Local point counts, assembled into a global invariant
For each prime of good reduction, count points on the curve modulo p and record the deficiency ap. These assemble into an Euler product, the L-function of the curve. Modularity guarantees it extends to the whole complex plane, so its value and order of vanishing at s = 1 are meaningful — and Birch and Swinnerton-Dyer conjecture that the order of vanishing is exactly the rank of the group of rational points.
Learning objectives
- Define the local factors and the conductor.
- Write the L-function as an Euler product and explain what modularity provides.
- State the BSD conjecture in its weak and strong forms.
- Use the L-function computationally to predict rank.
- Identify what has been proved and what has not.
Section 01Local data
For a prime p of good reduction set ap = p + 1 − #E(Fp), which Hasse bounds by 2√p. At bad primes the local factor is simpler and is determined by the reduction type.
| Reduction at p | ap | Local factor |
|---|---|---|
| Good | p + 1 − #E(Fp) | 1 − app−s + p1−2s |
| Split multiplicative | 1 | 1 − p−s |
| Non-split multiplicative | −1 | 1 + p−s |
| Additive | 0 | 1 |
Tate's algorithm determines the reduction type at each bad prime and computes the conductor, which records how badly the curve degenerates and is the level of the associated modular form.
Section 02The L-function
The product converges only for Re(s) > 3/2, which places s = 1 outside its region of convergence. The modularity theorem — every elliptic curve over ℚ is modular — supplies analytic continuation and a functional equation, so the value at 1 is well defined.
Before modularity was proved, the statement of BSD presupposed an analytic continuation that was itself conjectural. Wiles's proof for semistable curves and its subsequent extension to all curves over ℚ removed that dependency, making BSD a precise statement about a well-defined function.
Section 03The conjecture
The order of vanishing at 1 equals the rank. This alone would give a practical algorithm for the rank, which is why it is computationally central.
The leading Taylor coefficient equals a product of the regulator, the real period, the Tamagawa numbers and the order of the Tate–Shafarevich group, divided by the square of the torsion order.
Descent gives an upper bound on the rank and finds points giving a lower bound. When the bounds meet, the rank is proved. When they do not, the gap is attributed to the Tate–Shafarevich group, whose finiteness is unproven — so the reported rank is conditional on BSD or on Ш being finite. This qualification belongs in the output.
Section 04Computational use
- Stage 01Compute apCount points modulo many primes, by baby-step giant-step for small p or by Schoof/SEA for large p.
- Stage 02Evaluate L near s = 1Use the rapidly convergent series derived from the functional equation rather than the Euler product.
- Stage 03Read the order of vanishingNumerically estimate how many derivatives vanish; this predicts the rank under BSD.
- Stage 04Search for pointsLook for independent rational points of the predicted number, confirming the lower bound and often completing the determination.
Knowing the analytic rank tells a point search how many independent points to look for and, via the strong form, roughly how large their heights should be. Without that guidance a search may be abandoned just short of finding a generator of large height.
ReferenceFrequently asked questions
Why is the rank hard when the torsion is easy?
Torsion points have integral coordinates bounded by the discriminant, so a finite search suffices. Points of infinite order can have enormous height, with no effective bound known in general — so no finite search is guaranteed to find them.
What is the Tate-Shafarevich group?
It measures the failure of the local-global principle: curves that have points over every completion but none over ℚ. It is conjectured finite, which is known in the rank 0 and 1 cases, and its unknown order is what obstructs descent from being a complete algorithm.
Can the L-function determine the rank rigorously?
Only in the cases where BSD is proved. Numerically the order of vanishing can be estimated with high confidence, and combined with a matching descent bound it often yields a proof — but the numerical estimate alone is not one.
NavigateContinue in this stream
Curated next steps from this page. The site also surfaces algorithmically related reading below.
ProvenanceSources and further reading
This page is an original KEVOS explanatory article. It presents the underlying mathematics — definitions, algorithms, complexity results and selection criteria — in KEVOS editorial voice. No text is reproduced from any copyrighted source. Where numerical tables are relevant, KEVOS links to live authoritative databases rather than republishing static values.