Game Show Problem Jacob Richey August 31,2009 Problem: You are on a game show. You start with a score of 0 points. You are then asked a series of questions. For each correct answer (C), your score is increased by 1, and for each incorrect answer (I), your score is divided by 2. After answering π questions, how many distinct scores are possible? Solution: Let π be a sequence of correct and incorrect answers and π₯π or π π₯ be a sequence of answers π₯ added to the beginning or end of π . (For example, πΌπ πΆπΆ represents incorrectly answering, then answering according to π , then answering correctly twice.) Let π β² be the score of a sequence π . (For example, (πΆπΌπΆ)β² = 23 .) Let πΊπ be the set of all sequences of length n with distinct value. Deο¬ne a pair of transformation sequences to be two distinct sequences π₯ and π¦ both applied to some arbitrary sequence π . Then a pair of repeat transformations is one where (π π₯)β² = (π π¦)β² for all s. Finally, deο¬ne a repeat to be two sequences π and π of equal length so that πβ² = πβ² . Claim: β£πΊπ β£ = 2β£πΊπβ1 β£ β β£πΊπβ3 β£ with πΊ0 = 1, πΊ1 = 2 and πΊ2 = 4. Lemma 1: (πΌπ )β² = π β² . If a sequence begins with an incorrect answer, then the score after that answer is unchanged since 02 = 0. Trivially, we can add any number of πΌβs to the beginning of any sequence π without changing its value. Lemma 2: πΆπΆπΌ and πΌπΆ are a pair of repeat transformations that give rise to exactly πΊπβ3 repeats of length π. Also, they are the only pair of repeat transformations of length less than four. β² β² Note that (π πΆπΆπΌ)β² = (π πΌπΆ)β² because π +1+1 = π 2 + 1, which holds for all 2 s. By Lemma 1, we also have that (π πΆπΆπΌ)β² = (πΌπ πΌπΆ)β² . Now, say the sequence π is of length π β 3. The number of diο¬erent sequences π πΆπΆπΌ or πΌπ πΌπΆ simply relies on how many ways we can choose π ; but, since it is of length π β 3, we can choose it exactly β£πΊπβ3 β£ ways. Thus, in producing πΊπ recursively, we must exclude at least β£πΊπβ3 β£ sequences. Now, to show that this pair of transformation sequences is unique among those with length less than or equal to 3, we simply need to write out all the possible transformations of lengths 0, 1, 2, or 3 on an arbitrary sequence π .
1
π=0 π = π β² π=1 (π πΆ)β² = π β² + 1 β² (π πΌ)β² = π 2 π=2 (π πΆπΆ)β² = π β² + 2 β² (π πΆπΌ)β² = π 2 + 12 β² (π πΌπΆ)β² = π 2 + 1 β² (π πΌπΌ)β² = π 4 π=3 β² (π πΆπΆπΌ)β² = π 2 + 1 β² (π πΆπΌπΆ)β² = π 2 + 32 β² (π πΆπΌπΌ)β² = π 4 + 14 β² (π πΌπΆπΆ)β² = π 2 + 2 β² (π πΌπΆπΌ)β² = π 4 + 12 β² (π πΌπΌπΆ)β² = π 4 + 1 β² (π πΌπΌπΌ)β² = π 8 The only two of these that have exactly the same value are π πΌπΆ and π πΆπΆπΌ; all the other pairs of expressions are equal either once or zero times, but certainly not always. Therefore, this pair of repeat transformations is unique among transformations of length less than or equal to 3. Lemma 3: Any pair of repeat transformations of length greater than 2 (aside from the π πΆπΆπΌ/πΌπ πΌπΆ one) can be ignored, as they have all already been counted when considering repeats of πΊπ of the form π πΆπΆπΌ/πΌπ πΌπΆ. Assume there is some pair of transformation sequences, π₯ and π¦, one of which is longer than three answers, that act on some sequence π of length π β π. Then the longer transformation sequence, suppose WLOG it is π₯, has length π with π β₯ 4. Now lets consider the partial transformation of π₯ on π of total length π β 3. (All but the last 3 answers of π₯ have been applied to π .) Letβs call this truncated sequence π€. By Lemma 2, we know that the only way π€ can always be a repeat in πΊπ is if we apply the πΆπΆπΌ/πΌπΆ transformation to it. Therefore, the one of the transformations π₯ and π¦ must end in πΆπΆπΌ and the other in πΌπΆ. However π€ was constructed to begin with, it is a member of πΊπβ3 , and since it is a repeat of the desired form it was counted in our analysis in Lemma 2. Lemma 4: Any pair of transformations on an arbitrary sequence π of length 2 or less, (i.e., the expressions listed above), that yield a positive solution for π β² , (as a negative score is impossible), have already been counted in the repeats given by the π πΆπΆπΌ/πΌπ πΌπΆ repeat transformation pair.
2
We need only check all the pairs of two of the 7 transformations of length two or less, (i.e., all those in the above table under π = 0, π = 1 and π = 2), and see if they have solutions that arenβt counted as π πΆπΆπΌ/πΌπ πΌπΆ transformations. We πβπ , note that the general solution to such an equation, ππ β² + π = ππ β² + π is π β² = πβπ which is non negative only when π β₯ π and π β₯ π or π β€ π and π β€ π. We also note that when the coeο¬ecients of π β² are equal the equation has no solution, and when the transformation includes no correct answers, (i.e., π = π = 0), the solution is zero, which is never a repeat as it can only be generated in one way, namely as a sequence of incorrect answers. This eliminates all but two equations: π = 2π + 1 and π = 2π + 21 , whose solutions are 2 and 1, respectively. These give the equalities (πΆπΆ)β² = (πΆπΆπΌπΆ)β² and (πΆ)β² = (πΆπΆπΌ)β² . However, we note that both of these expressions are special cases of the equality (πΌπΆπ₯)β² = (πΆπΆπΌπ₯)β² , with perhaps some πΌβs added at the front to balance their lengths. (They could be re-written as (πΌπΌπΆπΆ)β² = (πΆπΆπΌπΆ)β² and (πΌπΌπΆ)β² = (πΆπΆπΌ)β² .) Therefore, they were counted in our analysis in Lemma 2, and add no additional repeats. Proof of Claim: To recursively deο¬ne πΊπ , note that for each successive π we can attach either a correct or an incorrect answer to the end of any sequence in πΊπ . However, we must subtract oο¬ all the new repeated values we get in πΊπ+1 , and by the above analysis there are exactly β£πΊπβ2 β£ of those cases. Therefore β£πΊπ+1 β£ = 2β£πΊπ β£ β β£πΊπβ2 β£, which is the desired result. Note: In general, πΊπ = πΉπ+2 β 1, where πΉπ is the πth Fibonacci number and πΉ1 = 1, πΉ2 = 1.
3