1: \begin{abstract}

2: The nonlinear effects of environmental variability on species abundance plays an important role in the maintenance of ecological diversity. Nonetheless, many common models use parametric nonlinear terms pre-determining ecological conclusions.  Motivated by this concern, we study the estimate of the second derivative (curvature) of the link function $g$ in a functional single index model: $Y = g\left(\int X\left(t\right)\beta^0\left(t\right)\mathrm{d}t\right) + \epsilon$. Since the coefficient function $\beta^0$ and the link function $g$ are both unknown, the estimate is expressed as a nested optimization. For a fixed and unknown $\beta$, the link function $g$ and $g''$ are estimated by local quadratic approximation, then the coefficient function $\beta^0$ is estimated by minimizing the MSE of the model. In this paper, we derive the rate of convergence of the estimation is $\frac{1}{n}\sum\limits_{i=1}^n\mathrm{E}\left[\hat{g}''\left(\int X_i\hat{\beta}\right) - g''\left(\int X_i\beta^0\right)\right]^2 = O\left(h_n^4 + \frac{1}{nh_n^4}\right)$, where $h_n$ is the bandwidth in the local quadratic approximation. In addition, we prove that the argument of $g$, $\int X\beta^0$, can be estimated root-$n$ consistently. However, practical implementation of the method requires solving a nonlinear optimization problem, and our results show that the estimates of the link function and the coefficient function are quite sensitive to the choices of starting values.

3: \end{abstract}

4: