104 | 104 | ^*^ using taylor expansion near 0 for fq = 3 (sin(qr)/qr - cos(qr))/qr^2^; using quad precision as a cross check, this is more accurate than computing the direct expression for qr < 0.1 in double precision, and so the values will differ from those in sasview. It is much more accurate in single precision. This affects many models, including sphere and ellipsoid models. The cylinder models are using J1(qr)/qr, which has its own numerical expansion and does not suffer from this problem. Should probably settle on one or the other and move it into the library. So even though the double precision values for sphere, ellipsoid and triaxial ellipsoid differ by more than 5e-14, they are still marked good. Similarly, the Teubner-Strey model checks out against the quad precision version and so is marked good. |