Upper end of concentration range

[WK]

Hi All,
hope you might be able to help here.
I am developing methods to cover an analyte range 0 - 97%.
Its internal standard. When using 13 levels over 0 - 97% I get R=0.9999.
But the check standard results are out significantly in 0-5% range and 90%+ range.
So when I use 9 levels to cover the 0-10% range it is very good when I use same standard and internal standard amount in the 9 low levels.
But when I use 9 levels to cover the 80-97% range R=0.9940 when I use same standard and internal standard amount in the 9 high levels.
I have diluted these 9 high levels various ways with non-interfering solvent so that there is no overload - but it is still poor.
I can live with 3 methods of 0-10% , 10-80% and 80%+ but it is only the 80%+ range that causes a problem.
Any ideas would be gratefully received !
Ethanol with propan-1-ol internal standard 624 1.8um 0.32mm id 30m.
165degC injector / 250degC FID / 50degC oven
Regards
WK

[Peter Apps]

Hi WK

I am slightly puzzled - at the upper end of your range you have 97% ethanol, and to that you add a standard and an internal standard ? Since ideally you would be aiming to have both the standards close to the concentration of the analyte a mathematic impossibility ensues.

What is the sample matrix ?, i.e. the 3 - 100% that is not ethanol ?

Peter

[WK]

Hi Peter,
Thanks
Sorry - by that I mean I use for example 5g of the ethanol solution and 5g of the internal standard.
The rest is water. I have used methanol to dilute after the internal standard has been added.
Regards
WK

[Peter Apps]

Hi WK

Can you give us your complete scheme for adding standards, diluting with methanol, and then injecting, including the injector conditions in terms of split or splitless.

Peter

[WK]

Hi Peter,
I have prepared 9 ethanol standards in water using the density and tables from 80%v/v to 99.6%v/v.
To 5g of these I have added 5g of n-propanol.
I have run these as is and also diluted 1+2 methanol and also 1+4 methanol by volume.
I get R2 about 0.994 each time.
0.1ul 200ml min split.
By using the same internal standard amount in each of these 9 levels I get a bunched up graph in top right hand corner. The method is not set up to pass through zero - just 1st order. Plotting with 2nd/3rd or weighting does not help.
Would using a bit less internal standard in lower end of this range help - to create more different analyte/internal standard ratios?
Regards
WK

[Peter Apps]

Hi WK

My best guess is that you have some strange effect of mixing three alcohols in roughly simialr amounts with a little bit of water, which is affecting evaporation rate, pressure pulse and split ratio. I would try using 10 times as much methanol (to drown out the effects of the other components) and injecting 1 ul, which is usually much more robust than 0.1 ul. With those changes, playing with the inlet temperature and the pre-injection dwell and injection speed might also help.

What is the repeatability like for replicate injections in the problem range ?

Peter

[WK]

Hi Peter,
The %RSD for 7 runs without methanol dilution was 0.18%.
The %RSD for 7 runs with 1 in 5 dilution with methanol was 0.27%.

Thanks
WK

[Peter Apps]

Hi WK

No smoking gun there then - but it is still worth testing the linearity with more methanol and larger injections. What is the rationale behind the rather low injector temp ?

Peter

[WK]

Hi Peter,
Thanks again for posting replies.
I get a breakdown products (inj >175degC) with certain components in samples which co-elute with the ethanol.
I will try to test larger injection / more diluent overnight.
WK

[Peter Apps]

Hi Peter,

I get a breakdown products (inj >175degC) with certain components in samples which co-elute with the ethanol.

WK

OUCH !

[WK]

Hi Peter,
Kind of thinking I'm quite lucky to get the results I'm getting then!!
WK

[Yama001]

I would not be surprised to see the result for a check standard near the low end of the curve be off by a lot more than those in the mid range. If there is any intercept with the y-axis at all, it will throw the results near the low end by a lot more than you may expect, particularly with an internal standard calibration.

I prefer in these cases to to fit the curve to a y = mx model, rather than y = mx + b. It is not considered statistically valid, however.

[WK]

Hi Yama001,
Thanks for the post.
I have read into it and if I cover the full range of say 1% to 96% with 13 levels I do get
the usual situation where the best precision will be in the mid point of the curve and the worst precision in the very upper and lower end. Hence so I have been trying to use so far 3 different methods to cover the ranges:
eg 0 - 10% , 10 - 80% and 80%+.
As I say its only the 80%+ which I try to improve the correlation.
I will recheck my work again tomorrow using %v/v to 2 decimal places from the densities (instead of 1 decimal place).
(The tables only use 1 decimal place for %v/v from the 4 decimal place density and 1 decimal place %v/v can cover a range +/- 0.0002 g/ml in this part of the tables).
Regards
WK

[Peter Apps]

Hi WK

How are you making up the mixtures ? - all volumetrics, a mix of volume and weight with tabulated corrections for density, all weighing ? And what kind of quantities are you working with - e.g. if you make X% ethanol in water by putting X ml of ethanol into a volumetric and making it up to 100 ml with water you will get a different composition than if you put in 100-X ml of water and top up with ethanol.

Peter

[WK]

Hi Peter,

I am adding X ml ethanol to 100ml volumetric. Make to mark with DI water. Mix well and allow to stand for 4hours. Make up to mark with DI water. Mix well and leave 2hours.
Measure density at 20.0degC. Use alcohol tables to determine ethanol %v/v.

I seem to have a marginal outlier in upper range so will keep you posted.

Regards
WK