1: %Analysis and numerics for the self-similar ode
2:
3: \appendix
4: \section{Analysis of the Self-similarity Equation and the Critical
5: Front Speed} \label{sec-ode}
6:
7: In this appendix we shall analyse the similarity equation
8: (\ref{eq-ss_dke}) further. We wish to check that it does indeed
9: admit solutions which reproduce the critical behaviour which we
10: have ascribed to the solutions of its antecedent PDE in section
11: \ref{sec-margin}. Specifically, we are interested in solutions
12: which match the $F(\eta) \sim (1-\eta)^{3/2}$ singularity at
13: $\eta=1$ to a power law solution, $F(\eta) \sim \eta^{-x}$ as
14: $\eta\to 0$.
15:
16: Let us write (\ref{eq-ss_dke}) in the form
17: \begin{equation}
18: \eta^{-1+d/\alpha}\left(xF+\eta\dd{F}{\eta}\right)=
19: (2(x-x_0)-\kappa_d)\ \ddd{ }{\eta}\left(\eta^sF^4\ddd{
20: }{\eta}\left(\frac{1}{F}\right)\right),
21: \end{equation}
22: and look for solutions of the form $F(\eta)=A\eta^{-y}$.
23: Substituting this form we obtain
24: \begin{eqnarray}
25: \nonumber(-y+x)\ \eta^{\frac{d}{\alpha}-1-y} &=&
26: 3A^2y(y-1)(x_0-y)(3x_0-3y-1)\\
27: \label{eq-ss_dke_2} & &(2(x-x_0)-\kappa_d)\ \eta^{3 x_0-3y-2}
28: \end{eqnarray}
29: We see that we can have the following solutions $y=x$ where $x$
30: takes one of the following values,
31: \begin{eqnarray}
32: x&=&0\\
33: x&=& 1\\
34: x&=& x_0\\
35: x&=& x_0-1/3\\
36: \end{eqnarray}
37: The first pair are the thermodynamic spectra, the second pair are
38: the K-Z energy and particle spectra respectively. A fifth special
39: value of $y$ is
40: \begin{equation}
41: \label{eq-singular_x}x= x_0+\kappa_d/2.
42: \end{equation}
43: What we observe from numerical solution of the O.D.E.
44: (\ref{eq-ss_dke_2}), described below, is as follows. If we choose
45: a value of $x$ which is not $x_c=x_0+\kappa_d/4$, then the
46: solution, $\eta^{-x}$, makes a transition near $\eta=0$ to the
47: state $\eta^{-y}$, where $y=x_0+\kappa_d/2$. This behaviour near
48: $\eta=0$ balances the leading order divergences on both sides of
49: equation (\ref{eq-ss_dke_2}) as $\eta\to 0$. It might be noted,
50: although it may have no relevance to the problem under
51: consideration, that this value for the exponent, $y$, leads to
52: front dynamics, $\ws(t) = (\ts-t)^b$ with zero $b$. This spectrum
53: was never observed in the solutions of the P.D.E.
54: %
55: %the final spectrum is a singular solution of the equation since
56: %the coefficient on the LHS of (\ref{eq-ss_dke}) diverges for this
57: %value of x. For example $y \to x_0+\kappa_d/2$ as $\eta \to 0$. It
58: %turns out that the final form, (\ref{eq-singular_x}), which is the
59: %only one of the five spectra to balance the order of the
60: %divergences on both sides of equation (\ref{eq-ss_dke_2}) as $\eta
61: %\to 0$, is the generic singularity of this equation near $\eta=0$.
62: %However it is {\it not} the spectrum we expect to see physically.
63: %From (\ref{eq-b}) we see that the front speed is infinite for this
64: %choice of $x$.
65:
66: To find solutions of (\ref{eq-ss_dke}) which do not exhibit the
67: $x_0+\kappa_d/2$ scaling as $\eta\to0$, we decided to perform a
68: set of numerical experiments. Let us write out the RHS explicitly
69: so that we can see exactly the equation which we wish to solve:
70: \begin{eqnarray}
71: \nonumber\frac{1}{2(x-x_0)-\kappa_d}\
72: \eta^{-1+\frac{d}{\alpha}}& &\left(xF+\eta\dd{F}{\eta}\right)= -\eta^sF^2\ddddd{F}{\eta} - 2s\eta^{s-1}F^2\dddd{F}{\eta}\\
73: \label{eq-ss_dke_3}& &+2\eta^sF\left(\ddd{F}{\eta}\right)^2 +
74: 8\eta^s\left(\dd{F}{\eta}\right)\ddd{F}{\eta}\\
75: \nonumber& &+4s\eta^{s-1}\left(\dd{F}{\eta}\right)^3
76: +4s\eta^{s-1}F\dd{F}{\eta}\ddd{F}{\eta}\\
77: \nonumber& &+2s(s-1)\eta^{s-2}F\left(\dd{F}{\eta}\right)
78: -s(s-1)\eta^{s-2}F^2\ddd{F}{\eta}
79: \end{eqnarray}
80: The problem with integrating (\ref{eq-ss_dke_3}) on a computer is
81: that for generic initial conditions, the strong power law
82: dependences of the right hand side on the independent variable
83: render the numerics very susceptible to round-off error and
84: numerical instability. For example, to study the system with
85: $\gamma=7/2$, $\alpha=1/2$, $d=1$, to which we gave a lot of
86: consideration in section \ref{sec-pde_num} due to its relatively
87: large anomaly, we are required to take $3x_0+2=22$.
88:
89: To get around this difficulty, and to aid visualisation of the
90: global properties of the equation, we make the following change of
91: variables :
92: \begin{eqnarray}
93: \nonumber F(\eta) &=& \eta^af(\tau)\\
94: \label{eq-autonomise} \dd{F}{\eta}&=& \eta^{a-1}g(\tau)\\
95: \nonumber \ddd{F}{\eta}&=& \eta^{a-2}h(\tau)\\
96: \nonumber \dddd{F}{\eta}&=& \eta^{a-3}k(\tau),
97: \end{eqnarray}
98: where $\tau=\log(\eta)$. By choosing
99: \begin{equation}
100: \label{eq-a} a=\frac{1}{2}\left( \frac{d}{\alpha} -s +3 \right) =
101: -x_0-\frac{\kappa_d}{2} ,
102: \end{equation}
103: we can cancel all the power dependence from the system and
104: eventually recast (\ref{eq-ss_dke_3}) as the the following
105: autonomous fourth order system:
106: \begin{eqnarray}
107: \nonumber \dd{f}{\tau} &=& g-af\\
108: \nonumber \dd{g}{\tau} &=& h-(a-1)g\\
109: \label{eq-auto_system} \dd{h}{\tau} &=& k-(a-2)h\\
110: \nonumber f^2\dd{k}{\tau} &=& -\frac{1}{2(x-x_0)-\kappa_d}\ \left( g+xf\right) -(a+2s-3)f^2k + 2fh^2\\
111: \nonumber & &+8g^2h + 4sg^3 + 4sfgh +2s(s-1)fg^2 -s(s-1)f^2h.
112: \end{eqnarray}
113: Only the region, $f\geq 0$ makes physical sense since $n_\w$, and
114: hence $F$, cannot be negative. This system is singular on the
115: hyperplane $f=0$. This system is much easier to integrate
116: numerically and has the added advantage that we can determine the
117: presence of fixed points which are not obvious in the original
118: differential equation. Let us determine these points. It is
119: obvious that the RHS of (\ref{eq-auto_system}) has a trivial zero
120: at $(f,g,h,k)=(0,0,0,0) = O$ but this is clearly a singular point
121: due to the factor of $f^2$ on the LHS. A second pair of nontrivial
122: (and nonsingular) fixed points can be shown after quite a bit of
123: algebra to exist at the points $P_{\pm}=(f_0, af_0, a (a-1)f_0,
124: a(a-1)(a-2)f_0)$, where
125: \begin{equation}
126: f_0=\pm\left(18a(a+1)(a+(s-2)/3)(a+(s-3)/3)\right)^{-\frac{1}{2}}.
127: \end{equation}
128: We are naturally only interested in the point $P_+$ since $P_-$
129: lies in the negative $f$ region. The factors $(a+(s-2)/3)$ and $
130: (a+(s-3)/3)$ are interesting for the following reason. If we
131: substitute back in $s=3x_0+2$ and the value of $a$ from
132: (\ref{eq-a}), these two factors are simply $-\kappa_d/2$ and
133: $-\kappa_i/2$ respectively. At the transition points between
134: finite and infinite capacity the fixed point runs away to infinity
135: which suggests that it has a central role to play in organising
136: the critical solution in the finite capacity case.
137:
138:
139: Let us now look for a numerical solution of this system which
140: mirrors the solutions of the differential kinetic equation. All
141: the simulations in this section were done using the transformed
142: system, (\ref{eq-auto_system}), for which a standard
143: out-of-the-box adaptive Runge-Kutta routine seemed to work fine.
144: Suppose we want a solution for the wake of the form $F(\eta) \sim
145: \eta^{-x}$ as $\eta\to 0$. In the new variables this is equivalent
146: to demanding
147: \begin{eqnarray*}
148: f(\eta) &\sim& A\eta^{-a-x}\\
149: g(\eta) &\sim& -x f(\eta)\\
150: h(\eta) &\sim& x(x+1)f(\eta)\\
151: k(\eta) &\sim& -x(x+1)(x+2)f(\eta)
152: \end{eqnarray*}
153: as $\eta\to 0$ or as $\tau\to \-\infty$. If $x<-a=x_0+\kappa_d/2$,
154: which is the case, then we observe that the wake is described by a
155: trajectory for which $(f,g,h,k) \to (0,0,0,0)$ as $\tau\to
156: -\infty$. The wake is the singular point, $O$, of the system,
157: (\ref{eq-auto_system}). As we approach $\eta\to 1$ we must
158: reproduce the front structure described in section \ref{sec-ss}.
159: Thus we require a trajectory for which $(f,g,h,k) \to F$ as
160: $\tau\to 0$, where $F= (0,0,\infty,\infty)$. We require a
161: trajectory which links these two singular points.
162:
163: In our numerical simulations we rescaled the variables as follows,
164: \begin{eqnarray*}
165: f &\to& \frac{f}{f_0}\\
166: g &\to& \frac{g}{af_0}\\
167: h &\to& \frac{h}{a(a-1)f_0}\\
168: k &\to& \frac{k}{a(a-1)(a-2)f_0}
169: \end{eqnarray*}
170: This maps the point $P_+ \to (1,1,1,1)$ for ease of visualisation.
171: The only difference is that since $a<0$, the signs of $g$ and $k$
172: are swapped so that the wake part of the solution must now
173: approach $O$ from the direction $(0^+,0^+,0^+,0^+)$ and the front
174: tip, $F$, is now at $(0^+, 0^+, \infty, -\infty)$.
175:
176: It turns out to be difficult to find a trajectory linking $F \to
177: O$. We performed a series of experiments integrating backwards
178: from $1-\eta = \varepsilon$, with $\varepsilon << 1$, towards
179: $\eta=0$. Because we cannot specify initial data exactly at the
180: singular point, $F$, we were required to manually tune the initial
181: conditions quite a bit in order to to reproduce the $F(\eta) \sim
182: (1-\eta)^{3/2}$ structure near the tip. The remaining adjustable
183: parameter is the equation is $x$. We again choose to study the
184: case $\gamma=7/2$, $\alpha=1/2$, $d=1$. Two generic types of
185: trajectory emerge as we vary $x$. We visualise these trajectories
186: in 4 dimensional phase space as a pair of projections of the
187: actual trajectory onto the $(f,g)$ and $(h,k)$ planes
188: respectively, hence the apparent intersections.
189:
190: \begin{figure}[ht]
191: \begin{center}
192: \epsfig{file=traj3.eps,height=2.5in,angle=0}
193: \end{center}
194: \caption{\label{fig-traj_xc_plus}Trajectory in $(f,g,h,k)$ space
195: for $x=7.56275$, which is less than the critical value.}
196: \end{figure}
197:
198:
199: \begin{figure}[ht]
200: \begin{center}
201: \epsfig{file=traj1.eps,height=2.5in,angle=0}
202: \end{center}
203: \caption{\label{fig-traj_xc_minus}Trajectory in $(f,g,h,k)$ space
204: for $x=7.56975$, which is greater than the critical value.}
205: \end{figure}
206:
207: \begin{figure}[ht]
208: \begin{center}
209: \epsfig{file=traj2.eps,height=2.5in,angle=0}
210: \end{center}
211: \caption{\label{fig-traj_xc}Trajectory in $(f,g,h,k)$ space for
212: $x=7.56875\approx x_c$. Since it is impossible to specify the data
213: exactly, if we integrate for long enough in $\tau$ this trajectory
214: eventually deflects away from $O$. }
215: \end{figure}
216:
217: \begin{figure}[ht]
218: \begin{center}
219: \epsfig{file=wake.eps,height=2.5in,angle=0}
220: \end{center}
221: \caption{\label{fig-asymp_slopes} Plots of the $F(\eta)$
222: associated with the trajectories in figures
223: \ref{fig-traj_xc_plus}, \ref{fig-traj_xc_minus} and
224: \ref{fig-traj_xc} after converting from the $(f,g,h,k)$ and $\tau$
225: variables back to $F$ and $\eta$ variables. The scales are
226: log-log.}
227: \end{figure}
228: Figure \ref{fig-traj_xc_plus} shows what happens when $x$ is
229: slightly less than $x_c$. The trajectory leaves the front tip and
230: heads towards the fixed point, $P$, but deflects to the right and
231: is quickly attracted onto the line corresponding to $F(\eta)\sim
232: \eta^{-x_0-\kappa_d/2}$. Figure \ref{fig-traj_xc_minus} shows the
233: corresponding trajectory when $x$ is slightly greater than $x_c$.
234: This time the trajectory deflects to the left as it approaches the
235: fixed point, $P$, and is attracted towards the singular point,
236: $O$. Before it reaches there however, it is deflected and is
237: rapidly attracted back onto the $F(\eta)\sim
238: \eta^{-x_0-\kappa_d/2}$ solution again. The transition point
239: between these two trajectories is illustrated in figure
240: \ref{fig-traj_xc}. Further analysis is required to determine the
241: exact nature of the critical trajectory. It is not clear whether
242: it will eventually deflect away from $O$ for sufficiently small
243: $\eta$. Of course, in practical terms there is a limit placed on
244: the extent of the scaling region by the left boundary of the
245: inertial range which cannot extend all the way to 0 in a real
246: experiment.
247:
248: The system is very sensitive to the value of $x$. The values of
249: $x$ for the trajectories shown differ only in the fourth decimal
250: place. In figure \ref{fig-asymp_slopes}, the form of the function
251: $F(\eta)$ is shown for the three cases discussed above, after
252: transforming back from the $(f,g,h,k)$, $\tau$ variables. The
253: integration was again done for the case $\gamma=7/2$,
254: $\alpha=1/2$, $d=1$. Despite the small difference in the values of
255: $x$ in the equation the difference in asymptotic slopes is $7.56$
256: for the critical slope versus $8.5$ for the other two.
257: