For a natural n consider the hyperplane
R0n={x=(x1,x2,…,xn)∈Rn:i=1∑nxi=0}
and the lattice
Z0n={y∈R0n:all yi are integers}. Define the (quasi–)norm in Rn by
∥x∥p=(i=1∑n∣xi∣p)1/p if
0<p<∞, and ∥x∥∞=imax∣xi∣.
a) Let x∈R0n be such that
imaxxi−iminxi≤1.
For every p∈[1,∞] and for every y∈Z0n
prove that
∥x∥p≤∥x+y∥p.
b) For every p∈(0,1), show that there is an n and an
x∈R0n with imaxxi−iminxi≤1
and an y∈Z0n such that
∥x∥p>∥x+y∥p.
Solution (official)
a) For x=0 the statement is trivial. Let x=0. Then
imaxxi>0 and iminxi<0. Hence
∥x∥∞<1. From the hypothesis on x it follows that:
i) If xj≤0 then imaxxi≤xj+1.
ii) If xj≥0 then iminxi≥xj−1.
Consider y∈Z0n, y=0. We split the indices
{1,2,…,n} into five sets:
I(0)={i:yi=0},I(+,+)={i:yi>0,xi≥0},I(+,−)={i:yi>0,xi<0},I(−,+)={i:yi<0,xi>0},I(−,−)={i:yi<0,xi≤0}.
As least one
of the last four index sets is not empty. If I(+,+)=∅ or
I(−,−)=∅ then ∥x+y∥∞≥1>∥x∥∞. If
I(+,+)=I(−,−)=∅ then ∑yi=0 implies
I(+,−)=∅ and I(−,+)=∅. Therefore i) and ii)
give ∥x+y∥∞≥∥x∥∞ which completes the case
p=∞.
Now let 1≤p<∞. Then using i) for every j∈I(+,−) we
get ∣xj+yj∣=yj−1+xj+1≥∣yj∣−1+imaxxi.
Hence
∣xj+yj∣p≥∣yj∣−1+∣xk∣pfor every k∈I(−,+) and j∈I(+,−).
Similarly
∣xj+yj∣p≥∣yj∣−1+∣xk∣pfor every k∈I(+,−) and j∈I(−,+);∣xj+yj∣p≥∣yj∣+∣xj∣pfor every j∈I(+,+)∪I(−,−).
Assume that j∈I(+,−)∑1≥j∈I(−,+)∑1. Then
∥x+y∥pp−∥x∥pp=j∈I(+,+)∪I(−,−)∑(∣xj+yj∣p−∣xj∣p)+j∈I(+,−)∑∣xj+yj∣p−k∈I(−,+)∑∣xk∣p+j∈I(−,+)∑∣xj+yj∣p−k∈I(+,−)∑∣xk∣p≥j∈I(+,+)∪I(−,−)∑∣yj∣+j∈I(+,−)∑(∣yj∣−1)+j∈I(−,+)∑(∣yj∣−1)−j∈I(+,−)∑1+j∈I(−,+)∑1=i=1∑n∣yi∣−2j∈I(+,−)∑1=2j∈I(+,−)∑(yj−1)+2j∈I(+,+)∑yj≥0.
The case
j∈I(+,−)∑1≤j∈I(−,+)∑1 is
similar. This proves the statement.
b) Fix p∈(0,1) and a rational t∈(21,1). Choose a
pair of positive integers m and l such that mt=l(1−t) and set
n=m+l. Let
xi=t,i=1,2,…,m;xi=t−1,i=m+1,m+2,…,n;yi=−1,i=1,2,…,m;ym+1=m;yi=0,i=m+2,…,n.
Then x∈R0n,
imaxxi−iminxi=1,
y∈Z0n and
∥x∥pp−∥x+y∥pp=m(tp−(1−t)p)+(1−t)p−(m−1+t)p,
which is possitive
for m big enough.