The Notebook.

A bunch of possibly interesting things.

Predicting Melting Points.

One feature I have wanted to add to ChemSi is the ability to have state symbols - that is, instead of outputting

C3H8 + 5O2 -> 3CO2 + 4H2O

It would instead output

C3H8(g) + 5O2(g) -> 3CO2(g) + 4H2O(l)

In order to do this I needed some way to predict the melting and (eventually) boiling points of different compounds, given only a little information about them.

My initial solution was to simply use enthalpy and entropy changes with the Gibb's equation. If I could get the entropy change from solid to liquid, and the enthalpy change, then I could use




This solutions works - to a point. While for chemicals I have the entropy and enthalpy data for it works very nicely, I do not have the data for all states of all compounds - and it is unreasonable to get this. Additionally, this felt like a bit of a cop out. The point of ChemSi is to try and calculate (or at least, approximate) these values from easily accessible theoretical data, such as is done with the reaction prediction. If I wanted it to predict them excellently I could just provide it with some rules (ie, compound containing C, H, and O reacts with O2 to produce blah) or even worse, just give it a list of the reactions.

After some Googling, I came across the Lindemann criterion. Essentially it boils down to

T_{m} = \frac{2\pi mc^{2}a^{2}\theta^{2}_{D}k_{B}}{h^{2}}

Where the 'dodgy constants' are \theta_{D} (Debye temperature) and c. The rest are either pretty standard constants (the Boltzmann constant, for example) or pretty easy to get. The reason that the dodgy constants were a problem is that they require information on the geometry of the crystalline material - which was quite difficult to get access too.

Just a note on that crystalline material part - almost all of these prediction methods (bar the enthalpy and entropy change one) are limited to Ionic crystals, and are entirely useless for other molecules.

Following this, I searched around some more. I had the idea that perhaps the Lattice Enthalpy would have some relationship with the melting temperature - after all, surely a lattice with a more negative lattice enthalpy would be more strongly bonded so more energy would be needed to break it apart? I decided to test this hypothesis.

For a sample of ionic lattices, Na-Cs and F-I (Only using the group 1 metals and halogens may be a bit of a limitation, which led to me having to do some fiddling later on to fit the other groups. At this stage, however, I was looking for a relationship. Lithium is not included as bonds with Lithium tend to be more covalent in character.) I was able to produce the following graph. The data was taken from this website - I know it refers to Lattice Energy but from the explanation with it I believe they are referring to the Lattice Dissociation Enthalpy so it is still usable in this context.



As you can see, this produces a reasonable correlation - the pairwise correlation coefficient calculated by Octave is 0.9300.

I decided to go ahead and use this lattice enthalpy method to predict the melting points. Unfortunately, as you may have expected, I ran into the same issue as the Gibbs method - I can't really store enough lattice enthalpies! Luckily, there is an equation called the Born-Lande equation

 E = -\frac{N_{A}Mz^{+}z^{-}e^{2}}{4\pi \varepsilon_{0}r_{0}}(1-\frac{1}{n})

This does not produce exact Lattice enthalpies - it is more of a prediction. This method is used in predict_mp_alg2().

I also decided to try and see if I could modify the equation used to calculate the shell energy that is already used for reaction prediction to approximate lattice enthalpy. I won't go into detail about it here, but it is given as predict_mp_alg1() in ChemSi if you want to check it out.

Just to show these equations working, I decided to predict the melting points of three compounds using both the Lattice Enthalpies (Born-Lande) and my shell energies. The compounds I chose were NaCl, FeCl2, and CsI.

Compound Actual MP (K) Shell Energy MP (K) Born-Lande MP (K)
NaCl 1074 1088 1122
FeCl2 950 993 895
CsI 894 1084 819

As you can see, both give a reasonable (+- 50) approximation for NaCl with the shell energy giving the best prediction; this is mirrored with FeCl2 where shell energy is closest (but both are spread further than for NaCl), and finally in CsI the Born-Lande approximation is closest, with the shell energy prediction being too far out. This is likely to do with the shell energy calculation getting less accurate at higher atomic numbers.

Both algorithms have been left in ChemSi so the differences can be seen.