Professional Documents
Culture Documents
R. H. ATKIN
Department of Mathematic& University of Essex,
Colchester, Essex, U.K.
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.
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).
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
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 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
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
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
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].
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.
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];
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
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.
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.
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
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
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).