Special Train Algebras Arising in Genetics: by H. Gonshor
Special Train Algebras Arising in Genetics: by H. Gonshor
Special Train Algebras Arising in Genetics: by H. Gonshor
by H. GONSHOR
(Received 18th January 1959)
(Received in revised form 22nd December 1959)
1. Introduction
Ever since Mendel promulgated his famous laws, probability theory and
statistics have played an important role in the study of heredity (9). Etherington
introduced some concepts of modern algebra when he showed how a non-
associative algebra can be made to correspond to a given genetic system (1, 4).
The fact that many of these algebras have common properties has led to their
study from a purely abstract point of view (2, 3, 5, 6, 11, 12). Furthermore,
the techniques of algebra give new ways of attacking problems in genetics
such as that of stability.
We shall prove a stability theorem on special train algebras and apply it
to several genetic systems.
Similar theorems may be proved in cases where several of the A's are equal
to i. The case where Xt = \ and all other A's are distinct from \ occurs
frequently in genetics. By the same technique as before it is easy to show
that non-zero idempotents exist only if xOoi = 0 and in that case a one para-
meter family of idempotents exists. (In fact, xx may be chosen at will, and
the other x's are uniquely determined as before.) Similarly if r of the A's, and
no more, are equal to \, it can happen that there is an r-parameter family of
idempotents. This occurs in examples connected with multiple alleles and
with linkage (3).
Henceforth, we shall consider algebras over the reals. We may then define
n n
Y xmiam-+ Y xmamoxmi-+xm for all x. It is clear that convergence defined
m = 0 m= 0
this way is independent of the basis used. In order to state the next theorem
it will be convenient to define two concepts. The " weight" of an element
is its image in the unique homomorphism of the algebra onto the reals. The
" plenary powers " of an element A are defined as Ao = A, An+l = A*.
Theorem 2.2. The sequence of plenary powers of an element of weight 1
in a special train algebra whose train roots other than Ao = 1 all have absolute
value less than -J tends to an idempotent.
n
Proof. Suppose At = V xmiam. It is required to show that xmi-*xm
m = 0
where xm is defined as in Theorem 2.1. It is clear that xol = 1 for all i. Assum-
ing xu-*xu x2i->x2, ..., xmi-+xm, we shall show that xm+u ;->-xm+1. As in
the proof of Theorem 2.1, xm+u i+1 = 2Xm+1xm+u t+f(xu, ..., xmi) where/
is a function depending only on the multiplication table of the algebra. Since
/ i s a quadratic function f(xt;, ..., xmi)-»/(xi, ..., xm). The result now reduces
to the following theorem in analysis. Let («„) and (bn) be two sequences satisfying
SPECIAL TRAIN ALGEBRAS ARISING I N GENETICS 43
an+1 = Xan+bn for some A with | A | < 1 and let bn-+b; then (an) converges to
the limit a determined by a = Aa+b.
Write an = a + Mn, &„ = b+yn. We have w n+1 = kun+yn, where | A | < 1
and yn-*0, and have to prove that Mn->0. Since
if we define zn = Xn~1yx + Xn~2y2 + ... + yn, it will suffice to show that zn-»0.
In this form the result is contained in Knopp ((10), p. 76, 9(b)).
The theorem may be extended to cases where several of the A's are equal
to i. If Aj only = \, then the theorem is still valid if x001 = 0; otherwise the
sequence xt,- is an arithmetic progression which certainly does not converge.
3. Mutation Algebras
If the simplest case is considered with mutation rates r and s, the gametic
algebra has basis D and R and the multiplication table:
D2 = ( 1 - r)D + rR, R2 = sD + (l-s)R,
DR = ${(l-r)D + rR} + i{sD + (l-s)R} = *(1 -r + s)D + #l -s + r)R.
It can be verified by direct computation that if a and b are chosen as a
basis where a = D and b = D—R, the table takes on the simplified form:
a2 = a-rb, ab = \{l-r-s)b, b2 = 0.
Thus the algebra is a special train algebra. (Note that the basis is in-
dependent of r and s.) Theorems 2.1 and 2.2 can therefore be applied. By
theorem 2.1 there is a unique non-zero idempotent unless r+s = 0. By
theorem 2.2 the plenary powers of an element of weight 1 approach this
idempotent for 0<r + s<2.
Although the algebra is defined for all real r and s, the corresponding
genetic system has meaning only for O^r^l, O^s^l. Since it is known that
in the case r = s = 0 (no mutations) stability is achieved in the second genera-
tion, this proves that the gene distribution approaches a stable distribution
unless possibly when r — s = 1. This case is clearly unstable since D2 = R
andi? 2 = D. More generally, the plenary powers of ocD+fiR, where a+/? = 1,
oscillate between aD+pR and fiD+<xR, only the distribution \D+%R being
one of unstable equilibrium.
The zygotic algebra has the table
D2 = (l-r)2D + 2(l-r)rH + r2R, R2 = s2 D + 2(1 -s)sH + (1 -s)2R,
DH = i(l-r)(l-
RH = $s(l
DR = (l-r)sD + (l-r-s+2rs)H+(l-s)rR,
2 2
H = i(l-
If a, b, and c are chosen as a basis where a = D, b = D—H, and
44 H. GONSHOR
c = D—2H+R we obtain:
a2 = a - 2rb+r2c, ab = i(l-r- s)(b - re),
2
b = i(l-/— sfc, ac = bc = c2 = 0.
Thus this is also a special train algebra.
4. Sex Linkage
We shall use the subscripts 1 and 2 for the haploid (usually male) and
diploid sex respectively. The basis elements, representing genotypes, will
thus be called Du Ru D2, H2, R2. Since members of the same sex do not
produce offspring it is convenient to define the product of any two basis
elements with the same subscript as zero. The rest of the table is:
DtD2 =
The algebra is certainly not a special train algebra. In fact, the existence
of a basis consisting entirely of nilpotent elements makes it clear that no
non-trivial homomorphism into the reals exists (i.e. the algebra is not baric
in the terminology of (1)).
A simplification can be effected by using as a basis Du a = D1-\-D2,
b = Dt-Ru c = D2-R2, d = D2-2H2+R2. The table then becomes
D\ = 0, aD1 = \a, bDt = 0, cDx = ib+ic+$d, dDt = 0,
a2 = a,ab = \c+\d, ac = ib+$c+ld, ad = 0, b2 = 0, be = ±d,
bd = 0, c2 = 0, cd = 0, d2 = 0.
It follows immediately from the table that the subspace {a, b, c, d) is an
ideal containing the product of any two elements in the original ring. This
ideal is the set of all elements for which the sums of the coefficients of the
original basis elements of subscripts one and two are equal. (The genetic
significance of the ideal is that it contains all the elements which represent
populations.) Furthermore, the ideal is a special train algebra. This can be
more easily seen if b+c is used as a basis element instead of c. It is interesting
to note that the powers of the ideal {b, c, d}, ({b, c, d}2 = d, {b, c, d}3 = O)
are ideals even in the original ring.
The ideas in § 2 may be used to show that the elements of the form
a+yb+yc+(y2+y)d represent all the non-zero idempotents and that the
plenary powers of an element of weight one approach an idempotent. The
elements A which are non-zero and satisfy (A2)2 = A2 are precisely those of
the form a+yb+yc+xd where x and y are arbitrary.
The genetic significance of these results can be made clear by expressing
them in terms of the original basis. By letting y = — /? and a = 1 — /? we find
that the idempotents are exactly all those elements <xD1+PR1 + u2D2-\-
2aPH2+02R2 with a + 0 = 1.
SPECIAL TRAIN ALGEBRAS ARISING IN GENETICS 45
It clearly follows that the elements which satisfy (A2)2 = A2 are precisely
those of the form
where y = x-(y2+y).
A more intuitive feeling for this result can be grasped if it is noted that
y(D2 — 2H2 + R2) is the most general element which can be added to a genotype
distribution which will not change the distribution of gametes produced. For
a more classical approach see (9).
+ r(l+r-s)R2],
6. Polyploidy
We shall first consider the gametic n-ploidy algebra. The types will be
denoted by £>„, Dn_u ..., Do where Dx stands for the type which has / dominant
genes and n—i recessive genes. The multiplication table is:
\j . = o \ i J\ n-i
where ( I stands for the usual binomial coefficient. (It is convenient to use
46 H. GONSHOR
Lemma 6.1.
= 0 for n<s^2n,
_. {2n\fn\ /2n\/2n-s\ , . .
Since = this becomes
\nj\sj \sj\n-sj
(Note the convenience of using the symbol I I for all integers n and r. Thus
W
the proof is valid even when s>n in which case the last line reduces to 0.)
s / \
Lemma 6.2. If cs is defined as V ( - 1 ) ' )Dn _, for 0 ^ j ^ n, then
«= o \»7
= 0 for i+j>n.
Proof. By comparison with the binomial theorem it is easily seen that
where P, is defined as in Lemma 6.1. The result now follows from Lemma 6.1.
As an immediate consequence of Lemma 6.2 we obtain the following:
Theorem 6.1. The gametic n-ploidy algebra is a special train algebra with
train roots I I I I for 0 ^ / g « . Since l0 = 1, kt = % and all other A's
are less than \, the remarks in § 2 may be applied to this case.
When n = 2 this reduces to the well-known case of the tetraploidy gametic
algebra with table
Co = c 0 , c o C l = icu c0c2 = c\ = \c2, exc2 = c\ = 0 (1).
By applying the technique of duplication (5) we can now consider the
zygotic algebra. We shall identify ctCj with fixed i+j. This corresponds to
identifying all DaDb with fixed a+b which in turn corresponds genetically to
disregarding the nature of the contributions of each gamete to the zygote.
It will be convenient to use the symbol dt to stand for any product cacb with
a+b = i. d0, du ...d2n is then a basis of the algebra.
Theorem 6.2. The zygotic n-ploidy algebra is a special train algebra with
basis elements dh 0^i^2n and multiplication table:
r(:)(H>< ••«»•
djij = 0 otherwise.
Proof. The table follows immediately from the process of duplication
48 H. GONSHOR
and implies the special train property. The remarks in § 2 remain valid in
the zygotic case.
When n = 2 the table becomes d% = d0, dodt = \du d\ = \d2, d0d2 = \d2,
d\d2 = TT^3> d\ = Ttd^ with all other products zero.
Lemma 6.3.
t
i = 0
t
r= 0
fz (
|s = r \r
r=Os=r«=O
Now
In other words the relationship expressing the D's in terms of the c's is the
same as the relationship expressing the c's in terms of the D's. For example
in the simplest case where c0 = D2, c t = D2 — D± and c2 = D2 — 2Dl + Do
we have D2 = c0, D^ = co — cu and Do = CQ—IC^ + C^
n
The special case where xt^0 for all i and V xt = 1 corresponds to a
i= 0
/ «
distribution of gametic types. I V xiDn-i corresponds to the distribution
V
V o
in which xt is the fraction of types which have exactly / recessive genes.) If
a package of i genes is chosen at random from one individual of the population,
the probability of all genes being recessive is
SPECIAL TRAIN ALGEBRAS ARISING IN GENETICS 49
If the probability is denoted by yh then it follows from Lemma 6.3 that
i=0
i i 0
i=0
t ( \lj
\lj
Theorem 6.3. A necessary and sufficient condition for a non-zero element
to be idempotent is that y{ = y\ for all i satisfying 0 ^ / ^ n . Otherwise no
plenary power of the element is idempotent.
Proof. We first prove the sufficiency of the first statement.
and
,?. <-"' (:) '*J -,i. k <-"'" CX") '•"*
' ~n W "
K'+J) V+.
the coefficient of ck in the product becomes
E.M.S.—D
50 H. GONSHOR
as well as the analogue of Lemma 6.3. Furthermore we have:
Theorem 6.4. A necessary and sufficient condition for a non-zero element
to be an idempotent is that yt = y\ for all i satisfying 0^i^2n. If yx = y\
for all i satisfying 0 ^ / ^ n then the element satisfies {A1)2 = A2. Otherwise
no plenary power of the element is idempotent.
Proof. As in the proof of Theorem 6.3 we consider
k-it
The rest of the proof is similar to the gametic case except for the consideration
of the property (A2)2 = A2.
Since the coefficient of dt in a product does not depend on the coefficient
of dj for j>i, the condition yt = y\ for O^i^n will at least guarantee that
the coefficient of dt for i^n will be the same for A and A2. Since dtdj = 0
for i>n it follows that A2 depends only on the coefficients of dt for i^n in A.
Hence A2 = (A2)2.
As before it may be noted that
i.e.
T(cw) = ( l - r - s ) w c w - f " W
)rcw+1+ (" W
J r 2 c w+2 ... .
Now that we have found T exphcitly in terms of the c's we can write down
the multiplication table.
Theorem 7.1. The multiplication table of the n-ploidy gametic mutation
algebra is
for i + j g n ,
0 otherwise.
E.M.S.—D 2
52 H. GONSHOR
A similar result is obtained for the zygotic algebra.
Theorem 7.2. The multiplication table of the n-ploidy zygotic mutation
algebra is
= 0 otherwise.
The tables in § 3 may be seen to be a special case of this. When n = 2
the table in the zygotic case becomes
3
d^), d\ = \{\-r-s)2(d2
3-r
MATHEMATICS DEPARTMENT
RUTGERS UNIVERSITY
NEW BRUNSWICK, N.J.
U.S.A.