GC Calibration -- TCD Detectors

[mugrad.2011]

I'm dealing with an SRI Multiple Gas Analyzer ... A co-worker insists that we fit our calibration points to a quadratic. Now, these instruments have TCD detectors and we are analyzing natural gas samples. It is my understanding from everything that I can find that these detectors have a linear response. He is running a single point injection for each concentration level and arguing that the shape of the 3 points is not linear. Of course the quadratic is extremely wide (I think because it's not quadratic) to the point of local linearity within the calibrated region. Can anybody please help me understand this? My data with three calibration points and a line of best fit forced through the origin gives R-squared values of 1.0000 and 0.9999 ... Which sounds linear to me ... ... How about you?

Thanks

[James_Ball]

I haven't worked with gas analyzers, but for most chromatography applications it is usually recommended to not force the origin with linear or quadratic calibration curves. Also it is not recommended to calibrate using quadratic fit with anything less than 5 points. With only three points how can you know if the curvature of the quadratic fit is actually a curved response fit or simply curvature induced by one point being slightly inaccurate?

Actually most of the calibrations I do have 7+ points so you can easily tell if it is going to be linear or quadratic.

As for a TCD being linear, it will depend on how many orders of magnitude you are calibrating across. For the best accuracy you should have calibration points that are at the upper and lower concentrations you will report, with ones in the middle to fill in the shape of the curve. For EPA regulated work like I do, reporting results above or below your highest and lowest calibration points is considered invalid.

I used to do metals analysis where the ICAP was considered "linear over five orders of magnitude" which does hold up when you are in the upper ranges, but once you are working near the Limit of Detection in trace analysis that falls apart quickly. The method said to calibrate with a 1ppm standard and a blank with the limit of detection at 5ppm. Problem was when you ran a standard at 10ppb sometimes it would calculate out to be 40ppb or -20ppb, which is fine if you are worried about values in the ppm range but not if you need to know values near 10ppb required reporting levels. So whatever detector you are using, be sure to check it at the limits of what you expect samples to be and at the lowest level you wish to report as a positive result.

[aidnai]

Hi, I have used GC/TCD for about 10 years and my experience is that they are mostly linear (I had one method I used from 1000ppm to 100%). But there are exceptions --
- column overloaded
- trying to measure H2 with helium carrier
- other method/setup issues
As James Ball said, you need to prove linearity with your calibration, and that means using standards above and below your measurement range. If you want a zero actually measure a zero (nitrogen or whatever).