We will derive properties that hold regardless of the parity of p and divide the cases later when needed. For each i > k, let
\bar{\beta}_i:= \sum_{j=k+1}^{i-1} b_{ij} y_j \in H^2(B_n(B))
denote the class obtained from \beta_i by truncating the terms in F_{k}(B). Since \phi is k-stable, the image of \alpha_k under \phi is contained in F_{k}(B). Using Lemma 3.3 and the assumption that \ell is the smallest integer satisfying the condition \phi(x_{k+1}) \in F_{\ell}(B), we may write
\phi(x_{k+1}) = \epsilon (2y_{\ell} - \bar{\beta}_{\ell}) + w
for some nonzero \epsilon \in \mathbb{Q} and w \in F_{k}(B). If p=0, the proof is complete by applying Lemma 2.3 for j=\ell-1.
We assume p \neq 0 from now on. If follows from (?) and (Proof) that
\begin{aligned}
0 =& \phi \left(x_{k+1}(x_{k+1} - \alpha_{k+1})\right) = (2\epsilon y_{\ell} - \epsilon \bar{\beta}_{\ell} + w)(2\epsilon y_{\ell} - \epsilon \bar{\beta}_{\ell} + w - \phi(\alpha_{k+1}))\\
=& 2\epsilon y_{\ell}(2\epsilon y_{\ell} - 2\epsilon \bar{\beta}_{\ell} + 2w - \phi(\alpha_{k+1}))\\
&+ \epsilon \bar{\beta}_{\ell}(\epsilon \bar{\beta}_{\ell} -2w + \phi(\alpha_{k+1}))\\
&+ w(w-\phi(\alpha_{k+1})).
\end{aligned}
We can remove all the square terms using the relation y_i^2 = \beta_i y_i. From the expression (Proof) and the condition that w, \phi(\alpha_{k+1}) \in F_k(B), all the three terms in the last expression are zero by Remark 2.1. By setting
u:= \frac{-2w + \phi(\alpha_{k+1})}{\epsilon},
the first two terms lead to the relations
\beta_{\ell} = \bar{\beta}_{\ell} + \frac{-2w + \phi(\alpha_{k+1})}{2\epsilon} = \bar{\beta}_{\ell} + \frac{u}{2}
and
\bar{\beta}_{\ell} \left(\bar{\beta}_{\ell} + u\right) = \bar{\beta}_{\ell} \left(\bar{\beta}_{\ell} + \frac{-2w + \phi(\alpha_{k+1})}{\epsilon}\right) = 0.
Since x_{k+1}^2 is contained in the ideal generated by F_{k}(A) in H^\ast(B_n(A)), using the equation (Proof) and the assumption that \phi is k-stable, we see that (2y_{\ell} - \bar{\beta}_{\ell})^2 is contained in the ideal generated by F_{k}(B). By the classification of the square zero elements (Lemma 3.1), we see that
\bar{\beta}_{\ell} = p\left(y_{\ell-1} - \frac{\bar{\beta}_{\ell-1}}{2}\right).
We divide the cases by the parity of p.
Case 1: p is even.
By (Proof) and (Proof), we have
\bar{\beta}_{\ell}(2\beta_{\ell} - \bar{\beta}_{\ell}) = 0.
Substituting (Proof) into (Proof), it follows that
\begin{aligned}
0 &= \left(y_{\ell-1} - \frac{\bar{\beta}_{\ell-1}}{2}\right)\left(2\beta_{\ell} - py_{\ell-1} + \frac{p}{2}\bar{\beta}_{\ell-1} \right)\\
&= y_{\ell-1} (2\beta_{\ell} - py_{\ell-1} ) - \frac{\bar{\beta}_{\ell-1}}{2}\left( 2\beta_{\ell} - 2py_{\ell-1} + \frac{p}{2}\bar{\beta}_{\ell-1} \right)\\
&= 2y_{\ell-1} \left(\beta_{\ell} - \frac{p}{2}y_{\ell-1} \right) - \frac{\bar{\beta}_{\ell-1}}{2}\left( u - \frac{p}{2}\bar{\beta}_{\ell-1} \right),
\end{aligned}
where the last equality follows from (Proof) and (Proof). Since \bar{\beta}_{\ell-1} has no y_{\ell-1} term and u \in F_{k}(B), both terms in the last expression vanish. Hence,
\frac{p}{2}y_{\ell-1} \left(\beta_{\ell} - \frac{p}{2}y_{\ell-1} \right) = 0.
By applying Lemma 2.4 for j=\ell and v = py_{\ell-1}/2, we can find a new matrix B' = (b'_{ij}) with b'_{\ell, \ell-1} = 0. This reduces the proof to the case p=0.
Case 2: p is odd.
Substituting (Proof) into (Proof), we get
\begin{aligned}
0 &= \left(y_{\ell-1} - \frac{\bar{\beta}_{\ell-1}}{2}\right)\left(y_{\ell-1} - \frac{\bar{\beta}_{\ell-1}}{2} + \frac{u}{p}\right)\\
&= y_{\ell-1}\left(y_{\ell-1} - \bar{\beta}_{\ell-1} + \frac{u}{p}\right) + \frac{\bar{\beta}_{\ell-1}}{2}\left(\frac{\bar{\beta}_{\ell-1}}{2} - \frac{u}{p} \right).
\end{aligned}
By the similar reasoning as before, it follows that
\beta_{\ell-1} = \bar{\beta}_{\ell-1} - \frac{u}{p}
and
\frac{\bar{\beta}_{\ell-1}}{2}\left(\frac{\bar{\beta}_{\ell-1}}{2} - \frac{u}{p} \right) = 0.
Since p is odd, the class \bar{\beta}_{\ell-1} is divisible by 2 by (Proof). By taking v:= \bar{\beta}_{\ell-1}/2, we obtain v(\beta_{\ell-1} - v) = 0 from (Proof) and (Proof). By Lemma 2.4, we obtain a new matrix B' = (b'_{ij}) with
b'_{\ell-1, k+1} = b'_{\ell-1, k+2} = \dots = b'_{\ell-1, \ell-2} = 0.
The assumption that \ell > k+2 implies that (Proof) is not vacuous. Note that b'_{\ell, \ell - 2} must be zero since (\bar{\beta}'_{\ell})^2 is contained in the ideal generated by F_k(B').
Applying Lemma 2.3 for j= \ell - 2, we obtain a new matrix B'' = (b''_{ij}) satisfying
b''_{\ell, \ell - 1} = b'_{\ell, \ell - 2}=0.
This reduces the proof to the case p=0 and the proof is complete. Note that we used Lemma 2.3 for j=\ell-2 and \ell-1, and Lemma 2.4 for j=\ell-1, so the ith row for i<\ell-2 is unchanged.