- #1
andrewr
- 263
- 0
I would like to see how far I can get writing some novel & simplified simulator code for molecular modeling based on QM.
------ Background follows ---- skip to next post to get to the details of the project & starter QM question..
My son took chemistry this last semester (High-school level) and to help him out I wanted to find a free molecular modeler tool that would allow him to visualize VSEPR like images for molecules (PI/Sigma bonds), and also for myself; I was severely disappointed. As I have have some QM work that I may later publish (free) I don't want to enter into any kind of ip contract for such a program. eg: most molecular modelers are either 1. $1K> / seat or 2. proprietary, and want signed papers ... although the methods and implementations have been around for nearly as long as computers.
This became acute a few weeks ago when I succeeded in doing an electrolytic process that I previously though impossible because water would be decomposed; but it worked in spite of that. I desperately wanted to understand this chemical process although I am not a chemist but an EE. (Chem was my weak point in college). After a fairly diligent search, I discovered that the only program which was totally free in a sense usable to me was Ghemical, built on top of Jmol as a visualizer.
Sad to say the code is extremely buggy in places; The Java version never got beyond the splash screen, and the C/fortan code version could not completely compile -- especially in the area of orbital interaction based on old Fortran code, so that some features were disabled in my version.
I still had the crude "optimize geometry" function which I had hoped was close to VSEPR (Vander Waal's outlines...) and that would have been good enough -- but even that has serious bugs; Eg: when I built a test crystal that I knew what its bonds were the optimization turned it into a 2D item with atoms that would practically have to undergo nuclear fusion to be in the places they were optimized to...
I am not adverse to fixing code -- I have done quite a bit of it including kernel drivers, but I would rather spend my time on something other than old algorithms such as Hartree Fock, slater determinants, and all the well known problems of these variational methods which make my eyes glaze over.
Nobody yet has a flawless simulator...
So I would like to write my own code, both for my son and I -- but do not understand QM sufficiently to complete the project, and would like some feedback on problems I see and perhaps a heads up on any issues where I show myself to be blatantly ignorant of something. (As a BSEE I solve solid state QM problems regularly, but molecular modeling is a bit different ... )
I have figured out how to numerically simulate a field, such as E-M field, in ways that are reasonably accurate and efficient (sparse matrix). I wrote an electronics simulator using these techniques and am able to simulate circuit functions and nonlinear devices with extreme accuracy. I know this is sufficient as a foundation for actually building virtual "space" for watching EM in the classical Maxwell sense from discreetly located electrons and protons. So, I would like to experiment on the next possible step -- replacing Hartree Fock approximations, Slater determinants, etc, with the pilot wave interpretation of the Schrodinger equation and a statistical sampling of a (perhaps modified) Schrodinger equation.
As inspiration for trying this, I came across an old textbook (1950's) showing all of the QM orbitals S,P,D,F, modeled *correctly* using a mechanical spinning top -- and the accuracy was surprising to me -- so I do believe that a semi-classical simulation can produce fairly accurate results.
------ Background follows ---- skip to next post to get to the details of the project & starter QM question..
My son took chemistry this last semester (High-school level) and to help him out I wanted to find a free molecular modeler tool that would allow him to visualize VSEPR like images for molecules (PI/Sigma bonds), and also for myself; I was severely disappointed. As I have have some QM work that I may later publish (free) I don't want to enter into any kind of ip contract for such a program. eg: most molecular modelers are either 1. $1K> / seat or 2. proprietary, and want signed papers ... although the methods and implementations have been around for nearly as long as computers.
This became acute a few weeks ago when I succeeded in doing an electrolytic process that I previously though impossible because water would be decomposed; but it worked in spite of that. I desperately wanted to understand this chemical process although I am not a chemist but an EE. (Chem was my weak point in college). After a fairly diligent search, I discovered that the only program which was totally free in a sense usable to me was Ghemical, built on top of Jmol as a visualizer.
Sad to say the code is extremely buggy in places; The Java version never got beyond the splash screen, and the C/fortan code version could not completely compile -- especially in the area of orbital interaction based on old Fortran code, so that some features were disabled in my version.
I still had the crude "optimize geometry" function which I had hoped was close to VSEPR (Vander Waal's outlines...) and that would have been good enough -- but even that has serious bugs; Eg: when I built a test crystal that I knew what its bonds were the optimization turned it into a 2D item with atoms that would practically have to undergo nuclear fusion to be in the places they were optimized to...
I am not adverse to fixing code -- I have done quite a bit of it including kernel drivers, but I would rather spend my time on something other than old algorithms such as Hartree Fock, slater determinants, and all the well known problems of these variational methods which make my eyes glaze over.
Nobody yet has a flawless simulator...
So I would like to write my own code, both for my son and I -- but do not understand QM sufficiently to complete the project, and would like some feedback on problems I see and perhaps a heads up on any issues where I show myself to be blatantly ignorant of something. (As a BSEE I solve solid state QM problems regularly, but molecular modeling is a bit different ... )
I have figured out how to numerically simulate a field, such as E-M field, in ways that are reasonably accurate and efficient (sparse matrix). I wrote an electronics simulator using these techniques and am able to simulate circuit functions and nonlinear devices with extreme accuracy. I know this is sufficient as a foundation for actually building virtual "space" for watching EM in the classical Maxwell sense from discreetly located electrons and protons. So, I would like to experiment on the next possible step -- replacing Hartree Fock approximations, Slater determinants, etc, with the pilot wave interpretation of the Schrodinger equation and a statistical sampling of a (perhaps modified) Schrodinger equation.
As inspiration for trying this, I came across an old textbook (1950's) showing all of the QM orbitals S,P,D,F, modeled *correctly* using a mechanical spinning top -- and the accuracy was surprising to me -- so I do believe that a semi-classical simulation can produce fairly accurate results.