The Verma et al. (2003) version of the normative mineralogy calculation was used as the basis for this work. Verma et al. made a number of revisions to previous normative calculations, including i) the inclusion of minor and trace elements; ii) correction of errors in formulae and molecular weights for apatite that had existed since Cross et al. (1902); iii) inclusion of the concept of variable molecular weights to account for substitution of trace elements and calculation of free or unused oxides from remaining mass balance of elements. Their revisions resulted in very consistent results compared to previous calculations, with the absolute difference between the sum of the normative minerals and the bulk chemical analyses typically less than 0.002% (Verma et al., 2003). This version of the normative calculation was initially distributed as part of the “Standard Igneous Norm and Volcanic Rock Classification System” (SINCLAS) (Verma et al., 2002). Subsequently, IgRoCS incorporated the standard igneous norm procedure with some minor corrections to the normative calculation, and added the ability to process multiple samples in one run (Verma and Rivera-Gómez, 2013).

Python is an interpreted high-level programming language which has become one of the most widely used and popular programming languages (Perez et al., 2011). Over multiple decades a large ecosystem of open-source third-party libraries has been developed, providing a vast range of functionality. Python also works cross-platform, allowing use within Mac, Windows or Linux operating systems. These factors made it a good choice for use in developing a new, easy to use implementation of the normative mineralogy algorithm.

The algorithm steps described by Verma et al. (2003) were translated to Python code and where possible, written to take advantage of the high performance of vectorized functions in NumPy (Harris et al., 2020). Our implementation requires the major oxides as a minimum (SiO₂, TiO₂, Al₂O₃, Fe₂O₃, FeO, MnO, MgO, CaO, Na₂O, K₂O, P₂O₅ as weight percent), and optionally minor and trace elements (CO₂, SO₃, S, F, Cl, Ba, Co, Cr, Cs, Li, Ni, Rb, Sr, V, Zr as parts per million). The major oxides are first adjusted to 100% on anhydrous basis, and then a second adjustment to 100% including the minor and trace elements is performed, as per Verma et al. (2003). Trace elements are added to certain major oxides and variable molecular weights calculated. A generalised overview of the algorithm is shown in Figure 1. For further information on the specifics of the normative mineralogy algorithm we refer the reader to the details in Verma et al. (2003). The code developed in this work also leverages many of the functions available within the geochemistry-focused Python package pyrolite (Williams et al., 2020), and the final algorithm has itself been included as a submodule function within the latest release of pyrolite.

The web application utilises the Python based web framework library Streamlit, which is designed to allow for rapid development of web applications. In deploying the implementation as a web application, users are not required to install anything in order to access it (requiring only a modern web browser), updates for bugfixes and new features are easily distributed (users always have the most up to date version), and the maintenance burden for the developer is minimised.

The majority of whole-rock geochemical analyses report the iron content of a sample as total Fe₂O₃ or FeO depending on the method of analysis (XRF, ICP, etc.) and on the convention for the type of rocks analysed and the laboratory undertaking the analysis. However the normative mineralogy calculation requires both to be specified. Verma et al. (2003) discuss a number of different methods to apportion the total Fe to Fe₂O₃ and FeO, and in IgRoCS (Verma et al., 2002) three methods are implemented: Le Maitre (1976), Middlemost (1989) and a user defined adjustment for cases where the ratio is known (for example, based on an assumption of lithology or if a titration for Fe has been conducted to measure the ratio of Fe₃ + to Fe₂ +). The Middlemost (1989) method defines a Fe ratio based on the total alkali-silica (TAS) classification of each sample. The Le Maitre (1976) method uses an adjustment based on the SiO₂, Na₂O, and K₂O content of the sample, with separate formulas for volcanic (Eq. 1) and plutonic (Equation 2) rocks, though only the former is used in IgRoCS. F e O F e O + F e 2 O 3 = 0.93 − 0.0042 × S i O 2 − 0.022 N a 2 O + K 2 O (1) F e O F e O + F e 2 O 3 = 0.88 − 0.0016 × S i O 2 − 0.027 N a 2 O + K 2 O (2)

Both the Le Maitre (1976) and Middlemost (1989) method have been implemented in this work, with an additional option in the web app to specify a column in the user’s uploaded data that contains a specific Fe adjustment ratio for each sample or a fixed value for all samples.

To confirm that the algorithm is performing as expected, the output from webNORM was compared to the output from IgRoCS. A test dataset of 551 bulk rock geochemical samples with major oxide values is provided with the IgRoCS software and was used as the basis for our comparison. The Fe Adjustment was calculated using both the Le Maitre (1976) and Middlemost (1989) methods for all samples. As IgRoCS only uses the volcanic variant (Equation 1) of the Le Maitre (1976) method, the plutonic variant could not be compared.

The Le Maitre (1976) adjusted data was used for the input to the normative mineralogy calculations. Normative mineralogy was calculated in both IgRoCS and webNORM twice: initially with only the major element data, and subsequently incorporating major, minor, and trace elements. The normative algorithm as described by Verma et al. (2003) should result in a normative mineralogy that sums to within ±0.002% of 100%. Herein we use this as a general indicator that our implementation of the Verma et al. (2003) algorithm is performing as intended. For a number of samples, IgRoCS does not calculate the normative mineralogy where webNORM does, and these samples are excluded from the comparison.

A comparison of the normative sum between webNORM and shinyNORRRM (González-Guzmán et al., 2023) was also made. Comparison of individual normative mineralogy between webNORM and shinyNORRRM proved difficult due to shinyNORRRM not including minor and trace elements when readjusting to 100%. A dataset of 7,019 samples provided with shinyNORRRM (‘Deccan.csv’; González-Guzmán et al. (2023)) was used for this comparison.

When using the Le Maitre (1976) adjustment, the FeO and Fe₂O₃ values from IgRoCS and webNORM show a very good agreement (Figure 2). When rounded to 3 decimal places, only one sample had a difference of < 0.002% for Fe₂O₃, and none for FeO. The maximum difference between the two calculations for Fe₂O₃ was 0.003%.

For the Middlemost (1989) adjustment method, there were more discrepancies between IgRoCS and webNORM (Figure 2). When rounded to 3 decimal places, 22 samples had a difference of < 0.002% for both Fe₂O₃ and FeO. The maximum difference for Fe₂O₃ and FeO was 0.675% and 0.741% respectively. Further investigation showed that the samples that show the errors all fall close to the TAS class boundaries (Figure 3).

When using the Le Maitre (1976) Fe adjusted data with only major oxides as the input, the results from IgRoCS and webNORM generally show a good match. The results from webNORM show a very consistent normative sum, with 20 samples deviating from 100% by more than ±0.002%, and a minimum sum of 99.998% and a maximum of 100.003%. This is compared to 58 samples deviating from 100% by more than ±0.002% in the results from IgRoCS. For normative diopside, there is a notable discrepancy between IgRoCS and webNORM in 47 samples. When looking at the deviation from 100% in the normative sum for just these samples, all show a deviation of ±0.002% from 100% in the webNORM results compared to a mean of ±3.2% (with a maximum of ±56.7%) for IgRoCS.

When calculating normative mineralogy using major and trace elements (with the Le Maitre (1976) Fe adjusted data), 28 samples show a normative sum deviating by more than ±0.002% (−0.09% to +0.003%) compared to 56 samples from IgRoCS. The previous issue with diopside still remains, and 120 samples show discrepancy between IgRoCS and webNORM of > 0% in either corundum, apatite, anorthite or fluorite (Figure 2).

The mean deviation of the normative sum from 100% is lower in the webNORM results compared to IgRoCS for the erroneous samples for apatite (±0.033% vs ±0.691%), fluorite (±0.092% vs ±1.418%) and anorthite (±0.101% vs ±0.598%). For samples showing a discrepancy in normative corundum, the mean deviation of the normative sum from 100% is larger in webNORM (±0.081%) than in IgRoCS (±0.046%).

When comparing webNORM and shinyNORRRM, both show a very consistent normative sum that averages 100.000%. The standard deviation of the sum is marginally lower from shinyNORRRM (0.001% vs 0.003%). Out of the 7,019 samples tested, the normative sum is > ± 0.002% for 67 samples from shinyNORRRM compared to 107 from webNORM. The minimum and maximum deviation from 100% is −0.004% and +0.004% from shinyNORRRM compared to −0.087% and +0.003% from webNORM.