Exact Mappings to Free Fermions and the Onsager Algebra

Why is the two-dimensional Ising model solvable when almost no other interacting many-body system is? Onsager's 1944 calculation was so intricate — Clifford algebras, quaternions, a $2N$ -dimensional representation — that for years even experts could not say why it worked, only that it did. The modern answer is a single structural fact, and it is beautiful: the transfer matrix that builds the lattice one row at a time is the exponential of an operator that is merely quadratic in fermion creation and annihilation operators. A quadratic fermion problem is, like a set of independent harmonic oscillators, exactly diagonalizable. Every thermodynamic quantity collapses to a sum over non-interacting single-particle modes. That is the whole secret. The bridge from spins to fermions is the Jordan–Wigner transformation. It represents Pauli spin operators along a chain — a single row of the lattice, or the auxiliary quantum chain of the transfer matrix — in terms of spinless fermions dressed by a nonlocal "string." The string is what converts the commutation of spins on different sites into the anticommutation fermions demand, and it is exactly what makes an apparently four-spin object bilinear in fermions. A Fourier transform then decouples momenta, and a Bogoliubov rotation mixes particles and holes within each momentum pair, leaving a gas of free fermionic quasiparticles with a definite dispersion. Onsager's own route was different and, in hindsight, even more astonishing. He noticed that the two operators in the transfer matrix — the bond term and the field term — generate, through repeated commutation, an infinite-dimensional Lie algebra that closes: the Onsager algebra. This algebra organizes infinitely many conserved charges and yields the transfer-matrix eigenvalues without ever mentioning a fermion. The two pictures — free fermions and the Onsager algebra — are two faces of the same integrable structure. From either, one reads the free energy, the self-dual critical point $\sinh 2K_1\sinh 2K_2 = 1$ , and the exact exponents $\alpha = 0$ (log), $\beta = 1/8$ , $\nu = 1$ , $\eta = 1/4$ .

Key takeaways

The 2D Ising transfer matrix is the exponential of a quadratic form in fermion operators — the deep reason the model is exactly solvable.
The Jordan–Wigner transformation ( $\sigma^x_j = 1 - 2c_j^\dagger c_j$ , with a nonlocal string) turns spins into spinless fermions and the transfer matrix into a free-fermion Hamiltonian.
Fourier plus Bogoliubov diagonalization gives the quasiparticle dispersion $\gamma_k$ ; the gap closes at the critical point $\sinh 2K_1\sinh 2K_2 = 1$ .
The gap closes linearly, $\gamma_{k=0} \sim |T - T_c|$ , so the correlation length diverges as $|T-T_c|^{-1}$ : the exponent $\nu = 1$ .
Equivalently, the two transfer-matrix seeds generate the closed Onsager algebra (Dolan–Grady relations), furnishing infinitely many commuting charges — the algebraic hallmark of integrability.
Key idea: products of Pauli operators along a chain become bilinears in Jordan–Wigner fermions, so the Ising transfer matrix is exp of a bilinear — a free-fermion gas — whose gap closes linearly at criticality, printing $\nu = 1$ and the log-divergent specific heat $\alpha = 0$ .

The core insight

Products of Pauli operators along a chain become bilinears in Jordan–Wigner fermions, so the Ising transfer matrix is the exponential of a quadratic fermion form — exactly diagonalizable as a free-fermion gas — and equivalently the two seed operators generate the closed Onsager algebra, whose structure dictates the spectrum.

Where this lives today

The free-fermion structure of the Ising model is the foundation of a whole industry. The one-dimensional transverse-field Ising chain — the exact quantum sibling of the 2D classical model — is the textbook example of a quantum phase transition, and its Jordan–Wigner fermions are the simplest arena in which Majorana modes and topological order appear. The Kitaev chain, a cornerstone of the theory of topological quantum computing, is essentially this same free-fermion problem read in a topological light, with the domain-wall fermions of the Ising model reincarnated as Majorana zero modes at the ends of a wire. The Onsager algebra, once a curiosity of a single solved model, is now recognized as a member of a family of infinite-dimensional symmetries (loop algebras, Yangians) that organize integrable systems generally. Superconducting-qubit and trapped-ion processors now realize small transverse-field Ising chains directly and measure the closing of the fermion gap as the field is tuned through criticality — the abstract dispersion of the exact solution turned into laboratory data, quasiparticle by quasiparticle.

The physical picture

A transfer matrix advances the lattice one row at a time; written as $V = e^{-\tau H}$ it defines an effective one-dimensional quantum Hamiltonian $H$ . For the Ising model this $H$ is the transverse-field Ising chain: bonds $\sigma^z_j\sigma^z_{j+1}$ that want neighbours aligned, and a transverse field $\sigma^x_j$ (set by the dual coupling) that wants each spin flipped. The reason a two-dimensional classical problem becomes a one-dimensional quantum one is exactly this row-to-row picture — the extra dimension turns into "imaginary time." Now think of a fermion as living on a bond between two spins. Flipping a pair of neighbouring antiparallel spins moves a domain wall by one lattice site — and a domain wall, in the fermion language, is a particle that propagates freely. Two such particles never interact; they pass through one another, picking up only the minus sign of fermionic statistics. That non-interaction is why the free energy is a single-particle integral: the model is genuinely a gas of non-interacting quasiparticles, however complicated the underlying spin problem looks.

Figure

The mechanism — Jordan-Wigner maps a spin chain to free fermions; a domain wall becomes a propagating particle

The Onsager algebra is the statement that two seed operators — the field term $A_0 = \sum_j \sigma^x_j$ and the bond term $A_1 = \sum_j \sigma^z_j\sigma^z_{j+1}$ — generate, by repeated commutation, an infinite tower of operators $A_m, G_m$ with relations that close: $[A_l, A_m] = 4G_{l-m}$ , $[G_m, A_l] = 2A_{l+m} - 2A_{l-m}$ , $[G_l, G_m] = 0$ . This closure is the algebraic reason a Hamiltonian of the form $\alpha A_0 + \beta A_1$ is integrable — it commutes with infinitely many independent charges. Criticality, in either picture, is the closing of the fermion gap: at $T_c$ the lowest quasiparticle energy vanishes linearly in momentum, the spectrum becomes that of a massless relativistic Majorana fermion, and conformal invariance with central charge $c = 1/2$ emerges.

Figure

The punchline — the fermion gap γ(0) closes linearly to zero exactly at T_c, then reopens

The core equations

The starting point is the anisotropic square-lattice Ising model, with possibly different horizontal and vertical couplings. Keeping them distinct exposes the structure; the isotropic case $K_1 = K_2$ is recovered at the end.

-\beta H = \sum_{\langle ij\rangle} K_1\,\sigma_i\sigma_j^{(\text{horiz})} + \sum_{\langle ij\rangle} K_2\,\sigma_i\sigma_j^{(\text{vert})} \qquad (1)

where: $K_1, K_2 = \beta J_1, \beta J_2$ — dimensionless horizontal/vertical couplings; $\sigma_i = \pm1$ ; $\beta = 1/k_BT$ . The two directions play asymmetric roles in the transfer-matrix construction: one becomes the quantum chain, the other the discrete "time" step. The transfer matrix factorizes into a transverse-field piece (from the horizontal bonds, via duality) and a bond piece (from the vertical bonds). The dual coupling $K_1^*$ appears because summing over an intermediate row-spin turns a horizontal bond into a transverse field.

V = V_2^{1/2} V_1 V_2^{1/2}, \quad V_1 \propto \exp\!\Big(K_1^{*}\!\sum_j \sigma^x_j\Big), \quad V_2 = \exp\!\Big(K_2\!\sum_j \sigma^z_j\sigma^z_{j+1}\Big), \quad \tanh K_1^{*} = e^{-2K_1} \qquad (2)

where: $V$ — symmetrized row-to-row transfer matrix, $Z = \operatorname{Tr} V^M$ ; $\sigma^{x}, \sigma^{z}$ — Pauli operators on the auxiliary $N$ -site chain; $K_1^*$ — Kramers–Wannier dual of the horizontal coupling; $M, N$ — rows, columns. The transverse field $\sigma^x$ flips a spin between rows; the bond $\sigma^z\sigma^z$ couples neighbours within a row. The heart of the matter is the Jordan–Wigner transformation. It writes each Pauli operator in terms of spinless fermions, with a nonlocal string that supplies the signs fermions need.

c_j = \Big(\prod_{l<j}\sigma^x_l\Big)\,\frac{\sigma^z_j + i\sigma^y_j}{2}, \qquad \{c_i, c_j^\dagger\} = \delta_{ij}, \quad \{c_i, c_j\} = 0 \qquad (3)

where: $c_j, c_j^\dagger$ — spinless fermion operators; $\prod_{l<j}\sigma^x_l$ — the nonlocal string. The transverse field becomes simply the local fermion number, $\sigma^x_j = 1 - 2c_j^\dagger c_j$ , and the bond becomes a local bilinear, $\sigma^z_j\sigma^z_{j+1} = (c_j^\dagger - c_j)(c_{j+1}^\dagger + c_{j+1})$ — the strings on adjacent sites cancel, which is exactly why the four-spin-looking object is quadratic. Fourier transforming and Bogoliubov-rotating gives the single-particle dispersion. In the isotropic 2D form it is the $\cosh$ relation; for the transverse-field chain it is the compact $2\sqrt{\varepsilon_k^2 + \Delta_k^2}$ .

\gamma_k = 2\sqrt{\varepsilon_k^2 + \Delta_k^2}, \quad \varepsilon_k = h - J\cos k, \quad \Delta_k = J\sin k \qquad\Longrightarrow\qquad \gamma_0 = 2|h - J| \qquad (4)

where: $\gamma_k \ge 0$ — quasiparticle energy at momentum $k$ ; $\varepsilon_k$ — "normal" part; $\Delta_k$ — pairing (anomalous) part removed by Bogoliubov. Since $\varepsilon_k^2 + \Delta_k^2 = h^2 + J^2 - 2hJ\cos k$ , the gap sits at $k = 0$ and equals $2|h - J|$ : it closes exactly at $h = J$ , the critical point of the chain, and at that point the dispersion becomes linear, $\gamma_k \approx 2J|k|$ — a massless relativistic mode. The isotropic 2D critical condition and the linear gap closing give $\nu = 1$ directly, since $\xi = 1/\gamma_0 \sim |T - T_c|^{-1}$ .

\sinh 2K_1\,\sinh 2K_2 = 1 \;\xrightarrow{K_1=K_2}\; \sinh 2K_c = 1, \quad K_c = \tfrac12\ln(1+\sqrt2) \approx 0.44069; \qquad \gamma_0 \sim |T - T_c| \Rightarrow \nu = 1 \qquad (5)

where: $K_c$ — critical coupling of the isotropic model; $\nu = 1$ — correlation-length exponent. The gap is the inverse correlation length, so the rate at which it closes sets $\nu$ ; here it is strictly linear. Finally, the algebraic face: the Onsager algebra, generated by the two transfer-matrix seeds through the Dolan–Grady closure conditions.

[A_l, A_m] = 4G_{l-m}, \quad [G_m, A_l] = 2A_{l+m} - 2A_{l-m}, \quad [G_l, G_m] = 0 \qquad (6)

where: $A_m, G_m$ — generators built from nested commutators of the seeds $A_0 = \sum_j\sigma^x_j$ and $A_1 = \sum_j\sigma^z_j\sigma^z_{j+1}$ ; $l, m \in \mathbb{Z}$ . The algebra closes iff the seeds satisfy the two Dolan–Grady relations, and its closure furnishes infinitely many commuting charges $Q_n$ with $[H, Q_n] = 0$ — the operational definition of an integrable model.

Derivation AJordan–Wigner: turning spins into free fermions

The goal is to show that the transverse-field Ising chain — the effective Hamiltonian of the 2D Ising transfer matrix — is a quadratic fermion problem, hence exactly solvable. The strategy: replace each Pauli operator with a Jordan–Wigner fermion, verify that the nonlocal string produces genuine anticommutation, and check that both the field and bond terms come out bilinear. The subtlety to watch: spins on different sites commute while fermions must anticommute, so a naive local substitution fails — the string is not optional decoration but the essential ingredient.

Step 1. Map the transverse field. Set $\sigma^x_j = 1 - 2c_j^\dagger c_j$ , so the field term $\sum_j\sigma^x_j$ becomes $\sum_j(1 - 2c_j^\dagger c_j)$ — diagonal in fermion number, manifestly quadratic. Why: the transverse field just counts flipped spins, which the number operator does directly; no string is needed for a single-site operator.

Step 2. Verify anticommutation. Define $c_j$ with the string $\prod_{l<j}\sigma^x_l = \prod_{l<j}(1 - 2c_l^\dagger c_l)$ ; the string contributes a $-1$ for each occupied site to the left, converting the commutation of spins into the anticommutation $\{c_i, c_j^\dagger\} = \delta_{ij}$ , $\{c_i, c_j\} = 0$ . Why: without the string the mapped operators would commute on different sites — not fermions at all; the string is exactly the sign bookkeeping that fixes this.

Step 3. Map the bond and confirm bilinearity. The product $\sigma^z_j\sigma^z_{j+1} = (c_j^\dagger - c_j)(c_{j+1}^\dagger + c_{j+1})$ — the strings on neighbouring sites overlap and cancel, leaving a local quadratic form. Why: this is the miracle of the nearest-neighbour Ising bond — an apparently string-laden object collapses to a clean bilinear because adjacent strings annihilate.

Step 4. Assemble the free-fermion Hamiltonian. Both terms are now bilinear, so the whole transfer matrix is $\exp$ of a quadratic fermion form — a free-fermion gas after Fourier and Bogoliubov (next derivation). Why: a quadratic fermion Hamiltonian is diagonalizable exactly, like coupled oscillators, which is the entire reason the 2D Ising model is solvable. *Verified: verify_aq-sm-free-fermions-onsager-algebra.py builds a two-site Jordan–Wigner representation numerically and confirms $\{c_1,c_1^\dagger\} = \{c_2,c_2^\dagger\} = I$ and $\{c_1,c_2\} = \{c_1,c_2^\dagger\} = 0$ , i.e. the string enforces genuine anticommutation.*

Go deeper · Derivation B · 1 minBogoliubov diagonalization and the gap that sets ν = 1

Derivation BBogoliubov diagonalization and the gap that sets ν = 1

The aim is to diagonalize the quadratic fermion Hamiltonian, extract the dispersion $\gamma_k$ , and read off the correlation-length exponent from how the gap closes. The strategy: Fourier-transform to decouple momenta into $\pm k$ pairs, use a Bogoliubov rotation to remove the particle–hole-mixing anomalous terms, and inspect the gap at $k = 0$ . Watch out for the anomalous $c_k^\dagger c_{-k}^\dagger$ terms — they are what make this a pairing problem, and a plain diagonalization without Bogoliubov would miss the physics entirely.

Step 1. Fourier transform, $c_j = N^{-1/2}\sum_k e^{ikj}c_k$ , which decouples the chain into independent momentum pairs $\pm k$ , each a $2\times2$ block with normal part $\varepsilon_k = h - J\cos k$ and pairing part $\Delta_k = J\sin k$ . Why: translation invariance means momentum is conserved, so only $\pm k$ couple, reducing an $N$ -body problem to $N/2$ two-level problems.

Step 2. Bogoliubov-rotate, $\eta_k = \cos\theta_k\, c_k - \sin\theta_k\, c_{-k}^\dagger$ with $\tan 2\theta_k = \Delta_k/\varepsilon_k$ , which removes the anomalous terms and yields free quasiparticles of energy $\gamma_k = 2\sqrt{\varepsilon_k^2 + \Delta_k^2}$ . Why: the anomalous terms mix particles and holes, and the Bogoliubov angle is precisely the rotation that diagonalizes each $2\times2$ pairing block.

Step 3. Evaluate the gap. Since $\varepsilon_k^2 + \Delta_k^2 = h^2 + J^2 - 2hJ\cos k$ , the minimum is at $k = 0$ : $\gamma_0 = 2|h - J|$ , which vanishes at $h = J$ , the critical point. At criticality $\gamma_k = 4J|\sin(k/2)| \approx 2J|k|$ — a linear, massless dispersion. Why: gaplessness is criticality, and the linear (relativistic) form at $k = 0$ is the signature of a conformal critical point with $c = 1/2$ .

Step 4. Read off $\nu$ . The gap is linear in the distance to criticality, $\gamma_0 \sim |T - T_c|$ , and the correlation length is its inverse, $\xi = 1/\gamma_0 \sim |T - T_c|^{-1}$ , so $\nu = 1$ . Why: the mass gap is the inverse correlation length, so how fast the gap closes directly sets $\nu$ — here, linearly, giving exactly 1. *Verified: verify_aq-sm-free-fermions-onsager-algebra.py confirms $\gamma_0 = 2|h-J|$ symbolically, that $\gamma_k = 2\sqrt{h^2+J^2-2hJ\cos k}$ , that at $h = J$ the dispersion is linear ( $\gamma_k \approx 2Jk$ at small $k$ ), and that the gap being linear in the detuning gives $\nu = 1$ .*

Limiting cases

Case	Result	Meaning
$h \gg J$ (or $T \gg T_c$ )	large gap, $\gamma_0 = 2(h-J)$	disordered; correlations die fast
$h = J$ ( $T = T_c$ )	$\gamma_0 = 0$ , $\gamma_k \approx 2J\|k\|$	gapless massless Majorana fermion; criticality
$h \ll J$ ( $T \ll T_c$ )	large gap again, $\gamma_0 = 2(J-h)$	ordered; the two ground states are the aligned states
1D transverse-field chain	strictly free fermions	the exact quantum analogue, solvable in one line
isotropic 2D	$\sinh 2K_c = 1$	the classic Onsager critical point

Common trap“Jordan–Wigner is just a local relabelling of spins as fermions”

The transformation is not local, and the non-locality is the whole point. Each fermion $c_j$ carries a string $\prod_{l<j}\sigma^x_l$ stretching back to the start of the chain. Drop the string and the mapped operators commute on different sites — you get bosons, or nonsense, not fermions. The string is what encodes the fermionic minus sign, and it is only because adjacent strings cancel in the nearest-neighbour bond $\sigma^z_j\sigma^z_{j+1}$ that the Hamiltonian comes out local and quadratic. For longer-range interactions the strings do not cancel and the fermion problem is no longer free — which is exactly why the nearest-neighbour Ising model is special and a next-nearest-neighbour version generally is not.

Worked examples

1 · The gap of the transverse-field Ising chain (confidence builder)

For the transverse-field Ising chain with $J = 1$ and field $h = 0.6$ (ordered side, $h < J$ ), compute the quasiparticle gap $\gamma_0$ and confirm the gap lives at $k = 0$ .

Step 1. Gap at $k = 0$ : $\gamma_0 = 2|h - J| = 2|0.6 - 1| = 2(0.4) = \mathbf{0.8}$ .

Step 2. Check it is the minimum. The dispersion is $\gamma_k = 2\sqrt{(0.6 - \cos k)^2 + \sin^2 k} = 2\sqrt{1.36 - 1.2\cos k}$ . This is smallest when $\cos k$ is largest, i.e. $k = 0$ : $\gamma_0 = 2\sqrt{1.36 - 1.2} = 2\sqrt{0.16} = 2(0.4) = \mathbf{0.8}$ . ✓

Step 3. Interpret: a nonzero gap means the system is ordered (the field has not yet destroyed alignment) and correlations decay over a length $\xi = 1/\gamma_0 = 1.25$ sites. Push $h$ toward $1$ and the gap shrinks toward zero — the approach to criticality.

2 · The gap closing linearly gives ν = 1 (exam level)

Show that near the critical field $h = J$ the correlation length of the transverse-field Ising chain diverges as $|h - J|^{-1}$ , i.e. $\nu = 1$ .

Step 1. Write the detuning $\delta = h - J$ . The gap is $\gamma_0 = 2|h - J| = 2|\delta|$ — exactly linear in $\delta$ , with no higher-order correction at $k = 0$ .

Step 2. The correlation length is the inverse gap: $\xi = 1/\gamma_0 = 1/(2|\delta|)$ .

Step 3. So $\xi \propto |\delta|^{-1} = |h - J|^{-1}$ . Since $|h - J|$ is linear in $|T - T_c|$ (the field/temperature maps smoothly near criticality), $\xi \sim |T - T_c|^{-1}$ , which is the definition of $\nu = 1$ .

Step 4. Sanity check the number: halve $\delta$ from $0.4$ to $0.2$ and $\xi$ doubles from $1.25$ to $2.5$ — an exact inverse proportionality, not a fitted power. The exponent $\nu = 1$ is a rigorous consequence of the linear gap closing, itself a consequence of the free-fermion structure.

Origins · 1 minjordan-wigner transformation and how this was discovered

jordan-wigner transformation

1928 · spin-to-fermion mapping

The Jordan–Wigner transformation, introduced by Pascual Jordan and Eugene Wigner in 1928, represents the spin operators of a one-dimensional chain in terms of spinless fermions dressed by a nonlocal string of operators. Its genius is turning the awkward algebra of spins — which commute on different sites but anticommute-like on the same site — into the clean anticommutation of fermions, at the cost of a string that tracks parity along the chain. For nearest-neighbour models the strings on adjacent bonds cancel, leaving local, quadratic fermion Hamiltonians that are exactly solvable. This single trick underlies the exact solutions of the transverse-field Ising chain, the XY chain, and — through the transfer matrix — the two-dimensional classical Ising model itself. It remains a workhorse of condensed-matter physics, from the theory of one-dimensional quantum magnets to modern discussions of Majorana fermions and topological order.

History

1928

— Jordan and Wigner introduce the spin-to-fermion transformation, converting spin chains into fermion problems.

1944

— Onsager solves the 2D Ising model, discovering en route the infinite-dimensional algebra now bearing his name.

1949

— Kaufman recasts Onsager's solution using a spinor (Clifford-algebra) representation, greatly simplifying it.

1964

— Schultz, Mattis, and Lieb reformulate the 2D Ising solution explicitly as free fermions via Jordan–Wigner and Bogoliubov.

1982

— Dolan and Grady give the two closure conditions on the seed operators that generate the Onsager algebra.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 13 · Kardar, Statistical Physics of Fields, Ch. 6 · Sachdev, Quantum Phase Transitions, Ch. 4 · Schultz, Mattis & Lieb, Rev. Mod. Phys. 36, 856 (1964).

Provenance. Onsager, L. (1944). "Crystal Statistics I." Phys. Rev. 65, 117. A verifier should confirm: the Jordan–Wigner anticommutation relations (built explicitly on a two-site space); $\sigma^x_j = 1 - 2c_j^\dagger c_j$ ; the Bogoliubov dispersion $\gamma_k = 2\sqrt{\varepsilon_k^2 + \Delta_k^2}$ with $\varepsilon_k = h - J\cos k$ , $\Delta_k = J\sin k$ ; the gap $\gamma_0 = 2|h - J|$ , linear in the detuning ⇒ $\nu = 1$ ; the linear ( $\gamma_k \approx 2Jk$ ) critical dispersion; the isotropic critical condition $\sinh 2K_c = 1$ ; and the Dolan–Grady closure $[A_0,[A_0,[A_0,A_1]]] = 16[A_0,A_1]$ (checked numerically on a three-site chain).

Prerequisites. The Onsager solution of the 2D Ising model · the transfer matrix formalism · the one-dimensional Ising transfer matrix. Leads to. Order parameter and spontaneous magnetization · high- and low-temperature series expansions · domain walls and interfacial excitations.

Why does a two-dimensional classical model become a one-dimensional quantum chain?

Because of the transfer-matrix construction. To compute $Z$ , you build the lattice one row at a time: the transfer matrix $V$ encodes the Boltzmann weight of adding one row given the previous one, and $Z = \operatorname{Tr} V^M$ for $M$ rows. Writing $V = e^{-\tau H}$ defines an effective Hamiltonian $H$ acting on a single row — a one-dimensional quantum system of $N$ spins. The spatial direction along the row stays spatial; the perpendicular direction (row index) becomes discrete "imaginary time." So a $d$ -dimensional classical problem maps to a $(d-1)$ -dimensional quantum one. For 2D Ising, $H$ is the one-dimensional transverse-field Ising chain, which is exactly the free-fermion model discussed here.

What exactly does the Jordan–Wigner string do, and why can't we drop it?

The string $\prod_{l<j}\sigma^x_l$ attached to each fermion $c_j$ supplies a factor of $-1$ for every occupied site to its left. Spins on different sites commute, but fermions must anticommute; the string is precisely the operator that converts one into the other. Drop it, and the mapped operators would commute on different sites — they would be hard-core bosons or nothing coherent, not fermions, and the elegant free-fermion structure would evaporate. The string is nonlocal (it reaches back to the start of the chain), but for nearest-neighbour bonds the strings on adjacent sites overlap and cancel, leaving a local, quadratic Hamiltonian. That cancellation is the small miracle that makes the nearest-neighbour Ising model free.

Why is a quadratic fermion Hamiltonian "exactly solvable" but a quartic one not?

A Hamiltonian quadratic in fermion operators (bilinear, like $c_i^\dagger c_j$ and $c_i^\dagger c_j^\dagger$ ) can always be brought to the form $\sum_k \gamma_k \eta_k^\dagger \eta_k$ by a linear (Bogoliubov) transformation — just as a quadratic form in classical variables is diagonalized by an orthogonal rotation. The result is a set of independent modes, so the many-body spectrum is a simple sum of single-particle energies and all thermodynamics reduces to one-particle integrals. A quartic term ( $c^\dagger c^\dagger c c$ ) represents genuine interactions between quasiparticles; no linear transformation removes it, the modes are coupled, and the problem is generically as hard as any many-body problem. The 2D Ising model's gift is that it is quadratic.

How does the linearly-closing gap give ν = 1 so directly?

The gap $\gamma_0 = 2|h - J|$ is the energy of the cheapest quasiparticle, and it is also the inverse correlation length: $\xi = 1/\gamma_0$ . Near the critical field $h = J$ , the gap is exactly linear in the detuning, $\gamma_0 = 2|h - J|$ , with no curvature. Therefore $\xi = 1/(2|h-J|) \propto |h - J|^{-1}$ , and since $|h - J|$ is proportional to $|T - T_c|$ near criticality, $\xi \sim |T - T_c|^{-1}$ . The correlation-length exponent is defined by $\xi \sim |T - T_c|^{-\nu}$ , so reading off the power gives $\nu = 1$ — no fitting, just the linear gap. Had the gap closed as $|h-J|^2$ , we would have found $\nu = 2$ instead.

What is the Onsager algebra, in plain terms, and why does it signal integrability?

It is an infinite-dimensional Lie algebra generated by just two "seed" operators — the field term $A_0 = \sum_j\sigma^x_j$ and the bond term $A_1 = \sum_j\sigma^z_j\sigma^z_{j+1}$ . Taking repeated commutators of these two produces an infinite tower $A_m, G_m$ whose commutation relations close (they only ever produce more $A$ 's and $G$ 's, never anything new). The seeds satisfy the Dolan–Grady conditions, e.g. $[A_0,[A_0,[A_0,A_1]]] = 16[A_0,A_1]$ , which are exactly the requirement for closure. Closure furnishes infinitely many operators $Q_n$ commuting with $H = \alpha A_0 + \beta A_1$ — an infinite set of conserved charges. A model with as many independent conservation laws as degrees of freedom is, by definition, integrable, which is the abstract reason the transfer-matrix eigenvalues can be found exactly.

Is the free-fermion picture the same as the Onsager-algebra picture?

They are two descriptions of the same integrable structure, and either one solves the model. The free-fermion route (Jordan–Wigner plus Bogoliubov, made explicit by Schultz–Mattis–Lieb in 1964) diagonalizes the transfer matrix by turning it into a gas of non-interacting quasiparticles. Onsager's original 1944 route instead used the algebra generated by the two seeds to organize the spectrum without ever mentioning fermions. The two are connected: the Onsager algebra is isomorphic to a fixed-point subalgebra of the loop algebra of $\mathfrak{sl}_2$ , and its representations reproduce exactly the free-fermion spectrum. Modern treatments prefer the fermion picture because it is more concrete, but the algebra explains why the fermions are free — the two viewpoints illuminate each other.