Simplicity of Skew Laurent Rings

The skew Laurent ring $R=k[x,x^{-1};\sigma ]$ is the crossed product of $k$ by the infinite cyclic group acting through $\sigma$. Its elements are finite sums $\sum _{i=r}^s{a}_i{x}^i$ with $r\le s$ in $\mathbb{\mathbb{Z}}$, and the single rule $x^{i}a={\sigma }^i(a)x^i$ governs everything.

Lam's $(3.18)$, following a paper of D. A. Jordan, gives three equivalent conditions for $R$ to be simple. The clean one is condition (2): $k$ is $\sigma$-simple and $\sigma$ has infinite inner order — no power ${\sigma }^n$ with $n\ge 1$ is an inner automorphism of $k$. The technical one is condition (3), which weakens inner to *inner via a unit fixed by $\sigma$*, and is what the proof of simplicity actually consumes.

The equivalence (2) $\iff$ (3) is a pure statement about $(k,\sigma )$: if ${\sigma }^n$ is inner via a unit $b$, then the product $a=b\sigma (b)\cdots {\sigma }^{n-1}(b)$ is a $\sigma$-fixed unit inducing ${\sigma }^{n^2}$. Once condition (3) holds, a minimal-degree argument on $I\cap k[x;\sigma ]$ forces any nonzero ideal to contain $x^n$, which is a unit — so the ideal is everything.

Unlike the differential criterion $(3.15)$, this one needs no hypothesis on $\mathrm{char}k$. Nothing in the argument divides by an integer.

Let $k$ be a ring and $\sigma \in \mathrm{Aut}(k)$. The skew Laurent polynomial ring $R=k[x,x^{-1};\sigma ]$ is the free left $k$-module on the symbols $x^i$, $i\in \mathbb{\mathbb{Z}}$, with multiplication determined by

Equivalently $R$ is the crossed product $k{\ast }_{\sigma }\mathbb{\mathbb{Z}}$, or the skew group ring of $\mathbb{\mathbb{Z}}$ acting on $k$ through $\sigma$. It contains the skew polynomial ring $k[x;\sigma ]$ of Skew Polynomial Rings: Hilbert's Twist as the non-negatively graded part, and is its localisation at the powers of $x$.

Two features distinguish it from the differential case. First, $x$ is invertible, so no ideal can be built out of $x$ alone and the descending chain used for non-artinianness has to be built from $x+1$ instead. Second, the twisting is by an automorphism rather than a derivation, and the corresponding non-degeneracy condition is about powers of $\sigma$, not $\sigma$ itself.

The reason powers matter is visible immediately. If ${\sigma }^n$ is inner via a $\sigma$-fixed unit $a$, then $a^{-1}{x}^n$ is central in $R$, and a central non-unit generates a proper ideal. This is the exact analogue of the change of variable $t=x-c$ in the differential case.

The two formulations of $\sigma$-simplicity agree

Lam requires only $\sigma (𝔄)\subseteq 𝔄$, while much of the literature requires $\sigma (𝔄)=𝔄$. The two versions of $\sigma$-simplicity define the same class of rings, and it is worth seeing why before using either.

Why powers of $\sigma$, not just $\sigma$

Suppose ${\sigma }^n$ is inner via a unit $a$ that is fixed by $\sigma$. Then the Laurent monomial $z=a^{-1}{x}^n$ is central:

So $1+z$ is central. It is not a unit: $R$ is $\mathbb{\mathbb{Z}}$-graded, and multiplying $1+z$ by any $w$ with lowest graded component in degree $r$ and highest in degree $s$ gives something with a nonzero component in degree $r$ and another in degree $s+n$ — two distinct degrees, since $n\ge 1$. A product equal to $1$ has only one nonzero component, so no such $w$ exists. The central non-unit $1+z$ therefore generates a proper nonzero ideal.

From inner to $\sigma$-fixed inner

Condition (3) looks weaker than the negation of infinite inner order, because it demands the conjugating unit be $\sigma$-fixed. The bridge is a symmetrisation. If ${\sigma }^n(c)=bc{b}^{-1}$ for all $c$, then ${\sigma }^n(b)=bb{b}^{-1}=b$, and one checks that each ${\sigma }^i(b)$ induces the same automorphism ${\sigma }^n$. Two units inducing the same inner automorphism differ by a central factor, so ${\sigma }^i(b)=z_{i}b$ with $z_i\in Z(k)$, $z_0=z_n=1$.

The product $a=b\sigma (b)\cdots {\sigma }^{n-1}(b)=\left(\prod _{i=0}^{n-1}{z}_i\right)b^n$ is then $\sigma$-fixed, because $\sigma (a)=\left(\prod _{i=1}^n{z}_i\right)b^n$ and the two central products agree. And $a$ differs from $b^n$ by a central factor, so it induces $({\sigma }^n{)}^n={\sigma }^{n^2}$.

The skew Laurent proof reuses the differential template with two substitutions: grading replaces filtration, and conjugation replaces differentiation.

Move 4 is the one worth memorising outside this context: it is the multiplicative analogue of averaging over a group, and the same product appears in Hilbert's Theorem 90 and in descent arguments for Galois cohomology.

The honest summary: the criterion is a dial. Turning the inner order from finite to infinite moves the ring from the classical world of central simple algebras and finite modules over a centre into the world of simple rings with no chain conditions, and applications sit on both sides of that dial.

• Laurent or polynomial? Inverting $x$ is what makes the ideal $x^{n}R$ disappear and simplicity possible. $k[x;\sigma ]$ is essentially never simple, because $xk[x;\sigma ]$ is a proper ideal. If your operator is invertible — a shift, a rotation, a Galois twist — use the Laurent ring.
• Field or polynomial coefficients? $F[t]$ is not $\sigma$-simple for $\sigma (t)=qt$, since $(t)$ is stable, so simplicity requires localising to $F(t)$. This is the analogue of the choice between $\mathbb{\mathbb{Q}}[y]$ and $\mathbb{\mathbb{Q}}(y)$ in the differential setting, but here it is forced rather than optional.
• Which power to test. Checking that $\sigma$ is outer is not enough. Budget for testing ${\sigma }^m$ for all $m\ge 1$; for commutative $k$ this reduces to computing the order of $\sigma$, which is usually easy, and for central simple $k$ it reduces to a Skolem–Noether computation.
• Generic versus root of unity. In applications with a parameter $q$, decide early which regime you are in. Generic $q$ gives a simple ring with infinite-dimensional representations only; $q$ a root of unity gives a finite module over a large centre and a completely different representation theory.

• Arithmetic is cheaper than in the differential case: multiplying $x^i$ past a coefficient applies ${\sigma }^i$ and produces one term, not $i+1$. A product of two elements with $m$ and $n$ terms costs $O(mn)$ coefficient operations plus $O(mn)$ applications of powers of $\sigma$.
• When $k$ is a field, $k[x;\sigma ]$ is a left and right euclidean domain and $k[x,x^{-1};\sigma ]$ is its localisation, so gcds, factorisations and Smith-form computations are all available.
• Deciding the inner order is the hard step. For $k$ commutative it is the order of $\sigma$ in $\mathrm{Aut}(k)$; for $k$ a central simple algebra it is decidable by Skolem–Noether; for a general noncommutative $k$ there is no uniform algorithm.
• Testing $\sigma$-simplicity is decidable for finite-dimensional $k$ by enumerating $\sigma$-stable ideals through linear algebra on the ideal lattice, and undecidable in general.
• Ore-algebra libraries — Sage's ore_algebra, Maple's OreTools, Mathematica's HolonomicFunctions — implement the difference and $q$-difference specialisations directly, and their termination arguments rely on the euclidean structure rather than on simplicity.

• Do not use $Rx⊋R{x}^2⊋\cdots$ to prove non-artinianness here — $x$ is a unit, so those left ideals are all equal to $R$. The correct chain uses $x+1$.
• Do not assume the two definitions of $\sigma$-ideal differ. They give the same notion of $\sigma$-simplicity, as the Remark in Core Concepts shows, but the proof of the equivalence uses that $\sigma$ is bijective and does not extend to endomorphisms.
• Do not expect the exponent produced by the symmetrisation to be optimal. The construction gives a $\sigma$-fixed unit inducing ${\sigma }^{n^2}$, not ${\sigma }^n$; the theorem only needs some positive exponent, so no sharper bound is required.
• Do not carry the $\mathbb{\mathbb{Q}}$-algebra reflex over from the differential criterion. $(3.18)$ is characteristic-free, and imposing an unnecessary hypothesis will exclude the finite-field cases used in coding theory.