Artificial Intelligence (AI) and microanalysis

[John Donovan]

I was discussing with a colleague recently about her use of AI in her own geological research and she said she has found it quite useful in several ways. As the conversation continued, I asked if she thought these AI tools might be helpful in helping users with our software when they have questions and not finding the answer in either our Training and Support page:

https://www.probesoftware.com/training-support/

or in our Resources page:

https://www.probesoftware.com/resources/

including this user forum and also our YouTube channel video tutorials:

https://www.youtube.com/@ProbeSoftware

So we decided to perform a test. The "prompt" which we used, first with the Perplexity AI which is what she utilized personally for her own research, was this: "How do I set up a run in Probe for EPMA where I analyze Fe in quartz and correct for secondary fluorescence?  In addition to other resources, please consult the guides on the probe software website (https://www.probesoftware.com/training-support/) and the user forum (https://smf.probesoftware.com/index.php)."

This prompt has the advantage of having a simple question (how do I analyze Fe in quartz?) and also a complex question (how to correct the Fe in quartz measurements for secondary boundary fluorescence from nearby phases containing Fe?).

The idea being, that if the AI really could read through all our hundreds of pages of user manuals, including the User's Reference manual, our step by step directions in the Getting Started and Advanced Topic manuals and in addition the 12,000+ posts in the Probe Software User Forum, maybe it might be able to do a pretty good job answering the question. And Perplexity responded with the following:

Code Select Expand

To set up a run in Probe for EPMA to analyze Fe in quartz and correct for secondary fluorescence, follow these steps:

    Create a new run and set up your quartz sample:
        Open Probe for EPMA and create a new run
        In the Acquire! window, click "New Sample" to set up your quartz sample
        Add Fe as an analyzed element and Si as a matrix element
        Select appropriate standards for Fe and Si
    Configure secondary fluorescence correction:
        In the Analyze! window, click "Calculation Options"
        Check the box for "Calculate Secondary Fluorescence"
        Click "Secondary Fluorescence Parameters" to open the configuration window
        Set the boundary geometry (e.g., planar)
        Define the matrix composition on the other side of the boundary (e.g., Fe-rich phase)
        Specify the distance from the boundary for each analysis point
    Optimize analytical conditions:
        Use a higher accelerating voltage (e.g., 20-25 kV) to excite Fe K-alpha
        Set longer count times for Fe to improve detection limits
        Consider using multiple spectrometers for Fe to increase sensitivity
    Acquire data:
        Run your analyses on the quartz sample, ensuring you collect data at various distances from the Fe-rich boundary
    Process and interpret results:
        After acquisition, the software will automatically apply the secondary fluorescence correction
        Review the corrected Fe concentrations, which should decrease with increasing distance from the boundary

Additional tips:

    Consult the Probe for EPMA user guide for detailed instructions on setting up secondary fluorescence corrections
    Consider using the multi-point background method for improved trace element accuracy
    If available, use a field emission gun (FEG) source for better spatial resolution and sensitivity

For more detailed guidance, refer to the Probe Software website (https://www.probesoftware.com/training-support/) and consult the user forum (https://smf.probesoftware.com/index.php) for specific questions or troubleshooting. The forum has several threads discussing trace element analysis and secondary fluorescence corrections that may be helpful for your particular application.
You can look through the response and decide what you think, but to quickly get to the point, it actually seemed to be somewhat helpful in a general sort of way, but then you will note this paragraph in step 2 quoted from the response above:

"Configure secondary fluorescence correction:

    In the Analyze! window, click "Calculation Options"
    Check the box for "Calculate Secondary Fluorescence"
    Click "Secondary Fluorescence Parameters" to open the configuration window
    Set the boundary geometry (e.g., planar)"

But, but, but those specific controls don't exist in that dialog!  In fact, those specific buttons don't exist anywhere in the software... so it hallucinated these buttons! Instead you'd actually want to utilize the Boundary Corrections button in the Analyze! window as posted in the user forum here:

https://smf.probesoftware.com/index.php?topic=1545.0

That is, if the AI really understood what it was doing.  This type of response reminds me of the "chat" types of assistance a lot of sites are offering these days instead of actual phone support with a human, which I personally have found to be pretty darn useless. 

But to be fair we decided to use the same prompt with another AI tool, this time the Grok tool from xAI and it responded:

Code Select Expand

    To set up a run in Probe for EPMA for analyzing Fe in quartz with corrections for secondary fluorescence, follow these steps:

        Define the Analytical Problem:
            Identify the elements you wish to analyze. In this case, you're focusing on Fe in quartz.
            Ensure you have appropriate standards for Fe, typically Fe metal or Fe-bearing minerals like magnetite or fayalite.
        Sample and Standard Preparation:
            Prepare your samples and standards. Quartz samples should be polished to ensure a flat surface for analysis. Place both your unknown quartz samples and Fe standards in the sample holder or mount.
            Photograph or image your mount to keep track of sample positions.
        Instrument Setup:
            Ensure the electron microprobe is properly calibrated. This includes filament saturation and column alignment to optimize the electron beam.
        Setting Up Analysis in Probe for EPMA:
            Load Standards: Use the pre-digitized standard mounts feature if available. Navigate to the standards database in Probe for EPMA and select or add your Fe standards.
            Set Conditions: Define the analytical conditions like accelerating voltage, beam current, and counting times. For Fe in quartz, standard conditions might involve 15-20 kV and beam currents around 10-20 nA, but these should be optimized based on your specific setup and the sensitivity required.
        Correcting for Secondary Fluorescence:
            Use of Boundary Fluorescence Correction: Probe Software has implemented boundary fluorescence corrections based on work by Francesco Salvat and Xavier Llovet. Make sure you're using the most recent software version where these corrections are included.
            Configure the Correction: When setting up your run, enable the secondary fluorescence correction in the analysis parameters. This will use analytical expressions developed for real-time correction based on spatial distances and compositions analyzed.
        Running the Analysis:
            Automate Acquisition: Use the automation features to move between standards and samples. The software allows for automated spectrometer peaking, which is crucial for precise quantification.
            Data Collection: Start the analysis by acquiring counts on both standards and unknowns. Ensure you repeat measurements for statistical reliability.
        Data Analysis:
            Review Data: After data collection, analyze the results to check for agreement with standards and to ensure the secondary fluorescence correction was applied correctly. Look for any spectral interferences or deviations that might indicate issues with the correction.
            Use of Software Tools: Utilize the "Analyze!" feature to process the data, where you can apply further corrections if needed or check the effectiveness of the fluorescence correction.
        Documentation and Reporting:
            Document all steps taken, conditions used, and any anomalies observed during the run. Probe for EPMA allows for comprehensive reporting and documentation, which should be used to record your findings.

    For additional guidance:
        Consult the Probe Software Website: The training and support section provides detailed guides on setting up analyses, which could include specific tips for trace element analysis like Fe in quartz.

    User Forum: Engaging with the user forum can provide practical tips from experienced users, particularly on handling corrections for secondary fluorescence.

    Remember, the effectiveness of your analysis will depend on both the setup of your instrument and the software tools' capabilities to handle such corrections. Always refer to the most current documentation and updates, as software enhancements might introduce new features or better algorithms for such corrections.

Again, pretty vague and general but at least it didn't hallucinate non-existent buttons in the software!

What's your experience with these tools?

[Ben Buse]

My concerns would be "hallucination", energy and water usage of AI, and why not if you can't find an answer ask in forum and have real people answering you

[Probeman]

My concerns would be "hallucination", energy and water usage of AI, and why not if you can't find an answer ask in forum and have real people answering you

Hallucination does seem to be a significant problem with current AIs. And energy use is certainly a concern as you mention.

The above was only an anecdotal instance of exploring current capabilities, and one would certainly hope that people would post their questions to this forum if they needed help with microanalysis (that is the purpose of this forum!). But do keep in mind that some people are shy and hesitate to ask questions in public.

Though I find that "real people" are often very helpful and don't usually hallucinate! 

[Nicholas Ritchie]

There are a handful of modern technologies that I'd give up in heartbeat - social media, smart phones, LLMs and other many other forms of AI.  (Yes, I can be a crank ) Sadly, each of these is here to stay.

For the moment, LLMs seem to be good at answering questions for which there is a lot of data on the Internet. Think "what is 2+2" rather than "how do I use the secondary fluorescence corrections in Probe for EPMA."  When there isn't enough real information available they resort to making stuff up - hallucinations.

However, if reframed, hallucinations may be one of the super-powers of current AI.  Rather than calling them hallucinations call them "guided creativity".  Imagine problems that require creativity to solve but once we are provided with a solution it is easy to verify that it is correct.  An example is a Rubic's cube.  The program needs to be able to assimilate and develop strategies to bring the cube closer to solution.  It needs to test many strategies and iterate.  However, it is very easy to differentiate correct solutions from incorrect ones.  There are many problems like this - protein folding, design of optimally light and strong mechanical parts, molecule design, other design problems requiring tradeoffs.  There may even be some in EPMA though I can't think of any off the top of my head.

[John Donovan]

There are a handful of modern technologies that I'd give up in heartbeat - social media, smart phones, LLMs and other many other forms of AI.  (Yes, I can be a crank ) Sadly, each of these is here to stay.

I hear you!

They can be quite harmful to some, especially children.  It is heartening to see schools starting to ban phones during school hours:

https://www.thefp.com/p/jonathan-haidt-school-phone-bans-anxious-generation

Haidt spells out four "foundational rules" to inspire a "Great Rewiring of Childhood." They are: no smartphones before high school, no social media before age 16, no phones at school, and more unsupervised play and independence for kids.

[John Donovan]

We wanted to once again, test how useful AI might be for training and support in using Probe for EPMA. For this test we utilized Grok AI:

https://x.com/i/grok

Here is the query:

How do I set up a run in Probe for EPMA where I correct for beam sensitivity in EPMA samples?  In addition to other resources, please consult the guides on the probe software website (https://www.probesoftware.com/training-support/) and the user forum (https://smf.probesoftware.com/index.php)?

Note that we did not mention "volatile" or "TDI" effects. Here are the web pages it says consulted:

Code Select Expand

Relevant Web Pages
1. Probe for EPMA Manual: Getting Started and Advanced Topics
The software used was both Probe for Windows and Probe for EPMA Enterprise Versions (32 bit). The following conventions are used in this document: Menu Commands and Dialog Box (Windows) Names and buttons are bold-faced whenever they occur in the text. Dialog Box Options are italicized and FILE NAMES are capitalized. ... Context sensitive HELP is available in any window by pressing the F1 key.
xraysrv.wustl.edu
2. Electron probe micro-analyzer (EPMA)
An electron probe micro-analyzer is a microbeam instrument used primarily for the in situ non-destructive chemical analysis of minute solid samples. EPMA is also informally called an electron microprobe, or just probe. It is fundamentally the same as an SEM, with the added capability of chemical analysis.
serc.carleton.edu
3. Load Sample Setup - EPMA Probe Documentation
It may be best to name the sample setup something descriptive, such as, "olivine setup" or "volatile calibration setup" so that the correct sample setup may be easily remembered. To utilize the full flexibility of PROBE for setting up separate setups for different phases, you will also want to make sure one has set the proper standard assignments (Analyze!-Standard Assignments), and Calculation Options (e.g. Oxygen by stoichiometry, maybe mineral formula type), as well as any various operating conditions in Acquire! (e.g., current, beam size).
documentation.help
4. Washington University EPSC Microprobe Lab
Outline of a typical EPMA run Here is the bigger picture of how to set up a probe run. Operational Layout: Location of Displays and Controllers ... JEOL monitors: the upper three monitors are for the JEOL microprobe system. They are controlled using the left keyboard and left mouse only. OS-9 monitor is upper left and is used for all electron beam imaging and JEOL photography.
xraysrv.wustl.edu
5. Probe for EPMA v. 13.8.7 (Xtreme Edition)
probesoftware.com
6. Training & Support - Probe Software
Customized introductory, intermediate or advanced training on specific areas of EPMA methods, e.g., matrix corrections, spectral interference corrections, multi-point backgrounds for trace elements in complex matrices, MAN backgrounds for high throughput quantification of points and/or x-ray maps and improved sensitivity, TDI corrections for beam sensitive materials, etc.
probesoftware.com
7. Electron microprobe - Wikipedia
An electron microprobe (EMP), also known as an electron probe microanalyzer (EPMA) or electron micro probe analyzer (EMPA), is an analytical tool used to non-destructively determine the chemical composition of small volumes of solid materials. It works similarly to a scanning electron microscope: the sample is bombarded with an electron beam, emitting x-rays at wavelengths characteristic to the elements being analyzed.
en.wikipedia.org
8. Electron Probe Microanalysis - an overview | ScienceDirect Topics
The discussion with regard to the ZAF is truncated here as most software packages employ correction factors that are each unique. ... X Zhou, G.E. Thompson, in Reference Module in Materials Science and Materials Engineering, 2017 · Electron probe microanalysis (EPMA) is a nondestructive technique that allows determination of the composition of small volumes of a solid material. In an electron probe microanalyzer, an electron beam is generated and directed to a specimen to be examined and a wavelength dispersive spectrometer is employed to detect the X-rays emitted from the specimen.
sciencedirect.com
9. Electron Probe Microanalysis - EPMA - Electron Microbeam Analysis | Lucideon
Electron Probe Microanalysis (EPMA) is a specialised technique for precise elemental analysis, employing an electron beam to induce X-ray emissions characteristic of the elements present in a given sample.
lucideon.com
10. Caltech
web.gps.caltech.edu
11. Electron Probe Micro Analyzer - an overview | ScienceDirect Topics
Since the emission efficiency of the secondary ion changes by matrices, to determine quantities, one must analyze a standard specimen which has a similar composition and known concentration under the same conditions as the target inspection to produce a relative sensitivity index. EPMA (Electron Probe Micro Analyzer) uses an electron beam excitation source.
sciencedirect.com
12. Electron Probe Microanalysis - EPMA Technique | Lucideon
Electron Probe Microanalysis (EPMA) is a specialised technique for precise elemental analysis, employing an electron beam to induce X-ray emissions characteristic of the elements present in a given sample.
lucideon.com

Here is Grok's response, to which I will comment on the accuracy and relevance of each part:

To set up a run in Probe for EPMA where you correct for beam sensitivity, you will need to follow several steps that involve configuring your analysis parameters and possibly implementing specific correction methods. Here's how you can do it based on available resources and general practices:

Step-by-Step Setup:

    Software Initialization:
        Ensure you have the latest version of Probe for EPMA installed. Check the Probe Software website for any updates or patches that might include enhancements for handling beam-sensitive samples.
    Sample Preparation:
        Prepare your samples by ensuring they are clean, flat, and appropriately mounted. Beam-sensitive samples might require special coatings or a lower beam current to minimize damage.
    Instrument Setup:
        Beam Current: Typically, for beam-sensitive samples, you should use a lower beam current. Navigate to the Acquire! window in Probe for EPMA, where you can adjust the beam current. A common range for sensitive samples might be 5 nA to 20 nA, but this depends on your sample.

    Beam Size: Increase the beam size if possible; this spreads the electron dose over a larger area, reducing damage. This setting is also adjusted in the Acquire! window.

Well nothing wrong here, all good general information and advice. It is absolutely correct that reducing beam current and/or increasing beam size will reduce beam sensitivity (TDI) effects.

Setting Up Analysis Conditions:

    Kilovolts (kV): Lower kV reduces penetration, which can be beneficial for sensitive materials. However, ensure you have enough energy to excite the characteristic X-rays of your elements of interest.
    Count Times: For beam-sensitive samples, shorter peak count times might be necessary to avoid beam damage. You can set these in the Analytical Conditions dialog box.

Hmmm... shorter count times can of course reduce beam damage effects, but I think it's difficult to say whether ion migration effects are significantly affected by higher electron beam energies. I suspect it will be very material dependent and depend greatly on the thermal conductivity of the material.

For example, a higher beam energy could create more lattice defects in crystalline materials, thus allowing more ion migration, but not as much in disordered materials such as glasses. Which is why we see larger TDI effects in SiO2 crystal as opposed to SiO2 glass.

On the other hand, watts = energy * amps, so lower beam energies mean that at a given beam current, fewer watts are concentrated into a smaller volume. Again, I think this effect will be very material dependent depending on thermal conductivity.

Has anyone looked at TDI effects in beam sensitive materials at different beam energies but the same beam current?  Here's some nominal sample temperature calculations from the equation in CalcZAF:

For pyrex glass (0.01 W/cmK):

Beam conditions of 5 keV, 30 nA,  1 um
Temperature rise =  72.00001 C

Beam conditions of 10 keV, 30 nA,  1 um
Temperature rise =  144 C

Beam conditions of 15 keV, 30 nA,  1 um
Temperature rise =  216 C

For Al metal (2.5 W/cmK):

Beam conditions of 5 keV, 30 nA,  1 um
Temperature rise =  .288 C

Beam conditions of 10 keV, 30 nA,  1 um
Temperature rise =  .576 C

Beam conditions of 15 keV, 30 nA,  1 um
Temperature rise =  .864 C

So we can see that glass is heated considerably more by higher beam energies, but not with more conductive materials.

Does anyone has actual data on the effects of increasing beam energies and TDI effects?

Beam Sensitivity Correction:

    Time-Dependent Intensity (TDI) Correction: This is crucial for correcting for beam damage or build-up effects over time. In the Analyze! window, under Elements/Cations, you can select the TDI Correction for each element if needed. This correction involves measuring the intensity of the element over time and applying a correction factor.

OK, but the TDI correction in Probe for EPMA is NOT in the Elements/Cations dialog, but rather it is in the Standard Standard Assignments dialog! Along with the interference correction assignments and blank correction assignments.  So two demerits for Grok!

Standardization:
        Use Standards: Ensure that your standards are similar in composition and structure to your unknown samples to account for matrix effects, especially in beam-sensitive materials. You might need to recalibrate more frequently if the beam affects the sample standards.

I don't need to repeat my rants here on using matrix matched standards of dubious composition again... but what it should have said is: acquire your standards also using the TDI correction method if they are beam sensitive!

Data Acquisition:
        Start the acquisition in the Acquire! window. Monitor the sample for signs of beam damage or changes in intensity over time. If you see significant changes, you might need to adjust your settings or use multiple analyses at different points or times to get stable data.
    Data Analysis:
        After data collection, analyze your results. If you've used TDI correction, ensure these corrections have been applied correctly during the quantitative analysis phase. Check for any inconsistencies that might suggest beam sensitivity issues.

Well, that's correct so far as it goes, but a bit vague so not very helpful, I will say.

Advanced Techniques:
        For more complex scenarios, you might look into MAN (Mean Atomic Number) backgrounds or polynomial background fits for elements where the background might change due to beam damage. This can be configured in the Background Type settings for standards and unknowns.

I'm not sure how using polynomial background corrections can help with beam damage during the off-peak background acquisition.  It should have mentioned using alternating on/off acquisition method instead:

https://smf.probesoftware.com/index.php?topic=812.0

But it is absolutely correct that using MAN backgrounds does remove any problems with off-peak backgrounds, since they are not acquired at all!

Consultation and Further Learning:
        Visit the Probe Software Users Forum to discuss specific issues related to beam sensitivity with other users. Topics on the forum like "General EPMA Issues" might contain discussions on similar problems.
        The training and support section on the Probe Software website offers customized training on specific EPMA methods which could include handling beam-sensitive materials. Consider reaching out for tailored advice or workshops.

Remember, the exact approach can vary based on the specifics of your sample and the elements you are analyzing. Always refer to the most current documentation or contact support if you're unsure about any procedure or setting.

Again, reasonable general advice, but not very specific.

So, I would say it is still not yet up to the task at hand, but it's getting closer I have to say.

[John Donovan]

Here is another effort to test how well AI can provide technical support, this time using the latest Grok 3 AI.  Here was the prompt we used:

How do I correct for spectral interferences using the Probe for EPMA software in EPMA analysis?  In addition to other resources, please consult the pdf guides on the probe software website (https://www.probesoftware.com/training-support/) and the user forum (https://smf.probesoftware.com/index.php) including the Getting Started manual and the Advanced Topics manual.

And here is the summary response it gave:

Correcting for spectral interferences in Electron Probe Microanalysis (EPMA) using the Probe for EPMA software involves leveraging its built-in tools and features to account for overlapping X-ray peaks, which can skew quantitative results. Spectral interferences occur when characteristic X-ray lines from different elements overlap in energy or wavelength, complicating accurate elemental quantification. Below, I'll outline a step-by-step approach based on general EPMA principles and specific functionalities of the Probe for EPMA software, incorporating insights from its documentation and user community where applicable.

Understanding Spectral Interferences in EPMA

In EPMA, spectral interferences typically arise with wavelength-dispersive spectrometry (WDS), where the high resolution can still encounter overlaps (e.g., Ti Kß overlapping with V Ka). The Probe for EPMA software provides robust tools to identify and correct these interferences, ensuring accurate compositional analysis.

Here are the step by step procedure it provided:

Steps to Correct Spectral Interferences

Acquire High-Quality Background-Corrected Data

        Start by collecting WDS data with proper background corrections. In Probe for EPMA, the software allows you to measure off-peak backgrounds on either side of the peak to account for continuum X-rays. This is critical because inaccurate backgrounds can exacerbate interference issues.

        From the "Acquire!" window, set up your elements and ensure background positions are carefully chosen to avoid other peaks. The software's graphical interface lets you visualize the spectrum and adjust these positions manually if needed.

Identify Potential Interferences
        Use the "Analyze!" window to inspect your raw data. Probe for EPMA displays peak intensities and can overlay standard spectra to help you spot overlaps. For example, if analyzing vanadium (V) in the presence of titanium (Ti), you might notice the Ti Kß line interfering with V Ka.

        The software includes a database of X-ray lines, accessible via the "Standard" menu, which you can use to predict potential interferences based on your sample's composition.

Enable Interference Corrections
        Probe for EPMA offers an automated interference correction feature, which is a standout capability. In the "Analytical" menu, select "Interference Corrections" to activate this option.

        You'll need to specify the interfering element and the interfered element. For instance, if Ti interferes with V, you'd set Ti as the interferer and V as the target. The software uses a mathematical model (based on peak overlap ratios and standard intensities) to subtract the interfering signal.

Assign Interference Standards
        For accurate correction, you must measure interference standards—pure or well-characterized samples containing the interfering element (e.g., pure Ti) but not the interfered element (e.g., no V). These standards are assigned in the "Standard Assignments" dialog under the "Analytical" menu.

        The software calculates the interference contribution based on the standard's intensity at the interfered position and scales it to your sample's data.

and this followup it also provided:

Fine-Tune with Manual Adjustments (if Necessary)

        If the automated correction doesn't fully resolve the overlap (e.g., due to complex multi-element interferences), you can manually adjust the interference parameters. The "Advanced Topics" manual (available on the Probe Software website) details how to tweak correction factors using the "Interference Correction Setup" dialog.

        This might involve adjusting the peak position or intensity ratio based on a wavelength scan, which you can perform via the "Acquire!" window's "WDS Scan" option.

Validate Results
        After applying corrections, re-run the quantification in the "Analyze!" window. Compare the corrected concentrations against known standards or expected values to ensure accuracy.
        The software outputs a detailed log (viewable in the "Log Window") that includes interference correction factors applied, helping you verify the process.

Additional Tips from Probe for EPMA Resources

    User Forum Insights: The Probe for EPMA user forum (smf.probesoftware.com) frequently discusses interference correction. Users recommend running a qualitative WDS scan across a broad wavelength range to map all possible overlaps before setting up corrections. Search the forum for threads on "spectral interference" or "WDS overlap" for practical examples.

    Getting Started Manual: This guide emphasizes proper spectrometer calibration (section on "Spectrometer Setup") to ensure peak positions align with the software's X-ray database, reducing interference misidentification.

    Advanced Topics Manual: Look into the section on "Quantitative Interference Corrections" for a deeper dive into the algorithm, which uses a least-squares fit to deconvolve overlapping peaks. It also covers multi-interference scenarios (e.g., multiple elements affecting one peak).

and it even provided a "practical example":

Practical Example
Suppose you're analyzing a mineral with vanadium (V Ka at ~4.95 keV) and titanium (Ti Kß at ~4.93 keV). After collecting data:

    In "Analyze!", you notice elevated V counts in Ti-rich areas.

    Go to "Interference Corrections", select Ti as the interferer and V as the interfered element, and assign a Ti standard.

    The software subtracts the Ti Kß contribution from the V Ka peak based on the standard's intensity ratio, yielding a corrected V concentration.

and it ended with this:

Final Notes
The Probe for EPMA software streamlines spectral interference correction with its user-friendly interface and robust algorithms, but success hinges on careful setup—accurate backgrounds, well-chosen standards, and validation. If you're new to this, the training PDFs on probesoftware.com (e.g., "Getting Started with Probe for EPMA") are invaluable for walking through these steps with screenshots. For complex cases, the forum's community can offer tailored advice based on real-world experience.

Let me know if you'd like me to dig deeper into a specific part of the process!

I highlighted a couple of sentences that appear to be problematic, but the rest of it seems pretty useful... what do you all think?

[John Donovan]

Joe Boro created a GPT helper for EPMA:

https://chatgpt.com/g/g-67ca1b54ad948191ab601cbb5b7ee2d4-epma-expert

Joe wrote:

Made this GPT, thought it might be useful for folks, it currently has access to Reed and most the Probe Software documentation and x-ray data book; I'd be happy to add other PDFS to it.

Let me know what you think – ask it some questions that you think might be difficult and/or complicated. Something like this you could think about integrating into the site maybe? Not sure.

Here's an example question and output:

"How do I view wavescans and set custom background locations and in probe software?


To view wavescans and set custom background locations in Probe Software (Probe for EPMA), follow these steps:

Viewing Wavescans

    Initiate a Wavescan
        Open Probe for EPMA and go to the Acquire! window.
        Click New Sample, then select Wavescan as the sample type.
        Enter a sample name and description, then click OK.
    Start the Wavescan
        Click Start Wavescan in the Acquire! window to begin data acquisition.
        The Wavescan Acquisition window will open, displaying live data.
    Analyze the Scan
        Use the Plot! window to display the completed wavescan.
        Select the X-Axis parameter (spectrometer) and Y-Axis (wavescan counts).
        Click Output to graph the wavescan and inspect for interferences.

 

Setting Custom Background Locations

    Open the Wavescan Data in the Plot Window
        In the Plot Graph Data window, select a sample and plot the graph.
        Off-peak background positions are indicated in green.
    Adjust Off-Peak Positions
        Click the Low or High buttons to move the background position.
        Drag the cursor over the graph and click to set the new position.
        The updated background position will appear in yellow.
    Model Backgrounds
        Click Model Background to select different background correction types (e.g., linear, polynomial).
        Observe the updated background model on the graph.

This workflow allows you to visualize x-ray intensity distributions, identify spectral interferences, and refine background settings for accurate quantitative EPMA analysis."

[Joe Boro]

Hey John, thanks for posting. Here's another example a user might experience. For environmental impact, once these models are trained and operating, they don't use much more energy than someone playing a high end video game (unless you're doing image generation).


Question:
I'm seeing artificially high Cu in my brass alloy, what might be causing that?



Answer:
If you're seeing artificially high Cu in your brass alloy, here are the most likely culprits, in priority order:
1. Zn Kβ overlap on Cu Kα
Brass is a Cu-Zn alloy, and Zn Kβ (8.905 keV) overlaps significantly with Cu Kα (8.041 keV). If you're measuring Cu with WDS at Cu Kα, and Zn is also present in significant quantities (which it always is in brass), this spectral interference can cause overestimated Cu unless corrected.
Fix:

• Either:
  
  
  • Switch Cu to Cu Lα (0.929 keV) and Zn to Zn Lα (1.012 keV) — lower energy, but manageable if your detector and spectrometer/crystal setup allows.
  • Or explicitly correct for the interference using an empirical interference correction in Probe for EPMA.
• You should also verify your PHA settings, and make sure you're not counting Zn Kβ as Cu Kα — especially important on JEOL systems without digital PHA.

2. Standard misassignment or mismatch
If the standard used for Cu has significantly different Zn content than the unknown (e.g., you're using pure Cu as the standard), and Zn Kβ is bleeding into Cu Kα, you're calibrating the spectrometer differently than the actual unknown. That mismatch makes the interference worse.
Fix:

• Use a brass standard of known composition close to the unknown.
• Or use Probe for EPMA's "virtual standard intensity" from a prior matrix-corrected value if you don't have a good match.

3. Uncorrected secondary fluorescence from Zn to Cu
Zn can fluoresce Cu via characteristic X-rays if present in sufficient concentration and beam energy is high (>15 kV). This isn't usually the major factor, but can be a few % overestimate in extreme cases.
Fix:

• Run the "Calculate Secondary Fluorescence" tool in Probe for EPMA to assess this.
• Reduce beam energy to minimize overvoltage (e.g., 10–12 kV) if resolution and X-ray generation still allow it.

4. Wavescan not done (or not reviewed)
Have you done a wavescan on your unknown sample for the Cu region? A misidentified or poorly resolved peak might be contributing — and this will also reveal any unexpected overlap, such as Ni or Ga if it's a multicomponent brass.
Fix:

• Always run a wavescan, especially if the data look off.
• Confirm that the peak position for Cu matches your standard.

Want to step through fixing this in Probe for EPMA? I can help walk you through:

• Running a Cu/Zn wavescan.
• Checking for interferences.
• Applying corrections.

Let me know your current beam conditions, spectrometer setup, and what crystal you're using for Cu.

[John Donovan]

OK, I just tried Joe's latest iteration of his EPMA Expert AI:

https://chatgpt.com/g/g-67ca1b54ad948191ab601cbb5b7ee2d4-epma-expert

and frankly I'm a bit stunned...so my prompt was: "How should I quantitatively correct for spectral interferences using Probe for EPMA?" and it replied:

To quantitatively correct for spectral interferences in Probe for EPMA, follow these key steps carefully. The process is rigorous and must be executed precisely to ensure accurate results, especially for minor and trace elements where even small overlaps can have a big impact.

Ok, well that's fair enough I think.  Then it gave a step by step procedure:

Step-by-Step Interference Correction Procedure:
1. Confirm the Interference

    Run a wavescan for each element to identify any overlapping peaks.

    Use the Calculate Interference button in the Assignment Properties window to view possible spectral interferences. This calculates worst-case scenarios using Gaussian peak assumptions.

2. Acquire Proper Standards

    For each interfered element, you must:

        Acquire counts for both the interfered and the interfering element.

        Use a calibration standard that contains a major concentration of the interfering element and none of the interfered element or any other overlapping elements.

        Example: To correct Ti Kα interfering with V Kα, you would use a TiO₂ standard that contains no vanadium.

3. Assign Interferences

    Open the Standard and Interference Assignments window from the Analyze! window.

    Click on the interfered element (e.g., V) and open its Assignment Properties.

    Choose the interfering element (e.g., Ti) and select the appropriate interference standard from the dropdown menu.

    Repeat for each interference channel.

4. Enable Interference Corrections

    Go to Analytical > Analysis Options.

    Make sure the following options are checked:

        ✅ Use Assigned Interference Corrections on Standards and Unknowns

        ✅ Force Negative Interference Intensities To Zero in Corrections (this avoids negative k-ratios due to noisy interference subtraction)

    Do not check "Do Not Use Full Quant Interference Correction" unless testing.

5. Reanalyze Standards

    Run the Analyze function again on the standards. This now includes the interference correction routine.

    Check that the previously interfered element (e.g., the phantom V in TiO₂) is now correctly reported at near-zero concentration.

6. Proceed to Unknowns

    Once corrections are validated on standards, move on to acquiring and analyzing unknowns using the same correction assignments.

🧠 Notes on Edge Cases

    If both elements interfere with each other (e.g., Zn Kα and Re Lα), you must have two interference standards:

        One with Re only, no Zn.

        One with Zn only, no Re.

    Probe for EPMA will use an iterative solution to resolve these complex cases.

Let me know if you'd like me to walk through this setup using your actual elements and lines of interest. We can go through assigning crystals, checking overlaps, and even setting spectrometer positions if needed.

This time it even gets the names of the windows, menus and buttons correct! The only thing I would mention is that where it says:

Make sure the following options are checked:

        ✅ Use Assigned Interference Corrections on Standards and Unknowns

        ✅ Force Negative Interference Intensities To Zero in Corrections (this avoids negative k-ratios due to noisy interference subtraction)

It apparently doesn't know that the first option gets automatically checked whenever you add an interferences correction.

As for the second option I do not recommend that it be checked.  Yes, this second option can help where one has incorrectly specified backgrounds or contaminated standards, and the interference correction is over correcting, but it really should only be used as a last resort for problematic data.

But otherwise, it's pretty damn good!  Wow... 

[John Donovan]

Here is the interface to Joe Boro's AI EPMA Expert which can be found here:

https://chatgpt.com/g/g-67ca1b54ad948191ab601cbb5b7ee2d4-epma-expert



I recommend you try it with some EPMA questions yourself, particularly on using various features and methods in Probe for EPMA and see how well it does.

Feel free to post your comments and observations with the results as Joe is continuing to improve it

[John Donovan]

Please also log in and vote in the poll here as it pertains to this post:

https://smf.probesoftware.com/index.php?topic=1752.0

One of the most egregious sources of error today in microanalysis (WDS and EDS), are boundary fluorescence effects from nearby phases. These errors can occur when measuring a small amount of an element in a phase, which is adjacent to another phase, where the measured element is present in higher concentrations.  For example, measuring trace Ca in olivine, which is adjacent to a Ca rich glass, diopside or anorthite phase.

These effects can be as large as weight percent level but more often are in the hundreds to thousands of PPM levels, which are obviously of concern when performing trace and minor element analyses. In fact I consistently see so called "diffusion profiles" in published work, which are clearly mostly or even all due to such fluorescence artifacts.

So I was going to write up a step by step procedure for how to model and correct for such effects, as described here:

https://smf.probesoftware.com/index.php?topic=1545.0

when I decided instead to ask Joe Boro's EPMA expert AI... so I went to it:

https://chatgpt.com/g/g-67ca1b54ad948191ab601cbb5b7ee2d4-epma-expert

logged in, and asked, "How can I determine if I am getting secondary fluorescence from boundary phases and if so, how do correct for this?"  And it replied:

To determine if you're experiencing secondary fluorescence from boundary phases in your EPMA data—and to correct for it accurately—here's how to proceed. It's a fairly complex and time-consuming process, but I'll break it into clear, paranoid steps to keep your data solid.

Step 1: Diagnose the Problem
Indications You Might Be Seeing Secondary Fluorescence

    Elevated minor/trace element concentrations near phase boundaries.
    Higher totals near boundaries (especially for incompatible elements).
    Anomalous element maps that don't match expected diffusion or zoning profiles.
    Elemental ratios in unknowns not matching the standard calibration.

If you're unsure, collect linescans or maps across the phase boundary and look for intensity increases on the side without the fluorescing element. If you see this, we need to run a proper correction.

Step 2: Use PENEPMA for Secondary Fluorescence Modeling

The most rigorous correction method involves Monte Carlo modeling with PENEPMA. Probe for EPMA supports this via a multi-step process involving .MAT, .PAR, and .DAT files.

Step 3: Create Material Files

    Open Standard.exe in Probe for EPMA.
    Choose: Analytical > PENEPMA (Secondary Fluorescence Profile) Calculations.

    Define:
        Material A (beam incident phase),
        Material B (boundary phase),
        Material B Std (a pure standard of the fluoresced element).

    Be sure to input accurate densities. Insulators require oscillator strength and energy = 0.

Step 4: Generate Parameter (.PAR) Files

Run:
    Run Penfluor – for fluorescence simulation.
    Run Fitall – fits the intensity vs. energy data.

Each .PAR file takes up to 8 hours to compute. These simulate photon emission and absorption for each geometry.

Step 5: Run FANAL to Calculate k-ratios

Use:
    Run Fanal – inputs your .PAR files and outputs corrected intensities and k-ratios.

These account for:

    Primary excitation.
    Characteristic and continuum fluorescence.
    Absorption geometry effects.

So pretty good so far.  Then it continued with the correction steps:

Step 6: Subtract Fluorescence Contribution

    Load your original analysis in Probe for EPMA.

    Use:
        Calculate Secondary Fluorescence under Analytical > CalcZAF.

Well that's not exactly correct. It seems to be confusing CalcZAF with Probe for EPMA. Yes, one can correct analyses both in CalcZAF and Probe for EPMA, but it's much simpler in Probe for EPMA and involves clicking the Boundary Corrections button in the Analyze! window.

It continued:

Input the boundary distances—either:

        fixed value,
        2 stage coordinates,
        angle and distance,
        or a calibrated BMP map.

    The fluorescence-corrected weight percent is then:
    Ccorrected=Cmeasured−Cfluorescence
    Ccorrected�=Cmeasured�−Cfluorescence�

    (Where fluorescence is derived from Fanal-calculated intensity differences.)

    Note: This is a first-order correction—the matrix correction should ideally be recalculated afterward for rigorous accuracy.

This last statement is incorrect. Both CalcZAF and Probe for EPMA perform the boundary fluorescence correction iteratively within the matrix correction. It is only necessary to perform iterations manually if performing the subtraction in say an Excel spreadsheet.

Step 7: Export Results for Review

    Click Send to Excel in the correction window.
    Review Calc. B Flu Conc. % and Meas. Total Conc. % to quantify the correction magnitude.

Extra Tip: Watch Out for Inverse Effects

If your beam incident phase fluoresces but the boundary doesn't, you may observe a drop in intensity near the boundary due to x-rays escaping instead of being absorbed locally.

If you'd like, I can walk you through an example using your actual materials. Just tell me:

    The composition of your matrix phase (Material A),
    The composition of the adjacent phase (Material B),
    The target element,
    And your beam energy + geometry.

Want to continue with that setup?

That is reasonable enough. I replied "Yes" and it replied:

Step 1: Define Materials and Conditions

Please fill in the following:
1. Material A (Beam-Incident Phase)

    Name:
    Composition (in wt% or oxide % preferred):
    Density (g/cm³ if known, or I can estimate):

2. Material B (Boundary Phase)

    Name:
    Composition:
    Density:

3. Target Element (element you're trying to correct for):

4. Operating Conditions

    Beam Energy (kV):
    Takeoff Angle (degrees, usually 40–50° on JEOL):
    Approximate boundary distance (μm) from beam center to boundary phase:

        Or do you want to provide two stage coordinates instead?

Once I have that info, I'll:

    Build the .MAT inputs for PENEPMA.
    Suggest default modeling parameters.
    Walk you through generating .PAR and .FAN files.
    Show how to interpret the correction back in Probe for EPMA.

Shoot me the values and we'll get modeling!

At this point I did not continue further, but I have to say that I'm pretty impressed all in all.

Currently Joe's EPMA AI expert is NOT utilizing the information in this user forum, but he is working on that so soon there should be a significant improvement in accuracy.

[Probeman]

Today's AI LLMs seem to be useful tools for ascertaining the current state of knowledge on particular subjects even if they cannot (yet) synthesize new conclusions.  But I wanted to see how close they could come to "thinking", so here are the results of a short conversation I recently had with the Grok AI...

I started by asking it what methods or procedures in general should be utilized to improve accuracy in quantitative microanalysis and it gave a reply that covered the "usual suspects".

But in its reply I noted that it mentioned using "matrix matched" standards for best accuracy, so decided to explore that question further by mentioning that due to improvements in modern matrix corrections, wouldn't it be problematic to utilize "matrix matched" standards if they weren't well characterized or as homogeneous as high purity synthetic minerals? 

The underlying problem I suspect being that the use of "internal" natural standards with variable heterogeneity can result in the fact that various EPMA labs generate very reproducible results, yet they disagree with the also very reproducible results from other EPMA labs. Yes, I think that most labs like to flatter themselves in thinking that "well, I know our results are accurate", but these results show consistent systematic differences between each other, and we know they can't all be correct!

As noted in the recent paper:

Wieser, P. E., Kent, A. J., Till, C. B., Donovan, J., Neave, D. A., Blatter, D. L., & Krawczynski, M. J. (2023). Barometers Behaving Badly I: Assessing the Influence of Analytical and Experimental Uncertainty on Clinopyroxene Thermobarometry Calculations at Crustal Conditions. Journal of Petrology, 64(2), egac126.

this figure reveals the issue:



Penny Wiser commented to me about the above figure:

Cpx composition measured at MIT by Mike Krawcynski - and calculates the Cpx-only pressure and Temp, and takes a paired Cpx-Glass composition and calculates the Cpx-Liq pressure. We then use the interlaboratory comparisons of Gale to say 'well what if Mike had instead measured that glass and Cpx at other labs'. And then we take these corrected compositions and calculate the P and T again. This shows that just the lab you measure the sample in can change the pressure and temperature by 2.5-6 kbar, this is comparable to the stated error on these expressions. But this is a systematic offset. If a set of samples was measured at Lamont not MIT, you could think the Cpx were crystallized on average 6 kbar deeper.

In other words, the science of thermobarometry is currently limited by the accuracy of EPMA results. Now some of these differences between these various labs are surely due to the choice of matrix corrections, but given that the variation in modern matrix corrections are so small, I think it is evident that most of these systematic differences are due to the lack of global high purity, homogeneous synthetic mineral standards. Yes, there are probably other factors at work here as well (dead time calibrations, spectrometer alignment, etc.), but until we have a truly global set of primary (and secondary) standards we will never be able to determine the source and magnitude of these systematic errors between various labs.

Now, Will Nachlas with the support of the MAS council, is attempting to right this wrong, by procuring and characterizing various commercial (and private) sources of such synthetic materials, and in a recent conversation with him, it seems he is making excellent progress,

Anyway, back to the AI conversation on matrix corrections vs. standard compositional accuracy, it replied:

Yes, given the advancements in analytical models for matrix corrections achieving accuracies better than 2% (often around 1% relative for silicates and oxides), high-purity, end-member stoichiometric synthetic standards are generally preferable over heterogeneous natural materials in quantitative electron probe microanalysis (EPMA).  This preference stems from several key factors:

Synthetic standards offer assured purity and fine-scale homogeneity, which are critical for reliable calibration and reducing variability in X-ray intensity measurements.   Natural standards, while sometimes well-characterized (e.g., Kakanui hornblende for elements like Na and Ca), often suffer from micron- or submicron-scale heterogeneity, inclusions, oxidation, or hydration, leading to inaccuracies in analytical totals and compositions.   For instance, natural sulfides like chalcopyrite can oxidize differently from unknowns, causing systematic errors that require frequent re-polishing—issues largely avoided with synthetics.

Reduced Reliance on Matrix Matching

Modern matrix correction algorithms (e.g., phi-rho-z models like PAP or Armstrong) can accurately handle large extrapolations (up to ~40% in systems like SiO₂-Al₂O₃-MgO-CaO) without significant loss in precision, making close matrix matching less essential for most analyses.   The primary limitations in EPMA accuracy are often instrument calibrations (e.g., dead time, spectral interferences, background measurements) rather than the corrections themselves.  Synthetics, such as MgO, Al₂O₃, or MgAl₂O₄ from projects like FIGMAS, enable robust testing of these extrapolations and provide a stable reference, even for complex minerals. 

Practical and Efficiency Benefits

Using synthetics simplifies workflows by reducing the need for a large library of matrix-matched natural standards, saving time on maintenance and verification.  They are easier to mount (e.g., in acrylic for re-polishing) and less prone to environmental degradation.  However, for specific cases like light elements (e.g., F in apatite) or volatiles, where absorption or fluorescence effects are pronounced, standards with similar compositions (potentially natural) may still be advisable to minimize correction magnitudes, though advanced methods like PAP can often compensate. 

In summary, the high accuracy of current matrix corrections shifts the emphasis toward standards that prioritize purity and homogeneity, making synthetics the better choice for most quantitative applications unless specific matrix effects demand otherwise.

Now is the AI just trying to be agreeable or did it actually change its mind?  Who knows, but it looks hopeful.

Even more hopeful is the work that Will Nachlas is doing obtaining quantities of high purity synthetic minerals for truly global EPMA calibrations.