We propose new identities for dynamic subgrid modeling in large-eddy simula
tion involving an explicit filter and its inverse. Exact defiltering of a c
lass of numerical realizations of the top-hat filter is developed. The appr
oach is applied to large-eddy simulation of the temporal mixing layer. Smag
orinsky's model is adopted as base model and the results are compared to th
e standard dynamic eddy-viscosity model as well as to filtered DNS (direct
numerical simulation) results. The difference between the results of the tw
o models for the present application is found to be quite small. This is ex
plained by performing a sensitivity analysis with respect to the dynamic co
efficient, which hints towards a "self-restoring" response underlying the o
bserved robustness of the physical predictions. Using DNS data the validity
of the assumption that the model coefficients are independent of filter wi
dth is tested and found to favor the inverse modeling procedure. The comput
ational effort of the dynamic inverse model is 15% smaller than of the stan
dard dynamic eddy-viscosity model. (C) 1999 American Institute of Physics.
[S1070-6631(99)00212-3].