This a proof from a book:

Theorem: let $f:\Bbb R^p\to\Bbb R$ with continuous partial derivatives at $a$. Then $f$ is differentiable at $a$.

Proof: without lose of generality suppose that $p=2$ and let $a:=(b,c)$ we have $$f(b+h,c+k)-f(b,c)=f(b+h,c+k)-f(b+h,c)+f(b+h,c)-f(b,c)\tag1$$ On one hand we have $$f(b+h,c)-f(b,c)=\partial_1 f(b,c)h+o(h)\tag2$$ On the other hand we can apply the mean value theorem to the function $t\mapsto f(b+h,t)$, hence $$f(b+h,c+k)-f(b+h,c)=\partial_2 f(b+h,c+\theta k)k,\quad\text{for some }\theta\in(0,1)\tag3$$ However, because of the continuity at $a$, we have that $$\partial_2 f(b+h,c+\theta k)=\partial_2 f(b,c)+o(h,k)\tag4$$ $\Box$

I cant see where $(4)$ comes and it seems wrong. If it would be true then we will had that

$$\lim_{(h,k)\to(0,0)}\frac{\partial_2 f(b+h,c+\theta k)-\partial_2 f(b,c)}{\|(h,k)\|}=0$$

what means that $\partial\partial_2 f(b,c)=0$, what doesnt make sense.

QUESTION: it is a typo in the proof or Im wrong in some place? In any case, can you enlighten the attempted proof?


What the author means by the notation $o(h,k)$ (which is not the same thing as $o(\|(h,k)\|$) is

$$\lim_{(h,k) \to (0,0)}\partial_2 f(b+h,c+\theta k)=\partial_2 f(b,c).$$

This is true since the partial derivatives are continuous at $a = (b,c)$ by hypothesis, and facilitates proving the desired result that $f$ is differentiable at $a$:

$$f(b+h,c+k) - f(b,c) - \underbrace{[\partial_1f(b,c)h+ \partial_2f(b,c)k]}_{\text{linear operator acting on }(h,k)} = o(\|(h,k\|)$$

Had this been written as

$$\partial_2 f(b+h,c+\theta k)=\partial_2 f(b,c) +o(\|(h,k)\|)$$

it would mean

$$\lim_{(h,k) \to (0,0)}\frac{\partial_2 f(b+h,c+\theta k)-\partial_2 f(b,c)- \mathcal{O\cdot(h,k)}}{\sqrt{h^2+k^2}} = 0,$$

where $\mathcal{O}$ is the null (or zero) linear operator and you would be correct saying that

$$\partial\partial_2f(b,c) = 0.$$

  • $\begingroup$ but the, supposed, meaning of $o(h,k)$ doesnt correspond with the definition of little o. The definition of little o is $$f(x)\in o(g(x))\iff |f(x)|\le\epsilon |g(x)|,\forall\epsilon>0$$ Then use $o(h,k)$ or $o(\|h,k\|)$ in this context is the same, in any case it means that $\partial\partial_2 f(b,c)=0$, what doesnt make sense. $\endgroup$ – Masacroso Feb 27 '18 at 19:05
  • 1
    $\begingroup$ Yes -- the author is abusing the little o notation. He should not use $o(h,k)$. I'm just explaining what he meant. The argument of little o needs to be clearly expressed in terms of a function. In proofs like this, I would prefer to see everything written more clearly in terms of limits or $\epsilon - \delta$ arguments. $\endgroup$ – RRL Feb 27 '18 at 19:09

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.