0705.1858/ms.tex
1: %                                                                 
2: \documentclass[useAMS,usenatbib]{mn2e}
3: %
4: \usepackage{graphicx}
5: \usepackage{pslatex}
6: \usepackage{natbib}
7: \voffset=-0.5in
8: %
9: % local macros
10: %
11: \def\deg{{\rm o}}
12: \def\idm#1{{\mbox{\scriptsize #1}}}
13: \newcommand\Chi{{(\chi^2_\nu)^{1/2}}}
14: \newcommand\stara{{14~Her}}
15: \newcommand\Ym{\langle Y\rangle}
16: \newcommand\Em{\langle E\rangle}
17: \def\mean#1{\langle #1\rangle}
18: %
19: \begin{document}
20: %
21:    \title{A dynamical analysis of the 14~Her planetary system}
22:    \author[K. Go\'zdziewski, C. Migaszewski and M. Konacki]{Krzysztof Go\'zdziewski$^1$\thanks{Toru\'n Centre for Astronomy, Poland}, 
23:     Cezary Migaszewski$^1$\thanks{Toru\'n Centre for Astronomy, Poland} and
24:     Maciej Konacki$^2$\thanks{Nicolaus Copernicus Astronomical Center,
25:     Polish Academy of Sciences, Toru\'n, Poland; Obserwatorium Astronomiczne, 
26:     Uniwersytet A. Mickiewicza, Pozna\'n, Poland}
27:    }
28: 
29:    \maketitle
30:    \begin{abstract}
31:    Precision radial velocity (RV) measurements of the Sun-like dwarf 
32:    14~Herculis published by Naef et. al (2004), Butler et. al (2006) and
33:    Wittenmyer et al (2007) reveal a Jovian planet in a 1760~day orbit and a
34:    trend indicating the second distant object. On the grounds of dynamical
35:    considerations,  we test a hypothesis that the trend can be  explained  by
36:    the presence of an additional giant planet. We derive dynamical limits to the
37:    orbital parameters of the putative outer Jovian companion in an orbit within
38:    $\sim 13$~AU. In this case, the mutual interactions between the Jovian
39:    planets are important for the long-term stability of the system.  The best
40:    self-consistent and stable Newtonian fit to an edge-on configuration of
41:    Jovian planets has the outer planet in 9~AU orbit with a moderate
42:    eccentricity $\sim 0.2$ and confined to a zone spanned by the low-order mean
43:    motion resonances 5:1 and 6:1. This solution lies in a shallow minimum of
44:    $\Chi$ and persists over a wide range of the system inclination.  Other
45:    stable configurations within $1\sigma$ confidence interval of the best fit
46:    are possible for the semi-major axis  of the outer planet in the range of
47:    (6,13)~AU and the eccentricity in the range of (0,0.3).  The orbital
48:    inclination cannot yet be  determined but when it decreases, both planetary
49:    masses approach  $\sim 10 \mbox{m}_{\idm{J}}$ and for $i \sim 30^{\deg}$ the
50:    hierarchy of the masses is reversed.  
51:   \end{abstract}
52: 
53:   \begin{keywords}
54:    extrasolar planets---radial velocity---stars:individual 14Her---N-body problem
55:   \end{keywords}
56: %
57: \section{Introduction}
58: %______________________________________________________________________________
59: %
60: An extrasolar planet around 14~Herculis discovered by the Geneva planet search
61: team was announced in a conference talk (1998). The star was also monitored by
62: other planet-hunting teams. The Jovian companion in $\sim 1700$~days orbit was
63: next confirmed by \cite{Butler2003}. In their new paper,  the discovery team
64: \citep{Naef2004} published 119 observations revealing a linear RV trend of 
65: $\sim 3.6$~m/s per year. The single planet Keplerian solution to these
66: measurements yielded an abnormally large rms of about $14$~m/s. Even if the
67: drift was accounted for, the rms of the single planet+drift model was $\sim
68: 11.3$~m/s, significantly larger than the mean observational uncertainty
69: $\mean{\sigma_{\idm{m}}} \sim 7.2$~m/s quoted in that paper. 
70: 
71: Following the hypothesis that the linear trend in the RV data may indicate
72: other, more distant bodies, in \cite{Gozdziewski2006a}, we merged these
73: measurements  with  accurate, 35 observations (the mean of $\sigma_{\idm{m}}\sim
74: 3.1$~m/s) by \cite{Butler2003}, spanning the middle part of the combined
75: observational window. We reanalyzed the full data set of 154 measurements,
76: covering about $3400$~days, $\sim 2 P_{\idm{b}}$. Using our GAMP method
77: \citep{Gozdziewski2003e}, i.e., the optimization incorporating  stability
78: constraints into the fitting algorithm, we searched for stable configurations of
79: putative two-planet systems which are consistent with the RV data. This search
80: revealed many stable solutions, in particular two distinct and equally good best
81: fits with $a_{\idm{c}}\sim 5.8$~AU and $a_{\idm{c}}\sim 9$~AU.  They were
82: localized in the proximity of low-order mean motion resonances (MMRs), 3:1~MMR
83: (14~Her$^a$) and 6:1~MMR (14~Her$^b$). Apparently, the companions would be well
84: separated, but the system would still be active dynamically.  The dynamical maps
85: computed in the neighborhood of the selected fits revealed a complex structure
86: of the phase space. We found a clear relation of  chaotic motions to strongly
87: unstable, self-disrupting  configurations on a short time-scale of $10^4$
88: periods of the outermost planet.
89: 
90: After the upgrade of HIRES spectrograph and improvements to the data pipeline, 
91: \cite{Butler2006} have re-analyzed their spectra and published a revised set of
92: 50 observations of 14~Her. Their mean accuracy $\sim 2.4$~m/s is already very
93: good. Yet, towards the end of the observational window, single data points  have
94: formal errors as small as 1~m/s. Recently, \cite{Wittenmyer2007}  have also
95: published 35 additional and independent precision measurements made at the
96: McDonald Observatory. They have a mean accuracy of 7.5~m/s. The full publicly
97: available data set is now comprised of 203 observations spanning 4463.2~days and
98: still does support the presence of a second planetary  object. The single-planet
99: solution has an rms $\sim 13$~m/s, and  shows a few meter per second excess 
100: over the mean $\sigma_{\idm{m}} \sim 6$~m/s, added in quadrature to the stellar
101: jitter $\sim 3.5$~m/s \citep{Butler2006}. Using the Keplerian model of the RV,
102: \cite{Wittenmyer2007} have found the best-fit solution in the vicinity of the
103: 4:1 mean motion resonance (MMR). However, having in mind the results of our
104: earlier dynamical analysis \citep{Gozdziewski2006a}, we can suspect that the
105: kinematic model is not  adequate to properly interpret the RV observations. 
106: 
107: In this paper, we re-analyse the updated high-precision RV data to study and
108: extend the kinematic model by \cite{Wittenmyer2007}. We also revise and
109: correct the conclusions from our previous paper \citep{Gozdziewski2006a} that
110: were derived on the basis of a smaller and less precise RV data set.  The goal
111: of our work is to derive dynamical characteristics of the putative planetary
112: system and to place limits on its orbital parameters.   Assuming a putative
113: configuration close to the 4:1~MMR, the outermost  orbital  period would be
114: $\sim 18$--$20$~yr. Hence, the currently available  data would cover only a part
115: of this period and a full characterization of  the orbits would still require
116: many years of observations. One cannot expect  that in such a case the orbits
117: can be well determined. Nevertheless, we show  in this work, that the modeling
118: of the RV data when merged with a dynamical analysis and mapping of the phase
119: space of initial conditions with fast-indicators enable us to derive meaningful
120: limits to the orbital parameters. Such an extensive dynamical study of this
121: systems has not yet been done.
122: 
123: In order to take into account the RV variability  induced by the stellar
124: activity, the internal errors of the RV measurements are rescaled by the
125: jitter   added in quadrature: 
126: $
127:  \sigma^2 = \sigma_{\idm{m}}^2 + \sigma_{\idm{j}}^2,
128: $ 
129: where $\sigma_{\idm{m}}$ and $\sigma_{\idm{j}}$ are the mean of  the internal
130: errors and the adopted estimate of the stellar jitter.  The
131: 14~Herculis is a quiet star with $\log R'_{\idm{HK}}=-5.07$ \citep{Naef2004}
132: thus it is reasonable to keep the safe estimate of  $\sigma{\idm{j}}=3.5$~m/s
133: \citep{Butler2006}. The mass of the parent star varies by $\sim 10\%$ in
134: different publications. In this work, we adopt the canonical mass
135: $1\mbox{M}_{\sun}$, although we also did some calculations for the  mass of
136: $0.9\mbox{M}_{\sun}$ and $1.1\mbox{M}_{\sun}$.
137: 
138: 
139: \begin{figure*}
140:    \centerline{
141:    \vbox{
142:    \hbox{
143:        \hbox{\includegraphics[width=6.0in]{fig1a.eps}}
144:    }
145:    \hbox{
146:        \hspace*{1.2cm}
147:        \hbox{\includegraphics[width=2.56in]{fig1b.eps}}
148:        \hspace*{0.8cm}
149:        \hbox{\includegraphics[width=2.56in]{fig1c.eps}}
150:    }
151:    }
152:    }
153:    \caption{
154:    The best fit solutions to the two-planet model of the RV data of \stara{} in
155:    the ($P,e$)- and ($P,K$)-plane. In the search,  a distant companion to the
156:    known Jovian planet in $\sim 1760$~days orbit is assumed. Upper plots are for
157:    the inner planet, middle plots are for the outer one.  The best-fit found
158:    with the hybrid code in the range of moderate orbital periods of the outer
159:    planet, with $\Chi \sim 1.085$, is marked with intersecting lines. Its
160:    elements, in terms of the parameter tuples of Eq.1, $(K\mbox{[m/s]},
161:    P\mbox{[days]}, e,  \omega\mbox{[deg]},T_{\idm{p}}-T_0\mbox{[days]})$,   are
162:    (89.460, 1765.743, 0.360,  22.108, 1375.03) for the inner and (226.2263,
163:    11675.333, 0.395, 182.006, 6115.352 ) for the outer planet, respectively,
164:    with the velocity offsets 
165:    ($V_{\idm{N2004}},V_{\idm{B2006}},V_{\idm{W2007}}$)   $=$ 
166:    (-129.105, -132.345, -108.384)~m/s for the data in \citep{Naef2004}, \citep{Butler2006} and
167:    \citep{Wittenmyer2007}, respectively;  the epoch $T_0=$~JD2,450,000.  
168:    Dots are for $1\sigma$  solutions with $\Chi \in [1.0848,1.090)$ and an
169:    rms $\sim 8.17$~m/s,
170:    red crosses are for the the best fit solutions with $\Chi \sim 1.0848$. 
171:    The red curve in ($P_{\idm{c}},e_{\idm{c}}$)-plane marks an approximate
172:    collision line of the orbits determined from $a_{\idm{b}} (1+e_{\idm{b}}) =
173:    a_{\idm{c}} (1-e_{\idm{c}})$, with $a_{\idm{b}},e_{\idm{b}}$ fixed at their
174:    best-fit values. 
175:    Bottom panels are for the scans of $\Chi$ at the selected
176:    parameter planes. Contours (also shown in the upper panels,
177:   for a reference) mark the limits of
178:    $\Chi$ corresponding to $1\sigma,2\sigma$ and $3\sigma$ of the
179:    best fit solution.
180:    Circle mark the position of the best $N$-body solution (see Sect.~4).
181:    See the text for remaining details.
182:    }
183: \label{fig:fig1}
184: \end{figure*}
185: %
186: \section{Modeling the RV data -- an overview}
187: %______________________________________________________________________________
188: %
189: The standard formulae by~\cite{Smart1949} are  commonly used to model the RV
190: signal. Each planet in the system contributes to the reflex motion of the star
191: at time $t$ with:   \begin{equation} V_{\idm{r}}(t) = K [ \cos (\omega+\nu(t)) 
192: + e \cos \omega] + V_0, \label{eq:eq1}   \end{equation} where $K$ is the
193: semi-amplitude, $\omega$ is the argument of pericenter, $\nu(t)$ is the true
194: anomaly (involving implicit dependence on the orbital period $P$ and time of
195: periastron passage $T_{\idm{p}}$), $e$ is the eccentricity, $V_0$ is the
196: velocity offset. We interpret the primary model parameters
197: $(K,P,e,\omega,T_{\idm{p}})$ in terms of the Keplerian elements and minimal
198: masses related to Jacobi coordinates \citep{Lee2003,Gozdziewski2003e}.  
199: 
200: In our previous works, we tested and optimized different tools used to explore
201: the multi-parameter space of $\chi^2$. In the case when $\chi^2$ has many local
202: extrema, we found that the hybrid optimization  provides particularly  good
203: results \citep{Gozdziewski2004, Gozdziewski2006b}. A run of our code starts the
204: genetic algorithm \citep[GA,][]{Charbonneau1995} that basically  has a global
205: nature and requires only the $\chi^2$ function and rough parameter ranges. GA
206: easily permits for a constrained optimization within prescribed parameter bounds
207: or for a penalty term to $\chi^2$ \citep{Gozdziewski2006a}.  Because the best
208: fits found with GAs are not very accurate in terms of $\chi^2$, we refine  a
209: number of ``fittest individuals'' in the final ``population'' with a relatively
210: fast local method like the simplex of Melder and Nead~\citep{Press1992}.   The
211: use of simplex algorithm is a matter of choice and in principle we could use 
212: other fast local methods. However, a code using non-gradient methods works  with
213: the minimal requirements for the user-supplied information, i.e., the model
214: function, usually equal to $\chi^2$ [its inverse  $1/\chi^2$ is the {\em fitness
215: function} required by GAs].   The  repeated runs  of the hybrid code provide an
216: ensemble of the best-fits that helps us to detect local minima of $\chi^2$, even
217: if they are distant in the parameter space. We can also obtain reliable
218: approximation to the errors of the best-fit parameters \citep{Bevington2003}
219: within the $1\sigma$, $2\sigma$ and $3\sigma$  confidence levels of $\chi^2$ at
220: selected  2-dim parameter planes,  as corresponding to appropriate increments of
221: $\chi^2$. The hybrid approach will be called the algorithm~I from hereafter.
222: 
223: However, in this paper we deal with a problem of only a partial coverage of the
224: longest orbital period by the data. In such a case $\chi^2$ does not have a  
225: well defined minimum or the space of acceptable solutions is  very ``flat'' so 
226: the confidence levels may cover large ranges of the parameters.  In such a case,
227: to illustrate {\em the shape} of $\chi^2$ in the selected 2-dimensional
228: parameter planes, we use a complementary approach that relies on systematic
229: scanning of the parameter space with the fast Levenberg-Marquardt (L-M)
230: algorithm \citep{Press1992}. Usually, as the parameter plane for representing
231: such scans, we choose the semimajor-axis---eccentricity $(a,e)$ plane of the
232: outermost planet.  We fix $(a,e)$ and then search for the best fit, with the
233: initial conditions selected randomly (but within reasonably wide parameter
234: bounds).  Then the L-M scheme ensures a rapid convergence. This approach is
235: called the algorithm~II from hereafter. In fact, we already used it in
236: \cite{Gozdziewski2003e}. The method is  CPU-time consuming and may be
237: effectively applied in low-dimensional problems but in reward it can provide a
238: clear and global picture of the parameter space.
239: 
240: Moreover, the RV measurements can be polluted by many sources of error, like 
241: complex systematic instrumental effects, short time-series of the observations,
242: irregular sampling due to observing conditions, and stellar noise.  The problem
243: is even more complex when we deal with models of resonant or strongly
244: interacting planetary systems. It is well known that the kinematic model of the
245: RV is not adequate to describe the  observations in such instances. Instead, the
246: self-consistent $N$-body Newtonian model should be applied
247: \citep{Rivera2001,Laughlin2001}. (Note that the hybrid optimization can be used
248: to explore the parameter space for both models).   Still, the best-fit solutions
249: may be related (and often do) to unrealistic quickly disrupting configurations
250: \citep{FerrazMello2005,Lee2006,Gozdziewski2006a}. In such a case one should
251: explore the dynamical stability of the planetary system in a neighborhood of the
252: best-fit configuration. One can do that with the help of direct numerical
253: integrations or by resolving the dynamical character of orbits in terms of the
254: maximal Lyapunov exponent or the diffusion of fundamental frequencies. Hence,
255: the most general way of modeling the RV data should involve an elimination of
256: unstable (for instance, strongly chaotic) solutions during the fitting
257: procedure. We described  such an approach (called GAMP) in our previous works. 
258: In particular, we made attempts to analyze the old data set of 14~Her as well as
259: other stars hosting multiple planet systems \citep{Gozdziewski2006a}.
260: 
261: 
262: %
263: \section{Keplerian fits}
264: %______________________________________________________________________________
265: %
266: In order to compare our fits with the results of \cite{Wittenmyer2007}, we
267: carried out the hybrid search for the two-planet edge-on configuration.    In
268: the experiment, we limited the orbital periods to the range of 
269: $[1000,14000]$~days and the eccentricities to the range of $[0,0.9]$.  The
270: choice of the upper limit of orbital periods followed the conclusions of
271: \cite{Wittenmyer2007}. According to the results of adaptive-optics imaging by
272: \cite{Luhman2002} and \cite{Patience2002}, the parent star has no stellar
273: companion beyond $9$ --- $12$~AU, so we try to verify the hypothesis about a
274: short-period, low-mass companion \stara{}~c up to that distance. The RV offsets
275: are different for each of the three instruments and are included as free
276: parameters in the model together with the tuples of $(K,P,e,\omega,T_{\idm{p}})$
277: for each companion. Note that we calculate the RV offset  $V_{\idm{N2004}}$ 
278: with respect to the simple mean of all RV measurements from the set published by
279: \cite{Naef2004}. 
280: 
281: The results of the hybrid search (algorithm~I) are illustrated in upper panels
282: Fig.~\ref{fig:fig1} (note that before plotting,  the fits are sorted with
283: respect to the smaller semi-major axis).  The best-fit found with this method
284: has $\Chi\simeq 1.085$ and an rms $\simeq 8.17$~m/s. Its parameters are given in
285: the caption to Fig.~\ref{fig:fig1}.  Clearly, there is no isolated best-fit
286: solution but rather a huge number of acceptable fits. The  ratio of the orbital
287: periods within  $1\sigma$ confidence interval of the best fit may be as low as
288: $\sim 3$. The orbital period in the best fit is $\sim 12000$~days and
289: $P_{\idm{c}}/P_{\idm{b}}\sim 7$. This is roughly consistent with the work of
290: \cite{Wittenmyer2007} who quote the best-fit configuration close to the 4:1~MMR
291: and with a similar rms $8.17$~m/s. Yet  the fits are not constrained with
292: respect to $P_{\idm{c}}$. Many values between $[5000,14000]$~days are equally
293: good in terms of $1\sigma$ confidence interval of the best-fit solution. 
294: Moreover, it means that the system may be confined to a zone spanned by many
295: low-order mean motion resonances of varying width in the
296: $(P_{\idm{c}},e_{\idm{c}})$-plane. In such cases, the  mutual interactions
297: between the planets are important for the long-term stability.
298: 
299: In order to illustrate the shape of $\Chi$ in more detail, we also applied the
300: algorithm~II for the systematic  scanning of that function in selected parameter
301: planes (see lower panels in Fig.~\ref{fig:fig1}). This has lead to even better 
302: statistics of the best fits and a very clear illustration of the
303: $\Chi$-surface. The panels  reveal that $\Chi$ of 2-Kepler model has no minimum
304: in the examined ranges of the outermost periods, $P_{\idm{c}} < 14,000$~days.
305: 
306: Along a wide valley, the best fits with an increasing $P_{\idm{c}}$ have also
307: rapidly increasing $e_{\idm{c}}$. For $P_{\idm{c}} \sim 14,000$~days, they reach
308: the collision zone of the orbits.  In this strongly chaotic region a number of
309: MMRs of the type $p:1$ with $p>7$ overlap. It justifies  a-posteriori our choice
310: of the search limit $P_{\idm{c}} \sim 14,000$~days. Note, that the two searches
311: are in an excellent accord that is illustrated with the $\Chi$ smooth contour
312: levels overlayed on the results of the hybrid search. In that case, the scanning
313: of parameter space helps to understand the shape of $\Chi$ in detail.
314: 
315: The $1\sigma$ scatter of solutions around the best-fit (Fig.~\ref{fig:fig1})
316: provides realistic error estimates. Although the parameters of the inner planet
317: are very well fixed, the elements of the outer Jovian planet have much larger
318: formal errors. The kinematic fits do not provide an upper bound to the
319: semi-major axis of the outer planet. Both methods reveal exactly the same
320: parameters of the inner companion (see Fig.~\ref{fig:fig1}). Obviously, certain
321: discrepancy between the outcomes of algorithm~I and algorithm~II is related to a
322: ``flat'' shape of $\Chi$ in the $(a_{\idm{c}},e_{\idm{c}})$-plane and to a worse
323: efficiency of simplex in detecting shallow minima compared to the gradient-like
324: L-M algorithm.  When $\Chi$ does not have a well localized minimum, it is very
325: difficult to resolve its behaviour with any non-gradient method.
326: 
327: %
328: \section{Self-consistent Newtonian fits}
329: %______________________________________________________________________________
330: %
331: In order to compare the outcome of kinematic modeling with the self-consistent
332: $N$-body one, at first, we transformed the ensemble of Keplerian fits to
333: astro-centric osculating elements \citep{Lee2003,Gozdziewski2003e}. These fits
334: became the initial conditions for the L-M algorithm employing the $N$-body model
335: of the RV. In this experiment, the mass of the star was $M_{\star} =
336: 1~M_{\sun}$. Curiously, we found that the procedure often converged to the two
337: best-fits, both having similar $\Chi \sim 1.083$, an rms $\sim 8.15$~m/s, and
338: $a_{\idm{c}}\sim 7.8$~AU and $a_{\idm{c}}\sim 8.4$~AU, respectively.  To shed
339: some light on the orbital stability of these solutions, we computed the
340: dynamical maps in the neighborhood of the best-fits in terms of the Spectral
341: Number \citep[$\log SN$,][]{Michtchenko2001}. We chose $a_{\idm{c}}$ and
342: $e_{\idm{c}}$ as the map coordinates, keeping other orbital parameters at their
343: nominal values. The results are shown in Fig.~\ref{fig:fig2}. The left panel is
344: for the unstable fit located in the separatrix of the 5:1~MMR. The right panel
345: is for the stable solution in the center of 9:2~MMR.  These dynamical maps prove
346: that the mutual interactions between planets are  significant --- even small
347: changes of initial parameters modify the shapes of low-order MMRs which overlap
348: already for $e_{\idm{c}} \sim 0.2-0.3$. Besides, the chaotic motions are related
349: to strongly unstable configurations (see also Fig.~\ref{fig:fig3},
350: \ref{fig:fig4}).
351: 
352: \begin{figure*}
353:    \centerline{
354:    \hbox{
355:        \hbox{\includegraphics[width=2.56in]{fig2a.eps}}
356:        \hspace*{1cm}
357:        \hbox{\includegraphics[width=2.56in]{fig2b.eps}}
358:    }
359:    }
360:    \caption{
361:    The dynamical maps in terms of the Spectral Number ($\log SN$) in the
362: $(a_{\idm{c}},e_{\idm{c}})$-plane for the two-planet Newtonian models of the
363: \stara{} system.  Colors used in the $\log SN$ map classify the orbits --- black
364: indicates  quasi-periodic regular configurations while yellow strongly chaotic
365: ones.  The maps are computed for the following osculating elements of the Jovian
366: planets at the epoch of the first observation $t_0$, in terms of
367: tuples                                                  
368: $(m~[m_{\idm{J}}], a~[\mbox{AU}], e, \omega~[\mbox{deg}], {\cal M}(t_0)~[\mbox{deg}])$:
369: the left panel is for
370:  (4.975,    2.864,    0.359,   22.281,  330.365) of the inner
371:  planet, and
372:  (5.618,    8.343,    0.174,  188.543,   72.091) of the outer planet,
373:  respectively;
374: the right panel is for
375:  (4.974,    2.863,    0.359,   22.310,  330.314) of the inner planet, and
376:  (4.558,    7.863,    0.173,  187.674,   64.417) of the outer planet,
377: respectively. The large circles mark the  parameters of the best
378: fits.   The thin line marks the collision curve for planets b and~c,   as
379: determined by $a_{\idm{b}} (1+e_{\idm{b}}) = a_{\idm{c}} (1-e_{\idm{c}})$. The
380: low-order MMRs of the planets b and c are labeled.  The integrations are conducted
381: over $\sim 10^4 P_{\idm{c}}$. The resolution is $500\times120$ data points.
382: }
383: \label{fig:fig2}%
384: \end{figure*}
385: 
386: As in the case of the Keplerian model, to resolve  the topology of $\Chi$ in
387: detail, we performed similar  searches for the best-fit configurations  using
388: the $N$-body model of the RV.  Again, the results of the hybrid algorithm are
389: consistent with the algorithm~II. However, to conserve space, we only show the
390: $\Chi$-maps obtained by scanning the ($a_{\idm{c}},e_{\idm{c}}$)-plane. The
391: results are illustrated in the left column panels of Fig.~\ref{fig:fig3}.
392: 
393: These scans are computed for three different and fixed values of  the
394: inclination of the putative orbital plane of the system, $i=90^{\circ}$,
395: $60^{\circ}$ and $30^{\circ}$, respectively. In the maps, we mark a few contour
396: levels of $\Chi$ that correspond to $1\sigma, 2\sigma$ and $3\sigma$-levels of
397: the best fit solution denoted with a circle in every panel.
398: 
399: Overall, the topology of $\Chi$ does change, nevertheless qualitatively certain 
400: features are common. Contrary to the Keplerian case,  we can detect a shallow
401: minimum of $\Chi$ close to (9~AU,0.2) and a hint of another minimum localized
402: far over the collision line. The white thick lines on the panels mark mass
403: limits of $m_{\idm{c}}$.  Clearly, for $a_{\idm{c}}>9$~AU and for moderate
404: eccentricity, $m_{\idm{c}}$ approaches the brown-dwarf mass limit and substellar
405: mass  80~$m_{\idm{J}}$ for $\sim 12$~AU.   Thus a brown-dwarf could exist in the
406: system beyond $10$-$11$~AU depending on the inclination and eccentricity ---
407: and  the fits preclude a low-mass planetary object over the 14~$m_{\idm{J}}$
408: level. 
409: 
410: Note, that the best-fit for the nominal edge-on configuration 
411: with $i=90^{\circ}$ turns out to be {\em unstable} but its parameters are very
412: close to {\em a stable} solution given in Table~1; its synthetic curve is
413: illustrated in Fig.~\ref{fig:fig5}.
414: 
415: To characterize the stability of the best-fit solutions, we computed their MEGNO
416: fast indicator \citep{Cincotta2000}, $\Ym$, (i.e., an approximation of the
417: maximal Lyapunov exponent) up to $\sim 2\times 10^5~P_{\idm{c}}$ using a fast
418: and accurate symplectic algorithm \citep{Gozdziewski2005b}. The total
419: integration time $3$-$4$~Myr is long enough to detect even weak unstable MMRs.
420: 
421: The best-fits that passed the MEGNO test (i.e. that are regular over the
422: integration time) are marked with black dots in the right panels of
423: Fig.~\ref{fig:fig3}.  For a reference, the $\sigma$-levels of $\Chi$ from the
424: respective left-panels are marked in these plots as well. The red curves are for
425: the collision line of the orbits computed for well constrained and fixed
426: parameters of the inner planet. Overall, the stable solutions reveal the
427: structure of the MMRs that we have already seen in the dynamical maps computed
428: for fixed orbital parameters  (see Fig.~\ref{fig:fig2}). The semi-major axis of
429: the outer planet cannot be yet constrained well and may vary between $6$~AU and
430: (at least) 13~AU.  Moreover there is a rather evident although irregular border
431: $e_{\idm{c}} \sim 0.3$ of all solutions, up to $3\sigma$.  Besides, we found
432: that in the relevant range of $a_{\idm{c}}$, the mean longitude
433: $\lambda_{\idm{c}} = \varpi_{\idm{c}} + {\cal M}_{\idm{c}} \sim
434: (200^{\circ},300^{\circ})$ within the $1\sigma$ interval of the best fit,
435: increasing along quasi-parabolic curve as the function of $a_{\idm{c}}$.
436: Simultaneously, $\lambda_{\idm{b}} \sim 352.6^{\circ} \pm 1.5^{\circ}$. From the
437: dynamical point of view, this provides significant limits on the relative mean
438: phases of the planets.
439: 
440: The results for two-planet 14~Her system in \cite{Gozdziewski2006a} are
441: basically in accord with the present work although not all conclusions are
442: confirmed. The two isolated minima of $\Chi$ found in that paper can be
443: identified in this work too.  In particular it concerns the narrow 3:1~MMR
444: island. The second solution close to the 6:1~MMR has also been found close to
445: the present best fit solution. Nevertheless, the more extensive current search
446: reveals many other stable solutions between 3:1 and 9:1~MMRs equally good in
447: terms of $1\sigma$. Thus the results of our previous work are not wrong in
448: general but rather incomplete --- it appears that the search performed was not
449: exhaustive enough to find all acceptable solutions.  Still, some of the unstable
450: solutions can be modified in order to obtain stable fits, possibly with slightly
451: increased $\Chi$. For instance, the stability maps in Fig.~\ref{fig:fig3} seem
452: to rule out the 4:1~MMR in the system what may be related to specific orbital
453: phases. After an appropriate change of these phases [still keeping $\Chi$ at
454: acceptable level] we could find also stable solutions close to the 4:1~MMR. In
455: order to do this in a self-consistent manner, a GAMP-like method would be
456: necessary. However, as we have shown, $\Chi$ is very flat in the interesting
457: range of orbital parameters. Then, the extensive enough GAMP search would
458: require a very significant amount of CPU time, so we decided not to do it.
459: 
460: \begin{table}
461: \label{tab:tab1}
462: \caption{
463: Astrocentric, osculating Keplerian parameters of the edge-on, best-fit,
464: self-consistent configuration of 14~Her system at the epoch of the first
465: observation, $t_0=$~JD~2459464.5956. The mass of the parent star is
466: 1.0~$M_{\sun}$. See the text and Fig.~\ref{fig:fig3} for the error estimates. 
467: The RV offsets $V_{\idm{N2004}}$, $V_{\idm{B2006}}$, and $V_{\idm{W2007}}$ label
468: the data sets published in \citep{Naef2004,Butler2006,Wittenmyer2007},
469: respectively; the offset of $V_{\idm{N2004}}$ is given with respect to the mean
470: of RV in \citep{Naef2004}. ${\cal M}$ denotes the mean anomaly at the epoch
471: $t_0$.
472: }
473: \centering
474: \begin{tabular}{lcc}
475: \hline
476: \hline
477: parameter \hspace{1em}  
478: & \ \ \ \ \ \ \ \ planet  {\bf b} \ \ \ \ \  & \ \ \ \ \ planet {\bf c} \ \ \ \ \  
479: \\
480: \hline
481: $m$ [m$_{\idm{J}}$]  
482: 			&   4.975   &    7.679   
483: \\
484: $a$ [AU] 		&   2.864   &   9.037  
485: \\
486: $P$ [days]              &   1766    &   9886  
487: \\
488: $e$     		&   0.359   &    0.184  
489: \\
490: $\omega$ [deg] 		&  22.230   &  189.076 
491: \\
492: ${\cal M}(T_0)$ [deg]
493: 			&  330.421  &  81.976
494: \\
495: $V_{\idm{N2004}}$ [m s$^{-1}$] 	& \multicolumn{2}{c}{-52.049} 
496: \\
497: $V_{\idm{B2006}}$ [m s$^{-1}$] 	& \multicolumn{2}{c}{-55.290} 
498: \\
499: $V_{\idm{W2007}}$ [m s$^{-1}$] 	& \multicolumn{2}{c}{-31.368} 
500:  \\
501: $\Chi$  		& \multicolumn{2}{c}{1.0824}
502: \\
503: rms~[m s$^{-1}$] 	& \multicolumn{2}{c}{8.15}
504: \\
505: \hline
506: \end{tabular}
507: \end{table}
508: 
509: 
510: Finally, using the algorithm~II, we try to resolve the topology of $\Chi$ with
511: respect to a varying mass of the star and inclination of the system  (both can
512: be free parameters in the model) but assuming that the system remains 
513: coplanar.  Adopting the uncertainty of the 14~Her mass $\sim 10\%$
514: \citep{Butler2006}, we choose $0.9\mbox{M}_{\sun}, 1.0\mbox{M}_{\sun},
515: 1.1\mbox{M}_{\sun}$, and inclinations of orbital plane as $90^{\circ}$ (the
516: edge-on system), $60^{\circ}, 45^{\circ}, 30^{\circ}$.  Apparently, the
517: distributions of the best-fits are similar for all ($\mbox{M}_{\star},i$)-pairs.
518: Their quality in terms of $\Chi$ is comparable.
519: 
520: The significant differences between the best-fit configurations are related to
521: their stability.  To address this issue, we calculated the dynamical maps for
522: the best fits derived for the nominal mass of the parent star
523: ($1~\mbox{M}_{\sun}$) and inclinations $90^{\circ}$ (the edge-on system,
524: Table~1), $60^{\circ}$, and $30^{\circ}$ (see
525: Fig.~\ref{fig:fig4},\ref{fig:fig6}). The $\log SN$ maps are accompanied by the
526: maximal eccentricity maps, $\max e_{\idm{c}}$, i.e., the eccentricity attained
527: by the orbit during the total integration time. The $\max e$ indicator makes it
528: possible to resolve a link between formally chaotic character of orbits with
529: their physical, short-term instability. In the regions where chaos is strong,
530: the maximal eccentricity grows quickly to~1, implying ejections or collisions
531: between the planets or with the star.
532: 
533: The nominal system lies in a stable zone between 5:1 and 6:1~MMRs (see
534: Fig.~\ref{fig:fig4}). However, for the inclination of $\sim 60^{\circ}$, the
535: best fit configuration appears locked in the 5:1~MMR. Overall,  the phase space
536: does not change significantly compared to the edge-on system
537: (Fig.~\ref{fig:fig4}, also Fig.~\ref{fig:fig2}). For the inclination of
538: $30^{\circ}$, the stable zone shrinks due to strong interactions caused by large
539: masses of the planets $\sim 10\mbox{m}_{\idm{J}}$ (see caption to
540: Fig.~\ref{fig:fig6}).   Moreover, the masses {\em are not} simply scaled by
541: factor $1/\sin i$, as one would expect, assuming the kinematic model of the RV.
542: Instead, already for $i\sim 30^{\circ}$, the mass hierarchy is reversed -- the
543: inner planet becomes more massive than the outer one. It is yet another argument
544: against the  kinematic model of the RV data. The best-fits derived for small
545: inclinations are confined to a strongly chaotic zone.  Comparing the dynamical
546: map for $i\sim 30^{\deg}$ with Fig.~\ref{fig:fig4}, we conclude that is unlikely
547: to find stable systems close to the formal best fit with $a_{\idm{c}} \sim
548: 9$~AU  unless they are trapped in 5:1 MMR. The resonance island can be seen at
549: the edge of the bottom panels in Fig.~\ref{fig:fig6}. The planetary masses close
550: to the minimum of $\Chi$ remain smaller from the brown-dwarf limit even for
551: relatively small inclinations of the orbital plane. 
552: 
553: 
554: \begin{figure*}
555:    \centerline{
556:    \vbox{
557:    \hbox{
558:        \hbox{\includegraphics[width=2.7in]{fig3a.eps}}      
559:        \hspace*{0.4cm}     
560:        \hbox{\includegraphics[width=2.65in]{fig3b.eps}}
561:    }
562:    \hbox{
563:        \hbox{\includegraphics[width=2.7in]{fig3c.eps}}      
564:        \hspace*{0.4cm}   
565:        \hbox{\includegraphics[width=2.65in]{fig3d.eps}}
566:    }
567:    \hbox{
568:        \hbox{\includegraphics[width=2.7in]{fig3e.eps}}      
569:        \hspace*{0.4cm}
570:        \hbox{\includegraphics[width=2.65in]{fig3f.eps}}
571:    }
572:    }
573:    }
574:    \caption{The panels in the left column are for the Newtonian best-fit model
575:    with varying fixed inclination of the coplanar system $i$ and illustrated as
576:    projections onto the $(a{\idm{c}},e_{\idm{c}})$-plane of initial osculating
577:    elements at the epoch of the first observation.  In each panel, the best fit
578:    is marked with a circle. The quality of the fits within formal  $1\sigma,
579:    2\sigma$ and $3\sigma$ confidence interval of the best fit, in terms of their
580:    $\Chi$ is marked with contours and labeled in subsequent panels.  White thick
581:    lines corresponds to mass levels of a brown-dwarf and sub-stellar companion,
582:    respectively. These solutions are derived with algorithm~II. The panels in
583:    the right column are for the stability analysis of the best-fit
584:    configurations. See the text for more explanation.
585:    }
586: \label{fig:fig3}
587: \end{figure*}
588: 
589: \begin{figure*}
590:    \centerline{
591:    \vbox{
592:    \hbox{
593:        \hbox{\includegraphics[width=2.56in]{fig4a.eps}}
594:        \hspace*{1cm}
595:        \hbox{\includegraphics[width=2.56in]{fig4b.eps}}
596:    }
597:    \hbox{
598:        \hbox{\includegraphics[width=2.56in]{fig4c.eps}}
599:        \hspace*{1cm}
600:        \hbox{\includegraphics[width=2.56in]{fig4d.eps}}
601:    }
602:    }
603:    }
604:    \caption{
605: The dynamical maps in the $(a_{\idm{c}},e_{\idm{c}})$-plane in terms of the
606: Spectral Number ($\log SN$, the top-left panel),  $\max \Delta\varpi$  relative
607: to the libration mode $180^{\circ}$ (the top-right panel) and  $\max
608: e_{\idm{b,c}}$ (the bottom panels) for the two-planet edge-on coplanar Newtonian
609: best-fit model of the \stara{} system (Table~1).  Colors used in the $\log SN$
610: map classify the orbits --- black indicates   quasi-periodic regular
611: configurations while yellow strongly chaotic ones.  The large crossed circles
612: mark the parameters of the fit.  The low-order MMRs of the planets b and c are
613: labeled.  The integrations are conducted over  $\sim 10^4 P_{\idm{c}}$.  The
614: resolution is $500\times120$ data points. 
615: }
616: \label{fig:fig4}%
617: \end{figure*}
618: 
619: \begin{figure}
620:    \hbox{\includegraphics[width=3.3in]{fig5.eps}}
621:    \caption{
622:    Synthetic RV curve of the self-consistent best fit to the two-planet model.
623:    The osculating elements at the epoch of the first observation are given in
624:    Table~1.  This fit yields $\Chi \sim 1.082$ and an rms $\sim 8.15$~m/s.
625:    Observations from Naef et al. (2004),  Butler et al. (2006) and Wittenmyer et
626:    al. (2007) are marked  with  red, blue and green circles, accordingly.
627:    }
628: \label{fig:fig5}
629: \end{figure}
630: 
631: 
632: \begin{figure*}
633:    \centerline{
634:    \vbox{
635:    \hbox{
636:        \hbox{\includegraphics[width=2.56in]{fig6a.eps}}
637:        \hspace*{1cm}
638:        \hbox{\includegraphics[width=2.56in]{fig6b.eps}}
639:    }
640:    \hbox{
641:        \hbox{\includegraphics[width=2.56in]{fig6c.eps}}
642:        \hspace*{1cm}
643:        \hbox{\includegraphics[width=2.56in]{fig6d.eps}}
644:    }
645:    }
646:    }
647:    \caption{
648:      The dynamical maps in the $(a_{\idm{c}},e_{\idm{c}})$-plane in terms of the
649: Spectral Number ($\log SN$, the left panels), and $\max e_{\idm{c}}$ (the right
650: panels)  for the two-planet coplanar Newtonian models of the \stara{} system and
651: different inclinations of the orbital plane.  Colors used in the $\log SN$ map
652: classify the orbits --- black indicates quasi-periodic regular configurations
653: while yellow strongly chaotic ones. The maps are computed for the following
654: elements of the Jovian planets in terms of
655: tuples                                                   $(m~[m_{\idm{J}}],
656: a~[\mbox{AU}], e, \omega~[\mbox{deg}], {\cal M}(t_0)~[\mbox{deg}])$. The top row
657: is for $i=60^{\circ}$ and the osculating elements (5.747,    2.864,    0.359,  
658: 22.271,  330.378) for the inner planet,  (7.343,    8.675,    0.171,  190.065,  
659: 75.540) for the outer planet and   the velocity offsets (-43.458, -46.700,
660: -22.780)~[m/s]. The fit has $\Chi\sim 1.082$ and rms $\sim 8.15$~m/s. The bottom
661: row is for $i=30^{\circ}$ and the elements  (9.978, 2.868, 0.357, 22.498,
662: 330.135) of the inner planet,  (8.581, 7.974, 0.142, 194.823, 58.154) of the
663: outer planet and the velocity offsets (-29.237, -32.485, -8.585)~m/s, $\Chi\sim
664: 1.080$ and rms $\sim 8.13$~m/s.  The large crossed circles mark the parameters
665: of the best fits.  The low-order MMRs of the planets b and c are labeled. The
666: $SN$  integrations are conducted over $\sim 10^4 P_{\idm{c}}$. The resolution is
667: $500\times120$ data points. Compare with Fig.~\ref{fig:fig4}  for the edge-on
668: system.
669:    }
670: \label{fig:fig6}%
671: \end{figure*}
672: 
673: Still, the parameters of the outer planet are not well constrained and it is
674: clear that many years of new observations are required to fix the elements of
675: the putative outer companion. To simulate the future RV analysis of the system,
676: we prepared two synthetic sets of observations, choosing the elements of Table~1
677: as the parameters of the ``real'' system. We computed the synthetic RV for that
678: system and added Gaussian noise to these data with parameters as from the real
679: observations. Next, we selected the data points sampling them with a similar
680: frequency to that in the real  measurements. We analyzed two sets: one extended
681: over +1000~days and the second one extended over +2000~days after the last
682: moment in the real observations.  Next, for both synthetic data sets
683: (Fig.~\ref{fig:fig7}), we scanned the $\Chi$ space with the algorithm~II in
684: order to resolve the best-fit system, moreover, in this experiment, no stability
685: constraints were imposed. The results are shown in two panels in
686: Fig.~\ref{fig:fig8}. Curiously, after the additional $\sim 3$~yr observations,
687: the parameters of the outer planet cannot be still determined without doubt.
688: After $\sim 6$~yr, the space of $\Chi$ shrinks significantly and we may expect
689: that after such a time, the semi-major axis of the outer companion can be fixed
690: with $1\sigma$ range of $\sim 1$~AU.
691: 
692: \begin{figure*}
693:    \centerline{
694:    \hbox{
695:        \hbox{\includegraphics[width=3.3in]{fig7a.eps}}
696:        \hspace*{1cm}
697:        \hbox{\includegraphics[width=3.3in]{fig7b.eps}}
698:    }
699:    }
700:    \caption{
701:  Synthetic RV data and the RV curve of the corresponding best fit to the
702: two-planet model.  The left panel is for the time series extended over 1000~days
703: after the last real observation, the right panel is for the synthesized data
704: over 2000~days after the last real data point. Sampling frequency and errors of
705: the artificial observations are similar to those of the real data. 
706: }
707: \label{fig:fig7}
708: \end{figure*}
709: 
710: 
711: \begin{figure*}
712:    \centerline{
713:    \hbox{
714:        \hbox{\includegraphics[width=2.56in]{fig8a.eps}}
715:        \hspace*{1cm}
716:        \hbox{\includegraphics[width=2.56in]{fig8b.eps}}
717:    }
718:    }
719:    \caption{
720:    The topology of $\Chi$ in the $(a_c,e_c)$-plane of the best fits to the
721: synthetic data sets shown in Fig.~\ref{fig:fig7}. The left panel is for
722: artificial observations spanning additional 1000~days after the date of the last
723: real data point, the right panel is for the additional 2000~days period.
724: Contours mark $1\sigma,2\sigma$ and $3\sigma$-levels of $\Chi$. Thick lines
725: (white in the left panel, black in the right panel) are for the mass limits of
726: $14~\mbox{m}_{\idm{J}}$ and $80~\mbox{m}_{\idm{J}}$, respectively. Circle marks
727: the position of the nominal best-fit  in Table~1 (the ``real'' system).
728: }
729: \label{fig:fig8}
730: \end{figure*}
731: 
732: %
733: \section{Secular dynamics of the best-fit system}
734: %______________________________________________________________________________
735: %
736: Although the RV data permit for different dynamical states of the system, we can
737: analyse the secular evolution of the best fit coplanar and edge-on configuration
738: (Table~1) and the solutions in its neighborhood. Due to the significant
739: uncertainty of the orbital elements of outer planet, such analysis cannot
740: provide exhaustive conclusions but we can obtain some qualitative view of the
741: secular behaviour of the system. It can be easily extended when more RV data
742: will be available. 
743: 
744: As we have demonstrated, the parameters of the inner planet as well as the
745: eccentricity and orbital  phase of the putative outer companion can be
746: relatively well constrained. The most unconstrained parameter in the edge-on, 
747: coplanar system is $a_{\idm{c}}$. Yet we should remember that the observations 
748: constrain very roughly the inclinations; and we did not include the 
749: longitudes of nodes as free parameters in the model, keeping the planets  in
750: coplanar orbits. Still, we are in relatively a good situation because the 
751: careful analysis of the RV data makes it possible to reduce the dimension of the
752: phase space. The dynamical map in Fig.~\ref{fig:fig4} covering the shallow
753: minimum of $\Chi$ and the neighborhood of the formal best fit reveals its 
754: proximity to the 11:2~MMR. Several authors argue that in similar cases, 
755: \citep{Libert2007,Libert2005,Gallardo2005} in the first approximation  the
756: near-resonance effects have a small or negligible influence on the secular
757: dynamics.  Hence, neglecting the effects of the MMRs, we can recover its basic
758: properties with the help of the recent quasi-analytical averaging method by
759: \cite{Michtchenko2004}. It generalizes the classic  Laplace-Lagrange (L-L)
760: theory \citep{Dermott1999} or a more recent work by \cite{Lee2003}.  It relies
761: on a semi-numerical averaging of the perturbing Hamiltonian and makes it
762: possible  to avoid any series expansions when recovering the secular variations
763: of the orbital elements.   Then one obtains a very accurate secular model which
764: is valid up to large eccentricities by far beyond the limits of L-L classical
765: theory. The details of this  powerful  approach can be found in
766: \citep{Michtchenko2004,Michtchenko2006}. 
767: 
768: To study the secular dynamics, we follow \cite{Michtchenko2004} and describe the
769: system in terms of the canonical Poincar\'e variables. First, we obtain the
770: time-average of the canonical elements derived from the numerical integration of
771: the full equations of motion with the initial condition in Table~1. The mean
772: canonical semi-major axes of the system are then $\mean{a_{\idm{b}}} \sim
773: 2.868$~AU and $\mean{a_{\idm{c}}} \sim 8.891$~AU.  The eccentricities are 
774: $\mean{e_{\idm{b}}}\sim 0.356$ and  $\mean{e_{\idm{c}}}\sim 0.191$. The initial
775: state of the system is also characterized by the difference of arguments of
776: periastron, $\Delta \varpi = \varpi_{\idm{b}}-\varpi_{\idm{c}} \sim 195^{\circ}$
777: with $1\sigma$ uncertainty $\sim 27^{\circ}$ that is estimated from the
778: statistics  of solutions derived for {\em fixed} masses and semi-major
779: axes  (parameters of the secular model), at their respective values
780: in Table~1.
781: 
782: Next, we compute the energy levels of the averaged Hamiltonian of the  system
783: $\mean{H_{\idm{sec}}}$ (Fig.~\ref{fig:fig9}, left panel)  in the characteristic
784: plane ($e_{\idm{b}}\cos\Delta\varpi,e_{\idm{c}}$) where $\Delta\varpi$ is set
785: either to $0^{\circ}$ or $180^{\circ}$. Let us note that $\Delta\varpi$ always
786: passes through these values during the orbital evolution related to two
787: distinct  libration modes of $\Delta\varpi$ in the coplanar system
788: \citep{Michtchenko2004}. The energy level of the nominal system is marked with
789: red line. The black thick lines mark the exact positions of libration centers 
790: (i.e., periodic solutions) of $\Delta\varpi$ around  $0^{\circ}$  (mode~I) and
791: $180^{\circ}$ (mode~II), respectively. A green part of the libration curve (on
792: the right half-plane of the energy diagram) is for the nonlinear secular
793: resonance of the system. The black dashed lines mark different levels of the
794: total angular momentum (expressed by the {\em Angular Momentum Deficit}, $AMD$).
795: The nominal system (marked with filled dots) is found in the region of the
796: anti-aligned apsides (close to the thin curve representing mode~II) with
797: $e_{\idm{c}}$ in the range $\sim (0.1,0.2)$. 
798: 
799: To illustrate the secular dynamics in more detail, we computed the phase
800: diagrams of the system in the characteristic plane of $(e_{\idm{b}}
801: \cos\Delta\varpi,e_{\idm{b}}\sin\Delta\varpi)$ (the top-left panel in
802: Fig~\ref{fig:fig9}) and  in the $(e_{\idm{c}}
803: \cos\Delta\varpi,e_{\idm{c}}\sin\Delta\varpi)$-plane (the bottom-left panel in
804: Fig.~\ref{fig:fig9}). The red curves mark the oscillation around  libration
805: mode~I, blue curves are for  libration mode~II and black curves mark the
806: circulation of $\Delta\varpi$. The nominal system (marked with filled circles)
807: lies close to the center of mode~II with a relatively large amplitude of
808: $e_{\idm{b}} \sim 0.5$.  The librations of $\Delta\varpi$ are confined to small
809: semi-amplitudes $\sim 30^{\circ}$. The results of the secular theory are in an
810: excellent accord with the behavior of the full unaveraged system. Its dynamical
811: maps are shown in Fig.~\ref{fig:fig4}. The dynamical map of  $\max \Delta
812: \varpi$ (i.e., the maximal semi-amplitude of $\Delta \varpi$ attained during the
813: integration time) reveals a similar semi-amplitude of librations $\sim
814: 30^{\circ}$ around the center of $180^{\circ}$ (and of course, the presence of
815: mode~II). This zone persists over wide ranges  of $(a_{\idm{c}},e_{\idm{c}})$.
816: In fact, almost the entire stable region extending at least up to 11~AU is
817: spanned by configurations involved in mode~II librations.   The error bars in
818: plots of Fig.~\ref{fig:fig9} are derived from formal estimates of $1\sigma$
819: errors of the parameters in Table~1 (we estimate the error of $\Delta\varpi \sim
820: 27^{\circ}$; according with the analysis of stability, we adopted  the
821: uncertainty of $e_c$ as 0.15). The errors are significant but still, in the
822: statistical sense, the libration modes would mostly not change  within the error
823: ranges. This behaviour is confirmed in Fig.~\ref{fig:fig10} showing
824: $\Delta\varpi$ for each individual best-fit found   (see the top-left panel of
825: Fig.~\ref{fig:fig3}). Simultaneously, physically stable systems are close to
826: quasi-periodic configurations what follows from the comparison of the  $\max e$
827: and $\max \Delta\varpi$ indicators.   Both of them, as indicators of geometrical
828: or physical behaviour of the system in short-time scale, appear tightly
829: corelated to the $\log SN$ as the formal measure of the system stability. 
830: 
831: \begin{figure*}
832:    \centerline{
833:    \hbox{
834:        \vspace*{-30mm}
835:    \hbox{
836:        \hbox{\includegraphics[width=4.4in]{fig9a.eps}}
837:    }
838:    \hspace*{0mm}
839:    \vbox{
840:        \hbox{\includegraphics[width=2.0in]{fig9b.eps}}
841:        \vspace*{-4mm}
842:        \hbox{\includegraphics[width=2.0in]{fig9c.eps}}
843:    }
844:    }
845:    }
846:    \caption{
847:   The secular evolution of the best-fit configuration (see Table~1). The
848: left
849:   panel is  for the energy levels of the averaged secular Hamiltonian
850:   $\mean{H}_{\idm{sec}}$ (blue, continuous curves) and the $AMD$ levels (black,
851:   dashed lines) on the representative plane of initial conditions.  The energy
852:   decreases  and $AMD$ increases  with increasing $e_{\idm{c}}$. The sign of
853:   $e_{\idm{b}}\cos\Delta\varpi$ is for initial conditions with
854:   $\Delta\varpi=0^{\circ}$ (+) or $\Delta\varpi=180^{\circ}$ (-), respectively.
855:   The level curves are computed for $m_{\idm{b}}/m_{\idm{c}}=0.648$ and
856:   $a_{\idm{b}}/a_{\idm{c}}=0.3226$.  The red thick curve marks the energy level
857:   of the nominal system, and filled circles indicate actual variation of the 
858:   eccentricities (through the angular momentum integral). The black thick lines
859:   mark the centers of librations of $\Delta\varpi$ around  $0^{\circ}$ 
860:   (mode~I) or $180^{\circ}$ (mode~II), respectively.  A green region (on the
861:   right, for positive $e_{\idm{b}}\cos\Delta\varpi$) is for the nonlinear
862:   secular resonance.  Panels in the left column are for the secular phase space of
863:   the planets~b and c at secular energy $\mean{H}_{\idm{sec}}$ of the nominal
864:   system in the  ($e_{\idm{b}}\cos \Delta\varpi, e_{\idm{b}}\sin
865:   \Delta\varpi)$-plane (the right-top panel) and   ($e_{\idm{c}}\cos \Delta\varpi,
866:   e_{\idm{c}}\sin \Delta\varpi)$-plane (the right-bottom panel), respectively. Two
867:   libration modes of $\Delta\varpi$ about the centers $\Delta\varpi = 0^{\circ}$
868:   (mode~I, red curves) and $\Delta\varpi = 180^{\circ}$ (mode II, blue curves)
869:   are labeled. Thick black lines are for the transition between the libration and
870:   circulation of $\Delta\varpi$. Initial mean parameters of the nominal systems
871:   are marked with filled circles. For a reference, we plot formal
872:   error bars derived from $1\sigma$ errors of the best fit parameters 
873:   in Table~1.
874:    }
875: \label{fig:fig9}%
876: \end{figure*}
877: 
878: 
879: \begin{figure}
880:    \centerline{\hbox{\includegraphics[width=2.7in]{fig10.eps}}}
881:    \caption{
882: The dynamical map in the $(a_{\idm{c}},e_{\idm{c}})$-plane in terms  of the 
883: $\max \Delta\varpi$ indicator for edge-on, coplanar system and assemble of fits
884: shown in Fig.~\ref{fig:fig3}, the top-left panel. The large crossed circles mark the
885: parameters of the  best fit solution in Table~1.  The low-order MMRs of the
886: planets b and c are labeled.  The integrations are conducted over  $\sim 10^4
887: P_{\idm{c}}$. 
888: }
889: \label{fig:fig10}
890: \end{figure}
891: 
892: 
893: \section{Conclusions}
894: %______________________________________________________________________________
895: %
896: 
897: According to the results from the adaptive-optics imaging
898: \citep{Luhman2002,Patience2002},  there is no stellar-mass object in the 14~Her
899: neighborhood beyond $\sim 12$~AU.  This finding supports the hypothesis that the
900: RV trend may be attributed to a massive planet~c.  The orbital period of the
901: putative companion to the known Jovian planet~b cannot yet be  very well
902: constrained. Nevertheless the available data already reveal a very shallow
903: minimum od $\Chi$ in the ($a_{\idm{c}},e_{\idm{c}})$-plane of the initial
904: osculating elements. The minimum persists for reasonable combinations of the
905: parent star mass and the inclination of the system. Its dynamical character
906: strongly depends on that parameter influencing the planetary masses. Quite
907: surprisingly, for small  inclinations the mass hierarchy is reversed and the
908: inner planet becomes more massive than its distant companion. Also the positions
909: of the nearby MMRs are significantly affected. Depending on the inclination, the
910: system may be  locked in the low-order 9:2, 5:1 or 6:1~MMR. We can also conclude
911: that the kinematic model of  the RV is already not adequate for the
912: characterization of the system and further analysis of the RV would be done best
913: within the self-consistent Newtonian model.
914: 
915: The dynamical analysis enables us to derive some limits on the orbits elements
916: and the stability of the system. No stable configurations are found with a
917: period ratio smaller than $\sim 3$. This limit is shifted towards larger values
918: when the inclination grows. For the inclination of $30^{\circ}$, the best fit
919: solutions are localized in a strongly chaotic and unstable zone. This
920: constitutes a dynamically derived argument against a small inclination of the 
921: system. It is likely larger than $30^{\circ}$-$40^{\circ}$. Allowing for some 
922: speculation, the presence of the best-fits in a robustly stable zone supports 
923: the hypothesis of highly inclined two-planet configuration in the 14~Her
924: system.  Our attempts to fit one more planet to the RV data that could explain
925: the scatter  of the data points close to the minima of the RV curve
926: (Fig.~\ref{fig:fig5}), did not lead to a meaningful improvement of the fit. The
927: third planet would  have a weak RV signal at the level of a few m/s and a
928: short-period eccentric orbit. Finally, the analysis of all available RV data of
929: 14~Her enables us to revise and extend some of the conclusions from
930: \cite{Gozdziewski2006a} --- an isolated and very well defined best-fit solution
931: cannot yet be  found to properly describe the orbital configuration of 14 Her. 
932: 
933: {\bf Acknowledgements.}
934: We thank Eric Ford for a critical review, suggestions and comments 
935: that greatly improved the manuscript. 
936: This work would not be possible without the access to the precision RV data
937: which the discovery teams make available to the community. This work is
938: supported  by the Polish Ministry of Science and Education through grant
939: 1P03D-021-29. M.K. is also supported by the Polish Ministry of Science and
940: Education through grant N203 005 32/0449.
941: 
942: %
943: \bibliographystyle{mn2e}
944: \bibliography{ms}
945: \end{document}
946: