The typical manufacturing errors of optical instruments are tilt, decenter, thickness error, and curvature error. The
difference between these errors and the surface irregularity is the randomness of the surface irregularity. But the surface
irregularity is not fully random. For example, there is a correlation between deviations on neighboring points. How much the correlation spreads determines the character of the surface irregularity. With the statistical treatment of many sample data, the surface irregularity can be expressed as a linear combination of base functions, [see manuscript for equation] where f1(x,y)...fn(x,y) are base functions and a1...an are coefficients. x and y are coordinates perpendicular to the optical axis. fi(x,y) contains no randomness. ai's are random variables and independent to each other. If f1(x,y)...fn(x,y) are known, the sensitivity control to deviation fi(x,y) can be performed just as a tilt. If the sensitivity to f1(x,y)...fn(x,y) are small, the sensitivity to surface irregularity is promised to be small. This paper contains the proposal of a general method of the sensitivity control to manufacturing errors. The sensitivity control to the surface irregularity is an advanced application of this method. In this method the sensitivity of
RMS OPD or MTF is exactly calculated based on the change of the optical path length. The sensitivity is controlled as a
target of the optimization. This method is applicable to any sort of manufacturing errors and the control of the sensitivity
is reliable even for the large aperture and/or wide angle lenses. In sections 2 and 3 the treatment of the randomness of the surface irregularity is explained. In sections 4 and
5 the method to calculate the sensitivity is explained. In section 6 it is shown how the sensitivity control is performed during the optimization and how effective the sensitivity control to surface irregularity is.