9 ms·
I can't read the research article now, but from the press article, it looks like they are explaining a known experimental result that was unexplained (or confus
by gus_massa 8y ago
I can't read the research article now, but from the press article, it looks like they are explaining a known experimental result that was unexplained (or confusing) until now.
IIUC in a previous real life experiment other group put a tiny amount of Potassium under very high pressure, and they uses x-ray diffraction to "see" the structure of the sample. At low temperature they saw regular arrangement of atoms that has slots (like a "sponge", but a regular sponge). In these slots there were more atoms in some fixed positions.
When they increase the temperature, the "sponge" part didn't change, but the atoms in the slots began to move like in a liquid. In the x-ray you get sharp point for the atoms that are essentially still and in a regular arrangement (the points are not the atoms), but the atoms that can move or are in irregular positions don't produce sharp point.
So IIUC in a previous real life experiment other group saw a change. My guess is that the atoms that can move don't drip. They keep trapped inside the "sponge" part (like a small amount of water in a towel). The dripping part is probably a bad metaphor.
---
In spite we know all the quantum mechanic rules to simulate this, the problem is that the runtime grows exponentially with the number of atoms. So an exact simulation is only possible with only a few small atoms (5-10 Carbons, perhaps 10-20 Hydrogens (but not 10 Carbons and 20 Hydrogens). Potassium has a lot of electrons, so it is more difficult.) (And the simulation is not as exact as a Mathematician would like to call it exact.)
There are some tricks to avoid the exponential time. Like assuming that the atoms/electrons only interact with nearby atoms/electrons. It's quite true, but not 100% true. You can also use some model of how the electrons in one atom interact with the electrons in the other atom. But electrons like to wander, specially inside a conductor, so you must consider the electrons that live between the atoms. It's a lot of fun, there are plenty of model to approximate how the electrons behave in some situation, but each model is good for one situation and bad for other situations.
If you want to go down the rabbit hole https://en.wikipedia.org/wiki/Density_functional_theory https://en.wikipedia.org/wiki/Density_functional_theory , but it is not the only method to try to solve the problem. There are a few rabbit holes, because the exact problem has exponential complexity.
In this work, they use a neural network to approximate the behavior of the electrons in each atom (and between atoms). I guess that they are assuming that they interact almost locally, and hopefully they can get the result in linear time instead of exponential. I'm not sure if this can be classified as DFT or other technique, or they are digging a brand new rabbit hole.
It's difficult to judge the work by reading only a press article, but the abstract of the research paper says:
> Calculations necessitated the development of an interatomic forcefield using machine learning, which we show fully reproduces potassium’s phase diagram, including the chain-melted state and 14 known phase transitions.
It looks very promising, specially because in each phase the conditions are different and that makes the use of a single model very difficult.
- semi-extrinsic 8y agoI can't even find the paper in question - DOI link in the article here doesn't even resolve. However, from the description of this model as "an interatomic forcefield", I'd guess this is more like a potential that can be used in classical molecular simulations. I.e. it's a highly refined and application-specific version of an EAM type model. This is a good jumping-off point for those: https://www.ctcms.nist.gov/potentials/ https://www.ctcms.nist.gov/potentials/
- DannyBee 8y agoI wrote a computer program which proves this article is wrong. It avoids exponential runtime by assuming that because newspapers mostly make a mishmash of science articles, that you only need to look at some of the words before deciding whether they made a mishmash of this article. It's not 100% true, but it's quite true. ...
- gus_massa 8y agoGreat. When you get it published in a peer review journal, please post it here so we can discuss it.
- markwkw 8y agoIt seems that runtime should grow quadratically with the number of atoms. There are n^2 pairs of atoms interacting with each other, so n^2 interactions to account for in each step. Why is it that the simulation takes exponential time? Is there an approachable explanation (honestly curious)
- gus_massa 8y agoIt's part of the weird effect of quantum mechanics. To simplify, let's suppose you have two electrons that can jump between 4 Hydrogen atoms. Let's call the ("1s" orbitals) in the atoms 1, 2, 3, 4. Now the 2 electrons can be in the atoms 1 and 2. The official notation is |12> and it's essentially equal to |21> because the wo electrons are exactly equal. The two electrons can be in |13>, |14>, |23>, |24> or |34>, so there are in total 6 possibilities. For a calculation of the state with minimal energy, you must find a combination of them, i.e. |state> = alpha |12> + beta |13> + gamma |14> + delta |23> + epsilon |24> + dseta |34> where the Greek letters are real number, and the square of them sum 1. With 3 electrons in 6 Hydrogen atoms you have 20 possibilities, so you must calculate 20 real numbers. With 4 electrons in 8 Hydrogen atoms you have 70 possibilities, so you must calculate 70 real numbers. (It's more usual to consider that the number of electrons is equal to the number of Hydrogens, but the electrons in the Hydrogens can have spin up or down. So you have to double the numbers in the previous examples.) The general calculation for N Hydrogens is Combinatorial(N, N/2) (or Combinatorial(2N, N)), and it is exponential. With more complex atoms, you have more orbitals and more electrons, so it's even bigger. --- As a side effect, if you can modify the global state of one of this systems with N atoms, you are operating on an exponential number of states. This is the main idea of quantum computers, and why they can go faster than a classical computer. They use better system than my examples, so the calculation are simpler, but it is the same "exponential" magic/curse. There are a lot of problems to build a "good" system, so the current quantum computer have a small amount of atoms.