Just a quick chemistry question

[Mosquito]

before I run home to stuff myself on something fattening.


Is it possible to calculate the properties of molecules/compound materials given the composition and the properties of the parts?

I.e. Can the properties of water be deduced from the properties of Oxygen and Hydrogen?

If yes, does this hold true for all materials? (Maybe given the layout of the resulting molecules as well.)

If the properties of the molecules can be calculated by knowing which atoms and how they are arranged, can the properties of large groups of said molecules be calculated as well? (Actual water, as opposed to just one molecule.)

Can the effects of a known molecule on a different known molecule be calculated?


I guess what I am asking is if for chemistry, the total is equal to the sum of its parts, and if it is, are the parts known to a degree that predictions can be accurately made?

I seem to remember having read that color, melting point/boiling point etc. of trans-uranic metals are known, but I don't know if anybody have ever seen and tested these materials in anything but "the occational atom in a atom-collider"-levels.


Mosquito - curious and ignorant
(Ok, so it is technically more than one question, but I'm hungry, so deal with it.)

 

[Unnamed]

Is it possible to calculate the properties of molecules/compound materials given the composition and the properties of the parts?

I.e. Can the properties of water be deduced from the properties of Oxygen and Hydrogen?

No, at least not in the way you are thinking. Once atoms are combined, the properties of the compound are that of the molecule, not of the atoms anymore. Another example is steel, which is very different from iron, even though only a small percentage of carbon is added.

To be more precise, even the properties of different arrangements of the same atom can be wildly different. Diamond is extremely stiff and hard, while graphite is very weak and soft. But both are forms of carbon with different arrangements. And on top of this, carbon nanotubes have the same chemical bonds as graphite, but they are even stiffer and more resistent than diamond.

If yes, does this hold true for all materials? (Maybe given the layout of the resulting molecules as well.)

With that out of the way, there has been steady progress in "ab-initio" methods to determine the properties of chemical compounds given only their composition and atomic arrangement. Look up DFT, for example. In summary, we can solve an approximate version of Schrödinger's equation, which gives us very detailed information about the interactions between the atoms, and gives access to many properties.

If the properties of the molecules can be calculated by knowing which atoms and how they are arranged, can the properties of large groups of said molecules be calculated as well? (Actual water, as opposed to just one molecule.)

Yes and no. Properties of large groups of molecules require more computer power than generally available. But there are faster methods such as classical molecular dynamics, where each atom is considered a particle and the interactions are approximated using empirical (i.e. determined by trial and error) functions. These give good results, depending on your definition of "good".

Interestingly, there is no model (that I know of, but I am fairly new in the field) that gives accurate predictions for all important properties of water. Methods (empirical functions, simplifications, assumptions, etc.) that are tuned to give thermal conductivity, for example, may be lousy for compressibility (the example may not be accurate, but that's the idea). Even DFT-based molecular dynamics is not good enough, because the hydrogen bond is not well simulated by it. Even more precise (and computationally expensive) methods are needed, but then you only get a few molecules.

Can the effects of a known molecule on a different known molecule be calculated?

They can, but within reason. Protein-protein interaction is not "within reason", specially in the presence of water. But simple chemical reactions with just hundreds of atoms can be simulated.

I guess what I am asking is if for chemistry, the total is equal to the sum of its parts, and if it is, are the parts known to a degree that predictions can be accurately made?

It isn't, but I hope I gave you some hope. The degree of accuracy of these methods is quite good. Typical errors are from 10 to 30%, or better if you are prepared to spend more computer time.

And I'm sorry that I can't help you with the transuranic elements, but I will ask around.

 

[Unnamed]

Just one more thing before I go back to work:

Of course you've heard of the periodic table That's a nice representation of the fact that we can estimate the properties of an atom based on similar atoms. Oxygen, fluorine and chlorine are all very reactive oxidant gases, and they are all located in the top right corner. Germanium and Silicon (stacked one above the other) are the most important semiconductors in electronics. And the similarity is still there when the atoms combine. For example, strontium (the element right below calcium) can easily replace calcium in bones, with some therapeutic uses.

 

[Dr Adequate]

Is it possible to calculate the properties of molecules/compound materials given the composition and the properties of the parts?

I guess what I am asking is if for chemistry, the total is equal to the sum of its parts, and if it is, are the parts known to a degree that predictions can be accurately made?

No. You also need to know about the way the atoms are fitted together.

Think of Lego. You could know which pieces were used and how many of each, but if you don't know how they fit together, you don't know if it's a castle or a spaceship.

 

[Soapy Sam]

Yes and no. The geometry of an ionic crystal lattice for example can be deduced from the sizes and valence of the ions forming it.

Some of it's physical properties will also be calculable- dislocation energy and the like.

As the compound grows more complex- (I use "compound " here in a general sense for any atomic aggregation, however bonded) combinatorial effects will come into play. The number of possible emergent properties will rapidly become incalculably large

However- observational chemistry shows that there are patterns of molecular characteristics just as there are patterns of atomic characteristics, so some probable properties may be estimated.
(Particularly in organic chemistry, where complex molecules tend to have distinct sub groups -a methyl group for instance- , which have predictable behaviour in many situations.) This is one way new drugs are designed.

 

[Mosquito]

Ok, so what I get from Unnamed is that I can spend a huge amount of time calculating some heavy duty equations and get nice, precise, consistent results that just happen to be itsy bitsy wrong. After having checked out the wikipedia aticle on DFT, I also realize that though it could be fun to put this stuff in a AL-program, I'll have to spend a fairly large amount of time to learn to read the math. Depressing. Not that I don't like math, but that I don't know how to read it. And I don't have time to spend the next 5 years learning about it. Sometimes having a potential lifespan of about 100 years sucks.

And I have heard of the periodic table

Just for the record, when I say "the total is equal to the sum of the parts", I consider organization to be one of the parts. An important part, in fact. There are many ways of organizing the atoms of, say, a human being, not all of which will result in anything resembling a human being.

The reason I asked, aside from the general curiosity, is that I envisioned the possibility of writing an AL (genetic algorithms) program for coming up with new materials. As I'm writing this, I just realized that such a program may also be used to "evolve" abiogenisis(sp?), given the right goal. We might have found alternative ways of having the process of life. That would have been a nasty thing to do to those IDiots

I was hoping that the four fundamental forces and some not-too-complex equations would be enough to get the properties requested from any molecules/materials.

Seems I was wrong.

I probably should have expected this, as something as "simple" as protein folding apparently is rather complicated.


Mosquito - Yet again I've learned that the real world does not want to play nicely with me

PS: Any hope of having a solid theoretical basis allowing for such simulations anywhen soon?

 

[Mosquito]

Yes and no. The geometry of an ionic crystal lattice for example can be deduced from the sizes and valence of the ions forming it.

Some of it's physical properties will also be calculable- dislocation energy and the like.

As the compound grows more complex- (I use "compound " here in a general sense for any atomic aggregation, however bonded) combinatorial effects will come into play. The number of possible emergent properties will rapidly become incalculably large

However- observational chemistry shows that there are patterns of molecular characteristics just as there are patterns of atomic characteristics, so some probable properties may be estimated.
(Particularly in organic chemistry, where complex molecules tend to have distinct sub groups -a methyl group for instance- , which have predictable behaviour in many situations.) This is one way new drugs are designed.

But what we end up with are approximations, not just "correct to 4 decimals" but "wrong somewhere, we don't know where". If I've gotten the general gist of this, that is.

I wanted to use GA/AL to evolve new materials, this could work if similar molecules behave similarily. If small changes can cause huge differences, this becomes moot. But if we can't calculate what a material does/is using the chemical formula and the organization of the atoms, this also becomes moot, as we have no way of implementing a fitness function.

Hopefully somebody can come up with a nice, simple, fast-to-calculate equation for this purpose one day soon.


Mosquito - Limited by reality

 

[JamesM]

But what we end up with are approximations, not just "correct to 4 decimals" but "wrong somewhere, we don't know where". If I've gotten the general gist of this, that is.

Well, don't worry about that. As the chemist Henry A Bent said:

A model must be wrong, in some respects, else it would be the thing itself. The trick is to see where it is right

I wanted to use GA/AL to evolve new materials, this could work if similar molecules behave similarily.

The similarity principle is the basis of all chemistry, if we couldn't discern similarities between different molecules and reactions, we couldn't leverage past experiments to produce new molecules. So in some sense, similar molecules do behave similarly.

But the devil is in the details. One of the largest efforts in chemoinformatics is in defining what it means for one molecule to be similar to another. There have been a huge number of papers and books written on this subject.

As for using GA to breed new materials, using genetic algorithms to build new molecules is part of a well-established principle of computer-aided molecular design, called de novo design. The problem is not so much in getting a computer to generate new molecules, but to generate a new molecule that a bench chemist can actually make. Synthetic feasibility is a very difficult concept to quantify.

But if we can't calculate what a material does/is using the chemical formula and the organization of the atoms, this also becomes moot, as we have no way of implementing a fitness function.

No need for pessimism, what you're talking about is regularly carried out across several fields, called Quantitative Structure-Property Relationships (QSPRs)...

Hopefully somebody can come up with a nice, simple, fast-to-calculate equation for this purpose one day soon.

... but there is no one master equation that predicts all properties for all molecules.

Instead, you use machine learning and statistical techniques to derive equations relating a property of interest to a description of a molecule, which are normally referred to as descriptors in the jargon of chemoinformatics. Each QSPR is derived from an existing set of measurements ranging from 10s to 1000s of molecules.

In medicinal chemistry, where most of the action takes place, good quality datasets are rarely larger than 100 molecules in size. When it comes to predicting the properties of new molecules, assessing when your equation is valid, and when its extrapolating too far remains an open area of research.

Descriptors are not limited to chemical formula, though, several hundred different types are available. A recent book on the subject runs to nearly 700 pages.

 

[JamesM]

PS: Any hope of having a solid theoretical basis allowing for such simulations anywhen soon?

In short, no.

Quantum chemical methods are beginning to be applied to molecules of biological interest, but the calculations are extremely time-consuming. They do not scale nicely at all.

Even using empirical force field approaches (where we treat molecules as balls on springs and use Newton's laws to model their behaviour) with no quantum mechanics at all to study protein folding is a major, major computational undertaking. You can't really go faster than 2 femtoseconds in terms of simulation time steps, and the time scale of protein folding is normally in the region of milliseconds (ish).

If you want to explicitly model the solvent (normally water) surrounding a protein, things get even slower.

It all depends on what properties you want to model. If you're interested in the ionisation energies of fairly small molecules (not much bigger than benzene, please), then that's feasible.

Anything a lot more fancy than that is pretty much beyond our current capabilities, and probably will remain so, pending a theoretical break through.

 

[Mosquito]

In short, no.

Quantum chemical methods are beginning to be applied to molecules of biological interest, but the calculations are extremely time-consuming. They do not scale nicely at all.

Even using empirical force field approaches (where we treat molecules as balls on springs and use Newton's laws to model their behaviour) with no quantum mechanics at all to study protein folding is a major, major computational undertaking. You can't really go faster than 2 femtoseconds in terms of simulation time steps, and the time scale of protein folding is normally in the region of milliseconds (ish).

If you want to explicitly model the solvent (normally water) surrounding a protein, things get even slower.

It all depends on what properties you want to model. If you're interested in the ionisation energies of fairly small molecules (not much bigger than benzene, please), then that's feasible.

Anything a lot more fancy than that is pretty much beyond our current capabilities, and probably will remain so, pending a theoretical break through.

So, it is not only (currently) theoretically impossible, it is also much slower than I anticipated, not to mention much more complex.

And all this to the point where it is probably faster&cheaper to have some chemists drinking good wine and coming up with new ideas, then going to the lab and make more wine.

I can't afford wine

Well, I'll just have to come up with something else I can find a quick and dirty solution to

Thanks to all for input though, while I don't like the realities of this, I suppose I have to accept them. That seems to be the curse of reality.


Mosquito - lacking theoretical background to understand my own questions

 

[Unnamed]

So, it is not only (currently) theoretically impossible, it is also much slower than I anticipated, not to mention much more complex.

Well, I'll just have to come up with something else I can find a quick and dirty solution to

Thanks to all for input though, while I don't like the realities of this, I suppose I have to accept them. That seems to be the curse of reality./

I can't help you more than JamesM splendidly did above, but here are a few comments:

You had a great idea, and 5-year research grants are regularly given to people that want to solve these problems. People get PhDs for that (JamesM sounds like one of these ).

Reality is too hard, but who cares about reality ? If your goal is to see abiogenesis, define your own simplified "chemistry" and use it. Conway's Game of Life is an oversimplified example, but any AL book will give you more ideas. Some are amazing and still very simple.

One of my pet curiosities is to know how simple a "reality" can be and still support some form of life. I see life as an emergent property, independent of the underlying physics and chemistry.

Finally, ID people will not be convinced by anything less than a cell created in a test tube. And they will accuse the scientists of cheating.

 

[Mosquito]

I can't help you more than JamesM splendidly did above, but here are a few comments:

You had a great idea, and 5-year research grants are regularly given to people that want to solve these problems. People get PhDs for that (JamesM sounds like one of these ).

Reality is too hard, but who cares about reality ? If your goal is to see abiogenesis, define your own simplified "chemistry" and use it. Conway's Game of Life is an oversimplified example, but any AL book will give you more ideas. Some are amazing and still very simple.

One of my pet curiosities is to know how simple a "reality" can be and still support some form of life. I see life as an emergent property, independent of the underlying physics and chemistry.

The idea of abiogenesis is more like a spinoff, though I certainly can see the sexyness of using it to attract funding.

Problems with that is that I can't afford it (unless such funding leaves a lot of money to me), and with my credentials, nobody within their right mind would allow me to mess up such a research project

Maybe I can find a small team and do it as a hobby project some day, but right now finding money for food is of higher priority.

To me life is dependant of the physics and chemistry, but I don't think that only one particular set of rules and conditions would favour life. As such, I think life is fairly abundant. Being able to simulate abiogenesis in highly simplified versions could be very interesting. Maybe even very different chemistry/physics would eventually result in some sort of life. Such a project may be able to uncover this.

Finally, ID people will not be convinced by anything less than a cell created in a test tube. And they will accuse the scientists of cheating.

I don't think they'll make such accusations. They'll just say that "Yes, the cell got created in the test tube, this is the result of the scientist performing intelligent design. And thus it has been proven that life needs to be designed. Now pray to Jesus."


Mosquito

 

[JamesM]

Reality is too hard, but who cares about reality ? If your goal is to see abiogenesis, define your own simplified "chemistry" and use it. Conway's Game of Life is an oversimplified example, but any AL book will give you more ideas. Some are amazing and still very simple.

On a related note, over the past ten years, Lemont Kier and co-workers have produced a whole series of papers on applying cellular automata to modelling chemistry. They started with a simple model of water, and have recently looked at second order reaction kinetics and enzymatic pathways.

I remain unsure of its practical merits and its been pretty much ignored by the wider community - I only discovered them when I went looking specifically for CA applications in chemistry, inspired by a topic in this very forum. But if you have even a passing interest in chemistry and/or CA, they're fascinating.