Identical Particles

Here is an experiment with no moving parts to argue about. Collide a beam of alpha particles — helium-4 nuclei — with helium-4 gas, and count arrivals at 90° in the center-of-mass frame. Chapter 14 taught us exactly what to expect: the amplitude f(θ) for deflecting a projectile by θ, the cross-section |f(θ)|². But at 90° something is slippery. If the detector clicks, which particle arrived — the projectile deflected by 90°, or the target recoiling at 90°? The two final states are the same physical situation: two identical nuclei flying apart at right angles. No measurement, even in principle, can say which history happened.

We have been here before. In Chapter 1, when two paths led to the same detector and nothing recorded which was taken, we did not add probabilities — we added amplitudes, and interference appeared. Exchange is a double slit whose two slits are who is who. Classical bookkeeping predicts the 90° rate |f(90°)|² + |f(90°)|² = 2|f(90°)|². The 1956 measurement found four times |f(90°)|² — the amplitudes had been added first, |f + f|² = 4|f|², twice the classical count. Now repeat the experiment with electrons (spins aligned) and the 90° rate is zero: the amplitudes subtracted. Nature is telling us there are exactly two kinds of particle in the world, and it is telling us with factors of 2 and 0 — not effects you tune, effects you count.

One chapter ago these would have been paradoxes. This chapter they become a single postulate about a single sign, and by the end that sign will have explained why atoms have chemistry, why you don't fall through the floor, and why lasers work.

Let us do the bookkeeping honestly. For two particles, label states ψ(1, 2), where 1 stands for all of particle 1's coordinates (position and spin). Define the exchange operator P₁₂ that swaps the labels: P₁₂ ψ(1, 2) = ψ(2, 1). Two facts follow in one line each. First, swapping twice is doing nothing, so P₁₂² = 1 and the eigenvalues of P₁₂ are ±1. Second, if the particles are truly identical, the Hamiltonian cannot care about the labels — H(1, 2) = H(2, 1) — so

“Identical” is doing real work in that sentence. An electron and a muon have different masses; H distinguishes them; nothing here applies. But every electron in the universe has exactly the same mass, charge, and spin as every other — they are not similar, they are indistinguishable in the sense that no term of any Hamiltonian can tell them apart. For such particles, labels are our invention, and physics must not depend on our inventions.

Now the postulate that experiment forces. States of identical particles are not merely allowed to be eigenstates of P₁₂ — they are required to be, all of them, with one fixed sign per species:

and the spin–statistics theorem assigns the sign: integer-spin particles (photons, helium-4, pions) take the plus; half-integer particles (electrons, protons, neutrons, helium-3) take the minus. Within non-relativistic quantum mechanics this assignment is an experimental fact — we adopt it the way we adopted the Born rule. (The proof exists, but it lives in relativistic field theory: causality plus positive energies leaves no other option. We will not pretend to derive it here.) Equation (15.1) is what makes the choice permanent: an electron pair can never drift into symmetric behavior, because the symmetry commutes with every possible time evolution. Statistics is a career, not a mood.

Before the machinery, collect the scattering payoff. A detector at θ catches either the particle deflected by θ or its partner deflected by π − θ; indistinguishable histories, so amplitudes add with the species sign:

plus for bosons, minus for identical-spin fermions. At θ = π/2 the two amplitudes are equal, so the bracket is 2f or 0 — the factor 4 (twice classical) and the zero of Section 15.1, independent of every detail of the potential. For unpolarized spin-½ beams, a quarter of the collisions are singlets (symmetric in space) and three quarters triplets (antisymmetric), and the 90° rate lands at exactly half the classical count — the Mott result, confirmed in proton–proton scattering. Chapter 14's Numerov engine is still warm; here it is with the exchange sign wired in:

Slide k across its range and the three curves never collapse onto the classical sum: the boson curve rides a full factor of two high at 90°, the aligned-fermion curve pins to zero there, and the unpolarized average splits the difference at one-half — the cross-section rewritten by nothing but the exchange sign. That is the lesson worth carrying forward: indistinguishability is not a philosophical footnote, it is a measurable rearrangement of where particles land. Move the same effect from a collision to a beamsplitter and it becomes the Hong–Ou–Mandel dip in this chapter's lab; move it into a cloud of ultracold atoms and it is why identical bosons and fermions scatter by entirely different laws at low energy — a knob today's quantum-gas experiments dial on purpose.

Build two-particle states explicitly. Given single-particle orbitals ψ_a and ψ_b, distinguishable particles could occupy the plain product ψ_a(1)ψ_b(2). Identical particles cannot — the product is neither even nor odd under exchange. The allowed combinations are

and the minus case generalizes to N particles as a determinant — the Slater determinant, rows labeled by orbitals, columns by particles. Determinants vanish when two rows coincide, and that innocent algebra fact is the Pauli exclusion principle: put a = b in (15.4) with the minus sign and Ψ cancels identically. Not “costs energy,” not “is unlikely” — does not exist.

Now look at what the sign does to geometry. Compute the mean-square separation in the states (15.4) (exercise 2 — it is three lines):

Fermions sit farther apart than distinguishable particles would; bosons huddle closer. Nothing pushes them. There is no exchange term in H, no potential, no interaction — the spacing is enforced by the sign alone. Physicists call it the exchange “force,” always with the quotation marks audible. Watch it happen:

The trench along x₁ = x₂ is called the exchange hole, and it has a famous energy consequence. Take helium (Chapter 12's variational patient) with one electron excited: 1s2s. The spins can form a singlet (antisymmetric, forcing the spatial state symmetric) or a triplet (symmetric, forcing space antisymmetric). Same orbitals, same Coulomb repulsion e²/r₁₂ — but the triplet's spatial antisymmetry digs the exchange hole, keeps the electrons apart, and lowers their repulsion energy. Splitting the energies gives

the exchange integral — about 0.8 eV in helium, ten thousand times larger than any magnetic interaction between the two spins. Sit with that: the energy depends on the relative spin orientation, ferociously, through a Hamiltonian that never mentions spin. This masquerade — electrostatics dressed as spin coupling — is what ferromagnetism actually is. Iron's magnetism is Coulomb repulsion filtered through the Pauli sign, which is why it survives at room temperature when genuine spin–spin forces would melt at a few kelvin.

Scale the two signs up and they become the two great architectural styles of the universe.

Fermions stack. One particle per state, so N fermions in a potential fill the lowest N levels — a Fermi sea with a sharp surface. This single rule is the periodic table (Chapter 10 gave hydrogen's level diagram; Pauli populates it shell by shell, and chemistry is what the topmost, loosely-bound electrons do); it is why metals conduct (only electrons at the sea's surface can find empty states to move into); and it is a pressure. Squeeze N electrons into volume V and the exclusion principle alone — no repulsion, no temperature — forces kinetic energy to grow as it packs states up to higher momenta. That degeneracy pressure is what holds up a white dwarf star after its fusion dies. (Push past about 1.4 solar masses and even Pauli loses to gravity — Chandrasekhar computed that limit at nineteen, on the boat to England.)

Bosons pile on. Chapter 7's ladder algebra already knew the number:

The rate for emitting a photon into a mode grows with the number already there. The “1” is spontaneous emission; the “n” is stimulated emission — light already present recruits identical light, same frequency, same direction, same phase. Einstein deduced this factor in 1917 from thermodynamic consistency alone (exercise 6 walks his argument); build a cavity where the recruitment outruns the losses and you have a laser. Every photon in a laser beam is in the same mode because photons are bosons and (15.7) pays n + 1 to join the crowd. The same arithmetic drives atoms, cooled until their wavefunctions overlap, to condense into one macroscopic quantum state — but that story (and the Fermi sea's interacting sequel) is Chapter 16's.

Try to write a 10²³-particle wavefunction as a symmetrized sum of products and you will fill the observable universe with paper. The cure is to stop tracking who is where — the whole lesson of this chapter is that “who” is meaningless — and track only how many sit in each single-particle state:

This is Fock space, and the operators that move between its floors are old friends. For bosons, each orbital gets Chapter 7's ladder:

For fermions, one change of bracket carries the entire content of antisymmetry — commutators become anticommutators:

Read that last equation aloud: creating the same fermion twice is not forbidden by a rule — the operator squares to zero. Pauli exclusion, the exchange sign, the Slater determinant's alternating sum: all of it is packed into the curly bracket, and every state you can build with these operators is automatically, unavoidably antisymmetric. Hamiltonians translate term-for-term — hops and potentials become quadratic, pair interactions quartic:

The name “second quantization” is a historical accident and mildly a lie — nothing is quantized a second time. It is the same quantum mechanics with the symmetrization postulate compiled into the operator algebra instead of enforced by hand on wavefunctions. That claim — same physics, different bookkeeping — is precisely the kind of claim a lab can referee.

The lab (Rust-QP/ch15-fock) builds Fock space with no symbols and no trust: occupation bases enumerated directly, operators as literal matrices, and every claim of Sections 15.2–15.5 turned into an assert. Fermion modes are bits; the exchange sign is a sign string:

1/// annihilation matrix c_i on the 2^modes-dimensional Fock space
2fn fermion_c(modes: usize, i: usize) -> M {
3    let dim = 1usize << modes;
4    let mut c = M::zeros(dim, dim);
5    for mask in 0..dim {
6        if mask & (1 << i) != 0 {
7            // Jordan–Wigner sign string. Removing an electron from mode i costs a
8            // factor (-1)^(number of occupied modes BELOW i), because c_i must
9            // anticommute past each of those occupied modes to reach position i.
10            // `(1 << i) - 1` is the bitmask of all modes below i; `& mask` keeps
11            // the occupied ones; `count_ones()` counts them. This one line is what
12            // makes the c's ANTIcommute ({c_i, c_j†} = δ_ij) instead of commute —
13            // i.e. it is the Pauli exclusion principle, encoded in a parity bit.
14            let below = (mask & ((1 << i) - 1)).count_ones();
15            let sign = if below % 2 == 0 { 1.0 } else { -1.0 };
16            c[(mask ^ (1 << i), mask)] = sign;
17        }
18    }
19    c
20}
21// … and the algebra referees come back EXACTLY zero — tol 0.0, no epsilon:
22("fermions     worst |{c_i,c_j†} - δ_ij|", anti_worst, 0.0),
23("Pauli        worst |(c_i†)²|", pauli_worst, 0.0),

The centerpiece is built on the simplest model of interacting electrons ever written down — the Hubbard model. Chop space into discrete sites (think of atoms in a crystal). An electron may hop from one site to a neighbor with amplitude $t$ — that is just kinetic energy on a lattice, the freedom to move. And whenever two electrons (necessarily opposite spins, by Pauli) land on the same site, they pay a Coulomb penalty $U$ for crowding. Two numbers, one competition — mobility $t$ against repulsion $U$ — and almost every question in the physics of magnets, Mott insulators, and high-temperature superconductors is some corner of it.

The centerpiece referees the chapter's thesis. Two electrons on two sites (the minimal Hubbard model) are solved by two disjoint code paths — explicit antisymmetrized tensor products on one side, occupation numbers and sign strings on the other — and both against the closed-form spectrum. Three derivations, one answer, parts in 10¹⁵:

1// antisymmetric projector basis: (|αβ> - |βα>)/√2, α < β
2let mut basis_a = M::zeros(16, 6);
3// …
4for iu in 0..=16 {
5    let u = 0.5 * iu as f64;
6    // path B: second quantization, N = 2 sector
7    let h_full = &hop + &u_int * u;
8    let e_sq = sorted_eigs(&restrict_to_n(&h_full, 4, 2));
9    // path A: first quantization, antisymmetrized
10    let h2 = &kron_sum + &vpair * u;
11    let e_fq = sorted_eigs(&(basis_a.transpose() * &h2 * &basis_a));
12    for k in 0..6 {
13        hub_fq_sq = hub_fq_sq.max((e_sq[k] - e_fq[k]).abs());
14    }
15    // closed form: (U ± √(U²+16t²))/2, three 0s (triplet), U
16    let s = (u * u + 16.0 * t * t).sqrt();
17    let mut exact = vec![(u - s) / 2.0, 0.0, 0.0, 0.0, u, (u + s) / 2.0];
18    // …
19}
20// this run: 1st-q vs 2nd-q worst 5.3e-15 over the full U sweep, and
21// 3.6e-15 against the closed form — three derivations, one spectrum.
22("Hubbard      1st-q vs 2nd-q spectra", hub_fq_sq, 1e-14),
23("Hubbard      2nd-q vs closed form", hub_analytic, 8e-15),

The same six-state toy hands us a real prize: at large U the singlet–triplet gap follows 4t²/U — superexchange, the reason most magnetic insulators are antiferromagnets. The intuition is pure exclusion arithmetic: when $U$ is huge each site keeps exactly one electron, so no real hop can happen — but a virtual hop to a neighbor and back is allowed only if the two spins are opposite (Pauli blocks it for parallel spins). That borrowed excursion costs energy $U$ and gains kinetic amplitude $t$ twice, so second-order perturbation theory lowers the anti-aligned (singlet) state by of order $4t^2/U$ — an effective spin coupling conjured entirely from hopping and repulsion, with no magnetic force in sight. And two more famous numbers come out with no free parameters:

1let theta = std::f64::consts::PI / 4.0;
2
3// BOSONS. Two modes, cutoff 2 quanta each, |n₁,n₂> at index 3n₁ + n₂, so
4// |1,1> is index 4. The ladders carry the √(n+1) that makes the two split
5// amplitudes equal and opposite.
6let ab = boson_a(2);
7let i3 = M::identity(3, 3);
8let ab1 = ab.kronecker(&i3);
9let ab2 = i3.kronecker(&ab);
10let hom_boson = beamsplitter_coincidence(&ab1, &ab2, theta, 3 * 1 + 1);
11// (this run: 1.8e-33 — the dip, indistinguishable from the exact zero)
12
13// FERMIONS. Same routine, same θ, different operators: the two-mode Fock
14// space is 4-dimensional and |1,1> is bitmask 0b11 = 3. Pauli is not
15// asserted here and the |2,0> exit is not deleted by hand — it is simply
16// unreachable, because c† applied to an occupied mode gives the zero vector.
17let cs2: Vec<M> = (0..2).map(|i| fermion_c(2, i)).collect();
18let hom_fermion = beamsplitter_coincidence(&cs2[0], &cs2[1], theta, 0b11);
19// …
20// distinguishable: independent coins — both straight, or both cross.
21let hom_dist = (u1[(0, 0)] * u1[(1, 1)]).powi(2) + (u1[(0, 1)] * u1[(1, 0)]).powi(2);
22// …
23("HOM bosons   coincidence (target 0)", hom_boson, 1e-30),
24("HOM fermions |coincidence - 1|", (hom_fermion - 1.0).abs(), 4e-15),
25("HOM classical |coincidence - 1/2|", (hom_dist - 0.5).abs(), 4e-15),

That first assert is a whole tabletop experiment compressed to a line. In the Hong–Ou–Mandel setup (Figure 15.1), two identical photons arrive simultaneously, one at each input port of a 50/50 beamsplitter — a half-silvered mirror that transmits or reflects each photon with equal amplitude. A single-photon detector watches each output port, and a coincidence counter clicks only when both fire at once. Classically you would expect coincidences half the time. But there are two indistinguishable histories that end with one photon at each detector — both photons transmit, or both reflect — and for identical bosons these two exchange amplitudes are equal and opposite. They cancel exactly. The photons always leave through the same port, bunched together, and the coincidence rate collapses to zero: the celebrated HOM dip, the double slit of Section 15.1 wearing a beamsplitter for its wall.

1. (F) Prove the two one-liners of 15.2 carefully: P₁₂² = 1 forces eigenvalues ±1, and [P₁₂, H] = 0 for any H symmetric under label exchange. Then explain why a symmetric H is not an assumption but the definition of “identical.”
2. (F) Derive (15.5). Then compute the three rms separations for the box orbitals n₁ = 1, n₂ = 2 (use ⟨x⟩ = ½, ⟨x²⟩_n = ⅓ − 1/2n²π², and ⟨1|x|2⟩ = −16/9π²) and check your three numbers against the readout in the PauliBox widget.
3. (C) In the collision explorer, verify the three 90° verdicts survive every k you can dial. Then hunt for an angle where the aligned-fermion curve rises above the distinguishable one — antibunching in angle — and explain the sign of Re f*(θ)f(π−θ) there.
4. (F) Show that the N-particle Slater determinant is normalized when built from orthonormal orbitals with prefactor 1/√N!, that it vanishes whenever two orbitals coincide, and that ⟨n_i⟩ computed in it reproduces the naive occupation picture.
5. (F) The unpolarized Mott formula: for spin-½ beams with random spins, show the ¼ singlet / ¾ triplet weights give dσ/dΩ = |f|² + |g|² − Re f*g (g ≡ f(π−θ)), and that identical bosons with random spin s would give a +Re f*g/(2s+1) cross term instead.
6. (F, hard) Einstein's 1917 blackmail, run forward. Two-level atoms in a photon gas: let the emission rate be proportional to n + 1 (equation 15.7) and absorption to n. Demand equilibrium with the Boltzmann ratio of atomic populations and derive Planck's spectrum. Then run it backward, as Einstein did: assume Planck and show the “+1” is forced — stimulated emission was discovered by thermodynamic bookkeeping, forty years before the laser.
7. (F, hard) The spin Hamiltonian masquerade. For two electrons in orthogonal orbitals with Coulomb repulsion, show the singlet–triplet splitting (15.6) is reproduced exactly by the effective operator H_spin = ¼(E_s + 3E_t) − (E_s − E_t)(¼ + Ŝ₁·Ŝ₂/ℏ²) acting on spins alone — no spatial coordinates anywhere. (Hint: Ŝ₁·Ŝ₂ distinguishes singlet from triplet; Chapter 11's addition machinery gives its two eigenvalues.) This is how Heisenberg models are born from Coulomb physics, and why the lab's 4t²/U gap is called an exchange coupling.
8. (P, hard) Give the lab a superfluid-to-Mott story: 4 bosons on a 4-site Bose-Hubbard ring (dimension 35), sweep U/t from 0 to 40, and plot the ground-state number variance ⟨Δn²⟩ per site. Referee both ends analytically: at U = 0 the ground state puts every boson in the k = 0 orbital and the on-site variance is that of a binomial, N(1/M)(1 − 1/M) = ¾; at U → ∞ the Mott state |1111⟩ freezes it to 0. Then locate the crossover and compare U_c/t with the thermodynamic-limit value (≈ 3.3 per boson at unit filling) to see how honest a 4-site universe can be.