Take the 2-minute tour ×
Mathematica Stack Exchange is a question and answer site for users of Mathematica. It's 100% free, no registration required.

I need ANY combination of 8 variables which satisfy:

a+b+c+d+e+f+g+h=0
a^2+b^2+c^2+d^2+e^2+f^2+g^2+h^2=1
a*b+b*c+c*d+d*e+e*f+f*g+g*h+a*h>2/3

Ideally these would be numbers having simple closed forms...

If I exchange third restriction with

a*b+b*c+c*d+d*e+e*f+f*g+g*h+a*h=2/3

I get a solution a=b=c=6^(-0.5); e=f=g=-6^(-0.5); d=h=0;

share|improve this question
1  
Ok, then please show us what you have tried. –  rm -rf Mar 20 '13 at 18:32
1  
@rm-rf I have tried entering those three restrictions into "Free-form input". Before that I figured an answer to a modified version analytically. Actually this question looks so straightforward to me that if I knew how to proceed the thread wouldn't even exist... –  Pranasas Mar 20 '13 at 18:44
    
$\left(\sqrt{1/8}, 1/2,\sqrt{1/8}, 0, -\sqrt{1/8}, -1/2, -\sqrt{1/8}, 0\right)$ works: the third expression evaluates to $\sqrt{1/2} \gt 2/3$. (This in fact is the maximum possible value it can attain subject to the restrictions.) –  whuber Mar 20 '13 at 19:51

1 Answer 1

up vote 8 down vote accepted

This is a special kind of problem which succumbs to a standard approach.

Analysis

The question can be solved by maximizing $x \mathbb{A} x'$ and checking that this maximum exceeds $2/3$. The restrictions are (1) that $x=(a,b,\ldots,h)$ must have unit square length and (2) be orthogonal to $e_1 = (1,1,\ldots, 1)$, where $A$ is the matrix

$$\left( \begin{array}{cccccccc} 0 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 1 & 0 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 0 & 1 \\ 1 & 0 & 0 & 0 & 0 & 0 & 0 & 0 \end{array} \right)$$

Solution

Because $\mathbb{A}$ is obviously a rotation matrix (it represents a cyclic permutation of the eight basis vectors), $e_1$ (by virtue of its constant coefficients) must be an eigenvector. Seven other complex eigenvectors can be found orthogonal to $e_1$. Their eigenvalues fall into three groups of complex conjugates, representing rotations by primitive multiples of $2\pi/8$, and one real eigenvalue of $-1$, representing a rotation by $\pi$. We seek the smallest rotation among these rotations, because that corresponds to the largest cosine of the angle, which geometrically is what $x\mathbb{A} x'$ is computing.

The eigensystem can be obtained and displayed as

MatrixForm /@ ({e, ev} = Eigensystem[a = RotateRight[#, 1] & /@ IdentityMatrix[8]] // Simplify)

The cosines are the real parts of the eigenvalues e:

Re[e]

$\left\{-\frac{1}{\sqrt{2}},-\frac{1}{\sqrt{2}},\frac{1}{\sqrt{2}},\frac{1}{\sqrt{2}},-1,0,0,1\right\}$

The two largest real parts, $1/\sqrt{2}\approx 0.7071 \gt 2/3$, occur in the third and fourth places. The real parts of either the third or fourth normalized eigenvectors therefore afford values of $x$ where the desired maximum is attained:

x = Normalize @ (Re /@ ev[[3]]) 

$\left\{\frac{1}{2 \sqrt{2}},0,-\frac{1}{2 \sqrt{2}},-\frac{1}{2},-\frac{1}{2 \sqrt{2}},0,\frac{1}{2 \sqrt{2}},\frac{1}{2}\right\}$

If any of this seems too abstract, let's check all the requirements:

  1. Confirm $x$ is orthogonal to $e_1$:

    x .ConstantArray[1, 8]
    

    $0$

  2. Confirm $x$ has unit squared length:

    x . x
    

    $1$

  3. Compute $x\mathbb{A} x'$:

    x . a . x
    

    $\frac{1}{\sqrt{2}}$

share|improve this answer
    
Thanks. And you are actually very good at mathematical aspect of this. –  Pranasas Mar 21 '13 at 17:51

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.