     Egwald Economics: Microeconomics

Duality and the Translog Production / Cost Functions
Non-Homothetic Generalized CES Technology

by

Elmer G. Wiens

Egwald's popular web pages are provided without cost to users.
Please show your support by joining Egwald Web Services as a Facebook Fan: Follow Elmer Wiens on Twitter: Duality: Production / Cost Functions:   Cobb-Douglas Duality | CES Duality | Theory of Duality | Translog Duality - CES | Translog Duality - Generalized CES | References and Links

S. Translog (Transcendental Logarithmic) Duality and the Generalized CES Technology

I. Profit (Wealth) Maximizing Firm.

Production and cost functions (and profit functions) can be used to model how a profit (wealth) maximizing firm hires or purchases inputs (factors), such as labour, capital (structures and machinery), and materials and supplies, and combines these inputs through its production process to produce the products (outputs) that the firm sells (supplies) to its customers. The theory of duality links the production function models to the cost function models by way of a minimization or maximization framework. The cost function is derived from the production function by choosing the combination of factor quantities that minimize the cost of producing levels of output at given factor prices. Conversely, the production function is derived from the cost function by calculating the maximum level of output that can be obtained from specified combinations of inputs.

II. The Production and Cost Functions.

The Translog production function:

ln(q) = ln(A) + aL*ln(L) + aK*ln(K) + aM*ln(M) + bLL*ln(L)*ln(L) + bKK*ln(K)*ln(K) + bMM*ln(M)*ln(M)
+ bLK*ln(L)*ln(K) + bLM*ln(L)*ln(M) + bKM*ln(K)*ln(M)   =   f(L, K, M).

q = exp(f(L, K, M)) = F(L, K, M).

an equation in 10 parameters, A, aL, aK, aM, bLL, bKK, bMM, bLK, bLM, bKM, where L = labour, K = capital, M = materials and supplies, and q = product.

The Translog (total) cost function:

 ln(C(q;wL,wK,wM)) = c + cq * ln(q) + cL * ln(wL) + cK * ln(wK) + cM * log(wM)                                 + .5 * [dqq * ln(q)^2 + dLL * ln(wL)^2 + dKK * ln(wK)^2 + dMM * ln(wM)^2]                   + .5 * [(dLK + dKL) * ln(wL)*ln(wK) + (dLM + dML) * ln(wL)*ln(wM) + (dKM + dMK) * ln(wK)*log(wM)]                   + dLq * ln(wL)*ln(q) + dKq * ln(wK)*ln(q) + dMq * ln(wM)*ln(q)

an equation in 18 parameters, c, cq, cL, cK, cM, dqq, dLL, dKK, dMM, dLK, dKL, dLM, dML, dKM, dMK, dLq, dKq, and dMq, where wL, wK, and wM are the factor prices of L, K, and M respectively.

III. Duality between Production and Cost Functions.

Mathematically, the duality between a production function, q = F(L, K, M), and a cost function, C(q; wL, wK, wM), is expressed:

C*(q; wL, wK, wM) = minL,K,M{ wL * L + wK * K + wM * M   :   q - F(L,K,M) = 0,   q > 0, wL > 0; wK > 0, and wM > 0,   L > 0, K > 0, M > 0}     (*)

F*(L, K, M) = maxq {q   :   C(q; wL, wK, wM)   <=   wL * L + wK * K + wM * M,   L > 0, K > 0, M > 0,   for all wL >= 0 , wK >= 0, wM >= 0}     (**),

with the questions, F == F*, and C == C*?

The dual functions C*, and F* will be derived from the estimated functions F, and C.

If the functions F(L, K, M), and C(q; wL, wK, wM) obey sufficient conditions, the above minimization and maximization problems can be solved by nonlinear optimization techniques, such as Newton's Method. Furthermore, the implicit function theorem can be exploited to facilitate such calculations.

IV. Constrained Optimization.

The (*) minimization problems are solved for the Generalized CES production function, the Translog production function, and for the Diewert (Generalized Leontief) production function.

The maximization problem (**) is somewhat more difficult than the minimization problem (*), since as specified, the requirements   —   for all wL >= 0 , wK >= 0, wM >= 0   —   imply an infinite number of constraints.

This difficulty is overcome with cost functions that factor:

C(q; wL, wK, wM) = q^1/nu * c(wL, wK, wM)

by solving the reduced dimension problem:

F*(L, K, M)^1/nu = 1 / maxwL,wK,wM {c(wL, wK, wM)   :   wL * L + wK * K + wM * M = 1,   wL >= 0, wK >= 0, wM >= 0)

However, with cost functions that do not factor, such as the Translog cost function, the following method can be used to reduce the dimensionality of the constraints of (**), if the cost function is linear homogeneous in factor prices.

Cost Function Linear Homogeneous in Factor Prices.

When (L, K, M) is the least cost combination of inputs at the specific combination of factor prices (wL, wK, wM):

C(q; wL, wK, wM)   =   wL * L + wK * K + wM * M.

With the cost function, C(q; wL, wK, wM), linear homogeneous in factor prices:

C(q; wL / wM, wK / wM, 1)   =   wL / wM * L + wK / wM * K + M,

assuming that wM > 0. Writing vL = wL / wM, and vK = wK / wM, the maximization problem (**) becomes the minimization problem:

F*(L, K, M) = minq, vL, vK {q   :   C(q; vL, vK, 1)   >=   vL * L + vK * K + M,   L > 0, K > 0, M > 0,   for all vL >= 0 , vK >= 0}     (***),

an optimization problem with one constraint.

V. Cost Function to Production Function: The Method of Langrange.

Using the Method of Lagrange, define the Langrangian function, H, of the minimization problem (***):

H(L, K, M, q, vL, vK, λ) = q + λ * (vL * L + vK * K + M - C(q; vL, vK, 1)),

where the new variable, λ, is called the Lagrange multiplier.

a.   First Order Necessary Conditions:

 0.   Hλ(L, K, M, q, vL, vK, λ) = vL * L + vK * K + M - C(q; vL, vK, 1) = 0 1.   Hq(L, K, M, q, vL, vK, λ) = 1 - λ * ∂C(q; vL, vK, 1)/∂q = 0 2.   HvL(L, K, M, q, vL, vK, λ) = λ * (L - ∂C(q; vL, vK, 1)/∂vL) = 0 3.   HvK(L, K, M, q, vL, vK, λ) = λ * (K - ∂C(q; vL, vK, 1)/∂vK) = 0

b.   Solution Functions:

We want to solve, simultaneously, these four equations for the variables λ, q, vL, and vK as functions of the variables L, K, and M, and the parameters of the cost function.

λ = λ(L, K, M)
q = q(L, K, M)
vL = vL(L, K, M)
vK = vK(L, K, M)

c.   Jacobian Matrices:

The Jacobian matrix (bordered Hessian of H) of the four functions, Hλ, Hq, HvL, HvK, with respect to the choice variables, λ, q, vL, vK :

J3   =

 Hλλ Hλq HλvL HλvK Hqλ Hqq HqvL HqvK HvLλ HvLq HvLvL HvLvK HvKλ HvKq HvKvL HvKvK
=
 0 -Cq L - CvL K - CvK -Cq -λ*Cqq -λ*CqvL -λ*CqvK L - CvL C-λ*CqvL -λ*CvLvL -λ*CvLvK K - CvK -λ*CqvK -λ*CvKvL -λ*CvKvK

=   Jλ, q, vL, vK

The bordered principal minor of the bordered Hessian of the Langrangian function, H:

J2   =

 0 -Cq L - CvL -Cq -λ*Cqq -λ*CqvL L - CvL -λ*CqvL -λ*CvLvL

d.   Second Order Necessary Conditions:

The second order necessary conditions require that the Jacobian matrix (bordered Hessian of H) be positive definite at the solution vector W = (L, K, M, q, vL, vK, λ). The Jacobian matrix, J3, is a positive definite matrix if the determinant of J2 is negative, and the determinant of J3 is negative.

e.   Sufficient Conditions:

If the second order necessary conditions are satisfied, then the first order necessary conditions are sufficient for a minimum at W.

f. The Solution Functions' Comparative Statics.

The Jacobian matrix of the four functions, Hλ, Hq, HvL, HvK, with respect to the variables, L, K, M:

JL, K, M   =

 Hλ,L; Hλ,K Hλ,M Hq,L Hq,K Hq,M HvL,L HvL,K HvL,M HwK,L HwK,K HvK,M
=
 vL vK 1 0 0 0 λ 0 0 0 λ 0

The Jacobian matrix of the four solution functions, Φ = {λ, q, vL, vK}, with respect to the variables, L, K, and M:

JΦ   =

 λL λK λM qL qK qM vLL vLK vLM vKL vKK vKM

From the Implicit Function Theorem:

JL, K, M;   +   Jλ, q, vL, vM   * JΦ   =   0 (zero matrix)   →

JΦ   =   - (Jλ, q, vL, vK)-1   *   JL, K, M

for points (L, K, M) in a neighborhood of the point (L, K, M), and wL = wL(L, K, M), wK = wK(L, K, M), wM = wM(L, K, M), and λ = λ(L, K, M),

with:

 F*(L, K, M) = q(L, K, M), F*L(L, K, M) = qL(L, K, M), F*K(L, K, M) = qK(L, K, M), and F*M(L, K, M) = qM(L, K, M),

VI. Question: F == F*?

The obtained values for F*(L, K, M), F*L(L, K, M), F*K(L, K, M), and F*M(L, K, M) can be compared with the values of F(L, K, M), FL(L, K, M), FK(L, K, M), and FM(L, K, M), and with the corresponding values from the Generalized CES production function.

VII. Duality.  The Plan:

1. Specify the parameters of a Generalized CES production function, and obtain the derived Generalized CES cost function using Newton's Method.

2. Generate the Generalized CES data displayed in the table below, and estimate the parameters of a Translog production function, and the parameters of a Translog cost function.

3. Using Newton's Method with the implicit function theorem, obtain the production function that is dual to the estimated Translog cost function. Check that the derived Translog production function corresponds with the estimated Translog production function, and the underlying Generalized CES production function.

4. Using Newton's Method with the implicit function theorem, obtain the cost function that is dual to the estimated Translog production function. Check that the derived cost function corresponds with the estimated Translog cost function, and the underlying derived Generalized CES cost function.

VIII. Generate Generalized CES production / cost data.

The three factor Generalized CES production function is:

q = A * [alpha * L^-rhoL + beta * K^-rhoK + gamma *M^-rhoM]^(-nu/rho) = f(L,K,M).

where L = labour, K = capital, M = materials and supplies, and q = product.

The parameter nu permits one to adjust the returns to scale, while the parameter rho is the geometric mean of rhoL, rhoK, and rhoM:

rho = (rhoL * rhoK * rhoM)^1/3.

If rho = rhoL = rhoK = rhoM, we get the standard CES production function. If also, rho = 0, ie sigma = 1/(1+rho) = 1, we get the Cobb-Douglas production function.

The estimated coefficients of the Translog production and cost functions and will vary with the parameters nu, rho, rhoL, rhoK, rhoM, alpha, beta and gamma of the Generalized CES production function.

Set the parameters below to re-run with your own Generalized CES parameters.

The restrictions ensure that the least-cost problems can be solved to obtain the underlying Generalized CES cost function, using the parameters as specified.
Intermediate (and other) values of the parameters also work. The program might "time-out" for values of rho >> 0 and rho << 0, yielding "NAN" values.

Restrictions:
.8 < nu < 1.1;
-.6 < rhoL = rho < .6;
-.6 < rho < -.2 → rhoK = .95 * rho, rhoM = rho/.95;
-.2 <= rho < -.1 → rhoK = .9 * rho, rhoM = rho / .9;
-.1 <= rho < 0 → rhoK = .87 * rho, rhoM = rho / .87;
0 <= rho < .1 → rhoK = .7 * rho, rhoM = rho / .7;
.1 <= rho < .2 → rhoK = .67 * rho, rhoM = rho / .67;
.2 <= rho < .6 → rhoK = .6 * rho, rhoM = rho / .6;
rho = 0 → nu = alpha + beta + gamma (Cobb-Douglas)
4 <= wL* <= 11,   7<= wK* <= 16,   4 <= wM* <= 10

Generalized CES Production Function Parameters
nu:
rho:
Base Factor Prices
 wL* wK* wM*
Distribution to Randomize Factor Prices
 Use [-2, 2] Uniform distribution     Use .25 * Normal (μ = 0, σ2 = 1)

The Generalized CES production function as specified:

q = 1 * [0.35 * L^- 0.17647 + 0.4 * K^- 0.11823 + 0.25 *M^- 0.26339]^(-1/0.17647) = f(L,K,M).

 The factor prices are distributed about the base factor prices by adding a random number distributed uniformly in the [-2, 2] domain.

The Generalized CES cost function as derived from the Generalized CES production function:

Unlike the homothetic CES technology, the non-homothetic Generalized CES technology does not admit a closed form cost function. Consequently, the Generalized CES cost function is derived from the Generalized CES production function as demonstrated on the Generalized CES Production Function web page.

IX. Estimate the Translog Production and Cost Functions.

Estimating the Translog production function from the Generalized CES data using SVD least squares yields the coefficient estimates:

Generalized CES
Translog Production Function

SVD Least Squares
Parameter Estimates
Parameter Coefficient std error t-ratio
lnA0.006995016.395
aL0.35773801300.882
aK0.2691360631.497
aM0.35890933.793
bLL-0.0195430-156.262
bKK-0.0100470-75.673
bMM-0.0278840-247.139
bLK0.019646087.161
bLM0.0148460104.212
bKM0.015113083.644
 R2 = 1 R2b = 1 # obs = 28 Observation Matrix Rank: 10
 aL + aK + aM = 0.986-2*bLL = 0.039086 =~ 0.034492 = bLK + bLM-2*bKK = 0.020095 =~ 0.034759 = bLK + bKLM-2*bMM = 0.055767 =~ 0.029959 = bLM + bKM

The estimated Translog production function:

ln(q) = 0.006995 + 0.357738 * ln(L) + 0.269136 * ln(K) + 0.3589 * ln(M) + -0.019543 * ln(L)*ln(L) + -0.010047 * ln(K)*ln(K) + -0.027884 * ln(M)*ln(M)
+ 0.019646 * ln(L)*ln(K) + 0.014846 * ln(L)*ln(M) + 0.015113 * ln(K)*ln(M)   =   f(L,K,M).

The Translog cost factor share functions are:

 sL(q;wL,wK,wM) = cL + dLq * ln(q) + dLL * ln(wL) + dLK * ln(wK) + dLM * ln(wM), sK(q;wL,wK,wM) = cK + dKq * ln(q) + dKL * ln(wL) + dKK * ln(wK) + dKM * ln(wM), sM(q;wL,wK,wM) = cM + dMq * ln(q) + dML * ln(wL) + dMK * ln(wK) + dMM * ln(wM),

three linear equations in their 15 parameters.

Estimating the Translog factor share functions simultaneously with the specified constraints, from the Generalized CES data using QR least squares, yields the coefficient estimates:

QR Restricted Least Squares
Parameter Estimates
Parameter Coefficient std error t-ratio
cL0.3613410.0001782032.974429
dLq-9.0E-65.0E-5-0.179929
dLL0.0344798.0E-5429.594317
dLK-0.0138386.9E-5-200.771679
dLM-0.0206415.3E-5-389.508306
cK0.2723230.0001831484.937967
dKq0.0207195.0E-5416.170362
dKL-0.0138386.9E-5-200.771679
dKK0.0307728.3E-5369.800374
dKM-0.0169355.4E-5-312.356059
cM0.3663360.0001841987.861853
dMq-0.0207095.1E-5-409.748804
dML-0.0206415.3E-5-389.508306
dMK-0.0169355.4E-5-312.356059
dMM0.0375766.1E-5611.264869
 R2 = 1 R2b = 1 # obs = 84
 dLK = dKL, dLM = dML, dKM = dMK 1 = cL + cK + cM 0 = dLL + dLK + dLM 0 = dKL + dKK + dKM 0 = dML + dMK + dMM 0 = dLq + dKq + dMq

The first three constraints are the usual across equation constraints on the factor share functions. The last five constraints are necessary conditions for the estimated Translog cost function to be linear homogeneous in factor prices. See the Generalized CES-Translog Cost Function for more details.

To obtain estimates of the remaining three parameters, c, cq, and dqq, of the Translog cost function write:

ln(C(q;wL,wK,wM)) - {cL * ln(wL) + cK * ln(wK) + cM * log(wM) + .5 * [dLL * ln(wL)^2 + dKK * ln(wK)^2 + dMM * ln(wM)^2]
+ .5 * [(dLK + dKL) * ln(wL)*ln(wK) + (dLM + dML) * ln(wL)*ln(wM) + (dKM + dMK) * ln(wK)*log(wM)]
+ dLq * ln(wL)*ln(q) + dKq * ln(wK)*ln(q) + dMq * ln(wM)*ln(q)}
= R(q;wL,wK,wM) = c + cq * ln(q) + .5 * dqq * ln(q)^2

Estimate the linear equation:

R(q;wL,wK,wM) = c + cq * ln(q) + .5 * dqq * ln(q)^2

to obtain c, cq, and dqq.

QR Restricted Least Squares
Parameter Estimates
Parameter Coefficient std error t-ratio
c1.1149690.0001447758.570275
cq101000
dqq0.021012.5E-5851.89353
 R2 = 1 R2b = 1 # obs = 28

1 = cq

The estimated Translog cost function:

ln(C(q;wL,wK,wM)) = 1.114969 + 1 * ln(q) + 0.361341 * ln(wL) + 0.272323 * ln(wK) + 0.366336 * log(wM)
+ .5 * [0.02101 * ln(q)^2 + 0.034479 * ln(wL)^2 + 0.030772 * ln(wK)^2 + 0.037576 * ln(wM)^2]
+ .5 * [-0.027676 * ln(wL)*ln(wK) + -0.041283 * ln(wL)*ln(wM) + -0.033869 * ln(wK)*log(wM)]
+ -9.0E-6 * ln(wL)*ln(q) + 0.020719 * ln(wK)*ln(q) + -0.020709 * ln(wM)*ln(q)

X. Example: Cost Function to Production Function:

The dual Translog production function, F*, is obtained from the estimated Translog cost function, C, by:

F*(L, K, M) = minq, vL, vK {q   :   C(q; vL, vK, 1)   >=   vL * L + vK * K + M,   L > 0, K > 0, M > 0,   for all vL >= 0 , vK >= 0}     (***).

With L = 39.05, K = 21.4, and M = 36.07, (***) becomes:

F*(39.05, 21.4, 36.07) = minq, vL, vK {q   :   C(q; vL, vK, 1)   >=   vL * 39.05 + vK * 21.4 + 36.07,   for all vL >= 0 , vK >= 0}     (***).

XI. Constrained Optimization (Minimum): The Method of Lagrange:

H(39.05, 21.4, 36.07, q, vL, vK, λ) = q + λ * (vL * 39.05 + vK * 21.4 + 36.07 - C(q; vL, vK, 1)),

where λ is the Lagrange multiplier.

a. First Order Necessary Conditions:

 0.   Hλ(39.05, 21.4, 36.07, q, vL, vK, λ) = vL * 39.05 + vK * 21.4 + 36.07 - C(q; vL, vK, 1) = 0 1.   Hq(39.05, 21.4, 36.07, q, vL, vK, λ) = 1 - λ * ∂C(q; vL, vK, 1)/∂q = 0 2.   HvL(39.05, 21.4, 36.07, q, vL, vK, λ) = λ * (39.05 - ∂C(q; vL, vK, 1)/∂vL) = 0 3.   HvK(39.05, 21.4, 36.07, q, vL, vK, λ) = λ * (21.4 - ∂C(q; vL, vK, 1)/∂vK) = 0

b.   Solution Functions:

Solve these four equations simultaneously (using Newton's Method) for λ, q, vL, and vK as functions of the variables L, K, and M, and the parameters of the cost function, so that:

λ = λ(39.05, 21.4, 36.07)
q = q(39.05, 21.4, 36.07)
vL = vL(39.05, 21.4, 36.07)
vK = vK(39.05, 21.4, 36.07)

We know that the ranges of factor prices are: 4 <= wL <= 11, 7 <= wK <= 16, and 4 <= wM <= 10. Estimating wL = 7.5, wK = 11.5, and wM = 7 yields vL = wL / wM = 7.5 / 7, and vK = wK / wM = 11.5/7. With a range of 20 to 45 for output, estimate q = 33. From the first order condition 1., estimate λ = 1 / ∂C(33; 7.5/7, 11.5/7, 1)/∂q = 0.218.

Newton's Method:

Using these estimates, Newton's Method provides:

Nonlinear Optimization: Newton's Method
Parameter Estimates
Iter #λqvLvK
0   0.218 331.0714 1.6429
10.2122 26.791.0602 1.8586
20.1955 26.991.1519 2.1164
30.1944 271.1661 2.1652
40.1944 271.1665 2.1666
50.1944 271.1665 2.1666

c. Solution Vector:

W = (L, K, M, q, vL, vK, λ) = (39.05, 21.4, 36.07, 27, 1.16653, 2.16659, 0.19439)

With λ = 0.19439, q = 27, vL = 1.16653, and vK = 2.16659, the first order conditions are:

 0.   Hλ(39.05, 21.4, 36.07, q, vL, vK, λ) = 1.16653 * 39.05 + 2.16659 * 21.4 + 36.07 - C(q; 1.16653, 2.16659, 1) = 0 1.   Hq(39.05, 21.4, 36.07, q, vL, vK, λ) = 1 - 0.19439 * ∂C(q; 1.16653, 2.16659, 1)/∂q = 0 2.   HvL(39.05, 21.4, 36.07, q, vL, vK, λ) = 0.19439 * (39.05 - ∂C(q; 1.16653, 2.16659, 1)/∂vL) = 0 3.   HvK(39.05, 21.4, 36.07, q, vL, vK, λ) = 0.19439 * (21.4 - ∂C(q; 1.16653, 2.16659, 1)/∂vK) = -0

d. Second Order Necessary Conditions:

The second order necessary conditions require that the Jacobian matrix (bordered Hessian of H) be positive definite at the solution vector W = (L, K, M, q, vL, vK, λ). The Jacobian matrix, J3, is a positive definite matrix if the determinant of J2 is negative, and the determinant of J3 is negative.

Jλ, q, vL, vK =

J3   =

 0 -Cq L - CvL K - CvK -Cq -λ*Cqq -λ*CqvL -λ*CqvK L - CvL C-λ*CqvL -λ*CvLvL -λ*CvLvK K - CvK -λ*CqvK -λ*CvKvL -λ*CvKvK
=
 0 -5.144 0 -0 -5.144 -0.004 -0.305 -0.176 0 -0.305 3.561 -1.133 -0 -0.176 -1.133 1.062

Determinant(J3) = -66.05693

J2   =

 0 -5.144 0 -5.144 -0.004 -0.305 0 -0.305 3.561

Determinant(J2) = -94.24171

e. Maximum Output:

With L = 39.05, K = 21.4, M = 36.07,

Dual Translog production function:

F*(39.05, 21.4, 36.07) = q(39.05, 21.4, 36.07) = 27,

Estimated Translog production function:

F(39.05, 21.4, 36.07) = 27.

Specified Generalized CES production function:

f(39.05, 21.4, 36.07) = 27.

f. The Solution Functions' Comparative Statics.

JL, K, M   =

 vL vK 1 0 0 0 λ 0 0 0 λ 0
=
 1.16653 2.16659 1 0 0 0 0.1944 0 0 0 0.1944 0

From the Implicit Function Theorem:

JΦ   =   - (Jλ, wL, wK, wM, q)-1   *   JL, K, M

JΦ   =

 λL λK λM qL qK qM vLL vLK vLM vKL vKK vKM
=
 0.001909 0.003554 -0.005155 0.22676 0.42117 0.19439 -0.03512 5e-05 0.04076 0.00011 -0.11323 0.07574

g. Comparing Partial Derivatives:

Partial Derivates of the dual Translog production function F*(L,K,M):

F*L(L,K,M) = F*(39.05, 21.4, 36.07) = qL(39.05, 21.4, 36.07) = 0.227,
F*K(L,K,M) = F*(39.05, 21.4, 36.07) = qK(39.05, 21.4, 36.07) = 0.421,
F*M(L,K,M) = F*(39.05, 21.4, 36.07) = qM(39.05, 21.4, 36.07) = 0.194.

Partial Derivatives of the estimated Translog production function F(L,K,M):

FL(39.05, 21.4, 36.07) = 0.227,
FK(39.05, 21.4, 36.07) = 0.421,
FM(39.05, 21.4, 36.07) = 0.194.

Partial Derivatives of the specified Generalized CES production function f(L,K,M):

fL(39.05, 21.4, 36.07) = 0.227,
fK(39.05, 21.4, 36.07) = 0.421,
fM(39.05, 21.4, 36.07) = 0.194.

XII. Table of Results.

Check that the derived dual production function corresponds with the estimated Translog production function by comparing the values for output, q, and the partial derivatives of the production function.

Translog Production / Cost Function Duality
Cost Function to Production Function
Generalized CES: Returns to Scale = 1, rho = 0.17647, rhoL = 0.17647, rhoK = 0.11823, rhoM = 0.26339
—       Generalized CES Data   —    —    Estimated Translog Cost   —   — Derived Dual Production — — Estimated Translog Production —
obs #qwLwKwM LK Mcost LKMcostqF*LF*KF*MqFLFKFM
1178.9 13.246.34 21.1513.69 23.59519.1421.1713.6923.61519.3416.9940.2710.4030.193170.2710.4030.193
2188.46 12.824.56 21.2713.54 29.57488.3721.2713.5429.59488.4917.9960.2880.4360.155180.2880.4360.155
3195.92 13.16.58 29.9713.85 22.95509.9129.9913.8622.93509.9718.9980.2050.4540.228190.2050.4530.227
4205.52 11.14.64 28.8114.47 27.59447.7228.8114.4727.59447.7519.9990.2280.4590.192200.2280.4580.191
5218.48 14.626.46 28.2816.29 29.32667.2728.2816.2829.32667.28210.2470.4260.188210.2470.4250.188
6228.6 14.324 26.4215.64 40.67613.9526.4215.6540.66613.9222.0010.2820.470.131220.2820.470.131
7237.16 13.124.64 30.7116.82 35.98607.5830.7116.8235.99607.5323.0020.2490.4570.162230.2490.4560.161
8248.26 12.444.02 28.3918.46 41.95632.7428.3818.4641.95632.6624.0030.2870.4330.14240.2870.4330.14
9256.36 13.545.36 37.4118.1 35.12671.3437.418.135.12671.2525.0030.2180.4640.184250.2180.4640.184
10267.6 14.685.04 35.8918.91 40.85756.1735.8818.940.84756.0526.0040.240.4630.159260.240.4630.159
11277.7 13.167.56 39.2123.15 32.51852.3339.2123.1432.5852.227.0040.2260.3860.222270.2260.3860.221
12288.84 13.75.84 36.1823.22 41.28879.0336.1723.2141.29878.8928.0040.2590.4010.171280.2590.4010.171
13298.96 14.37.14 39.5424.83 38.66985.3639.5324.8238.66985.2129.0040.2430.3880.194290.2430.3880.194
14307 136 43.7924.14 40.13861.1743.7824.1440.13861.0530.0040.2240.4170.192300.2240.4170.192
15317.66 14.97.08 46.8524.87 40.261014.4346.8424.8740.261014.2931.0040.2150.4190.199310.2150.4190.199
16326.54 14.547.62 53.3725.31 37.841005.3553.3525.3137.831005.2432.0030.1920.4260.223320.1920.4260.223
17337.5 12.545.96 46.1728.03 44.75964.5346.1628.0344.76964.4333.0030.2360.3940.187330.2360.3940.187
18345 13.725.88 61.2524.24 42.7889.8561.2224.2442.72889.7934.0020.1750.480.206340.1750.4810.206
19356.72 11.087 52.0332.19 40.38989.0652.0332.1940.38988.9935.0020.2190.3620.228350.220.3620.229
20367.98 11.287.6 50.0335.45 41.791116.8150.0335.4541.791116.7736.0010.2370.3360.226360.2380.3360.226
21377.98 12.546.18 50.9632.88 50.071128.3550.9532.8850.081128.3437.0010.240.3770.186370.240.3770.186
22385.32 11.924.84 61.0828.97 52.19922.8561.0828.9652.2922.8737.9990.20.4480.182380.20.4490.182
23396.32 13.064.18 57.1629.04 63.151004.5357.1929.0463.121004.5838.9980.2230.4610.147390.2230.4610.147
24405.22 13.94.54 67.8327.72 59.621010.0867.8527.7159.651010.1339.9980.1880.50.163400.1880.5010.164
25418.98 14.846 56.1534.81 61.611390.4756.1634.8261.61390.6440.9960.2410.3990.161410.2420.40.161
26426.26 11.167.82 67.7839.65 44.731216.5967.7839.6644.741216.841.9940.1990.3550.249420.1990.3550.249
27438.38 11.67.56 59.5443.38 51.31389.9359.5543.3851.321390.2142.9920.2380.330.215430.2390.330.215
28447.22 13.144.46 61.6535.23 71.511226.9261.735.2471.471227.2343.990.2350.4280.145440.2360.4280.145

XIII. Production Function to Cost Function: The Method of Langrange

The dual cost function, C*(q; wL, wK, wM), is obtained from a production function, q = F(L, K, M), by the constrained optimization:

C*(q; wL, wK, wM) = minL,K,M{ wL * L + wK * K + wM * M   :   q - F(L,K,M) = 0,   q > 0, wL > 0; wK > 0, and wM > 0,   L > 0, K > 0, M > 0 }     (*)

Define the Langrangian function, G, of the least-cost problem (*):

G(q; wL, wK, wM, L, K, M, μ) = wL * L + wK * K + wM * M + μ * (q - F(L,K,M))

where the new variable, μ, is called the Lagrange multiplier.

a.   First Order Necessary Conditions:

 0.   Gµ(q; wL, wK, wM, L, K, M, μ) = q - F(L, K, M) = 0 1.   GL(q; wL, wK, wM, L, K, M, μ) = wL - µ * FL(L, K, M) = 0 2.   GK(q; wL, wK, wM, L, K, M, μ) = wK - µ * FK(L, K, M) = 0 3.   GM(q; wL, wK, wM, L, K, M, μ) = wM - µ * FM(L, K, M) = 0

b.   Solution Functions:

We want to solve, simultaneously, these four equations for the variables µ, L, K, and M as functions of the variables q, wL, wK, and wM, and the parameters of the production function.

µ = µ(q; wL, wK, wM)
L = L(q; wL, wK, wM)
K = K(q; wL, wK, wM)
M = M(q; wL, wK, wM)

c.   Jacobian Matrices:

The Jacobian matrix (bordered Hessian of G) of the four functions, Gµ, GL, GK, GM, with respect to the choice variables, μ, L, K, and M:

J3   =

 Gµµ GµL GµK GµM GLµ GLL GLK GLM GKµ GKL GKK GKM GMµ GML GMK GMM
=
 0 -FL -FK -FM -FL -µ * FLL -µ * FLK -µ * FLM -FK -µ * FKL -µ * FKK -µ * FKM -FM -µ * FML -µ * FMK -µ * FMM

=   Jµ, L, K, M

The bordered principal minor of the bordered Hessian of the Langrangian function, G:

J2   =

 0 -FL -FK -fL -µ * FLL -µ * FLK -fK -µ * FKL -µ * FKK

d.   Second Order Necessary Conditions:

The second order necessary conditions require that the Jacobian matrix (bordered Hessian of G) be positive definite at the solution vector Z = (q, wL, wK, wM, L, K, M, µ). The Jacobian matrix, J3, is a positive definite matrix if the determinants of J2 and J3 are both negative.

e.   Sufficient Conditions:

If the second order necessary conditions are satisfied, then the first order necessary conditions are sufficient for a minimum at Z.

f. The Solution Functions' Comparative Statics.

The Jacobian matrix of the four functions, Gµ, GL, GK, GM, with respect to the variables, q, wL, wK, and wM:

Jq, wL, wK, wM;   =

 Gµq GµwL GµwK GµwM GLq GLwL GLwK GLwM GKq GKwL GKwK GKwM GMq GMwL GMwK GMwM
=
 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1

The Jacobian matrix of the four solution functions, Φ = {µ, L, K, M}, with respect to the variables, q, wL, wK, and wM:

JΦ   =

 µq µwL µwK µwM Lq LwL LwK LwM Kq KwL KwK KwM Mq MwL MwK MwM

From the Implicit Function Theorem:

Jq, wL, wK, wM;   +   Jµ, L, K, M   * JΦ   =   0 (zero matrix)   →

JΦ   =   - (Jµ, L, K, M)-1

for points (q; wL, wK, wM) in a neighborhood of the point (q; wL, wK, wM), and L = L(q; wL, wK, wM), K = K(q; wL, wK, wM), M = M(q; wL, wK, wM), and µ = µ(q; wL, wK, wM).

XIV. Question C == C*?

The obtained values of the solution functions at q, wL, wK, and wM:

µ = µ(q; wL, wK, wM)
L = L(q; wL, wK, wM)
K = K(q; wL, wK, wM)
M = M(q; wL, wK, wM),

and the calculated value of the dual Translog cost function:

C*(q; wL, wK, wM) = wL * L + wK * K + wM * M

can be compared with the corresponding values of the estimated Translog cost function, and the underlying CES cost function. Moreover, these functions' corresponding comparative static values can also be compared.

XV. Example: Production Function to Cost Function:

The dual Translog cost function, C*, is obtained from the estimated Translog production function, F, by:

C*(q; wL, wK, wM) = minL,K,M{ wL * L + wK * K + wM * M   :   q - F(L,K,M) = 0,   q > 0, wL > 0; wK > 0, and wM > 0,   L > 0, K > 0, M > 0 }     (*)

With q = 30, wL = 7, wK = 13, and wM = 6, (*) becomes:

C*(30; 7, 13, 6) = minL,K,M{ 7 * L + 13 * K + 6 * M   :   30 - F(L,K,M) = 0,     L > 0, K > 0, M > 0 }     (*)

The Langrangian function, G, of the least-cost problem (*):

G(30; 7, 13, 6, L, K, M, μ) = 7 * L + 13 * K + 6 * M + μ * (30 - F(L,K,M))

where the new variable, μ, is called the Lagrange multiplier.

a.   First Order Necessary Conditions:

 0.   Gµ(30; 7, 13, 6, L, K, M, μ) = 30 - F(L, K, M) = 0 1.   GL(30; 7, 13, 6, L, K, M, μ) = 7 - µ * FL(L, K, M) = 0 2.   GK(30; 7, 13, 6, L, K, M, μ) = 13 - µ * FK(L, K, M) = 0 3.   GM(30; 7, 13, 6, L, K, M, μ) = 13 - µ * FM(L, K, M) = 0

b.   Solution Functions:

We want to solve, simultaneously, these four equations for the variables µ, L, K, and M as functions of the variables q, wL, wK, and wM, and the parameters of the production function, so that.

µ = µ(30; 7, 13, 6)
L = L(30; 7, 13, 6)
K = K(30; 7, 13, 6)
M = M(30; 7, 13, 6)

Suppose we think that the factor inputs for q = 30 are about 20 units. Estimate L = 20, K = 20, and M = 20, and guess µ = 10 to start Newton's Method.

Newton's Method:

Using these estimates, Newton's Method provides:

Nonlinear Optimization: Newton's Method
Parameter Estimates
Iter #µLKM
0   10 2020 20
131.7519 48.768.2884 45.3423
237.0053 48.2116.8994 43.3003
332.0799 45.2122.0101 41.3977
431.2745 43.8523.9671 40.1888
531.2087 43.7924.1404 40.1368
631.2083 43.7924.1414 40.1365
731.2083 43.7924.1414 40.1365

c. Solution Vector:

Z = (q, wL, wK, wM, L, K, M, µ) = (30, 7, 13, 6, 43.79, 24.14, 40.14, 31.21)

With µ = 31.21, L = 43.79, K = 24.14, and M = 40.14, the first order conditions are:

 0.   Gµ(30; 7, 13, 6, L, K, M, μ) = 30 - F(43.79, 24.14, 40.14) = -0 1.   GL(30; 7, 13, 6, L, K, M, μ) = 7 - µ * FL(43.79, 24.14, 40.14) = -0 2.   GK(30; 7, 13, 6, L, K, M, μ) = 13 - µ * FK(43.79, 24.14, 40.14) = 0 3.   GM(30; 7, 13, 6, L, K, M, μ) = 13 - µ * FM(43.79, 24.14, 40.14) = -0

d. Second Order Necessary Conditions:

The second order necessary conditions require that the Jacobian matrix (bordered Hessian of G) be positive definite at the solution vector Z = (q, wL, wK, wM, L, K, M, µ). The Jacobian matrix, J3, is a positive definite matrix if the determinants of J2 and J3 are both negative.

Jµ, L, K, M   =

J3   =

 0 -FL -FK -FM -FL -µ * FLL -µ * FLK -µ * FLM -FK -µ * FKL -µ * FKK -µ * FKM -FM -µ * FML -µ * FMK -µ * FMM
=
 0 -0.224 -0.417 -0.192 -0.224 0.127 -0.115 -0.053 -0.417 -0.115 0.39 -0.098 -0.192 -0.053 -0.098 0.143

Determinant(J3) = -0.01608

J2   =

 0 -0.224 -0.417 -0.224 0.127 -0.115 -0.417 -0.115 0.39

Determinant(J2) = -0.06302

e. The Solution Functions' Comparative Statics:

JΦ   =

 C*q,q C*q,wL C*q,wK C*q,wM C*wL,q C*wL,wL C*wL,wK C*wL,wM C*wK,q C*wK,wL C*wK,wK C*wK,wM C*wM,q C*wM,wL C*wM,wK C*wM,wM
=
 µq µwL µwK µwM Lq LwL LwK LwM Kq KwL KwK KwM Mq MwL MwK MwM
=
 0.107 1.587 0.92 1.355 1.587 -3.421 1.096 1.617 0.92 1.096 -1.023 0.938 1.355 1.617 0.938 -3.919

The estimated Translog cost function's comparative statics.

 Cq,q Cq,wL Cq,wK Cq,wM CwL,q CwL,wL CwL,wK CwL,wM CwK,q CwK,wL CwK,wK CwK,wM CwM,q CwM,wL CwM,wK CwM,wM
=
 0.111 1.587 0.921 1.356 1.587 -3.422 1.096 1.617 0.921 1.096 -1.023 0.938 1.356 1.617 0.938 -3.919

The underlying specified Generalized CES cost function's comparative statics:

 Cq,q Cq,wL Cq,wK Cq,wM CwL,q CwL,wL CwL,wK CwL,wM CwK,q CwK,wL CwK,wK CwK,wM CwM,q CwM,wL CwM,wK CwM,wM
=
 0.107 1.587 0.921 1.355 1.587 -3.424 1.098 1.616 0.921 1.098 -1.024 0.937 1.355 1.616 0.937 -3.915

f. Comparing Least-Cost Combinations of Inputs:

With q = 30, wL = 7, wK = 13, and wM = 6,

Derived Dual Translog cost function factor demands: (L, K, M) = (43.79, 24.14, 40.14)

Estimated Translog cost function factor demands: (L, K, M) = (43.78, 24.14, 40.13)

Specified Generalized CES cost function factor demands: (L, K, M) = (43.79, 24.14, 40.13)

g. Comparing total costs and marginal costs:

Derived Dual Translog cost function: total cost = 861.17; marginal cost = µ = 31.21.

Estimated Dual Translog cost function: total cost = 861.05; marginal cost = 31.21.

Specified Generalized CES cost function: total cost = 861.17; marginal cost = 31.21.

XVI. Table of Results.

Check that the derived dual cost function corresponds with the estimated Translog cost function by comparing the values of the inputs, L, K, and M.

Translog Production / Cost Function Duality
Production Function to Cost Function
Generalized CES: Returns to Scale = 1, rho = 0.17647, rhoL = 0.17647, rhoK = 0.11823, rhoM = 0.26339
—   Generalized CES Data   — — Est. Translog Production — — Derived Dual Translog Cost —   — Estimated Translog Cost —
obs #qwLwKwMLKMcost q LK ML KMcostLKMcost
117 8.9 13.246.34 21.1513.69 23.59519.14 1721.1513.6923.59 21.1613.69 23.59519.14 21.1713.6923.61519.34
218 8.46 12.824.56 21.2713.54 29.57488.37 1821.2713.5429.57 21.2613.54 29.58488.37 21.2713.5429.59488.49
319 5.92 13.16.58 29.9713.85 22.95509.91 1929.9713.8522.95 29.9813.86 22.92509.9 29.9913.8622.93509.97
420 5.52 11.14.64 28.8114.47 27.59447.72 2028.8114.4727.59 28.8114.48 27.59447.72 28.8114.4727.59447.75
521 8.48 14.626.46 28.2816.29 29.32667.27 2128.2816.2929.32 28.2816.29 29.32667.27 28.2816.2829.32667.28
622 8.6 14.324 26.4215.64 40.67613.95 2226.4215.6440.67 26.4215.65 40.65613.95 26.4215.6540.66613.92
723 7.16 13.124.64 30.7116.82 35.98607.58 2330.7116.8235.98 30.7116.82 35.99607.58 30.7116.8235.99607.53
824 8.26 12.444.02 28.3918.46 41.95632.74 2428.3918.4641.95 28.3918.46 41.94632.73 28.3818.4641.95632.66
925 6.36 13.545.36 37.4118.1 35.12671.34 2537.4118.135.12 37.418.11 35.13671.34 37.418.135.12671.25
1026 7.6 14.685.04 35.8918.91 40.85756.17 2635.8918.9140.85 35.8918.91 40.85756.16 35.8818.940.84756.05
1127 7.7 13.167.56 39.2123.15 32.51852.33 2739.2123.1532.51 39.2123.15 32.5852.33 39.2123.1432.5852.2
1228 8.84 13.75.84 36.1823.22 41.28879.03 2836.1823.2241.28 36.1823.22 41.29879.03 36.1723.2141.29878.89
1329 8.96 14.37.14 39.5424.83 38.66985.36 2939.5424.8338.66 39.5424.83 38.66985.37 39.5324.8238.66985.21
1430 7 136 43.7924.14 40.13861.17 3043.7924.1440.13 43.7924.14 40.14861.17 43.7824.1440.13861.05
1531 7.66 14.97.08 46.8524.87 40.261014.43 3146.8524.8740.26 46.8524.87 40.261014.43 46.8424.8740.261014.29
1632 6.54 14.547.62 53.3725.31 37.841005.35 3253.3725.3137.84 53.3625.32 37.841005.36 53.3525.3137.831005.24
1733 7.5 12.545.96 46.1728.03 44.75964.53 3346.1728.0344.75 46.1728.03 44.76964.53 46.1628.0344.76964.43
1834 5 13.725.88 61.2524.24 42.7889.85 3461.2524.2442.7 61.2124.24 42.72889.87 61.2224.2442.72889.79
1935 6.72 11.087 52.0332.19 40.38989.06 3552.0332.1940.38 52.0432.19 40.38989.06 52.0332.1940.38988.99
2036 7.98 11.287.6 50.0335.45 41.791116.81 3650.0335.4541.79 50.0435.45 41.81116.82 50.0335.4541.791116.77
2137 7.98 12.546.18 50.9632.88 50.071128.35 3750.9632.8850.07 50.9632.87 50.081128.34 50.9532.8850.081128.34
2238 5.32 11.924.84 61.0828.97 52.19922.85 3861.0828.9752.19 61.0728.97 52.2922.84 61.0828.9652.2922.87
2339 6.32 13.064.18 57.1629.04 63.151004.53 3957.1629.0463.15 57.1929.05 63.081004.53 57.1929.0463.121004.58
2440 5.22 13.94.54 67.8327.72 59.621010.08 4067.8327.7259.62 67.8327.72 59.621010.06 67.8527.7159.651010.13
2541 8.98 14.846 56.1534.81 61.611390.47 4156.1534.8161.61 56.1734.81 61.571390.46 56.1634.8261.61390.64
2642 6.26 11.167.82 67.7839.65 44.731216.59 4267.7839.6544.73 67.7839.65 44.731216.59 67.7839.6644.741216.8
2743 8.38 11.67.56 59.5443.38 51.31389.93 4359.5443.3851.3 59.5543.36 51.311389.93 59.5543.3851.321390.21
2844 7.22 13.144.46 61.6535.23 71.511226.92 4461.6535.2371.51 61.6935.24 71.41226.95 61.735.2471.471227.23

Mathematical Notes

1. The Translog (Transcendental Logarithmic) Production Function:

 ln(q) = ln(A) + aL * ln(L) + aK * ln(K) + aM * ln(M)               + bLL * ln(L) * ln(L) + bKK * ln(K) * ln(K) + bMM * ln(M) * ln(M)                   + bLK * ln(L) * ln(K) + bLM * ln(L) * ln(M) + bKM * ln(K) * ln(M)   =   f(L,K,M).

an equation in 10 parameters, A, aL, aK, aM, bLL, bKK, bMM, bLK, bLM, bKM, where L = labour, K = capital, M = materials and supplies, and q = product.

2. The partial derivatives of f(L,K,M):

 fL(L,K,M) = (1/L) * [aL + 2 * bLL * ln(L) + bLK * ln(K) + bLM * ln(M)] = (1/L) * vL , fK(L,K,M) = (1/K) * [aK + 2 * bKK * ln(K) + bLK * ln(L) + bKM * ln(M)] = (1/K) * vK, fM(L,K,M) = (1/M) * [aM + 2 * bMM * ln(M) + bLM * ln(L) + bKM * ln(K)] = (1/M)* vM, fLL = (1/L^2) * [2 * bLL - vL],   fLK = bLK / (L*K),   fLM = bLM / (L*M),   fKK = (1/K^2) * [2 * bKK - vK],   fKL = bLK / (L*K),   fKM = bKM / (K*M), fMM = (1/M^2) * [2 * bMM - vM],   fML = bLM / (L*M),   fMK = bKM / (K*M).

3. The partial derivatives of F(L,K,M) = exp(f(L,K,M)):

 FL = fL * exp(f(L,K,M)),   FK = fK * exp(f(L,K,M)),   FM = fM * exp(f(L,K,M)), FLL = [fLL + fL * fL] * exp(f(L,K,M)),   FLK = [fLK + fL * fK] * exp(f(L,K,M)),   FLM = [fLM + fL * fM] * exp(f(L,K,M)),   FKL = FLK,   FKK = [fKK + fK * fK] * exp(f(L,K,M)),   FKM = [fKM + fK * fM] * exp(f(L,K,M)),   FML = FLM,   FMK = FKM,   FMM = [fMM + fM * fM] * exp(f(L,K,M)),

4. The Translog Cost Function:

 ln(C(q;wL,wK,wM)) = c + cq * ln(q) + cL * ln(wL) + cK * ln(wK) + cM * log(wM)                                 + .5 * [dqq * ln(q)^2 + dLL * ln(wL)^2 + dKK * ln(wK)^2 + dMM * ln(wM)^2]                   + .5 * [(dLK + dKL) * ln(wL)*ln(wK) + (dLM + dML) * ln(wL)*ln(wM) + (dKM + dMK) * ln(wK)*log(wM)]                   + dLq * ln(wL)*ln(q) + dKq * ln(wK)*ln(q) + dMq * ln(wM)*ln(q)   =   C(q;wL,wK,wM)

an equation in 18 parameters, c, cq, cL, cK, cM, dqq, dLL, dKK, dMM, dLK, dKL, dLM, dML, dKM, dMK, dLq, dKq, and dMq, where wL, wK, and wM are the factor prices of L, K, and M respectively.

Note: C(q; wL, wK, wM) = exp(C(q; wL, wK, wM)), a change in the notation above where C = CES cost function.

5. The Factor Share Functions:

 ∂ln(C)/∂ln(wL) = (∂ln(C)/∂wL )/ (∂ln(wL)/∂wL) = (1 / C(q;wL,wK,wM)) * ∂ln(C)/∂wL * wL = wL * L(q; wL, wK, wM) / C(q;wL,wK,wM) = sL(q;wL,wK,wM), ∂ln(C)/∂ln(wK) = (∂ln(C)/∂wK )/ (∂ln(wK)/∂wK) = (1 / C(q;wL,wK,wM)) * ∂ln(C)/∂wK * wK = wK * K(q; wL, wK, wM) / C(q;wL,wK,wM) = sK(q;wL,wK,wM), ∂ln(C)/∂ln(wM) = (∂ln(C)/∂wM )/ (∂ln(wM)/∂wM) = (1 / C(q;wL,wK,wM)) * ∂ln(C)/∂wM * wM = wM * M(q; wL, wK, wM) / C(q;wL,wK,wM) = sM(q;wL,wK,wM).

 sL(q;wL,wK,wM) = cL + dLq * ln(q) + dLL * ln(wL) + dLK * ln(wK) + dLM * ln(wM), sK(q;wL,wK,wM) = cK + dKq * ln(q) + dKL * ln(wL) + dKK * ln(wK) + dKM * ln(wM), sM(q;wL,wK,wM) = cM + dMq * ln(q) + dML * ln(wL) + dMK * ln(wK) + dMM * ln(wM),

6. The Factor Demand Functions:

 L(q;wL,wK,wM) = sL(q;wL,wK,wM) * C(q;wL,wK,wM) / wL, K(q;wL,wK,wM) = sK(q;wL,wK,wM) * C(q;wL,wK,wM) / wK, M(q;wL,wK,wM) = sM(q;wL,wK,wM) * C(q;wL,wK,wM) / wM.

7. The partial derivatives of C(q; wL, wK, wM) = exp(C(q; wL, wK, wM)):

 CwL(q;wL,wK,wM) = L(q;wL,wK,wM) = sL(q;wL,wK,wM) * C(q;wL,wK,wM) / wL, CwK(q;wL,wK,wM) = K(q;wL,wK,wM) = sK(q;wL,wK,wM) * C(q;wL,wK,wM) / wK, CwM(q;wL,wK,wM) = M(q;wL,wK,wM) = sM(q;wL,wK,wM) * C(q;wL,wK,wM) / wM,

Writing:

 Cq(q;wL,wK,wM) = (ecq + edqq * log(q) + edLq * log(wL) + edKq * log(wK) + edMq * log(wM) ) / q, Cq,q(q;wL,wK,wM) = (edqq - (ecq + edqq * log(q) + edLq * log(wL) + edKq * log(wK) + edMq * log(wM) )) / (q * q), Cq(q;wL,wK,wM) = Cq(q;wL,wK,wM) * C(q;wL,wK,wM), Cq,q(q;wL,wK,wM) = ( Cq,q(q;wL,wK,wM) + Cq(q;wL,wK,wM) * Cq(q;wL,wK,wM) ) * C(q;wL,wK,wM);

then:

 CwL,q(q;wL,wK,wM) = ( edLq * C(q; wL, wK, wM) / q + sL(q,wL,wK,wM) * Cq(q;wL,wK,wM) ) / wL, CwL,wL(q;wL,wK,wM) = ( L(q;wL,wK,wM) / wL ) * (-1 + sL(q,wL,wK,wM) + edLL / sL(q,wL,wK,wM)), CwL,wK(q;wL,wK,wM) = ( L(q,wL,wK,wM) / wK ) * (sK(q,wL,wK,wM) + edLK / sL(q,wL,wK,wM)), CwL,wM(q;wL,wK,wM) = ( L(q,wL,wK,wM) / wM ) * (sM(q,wL,wK,wM) + edLM / sL(q,wL,wK,wM)), CwK,q(q;wL,wK,wM) = ( edKq * C(q; wL, wK, wM) / q + sK(q,wL,wK,wM) * Cq(q;wL,wK,wM) ) / wK, CwK,wL(q;wL,wK,wM) = ( K(q;wL,wK,wM) / wL ) * (sL(q,wL,wK,wM) + edKL / sK(q,wL,wK,wM)), CwK,wK(q;wL,wK,wM) = ( K(q,wL,wK,wM) / wK ) * (-1 + sK(q,wL,wK,wM) + edKK / sK(q,wL,wK,wM)), CwK,wM(q;wL,wK,wM) = ( K(q,wL,wK,wM) / wM ) * (sM(q,wL,wK,wM) + edKM / sK(q,wL,wK,wM)), CwM,q(q;wL,wK,wM) = ( edMq * C(q; wL, wK, wM) / q + sM(q,wL,wK,wM) * Cq(q;wL,wK,wM) ) / wM, CwM,wL(q;wL,wK,wM) = ( M(q;wL,wK,wM) / wL ) * (sL(q,wL,wK,wM) + edML / sM(q,wL,wK,wM)), CwM,wK(q;wL,wK,wM) = ( M(q,wL,wK,wM) / wK ) * (sK(q,wL,wK,wM) + edMK / sM(q,wL,wK,wM)), CwM,wM(q;wL,wK,wM) = ( M(q,wL,wK,wM) / wM ) * (-1 + sM(q,wL,wK,wM) + edMM / sM(q,wL,wK,wM)), Copyright © Elmer G. Wiens:   Egwald Web Services All Rights Reserved.    Inquiries 