[guided]The goal is to replace the abstract centre $Z(K)$ by an explicit subgroup of the torus $T$. For each $\alpha\in R$, the root space $\mathfrak{k}_{\mathbb C,\alpha}$ means $\{X\in\mathfrak{k}_{\mathbb C}:\operatorname{Ad}_s(X)=\alpha(s)X\text{ for every }s\in T\}$. Define $Z_T:=\{t\in T:\alpha(t)=1\text{ for every }\alpha\in R\}$. We prove that $Z(K)=Z_T$.
First take $z\in Z(K)$. The [maximal torus theorem](/theorems/9713) [citetheorem:9713] says that every element of the compact connected Lie group $K$ is conjugate to an element of the fixed maximal torus $T$. Thus there are $h\in K$ and $t\in T$ such that
\begin{align*}
hzh^{-1}=t.
\end{align*}
Because $z$ is central, $hzh^{-1}=z$, and hence $z=t\in T$.
Now we check that every root is trivial on $z$. For each root $\alpha\in R$, the corresponding root space $\mathfrak{k}_{\mathbb C,\alpha}$ is an eigenspace for the adjoint action of $T$: for every $v\in \mathfrak{k}_{\mathbb C,\alpha}$,
\begin{align*}
\operatorname{Ad}_z(v)=\alpha(z)v.
\end{align*}
Since $z$ is central, conjugation by $z$ is the identity map on $K$, and therefore its differential is the identity linear map on $\mathfrak{k}_{\mathbb C}$. Applying this identity operator to any non-zero $v\in\mathfrak{k}_{\mathbb C,\alpha}$ gives
\begin{align*}
v=\operatorname{Ad}_z(v)=\alpha(z)v.
\end{align*}
Thus $\alpha(z)=1$. Since this holds for every root $\alpha\in R$, we have $z\in Z_T$.
Conversely, suppose $t\in Z_T$. We prove that $t$ is central. The torus $T$ is abelian, so $\operatorname{Ad}_t$ is the identity on its Lie algebra $\mathfrak{t}$, and therefore also on $\mathfrak{t}_{\mathbb C}$. On a root space $\mathfrak{k}_{\mathbb C,\alpha}$, the operator $\operatorname{Ad}_t$ acts by multiplication by $\alpha(t)$, and $\alpha(t)=1$ by the definition of $Z_T$. Hence $\operatorname{Ad}_t$ is the identity on every summand in the root decomposition
\begin{align*}
\mathfrak{k}_{\mathbb C}=\mathfrak{t}_{\mathbb C}\oplus \bigoplus_{\alpha\in R}\mathfrak{k}_{\mathbb C,\alpha}.
\end{align*}
Therefore $\operatorname{Ad}_t$ is the identity on $\mathfrak{k}_{\mathbb C}$ and hence on the real Lie algebra $\mathfrak{k}$.
It remains to translate this infinitesimal statement back to the group. The [kernel of the adjoint representation](/theorems/3752) of a connected Lie group is its centre. Indeed, if $\operatorname{Ad}_t=\operatorname{id}_{\mathfrak{k}}$, then the conjugation automorphism
\begin{align*}
c_t:K&\to K
\end{align*}
\begin{align*}
x&\mapsto txt^{-1}
\end{align*}
satisfies
\begin{align*}
c_t(\exp X)=\exp(\operatorname{Ad}_t X)=\exp X
\end{align*}
for each $X\in\mathfrak{k}$. Let $H\le K$ be the subgroup generated by $\exp(\mathfrak{k})$. Since the exponential map sends a neighbourhood of $0\in\mathfrak{k}$ onto a neighbourhood of the identity element $e\in K$, the subgroup $H$ contains an open neighbourhood of $e$ and is therefore open. Its complement is a union of open cosets of $H$, so $H$ is also closed. Because $K$ is connected and $H$ is non-empty, open, and closed, we have $H=K$. Thus $c_t$ is the identity map on all of $K$. Hence $txt^{-1}=x$ for every $x\in K$, which means $t\in Z(K)$.[/guided]