1: The mixing parameters are determined in an unbinned, extended
2: maximum-likelihood fit to the RS and WS data samples over the four
3: observables \mKpi, \dm, $t$, and \terr. The fit is performed in several
4: stages. First, RS and WS signal and
5: background shape parameters are determined from a fit to \mKpi and
6: \dm, and are not varied in subsequent fits. Next, the
7: \Dz proper-time resolution function and lifetime are
8: determined in a fit to the RS data using \mKpi and \dm to separate the
9: signal and background components. We fit to the WS data sample using
10: three different models. The first model assumes both \CP conservation
11: and the absence of mixing, and only measures \Rdcs. The second model
12: allows for mixing, but assumes no
13: \CP violation, and the third model allows for both mixing and \CP violation.
14:
15: The RS and WS \mdm\ distributions are described by four components:
16: signal, random $\pisoft^+$, misreconstructed \Dz and combinatorial
17: background. Signal has a characteristic peak in both \mKpi and \dm.
18: The random $\pisoft^+$ component models reconstructed \Dz decays
19: combined with a random slow pion and has the same shape
20: in \mKpi as signal events, but does not peak in \dm. Misreconstructed
21: \Dz events have one or more of the \Dz decay products either not
22: reconstructed or reconstructed with the wrong particle
23: hypothesis. They peak in \dm, but not in \mKpi. For RS events, most of
24: these are semileptonic decays $\Dz\to\Km\ell^+\nu$ with the charged
25: lepton misidentified as a pion. For WS events, the main contributor
26: is RS $\Dz\to\Kmpip$ decays where the
27: \Km and the \pip are misidentified as \pim and \Kp, respectively.
28: Combinatorial background events are those not described by the
29: above components; they do not exhibit any peaking structure in
30: \mKpi or \dm.
31: \par
32: The functional forms of the probability density functions (PDFs) for
33: the signal and background components are chosen based on studies of
34: Monte Carlo (MC) samples. However, all parameters are determined from
35: two-dimensional likelihood fits to data over the full
36: $1.81<\mKpi<1.92\gevcc$ and $0.14<\dm<0.16\gevcc$ region.
37:
38:
39: We fit the RS and WS data samples simultaneously with shape parameters
40: describing the signal and random $\pisoft^+$ components shared between
41: the two data samples. We find $1,141,500\pm
42: 1,200$ RS signal events and $4,030\pm 90$ WS signal events. The
43: dominant background component is the random $\pisoft^+$ background.
44: Projections of the WS data and fit are shown in Fig.~\ref{fig:r18Data_MdMfit}.
45: \begin{figure}[phtb]
46: \centering
47: \centerline{%
48: \includegraphics[width=0.5\linewidth, clip=]{figures/paper_WS_D0Mass_figure1a.eps}
49: \includegraphics[width=0.5\linewidth, clip=]{figures/paper_WS_DeltaM_figure1b.eps}
50: }
51: \caption{a) \mKpi for
52: wrong-sign (WS) candidates with $0.1445<\dm<0.1465\gevcc$, and b) \dm for WS candidates
53: with $1.843<\mKpi<1.883\gevcc$. The fitted PDFs are overlaid.
54: The shaded regions represent the different background components.}
55: \label{fig:r18Data_MdMfit}
56: \smallskip
57: \end{figure}
58:
59: The measured proper-time distribution for the RS signal is described by an
60: exponential function convolved with a resolution function whose
61: parameters are determined by the fit to the data. The resolution function
62: is the sum of three Gaussians with widths
63: proportional to the estimated event-by-event proper-time uncertainty
64: \terr. The random $\pisoft^+$ background is described by the same proper-time
65: distribution as signal events, since the slow pion has little weight
66: in the vertex fit. The proper-time distribution of the combinatorial
67: background is described by a sum of two Gaussians, one of which has a
68: power-law tail to account for a small long-lived component. The
69: combinatorial background and real \Dz decays have different
70: \terr distributions, as determined from data using a background-subtraction
71: technique \cite{Pivk:2004ty} based on the fit to \mKpi and \dm.
72:
73: The fit to the RS proper-time distribution is performed over all
74: events in the full \mKpi and \dm region. The PDFs for signal and
75: background in \mKpi and \dm are used in the proper-time fit with all
76: parameters fixed to their previously determined values. The fitted
77: \Dz lifetime is found to be consistent with the world-average
78: lifetime~\cite{PDG2006}.
79:
80: The measured proper-time distribution for the WS signal is modeled by
81: Eq.~(\ref{eq:Tws}) convolved with the resolution function determined in
82: the RS proper-time fit. The random $\pisoft^+$ and misreconstructed
83: \Dz backgrounds are described by the RS signal proper-time
84: distribution since they are real \Dz decays.
85: The proper-time distribution for WS data is shown in
86: Fig.~\ref{fig:histTimeBiasWSR18Data}. The fit results with and without
87: mixing are shown as the overlaid curves.
88:
89: \begin{figure}[phtb]
90: \centering
91: \centerline{%
92: \includegraphics[width=0.9\linewidth, clip=]{figures/paper_WSResidual_figure2.eps}
93: }
94: \caption{a) The proper-time distribution of combined \Dz and \Dzb
95: WS candidates in the signal region
96: $1.843<\mKpi<1.883\gevcc$ and $0.1445<\dm<0.1465\gevcc$.
97: The result of the fit allowing (not allowing) mixing
98: but not \CP violation
99: is overlaid as a solid (dashed) curve.
100: Background components are shown as shaded regions.
101: b) The points represent the difference
102: between the data and the no-mixing fit. The solid curve
103: shows the difference between fits with and without mixing.}
104: \label{fig:histTimeBiasWSR18Data}
105: \end{figure}
106:
107: The fit with mixing provides a substantially better description of the data
108: than the fit with no mixing.
109: The significance of the mixing signal is evaluated based on
110: the change in negative log likelihood with respect to the minimum.
111: Figure~\ref{fig:CPContour} shows confidence-level (CL) contours
112: calculated from the change in log likelihood ($-2\Delta\ln{\cal L}$) in two
113: dimensions (\xPrimeSq and \yPrime) with systematic uncertainties
114: included. The likelihood maximum is at the unphysical value of
115: $\xPrimeSq=-2.2\times10^{-4}$ and $\yPrime = 9.7 \times 10^{-3}$. The
116: value of $-2\Delta\ln{\cal L}$ at the most likely point in the
117: physically allowed region ($\xPrimeSq=0$ and $\yPrime=6.4 \times
118: 10^{-3}$) is $0.7$~units. The value of
119: $-2\Delta\ln{\cal L}$ for no-mixing is $23.9$~units.
120: Including the systematic uncertainties, this corresponds to a
121: significance equivalent to 3.9~standard deviations
122: ($1-\mbox{CL}=1\times10^{-4}$) and thus constitutes evidence for
123: mixing. The fitted values of the mixing parameters and \Rdcs are
124: listed in Table~\ref{tab:results}. The correlation coefficient
125: between the \xPrimeSq and \yPrime parameters is $-0.94$.
126:
127: \begin{figure}[phtb]
128: \centering
129: \centerline{%
130: \includegraphics[width=0.95\linewidth, clip=]{figures/paper_SigmaContour_figure3.eps}
131: }
132: \caption{The central value (point) and confidence-level (CL) contours for
133: $1-\mbox{CL}=0.317\ (1\sigma)$, $4.55\times10^{-2}\ (2\sigma)$,
134: $2.70\times10^{-3}\ (3\sigma)$, $6.33\times10^{-5}\ (4\sigma)$ and
135: $5.73\times10^{-7}\ (5\sigma)$, calculated from the change in the value
136: of $-2\ln{\cal L}$ compared with its value at the minimum.
137: Systematic uncertainties are included. The no-mixing point is shown
138: as a plus sign~($+$).}
139: \label{fig:CPContour}
140: \end{figure}
141:
142: Allowing for the possibility of \CP violation, we calculate the values
143: of $\Rdcs = \sqrt{\Rdcs^+\Rdcs^-}$ and
144: $\AD = (\Rdcs^{+} - \Rdcs^{-})/(\Rdcs^{+} + \Rdcs^{-})$
145: listed in Table~\ref{tab:results}, from the fitted $\Rdcs^{\pm}$ values.
146: The best fit in each case is more than three
147: standard deviations away from the no-mixing hypothesis.
148: All cross checks indicate that the high level of agreement between
149: the separate \Dz and \Dzb fits is a coincidence.
150: \par
151: As a cross-check of the mixing signal, we perform independent
152: \mdm\ fits with no shared parameters for intervals in proper time selected
153: to have approximately equal numbers of RS candidates.
154: The fitted WS branching fractions are shown in
155: Fig.~\ref{fig:RwsTimeBins} and are seen to increase with time. The
156: slope is consistent with the measured mixing parameters and
157: inconsistent with the no-mixing hypothesis.
158: \begin{figure}[phtb]
159: \centering
160: \centerline{%
161: \includegraphics[width=0.9\linewidth, clip=]{figures/paper_RwsTime_figure4.eps}
162: }
163: \caption{The WS branching fractions from independent \mdm\ fits to
164: slices in measured proper time (points).
165: The dashed line shows the expected wrong-sign rate
166: as determined from the mixing fit shown in Fig.~\ref{fig:histTimeBiasWSR18Data}.
167: The $\chi^2$ with respect to expectation from the mixing fit is 1.5;
168: for the no-mixing hypothesis (a constant WS rate), the $\chi^2$ is 24.0.}
169: \label{fig:RwsTimeBins}
170: \end{figure}
171:
172: \begin{table}[thb]
173: \caption{Results from the different fits.
174: The first uncertainty listed is statistical and the second systematic.}
175: \label{tab:results}
176: \centering\small
177: \begin{ruledtabular}
178: \begin{tabular}{lcr@{~$\pm$}r@{~$\pm$}r}
179: Fit type & Parameter & \multicolumn{3}{c}{Fit Results ($/10^{-3}$)} \\
180: \hline
181: No \CP viol. or mixing & $\Rdcs$ & $3.53 $ & $ 0.08 $ & $ 0.04$\\
182: \hline
183: \multirow{3}{1.7cm}{No \CP\\ violation}
184: & $\Rdcs$ & $3.03$ & $0.16$ & $ 0.10$ \\
185: & $\xPrimeSq$ & $-0.22$ & $0.30$ & $ 0.21$ \\
186: & $\yPrime$ & $9.7$ & $4.4$ & $ 3.1$ \\
187: \hline
188: \multirow{5}{1.7cm}{\CP\\ violation \\ allowed}
189: & $\Rdcs$ & $3.03$ &$0.16$ & $0.10$ \\
190: & $\AD$ & $-21$ & $52$ & $15$ \\
191: & $\xPrimePSq$ & $-0.24 $ & $ 0.43 $ & $ 0.30 $\\
192: & $\yPrimeP$ & $ 9.8 $ & $ 6.4 $ & $ 4.5 $\\
193: & $\xPrimeMSq$ & $-0.20 $ & $ 0.41 $ & $ 0.29 $\\
194: & $\yPrimeM$ & $ 9.6 $ & $ 6.1 $ & $ 4.3 $
195: \end{tabular}
196: \end{ruledtabular}
197: \end{table}
198:
199: We have validated the fitting procedure on simulated data samples using
200: both MC samples with the full detector simulation and large
201: parameterized MC samples. In all cases we have found the fit to be
202: unbiased. As a further cross-check, we have performed a fit to the RS
203: data proper-time distribution allowing for mixing
204: in the signal component; the fitted values of the mixing parameters
205: are consistent with no mixing. The correlations among parameters
206: determined at different stages of the fit are low. In addition we have found the staged fitting approach to give the same solution and confidence regions as
207: a simultaneous fit in which all parameters are allowed to vary.
208: