Inspired by this question, what is the maximum theoretical compression ratio of brotli?
I was able to find a lower-bound by experimenting with the brotli cli, but I'd like someone who understands the algorithm better to find a more formal answer.
1GB (1073741824 bytes) of null bytes, with no dictionary, at the best (11) (default) compression level, gets compressed to 841 bytes for a compression ratio of ~1276744x.
Analyzing the archive, there seems to be a head, a repeating segment and a tail. In the 1GB archive, this segment has been repeated 3 times.
Repeating it 1 and 2 times, gave the following valid archives:
| Download Link | Compressed Size | Decompressed Size |
|---|---|---|
| repeat1 | 421 bytes | 536870912 bytes |
| repeat2 | 631 bytes | 805306368 bytes |
| repeat3 (original) | 841 bytes | 1073741824 bytes |
These data points exhibit a linear dependency between the decompressed and compressed size where the slope is 1278264.076190476 (268435456 decompressed bytes for every 210 compressed bytes).
