Can the conductance of an adiabatic ballistic constriction be lower than 2e²/h?

We have performed four-terminal conductance measurements of a one-dimensional (1D) channel in which it is possible to modulate the potential profile using three overlaying finger gates. In such a 1D ballistic structure we have observed, {\em for the first time,} that the conductance steps show a gradual decrease from $2e^2/h$ to $0.97 \times 2e^2/h$ with increasing negative finger gate voltage in a short, clean 1D constriction. We suggest this phenomenon is due to differing shifts of 1D subbands with changing spilt-gate voltage. Both a simple analytical estimate for an adiabatic constriction and, realistic modeling of the device, give the same magnitude of the conductance decrease as observed in our experiments.