A standard 6-sided fair die is rolled until the last 3 rolls are strictly ascending. What is probability that the first such roll is a 1,2,3, or 4?

2.6k Views Asked by At

A standard $6$-sided fair die is rolled until the last $3$ rolls are strictly ascending. What is probability that the first such roll is a $1$, $2$, $3$, or $4$?

My attempt

We can investigate the $3-$roll (when there are exactly $3$ rolls), the $4-$roll, the $5-$roll, the $n-$roll. I did that via SQL, basically n times a cartesian product of a table with digits $1-6$, with some constraints in place.

There are $20$ ascending triplets that can be thrown. Starting with $1: 123, 124, 125, 126, 134, 135, 136, 145, 146, 156$

I call those triplets $T_1$. There are $10$ $T_1$ ‘s. Likewise $T_2$: $234, 235, 236, 245, 246, 256$. There are 6 $T_2$’s. $T_3$: $345, 346, 356$ . There are $3$ $T_3$'s. $T_4: 456$. There is only one $T_4$.

We can start with the $3-$roll: The probability of a $3-$roll is $\frac{20}{216}=\frac{5}{54}$. Each triplet has the same probability of being rolled. So the probability of each triplet is $\frac{1}{20}$.

The $4-$roll. Not any digit can be the first digit. For instance '$456$' can't be prepended by '$1$' or '$2$' or '$3$', as that would have resulted in a $3-$roll. If we break it down we have $6$ digits before a $T_1$, $5$ digits before a $T_2$, $4$ digits before a $T_4$. In total $6*20+5*6+4*3+3*1=105$ quartets. Of those $6$ belong to a specific $T_1$, $5$ to a specific $T_2$, $4$ to a specific $T_3$, $3$ to a specific $T_4$. The probabilities of a specific $T_1$ is $\frac{6}{105}; T_2: \frac{5}{105}; T_3:\frac{4}{105}, T_4: \frac{3}{105}$. The probability of a $4 -$roll happening is $ \frac{105}{6^4}=\frac{35}{432}$

For the $5-$roll we can prepend any digit, so there will be 6*105 quintets and $6*6^4$ total ways, so the probability is still $\frac{6.105}{6.6^4}=\frac{105}{6^4}=\frac{35}{432}$. The probabilities of a specific $T_1$ is $\frac{6.6}{6.105}=\frac{6}{20}$; $T_2$: $\frac{6.5}{6*105}=\frac{5}{105}$; $T_3$:$\frac{6.4}{6*105}=\frac{4}{105}$, $T_4$: $\frac{6.3}{6.105}=\frac{3}{105}$. All probabilities are the same compared the $4-$roll.

Now look at the $6-$roll. Like with the $4-$roll not all digits can be prepended. An sql query tells me there are $3381$ ways to have an 6-roll and that makes the probability $\frac{3381}{6^6}=\frac{1127}{15552}$. To tie them into the T-groups I've reworked the formula's from the $5-$roll to

$6-\frac{\frac{20}{36}}{\frac{2217}{12}}$ for a $T_1$;

$5-\frac{\frac{20}{36}}{\frac{2217}{12}}$ for a $T_2$;

$4-\frac{\frac{20}{36}}{\frac{2217}{12}}$ for a $T_3$;

$3-\frac{\frac{19}{36}}{\frac{2217}{12}}$ for a $T_4$.

This is all matching up. The formula's, just as the sql yield, $1960$ $T_1$'s, $960$ $T_2$'s, $372$ $T_3$'s, $89$ $T_4$'s$.

Now the $7-$roll, and this is where I get stuck. I assumed/hoped we could still use the formulas from the $6-$roll, and I assumed they could be used for the $nth-$roll, where $N>6$. But in the $7-$roll up to the $10-$roll (the $10-$roll is the last one I checked), the T-groups are not 'hit' by the invalid paths proportionally. Although there's only a slight deviation, this means we can't use the formula's from the $6-$roll, and we can't get a precise answer on our question.

So this whole approach seems flawed. Is there another way or a better approach, is what I'd like to know.

4

There are 4 best solutions below

2
On

The modelling probability space is the space of all infinite tuples with entries among $1,2,3,4,5,6$. There is a stopping time $\tau$, the first occurrence of the pattern $$ a<b<c $$ with three last rolls $a,b,c$. When such a pattern is seen, the "game stops". We have $\tau<\infty$ with probability one. So we break the infinite tuples at the point when such an ascending pattern occurs, and deal only with finite tuples. Such a tuple will be written as a word. So instead of $1,3,1,3,1,3,1,4,2,6,4,1,2,2,5,3,3,5,6$ we simply write $w=1313131426412253356$, and stop here. This word stays for the event of all tuples starting with the digits in the word, taken exactly in the same order.

We need the probabilities

  • $p^{(1)}$ for $a=1$, below the "[ONE WINS]" case,
  • $p^{(2)}$ for $a=2$, below the "[TWO WINS]" case,
  • $p^{(3)}$ for $a=3$, below the "[THREE WINS]" case,
  • $p^{(4)}$ for $a=4$, below the "[FOUR WINS]" case,

for the event occuring when in the moment we stop, the last three rolls are $a,b,c$, and $a$ is the specified value.

Below the star "$*$" stays for "not relevant for building an ascending triple". We have the following tree of states, that lead to the one or the other winning case.

                     / *12 ------ *123, *124, *125, *126 [1 WINS]
                    /  *13 ------ *134, *135, *136       [1 WINS]
                   / /
          *1 --------
        /          \ \ 
       /            \  *14 ------ *145, *146             [1 WINS]
      /              \ *15 ------ *156                   [1 WINS]
     /
    /                  *23 ------ *234, *235, *236       [2 WINS]
   /                 / 
  /   *2 ------------- *24 ------ *245, *246             [2 WINS]
 /   /               \ 
*---                   *25 ------ *256                   [2 WINS]
 \   \
  \   *3 ------------- *34 ------ *345, *346             [3 WINS]
   \                 \
    \                  *35 ------ *356                   [3 WINS]
     \
       *4 ------------ *45 ------ *456                   [4 WINS]

Below, there are a lot of connections between nodes, states, that are not shown. For instance, the sample word used above, $w=1313131426412253356$ starts in the empty word, which belongs to $*$, and passes through the following states till a final state is reached, in our case this being the [3 WINS] state: $$ *\to *1\to *13\to*1\to*13\to*1\to*13\to*1\to*14\to*2\to*\to*4\to*1\\ \to*12\to*2\to*25\to*3\to*3\to*35\to*356\ . $$


[FOUR WINS] CASE:

Now let us compute the probability to win with the $4$, i.e. when we stop with three ascending rolls $a<b<c$ that we are in the case $a=4$. This is arguably the easier case. Denote by $p_S=p^{(4)}_S$ the probability that starting in a state $S$ we win with the four. Well, all states are starting with a star, i will omit the star in $p_S$. Also, from the point of view of winning and not winning with $a=4$ the following states can be collapsed to one:

  • $*13$, $*23$ - and we write instead for the corresponding probability simply $p_{X3}$,
  • $*14$, $*24$, $*34$ - corresponding probability is $p_{X4}$,
  • $*15$, $*25$, $*35$ - corresponding probability is $p_{X5}$.

Roughly, the meaning of $X$ is "not four". The system to be solved (to get only $p^{(4)}$ - immediately below denoted by $p$ as long as we focus on the $4$) is now: $$ \left\{ \begin{aligned} 6p &= p_1 + p_2 + p_3 + p_4 + 2p\\ 6p_1 &= p_1 + p_{X2} + p_{X3} + p_{X4} + p_{X5} + p\\ 6p_2 &= p_1 + p_2 + p_{X3} + p_{X4} + p_{X5} + p\\ 6p_3 &= p_1 + p_2 + p_3 + p_{X4} + p_{X5} + p\\ 6p_4 &= p_1 + p_2 + p_3 + p_4 + p_{45} + p\\ 6p_{X2} &= p_1 + p_2 + 0 + 0 + 0 + 0\\ 6p_{X3} &= p_1 + p_2 + p_3 + 0 + 0 + 0\\ 6p_{X4} &= p_1 + p_2 + p_3 + p_4 + 0 + 0\\ 6p_{X5} &= p_1 + p_2 + p_3 + p_4 + p + 0\\ 6p_{45} &= p_1 + p_2 + p_3 + p_4 + p + 1\ , \end{aligned} \right. $$ and note that this system can be easily solved with bare hands, the first equation reads $4p=p_1+p_2+p_3+p_4$, and the $p_{X?}$-variables, and $p_{45}$ can be immediately substituted into the former equations, also writing $p_1+p_2+p_3+p_4=4p$, and $p_1+p_2+p_3= p-p_4$, thus we have: $$ \left\{ \begin{aligned} 4p &= p_1 + p_2 + p_3 + p_4\\ 6p_1 &= p_1 + \frac 16(p_1+p_2) + \frac 16(p - p_4) + \frac 16(4p)\\ &\qquad\qquad + \frac 16(4p + p) + p\\ 6p_2 &= p_1 + p_2 + \frac 16(p - p_4) + \frac 16(4p) + \frac 16(4p + p) + p\\ 6p_3 &= (p - p_4)+ \frac 16(4p) + \frac 16(4p + p) + p\\ 6p_4 &= 4p + \frac 16(4p + p + 1) + p\ , \end{aligned} \right. $$ and the solution of this system may be found in few further steps with bare hands. It is: $$ \begin{aligned} \color{blue}{p^{(4)}=p}&\color{blue}{= \frac{109}{3781}} \end{aligned}\qquad \begin{aligned} p_{ 1 } &= \frac{4805}{272232}\\ p_{ 2 } &= \frac{5425}{272232}\\ p_{ 3 } &= \frac{5}{228}\\ p_{ 4 } &= \frac{211}{3781} \end{aligned}\qquad \begin{aligned} p_{ X2 } &= \frac{1705}{272232}\\ p_{ X3 } &= \frac{75}{7562}\\ p_{ X4 } &= \frac{218}{11343}\\ p_{ X5 } &= \frac{545}{22686} \end{aligned}\qquad \begin{aligned} p_{ 45 } &= \frac{721}{3781} \end{aligned} $$

Code check:

var('p,p1,p2,p3,p4,px2,px3,px4,px5,p45');
eq0 = 6*p   == p1 +  p2 +  p3 +  p4       + 2*p;
eq1 = 6*p1  == p1 + px2 + px3 + px4 + px5 +   p;
eq2 = 6*p2  == p1 +  p2 + px3 + px4 + px5 +   p;
eq3 = 6*p3  == p1 +  p2 +  p3 + px4 + px5 +   p;
eq4 = 6*p4  == p1 +  p2 +  p3 +  p4 + p45 +   p;
eq5 = 6*px2 == p1 +  p2;
eq6 = 6*px3 == p1 +  p2 +  p3;
eq7 = 6*px4 == p1 +  p2 +  p3 +  p4;
eq8 = 6*px5 == p1 +  p2 +  p3 +  p4       +   p;
eq9 = 6*p45 == p1 +  p2 +  p3 +  p4       +   p + 1;

dic = solve([eq0, eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8, eq9],
            [  p,  p1,  p2,  p3,  p4, px2, px3, px4, px5, p45],
            solution_dict=True)[0]

And we obtain:

sage: import pprint
sage: pprint.pprint(dic)
{p: 109/3781,
 p1: 4805/272232,
 p2: 5425/272232,
 p3: 5/228,
 p4: 211/3781,
 px2: 1705/272232,
 px3: 75/7562,
 px4: 218/11343,
 px5: 545/22686,
 p45: 721/3781}


[THREE WINS] CASE:

Now let us compute the probability to win with the $3$, i.e. when we stop with three ascending rolls $a<b<c$ that we are in the case $a=3$. We use similar notations. Denote by $p_S=p^{(3)}_S$ the probability that starting in a state $S$ we win with the three. We use now $Y$ in a similar manner as ist was the case with the $X$ in the previous case, roughly, the meaning of $Y$ is "not three". The system to be solved (to get now $p^{(3)}$ - immediately below denoted by $p$ as long as we focus on the $3$) is: $$ \left\{ \begin{aligned} 6p &= p_1 + p_2 + p_3 + p_4 + 2p\\ 6p_1 &= p_1 + p_{Y2} + p_{Y3} + p_{Y4} + p_{Y5} + p\\ 6p_2 &= p_1 + p_2 + p_{Y3} + p_{Y4} + p_{Y5} + p\\ 6p_3 &= p_1 + p_2 + p_3 + p_{34} + p_{35} + p\\ 6p_4 &= p_1 + p_2 + p_3 + p_4 + p_{Y5} + p\\ 6p_{Y2} &= p_1 + p_2 + 0 + 0 + 0 + 0\\ 6p_{Y3} &= p_1 + p_2 + p_3 + 0 + 0 + 0\\ 6p_{Y4} &= p_1 + p_2 + p_3 + p_4 + 0 + 0\\ 6p_{Y5} &= p_1 + p_2 + p_3 + p_4 + p + 0\\ 6p_{34} &= p_1 + p_2 + p_3 + p_4 + 1 + 1\\ 6p_{35} &= p_1 + p_2 + p_3 + p_4 + p + 1\ , \end{aligned} \right. $$ and the solution of this system is: $$ \begin{aligned} \color{blue}{p^{(3)}=p}&\color{blue}{= \frac{432}{3781}} \end{aligned}\qquad \begin{aligned} p_{ 1 } &= \frac{20119}{272232}\\ p_{ 2 } &= \frac{22715}{272232}\\ p_{ 3 } &= \frac{43}{228}\\ p_{ 4 } &= \frac{420}{3781} \end{aligned}\qquad \begin{aligned} p_{ Y2 } &= \frac{7139}{272232}\\ p_{ Y3 } &= \frac{218}{3781}\\ p_{ Y4 } &= \frac{288}{3781}\\ p_{ Y5 } &= \frac{360}{3781} \end{aligned}\qquad \begin{aligned} p_{ 34 } &= \frac{4645}{11343}\\ p_{ 35 } &= \frac{5941}{22686} \end{aligned} $$

Code check:

var('p,p1,p2,p3,p4,py2,py3,py4,py5,p34,p35');
eq0 = 6*p   == p1 +  p2 +  p3 +  p4       + 2*p;
eq1 = 6*p1  == p1 + py2 + py3 + py4 + py5 +   p;
eq2 = 6*p2  == p1 +  p2 + py3 + py4 + py5 +   p;
eq3 = 6*p3  == p1 +  p2 +  p3 + p34 + p35 +   p;
eq4 = 6*p4  == p1 +  p2 +  p3 +  p4 + py5 +   p;
eq5 = 6*py2 == p1 +  p2;
eq6 = 6*py3 == p1 +  p2 +  p3;
eq7 = 6*py4 == p1 +  p2 +  p3 +  p4;
eq8 = 6*py5 == p1 +  p2 +  p3 +  p4       +   p;
eq9 = 6*p34 == p1 +  p2 +  p3 +  p4       +   1 + 1;
eqA = 6*p35 == p1 +  p2 +  p3 +  p4       +   p + 1;

dic = solve([eq0, eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8, eq9, eqA],
            [  p,  p1,  p2,  p3,  p4, py2, py3, py4, py5, p34, p35],
            solution_dict=True)[0]

And we obtain:

sage: dic
{p: 432/3781,
 p1: 20119/272232,
 p2: 22715/272232,
 p3: 43/228,
 p4: 420/3781,
 py2: 7139/272232,
 py3: 218/3781,
 py4: 288/3781,
 py5: 360/3781,
 p34: 4645/11343,
 p35: 5941/22686}




[TWO WINS] CASE:

Now let us compute the probability to win with the $2$, i.e. when we stop with three ascending rolls $a<b<c$ that we are in the case $a=2$. We use similar notations. Denote by $p_S=p^{(2)}_S$ the probability that starting in a state $S$ we win with the two. We use now $Z$ in a similar manner as it was the case with the $X,Y$ in the previous cases, roughly, the meaning of $Z$ is "not two". The system to be solved (to get now $p^{(2)}$ - immediately below denoted by $p$ as long as we focus on the $2$) is: $$ \left\{ \begin{aligned} 6p &= p_1 + p_2 + p_3 + p_4 + 2p\\ 6p_1 &= p_1 + p_{Z2} + p_{Z3} + p_{Z4} + p_{Z5} + p\\ 6p_2 &= p_1 + p_2 + p_{23} + p_{24} + p_{25} + p\\ 6p_3 &= p_1 + p_2 + p_3 + p_{Z4} + p_{Z5} + p\\ 6p_4 &= p_1 + p_2 + p_3 + p_4 + p_{Z5} + p\\ 6p_{Z2} &= p_1 + p_2 + 0 + 0 + 0 + 0\\ 6p_{Z3} &= p_1 + p_2 + p_3 + 0 + 0 + 0\\ 6p_{Z4} &= p_1 + p_2 + p_3 + p_4 + 0 + 0\\ 6p_{Z5} &= p_1 + p_2 + p_3 + p_4 + p + 0\\ 6p_{23} &= p_1 + p_2 + p_3 + 1 + 1 + 1\\ 6p_{24} &= p_1 + p_2 + p_3 + p_4 + 1 + 1\\ 6p_{25} &= p_1 + p_2 + p_3 + p_4 + p + 1\ , \end{aligned} \right. $$ and the solution of this system is: $$ \begin{aligned} \color{blue}{p^{(2)}=p}&\color{blue}{= \frac{1080}{3781}} \end{aligned}\qquad \begin{aligned} p_{ 1 } &= \frac{4349}{22686}\\ p_{ 2 } &= \frac{9301}{22686}\\ p_{ 3 } &= \frac{5}{19}\\ p_{ 4 } &= \frac{1050}{3781} \end{aligned}\qquad \begin{aligned} p_{ Z2 } &= \frac{2275}{22686}\\ p_{ Z3 } &= \frac{545}{3781}\\ p_{ Z4 } &= \frac{720}{3781}\\ p_{ Z5 } &= \frac{900}{3781} \end{aligned}\qquad \begin{aligned} p_{ 23 } &= \frac{4871}{7562}\\ p_{ 24 } &= \frac{5941}{11343}\\ p_{ 25 } &= \frac{9181}{22686} \end{aligned} $$ Code check:

var('p,p1,p2,p3,p4,pz2,pz3,pz4,pz5,p23,p24,p25');
eq0 = 6*p   == p1 +  p2 +  p3 +  p4       + 2*p;
eq1 = 6*p1  == p1 + pz2 + pz3 + pz4 + pz5 +   p;
eq2 = 6*p2  == p1 +  p2 + p23 + p24 + p25 +   p;
eq3 = 6*p3  == p1 +  p2 +  p3 + pz4 + pz5 +   p;
eq4 = 6*p4  == p1 +  p2 +  p3 +  p4 + pz5 +   p;
eq5 = 6*pz2 == p1 +  p2;
eq6 = 6*pz3 == p1 +  p2 +  p3;
eq7 = 6*pz4 == p1 +  p2 +  p3 +  p4;
eq8 = 6*pz5 == p1 +  p2 +  p3 +  p4       +   p;
eq9 = 6*p23 == p1 +  p2 +  p3 +   1       +   1 + 1;
eqA = 6*p24 == p1 +  p2 +  p3 +  p4       +   1 + 1;
eqB = 6*p25 == p1 +  p2 +  p3 +  p4       +   p + 1;

dic = solve([eq0, eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8, eq9, eqA, eqB],
            [  p,  p1,  p2,  p3,  p4, pz2, pz3, pz4, pz5, p23, p24, p25],
            solution_dict=True)[0]

And we obtain:

{p: 1080/3781,
 p1: 4349/22686,
 p2: 9301/22686,
 p3: 5/19,
 p4: 1050/3781,
 pz2: 2275/22686,
 pz3: 545/3781,
 pz4: 720/3781,
 pz5: 900/3781,
 p23: 4871/7562,
 p24: 5941/11343,
 p25: 9181/22686}



[ONE WINS] CASE:

We can solve a similar system, or simply use the relation $$ p^{(1)} + p^{(2)} + p^{(3)} + p^{(4)} = 1\ . $$ This gives the missing probability: $$ \color{blue}{p^{(1)} = \frac{2160}{3781} \ .} $$



AN OTHER PROBLEM I first understood that we are rolling the dice till three ascending steps appear, i.e. the last four rolls are $a<b<c<d$ with such a triple inequality occurred for the first time, then need the probability to have $1234$ at this stop. Here is the solution for this other problem, although off topic now...

I must draw a diagram with all "relevant paths" to

  • WIN - last four rolls are in order $1234$
  • LOSS / GAME OVER - last four rolls are in order among $1235, 1236, 1345, 1346, 1356, 1456, 2345, 2346, 2356, 2456, 3456$.

The modelling probability space is the space of all infinite tuples with entries among $1,2,3,4,5,6$. There is a stopping time $\tau$, the first occurrence of one of the above ascending patterns with four rolls. When such a pattern is seen, the "game stops". We have $\tau<\infty$ with probability one. So we break the infinite tuples at the point when such an ascending pattern occurs, and deal only with finite tuples. Such a tuple will be written as a word. So instead of $1,3,1,3,1,3,1,4,2,3,4,1,2,2,5,3,4,5,6$ we simply write $w=1313131423412253456$, and stop here. This word stays for the event of all tuples starting with the digits in the word, taken exactly in the same order.

Below the star "$*$" stays for "not relevant for building an ascending tuple with four entries".

               2               3             4
       *1 ---------- *12 ----------- *123 -------- [ *1234 = WIN ] 
      /  \\             \\                \
     /    \\             \\                \ 5,6
    /      \\             \\                 ----  
   /       4\\3           5\\4                    \
  /          \\             \\                     \
 /         3  \\          4  \\                 5,6 \
*----- *2 ----- *13, *23 ------ *124,*134,*234 ------ [ GAME OVER ]
 \        \     \              \                      /    /
  \        \4    \              \                    /6   /
   \        ---- *14,*24 --- *125,*135,*145,*235,*245    /6
    \                                                   /
     *3 ----------- *34 -------- *345 ------------------

There are many connections that are not listed. For example, the long word used above, $w=1313131423412253456$ is going the following path in the state diagram above:

$*\to *1\to *13\to*1\to*13\to*1\to*13\to*1\to*14\to*2\to*23\to*234\to*1\to*12\to*2\to*\to*3\to*34\to*345\to*3456$, which is a losing state.

What are the nodes again, just to be sure that notations are clear...

For instance, the node $*$ stays for the empty word (beginning of the game, no previous rolls), or for any word that terminates in $4,5,6$, and having no last two rolls in (strictly) ascending order. What is for instance $*12$? This is either $12$ or some word of the shape $\dots X12$ with $X$ any character. What is $*235$? This is either $235$ (from the start) or some word of the shape $\dots X235$ with $X\ne 1$. (So the $*$-part of $*235$ is not relevant in conjunction with the end $235$ of the pattern.) I hope these examples are illustrating the notation without any further doubts.

The above Markov type diagram is good enough to write a solution, however, let us reshape it to have even less entries. We observe for instance, that $*125$ is as good as $*345$. So let us have only one node for it. Also $*14,*24$ are as good as $*34$. All what counts is the last roll $L$, and the number $N$ of ascending steps. We will denote such a state by $N[L]$. So all rolls ending in $3$ are of the shape $N[3]$, and $N$ gives the information on the number of ascending steps $j\to k$. Again, let us illustrate. The node $*13$ belongs to $1[3]$. The node $*123$ is $2[3]$. The node $*245$ is $2[5]$. The winning node is $*1234=3[4]$ (three ascending steps, last roll is the $4$). Losing nodes are $3[5]$ and $3[6]$. The start is $0$ as a matter of notation. And many other nodes like $1[5]$, $2[5]$, ... are in fact $0$. The new graph / state diagram is:

     0[1] -- 1[2] -- 2[3] -- 3[4] = WIN 
   /     \\      \\      \\
  /       \\      \\      \\
 /         \\      \\      \\
0--- 0[2] -- 1[3] -- 2[4] -- 3[5] = LOSS
 \        \  \    \  \    \  \
  \        \  \    \  \    \  \
   \        \  \    \  \    \  \
     0[3] -- 1[4] -- 2[5] -- 3[6] = LOSS

There are again many connections between nodes/states that are not shown. For instance, after $*123=2[3]$, if a $2$ is the next roll, then we "step back" to the state $*2=0[2]$. If we are in $*13=1[3]$ and the next roll is a $6$ we go back to $*=0$.

Let now $p_S$ be the probability to win when being in the state $S$. Then we have the following linear system to solve:

$$ \left\{ \begin{aligned} 6p_0 &= p_{0[1]} + p_{0[2]} + p_{0[3]} + 3p_0 \\ 6p_{0[1]} &= p_{0[1]} + p_{1[2]} + p_{1[3]} + p_{1[4]} + 2p_0\\ 6p_{0[2]} &= p_{0[1]} + p_{0[2]} + p_{1[3]} + p_{1[4]} + 2p_0\\ 6p_{0[3]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + p_{1[4]} + 2p_0\\ 6p_{1[2]} &= p_{0[1]} + p_{0[2]} + p_{2[3]} + p_{2[4]} + p_{2[5]} + p_0\\ 6p_{1[3]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + p_{2[4]} + p_{2[5]} + p_0\\ 6p_{1[4]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + p_{2[5]} + p_0\\ 6p_{2[3]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + 1 + 0 + 0\\ 6p_{2[4]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + p_0 + 0 + 0\\ 6p_{2[5]} &= p_{0[1]} + p_{0[2]} + p_{0[3]} + 2p_0 + 0 \\ \end{aligned} \right. $$ The solution of the above system is: $$ \begin{aligned} p_0 &= \frac{1}{14}\\ \end{aligned}\qquad \begin{aligned} p_{0[1]} &= \frac{1325}{18144}\\ p_{0[2]} &= \frac{1273}{18144}\\ p_{0[3]} &= \frac{215}{3024}\\ \end{aligned}\qquad \begin{aligned} p_{1[2]} &= \frac{1585}{18144}\\ p_{1[3]} &= \frac{11}{168}\\ p_{1[4]} &= \frac{5}{72}\\ \end{aligned}\qquad \begin{aligned} p_{2[3]} &= \frac{17}{84}\\ p_{2[4]} &= \frac{1}{21}\\ p_{2[5]} &= \frac{5}{84} \end{aligned} $$ So the answer, the wanted probability to reach $*1234=3[4]$ is $$ \bbox[lightyellow]{ \qquad p_0 =\frac1{14}\ .\qquad } $$



After having seen the solution, we may find of course a simpler one. But building sentences to argue...



Sage code checking the solution, please stop here, if this feels annoying, i am doing it for my own sleep (and for people that are like me):

var('p0,p01,p02,p03,p12,p13,p14,p23,p24,p25')
eq1 = 6*p0  == p01 + p02 + p03 + 3*p0;
eq2 = 6*p01 == p01 + p12 + p13 + p14 + 2*p0;
eq3 = 6*p02 == p01 + p02 + p13 + p14 + 2*p0;
eq4 = 6*p03 == p01 + p02 + p03 + p14 + 2*p0;
eq5 = 6*p12 == p01 + p02 + p23 + p24 + p25 + p0;
eq6 = 6*p13 == p01 + p02 + p03 + p24 + p25 + p0;
eq7 = 6*p14 == p01 + p02 + p03 + p25 + 2*p0;
eq8 = 6*p23 == p01 + p02 + p03 + 1 + 0 + 0;
eq9 = 6*p24 == p01 + p02 + p03 + p0 + 0 + 0;
eqA = 6*p25 == p01 + p02 + p03 + 2*p0 + 0;

dic = solve([eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8, eq9, eqA],
            [p0 , p01, p02, p03, p12, p13, p14, p23, p24, p25],
            solution_dict=True)[0]

This gives:

sage: import pprint
sage: pprint.pprint(dic)
{p0: 1/14,
 p01: 1325/18144,
 p02: 1273/18144,
 p03: 215/3024,
 p12: 1585/18144,
 p13: 11/168,
 p14: 5/72,
 p23: 17/84,
 p24: 1/21,
 p25: 5/84}


Monte-Carlo check:

import random
random.seed('MSE4882983')

N = 1000000    # trials
R = [1, 2, 3, 4, 5, 6]    # the faces

wins = 0    # so far
for trial in range(N):
    a, b, c, d = [random.choice(R) for k in [1,2,3,4]]
    while not(a < b and b < c and c < d):
        e = random.choice(R)
        a, b, c, d = b, c, d, e
    if d == 4:
        wins += 1

p = wins / N
print(f"Statistical probability is {p} ~ {p.n()}")

And we get this time (for the chosen seed) after some coffee break:

Statistical probability is 70997/1000000 ~ 0.0709970000000000

Well, this is somehow far away from $0.0\;714285\;714285\;714\dots$ but the high variance must the reason for it. Working with an other seed, 'MSE' only, i got the result below, so the result $p_0=1/14$ is reasonable.

Statistical probability is 35907/500000 ~ 0.0718140000000000

(Now we are somehow far away on the other side.)

4
On

The process of rolling the dice till getting an increasing triplet is well modelized by a Markov chain: the current state consists in the last three dice rolls, which gives $6^3=216$ possibilities.

We then consider that an increasing triplet is stable (i.e. we stop rolling the dice). The other triplets have a $1/6$ probability to reach any of the triplets obtained by removing the first roll and adding an arbitrary new third roll. The transition matrix is thus a $216 \times 216$ matrix with entries being 0, 1 or $1/6$.

The initial state is well balanced over the 216 possible states and one can iterate this state by left multiplication by the matrix described above. I ran a simulation with 120 digits precision which converges to 0 for all non increasing triplets and positive values for the 20 possible ascending triplets. The results I gave in the comment above are obtained by searching a rational approximation.

To be more precise, the ten ascending triplets starting with a one all have $216/3781$ probability each, the six triplets starting with a two have probability $180/3781$; the three triplets starting with a three have probability $144/3781$ and the only triplet starting with a four has probability $109/3781$.

It should be possible to compute the eigenvalues, eigenvectors to rigorously prove the results. There may also be a more subtle and less computational way to solve the problem.

3
On

This can be modelled using a Markov chain. We need $15$ transient states: One initial state in which we have nothing (either because we just started or because the last roll didn’t leave enough room for the remaining ascent), four states in which we obtained $1$, $2$, $3$ and $4$, respectively, and ten states in which we obtained $12$, $13$, $14$, $15$, $23$, $24$, $25$, $34$, $35$ and $45$, respectively. There are also $4$ absorbing states in which we’ve obtained an ascending triple, but I’ll treat them separately. The transition matrix is

$$ T=\frac16\,\pmatrix{ 2&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 1&1&0&0&0&1&1&1&1&0&0&0&0&0&0\\ 1&1&1&0&0&0&0&0&0&1&1&1&0&0&0\\ 1&1&1&1&0&0&0&0&0&0&0&0&1&1&0\\ 1&1&1&1&1&0&0&0&0&0&0&0&0&0&1\\ 0&1&1&0&0&0&0&0&0&0&0&0&0&0&0\\ 0&1&1&1&0&0&0&0&0&0&0&0&0&0&0\\ 0&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 1&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 0&1&1&1&0&0&0&0&0&0&0&0&0&0&0\\ 0&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 1&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 0&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 1&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ 1&1&1&1&1&0&0&0&0&0&0&0&0&0&0\\ }\;. $$

The matrix of absorption probabilities is

$$ A=\frac16\,\pmatrix{ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0\\ 0&0&0&0\\ 4&0&0&0\\ 3&0&0&0\\ 2&0&0&0\\ 1&0&0&0\\ 0&3&0&0\\ 0&2&0&0\\ 0&1&0&0\\ 0&0&2&0\\ 0&0&1&0\\ 0&0&0&1 }\;. $$

We start in the intial state $\pi=(1\,0\,0\,0\,0\,0\,0\,0\,0\,0\,0\,0\,0\,0\,0)$, there are any number of transitions between the transient states, and then the chain is absorbed in one of the absorbing states. Thus, the desired probability vector $p=\left(p_1\,p_2\,p_3\,p_4\right)$ is given by

\begin{eqnarray*} p &=& \pi\sum_{k=0}^\infty T^kA \\ &=& \pi(1-T)^{-1}A\;. \end{eqnarray*}

Here’s Java code that performs this calculation with exact computation with rational numbers. The result is

\begin{eqnarray*} p &=& \pmatrix{\frac{2160}{3781}&\frac{1080}{3781}&\frac{432}{3781}&\frac{109}{3781}} \\ &\approx& \pmatrix{0.5712774&0.2856387&0.1142555&0.0288284}\;. \end{eqnarray*}

The result is suprisingly close to Haris’ (I believe incorrect) result conditional on $N\ge4$, but not quite as close to the unconditional result.

Here’s Java code that checks the result by simulation. In order to allow the two similar results to be distinguished, I ran $10^{11}$ trials. The results were $\pmatrix{0.5712752&0.2856400&0.1142563&0.0288285 }$, and the standard deviations of these relative frequencies (whose marginal distributions are binomial) are about $\pmatrix{1.6&1.4&1.0&0.5}\cdot10^{-6}$, so the simulation is in good agreement with the present answer.

Full disclosure: At first I got results which, though close to my answer, were off by more than $6$ standard deviations. I looked into the quality of the pseudorandom numbers generated by the standard Random class in Java and found to my surprise that they’re known to suffer from significant correlations between successive numbers (see e.g. here). I reran the simulation using the ThreadLocalRandom class instead and got the above expected results.

0
On

A simple Monte Carlo method:

import random

NUM_SIMULATIONS = 10**6

def roll_die():
    return random.randrange(1, 7)

def roll_until_ascending():
    dice = []
    while len(dice) < 3 or not (dice[-3] < dice[-2] < dice[-1]):
        dice.append(roll_die())
    return dice

counts = {1: 0, 2: 0, 3: 0, 4: 0}

for dummy in range(NUM_SIMULATIONS):
    counts[roll_until_ascending()[-3]] += 1

for value in range(1, 5):
    prob = counts[value] / NUM_SIMULATIONS
    print(f'Probability that the first roll is {value} = {prob:.6f}')

My results for three runs of the script are:

Probability that the first roll is 1 = 0.571111
Probability that the first roll is 2 = 0.286095
Probability that the first roll is 3 = 0.114084
Probability that the first roll is 4 = 0.028710
Probability that the first roll is 1 = 0.570891
Probability that the first roll is 2 = 0.285710
Probability that the first roll is 3 = 0.114338
Probability that the first roll is 4 = 0.029061
Probability that the first roll is 1 = 0.570729
Probability that the first roll is 2 = 0.285701
Probability that the first roll is 3 = 0.114523
Probability that the first roll is 4 = 0.029047