Four iterations (top to bottom) of sampling for the CCl stretch in HClCO up to 8000 cm−1. The left column shows the calculated points (round points), the fitted curve (line), and a rescaled (dashed line). The right column shows (line), the true error weighted by (dashed line), and the optimal position for the next point (point on x axis).
The average vibrational energy deviation to the reference PES for various error thresholds ε cut (see text). Crosses with dotted line and diamonds with continuous line represent AGAPES runs for HNO PES generation using the true error to the reference PES and the error estimate , respectively, for the point placement scheme.
The same as in Fig. 2 this time with the number of points suggested by AGAPES at each run on the x axis.
The left column of each interval count the number of terms N suggested by our method normalised by N 4, the number of terms suggested by the standard 4th order scheme over remoteness intervals for Formaldoxime. The right column shows a corrected version, where all correlation terms which came out to be insignificant after the calculation are not counted in N. For the definition of the remoteness R of a coordinate combination and insignificance see Sec. IV.
Convergence to the reference PES for HNO. Number of used points #P and average error to vibrational levels of the reference PES (in cm−1) found by employing threshold ε cut (cm−1) determined by different methods (see text).
Average error to vibrational levels of the respective reference PES and number of points #P suggested by our method employing the convergence threshold ε cut for the vibrational levels in the subspaces up to E max . The in brackets denote the error to the PES one line above. The P values in brackets indicate the #P of a 4th order standard approach. All energies in cm−1.
Article metrics loading...
Full text loading...