Re: Help with a problem: non-convexifiability of a function




In article <1141715475.115592.190660@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>,
tsy@xxxxxxxxxxxx writes:
I need to prove that there is no increasing function f such that the
function f[sqrt(x+y^2)+y] (x,y>=0) is convex.

sounds much like homework.
hint: consider x>0, y>0 .
compute the Hessian matrix of g(x,y)=f(sqrt(x+y^2)+y)
clearly this depends also on the derivatives of f and
g is convex there if and only if the Hessian is positive semidefinite
f is increasing means f'(z)>=0 .
a two by two matrix H is positive semidefinite if and only if
1) it is the zero matrix
or
2) H(1,1)>0 or H(2,2)>0 and H(1,1)*H(2,2)-H(1,2)^2>=0
hth
peter

.



Relevant Pages

  • Re: How to calculate Hessian Matrix
    ... It is very easy to compute an Hessian matrix using numerical ... double precision function f ... subroutine gradient ... also prefer central derivatives rather than "right" derivatives and ...
    (comp.lang.fortran)
  • Re: How to calculate Hessian Matrix
    ... It is very easy to compute an Hessian matrix using numerical ... subroutine gradient ... also prefer central derivatives rather than "right" derivatives and ...
    (comp.lang.fortran)
  • Re: How to calculate Hessian Matrix
    ... evalued but has no explicit expression. ... Hessian matrix at some point to see the shape. ... For each mathematical operation on the way to computing ... for each intermediate value also compute the derivatives ...
    (comp.lang.fortran)
  • Re: How to calculate Hessian Matrix
    ... Hessian matrix at some point to see the shape. ... How do you evaluate it without an explicit expression? ... For each mathematical operation on the way to computing ... for each intermediate value also compute the derivatives ...
    (comp.lang.fortran)
  • Re: How to calculate Hessian Matrix
    ... you write the function computing the Hessian in the same ... also prefer central derivatives rather than "right" derivatives and ... as the Hessian matrix is symmetrical (d/dxi(df/ ... dxj)=d/dxj) the previous algorithm is not optimal. ...
    (comp.lang.fortran)