A Probabilistic Analysis of the Game of the Goose

Author(s): Jan Friso Groote, Freek Wiedijk and Hans Zantema

Source: SIAM Review, Vol. 58, No. 1 (March 2016), pp. 143-155
Published by: Society for Industrial and Applied Mathematics
Stable URL: [Link]
Accessed: 27-11-2025 07:46 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

Society for Industrial and Applied Mathematics is collaborating with JSTOR to digitize,
preserve and extend access to SIAM Review

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
SIAM REVIEW (c) 2016 Society for Industrial and Applied Mathematics
Vol.58. No. I, pp. 143-155

A Probabilistic Analysis of the

Game of the Goose*

Jan Friso Groote*

Freek Wiedijk*
Hans Zantema^

, We analyze the traditional board game the Game of the Goose. We are particularly
interested in the probabilities of the different players winning, and we show that we can
determine these probabilities exactly for up to six players and using simulation for any
number of players. Our original motivation to investigate this game came from progress
in stochastic process theories, which prompted the question of whether such methods are
capable of dealing with well-known probabilistic games. As these games have large state
spaces, this is not trivial. As a side effect we find that some common wisdom about the
game is not true.

Key words, probabilistic game, Game of the Goose, winning probabilities, recreational mathematics

AMS subject classifications. 97A20, 60G05

DOI. 10.1137/140983781

I. Introduction. The Game of the Goose is a traditional board game that comes
in many variations and is commonly known, especially in Europe [1]. In the game,
the players are geese, which allows us to refer to them by the neutral gender (e.g., "its
turn," "it wins"). The board has a number of fields and the goal for each goose is to
be the first to reach the last field. On its turn, a player throws two dice and moves
the indicated number of spots forward. On its way to the goal a player can encounter
several difficulties, among which are the well and the prison, where the player is held
until released by another player.
In [9] the Game of the Goose is described as "historically the most important
spiral game ever devised" and it is claimed that the game stems from the Italy of
Francesco de Medici (1574-1587), but similar games were already popular in ancient
history, especially among the Egyptians and the Greeks.
Typical for all variants of the Game of the Goose is that there are 64 fields and
two or more players. As the game is solely determined by throwing a pair of dice,
no strategy is involved. Hence, the probability that each player wins the game is
completely determined. An obvious question is whether one can determine these
probabilities, and this is the main issue of this article.
Unlike probabilistic games, the question of determining the winner in a strategy
game has always attracted a lot of attention. Although it is not (yet) resolved which

* Received by the editors August 25, 2014; accepted for publication (in revised form) April 1, 2015;
published electronically February 4, 2016.
[Link]
^Department of Computer Science, Eindhoven University of Technology, 5600 MB Eindhoven,
The Netherlands ([Link]@[Link], [Link]@[Link]).
^Faculty of Science, University of Nijmegen, 6500 GL Nijmegen, The Netherlands (freek@[Link]).
143

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
144 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

\sz

Is/

%> a
>2E
Cs
38 39 40

IBii

Fig. I.I A Dutch version of the Game of the Goose from appr. 1960. (Copyright of the "Game of
the Goose" image is owned by Koninklijke Jumbo B. V., Zaandam, The Netherlands, all
rights reserved.)

player has a winning strategy in the great games chess and go, there are many strategy
games for which the winner is known, such as four-in-a-row (also called connect
four) [11] and restricted versions of awari [7]. With checkers the situation is mixed.
American checkers, also called British draughts, played on an 8 x 8 board has been
solved [8]. International checkers (10 x 10) and Canadian checkers (12 x 12) are still
open problems.
So, we set out to determine the winning probabilities for each player of the "Old
Dutch" variant of the Game of the Goose (het oud-hollands ganzenbord). Unfortu
nately, even for this game there are different sets of rules and so we simply made an
arbitrary choice among them. We ignore the payment of small fines ("fiches") that
occur in some variants. The precise rules that we use are described in section 2, a
typical layout of the playing board is shown in Figure 1.1.
The most straightforward way to estimate the winning probabilities is by simu
lating the game randomly a huge number of times and counting how often each player
wins. Doing this, we observed that the convergence of winning probabilities is slow:
Simulating the game for 108 times typically gave precisions of only three to four dig
its. However, an advantage of such simulation is that the number of players is not a
practical limit.
In this paper we also calculate the probabilities by generating the complete state
space of the game. Each legitimate way to put the players on the board together with
the information about who has the next turn constitutes a state. For each state and
each player we introduce a variable expressing the probability of winning the gam
for that player in that state. By formulating the relations between these probabilit
we obtain n linear equations among n variables, where n is the size of the state spa
The number of states grows exponentially with the number of players. For the ga
with two players, n is around 4000, and for five players there are 885 million stat
Using Mathematica [4] we can solve the two-player game exactly as a fraction
For the three player game this is no longer possible, so instead we solve it using

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE 145

numeric approximation. The four player game can be solved using MATLAB [5] w
the induced dimension reduction (IDR) method as an extension package [10].
managed to solve the game for five players with an ad-hoc fixed point algorith
Establishing the winning probabilities for six or more players in this way is currently
beyond our reach.
Inspecting these probabilities confirms many intuitive ideas about the game, b
also leads to several surprising observations. The most surprising is that for the g
with two players there is a substantial probability (23%) of the game ending in
draw, where one player is in the prison and the other is in the well. For all numb
of players investigated the first player has a small but definitive bias to win the game
which increases when more players join it. For two players, player one has a less t
4% higher probability of winning the game compared to the second player. For t
players, the first player has a 66% higher probability of winning the game than
tenth player.
It is also interesting to observe how the positions of the players on the boa
influence the winning probability. For two players we provide 3-D diagrams show
how these probabilities fluctuate depending on the relative positions of the play
on the board. From this, one can, for instance, make the rather counterintuiti
observation that if one of the players ends in the well, this only narrowly incre
the probability of the other player winning.
It is of course good fun to determine the winning probabilities for the Game
the Goose, but there is a more serious side to such an endeavor. Rather different
life phenomena such as processor scheduling, weather forecasting, human interact
and quantum mechanics can be viewed as probabilistic games. If we can analyze la
games, we will have the techniques to analyze models of such real life phenomen
well. There are plenty of fully random games that can serve as a useful playgrou
and if all such games have been investigated, there are still the games that comb
strategy and randomness to be explored.

2. The Rules. Before presenting the results we have to determine the precise r
of the Game of the Goose. As it is a traditional game, it has been described by sev
sources. Although most ingredients of the game are the same in all sources, there
some differences. Here we fix a version that covers all the interesting ingredient
have encountered and that is as close as possible to the various sources considered
• There are 64 positions, numbered from 0 to 63. All players start on positi
0. The first player that reaches position 63 wins.
• A step of a player consists of throwing two six-sided dice and moving forward
as many steps as there are spots shown.
• If the resulting position is already occupied by another player, the player
to go back to its original position.
• A player only wins when it lands on position 63 exactly; if the number thrown
is higher than required, the surplus is counted backwards from position 63
• At positions 5, 9, 14, 18, 23, 27, 32, 36, 41, 45, 50, 54, and 59 there is a go
(see Figure 1.1). If a player arrives at such a position, it will move the sam
count of the roll again in the same direction. If this is again a position wi
a goose, this will be repeated.
• If from position 0 the dice show 3 and 6, the resulting position is 53. If fr
position 0 the dice show 4 and 5, the resulting position is 26. Note that if this
rule did not exist, the player would jump immediately to the finish via t
positions marked with a goose when throwing 9 spots in the initial positio

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
146 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

Position 6 is the bridge: a player arriving there will jump to position 12.
Position 19 is the inn: a player who is in the inn will stay there for one extra
turn.

Position 31 is the well: a player in the well does not take part in the game
until another player arrives in the well, in which case the former player is free
to move at the next turn and the latter player remains in the well until it is
similarly released.
Position 42 is the maze: a player arriving there will go back to position 30.
Position 52 is the prison: just as with the well, a player in prison has to
postpone participating in the game until another player ends up in prison,
releasing the first player.
Position 58 is death: a player arriving there has to start anew by going back
to position 0.
These rules are applied repeatedly. For instance, if a player is at position 46 and
throws 4, it will move to position 50. This is a goose, so it moves to position 54. This
is again a goose, so it moves to position 58. This is death, so it has to start over
by going back to position 0. This combination of throws is considered as one move.
If position 0 is occupied since the turn before another player landed on death, this
player will stay at position 46. As another example, consider a player at position 60
throwing double 6, yielding 12 in total. By counting back it arrives at position 54.
Since this is a goose, it has to count back 12 more positions, arriving at position 42.
Since this is the maze, it goes to position 30.
There is no upper bound on the number of moves and in principle the game can
go on indefinitely. However, in the Game of Goose there is a positive probability that
the game will end in a finite number of steps in every state. It can be shown that as
there is a finite number of states, this implies that the probability that the game goes
on forever equals zero.
If there are two players, there are three possible outcomes: either player can win by
arriving at position 63, but the game can also end in a draw if one player is in the prison
and the other is in the well, since then no player can move. Note that when there are
more than two players, one of the players will win and there is no possibility of a draw.
Although we have carefully described the rules of the game, experience shows
that textual descriptions are virtually always incomplete and can be interpreted in
multiple ways. A typical example with three players is the following. Player 1 is
before the prison, player 2 is in the prison, and player 3 has passed the prison. On its
turn, player 1 goes to prison, freeing player 2. Player 2 jumps to the position of player
3 and has to go back to prison. The question is whether player 1 or player 2 is now
free. The listed rules above do not deal with this situation. The formal description
below says that player 1 is free. A similar question occurs with the inn. If a player
that leaves the inn is forced back into the inn, as it jumps to an occupied place, does
it have to rest again one extra turn or not? (Answer: no.)
To make our rules absolutely clear, we provide a formal description of the game
in Figure 2.1 using the process algebraic language mCRL2 [2], This is a language that
mathematically describes and analyzes the behavior of systems that send messages
to each other. It is also quite suitable for describing games. Such languages stem
from a tradition arising in [6], where it was observed that classical mathematics is not
suitable for describing and studying digital communication among computers. The
process algebra mCRL2 describes, in essence, in which sequences actions can happen.
In this case, the actions are win(p), rest(p), and throw(p,<i,<2)) where p is the
player and t\ and I2 are the numbers of spots on each dice.

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE 147

eqn JV = 4;
next(p) = iftpzeN, l,p+l);
previous(p) = i/(pssl, AT,p— 1);
initiaLpositions(p) = 0;
adapt-after-63(n) = if(n> 63,126—n,n);

next-position(p, position, t\,t2) =

if(position(p)~0 A (tiRs4At2~5 V ii?s5Ai2~4), if(occ-twice(p, position\p—>26]), 0, 5
if(position(p)zz0 A (ti~3At2~6 V tif«6At2«3), if(occ-twice(p, position\p—>-53]), 0, 26
next-position2(p, position\p^adapt-after-63(position(p)+ti +t2)],
if(position(p)+ti+t2>63, —ti—t2,ti+t2),position(p))))-,

next-position2 (p, position, throw, old-position) =

i/(posifcorc(p)e{5, 9,14,18, 23, 27, 32, 36,41,45, 50,54, 59},
next-position2{p, position\p^fadapt-after-63(position(p)+t)],
if(position(p)+t>63, —t,t), old-position),
if(position(p)xi6, next-position2(p, position\p—>12], t, old-position),
if{position{p)~19, nextjposition2(j>, position\p—>64], t, old-position),
if(position(p)Ri42, next-position2(p, position^-)30], t, old-position),
if(position(p)~o8, next-position2 (p, position[p—}0],t, old-position),
if(position(p)£{31, 52} A occ_tu;ice(p, position), old-position, position{p)))))))-,

0CC-twice(p, position) — occ-twice-rec(p, position, 1);

occ-twice-rec(p, position, other) = if( ot,her<N, occ-twice-rec(p, position, other-p 1), fals
{other^kp A
(position(p)~position(other) V
position(other)zsW/\position(jp)ze%4 V
position(other)&64/\position(p) ~ 19));

proc PL.4 l^piN"'", positioraiN^"—>N) =

(position(previous(p)) «63) —►win (previous(p)) ■ So
(position(p)£{31,52} A ->occ-twice(p, position))
—>rest (p) • PP/1 Y{next{p), position)o
(position(p)!»fi4)^>-rest(p)-PLAY(next(p),position[p—^19])o
±tl,t2M+ (*iA t2<6)->throw(p, ti,t2)
PLAY{next(p),position[p^next-position{p, position, t\, t2)])',

init PLAY{1, initial-positions)-,

Fig. 2.1 An MCRL2 description of the Game of the Goose.

Players are represented by positive numbers (N+). The numbers 1..... Ar repre
sent the actual players, where N is the total number of players. As indicated above,
we use the letter p to represent a player. The functions next and previous provide the
next and previous players in a round robin fashion. The fields on the board are given
as natural numbers (N) ranging from 0 (initial state) to number 64. Field 63 is the
winning field. The field with number 64 is used for the inn. A player that arrives in
the inn moves to position 64. When the player is at position 64, it moves to position
19 during its next turn, which represents waiting for one turn in the inn. The position
of the players is represented by a function position:N+—>N that maps a player to its
position on the board. The function initiaLpositions indicates the initial position on
the board for each player and is defined in the eqn section as initiaLpositions(p) = 0,
i.e., each player starts at position 0.
The process PLAY indicates the sequence in which the actions take place. It
has two arguments. The first argument indicates the player that plays next, and the
second argument is a function that for each player gives its position on the board.

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
148 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

The behavior of PLAYis given on its right-hand side by if-then-else rules, denoted as
b—>xoy, indicating that if condition b holds, process x must be executed, otherwise y
is executed.

The first line, starting with the condition position(previous(p))~63, says that i
the position of the previous player is 63, this previous player has won the game,
indicated by the action win(previous(p)), and after that nothing is done, indicated
by the deadlock <5.
If this first condition does not hold, the second line applies, which starts with
the condition position(p)G{31,52} A • • •. It says that if the position of the current
player p is 31 (the well) or 52 (the prison) and there is no other player in the well o
prison (->occ-twice(p, position)), then the player must wait one turn by carrying ou
the action rest(p) and continuing to play the game giving the turn to the next playe
and leaving the positions of all players unchanged {PLAY{next{p), position)).
If the second condition also does not hold, the third condition position{p)^64 can
apply. This says that the player p is waiting in the inn. It automatically moves to
position 19. This is denoted using the function update construction. The expressio
position[p—>19] represents the function position, except that it maps the argument
to position 19.
If none of the three cases above applies, the player p throws two dice. This
is represented using the sum operator J2tl t2.f:J+ (^i <6A<2 <6) —^ ■ • •, which says that
positive values for t\ and f2 are selected that satisfy the condition. Then, the actio
throw(p, fi,f2) happens representing that player p throws one die with result t\ an
one with result t2. Subsequently, the game continues as the next player takes a turn
and the position of the current player is updated using the rather complex functio
[Link]{p,position, fi,f2). This function is defined in the eqn section and is
explained below.
The function nexLposition{p, position, t\, t2) calculates the next position of playe
p upon throwing t\ +12 spots, given that the current position of the players is give
by position. If p is at the initial position and a 5 and 4 or a 6 and 3 are thrown,
player p moves to position 53, respectively, 26, unless there is already a player oc
cupying this field. In the latter case, the player stays at position 0. The expression
occ-twice{p, position\p—>53]) is used to check whether if player p moves to position 53,
the position of player p is already occupied. Note that in the definition of occJ-wice
there is an extra check for whether positions 19 and 64 are both occupied. As both
positions represent the inn, this also counts as a single field having double occupancy
If the special initial case described above does not apply in the definition of the
function calculating the next position [Link](p,position,t\,t2), its behavior
defined by the auxiliary function nexLposition2(p, position, throw, old-position). Th
yields the ultimate position of player p on the board when player p made an initi
move (which is already reflected in position) where the dice showed the value throw
(this value is negative if p is moving backwards), and oULposition is the position from
which p came. Note that the use of next-position2 is complicated, because the player
must move backwards when overshooting field 63.
In the definition of nexLposition2, all remaining special rules of the game are
dealt with. The first if deals with the 13 positions where the player can move the
same number of moves ahead (or backwards). The second condition (position{p) »6)
deals with the situation where the player is at the bridge and must move to positio
12. The third condition position{p) « 19 represents the player entering the inn. It is
moved to the "resting room" at position 64. The fourth condition positionfp) ~ 42
indicates that the player is in the maze. It must move to position 30. The fift

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE 149

condition position(p) « 58 corresponds to the situation where the player dies. It

restart by moving to position 0. The last condition applies when no subsequent
of the player is possible. It is checked whether the move of the player will lea
double occupancy of fields (except for the well and the prison at positions 31 an
which can have more than one occupant). If there is a double occupancy the p
moves to its old position, otherwise it moves to the new position.
The mCRL2 tools allow us to simulate the game and generate for it a full
space which consists of all reachable configurations of players on the board.
interpreting the throw actions as being able to happen with probability the
winning probabilities can be calculated by interpreting the state space as a discrete
Markov chain. If a win or rest action can happen, no other actions are possible and,
therefore, one can consider these actions as happening with probability 1.

3. Analysis of a Simple Game. In this section we introduce an extremely sim

plified version of the Game of the Goose in order to illustrate how we obtain the
probabilities. Two players start at position 0. The first player that reaches position
2 wins the game. Each player throws a two-sided coin with no or one dot and will
move zero or one position forward in accordance with the value thrown. The game is
illustrated in Figure 3.1.

0 1 2
start finish

Fig. 3.1 A simple two-player game.

The game can be described in mCRL2 as follows:

proc PLAY(pi,p2'.N, turn:N+) =

(pi«2)—>win(l)-5o
(p2~2)—>win(2)-<k>
((torn«l)-KSt:N -(t<2)-^thraw(l,tyPLAY(pi+t,p2,2))
0 (Et:N ■it<2)->thraw(2,t)-PLAY{p1,p2+t, 1)));

init PLAY(0,0,1);

The state space of the game is depicted in Figure 3.2. The initial state is light
gray and has number 0. In the initial state player 1 either throws zero (throw(l,0))
or one (throw(l, 1)), which are represented by arrows leading to, respectively, states
1 or 2. In states 1 and 2 the second player can make a move. Figure 3.2 shows how
the game proceeds. At states 8 and 9, player 1 wins the game (win(l)) and in states
10 and 11, player 2 wins (win(2)). State 12 is a deadlock state where the game is
finished, corresponding to ô in the mCRL2 description.
We are now interested in the probability of player 1 winning the game when it is
in state i. We denote this probability by pi. Clearly, ps — pg = 1 and pio — p\\ — 0.
The probability pu makes no sense because in state 12 the game is finished. For
all other probabilities pt we can derive a simple linear equation. The probability of
winning in state 0 is \p\ + |p2 because player 1 has 50% chance of ending up in state
1 and 50% chance of ending up in state 2. Writing out all equations leads to the

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 50 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

throw(l, 0)

[0)3?
throw(2,0)

throw(l,l)/ x\throw(2, 1)

throw(l,0)y / \ / \ \throw(l,0)
ir0w(2,O) \throw(2, 1) throw(l,l)/
throw(l, 0)

^throw(l,l) j throw(2,0) ^ thr

throw(l,l)/ \throw(2, 1)
7 11

9 ) ( 10 '
dn(l) win(

win(l)^ "win(2)

Fig. 3.2 The state space of the simple game.

following set of linear equalities:

Po = \p\ + \P2, Pa = \P2 + \p&, P8 = 1,

Pi = \po + \P3, P5 = \p& + \P9, P9 = 1,
P2 = 5^4 + 5P5, P6 = |P5 + |piO, PlO = 0,
P3 = ^P6+^P7, P7 = |P3 + |P11, Pll = 0.
This set of linear equations is small and therefore easily solved, leading to po =
16 ~ 0.59. In order to find the probability that player 2 wins the game we can us
27

same set of equations, except that we must take ps = 0, p$ = 0, pio = 1, and pio = 1.
This leads to the expected result that is the winning probability for player 2. As
> il; the first player has a higher probability of winning the game.
With a small change of the equations we can use them to calculate the average
duration of the game. In every equation that corresponds to a turn of a player, 1
is added to the right-hand side. In every equation that represents winning or losing
the game the right-hand side is set to 0. It follows that the average game duration is
-y « 5.33 moves.
We can also derive the linear equations directly from the game, without generating
an explicit state space. We define the probability q-ijk to represent the probability that
player 1 wins the game, provided player i (ie{ 1, 2}) has the next turn, player 1 is at
position j, and player 2 is at position k (j,k < 2). The probability qioo is equal to

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE I 5 I

59200 +\ <i2 j o because player 1 has equal probability of staying at position 0 or mov
to position 1, after which it is player 2's turn. By analyzing all board position
the game one can derive the following set of equations. Note that probabilities
unreachable states, such as prifc. are omitted. These probabilities represent situat
where player 1 can play and has won. Note that the equalities obtained are in th
case exactly those obtained via the state space:

9100 = §9200 + 59210, 9200 = 59100 + 59101, 9no = 59210 + 59220,

9112 = 0, 9210 — 59110 + 59111, 9220 = f,
9101 = 59201 + 59211, 9in = 59211 + 59221, 9201 = 59101 + 59102,
9211 = 59111 + 59112, 9221 = 1, 9102 = 0.
An alternative approach is random simulation of the game wh
how often each player wins. The question asked is how often mu
simulation to obtain a sufficient accuracy. Call the probability of
for some player p, the number of simulation runs n, and the num
are won m. The observed probability of winning the game is then
As the variance of a single coin flip with probability p is p(l —
simulating n games is np( 1 —p). The standard deviation of the va
runs of n simulations will be y/np( 1 — p). This means that the distri
to a normal distribution with mean value p and standard deviation
For this reason, in 99% of the simulation runs the value of p wil
[p- 2.58a,p + 2.58<r].
Now we can also use this in the other direction. If we find a cer
99% of the cases the true value of p will be in the interval [p — 2.58<
<7 := y/p{ 1 — p)/n. This means that n > 0.645 • 102fc if we want
precision, where we do not allow all digits to be correct in less th
We have used all three ways to establish the winning probabilit
two players and the first and last for more than two players. Th
that it is easy to make a mistake in modeling even simple games.
in different ways we can compare the results, remove modeling mista
our confidence that the final results are correct.

4. Computations for Two Players. If we analyze the Game of the Goose for two
players, we obtain a set of 4145 linear equations. We can derive the following results
regarding the winning probability and average game duration for each player. The
game duration is defined as the number of turns where the dice are thrown. If a player
is in jail or in the well, or resting a second turn in the inn, this is not counted as a
turn. All numbers are rounded in the ordinary way.
Probability that player 1 wins the game 0.3936
Probability that player 2 wins the game 0.3799
Probability of a draw 0.2265
Average game duration 29.0651
For a two-player game it is possible using Mathematica to obtain th
of the set of equations. As a curiosity, the precise winning prob
Figure 4.1 as a quotient of two natural numbers, which is approxi
0.3936251373937573914028403448768445020070441350696.

This exact solution can be used to check whether or not exactly the same gam
been modeled. Minor flaws in modeling the game, such as forgetting that a player m

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
I 52 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

354578138124950186817810489217182791317787798724802614410029328485608903369477431635018222558
399309450507103459074238228376405712500904558023752239339280938603803457573412946045548555428
192408470799948053129350811785597975235328183601609649319429499349866185498239751673351341769
582839004783122090660713501153416711577247932684590810429682087674204514516206428023232773142
970683832903046790844706695907372586643039096740098702406743747498676783882062602493264926223
838302824908562507570250392570595827698114873011234704695646697896143603966602499097872351154
669089452354191599916709746676851938051878405719576742617594702275857493698645246959330233303
091445902939904912863840744790654661105175111195454449663058266233638363673073343450482998878
224889581332206333579009718721098139784479036279112434678742992137019681147035075938046701550
918055159448125308748921320289899333197006467347143802920860473826874931173910161985261833871
891348234745481157765616175520794275559824385481401595657510498531296821490135004997149723846
312298317137414865283481215837836732118455291175984574489259882269507968728081889027165529711
660915404995631330437820160918345491253783805093745020459199567996953546847609820992902762908
750628689522634308453656275384033760133459102330761797545360967662642992417761488637785343096
179837783567510615016827169189382638260325730601238974784021870275751361465251485833641380988
601239137220474152073594938576999714048510670106244854187988128408120363500526256376389738531
527536713654289142088651349218285889196358082646355336459464707565595257453153280748427467824
804125023919836067186411643984746979465748585196447014561870836245035624955049440432350263977
824403133461025845332976436704285089656500992657870174243661788423998434592130682667322269405
049490116752141370798765615956317551142249979497567046069585441301567890743987803069586801932
669973979851519762918559922723993392470947267582828358727134952210072937955951818226524755153
725945829745811214804332328703353196315111784798277410263357528661496628912933728773218555977
962282140759639089508990171920383431393394642405434973415711204469313507363576994397787067207
425052329110883835493651467371006199576012293462207442967865421344357444321357800987822349475
376878156929670045445507371744044527718465666904104107234441730939460632072987861228015467811
882360487748332166057469542836916432762990864234404289049048042068157866859460085236085608315
623668882716193909184091442443597083669348518632793885704717828545288254527572206829503888843
189830953755201841220373817976349249436975535253449029794712313365966457380552647526110163477
309530899123504784001358147399949405347287670149449824157437381968509825640543358543867976582
112932024845249858864855940315369112486827116397833485123121716079117281508469354925885409764
674924868096271749236366074963266182936247416433242594731524706999193767511101871330792678259
280382594971805430937807818516642167895701603935105753261288618047393431315053008943441431152
112314766320450756187900149495780206679762576019422616321340814720511095615601551532129272718
542739705072677966046511469321938635400662566613122323344475936596007879096566202575850204524
189676551943771629765498166351584803390675990234443292078393699164128871377784023956250048044
331552068284643624644046415414958009203064443229091249119093243635766772160882604554315879605
060831221307872086909952800271602369875618608782388082232510649386131941136637973880915801002
154572511895832800230367533198381720968690002760891453868273714176187898521616789993779960852
761931148923821002490024262343407522301166428721085616227115222756065531414838764999639026632
039441391801485622144284338102119223432872300849269879911935735471566830470688649086825183967
844807850946421646147394307730886861802798090936151774215069265819265138777116887920005066844
010830068931878504072308068647265799544023019246293329832623031453567259134862222156900665676
786448672835928798628020069958939032396484550567944379409480283

900801560775964656126774493326889579738207808673590850978148447654100516265185014530331647892
963003549157974321305021253246292417449947288406287943081249483047759028012482686458088710035
667076237996138211920100304931630064917887932900820281965696801766239831673398209178450364993
664141772270799113017382685324511970805425293800448269497362017976312673299247677364939829441
356682373589121359929188239605113065014562579196471221602072740197329678855003920081111125071
562181004087297676021601547101904445991031881920268404465974551250191245335606564444955874790
579668842903908691859663445026042171718496802772890001540462085862285203596824508136312510930
848256782907319085401577950566480236492258368651403851498168474838648238887253361826278320974
211582989769501836744362197433090321629992797581960180227991741382084013179433356134386879465
593763466375933164532378142992032030676788266593780697459863590954425340611505393883463891170
971971064338366560257877714490386750914477743904362410695314284442222150192349474163351136665
216496309031156352596603819405940309695923665530872630461144194903750285125269104079009482059
449373800161294244270975727298452311247329735693384663705652207467764860758084365569131209291
229381384664939466470853167529986572849474795951706727760432071740948785034512160370763120637
339519772942758900998043467561936589952119274344302734511358915860188099372260220328866589088
156263594454570824879678437060954154474559388781342874179426124184682655203719929399964083174
163103613617183832005419973105753767668901642442011068968931764251547938499039025283281070521
626469361626355506807630054195079424417327991706187186940870807368776939235542166982376805364
554504077292848886835539066227774214224562642653978516302884675382241286557954761373473131260
398683754349784037651478869577494891197647440888866724744290474747707116650727026018201161485
468721477586155819906563951435363663409456413316958670299530968871213165410259673299779136449
546570382066086147033604221435389641473231140715280626878852774956863960299786109914398931518
303131212483531294493331159638610052553627850621091892922181967442078061555180719269959611691
375488038095571656477654073523484609997634483232600502379658428980870346360321738192432670388
030086044391330549464290582334683133818870134666528351268436864593343373928316778450239320151
490616017571218225599667338674345549599369213030835046237448574463407637323433037344642693651
701648988954227861867925754021764324627782273822196813395500498990947900775262466674566797502
538208692635181129126570140804676875824884483807381185441791415008023307041497487737249042310
505372990270640822151972277764575305052107974872293504455660045611058486209180538423771137278
665443687822206785250296004711649547019031985510776338368259491949746485804668295542587032964
330180313351672274528115527876257178151725307206824846802270199905183486827906401182641218870
282587165253232412527268173813719376856485718148441478527543006201364921512289327167966829752
840415669558102361080423578070292648605539487934572107269958992134149955478331466018851569593
718356612338981495161662932308349814571401207328017569896080483808315461551874907047767528969
200021893878774750118480746745401602759010790621652738798780702998291443204275453806422937174
486042713846526300307683386755779988859765442464426523581101010235589403184706181157183193989
204685507819243217476508781703035183919970452946772074659104050764081839159268568343117208670
821350611437447602059630356954342172567631055484692745718321278817807410839504476278638075658
957815114791859980138843196023327044410709069485441809674404861155840974443329069685443563451
000330064673984317988986377130965450283592011509535800247907476188834267070613796557088503934
717084645791126874032339497526018300237585065551096520626437434582945009801773336358068326941
285478875469093008017253189675909895843648482906710452459750563144869260744545121189657756675
3191653418733360283585073225627136513606612524945418565535662080

Fig. 4.1 The exact winning probability for player 1 in a two-play

rest one turn in the inn, will not substantially influence the w
the players, but it will also not lead to exactly the same solution
It is interesting to figure out how the winning probabilitie
is progressing. For a two-player game this can be neatly visu
Here three diagrams are given, each with a view from above an

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE I 53

Fig. 4.2 Visualizations of the winning probabilities of a two-player game.

upper diagram models the probability that player 1 will win the game, when
turn to make a move. In particular, if player 1 is alone in the well or in the
it cannot move, and this situation is not part of this diagram. The second
depicts the probability of player 1 winning the game, when player 2 is about
The third diagram depicts the probability of ending up in a draw.
If player 1 moves forward, it moves to the back of the diagram. If player
forward, it moves to the right. There are only 47 positions in the diagram, b
fields where a player must continue to move forward (i.e., a goose, death, the
or the maze) have been removed.
Note that the upper two diagrams have a solid wall at the back. This corr
to player 1 winning the game. Similarly, there is a valley at the right, corres
to player 2 winning the game, which means that the probability of player 1
the game is 0. Observe that the closer player 1 is to the finish, the higher

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
154 JAN FRISO GROOTE, FREEK WIEDIJK, AND HANS ZANTEMA

probability of winning, and conversely, the closer player 2 is to the finish, the lower is
the probability that player 1 will win. Remarkably, if player 2 is in the prison, there is
a substantially higher probability that player 1 will win, but if player 2 is in the well,
this only narrowly influences the probability that player 1 will win. The reason for
this can be seen in the third diagram. If player 2 is in the well, there is a substantially
increased probability that the game will end in a draw. The two spikes in the third
diagram correspond to the situation where the game is actually in a draw.
There is much more detailed information in these diagrams. For instance, it is
not advantageous to be too close to the finish. We leave the detailed interpretation
of these features to the reader.

5. Winning Probabilities for More Players. It is also possible to establish the

winning probabilities when there are more than two players, but this becomes in
creasingly more difficult as the number of states grows exponentially, approximately
according to the formula NcN, where c « 45 and N is the number of players. In Table
5.1 the winning probabilities and the average game durations are provided. A game
duration is defined as the number of moves where dice are thrown. Turns that are
skipped are not counted. The numbers for more than five players were obtained
simulation only. To show that simulation is capable of solving games independen
of the number of players, we have added the results for a ten player game as w
where we list the winning probabilities for all ten players.

Table 5.1 Winning probabilities when there are two or more players.

Average
#players ^equations Player 1 Player 2 Player 3 Player 4 Player 5 duration
2 4145 0.39363 0.37999 - - -
29.07
3 2.79 105 0.34596 0.33290 0.32114 - -
39.42
4 1.64 107 0.26695 0.25471 0.24408 0.23426 -
44.10
5 8.85 108 0.22039 0.20875 0.19894 0.19012 0.18181 49.10

6 -
0.18986 0.17865 0.16944 0.16135 0.15390 54.35
10 -
0.12995 0.11992 0.11213 0.10568 0.10009 76.23
0.09508 0.09049 0.08622 0.08218 0.07827

We first solved the sets of linear equations using Mathematica [

done exactly for two players and numerically for three players. For th
players, using MATLAB extended with the IDR package, we could also
of obtained linear equations [5, 10]. For four players 400GB of memory
However, we observed that the generated equations have a rather r
ture. For each variable pt there is an equation of the shape

Pi — CilPil H~ * * * H~ CikPiki

where all Cij are positive numbers smaller than or equal to 1 and the s
all variables are in the interval [0,1]. This linear set of equations can b
monotonie operator for which the solution can be obtained using fixed p
Initially, all pi are set to 1. Taking the equations as assignments, a new
Pi is repeatedly calculated until a fixed point is reached. This allowed u
winning probabilities in a five player game.
The numbers in the table were obtained in different ways and on di
puters. The numbers for five players were obtained by generating a sta
8.85 108 nodes and 3.11-1010 transitions using the mCRL2 tools [2]. This

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]
 A PROBABILISTIC ANALYSIS OF THE GAME OF THE GOOSE I 55

1.5 months and 300GB of memory on a Linux machine with Intel Xeon E5520 pr
cessors at 2.27GHz with 1TB of main memory. Solving the equations derived fro
the state space required approximately a week and 200GB of memory.
As it stands, solving the game using equations for six players is currently b
yond our capabilities, although it is conceivable that with a concerted effort, capa
hardware, and dedicated software this could be achieved. The number of require
equations is estimated to be around 5.0 ■ 1010.
The simulation results were obtained on a Linux machine with Intel E5-2670
processors at 2.6GHz. The simulation used ten processors simultaneously and req
approximately a week to obtain the accuracy reported in this article.

Acknowledgments. We thank Michiel Höchstenbach for his fruitful rema

especially for pointing out the IDR extension package for MATLAB to solve l
systems of linear equations. A shortened Dutch version of this article appeared

REFERENCES

[1] H. C. Bolton, The Game of Goose, J. Amer. Folklore, 8 (1985), pp. 145-150.
[2] J. F. Groote and M. R. Mousavi, Modeling and Analysis of Communicating Systems, The
MIT Press, Boston, MA, 2014.
[3] J. F. Groote and H. Zantema, De kans om Ganzenbord te Winnen, Nieuw archief voor de
wiskunde, 5/15 (2014), pp. 234-239.
[4] Mathematica, Version 8.0, Wolfram Research, Inc., Champaign, IL, 2010.
[5] MATLAB version 7.10.0 (R2010a), The MathWorks Inc., Natick, MA, 2010.
[6] R. MlLNER, A Calculus for Communicating Systems, Lecture Notes in Comput. Sei. 92,
Springer-Verlag, New York, 1980.
[7] J. W. Romein and H. E. Bal, Solving the Game of Awari using parallel retrograde analysis,
IEEE Computer, 38 (2003), pp. 26-33.
[8] J. SCHAEFFER, N. BURCH, Y. BjÖRNSSON, A. KlSHIMOTO, M. MÜLLER, R. LAKE, P. Lü, AND
S. Sutphen, Checkers is solved, Science, 317 (2007), pp. 1518-1522.
[9] A. H. Seville, Tradition and variation in the Game of Goose, in Board Games in Academia
III (Proceedings of Colloquium in Florence), 1999, pp. 163-174.
[10] P. Sonneveld AND M. B. VAN Gijzen, IDR(s): A family of simple and fast algorithms for
solving large nonsymmetric systems of linear equations, SIAM J. Sei. Comput., 31 (2008),
pp. 1035-1062.
[11] J. W. H. M. UlTERWIJK, H. J. van DEN Herik, AND L. V. Allis, A knowledge-based approach
to Connect-four. The game is solved!, in Heuristic Programming in Artificial Intelligence:
The First Computer Olympiad, D. N. L. Levy and D. F. Beal, eds., Ellis Horwood Limited,
Chichester, UK, 1989, pp. 113-133.

This content downloaded from [Link] on Thu, 27 Nov 2025 [Link] UTC
All use subject to [Link]