The torsion correction to an isogeny's cokernel #
Let f : E₁ → E₂ be a nonzero morphism of elliptic curves. Its dual shows that f reflects
torsion: the inverse image of E₂(F)tors is exactly E₁(F)tors. Consequently the cokernel on
points and the cokernel modulo torsion satisfy
#coker(f) · #E₁(F)tors = #coker(f_free) · #E₂(F)tors · #ker(f).
Here the cokernel modulo torsion is expressed as the index of the image in
E₂(F)/E₂(F)tors; the map from E₁(F) has the same image as the induced map from its quotient
by torsion. This avoids choosing bases or splittings of the Mordell–Weil groups.
The kernel is finite for every nonzero morphism. When the source torsion is finite (in particular,
when the source is finitely generated), division gives the rational cokernel/kernel ratio
#coker(f) / #ker(f). When the target is finitely generated, the cokernel is finite too:
its image contains deg(f) · E₂(F).
Since f̂ ∘ f is multiplication by deg f, the cokernel/kernel ratios of f and f̂ multiply to
that of [deg f] on E₁(F). In cross-multiplied form,
#coker([deg f]) · #ker(f) · #ker(f̂) = #coker(f̂) · #coker(f) · #E₁(F)[deg f].
This holds over every field, again with no finiteness hypothesis; when the six quantities are
finite it is the ratio identity. Over a local field, at a place prime to deg f, the ratio for
[deg f] is 1, so the ratios of f and f̂ are inverse.
This is the torsion factor in the kernel/cokernel proof of Cassels' isogeny invariance of the BSD quotient. The regulator, period, local, and Tate–Shafarevich factors are not treated here. No characteristic-zero or separability hypothesis is needed.
Main results #
TauCeti.Isogeny.Hom.comap_torsion_pointMapHom: a nonzero morphism reflects torsion.TauCeti.Isogeny.Hom.ker_pointMapHom_le_torsionandTauCeti.Isogeny.Hom.finite_ker_pointMapHom: the kernel is torsion and finite, without a finite-generation assumption.TauCeti.Isogeny.Hom.finiteIndex_range_pointMapHom: finiteness of the rational cokernel when the target group is finitely generated.TauCeti.Isogeny.Hom.index_range_mul_card_torsion: the cross-multiplied correction formula.TauCeti.Isogeny.Hom.index_range_div_card_ker: the cokernel/kernel ratio#coker(f) / #ker(f), inℚ.TauCeti.Isogeny.Hom.index_range_nsmul_mul_card_ker_mul_card_ker_dual: the cokernel/kernel ratios offandf̂multiply to that of multiplication bydeg f.
References #
- J. H. Silverman, The Arithmetic of Elliptic Curves, III.6, for the dual-isogeny identity.
- J. W. S. Cassels, Arithmetic on curves of genus 1. VIII. On conjectures of Birch and Swinnerton-Dyer, J. Reine Angew. Math. 217 (1965), 180–199, for the arithmetic application.
A nonzero morphism of elliptic curves has exactly the torsion points as the inverse image of the target's torsion subgroup.
Every point in the kernel of a nonzero morphism is torsion.
The kernel on points of a nonzero morphism is finite, as its fibre over zero.
The cokernel on rational points is finite if the target group is finitely generated:
the image contains all deg(f)-multiples, by the dual identity.
The correction between the full cokernel index and the image index modulo torsion. The cross-multiplied identity needs no finite-generation or finiteness hypothesis.
The rational cokernel/kernel ratio #coker(f) / #ker(f) equals the image index modulo torsion
times the ratio of target to source torsion orders. Only the source torsion subgroup needs to be
finite; in particular, finite generation of the source suffices. If the target is finitely
generated, finiteIndex_range_pointMapHom makes the cokernel index a positive integer.
The cokernel/kernel ratios of a morphism and of its dual multiply to that of multiplication
by the degree: on the points over any field,
#coker([deg f]) · #ker(f) · #ker(f̂) = #coker(f̂) · #coker(f) · #E₁(F)[deg f]. No finiteness is
required: infinite orders and indices are read as 0. When all of them are finite, dividing out
gives the identity of ratios.