Author Topic: Acceptable errors  (Read 5514 times)

knshugart

  • Post Doc
  • ***
  • Posts: 10
Acceptable errors
« on: April 05, 2016, 11:09:20 AM »
I've been trying to test some of the standards we have in our lab by running a calibration on one standard and testing another. I know that for samples with a small amount of an element it is best to use a standard with a small/similar amount of that element. However, I'm curious just how far off I can expect to see the quantification if I use a pure element standard and then try to measure a small amount.

For example: I recently used pure Si as the standard and tried to measure an SiO2 standard. I found myself off by 8%. Is that reasonable? Or does it tell me my standards are bad? (Goodness knows they are old!) Any other theories?

Thanks!

Probeman

  • Emeritus
  • *****
  • Posts: 2596
  • Never sleeps...
    • John Donovan
Re: Acceptable errors
« Reply #1 on: April 05, 2016, 12:58:37 PM »
I've been trying to test some of the standards we have in our lab by running a calibration on one standard and testing another. I know that for samples with a small amount of an element it is best to use a standard with a small/similar amount of that element. However, I'm curious just how far off I can expect to see the quantification if I use a pure element standard and then try to measure a small amount.

For example: I recently used pure Si as the standard and tried to measure an SiO2 standard. I found myself off by 8%. Is that reasonable? Or does it tell me my standards are bad? (Goodness knows they are old!) Any other theories?

Thanks!

Hi and welcome to the user forum!  I think I can explain, and others here will certainly have additional ideas/suggestions for you.

First of all, it is an excellent idea to run one's standards against each other as you mentioned. Selecting one standard as the "primary" standard and running the remaining standards as "secondary" standards is exactly how one can evaluate accuracy in our unknowns.  The idea being that the primary standard has the best accuracy (based on whatever considerations), and the secondary standard acts as a sort of check on our matrix physics extrapolation, to match the expected matrix of our unknowns.

For example, we might use MgO as our primary standard for Mg and the NIST K411/412 mineral glasses as secondary standards to check on our extrapolation to our unknown- assuming of course that our unknown is a mineral glass! 

By the way, the need for "matrix matching" of standards to unknowns is generally unnecessary now as the matrix correction physics has improved greatly since early days. That said, the example you mentioned of Si as the primary standard and SiO2 as the secondary standard fails for a well known reason, and it has nothing to do with the matrix correction physics.  The problem here is that the Si ka emission line has a very significant shift in the emission energy depending on the Si-O bond chemistry.  This is because the Si L shell is involved in the oxygen bond and it is the L to K transition we are measuring.

There are several ways to deal with this if you actually need to measure Si-O compositions with varying chemistry.  The most accurate (and slowest) method is to measure the integrated peak intensity as opposed to simply the peak intensity as seen here:

http://probesoftware.com/smf/index.php?topic=536.msg2992#msg2992

One can also apply so called Area Peak Factors (APF) to the peak intensities as described here (they come in two flavors: compound and specified):

http://probesoftware.com/smf/index.php?topic=536.msg2946#msg2946

So generally if we are measuring Si in a non-oxide matrix we would use elemental Si as a primary standard, and if we are measuring Si in an oxide matrix we would use SiO2 as a primary standard.

By the way, although Si ka has a significant peak shift/shape effect, most higher energy emission lines show much smaller effects and these can generally be ignored especially when characterizing trace elements.  This is because the accuracy error from the peak shift is usually much smaller than the precision of the measurement.  In fact for trace elements, we generally want to use a very high concentration standard to improve sensitivity (see the discussion here):

http://probesoftware.com/smf/index.php?topic=607.msg3465#msg3465
« Last Edit: April 05, 2016, 03:16:40 PM by Probeman »
The only stupid question is the one not asked!

knshugart

  • Post Doc
  • ***
  • Posts: 10
Re: Acceptable errors
« Reply #2 on: April 05, 2016, 01:38:50 PM »
Thanks! That is very helpful information. And makes me wish I had even more standards I could test against one another. Most of what I have is pure materials, so there aren't many secondary tests I can run.

I also noticed an issue when I used a pure C standard and then measure CaCO. And had my C values very far from what the CaCO told me it should be. Is there similar reasoning for this example? (Note-- I ran these tests at 15kV which I know is not idea for light elements but at the moment that is what my Cameca SX-100 is set up for and no one here is quite sure what needs to be done to run at lower voltages... they keep mumbling something about a spacer in the gun. This high voltage was my theory as to the difference in expected and measure for the CaCO.)

Probeman

  • Emeritus
  • *****
  • Posts: 2596
  • Never sleeps...
    • John Donovan
Re: Acceptable errors
« Reply #3 on: April 05, 2016, 01:54:44 PM »
Thanks! That is very helpful information. And makes me wish I had even more standards I could test against one another. Most of what I have is pure materials, so there aren't many secondary tests I can run.

I also noticed an issue when I used a pure C standard and then measure CaCO. And had my C values very far from what the CaCO told me it should be. Is there similar reasoning for this example? (Note-- I ran these tests at 15kV which I know is not idea for light elements but at the moment that is what my Cameca SX-100 is set up for and no one here is quite sure what needs to be done to run at lower voltages... they keep mumbling something about a spacer in the gun. This high voltage was my theory as to the difference in expected and measure for the CaCO.)

Get yourself some standards!   There are some offered here:

http://probesoftware.com/smf/index.php?topic=301.0

and

http://probesoftware.com/smf/index.php?topic=308.0

and I have some pure Wollasonite which I can send you if you are interested.

Once you have some standards you can "evaluate" them in the Evaluate application:

http://probesoftware.com/smf/index.php?topic=71.msg400#msg400

You are probably destroying the CaCO3 with 15 keV!    :(
 
Beam sensitivity can be a very big issue for some materials.  Much discussion can be found here:

http://probesoftware.com/smf/index.php?topic=11.0
The only stupid question is the one not asked!

knshugart

  • Post Doc
  • ***
  • Posts: 10
Re: Acceptable errors
« Reply #4 on: April 06, 2016, 05:33:00 AM »

By the way, the need for "matrix matching" of standards to unknowns is generally unnecessary now as the matrix correction physics has improved greatly since early days.

...

By the way, although Si ka has a significant peak shift/shape effect, most higher energy emission lines show much smaller effects and these can generally be ignored especially when characterizing trace elements.  This is because the accuracy error from the peak shift is usually much smaller than the precision of the measurement.  In fact for trace elements, we generally want to use a very high concentration standard to improve sensitivity (see the discussion here):

http://probesoftware.com/smf/index.php?topic=607.msg3465#msg3465


Out of curiosity in regards to your statement about not needing matrix matching as much as before-- is that only true when using the ProbeSoftware? Sadly, I have yet to convince those with money here of our need for it, so we are still running with the Cameca provided software.

In addition-- one reason I was running these tests was to prove that the standards we have are degraded/incomplete for our needs. We are a user facility and have a large variety of projects coming through. Mostly metals and ceramics, not much in the geological field. Do you have suggestions of good secondary standards and where to get them? (Most of what I found wandering through the forum so far seems to be glasses, which are of less interest here.) Our current standards are blocks from CM containing mostly pure metals, making it difficult to do secondary tests.

Probeman

  • Emeritus
  • *****
  • Posts: 2596
  • Never sleeps...
    • John Donovan
Re: Acceptable errors
« Reply #5 on: April 06, 2016, 07:45:25 AM »
Out of curiosity in regards to your statement about not needing matrix matching as much as before-- is that only true when using the ProbeSoftware? Sadly, I have yet to convince those with money here of our need for it, so we are still running with the Cameca provided software.

It will depend on the accuracy of the matrix correction methods used and the compositions to which they are applied.  Cameca utilizes XPP which is the simplified version of the PAP correction.  It is matrix correction method #10 in the Probe for EPMA/CalcZAF software.  A nice post discussing matrix correction accuracy is here:

http://probesoftware.com/smf/index.php?topic=508.msg2779#msg2779

By the way, if you'd like to play with the matrix correction methods in CalcZAF you can download it for free here:

http://probesoftware.com/Technical.html

In addition-- one reason I was running these tests was to prove that the standards we have are degraded/incomplete for our needs. We are a user facility and have a large variety of projects coming through. Mostly metals and ceramics, not much in the geological field. Do you have suggestions of good secondary standards and where to get them? (Most of what I found wandering through the forum so far seems to be glasses, which are of less interest here.) Our current standards are blocks from CM containing mostly pure metals, making it difficult to do secondary tests.

Welcome to the eternal quest for secondary standards!    :D

In the meantime you can perform some very useful tests on your pure metal standards by analyzing them for elements that are *not* supposed to be present.

For example, try and measure Fe in pure Mn or S in pure Mo or V in pure Ti and see what you get... you should get zero, but you won't! 

The idea being, that while it's useful to have a secondary standard with a known non-zero concentration, it's as useful to have a pure metal (or oxide) "blank" to check if one can measure a zero concentration (we should be able to if the element is not present!). 

Or as Mike Jercinovic at U. Mass says " If you can't measure something, try measuring nothing, because if you can't measure nothing then you can't measure anything".

By the way, you should try and make the EPMA 2016 topical conference in Madison in May- everyone will be there:

http://probesoftware.com/smf/index.php?topic=99.msg3927#msg3927
The only stupid question is the one not asked!

knshugart

  • Post Doc
  • ***
  • Posts: 10
Re: Acceptable errors
« Reply #6 on: April 06, 2016, 07:52:08 AM »

By the way, you should try and make the EPMA 2016 topical conference in Madison in May- everyone will be there:

http://probesoftware.com/smf/index.php?topic=99.msg3927#msg3927

I'm already registered! And looking forward to it.  (Though I note very few of the hotels are within the per-diem rate for government...)

I'll also be at M&M in July.
« Last Edit: April 06, 2016, 08:04:16 AM by knshugart »

Probeman

  • Emeritus
  • *****
  • Posts: 2596
  • Never sleeps...
    • John Donovan
Re: Acceptable errors
« Reply #7 on: April 06, 2016, 08:53:37 AM »

By the way, you should try and make the EPMA 2016 topical conference in Madison in May- everyone will be there:

http://probesoftware.com/smf/index.php?topic=99.msg3927#msg3927

I'm already registered! And looking forward to it.  (Though I note very few of the hotels are within the per-diem rate for government...)

I'll also be at M&M in July.

Awesome!   See you there then.

By the way, there is a Probe Software user meeting on Monday afternoon that all are invited to...   :)
The only stupid question is the one not asked!

John Donovan

  • Administrator
  • Emeritus
  • *****
  • Posts: 3016
  • Other duties as assigned...
    • Probe Software
Re: Acceptable errors
« Reply #8 on: April 06, 2016, 09:41:43 AM »
Awesome!   See you there then.

By the way, there is a Probe Software user meeting on Monday afternoon that all are invited to...   :)

Probeman is correct, the Probe Software user meeting on Monday afternoon is open to all, but Barbara needs a head count for pastries and snacks, so please RSVP to her at barbara@probesoftware.com if you want to attend the user meeting.

We'll also have some very short informal presentations by several our experts and lots of discussion... see you there! Thanks!
john
« Last Edit: April 06, 2016, 09:43:28 AM by John Donovan »
John J. Donovan, Pres. 
(541) 343-3400

"Not Absolutely Certain, Yet Reliable"

knshugart

  • Post Doc
  • ***
  • Posts: 10
Re: Acceptable errors
« Reply #9 on: April 06, 2016, 11:29:16 AM »
I see there is also a Cameca users group meeting Monday morning... What do I have to do to attend that?

Probeman

  • Emeritus
  • *****
  • Posts: 2596
  • Never sleeps...
    • John Donovan
Re: Acceptable errors
« Reply #10 on: April 06, 2016, 08:25:50 PM »
I see there is also a Cameca users group meeting Monday morning... What do I have to do to attend that?

Paul Carpenter (TC organizer) says: "That info will be put on the EPMA2016 website for attendees to use."

Meaning (I think) that the TC web site will soon have sponsor email addresses for signing up for all the user meetings.
The only stupid question is the one not asked!

knshugart

  • Post Doc
  • ***
  • Posts: 10
Re: Acceptable errors
« Reply #11 on: April 07, 2016, 05:28:06 AM »
Thanks!

I'll keep checking back on the website.