All fantastic sequences are: xk=a(klog(k)−(k−1)log(k−1))+c, where a is a nonnegative real number, and c is an arbitrary real number.
First it is easy to see that if {x1,x2,⋯,xk,⋯} is fantastic, then for any real number c and positive real number r, {rx1+c,rx2+c,⋯,rxk+c,⋯} is also fantastic.
Therefore, without loss of generality assume x1=0. Let f(p)=∑i=1pxi, then it is easy to see that H(pq)=pf(p)−f(q)−f(p−q).
Since H(pq)=H(kpkp), therefore
pf(p)−f(q)−f(p−q)=kpf(kp)−f(kq)−f(k(p−q))
After rearranging terms and simplifying we obtain
(f(kp)−kf(p))=(f(kq)−kf(q))+(f(k(p−q))−kf(p−q))
which holds for all positive integers k,p and non-negative integers 0≤q≤p. Now fix a positive integer k, and define gk(p)=f(kp)−kf(p), then the above formula becomes
gk(p)=gk(q)+gk(p−q)
This is a standard Cauchy equation; setting q=1 and using mathematical induction, we get gk(p)=pgk(1)=p(f(k)−kf(1))=pf(k) (since f(1)=x1=0), therefore
f(kp)=kf(p)+pf(k)
Furthermore, if we fix p in this formula and use mathematical induction, we can obtain that for any positive integers p,k, f(pk)=pk−1kf(p).
For two positive integers q<p, since p+q+1q<p+q+1q+1≤21 and by the increasing property of H(x), we have
p+q+1f(p+q+1)−f(q)−f(p+1)≤p+q+1f(p+q+1)−f(q+1)−f(p)
After rearranging we obtain
f(p+1)−f(p)≥f(q+1)−f(q)
which holds for all p>q. Note that this means f(p) is a convex function.
From this, we can obtain an obvious inequality: for positive integers p>q>r,
p−qf(p)−f(q)≥p−rf(p)−f(r)≥q−rf(q)−f(r)
This is because
≥===p−qf(p)−f(q)p−q(p−q)(f(q+1)−f(q))f(q+1)−f(q)q−r(q−r)(f(q+1)−f(q))q−rf(q)−f(r)
And since
p−rf(p)−f(r)=p−rp−qp−qf(p)−f(q)+p−rq−rq−rf(q)−f(r)
therefore its value lies between the two.
Now suppose f(2)=a is known, then a=f(2)=2f(2)−f(1)−f(1)=H(21)≥H(0)=0. According to f(pk)=pk−1kf(p), we can deduce f(2k)=2k−1ka, that is, when p=2k, f(p)=2aplog2p. When p=2k, we hope to estimate the value of f(p) using inequalities. For any positive integer m, there exists a positive integer n such that 2n+1>pm>2n, therefore
2n+1−2nf(2n+1)−f(2n)≥pm−2nf(pm)−f(2n)
After rearranging and substituting f(2k)=2k−1ka we obtain
a(2n−1n+(2n+1)(pm−2n))≥f(pm)=pm−1mf(p)
Further we have
f(p)≤2mapn+2−pm2n≤2mapn+2
From 2n+1>pm>2n we can deduce mn+1>log2p>mn, so mn is a rational estimate of log2p. Therefore
f(p)≤2mapn+2≤2ap(log2p+m2)
When m is very large we obtain
f(p)≤2aplog2p
Similarly, using pm>2n>2n−1 we can obtain
pm−2nf(pm)−f(2n)≥2n−2n−1f(2n)−f(2n−1)
After rearranging we obtain
f(p)≥2apmn+1−pm2n≥2ap(log2p−m1)
When m is very large we then have
f(p)≥2aplog2p
Combining the two inequality estimates together we obtain
f(p)=2aplog2p
Combining 2log2pa into a single parameter, and combining with the case p=2k, we obtain
f(p)=aplogp
which holds for any positive integer, where a is a nonnegative real number. Therefore
xk=f(k)−f(k−1)=a(klog(k)−(k−1)log(k−1))