The Naked Scientists

The Naked Scientists Forum

Author Topic: Calibration Curves  (Read 3472 times)

Offline kristenann2000

  • First timers
  • *
  • Posts: 3
    • View Profile
Calibration Curves
« on: 10/09/2005 03:28:06 »
I am constructing a conc'n/absorbance calibration curve.  What is the general procedure for incuding the blank (0 conc'n and 0 absorbance) when constructing the best fit line?  Of course, if it is included it the line passes through the origin of the graph.  However sometimes it seems as if you could better fit line if it excluded.  What is the proper way to the handle this?  Thanks so much!


 

Offline David Sparkman

  • Sr. Member
  • ****
  • Posts: 234
    • View Profile
Re: Calibration Curves
« Reply #1 on: 10/09/2005 07:40:59 »
I assume you are doing Atomic Adsorption or something similar. Most instruments have electronic noise from the leakage of signal (some light gets in) into the tubes or just from electronic leakage (noise). Your zero reference may not fall on the intercept if there is background signal. But that would be important and be corrected for in the equation offset.

You may also find that your instrument has a lower detection limit. I.e. your blank is showing itself to be in the middle of several low level concentrations. This simply means you are below the detection limit of the instrument, and those other points with the null point cannot be trusted.

My background is optical spectrometers and I have done calibrations for many foundries including General Motors Defiance. I haven't done AA calibrations.

David
« Last Edit: 10/09/2005 07:41:37 by David Sparkman »
 

Offline Ylide

  • Moderator
  • Hero Member
  • *****
  • Posts: 905
    • View Profile
    • http://clem.mscd.edu/~mogavero
Re: Calibration Curves
« Reply #2 on: 18/09/2005 11:59:16 »
I've found that I minimize the variance in my calibration curves when I use a Fibonacci sequence for the calibration points as well as repeat a few of the lower points.  For instance, say you want to use concentrations of 0.01M at the low and and around 0.40M in the high end.  I would plot points

1)  0.01
2)  0.01
3)  0.02
4)  0.03
5)  0.03
6)  0.05
7)  0.10
8)  0.15
9)  0.25
10) 0.40

This gives you a 10 point curve, slightly weighted towards the low end because that's where variance has the largest effect on your calibration...hence you want more data points there to minimize the variance of any one point.  

As David said, if you're not hitting origin with your calibration, don't force it.  You have some background noise or other experimental error that needs to be addressed in your calibration.  



This message brought to you by The Council of People Who Are Sick of Seeing More People
 

Offline David Sparkman

  • Sr. Member
  • ****
  • Posts: 234
    • View Profile
Re: Calibration Curves
« Reply #3 on: 19/09/2005 23:26:02 »
I have seen several 0.01 certs and I generally don't trust them mucyh. 0.0051 is also a 0.01. I think for a lot of my standards the AA people just used 0.01 instead of the word "trace": i.e. it is there but too low to measure with their instrument. If you have a good instrument, you may find the 0.01's all over the place, some being closer to 0.001. Also be careful of your curve fitting equation. With the above data set, a 4th or 5th order equation will give a great fit, but be nonsense. Keep the degrees of freedom as high as possible and only use a higher order equation when many of the points show that useful. Remember too that the reason for curvature will often be due to things in nature that are logrithmic, and most curve fitting algorithims try to get by with power series fits. Such problems as readsorption, capacitance limits, and amplification are similar to natural logs not to power series.

David
 

Offline Ylide

  • Moderator
  • Hero Member
  • *****
  • Posts: 905
    • View Profile
    • http://clem.mscd.edu/~mogavero
Re: Calibration Curves
« Reply #4 on: 20/09/2005 09:30:16 »
The response is assumed to be linear in the above calibration curve.... the concentration of standard may be increasing logarithmically, geometrically or what have you, but the response per point with fit into a y=mx+b line...

Also, if you're getting measurements of 0.0051 to a 0.01 curve, then your curve is flawed to begin with as your lowest point is above an expected measurement..



This message brought to you by The Council of People Who Are Sick of Seeing More People
 

Offline David Sparkman

  • Sr. Member
  • ****
  • Posts: 234
    • View Profile
Re: Calibration Curves
« Reply #5 on: 21/09/2005 03:48:28 »
Ylide,

Think you missed the point. Standards are not always what they say they are, especially the older ones. An analysis of 0.0051 rounds to 0.01, but on a more accurate machine might show up as very different from a true 0.010.

Second any light emmission machine will eventually get into a non-linear response due to readsorption. This is when a photon emmitted from one type of atom passes close to a second atom of the same type and is readsorbed. Though it is then reemitted almost immediately, it is heading in a different direction and not picked up by the detector and you get a curve like below: X-axis is chem, Y-axis is intensity. X is point, period is a place holder to prevent justification.
.....................x
.............x
........x
.....x
...x
..x
x

David
« Last Edit: 21/09/2005 03:51:05 by David Sparkman »
 

The Naked Scientists Forum

Re: Calibration Curves
« Reply #5 on: 21/09/2005 03:48:28 »

 

SMF 2.0.10 | SMF © 2015, Simple Machines
SMFAds for Free Forums