Is there a passive molecular flow 'diode'?

[cmb]

During high vacuum pumping, one sucks viscous gas out of a chamber down to just above the molecular flow density where viscous flow becomes absent, and then may employ a different sort of pump to start pulling in the remaining molecules: Molecular flow.

I have been wondering for a while if there is some geometry of pipe, maybe of changing diameter or internal baffles, that would automatically and passively compress gas in the molecular flow regime. A bit like a Tesla valve is in the viscous regime.

I was thinking about this, this evening, and did a search for 'trumpet molecular flow diode'. What I was thinking was that if molecular flow bounces its way into a trumpet cone, surely it will end up tending in one direction. However I came across this; https://www.comsol.com/paper/download/182051/eisenschmid_abstract.pdf

which shows a straight cone has the theoretical capacity to do this, but it actually works the other way around, the converging walls of the cone representing a block to molecular flow like a prism to light, or like a charged particle in a magnetic cusp.

So, are there any 'passive molecular flow pump' geometries used, do they exist, do they work well?

 

[Paul Colby]

Are molecule-wall collisions elastic? I seem to recall they stick for some time and then break free flying off in some unrelated direction. This would make the transport calculation more involved.

 

[Paul Colby]

Two additional thoughts,

<general simulation skepticism>
1) I suspect that the simulation shown doesn't represent a long term average.

<general thermodynamic skepticism>
2) A pressure differential developed between regions allow one to extract mechanical energy from the thermal energy in the gas. This would violate the usual thermodynamic arguments against a free lunch. An example is a semiconductor diode which has no net current flow from thermal charge carrier motion.