Take logarithms of a curve and watch it cast an amoeba-shaped shadow that crystallizes into piecewise-linear bones
Here is a strange thing to do to a curve. Take a polynomial equation in two complex variables and look at its solution set — a surface living in four real dimensions, impossible to picture directly. Now throw away all the phase information and keep only the magnitudes: send each solution (x, y) to the pair of real numbers (log|x|, log|y|). What comes back is a picture in the ordinary plane.
It looks like an amoeba. A blobby body with tentacles reaching off to infinity, and often holes in the middle. The name was coined by Gelfand, Kapranov and Zelevinsky in 1994, and it is exactly right. The tentacles are not decorative: their directions are dictated by the Newton polygon of the polynomial you started with.
Now squeeze. Take logarithms in a larger and larger base and the amoeba slims down, the body shrinking toward the bones, until in the limit nothing is left but a graph of straight line segments — a tropical curve. The curve has become a piece of combinatorics, and remarkably, it remembers enough about the original to answer real questions about it.
The amoeba of a line, a conic, and a cubic. Drag the coefficient moduli and watch the body swell and the holes open and close. Everything outside the shaded region is certified empty by the lopsidedness test: if one term of the polynomial is bigger than all the others put together, no cancellation is possible, so no solution can sit above that point.
Bright fuchsia: a zero of f was found on the torus fiber over (log|x|, log|y|) — the point is in the amoeba. Faint fringe: not lopsided (no single monomial dominates), the honest outer approximation of the amoeba. Dark: one monomial outweighs all the others combined, so f cannot vanish there. Three tentacles — west, south, and northeast — dual to the three edges of the Newton triangle.
Far from the origin of the log plane, one monomial of the polynomial usually dominates everything else, and a sum with one dominant term cannot vanish. So the amoeba can only reach out to infinity along the directions where two monomials stay comparable — and those directions are determined by the Newton polygon, the convex hull of the exponent vectors (i, j) of the monomials appearing in the polynomial.
Precisely: the amoeba has one tentacle for each edge of the Newton polygon, pointing in the direction of that edge’s outward normal. A line 1 + x + y has a triangular Newton polygon with three edges, so its amoeba is a three-tentacled blob. Nothing about the coefficients can change that count; it is fixed by which monomials are present.
The holes are subtler and depend on the actual coefficients. A theorem of Forsberg, Passare and Tsikh says the holes of an amoeba are indexed by the lattice points of the Newton polygon, with each hole corresponding to a different order of vanishing. Push the coefficients around in the demo above and you can watch holes open and close as lattice points switch roles.
The main event. Take logarithms in base t and let t grow. The amoeba narrows toward a skeleton of straight segments — and the overlaid tropical curve, computed exactly from the combinatorics, is the shape it is heading for. This limiting process is called Maslov dequantization.
The coefficient of each monomial is t raised to your chosen exponent (the Viro family), and the picture is the amoeba in base-t logarithm coordinates — recomputed honestly at every t, since the shape genuinely changes as the coefficients move. The cyan skeleton is the exact tropical curve: the locus where the maximum of c + i·X + j·Y is achieved twice. As t grows, the amoeba deflates onto it.
Watch what happens to arithmetic under the squeeze. Write numbers as powers of t, say ta and tb. Multiplying them gives ta+b — so multiplication turns into addition of exponents. Adding them gives roughly tmax(a,b), since the bigger power swamps the smaller — so addition turns into taking a maximum (or, with the opposite sign convention, a minimum).
That is the whole of tropical arithmetic: a semiring in which “plus” means take the minimum and “times” means ordinary addition. A tropical polynomial is therefore a minimum of affine functions, which is a piecewise-linear function, and the tropical curve is its corner locus — the set of points where the minimum is achieved twice over and the function fails to be smooth.
The name has nothing to do with the mathematics. It honours the Brazilian computer scientist Imre Simon, who pioneered min-plus algebra; French colleagues began calling it “tropical” in his honour, and the label stuck.
Skip the limit and work tropically from the start. Choose a degree, push the coefficients, and watch the curve rearrange itself. Every edge of the curve is perpendicular to an edge of a subdivided Newton polygon — hover either panel to see the correspondence light up in the other.
Every vertex is balanced: give each edge leaving it a primitive integer direction vector, weight it by the lattice length of its dual edge, and the vectors sum to zero. That single condition is what makes tropical curves behave like algebraic ones — it is the tropical shadow of the fact that a polynomial has as many roots as its degree. Push the sliders far enough and edges collapse: the curve degenerates exactly when the dual subdivision stops being a full triangulation.
Flattening a curve into a stick figure sounds like losing everything. The surprise is how much survives. Tropical curves obey their own version of Bézout’s theorem: two tropical curves of degrees m and n meet in mn points, counted with the right multiplicities. They have a genus, given by the number of independent loops in the graph. They satisfy a balancing condition at every vertex that is the exact shadow of an algebraic identity.
The payoff came in 2005, when Grigory Mikhalkin proved his correspondence theorem: counting complex algebraic curves through a set of points gives the same answer as counting tropical curves through the corresponding tropical points. A hard problem in complex geometry became a finite combinatorial search over graphs. Enumerative problems that had resisted classical attack fell to counting stick figures.