Two dice can carry different labels and still leave exactly the same fingerprint when their results are added.
The Dice Signature
Riddle statement
When two ordinary dice are rolled, the sums from 2 to 12 occur, among the 36 possible outcomes, with frequencies
You may make two other fair six-sided dice. Every face must carry a positive integer, and labels may be repeated.
Is there a pair different from two ordinary dice that produces exactly the same distribution of sums? If so, find it and determine whether any other pair exists.
Show solution
Solution
Answer: yes. Up to swapping the dice, the unique nonstandard pair is
These are known as the Sicherman dice.
Encoding the sums. For a die whose faces are \(a_1,\ldots,a_6\), define
If the two dice correspond to \(P(x)\) and \(Q(x)\), the coefficient of \(x^s\) in \(P(x)Q(x)\) is exactly the number of the 36 outcomes whose sum is \(s\).
For an ordinary die,
We therefore seek two polynomials \(A(x)\) and \(B(x)\) satisfying
Constructing the pair. Distribute the factors as follows:
The coefficients specify how many faces carry each label. Thus the dice are
Since \(A(x)B(x)=D(x)^2\), their sum frequencies are exactly the same as those of two ordinary dice.
Why no other pair exists. The smallest sum is 2 and occurs only once. Each die must therefore have exactly one face labeled 1, so
The three factors
are irreducible in \(\mathbb Z[x]\). Unique factorization means that \(P\) and \(Q\) must be obtained by distributing between them the two copies of each factor occurring in \(D(x)^2\).
Each die has six faces, so
At \(x=1\), the three factors have values \(2\), \(3\), and \(1\). Each polynomial must therefore receive one copy of \(x+1\) and one copy of \(x^2+x+1\).
Only the two copies of \(x^2-x+1\) remain:
- giving one to each die recovers the ordinary pair;
- giving both to the same die produces the Sicherman pair;
- swapping which die receives both copies does not create a new pair.
There are no other possibilities.
The essential idea is to replace a search through labels with a factorization of the entire sum distribution.