We say that the sequense $\{T_\alpha\}_{\alpha<\lambda}\subseteq[\omega]^\omega$ (where $\lambda$ is a ordinal) is a tower iff
- $\alpha<\beta<\lambda\rightarrow T_\beta\subseteq^*T_\alpha$.
- $\neg\exists K\in[\omega]^\omega\forall\alpha<\lambda(K\subseteq^*T_\alpha)$.
Here "$A\subseteq^*B$" means: $\exists C\in [A]^{<\omega}(A\setminus C\subseteq B)$.
With this, we define the number tower as the least length of a tower.
My doubt is, how can i to show that the number $\mathfrak{t}$ is well defined?
In order for $\mathfrak{t}$ to be well-defined, you only need to see that a tower exists. Consider, for sequences $(T_{\alpha} \mid \alpha < \lambda)$, $T_\alpha \subseteq [\omega]^{\omega}$, the property
$$ \alpha < \beta < \lambda \implies T_{\beta} \subseteq^* T_{\alpha} \wedge T_\beta \neq T_\alpha. \tag{$\dagger$} $$ Since $(\dagger)$ is preserved under unions and any such sequence has length $<(2^{\aleph_0})^{+}$, there is a maximal sequence $(T_{\alpha} \mid \alpha < \lambda)$ satisfying $(\dagger)$. It's easy to see that this sequence is a tower.
(If there were a counterexample $K$, we could define $$ T_{\lambda} := \begin{cases} K & \text{, if } \lambda \text{ is a limit ordinal} \\ K \setminus \min T_{\lambda -1 } & \text{, otherwise} \end{cases} $$ and then $(T_{\alpha} \mid \alpha < \lambda + 1)$ would satisfy $(\dagger)$.)