ArXiv:1301.7652 and Euler homogeneous function theorem

[strangerep]

Let ##F : R^n \to R## be a degree-1 positive-homogeneous function. I.e., ##F(\lambda y) = \lambda F(y),## for all real ##\lambda>0## and any nonzero ##y\in R^n##.

In this paper, near the middle of p2 at eq(4), the authors introduce
$$\ell_a ~=~ \frac{\partial F}{\partial y^a} ~,$$and then they claim that because ##F## is 1-homogeneous, it follows that ##\ell_a## is necessarily of the form:
##\ell_a ~=~ y^a/F ~.##

ISTM that one can only claim that ##\ell_a## is 0-homogeneous (by the Euler homogeneous function theorem), and there are other 0-homogeneous functions besides their ##y^a/F## .

Am I missing something, or are the authors wrong?

Edit: Maybe they meant ##F/y^a## ? But that still seems wrong if ##n > 1##.

 

[micromass]

The authors are most definitely wrong, even if it's a typo and they actually mean your edit. Almost any homogeneous function would give a solid counterexample. Not to say that the indices are wrong.

What I think they did is saying that ##y^a \ell_a = F##, which is indeed the Euler homogeneous function theorem. Then they somehow forget that this is a sum and divide to obtain ##\ell_a = F/y^a## (which I think they meant). From a rigorous point of view, this is nonsense. But perhaps the rest of the paper goes through by only using the relation ##y^a \ell_a = F##?

 

[strangerep]

Oh, what a relief! Micromass returns unexpectedly from the wilderness! After 50+ views and no replies, I was becoming depressed.

The authors are most definitely wrong, even if it's a typo and they actually mean your edit. Almost any homogeneous function would give a solid counterexample. Not to say that the indices are wrong.

Yes.

What I think they did is saying that ##y^a \ell_a = F##, which is indeed the Euler homogeneous function theorem. Then they somehow forget that this is a sum and divide to obtain ##\ell_a = F/y^a## (which I think they meant). From a rigorous point of view, this is nonsense.

OK, good -- in the sense that this is indeed what I thought.

I suppose I should email the authors, though I suspect they won't appreciate it.

But perhaps the rest of the paper goes through by only using the relation ##y^a \ell_a = F##?

I'll have to check that.

Thank you indeed.

 

[micromass]

It's a bit interesting though that the indices in ##\ell_a = F/y^a## are correct. Clearly the equation is nonsense, but perhaps we can give a rigorous meaning to it nevertheless. I need to think of this.

 

[strangerep]

It's a bit interesting though that the indices in ##\ell_a = F/y^a## are correct. Clearly the equation is nonsense, but perhaps we can give a rigorous meaning to it nevertheless. I need to think of this.

Consider the example function ##F(x,y) := x_\alpha y^\alpha##, where ##F## is 1-homogeneous in the vector ##y##. We have
$$\ell_\lambda ~:=~ \frac{\partial F}{\partial y^\lambda} ~=~ x_\lambda ~,$$and indeed
$$\ell_\mu y^\mu ~=~ F(x,y) ~.$$However,
$$\ell_\lambda ~\ne~ \frac{F}{y^\lambda} ~=~ \frac{x_\alpha y^\alpha}{y^\lambda} ~.$$

 

[micromass]

Yes, I know that it's nonsense. But I was thinking of changing the meaning of /, to make things work out nice. But this is obviously not what the paper does. Your counterexample is indeed a good one to the relation in the paper.

 

[strangerep]

Argh! I think I see what they probably intended...

From their definition of ##h_{ab}## in eq(4), it follows that ##y^b h_{ab} = 0##. Then, contracting both sides of their eq(5) with ##y^b##, we get
$$0 ~=~ g_{ab} y^b - \ell_a \ell_b y^b ~=~ y_a - \ell_a F ~.$$ Hence
$$y_a ~=~ \ell_a F$$and so, (for ##F\ne 0##),
$$\ell_a ~=~ y_a/F ~.$$
But their sequence of statements is totally messed up (sigh).