A Parameterization of Newton Polytope of a Polynomial with All Positive Coefficients
Consider a polynomial p=∑IaIxI with aI≥0 for all I , where xI=i=1∏NxiniI, its Newton polytope N(p) is defined as the convex hull of vectors {nI=(n1I,…,nNI)} .
Now consider the `scattering map’ X(x)=∂logx∂logp, where xi∈(0,∞) for all i . We claim that X is an one-to-one map between (0,∞)N and the interior of N(p) .
Firstly, suppose the matrix (niI) does not have full rank, i.e. there exists s=0 such that s⋅nI=0 for all I . In this case, s⋅X=s⋅∂logx∂logp=0. Therefore, one can project im(X) by setting some xi=1 so that there’s no more vector s such that s⋅nI=0 for all I . Conversely, one can recover the original image by solving these equations s⋅X=0 . Therefore, we can assume that the matrix (niI) has full rank, then dim(N(p))=N .
Next we show that interior points of N(p) are in the im(X) .
Let Λ be a interior point in N(p) defined by Λ=I∑λInI=∂logx∂I∑λIlogxI where ∑IλI=1 and λI>0 . Equations X(x)=Λ are 0=∂logx∂(logp−I∑λIlogxI)=∂logx∂(logF(x))=F(x)1∂logx∂F(x) where F(x)=I∑aIxIJ∏(x−J)λJ. Let y=logx , and then F(y)=I∑aIexp(y⋅(nI−Λ)) so we only need to show that F(y) has saddle points in RN because F>0 for all y .
To see it, we firstly notice that F(y) is a strict convex function of y . In fact, the Hessian of F(y) is Hij(y)=∂yi∂yj∂2F(y)=I∑aIexp(y⋅(nI−Λ))(nI−Λ)i(nI−Λ)j. For any v=0 , i,j∑vivjHij=I∑aIexp(y⋅(nI−Λ))(v⋅(nI−Λ))2>0. It cannot vanish because that it only happens when v⋅(nI−Λ)=0 for all I , but we have assumed {nI} has full rank.
For a strict convex function F(y) on RN , it has a unique minimal in RN iff it does not take its minimal when y goes to infinity along any direction. In our case, we only need that for any y=0 and Λ∈(N(p))∘ , there exist some I such that y⋅(nI−Λ)>0 and if Λ∈N(p) , there exist some y=0 such that the exponentials are all negative, then F(ty)→0 when t→∞ .
In fact, if Λ∈(N(p))∘ , for a given y=0 , the hyperplain L with normal vector y crossing Λ divides the space RN into semi-spaces L+ and L− , then nI in L+ are what we are looking for. Conversely, if Λ∈N(p) , one can find a hyperplain L such that N(p)⊂L− , then the normal vector of L gives the wanted direction. These are direct results of convexity of N(p) .
Therefore, we have proven that F(y) has a unique minimal in RN when Λ∈(N(p))∘ and no minimal in RN when Λ∈N(p) . Hence, X(x)=∂logx∂logp, is an one-to-one map between (0,∞)N and the interior of N(p) .
Generally, let Λ be a interior point in Minkowski sum ∑pαpN(p) corresponding to P=∏ppαp defined by Λ=p∑αpΛp=p∑αpIp∑λIpnIp=∂logx∂p∑αpIp∑λIplogxIp where ∑IpλIp=1 and λIp>0 . Equations X(x)=Λ are 0=∂logx∂⎝⎛logP−p∑αpIp∑λIplogxIp⎠⎞=∂logx∂(log(p∏Fpαp)), where Fp(y)=Ip∑aIpexp(y⋅(nIp−Λp)). Write F(y)=p∏Fp(y)αp, we only need to show that F(y) has a unique saddle point in RN .
We assume that αp>0 for all p . Here, F(y) is no longer a strict convex function. However, take a positive number α0 such that α0<αp for all p , consider a new function F(y)1/α0=p∏Fp(y)αp/α0, since Fp are convex functions and αp/α0>1 , Fp(y)αp/α0 are convex functions, so is their product. Since F(y)1/α0 has the same minimums as F(y) , so we can further assume that αp>1 for all p .
Finally, we claim that F(y) does not take minimum when y goes to infinity along any direction. It is because that for a given y and each p , Fp(ty)→∞ when t→∞ , so does F(ty)=∏pFp(ty)αp . Therefore, equation X(x)=Λ has a unique solution.