0705.3039/ms.tex
1: \documentclass{emulateapj}
2: 
3: \slugcomment{ApJ, in press}
4: 
5: \shorttitle{Predictions of planetary migration theory}
6: \shortauthors{Armitage}
7: 
8: \begin{document}
9: 
10: \title{Massive planet migration: Theoretical predictions \\ and comparison with observations}
11: 
12: \author{Philip J. Armitage\altaffilmark{1,2}}
13: \altaffiltext{1}{JILA, Campus Box 440, University of Colorado, Boulder CO 80309; 
14: pja@jilau1.colorado.edu}
15: \altaffiltext{2}{Department of Astrophysical and Planetary Sciences, University of Colorado, Boulder CO 80309}
16: 
17: \begin{abstract}
18: We quantify the utility of large radial velocity surveys for constraining theoretical models of Type~II 
19: migration and protoplanetary disk physics. We describe a theoretical model for the 
20: expected radial distribution of extrasolar planets that combines 
21: an analytic description of migration with an empirically calibrated disk 
22: model. The disk model includes viscous evolution and mass loss via photoevaporation. 
23: Comparing the predicted distribution to a uniformly selected subsample of planets from the 
24: Lick / Keck / AAT planet search programs, we find that a simple model in 
25: which planets form in the outer disk at a uniform rate, migrate  
26: inward according to a standard Type~II prescription, and become stranded when 
27: the gas disk is dispersed, is consistent with 
28: the radial distribution of planets for orbital radii $0.1~{\rm AU} \leq a < 2.5~{\rm AU}$ 
29: and planet masses $M_p > 1.65 \ M_J$. Some variant models are disfavored by 
30: existing data, but the significance is limited ($\sim$95\%) due to the small 
31: sample of planets suitable for statistical analysis. We show that the 
32: favored model predicts that the planetary mass function should be almost 
33: independent of orbital radius at distances where migration dominates the 
34: massive planet population. We also study how the radial distribution 
35: of planets depends upon the adopted disk model. We find that the distribution 
36: can constrain not only changes in the power-law index of the disk viscosity, but 
37: also sharp jumps in the efficiency of angular momentum transport that 
38: might occur at small radii. 
39: \end{abstract}
40: 
41: \keywords{solar system: formation --- planets and satellites: formation --- 
42: planetary systems: formation --- planetary systems: protoplanetary disks --- 
43: accretion, accretion disks}
44: 
45: \section{Introduction}
46: Radial velocity and transit searches for extrasolar planets have detected in 
47: excess of 170 low-mass companions around nearby, mostly Solar-type, stars 
48: \citep{butler06}. These detections, which result from radial velocity surveys 
49: targeting a few~$\times 10^3$ stars, have allowed for an  
50: initial determination of the distribution of massive extrasolar 
51: planets with mass, orbital radius, eccentricity, 
52: and stellar metallicity \citep{santos04,marcy05,fischer05,santos05}. The 
53: statistical (and, hopefully, systematic) errors on these determinations  
54: will improve as ongoing surveys press to larger orbital radii and 
55: smaller planet masses. Substantially larger radial velocity surveys  
56: of $10^5$ to $10^6$ stars, with precision in the $\sim$10~ms$^{-1}$ range, 
57: are technically possible over the next decade \citep{ge06}. Given this rapid 
58: observational progress it is of interest to ask how much information -- 
59: about planet formation, planet migration, and the protoplanetary disk -- 
60: is retained in the statistical properties of extrasolar planets to be 
61: potentially tapped via an expansion of existing planet samples. Put more 
62: bluntly, is it worth obtaining much larger samples of planets with properties similar 
63: to those already known, or does the primary scientific interest for future 
64: surveys lie in exploring entirely new regimes of parameter space? 
65: 
66: At sufficiently small orbital radii, massive extrasolar planets very probably 
67: migrated inward from formation sites further out rather than forming 
68: in situ \citep{lin96,trilling98,bodenheimer00}. There remains some uncertainty in 
69: quantifying `sufficiently small', but it seems likely that massive 
70: planet formation is most common outside the snow line \citep{hayashi85}. 
71: Protoplanetary disk models clearly show that the radius of the snow line 
72: changes dramatically with time as the disk evolves \citep{garaud07}, so 
73: to quote a single radius is potentially misleading. However, for a Solar 
74: type star, the apparent presence of hydrated minerals in Solar System 
75: asteroids allows an empirical determination of the location of the 
76: snow line at a radius of around 2.7~AU \citep{morbidelli00}. This 
77: suggests that most of the extrasolar planets currently known, which orbit within  
78: a few~AU of their host stars, derive their properties largely via 
79: migration. For massive planets, the appropriate regime of migration 
80: is thought theoretically to be the Type~II regime --- which involves migration within a gap in 
81: the protoplanetary disk \citep{goldreich80,lin86} --- rather than the 
82: gap-less Type~I regime appropriate to Earth mass planets and giant planet cores 
83: \citep{ward96}. Since direct observational evidence of migration is 
84: currently lacking, the only tests possible of this theory come 
85: from statistical comparison with the observed properties of extrasolar 
86: planetary systems. Indeed, prior work along these lines by \cite{armitage02}, 
87: \cite{trilling02} and \citet{ida04} has shown that the distribution 
88: of observed planets in orbital radius (and, in the case of the 
89: \cite{ida04} study, planetary mass) is broadly consistent with theoretical 
90: expectations based on disk migration within an evolving protoplanetary 
91: disk.
92: 
93: In this paper, we develop more refined predictions for the radial distribution 
94: of massive planets based on a simple analytic model for Type~II 
95: migration. Our main goal is to determine, at least in principle, what might be learned 
96: from comparisons of large planet samples with theoretical models. In \S2 we describe the 
97: adopted model for the protoplanetary disk, and how migration 
98: of massive planets within the disk is treated. In \S3 we 
99: compute the predicted distribution of planets in orbital 
100: radius. We compare the predictions to the observed distribution 
101: of planets in the \cite{fischer05} sample, which has previously 
102: been used to study the dependence of planet frequency on host 
103: metallicity. This sample has a clearly 
104: specified selection limit, which allows for a 
105: reliable statistical comparison between the model and observations. 
106: In \S4 and \S5 we investigate the extent to which migration 
107: leads to a radial variation in the exoplanet mass function, 
108: and how sensitive it is to structure within the protoplanetary 
109: disk. These Sections are primarily forward-looking, since 
110: existing data are too limited to support or refute the 
111: theoretical model. We conclude in \S6 with some discussion of 
112: the results.
113: 
114: \section{Type II migration within the protoplanetary disk}
115: We consider a model in which massive planets form within an 
116: evolving protoplanetary disk at radii beyond the snow line, 
117: and migrate inward within a gap (`Type~II' orbital migration). 
118: Migration slows and eventually ceases as the gas disk is dispersed. We assume that 
119: photoevaporation causes disk dispersal. The resulting distribution 
120: of massive planets in orbital radius then depends upon the 
121: disk model (which is reasonably tightly constrained by observations); 
122: the migration rate (which is known reasonably well theoretically); 
123: and the rate of planet formation in the disk as a function of 
124: time. The latter can in principle be predicted from a model of massive 
125: planet formation, but here we treat it as a free function.
126: 
127: The surface density $\Sigma$ of a protoplanetary disk that 
128: evolves under the combined action of an effective kinematic viscosity 
129: $\nu$, and mass loss per unit area $\dot{\Sigma}_{\rm wind} (r)$, is described 
130: by \citep{pringle81},
131: \begin{equation} 
132:  \frac{\partial \Sigma}{\partial t} = \frac{3}{r} 
133:  \frac{\partial}{\partial r} \left[ 
134:  r^{1/2} \frac{\partial}{\partial r} \left( \nu 
135:  \Sigma r^{1/2} \right) \right] - \dot{\Sigma}_{\rm wind} (r),
136: \label{eq_diffusion}
137: \end{equation} 
138: provided that the mass lost in the wind has the same specific 
139: angular momentum as the disk at the launch point. Approximating 
140: the angular momentum transport in the disk as a time-independent 
141: kinematic viscosity that is a power-law in radius,
142: \begin{equation} 
143:  \nu = \nu_0 r^\gamma
144: \label{eq_viscous} 
145: \end{equation}
146: equation (\ref{eq_diffusion}) admits a compact self-similar solution 
147: if the mass loss rate is negligible \citep{lyndenbell74,hartmann98}. 
148: In this solution the surface density 
149: evolves according to,
150: \begin{equation} 
151:  \Sigma (R,T) = \frac{C}{3 \pi \nu_{\rm scale} R^\gamma} 
152:  T^{-(5/2 - \gamma)/(2-\gamma)} 
153:  \exp[{-R^{(2-\gamma)} / T}]
154: \label{eq_similar} 
155: \end{equation}  
156: where the scaled variables $R$ and $T$ are defined via a 
157: fiducial radius $r_{\rm scale}$,
158: \begin{eqnarray}
159:  R & \equiv & \frac{r}{r_{\rm scale}} \\
160:  \nu_{\rm scale} & \equiv & \nu(r_{\rm scale}) \\
161:  T & \equiv & \frac{t}{t_{\rm scale}} + 1 \\
162:  t_{\rm scale} & = & \frac{1}{3(2-\gamma)^2} 
163:  \frac{r_{\rm scale}^2}{\nu_{\rm scale}}.
164: \end{eqnarray}
165: The constant $C$ is the mass accretion rate at $t=0$ as 
166: $r \rightarrow 0$. Other quantities, such as the mass accretion rate 
167: $\dot{M}$ and radial velocity $v_r$ in the disk, can be derived straightforwardly 
168: using these expressions.
169: 
170: \begin{figure}
171: \plotone{f1.ps}
172: \caption{The time dependence of the accretion rate on to the central star plotted for 
173: different disk models. The three dashed curves in the upper panel show the predicted evolution in 
174: self-similar models with different power-law exponents for the disk viscosity: 
175: $\gamma=0.5$, $\gamma=1.0$ and $\gamma=1.5$. The solid curves show the 
176: result including photoevaporation, which is here modeled using the solution 
177: of \cite{ruden04} assuming a disk dispersal time of 6~Myr. The analytic 
178: cut-off is derived assuming $\gamma=1.0$, but should be a good approximation 
179: for general power-law viscosity profiles. The lower panel shows the surface 
180: density evolution at a disk radius of 1~AU (the $\gamma=1.5$ model is the 
181: uppermost curve).}
182: \label{f1}
183: \end{figure}
184:  
185: We constrain the parameters of the similarity solution 
186: (equation \ref{eq_similar}) to be consistent with the observational 
187: analysis of the time dependence of T~Tauri accretion rates presented 
188: by \cite{muzerolle00}. We fix all models to have an initial accretion 
189: rate $\dot{M} (t=0) = 3 \times 10^{-7} \ M_\odot {\rm yr}^{-1}$, 
190: which decays to $\dot{M} = 3 \times 10^{-10} \ M_\odot {\rm yr}^{-1}$ 
191: in $10^7$~yr if the disk lives that long before being dispersed. We also 
192: fix $r_{\rm scale} = 10 \ {\rm AU}$. Table~1 shows the 
193: values of viscosity normalization constant $\nu_0$, 
194: and initial disk mass $M_{\rm disk}$, that meet these constraints 
195: for different values of power-law index $\gamma$. Not surprisingly, 
196: since the basic observational constraints have been known for 
197: some time, the disk models we favor have parameters similar to 
198: those considered in many previous studies \citep{hartmann98,armitage03,
199: alexander06}. Figure~\ref{f1} 
200: shows the resulting evolution of the accretion rate for these  
201: models.
202:  
203: \begin{table*}
204: \center{
205: \begin{tabular}{ccc}
206: \hline \hline
207: Power law index of viscosity $\gamma$ & Viscosity normalization $\nu_0$ & Initial disk mass $M_{\rm disk}$ \\ \hline
208: 0.5 & $1.52 \times 10^8 \ {\rm cm}^{3/2}{\rm s}^{-1}$ & $0.057 \ M_\odot$ \\
209: 1.0 & $15.62 \ {\rm cm}{\rm s}^{-1}$ & $0.051 \ M_\odot$ \\
210: 1.5 & $1.58 \times 10^{-6} \ {\rm cm}^{1/2}{\rm s}^{-1}$ & $0.058 \ M_\odot$ \\ 
211: \hline
212: \label{table1}
213: \end{tabular}}
214: \caption{Properties of self-similar disk models that meet the 
215: observational constraints \\ on the evolution of the stellar 
216: accretion rate discussed in the text}
217: \end{table*} 
218:  
219: The power-law decline in the late-time accretion rate implied by the 
220: similarity solution does not yield the sharp transition between 
221: accreting Classical T~Tauri stars and non-accreting Weak-Lined T~Tauri 
222: stars that is observed \citep{simon95,wolk96}. It is plausible that this 
223: transition is driven by photoevaporation \citep{bally82} from the outer disk, 
224: which acts to starve the inner disk (thereby allowing it to drain viscously 
225: onto the star on a short time scale) once the accretion rate 
226: becomes comparable to the wind mass loss rate \citep{clarke01}. 
227: If photoevaporation is  
228: driven by irradiation from the central star, then the 
229: simplest analytic models predict that the mass loss 
230: rate per unit area scales as \citep{hollenbach94},
231: \begin{eqnarray}
232:  \dot{\Sigma}_{\rm wind} \propto r^{-5/2}
233: \label{eq_massloss} 
234: \end{eqnarray}
235: exterior to some critical radius $r_{\rm in}$, with zero 
236: mass loss from smaller radii. The critical radius is given 
237: to order of magnitude by the radius where the sound speed 
238: in photoionized gas ($T \simeq 10^4$) first exceeds the local 
239: escape speed. This is a few AU for a Solar mass star. The 
240: normalization of the mass loss rate depends upon the square 
241: root of the ionizing flux $\Phi$, which is hard to measure 
242: accurately but whose value can be constrained observationally 
243: \citep{alexander05,pascucci07}. Much 
244: more detailed numerical models of photoevaporation are now 
245: available \citep{font04,alexander06}, but the additional 
246: complexity they involve is not warranted for this application.
247: Accordingly, we assume a time-independent mass loss rate of the form given 
248: by equation (\ref{eq_massloss}), with $r_{\rm in} = 5 \ {\rm AU}$, and 
249: zero mass loss at smaller radii.
250: 
251: Once photoevaporation is included (in this simplified form), it is still 
252: possible to derive a Green's function solution to equation (\ref{eq_diffusion}) 
253: \citep{ruden04}, but there is no compact form for the evolution of 
254: $\Sigma(r,t)$ analogous to equation (\ref{eq_similar}). At late times 
255: (i.e. after the disk has evolved for several viscous times) and small 
256: radii ($r < r_{\rm in}$), however, it is possible to derive an 
257: expression for the reduction in the inner accretion rate and 
258: surface density caused by the wind. Defining $t_0$ as the time 
259: at which the inner accretion rate falls to zero as a consequence of 
260: the mass loss from the outer disk, \cite{ruden04} finds that the 
261: time dependence of the accretion rate at small radii follows,
262: \begin{equation}
263:  \dot{M} = \left[ 1 - \left( \frac{t}{t_0} \right)^{3/2} \right]
264:  \dot{M}_{\rm self-similar}, 
265: \end{equation} 
266: where $\dot{M}_{\rm self-similar}$ is the accretion rate evolution 
267: predicted by the self-similar model in the absence of any mass loss. 
268: The suppression of accretion implied by the term in parenthesis is 
269: derived for $\gamma=1$, but should also be approximately valid for 
270: other values of the power law viscosity index. Using numerical 
271: solutions to equation (\ref{eq_diffusion}), we have verified that 
272: the \cite{ruden04} formula provides a good description of the 
273: evolution of the accretion rate and inner surface density during 
274: the transition, and hence we apply the same cutoff to generate 
275: the surface density evolution including wind loss $\Sigma(r,t)$ from 
276: the self-similar prediction (equation~\ref{eq_similar}). Based on 
277: observations of disk frequency in young clusters \citep{haisch01}, 
278: we take $t_0 = 6 \ {\rm Myr}$, which yields an 
279: accretion rate evolution shown as the solid curves in Figure~\ref{f1}.
280: As has been emphasized previously \citep{clarke01,armitage03},   
281: to a good approximation the disk evolution proceeds as if there is no mass 
282: loss while $\dot{M} \gg \dot{M}_{\rm wind}$, and is then dispersed rapidly 
283: once mass loss becomes significant. Figure~\ref{f1} also shows the 
284: evolution of the inner (1~AU) disk surface density. Irrespective of 
285: the value of $\gamma$, all of the disk models considered have an 
286: {\em initial} surface density that is substantially in excess of 
287: the minimum mass Solar Nebula reference value \citep{weidenschilling77}, 
288: but this drops rapidly as the disk evolves. By the time that massive 
289: planets migrate through this region (after several Myr of evolution) 
290: the surface density has fallen by at least an order of magnitude. The 
291: import of this is that the local disk mass during migration is 
292: typically smaller than the mass of giant planets, a regime which 
293: leads to significantly slower inspiral.
294: 
295: Sufficiently massive planets are able to open a gap within the gaseous 
296: protoplanetary disk and, having opened a gap, migrate inward\footnote{Here, 
297: we consider exclusively planets that form close enough in that the sense 
298: of gas flow and migration is inward --- planets that form further out 
299: may instead absorb the angular momentum of the inner disk and move 
300: outward \citep{veras04,martin07}, or become stranded outside the annular hole 
301: that is predicted to form as photoevaporation proceeds \citep{clarke01,matsuyama03}.} 
302: in lockstep with the gas via Type~II migration \citep{lin86}. For a planet at 
303: orbital radius $r_p$, the gravitational torques exerted by the planet on the 
304: surrounding disk are strong enough to maintain a gap provided that 
305: the mass ratio $q = M_p / M_*$ satisfies \citep{takeuchi96},
306: \begin{equation}
307:  q \gtrsim \left( \frac{c_s}{r_p \Omega_p} \right)^2 \alpha^{1/2}
308: \end{equation}
309: where $c_s$ and $\Omega_p$ are the sound speed and angular velocity 
310: in the disk at the radius of the planet, and $\alpha$ is the 
311: dimensionless Shakura-Sunyaev viscosity parameter \citep{shakura73}. 
312: Noting that the thickness of the disk $h \simeq c_s / \Omega_p$, 
313: we find that the $\gamma=1$ disk model specified in Table~1 
314: would correspond to an equivalent $\alpha \simeq 5 \times 10^{-3}$ 
315: (matching at 5~AU around a Solar mass star assuming that the 
316: disk has $h/r = 0.05$). The viscous gap opening criteria is then 
317: satisfied for planets of mass exceeding about 0.2~$M_J$. The 
318: thermal gap opening condition is satisfied at a similar mass 
319: \citep{bate03}. In this paper, we exclusively consider planets 
320: with masses of $M_J$ and above, which should accordingly be safely 
321: in the gap-opening regime of parameter space. 
322: 
323: \begin{figure}
324: \plotone{f2.ps}
325: \caption{The critical accretion rate above which a 1~$M_J$ planet migrates as a 
326: test particle is plotted as a function of radius for the different disk models.}
327: \label{f2}
328: \end{figure}
329: 
330: To compute the Type~II migration rate within the disk model 
331: specified above, we draw on the results of \cite{syer95}. 
332: These authors noted that the rate of migration, which is 
333: sometimes assumed to equal the radial velocity that gas in 
334: the disk {\em would have in the absence of a planet}, is 
335: suppressed once the planet mass exceeds a local estimate 
336: of the disk mass. Specifically, they defined a measure of 
337: the importance of the disk relative to the planet,
338: \begin{equation}
339:  B \equiv \frac{4 \pi \Sigma r_p^2}{M_p}
340: \end{equation}
341: which is small if the planet dominates the angular momentum 
342: budget of the planet + disk system, and large otherwise. The 
343: Type~II migration rate is then given for $B \geq 1$ (the disk 
344: dominated case) by,
345: \begin{equation}
346:  \dot{r}_p = v_r 
347: \label{eq_migrate1} 
348: \end{equation}
349: while for $B < 1$ (the planet dominated limit),
350: \begin{equation}
351:  \dot{r}_p = B^{1/2} v_r
352: \label{eq_migrate2}
353: \end{equation}
354: where we have assumed (consistent with equation~\ref{eq_viscous}) 
355: that the efficiency of angular momentum transport is independent 
356: of the surface density. For our disk models, Figure~\ref{f2} shows the critical accretion rate 
357: below which $B < 1$ for a Jupiter mass planet. In all these models, slowdown 
358: of the migration rate occurs due to the inertia of the planet 
359: for accretion rates substantially greater than those at 
360: which photoevaporation starts to influence the disk evolution. 
361: The effect on the nominal migration time scale,
362: \begin{equation} 
363:  t_{\rm migrate} \equiv \frac{r_p}{\left| \dot{r}_p \right|}
364: \end{equation}
365: is shown in Figure~\ref{f3}. The $B < 1$ limit is applicable 
366: at all radii of interest in a disk with $\dot{M} = 10^{-10} \ 
367: M_\odot {\rm yr}^{-1}$, and within radii less than a few 
368: AU (depending on the planet mass) in a disk with 
369: $\dot{M} = 10^{-8} \ M_\odot {\rm yr}^{-1}$. 
370: 
371: \begin{figure}
372: \plotone{f3.ps}
373: \caption{The predicted migration time scale as a function of radius in the 
374: $\gamma=1$ disk model. The solid lines show the migration time scale for 
375: planets of masses 0.5~$M_J$, 1~$M_J$, 2~$M_J$ and 4~$M_J$ when the disk 
376: accretion rate is $10^{-8} \ M_\odot {\rm yr}^{-1}$ (more massive planets 
377: have longer migration time scales at small orbital radii). The dashed lines 
378: show results for the same mass planets in a disk with $\dot{M} = 10^{-10} 
379: \ M_\odot {\rm yr}^{-1}$.}
380: \label{f3}
381: \end{figure}
382: 
383: Once planets exceed the critical gap opening mass by a significant 
384: factor, the rate of accretion onto the planet across the gap 
385: drops rapidly \citep{lubow99}. Our goal is to compare theoretical 
386: predictions for the planet distribution with a complete subsample 
387: of extrasolar planets, which requires a cut at approximately 
388: 1.5~$M_J$. Most of the planets that survive this cut have masses 
389: substantially greater than the gap-opening mass, so {\em as a 
390: first approximation} it seems reasonable to ignore the possibility 
391: of mass accretion during Type~II migration, and assume that 
392: negligible mass is accreted across the range of radii (interior 
393: to 2.5~AU) over which we make the comparison. The results of 
394: \cite{bate03}, which show that planets can grow rapidly to 
395: masses beyond that of Jupiter while suffering little migration, 
396: support this approximation. It is then easy, using 
397: equations (\ref{eq_migrate1}) and (\ref{eq_migrate2}), to calculate 
398: the final (following disk dispersal) orbital radius of a planet 
399: that forms with mass $M_p$ at radius $a_{\rm form}$ and time $t_{\rm form}$ 
400: in one of the disk models specified in Table~1. The resulting mapping 
401: between the formation conditions and the final state is the basic 
402: input needed to make a prediction of the resulting planet distribution.
403: 
404: The above model represents our attempt to define a `best-guess' description 
405: of Type~II migration within the protoplanetary disk. One may note that 
406: in the planet-dominated $B < 1$ limit, the predicted rate of Type~II 
407: migration is reduced by the square-root of $B$ rather than the full ratio 
408: of the local disk mass to the planet mass. We will refer to this as 
409: {\em partial suppression} of the migration rate. The rate is not 
410: fully suppressed because gas accumulates close to the tidal barrier 
411: as the disk / planet system evolves, increasing the torque beyond 
412: the value that would occur in an unperturbed disk \citep{pringle91}. 
413: In practice, however, some gas may overflow the gap to be either 
414: accreted by the planet or to flow into an inner disk interior 
415: to the gap \citep{lubow06}. Mindful of this, we have also 
416: computed models in which the migration rate in the planet dominated 
417: regime is {\em fully suppressed}, so that,
418: \begin{equation}
419:  \dot{r}_p = B v_r.
420: \end{equation}
421: The difference between the partially and fully suppressed calculations 
422: provides an indication as to how sensitive the results are to uncertainties 
423: in the treatment of massive planet migration.
424: 
425: \section{Predicted distribution of planets in orbital radius}
426: Using the analytic disk evolution and migration models described in 
427: \S2, we compute the radius $a_{\rm final}$ at which massive planets become 
428: stranded as a function of the time $t_{\rm form}$ at which they form 
429: within the protoplanetary disk. The parameters of the model are the planet mass, 
430: the power-law index of the disk viscosity, the formation radius, 
431: and whether the migration rate is partially or fully suppressed. 
432: Illustrative results for a formation radius of 5~AU and a planet mass 
433: of 1~$M_J$ are shown in Figure~\ref{f4}. As is well known, planets 
434: that form at earlier epochs migrate to smaller final radii, and  
435: as the orbital radii decrease the window of allowed formation 
436: times also narrows \citep{trilling02,armitage02}. For the favored model, in which migration is only 
437: partially suppressed, all surviving planets must form during the 
438: last 1-1.5~Myr of the disk lifetime. This is not so short a window as to 
439: imply that the planet formation process must be worryingly lossy, but it 
440: does imply that the final masses of giant planets might well reflect details of the disk 
441: dispersal process \citep{shu93}. Planets can form and survive across 
442: a larger fraction of the disk lifetime if instead migration is 
443: fully suppressed. There are also significant differences in the 
444: outcome that depend upon the adopted disk model. These arise 
445: because of the differing profiles of the radial velocity as a 
446: function of radius, but they are less significant than differences in the 
447: treatment of migration.
448: 
449: \begin{figure}
450: \plotone{f4.ps}
451: \caption{Predicted final radii for 1~$M_J$ planets as a function of the time at 
452: which they formed at 5~AU in the protoplanetary disk. The three closely spaced 
453: curves on the right-hand-side of the Figure show results for disks with 
454: $\gamma=0.5$ (long-dashed lines), $\gamma=1$ (solid lines) and $\gamma=1.5$ 
455: (short-dashed lines) respectively, assuming that migration is suppressed at 
456: small radii according to the \cite{syer95} model. The left-hand curves show 
457: results if, instead, migration is fully suppressed by the ratio of the local 
458: disk mass to the planet mass. Differences between migration treatments are 
459: evidently more significant that differences between the disk models. For the 
460: favored models in which migration is partially suppressed, the survival 
461: time of planets in the disk is in the range of 10-20\% of the disk lifetime.}
462: \label{f4}
463: \end{figure}
464: 
465: To translate these results into a prediction for the orbital distribution 
466: of massive planets, we note that,
467: \begin{equation}
468:  \frac{{\rm d}N_p}{{\rm d}a} = \frac{{\rm d}t_{\rm form}}{{\rm d}a_{\rm final}} \times 
469:  \frac{{\rm d}N_p}{{\rm d}t_{\rm form}}
470: \end{equation}
471: where ${{\rm d}N_p}/{{\rm d}t_{\rm form}}$ is the rate at which planets 
472: of a given mass form in the outer disk. The simplest assumption is that 
473: this rate is a constant, or at least can be approximated as such, over 
474: the interval of time near the end of the disk lifetime during which 
475: planets can form and survive migration without being swept into the star. 
476: Making this assumption, we compute and plot in Figure~\ref{f5} 
477: ${{\rm d}N_p}/{{\rm d}\log a}$ for a variety of models, which differ 
478: in the assumed formation radius, disk model, and migration treatment. 
479: Numerical values for three of the these models, which assume 
480: $a_{\rm form} = 5 {\rm AU}$, partial suppression of the migration 
481: velocity, and $\gamma=0.5$, $\gamma=1$ or $\gamma=1.5$, are tabulated 
482: in Table~2. These three cases roughly bracket the range of predicted 
483: outcomes for all of the models that we have considered. 
484: 
485: \begin{figure}
486: \plotone{f5.ps}
487: \caption{The predicted distribution of massive planets as a function of radius. 
488: All the curves show results for Jupiter mass planets (note that the mass dependence 
489: is sufficiently weak that essentially identical results apply also for higher 
490: masses). The solid curve shows the 
491: results for the favored model, in which planets form at 5~AU at a constant 
492: rate in a $\gamma=1$ disk, and migration is partially suppressed at small 
493: radii. The two dotted curves show the effect of varying the disk viscosity 
494: to $\gamma=0.5$ or $\gamma=1.5$. The long-dashed curves show the effect of 
495: assuming that planets form at 3~AU (the curve furthest to the left at 
496: 2.5~AU) or 10~AU rather than 5~AU. The short-dashed curve is  
497: the predicted distribution in the case in which migration is fully suppressed. 
498: The data points (shown with approximate errors) show the number of observed 
499: extrasolar planets with $1.65 M_J < M_p \sin(i) < 10 M_J$ in a uniform and 
500: complete sample of planets constructed from \cite{fischer05}.}
501: \label{f5}
502: \end{figure}
503: 
504: \begin{table*}
505: \center{
506: \begin{tabular}{cccc}
507: \hline \hline
508: $\log(a / {\rm AU})$ & ${\rm d}N (\gamma=0.5) / {\rm d}\log(a) $ & ${\rm d}N (\gamma=1.0) / {\rm d}\log(a) $ &
509: ${\rm d}N (\gamma=1.5) / {\rm
510: d}\log(a) $ \\ 
511: \hline
512: -1.0 & 0.152 & 0.263 & 0.420 \\
513: -0.9 & 0.184 & 0.298 & 0.423 \\
514: -0.8 & 0.221 & 0.338 & 0.489 \\
515: -0.7 & 0.266 & 0.384 & 0.529 \\
516: -0.6 & 0.320 & 0.436 & 0.573 \\
517: -0.5 & 0.385 & 0.497 & 0.623 \\
518: -0.4 & 0.463 & 0.567 & 0.679 \\
519: -0.3 & 0.559 & 0.650 & 0.743 \\
520: -0.2 & 0.676 & 0.747 & 0.816 \\
521: -0.1 & 0.820 & 0.863 & 0.899 \\
522: 0.0 & 1.000 & 1.000 & 1.000 \\
523: 0.1 & 1.226 & 1.170 & 1.117 \\
524: 0.2 & 1.518 & 1.380 & 1.261 \\
525: 0.3 & 1.912 & 1.656 & 1.448 \\
526: 0.4 & 2.458 & 2.035 & 1.705 \\
527: 0.5 & 3.283 & 2.592 & 2.080 \\
528: \hline
529: \label{table2}
530: \end{tabular}}
531: \caption{The predicted number of planets per logarithmic interval in 
532: semi-major axis for models in which planet formation occurs at 5~AU and 
533: the migration rate is partially suppressed. Results for three different 
534: values of $\gamma$ are quoted. These results approximately bracket the 
535: shallowest to steepest curves obtained for the models shown in Figure~\ref{f5}. 
536: Results are quoted for 1 Jupiter mass planets, but the mass dependence is 
537: weak -- much weaker than the differences between the different $\gamma$ models. 
538: The numbers have been normalized to unity at 1~AU.}
539: \end{table*}
540: 
541: Although all of the curves have the same general form --- relatively 
542: few planets are predicted to be stranded at very small orbital radii, 
543: with the number increasingly rapidly as $a$ increases --- there are 
544: significant differences which may in principle leave signatures in the 
545: observed distribution of extrasolar planets. Most importantly, the 
546: predicted number of planets per logarithmic interval in semi-major 
547: axis rises sharply as the formation radius is approached. A model in 
548: which planets are assumed to form typically at 10~AU rather than 
549: 5~AU results in a much smaller predicted number of planets at 4~AU, 
550: whereas assuming a typical formation radius of 3~AU (i.e. immediately 
551: outside the snow line) leads to a predicted pile-up of planets as that 
552: radius is approached. 
553: Observational detection of a rapid rise in the number of planets at a 
554: particular orbital radius would then be a signature of a preferred 
555: orbital radius for giant planet formation. Also easily detectable 
556: are the differences between the partially and fully suppressed 
557: migration models. If migration is fully suppressed, then the 
558: number of planets (expressed as ${{\rm d}N_p}/{{\rm d}\log a}$) at 
559: 0.1~AU is predicted to be around half the number at 1~AU, whereas 
560: for partial suppression this ratio is around 0.25. This result is 
561: easily understood -- differences in the treatment of migration are 
562: most important at those small radii, interior to 1~AU, where the disk 
563: mass is lowest. 
564: 
565: \subsection{Sensitivity to the disk dispersal mechanism}
566: Our parameterization of disk evolution sweeps much that is poorly 
567: known about photoevaporation into a single parameter -- the disk 
568: dispersal time $t_0$. Physically, rapid dispersal (small values of 
569: $t_0$ for a given set of disk initial conditions) will occur 
570: for large values of the ionizing flux $\Phi$, which will drive a 
571: stronger photoevaporative flow. Quantitatively, the total mass 
572: loss rate from the disk scales as \citep{hollenbach94},
573: \begin{equation}
574:  \dot{M}_{\rm wind} \simeq 1.5 \times 10^{-10} 
575:  \left( \frac{\Phi}{10^{41} \ {\rm s}} \right)^{1/2} 
576:  \left( \frac{M_*}{M_\odot} \right) ^{1/2} M_\odot {\rm yr}^{-1}
577: \end{equation}
578: where the prefactor has been adjusted from the analytic 
579: value to better match the numerical results of \cite{font04}, and the 
580: fiducial ionizing flux has been chosen to be consistent with observational 
581: estimates \citep{alexander05,pascucci07}, which are, however, subject 
582: to substantial uncertainties.
583: 
584: With our disk models, a wind mass loss rate in the range between 
585: $10^{-10} \ M_\odot {\rm yr}^{-1}$ and $10^{-9} \ M_\odot {\rm yr}^{-1}$ 
586: yields the dispersal time of 6~Myr that is our baseline assumption. 
587: Our fiducial parameters are therefore consistent with standard 
588: photoevaporation models. A consequence of this, however, is that 
589: the {\em disk mass} at the epoch when surviving planets form is 
590: quite small. Figure~\ref{f5b} shows the gas disk mass at the moment 
591: planets form as a function of the final planet semi-major axis. For 
592: the models we have considered, the typical disk masses are only a 
593: few Jupiter masses if migration is partially suppressed (in the 
594: fully suppressed case, on the other hand, surviving planets form 
595: substantially earlier when there is plenty of disk gas remaining). 
596: These small masses directly reflect the efficiency of the migration process -- 
597: relatively modest amounts of gas are able to drive substantial 
598: migration. However, they do pose a possible consistency problem: 
599: planets that form in very low mass disks evidently perturb the 
600: disk structure substantially, and this might affect the final 
601: planetary distribution.
602: 
603: \begin{figure}
604: \plotone{f6.ps}
605: \caption{The predicted distribution of massive planets as a function of radius 
606: (upper panel) is plotted together with the disk mass evaluated at the formation 
607: epoch (lower panel). The solid curve shows the results for the favored model, in which planets 
608: form at 5~AU at a constant rate in a $\gamma=1$ disk, and migration is partially 
609: suppressed at small radii. The short-dashed curve -- almost indistinguishable in the 
610: upper panel -- shows the effect of reducing the disk dispersal time $t_0$ from 
611: 6~Myr to 3~Myr. The long-dashed curve shows the results of assuming an instantaneous 
612: dispersal of the disk at 6~Myr.}
613: \label{f5b}
614: \end{figure}
615: 
616: To explore this possibility, we considered two options. First, 
617: within the formalism developed here, we have investigated the 
618: effect of reducing $t_0$ so that surviving planets form earlier 
619: when the disk mass is higher. As shown in Figure~\ref{f5b}, 
620: a model computed with $t_0 = 3 \ {\rm Myr}$ yields an almost 
621: identical planet distribution. Second, we have compared the 
622: results for $\gamma=1.5$ with the numerical models presented 
623: in \cite{armitage02}. In the numerical models, planet masses 
624: are restricted to be no larger than a local estimate of the 
625: disk mass, and the feedback of a massive planet on the structure 
626: of a very low mass disk is explicitly followed. Reasonably 
627: good agreement is obtained between the numerical and analytic 
628: schemes, with the magnitude of the differences being comparable 
629: to the differences between the curves plotted in Figure~\ref{f5}.
630: 
631: A related question is how important the nature of the disk 
632: dispersal mechanism is for the resulting planet distribution. 
633: Physically, a model in which there is {\em no} mass loss from 
634: the disk exterior to the planet (either via a wind, or via 
635: accretion on to the planet) fails to yield any surviving planets -- 
636: ultimately the disk, no matter how small, absorbs the orbital angular momentum of the 
637: planet and drives it to small radius\footnote{Strictly, this 
638: is only true if the inner disk is subject to a zero-torque 
639: boundary condition. A steady-state solution is possible if 
640: there is instead angular momentum injection at small radii, 
641: for example due to the interaction between the disk and 
642: the stellar magnetosphere \citep{armitage96}.}. However, 
643: many other models in which the disk is dispersed rapidly 
644: do yield sensible distributions, so it is of interest to 
645: assess how sensitive the planet distribution is to the 
646: specifics of disk dispersal. To gauge this, we have considered 
647: an extreme model in which the disk evolves viscously for 
648: 6~Myr without any mass loss, and is then instantaneously 
649: dispersed. Figure~\ref{f5b} depicts the resulting planet 
650: distribution. At small radii (within about 1.5~AU) this 
651: model tracks the fiducial photoevaporative case closely, 
652: but further out the instantaneous dispersal model yields 
653: a much flatter distribution. This reflects the fact that 
654: in the case of photoevaporation, migration of the last 
655: planets to form is limited by the rapid drop in the disk 
656: surface density, and these planets pile up at radii 
657: relatively close to their formation sites. We conclude that 
658: the innermost part of the extrasolar planet distribution 
659: (within roughly an AU) ought to be largely independent of 
660: how the disk is dispersed, but that the distribution 
661: further out does depend on whether photoevaporation or some 
662: other mechanism is at work.
663: 
664: \subsection{Effect of the dispersion in disk properties}
665: The disk models used in this paper have been adjusted to 
666: approximately reproduce the mean lifetimes and accretion rates 
667: of observed disks \citep{haisch01,hartmann98}. This procedure 
668: ignores the fact that, observationally, there is a large 
669: dispersion in the accretion rate at a given age. This 
670: dispersion exceeds that expected from measurement uncertainties in 
671: accretion rates and stellar ages.
672: 
673: The origin of the intrinsic dispersion in disk properties 
674: as a function of system age is not known, and hence it is 
675: impossible to make a blanket statement as to whether consideration 
676: of a mean model is reasonable or not. We can, however, distinguish 
677: some possibilities. One possibility \citep{armitage03} is that the dispersion 
678: in disk properties arises from a dispersion in disk initial 
679: conditions (disk mass, disk angular momentum). Since the 
680: planet distribution depends only upon the disk evolution 
681: close to the transition epoch (independent of the absolute 
682: timing), in this scenario the ensemble distribution averaged over the population 
683: would be expected to be the same as the mean model we have 
684: computed. Of course, giant planet formation would be less 
685: probable in short-lived disks with low initial surface 
686: densities \citep{pollack96}, so in practice most planets 
687: would form in those disks with higher initial surface 
688: densities at radii of a few AU.
689: 
690: A second possibility is that the observed dispersion in 
691: accretion rates arises from changes in individual accretion 
692: rates on timescales intermediate between the disk lifetime 
693: and observable timescales. Apart from thermal instabilities 
694: \citep{bell94}, which are only likely to be relevant for 
695: initial accretion rates higher than those we have considered here, 
696: no obvious candidate instabilities that might yield large fluctuations 
697: in $\dot{M}$ are known, but it remains possible that $\dot{M} (r)$ 
698: might vary rapidly. If such variability extended across the 
699: radial range considered here, it is likely that the migration 
700: history of planets would be altered.
701: 
702: \subsection{Comparison with a uniformly selected dataset}
703: We compare the models to a subsample of extrasolar planets detected 
704: by the Keck / Lick / AAT planet search program. Initially, we proceed 
705: rather conservatively, and define 
706: a subsample that is as complete and unbiased as 
707: possible over specified intervals in planet mass and orbital radii. 
708: Our procedure is as follows:
709: \begin{enumerate}
710: \item
711: We start with a sample of 850 stars of FGK spectral types, targeted 
712: by the Lick / Keck / AAT planet search program, for which 10 or 
713: more radial velocity measurements over a period of at least 4 
714: years are available. This sample is listed in Table~3 of 
715: \cite{fischer05}. For these stars, hypothetical planets that 
716: yield a radial velocity semi-amplitude $K > 30 \ {\rm ms}^{-1}$ 
717: have nearly uniform detectability provided that the orbital 
718: period is less than 4~years. The orbital period restriction 
719: corresponds to a maximum semi-major axis of 2.5~AU, assuming 
720: Solar mass hosts. From this sample, \cite{fischer05} list 46 
721: stars which host known extrasolar planets. In some cases, these are 
722: multiple systems.
723: \item
724: In many cases, additional radial velocity data is available beyond 
725: that used by \cite{fischer05}. We therefore update the orbital 
726: elements quoted by \cite{fischer05} to match those reported by 
727: \cite{butler06}. Typically the changes to the derived masses and 
728: semi-major axes are rather modest (here we ignore eccentricity, 
729: which in some cases is more substantially altered). Including the 
730: multiple planets in some systems, this yields a sample of 59 planets.
731: \item
732: Finally, we define a subsample that includes only planets that are 
733: massive enough to be detectable across the entire range of orbital radii 
734: for which the survey is complete. At 2.5~AU, $K = 30 \ {\rm ms}^{-1}$ 
735: corresponds to a planet mass of $1.65 \ M_J$, so we discard planets 
736: with smaller masses. We also cut the sample at the high mass end, 
737: somewhat arbitrarily, at $10 M_J$. In terms of radial extent, we 
738: include those planets with $0.1 \ {\rm AU} < a < 2.5 \ {\rm AU}$. 
739: This excludes planets with very tight orbits whose dynamics may have 
740: been influenced by tidal effects and / or penetration into the 
741: protostellar magnetosphere.
742: \end{enumerate}
743: It is somewhat striking --- given the large number of extrasolar 
744: planets now known --- how little of the publicly available data 
745: can be used for the sort of statistical comparison we are attempting 
746: here. Demanding both completeness (which restricts the stellar sample 
747: used), and lack of mass bias (which restricts the minimum mass), we are 
748: left with only 22 massive planets. Obviously this small sample size 
749: limits the statistical power to constrain theoretical models. However, 
750: the sample should be free of any {\em systematic} biases, which may 
751: not have been true for earlier analyses that typically used larger 
752: compilations of detected planets. For illustrative purposes, we 
753: divide the planets in our subsample into four logarithmic bins 
754: in semi-major axis (between 0.1~AU and 2.5~AU) and plot the number 
755: of planets in each bin over the theoretical curves in Figure~\ref{f5}. 
756: The observed distribution, in agreement with the theoretical models, 
757: rises rapidly with increasing semi-major axis.
758: 
759: \begin{figure}
760: \plotone{f7.ps}
761: \caption{Comparison of the predicted planet distribution with the observed one. 
762: The solid histogram shows the cumulative radial distribution of massive planets 
763: with $M_p \sin(i) > 1.65 M_J$. The sample is based on that of \cite{fischer05}, 
764: with the planet properties updated using the compilation of \cite{butler06}. The solid 
765: curve is the prediction of the fiducial model for planets of mass $3 M_J$ 
766: formed at 5~AU in a disk with $\gamma=1$ and partial suppression of the migration 
767: velocity. The dashed curve shows the predictions of a model in which $\gamma=1.5$ 
768: but all other parameters remain fixed. This, and other alternate models   
769: which predict flatter distributions, are not supported by the data, though the 
770: distributions cannot be distinguished at much better than the 95\% confidence level.}
771: \label{f6}
772: \end{figure}
773: 
774: Figure~\ref{f6} shows the cumulative distribution of orbital radii 
775: for massive planets in the sample, together with selected theoretical 
776: curves derived from the differential distributions plotted in Figure~\ref{f5}. 
777: The baseline model, in which planets form at a constant rate at relatively 
778: small orbital radii (5~AU) in a protoplanetary disk with $\gamma=1$, is shown as 
779: the solid line. This model is consistent with the available data 
780: (Kolmogorov-Smirnov test $P=0.3$). Of the variant models considered, 
781: those in which migration is fully suppressed, $\gamma > 1$, and / or planets are 
782: considered to typically form at larger radii are disfavored, as these 
783: models predict relatively more planets at sub-AU orbital radii that are 
784: not observed. A model in which $\gamma=1.5$, with the other parameters 
785: remaining unchanged, is shown as the upper  
786: dashed line. This model is disfavored at modest statistical significance 
787: (Kolmogorov-Smirnov test $P=0.06$). Several variant models that we have 
788: considered are disfavored at roughly comparable confidence levels 
789: (95\%), but with such a small observed sample few definitive conclusions 
790: can be drawn.
791: 
792: As noted above, our sample selection procedure is deliberately conservative. 
793: We have also compared the theoretical model to a larger sample, obtained by 
794: taking the entire catalog of extrasolar planets from \cite{butler06} and 
795: retaining those planets with orbital radii $0.1 \ {\rm AU} < a < 2.5 \ {\rm AU}$ 
796: and masses $1.65M_J < M_p \sin(i) < 10 M_J$ (i.e. the same mass and radius 
797: cuts as before, but relaxing the condition that all planets were detected 
798: from one survey with well-known selection properties). The resulting 
799: sample of 53 planets has a radial distribution 
800: that, by eye, looks very similar to the 22 planet sample, and the greater 
801: numbers give more power to discriminate between models. The baseline model 
802: ($\gamma=1$, formation at 5~AU, partially suppressed migration) remains 
803: consistent with the data, while the second model discussed above in 
804: which $\gamma=1.5$ is more clearly inconsistent ($P=3 \times 10^{-3}$). 
805: We do not place much weight on these results, since the compilation of 
806: planets reported in \cite{butler06} is not uniformly selected. However, 
807: they illustrate clearly that only modest expansion of the 
808: sample size --- by around a factor of two --- would allow us (within the 
809: formal confines of the model) to test whether the observed distribution 
810: of massive extrasolar planets is or is not consistent with theoretical 
811: models that differ in their assumptions as to planet formation radius, 
812: migration rate, and disk properties. Since some of these parameters 
813: are partially degenerate, somewhat larger unbiased samples would be 
814: needed to try and pin down empirically a single favored model.
815: 
816: \section{Radial variation of the mass function}
817: At small orbital radii where $B < 1$, the migration velocity is 
818: dependent on the planet mass. In the case in which migration is 
819: partially suppressed, $\dot{r}_p \propto M_p^{-1/2}$, and hence the 
820: fraction of time over which the most massive planets ($10 M_p$) can 
821: form and survive without migrating in to the star is a factor of a few 
822: greater than for Jupiter mass planets. In a steady-state disk, 
823: however, the fractional suppression of the migration velocity as a function 
824: of radius is fixed for each planet mass (at radii where $B < 1$). 
825: Hence, it is approximately true that the 
826: {\em relative} number of planets left stranded at different radii 
827: is independent of the planet mass, and the predicted mass function 
828: of giant planets is independent of radius \citep{armitage02}.
829: 
830: \begin{figure}
831: \plotone{f8.ps}
832: \caption{The magnitude of the predicted variation in the extrasolar planet mass 
833: function with radius in different models. The solid curve shows results from the 
834: fiducial model in which planets form at 5~AU, $\gamma=1$, and the migration 
835: velocity is partially suppressed. The short-dashed curve shows results for $\gamma=1.5$, 
836: with the remaining model parameters unchanged from their fiducial values.  
837: The long dashed curve shows the effect of assuming that a 
838: change of migration regime occurs between high and low mass planets. In all 
839: cases the slope of mass function at small radii has been fixed at $\alpha=1$.}
840: \label{f7}
841: \end{figure}
842: 
843: In more detail, however, migration does lead to some fractionation 
844: of the planet mass function. More massive planets must form earlier 
845: than their less massive counterparts if they are to end up at the 
846: same orbital radii after disk dispersal, and the evolution of the 
847: disk is itself time-dependent (being dominated by viscous evolution 
848: early on, with a change to a steeper decline in the mass accretion 
849: rate later on due to the influence of photoevaporation). To quantify 
850: the extent to which this results in a radial variation of the 
851: mass function we assume that at small orbital radii the mass function 
852: can be written as a power-law,
853: \begin{equation}
854:  \frac{{\rm d}N_p}{{\rm d}M_p} \propto M_p^{-\alpha}
855: \end{equation}
856: and compute the predicted radial variation in $\alpha$. Figure~\ref{f7} 
857: shows the results for two of the models that we have discussed 
858: earlier, including the baseline model. For definiteness, we set $\alpha=1$ 
859: at small radii, though this choice is arbitrary. The predicted variation  
860: in the mass function with orbital radius in these models is non-zero, 
861: but in these and other similar models we have computed in 
862: which the migration properties of low and high mass planets are the same the 
863: magnitude of the change is small --- of the order of $\Delta \alpha = 0.1$ or less. 
864: It is straightforward to show that variations of this magnitude are 
865: undetectable with feasible surveys.  
866: The conclusion is that any detectable change in the mass function with 
867: radius (at small orbital radii where migration rather than mass growth 
868: is the dominant effect) would have to be attributed to effects other 
869: than mass-dependent migration.
870: 
871: There is one exception to this rule. If the migration {\em regime} 
872: for high mass planets is different from that for low mass planets, then 
873: large changes in the mass function with radius result. Such a change in 
874: migration regime is conceivable --- for example it is possible that high mass planets 
875: completely prevent mass flow across the gap, while low mass planets 
876: allow significant mass flow \citep{lubow99}. In this case, one would 
877: expect less perturbation to the disk structure immediately outside the 
878: gap for low mass planets that for high mass planets. If less mass 
879: piles up outside the tidal barrier, the torque will be reduced, 
880: and the migration velocity will go down. An example model of this kind, in which we assume 
881: that migration is fully suppressed for $0.5 M_J$ planets but only 
882: partially suppressed for $2 M_J$ planets, is also plotted in Figure~\ref{f7}. 
883: As expected (given the substantial difference between the relevant curves 
884: in Figure~\ref{f5}), this model displays order unity variations in the 
885: predicted slope of the mass function with radius. 
886: 
887: \begin{figure}
888: \plotone{f9.ps}
889: \caption{The cumulative mass function of known extrasolar planets in the mass 
890: range $1.65 \ M_J < M_p \sin(i) < 10 \ M_J$ is plotted for two intervals in 
891: semi-major axis. The solid histogram shows the distribution for planets 
892: with $0.1 \ {\rm AU} < a < 1.185 \ {\rm AU}$, while the dashed histogram 
893: represents the distribution for $1.185 \ {\rm AU} < a < 2.5 \ {\rm AU}$.  
894: These distributions are not statistically distinguishable.}
895: \label{f7b}
896: \end{figure}
897: 
898: Observationally, an apparent paucity of high mass planets amongst the 
899: hot Jupiters has been noted by several authors \citep{zucker02}. These 
900: planets have presumably undergone interactions with their host stars 
901: (possibly including tidal circularization, mass loss, or stopping 
902: of migration due to entry into the stellar magnetosphere), any one of 
903: which could result in either mass loss or a mass-dependence to the 
904: planet's survival probability. At larger radii, however, there is no 
905: strong evidence for any variation in the mass function \citep{marcy05}. 
906: To illustrate this with current data, we consider the entire sample of 
907: known extrasolar planets \citep{butler06} with masses in the range 
908: $1.65M_J < M_p \sin(i) < 10 M_J$, and orbital radii between 
909: $0.1 \ {\rm AU} < a < 2.5 \ {\rm AU}$ (similar conclusions follow 
910: from analyses of smaller, more strictly selected samples). We divide this sample 
911: into `inner' and `outer' subsets, with equal numbers of planets 
912: in each. The dividing line between the subsets falls at a semi-major 
913: axis of 1.185~AU. Figure~\ref{f7b} shows the cumulative distributions  
914: of $M_p \sin(i)$ for these subsamples. By eye, and statistically, no 
915: significant differences are seen. This is consistent with the 
916: predictions of the simple migration model we have developed here, 
917: and probably already rules out the large changes in the mass function 
918: with radius predicted in the case where high and low mass planets 
919: migrate in a qualitatively distinct manner.
920: 
921: \section{Sensitivity to structure in the protoplanetary disk}
922: The structure of the protoplanetary disk at radii of the order of 
923: 1~AU is uncertain theoretically, primarily as a consequence of the 
924: low ionization fraction which sows doubt as to the efficiency of 
925: angular momentum transport driven by the magnetorotational instability 
926: \citep{balbus98,gammie96,turner07}. Moreover, most existing observational constraints 
927: are either explicitly \citep{wilner00} or implicitly \citep{hartmann98,armitage03} 
928: sensitive only to large radii in the disk. Not only is the slope of the 
929: steady-state surface density profile at small radii ($\Sigma \propto r^{-\gamma}$) 
930: poorly known, but there could in principle be discontinuities in 
931: $\Sigma$ due to abrupt changes in the angular momentum transport 
932: efficiency \citep{gammie96}. We have looked at how both of these 
933: possibilities affect the resulting distribution of planets.
934: 
935: \begin{figure}
936: \plotone{f10.ps}
937: \caption{Illustration of the effect of discontinuities in the efficiency of disk 
938: angular momentum transport on the predicted planet distribution. In this, not 
939: very realistic example, the viscosity is assumed to drop by a factor of 10 
940: inside 1~AU (upper curve), leading to a large accumulation of planets at small 
941: orbital radii. As in Figure~6, the histogram shows the observed planet 
942: distribution and the lower curve the prediction of the fiducial migration model.}
943: \label{f8}
944: \end{figure}
945: 
946: Holding other parameters of the model (primarily the planet formation 
947: radius) fixed, the resulting differential distributions of massive 
948: planets for different $\gamma$ are plotted in Figure~\ref{f5}. Here 
949: we have assumed a smooth surface density profile, which is fully 
950: characterized by the value of $\gamma$ which we vary between 
951: $\gamma=0.5$ and $\gamma=1.5$. The differences between these 
952: models at sub-AU radii are quite significant. If we consider the 
953: relative number of planets at 1~AU and 0.1~AU, the ratio varies 
954: from 2.4 ($\gamma=1.5$) to 3.8 ($\gamma=1$) to 6.6 ($\gamma=0.5$). 
955: The unbiased subsample defined earlier includes only a handful 
956: of planets at sub-AU radii, so comparison with existing data is not possible, 
957: but these differences are large enough that only a small unbiased 
958: sample of planets at these small radii would be needed to see 
959: variations of this magnitude. Changing the assumed planet formation 
960: radius has only a small influence on the distribution at sub-AU 
961: scales, so the influence of different $\gamma$ is in principle 
962: separable from other unknown parameters of the model.
963: 
964: Abrupt changes in the surface density profile --- for example 
965: due to opacity transitions within the disk or rapid changes in the 
966: efficiency of angular momentum transport with radius, leave an 
967: even more distinctive fingerprint in the radial distribution of 
968: planets, provided only that migration is not fully suppressed,  
969: so that $\dot{r}_p$ is not linear in $B$. Such changes lead to 
970: a jump in the differential distribution of planets at the radius 
971: where the discontinuity occurs, and correspond to a change of 
972: slope in the cumulative distribution. As an example, Figure~\ref{f8} 
973: shows the predicted cumulative distribution if the viscosity is assumed to 
974: drop by a factor of 10 within 1~AU. Only rather unrealistic toy models 
975: of this kind (in which the jump is both large and situated squarely 
976: in the middle of the accessible radial range) can be ruled out using 
977: existing data, but larger data sets could potentially constrain 
978: discontinuities in the disk physics quite well.
979: 
980: \section{Conclusions}
981: In this paper we have investigated the predictions of a rather 
982: simple, almost entirely analytic, model of giant planet formation 
983: and migration for the radial distribution of massive extrasolar 
984: planets. Our main results are:
985: \begin{enumerate}
986: \item
987: The distribution of massive planets with radius, inside the 
988: snow line beyond which planets are assumed to form, preserves 
989: information about the typical location of planet formation and 
990: the structure of the protoplanetary disk at small radii. Statistical 
991: analyses of large, uniformly selected samples of massive planets  
992: could therefore in principle be used to recover information about planet 
993: formation and disk physics.
994: \item
995: Abrupt changes in the disk properties with radius, which may occur 
996: at small orbital radii, leave the most distinctive signature in the 
997: resulting planet distribution (a corresponding discontinuity in 
998: ${\rm d}N_p / {\rm d}\log a$). Constraints on smooth surface density 
999: profiles at AU scales are also possible, but are likely to harder 
1000: to obtain and more ambiguous.
1001: \item
1002: The mass function of extrasolar planets is not predicted to vary 
1003: detectably with radius as a consequence of migration, at least in the 
1004: simplest models.
1005: \item
1006: Existing data remains compatible with a surprisingly simple model, 
1007: proposed previously \citep{armitage02}, in which massive planets form beyond the snow line 
1008: at a constant rate, and migrate inward through a smooth protoplanetary disk before 
1009: becoming stranded when the gas disk is dispersed due to photoevaporation. 
1010: However, the sample of planets suitable for a statistical comparison 
1011: with the theory is limited, and as a result a wide class of 
1012: alternative models remain viable or can only be ruled out with 
1013: limited significance.
1014: \end{enumerate}
1015: Without a doubt, the simple theoretical model we have explored here is 
1016: too simple. On a technical level, it would be desirable to extend the 
1017: model to treat properly the case when planets form within a disk 
1018: whose mass is only a small multiple of the planet mass, as this 
1019: situation is a common outcome of the scenarios we favor. The model 
1020: also ignores some effects, such as mass flow across the gap and 
1021: accretion onto the planet, that are almost certainly important 
1022: \citep{lubow99,lubow06}, along with others, such as planet-planet 
1023: scattering which has the side-effect of altering the orbital 
1024: elements of surviving planets \citep{ford01}, that could be 
1025: significant. As such, the current agreement between our 
1026: predictions and the observed distribution of extrasolar planets speaks 
1027: more to the paucity of the data than to the validity of the theoretical model. 
1028: Taking a broader view, however, the effects that we have ignored do not 
1029: appear to be so intractable that they could not, in the future, be 
1030: quantified and incorporated into a theoretical model of migration. Our results 
1031: therefore imply that comparison 
1032: of theoretical models to the larger samples of extrasolar planets that are 
1033: forthcoming holds substantial promise for learning new information,  
1034: about both planet formation and protoplanetary disk physics, that is 
1035: currently unavailable via other methods.
1036: 
1037: \acknowledgements
1038: 
1039: I thank Richard Alexander for a detailed critique of an early version 
1040: of this paper, and in particular for alerting me to the work of \cite{ruden04}. I 
1041: am also grateful to the referee for an insightful report. 
1042: This work was supported by NASA under grants NAG5-13207, NNG04GL01G and NNX07AH08G 
1043: from the Origins of Solar Systems and Astrophysics Theory Programs, and 
1044: by the NSF under grant AST~0407040.
1045: 
1046: \begin{thebibliography}{}
1047: 
1048: \bibitem[Alexander, Clarke \& Pringle(2005)]{alexander05}
1049:  Alexander, R. D., Clarke, C. J., \& Pringle, J. E. 2005, 
1050:  \mnras, 358, 283
1051: 
1052: \bibitem[Alexander, Clarke \& Pringle(2006)]{alexander06}
1053:  Alexander, R. D., Clarke, C. J., \& Pringle, J. E. 2006, 
1054:  \mnras, 369, 229
1055: 
1056: \bibitem[Armitage \& Clarke(1996)]{armitage96}
1057:  Armitage, P. J.; \& Clarke, C. J. 1996, 
1058:  \mnras, 280, 458
1059: 
1060: \bibitem[Armitage, Clarke \& Palla(2003)]{armitage03}
1061:  Armitage, P. J., Clarke, C. J., \& Palla, F. 2003, 
1062:  \mnras, 342, 1139
1063: 
1064: \bibitem[Armitage et al.(2002)]{armitage02}
1065:  Armitage, P. J., Livio, M., Lubow, S. H., \& Pringle, J. E. 2002, 
1066:  \mnras, 334, 248
1067: 
1068: \bibitem[Balbus \& Hawley(1998)]{balbus98}
1069:  Balbus, S. A., \& Hawley, J. F. 1998, 
1070:  Reviews of Modern Physics, 70, 1
1071:  
1072: \bibitem[Bally \& Scoville(1982)]{bally82}
1073:  Bally, J., \& Scoville, N. Z. 1982, 
1074:  \apj, 255, 497
1075: 
1076: \bibitem[Bate et al.(2003)]{bate03}
1077:  Bate, M. R., Lubow, S. H., Ogilvie, G. I., \& Miller, K. A. 2003, 
1078:  \mnras, 341, 213
1079: 
1080: \bibitem[Bell \& Lin(1994)]{bell94}
1081:  Bell, K. R., \& Lin, D. N. C. 1994, 
1082:  \apj, 427, 987
1083: 
1084: \bibitem[Bodenheimer, Hubickyj \& Lissauer(2000)]{bodenheimer00}
1085:  Bodenheimer, P., Hubickyj, O., \& Lissauer, J. J. 2000, 
1086:  Icarus, 143, 2
1087: 
1088: \bibitem[Butler et al.(2006)]{butler06}
1089:  Butler, R. P., et al. 2006, \apj, 646, 505
1090:  
1091: \bibitem[Clarke, Gendrin \& Sotomayor(2001)]{clarke01}
1092:  Clarke, C. J., Gendrin, A., \& Sotomayor, M. 2001, 
1093:  \mnras, 328, 485
1094: 
1095: \bibitem[Fischer \& Valenti(2005)]{fischer05}
1096:  Fischer, D. A., \& Valenti, J. 2005, \apj, 622, 1102
1097:  
1098: \bibitem[Font et al.(2004)]{font04}
1099:  Font, A. S., McCarthy, I. G., Johnstone, D., \& Ballantyne, D. R. 2004, 
1100:  \apj, 607, 890
1101:  
1102: \bibitem[Ford, Havlickova \& Rasio(2001)]{ford01}
1103:  Ford, E. B., Havlickova, M., \& Rasio, F. A. 2001, 
1104:  Icarus, 150, 303
1105: 
1106: \bibitem[Gammie(1996)]{gammie96}
1107:  Gammie, C. F. 1996, \apj, 457, 355
1108:  
1109: \bibitem[Garaud \& Lin(2007)]{garaud07} 	
1110:  Garaud, P., \& Lin, D. N. C. 2007, 
1111:  \apj, 654, 606
1112:  
1113: \bibitem[Ge et al.(2006)]{ge06}
1114:  Ge, J., et al. 2006, SPIE, 6269, 75
1115:  
1116: \bibitem[Goldreich \& Tremaine(1980)]{goldreich80} 	
1117:  Goldreich, P., \& Tremaine, S. 1980, 
1118:  \apj, 241, 425
1119: 
1120: \bibitem[Haisch, Lada \& Lada(2001)]{haisch01}
1121:  Haisch, K. E., Lada, E. A., \& Lada, C. J. 2001, 
1122:  \apj, 553, L153
1123: 
1124: \bibitem[Hartmann et al.(1998)]{hartmann98}
1125:  Hartmann, L., Calvet, N., Gullbring, E., \& D'Alessio, P. 1998, 
1126:  \apj, 495, 385
1127:  
1128: \bibitem[Hayashi, Nakazawa \& Nakagawa(1985)]{hayashi85} 	
1129:  Hayashi, C., Nakazawa, K., \& Nakagawa, Y. 1985, 
1130:  in Protostars and Planets~II, University of Arizona Press (Tucson, AZ), 
1131:  p.~1100
1132: 
1133: \bibitem[Hollenbach et al.(1994)]{hollenbach94} 
1134:  Hollenbach, D., Johnstone, D., Lizano, S., \& Shu, F. 1994, 
1135:  \apj, 428, 654
1136: 
1137: \bibitem[Ida \& Lin(2004)]{ida04}
1138:  Ida, S., \& Lin, D. N. C. 2004, \apj, 616, 567
1139: 
1140: \bibitem[Lin, Bodenheimer \& Richardson(1996)]{lin96} 
1141:  Lin, D. N. C., Bodenheimer, P., \& Richardson, D. C. 1996,
1142:  Nature, 380, 606
1143: 
1144: \bibitem[Lin \& Papaloizou(1986)]{lin86} 
1145:  Lin, D. N. C., \& Papaloizou, J. 1986, 
1146:  \apj, 309, 846
1147:  
1148: \bibitem[Lubow \& D'Angelo(2006)]{lubow06}
1149:  Lubow, S. H., \& D'Angelo, G. 2006, 
1150:  \apj, 641, 526
1151:  
1152: \bibitem[Lubow, Seibert \& Artymowicz(1999)]{lubow99}
1153:  Lubow, S. H., Seibert, M., \& Artymowicz, P. 1999, 
1154:  \apj, 526, 1001
1155:  
1156: \bibitem[Lynden-Bell \& Pringle(1974)]{lyndenbell74} 	
1157:  Lynden-Bell, D., \& Pringle, J. E. 1974, 
1158:  \mnras, 168, 603
1159:  
1160: \bibitem[Marcy et al.(2005)]{marcy05} 	
1161:  Marcy, G., Butler, R. P., Fischer, D., Vogt, S., Wright, J. T., Tinney, C. G.,
1162:  \& Jones, H. R. A. 2005, Progress of Theoretical Physics Supplement, 158, 24
1163:  
1164: \bibitem[Martin et al.(2007)]{martin07}
1165:  Martin, R. G., Lubow, S. H., Pringle, J. A., \& Wyatt, M. C. 2007, 
1166:  \mnras, 378, 1589
1167:  
1168: \bibitem[Matsuyama, Johnstone \& Murray(2003)]{matsuyama03}
1169:  Matsuyama, I., Johnstone, D., \& Murray, N. 2003, 
1170:  \apj, 585, L143
1171:  
1172: \bibitem[Morbidelli et al.(2000)]{morbidelli00}
1173:  Morbidelli, A., Chambers, J., Lunine, J. I., Petit, J. M., Robert, F., Valsecchi, G. B., \& 
1174:  Cyr, K. E. 2000, Meteoritics \& Planetary Science, 35, 1309 
1175:  
1176: \bibitem[Muzerolle et al.(2000)]{muzerolle00}
1177:  Muzerolle, J., Calvet, N., Brice\~no, C., Hartmann, L., \& Hillenbrand, L. 2000, 
1178:  \apj, 535, L47
1179:  
1180: \bibitem[Pascucci et al.(2007)]{pascucci07}
1181:  Pascucci, I., et al. 2007, 
1182:  \apj, 663, 383
1183:  
1184: \bibitem[Pollack et al.(1996)]{pollack96}
1185:  Pollack, J. B., Hubickyj, O., Bodenheimer, P., Lissauer, J. J., Podolak, M., \& 
1186:  Greenzweig, Y. 1996, Icarus, 124, 62 
1187:  
1188: \bibitem[Pringle(1981)]{pringle81}
1189:  Pringle, J. E. 1981, ARA\&A, 19, 137
1190:   
1191: \bibitem[Pringle(1991)]{pringle91}
1192:  Pringle, J. E. 1991, \mnras, 248, 754 
1193:  
1194: \bibitem[Ruden(2004)]{ruden04}
1195:  Ruden, S. P. 2004, \apj, 605, 880 
1196:  
1197: \bibitem[Santos, Israelian \& Mayor(2004)]{santos04}
1198:  Santos, N. C., Israelian, G., \& Mayor, M. 2004, A\&A, 415, 1153
1199:  
1200: \bibitem[Santos, Benz \& Mayor(2005)]{santos05}	
1201:  Santos, N. C., Benz, W., \& Mayor, M. 2005, 
1202:  Science, 310, 251
1203:   
1204: \bibitem[Shakura \& Sunyaev(1973)]{shakura73}
1205:  Shakura, N. I., \& Sunyaev, R. A. 1973, 
1206:  A\&A, 24, 337
1207:  
1208: \bibitem[Shu, Johnstone \& Hollenbach(1993)]{shu93}
1209:  Shu, F. H., Johnstone, D., \& Hollenbach, D. 1993, 
1210:  Icarus, 106, 92 
1211:  
1212: \bibitem[Simon \& Prato(1995)]{simon95}
1213:  Simon, M., \& Prato, L. 1995, 
1214:  \apj, 450, 824
1215:  
1216: \bibitem[Syer \& Clarke(1995)]{syer95}
1217:  Syer, D., \& Clarke, C. J. 1995, \mnras, 277, 758
1218:   
1219: \bibitem[Takeuchi, Miyama \& Lin(1996)]{takeuchi96}
1220:  Takeuchi, T., Miyama, S. M., \& Lin, D. N. C. 1996, 
1221:  \apj, 460, 832 
1222:   
1223:  \bibitem[Trilling et al.(1998)]{trilling98} 
1224:   Trilling, D. E., Benz, W., Guillot, T., Lunine, J. I., Hubbard, W. B., \& 
1225:   Burrows, A. 1998, \apj, 500, 428
1226:   
1227: \bibitem[Trilling, Lunine \& Benz(2002)]{trilling02}
1228:  Trilling, D. E., Lunine, J. I., \& Benz, W. 2002, 
1229:  A\&A, 394, 241
1230:  
1231: \bibitem[Turner, Sano \& Dziourkevitch(2007)]{turner07}
1232:  Turner, N. J., Sano, T., \& Dziourkevitch, N. 2007, 
1233:  \apj, 659, 729
1234:  
1235: \bibitem[Veras \& Armitage(2004)]{veras04} 
1236:  Veras, D., \& Armitage, P. J. 2004, 
1237:  \mnras, 347, 613
1238:  
1239: \bibitem[Ward(1996)]{ward96} 	
1240:  Ward, W. R. 1996, Icarus, 126, 261
1241:   
1242: \bibitem[Weidenschilling(1977)]{weidenschilling77}
1243:  Weidenschilling, S. J. 1977, Astrophysics and Space Science, 51, 153  
1244:   
1245: \bibitem[Wilner et al.(2000)]{wilner00}
1246:  Wilner, D. J., Ho, P. T. P., Kastner, J. H., \& Rodríguez, L. F. 2000, 
1247:  \apj, 534, L101
1248:  
1249: \bibitem[Wolk \& Walter(1996)]{wolk96}
1250:  Wolk, S. J., \& Walter, F. M. 1996, 
1251:  \aj, 111. 2066
1252:  
1253: \bibitem[Zucker \& Mazeh(2002)]{zucker02}
1254:  Zucker, S., \& Mazeh, T. 2002, 
1255:  \apj, 568, L113 
1256:   
1257: \end{thebibliography}
1258: 
1259: \end{document}
1260: 
1261: