I'm having some trouble understanding Terence Tao's proof of the Hausdorff-Young theorem in his lectures notes 1. The theorem states
Proposition 3.1. Let $B$ denote the open unit ball in $\mathbb R^n$ and let $1\leq p,q\leq\infty$ such that $\|\hat f\|_{L^q(B)}\leq C\|f\|_{L^p(\mathbb R^n)}$ for all test functions $f\in\mathcal S(\mathbb R^n)$ with support in $B$. Then we have $p\in[1,2]$ and $q\leq p'$, where $p':=p/(p-1)$ denotes the conjugate index of $p$.
In the proof of the proposition prior to this one (where $B=\mathbb R^n$), we showed that the inequality $\|\hat f\|_{L^q(\mathbb R^n)}\leq C\|f\|_{L^p(\mathbb R^n)}$ gives us $$\lambda^n\lambda^{-n/q}\leq\widetilde C\lambda^{n/p}\qquad\text{for all }\lambda>0.$$ By rearranging this inequality we obtain $\lambda^{1-1/p-1/p}\leq\widetilde C^{1/n}$ for all $\lambda>0$, which is only possible, if $\widetilde C\geq1$ and $1-1/p-1/q=0$, where the latter is equivalent to $q=p'$.
Now, we want to utilize the same trick in Proposition 3.1. to show $q\leq p'$. Since we are restricted to the unit ball $B$, we can make $\lambda$ only so large before the support of $\psi(\cdot/\lambda)$ leaves $B$. Thus, we only have $$\lambda^n\lambda^{-n/q}\leq\widetilde C\lambda^{n/p}\qquad\text{for all }\lambda\in(0,b]$$ for some $b\geq1$ (we can choose $\psi$ such that $b>1$). Rearranging this inequality now gives us $\lambda^{n-q/n-p/n}\leq\widetilde C$ for all $\lambda\in(0,b]$. And if $\widetilde C\geq1$, it follows that $n-\frac nq-\frac np\geq0$. But using this inequality I always get $p'\leq q$ and not $q\leq p'$. Am I missing something or is this approach flawed? If so, do you know of another proof of this theorem (or the $q\leq p'$ part at least)?