Convection in a box 
J.L. Henshaw (aka P.F.)- experiments in 1978, Denver   Other Airwork,

1. A single 2D vortex ring - displaying apparent contradictions
2. Detailed diagrams of vortex flow - the winding & wrapping of layers
3. Smoke pool - with branching columns of clear air
4. Very strange indeed - a clear confusion between up and down

One thing these experiments demonstrate is Organized Molecular Motion , fluid states that store and release energy.   This is clearly evident from experiment 4, and then can help explain experiments 1, 2 & 3.   There is a simpler way to perform experiment 4 that usually works.   On a table in a quiet place stand a burning cigarette on its end and patiently watch.   After a while smoke will be seen pouring down the outside to form a pool of smoke around the base.   Wisps will then sometimes be seen spontaneously rising from the smoke pool, storage and release, QED.   Perhaps several repetitions will be needed for you to believe what you see.

	With an opening in the bottom of a ~1/4" deep box, a burning incense stick is inserted and the  rising smoke generates a convection loop.   At first the flow may look like a large circle, two 'D' shaped spirals back to back, nearly filling the chamber (left).   Curiously the none of the smoke rises to spread out at the top.    After a few seconds, with a continuing introduction of heat and smoke, the double spiral curiously becomes smaller and denser (center).    Both of these behaviors seem wrong when you think about them.   First, the smoke should rise and spread out at the top, like in open air, but it doesn't.   Second the vortex should get larger the more heat and smoke are added, not smaller.  After removing the incense stick and covering the hole in  the bottom of the box the 2D vortex ring continues, firmly sitting on the bottom of the box (right).
	When you try it with a 1/2" deep box the self-confined cell does not develop.   Smoke goes to the top and spreads out (in a 6"x6" box at least) just like you'd expect it to.    The image at the left shows a recent test of both 1/2" and 1/4" depth containers, and using color changing heat sensitive film on the back surface, to help show where the heat goes.   In the 1/4" box, the heat introduced almost seems to disappear.  Of course it does not, but there is an increase in density in the confined smoke cell and no indication of a corresponding heat loss.   There are surely other things to be resolved too, but I think this is unambiguous evidence of fluid-partical gels.  That could explain why something hot acts heavy, a problem observed in all these studies.  The key observation is that the smoke comes to circulate entirely within itself.  The rising plume flows down from above and is then drawn in at the bottom and up at the center.  It's possible to interpret this as smoke being pulled downward on the outside by the central current.   A possibly important observation is that in an 1/8" deep box the smoke cell that forms is about the size of a quarter, very small, and the circulation is so confined that the incense stick is snuffed out almost immediately.   In the 1/4" deep box the ember can go out for lack of oxygen as well, though not for some time.

What's wrong with the idea suggested by some, that the walls drain enough heat from the smoke to make it heavier than air, is that :
1) the effect increases over time, rather than slowing as the walls get hotter, and
2) the smoke would spread out over time, not compress, seeking the cooler portions of the surface
3) Plexiglas has a very low specific heat and the smoke is not in circulating contact with it.

Others have suggested that it's just the proximity of vertical surfaces.  If the same smoke source is held close to a similar surface in quiet open air, smoke is seen easily drifting upward even within the near ('sticky' ~1/32") boundary layer where all flow is laminar.  The proximity of the surface alone is not what makes the smoke turn and go down, but proximity must have something to do with it, since proximity is a controlling variable.

After removing the incense stick and covering the hole at the bottom the 2D vortex ring  settles into the shape at the right, similar to an apple upside down.  This continues without  outwardly changing shape or size for a very long time.   It seems to just spiral slower and slower until the smoke particles settle to the bottom of the chamber.

top

The details of the spiral path of circulation can be seen after generating a double spiral, removing the smoke source, and covering the bottom hole.   Briefly removing the cover on the bottom hole introduces a pulse of clear air to be drawn up into the convection vortex and trace the circulation shape..

The actual movement of the layers is quite simple, after moving up the center and splitting at the top the new material begins an infinite regression of stretching and wrapping around itself while also migrating outward toward the outer surface.  Material from the outer surface may then be drawn back in the bottom to start the process over.   At first it looks like circular motion, but it is composed entirely of winding.

top

Some interesting shapes result from blowing smoke into the box.   Do it gently so as not to mix it by forced turbulence.  Then momentarily open a hole in the bottom to let in cool room air.   Oddly, the smoke sits on the bottom and the 'cool' air rises through it, displaying a number of fascinating circulation patterns.

Introducing an incense stick near the edge of the 1/4" deep box stimulates a single 'D' shaped loop, with the same characteristics as the more normal double 'D' shaped shrinking convection cell.    After removing the incense stick and watching it sit there spinning by itself, crammed into the corner instead of spreading out either along the top or bottom, seems very strange!

One of the strangest things found were these two examples of 'heavy smoke' spontaneously becoming buoyant.   In both cases warm smoke was blown into the boxes gently so as not to turbulently mix, and was found to first settle in a pool on the bottom.   In the first example the box was then sealed and left alone.   In the second, openings were made at the top and bottom, in completely still surroundings, and the smoke was allowed to gently drain out the bottom with room air entering at the top. The circulation shown in the first example developed after a few minutes.   Instead of remaining motionless the smoke pool took on the shape shown, oddly sloped, necking upward on one side and flowing toward the top of the box in a loose, lopsided, double spiral.   There are quite a long list of things wrong with this one!   What explains both this and the following example seems impossible, energy storage and release in a fluid in some form other than heat. In the second example the smoke pool in the box flows gently out over the table surface.  It forms a thick spreading film as if a puddle of syrup.   On occasion (~1/10 or so), and clearly only when the room air is quite still, little vigorous rising currents develop.   They're not normal convection cells, but twisting slivers of smoke.  Warm smoke should not have settled in a pool in the first place, and especially not spontaneously rise again from it.