Finding a Formula for 80 Years — Only to Prove It Does Not Exist

— The Ising model in a magnetic field and a victory of "impossibility"

> Paper: *Galois conjugation, droplet growth, and the differential transcendence of the Ising model in nonzero field* > Authors: Ran Luo, Jiahua Tian (arXiv:2611.08033, 2026-11-05, cond-mat.stat-mech)

I. A Model Explainable in a Lounge Chair

Rarely does physics produce something a middle-schooler can understand in its rules, yet which captivated Nobel-caliber masters for half a century. The Ising model is one such thing.

Imagine a chessboard where each square holds a tiny magnetic needle with only two states — up (+1) or down (−1). There are only two rules: neighboring spins prefer to align (coupling J), and an external field prefers them to point up (field strength h). The higher the temperature, the more thermal noise, and the more the spins disobey. Question: what macroscopic behavior does the whole board exhibit?

It's that simple. In 1925, Ising's master's thesis proved the one-dimensional version has "no phase transition"; for nearly a century since, this model has been statistical physics's E=mc² — known and revered by all.

II. The 1944 Miracle

In 1944, Onsager did something that still deserves a standing ovation: he solved the two-dimensional square-lattice Ising model exactly at zero magnetic field. Partition function, free energy — all written as a finite combination of elementary and elliptic functions, a closed formula that looks like "the hand of God." In 1952, C. N. Yang completed the final piece: the exact expression for the spontaneous magnetization. The zero-field 2D Ising model became the pinnacle of exact many-body solutions and later the touchstone of the entire theory of phase transitions.

But note the premise: zero field.

Once the field H is switched on, the spin-flip symmetry is broken and the mathematical structure changes beyond recognition. Eighty years passed. Onsager tried, Yang tried, countless brilliant minds tried — no closed formula was ever found for the 2D Ising model at nonzero field. Textbooks gloss over it; review articles list it as "a famous unsolved problem." It lay quietly at the back of every statistical physics student's homework: a puzzle whose rules fit in one sentence, yet no one could solve.

III. 2026: The Mystery Resolved in an Unexpected Way

In October and November 2026, Luo and Tian published two papers, playing an elegant "two-step move":

Step one (arXiv:2610.20569): In all dimensions d≥2, except for a countable exceptional set (field values corresponding to algebraic-integer fugacities), the free energy satisfies no algebraic differential equation — not even locally, on any small interval.

Step two (arXiv:2611.08033, this paper): The exceptional set is closed off to nothing — on the square lattice, at every nonzero field H≠0, neither the free energy nor the magnetization satisfies any algebraic differential equation in the coupling constant K. The result also extends to Lieb lattices, decorated lattices, and conjugate-related field values in d≥2.

This statement is worth translating into plain language. An "algebraic differential equation" is a polynomial equation in f, f′, f″, …. It is one of the broadest mathematical formalizations of a "closed formula": elementary functions, hypergeometric functions, elliptic functions, Painlevé transcendents — the entire family of "nice" functions to which Onsager's formula belongs — all satisfy some algebraic differential relation. The theorem's corollary is therefore thunderous:

> An Onsager-type formula at nonzero field cannot exist. It is not that we failed to find it — it is not there.

IV. The Proof, in Three Acts

The beauty of this proof is that every step can be told as an intuitive picture. The curtain rises —

Act One: Algebraic "clones" must escape

For fixed nonzero field, introduce the fugacity z = e^(−2H) ∈ (0,1). Suppose z is algebraic (most field values are); its minimal polynomial P(x) has several roots — the Galois conjugates of z, which can be understood as z's "clones" in the algebraic world.

Vieta's formulas say the product of all conjugate roots is a (nonzero) integer. But z itself hides in (0,1), with modulus less than 1 — so some clone β must be pushed outside the unit circle, |β| > 1. This is an observation almost joke-level simple, yet it is the engine of the whole play: if the free energy series satisfies an algebraic differential equation at z, applying a field embedding σ to the coefficients one by one yields an equation also satisfied by the "clone-version" series Φ_β. The clones obey the rules too.

Act Two: Droplets crush exponentially

Now the low-temperature expansion enters. Place a small r×r flipped-spin region (a "droplet") in the paramagnetic sea; its contribution to the partition function is z^(r²) · u^(2r) — an area penalty z^(r²) and a perimeter penalty u^(2r). In the clone world with |β|>1, this term's coefficients grow like |β|^(r²), i.e., the log of the coefficients is quadratic in the index m.

On the other side, differential algebra has a classic ceiling (a Maillet-type theorem): any function satisfying an algebraic differential equation has Taylor coefficients whose logs grow at most as O(m log m) — "factorial-grade." Quadratic crushes linear, and the gap grows unboundedly with m. The droplets' exponential temper cannot be contained by any "well-behaved" function.

Act Three: No cancellation, and no help from friends

Skeptics have two objections. First: could phases cancel and contributions exactly offset? Answer: the dominant contribution comes from "near-square" droplets (boxes of (r+j)×(r−j) with Young-diagram corners removed), whose total amplitude is a theta function divided by a fourth-power product — Jacobi's triple product gives a product formula with no zeros inside |t|<1. The leading term has nowhere to hide.

Second: could droplets with holes, or multi-droplet Mayer clusters, rewrite the story? Answer: the area penalty is exponential, while the combinatorial gain from hole arrangements is only polynomial entropy — q^(−m/4) easily crushes polynomials. The supporting cast cannot overturn the plot.

The final hammer: the pressure is jointly real-analytic in (K,H) (K>0, H≠0), so any differential-algebraic relation holding even on a small interval analytically continues to the whole low-temperature region — where the series converges and the coefficient-growth argument applies on the spot. Everywhere, no exceptions.

V. Why Zero Field Is Heaven, Peculiarly

Looking back at zero field: z = 1 is a root of unity, so the "clone escape" mechanism fails; Lee-Yang theory tells us the zeros lie exactly on the physical fugacity circle, and droplet area no longer dominates the weights. So at zero field the free energy is indeed differentially algebraic — Onsager's formula lives on in broad daylight, and Yang's magnetization formula remains intact. The theorem negates not these, but the expectation that "the same kind of formula still exists once the field is turned on."

A thought-provoking boundary: humanity currently knows closed solutions at only two fugacity values — z = 1 and z = −1. What about other roots of unity? This is the suspense the theorem leaves to the next generation.

VI. Deeper Echoes

Viewed from a higher vantage, this work is interesting on three levels.

First, "impossibility" is knowledge. For 80 years, countless people suspected they simply weren't smart enough. This time, mathematics has spoken: you were not lacking in cleverness — the target does not exist. A negative result collectively rescues researchers from a dead end — like the falsification of the "room-temperature ambient-pressure" route in superconductivity, it is a liberation.

Second, the weapon came from an unexpected direction. The verdict used neither functional analysis nor probability theory, but differential algebra — the branch founded by Ritt and Kolchin that carries Galois-theoretic ideas over to differential equations. Algebra's notion of "conjugation" found an unprecedented application in the series of statistical physics. Interdisciplinary science is not a slogan; it is the migration of tools.

Third, the marriage of picture and machine. The heart of the proof is pure statistical-mechanics imagery (droplets, area penalties, entropy penalties), while the finishing blow is pure algebra (Galois conjugates + coefficient-growth theorems). No one expected them to interlock so seamlessly — good proofs often look like this: tectonic plates from different continents colliding one day in perfect fit.

Feynman said the physicist's job is "to find laws in the face of admitted ignorance." This time, the law was found — only its shape is a prohibition. The next time you hear in class that "the Ising model at nonzero field has no exact solution," read it as a precise mathematical theorem: there, no Onsager-style formula can possibly exist. An 80-year cold case, closed in 2026.

---

Further reading

1. Luo, Tian, *Galois conjugation, droplet growth, and the differential transcendence of the Ising model in nonzero field*, arXiv:2611.08033 (2026) 2. Luo, Tian, *Droplets obstruct Onsager-type solutions of the Ising model*, arXiv:2610.20569 (2026, precursor paper) 3. Onsager, *Crystal Statistics I: A Two-Dimensional Model with an Order-Disorder Transition*, Phys. Rev. 65, 117 (1944) 4. Yang, *The Spontaneous Magnetization of a Two-Dimensional Ising Model*, Phys. Rev. 85, 808 (1952) 5. Okounkov, *The Ising model in our dimension and our times* (mathematical survey of the Ising model)

*This article is piece 1 of Round 1, Season 5 of the "Zhichai: Mathematical Foundations and Frontier Intersections" series.*