4

Find the minimum value of $(x+y)(y+z)$ where $x,y,z$ are positive real numbers satisfying the condition $$xyz(x+y+z)=1$$

Hint?

Cameron Buie
  • 102,994

3 Answers3

7

Just apply AM-GM.

$(x+y)(y+z)=xy+y^2+xz+yz=y(x+y+z)+zx=\frac y{xyz}+zx=zx+\frac1{zx}\geq 2$

Equality holds if and only if $zx=1$.


Substituting $xz=1$, we have the condition that $y(x+y+\frac{1}{x}) = 1$. As such, $(x+y)(y+ \frac{1}{x}) = xy + \frac{y}{x} + y^2 + 1 = 1 + 1 = 2 $.

Calvin Lin
  • 68,864
1

if you make the transformation to new variables $u,v,w$ with $$ u = y+z \\ v = z + x \\ w = x + y $$ then $$ u + v + w = 2(x+y+z) = 2s $$ and $$ x = s - u\\ y = s - v\\ z = s - w $$

so that $s$ is the semi-perimeter of a triangle with sides $u,v,w$ and the area of the triangle:

$$ A = \sqrt{s(s-u)(s-v)(s-w)} = 1 $$ and the problem is to minimize the product of two sides of a triangle with area of triangle constrained to be $1$

suppose $u$ is not smaller than either of the other two sides. then the minimum product of two sides in this case must be $vw$.

if, in $\mathbb{R}^2$ we let the base of the triangle be the interval $[-\frac{u}2,\frac{u}2]$ then to satisfy the constraint on area the third vertex must lie either of the lines $Y=\pm \frac2{u}$

a fairly straightforward argument shows that the minimum value of the product $vw$ occurs when the triangle is isosceles, hence (Pythagoras)

$$ v=w=\sqrt ( \left(\frac{u}2\right)^2 + \left(\frac2{u}\right)^2) $$ since then minimum of the expression inside the surd is the value 2, occurring when u=2, we have, as the sides of the triangle in minimal configuration, $2, \sqrt{2}, \sqrt{2}$, and the minimum value of the product required is $\sqrt{2} \times \sqrt{2} = 2$

David Holden
  • 18,040
  • the geometric approach perhaps offers insight. however i nearly abandoned my effort - i easily get confused during calculations - and would like to thank Calvin for his rigorous algebraic treatment which sufficiently convinced me i was on a good track to stimulate me to go back and complete my solution – David Holden Dec 07 '13 at 06:46
0

Define the auxiliary function $F(x,y,z,\lambda) = (x+y)(y+z) + \lambda(1-xyz(x+y+z))$ and optimise it by taking the partial derivatives and settting them to zero.

But why can we do this?

This method applies Lagrange multipliers, since the level set, that is the constraint, is in the tangent space of the gradient. So we can set up this function, because we can make the claim that for $f(x,y,z) = (x+y)(y+z)$ and $g(x,y,z) = xyz(x+y+z)$, $\nabla f$ is parallel to $\nabla g$. Namely, $\exists \lambda$ such that $\nabla f + \lambda \nabla g = 0$.

  • @OliverBel: Is this what you're looking for? – Christopher K Dec 06 '13 at 21:40
  • 1
    we can't conclude that $x=z.$ – Leox Dec 06 '13 at 21:48
  • There is no guarantee that the minimum (or maximum) must occur at a point where $x=z$, even if the function is symmetric. – Calvin Lin Dec 06 '13 at 21:50
  • @CalvinLin: My reasoning was that then $x+y=x+z$, but in retrospect that probably evaluates to the maximum. Although my answer was intended to be a suggestion, it obviously does not bear fruit. I'm slightly worried that I didn't think of AM-GM, especially when I have Putnam tomorrow... – Christopher K Dec 06 '13 at 22:06
  • @ChrisK In such a case, I would perform Lagrange first to check the equality condition, before making such a claim. Knowing that makes performing (forcing) AM-GM easier too. All too often, people claim that the equality case must be when $a=b$ due to symmetry, but there are several counterexamples to that. Good luck with the Putnam! – Calvin Lin Dec 06 '13 at 22:13