0803.2935/README
1: Chris Sneden, Letters Editor
2: THE ASTROPHYSICAL JOURNAL LETTERS
3: apjletters@letters.as.utexas.edu
4: 
5: Re: ApJL MS#: 22774
6:     Title:    Orbital Dynamics of a Second Planet in HD17156
7:     Authors:  D. Short, W. F. Welsh, J. A. Orosz, and G. Windmiller
8: 
9: 
10: Dear Dr. Sneden,
11: 
12: We have revised our paper and resubmit our work for your consideration.
13: We have made substantial improvements to the paper, following the 
14: suggestions of the referee plus our own continued efforts. We outline 
15: the revisions below, following the comments of the referee.
16: 
17: Sincerely,
18: 
19: Don Short, Bill Welsh, Jerry Orosz and Gur Windmiller
20: 
21:                           ....................
22: 
23: I. "The title of the paper is really the first problem with this paper."
24: We appreciate the referee's appropriate skepticism of the possibility
25: of a second planet in the system, and we address that in detail below.
26: The title accurately reflects the content of the paper and in our 
27: opinion does not claim an unambiguous detection. However, this is a 
28: very minor point and if the referee requests a revised title like
29: "Orbital Dynamics of a Possible Second Planet in HD17156" that would 
30: be fine.
31: 
32: 
33: II. "the second serious issue with the paper: the case for a second 
34:     planet is not securely made. ... Therefore, this result demands far 
35:     more than the usual care in both understanding sources of possible 
36:     systematic errors in the data and in assessing the confidence level 
37:     of the detection."
38: The intent of our paper was not to claim a detection of a second 
39: planet, but rather to point out the possibility of such a planet, and 
40: in particular, its fascinating orbital dynamics. The main thrust of the 
41: paper is dynamics, not detection. The referee is of course correct that 
42: ample evidence is needed to claim a discovery. The very small radial 
43: velocity signal this planet produces means that the evidence cannot be
44: anywhere near as secure as for other extrasolar planet detections, such 
45: as for planet b itself. Thus throughout the paper we speak of this as a 
46: "probable" planet, and we make extensive predictions for testing if 
47: such a planet exists. In many ways, we are predicting the existence of 
48: a second planet based on the available data, and providing specific 
49: observations to test that prediction. In our revised paper we have 
50: added the following to support the case that a second planet probably 
51: is present.
52: -  The full Newtonian 3-body solution with 2 planets gives a 
53:    significantly better fit to the data (not just the residuals) than
54:    a single planet. We demonstrate this by the drop in rms of the 
55:    residuals and the reduced chi square of the fit (which of course 
56:    includes the number of free parameters). The significance of this 
57:    chi square drop is checked via a standard F-test.  
58: -  We have computed formal uncertainties for each parameter of the fit
59:    and these show that the K velocity and mass for the hypothetical 2nd
60:    planet are both approximately 3.4 sigma above zero. In other words,
61:    the planet seems to be secure at better than the 99% level. However,
62:    we remain cautious and continue to call this a "probable planet".
63:  - We computed an "Keplerian periodogram", optimizing the chi square of
64:    a Keplerian fit to the RV data. Unlike a power spectrum that uses 
65:    sinusoids as the basis functions, we use the actual Keplerian orbital 
66:    parameters. The periodogram includes the uncertainty in the 
67:    observations, and so we can estimate the significance of the fit 
68:    directly, without need to resort to Monte Carlo or bootstrap 
69:    sampling. The periodograms shows the presence of both planets at a 
70:    high level of significance. Of course this is still only an 
71:    approximation; we need to solve the full 3-body problem to 
72:    "do it right". 
73: -  At the end of section 4 we added a paragraph starting with the 
74:    following, "The question remains, does HD 17156c exist?" addressing 
75:    the above concern.
76: Finally, we point out a vital aspect of our model. We are not simply 
77: fitting the radial velocities. The referee is mostly correct in    
78: thinking that this might not be sufficient to discern the presence of 
79: a second planet. In addition to the RV data, we simultaneously fit the
80: 4 known transit times. These timing data are extremely powerful in that 
81: they very tightly constrain the exact orbital configuration of planet b 
82: at those times. They are *much* more discriminating than simply 4 more 
83: radial velocities.
84: 
85: 
86: 
87: III. Regarding the Referee's Concern with the Power Spectrum
88: We respond to all of the referee's concerns, though in fact we opt to 
89: omit any spectral/periodogram analysis as described below.
90:  1. Monte Carlo simulation
91:  The text correctly states what we have done, in that we have carried 
92:  out a Monte Carlo analysis by adding Gaussian distributed white noise
93:  to each datum, consistent with that datum's uncertainty. "Datum" 
94:  means RV residual in this case. We preserved the times of observation. 
95:  We repeat this 500 times, yielding 500 sets of residuals, all 
96:  statistically consistent with the original. These 500 realizations are 
97:  transformed into power spectra and averaged. We repeat this proceedure 
98:  for the case when the residuals are initially all set to zero and then 
99:  noise is added, to show what the "null hypothesis" would yield. 
100:  We have not done a bootstrap resampling. 
101:  The power spectrum indicated a spike near 100-150 days that was well 
102:  above the null hypothesis baseline. This led to further investigation. 
103:  But we stress that a power spectrum is a crude tool to use; it does 
104:  not include any a priori information that we have on hand, like the
105:  bodies must obey gravity. As such, it is a useful, but far from 
106:  definitive, tool. The definitive tool is to model the data using full 
107:  Newtonian gravitational physics. POFP can do this. This is not to say 
108:  POFP is perfect - we ignore general relativity and tidal forces 
109:  for example. But it is far superior to any power spectrum analysis. 
110:  It is important to note that the power spectrum of the RV data can 
111:  only be informative and not conclusive. If one considers Figure 4,
112:  the radial velocity difference between the 1 and 2 planet model, the 
113:  difference remains quite small in the near future except for excursions 
114:  near the time of periastron of planet b. Thus, the RV data alone will 
115:  not suffice to distinguish between the two models unless specifically
116:  obtained near periastron of planet b.
117:  For these reasons, we have omitted the only somewhat useful power 
118:  spectrum figure from the revised paper so that we have room to include 
119:  a much more useful figure.
120: 
121: 2. "window function" concerns
122:  The Monte Carlo analysis includes the effects of the sampling 
123:  window (because the time of the observations were preserved in the
124:  500 simulations) and shows the 110 d peak well above the background 
125:  caused just by noise. Removing the stellar RV reflex motion due to  
126:  planet c leaves behind no residual signal, showing that the ~115 d
127:  peak is not the result of the window function. If it were due to
128:  spectral leakage from a longer period, the peak would not vanish.
129: 
130: 3. FAP
131:  The false alarm probability is a useful statistic when the noise are
132:  Gaussian and white, the data equally sampled, and there are no large 
133:  gaps in the time series. However, this is not the case: the RV data 
134:  sampling is irregular and sparse, and stellar jitter is correlated 
135:  noise, not white noise. As such, power spikes could appear to be 
136:  highly significant, then in fact they are not - they could simply be 
137:  aliases for example (motivating the window function mentioned above). 
138:  False positives are very likely. For that reason the false alarm 
139:  probability is at best only mildly helpful in this case, whether 
140:  it states a peak is apparently significant or insignificant. It is 
141:  much better to model the data using physics than to rely on peaks in 
142:  an imperfect Fourier domain representation.
143: 
144: 4. "...properly accounted for the increased degrees of freedom..."
145:  While this is by definition included in the reduced chi-square, we 
146:  agree that is important to explictly list the numbers of degrees of
147:  freedom, or equivalently, the number of data points and number of 
148:  parameters in the model. We now do this in Table 2.
149: 
150: 5. "broad peak in the periodogram"
151:  The broad peak was due in part to undersampling in the generation of 
152:  the power spectrum. Even so, the width is not all that important - 
153:  because the orbit is non-sinusoidal, the power will not be a delta 
154:  function even if the data were perfectly sampled and noise-free. 
155:  Irregular sampling and gaps add to leakage that tapering can only
156:  partly compensate for, substantially further broadening the peak.
157:  The period of the peak itself is what is important, but again, only 
158:  in that it provides an initial guess for the correct Newtonian model.
159: 
160: 
161: 
162: IV. "...uncertainties associated with the Subaru data are poorly known.
163:      ...Therefore your analysis should show the strongest result with 
164:     the Keck-only data."
165: This comment was taken to heart and our analysis repeated with and 
166: without using the Subaru RV data. Fortunately the results remain the 
167: same: the 2-planet solution is superior to the 1-planet solution whether 
168: or not the Subaru data are included. It is the inclusion of the transit 
169: times allowed us to still find evidence for the 2nd planet, even without 
170: the Subaru velocities. The results are included in a new table (Table 2).
171: 
172: 
173: 
174: V. "...the statement in the discussion that a 1 m/s Rossiter-McLaughlin 
175:    amplitude should be observable is incorrect."
176: We have removed this statement and most of the section as well.
177: 
178: 
179: 
180: VI. "It's probably better to carry out high precision photometry from 
181:     above the Earths atmosphere (Spitzer or HST)."
182: The referee is absolutely correct, and to help plan observations, we have 
183: computed a predicted O-C diagram.
184: 
185:                                 ....
186: 
187: In addition to the above concerns of the referee, we have done the 
188: following:
189:  
190:  Using a fully independent code (HNBody) we have carried out a 
191:  100 million day integration of the two-planet model. We verify 
192:  the system's long-term  stability and confirm the 5:1 mean motion  
193:  resonance.
194: 
195:  We have added a new figure (Fig. 3) illustrating the dynamical 
196:  coupling of the 2 planets. It succinctly shows the long term planetary 
197:  orbital characteristics compressed onto 1 figure.
198: 
199:  Further optimization of the models have not changed the short-term
200:  characteristics of the orbits, but the very long-term cycles of 
201:  212 years and 37,700 years have been refined. Because these are 
202:  unobservable (even on a grad student thesis timescale) they are 
203:  poorly constrained. The improved values are now 141 and 33,000 years.
204: 
205: 
206: Finally:
207: 
208: Because of the low planet c mass, the small number of observations 
209: and the short time span of those observations, we share the referee's 
210: skepticism on the existence of planet c, and we hope that skepticism
211: is clear in our writing. The statistics strongly support the existence
212: of a second planet. However, we all know that statistical probabilities 
213: are not always that convincing, and certainly not satisfying. Thus in 
214: the discussion we make predictions and emphasize observational ways to 
215: discriminate between the 1 or 2 planet scenario.
216: