1:
2: \documentstyle[preprint,pra,aps]{revtex}
3:
4: %\documentstyle[twocolumn,pra,aps]{revtex}
5: %\documentstyle{article}
6:
7: \begin{document}
8:
9: \bibliographystyle{prsty}
10: %\draft
11: %\preprint{submitted to Phys.Rev. A}
12: \tighten
13:
14: \title{Finite resolution measurement of the non-classical polarization
15: statistics of entangled photon pairs}
16:
17: \author{Holger F. Hofmann}
18: \address{Department of Physics, Faculty of Science, University of Tokyo\\
19: 7-3-1 Hongo, Bunkyo-ku, Tokyo113-0033, Japan}
20:
21: %\author{Holger F. Hofmann\\Department of Physics, Faculty of Science,
22: %University of Tokyo\\7-3-1 Hongo, Bunkyo-ku, Tokyo113-0033, Japan}
23:
24: \date{\today}
25:
26: \maketitle
27:
28:
29: \begin{abstract}
30: By limiting the resolution of quantum measurements, the measurement
31: induced changes of the quantum state can be reduced, permitting
32: subsequent measurements of variables that do not commute with the
33: initially measured property. It is then possible to experimentally
34: determine correlations between non-commuting variables.
35: The application of this method to the polarization statistics of entangled
36: photon pairs reveals that negative conditional probabilities between
37: non-orthogonal polarization components are responsible for the violation
38: of Bell's inequalities. Such negative probabilities can also be observed
39: in finite resolution measurements of the polarization of a single photon.
40: The violation of Bell's inequalities therefore originates from local
41: properties of the quantum statistics of single photon polarization.
42: \end{abstract}
43:
44: \pacs{PACS numbers:
45: 03.65.Bz % Foundations of quantum mechanics
46: 42.50.Dv % Non-classical field states
47: 03.67.-a % Quantum Information
48: }
49:
50: \section{Introduction}
51: Perhaps the most convincing proof of the non-classical nature of quantum
52: statistics is the violation of Bell's inequalities by a pair of entangled
53: spin-1/2 particles \cite{Bel64}. Several experimental tests of these
54: inequalities have been performed on pairs of entangled photons generated
55: either by two photon emission \cite{Fre72,Asp82} or by parametric
56: down conversion \cite{Ou88,Shi88}. These experimental tests compare the
57: polarization correlations
58: of photon pairs emitted at the same time for different sets of orthogonal
59: polarizations. While no information about the relationship between
60: non-orthogonal polarization directions of the single photon are revealed
61: in such measurements, the statistics obtained correspond to the quantum
62: theoretical prediction. Since the quantum formalism from which the violation
63: of Bell's inequalities is derived is widely accepted, one might wonder whether
64: it should not be possible to obtain a clearer understanding of the origin
65: of this non-classical effect by investigating the unique statistical
66: connection between non-commuting quantum variables in more detail.
67: In particular, finite resolution measurements can provide quantitative
68: information about a quantum variable without destroying the quantum coherence
69: between different eigenstate components of that variable \cite{Hof00a}.
70: By applying finite resolution measurements, it is therefore possible to
71: identify the non-classical correlations between non-commuting variables
72: directly \cite{Hof00b,Hof00c}. In the following, an experiment is proposed to
73: determine the correlations between non-orthogonal polarizations
74: of entangled photon pairs. It is shown that the violation of Bell's
75: inequalities results from the negative joint probabilities arising from
76: local non-classical correlations of the photon polarization. It is then
77: possible to give a local interpretation of entanglement based on
78: standard quantum mechanics.
79:
80: The rest of the paper is organized as follows.
81: In section \ref{sec:postulate}, the application of finite resolution
82: measurements to the polarization of a single photon is discussed and
83: fundamental non-classical correlations are derived. In section
84: \ref{sec:entangle}, the experimental setup for a measurement of entangled
85: photon pairs is presented and the statistical results of such a measurement
86: are derived. In section \ref{sec:quantcorr}, the non-classical features of
87: the statistics are identified and the implications concerning the nature
88: of entanglement are discussed. Finally, the conclusions are summarized in
89: section \ref{sec:concl}.
90:
91: \section{Finite resolution measurements}
92: \label{sec:postulate}
93: \subsection{The generalized measurement postulate}
94: A measurement assigns a quantity to a system property based on
95: the observable action of the system on some measurement device. The
96: uncertainty principle of quantum mechanics requires that this interaction
97: between the system and the measurement setup introduces noise into
98: properties that do not commute with the measured variable.
99: Therefore, the classical ideal of a complete determination of all
100: system properties is unattainable. Nevertheless it is possible to
101: obtain quantitative information on the correlations between non-commuting
102: variables by limiting the measurement resolution.
103: Such a finite resolution measurement is described by the generalized
104: measurement operator $\hat{P}_{\delta\!s} (s_m)$, which assigns a continuous
105: measurement value $s_m$ to the operator $\hat{s}$\cite{Hof00a}. It reads
106: \begin{equation}
107: \hat{P}_{\delta\!s} (s_m) = \left(2\pi \delta\!s^2 \right)^{-\frac{1}{4}}
108: \exp\left(- \frac{(s_m-\hat{s})^2}{4\delta\!s^2}\right).
109: \end{equation}
110: For a given initial state $\mid \psi_{\mbox{in}}\rangle$,
111: the probability $P(s_m)$ of obtaining a measurement result $s_m$ and the
112: state $\mid \psi_{\mbox{out}}\rangle$ after the measurement are then
113: given by
114: \begin{eqnarray}
115: P(s_m) &=& \langle \psi_{\mbox{in}} \mid \hat{P}^2_{\delta\!s} (s_m)
116: \mid \psi_{\mbox{in}} \rangle \nonumber \\
117: \mid \psi_{out}\rangle &=& \frac{1}{\sqrt{P(s_m)}}
118: \hat{P}_{\delta\!s} (s_m) \mid \psi_{in}\rangle.
119: \end{eqnarray}
120: Note that this generalized measurement postulate does not restrict the
121: values of an operator variable to the eigenvalues of that operator.
122: Eigenvalues emerge only in infinitely precise measurements. One of the
123: fundamental problems in the discussion of quantum mechanics is that
124: eigenvalues are often identified with ``elements of reality''\cite{EPR,Mer}
125: regardless of the measurement context discussed. By assigning a continuous
126: measurement value to the operator variable in a finite resolution measurement
127: this identification is avoided, allowing a determination of quantitative
128: results beyond the spectrum of its eigenvalues.
129:
130: \subsection{Finite resolution measurement of photon polarization}
131: The polarization of a single photon can be described in terms of the
132: Stokes parameters $\hat{s}_i$. In terms of the circular polarization
133: eigenstates $\mid R \rangle$ and $\mid L \rangle$, the operators of the
134: three single photon Stokes parameters may be written as
135: \begin{eqnarray}
136: \hat{s}_1 &=& \hspace{0.2cm}
137: \mid L \rangle\langle R \mid + \hspace{0.2cm}\mid R \rangle\langle L \mid
138: \nonumber \\
139: \hat{s}_2 &=& i \mid L \rangle\langle R \mid - i \mid R \rangle\langle L \mid
140: \nonumber \\
141: \hat{s}_3 &=& \hspace{0.2cm}
142: \mid R \rangle\langle R \mid - \hspace{0.2cm} \mid L \rangle\langle L \mid.
143: \end{eqnarray}
144: $\hat{s}_1$ represents the intensity difference between the x and y
145: polarizations, $\hat{s}_2$ represents the intensity difference between the
146: polarizations along the diagonals between the x and y axes, and
147: $\hat{s}_3$ represents the intensity difference between the circular
148: polarizations. Since only one photon is considered, the eigenvalues of
149: each Stokes parameter are $\pm 1$.
150:
151: A finite resolution measurement of photon polarization can be realized by
152: using a polarization sensitive beam displacer that shifts the x-polarization
153: component relative to the y-polarization component of the light field.
154: The displacement of the photon trajectory in the beam displacer can be
155: interpreted as the action of the one-photon Stokes parameter $\hat{s}_1$.
156: The polarization of the photon is then described by a
157: continuous value $s_{1m}$ of the Stokes parameter $\hat{s}_1$ obtained from
158: the measurement of the transversal photon position after the beam displacer.
159: The measurement resolution depends on the ratio of the displacement
160: and the width of the input beam. If the transversal profile of the
161: light field is Gaussian, the generalized measurement
162: postulate describes the single photon polarization statistics obtained
163: by measuring the polarization dependent displacement of the photon.
164: In terms of the circular polarization eigenstates, the measurement operator
165: is given by
166: \begin{eqnarray}
167: \lefteqn{\hat{P}_{\delta\!s}(s_{1m}) =
168: (2\pi\delta\!s^2)^{-\frac{1}{4}}
169: \exp\left(-\frac{s_{1m}^2+1}{4\delta\!s^2}\right)}
170: \nonumber \\ &\times\big(&
171: \;\cosh\left(\frac{s_{1m}}{2\delta\!s^2}\right)
172: \left(\mid R \rangle \langle R \mid + \mid L \rangle \langle L \mid\right)
173: \nonumber \\ &&
174: + \sinh\left(\frac{s_{1m}}{2\delta\!s^2}\right)
175: \left(\mid R \rangle \langle L \mid + \mid L \rangle \langle R \mid\right)
176: \big).
177: \end{eqnarray}
178: This operator describes the changes in the quantum state of the single
179: photon polarization conditioned by the finite resolution measurement of
180: the Stokes parameter $\hat{s}_1$.
181:
182: \subsection{Joint measurements of non-orthogonal polarizations}
183: In order to measure the correlated non-orthogonal polarization components
184: of a single photon, the finite resolution measurement can be combined with
185: a fully resolved polarization measurement. By rotating the polarization
186: by an angle of $\pi/4$ and separating the x and y components as shown in
187: figure \ref{setup1}, the eigenvalues of the Stokes parameter $\hat{s}_2$ are
188: measured. Two spatial patterns emerge, corresponding to the
189: conditional distributions of continuous measurement results $s_{1m}$
190: of the Stokes parameter $\hat{s}_1$
191: associated with a final measurement of the eigenvalues $+1$ or $-1$ of
192: the Stokes parameter $\hat{s}_{2}$.
193: The positive valued operator measure (POM) describing the joint measurement
194: of $\hat{s}_1$ and $\hat{s}_2$ is defined by projections onto the states
195: \begin{equation}
196: \mid s_{1m};s_2=\pm 1\rangle =
197: \hat{P}_{\delta\!s}(s_{1m})
198: \frac{1}{\sqrt{2}}\left(\mid R \rangle \pm i \mid L \rangle\right).
199: \end{equation}
200: %
201: The joint probabilities $P(s_{1m};s_2=\pm1)$ for measuring a finite
202: resolution value of $s_{1m}$ for the Stokes parameter $\hat{s}_1$
203: followed by an eigenvalue of $s_2=\pm 1$ for the Stokes parameter
204: $\hat{s}_2$ is then given by
205: \begin{eqnarray}
206: \label{eq:joint}
207: P(s_{1m};s_2=\pm1) &=&
208: |\langle s_{1m};s_2=\pm 1 \mid \psi_{\mbox{in}} \rangle|^2
209: \nonumber \\
210: &=& \frac{1}{2}
211: \left|\langle R \mid \hat{P}_{\delta\!s}(s_{1m})\mid\psi_{\mbox{in}} \rangle
212: \mp i \langle L \mid \hat{P}_{\delta\!s}(s_{1m})\mid\psi_{\mbox{in}} \rangle
213: \right|^2
214: ,
215: \end{eqnarray}
216: where $\mid \psi_{\mbox{in}}\rangle$ is an arbitrary initial state.
217: This POM thus assigns quantitative results to both Stokes parameters, allowing
218: a derivation of correlations between the polarization components of a single
219: photon.
220:
221: If the light field entering the measurement setup shown in
222: figure \ref{setup1} is polarized along the diagonal between the x and y
223: axes, the initial photon state is given by
224: \begin{equation}
225: \label{eq:input}
226: \mid \psi_{\mbox{in}}\rangle = \frac{1}{\sqrt{2}}\left(\mid R \rangle
227: + i \mid L \rangle \right).
228: \end{equation}
229: The joint probabilities of the measurement results $s_{1m}$ and $s_2$ can
230: then be determined using equation (\ref{eq:joint}).
231: In its most compact form, it reads
232: \begin{eqnarray}
233: \label{eq:condprop}
234: P(s_{1m};s_2=+1) &=& \left(2\pi \delta\!s^2\right)^{-\frac{1}{2}}
235: \exp\left(-\frac{s_{1m}^2+1}{2\delta\!s^2}\right)
236: \cosh^2\left(\frac{s_{1m}}{2 \delta\!s^2}\right)
237: \nonumber \\
238: P(s_{1m};s_2=-1) &=& \left(2\pi \delta\!s^2\right)^{-\frac{1}{2}}
239: \exp\left(-\frac{s_{1m}^2+1}{2\delta\!s^2}\right)
240: \sinh^2\left(\frac{s_{1m}}{2 \delta\!s^2}\right).
241: \end{eqnarray}
242: Note that $P(s_{1m}=0;s_2=-1)$ is always exactly equal to zero, even if
243: $\delta\!s$ is larger than one. Obviously, this result
244: is too exact to be explained in terms of a random measurement error
245: superimposed on classical statistics. The result for a measurement resolution
246: of $\delta\!s=0.6$ is illustrated in figure \ref{condprop1}. The
247: peaks in $P(s_{1m};s_2=-1)$ are shifted to values of about $\pm 1.1$ and
248: the asymmetry of the peaks seems to favor even higher values. These results
249: can hardly be explained by statistics originating only from the eigenvalues
250: of $s_1=\pm1$.
251:
252: \subsection{Negative conditional probabilities and non-classical
253: correlations in the polarization of single photons}
254: The non-classical features of the joint probabilities $P(s_{1m};s_2=\pm1)$
255: can be analyzed by expressing the result
256: as a sum of shifted normalized Gaussian distributions
257: \begin{equation}
258: G_{\delta\!s}(s_{1m}-d) := \left(2\pi \delta\!s^2\right)^{-\frac{1}{2}}
259: \exp\left(-\frac{(s_{1m}^2-d)^2}{2\delta\! s^2}\right).
260: \end{equation}
261: In terms of these Gaussians, the joint probabilities read
262: \begin{eqnarray}
263: \label{eq:sumcp}
264: P(s_{1m};s_2=+1) &=& \frac{1}{4}G_{\delta\!s}(s_{1m}+1)
265: +\exp\left(-\frac{1}{2\delta\! s^2}\right)\frac{1}{2}G_{\delta\!s}(s_{1m})
266: + \frac{1}{4}G_{\delta\!s}(s_{1m}-1)
267: \nonumber \\
268: P(s_{1m};s_2=-1) &=& \frac{1}{4}G_{\delta\!s}(s_{1m}+1)
269: -\exp\left(-\frac{1}{2\delta\! s^2}\right)\frac{1}{2}G_{\delta\!s}(s_{1m})
270: + \frac{1}{4}G_{\delta\!s}(s_{1m}-1).
271: \end{eqnarray}
272: Each Gaussian contribution to the joint probabilities given in equation
273: (\ref{eq:sumcp}) can be identified with elements of the density matrix
274: of the original state in the eigenstate basis of the observable $\hat{s}_1$.
275: As discussed in a previous paper \cite{Hof00a}, the measurement of
276: $s_{1m}$ modifies each matrix element by a decoherence factor
277: given by the difference of the eigenvalues and an information factor
278: depending on the difference between the measurement result $s_{1m}$ and
279: the average of the eigenvalues.
280:
281: The decoherence factor $\exp(-1/(2\delta\!s^2))$ is a result of the
282: quantum noise in the measurement required by the uncertainty principle.
283: In the case of the beam displacer acting on single
284: photons, it is the uncertainty of the wave vector dependent time the photon
285: spends in the birefringent medium which randomly rotates the Stokes vector
286: around the $s_1$ axis.
287: Since this noisy interaction is statistically independent of
288: the measurement result, it is possible to separate its effect
289: from the information obtained about the system. A hypothetical noise free
290: measurement then reveals negative probabilities of $s_2=-1$ for measurement
291: results $s_{1m}$ close to zero \cite{Hof00a}. These negative probabilities
292: describe the non-classical correlations between non-commuting operator
293: variables \cite{Hof00b,Hof00c}.
294:
295: The information about $\hat{s}_1$ obtained in the measurement
296: modifies the statistical weight of each density matrix element
297: by a Gaussian function of the difference between the
298: measurement result $s_{1m}$ and the average of the two eigenvalues
299: of the density matrix element. In particular, the Gaussians centered
300: around $s_{1m}=0$ represent contributions from
301: the quantum coherence between the $s_1=+1$ and the $s_1=-1$
302: eigenstates conditioned by a measurement of $s_{1m}$.
303: Measurement results close to $s_{1m}=0$ enhance the coherence
304: and increase the probability of $s_2=+1$ to values above 1, while
305: measurement results far away from $s_{1m}=0$ reduce the coherence,
306: lowering the probability of $s_2=+1$ to values below 1. In order to explain
307: this non-classical correlation between $s_1$ and $s_2$, some measure of
308: reality must be attributed to $s_1=0$, even though it is not an
309: eigenvalue of $\hat{s}_1$ \cite{Hof00b}.
310: Since the width of the Gaussians represents the effect of random noise
311: in the readout of the finite resolution measurement, it is reasonable
312: to identify each Gaussian contribution with its average value of $s_{1m}$.
313: The continuum of measurement values $s_{1m}$ can then be represented by
314: a discrete set of three values at $s_1=\pm 1$ and $s_1=0$. The joint
315: probabilities for these three values of $s_1$ and the two eigenvalues
316: of $s_2$ read
317: \begin{eqnarray}
318: \label{eq:joint0}
319: P(s_1=-1;s_2=-1)=1/4 &\hspace{1cm}& P(s_1=-1;s_2=+1)=1/4
320: \nonumber \\
321: P(s_1=0;s_2=-1)=-1/2 &\hspace{1cm}& P(s_1=0;s_2=+1)=1/2
322: \nonumber \\
323: P(s_1=+1;s_2=-1)=1/4 &\hspace{1cm}& P(s_1=+1;s_2=+1)=1/4.
324: \end{eqnarray}
325: These joint probabilities explain the non-classical features of the
326: quantum statistics obtained from the single photon polarization
327: measurement setup shown in figure \ref{setup1} for any value of the
328: measurement resolution $\delta\!s$.
329:
330: It should be noted that the measurement setup itself defines an
331: asymmetry between $\hat{s}_1$ and $\hat{s}_2$
332: since the non-eigenvalue of zero appears only in the statistics of
333: the initial finite resolution measurement of $\hat{s}_1$.
334: This dependence on the order of
335: measurement is reflected in the operator order dependence of quantum
336: mechanical expectation values. In order to identify the operator properties
337: responsible for the appearance of negative probabilities in the statistical
338: properties, it is useful to characterize the measurement
339: statistics in terms of the correlation between $s_{1m}^2$ and $s_2$,
340: \begin{eqnarray}
341: \label{eq:corr}
342: C(s_{1m}^2,s_2) &=& \langle s_{1m}^2 s_2 \rangle - \langle s_{1m}^2 \rangle
343: \langle s_2 \rangle
344: \nonumber \\[0.2cm]
345: &=& - 2 (\langle s_{1m}^2 \rangle-\delta\!s^2) \langle s_2 \rangle
346: \nonumber \\
347: &=& - 2 \exp\left(- \frac{1}{2\delta\!s^2}\right),
348: \end{eqnarray}
349: where $\langle \; \rangle$ denotes statistical averages over
350: actual measurement results.
351: This correlation may be expressed in terms of the operator expectation
352: values of $\mid \psi_{\mbox{in}}\rangle$ as
353: \begin{equation}
354: C(s_1^2,s_2) = \exp \left( - \frac{1}{2\delta\!s^2}\right)
355: \left(\langle \psi_{\mbox{in}}\mid \hat{s}_1\hat{s}_2\hat{s}_1
356: \mid \psi_{\mbox{in}} \rangle -
357: \langle \psi_{\mbox{in}} \mid \hat{s}_1^2 \mid \psi_{\mbox{in}} \rangle
358: \langle\psi_{\mbox{in}} \mid \hat{s}_2 \mid \psi_{\mbox{in}}\rangle \right).
359: \end{equation}
360: As explained above, the exponential factor expresses the randomization of
361: $\hat{s}_2$ induced by the measurement of $\hat{s}_1$ according to the
362: uncertainty principle. For $\delta\!s \to \infty$, the noise introduced in
363: the measurement of $\hat{s}_1$ goes to zero and the correlation is
364: given by the operator expectation values of the initial state. Due to the
365: operator ordering, the anti-correlation between $s_1^2$ and $s_2$ is
366: an inherent statistical property of $\mid \psi_{\mbox{in}}\rangle$
367: even though $\mid \psi_{\mbox{in}}\rangle$ is an eigenstate
368: of $\hat{s}_2$.
369: Thus operator ordering allows a correlation between
370: fluctuating properties and seemingly well defined operator variables of
371: the quantum state.
372:
373: This property implies that even the eigenvalues of a
374: quantum state do not represent ``elements of reality''. Consequently,
375: it is wrong to assign measurement values to physical properties before
376: the measurement {\it even if the measurement result can be predicted
377: with certainty}. Since the violation of Bell's inequalities depends
378: on the assignment of such ``elements of reality'' it is not surprising
379: that it can be violated by quantum theory.
380: In the following, it will be shown how the violation of Bell's inequalities
381: can be explained in terms of negative joint probabilities obtained from
382: finite resolution measurements.
383:
384: \section{Measurement of polarization entanglement}
385: \label{sec:entangle}
386: \subsection{Entangled photons}
387: Entangled photon pairs can be created in two photon emission
388: \cite{Fre72,Asp82} or in parametric down conversion \cite{Ou88,Shi88}.
389: The precise polarization statistics
390: may vary depending on the geometry of the setup. In order to express
391: the violation of Bell's inequalities in terms of the Stokes parameters
392: $\hat{s}_1$ and $\hat{s}_2$, it is useful to rotate the polarizations
393: of the two photons in such a way that the quantum state is given by
394: \begin{equation}
395: \label{eq:eprstate}
396: \mid \psi_{a,b}\rangle = \frac{1}{\sqrt{2}}\left(\mid R;L \rangle
397: + \exp\left(-i \frac{\pi}{4}\right) \mid L;R \rangle \right).
398: \end{equation}
399: This state is an eigenstate of two polarization correlations,
400: \begin{eqnarray}
401: \frac{1}{\sqrt{2}}\left(\hat{s}_1(a)+\hat{s}_2(a)\right)\;\hat{s}_1(b)
402: \mid \psi_{a,b}\rangle &=& \mid \psi_{a,b}\rangle
403: \nonumber \\
404: - \frac{1}{\sqrt{2}}\left(\hat{s}_1(a)-\hat{s}_2(a)\right)\;\hat{s}_2(b)
405: \mid \psi_{a,b}\rangle &=& \mid \psi_{a,b}\rangle.
406: \end{eqnarray}
407: The sum of these two eigenvalues violates a Bell's inequality of the form
408: \begin{equation}
409: \label{ineq:bell}
410: K = s_1(a)s_1(b)+s_2(a)s_1(b)-s_1(a)s_2(b)+s_2(a)s_2(b) \leq 2.
411: \end{equation}
412: It is therefore not possible to interpret the polarization statistics by
413: assigning eigenvalues of $\pm 1$ to each Stokes parameter. However, as
414: indicated by the results of finite resolution measurements on the
415: polarization of single photons discussed in section \ref{sec:postulate}
416: above, such an identification of physical properties with their eigenvalues
417: is not even consistent with the correlated statistics of local single
418: photon properties.
419: The non-classical statistical properties responsible for the violation
420: of Bell's inequality (\ref{ineq:bell}) can be derived in detail
421: by applying the finite resolution measurement setup introduced above
422: to realize a polarization measurement on the entangled photon pairs
423: given by equation (\ref{eq:eprstate}).
424:
425: \subsection{Experimental setup and measurement statistics}
426: Figure \ref{setup2} shows the experimental setup for a measurement of the
427: correlations in the Bell's inequality (\ref{ineq:bell}). The detector arrays
428: record coincidence counts between the right and left hand side. Each detector
429: array corresponds to an eigenvalue measurement of $\hat{s}_2$. The spatial
430: coordinate at which the photon is registered corresponds to the continuous
431: measurement value $s_{1m}$ of $\hat{s}_1$. Each measurement result can then be
432: identified with a point in one of four two dimensional graphs.
433: The probability distribution for the measurement outcomes of the joint
434: measurements may be determined by projections onto the non-orthogonal,
435: non-normalized set of states
436: \begin{eqnarray}
437: \lefteqn{
438: \mid s_{1m}(a); s_{1m}(b);s_2(a)=\pm 1;s_2(b)=\pm 1
439: \rangle =} \nonumber \\ &&
440: \hat{P}_{\delta\!s}(s_{1m}(a))\hat{P}_{\delta\!s}(s_{1m}(b))
441: \frac{1}{2}\left(\mid R;R \rangle + s_2(a) i \mid L;R \rangle
442: + s_2(b) i \mid R;L \rangle - s_2(a)s_2(b) \mid L;L \rangle
443: \right).
444: \end{eqnarray}
445: In their most compact form,
446: the joint probabilities of the measurement results $s_{1m}(a)$,
447: $s_{1m}(b)$, $s_2(a)$, and $s_2(b)$ for the entangled input state
448: $\mid \psi(a,b)\rangle$ given by equation (\ref{eq:eprstate})
449: read
450: \begin{eqnarray}
451: \label{eq:totprop}
452: \lefteqn{P(s_{1m}(a); s_{1m}(b);s_2(a)=\pm 1;s_2(b)=\pm 1)=}
453: \nonumber \\ &\hspace{1cm}&
454: \frac{\sqrt{2}}{16\pi \delta\!s^2}
455: \exp\left(-\frac{s_{1m}(a)^2+s_{1m}(b)^2+2}{2\delta\!s^2}\right)
456: \nonumber \\ && \times \Big(
457: 2 \sinh\left(\frac{s_{1m}(a)s_2(b)-s_{1m}(b)s_2(a)}{2 \delta\!s^2}\right)
458: \cosh\left(\frac{s_{1m}(a)s_2(b)+s_{1m}(b)s_2(a)}{2 \delta\!s^2}\right)
459: \nonumber \\&& +
460: \left(\sqrt{2}+s_2(a)s_2(b)\right)
461: \cosh^2\left(\frac{s_{1m}(a)s_2(b)+s_{1m}(b)s_2(a)}{2 \delta\!s^2}\right)
462: \nonumber \\&& +
463: \left(\sqrt{2}-s_2(a)s_2(b)\right)
464: \sinh^2\left(\frac{s_{1m}(a)s_2(b)-s_{1m}(b)s_2(a)}{2 \delta\!s^2}\right)
465: \Big).
466: \end{eqnarray}
467: Figure \ref{condprop2} shows the results for a measurement resolution of
468: $\delta\!s=0.6$. At this intermediate resolution, quantum mechanical
469: interference effects are especially visible \cite{Hof00b}.
470: In particular, separate peaks can be resolved clearly, but quantum
471: interference effects are visible in the asymmetric peak shapes
472: and in the zero probability valleys in the $s_{1m}\approx 0$ regions
473: separating the peaks. As in the single photon case
474: discussed in section \ref{sec:postulate} above, it is indeed possible
475: to interpret these features entirely in terms of Gaussian distributions.
476: However, negative probability contributions centered around values of
477: $s_{1m}=0$ have to be included in order to explain the asymmetries and
478: the regions of extremely low probabilities near $s_{1m}=0$ separating
479: the peaks corresponding to quantized results around $s_{1m}=\pm1$.
480:
481: \subsection{Violation of Bell's inequality by the finite resolution
482: measurement statistics}
483: As in the one photon case, the regions of low probability at values of
484: $s_{1m}(a/b)=0$ can be traced
485: back to negative joint probabilities. The measurement probabilities
486: given by equation (\ref{eq:totprop}) may be expressed as a sum of shifted
487: normalized Gaussian distributions
488: \begin{equation}
489: G_{\delta\!s}(s_{1m}(a)-d_a;s_{1m}(b)-d_b) :=
490: \left(2\pi \delta\!s^2\right)^{-1}
491: \exp\left(-\frac{(s_{1m}(a)^2-d_a)^2+
492: (s_{1m}(b)^2-d_b)^2}{2\delta\! s^2}\right).
493: \end{equation}
494: Since the shifts $d_a$ and $d_b$ may be $-1$, $0$ or $+1$, respectively,
495: each of the four sums has nine components associated with joint probabilities
496: of the four Stokes parameters. The probability distribution of the measurement
497: results is then given by the sum
498: \begin{eqnarray}
499: \label{eq:eprcond}
500: \lefteqn{P(s_{1m}(a);s_{1m}(b);s_2(a)=\pm1;s_2(b)=\pm1)
501: =}
502: \nonumber \\[0.2cm] &\hspace{1cm}&
503: \frac{\sqrt{2}+1}{16\sqrt{2}}
504: \left(G_{\delta\!s}\left(s_{1m}(a)+1;s_{1m}(b)+1\right)+
505: G_{\delta\!s}\left(s_{1m}(a)-1;s_{1m}(b)-1\right)\right)
506: \nonumber \\ && +
507: \frac{\sqrt{2}-1}{16\sqrt{2}}
508: \left(G_{\delta\!s}\left(s_{1m}(a)+1;s_{1m}(b)-1\right)+
509: G_{\delta\!s}\left(s_{1m}(a)-1;s_{1m}(b)+1\right)\right)
510: \nonumber \\ && + \exp\left(-\frac{1}{2\delta\!s^2}\right)
511: \frac{1}{8\sqrt{2}} s_2(b)
512: \left(G_{\delta\!s}\left(s_{1m}(a)+1;s_{1m}(b)\right) -
513: G_{\delta\!s}\left(s_{1m}(a)-1;s_{1m}(b)\right)\right)
514: \nonumber \\ && - \exp\left(-\frac{1}{2\delta\!s^2}\right)
515: \frac{1}{8\sqrt{2}}s_2(a)
516: \left(G_{\delta\!s}\left(s_{1m}(a);s_{1m}(b)+1\right) -
517: G_{\delta\!s}\left(s_{1m}(a);s_{1m}(b)-1\right)\right)
518: \nonumber \\ && + \exp\left(-\frac{1}{\delta\!s^2}\right)
519: \frac{1}{4\sqrt{2}} s_2(a) s_2(b)\;
520: G_{\delta\!s}\left(s_{1m}(a);s_{1m}(b)\right).
521: \end{eqnarray}
522: Using this decomposition, it is a straightforward matter to determine the
523: averages corresponding to the correlations of the Bell's inequality
524: (\ref{ineq:bell}) by summing over $s_2(a)$ and $s_2(b)$ and integrating over
525: the continuous results $s_{1m}(a)$ and $s_{1m}(b)$. The result reads
526: \begin{equation}
527: \label{eq:kav}
528: \langle K \rangle =
529: \frac{1}{\sqrt{2}}\left(1+\exp\left(-\frac{1}{2\delta\!s^2}\right)\right)^2.
530: \end{equation}
531: This expectation value exceeds the maximal value of 2 allowed by inequality
532: (\ref{ineq:bell}) for measurement resolutions of $\delta\!s > 1.143$.
533: The violation of Bell's inequality can therefore be obtained directly from
534: the measurement statistics for sufficiently low resolutions of the
535: $\hat{s}_1$ measurements. An example for this direct violation of
536: Bell's inequality is shown in figure \ref{lowres} for a measurement
537: resolution of $\delta\! s=2$. At this low resolution, quantization effects
538: are not resolved. The non-classical properties of the statistics are
539: observable in the shift of the peak maxima for $s_2(a)= - s_2(b)$ to values
540: greater than $s_{1m}=+1$ or lower than $s_{1m}=-1$. Specifically, the
541: maximum probability density for $s_2(a)=+1$ and $s_2(b)=-1$ is at
542: $s_{1m}(a)==s_{1m}(a)=1.383$ and the maximum for $s_2(a)=-1$ and
543: $s_2(b)=+1$ is at $s_{1m}(a)==s_{1m}(a)=-1.383$. The value of $K$
544: at these points would be equal to 3.68.
545:
546: While it might be tempting to interpret the statistics in terms of
547: polarization components greater than $+1$ or smaller than $-1$,
548: the high resolution results of figure \ref{condprop2} and the
549: analysis of single photon polarization in section \ref{sec:postulate}
550: suggests that the true reason for the shifted peaks are negative
551: probabilities around $s_{1m}(a)=s_{1m}(b)=0$. In order to obtain
552: a consistent interpretation of the measurement results for both
553: high and low resolutions, it is necessary to identify the decoherence
554: factor $\exp(1/(2\delta\!s^2))$ with the quantum noise induced
555: reduction of the expectation values of $\hat{s}_2(a)$ and $\hat{s}_2(b)$.
556: It is then possible to remove the effects of noise and of finite measurement
557: resolution from the measurement statistics, tracing the violation of
558: Bell's inequality directly to the appearance of negative probabilities in the
559: joint probabilities for $s_1(a)$,$s_2(a)$,$s_1(b)$, and $s_2(b)$.
560:
561: \section{Disentangling entanglement: interpretation of the non-classical
562: statistics}
563: \label{sec:quantcorr}
564: \subsection{Negative conditional probabilities in photon entanglement}
565: As in the case of single photon polarization discussed in section
566: \ref{sec:postulate}, the sum of Gaussians given in equation
567: (\ref{eq:eprcond}) can be interpreted in terms of joint probabilities
568: for $s_1(a)$,$s_2(a)$,$s_1(b)$, and $s_2(b)$ by
569: identifying the average of each Gaussian with the appropriate value of
570: $s_1$. The joint probabilities for all 36 combinations of the six
571: contributions from $s_1(a)$ and $s_2(a)$ with the six contributions from
572: $s_1(b)$ and $s_2(b)$ characterizing the statistics of the
573: entangled state $\mid \psi_{a,b}\rangle$ are shown in table \ref{negprop}.
574: From these probabilities, the statistical weight of different
575: contributions to the sum correlation $K$ in inequality (\ref{ineq:bell})
576: can be determined.
577:
578: The joint probabilities can be classified according to whether the values
579: of $s_1(a)$ and $s_1(b)$ are zero or not. There are sixteen contributions with
580: both $s_1(a)$ and $s_1(b)$ non-zero. These cases correspond to the classical
581: expectation that the values of $s_1$ should be equal to the eigenvalues
582: observed in high resolution measurements. Consequently, they are the only
583: contributions that are not diminished by the decoherence factor for small
584: $\delta\!s$. Moreover, their probabilities are all positive. In eight of
585: these sixteen cases, three of the four correlations in inequality
586: (\ref{ineq:bell}) are equal to $+1$ and one is equal to $-1$,
587: for a total of $K=2$. The reverse is true for the remaining eight cases,
588: resulting in a total of $K=-2$ for the sum of correlations in inequality
589: (\ref{ineq:bell}). The probability of each case is equal to
590: $(2+s_1(a)s_1(b)\sqrt{2})/32$. Summing up the probability of the eight
591: cases with $K=2$ thus results in a total probability of $(4+\sqrt{2})/8$ or
592: roughly 67.7 \%. The eight cases with $K=-2$ have a total probability
593: of $(4-\sqrt{2})/8$, or 32.3 \%. The average value of $K$ for these
594: ``classical'' contributions to the joint probability is therefore equal
595: to $1/\sqrt{2}$, as evidenced by the limit of equation (\ref{eq:kav})
596: for $\delta\!s \to 0$. Obviously, the violation of Bell's inequality must
597: originate from the remaining twenty contributions with at least one value
598: of $s_1$ equal to zero.
599:
600: There are sixteen contributions with one value of $s_1$ equal to zero and
601: the other value non-zero. Two of the four correlations in the inequality
602: (\ref{ineq:bell}) are then equal to zero, while the other two may be either
603: plus or minus one each. In four cases, they are both equal to plus one ($K=2$),
604: in eight cases, they have opposite sign ($K=0$), and in the remaining four
605: cases, they are both equal to minus one ($K=-2$). The probabilities for these
606: cases are $\pm \sqrt{2}/16$. As a result, the total probability for the four
607: cases with $K=2$ is equal to $\sqrt{2}/4$ or 35.4\%, the total probabilities
608: for the eight cases with $K=0$ cancel to zero, and the total probability for
609: $K=-2$ is $-\sqrt{2}/4$ or -35.4\%. This negative probability more than
610: outweighs the 32.3 \% of the classical contributions, explaining the increase
611: of the expectation value of $K$ beyond the limit of 2. However, the
612: effect is further enhanced by the contributions from $s_1(a)=s_1(b)=0$.
613:
614: There are four contributions with $s_1(a)=s_1(b)=0$. Only the correlation
615: $\langle s_2(a)s_2(b)\rangle$ is non-zero in these cases. Two cases
616: have $K=1$ and a positive probability of $\sqrt{2}/8$, and two cases have
617: $K=-1$ and a negative probability of $-\sqrt{2}/8$. This adds a total
618: probability of 35.4 \% for $K=1$ and -35.4 \% for $K=-1$.
619:
620: The probability distribution over values of $K$ can be summarized as
621: follows:
622: \begin{eqnarray}
623: P(K=2) = 103.1 \% &\hspace{1cm}& P(K=-2) = -3.1 \%
624: \nonumber \\
625: P(K=1) = 35.4 \% && P(K=-1) = -35.4 \%
626: \nonumber \\
627: P(K=0) = 0 \%. &&
628: \end{eqnarray}
629: The high expectation value of $K$ is a result of the negative probabilities
630: for combinations of $s_1(a)$,$s_2(a)$,$s_1(b)$, and $s_2(b)$ with $K<0$.
631: In the measured probability distributions described by equations
632: (\ref{eq:totprop}) and (\ref{eq:eprcond}), these negative probabilities appear
633: as a suppression of the probability for values of $s_1$ close to zero,
634: pushing the peak of the probability distribution beyond the eigenvalue
635: limit of $\pm 1$. Since $s_{1m}$ is not restricted to eigenvalues of
636: $\hat{s}_1$, the contributions to the expectation value of $K$ taken from
637: the measured distribution (\ref{eq:totprop}) shown in figure \ref{condprop2}
638: may exceed the classical limit.
639: Even though a direct observation of the negative
640: probabilities is of course impossible, the continuous distribution of finite
641: resolution measurement results thus reveals clear evidence of these
642: non-classical statistical features.
643:
644: \subsection{Quantum noise and negative probabilities in entangled systems}
645:
646: The negative conditional probabilities shown in table \ref{negprop}
647: allow an interpretation of the measurement statistics in terms of
648: individual measurement results observed separately in branch $a$ and
649: in branch $b$. There is neither a need for action at a distance, nor
650: for non-local properties. The non-classical feature required to
651: explain the violation of Bell's inequalities is expressed in the
652: negative probabilities which are possible even in individual quantum
653: systems because the uncertainty principle does not allow an isolated
654: measurement of a joint probability of non-commuting variables.
655:
656: Once the relationship between uncertainty and negative conditional
657: probabilities is understood, the problem of non-locality in
658: entangled systems can be resolved by introducing local decompositions
659: of the entangled state density matrix based on negative probability
660: components of the local density matrices. For the state discussed
661: above, one possible decomposition reads
662: \begin{eqnarray}
663: \mid \psi_{a,b} \rangle \langle \psi_{a,b} \mid &=&
664: \frac{1}{4} \hat{1}(a) \otimes \hat{1}(b)
665: + \frac{1}{4\sqrt{2}} (\hat{s}_1(a) + \hat{s}_2(a))\otimes \hat{s}_1(b)
666: \nonumber \\ &&
667: + \frac{1}{4\sqrt{2}} (\hat{s}_1(a) - \hat{s}_2(a))\otimes \hat{s}_2(b)
668: - \frac{1}{4} \hat{s}_3(a) \otimes \hat{s}_3(b).
669: \end{eqnarray}
670: All by themselves, the Stokes parameter operators $\hat{s}_i$ would not
671: qualify as density matrices because of their negative eigenvalues.
672: Once negative eigenvalue components are permitted, however, the
673: decomposition given above can be interpreted as a separation of
674: the entangled density matrix into products of local density matrices.
675: The reason why density matrices with negative probability eigenvalues
676: may be used in the decomposition of entangled states is that any
677: measurement performed on system $a$ mixes contributions to the density
678: matrix of system $b$ in such a way that the information required to
679: isolate the negative conditional probabilities represented by the
680: negative eigenvalues is lost.
681:
682: Effectively, the uncertainty in system
683: $a$ necessarily ``covers up'' the negative eigenvalues of the density
684: matrix components of system $b$ by mixing them with positive components.
685: Entanglement can therefore be explained by the local properties of
686: quantum measurements described previously \cite{Hof00a}. In the light
687: of this result, it is not surprising that some applications of quantum
688: mechanics such as quantum computation can work without entanglement
689: \cite{Fit00,Bra99}. The most fundamental property of quantum mechanics
690: is not entanglement, but local non-classical correlations represented by
691: the operator-ordering dependence of expectation values and the negative
692: conditional probabilities obtained from finite resolution measurements.
693:
694: \section{Conclusions}
695: \label{sec:concl}
696: The interpretation of quantum statistics cannot be based on the
697: assumption that potential measurement results represent
698: ``elements of reality'' whether the actual measurement is performed
699: or not. This is not only true for entangled systems, but also for
700: combinations of finite resolution measurements performed to obtain the
701: correlations between non-commuting operator variables in a single
702: quantum system. As a result, concepts such as photon polarization have
703: to be reviewed critically in order to understand the relationship between
704: eigenvalues and operator variables.
705:
706: The experimental approach proposed
707: above allows a direct determination of the non-classical features of the
708: polarization statistics for both single photons and entangled pairs.
709: Its application to the violation of Bell's inequalities reveals details
710: of the statistical relationships between all four polarization components.
711: A full set of conditional probabilities can be obtained from the statistics
712: of a single measurement, revealing the negative conditional probabilities
713: that are responsible for the violation of Bell's inequalities. A comparison
714: with the single photon polarization statistics reveals that such negative
715: probabilities are also observable in the polarization of a single photon.
716: The property responsible for the violation of Bell's inequalities is
717: therefore a local feature of quantum statistics. Once the implications of
718: the operator formalism are accepted, entanglement can be understood as a
719: special case of the non-classical features observable in local correlations.
720:
721: \section*{Acknowledgment}
722: The author would like to acknowledge support from the Japanese Society
723: for the Promotion of Science, JSPS, and thank Dr. Takao Fuji for
724: helpful discussions of the experimental aspects.
725:
726: %=========================================================
727:
728: \begin{thebibliography}{5}
729: \bibitem{Bel64} J.S. Bell, Physics (Long Island City, N.Y.) {\bf 1},195 (1964).
730:
731: \bibitem{Fre72}
732: S.J. Freedman and J.F. Clausner, Phys. Rev. Lett. {\bf 28}, 938 (1972).
733:
734: \bibitem{Asp82}
735: A. Aspect, J. Dalibard, and G. Roger, Phys. Rev. Lett. {\bf 49}, 1804 (1982).
736:
737: \bibitem{Ou88}
738: Z.Y. Ou, L. Mandel, Phys. Rev. Lett. {\bf 61}, 50 (1988).
739:
740: \bibitem{Shi88}
741: Y.H. Shih, C. Alley, Phys. Rev. Lett. {\bf 61}, 2921 (1988).
742:
743: \bibitem{Hof00a}
744: H.F. Hofmann, Phys. Rev. A {\bf 62}, 022103 (2000).
745:
746: \bibitem{Hof00b}
747: H.F. Hofmann, Phys. Rev. A {\bf 61}, 033815 (2000).
748:
749: \bibitem{Hof00c}
750: H.F. Hofmann, T. Kobayashi, and A. Furusawa, Phys. Rev. A {\bf 62}, 013806
751: (2000).
752:
753: \bibitem{EPR}
754: A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. {\bf 47}, 777 (1935).
755:
756: \bibitem{Mer}
757: N.D. Mermin, Phys. Today {\bf 43} (6), 9 (1990).
758:
759: \bibitem{Fit00}
760: R. Fitzgerald, Physics Today {\bf 53}(1), 20(2000).
761:
762: \bibitem{Bra99}
763: S.L. Braunstein, C.M. Caves, R. Jozsa, N. Linden, S.Popescu, and R. Schack,
764: Phys. Rev. Lett. {\bf 83}, 1054 (1999).
765: \end{thebibliography}
766:
767: \begin{figure}
768: \caption{\label{setup1} Schematic representation of the experimental
769: setup for a joint measurement of non-orthogonal polarizations.
770: The beam displacer separates the incoming light into two parallel beams.
771: The polarization is then rotated by an angle of $\pi/4$ before the
772: light beam is split at the polarizer. The overlapping transversal profile
773: of the beams is illustrated at the detector arrays.}
774: \end{figure}
775: %
776: \begin{figure}
777: \caption{\label{condprop1} Probability distribution $P(s_{1m};s_2)$ for
778: an initial eigenstate of $s_2=+1$ at a resolution of $\delta\!s=0.6$.
779: Note the asymmetry and the shifted maxima obtained for $s_2=-1$.}
780: \end{figure}
781: %
782: \begin{figure}
783: \caption{\label{setup2} Schematic representation of the experimental
784: setup for a measurement of polarization correlations on entangled photons.
785: The setup of the branches $a$ and $b$ are as shown in figure 1. Coincidence
786: counts are registered in one of four channels as illustrated.
787: }
788: \end{figure}
789: %
790: \begin{figure}
791: \caption{\label{condprop2} Contour plot of the probability distribution
792: $P(s_{1m}(a);s_{1m}(b);s_2(a);s_2(b))$ at a resolution of $\delta\!s=0.6$.
793: While the major peaks appear to be close to the eigenvalues at
794: $s_{1m}=\pm1$, the shape of the peaks and the separation between them
795: reveals the same non-classical statistical effects seen in figure 2.
796: }
797: \end{figure}
798: %
799: \begin{figure}
800: \caption{\label{lowres} Contour plot of the probability distribution
801: $P(s_{1m}(a);s_{1m}(b);s_2(a);s_2(b))$ at a resolution of $\delta\!s=2$.
802: The peaks for $s_2(a)=-s_2(b)$ are at $s_{1m}(a)=s_{1m}(b)=\pm 1.383$.
803: The contribution to the average value of $K$ at these points is
804: 3.68.}
805: \end{figure}
806: %
807: \begin{table}
808: \caption{\label{negprop} Table of conditional probabilities
809: derived from the results shown in figure 4. Note that the negative
810: probabilities roughly coincide with regions of zero probability in
811: the measurement statistics.}
812: \end{table}
813: %
814: %=========================================================
815:
816: \end{document}
817:
818:
819:
820:
821:
822:
823:
824:
825:
826:
827: