PREP Blog Post #2

Context:
First off, my research deals with the field of Molecular dynamics as a whole, which is a computer simulation method for analyzing the physical movements of atoms and molecules, which in many real world applications to predict properties of a molecular system.

Since these are computer simulated results of real phenomena, it is vital to assign and apply proper statistical models and tests to draw appropriate conclusions. 

Gap:

This leads into the gap that my research is addressing, as many researchers who carry out molecular dynamics simulations have not done rigorous quantitative analyses on whether their model systems are fully equilibrated (needed to draw conclusions).

This could be greatly affecting their possible conclusions and scope of inferences.

This is where my project comes in, as an easily accessible and user-friendly program to aid researchers in quantifying and performing statistical tests on molecular dynamics data.

Process:

This process that was taken in order to fulfill this overall goal is described as follows:

• The program was built in Python due mainly to convenience and accessibility of use.

• The program outline roughly follows the tests suggested by Schiferl and Wallace in, Statistical errors in molecular dynamics averages, 

  • Four separate statistical tests: a test for lack of trend in the X, using a Mann-Kendall test, a test for lack of trend in the standard error S, using another Mann-Kendall test, a test for normality of X, using either a W test or shape test, and lastly a test for lack of positive correlation in the X using a one-tailed von Neumann test.

  • The Mann-Kendall test is of particular importance in determining when the data is equilibrated as it is a non-parametric test for trend.

(for a more detailed program outline see the "Journey to Research" page)

Progress:

Here is my current progress as of now:

• Currently most of the major statistical test modules have been developed.

  • This includes their associated procedures of calculating test statistics and confidence intervals. 

  • Specifically the Mann-Kendall a confidence interval can be found when given start time points and time steps. 

  • Currently working on linking all the tests and modules together and bug fixing as well as polishing the product.

  • Furthermore, I need to work on the documentation for the program. 

• As for changes I made:

  • Originally I intended only for the program to perform the tests, but later I decided on graphing the data in order to help me and other potential users better visualize and grasp the shape of the data and help them understand why a statistical test might have resulted in a specific outcome.

    • Specifically, in order to better visualize the data and sanity check statistical tests, a module was developed to read in a graph the data using MatPlotLib and NumPy.

  • Other than that, most changes made have not been at a large level, and mostly come in the form of bug fixing, debugging and polishing the program.