Peek-a-boo: is Proxima b hiding in the flares? Determining the impact of flares from Proxima Centauri on the detection of Proxima b

(1)

Peek-a-boo: is Proxima b hiding in the flares?

Determining the impact of flares from Proxima Centauri on the detection of Proxima b

Lucas M. Stapper, 11051183 Supervisor: Jayne L. Birkby

Second assessor: Jean-Michel L. B. D´esert

Anton Pannekoek institute of astronomy University of Amsterdam, Faculty of Science

July 4, 2018

Report Bachelor Project Physics and Astronomy, size 15 EC Conducted between 03 04 2018 and 04 07 2018

(2)

Abstract

High resolution spectroscopy (HRS) is a new promising technique for characterizing exoplanets. The light coming from the exoplanet is buried inside the star’s light, which is in the case of Proxima Centauri ∼ 107 _{times brighter than its planet, Proxima b (our nearest neighbouring and potentially}

habitable exoplanet). Therefore, this technique could highly dependent on the state of the star. On Proxima Centauri flares occur which can obscure Proxima b. To be able to identify the influence of flares in characterising an exoplanet with HRS around flaring stars, the quiescent spectrum and a simulated flare spectrum of the Proxima Centauri system were reduced to attempt to reach the photon noise limit. In this research it was attempted to reach this limit by doing four steps: a bad pixel correction, by using the standard deviations, a normalization of the dataset, a correction of the shift due to the rotation of the Earth around the sun, by using cross correlation and an airmass correction by using a relation between airmass and incoming flux. In this research, the quiescent data was reduced to a noise of 2.94 times the photon noise and the simulated flare data was reduced to a noise ranging from 150 to 650 times the photon noise. Taking these noises into account and the fact that bad pixel and shift corrections would be challenging with a flare inside the data, it was concluded that using the HRS technique for identifying exoplanets around flaring stars is difficult. It is advised to monitor the star at the same time to know when a flare occurs, to counteract this problem.

Samenvatting

Een relatief nieuwe en veel belovende techniek voor het karakteriseren van exoplaneten is hoge resolutie spectroscopie (HRS). Met deze techniek wordt gezocht naar het signaal van een exoplaneet in het licht van een ster. Het licht van de planeet wordt meer Doppler verschoven dan het licht van de ster, waardoor deze terug te vinden is. Deze techniek is alleen nogal afhankelijk van wat de ster doet. Zo kan het zijn dat als de ster een zonnevlam produceert, deze het signaal van de planeet overstemt. Om te testen wat de invloed van zonnevlammen is op HRS, is data afkomstig van Proxima Centauri gebruikt, de dichtstbijzijnde ster van ons zonnestelsel op 4.2 lichtjaar afstand. Proxima Centauri is alleen ∼ 107 keer zo fel als zijn planeet, Proxima b. Om het licht van de ster zo ver mogelijk te onderdrukken, zijn daarom verschillende data-analyse stappen toegepast op hoge resolutie spectra van Proxima Centauri. Deze stappen zijn zowel op data zonder een zonnevlam toegepast als data met een gesimuleerde zonnevlam. Na deze stappen bleek dat een zonnevlam het terugvinden van de planeet aanzienlijk moeilijker maakt.

(3)

1 Introduction

In the last decade the number of confirmed exoplanets went from 200 to over 3500 (exoplanet.eu , 2018). These planets range from hot Jupiters orbiting close to their stars, to super Earths orbiting sun like stars. From all these discoveries only a few are potentially habitable. A notable, potentially habitable planet is the one orbiting Proxima Centauri, or simply Proxima, which is the closest star from our sun. This red dwarf with spectral type M5.5 is located 1.295 parsecs away and has a radius of 0.141R and a mass of 0.12M. Proxima has an effective temperature of 3050 K and an apparent

magnitude of 11.43 (Hansen , 2017). Recent studies revealed the potentially habitable planet, Proxima b, around Proxima and showed that it is around 1.27 earth masses. It has a 11.186 day long orbit with a semi-major axis of 0.0485 AU (Anglada-Escud´e et al. , 2016).

(a) (b)

Figure 1: (a) Radial velocity. Both planet and star are orbiting around their center of mass. This movement can be measured from the Doppler shifted light coming from the star. The planets mass and orbital period can then be found. (b) Transit method. The planet transits the star which causes a dip in the luminosity coming from the star. This can be used to find the size of the planet and the atmospheric composition.

Most exoplanets found until now were observed by either using the radial velocity technique or the planet transiting technique.

Proxima b was found with radial velocity, see figure 1a, which makes use of the movement of the star around the center of mass of the system. This movement causes a Doppler shift in the light of the star coming to Earth. From this shift one can find the orbital period and size. By using the binary mass function, a lower limit on the planets mass can be found because this function depends on the inclination in which the system is observed. However, the planet is not directly observed, so the composition of both the planet and atmosphere is unknown.

The planet transit technique, see figure 1b, uses the reduction in luminosity of the light coming from a star when the planet transits. This gives an indication of how large the planet is. The light of the star can also go through the atmosphere of the planet so that it is possible to find the composition of the atmosphere. A disadvantage of this technique is that the planet should transit, which only happens for an estimated two percent of the habitable planets around stars within ten parsec from the Solar System (Birkby , 2018).

Both techniques have their limitations in getting information about the planet, which obstruct our search in characterizing habitable planets. A more promising technique is that of high resolution spectroscopy.

(5)

Figure 2: Top panel shows a non transiting planet orbiting around a star. When the planet is moving towards the observer, it will be blue shifted and when it is moving away from the observer it will be red shifted. The bottom panel shows how this would look in a high resolution spectroscopy spectrum.(Birkby , 2018)

2 High resolution spectroscopy

With high resolution spectroscopy, or HRS for short, the whole planetary system is considered. The light coming from the star and the planet is captured with a high resolution spectrograph and made into a spectrum. Because the planet is moving faster than the star, the Doppler effect causes the signal from the planet to Doppler shift. To find this Doppler shift, some different analysis steps have to be taken. Proxima Centauri is ∼ 107 times brighter than Proxima b, so these steps need to remove the stellar spectrum, before Proxima b could be extracted at optical wavelengths. After removing the non-time varying components, the tellurics and the almost stationary stellar spectrum, one remains with a spectrum with a sine wave like pattern coming from the planet, see the bottom panel of figure 2. With this technique, it is not necessary for the planet to transit the star. But HRS is sensitive to the state of the star. Most A, F, G, K and M type stars can produces flares, which can obscure the planets signal (Balona , 2015). The question then becomes if it is possible to detect a planet with HRS when a star is flaring.

3 Data analysis

To answer this question, archival observations of Proxima were used which were observed during two nights on dates 04−05−2013 and 05−05−2013, eight ours each night, with the High Accuracy Radial velocity Planet Searcher (HARPS hereafter). HARPS is located in the La Silla observatory in the Atacama dessert, Chile. The spectrum is wrapped around into 72 orders, with each order increasing in wavelengths between 3779.6 and 6912.9 Angstroms. Each order is a grid made out of 64 rows and 4096 columns. Every row is its own spectrum taken approximately ten minutes after each other. This quiescent dataset, so called because it does not contain flares, will later be compared to a dataset with a flare. To see the difference between the quiescent and flare dataset the standard deviation of both datasets will be calculated, which gives an indication of how well the planets signal can eventually be extracted. The lower the standard deviation, the better this can be done. The standard deviation

(6)

cannot be lower than the photon noise limit, which is the intrinsic noise one has when measuring light. The ratio between the standard deviation and the photon noise will give an indication of how well the datasets were reduced. This ratio should preferably be as close as possible to one for the easiest planet signal extraction. The goal of this data analysis is not to find the planets signal, because the used telescope is not large enough for this, but to reduce the data as much as possible to the photon noise limit and see what the effect of a flare is on this photon noise.

The analysis steps mostly correspond to the steps explained and used in other articles (e.g. Snellen et al. 2010, Brogi et al. 2012, 2014, 2016, 2018, Birkby et al. 2013, 2017).

3.1 Bad Pixel correction

The first step in the data analysis is the bad pixel correction. A bad pixel can be defined as an anomalous pixel which has not the value one would expect, it may be too high, or too low than the pixels around it (Celestre et al. , 2016). It therefore has no informational value to the spectrum. Furthermore, these bad pixels increase the standard deviation of the spectrum, which needs to be as low as possible. Bad pixels can be caused by for example a defective pixel, cosmic rays or temperature fluctuations. Bad pixels appear either alone or in groups. During this project, some different techniques where used, all using different algorithms to detect bad pixels. Here, the most successful one is described.

3.1.1 Detecting bad pixels

Bad pixels are pixels whose values differ too much from the other pixels. But when can a value be labelled as too different? One of the more common methods is to look at the standard deviation of a group of values. When some value is for example two standard deviations higher than the mean, the value can be considered to be wrong. A spectrum taken in infra-red can have a lot of fluctuations due to Earth’s atmosphere emitting infra-red light. In such a case, the margin can be made higher, to compensate for these fluctuations. The dataset in this research was taken in visible light, so fluctuations due to the atmosphere are less of a problem and thus the margin can be lower.

When going through the spectrum, the standard deviation is calculated of five pixels next to each other with the possible bad pixel in the middle. If the middle pixel is higher than two standard deviations from the mean, this pixel is labelled as bad. The problem with this technique however, is that it is sensitive towards spectral lines. These lines can be one pixel wide and meet most of the time the margin requirements. These are then marked as bad, even though they are not. But more importantly, if there are multiple bad pixels next to each other, the algorithm will detect only the ones on the outer edges.

So to summarize, the algorithm should be able to detect multiple bad pixels in a row, yet should not label pixels of spectral lines as bad. To overcome this problem, it was decided to compare the standard deviation of different rows of the spectrum with the average of the standard deviation of the other pixels in the same column. The values of the pixels inside the column can be different, but the standard deviation is constant. Also in this case, the standard deviation is calculated of five pixels side by side in the same row with the potential bad pixel in the middle. This is done for all the pixels in one column and then averaged. If the standard deviation is at least three times as high as the average, the so called sigma margin is 3, the pixel is a bad one.

(7)

Figure 3: In the top panel, a section of order one of the dataset is shown where the green boxes show the detected bad pixels with a sigma margin of three. This means that the standard deviation of the bad pixel and its four neighbours is three times, or more, higher than the average of the standard deviation of the column. The bottom panel shows the the same section, but with the bad pixels removed. The standard deviation went down from 12.33 to 12.22. The y-axis is time running from bottom to top, the x-axis is wavelength increasing in length from left to right.

When running the program with a sigma margin of 3, the pixels it detects in order one, can be seen in the top panel of figure 3. The amount of bad pixels found in the complete dataset, is shown in figure 4 in percentages for a sigma margin of 2. The bluer orders have more bad pixels than the redder orders because Proxima is an M dwarf. These have more light in redder wavelengths, so there is more signal which obscures the bad pixels.

3.1.2 Removing bad pixels

When a bad pixel is detected, it should be removed and replaced by the right value. To determine this value, splines come into use. A spline function is made out of pieces of polynomials fitted through points. Because a spline is made out of polynomials, a spline can have different orders. In this research the spline used was up to order three. By fitting a spline through the two pixels on either side of the detected bad pixel, the bad pixel can be exchanged by the value of the spline at that point, see figure 5. If there were more than one bad pixel side by side, the spline was fitted through the two pixels on either side of the group. The bad pixels where then replaced by the values of the spline.

(8)

Figure 4: Percentage of bad pixels for every order with a sigma margin of 2. This means that the standard deviation of the bad pixel should be at least two times higher than the average standard deviation. The redder orders have less bad pixels, because Proxima is an M dwarf. These have more light in redder wavelengths, so there is more signal which obscures the bad pixels.

Figure 5: Spline function fitted through the the two pixels on either side of the bad pixel. The value of the spline at the place of the bad pixel, is the new value for the bad pixel. The original value of the bad pixel is shown as the blue dot.

3.2 Blaze removal

A blaze is caused by the throughput efficiency of the spectrograph. When the light comes into the spectrograph, it has to go through a number of lenses and mirrors, before it gets measured. This causes the edges of an order to have less flux than the middle section. An approximate way to remove this is by fitting a second order polynomial through the spectrum and dividing the spectrum by this polynomial. This removes the continuum of the star and planet, but it is a necessary step in removing the telluric spectrum. A spectrum from order 25, row 38 with the second order polynomial fitted through it, is shown in the top panel of figure 6. After the division the spectrum looks like the bottom panel of figure 6.

3.3 Shifting rows

To see whether there is a time dependent mode in the data, the mean of every column is calculated and subtracted from their respective columns. From this, black and white patterns like the ones in figure 7 emerge. So, the spectral lines are shifting over time.

The shift is caused by Earth’s motion around the sun which has an impact on the spectral lines in the spectrum. The Doppler shift causes these lines to shift diagonally across the matrix containing the spectral time series. These time dependent changes should be removed, so that eventually the only time dependent factor in the dataset is that of Proxima b moving around Proxima. To remove the motion of the earth, one has to shift from the heliocentric frame of reference into the one of Prox-ima. The Earth motion can be measured by using cross correlation which makes a cross correlation function. To know which row is going to be used as a baseline, the signal to noise ratio of each row is calculated. This is done by using the following formula:

S/N = R?× t h

(R?× t) + σ_bg2 + σ_read2 + σ_dark2 )

(9)

Figure 6: Top: Spectrum of Order 25, Row 38 in green with a second order polynomial fitted through it, in blue. Bottom: Spectrum after dividing by the second order polynomial.

in which R?× t is the signal from the star, σbg the background noise, σread the read out noise and

σdark the dark current noise. In this case, the source is bright and therefore dominated by photon

noise. The formula above then reduces to:

S/N 'pR?× t (2)

The row with the highest signal to noise ratio is then cross correlated with all the other rows of the order. When cross correlating two rows of an order, the cross correlation function peaks at the moment when the two rows are corresponding most to each other. This shown in figure 8.

In figure 8 is also a Gaussian function fitted around the peak of the cross correlation function, this function is of the form:

y = a exp (x − b)

2

2c2

+ d (3)

Parameter b gives the shift to the left or right. The values of parameter b will not be in integers so interpolation is used to shift the rows onto the new wavelength grid after the shift is subtracted from the original grid. The shift of every row of order 60 is plotted in figure 9. To determine how aligning the stellar lines improves the cleaning of the spectra towards the photon noise, the mean of each column is calculated again and subtracted from their respective column. After the shifting the

(10)

Figure 7: Top panel: Mean column subtraction before removing the shift caused by the movement of the earth. Bottom panel: Mean column subtraction after removing the shift.

Figure 8: Cross Correlation function, in green, of order 60 when cross correlating row 1 with row 58 with a Gaussian function, in red, fitted around the peak to measure the pixel shift.

Figure 9: Shift of every row in order 60 from the row with the highest signal to noise ratio, in this case row 58.

order looks like the bottom panel of figure 7. The time dependent changes are gone, as was supposed to happen.

3.4 Airmass correction

During the observation night, Proxima will follow a path over the sky because of the spinning of the Earth around its axis. When Proxima is at the horizon, there is more air in the line of sight than when it is at its upper culmination. This causes the amount of flux from Proxima to change over the observation time. To remove this time dependency on airmass, a line is fitted through the airmass plotted against the tenth logarithm of the flux of a column. By then raising ten to the power of the fitted line and dividing the column by this, the airmass dependency goes away.

(11)

(a) (b)

Figure 10: The airmass plotted against the tenth logarithm of the flux. (a) A downward correlation; flux is coming from the star. (b) An upward correlation; flux is coming from the telluric lines.

Figure 11: A plot of order 60. The top panel is after the bad pixel analysis. The middle panel shows the order after the shift correction and the bottom panel is airmass corrected. In the residual noise, the planets signal is buried. The standard deviation was first 314.213 and after the analysis steps 0.050. The y-axis is time running from bottom to top, the x-axis is wavelength increasing in length from left to right.

Now that every step is done, figure 11 shows a part of order 60 after respectively the bad pixel correction, shift correction and airmass correction. The standard deviation went down from 314.213 to 0.050, which is 2.94 times the photon noise limit. This is the baseline cleaning efficiency achieved after the data analysis steps.

4 Flare data

Now that the quiescent dataset in the previous section is analysed, it had to be compared with data in which a flare is happening. Unfortunately, there was no data of a flare during a full night of observing. There was however data taken over one month, with one 1 hour spectrum taken every night, of observation on Proxima Centrauri in which a flare does happen, see figure 12. This dataset is made out 72 orders with every row taken on a different day between 02 − 03 − 2016 and 30 − 03 − 2016.

Figure 12: A dataset of Proxima Centauri in which a flare occurs, note the bright spectrum. This dataset was taken between 02 − 03 − 2016 and 30 − 03 − 2016 with each row on a different day. The exposure time was around one hour per spectrum. The y-axis is time running from bottom to top, the x-axis wavelength from left to right.

(12)

Figure 13: Left: The function yflare= sin log e

π₋₁

a x + 1 which represents the flare. Middle: One spectrum

of a flare coming from Proxima Centauri. Right: The bad pixel and shift corrected quiescence data.

With this dataset it is possible to simulate a flare which takes two hours. This home-made flare dataset can then be exposed to the same analysis steps as the quiescent dataset in section 3.

4.1 Making the flare dataset

To make a flare, three components are needed.

The first component is a good function which resembles a flare over time. The function which was settled upon was:

yflare= sin log e π_{− 1} a x + 1 (4)

where a is the amount of time the flare has to take in tens of minutes. For example, when a = 12 the flare takes 2 hours in total. The plotted function is showed in the left graph of figure 13.

The second component is a single flare spectrum from the flare dataset. The values of the flare in the dataset of figure 12 are assumed to be the maxima of the flare. The last component is the bad pixel corrected and shifted quiescence dataset, see the right graph of figure 13. This can then together with some Gaussian noise serve as a base of the flare spectrum.

These three components can then be combined into one flare dataset:

Flare = Flare function × Spectrum flare + Quiescent data + Gaussian noise (5) The flare function from the left graph of figure 13 is made out of eighteen values spanning a time of three hours. The dataset will be eight hours long so the remaining values will be zero. These values coming from the flare function will work as a scaling factor when multiplying with the flare spectrum from the middle graph of figure 13. This flare has not exactly the same wavelengths as the quiescence data, so before adding them together the flare can be shifted onto the quiescence data grid by using interpolation. The last step is adding the values of the bad pixel corrected and shifted quiescence dataset together with some Gaussian noise with mean of zero and a standard deviation of ten. After completing these steps, the flare dataset will look like the top panel in figure 14.

4.2 Analysing the flare dataset

(13)

4.3. In figure 14 order one of the dataset after each step can be seen.

Figure 14: A plot of order 1 of the simulated flare dataset. The top panel shows the original, the middle panel is after the division of the mean of every row and the bottom panel is after the airmass correction. The y-axis is time running from bottom to top, the x-axis is wavelength increasing in length from left to right.

4.3 Remaining steps

The steps not mentioned above, bad pixel and shift correction, where not included due to a few reasons. The algorithm eventually used for the bad pixel analysis was based upon the fact that every row did not change. In other words, that the flux was constant during the observation time. Of course, this is not the case when a flare occurs. The program therefore labelled most of the pixels coming from the flare as a bad pixel, see figure 16.

During a flare, the absorption lines become emission lines. This gives the second problem, which is removing the shift with a flare in the dataset. Due to the absorption lines becoming emission lines, the rows are essentially shifted which causes the algorithm to calculate the wrong shift, see figure 15. These two reasons made the decision to use the bad pixel and shift corrected quiescent data as a basis instead of the original quiescent data, removing the need to do these on the simulated dataset.

5 Results

In figure 17 the standard deviations of every analysis step of both the quiescence and flare data is shown.

According to figure 17 the standard deviation of the quiescence data after the airmass correction, is between 0.03 and 0.13. In the first 25 orders, the standard deviation is about the same for the last few steps.This is because there is only a small amount of flux coming from the star, Proxima is an M-type dwarf and has mostly red light. Thus there is more signal in the higher orders, longer wavelengths, than in the lower orders, shorter wavelengths. So the analyse steps have more impact on the higher orders. This could be because the lower orders may already have reached the photon noise limit due to the dark correction.

Regarding the flare data, the difference between the standard deviations of the row normalized data and the airmass corrected data is less than that of the quiescent data. The ratio between the photon noise and the standard deviation of the last step can be calculated. Figure 18 shows this ratio of every order from both the quiescent and flare data. The quiescent data has a ratio around 2.94, which means that the noise is 2.94 times the noise one would preferably want. The flare data on the other hand has large fluctuations with peaks around 600. This large ratio is due to the flare which persisted through the data analysis.

(14)

Figure 15: Calculated shifts of order 59 with a flare in the data. The large shift of orders 25 − 41 is caused by the flare.

Figure 16: Part of the first order of the flare dataset. The green boxes show the detected bad pixels, all inside the flare. The flare is between row 25 − 41. The y-axis is time running from bottom to top, the x-axis is wavelength increasing in length from left to right.

6 Discussion

6.1 Bad pixel correction

As was mentioned in section 4.3, the algorithm for detecting bad pixels does not do well with a flare in the dataset. The algorithm looks for difference between the standard deviation of the supposed bad pixel and its four neighbours and the average of the standard deviation of the column and its four neighbours. Because a flare changes absorption lines into emission lines, the rows with the flare have different standard deviations than the other rows. The more time the flare takes during an observation the less of a problem this becomes. So for bad pixel correction one would want preferably a flare during the entirety of the observation rather than one during a part of the observation. A possible solution to this problem is instead of taking only a lower bound, i.e. higher than two times the average standard deviation, also taking an upper bound. So if the standard deviation is higher than the upper bound times the average, the pixel is possibly part of a flare. The problem with this is that it is possible for a bad pixel to be higher than this upper bound. Of course a possible follow up research is to see whether this is often the case and if the upper bound is a solution to this problem.

(15)

Figure 17: Left: Standard deviation of the quiescent data, with a zoomed in version on the bottom panel. The Original data and Bad Pixel data are overlapping. Right: Standard deviation of the flare data, with a zoomed in version on the bottom panel.

(a) (b)

Figure 18: Ratio between the standard deviation of an order and the photon noise for: (a) Quiescent data, (b) Flare data.

6.2 Shift correction

Correcting a shift in a dataset with a flare, struggles with the same problem as the bad pixel analysis. The absorption lines became emission lines and this causes the cross correlation to find shifts like the ones showed in figure 15. A possible solution is to cross correlate the flare independently of the rest of the spectrum. However, this is only possible when it is known that a flare happened and when the flare happened during an observation. For the shift correction to work with a flare, the flare has to either occur during the complete observation time or occur during the timespan it takes to make one spectrum. In this way the cross correlation will give the proper shift.

A possible follow-up research is finding a way to identify a flare and how to separate it from the rest of the spectrum during the cross correlations. A possible way is cross correlating a flare model to the data of one column, to find out where the flare resides inside the order. Another possible solution to still get results after measuring with HRS, would be to measure the wave-front coming from the star at the same time. In this way, it is known when a flare happened. The data can then be cut into sections without the flare.

(16)

6.3 Airmass correction

In the bottom panel of figure 14 some artefacts below and above the brighter parts of the flare are visible after the airmass correction. The airmass was removed by fitting a line through the tenth logarithm of the flux plotted against the airmass and then dividing the columns by ten to the power of the values of this line. Because the flare is brighter than the quiescent flux, when dividing the airmass away, the quiescence flux becomes smaller and the flare persists through. This causes the standard deviation to be bigger than in the quiescent data.

6.4 Positive data

One last important fact not mentioned yet, was that the original data had around 50% negative values in the first seven orders decreasing every order to 0% in order 40 and onward. These negative values could have been caused by the dark current subtraction. As mentioned previously, the bluer orders have less signal than the redder orders due to the source being a M dwarf.

It was decided to shift the data up by the most negative value of the dataset. In this way the dataset was made positive. Because every value shifted with the same amount, the standard deviation did not change.

7 Conclusion

Concluding this research, there are a few important findings.

The bad pixel analysis is the first step in processing the data and lowering the standard deviation. The standard deviation went down on average by 0.036. If these pixels are not removed, they will also give problems during the subsequent steps in the data analysis process. With a flare inside the data however, the algorithm to find the bad pixels labelled the pixels of the flare as bad ones. Bad pixel analysis can thus be difficult when a flare is encountered inside the dataset.

The shifting inside the dataset due to the movement of the earth is also difficult to remove with a flare. The flare changes the absorption lines into emission lines, which makes the cross correlation between the rows difficult.

Looking at the ratios between the standard deviation of an order and the photon noise in figure 18, the dataset without the flare has a lower ratio than the one with the flare. This means that the quiescent dataset is cleaned better than the flare dataset and therefore it would be easier to get a planetary signal out of the quiescent dataset than the flare dataset.

A flare happening during an observation will make it harder to both do the bad pixel analysis and shift correction and if that could be done, the extraction of a planets signal from the flare dataset will be harder than for the quiescent dataset. Considering that most of the A, F, G, K, M type stars show some flaring, this can definitely prohibit the detection of planets around these flaring stars when using High Resolution Spectroscopy. A possible solution discussed was to measure at the same time the light coming from the star. In this way, it is known when the flare happens and the data can be cut so that there is no flare inside it. The steps without a flare have been proven time and again to work by a lot of different studies (e.g. Snellen et al. 2010, Brogi et al. 2012, 2014, 2016, 2018, Birkby et al. 2013, 2017). Then High Resolution Spectroscopy has a bright future on its way.

(17)

8 Acknowledgements

I want to thank assistant professor Jayne L. Birkby for assisting me these past three months, giving me great feedback and helping me when I had problems. I also want to thank professor Jean-Michel L. B. D´esert for being my second assessor.

(18)

References

Anglada-Escud´e, G., et al. 2016, A terrestrial planet candidate in a temperate orbit around Proxima Centauri, Nature, Volume 536, Issue 7617, p.437-440

Balona, L. 2015, Flare stars across the H-R diagram, South African Astronomical Observatory, Re-trieved from: https://files.aas.org/astronomy2015/Presentations/IAUS320_Luis_Balona_ Planets.pdf, Consulted on 20-06-2018

Birkby, J.L., et al. 2013, Detection of water absorption in the day side atmosphere of HD 189733 b using ground-based high-resolution spectroscopy at 3.2 um, Monthly Notices of the Royal Astronomical Society: Letters, Volume 436, Issue 1

Birkby, J.L. 2018, Spectroscopic Direct Detection of Exoplanets

Birkby, J.L., de Kok, R.J., Brogi, M., Schwarz, H., & Snellen, G. 2017, DISCOVERY OF WATER AT HIGH SPECTRAL RESOLUTION IN THE ATMOSPHERE OF 51 Peg b, The Astronomical Journal

Brogi, M. 2012, The signature of orbital motion from the dayside of the planet τ Bo¨otis b, Nature, Volume 486, Issue 7404, pp. 502-504

Brogi, M., de Kok, R.J., Birkby, J.L., Schwarz, H., & Snellen, I.A.G. 2014, Carbon monoxide and water vapor in the atmosphere of the non-transiting exoplanet HD 179949 b, Astronomy & Astrophysics, Volume 565, Article 124

Brogi, M., et al. 2016, ROTATION AND WINDS OF EXOPLANET HD 189733 b MEASURED WITH HIGH-DISPERSION TRANSMISSION SPECTROSCOPY, The Astronomical Journal, 817:106, pp. 15

Brogi, M., et al. 2018, Exoplanet atmospheres with GIANO I. Water in the transmission spectrum of HD 189 733 b, Astronomy & Astrophysics manuscript, no. 32189

Celestre, R., et al. 2016, A novel algorithm for bad pixel detection and correction to improve quality and stability of geometric measurements, Journal of Physics: Conference Series, Ser. 772 012002 The Exoplanet Encyclopaedia, 2018, Retrieved from: http://exoplanet.eu/, Consulted on

22-06-2018

Hansen, B.M.S. 2017, Perturbation of Compact Planetary Systems by Distant Giant Planets, Monthly Notices of the Royal Astronomical Society, Volume 467, Issue 2, p.1531-1560

Snellen, I.A.G.,et al. 2010, The orbital motion, absolute mass and high-altitude winds of exoplanet HD209458b, Nature, Volume 465, Issue 7301, pp. 1049-1051

Peek-a-boo: is Proxima b hiding in the flares? Determining the impact of flares from Proxima Centauri on the detection of Proxima b

Peek-a-boo: is Proxima b hiding in the flares?

Contents

1

Introduction

2

High resolution spectroscopy

3

Data analysis

4

Flare data

5

Results

6

Discussion

7

Conclusion

8

Acknowledgements

References