Proof. Let the solutions to the congruence equation kx≡y(modp) in S×T be (x1,y1),…,(xn,yn), where n≥1+λs≥1. Hence, we know that S and T are both non-empty, implying that n≥2. Since T is contained in a complete residue system modulo p, we know that the xi's are distinct. Without loss of generality, let's assume x1<x2<⋯<xn. Let ui=xi−x1 and vi=yi−y1. Then we have
kui≡vi(modp),0≤ui≤s,∣vi∣≤t.
Let a=gcd(u2,v2)u2 and b=gcd(u2,v2)v2. Thus, gcd(a,b)=1, and we have
0<a≤s,∣b∣≤t,ka≡b(modp).
Combining with kui≡vi(modp), we obtain
bui≡kaui≡avi(modp).
Let bui=avi+pwi. Then we have
∣wi∣≤p∣b∣s+at.
Let L=p∣b∣s+at.
(1) If L<1, then wi=0 for all i, and hence bui=avi. Since gcd(a,b)=1, it follows that ui must be multiples of a. Combining with the fact that ui∈[0,s] are distinct non-negative integers, we have
s≥maxui≥a(n−1)≥aλs,
which implies a≤λ1. Furthermore, from
at≥max∣avi∣=max∣bui∣≥∣b∣⋅(aλs),
we have ∣b∣≤λst.
(2) If L≥1, then wi can take at most 2L+1≤3L integer values in the range [−L,L]. By the pigeonhole principle, at least 3Ln of the wi's take the same value. Let E be the set of indices i where wi takes the same value, and let ui0 be the smallest value in {ui:i∈E}. From
b(ui−ui0)=a(vi−vi0)+p(wi−wi0)=a(vi−vi0),∀i∈E,
we obtain a∣ui−ui0. Combining 0≤ui−ui0=xi−xi0≤s, we have
s≥i∈Emax(ui−ui0)≥a(∣E∣−1).
Also, from ∣vi−vi0∣=∣yi−yi0∣≤t, we have
at≥max(a∣vi−vi0∣)=∣b∣i∈Emax(ui−ui0)≥∣b∣(a(∣E∣−1)).
Note that
L=p∣b∣s+at≤p2st<6λs,
which implies 1<6Lλs. Thus, we have
∣E∣−1≥3Lλs−6Lλs=6Lλs.
Substituting back into the previous inequalities, we obtain
λa≤6L,λ∣b∣s≤6Lt,
which leads to
λp=Lλ∣b∣s+λat≤12t,
contradicting the condition t<12λp!
Thus, we have shown that L≥1 is not possible, and only L<1 holds true in this case. □