Backporch

Pillars · deep dive

Affine space & affine varieties

Forget where the origin is, and the natural maps become the affine ones — a linear part and a shift. Ask which points make a system of polynomials vanish and you get a variety, paired with an ideal of equations by a clean dictionary. Complete the plane with points at infinity and a single cubic becomes an elliptic curve that carries a group — whose rank is the deepest unsolved question about it. The geometric floor under the elliptic-curve and modular-forms climb — Cox's Ideals, Varieties, and Algorithms for the dictionary, Silverman's Arithmetic of Elliptic Curves for the payoff.

1 · The simplest affine map: $x \mapsto a x + b$

Start on the line. A linear map fixes the origin ( $x \mapsto a x$ ); allow a shift and you have an affine map. Slide the stretch $a$ and the shift $b$ , and watch a line stay a line:

the space

A^{1}

its image under

x \mapsto a x + b

fixed point

f (x) = 1.5 x + 1.0

· fixes

x^{*} = - 2

a

1.5

b

1.0

An affine map is a linear part (the stretch $a$ , a flip if $a < 0$ ) plus a shift $b$ — the most general map that sends lines to lines and keeps ratios along them (watch the equal gaps stay equal). Set $a = 1, b = 0$ and nothing moves; otherwise one fixed point $x^{*} = b / (1 - a)$ stays put while the rest slides past it. It is exactly what the calculator does to a variable ( $κ_{1} \mapsto a κ_{1} + b, κ_{n} \mapsto a^{n} κ_{n}$ ), and standardizing $Z = (X - μ) / σ$ is the affine map to mean $0$ , variance $1$ .

2 · Affine space — a vector space that forgot its origin

Affine space $A^{n}$ is $R^{n}$ with the zero demoted — no privileged origin, no adding points. What survives: the difference of two points is a vector, and affine combinations $\sum_{i} λ_{i} P_{i}$ with $\sum_{i} λ_{i} = 1$ (weighted averages — the segment, the triangle).

points

P, Q

(no origin) · difference

P - Q

is a vector · midpoint

\frac{1}{2} P + \frac{1}{2} Q

The maps that respect this are the affine maps $x \mapsto A x + b$ . Pick an origin and $A^{n}$ becomes the vector space $k^{n}$ — the change of basis move, now free to move the origin too.

3 · Affine varieties — and the ideal ⇄ variety dictionary

Now put equations on the space. An affine variety is the common zero set of polynomials, $V (f_{1}, \dots, f_{s}) = {P \in A^{n} : f_{1} (P) = \dots = f_{s} (P) = 0} .$ A line is $V (a x + b y - c)$ , a circle is $V (x^{2} + y^{2} - r^{2})$ , and a cubic $V (y^{2} - x^{3} - a x - b)$ is the curve we are heading for. The deep move is a dictionary betweengeometry and algebra: to a variety $V$ attach the ideal $I (V)$ of all polynomials vanishing on it, and to an ideal attach its zero set $V (I)$ . Hilbert's Nullstellensatz makes the loop nearly exact over an algebraically closed field: $I (V (J)) = J$ . Geometry becomes computable — intersect varieties by adding ideals, project by eliminating variables — and the engine that runs it is the Gröbner basis (this is the heart of Cox, Little & O'Shea).

slide B →0.00

V (x^{2} + y^{2} - 2)

\cap

V (y - 0.55 x)

2 real points · Bézout 2 over

C

Each equation is a shape — the variety $V (f)$ , the points where $f = 0$ . Two shapes meet exactly where both equations hold, so the gold points are the solutions of the system ${f = 0, g = 0}$ — the single most common task in applied math (GPS is two circles meeting; robot kinematics, CAD, and optimization are all polynomial systems). Slide B and watch two real solutions slide together, touch (a double root — tangency), then vanish: they became complex. By Bézout, the count over $C$ is always $de g f \cdot de g g$ — the real plane just hides some. Pick Cubic and a Line: three intersections — the chord of the group law. That is all algebraic geometry is: equations in, shapes out, and the ideal ⇄ variety dictionary translating between them.

One thing the explorer hides: which equation? A shape is cut out by infinitely many, and the ideal $I (V)$ is the canonical object that holds them all — watch four polynomials trace the same circle:

V ((x^{2} + y^{2} - 1) (x^{2} + y^{2} + 2))

same circle — vanishes everywhere on it, so it lives in

I (V) = ⟨ x^{2} + y^{2} - 1 ⟩

Four different polynomials, one circle. The shape — the variety $V$ — does not pin down the equation; infinitely many polynomials vanish on it, and together they form the ideal $I (V) = {f : f ∣_{V} = 0}$ . Here every one is a multiple of a single generator, $I (circle) = ⟨ x^{2} + y^{2} - 1 ⟩$ — closed under addition and under multiplying by anything (that is what makes it an ideal, not just a set). Bump the constant to $x^{2} + y^{2} - 2$ and the shape moves — a different ideal. This is why algebraic geometry carries the ideal, not a chosen equation: $V$ remembers the shape, $I (V)$ remembers everything that vanishes — and the Nullstellensatz $I (V (J)) = J$ says the ideal is exactly the radical, multiplicities and all.

4 · Completing the picture — points at infinity

Affine space has a defect: parallel lines never meet. Projective space $P^{n}$ repairs it by adjoining one point at infinity per direction — formally, lines through the origin in $A^{n + 1}$ , in homogeneous coordinates $[X_{0} : \dots : X_{n}]$ .

lines through

O

= points of

P^{1}

· the affine chart sees them · the one parallel direction =

\infty

Homogenizing lifts a variety to its projective closure and Bézout holds exactly: a line meets a degree- $d$ curve in $d$ points (with multiplicity, counting infinity). For our cubic, that one extra point $O = [0 : 1 : 0]$ is what makes the curve a group.

5 · Elliptic curves — the chord-and-tangent group law

An elliptic curve is a smooth projective cubic with a chosen base point — affinely, $y^{2} = x^{3} + a x + b$ with discriminant $- 16 (4 a^{3} + 27 b^{2}) \neq = 0$ , plus the point at infinity $O$ . Its points form an abelian group: to add $P$ and $Q$ , draw the line through them, take its third intersection with the curve, and reflect across the $x$ -axis. The tangent line doubles a point; the vertical line sends a point to its inverse, with $O$ the identity. Drag $P$ and $Q$ and watch the law work — this is an affine variety that is also a group.

P

Q

R

— third intersection $P + Q$ — reflect

R

P + Q = (reflect R) =

(- 0.55, - 1.18)

a

-1.00

b

1.00

x_{P}

0.62

x_{Q}

0.18

m = \frac{y _{Q} - y _{P}}{x _{Q} - x _{P}} (or \frac{3 x _{P}^{2} + a}{2 y _{P}} if P = Q), x_{3} = m^{2} - x_{P} - x_{Q}, P + Q = (x_{3}, m (x_{P} - x_{3}) - y_{P})

An affine variety — the zero set $V (y^{2} - x^{3} - a x - b)$ in the plane $A^{2}$ — that secretly carries a group. A line meets a cubic in three points; call the third R, reflect it, and you have P+Q. A vertical line's third point is the one we adjoined when we passed to $P^{2}$ — the point at infinity $O$ , which acts as the identity. The chord gives addition, the tangent gives doubling, and associativity (the hard axiom) is a theorem. Count the points of this same equation over a finite field and you get $a_{p} = p + 1 - # E (F_{p})$ — the very numbers that become traces of Frobenius in the Eigenbook. Affine geometry (Cox), arithmetic payoff (Silverman).

6 · Two ranks — the Birch–Swinnerton-Dyer conjecture

Read the curve over a finite field and its point-count is an arithmetic fingerprint, $a_{p} = p + 1 - # E (F_{p})$ with $∣ a_{p} ∣ \leq 2 p$ (Hasse). Hidden in those counts is the deepest invariant — the rank, how many independent rational points the curve carries — and Birch–Swinnerton-Dyer says that algebraic rank equals the analytic rank, the order of vanishing of $L (E, s)$ at $s = 1$ . Their evidence was a growth law you can watch: $\prod_{p \leq X} N_{p} / p \sim C (lo g X)^{r}$ . Pick a curve and see the product fan apart by rank.

∏ N_p/p — growth rate = rank

\prod_{p \leq X} N_{p} / p

(log scale) vs prime bound

X

· the curves separate by rank

prime bound X400

r=0

\prod =

3.3r=1

\prod =

48.9r=2

\prod =

192.7r=3

\prod =

936.5

y^{2} + y = x^{3} - x

· 37a — smallest rank 1

a_{2}

-2

a_{3}

-3

a_{5}

-2

a_{7}

-1

a_{11}

-5

a_{13}

-2

a_{17}

a_{19}

0…

algebraic rank

= 1

— independent rational points

= BSD

analytic rank

= ord_{s = 1} L (E, s)

= 1

Two numbers that look unrelated. The algebraic rank counts independent rational points — global, infinite, hard. The analytic rank is how hard the L-function $L (E, s) = \prod_{p} (1 - a_{p} p^{- s} + p^{1 - 2 s})^{- 1}$ vanishes at $s = 1$ — an analytic property of a generating function in the $a_{p}$ . Birch–Swinnerton-Dyer: they are equal — and the rank shows up as the growth rate of the point-count product, $\prod_{p \leq X} N_{p} / p \sim C (lo g X)^{r}$ , which is exactly the fan you are watching. Local counts mod $p$ forecast the global rank: that is the miracle. Honest status: BSD is a Millennium Prize problem, open in general — proven only for analytic ranks 0 and 1 (Gross–Zagier, Kolyvagin). The bridge to the Eigenbook: these same $a_{p}$ are traces of Frobenius.

← All deep dives The Eigenbook →