The Birch and Swinnerton-Dyer Conjecture

⚡ Key Takeaways

The Birch and Swinnerton-Dyer (BSD) Conjecture asserts that for an elliptic curve $E$ over $\mathbb{Q}$ , the order of vanishing of its Hasse–Weil $L$ -function $L(E,s)$ at $s=1$ equals the rank of the Mordell–Weil group $E(\mathbb{Q})$ . In words: the analytic behaviour of an $L$ -function at its central point detects exactly how many independent rational points of infinite order live on the curve. Where it applies: arithmetic of elliptic curves, Diophantine equations, the theory of motives and $L$ -functions. Who/when: Bryan Birch and Peter Swinnerton-Dyer, formulated 1960–1965 via EDSAC computations in Cambridge; one of the seven Clay Millennium Prize Problems (2000), worth $1,000,000.

💡 The Core Insight

Define the local factor $a_p = p + 1 - \#E(\mathbb{F}_p)$ . Heuristically, if $E$ has many rational points, those points reduce to many points mod $p$ , so $\#E(\mathbb{F}_p)$ is "large," $a_p$ is "small/negative," and the product $\prod_p \#E(\mathbb{F}_p)/p$ tends to zero — forcing $L(E,s)$ to vanish at $s=1$ to high order. Birch and Swinnerton-Dyer turned this heuristic into a precise quantitative dictionary: rank = order of vanishing, with the leading coefficient governed by the arithmetic invariants of $E$ . The entire conjecture is the statement that this naïve "more points $\Rightarrow$ more vanishing" intuition is exactly right.

DefinitionElliptic Curve and Mordell–Weil Group

An elliptic curve over $\mathbb{Q}$ is a smooth projective genus-1 curve with a specified rational point $O$ , given (after change of variables, away from characteristic 2,3) by a Weierstrass equation

1

\quad E:\ y^2 = x^3 + Ax + B, \qquad A,B \in \mathbb{Q},\quad \Delta = -16(4A^3 + 27B^2) \neq 0.

Derivation hint: smoothness is equivalent to the cubic $x^3+Ax+B$ having distinct roots, i.e. nonvanishing discriminant $\Delta$ . The set of rational points $E(\mathbb{Q})$ forms an abelian group under the chord-and-tangent law with identity $O$ . The Mordell–Weil Theorem (Mordell 1922, Weil 1929) states it is finitely generated:

2

\quad E(\mathbb{Q}) \;\cong\; \mathbb{Z}^{\,r} \;\oplus\; E(\mathbb{Q})_{\mathrm{tors}},

where $r = \mathrm{rank}\,E(\mathbb{Q})$ is the algebraic rank and the torsion subgroup is finite (Mazur: one of 15 explicit groups over $\mathbb{Q}$ ). Derivation hint: descent gives $E(\mathbb{Q})/mE(\mathbb{Q})$ finite; combined with the theory of heights and an infinite descent argument yields finite generation.

DefinitionThe Hasse–Weil L-function

For a prime $p$ of good reduction (i.e. $p \nmid \Delta$ for a minimal model), set

3

\quad a_p = p + 1 - \#E(\mathbb{F}_p), \qquad |a_p| \leq 2\sqrt{p} \ \ (\text{Hasse bound}).

Derivation hint: $\#E(\mathbb{F}_p) = p+1-a_p$ where $a_p$ is the trace of Frobenius; the bound follows from the Riemann Hypothesis for curves over finite fields (Hasse 1933). The $L$ -function is the Euler product

4

\quad L(E,s) = \prod_{p \mid N} \big(1 - a_p p^{-s}\big)^{-1} \prod_{p \nmid N} \big(1 - a_p p^{-s} + p^{1-2s}\big)^{-1},

where $N$ is the conductor and the bad-reduction factors use $a_p \in \{0,\pm 1\}$ (additive/split/non-split multiplicative reduction). Derivation hint: each local factor is the reciprocal of the numerator of the local zeta function $Z(E/\mathbb{F}_p, T)$ evaluated at $T = p^{-s}$ . By the Modularity Theorem (Wiles, Taylor–Wiles, Breuil–Conrad–Diamond–Taylor 1995–2001), $L(E,s)$ equals the $L$ -function of a weight-2 newform of level $N$ , hence extends to an entire function on $\mathbb{C}$ satisfying a functional equation relating $s$ and $2-s$ :

5

\quad \Lambda(E,s) = N^{s/2}(2\pi)^{-s}\,\Gamma(s)\,L(E,s) = \varepsilon \,\Lambda(E,2-s), \qquad \varepsilon = \pm 1.

Derivation hint: modularity identifies $L(E,s)$ with the Mellin transform of a cusp form $f$ , whose modular transformation under $w_N$ (Fricke involution) produces the sign $\varepsilon$ , the parity of the analytic rank.

TheoremBirch and Swinnerton-Dyer Conjecture

Let $E/\mathbb{Q}$ be an elliptic curve of rank $r$ .

(Weak BSD) The order of vanishing of $L(E,s)$ at the central point $s=1$ equals the rank:

6

\quad \mathrm{ord}_{s=1} L(E,s) = r = \mathrm{rank}\,E(\mathbb{Q}).

Derivation hint: conjectural; the right side is the analytic rank $r_{\mathrm{an}}$ . The functional equation forces $\varepsilon = (-1)^{r_{\mathrm{an}}}$ , giving the parity prediction $r \equiv r_{\mathrm{an}} \pmod 2$ .

(Strong BSD) The leading Taylor coefficient is given explicitly by

7

\quad \lim_{s\to 1}\frac{L(E,s)}{(s-1)^r} = \frac{\#\Sha(E/\mathbb{Q}) \cdot \Omega_E \cdot \mathrm{Reg}(E/\mathbb{Q}) \cdot \prod_p c_p}{\big(\#E(\mathbb{Q})_{\mathrm{tors}}\big)^2}.

Derivation hint: this is the elliptic-curve specialization of the Tamagawa-number conjecture of Bloch–Kato; each factor is the "size" of a local or global arithmetic obstruction, assembled in analogy with the analytic class number formula. Here $\Sha(E/\mathbb{Q})$ is the Tate–Shafarevich group, $\Omega_E$ the real period, $\mathrm{Reg}$ the regulator (determinant of the canonical height pairing on a basis of $E(\mathbb{Q})/\mathrm{tors}$ ), $c_p$ the local Tamagawa numbers, all defined below.

DefinitionThe Arithmetic Invariants in Strong BSD

Real period $\Omega_E = \int_{E(\mathbb{R})} |\omega|$ , the integral of the Néron differential $\omega = dx/(2y)$ over the real points.

8

\quad \Omega_E = c_\infty \int_{e_1}^{\infty} \frac{dx}{\sqrt{x^3+Ax+B}}, \qquad c_\infty = \#\big(E(\mathbb{R})/E(\mathbb{R})^0\big) \in \{1,2\}.

Derivation hint: on a minimal model the Néron differential is well-defined up to sign; $c_\infty$ counts real connected components.

Regulator $\mathrm{Reg}(E/\mathbb{Q}) = \det\big(\langle P_i, P_j\rangle\big)_{1\le i,j\le r}$ , the Gram determinant of the canonical (Néron–Tate) height pairing on a basis $P_1,\dots,P_r$ of the free part. (For $r=0$ , $\mathrm{Reg}=1$ by convention.)

9

\quad \hat h(P) = \lim_{n\to\infty} \frac{h(2^n P)}{4^n}, \qquad \langle P,Q\rangle = \tfrac12\big(\hat h(P+Q) - \hat h(P) - \hat h(Q)\big).

Derivation hint: the naïve logarithmic height $h$ is quasi-quadratic; the limit regularizes it into an exact quadratic form, positive-definite on $E(\mathbb{Q})\otimes\mathbb{R}$ .

Tamagawa numbers $c_p = [E(\mathbb{Q}_p):E^0(\mathbb{Q}_p)]$ , the number of components of the special fibre of the Néron model rational over $\mathbb{F}_p$ ( $c_p = 1$ for good reduction). They are computed from the Kodaira type via Tate's algorithm.

Tate–Shafarevich group $\Sha(E/\mathbb{Q}) = \ker\Big(H^1(\mathbb{Q},E) \to \prod_v H^1(\mathbb{Q}_v, E)\Big)$ , classifying genus-1 curves (torsors under $E$ ) that have points everywhere locally but possibly not globally. Its finiteness is itself conjectural and a prerequisite for even stating [7] meaningfully.

ProofHeuristic for Weak BSD via the Naïve Euler Product

The original Birch–Swinnerton-Dyer reasoning was a heuristic, not a proof — but it illuminates the mechanism.

Step 1 — Define a partial product. For good primes, consider

10

\quad \frac{\#E(\mathbb{F}_p)}{p} = \frac{p+1-a_p}{p} = 1 - \frac{a_p - 1}{p}.

Derivation hint: substitute [3] and divide; the deviation from 1 measures excess of points over the "expected" $p$ .

Step 2 — Form the truncated product. BSD studied numerically

11

\quad \prod_{p \le X} \frac{\#E(\mathbb{F}_p)}{p} \;\sim\; C \,(\log X)^{r} \quad \text{as } X\to\infty,

finding empirically that the power of $\log X$ equals the rank $r$ . Derivation hint: heuristically $\#E(\mathbb{F}_p)/p \approx |c_p(s)|$ near $s=1$ ; a curve with many rational points has systematically more points mod $p$ , biasing the product to grow like $(\log X)^r$ — the signature of an order- $r$ zero in the associated $L$ -function.

Step 3 — Relate to $L(E,s)$ . Formally setting $s=1$ in the Euler product [4] (where it does not converge — only conditionally, à la Mertens) one expects

12

\quad \prod_{p \le X}\big(1 - a_p p^{-1} + p^{-1}\big)^{-1} \approx \frac{1}{L(E,1)}\cdot(\text{partial product}),

so growth $(\log X)^r$ of the point-product corresponds to $L(E,s) \sim c\,(s-1)^r$ near $s=1$ . Derivation hint: a Tauberian-type comparison; rigour requires the Riemann Hypothesis for $L(E,s)$ (GRH) and is the source of the conjectural status.

Step 4 — Identify the constant. The leading constant $C$ in [11], once all local periods and Tamagawa factors are correctly normalized, matches the strong-BSD formula [7] — this is how the period, regulator, and $\Sha$ entered the picture. Derivation hint: each $c_p$ corrects the bad-reduction primes; $\Omega_E$ the archimedean place; $\mathrm{Reg}$ the global heights; $\#\Sha$ the "missing" cohomological mass. The structure mirrors the analytic class number formula for Dedekind zeta functions. $\blacksquare$

∎

ProofWhat Is Actually Proven: Gross–Zagier–Kolyvagin

The deepest unconditional results establish BSD in ranks 0 and 1.

Step 1 — Heegner points (Gross–Zagier 1986). For $E/\mathbb{Q}$ modular with $L(E,1)=0$ , a Heegner point $y_K \in E(K)$ over a suitable imaginary quadratic field $K$ satisfies

13

\quad L'(E/K, 1) = c \cdot \hat h(y_K), \qquad c > 0.

Derivation hint: compute the Néron–Tate height of $y_K$ via the Gross–Zagier formula relating it to the derivative of the Rankin–Selberg convolution $L$ -function; nonvanishing of $L'$ forces $y_K$ to have infinite order.

Step 2 — Euler systems (Kolyvagin 1989). Using the Euler system of Heegner points, if $y_K$ is non-torsion then $E(K)$ has rank 1 and $\Sha(E/K)$ is finite. Derivation hint: derived classes from the Heegner system bound the Selmer group, controlling both rank and $\Sha$ .

Step 3 — Descent to $\mathbb{Q}$ . Combining with analytic nonvanishing results (Bump–Friedberg–Hoffstein, Murty–Murty) gives:

14

\quad r_{\mathrm{an}} \in \{0,1\} \ \Longrightarrow\ r = r_{\mathrm{an}} \ \text{ and } \ \Sha(E/\mathbb{Q}) \text{ is finite.}

Derivation hint: choose $K$ so that the analytic rank splits as $0+1$ across the curve and its quadratic twist; transfer the rank-1 information from $E(K)$ to $E(\mathbb{Q})$ . For $r_{\mathrm{an}} \ge 2$ , the conjecture is open. $\blacksquare$

∎

mathematician · Bryan Birch & Peter Swinnerton-Dyer | 1931–present / 1927–2018

Peter Swinnerton-Dyer, a Cambridge mathematician (and later Master of St Catharine's College), was among the first to harness electronic computation for pure number theory, programming the EDSAC machine. Bryan Birch, his collaborator, brought deep instincts in modular forms and Heegner points. In the early 1960s they computed $\#E(\mathbb{F}_p)$ for congruent-number curves $y^2 = x^3 - n^2 x$ across thousands of primes, plotted the partial products [11], and read off the rank from the slope on a log–log scale. The conjecture was a triumph of experimental mathematics decades before the term existed. Birch's later work on Heegner points (with Stephens and Zagier) seeded the Gross–Zagier formula. Swinnerton-Dyer's broad work spanned analytic number theory, the Hasse principle, and combinatorics; he was knighted in 2006.

🧠 Intuition

Think of three "dials" all reading the same hidden number, the rank $r$ : 1. Algebraic dial — how many independent infinite-order rational points you can actually find (often by brute-force point search or descent). 2. Analytic dial — how flat $L(E,s)$ is at $s=1$ : $r$ is the number of derivatives that vanish before the first nonzero one. 3. Geometric/statistical dial — how much "excess" the point counts $\#E(\mathbb{F}_p)$ carry on average; richer rational structure inflates local point counts, dragging the Euler product to zero. BSD says all three dials are wired together. The functional equation [5] already proves the parity of dial 2 (via $\varepsilon = (-1)^{r_{\mathrm{an}}}$ ), and the parity conjecture (proven by Dokchitser–Dokchitser, Nekovář under modularity) shows dial 1's parity matches dial 2's. What remains open is whether the exact values, not just parities, always coincide for $r \ge 2$ . The strong form [7] adds a fourth dial — $\Sha$ — measuring the failure of the local-to-global (Hasse) principle, the "dark matter" of the theory.

Figure

The three faces of the rank: analytic order of vanishing, rational points, and point-count excess

ExampleRank 0 — the curve 11a (conductor 11)

Given: $E: y^2 + y = x^3 - x^2 - 10x - 20$ , conductor $N=11$ , the curve of smallest conductor. We verify BSD numerically.

Step 1 — Torsion. $E(\mathbb{Q})_{\mathrm{tors}} \cong \mathbb{Z}/5\mathbb{Z}$ , so $\#E(\mathbb{Q})_{\mathrm{tors}} = 5$ .

Step 2 — Rank. Descent gives $r=0$ : no points of infinite order. Hence $\mathrm{Reg}=1$ (empty determinant).

Step 3 — $L$ -value. Numerically $L(E,1) \approx 0.2538\ldots \neq 0$ , confirming $r_{\mathrm{an}}=0=r$ (weak BSD holds).

Step 4 — Period and local factors. $\Omega_E \approx 1.2692\ldots$ ; the single bad prime $p=11$ (split multiplicative) has Tamagawa number $c_{11}=5$ .

Step 5 — Strong BSD check. Plug into [7]:

L(E,1) \stackrel{?}{=} \frac{\#\Sha \cdot \Omega_E \cdot 1 \cdot c_{11}}{5^2} = \frac{\#\Sha \cdot 1.2692 \cdot 5}{25}.

Solving with $L(E,1)\approx 0.2538$ forces $\#\Sha = 1$ (trivial), consistent with the known triviality.

\boxed{r = r_{\mathrm{an}} = 0,\quad \#\Sha=1,\quad \text{strong BSD verified numerically.}}

ExampleRank 1 — the curve 37a (conductor 37)

Given: $E: y^2 + y = x^3 - x$ , conductor $N=37$ , the rank-1 curve of smallest conductor.

Step 1 — A rational point. $P=(0,0)$ lies on $E$ and has infinite order; in fact $E(\mathbb{Q}) \cong \mathbb{Z}$ , generated by $P$ . So $r=1$ , $\#E(\mathbb{Q})_{\mathrm{tors}}=1$ .

Step 2 — Analytic rank. Sign $\varepsilon = -1$ , so $L(E,1)=0$ by the functional equation; numerically $L'(E,1)\approx 0.3059\ldots \neq 0$ , giving $r_{\mathrm{an}}=1$ . Weak BSD holds.

Step 3 — Regulator. The canonical height $\hat h(P) \approx 0.0511\ldots$ , so $\mathrm{Reg} = \hat h(P) \approx 0.0511$ .

Step 4 — Period. $\Omega_E \approx 5.9869\ldots$ . Good reduction except $p=37$ with $c_{37}=1$ .

Step 5 — Gross–Zagier check. By [13] the rank-1 BSD formula

L'(E,1) \stackrel{?}{=} \frac{\#\Sha \cdot \Omega_E \cdot \mathrm{Reg} \cdot c_{37}}{1^2} = \#\Sha \cdot 5.9869 \cdot 0.0511 \approx 0.3059 \cdot \#\Sha.

Matching $L'(E,1)\approx 0.3059$ yields $\#\Sha = 1$ .

\boxed{r = r_{\mathrm{an}} = 1,\quad \mathrm{Reg}\approx 0.0511,\quad \#\Sha = 1.}

This case is proven by Gross–Zagier–Kolyvagin, not merely numerical.

⚠️ Analytic rank vs. order of vanishing of the Euler product

The $L$ -function Euler product [4] diverges at $s=1$ (the center is inside the critical strip's "edge"); one cannot literally evaluate $L(E,1)$ from the product. Modularity is what gives the analytic continuation needed to even define the order of vanishing. Before Wiles, weak BSD was not even well-posed unconditionally — the left side of [6] presupposed a continuation that was itself conjectural. Always invoke the Modularity Theorem when speaking of $\mathrm{ord}_{s=1}L(E,s)$ .

⚠️ Strong BSD requires finiteness of Sha — which is unknown

Equation [7] contains $\#\Sha(E/\mathbb{Q})$ . If $\Sha$ is infinite, the formula is meaningless. Finiteness of $\Sha$ is known only when $r_{\mathrm{an}}\le 1$ (Kolyvagin, Rubin), never in general. So "verifying strong BSD numerically" in rank $\ge 2$ really means computing the expected $\#\Sha$ from [7] and checking it is a positive integer (and a perfect square, as $\Sha$ carries an alternating Cassels–Tate pairing). A non-integer would disprove BSD.

CounterexampleWhy a naïve "rank = number of small points" fails — Mordell curves

Consider $E: y^2 = x^3 + k$ for varying $k$ . One might guess large rank correlates with abundant small solutions. But ranks can be large while the generating points are enormous: there exist rank- $\ge 28$ curves (Elkies) whose generators have astronomically large height. Conversely, $y^2 = x^3 - 2$ has rank 1 generated by the modest point $(3,5)$ . Moral: rank is not read off from the size or count of visible solutions — only the subtle growth rate [11] or the $L$ -function detects it. The "naïve point-counting intuition" predicts the exponent in [11], never the magnitude of generators.

connection

BSD is the elliptic-curve avatar of a vast web of conjectures.

The Analytic Class Number Formula for Dedekind zeta functions is the structural template: $\zeta_K(s)$ has a pole/zero whose order is the unit rank and whose leading term packages regulator, class number, and discriminant — exactly mirrored by $\mathrm{Reg}$ , $\#\Sha$ , and periods in [7].
The Bloch–Kato Conjecture (Tamagawa numbers of motives) generalizes strong BSD to arbitrary motives; [7] is its $\mathrm{rank}=1$ , weight-2 case.
The Modularity Theorem is the indispensable input giving analytic continuation of $L(E,s)$ .
The Tate–Shafarevich Group measures failure of the Hasse Principle for torsors, linking BSD to local-global questions.
Iwasawa Theory (the Main Conjecture, proven by Mazur–Wiles, Skinner–Urban) relates $p$ -adic $L$ -functions to Selmer groups, yielding the strongest partial results toward strong BSD.
The Congruent Number Problem — whether $n$ is the area of a rational right triangle — is equivalent to a rank statement for $y^2 = x^3 - n^2 x$ , and is solved conditionally on BSD by Tunnell's theorem.

🔬 Where this appears

⏳ History

- 1922–1929 — Mordell proves finite generation of

E(\mathbb{Q})

; Weil generalizes to abelian varieties over number fields (Mordell–Weil Theorem).

- 1933 — Hasse proves the Riemann Hypothesis for elliptic curves over

\mathbb{F}_p

, giving the bound

|a_p|\le 2\sqrt p

.

- 1960–1963 — Birch and Swinnerton-Dyer run EDSAC computations on congruent-number curves, discover the empirical law [11], and formulate the weak conjecture [6].

- 1965 — Publication of the refined (strong) conjecture incorporating period, regulator, and Tamagawa factors; later writings tie in

\Sha

.

- 1977 — Coates and Wiles prove that for CM curves with

r_{\mathrm{an}}=0

, the rank is 0 — the first major theorem toward BSD.

- 1986 — Gross and Zagier prove the height formula [13] linking

L'(E,1)

to Heegner-point heights.

- 1989 — Kolyvagin's Euler-system method proves BSD in analytic ranks 0 and 1 (with finiteness of

\Sha

), conditional on Gross–Zagier.

- 1994–2001 — Wiles, Taylor–Wiles, and Breuil–Conrad–Diamond–Taylor prove the Modularity Theorem, making

L(E,s)

unconditionally entire and weak BSD well-posed.

- 2000 — The Clay Mathematics Institute names BSD a Millennium Prize Problem ($1,000,000).

- 2010s — Skinner–Urban and Wei Zhang prove cases of the Iwasawa Main Conjecture and refine

p

-part of strong BSD; Bhargava–Shankar–Skinner–Zhang show a positive proportion of curves satisfy BSD; average-rank results suggest most curves have rank 0 or 1.

open-problem

The conjecture is open for all analytic ranks $\geq 2$ . No example of a curve with $r_{\mathrm{an}}\ge 2$ is proven to satisfy even weak BSD; in particular finiteness of $\Sha$ is unknown there. Key obstructions and frontier questions:

Rank $\ge 2$ vanishing: Gross–Zagier–Kolyvagin's Euler-system input degenerates because the relevant Heegner point becom

The Birch and Swinnerton-Dyer Conjecture

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits