Proc. Natl. Acad. Sci.
USA
Vol. 73, No. 10, pp. 3747-3748, October 1976
Economic Sciences
A theorem on homogeneous functions and extended Cobb-Douglas
forms
(production and distribution theory)
A. CHARNES*, W. W. COOPERt, AND A. P. SCHINNARt
* Center for Cybernetic Studies, The University of Texas, Austin, Texas
Massachusetts 02163; and t School of Urban and Public Affairs, Carnegie-Mellon University, Pittsburgh, Pennsylvania 15213
Communicated by Jacob Marschak, July 27, 1976
ABSTRACT A form for homogeneous functions is presented
which shows them to be a very simple extension of the well-
known Cobb-Douglas functions with similar properties in pro-
duction (and distribution) economics. This form thus suggests
new possibilities for interpreting a wide variety of empirical
and theoretical results in economics; it also provides contact
with developments in other fields, such as information theory
and geometric programming.
One of the most extensively used forms for empirical studies
of production (and other areas, too) in economics is the
"Cobb-Douglas function" (see refs. 1 and 2),
Q = AL"K'". [1]
Here we have written it as a homogeneous form, of degree a
+ f3, in terms of Q, a single output, which results from the
amounts of two inputs, labor, L, and capital, K. Interchange-
ably, we shall also refer to such inputs as "factors of production,"
or, more briefly, as "factor inputs." A, a, and ,B are positive
constants, and a and ,@ represent "output elasticities" or re-
sponses of Q to variations in the factor amounts, L and K, re-
spectively. (See [3] and the following below.)
More generally,
n smooth," i.e., has continuous partial derivatives, and bi5(x) and
Q = A H| xj6'= AxI6'X262 -- X)',
where xi is the amount of the ith factor input and the constant
as is its corresponding "output elasticity", namely,
a ln Q xiOQ
a In xi Qax,
Q [3]
which is assumed to be positive, as are also all the variables
Xi.
As [3] makes clear, output elasticity is essentially a measure
of the incremental response in output, Q, to variation in the
amount of the ith factor input, xi. The statement of the deriv-
ative in logarithmic units produces a "dimensionless" measure§,
as is shown by the second term in [3]. As this second term also
makes clear, the value of this elasticity will, in general, depend
not only on aQ/8x1 (and the point where it is evaluated) but also
on the already attained ratio of the ith input to total output, i.e.,
xt/Q.
Sometimes [3] is called the "partial output elasticity" or, more
precisely, the "partial output elasticity for the ith factor input,"
but we shall follow common usage and refer to this more briefly
§ This was an original motivation in the choice of such logarithmic units
for characterizing these "proportionate responses" of output to each
indicated input (see ref. 3). P. A. Samuelson (4), p. 126, footnote 4,
shows, however, an example that eliminates this "dimensionless"
property.
78712; t Harvard University Graduate School of Business Administration, Boston,
as "output elasticity" since the associated mathematical notation
can generally be relied on to make clear which input is being
referred to.
Taking logarithms in [2] and then differentiating as in [3],
we see that the exponents bi are output elasticities. They are all
also constant. Because, in general, output elasticities are as-
sumed to be positive for any actually used input, the 5s in these
Cobb-Douglas functions are also assumed to be positive con-
stants. This assumption follows from the fact that the output
Q and all inputs xj are positive and so are the derivatives (called
"marginal productivities") 8Q/&rt, since all inputs have positive
prices and hence it would not be economical to use any input
that does not produce some positive output. In keeping with
common assumption we shall also assume that all output elas-
ticities are positive, although not necessarily constants, in the
discussion that follows.
With these assumptions, we now extends the above forms
and elasticities to
n
Q = A(x) HI xii(x). [4]
i=I
where Q is now a homogeneous function that is "sufficiently
[2] A(x) are functions depending on the vector x = (xi, . . .,xn) with
xj > 0, and defined in such a way that, as will be shown, the bi5(x)
also satisfy
n
i.
E: bl.(x)
= I
= I, bi(x) > O. i -= 1, ... , n. [5]
We call this an "extended" Cobb-Douglas form and observe
that it is a new form that holds for all sufficiently smooth ho-
mogeneous functions.
THEOREM ON HOMOGENEOUS FUNCTIONS
To state and motivate our theorem we first consider the ex-
pression
n
hf(x) = E fix,
-I
[6]
i
which, by Euler's theorem, always holds when the function f(x)
is homogeneous of degree h with continuous partial derivatives
("marginal productivities" if f is the production function),
I A. Charnes, W. W. Cooper, and A. P. Schinnar, "An extended
Cobb-Douglas form for use in production economics" (unpublished).
The University of Texas Center for Cybernetics Studies, BEB 203E,
Austin, Texas, 78712, where the subject of the paper is treated in
greater detail.
3747
3748 Economic Sciences: Charnes et al. Proc. Natl. Acad. Sci. USA 73 (1976)

d=f(x) in terms of specified behavioral hypotheses (e.g., with respect

f
i- Xi [7]
THEOREM. Let hf(x) > 0, where f(x) is a function that is
homogeneous of degree h, and let f(x) have continuous partial
derivatives. Then
n CONCLUSION
f(x) = A(x) H xjai'x' [8]
i. = I
where h b6(x) is the output elasticity of the ith factor input,
i.e.,
n
b6(x) = (fixj/Zfjxj) = (fixi/hf(x)),
j =1
and where
\6,x
n f
x
I hbi(x) * [10]
Proof: Insert [9] and [10] into [8] and use [6].
Of course, the output elasticities are no longer constants, as
in the ordinary Cobb-Douglas functions, but the above theorem
now provides an opening for further justification and inter-
pretations in the use of such functions and their extensions,
which have become almost canonical in empirical econometric
investigations at aggregate-i.e., economy wide or industry
wide-levels.
Finally, the b5(x) as represented in the theorem provide a
ready means for testing a variety of economic hypotheses (e.g.,
with respect to cost minimizing behavior) that are of interest
in economics. In fact, turning to A(x), as represented in [10],
we can give it economic meaning as the geometric mean of the
ratios of factor (i.e., input) marginal productivities and output
elasticities. Further we can also represent it, for each x, as
ln A(x) =-n h - 6x) In 6f (x)
But now recalling, see [5], that b6(x) 2 0 and libi(x) = 1, we
then see that this is, essentially**, the negative, of the "Khin-
chin-Kullback-Leibler" measure of the mean information for
discrimination in favor of the "as distribution" against the "fj
distribution" (see ref. 6). Here "distribution" means "frequency
distribution" as in statistics. It can also be given an interpretation
1I See e.g., the extension effected by Tinbergen (5) for a typical vari-
ant.
** i.e., apart from a "norming factor" (see footnotes).
to cost minimizing behavior) in economics so that [11] may also
be used for empirical inquiries to provide a very natural sum-
mary measure of the degree of departure (if any) from the
hypothesized behavior.
From what has just been said, it should be evident that the above
theorem provides contacts with a variety of other disciplines.
As we have just shown, for instance, such a contact is explicitly
made with information theory, and thus to other parts of sta-
tistics, too (see ref. 6). Further developments of an analogous
kind provide similar contacts with geometric programming (7),
dimensional analysis (8), and other disciplines as wells.
Doubtless other such generalizations can also be effected
(refs. 9 and 10). On the other hand, by a variety of specializing
assumptions with respect to the nature of the functions con-
sidered, the results of the above theorem can also be brought
to bear for new insight into large bodies of already developed
economic theories and empirical findingsT.
We thank Professor Jacob Marschak for numerous suggestions which
have helped to improve the above presentation.
1. Cobb, C. W. & Douglas, P. H. (1928) "A theory of production,"
Am. Econ. Rev. 18, 139-165.
2. Douglas, P. H. (1948) "Are there laws of production," The Am.
Econ. Rev. 38,1-39.
3. Marshall, A. (1920) Principles of Economics (The MacMillan
Company, New York), 8th ed. pp. 830-832.
4. Samuelson, P. A. (1947) Foundations of Economic Analysis
(Harvard University Press, Cambridge, Mass.), p. 126.
5. Tinbergen, J. (1975) Income Distribution (North-Holland
Publishing Co., Amsterdam), pp. 81-84.
6. Kullback, S. (1959) Information Theory and Statistics (John
Wiley & Sons, New York), pp. 36-65.
7. Duffin, R. J., Peterson, E. L. & Zener, C. M. (1967) Geometric
Programming (John Wiley & Sons, Inc., New York), pp. 79-
81.
8. Bridgman, P. W. (1931) Dimensional Analysis (Yale University
Press, New Haven, Conn.) pp. 1-112.
9. Charnes, A. & Cooper, W. W. (1974) "Constrained Kullback-
Leibler estimation; generalized Cobb-Douglas balance, and
unconstrained convex programming," Rendiconti di Accademia
Nazionale dei Lincei, Classe di scienze fisiche, matematiche e
naturali, Serie VIII, 56,556-561.
10. Charnes, A., Cooper, W. W. & Kortanek, K. 0. (1971) "Semi-
infinite programming, differentiability and geometric pro-
gramming, Part I: With example applications in economics and
management science," R. S. Varma Memorial Volume, J. Math.
Sci. 6, 19-40.