RDI is significantly better for short separations at À 0.3 arcsec which is where planets are more expected to be found, with the advantage diminishing at larger separations. Photometry and astrometry are better recovered as well for RDI, whereas we do find more false positives at short separations with RDI, making further caution important.

Overall, we have a current sample of 40 stars that have linear or long period trends with one inflection from CHEPS (Jenkins et al., 2009), EXPRESS (Jones et al., 2011) and Tuomi et al. (2019) for the SAFARI. The detection of companions to HD 86006 and GJ 3634 have showed that this technique works and is useful for extracting the characteristics of any detectable companions. These stars were first observed with snapshot imaging with non-detections being followed up with longer exposures for ADI post-processing and halo subtraction. Of these stars, we have preliminarily detected 19 companions with the other 21 showing non-detections and being followed up with SPHERE. These results hint at long- period trends often being relatively distant low-mass stellar companions, with further work to be performed. In the near term we will be using the detections found by our method in an article where we will describe the statistics of detections (such as binarity, separation, model-dependent mass ratios). We also plan to make a reference library of the snapshot stars’ PSFs to perform RDI and achieve the highest possible contrast for these images where there is little field rotation. We will be able to test firsthand whether our RDI method is effective at detecting further real low-mass companions in our data that would be impossible to detect otherwise. In the longer term, we plan to continue following up their orbits. With enough observations and time passing to observe the companions’ orbit around its host, we can measure their dynamical mass directly and precisely. Of great importance for low mass stars, it will allow us to constrain the mass-luminosity-metallicity relation, as shown in Fig. 5.1. We see that for the V band, metallicity can have a large effect on relation. The K-band magnitudes, as shown by Mann et al. (2019), show little dependence on the metallicity both for data and model tracks, because most increases in opacity from increased metallicity happen in the optical with the strong K-band CO and H2O lines having little effect on metallicity. With our SPHERE data, we would be able to study this in the H band with a metal-rich sample from CHEPS. We can also use the masses as mass constraints to evolutionary models, where real measurements will allow them to be adjusted and more realistic.

We have a tremendous long term potential from the targets in our sample that show non- detections. RVs show us that these targets must possess a companion. As these targets will have already been observed with a current high-contrast imager (SPHERE), we can then infer that the hidden companion must be either very low mass, very close to the star, or both. This means that we will have an ideal, already ”cleaned” sample with which we can follow up with space instrumentation such WFIRST (Content et al., 2013) or ground-based instruments on the European Extremely Large Telescope (ELT, Gilmozzi & Spyromilio, 2007) or Giant Magellan Telescope (Milton et al., 2003). In Fig. 5.2, Chauvin (2018b) shows that

107 Figure 5.1: Mass-magnitude (V) plot for M dwarfs that have dynamical masses and metallicities measured. The curves correspond to the theoretical models from Allard et al. (2012) with an age of 3 Gyr and metallicities of [Fe/H] = -0.5, 0.0, and 0.3 (The curves with metallicities of 0.0 and 0.3 are close to each other).

108 Figure 5.2: Plot of known exoplanets with sensitivities overlaid from current instruments such as SPHERE and GPI, space instruments such as JWST, and future ELT instruments. Fig. 5 of Chauvin (2018b). some second generation ELT instruments could detect planets with masses down to that of Saturn or of super-Earths with short separations of „ 20 mas. Beside the possibility to just detect low-mass planets around aged stars (where currently we have been mainly detecting them with young stars), with integral ﬁeld spectroscopy we could also obtain spectra, looking for and potentially ﬁnding possible life signatures without having to wait for a rare transit!

109 Bibliography

Aime, C., & Soummer, R. 2004, ApJ, 612, L85

Allard, F. 2014, in IAU Symposium, Vol. 299, Exploring the Formation and Evolution of Planetary Systems, ed. M. Booth, B. C. Matthews, & J. R. Graham, 271–272

Allard, F., Hauschildt, P. H., Alexander, D. R., Tamanai, A., & Schweitzer, A. 2001, ApJ, 556, 357

Allard, F., Homeier, D., & Freytag, B. 2012, Philosophical Transactions of the Royal Society of London Series A, 370, 2765

Allen, P. R., Koerner, D. W., Reid, I. N., & Trilling, D. E. 2005, ApJ, 625, 385

Allers, K. N., Jaﬀe, D. T., Luhman, K. L., et al. 2007, ApJ, 657, 511

Almenara, J. M., Deeg, H. J., Aigrain, S., et al. 2009, A&A, 506, 337

Anglada-Escud´e,G., Amado, P. J., Barnes, J., & et al. 2016, Nature, 536, 437

Anglada-Escud´e,G., & Butler, R. P. 2012, ApJS, 200, 15

Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481

Astropy Collaboration, Price-Whelan, A. M., SipHocz, B. M., et al. 2018, AJ, 156, 123

Astudillo-Defru, N., Delfosse, X., Bonﬁls, X., et al. 2017, A&A, 600, A13

Baraﬀe, I., Chabrier, G., Barman, T. S., Allard, F., & Hauschildt, P. H. 2003, A&A, 402, 701

Baranne, A., Queloz, D., Mayor, M., et al. 1996, A&A, 119, 373

Barclay, T., Pepper, J., & Quintana, E. V. 2018, ApJS, 239, 2

Baron, F., Lafreni`ere,D., Artigau, E.,´ et al. 2015, ApJ, 802, 37

Basri, G. 1998, in Astronomical Society of the Paciﬁc Conference Series, Vol. 134, Brown Dwarfs and Extrasolar Planets, ed. R. Rebolo, E. L. Martin, & M. R. Zapatero Osorio, 394

110 Basri, G., Marcy, G. W., & Graham, J. R. 1996, ApJ, 458, 600

Bayliss, D., Gillen, E., Eigmuller, P., et al. 2017, ArXiv e-prints, arXiv:1710.11099

Bean, J. L., Sneden, C., Hauschildt, P. H., Johns-Krull, C. M., & Benedict, G. F. 2006, ApJ, 652, 1604

Becklin, E. E., & Zuckerman, B. 1988, Nature, 336, 656

Benedict, G. F., Henry, T. J., Franz, O. G., et al. 2016, AJ, 152, 141

Beuzit, J.-L., Feldt, M., Dohlen, K., et al. 2008, in Proc. SPIE, Vol. 7014, Ground-based and Airborne Instrumentation for Astronomy II, 701418

Beuzit, J. L., Vigan, A., Mouillet, D., et al. 2019, arXiv e-prints, arXiv:1902.04080

Bihain, G., Rebolo, R., Zapatero Osorio, M. R., B´ejar,V. J. S., & Caballero, J. A. 2010, A&A, 519, A93

Bluhm, P., Jones, M. I., Vanzi, L., et al. 2016, A&A, 593, A133

Bochanski, J. J., Munn, J. A., Hawley, S. L., et al. 2007, AJ, 134, 2418

Bonﬁls, X., Delfosse, X., Udry, S., et al. 2005, A&A, 442, 635

Bonﬁls, X., Gillon, M., Forveille, T., et al. 2011, A&A, 528, A111

Bonnefoy, M., Chauvin, G., Lagrange, A. M., et al. 2014, A&A, 562, A127

Bonnefoy, M., Boccaletti, A., Lagrange, A. M., et al. 2013, A&A, 555, A107

Bonnet, H., Abuter, R., Baker, A., et al. 2004, The Messenger, 117, 17

Bowler, B. P. 2016, PASP, 128, 102001

Bradley, L., Sipocz, B., Robitaille, T., et al. 2016, Photutils: Photometry tools, Astrophysics Source Code Library, , , ascl:1609.011

Brahm, R., Jord´an,A., & Espinoza, N. 2017, PASP, 129, 034002

Bryan, M. L., Knutson, H. A., Fulton, B., et al. 2018, ArXiv e-prints, arXiv:1806.08799

Buchner, J., Georgakakis, A., Nandra, K., et al. 2014, A&A, 564, A125

Burgasser, A. J., Kirkpatrick, J. D., Brown, M. E., et al. 2002, ApJ, 564, 421. http://ukads.nottingham.ac.uk/cgi-bin/nph-bib_query?bibcode=2002ApJ... 564..421B&db_key=AST

Burningham, B., Pinﬁeld, D. J., Leggett, S. K., & et al. 2009, MNRAS, 395, 1237

Burningham, B., Pinﬁeld, D. J., Lucas, P. W., et al. 2010, MNRAS, 406, 1885

111 Burrows, A., Hubbard, W. B., Lunine, J. I., & Liebert, J. 2001, Reviews of Modern Physics, 73, 719

Burrows, A., Marley, M., Hubbard, W. B., et al. 1997, ApJ, 491, 856. http: //ukads.nottingham.ac.uk/cgi-bin/nph-bib_query?bibcode=1997ApJ...491. .856B&db_key=AST

Cannon, A. J., & Pickering, E. C. 1901, Annals of Harvard College Observatory, 28, 129

Carson, J., Thalmann, C., Janson, M., et al. 2013, ApJ, 763, L32

Casagrande, L., Sch¨onrich, R., Asplund, M., et al. 2011, A&A, 530, A138

Castelli, F., & Kurucz, R. L. 2004, ArXiv Astrophysics e-prints, astro-ph/0405087

Chabrier, G., & Baraﬀe, I. 1997, A&A, 327, 1039

Chabrier, G., Johansen, A., Janson, M., & Raﬁkov, R. 2014, in Protostars and Planets VI, ed. H. Beuther, R. S. Klessen, C. P. Dullemond, & T. Henning, 619

Charbonneau, D., Berta, Z. K., Irwin, J., et al. 2009, Nature, 462, 891

Chauvin, G. 2018a, in Society of Photo-Optical Instrumentation Engineers (SPIE) Confer- ence Series, Vol. 10703, Proc. SPIE, 1070305

Chauvin, G. 2018b, arXiv e-prints, arXiv:1810.02031

Chauvin, G., Desidera, S., Lagrange, A. M., et al. 2017, A&A, 605, L9

Cheetham, A., Bonnefoy, M., Desidera, S., et al. 2018, A&A, 615, A160

Chen, J., & Kipping, D. 2017, ApJ, 834, 17

Choquet, E.,´ Perrin, M. D., Chen, C. H., et al. 2016, The Astrophysical Journal, 817, L2

Claudi, R. U., Turatto, M., Gratton, R. G., et al. 2008a, in Proc. SPIE, Vol. 7014, Ground- based and Airborne Instrumentation for Astronomy II, 70143E

Claudi, R. U., Turatto, M., Gratton, R. G., et al. 2008b, in Proc. SPIE, Vol. 7014, Ground- based and Airborne Instrumentation for Astronomy II, 70143E

Close, L. M., Gasho, V., Kopon, D., et al. 2008, in Proc. SPIE, Vol. 7015, Adaptive Optics Systems, 70150Y

Close, L. M., Males, J. R., Morzinski, K., et al. 2013, ApJ, 774, 94

Connolly, A. J., Szalay, A. S., Bershady, M. A., Kinney, A. L., & Calzetti, D. 1995, AJ, 110, 1071

Content, D. A., Armani, N. V., Baker, C. L., et al. 2013, in Society of Photo-Optical Instru-

112 mentation Engineers (SPIE) Conference Series, Vol. 8860, Proc. SPIE, 88600F

Cort´es-Contreras, M., B´ejar,V. J. S., Caballero, J. A., et al. 2017, A&A, 597, A47

Covey, K. R., Lada, C. J., Román-Zúñiga,C., et al. 2010, ApJ, 722, 971

Crepp, J. R., Gonzales, E. J., Bechter, E. B., et al. 2016, ApJ, 831, 136

Crepp, J. R., Johnson, J. A., Howard, A. W., et al. 2014, ApJ, 781, 29

—. 2012, ApJ, 761, 39

Cumming, A., Butler, R. P., Marcy, G. W., et al. 2008, PASP, 120, 531

Cushing, M. C., Rayner, J. T., & Vacca, W. D. 2005, ApJ, 623, 1115

Cutri, R. M., Skrutskie, M. F., van Dyk, S., & et al. 2003, VizieR Online Data Catalog, 2246

Delfosse, X., Forveille, T., S´egransan,D., et al. 2000, A&A, 364, 217

Dittmann, J. A., Irwin, J. M., Charbonneau, D., et al. 2017, Nature, 544, 333

Dohlen, K., Langlois, M., Saisse, M., et al. 2008, in Proc. SPIE, Vol. 7014, Ground-based and Airborne Instrumentation for Astronomy II, 70143L

Dotter, A. 2016, ApJS, 222, 8

Duchˆene,G., & Kraus, A. 2013, ARA&A, 51, 269

Dumas, C., Terrile, R. J., Brown, R. H., Schneider, G., & Smith, B. A. 2001, AJ, 121, 1163

Dupuy, T. J., Brandt, T. D., Kratter, K. M., & Bowler, B. P. 2019, ApJ, 871, L4

Dupuy, T. J., & Liu, M. C. 2012, ApJS, 201, 19

Dupuy, T. J., Liu, M. C., Leggett, S. K., et al. 2015, ApJ, 805, 56

Eddington, A. S. 1924, MNRAS, 84, 308

Eddy, J. A., & Ise, R. 1979, A new sun : the solar results from SKYLAB

Egret, D., Didelon, P., McLean, B. J., Russell, J. L., & Turon, C. 1992, A&A, 258, 217

Eisenhauer, F., Abuter, R., Bickert, K., et al. 2003, in Proc. SPIE, Vol. 4841, Instrument De- sign and Performance for Optical/Infrared Ground-based Telescopes, ed. M. Iye & A. F. M. Moorwood, 1548–1561

Faherty, J. K. 2014, Mem. Soc. Astron. Italiana, 85, 725

Fedrigo, E., Donaldson, R., Soenke, C., et al. 2006, in Proc. SPIE, Vol. 6272, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, 627210

113 Feroz, F., Hobson, M. P., & Bridges, M. 2009, MNRAS, 398, 1601

Fischer, D. A., & Valenti, J. 2005, ApJ, 622, 1102

Foreman-Mackey, D., Hogg, D. W., Lang, D., & Goodman, J. 2013, PASP, 125, 306

Freedman, D., & Diaconis, P. 1981, On the Histogram as a Density Estimator: L 2 Theory, ,

Freudling, W., Romaniello, M., Bramich, D. M., et al. 2013, A&A, 559, A96

Gaia Collaboration, Prusti, T., de Bruijne, J. H. J., et al. 2016, A&A, 595, A1

Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2018, A&A, 616, A1

Gaidos, E., Mann, A. W., L´epine,S., et al. 2014, MNRAS, 443, 2561

Gauza, B., B´ejar,V. J. S., P´erez-Garrido,A., et al. 2019, MNRAS, 487, 1149

Gillon, M., Triaud, A. H. M. J., Demory, B.-O., & et al. 2017, Nature, 542, 456

Gilmozzi, R., & Spyromilio, J. 2007, The Messenger, 127, 11

Golimowski, D. A., Leggett, S. K., Marley, M. S., et al. 2004, AJ, 127, 3516

Gomez Gonzalez, C. A., Wertz, O., Christiaens, V., Absil, O., & Mawet, D. 2016, VIP: Vortex Image Processing pipeline for high-contrast direct imaging of exoplanets, Astrophysics Source Code Library, , , ascl:1603.003

Gomez Gonzalez, C. A., Wertz, O., Absil, O., et al. 2017, AJ, 154, 7

Grandjean, A., Lagrange, A. M., Beust, H., et al. 2019, A&A, 627, L9

Grether, D., & Lineweaver, C. H. 2006, ApJ, 640, 1051

Guenther, E. W., Paulson, D. B., Cochran, W. D., et al. 2005, A&A, 442, 1031

Guerri, G., Daban, J.-B., Robbe-Dubois, S., et al. 2011, Experimental Astronomy, 30, 59

Guillot, T., Lin, D. N. C., Morel, P., Havel, M., & Parmentier, V. 2014, in EAS Publications Series, Vol. 65, EAS Publications Series, 327–336

Halm, J. 1911, MNRAS, 71, 610

Hayashi, C., & Nakano, T. 1963, Progress of Theoretical Physics, 30, 460

Henry, T. J., & McCarthy, Jr., D. W. 1993, AJ, 106, 773

Herschel, W. 1803, Philosophical Transactions of the Royal Society of London Series I, 93, 339

Hinkley, S., Oppenheimer, B. R., Soummer, R., et al. 2007, ApJ, 654, 633

114 Houdebine, E. R., Mullan, D. J., Paletou, F., & Gebran, M. 2016, ApJ, 822, 97

Houk, N. 1978, Michigan catalogue of two-dimensional spectral types for the HD stars

Jenkins, J. S., Jones, H. R. A., Tuomi, M., & et al. 2017, MNRAS, 466, 443

Jenkins, J. S., Jones, H. R. A., Go´zdziewski,K., et al. 2009, MNRAS, 398, 911

Jenkins, J. S., Murgas, F., Rojo, P., et al. 2011, A&A, 531, A8

Jenkins, J. S., Jones, H. R. A., Tuomi, M., et al. 2013a, ApJ, 766, 67

Jenkins, J. S., Jones, H. R. A., Rojo, P., et al. 2013b, in European Physical Journal Web of Conferences, Vol. 47, European Physical Journal Web of Conferences, 5001

Jenkins, J. S., D´ıaz,M., Jones, H. R. A., et al. 2015, MNRAS, 453, 1439

Johnson, J. A., & Apps, K. 2009, ApJ, 699, 933

Jones, M. I., Brahm, R., Wittenmyer, R. A., et al. 2017, A&A, 602, A58

Jones, M. I., Jenkins, J. S., Rojo, P., & Melo, C. H. F. 2011, A&A, 536, A71

Jones, M. I., Jenkins, J. S., Rojo, P., Melo, C. H. F., & Bluhm, P. 2013, A&A, 556, A78

—. 2015a, A&A, 573, A3

Jones, M. I., Jenkins, J. S., Rojo, P., Olivares, F., & Melo, C. H. F. 2015b, A&A, 580, A14

Jones, M. I., Jenkins, J. S., Brahm, R., et al. 2016, A&A, 590, A38

Kaltenegger, L., & Traub, W. A. 2009, ApJ, 698, 519

Kirkpatrick, J. D., Barman, T. S., Burgasser, A. J., et al. 2006, ApJ, 639, 1120

Kirkpatrick, J. D., Henry, T. J., & McCarthy, Donald W., J. 1991, ApJS, 77, 417

Kroupa, P. 2002, Science, 295, 82

Kuiper, G. P. 1942, ApJ, 95, 201

Kumar, S. S. 1962, AJ, 67, 579

Kuznetsov, M. K., del Burgo, C., Pavlenko, Y. V., & Frith, J. 2019, ApJ, 878, 134

Kuzuhara, M., Tamura, M., Kudo, T., et al. 2013, ApJ, 774, 11

Lafreni`ere,D., Marois, C., Doyon, R., & Barman, T. 2009, ApJ, 694, L148

Lafreni`ere,D., Marois, C., Doyon, R., Nadeau, D., & Artigau, E.´ 2007, ApJ, 660, 770

115 Lagrange, A. M., Gratadour, D., Chauvin, G., et al. 2009, A&A, 493, L21

Lagrange, A.-M., Bonnefoy, M., Chauvin, G., et al. 2010, Science, 329, 57

Laughlin, G., Bodenheimer, P., & Adams, F. C. 1997, ApJ, 482, 420

Leggett, S. K., Cushing, M. C., Saumon, D., et al. 2009, ApJ, 695, 1517

Lo Curto, G., Mayor, M., Benz, W., et al. 2010, A&A, 512, A48

Lodieu, N., P´erez-Garrido,A., B´ejar,V. J. S., et al. 2014, A&A, 569, A120

Lucas, P. W., Roche, P. F., Allard, F., & Hauschildt, P. H. 2001, MNRAS, 326, 695

Lyot, B. 1939, MNRAS, 99, 580

Ma, B., & Ge, J. 2014, MNRAS, 439, 2781

Mace, G. N., Kirkpatrick, J. D., Cushing, M. C., et al. 2013, ApJ, 777, 36

Macintosh, B., Graham, J. R., Barman, T., et al. 2015, Science, 350, 64

Maire, A.-L., Langlois, M., Dohlen, K., et al. 2016, in Proc. SPIE, Vol. 9908, Ground-based and Airborne Instrumentation for Astronomy VI, 990834

Maldonado, J., Aﬀer, L., Micela, G., et al. 2015, A&A, 577, A132

Mamajek, E. E., & Hillenbrand, L. A. 2008, apj, 687, 1264

Mann, A. W., Dupuy, T., Kraus, A. L., et al. 2019, The Astrophysical Journal, 871, 63

Marley, M. S., Fortney, J. J., Hubickyj, O., Bodenheimer, P., & Lissauer, J. J. 2007, ApJ, 655, 541

Marois, C., Lafreni`ere,D., Doyon, R., Macintosh, B., & Nadeau, D. 2006, ApJ, 641, 556

Marois, C., Macintosh, B., Barman, T., et al. 2008, Science, 322, 1348

Marois, C., Zuckerman, B., Konopacky, Q. M., Macintosh, B., & Barman, T. 2010, Nature, 468, 1080

Marsh, T. R., Parsons, S. G., Bours, M. C. P., et al. 2014, MNRAS, 437, 475

Mawet, D., Milli, J., Wahhaj, Z., et al. 2014, ApJ, 792, 97

Mayor, M., & Queloz, D. 1995, Nature, 378, 355

Mayor, M., Pepe, F., Queloz, D., et al. 2003, The Messenger, 114, 20

McLean, I. S., McGovern, M. R., Burgasser, A. J., et al. 2003, ApJ, 596, 561

116 Mesa, D., Gratton, R., Zurlo, A., et al. 2015, A&A, 576, A121

Meschiari, S., Wolf, A. S., Rivera, E., et al. 2009, PASP, 121, 1016

Metchev, S. A., Heinze, A., Apai, D., et al. 2015, ApJ, 799, 154

Milli, J., Mouillet, D., Lagrange, A. M., et al. 2012, A&A, 545, A111

Milton, N. M., Lloyd-Hart, M., Cheng, A., Georges, James A., I., & Angel, J. R. P. 2003, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 5169, Proc. SPIE, ed. R. K. Tyson & M. Lloyd-Hart, 238–249

Morgan, W. W. 1938, ApJ, 87, 589

Morton, T. D. 2015, isochrones: Stellar model grid package, Astrophysics Source Code Li- brary, , , ascl:1503.010

Morzinski, K. M., Males, J. R., Skemer, A. J., et al. 2015, ApJ, 815, 108

Mugrauer, M., Neuh¨auser, R., Mazeh, T., et al. 2006a, Astronomische Nachrichten, 327, 321

Mugrauer, M., Seifahrt, A., Neuh¨auser,R., & Mazeh, T. 2006b, MNRAS, 373, L31

Nakajima, T., Oppenheimer, B. R., Kulkarni, S. R., et al. 1995, Nature, 378, 463. http://ukads.nottingham.ac.uk/cgi-bin/nph-bib_query?bibcode=1995Natur. 378..463N&db_key=AST

Newton, E. R., Charbonneau, D., Irwin, J., et al. 2014, AJ, 147, 20

Noethe, L. 2002, Progess in Optics, 43, 1

Pantoja, B. M., Jenkins, J. S., Girard, J. H., et al. 2018, MNRAS, 479, 4958

Pavlov, A., M¨oller-Nilsson,O., Feldt, M., et al. 2008, in Proc. SPIE, Vol. 7019, Advanced Software and Control for Astronomy II, 701939

Peatross, J., & Ware, M. 2015, Physics of Light and Optics (Arva: Ontario Gray, 1988). optics.byu.edu

Pecaut, M. J., & Mamajek, E. E. 2013, ApJS, 208, 9

Pepe, F., Mayor, M., Galland, F., et al. 2002, A&A, 388, 632

Pickering, E. C. 1890, Annals of Harvard College Observatory, 27, 1

Pueyo, L. 2016, ApJ, 824, 117

Queloz, D., Mayor, M., Weber, L., et al. 2000, A&A, 354, 99

Racine, R., Walker, G. A. H., Nadeau, D., Doyon, R., & Marois, C. 1999, PASP, 111, 587

117 Raghavan, D., McAlister, H. A., Henry, T. J., et al. 2010, ApJS, 190, 1

Rameau, J., Chauvin, G., Lagrange, A. M., et al. 2013, ApJ, 772, L15

Rayner, J. T., Cushing, M. C., & Vacca, W. D. 2009, ApJS, 185, 289

Rebolo, R., Zapatero Osorio, M. R., & Mart´ın,E. L. 1995, Nature, 377, 129

Reid, I. N., & Hawley, S. L. 2005, New light on dark stars : red dwarfs, low-mass stars, brown dwarfs, doi:10.1007/3-540-27610-6

Reid, I. N., Kirkpatrick, J. D., Liebert, J., et al. 1999, ApJ, 521, 613. http://adsabs. harvard.edu/cgi-bin/nph-bib_query?bibcode=1999ApJ...521..613R&db_key=AST

Ribas, I., Tuomi, M., Reiners, A., et al. 2018, Nature, 563, 365

Rigaut, F. 2015, PASP, 127, 1197

Rodigas, T. J., Weinberger, A., Mamajek, E. E., et al. 2015, ApJ, 811, 157

Rojas-Ayala, B., Covey, K. R., Muirhead, P. S., & Lloyd, J. P. 2010, ApJ, 720, L113

—. 2012, ApJ, 748, 93

Ruane, G., Mawet, D., Kastner, J., et al. 2017, The Astronomical Journal, 154, 73

Ruane, G., Ngo, H., Mawet, D., et al. 2019, AJ, 157, 118

Ryden, B., & Peterson, B. 2010, Foundations of Astrophysics (Addison-Wesley). https: //books.google.cl/books?id=1p7tOwAACAAJ

Ryu, T., Sato, B., Kuzuhara, M., et al. 2016, ApJ, 825, 127

Sahlmann, J., S´egransan, D., Queloz, D., & Udry, S. 2011, in IAU Symposium, Vol. 276, The Astrophysics of Planetary Systems: Formation, Structure, and Dynamical Evolution, ed. A. Sozzetti, M. G. Lattanzi, & A. P. Boss, 117–120

Samland, M., Molli`ere,P., Bonnefoy, M., et al. 2017, A&A, 603, A57

Santos, N. C., Sousa, S. G., Mortier, A., et al. 2013, A&A, 556, A150

Schlaufman, K. C., & Laughlin, G. 2010, A&A, 519, A105

Schmolze, D., Standley, C., Fogarty, K., & H Fischer, A. 2011, Archives of pathology labo- ratory medicine, 135, 255

Schneider, J., Dedieu, C., Le Sidaner, P., Savalle, R., & Zolotukhin, I. 2011, A&A, 532, A79

Sivaramakrishnan, A., Koresko, C. D., Makidon, R. B., Berkefeld, T., & Kuchner, M. J. 2001, ApJ, 552, 397

118 Soto, M. G., & Jenkins, J. S. 2018, ArXiv e-prints, arXiv:1801.09698

Soummer, R., Hagan, J. B., Pueyo, L., et al. 2011, ApJ, 741, 55

Soummer, R., Pueyo, L., & Larkin, J. 2012, ApJ, 755, L28

Sousa, S. G., Santos, N. C., Israelian, G., Mayor, M., & Udry, S. 2011, A&A, 533, A141

Sparks, W. B., & Ford, H. C. 2002, ApJ, 578, 543

Stetson, P. B. 1987, PASP, 99, 191

Terrien, R. C., Mahadevan, S., Bender, C. F., Deshpande, R., & Robertson, P. 2015, ApJ, 802, L10

Testi, L. 2009, A&A, 503, 639

Torres, G., Andersen, J., & Gim´enez,A. 2010, A&A Rev., 18, 67

Troup, N. W., Nidever, D. L., De Lee, N., et al. 2016, AJ, 151, 85

Tuomi, M., Jones, H. R. A., Anglada-Escud´e,G., et al. 2019, arXiv e-prints, arXiv:1906.04644

Udry, S., Mayor, M., Benz, W., et al. 2006, A&A, 447, 361 van Leeuwen, F. 2007, A&A, 474, 653

Vigan, A. 2016, SILSS: SPHERE/IRDIS Long-Slit Spectroscopy pipeline, Astrophysics Source Code Library, , , ascl:1603.001

Vigan, A., Gry, C., Salter, G., et al. 2015, MNRAS, 454, 129

Vigan, A., Langlois, M., Moutou, C., & Dohlen, K. 2008, A&A, 489, 1345

Vigan, A., Moutou, C., Langlois, M., et al. 2010, MNRAS, 407, 71

Whitworth, A. 2018, arXiv e-prints, arXiv:1811.06833

Wolszczan, A., & Frail, D. A. 1992, Nature, 355, 145

Woolf, V. M., & Wallerstein, G. 2005, MNRAS, 356, 963

Wright, C. O., Egan, M. P., Kraemer, K. E., & Price, S. D. 2003, AJ, 125, 359

Xuan, W. J., Mawet, D., Ngo, H., et al. 2018, AJ, 156, 156

Zacharias, N., Finch, C. T., Girard, T. M., et al. 2013, AJ, 145, 44

Zurlo, A., Vigan, A., Galicher, R., & et al. 2016, A&A, 587, A57

119