We transform the problem to a single variable question by the following.
Lemma. Let a1,…,an be nonnegative integers and let G(x) be a nonzero polynomial with degG≤a1+⋯+an. Suppose that some polynomial F(x1,…,xn) satisfies
F(x1,…,xn)=G(x1+⋯+xn)
for (x1,…,xn)∈{0,1,…,a1}×⋯×{0,1,…,an}.
Then F cannot be zero polynomial, and degF≥degG.
For proving the lemma, we will use forward differences of polynomials. If p(x) is a polynomial with a single variable, then define
(Δp)(x)=p(x+1)−p(x).
It is well-known that if p is a nonconstant polynomial then
degΔp=degp−1.
If p(x1,…,xn) is a polynomial with n variables and 1≤k≤n then let
Δk(p)(x1,…,xn)=p(x1,…,xk−1,xk+1,xk+1,…,xn)−p(x1,…,xn).
It is also well-known that either Δkp is the zero polynomial or
deg(Δkp)≤degp−1.
Proof of the lemma. We apply induction on the degree of G, If G is a constant polynomial then we have F(0,…,0)=G(0)=0, so F cannot be the zero polynomial.
Suppose that degG≥1 and the lemma holds true for lower degrees. Since
a1+⋯+an≥degG>0,
at least one of a1,…,an is positive; without loss of generality suppose a1≥1.
Consider the polynomials F1=Δ1F and G1=ΔG. On the grid {0,…,a1−1}×{0,…,a2}×…{0,…,an} we have
F1(x1,…,xn)=F(x1+1,x2,…,xn)−F(x1,x2,…,xn)=G(x1+⋯+xn+1)−G(x1+⋯+xn)=G1(x1+⋯+xn).
Since G is nonconstant, we have
degG1=degG−1≤(a1−1)+a2+⋯+an.
Therefore we can apply the induction hypothesis to F1 and G1 and conclude that F1 is not the zero polynomial and degF1≥degG1. Hence,
degF≥degF1+1≥degG1+1=degG.
To prove the problem statement, take the unique polynomial g(x) so that
g(x)=[mx] for x∈{0,1,…,n(m−1)} and degg≤n(m−1).
Notice that precisely n(m−1)+1 values of g are prescribed, so g(x) indeed exists and is unique. Notice further that the constraints g(0)=g(1)=0 and g(m)=1 together enforce degg≥2.
By applying the lemma to a1=⋯=an=m−1 and the polynomials f and g, we achieve degf≥degg. Hence we just need a suitable lower bound on degg.
Consider the polynomial
h(x)=g(x+m)−g(x)−1.
The degree of g(x+m)−g(x) is degg−1≥1, so
degh=degg−1≥1,
and therefore h cannot be the zero polynomial. On the other hand, h vanishes at the points 0,1,…,n(m−1)−m, so h has at least (n−1)(m−1) roots. Hence,
degf≥degg=degh+1≥(n−1)(m−1)+1≥n.
At the point (b1,…,bn) we have
H(b1,…,bn)=G(d)−G0(d)=0.
At all other points of the grid we have F=G and therefore
H=G−G0=0.
So, by the Alon-Füredi bound,
degH≥b1+⋯+bn=d.
Since degG0<d, this implies
degF=deg(H+G0)=degH≥d=degG.