1: \documentclass[aps,prl,twocolumn]{revtex4}

2: \usepackage{epsfig}

3: 

4: \begin{document}

5: 

6: \title{Is a multiple excitation of  a single atom equivalent to a single excitation of an ensemble of atoms?}

7: 

8: \author

9: {Ido Kanter, Aviad Frydman and Assaf Ater}

10: \address

11: {Minerva center and the Department of Physics, Bar Ilan

12: University, Ramat-Gan 52900, Israel}

13: 

14: \begin{abstract}

15: Recent technological advances have enabled to isolate, control and

16: measure the properties of a single atom, leading to the

17: possibility to perform statistics on the behavior of single

18: quantum systems. These experiments have enabled to check a

19: question which was out of reach previously: Is the statistics of a

20: repeatedly  excitation of an atom N times equivalent to a single

21: excitation of an ensemble of N atoms? We present a new method to

22: analyze quantum measurements which leads to the postulation that

23: the answer is most probably no. We discuss the merits of the

24: analysis and its conclusion.

25: 

26: PACS numbers: 03.65.Ta, 42.50.Lc, 02.50.Ga

27: \end{abstract}

28: \maketitle

29: 

30: 

31: The development of laser cooling techniques have enabled to study

32: the properties of single atoms. This research is motivated both by

33: the quest for better understanding of basic quantum mechanics

34: concepts as well as by potential applications in the fields of

35: quantum computers, quantum clocks and random number generators.

36: \cite{wieman,berkeland,itano}. Though the accuracy of quantum

37: mechanical experiments is increasing rapidly, the analysis of the

38: data is still in its primary stage and a method for detecting

39: correlations is far from being established.

40: 

41: An example for the above effort is the advanced study of quantum jumps

42: in a single atom. Figure 1 shows the results of a typical quantum jump

43: experiment performed on a $^{189}Hg^{+}$ ion (data provided by W.M Itano).

44: The atomic system contains two excited states (as seen in figure 2).

45: State 1 has a strong coupling to the ground state while state 2 is

46: weakly coupled and has a long lifetime. If the system is excited to

47: state 1 it will emit a photon as the system decays to the ground

48: state (``light'' level) unless it is transferred to state 2 in which

49: case the fluorescence will stop for a period of the lifetime of this

50: state (``dark'' level). The data is viewed as a set of switches between

51: two levels. The higher one representing detection of emitted photons as

52: the sample decays from state 1 to the ground state and the lower one

53: (basically 0) represents the situation where the sample is at state 2.

54: 

55: \vspace{1.5cm}

56: \begin{figure}

57: \centerline{\epsfxsize=3.25in \epsffile{fig1.eps}} \vspace{1.5cm}

58: \caption{A section of a typical quantum jump measurement performed

59: on a Mercury ion. The red line shows the clipping threshold for

60: the analysis process (results are insensitive to the precise

61: threshold value). The figure also depicts the Markov matrix (M)

62: and the noise matrix (N) obtained from the analysis of the entire

63: experimental data set using by the BW algorithm. } \end{figure}

64: 

65: 

66: 

67: 

68: This type of experiments opens the opportunity to examine new

69: questions related to quantum mechanics. It is well established

70: that the decay of an ensemble of quantum particles is governed by

71: a Poissonian process. The number of decay events as a function of

72: time is proportional to $exp(-t/\tau)$, where $\tau$ is the

73: relevant life time. This behavior indicates that the process is

74: random with no correlations among successive events. Such

75: statistics were confirmed in many quantum experiments such as

76: Alpha and Beta radiation and decay of excited atoms to the ground

77: state \cite{ensemble}. The question we raise in this work is

78: whether the excitation of a single atom many times would exhibit

79: the same behavior and statistics. For instance, in the quantum

80: jumps experiments, the question is whether the level lifetimes are

81: random and temporally uncorrelated. This question is of major

82: importance both for the foundation of quantum mechanics as well as

83: for setting guidelines for future applications.

84: 

85: \vspace{0.0cm}

86: \begin{figure}

87: \centerline{\epsfxsize=3.25in \epsffile{fig2.eps}} \vspace{1.0cm}

88: \caption{A schematic energy diagram of the ionic system. }

89: \end{figure}

90: 

91: 

92: In order to check this question we develop a method to analyze the

93: data of quantum jumps experiments. This method is based on the

94: theory of the hidden Markov model\cite{BW,2,paper}. A Markov model

95: is a finite state machine that changes state once every time unit.

96: The manner in which the state transitions occur is probabilistic

97: and is governed by a state-transition matrix, M. For applying this

98: model to quantum jumps we focus on the simplest case of a Markov

99: model, i.e. only two levels ``light'' and ``dark''.  Quantum jumps

100: are expected to be described by a Markovian process, since events

101: are assumed to depend only on the current state and not on the

102: history.  If on the other hand, a system does not follow pure

103: Markovian statistics, the process is named a Hidden Markov

104: Process. For example, if the atom experiences external fields or

105: interactions, one may expect to observe a deviation from Markovian

106: behavior. In this case the process has to be described by two

107: $2X2$ transition matrices. The first one, M, stands for the

108: transitions due to the Markovian process (M(i,j) stands for the

109: transition from i to j) while the second one, N, represents

110: unexpected transitions that can not be ascribed to the usual

111: statistics of the system.  The off-diagonal elements of matrix N,

112: known also as the noise, measure the amount of deviation from pure

113: Markovian behavior. For pure Markovian processes the off-diagonal

114: elements are zero and as these elements increase the noise

115: increases. A practical way to determine the two matrices, M and N,

116: that characterize the most likely underlying processes, is known

117: as the Baum-Welch (BW) algorithm \cite{BW}.

118: 

119: The actual analysis of the quantum jumps data is performed using

120: the following procedure.  A data set, such as that of figure 1, is

121: clipped to produce a symbolic data sequence containing two levels,

122: 0 and 1 (0 representing a "dark" level and 1 representing the

123: ``light'' level). The transitions between these levels are

124: analyzed by running the Baum-Welch procedure \cite{BW} to produce

125: two matrices, M, and, N. A typical result is shown in figure 1.

126: Here we analyze the data of a sequence of $10^5$ data points where

127: each point represents the number of photon counts in a time

128: interval of $1ms$.

129: 

130: 

131: The off-diagonal elements of matrix N show that while there are no

132: unexpected transitions from $0 \rightarrow 1$, there are

133: non-trivial transitions from $1 \rightarrow 0$ that are not

134: consistent with a simple Markovian framework. It is important to

135: note that though the effect is rather small, it can not be

136: attributed to a finite size artifact. We have generated artificial

137: sequences (using matrix M only) having sizes similar to the

138: experimental one which produced pure Markov matrices with

139: negligible off-diagonal elements of matrix N. Similar

140: non-Markovian results were obtained on a longer sequence

141: ($7.5*10^5$ data points) taken from a Sr quantum jumps experiment

142: \cite{berkeland}.

143: 

144: The BW procedure indicates that a single ion produces unexpected

145: transitions. One may ask whether it is possible to go one step

146: further and to identify the location (in the time sequence) of the

147: noise. Such a question arises, for instance, in the case of

148: digital communication, where a Markovian message is transmitted

149: through a noisy channel. The receiver's goal is to identify the

150: locations of the noisy bits in order to recover the original

151: signal \cite{cover}. An established way to do this is by using the

152: Viterbi algorithm \cite{viterbi1} to compute the most likely

153: underlying Markov sequence, or, in other words, the expected

154: physical outcome without the detected noise \cite{viterbi}.

155: 

156: 

157: 

158: The outcome of running the Viterbi algorithm on the data of figure 1 is presented in figure 3.  The clipped data contains sub-sequences like ...000010000... (type A) or ...111101111....  (type B).  The Viterbi procedure identifies most of events of type B as noise, but non of type A. A similar effect was obtained in all studied quantum jump experimental sequences (over 10 data sets).

159: 

160: 

161: \vspace{1.5cm}

162: \begin{figure}

163: \centerline{\epsfxsize=3.25in \epsffile{fig3.eps}} \vspace{1.0cm}

164: \caption{Part of the symbolic data sequence after clipping the

165: data of figure 1. The dashed line shows the result of the Viterbi

166: algorithm.}

167: \end{figure}

168: 

169: 

170: The above result reveals the fact that sawtooth-like features of type B

171: (a few dozens in a sequence of $10^{5}$ data points), can not be explained

172: as part of the usual statistics of the ion, i.e.  Markovian statistics.

173: We demonstrate that these sawteeth are the cause for non-Markovian noise by

174: flipping these particular data points from 0 to 1 and running the BW procedure

175: on the revised sequence. This procedure yields practically pure Markov matrices

176: even for flipping only $\sim$40\% of the features detected by the Viterbi process as problematic.

177: 

178: It is important to note that deviations from pure exponential behavior of the plateau statistics were observed in quantum jumps experiments \cite{berkeland}. These were attributed to experimental difficulties in controlling the laser amplitude and frequency over relatively large time periods. This experimental artifact gives rise to non-Poissonian statistics due to long plateaus (with times larger than 130 ms).  Our analysis (the BW and the Viterbi algorithms) implies that it is the short plateaus, rather then the long ones, which are responsible for the unexpected behavior.  Further support is obtained by running the BW algorithm on quantum jump data once while erasing long plateaus (having time scales above 100ms) and again while erasing plateaus shorter than 2ms. In the former case we found that the off-diagonal noise remained non-zero. On the other hand the latter case revealed that the noise was entirely suppressed when the short plateau were removed from the sequence. Our analysis therefore strongly implies that the sources of non-Markovian behavior in a single atom experiment are the short plateaus where the system spends a relatively short time at the excited state.

179: 

180: In an attempt to understand the results of the analysis we note

181: that though an atom is a single quantum particle it is a many body

182: system containing many degrees of freedom. Therefore, the decay of

183: the atom to the ground state may have some characteristic

184: timescales. This might be a source for our observations. If the

185: ion is repeatedly excited with very short time intervals it may

186: not be able to fully relax. This may give rise to a deviation from

187: the expected Markovian statistics. In this respect the excitation

188: of a single atom many times may differ from a single excitation of

189: many atoms. A natural way to further check this hypothesis

190: experimentally is to enhance the effect by taking measurement with

191: a shorter characteristic timescale (counting photons in time

192: intervals shorter than a ms). We suspect that such measurements

193: would result in a larger deviation from Markovian behavior.

194: 

195: We gratefully acknowledge W.M Itano and D. J Berkeland for

196: providing the quantum jump experimental data and P.W. Anderson and

197: L. Khaikovich for useful discussions.

198: 

199: 

200: 

201: 

202: %\bibliography{scibib}

203: 

204: \begin{thebibliography}{9}

205: 

206: \bibitem{wieman} C.E. Wieman, D.E. Pritchard and D.J. Wineland, Rev. Mod.  Phys., \textbf{71}, S253 (1999).

207: 

208: \bibitem{berkeland} D.J. Berkeland, D.A. Raymonson and V.M. Tassin, $\it{arXiv:physics 0304013}$.

209: 

210: \bibitem{itano} S.A. Diddmas et. al., Science  \textbf{293}, 825 (2001).

211: 

212: \bibitem{ensemble} For example see A.G. Calamai and C.E. Johnson, Phys. Rev. A {\bf 42}, 5425 (1990).

213: 

214: \bibitem{BW} G.J. McLachen and T. Krishnan, "The EM Algorithm and

215: Extensions", John Wiley and Sons (1997).

216: 

217: \bibitem{2} L.B. Rabiner, B.H. Juang, "An Introduction to Hidden

218: Markov Models", IEEE ASSP Magazine, vol 3, February 1986.

219: 

220: \bibitem{paper} I. Kanter, A. Frydman and A. Ater, $\it{Cond-mat/0402246}$.

221: 

222: \bibitem{cover} T.M. Cover and J.A. Cover, "Elements of Information Theory", John Wiley and Sons (1991).

223: 

224: \bibitem{viterbi1} J.G. Proakis, ``Digital communication'', McGraw-Hill (2001).

225: 

226: \bibitem{viterbi} Note that the separation of the received message to signal and noise is not always possible due to either theoretical bounds of information theory or limitations of the used algorithm and the examined sequence. However, in many cases the Viterbi algorithm proves very usuful.

227: 

228: \end{thebibliography}

229: 

230: 

231: \end{document}

232: