You are on page 1of 22

Int. J.

Man-Machine Studies (1972) 4, 341-362

Multi-dimensional Structure in the Game of Chess

R. H. ATKIN
Department of Mathematic& University of Essex,
Colchester, Essex, U.K.

(Received 23 May 1972)

This paper is art immediate follow-up of the author's previous "From Cohbmology
in Physics to q-connectivity in Social Science"; published in the previous issu6 of
this journal.
It is art illustration of the theory of structure contained in the first paper, being
an application to the game of chess.
The relations which exist between the chess-men and the squares of the chess-
board, together with the rules of the game, are used to define mathematical rela-
tions Fw and FB. The connectivity patterns in multi-dimensional spaces, which
these relations define, are then obtained by the use of computer programs written
in Basys.
The accepted theories of positional chess are traced out and expressed in terms
of the structure in Q-space; in particular, checkmate can be defined precisely in
terms of the intersection of appropriate simplicial complexes.
Finally, a detailed analysis of the so-called Immortal Game (that between Anders-
sen and Kieseritsky--London, 1851) is given in terms of connectivity patterns:
the subordination of tactical play to positional judgement (the structure in Q-space)
is illustrated by the variation of the Structure Vector.
If chess masters base their profound positional judgements on an unconscious
appraisal of the connectivity structure in Q-space--as the author believes--then
this analysis holds out the prospect of a chess-playing computer using the same
methods of assessment in its play.

1. A Review of Structural Definitions


A b i n a r y relation 2, between finite sets X, Y, defines a m u h i - d i m e n s i o n a l
structure k n o w n as a simplieial complex K. I n fact there are two complexes
for each 2, which we shall denote by Kx(Y;2) a n d K r ( X ; 2 - 1 ) : either will be
referred to as the conjugate o f the other. F o r a m o r e detailed discussion o f
the definitions a n d possible applications refer t o A t k i n (1972); here we shall
summarize the definitions.
341
,342 R.H. ATKIN

The r.etation 2 may be represented by a matrix array with entries 0,1 ; this
is summarized in the notation
h I x
Y I A-----(&,j)
where A denates the incidence matrix. T h e incidence matrix of the inverse
relation •~-1 is the transpose (Z~j) with z~j----2j~. We define (e.g.) the complex
Ky(X;2)/ts.follows:
(a) K'y(X;2) is a collection of simplices (o-p},p=O,1 . . . . . N.
(b) Each ~rp e K is a subset of ( p + l ) distinct Xt, provided there is at
least o n e Yk e Y such that (X~,Yk) e 2 for each of the ( p + l )
values of i.
(c) The-a~0 are identified with the X~, i = 1 . . . . . n ( n = c a r d X).
(d) EverY ( q + l ) subset of a ap(q ~<p) is called a q-face of the o-p and
defines a o-~ ~ K.
The finite number N in (a) is called the dimension of K and written dim K;
the dimension of each simplex ap is the value of p, and N is the largest value
o f p in K. The set X={X~} is called the vertex set of Kr(X;2). We notice too
that each simplex a p e K corresponds to at least one Irk e Y. When there is
no danger of confusion we might speak loosely of the set Y denoting the
simplices of Kr(X;2)--hence the notation. When we so speak of the simplex
Yk we shall mean that ap, for max p, which corresponds to Yk.
In a similar way, if we let Y be the vertex set, ).-1 gives us the conjugate
complex Kx( Y;2-O.
It is common to think of a geometrical representation of a complex, in
which each p-simplex ap is a closed convex polyhedron of ( p + l ) vertices--
in a suitable Euclidean space E n. Then, for example, a o-z becomes a tetra-
hedron (and all its faces), a as becomes a triangle (and all its faces), and acr 1
becomes an edge (and its end-points). If dim K = N it can be shown that a
geometric realization of K requires, in general, a space E n with H ~ 2 N + 1.
Thus an application of the idea of complexes naturally corresponds to
studying multi-dimensional structures (in the Euclidean sense).
In particular we study the structure of a relation 2 by way of the con-
nected components of K r(X) and of Kx(Y). To this end we define the notion
of q-connectivity in the following w a y - -
Simplices ap, o~ are said to be joined by a chain of connection if there exists
a finite sequence of simplices
O'~z, 0 " ~ . , . , O'~h
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 343

such that
(i) a~, is a face ofa~,
(ii) a~h is a face of a,,
(iii) a~l and a~i+~ have a common face (say) ao,, for i = 1 . . . . (h--i).

We say that this chain of connection is a q-connectivity if q'is the least of


the integers ,. _,

As a special case a simplex ap is p-connected to itself, but not (p + 1)-connected


to any at.
If we define a relation ?q as meaning "is q-c0nnected with" then ?q is an
equivalence relation on the simplices on K. The classes of 7¢ are now the pieces
of K which are separately q-connected. If the cardinality of the set {classes of
yq} is Q~ we notice that, although Q0---zfro Betti number, in general Q ~
q-th Betti number.
The set of values
Q:{QN, QN-1. . . . . Q1, Q0}
is referred to as the (first) structure vector of the complex K and may be
regarded as a vector in an affine Q-space. This idea, and the associated idea
of an obstruction vector Q is fully discussed in Atkin (1972).
The significance of the structure vector, in the context of this paper, will
emerge in section 3.

2.1. Specific Relations in Chess


The game of chess is a game played on a board of 64 squares between
two players White and Black, each of whom starts with 8 pieces and 8 pawns.
The rules of the game are sufficiently well known to need no repetition here;
we shall use the international algebraic notation to describe the board and
the pieces.
The squares of the board are named a l . . . h8 and the pieces are denoted
by R (rook), N (Knight), B (bishop), Q (queen), K (king) and P (pawn).
The initial position is shown in the diagram.
In Fig. 1 we distinguish the pieces by their initial positions Q-side (e.g. QR)
or K-side (e.g. KB). Figure 2 shows the simple numbering of the squares
1-64, used in the computing programmes; thus the Black KB is ou square
f8, which equals 6 + ( 8 - - 1 ) × 8=62. Incidentally the initial position requires
the square hl ( = 8 ) to be a white square, according to the F.I.D.E. ,rules.
344 R. H. ATKIN

Black

a b c d e f g h
White
FIo. 1. M o d e [0,0]

t
57 J 58 59 60 61 62 6:'5 6,4

49 50 51 52 53 54 55 56

41 42 43 ,4,4 ,45 46 47 ,48

33 34 35 36 37 38 39 40

25 26 27 28 29 30 31 32

17 18 19 20 2t 22 23 2,4

9 I0 II 12 13 14 15 16
I
I 213 4 5 6 7 8

FIG. 2.

The world champion Emmnuel Lasker once remarked, "there are 64


squares on the chessboard; if you control 33 of them you must have an
advantage". This is a gross simplification of the problem but it illustrates the
fact that chess masters have always had a sharp sense of the so-called "posi-
tional" or "strategic" nature of the play. This feature of the game can be seen
immediately as an expression of the mathematical relation F between the men
and the squares. We therefore consider two relations Fw and FB; F w gives
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 345

the relation between the White men and the 64 squares, FB that for the Black.
Precisely, we define (e.g.) Fw by the statement:
given X--a piece or pawn, S=square on the board,
then X FwS (or(X,S) 8 Fw)
if and only if X attacks S.
By attacks we mean that one of the following holds true:
(a) if it were White's move then " X goes to square S " is a legal move;
(b) if there is a White mart Y on square S then X is protecting Y, in
ordinary chess-players' parlance;
(c) if X is the White K then S is an immediate neighbour to the square
occupied by X; horizontally, vertically, or diagonally;
(d) if the square S contains a Black piece or pawn Z (other than the K)
and if it were White's move then " X captures Z " is a legal move;
(e) the Black K is on square S and is in check to White's X.
Having defined Fw and FB, for any position on the board, we therefore
have four complexes at our disposal, viz.
Kx(S;rw), Ks(X;r-lw)
and
Kx(S;FB), Ks(X;F-1B)
We shall describe Kx(S;Fw) as White's view of Board, and the conjugate
Ks(X;F-lw) as Board's view of White.
In the complex Kx(S;Fw) each man (X) is representing a simplex with
vertices selected from the 64 squares. If two pieces XI, X~ are q-connected in
this K then they share (q-t-l) squares on the board, that is to say, there exist
(qq-1) squares which are attacked by both X~ and X2.
In the complex Ks(X) each square S is representing a simplex with vertices
selected from the 16 men. If two squares $1, S, are q-connected in this K t h e n
they share (qq-1) men, that is to say there exist (q-t-l) men who are attacking
both S~ and Sz.
A particular distribution of men (including both White and Black) on
the squares is referred to as a mode. A game commences in mode [0,0]; after
White's first move the game is in mode [I,0]; after Black's first move the
game is in mode [1,1]. Figure 1 shows mode [0,0], and the commencement
of a game; White to move. In this position the QR-pawn (on a2) is a 0-simplex,
since it attacks the one square b3 (number 18), similarly the KR-pawn is a
0-simplex, whereas each of the remaining White pawns is a 1-simplex (e.g. the
Q-pawn is the 1-simplex < 1 9 , 2 1 > . The QR is also a 1-simplex ( < 9 , 2 > )
whereas the QN is a 2-simplex (viz., <17, 19, 12>). The QB is the 1-s:mplex
346 R.H. ATKIN

< 1 0 , 12>, the Q is the 4-simplex < 3 , 11, 12, 13,5>, whilst the K is also a
4-simplex, viz. < 4 , 12, 13, 14, 6 > . The Q and K, in mode [0.0], are therefore
1-connected, with a common face < 1 2 , 1 3 > ; notice however that in Fw
(or/'B) a piece does not attack its own square. This is of practmal significance
since the pieces and pawns must protect each other by attacking the occupied
squares.
We can therefore list the dimensions of and connectivity patterns between
White's men when mode [0,0] (Fig. 1) applies, as follows:

TABL~ 1
Kx (S; Fw) = White's view of Board Onode [0; 0])
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
1 1 QR
2 0 0 0 1 -- 0 2 QN
1 0 0 3 QB
4 1 0 0 4 Q
4 0 0 5 K
1 0 6 KB
2 0 -- 1 -- 7 KN
1 8 KR
0 -- 0 9 QRP
1 -- 0 10 QNP
1 -- 0 11 QBP
1 -- 0 12 QP
1 -- 0 -- 13 KP
1 -- 0 14 KBP
1 -- 15 KNP
0 16 KRP

The values of dim X are in the diagonal of this triangular array; the remaining
integers giving q-values in the connectivity patterns; the dashes (--) indicated
q------l, which means disconnected. The easiest way to read the table is to
start at a number in the diagonal and read upwards and then to the left,
or to the right and then downwards. Thus
(i) dim Q = 4 Q is a 4-simplex (5 squares attacked by Q)
and is 0-connected to QB (via < 1 2 > )
and is 0-connected to Q N (via < 1 2 > )
(ii) dim Q N = 2 Q N is a 2-simplex (3 squares attacked by QN)
and is 1-connected to Q N P (via <17, 1 9 > )
and is 0-connected to QP (via < 1 9 > ) ,
etc.
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 347

2.2. Positional Play


Various positional motifs in master play have an illuminating interpretation
in terms of our connectivity patterns.
The modern idea of positional play commenced with Philidor (whom
Reti (1933).has designated the greatest chess thinker of all time) and subse-
quently made great advances through the work of Stenitz (a point in time
which Emmanuel Lasker (1932) described as the greatest landmark in the
history of chess), and more latterly through Rubinstein, Nimzowitsch, and
then Botwinnick and the modern Russian school, to name only a few.
(i) To Philidor is attributed the sentiment, "the pawns are the soul of
chess", and his own theory and praxis certainly bore witness to this idea.
Philidor's main point was to emphasize the idea, of unity in play, the co-
operation of pieces and pawns in the development of a successful attack on
the opponent.This led him to place great emphasis on the "orderly advance"
of a pawn mass--often, in practice, to the neglect of appropriate piece play
behind the pawns. We see at once from our analysis in section 2.1 that a
pawn can, at the most, be a 1-simplex in Kx(S); its individual influence
(neglecting the queening potential) can therefore be only slight in comparison
with a piece. But it follows also that, in mode [0 ;0] the pawns form a con-
nected chain (at q=0) in the complex; their overall power vis-a-~,istheir power
to attack and control squares on the board is considerably enhanced. It is
not difficult to see that in mode [0;0] and in the conjugate complex Ks(X)--
that is to say, Board's view of White--the pawn subcomplex (which forms a
0-connected chain) is responsible for certain squares being at least 1-simplices.
These squares are {18, 19, 20, 21, 22, 23}; for example, in Ks(X) the square 20
represents the simplex <QBP, K P > .
Furthermore we see that the q-connected components of Kx(S;Fw), in
mode [0;0], are as follows:
Q4=2 Components are Q and K
Q3=2 Components are Q and K
Qo=4 Components are Q, K, QN, KN
Q I = 11 Components are {KN, KNP}, {QN, QNP}, {e, K}, {QR}, {KB},
{QB}, {QBP}, {QP}, {KP}, {KBP}
and Q0--3 Components are {KR}, {QR}, {the rest}
Hence the pawns (in toto) are 0-connected to all the pieces except the two
Rooks. In this sense we can see that (at the q=0 level) the chain of pawns
348 R . H . ATKIN

acts as a support for most of the pieces vis-a-vis the connectivity. It is immedi-
ately clear that if the White pawns-and-pieces could advance u.p the board
~t la Philidor (in an "orderly manner") this connectivity structure would be
preserved. Indeed such an advance would soon activate the rooks in such a
manner that Q0--1 so that, at the level of q = 0 White's view of the Board
would be one connected component. The degree of connectivity inherent in
Kx(S) is in fact a recurring theme in chess theory, albeit expressed in different
terms. Thus, with Philidor, we can see that the pawns exhibit (even in [0, 0])
the basic idea of connectivity in their relations with the squares of the board
(modulo the rules of the game). If connectivity patterns are at the heart of the
game of chess then the pawns can certainly be said to be the soul of the game.
(ii) gteinitz introduced the idea of weak sq'uares and strong squares into
chess thinking. He could therefore speak of weak positions or strong posi-
tions; he set himself the task of finding the characteristics of position which
amounted to an advantage or a disadvantage. By a strong square S he meant
a square which is amply controlled by the pieces and pawns so that, for
example, it can be safely occupied by a knight or other piece. It is even
stronger if it is well placed, a strong square at 28 is "stronger" than a strong
square at, say, 25. The reasons are clear.
A strong square for White (Steinitz) is a square S such that
dim S in Ks(X;F-lw)>dim S in Ks(X;I"-IB)
- - a n d this is obtained without weighting the vertices of S in any way (see
below). If such a square is to be used by White for posting a Knight then it
is important for the N (on that strong square) to have maximum affect, to
have a maximum dim in Kx(S;Fw). On S----28, for example, dim N-----7 since
N=<ll, 18, 34, 43, 45, 38, 22, 13>, whereas on S = 2 5 we have only
dim N = 3 .
Powerfully placed pieces are pieces with largest possible values of dim X,
in the first place. General control of the board corresponds to a high degree of
connectivity between powerfully placed pieces.
Steinitz broadened chess theory to include an appreciation of the Board's
view of White and urged us to hold it in the mind during the process of
controlling White's view of Board. He effectively said that not only are White's
men co-operating on the Board's squares but equally well are the Board's
squares co-operating on White's men. If a square S is a weak square for
White then it is a strong square for Black. For example, the isolated pawn
is a well-known instance'in point. If White's Q-pawn is isolated on S = 2 8
then the square S = 3 6 is weak for White, strong for Black. The advantage
for Black is realized when he places a piece (something with a large dim X
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 349

under these conditions--not a Rook) on S----36. For White, with respect to


the pawn mass, square S----36 is disconnected (q------1) in Ks(X;F-lw).
(iii) The relative power of the pieces is a function of the mode [I,J] of" the
game (which lies behind every sound Q-sacrifice), but for fixed 1,J, we must
measure it as the ratio dim X: dim Y in Kx(S). On this score the wing pawns
are bottom of the table, being only 0-simplices. It is not surprising that, for
example, Tchigorin (and later, Tartakower) scored many surprises with
opening play which consisted in an early advance of the KRP--and its early
exchange if possible. Such an advance immediately increased the value of
dim" KR, giving the Rook a powerful early entry into the game: after only
two moves by White (h2-h4 and R-h3) it is possible for the value of dim
KR to have increased from 1 to 9. This seems a handsome return for (pos-
sibly) sacrificing a wing pawn. It also illustrates much of the positional
thinking about opening play and wing gambits in particular.
The elementary discussions of relative strengths of pieces usually suggest
what is by-and-large the accepted practical values, viz.,
with P = I we take B=N----3, R = 5 , Q=9, K----infinite.
The infinite value for K is not simply a connectivity value but expresses the
peculiar role played by the King in that it cannot be captured in any mode
[I,J], nor can it be exchanged for any (arbitrary) number of pawns--as, for
example, can the B or R.
If we leave the King out of account then we Call reasonably assess the
relative strengths of the Pawns and Pieces in any mode by their dimensions.
In mode [0,0] these are given in the array in section 2.1. After the opening
move (1) e2-e4 the g.ame is in mode [1,0] and it is easy to see that the dimen-
sions of the simplices in K x ( S ; F w ) are unchanged except for the following:
dim Q--7, dim KB=5, in mode [1,0]
whereas dim Q--4, dim K B = I , in mode [0,0].
This point of view sheds valuable light on that thorny problem of B v. N.
There are many positions in which the B is to be preferred to the N, and
conversely; the chess master Rubinstein played those endings involving
B v. N with a sound positional judgement. This judgement involved assessing
the subsequent structures of Kx(S) from the point of view of controlling the
dimensions of the B- or N-simplices. Pawn moves which limit the power of
the opponent's B (or N) whilst at the same time extending the power of one's
own N (or B) play a crucial role in such situations. "Before allowing that type
of position to occur it is therefore necessary to see the structural pattern of
future complexes K x ( S ) and how they can be attained by tactical means.
350 R.U. ATKIN

In this respect, the orthodox and rigid valuation of B and N (as equally
weighted) is not helpful.
The same reasoning lies behind the notions of good B and bad B. The latter
is one which operates on the same coloured squares as most of one's pawns are
placed. The relation Fw contains this situation and expresses both the fact
that the (bad) B is protecting the pawns and the fact that the pawns are
restricting the value of dim B, and consequently of its connectivity with
other pieces. The malaise is acute when the pawns are relatively fixed in
positions, being interlocked with the enemy pawns or blocked by pieces. In
such situations the tactical play must be designed to "free" the position,
to change one's bad B into a good B.
In the absence of obstacles on the board the maximum values of dim X are
given in the list below:
max dim P = 1, over all possible modes;
max dim N = 7 , occurs when N is within the square;
bounded by S--43, 46, 22, 19;
max dim B = 1 2 occurs when B is on S=36, 37, 29, 28;
max dim R = 1 3 occurs when R is on any S;
max dim Q = 2 6 occurs when Q is on S=36, 37, 29, 28.
We notice immediately the importance of the central squares 36, 37, 29, 28.
These have always played an important part in opening theory, since control
of these squares gives either player the opportunity of placing his pieces to
maximum effect. Whether or not the centre should be occupied by pawns
in the first instance is a tactical matter; in contesting that earlier golden rule
the hypermodern school were making just this point about the wider posi-
tional significance of the squares 36, 37, 29, 28. For example, the Reti
opening, commencing with (1) N-f3 and (2) c2-c4 amounts to a rapid fight
for control of 28, 37, 36 without occupying 28, 29 with pawns. A later
fianchetto of the KB by B-g2 increases this pressure on the diagonal central
squares 29, 36.
We can also appreciate that school of thought which has claimed that all
opening play should be dedicated to the task of bringing the Rooks into play;
castling early is a contribution to this and, of course, the classical King's
Gambit opening--with an early sacrifice of the KBP--is patently just such
a play. Perhaps the style of Morphy exemplified this urge to create open files
for the Rooks, aboveallother players. Thus Morphy frequently found himself
with a greater command of the board (or of the most significant part of it)
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 351

than his opponent, with pieces highly connected h7 the complex K x(S) and
squares (particularly those in the vicinity of the enemy) highly connected in the
conjugate Ks(X). Bringing his Rooks into play so that they enjoyed their
maximum 'dimensions as simplices was often the final touch of disaster for
his opponent. Now from section 2.1 we notice that in mode [0,0] the two
Rooks are quite disconnected ( q = - - l ) from all other men in Kx(S). Thus
arranging for a Rook to move into a more central position will ensure, in an
early mode, that it becomes connected in Kx(S). Also in the middle game
modes the penetration of the Rook to the enemy's seventh (or eighth) rank--
a strategy stressed by Nimzowitsch--is more likely to give a maximum value
for dim R, since often the enemy pawns have advanced from the seventh.
Since the squares attacked by the R, on the seventh rank, also include signifi-
cant squares in the vicinity of the enemy King, we expect this invasion to be of
great tactical as well as positional value. -

2.3. Checkmate
If we consider an easy checkmate position, viz., White's Q ot~ f8 ( S : 6 2 ) ,
K on g6 ( S : 4 7 ) , and Black's K on h8 ( S : 6 4 ) , then the complexes Kx(S)
and Ks(X) for both relations Fw and F~ are easily found.
Kx(S; Fw)-- White's view of Board:
dim Q : 2 0 and dim K = 7
together with a q-connection between Q and K of q = 4 :
precisely Q : < 5 7 , 58, 59, 60, 61, 63, 64, 55, 48, 46, 38, 30, 22, 14, 6, 63, 44,
36, 26, 17>
and K:<54, 55, 56, 46, 38, 39, 40, 4 8 >
and Q . K : < 5 4 , 46, 38, 55, 4 8 > n
Ks(X;Fw)--Board's view of White:
each of the squares 54, 46, 38, 55, 48 is a 1-simplex < Q , K > the
remaining squares are either 0-simplices or not members of F-lw; in
particular 64= < Q > , 63 =- < Q > , 5 5 : < Q , K > , 56= < K > .
Kx(S; FB)--Black's view of Board
dim K = 2 ;
precisely K = <63, 55, 56>.
Ks(X; F -1B)--Board's view of Black
each of 63, 55, 56 is a 0-simplex < K > , the rest are not members
of F-~B.
352 R.H. ATKIN

Black is in check because his K is on square S = 6 4 which is a vertex of


White's Q in Kx(S;Fw). The only possible moves for the Black K are the
vertices of K in Kx(S;FB), viz., S = 6 3 , 55, 56. But the squares 55, 56 are
vertices in Kx(S;Fw) viz. in the simplex K, whilst the vertex 63 is in the
simplex Q of the same complex. Hence Black's K has no legal move and is
checkmated--a typical player's assessment of the position, since as players
we are inclined to think in terms of the complexes Kx(S). But we see now
that the Board's view of White/Black contains the immediate checkmate
condition, viz.
Board's view of Black's K is a subcomplex of Board's view of White.
The notation naturally allows us to denote Board's view of Black's K by
the subcomplex Ks(K;F-1B). In this particular illustration we happen to
have Ks(K; F -~ B)= Ks(X; F -1B) since Black only has one piece. But generally,
in other modes, we have
Ks(K; V-1B) Ks(X; V -1 B)
and the checkmate condition is now precisely expressed as follows.
Black is checkmated in mode [I,(I-1)] when the following structure is invariant
under all legal transitions to mode [1,1].
O) Ks(K;F-1B) c Ks(X;F-lw);
(ii) with the Black K on square S K.
< S K > ~ Ks(X;F-lw).
We notice that this is an economical view of the checkmate condition
and is a characteristic of the definitions of the relations Fw, FB. Thus, when
the condition applies, the Black K cannot move without being in check and
it cannot capture the checking piece because the vertices of X do not include
the location-square of X. Furthermore, intervention by any other of the Black
pieces is ineffective because of the coupling of both modes [1,(I-1)] and [1,1]
to the structure.
We can now express, in general terms, the course of a game as the develop-
ment of the complexes K(Fw) and K(FB), each trying to engulf pieces of (or
the whole of) the other. It is rather like two giant amoebae in multi-dimen-
sional space trying to devour each other; each has the ability to grow in
unusual ways.
A chess master normally subordinates his tactical moves to overall posi-
tional requirements. When he looks ahead through n moves he is consciously
or unconsciously assessing the future structures of all four complexes defined
by Fw and FB. His tactical analysis only ceases, for some particular n,
MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 353

when he believes the nth structures to be in his favour. Often this belief is
based on an intuition developed through years of experience; always it
must require an intuitive appreciation of multi-dimensional connectivity
patterns; it is not based on linear (network) type structures, except perhaps
in the strictly book-.keeping sense of tracking the modes from [ll,J~) to
[In,In].

2.4. Chess Time


The idea of time on the chessboard (players speak of gain or loss of tempi)
is an intriguing one. It i s discussed at some length by Zno sko-Borovsky (1938),
who builds it into an overall description of positional judgement. Thus it is
plausible to think that if White plays (1) e2-e3 and (2) e3-e4 then in some
sense he has wasted a tempo, since he could have achieved the same position
by (1) e2-e4: Inexperienced players commonly bring the Q out into the centre
of the board in the early stages of the game (for example, by playing the
Centre Counter Opening). This leads to tactical moves by the opponentwhich
force the Q to retire again--more loss of tempi, leaving one player with a
superior development (gain of tempi). The end result can be a frustrated loser
complaining, " I f I had only had one more move I would have been able to
force mate !".
This idea of tempo is dearly a relation between structure of the game and
mode [I,J], between K(Fw), K(FB) and [I,J]. Thus if White truly wastes a
move by (1) e2-e3 and (2) e3-e4 it means that the structure of the game (for
White) in mode [2,1] could have been achieved in mode [1,0]. This situation
is felt most acutely in the opening stages of the game before the complexes
Ks(X; F-lw) and Ks(X;F-1B) are connected. At this stage it is possible for
White truly to waste a tempo since he is effectively playing without any
immediate interaction with Black. Later on, when the forces are obviously
joined in combat it is not easy for White to produce identical structures in
[Iz,J~] by different routes from [I1,J1). We notice this in chess writers who
rightly are unable to sustain an interest in the tempo-situation once the game
is reaching certain kinds of middle-game structures. It is not therefore sur-
prising that the tempo-question arises again in the end-game phase, when
most of the pieces and pawns are cleared off the board and the structures
Ks(X;F-lw) and Ks(X;F-IB) are (often) disjoint, or connected only slightly.
Many tactical possibilities arise in end-game modes for a player deliberately
to lose a tempo in order to recover a certain structure with (or without) the
obligation to move. This effectively means that Black is forced to gain a
tempo vis-d-vis some specific structure--which is advantageous.
354 R.H. ATKIN

Thus in any event the gain or loss of time is only good or bad according
to the achievement of desirable structures. If we consider two structural
patterns K, Kll(Kdenotes Kx(S; Fw) u K s ( X ; F -1 w)WKx(S; FB)U Ks(X; F -~ r~))
corresponding to modes [I,J] and [P,J~], then it is reasonable to describe
White's time with respect to (K, K 1) as P - I and Black's time with respect to
(K,K 1) as j t _ j .
Since K and K 1 both involve four complexes we see that juggling with
White's time vis-d-vis Black's time is only part of the struggle to achieve the
structure K ~ with the right to move.

3. The Immortal Game

A classic game of chess, known in fact as The Immortal Game, was that
played between Anderssen and Kierseritsky in London (1851). It was a
demonstration of Anderssen's remarkable combinational powers and
tactical foresight, but it showed too that his instinct for the positional structure
was very highly developed. This latter quality of Anderssen's play was re-
garded as largely magical in his time (20 years before Steinitz), but without it
he would not have been able to produce such elegant and economical tactical
play against his adversaries. In 1858 Anderssen was to-be soundly defeated
in a match with the young Paul Morphy, his equal in tactical imagination
but his superior in positional instinct.
The analysis of the game which we now give is based on a Q-analysis
produced by the writer's computer program which was run on the PDP-10
at th= University of Essex. The program was written in BASYS, an extension
of B: ,qIC (BASYS was devised by Brian Gaines and largely developed by
Peter ?'acey), and operates on a Data file which contains the score of the
game -a the international algebraic notation. After reading a move the
program sets up the incidence matrices corresponding to the relations
Fw, F-~w, FB, F-1B and then systematically computes and prints the q-
connectivities of the complexes. The four compleces are labelled, for a
particular mode [/,J];

Kx(S;Fw) --White's view of Board;

Ks(X;F-lw)--Board's view of White;

Kx(S;FB) --Black's view of Board;

Ks(X;/-,-1 B) --Board's view of Black.


MULTI-DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 355

The score of the game is as follows:


Anderssen v. Kieseritsky (London, 1851).
(1) e2-e4 : e7-e5 (9) N-f5 : c7-c6 (16) N-c3 : B-c5
(2) f2-f4 : e5*f4 (10) g2-g4 : N-f6 (17) N-d5 : Q'b2
(3) B-c4 : Q - h 4 + (11) R-gl :c6*b5 (18) B-d6 :Q'a1+
(4) K - f l : b7-b5 (12) h2-h4 : Q-g6 (19) K-e2 : B*gl
(5) B'b5 : N-f6 (13) h4-h5 : Q-g5 (20) e4-e5 : N-a6
(6) N-f3 : Q-la6 (14) Q-f3 :N-g8 (21) Neg7+ : K-d8
(7) d2-d3 :N-h5 (15) B'f4 : Q-f6 (22) Q-f6+ :RESIGNS
(8) N-h4 :Q-g5
(The symbol for "captures" is the computer's multiplicative sign *)

Blacl~
N !N,N
NtNtNt

N N
N NeN N
White
FIo, 3. Anderssen v, Kieseritsky M o d e [12;12]

The opening play was a King's Gambit, in the spirit of the times, being an
early attempt to create space for White's pieces and in particular for the
White K R after the f-file is opened. Black accepts the gambit pawn, brings
his Q out for an early check--which loosens the pawn barrier around White's
K - - a n d cannot resist capturing the White KB (by 11 . . . . : c6"b5) which
Anderssen deliberately leaves en prise [by (11) R-gl]. The justification for
White's tactical play lies entirely in the structure of the game and the subse-
quent middle-game possibilities. The character of the game is essentially
manifest by the time it is in mode [12,11]. To investigate this character we
therefore commence our analysis in mode [12,12].
356 R.H. ATKIN

Game in mode [12,12], following ( 1 2 ) . . . : Q-g6


In White's view of Board we have the structure vector
8 0
Q={1, 2, 2, 2, 4, 5, 6, 9, 5}
This corresponds t o d i m Q - - 8 - - i n f a c t Q - - < 3 , 11, 12, 20, 5, 6, 13, 22, 3 1 > -
and includes dim K N = 7 , dim K = d i m K R = 4 , dim QB=3.
The White pieces (excluding the pawns) are not connected until we reach
q = 1 and then the only pieces which are so connected are Q, K, KR. This is
illustrated by the value Q~--9. We notice too that the complex consists of five
separate components at the zero level (q----0).
In Black's view of Board we have the structure vector

8 0
Q={1, 2, 2, 3, 4, 4, 6, 9, 5}
a pattern very similar to White's. But the difference between the positions of
White and Black appear in the conjugate complexes.
Thus, already we see that
Ks (X; /"--lw) ('~ Ks(X;F-1B)@~; in fact g 7 = < K B , Q > in Ks(X;F-1B)
and g7= < K N > in Ks(X;-lw); e7= < K , K B > in Ks(X;F-IB)
and e7= < K N > in Ks(X;F-lw).
Squares which are deep inside the Black camp, and close to the K, are
already part of the White structure, either as simplices in Ks(X; F -x w) or as
vertices in Kx(S;Fw). By comparison the White K simplex (in Kx(S;Fw) is
not connected to the Black structure. Furthermore the Black Q is under
attack by the White pawns; this is a dangerous situation for Black because,
in avoiding further harassment Black must use moves (time) which should
be available for strengthening his own structure. In effect, because of the
positions of the K-side pawns and the KN, the Black Q acts as if it were a
3-simplex, viz., Q = < 4 8 , 40, 39, 31>. But each of these vertices is
in Ks(X;F-lw), that is to say, the Q cannot move without being captured.

Game in mode [13,12], following (13) h4-h5


Black cannot allow the exchange of the Q (a potentially powerful piece
with high values of dim Q) for a White pawn (or a R or a N in this position).
Hence the Q is driven on to g5; the structure vectors remain almost unaltered,
but another square (g6) is attached to Ks(X;F-lw).
M U L T I - D I M E N S I O N A L S T R U C T U R E I N T H E GAME OF CHESS 357

Game in mode <13,13>, following ( 1 3 ) . . . : Q-g5


The future possibilities for Black have diminished by this move; even
when the KN has moved the value of dim Q can only increase by 2, whereas
were the Q still on g6 the increase would have been 5. This cramping effect is
manifest in an increase in the structure vector (and therefore in the obstruction
vector) for Black. In this mode, in Kx(S;FB), we have Qo--7 compared with
Q~=6 in mode [12,12], the other values of Qr being unchanged. An advantage
for Black lies in the fact that the Q now defends the pawn on f5, but White's
next move makes this square into a 1-simplex, f5---<QB,Q>, in
Ks( X ;F-lw).

Game in mode [14,13], following (14) Q-f3


This has increased the value of dim Q from 8 to 12, in Kx(S;Fw), the
new structure vector in White's view of Board being

12 0
Q=(1, 1, 1, 1, 1,2,2,2,3,4,4,7, 5}
In Black's view of Board the max dim X is still q=8, and Q2--7 (ef. Q2=4
for White) although Qo=4. We see also that at the q = 1 level there has been
an improvement for White (Q1---7 against the previous Q1=9). Thus we
have a tangible sign via the structure vector that White's game has become
more connected (at q----1 and q----2), Black's not so, and the increase in the
max dim X for White shows that his game is more open, flexible. For example,
apart from the threat of B'f5, White has the tactical possibility of e4--e5,
putting dim Q up to 16 and threatening e5*N and Q*R.

Game in mode [14,14], following ( 1 4 ) . . . : N-g8


This saves the Q by increasing dim Q from 8 to 10, the new structure vector
for Black being

10
Q={1, 1, 1, 1, 1, 2, 3, 3, 5, 8, 6}
But Q0 is now 6, due to the retreat of the KN. White will now be able to
increase the value of dim QB and to lower the value of Q2 yet again. The
development of the QB will bind even more squares to Ks(X;F-lw) and, in
particular, make d6 a 1-simplex in this complex and c7 a 0-simplex: we notice
thatin this mode c7 6 Ks(X;F-1B).
358 R.H. ATKIN

Game in mode [15,15] following (15) B'f5 : Q-f6


Blacks' structure vector is now
17 5
Q = ( 1 , 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 3,4,4,4)
whilst White's is
12
Q={1, 1, 2, 2, 2, 3, 3, 3, 4, 3, 3, 7, 5}
This means that Black's game is being played almost entirely by his Q, the
latter only being connected (via square e7) at q=2. At this level (q-----2)
there are four components in Kx(S;FB) and these are defined by the Black
pieces as follows:
{O, B, KB}, {ON}, {QB}, {KN).
At the q = 1 level the pieces fall into 3 components given by
{O, K, KB, QN, QB, KN), {QR), {KR}.
But the squares common to the conjugate complexes Ks(X;F-lw) and
Ks(X; F-1B) are the folowing:
b8, d6, e5, e7, f5, gS, g6, g7, h4, h6
whereas the squares c7, d5 are not in Ks(X;F-1B) but are in Ks(X;F-lw).
Thus Black is heavily on the defensive and his pieces are spread about in a
disconnected way. Kieseritsky now must make every effort to develop his
pieces so as to contest key squares and to build up his own complexes again.

Game in mode [16,16], following (16) N-c3 : B-c5


White's structure vector is now
1'2 1
Q={1, 1, 2, 2, 2, 4, 4, 5, 6, 5, 2, 6=1 q-5, 3}
and Black's is
16 10 1
Q={1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 2, 4, 4=3q-1, 3}.
But at q----1 all of White's pieces (excluding pawns) are connected in one
component whereas Black's pieces form three components. At this level
(q----1) connection means that at least two pieces control (or attack) the same
square. If all the pieces (seven in this case) fall in the same component then
M U L T I - D I M E N S I O N A L S T R U C T U R E IN T H E G A M E OF CHESS 359

there is greater control over a wider portion of the Board. At a higher level
(e.g., q = 7 ) the pieces are exerting their (possibly) maximum effect in a
disconnected manner. In this case, in White's view of Board, Q7=4 and the
four components are:
{QN}, {QB}, {Q}, {KN}.
The level of q = 2 can be seen as intermediate in a crucial manner (reviving
Nimzowitsch's idea of "over-protection".?) between the critical q = l and
the higher levels. In White's case we have Q~=2, viz.
{QR, QN, Q, K, KR} and {QB, KN}.
In Black's view of Board the Q (dim Q = 1 6 ) and KB (dim K B = I 0 ) are in
splendid isolation until we reach the low level of q = 3 . Black's pieces are
relatively disorganized therefore, and technically disconnected. Add to this
that White's move N-c3 further attacks d5, from where the QN can attack
the Q on f6, and we can see that the complex Ks(X;F-lw) is steadily connect-
ing with the central squares around the Black K.
Black
"% y,.7/Mk ' " ' '~
~',,m, Ai - - ' . . . - "2~kk'A ~///////, ~ . ; ~ ;

~/~ ~ ~

N
~-- - ~ z,,~,,. ~,:., -,~

White
F~o. 4. Anderssenv. Kieseritsky Mode [t7,171

Game in mode [17,17], following (17) N-d5 : Q ' b 2


The structure vector for White is now
12 0
Q={1, 1,2, 2, 2,4,4, 5,6, 5,4, 5 = 1 + 4 , 4 }
and for Black,
13 10 0
Q = { 1 , 1, 1 , 2 , 2 , 2 , 2 , 2 , 2 , 3 , 3 , 5 , 8 = 5 + 3 , 3 }
360 R.H. ATKIN

At q = 1 all seven of White's pieces fall into 1 component whereas Black's


pieces fall into five components. Thus Black's "attacking" play via Q'b2 has
thrown his own piece structure into disarray. Even at the q = 0 level Black's
pieces form two distinct components. His Q foray has also reduced the
connection to his own Ks(X;I"-IB) of the squares d6, d8, e7. These squares
are consequently the better attached to Ks(X; F -1 w), the more so since White's
move N-d5 has further connected squares c7, e7. For his part Black no
doubt thinks that he is threatening to capture either or both of White's Rooks,
but at the price of neglecting his own important squares around the K.

Game in mode [19,18], following (18) B-d6 : Q*al-]-, (19) K-e2


White is offering both Rooks, of course, by (18) B-d6, but in return he
has altered the structure so that f 7 = < Q N , QB, K N > in Ks(X;F-lw) and
f 7 = < K , K N > (only) in Ks(X;F-IB). Thus the Black KB is cut off from
the square f7. Tactically all is lost; if Black rejects the capture of the K R
and plays (18) . . . : B'd6 then (19) N*d6q- : K-d8, (20) N*fTq- : K-e8,
(21) N-d6q- : K-d8 (22) Q-f8 mate. The structure vectors are now
12 1 0
for White: Q = ( 1 , 1, 2, 2, 2, 5, 5, 5, 5, 3, 3, 5 = 1 +4, 3 = 1 +2}
18 1
for Black: Q={1, 1, 1, 1, 1, 1,2,2,3,3,6, 1 0 = 7 + 3 , 2 = 1 + 1 }
White's pieces form one component at q = l , whilst Black's form seven
components. Also the high dimensions of White's pieces (Q7=5) dominate
those of Black's (Q7= 1). But Black cannot resist capturing the other Rook.

Game in mode [20,20], following ( 1 9 ) . . . : B*gl, (20) e4-e5 : N-a6


White's pawn move has finally disconnected the Black Q from square g7,
allowing a fatal check by the N on f5. Black's move N-a6 was an attempt to
contest square c7, but this is far too late. The flight squares for the Black K,
f8, e7, d8, are doomed to be absorbed into the White structure Ks(X;F-~w)
without redress by Black.
The structure vectors show the same pattern; B*gl has actually increased
the number of components at q = 1 (piece-wise) to eight (the maximum
number.)

Game in mode [22,21], following N*g7q- : K--d8, (22) Q-f6-k : RESIGNS


Black can play ( 2 2 ) . . . : N ' f 6 but then he reduces e7 to a 0-simplex only,
viz. e 7 = < K > , and then (23) B-e7 mate; cannot be stopped. Thus Black,
very wisely, resigned--just in time to avoid mate.
MULTI DIMENSIONAL STRUCTURE IN THE GAME OF CHESS 361

4. Computer Player
The above analysis of the game Anderssen-Kieseritsky (1851) in its middle-
game modes brings out the role of the connectivity pattern in Q-space for
both White and Black. The structure vector Q contains much of this pattern,
in the first instance. Thus we can see that the level q = 1 is crucially important
for the pieces, and that a first sign of this is a change in the q = 2 level. Also
the values of Qt for t > 3 are an indication of piece development and board
control. High values of (say) Q5 or Qe indicate future possibilities of low
values of Qo and Q1. The dimensions of the squares in either Ks(X;F-lw) or
Ks(X;F-1Q also indicate the contest for control over important parts of
the board.
All this would suggest that there is a promising future for computer-chess
via analyses of structure patterns. The language of connectivity in a multi-
dimensional Q-space seems to be naturally suited to a discussion of the
game of chess. Furthermore, it seems to reflect all the niceties and general
truths of the modern sum of chess theory. Surely it therefore possesses the
ability to express more accurately the way that chess masters actually think
about the game. By "think" we include, of course, what nowadays often
passes for "instinctive feeling" or "subconscious appreciation".
A learning-machine faced with the game of chess should therefore be able
to acquire a chess mastership which is so far denied to the species. Such a
project must hopefully throw new light on the structure of the mental pro-
cesses and the capacities of the human mind.
Much work has already been done in teaching computers to play chess,
since an initial stimulus from C. Shannon in an address to the (American)
National Institute of Radio Engineers in 1949. An interesting demonstration
of the computer-player's ability to conduct a tactical attack is published in
the British Chess Magazine of July 1971. This reprints the score of a short
tactical skirmish between A. J. Roycroft and the CDC 6600 "Chess 3.0"
which took place at the December 1970 Islington Congress. Roycroft, who
played Black, was clearly playing for the amusement of the onlookers and
at great speed, and after good-naturedly throwing away a pawn and a knight
he resigned; "Chess 3.0" had won.
It is in the tactics of formulating the attack that the chess computer pro-
grams seem to have been so far largely concentrated. The Russian ex-world
champion M. M. Botvinnick has led a strong line of research in this direction
for some time (Botvinnick, 1971), aiming at an increasing depth of analysis
in the so-called tree of analysis (the network which links possible transfor-
mations from mode [/,J] to mode [11,ji]). Hismethod is based on identifying
362 R.H. ATKIN

lines of attack (and of defence) against a target (an enemy piece) by way of
the squares over which the attack can take place. By defining suitable func-
tions (with values in the complex domain) on these squares, which take
account of the action of all relevant pieces, he is able to obtain numerical
comparisons for the potential successes of White or Black. The analysis is
elegantly conceived and mathematically suited to computer programming.
We can see from our analysis, however, that Botvinnick's mappings are
defined on the squares in Ks(X;F-aw) or Ks(X;F-1B) and are therefore
cosimplices (not always 0-cosimplices, as the Russian analysis implies).
The connectivity patterns of these complexes should therefore be reflected
in the interaction of the various functions and consequently affect the
judgement involved in choosing the dynamical tree of analysis. Botvinnick's
treatment seems to be particularly primitive (his own word) in assessing the
positional features, and this is the area where our connectivity structure
seems to be most powerful.

References
ATKIN, R. H. (1972). From cohomology in physics to q-connectivity in social
science. Int. J. Man-machine Studies, 4, 139.
BOTVlNNICK, M. M. (1971). Computers Chess, and Long-range Planning. London:
Longman; (1970) New York, Berlin: Springer-Verlag.
LASKER, E. (1932). A Manual of Chess. London: Constable; (1947) New York:
Dover.
RETI, R. (1933). Masters of the Chess Board. London: Bell (reprinted 1953).
ZNOSI(O-BoRovsKY,E. (1938). The Middle Game in Chess. London: Bell (reprinted
1953).

You might also like