TITLE: Asymptotics for the tail of $L_p$ norms
QUESTION [5 upvotes]: Let $1<p<\infty$ be fixed and $f\in L^p(\mathbb{R}^d)$. I would like to obtain some asymptotic estimate for the function 
\begin{equation}\tag{1}
F(R)=\lVert f\chi_{\{\lvert x \rvert >R\}}\rVert_{L^p}=\left(\int_{\lvert x\rvert>R}\lvert f(x)\rvert^p\, dx\right)^{\frac{1}{p}}\qquad \text{ as }R\to +\infty,
\end{equation}
under suitable further assumptions on $f$. Here's a conjecture based on the heuristic that a lower summability index corresponds to a faster decay at infinity: 

Question. Assume that $f\in L^p\cap L^1(\mathbb{R}^d)$. Is it true that 
  $$\tag{2}F(R)\le C_f \cdot R^{-\frac{1}{p'}}?$$
  If (2) does not hold, can we give some other bound on $F$?

Here $C_f>0$ is a constant that depends on $f$ (probably through one or more of its norms).

REPLY [2 votes]: Claim. The conjecture is false. Indeed, proceeding as in Stephen Montgomery-Smith's answer, one may see that the problem is reduced to the asymptotic estimate of the tails 
$$S_N^{(p)}=\sum_{n=N}^\infty a_n^p, $$
where $a_n\ge 0$ is a summable sequence (i.e. $\sum_n a_n < \infty$). If the conjecture were true, then $S^{(p)}_N$ should decay as $N^{1-p}$ (or faster). However, we can find examples of sequences $a_n$ such that $S_N^{(p)}$ converges only logarithmically. 
Namely, if we take a $\alpha>1$ and we take
$$
a_n=\begin{cases} k^{-\alpha}, & n=2^k \\ 0, & \mathrm{otherwise}\end{cases}, $$
we obtain
$$
S_N^{(p)}=\sum_{k\ge \left\lfloor \frac{\log N}{\log 2}\right\rfloor}k^{-\alpha p} \ge c \left(\left\lfloor \frac{\log N}{\log 2}\right\rfloor\right)^{1-\alpha p},
$$ 
for some constant $c>0$. This is enough to disprove the conjecture. $\square$

EDIT (in response to Stephen Montgomery-Smith's comment below). Here's a more general construction, proving that no asymptotic estimate for $F(R)$ can exist. 
Let $\phi\colon (0, \infty)\to (0, \infty)$ be a monotonically increasing function such that $\phi(R)\to \infty$ as $R\to \infty$. We claim that a function $f\in L^1\cap L^p(\mathbb{R}^d)$ exists such that 
$$\sup_{R>0} \lVert f\chi_{\{\lvert x \rvert > R\}}\rVert_{L^p}\phi(R)=\infty.$$
To build $f$ we first take a nonvanishing bump function $g\ge 0$ supported in the unit ball. Then we set 
$$
\tilde{\phi}(R)=\phi(R)^{\frac{p}{2}\frac{1}{2p-1}},$$
observing that $\tilde{\phi}(R)$ increases monotonically towards $\infty$, and we set 
\begin{equation}
y_n=[\tilde{\phi}]^{-1}(n) \mathbf{e}_1.
\end{equation}
Here $\cdot^{-1}$ refers to functional inversion and $\mathbf{e}_1=(1, 0,\ldots, 0)\in \mathbb{R}^d$. Then we set $a_n=n^{-2}$ and we define $f$ to be 
\begin{equation}
f(x)=\sum_{n=1}^\infty a_n g(x-y_n).
\end{equation}
Since $\lVert f\rVert_{L^1} = C_1\sum_{n} a_n$ and $\lVert f\rVert_{L^p}^p=C_p \sum_n a_n^p$, the function $f$ is in $L^1\cap L^p(\mathbb{R}^d)$. We have
$$
\begin{split}
\lVert f\chi_{\{\lvert x \rvert > R\}}\rVert_{L^p}^p &\ge C \sum_{n> \tilde{\phi}(R)} n^{-2p} \\
&\ge C \tilde{\phi}(R)^{1-2p}= C\phi(R)^{-\frac{p}{2}}.
\end{split}$$
Therefore 
$$\sup_{R>0} \lVert f\chi_{\{\lvert x \rvert > R\}}\rVert_{L^p}^p\phi(R)^p=\sup_{R>0}\phi^{\frac{p}{2}}(R)=\infty.$$