MAX UnMix

Welcome to MAX UnMix. If it is your first time using the program it will be helpful to visit the 'Resources' page where there are instructional videos and the research paper describing this program. If you're familiar with unmixing coercivity data already, feel free to begin by following the instructions on the individual pages.


Operating Instructions

Use the control panel on on the right side of the page to upload your data. Data for upload should represent either demagnetization or acquisition data. If you would like to explore the program without uploading data please select the 'Use Example Data?' option. Be sure to deselect this option when you are ready to upload data.

To upload your data, select the 'choose file' option at right. You will need to then select if your data file has a header or not and what type of separator is being used. It is recommended to upload either .csv or .txt files.

NOTE: Your data file should have field values in column 1 and magnetization values in column 2

Once you've uploaded your data you can view both magnetization data and the coercivity spectra on either log or linear scales by selecting options from the drop down menus



Below you can control the smoothing of your data. As a default, smoothing is performed on magnetization data before coercivity data is derived (see option 'Magnetization Smoother' below). There is an additional option to smooth the coercivity spectra derived from your raw magnetization data. Noisy datasets produce spurious features from smoothing that should be avoided if possible. It is suggested to test each method with variable smoothing factors to determine an adequate coercivity distribution for fitting.



The options above allow you to remove outlier data points that may have a large influence on your smoothing. Each data point can be removed individually by clicking on the plot. To remove a portion of the data, click and drag over the desired data and select 'Toggle'. To clear all edits to the data, select 'Reset'



Residual Sum Square (RSS):


              

Common Components:


Step 1: enter your maximum field value in the upper left corner, using either log or linear option

Step 2: Select any combination of the 6 components listed at left. When you select a component, a panel will appear at the right side of the screen that will enable you to set values for that components parameters. Note that this panel is movable by clicking and dragging.

Step 3: Use the sliders to adjust the mean coercivity, dispersion, relative proportion, and skewness for each of the selected components. As you adjust these parameters you will be able to watch you model (represented in orange) begin to replicate your data (shown as black data points; Note that all fitting is perfomed relative to your spline fit. If you selected a high smoothing factor on the previous page your measured data will be visible in the background as light grey data points). As you adjust your model, pay attention to the residual sum square value listed at left. The goal of component fitting is to minimize this value.

Within a session it is possible to save components . Simply click the button for each component and enter a name in the pop up window. When you upload a new dataset or change any inputs you can select the 'revert to last save' option to recover your saved component.

Step 4: When you are comfortable with you initial fit, proceed to the 'Optimization' page.

The table at left lists a series of common magnetic components that have been identified by previous work (see Elgi, 2003 and 2004 parts 1-3 linked in the Resources tab). These are included here for reference, but have no bearing on the actual analysis and do not necessarily correspond to the components available to select above.

Note: click and drag to move this panel.


                  

The plot above represents the optimal fit achieved by MAX UnMIX given the inital parametes you set on the 'Fitting' page. The RSS is displayed in the upper right corner. Comparison of this RSS value to the RSS value on the 'Fitting' page should confirm that your inital fit has now imporved.

It is advised, unless you have already analyzed a number of similar samples, to spend time comparing various models (e.g., 2 component vs. 3 component) using this page and the 'Fitting' page. In order to compare fits follow the steps below.

Step 1: Copy the RSS value and paste it in the box labeled 'paste RSS here for comparison'.

Step 2: Return to the 'Fitting' page and either adjust the initial parameter values for your components or add/remove components. It is suggested to use new components and 'uncheck' the components used in previous models. Doing this will save the initial parameters from previous models in the case that you'd like to revert to that model. Once you've constructed a new model that you would like to test against your previous model return to this page.

Step 3: The results that are displayed now represent your most current model optimized from your updated inital parameters. Compare the new RSS with the one you copied from your previous fit. If the RSS has decreased, your new model has a better fit with your data. If the new model produces a higher RSS value, you can revert to your previous model by returning to the 'Fitting' page and selecting the proper components.

Step 4: When you have arrived at a model that provides a robust fit statistically and can be reasonably interpreted in light of prior knowledge about your samples magnetic mineralogy proceed to the 'Error Analysis' page.


Residual Sum Square (RSS):

              

F-statistic

              
p-value

              

This F-test is testing the Null hypothesis that your less complicated model is best. If the reported F-statistic is much greater than 1, and the p-value is less than 0.05 you can confidently reject the Null hypothesis and support a more complicated model.



export results in log units export eps export results in normal units export png

To perform error analysis, click the 'START' button above. You may vary your number of resample event from 100 (default) to 1000.

You are also given the option to vary the proportion of data that is being using during each resample. This routine may become more important if unmixing is being performed on datasets with large field increments.

If the error analysis is finished and you are satified with your results you may export the data file that is printed in the results window. Simply type in a file name (without extension) and click 'export data file'. A .csv file of your results will download to your computers download directory. In future versions of MAX UnMix it will be possible to export the final figure as a .eps file, stay tuned!

Results:

                

              

User Videos

Below are some brief instructional videos to help get you started with unmixing your magnetic data. Links to previous magnetic unmixing methods and application papers are provided in the sidebar. This program is still in its infancy so please report any issues you come across to Dan Maxbauer (maxba001@umn.edu). Enjoy the unmixing!


Data Upload tutorial:


Fitting and Optimization tutorial:

Error Analysis and Data Export tutorial: