1. Radial Velocity Detection of Planets: I. Techniques 1. Keplerian Orbits 2. Spectrographs/Doppler shifts 3. Precise Radial Velocity measurements

2. apap asas V P 2 = 4242 (a s + a p ) 3 G(m s + m p ) msms mpmp Newton‘s form of Kepler‘s Law

3. P 2 = 4242 (a s + a p ) 3 G(m s + m p ) Approximations: a p » a s m s » m p P2 ≈P2 ≈ 4242 ap3ap3 Gm s

4. Circular orbits: V = 2as2as P Conservation of momentum: m s × a s = m p × a p a s = m p a p msms Solve Kepler‘s law for a p : a p = P 2 Gm s 4242 () 1/3 … and insert in expression for a s and then V for circular orbits

5. V = 22 P(4  2 ) 1/3 m p P 2/3 G 1/3 m s 1/3 V = 0.0075 P 1/3 m s 2/3 mpmp = 28.4 P 1/3 m s 2/3 mpmp m p in Jupiter masses m s in solar masses P in years V in m/s 28.4 P 1/3 m s 2/3 m p sin i V obs =

6. PlanetMass (M J )V(m s –1 ) Mercury1.74 × 10 –4 0.008 Venus2.56 × 10 –3 0.086 Earth3.15 × 10 –3 0.089 Mars3.38 × 10 –4 0.008 Jupiter1.012.4 Saturn0.2992.75 Uranus0.0460.297 Neptune0.0540.281 Pluto1.74 × 10 –4 3×10 –5 Radial Velocity Amplitude of Sun due to Planets in the Solar System

7. Radial Velocity Amplitude of Planets at Different a Radial Velocity (m/s) G2 V star

8. Radial Velocity (m/s) A0 V star

9. M2 V star

10. eccentricity Elliptical Orbits = OF1/a O

11.  : angle between Vernal equinox and angle of ascending node direction (orientation of orbit in sky) i: orbital inclination (unknown and cannot be determined P: period of orbit  : orientation of periastron e: eccentricity M or T: Epoch K: velocity amplitude Important for radial velocities Not important for radial velocities

12. Radial velocity shape as a function of eccentricity:

13. Radial velocity shape as a function of , e = 0.7 :

14. Eccentric orbit can sometimes escape detection: With poor sampling this star would be considered constant

15. Eccentricities of bodies in the Solar System

16. Important for orbital solutions: The Mass Function f(m) = (m p sin i) 3 (m p + m s ) 2 = P 2G2G K 3 (1 – e 2 ) 3/2 P = period K = Amplitude m p = mass of planet (companion) m s = mass of star e = eccentricity

17. Observer Because you measure the radial component of the velocity you cannot be sure you are detecting a low mass object viewed almost in the orbital plane, or a high mass object viewed perpendicular to the orbital plane We only measure M Planet x sin i i

18. The orbital inclination We only measure m sin i, a lower limit to the mass. What is the average inclination? i The probability that a given axial orientation is proportional to the fraction of a celestrial sphere the axis can point to while maintaining the same inclination P(i) di = 2  sin i di

19. The orbital inclination P(i) di = 2  sin i di Mean inclination: =  ∫ P(i) sin i di 0  ∫ P(i) di 0 =  /4 = 0.79 Mean inclination is 52 degrees and you measure 80% of the true mass

20. The orbital inclination P(i) di = 2  sin i di But for the mass function sin 3 i is what is important : =  ∫ P(i) sin 3 i di 0  ∫ P(i) di 0 =  ∫ sin 4 i di  0 = 3  /16 = 0.59

21. The orbital inclination P(i) di = 2  sin i di Probability i <  : P(i<  ) =  ∫ P(i) di 0  ∫ 0 2 (1 – cos  ) =  < 10 deg : P= 0.03 (sin i = 0.17)

22. Measurement of Doppler Shifts In the non-relativistic case: – 0 0 = vv c We measure  v by measuring 

23. The Radial Velocity Measurement Error with Time How did we accomplish this?

24. The Answer: 1. Electronic Detectors (CCDs) 2. Large wavelength Coverage Spectrographs 3. Simultanous Wavelength Calibration (minimize instrumental effects)

25. Instrumentation for Doppler Measurements High Resolution Spectrographs with Large Wavelength Coverage

26. collimator Echelle Spectrographs slit camera detector corrector From telescope Cross disperser

27. A spectrograph is just a camera which produces an image of the slit at the detector. The dispersing element produces images as a function of wavelength without disperser with disperser slit

29. 5000 A 4000 A n = –1 5000 A 4000 A n = –2 4000 A 5000 A n = 2 4000 A 5000 A n = 1 Most of light is in n=0

30. These are the instensity distribution of the first 3 orders of a reflection grating

31. 3000 m=3 5000 m=2 40009000 m=1 6000 14000 Schematic: orders separated in the vertical direction for clarity 1200 gr/mm grating 2 1 You want to observe 1 in order m=1, but light 2 at order m=2, where 1 ≠ 2 contaminates your spectra Order blocking filters must be used

32. 4000 m=99 m=100 m=101 5000 9000 14000 Schematic: orders separated in the vertical direction for clarity 79 gr/mm grating 30002000 Need interference blocking filter but why throw away light? In reality:

33. yy ∞ 2 y m-2 m-1 m m+2 m+3   Free Spectral Range  m Grating cross-dispersed echelle spectrographs

34. On a detector we only measure x- and y- positions, there is no information about wavelength. For this we need a calibration source y x

35. CCD detectors only give you x- and y- position. A doppler shift of spectral lines will appear as  x  x →  →  v How large is  x ?

36. Spectral Resolution d 1 2 Consider two monochromatic beams They will just be resolved when they have a wavelength separation of d Resolving power: d = full width of half maximum of calibration lamp emission lines R = d ← 2 detector pixels

37. R = 50.000 →  = 0.11 Angstroms → 0.055 Angstroms / pixel (2 pixel sampling) @ 5500 Ang. 1 pixel typically 15  m 1 pixel = 0.055 Ang → 0.055 x (310 8 m/s)/5500 Ang → = 3000 m/s per pixel = v  c  v = 10 m/s = 1/300 pixel = 0.05  m = 5 x 10 –6 cm  v = 1 m/s = 1/1000 pixel → 5 x 10 –7 cm

38. RAng/pixelVelocity per pixel (m/s)  pixel Shift in mm 500 0000.0053000.060.001 200 0000.1257500.0274×10 –4 100 0000.02515000.01332×10 –4 50 0000.05030000.006710 –4 25 0000.1060000.0335×10 –5 10 0000.25150000.001332×10 –5 5 0000.5300006.6×10 –4 10 –5 1 0002.51500001.3×10 –4 2×10 –6 So, one should use high resolution spectrographs….up to a point For  v = 20 m/s How does the RV precision depend on the properties of your spectrograph?

39. Wavelength coverage: Each spectral line gives a measurement of the Doppler shift The more lines, the more accurate the measurement:  Nlines =  1line /√N lines → Need broad wavelength coverage Wavelength coverage is inversely proportional to R:  detector Low resolution High resolution 

40. Noise:  Signal to noise ratio S/N = I/  I For photon statistics:  = √I → S/N = √I I = detected photons Note: recall that if two stars have magnitudes m 1 and m 2, their brightness ratio is  B = 2.512 (m 1 – m 2 )

41.  (S/N) –1 Price: S/N  t 2 exposure 14 Exposure factor 1636 144400 S/N  t 2 exposure

42. How does the radial velocity precision depend on all parameters?  (m/s) = Constant × (S/N) –1 R –3/2 (  ) –1/2  : error R: spectral resolving power S/N: signal to noise ratio  : wavelength coverage of spectrograph in Angstroms For R=110.000, S/N=150,  =2000 Å,  = 2 m/s C ≈ 2.4 × 10 11

43. The Radial Velocity precision depends not only on the properties of the spectrograph but also on the properties of the star. Good RV precision → cool stars of spectral type later than F6 Poor RV precision → hot stars of spectral type earlier than F6 Why?

44. A7 star K0 star Early-type stars have few spectral lines (high effective temperatures) and high rotation rates.

45. A0 A5 F0 F5 RV Error (m/s) G0G5 K0 K5 M0 Spectral Type Main Sequence Stars Ideal for 3m class tel. Too faint (8m class tel.). Poor precision 98% of known exoplanets are found around stars with spectral types later than F6

46. Including dependence on stellar parameters v sin i : projected rotational velocity of star in km/s f(T eff ) = factor taking into account line density f(T eff ) ≈ 1 for solar type star f(T eff ) ≈ 3 for A-type star f(T eff ) ≈ 0.5 for M-type star  (m/s) ≈ Constant ×(S/N) –1 R –3/2 v sin i ( 2 ) f(T eff ) (  ) –1/2

47. Eliminate Instrumental Shifts Recall that on a spectrograph we only measure a Doppler shift in  x (pixels). This has to be converted into a wavelength to get the radial velocity shift. Instrumental shifts (shifts of the detector and/or optics) can introduce „Doppler shifts“ larger than the ones due to the stellar motion z.B. for TLS spectrograph with R=67.000 our best RV precision is 1.8 m/s → 1.2 x 10 –6 cm

48. Traditional method: Observe your star→ Then your calibration source→

49. Problem: these are not taken at the same time…... Short term shifts of the spectrograph can limit precision to several hunrdreds of m/s

51. Solution 1: Observe your calibration source (Th-Ar) simultaneously to your data: Spectrographs: CORALIE, ELODIE, HARPS Stellar spectrum Thorium-Argon calibration

52. Advantages of simultaneous Th-Ar calibration: Large wavelength coverage (2000 – 3000 Å) Computationally simple: cross correlation Disadvantages of simultaneous Th-Ar calibration: Th-Ar are active devices (need to apply a voltage) Lamps change with time Th-Ar calibration not on the same region of the detector as the stellar spectrum Some contamination that is difficult to model Cannot model the instrumental profile, therefore you have to stablize the spectrograph

53. One Problem: Th-Ar lamps change with time!

54. Solution 2: Absorption cell a) Griffin and Griffin: Use the Earth‘s atmosphere:

55. O2O2 6300 Angstroms

56. Filled circles are data taken at McDonald Observatory using the telluric lines at 6300 Ang. Example: The companion to HD 114762 using the telluric method. Best precision is 15–30 m/s

57. Limitations of the telluric technique: Limited wavelength range (≈ 10s Angstroms) Pressure, temperature variations in the Earth‘s atmosphere Winds Line depths of telluric lines vary with air mass Cannot observe a star without telluric lines which is needed in the reduction process.

58. Absorption lines of the star Absorption lines of cellAbsorption lines of star + cell b) Use a „controlled“ absorption cell

59. Campbell & Walker: Hydrogen Fluoride cell: Demonstrated radial velocity precision of 13 m s –1 in 1980!

60. Drawbacks: Limited wavelength range (≈ 100 Ang.) Temperature stablized at 100 C Long path length (1m) Has to be refilled after every observing run Dangerous

61. A better idea: Iodine cell (first proposed by Beckers in 1979 for solar studies) Advantages over HF: 1000 Angstroms of coverage Stablized at 50–75 C Short path length (≈ 10 cm) Can model instrumental profile Cell is always sealed and used for >10 years If cell breaks you will not die! Spectrum of iodine

62. Spectrum of star through Iodine cell:

63. The iodine cell used at the CES spectrograph at La Silla

64. Modelling the Instrumental Profile: The Advantage of Iodine What is an instrumental profile (IP): Consider a monochromatic beam of light (delta function) Perfect spectrograph

65. Modelling the Instrumental Profile (IP) We do not live in a perfect world: A real spectrograph IP is usually a Gaussian that has a width of 2 detector pixels The IP is a „smoothing“ or „blurring“ function caused by your instrument

66. Convolution  f(u)  (x–u)du = f *  f(x):  (x):

67. Convolution  (x-u) a1a1 a2a2 g(x) a3a3 a2a2 a3a3 a1a1 Convolution is a smoothing function. In our case f(x) is the stellar spectrum and  (x) is our instrumental profile.

68. The IP is not so much the problem as changes in the IP No problem with this IP Or this IP Unless it turns into this Shift of centroid will appear as a velocity shift

69. Use a high resolution spectrum of iodine to model IP Iodine observed with RV instrument Iodine Observed with a Fourier Transform Spectrometer

70. FTS spectrum rebinned to sampling of RV instrument FTS spectrum convolved with calculated IP Observed I 2

71. Model IP Gaussians contributing to IP IP =  g i Sampling in Data space Sampling in IP space = 5×Data space sampling

72. Instrumental Profile Changes in ESO‘s CES spectrograph 2 March 1994 14 Jan 1995

73. Instrumental Profile Changes in ESO‘s CES spectrograph over 5 years:

74. Mathematically you solve this equation: I obs ( ) = k[T I2 ( )I s ( +  )]*IP Where: I obs ( ) : observed spectrum T I2 : iodine transmission function k: normalizing factor I s ( ): Deconvolved stellar spectrum without iodine and with IP removed IP: Instrumental Point Spread Function Problem: IP varies across the detector. Solve this equation iteratively

75. Modeling the Instrumental Profile In each chunk: Remove continuum slope in data : 2 parameters Calculate dispersion (Å/pixel): 3 parameters (2 order polynomial: a0, a1, a2) Calculate IP with 5 Gaussians: 9 parameters: 5 widths, 4 amplitudes (position and widths of satellite Gaussians fixed) Calculate Radial Velocity: 1 parameters Combine with high resolution iodine spectrum and stellar spectrum without iodine Iterate until model spectrum fits the observed spectrum

76. Sample fit to an observed chunk of data

77. Sample IP from one order of a spectrum taken at TLS But why is the IP double peaked?

78. Recall that at the detector we have an image of a slit. The slit is square shape and this is convolved with a Gaussian emission line profile, in other words they should have a flat top The modeling uses a sum of Gaussians and the only way to mimic a flat top is with two Gaussians. The IP function is oversampled, when you rebin to 2 pixels of „real detector“ space, one no longer sees the dip in the middle and the IP is square topped. 2 detector pixels

79. WITH TREATMENT OF IP-ASYMMETRIES

80. HARPS Simultaneous ThAr cannot model the IP. One has to stabilize the entire spectrograph

81. Valenti, Butler, and Marcy, 1995, PASP, 107, 966, „Determining Spectrometer Instrumental Profiles Using FTS Reference Spectra“ Butler, R. P.; Marcy, G. W.; Williams, E.; McCarthy, C.; Dosanjh, P.; Vogt, S. S., 1996, PASP, 108, 500, „Attaining Doppler Precision of 3 m/s“ Endl, Kürster, Els, 2000, Astronomy and Astrophysics, “The planet search program at the ESO Coudé Echelle spectrometer. I. Data modeling technique and radial velocity precision tests“ Additional information on RV modeling:

82. Barycentric Correction Earth’s orbital motion can contribute ± 30 km/s (maximum) Earth’s rotation can contribute ± 460 m/s (maximum)

83. Needed for Correct Barycentric Corrections: Accurate coordinates of observatory Distance of observatory to Earth‘s center (altitude) Accurate position of stars, including proper motion:   ′  ′ Worst case Scenario: Barnard‘s star Most programs use the JPL Ephemeris which provides barycentric corrections to a few cm/s

84. For highest precision an exposure meter is required time Photons from star Mid-point of exposure No clouds time Photons from star Centroid of intensity w/clouds Clouds

85. Differential Earth Velocity: Causes „smearing“ of spectral lines Keep exposure times < 20-30 min