I was wondering if anyone has come across any problems with Chromeleon v6.3 not calculating the %RSD correctly. If you look under the calibration tab within the quant method window, for a sequence, the %RSD for that sequence is not correct. If you then go to the sample table for that sequence and re-enter all of the fields for the first standard in that sequence, and then go back to the quant method the %RSD is suddenly correct. It seems to not be reading the first standard correctly. Can anyone tell me why this is happening? I have contacted Dionex and they are working on the problem, but I was curious if anyone else has had this problem.

Because of the unusual combination of factors involved in this case, it is very unlikely that any other Chromeleon users will have an issue like this.

The RSD in question is that of the calibration residuals. (A residual is the difference between the response of the calibration standard and the corresponding response predicted by the calibration function.) This RSD is computed correctly by Chromeleon, but in this case it was differing from the RSD of the responses of the calibration points -- an unexpected result considering that this was a single-level calibration with 5 replicates.

The answer lies in the fact that the method was set to use a point-to-point fit, a discontinous function that is recommended only for cases in which a continuous function (such as a polynomial or exponential function) cannot adequately describe the component response behavior. With the point-to-point fit, the calibration responses at each level are averaged with each other, and a line is constructed between each adjacent pair of averaged calibration points. If only one calibration level is present, the point-to-point fit is normally equivalent to a linear fit with the intercept forced to 0.

The reason those RSDs differed in this case is that the volume value stored for four replicates was 50.0, but for one replicate it was 50.0000038146973. With a point-to-point fit, this yields one calibration level with four replicates and a second level with only one replicate. Because there was only one point at the second level, its residual was 0, which explains why the RSD of the calibration residuals was lower than expected.

Further investigation has indicated that the small discrepancy in the one volume value was most likely introduced during a number type conversion in the code segment of the Sequence Wizard that is specific to the ASI-100 autosampler. Other instances of non-zero digits beyond the 6th decimal place have been found only in other sequences that used ASI-100 autosamplers. With normal calibration methodology, the impact of such minute errors on the calibration function is insignificant, but because the point-to-point fit is a discontinous function, it was significantly affected in this case.

We have advised the laboratory to use a normal linear fit for their single-level methodology, and provided them with a special report form to help them quickly find any other cases in which there was an RSD discrepancy with their point-to-point approach. We will take other actions as appropriate.

(Note: The actual investigation was done the week of January 19th -- apologies for the delay in getting this response posted.)