No algorithm can compress every file: if every string got shorter, two different strings would have to share one output and decompression could not tell which to restore. The real ceiling is the Shannon entropy of the symbol distribution. When one symbol dominates, the Huffman tree leans hard to one side and that symbol earns a code of a single bit; when all symbols occur equally often, entropy hits its maximum of log₂N, every code ends up the same length and there is nothing left to gain. Images and audio get further only by throwing away detail the eye and ear barely notice.
No algorithm can compress every file: if every string got shorter, two different strings would have to share one output and decompression could not tell which to restore. The real ceiling is the Shannon entropy of the symbol distribution. When one symbol dominates, the Huffman tree leans hard to one side and that symbol earns a code of a single bit; when all symbols occur equally often, entropy hits its maximum of log₂N, every code ends up the same length and there is nothing left to gain. Images and audio get further only by throwing away detail the eye and ear barely notice.
Related topics in Mind & Machine