I am trying to create an Lennard-Jones system with a ayer of liquid
approx 50 sigma deep sitting on top of a solid at the very bottom of
the simulation domain.

Despite my efforts the liquid layer always acquires a velocity upwards
away from the wall (i.e. the whole slab moves up as one)

if you integrate equations of motion with a finite time step,
you will accumulate kinetic energy of the interactions (basically,
because of the finite time step, you move up too high on the
repulsive branch of the potential and then pick up kinetic energy
in the opposite direction). in a bulk liquid this is random in and thus
dissipates. if you have a wall and too light an object or too steep
a potential, this may give your system a push away from it.
with a langevin thermostat you have an option to dissipate this
artificial momentum.

Dan,
I have simulated similar systems before. I used the 93 walls, and I found that it takes quite a bit of trial and error to get the proper offset. The LJ-93 sigma value should give you a first order approximation of what that offset should be. Due to the “bulk” liquid pushing on the surface, the equilibrium distance will be higher up on the LJ-curve (i.e. smaller than sigma). I would suggest running numerous quick simulations with different offsets to see which ones push the liquid away the least. A minimization won’t help because 1) you are not explicitly modelling the solid and 2) the liquid does not really have a minimum potential energy (or it does but it is highly degenerate).

If you put the wall at y=0 and use lj93,
then the minimum of the wall potential
is about 1 sigma away from the wall.
Particles at y=0 or 1/2 sigma will experience
a huge repulsive force, actually infinite
at 0. You can minimize with this potential
but make sure you read the fix wall doc
page about including the wall energy
being minimized.

If you use wall reflect there is no minimum.
Particles can get as close to y=0 as they
want w/out a repulsive force. Hence you
also can't do minimizations with this kind
of wall.