1: \documentclass[useAMS,usenatbib,usegraphicx]{mn2e}
2: \usepackage{times}
3:
4:
5: \bibpunct{(}{)}{;}{a}{}{,}
6:
7: \begin{document}
8:
9: \title[]{TIRAVEL - Template Independent RAdial VELocity measurement}
10:
11: \author[S. Zucker \& T. Mazeh]{
12: S.~Zucker$^1$ and T.~Mazeh$^2$ \\
13: $^1$Department of Geophysics and Planetary Sciences and Wise
14: Observatory, Raymond and Beverly Sackler Faculty of Exact Sciences, \\
15: Tel Aviv University, Tel Aviv, Israel \\
16: $^2$School of Physics and Astronomy, Raymond and Beverly
17: Sackler Faculty of Exact Sciences, \\
18: Tel Aviv University, Tel Aviv, Israel}
19:
20: \maketitle
21: \begin{abstract}
22: We propose a new approach to measure differential radial velocities,
23: mainly for single-lined spectroscopic binaries. The proposed procedure
24: -- {\sc TIRAVEL} (Template Independent RAdial VELocities) -- does not
25: rely on a prior theoretical or observed template, but instead looks
26: for a set of relative Doppler shifts that simultaneously optimizes the
27: alignment of all the observed spectra. We suggest a simple measure to
28: quantify this overall alignment and use its maximum to measure the
29: relative radial velocities. As a demonstration, we apply {\sc
30: TIRAVEL} to the observed spectra of three known spectroscopic
31: binaries, and show that in two cases {\sc TIRAVEL} performs as good as
32: the commonly used approach, while in one case {\sc TIRAVEL} yielded a
33: better orbital solution.
34: %
35: %Once
36: %generalized to multi-order spectra, {\sc TIRAVEL} may improve the
37: %precise measurement of radial velocities of late-type stars, in the
38: %search of extrasolar planets in particular.
39: %
40: %
41: \end{abstract}
42: \begin{keywords}
43: methods: data analysis --
44: methods: statistical --
45: techniques: radial velocities --
46: techniques: spectroscopic --
47: binaries: spectroscopic
48: \end{keywords}
49:
50: \section{Introduction}
51: \label{intro}
52:
53: Since the seminal works of \citet{Sim1974} and \citet{TonDav1979}, the
54: cross-correlation technique to measure astronomical Doppler shifts has
55: become extremely popular. The advent of digitized spectra and
56: computers rendered it the preferred method in astronomical fields
57: that require measurement of radial velocities (RVs) from observed
58: spectra, ranging from binary and multiple stellar systems to
59: cosmology. In recent years, improvements in the precision of RVs
60: measured through cross-correlation opened the way to the detection of many
61: extrasolar planets \citep[e.g.,][]{MayQue1995}.
62:
63: The cross-correlation technique is conceptually simple and can be
64: presented in an intuitive manner. It searches for the velocity shift
65: of the observed stellar spectrum which maximizes the correlation with
66: a predetermined template. The template is assumed to adequately
67: represent the object spectrum with no Doppler shift. The properties
68: of the cross-correlation technique have been studied extensively, in
69: an effort to improve its precision and overcome its few limitations,
70: and various methods have been suggested to estimate its precision
71: \citep[e.g.,][]{TonDav1979,Con1985,MurHea1991,Bouetal2001}.
72:
73: The precision of the individual RVs depends on several factors, mainly
74: on the amount of spectral information and the signal-to-noise ratio
75: ($S/N$) of the spectra. Another crucial factor is the quality of the
76: template spectrum, and the extent to which it represents the actual
77: stellar spectrum. Common practice is to use a theoretically calculated
78: spectrum as template \citep[e.g.,][]{Latetal2002}, a high $S/N$
79: exposure of another similar star or the observed star itself
80: \citep[e.g.,][]{Howetal1997}.
81:
82: The theoretical synthetic spectrum might have some systematic
83: differences relative to the actual stellar
84: spectrum. This happens not only because the theory of stellar
85: atmospheres is still not perfect and the spectral line lists are not
86: complete, but also because every star is somewhat different from other
87: stars, by its distinct abundances of the various atoms and ions in
88: particular. Similar problems occur when an observed spectrum is used
89: as template. An observed spectrum, either of the same star or of a
90: different star, introduces random noise to the template, thus
91: alienating it again from the true stellar spectrum.
92:
93: We present here a new approach, {\sc TIRAVEL}, to derive the relative
94: velocity of stellar spectra without using any template. The approach
95: is based on the simple idea that when two spectra of the same star are
96: available, one can use cross-correlation to measure a {\it relative}
97: Doppler shift between the two spectra, each spectrum effectively
98: acting as a template for the other one. {\sc TIRAVEL} is a
99: generalization of this approach for more than two spectra. We will
100: show that for large number of spectra, this approach is equivalent to
101: having a high-$S/N$ effective template. Section \ref{tiravel}
102: introduces the approach, while Section \ref{realdata} applies it to
103: three real test cases. We discuss the test results and the
104: applicability of {\sc TIRAVEL} in Section \ref{discussion}.
105:
106: \section{TIRAVEL}
107: \label{tiravel}
108:
109: {\sc TIRAVEL} is basically a multi-spectral generalization of the
110: conventional cross-correlation technique. When we cross correlate two
111: spectra we scan a set of potential shifts between the spectra and
112: score them by the degree of mutual similarity between the two shifted
113: spectra. This score is based on the correlation coefficient, which has
114: been shown to be an optimal similarity measure under various
115: assumptions \citep[e.g.,][]{Zuc2003}.
116:
117: Suppose we wish to estimate the relative Doppler shift between two
118: spectra $f_1(n)$ and $f_2(n)$. The two spectra are described as
119: functions of the pixel number -- $n$, where $n = A\ln\lambda + B.$
120: Thus, the Doppler shift results in a uniform linear shift of the
121: spectrum \citep{TonDav1979}. For simplicity we assume the spectra were
122: already 'continuum-subtracted' and normalized, i.e.:
123: \begin{eqnarray}
124: \sum_n f_i(n) &= &0 \ , \\
125: \frac{1}{N} \sum_n f_i^2(n) &= &1 \ .
126: \end{eqnarray}
127: The cross-correlation function is then defined as:
128: \[
129: R_{12}(s) = \frac{1}{N} \sum_n f_1(n) f_2(n-s) \ .
130: \]
131: This expression is essentially the correlation coefficient between the
132: two spectra after they have been shifted according to the trial
133: relative shift $s$. A higher value of $R_{12}(s)$ corresponds to
134: better agreement between the two sequences. Furthermore, it is well
135: known that $|R_{12}(s)| \le 1$.
136:
137: {\sc TIRAVEL} generalizes the cross-correlation procedure to the case
138: of a set of $K>2$ spectra by considering the correlation matrix
139: $R_{ij}(s_1,s_2,\cdots,s_{\scriptscriptstyle K})$. The entry $R_{ij}$
140: is the cross-correlation between the $i$-th and $j$-th spectrum, as a
141: function of their relative shift $s_i-s_j$. In order to measure the
142: overall agreement for a given set of shifts we propose to use
143: $\lambda_M$ -- the largest eigenvalue of the matrix
144: $R_{ij}(s_1,s_2,\cdots,s_{\scriptscriptstyle K})$. We will show that
145: this estimate reduces to the correlation coefficient for $K=2$
146:
147: The rationale behind this measure is simple and intuitive. At the
148: correct alignment, all the spectra are supposed to be easily modeled
149: by one {\it principal component}, or template. The highest eigenvalue
150: of the correlation matrix measures the degree to which we can describe
151: the spectra by one single principal component, as is well known from
152: Principal Component Analysis (PCA) theory
153: \citep[e.g.,][]{MurHec1987}. The eigenvector of $R_{ij}$, corresponding
154: to the eigenvalue $\lambda_M$, can be used to produce the principal
155: component itself, or in other words, the effective template {\sc
156: TIRAVEL} 'used' in order to produce the RVs. The elements of this
157: eigenvector, with proper normalization, are the optimal weights with
158: which the indvidual spectra should be summed to produce this
159: 'template'. This is especially important in cases where the $S/N$
160: ratios are varying considerably among the observed spectra.
161:
162: The value of the pairwise correlation coefficient can be interpreted
163: easily and intuitively, since values closer to unity mean better
164: agreement between the two sequences, and values close to zero mean
165: poorer agreement. We suggest to transform $\lambda_M$ to obtain
166: a similar behaviour:
167: \[
168: \rho =\frac{\lambda_M-1}{K-1} \ .
169: \]
170:
171: {\sc TIRAVEL} essentially seeks the maximum of $\rho$ as a function of
172: the $K$ relative shifts $(s_1,s_2,\cdots,s_{\scriptscriptstyle K})$.
173: Each trial set of the $K$ relative shifts requires the calculation of
174: the corresponding correlation matrix $R_{ij}$. Therefore, an important
175: preliminary step in the computation should be the calculation of the
176: $K(K-1)/2$ cross-correlation functions, corresponding to all the pairs
177: of different spectra. The matrix elements $R_{ij}(s_i-s_j)$ are then
178: simply sampled from the pre-calculated cross-correlation functions.
179:
180: Note also that the elements of the correlation matrix $R_{ij}$ depend
181: only on pairwise differences of shifts, i.e., relative shifts. It is
182: therefore insensitive to zero-point shifts which are common to all the
183: observed spectra. In that respect, the shifts derived by {\sc TIRAVEL}
184: are only relative shifts, with effective $K-1$ degrees of freedom
185: instead of $K$.
186:
187: After optimizing over $(s_1,s_2,\cdots,s_{\scriptscriptstyle K})$,
188: which can be done by any optimization procedure
189: \citep[e.g.,][]{Preetal1992}, we obtained the final velocities by
190: fitting parabolas to the peak, similarly to the common practice in
191: regular cross-correlation.
192:
193: Since the new procedure yields also an effective 'template', produced
194: by summing all the individual exposures at the right shifts with
195: optimal weights, we expect the quality of this effective template to
196: improve as the number of observed spectra increases. Better template
197: should yield more accurate individual velocities. We therefore expect
198: the precision of the {\sc TIRAVEL} velocities to improve as a function
199: of the number of spectra. To test this we ran {\sc TIRAVEL} on
200: different sets of simulated spectra and checked the resulting
201: velocities. We generated $K$ spectra by using a synthetic spectrum of
202: a G star to which we added normal noise at a specified $S/N$. No shift
203: was implemented in producing the spectra, and therefore the scatter of
204: the velocities derived by {\sc TIRAVEL} can serve as a measure of the
205: precision of the approach. We repeated the procedure $50$ times for
206: each $S/N$ and $K$.
207:
208: Figure \ref{Kplot} displays the results of these simulations. The
209: plot shows the mean RMS of the velocities obtained by {\sc TIRAVEL} as
210: a function of the number of spectra analysed, for three different
211: $S/N$-ratios. As expected, the RMS converges to a constant value, when
212: the 'effective template' becomes similar to the true template. It
213: seems that the convergence rate is not very sensitive to the $S/N$,
214: and that when $10$--$15$ spectra are available, the ultimate
215: performance is already achieved (assuming they all share about the
216: same $S/N$). This behaviour is easily understood, since once the
217: template is significantly less noisier than the individual spectra,
218: the main factor which determines the velocity precision is the $S/N$
219: of the observed spectrum. This happens already when a moderate number
220: of spectra are co-added to produce the template.
221:
222: \begin{figure}
223: \resizebox{\hsize}{!}{\includegraphics{fig1.eps}}
224: \caption{Results of applying {\sc TIRAVEL} to sets of $K$ simulated
225: spectra with zero Doppler shift. The plots present the mean RMS of the
226: velocities produced by {\sc TIRAVEL}.}
227: \label{Kplot}
228: \end{figure}
229:
230: The simulations also indicate the computational burden of {\sc
231: TIRAVEL}. Using a very na\"{\i}ve optimization scheme, on average
232: about $10$ iterations were needed to reach the stopping criterion of
233: less than $0.01\,\mathrm{km\ s}^{-1}$ RMS change in the velocities,
234: for $K=30$. We wrote our code in {\sc MATLAB} and ran it on a Pentium
235: Pro Linux machine, where those $10$ iterations, including the
236: pre-calculation step, lasted about $20$ seconds.
237:
238: \section{Real Test Cases}
239: \label{realdata}
240:
241: We chose to demonstrate the potential of the new approach on three
242: single-lined spectroscopic binaries (SB1s) which were already solved
243: and published \citep{Latetal2002} as part of a survey of $171$
244: high-proper-motion SB1s. The radial velocities in \citet{Latetal2002}
245: were derived from spectra obtained using the CfA Digital Speedometers
246: \citep{Lat1985,Lat1992}. In order to obtain the radial velocities
247: \citeauthor{Latetal2002} used an extensive grid of spectra computed
248: using the model atmospheres of the {\sc ATLAS9} code, developed by
249: Kurucz \citetext{Morse \& Kurucz, in preparation}.
250:
251: The spectra we used for the {\sc TIRAVEL} analysis were the same
252: observed spectra that \citeauthor{Latetal2002} used for their
253: analysis. Using the velocities produced by {\sc TIRAVEL} we have
254: recalculated a least-square best-fitting orbital solution. The RMS of
255: the residuals of the new orbit serves as a criterion to compare with
256: the previously published orbit. Furthermore, the effective template
257: derived by {\sc TIRAVEL} indicates the amount of spectral information
258: that can be extracted from the spectra without using a template.
259:
260: \subsection{G72--59}
261: \label{G72-59}
262:
263: \citet{Latetal2002} solved G72--59 with 32 spectra
264: and derived a period of about $88$ days with a small
265: eccentricity. Figure \ref{G72-59_specs} shows three of the 32 spectra,
266: to demonstrate their S/N. Table \ref{G72-59_table} lists both the
267: orbital elements that appear in \citet{Latetal2002} and the orbital
268: elements obtained from the {\sc TIRAVEL} velocities. Note that the {\sc
269: TIRAVEL} analysis reveals no information about {\it absolute} radial
270: velocities, and the value of the center-of-mass velocity obtained in
271: analysing {\sc TIRAVEL} velocities is not very
272: informative. Nevertheless we list it for completeness.
273:
274: Table \ref{G72-59_table} shows that the two solutions are
275: consistent. The residual RMS of the {\sc TIRAVEL} solution is somewhat
276: larger than the old one, and so are the errors of the elements. This
277: is one of the cases where the {\sc TIRAVEL} solution is a little worse
278: than \citet{Latetal2002} solution. The two orbital solutions can be
279: visually compared in Figure \ref{G72-59_orbits}. Indeed, the
280: differences between the two orbits are completely negligible. Figure
281: \ref{G72-59_temps} presents the new effective template that is built
282: by {\sc TIRAVEL}, together with the synthetic template used in
283: the original analysis.
284:
285: The synthetic template shown in Figure \ref{G72-59_temps} was shifted
286: according to the difference between the two derived center-of-mass
287: velocities, in order to match the derived {\sc TIRAVEL} template. The
288: similarity is striking, although the synthetic template evidently has
289: a much higher $S/N$ than the derived one. It is somewhat surprising
290: and gratifying that the {\sc TIRAVEL} results are almost as good as
291: the results of the calculated spectrum, despite the fact that the
292: effective spectrum of {\sc TIRAVEL} looks so much noisier.
293:
294:
295: \begin{figure}
296: \resizebox{\hsize}{!}{\includegraphics{fig2.eps}}
297: \caption{Sample spectra of G72--59. The times of the presented exposures are
298: $\rmn{JD}=2447053.7466$ (upper panel), $\rmn{JD}=2447163.5952$ (middle panel)
299: and $\rmn{JD}=2449678.7457$ (lower panel).}
300: \label{G72-59_specs}
301: \end{figure}
302:
303: \begin{table}
304: %\begin{minipage}{260mm}
305: \caption{Comparing the two orbital solutions for G72--59}
306: \begin{tabular}{lc*{2}{r@{\,$\pm$\,}l}}
307: Parameter & &
308: \multicolumn{2}{c}{\citealt{Latetal2002}} &
309: \multicolumn{2}{c}{\sc TIRAVEL} \\
310: \hline
311: $P$ & [days] & $87.754$ & $0.026$ & $87.755$ & $0.031$ \\
312: $T$ & [JD] & $2\,447\,701.7$ & $1.4$ & $2\,447\,701.8$ & $1.5$ \\
313: $e$ & & $0.168$ & $0.017$ & $0.169$ & $0.019$ \\
314: $\omega$ & [$\degr$] & $260.8$ & $5.4$ & $261.1$ & $6.1$ \\
315: $K$ & [km\ s$^{-1}$] & $15.33$ & $0.24$ & $15.30$ & $0.28$ \\
316: $V_0$ & [km\ s$^{-1}$] & $11.78$ & $0.18$ & $5.26$ & $0.20$ \\
317: \hline
318: $\sigma_{\scriptscriptstyle O-C}$ & [km\ s$^{-1}$] & \multicolumn{2}{c}{$0.94$} & \multicolumn{2}{c}{0.96}
319: \end{tabular}
320: %\end{minipage}
321: %\medskip
322: %\medskip
323: %\medskip
324: \label{G72-59_table}
325: \end{table}
326:
327: \begin{figure}
328: \resizebox{\hsize}{!}{\includegraphics{fig3.eps}}
329: \caption{Upper panel: The orbital solution of G72--59 using
330: {\sc TIRAVEL} velocities.
331: Lower panel: The old orbital solution by \citet{Latetal2002}.}
332: \label{G72-59_orbits}
333: \end{figure}
334:
335: \begin{figure}
336: \resizebox{\hsize}{!}{\includegraphics{fig4.eps}}
337: \caption{Upper panel: The effective template spectrum obtained by
338: {\sc TIRAVEL} for
339: the spectra of G72--59. Lower panel: the Kurucz spectrum used as
340: template by \citet{Latetal2002}, shifted to match the {\sc TIRAVEL}
341: effective template.}
342: \label{G72-59_temps}
343: \end{figure}
344:
345: \subsection{G178--27}
346: \label{G178-27}
347:
348: According to \citet{Latetal2002}, G178--27 has a low amplitude orbit
349: ($K = 3.4$\,km\ s$^{-1}$), with an eccentricity of about $0.4$. Its
350: period is about $81$ days, and the orbital solution was obtained using
351: $31$ exposures. \citeauthor{Latetal2002} derived very low
352: metallicity for G178--27, of $[\rmn{Fe}/\rmn{H}] = -2.0$. Figure
353: \ref{G178-27_specs} presents three sample spectra of G178--27. One
354: can see the relative scarcity of the spectral information compared with
355: G72--59, due to the very low metallicity.
356:
357: The two orbital solutions are listed in Table \ref{G178-27_table}, and
358: shown on Figure \ref{G178-27_orbits}. Once again, the new solution is
359: very similar and consistent with the previous one, while its residuals are
360: slightly smaller. The good choice of synthetic
361: template by \citeauthor{Latetal2002} is clearly seen once the {\sc
362: TIRAVEL} effective template is derived, as shown in Figure
363: \ref{G178-27_temps}.
364:
365: \begin{figure}
366: \resizebox{\hsize}{!}{\includegraphics{fig5.eps}}
367: \caption{Sample spectra of G178--27. The times of the presented exposures are
368: $\rmn{JD}=2446928.9137$ (upper panel), $\rmn{JD}=2447494.0177$ (middle
369: panel) and $\rmn{JD}=2447702.7553$ (lower panel).}
370: \label{G178-27_specs}
371: \end{figure}
372:
373: \begin{table}
374: %\begin{minipage}{260mm}
375: \caption{Comparing the two orbital solutions for G178--27}
376: \begin{tabular}{lc*{2}{r@{\,$\pm$\,}l}}
377: Parameter & &
378: \multicolumn{2}{c}{\citealt{Latetal2002}} &
379: \multicolumn{2}{c}{\sc TIRAVEL} \\
380: \hline
381: $P$ & [days] & $81.18$ & $0.37$ & $81.18$ & $0.35$ \\
382: $T$ & [JD] & $2\,447\,436.5$ & $1.7$ & $2\,447\,436.1$ & $1.8$ \\
383: $e$ & & $0.431$ & $0.046$ & $0.428$ & $0.045$ \\
384: $\omega$ & [$\degr$] & $264.5$ & $8.5$ & $258.8$ & $9.1$ \\
385: $K$ & [km\ s$^{-1}$] & $3.42$ & $0.18$ & $3.44$ & $0.18$ \\
386: $V_0$ & [km\ s$^{-1}$] & $-180.30$ & $0.12$ & $-1.11$ & $0.13$ \\
387: \hline
388: $\sigma_{\scriptscriptstyle O-C}$ & [km\ s$^{-1}$] & \multicolumn{2}{c}{$0.61$} & \multicolumn{2}{c}{0.56}
389: \end{tabular}
390: %\end{minipage}
391: %\medskip
392: %\medskip
393: %\medskip
394: \label{G178-27_table}
395: \end{table}
396:
397: \begin{figure}
398: \resizebox{\hsize}{!}{\includegraphics{fig6.eps}}
399: \caption{Upper panel: The orbital solution of G178--27 using
400: {\sc TIRAVEL} velocities.
401: Lower panel: The previous orbital solution by \citet{Latetal2002}.}
402: \label{G178-27_orbits}
403: \end{figure}
404:
405: \begin{figure}
406: \resizebox{\hsize}{!}{\includegraphics{fig7.eps}}
407: \caption{Upper panel: The effective template spectrum obtained
408: by {\sc TIRAVEL} for
409: the spectra of G178--27. Lower panel: the Kurucz spectrum used as
410: template by \citet{Latetal2002}, shifted to match the {\sc TIRAVEL}
411: effective template}
412: \label{G178-27_temps}
413: \end{figure}
414:
415: \subsection{G48--54}
416: \label{G48-54}
417:
418: The star G48--54 is the best case considered here to demonstrate the
419: potential of {\sc TIRAVEL}. The previous solution by
420: \citeauthor{Latetal2002} has a period of $22.6$ days and an
421: eccentricity of $0.3$, obtained with $30$ exposures. The star G48--54
422: is the star with the latest spectral type presented in
423: \citet{Latetal2002}, with an estimated temperature of
424: $3750\,\rmn{K}$. Figure \ref{G48-54_specs} presents three sample
425: spectra of G48--54.
426:
427: As can be seen in Table \ref{G48-54_table}, the {\sc TIRAVEL} solution
428: is better than the old published solution. Because of the large RV
429: amplitude, the differences between the two solutions can hardly be
430: seen in the phase-folded RV curve (Figure \ref{G48-54_orbits}). We
431: therefore present in Figure \ref{G48-54_residuals} the residuals after
432: subtracting the orbital solution in the two cases, on the same scale.
433: The improvement is easily seen. Figure \ref{G48-54_temps} suggests the
434: reason for the improvement. The spectrum of this late-type star is
435: more complex than most of the other stars in the sample, because of
436: its lower temperature. We suggest that the spectral modeling of this
437: star is not
438: complete, probably due to the notorious difficulties associated with
439: modeling later-type stars.
440:
441: Close examination of Figures \ref{G48-54_orbits} and
442: \ref{G48-54_residuals} suggests that a single measurement, at a phase
443: of about $0.3$, may be 'pulled' by the secondary spectrum, at a phase
444: where the primary and secondary velocities are very close. This
445: outlying velocity stands out even better in the {\sc TIRAVEL}
446: velocities, because of the smaller RMS. Maybe a somewhat better
447: procedure would exclude this point from the analysis.
448:
449: \begin{figure}
450: \resizebox{\hsize}{!}{\includegraphics{fig8.eps}}
451: \caption{Sample spectra of G48--54. The times of the presented exposures are
452: $\rmn{JD}=2445770.8684$ (upper panel), $\rmn{JD}=2446861.8126$ (middle
453: panel) and $\rmn{JD}=2446959.6461$ (lower panel).}
454: \label{G48-54_specs}
455: \end{figure}
456:
457: \begin{table}
458: %\begin{minipage}{260mm}
459: \caption{Comparing the two orbital solutions for G48--54}
460: \begin{tabular}{lc*{2}{r@{\,$\pm$\,}l}}
461: Parameter & &
462: \multicolumn{2}{c}{\citealt{Latetal2002}} &
463: \multicolumn{2}{c}{\sc TIRAVEL} \\
464: \hline
465: $P$ & [days] & $22.6057$ & $0.0049$ & $22.6020$ & $0.0042$ \\
466: $T$ & [JD] & $2\,446\,729.39$ & $0.16$ & $2\,446\,729.19$ & $0.15$ \\
467: $e$ & & $0.287$ & $0.012$ & $0.257$ & $0.010$ \\
468: $\omega$ & [$\degr$] & $320.4$ & $2.8$ & $316.2$ & $2.6$ \\
469: $K$ & [km\ s$^{-1}$] & $25.74$ & $0.32$ & $25.44$ & $0.27$ \\
470: $V_0$ & [km\ s$^{-1}$] & $0.84$ & $0.24$ & $9.55$ & $0.20$ \\
471: \hline
472: $\sigma_{\scriptscriptstyle O-C}$ & [km\ s$^{-1}$] & \multicolumn{2}{c}{$1.20$} & \multicolumn{2}{c}{0.89}
473: \end{tabular}
474: %\end{minipage}
475: %\medskip
476: %\medskip
477: %\medskip
478: \label{G48-54_table}
479: \end{table}
480:
481: \begin{figure}
482: \resizebox{\hsize}{!}{\includegraphics{fig9.eps}}
483: \caption{Upper panel: The orbital solution of G48--54
484: using {\sc TIRAVEL} velocities.
485: Lower panel: The previous orbital solution by \citet{Latetal2002}.}
486: \label{G48-54_orbits}
487: \end{figure}
488:
489: \begin{figure}
490: \resizebox{\hsize}{!}{\includegraphics{fig10.eps}}
491: \caption{Upper panel: The residual RVs after subtracting the
492: best-fitting orbital solution from the {\sc TIRAVEL} velocities of
493: G48--54. Lower panel: The residual RVs after subtracting the previous
494: best-fitting orbital solution from the velocities published by
495: \citet{Latetal2002}.}
496: \label{G48-54_residuals}
497: \end{figure}
498:
499: \begin{figure}
500: \resizebox{\hsize}{!}{\includegraphics{fig11.eps}}
501: \caption{Upper panel: The effective template spectrum obtained
502: by {\sc TIRAVEL} for
503: the spectra of G48--54. Lower panel: the Kurucz spectrum used as
504: template by \citet{Latetal2002}, shifted to match the {\sc TIRAVEL}
505: effective template}
506: \label{G48-54_temps}
507: \end{figure}
508:
509: \section{Discussion}
510: \label{discussion}
511:
512: We have presented here a new technique to analyse the spectra of SB1
513: systems. {\sc TIRAVEL} is independent of any assumption about the
514: primary spectrum, which is not included in the actual observed
515: data. Instead, {\sc TIRAVEL} makes use of the observed spectra to
516: derive the radial velocities of the primary. This approach assumes
517: that there is only one component in the observed spectra and that they
518: are free of atmospherical or mechanical systematic features. Such
519: features might reduce the efficiency of the technique, in particular
520: compared to the use of theoretically calculated templates.
521:
522: We have applied {\sc TIRAVEL} to three SB1 cases, where a satisfactory
523: solution had already obtained with theoretically calculated
524: templates. {\sc TIRAVEL} improved significantly the quality of one of
525: the three solutions, in terms of the precision of the orbital elements
526: and the RMS residuals. Obviously, the advantages of {\sc TIRAVEL} will
527: be most pronounced when there is no good theoretically calculated
528: template available. In the presented cases, the template grid that
529: \citet{Latetal2002} used was very extensive, and we could not expect
530: to obtain very large improvements, except at the low-temperature
531: frontier of the grid. One immediate implication of {\sc TIRAVEL} is
532: the analysis of the spectra of cataclysmic variables or some T Tauri
533: stars, where the high-temperature accretion disc has an emission
534: spectrum which in not easy to model.
535:
536: The version of {\sc TIRAVEL} we presented here is specifically
537: tailored for SB1s, where there is only one RV to be measured in each
538: exposure, resulting in a simple shift of the spectrum. In the case of
539: double-lined spectroscopic binaries (SB2), a more elaborate scheme is
540: needed, one that has to separate the effects of the orbit on the two
541: component spectra. Such a scheme may make use of ideas related to
542: Doppler tomography \citep{Bagetal1992} or two-dimensional correlation
543: ({\sc TODCOR};\citet{ZucMaz1994}). One possible generalization of {\sc
544: TIRAVEL} may simply apply it iteratively - first solving for the
545: primary component, subtracting the resulting template from the
546: spectra, then applying {\sc TIRAVEL} to the subtracted spectra again
547: to retrieve the information related to the secondary component, and
548: repeat the process iteratively. Similar procedures were applied by
549: \citet{Maretal1998} and \citet{GonLev2006}.
550:
551: Note that spectral disentangling codes, like {\sc KOREL}
552: \citep{Had1995,Had1997} can be applied also to SB1s
553: \citep[e.g.,][]{Saaetal2005}. However, they
554: usually impose an orbital model, treating any deviation from the model
555: as noise. {\sc TIRAVEL} does not incorporate any orbital model (or
556: any other source of RV variation), and produces the best relative RV
557: estimates one can get without a template. Therefore, the {\sc TIRAVEL}
558: approach enables us to detect any additional source for RV variation,
559: a third companion in particular.
560:
561: In order to be applicable to spectra obtained by the modern echelle
562: spectrographs and other multi-order spectra, {\sc TIRAVEL} has to be
563: generalized, probably along lines similar to the ones that were used
564: to generalize {\sc TODCOR} to multi-order spectra
565: \citep{Zuc2003}. This modification was crucial in detecting the planet
566: in HD\,41004 \citep{Zucetal2003,Zucetal2004}.
567:
568: In principle, improvements in RV precision, like the one offered by
569: {\sc TIRAVEL}, may lead to the detection of small finite
570: eccentricities or additional stellar components. One area where the
571: precision of radial velocities and orbital solution is of utmost
572: importance is the search for extrasolar planets, where the orbital RV
573: amplitude is usually extremely small, of the order of tens of meters
574: per second. Effective improvements in the precision of radial
575: velocities may lead to an improved planet-detection capability. In
576: general, the improved precision also means a better utilization of the
577: instrumental potential and thus a more efficient observing time usage.
578:
579: \section*{Acknowledgments}
580:
581: We are deeply indebted to Dave Latham for his most useful
582: assistance in accessing and analysing the CfA spectra.
583:
584: \bibliographystyle{mn2e}
585: \bibliography{ref}
586:
587: \end{document}
588:
589:
590:
591:
592: