Bourbaki marks this one as harder than the rest.
The hypotheses being those of Exer. 27 a), we suppose in addition that there exists a μ-integrable G-tiling F.
a) For every numerical μ-integrable function f≥0 defined in X, put f~(x)=∑s∈Gf(sx), so that ∫Ff~(x)dμ(x)=∫Xf(x)dμ(x) § 2, No. 10, Prop. 15). Let (fi)1≤i≤n be a finite family of numerical functions ≥0, μ-integrable in X; put, for i=j,
mij=∫Ff~i(x)f~j(x)dμ(x).
and ci=supx∈Xf~i(x). Now prove that there exists at least one couple of distinct indices (i,j) such that
mij≥n(n−1)μ(F)1((i=1∑n∫Xfi(x)dμ(x))2−μ(F)i=1∑nci∫Xfi(x)dx)
(Minorize the sum ∑i=jmij by using the Cauchy-Schwarz inequality).
Deduce that if (Ai)1≤i≤n is a finite family of μ-integrable G-fillings, and if ∑i=1nμ(Ai)>μ(F), there exists a couple (i,j) of distinct indices and an s∈G such that μ(Ais∩Aj)>0.
b) If G0 is a subgroup of G of finite index (G:G0)=h, and if (s1,…,sh) is a system of representatives of the left classes mod G0 in G, F0=⋃1≤i≤hsiF is a G0-tiling.
c) In X=Rn, let A be a convex symmetric body; let ui:(xj)→∑j=1ncijxj, be m linear forms on X, with integer coefficients cij (m<n).
Now prove that, for every integer p≥1 and every number r≥0 such that μ(A)rn≥2npm, there exists a point x∈rA, distinct from 0 and such that ui(x)≡0 (mod. p) for 1≤i≤m (apply Minkowski's theorem to the subgroup G0 of Zn formed by the z∈Zn such that ui(z)≡0 (mod. p) for 1≤i≤m, and use b)). Particular case where n=2, m=1, and A is defined by ∣x1∣≤1, ∣x2∣≤1 (Thue's theorem).