The leading Lyapunov vector is computed as follows:

1) Start with an arbitrary perturbation of arbitrary size

2) Evolve it from
*t* to using the TLM

3) Repeat 2) for the succeeding time intervals

After a sufficiently long time , the perturbation
converges to the leading Lyapunov

vector. The direction of this vector is independent of
the initial perturbation, the

length of the time interval or the choice of norm of the
perturbation,

properties not shared by singular vectors. If during the
repeated application of

the TLM the LV becomes too large it may be scaled down to
avoid computational

blow up.

Additional LVs can be obtained by the same procedure,
except that after

each time step the perturbation has to be orthogonalized
with respect to

the subspace of the previous LVs, since otherwise all the
LVs would

converge to the leading LV.