Let L/K be an arbitrary Galois extension with group G, equipped with the Krull topology. Let C(G,K) denote the K-vector space of continuous functions G\to K, with K discrete and G acting by left translation. We prove that there exists a G-equivariant K-linear isomorphism \varphi:C(G,K)\to L such that, for every open subgroup H of G, the map \varphi^H:C(H,L^H)\to L induced by extension by zero and extension of scalars is an H-equivariant L^H-linear isomorphism. This gives a completely normal analogue of Lenstra’s infinite normal basis theorem; the case of finite base fields was previously established in an equivalent form by Hachenberger.
Let L/K be a finite Galois field extension with Galois group G. Then there exists an element x \in L such that the family (\sigma(x))_{\sigma\in G} forms a basis for L as a vector space over K. In that case, (\sigma(x))_{\sigma\in G} is called a normal basis for L/K, and x is called a normal basis generator for L/K.
Let L/K be a finite Galois field extension with Galois group G. Then there exists an element x\in L which is a normal basis generator for L/M for every intermediate field K\subseteq M\subseteq L. In that case, the family (\sigma(x))_{\sigma\in G} is called a completely normal basis for L/K, and x is called a completely normal basis generator for L/K.
Let L/K be a Galois field extension with Galois group G. Then there is a G-equivariant K-linear isomorphism C(G,K) \to L.
Let L/K be a Galois field extension with Galois group G. Then there exists a G-equivariant K-linear isomorphism \varphi:C(G,K) \to L such that, for every open subgroup H of G, the induced H-equivariant L^H-linear map \varphi^H:C(H,L^H) \to L is an isomorphism.
Every surjective affine-linear map between finite-dimensional affine K-spaces is open for the Zariski topology.
Let q:A\to B be such a map and let O\subseteq A be open. For each a\in O, choose an affine-linear section s:B\to A of q satisfying s(q(a))=a. Since s is continuous, s^{-1}(O) is an open neighborhood of q(a) contained in q(O). Thus q(O) is open.
Let K be an infinite field and let A be a nonempty finite-dimensional affine K-space. If \{O_\lambda:\lambda\in\Lambda\} is a family of nonempty Zariski-open subsets of A with |\Lambda|<|K|, then \bigcap_{\lambda\in\Lambda}O_\lambda\neq \varnothing.
This follows by taking complements in [Ghosh2022, Theorem 3.14] when \dim A>0. If \dim A=0, then A is a singleton and the assertion is immediate.
Let (X_n,T_{m,n}) be an inverse system of nonempty finite sets indexed by the positive integers ordered by divisibility. Then \varprojlim_n X_n\neq\varnothing.
The assertion is a special case of [BourbakiSets, Ch. III, §7, no. 4, Theorem 1]. For completeness, we give an elementary proof adapted to the present setting. For each n\geq1, put Y_n=\bigcap_{n\mid m}T_{m,n}(X_m). Choose a multiple d_n of n with |T_{d_n,n}(X_{d_n})| minimal. If n\mid m and r=\operatorname{lcm}(d_n,m), then T_{r,n}(X_r) \subseteq T_{d_n,n}(X_{d_n})\cap T_{m,n}(X_m). Minimality therefore gives T_{r,n}(X_r)=T_{d_n,n}(X_{d_n}). Thus, Y_n=T_{d_n,n}(X_{d_n})\neq\varnothing, and T_{r,n}(X_r)=Y_n whenever d_n\mid r. If n\mid m, choose a common multiple r of d_n and d_m. Then T_{m,n}(Y_m) =T_{m,n}(T_{r,m}(X_r)) =T_{r,n}(X_r) =Y_n. Thus the restricted transition maps are surjective. We may therefore choose recursively y_k\in Y_{k!} such that T_{(k+1)!,k!}(y_{k+1})=y_k for every k\geq1. Put x_n=T_{n!,n}(y_n). If n\mid m, the recursive relations give T_{m!,n!}(y_m)=y_n, and hence T_{m,n}(x_m) =T_{m!,n}(y_m) =T_{n!,n}(y_n)=x_n. Consequently (x_n)_{n\geq1}\in\varprojlim_n X_n.
Let \kappa be an infinite cardinal, and let (X_i,p_{ji})_{i\in I} be an inverse system of nonempty sets indexed by a directed set I with |I|\leq\kappa. Suppose that, whenever S\subseteq I satisfies |S|<\kappa and i\in I\setminus S, every finitely liftable family on S extends to a finitely liftable family on S\cup\{i\}. Then, for every directed subset J\subseteq I with |J|<\kappa, the restriction map \varprojlim_{i\in I}X_i \to \varprojlim_{j\in J}X_j is surjective.
Fix a directed subset J\subseteq I with |J|<\kappa and a coherent family (a_j)_{j\in J}. This family is finitely liftable, since every finite subset of J has an upper bound in J. We extend this family by transfinite recursion, maintaining finite liftability by transfinite induction; see [BourbakiSets, Ch. III, §2, no. 2, C59–C60].
Write I\setminus J=\{i_\alpha:\alpha<\mu\}, where \mu=|I\setminus J|\leq\kappa is regarded as its initial ordinal, and put S_\alpha=J\cup\{i_\beta:\beta<\alpha\} for \alpha\leq\mu. To make the choices definite, fix a well-ordering of \bigcup_{i\in I}X_i. Start with x_j=a_j for j\in J. Suppose that \alpha<\mu and a finitely liftable family has been constructed on S_\alpha. Since |J|<\kappa and |\alpha|<\mu\leq\kappa, we have |S_\alpha|<\kappa. The hypothesis therefore allows us to choose x_{i_\alpha}\in X_{i_\alpha} so that the extended family on S_{\alpha+1} remains finitely liftable. Choose the least such value in the fixed well-ordering. At a nonzero limit stage \lambda\leq\mu, take the union of the previously constructed families. Every finite subset of S_\lambda is contained in some S_\alpha with \alpha<\lambda. Thus every finite subfamily already has a common lift, and the union remains finitely liftable. The recursion produces a finitely liftable family (x_i)_{i\in I} extending (a_j)_{j\in J}. This family is coherent: if i\leq j, choose a common lift y\in X_k of x_i and x_j. Compatibility of the transition maps gives x_i = p_{ki}(y) = p_{ji}(p_{kj}(y))= p_{ji}(x_j).
Hence (x_i)_{i\in I}\in\varprojlim_{i\in I}X_i extends (a_j)_{j\in J}, proving surjectivity.
Let \kappa be an infinite cardinal, and let (X_i,p_{ji})_{i\in I} be an inverse system of nonempty sets with surjective transition maps, indexed by a directed set I with |I|\leq\kappa. For each i\in I, suppose that a collection \mathcal C_i of subsets of X_i satisfies:
(i) X_i\in\mathcal C_i, and \mathcal C_i is closed under finite intersections.
(ii) Every subfamily of \mathcal C_i of cardinality less than \kappa having the finite intersection property has nonempty intersection.
(iii) For i\leq j and x\in X_i, the fibre p_{ji}^{-1}(\{x\}) belongs to \mathcal C_j.
(iv) For i\leq j and C\in\mathcal C_j, the image p_{ji}(C) belongs to \mathcal C_i.
Then, for every directed subset J\subseteq I such that |J|<\kappa, the restriction map \varprojlim_{i\in I}X_i \to \varprojlim_{j\in J}X_j is surjective.
By Lemma 8, it suffices to verify the extension property for finitely liftable partial families. Fix such a family (x_j)_{j\in S} with |S|<\kappa, and let i\in I\setminus S.
First observe that a finite family having a common lift in some X_\ell has a common lift in X_k for every upper bound k of its indices. Indeed, choose r\geq k,\ell, lift the given element from X_\ell to X_r using surjectivity, and project to X_k.
For each finite F\subseteq S, choose an upper bound k of F\cup\{i\}. We now define C_F= p_{ki}( \bigcap_{j\in F}p_{kj}^{-1}(\{x_j\})). When F=\varnothing, the intersection inside the parentheses is understood to be X_k. Conditions (i), (iii), and (iv) imply that C_F\in\mathcal C_i.
By the preceding observation, C_F is independent of the chosen upper bound k. It consists precisely of those u\in X_i such that u and (x_j)_{j\in F} have a common lift. Finite liftability and the same observation show that C_F\neq\varnothing. Moreover, for any finite subsets F_1,\ldots,F_r\subseteq S, we have C_{F_1\cup\cdots\cup F_r} \subseteq C_{F_1}\cap\cdots\cap C_{F_r}. Thus the family of all C_F has the finite intersection property.
Since |S|<\kappa, there are fewer than \kappa finite subsets of S. Condition (ii) therefore allows us to choose x_i\in \bigcap_{F\subseteq S, \ F\text{ finite}}C_F. In particular, x_i\in C_{\varnothing}=X_i. For every finite F\subseteq S, membership in C_F means that x_i and (x_j)_{j\in F} have a common lift. Hence adjoining x_i preserves finite liftability.
Let K be an infinite field and let (V_i,p_{ji}) be an inverse system of nonempty finite-dimensional affine K-spaces with surjective affine-linear transition maps, indexed by a directed set I with |I|\leq|K|. Let U_i\subseteq V_i be nonempty Zariski-open subsets satisfying p_{ji}(U_j)=U_i whenever i\leq j. Then, for every directed subset J\subseteq I with |J|<|K|, the restriction map \varprojlim_{i\in I}U_i \to \varprojlim_{j\in J}U_j is surjective.
Put \kappa=|K|. We apply Theorem 9 to the inverse system (U_i,p_{ji}|_{U_j}). For each i\in I, let \mathcal C_i consist of \varnothing and all nonempty subsets of U_i that are relatively Zariski open in some affine subspace of V_i.
Clearly, U_i\in\mathcal C_i. If finitely many members of \mathcal C_i have nonempty intersection, their intersection is relatively Zariski open in the intersection of the corresponding affine subspaces. Since \varnothing\in\mathcal C_i, it follows that \mathcal C_i is closed under finite intersections.
For i\leq j and x\in U_i, the fibre of p_{ji}|_{U_j} over x is U_j\cap p_{ji}^{-1}(\{x\}), which is relatively Zariski open in the affine subspace p_{ji}^{-1}(\{x\}) of V_j. Hence this fibre belongs to \mathcal C_j. Moreover, if C\in\mathcal C_j is nonempty and relatively Zariski open in an affine subspace B\subseteq V_j, then Lemma 5, applied to p_{ji}|_B:B \to p_{ji}(B), shows that p_{ji}(C) is relatively Zariski open in the affine subspace p_{ji}(B). Since p_{ji}(C)\subseteq U_i, we have p_{ji}(C)\in\mathcal C_i. The image of the empty set also belongs to \mathcal C_i.
It remains to verify the intersection property. Fix i\in I, and let \{C_\lambda:\lambda\in\Lambda\}\subseteq\mathcal C_i have the finite intersection property, with |\Lambda|<\kappa. The case \Lambda=\varnothing follows from U_i\neq\varnothing, so assume \Lambda\neq\varnothing. For each \lambda, choose an affine subspace A_\lambda\subseteq V_i in which C_\lambda is relatively Zariski open.
For every nonempty finite F\subseteq\Lambda, put A_F=\bigcap_{\lambda\in F}A_\lambda. These affine subspaces are nonempty by the finite intersection property. Choose a nonempty finite F_0\subseteq\Lambda for which \dim A_{F_0} is minimal, and put A=A_{F_0}. For every \lambda\in\Lambda, A_{F_0\cup\{\lambda\}} = A\cap A_\lambda\subseteq A. Minimality of dimension gives A\cap A_\lambda=A, so A\subseteq A_\lambda. Consequently, C_\lambda\cap A is relatively Zariski open in A. It is also nonempty, since it contains \bigcap_{\nu\in F_0\cup\{\lambda\}}C_\nu, which is nonempty by the finite intersection property.
Lemma 6 now gives \varnothing\neq \bigcap_{\lambda\in\Lambda}(C_\lambda\cap A) \subseteq \bigcap_{\lambda\in\Lambda}C_\lambda. All hypotheses of Theorem 9 are therefore satisfied, and the desired surjectivity follows.
The assumption |J|<|K| in Proposition 10 cannot be replaced by |J|\leq|K|. We adapt Waterhouse’s example [Waterhouse1972]. Let K be an infinite field, put E=K\sqcup\{*\}, and let I be the set of finite subsets of E, ordered by inclusion. Put J=\{F\in I \mid F\subseteq K\}. Then I and J are directed and |I|=|J|=|K|. For F\in I, put V_F=K^F and U_F=\{x\in K^F:x_a\neq x_b \text{ whenever }a,b\in F,\ a\neq b\}. These sets are nonempty and Zariski open. The coordinate restriction maps are surjective linear maps between the V_F and are also surjective on the U_F: every injection from a finite subset of E into K extends to any larger finite subset, since K is infinite. However, the coherent family (\operatorname{id}_K|_F)_{F\in J} cannot extend to an element of \varprojlim_{F\in I}U_F. Such an extension would define an injection E\to K whose restriction to K is the identity, leaving no possible value for the additional element *.
Let A be a finite-dimensional K-algebra. Then A^\times is a nonempty basic Zariski-open subset of the affine K-space A.
For a\in A, let L_a:A\to A denote left multiplication by a. An element a is invertible if and only if L_a is invertible. Indeed, if L_a is invertible, choose b\in A with ab=1; injectivity of L_a then gives ba=1. The function d:A\to K, defined by d(a)=\det(L_a), is polynomial in linear coordinates. Thus A^\times=D(d), which is nonempty since d(1)=1.
Let R and S be finite-dimensional K-algebras, and let f:R\to S be a surjective K-algebra homomorphism. Then f(R^\times)=S^\times.
This holds more generally for surjective homomorphisms from Artinian rings; see [Lenstra, Lemma (2.6)]. For completeness, we give a direct proof in two cases. The inclusion f(R^\times)\subseteq S^\times is immediate. Now we show the reverse inclusion.
Case 1: K is infinite. By Lemma 12, R^\times is a nonempty Zariski-open subset of R. Since f is surjective and K-linear, Lemma 5 shows that H:=f(R^\times) is a nonempty Zariski-open subset of S. Moreover, H is a subgroup of S^\times. Take s\in S^\times. Left multiplication by s is a K-linear automorphism of S, so sH is also nonempty and Zariski open in S. Since K is infinite, H\cap sH\neq\varnothing. Thus s=h_1h_2^{-1}\in H for some h_1,h_2\in H. This proves f(R^\times) \supseteq S^\times.
Case 2: K is finite. Let s\in S^\times and choose a\in R with f(a)=s. Since R is finite-dimensional over the finite field K, it is finite. The powers of a are therefore eventually periodic. Taking n sufficiently large and divisible by the period gives a^{2n}=a^n. Thus e:=a^n is idempotent. Since f(e)=s^n is both a unit and an idempotent, we have f(e)=1. Put u=ae+(1-e). Since a commutes with e, the two summands annihilate one another, and u^n=(ae)^n+(1-e)=e+(1-e)=1. Hence u\in R^\times, and f(u)=f(a)f(e)+1-f(e)=s. This proves f(R^\times) \supseteq S^\times.
The K-algebra S is finite-dimensional.
By [ChaseHarrisonRosenberg, Theorem 1.3(c)], S is finitely generated projective as an R-module. Since R is finite-dimensional over K, so is S.
The extension S/R has a G-normal basis.
Since R is finite-dimensional over K, it is Artinian and hence semilocal. The assertion therefore follows from [ChaseHarrisonRosenberg, Theorem 4.2(c)].
An element x\in S is a G-normal basis generator for S/R if and only if \rho_G(x)\in S[G]^\times.
This is a special case of [Mazur, Proposition 1].
Let H\trianglelefteq G. Then \operatorname{Tr}_H(\mathcal N_G(S/R)) = \mathcal N_{G/H}(S^H/R). In particular, \operatorname{Tr}_G(\mathcal N_G(S/R))=R^\times.
Let s\in\mathcal N_G(S/R) and put t=\operatorname{Tr}_H(s). Since H permutes the R-basis (\sigma(s))_{\sigma\in G}, the sums over its orbits form an R-basis of S^H. As H is normal, these sums are precisely the G/H-conjugates of t. Thus t\in\mathcal N_{G/H}(S^H/R), proving one inclusion.
Conversely, let x\in\mathcal N_{G/H}(S^H/R). Choose s\in\mathcal N_G(S/R) by Theorem 15, and put t=\operatorname{Tr}_H(s). Since x and t are both normal basis generators, there exist v,w\in R[G/H] such that x=v(t) and t=w(x). Their normal basis properties give wv=vw=1, so v\in R[G/H]^\times.
By Proposition 13, the canonical surjection R[G]\to R[G/H] lifts v to some u\in R[G]^\times. Put y=u(s). The map R[G] \ni a \mapsto a(y)=(au)(s) \in S is an R-linear isomorphism, being the composite of right multiplication by u and the normal basis isomorphism associated with s. Hence y\in\mathcal N_G(S/R). Since H is normal, the trace is compatible with the quotient group-algebra action, so \operatorname{Tr}_H(y)=v(\operatorname{Tr}_H(s))=v(t)=x. This proves the reverse inclusion.
Taking H=G gives the final assertion, since \mathcal N_{\{e\}}(R/R)=R^\times.
\mathcal N_G(S/R) is a nonempty Zariski-open subset of the affine K-space S.
Let K be infinite. Then \mathcal{CN}_G(S/R) is a nonempty Zariski-open subset of the affine K-space S. In particular, S/R has a G-completely normal basis.
For every subgroup H\leq G, the algebra S^H is finite-dimensional over K, since it is a K-subspace of S. Lemma 18, applied to the H-Galois extension S/S^H, therefore shows that \mathcal{N}_H(S/S^H) is nonempty and Zariski open in S. Since K is infinite, these subsets are dense. As G has only finitely many subgroups, their intersection \mathcal{CN}_G(S/R) is nonempty and Zariski open.
The hypothesis that K is infinite in Theorem 19 cannot be omitted. Let K=\mathbb F_3, put S=K\times K, and let G=C_2=\{e,\tau\} act by \tau(a,b)=(b,a). Identifying K with the diagonal subalgebra of S, the extension S/K is G-Galois. If x=(a,b) were a G-completely normal basis generator, the condition for the trivial subgroup would imply x\in S^\times, so a,b\neq0. But then \left|\begin{smallmatrix}a&b\\ b&a\end{smallmatrix}\right| =a^2-b^2 = 1-1 = 0, so (x,\tau(x)) is not a K-basis of S. Thus \mathcal{CN}_G(S/K)=\varnothing.
Let H\trianglelefteq G. Then \operatorname{Tr}_H(\mathcal{CN}_G(S/R)) \subseteq \mathcal{CN}_{G/H}(S^H/R).
Let x\in\mathcal{CN}_G(S/R) and put y=\operatorname{Tr}_H(x). For every subgroup V with H\leq V\leq G, we have x\in\mathcal{N}_V(S/S^V). Since H\trianglelefteq V, Proposition 17, applied to S/S^V, gives y\in\mathcal{N}_{V/H}(S^H/S^V). Every subgroup of G/H has the form V/H, and (S^H)^{V/H}=S^V. Thus y\in\mathcal{CN}_{G/H}(S^H/R).
If K is infinite and H\trianglelefteq G, then \operatorname{Tr}_H(\mathcal{CN}_G(S/R)) =\mathcal{CN}_{G/H}(S^H/R).
By Lemma 21, it suffices to prove the reverse inclusion. Fix x\in\mathcal{CN}_{G/H}(S^H/R) and put \mathcal T_x=\{y\in S:\operatorname{Tr}_H(y)=x\}. By Proposition 17, \mathcal T_x is nonempty. Since \operatorname{Tr}_H is K-linear, \mathcal T_x is a finite-dimensional affine K-space. For every subgroup U\leq G, put \mathcal O_U =\mathcal T_x\cap\mathcal N_U(S/S^U). By Lemma 18, \mathcal O_U is Zariski open in \mathcal T_x. We show that each \mathcal O_U is nonempty.
Choose s\in\mathcal{CN}_G(S/R) by Theorem 19. Fix U\leq G, and put I=U\cap H and J=UH. Then I\trianglelefteq U, and the map uI\mapsto uH identifies U/I with J/H.
Since x is completely normal for S^H/R, it is a (J/H)-normal basis generator for S^H/S^J. By Lemma 16, \rho_{J/H}(x) is a unit of S^H[J/H]. The inclusion S^H\subseteq S^I and the preceding group isomorphism induce a homomorphism S^H[J/H] \to S^I[U/I] carrying \rho_{J/H}(x) to \rho_{U/I}(x). By Lemma 16 again, we get x\in\mathcal N_{U/I}(S^I/S^U). Since s is completely normal, Proposition 17 gives \operatorname{Tr}_U(s)\in(S^U)^\times and \operatorname{Tr}_J(s)\in(S^J)^\times.
Consequently, c:=\operatorname{Tr}_J(s)^{-1}\operatorname{Tr}_U(s)\in(S^U)^\times, and cx is also a (U/I)-normal basis generator for S^I/S^U. Applying Proposition 17 to S/S^U, we obtain y_U\in\mathcal N_U(S/S^U) with \operatorname{Tr}_I(y_U)=cx.
Let T\subseteq H be a set of representatives for the left cosets of I in H. Since J=HU and I=H\cap U, the same set represents the left cosets of U in J. Both x and \operatorname{Tr}_J(s) are fixed by H, so we get that \operatorname{Tr}_H(y_U) =\sum_{\tau\in T}\tau(\operatorname{Tr}_I(y_U)) =\sum_{\tau\in T}\tau(cx) =x\,\operatorname{Tr}_J(s)^{-1} \sum_{\tau\in T}\tau(\operatorname{Tr}_U(s)) =x\,\operatorname{Tr}_J(s)^{-1}\operatorname{Tr}_J(s)=x. Therefore y_U\in\mathcal O_U, proving nonemptiness.
Since G has only finitely many subgroups and K is infinite, the nonempty open subsets \mathcal O_U of \mathcal T_x have nonempty intersection. Any element of this intersection belongs to \mathcal{CN}_G(S/R) and has H-trace equal to x. This proves the reverse inclusion.
There is a natural bijection between trace-coherent families (x_N)_{N\in\mathcal U} satisfying x_N\in\mathcal{CN}_{G/N}(S^N/R), for N\in\mathcal U, and G-equivariant R-linear isomorphisms \varphi:C(G,R)\to S such that \varphi^H is an isomorphism for every open subgroup H\leq G. The correspondence is characterized by x_N=\varphi(1_N), where 1_N is the characteristic function of N.
We first establish the correspondence for ordinary normal basis generators. Identify (G/N,R) with C(G,R)^N by inflation. Given a trace-coherent family (x_N)_{N\in\mathcal U} of normal basis generators, define \varphi_N : (G/N,R) \to S^N by \varphi_N(f) =\sum_{\sigma N\in G/N}f(\sigma N)\sigma(x_N). Each \varphi_N is a (G/N)-equivariant R-linear isomorphism. For M\subseteq N, grouping the terms in \varphi_M(f) according to the cosets of N and using trace coherence gives \varphi_M(f)=\varphi_N(f) for f \in (G/N,R). Since C(G,R) and S are the unions of these finite-level function modules and fixed rings, respectively, the maps \varphi_N induce a G-equivariant R-linear isomorphism \varphi:C(G,R)\to S.
Conversely, let \varphi:C(G,R)\to S be such an isomorphism and put x_N=\varphi(1_N). Taking N-fixed points gives a (G/N)-equivariant R-linear isomorphism (G/N,R) \to S^N, which sends the characteristic-function basis to (\sigma(x_N))_{\sigma N\in G/N}. Thus x_N is a normal basis generator. For M\subseteq N, the identity 1_N=\sum_{\tau M\in N/M}\tau 1_M and equivariance give x_N=\sum_{\tau M\in N/M}\tau(x_M) =\operatorname{Tr}_{S^M/S^N}(x_M). Hence the family is trace coherent. These constructions are inverse to each other.
It remains to characterize complete normality. Let H\leq G be open and let N\in\mathcal U satisfy N\subseteq H. The restriction of \varphi^H to N-fixed points is (H/N,S^H) \to S^N, where f \mapsto \sum_{hN\in H/N}f(hN)h(x_N). Indeed, \varphi^H sends the characteristic function of hN to h(x_N). This restriction is an isomorphism precisely when x_N is an (H/N)-normal basis generator for S^N/S^H.
If every x_N is completely normal, all these finite-level restrictions are isomorphisms. As N ranges over the open normal subgroups of G contained in H, their domains and codomains exhaust C(H,S^H) and S, respectively. Thus \varphi^H is an isomorphism.
Conversely, if every \varphi^H is an isomorphism, so is its restriction to N-fixed points. Every subgroup of G/N has the form H/N for an open subgroup H containing N, and (S^N)^{H/N}=S^H. Consequently, each x_N is completely normal.
The same construction gives a natural bijection between trace-coherent families of ordinary normal basis generators at the finite Galois levels and G-equivariant R-linear isomorphisms C(G,R)\to S. For Galois field extensions, these are the isomorphisms appearing in Lenstra’s infinite normal basis theorem.
Suppose that K is infinite and |\mathcal U|\leq |K|. Let \mathcal J\subseteq\mathcal U be directed with |\mathcal J|<|K|. Then every trace-coherent family (x_N)_{N\in\mathcal J} with x_N\in\mathcal{CN}_{G/N}(S^N/R), for N\in\mathcal J, extends to a trace-coherent family of completely normal basis generators indexed by all of \mathcal U.
For N\in\mathcal U, put V_N=S^N and U_N=\mathcal{CN}_{G/N}(S^N/R). Each V_N is finite-dimensional over K, and Theorem 19 shows that U_N is a nonempty Zariski-open subset of V_N. For M\subseteq N, put p_{MN}=\operatorname{Tr}_{S^M/S^N}. These maps are K-linear and form an inverse system by transitivity of trace. Applying Theorem 22 to the (G/M)-Galois extension S^M/R and the normal subgroup N/M gives p_{MN}(U_M)=U_N. Moreover, p_{MN}(V_M) is a Zariski-closed linear subspace of V_N containing the dense subset U_N. Hence p_{MN}(V_M)=V_N. All hypotheses of Proposition 10 are therefore satisfied, and its conclusion gives the required extension.
The text on this page is from arXiv:2610.22247v1 by Patrik Lundström, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2610/2610.22247.jsonl.