# Proof of a step of a lemma on the asymptotics of maximum likelihood where a Taylor expansion is used

I am trying to understand a proof of quite a long theorem that I report completely for the sake of completeness. This is From Jensen and Rahbek Asymptotic Inference for Nonstationary GARCH (2004). My question is actually basic and only regards the Taylor expansion done in the proof so one can skip to the end if he wishes.

At the start of the proof through the mean value theorem $\lambda^*$ is proven to exist. Then a Taylor expansion is made.

$$\Big| \frac{\partial f(\lambda^*)}{\partial \lambda} \Big| = \Big| \sum_{i,j,l = 1}^k v_{1,i}v_{2,j} (\varphi_l - \varphi_{0,l}) \partial^3 \ell_T (\varphi_0 - \lambda^* (\varphi - \varphi_0))/ \partial \varphi_i \varphi_j \varphi_l \Big|$$

I think he cuts it off after the second term, to be clear I write the Taylor expansion that I think he does in the univariate form:

$$f(\lambda^*)+\frac {f'(\lambda^*)}{1!} (\varphi-\lambda^*)+ \cdots.$$

My questions that I think are all connected are:

1. Where does the first term go in the expansion? (where is $f(\lambda^*)$ ?)
2. Why introduce $\lambda^*$ at all? (notice that condition A.3 of Lemma 1 does not involve $\lambda^*$)?
3. Why does equality hold in this Taylor expansion?
• The series is not cut off in this exposition: it uses a mean-value form of the remainder in Taylor's Theorem.
– whuber
Commented May 13, 2015 at 21:27
• @whuber Thanks! what would be the error function? Commented May 14, 2015 at 9:38
• @whuber Also is the Cauchy form of the remainder being used? I still can't seem to calculate it correctly. Commented May 14, 2015 at 12:23

And it is an expansion of f(1) around f(0), ie $f(1)=f(0)+f'(\lambda^*).(1-0)$, for some $\lambda^*\in (0,1)$.
• But where does $f(0) = v_2' [D^2 \ell_T( \varphi_0) ] v_2$ go? And if we are deriving with respect to $\lambda$ why is there the third derivative with respect to $\varphi$? Thanks by the way. Commented May 17, 2015 at 10:21
• I think he is just taking the derivative with respect to $\lambda$ but why is it written with respect to $\varphi$ ? Can we chat so you can explain it to me? Commented May 17, 2015 at 18:21