The proper orthogonal decomposition (P.O.D.) is applied to the flow in a di
fferentially heated cavity. The fluid considered is air, and the aspect rat
io of the cavity is 4. At a fixed Rayleigh number, P.O.D. empirical functio
ns are extracted, and low-dimensional models are built and compared to the
numerical simulation. Generally speaking, low-D models provide a coarse pic
ture of the flow, which is also quick, cheap, and easy to understand. They
can help pinpoint leading instability mechanisms. They are potentially key
players in a number of applications such as optimization and control. Our g
oal in this study is to determine how well the flow can be represented by v
ery low-dimensional models. Two moderately complex situations are examined.
In the first case, at some distance from the bifurcation point, the dynami
cs can still be reduced down to two modes, although it is necessary to acco
unt for the effect of higher-order modes in the model. In the second case,
farther away from the bifurcation, the flow is chaotic. A ten-dimensional m
odel successfully captures the essential dynamics of the flow. The procedur
e was seen to be robust. It clearly illustrates the power of the P.O.D. as
a reduction tool. (C) 2001 American Institute of Physics.