Tutorials Developer

Jacob Ziv: Compression That Learns the Data as It Goes

By Infinite Creation · Profile

At a Glance

Jacob Ziv (1931-2023) and Abraham Lempel published LZ77 in 1977 and LZ78 in 1978, introducing universal lossless compression that builds its dictionary from the data as it reads it, without prior statistics. The family underlies ZIP, gzip, PNG, GIF and much else.

Lived: 1931-2023, Israel
Anchor year: 1977, LZ77
Strand: overlooked

Career

The problem as it stood

Huffman coding is optimal for symbol-by-symbol coding but needs symbol frequencies in advance, and it cannot exploit repetition of longer sequences. Real data is full of repeated strings that a symbol-frequency approach cannot see.

What he actually did

Ziv and Lempel devised schemes that replace a repeated sequence with a reference to its earlier occurrence: a distance back and a length. The dictionary is the data already seen, so no advance analysis is needed and the decoder reconstructs it as it goes.

How it worked

"Universal" means asymptotically optimal without knowing the source statistics: the algorithm adapts. Because the decoder builds the same dictionary from the same data, nothing needs transmitting alongside. That single-pass, self-describing property is what made it practical for files and streams alike.

What it made possible

General-purpose lossless compression. ZIP, gzip, PNG and GIF all use LZ variants, usually followed by Huffman coding. The combination, dictionary matching then entropy coding. Is the standard architecture of lossless compression.

What happened to him

He was a professor at the Technion and president of the Israel Academy of Sciences, and received the IEEE Medal of Honor. He died in 2023.

Where the credit landed

Ziv and Lempel are properly credited and the algorithms carry their initials, which is better than most in this wave manage. LZW, the Welch variant used in GIF. Was patented by Unisys, and enforcement of that patent in the 1990s caused significant disruption on the early web and directly motivated the creation of PNG.

Known For

The dictionary is the data already seen, so nothing needs transmitting with it.
"Universal" means it adapts without knowing the source statistics in advance.
The LZW patent dispute is why PNG exists.

Knowledge Base

What You'll Learn

Dictionary matching then entropy coding is the standard shape of lossless compression.

Why LZ and Huffman are used together

They exploit different redundancies. LZ finds repeated sequences and replaces them with back-references, Huffman exploits uneven symbol frequencies in whatever remains. Running LZ first and Huffman second captures both, which is exactly what DEFLATE. The algorithm inside ZIP, gzip and PNG, does.

Where This Fits

LZ compression shrinks data for transmission, one step inside the chain traced in A History of Broadcast Technology.

FAQ

Q: What is LZ compression?
A: A family of lossless algorithms replacing repeated sequences with references to their earlier occurrences, building the dictionary from the data itself so nothing needs transmitting alongside.

Q: Why does PNG exist?
A: Largely because Unisys enforced its patent on the LZW variant used in GIF during the 1990s. PNG was created as a free, unencumbered alternative.

Vocabulary

Dictionary compression. Compressing by replacing repeated sequences with references to earlier occurrences. Builds its dictionary from the data itself, so it needs no prior knowledge of what it is compressing.

Redundancy. Repetition or predictability within data, which compression exploits. Text and images contain a great deal, which is why they compress well, and already-compressed data contains little.

Sliding window. The span of recently seen data a compressor can refer back to. A larger window finds more repetition and costs more memory and time, which is the central trade in the approach.

Universal compression. A method working well on data whose statistics are unknown in advance, rather than being tuned to one type. What makes a single algorithm useful for text, code and arbitrary files alike.

Sources

The paper was checked against its publication record. Checked 20 August 2026.

J. Ziv and A. Lempel, A Universal Algorithm for Sequential Data Compression. IEEE Transactions on Information Theory, volume 23, issue 3, pages 337 to 343, May 1977. This is LZ77, and the record names Ziv and Lempel jointly, which is why the whole family of algorithms carries both initials.