A Dynamical Model For Human Population: dN1/dt (N1, N2
A Dynamical Model For Human Population: dN1/dt (N1, N2
USA
Vol. 71, No. 7, pp. 2813-2817, July 1974
ABSTRACT We analyze a simple, deterministic model dividuals N1(t) and N2(t), vanishing when either Nf van-
of the dynamics of population changes in a bisexual, ishes, and increasing without limit as either Ni increases.
reproductive system based on marriage. Our model is one
of a general class, special cases of which have been pre- We will write this (nonnegative) rate as MI[Nj(t),N2(t)].
viously discussed within the framework of population 5. The mortality rates (X1,X2) of single individuals are con-
biology by D. G. Kendall, L. A. Goodman, J. H. Pollard, stants, independent of age, as is the mortality rate of
and others. Here, we extend and complete previous couples, X3, due to death (or divorce, or permanent steriliza-
analyses of systems characterized by first-degree homo- tion, etc.).
geneous, unbounded marriage functions, allowing for
arbitrary birth and death rates. 6. The dynamics of the system is independent of its overall
The dynamics of the model is determined by three scale, and of the common scale of its components.
coupled first-order, nonlinear differential equations,
similar to those used in the description of chemical re- Dynamical equations
actions and of radioactive decay chains. Sqlutions of the Given the assumptions above, the behavior of the system is
differential equation system are classified according to the
associated patterns of birth and death rates of the two determined by the following set of differential equations:
sexes, and growth and stability properties are discussed.
This preliminary report gives conditions sufficient to dN1/dt = - XN1 + f, T3 -M [N1,N2 ] [2a]
insure the existence of a unique, exponential mode of dN2/dt = -X2N2 + f2T3 -M[N,,N2] [2b]
population growth or decay, with a finite ratio of the
sexes. We also exhibit other conditions which, in contrast dT3/dt = -X3T3 + A[N1,N2]. [2c]
to the standard, linear demographic analysis of Lotka,
guarantee that the sex ratio asymptotically becomes The interpretation of [2] is straightforward. For example,
infinite.
The model manifests a delicate balance between the in [2a] the rate of change of the number of single women is
vital parameters that alerts one to the possibility of self- negatively proportional to the fraction of single women dying
aggravating distortions of the sex ratio, once a monog- at time t, - XN(t), and to the marriage rate -A3. The rate of
amous society's fertility falls below the replacement female births is proportional to the number of existing mar-
value. riages, and makes a positive contribution of f1T3(t) to dN1 (t)/
This is a preliminary report of the results of an investigation dt. Similarly, for the growth conditions on the number of
into the stability properties of reproductive systems whose single men N2(t).
dynamics are independent of overall scale, and in which age- Though we take mortality and fertility rates to be fixed
specific fertility and mortality rates are (provisionally) as- quantities independent of population growth, a more realistic
sumed to be non-age-specific, and constant through time. model would include, for example, the nonlinearity which
Special cases of this general class have previously been par- arises from the relations between agricultural production, the
tially analyzed by D. G. Kendall (1), L. A. Goodman (2), J. H. supply of arable land, and population density.
Pollard (3), and others. It is often convenient to work with the totals (T,,T2), in
which case the set [21 translates as:
Scale-independent model
dT1/dt = -X1T1 + (f' - X3 + X)DT3 = -XiT + fOTS [3a]
Our population model is defined by the following assumptions:
1. There are two sexes with total numbers TI(t) and T2(t). dT2/dt = -X2T2 + (f2 - X3 + X2)T3 = -X2T2 + 32T3 [3b]
2. Monogamy prevails, with T3(t) marriages existing at time t. [3c]
The number of single individuals of each sex is therefore dT3/dt = -X3T3 + M[T, - T3,T2- T3]
NAT(t) = T1(t) - T3(t); N2(t) = T2(t) -T3(t). [1] which defines the quantities fli. Note that for population
growth, both the O's must be positive. However, as we shall
S. The population is replenished only through births to mar- see below, positivity of the O's, while necessary, is not suf-
ried couples. The fertility rates, per marriage, for female ficient to insure a growing population.
and male births are fi and f2 respectively, and are inde- In [21 and [3 ], X3 is to be interpreted as a disappearance rate
pendent of parental ages. of marriages, due to the death of either partner, or to divorce.
4. The (instantaneous) rate at which marriages occur de- In this simple treatment, divorce or the death of one marriage
pends nonlinearly on the then-existing stocks of single in- partner can be considered to result in an accretion to the
2813
2814 Social Sciences: Yellin and Samuelson Proc. Nat. Acad. Sci. USA .l (1974)
stocks of single individuals available for later fecund mar- We note parenthetically that there exists an alternative
riages; such an accretion is tacitly included in our fertility possibility of bounded Mf, with
rates fi.
In the discusssion to follow, we allow the fundamental liniN/N o M[N1,N2]/Nj = Cj < co, (i, j = 1,2, i j)
parameters (fi,Xj) to range over all positive values, and we [8]
then qualitatively analyze the associated dynamical be- in contrast to [7b]. This asymptotic requirement is favored by
havior of the system. Our procedure differs from that of pre- many demographers, and is manifested by the harmonic
vious writers, who imposed certain further restrictions on the mean M[x,y] = xy/(x + y), and also by M[x,y] = Min(x,y).
parameters in order to explicitly take the accretions mentioned [See e.g., Pollard (3), Chap. 7].
above into account. An example of such additional restrictions It turns out that the bounded behavior [8] increases the
may be found in refs (1) and (3), where it is assumed that chances of pathological sex-ratio distortion. In this preliminary
3 = X1 + X2, fl > X2, f2 > Xi, and therefore that the resulting report we therefore concentrate primarily on the a priori more
values of the Eli-i.e., j31 = fi - X2; 32 = f2 - Xi-are positive. favorable case [7b], and show that unstable, infinitely un-
Our analysis is not restricted to these special cases, since in balanced, sex-ratios can still occur.
the age-specific models needed for more realistic description
there will often be a higher average age for married persons, Two-dimensional intensive system
and for surviving partners of terminating marriages, than for Since our marriage functions are first-order homogeneous, we
single persons of the same sex-with implied effects on average can choose to write the system in terms of the ratios
(fj,Xi) parameters not consistent with any such simple re- xi = N,/T3, X2 = N2/T3 [9]
strictions. We do, however, note the special implications of
any such postulated restrictions on the parameters. of single individuals to married couples. This reduces [21 and
[3] to the two-dimensional system
Restrictions on the marriage function M[N,,N21
dxi/dt = (X3 - Xi)Xi + fi
To insure scale-independence, we ask that the (positive) -
(1 + xj)M[XI,X2], (i = 1,2). [10]
marriage function M be first-degree homogeneous in its argu-
ments, so that The question of stability of the sex ratio reduces to a dis-
cussion of the possible existence of stable nodes in the (finite)
Mf[qNi,qN2] = qM[N,,N2] [41 positive quadrant of the (x,x2) plane. As we shall see belQw,
for all positive q. We do not insist that M[x,y] be a symmetric the number and distribution of critical points depend on the
function of x and y. detailed behavior of the marriage function, M[N,,N2].
We further suppose that no marriages take place if either Qualitative dynamics
stock of single individuals vanishes, so that for positive Ni
For any particular marriage function, our system is defined in
M[ON2] = 0 = M[N,,0]. [5] a five-dimensional space of positive parameters (XI;X2,X3,fif2).
Furthermore, we require that the marriage rate increase Our basic question then is: given a particular pattern of rela-
when the number of single individuals of either sex rises, so tive sizes of these parameters, is there positive population
that growth, immediately or ultimately, and does the sex ratio
R(t) = T2(t)/IT(t) [11]
OM[NNN2]/iN1 > 0 [i = 1,2], [6]
again for positive (N1,N2). approach a constant, positive, finite value which is stable
The general class of models we consider is characterized by under perturbation of the initial conditions?
In particular, we are interested in determining whether
the set [3]- or [2]- and by Eqs. [4]-[6]. Particular cases there exist trajectories Ti(t) which represent "interior rays" of
within this class are distinguished by further detailed assump- balanced-growth, exponential chaige, in which
tions on the behavior of the function Of. In this preliminary
note, we consider the class of M functions for which T(t) = k ert [12]
limNi1/NjM[N1,N2]/Ni = 0 (ij = 1,2; i id j) [7a] where the constant vector k has components ki > 0, and where
and k1/k3 > 1, k2/k3> 1, and r is real. (The exponential rate r is
positive for population growth, negative for decay, and zero
limN1/Nj,,M[N1,N2]/Nj = (ij = 1,2; i 7 j). [7b] for stationary replacement.) Furthermore, if such an interior
ray (or rays) does exist, we would like to know if it is stable
Equation [7a], which places restrictions on the behavior of under small and/or arbitrary perturbations of the initial
M[1,x]/x and M[x,1]/x for large x, can be interpreted as conditions, and whether its stability and growth properties
expressing the assumption that the more single individuals are altered under small changes in the fundamental param-
there are of one sex alone, the lower is the ultimate marriage eters (fi,Xi).
rate, per unit of that sex. We know that there always do exist two modes of one-sex
Equation [7b], which translates as unbounded behavior of exponential decay, which we call "exterior rays." For ex-
M[1,x] and AlM[x,1 ], can be interpreted as the assumption that ample, if the system begins with females only, they die out at
single females can, for a fixed positive supply of single males, an exponential rate X,, and the whole system moves along a
achieve any preset rate of marriages per unit time. (That trajectory
does not imply, however, that there are ever more weddings
than bachelors available to be grooms!) T(t) = [T,(O),OO]exp(-Xit). [13]
Proc. Nat. Acad. Sci. USA 71 (1974) A Dynamical Model for Human Population 2815
Likewise there is also a T2 exterior ray [0,T2(0),O ]exp(-X2t). Clearly, if we break the degeneracy [17], even by an in-
Note that there is no T3 exterior ray, since any initial stock of finitesmal amount, the curves dxi/dt = 0 will no longer be
married couples T3(0) immediately produces births at a rate coincident, and, generally, there will be, at most, only a finite
number of critical points in the (xl,x2) plane.
dNi(0)/dt = fiT3(0), (i = 1,2). [14] For illustrative purposes, let us examine a particular simple
A major part of our problem is to distinguish between situa- perturbation from degeneracy. We will still treat both sexes
tions in which there exist interior rays that asymptotically symmetrically, choosing
determine the behavior of the system, and those in which f - X3 + X = f- X3 + Xi = (, [191
R(t) tends to 0 or c, so that the system eventually reaches an
exterior ray on the T, or T2 axis. but keeping fi = f2 and Xi = X2.
We can easily derive necessary conditions for the existence It is easily verified that negative (3 in [19] corresponds to a
of interior rays by substituting [12 ] into [2 1. This leads to the declining population, while positive (3 may be associated with
three conditions either growth or decline. Explicitly, from the now-linear case
of [15], the general symmetric growth rate is found to be
kilk3 = (fs - r - X3)/(r + X) > 0, (i = 1,2) [15a]
+ + o\(f
2,M M)2 + 4fM [20]
r + X3 = M[k1/k3,k2/k3] > 0. [15b] ±
totes lie along the axes xi = 0 and x2 = 0. Saddle points and [23]
stable nodes alternate along this hyperbola. For case (A'), the
two nodes at the ends of the chain are stable; all trajectories It can be shown from [10] that the relation [23] insures that
therefore result in balanced sex ratios, just as in case (A). For the curves dxl/dt = 0 and dx2/dt = 0 cross at least once in the
case (A'), the population shrinks. positive quadrant of the (xx2) plane. Cases (C) and (D), to be
Remark: In cases (A'), (B) and (C) certain singular values of discussed below, are associated with negative U.
the parameters result in points of tangency between the two (C) K stable nodes, K saddle points in xi plane.
curves dxi/dt = 0 of [10], and there can even be an uncount- This occurs for sectors 7 and 8 in Table 1. The chain of
able infinity of points in common. Furthermore, it can be critical points now has a saddle point on one end, and a stable
shown that for particularly simple choices of 31 [x,,x2], e.g. node on the other. There is now a watershed defined by the
M = XlaX2l -a, 0 < a < 1, there is at most a pair of critical separatrix dividing the last saddle point from its neighboring
points in the finite plane. We note also that all our critical stable node. Trajectories on the saddle point side of this
points are associated with real exponents. There are no separatrix always result in unbalanced sex ratios.
complex foci, and there is no limit cycle behavior in the xi Cases (A'), (B), and (C) have the property that K can be
plane. reduced to zero, independently of the shape of M [x,,x2], by
(B) K+1 saddle points, K stable nodes in the finite xi plane. changing the relative scale of the marriage function. In par-
This occurs for sectors 4-6 in Table 1, in which at least one ticular, if we write .Ml[N,,N2] = A141,1] .1[N,,N2]/M1[1,1]J}
of the Pi is negative. In general, we again have a chain of and permanently fix the expression in curly brackets, then
critical points, but here, unlike (A'), the two ends of the chain ll [1,1] becomes a parameter characterizing the "strength"
are saddle points. Therefore, trajectories lying beyond the two or scale-of the marriage function, and this strength plays a
separatrices dividing the saddle points at the ends of the chain role in determining K in sectors 3 through 8.
from their stable-node neighbors will always result in unbal- In case (C)-as in Cases (A') and (B)-the population
anced sex ratios. In case (B) the population shrinks. shrinks.
Cases (A) and (B) occur when (D) No critical points in the finite (X1,X2) plane. In these
Proc. Nat. Acad. Sci. USA 71 (1974) A Dynamical Model for Human Population 2817
cases, sectors 9-12 in Table 1, one sex has both a higher geometric explanation of why this is so. Visualize [Ti] space,
birth rate and a lower death rate than the other, with the re- with a vertical T3 axis. Then note, from [3], that above
sult that the sex ratio asymptotically approaches 0 or a. (below) their respective planes
Specifically, a sufficient condition for the nonexistence of a
balanced-growth path is -X1Ti + (Xi + fi - 3) T3 = 0 (i = 1,2), [27]
A2) < 0- [24] dT1/dt and dT2/dt are positive (negative). Similarly, below
(A1 - A2) (fl f2) (fi
- - A1) (fl -
(above) the hyperbolic surface
The Kendall-Pollard restriction to positive fl's rules out the
saddles of sections 4-6, and also the exterior ray sectors 10-12, -X3T3 + M[T1,T2] = 0, [28]
while permitting sector 9. Sector 9's disastrous blowup of the dT3/dt is positive (negative). For sufficiently large fi and f2,
sex ratio may seem intuitively obvious from the fact that one the planes [27] intersect the surface [28] in such a way as to
of the sexes there has a higher birth rate and a lower death create an irregularly shaped cone, with its apex at T1 = 0,
rate than the other. But such intuitive explanations are dan- within which all three dTi/dt are positive. With the exception
gerous for at least two reasons: (1) XA > X2 and fi < f2 causes no of the apex, the surface of this cone lies completely within the
infinite distortion of the asymptotic sex ratio in sector 2. In positive octant, and all trajectories entering it can never
sector 2 birth rates will be sufficiently high, and the sex ratio escape, but instead go off to infinity. Again, the cone is
N2/N1 asymptotically approaches a positive limit, greater bounded above by the surface [28], and below by the planes
than unity but finite. (2) The following brief discussion of a [27].
non-marriage, bisexual deterministic model shows one can This trapping phenomenon for growing population is a
always avoid infinite sex ratios, even with XI >> X2 and fi << f2, feature of systems with marriage functions even more general
if monogamy does not prevail. than those we consider here. The essential properties of M
A bisexual, non-marriage model
which insure trapping are that it vanish for vanishing stocks of
single individuals, and that it grow at least linearly as its
Consider, along with the earlier cited writers, a system with no arguments increase together.
marriage institution, and write, instead of [2 ] and [3]:
Conclusion
dNl(t)/dt = -X1NI(t) + f1H [N1(t),N2(t) ] [25a]
In conclusion, it is important to emphasize that models such
dN2(t)/dt = -X2N2(t) + f2H [N1(t),N2(t) ] [25b] as ours lack two features which play crucial roles in the popu-
where H(x,y) is again an unbounded, positive function, homo- lation dynamics of real societies: (1) the effects of biomedical
and cultural change on mortality and fertility; (2) the feed-
geneous of order 1, vanishing on the boundaries of the positive back caused by resource limitations on output. Such effects
quadrant, and possessing a positive gradient, just as did M. might be admitted by allowing our fundamental parameters to
The system [25] always moves asymptotically to a finite
sex-ratio, as one easily sees by transforming to a one-dimen- be explicit functions of time, and of population density itself.
sional system in Z = N21NI, We thank Professors J. M. Deutch and R. M. Solow of M.I.T.
for a number of helpful discussions; Professors J. E. Cohen of
dZ/dt = (XI - X2)Z + H(1,Z)(f2- fZ) [26] Harvard University and A. J. Coale of Princeton alerted us to the
fact that the cited works of Kendall, Goodman, Pollard, and
and using the properties of H to show that dZ/dt changes sign others had anticipated some of our results. We would like to
from positive to negative once along the positive Z axis, inde- express our appreciation to the National Science Foundation for
pendent of the signs or magnitudes of X2 - XI, f2 and f'. its financial support of this research.
1. Kendall, D. G. (1949) "Stochastic processes and population
Population decline and sex-ratio pathology growth," J. Roy. Stat. Soc. B!1, 230-264, in particular pp.
Finally, we note that a striking feature of the results of Table 1 248-249.
is the absence of pathological behavior of the sex ratio in the 2. Goodman, L. A. (1953) "Population growth of the sexes,"
Biometrics 9, 212-225, in particular pp. 216-218.
presence of overall population growth. Growth is always as- 3. Pollard, J. H. (1973) Mathematical Models for the Growth of
sociated with a unique, stable interior ray. There is a simple Human Population (Cambridge Univ. Press, Cambridge).