I tried to avoid mathematical expressions (I leave that for the mathematicians). Let try to give an equation here.
Consider Xn1 and Xn2 as two points in phase space with distance ||Xn1 – Xn2|| = δ0 << 1
Let δΔn be the distance between the trajectories of these two points some later time Δn
Then:
δΔn = || Xn1 +Δn - Xn2+ Δn||.
The Lyapunov exponent 𝛌 is determined by:
δΔn ≈ δ0e𝛌Δn,
With δΔn << 1 and Δn>> 1


This equation remains valid provided that the limit converges like you have rightly said and then there will be exponential divergence if 𝛌 is positive.
Other methods such has phase space method, false nearest neighbors; correlation dimension etc. will also be used on the data set I am working on.
It is however not all close trajectories that will eventually have this (that is the essence of false nearest neighbor) and so the consideration is only on those that fulfill this requirement.


I think the challenge here is that I am working with real data, i.e. (actual scalar measurement as most physics research using nonlinear dynamical tool), so in a way, we have to use a proper embedding dimension that fits the data. Then we also consider modeling the physical system.
Perhaps when I finish collecting my data and start my analysis, it would be much more clearer the angle from which the TISEAN software will be computing and searching for chaos in the data set.
Thank you and I look forward to our discussion on the chaos this week. I really want to have a robust research and I can use any help I can get.


Edit: to add, I agree that the estimated finite dimension and Lyapunov exponents on their own cannot be a suitable prove for strict nonlinear determinism if a clear scaling region cannot be established. One suggested way to avoid this is to test first if the data could be explained by a linear model. Some literature suggested the use of surrogate data