references/*.pdf and references.bib
===================================
References on a similar analysis performed in the Ceder group.

Methodical differences:
  - they look at ions (element,formal oxidation number)
  - they evaluate 'which ions are found in position x of the
    structure prototype', independent of the actual chemical
    environment
  - machine-learning via maximum-likelihood estimation


Pair correlation function
=========================

Comparison with Hautier paper
-----------------------------
The first paper (Hautier,2011) displays their estimate of the
pair correlation function
  g_{AB}=P(A,B)/(P(A)*P(B))
in Figure 3, log(g_{AB}) as color, on a subset of ions in
Pettifor ordering.

I produced a similar plot for easier comparison in
compare_hautier/paircorrelation_threshold_none_order-hautier.pdf

As we did not evalute the oxidation number, in to keep the
scale identical to their figure, I replicated the data as necessary.

Full data
---------
Pair correlation function for the full periodic table:
paircorrelation_threshold_none_order-Z.pdf        (ordered by Z)
paircorrelation_threshold_none_order-pettifor.pdf (Pettifor order)

As already suggested by Antonio, the pair correlation function
over-emphasizes badly represented elements; this is especially evident
in the Pettifor-ordered representation


Periodic table plots
====================

PT_threshold_*_N_other_elements.pdf
-----------------------------------
Number of substituting elements, both with and without thresholds

PT_threshold_*/periodic_table_substitutions-*.pdf
-------------------------------------------------
Substitution rates for each element, both with and without thresholds.
e.g. PT_threshold_none/periodic_table_substitutions-Na.pdf

Files in raw_data:
==================

convention: 0-based arrays
indices correspond to Z
Z=0 used for dummy or global data

substitution.dat.element_name
-----------------------------
Names/Symbols of elements, line-by-line in order of increasing Z,
Z=0 is 'dum'my

substitution.dat.element_count
------------------------------
number of (non-duplicate) materials containing each element
line-by-line in order of increasing Z
Z=0 is _total number_ of materials

substitution.dat
----------------
NxN matrix, row/column indices correspond to elements (increasing Z)
(see substitution.dat.element_name)
first row/column (corresponding to Z=0) filled with zeroes.

Matrix elements are the number of materials with element A substituted by element B.

orderZ-normpair-thres001.element_matrix
---------------------------------------
pair correlation function, NxN matrix

g_{AB}= P(A,B)/(P(A)*P(B))

orderZ-normsumx-thres*.element_matrix
-------------------------------------
Data used to generate the periodic table plots for the different thresholds.
NxN matrix, matrix elements is replacement rate in %: given element <row>, how many
many % of the observed substitutions are with element <column>
-> each row sums up to 100