To compute the Lyapunov exponents we followed this procedure.  From the output of swift_lyap2 we read the data in lyap2.out (time and distances) and construct a serie of the form:

t    -    ln(d(t)/d(0))

0.0    0.0
1.0    ln(d(1)/d(0))
2.0    ln(d(2)/d(0))
....

Then we perform  a linear least-square fit on the curve [t,ln(d(t)/d(0))].    Since d(t) ~ d(0)*exp(L*t) => ln(d/d0) ~ L*t , the slope should be equal to L. If your data set is about 10*(1/L) long, this should give a quite
accurate computation. If your data points reach numbers as high as 10^300 and overflow, it is certain that you will have a quite good estimate with the first part of the series  (from an email of Kleomenis Tsiganis).

Here we the results of this procedure for the low-resolution survey.   We report a contour plot of the log10 of Lyapunov times, without the median filtering.

In the following figures we report a) Lyapunov times (log10) after the use of the median filter, b) Lyapunov exponents, and c) the results of the Frequency analysis method for the same survey.
 
 

In the following plots we report the results of simulations that used analytical solutions from the Bretagnon model.  On the left there are the values of Lyapunov times, and on the right the analytical solution used for
Saturn.   Note that when only secular terms are considered Lyapunov times are much longer, and we do not see the feature with larger lyapunov times associated with the most chaotic region found with the FAM.
This seems to suggest that this feature might be associated with the great-inequality terms.