astro-ph0106042/ms.tex
1: \documentclass[12pt,preprint]{aastex}
2: 
3: \newcommand{\likelihood}{\mathcal L}
4: \newcommand{\Kmin}{$K_{\rm min}$}
5: \newcommand{\Mo}{$M_{\odot}$}
6: \newcommand{\Mt}{$M_{2}$}
7: \newcommand{\MJ}{$M_{J}$}
8: \newcommand{\kms}{km s$^{-1}$}
9: \newcommand{\ms}{m s$^{-1}$}
10: \newcommand{\Ro}{$R_{\odot}$}
11: \newcommand{\MAXLIMA}{{\footnotesize MAXLIMA}}
12: 
13: \shorttitle{Mass Distribution of Planets}
14: \shortauthors{Zucker \& Mazeh}
15: 
16: 
17: \begin{document}
18: 
19: \title{Derivation of the Mass Distribution of Extrasolar Planets
20: with \MAXLIMA\ - a Maximum Likelihood Algorithm}
21: 
22: \author{Shay Zucker and Tsevi Mazeh}
23: \affil{School of Physics and Astronomy, Raymond and Beverly Sackler
24: Faculty of Exact Sciences, Tel Aviv University, Tel Aviv, Israel}
25: \email{shay@wise.tau.ac.il; mazeh@wise7.tau.ac.il} 
26: 
27: \vskip 4.8 truecm
28: 
29: \begin{abstract}
30: We construct a maximum-likelihood algorithm - \MAXLIMA, to derive the
31: mass distribution of the extrasolar planets when only the minimum
32: masses are observed. The algorithm derives the distribution by solving
33: a numerically stable set of equations, and does {\it not} need any
34: iteration or smoothing. Based on 50 minimum masses, \MAXLIMA\ yields a
35: distribution which is approximately flat in $\log M$, and might rise
36: slightly towards lower masses. The frequency drops off very sharply
37: when going to masses higher than 10 \MJ, although we suspect there is
38: still a higher mass tail that extends up to probably 20 \MJ.  We
39: estimate that 5\% of the G stars in the solar neighborhood have
40: planets in the range of 1--10 \MJ\ with periods shorter than 1500
41: days.  For comparison we present the mass distribution of stellar
42: companions in the range of 100--1000\MJ, which is also approximately
43: flat in $\log M$. The two populations are separated by the
44: ``brown-dwarf desert'', a fact that strongly supports the idea that
45: these are two distinct populations.  Accepting this definite
46: separation, we point out the conundrum concerning the similarities
47: between the period, eccentricity and even mass distribution of the two
48: populations.
49: \end{abstract}
50: 
51: \subjectheadings{binaries: spectroscopic --- 
52: methods: statistical ---
53: planetary systems ---
54: stars: low-mass, brown dwarfs ---
55: stars: statistics}
56: 
57: \section{INTRODUCTION}
58: 
59: Since the detection of the first few extrasolar planets their mass
60: distribution was recognized to be a key feature of the growing new
61: population. In particular, the potential of the high end of the mass
62: distribution to separate between planets on one side and brown dwarfs
63: and stellar companions on the other side was pointed out by numerous
64: studies (e.g., Basri \& Marcy 1997; Mayor, Queloz \& Udry 1998; Mazeh,
65: Goldberg \& Latham 1998).  A clear mass separation between the two
66: populations could even help to clarify one of the very basic questions
67: concerning the population of extrasolar planets --- the precise
68: definition of a planet (Burrows et al. 1997; see a detailed discussion
69: by Mazeh \& Zucker 2001). The lower end of the mass distribution could
70: indicate how many Saturn- and Neptune-like planets we expect on
71: the basis of the present discoveries, a still unsurveyed region of the
72: parameter space of extrasolar planets.
73: 
74: The present number of known extrasolar planets --- more than 60 are known
75: (Encyclopedia of extrasolar planets, Schneider 2001), offers an
76: opportunity to derive a better estimate of the mass distribution of this
77: population. In order to use the derived masses of the extrasolar planets
78: we have to correct for two effects. The first one is the unknown orbital
79: inclination, which renders the derived masses only {\it minimum}
80: masses. The second effect is due to the fact that stars with too small
81: radial-velocity amplitudes could not have been detected as
82: radial-velocity variables. Therefore, planets with masses too small,
83: orbital periods too large, or inclination angles too small are not
84: detected.
85: 
86: The effect of the unknown inclination of spectroscopic binaries was
87: studied by numerous papers (e.g., Mazeh \& Goldberg 1992; Heacox 1995;
88: Goldberg 2000), assuming random orientation in space. Heacox (1995)
89: calculates first the minimum mass distribution and then uses its
90: relation to the mass distribution to derive the latter. This
91: calculation amplifies the noise in the observed data, and necessitates
92: the use of smoothing to the observed data.  Mazeh \& Goldberg (1992)
93: introduce an iterative algorithm whose solution depends on the initial
94: guess.  In the present work we followed Tokovinin (1991, 1992) and
95: constructed an algorithm --- MAXimum LIkelihood MAss, to derive the
96: mass distribution of the extrasolar planets with a maximum likelihood
97: approach. \MAXLIMA\ assumes that the planes of motion of the planets
98: are randomly oriented in space and derives the mass distribution
99: directly by solving a set of numerically stable linear equations. It
100: does not require any smoothing of the data nor any iterative
101: algorithm. \MAXLIMA\ also offers a natural way to correct for the
102: undetected planets.
103: 
104: The randomness of the orbital planes of the discovered planets were
105: questioned recently by Han et al. (2001), based on the analysis of
106: Hipparcos data. However a few very recent studies (Pourbaix 2001;
107: Pourbaix \& Arenou 2001; Zucker \& Mazeh 2001a,b) showed that the
108: Hipparocos data do not prove the nonrandomness of the orbital
109: planes, allowing us to apply \MAXLIMA\ to the sample of known minimum
110: masses of the planet candidates.
111: 
112: In the course of preparing this paper for publication we have learned
113: about a similar paper by Jorissen, Mayor \& Udry (2001) that was
114: posted on the Astrophysics e-Print Archive (astro-ph).
115: Like Heacox (1995), Jorrisen et al. derive first the
116: distribution of the minimum masses and then apply two alternative algorithms to
117: invert it to the distribution of planet masses. One algorithm is a
118: formal solution of an Abel integral equation and the other is the
119: Richardson-Lucy algorithm (e.g., Heacox 1995). The first algorithm
120: necessitates some degree of data smoothing  and the second one 
121: requires a series of iterations.  The results of the first algorithm depend on
122: the degree of smoothing applied, and those of the second one on the
123: number of iterations performed.  \MAXLIMA\ has no built-in free parameter,
124: except the widths of the histogram bins. In addition, Jorissen et al.\ did not
125: apply any correction to the selection effect we consider here, and displayed
126: their results on a linear mass scale. We feel that a
127: logarithmic scale can illuminate some other aspects of the
128: distribution. Despite all the differences, our results are completely
129: consistent with those of Jorrisen et al., the sharp cutoff in the planet
130: mass distribution at about 10 \MJ, and the small high-mass tail that
131: extends up to about 20 \MJ\ in particular.
132: 
133: Section 2 presents \MAXLIMA, while section 3 presents our
134: results. Section 4 discusses briefly our findings.
135: 
136: 
137: \section{MAXLIMA}
138: 
139: \subsection{The Unknown Inclination}
140: 
141: Our goal is to estimate the probability density function (PDF) of the
142: secondary mass --- $f_m(m) dm$, given a set of observed minimum masses
143: \{$y_j; j=1,N\}$, where $y_j=m_j\times \sin i_j$, $m_j$ is the mass of
144: the j-th planet and $i_j$ is its inclination. Within certain
145: assumptions and limitations, \MAXLIMA\ finds the function $f_m$ that
146: maximizes the likelihood of observing these minimum masses.
147: 
148: Note that the present realization of \MAXLIMA\ uses the approximation that
149: the mass of the unseen companion is much smaller than the primary
150: mass. Within this approximation $y_j$ can be derived from the
151: observations of each system, given the primary mass.  In general, when
152: the secondary mass is not so small, the value of $m_j\times \sin i_j$
153: cannot be derived from the observations, and a more complicated
154: expression has to be used. Nevertheless, the extension of \MAXLIMA\ to
155: those cases is straightforward, and will be worked out in details in a
156: separate paper.
157: 
158: We assume that the directions of the angular momenta of the
159: systems are distributed isotropically, which will cause $\sin i$ to have
160: a PDF of the form:
161: $$
162: f_s(s)\,ds = \frac{s}{\sqrt{1-s^2}}\,ds \ ,
163: $$
164: where we denote $\sin i$ by $s$.
165: We further assume that the planet mass and its orbital inclination are
166: uncorrelated, and therefore the {\it joint} PDF of $\sin i$ and the
167: planet mass
168: have the form:
169: $$
170: f_{ms}(m,s)\,dm\,ds = f_m(m)\ \frac{s}{\sqrt{1-s^2}}\,dm\,ds \ .
171: $$
172: 
173: Now, define a variable $y=m\times \sin i =m\times s $, 
174: which has the PDF
175: 
176: $$
177: f_y(y)dy=
178: \int{ f_m(m){f_s(y/m)\over m}}\,dm\,dy = 
179: \int{ {f_m(m) \over m}{(y/m)\over \sqrt{1-(y/m)^2}} }\,dm\,dy \ .
180: $$
181: A sketch of contours of constant $y$ in the $m-s$ parameter space is
182: plotted in Figure~1.
183: 
184: We wish to estimate $f_m(m)$ in the form of a histogram with
185: K bins, between the limits  $m_{min} \leq m < m_{max}$.
186: We thus consider a partition of
187: the interval $[m_{min},m_{max})$:
188: $$
189: m_{min}=m_1,m_2,\ldots,m_{K+1}=m_{max}\ ,
190: $$
191: for which the PDF gets the form:
192: $$
193: f_m(m) \equiv b_k,\; 
194: {\rm if\ } m_k \leq m < m_{k+1},\; {\rm for\ } k=1,\ldots,K.
195: $$
196: 
197: 
198: Note, that the function $f_m(m)$ is supposed to be a probability density
199: function, and therefore its integral must equal one:
200: \begin{equation}
201: \sum_{k=1}^K b_k \Delta m_k = 1  \ ,
202: \end{equation}
203: where $\Delta m_k \equiv m_{k+1}-m_k$ for $k=1,\ldots,K$. 
204: 
205: The PDF of $y$ then gets the form:
206: \begin{eqnarray*}
207: f_y(y)\,dy &= &
208: \sum_{k=1}^K \int_{m_k}^{m_{k+1}} \frac{b_k}{m} 
209: \frac{y/m}{\sqrt{1-(y/m)^2}}\,dm\,dy \\
210:            &= &
211: \sum_{k=1}^K b_k \int_{m_k}^{m_{k+1}} \frac{1}{m} 
212: \frac{y/m}{\sqrt{1-(y/m)^2}}\,dm\,dy\ ,
213: \end{eqnarray*}
214: where the integrals do not depend on $f_m(m)$ at all, but only on the
215: intervals borders and $y$.
216: 
217: Now we can solve our problem in a maximum-likelihood fashion by
218: finding the set of $b_k$'s that maximizes the likelihood of the
219: actually observed values - ${y_j}$. The likelihood
220: function is:
221: \begin{eqnarray*} 
222: \likelihood(b_1,\ldots,b_K;y_1,\ldots,y_N) &= &\prod_{j=1}^N f_y(y_j) \\
223: &= &\prod_{j=1}^N \sum_{k=1}^K b_k \int_{m_k}^{m_{k+1}} \frac{1}{m} \frac{y_j/m}{\sqrt{1-(y_j/m)^2}}\,dm \\
224: &= &\prod_{j=1}^N \sum_{k=1}^K A_{jk} b_k \ ,
225: \end{eqnarray*}
226: where
227: \begin{equation}
228: A_{jk} = \cases {
229:  0                                        & $m_{k+1} \leq y_j$ \cr
230:  \int_{y_j}^{m_{k+1}} \frac{1}{m} 
231: \frac{y_j/m}{\sqrt{1-(y_j/m)^2}}\,dm      & $m_k \leq y_j < m_{k+1}$ \cr
232: \int_{m_k}^{m_{k+1}} \frac{1}{m} 
233: \frac{y_j/m}{\sqrt{1-(y_j/m)^2}}\,dm      & $y_j < m_k$ \ , \cr
234: }
235: \end{equation}
236: and the integrals are easily calculated even analytically.
237: 
238: In the appendix we present an elegant way to find the $b_k$'s that
239: maximize $\log \likelihood$ directly, without any iterations.
240: 
241: \subsection{Simulation}
242: 
243: In order to check the performance of \MAXLIMA\ and its realization we have
244: performed several simulations, some of which are presented in Figure
245: 2. In those simulations we generated an artificial sample of planets
246: drawn from populations with different PDFs of the planet masses, and  
247: inclinations oriented isotropically in space. To make the simulation similar
248: to the present work we chose the size of each sample to be 50
249: planets. We assumed no selection effects. We then applied \MAXLIMA\ to the
250: simulated sample, the results of which are plotted in Figure~2.
251: 
252: The three examples of Figure~2 clearly show the power of MAXLIMA.
253: 
254: 
255: \subsection{Selection Effect: Undetected Planets}
256: 
257: We assume that the sample is constructed of planets with period, $P$,
258: between $P_{\rm min} \leq P \leq P_{\rm max}$. We further assume that the
259: search for planets discovered all radial-velocity variables with
260: amplitude $K$ larger than $K_{\rm min}$. We have to correct for planets
261: not detected because they induce $K$ smaller than the threshold. To do
262: that we note that the amplitude can be written as 
263: \begin{equation} 
264: K(P,M_p,M_1,\sin i) = 
265: 204\,
266: \left( {P\over {{\rm day}}} \right) ^{-1/3}    \,  
267: \left( {M_1\over {M_{\odot}}} \right) ^{-2/3}    \, 
268: \left( {M_p\over {M_{Jup}}}   \right)            \, 
269: \sin i\ \,\   
270: \rm m\, s^{-1}                        \ . 
271: \end{equation} 
272: %
273: 
274: The expression $M_p\times \sin i$ is actually our $y$. For any {\it
275: given} value of $y$ and $M_1$ we can derive the maximum possibly
276: detected period --- $P_{\rm max-detect}$, given $K_{\rm min}$. This
277: implies that if we know the period distribution and we assume that the
278: period is uncorrelated to the mass distribution, we can estimate for each
279: of the given $y$'s the fraction of planets with long periods that were
280: not detected with the same $y$. This means that to correct for the
281: undetected planets with long periods we have to consider each of the j-th
282: detected systems as representing some $\alpha_j$ planets. If $P_{\rm
283: max-detect}$ is smaller than $P_{\rm max}$, then $\alpha_j$ is larger
284: than unity. Otherwise $\alpha_j$ is equal to unity.
285: 
286: We then can write a new generalized likelihood as:
287: \begin{eqnarray*} 
288: \likelihood(b_1,\ldots,b_K;y_1,\ldots,y_N;\alpha_1,\ldots,\alpha_N) 
289: &= &\prod_{j=1}^N \left( f_y(y_j) \right)^{\alpha_j}\\
290: &= &\prod_{j=1}^N \left( \sum_{k=1}^K b_k \int_{m_k}^{m_{k+1}} 
291:     \frac{1}{m} \frac{y_j/m}{\sqrt{1-(y_j/m)^2}}\,dm \right) ^{\alpha_j} \\
292: &= &\prod_{j=1}^N \left( \sum_{k=1}^K A_{jk} b_k \right)^{\alpha_j} \ .
293: \end{eqnarray*}
294: %
295: The Appendix shows an easy way used by \MAXLIMA\ to find a maximum to
296: $\log\likelihood$.
297: 
298: For simplicity we consider only circular orbits.  Eccentricity
299: introduces two effects (Mazeh, Latham \& Stefanik 1996), the first of
300: which is the dependence of $K$ on the eccentricity $e$.  Equation (3)
301: should include an additional factor of $(1- e^2)^{-1/2}$, which causes
302: $K$ to increase for increasing $e$.  The other factor is the dependence
303: of the detection threshold $K_{\rm min}$ on $e$.  Our simplifying
304: assumption about the constancy of $K_{\rm min}$ throughout the sample
305: breaks down when we consider eccentric orbits.  This is so because for
306: eccentric orbits the velocity variation tends to concentrate around the
307: periastron passage, and therefore $K_{\rm min}$ increases for increasing
308: eccentricity.  These two effects tend to cancel each other (Fischer \&
309: Marcy 1992), the net effect depends on the characteristics of the
310: observational search.  By running numerical simulations Mazeh et
311: al.\ (1996) have found that if the detection limit depends on the
312: r.m.s. scatter of the observed radial-velocity measurements, the two
313: effects cancel each other for any reasonable eccentricity. We therefore
314: chose not to include the eccentricity of the planets in our analysis.
315: 
316: 
317: \section{ANALYSIS AND RESULTS}
318: 
319: To apply \MAXLIMA\ to the current known sample of extrasolar planets
320: we considered all known planets and brown dwarfs as of April 2001. We
321: consider only G- or K-star primaries and therefore excluded Gls 876
322: from the sample.
323: 
324: Obviously, the present sample in not complete. In particular, not all
325: planets with long periods and small induced radial-velocity amplitudes
326: were discovered and/or announced. To acquire some degree of completeness
327: to our sample we have decided, somewhat arbitrarily, to exclude planets
328: with periods longer than 1500 days and with radial-velocity amplitudes
329: smaller than 40 m/s. The values of these two parameters determine
330: the correction of \MAXLIMA\ for the selection effect, for which we assumed
331: a period distribution which is flat in $\log P$. This choice of parameters
332: also implies that our analysis applies only to planets with periods
333: shorter than 1500 days. We further assumed that the primary mass is 1
334: \Mo\ for all systems.
335: 
336: In order to be consistent with the selection effects and the correction
337: we applied, we included in our analysis only planets that were discovered
338: by the high precision radial-velocity searches. We had to exclude
339: HD~114762 and similar objects that were discovered by other searches
340: (e.g., Latham et al.\ 1989; Mayor et al.\ 1992; Mazeh et al.\ 1996). 
341: This does not mean that we assume
342: anything about their nature in this stage of the study. Table~1 lists
343: all the known planets with G star primaries from the high precision
344: radial-velocity studies. Planets excluded from our analysis are marked
345: by an asterisk. We did not take into account the known inclination of
346: the planet around HD~209458. All together we are left with 50 planets.
347: 
348: The results of \MAXLIMA\ are presented in the lower panel of Figure~3 on a
349: logarithmic mass scale. Each bin is 0.3 dex wide, which means about a
350: factor 2 in mass. The value of each bin is the estimated number of
351: planets found in the corresponding range of masses in the known sample
352: of planets, after correcting for the undetected systems. To estimate the
353: uncertainty of each bin we ran 5000 Monte Carlo simulations and found
354: the r.m.s.\ of the derived values of each bin. Therefore, the errors
355: plotted in the figure represent only the statistical noise of the
356: sample. Obviously, any deviation from the assumptions of our model for
357: the selection effect induces further errors into the histogram, the
358: assumed period distribution in particular. This is specially true for
359: the first bin, where the actual number of systems is small and the
360: correction factor large. The value of the first bin is sensitive, for
361: example, to the assumed $K_{\rm min}$. Assuming $K_{\rm min}$ of 50 m/s
362: increased the value of the first bin by more than 50\%.
363: 
364: To compare the mass distribution of the planets with that of the stellar
365: secondaries we plot the latter on the same scale in an adjacent panel of
366: Figure~3. We plot here only two bins, with masses between 100 and 1000
367: \MJ. We follow the work of Mazeh (1999b) and Mazeh and Zucker (2001),
368: and used for those bins subsamples of binaries found by the CfA
369: radial-velocity search for spectroscopic binaries (Latham 1985) in the
370: Carney \& Latham (1987) sample of the high-proper-motion stars (Latham
371: et al.\ 2000; Goldberg et al.\ 2000). For the smaller bin we used a
372: subsample that included only the Galaxy disk stars (Goldberg 2000), and
373: for the larger-mass bin a subsample of this sample that included only
374: primaries with masses higher than 0.7 \Mo.  The values of those two bins
375: were derived with the algorithm of Mazeh \& Goldberg (1992).
376: 
377: Note that the upper panel {\it does not have any estimate of the values
378: of the bins with masses smaller than 100 \MJ}. This is so because the
379: CfA search does not have the sensitivity to detect secondaries in that
380: range. On the other hand, the lower panel does include information on
381: the bins below 100 \MJ. This panel presents the results of the
382: high-precision radial-velocity searches, and these searches could easily
383: detect stars with secondaries in the range of, say, 20--100 \MJ. We
384: assume that these binaries were not excluded from the various
385: radial-velocity searches at the first place, and further assume that
386: all, or at least most, findings of the various research groups
387: corresponding to this range of masses were already published. If these
388: two assumptions are correct, then the lower panel does represent the
389: frequency of secondaries in the mass range of 20--100 \MJ. This panel
390: shows that the frequency of secondaries in this range of masses is close
391: to zero. The present analysis is not able to tell whether this
392: ``brown-dwarf desert'' extends up to 60, 80 or 100 \MJ.
393: 
394: The relative scaling of the planets and the stellar companions is not
395: well known. The spectroscopic binaries come from well defined samples --
396: 577 stars for the lower-mass bin, and 312 stars for the higher-mass
397: stars (Goldberg 2000). However, this is not the case for the detected
398: planets, specially because the sample of published planets is not
399: complete and also because the search samples of the different groups are
400: not well documented in the public domain. We assumed, somewhat {\it arbitrarily},
401: that they come from a sample of 1000 stars, and scaled the stellar
402: bins accordingly. We also rescaled the stellar bins to account for the
403: fact that their bins are larger, and their period range extends up to 3000
404: days, assuming a flat distribution in $\log P$.  Therefore the values of
405: the stellar bins are our best estimate for the number of binaries for
406: 1000 stars within a mass range of 0.3 dex, and up to a period of 1500
407: days.
408: 
409: Obviously the relative scaling of the two panels has a large
410: uncertainty. This scaling uncertainty is {\it not} reflected in the
411: error bars of the higher panel. Nevertheless we think that the
412: comparison is illuminating, as will be discussed in the next section.
413: 
414: 
415: \section{DISCUSSION}
416: 
417: The grand picture that is emerging from Figure~3 strongly indicates that
418: we have here two distinct populations. The two populations are separated
419: by a ``gap'' of about one decade of masses, in the range between 10 and
420: 100 \MJ. Such a gap was already noticed by many early studies (Basri
421: \& Marcy 1997; Mayor, Queloz \& Udry 1998; Mayor, Udry \& Queloz 1998;
422: Marcy \& Butler 1998). Those early papers binned the mass distribution
423: linearly.  Here we follow our previous work (Mazeh et al.\ 1998)
424: and use a logarithmic scale to study the
425: mass distribution, because of the large range of masses, 0.5--1000 \MJ,
426: involved.  The logarithmic scale has also been used by Tokovinin (1992)
427: to study the secondary mass distribution in spectroscopic binaries, and
428: was suggested by Black (1998) to study the mass distribution of the
429: planetary-mass companions (see also Mazeh 1999a,b; Mazeh \& Zucker
430: 2001; Mayor et al. 2001). 
431: The gap or the ``brown-dwarf desert'' are
432: consistent also with the finding of Halbwachs et al.\ (2000), who used
433: Hipparcos data and found that many of the known brown-dwarf candidates
434: are actually stellar companions.
435: 
436: We will assume that the two populations are the planets, at the low-mass
437: side of Figure~3, and the stellar companions at the high-mass end of the
438: figure.  Interestingly, the mass distribution of {\it single} stars
439: extends far below 100 \MJ\ (e.g., Zapatero Osorio et al.\ 2000; Lucas \& Roche
440: 2000), indicating that the gap separating the two populations of {\it
441: companions} apparently does not exist in the population of single
442: stars/brown dwarfs. This difference probably indicates different
443: formation processes for single and secondary objects.
444: 
445: The distribution we derived in Figure~3 suggests that the planet mass
446: distribution is almost flat in $\log M$ over five bins --- from 0.3 to
447: 10 \MJ. Actually, the figure suggests a slight rise of the distribution
448: towards smaller masses. The distinction between these two distributions
449: is not possible at this point, when our knowledge about planets with
450: sub-Jupiter masses is very limited. At the high-mass end of the planet
451: distribution the mass distribution dramatically drops off at 10 \MJ,
452: with a small high-end tail in the next bin. Although the results are
453: still consistent with zero, we feel that the small value beyond 10 \MJ\
454: might be real. The dramatic drop at 10 \MJ\ and the small high-mass tail
455: agree with the findings of Jorissen et al.\ (2001).
456: 
457: Examination of the two panels of Figure~3 suggests that per equal dex
458: range of masses the frequency of stellar secondaries is higher than that
459: of the planets by a factor of about 2. As we emphasized in the previous
460: section, this is a very preliminary result that should be checked by
461: future observations. Nevertheless, the frequency of planets is
462: impressive by itself. Our results indicate that about 5\% of the stars
463: have planets with masses between 1 and 10 \MJ. This is so because the
464: number of multiple planets in this sample is small, so the number of
465: planets considered in the figure is about the number of stars found to
466: have one or more planets. If this frequency extends further down the
467: mass axis to Earth masses, we might find that more than 10\% of the
468: stars have planets with periods shorter than 1500 days.
469: 
470: The analysis presented here raises the question what mechanism can
471: produce flat or approximately flat mass distribution of planets up to 10
472: \MJ.  What determines the mass of the forming planet? The
473: present paradigm assumes that planets were formed out of a
474: protoplanetary disk. Is it the mass, density, angular momentum or the
475: viscosity of the disk that determined the planet mass? If planets were
476: formed by accreting gas onto a rocky core, is the planet mass determined
477: also by the location or evolutionary phase of the formation of the rocky
478: core?  Any detailed model of planet formation should account for this
479: mass distribution.
480: 
481: The clear distinction between the two populations suggests that planets
482: and stellar companions were formed by two different processes.  This is
483: so despite the striking similarity between the distributions of the
484: eccentricities and periods of the two populations (Heacox 1999;
485: Stepinski \& Black 2000, 2001; Mayor \& Udry 2000; Mazeh \& Zucker
486: 2001). Even the mass distributions of the two populations might be very
487: similar --- approximately flat in $\log M$. This is still a conundrum that
488: any formation model for planets as well as for binaries needs to solve.
489: 
490: \acknowledgments
491: 
492: We are indebted to Yoav Benjamini for illuminating discussions.  This
493: work was supported by the US-Israel Binational Science Foundation
494: through grant 97-00460 and the Israeli Science Foundation (grant
495: no. 40/00)
496: 
497: 
498: \appendix
499: \section*{Appendix}
500: 
501: We want to find the maximum likelihood to observe a given set of
502: observed minimum masses $\{y_j;j=1,N\}$. As usual, it is easier to
503: maximize the logarithm of the likelihood function:
504: $$
505: \log \likelihood = \sum_{j=1}^N \alpha_j \times 
506:                    \log \left( \sum_{k=1}^K A_{jk} b_k \right) \ ,
507: $$
508: %
509: where each $A_{jk}$ depends on the corresponding $y_j$ through
510: Equation (2).
511: 
512: The $b_k$'s are not all independent. They are constrained by Equation
513: (1), and therefore we modify
514: our target function by adding a Lagrange multiplier term:
515: $$
516: \log \likelihood = \sum_{j=1}^N  \alpha_j \times
517:                    \log \left( \sum_{k=1}^K A_{jk} b_k \right) + 
518:     \lambda \left( \sum_{k=1}^K b_k \Delta m_k - 1 \right) \ .
519: $$
520: 
521: The optimization is performed by equating the partial derivatives of
522: this target function to zero:
523: \begin{eqnarray*}
524:  \frac{\partial \log \likelihood}{\partial b_k} &= 
525: &\sum_{j=1}^N
526: \frac{  \alpha_j \times A_{jk}}{\sum_{l=1}^K A_{jl} b_l} + 
527:                                 \lambda \Delta m_k = 0, \ 
528: {\rm for\ } k=1,\ldots,K \\
529:  \frac{\partial \log \likelihood}{\partial \lambda} &= 
530: &\sum_{k=1}^K b_k \Delta m_k -1 = 0 \ .
531: \end{eqnarray*}
532: The parameter 
533: $\lambda$ is eliminated quite easily. We first multiply each of the
534: $K$ equations by the corresponding $b_k$ and then sum them up
535: to get:
536: $$
537: \sum_{k=1}^K \sum_{j=1}^N\frac{   \alpha_j \times A_{jk} b_k}
538:           {\sum_{l=1}^K A_{jl} b_l} +
539:   \lambda \sum_{k=1}^K b_k \Delta m_k = 0 \ .
540: $$
541: Changing the order of summation reduces the first term, after a simple
542: manipulation, to simply 
543: $$
544: N_{eff}=\sum_{j=1}^N \alpha_j \ .
545: $$
546: Using the constraint reduces the second
547: term to $\lambda$, and we finally get:
548: $$
549: N_{eff} + \lambda = 0 \ ,
550: $$
551: and we can simply set $\lambda = -N_{eff}$. The $K$ equations we are now left
552: with are:
553: $$
554: \sum_{j=1}^N  \alpha_j \times A_{jk} 
555: \frac{1}{\sum_{l=1}^K A_{jl} b_l} = N_{eff} \Delta m_k \ .
556: $$
557: We have a set of $K$ non-linear equations in $K$ variables - $b_k$'s. 
558: 
559: An elegant
560: reduction of the complexity of the problem can be achieved if we set $K$
561: to $N$, by assigning $m_k \equiv y_k$ for $k = 1,\ldots,N$. Let us
562: also denote (remember that now $K=N$):
563: \begin{equation}
564: \label{E3} h_j \equiv \frac{1}{\sum_{l=1}^N A_{jl} b_l}
565: \end{equation}
566: The $N$ equations now look like:
567: \begin{equation}
568: \label{E4} \sum_{j=1}^N  \alpha_j \times A_{jk} h_j = N_{eff} \Delta m_k \ ,
569: \end{equation}
570: which is a system of $N$ linear equations in the $N$ variables
571: $h_j$. We can easily solve for them. The problem is even more 
572: easily solved when we note that the matrix $A_{jk}$ is upper-triangular
573: and thus the amount of computations needed for the solution is
574: smaller. Furthermore, examination of the integrals involved in
575: the calculation of
576: $A_{jk}$ (Eq. 2) shows that the matrix is very close to being 
577: diagonally-dominated, and therefore the set of linear equations is
578: numerically stable. Having solved for $h_j$ we face again a similar system of
579: linear equations in order to solve for $b_k$, coming from the
580: definition of $h_j$:
581: $$
582: \sum_{l=1}^N A_{jl} b_l = \frac{1}{h_j} .
583: $$
584: In this system the matrix is the transposed matrix of the
585: previous system of linear equations if the $\alpha_j$'s are all
586: equal to unity.
587: 
588: Obviously, we wish to estimate the densities of our original
589: intervals. But these are easily calculable from the $N$ densities $b_k$
590: calculated above, as simple linear combinations.
591: 
592: 
593: \begin{references}
594: 
595: \reference{} Basri, G. \& Marcy, G.\ W. 1997, in AIP Conf. Proc 393,
596: Star Formation, Near and Far, eds. S. Holt \& L.G. Mundy (New York, AIP),
597: 228
598: 
599: \reference{} Black, D. 1998, in Encyclopedia of the Solar System, 
600: eds. P. Weissman, L.-A. McFadden, T. Johnson (San Diego: Academic Press),
601: in press
602: 
603: \reference{} Burrows, A., Marley, M., Hubbard, W. B., Lunine, J. I.,
604:  Guillot, T., Saumon, D., Freedman, R., Sudarsky, D., \& Sharp, C. 1997, 
605:  ApJ, 491, 856
606: 
607: \reference{} Carney, B.\ W., \& Latham, D.\ W.\ 1987, \aj, 93, 116 
608: 
609: \reference{} Fischer, D.~A., \& Marcy, G.~W. 1992, \apj, 396, 178
610: 
611: \reference{} Goldberg, D. 2000, Ph.D. thesis, Tel Aviv University
612: 
613: \reference{} Goldberg, D., Mazeh, T., Latham, D.~W., Stefanik, R.~P., Carney,
614: B.~W., \& Laird, J.~B. 2000, submitted to \aap
615: 
616: \reference{} Halbwachs, J.-L., Arenou, F., Mayor, M., Udry, S., \& Queloz,
617:   D. 2000, \aap, 355, 581
618: 
619: \reference{} Han, I., Gatewood, G., \& Black, D. 2001, \apj, 548, L57
620: 
621: \reference{} Heacox, W.~D. 1995, \aj, 109, 2670
622: 
623: \reference{} Heacox, W.~D. 1999, \apj, 526, 928
624: 
625: \reference{} Jorissen, A., Mayor, M. \& Udry, S. 2001, submitted to
626: \aap, astro-ph/0105301
627: 
628: \reference{} Latham, D.~W. 1985, in IAU Colloq. 88, Stellar Radial Velocities,
629: ed. A.~G.~D.~Philip \& D.~W.~Latham (Schenectady, L.~Davis Press) 21
630: 
631: \reference{} Latham, D. W., Mazeh, T., Stefanik, R. P., Mayor, M., \&
632: Burki, G. 1989, \nat, 339, 38
633: 
634: \reference{} Latham, D.~W., Stefanik, R.~P., Torres, G., Davis, R.~J., 
635: Mazeh, T., Carney, B.~W., Laird, J.~B., \& Morse, J.~A. 2001, 
636: submitted to \aap
637: 
638: \reference{} Lucas, P. W., \& Roche, P. F. 2000, MNRAS, 314, 858
639: 
640: \reference{} Marcy, G.~W. \& Butler, R.~P 1998, \araa, 36, 57
641: 
642: \reference{} Mayor, M., Duquennoy, A., Halbwachs, J. L., \& Mermilliod,
643: J. C. 1992, in ASP Conf. Ser. 32, IAU Coll. 135, Complementary
644: Approaches to Double and Multiple Star Research, ed. H.A. McAlister
645: \& W.I. Hartkopf (San Francisco: ASP), 73
646: 
647: \reference{} Mayor, M.,  Queloz, D., \& Udry, S. 1998, in Brown Dwarfs
648: and Extrasolar Planets, eds. R. Rebolo, E.L. Martin, \& 
649: M.R. Zapatero-Osorio (San Francisco: ASP), 140
650: 
651: \reference{} Mayor, M. \& Udry, S. 2000, in Disks, Planetesimals and Planets, 
652: ed. F. Garcon, C. Eiron, D. de Winter, \& T.~J. Mahoney, in press
653: 
654: \reference{} Mayor, M., Udry, S., Halbwachs, J.-L, \& Arenou, F. 2001,
655: to be published in {\it Birth and Evolution of Binary Stars}, 
656: IAU Symp. 200, ASP Conf. Proc.,  eds. B. Reipurth and H. Zinnecker
657: 
658: \reference{} Mayor, T., Udry, S., \& Queloz, D. 1998, in ASP Conf. Ser 154, 
659: Tenth Cambridge Workshop on Cool Stars, Stellar Systems, and the Sun, 
660: eds. R. Donahue \& J. Bookbinder (San Francisco: ASP), 77
661: 
662: \reference{} Mazeh, T. 1999a, Physics Reports, 311, 317
663: 
664: \reference{} Mazeh, T. 1999b,, in ASP Conf. Ser. 185, IAU Coll. 170, Precise
665: Stellar Radial Velocities, eds. J.~B. Hearnshaw \& C.~D. Scarfe,  (San
666: Francisco: ASP), 131
667: 
668: \reference{} Mazeh, T. \& Goldberg, D. 1992, \apj, 394, 592
669: 
670: \reference{} Mazeh, T., Goldberg, D., \& Latham, D.W. 1998, \apj,
671: 501, L199
672: 
673: \reference{} Mazeh, T., Latham, D.~W., \& Stefanik R.~P. 1996, ApJ, 466,
674: 415
675: 
676: \reference{} Mazeh, T., \& Zucker, S. 2001, astro-ph/0008087, to be published
677: in {\it Birth and Evolution of Binary Stars}, IAU Symp. 200, ASP
678: Conf. Proc.,  eds. B. Reipurth and H. Zinnecker
679: 
680: \reference{} Pourbaix, D. 2001, \aap, 369, L22
681: 
682: \reference{} Pourbaix, D., \& Arenou, F. 2001, \aap, accepted, astro-ph/0104412
683: 
684: \reference{} Schneider, J. 2001, in Extrasolar Planets Encyclopaedia 
685: http://www.obspm.fr/planets
686: 
687: \reference{} Stepinski, T. F. \& Black, D. C. 2000, in IAU Symp. 200,
688: Birth and Evolution of Binary Stars, ed. B. Reipurth \& H. Zinnecker
689: (Potsdam) 167
690: 
691: \reference{} Stepinski, T. F. \& Black, D. C. 2001, \aap, 356, 903 
692: 
693: \reference{} Tokovinin, A.A. 1991, Sov. Astron. Lett., 17, 345 
694: 
695: \reference{} Tokovinin, A.A. 1992, \aap, 256, 121
696: 
697: %\reference{} Zapatero-Osorio, M. R. 2000, in From Extrasolar Planets
698: %to Cosmology: The VLT Opening Symp., ed. J. Bergeron \& A. Renzini
699: %(Berlin: Springer-Verlag), 511 
700: 
701: \reference{} Zapatero Osorio, M. R., Béjar, V. J. S., Martín, E. L.,
702:  Rebolo, R., y Navascués, D. Barrado, Bailer-Jones, C. A. L., Mundt,
703:  R. 2000, Science, 290, 103
704: 
705:  \reference{} Zucker, S., \& Mazeh, T. 2001a, submitted to \mnras,
706:  astro-ph/0104098
707: 
708: \reference{} Zucker, S., \& Mazeh, T. 2001b, submitted to \apj 
709: 
710: \end{references}
711: 
712: \begin{figure}
713: \plotone{f1.eps}
714: \caption{Three contours of constant $y$'s in the $m-s$
715: plane.}
716: \end{figure}
717: 
718: \begin{figure}
719: \epsscale{0.35}
720: \plotone{f2.eps}
721: \caption{Three simulations of MAXLIMA. In all three panels the dashed
722: line shows the input distribution and the histogram the results of
723: MAXLIMA.}
724: \end{figure}
725: 
726: \begin{figure}
727: \epsscale{1.0}
728: \plotone{f3.eps}
729: \caption{The mass distributions of the planets and the stellar
730: companions}
731: \end{figure}
732: 
733: \begin{deluxetable}{llcc}
734: \tabletypesize{\scriptsize}
735: \tablewidth{0pt}
736: \tablecolumns{4}
737: \tablecaption{Substellar companions list}
738: \tablehead{
739: \colhead{Name} & 
740: \colhead{$M\sin i$} & 
741: \colhead{$P$} &
742: \colhead{$K$} \\
743: \colhead{} &
744: \colhead{(\MJ)} &
745: \colhead{(days)} &
746: \colhead{(m\ s$^{-1}$)}
747: }
748: \startdata
749: $^\ast$HD 83443 b     & 0.16  & 29.83  & 14 \\
750: $^\ast$HD 16141       & 0.215 & 75.82  & 11 \\
751: $^\ast$HD 168746      & 0.24  & 6.41   & 28 \\
752: $^\ast$HD 46375       & 0.249 & 3.02   & 35 \\
753: HD 83443 c     & 0.34  & 2.985  & 56 \\
754: $^\ast$HD 108147      & 0.34  & 10.88  & 37 \\
755: HD 75289       & 0.42  & 3.51   & 54 \\
756: 51 Peg         & 0.47  & 4.23   & 56 \\
757: BD -10\degr 3166& 0.48 & 3.487  & 61 \\
758: $^\ast$HD 6434        & 0.48  & 22.09  & 37 \\
759: HD 187123      & 0.48  & 3.10   & 69 \\
760: $^\ast$Gliese 876 c   & 0.56  & 30.12  & 81 \\
761: HD 209458      & 0.69  & 3.52   & 86 \\
762: $\upsilon$ And b&0.69  & 4.617  & 71 \\
763: HD 192263      & 0.787 & 24.36  & 68 \\
764: HD 38529       & 0.81  & 14.32  & 54 \\
765: HD 179949      & 0.84  & 3.09   & 101 \\
766: 55 Cnc         & 0.84  & 14.65  & 77 \\
767: $^\ast\epsilon$ Eri & 0.86  & 2502.1 & 19 \\
768: $^\ast$HD 82943 c     & 0.88  & 222    & 34 \\
769: HD 121504      & 0.89  & 64.6   & 45 \\
770: HD 130322      & 1.02  & 10.72  & 115 \\
771: HD 37124       & 1.04  & 155.7  & 43 \\
772: $\rho$ CrB     & 1.1   & 39.65  & 67 \\
773: HD 52265       & 1.13  & 118.96 & 45 \\
774: $^\ast$HD 177830      & 1.22  & 391.6  & 34 \\
775: HD 217107      & 1.27  & 7.13   & 140 \\
776: HD 210277      & 1.28  & 437    & 41 \\
777: $^\ast$HD 27442       & 1.43  & 426.5  & 34 \\
778: 16 Cyg B       & 1.5   & 801    & 44 \\
779: HD 74156 b     & 1.56  & 51.6   & 108 \\
780: HD 134987      & 1.58  & 259.6  & 50 \\
781: HD 82943 b     & 1.63  & 445    & 46 \\
782: $^\ast$Gliese 876 b   & 1.89  & 61.02  & 210 \\
783: HD 160691      & 1.97  & 743    & 54 \\
784: HD 19994       & 2.0   & 454    & 45 \\
785: HD 213240      & 3.7   & 759    & 91 \\
786: $\upsilon$ And c&2.06  & 240.6  & 58 \\
787: HD 8574        & 2.23  & 228.8  & 76 \\
788: HR 810         & 2.26  & 320.1  & 67 \\
789: 47 UMa         & 2.39  & 1090   & 45 \\
790: HD 12661       & 2.79  & 252.7  & 88 \\
791: HD 169830      & 2.96  & 230.4  & 83 \\
792: $^\ast$14 Her         & 3.3   & 1654    & 73 \\
793: GJ 3021        & 3.32  & 133.82 & 164 \\
794: HD 92788       & 3.34  & 326.7  & 100 \\
795: HD 80606       & 3.41  & 111.8  & 414 \\
796: HD 195019      & 3.47  & 18.2   & 272 \\
797: $\tau$ Boo     & 3.87  & 3.31   & 469 \\
798: Gliese 86      & 4     & 15.78  & 380 \\
799: $\upsilon$ And d& 4.10 & 1313   & 68 \\
800: HD 50554       & 4.9   & 1279   & 95 \\
801: HD 190228      & 5     & 1161   & 95 \\
802: HD 222582      & 5.3   & 575.9  & 184 \\
803: HD 28185       & 5.6   & 385    & 168 \\
804: HD 10697       & 6.35  & 1072.3 & 119 \\
805: HD 178911 B    & 6.46  & 71.50  & 343 \\
806: 70 Vir         & 6.6   & 116.7  & 318 \\
807: HD 106252      & 6.81  & 1500   & 139  \\
808: HD 89744       & 7.2   & 256    & 257 \\
809: HD 168443 b    & 7.2   & 58.12  & 473 \\
810: $^\ast$HD 74156 c     & $>$7.5& 2300   & 121 \\
811: HD 141937      & 9.7   & 659    & 247 \\
812: $^\ast$HD 114762      & 11    & 84.03  & 600 \\
813: HD 202206      & 14.7  & 259    & 554 \\
814: $^\ast$HD 168443 c    & 15.1  & 1667   & 288 \\
815: $^\ast$HD 127506      & 36    & 2599   & 891 \\
816: \enddata
817: \end{deluxetable}
818: 
819: \end{document}
820: 
821: 
822: 
823: