Open MathsAll courses

A fourfold whose cylinder is affine space

Written by Claude Opus 5.5 (Anthropic), October 2026. Self-checked by the writing AI. Public domain (CC0).

Zariski's cancellation problem asks whether an affine variety \(X\) with \(X\times\mathbb A^1\cong\mathbb A^{n+1}\) must itself be \(\mathbb A^n\). In algebraic terms: if adjoining one variable to a finitely generated algebra \(A\) gives a polynomial ring in \(n+1\) variables, is \(A\) a polynomial ring in \(n\) variables? This lesson and the next answer the question negatively in dimension four, in characteristic zero. This lesson constructs the fourfold, a hypersurface in affine five-space given by one explicit equation, and writes down an isomorphism of its cylinder with affine five-space. The next lesson, Cancellation fails in dimension four, proves that the fourfold itself is not affine four-space.

We use Locally nilpotent derivations for exponentials of derivations, and unique factorization in polynomial rings from Integral extensions: lying over, going up and going down. Dimension is computed with Krull dimension and Noether normalization, Theorem 4.2.

Basic references are [OpenAI-cancellation] and [Gupta-survey].

1. The cancellation problem

Let \(k\) be a field and write \(k^{[n]}\) for a polynomial ring in \(n\) variables over \(k\). The cancellation problem asks: if \(A\) is a finitely generated \(k\)-algebra and \(A[w]\cong k^{[n+1]}\) for a variable \(w\), must \(A\cong k^{[n]}\)? An isomorphism of the cylinder need not respect the projection to the line \(w\), so it gives no direct way to find coordinates on \(A\).

For \(n=1\) the answer is yes, by Abhyankar, Eakin and Heinzer [AEH]. For \(n=2\) and \(k\) of characteristic zero it is yes by the work of Fujita, Miyanishi and Sugie [Fujita], [Miyanishi–Sugie]. In positive characteristic, Gupta proved that the answer is no in every dimension \(n\ge3\) [Gupta]. The example of this lesson, found in 2026 [OpenAI-cancellation], answers the question negatively in characteristic zero for \(n=4\). We do not use any of these earlier results.

From now on \(k\) is a field of characteristic zero.

2. The fourfold

Let \(P=k[p,s,u,F,J]\) be a polynomial ring in five variables, and define

\[ \begin{gathered} x=s^2+u^3+p^2F,\\ y=s+x(x-u^3),\\ z=sx+p^2J, \end{gathered}\tag{2.1} \]

and make the definitions (2.2): \[ \begin{aligned} H&=x^2F-(1+2sx)J-p^2J^2\\ &\quad-pu,\\ A&=P/(H). \end{aligned} \]

So \(H\) is one explicit polynomial in the five variables \(p,s,u,F,J\), and \(A\) is the coordinate ring of the hypersurface \(H=0\). We use the same letters for elements of \(P\) and their images in \(A\).

Lemma 2.1. In \(P\),

\[ \begin{aligned} &xy-z(z+1)\\ &\quad =p^2\,(H+pu). \end{aligned}\tag{2.3} \]

Proof. Since \(xy=sx+x^2(x-u^3)\) and \(z(z+1)\) equals \(s^2x^2+2sxp^2J+p^4J^2+sx+p^2J\), the terms \(sx\) cancel and

\[ \begin{aligned} &xy-z(z+1)=x^2(x-u^3-s^2)\\ &\quad -p^2(1+2sx)J-p^4J^2 . \end{aligned} \]

Now \(x-u^3-s^2=p^2F\), so the right side is \(p^2\bigl(x^2F-(1+2sx)J-p^2J^2\bigr)\), which is \(p^2(H+pu)\). \(\square\)

Over the Laurent polynomial ring \(k[p,p^{-1}]\), the elements \(x,y,z,u\) are coordinates on \(P[p^{-1}]\). Indeed (2.1) can be solved:

\[ \begin{gathered} s=y-x(x-u^3),\\ F=p^{-2}\bigl(x-s^2-u^3\bigr),\\ J=p^{-2}\bigl(z-sx\bigr), \end{gathered}\tag{2.4} \]

so \(P[p^{-1}]=k[p,p^{-1}][x,y,z,u]\), with \(x,y,z,u\) algebraically independent over \(k[p,p^{-1}]\). In these coordinates (2.3) reads

\[ \begin{aligned} H&=p^{-2}\bigl(xy-z(z+1)\bigr)\\ &\quad -pu . \end{aligned}\tag{2.5} \]

Proposition 2.2. The ring \(A\) is a domain of dimension four, the image of \(p\) in \(A\) is nonzero, and

\[ A[p^{-1}]=k[p,p^{-1},x,y,z] \]

is a Laurent polynomial ring over a polynomial ring in \(x,y,z\).

Proof. In (2.5) the coefficient of \(u\) is the unit \(-p\), so replacing \(u\) by \(H\) is another change of coordinates of \(P[p^{-1}]\) over \(k[p,p^{-1}]\). Hence \(H\) generates a prime ideal of \(P[p^{-1}]\), and \(P[p^{-1}]/(H)=k[p,p^{-1},x,y,z]\).

Reducing \(H\) modulo \(p\) gives \(x_0^2F-(1+2sx_0)J\) with \(x_0=s^2+u^3\), which is not zero. So \(p\) does not divide \(H\) in \(P\). Factor \(H\) into irreducibles in the factorial ring \(P\). In \(P[p^{-1}]\) it becomes irreducible, so all but one of its irreducible factors become units there, that is, divide a power of \(p\); such a factor is associated to \(p\), which does not divide \(H\). So \(H\) is irreducible, hence prime, and \(A\) is a domain. Since \(A[p^{-1}]\ne0\), the element \(p\) is not zero in \(A\). The fraction field of \(A\) is \(k(p,x,y,z)\), of transcendence degree four, so \(\dim A=4\) by Krull dimension and Noether normalization, Theorem 4.2. \(\square\)

3. An exponential that changes the equation

On \(P[p^{-1}]=k[p,p^{-1}][x,y,z,u]\), let \(\Delta=-p^2\,\partial/\partial u\), the derivation that kills \(p,x,y,z\) and sends \(u\) to \(-p^2\). It is locally nilpotent, by Lemma 1.2 of Locally nilpotent derivations.

Lemma 3.1. The derivation \(\Delta\) maps \(P\) into itself, with

\[ \begin{aligned} &\Delta(p)=0,\quad \Delta(u)\\ &\quad =-p^2,\quad \Delta(s)\\ &\quad =-3p^2xu^2,\quad \Delta(F)\\ &\quad =(6sx+3)u^2,\quad \Delta(J)\\ &\quad =3x^2u^2 . \end{aligned}\tag{3.1} \]

Moreover \(\Delta(H)=p^3\) and \(\Delta^2(H)=0\).

Proof. Apply \(\Delta\) to (2.4), using \(\Delta(x)=\Delta(y)=\Delta(z)=0\). From \(s=y-x(x-u^3)\), \(\Delta(s)=3xu^2\Delta(u)=-3p^2xu^2\). From \(p^2F=x-s^2-u^3\), \(p^2\Delta(F)=-2s\Delta(s)-3u^2\Delta(u)\), which is \(6p^2sxu^2+3p^2u^2\). From \(p^2J=z-sx\), \(p^2\Delta(J)=-x\Delta(s)=3p^2x^2u^2\). These values lie in \(P\), and \(P\) is generated by \(p,s,u,F,J\), so \(\Delta(P)\subset P\). By (2.5), \(\Delta(H)=-p\,\Delta(u)=p^3\), which \(\Delta\) kills. \(\square\)

Extend \(\Delta\) to \(P[w]\) by \(\Delta(w)=0\). By Proposition 2.1 of Locally nilpotent derivations, \(\Phi=\exp(w\Delta)\) is an automorphism of \(P[w]\), and by Lemma 3.1

\[ \Phi(H)=H+w\,\Delta(H)=H+p^3w . \]

Hence \(\Phi\) induces the isomorphism (3.2):

\[ \begin{aligned} A[w]=P[w]/(H)&\xrightarrow{\ \sim\ }T,\\ T&:=P[w]/(H+p^3w). \end{aligned} \]

So it suffices to show that \(T\) is a polynomial ring in five variables. Since \(\Phi\) fixes \(p\) and \(A\) is a domain, \(T\) is a domain in which \(p\ne0\).

4. The cylinder is affine five-space

Linear coordinates adapted to \(p=0\). Put \(x_0=s^2+u^3\) and

\[ \begin{gathered} L=x_0^2F-(1+2sx_0)J,\\ M=(1-2sx_0)F+4s^2J . \end{gathered}\tag{4.1} \]

This is a linear change of the variables \(F,J\) over \(k[p,s,u]\), with determinant \(4s^2x_0^2+(1+2sx_0)(1-2sx_0)=1\). Its inverse is given by (4.2):

\[ \begin{gathered} F=4s^2L+(1+2sx_0)M,\\ J=-(1-2sx_0)L+x_0^2M . \end{gathered} \]

So \(P=B[L]\) with \(B=k[p,s,u,M]\), a polynomial ring in four variables. From now on we regard \(F\), \(J\) and \(H\) as polynomials in \(L\) over \(B\) through (4.2). Expanding \(x=x_0+p^2F\) in powers of \(p\) gives (4.3):

\[ \begin{gathered} H=L-pu+p^2Q+p^4F^3,\\ Q=2x_0F^2-2sFJ-J^2 . \end{gathered} \]

Indeed \(x^2F=x_0^2F+2p^2x_0F^2+p^4F^3\) and \((1+2sx)J=(1+2sx_0)J+2p^2sFJ\), and \(x_0^2F-(1+2sx_0)J=L\).

A root modulo \(p^3\). Write \(C=Q|_{L=0}\in B\) and \(Q=C+L\,Q_1\) with \(Q_1\in B[L]\). Explicitly \(C=M^2\bigl(2x_0+6sx_0^2+4s^2x_0^3-x_0^4\bigr)\). Put

\[ L_*=pu-p^2C\in B . \]

Then (4.3) gives \[ \begin{aligned} &H(L_*)=L_*-pu\\ &\quad+p^2\bigl(C+L_*Q_1(L_*)\bigr)+p^4F(L_*)^3\\ &\quad=p^3\bigl((u-pC)Q_1(L_*)+pF(L_*)^3\bigr), \end{aligned} \] so

\[ H(L_*)\in p^3B . \tag{4.4} \]

The fifth coordinate. In \(B[L,w]\) put

\[ \begin{gathered} h=u-pQ-p^3F^3-p^2w,\\ e_0=-w-pF^3-Q_1h . \end{gathered} \]

Then \(H+p^3w=L-ph\) by (4.3), and a direct expansion, using \(Q=C+LQ_1\), gives the identity (4.5):

\[ \begin{aligned} &p^3e_0-(L-L_*)\\ &\quad=\bigl(p^2Q_1-1\bigr)\bigl(H+p^3w\bigr). \end{aligned} \]

So the image \(e\) of \(e_0\) in \(T\) satisfies \(p^3e=L-L_*\).

Theorem 4.1. \(T=B[e]=k[p,s,u,M,e]\) is a polynomial ring in five variables. Consequently \(A[w]\cong k^{[5]}\).

Proof. \(B\) embeds in \(T\), and \(e\) is transcendental over it. Inverting \(p\) in \(T=B[L,w]/(H+p^3w)\) lets us solve for \(w=-p^{-3}H\), so \(T[p^{-1}]\cong B[p^{-1}][L]\), the identity on \(B\) and \(L\). So \(B\to T\) is injective, and in \(T[p^{-1}]\) the element \(e=p^{-3}(L-L_*)\) is a linear polynomial in the variable \(L\) with unit leading coefficient. Hence \(e\) is transcendental over \(B[p^{-1}]\), and so over \(B\).

\(B\) and \(e\) generate \(T\). We have \(L=L_*+p^3e\in B[e]\). In \(B[Z]\), the difference \(H(L_*+p^3Z)-H(L_*)\) is divisible by \(p^3\), and \(H(L_*)\) is divisible by \(p^3\) by (4.4). So \(W(Z)=-p^{-3}H(L_*+p^3Z)\) lies in \(B[Z]\). The relation \(H+p^3w=0\) in \(T\) gives \(p^3\bigl(w-W(e)\bigr)=0\), and \(p\) is a nonzero element of the domain \(T\), so \(w=W(e)\in B[e]\). Thus \(T=B[L,w]=B[e]\).

Finally \(A[w]\cong T\) by (3.2). \(\square\)

The five coordinates of \(A[w]\) itself are the images of \(\Phi^{-1}(p)=p\), \(\Phi^{-1}(s)\), \(\Phi^{-1}(u)\), \(\Phi^{-1}(M)\) and \(\Phi^{-1}(e_0)\), where \(\Phi^{-1}=\exp(-w\Delta)\); each is an explicit polynomial, because the exponential is a finite sum.

5. What remains

Theorem 4.1 shows that the hypersurface \(H=0\) in \(\mathbb A^5\) becomes affine five-space after multiplying by a line. The equation was chosen so that the hypersurface looks like affine four-space in two other ways as well: after inverting \(p\), Proposition 2.2 identifies it with \(\mathbb A^3\times(\mathbb A^1\setminus0)\), and modulo \(p\) the change of coordinates (4.1) makes \(H\equiv L\) a coordinate. The next lesson proves that nevertheless \(A\) is not a polynomial ring.

6. Exercises

Exercise 6.1 (easy). Verify that the linear change (4.1) has determinant one, and that (4.2) inverts it.

Exercise 6.2 (medium). Show that \(\Delta(H)=p^3\) directly from (3.1), without passing to the coordinates \(x,y,z,u\).

Exercise 6.3 (medium). Show that the fibre \(p=0\) of \(\operatorname{Spec}A\) is affine three-space: \(A/pA\cong k^{[3]}\).

7. Solutions

6.1. The coefficient matrix of \((F,J)\mapsto(L,M)\) is \(\begin{pmatrix}x_0^2&-(1+2sx_0)\\1-2sx_0&4s^2\end{pmatrix}\), with determinant \(4s^2x_0^2+(1+2sx_0)(1-2sx_0)\), which is \(4s^2x_0^2+1-4s^2x_0^2=1\). Its inverse is \(\begin{pmatrix}4s^2&1+2sx_0\\-(1-2sx_0)&x_0^2\end{pmatrix}\), which is (4.2).

6.2. Write \(H=x^2F-(1+2sx)J-p^2J^2-pu\) with \(x=s^2+u^3+p^2F\). Then \(\Delta(x)=2s\Delta(s)+3u^2\Delta(u)+p^2\Delta(F)\), which is \(-6p^2sxu^2-3p^2u^2+p^2(6sx+3)u^2=0\). So \[ \begin{aligned} &\Delta(H)=x^2\Delta(F)-2x\Delta(s)J\\ &\quad-(1+2sx)\Delta(J)-2p^2J\Delta(J)\\ &\quad-p\Delta(u). \end{aligned} \] Substituting (3.1), the first four terms are \[ \begin{aligned} &x^2(6sx+3)u^2+6p^2x^2u^2J\\ &\quad-3(1+2sx)x^2u^2-6p^2x^2u^2J=0, \end{aligned} \] and the last is \(p^3\).

6.3. Modulo \(p\), (4.3) gives \(H\equiv L\), and \(P/pP=k[s,u,M,L]\) by (4.1) and (4.2). So \(A/pA=P/(p,H)\), which is \(k[s,u,M,L]/(L)\cong k[s,u,M]\).

References