Condensed Matter Ep 3: Why Metals Work

A plot on a laboratory wall: resistivity against temperature for a copper-oxide superconductor above the temperature where it superconducts. The line is straight. It is straight down through a hundred kelvin, through ten, and as far down as anyone can push it with the superconductivity switched off by a magnet. That is the whole problem, and it has been on the wall for four decades. The third conversation in this series proved — not fitted, proved — that a metal's resistivity must rise as the *square* of temperature. Two Pauli squeezes on the available phase space, and no interaction strength anywhere in the argument, which is why it holds for copper and for the heavy-fermion compounds where electrons are a thousand times heavier than free ones. A straight line is not a small deviation from that. It is a different law. So this finale rebuilds the theorem on air and then asks which assumption dies. The answer is precise, and it is the episode's central takeaway: **the excitation is sharper than the energy it carries**. That is what makes a quasiparticle a thing at all — a bump narrow enough to have a name. In a strange metal the width equals the energy. There is no small parameter, no long-lived excitation, and therefore no Fermi surface to count seats around. The apparatus that explains every ordinary metal does not fail by a bit; it has nothing to describe. Also: why a Mott insulator is not a band insulator — a material every band calculation calls a metal that refuses to conduct, because the electrons jam each other rather than running out of states; the Zhang–Rice singlet; why doping one is so hard, since a moving hole drags a string of broken magnetic bonds behind it and pays for every link; the pseudogap and the Fermi arcs; quantum criticality; **Planckian dissipation**, the scattering rate at the fastest anything is allowed to relax; the Hall-number jump that says the carrier count changes from the doping to one-plus-the-doping across a particular concentration; and **quantum spin liquids** — order with no order parameter, the last callback to the broken-symmetry conversation. Then how the field is actually attacking it: **moiré materials** as designer strong correlation, where two graphene sheets twisted by about one degree produce a superconductor you can tune with a knob; and cold-atom and quantum-computer simulation of the Hubbard model, with a straight account of what those machines have and have not yet shown — including that the pure two-dimensional Hubbard model appears *not* to superconduct until one further hopping term is added. The series ends where it has to. Nothing is resolved; high-temperature superconductivity is unsolved after forty years. But the closing observation is the one all twelve conversations were built toward: every time this subject has understood something, it did so by finding a new object — a macroscopic phase, a dressed excitation, an order parameter, a diverging length, an integer belonging to a band — and never by working harder on the thing written in the Hamiltonian. The strange metal is not unsolved because the equations are difficult. The Hamiltonian is three lines long. It is unsolved because nobody has yet found the object. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

A metre of copper wire obeys a rule that never lets you down: halve the length, halve the resistance. Cut it to a centimetre and the resistance falls by a hundred. Cut it to a micron and it falls again, faithfully. Now swap the copper for a sheet of electrons buried a hundred nanometres under the surface of a gallium arsenide chip, cooled to a few hundredths of a degree above absolute zero, and squeeze the channel shut with a pair of gates. The resistance stops falling smoothly. It comes down a staircase — and the treads are at values built out of nothing but the electron's charge and Planck's constant. This conversation is about what happens when a conductor gets small enough that Ohm's law stops being true, and it starts by having the answer guessed before it is derived. Conductance is current over voltage; that is charge squared divided by energy times time; and energy times time is an action. So the natural unit of conductance is the electron charge squared over Planck's constant, and you can get that far with no physics at all — which is the point, and also the limit. It gives you the currency and not the amount. The amount is Landauer's, and it inverts what resistance means. Conductance is not an obstruction to be overcome; it is a **transmission probability**. A perfect channel does not have zero resistance — it has about thirteen thousand ohms, and that resistance does not live in the channel at all. It lives in the join, where a wide reservoir has to squeeze its current into a countable number of ways through. Which is also why the number is universal: every one of those experiments got the same step height out of a different chip. Then the hero: pinch two constrictions nearly shut and the puddle between them becomes a **quantum dot** — an artificial atom, and the word is meant literally. It has discrete levels for the same particle-in-a-box reason an atom does, three orders of magnitude larger in size and getting on for five orders smaller in energy. Its shell closings come at two, six and twelve, and it obeys Hund's rule. Put one more electron on and you pay a charging energy; that price is why the current comes in sharp peaks separated by dead zones, and why sweeping gate against bias draws diamonds you can read the addition energy off with a ruler. The last act is the Kondo effect, one of very few strong-correlation problems that is genuinely solved — you can compute the resistivity, susceptibility and specific heat at every temperature, each a universal function of one measured number. In a bulk metal a magnetic impurity makes resistance *rise* as you cool. In a dot it does the opposite, because the impurity is not beside the road, it *is* the road — and the conductance climbs to the maximum a single channel is allowed. Ends where every tool in the hour fails: the zero point seven anomaly, which has been argued about since nineteen ninety-six. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

In nineteen eighty-eight Duncan Haldane drew a honeycomb lattice on a sheet of paper and threaded a magnetic field through it that adds up to exactly nothing — up through the middle of each hexagon, down again around the edges, net zero. There is no field to speak of, and yet the model carries the same quantised Hall conductance the previous conversation spent an hour earning from an eighteen-tesla magnet. It took about twenty-five years before anyone built one. That is the opening move, and the harder prize follows immediately. Keep time-reversal symmetry — do not break it internally, do not apply anything — and the total curvature over the Brillouin zone is forced to zero, so the Chern number is zero and by the previous hour's argument there should be nothing left to say. There is. The invariant that survives is not a whole number but a parity: **even or odd**, a Z-two count, and odd is a material that insulates through its bulk and conducts only on its skin, with nothing in the room switched on. What replaces the magnet is **spin–orbit coupling** — an electron moving through a crystal's electric field sees, in its own frame, a magnetic field whose direction depends on which way it is going. Opposite-moving electrons see opposite fields. So the two spins get opposite Hall responses that cancel in total while leaving something behind at the edge: a **helical** pair of channels, one spin running each way, locked to their directions. The most satisfying idea in the hour is why that edge resists dirt. To backscatter, an electron must reverse its direction, which means reversing its spin — and the two paths around a non-magnetic impurity differ by a full rotation of a spin one-half, which contributes a minus sign. The two reflection amplitudes cancel exactly. Not "are suppressed". Cancel. But the protection is only as good as the symmetry it rests on: bring in a magnetic impurity and it is gone. That is what "symmetry-protected" buys you, and exactly what breaks it. Then the experiments, in order: Kane and Mele's two two-thousand-and-five papers on graphene where the effect is far too small to see; Bernevig, Hughes and Zhang predicting the right material — a mercury telluride quantum well past a critical thickness where the bands invert; König and colleagues at Würzburg measuring it in two thousand and seven and finding the conductance plateau the theory demanded; then the three-dimensional versions, and the single Dirac cone on the surface of bismuth selenide seen directly with photoemission. The last section is the Majorana mode and the topological qubit — a state that is its own antiparticle, split across the two ends of a wire so that no local disturbance can read it. The theory is solid; the experimental record is not. That is stated plainly, retractions included, without hype and without performing scepticism about it. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

Two in the morning, the fifth of February nineteen eighty, in the high-field laboratory at Grenoble. Klaus von Klitzing is watching a chart-recorder pen that should be drawing a ramp. It is drawing a staircase. On each flat tread the Hall resistance is a fixed number — and that number turns out to be Planck's constant divided by the electron charge squared, divided by a whole number, to a precision that would eventually be measured in parts per billion and that does not care what the sample is made of. The previous conversation built the geometry: a phase that depends on the shape of a path rather than its speed, a curvature whose local density can be integrated, and a Brillouin zone that is a torus with no edges. This one spends it. The whole number on the plateau is that curvature added up over the zone and divided by two pi — the result Thouless, Kohmoto, Nightingale and den Nijs got in nineteen eighty-two, sixteen months before anyone could say what kind of object they had found. The most counter-intuitive idea in the hour is that **the dirt is essential**. A perfectly clean sample would show no plateau at all — just the ramp. Disorder localises most of the states, leaving a thin percolating thread of extended ones, and it is the localised states that give the plateau its width. Prange showed in nineteen eighty-one that when an electron gets trapped, the ones still moving carry *exactly* the current it stopped carrying. Not approximately. Exactly. Nothing in that calculation explains why — which is precisely why the topology is needed. Then the argument that produces a metal without going anywhere near the surface: the invariant is a whole number inside the sample and zero outside, integers cannot change by degrees, so the gap must close somewhere on the way out. That is the edge state, derived from the bulk. And a caution — what topology guarantees is that gapless boundary modes *exist*, not that they carry your current; imaging finds the current in the bulk in some samples and at the edges in others. Also: Landau levels and why the degeneracy is one state per flux quantum; why a topological integer cannot creep; Laughlin's gauge argument; and the honest admission that the clean band-structure derivation applies to exactly the case with zero plateau width, with the repair (twisted boundary conditions) and the cost of that repair both stated. The last third is the fractional effect, which is a different kind of thing entirely. Tsui and Störmer found a plateau at one third of a filled level in nineteen eighty-two; Laughlin wrote down a wavefunction for it in nineteen eighty-three that remains a guess with very good support rather than a derivation. Its excitations carry one third of an electron charge — measured directly by shot noise in nineteen ninety-seven, in Saclay and at the Weizmann Institute independently — and they are neither bosons nor fermions. Switch the interaction off and the state does not weaken; it ceases to exist. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

Walk a spear from the North Pole down to the equator, a quarter of the way round, and back up — keeping it always pointing as straight ahead as the ground allows, never twisting it in your hands. It comes home turned through a right angle. Nothing rotated it. The rotation is a property of the journey. This is the most abstract hour in the series and it says so out loud, then refuses to leave the ground: one spin, one magnetic field, one hand moving the field slowly round a loop. The claim to be earned is that the spin comes back carrying a phase that has nothing to do with how long you took. Go round twice as slowly and it is unchanged — which is exactly what a physicist's instinct says is impossible, because the obvious phase, the one from the energy, scales with duration. Alice reasons her way to that wrong answer on air before it is corrected. The distinction, once you have it, is hard to un-see. The dynamical phase is a sum over *when*. The leftover is a sum over *where* — each term fixed by two neighbouring settings of the apparatus, with no time in it anywhere. That is why slowness cannot stretch it. Berry wrote it down in nineteen eighty-four; Pancharatnam had the same quantity for polarised light in nineteen fifty-six, aged twenty-two, in a paper communicated by his uncle C. V. Raman and then ignored for thirty years. Then the geometry proper. The phase is the area swept out on the sphere of directions — half the solid angle for a spin one-half — which immediately raises the question the rest of the series turns on: a loop divides a sphere into two regions, so which one is the area? The answer only works because the two disagree by exactly two pi, and a phase cannot tell the difference. That ambiguity is not a blemish. It is the seed of an integer. Also: why the arbitrary phase you are free to assign at every point cancels round a closed loop and nowhere else; the degeneracy that acts like a magnetic monopole in parameter space, and how far the Aharonov–Bohm analogy may honestly be taken before it stops; anomalous velocity — how a phase, which is not a trajectory, nevertheless pushes a moving electron sideways, and why monolayer molybdenum disulphide has curvature everywhere while its bilayer has none; the Zak phase and why a crystal's Brillouin zone has no edges but is a torus, its walls being gluings rather than boundaries. It ends where the geometry stops being a theorem about a doughnut: von Klitzing's plateaus, flat to parts in a billion, and Thouless and company showing two years later that the whole number on the plateau is this curvature added up over the zone. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

You can buy the demonstration from a catalogue: a thick-walled cell of sulphur hexafluoride, liquid below, gas above, a flat line between them. Warm it towards forty-five degrees and the line goes soft, wavers, and stops existing — no boiling, no bubbles. And in the half-minute before it goes, the whole cell turns milky white. Two centimetres of it will not pass the light of a lamp. To scatter light that hard, a clear fluid needs structures the size of the wavelength of light — about a thousand molecules across — and it has just produced them out of nothing but being at a transition. That whiteness is the subject. Every strange thing about critical points follows from one fact: approaching the transition, the size of the largest correlated patch grows without bound, and at the transition there is no characteristic size left in the problem at all. The method is to squint. Kadanoff's move in nineteen sixty-six: replace each block of spins by a single spin carrying the block's majority, rescale, and ask what the couplings look like now. Repeat. You are not solving the system; you are watching where it flows. A system with no characteristic size is one the squint cannot change — a fixed point — and everything the transition does is set by which fixed point it flows to, not by what the material is made of. Wilson turned that into a calculation and took the nineteen eighty-two Nobel for it. That is the answer to the scandal the previous conversation left open. Uniaxial magnets, the liquid-gas critical point of every simple fluid, binary alloys ordering, two liquids unmixing — nothing in that list resembles anything else in it, and they share critical exponents to the measurable digit, because they flow to the same fixed point. Guggenheim's eight liquefied gases falling on one curve is the same statement, drawn in nineteen forty-five. Also: what "irrelevant" means physically — two samples differing only in that feature become not merely similar but indistinguishable under enough squinting, which is why the microscopic details are allowed to not matter; why four dimensions is the dividing line and what changes below it; why the exponents are not independent of one another, so pinning down two decides the third; the Harris criterion and what happens when you deliberately dirty a sample; and the small-number expansion that made the whole thing computable. And a reversal worth the hour: Landau's construction was declared broken last time, and it comes back — what he wrote down was always the coarse-grained free energy stopped at a short distance, which is analytic and fine. The non-analyticity everyone objects to belongs to the exact free energy with every scale still in it. The flaw was never in what he wrote. It was in the step he did not take. One limit belongs beside all of it, because it is the price of the method: nothing here predicts a critical temperature. The flow tells you which destination a system reaches and nothing about how far it had to travel, so every number a chemist would actually want is still measured rather than derived. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

On the eighth of April, nineteen eleven, Kamerlingh Onnes pumped a mercury thread down to about three kelvin and the resistance vanished — and he read it as confirming his own prediction that a pure metal slides smoothly to zero. He was wrong, and it took him half a year to find out. The cliff arrived on the twenty-sixth of October: within a hundredth of a degree, at four point two kelvin, from unmeasurable to a tenth of an ohm. Nothing slides like that. He suspected a short circuit first, and looked for one. And then twenty-two years passed before anybody discovered the property that actually defines the state. Zero resistance is not it. A perfect conductor would trap whatever field was threading it when it cooled — its final state would depend on its history, which means it has no thermodynamic state at all and no free energy to write down. What Meissner and Ochsenfeld found in nineteen thirty-three is that a superconductor *expels* the field either way. History-independence is what buys you a phase, a boundary, and everything Landau's machinery can do — seven years before anyone knew what the electrons were doing. The central puzzle is why pairing happens at all. The attraction between two electrons in a metal is absurd: an electron dragging a wake of displaced positive ions, an effect a ten-thousandth the size of the repulsion it is fighting. In free space an attraction that weak binds nothing — there is a threshold, and it is a *counting* fact, because the number of available states vanishes at zero energy. Cooper's move is to notice that the Fermi sea blocks everything below its surface, so the counting starts where the density of states is a large constant instead of zero. Constant density with an energy in the denominator gives a logarithm, and a logarithm has no ceiling: it can always be made to match one over any coupling, however small. Which means the normal state was never stable. The calculation destroys its own starting assumption. Then the number that settles the argument. Thread a superconducting ring, insist the wavefunction be single-valued, and the trapped flux must come in whole units of Planck's constant over the carrier's charge. Deaver and Fairbank in Stanford and Doll and Näbauer in Munich measured the period independently in nineteen sixty-one, back to back in the same volume. It comes out at *twice* the electron charge. One number, read off the spacing of an oscillation, and the thing doing the condensing is a pair. Also: the gap and what tunnelling sees; Ginzburg–Landau as Landau theory with a charged order parameter, and Gor'kov deriving it from BCS nine years later; type one versus type two, Abrikosov's vortex lattice — which he sat on for four years and then got the wrong shape for; what actually holds the magnet up over the cooled ceramic (not expulsion, which cannot suspend anything below a plate, but thousands of millions of pinned flux threads); the Josephson effect and the volt; and Anderson showing in nineteen sixty-three that the photon acquires a mass inside a superconductor — the mechanism particle physics borrowed the following year. The limit is stated plainly: BCS does not explain the copper oxides. Four specific failures, not a slogan. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

Stand a glass beaker in a bath of liquid helium below about two kelvin, rim well clear of the surface, and it empties itself. Uphill, over its own rim, through a film thirty nanometres thick — a hundred atoms deep, coating every surface in the cryostat. Nothing pumps and nothing pushes, and it does not stop until the beaker is dry. The previous conversation built Landau's wine-bottle trough and the stiffness that comes with it. This one spends that stiffness. The claim to be earned is precise, and it is not the one most people carry: superflow is not frictionless by magic, it is *metastable* — the fluid cannot find a way to lose momentum in small enough pieces, so it keeps it. The argument is Landau's and it is one line of geometry. Draw the excitation spectrum, draw a straight line from the origin, and the slope of the steepest line that still touches the curve is the fastest the liquid can flow without being able to shed anything at all. A free particle's spectrum bends away from the origin like a parabola and that line has zero slope — an ideal Bose gas is not a superfluid, which is the first surprise. Interactions are what lift the curve into a straight rise at small momentum, and the straight rise is what buys the critical velocity. Then the second surprise: helium's measured critical velocities are far *below* Landau's number, because vortices nucleate at rough patches long before rotons do, and the honest version of the theory is the one that says so. Also: why the wavefunction's phase must return to itself around a loop, and how that single requirement forces circulation to come in whole units rather than any amount; the two-fluid picture and Andronikashvili's stack of discs in Kapitza's Moscow institute, which weighed the normal fraction directly; the roton minimum and what it is a memory of; why only about nine per cent of helium-four is in the condensate while all of it flows, so "condensate" and "superfluid" are not the same word; Gross–Pitaevskii assembled from Schrödinger plus one term; Kapitza coining "superfluid" in the last sentence of his paper, by analogy with superconductors; and the cold gases, where the same physics can be ordered to specification — including the two clouds released to overlap and interfere, a year and a half after the first condensates. The closing strand is helium-three, which has to pair before it can do any of this: fermions with no charge, a complicated order parameter, and thirty years between the question and the answer. Next time the particles that pair are charged, and one number — a factor of two in a measured flux — proves it. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

Heat a paperclip past a little over a thousand kelvin and it stops being magnetic. Let it cool and the magnetism returns — pointing somewhere. Nothing in the laws of physics prefers that direction over any other, and yet the iron picks one. Not "settles into the lowest energy state": there is no lowest state, there are infinitely many, all identical, and it takes one. The first conversation in this series argued that symmetry breaking is a genuine large-number effect, using a sugar molecule. This one does the version a molecule cannot do: a sharp transition at a definite temperature, a continuously degenerate family of choices rather than two, and — the consequence nobody expects — free waves and a stiffness that no single molecule can possess. The method is Landau's, and it is almost insultingly simple. Near the transition the order is small, so expand the free energy in powers of it. Symmetry kills every odd term. The quadratic coefficient changes sign at the critical temperature — that is the entire content — and a positive quartic term stops the runaway. The bowl becomes a wine bottle, the marble rolls off the hump into the trough, and the order grows as the square root of the distance below the transition. Landau wrote it in nineteen thirty-seven at twenty-nine, a year before the Lubyanka. Then the two consequences that matter. Because the trough is a continuous circle, you can rotate the choice slowly across the sample and the cost falls without limit — that is a Goldstone mode, a wave that costs arbitrarily little, and in a magnet it is a magnon you can watch directly with neutrons. And run the same sum in three dimensions instead and the cost of twisting a sample *grows* with its size — that is rigidity, which is why a solid resists shear, why a magnet holds its direction against any local disturbance, and (next time) why a superfluid flows forever. One stiffness: ask it statically and you get a modulus, ask it dynamically and you get a wave. The honest ending is that the method is wrong where it matters most. Landau theory predicts the order grows as a square root — an exponent of one half. Real magnets give about a third; the two-dimensional Ising model gives one eighth; helium's heat-capacity exponent has been measured to four decimal places aboard a space shuttle. Mean-field theory assumes each part feels the average of its neighbours, and close to the transition there is no meaningful average — the sample becomes a froth of ordered patches of every size at once. Why that froth appears, and why utterly different systems share the same wrong-looking numbers, is a later conversation. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.

In an ordinary copper wire, the electrostatic repulsion between neighbouring electrons is about six electron volts. The energy scale of the electrons themselves is about seven. The interaction everyone ignores is the same size as everything they keep — there is no small parameter, nothing to expand in, no excuse. And yet the simple picture works: copper conducts, band theory predicts, and photoemission sees a razor-sharp Fermi surface exactly where the arithmetic says there should be a smear. This conversation pays a debt. The previous one stated the quasiparticle claim — that a soup of violently interacting electrons behaves like a gas of barely interacting ones — and then refused to justify it, putting three conditions on the record instead. Here they are settled, in order, and the settlement is Landau's. The key move is adiabatic continuity: turn the interaction up gradually rather than switching it on, and every state of the free system slides continuously into a state of the real one. Nothing is born, nothing dies. You lose the right to say what any state looks like; you keep the right to count them. Alice attacks it exactly where it is weakest — the levels in a real metal are spaced by the Fermi energy divided by ten to the twenty-three, so no process in the universe is slow compared with that — and the repair is what makes the theory work. Then the argument the previous conversation would not allow: the phase-space count. An excitation just above the Fermi surface can only scatter if its partner comes from a thin shell just below, and both outgoing states must land above. Two independent squeezes, so the decay rate goes as the excitation energy squared — while the energy itself goes as the energy. The ratio shrinks without limit. Quasiparticles do not merely survive near the Fermi surface; they get better defined the closer you look. Not asymptotically adequate — asymptotically exact. And the parameter controlling it turns out to be the excitation energy over the Fermi energy: the small parameter this series opened by insisting did not exist. Also: what the quasiparticle weight Z actually means and what photoemission actually measures; why the volume of the Fermi surface is fixed by the electron density no matter how hard the electrons push on each other; why the resistivity of a clean metal rises as the square of temperature, and why that needs the lattice to absorb the momentum; effective masses from a quarter heavier in potassium to a thousand times bare in the heavy fermions; and finally where the whole picture fails — the strange metals, whose resistivity is straight-line in temperature and which have no quasiparticle peak at all. Landau worked most of this out with almost no calculation, four years before a car crash ended his career. ——— Series: Physics - Tutorial · Grad level. Every equation spoken in full, no chalkboard required. Built by Fermi AI.