<p>The classic approach to tsunami simulation by earthquake sources consists<br>of computing the vertical static deformation of the ocean bottom due to<br>the dislocation, using formalisms such as Mansinha and Smylie's [1971] or<br>Okada's [1985], and of transposing that field directly to the ocean's<br>surface as the initial condition of the numerical simulation.<br>We look into the limitations of this approach by developing a very<br>simple general formula for the energy of a tsunami, expressed as the<br>work performed against the hydrostatic pressure at the bottom of<br>the ocean, in excess of the simple increase in potential energy<br>of the displaced water, due to the irreversibility of the process.<br>We successfully test our results against the exact analytical solution<br>obtained by Hammack [1972] for the amplitude of a tsunami generated<br>by the exponentially-decaying uplift of a circular plug on the ocean<br>bottom. We define a "tsunami efficiency" by scaling the resulting energy<br>to its classical value derived, e.g., by Kajiura [1963]. As expected, we<br>find that sources with shorter rise times are more efficient tsunami<br>generators; however, an important new result is that the efficiency is<br>asymptotically limited, for fast sources, to a value depending on the<br>radius of the source, scaled to the depth of the water column; as this<br>ratio increases, it becomes more difficult to flush the water out of<br>the source area during the generation process, resulting in greater<br>tsunami efficiency. Fortunately, this result should not affect<br>significantly the generation of tusnamis by mega-earthquakes.</p>