Continuing from this earlier blog post: For the remainder of this post, assuming that $m$ is almost perfect, then $m$ must be a square (since $m$ is odd). Consequently, if the equation $\sigma(p^k)=2m$ holds (which is true if and only if $G=H=I$), then $\sigma(p^k)/2$ is also a square. This implies that $k=1$, since $$p^k + 1 \leq \sigma(p^k)=2m<2p$$ holds under the assumption $m<p$.
Summarizing our results so far, we have the chain of implications:
$$m < p \Rightarrow \Bigg((m < p^k) \land (k = 1) \land (D(m) = 1)\Bigg).$$
Since $\sigma(m^2)/p^k$ is also a square if $\sigma(p^k)/2$ is a square, then under the assumption that $m < p$, we obtain $\sigma(m^2)/p^k = \sigma(p^k)/2 = m$ is a square. (Note that the difference $m^2 - p^k$ is also a square.)
Additionally, we obtain
$$\sigma(p^k) = p^k + 1$$
since $k=1$ follows from the assumption $m < p$. Since $m^2 - p^k$ is a square if and only if
$$m^2 - p^k = (m - 1)^2,$$
it then follows from the assumption $m < p$ that $\sigma(m) = 2m - 1 = p^k$, whence we obtain
$$2m = p^k + 1 = \sigma(p^k)$$
and
$$2m - 1 = \sigma(m).$$
Since $m^2 - p^k$ is not a square, then we obtain
$$m^2 - p^k = 2^r t$$
where $r \geq 2$ and $2^r \neq t$.
Now, consider the following scenario in this MathOverflow question, under which the inequality $m < p^k$ is sure to hold:
$$\text{Case (A): } m > \max(2^r, t)$$
Case (A) implies that
$$\left(m - \max(2^r, t)\right)\left(m + \min(2^r, t)\right) > 0$$
$$p^k = m^2 - 2^r t = m^2 - \min(2^r, t)\max(2^r, t)$$
$$ > m\left(\max(2^r, t) - \min(2^r, t)\right) = m\left|2^r - t\right|,$$
from which we get
$$p^k > m\left|2^r - t\right|.$$
Note that this implies that $m < p^k$ since $\left|2^r - t\right| \geq 1$ must necessarily hold.
Next, notice that the other (remaining) cases are
$$\text{Case (B): } \min(2^r, t) < m < \max(2^r, t)$$
$$\text{Case (C): } m < \min(2^r, t).$$
The disposal of Case (C) is very easy and is left as an exercise for the interested reader.
We then have the biconditional
$$\left(m < \max(2^r, t)\right) \iff \left(p^k < m\left|2^r - t\right|\right).$$
$$p^k < m \Rightarrow \text{ Case (B) } \Rightarrow \left|2^r - t\right| \neq 1.$$
By the contrapositive, we obtain
$$\left|2^r - t\right| = 1 \Rightarrow m < p^k.$$
However, under the assumption $m < p$, we know that $k = 1$. Together with Acquaah and Konyagin's estimate (2012), we obtain
$$p^k < m\sqrt{3}$$
and this works under the assumption $k=1$. It follows that $m < p^k < 2m$ holds when $k=1$ is true.
Consequently, under Case (A), we derive
$$m \leq m\left|2^r - t\right| < p^k < 2m$$
from which it follows that
$$1 \leq \left|2^r - t\right| < 2$$
which forces $\left|2^r - t\right| = 1$.
We conclude that
$$m < p^k \iff \left|2^r - t\right| = 1.$$
Conclusion: We now claim that the inequality $p < m$ indeed holds. In deed, assume to the contrary that $m < p$ holds. It follows that $k = 1$. If the equation $\sigma(p^k)=2m$ holds, then we obtain
$$2m=\sigma(p^k)=p+1$$
$$2m - 1 = p$$
But then again, as a by-product of Acquaah and Konyagin's results published in the International Journal of Number Theory in 2012, we have the upper bound
$$p < m\sqrt{3}.$$
Consequently,
$$2m - 1 = p < m\sqrt{3}$$
$$m(2 - \sqrt{3}) < 1.$$
This contradicts $m \geq 4$.
Hence, we infer that either the estimate
$$p < m$$
or the inequation
$$\sigma(p^k) \neq 2m$$
must be true.
