1: %%%%%%%%%%%% tex/notes/TSALLIS/ZAJ/zaj.tex %%%%%%%%%%%%%
2: %
3: % Langevin with additive + multiplicative noise
4: % Stationary solution: POWER LAW for high p!
5: % A. Jakovac 14.05.2004. (langevin.tex notes, derivation)
6: % T.S.Biro 15.05.2004. (PRL article first version)
7: % T.S.Biro 18.05.2004. (Corrections, Tsallis, D+Cp2 diff. coeff.)
8: % T.S.Biro 19.05.2004. (Graphs, gauge theory case, refs)
9: %
10: %
11: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
12: %\documentclass[10pt]{article}
13: %\documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb,floatfix]{revtex4}
14: \documentclass[prl,twocolumn]{revtex4}
15: %\documentclass[prl,preprint]{revtex4}
16:
17: %%%%%%%%%%%%%%%%%%%%% PACKAGES %%%%%%%%%%%%%
18:
19: \usepackage{dcolumn}% Align table columns on decimal point
20: \usepackage{bm}% bold math
21:
22: \usepackage{graphics}
23: \usepackage{graphicx}
24: \usepackage[dvips,usenames]{color}
25:
26: %%%%%%%%%%%%%%%%%%%%%%%% MACROS %%%%%%%%%%%%%%%
27:
28: %\newcommand{\vs}{\vspace{3mm}}
29: \newcommand{\vs}{\quad}
30: \newcommand{\tc}[2]{\textcolor{#1}{#2}}
31: \newcommand{\cb}[2]{\colorbox{#1}{#2}}
32: \newcommand{\fb}[3]{\fcolorbox{#1}{#2}{#3}}
33:
34: \newcommand{\be}{\begin{equation}}
35: \newcommand{\ee}{\end{equation}}
36: \newcommand{\ba}{\begin{eqnarray}}
37: \newcommand{\ea}{\end{eqnarray}}
38: \newcommand{\NL}{\nonumber \\}
39:
40: \setlength{\arraycolsep}{.13889em}
41:
42: \newcommand{\eqref}[1]{(\ref{#1})}
43: \renewcommand{\d}{\partial}
44: \newcommand{\nn}{\nonumber\\}
45: \newcommand{\ph}{\varphi}
46: \newcommand{\exv}[1]{\left\langle{#1}\right\rangle}
47: \newcommand{\exvs}[1]{\langle{#1}\rangle}
48: \newcommand{\ep}{\varepsilon}
49: \newcommand{\GeV}{\textrm{~GeV~}}
50: \newcommand{\Tr}{\mathop{\textrm{Tr}}}
51: \renewcommand{\Im}{\,\textrm{Im}\,}
52: \renewcommand{\Re}{\,\textrm{Re}\,}
53: \newcommand{\pint}[2]{{\int\!\frac{d^{#1}#2}{(2\pi)^{#1}}\,}}
54: \newcommand{\MSbar}{{\ensuremath{\overline{\mathrm{MS}}}}}
55:
56: \renewcommand{\c}[1]{{\cal{#1}}}
57:
58:
59: \parindent0mm
60: \fboxsep3mm
61: \fboxrule1mm
62:
63: \begin{document}
64:
65: %%%%%%%%%%%%%%%%%%%%%%%%% TITLE PAGE %%%%%%%%%%%%%%%%%%
66:
67:
68: \title{Power-law tails from multiplicative noise}
69: %\title{Power-law tails from a modified Brownian motion}
70: %\title{Power-law tails and noisy equations}
71: %\title{Power-law tails are omnipresent}
72:
73: \author{Tam\'as S. Bir\'o}
74: \affiliation{MTA KFKI Research Institute for
75: Particle and Nuclear Physics, H-1525 Budapest Pf. 49, Hungary}
76:
77: \author{Antal Jakov\'ac}
78: \affiliation{Research Group for Theoretical Condensed Matter of HAS and
79: TU Budapest, H-1521 Budapest, Hungary}
80:
81: %\date{ \today}
82:
83:
84: \begin{abstract}
85: \vs
86: {\bf Abstract}
87: We show that the well-known linear Langevin equation, modeling the Brownian
88: motion and leading to a Gaussian stationary distribution of the corresponding
89: Fokker-Planck equation, is changed by the smallest multiplicative noise.
90: This leads to a power-law tail of the distribution for large enough
91: momenta. At finite ratio of the correlation strength for the
92: multiplicative and additive noise the stationary energy distribution
93: becomes exactly the Tsallis distribution.
94: \end{abstract}
95:
96: \pacs{}
97:
98: \maketitle
99:
100: %%%%%%%%%%%%%%%%%%%%%%%%% MAIN TEXT BODY %%%%%%%%%%%%%%%%%%%%%%%
101:
102: %\section{Introduction}
103:
104: %\vs
105: Power-law tails are present in numerous distributions studied in physics
106: or elsewhere when dealing with complex systems\cite{POWER-TAILS}.
107: They are of generic
108: interest, regarded as to signal long range order, non-vanishing
109: correlations or scale invariance in complex systems with strong
110: dynamics, which's details are mostly unknown.%\cite{LONG-RANGE}.
111: These are often contrasted
112: to the traditional statistical system, showing the Gibbs distribution
113: ($\exp(-E/T)$) in energy, which is Gaussian in the momenta of
114: free, massive particles ($\exp(-p^2/2mT)$). The latter is considered
115: as the generic case for thermal equilibrium of non-correlated or short-range
116: correlated systems. This concept has been carried far beyond of its
117: original field describing monoatomic ideal gas (Maxwell-Boltzmann statistics),
118: by applying the Gibbs distribution in thermal equilibrium
119: to areas such as particle physics and field theory. It serves as starting
120: point of high-temperature field theory calculations both with analytical
121: and numerical methods. Lattice gauge theory is based on the formal
122: similarity between Euclidian path integrals and the canonical partition
123: sum.
124:
125: %\vs
126: A very simple and elegant, microdynamical explanation for the Maxwell-Boltzmann
127: statistics is offered by the Langevin equation, describing a free
128: particle moving under the influence of a deterministic damping force
129: and a stochastic drive\cite{LANGEVIN}.
130: The latter accelerates the particle in a short
131: time, changing its momenta randomly and uncorrelated. The stationary solution of
132: this stochastic equation follows the Gaussian statistics, compatible
133: to Gibbs' principle. It seems that in many statistical considerations of
134: complex physical models from that on it is tacitly assumed that the
135: presence of this additive noise is a dominant effect: the equilibrium
136: distribution follows Gibbs' formula (the Gaussian distribution in momentum
137: for a free, massive particle). Since the harmonic oscillator is just the
138: extension of this free motion Langevin equation into the phase space,
139: also for free quantum systems the above picture is generally accepted.
140:
141: %\vs
142: We shared this expectations, thinking that any non-Gaussian (or in the
143: energy non-exponential) distribution, especially
144: the power-law tail observed in many phenomena including particle
145: spectra in high energy physics, may only come from non-thermalized or
146: in an other way non-equilibrium situation. In this note we would like
147: to share our deep astonishment about that this is not so: we found out
148: that treating the damping constant in the Langevin equation also
149: stochastically (considering this way both multiplicative and additive noise)
150: the stationary distribution is in general non-Gaussian.
151: Moreover, above a certain momentum, depending on the strength
152: (i.e. self-correlation width) of the multiplicative noise, the
153: stationary solution goes over into a power-law.
154:
155: %\vs
156: This finding seems to have manifold consequences. Even in equilibrium,
157: even averaging over a huge number of elementary events, measurements
158: on quantum systems, such as particles or fields, are bound to find
159: a power-law tail irrespective if the underlying dynamical system
160: had equilibrated itself long enough in units of a characteristic time.
161: Power-law tails in pion spectra, found experimentally in high energy $e^+e^-$,
162: $p\overline{p}$ or heavy-ion collisions\cite{POW-TAIL-SPECTRA},
163: in the view of this simple
164: mathematical result may reflect an already stationary distribution.
165: Power-law tails found in other areas, in particular as properties of
166: average correlations (cf. time series in stock market), may also be
167: a non-temporary, long term effect.
168:
169: %\vs
170: In this paper we derive a generic stationary distribution for the
171: Langevin-type equation with both additive and multiplicative noise.
172: Conform to the original assumptions both noise terms are white
173: (Dirac delta correlated in time), but they may show cross-correlations.
174: This situation may stem naturally from the widespread treatment
175: of field theoretical operator equations, where the fields are split to
176: a large, ''classical'' expectation value and to a noisy quantum or
177: thermal fluctuation part: the different noise terms do have a common
178: origin, so it is natural to consider cross-correlations among them
179: \cite{NOISE-FIELD}.
180: Often the overdamped approximation is applied for studies of plasmas,
181: solving then effectively a differential equation first order in time,
182: instead of second order\cite{PLASMA}. We think therefore that the analytically
183: solvable case, we present in this paper, may carry a quite general lesson.
184:
185:
186:
187:
188:
189: %\begin{document}
190: %\newpage
191: %\section{Langevin equation with two noise terms}
192: %\label{sec:LangFock}
193:
194: We present the solution of the Langevin equation
195: with both additive and multiplicative noise terms applying the
196: classical method of Wang and Uhlenbeck\cite{UHLENBECK}. Our starting point is
197: the linear equation:
198: \begin{equation}
199: \label{eq:lan}
200: \dot p + \gamma p = \xi,
201: \end{equation}
202: where now both $\xi$ and $\gamma$ are stochastic variables.
203: They both may have a non-zero mean value (motivated by possible field theory
204: applications),
205: \be
206: \exv{\xi(t)}=F, \qquad \exv{\gamma(t)}=G,
207: \label{MEAN}
208: \ee
209: and show white-noise (i.e. extremely short term) correlations:
210: \ba
211: \exv{\xi(t)\xi(t')} -\exv{\xi(t)}\exv{\xi(t')}\: &=& \: 2D \: \delta(t-t'), \NL
212: \exv{\gamma(t)\gamma(t')} -\exv{\gamma(t)}\exv{\gamma(t')}\: &=&
213: \: 2C \: \delta(t-t'), \NL
214: \exv{\gamma(t)\xi(t')} -\exv{\gamma(t)}\exv{\xi(t')} \: &=& \: 2B \: \delta(t-t'), \NL
215: \exv{\xi(t)\gamma(t')} -\exv{\xi(t)}\exv{\gamma(t')} \: &=& \: 2B \: \delta(t-t'),
216: \label{CORR}
217: \ea
218: This problem can be solved analytically.
219: We will determine the time dependence of the distribution of $p$
220: values, denoted by $f(p,t)$. In this notation $f(p_0,t)dp$ is the
221: probability that after time $t$ the variable $p$ has the value in the
222: range $[p_0,p_0+dp]$. We rewrite \eqref{eq:lan} as a difference
223: equation
224: \begin{equation}
225: \label{eq:lan1_diff}
226: p(t+dt) = p(t) \: + \:
227: \int\limits_t^{t+dt}\!dt'\,\left(\xi(t') - \gamma(t')\: p(t')\right),
228: \end{equation}
229: If $p(t)$ is a smooth function of time we can
230: replace $p(t')$ either by $p(t)$ or $p(t+dt)$, or any value in between.
231: We choose here the Ito prescription\cite{ITO},
232: which uses $p(t)$ under the integral in the $dt \rightarrow 0$ limit.
233: In order to simplify notation we
234: write the integral term as $dt\exv{x}$, with $x$ denoting the general
235: integrand.
236: Now we can write down a Fokker-Planck equation for the distribution:
237: the probability to have the value \hbox{$p(t+dt)$} at $t+dt$ is the probability
238: that we have the value $p(t)$ at time $t$ and noise values $\xi$
239: and $\gamma$ that just satisfy \eqref{eq:lan1_diff}:
240: \begin{equation}
241: \label{eq:FP_pro}
242: f(p,t+dt) = \int\!d\xi d\gamma\, P(\xi,\gamma)\,
243: f(p - dt \exv{\xi}+p dt \exv{\gamma}).
244: \end{equation}
245: Unfortunately this form is not appropriate to directly create
246: differential equation as $dt\to0$. Instead we follow the method of
247: Wang and Uhlenbeck: we have a trial function $R(p)$ that is smooth
248: enough and we compute the expectation value of $R$ (averaged over the
249: noise) as a function of time
250: \begin{equation}
251: \exv{R(t)} = \int\!dp\, R(p)\, f(p,t).
252: \end{equation}
253: Applying this form to \eqref{eq:FP_pro} we have
254: \ba
255: \int\!dp\, R(p)\, f(p,t+dt) = \qquad \qquad \qquad & &\NL
256: \int\!d\xi d\gamma\,P(\xi,\gamma)
257: \int\!dp\, R(p + dt\: \exv{\xi} - p \:dt\: \exv{\gamma} )\, f(p,t). & &
258: \ea
259: %We now assume that the two variables are independent, so
260: %$P(\xi,\gamma)$ factorizes.
261: By Taylor expanding $R(p)$ and integrating over the noise distribution we get
262: \ba
263: \exv{ R(p + dt\: \exv{\xi} - p \:dt\: \exv{\gamma} ) } = \qquad \qquad \qquad & & \NL \NL
264: \exv{R(p) + dt \,
265: R'(p) \: K_1(p) + dt \, R''(p) \: K_2(p) + \c O(dt^2)} & &
266: \ea
267: with
268: \be
269: K_1 = F - Gp, \qquad K_2 = D - 2B p + Cp^2
270: \label{K1K2DEF}
271: \ee
272: in the present case.
273: This leads to the following general Fokker-Planck equation
274: \begin{equation}
275: \label{eq:FP_gen}
276: \frac{\d f}{\d t} = -\frac{\d (K_1 f)}{\d p} +
277: \frac{\d^2\!(K_2 f)}{\d p^2}.
278: \end{equation}
279: The stationary solution satisfies
280: \begin{equation}
281: \frac{d}{dp} \: (K_2f) \, = \, K_1 \: f,
282: \end{equation}
283: which is analytically solvable. It leads to
284: \be
285: f(p) \, = \, f(0)\: \frac{K_2(0)}{K_2(p)}\: \exp(\:L(p)\:)
286: \label{FP-stac}
287: \ee
288: with
289: \be
290: L(p) \: = \: \int_0^p\!\!dq \, \frac{K_1(q)}{K_2(q)}.
291: \ee
292: In the case of two noises correlated the way given in eq.\eqref{CORR}
293: we arrive at the following logarithm of the stationary
294: distribution:
295: \be
296: \ln \frac{f(p)}{f(0)} = -\left(1+\frac{G}{2C}\right)\ln \frac{K_2(p)}{D}
297: \: - \: \frac{\alpha}{\vartheta} \: {\rm atn} \left( \frac{\vartheta \: p}{D-Bp} \right),
298: \label{SOLUTION}
299: \ee
300: with
301: \be
302: \vartheta = \sqrt{CD-B^2} \qquad {\rm and} \qquad \alpha = G \: \frac{B}{C}-F.
303: \ee
304: Here 'atn' denotes the inverse tangent function, not always taking the
305: first principle value, but rather continuing at $p>D/B$ smoothly.
306: %(This can be achieved in programming by using the $atan2(x,y)$ function.)
307: Considering physical applications the noise correlation values,
308: $D$, $C$ and $B$ build a positive semidefinite matrix. This ensures
309: that $C$ and $D$ are non-negative values, and the determinant
310: $\vartheta$ is real and also non-negative. The same applies for
311: the function $K_2(p)$ occurring under the logarithm.
312: In this context zero
313: values are limiting cases and the stability of the stationary
314: solution \eqref{SOLUTION} against choosing a small
315: positive value has to be investigated.
316:
317: %\vs
318: The lesson for physical applications lies in the analysis of different
319: limiting cases. First we consider the traditional case: $C=B=0$,
320: not allowing for any noise (fluctuation) in the multiplicative factor
321: $\gamma$ (damping constant). This leads back to the familiar
322: Gauss distribution:
323: \be
324: f(p) = f(0) e^{-\frac{G}{2D} p^2} e^{\frac{F}{D}p}
325: \label{GAUSS}
326: \ee
327: with an eventual shift in the mean momentum for a non-zero mean
328: driving force $\exv{\xi}=F$.
329: Another limiting case is that of the purely multiplicative noise
330: with $D=0, B=0$
331: (G.Wilk\cite{WILK} has applied it for the heat conduction equation and obtained
332: a Gamma distribution for the inverse temperature $1/T$). Now the
333: stationary solution becomes
334: \be
335: f(p) = f(0) \: p^{-\: 2\: -\: G/C} \: e^{-\frac{F}{C \: |p|}}
336: \label{GAMMA}
337: \ee
338: a Gamma distribution in $1/p$. For large $p$ this approaches a
339: pure power-law.
340: It is particularly interesting to investigate the mathematically degenerate
341: case of $\vartheta=0$. Now $K_2(p)$ reaches zero at the critical
342: momentum, $p_c=\sqrt{D/C}$, leading to zero probability in the
343: stationary distribution $f(\sqrt{D/C})=0$. This occurs as a
344: ''limiting momentum'' in the physical distribution.
345:
346:
347: %\vs
348: The above limiting cases rely on an expansion of the
349: generic solution. They are valid only in a limited range
350: of momenta: the Gaussian solution for small,
351: the Gamma distribution for large argument of the inverse tangent
352: function. Correspondingly the widely beloved Gaussian
353: distribution can be a good approximation
354: to the stationary solution only for momenta $p \ll \sqrt{D/C}$.
355: (For $C=0$ strictly, of course this is 'the' solution for any finite momentum.)
356: Generally the small argument of the inverse tangent is fulfilled
357: for
358: %\be
359: $p \ll \frac{D}{\vartheta + B} $.
360: %\label{KNACK}
361: %\ee
362: This result also means that {\em for the smallest fluctuation
363: in the multiplicative factor the stationary Gauss distribution
364: develops a power-law tail}. The power in this tail, $p^{-2v}$, is given by
365: %\be
366: $v = 1+\frac{G}{2C}$.
367: %\label{POWER}
368: %\ee
369: In order to offer a visual insight into the nature of
370: the generic stationary solution we show stationary spectra
371: for different parameters (cf.Fig(\ref{LOG-f-CASES})).
372:
373: %%%%%%%%%%%%%%%%%%% FIG: log distributions, case analysis
374: %\newpage
375: %\vspace*{-20mm}
376: \begin{figure}
377: \begin{center}
378: \includegraphics[width=0.25\textwidth,angle=-90]{stat.ps}
379: \end{center}
380: \caption{\label{LOG-f-CASES}
381: Comparison of the generic stationary distributions of the Langevin-type
382: equation with both additive and multiplicative noise. Limiting cases,
383: as the Gauss-, the Gamma- and the Power-Law distribution are labeled
384: correspondingly. The label 'Zero' refers to the case with $\vartheta=0$.
385: % The numbers in the legend code the parameters applied
386: % in the order $G, F, D, C,$ and $B$. $s$ stands for a small number,
387: % $L$ for a large one. The critical momenta in these cases are
388: % $1$ GeV for the Power-Law, $0.1$ GeV for the Gamma-, $10$ GeV
389: % for the Gauss- and for the degenerate 'Zero' distribution.
390: }
391: \end{figure}
392:
393: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
394:
395: %\vs
396: In the case $B=0$ (no cross correlation between the noises), and
397: $F=0$ (no drift term due to the additive noise), the {\em exact} stationary
398: distribution is the Tsallis distribution\cite{TSALLIS-DIST}.
399: In order to achieve this
400: result one uses the energy of the free particle, $E=p^2/2m$ as the
401: distribution variable and eq.\eqref{SOLUTION}. We get
402: \be
403: f(E) = f_0 \, \left(1 + (q-1) \frac{E}{T} \right)^{\frac{q}{1-q}}
404: \label{TSALLIS}
405: \ee
406: where the parameters of the Tsallis distribution are given by
407: \be
408: T = \frac{D}{mG}, \qquad q = 1 + \frac{2C}{G}.
409: \label{TS-PARAMS}
410: \ee
411: Again for $C=0$ (only additive noise) $q=1$ and the Tsallis distribution
412: goes over into the Gibbs distribution,
413: \be
414: f(E) = f_0 \exp(-E/T).
415: \label{GIBBS}
416: \ee
417: Tsallis and others have worked out a thermodynamical framework
418: offering the distribution \eqref{TSALLIS} as the canonical distribution.
419: This approach, the non-extensive thermodynamics, however, is based on
420: a non-extensive entropy measure. This eventually unwanted property is
421: not fatal: the distribution \eqref{TSALLIS} can also be obtained
422: based on the extensive R\'enyi entropy\cite{RENYI}.
423: We note by passing this point that the presence of the two uncorrelated
424: noise and the corresponding Fokker-Planck equation also can be obtained
425: from an inhomogeneous diffusion coefficient. Instead of \eqref{eq:lan} one
426: may consider
427: \be
428: \dot{p} + Gp = (D+Cp^2)^{1/2} \eta
429: \ee
430: With a single noise $\eta$, normalized to unity:
431: \be
432: \exv{\eta(t)} = 0, \qquad \exv{\eta(t)\eta(t')}= 2 \delta(t-t').
433: \ee
434: This is a particular case of a more general field-dependent noise
435: considered by Arnold, Son and Yaffe in the context of non-abelian
436: plasmas\cite{ARNOLD}.
437:
438: %\vs
439: The mechanism outlined in the present article may work for
440: non-abelian gauge theories, as well.
441: %Here we just would like to sketch
442: %the principal mechanism and check order of magnitude estimates,
443: %detailed calculations are postponed for future work.
444: The basic gluonic field, described by a vector potential $A$, satisfies
445: the equation
446: \be
447: \ddot{A} + \sigma \dot{A} = D \times (D \times A),
448: \label{BODEK}
449: \ee
450: with $D$ being the gauge-covariant derivative and $\sigma$ the
451: %inverse ohmic resistance or
452: color conductivity factor. % Following Bodeker
453: We consider an overdamped dynamics, when the second
454: time derivative is ignored.
455: %This means we deal with the low-frequency behavior, we select out the soft
456: %fields to observe.
457: Driving this to the extreme we single out
458: in a Fourier expansion the zero mode, $A_0$
459: and consider the following effective equation
460: \be
461: \sigma \dot{A}_0 = \sum_{k,q} (k-gA_k)(q-gA_q)A_{-k-q}.
462: \label{PLASMA0}
463: \ee
464: Treating the the $k=0$ and $q=0$ contributions on the right hand side
465: separately we arrive at
466: \be
467: \dot{A}_0 - \frac{g^2}{\sigma} A_0^3 = - \gamma A_0 + \xi,
468: \label{PLASM}
469: \ee
470: with
471: \ba
472: \gamma \quad &=& \quad - 2\frac{g}{\sigma} \sum_{k>0} (k-gA_k)A_{-k}, \NL
473: \xi \quad &=& \quad \frac{1}{\sigma} \sum_{k>0,q>0} (k-gA_k)(q-gA_q)A_{-k-q}.
474: \label{PL-XI}
475: \ea
476: Ignoring the classical $ \sim A_0^3$ contribution for the following
477: discussion, one realizes that $\gamma$ and $\xi$ do contain hard
478: Fourier component contributions. In the Langevin equation with
479: both additive and multiplicative noise these are regarded as noisy,
480: fast-fluctuating quantities.
481:
482: %%%%%%%%%%%%%%%%%%% FIG: graphical notation
483: %\newpage
484: %\vspace*{-20mm}
485: \begin{figure}
486: %\vspace*{-36mm}
487: \begin{center}
488: \includegraphics[width=0.25\textwidth]{graf.eps}
489: \end{center}
490: \caption{\label{graf}
491: }
492: \vspace*{-2mm}
493: \end{figure}
494: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
495:
496: %\vs
497: The full quantum field theoretical treatment of these factors is
498: rather involved.
499: In order to gain estimates over the parameters of a possible stationary
500: distribution of $A_0^2$ values,
501: %-- and through this over the energy
502: %stored in the soft sector $E \approx V T^2 A_0^2 $, where $T$
503: %here is a typical hard momentum scale characteristic to the medium --
504: averages and correlations of $\gamma$ and $\xi$ have to be obtained.
505: This process can be facilitated by using a graphical notation:
506: external legs stand for the zero mode and internal lines for
507: the hard modes. (cf. Fig.\ref{graf}).
508: For no spontaneous symmetry breaking, $F=0$ and $B=0$ immediately
509: follows, so the stationary distribution is a Tsallis-distribution
510: for $A_0^2$ and eventually in the soft energy
511: $E \sim V\Lambda^2A_0^2 = A_0^2/(2m)$,
512: where $\Lambda$ is the underlying energy scale.
513: A qualitative, order of magnitude estimate for the parameters of
514: the Tsallis distribution can be given as: $T \sim g^2\Lambda^4/(m\sigma^2)$
515: and $q - 1 \sim g^2\Lambda^2/\sigma^2$. The starting point of the
516: power law in eq.\eqref{TSALLIS} is at $E_c = T/(q-1)\sim\Lambda^2/m$.
517:
518: %\vs
519: Finally we would like to check whether quantitative
520: estimates give reliable results. In high energy particle physics
521: experiments the transverse momentum distribution has been
522: investigated for long. From Gaussian fits to
523: the parton distribution\cite{PARTON-GAUSS} one conjectures a ratio
524: $D/G = \exv{p_t^2} \approx 1\: - \: 1.5$ GeV$^2$. On the
525: other hand power-law tails at high transverse momenta make a
526: value of $v \approx 5.8 \pm 0.5$ realistic. This fixes the
527: ratio to $G/C \approx 9.6 \pm 1$
528: and the Tsallis index to $q \approx 1.2 \pm 0.03 $.
529: The critical transverse momentum, beyond which the power-law
530: dominates the familiar Gauss distribution, can be calculated
531: from this to be $p_c \approx 3\: - \:4$ GeV. Compared to experimental
532: spectra\cite{POW-TAIL-SPECTRA} this is a quite realistic estimate.
533:
534:
535: \vspace{2mm}
536: %{\bf Conclusion.}
537: {\em In conclusion} we have shown that the smallest multiplicative
538: white-noise related to the classically deterministic damping constant
539: in the Langevin equation leads to a stationary distribution
540: of particle momenta, which ends in a power-law tail at high
541: values. This result seems to undermine the well-established
542: thermal approaches to phenomenological and field theoretical
543: studies in particle physics, where the presence of a multiplicative
544: noise is not less probable than the presence of an additive one
545: in any simplification (linearization) of the underlying
546: microdynamical problem. This approach on the other hand offers
547: a new way to deal with the interpretation of power-law tails
548: occurring in experimental findings, as well as it animates to seek
549: new methods in thermal field theory exceeding the traditional
550: thermodynamical approach.
551: %stuck to the assumption of the Gibbs distribution.
552:
553:
554:
555: %\vs
556: {\bf Acknowledgment}
557: Discussions with J. Zim\'anyi and A. Parvan about particle spectra
558: and thermodynamics are gratefully
559: acknowledged. Special thank from T.S.B. to B. M\"uller and C. Greiner with whom
560: he started to speculate about the role of noise in field theory.
561: This work has been supported by the
562: Hungarian National Research Fund OTKA (T034269, T037689, T034980, F043465).
563:
564:
565: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
566:
567:
568: \begin{thebibliography}{xxxxx}
569:
570: \bibitem{POWER-TAILS} A recent review: R.Botet, M.Ploszajczak,
571: {\em Universal Fluctuations (The Phenomenology of Hadronic Matter)},
572: World Scientific Lecture Notes in Physics Vol. 65, 2002.
573:
574: %\bibitem{LONG-RANGE} LONG-RANGE: review of long range order
575:
576: \bibitem{LANGEVIN} G.E.Uhlenbeck, L.S.Ornstein, Phys.Rev.36, 823, 1930.
577:
578: \bibitem{POW-TAIL-SPECTRA} ZEUS Collaboration, Eur.Phys.J.C 11, 251, 1999;
579: PHENIX Collaboration, Acta Phys.Pol.B 35, 1081, 2004; STAR Collaboration,
580: nucl-ex/0403059;
581:
582: \bibitem{NOISE-FIELD} C.Greiner, B.M\"uller, Phys.Rev.D 55, 1026, 1997;
583: T.S.Bir\'o, C.Greiner, Phys.Rev.Lett. 79, 3138, 1997;
584: C.Greiner, S.Leupold, Ann.Phys. 270, 328, 1998;
585: C.Greiner, S.Juchem, Z.Xu, hep-ph/0404022.
586:
587: \bibitem{PLASMA} D.B\"odeker, Nucl.Phys.B 509, 502, 1999;
588: ibid 647, 512, 2002; Phys.Lett.B 516, 175, 2001;
589:
590: \bibitem{UHLENBECK} R.J.Rivers, {\em Path Integral Methods in Quantum
591: Field Theory}, Cambridge University Press, 1987;
592:
593: \bibitem{ITO} K.Ito, H.McKean Jr., {\em Diffusion Process and their
594: Sample Paths}, Springer-Verlag, Berlin, 1965.
595:
596: \bibitem{WILK} G.Wilk, Z.Wlodarczyk, Phys.Rev.Lett. 84, 2770, 2000;
597: Physica A 305, 227, 2002.
598:
599: \bibitem{TSALLIS-DIST} C.Tsallis, J.Stat.Phys. 52, 50, 1988; Physica A 221, 277,
600: 1995;
601:
602: \bibitem{RENYI} A.R\'enyi, Acta Math.Acad.Sci.Hung. 10, 193, 1959;
603: S. Abe, Phys.Rev.E 63, 061105, 2001; Q.A.Wang, L.M\'ehaut\'e,
604: J.Math.Phys. 43, 5079, 2002.
605:
606: \bibitem{ARNOLD} P.Arnold, D.T.Son, L.G.Yaffe, Phys.Rev.D 60, 025007, 1999;
607: P.Arnold, Phys.Rev.D. 62, 036003, 2000;
608:
609: \bibitem{PARTON-GAUSS} Y.Zhang, G.I.F\'ai, G.Papp, G.C.Barnaf\"oldi, P.L\'evai,
610: Phys.Rev.C 65, 034903, 2002.
611:
612: \end{thebibliography}
613: \end{document}
614: