It's great that the popular press still runs science articles.
What's not so great:
1. There is no link to the original paper.
2. The description of the phenomenon has been so contorted that it's impossible to understand exactly what was found or its significance.
The article notes:
> “Somehow, these potassium atoms decide to divide up into two loosely linked sub-lattices,” Hermann says. But as scientists turned up the heat, x-ray images showed the four chains disappearing, and researchers argued about what exactly was happening.
Fine, there's debate about how to explain the experiment.
> The computer models confirmed that between about 20,000 and 40,000 times atmospheric pressure and 400 to 800 Kelvin (260 to 980 degrees Fahrenheit), the potassium entered what’s called a chain-melted state, in which the chains dissolved into liquid while the remaining potassium crystals stayed solid.
That's just one explanation consistent with the data. Clearly, there are others.
The article gives the false impression that some smoking gun has been found. Far from it.
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 , 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.
I 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:
I 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.
It 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)
It'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.
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.
We know the basic rules of physics, how all the little bits interact. We also know that if you combine enough of the little bits, you'll get a system with no analytic solution. Where, provably, all you can do to learn things about is simulate it.
So it's possible to discover new and unexpected behavior of a system with modeling that hasn't yet been experimentally verified. You don't know if such a phenomenon is present in the real world until you do experimental validation, but that doesn't mean your model is wrong.
Maybe a more accurate statement would be 'laws of nature found to give rise to new phase of matter'.
Sorry but I think the verbiage needs cleaning up. A model can predict a new phase of matter, but not discover it. E.g. dark matter has been predicted, but not discovered.
All models are by definition abstractions and approximations (except perhaps the lonely H atom) so there is absolutely room to be wrong about predictions.
I think this means "discovered" in the sense that a mathematical proof is discovered, rather than the sense that an ancient buried artefact is discovered.
Sure, but chemical and thermodynamc models fall under far less rigorous logic than a mathematical proof. To again ape physics, would it not have been wrong to say gravitational waves were discovered when they first appeared in equation? I'm on mobile so I haven't read the actual paper yet but this isn't the sort of thing that can constitute 'discovery'.
What I think thereisnospork is trying to say is that you can't give the same weight to "discoveries" obtained through simulation using current models as "discoveries" obtained through observations via empirical research.
The models are estimated through the observations we make through experimentation on the real world. They're not necessarily right.
I agree, one should distinguish the experimental confirmation from theoretical predictions.
Simultaneously, the "it's just a theory"-flavor rhetoric is tiresome and small-minded.
The basic theories (GR, QM) have been experimentally verified with exquisite accuracy. Theories produced with those axioms as a foundation inherit the incredible effort that has been put into verifying the underlying rules.
Experimental results showing that the predictions of such a theory are wrong would be a HUGE deal. That does not indicate a problem with the theory, as much of a fundamental gap in our knowledge of how the universe works.
But sure, whatever, it's just a theory
Edit: clarification, this only applies to theories that only take as axioms the _fundamentals_. No approximations.
I think some of the sibling comments do a decent job of address the how, so I'll loop back to why.
Material science research can be expensive due to: material costs, specialized equipment, specialized knowledge, dealing with hazardous waste, facility requirements//specialization, competition over the said resources, etc.
A computational approach greatly reduces the barrier to entry and expands your reach at the same time.
Let's talk phase diagrams as an example. We have some pretty good models for calculating the phase diagrams of bulk materials, but they breakdown as we attempt to describe smaller (I.e. nano) particles. While some experimentation was/is needed to get enough raw data to model additional terms/changes that need to be applied to make the general CALPHAD approach validate at the nanoscale, we can now cheaply test theories about how size/shape/composition of nanoparticles would influence their phase diagrams.
This helps cheaply narrow and guide the search space, let's say for nanosolders, that researchers would want to follow up with physical experimentation.
This makes a loop, the resultant data is used to make better models which in turn helps point out potentially interesting systems to study.
For example, you can use Molecular Dynamics simulations of atoms and their interactions over time in order to compute properties of the material through the statistical mechanics magic that is the Boltzmann constant, which bridges the microscopic with the macroscopic.
That's how in University I got to play with melting high level nuclear waste in molten potassium and lithium salts without having to actually physically handle any of that yucky stuff.
>How can a computer model prove the existence of something physical?
By having verified models of underlying physics.
Many or most physical models these days rely on some empirical knowledge so experimental verification is necessary. Year after year science is getting better at simulation.
If you were interested in one day creating Star Trek's Replicator, this might be a good place to start. That is, here might be a good place to study the junction between solid and liquid phases of matter (your future Replicator would probably create liquids out of thin air before you figured out how to create solids, and this might be the phenomena you wanted to study to then figure out how to create those solids...)
Also, the list of exotic states of matter in the article is worth reading and re-reading...
Clickbait title. The discovery is that at high pressures, potassium adopts a complex crystalline structure, of which different parts have different melting temperatures.
Ok, we'll use that. If anyone can suggest a better title (i.e. more accurate and neutral, preferably using representative language from the article itself) we can change it again.
The melting part is important. It's not just a complex crystal structure. If's a crystal structure with slots that are filled with the same material. The interesting part is that the filling and the main crystal structure have a different fusion temperature.
"At high pressures, potassium adopts a mix of crystalline and liquid structure"???
Isn't that what science is based upon though? We can only make assumptions based upon observed results. You can extrapolate and then test based upon those extrapolations, and we can acknowledge that there are gaps, but I don't think someone saying "wow, that didn't meet my expectations at all" in any way discredits science.
While the material discussed in the article and new, the phase of matter is not new. The term is called non-newtonian fluid. Believe it or not you can make material with these properties in your own home.
An inexpensive, non-toxic example of a non-Newtonian fluid is a suspension of starch (e.g. cornstarch) in water, sometimes called "oobleck", "ooze", or "magic mud" (1 part of water to 1.5–2 parts of corn starch). The name "oobleck" is derived from the Dr. Seuss book Bartholomew and the Oobleck.
Because of its properties, oobleck is often used in demonstrations that exhibit its unusual behavior. A person may walk on a large tub of oobleck without sinking due to its shear thickening properties, as long as the individual moves quickly enough to provide enough force with each step to cause the thickening. Also, if oobleck is placed on a large subwoofer driven at a sufficiently high volume, it will thicken and form standing waves in response to low frequency sound waves from the speaker. If a person were to punch or hit oobleck, it would thicken and act like a solid. After the blow, the oobleck will go back to its thin liquid like state.
What they describe doesn't sound like a non-newtonian fluid, although it's possible that the atomic-scale behavior they described manifests non-newtonian behavior on a macroscopic scale.
From the article, it seems that what’s described is slightly different. While oobleck requires certain stimulation to achieve its states (standing still on Oobleck, you will sink; walking across oobleck, you’ll be fine) the potassium contraption described chemically appears to be a solid but is also actually melting.
Forgive me if I didn’t do a good job describing, but I felt the difference was worth pointing out. Perhaps it would be fair too to say that the potassium did require a little prodding at 20,000 an increase in pressure to that of earth.
Not ballistic armor but D3O is one material used for impact protection that is flexible but becomes rigid when there's an impact. (I have a ski hat with ribs of this material.)
What's not so great:
1. There is no link to the original paper.
2. The description of the phenomenon has been so contorted that it's impossible to understand exactly what was found or its significance.
The article notes:
> “Somehow, these potassium atoms decide to divide up into two loosely linked sub-lattices,” Hermann says. But as scientists turned up the heat, x-ray images showed the four chains disappearing, and researchers argued about what exactly was happening.
Fine, there's debate about how to explain the experiment.
> The computer models confirmed that between about 20,000 and 40,000 times atmospheric pressure and 400 to 800 Kelvin (260 to 980 degrees Fahrenheit), the potassium entered what’s called a chain-melted state, in which the chains dissolved into liquid while the remaining potassium crystals stayed solid.
That's just one explanation consistent with the data. Clearly, there are others.
The article gives the false impression that some smoking gun has been found. Far from it.