21.13.02 · number-theory / dirichlet-l-functions-characters

Siegel's Theorem on the Exceptional Zero

shipped3 tiersLean: none

Anchor (Master): Davenport 2000 *Multiplicative Number Theory* 3e §21-22 (the full two-$L$-function proof, ineffectivity, the class-number bound $h(-d) \gg_\varepsilon d^{1/2-\varepsilon}$, the Brauer-Siegel theorem stated); Iwaniec-Kowalski 2004 *Analytic Number Theory* (AMS Colloquium Publications 53) §5.9, §22 (the Landau-Siegel zero, the contrast with Goldfeld-Gross-Zagier); Siegel 1935 *Acta Arithmetica* 1; Goldfeld 1976 *Ann. Scuola Norm. Sup. Pisa*; Gross-Zagier 1986 *Invent. Math.* 84

Intuition Beginner

The zero-free region of 21.13.01 left one stubborn gap: for $L$ -functions built from a real-valued character, a single real zero $β$ might creep up dangerously close to the line where the real part equals one. The closer $β$ sits to one, the worse our control over primes in arithmetic progressions, and the smaller a certain quantity $L (1, χ)$ becomes. Siegel's theorem is the strongest known promise that this cannot happen too badly: it says the zero must stay a measurable distance back from one, and equivalently that $L (1, χ)$ can never be too small.

The promise has a strange and famous flaw. Siegel's proof argues by cases. Either some real character already has a zero very close to one, or none does. In the first case, that one bad character can be used as a lever to push every other character's zero back. In the second case, there was never a problem to begin with. Either way the conclusion holds — but the argument never tells you which case you are in, because nobody has ever found a single one of these bad zeros. The constant in the bound therefore cannot be computed. We know a number exists; we cannot write it down.

This is the price of the theorem. Siegel gives a strong bound with an invisible constant, while an earlier and weaker bound of Page came with a constant you could actually use. The same invisible constant haunts the consequence Siegel really wanted: a lower bound on the class number of imaginary quadratic fields, the count that measures how badly unique factorisation fails. Gauss had asked whether this count grows; Siegel proved it grows nearly like the square root of the discriminant, but could not say how fast.

Visual Beginner

Picture two real characters as two players, each owning one $L$ -function with one possible exceptional zero sitting just left of the line "real part $= 1$ ". Siegel's argument links them: if either player's zero sits very close to the line, that closeness becomes a tool to force the other player's zero back, and to bound the other player's value $L (1, χ)$ from below.

situation	what Siegel's proof does	constant
some character has a near-one zero	use it to bound all the others	exists but unknown
no character has a near-one zero	the bound is direct and clean	could be computed
we do not know which holds	conclusion holds regardless	not computable

The two branches both reach the same conclusion. The reason the constant stays invisible is the bottom row: the proof works without ever deciding the top two rows, and that undecided choice is exactly where the computable constant leaks away.

Worked example Beginner

See how a lower bound on $L (1, χ)$ turns into a lower bound on a class number, with concrete numbers, using the Dirichlet class-number formula $h (- d) = w d L (1, χ) / (2 π)$ with $w = 2$ for $d > 4$ .

Step 1. Take the discriminant $d = 10000$ , so $d = 100$ . Suppose Siegel's bound, at $ε = 1/4$ , hands us $L (1, χ) \geq c \cdot d^{- 1/4}$ with (pretend) $c = 1$ . Then $d^{- 1/4} = 1000 0^{- 1/4} = 0.1$ , so $L (1, χ) \geq 0.1$ .

Step 2. Plug into the class-number formula with $w = 2$ : $$ h(-d) = \frac{2 \cdot 100 \cdot L(1, \chi)}{2\pi} = \frac{100 \cdot L(1, \chi)}{\pi} \geq \frac{100 \cdot 0.1}{\pi} = \frac{10}{\pi} \approx 3.18. $$ Since the class number is a whole number, $h (- d) \geq 4$ .

Step 3. Compare with the exponent. Siegel's bound says $h (- d) ≫ d^{1/2 - 1/4} = d^{1/4} = 1000 0^{1/4} = 10$ . The pretend constant $c = 1$ gave a weaker numerical floor of $4$ here, but the shape $d^{1/4}$ is what matters: as $d$ grows the floor climbs without bound.

What this tells us: a lower bound on the single number $L (1, χ)$ converts, through one multiplication by $d$ , into a lower bound on the class number. The catch is Step 1: the constant $c$ is the invisible one Siegel cannot compute, so the floor " $h \geq 4$ " is real only in principle, not as a number you could certify for this exact $d$ .

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $χ$ is a real primitive Dirichlet character modulo $q$ as in 21.03.02, $L (s, χ) = \sum_{n} χ (n) n^{- s}$ its $L$ -function, $β$ the possible exceptional (Siegel) zero of 21.13.01, $s = σ + i t$ , and $ζ$ the Riemann zeta function. The notation $≫_{ε}$ means " $\geq$ a positive constant depending on $ε$ times", and $c (ε)$ denotes such a constant. We follow Davenport ^{[Davenport §21]}.

Definition (Siegel bound). A real primitive character $χ$ modulo $q$ satisfies the Siegel bound with parameter $ε > 0$ if $$ L(1, \chi) > c(\varepsilon), q^{-\varepsilon} $$ for a positive $c (ε)$ . By the relation $1 - β ≍ L (1, χ) / lo g q$ from 21.13.01, this is equivalent to the exceptional-zero bound $β < 1 - c^{'} (ε) q^{- ε}$ .

Definition (effective versus ineffective constant). A constant $c (ε)$ in such a bound is effective if a finite procedure produces a numerical value of it from $ε$ alone; otherwise it is ineffective. Siegel's $c (ε)$ is ineffective for $ε < 1/2$ : the proof establishes $\exists c (ε) > 0$ without a procedure to compute it. Page's bound $β < 1 - c / (q lo g^{2} q)$ of 21.13.01 has an effective constant but a weaker rate.

Definition (auxiliary product). For two real primitive characters $χ_{1}$ modulo $q_{1}$ and $χ_{2}$ modulo $q_{2}$ (possibly equal), the auxiliary product is $$ F(s) = \zeta(s), L(s, \chi_1), L(s, \chi_2), L(s, \chi_1\chi_2). $$ Its Dirichlet coefficients are non-negative: $F (s) = \sum_{n} a_{n} n^{- s}$ with $a_{n} \geq 0$ and $a_{1} = 1$ , because $F$ is the Dedekind-type zeta function of the genus character data and its logarithm has non-negative coefficients $Λ (n) (1 + χ_{1} (n)) (1 + χ_{2} (n)) / lo g n \geq 0$ .

Definition (class number, imaginary quadratic). For a fundamental discriminant $- d < 0$ with Kronecker character $χ_{- d}$ and $w$ roots of unity ( $w = 2$ for $d > 4$ ), the class number $h (- d)$ of $Q (- d)$ is given by the Dirichlet class-number formula $$ h(-d) = \frac{w\sqrt{d}}{2\pi}, L(1, \chi_{-d}). $$

The symbols $χ$ , $L (s, χ)$ , $β$ , $≫_{ε}$ , $c (ε)$ , the auxiliary product $F$ , $h (- d)$ , and the regulator $R$ used below are recorded in _meta/NOTATION.md.

Counterexamples to common slips

"Siegel's bound is effective because the statement names a definite function $c (ε)$ ." Naming a function does not make it computable. The proof produces $c (ε)$ via a case-split on the existence of an exceptional zero; the value is determined only once that existence is decided, which it never is for $ε < 1/2$ .
"Page's bound and Siegel's bound are the same strength up to constants." They differ in rate, not just constant: Page gives $1 - β ≫ q^{- 1/2} / lo g^{2} q$ , Siegel gives $1 - β ≫_{ε} q^{- ε}$ . For $ε < 1/2$ the latter is far larger, so Siegel pushes $β$ much further from $1$ — but loses the computable constant Page keeps.
"Siegel's class-number bound determines the list of $h (- d) = 1$ discriminants." It does not, precisely because it is ineffective: it shows the list is finite but cannot bound the largest member. Identifying the nine Heegner discriminants required separate effective methods (Heegner, Baker, Stark), and the effective growth $h (- d) ≫ (lo g d)^{1 - ε}$ came later from Goldfeld-Gross-Zagier.

Key theorem with proof Intermediate+

The signature result is Siegel's lower bound, proved by the auxiliary product of two $L$ -functions and the dichotomy of 21.13.01.

Theorem (Siegel, 1935). For every $ε > 0$ there is a constant $c (ε) > 0$ such that for every real primitive character $χ$ modulo $q$ , $$ L(1, \chi) > c(\varepsilon), q^{-\varepsilon}. $$ The constant $c (ε)$ is ineffective for $ε < 1/2$ .

Proof. Fix $ε \in (0, 1/2)$ . Let $χ_{1}$ be a real primitive character modulo $q_{1}$ , and let $χ_{2}$ be any real primitive character modulo $q_{2}$ , with $χ_{1} χ_{2}$ non-principal. Form $F (s) = ζ (s) L (s, χ_{1}) L (s, χ_{2}) L (s, χ_{1} χ_{2})$ , whose Dirichlet coefficients are non-negative with $a_{1} = 1$ , as established for the analogous repulsion product in 21.13.01. The function $F$ is holomorphic except for a simple pole at $s = 1$ with residue $λ = L (1, χ_{1}) L (1, χ_{2}) L (1, χ_{1} χ_{2}) > 0$ .

By a contour argument applied to the non-negative-coefficient series $F$ , for any $σ_{0} \in (1/2, 1)$ there is an absolute constant such that $$ F(\sigma_0) \geq 1 - \frac{c_1 \lambda}{1 - \sigma_0}(q_1 q_2)^{c_2(1 - \sigma_0)}, $$ because the truncated Dirichlet polynomial of $F$ exceeds its leading term $a_{1} = 1$ while the pole at $s = 1$ contributes the negative correction $- λ / (1 - σ_{0})$ , the conductor $q_{1} q_{2}$ entering through the functional-equation bound on the remainder. Now invoke the dichotomy of 21.13.01.

Case A: some real primitive $χ_{1}$ has an exceptional zero $β_{1}$ with $β_{1} > 1 - ε /4$ . Fix this $χ_{1}$ once and for all; it depends only on $ε$ . Since $L (β_{1}, χ_{1}) = 0$ , evaluating $F$ at $σ_{0} = β_{1}$ gives $F (β_{1}) = 0$ , so the displayed inequality forces $λ / (1 - β_{1}) \geq c_{3} (q_{1} q_{2})^{- c_{2} (1 - β_{1})}$ . Because $1 - β_{1} < ε /4$ is fixed and small, $(q_{2})^{- c_{2} (1 - β_{1})} > q_{2}^{- ε /2}$ , and $λ \leq L (1, χ_{1}) L (1, χ_{1} χ_{2}) \cdot L (1, χ_{2}) ≪ (lo g q_{2})^{2} L (1, χ_{2})$ . Rearranging, $$ L(1, \chi_2) \gg \frac{(1 - \beta_1)}{(\log q_2)^2}, q_2^{-\varepsilon/2} \gg_\varepsilon q_2^{-\varepsilon}, $$ the constant depending on the fixed $β_{1}$ .

Case B: no real primitive character has an exceptional zero with $β > 1 - ε /4$ . Then for the given $χ_{2}$ , take any auxiliary $χ_{1}$ with $χ_{1} χ_{2}$ non-principal; evaluating $F$ at $σ_{0} = 1 - ε /4$ , the absence of a zero of $F$ to the right of $σ_{0}$ keeps $F (σ_{0}) > 0$ , and the same inequality, now with $1 - σ_{0} = ε /4$ fixed, yields $λ ≫_{ε} (q_{1} q_{2})^{- ε /2}$ , whence $L (1, χ_{2}) ≫_{ε} q_{2}^{- ε}$ after absorbing the bounded factors $L (1, χ_{1}), L (1, χ_{1} χ_{2}) ≪ lo g (q_{1} q_{2})$ .

In both cases $L (1, χ_{2}) ≫_{ε} q_{2}^{- ε}$ , and $χ_{2}$ was an arbitrary real primitive character, so the bound holds for all of them. The constant is ineffective because the proof does not decide which case obtains: in Case A the constant depends on the value $β_{1}$ of a zero whose existence is hypothesised but never exhibited, and one cannot certify that Case B holds without ruling out every exceptional zero. $□$

Bridge. Siegel's theorem builds toward the Siegel-Walfisz theorem on primes in arithmetic progressions and appears again in 21.13.03, where the ineffective constant propagates into the range $q \leq (lo g x)^{A}$ of uniform validity. The foundational reason the two- $L$ -function product works is exactly the non-negativity of its coefficients, $Λ (n) (1 + χ_{1} (n)) (1 + χ_{2} (n)) \geq 0$ , the same device that gave Landau's repulsion in 21.13.01: this is exactly the repulsion phenomenon used in reverse, where a near-one zero of one character is converted into a lower bound for another. The central insight is that the case-split is unavoidable, and putting these together the bridge is the dichotomy itself — an exceptional zero, if it exists, is a lever; if it does not, there was nothing to bound — and the ineffectivity is the structural cost of never deciding the lever's existence, which generalises into every consequence the theorem touches.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Show that the residue of $F (s) = ζ (s) L (s, χ_{1}) L (s, χ_{2}) L (s, χ_{1} χ_{2})$ at $s = 1$ equals $L (1, χ_{1}) L (1, χ_{2}) L (1, χ_{1} χ_{2})$ , assuming $χ_{1}, χ_{2}, χ_{1} χ_{2}$ are all non-principal.

Hint

Only $ζ$ has a pole at $s = 1$ , of residue $1$ ; the three $L$ -factors are holomorphic there.

Answer

For non-principal characters, each $L (s, χ_{i})$ and $L (s, χ_{1} χ_{2})$ extends holomorphically to $s = 1$ with finite value $L (1, χ_{i}) \neq = 0$ (the non-vanishing of 21.03.02). The only singular factor is $ζ (s)$ , with a simple pole at $s = 1$ of residue $1$ : $ζ (s) = 1/ (s - 1) + O (1)$ . Hence near $s = 1$ , $$ F(s) = \frac{1}{s-1} L(1, \chi_1) L(1, \chi_2) L(1, \chi_1\chi_2) + O(1), $$ so the residue is $λ = L (1, χ_{1}) L (1, χ_{2}) L (1, χ_{1} χ_{2})$ , which is positive since each factor is. Rubric: full credit for identifying the single pole and multiplying the three regular values.

Exercise 4 (medium, symbolic).

Explain precisely where the ineffectivity enters Siegel's proof. Which quantity in Case A is not computable, and why does Case B not rescue effectivity?

Hint

In Case A the constant depends on $β_{1}$ . To use Case B instead, one must certify that Case A fails.

Answer

In Case A the auxiliary character $χ_{1}$ is one with an exceptional zero $β_{1} > 1 - ε /4$ , and the final constant in $L (1, χ_{2}) ≫_{ε} q_{2}^{- ε}$ depends on the value $1 - β_{1}$ . This $β_{1}$ is not computable: no exceptional zero has ever been exhibited, so we cannot extract a numerical $β_{1}$ . One might hope to use Case B (no exceptional zeros) where the constant is explicit — but to know we are in Case B we would have to prove no real character has a zero with $β > 1 - ε /4$ , which is exactly the unsolved Landau-Siegel zero problem. So the proof must keep both branches open, and since the constant in the live branch (whichever it is) cannot be named, $c (ε)$ is ineffective for $ε < 1/2$ . Rubric: full credit for identifying $β_{1}$ as the non-computable quantity and for explaining that certifying Case B is itself open.

Exercise 5 (medium, symbolic).

Translate Siegel's bound $L (1, χ_{- d}) ≫_{ε} d^{- ε}$ into the class-number bound $h (- d) ≫_{ε} d^{1/2 - ε}$ using the Dirichlet class-number formula.

Hint

Solve $h (- d) = w d L (1, χ_{- d}) / (2 π)$ for $h$ and substitute, with $w$ bounded.

Answer

From $h (- d) = w d L (1, χ_{- d}) / (2 π)$ with $w \in {2, 4, 6}$ bounded below by $2$ , $$ h(-d) = \frac{w\sqrt{d}}{2\pi} L(1, \chi_{-d}) \gg \sqrt{d}, L(1, \chi_{-d}). $$ Substituting Siegel's bound $L (1, χ_{- d}) ≫_{ε} d^{- ε}$ (taking $q = d$ ), $$ h(-d) \gg_\varepsilon \sqrt{d} \cdot d^{-\varepsilon} = d^{1/2 - \varepsilon}. $$ The class number grows nearly like $d$ , answering Gauss's question affirmatively, though the implied constant inherits Siegel's ineffectivity. Rubric: full credit for the substitution, the exponent $1/2 - ε$ , and noting the inherited ineffectivity.

Exercise 6 (hard, symbolic).

Derive the key inequality $F (σ_{0}) \geq 1 - \frac{c _{1} λ}{1 - σ _{0}} (q_{1} q_{2})^{c_{2} (1 - σ_{0})}$ in outline, for $σ_{0} \in (1/2, 1)$ , from the non-negativity of the coefficients of $F$ and the pole at $s = 1$ .

Hint

Write $F (s) = \sum a_{n} n^{- s}$ with $a_{n} \geq 0$ , $a_{1} = 1$ . Use a smoothed Perron / contour integral picking up the pole and bounding the tail by the functional equation.

Answer

Let $F (s) = \sum_{n} a_{n} n^{- s}$ with $a_{n} \geq 0$ and $a_{1} = 1$ . Apply the Mellin-Perron integral $\frac{1}{2 π i} \int_{(2)} F (s + σ_{0}) Γ (s) X^{s} d s = \sum_{n} a_{n} n^{- σ_{0}} e^{- n / X}$ for a smoothing parameter $X$ . Moving the contour left past the pole of $ζ$ at $s = 1 - σ_{0}$ (residue $λ Γ (1 - σ_{0}) X^{1 - σ_{0}}$ ) and the pole of $Γ$ at $s = 0$ (residue $F (σ_{0})$ ), and bounding the shifted integral by the convexity bound $∣ F (σ_{0} + i t) ∣ ≪ (q_{1} q_{2} (∣ t ∣ + 1))^{c_{2} (1 - σ_{0})}$ from the functional equation, gives $$ \sum_n a_n n^{-\sigma_0} e^{-n/X} = F(\sigma_0) + \lambda \Gamma(1-\sigma_0) X^{1-\sigma_0} + O\big((q_1 q_2)^{c_2(1-\sigma_0)} X^{-A}\big). $$ The left side is $\geq a_{1} \cdot 1 \cdot e^{- 1/ X} \geq 1 - 1/ X$ by non-negativity (keeping only $n = 1$ ). Choosing $X$ a suitable power of $q_{1} q_{2}$ and rearranging isolates $F (σ_{0}) \geq 1 - \frac{c _{1} λ}{1 - σ _{0}} (q_{1} q_{2})^{c_{2} (1 - σ_{0})}$ , the $Γ (1 - σ_{0}) ≍ 1/ (1 - σ_{0})$ near $σ_{0} = 1$ supplying the displayed denominator. Rubric: full credit for the Perron set-up, the two residues, the non-negativity lower bound on the left, and the convexity tail bound.

Exercise 7 (hard, symbolic).

State the Brauer-Siegel theorem for a sequence of number fields $K$ with degree fixed and discriminant $d_{K} \to \infty$ , and explain how Siegel's theorem is the imaginary-quadratic case.

Hint

Brauer-Siegel concerns the product $h_{K} R_{K}$ of class number and regulator; for imaginary quadratic fields $R_{K} = 1$ .

Answer

Brauer-Siegel. For a sequence of number fields $K$ of fixed degree $n$ (or with $[K : Q] / lo g ∣ d_{K} ∣ \to 0$ ) and $∣ d_{K} ∣ \to \infty$ , $$ \frac{\log(h_K R_K)}{\log\sqrt{|d_K|}} \to 1, $$ i.e. $h_{K} R_{K} = ∣ d_{K} ∣^{1/2 + o (1)}$ , where $h_{K}$ is the class number and $R_{K}$ the regulator. The proof rests on the same residue analysis: the Dedekind zeta $ζ_{K} (s)$ has residue at $s = 1$ proportional to $h_{K} R_{K} / ∣ d_{K} ∣$ , and bounding this residue above and below gives the asymptotic, the lower bound being the ineffective Siegel-type input. For an imaginary quadratic field $K = Q (- d)$ the unit group is finite, so the regulator $R_{K} = 1$ , and Brauer-Siegel reduces to $h (- d) = d^{1/2 + o (1)}$ — precisely Siegel's $h (- d) ≫_{ε} d^{1/2 - ε}$ together with the matching upper bound from $L (1, χ) ≪ lo g d$ . The general theorem is ineffective for the same reason: the lower bound on the residue passes through a Siegel-type case-split. Rubric: full credit for the correct statement, the reduction by vanishing regulator, and the shared ineffectivity.

Exercise 8 (hard, symbolic).

Contrast Siegel's ineffective bound with the Goldfeld-Gross-Zagier effective bound. What does each give for $h (- d)$ , and what makes the latter effective?

Hint

Siegel gives a strong rate $d^{1/2 - ε}$ ineffectively; Goldfeld-Gross-Zagier gives a weak rate $(lo g d)^{1 - ε}$ effectively. The effectivity comes from a known $L$ -function with a high-order zero.

Answer

Siegel (1935). $h (- d) ≫_{ε} d^{1/2 - ε}$ : a strong rate, nearly $d$ , but with an ineffective constant, so it cannot certify $h (- d) > N$ for any explicit $N$ and given $d$ . Goldfeld-Gross-Zagier (1976/1986). $h (- d) ≫ (lo g d)^{1 - ε}$ : a much weaker rate (logarithmic, not power), but effective — the constant can be computed, so one can in principle list all $d$ with $h (- d) \leq N$ . The effectivity comes from replacing the case-split on an unknown exceptional zero by an exhibited analytic object: Goldfeld showed that an elliptic curve $L$ -function with a zero of order $\geq 3$ at the central point yields an effective class-number bound, and Gross-Zagier produced such a curve explicitly via the Heegner-point height formula $L^{'} (E, 1) ≍ ⟨ y_{K}, y_{K} ⟩$ . Because the high-order-zero curve is known, no undecided case remains and the constant is extractable. The trade-off is universal: Siegel buys a strong rate with an invisible constant; Goldfeld-Gross-Zagier buys a computable constant with a weak rate. Rubric: full credit for both rates, the effective/ineffective contrast, and the role of the exhibited high-order zero.

Advanced results Master

Theorem 1 (Siegel's lower bound, sharp form). For every $ε > 0$ there is an ineffective $c (ε) > 0$ with $L (1, χ) > c (ε) q^{- ε}$ for all real primitive $χ$ modulo $q$ ^{[Siegel 1935]}. Equivalently $1 - β > c^{'} (ε) q^{- ε}$ for the exceptional zero. The exponent $ε$ cannot be replaced by any explicit function tending to $0$ while keeping the constant effective; the boundary $ε = 1/2$ is where ineffectivity sets in, since for $ε \geq 1/2$ the effective Page bound already suffices.

Theorem 2 (the dichotomy made quantitative). The proof of Theorem 1 turns on the alternative of 21.13.01: either (A) some real primitive character $χ_{1}$ has $β_{1} > 1 - ε /4$ , in which case $χ_{1}$ is a fixed lever bounding $L (1, χ_{2}) ≫_{ε} q_{2}^{- ε}$ for every other $χ_{2}$ through the auxiliary product $F = ζ L (\cdot, χ_{1}) L (\cdot, χ_{2}) L (\cdot, χ_{1} χ_{2})$ ; or (B) no such zero exists, in which case the same product with $σ_{0} = 1 - ε /4$ gives the bound directly ^{[Davenport §21]}. The two constants disagree, and no algorithm selects between them.

Theorem 3 (class number of imaginary quadratic fields). For fundamental $- d < 0$ , the Dirichlet class-number formula $h (- d) = w d L (1, χ_{- d}) / (2 π)$ converts Theorem 1 into $h (- d) ≫_{ε} d^{1/2 - ε}$ ^{[Davenport §21]}. This settled Gauss's conjecture that $h (- d) \to \infty$ , but ineffectively: the finiteness of the set ${d : h (- d) = 1}$ follows, while its determination — the nine Heegner discriminants $d \in {3, 4, 7, 8, 11, 19, 43, 67, 163}$ — required the independent effective methods of Heegner (1952), Baker (1966), and Stark (1967).

Theorem 4 (Brauer-Siegel). For a sequence of number fields of fixed degree with $∣ d_{K} ∣ \to \infty$ , $lo g (h_{K} R_{K}) / lo g ∣ d_{K} ∣ \to 1$ ^{[Iwaniec-Kowalski §22]}. The residue of $ζ_{K}$ at $s = 1$ , equal to $\frac{2 ^{r_{1}} ( 2 π ) ^{r_{2}} h _{K} R _{K}}{w _{K} ∣ d _{K} ∣}$ by the analytic class-number formula, is squeezed between bounds whose lower side is a Siegel-type ineffective estimate; the imaginary-quadratic case $R_{K} = 1$ recovers Theorem 3.

Theorem 5 (effective Goldfeld-Gross-Zagier bound). There is an effectively computable constant with $h (- d) > c (lo g d)^{1 - ε}$ ^{[Goldfeld 1976; Gross-Zagier 1986]}. Goldfeld reduced an effective bound to the existence of a modular elliptic curve whose $L$ -function vanishes to order $\geq 3$ at $s = 1$ ; Gross-Zagier's height formula $L^{'} (E / Q, 1) = c_{E} ⟨ y_{K}, y_{K} ⟩$ for the Heegner point $y_{K}$ exhibits such a curve, removing the undecided case and yielding a computable constant — at the cost of the far weaker logarithmic rate.

Synthesis. Siegel's theorem is the foundational reason the analytic theory of primes in arithmetic progressions is uniform at all, and the central insight is that the non-negativity of the auxiliary product $ζ L (\cdot, χ_{1}) L (\cdot, χ_{2}) L (\cdot, χ_{1} χ_{2})$ converts the repulsion of 21.13.01 into a lower bound: a near-one zero of one character is exactly the lever that bounds another, and this is exactly the same coefficient positivity $Λ (n) (1 + χ_{1}) (1 + χ_{2}) \geq 0$ that forbade two near-one zeros from coexisting. Putting these together, the dichotomy generalises the single-character exceptional-zero theory into a two-character argument, and the bridge is the residue $λ = L (1, χ_{1}) L (1, χ_{2}) L (1, χ_{1} χ_{2})$ : bounding it below bounds $L (1, χ_{2})$ below, which is dual to the class-number formula $h (- d) ≍ d L (1, χ_{- d})$ , so the analytic statement is the arithmetic one. The central insight recurs in the contrast with Goldfeld-Gross-Zagier: Siegel's ineffectivity is the shadow of an undecided case, and the only known route to an effective constant trades the strong power rate for a weak logarithmic one by exhibiting, rather than hypothesising, the analytic object that does the bounding — the high-order zero of an elliptic curve $L$ -function supplied by the Heegner-point height. This is dual to Gauss's original effectively-open class-number problem: the same $L (1, χ)$ that measures the exceptional zero measures the class number, and the entire quantitative theory is written in the conditional mood the Siegel zero imposes.

Full proof set Master

Proposition 1 (non-negativity of the auxiliary coefficients). For real primitive characters $χ_{1}, χ_{2}$ with $χ_{1} χ_{2}$ non-principal, the Dirichlet series $F (s) = ζ (s) L (s, χ_{1}) L (s, χ_{2}) L (s, χ_{1} χ_{2})$ has coefficients $a_{n} \geq 0$ with $a_{1} = 1$ .

Proof. Taking logarithms over the Euler product, $lo g F (s) = \sum_{p} \sum_{m \geq 1} \frac{1}{m} (1 + χ_{1} (p^{m}) + χ_{2} (p^{m}) + χ_{1} χ_{2} (p^{m})) p^{- m s}$ . The bracket equals $(1 + χ_{1} (p^{m})) (1 + χ_{2} (p^{m})) \geq 0$ since each $χ_{i}$ is real-valued in ${- 1, 0, 1}$ . Hence $- F^{'} / F = \sum_{n} Λ (n) (1 + χ_{1} (n)) (1 + χ_{2} (n)) n^{- s}$ has non-negative coefficients, so $lo g F$ is a Dirichlet series with non-negative coefficients, and therefore $F = exp (lo g F)$ also has non-negative coefficients (the exponential of a non-negative-coefficient series, via the convolution exponential, preserves non-negativity). The constant term is $a_{1} = 1$ . $□$

Proposition 2 (lower bound from non-negativity and the pole). With $F$ as above, residue $λ > 0$ at $s = 1$ , and $σ_{0} \in (1/2, 1)$ , there are absolute $c_{1}, c_{2} > 0$ with $F (σ_{0}) \geq 1 - \frac{c _{1} λ}{1 - σ _{0}} (q_{1} q_{2})^{c_{2} (1 - σ_{0})}$ .

Proof. Smooth Perron with the Gamma kernel gives, for $X \geq 1$ , $$ \sum_n a_n n^{-\sigma_0} e^{-n/X} = F(\sigma_0) + \lambda,\Gamma(1 - \sigma_0) X^{1 - \sigma_0} + \frac{1}{2\pi i}\int_{(\sigma_0 - 2 - \sigma_0)} F(s + \sigma_0)\Gamma(s) X^s, ds. $$ The left side is at least $a_{1} e^{- 1/ X} = e^{- 1/ X} \geq 1 - 1/ X$ by non-negativity of the $a_{n}$ . The shifted integral, on the line $Re (s) = - 1 + ε - σ_{0}$ , is bounded using the convexity estimate $∣ F (σ + i t) ∣ ≪ (q_{1} q_{2})^{c_{2} (1 - σ)} (1 + ∣ t ∣)^{c_{2} (1 - σ)}$ from the functional equations of the four factors and the rapid decay of $Γ$ ; it contributes $O ((q_{1} q_{2})^{c_{2} (1 - σ_{0})} X^{- 1 + ε})$ . Thus $F (σ_{0}) = \sum a_{n} n^{- σ_{0}} e^{- n / X} - λ Γ (1 - σ_{0}) X^{1 - σ_{0}} - O (\dots) \geq 1 - 1/ X - λ Γ (1 - σ_{0}) X^{1 - σ_{0}} - O (\dots)$ . Choosing $X = (q_{1} q_{2})^{c_{2}}$ and using $Γ (1 - σ_{0}) ≍ (1 - σ_{0})^{- 1}$ near $σ_{0} = 1$ yields $F (σ_{0}) \geq 1 - \frac{c _{1} λ}{1 - σ _{0}} (q_{1} q_{2})^{c_{2} (1 - σ_{0})}$ . $□$

Proposition 3 (Case A lever bound). If a fixed real primitive $χ_{1}$ has $L (β_{1}, χ_{1}) = 0$ with $1 - β_{1} = η$ small, then $L (1, χ_{2}) ≫_{η} q_{2}^{- 2 c_{2} η}$ for every real primitive $χ_{2}$ with $χ_{1} χ_{2}$ non-principal.

Proof. Evaluate Proposition 2 at $σ_{0} = β_{1}$ . Since $L (β_{1}, χ_{1}) = 0$ makes $F (β_{1}) = 0$ , the inequality forces $1 \leq \frac{c _{1} λ}{1 - β _{1}} (q_{1} q_{2})^{c_{2} (1 - β_{1})}$ , i.e. $λ \geq \frac{η}{c _{1}} (q_{1} q_{2})^{- c_{2} η}$ . Now $λ = L (1, χ_{1}) L (1, χ_{2}) L (1, χ_{1} χ_{2})$ with $L (1, χ_{1}) ≪ lo g q_{1}$ and $L (1, χ_{1} χ_{2}) ≪ lo g (q_{1} q_{2})$ both bounded above (the upper bound $L (1, χ) ≪ lo g q$ is elementary). Therefore $L (1, χ_{2}) \geq \frac{λ}{L ( 1 , χ _{1} ) L ( 1 , χ _{1} χ _{2} )} ≫ \frac{η ( q _{1} q _{2} ) ^{- c_{2} η}}{( l o g q _{1} q _{2} ) ^{2}} ≫_{η} q_{2}^{- 2 c_{2} η}$ , the fixed $q_{1}$ absorbed into the $η$ -constant. Taking $η < ε / (2 c_{2})$ gives $L (1, χ_{2}) ≫_{ε} q_{2}^{- ε}$ . $□$

Proposition 4 (ineffectivity). The constant $c (ε)$ in $L (1, χ) > c (ε) q^{- ε}$ produced by Propositions 1-3 is ineffective for $ε < 1/2$ .

Proof. The bound is proved by the alternative: if some real primitive character has a zero $β_{1} > 1 - ε /4$ , fix it as the lever $χ_{1}$ in Proposition 3, obtaining a constant depending on $η = 1 - β_{1}$ ; if none does, Proposition 2 at $σ_{0} = 1 - ε /4$ gives an explicit constant. The two constants are not equal, and the constant delivered by the proof is the one from the branch that actually holds. To name it one must either exhibit $β_{1}$ (impossible: no exceptional zero is known) or prove no exceptional zero exists with $β > 1 - ε /4$ (open: this is the Landau-Siegel zero problem). Since neither sub-problem is solved for $ε < 1/2$ , no finite procedure outputs a numerical $c (ε)$ ; the existence $\exists c (ε) > 0$ is all that is proved. For $ε \geq 1/2$ , Page's effective bound of 21.13.01 already gives an explicit constant, so ineffectivity is confined to $ε < 1/2$ . $□$

Connections Master

The exceptional-zero dichotomy, the auxiliary product $ζ L (\cdot, χ_{1}) L (\cdot, χ_{2}) L (\cdot, χ_{1} χ_{2})$ , Landau's repulsion, and Page's effective bound are all imported from 21.13.01; this unit consumes that infrastructure and runs the repulsion in reverse — where 21.13.01 used coefficient non-negativity to show two near-one zeros cannot coexist, Siegel uses the same product to convert one hypothetical near-one zero into a lower bound for every other $L (1, χ)$ .
The Dirichlet $L$ -function $L (s, χ)$ , its Euler product, functional equation, and the boundary non-vanishing $L (1, χ) \neq = 0$ all come from 21.03.02; Siegel's theorem is the quantitative refinement of that non-vanishing, replacing the qualitative $L (1, χ) \neq = 0$ with the explicit lower order $L (1, χ) ≫_{ε} q^{- ε}$ , and the class-number formula that converts the $L$ -value into an arithmetic invariant lives in the same character-theoretic chapter.
Siegel's ineffective constant propagates into the Siegel-Walfisz theorem of 21.13.03 on primes in arithmetic progressions uniform for $q \leq (lo g x)^{A}$ ; the modulus range and the ineffectivity of its implied constant are direct inheritances from the bound proved here, making this unit the analytic bottleneck for uniform PNT in progressions.
The effective zero-free region for $ζ$ of 21.12.03 is the $q = 1$ specialisation of the region whose real-character defect Siegel's theorem repairs; together with 21.13.01 it shows how the same $3 + 4 cos θ + cos 2 θ$ pole-budget method governs both the effective $ζ$ region and the ineffective $L$ -function exceptional-zero theory, the contrast between them being precisely the conjugate-symmetry loophole for real characters.

Historical & philosophical context Master

Carl Ludwig Siegel proved the bound $L (1, χ) ≫_{ε} q^{- ε}$ in a three-page 1935 paper in Acta Arithmetica ^{[Siegel 1935]}, whose explicit aim was the class number of imaginary quadratic fields: Gauss had conjectured in the Disquisitiones Arithmeticae of 1801 that $h (- d) \to \infty$ as $d \to \infty$ , and Siegel's bound $h (- d) ≫_{ε} d^{1/2 - ε}$ proved it with a near-optimal rate. The auxiliary-product argument refined Landau's 1918 repulsion idea, turning the mutual repulsion of exceptional zeros into a positive lower bound. Richard Brauer extended the residue analysis to general number fields, giving the Brauer-Siegel theorem $h_{K} R_{K} = ∣ d_{K} ∣^{1/2 + o (1)}$ , which subsumes Siegel's result as the imaginary-quadratic case, where the regulator is $1$ .

The ineffectivity was understood from the start as structural rather than technical: the proof selects a favourable case keyed to the existence of an exceptional zero, and that existence remains undecided. The consequence — that Siegel's bound cannot determine the list of one-class discriminants — left the Gauss class-number-one problem open until Heegner (1952), Baker (1966), and Stark (1967) settled it by independent effective means. The first effective lower bound on $h (- d)$ growing with $d$ came only in 1976-1986: Dorian Goldfeld ^{[Goldfeld 1976]} reduced effectivity to the existence of an $L$ -function with a triple zero at the central point, and Benedict Gross and Don Zagier ^{[Gross-Zagier 1986]} supplied one via their height formula for Heegner points, yielding the effective but weak $h (- d) ≫ (lo g d)^{1 - ε}$ . The Landau-Siegel zero problem — eliminating the exceptional zero outright, which would make Siegel's theorem effective with its strong rate — remains among the central open problems of analytic number theory.

Bibliography Master

@article{siegel1935classenzahl,
  author  = {Siegel, Carl Ludwig},
  title   = {\"{U}ber die Classenzahl quadratischer Zahlk\"{o}rper},
  journal = {Acta Arithmetica},
  volume  = {1},
  pages   = {83--86},
  year    = {1935}
}

@article{goldfeld1976,
  author  = {Goldfeld, Dorian},
  title   = {The class number of quadratic fields and the conjectures of Birch and Swinnerton-Dyer},
  journal = {Annali della Scuola Normale Superiore di Pisa, Classe di Scienze (4)},
  volume  = {3},
  number  = {4},
  pages   = {624--663},
  year    = {1976}
}

@article{grosszagier1986,
  author  = {Gross, Benedict H. and Zagier, Don B.},
  title   = {Heegner points and derivatives of $L$-series},
  journal = {Inventiones Mathematicae},
  volume  = {84},
  number  = {2},
  pages   = {225--320},
  year    = {1986}
}

@book{davenport2000siegel,
  author    = {Davenport, Harold},
  title     = {Multiplicative Number Theory},
  edition   = {3},
  series    = {Graduate Texts in Mathematics},
  volume    = {74},
  publisher = {Springer-Verlag},
  year      = {2000},
  note      = {Revised by H. L. Montgomery; \S21--22: Siegel's theorem and the class number}
}

@book{iwaniec-kowalski2004siegel,
  author    = {Iwaniec, Henryk and Kowalski, Emmanuel},
  title     = {Analytic Number Theory},
  series    = {American Mathematical Society Colloquium Publications},
  volume    = {53},
  publisher = {American Mathematical Society},
  year      = {2004},
  note      = {\S5.9, \S22: the Landau-Siegel zero and the Goldfeld-Gross-Zagier bound}
}

@book{montgomery-vaughan2007siegel,
  author    = {Montgomery, Hugh L. and Vaughan, Robert C.},
  title     = {Multiplicative Number Theory I: Classical Theory},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {97},
  publisher = {Cambridge University Press},
  year      = {2007},
  note      = {\S11: Siegel's theorem and the Siegel-Walfisz theorem}
}

Prerequisites

21.13.01
21.03.02

Tier anchors

beginner: Derbyshire 2003 *Prime Obsession* (Joseph Henry Press) Ch. 21-22 (the Siegel zero as the 'one zero that refuses to move', told informally); Soundararajan's expository notes on the Landau-Siegel zero for the picture of a strong-but-invisible bound and the class-number link
intermediate: Davenport 2000 *Multiplicative Number Theory* 3e (Springer GTM 74) §21 (Siegel's theorem $L(1,\chi) \gg_\varepsilon q^{-\varepsilon}$ via the auxiliary product $\zeta(s) L(s,\chi_1) L(s,\chi_2) L(s,\chi_1\chi_2)$, the dichotomy on whether a Siegel zero exists, the ineffectivity, and the class-number consequence); Montgomery-Vaughan 2007 *Multiplicative Number Theory I* (Cambridge UP) §11
master: Davenport 2000 *Multiplicative Number Theory* 3e §21-22 (the full two-$L$-function proof, ineffectivity, the class-number bound $h(-d) \gg_\varepsilon d^{1/2-\varepsilon}$, the Brauer-Siegel theorem stated); Iwaniec-Kowalski 2004 *Analytic Number Theory* (AMS Colloquium Publications 53) §5.9, §22 (the Landau-Siegel zero, the contrast with Goldfeld-Gross-Zagier); Siegel 1935 *Acta Arithmetica* 1; Goldfeld 1976 *Ann. Scuola Norm. Sup. Pisa*; Gross-Zagier 1986 *Invent. Math.* 84

References

Davenport, H. — Multiplicative Number Theory · Springer Graduate Texts in Mathematics 74, 3rd edition (2000), revised by H. L. Montgomery. §21. Proves Siegel's theorem: for every $\varepsilon > 0$ there is $c(\varepsilon) > 0$ with $L(1, \chi) > c(\varepsilon) q^{-\varepsilon}$ for real primitive $\chi$ modulo $q$, equivalently the exceptional zero satisfies $\beta < 1 - c(\varepsilon) q^{-\varepsilon}$. The proof studies the product $\zeta(s) L(s, \chi_1) L(s, \chi_2) L(s, \chi_1\chi_2)$ for two real characters, splits on whether some $L(s, \chi_1)$ has a Siegel zero, and in either case bounds $L(1, \chi_2)$ below; the constant $c(\varepsilon)$ is ineffective because the favourable case is selected by the existence of a zero that cannot be exhibited. Deduces the class-number bound $h(-d) \gg_\varepsilon d^{1/2 - \varepsilon}$.
Siegel, C. L. — Über die Classenzahl quadratischer Zahlkörper · Acta Arithmetica 1 (1935), 83-86. The original proof of the ineffective lower bound $L(1, \chi) \gg_\varepsilon q^{-\varepsilon}$ for real primitive $\chi$ modulo $q$, and the class-number lower bound $h(-d) \gg_\varepsilon d^{1/2 - \varepsilon}$ for imaginary quadratic fields via the Dirichlet class-number formula.
Iwaniec, H. & Kowalski, E. — Analytic Number Theory · AMS Colloquium Publications 53 (2004), §5.9 and §22. The Landau-Siegel zero, Siegel's theorem with the auxiliary-character argument, the ineffectivity, the Deuring-Heilbronn repulsion, and the contrast between Siegel's ineffective class-number bound and the effective Goldfeld-Gross-Zagier bound from the analytic theory of elliptic curves.
Goldfeld, D. — The class number of quadratic fields and the conjectures of Birch and Swinnerton-Dyer · Annali della Scuola Normale Superiore di Pisa (4) 3 (1976), 624-663. Reduces an effective lower bound for $h(-d)$ to the existence of an $L$-function with a triple (or higher) zero at the central point; the existence is supplied by Gross-Zagier.
Gross, B. & Zagier, D. — Heegner points and derivatives of L-series · Inventiones Mathematicae 84 (1986), 225-320. The Gross-Zagier formula relating the height of Heegner points to the derivative of the L-function; combined with Goldfeld it yields the first effective lower bound $h(-d) \gg (\log d)^{1-\varepsilon}$ for imaginary quadratic class numbers.
Montgomery, H. L. & Vaughan, R. C. — Multiplicative Number Theory I: Classical Theory · Cambridge Studies in Advanced Mathematics 97 (2007), §11. Siegel's theorem, the Siegel-Walfisz theorem on PNT in arithmetic progressions, and the ineffectivity of the implied constants.

Estimated time

beginner: 18m
intermediate: 46m
master: 90m