betaFeedback

WorksMathematics

Inverse problem in difference Galois theory

Abstract

Let CC be an algebraically closed field of characteristic zero, and let K=C(x)K=C(x) with σ(x)=x+1σ(x)=x+1. We show that every closed subgroup G\subseteq\GL_n(C) with cyclic component group G/G∘G/G^\circ can be realized as the \PV{} group of a system σY=A(x)YσY=A(x)Y. Here G∘G^\circ denotes the identity component of GG. Moreover, we can choose the coefficient matrix to have the formal expansion A(x)=h+O(x−2)A(x)=h+O(x^{-2}) at infinity, where h∈G(C)h\in G(C) is semisimple. This gives an affirmative answer to the conjecture of van der Put and Singer in their monograph Galois Theory of Difference Equations (Lecture Notes in Mathematics, vol.~1666). The original results in this paper were entirely generated by ChatGPT 6 and the MechMath Agent Team (MMAT) developed by the Key Laboratory of Mathematics Mechanization, AMSS, CAS. The author reviewed and simplified the initial proofs and verified the correctness of the results.

Introduction

The inverse problem is one of the fundamental problems in Galois theory. It asks which groups can occur as Galois groups over a fixed base field. For linear differential and difference equations, Picard–Vessiot theory associates a linear algebraic group with each system. The corresponding inverse problem is to realize prescribed linear algebraic groups by such systems. The differential inverse problem over the rational function field C(x)C(x) has been completely solved. For the shift operator σ(x)=x+1\sigma(x)=x+1, however, the corresponding inverse problem had remained open in general, although positive results were known for several classes of groups.

The differential inverse problem

The differential inverse problem. In his address to the 1966 International Congress of Mathematicians in Moscow, Kolchin [5] called attention to the differential inverse problem: given a linear algebraic group, find a linear differential equation over a prescribed differential field having that group as its Galois group.

Early algebraic results of Białynicki-Birula and Kovacic treated connected nilpotent and connected solvable groups, respectively. Kovacic also developed reductions toward the reductive case. These developments and their relation to later constructions are reviewed in [7 Introduction]. Over C(x)C(x), with derivation d/dx\mathrm{d}/\mathrm{d}x, Tretkoff and Tretkoff [12] proved in 1979 that every complex linear algebraic group is a differential Galois group. Their analytic method uses the Riemann–Hilbert problem to realize a finitely generated Zariski-dense subgroup as monodromy. The density theorem for regular singular equations then identifies the resulting differential Galois group.

For an arbitrary algebraically closed constant field CC of characteristic zero, Mitschi and Singer [7] gave a constructive, purely algebraic realization of every connected linear algebraic group. Their theorem applies more generally to differential fields of finite, nonzero transcendence degree over CC with constant field CC. They subsequently treated solvable-by-finite groups over C(x)C(x) [8]. Hartmann [4] completed the inverse problem over C(x)C(x) for arbitrary linear algebraic groups, using algebraic methods.

More recent work has strengthened this existence theorem in several directions. Feng and Wibmer [3] established specialization results for differential Galois groups and proved Matzat’s conjecture. This conjecture states that the absolute differential Galois group of a one-variable function field FF over C\mathbb{C}, equipped with a nonzero C\mathbb{C}-derivation, is a free proalgebraic group on a set of cardinality ∣F∣|F|. Together with [1 Corollary 3.8(b)], this result implies that, for every nontrivial linear algebraic group GG over C\mathbb{C}, there are exactly ∣C∣|\mathbb{C}| differential C(x)\mathbb{C}(x)-isomorphism classes of Picard–Vessiot extensions L/C(x)L/\mathbb{C}(x) with differential Galois group isomorphic to GG. Serafini and Wibmer [11] also proved that every linear algebraic group over C\mathbb{C} can be realized as the differential Galois group of a regular singular differential equation over C(x)\mathbb{C}(x).

The difference inverse problem

For the difference field (C(x),σ)(\mathbb{C}(x),\sigma), where C\mathbb{C} is algebraically closed of characteristic zero and σ(x)=x+1\sigma(x)=x+1, van der Put and Singer [9 Proposition 1.20] proved that G/G∘G/G^{\circ} is necessarily cyclic whenever GG is a Picard–Vessiot group. They stated the converse as a conjecture in Chapter 3, page 35 of [9]: every linear algebraic group GG for which G/G∘G/G^{\circ} is cyclic should occur as a difference Galois group over C(x)\mathbb{C}(x).

In their monograph [9], van der Put and Singer laid the foundations for solving the difference inverse problem. In particular, they introduced the coefficient group TAT_A associated with a system σY=AY\sigma Y=AY (see Definition 2.1). They showed that the difference Galois group of this system is conjugate to a subgroup of TAT_A [9 Proposition 3.2(2)].1 Furthermore, they gave a criterion, in terms of rational changes of variables, for the coefficient group TAT_A to be the difference Galois group [9 Proposition 1.21]. Their algebraic constructions realize Gm\mathbb{G}_m and tori over C\mathbb{C} as difference Galois groups over C(x)\mathbb{C}(x) [9 Lemma 3.4(2)–(3)]. Over C(x)\mathbb{C}(x), the analytic theory developed in Chapter 8 of [9] also realizes every connected linear algebraic group and every cyclic extension of a torus as a difference Galois group. The respective results are Corollary 8.6(3) and Lemma 8.12 of [9].

Feng [2] subsequently developed specialization results for difference Galois groups. These results transfer the above analytic realizations over C(x)\mathbb{C}(x) to the corresponding realizations over C(x)\mathbb{C}(x) for an arbitrary algebraically closed field C\mathbb{C} of characteristic zero [2 Theorem 6.3 and Corollary 6.4]. Thus every connected linear algebraic group over C\mathbb{C}, and every cyclic extension of a torus over C\mathbb{C}, can be realized as a difference Galois group over C(x)\mathbb{C}(x).

There is an analogous inverse Galois problem for the qq-shift σq(f(x))=f(qx)\sigma_q(f(x))=f(qx). We do not discuss this problem in detail here and refer interested readers to Chapter 12 of the monograph of van der Put and Singer [9] and the article by Ramis and Sauloy [10].

Main result and conventions

Throughout this paper, C\mathbb{C} is algebraically closed of characteristic zero, K=C(x)K=\mathbb{C}(x), and σ(f(x))=f(x+1)\sigma(f(x))=f(x+1). We identify the prime field with Q⊂C\mathbb{Q}\subset\mathbb{C}. The field of constants of KK, denoted by KσK^\sigma, is C\mathbb{C}. By an algebraic group we mean a reduced closed subgroup of a general linear group over C\mathbb{C}. We write G∘G^{\circ} for the connected component of GG containing the identity. Such groups and their closed subgroups are smooth in characteristic zero. A rational representation of GG is a regular homomorphism of algebraic groups ρ:G→GL(V)\rho:G\to\mathrm{GL}(V) on a finite-dimensional C\mathbb{C}-vector space; it is not a partially defined rational map. We use the same symbol for the induced map on points over any commutative C\mathbb{C}-algebra. The main result of this paper is the following theorem.

Theorem 1.1. Let G⊂GLn(C)G\subset\mathrm{GL}_n(\mathbb{C}), n≥1n\ge1, be a closed algebraic subgroup. Assume that G/G∘G/G^{\circ} is cyclic. There exist A(x)∈G(K)A(x)\in G(K), a Picard–Vessiot ring RR for σY=AY\sigma Y=AY, and a fundamental matrix Z∈G(R)Z\in G(R) such that

Aut⁡σ(R/K)⟶GLn(C),τ⟼Z−1τ(Z)\operatorname{Aut}_{\sigma}(R/K)\longrightarrow\mathrm{GL}_n(\mathbb{C}),\qquad\tau\longmapsto Z^{-1}\tau(Z)

has image exactly G(C)G(\mathbb{C}).

The theorem establishes the sufficient direction of van der Put and Singer’s conjecture. As noted in § 1.2, their monograph already provides many of the tools needed to solve the inverse problem. In particular, they proved that whenever G/G∘G/G^{\circ} is cyclic, there exists A∈G(K)A \in G(K) with TA=GT_A = G [9 Proposition 3.3(2)]. Our main contribution is to construct such an AA satisfying A(x)=h+O(x−2)A(x) = h + O(x^{-2}), with hh semisimple, and the singularity condition in (3), and to establish a rigidity result for rational difference-invariant lines under these conditions. This rigidity, together with TA=GT_A = G, rules out any rational change of variables U=BYU = BY, with B∈G(K)B \in G(K), for which the transformed coefficient matrix σ(B)AB−1\sigma(B)AB^{-1} belongs to H(K)H(K) for some proper closed algebraic subgroup H⊊GH \subsetneq G.

Contributions

The original proofs2 in this paper were generated entirely by ChatGPT 6 and the MechMath Agent Team (MMAT), after the author supplied the necessary reference materials. The original proofs contain two main new ingredients. The first is the construction of a coefficient matrix A(x)A(x) satisfying A(x)=h+O(x−2)A(x) = h + O(x^{-2}) at infinity, where hh is a semisimple constant matrix, together with the required singularity condition. The second is a rigidity theorem for rational difference-invariant lines (see Theorem 2.5.

Using these ingredients, the original AI-generated proof used descent over a Picard-Vessiot ring. Suppose that the difference Galois group HH of σY=AY\sigma Y = AY is realized as a proper closed subgroup of TA=GT_A = G. By Chevalley’s theorem, there exist an algebraic representation ρ:G→GL⁡(V)\rho: G \to\operatorname{GL}(V) and a line ℓ⊂V\ell\subset V such that H=Stab⁡G(ℓ)H = \operatorname{Stab}_G(\ell). Let RR be a Picard-Vessiot ring for the system, let Q(R)Q(R) be its total ring of fractions, and choose a fundamental matrix Z∈G(R)Z \in G(R) realizing this identification of the Galois group. Descent produces a rational difference-invariant line W⊂VK=K⊗CVW \subset V_K = K \otimes_C V satisfying

Q(R)W=Q(R)ρ(Z)ℓ.Q(R)W = Q(R)\rho(Z)\ell.

The rigidity theorem implies that W=Kℓ0W = K\ell_0 for a constant line ℓ0⊂V\ell_0 \subset V. Since A∈Stab⁡G(ℓ0)(K)A \in\operatorname{Stab}_G(\ell_0)(K) and TA=GT_A = G, the line ℓ0\ell_0 is GG-stable. Consequently, ρ(Z)−1\rho(Z)^{-1} preserves Q(R)ℓ0Q(R)\ell_0, and the preceding equality gives Q(R)ℓ=Q(R)ℓ0Q(R)\ell= Q(R)\ell_0. Thus ℓ=ℓ0\ell= \ell_0, contradicting Stab⁡G(ℓ)=H⊊G\operatorname{Stab}_G(\ell) = H \subsetneq G. This argument requires extending scalars to Q(R)Q(R), invoking properties of the Picard-Vessiot ring, and carrying out a separate descent step.

The author observed that the proof could be simplified by applying Proposition 1.21 of van der Put and Singer’s monograph directly, with all line constructions carried out over KK (see the proof of Theorem 2.8). Specifically, suppose that

σ(B)AB−1∈H(K)\sigma(B)AB^{-1} \in H(K)

for a proper closed subgroup H⊊GH \subsetneq G and some B∈G(K)B \in G(K). Choose a Chevalley representation ρ\rho and a line ℓ=Cw\ell= Cw with stabilizer HH. Then W=Kρ(B−1)wW = K\rho(B^{-1})w is rational difference-invariant. Rigidity gives W=Kℓ0W = K\ell_0 for a constant line ℓ0⊂V\ell_0 \subset V, and the identity

Stab⁡G(ℓ0)K=B−1HKB\operatorname{Stab}_G(\ell_0)_K = B^{-1}H_KB

shows that its stabilizer is proper. On the other hand, A∈Stab⁡G(ℓ0)(K)A \in\operatorname{Stab}_G(\ell_0)(K) forces this stabilizer to contain TA=GT_A = G, a contradiction. The Galois group is therefore GG by [9 Proposition 1.21(3)]. This simplification eliminates both the separate descent step and the subsequent use of ρ(Z)−1\rho(Z)^{-1}. Under the guidance of the author, the proof of Lemma 3.2 was also simplified by making greater use of existing results in Milne’s monograph.

Organization

In Section 2 we recall the coefficient group and the finite singularity set, study their behavior under algebraic representations, and prove the rigidity theorem for rational difference-invariant lines. We then combine rigidity with Chevalley’s theorem to exclude rational reduction to any proper subgroup and apply van der Put and Singer’s criterion to identify the Galois group. In Section 3 we construct a rational curve through generating points of the identity component, impose the required expansion at infinity, and use a semisimple lift and an arithmetic dilation to complete the proof of Theorem 1.1. In Section 4 we give explicit realizations of SL⁡2\operatorname{SL}_2, GmG_m, a nonsplit extension of a torus and finite cyclic groups, and use counterexamples to explain the roles of the hypotheses. In Section 5 we give an alternative construction of the generating curve, using a rational chart and Hermite interpolation to prescribe tangent vectors that span the Lie algebra of the identity component.

Rigidity of rational difference-invariant lines

A rational difference-invariant line need not have a constant generator. We prove that every such line in every algebraic representation of GG is the scalar extension of a C\mathbb{C}-line when A(x)=h+O(x−2)A(x)=h+O(x^{-2}) with hh semisimple and 1∉span⁡QP(A)1\notin\operatorname{span}_{\mathbb{Q}} P(A), where P(A)P(A) is the set of finite poles of AA and A−1A^{-1}; the proof uses valuations at finite points and a comparison of coefficients at infinity. We then use this rigidity to exclude rational reduction to a proper subgroup when TA=GT_A=G, and deduce that the difference Galois group is GG.

Finite singularities and change of representation

Definition 2.1. For A∈GL⁡n(K)A\in\operatorname{GL}_n(K), let

P(A)={s∈C:some entry of A or A−1 has a pole at s}.P(A)=\{s\in\mathbb{C}:\text{some entry of }A\text{ or }A^{-1}\text{ has a pole at }s\}.

Let TAT_A be the smallest closed C\mathbb{C}-subgroup of GL⁡n\operatorname{GL}_n for which A∈TA(K)A\in T_A(K). Equivalently, TAT_A is the algebraic subgroup generated by the values A(a)A(a) for a∈C∖P(A)a\in\mathbb{C}\setminus P(A).

The definition and the description of TAT_A by its generating values are given in [9 Proposition 3.2(1)]. Note that the poles of A−1A^{-1} are indispensable in the definition of P(A)P(A).

Lemma 2.2 (Functorial control of singularities and the first jet). Let A∈G(K)A\in G(K) and let ρ:G→GL⁡(V)\rho:G\to\operatorname{GL}(V) be an algebraic representation. Then

P(ρ(A))⊆P(A).P(\rho(A))\subseteq P(A).

If A(x)=h+O(x−2)A(x)=h+O(x^{-2}) with h∈G(C)h\in G(\mathbb{C}), then

ρ(A(x))=ρ(h)+O(x−2).\rho(A(x))=\rho(h)+O(x^{-2}).

If hh is semisimple, so is ρ(h)\rho(h).

Proof. For s∉P(A)s\notin P(A), put Os=C[x](x−s)\mathcal{O}_s=\mathbb{C}[x]_{(x-s)}. Both AA and A−1A^{-1} have entries in Os\mathcal{O}_s, so A∈GL⁡n(Os)A\in\operatorname{GL}_n(\mathcal{O}_s). The defining equations of GG vanish on AA in KK and hence in Os\mathcal{O}_s. It follows that A∈G(Os)A\in G(\mathcal{O}_s). Applying the morphism ρ\rho gives ρ(A)∈GL⁡(V)(Os)\rho(A)\in\operatorname{GL}(V)(\mathcal{O}_s). Thus both ρ(A)\rho(A) and its inverse are regular at ss.

At infinity, set u=x−1u=x^{-1}. Then AA and hh are G(C[[u]])G(\mathbb{C}[[u]])-points with the same reductions in G(C[[u]]/(u2))G(\mathbb{C}[[u]]/(u^2)). Applying ρ\rho preserves this congruence, proving the second assertion. The third assertion follows from the functoriality of Jordan decomposition. □\square

Lemma 2.3 (Local primitivity).

(i) Let O\mathcal{O} be a discrete valuation ring with fraction field FF and valuation ν\nu. For 0≠w=(wi)∈Fm0\ne w=(w_i)\in F^m, put ν(w)=min⁡iν(wi)\nu(w)=\min_i\nu(w_i), with ν(0)=+∞\nu(0)=+\infty. Then, for a∈F∗a\in F^* and B∈GL⁡m(O)B\in\operatorname{GL}_m(\mathcal{O}),

ν(aw)=ν(a)+ν(w),ν(Bw)=ν(w).\nu(aw)=\nu(a)+\nu(w),\qquad\nu(Bw)=\nu(w).

(ii) Every KK-line L⊂KmL\subset K^m has a primitive polynomial generator v∈C[x]mv\in\mathbb{C}[x]^m, meaning L=KvL=Kv and gcd⁡(v1,…,vm)=1\gcd(v_1,\ldots,v_m)=1. It is unique up to multiplication by an element of C∗\mathbb{C}^*. Writing νs=ord⁡x=s\nu_s=\operatorname{ord}_{x=s} and extending it to vectors as in (i), we have

νs(v)=νs(σv)=0for every s∈C.\nu_s(v)=\nu_s(\sigma v)=0\qquad\text{for every }s\in\mathbb{C}.

Proof. (i) The scalar identity follows directly from the definition. Since the entries of BB lie in O\mathcal{O},

ν(Bw)=min⁡iν(∑jBijwj)≥min⁡i,j(ν(Bij)+ν(wj))≥ν(w).(1)\nu(Bw)=\min_i\nu\left(\sum_j B_{ij}w_j\right)\ge\min_{i,j}\bigl(\nu(B_{ij})+\nu(w_j)\bigr)\ge\nu(w). \tag*{(1)}

Applying this inequality to B−1∈GLm(O)B^{-1} \in\mathrm{GL}_m(\mathcal{O}) gives

ν(w)=ν(B−1(Bw))≥ν(Bw)≥ν(w).\nu(w)=\nu\left(B^{-1}(Bw)\right)\geq\nu(Bw)\geq\nu(w).

(ii) Choose 0≠u∈L0 \neq u \in L and b∈C[x]∖{0}b \in\mathbb{C}[x]\setminus\{0\} such that bu∈C[x]mbu \in\mathbb{C}[x]^m. If gg is the monic greatest common divisor of the coordinates of bubu, then v=g−1buv=g^{-1}bu is a primitive polynomial generator of LL. Since C[x]\mathbb{C}[x] is a principal ideal domain, there exist a1,…,am∈C[x]a_1,\ldots,a_m \in\mathbb{C}[x] such that

∑i=1maivi=1,∑i=1mσ(ai)σ(vi)=1.\sum_{i=1}^{m}a_i v_i=1,\qquad\sum_{i=1}^{m}\sigma(a_i)\sigma(v_i)=1.

After localization at (x−s)(x-s), these identities show that the coordinates of each of vv and σv\sigma v generate Os=C[x](x−s)\mathcal{O}_s=\mathbb{C}[x]_{(x-s)}. Thus νs(v)=νs(σv)=0\nu_s(v)=\nu_s(\sigma v)=0.

Finally, if v~=fv\widetilde{v}=fv is another primitive polynomial generator, with f∈K∗f\in K^*, then, as fractional ideals of C[x]\mathbb{C}[x],

C[x]=∑iC[x]v~i=f∑iC[x]vi=fC[x].\mathbb{C}[x]=\sum_i\mathbb{C}[x]\widetilde{v}_i=f\sum_i\mathbb{C}[x]v_i=f\mathbb{C}[x].

Hence f∈C[x]∗=C∗f\in\mathbb{C}[x]^*=\mathbb{C}^*, proving uniqueness. □

Rational difference-invariant lines

Throughout this section, let VV be a finite-dimensional C\mathbb{C}-vector space equipped with an algebraic representation ρ:G→GL(V)\rho:G\to\mathrm{GL}(V). Put VK=K⊗CVV_K=K\otimes_{\mathbb{C}}V, and identify VV with 1⊗V1\otimes V. Extend σ\sigma to VKV_K by

σ(f⊗v)=σ(f)⊗v.\sigma(f\otimes v)=\sigma(f)\otimes v.

This defines a σ\sigma-semilinear automorphism whose fixed vectors are precisely the elements of VV, which we call constant vectors.

Definition 2.4 (Rational difference-invariant line). Let A∈G(K)A\in G(K) and let ρ:G→GL(V)\rho:G\to\mathrm{GL}(V) be an algebraic representation. On VKV_K, consider the σ\sigma-semilinear automorphism

Φρ=ρ(A)−1σ.\Phi_\rho=\rho(A)^{-1}\sigma.

A one-dimensional KK-subspace L⊂VKL\subset V_K is called a rational difference-invariant line if Φρ(L)=L\Phi_\rho(L)=L. Equivalently, for a nonzero generator vv of LL, there exists c∈K∗c\in K^* such that

σ(v)=cρ(A)v.(2)\sigma(v)=c\rho(A)v. \tag*{(2)}

Note that the condition in the above definition is independent of the chosen generator, because replacing vv by fvfv, where f∈K∗f\in K^*, replaces cc by cσ(f)/fc\sigma(f)/f.

Theorem 2.5 (Rigidity of rational difference-invariant lines). Assume that A∈G(K)A\in G(K) satisfies

A(x)=h+O(x−2),h∈G(C) semisimple,1∉span⁡QP(A).(3)A(x)=h+O(x^{-2}),\qquad h\in G(\mathbb{C})\ \text{semisimple},\qquad1\notin\operatorname{span}_{\mathbb{Q}}\mathcal{P}(A). \tag*{(3)}

For every finite-dimensional algebraic representation ρ\rho of GG, each rational difference-invariant KK-line in K⊗CVK\otimes_{\mathbb{C}}V is the scalar extension of a C\mathbb{C}-line.

Proof. Fix a C\mathbb{C}-basis of VV and identify VKV_K with KmK^m. Let L⊂VKL\subset V_K be a rational difference-invariant line. By Lemma 2.3(ii), it has a primitive polynomial generator v∈C[x]mv\in\mathbb{C}[x]^m satisfying

νs(v)=νs(σv)=0(s∈C).(4)\nu_s(v)=\nu_s(\sigma v)=0\qquad(s\in\mathbb{C}). \tag*{(4)}

By Definition 2.4, we have σ(v)=cρ(A)v\sigma(v)=c\rho(A)v for some c∈K∗c\in K^*. Put B=ρ(A)B=\rho(A) and P=P(A)\mathcal{P}=\mathcal{P}(A). For s∉Ps\notin\mathcal{P}, Lemma 2.2 gives B∈GLm(Os)B\in\mathrm{GL}_m(\mathcal{O}_s). Lemma 2.3(i) and (2.3) therefore yield

0=νs(σv)=νs(cBv)=νs(c)+νs(v)=νs(c).(5)0=\nu_s(\sigma v)=\nu_s(cBv)=\nu_s(c)+\nu_s(v)=\nu_s(c). \tag*{(5)}

Thus cc has no finite zero or pole outside P\mathcal{P}. Since C\mathbb{C} is algebraically closed, we can write

c(x)=γ∏s∈P(x−s)ks,γ∈C∗,ks∈Z.(6)c(x)=\gamma\prod_{s\in\mathcal{P}}(x-s)^{k_s},\qquad\gamma\in\mathbb{C}^*,\qquad k_s\in\mathbb{Z}. \tag*{(6)}

We now work with formal Laurent expansions at infinity, in C((x−1))C((x^{-1})). Let d=max⁡ideg⁡vi≥0d=\max_i\deg v_i\ge0 and write

v(x)=xd(vd+x−1vd−1+O(x−2)),vd≠0.v(x)=x^d\left(v_d+x^{-1}v_{d-1}+O(x^{-2})\right),\qquad v_d\ne0.

If d=0d=0, put v−1=0v_{-1}=0. Set e=∑skse=\sum_s k_s. Since c(x)=γxe(1+O(x−1))c(x)=\gamma x^e(1+O(x^{-1})) and ρ(A(x))=ρ(h)+O(x−2)\rho(A(x))=\rho(h)+O(x^{-2}), the highest powers of xx in the Laurent expansions of the two sides of (2.1) are respectively dd and d+ed+e. The leading coefficient on the right is γρ(h)vd≠0\gamma\rho(h)v_d\ne0. Therefore

e=∑sks=0.(7)e=\sum_s k_s=0. \tag*{(7)}

Let q=∑skssq=\sum_s k_s s and M=γρ(h)M=\gamma\rho(h). Factoring powers of xx out of (6) and expanding formally, also for negative ksk_s, gives

c(x)=γ(1−qx−1+O(x−2)).c(x)=\gamma\left(1-qx^{-1}+O(x^{-2})\right).

We consequently have

v(x+1)=xd(vd+x−1(dvd+vd−1)+O(x−2)),v(x+1)=x^d\left(v_d+x^{-1}(dv_d+v_{d-1})+O(x^{-2})\right),
cρ(A)v=xd(Mvd+x−1(Mvd−1−qMvd)+O(x−2)).c\rho(A)v=x^d\left(Mv_d+x^{-1}(Mv_{d-1}-qMv_d)+O(x^{-2})\right).

Comparing coefficients yields

Mvd=vd,(d+q)vd=(M−I)vd−1.(8)Mv_d=v_d,\qquad(d+q)v_d=(M-I)v_{d-1}. \tag*{(8)}

By Lemma 2.2, MM is semisimple. Its 1-eigenspace has zero intersection with im⁡(M−I)\operatorname{im}(M-I). The two sides of the second equality in (8) lie in these respective spaces. Since vd≠0v_d\ne0, we get

d+∑s∈Pkss=0.(9)d+\sum_{s\in\mathcal{P}} k_s s=0. \tag*{(9)}

If d>0d>0, division by the nonzero integer dd would put 11 in span⁡QP\operatorname{span}_{\mathbb{Q}}\mathcal{P}. Hence d=0d=0, and vv is a nonzero constant vector. □

Example (A rational difference-invariant line with no constant generator). The conclusion of Theorem 2.5 need not hold for a general coefficient matrix. Take G=GL2G=\mathrm{GL}_2 with its natural representation on V=C2V=\mathbb{C}^2, and put

A=(1101),v(x)=(x1),L=Kv⊂K⊗CV=K2.A=\begin{pmatrix}1&1\\0&1\end{pmatrix},\qquad v(x)=\begin{pmatrix}x\\1\end{pmatrix},\qquad L=Kv\subset K\otimes_{\mathbb{C}}V=K^2.

Then

σ(v)=(x+11)=Av,Φρ(v)=A−1σ(v)=v.\sigma(v)=\begin{pmatrix}x+1\\1\end{pmatrix}=Av,\qquad\Phi_\rho(v)=A^{-1}\sigma(v)=v.

Consequently, Φρ(fv)=σ(f)v\Phi_\rho(fv)=\sigma(f)v for every f∈Kf\in K, and the surjectivity of σ:K→K\sigma:K\to K gives Φρ(L)=L\Phi_\rho(L)=L. Nevertheless L∩V={0}L\cap V=\{0\}. Indeed, if fv=(a,b)Tfv=(a,b)^T with f∈Kf\in K and a,b∈Ca,b\in\mathbb{C}, comparison of coordinates gives f=bf=b and bx=abx=a. Since xx is transcendental over C\mathbb{C}, these equalities imply b=a=f=0b=a=f=0. If LL were the scalar extension K⊗CL0K\otimes_{\mathbb{C}}L_0 of a C\mathbb{C}-line L0⊂VL_0\subset V, any nonzero vector in L0L_0 would belong to L∩VL\cap V, a contradiction. Thus no rescaling of vv by an element of K∗K^* produces a constant vector.

Here P(A)=∅\mathcal{P}(A)=\varnothing and A(x)=h+O(x−2)A(x)=h+O(x^{-2}) with h=Ah=A, but hh is not semisimple, so the hypotheses of Theorem 2.5 are not satisfied. The Picard–Vessiot group of this same system is discussed in § 4.6.

Remark 2.6 (Uniformity and the exact conclusion). The hypotheses are imposed on AA once, independently of ρ\rho. In every representation the finite singularities lie in P(A)\mathcal{P}(A) and the first-order term at infinity vanishes. What becomes constant is the line, not necessarily cc; the integers ksk_s need not vanish separately. For a constant generator v0v_0, (2.1) says that ρ(A)\rho(A) preserves Kv0Kv_0, or A∈Stab⁡G(Cv0)(K)A\in\operatorname{Stab}_G(\mathbb{C}v_0)(K). It does not say that ρ(A)\rho(A) fixes v0v_0.

Remark 2.7 (A translation-invariant weakening). If PP is nonempty, the last condition in (3) may be replaced by

1∉span⁡Q{s−s0:s∈P},s0∈P.1 \notin\operatorname{span}_{\mathbb{Q}}\{s-s_0:s\in P\},\qquad s_0\in P.

Indeed, (7) gives ∑skss=∑sks(s−s0)\sum_s k_s s=\sum_s k_s(s-s_0). This condition is independent of s0s_0. Equivalently,

{∑skss:ks∈Z, ∑sks=0}∩Z={0}.\left\{\sum_s k_s s:k_s\in\mathbb{Z},\ \sum_s k_s=0\right\}\cap\mathbb{Z}=\{0\}.

One implication follows by clearing denominators, and the other by dividing by a nonzero integer. The stronger condition (3) is sufficient for the construction below.

Rational reduction and the coefficient-group criterion

The change of variables U=BYU=BY, with B∈G(K)B\in G(K), transforms σY=AY\sigma Y=AY into

σU=σ(B)AB−1U.\sigma U=\sigma(B)AB^{-1}U.

We shall show that, under the rigidity hypotheses, the transformed matrix cannot belong to a proper subgroup of G=TAG=T_A. The Galois group will then be determined by [9 Proposition 1.21(3)].

Theorem 2.8 (Coefficient-group criterion). Suppose that A∈G(K)A\in G(K) satisfies (3) and TA=GT_A=G. For every proper closed CC-subgroup H⊊GH\subsetneq G and every B∈G(K)B\in G(K), we have

σ(B)AB−1∉H(K).\sigma(B)AB^{-1}\notin H(K).

Consequently, the difference Galois group of σY=AY\sigma Y=AY is GG.

Proof. Suppose that D=σ(B)AB−1∈H(K)D=\sigma(B)AB^{-1}\in H(K) for some B∈G(K)B\in G(K) and some proper closed subgroup H⊊GH\subsetneq G. By Chevalley’s theorem [6 Theorem 4.27], there are an algebraic representation ρ:G→GL⁡(V)\rho:G\to\operatorname{GL}(V) and a line ℓ=Cw\ell=\mathbb{C}w such that H=Stab⁡G(ℓ)H=\operatorname{Stab}_G(\ell). Write ρ(D)w=aw\rho(D)w=aw with a∈K∗a\in K^*, and put v=ρ(B−1)wv=\rho(B^{-1})w. Since σ(B)−1=AB−1D−1\sigma(B)^{-1}=AB^{-1}D^{-1}, we have

σ(v)=ρ(σ(B)−1)w=ρ(A)ρ(B−1)ρ(D−1)w=a−1ρ(A)v.\sigma(v)=\rho(\sigma(B)^{-1})w=\rho(A)\rho(B^{-1})\rho(D^{-1})w=a^{-1}\rho(A)v.

Thus KvKv is a rational difference-invariant line. Theorem 2.5 gives Kv=Kℓ0Kv=K\ell_0 for a constant line ℓ0⊂V\ell_0\subset V.

Let H1=Stab⁡G(ℓ0)H_1=\operatorname{Stab}_G(\ell_0). For every commutative KK-algebra SS, the equality Kℓ0=ρ(B−1)(Kℓ)K\ell_0=\rho(B^{-1})(K\ell) gives Sℓ0=ρ(B−1)(Sℓ)S\ell_0=\rho(B^{-1})(S\ell). Thus, for g∈G(S)g\in G(S),

ρ(g)(Sℓ0)=Sℓ0⟺ρ(BgB−1)(Sℓ)=Sℓ.\rho(g)(S\ell_0)=S\ell_0\quad\Longleftrightarrow\quad\rho(BgB^{-1})(S\ell)=S\ell.

Since H=Stab⁡G(ℓ)H=\operatorname{Stab}_G(\ell), this says that g∈H1(S)g\in H_1(S) if and only if BgB−1∈H(S)BgB^{-1}\in H(S). The equality of these subgroup functors proves

(H1)K=B−1HKB.(H_1)_K=B^{-1}H_KB.

Hence H1H_1 is a proper closed subgroup of GG. Since ℓ0\ell_0 is constant, σ(Kℓ0)=Kℓ0\sigma(K\ell_0)=K\ell_0. Difference invariance therefore gives ρ(A)(Kℓ0)=Kℓ0\rho(A)(K\ell_0)=K\ell_0, or A∈H1(K)A\in H_1(K). It follows that

G=TA⊆H1⊊G,G=T_A\subseteq H_1\subsetneq G,

a contradiction. The Galois-group assertion follows from [9 Proposition 1.21(3)]. □

Construction of realizing systems

In this section we construct the coefficient matrices needed to prove Theorem 1.1. We first interpolate through finitely many generating points of the identity component while making the first-order term at infinity vanish. We then choose a semisimple representative of a component-group generator and dilate the finite singularities so that the coefficient-group criterion applies.

Unipotent elements and finitely many one-parameter subgroups

Both lemmas in this subsection follow from classical structure theory. We give elementary proofs using polynomial exponential maps, Jordan decomposition, and the containment of semisimple elements in tori [6 Theorem 9.18 and Corollary 17.36]. The subsequent use of finitely many generating points follows the strategy of [9 Proposition 3.3(2)].

Lemma 3.1 (A unipotent element lies in the identity component). If u∈G(C)u \in G(\mathbb{C}) is unipotent, then u∈G∘(C)u \in G^{\circ}(\mathbb{C}). More precisely, if u≠Iu \ne I and N=log⁡(u)N = \log(u), the polynomial homomorphism

Ga⟶GLn,t⟼exp⁡(tN)\mathbb{G}_{a} \longrightarrow\mathrm{GL}_{n}, \qquad t \longmapsto\exp(tN)

factors through GG and maps 11 to uu.

Proof. The logarithm series for a unipotent matrix and the exponential series for a nilpotent matrix both terminate. Their formal identities give exp⁡((s+t)N)=exp⁡(sN)exp⁡(tN)\exp((s+t)N) = \exp(sN)\exp(tN) and exp⁡(N)=u\exp(N) = u. For every integer mm, exp⁡(mN)=um∈G(C)\exp(mN) = u^{m} \in G(\mathbb{C}). For any defining equation f∈I(G)f \in I(G), the function f(exp⁡(tN))f(\exp(tN)) is a polynomial in tt: since det⁡(exp⁡(tN))=1\det(\exp(tN)) = 1, the pullback of every regular function on GLn\mathrm{GL}_{n} is a polynomial in tt. It vanishes at all integers, an infinite subset of C\mathbb{C}, and is therefore identically zero. The image is consequently contained in GG. Since it is connected and contains II, it lies in G∘G^{\circ}.

If u≠Iu \ne I, choose an entry Nij≠0N_{ij} \ne0. The polynomial

1Nij(∑k=1n−1(−1)k+1k(X−I)k)ij\frac{1}{N_{ij}}\left(\sum_{k=1}^{n-1}\frac{(-1)^{k+1}}{k}(X-I)^{k}\right)_{ij}

pulls back to tt, since log⁡(exp⁡(tN))=tN\log(\exp(tN)) = tN. Thus the induced map C[G]→C[t]\mathbb{C}[G] \to\mathbb{C}[t] is surjective, showing that the parametrization is a closed immersion of Ga\mathbb{G}_{a}. The case u=Iu = I is immediate. □

Lemma 3.2 (Finite one-parameter generation). Every nontrivial connected linear algebraic group HH over C\mathbb{C} is generated as an algebraic group by finitely many closed subgroups isomorphic to Ga\mathbb{G}_{a} or Gm\mathbb{G}_{m}.

Proof. Let U\mathcal{U} be the family of closed subgroups of HH isomorphic to Ga\mathbb{G}_{a} or Gm\mathbb{G}_{m}. For F⊆U\mathcal{F} \subseteq\mathcal{U}, denote by PFP_{\mathcal{F}} the algebraic subgroup generated by F\mathcal{F}, whose existence follows from [6 Proposition 2.51].

For g∈H(C)g \in H(\mathbb{C}), write g=gsgug = g_{s}g_{u} for its multiplicative Jordan decomposition in HH [6 Theorem 9.18]. By Lemma 3.1, every nonidentity gug_{u} belongs to a closed subgroup isomorphic to Ga\mathbb{G}_{a}. By [6 Corollary 17.36], gsg_{s} belongs to a torus in HH. Since C\mathbb{C} is algebraically closed, this torus is a product of copies of Gm\mathbb{G}_{m} [6 Definition 12.14 and Theorem 12.18(d)]. Thus every g∈H(C)g \in H(\mathbb{C}) belongs to PUP_{\mathcal{U}}, and [6 Corollary 1.18] gives

PU=H.(10)P_{\mathcal{U}} = H. \tag*{(10)}

For a nonempty finite subfamily F={U1,…,Ur}\mathcal{F} = \{U_{1}, \ldots, U_{r}\}, the multiplication morphism

U1×⋯×Ur⟶H,(u1,…,ur)⟼u1⋯urU_{1} \times\cdots\times U_{r} \longrightarrow H, \qquad(u_{1}, \ldots, u_{r}) \longmapsto u_{1} \cdots u_{r}

has a smooth, connected source. Its image is contained in PFP_{\mathcal{F}} and contains every UiU_{i}, as is seen by setting all other coordinates equal to the identity. Hence its image generates PFP_{\mathcal{F}}. By [6 Proposition 2.53 and Corollary 1.35], PFP_{\mathcal{F}} is connected and irreducible. This also holds for P∅={1H}P_{\varnothing} = \{1_{H}\}.

Choose a finite family F0⊆U\mathcal{F}_{0} \subseteq\mathcal{U} for which dim⁡PF0\dim P_{\mathcal{F}_{0}} is maximal; such a choice exists because these dimensions are integers between 00 and dim⁡H\dim H. For every U∈UU \in\mathcal{U},

PF0⊆PF0∪{U},dim⁡PF0=dim⁡PF0∪{U}.P_{\mathcal{F}_{0}} \subseteq P_{\mathcal{F}_{0} \cup\{U\}}, \qquad\dim P_{\mathcal{F}_{0}} = \dim P_{\mathcal{F}_{0} \cup\{U\}}.

The larger subgroup is irreducible, so the two subgroups are equal [6 Appendix A.47]. Hence PF0P_{\mathcal{F}_{0}} contains every U∈UU \in\mathcal{U}, and (3.1) implies PF0=HP_{\mathcal{F}_{0}} = H. The family F0\mathcal{F}_{0} is nonempty because HH is nontrivial. □

Fix such a finite list, with embeddings

ui:Ga↪Horλi:Gm↪H,1≤i≤r.u_i:\mathbb{G}_a \hookrightarrow H \quad\text{or}\quad\lambda_i:\mathbb{G}_m \hookrightarrow H,\qquad1 \le i \le r.

For an additive subgroup, put gi=ui(1)g_i=u_i(1); for a multiplicative subgroup, put gi=λi(2)g_i=\lambda_i(2). The powers of gig_i are respectively ui(m)u_i(m), m∈Zm\in\mathbb{Z}, or λi(2m)\lambda_i(2^m), m∈Zm\in\mathbb{Z}. These parameter sets are infinite in characteristic zero. Every infinite subset of A1\mathbb{A}^1 or Gm\mathbb{G}_m is Zariski dense, since a nonzero polynomial in one variable has finitely many roots. Therefore

H=⟨g1,…,gr⟩‾Zar.(11)H=\overline{\langle g_1,\ldots,g_r\rangle}^{\mathrm{Zar}}. \tag*{(11)}

Only these finitely many points will be prescribed on a rational curve.

Interpolation with a vanishing first jet

The formulas here give the required second-order condition in the coefficient variable xx, using elementary Lagrange interpolation on one-parameter subgroups. Their singularities can then be controlled independently of the matrix entries defining those subgroups. For 1≤i≤r1\le i\le r, define

M=r+1,Li(z)=∏1≤j≤rj≠iz−ji−j,pi(z)=iMLi(z)zM.(12)M=r+1,\qquad L_i(z)=\prod_{\substack{1\le j\le r\\j\ne i}}\frac{z-j}{i-j},\qquad p_i(z)=\frac{i^M L_i(z)}{z^M}. \tag*{(12)}

The empty product is understood to be 1. Since distinct nonzero integers remain distinct in C\mathbb{C}, these formulas are valid. They satisfy

pi(j)=δij,pi(z)=O(z−2)(z=∞).(13)p_i(j)=\delta_{ij},\qquad p_i(z)=O(z^{-2})\quad(z=\infty). \tag*{(13)}

The second assertion follows from deg⁡Li=r−1=M−2\deg L_i=r-1=M-2. Put

Fi(z)={ui(pi(z)),for an additive subgroup,λi(1+pi(z)),for a multiplicative subgroup,F_i(z)= \begin{cases} u_i(p_i(z)), & \text{for an additive subgroup},\\ \lambda_i(1+p_i(z)), & \text{for a multiplicative subgroup}, \end{cases}
B(z)=F1(z)⋅…⋅Fr(z).(14)B(z)=F_1(z)\cdot\ldots\cdot F_r(z). \tag*{(14)}

The rational function 1+pi1+p_i is nonzero because its value at infinity is 1, so the multiplicative expression defines a point over C(z)\mathbb{C}(z). All factors are H(C(z))H(\mathbb{C}(z))-points.

At z=jz=j, every additive parameter is 0 except possibly the jjth, and every multiplicative parameter is 1 except possibly the jjth, which is 2. All evaluations are defined. Hence

B(j)=gj,B(z)=I+O(z−2),B(∞)=I.(15)B(j)=g_j,\qquad B(z)=I+O(z^{-2}),\qquad B(\infty)=I. \tag*{(15)}

The order estimate holds because each regular group homomorphism maps its identity to II and preserves congruence modulo z−2z^{-2}. Multiplying finitely many matrices congruent to II modulo z−2z^{-2} preserves that congruence. By (11), the values of BB generate HH.

A finite set of algebraic numbers containing the singularities

The entries of ui(t)u_i(t) and ui(t)−1=ui(−t)u_i(t)^{-1}=u_i(-t) are polynomials in tt. Thus an additive factor in (14) has no finite pole except possibly at z=0z=0. The entries of λi(t)\lambda_i(t) and its inverse are Laurent polynomials, since C[Gm]=C[t,t−1]\mathbb{C}[\mathbb{G}_m]=\mathbb{C}[t,t^{-1}]. Consequently, a multiplicative factor and its inverse have finite poles only among

0and the zeros ofQi(z)=zM+iMLi(z).0\quad\text{and the zeros of}\quad Q_i(z)=z^M+i^M L_i(z).

These QiQ_i are monic polynomials in Q[z]\mathbb{Q}[z]. Thus

P0={0}∪⋃i multiplicative{α∈Q‾:Qi(α)=0}(16)P_0=\{0\}\cup\bigcup_{i\ \mathrm{multiplicative}}\{\alpha\in\overline{\mathbb{Q}}:Q_i(\alpha)=0\} \tag*{(16)}

satisfies P(B)⊆P0\mathcal{P}(B)\subseteq P_0.

Here we choose an algebraic closure of Q\mathbb{Q} inside C\mathbb{C}. Products and inverse products create no pole outside the union of the factor singularities. Possible cancellations only shrink P(B)\mathcal{P}(B). The points of P0P_0 are algebraic even when the entries defining HH and its one-parameter subgroups are arbitrary constants in C\mathbb{C}.

Arithmetic dilation and realization of the full group

The dilation below is an elementary avoidance argument in a finite-dimensional Q\mathbb{Q}-vector space. Its role in the proof is to arrange the hypothesis of Theorem 2.5.

Lemma 3.3 (Dilation of a finite singularity set). For every finite P0⊂CP_0 \subset\mathbb{C}, there is t∈C∗t \in\mathbb{C}^{*} such that 1∉span⁡Q(tP0)1 \notin\operatorname{span}_{\mathbb{Q}}(tP_0). If P0P_0 lies in a number field FF of degree DD, any root t∈Ct \in\mathbb{C} of TD+1−2T^{D+1}-2 is such a choice.

Proof. The space W=span⁡QP0W=\operatorname{span}_{\mathbb{Q}}P_0 is finite dimensional. The field C\mathbb{C} has infinite dimension over Q\mathbb{Q}, since it contains algebraic numbers of arbitrarily large degree. Choose a∈C∖Wa \in\mathbb{C}\setminus W and put t=a−1t=a^{-1}. Note that a≠0a\ne0, since 0∈W0\in W. The relation 1∈tW1\in tW would imply a∈Wa\in W.

For the explicit assertion, Eisenstein’s criterion gives [Q(t):Q]=D+1[\mathbb{Q}(t):\mathbb{Q}]=D+1. Hence t∉Ft\notin F and t−1∉Ft^{-1}\notin F. Since span⁡Q(tP0)⊆tF\operatorname{span}_{\mathbb{Q}}(tP_0)\subseteq tF, this space cannot contain 11.

We shall also use the following semisimple lift of a generator of the component group.

Lemma 3.4 (A semisimple lift of a quotient generator). If G/G∘G/G^{\circ} is cyclic, there is a semisimple h∈G(C)h\in G(\mathbb{C}) whose image generates that quotient. Moreover, G(C)=⟨G∘(C),h⟩G(\mathbb{C})=\langle G^{\circ}(\mathbb{C}),h\rangle.

This is the group-theoretic observation in [9 proof of Theorem 8.10, p. 91]; it applies over every algebraically closed field of characteristic zero. No assertion that hm=Ih^{m}=I is used. In particular, the exact sequence 1→G∘→G→G/G∘→11\to G^{\circ}\to G\to G/G^{\circ}\to1 need not split, and no finite-order lifting result is required.

Proof of Theorem 1.1. Put H=G∘H=G^{\circ}. Suppose first that H≠1H\ne1. Choose BB from (14), hh from Lemma 3.4, and tt from Lemma 3.3 for the set (16). Define

A(x)=hB(x/t).(17)A(x)=hB(x/t). \tag*{(17)}

Then A∈G(K)A\in G(K) and

A(x)=h+O(x−2),P(A)⊆tP0,1∉span⁡QP(A).A(x)=h+O(x^{-2}),\qquad\mathcal{P}(A)\subseteq t\mathcal{P}_0,\qquad1\notin\operatorname{span}_{\mathbb{Q}}\mathcal{P}(A).

Left multiplication by the constant hh and right multiplication by its inverse introduce no singularities.

It remains to prove that TA=GT_A=G. The group TAT_A is contained in GG. For every equation f∈I(TA)f\in I(T_A), the rational function f(A(x))f(A(x)) is zero. The matrix AA is regular and invertible at infinity, so evaluation there gives f(h)=0f(h)=0. Thus h∈TA(C)h\in T_A(\mathbb{C}). Also, using (15), we obtain

A(ti)=hgi,gi=h−1A(ti)∈TA(C).A(t_i)=hg_i,\qquad g_i=h^{-1}A(t_i)\in T_A(\mathbb{C}).

These evaluations are defined and invertible, as checked above. Since the gig_i generate HH and the image of hh generates G/G∘G/G^{\circ}, we have TA=GT_A=G. Theorem 2.8 gives the required realization.

If H=1H=1, then GG is a finite cyclic group. Take its generator hh and put A=hA=h. In characteristic zero a finite-order matrix is semisimple, because its minimal polynomial divides a separable polynomial Tm−1T^m-1. Now P(A)=∅\mathcal{P}(A)=\varnothing, A=h+O(x−2)A=h+O(x^{-2}), and TA=GT_A=G. The same criterion applies, including to the trivial group.

Examples and counterexamples

In this section we give explicit realizations of several algebraic groups and examine the hypotheses of the coefficient-group criterion. The counterexamples show why the finite singularities, semisimplicity and the first-order term at infinity must be controlled.

A realization of SL2 with a single finite singularity

Let

u+(s)=(1s01),u−(s)=(10s1),u_{+}(s)=\begin{pmatrix}1&s\\0&1\end{pmatrix},\qquad u_{-}(s)=\begin{pmatrix}1&0\\s&1\end{pmatrix},

and set

p(x)=2−xx3,q(x)=8(x−1)x3,A(x)=u+(p(x))u−(q(x))=(1+pqpq1).p(x)=\frac{2-x}{x^{3}},\qquad q(x)=\frac{8(x-1)}{x^{3}},\qquad A(x)=u_{+}(p(x))u_{-}(q(x))=\begin{pmatrix}1+pq&p\\q&1\end{pmatrix}.

Both AA and its inverse have no finite pole outside {0}\{0\}, and A=I+O(x−2)A=I+O(x^{-2}). Moreover, A(1)=u+(1)A(1)=u_{+}(1) and A(2)=u−(1)A(2)=u_{-}(1). The Zariski closures of the cyclic subgroups generated by A(1)A(1) and A(2)A(2) are the two root subgroups, respectively, and these two root subgroups generate SL⁡2\operatorname{SL}_{2}. For example, the Lie algebra generated by CE12CE_{12} and CE21CE_{21} contains E12E_{12}, E21E_{21}, and [E12,E21][E_{12},E_{21}], which form a basis of sl2\mathfrak{sl}_{2}. Thus TA=SL⁡2T_{A}=\operatorname{SL}_{2}, and the criterion applies without any dilation.

A multiplicative group

Take B(z)=1+z−2B(z)=1+z^{-2}, with singularities 0,i,−i0,i,-i. Choose tt satisfying t3=2t^{3}=2 and consider the system

A(x)=(1+t2x2)∈GL⁡1(K),σY=A(x)Y.A(x)=\left(1+\frac{t^{2}}{x^{2}}\right)\in\operatorname{GL}_{1}(K),\qquad\sigma Y=A(x)Y.

Since t−1∉Q(i)t^{-1}\notin\mathbb{Q}(i), we have 1∉span⁡Q{0,it,−it}1\notin\operatorname{span}_{\mathbb{Q}}\{0,it,-it\}, and A=I+O(x−2)A=I+O(x^{-2}). Its nonconstant coefficient has infinitely many values, which generate Gm\mathbb{G}_{m}. Thus its PV group is Gm\mathbb{G}_{m}. Dilation is not always necessary; for example, 1∉span⁡Q{0,i,−i}1\notin\operatorname{span}_{\mathbb{Q}}\{0,i,-i\} already holds. The chosen tt illustrates a uniform algebraic rule that works without first testing the individual singularity set.

A nonsplit extension with quotient of order two

Let

T={diag⁡(s,s−1):s∈C∗},h=(0−110),T=\{\operatorname{diag}(s,s^{-1}):s\in\mathbb{C}^{*}\},\qquad h=\begin{pmatrix}0&-1\\1&0\end{pmatrix},
G=NSL⁡2(T)=T∪hT.G=N_{\operatorname{SL}_{2}}(T)=T\cup hT.

To see this description of the normalizer, note that an element normalizing TT must permute its two distinct weight lines in C2\mathbb{C}^{2}. It is therefore diagonal or antidiagonal; the determinant-one condition gives exactly T∪hTT\cup hT. The quotient G/TG/T has order 22. Conjugation by hh inverts TT and h2=−Ih^{2}=-I, so (hdiag⁡(s,s−1))2=−I(h\operatorname{diag}(s,s^{-1}))^{2}=-I for every ss. Thus the extension by C2C_{2} does not split: there is no element of order 22 in the nonidentity component.

Take t3=2t^{3}=2 and q(x)=1+t2/x2q(x)=1+t^{2}/x^{2}. Then

A(x)=hdiag⁡(q(x),q(x)−1)=(0−q(x)−1q(x)0).A(x)=h\operatorname{diag}(q(x),q(x)^{-1})= \begin{pmatrix} 0&-q(x)^{-1}\\ q(x)&0 \end{pmatrix}.

has semisimple limit hh, no first-order term, and singularities among 0,it,−it0,it,-it. Its coefficient group contains hh and TT, and hence equals GG. Its PV group is therefore the nonsplit normalizer.

A finite cyclic group

The construction is classical. We give a quotient-algebra presentation of the finite cyclic PV ring and verify its prescribed matrix representation. Let G=⟨h⟩G=\langle h\rangle have order mm. Choose a primitive mm-th root of unity ζ∈C\zeta\in\mathbb{C} and write h=Pdiag⁡(ζe1,…,ζen)P−1h=P\operatorname{diag}(\zeta^{e_{1}},\ldots,\zeta^{e_{n}})P^{-1}. Put

R=K[t]/(tm−1),σ(t)=ζt,Z=Pdiag⁡(te1,…,ten)P−1.R=K[t]/(t^{m}-1),\qquad\sigma(t)=\zeta t,\qquad Z=P\operatorname{diag}(t^{e_{1}},\ldots,t^{e_{n}})P^{-1}.

The ring is isomorphic to KmK^{m}, and σ\sigma permutes its primitive idempotents transitively. An ideal is a product of ideals in these field factors, so no nonzero proper ideal is σ\sigma-stable. Also σZ=hZ\sigma Z=hZ.

Since hh has order mm, gcd⁡(m,e1,…,en)=1\gcd(m,e_{1},\ldots,e_{n})=1. The diagonal entries of P−1ZPP^{-1}ZP give teit^{e_i}, and their inverses are the diagonal entries of P−1Z−1PP^{-1}Z^{-1}P. They belong to the algebra generated by the entries of ZZ and det⁡(Z)−1\det(Z)^{-1} because Z−1=adj⁡(Z)/det⁡(Z)Z^{-1}=\operatorname{adj}(Z)/\det(Z). Bézout’s identity modulo mm therefore recovers tt in this algebra. Thus RR is generated by the fundamental matrix and its inverse determinant, and is a PV ring. The KK-automorphisms are permutations of the mm field factors. A permutation commuting with a full cycle is determined by the image of one factor, so the commuting automorphisms are precisely t↦ζjtt\mapsto\zeta^{j}t, with right-action matrices hjh^{j}.

Omitting the singularity condition or the poles of the inverse

Let a(x)=1−x−2a(x)=1-x^{-2} and y=x/(x−1)y=x/(x-1). Then

y(x+1)y(x)=(x+1)(x−1)x2=a(x).\frac{y(x+1)}{y(x)}=\frac{(x+1)(x-1)}{x^2}=a(x).

The one-dimensional system with coefficient matrix A=(a)A=(a) has fundamental matrix Z=(y)∈GL1(K)Z=(y)\in\mathrm{GL}_1(K). Thus its PV group is trivial, while its coefficient group is GmG_m. The normalization at infinity holds, but P(a)={0,1,−1}\mathcal{P}(a)=\{0,1,-1\} violates the arithmetic condition. Counting only poles of aa, and not of a−1a^{-1}, would give the erroneous set {0}\{0\}.

The degree identity (9) can be verified explicitly in the representation ρ(s)=diag⁡(s,1)\rho(s)=\operatorname{diag}(s,1):

v=(xx−1),σv=xx−1ρ(a)v.v=\begin{pmatrix}x\\x-1\end{pmatrix},\qquad\sigma v=\frac{x}{x-1}\rho(a)v.

The vector is primitive and d=1d=1. The multiplier has k0=1k_0=1 and k1=−1k_1=-1, so ∑skss=−1\sum_s k_s s=-1 and d+∑skss=0d+\sum_s k_s s=0.

Omitting semisimplicity

Let

A=(1101),Z(x)=(1x01).A=\begin{pmatrix}1&1\\0&1\end{pmatrix},\qquad Z(x)=\begin{pmatrix}1&x\\0&1\end{pmatrix}.

Then σZ=AZ\sigma Z=AZ and Z∈GL2(K)Z\in\mathrm{GL}_2(K). Thus the Picard-Vessiot group is trivial, whereas TA≅GaT_A\cong G_a. The singularity set is empty, but the limit at infinity is not semisimple. Correspondingly, the vector v=(x,1)Tv=(x,1)^T satisfies σv=Av\sigma v=Av, but the line KvKv has no nonzero constant generator. This is precisely the failure of ker⁡(M−I)∩im⁡(M−I)=0\ker(M-I)\cap\operatorname{im}(M-I)=0 in the proof of rigidity.

Allowing a first-order term at infinity

Consider

A(x)=(1+x−1+x−2−x−1x−21−x−1),v(x)=(x1).A(x)=\begin{pmatrix}1+x^{-1}+x^{-2}&-x^{-1}\\x^{-2}&1-x^{-1}\end{pmatrix},\qquad v(x)=\begin{pmatrix}x\\1\end{pmatrix}.

A direct computation shows that det⁡A=1\det A=1 and σv=Av\sigma v=Av. Both AA and A−1A^{-1} have finite poles only at 00, and A(∞)=IA(\infty)=I is semisimple. Nevertheless, KvKv is not a constant line. Thus the vanishing of the x−1x^{-1} coefficient cannot be omitted from the hypotheses.

More generally, if B=H+B1/x+O(x−2)B=H+B_1/x+O(x^{-2}) and σv=cBv\sigma v=cBv, the same two-coefficient calculation gives, for M=γHM=\gamma H,

Mvd=vd,(d+q)vd=(M−I)vd−1+γB1vd.Mv_d=v_d,\qquad(d+q)v_d=(M-I)v_{d-1}+\gamma B_1v_d.

When HH is semisimple, projection Π1\Pi_1 onto the 11-eigenspace of MM gives (d+q)vd=γΠ1B1vd(d+q)v_d=\gamma\Pi_1B_1v_d. The proof requires this extra term to vanish; our construction arranges the stronger condition B1=0B_1=0 uniformly in every representation.

Rational gauges of constant systems

Lemma 4.1 (A scalar coboundary creates an integer resonance). If f∈C(x)∗f\in\mathbb{C}(x)^* is nonconstant, the finite zero–pole support of σ(f)/f\sigma(f)/f contains two distinct points differing by a nonzero integer.

Proof. A nonconstant rational function has a finite zero or pole. In an orbit a+Za+\mathbb{Z} containing one, put nk=ord⁡a+k(f)n_k=\operatorname{ord}_{a+k}(f). This is a nonzero finitely supported sequence. Let mm and MM be the least and greatest indices with nk≠0n_k\ne0. Then

ord⁡a+k(σ(f)/f)=nk+1−nk\operatorname{ord}_{a+k}(\sigma(f)/f)=n_{k+1}-n_k

is nonzero at k=m−1k=m-1 and at k=Mk=M. These are distinct indices, and the corresponding points differ by M−m+1∈Z>0M-m+1\in\mathbb{Z}_{>0}. □

Suppose AA satisfies (2.2) and A=σ(P)DP−1A=\sigma(P)DP^{-1} with P∈GLn(K)P\in\mathrm{GL}_n(K) and constant semisimple DD. Replacing (P,D)(P,D) by (PS,S−1DS)(PS,S^{-1}DS) for a suitable S∈GLn(C)S\in\mathrm{GL}_n(\mathbb{C}), we may assume that D=diag⁡(λ1,…,λn)D=\operatorname{diag}(\lambda_1,\ldots,\lambda_n). Applying Theorem 2.5 to the columns of PP gives

P=Qdiag⁡(f1,…,fn),Q∈GLn(C),fj∈K∗.P=Q\operatorname{diag}(f_1,\ldots,f_n),\qquad Q\in\mathrm{GL}_n(\mathbb{C}),\quad f_j\in K^*.

Consequently,

Q−1AQ=diag⁡(λ1σ(f1)f1,…,λnσ(fn)fn).Q^{-1}AQ=\operatorname{diag}\left(\lambda_1\frac{\sigma(f_1)}{f_1},\ldots,\lambda_n\frac{\sigma(f_n)}{f_n}\right).

The finite zeros and poles of each σ(fj)/fj\sigma(f_j)/f_j lie in P(A)\mathcal{P}(A). Lemma 4.1 and 1∉span⁡QP(A)1\notin\operatorname{span}_{\mathbb{Q}}\mathcal{P}(A) therefore force every fjf_j to be constant. Hence PP is constant.

An alternative interpolation construction

In this section, we give an alternative construction of the generating rational curve using a rational chart and Hermite interpolation. The interpolation prescribes tangent vectors spanning the Lie algebra of the identity component and ensures the required expansion at infinity. Smoothness and connectedness then imply that the values of the curve generate the identity component.

A connected linear algebraic group HH over C\mathbb{C} is a rational variety; see [9 proof of Proposition 3.3(2)]. Put d=dim⁡Hd=\dim H. Translating a rational chart gives an open subset U⊂AdU\subset\mathbb{A}^d and a regular map that is an isomorphism onto an open subset,

F:U⟶H,0∈U,F(0)=I,dF0:Cd→∼Lie⁡(H).F:U\longrightarrow H,\qquad0\in U,\quad F(0)=I,\quad dF_0:\mathbb{C}^d\xrightarrow{\sim}\operatorname{Lie}(H).

For d>0d>0, put bi=ib_i=i, M=2d+1M=2d+1, and

Li(z)=∏j≠iz−bjbi−bj,ui(z)=biM(z−bi)Li(z)2zM.L_i(z)=\prod_{j\ne i}\frac{z-b_j}{b_i-b_j},\qquad u_i(z)=\frac{b_i^M(z-b_i)L_i(z)^2}{z^M}.

At z=biz=b_i, the factor z−biz-b_i gives derivative 11 after multiplication by biM/zMb_i^M/z^M. At z=bjz=b_j, j≠ij\ne i, the square of LiL_i has a double zero. Hence

ui(bj)=0,ui′(bj)=δij,ui=O(z−2).u_i(b_j)=0,\qquad u_i'(b_j)=\delta_{ij},\qquad u_i=O(z^{-2}).

The last estimate follows because the numerator has degree 2d−12d-1. The curve B(z)=F(u1(z),…,ud(z))B(z)=F(u_1(z),\ldots,u_d(z)) is defined generically and at the specified nodes and infinity. At each interpolation node and at infinity, the parameter vector (u1(z),…,ud(z))(u_1(z),\ldots,u_d(z)) equals 0∈U0\in U. Hence the composition defining BB is well defined at these points. It satisfies

B(z)=I+O(z−2),B(bj)=I,B′(bj)=dF0(ej).B(z)=I+O(z^{-2}),\qquad B(b_j)=I,\qquad B'(b_j)=dF_0(e_j).

If a closed subgroup P⊆HP\subseteq H contains all regular values of BB, its Lie algebra contains these derivatives, which form a basis of Lie⁡(H)\operatorname{Lie}(H). Smoothness in characteristic zero gives dim⁡P=dim⁡H\dim P=\dim H; connectedness of HH implies P=HP=H. Its finite singularities need not be algebraic over Q\mathbb{Q}, but the first assertion of Lemma 3.3 works for every finite set. Thus A(x)=hB(x/t)A(x)=hB(x/t) gives a second construction under the same rigidity criterion.

Acknowledgements

The author was supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant Nos. XDA0480502 and XDA0480503) and by the National Natural Science Foundation of China (Grant No. 12471479). The author is very grateful to Michael Singer for his valuable suggestions concerning the results of this paper.

State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, School of Mathematical Sciences, University of Chinese Academy of Sciences

Email address: [email protected]

References

  1. [1]A. Bachmayr, D. Harbater, J. Hartmann, and M. Wibmer, Free differential Galois groups, Trans. Amer. Math. Soc. 374 (2021), no. 6, 4293–4308. doi:10.1090/tran/8352.DOI
  2. [2]R. Feng, Difference Galois groups under specialization, Trans. Amer. Math. Soc. 374 (2021), no. 1, 61–96. doi:10.1090/tran/8143.DOI
  3. [3]R. Feng and M. Wibmer, Differential Galois groups, specializations and Matzat’s conjecture, Mem. Amer. Math. Soc. 319 (2026), no. 1622, v+91 pp. doi:10.1090/memo/1622.DOI
  4. [4]J. Hartmann, On the inverse problem in differential Galois theory, J. Reine Angew. Math. 586 (2005), 21–44. doi:10.1515/crll.2005.586.21.
  5. [5]E. R. Kolchin, Some problems in differential algebra, in Proceedings of the International Congress of Mathematicians (Moscow, 1966), vol. 1, Mir, Moscow, 1968, pp. 269–276.
  6. [6]J. S. Milne, Algebraic Groups: The Theory of Group Schemes of Finite Type over a Field, Cambridge Studies in Advanced Mathematics, vol. 170, Cambridge University Press, Cambridge, 2017.
  7. [7]C. Mitschi and M. F. Singer, Connected linear groups as differential Galois groups, J. Algebra 184 (1996), no. 1, 333–361. doi:10.1006/jabr.1996.0263.DOI
  8. [8]C. Mitschi and M. F. Singer, Solvable-by-finite groups as differential Galois groups, Ann. Fac. Sci. Toulouse Math. (6) 11 (2002), no. 3, 403–423. doi:10.5802/afst.1029.DOI
  9. [9]M. van der Put and M. F. Singer, Galois Theory of Difference Equations, Lecture Notes in Mathematics, vol. 1666, Springer, 1997.
  10. [10]J.-P. Ramis and J. Sauloy, The q-analogue of the wild fundamental group and the inverse problem of the Galois theory of q-difference equations, Ann. Sci. Éc. Norm. Supér. (4) 48 (2015), no. 1, 171–226. doi:10.24033/asens.2241.DOI
  11. [11]T. Serfini and M. Wibmer, The regular singular inverse problem in differential Galois theory, to appear in Bull. Soc. Math. France, arXiv:2501.07993, version 2.
  12. [12]C. Tretkoff and M. Tretkoff, Solution of the inverse problem of differential Galois theory in the classical case, Amer. J. Math. 101 (1979), no. 6, 1327–1332. doi:10.2307/2374143.

Paper details

vlnc:W_f34n4jzr
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT 6, MechMath Agent Team (MMAT)

View statement
Looking for related papers…
Contents