Petrologie und Geodynamik

Background Correction

Linear vs. Curved Background - Introduction

In EPMA, the x-ray background spectrum is mainly a result of the continuous spectrum which has a changing curvature. 

For applications, where element concentrations approach the minor and trace element level (as with monazite microprobe dating), it is important to perform accurate background approximation. Online sofware provided by commercial electron microprobe manufactures  however, perform linear background approximation. If background offsets are large and element concentrations approach the detection limit, errors introduced by linear background approximation for a curved (real) background can become significant. For high-quality U-Th-total Pb dates it can therefore be important to consider the curved background and include some kind of correction (see <a href="">Jercinovic et al., 2008</a>). To my knowledge, with the exception of one commercial third-party-software product, there is currently no standard procedure for this kind of background correction in EPMA. 

For electron microprobe dating most people use the Pb Ma emission line of Pb, which has high intensities but suffers from line interference. To avoid this I prefer the Pb Mβ emission line (@2.4427 keV), which however has lower net intensities (roughly about 30%) but can be used with much lower background offsets. 

Motivated by discussions and reviews I received, I was interested to know how the error of a linear background approximation by using Pb Mβ with our offset settings could be.

I made a number of WD scans on monazite with different composition. These scans provide the basis for the background modelling. Here I will present the scans and show how I performed the background modelling by using the numerical software Octave.. 

The scans were performed from 2.206 keV to 2.579 keV (equals a x-tal to sample distance from 180 mm to 154 mm, respectively, with 140 mm Rowland radius) on PET and PETH crystals with 2601 steps. Counting times per step are 3 seconds and the probe current was 60 nA at 20 keV. Compared to similar experiments from other labs this probe current is relatively low, but avoids the decrease of intensities due to thermal degradation on the sample and on the coating. The scans that I present here are all made with the PETH crystal on a Jeol JXA 8900 spectrometer. This so-called H-Type spectrometer has a Rowland circle of 10 cm and is also used for the quantitative analysis. It uses a sealed xenon counter.

Octave/Matlab script for background calculations

For each scan, 3 diagrams are shown, that were generated by Octave:

The uppermost diagram simply shows the raw scan data. 

  • The scan in the middle is based on the same data (with different scaling of the intensity axis) but without those parts of the scan, where interfering peaks are. The theoretical positions of various lines are also shown in the diagram. The parts of the scan which are shown, were used for the fitting of the spectra.
  • The lower scan shows in blue the filtered spectra and in red and green the lines for the polynominal (green) and exponential (red) fit. 
  • To smooth the scatter of the scan a Savitzky-Golay Filter was applied to the noisy data. This filter is suited because it preserves quite good the peak widths and other original features of the scan.

In many cases, there is almost no visible difference between the polynomial and the exponential fits; they appear to plot on each other. Only at high magnification you will see the minor differences between these two lines. You can click on the images to enlarge the view - double click to return to the original size.

What is most interesting  here is the difference in background intensities at the position of Pb Mβ between the non linear background fits and the  linear background approximation. The Octave script calculates these values and also the differences. Using these differences, we can calculate the difference in composition, resulting from the different background models and from this also the change in Th-U-total Pb ages. 

Our Jeol microprobe typically measures values of 2.3 cps/nA/wt% PbO by using Pb Mβ  at 20 keV with the H-Type spectrometer.

If, for instance, we have a difference of 0.00032 cps/nA between the value for the  polynominal background approximation and the   linear approximation, we can estimate the difference in PbO caused by the two background approximations from 0.00032/2.3 wt% PbO. This results in about 1 ppm PbO. The difference in age, calculated from this, is less than 1 Ma for younger (Paleozoic and younger) and Th-poor monazite and less than 0.5 Ma for Th-richer, older monazite.


Examples of Monazite Scans

Intermediate Th content: Madmon

This monazite is from Madagascar and contains about 10.4 wt% ThO2 and 0.25 PbO. The grains were kindly provided by  B. Schulz, Freiberg. For this monazite, two scans were made to show the influence of detector settings. The most relevant scan, however, is the second one, where detector settings are in the same mode as during quantitative analysis.

  (1) PHA settings in integral mode. 

WD scan integ

For this scan, the discriminator  was set to integral mode in the pulse height analysis, thus setting only a lower threshold to eliminate electronic noise. By using this setting, all detector input signals are counted. This results in higher background count rates and in the detection of peaks of higher order. A PETH crystal (Rowland circle radius of 10 cm) and a sealed Xenon detector were used.

The difference between linear and polynominal fit is 0.0058 cps/nA, resulting in 25 ppm PbO difference. This is equivalent to a difference in age of about 5.2 Ma.

(2) PHA settings in differential mode.</b> Now, we will have a look at a scan on the same monazite but with pulse height analysis set to differential discriminator mode (defining an upper and lower threshold) and detector voltage set to the best value for Pb Mb analysis. Therefore these are the same settings that are used for quantitative measurements of the Pb Mb line:



WD Scan Diff Mode Mammon

This scan is very different from the one above and it nicely demonstrates the power of pulse height analysis. First, background intensities are significantly lower throughout the whole scan. I assume that this is mainly an effect of eliminating contributions from various higher order x-rays from the continuous spectrum. Second, the higher order peaks of Ce and La are strongly reduced by more than 90%.   Because these detector settings are also used when Pb is analysed, the  background fitting should be based on scans that were obtained in this way. 

The difference in intensities is 0.0055 cps/nA between linear and polynomial approximation and 0.010 cps/nA between linear and exponential approximation. This leads to a difference in PbO of about 24 and 43 ppm, resulting in an age difference of 6 and 9 Ma, both in the same direction relative to the linear background model.


High Th Monazite

This monazite is from India and contains about 20 wt.% ThO2 and  wt.% PbO (sample TK35-6 monazite 12). In this case, the difference between linear and polynominal approximation is 0.0064 cps/nA and therefore somewhat higher. This causes  a difference of 0.003 wt% PbO, or 30 ppm in lead. This results in a difference in age of about 3 Ma.

The difference between linear and exponential fit is 0.0009 cps/nA, resulting in 4 ppm PbO difference. This is equivalent with a difference in age of less than 0.5 Ma.



Monazite TK 35


Low Th Monazite

The sample is from southern India (M31-1, monazite 2) and has a composition of ThO2 5.8 wt%, PbO 0.18 wt% and UO2  0.20 wt%. The spectrum reveals that this sample contains significant amount of sulfur, which is missing in the other monazite. Due to this, the fitted part of the spectrum is different to the other samples. 

The calculations yield a difference of about 0.008 cps/nA, which results in a PbO  difference of about 35 ppm. The  related difference in age is about 12 Ma.

Monazite Scan on ZA


Very low Th Monazite

This monazite is from a charnockite from Jebel Uweinat in NE Sudan. It contains very low contents of Th and Pb and it turned out to be not suited for electron microprobe age dating. 

Nevertheless, for the sake of completeness, here are the results: The difference between the linear and the polynominal background approximation is 1.1832e-04 cps/nA. Even less than that of sample madmon. The PbO content thus is also less, the calculation yields about 0.5 ppm. 


Some general trends can be observed from the scans.

  • Scans performed with differential PHA settings show that interfering peaks can be eliminated to the order of about 90% when PHA settings of the analyte line are used.</li>

The lower the Pb content, the higher the differences (errors) due to different background models.</li>

  • Even in the case of low Pb monazite, the differences in ages due to different background models appear to be much lower than those that were  for the Pb Ma emission line (<a href="">Jercinovic & Williams, 2005</a>, <a href="">Jercinovic et al., 2008</a>, <a href="">Pyle et al., 2005</a>). I assume that this is mainly caused by much larger background offsets that are necessary for Pb Ma.


If you like to perform the background approximation yourself, or if you are just curious to have a look on the script you can download the Octave code. Octave is almost fully compatible to Matlab, so the script should also run in Matlab.

A Perl script that removes data in a predifined range from the input file can be downloaded here.. This script also calculates the l-value and the wavelength (in Ångström) for each step and adds these values to the output file that is used by the Octave script.


P. Appel