When shuffling a deck of cards using a riffle shuffle, one divides the deck in two and lets the two halves riffle down to the table, interleaving as they do so. Assume that a person using this shuffle will always divide a deck of 52 cards exactly evenly, and that the riffle will start equally often from the left as from the right.
The expert dealer that I am, when I perform a riffle shuffle the cards from the two halves always interleave perfectly, the cards alternating from the left and right halves of the deck.
How many times must I shuffle the deck before the probability of correctly guessing the next card down in the deck after seeing a card chosen randomly from some place in the deck will be less than 1.97%? (If the cards were perfectly random, the probability of correctly guessing the next card would be 1/51 = 1.96%)
Bonus: What if there were a 10% chance that, as each card falls during the riffle, the card will be covered by another card from the same half, instead of strictly alternating?
(Assume that the person guessing knows the original order of the cards, the number of times the deck has been shuffled, and the probability of the cards interleaving perfectly.)
Using monte carlo simulation of 1M trials...
When tracking the position of a single card, it's distribution throughout the shuffled deck becomes almost uniform at 10 shuffles. The way the histograms appear in my simulation shows that even at 6 and 7 shuffles, there are large regions in the deck where one would expect to find the card. At 8 shuffles and beyond, it switches and the entire distribution is roughly uniform at a little less than 2% (avg 1.92% = 1/52) with only small regions of slightly lower probability. At 12 shuffles, the max of the distribution is below 1.97%
The question however was about predicting the next card in the deck. Here, even at 8 shuffles there are still large spikes in the likelihood of certain cards following a selected card (on the order of 10-15%) Unfortunately, these spikes don't seem to go away. Even at 12-14 shuffles I see spikes on the order of 5%.
More simulation
