The problem of numerical integration of galvanometrically recorded seismograms involving three successive integrations is considered and the following conclusions are reached:
The error involved as a result of three integrations is essentially cubic in nature but also contains quadratic and linear factors. The form of the error curve is a cubical parabola.
The error is a random effect, or we might say a curve-point function.
Correction factors must be determined for short intervals. The method of application of these factors eliminates cumulative error and reduces the liability of computational error.
The presence of transients are noted which limit the usefulness of the results within the range over which the transients are present.
The method was applied to a test record of known motion which was approximately harmonic. In the development of the method there is nothing that seems to require such motion, but the transients excited in the recording system at the beginning of the motion effectively limit the accuracy of the results to that portion of the record which corresponds to a steady state.