Asymmetric Polya

Initially, a large urn contains 3 black marbles and 1 white marble. Every minute, a marble is chosen at random from the urn, and then returned to the urn, together with another marble of the same colour.

If the probability that, after exactly one hour, precisely three-quarters of the marbles in the urn are black can be written as a b \dfrac{a}{b} , where a a and b b are coprime positive integers, find a + b a+b .


The answer is 40792.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

6 solutions

Mark Hennings
May 25, 2016

Relevant wiki: Induction - Problem Solving

Let p n , k p_{n,k} be the probability that there are k + 3 k+3 black marbles in the urn after n n minutes. Conditional probability considerations tell us that p n , 0 = n n + 3 p n 1 , 0 p n , n = n + 2 n + 3 p n 1 , n 1 p_{n,0} \; = \; \frac{n}{n+3} p_{n-1,0} \qquad \qquad p_{n,n} \; = \; \frac{n+2}{n+3} p_{n-1,n-1} for any n 1 n \ge 1 , with p 0 , 0 = 1 p_{0,0} = 1 , and hence p n , 0 = 6 ( n + 1 ) ( n + 2 ) ( n + 3 ) p n , n = 3 n + 3 p_{n,0} \; = \; \frac{6}{(n+1)(n+2)(n+3)} \qquad \qquad p_{n,n} \; = \; \frac{3}{n+3} More generally, if 1 k n 1 1 \le k \le n-1 then p n , k = n k n + 3 × p n 1 , k + k + 2 n + 3 × p n 1 , k 1 p_{n,k} \; = \; \frac{n-k}{n+3} \times p_{n-1,k} + \frac{k+2}{n+3} \times p_{n-1,k-1} and these equations can be solved inductively to yield p n , k = 3 ( k + 1 ) ( k + 2 ) ( n + 1 ) ( n + 2 ) ( n + 3 ) 0 k n p_{n,k} \; = \; \frac{3(k+1)(k+2)}{(n+1)(n+2)(n+3)} \qquad \qquad 0 \le k \le n In particular p 60 , 45 = 1081 39711 p_{60,45} = \frac{1081}{39711} , making the answer 40792 \boxed{40792} .

Fencepost error. If the balls are operated on every minute for an hour there are 61 operations. That would make 3/4 of 65 balls 48 and 3/4 making the problem as stated insolvable.

Terry Smith - 5 years ago

Log in to reply

Arguable. If the phrase "every minute" is taken to mean "at the end of every minute", then there are only 60 selections. Given that the question implies that it must be possible for there to be three-quarters black marbles after one hour, this must be the reading of the question.

Mark Hennings - 5 years ago

can you tell me how you solve these equations inductively? thanks!!!

Willia Chang - 4 years, 11 months ago

Log in to reply

Let P n P_n be the statement that the formula for p n , k p_{n,k} holds for each k k between 0 0 and n n . In other words, perform induction on n n .

Mark Hennings - 4 years, 11 months ago

Log in to reply

OH...ic, thanks!!! but i'm still curious as to how you came up with the statement in the first place?

Willia Chang - 4 years, 11 months ago

+1 @Mark Hennings how did you come up with the inductive formula? I know you can prove it is true using induction, but coming up with it seems like the hard part

Forest Yang - 3 years, 11 months ago

Log in to reply

Ultimately, experience. Two-dimensional recurrence relations turn up from time to time, and after a while you recognise the possibilities.

If you investigate, you discover that the similar recurrence relation for the BBW initial setup gives a linear expression for p n , k p_{n,k} , and a BBBBW initial setup gives a cubic expression. There is a pattern here!

Mark Hennings - 3 years, 11 months ago
Geoff Pilling
Jun 5, 2016

This was one of my favorite problems on the site. Original, and fun to solve... Thanks for posting it @Mark Hennings ! (+1 like!)

My solution:

Let T i T_i be the i i 'th triangular number. i.e. T i = 1 + 2 + . . . + i T_i = 1+2+...+i

Let P ( m , n ) = P(m,n) = The probability that you will have m m black balls and n n white balls once the corresponding number of moves have taken place after starting with 3 3 black balls and 1 1 white ball.

Then, P ( m , n ) = T m 2 i = 1 m + n 3 T i P(m,n) = \frac{T_{m-2}}{\sum_{i=1}^{m+n-3}T_i}

So, P ( 48 , 16 ) = T 46 i = 1 61 T i = 1081 37711 P(48,16) = \frac{T_{46}}{\sum_{i=1}^{61}T_i} = \frac{1081}{37711}

1081 1081 (factors 23 23 and 47 47 ) and 37711 37711 (factors 3 3 , 7 7 , 31 31 and 61 61 ) are coprime, so the answer is 1081 + 37711 = 40792 1081 + 37711 = \boxed{40792}

I am interested in how you came up with your (correct) formula...

Mark Hennings - 5 years ago

Log in to reply

Ah, let me think if I can come up with a good way to show it is true for all m,n... If I do, I'll be sure to post it. So far I haven't been able to prove it rigorously, but I just saw the pattern for all m,n up to m+n = 9, and convinced myself that the pattern would continue for all m,n.

It took me a couple of weeks to finally see this pattern. At first I expected the probabilities to "flatten out" (since they remain equal for the symmetric Polya) so I was sure I was doing something wrong... :-P

Thats what I liked so much about the problem is that the answer wasn't what I expected at all, yet in the end it has a very elegant solution!

Geoff Pilling - 5 years ago

Log in to reply

The probabilities do not flatten out. If you start with b b black and w w white balls in the urn, the long term distribution of black balls follows a beta distribution - this just happens to be uniform if b = w = 1 b=w=1 .

If we start with a symmetric Polya urn, and the first two balls are black, we end up with the asymmetric urn of this question, which is not flat. However, it would have been equally likely that the first two balls were white, which would have led to an asymmetric urn in the other direction, and these two would balance each other out, which is why the symmetric urn can have asymmetric variants...

Mark Hennings - 5 years ago

What we want at last is 48 , 16 48,16 Black and white Balls respectively, given each second adds one ball to the set.

So first we add 48 blacks and then 16 whites.

P ( B ) = ( 3 ) ( 4 ) ( 4 ) ( 5 ) ( 5 ) ( 6 ) . . . . . ( 4 8 ) ( 49 ) P(B) = \frac(3)(4)*\frac(4)(5)*\frac(5)(6).....\frac(48)(49)

P(Adding first black ball )*P(adding adding second black ball)...

Followed by

P ( W ) = ( 1 ) ( 50 ) ( 2 ) ( 51 ) . . . . ( 1 5 ) ( 63 ) P(W) = \frac(1)(50)*\frac(2)(51)....\frac(15)(63)

P(adding first white ball after having a total of 48 black balls and 1 white ball)*P(adding second white ball)....

Our probability of getting the desired result is P ( A ) P ( B ) P(A)*P(B) ..

Vishal Yadav - 4 years, 3 months ago

Log in to reply

Those are 3 4 . . . \frac {3}{4}... and 1 50 . . . \frac {1}{50}... .

Vishal Yadav - 4 years, 3 months ago
Forest Yang
Jun 27, 2017

L a T e X LaTeX I'll show two ways of doing it, the second way is cooler. The first way is: By observing the terms that go in when applying Bayes rule, one notices it doesn't matter what order you pick the balls in, the probability of a specific order of choosing 15 white balls and 45 black balls is the same. The number of orderings is ( 60 15 ) {60\choose 15} and so the answer is \begin{equation*} {60\choose 15}\frac{3\times 4\times 5\times\ldots \times 47\times 1 \times 2\times\ldots \times 15}{4\times 5\times\ldots\times 63} = \frac{47\times 23}{21\times 31\times 61} \end{equation*}

Here is the second way, a combinatorial solution. Consider the setup \begin{equation*} BB| \end{equation*} The | represents a divider, the B's are "black weights" and since there is only one white ball we don't need a W or a "white weight." Let me explain. There are 3 insertion positions to the left of the divider and 1 insertion position to the right of the divider so the probability of inserting a ball on the left vs. on the right is the same as choosing a black ball vs. a white ball. And one can see that even after you add a ball to either side, the probabilities of inserting on either side keeps on matching the probability of choosing a black/white ball. Since we are adding balls randomly to this configuration it turns out that all final configurations are equally likely. There are ( 63 3 ) {63 \choose 3} final positions (fix the BB|) and there are ( 47 2 ) {47\choose 2} valid configurations (45 added on the left of the divider, 15 on the right, fix two of the 47 positions on the left) so another way to get the answer is \begin{equation*} \frac{{47\choose 2}}{{63 \choose 3}} = \frac{47\times 23}{21\times 31\times 61}\end{equation*}

Sal Gard
Jul 17, 2016

My solution. Consider the ways one can choose the marbles. First of all, we will have to pick a white marble 15 times and a black marble 45 times. This means that at some point the probability will be a number ranging from 1 to 15 divided by some number and from 3 to 47 over some number because these are the number of choices. Now the denominator is the product of the total number of choices, which ranges from 4 to 63. To account for permutations, we add all work together multiplying 47!/2! for the numbers 3 to 47, 15! for the number 1 to 15, and the permutation 60!/15!/45! which will count the permutations, and finally dividing by 62!/3!. Multiplying this out yields 1081/39711 and our answer is 40792.

Moderator note:

The solution could be rigorized further, by explaining how to set up the probability, which indicates that the numerator / denominator is as described.

Timon Gurcke
Jan 3, 2019

The probability of a single combinations is given by:

n = 3 47 n n + 1 n = 1 16 n n + 48 \prod_{n=3}^{47}\frac{n}{n + 1} * \prod_{n=1}^{16}\frac{n}{n + 48}

There are ( 60 45 ) \binom{60}{45} ways new marbles can be picked.

There are ( 4 3 ) \binom{4}{3} ways to arrange the marbles that are already in the urn.

Hence the probability for all the combinations is:

n = 3 47 n n + 1 n = 1 16 n n + 48 ( 60 45 ) ( 4 3 ) = 1081 39711 \prod_{n=3}^{47}\frac{n}{n + 1} * \prod_{n=1}^{16}\frac{n}{n + 48}* \binom{60}{45} * \binom{4}{3}=\frac{1081}{39711}

The answer is 1081 + 39711 = 40792 1081+39711=\boxed{40792}

Gopinath No
Jul 23, 2016

Another form of recurrence:

Let P ( a , b ) P(a,b) be the probability of the urn having a a white and b b black marbles.

Then P ( a , b ) = a 1 a + b 1 P ( a 1 , b ) + b 1 a + b 1 P ( a , b 1 ) P ( 1 , 3 ) = 1 \begin{aligned} P(a,b) &= \frac{a-1}{a+b-1} \cdot P(a-1,b)+\frac{b-1}{a+b-1} \cdot P(a,b-1)\\ P(1,3) &= 1 \end{aligned} and 0 0 if a < 1 a<1 or b < 3 b<3

We need to find P ( 16 , 48 ) P(16,48)

True, but this is the same as my solution, barring notation.

Mark Hennings - 4 years, 10 months ago

Log in to reply

Ah, I see. P ( a , b ) = p a + b 4 , b 3 P(a,b)=p_{a+b-4,b-3}

gopinath no - 4 years, 10 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...