1: \subsection{{\boldmath $\Dsop$} candidate reconstruction}
2: \label{ssec:evsel}
3: Two decay modes are reconstructed: in both modes the \Dsop decays to $\Dstarp\!\KS$, with the \KS decaying into \pipi and the \Dstarp into $\Dz\!\pip$. The \Dz decays either into $\Km\!\pip$ or $\Km\!\pip\!\pip\!\pim$. In the following, we refer to these two decay modes as $K4\pi$ and $K6\pi$, respectively, where also the charge conjugated states are included.
4:
5: For the \Dz candidate reconstruction, charged kaon and pion candidates are combined to form $\Km\!\pip$ ($\Km\!\pip\!\pip\!\pim$) final states.
6: Fits to the \Dz mass spectra yield a mean value of $1863.6\mevcc$ and a signal
7: resolution of $7.4\mevcc$ ($8.1\mevcc$) for
8: the \modea\ (\modeb) decay mode. The signal region is
9: chosen as a mass window of $\pm18\mevcc$ ($\pm14\mevcc$) centered around the obtained mean value.
10: The \Dz candidates are each combined with an additional charged pion to form \Dstarp candidates.
11: The fits to the \Dstarp-\Dz mass difference yield a mean value of $145.4\mevcc$ and a signal resolution
12: of $0.19\mevcc$ ($0.24\mevcc$) for \modea\ (\modeb). The signal
13: region is chosen as a mass window of $\pm1.5\mevcc$ centered around
14: the derived mean value for both decay modes.
15: Finally, \KS candidates are created from oppositely charged tracks.
16: From the fits to the \KS mass distributions one obtains a mean value of $497.2\mevcc$ and a resolution of $2.5\mevcc$ for both decay modes. For the further analysis we define the \KS signal region as a $\pm10\mevcc$ mass window centered around the signal mean value for both modes.
17:
18: The background of the \KS spectrum is further reduced by restricting the angle between the \KS direction of flight and the line connecting the primary vertex and the \KS vertex to values smaller than $0.15\,\mathrm{rad}$. The \Dstarp candidates are finally combined with the \KS candidates to form \Dsop candidates. In order to suppress combinatorial background, a momentum $p^{*} > 2.7 \gevc$ in the center of mass system (CMS) is required for \Dsop candidates. In addition, this restricts the source of the $\Dsop$ candidates to $\epem \to \ccbar$ continuum production. A kinematic fit is applied to the \Dsop candidates which satisfy the above selection criteria in such a way, that the tracks for each composed particle have the same origin. The position of the \Dsop vertex is required to be consistent with the $\epem$ interaction region. Since the mass difference \deltadso is measured, no mass constraint is applied. The probability of the vertex fit is required to be greater than $0.1\%$.
19: Initially, more than one \Dsop candidate per event is reconstructed. Although the multiple use of tracks within one reconstructed decay tree is excluded, there might be multiple \Dsop candidates sharing the same daughter candidate. After applying all
20: selection criteria, the average multiplicity of \Dsop candidates per event is $1.008$ and $1.02$ for the two decay modes.
21: The selection efficiency is $16\%$ for the \modea\ decay mode and $11\%$ for the \modeb\ mode.
22:
23:
24: The resulting mass difference spectra \deltadso for MC and data are shown in Section~\ref{sub:val} and~\ref{sec:ffit}, respectively.
25: A double Gaussian is fitted to the data spectrum as a rough estimate of the width of the signal. The results for the total width, calculated by adding the weighted Gaussian widths in quadrature, are $1.55 \pm 0.15\mevcc$ and $1.66 \pm 0.26\mevcc$ for mode \modea\ and \modeb, respectively. Note that for this preliminary fit the intrinsic width and the resolution have not been taken separately into account.
26:
27:
28: \subsection{Resolution model}
29: \label{ssec:resmod}
30: Although very clean signals with more than 2400 (2900) entries have been obtained for decay mode \modea\ (\modeb), it is not feasible to obtain the resolution model and the intrinsic width from a single fit to data with all parameters allowed to vary. Instead a resolution model is derived from the corresponding \Dsop Monte Carlo samples.
31: The generation of $\epem \to \ccbar$ fragmentation events with high accuracy is a difficult task, so deviations between simulated and real data are possible. In particular the distribution of the CMS momentum $p^{*}$ of the reconstructed \Dsop mesons differs between data and Monte Carlo events.
32: In order to extract a reliable resolution model from Monte Carlo, the model is determined as a function of $p^{*}$. Another method to compensate for the inaccuracies of the simulation is to weight the Monte Carlo spectrum according to the $p^{*}$ distribution obtained from data and extract the resolution model from the weighted Monte Carlo sample. Since the second method relies on both real data and Monte Carlo data and is sensitive to the chosen $p^{*}$ binning, it is used as a systematic check and the resolution model is derived from the first method.
33:
34: The resolution can be extracted by calculating the difference of the invariant mass of a reconstructed candidate and the corresponding generated mass. The derived distribution consists only of the deconvolved resolution part of the \Dsop signal.
35: Since for the measurement of the \Dsop
36: mass the mass difference \deltadso is taken, the mass difference $\Delta m_g(\Dsop)$ of
37: the corresponding generated candidates has to be subtracted to obtain
38: the deconvolved distribution:
39: \begin{equation}
40: \Delta m_{res}= \deltadso-\Delta m_g(\Dsop).
41: \end{equation}
42:
43: The procedure is the same for both decay modes. The Monte Carlo sample is divided into 25 $p^{*}$ bins with a width of $0.07 \gevc$ in the range from $2.70\gevc$ to $4.45\gevc$. An unbinned maximum likelihood fit of $\Delta m_{res}$ is performed for each bin of $p^{*}$ using a probability density function (PDF) assembled from Gaussian functions with fit parameters $\Delta m_{res,0}$, $r$ and $\sigma_0$:
44: \begin{equation}
45: R(\Delta m_{res})=\int_{\sigma_0}^{r\sigma_0} \frac{1}{r\sigma^2}
46: e^{-\frac{(\Delta m_{res}-\Delta m_{res,0})^2}{2\sigma^2}}\mathrm{d}\sigma,
47: \label{eq:iares}
48: \end{equation}
49: where $\Delta m_{res,0}$ is the mean value and the width is integrated from a minimum value of $\sigma_0$ up to a maximum of $r\sigma_0$. The scale parameter for the upper width limit $r$ is determined for each $p^{*}$ bin, but does not vary drastically with $p^{*}$. Fixed values for $r$ are obtained from a fit with a constant function to the $r$ distribution which yields $r = 5.64 \pm 0.05$ ($6.40 \pm 0.06$) for decay mode \modea\ (\modeb). A non-constant $1^{st}$ order polynomial can be fitted to the $r$ distribution for mode \modea; this scenario is investigated in Section~\ref{ssec:fitproc}. The fits to $\Delta m_{res}$ are repeated with $r$ fixed to the constant value obtained which leaves $\sigma_{0}$ as the only free parameter for the width. For each $p^{*}$ bin, $\sigma_0$ is recalculated. The new $\sigma_{0}$ distribution can be best parameterized by the second order polynomial below with coefficients $b_{i}$.
50: \begin{equation}
51: \sigma_{0}(p^{*}) = b_{0} + b_{1}p^{*} + b_{2}p^{*2}.
52: \label{eq:sigma0}
53: \end{equation}
54:
55: \subsection{Validation of the resolution model}
56: \label{sub:val}
57: To verify that the resolution model is reliable, we applied it to the \deltadso distribution of the \Dsop Monte Carlo data sample
58: \begin{equation}
59: \deltadso = S(\dmmean,\Gamma(\Dsop)) \ast R(\Delta m_{res}).
60: \end{equation}
61: The generated \Dsop mass difference distribution $S(\dmmean,\Gamma(\Dsop))$ in the simulation follows a non-relativistic Breit-Wigner distribution with mean value \dmmean and width $\Gamma(\Dsop)$ (Table~\ref{tab:mcfit}). The fit of this Breit-Wigner convoluted with the resolution function to Monte Carlo data must return the generated values for the mass difference \dmmean and the decay width $\Gamma(\Dsop)$. Since the convolution of the resolution model and the Breit-Wigner function cannot be handled analytically one has to use numerical integration methods instead. To compute the
62: convolution integral the {\it Trapezoid Sum Rule} method has been
63: applied. The convolution window has been chosen as $\pm10\mevcc$, which
64: is about $10$ times the width of the resolution function, and was
65: divided into 200 bins. The applied method returns stable fit results
66: for reconstructed Monte Carlo and reproduces the input values for the
67: mass difference and width in the simulation with sufficient accuracy (Fig.~\ref{fig:mcfit}, Table~\ref{tab:mcfit}). The differences between the reconstructed values and the generated values are assigned as systematic uncertainties, which are $-7\kevcc$ ($-14 \kevcc$) for the mass difference and $-2\kevcc$ ($-9\kevcc$) for the width .
68: \begin{table}[!htb]
69: \caption{Results of the test fits to the MC data (statistical errors only) for both decay modes, compared with the generated values.}
70: \begin{center}
71: \begin{tabular}{|c|c|c|c|} \hline
72: Parameter & \modea & \modeb & generated\\ \hline\hline
73: $\dmmean\ /\mevcc$ & $27.737 \pm 0.003$ & $27.730 \pm 0.003$ & $27.744$ \\ \hline
74: $\Gamma(\Dsop)\ /\mevcc$ & $0.998 \pm 0.005$ & $0.991 \pm 0.007$ & $1.000$ \\ \hline
75: \end{tabular}
76: \end{center}
77: \label{tab:mcfit}
78: \end{table}
79:
80: \begin{figure}[!htb]
81: \begin{center}
82: \includegraphics[height=5cm]{Eps/Fig2a.eps}
83: \includegraphics[height=5cm]{Eps/Fig2b.eps}
84: \caption{Fit of the convolution of the non-relativistic Breit-Wigner function and the resolution function to the \deltadso distribution in MC (dots) as a crosscheck for the $p^*$-dependent resolution model. Left: decay mode \modea; right: decay mode \modeb}
85: \label{fig:mcfit}
86: \end{center}
87: \end{figure}
88:
89:
90: \subsection{Fit to the data}
91: \label{sec:ffit}
92: The assumption that the \Dsop lineshape $S(\dmmean,\Gamma(\Dsop))$ follows a non-relativistic
93: Breit-Wigner, as used for the MC, is not sufficient for the measurement of the intrinsic
94: width. A better and commonly used description of a resonance lineshape
95: is the following, {\it relativistic} Breit-Wigner function:
96: \begin{equation}
97: BW(m)\propto \frac{m m_0 \Gamma}{\left(m^2-m_0^2\right)^2+m_0^2\Gamma^2}
98: \label{eq_relbw}
99: \end{equation}
100:
101: To measure the mass difference \dmmean and the intrinsic width $\Gamma(\Dsop)$, the extracted $p^{*}$-dependent resolution model $R(\Delta m_{res})$ is convoluted with the relativistic Breit-Wigner distribution and then fitted to data using an unbinned maximum likelihood fit. For the convolution the same numerical integration method as described for the MC test fit (Section~\ref{sub:val}) is applied.
102: The background is described by a linear function. The fit parameters obtained for both decay modes are listed in table~\ref{tab:datfit}. The corresponding mass distributions are shown with the fits in Fig.~\ref{fig:datfit}.
103:
104: \begin{table}[!htb]
105: \caption{Results of the fit to the data (statistical errors only) for both decay modes.}
106: \begin{center}
107: \begin{tabular}{|c|c|c|} \hline
108: Parameter & \modea & \modeb \\ \hline\hline
109: $\dmmean\ /\mevcc$ & $27.209 \pm 0.028$ & $27.180 \pm 0.023$ \\ \hline
110: $\Gamma(\Dsop)\ /\mevcc$ & $1.112 \pm 0.068$ & $0.990 \pm 0.059$ \\ \hline
111: Signal yield (events) & $2401 \pm 47$ & $2959 \pm 51$ \\ \hline
112: \end{tabular}
113: \end{center}
114: \label{tab:datfit}
115: \end{table}
116:
117: \begin{figure}[!htb]
118: \begin{center}
119: \includegraphics[height=6cm]{Eps/Fig3a.eps}
120: \includegraphics[height=6cm]{Eps/Fig3b.eps}
121: \caption{Fit to the data (dots) of the relativistic Breit-Wigner function convoluted with the $p^*$-dependent resolution function to the \deltadso spectrum to obtain the mass difference \dmmean and width $\Gamma(\Dsop$). The background is described by a first order polynomial, shown by the dotted line. Left: decay mode \modea; right: decay mode \modeb.}
122: \label{fig:datfit}
123: \end{center}
124: \end{figure}
125:
126: