venerdì 2 novembre 2018

PERDITE AL LOTTO - UN APPROCCIO MATEMATICO

Losing Lotto, A Mathematical Approach

Abstract
An (  n,  k,  p,  t)  lotto design  is  a set of  k  sets (called blocks) of an  n  set such
that  any  p  set intersects  at  least one block in  at  least  t  points.  We  will
denote  the  minimal number of blocks needed  to  make an  (n,  k,p,  t)  lotto
design by  L(n,  k,  p,  t).  This paper lists a  few  known theorems for upper and
lower bounds for lotto designs.  We  then  apply these theorems to  the  New
Zealand lotto system  and  calculate  upper  and  lower bounds for each of  the
six divisions of  the  New Zealand system.

Lotteries are a common form of gambling. Generally they work by bettors
choosing tickets with  k  numbers on  them  between 1  and  n.  The  house  then
draws  p  random numbers between 1  and  n.  If  a  bettor  has  t  or more numbers
that  are drawn on their ticket,  then they  win a prize. Usually for  the  lower
values of  t,  the  bettor  wins a fixed amount of money. While for  the  larger
values of  t  the  bettor  wins a share of  the  prize, depending on  the  number of
other people who had  the  same number of numbers drawn on their tickets.
The  prize for having all  k  of your numbers being drawn  is  almost always
a large sum of money. Because of this, everybody wants to win. There are
numerous systems  out  there  all claiming to increase your chances  at  winning.
Some of these systems are based on  lotto designs.  The  focus of this paper
will be on lotto designs,  and  how they  can/can  not  help us win  the  lottery.
Chapter  2 will cover  the  basic definitions and ideas of lotto designs. This
will include a number of theorems  and  their proofs. In  the  second  part  of
chapter 2 will look  at  a  few  algorithms used to construct lotto designs.
The  most common lottery in New Zealand  is  Lotto,  run  by  the  New Zealand
Lotteries Commission. In chapter 3  we  will be trying to apply  the  theorems
and constructions in  chapter  2 to  the  NZ  lotto system. Our main focus will
be trying  to  construct  upper  and lower bounds for  the  number of tickets
needed  to  guarantee winning each of  the  six divisions of Lotto.
2
CHAPTER  2
Lotto,  and  other  Designs
2.1  Lotto  Designs
To begin with  we  will need a few definitions. We will begin with some basic
combinatorics definitions, then move onto definitions used in making lotto
designs.
Note:  throughout  this  paper  we  will  be  referring  top  sets,  k  sets, etc. These
are  just  sets with  p  or  k  points in  them  respectively.
Definition  2.1.  A  combinatorial design  is  a pair  (P,  B),  where  B  is  a non-
empty  set  of  blocks  made  up  of elements called  points  from a set  P.
Definition  2.2.  At- (n, k,  A)  design  is  a combinatorial design such  that  P
has  n  points in it, every block in  B  has exactly  k  points,  and  such  that  every
t  element subset of  P  is  contained in exactly  A  blocks.
The  invitation  problem  is  a common combinatorial problem solved using com-
binatorial designs.  The  idea  is  to  invite  n  friends to dinner,  k  at  a time in
such a fashion  that  everybody has dinner with everybody else  at  least once.
Example  2.3.  The  invitation problem for inviting 7 friends to dinner, 3  at  a
time can be represented by a  2- (7, 3,  1)  design. In  the  case of this example
P  =  {1,  2,  3,  4,  5,  6,  7},  and  the  blocks of  Bare:
{1,2,3},  {1,4,5},  {1,6, 7},  {2,4,6},  {2,5, 7},  {3,4,  7},  {3,5,6}.
Definition  2.4.  An  (n,  k,p,  t)  lotto design  is  a set of blocks each consisting
of  k  points from an  n  set such  that  every  p  set intersects a block in  at  least
t points.  We  will denote  the  minimal number of blocks an (  n,  k,  p,  t)  lotto
design can have by  L(n,k,p,t).
Example 2.3  is  an example of  an  (7, 3,  2,  2)  lotto design. A special case of
lotto designs  is  when you have a  (n, k, k, k)  lotto design. In this case  the  lotto
design has  to  include every  k  subset of  the  set  n.  This makes  L(n,  k, k, k)
very easy  to  calculate.  It  is  just  ( ~ ) .
Example  2.5.  A  (12,4,6,3)  lotto design (on  the  set N  =  {1,  ..  ,  12})  is  given
by  the  following 8 blocks:
{1,2,3,4},{1,5,6,7},{2,3,4,5},{2,3,4,6},{8,9,  10,  11},{8,9,10,12},
{8,9,11,12},{8,10,11,12}.
3
CHAPTER  2.  LOTTO,  AND  OTHER  DESIGNS  4
If  we  take any random 6 set  A,  then  there is a block  B  such  that  IAn  B  I
~
3.
For example, if  A  =  {1,  4,  5,  7,  8,  10},  then  B  =  {1,  5,  6,  7}.  If  we  take  the
intersection of these two sets,  then  we  get  {1,  4,  5,  7,  8,  10}  n  {1,  5,  6,  7}  =
{1,  5,  7}.
It  is very  important  to  note  that  while  it  seems lotto designs give you  an
advantage this is not true.  They  do not improve  the  expected  return  on
each ticket. However, they do increase your odds of winning something small
instead of winning nothing  at  all.
There are two  important  subclasses of lotto designs. These are covering
designs, and  Tunin  designs. A  (n,  k,  t)  covering design is a design such  that
choosing blocks of k points from a set of  n  points, any subset of  t  points is
contained in one of those blocks. In  terms  of lotto designs, a covering design
is  an  (n, k, t, t)  lotto design. Any  t- (n, k,  1)  design is a covering design.  We
will denote  the  minimal number of blocks in a  (n,  k,  t)  covering design by
C(n,  k,  t)
A (  n,  p,  t)  Tunin  design  is  a design with blocks of  k  points chosen from a
set of  n  points where any  arbitrary  subset of p points will contain  at  least
one block of  the  design. A  Tunin  design  is  an  (  n,  t,  p,  t)  lotto design.  The
minimal number of blocks in a  Tunin  design will  be  denoted by  T(  n,  p,  t).
Both  covering  and  Turan  designs have been studied  to  a great extent as  stand
alone designs,  and  many papers have been written  about  them, see  [4].
Covering and  Turan  designs are closely related as theorem 2.6 will show us.
Theorem  2.6.  The complement  of  a Turim design is a covering design,
and the complement  of  a covering design is a Turim design. Consequently
T(n,p,  t)  =  C(n,  n- t,  n- p).
Proof.  We  will denote  the  set  {1,  ..  ,  n} by  N.  Remember  that  a 1\1ran design
is  a  L(n,  t,p,  t)  design, so any p-set will contain a  k  set  oft=  k  elements of
the  Turan  design. This can be seen in  fig  2.1  where  the  k  set  B  is  shown in
green,  the  p set  A  is  shown in purple,  and  everything else is grey.
CB c:
A
Figure  2.1:  A block of a  Turan  design
If  we  take  the  complement of these sets  we  will get something  that  looks like
fig  2.2, where  B'  =  N\A,  and  A'=  N\B.  In this figure  then- p  set  B'  is
CHAPTER  2.  LOTTO,  AND  OTHER  DESIGNS  5
shown in green, the  n - t  set  A'  is  shown in purple, and everything else is
grey.
A' C <
Figure  2.2:  The  complement of a block of a  Tunin  design.
Take any  n- p  set  B'  in  the  N,  then  there  is  a corresponding  p  set  N\  B'  =  A.
In  the  original  Tunin  design there  is  a block  B  contained in A. Since  A  ;2  B,
we  can conclude
B'  =  N\A
~
N\B  =  A'.
This works for any  n- p set  B'  you could choose, so  we  have for any  n - p set
B'  there is  an-t  block  A'  that  contains it. Therefore  we  have a  (n,  n-t,  n-p)
covering design. This shows us  that  the  complement of a  Turan  design is a
covering design and thus  T(n,p,  t)  ::;  C(n,  n - t,  n- p).  Using a similar
argument, working backwards  it  can be shown  that  T(n,p,  t)  >  C(n,  n-
t,  n- p).  Therefore  T(n,p,  t)  =  C(n,  n- t,  n- p).
D
It  is often computationally difficult  to  construct large lotto designs.  As  the
size of  n  increases,  the  size of  the  corresponding lotto design increases rapidly.
Often it is  better  to  construct  upp er  and  lower bounds for our desired lotto
designs first. This approach allows us  to  find  out  whether our lotto design
will  be  practical without calculating  the  actua l lotto design. Once  we  have
done this, if  the  design could  be  practical,  we  can  then  go  about  constructing
the  desired design. Doing this can save us a lot of time. Presented below
are a number of theorems used  to  calculate upper  and  lower bounds  for  lotto
designs. This is by no means a complete list, there are many more theorems
out  there  that  can be used  to  calculate lower and upper bounds for lotto
designs, see  [3]  pages 578-584,  [7],  [8]  and  [9].
Theorem  2. 7.  If  n  =  n 1  +  n 2 ,  and p  =  p 1  +  p 2  - 1,  then L(n,  k,p,  t)  ::;
L(n1,  k,p 1 , t)  +  L(n 2 ,  k,p 2 ,  t).
Proof.  To show  that  this is  true  we  will show  that  if you take any (n 1 ,  k,p 1 ,  t)
and  (n 2 ,  k,p 2 ,  t)  lotto designs on disjoint sets,  then  when  we  combine  them
we  have  an  (n,  k,p,  t)  lotto design. This situation can  be  shown by  fig  2.3,
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS  6
where  the  sets  PI  and  p 2  are in purple  and  the  blocks  BI  and  B 2  are in blue.
In  the  figure, either  ti  2':  t  or  t 2  2':  t  (or  both) .
Figure 2.3:
Let  N  =  {1,  ..  ,  n},  NI  =  {1,  ..  ,  ni}  and  N 2  =  {ni  +  1,  ..  ,  n}.  Also let  B  =
BI  U  8 2 ,  where  BI  is  the  set  of blocks of  an  (ni,  k,pi,  t)  lotto  design on  NI
and  8 2  is  the  sets of blocks of  an  (n 2 ,  k,p 2 ,  t)  lotto  design on N 2 .  We will
show  that  (N , B)  is  an  (n ,  k,p,  t)  lotto design on  N.  Take any  p  set
A
~
N.
Either  lA  n Nil  2':  PI,  or  lA  n N 2 1  2':  p 2 .  Otherwise we would have
This  is  a contradiction because  IAI  =  p.  If  lA  n Nil  2':  PI,  then  there is a
block  BI
~
BI  such  that  IAnBII  2':  ti.  If  IAnB 2 12':  p 2 ,  then  there  is  a block
B 2
~
B 2  such  that  lA  n  N 2 1  2':  t 2 .  Either  way  there  is a block  B
~
B  such
that  IE  n  AI  2':  t.  Therefore  we  have  an  (n ,  k,p,  t)  lotto  design.
0
Example  2.8. Say we are wanting  to  find  an  upper  bound  for L(27,  6,  8,  3).
Using theorem 2.3  we  know
L(27,  6,  8,  3)  :::;  L(15,  6,  4,  3)  +  L(12,  6,  5, 3).
From  the  tables  in  [6]  we  know  that  L(12,  6,  5,  3)  =  2,  and  L(15,  6,  4,  3)  :::;  14.
This  gives us
L(27,  6,  8,  3)  :::;  L(15,  6,  4,  3)  +  L(12,  6,  5, 3)  :::;  14  +  2  =  16.
Th
'  2 9  T(  t)  >  ( ~ ) n-p+I
eorem  . .  n,p,  _  (p-1)  ·  n-t+I.
t-1
Proof.  This  is  just  a rewording of  the  proof from  [2].  To  start  off  we  show
that:
G)  n - p  +  1  n - p  +  1 (  n )
( ~: =i ) .  n- t  +  1  =  t ( ~ : = i ) ·  t- 1 ·
This  is relatively straightforward,  and  goes as follows.
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS  7
(;)  n- p  +  1  n!/((n-t)!·t!)  n-p+1
( P- 1 ) .  n-
t  +  1
t-1
(p- 1)!/((p- t)! ·  (t- 1)!) ·  n- t  + 1
n!  ·  (p- t)! ·  (t- 1)!  n- p  + 1
(n- t)! ·  t!  ·  (p- 1)!  .  n- t  + 1
n!  (p- t)! ·  (t- 1)!
( n  t  +  1)  ! · (  t - 1)!  ·  t ·  (p  - 1)  !  · (  n - P  +
1 )
( t ~ 1 ) 1  (  )
-t- .  (p-1) . n - p  +  1
t-1
=  ~ ~ ( ~ = 1 ) 1  .  c :  1)
Let  H  be a collection  oft  subsets of the set  N  = {  1,  ..  , n}.  We  will define  mp
to be  the  number of distinct collections of all possible  t  subsets of  p  subsets
of  N  such  that  every  t  set in the collection  is  contained in  H.  That  is,  if  G  is
such a collection,  then  there exists a  p  set such  that  every possible  t  subset
of the  p  set  is  a  t  set in  H.  The collection  G  would be all the  t  subsets of the
p  set. Note  that  mt  is  just  !HI,  and  that  mt- 1  is  just  ( t ~ J Now,  theorem 1
from  [2]  states  that
p 2  .  mp  ( mp  (  t - 1)  (  n - p)  +  p)
m  +  1  >  -- - -'---.:__:_-:::------'-----
P  - (p  - t  +  1)  (p  +  1)  mp- 1  p 2  '
whenever  mp_ 1  =I- 0.  Let
_  ( ~ = i ) ( n - t + 1 ) - ( n - p + 1 ) ( n)
F(n,  p,  t) - t(p-1)  t- 1 .
t-1
It  can be shown through long and tedious,  but  simple calculations  that
(t- 1)(n- p)  +  p (  n )
F(n,p  +  1,  t)  =  F(n,p,  t) +
t
2
( ~ ) t _
1
We  will now prove by induction on  p  that
t2  ( ~ )
mp  ?:  mp-1
2  (
n ) (  mt  - F (  n,  p,  t))
p  t-1
(2.1)
(2.2)
(2.3)
when  p  =  t  the theorem does not apply because mt_ 2  =  0.  When  p  =  t  +  1,
from (2.1)  we  have
mt+I  >  t
2 mt
(mt  _ (  t - 1)  (  n - p)  +  p (  n ) )
(t+1)mt-1  t 2  t-1
t2  c ~ 1 )
mt(  ) 2 (n)(mt-F(n,t+1,t)).
t  +  1  t-1
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS  8
This  proves  the  case  p  =  t+  1.  From this we also see  that  mt+l  >  0 whenever
m  >  (t-l)(n-t)+t  (  n )
t  t2  t-1  .
Now, if  p  >  t  +  1  then  we  have
by(2.1)
induction(2.3)  >
by(2.2)
This  proves (2.3).  It  follows from (2.3)  that  mp  >  0 whenever  mt  >  F(n,p,  t).
In  other  words,  H  contains all  t  subsets  of some  p  subset  of  N  if  !HI  >
F(n,p,  t).  Let  N(t)  be  the  set of all possible  t  subsets  of  the  set  N.  The
collection of  sets  H  =  N(t\H  will have a  p  subset  which does  not  contain
any  t  subset  from  H  if  !HI  is  strictly  less  then  (7)  - F(n,p,  t).  Therefore
T  n  t  >  (n)  - F n  t  = (  n - p  +  1)  (  n )  =  J:;)_  .  n - p  +  1
( ,p,  )_  t  (  ,p,)  t ( ~ = D t  1  ( ~ = ~ ) n-t+1'
The  next  theorem  is useful for  creating  lower  bounds  for  lotto  designs.  The
lower  bounds  in  chapter  3 where calculated using  theorem  2.9  with  the  fol-
lowing theorem.
Theorem  2.10.  L(n, k,p,  t)  2::  T ( ( ~ ) , t )
Proof.  This  can  be  rewritten  as  L( n, k,  p,  t)  (;)  2::  T(  n,  p,  t).  Take a (  n, k,  p,  t)
lotto  design  with  L(n, k,p,  t)  blocks.  Any  p  set will have  at  least  t  points  in a
k  block.  If  we  take  every  k  block  and  replace  it  with  (;)  blocks corresponding
to  all different combinations of  t  points  made  up  from  the  k  points  in  that
block,  then  we will have  at  most  L(n, k,p,  t)(;)  blocks. Now take any  p  set,
then  there  is a block  in  the  original design  that  covered  at  least  t  points from
this set.  This  set  of  t  points  is  a possible combination  of  t  points  made  from
the  k  points  of  the  block. Therefore  there  is a block of  t  points  in  the  new
design  that  contains these  t  points.  That  is, those  t  points  form a block of  the
new design.  This  gives us a  (n,t,p,t)  lotto  design,  that  is  a  (n,p,t)  Tunin
design. Therefore  L(n,k,p,t);::::  r ( ( ~ r l · o
CHAPTER  2.  LOTTO,  AND  OTHER  DESIGNS
Theorem  2.11.  Let  p  =  T(t- 1)  +  1  +  v where  0::::;  v::::;  t- 1,  then
T
L(n,  k,p,  t)  ::::;  minn-v=l:[=
1  CTi  L  C((Ji, k,  t).
i=l
9
Before  we  provide a proof of this  it  would be a good idea  to.  explain a little
more  about  what  this theorem  is  saying. In its current form it is a  bit
confusing.  It  simply  states  that  given  T  suitable coverings on disjoint sets,
if  we  form  the  union of all  the  blocks of each covering design  we  get an
(n,  k,p,  t)  lotto design.  The  minimum is taken over all possible combinations
of  (Ji·  We  try  all possible combinations of  (Ji  and  find  the  one  that  gives the
minimum sum. Figure 2.4 may make things clearer,  we  can see  the  set  N
partitioned into  T  disjoint sets each with a covering on  them  plus  the  set of
v  points which is disjoint from  the  other sets.
Figure 2.4:  (n,  k,p,  t)  =  U[ = 1 ((Ji,  k,  t)
Now  that  we  understand  what  is going on,  we  can provide a proof.
Proof.  To prove this  we  need  to  show  that  if  we  have  the  union of  ((Ji,  k,  t)
coverings on disjoint sets such  that  L:=T =l  (Ji  +  v  =  n,  then  we  also have a
(n,  k,p,  t)  lotto design. Suppose  we  have  T  suitable coverings on  T  disjoint
sets Ni for 1  ::::;  i  ::::;  T  and  the  set N 0  =  {1,  ..  , v} which is disjoint to all Ni.
Let Bi  be  the  collection of blocks of covering  ((Ji,  k,  t)  on  the  set Ni. Given
any arbitrary  p  subset  A  of  the  set N  =  U[ =oNi,  we  will show  that  there is a
block E in one of  the  covering designs such  that  IE  n  A  I  2::  t.
Let  p  =  L:=T =oPi, where  A=  An  Ni  and  Pi=  I Ni  n  AI.  Now suppose there
is no block  E  in  the  union of covering designs such  that  IE  n  AI  ;=:::  t;  in other
words  Pi  ::::;  t- 1  for  all1  ::::;  i  ::::;  T.  Then  we  would have
IAI  ::::;  T(t- 1)  +  v  =  p- 1.
This is a contradiction because  IAI  =  p.  Therefore there exists  ani  such  that
Pi  2::  t,  and  consequently there is a block E
~
Bi such  that  IE  n  AI  2::  t.  This
holds  for  any  arbitrary  p set A
~
N.  Therefore  we  have  an  (n,  k,p,  t)  lotto
design.
CHAPTER  2.  LOTTO,  AND  OTHER  DESIGNS  10
This applies to any  arbitrary  set  of coverings  that  satisfy  the  theorems con-
ditions. This includes  the  smallest possible coverings  that  satisfy these con-
ditions. Therefore
T
L(n,  k,  p,  t)  ~ minn-v=L:[= 1  i=l
D
Example  2.12.  If  we  want to find an  upper  bound for L(40,  6,  14, 4), then
we  can use theorem 2.11. To do so,  we  would do  the  following:
14  =  T(3)  +  1  +  'U  :::}  T  =  4,  'U  =  1.
This gives us
4
£(40,6,14,4)  ~ LCki,6,4),
i=l
where  I:,i=l  (Ji  =  39.  Looking  up  the tables in  [4]  we  see  that  in this case  the
sum  is  minimal when
(Jl  (Jz  =  (J3  =  10,  and  (J4  =  9
(or some similar  permutation  of this).  The  upper bounds for  0(10,6,4)
and  C(9,  6,  4)  are  20  and  12  respectively  [4].  Therefore  L(  40,  6,  14,  4)  <
3.  20  +  12  72.
Theorem  2.13.  Let n  =  n 1  +  n 2 ,  0
~
s
~
t and s
~
l
~
k  t  +  s. Then
t
C(n,  k,  t)  ~ L  m/n  C(n 1 ,  l,  s) ·  C(nz,  k  -l,  t- s).
s=O
Proof.  Let  N 1  =  {1,  ..  ,  n 1 }  and  Nz  =  {n 1  +  1,  ..  , n 2 }.  Let  Bl,s  be  the  set of
blocks you get when for each block of a ( n 1 ,  l,  s) covering design  and  each
block of a ( n 2 ,  k - l,  t - s) covering design  we  take  the  union of these two
blocks. Do this for all possible values of  lands.  Choose  an  arbitrary  p  =  t  set
A
~
N  = {1,  ..  , n}  and  let  A 1  =  AnN 1  and  A 2  =  AnNz.  Since  IA1  UAzl  =  t
we  know  that  there  is  a set of blocks  B 1 ,A 1  for each possible value of  l.  The  set
Bl,A 1  is  made from joining all blocks from  the  (n 1 ,  l,  A 1 )  covering design with
all blocks from  the  (  n 2 ,  k  -l,  t- A  1 )  covering design. Because of this, there  is
a block  B
~
B1,A 1  such  that  B  =  B1UBz, IA1nB1I  =sand  IA 2 nBzl  t-s.
When  we  form  the  union of B 1 ,  and B 2  we  see  that
lA n  (Bl  u Bz)l  =  I(Al  n  Bl)  u  (Az  n Bz)l  =  (s  +  (t- s))  =  t.
Therefore, for any  arbitrary  p  =  t  set  A  we  could choose there  is  a block
B  such  that  IE n  AI  =  t.  So  we  have a  (n,  k,  t)  covering design.  So  given
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS  11
any collection of coverings of  the  form (  n 1 ,  l,  s),  and  ( n 2 ,  k - l,  t - s) for all
possible values of  l,  and  s  we  can form a  (n,  k,  t)  covering design.
This includes  the  coverings  that  give us  the  minimum value for  the  sum of
the  products of  their  size.  The  minimal number of blocks in this covering
design is given by:
Therefore
C(v 1 ,  l,  0)  ·  C(v 2 ,  k  -l,  t)
+  C (  V 1 ,  l,  1)  ·  C (  Vz,  k - l,  t - 1)
+  C (  V 1 ,  l,  t - 1)  ·  C (  Vz,  k - l,  1)
+  C(v 1 ,  l,  t)  ·  C(v 2 ,  k  -l,  0)
t
C(n,  k,  t)::;  L  min  C(v1,  l,  s) ·  C(vz, k  -l,  t- s).
s=O
D
We can  try  different variations of  n 1  and  n 2  to  get a  better  upper  bound.
One thing  to  note  is  that  if a  +  b  =  n,  then  trying n 1  =  a  and  n 2  =  b  will
give us  the  same result as n 1  =  b  and  n 2  =  a.
2.2  Algorithms  for  Constructing  Lotto  De-
.
signs
There are  three  main types of algorithms used  to  construct  lotto designs.
They  are greedy algorithms, backtracking,  and  simulated annealing. Due  to
space constraints,  we  will only briefly discus each  type  of algorithm. During
the  writing of this  paper  a greedy algorithm was  written  and implemented.
2.2.1  Greedy  Algorithms
Greedy algorithms are  the  simplest algorithms for finding lotto designs.  They
are very easy  to  program. However,  they  usually  don't  find optimal solutions.
The  way a greedy a.lgorithm works  is  by finding a feasible solution by choos-
ing  the  best option  at  each step. So for a  lotto  design,  at  each step  the
a.lgorithm would choose  the  ticket  that  covers  the  most p sets  that  are so far
not  covered by any other ticket. Presented below  is  some basic pseudo code
for a greedy  lotto  design algorithm.
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS
Greedy(P)
Input:  P  =  [Po,  PI,  ...  , Pm],profit function
Output:  X=  [x 0 ,  xi,  ... ,  x 1 ],  a subset of  P
Profit = 0
X=0
foreach  Pi  E  P  do
if  profit(pi) >Profit  then
I
Point=  Pi
Profit  =  profit(pi)
end
end
P  +-- P\Point.
X+-- XU  Point
Greedy(P)
Algorithm  1:  Pseudo Code for a Greedy Algorithm
12
In this code  the  set  P  is  the  set of all  the  possible  k  sets for  the  lotto design
and  the  profit function  is  the  number of uncovered  p  sets each  k  set covers. As
the  code runs,  it  finds  the  k  set (called Point)  that  covers  the  most uncovered
p  sets, it  then  adds it to  the  set  X,  then  runs again with  the  set  P\Point  and
a new profit function.  It  will keep doing this until all  the  p  sets have been
covered. In  the  appendix  is  a greedy algorithm  written  in Maple.
2.2.2  Backtracking  Algorithms
Backtracking  is  an exhaustive search method, it considers every feasible so-
lution. Because of this, it will always find an optimal solution.  The  draw
back of this  is  that  it consumes large amounts of memory,  and  can take a
long time  to  run.  Certain  methods called  pruning methods  can be used to
decrease  the  workload. Presented below  is  some pseudo code based on  the
code from  [5],  pg 107 for a general backtracking algorithm.  P  is  called  the
possibility set,  it  is  made up of all  the  possible solutions  the  problem could
have. In  the  case of a lotto design it  is  made  up  of all possible tickets. For
each partial solution  X  =  [x 0 ,  X1,  ...  ,  Xz-I]  there  is  a  choice set  C 1  <;;:;  Pi  that
contains all feasible points from  Pi  given  that  X  =  [x 0 ,  XI,  ..  ,  xz_I].  Using
the  choice set is one pruning  method  because  at  each step, you  don't  have
to consider all of  the  xi's  remaining. You  just  consider  the  xi  E  C 1 •
CHAPTER  2.  LOTTO,  AND  OTHER  DESIGNS
BACKTRACK(  l)
Input:  x,  constraints
Output:  X
if  [x 0 ,  x 1 ,  ...  ]  is a feasible solution  then
I  process  it
end
compute  Ct
foreach  X  E  C1  do
I
Xt  +-----X,
BACKTRACK(l+  1)
end
Algorithm  2:  Pseudo Code for  a.  Backtracking Algorithm
13
To  start  off,  l  =  0,  and  X  =  0.  The  basic idea of  the  algorithm  is  if  X  is
a feasible solution  then  the  process command checks if it  is  better  then  the
current  best  solution, and saves it if it  is.  If  it  is  not a feasible solution it
creates new solutions by adding each  x  from  the  choice set to  the  current
partial solution  X  and  runs  the  algorithm again with  the  new  X.  Once it
has gone  through  all possible solutions it  outputs  the  current best solution
X.
2.2.3  Simulated  Annealing
Simulated Annealing  is  a modified version of  Hill-climbing.  The  way hill-
climbing algorithms work  is  they  start  with  a.  feasible solution and  try  to
improve it to get  an  optimal solution.  The  way this  is  done is by taking  the
current solution  and  looking  a.t  the  solutions in its  neighborhood,  checking  to
see if any of these solutions are better.  The  neighborhood of a solution  is
all  the  different solutions possible by slightly changing  the  current solution.
The  problem with hill-climbing  is  that  is can get stuck  a.t  a.  local maximum.
This will  happen  if  the  current solution  is  better  then  all other solutions in
its neighborhood.
Simulated annealing  is  a way of getting round this problem. In simulated an-
nealing even if  the  solution being checked  is  worse  then  the  current one there
is  a.  probability  that  this solution will be still accepted. This  is  how simulated
annealing avoids getting stuck  at  local maximums.  The  probability  that  a
solution  is  chosen despite  not  being as good as  the  current solution  is  often
referred to as  the  temperature,  and  denoted by  T.  As  the  algorithm runs,
the  temperature  T  decreases. Usually simulated annealing will run until it
reaches its iteration limit,  at  which point  the  current solution  is  outputted.
Below is some pseudo code for simulated annealing. In  the  code  N(X)  per-
forms a neighborhood search  and  chooses a random neighbor of  X.  It  returns
fail  if,  then  there  are no points in  the  neighborhood of  X.
CHAPTER  2.  LOTTO,  AND  OTHER DESIGNS
Annealing(cmax,  To,  a,  X)
Input:  Cmax,  To,  a,  X
Output:  Xbest
c=O
T=To
Xbest  =X
while  C  ::::;  Cmax  do
Y  =  N(X)
if  Y  :f  Fail  then
if  P(Y)  >  P(X)  then
X=Y
if  P(X)  >  P(Xbest)  then
I  Xbest  =X
end
else
r  =  Random(O,  1)
if  r  <  e(P(Y)-P(X))/T  then
I  X=Y
end
end
end
c=c+1
T=aT
end
return  Xbest
Algorithm  3:  Pseudo  Code  for a  Simulated  Annealing Algorithm
14
In  the  algorithm  c  records  the  number  of  iterations  and  Cmax  is  the  maximum
number  of iterations.  the  algorithm  starts  off  with  a feasible solution  X  it
checks a  random  feasible solution from  the  current  solution's  neighborhood.
If  this  solution is  better  then  the  current  one  then  it  becomes  the  current
solution.  If  not,  then  it  chooses a  random  number  between 0  and  1.  If
this  number  is less  then  e(P(Y)-P(X))/T  then  it  becomes  the  current  solution
anyway.  At  the  end  of each  iteration  the  temperature  T  is decreased by a
factor of  a.  Once  it  reaches its  iteration  limit  it  outputs  the  current  best
solution.
CHAPTER  3
Lotto  Designs  on  the  NZ  Lotto
System
3.1  The  NZ  Lottery
In  the  NZ  lotto system there are  40  balls  and  6 are chosen  at  random. After
this a final bonus ball  is  chosen  at  random from  the  remaining 34 numbers.
Below  is  a  table  of  the  possible winning tickets  and  how much you can expect
to win
1 .
For more information see  [1]
Number of  Bonus  Average  Corresponding
Winning Numbers  Ball  Winnings  Lotto Design
Division 1  6  No  $833,333  (40,6,6,6)
Division 2  5  Yes  $27,587  (40,6,7,6)
Division 3  5  No  $699  (40,6,6,5)
Division 4  4  Yes  $64  (40,6,7,5)
Division 5  4  No  $34  (40,6,6,4)
Division 6  3  Yes  $24  (40,6,7,4)
We  want  to  be able to construct lotto designs corresponding to these six
divisions.  The  obvious goal  is  be  able to construct a  lotto  design for one (or
more) of these divisions such  that  the  amount of money required to buy all
the  tickets in  the  design  is  less  then  the  amount  of money  we  can expect to
win.
The  first step  is  to calculate upper  and  lower bounds for these divisions.
Calculating upper  and  lower bounds for a design  is  much easier  to  do  then
to calculate  the  actual lotto design.  If  we  discover  that  the  lower bound for
a lotto design is greater  the  maximum number of tickets  we  can by and still
make a profit,  then  there is no point in trying to construct  that  lotto design.
Using proposition 2.9 and theorem 2.10  we  can construct lower bounds for
the  NZ  lotto system.  With  the  power ball  we  have:
62145  :::::
T(40,7,6)
::;  L(40,  6,  7,  6)
m
6906  :::::
T(40,7,5)
::;  L(40,  6,  7,  5)
m
280  :::::
T(40,7,4)
::;  L(40,6, 7,4)
m
1  Based on  the  lotto draws over  the  last 10 weeks
15
CHAPTER  3.  LOTTO  DESIGNS  ON  THE  NZ  LOTTO  SYSTEM  16
And  without making use of  the  power ball  we  have:
3838380  ::::;
T(40,6,6)
::::;  L(40,6,6,6)
m
21325  ::::;
T(40,6,5)
::::;  1(40,6,6,5)
m
433::::;
T(40,6,4)
::::;  1(40,  6,  6,  4)
m
Unfortunately, now  that  we  have calculated lower bounds for all six divisions
we  see  that  no possible lotto design could guarantee us a profit, as  the  table
below shows. This table  is  just  for  the  lower bounds,  the  actual designs could
cost us a lot more.
Cost  Guaranteed Average Winnings  Profit
Division 1 $2,303,028  $833,333  -$1,469,695
Division 2  $37,287  $27,587  -$9,700
Division 3  $12,  795  $699  -$12,096
Division 4  $4,143.6  $64  -$4,079.6
Division 5  $259.8  $34  -$225.8
Division 6  $168  $24  -$144
Even though none of these lotto designs can  be  of use to us,  we  would still like
to calculate some upper bounds for them. These are generally a lot trickier
to calculate. We will use a number of different theorems to  try  to get  the
lowest  upper  bound.
Not all of  the  cases  we  are interested in are difficult.  The  upper bound for
1(40,  6, 6,  6), (which corresponds to winning first division),  is  very easy to
calculate. To have a ( 40,  6, 6,  6)  lotto design,  we  require  that  choosing sets of
6,  every possible set of 6  is  contained in one of  the  sets  we  chose.  The  only
way to achieve this  is  to choose every possible set of  6.  So  1(  40,  6,  6,  6)  =
( ~
0
) =  3838380.  So  the  upper  bound  is  (trivially) 3838380.  That  is,
1(  40,  6, 6,  6)  :S;  3838380.
Another easy  upper  bound  is  1(  40,  6,  7,  4), which would guarantee a sixth
division win. Using theorem  2.  7,  varying  the  values of n 1 ,  and n 2  we  get
1(40,6,  7,4)::::;  1(20,6,4,4)+1(20,6,4,4).  Onethingtonoteisthat1(40,6,4,4)
is  a covering design. This  is  good because  we  have access to some good
tables on covering designs in  [4].  From  the  tables in  [4]  we  know  that
1(20,  6,  4,  4)  ::::;  456. This gives us our second  upper  bound,
1(  40,  6,  7,  4)  ::::;  912.
Now, using theorem 2.11,  we  know  that  1(40,  6,  7,  6)  ::::;  C(39,  6,  6). Using a
similar argument as  we  did for  1(40,6,6,6),  we  deduce  that  1(40,6,  7,6)::::;
C(39,  6,  6)  =  e:)  =  3262623. So now  we  have our  third  upper  bound.
1(40,6,  7,6)::::; 3262623
CHAPTER  3.  LOTTO  DESIGNS  ON  THE  NZ  LOTTO  SYSTEM  17
The  final three upper bounds require a  bit  more work. Using theorems  2.11
and  2.13  we  can construct upper bounds for  the  remaining lotto designs,
L(40,6,  7,5),  1(40,6,6,5),  and  1(40,6,6,4).  These are
1(40,6,7,5)  <  105339
1(40,6,6,5)  ::::;  123478
1(40,6,6,4)  ::::;  7474
There is quits a lot of tedious work involved in these calculations. See  the  ap-
pendix for an example of  the  calculations involved in calculating  1(  40,  6,  6,  4).
Now  we  have complete lower  and  upper bounds for all six divisions of  the
NZ  lotto system.  With  the  bonus ball  we  have:
Division 2 : 62145  ::::;  1(40,  6,  7,  6)  ::::;  3262623
Division  4:  6906  ::::;  1(40,  6,  7,  5)  ::::;  105339
Division  6:  280  ::::;  1(40,6,  7,4)  ::::;  912
And without  the  bonus ball.
Division 1 : 3838380  ::::;  1(  40, 6,  6,  6)  ::::;  3838380
Division  3:  21325  ::::;  1 ( 40,  6, 6,  5)  ::::;  123478
Division 5 : 433  ::::;  1(40,6,6,4)  ::::;  7474
3.2  Lucky  Numbers
Some people foolishly or otherwise  think  that  some numbers are lucky, ie
have more chance of being drawn then regular numbers.  If  this  is  the  case
then  you can drastically reduce  the  number of tickets needed  to  get  at  match.
Of course in these situations you lose  the  guarantee of getting  at  match.  But
if you believe  that  the  lucky numbers are suitably likely  to  be drawn  then
you can construct lotto designs with a positive expected gain.
Example  3.1.  Working with  the  NZ  lotto system, assume  that  you believe
that  16  out  of  the  40 numbers are lucky. We will assume  that  there  is  a
high enough chance  that  of  the  7 numbers drawn, all of  them  will come from
these  16  lucky numbers.  If  these numbers were  not  lucky,  then  this would
only  happen  about  0.06% of  the  time. To get a 4  match  (and hence win sixth
division)  we  will need an  1(16,  6,  7,  4)  lotto design. Using theorem 2.10 (and
theorem 2.9)  we  can construct a lower bound for this design.
1(16,6,7,4)  2::
T(16,  7,  4)
( ~ )
C:)  16-7  +  1
( ~ ) .  ( ~ ) 16  - 4  +  1
1820  10
--·-
300  13
==  5.
Therefore  1(16,  6,  7,  4)  >  5.
CHAPTER  3.  LOTTO  DESIGNS ON THE  NZ  LOTTO  SYSTEM  18
This gives a lower  bound  of  5.  So  with each ticket costing $0.6, this design
would cost a minimal of $0.6 · 5  =  $3.  With  the  average  return  for winning
sixth  division being around $24, you could make a profit if all of  the  numbers
drawn were from your lucky numbers occurred  about  every eighth draw.
However, this would mean  that  it was happening  about  200 times more often
then  it should be. So for this to be  the  case your lucky numbers would have
to  be  very lucky. This  is  just  a lower bound,  we  may need  to  buy more tickets
then  this to get our design.
From  [6]  we  get  an  upper  bound  of  14  for a (16,  6,  7,  4)  lotto design. This
would cost you $8.4.  If  L(16, 6,  7,  4)  was  14  then  you would need all of  the
numbers drawn  to  be  from your set of lucky numbers happening every  third
time. This  is  about  540 times higher  then  what  you would expect normally.
Your numbers would have to  be  extremely lucky for this to happen.
If  we  want to construct an (16,  6,  7,  4)  lotto design  there  are several ways
we  can do this. One way  is  to first use theorem 2.7. This tells us  that
L(16,  6,  7,  4)  :::;  L(8,  6,  4,  4)  +  L(8,  6,  4,  4). Doing this make it a lot easier
because it  is  a lot easier to calculate an (8,  6,  4,  4)  lotto design  then  it  is  to
calculate a (16,  6,  7,  4)  lotto design.  The  following (8,  6,  4,  4)  lotto design was
constructed by  the  greedy algorithm in  the  appendix.  The  blocks are:
{1,2,3,4,5,6},{1,2,3,4,  7,8},{1,2,5,6,  7,8},{3,4,5,6,  7,8},{1,2,3,4,5,  7},
{1,2,3,4,5,8},{1,2,3,4,6,  7},{1,2,3,4,6,8}.
We  can join this lotto design together with a similar covering on {9,  10, 11, 12,  13,  14,  15, 16}.
This will give a (16, 6,  7,  4) lotto design with  the  following blocks:
{1,2,3,4,5,6},  {1,2,3,4,  7,8},  {1,2,5,6,  7,8},  {3,4,5,6,  7,8},{1,2,3,4,5,  7},
{1,2,3,4,5,8},{1,2,3,4,6,  7},{1,2,3,4,6,8}{9,10,11,12,13,14},
{9,10,11,12,15,16},{9,10,  13,14,15,16},{11,12,13,  14,15,16},{9,10,  11,12,13,15},
{9,10,11,  12,13,16},{9,10,11,12,14,15},{9,10,11,12,14,  16}.
Example  3.2.  Once again  we  will be working  with  the  NZ  lotto system,
and  the  same  set  of lucky numbers as example 3.1. This time  we  will look  at
L(16,  6,  7,  5), which corresponds to winning fourth division( 4 numbers from
the  6 drawn plus  the  bonus ball).  The  odds of all 7 balls drawn being from
our set of lucky numbers  is  around 0.06%.  The  average winnings for fourth
division  is  about  $64. Again using theorem 2.10  we  can  construct a lower
bound for L(16,  6,  7,  5), which  is  41. To use this design it would cost us
a minimum $24.6,  so  to make a profit  the  odds of all 7 balls being drawn
from our set of  16  lucky numbers must be greater  then  around 61%. This
is  roughly 1000 times higher  then  expected odds. This  is  also  the  best case
scenario, from theorem 2.11 and  [4]  we  find  that  the  upper  bound  is  385.  It
would cost $231 to implement this design,  about  3.6 times more  then  what
you could win. In this case,  the  design would be very impractical.
CHAPTER  4
Conclusion
Winning lotto  is  what  dreams are made  of.  But  we  have seen  that  winning
is  not an easy task. While it can be easy  to  construct small lotto designs,
constructing large lotto designs (  n  >  30  say)  is  very difficult. This  is  what
lotteries rely on. Otherwise everybody would be constructing lotto designs.
Despite  the  difficulty, this  is  still an active area of research. In this paper  we
saw only a small  part  of  the  number of theorems  and  constructions related
to  lotto designs.
In chapter 3  we  constructed lower and upper bounds for  the  NZ lotto system.
While these were  not  optimal bounds, they still gave us  an  idea of  the  number
of tickets needed for  the  corresponding lotto designs. In particular,  we  saw
that  the  possible winnings for  the  NZ  lotto system were well below  the  value
needed to make playing lotto practical.  That  is,  the  number of tickets needed
to  make an optimal lotto design for any one of  the  divisions was far too high
when compared with  the  amount  we  could win. However, this  is  just  one
lottery, there may be other lotteries  out  there for which effective lotto designs
can be constructed.
19
APPENDIX  A
Calculating  the  Upper  Bound
for  L(  40,6,6,4)
To calculate  the  upper bound for  £(40,6,6,4),  the  first thing  we  need to do
is  use proposition 2.11 to write  L(  40,  6,  6,  4) as  the  sum of suitable coverings.
When  we  do this  we  get  the  following results.
p  T(t- 1)  +  1  +  V
6  T(4-1)+1+v
5  3T+  V
Therefore  T  =  1,v  =  2.  So  we  have £(40,6,6,4)::::;  0(38,6,4).  Now  we  use
proposition 2.13 to calculate an upper bound. For  the  best  results  we  have
to check all values of  n 1  such  that  19  ::::;  n 1  ::::;  32.  We  don't  check for higher
values of  n 1  because  then  n 2  would be less  then  k,  which is not possible for
a covering. After  we  do a  few  calculations  we  will come to  n 1  =  27,  and
n 2  =  11.  This gives us  the  smallest upper bound.  The  following calculations
were used to calculate  the  upper  bound.
0(27,  0,  0)  ·  0(11,  6,  4)  =  1·  32  =  32
min of  0(27,  1,  0)  ·  0(11,  5,  4)  1·  66  =  66
0(27,  2,  0)  ·  0(11,  4,  4)  =  1·  330  = 330  min=  32
0(27,  1,  1)  ·  0(11,  5,  3)  =  27.20  = 540
+ min of  0(27,  2,  1)  ·  0(11,  4,  3)  =  14.47  =  658
0(27,  3,  1)  ·  0(11,  3,  3)  =  9.  165  = 1485  min=  540
0(27,2,2).  0(11,4,2)  =  351.  11  = 3861
+ min of  0(27,  3,  2)  ·  0(11,  3,  2)  = 117.  19  = 2223
0(27,  4,  2)  ·  0(11,  2,  2)  =  61.  55  = 3355  min=  2223
0(27,  3,  3)  ·  0(11,  3,  1)  =  2925.4  = 11700
+ min of  0(27,  4,  3)  ·  0(11,  2,  1)  = 763. 6  = 4578
0(27,  5,  3)  ·  0(11,  1,  1)  = 319.  11  = 3509  min=  3509
0(27,  4,  4)  ·  0(11,  2,  0)  = 17550. 1 = 17550
+ min of  0(27,  5,  4)  ·  0(11,  1,  0)  = 3906. 1  = 3906
0(27,  6,  4)  ·  0(11,  0,  0)  = 1170. 1  = 1170  min=  1170
total=7474
From this  we  have £(40,6,6,4)::::; 7474. This  is  the  smallest upper bound
possible using this approach.
20
APPENDIX  B
Greedy  Algorithm  in  Maple
What  follows  is  a greedy algorithm  to  construct lotto designs written in
maple.  The  code  is  usually in one piece,  but  has been broken up to allow for
comments explaining  what  it  is doing.
greedy:=proc(n,k,p,t)
local  lp,mp,lp2,lp3,lk,mk,cover,total,P,count,j,
com,ii,jj,hh,gg,iteration,winner,h,rank,z;
with(combinat,  choose);
with(combinat,numbcomb);
lp  is  the  set of all possible p sets.
lp:=choose(n,p);
mp:=numbcomb(n,p);
lp2:=array(1  ..  mp,1  ..  1);
lp3:=array(1  ..  mp,1  ..  1);
lk is  the  set of all possible k sets.
lk:=choose(n,k);
mk:=numbcomb(n,k);
cover is  the  maximum number of p sets any k  set  can cover.
cover:=O;
total  is  the  total  number of blocks in  the  design.
total:  =1;
P:=numbcomb(p,t);
count is  the  total  number of p sets  that  are covered  at  each point of  the  code.
count:=O;
rank  keeps  track  of how many uncovered p sets each k set covers  at  each
iteration of  the  code.
rank:=Matrix(mk,1);
The  next  part  of  the  code checks to see  the  maximum  number  of p sets each
k  set  can cover.
21
APPENDIX  B.  GREEDY  ALGORITHM  IN  MAPLE  22
for  j from 1  to  mp  do;  com:=choose(lp[j]  ,t);
for  ii  from 1  to  P do;
if  ((convert(com[ii]  ,set)  subset  convert(lk[1],set))=true)  then;
cover:=cover+1;
break;
end  if;
end do;
end do;
Before any k set  is  chosen, every possible k set covers  the  maximum number
(cover) of uncovered p sets.
for  jj  from 1  to  mk  do;
rank[jj,1]  :=cover;
end do;
The  first block  the  code chooses  is  {  1,  ...  , k}.
print(convert(lk[1],set));
count:=cover;
Now  the  code adjusts  the  number of uncovered p sets  that  each possible k
set covers according to  the  p sets covered by  the  first block, and also sets
lp2[j,1] to 0 if  the  p set  j  is  now covered.
for  jj  from  2  to  mk  do;
for  j from 1  to  mp  do;
com:=choose(lp[j]  ,t);
for  hh from 1  to  P  do;
if  ((convert(com[hh]  ,set)  subset  convert(lk[1]  ,set))=true)  then;
for  gg from 1  to  P  do;
if  (convert(com[gg],set)  subset  convert(lk[jj],set)=true)  then;
rank[jj,1]  :=rank[jj,1]-1;
break;
end  if;
end do;
lp2  [j  '1]  :  =0;
break;
end  if;
end do;
end do;
end do;
Now  the  code runs  through  100 iterations choosing  the  best  k set  at  each
iteration until all p sets are covered. For larger designs  we  would need to
change  the  upper  limit of iterations  to  something greater  then  100.
for  iteration  from 1  to  100 do;
if  (countAPPENDIX  B.  GREEDY  ALGORITHM  IN  MAPLE  23
The  first step  is  to find the unchosen k set  that  covers the most uncovered p
sets.
winner:=2;
for  h from 2  to  mk  do:
if  (rank[h,1]>=rank[winner,1])  then;
winner:=h;
end  if;
end do;
print(convert(lk[winner]  ,set));
Once the best k set  is  chosen, count increases by the number of uncovered p
sets the k set covers, then the number of uncovered p sets this k set covers
is  set  to  0.
count:=count+rank[winner,1];
total:  =total+1;
rank[winner,1]  :=0;
Now  the number of uncovered p sets  that  each k set covers  is  adjusted ac-
cording to the p sets  that  the k set covered.
for  jj  from 2  to  mk  do;
for  j from 1  to  mp  do;
First  we  check to see if  the  current k set covers any uncovered p sets.
if  (rank[jj,1]<>0)  then;
If  it  does, then  we  check to see if each uncovered p set  is  covered by the k
set.
if  (lp2[j,1]<>0)  then;
com:=choose(lp[j]  ,t);
for  z from 1  to  P do;
if  ((convert(com[z],set)  subset  convert(lk[winner]  ,set))=true)
then;
lp3[j,1]:=1;
for  ii  from 1  to  P do;
if  ((convert(com[ii]  ,set)  subset  convert(lk[jj]  ,set))=true)
then;
rank[jj,1]  :=rank[jj,1]-1;
break;
end  if;
end do;
break;
end  if;
end do;
APPENDIX  B.  GREEDY  ALGORITHM  IN  MAPLE
end  if;
end  if;
end do;
end do;
for  j from 1  to  mp  do;
if  (lp3[j,1]=1)  then;
lp2  [j  '1]:  =0;
end  if;
end do;
end  if;
end do;
print(total);
end  proc:
24
It  may seem strange to force  the  algorithm  to  choose  the  first block to be
{1,  ..  , k},  but  this has no effect on  the  efficiency of  the  algorithm. Through a
suitable isomorphism  we  can change  the  block {1,  ..  ,  k}  to  any other k subset
of  N  = {  1,  ..  , n}.  The  fact  that  the  first block has points labeled 1 to  k  does
not change  the  size or effectiveness of  the  design.

Nessun commento:

Posta un commento