Well, I might venture a new guess, although I'm unsure how to express it eloquently. From what I can see, if you are trying to come up with n divisions, you end up breaking the original square into predictable groupings of 2 smaller rectangles, which have to be further divided. I still postulate that rect(n)=2*square(n), and based on that, here's what I'm seeing:
for 7 divisions: 1 rectangle divided 4 times, 1 rectangle divided 3 times
1 rectangle divided 5 times, 1 rectangle divided 2 times
1 rectangle divided 6 times, 1 rectangle divided 1 time
for 6 divisions: 1 rectangle divided 3 times, 1 rectangle divided 3 times
1 rectangle divided 4 times, 1 rectangle divided 2 times
1 rectangle divided 5 times, 1 rectangle divided 1 time
for 5 divisions: 1 rectangle divided 3 times, 1 rectangle divided 2 times
1 rectangle divided 4 times, 1 rectangle divided 1 time
for 4 divisions: 1 rectangle divided 2 times, 1 rectangle divided 2 times
1 rectangle divided 3 times, 1 rectangle divided 1 time
Then, if rect(n)=2*square(n), I think you can express:
square(4)=rect(3)+rect(2)-1 =4+2-1=5
square(5)=rect(4)+rect(3)-3 =10+4-3=11
square(6)=rect(5)+rect(4)+rect(3)-5=22+10+4-5=29
for square(7), the answer would be 83 - not sure I want to draw that out to confirm!
As far as an expression goes, I guess this might do:
square(n)=rect(n-1)+rect(n-2)+...+rect(n/2)-[(n-3)*2-1]
where rect(n) = 2*square(n)
Of course, rect(n/2) is only defined for even values of n, so the recursion would have to stop with rect(n/2+0.5) for odd numbers. I can't figure out if my last term is correct - the subtraction of an ever-increasing odd number - because I can't understand a geometric reason for it. Also, the formula doesn't work for square(3), which makes me more suspicious of that term. But the geometry seems good to me for the recursion.
And if I'm wrong, well - I've been wrong before. I think the last time was yesterday!