Directed flow at RHIC from Lee-Yang zeroes
Extracted source text ยท arXiv nucl-th/0404087
\documentclass[aps,prc,superscriptaddress,showpacs,nofootinbib]{revtex4}
\usepackage{amsmath}
\newcommand{\mean}[1]{\left\langle #1 \right\rangle}
\newcommand{\smean}[1]{\langle #1 \rangle}
\newcommand{\sample}[1]{\left\{ #1 \right\}}
\newcommand{\N}{N_{\rm evts}}
\newcommand{\Np}{N'}
\begin{document}
\preprint{Saclay-T04/056}
\title{Directed flow at RHIC from Lee--Yang zeroes}
\author{N. Borghini}
\email{borghini@spht.saclay.cea.fr}
\author{J.-Y. Ollitrault}
\email{Ollitrault@cea.fr}
\altaffiliation[also at ]{L.P.N.H.E., Universit{\'e} Pierre et Marie Curie,
4 place Jussieu, F-75252 Paris cedex 05, France}
\affiliation{Service de Physique Th{\'e}orique, CEA-Saclay,
F-91191 Gif-sur-Yvette cedex, France}
\date{\today}
\begin{abstract}
Directed flow in ultrarelativistic nucleus-nucleus collisions
is analyzed using the event plane from elliptic flow,
which reduces the bias from nonflow effects.
We combine this method with the determination of elliptic
flow from Lee--Yang zeroes.
The resulting method is more consistent and somewhat
easier to implement than the previously used method based
on three-particle cumulants, and is also less biased by
nonflow correlations. Error terms from residual nonflow
correlations are carefully estimated, as well as statistical
errors. We discuss the application of the method at
RHIC and LHC.
\end{abstract}
\pacs{25.75.Ld, 25.75.Gz, 05.70.Fh}
\maketitle
\section{Introduction}
The standard method of analyzing directed flow ($v_1$) in
nucleus-nucleus collisions using an
estimate of the reaction plane~\cite{Danielewicz:hn,Poskanzer:1998yz}
was shown to be inadequate at ultrarelativistic
energies~\cite{Borghini:2000cm}, due to the smallness of
directed flow and the relatively large magnitude of
``nonflow'' effects~\cite{Poskanzer:1998yz,Ollitrault:dy}.
Trivial nonflow effects such as momentum conservation can
be taken into account in the analysis \cite{Borghini:2000cm},
and the standard analysis can be modified for this
purpose~\cite{Borghini:2002mv}.
A more systematic, model-independent way of eliminating
nonflow effects was introduced~\cite{Borghini:2002vp},
based on the observation that the three-particle average
$\smean{\cos(\phi_1+\phi_2-2\phi_3)}$ (where
$\phi_1$, $\phi_2$ and $\phi_3$ denote the azimuthal
angles of three particles emitted in a collision,
and $\mean{\cdots}$ denotes an average over triplets of
particles and events) is much less sensitive
to nonflow effects than the two-particle correlation
$\smean{\cos(\phi_1-\phi_2)}$ used in the standard analysis.
Since $\smean{\cos(\phi_1+\phi_2-2\phi_3)}\propto v_1^2 v_2$,
where the elliptic flow $v_2$ is large at ultrarelativistic
energies, this provides an alternative way of analyzing
directed flow.
This method was first implemented at the CERN
SPS by the NA49 Collaboration~\cite{Borghini:2002hm,Alt:2003ab},
and recently led to the discovery of directed flow at
RHIC~\cite{Adams:2003zg}.
Directed flow at ultrarelativistic energies is interesting
in itself~\cite{Snellings:1999bt}; in addition, it is the
only way of measuring the sign of elliptic flow, and to check
experimentally that it is positive~\cite{Ollitrault:bk}.
The practical implementation of the three-particle method presented
in Ref.~\cite{Borghini:2002vp} is rather cumbersome and requires
to estimate elliptic flow $v_2$ independently, using
cumulants~\cite{Borghini:2000sa} to avoid nonflow effects.
In this paper, we suggest to analyze simultaneously elliptic
and directed flows using the recently introduced method of
Lee--Yang zeroes~\cite{Bhalerao:2003yq,Borghini:2004ke} to analyze
elliptic flow.
This minimizes the bias from nonflow effects for both $v_1$ and $v_2$.
The practical recipe is presented in Sec.~\ref{s:implementation}.
The theoretical background is briefly discussed in Sec.~\ref{s:theory}.
In Sec.~\ref{s:errors},
we derive the general order of magnitude of the systematic
error due to nonflow correlations, which was underestimated
in Ref.~\cite{Borghini:2002vp}. We also give analytical expressions
of statistical errors.
\section{Implementation}
\label{s:implementation}
Let us first define useful quantities and notations.
For a given event, we define the following complex-valued function
\begin{equation}
\label{defgenfunc}
g_\epsilon^{\theta_1,\theta_2}(z)=\prod_{j=1}^M\left[1
+z \epsilon w_{1}(j)\cos(\phi_j-\theta_1)
+z w_{2}(j)\cos(2(\phi_j-\theta_2))\right]
\end{equation}
where $\theta_1$ and $\theta_2$ are angles, $\epsilon$ is a
real parameter, $z$ is a complex variable, $\phi_j$ are
the azimuthal angles of the particles
(measured using a fixed reference in the laboratory),
and the product runs over all detected particles.
As in other methods of analysis, $w_1$ and $w_2$ are weights appropriate
to directed and elliptic flows, respectively, and can be any functions
of particle type, transverse momentum $p_T$ and rapidity $y$.
In Eq.~(\ref{defgenfunc}), $w_n(j)$ is a shorthand for $w_n({p_T}_j,y_j)$.
The best weight is the flow itself~\cite{Danielewicz:1995,Borghini:2000sa},
$w_n(p_T,y) = v_n(p_T,y)$, where $v_n(p_t,y)$ denotes the value of the
flow in a small $(p_T,y)$ bin.
In practice, one can choose as a first guess the center-of-mass
rapidity for directed flow, $w_1=y-y_{\rm CM}$,
and the transverse momentum for elliptic flow $w_2=p_T$,
in regions of phase space covered by the detector acceptance.
The average of $g^{\theta_1,\theta_2}_\epsilon(z)$
over events in a centrality bin will be denoted
by $\smean{g^{\theta_1,\theta_2}_\epsilon(z)}_{\rm evts}$.
The same notation holds for any observable associated with
the event.
The implementation of the method consists of three logical steps,
as in the 3-particle correlation method~\cite{Borghini:2002vp}.
The first one is to analyze {\it integrated\/} elliptic flow
$V_2$. By ``integrated,'' we mean that it is
summed over all detected particles with appropriate
weights, and averaged over many events:
\begin{equation}
\label{defV2}
V_2\equiv\mean{\sum_{j=1}^M w_2(j)\cos(2(\phi_j-\Phi_R))}_{\!\!\!\rm evts},
\end{equation}
where the sum runs over all particles detected in an event and
$\Phi_R$ is the azimuthal angle of the reaction plane
of the event.\footnote{$\Phi_R$ is the {\it exact\/} reaction
plane, not an estimated plane; $\Phi_R$ is unknown on an event-by-event
basis.}
The second step is the analysis of integrated directed flow
$V_1$, which is defined similarly:
\begin{equation}
\label{defV1}
V_1\equiv\mean{\sum_{j=1}^M w_1(j)\cos(\phi_j-\Phi_R)}_{\!\!\!\rm evts}.
\end{equation}
Finally, in a third step, integrated values are used as a reference
to analyze directed flow differentially, as a function of
transverse momentum and rapidity.
In practice, the method requires only two passes through the data:
the first pass corresponds to Sec.~\ref{s:IIA}, while
the computations in Sec.~\ref{s:IIB} and Sec.~\ref{s:IIC}
can be done in a single second pass.
\subsection{Integrated elliptic flow}
\label{s:IIA}
The analysis of integrated elliptic flow is identical to that presented in
Refs.~\cite{Bhalerao:2003yq,Borghini:2004ke}, to which we refer the
reader for further practical details.
Note, however, that it uses the generating function introduced in
Ref.~\cite{Borghini:2004ke}, not that of Ref.~\cite{Bhalerao:2003yq}.
Let us just give a short reminder of the recipe.
One must evaluate
$g_\epsilon ^{\theta_1,\theta_2}(z)$ for $\epsilon=\theta_1=0$,
for several (typically 4 or 5) equally spaced values of
$\theta_2$ between $0$ and $\pi/2$ (i.e., $\theta_2=0,\pi/(2n),\ldots,
(n-1)\pi/(2n)$ with $n=4$ or $5$)
and many values of $z$ on the imaginary axis, $z={\rm i}r$.
One then plots the modulus
$|\smean{g^{0,\theta_2}_0({\rm i}r)}_{\rm evts}|$ for real, positive $r$
as a function of $r$, for a fixed value of $\theta_2$.
One determines numerically the value of $r$ corresponding
to the first minimum of this function, $r=r_0^{\theta_2}$.
The value of $r_0^{\theta_2}$ will be used in the analysis
of directed flow below.
It is closely related to elliptic flow, which can be estimated
through the following formula~\cite{Bhalerao:2003yq,Borghini:2004ke}:
\begin{equation}
\label{v2infty}
|V_2|=\mean{\frac{j_{01}}{r_0^{\theta_2}}}_{\!\!\theta_2},
\end{equation}
where $j_{01}\simeq 2.40483$ is the first root of the Bessel
function of the first kind $J_0(x)$.
The notation $\mean{\cdots}_{\theta_2}$ means an average
over $\theta_2$.
The analysis could in principle be performed with a single value of
$\theta_2$.
The averaging is only of practical importance: it reduces
the statistical error on $V_2$ by a factor of $\sim 2$.
Note that Eq.~(\ref{v2infty}) yields an estimate of the absolute
value of $V_2$, not of $V_2$ itself.
The sign of $V_2$ is determined simultaneously with
directed flow, as we shall see below.
\subsection{Integrated directed flow}
\label{s:IIB}
The analysis of integrated directed flow involves the generating function
$g_\epsilon ^{\theta_1,\theta_2}(z)$.
The angle $\theta_2$ takes the same values as in Sec.~\ref{s:IIA}.
For a given $\theta_2$, $z$ takes only one value,
$z={\rm i}r_0^{\theta_2}$.
For each event, one first evaluates the generating function for
$\epsilon=\theta_1=0$, as well as its derivative with respect to $z$:
\begin{equation}
\frac{\partial g^{0,\theta_2}_0}{\partial z}({\rm i}r_0^{\theta_2})=
g_0^{0,\theta_2}({\rm i}r_0^{\theta_2})
\sum_{j=1}^M\frac{w_{2}(j)\cos(2(\phi_j-\theta_2))}
{1+{\rm i}r_0^{\theta_2} w_{2}(j)\cos(2(\phi_j-\theta_2))}.
\end{equation}
One then evaluates $g_\epsilon ^{\theta_1,\theta_2}({\rm i}r_0^{\theta_2})$
for a single non-zero value of $\epsilon$, which is arbitrary
but must satisfy the following condition
\begin{equation}
\label{smallepsilon}
\epsilon\ll \frac{V_2}{V_1}.
\end{equation}
Since $V_1$ is a priori unknown, one must guess a reasonable
value and check afterwards that the condition is satisfied.
If $\epsilon$ is too small, numerical errors
may arise as the procedure amounts to expanding numerically
$g_\epsilon ^{\theta_1,\theta_2}(z)$ to order $\epsilon^2$.
Thus we recommend to perform tests with several values of $\epsilon$,
and to check the stability of the results, before doing the full
analysis.
The angle $\theta_1$ takes 5 (or more) equally spaced
values of $\theta_1$ between 0 and $2\pi$, (i.e.
$\theta_1=0,2\pi/n,\ldots, 2(n-1)\pi/n$ with $n\ge 5$).
Note that the range differs from that of $\theta_2$ values.
Our estimate of $V_1$ is defined by
\begin{equation}
\label{intv1}
(V_1)^2\,{\rm sgn}(V_2)=\mean{-\frac{8j_{01}}{\epsilon^2}
\left(r_0^{\theta_2}\right)^{\!-3}
{\rm Re}\left( \frac{\mean{
\cos(2(\theta_1-\theta_2))\,
g_\epsilon^{\theta_1,\theta_2}({\rm i}r_0^{\theta_2})}_{\!\theta_1,\rm evts}}
{\displaystyle
\mean{\frac{\partial g^{0,\theta_2}_0}{\partial z}({\rm i}r_0^{\theta_2})}
_{\!\!\rm evts}}\right)}_{\!\!\!\raisebox{11pt}{$\scriptstyle\theta_2$}},
\end{equation}
where $\epsilon$ is an arbitrary small number satisfying
condition~(\ref{smallepsilon}),
$\smean{\cdots}_{\theta_1,\rm evts}$ denotes an average over $\theta_1$
and events,
and Re is the real part of the ratio.
Finally, ${\rm sgn}(V_2)$ denotes the sign of $V_2$,
which is determined by this analysis of $V_1$.
The sign of $V_1$, on the other hand, is not measured
and must be postulated, as in any other method of analysis.
As in Sec.~\ref{s:IIA}, the averaging over $\theta_2$ in
Eqs.~(\ref{intv1}) and (\ref{diffv1}) is only for practical purposes:
it reduces statistical errors by a factor of $\sim 2$.
In contrast, the averaging over $\theta_1$ cannot be avoided, for reasons
we shall explain below in Sec.~\ref{s:theory}.
\subsection{Differential directed flow}
\label{s:IIC}
One can then turn to the analysis of differential flow, i.e.,
the flow of particles of a given type in a definite phase-space
window, which we shall call ``protons'' for the sake of brevity.
A ``proton'' azimuth will be denoted by $\psi$, and the corresponding
differential directed flow by $v'_1$.
The estimate of $v'_1$ involves the derivative of the
generating function with respect to the proton weight, evaluated at
$z={\rm i}r_0^{\theta_2}$:
\begin{equation}
\label{partialw}
\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})}
{\partial w_1(\psi)} =
g_\epsilon^{\theta_1,\theta_2}({\rm i}r_0^{\theta_2})
\frac{{\rm i}r_0^{\theta_2}\epsilon \cos(\psi-\theta_1)}
{1+{\rm i}r_0^{\theta_2}\epsilon w_{1}(\psi)\cos(\psi-\theta_1)
+{\rm i}r_0^{\theta_2} w_{2}(\psi)\cos(2(\psi-\theta_2))},
\end{equation}
and is defined by
\begin{equation}
\label{diffv1}
v'_1=\mean{-\frac{4j_{01}\,{\rm sgn}(V_2)}{V_1\epsilon^2}
\left(r_0^{\theta_2}\right)^{\!-3}
{\rm Re}\left( \frac{\displaystyle
\mean{\cos(2(\theta_1-\theta_2))\,\frac{
\partial g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})}
{\partial w_1(\psi)}}_{\!\!\theta_1,\psi}}
{\displaystyle
\mean{\frac{\partial g^{0,\theta_2}_0}{\partial z}({\rm i}r_0^{\theta_2})}
_{\!\!\rm evts}}\right)}_{\!\!\!\raisebox{12pt}{$\scriptstyle\theta_2$}},
\end{equation}
where $\smean{\cdots}_{\theta_1,\psi}$ denotes an average over protons
and over values of $\theta_1$.
Note that the value of integrated directed flow $V_1$ is only required after
averaging over protons or events, which means that the ratio between parentheses
can be computed at the same time as that in Eq.~(\ref{intv1}).
Moreover, the denominator of this ratio is the same as in Eq.~(\ref{intv1}),
so that in the second pass through data, one only need compute three
quantities for each event: this common denominator, the numerator of
Eq.~(\ref{intv1}) (actually, a value for each angle $\theta_1$), and the
numerator of Eq.~(\ref{diffv1}), taking into account only the ``protons.''
\subsection{Relation with the three-particle method of Ref.~\cite{Borghini:2002vp}}
\label{s:IID}
The three-particle method we proposed in Ref.~\cite{Borghini:2002vp} is
based on averages of the type $\mean{\cos(\phi_1+\phi_2-2\phi_3)}$,
where $\phi_1$, $\phi_2$ and $\phi_3$ are the azimuthal angles of three
particles belonging to the same event.
Let us explain how such three-particle averages appear in the
present method:
expanding $g_{\epsilon}^{\theta_1,\theta_2}(z)$ in
Eq.~(\ref{defgenfunc}) to order $\epsilon^2 z^3$ produces
terms of the type $\cos(\phi_1-\theta_1)\cos(\phi_2-\theta_1)
\cos(2(\phi_3-\theta_2))$. Multiplying by
$\cos(2(\theta_1-\theta_2))$ and averaging over $\theta_1$ and
$\theta_2$, as in Eqs.~(\ref{intv1}) and (\ref{diffv1}), one obtains
$\frac{1}{8}\cos(\phi_1+\phi_2-2\phi_3)$, thus recovering the
three-particle averages.
The present method uses an expansion to order $\epsilon^2$ (see
Sec.~\ref{s:theory}), but for a fixed value of $z$, namely
${\rm i}r_0^{\theta_2}$: it therefore also includes the information from
higher-order correlations, generated by higher powers of $z$, not only
three-particle correlations.
\section{Theory}
\label{s:theory}
The general philosophy is that zeroes of generating functions
are direct probes of collective effects~\cite{Bhalerao:2003yq},
while they are little sensitive to nonflow effects.
The key property of the generating function in
Eq.~(\ref{defgenfunc}) is its factorization property:
if a nucleus-nucleus collision can be viewed as
the superposition of independent subsystems containing
a few particles (for instance, independent nucleon-nucleon
collisions), i.e., if there are no collective effects,
the average over events
$\smean{g^{\theta_1,\theta_2}_\epsilon(z)}_{\rm evts}$
is the product of the contributions of individual subsystems.
The zeroes of the generating function are then simply the
zeroes of the contributions of the subsystems.
Since the value of the generating function is $1$ for $z=0$
by construction, the value of $z$ for which it vanishes
verifies $|z|\sim 1$ for dimensional reasons
(in this discussion, we assume unit weights $w_1=w_2=1$ for
simplicity).
If, on the other hand, there is collective flow,
nothing prevents the zeroes from moving much closer to
the origin, $|z|\ll 1$, and this is indeed what
happens: the positions of the zeroes scale with the
event multiplicity like $1/M$ (while they would be independent
of $M$ for independent subsystems).
This relates our method to the Lee--Yang theory of phase
transitions~\cite{Yang:be}, where the zeroes of the grand
partition function move closer and closer to the real
axis as the size of the system increases when, and only when,
there is a phase transition.
Let us now be more quantitative.
We first evaluate the average value of
$g_\epsilon^{\theta_1,\theta_2}(z)$ for events having exactly
the same reaction-plane orientation $\Phi_R$ (denoting such an average
by $\smean{\cdots|\Phi_R}$).
Taking the logarithm of Eq.~(\ref{defgenfunc}) and expanding to
order $z$, we obtain
\begin{eqnarray}
\label{gphiR}
\ln\mean{g_\epsilon^{\theta_1,\theta_2}(z)|\Phi_R}
&\simeq&z\epsilon
\mean{\sum_{j=1}^M w_{1}(j)\cos(\phi_j-\theta_1)|\Phi_R}
+z\mean{\sum_{j=1}^M w_{2}(j)\cos(2(\phi_j-\theta_2))|\Phi_R}\cr
&\simeq&z\epsilon V_1\cos(\Phi_R-\theta_1)+
z V_2\cos(2(\Phi_R-\theta_2)),
\end{eqnarray}
where we have used Eqs.~(\ref{defV2}) and (\ref{defV1})
and assumed symmetry with respect to the reaction plane,
which implies
$\smean{\sum_j w_2(j)\sin(2(\phi_j-\Phi_R))}_{\rm evts}=
\smean{\sum_j w_1(j)\sin(\phi_j-\Phi_R)}_{\rm evts}=0$.
Terms of order $z^2$ and higher in the expansion are responsible for
systematic errors that will be estimated in Sec.~\ref{s:systerrors}.
The average over events is eventually obtained by averaging
over $\Phi_R$, which is randomly distributed:
\begin{equation}
\label{avphiR}
\mean{g_\epsilon^{\theta_1,\theta_2}(z)}_{\rm evts}
=\int_{0}^{2\pi}\!\mean{g_\epsilon^{\theta_1,\theta_2}(z)|\Phi_R}
\frac{{\rm d}\Phi_R}{2\pi}.
\end{equation}
For $\epsilon=0$, substituting the estimate~(\ref{gphiR}) into
Eq.~(\ref{avphiR}), one obtains
\begin{equation}
\label{oldg0}
\mean{g_0^{0,\theta_2}(z)}_{\!\rm evts}=I_0(|V_2| z),
\end{equation}
where $I_0$ is the modified Bessel function of order 0. Its zeroes
lie on the imaginary axis, the first one being located
at ${\rm i} j_{01}$.
This explains how $|V_2|$ is obtained
in Sec.~\ref{s:IIA}, Eq.~(\ref{v2infty}).
Since zeroes are expected to be on the imaginary axis except
for irrelevant statistical fluctuations and/or detector
effects~\cite{Bhalerao:2003yq}, we suggest to look for the position of the first
minimum of $|\smean{g_0^{0,\theta_2}({\rm i}r)}_{\rm evts}|$ on the upper
imaginary axis, denoted by $z={\rm i}r_0^{\theta_2}$ in Sec.~\ref{s:IIA},
rather than that of the first zero of
$\smean{g_0^{0,\theta_2}(z)}_{\rm evts}$ in the complex plane.
In order to obtain an estimate of $V_1$, we shall now study how the
first zero of $\smean{g_\epsilon^{\theta_1,\theta_2}(z)}_{\rm evts}$,
which we denote by $z_{\epsilon}$, varies for small $\epsilon$.
For small enough $\epsilon$, we may write
\begin{equation}
\label{zepsilon}
z_\epsilon-z_0=
-\frac{\mean{g_\epsilon^{\theta_1,\theta_2}(z_0)}_{\rm evts}}
{\mean{\frac{\displaystyle\partial g^{0,\theta_2}_0}
{\displaystyle\partial z}(z_0)
}_{\!\!\rm evts}},
\end{equation}
where $z_0$ is the value of $z_\epsilon$ for $\epsilon=0$.
According to our general philosophy, the value of $z_\epsilon$
is directly related to collective flow, and little sensitive
to nonflow effects. Then, this is also true of the
right-hand side (r.h.s.) of Eq.~(\ref{zepsilon}), which is experimentally
measurable.
Following the discussion above, we replace the exact zero $z_0$
by ${\rm i}r_0^{\theta_2}$.
To relate the quantity in Eq.~(\ref{zepsilon}) to anisotropic flow,
we use our theoretical estimate of the generating function,
defined by Eqs.~(\ref{gphiR}) and (\ref{avphiR}).
The denominator is simply obtained by differentiating
Eq.~(\ref{oldg0}):
\begin{equation}
\label{denomflow}
\mean{\frac{\partial g^{0,\theta_2}_0}{\partial z}(z_0)}_{\!\!\!\rm evts}=
|V_2|\, I_1(|V_2| z_0).
\end{equation}
The numerator is obtained by substituting Eq.~(\ref{gphiR}) into
(\ref{avphiR}), and expanding to order $\epsilon^2$:
%integrating over $\Phi_R$:
\begin{equation}
\label{numflow}
\mean{g_\epsilon^{\theta_1,\theta_2}(z_0)}_{\rm evts}
=\frac{\epsilon^2}{4}z_0^2\, I_1(V_2 z_0)\,V_1^2\cos(2(\theta_1-\theta_2)).
\end{equation}
Since $I_1$ is an odd function, we may write
$I_1(V_2 z_0)=I_1(|V_2| z_0)\,{\rm sgn}(V_2)$.
Putting together the last two equations, replacing $z_0$ with
${\rm i}r_0^{\theta_2}$ and $|V_2|$ with its estimate
$j_{01}/r_0^{\theta_2}$ (see Eq.~(\ref{v2infty})), we obtain
\begin{equation}
V_1^2\,{\rm sgn}(V_2)\cos(2(\theta_1-\theta_2))=
-\frac{4 j_{01}}{\epsilon^2}\left(r_0^{\theta_2}\right)^{\!-3}
\frac{\mean{g_\epsilon^{\theta_1,\theta_2}({\rm i}r_0^{\theta_2})}_{\!\rm evts}}
{\mean{\frac{\displaystyle\partial g^{0,\theta_2}_0}{\displaystyle\partial z}
({\rm i}r_0^{\theta_2})
}_{\!\!\rm evts}}.
\end{equation}
This yields in principle an estimate of $V_1$ for each value of $\theta_1$ and
$\theta_2$. However,
the expansion of
$\smean{g_\epsilon^{\theta_1,\theta_2}({\rm i}r_0^{\theta_2})}_{\rm evts}$
in powers of $\epsilon$ generally yields a non-vanishing term
of order $\epsilon$ due to statistical fluctuations.
Multiplying the previous equation by
$\cos(2(\theta_1-\theta_2))$, and averaging over 5 or more equally
spaced values of $\theta_1$, one eliminates this term.
Averaging over $\theta_2$, one obtains our final estimate,
Eq.~(\ref{intv1}).
We can now turn to differential directed flow.
For this purpose, let us study how the zero of
$g_\epsilon^{\theta_1,\theta_2}(z)$ moves when the weights
$w_1$ of all protons are shifted by some small quantity $\delta w_1$:
\begin{equation}
z_{\delta w}-z_0=-\frac{
\mean{\frac{\displaystyle \partial g_\epsilon^{\theta_1,\theta_2}(z_0)}
{\displaystyle\partial w_1(\psi)}}_{\!\psi}}
{\mean{\frac{\displaystyle\partial g^{0,\theta_2}_0}
{\displaystyle\partial z}(z_0)
}_{\!\!\rm evts}}\delta w_1.
\end{equation}
The value of $z_{\delta w}$ is directly related to collective effects.
Following the same procedure as for integrated directed flow,
we estimate the value of the r.h.s. when flow is present.
To evaluate the numerator, we first compute the
average for a fixed $\Phi_R$. Using Eq.~(\ref{partialw}),
assuming that the proton and the other particles are uncorrelated
for fixed $\Phi_R$ (i.e., that the correlation between the proton
and the other particles is only due to flow), and neglecting the
contribution of the proton to the generating function, we obtain
\begin{eqnarray}
\mean{\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}(z_0)}
{\partial w_1(\psi)}|\Phi_R}
&\simeq &\epsilon z_0\mean{\cos(\psi-\theta_1)|\Phi_R}
\mean{g_\epsilon^{\theta_1,\theta_2}(z_0)|\Phi_R}\cr
&=&\epsilon z_0 v'_1\cos(\Phi_R-\theta_1)
\exp\left[z_0\epsilon V_1\cos(\Phi_R-\theta_1)+
z_0 V_2\cos(2(\Phi_R-\theta_2))\right],
\end{eqnarray}
where we have used Eq.~(\ref{gphiR}), the definition of
$v'_1=\mean{\cos(\psi-\Phi_R)}$, and assumed symmetry with
respect to the reaction plane, $\mean{\sin(\psi-\Phi_R)}=0$.
We then expand the exponential to order $\epsilon$ and
integrate over $\Phi_R$:
\begin{equation}
\label{numdiff}
\mean{\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}(z_0)}
{\partial w_1(\psi)}}
=\frac{\epsilon^2}{2}z_0^2\, v'_1V_1\,
I_1(V_2 z_0)\,\cos(2(\theta_1-\theta_2)).
\end{equation}
Using Eq.~(\ref{denomflow}), and replacing $z_0$ with
${\rm i}r_0^{\theta_2}$, we obtain
\begin{equation}
v'_1 V_1\, {\rm sgn}(V_2)\cos(2(\theta_1-\theta_2))=-
\frac{2 j_{01}}{\epsilon^2}\left(r_0^{\theta_2}\right)^{\!-3}
\frac{
\mean{\frac{\displaystyle \partial g_\epsilon^{\theta_1,\theta_2}
({\rm i}r_0^{\theta_2})}
{\displaystyle\partial w_1(\psi)}}_{\!\psi}}
{\mean{\frac{\displaystyle\partial g^{0,\theta_2}_0}
{\displaystyle\partial z}({\rm i}r_0^{\theta_2})
}_{\!\!\rm evts}}.
\end{equation}
Multiplying both sides of this equation by $\cos(2(\theta_1-\theta_2))$
and averaging over $\theta_1$, one eventually obtains Eq.~(\ref{diffv1}).
For the sake of consistency, one must recover the integrated
flow $V_1$ by integrating the differential flow $v'_1$ over phase
space.
Our estimates (\ref{intv1}) and (\ref{diffv1}) satisfy this
sum rule, provided $\epsilon$ is small enough. In order to
prove this, we weight
Eq.~(\ref{diffv1}) with $w_1(\psi)$ and integrate over
phase space. The following quantity appears
\begin{equation}
\sum_{j=1}^M w_1(j)\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})}
{\partial w_1(j)}=
\epsilon
\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})}
{\partial\epsilon}.
\end{equation}
Next, we use the fact that
$g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})$ scales like
$\epsilon^2$ for small $\epsilon$, hence
\begin{equation}
\epsilon
\frac{\partial g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2})}
{\partial\epsilon}=
2g^{\theta_1,\theta_2}_{\epsilon}({\rm i}r_0^{\theta_2}).
\end{equation}
This completes the proof.
\section{Errors}
\label{s:errors}
The main motivation for analyzing collective flow with
multiparticle correlations is that one thereby reduces
spurious, ``nonflow'' effects. The magnitude of errors due
to residual nonflow effects is estimated in Sec.~\ref{s:systerrors}.
The price to pay for this greater reliability is
an increase in statistical errors, which are evaluated
in Sec.~\ref{s:staterrors}.
Finally, azimuthal asymmetries in the detector acceptance
are a potential source of bias in the analysis. They are
carefully studied in Sec.~\ref{s:acceptance}.
\subsection{Systematic errors from nonflow effects}
\label{s:systerrors}
There are several tricks to check experimentally whether or
not the analysis is biased by nonflow effects.
A first one is to perform the analysis twice, using different weights
in Eq.~(\ref{defgenfunc}),
e.g., one analysis with $w_2=p_t$ and another with $w_2=1$,
or redoing the analysis with zero weights for the
particles in one hemisphere.
The final differential flow results should be independent of
the weights, while no such independence is expected for nonflow
effects. However,
using non-optimal weights increases statistical errors,
sometimes by large amounts, so that this is doable in practice
only if statistical errors are not a limitation.
A second trick is to check that the final results look
``reasonable:'' this means in particular that they must be consistent
with the known symmetries of the system, i.e., $v_1(-y)=-v_1(y)$,
$v_2(-y)=v_2(y)$ for a symmetric collision.
Nonflow effects can destroy these symmetries: for instance, effects of
total momentum conservation may contaminate the measurement of $v_1$,
yielding an estimate that does not vanish near
midrapidity~\cite{Borghini:2002mv}.
This may however occur only when the weights in Eq.~(\ref{defgenfunc})
are not symmetric themselves (i.e., $w_1(-y)\not=-w_1(y)$,
$w_2(-y)\not= w_2(y)$). With symmetric weights, the fact that
$v_1$ vanishes at midrapidity is not an indication that it is not
biased by nonflow effects.
In this paper, we are interested in directed flow at ultrarelativistic
energies, which is most often at the border of observability, so
that it is not always possible to use the above tricks.
It is therefore important to derive estimates of the magnitude of nonflow
effects using purely theoretical arguments. There is of course
no way to derive quantitative estimates, due to the variety of the
physical effects involved~\cite{Borghini:2000cm}.
In most cases, one can at best rely on orders of magnitude
and simple scaling rules~\cite{Borghini:2000sa}, which
are derived below.
More quantitative statements can be made concerning momentum
conservation, a well-known bias in analyses of directed
flow~\cite{Borghini:2002mv,Danielewicz:1988in,Ogilvie:1989}.
With standard methods of flow
analysis~\cite{Danielewicz:hn,Poskanzer:1998yz} and also,
to a lesser degree, with cumulants~\cite{Borghini:2000sa},
the analysis may yield a non-zero value of $v_1$ and $v_2$ even
when they are in fact zero, i.e., even if only ``nonflow'' effects
are present~\cite{Kovchegov:2002nf}.
This is not the case with Lee--Yang zeroes, where a non-vanishing
result beyond statistical errors can be considered a clear
signal of anisotropic flow.
However, the interference of collective flow and
nonflow effects may produce a small relative error,
which is the sum of two terms:
\begin{equation}
\label{systerror}
\frac{(\delta v_1)_{\rm nonflow}}{v_1}
={\cal O}\!\left(\frac{v_2}{M v_1^2}\right) +
{\cal O}\!\left(\frac{1}{M v_2}\right).
\end{equation}
Let us explain how such terms arise.
Our determination of directed flow with the
present method involves three-particle averages
such as $\mean{\cos(\phi_1+\phi_2-2\phi_3)}$
(see Sec.~\ref{s:IID}).\footnote{One can check that the systematic
errors due to higher-order correlations are at most of the same order
as those due to three-particle correlations.}
The contribution of flow to this average is $v_1^2 v_2$.
Let us now evaluate the contribution of nonflow effects.
As a simple model, assume that the $M$ particles
detected in an event are emitted in $M/2$ collinear pairs
(``jet-like correlations'').
There is a probability $1/(M-1)\simeq 1/M$ that $\phi_2=\phi_1$,
in which case the three-particle average becomes
$\mean{\cos 2(\phi_1-\phi_3)}=v_2^2$. This gives the first
term in Eq.~(\ref{systerror}).
Similarly, there is a probability of order $1/M$ that
$\phi_1=\phi_3$, in which case the
three-particle average reads
$\mean{\cos (\phi_1-\phi_2)}=v_1^2$;
this yields the second term in Eq.~(\ref{systerror}).
Since the present method is expected to be useful essentially when
$v_1<v_2$, this term is subleading.
Finally, one can check explicitly that momentum conservation
does not contribute to the error terms in Eq.~(\ref{systerror}).
The order of magnitude of the systematic error, Eq.~(\ref{systerror}), is more
general than suggested by this toy model of nonflow correlations.
It can be proven more rigorously by pushing the expansion of the logarithm of
the generating function, Eq.~(\ref{gphiR}), to order $z^2$, and studying the
influence of the additional terms on the flow estimate, much in the same way
as in Ref.~\cite{Bhalerao:2003yq}, where the interference between flow and
nonflow effects was also considered (see the discussion following Eq.~(48) in
the reference).
Of course, Eq.~(\ref{systerror}) is but a scaling law, and each term involves
an unknown numerical coefficient.
One can try to estimate these coefficients experimentally by studying the
difference between flow estimates from various methods, using situations where
this difference is believed to be dominated by nonflow effects.
This strategy was applied by the STAR Collaboration~\cite{Adams:2003zg} in
their estimates of systematic errors on $v_4$ and $v_1$.
With the standard, event-plane method~\cite{Danielewicz:hn,Poskanzer:1998yz},
the error due to nonflow effects is much larger:
\begin{equation}
\label{syststandard}
\frac{(\delta v_1)_{\rm nonflow}}{v_1}
={\cal O}\!\left(\frac{1}{M v_1^2}\right).
\end{equation}
Comparing with Eq.~(\ref{systerror}), our method reduces
the nonflow error by a factor of $1/v_2$, i.e., typically
15 to 20 at RHIC.
The errors with the present method, Eq.~(\ref{systerror}),
also apply to the three-particle cumulant method, but
they were not mentioned in Ref.~\cite{Borghini:2002vp}.
The cumulant method has an additional error term,
coming from the ``pure nonflow'' contribution to the
three-particle average $\mean{\cos(\phi_1+\phi_2-2\phi_3)}$,
of order $1/M^2$.
This gives an additional error term in Eq.~(\ref{systerror})
\begin{equation}
\label{systerror3}
\frac{(\delta v_1)_{\rm nonflow}}{v_1}
={\cal O}\!\left(\frac{1}{M^2 v_2 v_1^2}\right).
\end{equation}
This contribution, which was the only error
term mentioned in Ref.~\cite{Borghini:2002vp},
disappears with Lee--Yang zeroes.
In practice at ultrarelativistic energies $Mv_2^2\gtrsim 1$, so that the
additional term is typically of the same order of magnitude as
the first term of Eq.~(\ref{systerror}). In addition,
momentum conservation produces an error of this order,
so that the Lee--Yang zeroes method does a better job
in eliminating effects of momentum conservation.
Finally, the three-particle method requires to estimate $v_2$
independently. If $v_2$ is estimated from two-particle
cumulants or by the event-plane method, the resulting
error on $v_1$ is
$(\delta v_1)_{\rm nonflow}/v_1={\cal O}(1/M v_2^2)$,
which may dominate over both terms in Eq.~(\ref{systerror}).
\subsection{Statistical errors}
\label{s:staterrors}
Statistical uncertainties on $V_1$ and $v'_1$ naturally involve
both directed and elliptic flows.
Both errors turn out to scale like the statistical error on
the differential elliptic flow $v'_2$ determined
from Lee--Yang zeroes~\cite{Bhalerao:2003yq}, which is natural since
we use elliptic flow as a reference.
The statistical error on elliptic flow involves the
associated {\it resolution parameter\/}
\begin{equation}
\label{chi2}
\chi_2\equiv\frac{V_2}{\sigma_2},
\end{equation}
where $\sigma_2$ is a measure of event-by-event fluctuations.
Provided nonflow effects are not too large, $\sigma_2$ can be obtained
experimentally by the following formula:
\begin{equation}
\label{sigma2}
\sigma_2^2\simeq \mean{\sum_{j=1}^M w_2(j)^2}_{\!\!\rm evts}.
\end{equation}
More accurate determinations are discussed in Ref.~\cite{Bhalerao:2003yq}.
Once $\chi_2$ has been determined, the statistical error on $v'_2$
is~\cite{Bhalerao:2003yq}:
\begin{equation}
\label{deltav2}
(\delta v'_2)^2=\frac{1}{4\Np J_1(j_{01})^2}
\mean{
\exp\left(\frac{j_{01}^2}{2\chi_2^2}\cos 2\theta_2\right)
J_0(2 j_{01}\sin\theta_2)
-\exp\left(-\frac{j_{01}^2}{2\chi_2^2}\cos 2\theta_2\right)
J_0(2 j_{01}\cos\theta_2)}_{\!\theta_2},
\end{equation}
where $\Np$ is the number of ``protons'' in the phase-space region
under study.
For large $\chi_2$, this yields
$\delta v'_2=1/\sqrt{2\Np}$, which is the expected statistical
error when the reaction plane is exactly known.
When $\chi_2$ becomes significantly smaller than unity, on the
other hand, the statistical error on $v'_2$ increases exponentially
as $\delta v'_2\propto \exp(2.9/\chi_2^2)$, and the method cannot
be applied~\cite{Bhalerao:2003yq}.
Let us now discuss statistical errors on the directed flow estimates.
These errors involve two parameters $\sigma_1$ and
$\chi_1$ defined as in Eqs.~(\ref{chi2}) and (\ref{sigma2}),
namely, $\sigma_1^2\simeq\smean{\sum_j w_j^2}_{\rm evts}$
and $\chi_1=V_1/\sigma_1$.
The latter is the resolution parameter associated with directed flow.%
\footnote{Please note that the resolution parameter $\chi$ as defined
in Ref.~\cite{Poskanzer:1998yz} is larger by a factor of $\sqrt{2}$}
In order to obtain the square of the statistical error on
$V_1^2{\rm sgn}(V_2)$, one then does the following substitution
in Eq.~(\ref{deltav2}):
\begin{equation}
\frac{1}{\Np}\to \frac{2\,\sigma_1^4(1+\chi_1^2)}{N_{\rm evts}}.
\end{equation}
For large $\chi_1$ and $\chi_2$, this yields
$\delta V_1=\sigma_1/\sqrt{2 N_{\rm evts}}$, which is the
expected value of the statistical error if the reaction plane
is exactly known.
Finally, the statistical error on the differential directed flow
$v'_1$ is simply related to the error on $v'_2$:
\begin{equation}
\label{statervp1}
(\delta v'_1)^2=\frac{1+\chi_1^2}{\chi_1^2}(\delta v'_2)^2.
\end{equation}
One notes that the statistical error on directed flow obtained
with this method is always larger than the statistical error
on elliptic flow in absolute value.
When $\chi_2$ is large enough, $\delta v'_1$ is the same as
with the event-plane analysis.
When using 3-particle cumulants~\cite{Borghini:2002vp,Adams:2003zg},
the statistical error is given by a similar formula, where
$\delta v'_2$ is the error on elliptic flow obtained with
the event-plane method (or by 2-particle correlations).
Going from the 3-particle method to the present method,
the increase in statistical errors will be the same as
when going from the standard method to Lee--Yang zeroes in
the analysis of elliptic flow.
For a semi-central Au-Au collision at RHIC analyzed with the
STAR detector, the resolution parameter $\chi_2$ is at least 1,
which means a factor of at most 2 increase in statistical
errors~\cite{Bhalerao:2003yq}.
This increase is compensated by the smaller error from nonflow effects.
\subsection{Azimuthal asymmetries in the detector acceptance}
\label{s:acceptance}
A nice feature of all flow analyses based on cumulants or Lee--Yang
zeroes (which amount to cumulants of many-particle
correlations~\cite{Bhalerao:2003yq}) is that they automatically
eliminate most effects of azimuthal asymmetries
in the detector acceptance:
more precisely, such asymmetries cannot produce a signal
by themselves, even if the detector has a very partial
azimuthal coverage. This is not the case with the event-plane
analysis, where several flattening procedures are
required~\cite{Poskanzer:1998yz}, and with two-particle
correlation methods~\cite{Wang:1991qh,PHENIX}, which use
mixed events to correct for acceptance effects.
Acceptance asymmetries generally have two effects with
cumulants or Lee--Yang zeroes~\cite{Borghini:2000sa,Bhalerao:2003yq}:
1) the flow given by
the analysis differs from the true flow by some factor,
which can be computed analytically once the acceptance
profile is known;
2) different harmonics interfere, so
that a measurement of $v_1$ is generally biased by $v_2$
and vice-versa. These interference terms can also be calculated
analytically.
Such interference terms between $v_1$ and $v_2$ are also
present here, and they turn out to be a rather serious problem.
Indeed, as an explicit calculation will show, acceptance
asymmetries produce an additional term proportional to
$v_2$ in the left-hand side (l.h.s.) of Eq.~(\ref{intv1}): even
if there is only elliptic flow in the system, the analysis
then yields a spurious directed flow $v_1\propto\sqrt{v_2}$.
Since we are typically interested in a situation where
$v_2$ is large and $v_1$ is small, such an interference
is most unwelcome, and it must be thoroughly studied.
In order to focus on the interference with $v_2$, we assume
that only elliptic flow is present in the system, $v_1=0$, and
we evaluate the spurious directed flow.
The following study is limited to the integrated flow $V_1$.
For a given reaction plane $\Phi_R$, the azimuthal distribution
of a particle of type $j$ is, before detection,
\begin{equation}
\frac{{\rm d}N}{{\rm d}\phi_j}\propto 1+2 v_2(j)\cos(2(\phi_j-\Phi_R)).
\end{equation}
With this distribution, and with a non-symmetric detector,
the first term in Eq.~(\ref{gphiR}) becomes
\begin{equation}
\mean{\sum_{j=1}^M w_{1}(j)\cos(\phi_j-\theta_1)|\Phi_R}=
{\rm Re}\left[ V_0^{\rm acc}{\rm e}^{-{\rm i}\theta_1}
+V_2^{\rm acc} {\rm e}^{{\rm i}(2\Phi_R-\theta_1)}
+V_2^{\prime\,\rm acc} {\rm e}^{{\rm i}(-2\Phi_R-\theta_1)}\right],
\end{equation}
where $V_0^{\rm acc}$, $V_2^{\rm acc}$ and $V_2^{\prime\,\rm acc}$
are complex coefficients defined by
\begin{eqnarray}
V_0^{\rm acc}&=&
\mean{\sum_{j=1}^M w_{1}(j)\,{\rm e}^{{\rm i}\phi_j}}\cr
V_2^{\rm acc}&=&
\mean{\sum_{j=1}^M w_{1}(j)\,v_2(j)\,{\rm e}^{-{\rm i}\phi_j}}\cr
V_2^{\prime\,\rm acc}&=&
\mean{\sum_{j=1}^M w_{1}(j)\,v_2(j)\,{\rm e}^{3{\rm i}\phi_j}},
\end{eqnarray}
with the averages taken over detected particles.
For a symmetric acceptance, the three coefficients vanish.
With these values, one can compute the generating function
$\mean{g_\epsilon^{\theta_1,\theta_2}(z)}$ to order $\epsilon^2$
following essentially the same steps as in Sec.~\ref{s:theory}.
Let us give the final result, skipping tedious intermediate
calculations: to leading order in acceptance asymmetries,
one must replace $(V_1)^2$ with ${\rm Re}\,[2 V_0^{\rm acc}V_2^{\rm acc}]$
in the l.h.s.\ of Eq.~(\ref{intv1}).
For moderate inhomogeneities in the detector acceptance, the interference
of elliptic flow on the estimate of directed flow remains under control.
Thus, assuming an elliptic flow value $v_2=0.06$, variations of 20\% in
the detector efficiency will result in a spurious $v_1$ of at most 0.01.
Using appropriate weights (e.g., weighting with the inverse of the detector
efficiency profile) would significantly decrease this spurious directed
flow value.
However, we fear that working with a detector with incomplete azimuthal
coverage might prove prohibitive.
Finally, note that the three-particle method is affected by
acceptance asymmetries in essentially the same way, but
the corresponding interference term was unfortunately omitted
in Ref.~\cite{Borghini:2002vp}.
\section{Conclusions and perspectives}
In the foregoing, we have introduced a new method to analyze directed flow
$v_1$ in cases where it is too small to be reliably obtained with standard
two-particle methods without contamination from nonflow effects, and also too
small to be determined by more recent methods as cumulants or Lee--Yang
zeroes, which would both give large statistical uncertainties.
Moreover, the method relies on the implicit assumption that elliptic flow is
reasonably large, as when using three-particle
cumulants~\cite{Borghini:2002vp}.
With respect to the latter, the new method is easier to implement: instead of
interpolating successive derivatives of a function, one need just determine
the position $r_0$ of the first minimum of a function, and then compute the
values at $r_0$ of the three quantities that appear in the numerators and
denominator of Eqs.~(\ref{intv1}) and (\ref{diffv1}).
In addition, the method is conceptually more elegant, since it relies on the
deep relation there exists between the behaviour of the zeroes of generating
functions and the presence of collective effects in the systems these
functions describe~\cite{Bhalerao:2003yq}.
We have also carefully estimated both systematic errors due to nonflow effects
and statistical uncertainties arising from finite available statistics.
We may use these estimates to discuss the applicability of the method to
present and future heavy-ion experiments.
At RHIC, the three-particle cumulant method was already successfully
applied~\cite{Adams:2003zg}.
In mid-central collisions, statistical uncertainties should be at most a
factor 2 larger with the new method, but with the high statistics of Run 4,
this should not be a problem.
On the other hand, systematic errors will be reduced:
the error term (\ref{systerror3}), which arises in particular
from momentum conservation, disappears with the new method.
The method could be useful for new measurements of directed flow at
the CERN SPS, although this would require detectors with a larger coverage
than what single experiments had in the past runs.
Finally, one can expect that the present method will allow measurements of
directed flow at LHC, if any, using muons from decay pions seen in the
ALICE spectrometer at forward rapidities ($2.5<\eta<4$) or hits in the
CMS very forward hadronic calorimeter ($3<\eta<5$).
Since $v_2$ at LHC is expected to be at least as large as at RHIC,
and the multiplicity will be higher, elliptic flow will be analyzed
with an excellent resolution. Then, the statistical errors
with our new method will be barely larger than with the standard
event-plane method: statistics will not be a limitation down
to values of $v_1$ of a fraction of a percent.
Systematic errors from nonflow effects may be a more severe problem,
although our method minimizes their magnitude.
Comparing Eq.~(\ref{statervp1}), with $\chi_1\sim v_1\sqrt{M}$,
and the first term of Eq.~(\ref{systerror}), one sees that as
$v_1$ decreases, systematic errors tend to become larger than
statistical errors. One may reasonably hope, however, that the
huge particle multiplicity $M$ expected at LHC will compensate
for the smaller value of $v_1$, and that $v_1$ will eventually
be observed.
%As hinted above, the balance between both allows us to conclude that the new
%method is the most reliable one to analyze $v_1$ at RHIC energies and above.
\begin{thebibliography}{99}
\bibitem{Danielewicz:hn}
P.~Danielewicz and G.~Odyniec,
Phys.\ Lett.\ B {\bf 157}, 146 (1985).
%%CITATION = PHLTA,B157,146;%%
\bibitem{Poskanzer:1998yz}
A.~M.~Poskanzer and S.~A.~Voloshin,
Phys.\ Rev.\ C {\bf 58}, 1671 (1998)
[nucl-ex/9805001].
%%CITATION = NUCL-EX 9805001;%%
\bibitem{Borghini:2000cm}
N.~Borghini, P.~M.~Dinh and J.-Y.~Ollitrault,
Phys.\ Rev.\ C {\bf 62}, 034902 (2000)
[nucl-th/0004026].
%%CITATION = NUCL-TH 0004026;%%
\bibitem{Ollitrault:dy}
J.-Y.~Ollitrault,
Nucl.\ Phys.\ A {\bf 590}, 561c (1995).
%%CITATION = NUPHA,A590,561C;%%
\bibitem{Borghini:2002mv}
N.~Borghini, P.~M.~Dinh, J.-Y.~Ollitrault, A.~M.~Poskanzer and S.~A.~Voloshin,
Phys.\ Rev.\ C {\bf 66}, 014901 (2002)
[nucl-th/0202013].
%%CITATION = NUCL-TH 0202013;%%
\bibitem{Borghini:2002vp}
N.~Borghini, P.~M.~Dinh and J.-Y.~Ollitrault,
Phys.\ Rev.\ C {\bf 66}, 014905 (2002)
[nucl-th/0204017].
%%CITATION = NUCL-TH 0204017;%%
\bibitem{Borghini:2002hm}
N.~Borghini, P.~M.~Dinh and J.-Y.~Ollitrault,
Nucl.\ Phys.\ A {\bf 715}, 629c (2003)
[nucl-th/0208014].
%%CITATION = NUCL-TH 0208014;%%
\bibitem{Alt:2003ab}
C.~Alt {\it et al.} [NA49 Collaboration],
Phys.\ Rev.\ C {\bf 68}, 034903 (2003)
[nucl-ex/0303001].
%%CITATION = NUCL-EX 0303001;%%
\bibitem{Adams:2003zg}
J.~Adams {\it et al.} [STAR Collaboration],
Phys.\ Rev.\ Lett.\ {\bf 92}, 062301 (2004)
[nucl-ex/0310029].
%%CITATION = NUCL-EX 0310029;%%
\bibitem{Snellings:1999bt}
R.~J.~M.~Snellings, H.~Sorge, S.~A.~Voloshin, F.~Q.~Wang and N.~Xu,
Phys.\ Rev.\ Lett.\ {\bf 84}, 2803 (2000)
[nucl-ex/9908001];
%%CITATION = NUCL-EX 9908001;%%
L.~P.~Csernai and D.~R{\"o}hrich,
Phys.\ Lett.\ B {\bf 458}, 454 (1999)
[nucl-th/9908034];
%%CITATION = NUCL-TH 9908034;%%
M.~Bleicher and H.~St{\"o}cker,
Phys.\ Lett.\ B {\bf 526}, 309 (2002)
[hep-ph/0006147].
%%CITATION = HEP-PH 0006147;%%
\bibitem{Ollitrault:bk}
J.-Y.~Ollitrault,
Phys.\ Rev.\ D {\bf 46}, 229 (1992).
%%CITATION = PHRVA,D46,229;%%
\bibitem{Borghini:2000sa}
N.~Borghini, P.~M.~Dinh and J.-Y.~Ollitrault,
Phys.\ Rev.\ C {\bf 63}, 054906 (2001)
[nucl-th/0007063];
%%CITATION = NUCL-TH 0007063;%%
ibid.\ {\bf 64}, 054901 (2001)
[nucl-th/0105040].
%%CITATION = NUCL-TH 0105040;%%
\bibitem{Bhalerao:2003yq}
R.~S.~Bhalerao, N.~Borghini and J.-Y.~Ollitrault,
Phys.\ Lett.\ B {\bf 580}, 157 (2004)
[nucl-th/0307018];
%%CITATION = NUCL-TH 0307018;%%
Nucl.\ Phys.\ A {\bf 727}, 373 (2003)
[nucl-th/0310016].
%%CITATION = NUCL-TH 0310016;%%
\bibitem{Borghini:2004ke}
N.~Borghini, R.~S.~Bhalerao and J.-Y.~Ollitrault,
nucl-th/0402053.
%%CITATION = NUCL-TH 0402053;%%
\bibitem{Danielewicz:1995}
P. Danielewicz, Phys. Rev. C {\bf 51}, 716 (1995).
%%CITATION = PHRVA,C51,716;%%
\bibitem{Yang:be}
C.~N.~Yang and T.~D.~Lee,
Phys.\ Rev.\ {\bf 87}, 404 (1952).
%%CITATION = PHRVA,87,404;%%
\bibitem{Danielewicz:1988in}
P.~Danielewicz {\it et al.},
Phys.\ Rev.\ C {\bf 38}, 120 (1988).
%%CITATION = PHRVA,C38,120;%%
\bibitem{Ogilvie:1989}
C.~A.~Ogilvie {\it et al.},
Phys.\ Rev.\ C {\bf 40}, 2592 (1989).
%%CITATION = PHRVA,C40,2592;%%
\bibitem{Kovchegov:2002nf}
Y.~V.~Kovchegov and K.~L.~Tuchin,
Nucl.\ Phys.\ A {\bf 708}, 413 (2002)
[hep-ph/0203213];
%%CITATION = HEP-PH 0203213;%%
ibid.\ {\bf 717}, 249 (2003)
[nucl-th/0207037].
%%CITATION = NUCL-TH 0207037;%%
\bibitem{Wang:1991qh}
S.~Wang {\it et al.},
Phys. Rev. C {\bf 44}, 1091 (1991).
%%CITATION = PHRVA,C44,1091;%%
\bibitem{PHENIX}
K.~Adcox {\it et al.} [PHENIX Collaboration],
Phys. Rev. Lett. {\bf 89}, 212301 (2002) [nucl-ex/0204005];
%%CITATION = NUCL-EX 0204005;%%
N.~N.~Ajitanand [PHENIX Collaboration],
Nucl. Phys. A {\bf 715}, 765c (2003) [nucl-ex/0210007].
%%CITATION = NUCL-EX 0210007;%%
\end{thebibliography}
\end{document}