numhessian(f, {args}, minarg)

Numeric second derivative of f with respect
to argument "minarg".
* first argument: function name (string)
* second argument: all arguments of the function (cell array)
* third argument: (optional) the argument to differentiate w.r.t.
        (scalar, default=1)

If the argument
is a k-vector, the Hessian will be a kxk matrix

function a = f(x, y)
        a = x'*x + log(y);
endfunction

numhessian("f", {ones(2,1), 1})
ans =

    2.0000e+00   -7.4507e-09
   -7.4507e-09    2.0000e+00

Now, w.r.t. second argument:
numhessian("f", {ones(2,1), 1}, 2)
ans = -1.0000

Package: optim