[housecleaning]

Which subforum would computational chemistry fall under?  I'm just starting to learn about it, and it sounds really interesting.

[Donaldson Tan]

Computational chemistry is a branch of theoretical chemistry whose major goals are to create efficient mathematical approximations and computer programs that calculate the properties of molecules (such as total energy, dipole moment, vibrational frequencies) and to apply these programs to concrete chemical objects.

If it involves alot of math, i think it's physical chemistry. LOL.

[housecleaning]

Yeah, thanks Geo.  That's what I was thinking too -- math = physical chem.  Wikipedia had some pretty good info, but it hasn't been easy gathering info for this field.

[gravenewworld]

Basically what comp. chem is all about is trying to solve the Schrodinger equation (along with some other stuff).  There are many different types of programs that will numerically do this for you. When  you hear a comp person talking about a "basis"  for a calc they are talking about what type of program they are using.  The bigger the basis, the better the approximation.  In order from worst to best, some of the well known bases are sto3g, 3-21g, 6-31g, 6-31g*.   There are bases that also do a density functional theory methods in order to complete a calc.  The better the bases though, the longer the calc will take.  Sto3g calcs usually have only taken me anywhere from 20 minutes to 2 hours, depending on how big the molecule is.  6-31g* and dft calcs however have taken me anywhere from several days to weeks in order to complete.  I have only used the program GAUSSIAN to do calculations, I don't know how well other progs work.  The first calculation that GAUSSIAN does is what is known as a geometry optimization.  Basically what the program does is try to arrange all the atoms in space so that the total energy for the molecule is minimized.  GAUSSIAN creates a potential energy surface for the molecule and then uses numerical methods to "walk along the surface" in order to find a energy minima (mathematically it is just calculating a derivative and finding the minimas of the surface).  Next, GAUSSIAN does a frequency calculation.  The most important information I get from this part of the calculation are the eigenvalues of the wavefunction.  Negative values indicate an imaginary frequency, which tells me that the optimized geometry obtained from the first part of the calculation is a transition state.  Molecules arranged in a ground state configuration don't have imaginary frequency values.  You can kind of see where this all leads. By taking a bunch of input structures, doing geometry optimizations, and finding all the transition and ground state strucutres, you can basically make a "movie" of how a reaction progresses.  Once you find all the peices you can make the movie.  

[Borek]

When  you hear a comp person talking about a "basis"  for a calc they are talking about what type of program they are using.

some of the well known bases are sto3g, 3-21g, 6-31g, 6-31g*

I have only used the program GAUSSIAN to do calculations, I don't know how well other progs work.

base = program

GAUSSIAN = program thus GAUSSIAN = base

base in [sto3g, 3-21g, 6-31g, 6-31g*]

GAUSSIAN not in [sto3g, 3-21g, 6-31g, 6-31g*] thus GAUSSIAN not a base

so is GAUSSIAN a base, or not?

I suppose - but my Computational Chemistry knowledge is so rusty that I am afraid to touch it, as it will fall down - that GAUSSIAN is a program, while base is a standard set of functions - like those used in Hartree-Fock method.

[gravenewworld]

Forgive my rough descriptions-I do comp. chem. work, I am not a computer programmer.  GAUSSIAN is the program that contains different bases. I probably shouldn't have called a base a "program".  Bases are just the different methods GAUSSIAN uses to do calculations.    

[Juan R.]

Which subforum would computational chemistry fall under?  I'm just starting to learn about it, and it sounds really interesting.

Some short answers:

1) The development of computational chemistry belongs to physical chemistry, because this is the fundamental branch of chemistry. That is the branch that develops the fundamentals where the others branches of chemistry are built.

2) However the application of computational chemistry to solve specific problems belongs to specific branch: organic, chemical engineering, etc. That is, a organic chemist can use computational chemistry to help to desing/optimize a syntesis.

3) it has been cited the GAUSSIAN. An example of 1) is a physical chemist as Pople (Nobel Prize for chemistry 1998 for computational chemistry) developing the mathematical algoritms and programing the GAUSSIAN. An example of 2) can be an organic chemistry who is in her/his laboratory doing a computation for understanding the most stable form of certain isomer that he is interested in.

4) the identities math = physical chemistry no-math = rest chemistry are incorrect. math = mathematical chemistry. Physical chemistry studies the physical basis of chemistry. The rest of chemistry applies that basis to SPECIFIC problems. There are organic or inorganic chemists who deal with a lot of math. For example an organic chemist studying the RMN spectra of a complex protein use advanced mathematical techniques.

5) Chemistry is not only the study of molecules. Therefore, computational chemistry also deal with atoms, single electrons, surfaces, liquids, solids, etc.

6) By 5) Computational chemistry is not only the solving the Schrödinger equation. The Schrödinguer equation alone is usually devoted to studies on gas phase. The Schrödinger equation is the most simple and very popular but does not work in many COMPLEX situations. For instance, In electron transfer studies in biomembranes the popular equation is the Lindblad one. In RMN in condensed phases chemists work with the Redfield equation, physicists working in laser or solid state use other equations, etc.

7) In condensed phases, moreover, computational chemistry uses other branches of theory. For example in chemical dynamics one uses Schrödinger for studying PES, molecular structures, frecuencies of vibration, etc. but, and the WIKI forgets this, after one uses statistical mechanics and theory of reaction in final computations. For example, one computes partition functions using averages for different molecular configurations and obtains the final kinetic constant from SM theoretical average procedures.

8 ) A basis is not a program not a method. A basis is simply a collection of mathematical functions. Initially, the first computational programs obtained the wavefunction by direct numerical solving of the Schrödinger equation and display results like an interminable list of data. One obtained a numerical representation of the wave function and then one would interpolate between data, adjust them for obtaining an analitic (e.g. sin(x)) expression, etc.

Currently the wavefunction is expanded using vector theory into a basis of the vectorial space (remember math courses) {f_1, f_2, ... f_n}

psi = a_1 f_1 + a_2 f_2 + ... a_n f_n

and the GAUSSIAN obtains the values a_1, a_2, ... a_n for the user-choosed BASIS.

In general better BASIS => accurate result and more computational time.

In the command GAUSSIAN #HF/6-311G(3df,3pd)

GAUSSIAN is the program, HF is the computational method, and 6-311G(3df,3pd) the basis used.

9) DFT (density funtional theory) does not directly work with the Schrödinger equation such as WFT (wave funtion theory) does. Instead, DFT obtains the electronic density. In short, DFT is computationally more tractable -and popular- because works with less information that WFT (as all of us know the wavefunction contains unobservable information).

10) Other programs (e.g. GAMESS) work in a similar way to GAUSSIAN.

11) It is not true that the "first calculation that GAUSSIAN does is what is known as a geometry optimization." In fact one can do a point-like computation WITHOUT optimization or apply a "scan" command. This is is done in computation of PES in chemical dynamics. Moreover in optimization the energy is minimized only for stable molecules, for transition states the Gaussian energy is not minimized: the TS is not a minimum of the PES.

[housecleaning]

Thanks Juan, that's a lot of info to digest.  

Question:  What do you think the best preparation for entering a PhD program in computational chemistry would be?  I already have a fairly broad chemistry & biochemistry background, so I'm wondering whether I should be taking more math or more computer science courses (I'm fairly weak in both right now)?

[Borek]

Math, math and some math.

Numerical methods are all math based, so the better your math skills will be the easier it will be later for you to catch on with numerical methods.

[pantone159]

I don't know that you would need much sophisticated computer science for computational chemistry (disclaimer:  Although I do know something about programming numerical methods, I haven't done anything that is specifically computational chemistry.)

However - It will be VERY important that you are fluent and practiced in the general process of systematically converting a process into a computer program.  (I.e. you need to be a competent computer programmer, although not a computer scientist.)  Even more important is that you are good at systematically figuring out what you did wrong (debugging).

Numerical methods can be very difficult to debug.  In many areas of programming, you can look, in detail in the debugger, at the operations the code is doing, and often tell that something doesn't make sense, and this points you at your mistake.  In numerical methods programming, when you change a + to a - somewhere, and then use the same strategy, all you see is a mass of arithmetic that is not obviously silly.  Then, after you have done a billion or so calculations, all you can tell is you got the wrong answer.  (Even that may be non-obvious.)

As others have said, math will be very important.  Especially linear algebra, particularly eignenvalue/eigenvector problems.  Understanding approximations well will be important, I'm not sure there is a specific class for that (other than numerical methods courses.)

[Donaldson Tan]

I don't know that you would need much sophisticated computer science for computational chemistry Numerical methods can be very difficult to debug.  In many areas of programming, you can look, in detail in the debugger, at the operations the code is doing, and often tell that something doesn't make sense, and this points you at your mistake.  In numerical methods programming, when you change a + to a - somewhere, and then use the same strategy, all you see is a mass of arithmetic that is not obviously silly.  Then, after you have done a billion or so calculations, all you can tell is you got the wrong answer.  (Even that may be non-obvious.)

actually, that is my favourite engineering coursework. i enjoy systematic approaches to convert mathematical methods into computer programs to solve engineering problems, or modelling a chemical plant. computer does the methods fast, but these methods must be encoded and created by humans still. AI is still far from being intelligent.

[Juan R.]

Thanks Juan, that's a lot of info to digest.  

Question:  What do you think the best preparation for entering a PhD program in computational chemistry would be?  I already have a fairly broad chemistry & biochemistry background, so I'm wondering whether I should be taking more math or more computer science courses (I'm fairly weak in both right now)?

There exist not a standard PhD program, therefore preparation may vary. My advice is that you would select a specific PhD program, look for any PhD or proffesor of the program and ask her/him for specific requirements needed .

However, i can offer some adittional comments on that. As already said in previous post, there are different subfields on computational chemistry and requirements vary:

1)
If your objetive is the computational study of chemical and physical properties of an organic compound, for instance, then you need a basic background in 'office' computers:  Excel, Windows or Linux, etc.

You will need a knowledge of the Gaussian (or similar package) from a user viewpoint. For example, if you want optimize a molecular structure you would know the Gaussian command for optimization. The math requirement is chemical graduate level.

2)
If your objetive is the development of a new computational method -think during and instant in the men what invented/developed the DFT method- then you need more physical and mathematical training. At what level? Well that depends. Take for example the recent CC method; into the limits of my current knowledge it is still not completely implemented in Gaussian.

Many computational tasks were done by physicists because early research was done via second quantization and Feynman-like diagrams. This involves a lot of math, quantum mechanics, and quantum field theory.

3)
Now imagine that your computational objetive is the optimization of a method. Then you would need a lot of math and knowledge of computer architecture. For example, i read a recent computational method for RMN, where authors implemented an optimization for solving the Redfield equation via a tensor product decomposition of matrix for rapid storage and computation. That task involved few or none chemistry and physics.

Since Redfield equation is defined in a Liouville space, it is more general that Schrodinger one (which is defined in a Hilbert space) and lot of mathematical thecniques developed for the Schrödinger equation could not be applied. Then those authors developed their own computational algoritms.

They were interested in chemical applications, of course, but since nobody (none mathematician, etc.) had been able to program the equation they needed due to technical difficulties, then they did themselves. Now they did a computational program based in their computational methods (similar to Gaussian but solving more general systems like molecules on solution) and use for their chemical, physical, and biological research

If your objetive is the optimization of an existent method (e.g. MPn) via parallel computers, then you needed a good amount of math, but specially a excellent training in scientific informatics and lot of knowledge about parallel systems (that is generally lot of knowledge on a specific architecture suplied by a specialized vendor: Hitachi, SGI, NEC, etc.). Look for example this basic ACS monograph

A computational quantum chemistry said to me once that in many supercomputation centers (e.g. here in Galicia CESGA ) the programers choosen are often (computational) quantum chemist before own enginners and programers because are (in his own words) true specialists on the optimization of Unix systems

4)
Etc.