Pull-back of a differential : I get confused with the variables I edited again my message with the remarks done in the comments.
I have a 2-form :
$$\alpha=\alpha_{\mu \nu} dx^\mu \wedge dx^\nu$$
I want to compute the pull back $F^{*}(d \alpha)$ to show that : $F^{*}(d \alpha)=dF^{*}( \alpha)$
But I make a mistake somewhere because I can't prove the equality.
$$ F : y \mapsto x $$ 
So when I write $x^\mu$ I have in fact a dependance $x^\mu(y^\nu)$.
$$d \alpha=d \alpha_{\mu \nu} \wedge dx^\mu \wedge dx^\nu 
\\= \frac{\partial \alpha_{\mu \nu}}{\partial x^\epsilon} dx^\epsilon \wedge dx^\mu \wedge dx^\nu$$
$$F^{*}(d \alpha)=F^{*}(\frac{\partial \alpha_{\mu \nu}}{\partial x^\epsilon}) F^{*}(dx^\epsilon) \wedge F^{*}(dx^\mu)  \wedge F^{*}(dx^\nu)$$
I have :
$$ F^{*}(dx^\mu) = \frac{\partial x^\mu}{\partial y^i} dy^i$$
and
$$F^{*}(\frac{\partial \alpha_{\mu \nu}(x)}{\partial x^\epsilon})=\frac{\partial \alpha_{\mu \nu}(x(y))}{\partial x^\epsilon(y)}$$
And finally, I get :
$$F^{*}(d \alpha)=\frac{\partial \alpha_{\mu \nu}(x(y))}{\partial x^\epsilon(y)} \frac{\partial x^\epsilon}{\partial y^i}\frac{\partial x^\mu}{\partial y^j}\frac{\partial x^\nu}{\partial y^k} dy^i \wedge dy^j \wedge dy^k$$
On the other hand, I have :
$$F^{*}(\alpha)=F^{*}(\alpha_{\mu \nu}(x)) F^{*}(dx^\mu)  \wedge F^{*}(dx^\nu)\\
=\alpha_{\mu \nu}(x(y)) \frac{\partial x^\mu}{\partial y^i} \frac{\partial x^\nu}{\partial y^j} dy^i \wedge dy^j$$
But here there is a problem when I differentiate :
$$dF^{*}(\alpha)=d(\alpha_{\mu \nu}(x(y)) \frac{\partial x^\mu}{\partial y^i} \frac{\partial x^\nu}{\partial y^j}) \wedge dy^i \wedge dy^j$$
Indeed I will have extra derivative term in $\frac{\partial^2 x^\mu}{\partial y^i \partial y^l} $ when I differentiate. And I don't have these terms in $F^{*}(d \alpha)$.
So where is my mistake ??

[edit] According to the answer below, I see that my misunderstanding is in the fact that :
$$dF^{*}(\alpha)=d(\alpha_{\mu \nu}(x(y))) \wedge (\frac{\partial x^\mu}{\partial y^i}) dy^i \wedge ( \frac{\partial x^\nu}{\partial y^j} ) dy^j$$
We don't differentiate the terms $\frac{\partial x^\nu}{\partial y^j}$. But I don't understand why as the definition of the exterior derivative is the following :
With :
$$\alpha=\alpha_\mu dx^{\mu}$$
We have by definition :
$$ d\alpha=d\alpha_\mu \wedge dx^{\mu}$$
Thus in my example it should be :
$$dF^{*}(\alpha)=d(\alpha_{\mu \nu}(x(y)) \frac{\partial x^\mu}{\partial y^i}  \frac{\partial x^\nu}{\partial y^j} ) \wedge  dy^i \wedge  dy^j$$
And not :
$$dF^{*}(\alpha)=d(\alpha_{\mu \nu}(x(y)))  \frac{\partial x^\mu}{\partial y^i}  \frac{\partial x^\nu}{\partial y^j}  \wedge  dy^i \wedge  dy^j$$
ie the differential applies to all the terms including the chain derivative and not only on $\alpha_\mu$. Could someone clarify this for me (or at least give me an exact definition of the exterior derivative ?)
 A: From $d \alpha=d \alpha_{\mu \nu} \wedge dx^\mu \wedge dx^\nu$, we have $$F^*(d\alpha) = d(\alpha_{\mu \nu} \circ F) \wedge d(x^\mu \circ F) \wedge d(x^\nu \circ F).$$ On the other hand, $$dF^*(\alpha) = d[(\alpha_{\mu \nu} \circ F) \wedge d(x^\mu \circ F) \wedge d(x^\nu \circ F)] = d(\alpha_{\mu \nu} \circ F) \wedge d(x^\mu \circ F) \wedge d(x^\nu \circ F). $$
As you can see, writing down the partial derivatives explicitly is of no particular use.
A: $$\newcommand{\ai}{\alpha}
\newcommand{\be}{\beta}
\newcommand{\Ga}{\Gamma}
\newcommand{\ga}{\gamma} 
\newcommand{\de}{\delta}
\newcommand{\De}{\Delta}
\newcommand{\e}{\epsilon}
\newcommand{\lam}{\lambda}
\newcommand{\La}{\Lamda}
\newcommand{\om}{\omega}
\newcommand{\Om}{\Omega}
\newcommand{\si}{\sigma}
\newcommand{\vp}{\varphi}
\newcommand{\rh}{\rho}
\newcommand{\ta}{\theta}
\newcommand{\Ta}{\Theta}
\newcommand{\W}{\mathcal{O}}
\newcommand{\ps}{\psi}
\newcommand{\mf}[1]{\mathfrak{#1}}
\newcommand{\ms}[1]{\mathscr{#1}}
\newcommand{\mb}[1]{\mathbb{#1}}
\newcommand{\cd}{\cdots}
\newcommand{\s}{\subset}
\newcommand{\es}{\varnothing}
\newcommand{\cp}{^\complement}
\newcommand{\bu}{\bigcup}
\newcommand{\ba}{\bigcap}
\newcommand{\co}{^\circ}
\newcommand{\ito}{\uparrow}
\newcommand{\dto}{\downarrow}  
\newcommand{\ti}[1]{\tilde{#1}}
\newcommand{\la}{\langle}
\newcommand{\ra}{\rangle}
\newcommand{\ov}[1]{\overline{#1}}
\newcommand{\no}[1]{\left\lVert#1\right\rVert}
\newcommand{\du}{^\ast}
\newcommand{\pf}{_\ast}
\newcommand{\is}{\cong}
\newcommand{\n}{\lhd}
\newcommand{\m}{^{-1}}
\newcommand{\ts}{\otimes}
\newcommand{\ip}{\cdot}
\newcommand{\op}{\oplus}
\newcommand{\xr}{\xrightarrow}
\newcommand{\xla}{\xleftarrow}
\newcommand{\xhl}{\xhookleftarrow}
\newcommand{\xhr}{\xhookrightarrow}
\newcommand{\mi}{\mathfrak{m}}
\newcommand{\w}{\wedge}
\newcommand{\X}{\mathfrak{X}}
\newcommand{\pd}{\partial}
\newcommand{\dx}{\dot{x}}
\newcommand{\dr}{\dot{r}}
\newcommand{\dy}{\dot{y}}
\newcommand{\dth}{\dot{theta}}
\newcommand{\pa}[2]{\frac{\pd #1}{\pd #2}}
\newcommand{\na}{\nabla}
\newcommand{\dt}[1]{\frac{d#1}{d t}\bigg|_{ t=0}}
\newcommand{\ld}{\mathcal{L}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\bh}{\mathbb{H}}
\newcommand{\lix}{\lim_{x\to\infty}}
\newcommand{\li}{\lim_{n\to\infty}}
\newcommand{\infti}{\sum_{i=1}^{\infty}}
\newcommand{\inftj}{\sum_{j=1}^{\infty}}
\newcommand{\inftn}{\sum_{n=1}^{\infty}} 
\newcommand{\snz}{\sum_{n=-\infty}^{\infty}} 
\newcommand{\ie}{\int_E}
\newcommand{\ir}{\int_R}
\newcommand{\ii}{\int_0^1}
\newcommand{\sni}{\sum_{n=0}^\infty}
\newcommand{\ig}{\int_{\ga}}
\newcommand{\io}{\text{ i.o.}}
\newcommand{\aut}{\text{Aut}}
\newcommand{\out}{\text{Out}}
\newcommand{\inn}{\text{Inn}}
\newcommand{\F}{\mathcal{F}}
\newcommand{\V}{\mathbf{V}} 
\newcommand{\II}{\mathbf{I}}
\newcommand{\wh}{\Rightarrow}
\newcommand{\eq}{\Leftrightarrow}
\newcommand{\eqz}{\setcounter{equation}{0}}
\newcommand{\se}{\subsection*}
\newcommand{\ho}{\text{Hom}}
\newcommand{\ds}{\displaystyle}
\newcommand{\tr}{\text{tr}}
\newcommand{\id}{\text{id}}
\newcommand{\im}{\text{im}}
\newcommand{\ev}{\text{ev}}$$
Your derivation is mostly correct. If $F$ is not a change of coordinate, we usually write the Jacobian as $[\frac{\partial F^i}{\partial y^j}]$ instead of $[\pa{x^i}{y^j}]$. To get the desired result, we just need to expand the terms and use Poincare lemma as follows,
\begin{align*}
&d(\ai_{ab}\circ F\pa{F^a}{y^i}\pa{F^b}{y^j})dy^i\w dy^j\\
=&\pd_{y^h}(\ai_{ab}\circ F)dy^h\w\pa{F^a}{y^i}dy^i\w\pa{F^b}{y^j}dy^j+\ai_{ab}\circ F\frac{\pd F^a}{\pd y^h\pd x^i}\pa{F^b}{y^j}dy^h\w dy^i\w dy^j\\&+\ai_{ab}\circ F\frac{\pd F^b}{\pd y^h\pd y^j}\pa{F^a}{y^i}dy^h\w dy^i\w dy^j\\
=&(\ai_{ab})\pf\circ F\pf (\pd_{y^h})dy^h\w\pa{F^a}{y^i}dy^i\w\pa{F^b}{y^j}dy^j+\ai_{ab}\circ F\pa{F^b}{y^j}(d^2F^a)\w dy^j\\&-\ai_{ab}\circ F\pa{F^a}{y^i}(d^2F^b)\w dy^i\\
=&(\ai_{ab})\pf(\pd_{x^m})\pa{F^m}{y^h}dy^h\w\pa{F^a}{y^i}dy^i\w\pa{F^b}{y^j}dy^j\\
=&F\du(d\ai).
\end{align*}
