Week of March 22nd

What I did

• Monday: Train ride back to Boston. Wrote python script to generate .inp files with different radially dispalced atoms
• Tuesday: Cleaned up the code from monday
• Wednesday: Gaussianator.py from start to pretty far
• Thursday: Gaussianator.py mostly finished. Spent a long time on the gaussian averaging function and spline options
• Friday: Gaussianator.py technically finished.
• Saturday: Worked on getting FEFF to work. BASH in windows is fine but I found a bug on the third .exe file called.
• Sunday: Bug fixed to create proper energy mesh for 1d interpolation in Gaussianator.py

What I learned

• Monday: You can create dir and write files in them without having to os.chdir() into them. Just specify the FULL filepath when you go to write the file. It won’t work if you specify the relative path.
• Tuesday: XANES have a greater sensitivity to changes in structure than EXAFS do, and can detect changes as small as .0001 angstroms maybe.
• Wednesday: To fix non-standardized independent variable, I mapped the points onto a standardized “Energy Mesh” via an interpolation function.
• Thursday: 1d interpolation worked well, but required the mesh to be within the points to map’s range. Cubicspline does not require this and allows you to map points outside of the original range. (If the data is [1,…,9] you can map into onto a mesh of [0,..,10]).
• Friday: Questions about the cubic spline interpolation vs. linear interpolation remain. The resolution is high (i.e. \(\Delta x\) between the discrete x-axis values is low), which means the 1d-interpolation is probably good enough. The cubic spline looks just as good, though, and means I could just use the df.loc['O'].omega values for my energy mesh. If I can show that at high curvature areas on the XANES spectrum, that cubic spline does a better job interpolation than 1-d, I will keep cubic spine. I suspect that the higher number of datapoints on the mesh near the peaks means that 1d-interpolation will perform about as well and be more predictable.
• Saturday: I’m wondering if the problem is the feff.inp file.
• Sunday: Insertion sort is most efficient for nearly already sorted data.

What I will do next

Get FEFF to work and run the simulations.