Failure of Khintchine-type results along the polynomial image of IP$_0$ sets

Abstract: In "IP-sets and polynomial recurrence", Bergelson, Furstenberg, and McCutcheon established the following far reaching extension of Khintchine's recurrence theorem: For any invertible probability preserving system $(X,\mathcal A,\mu,T)$, any non-constant polynomial $p\in\mathbb Z[x]$ with $p(0)=0$, any $A\in\mathcal A$, and any $\epsilon>0$, the set $$R_\epsilon^p(A)={n\in\mathbb N\,|\,\mu(A\cap T^{-p(n)}A)>\mu^2(A)-\epsilon}$$ is IP$^*$, meaning that for any increasing sequence $(n_k){k\in\mathbb N}$ in $\mathbb N$, $$\text{FS}((n_k){k\in\mathbb N})\cap R_\epsilon^p(A)\neq \emptyset,$$ where $$\text{FS}((n_k){k\in\mathbb N})={\sum{j\in F}n_j\,|\,F\subseteq \mathbb N\,\text{ is finite}\text{ and }F\neq\emptyset}={n_{k_1}+\cdots+n_{k_t}\,|\,k_1<\cdots<k_t,\,t\in\mathbb N\}.$$ In view of the potential new applications to combinatorics, this result has led to the question of whether a further strengthening of Khintchine's recurrence theorem holds, namely whether the set $R_\epsilon^p(A)$ is IP$_0^*$ meaning that there exists a $t\in\mathbb N$ such that for any finite sequence $n_1<\cdots<n_t$ in $\mathbb N$, $$\{\sum_{j\in F}n_j\,|\,F\subseteq \{1,...,t\}\text{ and }F\neq \emptyset\}\cap R_\epsilon^p(A)\neq \emptyset.$$ In this paper we give a negative answer to this question by showing that for any given polynomial $p\in\mathbb Z[x]$ with deg$(p)\>1$ and $p(0)=0$ there is an invertible probability preserving system $(X,\mathcal A,\mu,T)$, a set $A\in\mathcal A$, and an $\epsilon>0$ for which the set $R_\epsilon^p(A)$ is not IP$_0^*$.