Say I have two normal lattice polytopes $P$ and $Q$ in $\mathbb{R}^n$ (lattice $\mathbb{Z}^n$) with the same number of lattice points $N+1$. Then they define two toric varieties $X_P$ and $X_Q$ which are embedded via $P$, resp. $Q$, into a projective space $\mathbb{P}^N$.

I'm pretty sure that it is true that $X_P$ and $X_Q$ are projectively equivalent if and only if $P$ and $Q$ are affinely isomorhpic as lattice polytopes (so you get one from the other by a composition of a translation and a lattice automorphism in ${\rm GL}(n,\mathbb{Z})$).

I couldn't find a reference for that statement. Can anyone point me in the right direction? Or does anyone know a short and concise proof?


First of all, the assertion is not quite true if $X_P$ and $X_Q$ are not embeddings of the torus $T=\mathbf G_m^n$. Let, e.g., $n=3$, $P=conv(0,e_1,e_2,e_3)$, $Q=conv(0,e_1,e_2,e_1+e_2+2e_3)$. Then $N+1=4$ and $X_P=X_Q=\mathbf P^3$.

So assume that $X_P$ and $X_Q$ are embeddings. This means that the lattice $\mathbb Z^n$ is generated by the differences of lattice points. . Then the corresponding homomorphisms $p,q:T\to PGL(N+1)$ are injective. Let $G=\{g\in PGL(N+1)\mid g(X_P)=X_P\}$. Then $G\to Aut(X_P)$ is injective since $X_P$ spans $\mathbb P^N$. Thus $p(T)$ is a maximal torus of $G$. Indeed $T$ is already a maximal torus in the entire Cremona group (see Demazure's paper).

Now assume that $X_P$ and $X_Q$ are projectively equivalent. Then there is $a\in PGL(N+1)$ with $a(X_Q)=X_P$. Then, we get two injective homomorphisms $T\to PGL(N+1)$ namely $p$ and $\bar q:=aqa^{-1}$. Since both $p(T)$ and $\bar q(T)$ are maximal tori of $G$, there is $b\in G$ with $p(T)=b\bar q(T)b^{-1}=cq(T)c^{-1}$ with $c=ba$. It follows that there is an automorphism $\alpha$ of $T$ with $p(\alpha(t))=cq(t)c^{-1}$. Now replace $P$ by $\alpha(P)$. Then we may assume that $\alpha=id_T$. Then $p$ and $q$ are conjugated by an element $c\in PGL(N+1)$.

Finally observe that the projective representation $p:T\to PGL(N+1)$ uniquely determines the weights of $T$ on $\mathbb C^{N+1}$ up to a translation. This set of weights is nothing else than the lattice points in $P$. So the lattice points of $P$ and $Q$, and therefore $P$ and $Q$ themselves, differ by a translation.

  • $\begingroup$ Thank you for your answer. I will go through the representation theory carefully. I swept the fact that the torus is actually embedded in the toric variety under the rug. If the polytope is normal, that is the case. The polytope Q in your example is not a normal lattice polytope. $\endgroup$ – Rainer Sinn Jul 28 '16 at 7:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.