1: \begin{abstract}

2: We consider the spread-out models of the self-avoiding walk and its finite-memory version, called the memory-$\tau$ walk.  For both models, each step is uniformly distributed over the d-dimensional box $\{x\in\Zd:\|x\|_{\infty} \le L\}$. The critical point $p_c^\tau$ for the memory-$\tau$ walk is increasing in $\tau$ and converges to the critical point $p_c^\infty$ for the self-avoiding walk as $\tau\uparrow\infty$. The best estimate of the rate of convergence so far was obtained by Madras and Slade \cite[Lemma~6.8.6]{ms93}: $p_c^{\infty}-p_c^{\tau} \le K\tau^{-(1+\delta)}$, where $\delta < (d-4)/2 \wedge 1$ and $K<\infty$ may depend on $d,~\delta,~L$ but not on $\tau$. By using the lace expansion, we show that there is a constant $C_d$ for $d>4$ and $L$ sufficiently large such that as $\tau\uparrow \infty$

3: \begin{align}

4: p_c^{\infty}-p_c^{\tau} =C_d\tau^{-\frac{d-2}{2}}+O(\tau^{-\frac{d-2}{2}}(\log \tau)^{-1}).\nn

5: \end{align}

6: Moreover, we show that as $L \uparrow \infty$, 

7: \begin{align}

8: C_d=\frac{2}{d-2}\left(\frac{d}{2\pi\Sigma_U^2}\right)^{\frac{d}{2}}L^{-d}+O(L^{-d-1})\nn

9: \end{align} 

10: where $\Sigma_U^2=\int_{\R^d}{|x|^2}\ind{x\in\R^d:\|x\|_{\infty}\le1}d^dx$ is the variance of the uniform distribution over the $d$-dimensional box of side-length $2$.

11: \end{abstract}