Artin L-functions of small conductor

We study the problem of finding the Artin L-functions with the smallest conductor for a given Galois type. We adapt standard analytic techniques to our novel situation of fixed Galois type and get much improved lower bounds on the smallest conductor. For small Galois types we use complete tables of number fields to determine the actual smallest conductor.

Artin L-functions L(X , s) are remarkable analytic objects built from number fields. Let Q be the algebraic closure of the rational number field Q inside the field of complex numbers C. Then Artin L-functions are indexed by continuous complex characters X of the absolute Galois group G = Gal(Q/Q), with the unital character 1 giving the Riemann zeta function L(1, s) = ζ(s). An important problem in modern number theory is to obtain a fuller understanding of these higher analogs of the Riemann zeta function. The analogy is expected to be very tight: all Artin L-functions are expected by the Artin conjecture to be entire except perhaps for a pole at s = 1; they are all expected to satisfy the Riemann hypothesis that all zeros with Re(s) ∈ (0, 1) satisfy Re(s) = 1/2.
The two most basic invariants of an Artin L-function L(X , s) are defined via the two explicit elements of G, the identity e and the complex conjugation element σ. These invariants are the degree n = X (e) and the signature r = X (σ) respectively. A measure of the complexity of L(X , s) is its conductor D ∈ Z ≥1 , which can be computed from the discriminants of related number fields. It is best for purposes such as ours to focus instead on the root conductor δ = D 1/n .
In this paper, we aim to find the simplest Artin L-functions exhibiting a given Galois-theoretic behavior. To be more precise, consider triples (G, c, χ) consisting of a finite group G, an involution c ∈ G, and a faithful character χ. We say that X has Galois type (G, c, χ) if there is a surjection h : G → G with h(σ) = c, and X = χ • h. Let L(G, c, χ) be the set of L-functions of type (G, c, χ), and let L(G, c, χ; B) be the subset consisting of L-functions with root conductor at most B. Two natural problems for any given Galois type (G, c, χ) are 1: Use known and the above conjectured properties of L-functions to get a lower bound d(G, c, χ) on the root conductors of L-functions in L(G, c, χ). 2: Explicitly identify the sets L(G, c, χ; B) with B as large as possible.
This paper gives answers to both problems, although for brevity we often fix only (G, χ) and work instead with the sets L(G, χ; B) := ∪ c L(G, c, χ; B).
There is a large literature on a special case of the situation we study. Namely let (G, c, φ) be a Galois type where φ is the character of a transitive permutation representation of G. Then the set L(G, c, φ; B) is exactly the set of Dedekind zeta functions ζ(K, s) arising from a corresponding set K(G, c, φ; B) of arithmetic equivalence classes of number fields. In this context, root conductors are just root discriminants, and lower bounds date back to Minkowski's work on the geometry of numbers. Use of Dedekind zeta functions as in 1 above began with work of Odlyzko [Odl76,Odl75,Odl77a], Serre [Ser86], Poitou [Poi77a,Poi77b], and Martinet [Mar82]. Extensive responses to 2 came shortly thereafter, with papers often focusing on a single degree n = φ(e). Early work for quartics, quintics, sextics, and septics include respectively [BF89,For91,BFP93], [SPDyD94], [Poh82,BMO90,Oli91,Oli92,Oli90], and [Lét95]. Further results towards 2 in higher degrees are extractable from the websites associated to [JR14a], [KM01], and [LMF].
The full situation that we are studying here was identified clearly by Odlyzko in [Odl77b], who responded to 1 with a general lower bound. However this more general case of Artin L-functions has almost no subsequent presence in the literature. A noticeable exception is a recent paper of Pizarro-Madariaga [PM11], who improved on Odlyzko's results on 1. A novelty of our paper is the separation into Galois types. For many Galois types this separation allows us to go considerably further on 1. This paper is also the first systematic study of 2 beyond the case of number fields.
Sections 2 and 3 review background on Artin L-functions and tools used to bound their conductors. Sections 4-6 form the new material on the lower bound problem 1, while Sections 7-9 focus on the tabulation problem 2. Finally, Section 10 returns to 1 and considers asymptotic lower bounds for root conductors of Artin L-functions in certain families. In regard to 1, Figure 8.1 and Corollary 10.1 give a quick indication of how our type-based lower bounds compare with the earlier degree-based lower bounds. In regard to both 1 and 2, Tables 8.1-8.8 show how the new lower bounds compare with actual first conductors for many types.
Artin L-functions have recently become much more computationally accessible through a package implemented in Magma by Tim Dokchitser. Thousands are now collected in a section on the LMFDB [LMF]. The present work increases our understanding of all this information in several ways, including by providing completeness certificates for certain ranges.

Artin L-functions
In this section we provide some background. An important point is that our problems allow us to restrict consideration to Artin characters which take rational values only. In this setting, Artin L-functions can be expressed as products and quotients of roots of Dedekind zeta functions, minimizing the background needed. General references on Artin L-functions include [Mar77,RM01].
2.1. Number fields. A number field K has many invariants relevant for our study. First of all, there is the degree n = [K : Q]. The other invariants we need are local in that they are associated with a place v of Q and can be read off from the corresponding completed algebra K v = K ⊗ Q v , but not from other completions.
For v = ∞, the complete invariant is the signature r, defined by K ∞ ∼ = R r × C (n−r)/2 . It is more convenient sometimes to work with the eigenspace dimensions for complex conjugation, a = (n + r)/2 and b = (n − r)/2. For an ultrametric place v = p, the full list of invariants is complicated. The most basic one is the positive integer D p = p cp generating the discriminant ideal of K p /Q p . We package the D p into the single invariant D = p D p ∈ Z ≥1 , the absolute discriminant of K.
Then this completed Dedekind zeta function ζ(K, s) meromorphically continues to the whole complex plane, with simple poles at s = 0 and s = 1 being its only singularities. It satisfies the functional equation ζ(K, s) = ζ(K, 1 − s).

Permutation characters.
We recall from the introduction that throughout this paper we are taking Q to be the algebraic closure of Q in C and G = Gal(Q/Q) its absolute Galois group. A degree n number field K then corresponds to the transitive n-element G-set X = Hom(K, Q). A number field thus has a permutation character Φ = Φ K = Φ X with Φ(e) = n. Also signature has the character-theoretic interpretation Φ(σ) = r, where σ as before is the complex conjugation element.
2.5. General characters and Artin L-functions. Let X be a character of G. Then one has an associated Artin L-function L(X , s) and conductor D X , agreeing with the Dedekind zeta function ζ(K, s) and the discriminant D K if X is the permutation character of K. The function L(X , s) has both an Euler product and Dirichlet series representation as in (2.1). In general, if Φ = X m X X then One is often interested in (2.3) where the X are irreducible characters.
For a finite set of primes S, let Q S be the compositum of all number fields in Q with discriminant divisible only by primes in S. Let G S = Gal(Q S /Q) be the corresponding quotient of G. Then for primes p ∈ S one has a well-defined Frobenius conjugacy class Fr p in G S . The local factor P p (x) in (2.1) is the characteristic polynomial det(1 − ρ(Fr p )x), where ρ is a representation with character X .
2.6. Relations with other objects. Artin L-functions of degree 1 are exactly Dirichlet L-functions, so that X can be identified with a faithful character of the quotient group (Z/DZ) × of G, with D the conductor of X . Artin L-functions coming from irreducible degree 2 characters and conductor D are expected to come from cuspidal modular forms on Γ 1 (D), holomorphic if r = 0 and nonholomorphic otherwise. This expectation is proved in all cases, except for those with r = ±2 and projective image the nonsolvable group A 5 . In general, to understand how an Artin L-function L(X , s) qualitatively relates to other objects, one needs to understand its Galois theory, including the placement of complex conjugation; in other words, one needs to identify its Galois type. To be more quantitative, one brings in the conductor.
2.7. Analytic Properties of Artin L-functions. An Artin L-function has a meromorphic continuation and functional equation, although each with an extra complication in comparison with the special case of Dedekind zeta functions. For the meromorphic continuation, the behavior at s = 1 is known: the pole order is the multiplicity (1, X ) of 1 in X . The complication is that one has poor control over other possible poles. The Artin conjecture for X says however that there are no poles other than s = 1.
The completed L-function with root number w. Irreducible characters of any compact group come in three types, orthogonal, non-real, and symplectic. The type is identified by the Frobenius-Schur indicator, calculated with respect to the Haar probability measure dg: For orthogonal characters X of G, one has X = X and moreover w = 1. The complication in comparison with permutation characters is that for the other two types, the root number w is not necessarily 1. For symplectic characters, X = X and w can be either of the two possibilities 1 or −1. For non-real characters, X = X and w is some algebraic number of norm 1.
Recall from the introduction that an Artin L-function is said to satisfy the Riemann hypothesis if all its zeros in the critical strip 0 < Re(s) < 1 are actually on the critical line Re(s) = 1/2. We will be using the Riemann hypothesis through Lemma 3.1. If we replaced the function (3.1) with the appropriately scaled version of (5.17) from [PM11], then our lower bounds would be only conditional on the Artin conjecture, which is completely known for some Galois types (G, c, χ). However the bounds obtained would be much smaller, and the comparison with first conductors as presented in Tables 8.1-8.8 below would be less interesting.
2.8. Rational characters and rational Artin L-functions. The abelianized Galois group G ab acts on continuous complex characters of profinite groups through its action on their values. If X and X are conjugate via this action then their conductors agree: Our study is simplified by this equality because it allows us to study a given irreducible character X by studying instead the rational character X obtained by summing its conjugates. By the Artin induction theorem [Fei67,Prop. 13.2], a rational character X can be expressed as a rational linear combination of permutation characters: For general characters X , computing the Frobenius traces a p = X (Fr p ) requires the results of [DD13]. Similarly the computation of bad Euler factors and the root number w present difficulties. For Frobenius traces and bad Euler factors, these complications are not present for rational characters X because of (2.5).

Signature-based analytic lower bounds
Here and in the next section we aim to be brief, with the main point being to explain how type-based lower bounds are usually much larger than signature-based lower bounds. We employ the standard framework for establishing lower bounds for conductors and discriminants, namely Weil's explicit formula. General references for the material in this section are [Odl77b,PM11].
3.1. Basic quantities. The theory allows general test functions that satisfy certain axioms. We work only with a function introduced by Odlyzko (see [Poi77b,(9)]), The simplifications in the integrals for N (z) and R(z) are fairly standard and apply to any test function, with the exception of the final steps which make use of the support for f (x). Evaluation of P (z) depends on the choice of f (x). The integrals for N (z) and R(z) cannot be evaluated in closed form like the third, but, as indicated in [Poi77b,§2], they do have simple limits N (∞) = log(8π) + γ and R(∞) = π/2 as z → ∞. Here γ ≈ 0.5772 is the Euler γ-constant. The constants Ω = e N (∞) ≈ 44.7632 and e R(∞) ≈ 4.8105, as well as their product Θ = e N (∞)+R(∞) ≈ 215.3325, will play important roles in the sequel.
3.2. The quantity M (n, r, u). Consider triples (n, r, u) of real numbers with n and u positive and r ∈ [−n, n]. For such a triple, define It is clear that M (n, r, u) = M (n/u, r/u, 1). Accordingly we regard u = 1 as the essential case and abbreviate M (n, r) = M (n, r, 1). For fixed ∈ [0, 1] and u > 0, one has the asymptotic behavior  3.3. Lower bounds for root discriminants. Suppose that Φ is a nonzero Artin character which takes real values only. We say that Φ is nonnegative if This nonnegativity ensures that the inner product (Φ, 1) of Φ with the unital character 1 is positive. A central result of the theory, a special case of the statement in [Poi77b, (7)], then serves us as a lemma.
If Φ is a permutation character, then the nonnegativity condition (3.3) is automatically satisfied. This makes the application of the analytic theory to lower bounds of root discriminants of fields relatively straightforward. 3.4. Lower bounds for general Artin conductors. To pass from nonnegative characters to general characters, the classical method uses the following lemma.
Lemma 3.2 (Odlyzko [Odl77b]). The conductor relation holds for any degree n character X and its absolute square Φ = |X | 2 .
A proof of this lemma from first principles is given in [Odl77b]. Combining Lemma 3.1 with Lemma 3.2 one gets the following immediate consequence is valid for all characters X with (X (e), X (σ), (X , X )) = (n, r, w) such that L(|X | 2 , s) satisfies the Artin conjecture and the Riemann hypothesis.
This theorem is essentially the main result in the literature on lower bounds for Artin conductors. It appears in [Odl77b,PM11] with the right side replaced by explicit bounds. For fixed ∈ [−1, 1] and w > 0, one has the asymptotic behavior The bases Ω ≈ 44.7632 and Θ ≈ 215.3325 of (3.2) serve as limiting lower bounds for root discriminants via Lemma 3.1. However it is only their square roots √ Ω ≈ 6.6905 and √ Θ ≈ 14.6742 which Theorem 3.3 gives as limiting lower bounds for root conductors. This discrepancy will be addressed in Section 10.

Type-based analytic lower bounds
In this section we establish Theorem 4.2, which is a family of lower bounds on the root conductor δ X of a given Artin character, dependent on the choice of an auxiliary character φ.
4.1. Conductor relations. Let G be a finite group, c an involution in G, χ a faithful character of G, and φ a non-zero real-valued character of G. Say that a pair of Artin characters (X ,Φ) has joint type (G, c, χ, φ) if there is a surjection Write the conductors respectively as Just as in the last section, we need lower bounds on D X in terms of D Φ . Our paper [JR14b] produces bounds of this sort in the context of many characters.
Here we present some of these results restricted to the setting of two characters, but otherwise following the notation of [JR14b]. For τ ∈ G, letτ be its order. Let ψ be a rational character of G. Define two similar numbers, Here ϕ is the Euler totient function given by ϕ(k) = |(Z/k) × |. For the identity element e, one clearly has c e (ψ) = c e (ψ) = 0. Whenτ is prime, the functions on rational characters defined in (4.1) are proportional: (τ − 1) c τ (ψ) =τ c τ (ψ).
The functions c τ and c τ are related to ramification as follows. Let Ψ be an Artin character corresponding to ψ under h. If Ψ is tame at p then for τ corresponding to a generator of tame inertia. The identity (4.2) holds because c τ (ψ) is the number of non-unital eigenvalues of ρ(τ ) for a representation ρ with character ψ. For general Ψ, there is a canonical expansion with always k τ ≥ 0, coming from the filtration by higher ramification groups on the p-adic inertial subgroup of G. Because (4.1)-(4.3) are only correct for ψ rational, when we apply them to characters χ and φ of interest, we are always assuming that χ and φ are rational. As explained in §2.8, the restriction to rational characters still allows obtaining general lower bounds. Also, as will be illustrated by an example in §5.5, focusing on rational characters does not reduce the quality of these lower bounds.
For the lower bounds we need, we define the parallel quantities Let B(G, χ, φ) be the best lower bound, valid for all primes p, that one can make on c p (X )/c p (Φ) by purely local arguments. As emphasized in [JR14b,§2], B(G, χ, φ) can in principle be calculated by individually inspecting all possible p-adic ramification behaviors. The above discussion says The left inequality holds because of the nonnegativity of the k τ in (4.3). The right inequality holds because of (4.2). A central theme of [JR14b] is that one is often but not always in the extreme situation For example, it often occurs in practice that the minimum in the expression (4.4) for α(G, χ, φ) occurs at a τ of prime order. Then the proportionality remark above shows that in fact all three quantities in (4.5) are the same, and so in particular (4.6) holds. As a quite different example, Theorem 7.3 of [JR14b] says that if φ is the regular character of G and χ is a permutation character, then (4.6) holds.
Some other examples of (4.6) are worked out in [JR14b] by explicit analysis of wild ramification; a few examples show that B(G, χ, φ) < α(G, χ, φ) is possible too.

Root conductor relations.
To switch the focus from conductors to root conductors, we multiply all three quantities in (4.5) by φ(e)/χ(e) to obtain Here the elementary purely group-theoretic quantity α(G, χ, φ) is improved to the best bound b(G, χ, φ) which in turn often agrees with a second more complicated but still purely group-theoretic quantity α(G, χ, φ). The notations α, α, α, α are all taken from Section 7 of [JR14b] while the notations B and b correspond to quantities not named in [JR14b]. Our discussion establishes the following lemma.
Lemma 4.1. The conductor relation holds for all pairs of Artin characters (X , Φ) with joint type of the form (G, c, χ, φ).

Bounds via an auxiliary Artin character
Just like Lemma 3.1 combined with Lemma 3.2 to give Theorem 3.3, so too Lemma 3.1 combines with Lemma 4.1 to give the following theorem.
Theorem 4.2. The lower bound is valid for all character pairs (X , Φ) of joint type (G, c, χ, φ) such that Φ is nonnegative and L(Φ, s) satisfies the Artin conjecture and the Riemann hypothesis.
Computing the right side of (4.10) is difficult because the base in (4.9) requires evaluating the maximum of a complicated function, while the exponent b(G, χ, φ) involves an exhaustive study of wild ramification. Almost always in the sequel, χ and φ are rational-valued and we replace b(G, χ, φ) by α(G, χ, φ); in the common case that all three quantities of (4.7) are equal, this is no loss.

Four choices for φ
This section fixes a type (G, c, χ) where the faithful character χ is rational-valued and uses the notation (n, r) = (χ(e), χ(c)). The section introduces four nonnegative characters φ i built from (G, χ). For the first character φ L , it makes m(G, c, χ, φ L , b), the lower bound appearing in Theorem 4.2, more explicit. For the remaining three characters φ i , it makes the perhaps smaller quantity m(G, c, χ, φ i , α) more explicit.
Two simple quantities enter into the constructions as follows. Let X be the set of values of χ, so that X ⊂ Z by our rationality assumption. Let − χ be the least element of X. The greatest element of X is of course χ(e) = n, and we let χ be the second greatest element. Thus, − χ < 0 ≤ χ ≤ n − 1.

Linear auxiliary character. A simple nonnegative character associated to
The character φ L seems most promising as an auxiliary character when χ is very small. In [PM11, §3] the auxiliary character χ + n is used, which has the advantage of being nonnegative for any rational character χ. Odlyzko also uses χ+n in [Odl77b], and suggests using the auxiliary character φ L = χ + χ since it gives a better bound whenever χ < n. This strict inequality holds exactly when the center of G has odd order.
5.2. Square auxiliary character. Another very simple nonnegative character associated to χ is φ S = χ 2 . This character gives The derivation of (5.2) uses the simple formula in (4.1) for c τ . The formula for c τ in (4.1) is more complicated, and we do not expect a simple general formula for m(G, c, χ, φ S , α), nor for the best bound m(G, c, χ, φ S , b) in Theorem 4.2. The character φ S is used prominently in [Odl77b,PM11]. When χ = n − 2, the lower bound m(G, c, χ, φ S , α) coincides with that of Lemma 3.2. Thus for χ = n−2, Theorem 4.2 with φ = φ S gives the same bound as Theorem 3.3. On the other hand, as soon as χ < n − 2, Theorem 4.2 with φ = φ S is stronger. The remaining case χ = n−1 occurs only three times among the 195 characters we consider in Section 8. In these three cases, the bound in Theorem 3.3 is stronger because the exponent is larger. However, in each of these cases, the tame-wild principle applies [JR14b] and we can use exponent m(G, c, χ, φ S , α), which gives the same bound as Theorem 3.3 in two cases, and a better bound in the third. 5.3. Quadratic auxiliary character. Let − χ be the greatest negative element of the set X of values of χ. A modification of the given character χ is χ * = χ + χ, with degree n * = n+ χ and signature r * = r + χ. A modification of φ S is φ Q = χχ * . The function φ Q takes only nonnegative values because the interval (−χ, 0) in the x-line where x(x +χ) is negative is disjoint from the set X of values of χ. The lower bound associated to φ Q is Comparing formulas (5.2) and (5.3), n 2 strictly increases to nn * and n/(n + χ) increases to n * /(n * + χ). In the totally real case n = r, the monotonicity of the function M (n, n) as exhibited in the right edge of Figure 3.1 then implies that m(G, c, χ, φ S , α) strictly increases to m(G, c, χ, φ Q , α). Even outside the totally real setting, one can expect that φ Q almost always yields a better lower bound than φ S . The character φ Q seems promising as an auxiliary character when χ is very small so that the exponent is near 1 rather than its lower limit of 1/2. As for the square case, we do not expect a simple formula for the best bound m(G, c, χ, φ Q , b) in Theorem 4.2.
5.4. Galois auxiliary character. Finally there is a strong candidate for a good auxiliary character that does not depend on χ, namely the regular character φ G . By definition, φ G (e) = |G| and else φ G (g) = 0. In this case one has Here δ ce is defined to be 1 in the totally real case and 0 otherwise. This auxiliary character again seems most promising when χ is small. As in the square and quadratic cases, we do not expect a simple formula for m(G, c, χ, φ G , b).

Spectral bounds and rationality.
To get large lower bounds on root conductors, one wants χ/n to be small for (5.1) or χ/n to be small for (5.2)-(5.4). The analogous quantities χ 1 /n 1 and χ/n 1 are well-defined for a general real character χ 1 , and replacing χ 1 by the sum χ of its conjugates can substantially reduce them. For example, let p be a prime congruent to 1 modulo 4, and let G be the simple group PSL 2 (p). Then G has two irrational irreducible characters, say χ 1 and χ 2 , both of degree (p + 1)/2. For each, its set of values is (except that 1 is missing if p = 5). However for χ = χ 1 + χ 2 , the set of values is just {−2, 0, 2, p + 1}. Thus in passing from χ 1 /n 1 to χ/n, one saves a factor of √ p + 1.
Similarly in passing from χ 1 /n 1 to χ/n, one saves a factor of √ p − 1.

Other choices for φ
To apply Theorem 4.2 for a given Galois type (G, c, χ), one needs to choose an auxiliary character φ. We presented four choices in Section 5. We discuss all possible choices here, using G = A 4 and G = A 5 as illustrative examples.
6.1. Rational character tables. As a preliminary, we review the notion of rational character table. Let G = {C j } j∈J be the set of power-conjugacy classes in G. Let G rat = {χ i } i∈I be the set of rationally irreducible characters. These sets have the same size k and one has a k × k matrix χ i (C j ), called the rational character table. Two examples are given in Table 6.1. We index characters by their degree, with I = {1, 2, 3} for A 4 and I = {1, 4, 5, 6} for A 5 . All characters are absolutely irreducible except for χ 2 and χ 6 , which each break as a sum of two conjugate irreducible complex characters. We likewise index power-conjugacy classes by the order of a representing element, always adding letters as is traditional. Thus J = {1A, 2A, 3AB} for A 4 and J = {1A, 2A, 3A, 5AB} for A 5 , with 3AB and 5AB each consisting of two conjugacy classes.
6.2. The polytope P G of normalized nonnegative characters. A general realvalued function φ ∈ R(G ) has an expansion The coefficients are recovered via inner products, x i = (φ, χ i )/(χ i , χ i ). Alternative coordinates are given by y j = φ(C j ). The φ allowed for Theorem 4.2 are the nonzero φ with the x i and y j non-negative integers.
An allowed φ gives the same lower bound in Theorem 4.2 as any of its positive multiples mφ. Without getting any new bounds, we can therefore give ourselves the convenience of allowing the x i and y j to be nonnegative rational numbers. Similarly, we can extend by continuity to allow the x i and y j to be nonnegative real numbers. The allowed φ then become the cone in k-dimensional Euclidean space given by x i ≥ 0 and y j ≥ 0, excluding the tip of the cone at the origin.
Writing the identity character as χ 1 , we can normalize via scaling to x 1 = 1. Writing the identity class as C 1A , the inequality y 1A ≥ 0 is implied by the other y j ≥ 0 and so the variable y 1A can be ignored. The polytope P G of normalized nonnegative characters is then defined by x 1 = 1, the inequalities x i ≥ 0 for i = 1, and inequalities y j ≥ 0 for j = 1A. The point where all the x i are zero is the unital character φ 1 . The point where all the y j are zero is the regular character φ G . Thus the (k − 1)-dimensional polytope P G is determined by 2k − 2 linear inequalities, with k − 1 corresponding to non-unital characters and intersecting at φ 1 , and k − 1 corresponding to non-identity classes and intersecting at φ G .   In this view, the point φ G = φ 60 = (4, 5, 3) should be considered as closest to the reader, with the solid lines visible and the dashed lines hidden by the polytope. Note that P A4 has the combinatorics of a square and P A5 has the combinatorics of a cube. While the general P G is the intersection of an orthant with tip φ 1 and an orthant with tip φ G , its combinatorics are typically more complicated than [0, 1] (k−1) . For example, the groups G = A 6 , S 5 , A 7 , and S 6 , have k = 6, 7, 8, and 11 respectively; but instead of having 32, 64, 128 and 1024 vertices, their polytopes P G have 28, 40, 115, and 596 vertices respectively. 6.3. Points in P G . In the previous subsection, we have mentioned already the distinguished vertices φ 1 and φ G . For every rationally irreducible character, we also have φ χ,L = χ + χ, φ χ,S = χ 2 , and φ χ,Q = χχ * , as in Section 5.
For every subgroup H of G, another element of P G is the permutation character φ G/H . For H = G, this character is just the φ 1 considered before, which is a vertex. Otherwise, a theorem of Jordan, discussed at length in [Ser03], says that φ G/H (C j ) = 0 for at least one j; in other words, φ G/H is on at least one character face. For A 4 and A 5 , there are respectively five and nine conjugacy classes of subgroups, distinguished by their orders. Figures 6.2 draws the corresponding points, labeled by φ |G/H| . All four vertices of P A4 and six of the eight vertices of P A5 are of the form φ N . The remaining one φ N in P A4 is on an edge, while the remaining three φ N in P A5 are on edges as well.
6.4. The best choice for φ. Given (G, c, χ) and u ∈ { α, b, α}, let m(G, c, χ, u) = max φ∈P G m(G, c, χ, φ, u). Computing these maxima seems difficult. Instead we vary φ over a modestly large finite set, denoting the largest bound appearing as d(G, c, χ, u). For most G, the set of φ we inspect consists of all φ χ,L , φ χ,S , and φ χ,Q , all φ G/H including the regular character φ G , and all vertices. For some G, like S 7 , there are too many vertices and we exclude them from the list of φ we try.
For each (G, χ), we work either with u = α or with u = α, as explained in the "middle four columns" part of §8.2.2. We then report d(G, χ) = min c d(G, c, χ, u) in Section 8.

The case G = S 5
Our focus in the next two sections is on finding initial segments L(G, χ; B) of complete lists of Artin L-functions, and in particular on finding the first root conductor δ 1 (G, χ). It is a question of transferring completeness statements for number fields to completeness statements for Artin L-functions via conductor relations. In this section, we explain the process by presenting the case G = S 5 in some detail. 7.1. Different orders on the same set of fields. Consider the set K of isomorphism classes of quintic fields K over Q with splitting field L/Q having Galois group Gal(L/Q) ∼ = S 5 . The group S 5 has seven irreducible characters which we index by degree and an auxiliary label: χ 1a = 1, χ 1b , χ 4a , χ 4b , χ 5a , χ 5b , and χ 6a . For φ a permutation character, let D φ (K) = D(K φ ) be the absolute discriminant of the associated resolvent algebra K φ of K. Extending by multiplicativity, functions D χ : K → R >0 are defined for general χ = m n χ n . They do not depend on the coefficient m 1a . We follow our practice of often shifting attention to the corresponding root conductors δ χ (K) = D χ (K) 1/χ(e) .
The orderings coming from different δ χ can be very different. For example, consider the field K ∈ K defined by the polynomial x 5 −2x 4 +4x 3 −4x 2 +2x−4. This field is the first field in K when ordered by the regular character φ 120 = n χ n (n)χ n . However it is the 22 nd field when ordered by φ 6 = 1 + χ 5b only the 2298 th field when ordered by φ 5 = 1 + χ 4a .
This phenomenon of different orderings on the same set of number fields plays a prominent role in asymptotic studies [Woo10]. Here we are interested instead in initial segments and how they depend on χ. Our formalism lets us treat any χ. Following the conventions for general G of the next section, we focus on the five irreducible χ with χ(e) > 1, thus χ n for n ∈ {4a, 4b, 5a, 5b, 6a}. 7.2. Computing Artin conductors. To compute general D χ (K), one needs to work with enough resolvents of K = K 5 = Q[x]/f 5 (x). For starters, we have the quadratic resolvent K 2 = Q[x]/(x 2 −D(K 5 )) and the Cayley-Weber resolvent K 6 = Q[x]/f 6 (x) [JLY02,JR14b]. The other resolvents we will need are K 10 = K 5 ⊗ K 2 , K 12 = K 2 ⊗K 6 , and K 30 = K 5 ⊗K 6 . Defining polynomials are obtained for K a ⊗K b by the general formula where f a (x) has roots α i and f b (x) has roots β j . So discriminants D 2 , D 5 , D 6 , D 10 , D 12 , D 30 are easily computed.
From the character table, the permutation characters φ N in question are expressed in the basis χ n as on the left in the following display. Inverting, one gets the χ n in terms of the φ N as on the right.
Conductors D n belonging to the χ n are calculable through these formulas, as e.g.
. For all the groups G considered in the next section, we proceeded similarly. Thus we started with rational character tables from Magma. We used linear algebra to express rationally irreducible characters in terms of permutation characters. We used Magma again to compute resolvents and then Pari to evaluate their discriminants. In this last step, we often confronted large degree polynomials with large coefficients. The discriminant computation was only feasible because we knew a priori the set of primes dividing the discriminant, and could then easily compute the p-parts of the discriminants of these resolvent fields for relevant primes p using Pari/gp without fully factoring the discriminants of the resolvent polynomials.
Magma's Artin representation package computes conductors of Artin representations in a different and more local manner. Presently, it does not compute all conductors in our range because some decomposition groups are too large. 7.3. Transferring completeness results. As an initial complete list of fields, we take K(φ; 85) with φ = φ G = φ 120 . We know from [JR14a] that this set consists of 2080 fields. We list these fields by increasing discriminant, K 1 , . . . , K 2080 , with the resolution of ties conveniently not affecting the explicit results appearing in Table 8.1.
The quantities of Section 4 reappear here, and we will use the abbreviations α(n) = α(S 5 , χ n , φ) and α(n) = α(S 5 , χ n , φ). Since φ is zero outside of the identity class, the formulas simplify substantially: For each of the five n, the classes contributing to the minima are in bold on Table 7.1. So, extremely simply, for computing α(n) on the left, the largest χ n (τ ) besides χ n (e) are in bold. For computing α(n) on the right, the c τ (χ n ) with c τ (χ n )/c τ (φ) minimized are put in bold. For the group S 5 , one has agreement α(n) = α(n) in all five cases. This equality occurs for 170 of the lines in Tables 8.1-8.8, with the other possibility α(n) < α(n) occurring for the remaining 25 lines.
One has an analogous inclusion for general (G, χ), with φ again the regular character for G. When G satisfies the tame-wild principle of [JR14b], the α in exponents can be replaced by α. The group S 5 does satisfy the tame-wild principle, but in this case the replacement has no effect. The final results are on Table 8.1. In particular for n = 4a, 4b, 5a, 5b, 6a the unique minimizing fields are K 103 , K 21 , K 14 , K 6 , and K 12 , with root conductors approximately 6.33, 18.72, 17.78, 16.27, and 18.18. The lengths of the initial segments identified are 45, 15, 186, 592, and 110. Note that because of the relations φ 5 = 1 + χ 4a and φ 6 = 1 + χ 5b , the results for 4a and 5b are just translation of known minima of discriminants of number fields with Galois groups 5T 5 and 6T 14 respectively. For 4b, 5b, 6a, and the majority of the characters on the tables of the next section, the first root conductor and the entire initial segment are new.

Tables for 84 groups G
In this section, we present our computational results for small Galois types. For simplicity, we focus on results coming from complete lists of Galois number fields. Summarizing statements are given in §8.1 and then many more details in §8.2.

Lower bounds and initial segments.
We consider all groups with a faithful transitive permutation representation in some degree from two to nine, except we exclude the nonsolvable groups in degrees eight and nine. There are 84 such groups, and we consider all associated Galois types (G, χ) with χ a rationally irreducible faithful character. Our first result gives conditional lower bounds: The bounds in Tables 8.1-8.8 are graphed with the best previously known bounds from [PM11] in Figure 8.1. The horizontal axis represents the dimension n 1 = χ 1 (e) of any irreducible constituent χ 1 of χ. The vertical axis corresponds to lower bounds on root conductors. The piecewise-linear curve connects bounds from [PM11], and there is one dot at height d(G, χ) for each (G, χ) from Tables 8.1-8.8 with χ 1 (e) ≤ 20. Here we are freely passing back and forth between a rational character χ and an irreducible constituent χ 1 via δ 1 (G, χ) = δ 1 (G, χ 1 ), which is a direct consequence of (2.4).
Not surprisingly, the type-based bounds are larger. In low dimensions n 1 , some type-based bounds are close to the general bounds, but by dimension 5 there is a clear separation which widens as the dimension grows. This may in part be explained by the fact that we are only seeing a small number of representations for each of these dimensions. However, as we explain in §10.3, we also expect that the asymptotic lower bound of √ Ω ≈ 6.7 [PM11] is not optimal, and that this bound is more likely to be at least Ω ≈ 44.8.

8.2.
Tables detailing results on lower bounds and initial segments. Our tables are organized by the standard doubly-indexed lists of transitive permutation groups mT j, with degrees m running from 2 through 9. Within a degree, the blocks of rows are indexed by increasing j. There is no block to print if mT j has no faithful irreducible characters. For example, there is no block to print for groups having noncyclic center, such as 4T 2 = V = C 2 × C 2 or 8T 9 = D 4 × C 2 . Also the block belonging to mT j is omitted if the abstract group G underlying mT j has appeared earlier. For example G = S 4 has four transitive realization in degrees m ≤ 8, namely 4T 5, 6T 7, 6T 8, and 8T 14; there is correspondingly a 4T 5 line on our tables, but no 6T 7, 6T 8, or 8T 14 lines.
8.2.1. Top row of the G-block. The top row in the G-block is different from the other rows, as it gives information corresponding to the abstract group G. Instead of referring to a faithful irreducible character, as the other lines do, many of its entries are the corresponding quantities for the regular character φ G . The first four entries are a common name for the group G (if there is one), the order φ G (e) = |G|, the symbol TW if G is known to have the universal tame-wild property as defined in [JR14b], and finally k, N . Here, k is the size of the rational character table, and N is number of vertices of the polytope P G discussed in §6.2, or a dash if we did not compute N . The last four entries are the smallest root discriminant of a Galois G field, the factored form of the corresponding discriminant, a cutoff B for which the set K(G; B) is known, and the size |K(G; B)|.

8.2.2.
Remaining rows of the G-block. Each remaining line of the G-block corresponds to a type (G, χ). However the number of rows in the G-block is typically substantially less than the number of faithful irreducible characters of G, as we list only one representative of each Gal(Q/Q) × Out(G) orbit of such characters. As an example, S 6 has eleven characters, all rational. Of the nine which are faithful, there are three which are fixed by the nontrivial element of Out(S 6 ) and the others form three two-element orbits. Thus the S 6 -block has six rows. In general, the information on a (G, χ) row comes in three parts, which we now describe in turn.
First four columns. The first column gives the lexicographically first permutation group mT j for which the corresponding permutation character has χ as a rational constituent. Then n 1 = χ 1 (e) is the degree of an absolutely irreducible character χ 1 such that χ is the sum of its conjugates. The number n 1 is superscripted by the size of the Out(G) orbit of χ, in the unusual case when this orbit size is not 1. Next, the complex number z is a generator for the field generated by the values of the character χ 1 , with no number printed in the common case that χ 1 is rationalvalued. The last entry gives the interval [− χ, χ], where χ and χ are the numbers introduced in the beginning of Section 5. In the range presented, the data of mT j, n 1 , z, and [− χ, χ] suffice to distinguish Galois types (G, χ) from each other. Middle four columns. The next four columns focus on minimal root conductors. In the first entry, d is the best conditional lower bound we obtained for root conductors, and the subscript i ∈ { , s, q, g, p, v} gives information on the corresponding auxiliary character φ. The first four possibilities refer to the methods of Section 5, namely linear, square, quadratic, and Galois. The last two, p and v, indicate a permutation character and a character coming from a vertex of the polytope P G . The best φ of the ones we inspect is always at a vertex, except in the three cases on Table 8.2 where * is appended to the subscript. Capital letters S, Q, G, P , and V also appear as subscripts. These occur only for groups marked with TW, and indicate that the tame-wild principle improved the lower bound. For most groups with fifteen or more classes, it was prohibitive to calculate all vertices, and the best of the other methods is indicated.
When the second entry is in roman type, it is the minimal root conductor and the third entry is the minimal conductor in factored form. When the second entry is in italic type, then it is the smallest currently known root conductor. The fourth entry gives the position of the source number field on the complete list ordered by Galois root discriminant. This information lets readers obtain further information from [JR14a], such as a defining polynomial and details on ramification.
Last three columns. The quantity β is the exponent we are using to pass from Galois number fields to Artin representations. Writing α = α(G, χ, φ G ) and α = α(G, χ, φ G ), one has the universal relation α ≤ α. When equality holds then the common number is printed. To indicate that inequality holds, an extra symbol is printed. When we know that G satisfies TW then we can use larger exponent and α • is printed. Otherwise we use the smaller exponent and α • is printed. The column Finally the column # gives |L(G, χ; B β )|, the length of the complete list of Artin L-functions we have identified. For the L-functions themselves, we refer to [LMF]. In this section, we discuss four topics, each of which makes specific reference to parts of the tables of the previous section. Each of the topics also serves the general purpose of making the tables more readily understandable. 9.1. Comparison of first Galois root discriminants and root conductors. Suppose first, for notational simplicity, that G is a group for which all irreducible complex characters take rational values only. When one fixes K gal with Gal(K gal /Q) ∼ = G and lets χ runs over all the irreducible characters of G, the root . Deviation of a root conductor δ χ from δ Gal is caused by nonzero values of χ. When χ(e) is large and [− χ, χ] is small, δ χ is necessarily close to δ Gal . One can therefore think generally of δ Gal as a first approximation to δ χ . The general principle of δ Gal approximating δ χ applies to groups G with irrational characters as well.
Our first example of S 5 illustrates both how the principle δ Gal ≈ δ χ is reflected in the tables, and how it tends to be somewhat off in the direction that δ Gal > δ χ . For a given K gal , the variance of its δ χ about its δ Gal is substantial and depends on the details of the ramification in K gal . There are many K gal with root discriminant near the minimal root discriminant, all of which are possible sources of minimal root conductors. It is therefore expected that the minimal conductors δ 1 (S 5 , χ) = min δ χ printed in the table, 6.33, 18.72, 16.27, 17.78, and 18.18, are substantially less than the printed minimal root discriminant δ 1 (S 5 , φ 120 ) ≈ 24.18. As groups G get larger, one can generally expect tighter clustering of the δ 1 (G, χ) about δ 1 (G, φ G ). One can see the beginning of this trend in our partial results for S 6 and S 7 . 9.2. Known and unknown minimal root conductors. Our method of starting with a complete list of Galois fields is motivated by the principle from the previous subsection that the Galois root discriminant δ Gal is a natural first approximation to δ χ . Indeed, as the tables show via nonzero entries in the # column, this general method suffices to obtain a non-empty initial segment for most (G, χ). As our focus is primarily on the first root conductor δ 1 = δ 1 (G, χ), we do not pursue larger initial segments in these cases. When the initial segment from our general method is empty, as reported by a 0 in the # column, we aim to nonetheless present the minimal root conductor δ 1 . Suppose there are subgroups H m ⊂ H k ⊆ G, of the indicated indices, such that a multiple of the character χ of interest is a difference of the corresponding permutation characters: cχ = φ m − φ k . Suppose one has the complete list of all degree m fields corresponding to the permutation representation of G on G/H m and root discriminant ≤ B. Then one can extract the complete list of L(G, χ; B m/(m−k) ) of desired Artin L-functions.
For example, consider χ 5 , the absolutely irreducible 5-dimensional character of A 6 . The permutation character for a corresponding sextic field decomposes φ 6 = 1 + χ 5 , and so the discriminant of the sextic field equals the conductor of χ 5 . As an example with k > 1, consider the 6-dimension character χ 6 for C 3 C 3 = 9T 17, which is the sum of a three-dimensional character and its conjugate. The nonic field has a cubic subfield, and the characters are related by φ 9 = φ 3 + χ 6 . In terms of conductors, D 9 = D 3 · D χ6 , where D 9 and D 3 are field discriminants. So, we can determine the minimal conductor of an L-function with type (C 3 C 3 , χ 6 ) from a sufficiently long complete list of nonic number fields with Galois group C 3 C 3 .
This method, applied to both old and newer lists presented in [JR14a], accounts for all but one of the δ 1 reported in Roman type on the same line as a 0 in the # column. The remaining case of an established δ 1 is for the type (GL 3 (2), χ 7 ). The group GL 3 (2) appears on our tables as 7T 5. The permutation representation 8T 37 has character χ 7 + 1. Here the general method says that L(GL 3 (2), χ 7 ; 26.12) is empty. It is prohibitive to compute the first octic discriminant by searching among octic polynomials. In [JR] we carried out a long search of septic polynomials, examining all local possibilities giving an octic discriminant at most 30. This computation shows that |L(GL 3 (2), χ 7 ; 48.76)| = 25 and in particular identifies δ 1 = 21 8/7 ≈ 32.44.
The complete lists of Galois fields for a group first appearing in degree m were likewise computed by searching polynomials in degree m, targeting for small δ Gal . This single search can give many first root conductors at once. For example, the largest groups on our octic and nonic tables are S 4 S 2 = 8T 47 and S 3 S 3 = 9T 31. In these cases, minimal root conductors were obtained for 5 of the 10 and 7 of the 12 faithful χ respectively. Searches adapted to a particular character χ as in the previous paragraph can be viewed as a refinement of our method, with targeting being not for small δ Gal but instead for small δ χ . Many of the italicized entries in the column δ 1 seem improvable to known minimal root conductors via this refinement. 9.3. The ratio δ 1 /d. In all cases on the table, δ 1 > d. Thus, as expected, we did not encounter a contradiction to the Artin conjecture or the Riemann hypothesis. In some cases on the table, the ratio δ 1 /d is quite close to 1. As two series of examples, consider S m with its reflection character χ m−1 = φ m − 1, and D m and the sum χ of all its faithful 2-dimensional characters. Then these ratios are as follows: In the cases with the smallest ratios, the transition from no L-functions to many L-functions is commonly abrupt. For example, in the case (S 7 , χ 6 ) the lower bound is d ≈ 7.50 and the first seven rounded root conductors are 7.55, 7.60, 7.61, 7.62, 7.64, 7.66, and 7.66. When the translation from no L-functions to many L-functions is not abrupt, but there is an L-function with outlyingly small conductor, again δ 1 /d may be quite close to 1. As an example, for (8T 25, χ 7 ), one has d ≈ 16.10 and δ 1 = 29 6/7 ≈ 17.93 yielding δ 1 /d ≈ 1.11. However in this case the next root conductor is δ 2 = 113 6/7 ≈ 57.52, yielding δ 2 /d ≈ 3.57. Thus the close agreement is entirely dependent on the L-function with outlyingly small conductor. Even the second root conductor is somewhat of an outlier as the next three conductors are 71.70, 76.39, and 76.39, so that already δ 3 /d ≈ 4.45.
There are many (G, χ) on the table for which the ratio δ 1 /d is around 2 or 3. There is some room for improvement in our analytic lower bounds, for example changing the test function (3.1), varying φ over all of P G , or replacing the exponent α with the best possible exponent b. However examples like the one in the previous paragraph suggest to us that in many cases the resulting increase in d towards δ 1 would be very small. 9.4. Multiply minimal fields. Tables 8.1-8.8 make implicit reference to many Galois number fields, and all necessary complete lists are accessible on the database [JR14a]. Table 9.1 presents a small excerpt from this database by giving six polynomials f (x). For each f (x), Table 9.1 first gives the Galois group G and the root discriminant δ of the splitting field K gal . We are highlighting these particular Galois number fields K gal here because they are multiply minimal: they each give rise to the minimal root conductor for at least two different rationally irreducible characters χ. The degrees of these characters are given in the last column of Table 9.1.

Lower bounds in large degrees
In this section, we continue our practice of assuming the Artin conjecture and Riemann hypothesis for the relevant L-functions. For n a positive integer, let ∆ 1 (n) be the smallest root discriminant of a degree n field. As illustrated by Figure 3.1 Corollary 10.1. Let (G k , χ k ) be a sequence of rationally irreducible Galois types of degree n k = χ k (e) . Suppose that the number of irreducible constituents (χ k , χ k ) is bounded, n k → ∞, and either A: χ k /n k → 0, or B: χ k /n k → 0. Then, assuming the Artin conjecture and Riemann hypothesis for relevant Lfunctions, Proof. For Case A, Theorem 4.2 using a linear auxiliary character as in (5.1) says δ 1 (G k , χ k ) ≥ M n k χ k + 1, r k χ k + 1, (χ k , χ k ) For Case B, Theorem 4.2 using a Galois auxiliary character as in (5.4) says δ 1 (G k , χ k ) ≥ M (|G k |, 0, (χ k , χ k )) 1− χ/n k .
In both cases, the first argument of M tends to infinity, the second argument does not matter, the third argument does not matter either by boundedness, and the exponent tends to 1. By (3.2), these right sides thus have an infimum limit of at least Ω, giving the conclusion (10.3).
For the proof of Case B, the square auxiliary character would work equally well through (5.2). Also  Table 10.1 summarizes four sequences which we discuss together with some related sequences next.
Spin 2 k 0 2 k 2 1+2k 2 k .S k Reflection k k − 2 k 2 k k! 10.2.1. The group PGL 2 (k) and its characters of degree k − 1, k, and k + 1. In the sequence (PGL 2 (k), χ k ) from the first line of Table 10.1, the index k is restricted to be a prime power. The permutation character φ k+1 arising from the natural action of PGL 2 (k) on P 1 (F k ) decomposes as 1 + χ k where χ k is the Steinberg character. Table 10.1 says that the ratios χ k /n k and χ k /n k are both 1/k, so Corollary 10.1 applies through both Hypotheses A and B. The conductor of χ k is the absolute discriminant of the degree k + 1 number field with character φ k+1 . Thus, in this instance, (10.3) is already implied by the classical (10.1). However, the other nonabelian irreducible characters χ of PGL 2 (k) behave very similarly to χ k . Their dimensions are in {k − 1, k, k + 1} and their values besides χ(e) are all in [−2, 2]. So suppose for each k, an arbitrary nonabelian rationally irreducible character χ k of PGL 2 (k) were chosen, in such a way that the sequence (χ k , χ k ) is bounded. Then Corollary 10.1 would again apply through both Hypotheses A and B. But now the χ k are not particularly closely related to permutation characters. 10.2.2. The group S k and its canonical characters. As with the last example, the permutation character φ k arising from the natural action of S k on {1, . . . , k} decomposes as 1 + χ k where χ k is the reflection character with degree k − 1. The second line of Table 10.1 shows that Corollary 10.1 applies through Hypothesis A. In fact, using the linear auxiliary character underlying Hypothesis A here is essential; the limiting lower bound coming from the square or quadratic auxiliary characters is √ Ω, and this lower bound is just 1 from the Galois auxiliary character. Again in parallel to the previous example, the familiar sequence (S k , χ k ) of types needs to be modified to make it a good illustration of the applicability of Corollary 10.1. Characters of S k are most commonly indexed by partitions of k, with χ (k) = 1, χ (k−1,1) being the reflection character, and χ (1,1,...,1,1) being the sign character. However an alternative convention is to include explicit reference to the degree k and then omit the largest part of the partition, so that the above three characters have the alternative names χ k,() , χ k,(1) , and χ k,(1,...,1,1) . With this convention, one can prove that for any fixed partition µ of a positive integer m, the sequence of types (G k , χ k,µ ) satisfies Hypothesis A but not B.
The case of general µ is well represented by the two cases where m = 2. In these two cases, information in the same format as Table (10.1) is Let X k,m be the S k -set consisting of m-tuples of distinct elements of {1, . . . , k}. Then its permutation character φ k,m decomposes into χ k,µ with µ a partition of an integer ≤ m. These formulas are uniform in k, as in φ k,2 = χ k,(1,1) + χ k,(2) + 2χ k,(1) + χ k,() .
For µ running over partitions of a large integer m, the characters χ k,µ can be reasonably regarded as quite far from permutation characters, and they thus serve as a better illustration of Corollary 10.1. The sequences (S k , χ k,µ ) satisfy Hypothesis A but not B, because n k and χ k grow polynomially as k m , while χ k grows polynomially with degree < m.
Let G k be the extra-special 2-group of type and order 2 1+2k , so that 2 1+2 + and 2 1+2 − are the dihedral and quaternion groups respectively. These groups each have exactly one irreducible character of degree larger than 1, this degree being 2 k . There are just three character values, −2 k , 0, and 2 k . For these two sequences, Corollary 10.1 again applies, but now only through Hypothesis B.
10.2.4. The Weyl group 2 k .S k and its degree k reflection character. The Weyl group W (B k ) ∼ = 2 k .S k of signed permutation matrices comes with its defining degree k character χ k . Here, as indicated by the fourth line of Table 10.1, neither hypothesis of Corollary 10.1 applies.
However the conclusion (10.3) of Corollary 10.1 continues to hold as follows. Relate the character χ k in question to the two standard permutation characters of 2 k .S k via φ 2k = φ k + χ k . For a given 2 k .S k field, D Φ 2k = D Φ k D X k . But, since Φ k corresponds to an index 2 subfield of the degree 2k number field for Φ 2k , we have D 2 Φ k | D Φ 2k . Combining these we get D Φ k | D X k and hence δ Φ k < δ X k . So (10.1) implies (10.3).
10.3. Concluding speculation. As we have illustrated in §10.2.1-10.2.3, both Hypothesis A and Hypothesis B are quite broad. This breadth, together with the fact that the conclusion (10.3) still holds for our last sequence, raises the question of whether (10.3) can be formulated more universally. While the evidence is far from definitive, we expect a positive answer. Thus we expect that the first accumulation point of the numbers δ 1 (G, χ) is at least Ω, where (G, χ) runs over all types with χ irreducible. Phrased differently, we expect that the first accumulation point of the root conductors of all irreducible Artin L-functions is at least Ω.