Candidacy II Slides - Speaker Deck

Slide 1

Slide 1 text

Flare Statistics for Young Stars from a Convolutional Neural Network Analysis of TESS Data Adina Feinstein NSF Graduate Research Fellow Ben Montet (UNSW), Megan Ansdell (NASA HQ), Brian Nord (UChicago/KICP/Fermi), Jacob Bean (UChicago), Max Günther (MIT), Michael Gully-Santiago (UT Austin), Josh Schlieder (NASA GSFC) arXiv:2005.07710 ! 1

Slide 2

Slide 2 text

Know thy star, know thy planet. !2

Slide 3

Slide 3 text

Flares can lead to increased photoevaporation of inner disks. (Benz & Güdel, 2010) !3

Slide 4

Slide 4 text

Flares can lead to increased photoevaporation of inner disks. (Benz & Güdel, 2010) !3

Slide 5

Slide 5 text

Flares can lead to increased photoevaporation of inner disks. (Benz & Güdel, 2010) !3

Slide 6

Slide 6 text

Flares have been shown to increase atmospheric erosion, especially when they are still forming & contracting. (Lammer et al. 2007; Owen & Wu 2017) !4

Slide 7

Slide 7 text

Flares have been shown to increase atmospheric erosion, especially when they are still forming & contracting. (Lammer et al. 2007; Owen & Wu 2017) !4

Slide 8

Slide 8 text

Flares have been shown to increase atmospheric erosion, especially when they are still forming & contracting. (Lammer et al. 2007; Owen & Wu 2017) !4

Slide 9

Slide 9 text

Flares have been seen to have long term effects on atmospheric chemical compositions. (Venot et al. 2016) !5 Metals & molecules

Slide 10

Slide 10 text

Flares have been seen to have long term effects on atmospheric chemical compositions. (Venot et al. 2016) !5 Metals & molecules Different metals & molecules

Slide 11

Slide 11 text

What is the relationship between stellar flare energy and spectral type? Age? Spot phase? !6

Slide 12

Slide 12 text

What is the relationship between stellar flare energy and spectral type? Age? Spot phase? !7

Slide 13

Slide 13 text

Similar studies with Kepler demonstrated a relationship between spectral type, rotation period, and flare energy. (Davenport, 2016) 1.66 1.13 0.86 0.73 0.62 0.47 0.28 Mass [MSun ] Maximum Log(Flare Energy [ergs]) 32 33 34 35 36 37 38 39 40 !8

Slide 14

Slide 14 text

Slide 15

Slide 15 text

Similar studies with Kepler demonstrated a relationship between spectral type, rotation period, and flare energy. (Yang & Liu, 2019) (Davenport, 2016) 1.66 1.13 0.86 0.73 0.62 0.47 0.28 Mass [MSun ] Maximum Log(Flare Energy [ergs]) 32 33 34 35 36 37 38 39 40 !8

Slide 16

Slide 16 text

Slide 17

Slide 17 text

What is the relationship between stellar flare energy and spectral type? Age? Spot phase? !9

Slide 18

Slide 18 text

Analysis of the Pleiades (0.125 Gyr) and Praesepe (0.63 Gyr) showed a decrease in flare rate with age. (Ilin et al. 2019) 3000-3249 K 3250-3499 K 3500-3749 K 3750-4000 K Pleiades Praesepe !10

Slide 19

Slide 19 text

What is the relationship between stellar flare energy and spectral type? Age? Spot phase? !11

Slide 20

Slide 20 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. !12

Slide 21

Slide 21 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. (Doyle et al. 2018) !12

Slide 22

Slide 22 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. (Doyle et al. 2018) !12

Slide 23

Slide 23 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. (Doyle et al. 2018) (Roettenbacher et al. 2018) 1-5% increase > 5% increase !12

Slide 24

Slide 24 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. (Doyle et al. 2018) (Roettenbacher et al. 2018) 1-5% increase > 5% increase !12

Slide 25

Slide 25 text

In the same year, two papers presented conflicting results on where flares occur with relation to spots. (Doyle et al. 2018) (Roettenbacher et al. 2018) 1-5% increase > 5% increase Sample size: 32 Sample size: 119 !12

Slide 26

Slide 26 text

What is the relationship between stellar flare energy and spectral type? Age? Spot phase? !13

Slide 27

Slide 27 text

The Transiting Exoplanet Survey Satellite (TESS) is a four+ year, nearly all-sky survey monitoring millions of stars. 27 days 54 days 81 days 108 days 189 days 351 days !14

Slide 28

Slide 28 text

There is no one best way to determine the age of a star or group of stars. !15

Slide 29

Slide 29 text

There is no one best way to determine the age of a star or group of stars. (Curtis et al. 2019) Gyrochronology !15

Slide 30

Slide 30 text

There is no one best way to determine the age of a star or group of stars. (Curtis et al. 2019) Gyrochronology !15 Isochrone Fitting (Murphy et al. 2020)

Slide 31

Slide 31 text

There is no one best way to determine the age of a star or group of stars. (Curtis et al. 2019) Gyrochronology !15 Isochrone Fitting (Murphy et al. 2020) (Schneider et al. 2019) Kinematics

Slide 32

Slide 32 text

We have completed a literature search for young moving group members with ages 1-750 Myr. !16

Slide 33

Slide 33 text

Young stars are great and active. Octans Columba AB Doradus !17

Slide 34

Slide 34 text

Young stars are great and active. Octans Columba AB Doradus !17

Slide 35

Slide 35 text

However, the outlier metrics used will bias studies towards identifying only the strongest flares. (Chang, Byun, & Hartman, 2015) !18

Slide 36

Slide 36 text

However, the outlier metrics used will bias studies towards identifying only the strongest flares. (Chang, Byun, & Hartman, 2015) !18

Slide 37

Slide 37 text

However, the outlier metrics used will bias studies towards identifying only the strongest flares. (Chang, Byun, & Hartman, 2015) :( !18

Slide 38

Slide 38 text

Machine learning can be used when searching for signals with a characteristic shape. (Pearson et al. 2017) !19

Slide 39

Slide 39 text

Machine learning can be used when searching for signals with a characteristic shape. (Pearson et al. 2017) !19

Slide 40

Slide 40 text

We used the TESS flare catalog from Günther et al. (2020) as our training, validation, and test sets. !20

Slide 41

Slide 41 text

Every light curve example receives a label that the neural network learns over several hundred epochs. !21

Slide 42

Slide 42 text

The results of 10 stella models show good training demonstrated through the accuracy and loss metrics. (Feinstein et al. 2020) !22

Slide 43

Slide 43 text

We ensure the neural network is classifying our light curves correctly by looking at the test set. !23

Slide 44

Slide 44 text

The confusion matrix for the stella test set is used to visualize what the false cases look like. (Feinstein et al. 2020) !24

Slide 45

Slide 45 text

The CNN is used as a sliding-box detector, where the probability will increase as the flare nears the center of the box. !25

Slide 46

Slide 46 text

The CNN is used as a sliding-box detector, where the probability will increase as the flare nears the center of the box. !25

Slide 47

Slide 47 text

https://GitHub.com/afeinstein20/stella Looking at some example light curves, the cadences “light up” when believed to be a flare. (Feinstein et al. 2020) ! 26

Slide 48

Slide 48 text

What relationship did we find between stellar flare energy and spectral type? Age? Spot phase? !27

Slide 49

Slide 49 text

We find a clear difference in flare rate at Gaia Bp -Rp = 2 (Teff ~ 4000K). (Feinstein et al. 2020) !28

Slide 50

Slide 50 text

The coolest stars have stronger flares in relation to their luminosity. (Feinstein et al. 2020) !29

Slide 51

Slide 51 text

What relationship did we find between stellar flare energy and spectral type? Age? Spot phase? !30

Slide 52

Slide 52 text

Flares amplitudes and rates are higher the coolest stars across all ages, with noticeable evolution in other temperature bins. (Feinstein et al. 2020) !31

Slide 53

Slide 53 text

Subdividing by age, it’s easier to see the differences in flare rate between cool (< 4000K) stars and hot stars. (Feinstein et al. 2020) !32

Slide 54

Slide 54 text

What relationship did we find between stellar flare energy and spectral type? Age? Spot phase? !33

Slide 55

Slide 55 text

Across 1500 stars, there seems to be no preference for where flares happen. They happen everywhere! (Feinstein et al. 2020) !34

Slide 56

Slide 56 text

Across 1500 stars, there seems to be no preference for where flares happen. They happen everywhere! (Feinstein et al. 2020) !34

Slide 57

Slide 57 text

What do we think these stellar surfaces look like? What we think we’re seeing: !35

Slide 58

Slide 58 text

What do we think these stellar surfaces look like? What we think we’re seeing: !35

Slide 59

Slide 59 text

What do we think these stellar surfaces look like? What we think we’re seeing: What a flare-phase relationship would’ve looked like: !35

Slide 60

Slide 60 text

What do we think these stellar surfaces look like? What we think we’re seeing: What a flare-phase relationship would’ve looked like: The Sun: !35

Slide 61

Slide 61 text

Flares have been shown to alter atmospheric chemistry and could partially affect mass-loss rates during formation. (Feinstein et al. 2020) !36

Slide 62

Slide 62 text

Flares have been shown to alter atmospheric chemistry and could partially affect mass-loss rates during formation. (Feinstein et al. 2020) !36

Slide 63

Slide 63 text

Flares have been shown to alter atmospheric chemistry and could partially affect mass-loss rates during formation. (Feinstein et al. 2020) !36

Slide 64

Slide 64 text

What’s next? !37

Slide 65

Slide 65 text

Young planets are challenging to detect and occur on similar timescales to flares, so let’s “flip” stella! (David et al. 2019) !38

Slide 66

Slide 66 text

Young planets are challenging to detect and occur on similar timescales to flares, so let’s “flip” stella! (David et al. 2019) !38

Slide 67

Slide 67 text

There’s a huge (500 Myr!!) age range of missing planets. stella may be able to find them. (Exoplanet Archive, as of 6/25/2020) !39

Slide 68

Slide 68 text

There’s a huge (500 Myr!!) age range of missing planets. stella may be able to find them. (Exoplanet Archive, as of 6/25/2020) !39

Slide 69

Slide 69 text

My project at the tess.ninja workshop already demonstrates the power of stella when applied to transits. !40

Slide 70

Slide 70 text

We hope to search both the 2-minute and 30-minute TESS data for new young planet candidates. Photo credit: Ethan Kruse !41 (Feinstein et al. 2019)

Slide 71

Slide 71 text

Summary • This is the ﬁrst application of CNNs to find stellar flares. It provides a new statistical analysis of these events. • We find no phase preference for flares, indicating the surfaces of young stars have an overall very high spot coverage. • Cool stars (Teff < 4000 K) have high and consistent ﬂare rates at all ages between 1-800 Myr, while hot stars show decline over time. • Flares have negative consequences for photoevaporative mass loss of exoplanet atmospheres. • Next — let’s find more young planets! !42