La geometria fornisce intuizioni e le disuguaglianze classiche consentono un facile accesso al rigore.
Soluzione geometrica
Sappiamo, dalla geometria dei minimi quadrati , che ˉ x = ( ˉ x , ˉ x , … , ˉ x ) è la proiezione ortogonale del vettore dei dati x = ( x 1 , x 2 , … , x n ) su il sottospazio lineare generato dal vettore costante ( 1 , 1 , … , 1 ) e che σ xx¯=(x¯,x¯,…,x¯)x=(x1,x2,…,xn)(1,1,…,1)σxè direttamente proporzionale alla (euclidea) distanza tra x e ˉ x . I vincoli di non negatività sono lineari e la distanza è una funzione convessa, da cui gli estremi della distanza devono essere raggiunti ai bordi del cono determinati dai vincoli. Questo cono è l'ortante positivo in R n e i suoi bordi sono gli assi delle coordinate, da cui segue immediatamente che tutti tranne uno di x i devono essere zero alle distanze massime. Per un tale insieme di dati, un calcolo diretto (semplice) mostra σ x / ˉ x = √xx¯.Rnxin .σx/x¯=n−−√.
Soluzione che sfrutta le disuguaglianze classiche
σx/ˉxσx/x¯ is optimized simultaneously with any monotonic transformation thereof. In light of this, let's maximize
x21+x22+…+x2n(x1+x2+…+xn)2=1n(n−1n(σxˉx)2+1)=f(σxˉx).
x21+x22+…+x2n(x1+x2+…+xn)2=1n(n−1n(σxx¯)2+1)=f(σxx¯).
(The formula for ff may look mysterious until you realize it just records the steps one would take in algebraically manipulating σx/ˉxσx/x¯ to get it into a simple looking form, which is the left hand side.)
An easy way begins with Holder's Inequality,
x21+x22+…+x2n≤(x1+x2+…+xn)max({xi}).
x21+x22+…+x2n≤(x1+x2+…+xn)max({xi}).
(This needs no special proof in this simple context: merely replace one factor of each term x2i=xi×xix2i=xi×xi by the maximum component max({xi})max({xi}): obviously the sum of squares will not decrease. Factoring out the common term max({xi})max({xi}) yields the right hand side of the inequality.)
Because the xixi are not all 00 (that would leave σx/ˉxσx/x¯ undefined), division by the square of their sum is valid and gives the equivalent inequality
x21+x22+…+x2n(x1+x2+…+xn)2≤max({xi})x1+x2+…+xn.
x21+x22+…+x2n(x1+x2+…+xn)2≤max({xi})x1+x2+…+xn.
Because the denominator cannot be less than the numerator (which itself is just one of the terms in the denominator), the right hand side is dominated by the value 11, which is achieved only when all but one of the xixi equal 00. Whence
σxˉx≤f−1(1)=√(1×(n−1))nn−1=√n.
σxx¯≤f−1(1)=(1×(n−1))nn−1−−−−−−−−−−−−−−−√=n−−√.
Alternative approach
Because the xixi are nonnegative and cannot sum to 00, the values p(i)=xi/(x1+x2+…+xn)p(i)=xi/(x1+x2+…+xn) determine a probability distribution FF on {1,2,…,n}{1,2,…,n}. Writing ss for the sum of the xixi, we recognize
x21+x22+…+x2n(x1+x2+…+xn)2=x21+x22+…+x2ns2=(x1s)(x1s)+(x2s)(x2s)+…+(xns)(xns)=p1p1+p2p2+…+pnpn=EF[p].
x21+x22+…+x2n(x1+x2+…+xn)2=x21+x22+…+x2ns2=(x1s)(x1s)+(x2s)(x2s)+…+(xns)(xns)=p1p1+p2p2+…+pnpn=EF[p].
The axiomatic fact that no probability can exceed 11 implies this expectation cannot exceed 11, either, but it's easy to make it equal to 11 by setting all but one of the pipi equal to 00 and therefore exactly one of the xixi is nonzero. Compute the coefficient of variation as in the last line of the geometric solution above.