Reveal - Lomsat gelral

For each node, you need to merge frequency maps from all its children. If you merge naively, you recompute everything from scratch, giving $O(n^2)$ time. Small-to-Large merging fixes this. For each node, keep the largest child's frequency map.

Then merge all smaller children into it, one color at a time. Each color gets moved at most $O(\log n)$ times (it only moves when its subtree is the smaller one). Total: $O(n \log n)$.

Space complexity is $O(n)$ for the data structures used.