Explore Soft Matter with Coarse-Grained Molecular Simulations

← Back to AIMS

By Charles Xie ✉

The pixel of a phone screen, a plastic cup, an egg white, and a bar of soap are all made of molecules too large and too many to simulate one atom at a time in a web browser. A single polystyrene chain has thousands of atoms, and a micelle of soap holds dozens of molecules with hundreds of atoms each. AIMS simulates a few hundred atoms while a student watches. To reach these materials AIMS trades atoms for beads: each bead stands for a group of atoms, a styrene unit, an amino acid, or four carbons of an oily tail, and the beads interact through forces fitted to the real molecules. This is coarse-grained modeling, a standard way to simulate soft matter in research, and AIMS brings it to the classroom with an AI agent that sets up the experiments. A student asks a plain-language question, and the agent heats a liquid crystal until its order collapses, cools a polymer melt until it turns to glass, unfolds a protein-like chain, or breaks up a micelle, and posts the measured data in the chat as a live chart. In the window below, 400 rod-shaped molecules modeled on the liquid crystal in most displays stand on a lattice at room temperature. Press Play and watch them jostle: they keep pointing roughly the same way. That shared direction is what makes a liquid crystal, and it survives at room temperature because of the shape of the molecules and the attraction between them.

Why Trade Atoms for Beads?

An atomistic simulation computes the force on every atom from every nearby atom, a femtosecond or two at a time. For a few hundred atoms a browser can do that fast enough to watch, and the phase-change and heat-conduction articles on this site work that way. The materials in this article are different in scale. What makes a liquid crystal is the alignment of thousands of rod-shaped molecules; what makes a glass is chains too tangled to slide past one another; what makes a micelle is dozens of soap molecules agreeing on which way to point. Each of these involves hundreds of molecules and takes nanoseconds. A model that keeps every hydrogen atom spends its whole budget on detail the phenomenon does not depend on.

A coarse-grained model keeps the degrees of freedom that matter and drops the rest. Two landmarks of the approach are the 1990 simulation by Kremer and Grest, who modeled polymer melts as chains of soft beads on springs and reproduced how the chains move and tangle, and the Martini force field of Marrink and coworkers, which maps about four heavy atoms onto one bead and is used today to simulate cell membranes and proteins. A bead has a mass, a size, and an attraction to other beads, and chains of beads have springs, bending stiffness, and a torsion term. Every built-in bead type in AIMS is in real units and fitted to a reference molecule: one bead of the polymer melt has the mass of a styrene unit and an attraction that makes the melt stop flowing near the glass transition of polystyrene; one rod has the length, width, and mass of 5CB, the molecule in most liquid crystal displays; the soap tails have the size of Martini's beads. Because the units are real, the thermometer in every lab reads kelvin, and a student can compare the model's transition temperature with the handbook value of the real material.

Beads are drawn in their own colors, and a coarse-grained card in the gallery shows a table of its bead types with mass, size, and well depth, each editable. The Build Chain dialog takes any bead sequence: "B" repeated twenty times is a polymer chain, "HPHHPP" is a protein-like chain of hydrophobic and polar residues, and "HD TL TL TL" is a soap molecule with one head and three tail beads. The agent builds the same models from a sentence. Chemistry below the size of a bead, a hydrogen bond or a reaction, is outside a coarse-grained model, and the soap model leaves out the water. The model shows, in a browser, phenomena that need thousands of atoms and nanoseconds of time.

Running Experiments by Asking

AIMS is an AI-native environment. Building a box of rods, stepping its temperature, measuring an order parameter, and plotting it are the agent's work. A student asks — "at what temperature does the liquid crystal melt?", "when does the melt become a glass?", "does this sequence fold?" — and the agent recognizes the intent, configures the lab, runs the experiment, and posts a live chart of the results in the conversation.

Liquid Crystals: Order in Direction Only

A liquid crystal flows like a liquid, and its molecules all point roughly the same way, as the molecules of a crystal do. The phase with that one kind of order is called nematic. In 1949 Onsager showed that shape alone can produce it: long rods packed densely enough line up because parallel rods leave one another more room to move than crossed rods do. Attraction between the rods adds to the effect; in the model a pair of rods attracts five times more strongly side by side than end to end. The model in this lab is the Gay–Berne potential of 1981, a single ellipsoid per molecule whose attraction depends on how the two ellipsoids are oriented. Each rod here is 18 Å long and 5 Å wide and weighs what a 5CB molecule weighs, and it rotates under the torque its neighbors exert. The cell has periodic boundaries: it has no walls and repeats in every direction, so the 400 rods stand for a small piece of bulk liquid crystal. They sit at a little over half the density of 5CB itself. At the full density, rods this long stay aligned at every temperature tried, up to 850 K, because they order by packing alone; the lower density is what lets temperature decide.

The measure of alignment is the nematic order parameter S, which is 1 when every rod is parallel and 0 when they point in every direction; 5CB at room temperature measures about 0.6. Ask for the heating and cooling ramp above and the agent holds the rods 60 ps at each temperature and then averages S over 20 ps more. At 263 K the rods are nematic with S = 0.83. Heating in steps of 10 K, S falls slowly to 0.65 at 303 K, then drops to 0.41 at 313 K and 0.15 at 323 K, and stays near 0.1 to 383 K. It crosses 0.5 at about 309 K. That is the clearing point of the model, the temperature at which the liquid crystal becomes an ordinary liquid with no preferred direction. Real 5CB clears at 308 K, 35 °C; displays use mixtures of similar molecules whose clearing points lie well above the temperatures a screen meets. The drop in the model is spread over a few tens of kelvin where a real sample switches within a degree, for two reasons the agent explains when asked: the cell holds the rods at a fixed volume, so it passes through a mixed state of nematic and liquid on the way, and 400 rods are few.

The cooling leg is the more surprising half of the chart. Stepping back down from 383 K, S stays near 0.1 all the way to 273 K and has only reached 0.29 at 263 K, where the heating leg began at 0.83. Regaining order needs a starting point: a nematic grows from a patch of rods that happen to agree, and in a cooling liquid that takes extra undercooling and time. The cell has no walls to seed it, and 80 ps per step is far too short for a nematic to grow on its own, so the liquid supercools through nearly the whole ramp. The gap between the two legs is hysteresis, the mark of a first-order transition, and it is the same reason the liquids in the phase-change article have to be given a seed crystal to freeze.

The second experiment is the switch inside a pixel. In a display, a voltage across the cell turns the rods, and light passes through the pixel or not depending on which way they point; the turning is the Fréedericksz transition. The lab applies an aligning field, which plays the part of the voltage. With the rods disordered at 383 K, S near 0.16 and the average direction leaning 35° from vertical, the agent switches the field on. Within about 9 ps S climbs to 0.47 and the tilt drops to 7°. When the field is switched off, the order it imposed stays near 0.43 for the 10 ps recorded. A student can try the same switch at 293 K, where the field only sharpens an order that is already there, and weaken the field to find the smallest one that still turns the rods.

A Polymer Melt and Its Glass Transition

A polymer melt is a tangle of long chains. Warm, the chains slide past one another and the material flows; cold, they lock in place and the material is a glass, hard and brittle and without any crystal order. The temperature at which that happens is the glass transition, and for polystyrene it is 373 K (100 °C), which is why a foam cup softens in boiling water. This lab packs twenty chains of twenty beads into a box, 400 beads in all, in the manner of Kremer and Grest; each bead stands for one styrene unit, and the melt opens at 450 K, well above the transition. Press Play and open the analysis panel: the mean squared displacement of the beads grows steadily with time, which is a liquid diffusing, and the radius of gyration of each chain fluctuates as the chain changes shape.

Ask the agent to cool the melt from 550 K to 200 K in steps of 50 K and it measures the diffusion coefficient at each step: 0.296 Å²/ps at 550 K, 0.146 at 450 K, 0.062 at 350 K, and 0.020 at 200 K. A falling coefficient by itself leaves the glass transition undefined, because a bead boxed in by its neighbors still rattles, and every 10 ps window reads a value above zero. What tells a liquid from a glass is whether the displacement keeps growing with time: in a liquid the mean squared displacement rises in proportion to the time elapsed, and in a glass it rises for a picosecond and then levels off. The agent fits that growth exponent at every step and charts it too. It falls from 0.74 at 550 K to 0.39 at 400 K, 0.24 at 350 K, and 0.01 at 200 K, where each bead moves only 0.8 Å, 0.15 of its own diameter, inside its cage. The exponent drops below 0.3 at about 370 K, give or take 50 K, and that is the glass transition of the model on a 10 ps time scale.

Polystyrene's own glass transition is 373 K, and the agent says plainly that a run landing near it is a coincidence: the uncertainty of one run is wider than the agreement, and repeat runs land tens of kelvin apart. A glass transition seen in picoseconds after cooling at a trillion kelvin per second is a kinetic transition: it moves with the cooling rate, the length of the window, the depth of the bead well, and the fact that nearly half of a 400-bead blob is surface, where a bead has fewer neighbors to cage it. Hold each step longer and the measured transition moves up. Warm the melt again and the motion returns.

A Chain That Folds and One That Does Not

A protein is a chain of amino acids, and the order of the amino acids decides whether the chain folds into a compact shape. The simplest model that shows this is the HP model of Lau and Dill, with two kinds of bead: hydrophobic beads, which attract one another because they would rather hide from water, and polar beads, which do not. Dill argued in 1990 that this hydrophobic effect is the dominant force in protein folding. Each bead in this lab is one residue, with the mass of an average amino acid and the 3.8 Å spacing of a protein backbone. Two chains of 24 residues share the box, each pinned at its middle bead so the two cannot drift into each other. One has its hydrophobic residues arranged so they can gather into a core; the other has the same count of each kind in an order that keeps them apart.

Press Play at 300 K and the first chain curls up, its orange beads clustered in the middle and the pale ones facing outward, a crude version of what every globular protein does. The second stays an open coil. The radius of gyration, the average distance of the beads from the chain's center, tells the two apart in numbers: 8.8 Å for the folded chain against 12.3 Å for the coil. Heating in steps of 50 K, the folded chain holds together and then opens: 9.7 Å at 350 K, 10.3 at 450 K, 13.7 at 500 K. It is halfway open near 464 K, the denaturing temperature of the model. Real proteins unfold a few tens of degrees above room temperature, which is why an egg white turns solid in a pan; this chain stays folded to a much higher temperature because its beads bind more strongly than real residues and there is no water around them. The coil wanders between 9 and 13 Å with no step of its own, because no arrangement of its beads ever brought enough hydrophobic ones together. Each radius fluctuates by an ångström or two at any temperature, so the trend matters more than any one point. Cool the box and the first chain refolds, though not always into the same shape.

A sequence is a hypothesis about folding, and the box tests it. A student can type any string of H and P into the Build Chain dialog, or ask the agent to build one, and run the same ramp.

Soap Makes Micelles

A soap molecule has two ends that want different things: a head that is happy in water and a hydrocarbon tail that is not. Put enough of them in water and they gather into balls called micelles, tails inside and heads outside, which is how soap carries grease away. This lab holds 60 such molecules, each four beads: a wide yellow head and a tail of three gray beads, the tails sized like Martini's. The box has no water in it. Following Cooke, Kremer, and Deserno, who showed in 2005 that lipids assemble into micelles and membranes without any solvent, the two rules of the model stand in for the water: the tails attract one another, more strongly than two oily beads would on their own because the attraction also has to do the work of the hydrophobic effect, and the heads repel everything.

The head is wider than the tail on purpose. Israelachvili, Mitchell, and Ninham showed in 1976 that the shape of the molecule decides the shape of the aggregate: a wide head on a thin tail is a cone, and cones pack into balls; a head no wider than its tails is a cylinder, and cylinders pack into sheets. That is why soap makes micelles while the two-tailed lipids of a cell membrane make bilayers. With a head as thin as its tails, the same 60 lipids pile into one lump whose surface is half tails.

The lipids start scattered at random, 40 separate clusters by the count the agent uses, which joins any two lipids whose tails touch. At 300 K the tails find one another quickly: 21 clusters after 50 ps, the largest with 11 lipids; three clusters by 300 ps; and by 1000 ps a single micelle of all 60, a ball with a thick shell of yellow heads and the gray tails fanning in toward its center. Two balls form first and merge, because the natural size of a micelle of this lipid is larger than 60. With a tail of two beads the same box makes two or three smaller micelles with free lipids drifting between them, the way a dilute soap solution looks above its critical micelle concentration.

Heating undoes it. Stepping from 300 to 600 K, the micelle sheds lipids and splits: still one cluster of 60 at 300 K, five clusters with the largest at 56 at 400 K, ten with the largest at 28 at 450 K, and 26 clusters with the largest at 12 at 600 K. The tail attraction, worth about 630 K per bead contact, can no longer hold against the thermal motion. Real soap stays below the boiling point of water; the box leaves the water out, so the model can be heated past it. Cool it again and micelles reform, usually not the same ones.

Four Models, Four Transitions

The table summarizes the four labs. Each model temperature is a measured run, and the real value is the handbook figure for the material the beads were fitted to, where one exists.

Two of the four model temperatures land within a few kelvin of the real material, and the agent is careful about what that means. The clearing point of the rods is close to 5CB's because the cell is periodic and the density was chosen so that temperature decides the order. The glass transition of the melt is a coincidence within its uncertainty. The folding chain and the micelle have no single real counterpart, because the HP model stands for every protein and the soap model for every surfactant, and both leave water out; their transition temperatures are the model's, and the agent says so.

Why Use AI to Assist Scientific Simulations?

Each experiment in this article needs a model built to the right scale, a temperature schedule with holds long enough for the system to settle, and a measurement that fits the question: an order parameter for rods, a diffusion coefficient and its growth exponent for a melt, a radius of gyration for a chain, a cluster count for lipids. The AI agent handles this setup for the student. Because it maps plain questions onto the right experiment, a student who wonders "why does a liquid crystal stop being one when it is heated?" or "why does soap work?" can run the simulation that answers it and read the answer off a chart.

A clearing point, a glass transition, a folded core, and a micelle all emerge from the forces among a few hundred beads, computed a femtosecond or two at a time, while the order, the motion, the shape, and the clusters are measured as they go. The AI agent sets up the simulation that answers a student's question.

Conclusion

This article uses soft matter to illustrate a new way to learn science: students converse with an AI agent that turns a plain-language question into a molecular dynamics experiment and hands back measured data. Liquid crystals, polymer glasses, protein folding, and micelles — phenomena of thousands of atoms and nanoseconds of time — become something students investigate in a browser: at what temperature each transition happens, what the molecules are doing when it does, and why the shape and sequence of a molecule decide the outcome. By pairing physically grounded coarse-grained simulations with an AI that can set them up and run them, AIMS turns a student's curiosity into a genuine scientific investigation of the materials of everyday life.

← Back to AIMS