Why is the hot air rising?

I want to investigate heat transfer through a window and associated framing. I'd like the boundary conditions to be a room and an exterior with still, uniform temperature air (except as disturbed by the natural convection at the window). I couldn't think of how to set those conditions for an external study (I don't have a lot of experience with Flow) so I tried an internal flow study instead. I created a box with two compartments and made part of the separating wall my window. The rest of the box is assigned the predefined "Insulator" material, the window is a solid piece of predefined "Glass". I applied an enviroment pressure BC to all the interior faces of each compartment. For one compartment I defined the environment temperature to be 50F, for the other compartment I defined the temp to be 75F.

When I run the simulation, the air on the warm side (compartment defined as 75F) rises up through the compartment, dragging the cooled air at the boundary layer up with it. On the cold side, the air accelerates downward through the compartment, dragging warmed air from the boundary layer down with it. It isn't just a little drift either, it increases from roughly zero velocity at the top right corner to about 50 ft/sec in/sec at the bottom. I expected that a uniform environmental pressure would cause no bulk flow, and that buoyancy effects would drive some gentle motion in the opposite direction of what I'm getting.

I can't think of a reason for this behavior. I've run with two different meshes, two different "depths" (dimension of the box perpendicular to the glass pane). I even tried it 2-D. Same results. I checked that gravity is pointing the direction I intended (-Y). The behavior is roughly the same across the width of the window.

I will attach some images showing the results; looking at the two chambers and the window in cross section, looking at the warm face of the window, and an ISO showing both plots.

I have two questions:

(1) what could be causing this?

(2) how could I set this up better?

Thanks in advance,

Jake.

SolidworksFlow Simulation