Al-Kindi

Al-Kindī

Al-Kindī was a ninth-century philosopher and scientific writer associated with Baghdad and a circle of translators working with Greek texts. His life dates are approximate. Reference works also disagree about his birthplace: the Stanford Encyclopedia gives Basra, while MacTutor gives Kufa. This reading does not convert that disagreement into false certainty. His philosophical work, including On First Philosophy, should not be reduced to a decorative prelude to cryptography. Stanford Encyclopedia account and MacTutor biography.

His cryptanalytic treatise is important evidence for the use of language statistics in decipherment. The task is to understand what those statistics preserve and what they cannot establish, rather than claim that frequency analysis automatically defeats any cipher.

The primary passage contains its own limitation

A translated excerpt hosted by Western Connecticut State University describes comparing letter counts in a sufficiently long known-language text with those in an encrypted message. It also explicitly warns that short messages may not follow the expected frequency ordering and discusses additional constraints from letter combinations. That qualification is central to the method. Read the excerpt.

The earlier biography's assertion that the most common ciphertext symbol must represent the most common letter of the language was wrong. It represents the most common letter of this particular plaintext, assuming a one-to-one substitution. The sample may differ substantially from a larger language corpus.

The BnF catalog identifies a 2003 edition, al-Kindi's Treatise on Cryptanalysis, published by the King Faisal Center. A catalog record verifies an edition's identity; it does not constitute access to all of its text. No unverified discovery anecdote or reconstructed diplomatic assignment is needed to teach the mechanism. Edition record.

Worked mechanism: what a substitution preserves

This original example uses a fixed one-to-one substitution over five symbols:

Plain:   A B R C D
Cipher:  Q M Z P T

The plaintext ABRACADABRA becomes QMZQPQTQMZQ. Count the occurrences:

Plain symbol	Count	Corresponding cipher symbol
A	5	Q
B	2	M
R	2	Z
C	1	P
D	1	T

If f is the substitution and count_P(a) counts plaintext symbol a, then count_C(f(a)) = count_P(a). Every occurrence is replaced once, and injectivity prevents another plaintext symbol from contributing to the same ciphertext symbol. This proves a preservation property of the cipher.

It does not prove a recovery algorithm. Suppose an analyst knows the plaintext counts but not the mapping. The unique count 5 identifies A -> Q. The two symbols with count 2 can be paired in two ways, and the two with count 1 in two ways. Four candidate assignments remain compatible with the histogram. Frequencies alone cannot distinguish them.

Order supplies more evidence. The first four plaintext symbols repeat as the last four; the ciphertext preserves that equality pattern. A proposed word or phrase must agree everywhere a symbol recurs. Mapping Q to two different plaintext letters would violate the assumed cipher model even if the resulting sentence sounded plausible.

Evidence, assumptions, and ambiguity

An analyst must identify the model before trusting a statistic. A fixed substitution preserves counts up to a renaming. A transposition preserves the counts without renaming but changes positions. A position-dependent substitution can map different occurrences of one letter to different symbols, so the simple histogram correspondence no longer holds.

A short ciphertext may have several reasonable plaintexts under different keys. Recovering a readable candidate does not establish uniqueness. One needs additional constraints, a known fragment, a sufficiently informative source model, or independent confirmation. Even a long message can have unusual vocabulary or deliberate stylistic restrictions.

This is a practical lesson in inference: a frequency table is evidence under assumptions. It is not a proof that the guessed plaintext is correct. The translated historical excerpt is stronger pedagogy than the claim that “structure cannot hide,” because it names an actual limitation.

A direct historical line to Shannon, Turing, or modern encryption design would require documentary evidence of transmission. Shared use of statistical information establishes a useful comparison, not a proven chain of personal influence.

Exercises

1. Encrypt CAB using the table above, then count its symbols. Explain why using the most common English letter as an automatic guess would be unjustified.
2. Given ciphertext counts Q:5, M:2, Z:2, P:1, T:1 and the known plaintext counts above, enumerate the ambiguity remaining after identifying A. How many candidate maps remain? Would counts reveal the order of the message?
3. An encoder alternates between mapping A to Q on odd positions and A to Z on even positions. Encrypt AAAA, compare the resulting histogram with fixed substitution, and identify the failed assumption. Explain why this example does not establish the security of the altered encoder.

Compare your reasoning with the worked solutions

1. A small sample

The ciphertext is PQM. Each symbol appears once, so there is no unique most frequent symbol. The plaintext contains no E, and the sample is too small to represent typical language frequencies. A general corpus statistic can suggest hypotheses for suitable data; it cannot force a missing letter to occur.

2. Four maps and many orders

A -> Q is fixed by the unique count 5. Either B -> M, R -> Z or B -> Z, R -> M is consistent. Independently, either C -> P, D -> T or the reverse is consistent. Multiplying gives 2*2=4 assignments. Counts also discard sequence order: rearranging the eleven symbols leaves the histogram unchanged. Thus even a completely known histogram is not a complete message.

3. The model changed

The ciphertext is QZQZ, with counts Q:2 and Z:2, rather than four occurrences of one symbol. There is no single symbol-only function f(A) for all positions. The fixed-substitution proof therefore does not apply. But the alternation itself creates a pattern; splitting odd and even positions recovers two separate streams. Invalidating one attack assumption is not a proof that all other attacks fail.

Sources and reading limits

• Al-Kindī, Peter Adamson, Stanford Encyclopedia of Philosophy: philosophical works, translator circle, and cautious chronology.
• Al-Kindī biography, University of St Andrews: a differing birthplace account. The disagreement remains explicit.
• Arabic Analysis of the Alphabet, Western Connecticut State University: translated primary excerpt, including the warning about short texts. The historical discussion here relies on the translated excerpt; it does not claim to collate the complete treatise.
• al-Kindi's Treatise on Cryptanalysis, Bibliothèque nationale de France: a specific edition record, not free full text. The cipher, proof, and exercises in this reading are original examples.