1: \vbox{
2: \centerline{\epsfysize=4truein\epsfbox{E0.1.ps},
3: \epsfysize=4truein\epsfbox{E10.ps}}
4: \smallskip
5: \centerline{\bf $\epsilon=0.1$ \hskip 2 in $\epsilon = 10$}
6: \smallskip
7: \centerline{\bf Figure 1. Trajectories with fixed $\epsilon$.}
8: \bigskip}
9:
10: \section{Numerical Results and Qualitative Analysis}
11:
12:
13: For any fixed $\epsilon$ and $\lambda$, and
14: given $a_1$ and $b_2$, one can in principle integrate the
15: differential equations out to $r=1$. In practice, numerical errors due to
16: the discretization of the interval $[0,1]$ can be very bad near the origin,
17: due to the singular nature of the ODE system there. A better method is
18: to use the power series \eqref{powers} in a neighborhood of the origin
19: and to numerically integrate from there. In a discretization of $10,000$
20: points, we use the power series out to $r=0.01$, or 100 lattice spacings
21: from the origin.
22:
23: In this way we get a pair $(\gamma(1), \varphi(1))$ for each $(a_1, b_2)$.
24: Using Newton's method, we then find values of $(a_1,b_2)$ such that
25: $(\gamma(1), \varphi(1))=(-1/2, 1)$. Table 1 lists the correct values
26: of $a_1$ and $b_2$ for several values of $\epsilon$ and $\lambda$.
27:
28:
29: The resulting functions $\varphi(r)$ and $\gamma(r)$ are sketched in
30: Figures 1 and 2. Figure 1 shows the functions for different values of
31: $\lambda$ and $\epsilon$ fixed at 0.1 or at 10. Figure 2 is similar, only
32: with $\lambda$ fixed and $\epsilon$ variable.
33: In each case the positive function is $\varphi$
34: and the negative function is $\gamma$.
35:
36:
37:
38: \vbox{
39: \centerline{\epsfysize=4truein\epsfbox{L0.ps}, \epsfysize=4truein\epsfbox{L30.ps}}
40: \smallskip
41: \centerline{\bf $\lambda=0$ \hskip 2 in $\lambda = 30$.}
42: \smallskip
43: \centerline{\bf Figure 2: Trajectories with $\lambda$ fixed.}
44: \bigskip}
45:
46: %\bigskip
47: \begin{table}\vbox{
48: \begin{tabular}{|r|r|r|r|} \hline
49: \hfil $\epsilon$ \hfil & \hfil $\lambda$ \hfil & \hfil $a_1$ \hfil &
50: \hfil $b_2$ \hfil \\ \hline
51: 0.1 & 0 & 2.82909077 & -4.47460232 \\ \hline
52: 0.1 & 1 & 3.14773551 & -4.92072556 \\ \hline
53: 0.1 & 3 & 3.62692766 & -5.57110938 \\ \hline
54: 0.1 & 10 & 4.62892407 & -6.81947999 \\ \hline
55: 0.1 & 30 & 6.19274693 & -8.46474894 \\ \hline
56: 0.3 & 0 & 2.01904955 & -1.88549902 \\ \hline
57: 0.3 & 1 & 2.26118176 & -2.04994984 \\ \hline
58: 0.3 & 3 & 2.66517994 & -2.31673622 \\ \hline
59: 0.3 & 10 & 3.59550462 & -2.86817001 \\ \hline
60: 0.3 & 30 & 5.12510342 & -3.58374045 \\ \hline
61: 1 & 0 & 1.67098122 & -1.02894746 \\ \hline
62: 1 & 1 & 1.85973704 & -1.07504639 \\ \hline
63: 1 & 3 & 2.19572981 & -1.15577783 \\ \hline
64: 1 & 10 & 3.04898441 & -1.34041824 \\ \hline
65: 1 & 30 & 4.55384341 & -1.58910470 \\ \hline
66: 3 & 0 & 1.57081044 & -0.80615986 \\ \hline
67: 3 & 1 & 1.74184236 & -0.82078859 \\ \hline
68: 3 & 3 & 2.05156143 & -0.84695210 \\ \hline
69: 3 & 10 & 2.86750186 & -0.90924487 \\ \hline
70: 3 & 30 & 4.35776101 & -0.99551235 \\ \hline
71: 10 & 0 & 1.53622287 & -0.73146686 \\ \hline
72: 10 & 1 & 1.70099654 & -0.73576432 \\ \hline
73: 10 & 3 & 2.00102288 & -0.74350147 \\ \hline
74: 10 & 10 & 2.80198139 & -0.76219107 \\ \hline
75: 10 & 30 & 4.28571713 & -0.78838676 \\ \hline
76: \end{tabular}}
77: \bigskip
78: \caption{Taylor coefficients
79: $(a_1,b_2)$ for various values of $(\epsilon, \lambda)$.}
80: \end{table}
81:
82: %\bigskip
83:
84: From these figures several qualitative features are clear. Although
85: $\varphi$ depends significantly on both $\epsilon$ and $\lambda$,
86: $\gamma$ is practically independent of $\lambda$, especially when
87: $\epsilon$ is large. The length scale on which $\gamma$ changes from
88: 0 to $-1/2$ is the smaller of $\sqrt{\epsilon}$ and 1. The length
89: scale on which $\varphi$ changes from 0 to 1 is the smallest of
90: $\sqrt{\epsilon}$, $1/\sqrt{\lambda}$, and 1. Thus changing $\lambda$
91: has the greatest effect when $\lambda$ is greater than 1, while
92: changing $\epsilon$ has the greatest effect when $\epsilon<1$.
93:
94: %{\tt There really isn't much more to say here. I can generate as many
95: %plots as you like, but they're not particularly interesting.}
96:
97: The source code for these numerical results can be obtained from the
98: authors.
99:
100:
101: