cond-mat0305132/v1.tex
1: %\documentstyle[aps,epsf]{revtex}
2: \documentstyle[aps,epsf,multicol]{revtex}
3: \parskip=5pt
4: %%\renewcommand{\baselinestretch}{1.7}
5: \begin{document}
6: \draft
7: 
8: \title{Application of the Bogolyubov's theory of weakly \\
9: non-ideal Bose gas on the A+A, A+B, B+B reaction-diffusion system}
10: 
11: \author{Zoran Konkoli}
12: 
13: \address{
14:   Department of Applied Physics \\
15:   Chalmers University of Technology \\ 
16:   and G\"oteborg University, \\
17:   SE 412 96 G\"oteborg, \\
18:   Sweden}
19: 
20: \date{\today}
21: \maketitle
22: \begin{abstract}
23: Theoretical methods for dealing with diffusion-controlled reactions
24: inevitably rely on some kind of approximation and to find the one that
25: works on a particular problem is not always easy. In here the
26: approximation used by Bogolyubov to study weakly non-ideal Bose gas,
27: to be refereed to as weakly non-ideal Bose gas approximation (WBGA),
28: is applied in the analysis of of the three reaction-diffusion models
29: (i) $A+A\longrightarrow \emptyset$, (ii) $A+B \longrightarrow
30: \emptyset$ and (iii) $A+A, B+B, A+B \longrightarrow \emptyset$ (the
31: ABBA model).  The two types of WBGA are considered, the simpler WBGA-I
32: and more complicated WBGA-II. All models are defined on the lattice to
33: facilitate comparison with computer experiment (simulation). It is
34: found that the WBGA describes A+B reaction well, it reproduces
35: correct $d/4$ density decay exponent. However, it fails in the case of
36: the A+A reaction and the ABBA model.  (To cure deficiency of WBGA in
37: dealing with A+A model the hybrid of WBGA and Kirkwood superposition
38: approximation is suggested.)  It is shown that the WBGA-I is identical
39: to the dressed tree calculation suggested by Lee in J. Phys. A {\bf
40: 27}, 2633 (1994), and that the dressed tree calculation does not lead
41: to the d/2 density decay exponent when applied to the A+A reaction, as
42: normally believed, but it predicts d/4 decay exponent. Last, the usage
43: of the small $n_0$ approximation suggested by Mattis and Glasser in
44: Rev. Mod. Phys. {\bf 70}(3), 979 (1998) is questioned if used beyond
45: A+B reaction-diffusion model.
46: \end{abstract}
47: \pacs{}
48: 
49: \begin{multicols}{2}
50: \narrowtext
51: 
52: \section{Introduction}
53: 
54: A variety of methods have been used to study diffusion-controlled
55: reactions (see, e.g.,
56: refs.~\cite{rev1,rev2,rev3,rev4,rev5,rev6,rev7,rev8} for review)
57: ranging from simplest pair-like or Smoluchowskii approach~\cite{OTB}
58: towards more sophisticated methods such as many particle density
59: function formalism~\cite{rev4,rev5} and field
60: theory~\cite{rev6,rev7,rev8}. In here, we focus on the field theory
61: approach which is exact but, like in any other theory, one needs
62: approximations to solve the problem. The particular way of making
63: approximate calculations will be analyzed, the Bogolyubov's theory of
64: weakly non-ideal Bose gas~\cite{bog}, to be referred to in the
65: following as Weakly non-ideal Bose Gas Approximation (WBGA).
66: 
67: The field theory is very attractive since it offers systematic way to
68: do calculation and there are well known procedures how to represent
69: stochastic dynamics of reaction-diffusion systems as field
70: theory. Such mapping can be carried out for lattice- and off-lattice
71: models, see \cite{rev6} and \cite{OTB} respectively for description.
72: Final result is that the dynamics is governed by a second quantized
73: Hamiltonian $H(a,a^\dagger)$,
74: %
75: \begin{equation}
76:   \frac{\partial}{\partial t} \Psi(t) = - H(a,a^\dagger) \Psi(t)
77:   \label{Hpsi}
78: \end{equation}
79: %
80: and system configuration at time $t$ can be extracted from state
81: vector $\Psi(t)$.  All observables can be expressed as a field
82: theoretic averages over products of creation and annihilation
83: operators $a^\dagger$ and $a$.  
84: 
85: 
86: There is no need whatsoever to resort to the use of field theory as
87: all calculations could be done without it.  The reasons for using it
88: are mostly practical. Field theory is a powerful book-keeping
89: device. First, approximations involved can be made more transparent
90: and, second, due to this transparency it is relatively straightforward
91: to borrow approximations from other problems (and related forms of
92: field theory).
93: 
94: However, despite its elegance, when applied to the reaction-diffusion
95: systems, field theory seems to work with limited success. Most often
96: the only practical procedure of solving field theory is perturbative
97: calculation. Diffusion is set up as the zeroth order problem and
98: reaction is perturbation.  Problem is that in the most interesting
99: regime, when dimension of the system is low (bellow some critical
100: dimension), perturbation series diverges due to infra-red divergences.
101: Any non-field theoretic approach which aims at perturbative treatment
102: will suffer in a similar way.
103: 
104: The infra-read divergences are normally controlled in the framework of
105: the renormalization group (RG) technique.  When applying RG, the
106: initial density is relevant parameter and grows under renormalization
107: process.  Growing initial densities lead to infinities in expressions
108: which have to be controlled. So far, such control exists only in the
109: limited amount of cases, the A+A reaction being the simplest
110: example. Clearly, there is a need to avoid perturbative treatment
111: (together with RG) and look for alternative ways of calculation. In
112: here, focus is on the WBGA.
113: 
114: The WBGA was used to study reaction-diffusion systems
115: previously~\cite{rev8,OTB,ZO,BOP,GMY}.  In somewhat technical terms,
116: the basic idea behind WBGA is to neglect the products containing three
117: or higher number of annihilation operators having non-zero wave vector
118: in the Hamiltonian. This procedure leads to closed set of equations
119: for particle density and correlation functions. Solving these
120: equations amounts to resummation of infinite series of diagrams. In
121: this way WBGA appears as a non-perturbative technique which can be used
122: to control infra-red divergences of perturbation series. In the
123: following, we will distinguish between two types of approximations to
124: be refereed to as WBGA-I or WBGA-II which result in linear or
125: nonlinear set of equations of motion respectively.
126: 
127: On the more intuitive basis, the neglect of fluctuations on the
128: shorter length scale (large $k$ vectors) has to do with the slowness
129: of diffusion. For initial conditions when particle density is uniform
130: there is no $k\ne 0$ component in average particle density and the
131: density profile is flat. Since diffusion is a very slow process one
132: expects that this profile is kept all the time up to a small
133: perturbation. This picture is likely to be true when particle density
134: is low when compared to the reaction range, which is summarized in the
135: condition that $n r^d\ll 1$ where $n$ and $r$ denote the particle
136: density and the reaction range~\cite{OTB}. Naturally, there are other
137: scales in the problem (which change in time) and this kind of thinking
138: can not be employed without encountering problems~\cite{rev4}. In
139: reality, one really has to test the WBGA on the particular model to
140: see whether it works.
141: 
142: To analyze properties of the WBGA method, we apply it in the analysis
143: of three diffusion-controlled reactions (i) the A+A, (ii) the A+B and
144: (iii) the ABBA model. The ABBA model includes A+A, A+B and B+B
145: reactions simultaneously which allows for competitions between A+A and
146: A+B reactions (see ref.~\cite{KJ1,KJ2} for more details). The A+A and
147: A+B models have been studied by variety of methods (see, e.g.,
148: ref.~\cite{CGP} for review).  In here, the focus is on the WBGA
149: studies.  Apart from few initial studies of the A+A and A+B models the
150: WBGA have not been widely used. These early applications of WBGA on
151: A+A and A+B reaction-diffusion systems are mostly done for stationary
152: situation (and decay from it) and are listed in refs.~\cite{ZO} (A+A)
153: and \cite{BOP,GMY} (A+B). Also, the WBGA reappears in
154: paper~\cite{rev6} where it is used to study the A+B reaction.
155: 
156: The purpose of this work is twofold. First, we study A+A and A+B
157: reactions with uncorrelated initial conditions with particles
158: Poisson-distributed~\cite{foot2} at initial time $t=0$. In previous
159: work which employed WBGA method in refs.~\cite{ZO,BOP,GMY} it was
160: assumed that particles are initially correlated, with a initial
161: condition being prepared in a somewhat special way~\cite{foot1}.
162: Second, once the workings of the WBGA approximation is illuminated, we
163: apply the WBGA method to analyze the more complicated ABBA model. This
164: model was studied previously with field theoretic RG technique using
165: $\epsilon$-expansion and numerical simulation~\cite{KJ1,KJ2}. The ABBA
166: model describes a mixture of the A+A and A+B reactions which occur at
167: the same time. The A+A (A+B) component bias kinetics towards $d/2$
168: ($d/4$) decay exponent and it is not a priori clear which exponent
169: will prevail. The analysis in~\cite{KJ1,KJ2} shows that both A and B
170: particles decay with $d/2$ decay exponent and different amplitudes.
171: 
172: The A+A model has been studied by variety of methods and lot is know
173: about it's behavior.  The study most relevant for this work is given
174: in ref.~\cite{Lee}. For uncorrelated initial conditions particle
175: density $n$ decays asymptotically as $n\sim {\cal A}(Dt)^{-d/2}$ for
176: $d<2$ while there is logarithmic correction at $d=2$ as $n\sim\ln(
177: Dt)/(8\pi Dt)$, where $d$ denotes dimension of system, $t$ is time and
178: $D$ denotes diffusion constant. Amplitude ${\cal A}$ is universal
179: number independent on the details of the model. Also, the exact
180: solution of the model is possible at $d=1$ where $n\sim 1/\sqrt{8\pi
181: Dt}$~\cite{Lush}.
182: 
183: The $d/2$ decay exponent of the A+A reaction is robust in a sense that
184: even simplest pair like approach (e.g. Smoluchowskii) predicts this
185: exponent. Thus, it is natural to test WBGA on the A+A model. It will
186: be shown that, somewhat surprisingly, WBGA fails to describe A+A
187: reaction: WBGA-I predicts $d/4$ decay exponent while WBGA-II gives
188: mean field exponent.
189: 
190: Taking slightly more complicated reaction scheme, such as the A+B,
191: leads to very hard problem where calculation methods start to differ
192: in their predictions. The nature of the asymptotics of A+B reaction
193: when initial number of A and B particles is equal has been hotly
194: debated for decades. For example, Smoluchowskii approach predicts $d/2$
195: decay exponent~\cite{rev4,rev5,OTB}. Field theoretic RG method predicts
196: correct $d/4$ decay for $3<d<4$~\cite{LeeCardy}, while for other
197: dimensions nothing can be said in the RG framework. Strangely enough,
198: seemingly the most sophisticated coupled-cluster technique argues for
199: mean field exponent~\cite{Rudav}. The true decay is governed by
200: $d/4$ exponent as shown by work of Bramson and Lebowitz which provided
201: strict mathematical proof~\cite{bram}. It will be shown that when
202: particles are initially uncorrelated, {\em i.e.}  Poisson-distributed,
203: both the WBGA-I and WBGA-II predict $d/4$ decay exponent. Strangely
204: enough, WBGA fails on simpler A+A reaction while it describes properly
205: more complicated A+B reaction.
206: 
207: 
208: The paper is organized as follows. In section \ref{sec:model} the
209: model to be studied is specified in detail, and mapping to the field
210: theory is described. The outcome of this section is Hamiltonian which
211: describes the most general two species reaction-diffusion model. In
212: section \ref{sec:EOM} equations of motion for density and correlation
213: functions are derived using WBGA method. In section \ref{sec:AA+WBGA}
214: the WBGA approach is used to describe A+A model, with application of
215: WBGA-I discussed in section \ref{sec:AA+WBGAI}, and application of
216: WBGA-II in section \ref{sec:AA+WBGAII}. Also in section
217: \ref{sec:AA+WBGAI} equivalence of dressed tree calculation with WBGA-I
218: is shown. Section \ref{sec:AA+hybrid} discusses modification of
219: Kirkwood superposition approximation in the spirit of WBGA. In section
220: \ref{sec:AB+WBGA} A+B model is studied by using WBGA-I and WBGA-II
221: methods. Finally, the analysis of the most complicated of three models
222: studied here, the ABBA model, is presented in section
223: \ref{sec:ABBA+WBGA}. Workings of WBGA-I are analyzed in section
224: \ref{sec:ABBA+WBGAI} and of WBGA-II in section
225: \ref{sec:ABBA+WBGAII}. The step by step merger of WBGA and Kirkwood
226: superposition approximation is discussed in \ref{sec:ABBA+hybrid}. The
227: summary and outline of future work is given in section
228: \ref{sec:conclusions}.
229: 
230: 
231: 
232: \section{The model and the mapping to the field theory}
233: \label{sec:model}
234: 
235: The mapping to the field theory will be carried out for the most
236: general two-species reaction-diffusion model.  The A+A, A+B and the
237: ABBA model will be obtained as special cases.  It will be assumed that
238: particles can not be created, neither by external source, nor by birth
239: process.  In the most general version of the two species
240: reaction-diffusion model each of the $A$ or $B$ particles jumps onto
241: the one of the neighboring lattice sites with rate (or diffusion
242: constant) $D_A$ or $D_B$, irrespectively of occupation number of the
243: site where particle jumps to.  Apart from diffusing particles
244: annihilate in pairs. Particle $\rho$ sitting at lattice site $x$ and
245: particle $\nu$ sitting at the site $y$ are assumed to annihilate with
246: rate $\sigma^{\rho\nu}_{xy}$. This is schematically denoted as
247: %
248: \begin{equation}
249:   \rho(x)+\nu(y)\rightarrow 0 
250: \end{equation}
251: %
252: where $\rho,\nu=A,B$ and $x,y=1,2,3,\ldots,V$. It is
253: assumed that there are $L$ lattice sites in one direction, and for $d$
254: dimensions total number of sites equals $V=L^d$. Labels $x$ and $y$
255: denote lattice sites.
256: 
257: The stochastic dynamics of the most general two-species model is
258: governed by the master equation of the system. The master equation
259: describes transitions between different configurations $c$. The
260: configuration of the system is specified by the occupancy of lattice
261: sites at certain time $t$;
262: %
263: \begin{equation}
264:   c\equiv(n_1,m_1,\ldots,n_x,m_x,\ldots,n_y,m_y,\ldots,n_V,m_V)
265: \end{equation}
266: %
267: $n_x$ and $m_x$ count number of A and B particles respectively at
268: $x$-th lattice site; $n_x,m_x=0,1,2,\ldots$ and $x=1,2,\ldots,V$. The
269: quantity of interest is $P(c;t)$, which denotes probability that at
270: given time $t$ system is in the configuration $c$. 
271: 
272: Diffusion and reaction contribute independently to change of $P(c;t)$;
273: %
274: \begin{equation}
275:   \frac{\partial}{\partial t}P(c;t) = 
276:      \dot P_{\rm D}(c;t) + \dot P_{\rm R}(c;t) 
277:   \label{ME}
278: \end{equation}
279: %
280: The influence of diffusion is described by
281: %
282: \begin{eqnarray}
283:   \dot P_{\rm D}(c;t) & = & 
284:       \sum_x\sum_{e(x)} \{ D_A [ (n_i+1) P(/n_x+1,n_e-1/;t) \nonumber \\
285:   & & - n_i P(c;t) ] + D_B [ (m_i+1) \nonumber \\
286:   & & \times P(/m_x+1,m_e-1/;t) - m_i P(c;t) ] \}
287:   \label{ME1}     
288: \end{eqnarray}
289: %
290: where notation $P(/\ldots/;t)$ indicates that content of the sites
291: specified by $/\ldots/$ has been modified in $c$. The $\sum_{e(x)}$
292: denotes sum over all first-neighbor sites of the site $x$. The change
293: of $P(c;t)$ driven by reactions is given by
294: %
295: \begin{eqnarray}
296:   & & \dot P_{\rm R}(c;t) = \sum_x \sigma^{AA}_{xx} \frac{(n_x+2)(n_x+1)}{2}
297:       P(/n_x+2/;t) \nonumber \\
298:   & & \ + \sum_x \sigma^{BB}_{xx} \frac{(m_x+2)(m_x+1)}{2}
299:       P(/m_x+2/;t) \nonumber \\
300:   & & \ + \sum_x \sigma^{AB}_{xx} (n_x+1)(m_x+1)
301:       P(/n_x+1,m_x+1/;t) \nonumber \\  
302:   & & \ + \sum_{x>y} \sigma^{AA}_{xy} (n_x+1)(n_y+1)
303:       P(/n_x+1,n_y+1/;t) \nonumber \\
304:   & & \ + \sum_{x>y} \sigma^{BB}_{xy} (m_x+1)(m_y+1)
305:       P(/m_x+1,m_y+1/;t) \nonumber \\
306:   & & \ + \sum_{x\ne y} \sigma^{AB}_{xy} (n_x+1)(m_y+1)
307:       P(/n_x+1,m_y+1/;t) \nonumber \\
308:   & & \ - \left[ 
309:             \sum_x\sigma^{AA}_{xx} \frac{n_x(n_x-1)}{2}
310:             + \sum_x\sigma^{BB}_{xx} \frac{m_x(m_x-1)}{2}  
311:            \right. \nonumber \\
312:   & & \ \ \ \ \ \ \ \  
313:             + \sum_x \sigma^{AB}_{xx} n_x m_x
314:             +  \sum_{x>y} \sigma^{AA}_{xy} n_x n_y 
315:             \nonumber \\
316:   & & \ \ \ \ \ \ \ \ 
317:             \left.
318:             + \sum_{x>y} \sigma^{BB}_{xy} m_x m_y 
319:             + \sum_{x\ne y} \sigma^{AB}_{xy} n_x m_y 
320:           \right] P(c;t) 
321:   \label{ME2}
322: \end{eqnarray}
323: %
324: The form of the master equation describing reactions can not be made
325: more compact since distinction has to be made between reaction on the
326: same and different lattice sites. For example, the term describing A+A
327: reaction at the same lattice site has the pre-factor $n_x(n_x-1)$ while
328: for two different sites another form $n_x n_y$ has to be used.
329: 
330: 
331: The master equation (\ref{ME}) is the first order differential
332: equation and one has to specify initial condition in the form $P(c;0)$
333: for all $c$. In the following it will be assumed that $P(c;0)$ is
334: Poisson distribution with averages denoted by $n_{0,A}$ and
335: $n_{0,B}$ for $A$ and $B$ particles respectively. This means that
336: $P(c;0)$ factorises,
337: %
338: \begin{equation}
339:   P(c;0) = \Pi_{x=0,1,2,\ldots,V} p(n_x,n_{0,A}) p(m_x,n_{0,B}) 
340: \end{equation}
341: %
342: where $p(n,\bar n)$ denotes Poisson distribution function
343: %
344: \begin{equation}
345:   p(n,\bar n) = e^{-\bar n} \bar n^n/n! \ , \ \ n=0,1,2,\ldots
346: \end{equation}
347: 
348: Eq.~(\ref{ME}) can be translated into a Schr\"odinger-type equation
349: (\ref{Hpsi}) by defining state vector,
350: %
351: \begin{eqnarray}
352:  &  \Psi(t) = & \sum_{n_1,m_1,...,n_V,m_V}  
353:                 P(n_1,m_1,...,n_V,m_V;t) \nonumber \\ 
354:  &           & \ \ \ \ \ \times (a_1^\dagger)^{n_1} (b_1^\dagger)^{m_1} ... 
355:                (a_V^\dagger)^{n_V} (b_V^\dagger)^{m_V} 
356:             |0\rangle 
357:   \label{Psi}
358: \end{eqnarray}
359: %
360: where $a_x^\dagger$, $a_x$, $b_x^\dagger$, and $b_x$ are creation and
361: annihilation operators for A and B particles respectively;
362: %
363: \begin{eqnarray}
364:   & & [a_x,a^\dagger_y] = \delta_{x,y} \, , \ 
365:       [b_x,b^\dagger_y] = \delta_{x,y} \, , \nonumber \\  
366:   & & [a_x,b^\dagger_y] = [a^\dagger_x,b_y] = 0 
367: \end{eqnarray}
368: %
369: and $[{\rm I},{\rm II}]\equiv {\rm I}\ {\rm II} - {\rm II}\ {\rm I}$
370: denotes commutator.  One can see that Eq.~(\ref{ME}) is equivalent to
371: the Eq.~(\ref{Hpsi}) provided dynamics is governed by the
372: second-quantized Hamiltonian $H = H_{\rm D} + H_{\rm R}$. Term
373: $H_{\rm D}$ describes diffusion
374: %
375: \begin{equation}
376:   H_{\rm D} = \sum_x \sum_{e(x)} 
377:                  [ D_A a^\dagger_x ( a_x - a_e ) + 
378:                    D_B b^\dagger_x ( b_x - b_e ) ]
379:   \label{Hdiffx}
380: \end{equation}
381: %
382: and $H_{\rm R}$ reactions
383: %
384: \begin{eqnarray}
385:   H_{\rm R} & = & 
386:         \frac{1}{2} \sum_{x,y} \sigma^{AA}_{xy}
387:           (a^\dagger_x a^\dagger_y-1) a_x a_y + \nonumber \\
388:  & &  + \frac{1}{2} \sum_{x,y} \sigma^{BB}_{xy}
389:           (b^\dagger_x b^\dagger_y-1) b_x b_y + \nonumber \\
390:  & &  + \sum_{x,y} \sigma^{AB}_{xy}(a^\dagger_x b^\dagger_y-1) a_x b_y
391:   \label{Hreactx}
392: \end{eqnarray}
393: %
394: Please note that expression for $H_{\rm R}$ has the more compact
395: form than the expression for related part of the master equation
396: (\ref{ME2}) as all sums in (\ref{Hreactx}) over $x$ and $y$ are
397: unrestricted (a first sign of convenient book-keeping). Also the fact
398: that reactions A+A, B+B and A+B happen independently from each other
399: influences the form of $H_{\rm R}$; contributions describing A+A
400: (first line of Eq.~\ref{Hreactx}), B+B (second line) and A+B (third
401: line) enter additively.
402: 
403: Once $H$ has been specified $\Psi(t)$ can be found from (\ref{Hpsi}),
404: %
405: \begin{equation}
406:   \Psi(t) = e^{-H t} \Psi_0
407:   \label{PsiHPsi0}
408: \end{equation}
409: %
410: In general, form of $\Psi_0$ depends on the initial particle
411: distribution and for Poisson-distributed particles equals
412: %
413: \begin{equation}
414:   \Psi_0 = {\rm exp} 
415:            \left[ 
416:               n_{0,A}\sum_x ( a_x^\dagger - 1) + 
417:               n_{0,B}\sum_x ( b_x^\dagger - 1 ) 
418:            \right] 
419:            |0\rangle
420: \end{equation}
421: %
422: Since $\Psi(t)$ is available (at least in principle) one can calculate
423: various observables with following recipe
424: %
425: \begin{equation}
426:   \langle O(n_x(t)) \rangle = 
427:   \langle 1 | O(a^\dagger_x a_x) |\Psi(t)\rangle
428:   \label{obs}
429: \end{equation}
430: %
431: where $O(n_x)$ denotes any function which depends on particle numbers
432: in arbitrary way. The generalization of Eq.~(\ref{obs}) for more
433: complicated form for $O(n_x,m_y,\ldots)$ is trivial. Vector $\langle
434: 1|$ is given by
435: %
436: \begin{equation}
437:   \langle 1| \equiv \langle 0| {\rm exp} \left[ \sum_x ( a_x + b_x ) \right] 
438:   \label{bra1}
439: \end{equation}
440: %
441: being left eigenvector of $a^\dagger_x$ and $b^\dagger_x$ with
442: eigenvalue $1$.  To see that Eq.~(\ref{obs}) indeed evaluates
443: observables correctly one can expand function $O$ in Taylor series in
444: $n_x$ and check that for every term in Taylor series
445: Eq.~(\ref{obs}) works.
446: 
447: Following ref.~\cite{rev6}, since $\langle 1 |a^\dagger_x \ne 0$, it is
448: useful to transform Eq.~(\ref{obs}) slightly in order to obtain form
449: where $\langle 1 |$ changes into the vacuum state $\langle 0 |$. This
450: makes calculation somewhat easier since one can use property of the
451: vacuum that $\langle 0 | a^\dagger_x = 0$. To do the transformation
452: following identity proves useful,
453: %
454: \begin{equation}
455:   \langle 1 | O(a^\dagger_x, a_x ) |0\rangle = 
456:   \langle 0 | O(a^\dagger_x+1, a_x ) |0\rangle  
457:   \label{obs1}
458: \end{equation}
459: %
460: and by using (\ref{obs1}) Eq.~(\ref{obs}) can be written in the form
461: %
462: \begin{equation}
463:   \langle O(n_x(t)) \rangle = 
464:   \langle 0 | O((a^\dagger_x+1)a_x) e^{-\bar Ht}|\bar\Psi_0\rangle
465:   \label{obs2}
466: \end{equation} 
467: %
468: where bar over $H$ and $\Psi_0$ indicates that substitution
469: %
470: \begin{equation}
471:   a^\dagger_x\rightarrow a^\dagger_x+1 \ , \
472:   b^\dagger_x\rightarrow b^\dagger_x+1 \ , \ \ x=0,1,2,...,V
473:   \label{adag1}
474: \end{equation}
475: %
476: have been made. The $\bar\Psi_0$ is given by
477: %
478: \begin{equation}
479:   \bar\Psi_0 = {\rm exp} 
480:            \left[ 
481:               n_{0,A}\sum_x a_x^\dagger  + 
482:               n_{0,B}\sum_x b_x^\dagger  
483:            \right] 
484:            |0\rangle
485: \end{equation}
486: %
487: and $\bar H=\bar H_{\rm D}+\bar H_{\rm R}$. After the shift
488: $H_{\rm D}$ does not change the form, i.e. $H_{\rm D}=\bar
489: H_{\rm D}$, while $H_{\rm R}$ changes into
490: %
491: \begin{eqnarray}
492:   \bar H_{\rm R} & = & 
493:         \frac{1}{2} \sum_{x,y} \sigma^{AA}_{xy}
494:           (a^\dagger_x + a^\dagger_y + a^\dagger_x a^\dagger_y) 
495:           a_x a_y + \nonumber \\
496:  & &  + \frac{1}{2} \sum_{x,y} \sigma^{BB}_{xy}
497:           (b^\dagger_x + b^\dagger_y + b^\dagger_x b^\dagger_y) 
498:           b_x b_y + \nonumber \\
499:  & &  + \sum_{x,y} \sigma^{AB}_{xy}
500:         (a^\dagger_x + b^\dagger_y + a^\dagger_x b^\dagger_y) a_x b_y
501:   \label{Hreactx'}
502: \end{eqnarray}
503: %
504: In the following we compactify notation and use
505: %
506: \begin{equation}
507:   \langle f(a_x,a^\dagger_y,\ldots) \rangle \equiv
508:   \langle 0 | f(a_x,a^\dagger_y,\ldots) 
509:   e^{-\bar H t}| \bar\Psi_0\rangle 
510: \end{equation}
511: %
512: where $f$ denotes any function of operators $a_x$, $a^\dagger_y$,
513: $b_x$ and $b^\dagger_y$ for any $x$ and $y$. Also notation 
514: %
515: \begin{equation}
516:   \bar\Psi(t) \equiv e^{-\bar H t} \bar\Psi_0
517:   \label{BarPsiHPsi0}
518: \end{equation}
519: %
520: will be useful.
521: 
522: To avoid effects of boundaries periodic boundary conditions are
523: assumed and one can introduce Fourier transforms for operators,
524: %
525: \begin{eqnarray}
526:  & &  a_x = \frac{1}{\sqrt{V}} \sum_k e^{ikx} a_k \, , \  
527:       a^\dagger_x = \frac{1}{\sqrt{V}} \sum_k e^{-ikx} a^\dagger_k 
528:       \label{akx}\\
529:  & &  b_x = \frac{1}{\sqrt{V}} \sum_k e^{ikx} b_k \, , \ 
530:       b^\dagger_x = \frac{1}{\sqrt{V}} \sum_k e^{-ikx} b^\dagger_k
531:       \label{bkx} 
532: \end{eqnarray}
533: %
534: and reaction rates
535: %
536: \begin{equation}
537:   \sigma^{\rho\nu}_x = \frac{1}{V} \sum_k \sigma^{\rho\nu}_k e^{ikx} \, , \ 
538:     \sigma^{\rho\nu}_k = \sum_x \sigma^{\rho\nu}_x e^{-ikx} 
539:   \label{sigmakx} 
540: \end{equation}
541: %
542: with $\rho,\nu=A,B$. For convenience, Fourier transforms are defined
543: slightly differently for operators and reaction rates (just to reduce
544: explicit occurrence of $V$ in expressions later on). Also, it is useful
545: to express Hamiltonian $\bar H$ in terms of creation and annihilation
546: operators in $k$-space. Starting from (\ref{Hdiffx}) and
547: (\ref{Hreactx'}), and using (\ref{akx})-(\ref{sigmakx}) gives
548: %
549: \begin{equation}
550:   H_{\rm D} =   D_A \sum_k k^2 a^\dagger_k a_k 
551:                  + D_B \sum_k k^2 b^\dagger_k b_k + {\cal O}(k^4) 
552:   \label{Hdiffk}
553: \end{equation}
554: and
555: %
556: \begin{eqnarray}
557:  \bar H_{\rm R} & = & 
558:         \frac{1}{\sqrt{V}} 
559:         \sum_{q,k} \sigma^{AA}_{q} a^\dagger_k a_{k-q} a_q \nonumber \\
560:  & &  + \frac{1}{2V} \sum_{q,k,l} \sigma^{AA}_q 
561:           a^\dagger_k a^\dagger_l a_{k-q} a_{l+q} \nonumber \\
562:  & &  + \frac{1}{\sqrt{V}} \sum_{q,k} \sigma^{BB}_{q} b^\dagger_k 
563:             b_{k-q} b_q \nonumber \\
564:  & &  + \frac{1}{2V} \sum_{q,k,l} \sigma^{BB}_q 
565:             b^\dagger_k b^\dagger_l b_{k-q} b_{l+q}  \nonumber \\
566:  & &  + \frac{1}{\sqrt{V}} \sum_{q,k} \sigma^{AB}_{q} 
567:            ( a^\dagger_k a_{k-q} b_q + b^\dagger_k b_{k-q} a_q )  
568:            \nonumber \\
569:  & &  + \frac{1}{V} \sum_{q,k,l} \sigma^{AB}_q 
570:             a^\dagger_k b^\dagger_l a_{k-q} b_{l+q}
571:   \label{Hreactk}
572: \end{eqnarray}
573: %
574: Please note that Eq.~(\ref{Hdiffk}) is an approximation which is valid
575: for small $k$.  The form in Eq.~(\ref{Hreactk}) is exact. Once the
576: explicit form of $\bar H$ is known one can proceed with calculation of
577: particle density, as shown in the next section.
578: 
579: 
580: 
581: 
582: \section{Equations of motion for density and the correlation functions}
583: \label{sec:EOM}
584: 
585: Equation~(\ref{obs2}) summarizes how observables are calculated within
586: the field theory formalism.  We continue with the specific case of
587: local particle densities $n_A(x,t)$ and $n_B(x,t)$, which can be
588: calculated with $O=a^\dagger_x a_x$ and $O=b^\dagger_x b_x$
589: respectively; using (\ref{obs2}) one gets $n_A(x,t)=\langle a_x
590: \rangle$ and $n_B(x,t)=\langle b_x \rangle$, or more explicitly
591: %
592: \begin{eqnarray}
593:  & &  n_A(x,t) = \langle 0 | a_x | \bar\Psi \rangle = 
594:                  \langle 0 | a_x e^{-\bar H t} | \bar\Psi_0 \rangle  
595:       \label{nax} \\ 
596:  & &  n_B(x,t) = \langle 0 | b_x | \bar\Psi \rangle = 
597:                  \langle 0 | b_x e^{-\bar H t} | \bar\Psi_0 \rangle  
598:       \label{nbx}  
599: \end{eqnarray}
600: %
601: 
602: If initial conditions are translationally invariant, which is the case
603: for Poisson-distributed particles, local particles densities are
604: position independent, i.e. $n_A(x,t)=n_A(t)$ and
605: $n_B(x,t)=n_B(t)$. Also at $t=0$ $n_\rho(0)=n_{0,\rho}$ with
606: $\rho=A,B$. It is convenient to re-express $n_A(t)$ and $n_B(t)$ as
607: $n_A(t)\equiv\frac{1}{V}\sum_x n_A(x,t)$ and
608: $n_B(t)\equiv\frac{1}{V}\sum_x n_B(x,t)$. Using field theory $n_A(t)$
609: and $n_B(t)$ can be calculated from
610: %
611: \begin{eqnarray}
612:   & & n_A(t) \equiv \frac{1}{V} \langle \sum_x a_x \rangle 
613:              =  \frac{1}{\sqrt{V}} \langle a_0 \rangle 
614:              \label{nAkx} \\
615:   & & n_B(t) \equiv \frac{1}{V} \langle \sum_x b_x \rangle
616:              =  \frac{1}{\sqrt{V}} \langle  b_0 \rangle 
617:              \label{nBkx}
618: \end{eqnarray}
619: %
620: where the sum over $x$ divided by $\sqrt{V}$ was recognized as $k=0$
621: component of $a_k$ and $b_k$.
622: 
623: The equations of motion for $n_A(t)$ and $n_B(t)$ can be derived as
624: follows. $n_A(t)$ and $n_B(t)$ are proportional to $\langle a_0
625: \rangle$ and $\langle b_0\rangle$ respectively, and it is sufficient
626: to derive equations for them. 
627: Taking time derivative of $\langle a_0 \rangle$
628: and $\langle b_0 \rangle$ gives,
629: %
630: \begin{eqnarray}
631:   & & \frac{\partial}{\partial t} \langle a_0 \rangle 
632:       =  -\langle [a_0,\bar H] \rangle \\
633:   & & \frac{\partial}{\partial t} \langle b_0 \rangle 
634:       =  -\langle [b_0,\bar H] \rangle
635: \end{eqnarray}
636: %
637: The commutator emerges when time derivative acts on ${\rm exp}(-\bar H
638: t)$~\cite{foot3}. Evaluating the commutator with $\bar H$ given in
639: (\ref{Hdiffk}) and (\ref{Hreactk}), and taking into account
640: (\ref{nAkx}) and (\ref{nBkx}) gives
641: %
642: \begin{eqnarray}
643:   \frac{\partial n_A}{\partial t} & = & 
644:      - \left[ 
645:           \sigma^{AA}_0 n_A n_A + \sigma^{AB}_0 n_A n_B + 
646:        \right. \nonumber \\ 
647:     & & \ \ \ 
648:            \left. 
649:                 + \frac{1}{V} \sum_{k\ne 0} 
650:                 ( \sigma^{AA}_k \Gamma_k^{AA} + \sigma^{AB}_k \Gamma_k^{AB})
651:            \right] 
652:        \label{dnAdt} \\
653:   \frac{\partial n_B}{\partial t} & = & 
654:      - \left[ 
655:           \sigma^{BB}_0 n_B n_B + \sigma^{AB}_0 n_A n_B + 
656:        \right. \nonumber \\ 
657:     & & \ \ \ 
658:             \left. 
659:                 + \frac{1}{V} \sum_{k\ne 0} 
660:                 ( \sigma^{BB}_k \Gamma_k^{BB} + \sigma^{AB}_k \Gamma_k^{AB})
661:             \right] 
662:        \label{dnBdt}
663: \end{eqnarray}
664: %
665: where 
666: %
667: \begin{eqnarray}
668:   \Gamma_k^{AA} & \equiv & \langle a_k a_{-k}\rangle  \\ 
669:   \Gamma_k^{BB} & \equiv & \langle b_k b_{-k}\rangle  \\
670:   \Gamma_k^{AB} & \equiv & \langle a_k b_{-k}\rangle
671: \end{eqnarray}
672: %  
673: To derive (\ref{dnAdt}) and (\ref{dnBdt}) the sum over $k$ is split
674: into $k=0$ and $k\ne 0$ parts and assumption is made that $k=0$
675: components are non-fluctuating, {\em i.e.} $\langle a_0 a_0 \rangle
676: \approx \langle a_0 \rangle \langle a_0 \rangle$ and likewise for
677: $\langle a_0 b_0 \rangle$ and $\langle b_0 b_0 \rangle$ (thermodynamic
678: limit).
679: 
680: The equations (\ref{dnAdt}) and (\ref{dnBdt}) involve correlation
681: functions $\Gamma_k^{\rho\nu}$ with $\rho,\nu=A,B$ which we proceed to
682: calculate. The time evolution of correlators is governed by $[O,\bar H]$
683: with $O=a_ka_{-k},\ a_k b_{-k},\ b_k b_{-k}$. Evaluating commutators
684: gives equations of motion,
685: %
686: \begin{eqnarray}
687:   & &  \frac{\partial}{\partial t} \Gamma_k^{AA} =
688:        - 2 D_A k^2 \Gamma_k^{AA}   
689:        - \left[  
690:          \sigma^{AA}_k n_A^2 + \right. \nonumber \\
691:   & &  \ \ \ \ 
692:          + \frac{1}{V} \sum_{q\ne k} 
693:          \sigma^{AA}_q \Gamma^{AA}_{k-q}   
694:          + 2(\sigma^{AA}_0+\sigma^{AA}_k) n_A \Gamma^{AA}_k + \nonumber \\
695:   & &  \ \ \ \ + \left. 
696:          \sigma^{AB}_k n_A ( \Gamma^{AB}_{k} + \Gamma^{AB}_{-k} )
697:          + 2 \sigma^{AB}_0 n_B \Gamma^{AA}_k  \right]
698:        \label{dtGa} \\
699:   & &  \frac{\partial}{\partial t} \Gamma_k^{BB} =
700:        - 2 D_B k^2 \Gamma_k^{BB}   
701:        - \left[  
702:          \sigma^{BB}_k n_B^2 + \right. \nonumber \\
703:   & &  \ \ \ \ 
704:          + \frac{1}{V} \sum_{q\ne k} 
705:          \sigma^{BB}_q \Gamma^{BB}_{k-q}   
706:          + 2(\sigma^{BB}_0+\sigma^{BB}_k) n_B \Gamma^{BB}_k + \nonumber \\
707:   & &  \ \ \ \ + \left. 
708:          \sigma^{AB}_k n_B ( \Gamma^{AB}_{k} + \Gamma^{AB}_{-k} )
709:          + 2 \sigma^{AB}_0 n_A \Gamma^{BB}_k  \right]        
710:        \label{dtGb} \\ 
711:   & &  \frac{\partial}{\partial t} \Gamma_k^{AB} =
712:        - ( D_A + D_B ) k^2 \Gamma_k^{AB} 
713:        - \left[ \sigma^{AB}_k n_A n_B + \right. \nonumber \\
714:   & &  \ \ \ \ 
715:          + \left. \frac{1}{V} \sum_{q\ne k} 
716:          \sigma^{AB}_q \Gamma^{AB}_{k-q} \right]   
717:          - \left[  
718:              ( \sigma^{AA}_0 + \sigma^{AA}_k ) n_A + \right. \nonumber \\
719:   & &  \ \ \ \ 
720:             \left.  + ( \sigma^{BB}_0 + \sigma^{BB}_k ) n_B 
721:            + \sigma^{AB}_0 ( n_A + n_B ) \right] \Gamma^{AB}_k - \nonumber \\
722:   & &  \ \ \ \ 
723:        - \sigma^{AB}_k ( n_A \Gamma^{BB}_k + n_B \Gamma^{AA}_k ) 
724:        \label{dtGc}
725: \end{eqnarray}
726: %
727: Eqs.~(\ref{dtGa}), (\ref{dtGb}) and (\ref{dtGc}) are approximate since
728: correlators of type $\langle a_k a_q a_l \rangle$ with $k\ne 0$, $q\ne
729: 0$, and $l\ne 0$ are assumed to be small. This is exactly the content of 
730: the WBGA.
731: 
732: 
733: Few comments about the inversion symmetry (in $k$-space) of
734: $\Gamma^{\rho\nu}_k$ $\rho,\mu=A,B$ are in order. By construction,
735: $\Gamma^{\rho\rho}_k(t) = \Gamma^{\rho\rho}_{-k}(t)$ for $\rho=A,B$
736: since these functions represent correlations for same operator
737: types. It is not obvious whether this property holds for
738: $\Gamma^{AB}_k$ which describes correlations for different operators
739: type. Using Eq.~(\ref{dtGc}) to derive equation of motion for
740: $\Gamma^{(-)}_k(t)\equiv \Gamma^{AB}_k(t) - \Gamma^{AB}_{-k}(t)$ one
741: can see that $\Gamma^{(-)}_k(t)=0$ is solution provided
742: $\Gamma^{(-)}_k(0)=0$. The initial conditions for $\Gamma^{\rho\nu}_k$
743: with $\rho\nu=A,B$ are given by
744: %
745: \begin{equation}
746:   \Gamma^{\rho\nu}_k(0)=\delta_{k,0} V n_{0,\rho} n_{0,\nu} \ , \ \ 
747:                         \rho,\nu=A,B
748: \end{equation}
749: %
750: where $\delta_{x,y}$ denotes Kronecker delta-function,
751: %
752: \begin{equation}
753:   \delta_{x,y} =
754:     \left\{
755:       \begin{array}{ll}
756:         1 & x=y \\ [5pt]
757:         0 & x\ne y
758:       \end{array}
759:     \right.
760:   \label{krondeltaxy}
761: \end{equation}
762: %
763: Also, notation
764: %
765: \begin{equation}
766:   \bar\delta_{x,y}=1-\delta_{x,y}
767: \end{equation} 
768: %
769: will be used.  Thus $\Gamma^{(-)}_k(0)$ is indeed zero. It follows
770: that $\Gamma^{AB}_k(t)=\Gamma^{AB}_{-k}(t)$ for every $t\ge 0$. In the
771: following whenever $\Gamma^{AB}_k + \Gamma^{AB}_{-k}$ appears in the
772: equations of motion, we will use assumption of inversion symmetry and
773: shorten expression to $2\Gamma^{AB}_k$. Naturally, if this inversion
774: symmetry is broken at $t=0$ one has to keep full form
775: $\Gamma^{AB}_k+\Gamma^{AB}_{-k}$ in the equations of motion, but such
776: case is not considered here.
777: 
778: The last terms in Eqs.~(\ref{dtGa})-(\ref{dtGc}), which are product of
779: density and correlator, and appear to be third order in density ${\cal
780: O}(n^3)$, lead to non-linear equations of motion. These terms come
781: from averages of the type $\langle a_k a_q a_l\rangle$ where one of
782: $\{k,q,l\}$ momenta is zero while remaining two are not.  One example
783: would be $\langle a_0 a_k a_l\rangle$ which is approximatively equal
784: to $\langle a_0 \rangle \langle a_k a_l \rangle$.  (Same discussion
785: would apply irrespectively which operator one uses in average, $a_k$
786: or $b_k$.) Being third order in density, it is tempting to neglect
787: these ${\cal O}(n^3)$ terms, since one expects that leading
788: contribution should come from terms which are second order in density
789: ${\cal O}(n^2)$, e.g. first and second terms on the right hand side of
790: Eq.~(\ref{dtGa}) and likewise for Eqs.~(\ref{dtGb}) and
791: (\ref{dtGc}). To test the effect of neglecting or keeping ${\cal
792: O}(n^3)$ terms two approximations will be studied WBGA-I or WBGA-II
793: with ${\cal O}(n^3)$ terms taken away or kept in the calculation.  WBGA-I
794: approximation results in a linear set of equations which makes
795: the analytical analysis possible. Also, as formulated above, the WBGA-II
796: and WBGA appear to be equivalent. In the following, the term WBGA will
797: imply both WBGA-I and WBGA-II.
798: 
799: The equations of motion for the three models we wish to study, the
800: A+A, A+B and ABBA are easily extracted from the most general form
801: given in (\ref{dnAdt})-(\ref{dnBdt}) and (\ref{dtGa})-(\ref{dtGc}). To
802: obtain equations of motion for the ABBA model one simply sets
803: $D_A=D_B=D$ and
804: %
805: \begin{equation}
806:   \sigma^{AA}_{xy}=\sigma^{BB}_{xy}=\lambda\delta_{x,y} \ , \ \ 
807:   \sigma^{AB}_{xy}=\delta\delta_{x,y}
808:   \label{sigmas}
809: \end{equation} 
810: %
811: An obvious distinction is being made between Kronecker delta-function
812: $\delta_{x,y}$ and the symbol $\delta$ which denotes the reaction
813: rate. Thus particles have to meet at the same lattice site in order to
814: react.  Further, to obtain the A+A model one simply sets $\delta=0$
815: (this decouples A+A and B+B reactions, i.e. particles A and B move and
816: react independently of each other). To get A+B model one takes
817: $\lambda=0$ which rules out the A+A reaction. In the following section
818: we continue with analysis of the A+A model within WBGA framework.
819: 
820: 
821: 
822: 
823: \section{WBGA applied to the A+A reaction}
824: \label{sec:AA+WBGA}
825: 
826: Using (\ref{sigmas}) with $\delta=0$ in (\ref{dnAdt})-(\ref{dnBdt})
827: and (\ref{dtGa})-(\ref{dtGc}) gives equation of motion for the density
828: %
829: \begin{equation}
830:   \frac{\partial n}{\partial t} = - \lambda [ n^2 + \Phi ] 
831:   \label{dn1}
832: \end{equation}
833: %
834: and the correlator
835: %
836: \begin{equation}
837:   \frac{\partial}{\partial t} \Gamma_k = 
838:        - 2 D k^2 \Gamma_k 
839:        - \lambda [ n^2 + \Phi]
840:        - 4 \lambda n \Gamma_k 
841:   \label{dtG1} 
842: \end{equation}
843: %
844: Letter $A$ has been dropped on $n_A$ and $\Gamma_k^A$ to simplify
845: notation, and likewise $n_{0,A}$ is shortened to $n_0$. $\Phi(t)$ is
846: implicitly defined by correlators,
847: %
848: \begin{equation} 
849:   \Phi(t)=\frac{1}{V}\sum_{k\ne 0}\Gamma_k(t)
850:   \label{Phi}
851: \end{equation}
852: %
853: Thus equations above are meant to describe the model where only one
854: type of species, A, jumps on the lattice and particles have a chance
855: to react only when at the same lattice site. The equations above will
856: be solved in the following two subsections using WBGA-I and WBGA-II
857: approaches.
858: 
859: 
860: 
861: \subsection{The WBGA-I approximation}
862: \label{sec:AA+WBGAI}
863: 
864: In the WBGA-I approximation, when term proportional to $n \Gamma_k$ is
865: dropped, Eq.~(\ref{dtG1}) can be studied analytically.  Uncorrelated
866: (Poisson-like) initial condition is described by
867: $\Gamma_k=\delta_{k,0}Vn_0^2$ and solution of (\ref{dtG1}), with $k\ne
868: 0$, reads
869: %
870: \begin{equation}
871:   \Gamma_k(t) = - \lambda \int_0^t dt' e^{-2Dk^2(t-t')}[n(t')^2 + \Phi(t')]
872:   \label{G1k}
873: \end{equation}
874: %
875: Please note that the $\Gamma_0(t)$ is determined from
876: $\Gamma_0(t)=Vn(t)^2$ (thermodynamic limit) and not from (\ref{G1k}).
877: Summation of Eq.~(\ref{G1k}) over $k\ne 0$ and division by $V$
878: gives
879: %
880: \begin{equation}
881:   \Phi(t) = -\lambda \int_0^t dt' G(t-t') [ n(t')^2 + \Phi(t') ]
882:   \label{PhiIntPhi}
883: \end{equation}
884: %
885: where
886: %
887: \begin{equation}
888:   G(t-t') \equiv \frac{1}{V}\sum_{k\ne 0}e^{-2k^2D(t-t')}
889:   \label{Gtt}
890: \end{equation}
891: % 
892: was introduced.  As shown in the appendix \ref{app:Gtt}, for large
893: lattice size when $V\rightarrow\infty$, expression above can be
894: approximated as
895: %
896: \begin{equation}
897:   G(t-t') \approx 
898:     \left[
899:        8\pi D (t-t' + \eta )
900:     \right]^{-d/2}
901:   \label{Gtt3}
902: \end{equation}
903: %
904: where $\eta = \frac{1}{8\pi D}$.
905: 
906: 
907: Equations (\ref{dn1}), (\ref{PhiIntPhi}) and (\ref{Gtt3}) completely
908: specify $n(t)$. It is not possible to solve them analytically,
909: however, large time behavior of $n(t)$ can be extracted. To do this
910: we introduce $\varphi\equiv n^2+\Phi$ and rewrite Eqs.~(\ref{dn1}) and
911: (\ref{PhiIntPhi}) as
912: %
913: \begin{eqnarray}
914:   & & \frac{\partial n}{\partial t} = - \lambda \varphi \label{dn3} \\
915:   & & \varphi(t) = n(t)^2 - \lambda 
916:       \int_0^t dt' G(t-t') \varphi(t') \label{varphi}
917: \end{eqnarray}
918: %
919: which completely specify $n(t)$.
920: 
921: By using Laplace transform it is possible to transform equations
922: (\ref{dn3}) and (\ref{varphi}) into a single equation.  Laplace
923: transform is defined as
924: %
925: \begin{equation}
926:   X(s)=\int_0^\infty dt e^{-st}X(t)
927: \end{equation}
928: %
929: For $X=n,\varphi$ same symbol will be used for Laplace transform as
930: for the original function. The only exception to the rule are two
931: cases. For $X(t)=n(t)^2$, $X(s)=n_2(s)$, while for $X(t)=G(t)$,
932: $X(s)=g(s)$. 
933: 
934: Taking Laplace transform of Eq.~(\ref{varphi}) one gets $\varphi(s) =
935: n_2(s) - \lambda g(s) \varphi(s)$, and combining it with $\varphi(s) =
936: (sn(s)-n_0)/\lambda$ from (\ref{dn3}) gives
937: %
938: \begin{equation}
939:   n_2(s) = \left[ g(s)+\frac{1}{\lambda} \right] [ n_0 - s n(s) ]
940:   \label{ns1}
941: \end{equation}
942: %
943: The $g(s)$ is the Laplace transform of $G(t)$,
944: %
945: \begin{equation}
946:   g(s) = (8\pi D)^{-d/2} e^{\eta s} s^{d/2-1} \Gamma(1-d/2,\eta s)
947: \end{equation}
948: %
949: $\Gamma(\beta,x)$ denotes incomplete Gamma function;
950: %
951: \begin{equation}
952:   \Gamma(\beta,x) = \int_x^\infty du\, u^{-1+\beta} e^{-u}
953: \end{equation}
954: %
955: The analytic continuation of $\Gamma(\beta,x)$ is possible. For
956: non-integer $\beta$ and $\beta=0$, $\Gamma(\beta,z)$ is
957: multiple-valued function of $z$ with a branch point at $z=0$, and has
958: no poles. Dividing by $g(s) + 1/\lambda$ Eq.~(\ref{ns1}) results in
959: %
960: \begin{equation}
961:   s n(s) - n_0 = - \lambda_{\rm eff}(s) n_2(s) 
962:   \label{ns}
963: \end{equation}
964: %
965: where $\lambda_{\rm eff}(s)$ denotes Laplace transform of the
966: effective reaction rate,
967: %
968: \begin{equation}
969:   \lambda_{\rm eff}(s) = \frac{\lambda}{1+\lambda g(s)}
970:   \label{lambda_s}
971: \end{equation}
972: %
973: Finally, taking inverse Laplace transform of (\ref{ns}) gives
974: %
975: \begin{equation}
976:   \frac{\partial n}{\partial t} = - \int_0^t dt' 
977:      \lambda_{\rm eff}(t-t') n(t')^2
978:   \label{dtc}
979: \end{equation}
980: %
981: 
982: Equations (\ref{dn3})-(\ref{varphi}) and Eq.~(\ref{dtc}) are fully
983: equivalent.  Both equations have been dealt with before, however, in
984: a very different context. Eq.~(\ref{dtc}) was obtained in
985: ref.~\cite{Lee} through a diagrammatic technique and referred to as the
986: {\em dressed-tree} calculation. Thus in here we have shown that WBGA-I
987: is equivalent to the dressed tree
988: calculation. Eqs.~(\ref{dn3})-(\ref{varphi}) have been also obtained in
989: the study of the entirely different A+B model in ref.~\cite{rev6}.
990: 
991: Studies \cite{Lee} and \cite{rev6} suggest contradictory result:
992: ref.~\cite{Lee} argues that the dressed tree calculation gives $d/2$
993: decay exponent for particle density, while ref.~\cite{rev6} argues for
994: the $d/4$ decay exponent.  In ref.~\cite{Lee} it was incorrectly
995: concluded that dressed-tree calculation results in the $d/2$ exponent,
996: which basically came from balancing wrong terms in Laplace transformed
997: version of Eq.~(\ref{dtc}). In here calculation done in
998: ref.~\cite{Lee} will be repeated to show how to balance terms
999: correctly. Also, calculation will justify approximations employed in
1000: ref.~\cite{rev6} more rigorously as the method of calculation employs
1001: Laplace Transform and well know Tauberian theorems which relate small
1002: $s$ with large $t$ behavior. In this way all the approximations are
1003: controlled.
1004: 
1005: To extract asymptotic behavior for $n(t)$ from Eq.~(\ref{dtc}) one
1006: assumes that at large times density decays as
1007: %
1008: \begin{equation}
1009:   n(t) \approx {\cal A} (\mu + t)^{-\alpha}
1010:   \label{nmu}
1011: \end{equation}
1012: %
1013: $\cal A$ and $\alpha$ denote amplitude and exponent of decay to be
1014: found.  $\mu$ is introduced as regulator for small $t$ so that Laplace
1015: transform of $n(t)$ and $n^2(t)$ exist;
1016: %
1017: \begin{eqnarray}
1018:   & & n(s) = {\cal A} e^{s\mu} s^{\alpha-1} \Gamma(1-\alpha,\mu s) 
1019:       \label{n1s} \\
1020:   & & n_2(s) = {\cal A}^2 e^{s\mu} s^{2\alpha-1} \Gamma(1-2\alpha,\mu s)
1021:       \label{n2s}
1022: \end{eqnarray}
1023: % 
1024: and please note that $n(s)^2\ne n_2(s)$. To extract asymptotics one
1025: inserts (\ref{n1s})-(\ref{n2s}) and (\ref{lambda_s}) into (\ref{ns}) and
1026: expands in small $s$ (to extract leading order behavior for large
1027: $t$) and matches the most dominant terms.
1028: 
1029: The expansion of $g(s)$ for small $s$ is given by 
1030: %
1031: \begin{eqnarray}
1032:  g(s)  & = & (8\pi D)^{-d/2} e^{s\eta} \times \nonumber \\
1033:  & & \times \left[ \Gamma(1-d/2) s^{d/2-1} + 
1034:             \frac{2\eta^{1-d/2}}{d-2} + {\cal O}(s) \right] 
1035:  \label{gsa}
1036: \end{eqnarray}
1037: %
1038: for $d\ne 2,4,6,\ldots$.  For $d=2$ one has
1039: %
1040: \begin{eqnarray}
1041:  g(s)  & = & (8\pi D)^{-1} e^{s\eta} 
1042:              \left[ -\gamma_E - \ln ( \eta s ) + {\cal O}(\eta s) \right] 
1043:  \label{gsb}
1044: \end{eqnarray}
1045: %
1046: where $\gamma_E$ is Euler constant.  Please note that the behavior of
1047: $g(s)$ for small $s$ is qualitatively different for $d<2$ and $d>2$
1048: which has to do with recurrence of random walks bellow and above
1049: $d=2$. For small $s$ and $d<2$ $g(s)\propto s^{d/2-1}$ while for $d>2$
1050: $g(s)=const$. At $d=2$ there is logarithmic dependence on $s$. The
1051: term $e^{s\eta}$ can be neglected if leading order behavior for small
1052: s (large $t$) is sought for. 
1053: 
1054: 
1055: At the moment we focus on the $d<2$ case.  Inserting approximate
1056: formulas above for $g(s)$ into (\ref{lambda_s}) gives
1057: %
1058: \begin{equation}
1059:   \lambda_{\rm eff}(s) \sim 
1060:           \frac{(8\pi D)^{d/2}}{\Gamma(1-d/2)} s^{-d/2+1} \ , \ \ d<2 
1061:   \label{lambda_s1}
1062: \end{equation}
1063: %
1064: Since value for $\alpha$ is not known one has to separate various
1065: cases: expansion for $n(s)$ reads
1066: %
1067: \begin{equation}
1068:  n(s) = {\cal A} \left\{ 
1069:     \begin{array}{ll}
1070:       \left[ \Gamma(1-\alpha) s^{\alpha-1} 
1071:              + {\cal O}(1) \right] & \alpha<1 \\ [5pt]
1072:        \left[ \frac{\mu^{1-\alpha}}{\alpha-1} 
1073:              + {\cal O}(s^{\alpha-1}) \right] & \alpha > 1
1074:     \end{array}
1075:     \right.
1076:   \label{n1s_app}
1077: \end{equation}
1078: %
1079: and likewise for $n_2(s)$
1080: %
1081: \begin{equation}
1082:  n_2(s) = {\cal A}^2 \left\{ 
1083:     \begin{array}{ll}
1084:       \left[ \Gamma(1-2\alpha) s^{2\alpha-1} 
1085:              + {\cal O}(1) \right] & 2\alpha<1 \\ [5pt]
1086:       \left[ \frac{\mu^{1-2\alpha}}{2\alpha-1} 
1087:              + {\cal O}(s^{2\alpha-1}) \right] & 2\alpha>1
1088:     \end{array}
1089:     \right.
1090:   \label{n2s_app}
1091: \end{equation}
1092: %
1093: Inserting small $s$ expansions (\ref{lambda_s1})-(\ref{n2s_app}) into
1094: (\ref{ns}) gives
1095: %
1096: \begin{eqnarray}
1097:   & & {\cal A} [s^\alpha \Gamma(1-\alpha) + {\cal O}(s)] - n_0 =  \nonumber \\
1098:   & & - {\cal A}^2 \frac{(8\pi D)^{d/2}}{\Gamma(1-d/2)}  
1099:       \left[ s^{2\alpha-d/2} \Gamma(1-2\alpha) + {\cal O}(s^{1-d/2}) \right] 
1100:   \label{match}
1101: \end{eqnarray}
1102: %
1103: Also, please note that there are two different forms to use for $n(s)$
1104: and $n_2(s)$ in (\ref{n1s_app}) and (\ref{n2s_app}) and the ones used
1105: in (\ref{match}) were for $\alpha<1$ and $2\alpha<1$ respectively
1106: (same choice was made in ref.~\cite{Lee}). Once $\alpha$ is found, one
1107: has to check these conditions on $\alpha$ for self consistency.  There
1108: are two ways to match the terms in (\ref{match}), (a) as in the
1109: ref.~\cite{Lee}, and (b) in a way related to the work in
1110: ref.~\cite{rev6} . We begin with first case.
1111: 
1112: Balancing $s^\alpha$ term on the left hand side of (\ref{match}) with
1113: $s^{2\alpha-d/2}$ on the right hand side, gives $\alpha=d/2$ and 
1114: %
1115: \begin{equation}
1116:   {\cal A}_a=-\frac{1}{\pi}{\rm sin}(\pi d)
1117:               \Gamma(d)\Gamma(1-d/2)^2(8\pi D)^{-d/2}
1118: \end{equation}
1119: %
1120: Also from $\alpha<1$ and $2\alpha<1$ one has constraint that
1121: $d<1$. However, for $d<1$ the term ${\rm sin}(\pi d)$ is positive
1122: which makes amplitude ${\cal A}_a$ negative. Thus all physical
1123: conditions can not be met with this type of matching. In
1124: ref.~\cite{Lee} the condition $d<1$ [coming from the fact that first
1125: row is used in (\ref{n2s_app})] was overlooked (if $d>2$ is allowed
1126: amplitude ${\cal A}_a$ is perfectly acceptable).
1127: 
1128: The $\alpha=d/2$ scenario can still turn out to be true. With this
1129: choice of $\alpha$ and $d<2$ condition coming from (\ref{lambda_s1})
1130: the second row in (\ref{n2s_app}) have to be used. Again, carrying out
1131: similar type of matching procedure would give negative
1132: amplitude. Finally, the $\alpha=d/2$ avenue has to be given up.
1133: 
1134: At this stage one is left by the second (b) way of balancing, {\em i.e.}
1135: matching constant $n_0$ term on the left hand side of (\ref{match})
1136: with $s^{2\alpha-d/2}$ on the right hand side. (The remaining terms,
1137: {\em e.g.} such as the $s^\alpha$ on the left hand side can be
1138: balanced by considering sub-leading corrections to $n(s)$). This way of
1139: balancing immediately gives $\alpha=d/4$ and
1140: %
1141: \begin{equation}
1142:   {\cal A}_b=\sqrt{n_0}(8\pi D)^{-d/4}
1143:   \label{Ab}
1144: \end{equation}
1145: %
1146: with constraints that $d<2$ (Eq.~\ref{lambda_s1} was used
1147: to get Eq.~\ref{match}).
1148: 
1149: Matching the constant term $n_0$ on the left hand side of
1150: (\ref{match}) is rather counter-intuitive since in the framework of
1151: Laplace transform constant can normally be disregarded when large $t$
1152: behavior is sought for. To see how this comes about it is useful to
1153: turn back to Eq.~(\ref{ns}).
1154: 
1155: Equation~(\ref{ns}) comes from Eq.~(\ref{ns1}). For simplicity reasons
1156: we focus on the case $\lambda=\infty$ in (\ref{ns1}). It is clear
1157: that at the right hand side of equation (\ref{ns1}) the $s n(s)$ term
1158: is sub-leading to $n_0$. (True enough, $n_0$ is constant but it is
1159: multiplied by $g(s)$.) Thus $n_2(s)$ indeed has to be matched with
1160: $g(s) n_0$. This procedure results in amplitude ${\cal A}_b$ obtained
1161: previously. Also, by using form (\ref{ns1}), one can show that
1162: amplitude ${\cal A}_b$ as given in Eq.~(\ref{Ab}) is valid even for
1163: $d>2$.  Analysis can be repeated with finite value of $\lambda$ with
1164: the same outcome.  Before proceeding, it is important to mentioned
1165: that procedure outlined above does not work at $d\ge 4$ and it has to
1166: be modified.
1167: 
1168: The main findings of this subsection are twofold. First, it has been
1169: shown that dressed tree calculation is equivalent to WBGA-I. Second,
1170: it was shown that WBGA-I (and dressed tree calculation) fail to
1171: describe A+A reaction predicting $d/4$ decay exponent;
1172: %
1173: \begin{equation}
1174:   n(t) \sim \sqrt{n_0} (8\pi Dt)^{-d/4}
1175:   \label{ntasym}
1176: \end{equation}
1177: %
1178: 
1179: All findings of this subsection are summarized in Fig.~1. The
1180: numerical treatment of Eqs.~(\ref{dn3}) and (\ref{varphi}) confirms
1181: asymptotic decay given in Eq.~(\ref{ntasym}). Equations (\ref{dn3})
1182: and (\ref{varphi}) were solved previously numerically in
1183: ref.~\cite{rev6}. In here, the features of the decay curves are
1184: somewhat different from the ones obtained in ref.~\cite{rev6}. For
1185: example, curves shown in this work have concave form (bend upward),
1186: while curves in fig. 1 of ref.~\cite{rev6} are convex (bend downward)
1187: as if asymptotics have not yet been reached. Also, in here, there is
1188: no intersection of curves, which can be found in
1189: ref.~\cite{rev6}. These difference could come from the numerical
1190: treatment. Apart from using double precision, to obtain curves in
1191: Fig. 1 the method of integration was used which integrates exactly
1192: $\int_0^t dt'(t-t')^{-d/2}f(t')$ provided $f(t)$ is piecewise linear
1193: in $t$. The more detailed description of numerical treatment is shown
1194: in the appendix \ref{app:kernal}
1195: 
1196: In the next subsection it will be shown that, in the case of A+A
1197: reaction, weaknesses of WBGA-I method extend to WBGA-II level.
1198: 
1199: 
1200: 
1201: \subsection{The WBGA-II approximation}
1202: \label{sec:AA+WBGAII}
1203: 
1204: When term $n\Gamma_k$ is kept in Eq.~(\ref{dtG1}), equivalent of
1205: Eq.~(\ref{varphi}) reads
1206: %
1207: \begin{equation}
1208:   \varphi(t) = n(t)^2 - \lambda \int_0^t  dt' I(t,t') \varphi(t')
1209:   \label{varphi1}
1210: \end{equation}
1211: %
1212: while Eq.~(\ref{dn3}) stays the same. The $I(t,t')$ is given by
1213: %
1214: \begin{equation}
1215:   I(t,t') = G(t-t')\ {\rm exp} 
1216:      \left[ -4\lambda \int_{t'}^t dt'' n(t'') \right] 
1217:   \label{Itt}
1218: \end{equation}
1219: %
1220: The asymptotics of Eq.~(\ref{varphi1}) can not be extracted by Laplace
1221: transform, and it is more convenient to use approach of
1222: ref.~\cite{rev6}. For large $t$, Eq.~(\ref{varphi1}) can be
1223: approximated by
1224: %
1225: \begin{equation}
1226:   \varphi(t) \approx n(t)^2 - I(t,0) {\cal I}(t)
1227:   \label{varphiapprox}
1228: \end{equation}
1229: %
1230: where ${\cal I}(t)\equiv \lambda \int_0^t dt' \varphi(t')$.  This step
1231: is valid provided two conditions are satisfied. First, the term
1232: $I(t,t')$ has to vanish as time difference $t-t'$ grows. Second, the
1233: integral $\int_0^\infty dt \varphi(t)$ has to be finite. Using
1234: (\ref{dn3}) one gets ${\cal I}(t)=n_0 - n(t) \approx n_0$ and
1235: (\ref{varphiapprox}) becomes
1236: %
1237: \begin{equation}
1238:   \frac{\partial n}{\partial t} \approx -\lambda n^2 + \lambda I(t,0) n_0
1239:   \label{dndtapprox}
1240: \end{equation}
1241: %
1242: Equation above is solved with the assumption that asymptotically
1243: $n(t)\sim {\cal A}/t$, which is checked self consistently at the
1244: end. Using postulated asymptotics for $t$, one can see from
1245: Eq.~(\ref{Itt}) that $I(t,0)\sim {\rm const}\ t^{-(d/2+4\lambda{\cal
1246: A})}$. Assuming that
1247: %
1248: \begin{equation}
1249:   \frac{I(t,0)}{n(t)^2}\rightarrow 0  \ , \ \ t\rightarrow\infty
1250:   \label{Iasym}
1251: \end{equation}
1252: %
1253: one can solve (\ref{dndtapprox}) in form $\partial n/\partial
1254: t=-\lambda n^2$, and get ${\cal A}=1/\lambda$. Assumption
1255: (\ref{Iasym}) is correct provided $2< d/2+4\lambda{\cal A}=d/2+4$,
1256: which is true for any $d$. This shows that $n(t)\approx 1/(\lambda t)$
1257: is asymptotic form for the solution of Eqs.~(\ref{dn3}) and
1258: (\ref{varphi1}). This means that the last term ($n\Gamma_k$) in
1259: (\ref{dtG1}) only influences intermediate behavior when $t$ is not too
1260: large. For large $t$, WBGA-II gives exactly the same asymptotics as the
1261: pure mean field treatment.
1262: 
1263: Thus main finding so far is that both WBGA-I and WBGA-II fail to
1264: describe A+A reaction.  This is somewhat surprising as even simplest
1265: pair-approach, e.g. Smoluchowskii method, describes exponent of A+A
1266: correctly.  Clearly A+A reaction can not be viewed as a weakly
1267: interacting Bose-gas. The question is what is the minimum modification
1268: of WBGA which will provide correct result for the A+A model? This
1269: question will be answered in the next subsection.
1270: 
1271: 
1272: 
1273: 
1274: \subsection{The hybrid of WBGA and Kirkwood superposition approximation}
1275: \label{sec:AA+hybrid}
1276: 
1277: To see how to improve WBGA one has to clarify what went wrong in the
1278: first place.  We start from problematic equation (\ref{dtGa}) which
1279: becomes (\ref{dtG1}) when terms with $\sigma^{AB}_k$ are set to
1280: zero. To trace why WBGA fails it is useful to rewrite Eq.~(\ref{dtGa})
1281: as it looks one step before WBGA is made, and we keep only terms
1282: describing A+A reaction:
1283: %
1284: \begin{eqnarray}
1285:   \frac{\partial}{\partial t} \Gamma_k & = &
1286:        - 2 D k^2 \Gamma_k   
1287:        - \left[  
1288:            \sigma_k n^2 
1289:            + \frac{1}{V} \sum_{q\ne k} \sigma_q \Gamma_{k-q} 
1290:          \right] - \dot\Gamma^{(3)}
1291:    \label{dtGa1}
1292: \end{eqnarray}
1293: %
1294: where
1295: %
1296: \begin{equation}
1297:    \dot\Gamma^{(3)} = \frac{1}{\sqrt{V}}
1298:           \sum_q \sigma_q ( \langle a_{-k} a_{k-q} a_q \rangle 
1299:                           + \langle a_{k} a_{-k-q} a_q \rangle )
1300:    \label{dotGam}
1301: \end{equation}
1302: %
1303: is focus of present subsection.  In technical terms, the usage of
1304: WBGA can be translated into approximating three point density $\langle
1305: a_x a_y a_z \rangle$ in a particular way. In the case of WBGA-I one
1306: simply takes
1307: %
1308: \begin{equation}
1309:   \langle a_{k_1} a_{k_2} a_{k_3} \rangle=
1310:   \langle a_x a_y a_z \rangle= 0
1311:   \label{a3i}
1312: \end{equation}
1313: %
1314: while in WBGA-II one assumes
1315: %
1316: \begin{eqnarray}
1317:   \langle a_{k_1} a_{k_2} a_{k_3} \rangle & \approx &  
1318:        \delta_{k_1,0} \delta_{k_2,0} \delta_{k_3,0} a_0^3 + 
1319:        \nonumber \\
1320:    & & + \delta_{k_1,0} \bar\delta_{k_2,0} \bar\delta_{k_3,0}
1321:          a_0 \langle a_{k_2} a_{k_3} \rangle +   
1322:        \nonumber \\
1323:    & & + \bar\delta_{k_1,0} \delta_{k_2,0} \bar\delta_{k_3,0} 
1324:          a_0 \langle a_{k_1} a_{k_3} \rangle +
1325:         \nonumber \\
1326:    & & + \bar\delta_{k_1,0} \bar\delta_{k_2,0} \delta_{k_3,0} 
1327:          a_0 \langle a_{k_1} a_{k_2} \rangle
1328:    \label{a3iik} 
1329: \end{eqnarray}
1330: %
1331: where $\bar\delta_{k,0}\equiv 1-\delta_{k,0}$. Inserting (\ref{a3iik})
1332: into (\ref{dtGa1}) gives the terms describing A+A process in
1333: (\ref{dtGa}). It is useful to transform approximation above into the
1334: x-space to understand nature of approximation better. Inverse Fourier
1335: transform of (\ref{a3iik}) gives~\cite{foot4}
1336: %
1337: \begin{eqnarray}
1338:   \langle a_x a_y a_z \rangle & \approx &  
1339:     n^3 + n ( \langle a_x a_y \rangle - n^2 ) + \nonumber \\
1340:     & & + n ( \langle a_x a_z \rangle - n^2 ) 
1341:         + n ( \langle a_y a_z \rangle - n^2 )
1342:   \label{a3iix} 
1343: \end{eqnarray}
1344: %
1345: The WBGA approximations given in Eqs.~(\ref{a3i}) and (\ref{a3iix})
1346: can be contrasted to the Kirkwood superposition approximation,
1347: %
1348: \begin{equation}
1349:   \langle a_x a_y a_z \rangle \approx \frac{1}{n^3} 
1350:     \langle a_x a_y \rangle \langle a_x a_z \rangle 
1351:     \langle a_y a_z \rangle
1352:     \label{kirkwood}
1353: \end{equation}
1354: %
1355: which is known to describe A+A problem correctly (at least the decay
1356: exponent). 
1357: 
1358: It is interesting to note that instead of (\ref{kirkwood}) one could
1359: carry out Kirkwood superposition approximation in a different way,
1360: e.g. as $\langle n_x n_y n_z \rangle \approx \langle n_x n_y \rangle
1361: \langle n_x n_z \rangle \langle n_y n_z\rangle/n^3$. (There is no such
1362: ambiguity when only single occupancy of site is allowed.)  
1363: When multiple site occupancy is allowed, as considered here, we argue
1364: that Eq.~(\ref{kirkwood}) is more accurate way to implement Kirkwood
1365: superposition approximation.
1366: 
1367: 
1368: By looking at equation (\ref{a3iix}) it is possible to understand WBGA
1369: better. Eq.~(\ref{a3iix}) suggests that WBGA is somewhat equivalent to
1370: the additive expansion of correlation functions. It has been argued
1371: that such additive approximation is inferior to the Kirkwood
1372: superposition approximation~\cite{rev4,rev5}. This in turn explains
1373: why WBGA fails to describe the A+A reaction.
1374: 
1375: Clearly, to correctly describe A+A reaction one has to use Kirkwood
1376: superposition approximation, which we modify further in the spirit of
1377: WBGA, assuming that higher order products of annihilation operators
1378: $a_k,b_k$ with $k\ne 0$ are small, and accounting for thermodynamic
1379: limit (extracting $a_0$ or $b_0$ out of field theoretic
1380: averages). This is done in two stages. First, Kirkwood superposition
1381: approximation in Eq.~(\ref{kirkwood}) is rephrased in the k-space
1382: which gives~\cite{foot5}
1383: %
1384: \begin{equation}
1385:   \langle a_{k_1} a_{k_2} a_{k_3} \rangle \approx 
1386:      \frac{\delta(k_1+k_2+k_3)}{n^3V^{3/2}} 
1387:      \sum_l \Gamma_l \Gamma_{k_1-l} \Gamma_{k_2+l}
1388:   \label{kirkwoodk}
1389: \end{equation}
1390: %
1391: To get the improved form for three body terms one inserts
1392: (\ref{kirkwoodk}) into (\ref{dotGam}) which gives
1393: %
1394: \begin{equation}
1395:   \dot\Gamma^{(3)} \approx \frac{2}{n^3V^2} 
1396:      \sum_{q,l} \sigma_q \Gamma_l \Gamma_{k+l} \Gamma_{k-q+l} 
1397:   \label{dotGam1}
1398: \end{equation}
1399: %
1400: The expression above was obtained by using symmetry properties of
1401: $\Gamma_k=\Gamma_{-k}$ and $\sigma_q=\sigma_{-q}$. Also upon inserting
1402: (\ref{kirkwoodk}) into (\ref{dotGam}) the two terms on the right hand side
1403: of (\ref{dotGam}) contribute equally resulting in factor two in
1404: (\ref{dotGam1}).  Equation above is too complicated to be used in
1405: practise and we approximate it further. 
1406: 
1407: In the spirit of WBGA the terms in (\ref{kirkwoodk}) which contain
1408: large number of correlation functions with $k$-vector different from
1409: zero are neglected. This is done in two stages, first sum over $q$ is
1410: split into $q=k$ and $q\ne k$ parts, and then for each of sums various
1411: contributions from sum over $l$ are distilled to extract
1412: non-fluctuating $k=0$ operators. This gives
1413: %
1414: \begin{eqnarray}
1415:   & & \dot\Gamma^{(3)} \approx 
1416:         \frac{2}{n^3V^2} \left[  \sigma_k ( \Gamma_0^2 \Gamma_k 
1417:      + \Gamma_0 \Gamma_k^2) + \right. \nonumber \\
1418:   & & \left. + \sum_{q\ne k} \sigma_q \Gamma_0 ( \Gamma_k \Gamma_q  
1419:        + \Gamma_k \Gamma_{k-q} + \Gamma_q \Gamma_{q-k} ) \right] 
1420:   \label{dotGam2} 
1421: \end{eqnarray}
1422: %
1423: where terms of the type $\Gamma_k\Gamma_q\Gamma_l$ with $k,q,l\ne 0$
1424: have been neglected.  Eq.~(\ref{dotGam2}) is still complicated to be
1425: used, at least to obtain analytic result. First term on the right hand
1426: side of (\ref{dotGam2}) can be neglected since it leads to mean field
1427: behavior (as shown in section \ref{sec:AA+WBGAII}). Second term can be
1428: absorbed into the one of the three terms under the sum sign,
1429: .e.g. under the first term. Second and third terms under the sum sign
1430: couple correlation functions in a non-trivial way and are neglected in
1431: the following for simplicity reasons. With $\Gamma_0\approx n^2V$, and
1432: applying recipe just described gives
1433: %
1434: \begin{equation}
1435:   \dot\Gamma^{(3)} \approx \frac{2}{nV} \Gamma_k \sum_{q} \sigma_q \Gamma_q  
1436:   \label{dotGam3} 
1437: \end{equation}
1438: %
1439: which is midway between WBGA-II (Eq.~\ref{a3iik}) and Kirkwood
1440: superposition approximation (Eq.~\ref{kirkwoodk}). It is interesting
1441: to note that equation above is very close to the shortened Kirkwood
1442: superposition approximation discussed in refs.~\cite{rev4,rev5}. Using
1443: approximation above to decouple three body density, and particular
1444: form for $\sigma^{AA}_{xy}$ used throughout this section, gives following
1445: equations of motion,
1446: %
1447: \begin{equation}
1448:   \frac{\partial}{\partial t} \Gamma_k = 
1449:        - 2 D k^2 \Gamma_k 
1450:        - \lambda ( n^2 + \Phi )
1451:        - 2 \lambda \Gamma_k \frac{ \Phi + n^2 }{n} 
1452:   \label{dtG2} 
1453: \end{equation}
1454: %
1455: which should be contrasted with Eq.~(\ref{dtG1}). The most convenient
1456: way to solve (\ref{dn1}) and (\ref{dtG2}) is to introduce $\chi_k$ as
1457: $\Gamma_k=n^2\chi_k$ and $n^2+\Phi = n^2\chi$ where
1458: %
1459: \begin{equation}
1460:   \chi\equiv 1+\frac{1}{V}\sum_{k\ne 0}\chi_k
1461: \end{equation}
1462: %
1463: $\chi_k$ with $k=0$ is set equal to $V$ and does not change in time
1464: (thermodynamic limit). Applying change of variables just described
1465: modifies Eq.~(\ref{dn1}) into
1466: %
1467: \begin{equation}
1468:   \frac{\partial}{\partial t} n(t) = - \kappa(t) n(t)^2  
1469:   \label{dn2} 
1470: \end{equation}
1471: %
1472: where effective reaction rate 
1473: %
1474: \begin{equation}
1475:   \kappa(t)=\lambda\chi(t)
1476: \end{equation}
1477: %
1478: was introduced. Same change of variables transforms Eq.~(\ref{dtG2})
1479: into
1480: %
1481: \begin{equation}
1482:   \frac{\partial}{\partial t} \chi_k = 
1483:        - 2 D k^2 \chi_k - \lambda \chi
1484:   \label{chi}
1485: \end{equation}
1486: %
1487: Equation (\ref{chi}) can be solved for all $\chi_k$ and $k\ne 0$
1488: (pretending that $\chi(t)$ is known), and after summing over $k\ne 0$
1489: one gets following integral equation
1490: %
1491: \begin{equation}
1492:   \chi(t) = 1 - \lambda \int_0^t dt' G(t-t') \chi(t')
1493: \end{equation}
1494: %
1495: Solution of the equation above can be found by Laplace transform which gives
1496: %
1497: \begin{equation}
1498:   \kappa(s) = \frac{\lambda}{s[1+\lambda g(s)]}
1499:   \label{kappachis}
1500: \end{equation}
1501: %
1502: Also, one can integrate Eq.~(\ref{dn2}) which gives
1503: %
1504: \begin{equation}
1505:   n(t) = \frac{n_0}{1+ n_0 \bar\kappa(t) } \ , \ \ 
1506:   \bar\kappa(t) = \int_0^t dt' \kappa(t') 
1507:   \label{nt}
1508: \end{equation} 
1509: %
1510: It is not possible to obtain closed expression for $\kappa(t)$ and
1511: $n(t)$.  However, asymptotic form of $n(t)$ can be extracted.
1512: 
1513: Inserting small $s$ expansion of $g(s)$ (see Eqs.~\ref{gsa} and
1514: \ref{gsb}) into (\ref{kappachis}) gives
1515: %
1516: \begin{equation}
1517:   \kappa(s) \sim 
1518:      \left\{ 
1519:        \begin{array}{ll}
1520:           \frac{(8\pi D)^{d/2}}{\Gamma(1-d/2)} s^{-d/2} & d<2 \\ [5pt]
1521:           - \frac{8\pi D}{s[\gamma_E+\ln(\eta s)]} & d=2 \\ [5pt]
1522:           \frac{\lambda}{1+\lambda g(0) }s^{-1} & d>2
1523:        \end{array}
1524:      \right.
1525:   \label{kappas}
1526: \end{equation}
1527: %
1528: for $s\rightarrow 0$. Taking inverse Laplace transform of equation
1529: above gives leading order behavior for effective reaction rate constant,
1530: %
1531: \begin{equation}
1532:   \kappa(t)\sim 
1533:     \left\{ 
1534:       \begin{array}{ll}
1535:          \frac{(8\pi D)^{d/2}}{\Gamma(1-d/2)\Gamma(d/2)} 
1536:              t^{d/2-1} & d<2 \\ [5pt]
1537:           \frac{8\pi D}{\ln t/\eta} & d=2 \\ [5pt]
1538:          \frac{\lambda}{1+\lambda g(0)} & d>2 
1539:       \end{array}
1540:     \right.
1541:   \label{kappat}
1542: \end{equation}
1543: %
1544: when $t\rightarrow\infty$.  To find inverse Laplace transform of
1545: $\kappa(s)$ for $d=2$ (second line Eq.~\ref{kappas}) is somewhat
1546: involved, please see appendix \ref{app:kappa} for details.  Finally,
1547: inserting (\ref{kappat}) into (\ref{nt}) gives following asymptotics
1548: %
1549: \begin{equation}
1550:   n(t) \sim \left\{ \begin{array}{ll}
1551:               \Gamma(1-d/2) \Gamma(1+d/2) (8\pi Dt)^{-d/2} & d<2 \\ [5pt]
1552:               \frac{\ln 8\pi Dt}{8\pi Dt} & d = 2 \\ [5pt]
1553:               \left[ \frac{1}{\lambda} + g(0) \right] t^{-1} & d>2
1554:             \end{array} \right.
1555: \end{equation}
1556: %
1557: where $g(0)$ entering in the third row can easily be found from
1558: (\ref{gsa}).
1559: 
1560: 
1561: 
1562: 
1563: 
1564: \section{WBGA applied to the A+B reaction}
1565: \label{sec:AB+WBGA}
1566: 
1567: Equations of motion for density and correlation functions describing
1568: A+B model result from (\ref{dnAdt})-(\ref{dnBdt}) and
1569: (\ref{dtGa})-(\ref{dtGc}) by using (\ref{sigmas}) with
1570: $\lambda=0$. For simplicity reasons, we focus on $n_A$=$n_B\equiv n$
1571: case and omit labels $A$ and $B$. Also, as in the previous section
1572: $n_0=n_{0,A}=n_{0,B}$. Applying procedure outlined above leads to the
1573: equations for particle densities
1574: %
1575: \begin{equation}
1576:   \frac{\partial n}{\partial t} = - \delta ( n^2 + \Phi_c ) 
1577:   \label{dn4}
1578: \end{equation}
1579: %
1580: where $\Phi_c$ is given by the AB correlation function,
1581: %
1582: \begin{equation}
1583:   \Phi_c(t) = \frac{1}{V} \sum_{k\ne 0}\Gamma_k^{AB}(t)
1584: \end{equation}
1585: %
1586: Equations for correlators $\Gamma_k\equiv\Gamma_k^{AA}=\Gamma_k^{BB}$
1587: and $\Gamma_k^c\equiv\Gamma_k^{AB}$ are given by
1588: %
1589: \begin{eqnarray}
1590:   & & \left( \frac{\partial}{\partial t} + 2Dk^2 \right) \Gamma_k^c =
1591:      - \delta ( n^2 + \Phi_c ) 
1592:       -2 \delta n ( \Gamma_k + \Gamma_k^c ) \\
1593:   & & \left( \frac{\partial}{\partial t} + 2Dk^2 \right) \Gamma_k = 
1594:       -2 \delta n ( \Gamma_k + \Gamma_k^c ) 
1595: \end{eqnarray}
1596: %
1597: The most convenient way to solve equations above is to diagonalise
1598: them by subtraction and addition. The final result is that
1599: correlations of AB pairs are governed by
1600: %
1601: \begin{equation}
1602:   \Phi_c(t) = - \delta \int_0^t dt' [ G(t,t') + I(t,t') ] 
1603:                  [n(t')^2+\Phi_c(t')]
1604:   \label{corAB}
1605: \end{equation}
1606: %
1607: and for AA (or BB) pairs as
1608: %
1609: \begin{equation}
1610:   \Phi(t) = \delta \int_0^t dt' [ G(t,t') - I(t,t') ] [n(t')^2+\Phi_c(t')]
1611:   \label{corAABB}
1612: \end{equation}
1613: %
1614: The $I(t,t')$ appearing in equations (\ref{corAB}) and (\ref{corAABB})
1615: has the same form as in (\ref{Itt}) with trivial change of $\lambda$
1616: into $\delta$. Same type of analysis as in the subsection
1617: \ref{sec:AA+WBGAII} leads to conclusion that approximation
1618: $G(t,t')+I(t,t')\approx G(t,t')$ can be used, which in turn leads to
1619: $d/4$ density decay exponent.  Interestingly enough, both WBGA-I and
1620: WBGA-II approaches lead to correct $d/4$ exponent when used to solve
1621: A+B model.
1622: 
1623: In some sense the WBGA approach seems to be suited rather well for A+B
1624: reaction. Quite the contrary can be said for A+A reaction as shown
1625: previously. To understand workings of WBGA on the more general model
1626: the ABBA model will be studied in the next section. 
1627: 
1628: 
1629: 
1630: 
1631: 
1632: \section{WBGA applied to the ABBA model}
1633: \label{sec:ABBA+WBGA}
1634: 
1635: 
1636: The ABBA model was suggested in refs.~\cite{KJ1,KJ2} and it has a
1637: number of interesting properties. The model is obtained as the special
1638: case of the general two species model discussed in section
1639: \ref{sec:model} where A+A and B+B reactions occur with same reaction
1640: rate $\lambda$, while A+B with reactions rate $\delta$, and
1641: $\delta>\lambda$ is taken. The fact that A and B species have equal
1642: diffusion constants is very important since it leads to survival of
1643: minority species. Minority species is the one with smaller
1644: concentration at $t=0$, and we chose B to be minority,
1645: i.e. $n_{0,A}>n_{0,B}$. By survival it is meant that particle density
1646: ratio saturates to constant $u(t)\equiv n_A(t)/n_B(t)\rightarrow{\rm
1647: const}$ as $t\rightarrow\infty$. (The mean field analysis predicts
1648: vanishing of minority species $u(t)\rightarrow\infty$ as
1649: $t\rightarrow\infty$.  Also, similar model has been studied in
1650: ref.~\cite{Howard} where it was shown that if $D_A\ne D_B$ only one
1651: type of species survives.) 
1652: 
1653: The goal of the present section is to understand strengths and
1654: weaknesses of WBGA approach by testing it on the ABBA model which
1655: combines A+A and A+B reactions.  These reactions have been studied in
1656: previous sections but now they are allowed to occur simultaneously. It
1657: is interesting to see how WBGA approach performs in such situation.
1658: Clearly, there are plenty of possibilities how to combine A+A and A+B
1659: reactions, but the way they are combined in the ABBA model gives rise
1660: to many interesting effects. (For example, there is survival of
1661: minority species, recovery of initial density ratio, dependence of
1662: amplitudes on initial densities and reaction rates~\cite{foot6},
1663: please see refs.~\cite{KJ1,KJ2} for details).
1664: 
1665: We start from equations of motion given in section \ref{sec:EOM} which
1666: describe very general two species reaction-diffusion model.
1667: Assumptions in Eq.~(\ref{sigmas}) describe content of the ABBA
1668: model. Using (\ref{sigmas}) in Eqs.~(\ref{dnAdt})-(\ref{dnBdt}) gives
1669: %
1670: \begin{eqnarray}
1671:   & & \frac{\partial}{\partial t} n_A = 
1672:         - \left( 
1673:              \lambda n_A^2 + \delta n_A n_B + 
1674:              \lambda \Phi_{AA} + \delta \Phi_{AB} 
1675:           \right) 
1676:           \label{dtnA} \\
1677:   & & \frac{\partial}{\partial t} n_B = 
1678:         - \left( 
1679:              \lambda n_B^2 + \delta n_A n_B + 
1680:              \lambda \Phi_{BB} + \delta \Phi_{AB} 
1681:           \right)
1682:           \label{dtnB} 
1683: \end{eqnarray}
1684: %
1685: where
1686: %
1687: \begin{equation}
1688:   \Phi_{\rho\nu} = \frac{1}{V} \sum_{k\ne 0} \Gamma^{\rho\nu}_k \ , \ \ 
1689:   \rho,\nu=A,B
1690: \end{equation}
1691: %
1692: Also, Eqs. for correlators (\ref{dtGa})-(\ref{dtGc}) simplify to
1693: %
1694: \begin{eqnarray}
1695:   \frac{\partial}{\partial t} \Gamma_k^{AA} 
1696:       & = & - 2 D k^2 \Gamma_k^{AA}   
1697:             - \lambda \left(  n_A^2 + \Phi_{AA} \right) - \nonumber \\
1698:       &   & - 2 ( 2 \lambda n_A + \delta n_B ) \Gamma^{AA}_k 
1699:             - 2 \delta n_A \Gamma^{AB}_k 
1700:       \label{GammaAA} \\
1701:   \frac{\partial}{\partial t} \Gamma_k^{BB} 
1702:       & = & - 2 D k^2 \Gamma_k^{BB}   
1703:             - \lambda \left(  n_B^2 + \Phi_{BB} \right) - \nonumber \\
1704:       &   & - 2 ( 2 \lambda n_B + \delta n_A ) \Gamma^{BB}_k 
1705:             - 2 \delta n_B \Gamma^{AB}_k  
1706:       \label{GammaBB} \\
1707:   \frac{\partial}{\partial t} \Gamma_k^{AB} 
1708:       & = & - 2 D k^2 \Gamma_k^{AB}   
1709:             - \delta \left(  n_A n_B + \Phi_{AB} \right) - \nonumber \\
1710:       &   & - ( 2 \lambda + \delta ) (n_A+n_B) \Gamma^{AB}_k - \nonumber \\
1711:       &   & - \delta ( n_A \Gamma^{BB}_k + n_B \Gamma^{AA}_k )
1712:       \label{GammaAB}
1713: \end{eqnarray}
1714: %
1715: The equations above will be solved in the next two subsections within
1716: WBGA-I and WBGA-II approaches.
1717: 
1718: 
1719: 
1720: 
1721: 
1722: \subsection{The WBGA-I approximation}
1723: \label{sec:ABBA+WBGAI}
1724: 
1725: In the framework of WBGA-I all seemingly ${\cal O}(n^3)$ terms of the
1726: type $n_\rho \Gamma^\nu_k$ with $\rho=A,B$ and $\nu=AA,AB,BB$ are
1727: thrown away in (\ref{GammaAA})-(\ref{GammaAB}). Following similar
1728: steps as in subsection \ref{sec:AA+WBGAI} gives
1729: %
1730: \begin{eqnarray}
1731:  & & \frac{\partial}{\partial t} n_A 
1732:        = - ( \lambda \varphi_{AA} + \delta \varphi_{AB} ) 
1733:      \label{dadt}\\
1734:  & & \frac{\partial}{\partial t} n_B 
1735:        = - ( \lambda \varphi_{BB} + \delta \varphi_{AB} )
1736:      \label{dbdt}
1737: \end{eqnarray}
1738: %
1739: and
1740: %
1741: \begin{eqnarray}
1742:   & & \varphi_{AA} = n_A^2  - \lambda \int_0^t dt' G(t-t') \varphi_{AA}(t') 
1743:       \label{dvaraa} \\
1744:   & & \varphi_{BB} = n_B^2 - \lambda \int_0^t dt' G(t-t') \varphi_{BB}(t') 
1745:       \label{dvarbb} \\
1746:   & & \varphi_{AB} = n_A n_B - \delta \int_0^t dt' G(t-t') \varphi_{AB}(t')
1747:       \label{dvarab}
1748: \end{eqnarray}
1749: %
1750: where $\varphi_{\rho\nu}\equiv n_\rho n_\nu + \Phi_{\rho\nu}$ for
1751: $\rho,\nu \in \{A,B\}$.  To solve Eqs.~(\ref{dadt})-(\ref{dvarbb}) it
1752: is possible to employs same technique as in section
1753: \ref{sec:AA+WBGAII}. Equations above can be approximated by
1754: %
1755: \begin{eqnarray}
1756:   & & 0 \approx n_A^2 - G(t,0) {\cal I}_{AA}(t)
1757:       \label{zeroaa} \\
1758:   & & 0 \approx n_B^2 - G(t,0) {\cal I}_{BB}(t)
1759:       \label{zerobb} \\
1760:   & & 0 \approx n_A n_B - G(t,0) {\cal I}_{AB}(t)
1761:       \label{zeroab}
1762: \end{eqnarray}
1763: %
1764: where
1765: %
1766: \begin{equation}
1767: {\cal I}_{\rho\nu}(t) \equiv  
1768:     ( \lambda\delta_{\rho,\nu} + \delta \bar\delta_{\rho,\nu})
1769:     \int_0^t dt' \varphi_{\rho\nu}(t')\
1770:     , \ \ \rho=A,B 
1771: \end{equation}
1772: %
1773: By integrating Eqs.~(\ref{dadt}) and (\ref{dbdt}) a useful
1774: relationship can be derived for ${\cal I}_{\rho\nu}$, $\rho,\nu=A,B$;
1775: %
1776: \begin{eqnarray}
1777:   & & {\cal I}_{AA}(t) + {\cal I}_{AB}(t) = n_{0,A}-n_A(t) \approx n_{0,A}
1778:       \label{IAA} \\ 
1779:   & & {\cal I}_{BB}(t) + {\cal I}_{AB}(t) = n_{0,B}-n_B(t) \approx n_{0,B}
1780:       \label{IBB}
1781: \end{eqnarray}
1782: %
1783: Using (\ref{zeroaa})-(\ref{zeroab}) gives $( n_A + n_B )^2 \approx
1784: G(t) ({\cal I}_{AA}+{\cal I}_{BB}+2{\cal I}_{AB})$, and adding
1785: (\ref{IAA}) and (\ref{IBB}) finally gives 
1786: %
1787: \begin{equation}
1788:   (n_A + n_B)^2 \approx G(t) ( n_{0,A}+n_{0,B} )
1789:   \label{nab+}
1790: \end{equation}
1791: %
1792: Also from (\ref{zeroaa}) and (\ref{zerobb})
1793: $n_A^2 - n_B^2 \approx G(t) ( {\cal I}_{AA}-{\cal I}_{BB})$ and
1794: subtraction of (\ref{IAA}) and (\ref{IBB}) gives 
1795: %
1796: \begin{equation}
1797:   n_A^2 - n_B^2 \approx G(t) ( n_{0,A}-n_{0,B} )
1798:   \label{nab-}
1799: \end{equation}
1800: %
1801: After solving (\ref{nab+}) and (\ref{nab-}) for $n_A$ and $n_B$ one
1802: gets
1803: %
1804: \begin{eqnarray}
1805:   n_\rho \sim \frac{n_\rho(0)}{\sqrt{n_{0,A}+n_{0,B}}} (8\pi Dt)^{-d/4} 
1806:   \ , \ \ \rho=A,B
1807: \end{eqnarray}
1808: %
1809: 
1810: According to WBGA-I both particles decay with $d/4$ exponent and
1811: amplitudes given above.  The WBGA-I predicts same decay exponent
1812: as for the pure A+A model.  As in the case of A+A reaction, the value for
1813: $d/4$ exponent obtained here is not correct.  The computer simulation
1814: and $\epsilon$-expansion analysis of this reaction suggest $d/2$
1815: exponent~\cite{KJ1,KJ2}. To see what happens when ${\cal O}(n^3)$
1816: terms are kept in (\ref{GammaAA})-(\ref{GammaAB}) we proceed with
1817: WBGA-II calculation.
1818: 
1819: 
1820: 
1821: 
1822: \subsection{The WBGA-II approximation}
1823: \label{sec:ABBA+WBGAII}
1824: 
1825: In the WBGA-II approximation, when all terms are kept in
1826: Eqs.~(\ref{GammaAA})-(\ref{GammaAB}), it is useful to rewrite these
1827: equations in the vector form
1828: %
1829: \begin{equation}
1830:   \left( \frac{\partial}{\partial t}+2Dk^2 \right)
1831:      \left( 
1832:         \begin{array}{c} 
1833:           \Gamma^{AA}_k \\ [5pt]
1834:           \Gamma^{BB}_k \\ [5pt]
1835:           \Gamma^{AB}_k
1836:         \end{array}
1837:       \right)
1838:       = - {\bf P}(t) 
1839:      \left( 
1840:         \begin{array}{c} 
1841:           \Gamma^{AA}_k \\ [5pt]
1842:           \Gamma^{BB}_k \\ [5pt]
1843:           \Gamma^{AB}_k
1844:         \end{array}
1845:       \right) 
1846:       -
1847:       \left(
1848:       \begin{array}{c}
1849:         \lambda \varphi_{AA} \\ [5pt]
1850:         \lambda \varphi_{BB} \\ [5pt]
1851:         \delta  \varphi_{AB} 
1852:       \end{array}
1853:       \right)
1854:   \label{dtvecGam}
1855: \end{equation}
1856: %
1857: where the matrix $\bf P$ is given by
1858: %
1859: \begin{equation}
1860:  {\bf P} = 
1861:     n_A
1862:     \left(
1863:        \begin{array}{ccc}
1864:           4\lambda + 2 \delta u   & 0  & 2\delta  \\
1865:           0  & 4\lambda u + 2\delta &  2\delta u \\
1866:           \delta u    & \delta     &  (2\lambda+\delta)(1+u)
1867:        \end{array}
1868:     \right)
1869:   \label{P}
1870: \end{equation}
1871: %
1872: with $u=n_B/n_A$.
1873: 
1874: Vector equation (\ref{dtvecGam}) is very hard to solve
1875: analytically. However, there are some guidelines how to extract late
1876: time asymptotics. At the WBGA-I level it appears that ABBA model and
1877: the A+A model are very similar. In the following it will be assumed
1878: that such similarity can be extrapolated to the presently studied
1879: WBGA-II level.  This implies that mean field behavior should be
1880: expected from Eq.~(\ref{dtvecGam}).
1881: 
1882: To get the feeling for what follows it is useful to analyze
1883: Eqs.~(\ref{dtnA}) and (\ref{dtnB}) at the mean field level by
1884: neglecting fluctuations and setting $\Phi_{\rho\nu}$ to zero with
1885: $\rho,\nu=A,B$. Carrying out such procedure gives
1886: %
1887: \begin{eqnarray}
1888:   & & \frac{\partial}{\partial t} n_A = 
1889:         - \left( 
1890:              \lambda n_A^2 + \delta n_A n_B 
1891:           \right) 
1892:           \label{dtnA1} \\
1893:   & & \frac{\partial}{\partial t} n_B = 
1894:         - \left( 
1895:              \lambda n_B^2 + \delta n_A n_B 
1896:           \right)
1897:           \label{dtnB1} 
1898: \end{eqnarray}  
1899: %
1900: Equations above can be solved approximatively for large $t$ (please see
1901: refs.~\cite{KJ2} for details) and one obtains
1902: %
1903: \begin{eqnarray}
1904:   & & n_A\sim \frac{1}{\lambda t} 
1905:       \label{mfa} \\
1906:   & & n_B\sim \frac{
1907:                  n_{0,B} 
1908:                    }{ 
1909:                  [ n_{0,A}\lambda t ]^{\gamma} 
1910:                }  
1911:       \label{mfb}
1912: \end{eqnarray}
1913: % 
1914: provided $\gamma\equiv\delta/\lambda>1$ and $n_{0,A}>n_{0,B}$. For
1915: $\delta=\lambda$ ($\gamma=1$) or $n_{0,A}=n_{0,B}$ solution is
1916: trivial, and it can be easily shown that in such case the ABBA model
1917: belongs to the A+A universality class. These simple cases are not
1918: considered here.  Initial imbalance in particle concentration leads to
1919: faster diminishing of minority species, i.e.  $u(t)=n_B(t)/n_A(t)
1920: \rightarrow 0$ as $t\rightarrow\infty$ given $0<u(0)<1$.
1921: 
1922: 
1923: In the following we assume the mean field ansatz
1924: (\ref{mfa})-(\ref{mfb}) and try to solve (\ref{dtvecGam}) with it. The
1925: validity of such mean field ansatz will be checked self consistently
1926: at the end.  For large times, and with mean field behavior
1927: ($u\rightarrow 0$), the matrix ${\bf P}$ can be approximated by
1928: %
1929: \begin{equation}
1930:   {\bf P}\approx \lambda n_A {\bf\Pi}
1931: \end{equation}
1932: % 
1933: with
1934: %
1935: \begin{equation}
1936:   {\bf\Pi} =
1937:     \left(
1938:        \begin{array}{ccc}
1939:           4 & 0       &  2\gamma         \\
1940:           0 & 2\gamma &  0               \\
1941:           0 & \gamma  &  2+\gamma
1942:        \end{array}
1943:     \right)
1944:   \label{Pi}
1945: \end{equation}
1946: %
1947: The fact that ${\bf P}$ (in the approximate form) is constant matrix
1948: multiplied by time dependent function implies that $[\dot{\bf
1949: P}(t),\dot {\bf P}(t')]=0$ (dot over symbol ${\bf P}$ denotes time
1950: derivative). This being the case, Eq.~(\ref{dtvecGam}) can be treated
1951: as scalar equation and calculation similar to the one in the
1952: subsection \ref{sec:AA+WBGAII} gives
1953: %
1954: \begin{eqnarray}
1955:   \left( 
1956:     \begin{array}{c}
1957:       \varphi_{AA}(t) \\ [5pt]
1958:       \varphi_{BB}(t) \\ [5pt]
1959:       \varphi_{AB}(t) 
1960:     \end{array}
1961:   \right)
1962:   & = &
1963:   \left( 
1964:     \begin{array}{c}
1965:       n_A(t)^2 \\ [5pt]
1966:       n_B(t)^2 \\ [5pt]
1967:       n_A(t) n_B(t) 
1968:     \end{array}
1969:   \right) - \nonumber \\
1970:  & & 
1971:   -
1972:   \int_0^t dt' {\bf J}(t,t')
1973:   \left( 
1974:     \begin{array}{c}
1975:       \lambda \varphi_{AA}(t') \\ [5pt]
1976:       \lambda \varphi_{BB}(t') \\ [5pt]
1977:       \delta \varphi_{AB}(t') 
1978:     \end{array}
1979:   \right)
1980:   \label{varphi2}
1981: \end{eqnarray}
1982: %
1983: where matrix ${\bf J}(t,t')$ is given by
1984: %
1985: \begin{equation}
1986:   {\bf J}(t,t') = G(t,t') {\rm exp} 
1987:     \left[ -\xi(t,t') {\bf\Pi} \right]
1988:   \label{Jtt}
1989: \end{equation}
1990: %
1991: and 
1992: %
1993: \begin{equation}
1994:   \xi(t,t')\equiv \lambda \int_{t'}^t dt'' n_A(t'')
1995:   \label{xi}
1996: \end{equation}
1997: %
1998: Please compare Eqs.~(\ref{varphi1})-(\ref{Itt}) and
1999: (\ref{varphi2})-(\ref{Jtt}). They are very similar, the only difference
2000: being in the matrix character of (\ref{varphi2}-\ref{Jtt}). Following
2001: same steps as in section \ref{sec:AA+WBGAII} the Eq.~(\ref{varphi2})
2002: can be approximated as
2003: %
2004: \begin{equation}
2005:   \left( 
2006:     \begin{array}{c}
2007:       \varphi_{AA} \\ [5pt]
2008:       \varphi_{BB} \\ [5pt]
2009:       \varphi_{AB} 
2010:     \end{array}
2011:   \right)
2012:    \approx 
2013:   \left( 
2014:     \begin{array}{c}
2015:       n_A^2 \\ [5pt]
2016:       n_B^2 \\ [5pt]
2017:       n_A n_B 
2018:     \end{array}
2019:   \right) 
2020:   -
2021:   {\bf J}(t,0) 
2022:   \left( 
2023:     \begin{array}{c}
2024:       {\cal I}_{AA} \\ [5pt]
2025:       {\cal I}_{BB} \\ [5pt]
2026:       {\cal I}_{AB} 
2027:     \end{array}
2028:   \right)
2029:   \label{varphi3}
2030: \end{equation}
2031: %
2032: Now we proceed to show that, as in the case of (\ref{varphiapprox}),
2033: the second term on the right hand side of equation (\ref{varphi3}) can
2034: be neglected.  
2035: 
2036: Matrix ${\bf\Pi}$ can be diagonalized as ${\bf\Pi U}={\bf U \Omega}$. The
2037: ${\bf\Omega}$ is diagonal matrix containing eigenvalues
2038: %
2039: \begin{equation}
2040:   \omega_1=4 \ , \ \ 
2041:   \omega_2=2\gamma \ , \ \ 
2042:   \omega_3=2+\gamma
2043:   \label{eigenvalues}
2044: \end{equation}
2045: %
2046: and matrix ${\bf U}$ contains eigenvectors
2047: %
2048: \begin{equation}
2049:   {\bf U} = 
2050:     \left(
2051:       \begin{array}{ccc}
2052:         1 & \frac{\gamma}{\gamma-2} & \frac{2\gamma}{\gamma-2} \\ 
2053:         0 & \frac{\gamma-2}{\gamma} & 0                        \\
2054:         0 & 1                       & 1
2055:       \end{array}
2056:     \right)
2057:   \label{eigenvectors}
2058: \end{equation}
2059: %
2060: Inserting (\ref{mfa}) into (\ref{xi}), and assuming large $t$, leads to 
2061: %
2062: \begin{equation}
2063:   \xi(t,0) \sim const + \ln t
2064:   \label{xi1}
2065: \end{equation}
2066: %
2067: and using (\ref{xi1}) in (\ref{Jtt}) gives
2068: %
2069: \begin{equation}
2070:   {\bf J}(t,0)\sim {\rm const}\ t^{-d/2} {\bf U} 
2071:                 \left(
2072:                   \begin{array}{ccc}
2073:                      t^{-4} &     0        & 0               \\
2074:                      0      & t^{-2\gamma} & 0               \\
2075:                      0      &     0        & t^{-(2+\gamma)} 
2076:                   \end{array}
2077:                 \right)
2078:                 {\bf U}^{-1}
2079:   \label{Jt0}
2080: \end{equation}
2081: %
2082: Finally, the second term on the right hand side of (\ref{varphi3}) can
2083: be calculated explicitly. Inserting (\ref{Jt0}) into (\ref{varphi3}),
2084: and assuming that ${\cal I}_{\rho\nu}$ $\rho,\nu=A,B$ are constants
2085: (can be checked for self consistency at the end), results in
2086: %
2087: \begin{eqnarray}
2088:  & &  \varphi_{AA} \approx n_A^2 
2089:             + t^{-d/2}( c_1 t^{-\omega_1} 
2090:                       + c_2 t^{-\omega_2} 
2091:                       + c_3 t^{-\omega_3} ) \label{c1-3} \\ 
2092:  & &  \varphi_{BB} \approx n_B^2 
2093:             + t^{-d/2} c_4 t^{-\omega_2}  \label{c4} \\ 
2094:  & &  \varphi_{AB} \approx n_A n_B 
2095:             + t^{-d/2}( c_5 t^{-\omega_2} 
2096:                       + c_6 t^{-\omega_3} )  \label{c5-6}
2097: \end{eqnarray}
2098: %
2099: The explicit form of constants $c_1$, $c_2$, $c_3$, $c_4$, $c_5$ and
2100: $c_6$ is not interesting since aim is to show that terms containing
2101: these constants are sub-leading to the mean field terms.  By studying
2102: equation above row by row, it is possible to show that for $\gamma\ge
2103: 1$ terms involving constants are sub-leading to the mean field terms.
2104: 
2105: To see that terms originating from ${\bf J}(t,0)$ are sub-leading one
2106: really has to calculate ${\bf U}$ explicitly. For example, not knowing that
2107: contribution from $\omega_1$ is absent in (\ref{c4}), there would be a
2108: need to compare $t^{-2\gamma}$ (asymptotics of the mean field $n_B^2$
2109: term in \ref{c4}) with $t^{-(d/2+4)}$ (coming from ${\bf J}(t,0)$ and
2110: $\omega_1$ eigenvalue). One would conclude that
2111: $\gamma=\delta/\lambda$ can not be too large if mean field asymptotics
2112: is to hold. In reality, there is no such bound on ratio
2113: $\delta/\lambda$ since eigenvalue $\omega_1$ does not appear in
2114: Eq.~(\ref{c4}), but this can only be seen after explicit calculation.
2115: 
2116: Main findings so far is that WBGA describes ABBA and A+A model in the
2117: same way. For both models the WBGA-I (WBGA-II) predict $d/4$ (mean
2118: field) density decay exponents.  In the next section attempt will be
2119: made to improve WBGA method in order to obtain correct value of
2120: density decay exponent for ABBA model.
2121: 
2122: 
2123: 
2124: \subsection{The hybrid of WBGA/Kirkwood applied to A+A and B+B sectors}
2125: \label{sec:ABBA+hybrid}
2126: 
2127: 
2128: How deep the weakness of WBGA goes? What needs to be changed in
2129: equations of motion (\ref{GammaAA})-(\ref{GammaAB}) in order to get
2130: correct decay exponent?  To answer these questions we begin by
2131: modifying the A+A and B+B reaction sectors in
2132: (\ref{GammaAA})-(\ref{GammaAB}) by using the recipe from section
2133: \ref{sec:AA+hybrid}. It was already remarked in the section
2134: \ref{sec:model} that contributions to $H$ describing different
2135: reaction sectors enter additively, and this feature is reflected in
2136: equations of motion (\ref{GammaAA})-(\ref{GammaAB}). Because of this
2137: it is possible to focus on terms describing influence of A+A and B+B
2138: reactions, i.e. terms proportional to $\lambda$, which govern decay
2139: exponents of the ABBA model.  At the moment terms proportional to $\delta$
2140: in (\ref{GammaAA})-(\ref{GammaAB}) that describe A+B reaction will not
2141: be changed.
2142: 
2143: If one is to follow procedure described in section
2144: \ref{sec:AA+hybrid}, the ${\cal O}(n^3)$ term in Eq.~(\ref{GammaAA})
2145: has to be modified as
2146: %
2147: \begin{equation}
2148:   4 \lambda n_A \Gamma_k^{AA} \rightarrow 
2149:   2\lambda \Gamma_k^{AA} \frac{n_A^2+\Phi_{AA}}{n_A}
2150: \end{equation}
2151: %
2152: and likewise for Eq.~(\ref{GammaBB})
2153: %
2154: \begin{equation}
2155:   4 \lambda n_B \Gamma_k^{BB} \rightarrow 
2156:   2\lambda \Gamma_k^{BB} \frac{n_B^2+\Phi_{BB}}{n_B}
2157: \end{equation}
2158: %
2159: This gives a new set of, hopefully better, equations:
2160: %
2161: \begin{eqnarray}
2162:   \frac{\partial}{\partial t} \Gamma_k^{AA} 
2163:       & = & - 2 D k^2 \Gamma_k^{AA}   
2164:             - \lambda \left(  n_A^2 + \Phi_{AA} \right) - \nonumber \\
2165:       &   & - 2 \lambda \Gamma_k^{AA} \frac{n_A^2+\Phi_{AA}}{n_A} -\nonumber \\
2166:       &   & - 2 \delta ( n_B \Gamma^{AA}_k + n_A \Gamma^{AB}_k )
2167:       \label{GammaAA1} \\
2168:   \frac{\partial}{\partial t} \Gamma_k^{BB} 
2169:       & = & - 2 D k^2 \Gamma_k^{BB}   
2170:             - \lambda \left(  n_B^2 + \Phi_{BB} \right) - \nonumber \\
2171:       &   & - 2\lambda \Gamma_k^{BB} \frac{n_B^2+\Phi_{BB}}{n_B} - \nonumber \\
2172:       &   & - 2 \delta ( n_A \Gamma^{BB}_k + n_B \Gamma^{AB}_k ) 
2173:       \label{GammaBB1} 
2174: \end{eqnarray}
2175: %
2176: Equation~(\ref{GammaAB}) stays the same, although the
2177: Eq.~(\ref{GammaAB}) contains term proportional to $\lambda$ which
2178: should be modified if one follows the principle outlined above.
2179: However, at the moment, Eq.~(\ref{GammaAB}) will not be changed.
2180: 
2181: It is useful to employ similar notation to the one used in section
2182: \ref{sec:AA+WBGAII};
2183: %
2184: \begin{equation}
2185:   \Gamma_k^{\rho\nu} \equiv n_\rho n_\nu \chi_k^{\rho\nu} 
2186:   \label{chikab}  
2187: \end{equation}
2188: %
2189: and
2190: %
2191: \begin{equation}
2192:   \chi_{\rho\nu} \equiv 1 + \frac{1}{V} \sum_{k\ne 0} \chi_k^{\rho\nu} 
2193:   \label{chiab}
2194: \end{equation}
2195: %
2196: with $\rho,\nu=A,B$.  Using (\ref{chikab}) and (\ref{chiab})
2197: in Eqs.~(\ref{dadt}) and (\ref{dbdt}) results in
2198: %
2199: \begin{eqnarray}
2200:  & & \frac{\partial}{\partial t} n_A 
2201:        = - ( \lambda n_A^2 \chi_{AA} + \delta n_A n_B \chi_{AB} ) 
2202:      \label{dtnA2}\\
2203:  & & \frac{\partial}{\partial t} n_B 
2204:        = - ( \lambda n_B^2 \chi_{BB} + \delta n_A n_B \chi_{AB} )
2205:      \label{dtnB2}
2206: \end{eqnarray}
2207: %
2208: Implementing same notation in Eqs.~(\ref{GammaAA1}), (\ref{GammaBB1})
2209: and (\ref{GammaAB}) gives
2210: %
2211: \begin{eqnarray}
2212:   \frac{\partial}{\partial t} \chi_k^{AA} 
2213:       & = & - 2 D k^2 \chi_k^{AA} - \lambda \chi_{AA} - \nonumber \\
2214:       &   & - 2 \delta n_B ( 1 - \chi_{AB} ) \chi_k^{AA} 
2215:             - 2 \delta n_B \chi_k^{AB}
2216:       \label{chiAA} \\
2217:   \frac{\partial}{\partial t} \chi_k^{BB} 
2218:       & = & - 2 D k^2 \chi_k^{BB} - \lambda \chi_{BB} - \nonumber \\
2219:       &   & - 2 \delta n_A ( 1 - \chi_{AB} ) \chi_k^{BB} 
2220:             - 2 \delta n_A \chi_k^{AB}
2221:       \label{chiBB} \\
2222:   \frac{\partial}{\partial t} \chi_k^{AB} 
2223:       & = & - 2 D k^2 \chi_k^{AB} - \delta \chi_{AB} - \nonumber \\
2224:       &   & - [ 
2225:                 \lambda n_A ( 2 - \chi_{AA} ) +
2226:                 \lambda n_B ( 2 - \chi_{BB} ) + \nonumber \\
2227:       &   & \ \ \ 
2228:                 + \delta ( n_A + n_B ) ( 1 - \chi_{AB} ) 
2229:               ] \chi_k^{AB} - \nonumber \\ 
2230:       &   & - \delta ( n_A \chi_k^{AA} + n_B \chi_k^{BB} )
2231:       \label{chiAB} 
2232: \end{eqnarray}
2233: %
2234: 
2235: The numerical solution of the set of equations above is shown in
2236: Fig. 2 (dotted line). The full line is a result of Monte Carlo
2237: simulation where particle densities are obtained as ensemble averages
2238: over 500 runs (simulation is repeated 500 times with a shift in the
2239: random number generator).  The Eqs.~(\ref{chiAA})-(\ref{chiAB}) do not
2240: describe ABBA model properly, not even qualitatively, since minority
2241: species die out faster (the particle density ratio grows to infinity)
2242: while the simulation shows that density ratio should saturate to a
2243: constant value (full line). Thus the attempt of modifying terms
2244: describing A+A and B+B reaction sector in AA and BB correlation
2245: functions by using shortened Kirkwood superposition approximation does
2246: not cure weaknesses of WBGA approach. The inspection of individual
2247: density decays reveals that the equations above correctly describe
2248: decay of majority species, i.e. $n_A\sim ({\rm const}) t^{-d/2}$, but
2249: fail do describe decay of minority species $n_B$.
2250: 
2251: In the following more terms will be modified by using Kirkwood
2252: superposition approximation.  The dash-dot line in Fig.~2 shows a
2253: solution of equations (not shown here) obtained from modifying A+A and
2254: B+B reaction sectors using shortened Kirkwood superposition
2255: approximation in the Eq.~(\ref{GammaAB}) describing time evolution of
2256: AB correlation function. Equations obtained in this way are identical
2257: to the (\ref{chiAA})-(\ref{chiAB}) with only difference that
2258: Eq.~(\ref{chiAB}) changes in a way that terms proportional to
2259: $\lambda$ drop out. In Fig.~2 it can be seen that even if all
2260: $\lambda$ proportional terms are modified the density ratio curve
2261: climbs to infinity, which indicates that minority species dies out,
2262: which is not correct behavior. However, the overall trend gets better
2263: as the dash-dot curve lies bellow dotted one and is pushed towards
2264: simulation curve.
2265: 
2266: To continue this line of incremental changes the equations of motion
2267: where studied where even the A+B reaction sector was modified using
2268: shortened Kirkwood superposition approximation (all terms proportional
2269: to $\delta$ were modified). Equations obtained in this way are same as
2270: in (\ref{chiAA})-(\ref{chiAB}) the only difference being in the fact
2271: that all seemingly ${\cal O}(n)$ terms drop out. Thus in the
2272: (\ref{chiAA})-(\ref{chiAB}) only diffusion term and terms $\lambda
2273: \chi_{AA}$, $\lambda \chi_{BB}$ and $\delta \chi_{AB}$ are kept in
2274: (\ref{chiAA}), (\ref{chiBB}) and (\ref{chiAB}) respectively. These
2275: equations are not shown explicitly to save the space but it should be
2276: clear how they look like.  The numerical solution for this set of
2277: equations is shown in Fig.~2 as dashed line. [It is possible to
2278: analytically extract density decay asymptotics for this truncated set
2279: of equations which gives $n_A(t)\sim({\rm const})t^{-d/2}$ and
2280: $n_B(t)\sim({\rm const'})t^{-d/4}$.]
2281: 
2282: The set of equations where Kirkwood superposition approximation has
2283: been applied fully agrees with the numerical simulation much better
2284: than the rest which are mixture of WBGA and Kirkwood superposition
2285: approximation. This is strong indication that, at least for the ABBA
2286: model, the Kirkwood superposition approximation is superior to the
2287: WBGA method.  For example, in Fig. 2, the trend in all curves improve
2288: as the content of superposition approximation is increased in
2289: equations of motion (dotted line is worse as it climbs fastest,
2290: dash-dot line is a bit better, while only dashed line where
2291: superposition approximation is implemented fully saturates to
2292: constant). As the goal of the present study is to understand WBGA
2293: method better, the more detailed analysis of ABBA reaction based on
2294: Kirkwood superposition approximation will be presented in forthcoming
2295: publication.
2296: 
2297: \section{Conclusions}
2298: \label{sec:conclusions}
2299: 
2300: The goal of this study was to test workings of WBGA and to relate
2301: some seemingly independent calculation schemes available in the
2302: literature. In particular, this work has addressed few important
2303: issues.
2304: 
2305: (1) It was shown that WBGA-I is equivalent to the dressed tree calculation
2306: introduced in ref.~\cite{Lee}, and this equivalence holds for any
2307: model where particles annihilate in pairs.  Furthermore, it was shown
2308: that for the A+A reaction the dressed tree calculation results in
2309: $d/4$ density decay exponent.  Thus the contradictory claims of
2310: refs.~\cite{Lee} and~\cite{rev6} have been sorted out.
2311: 
2312: (2) The WBGA method describes A+B reaction well, but does not work for
2313: A+A and ABBA models, and it is reasonable to expect that there are
2314: more models that can not be described by WBGA.  In the case of ABBA
2315: model WBGA predicts faster vanishing of minority species which is
2316: suggestive of the A+B like behavior rather than the behavior of the
2317: ABBA model as found in refs.~\cite{KJ1,KJ2}. This bias towards A+B
2318: type behavior is very hard to get rid off as successively correcting
2319: more and more terms in equations of motion for ABBA model by using
2320: Kirkwood superposition approximation results in faster vanishing of
2321: minority species. The vanishing of minority species persists until all
2322: terms are modified by Kirkwood superposition approximation. The WBGA
2323: emphasize A+B reaction sector too strongly in the ABBA model.
2324: 
2325: (3) Findings of this work suggest that formalism employed by
2326: Mattis-Glasser in~\cite{rev6} where small $n_0$ expansion is
2327: introduced (and applied to study A+B model) is somewhat
2328: questionable. This procedure works on A+B model, but might not work
2329: for other models. It can be shown (by rescaling $a^\dagger
2330: n_0\rightarrow a^\dagger$ and $a/n_0\rightarrow a$) that for the type
2331: of models studied in here, the small $n_0$ approximation of
2332: ref.~\cite{rev6} amounts to taking away three body terms in
2333: Hamiltonian given in Eq.~(\ref{Hreactx'}) or (\ref{Hreactk})
2334: (e.g. operators of the type $a^\dagger a a$ and likewise any mixture
2335: of $a$ or $b$ operators). The neglect of these terms amounts exactly
2336: to WBGA-I approach, i.e. neglect of seemingly ${\cal O}(n^3)$ terms in
2337: equations of motion for correlation functions; for example, equations
2338: (\ref{dn3})-(\ref{varphi}) of section \ref{sec:AA+WBGAI} (A+A), or
2339: equations (\ref{dadt})-(\ref{dvarab}) from section
2340: \ref{sec:ABBA+WBGAI} (ABBA). In here it has been shown that WBGA-I set
2341: of equations result in incorrect $d/4$ density decay exponent when
2342: used for A+A and ABBA models. For these reasons, small $n_0$
2343: expansion, which effectively means taking away three body term in
2344: Hamiltonian, can not be trusted if used beyond A+B model.
2345: 
2346: (4) The Kirkwood superposition approximation was formulated for the
2347: case of lattice dynamics with multiple occupancy of lattice sites
2348: allowed. There is strong indication from the present analysis that
2349: Kirkwood superposition approximation is superior to the WBGA method,
2350: at least when applied to the A+A and ABBA models.  However, one has to
2351: be careful in claiming superiority of Kirkwood superposition
2352: approximation over WBGA approach since in here it was shown that
2353: Kirkwood superposition approximation describes ABBA model
2354: qualitatively, but it remains to be seen whether it works on the A+B
2355: model which WBGA describes well. Similar study (where only single
2356: occupancy of lattice sites was allowed and reaction range was assumed
2357: short but finite) showed that Kirkwood superposition approximation can
2358: describe A+B reaction~\cite{rev4}, and claim of superposition
2359: approximation superiority is likely to be correct but, nevertheless,
2360: such claim has to be tested throughly.
2361: 
2362: The application of Kirkwood superposition approximation to the ABBA
2363: and A+B models will be presented in the forthcoming publication as
2364: there are many interesting technical details that need to be sorted
2365: out. For example, from present study there are some indications
2366: (results not shown here) that when using Kirkwood superposition
2367: approximation for extremely short range reaction, such as on-site
2368: reaction studied here, care has to be taken so that thermodynamic
2369: limit is accounted for properly. It seems that one has to approach
2370: Kirkwood superposition approximation through formalism developed in
2371: section \ref{sec:AA+hybrid}.
2372: 
2373: To conclude, it would be interesting to have a relatively simple
2374: approximation at hand, not far away from pair approximation, which
2375: could be used to extract qualitative asymptotics for arbitrary
2376: reaction-diffusion model. Clearly such program is ambitious since in
2377: reality one is bound to make approximation which is related to the
2378: particular model but, nevertheless, it is worth a try. The A+A and A+B
2379: reaction-diffusion models (or combination of them) are excellent
2380: bench-mark models and any successful approximation should strive to
2381: describe these reactions properly.  Present study is a attempt in this
2382: direction.
2383: 
2384: \begin{acknowledgments}
2385: 
2386: I would like to thank Prof. Malte Henkel and the staff at the
2387: Laboratoire de Physique des Materiaux, Universite Henri Poincare Nancy
2388: I, for warm hospitality where part of this work was done.
2389: 
2390: \end{acknowledgments}
2391: 
2392: 
2393: 
2394: 
2395: 
2396: \appendix
2397: 
2398: 
2399: \section{Approximation for $G(\lowercase{t}-\lowercase{t}')$}
2400: \label{app:Gtt}
2401: 
2402: In Eq.~(\ref{Gtt}) sum over $k$ can be approximated as integral
2403: %
2404: \begin{equation}
2405:   G(t-t') = \frac{1}{(2\pi)^d} 
2406:     \left[ 
2407:        \int_{-\pi}^\pi dp e^{-2Dp^2(t-t')}
2408:     \right]^d
2409:   \label{Gtt1}
2410: \end{equation}
2411: %
2412: The equality sign is strictly valid in thermodynamic limit only when
2413: $V\rightarrow\infty$ has been taken.  It is useful to approximate
2414: integral above further by changing bounds of integration
2415: %
2416: \begin{equation}
2417:   G(t-t') \approx \frac{1}{(2\pi)^d} 
2418:      \left[
2419:        \int_{-\infty}^\infty dp
2420:        e^{-p^2 \Lambda^{-2}-2D\kappa^2(t-t')}
2421:      \right]^d
2422:   \label{Gtt2}
2423: \end{equation}
2424: %
2425: $\Lambda$ corresponds to a large cutoff for the $k$-vector summation
2426: (it regularizes UV divergences of theory). The comparison of
2427: (\ref{Gtt1}) and (\ref{Gtt2}) quickly reveal that $\Lambda$ is a
2428: constant roughly equal to $\pi$. The evaluation of integral above gives
2429: Eq.~(\ref{Gtt3}) with $\eta=\frac{1}{2D\Lambda^2}$. It is convenient
2430: to chose $\eta=\frac{1}{8\pi D}$ and $\Lambda=2\sqrt{\pi}$ since this
2431: gives correct value for $G(0)=1$. For large $t-t'$ particular value of
2432: $\Lambda$ appears to be irrelevant, {\em i.e.} absent from final
2433: expressions for particle density for large times. However, this really
2434: depends on the critical dimension of the field theory.
2435: 
2436: 
2437: 
2438: \section{Integration of singular kernel}
2439: \label{app:kernal}
2440: 
2441: In here the numerical treatment of Eqs.~(\ref{dn3}-\ref{varphi}) is
2442: described in a more detail. The general procedure for integrating
2443: expressions of the type
2444: %
2445: \begin{equation}
2446:   I_i[f] = \int_0^{t_i} ds K(t_i,s)f(s)
2447:   \label{Iif}
2448: \end{equation}
2449: %
2450: where $K(t,s)$ is singular when s approaches t is described in
2451: ref.~\cite{delves}. The $i=0,1,2,\ldots$ and $t_i = i h$. In here the
2452: particular focus in on the particular form of singular kernel
2453: originating from propagator $G(t,t')$;
2454: %
2455: \begin{equation}
2456:   K(t,s)=(t-s+\eta)^{-\alpha}
2457:   \label{Kts}
2458: \end{equation}
2459: %
2460: The aim is to find quadrature formula which can integrate (\ref{Iif})
2461: exactly if $f(s)$ is linear within intervals $t_{i-1},t_i$. Thus goal is
2462: to find coefficients $w_{ij}$ in
2463: %
2464: \begin{equation}
2465:   I_i[f] \approx \sum_{j=0}^i w_{ij} f(t_j)
2466:   \label{Iifapp}
2467: \end{equation}
2468: %
2469: such that above equation turns equality for piecewise linear function
2470: $f(s)$. Implementing this procedure gives
2471: %
2472: \begin{equation}
2473:   I_i[f] = \sum_{j=0}^{i-1} \int_{s_j}^{s_{j+1}} ds K(t_i,s) f(s)
2474: \end{equation}
2475: %
2476: and after approximating $f(s)$ as linear function in intervals
2477: $t_j,t_{j+1}$ for $j=0,..,i-1$
2478: %
2479: \begin{equation}
2480:   f(s) \approx \frac{s_{j+1}-s}{h} f(s_j) + \frac{s-s_j}{h} f(s_{j+1})
2481: \end{equation}
2482: %
2483: gives (\ref{Iifapp}) with
2484: %
2485: \begin{eqnarray}
2486:   w_{i0} &=& \int_0^h ds K(t_i,s) \frac{h-s}{h} \\
2487:   w_{ij} &=& \int_{s_{j-1}}^{s_j} ds K(t_i,s) \frac{s-s_{j-1}}{h}
2488:                  + \nonumber \\ 
2489:          & &     + \int_{s_j}^{s_{j+1}} ds K(t_i,s) \frac{s_{j+1}-s}{h} \\
2490:   w_{ii} &=& \int_{s_{i-1}}^{s_i} ds K(t_i,s) \frac{s-s_{i-1}}{h}
2491: \end{eqnarray}
2492: %
2493: and after performing integrals above with $K(t,s)$ given in (\ref{Kts})
2494: %
2495: \begin{eqnarray}
2496:   w_{i0} &=& \frac{h^{1-\alpha}}{(\alpha-1)(\alpha-2)}
2497:              [(2-\alpha-i-\eta)(i+\eta)^{1-\alpha} + \nonumber \\
2498:          & &  + (i+\eta-1)^{2-\alpha}] 
2499:              \label{wi0} \\
2500:   w_{ij} &=& \frac{h^{1-\alpha}}{(\alpha-1)(\alpha-2)}
2501:              [(i-j+\eta-1)^{2-\alpha} + \nonumber \\
2502:          & & + (i-j+\eta+1)^{2-\alpha} - 2(i-j+\eta)^{2-\alpha}] 
2503:              \label{wij} \\
2504:   w_{ii} &=& \frac{h^{1-\alpha}}{(\alpha-1)(\alpha-2)}
2505:              [(1+\eta)^{2-\alpha}+ \nonumber \\
2506:          & & + \eta^{1-\alpha}(\alpha-\eta-2)]
2507:              \label{wii}
2508: \end{eqnarray}
2509: %
2510: The pair of equations in (\ref{dn3}-\ref{varphi}) is discretized as
2511: follows. First the differential equation (\ref{dn3}) is rewritten in
2512: integral form as $n(t)=n_0-\lambda \int_0^t ds \varphi(s)$ and
2513: trapezoidal rule is used to evaluate integral since all functions are
2514: well behaved. However, for Eq.~(\ref{varphi}) the rule (\ref{Iifapp})
2515: and (\ref{wi0})-(\ref{wii}) designed for singular kernel is
2516: used. Implementation of this philosophy gives
2517: %
2518: \begin{eqnarray}
2519:   n_i = n_0 - \lambda h [ \frac{\varphi_0}{2} 
2520:               + \sum_{j=1}^{i-1} \varphi_j + \frac{\varphi_i}{2} ] \\
2521:   \varphi_i = n_i^2 - \lambda \sum_{j=0}^{i-1} w_{ij} \varphi_j 
2522:                     - \lambda w_{ii} \varphi_i 
2523: \end{eqnarray}
2524: %
2525: where $n_i=n(t_i)$ and $\varphi_i=\varphi(t_i)$ for
2526: $i=0,1,2,\ldots$. Given that all $n_j$ and $\varphi_j$ are known for
2527: $j=0,1,2,\ldots,i-1$ using equations above it is possible to calculate
2528: $n_i$ and $\varphi_i$. The iteration is started with $n_0=n(0)$ and
2529: $\varphi_0=n_0^2$.
2530: 
2531: 
2532: 
2533: 
2534: \section{Finding inverse Laplace transform of $\lowercase{\kappa(s)}$ 
2535: for $\lowercase{d}=2$}
2536: \label{app:kappa}
2537: 
2538: In here inverse Laplace transform of $\kappa(s)$ for $d=2$ given in
2539: Eq.~(\ref{kappas}) will be found. Due to the presence of log one has
2540: to use Bramowitz contour to perform integration over $s$. Also,
2541: function $\kappa(s)$ does not have poles. This means that only
2542: contribution to $\kappa(t)$ comes from the branch cut and one obtains
2543: %
2544: \begin{equation}
2545:   \kappa(t) = 8\pi D \int_0^\infty \frac{du}{u} e^{-ut}
2546:       \frac{1}{(\gamma_E+\ln(\eta u))^2+\pi^2}
2547: \end{equation}
2548: %
2549: It is useful to re-scale integration variable as $\eta u\rightarrow u$
2550: with only change that time variable appears in combination $\tau\equiv
2551: t/\eta$. In the following we set $\eta=1$ but keep in mind that at the
2552: end of calculation $t$ has to be changed into $t/\eta$.
2553: 
2554: As $t$ grows, due to the presence of ${\rm exp}(-ut)$, only smaller and
2555: smaller values for $u$ contribute to the integral above and it is
2556: useful to split integration over $u$ in the two intervals (a) from $0$
2557: to $c$ and (b) from $c$ to infinity, where $0<c<1$;
2558: $\kappa(t)=\kappa_a(t)+\kappa_b(t)$. It is possible to find upper
2559: bound for $\kappa_b$ as follows. Starting from
2560: %
2561: \begin{equation}
2562:   \kappa_b(t) \le \frac{8\pi D}{\gamma_E^2+\pi^2} 
2563:                  \int_c^\infty \frac{du}{u} e^{-ut}    
2564: \end{equation}
2565: %
2566: which, upon using the fact that for particular range of integration
2567: above $1/u\le 1/c$ and performing remaining integration, gives
2568: %
2569: \begin{equation}
2570:   \kappa_b(t) \le \frac{8\pi D}{(\gamma_E^2+\pi^2)c} \frac{e^{-ct}}{t} 
2571: \end{equation}
2572: %
2573: It will be shown that $\kappa_a(t)$ vanishes lot slower than $\kappa_b(t)$. 
2574: One can approximate expression for $\kappa_a$ as 
2575: %
2576: \begin{equation}
2577:   \kappa_a \approx (8\pi D) \int_0^c \frac{du}{u} e^{-ut} \frac{1}{(\ln u)^2}
2578: \end{equation}
2579: %
2580: and it can be shown that terms omitted do not influence leading order
2581: behavior for $\kappa_a$. By using partial integration expression above becomes
2582: %
2583: \begin{equation}
2584:   \kappa_a = 8\pi D \left[ 
2585:      - t \int_0^c du e^{-ut}\frac{1}{\ln u} + {\cal O}(e^{-ct}) \right]
2586: \end{equation}
2587: %
2588: The integral over $u$ is most conveniently calculated by changing
2589: variables $tu=v$ which gives
2590: %
2591: \begin{equation}
2592:   \kappa_a(t) \approx \frac{8\pi D}{\ln t} \int_0^{ct} dv e^{-v} 
2593:                       \frac{1}{1-\frac{\ln v}{\ln t}}   
2594: \end{equation}
2595: %
2596: By sending upper integration limit to infinity, and expanding
2597: denominator in series over $\ln v/\ln t$ one gets
2598: %
2599: \begin{equation}
2600:   \kappa_a = \frac{8\pi D}{\ln t} [ 1 + {\cal O} (1/\ln t) ] 
2601: \end{equation}
2602: %
2603: Keeping in mind that transformation $t\rightarrow t/\eta$ has to be
2604: made in the expression above gives result for $\kappa(t)$ in
2605: Eq.~(\ref{kappat}).
2606: 
2607: \begin{thebibliography}{19}
2608: 
2609: \bibitem{rev1} A. S. Mikhailov, Phys. Rep. {\bf 184}, 307 (1989).
2610: 
2611: \bibitem{rev2} V. Privman, ``Nonequilibrium Statistical Mechanics in
2612: One Dimension'', (Cambridge Univ. Press, 1997)
2613: 
2614: \bibitem{rev3} Comprehensive Chemical Kinetics, Vol. 25,
2615: ``Diffusion-limited reactions'', C.H. Bamford, C.F.H. Tipper and
2616: R.G. Compton Editors, (Elsevier, 1985).
2617: 
2618: \bibitem{rev4} E. Kotomin and V. Kuzovkov, in Comprehensive Chemical
2619: Kinetics, Vol. 34, ``Modern aspects of diffusion-controlled
2620: reactions'', R.G.Compton and G. Hancock Editors, (Elsevier, 1996).
2621: 
2622: \bibitem{rev5} E. Kotomin and V. Kuzovkov, Rep. Prog. Phys. {\bf 55},
2623: 2079 (1992).
2624: 
2625: \bibitem{rev6} D.C. Mattis and M.L. Glasser, Rev. Mod. Phys. {\bf
2626: 70}(3), 979 (1998).
2627: 
2628: \bibitem{rev7} J. Cardy, cond-mat/9607163, ``Renormalisation group
2629: approach to reaction-diffusion problems''.
2630: 
2631: \bibitem{rev8} A.S. Mikhailov and V.V. Yashin, J. Stat. Phys. {\bf
2632: 38}(1/2), 347 (1985). 
2633: 
2634: \bibitem{OTB} For a very concise review of Smoluchowskii approach and
2635: WBGA, please see A.A. Ovchinnikov, S.F. Timashev, and A.A. Belyy,
2636: ``Kinetics of diffusion controlled chemical processes'', (Nova
2637: Science, 1989).
2638: 
2639: \bibitem{bog} For original work on weakly non-ideal Bose gases please
2640: see N.N. Bogolybov, Izv. Akad. Nauk SSSR (Ser. Fiz.) {\bf II}, 77
2641: (1974), or N.N. Bogoliubov, {\em Lectures on Quantum Statistics},
2642: (Gordon and Breach, New York, 1967), p.p. 107-119.
2643: 
2644: %WBGA on A+A
2645: 
2646: \bibitem{ZO} Y.B. Zeldovich, A.A. Ovchinnikov, 
2647: Zh. Eksp. Teor. Fiz. {\bf 74}(5), 1588 (1978).
2648: 
2649: %WBGA on A+B
2650: 
2651: \bibitem{BOP} S.F. Burlatskii, A.A. Ovchinnikov, K.A. Pronin, 
2652: Zh. Eksp. Teor. Fiz. {\bf 92}(2), 625 (1987). 
2653: 
2654: \bibitem{GMY} A.M. Gutin, A.S. Mikhailov, V.V. Yashin,
2655: Zh. Eksp. Teor. Fiz. {\bf 92}(3), 941 (1987).
2656: 
2657: %more details on ABBA model
2658: 
2659: \bibitem{KJ1} Z. Konkoli, H. Johannesson and B. P. Lee, Phys. Rev. E
2660: {\bf 59}, R3787 (1999).
2661: 
2662: \bibitem{KJ2} Z. Konkoli and H. Johannesson, Phys. Rev. E {\bf 62}(3),
2663: 3276 (2000).
2664: 
2665: % review on A+A and A+B, exact results
2666: 
2667: \bibitem{CGP} A.M.R. Cadilhe, M.L. Glasser and V. Privman,
2668: Int. J. Mod. Phys. B {\bf 11}, 109 (1997).
2669: 
2670: \bibitem{foot2} For example, one way to prepare system initially is to
2671: take given number of particles and distribute them randomly one by one
2672: on the lattice. This way of preparation leads to Poisson distribution
2673: of particle number at each lattice site. Also, it is clear that
2674: preparing system in this way does not lead to correlation among
2675: particles. Thus, saying that particles are distributed according to
2676: Poisson-distribution amounts to saying that there are no correlations
2677: among them.
2678: 
2679: 
2680: \bibitem{foot1} The initial state of the system was prepared by
2681: allowing for birth and annihilation of particles and waiting long
2682: enough to establish stationary state. Once this stationary state was
2683: reached, particle birth was ceased and system continued to evolve by
2684: annihilation process solely.
2685: 
2686: \bibitem{Lee} B.P. Lee, J. Phys. A {\bf 27}, 2633 (1994). 
2687: 
2688: %exact A+A in fermionic representation
2689: 
2690: \bibitem{Lush} A.A. Lushnikov, Zh. Eksp. Teor. Fiz. {\bf 91}, 1376
2691: (1986).
2692: 
2693: 
2694: \bibitem{LeeCardy} B.P. Lee and J. Cardy, J. Staty. Phys. {\bf 80}, 971
2695: (1995).
2696: 
2697: \bibitem{Rudav} M.G. Rudavets, J. Phys. A {\bf 26}, 5313 (1993).
2698: 
2699: \bibitem{bram} M. Bramson and J.L. Lebowitz, Phys. Rev. Lett. {\bf
2700: 61}(21), 2397 (1988).
2701: 
2702: \bibitem{foot3} Acting by time derivative on (\ref{BarPsiHPsi0}) gives,
2703: up to minus sign, $\langle 0|a_0\bar H|\bar\Psi\rangle$ which due to
2704: $\langle 0|\bar H=0$ (probability conservation) can be rewritten as
2705: $\langle 0|a_0\bar H-\bar Ha_0|\bar\Psi\rangle$.
2706: 
2707: \bibitem{foot4} Inverse Fourier transform of $\langle
2708: a_{k_1}a_{k_2}a_{k_3} \rangle$ is defined as $\langle a_x a_y a_z
2709: \rangle = \frac{1}{V^{3/2}} \sum_{k_1,k_2,k_3}e^{i(k_1x+k_2y+k_3z)}
2710: \langle a_{k_1}a_{k_2}a_{k_3} \rangle$.
2711: 
2712: \bibitem{foot5} The Fourier transform of (\ref{kirkwood}) is
2713: calculated by using $\langle a_x a_y \rangle = \frac{1}{V} \sum_k
2714: e^{ik(x-y)} \langle a_k a_{-k} \rangle$. Thus, translational
2715: invariance is ensured in form of $\langle a_x a_y \rangle$. The term
2716: $\delta(k_1+k_2+k_3)$ appears in (\ref{kirkwoodk}) as artifact of
2717: that.
2718: 
2719: \bibitem{Howard} M. Howard, J. Phys. A {\bf 29}, 3437 (1996).
2720: 
2721: \bibitem{foot6} The density decay amplitude for A+A model is
2722: independent of $\lambda$ and $n_0$ while the one of A+B model depends
2723: only on $n_0$ (given that systems are observed bellow critical
2724: dimension, naturally).
2725: 
2726: 
2727: \bibitem{delves} L.M. Delves and J.L. Mohamed, ``Computational methods
2728: for integral equations'', (Cambridge Univ. Press, 1985).
2729: 
2730: \end{thebibliography}
2731: 
2732: \begin{figure}
2733: %%%%%%%%%%%%%%%%%%% will be taken away later BEGIN %%%%%%%%%%%%%%%%%%
2734: \epsfxsize=9cm
2735: \epsfysize=8cm
2736: \epsfbox{fig1.eps}
2737: %%%%%%%%%%%%%%%%%%% will be taken away later END %%%%%%%%%%%%%%%%%%
2738: \caption{FIG 1. The numerical solution of
2739: Eqs.~(\ref{dn3}-\ref{varphi}) for $d=1,1.5,2.5,3$ (solid lines). The
2740: dotted lines indicate asymptotics as given by Eq.~(\ref{ntasym}). Time
2741: is given in seconds and particle density $n(t)$ is dimensionless in
2742: units of particles per site. Initial density $n_0$ was set equal to 1,
2743: and reaction rate $\lambda=1s^{-1}$ was used.  }
2744: \label{fig1}
2745: \end{figure}
2746: 
2747: \begin{figure}
2748: %%%%%%%%%%%%%%%%%%% will be taken away later BEGIN %%%%%%%%%%%%%%%%%%
2749: \epsfxsize=9cm
2750: \epsfysize=8cm
2751: \epsfbox{fig2.eps}
2752: %%%%%%%%%%%%%%%%%%% will be taken away later END %%%%%%%%%%%%%%%%%%
2753: \caption{FIG 2. The numerical solution of
2754: Eqs.~(\ref{dtnA2}-\ref{chiAB}) for $d=1$ with increasing amount of
2755: Kirkwood superposition approximation embedded (dotted, dash-dot and
2756: dashed line). Full curve is result of a Monte Carlo simulation
2757: (average of 500 runs). Parameters used are $L=1000$, $n_A(0)=2$,
2758: $n_B(0)=1$, $\lambda=1$ and $\delta=2$.  }
2759: \label{fig2}
2760: \end{figure}
2761: 
2762: \end{multicols}
2763: 
2764: \end{document}
2765: 
2766: