Discriminant and behavior at infinity#
The cubic model identifies exactly where inverse branches escape.
Discriminant hypersurface#
For
the discriminant is \(4\Delta(a,b,c)\), where
Thus \(\Delta=0\) is the locus where the binary cubic has a multiple root.
Explicit escape through every discriminant point#
Every point of \(\Delta=0\) has a repeated finite root \(\rho\) and can be written
For \(t\ne0\), define
Exact substitution gives
As \(t\to0\), the source escapes to infinity and the image converges to the chosen discriminant point. Hence every point of \(V(\Delta)\) is a nonproperness value.
This family can also be read as a uniform perturbation proof. If \(\rho\) is a repeated root at \((a_0,b_0,c_0)\), set
Then \(\rho\) stays a root and its derivative becomes \(\varepsilon\), so its inverse has \(x=2/\varepsilon\) and escapes.
Conversely, let \(I=\{Q=0\}\) be the full projective-root incidence inside \(\mathbb A^3\times\mathbb P^1\). The projection \(I\to\mathbb A^3\) is projective and quasi-finite, hence finite. Over \(\Delta\ne0\), all roots are simple, so \(I=I^{\mathrm{sm}}\) and the marked-root isomorphism identifies this finite étale map with the restriction of \(F\). All root-gradient normalizations and the inverse formulas remain bounded in a neighborhood of a limiting target, so no inverse branch can escape. Therefore the nonproperness set is exactly
The simplest escaping curve#
Taking \(\rho=0,c=0\), and replacing \(t\) by \(1/s\), gives
As \(|s|\to\infty\), the source escapes and the image tends to the origin.
Why nonproperness is not an objection#
The Keller condition says that \(F\) is étale: it has no critical points and is locally invertible. It does not assume properness. A proper étale self-map of complex affine space would be a finite covering and global topology would force degree one. The Jacobian conjecture asserted, in effect, that polynomiality plus the constant-Jacobian condition was already strong enough to prevent escape at infinity.
This example shows exactly the opposite: the discriminant is where two or three otherwise finite inverse branches run to infinity. The escape curve is the mechanism of the counterexample, not a loophole in it. This is consonant with the long-standing focus on nonproperness and ramification at infinity [Jelonek, 1993, Rodríguez Díaz, 2026, Truong, 2026].