2
$\begingroup$

I should analyze whether the function

$$f(x,y) = x^2y + x y^2 \text{ where } x,y > 0$$

is (quasi-) concave or convex.

Thus, as usual, I set up the Hessian as

$$ D^2f(x,y) = \left( \begin{array}{cc} 2y & 2x + 2y \\ 2x + 2y & 2x \\ \end{array} \right) $$

which, given the constraints $x,y > 0$, is indefinite.

So I need the bordered Hessian:

$$ H= \left( \begin{array}{ccc} 0 & 2xy + y^2 & x^2 + 2xy \\ 2xy + y^2 & 2y & 2x + 2y \\ x^2 + 2xy & 2x+2y & 2x \end{array} \right) $$

and check its determinant to find out about its properties. I end up with

$$ \det|H| = 2x(2x^3 + x^3 y - 2x^2 y + 4x^2 y^2 + 4 x y^3 + 2 y^4)$$

which, considering $x,y > 0$, looks quite positive to me, so $f$ would be quasiconcave. But how can I prove that $\det|H| > 0$? Or is there any easier approach?

Thanks!

$\endgroup$
1
$\begingroup$

The Hessian $D_2$ of your function $f$ is as you said, then: $$ D_2 - \lambda I = \begin{pmatrix} 2y - \lambda & 2x+2y \\ 2x+2y & 2x-\lambda \end{pmatrix}. $$ The determinant is: $$ \det(D_2-\lambda I) = \lambda^2 - 2(x+y)\lambda - 4(x^2+y^2+xy) , $$ hence we have one positive and one negative eigenvalues, surely it is not positive semi-definite.

The bordered Hessian $H$ has determinant (I don't know what your expression in the question is, but it sure doesn't look right): $$ \det (H) = 6 x^4 y + 12 x^3 y^2 + 12 x^2 y^3 + 6 x y^4 >0 \quad \forall x,y>0. $$

Another way to do this is considering the level set

$$F_a := \{(x,y): f(x,y)\ge t\}$$ is convex or not:

concave

as you can tell by the level curve the level set is indeed convex, hence $f$ is quasi-concave. To see it mathematically: set $$ x^2y + xy^2 -t =0 \implies y = -\frac{x}{2} + \sqrt{\frac{x^2}{4} + \frac{t}{x}}, $$ and $$ y'' = \frac{6 t (t+x^3)}{x^{5/2} (4 t+x^3)^{3/2}} >0 \quad \forall t>0, $$ therefore $F_a$ is convex and the quasi-concavity of $f$ follows.

$\endgroup$
  • $\begingroup$ Hessian a symetric matrix, so its eigenvalues are real. Your characteristic polynomial is wrong, by the way. $\endgroup$ – TZakrevskiy Jun 9 '13 at 19:49
  • $\begingroup$ @TZakrevskiy Thanks, fixed. $\endgroup$ – Shuhao Cao Jun 9 '13 at 19:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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