1: Although the correspondence between constant kernel aggregation and the
2: $A+A\to A$ system is convenient for calculations and greatly simplifies the
3: field theoretic description of the problem, the physics is sometimes obscured
4: by this mapping. In this section we show that the results for the average
5: mass density, derived in section \ref{sec-avgDensity} from a field
6: theoretic perspective, can also be interpreted in a more physically
7: transparent way using some heuristic arguments closely related to the
8: so-called Smoluchowski approximation used in \cite{krapivsky1994} to take into
9: account fluctuations in low dimensional heterogeneous annihilation.
10:
11: The idea is as follows. If we are interested in the average mass density, then
12: the only diagrams remaining in the field theory associated with the stochastic
13: rate equation, eq. (\ref{sre}), after all trees have been summed are those which
14: renormalise the reaction rate. Therefore it should be possible to understand
15: the problem directly in terms of a standard Smoluchowski kinetic theory
16: with a suitably modified kernel, without reference to the Laplace-transformed
17: field $R_\mu$. Furthermore, the stationary state of this renormalised
18: Smoluchowski theory can be found directly without need for any R.G. analysis
19: using a technique, known as the Zakharov Transformation, borrowed from the the
20: theory of wave turbulence.
21:
22: The variation of the dimensionless reaction rate, $g$, as the length scale
23: $L_\mu$ is changed is determined by the $\beta$-functions computed in
24: section \ref{sec-betaFunction}. From eq. (\ref{eq-gdlt2}) and eq.
25: (\ref{eq-gdeq2}) we can obtain the renormalisation law for the physical
26: reaction rate, $\lambda_R(L_\mu) = g_R(L_\mu)\,L_\mu^{-\epsilon}$. Since the
27: Laplace variable, $\mu$, can be thought of as an inverse mass, we can
28: make the heuristic substitution,
29: \begin{equation}
30: L_\mu \sim \left(\frac{J}{D m_0}\right)^{-\frac{1}{4}}\,m^{\frac{1}{d+2}},
31: \end{equation}
32: to motivate the following mass-dependent reaction rates :
33: \begin{eqnarray}
34: \label{eq-renLambda} \lambda_R(m) &=& \frac{\lambda}{1+\frac{\lambda}{2 \pi
35: \epsilon} \left(\frac{J}{m_0
36: D}\right)^{-{\frac{\epsilon}{d+2}}}m^\frac{\epsilon}{d+2}}, \quad \mbox{$d<2$},\\
37: \label{eq-renLambda2}\lambda_R(m) &= &\frac{\lambda}{1+\frac{\lambda}{8 \pi}
38: \ln \left(\frac{m}{m_0}\right)},\qquad \qquad \qquad \! \mbox{$d=2$}.
39: \end{eqnarray}
40: The renormalised Smoluchowski equation (RSE) is then obtained from the mean field
41: Smoluchowski equation by substituting the above mass-dependent reaction rate
42: for $\lambda$. The density should therefore satisfy the following equation at
43: large times
44: \begin{widetext}
45: \begin{eqnarray}
46: \nonumber \frac{\partial N(m,t)}{\partial t} &=&
47: \int_0^{\infty}dm_1 dm_2
48: \lambda_R(m) N(m_1,t) N(m_2,t) \delta(m-m_1-m_2)\\
49: \label{eq-RSE}&-&
50: \int_0^{\infty}dm_1dm_2
51: \lambda_R(m_2)N(m,t) N(m_1,t)\delta(m_2-m-m_1)\\
52: \nonumber&-&
53: \int_0^{\infty}dm_1dm_2
54: \lambda_R(m_1)N(m,t) N(m_2,t)\delta(m_1-m_2-m)\\
55: \nonumber
56: &+& \frac{J}{m_0}\,\delta(m-m_0).
57: \end{eqnarray}
58: \end{widetext}
59: The stationary state of this equation is best studied using the method of
60: Zakharov transformations as detailed in \cite{kontorovich2001,CRZ1}. For $d<2$
61: and $m \gg m_0$, $\lambda_R(m) = 2 \pi \epsilon (J/D)^\frac{\epsilon}{d+2}
62: m^{-\frac{\epsilon}{d+2}}$. The constant flux solution obtained by
63: applying the Zakharov transformation to eq. (\ref{eq-RSE}) with this kernel is
64: \begin{equation}
65: N(m) = c_K\, \left( \frac{J}{D}\right)^\frac{d}{d+2} m^{-\frac{2d+2}{d+2}}.
66: \end{equation}
67: The constant, $c_K$, can be calculated exactly. However since we have only
68: calculated the renormalised kernel heuristically, it does not make sense to do
69: so at this point. Note that we recover the predictions of the K41 theory for the
70: large mass behaviour of the solution to the RSE.
71: Conversely, for small masses we see that $\lambda_R(m) = \lambda$. In this
72: limit, the Zakharov transformation gives the original mean field solution
73: for constant kernel aggregation,
74: \begin{equation}
75: N(m) =\sqrt{\frac{J}{4\pi\lambda}}m^{-\frac{3}{2}},
76: \end{equation}
77: as found already from our RG analysis. It is an easy task to show that both
78: the mean field and renormalised mean field density distributions are local
79: in the sense that the inertial range mass transfer in the stationary state
80: does not depend strongly on the source (or the sink which removes large masses
81: if one is present). Therefore both are physically realisable solutions and
82: both carry the same flux, $J$, of mass from small masses to large.
83: The complete stationary distribution for
84: constant kernel aggregation should therefore exhibit a crossover from the
85: mean field solution to the renormalised mean field solution at a mass, $m_c$
86: which is given by $(J/D)^\frac{\epsilon}{d+2}
87: m_c^{-\frac{\epsilon}{d+2}} \approx 1$. We note that this crossover from a
88: mean field spectrum to a fluctuation dominated one has been conjectured to
89: occur in other turbulent systems, in particular in wave turbulence
90: \cite{BNN2001}, although these systems do not readily lend themselves to such
91: systematic analysis.
92:
93: In $d=2$, analysis of the RSE equation allows us to
94: obtain the logarithmic correction to the mean field spectrum, again without
95: resort to R.G. arguments. Strictly speaking, the Zakharov transformation
96: technology of \cite{kontorovich2001,CRZ1} only works for homogeneous kernels
97: which leaves us with the question of what to do with the logarithm in eq.
98: (\ref{eq-renLambda2}). It turns out that the approach is easily adapted to
99: extract the logarithmic correction to the spectrum at large masses. We give a
100: brief outline here. In $d=2$, the renormalised interaction, eq.
101: (\ref{eq-renLambda2}), behaves for large masses as
102: \begin{equation}
103: \lambda_R(m) \sim \frac{1}{\ln\left(\frac{m}{m_0}\right)}.
104: \end{equation}
105: After application of the Zakharov transformations described in
106: \cite{CRZ1}, the stationary RSE, eq. (\ref{eq-RSE}), it becomes
107: \begin{eqnarray}
108: \nonumber 0 &=&
109: \int_0^{\infty}dm_1 dm_2 \left[ \ln\left(\frac{m}{m_0}\right)^{-1} N(m_1)\, N(m_2)\right.\\
110: \nonumber&-& \ln\left(\frac{m^2}{m_2m_0}\right)^{-1} N(m)\, N(\frac{mm_1}{m_2})\, \left(\frac{m}{m_2}\right)^2\\
111: \nonumber&-& \left.\ln\left(\frac{m^2}{m_1m_0}\right)^{-1} N(m)\, N(\frac{mm_2}{m_1})\, \left(\frac{m}{m_1}\right)^2\right]\\
112: & &\delta(m-m_1-m_2).
113: \end{eqnarray}
114: Let us now look for a solution of the form $N(m)=c_K\, \ln(m/m_0)^y m^{-x}$.
115: This substitution yields a rather messy expression which we analyse by
116: introducing new integration variables, $\mu_1$, $\mu_2$ defined by
117: $m_1=m\mu_1$, $m_2=m\mu_2$ and then expanding the resulting expression as a
118: power series in $\ln (m/m_0)^{-1}$. After some algebra one obtains :
119: \begin{eqnarray}
120: \nonumber 0 &=&
121: c_K^2 m^{2-2x} \int_0^{\infty}d\mu_1 d\mu_2 (\mu_1\mu_2)^{-x}\left[ \ln\left(\frac{m}{m_0}\right)^{2y-1} \right. \\
122: \nonumber&-& \ln\left(\frac{m}{m_0}\right)^{2y-1} \mu_1^{2x-2} - \ln\left(\frac{m}{m_0}\right)^{2y-1} \mu_2^{2x-2}\\
123: &+& \left.{\rm O}\left( \ln\left(\frac{m}{m_0}\right)^{-1}\right)\right]\ \delta(1-\mu_1-\mu_2).
124: \end{eqnarray}
125: It is clear that the leading logarithms cancel out asymptotically as
126: $m\to\infty$ if we choose $y=1/2$. The integrand then vanishes asymptotically
127: for $x=3/2$ as in the usual mean field case. The renormalised Kolmogorov
128: spectrum in $d=2$ for large masses is therefore
129: \begin{equation}
130: P(m) = c_K \sqrt{\ln \left(\frac{m}{m_0}\right)}\, m^{-\frac{3}{2}},
131: \end{equation}
132: as found from the RG analysis of section \ref{sec-RGdeq2}.
133:
134: Let us close this section by discussing the connection between the
135: renormalised Smoluchowski equation and the Smoluchowski approximation used
136: in \cite{krapivsky1994} to study the kinetics of heterogeneous annihilation.
137: The essence of the argument used in \cite{krapivsky1994} is as follows. Consider
138: a heterogeneous system of annihilating particles where the reactants have a
139: continuous distribution of diffusivities with the slower particles being less
140: probable than the faster ones. Now consider the reaction
141: between particles with diffusivity $D$ and ``slower'' particles with
142: diffusivity $D^\prime < D$. Since the slower particles are rare we can
143: neglect reactions between slow particles. We therefore estimate the effective
144: reaction rate for the slow particles by considering each slow particle to be
145: stationary in a uniform background cloud of faster particles and calculating
146: the diffusive flux of fast particles reaching the slow particle from the
147: background cloud. One estimates the effective reaction rate in $d$ dimensions
148: at time $t$ to be $(D+D^\prime)^{d/2}t^{-1+d/2}$. Substitution of this
149: effective reaction rate in the mean field rate equation constitutes the
150: Smoluchowski approximation
151: which greatly improves the estimate of the asymptotic decay rate of the
152: particle density in $d<2$. In the case at hand, the dynamics naturally generates a distribution of particle masses in which heavy particles are
153: much rarer than the lighter ones. So, although in our model all particles
154: have equal diffusivity, $D$, in the reference frame of a heavy particle we can
155: consider the main interaction to be with a uniform background cloud of
156: light particles. Thus one can envisage an effective reaction rate for
157: the large mass particles of $(2D)^{d/2}t^{-1+d/2}$. However in our
158: case the mass flux into the system is a constant, $J$, so that the system
159: reaches a stationary state in which the natural unit of time for particles of
160: mass $m$ is
161: $t \sim (J^{-1}D^{-d/2} m)^{2/(d+2)}$. Thus the time dependent effective
162: reaction rate of \cite{krapivsky1994} should be replaced in the stationary
163: state of the constant kernel aggregation problem with a mass dependent
164: reaction rate $\lambda_R(m) \sim D^{d/2} (J D^{d/2})^{-(d-2)/(d+2)}
165: m^{(d-2)/(d+2)}$ which we recognise as the large mass behaviour of the
166: renormalised kernel, eq. (\ref{eq-renLambda}). Thus the large mass asymptotics of the
167: RSE corresponds to the correction of the mean field rate equation by
168: the Smoluchowski approximation. The field theory approach with which we
169: derived the original results allows unambiguous identification of the set of
170: diagrams which have been summed to give this approximation.
171:
172: