﻿
 UFDC Home <%MYSOBEK%>  |   Help  |   RSS
<%BANNER%>

 A MEMOIR ON NONLINEAR REGRESSION MODELS: Mathematical and Statistical Characteristics
CITATION PDF VIEWER
Full Citation
STANDARD VIEW MARC VIEW
 Material Information Title: A MEMOIR ON NONLINEAR REGRESSION MODELS: Mathematical and Statistical Characteristics Physical Description: Memoir Creator: Overman, Allen R. Publication Date: 2011
 Subjects Subjects / Keywords: modelsregressionelements of probability
 Notes Abstract: This memoir is focused on mathematical and statistical characteristics of nonlinear regression models, and includes a discussion on elements of probability. Particular models are chosen to illustrate various aspects of the procedures. A simple exponential model with two parameters is chosen as the first example. The model is rearranged to a linear form by performing logarithms on the response variable Y. This is referred to as the ‘linearized’ form of the model. Linear regression is then performed on ln Y vs. X (the control variable) to obtain estimates for the exponential parameter b along with the linear correlation coefficient r. Since the correlation coefficient is a measure of system response to the input variable X and reflects scatter in the response data, a decision is then made as to whether the linearized model is adequate or whether nonlinear regression is then needed. The ‘least squares criterion’ is used to determine ‘goodness of fit’ of the model to the data. Second order Newton-Raphson procedure is then selected to minimize the error sum of squares of deviations E between measured and estimated values of the response variable and to obtain optimum estimates of model parameters. In addition standard errors of parameter estimates are calculated using the Hessian matrix for the 2nd derivatives of E with respect to the parameters. This requires the inverse of the Hessian matrix along with the variance of the estimate, from which the standard errors of parameter estimates is obtained. The nonlinear correlation coefficient R is also used as a measure of goodness of fit of the model to the data. Contours of equal probability are then estimated for various levels of uncertainty using the Fisher F statistic. The memoir includes extensive discussion of elements of probability using the binomial expansion first for the natural numbers and then for n equal to fractions and negative values. For the natural numbers the expansion leads to finite series, whereas for fractions and negative values it leads to infinite series. All of this was established by Isaac Newton before he invented the calculus and for which he was appointed the second Lucasian professor of mathematics at Cambridge University, and led to his first memoir On the Analysis of Infinite Series. Coupling between discrete and continuous distributions are illustrated using the simple pegboard for linear, triangular, and rectangular configurations. This approach provides a logical foundation for the continuous Gaussian distribution of mathematical statistics. The procedure is further applied to role of dice for a single, two, three, and four dice. The pegboard is judged to be a simpler procedure to grasp and use in practice. Acquisition: Collected for University of Florida's Institutional Repository by the UFIR Self-Submittal tool. Submitted by Amy Buhler. Publication Status: Unpublished
 Record Information Source Institution: University of Florida Institutional Repository Holding Location: University of Florida Rights Management: All rights reserved by the submitter. System ID: IR00000578:00001

This item is only available as the following downloads:

Nonlinear_Models ( PDF )

Full Text

PAGE 1

A MEMOIR ON NONLINEAR REGRESSION MODELS Mathematical and Statis tical Characteristics Allen R. Overman Agricultural and Biological Engineering University of Florida Copyright 2010 Allen R. Overman

PAGE 2

Allen R. Overman Nonlinear Models ABSTRACT This memoir is focused on mathematical and st atistical characteristic s of nonlinear regression models, and includes a discussion on elements of probability. Partic ular models are chosen to illustrate various aspects of the procedures. A si mple exponential model with two parameters is chosen as the first example. The model is rea rranged to a linear form by performing logarithms on the response variable Y This is referred to as the linear ized form of the model. Linear regression is then performed on ln Y vs. X (the control variable) to obtain estimates for the exponential parameter b along with the linear correlation coefficient r Since the correlation coefficient is a measure of system response to the input variable X and reflects scatter in the response data, a decision is then made as to wh ether the linearized model is adequate or whether nonlinear regression is then neede d. The least squares criterion is used to determine goodness of fit of the model to the data. Second orde r Newton-Raphson procedure is then selected to minimize the error sum of squares of deviations E between measured and estimated values of the response variable and to obtain optimum estimates of model para meters. In addition standard errors of parameter estimates are calculated us ing the Hessian matrix for the 2nd derivatives of E with respect to the parameters. This requires the inverse of th e Hessian matrix along with the variance of the estimate, from which the standard errors of parameter estimates is obtained. The nonlinear correlation coefficient R is also used as a measure of goodness of fit of the model to the data. Contours of equal probabi lity are then estimated for various levels of uncertainty using the Fisher F statistic. The memoir includes extensive discussion of elements of probability using the binomial expansion first for the natural numbers ,3,2,1 n and then for n equal to fractions and negative values. For the natural numbers the expansion leads to finite series, whereas for fractions and negative values it leads to infini te series. All of this was established by Isaac Newton before he invented the calculus and for which he was appointed the second Lucasian professor of mathematics at Cambridge University, and led to his first memoir On the Analysis of Infinite Series Coupling between discrete and continuous distributions are illustrated using the simple pegboard for linear, triangu lar, and rectangular configurations. This approach provides a logical foundation for the continuous Gaussian di stribution of mathematical statistics. The procedure is further applied to role of dice for a single, two, three, and fo ur dice. The pegboard is judged to be a simpler procedure to grasp and use in practice. Keywords : Models, regression, elements of probability. Acknowledgement: The author expresses appreciati on to Amy G. Buhler, Associate University Librarian, University of Florida, for assistance with preparation of this memoir as part of the UF digital library. i

PAGE 3

Allen R. Overman Nonlinear Models ii Table of Contents Introduction Mathematical Characteristics Nonlinear Model Linearized Form of the Model Least Squares Criterion Newton Raphson Procedure for Nonlinear Regression Standard Errors of the Estimates Equal Probability Contours of Parameters Error Sum of Squares Near the Optimum Maximum Likelihood and Least Squares Analysis Statistical Analysis of the Model Linearized Form of the Model Nonlinear Regression Standard Errors of the Estimates Equal Probability Contours Dependence of E on b Near Minimum E Summary References Tables 1. Dependence of a response variable (Y ) to a control variable ( X ). 2. Newton Raphson iterations for nonlinear regression of the exponential model. 3. Newton Raphson iterations of the exponential model for initial b = .5000. 4. Combinations of A and b to satisfy equal probability equation near minimum E 5. Combinations of A and b to satisfy equal probability e quation for 75% probability. 6. Combinations of A and b to satisfy equal probability e quation for 95% probability. 7. Combinations of A and b to satisfy equal probability e quation for 99% probability. 8. Correlation of E with b using a parabolic model. Figures 1. Dependence of response variable ( Y ) on control variable ( X ). Data from Table 1. Curve drawn from Eq. (30). 2. Dependence of ln Y on X Data from Table 1. Li ne drawn from Eq. (27). 3. Scatter plot for estimated response variable ( Y ) vs. measured response variable ( Y ). Line represents the 45% diagonal. 4. Equal probability contours between parameters A and b Contours drawn from Table 5 (75%), Table 6 (95%), and Table 7 (99%) probability levels, respectively. Optimum and standard error values of 075.0016.5 A and 0129.05161.0 b are also shown. 5. Dependence of error sum of squares ( E ) on exponential parameter ( b ) for linear parameter ( A = 5.016). Parabola drawn from Eq. (67).

PAGE 4

Allen R. Overman Nonlinear Models Introduction Scientific analysis generally involves tw o essential components: (1) a set of data (measurements or observations) and (2) a conceptual model. The process of drawing inference about the system involves uncerta inty in both the data and in the model. In the case of an algebraic mathematical model, regression analysis is used to evaluate the parameters in the model. In regression analysis it is common to minimize the sum of squares of deviations between measured and estimated values of the response vari able as the criterion of goodness of fit of the model to the data. If all the parameters in the model occur in linear form (such as linear, quadratic, cubic, etc.), then the procedure is called linear regression If one or more of the parameters in the model occur in nonlinear form (such as exponential), then the procedure is called nonlinear regression Linear regression is the simpler of the two since it involves linear algebra, whereas nonlinear regressi on involves an iterative proce dure to estimate the parameters. Both methods are illustrated in this memoir. A variety of statistical measures are used to describe the quality of a model with a particular set of data. The first step is optimization of the model to obtain best estimates of the parameters. The next step is to calculate standard errors of the parameter estimates to determine uncertainty in the parameters. Relative error of an estimate is then calculated as standard error divided by the estimate. A scatter plot of estimated vs. measured response variable is often included to illustrate scatter of values and any evidence of bias in the model. It is al so possible to draw contours of equal probability (uncertainty) between two parameters by use of Fishers F statistic. Many of these points have been addre ssed in a previous publication (Overman et al., 1990) describing crop response to applied nitrog en with a logistic model. In this document a simple exponential model with one linear and one nonlinear parameter is applied to a set of data. Mathematical and statis tical characteristics are discussed in detail to illustrate the various steps involved. Mathematical Characteristics Nonlinear Model Consider the nonlinear regression model bXAY exp (1) where X is the control variable Y is the response variable A is the linear model parameter, and b is the exponential model parameter. For this discussion we consider X to be positive ( ) and Y to be positive ( ). It follows that parameter A m ust be positive as well. Parameter b can be positive or negative. For positive b it turns out that whereas for negative b we have Equation (1) is considered non linear in the regression sense because of the exponential parameter b 0 X0 Y/ dXdY 0 /, dXdYAY 0 AY Linearized Form of the Model 1

PAGE 5

Allen R. Overman Nonlinear Models Now Eq. (1) can be converted to a li near form by performing logarithms on Y bXabXAYZ lnln (2) where a = ln A Parameters a and b can then be estimated by linear regression of Z vs. X In Eq. (2) ln represents the natural logarithm. In some cases these estimates of parameters may be deemed sufficient for the purpose at hand. In ot her cases a more rigorous procedure may be desired, such as nonlinear regression. Least Squares Criterion For regression analysis we be gin with a criterion for goodness of fit of the model to the data. Define the error sum of squares of deviations ( E ) between data and model by n i iiYYE1 2 (3) where n is the num ber of observations, Yi is the observed value, and is the estimated value from the model. The goal is to choose parameters A and b to minimize E This is called the least squares criterion for goodness of fit of the model to the data. For the exponential model this takes the form iY n i i ibXAYE1 2exp (4) For regression purposes think of E as a function of A and b say E = E ( A, b ). At the minimum value of E it can be shown from calculus that 0 db b E dA A E dE (5) To minimize E w.r.t. (with respect to) A and b requires that b E A E 0 (6) simultaneously. This is called the necessary condition for a minimum. To insure a minimum, the sufficient condition from calculus is 0 and02 2 2 2 b E A E (7) 2

PAGE 6

Allen R. Overman Nonlinear Models The partial derivatives can be obtained fr om Eq. (4) and are given by the equations bX AbXY A E 2exp exp 2 (8) bX A E 2exp22 2 (9) bXXAbXXY Ab E bA E 2exp2exp22 2 (10) (11) bXXAbXXYA b E 2exp exp 2 bXXAbXYXA b E 2exp 2exp 22 2 2 2 (12) The subscripts have been omitted for convenience and the cross derivatives have been included for later use in the analysis. The derivative in Eq. (8) can be set to zero, which leads to bX bXY A 2exp exp (13) Equation (13) gives the optimum estimate of linear parameter A for an assumed value of b Setting Eq. (11) to zero leads to an implicit equation in parameter b An iterative procedure is needed to find b which will cause Eq. (11) to vanish. The second order Newton Raphson procedure is chosen for this purpose (Adby and Dempster, 1974). Newton-Raphson Procedure for Nonlinear Regression An initial estimate of parameter b is chosen in the neighborhood of minimum E Since we can treat E as a continuous function of b the derivative at a new value, say can be rela ted to the derivative at b by Taylor series expansion bb 3 4 4 2 3 3 2 2)( !3 1 )( !2 1 b b E b b E b b E b E b Eb b b bb (14) It is implicitly assumed that the series represen ted by Eq. (14) converges to a finite value. The strategy is to set this ne w derivative to zero and to truncate the series with the linear term in b which leads to 3

PAGE 7

Allen R. Overman Nonlinear Models b bbE bE b2 2/ / (15) A new estimate of parameter b is obtained from bbb (16) New estimates are then obtained for A bE /, 2 2/ bE b and b The procedure is repeated until the criterion is met b b (17) where is typically chosen as 10-3 to 10-5. The final values obtaine d by this procedure are chosen as optimum and minimum bb AA E E assuming that the procedure converges. It is necessary to choose the initial value of b near the optimum value to insure convergence of the procedure. Convergence requires that the second derivative in Eq. (15) be positive. Standard Errors of the Estimates The next step is to calculate the standard er rors of the estimates of the parameters. This procedure requires calculation of the Hessian matrix [ H ] of the second order derivatives given by Ab bb bA Ab AAb E Ab E bA E A E HH HH H, 2 2 2 2 2 2 (18) where the derivatives are evaluated at ( ). Since the cross derivatives are equal, it follows that the Hessian matrix is symmetric. The invers e of the Hessian matrix yields the elements Ab 1 1 1 1 1 bb bA Ab AAHH HH H (19) where the inverse Hessian is also symmetric. The variance of the estimate is defined by 2 XYS pn YY Sii XY 2 2 (20) where p is the number of parameters in the model. The standard errors of the estimates are then given by 4

PAGE 8

Allen R. Overman Nonlinear Models 2/1 12 AA XYHSA (21) 2/1 12 bb XYHSb (22) The covariance of the estimate is given by 12 ,COV Ab XYHSbA (23) Standard errors of the estimates provide a measur e of uncertainty in the parameter estimates for a given model and a particular set of data. Equal Probability Contours of Parameters The next mathematical characteristic which we explore is equal probability contours of A vs. b around the optimum for a chosen level of uncerta inty. Note that minimum error is calculated from n i i iXbAYbAE1 2exp (24) Now the error at some level of probability, q, is related to E by (Draper and Smith, 1981, p. 472) qpnpF pn p bAEbAE ,, 1, (25) where p is the number of parameters in the model, q is the probability level, and F is taken from tables for Fishers analysis of variance ( F statistic). The goal is to obtain combinations of parameters A and b which satisfy bAEbXAYn i i i, exp1 2 = constant (26) This leads to a plot of A vs. b which satisfies Eq. (26), and leads to an equal probability contour. Error Sum of Squares Near the Optimum The final characteristic which we explore is to examine E vs. b at fixed value of AA For the case of a linear model this result follows a parabola. Does this relationship hold for the nonlinear exponential model? If so, then we should obtain the parabola 2bbE (27) 5

PAGE 9

Allen R. Overman Nonlinear Models where are estimated from values of E vs. b near the optimum (minimum). Maximum Likelihood and Least Squares Analysis This section focuses on the connection betw een the maximum likelihood method of Fisher and that of the least squares crit erion (Frieden, 1983, chapter 14). The challenge is to calibrate a mathematical model to relate the dependent variable ( y ) to the independent variable ( x ). Assume that the error in the measurements of yi follows a Gaussian probability density function 2 22 )( exp 2 1 )( i iy yp (28) with mean of and variance of If we further assume that the error in the predicted values ( ) from the model also follow this same error law with the same mean and variance, then the probability density function for the e rror between measured and estimated y is given by 2iy 2 22 ) ( exp 2 1 ) ,( ii iiyy yyp (29) For n observations we can assume the joint probability ( p ) given by the product of individual terms n i ii n i iyy pyyp1 2 2 12 ) ( exp 2 1 ) ,( (30) This is referred to as the maximum likelihood principle when the parameters of the model have been chosen to maximize the function given by Eq. (30). Such a choice will also maximize the logarithm of p n i ii n i ii nyy nyy p1 2 2 1 2 2 2 1 2ln 2 1 2 1 lnln (31) In order for ln p to be a maximum, it follows that the error ( E ) defined by n i iiyyE1 2 (32) must be a minimum. Equation ( 32) therefore defines the least squares error between measured and predicted values of y based on the assumptions stated. 6

PAGE 10

Allen R. Overman Nonlinear Models Statis tical Analysis of the Model Linearized Form of the Model In this section the procedure is applied to the pa rticular set of data list ed in Table 1. The first step is to plot the data to see the tren d and scatter (see Figure 1). The decrease in Y with increase in X appears to follow an expone ntial pattern with negative b. The next step is to plot ln Y vs. X to test this hypothesis (see Figur e 2). Since Figure 2 appears to follow a straight line, linear regression of ln Y vs. X leads to the regression equation X bXaYZ 5524.0670.1 ln r = .9945 (33) with a correlation coefficient of r = .9945. This leads to the prediction equation X Y 5524.0exp31.5 (34) It should be noted that E q. (34) does not minimize E for Eq. (1), but instead minimizes the error sum of squares for Z For some purposes Eq. (34) may be deemed adequate for analysis. A more rigorous procedure follows nonlinear regression. The value of parameter b = .5524 is then used as a first estimate in the iteration procedure. Nonlinear Regression We now outline the nonlinear regression proced ure in detail, as given in Table 2. 5524.0 b 1141.53508.2/0858.122exp/exp bX bXYA 4083.35643.11411.53737.81411.52 2exp exp 2/ bXXAbXXYAbE 7920.1157548.21411.520640.171411.52 2exp 2exp 2/2 2 2 2 bXXAbXYXAbE 0294.07920.115/4083.3 ///2 2 bEbEb 5230.00294.05524.0 bbb 5230.0 b 0409.54476.2/3381.12 A 7008.07382.10409.58316.80409.52/ bE 5555.1601844.30409.521792.160409.52/2 2 bE 0044.05555.160/7008.0 b 5186.00044.05230.0 b 7

PAGE 11

Allen R. Overman Nonlinear Models 5186.0 b 0251.54631.2/3773.12 A 2619.07666.10251.59034.80251.52/ bE 4710.1642559.30251.523575.160251.52/2 2 bE 0016.04710.164/2619.0 b 5170.00016.05186.0 b 5170.0 b 0194.54687.2/3915.12 A 1017.07770.10194.59296.80194.52/ bE 9168.1652823.30194.524228.160194.52/2 2 bE 0006.09168.165/1017.0 b 5164.00006.05170.0 b 5164.0 b 0173.54708.2/3968.12 A 0380.07810.10173.59396.80173.52/ bE 4853.1662925.30173.524478.160173.52/2 2 bE 0002.04853.166/0380.0 b 5162.00002.05164.0 b 5162.0 b 0165.54715.2/3984.12 A 0200.07822.10165.59424.80165.52/ bE 6255.1662954.30165.524550.160165.52/2 2 bE 00012.06255.166/0200.0 b 5161.00001.05162.0 b 5161.0 b 0161.547194.2/39956.12 A 00645.0 78305.10161.594460.80161.52/ bE 7581.166 29766.30161.5246050.160161.52/2 2 bE 000039.07581.166/00645.0 b 5161.00000.05161.0 b 410000076.0 5161.0 000039.0 b b 8

PAGE 12

Allen R. Overman Nonlinear Models The iteratio ns are terminated at this poi nt and lead to the estimation equation X Y 5161.0exp016.5 (35) Note that the procedure converged to the final values in seven steps. The regression curve in Figure 1 is drawn from Eq. (35). A scatter plot of Y vs. Y is shown in Figure 3. The question now arises as to convergence if the first estimate is greater than the true value, say b = .5000. The steps are outlined below for th is case and are summarized in Table 3. 5000.0 b 9566.45311.2/5457.12 A 4253.08937.19566.43435.99566.42/ bE 9699.1815804.39566.421369.179566.42/2 2 bE 0023.09699.181/4253.0 b 5023.00023.05000.0 b 5023.0 b 9653.45224.2/5244.12 A 4465.18733.19653.41757.99653.4/ bE 5627.1795381.39653.420537.179653.42/2 2 bE 0081.05627.179/4465.1 b 5104.00081.05023.0 b 5104.0 b 9951.44927.2/4513.12 A 5872.08215.19951.40398.99951.4/ bE 0235.1723950.39951.426975.169951.42/2 2 bE 0034.00235.172/5872.0 b 5138.00034.05104.0 b 5138.0 b 0077.54802.2/4201.12 A 2348.07985.10077.59829.800771.5/ bE 8986.1683369.30077.525565.160077.52/2 2 bE 0014.08986.168/2348.0 b 5152.00014.05138.0 b 9

PAGE 13

Allen R. Overman Nonlinear Models 5152.0 b 0130.54751.2/4077.12 A 0908.07891.10130.59597.80130.5/ bE 6221.1673131.30130.524984.160130.52/2 2 bE 0005.06221.167/0908.0 b 5157.00005.05152.0 b 5157.0 b 0147.54734.2/4033.12 A 0376.07858.10147.59515.80147.5/ bE 1526.1673047.30147.524779.160147.52/2 2 bE 0002.01526.167/0376.0 b 5159.00002.05157.0 b 5159.0 b 0154.54726.2/4012.12 A 0118.07843.10154.59478.80154.5/ bE 9252.1663008.30154.524684.160154.52/2 2 bE 00007.09252.166/0118.0 b 5160.000007.05159.0 b 5160.0 b 0157.54724.2/4007.12 A 00744.07839.10157.59468.80157.5/ bE 8813.1662999.30157.524659.160157.52/2 2 bE 000045.08813.166/00744.0 b 5161.0 000045.05160.0 b 410000087.0 5160.0 000045.0 b b The procedure again converges to Eq. (35). 10

PAGE 14

Allen R. Overman Nonlinear Models Another measure of quality of fit of the model to the data is given by the nonlinear correlation coefficient defined by (Cornell and Berger, 1987) 99674.0 396364.23 152098.0 1 ( 12/1 2/1 2 2 YY YY Ri ii (36) which shows excellent agreement between model and data. The final derivatives are given by 000012.0 47194.20161.539956.1222exp exp2 bX AbXY A E (37) 094388.447194.222exp22 2 bX A E (38) 08866.1778305.10161.529446.82 2exp2exp22 2 bXXAbXXY Ab E bA E (39) 000645.0 78305.10161.594460.80161.52 2exp exp 2 bXXAbXXYA b E (40) 07581.166 29766.30161.5246050.160161.52 2exp 2exp 22 2 2 2 bXXAbXYXA b E (41) Note that the first derivatives are approximately zero and the second derivatives are positive as required. Standard Errors of the Estimates The second derivatives allow cal culation of the Hessian matrix 11

PAGE 15

Allen R. Overman Nonlinear Models 7581.1668866.17 8866.1794388.42 2 2 2 2 2b E Ab E bA E A E HH HH Hbb bA Ab AA (42) It follows that the inverse Hessian matrix becomes 0097995332 .04 0354540022 .0 4 0354540022 .07 3305402955 .01 1 1 1 1 bb bA Ab AAHH HH H (43) which is symmetric as required. The variance of the estimate is calculated from 016900.0 211 152098.0 2 2 pn YY Sii XY (44) It follows that the standard errors of the estimates and the covariance become 0747.07 3305402955 .0016900.02/1 2/1 12 AA XYHSA (45) 0129.0 0097995332 .0016900.02/1 2/1 12 bb XYHSb (46) 000599.04 0354540022 .0016900.0 ,COV12 Ab XYHSbA (47) Under ideal circumstances the co variance would be zero to signify that the model parameters were uncorrelated. Final estimates of parameters are 075.0016.5 A (48) 0129.05161.0b (49) with relative errors of %49.10149.0 016.5 0747.0 A A (50) %50.20250.0 5161.0 0129.0 b b (51) which are relatively small as desired. A check of the Hessian inverse shows that 12

PAGE 16

Allen R. Overman Nonlinear Models 10 01 5 9999999808 .00 0000000002 .0 206 0000000000 .09 9999999999 .0 0097995332 .04 0354540022 .0 4 0354540022 .07 3305402955 .0 7581.1668866.17 8866.1794388.41HH within roundoff as required. We also note that the determinant of the Hessian 0 501575868.504 930459560.319 432035428.824 7581.1668866.17 8866.1794388.4 is positive definite as required for convergence. Equal Probability Contours In this section we examine combinations of parameters A and b which lead to equal values of the error sum of squares E The optimized model is described by X XbAY 5161.0exp016.5 exp (52) which leads to the minimum error sum of squares of n ii i i iiX YYYE1 11 1 2 21521.0 5161.0exp016.5 (53) Values of E are calculated for 5161 .0,075.0016.5 b A 0129 and for Results for these values ar e listed in Table 4. Other combinations of A and b which lead to the same E are also given. These results provide a contour of equal probabilities which pass through the standard errors for A and b. From the table we note that for A = 5.016 we obtain ( b, E ) = (.5032, 0.1666) and (b, E ) = (.5290, 0.1652). .05161.0,016.5 bA This analysis is now extended to various levels of probability. 75 % probability contour It can be shown that the error at some leve l of probability, say 75%, can be calculated from 11 1 2exp 2069.062.1 211 2 11521.0%75,, 1,i i ibXAY pnpF pn p EbAE (54) where the value of F is obtained from statistical tables as F (2,9,75%) = 1.62. Combinations of A and b which satisfy Eq. (54) are listed in Table 5. A graph of the probability contour is shown in 13

PAGE 17

Allen R. Overman Nonlinear Models Figure 4, from which estim ates are made of ( b, E ) = (.491, 0.2069) and ( b, E ) = (.543, 0.2069). 95 % probability contour It can be shown that the error at the level of probability of 95% can be calculated from 11 1 2exp 2961.026.4 211 2 11521.0%95,, 1,i i ibXAY pnpF pn p EbAE (55) where the value of F is obtained from statistical tables as F (2,9,95%) = 4.26. Combinations of A and b which satisfy Eq. (55) are listed in Table 6 a nd shown in Figure 4. Estimates are made of ( b, E ) = (.477, 0.2961) and (b, E ) = (.560, 0.2961). 99 % probability contour It can be shown that the error at the level of probability of 99% can be calculated from 11 1 2exp 4232.002.8 211 2 11521.0%99,, 1,i i ibXAY pnpF pn p EbAE (56) where the value of F is obtained from statistical tables as F (2,9,99%) = 8.02. Combinations of A and b which satisfy Eq. (56) are listed in Table 7 a nd shown in Figure 4. Estimates are made of ( b, E ) = (.466, 0.4232) and (b, E ) = (.577, 0.4232). Dependence of E on b Near Minimum E A summary of E vs. b for which satisf y various probability levels is given Table 8. The question is whether or not this relationship follows a parabola given by 016 .5 AA 2bbE (57) To optimize the parabolic model requires that 14

PAGE 18

Allen R. Overman Nonlinear Models 2 4 3 2 3 2 2Eb Eb E bbb bbb bbn (58) where again n is the number of observations used in the analysis. Calculations are carried out for different numbers of points listed in Table 8. n = 9 4 6312315359 .0 21038451.1 3363.2 9 6648422154 .08 2660996762 .1 42264121.2 8 2660996762 .1 42264121.2 6581.4 42264121.2 6581.4 9 (59) which leads to the regression equation 25133.753473.784827.20 b b E (60) The minimum of the parabola occurs at 1608.0,5188.0 05133.7523473.78 E b b b E (61) which is inconsistent with Correlation between 1521 .0,5161.0 E b E and E is given by E E9166.00216.0 r = 0.99837 (62) n = 7 4 4000958010 .0 77084361.0 4902.1 2 5084421553 .08 9753885152 .0 87626821.1 8 9753885152 .0 87626821.1 6191.3 87626821.1 6191.3 7 (63) which leads to the regression equation 25410.835400.865631.22 b b E (64) The minimum of the parabola occurs at 1514.0,5179.0 05410.8325400.86 E bb b E (65) 15

PAGE 19

Allen R. Overman Nonlinear Models which is in consistent with Correlation between 1521 .0,5161.0 E b E and E is given by E E9958.000088.0 r = 0.99778 (66) While this is better than using all nine values, it is still off a bit. n = 5 4 2398675041 .0 46378791.0 8980.0 8 3583277494 .08 6912411822 .0 33513921.1 8 6912411822 .0 33513921.1 5821.2 33513921.1 5821.2 5 (67) which leads to the regression equation 27507.804693.837221.21 b b E (68) The minimum of the parabola occurs at 1522.0,5168.0 07507.8024693.83 E b b b E (69) which is near Correlation between 1521 .0,5161.0 E b E and E is given by E E99722.000046.0 r = 0.99905 (70) n = 3 4 1289835871 .0 24985331.0 4842.0 1 2132717681 .08 4127674042 .0 79920921.0 8 4127674042 .0 79920921.0 5481.1 79920921 5481.1 3 (71) which leads to the regression equation 25409.822091.851430.22 b b E (72) The minimum of the parabola occurs at 1520.0,5162.0 05409.8222091.85 E b b b E (73) 16

PAGE 20

Allen R. Overman Nonlinear Models which virtually ag rees with Correlation between 1521 .0,5161.0 E b E and E is given by E E00000.100010.0 r = 1 (74) Values of E vs. b are shown in Figure 5, where the pa rabola is drawn from Eq. (72). As b is changed away from the values deviate from the parabola. Note for b = .5524 that E is within the parabolic envelope. For b = .5000 note that E is also within the parabolic envelope. In both cases the proced ure converges toward the minimum at 5161 .0b1521.0 E5161.0b Note that it is important to carry a large num ber of digits to avoid roundoff errors in the matrix computational procedure. Summary This memoir has focused on the mathematical and statistical characte ristics of a nonlinear regression model. The model assu med an exponential relationship between the control variable (X) and the response variable (Y) as described by Eq. (1). Analys is was performed for a given set of data (Table 1). The first step was to linearize the model to the form of Eq. (2). A plot of the data supported this step as shown in Figure 2. Linear regression of Eq. (2) led to a first estimate of the parameters A and b. This estimate of b was then used to perf orm nonlinear regression of the model on the data to optimize the values of A and b in order to minimize the error sum of squares (E) between measured and estimated values of Y. It was shown that the Newton Raphson procedure converged rapidly to the minimum E. Standard errors of the parameters were then estimated which showed low relative errors in the parameters. A further measure of uncertainty in the parameters was illustrated by the equal probability contours for A vs. b for various levels of probability (see Figure 4). The cross in Figure 4 represents the most probable values of parameters A and b for the exponential model and for this particular set of data, i.e. the values which mi nimize the error sum of squares between measured and predicted values of response variable y. Vertical bars represen t the standard error in parameter b around optimum value while horizontal bars repres ent the s tandard error in parameter A around optimum value These are equivalent to the standard deviation around the m ean of a set of measurements which follow a Gaussian distribution. The contours in Figure 4 represent combinations of parameters A and b which produce various levels of uncertainty. Following Fishers maximum likelihood method, these represents combinations of equal probability. bA It was further shown that E vs. b at optimum A in the neighborhood of minimum E followed parabolic dependence. At this point it seems appropr iate to call attention to seve ral general points about data analysis and mathematical models. R. A. Fisher ca lled attention to two elem ents of uncertainty in this process in his classic ar ticle of 1922 (see Bennet, 1971). Un certainty in data led to his analysis of variance (ANOVA) procedure, while uncertainty in a model led to a subject called Fisher Information (see Frieden, 1998). In her biography of her father, Joan Fisher Box noted that the passion of Fishers life was the subject of inference (see Fisher Box, 1978, p. 447), i.e. drawing inference about a system from analysis of a specific set of data. 17

PAGE 21

Allen R. Overman Nonlinear Models Scientific research can be divided into two approaches: bottom/up and top/down. In the bottom/up approach data (measurements or obser vations) are examined in order to identify a unifying theory or model (from specific to genera l), which is commonly referred to as a process of induction. In the top/down approach a general princi ple is postulated and the consequences of these are developed (from general to specific), which is commonly referred to as the process of deduction. Most research appears to have follo wed the bottom/up approach. The top/down approach was championed by Einstein and by Paul Dirac (Farmelo, 2009, p. 2, 94, 382). Following a series of lectures by Murray Gell -Mann it appears that Dirac gained increased respect for the bottom/up approach, which Gell-Ma nn had followed. The work described in this memoir has followed the bottom/up approach. Of course we cant be certain that the simple exponential model is the very best model possible for the given data. There is always a level of uncertainty. Science progresses by assuming a theory or model and then checking the consequences of the theory through measurements. A final point has to do with pursuit of know ledge and understanding of how nature really works. I will call this the batt le between subjective and objective criteria for judging the values of ideas in science. According to James Glanz (see Chang, 2000, p. 354) the theoretical physicist Steven Weinberg has battled with thinkers and ph ilosophers of science over this issue. Today, one of his major battles is with postmodernist th inkers and philosophers of science who maintain that scientific theories reflect not objective reality but social ne gotiations among scientists. In its rawest form, this philosophy would say that the th eories of the most persuasive or politically powerful scientists become accepted fact. I was tr ained on the belief in objective criteria, and I still hold to this view. Otherwise, it becomes a battle for power and control of ideas in science based on personalities. Unfortunately I have observe d an increasing trend to cite experts as the source of truth in the evaluation of scientific ideas. Some editors and reviewers seem to find this an attractive alternative in the peer review process. References Adby, P.R. and M.A.H. Dempster. 1974. 1974. Introduction to Optimization Methods. John Wiley & Sons. New York, NY. Bennett, J.H. 1971. Collected Papers of R.A. Fisher. Vol 1 (1912-1924). University of Adelaide. Chang, L. 2000. Scientists at Work: Profil es of Todays Groundbreaking Scientists from Science Times. McGraw Hill. New York, NY. Cornell, J.A. and R.D. Berger. 1987. Factors th at influence the value of the coefficient of determination in simple linear and nonlinear regression models. Phytopathology 77:63-70. Draper, N.R. and H. Smith. 1981. Applied Regression Analysis. John Wiley & Sons. New York, NY. Farmelo, G. 2009. The Strangest Man: The Hidden Life of Paul Dirac, Mystic of the Atom. Basic Books. New York, NY. Fisher Box, J. 1978. R.A. Fisher: The Life of a Scientist. John Wiley & Sons. New York, NY. Frieden, B.R. 1983. Probability, Statistical Optics, and Data Testing. Springer-Verlag. New York, NY. Frieden, B.R. 1998. Physics from Fisher Information: A Unification. Cambridge University Press. New York, NY. Overman, A.R., F.G. Martin, and S.R. Wilkinson. 1990. A logistic equation for yield response of forage grass to nitrogen. Commun. Soil Science and Plant Analysis 21:595-609. 18

PAGE 22

Allen R. Overman Nonlinear Models Table 1. Dependence of a response variable ( Y ) on a control variable (X ). X Y ln Y 0.0 5.0 1.609 0.5 4.0 1.386 1.0 2.8 1.030 1.5 2.2 0.788 2.0 2.0 0.693 2.5 1.5 0.405 3.0 1.0 0.000 3.5 0.9 .105 4.0 0.6 .511 4.5 0.4 .916 5.0 0.3 1.204 Table 2. Newton Raphson iterations of the exponential model for initial b = .5524. X Y exp ( bX ) Y 0.0 5.0 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.00000 5.016 0.5 4.0 0.7587 0.7699 0.7716 0.7722 0.7724 0.7725 0.77256 3.875 1.0 2.8 0.5756 0.5927 0.5954 0.5963 0.5967 0.5968 0.59684 2.994 1.5 2.2 0.4367 0.4564 0.4594 0.4605 0.4609 0.4610 0.46110 2.313 2.0 2.0 0.3313 0.3513 0.3544 0.3556 0.3560 0.3561 0.35622 1.787 2.5 1.5 0.2513 0.2705 0.2735 0.2746 0.2750 0.2751 0.27520 1.380 3.0 1.0 0.1907 0.2082 0.2110 0.2120 0.2124 0.2125 0.21261 1.066 3.5 0.9 0.1447 0.1603 0.1628 0.1637 0.1641 0.1642 0.16425 0.824 4.0 0.6 0.1097 0.1234 0.1256 0.1264 0.1267 0.1268 0.12689 0.637 4.5 0.4 0.0833 0.0950 0.0969 0.0976 0.0979 0.0980 0.09803 0.492 5.0 0.3 0.0632 0.0732 0.0748 0.0754 0.0756 0.0757 0.07574 0.380 b .5524 .5230 .5186 .5170 .5164 .5162 .5161 .5161 A 5.1411 5.0409 5.0251 5.0194 5.0173 5.0165 5.0161 5.0161 bE / .4083 .7008 .2619 .1017 .0380 .0200 .00645 2 2/bE 115.7920 160.5555 164.4710 165.9168 166.4853 166.6255 166.7581 b 0.0294 0.0044 0.0016 0.0006 0.0002 0.0001 0.00004 b .5230 .5186 .5170 .5164 .5162 .5161 .51614 19

PAGE 23

Allen R. Overman Nonlinear Models Table 3. Newton Raphson iterations of the exponential model for initial b = .5000. X Y exp ( bX ) 0.0 5.0 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 0.5 4.0 0.7788 0.7779 0.7748 0.7734 0.7729 0.7727 0.7726 1.0 2.8 0.6065 0.6051 0.6003 0.5982 0.5974 0.5971 0.5970 1.5 2.2 0.4724 0.4707 0.4651 0.4627 0.4617 0.4614 0.4612 2.0 2.0 0.3679 0.3662 0.3603 0.3579 0.3569 0.3565 0.3564 2.5 1.5 0.2865 0.2849 0.2792 0.2768 0.2758 0.2755 0.2753 3.0 1.0 0.2231 0.2216 0.2163 0.2141 0.2132 0.2129 0.2127 3.5 0.9 0.1738 0.1724 0.1676 0.1656 0.1648 0.1645 0.1644 4.0 0.6 0.1353 0.1341 0.1298 0.1281 0.1274 0.1271 0.1270 4.5 0.4 0.1054 0.1043 0.1006 0.0991 0.0984 0.0982 0.0981 5.0 0.3 0.0821 0.0811 0.0779 0.0766 0.0761 0.0759 0.0758 b .5000 .5023 .5104 .5138 .5152 .5157 .5159 A 4.9566 4.9653 4.9951 5.0077 5.0130 5.0147 5.0154 bE / +0.4253 +1.4465 +0.5872 +0.2348 +0.0908 +0.0376 +0.0118 2 2/bE 181.9699 179.5627 172.0235 168.8986 167.6221 167.1526 166.9252 b .0023 .0081 .0034 .0014 .0005 .0002 .00007 b .5023 .5104 .5138 .5152 .5157 .5159 .5160 Table 3. (Continued). X Y exp ( bX ) Y Y Y 0.0 5.0 1.000000 1.000000 1.000000 1.000000 1.0000000 5.016 .016 0.5 4.0 0.772607 0.772589 0.772583 0.772580 0.7725794 3.875 +0.125 1.0 2.8 0.596921 0.596894 0.596884 0.596880 0.5968789 2.994 .194 1.5 2.2 0.461185 0.461154 0.461143 0.461138 0.4611364 2.313 .113 2.0 2.0 0.356315 0.356283 0.356271 0.356266 0.3562644 1.787 +0.213 2.5 1.5 0.275291 0.275260 0.275249 0.275244 0.2752426 1.381 +0.119 3.0 1.0 0.212692 0.212663 0.212652 0.212648 0.2126467 1.067 .067 3.5 0.9 0.164327 0.164301 0.164292 0.164288 0.1642865 0.824 +0.076 4.0 0.6 0.126960 0.126938 0.126929 0.126925 0.1269244 0.637 .037 4.5 0.4 0.098090 0.098071 0.098063 0.098060 0.0980591 0.492 .092 5.0 0.3 0.075785 0.075768 0.075762 0.075759 0.0757585 0.380 .080 b .51597 .516015 .516032 .516039 .516041 .5160 A 5.01566 5.01582 5.01588 5.01591 5.0159173 5.0160 bE / +0.007444 +0.002792 +0.001101 +0.000301 +0.000256 2 2/ bE 166.881252 166.840459 166.825228 166.819155 166.817429 b .000045 .000017 .000007 .000002 .0000015 b .516015 .516032 .516039 .516041 .516042 20

PAGE 24

Allen R. Overman Nonlinear Models Table 4. Combinations of A and b to satisfy equal probability equation near minimum E X Y Y 0.0 5.0 4.941 5.091 5.016 5.016 4.970 5.110 5.060 4.920 0.5 4.0 3.817 3.933 3.900 3.850 3.828 3.936 3.921 3.813 1.0 2.8 2.950 3.039 3.033 2.955 2.949 3.032 3.039 2.954 1.5 2.2 2.278 2.347 2.358 2.269 2.271 2.335 2.355 2.289 2.0 2.0 1.760 1.814 1.834 1.741 1.750 1.799 1.825 1.774 2.5 1.5 1.360 1.401 1.426 1.337 1.348 1.386 1.414 1.375 3.0 1.0 1.051 1.082 1.109 1.026 1.038 1.067 1.096 1.065 3.5 0.9 0.812 0.836 0.862 0.788 0.800 0.822 0.849 0.826 4.0 0.6 0.627 0.646 0.670 0.604 0.616 0.633 0.658 0.640 4.5 0.4 0.484 0.499 0.521 0.464 0.474 0.488 0.510 0.496 5.0 0.3 0.374 0.386 0.405 0.356 0.365 0.376 0.395 0.384 b .5161 .5161 .5032 .5290 .522 .522 .510 .510 A 4.941 5.091 5.016 5.016 4.97 5.11 5.06 4.92 E 0.1660 0.1665 0.1664 0.1657 0.1647 0.1668 0.1653 0.1673 Target is E = 0.1660 Table 4. (Continued). X Y Y 0.0 5.0 5.110 4.930 5.017 5.017 4.970 5.050 5.080 0.5 4.0 3.922 3.833 3.901 3.851 3.871 3.871 3.894 1.0 2.8 3.011 2.981 3.033 2.956 3.014 2.967 2.984 1.5 2.2 2.311 2.318 2.359 2.269 2.348 2.274 2.287 2.0 2.0 1.774 1.802 1.834 1.742 1.828 1.743 1.753 2.5 1.5 1.362 1.401 1.426 1.337 1.424 1.336 1.344 3.0 1.0 1.045 1.090 1.109 1.026 1.109 1.024 1.030 3.5 0.9 0.802 0.847 0.862 0.788 0.864 0.785 0.789 4.0 0.6 0.616 0.659 0.670 0.605 0.673 0.601 0.605 4.5 0.4 0.473 0.512 0.521 0.464 0.524 0.461 0.464 5.0 0.3 0.363 0.398 0.405 0.356 0.408 0.353 0.355 b .5290 .5032 .5032 .5290 .5000 .5320 .5320 A 5.11 4.93 5.017 5.017 4.97 5.05 5.08 E 0.1663 0.1650 0.1666 0.1652 0.1662 0.1658 0.1648 Target is E = 0.1660 21

PAGE 25

Allen R. Overman Nonlinear Models Table 5. Combinations of A and b to satisfy equal probability equation for 75% probability. X Y Y 0.0 5.0 5.160 4.870 5.210 4.960 5.080 4.830 5.140 5.130 0.5 4.0 3.986 3.762 3.987 3.796 3.956 3.762 3.904 3.897 1.0 2.8 3.080 2.907 3.051 2.905 3.081 2.930 2.966 2.960 1.5 2.2 2.379 2.246 2.335 2.223 2.400 2.282 2.253 2.248 2.0 2.0 1.838 1.735 1.787 1.701 1.869 1.777 1.711 1.708 2.5 1.5 1.420 1.340 1.368 1.302 1.455 1.384 1.300 1.297 3.0 1.0 1.097 1.035 1.047 0.996 1.134 1.078 0.987 0.985 3.5 0.9 0.848 0.800 0.801 0.763 0.883 0.839 0.750 0.748 4.0 0.6 0.655 0.618 0.613 0.584 0.688 0.654 0.570 0.568 4.5 0.4 0.506 0.477 0.469 0.447 0.535 0.509 0.433 0.432 5.0 0.3 0.391 0.369 0.359 0.342 0.417 0.396 0.329 0.328 b .5161 .5161 .535 .535 .500 .500 .550 .550 A 5.16 4.87 5.21 4.96 5.08 4.83 5.14 5.13 E 0.2035 0.2052 0.2087 0.2064 0.2044 0.2062 0.2082 0.2080 E = 0.2069 target Table 5. (Continued). X Y Y 0.0 5.0 5.000 4.840 4.920 4.960 4.860 5.050 4.830 4.910 0.5 4.0 3.914 3.788 3.861 3.796 3.813 3.943 3.771 3.855 1.0 2.8 3.063 2.965 3.029 2.905 2.992 3.078 2.944 3.026 1.5 2.2 2.398 2.321 2.377 2.223 2.348 2.403 2.299 2.376 2.0 2.0 1.877 1.817 1.865 1.701 1.842 1.876 1.795 1.865 2.5 1.5 1.469 1.422 1.463 1.302 1.446 1.465 1.401 1.464 3.0 1.0 1.150 1.113 1.148 0.996 1.134 1.144 1.094 1.149 3.5 0.9 0.900 0.871 0.901 0.763 0.890 0.893 0.854 0.902 4.0 0.6 0.704 0.682 0.707 0.584 0.698 0.697 0.667 0.708 4.5 0.4 0.551 0.534 0.555 0.447 0.548 0.544 0.521 0.556 5.0 0.3 0.431 0.418 0.435 0.342 0.430 0.425 0.407 0.437 b 0.490 0.490 0.485 0.485 0.485 0.495 0.495 0.484 A 5.00 4.84 4.92 4.96 4.86 5.05 4.83 4.91 E 0.2051 0.2042 0.2047 0.2064 0.2077 0.2074 0.2052 0.2077 E = 0.2069 target 22

PAGE 26

Allen R. Overman Nonlinear Models Table 5. (Continued). X Y Y 0.0 5.0 4.840 5.130 4.900 5.190 4.990 5.210 5.040 5.190 0.5 4.0 3.754 3.979 3.769 3.992 3.809 3.977 3.838 3.952 1.0 2.8 2.912 3.087 2.899 3.070 2.908 3.036 2.922 3.009 1.5 2.2 2.259 2.394 2.229 2.361 2.220 2.318 2.225 2.292 2.0 2.0 1.752 1.857 1.715 1.816 1.695 1.769 1.695 1.745 2.5 1.5 1.359 1.441 1.319 1.397 1.294 1.351 1.290 1.329 3.0 1.0 1.054 1.118 1.014 1.074 0.988 1.031 0.983 1.012 3.5 0.9 0.818 0.867 0.780 0.826 0.754 0.787 0.748 0.770 4.0 0.6 0.634 0.672 0.600 0.636 0.575 0.601 0.570 0.587 4.5 0.4 0.492 0.522 0.462 0.489 0.439 0.459 0.434 0.447 5.0 0.3 0.382 0.405 0.355 0.376 0.335 0.350 0.330 0.340 b 0.508 0.508 0.525 0.525 0.540 0.540 0.545 0.545 A 4.84 5.13 4.90 5.19 4.99 5.21 5.04 5.19 E 0.2095 0.2074 0.2095 0.2054 0.2089 0.2095 0.2068 0.2058 E = 0.2069 target Table 5. (Continued). X Y Y 0.0 5.0 4.930 5.200 5.080 5.170 5.016 5.016 5.016 0.5 4.0 3.782 3.989 3.862 3.931 3.924 3.823 3.922 1.0 2.8 2.902 3.061 2.937 2.989 3.070 2.914 3.067 1.5 2.2 2.226 2.348 2.233 2.272 2.402 2.221 2.398 2.0 2.0 1.708 1.802 1.698 1.728 1.879 1.693 1.875 2.5 1.5 1.310 1.382 1.291 1.314 1.470 1.291 1.466 3.0 1.0 1.005 1.060 0.981 0.999 1.150 0.984 1.146 3.5 0.9 0.771 0.814 0.746 0.759 0.900 0.750 0.896 4.0 0.6 0.592 0.624 0.567 0.577 0.704 0.572 0.701 4.5 0.4 0.454 0.479 0.431 0.439 0.551 0.436 0.548 5.0 0.3 0.348 0.367 0.328 0.334 0.431 0.332 0.429 b 0.530 0.530 0.548 0.548 0.491 0.543 0.492 A 4.93 5.20 5.08 5.17 5.016 5.016 5.016 E 0.2068 0.2056 0.2071 0.2062 0.2086 0.2220 0.2037 E = 0.2069 target 23

PAGE 27

Allen R. Overman Nonlinear Models Table 6. Combinations of A and b to satisfy equal probability equation for 95% probability. X Y Y 0.0 5.0 5.260 4.780 5.310 4.860 5.170 4.730 5.320 4.980 0.5 4.0 4.064 3.693 4.064 3.719 4.026 3.684 4.031 3.773 1.0 2.8 3.139 2.853 3.110 2.846 3.136 2.869 3.054 2.859 1.5 2.2 2.425 2.204 2.380 2.178 2.442 2.234 2.314 2.166 2.0 2.0 1.874 1.703 1.821 1.667 1.902 1.740 1.753 1.641 2.5 1.5 1.448 1.315 1.394 1.276 1.481 1.355 1.328 1.243 3.0 1.0 1.118 1.016 1.067 0.976 1.154 1.055 1.006 0.942 3.5 0.9 0.864 0.785 0.816 0.747 0.898 0.822 0.763 0.714 4.0 0.6 0.667 0.607 0.625 0.572 0.700 0.640 0.578 0.541 4.5 0.4 0.516 0.469 0.478 0.438 0.545 0.499 0.438 0.410 5.0 0.3 0.398 0.362 0.366 0.335 0.424 0.388 0.332 0.310 b .5161 .5161 .535 .535 .500 .500 .555 .555 A 5.26 4.78 5.31 4.86 5.17 4.73 5.31 4.98 E 0.2986 0.2900 0.2946 0.2897 0.2949 0.2956 0.2932 0.2931 E = 0.2961 target Table 6. (Continued). X Y Y 0.0 5.0 5.120 4.710 5.050 4.710 4.900 5.070 5.290 5.320 4.910 0.5 4.0 4.007 3.687 3.972 3.705 3.874 3.822 3.988 4.051 3.739 1.0 2.8 3.137 2.885 3.125 2.914 3.062 2.882 3.007 3.085 2.847 1.5 2.2 2.455 2.258 2.458 2.293 2.421 2.172 2.267 2.349 2.168 2.0 2.0 1.922 1.768 1.934 1.803 1.914 1.638 1.709 1.789 1.651 2.5 1.5 1.504 1.384 1.521 1.419 1.513 1.235 1.288 1.362 1.257 3.0 1.0 1.177 1.083 1.196 1.116 1.196 0.931 0.971 1.037 0.957 3.5 0.9 0.921 0.848 0.941 0.878 0.946 0.702 0.732 0.790 0.729 4.0 0.6 0.721 0.663 0.740 0.691 0.748 0.529 0.552 0.601 0.555 4.5 0.4 0.564 0.519 0.582 0.543 0.591 0.399 0.416 0.458 0.423 5.0 0.3 0.442 0.406 0.458 0.427 0.467 0.301 0.314 0.349 0.322 b 0.490 0.490 0.480 0.480 0.470 0.565 0.565 0.545 0.545 A 5.12 4.71 5.05 4.71 4.90 5.07 5.29 5.32 4.91 E 0.2926 0.2989 0.2981 0.2987 0.2921 0.2944 0.2930 0.2912 0.2944 E = 0.2961 target 24

PAGE 28

Allen R. Overman Nonlinear Models Table 6. (Continued). X Y Y 0.0 5.0 4.720 5.000 4.740 5.140 5.250 4.860 4.770 5.200 5.200 0.5 4.0 3.722 3.943 3.747 3.865 3.948 3.846 3.775 3.907 3.908 1.0 2.8 2.935 3.109 2.963 2.907 2.969 3.044 2.987 2.935 2.936 1.5 2.2 2.315 2.452 2.342 2.186 2.233 2.409 2.364 2.205 2.207 2.0 2.0 1.825 1.934 1.852 1.644 1.679 1.906 1.871 1.656 1.658 2.5 1.5 1.440 1.525 1.464 1.236 1.263 1.508 1.480 1.244 1.246 3.0 1.0 1.135 1.203 1.157 0.930 0.950 1.194 1.172 0.935 0.936 3.5 0.9 0.895 0.948 0.915 0.699 0.714 0.945 0.927 0.702 0.704 4.0 0.6 0.706 0.748 0.723 0.526 0.537 0.748 0.734 0.528 0.529 4.5 0.4 0.557 0.590 0.572 0.395 0.404 0.592 0.581 0.396 0.397 5.0 0.3 0.439 0.465 0.452 0.297 0.304 0.468 0.459 0.298 0.299 b .475 .475 .470 .570 .570 .468 .468 .572 .5715 A 4.72 5.00 4.74 5.14 5.25 4.86 4.77 5.20 5.20 E 0.2948 0.2960 0.2942 0.2967 0.2952 0.2821 0.2887 0.2994 0.2960 E = 0.2961 target Table 6. (Continued). X Y Y 0.0 5.0 5.190 5.180 5.170 4.800 4.810 4.910 5.320 4.810 5.290 0.5 4.0 3.909 3.901 3.894 3.799 3.806 3.739 4.051 3.700 4.069 1.0 2.8 2.944 2.938 2.933 3.006 3.012 2.847 3.085 2.845 3.129 1.5 2.2 2.217 2.213 2.209 2.379 2.384 2.168 2.349 2.188 2.407 2.0 2.0 1.670 1.667 1.663 1.883 1.886 1.651 1.789 1.683 1.851 2.5 1.5 1.258 1.255 1.253 1.490 1.493 1.257 1.362 1.295 1.424 3.0 1.0 0.947 0.945 0.944 1.179 1.181 0.957 1.037 0.996 1.095 3.5 0.9 0.713 0.712 0.711 0.933 0.935 0.729 0.790 0.766 0.842 4.0 0.6 0.537 0.536 0.535 0.738 0.740 0.555 0.601 0.589 0.648 4.5 0.4 0.405 0.404 0.403 0.584 0.585 0.423 0.458 0.453 0.498 5.0 0.3 0.305 0.304 0.304 0.462 0.463 0.322 0.349 0.348 0.383 b 0.567 0.567 0.567 0.468 0.468 0.545 0.545 0.525 0.525 A 5.19 5.18 5.17 4.80 4.81 4.91 5.32 4.81 5.29 E 0.2813 0.2748 0.2756 0.2809 0.2800 0.2944 0.2912 0.2940 0.2991 E = 0.2961 target 25

PAGE 29

Allen R. Overman Nonlinear Models Table 6. (Continued). X Y Y 0.0 5.0 4.820 4.810 4.810 4.810 4.820 4.800 4.790 4.750 5.220 0.5 4.0 3.815 3.807 3.807 3.808 3.816 3.800 3.793 3.685 4.049 1.0 2.8 3.020 3.014 3.013 3.015 3.022 3.009 3.003 2.858 3.141 1.5 2.2 2.390 2.386 2.385 2.387 2.392 2.382 2.377 2.217 2.436 2.0 2.0 1.892 1.888 1.888 1.890 1.894 1.886 1.882 1.720 1.890 2.5 1.5 1.498 1.495 1.494 1.497 1.500 1.493 1.490 1.334 1.466 3.0 1.0 1.186 1.183 1.182 1.185 1.187 1.182 1.180 1.035 1.137 3.5 0.9 0.938 0.937 0.936 0.938 0.940 0.936 0.934 0.803 0.882 4.0 0.6 0.743 0.741 0.741 0.743 0.744 0.741 0.740 0.623 0.684 4.5 0.4 0.588 0.587 0.586 0.588 0.589 0.587 0.586 0.483 0.531 5.0 0.3 0.465 0.464 0.464 0.466 0.467 0.465 0.464 0.375 0.412 b .4675 .4675 .4677 .4670 .4670 .4670 .4670 .508 .508 A 4.82 4.81 4.81 4.81 4.82 4.80 4.79 4.75 5.22 E 0.2818 0.2829 0.2813 0.2853 0.2846 0.2864 0.2882 0.2932 0.2918 E = 0.2961 target Table 7. Combinations of A and b to satisfy equal probability equation for 99% probability. X Y Y 0.0 5.0 5.350 4.680 5.280 4.640 5.160 4.600 5.420 4.780 0.5 4.0 4.133 3.616 4.112 3.614 4.059 3.618 4.138 3.649 1.0 2.8 3.193 2.793 3.202 2.814 3.193 2.846 3.158 2.786 1.5 2.2 2.467 2.158 2.494 2.192 2.512 2.239 2.411 2.126 2.0 2.0 1.906 1.667 1.942 1.707 1.976 1.761 1.841 1.623 2.5 1.5 1.472 1.288 1.513 1.329 1.554 1.385 1.406 1.239 3.0 1.0 1.137 0.995 1.178 1.035 1.223 1.090 1.073 0.946 3.5 0.9 0.879 0.769 0.806 0.763 0.962 0.857 0.819 0.722 4.0 0.6 0.679 0.594 0.715 0.628 0.756 0.674 0.625 0.551 4.5 0.4 0.524 0.459 0.557 0.489 0.595 0.530 0.477 0.421 5.0 0.3 0.405 0.354 0.433 0.381 0.468 0.417 0.364 0.321 b .5161 .5161 .500 .500 .480 .480 .540 .540 A 5.35 4.68 5.28 4.64 5.16 4.60 5.42 4.78 E 0.4274 0.4311 0.4301 0.4193 0.4285 0.4259 0.4248 0.4254 E = 0.4232 target 26

PAGE 30

Allen R. Overman Nonlinear Models Table 7. (Continued). X Y Y 0.0 5.0 5.440 4.890 4.960 5.440 5.040 5.420 5.360 5.160 5.220 0.5 4.0 4.111 3.696 3.730 4.091 3.771 4.056 3.991 3.842 4.086 1.0 2.8 3.107 2.793 2.805 3.076 2.822 3.035 2.971 2.860 3.198 1.5 2.2 2.349 2.111 2.109 2.314 2.112 2.271 2.212 2.130 2.503 2.0 2.0 1.775 1.596 1.586 1.740 1.580 1.699 1.647 1.586 1.959 2.5 1.5 1.341 1.206 1.193 1.308 1.182 1.271 1.226 1.180 1.533 3.0 1.0 1.014 0.911 0.897 0.984 0.885 0.951 0.913 0.879 1.200 3.5 0.9 0.766 0.689 0.675 0.740 0.662 0.712 0.680 0.654 0.939 4.0 0.6 0.579 0.521 0.507 0.556 0.495 0.533 0.506 0.487 0.735 4.5 0.4 0.438 0.393 0.382 0.418 0.371 0.399 0.377 0.363 0.576 5.0 0.3 0.331 0.297 0.287 0.315 0.277 0.298 0.281 0.270 0.450 b 0.560 0.560 0.570 0.570 0.580 0.580 0.590 0.590 0.490 A 5.44 4.89 4.96 5.44 5.04 5.42 5.36 5.16 5.22 E 0.4193 0.4209 0.4188 0.4239 0.4221 0.4251 0.4241 0.4231 0.4220 E = 0.4232 target Table 7. (Continued). X Y Y 0.0 5.0 5.280 4.610 4.990 4.800 4.600 5.080 4.620 4.630 4.930 0.5 4.0 3.925 3.663 3.965 3.837 3.637 4.016 3.616 3.696 3.927 1.0 2.8 2.918 2.910 3.150 3.067 2.875 3.175 2.830 2.944 3.128 1.5 2.2 2.169 2.312 2.503 2.451 2.273 2.510 2.215 2.345 2.491 2.0 2.0 1.613 1.837 1.989 1.959 1.797 1.984 1.734 1.868 1.984 2.5 1.5 1.199 1.460 1.580 1.566 1.421 1.569 1.357 1.488 1.581 3.0 1.0 0.891 1.160 1.255 1.252 1.123 1.240 1.062 1.185 1.259 3.5 0.9 0.663 0.921 0.997 1.001 0.888 0.980 0.831 0.944 1.003 4.0 0.6 0.493 0.732 0.792 0.800 0.702 0.775 0.651 0.752 0.799 4.5 0.4 0.366 0.582 0.630 0.639 0.555 0.613 0.509 0.599 0.636 5.0 0.3 0.272 0.462 0.500 0.511 0.439 0.484 0.399 0.477 0.507 b 0.593 0.460 0.460 0.448 0.470 0.470 0.490 0.455 0.455 A 5.28 4.61 4.99 4.80 4.60 5.08 4.62 4.63 4.93 E 0.4207 0.4213 0.4264 0.4222 0.4192 0.4222 0.4171 0.4188 0.4252 E = 0.4232 target 27

PAGE 31

Allen R. Overman Nonlinear Models Table 7. (Continued). X Y Y 0.0 5.0 4.730 5.390 4.830 5.440 5.430 4.660 5.310 4.720 5.380 0.5 4.0 3.629 4.135 3.669 4.132 4.124 3.615 4.119 3.630 4.138 1.0 2.8 2.784 3.173 2.787 3.139 3.133 2.804 3.195 2.792 3.183 1.5 2.2 2.136 2.434 2.117 2.384 2.380 2.175 2.478 2.148 2.448 2.0 2.0 1.639 1.867 1.608 1.811 1.807 1.687 1.922 1.652 1.883 2.5 1.5 1.257 1.433 1.221 1.375 1.373 1.309 1.491 1.270 1.448 3.0 1.0 0.965 1.099 0.928 1.045 1.043 1.015 1.157 0.977 1.114 3.5 0.9 0.740 0.843 0.705 0.794 0.792 0.787 0.897 0.751 0.857 4.0 0.6 0.568 0.647 0.535 0.603 0.602 0.611 0.696 0.578 0.659 4.5 0.4 0.436 0.496 0.407 0.458 0.457 0.474 0.540 0.445 0.507 5.0 0.3 0.334 0.381 0.309 0.348 0.347 0.368 0.419 0.342 0.390 b 0.530 0.530 0.550 0.550 0.550 0.508 0.508 0.525 0.525 A 4.73 5.39 4.83 5.44 5.43 4.66 5.31 4.72 5.38 E 0.4346 0.4174 0.4246 0.4301 0.4159 0.4221 0.4174 0.4191 0.4259 E = 0.4232 target Table 7. (Continued). X Y Y 0.0 5.0 4.700 4.680 4.670 4.660 4.860 4.720 4.700 4.750 0.5 4.0 3.753 3.737 3.729 3.721 3.881 3.773 3.757 3.799 1.0 2.8 2.997 2.984 2.978 2.971 3.099 3.016 3.003 3.038 1.5 2.2 2.393 2.383 2.378 2.373 2.474 2.410 2.400 2.429 2.0 2.0 1.911 1.903 1.899 1.895 1.976 1.927 1.919 1.943 2.5 1.5 1.526 1.519 1.516 1.513 1.578 1.540 1.534 1.554 3.0 1.0 1.218 1.213 1.211 1.208 1.260 1.231 1.226 1.243 3.5 0.9 0.973 0.969 0.967 0.965 1.006 0.984 0.980 0.994 4.0 0.6 0.777 0.774 0.772 0.770 0.803 0.786 0.783 0.795 4.5 0.4 0.620 0.618 0.616 0.615 0.641 0.629 0.626 0.635 5.0 0.3 0.495 0.493 0.492 0.491 0.512 0.502 0.500 0.508 b 0.450 0.450 0.450 0.450 0.450 0.448 0.448 0.447 A 4.70 4.68 4.67 4.66 4.86 4.72 4.70 4.75 E 0.4063 0.4139 0.4183 0.4229 0.4280 0.4159 0.4200 0.4226 E = 0.4232 target 28

PAGE 32

Allen R. Overman Nonlinear Models Table 8. Correlation of E with b using a parabolic model with 016.5A b E E E E E .577 0.4232 0.4169 .560 0.2961 0.2892 0.2992 .543 0.2069 0.2051 0.2039 0.2075 .529 0.1657 0.1689 0.1616 0.1642 0.1656 .5161 0.1521 0.1612 0.1517 0.1523 0.1520 .503 0.1664 0.1796 0.1701 0.1677 0.1663 .491 0.2069 0.2190 0.2121 0.2061 .477 0.2961 0.2925 0.2915 .462 0.4232 0.4041 29

PAGE 33

Allen R. Overman Nonlinear Models List of Figures 1. Dependence of response variable (Y) on control variable (X). Data from Table 1. Curve drawn from Eq. (35). 2. Dependence of ln Y on X. Data from Table 1. Li ne drawn from Eq. (33). 3. Scatter plot for estimated response variable ( Y ) vs. measured response variable (Y). Line represents the 45% diagonal. 4. Equal probability contours between parameters A and b. Contours drawn from Table 5 (75%), Table 6 (95%), and Table 7 (99%) pr obability levels, respectively. Optimum and standard error values of 075.0016.5 A and 0129.05161.0 b are also shown. 5. Dependence of error sum of squares (E) on exponential parameter (b) for linear parameter A = 5.016. Parabola drawn from Eq. (72). 30

PAGE 34

Allen R. Overman Nonlinear Models Figure 1. Dependence of response variable (Y) on control variable (X). Data from Table 1. Curve drawn from Eq. (35). 31

PAGE 35

Allen R. Overman Nonlinear Models Figure 2. Dependence of ln Y on X. Data from Table 1. Line drawn from Eq. (33). 32

PAGE 36

Allen R. Overman Nonlinear Models Figure 3. Scatter plot for estimated response variable ( Y ) vs. measured response variable (Y). Line represents 45% diagonal. 33

PAGE 37

Allen R. Overman Nonlinear Models Figure 4. Equal probability c ontours between parameters A and b. Contours drawn from Table 5 (75%), Table 6 (95%), and Table 7 (99%) pr obability levels, respectively. Optimum and standard error values of 075.0016.5 A and 0129.05161.0 b are also shown. 34

PAGE 38

Allen R. Overman Nonlinear Models Figure 5. Dependence of error sum of squares (E) on the exponential parameter (b) for linear parameter A = 5.016. Parabola drawn from Eq. (72). 35

PAGE 39

Allen R. Overman Nonlinear Models ABE 6933 Special Topics Mathem atical and Statistical Characteristics of Nonlinear Regression Models A. R. Overman I. Elements of Probability and Calculus A. Arithmetic the process of counting B. Natural numbers positive integers ( ,2,1, 0) C. Rational numbers ratio of two integers ( ,3/2,3/1,2/1,,1/2,1/ 1) D. Irrational numbers (such as e, 2, etc.) E. Complex numbers z = x + i y with i = 1 F. Binomial theorem and Pascals triangle (a + b)0 = 1 (a + b)1 = a1 + b1 (a + b)2 = a2 + 2ab + b2 (a + b)3 = a3 + 3a2b + 3ab2 + b3 (a + b)4 = a4 + 4a3b + 6a2b2 + 4ab3 + b4 (a + b)5 = a5 + 5a4b + 10a3b2 + 10a2b3 + 5ab4 + b5 (a + b)6 = a6 + 6a5b + 15a4b2 + 20a3b3 + 15a2b4 + 6ab5 + b6 (a + b)7 = a7 + 7a6b + 21a5b2 + 35a4b3 + 35a3b4 + 21a2b5 + 7ab6 + b7 (a + b)8 = a8 + 8a7b + 28a6b2 + 56a5b3 + 70a4b4 + 56a3b5 + 28a2b6 + 8ab7 + b8 (a + b)9 = a9 + 9a8b + 36a7b2 + 84a6b3 + 126a5b4 + 126a4b5 + 84a3b6 + 36a2b7 + 9ab8 + b9 Note symmetry in the distribution of coefficients for each expansion. 36

PAGE 40

Allen R. Overman Nonlinear Models Pascals triangle for binomial coefficients 1 1 2 1 1 3 3 1 1 4 6 4 1 1 5 10 10 5 1 1 6 15 20 15 6 1 1 7 21 35 35 21 7 1 1 8 28 56 70 56 28 8 1 1 9 36 84 126 126 84 36 9 1 1 10 45 120 210 252 210 120 45 10 1 1 11 55 165 330 462 462 330 165 55 11 1 1 12 66 220 495 792 924 792 495 220 66 12 1 Note the pattern in the co efficients, including symmetry. G. Frequency distributions 1. Discrete distribution Consider the problem of a peg board. This is a two state system a cell (hole) is either filled or empty. Each cell holds one and only one object (peg), which can be viewed as a type of exclusion principle. Define n as the total number of cells and x as the number of filled cells (pegs). Cells (holes) are indistinguishable (all alik e), as are the objects (pegs). Order of filling the cells is irrelevant. Note that a peg board can be linear, triangular, rectangular (Eigen and Winkler, 1993, p. 40; Polster, 2004, p. 33), or even 3-dimensional. The number of distinguishable combinations which are possible for each x, xnc,, can be calculated from (Ruhla, 1992, p. 18; Watkins, 2000, p. 22) )!(! ,xnx n xnc (1) 37

PAGE 41

Allen R. Overman Nonlinear Models where and is called n factorial. Note that n can assume positive integers n n 321! ( ) and x can also assume positive integers (,3,2,1 n n x,,2,1,0 ). For small values of n it is easy to estimate c by intuition, but for larger n calculations of c are best performed on a pocket calculator or computer with the algorithm for co mputations (Eq. (1)) buil t in. The total number of combinations C for the system is defined as the sum of c values for all values of x, and can be calculated from C = 2n. The frequency distribution of c values is then calculated from f = c/C. Cumulative frequency is calculated from the cumulative sum fF (2) so that F is normalized It should be noted that F forms a discrete set of numbers for a particular case. 10 F 2. Continuous distribution The next step is to compare the discrete di stribution to a continuous Gaussian distribution where x is considered a continuous variable and the cumulative distribution is described by 2 erf1 2 1x F (3) where and are the mean and spread of the distri bution. The error function is defined by 2 0 2)exp( 2 2 erfxduu x (4) where represents the Gaussian distribution (bell-shaped curve). Values of the erf can be obtained from mathematical tables (cf. Abramowitz and Stegun, 1965, chp. 7). Some properties of the error function should be noted: ) exp(2u erf (0) = 0, erf ( ) = 1, erf (x) = erf (+x), erf ( ) = 1 Equation (3) can be rear ranged to the linear form x F Z 2 1 2 12erf1 (5) where erf is the inverse error function. For example, Linear regression of Z vs. x leads to values of the parameters 00 .1)8427.0(erf1 and With these parameters now known the frequency distribution for f vs. x can be calculated for the continuous distribution from 38

PAGE 42

Allen R. Overman Nonlinear Models 22 exp 2 1 x f (6) The procedure can now be applied to a linear peg board, triangular peg board, and square peg board. It can even be applied to a 3dimensional system. This analysis falls within a branch of mathematics known as group theory. Values of the error function can be calculat ed from the series approximation (Abramowitz and Stegun, 1965, p. 299) 44 3 2]078108.0 000972.0 230389.0278393.01[ 1 1erfx x x x x (7) for For the case where erf x is given, the inverse erf and therefore x can be obtained on a scientific calculator or computer using the solver routine. Note that for the case F < 0.5 and 2F < 0 (negative) the procedure is to change the value from to +, solve for the inverse by Eq. (7) and change the sign from +x to x. Equation (7) does not work directly for x because the power series in Eq. (7) is not symmetric. 8.10 x H. Symmetry and c onservation principle In all of the discrete and con tinuous Gaussian distributions we note symmetry in the distributions around a mean point. A mathematical consequence of this property is that something is conserved (remains constant) in the system. Note that the number of filled cells is defined by x. Since this is a two-state (binary) system (cells are either empty of filled), it follows that the number of unfilled cells is n x. The total capacity of the system is the sum of filled and unfilled cells so that total capacity is = x + n x = n. While this is obvious for our case, it illustrates the connection between symmetry and conservation. This property turns out to be very important in the various models of physics (including mechan ics, electromagnetism, relativity, and quantum mechanics). It also shows up in chemistry and biology. 39

PAGE 43

Allen R. Overman Nonlinear Models Figure 1. Linear pegboard 40

PAGE 44

Allen R. Overman Nonlinear Models Figure 2. Triangular pegboard 41

PAGE 45

Allen R. Overman Nonlinear Models Figure 3. Square pegboard 42

PAGE 46

Allen R. Overman Nonlinear Models Frequency distributions for a linear peg board Case 1 x 2 In this case n = 1 x 2 = 2 and x = 0, 1, 2. Table A1. Frequency distribution for the linear peg board with a 1 x 2 array. x c f F Z c f 0.0000 0 1 0.2500 0.2167 0.867 0.2500 .4767 1 2 0.5000 0.5379 2.152 0.7500 +0.4767 2 1 0.2500 0.2167 0.867 1.0000 C 4 = 22 Note symmetry in the frequency distribution. x x F Z9534.09534.0 2 1 2 12erf 1 r = 1 000.1,0489.12 2 20489.1 00.1 exp5379.0 2 exp 2 1 x x f f f2848.11045.0 r = 1 fc 4 43

PAGE 47

Allen R. Overman Nonlinear Models Case 1 x 3 In this case n = 1 x 3 = 3 and x = 0, 1, 2, 3. Table A2. Frequency distribu tion for the linear peg boa rd with a 1 x 3 array. x c f F Z f c 0.0000 0 1 0.1250 0.1037 0.83 0.1250 .8142 1 3 0.3750 0.3888 3.11 0.5000 0.0000 2 3 0.3750 0.3888 3.11 0.8750 +0.8142 3 1 0.1250 0.1037 0.83 1.0000 C 8 = 23 Note symmetry in the frequency distribution. x x F Z 8142.02213.1 2 1 2 12erf 1 r = 1.0000 500.1,2282.12 2 22282.1 50.1 exp4594.0 2 exp 2 1 x x f f f 1404.103885.0 r = 1.0000 fc 8 44

PAGE 48

Allen R. Overman Nonlinear Models Case 1 x 4 In this case n = 1 x 4 = 4 and x = 0, 1, 2, 3, 4. Table A3. Frequency distribu tion for the linear peg boa rd with a 1 x 4 array. x c f F Z f c 0.0000 0 1 0.0625 0.0512 0.82 0.0625 .0842 1 4 0.2500 0.2419 3.87 0.3125 .3452 2 6 0.3750 0.4060 6.50 0.6875 +0.3452 3 4 0.2500 0.2419 3.87 0.9375 +1.0842 4 1 0.0625 0.0512 0.82 1.0000 C 16 = 24 Note symmetry in the frequency distribution. x x F Z 7196.04391.1 2 1 2 12erf 1 r = 0.999909 000.2,3897.12 2 23897.1 00.2 exp4060.0 2 exp 2 1 x x f f f1052.10226.0 r = 0.99710 fc 16 45

PAGE 49

Allen R. Overman Nonlinear Models Case 1 x 5 In this case n = 1 x 5 = 5 and x = 0, 1, 2, 3, 4, 5. Table A4. Frequency distribu tion for the linear peg boa rd with a 1 x 5 array. x c f F Z f c 0.00000 0 1 0.03125 0.02590 0.83 0.03125 .3148 1 5 0.15625 0.14143 4.53 0.18750 .6277 2 10 0.31250 0.33050 10.58 0.50000 0.0000 3 10 0.31250 0.33050 10.58 0.81250 +0.6277 4 5 0.15625 0.14143 4.53 0.96875 +1.3148 5 1 0.03125 0.02590 0.83 1.00000 C 32 = 25 Note symmetry in the frequency distribution. x x F Z 65146.06286.1 2 1 2 12erf 1 r = 0.99983 500.2,5350.12 2 25350.1 50.2 exp3675.0 2 exp 2 1 x x f f f0882.101543.0 r = 0.99723 fc 32 46

PAGE 50

Allen R. Overman Nonlinear Models Case 1 x 6 In this case n = 1 x 6 = 6 and x = 0, 1, 2, 3, 4, 5, 6. Table A5. Frequency distribu tion for the linear peg boa rd with a 1 x 6 array. x c f F Z f c 0.000000 0 1 0.015625 0.013247 0.85 0.015625 .5213 1 6 0.093750 0.080181 5.13 0.109375 .8703 2 15 0.234375 0.236181 15.11 0.343750 .2841 3 20 0.312500 0.338560 21.67 0.656250 +0.2841 4 15 0.234375 0.236181 15.11 0.890625 +0.8703 5 6 0.093750 0.080181 5.13 0.984375 +1.5213 6 1 0.015625 0.013247 0.85 1.000000 C 64 = 26 Note symmetry in the frequency distribution. x x F Z 60009.08003.1 2 1 2 12erf 1 r = 0.99975 000.3,66643.12 2 266643.1 00.3 exp33856.0 2 exp 2 1 x x f f f0809.101187.0 r = 0.99730 fc 64 47

PAGE 51

Allen R. Overman Nonlinear Models Case 1 x 7 In this case n = 1 x 7 = 7 and x = 0, 1, 2, 3, 4, 5, 6, 7. Table A6. Frequency distribu tion for the linear peg boa rd with a 1 x 7 array. x c f F Z f 0.0000000 0 1 0.0078125 0.008580 0.0078125 .7123 1 7 0.0546875 0.049286 0.0625000 .0842 2 21 0.1640625 0.158080 0.2265625 .5305 3 35 0.2734375 0.283108 0.5000000 0.0000 4 35 0.2734375 0.283108 0.7734375 +0.5305 5 21 0.1640625 0.158080 0.9375000 +1.0842 6 7 0.0546875 0.049286 0.9921875 +1.7123 7 1 0.0078125 0.008580 1.0000000 C 128 = 27 Note symmetry in the frequency distribution. x x F Z 53978.088923.1 2 1 2 12erf 1 r = 0.999963 500.3,8526.12 2 28526.1 50.3 exp3045.0 2 exp 2 1 x x f f f0689.1011875.0 r = 0.999198 48

PAGE 52

Allen R. Overman Nonlinear Models Case 1 x 8 In this case n = 1 x 8 = 8 and x = 0, 1, 2, 3, 4, 5, 6, 7, 8. Table A7. Frequency distribu tion for the linear peg boa rd with a 1 x 8 array. x c f F Z f 0.000000 0 1 0.003906 0.004626 0.5 0.003906 .8934 0.012167 1 8 0.031250 0.028129 1.5 0.035156 .2777 0.057167 2 28 0.109375 0.102126 2.5 0.144531 .7504 0.160373 3 56 0.218750 0.221375 3.5 0.363281 .2470 0.268612 4 70 0.273438 0.286500 4.5 0.636719 +0.2470 0.268612 5 56 0.218750 0.221375 5.5 0.855469 +0.7504 0.160373 6 28 0.109375 0.102126 6.5 0.964844 +1.2777 0.057167 7 8 0.031250 0.028129 7.5 0.996094 +1.8934 0.012167 8 1 0.003906 0.004626 1.000000 C 256 = 28 Note symmetry in the frequency distribution. x x F Z 5078.00312.2 2 1 2 12erf 1 r = 0.999946 0000.4,9692.12 2 29692.1 00.4 exp2865.0 2 exp 2 1 x x f f f0589.1008700.0 r = 0.999084 49

PAGE 53

Allen R. Overman Nonlinear Models Frequency distributions of a triangular peg board Case 1: 1 Cell ( n = 1) For this case n = 1 and x = 0 or 1. The distributio n is given in Table A8. Table A8. Frequency distribution for a triangular peg boa rd with 1 cell. x c f F 0.000 0 1 0.500 0.500 1 1 0.500 1.000 C 2 = 21 Case 2: 3 Cells (n = 3) For this case n = 3 and x = 0, 1, 2, or 3. The pegboard is shown in the diagram. The corresponding distribution is given in Table A9. Table A9. Frequency distribution for a triangular peg board with 3 cells. x c f F Z f 0.0000 0 1 0.1250 0.1034 0.5 0.1250 .8142 0.2367 1 3 0.3750 0.3892 1.5 0.5000 0.0000 0.4594 2 3 0.3750 0.3892 2.5 0.8750 +0.8142 0.2367 3 1 0.1250 0.1034 1.0000 C 8 = 23 x x F Z 8142.02213.1 2 1 2 )12(erf 1 r = 1.0000 500.1,2282.12 2 22282.1 500.1 exp4594.0 2 exp 2 1 x x f f f1432.103950.0 r = 1.000000 50

PAGE 54

Allen R. Overman Nonlinear Models This system is an example of group theory in mathematics, which links principles of symmetry and conservation. Note symmetry of c around the mean value of x (50 .1 ). Conservation comes from Filled cells + Unfilled cells = x + ( n x ) = n = total capacity of the system (number of cells) Case 3: 6 Cells (n = 6) For this case n = 6 and x = 0, 1, 2, 3, 4, 5, 6. The distribution is given in Table A10. Table A10. Frequency distribution for a triangular peg boa rd with 6 cells. x c f F Z f 0.000000 0 1 0.015625 0.013246 0.015625 .5213 1 6 0.093750 0.080177 0.109375 .8703 2 15 0.234375 0.236178 0.343750 .2841 3 20 0.312500 0.338560 0.656250 +0.2841 4 15 0.234375 0.236178 0.890625 +0.8703 5 6 0.093750 0.080177 0.984375 +1.5213 6 1 0.015625 0.013246 1.000000 C 64 = 26 x x F Z 6001.08003.1 2 1 2 )12(erf 1 r = 0.999748 000.3,6664.12 2 26664.1 00.3 exp33856.0 2 exp 2 1 x x f f f0809.101187.0 r = 0.99730 51

PAGE 55

Allen R. Overman Nonlinear Models Case 4: 10 Cells ( n = 1 0) For this case n = 10 and x = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. The distribution is given in Table A11. Table A11. Frequency distribution for a triangular peg boa rd with 10 cells. x c f F Z f 0.0000000 0 1 0.0009766 0.001342 0.0009766 1 10 0.0097656 0.008921 0.0107422 .6260 2 45 0.0439453 0.038930 0.0546875 .1311 3 120 0.1171875 0.111513 0.1718750 .6700 4 210 0.2050781 0.209679 0.3769531 .2215 5 252 0.2460938 0.258800 0.6230469 +0.2215 6 210 0.2050781 0.209679 0.8281250 +0.6700 7 120 0.1171875 0.111513 0.9453125 +1.1311 8 45 0.0439453 0.038930 0.9892578 +1.6260 9 10 0.0097656 0.008921 0.9990234 10 1 0.0009766 0.001342 1.0000000 C 1024 = 210 x x F Z 4588.02939.2 2 1 2 )12(erf 1 r = 0.99988 000.5,1797.22 2 21797.2 000.5 exp2588.0 2 exp 2 1 x x f f f0383.100352.0 r = 0.99894 52

PAGE 56

Allen R. Overman Nonlinear Models Case 5: 15 Cells ( n = 1 5) For this case n = 15 and x = The distribution is gi ven in Table A12. 15 ,,3,2,1,0 Table A12. Frequency distribution for a triangular peg boa rd with 15 cells. x c f F Z f 0.0000000 0 1 0.0000305 0.000096 0.0000305 1 15 0.0004578 0.000652 0.0004883 2 105 0.0032043 0.003355 0.0036926 3 455 0.0138855 0.013138 0.0175781 .4875 4 1365 0.0416565 0.039158 0.0592346 .1032 5 3003 0.0916443 0.088829 0.1508789 .7309 6 5005 0.1527405 0.153358 0.3036194 .3630 7 6435 0.1963806 0.201504 0.5000000 0.0000 8 6435 0.1963806 0.201504 0.6963806 +0.3630 9 5005 0.1527405 0.153358 0.8491211 +0.7309 10 3003 0.0916443 0.088829 0.9407654 +1.1032 11 1365 0.0416565 0.039158 0.9824219 +1.4875 12 455 0.0138855 0.013138 0.9963074 13 105 0.0032043 0.003355 0.9995117 14 15 0.0004578 0.000652 0.9999695 15 1 0.0000305 0.000096 1.0000000 C 32768 = 215 x x F Z 3695.07711.2 2 1 2 )12(erf 1 r = 0.999972 500.7,7065.22 53

PAGE 57

Allen R. Overman Nonlinear Models 2 27065.2 500.7 exp2085.0 2 exp 2 1 x x f f f 02423.1002046.0 r = 0.999637 54

PAGE 58

Allen R. Overman Nonlinear Models Case 6: 21 Cells ( n = 2 1) For this case n = 21 and x = The distribution is gi ven in Table A13. 21 ,,3,2,1,0 Table A13. Frequency distribution for a triangular peg boa rd with 21 cells. x c f F Z f 0.000000 0 1 0.000000 0.000004 0.000000 1 21 0.000010 0.000028 0.000010 2 210 0.000100 0.000159 0.000110 3 1,330 0.000634 0.000749 0.000744 4 5,985 0.002854 0.002913 0.003598 5 20,349 0.009703 0.009333 0.013301 .5668 6 54,264 0.025875 0.024629 0.039176 .2429 7 116,280 0.055447 0.053525 0.094623 .9287 8 203,490 0.097032 0.095807 0.191655 .6168 9 293,930 0.140157 0.141240 0.331812 .3071 10 352,716 0.168188 0.171489 0.500000 0.0000 11 352,716 0.168188 0.171489 0.668188 +0.3071 12 293,930 0.140157 0.141240 0.808345 +0.6168 13 203,490 0.097032 0.095807 0.905377 +0.9287 14 116,280 0.055447 0.053525 0.960824 +1.2429 15 54,264 0.025875 0.024629 0.986699 +1.5668 16 20,349 0.009703 0.009333 0.996402 17 5,985 0.002854 0.002913 0.999256 18 1,330 0.000634 0.000749 0.999890 55

PAGE 59

Allen R. Overman Nonlinear Models 19 210 0.000100 0.000159 0.999990 20 21 0.000010 0.000028 1.000000 21 1 0.000000 0.000004 1.000000 C 2,097,152 = 221 x x F Z 3115.02707.3 2 1 2 )12(erf 1 r = 0.999982 500.10,2103.32 2 22103.3 500.10 exp1757.0 2 exp 2 1 x x f f f 0183.100135.0 r = 0.99980 56

PAGE 60

Allen R. Overman Nonlinear Models Case 7: 28 Cells ( n = 2 8) For this case n = 28 and x = The distribution is gi ven in Table A14. 28 ,,3,2,1,0 Table A14. Frequency distribution for a triangular peg boa rd with 28 cells. x c f F Z f 0.0000000 0 1 0.0000000 0.000000 0.0000000 1 28 0.0000001 0.000001 0.0000001 2 378 0.0000014 0.000004 0.0000015 3 3,276 0.0000122 0.000022 0.0000137 4 20,475 0.0000763 0.000104 0.0000900 5 98,280 0.0003661 0.000414 0.0004561 6 376,740 0.0014035 0.001431 0.0018596 7 1,184,040 0.0044109 0.004273 0.0062705 .7706 8 3,108,105 0.0115786 0.011029 0.0178491 .4830 9 6,906,900 0.0257302 0.024603 0.0435793 .2079 10 13,123,110 0.0488874 0.047433 0.0924667 .9378 11 21,474,180 0.0799975 0.079034 0.1724642 .6684 12 30,421,755 0.1133299 0.113814 0.2857941 .3996 13 37,442,160 0.1394829 0.141652 0.4252770 .1335 14 40,116,600 0.1494460 0.152370 0.5747230 +0.1335 15 37,442,160 0.1394829 0.141652 0.7142059 +0.3996 16 30,421,755 0.1133299 0.113814 0.8275358 +0.6684 17 21,474,180 0.0799975 0.079034 0.9075333 +0.9378 18 13,123,110 0.0488874 0.047433 0.9564207 +1.2079 57

PAGE 61

Allen R. Overman Nonlinear Models 19 6,906,900 0.0257302 0.024603 0.9821509 +1.4830 20 3,108,105 0.0115786 0.011029 0.9937295 +1.7706 21 1,184,040 0.0044109 0.004273 0.9981404 22 376,740 0.0014035 0.001431 0.9995439 23 98,280 0.0003661 0.000414 0.9999100 24 20,475 0.0000763 0.000104 0.9999863 25 3,276 0.0000122 0.000022 0.9999985 26 378 0.0000014 0.000004 0.9999999 27 28 0.0000001 0.000001 1.0000000 28 1 0.0000000 0.000000 1.0000000 C 268,435,456 = 228 x x F Z 27007.07810.3 2 1 2 )12(erf 1 r = 0.999974 000.14,70275.32 2 270275.3 00.14 exp15237.0 2 exp 2 1 x x f f f 01674.1000995.0 r = 0.99985 58

PAGE 62

Allen R. Overman Nonlinear Models Frequency distributions for a square peg board. Case 1: 2 x 2 For this case n = 2 x 2 = 4 and x assumes values of 0, 1, 2, 3, and 4. Corresponding values of are calculated from Eq. (1). Results are given in Table A15 xnc Table A15. Frequency distribution for the square peg board with a 2 x 2 array. x c f F Z f 0.0000 0 1 0.0625 0.05117 0.0625 .0842 1 4 0.2500 0.24191 0.3125 .3452 2 6 0.3750 0.40600 0.6875 +0.3452 3 4 0.2500 0.24191 0.9375 +1.0842 4 1 0.0625 0.05117 1.0000 C 16 = 24 Note the symmetry in the disc rete frequency distribution. x x F Z 7196.04391.1 2 1 2 12erf 1 r = 0.999909 000.2,3897.12 2 23897.1 00.2 exp4060.0 2 exp 2 1 x x f f f 1053.102262.0 r = 0.9971 59

PAGE 63

Allen R. Overman Nonlinear Models Case 2: 3 x 3 For this case n = 3 x 3 = 9 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Results are giv en in Table A16. 9 ,xnc Table A16. Frequency distribution for the square peg board with a 3 x 3 array. x c f F Z f 0.000000 0 1 0.001953 0.002497 0.001953 1 9 0.017578 0.015920 0.019531 .4568 2 36 0.070312 0.063869 0.089843 .9491 3 84 0.164063 0.161257 0.253906 .4680 4 126 0.246094 0.256230 0.500000 0.0000 5 126 0.246094 0.256230 0.746094 +0.4680 6 84 0.164063 0.161257 0.910157 +0.9491 7 36 0.070312 0.063869 0.980469 +1.4568 8 9 0.017578 0.015920 0.998047 9 1 0.001953 0.002497 1.000000 Total 512 = 29 Again note the symmetry in the discrete frequency distribution. x x F Z 4812.01653.2 2 1 2 12erf 1 r = 0.999919 500.4,0782.22 2 20782.2 50.4 exp2715.0 2 exp 2 1 x x f f f 0369.100373.0 r = 0.99895 60

PAGE 64

Allen R. Overman Nonlinear Models Case 3: 4 x 4 For this case n = 4 x 4 = 16 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A17. 16 ,xnc Table A17. Frequency distribution for the square peg board with a 4 x 4 array. x c f F Z f 0.0000000 0 1 0.0000153 0.000044 0.0000153 1 16 0.0002441 0.000319 0.0002594 2 120 0.0018311 0.001966 0.0020905 3 560 0.0085449 0.008102 0.0106354 .6287 4 1820 0.0277710 0.025807 0.0384064 .2493 5 4368 0.0666504 0.063543 0.1050568 .8868 6 8008 0.1221924 0.120946 0.2272492 .5288 7 11440 0.1745605 0.177953 0.4018097 .1759 8 12870 0.1963806 0.202400 0.5981903 +0.1759 9 11440 0.1745605 0.177953 0.7727508 +0.5288 10 8008 0.1221924 0.120946 0.8949432 +0.8868 11 4368 0.0666504 0.063543 0.9615936 +1.2493 12 1820 0.0277710 0.025807 0.9893646 +1.6287 13 560 0.0085449 0.008102 0.9979095 14 120 0.0018311 0.001966 0.9997406 15 16 0.0002441 0.000369 0.9999847 16 1 0.0000153 0.000053 1.0000000 C 65536 = 216 Note the symmetry in the frequency distribution. 61

PAGE 65

Allen R. Overman Nonlinear Models x x F Z 3588.08703.2 2 1 2 12erf 1 r = 0.999958 000.8,7872.22 2 27872.2 00.8 exp2024.0 2 exp 2 1 x x f f f 0248.100194.0 r = 0.99962 Several characteristics should be not ed from these calculations. Firs t, note the symmetry in the distributions in the tables. S econd, note that the continuous Ga ussian function approximates the discrete distributions rather we ll. Third, as the number of va lues increases, agreement between the discrete and continuous distributions improves. 62

PAGE 66

Allen R. Overman Nonlinear Models Case 3: 5 x 5 For this case n = 5 x 5 = 25 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A18. 25 ,xnc Table A18. Frequency distribution for the square peg board with a 5 x 5 array. x c f F Z f 0.00000000 0 1 0.00000003 0.000000 0.00000003 1 25 0.00000075 0.000003 0.00000078 2 300 0.00000894 0.000020 0.00000972 3 2,300 0.00006855 0.000101 0.00007827 4 12,650 0.00037700 0.000439 0.00045527 5 53,130 0.00158340 0.001624 0.00203867 6 177,100 0.00527799 0.005101 0.00731666 .7297 7 480,700 0.01432598 0.013604 0.02164264 .4266 8 1,081,575 0.03223345 0.030810 0.05387609 .1362 9 2,042,975 0.06088540 0.059254 0.11476149 .8503 10 3,268,760 0.09741664 0.096770 0.21217813 .5650 11 4,457,400 0.13284087 0.134200 0.34501900 .2816 12 5,200,300 0.15498102 0.158037 0.50000002 0.0000 13 5,200,300 0.15498087 0.158037 0.65498089 +0.2816 14 4,457,400 0.13284087 0.134200 0.78782176 +0.5650 15 3,268,760 0.09741664 0.096770 0.88523840 +0.8503 16 2,042,975 0.06088540 0.059254 0.94612380 +1.1362 17 1,081,575 0.03223345 0.030810 0.97835725 +1.4266 18 480,700 0.01432598 0.013604 63

PAGE 67

Allen R. Overman Nonlinear Models 0.99268323 +1.7297 19 177,100 0.00527799 0.005101 0.99796122 20 53,130 0.00158340 0.001624 0.99954462 21 12,650 0.00037700 0.000439 0.99992162 22 2,300 0.00006855 0.000101 0.99999017 23 300 0.00000894 0.000020 0.99999911 24 25 0.00000075 0.000003 0.99999986 25 1 0.00000003 0.000000 0.99999989 C 33,554,432 = 225 Note the symmetry in the frequency distribution. x x F Z 2859.05740.3 2 1 2 12erf 1 r = 0.999974 50.12 4975.32 2 24975.3 50.12 exp1613.0 2 exp 2 1 x x f f f 017789.1001129.0 r = 0.999827 64

PAGE 68

Allen R. Overman Nonlinear Models Case 4: 6 x 6 For this case n = 6 x 6 = 36 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A19. 36 ,xnc Table A19. Frequency distribution for the square peg board with a 6 x 6 array. x c f F Z f 0.000000 0 1 0.000000 1 36 0.00000000 0.000000 0.00000000 2 630 0.00000001 0.000000 0.00000001 3 7,140 0.00000010 0.000000 0.00000010 4 58,905 0.00000086 0.000002 0.00000096 5 376,992 0.00000549 0.000010 0.00000645 6 1,947,792 0.00002834 0.000039 0.00003479 7 8,347,680 0.00012147 0.000144 0.00015626 8 30,260,340 0.00044035 0.000472 0.00059661 9 94,143,280 0.00136997 0.001380 0.00196658 10 254,186,856 0.00369891 0.003606 0.00566549 .7970 11 600,805,296 0.00874287 0.008415 0.01440836 .5442 12 1,251,677,700 0.01821431 0.017541 0.03262267 .3013 13 2,310,789,600 0.03362641 0.032657 0.06624908 .0633 14 3,796,297,200 0.05524340 0.054303 0.12149248 .8263 15 5,567,902,560 0.08102365 0.080647 0.20251613 .5890 16 7,307,872,110 0.10634354 0.106974 0.30885967 .3524 17 8,597,496,600 0.12511004 0.126733 0.43396971 .1179 18 9,075,135,300 0.13206060 0.134100 65

PAGE 69

Allen R. Overman Nonlinear Models 0.56603031 +0.1179 19 8,597,496,600 0.12511004 0.126733 0.69114035 +0.3524 20 7,307,872,110 0.10634354 0.106974 0.79748389 +0.5890 21 5,567,902,560 0.08102365 0.080647 0.87850754 +0.8263 22 3,796,297,200 0.05524340 0.054303 0.93375094 +1.0633 23 2,310,789,600 0.03362641 0.032657 0.96737735 +1.3013 24 1,251,677,700 0.01821431 0.017541 0.98559166 +1.5442 25 600,805,296 0.00874287 0.008415 0.99433453 +1.7970 26 254,186,856 0.00369891 0.003606 0.99803344 27 94,143,280 0.00136997 0.001380 0.99940341 28 30,260,340 0.00044035 0.000472 0.99984376 29 8,347,680 0.00012147 0.000144 0.99996523 30 1,947,792 0.00002834 0.000039 0.99999357 31 376,992 0.00000549 0.000010 0.99999906 32 58,905 0.00000086 0.000002 0.99999992 33 7,140 0.00000010 0.000000 1.00000002 34 630 0.00000001 1.00000003 35 36 0.00000000 1.00000003 36 1 0.00000000 1.00000003 C 68,719,476,736 = 236 Note the symmetry in the frequency distribution. 66

PAGE 70

Allen R. Overman Nonlinear Models x x F Z 2377.02786.4 2 1 2 12erf 1 r = 0.999981 00.18,2070.42 2 22070.4 00.18 exp1341.0 2 exp 2 1 x x f f f 01356.1000723.0 r = 0.999909 67

PAGE 71

Allen R. Overman Nonlinear Models Case 4: 7 x 7 For this case n = 7 x 7 = 49 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A20. 49 ,xnc Table A20. Frequency distribution for the square peg board with a 7 x 7 array. x c f F Z f 0.000000 0 1 0.000000 0.000000 1 49 0.000000 0.000000 0.000000 2 1,176 0.000000 0.000000 0.000000 3 18,424 0.000000 0.000000 0.000000 4 211,876 0.000000 0.000002 0.000000 5 1,906,884 0.000000 0.000000 0.000000 6 13,983,816 0.000000 0.000000 0.000000 7 85,900,584 0.000000 0.000000 0.000000 8 450,978,066 0.000001 0.000002 0.000001 9 2,054,455,634 0.000004 0.000006 0.000005 10 8,217,822,536 0.000015 0.000020 0.000020 11 29,135,916,264 0.000052 0.000064 0.000072 12 92,263,734,836 0.000164 0.000185 0.000236 13 262,596,783,764 0.000466 0.000497 0.000702 14 675,248,872,536 0.001199 0.001228 0.001901 15 1,575,580,703,000 0.002799 0.002795 0.004700 16 3,348,108,993,000 0.005947 0.005859 0.010647 .6285 17 6,499,270,398,000 0.011545 0.011315 0.022192 .4192 18 11,554,258,486,000 0.020524 0.020124 68

PAGE 72

Allen R. Overman Nonlinear Models 0.042716 .2145 19 18,851,684,898,000 0.033487 0.032967 0.076203 .0119 20 28,277,527,346,000 0.050231 0.049740 0.126434 .8093 21 39,049,918,716,000 0.069367 0.069120 0.195801 .6061 22 49,699,896,548,000 0.088285 0.088467 0.284086 .4032 23 58,343,356,817,000 0.103639 0.104288 0.387725 .2016 24 63,205,303,219,000 0.112275 0.113230 0.500000 0.0000 25 63,205,303,219,000 0.112275 0.113230 0.612275 +0.2016 26 58,343,356,817,000 0.103639 0.104288 0.715914 +0.4032 27 49,699,896,548,000 0.088285 0.088467 0.804199 +0.6061 28 39,049,918,716,000 0.069367 0.069120 0.873566 +0.8093 29 28,277,527,346,000 0.050231 0.049740 0.923797 +1.0119 30 18,851,684,898,000 0.033487 0.032967 0.957284 +1.2145 31 11,554,258,486,000 0.020524 0.020124 0.977808 +1.4192 32 6,499,270,398,000 0.011545 0.011315 0.989353 +1.6285 33 3,348,108,993,000 0.005947 0.005859 0.995300 34 1,575,580,703,000 0.002799 0.002795 0.998099 35 675,248,872,536 0.001199 0.001228 0.999298 36 262,596,783,764 0.000466 0.000497 0.999764 37 92,263,734,836 0.000164 0.000185 0.999928 38 29,135,916,264 0.000052 0.000064 0.999980 39 8,217,822,536 0.000015 0.000020 0.999995 40 2,054,455,634 0.000004 0.000006 0.999999 41 450,978,066 0.000001 0.000002 69

PAGE 73

Allen R. Overman Nonlinear Models 1.000000 42 85,900,584 0.000000 0.000000 43 13,983,816 44 1,906,884 45 211,876 46 18,424 47 1,176 48 49 49 1 C 562,949,953,421,000 = 249 Note the symmetry in the frequency distribution. x x F Z 20281.09687.4 2 1 2 12erf 1 r = 0.9999962 50.24,9308.42 2 29308.4 50.24 exp1144.0 2 exp 2 1 x x f f f 00727.1000345.0 r = 0.999965 70

PAGE 74

Allen R. Overman Nonlinear Models Case 5: 8 x 8 For this case n = 8 x 8 = 64 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A21. 64 ,xnc Table A21. Frequency distribution for the square peg board with a 8 x 8 array. x c f F Z f 0.000000 0 1 0.000000 0.000000 1 64 0.000000 0.000000 0.000000 2 3 4 7 8 0.00000000 1018 9 0.00000003 1018 10 0.00000015 1018 11 0.00000074 1018 0.0000000 0.0000000 12 0.00000328 1018 0.0000002 0.000000 0.0000002 13 0.00001314 1018 0.0000007 0.000001 0.0000009 14 0.00004786 1018 0.0000026 0.000004 0.0000035 15 0.00015952 1018 0.0000086 0.000011 0.0000121 16 0.00048853 1018 0.0000265 0.000032 0.0000386 17 0.00137937 1018 0.0000748 0.000084 0.0001134 18 0.00360169 1018 0.0001952 0.000210 0.0003086 19 0.00871988 1018 0.0004727 0.000491 0.0007813 71

PAGE 75

Allen R. Overman Nonlinear Models 20 0.01961973 1018 0.0010636 0.001079 0.0018449 21 0.04110800 1018 0.0022285 0.002224 0.0040734 22 0.08034745 1018 0.0043556 0.004306 0.0084390 .6919 23 0.14672143 1018 0.0079538 0.007828 0.0163828 .5077 24 0.25064910 1018 0.0135877 0.013363 0.0299705 .3279 25 0.40103857 1018 0.0217403 0.021421 0.0517108 .1504 26 0.60155785 1018 0.0326105 0.032244 0.0843213 .9736 27 0.84663698 1018 0.0458963 0.045576 0.1302176 .7965 28 1.11877029 1018 0.0606487 0.060493 0.1908663 .6189 29 1.38881829 1018 0.0752880 0.075395 0.2661543 .4412 30 1.62028801 1018 0.0878360 0.088238 0.3539903 .2645 31 1.77709008 1018 0.0963362 0.096971 0.4503265 .0887 32 1.83262414 1018 0.0993468 0.100070 0.5496733 +0.0887 33 1.77709008 1018 0.0963362 0.096971 0.6460095 +0.2645 34 1.62028801 1018 0.0878360 0.088238 0.7338455 +0.4412 35 1.38881829 1018 0.0752880 0.075395 0.8091335 +0.6189 36 1.11877029 1018 0.0606487 0.060493 0.8697822 +0.7965 37 0.84663698 1018 0.0458963 0.045576 0.9156785 +0.9736 38 0.60155785 1018 0.0326105 0.032244 0.9482890 +1.1504 39 0.40103857 1018 0.0217403 0.021421 0.9700293 +1.3279 40 0.25064910 1018 0.0135877 0.013363 0.9836170 +1.5077 41 0.14672143 1018 0.0079538 0.007828 0.9915708 +1.6919 42 0.08034745 1018 0.0043556 0.004306 0.9959264 72

PAGE 76

Allen R. Overman Nonlinear Models 43 0.04110800 1018 0.0022285 0.002224 0.9981549 44 0.01961973 1018 0.0010636 0.001079 0.9992185 45 0.00871988 1018 0.0004727 0.000491 0.9996912 46 0.00360169 1018 0.0001952 0.000210 0.9998864 47 0.00137937 1018 0.0000748 0.000084 0.9999612 48 0.00048853 1018 0.0000265 0.000032 0.9999877 49 0.00015952 1018 0.0000086 0.000011 0.9999963 50 0.00004786 1018 0.0000026 0.000004 0.9999989 51 0.00001314 1018 0.0000007 0.000001 0.9999996 52 0.00000328 1018 0.0000002 0.000000 0.9999998 53 0.00000074 1018 0.0000000 0.9999998 54 0.00000015 1018 55 0.00000003 1018 56 0.00000000 1018 57 58 59 60 61 62 63 64 64 1 C 18.4467441 1018 = 264 73

PAGE 77

Allen R. Overman Nonlinear Models Note the symmetry in the frequency distribution. x x F Z 17737.067576.5 2 1 2 12erf 1 r = 0.9999963 000.32,6380.52 2 26380.5 00.32 exp10007.0 2 exp 2 1 x x f f f 00622.1000252.0 r = 0.999978 74

PAGE 78

Allen R. Overman Nonlinear Models Case 6: 9 x 9 For this case n = 9 x 9 = 81 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A22. 81 ,xnc Table A22. Frequency distribution for the square peg board with a 9 x 9 array. x c f F Z f 0.000000 0 1 0.000000 0.000000 1 81 0.000000 0.000000 0.000000 2 3 4 5 6 14 0.00000002 1023 15 0.00000008 1023 16 0.00000034 1023 0.0000000 17 0.00000128 1023 0.0000000 0.0000000 18 0.00000457 1023 0.0000002 0.000000 0.0000002 19 0.00001514 1023 0.0000006 0.000001 0.0000008 20 0.00004694 1023 0.0000019 0.000003 0.0000027 21 0.00013636 1023 0.0000056 0.000007 0.0000083 22 0.00037190 1023 0.0000154 0.000018 0.0000237 23 0.00095400 1023 0.0000395 0.000044 0.0000632 24 0.00230549 1023 0.0000953 0.000103 75

PAGE 79

Allen R. Overman Nonlinear Models 0.0001585 25 0.00525652 1023 0.0002174 0.000228 0.0003759 26 0.01132173 1023 0.0004683 0.000480 0.0008442 27 0.02306279 1023 0.0009539 0.000961 0.0017981 28 0.04447825 1023 0.0018396 0.001834 0.0036377 29 0.08128783 1023 0.0033620 0.003329 0.0069997 .7415 30 0.14089890 1023 0.0058274 0.005750 0.0128271 .5769 31 0.23180142 1023 0.0095871 0.009449 0.0224142 .4162 32 0.36218972 1023 0.0149798 0.014777 0.0373940 .2579 33 0.53779686 1023 0.0222428 0.021988 0.0596368 .1008 34 0.75924263 1023 0.0314015 0.031133 0.0910383 .9439 35 1.01955439 1023 0.0421678 0.041944 0.1332061 .7866 36 1.30276395 1023 0.0538811 0.053771 0.1870872 .6288 37 1.58444264 1023 0.0655310 0.065592 0.2526182 .4709 38 1.83461779 1023 0.0758780 0.076134 0.3284962 .3136 39 2.02278372 1023 0.0836604 0.084087 0.4121566 .1571 40 2.12392290 1023 0.0878434 0.088370 0.5000000 0.0000 41 2.12392290 1023 0.0878434 0.088370 0.5878434 +0.1571 42 2.02278372 1023 0.0836604 0.084087 0.6715038 +0.3136 43 1.83461779 1023 0.0758780 0.076134 0.7473818 +0.4709 44 1.58444264 1023 0.0655310 0.065592 0.8129128 +0.6288 45 1.30276395 1023 0.0538811 0.053771 0.8667939 +0.7866 46 1.01955439 1023 0.0421678 0.041944 0.9089617 +0.9439 47 0.75924263 1023 0.0314015 0.031133 76

PAGE 80

Allen R. Overman Nonlinear Models 0.9403632 +1.1008 48 0.53779686 1023 0.0222428 0.021988 0.9626060 +1.2579 49 0.36218972 1023 0.0149798 0.014777 0.9775858 +1.4162 50 0.23180142 1023 0.0095871 0.009449 0.9871729 +1.5769 51 0.14089890 1023 0.0058274 0.005750 0.9930003 +1.7415 52 0.08128783 1023 0.0033620 0.003329 0.9963623 53 0.04447825 1023 0.0018396 0.001834 0.9982019 54 0.02306279 1023 0.0009539 0.000961 0.9991558 55 0.01132173 1023 0.0004683 0.000480 0.9996241 56 0.00525652 1023 0.0002174 0.000228 0.9998415 57 0.00230549 1023 0.0000953 0.000103 0.9999368 58 0.00095400 1023 0.0000395 0.000044 0.9999763 59 0.00037190 1023 0.0000154 0.000018 0.9999917 60 0.00013636 1023 0.0000056 0.000007 0.9999973 61 0.00004694 1023 0.0000019 0.000003 0.9999992 62 0.00001514 1023 0.0000006 0.000001 0.9999998 63 0.00000457 1023 0.0000002 0.000000 1.0000000 64 0.00000128 1023 0.0000000 0.000000 1.0000000 65 0.00000034 1023 0.0000000 1.0000000 66 0.00000008 1023 0.000000 1.0000000 67 0.00000002 1023 0.000000 1.0000000 68 0.00000000 1023 0.000000 1.0000000 69 0.00000000 1023 0.000000 1.0000000 C 24.1785164 1023 = 281 77

PAGE 81

Allen R. Overman Nonlinear Models Note the symmetry in the frequency distribution. x x F Z 157606.038305.6 2 1 2 12erf 1 r = 0.9999962 5000.40,34493.62 2 234493.6 50.40 exp08892.0 2 exp 2 1 x x f f f 00595.1000232.0 r = 0.999985 78

PAGE 82

Allen R. Overman Nonlinear Models Case 7: 10 x 10 For this case n = 10 x 10 = 100 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A23. 100 ,xnc Table A23. Frequency distribution for the square peg board with a 10 x 10 array. x c f F Z f 0.000000 0 1 0.000000 0.000000 1 100 0.000000 0.000000 0.000000 2 3 4 5 6 20 0.00000001 1029 21 0.00000002 1029 22 0.00000007 1029 23 0.00000025 1029 0.0000000 0.0000000 24 0.00000080 1029 0.0000001 0.0000001 25 0.00000243 1029 0.0000002 0.000000 0.0000003 26 0.00000700 1029 0.0000006 0.000001 0.0000009 27 0.00001917 1029 0.0000015 0.000002 0.0000024 28 0.00004999 1029 0.0000039 0.000005 0.0000063 29 0.00012411 1029 0.0000098 0.000011 0.0000161 30 0.00029372 1029 0.0000232 0.000026 79

PAGE 83

Allen R. Overman Nonlinear Models 0.0000393 31 0.00066325 1029 0.0000523 0.000056 0.0000916 32 0.00143012 1029 0.0001128 0.000118 0.0002044 33 0.00294692 1029 0.0002325 0.000239 0.0004369 34 0.00580717 1029 0.0004581 0.000465 0.0008950 35 0.01095067 1029 0.0008639 0.000867 0.0017589 36 0.01977205 1029 0.0015597 0.001553 0.0033186 37 0.03420030 1029 0.0026979 0.002673 0.0060165 .7815 38 0.05670049 1029 0.0044729 0.004420 0.0104894 .6325 39 0.09013924 1029 0.0071107 0.007020 0.0176001 .4871 40 0.13746234 1029 0.0108439 0.010708 0.0284440 .3440 41 0.20116440 1029 0.0158691 0.015692 0.0443131 .2024 42 0.28258809 1029 0.0222923 0.022088 0.0666054 .0613 43 0.38116533 1029 0.0300686 0.029866 0.0966740 .9202 44 0.49378236 1029 0.0389526 0.038790 0.1356266 .7787 45 0.61448471 1029 0.0484743 0.048394 0.1841009 .6367 46 0.73470998 1029 0.0579584 0.057996 0.2420593 .4946 47 0.84413487 1029 0.0665905 0.066763 0.3086498 .3529 48 0.93206559 1029 0.0735270 0.073826 0.3821768 .2119 49 0.98913083 1029 0.0780287 0.078417 0.4602055 .0711 50 1.00891345 1029 0.0795892 0.080010 0.5397947 +0.0711 51 0.98913083 1029 0.0780287 0.078417 0.6178234 +0.2119 52 0.93206559 1029 0.0735270 0.073826 0.6913504 +0.3529 53 0.84413487 1029 0.0665905 0.066763 80

PAGE 84

Allen R. Overman Nonlinear Models 0.7579409 +0.4946 54 0.73470998 1029 0.0579584 0.057996 0.8158993 +0.6367 55 0.61448471 1029 0.0484743 0.048394 0.8643736 +0.7787 56 0.49378236 1029 0.0389526 0.038790 0.9033262 +0.9202 57 0.38116533 1029 0.0300686 0.029866 0.9333948 +1.0613 58 0.28258809 1029 0.0222923 0.022088 0.9556871 +1.2024 59 0.20116440 1029 0.0158691 0.015692 0.9715562 +1.3440 60 0.13746234 1029 0.0108439 0.010708 0.9824001 +1.4871 61 0.09013924 1029 0.0071107 0.007020 0.9895108 +1.6325 62 0.05670049 1029 0.0044729 0.004420 0.9939837 +1.7815 63 0.03420030 1029 0.0026979 0.002673 0.9966816 64 0.01977205 1029 0.0015597 0.001553 0.9982413 65 0.01095067 1029 0.0008639 0.000867 0.9991052 66 0.00580717 1029 0.0004581 0.000465 0.9995633 67 0.00294692 1029 0.0002325 0.000239 0.9997958 68 0.00143012 1029 0.0001128 0.000118 0.9999086 69 0.00066325 1029 0.0000523 0.000056 0.9999609 70 0.00029372 1029 0.0000232 0.000026 0.9999841 71 0.00012411 1029 0.0000098 0.000011 0.9999939 72 0.00004999 1029 0.0000039 0.000005 0.9999978 73 0.00001917 1029 0.0000015 0.000002 0.9999993 74 0.00000700 1029 0.0000006 0.000001 0.9999999 75 0.00000243 1029 0.0000002 0.000000 1.0000001 76 0.00000080 1029 0.0000001 81

PAGE 85

Allen R. Overman Nonlinear Models 1.0000002 77 0.00000025 1029 0.0000000 1.0000002 78 0.00000007 1029 79 0.00000002 1029 80 0.00000001 1029 C 12.67650601 1029 = 2100 Note the symmetry in the frequency distribution. x x F Z 14181.00907.7 2 1 2 12erf 1 r = 0.9999959 000.50,0515.72 2 20515.7 00.50 exp080010.0 2 exp 2 1 x x f f f 00521.1000181.0 r = 0.999989 82

PAGE 86

Allen R. Overman Nonlinear Models Case 8: 12 x 12 For this case n = 12 x 12 = 144 and x assumes values of 0, 1, 2, 3, Corresponding values of are calculated from Eq. (1). Re sults are given in Table A24. 144 ,xnc Table A24. Frequency distribution for the square peg board with a 12 x 12 array. x c f F Z f 0.000000 0 1 0.000000 0.000000 1 144 0.000000 0.000000 0.000000 2 3 4 5 6 35 0.00000000 1042 0.000000 36 0.00000001 1042 0.000000 37 0.00000003 1042 0.000000 38 0.00000009 1042 0.00000000 39 0.00000025 1042 0.00000001 0.00000001 40 0.00000066 1042 0.00000003 0.00000004 41 0.00000168 1042 0.00000008 0.00000012 42 0.00000411 1042 0.00000018 0.000000 0.00000030 43 0.00000975 1042 0.00000044 0.000001 0.00000074 44 0.00002237 1042 0.00000100 0.000001 0.00000174 45 0.00004972 1042 0.00000223 0.000003 83

PAGE 87

Allen R. Overman Nonlinear Models 0.00000397 46 0.00010700 1042 0.00000480 0.000005 0.00000877 47 0.00022310 1042 0.00001000 0.000011 0.00001877 48 0.00045085 1042 0.00002022 0.000022 0.00003899 49 0.00088331 1042 0.00003961 0.000042 0.00007860 50 0.00167828 1042 0.00007526 0.000079 0.00015386 51 0.00309331 1042 0.00013871 0.000143 0.00029257 52 0.00553226 1042 0.00024808 0.000254 0.00054065 53 0.00960317 1042 0.00043062 0.000436 0.00097127 54 0.01618312 1042 0.00072568 0.000731 0.00169695 55 0.02648146 1042 0.00118747 0.001190 0.00288442 56 0.04208661 1042 0.00188723 0.001884 0.00477165 57 0.06497582 1042 0.00291362 0.002901 0.00768527 .7166 58 0.09746373 1042 0.00437042 0.004344 0.01205569 .5942 59 0.14206578 1042 0.00637045 0.006327 0.01842614 .4738 60 0.20125985 1042 0.00902480 0.008962 0.02745094 .3550 61 0.27714472 1042 0.01242760 0.012346 0.03987854 .2371 62 0.37101632 1042 0.01663695 0.016540 0.05651549 .1197 63 0.48291012 1042 0.02165444 0.021550 0.07816993 .0023 64 0.61118313 1042 0.02740640 0.027306 0.10557633 .8848 65 0.75222539 1042 0.03373095 0.033650 0.13930728 .7668 66 0.90039099 1042 0.04037493 0.040329 0.17968221 .6486 67 1.04821638 1042 0.04700365 0.047005 0.22668586 .5302 68 1.18695090 1042 0.05322472 0.053282 84

PAGE 88

Allen R. Overman Nonlinear Models 0.27991058 .4119 69 1.30736621 1042 0.05862433 0.058738 0.33853491 .2941 70 1.40074951 1042 0.06281178 0.062973 0.40134669 .1767 71 1.45993611 1042 0.06546580 0.065660 0.46681249 .0593 72 1.48021300 1042 0.06637505 0.066581 0.53318754 +0.0593 73 1.45993611 1042 0.06546580 0.065660 0.59865334 +0.1767 74 1.40074951 1042 0.06281178 0.062973 0.66146512 +0.2941 75 1.30736621 1042 0.05862433 0.058738 0.72008945 +0.4119 76 1.18695090 1042 0.05322472 0.053282 0.77331417 +0.5302 77 1.04821638 1042 0.04700365 0.047005 0.82031782 +0.6486 78 0.90039099 1042 0.04037493 0.040329 0.86069275 +0.7668 79 0.75222539 1042 0.03373095 0.033650 0.89442370 +0.8848 80 0.61118313 1042 0.02740640 0.027306 0.92183010 +1.0023 81 0.48291012 1042 0.02165444 0.021550 0.94348454 +1.1197 82 0.37101632 1042 0.01663695 0.016540 0.96012149 +1.2371 83 0.27714472 1042 0.01242760 0.012346 0.97254909 +1.3550 84 0.20125985 1042 0.00902480 0.008962 0.98157389 +1.4738 85 0.14206578 1042 0.00637045 0.006327 0.98794434 +1.5942 86 0.09746373 1042 0.00437042 0.004344 0.99231476 +1.7166 87 0.06497582 1042 0.00291362 0.002901 0.99522838 88 0.04208661 1042 0.00188723 0.001884 0.99711561 89 0.02648146 1042 0.00118747 0.001190 0.99830308 90 0.01618312 1042 0.00072568 0.000731 0.99902876 91 0.00960317 1042 0.00043062 0.000436 85

PAGE 89

Allen R. Overman Nonlinear Models 0.99945938 92 0.00553226 1042 0.00024808 0.000254 0.99970746 93 0.00309331 1042 0.00013871 0.000143 0.99984617 94 0.00167828 1042 0.00007526 0.000079 0.99992143 95 0.00088331 1042 0.00003961 0.000042 0.99996104 96 0.00045085 1042 0.00002022 0.000022 0.99998126 97 0.00022310 1042 0.00001000 0.000011 0.99999126 98 0.00010700 1042 0.00000480 0.000005 0.99999606 99 0.00004972 1042 0.00000223 0.000003 0.99999829 100 0.00002237 1042 0.00000100 0.000001 0.99999929 101 0.00000975 1042 0.00000044 0.000001 0.99999973 102 0.00000411 1042 0.00000018 0.000000 0.99999991 103 0.00000168 1042 0.00000008 0.99999999 104 0.00000066 1042 0.00000003 1.00000002 105 0.00000025 1042 0.00000001 1.00000003 106 0.00000009 1042 0.00000000 107 0.00000003 1042 0.000000 108 0.00000001 1042 0.000000 109 0.00000000 1042 0.000000 143 144 144 1 C 22.30074520 1042 = 2144 ok Note the symmetry in the frequency distribution. 86

PAGE 90

Allen R. Overman Nonlinear Models x x F Z 11801.049688.8 2 1 2 12erf 1 r = 0.9999985 00.72,47377.82 2 247377.8 00.72 exp066581.0 2 exp 2 1 x x f f f 00290.10000840.0 r = 0.9999960 Note that the total number of combinations is controlled by 2n. For n = 500 cells, this gives C = 3.2733906 10150 total combinations !!! Is this a large number ? Compared to what ? Compared to molecules in a container of gas it may not be so big !! Remember Avogadros number from chemistry (6.0 1023) ? Or to the number of micr oorganisms in a living body. Statistics entered the field of physics when James Clerk Maxwell used it in his kinetic theory of gases in the mid 1800s, and was then devel oped further by Ludwig Boltzmann and Willard Gibbs. This led to the branch of physics known as statistical mechanics These concepts were later incorporated in chemical kinetics by Henry Eyring in the absolute reaction rate theory. 87

PAGE 91

Allen R. Overman Nonlinear Models Table A25. Spread of the di stributions for peg boards with different number of holes (n ). Board n 2 2 2 n Linear 2 0.9534 1.00 1.0489 0.9534 5 1.6286 2.50 1.5351 1.0300 6 1.8003 3.00 1.6664 1.0394 7 1.8892 3.50 1.8526 1.0098 8 2.0312 4.00 1.9692 1.0156 Triangular 3 1.2213 1.50 1.2282 0.9972 6 1.8003 3.00 1.6664 1.0394 10 2.2939 5.00 2.1797 1.0259 15 2.7711 7.50 2.7065 1.0119 21 3.2707 10.50 3.2103 1.0094 28 3.7810 14.00 3.7027 1.0105 Square 4 1.4391 2.00 1.3898 1.0176 9 2.1653 4.50 2.0782 1.0207 16 2.8703 8.00 2.7872 1.0148 25 3.5740 12.50 3.4975 1.0109 36 4.2786 18.00 4.2070 1.0085 49 4.9687 24.50 4.93087 1.0038 64 5.67576 32.00 5.63801 1.0033 81 6.38305 40.50 6.34493 1.0030 100 7.0907 50.00 7.05149 1.0028 144 8.49672 72.00 8.473858 1.00135 Note: 1 2 n or n 2 spread of distribu tion is approaching n 2 n center of distribu tion is equal to n /2 References Abramowitz, M. and I.A. Stegun. 1965. Handbook of Mathematical Functions. Dover Publications. New York, NY. Eigen, M. and R. Winkler. 1993. Laws of the Game: How the Principles of Nature Govern Chance. Princeton University Press. Princeton, NJ. Polster, B. 2004. Q.E.D.: Beauty in Mathematical Proof. Walker & Co. New York, NY. Ruhla, C. 1992. The Physics of Chance: From Bl aise Pascal to Niels Bohr. Oxford University Press. New York, NY. Watkins, M. 2000. Useful Mathematical and Physical Formulae. Walker & Co. New York, NY. 88

PAGE 92

Allen R. Overman Nonlinear Models Table A26. Correlation between discrete ( f ) and continuous Gaussian ( ) distributions. f Pegboard Cells Regression Equation Correlation Linear 1 x 2 1 f f 2848.11045.0 1 x 3 1.0000 f f 1404.103885.0 1 x 4 0.9971 f f 1052.10226.0 1 x 5 0.99723 f f 0882.101543.0 1 x 6 0.99730 f f 0809.101187.0 1 x 7 0.999198 f f 0689.101188.0 1 x 8 0.999084 f f 0589.1008700.0 Triangular n = 3 1.000000 f f 1432.103950.0 n = 6 0.99730 f f 0809.101187.0 n = 10 0.99894 f f 0383.100352.0 n = 15 0.999637 f f 02423.1002046.0 n = 21 0.99980 f f 0183.100135.0 n = 28 0.99985 f f 01674.1000995.0 Square 2 x 2 0.9971 f f 1053.102262.0 3 x 3 0.99895 f f 0369.100373.0 4 x 4 099962 f f 0248.100194.0 5 x 5 0.999827 f f 017789.1001129.0 6 x 6 0.999909 f f 01356.1000723.0 7 x 7 0.999965 f f 00727.1000345.0 8 x 8 0.999978 f f 00622.1000252.0 9 x 9 0.999985 f f 00595.1000232.0 10 x 10 0.999989 f f 00521.1000181.0 12 x 12 0.9999960 f f 00290.10000840.0 As number of cells increases, the intercep t approaches 0 and slope approaches 1, which means that fit of the continuous Gaussian distribution to discrete distribution improves as the number of cells increases. 89

PAGE 93

Allen R. Overman Nonlinear Models ABE 6933 Special Topics Mathem atical and Statistical Characteristics of Nonlinear Regression Models A.R. Overman I. Elements of Probability and Calculus A. Arithmetic the process of counting B. Natural numbers positive integers ( ,2,1, 0) C. Rational numbers ratio of two integers ( ,3/2,3/1,2/1,,1/2,1/ 1) D. Irrational numbers (such as e, 2, etc.) E. Complex numbers z = x + i y with i = 1 F. Binomial theorem and Pascals triangle (a + b)0 = 1 (a + b)1 = a1 + b1 (a + b)2 = a2 + 2ab + b2 (a + b)3 = a3 + 3a2b + 3ab2 + b3 (a + b)4 = a4 + 4a3b + 6a2b2 + 4ab3 + b4 (a + b)5 = a5 + 5a4b + 10a3b2 + 10a2b3 + 5ab4 + b5 (a + b)6 = a6 + 6a5b + 15a4b2 + 20a3b3 + 15a2b4 + 6ab5 + b6 (a + b)7 = a7 + 7a6b + 21a5b2 + 35a4b3 + 35a3b4 + 21a2b5 + 7ab6 + b7 (a + b)8 = a8 + 8a7b + 28a6b2 + 56a5b3 + 70a4b4 + 56a3b5 + 28a2b6 + 8ab7 + b8 (a + b)9 = a9 + 9a8b + 36a7b2 + 84a6b3 + 126a5b4 + 126a4b5 + 84a3b6 + 36a2b7 + 9ab8 + b9 Note symmetry in the distribution of coefficients for each expansion. 90

PAGE 94

Allen R. Overman Nonlinear Models Pascals triangle for binomial coefficients 1 1 2 1 1 3 3 1 1 4 6 4 1 1 5 10 10 5 1 1 6 15 20 15 6 1 1 7 21 35 35 21 7 1 1 8 28 56 70 56 28 8 1 1 9 36 84 126 126 84 36 9 1 1 10 45 120 210 252 210 120 45 10 1 1 11 55 165 330 462 462 330 165 55 11 1 1 12 66 220 495 792 924 792 495 220 66 12 1 Note the pattern in the coefficients, including symmetry. G. Frequency distributions 1. Discrete distribution Consider the problem of a peg board. This is a two state system a cell (hole) is either filled or empty. Each cell holds one and only one object (peg), which can be viewed as a type of exclusion principle Define n as the total number of cells and x as the number of filled cells (pegs). Cells (holes) are indistinguishable (all alik e), as are the objects (pegs). Order of filling the cells is irrelevant. Note that a peg board can be linear, triangular, rectangular (Eigen and Winkler, 1993, p. 40; Polster, 2004, p. 33), or even 3-dimensional. The number of distinguishable combinations which are possible for each x, xnc ,, can be calculated from (Ruhla, 1992, p. 18; Watkins, 2000, p. 22) )!(! xnx n xnc (1) 91

PAGE 95

Allen R. Overman Nonlinear Models where and is called n f actorial. Note that n can assume positive integers n n 321! ( ) and x can also assum e positive integers (,3,2,1 n n x ,,2,1,0 ). For small values of n it is easy to estimate c by intuition, but larger n calculations of c are best performed on a pocket calculator or computer with the algorithm for co mputations (Eq. (1)) buil t in. The total number of combinations C for the system is defined as the sum of c values for all values of x and can be calculated from C = 2n. The frequency distribution of c values is then calculated from f = c/C. Cumulative frequency is calculated from the cumulative sum fF (2) so that F is normalized It should be noted that F for ms a discrete set of numbers for a particular case. 10 F 2. Continuous distribution The next step is to compare the discrete di stribution to a continuous Gaussian distribution where x is considered a continuous variable and the cumulative distribution is described by 2 erf1 2 1 x F (3) where and are the mean and spread of the distri bution. The error f unction is defined by 2 0 2)exp( 2 2 erfxduu x (4) where represents th e Gaussian distribution (bell-shaped curve). Values of the erf can be obtained from mathematical tables (cf. Abramowitz and Stegun, 1965, chp. 7). Some properties of the error function should be noted: ) exp(2u erf (0) = 0, erf ( ) = 1, erf ( x ) = erf (+x ), erf ( ) = 1 Equation (3) can be rear ranged to the linear form x F Z 2 1 2 12erf1 (5) where erf is the inverse error function. For example, Linear regression of Z vs. x leads to values of the parameters 00 .1)8427.0(erf1 and With these parameters now known the frequency distribution for f vs. x can be calculated for the continuous distribution from 92

PAGE 96

Allen R. Overman Nonlinear Models 22 exp 2 1 x f (6) The procedure can now be applied to a linear peg board, triangular peg board, and square peg board. It can even be applied to a 3dimensional system. This analysis falls within a branch of mathematics known as group theory Values of the error function can be calculat ed from the series a pproximation (Abramowitz and Stegun, 1965, p. 299) 44 3 2]078108.0 000972.0 230389.0278393.01[ 1 1erf x x x x x (7) for For the case where erf x is given, the inverse erf and therefore x can be obtained on a scientific calculator or computer using the solver routine. Note that for the case F < 0.5 and 2F < 0 (negative) the procedure is to change the value from to +, solve for the inverse by Eq. (7) and change the sign from + x to x Equation (7) does not work directly for x because the power series in Eq. (7) is not symmetric. 8.10 x H. Symmetry and conservation principle In all of the discrete and con tinuous Gaussian distributions we note symmetry in the distributions around a mean point. A mathematical consequen ce of this property is that something is conserved (remains constant) in the system. Note that the number of filled cells is defined by x Since this is a two-state (binary) system (cells are either empty of filled), it follows that the number of unfilled cells is n x The total capacity of the system is the sum of filled and unfilled cells so that total capacity is = x + n x = n While this is obvious for our case, it illustrates the connection between symmetry and c onservation. This property turns out to be very important in the various models of physics (including mechanics, electroma gnetism, relativity and quantum mechanics). It also shows up in chemistry and biology. 93

PAGE 97

Allen R. Overman Nonlinear Models Gaussian Distribution The Gauss differential equation is given by kxy dx dy with y = A at x = 0 (1) where y ) is a continuous function of x () 0( y x and k is the distribution coefficient. 1. Obtain the integral solution to Eq. (1). 2. Sketch the form of the solution y vs. x on linear graph paper. 3. Perform the 2nd derivative of y on x to obtain the inflection points at x Write the constant k in terms of 4. Evaluate the constant A by normalizing the integral (2) 1 ydx 5. Write the resulting solution y in terms of variable x and parameter 6. Obtain the cumulative probability distribution F 22 erf1 2 1xx ydx F (3) where the error function is defined by zduu z0 2exp 2 erf (4) 7. Calculate and plot F vs. 2/ x on linear graph paper. 94

PAGE 98

Allen R. Overman Nonlinear Models Solutions N ote: at x = 0, dy/dx = 0 (maximum or minimum) 1. 2 22 1 exp 2 1 kx Aydxkdxkx y dy Solution is symmetric around x = 0 3. 2 2 2 2 21 0 2 1 exp1 kkx kxkA dx yd 2 22 expx Ay at x = 0, d2y/dx2 < 0 (maximum) 4. 2 1 1 2 exp2 2 2 exp2 2 exp2 2 2 2 A A duu A x d x Adx x Aydx A is chosen so that the distribution is normalized, hence th e term normal distribution. 5. 22 exp 2 1 x y 6. 2 erf1 2 1 exp 2 1 2 1 exp 2 1 exp exp 1 exp 12 0 2 2 0 2 22 0 2 0 2 2 2x duu duu duu duu duu ydx Fx x xxx where 2 0 2exp 2 2 erfxduu x This ties the cumulative frequency distribution to the error function of mathematical physics. Note the characteristics of the error function: erf (0) = 0, erf ( ) = 1, erf ( x ) = erf (+ x ). It follows that F is bounded by 10 F. Note also that F is a well-behaved function. 95

PAGE 99

Allen R. Overman Nonlinear Models Solve Eq. (1) by the pow er series method. kxy dx dy with y = A at x = 0 (1) Assume that the solution is given by a power series 7 7 6 6 5 5 4 4 3 3 2 210xaxaxaxaxaxaxaay (2) The first derivative is given by 6 7 5 6 4 5 3 4 2 3 21765432 xaxaxaxaxaxaa dx dy (3) Substitution of Eqs. (2) and (3) into Eq. (1) leads to 7 6 6 5 5 4 4 3 3 2 2 1 0 6 6 5 5 4 4 3 3 2 210 6 7 5 6 4 5 3 4 2 3 21765432xkaxkaxkaxkaxkaxkaxka xaxaxaxaxaxaakx xaxaxaxaxaxaa (4) Equating like coefficients in Eq. (4) gives the recursion relations 6426 0 5 424 0 3 2 00 3 4 6 3 5 0 2 2 4 1 3 0 2 1 ak ka a ka a ak ka a ka a ka a a It follows that the solution is given by 96

PAGE 100

Allen R. Overman Nonlinear Models 6 3 4 2 2 0 6 0 3 4 0 2 2 0 0 6 6 5 5 4 4 3 3 2 210642422 1 642 0 42 0 2 0 x k x k x k a x ak x ak x ka a xaxaxaxaxaxaay (5) Now use the substitution 22 x k (6) Then Eq. (5) becomes 2 0 0 32 0 3 2 02 exp exp !3!2 1 32121 1 x k a a a ay (7) The constant is evaluated from the boundary condition, which leads to 0a 22 exp x k Ay (8) This is the famous Gaussi an distribution centered at x = 0. It remains to determine k in terms of the variance of the distribution and A to normalize the distribution. Check: kxyx k kxA dx dy 22 exp correct (9) 97

PAGE 101

Allen R. Overman Nonlinear Models Probability distributions with dice. Consider a single d ie with 6 faces numbered 1 through 6. Assume that each number is equally likely. Th e frequency distribution can now be calculated. Table A27. Frequency distri bution for a single die x S c f F Z Z f 0.00000000 1 1 1 0.16666667 0.093758 0.16666667 .684656 .669385 2 2 1 0.16666667 0.146760 0.33333334 .304151 .334693 3 3 1 0.16666667 0.183615 0.50000001 0.000000 0.000000 4 4 1 0.16666667 0.183615 0.66666668 +0.304151 +0.334693 5 5 1 0.16666667 0.146760 0.83333335 +0.684656 +0.669385 6 6 1 0.16666667 0.093758 1.00000002 C 6 S S F Z 334693.0171424.1 2 1 2 12erf 1 r = 0.998961 5000.3,9878.22 2 29878.2 5000.3 exp18883.0 2 exp 2 1 S S f No correlation between and f since f = constant = 1/6. f 98

PAGE 102

Allen R. Overman Nonlinear Models Table A28. Frequency distribution for two dice x S c f F Z Z f c 0.00000000 1,1 2 1 0.02777778 0.020242 0.73 0.02777778 .3513 .2999 1,2; 2,1 3 2 0.05555556 0.042894 1.54 0.08333334 .9781 .0110 1,3; 2,2; 3,1 4 3 0.08333333 0.076922 2.77 0.16666667 .6847 .7222 1,4; 2,3; 3,2; 4,1 5 4 0.11111111 0.116745 4.20 0.27777778 .4164 .4333 1,5; 2;4; 3,3; 4,2; 5,1 6 5 0.13888889 0.149951 5.40 0.41666667 .1490 .1444 1,6; 2,5; 3,4; 4,3; 5,2; 6,1 7 6 0.16666667 0.163000 5.87 0.58333334 +0.1490 +0.1444 2,6; 3,5; 4,4; 5,3; 6,2 8 5 0.13888889 0.149951 5.40 0.72222223 +0.4164 +0.4333 3,6; 4,5; 5,4; 6,3 9 4 0.11111111 0.116745 4.20 0.83333334 +0.6847 +0.7222 4,6; 5,5; 6,4 10 3 0.08333333 0.076922 2.77 0.91666667 +0.9781 +1.0110 5,6; 6,5 11 2 0.05555556 0.042894 1.54 0.97222223 +1.3513 +1.2999 6,6 12 1 0.02777778 0.020242 0.73 1.00000001 C 36 = 62 S S F Z 28885.002195.2 2 1 2 12erf 1 r = 0.999211 0000.7,4620.32 2 24620.3 0000.7 exp1630.0 2 exp 2 1 S S f f f 1360.10145.0 r = 0.99319 fc 36 The two dice problem is discussed by Speyer (1994, p. 62) 99

PAGE 103

Allen R. Overman Nonlinear Models Table A29. Frequency dist ribution for three dice x S c f F Z Z f 0.000000 1,1,1 3 1 0.004630 0.005418 0.004630 1,1,2; 1,2,1; 2,1,1 4 3 0.013889 0.012058 0.018519 .4726 .4344 1,1,3; 1,2,2; 1,3,1; 2,1,2; 5 6 0.027778 0.023940 2,2,1; 3,1,1 0.046297 .1878 .1953 1,1,4; 1,2,3; 1,3,2; 1,4,1; 6 10 0.046296 0.042397 2,1,3; 2,2,2; 2,3,1; 3,1,2; 0.092593 .9372 .9563 3,2,1; 4,1,1 1,1,5; 1,2,4; 1,3,3; 1,4,2; 7 15 0.069444 0.066973 1,5,1; 2,1,4; 2,2,3; 2,3,2; 0.162037 .6979 .7172 2,4,1; 3,1,3; 3,2,2; 3,3,1; 4,1,2; 4,2,1; 5,1,1 1,1,6; 1,2,5; 1,3,4; 1,4,3; 8 21 0.097222 0.094367 1,5,2; 1,6,1; 2,1,5; 2,2,4; 0.259259 .4563 .4781 2,3,3; 2,4,2; 2,5,1; 3,1,4; 3,2,3; 3,3,2; 3,4,1; 4,1,3; 4,2,2; 4,3,1; 5,1,2; 5,2,1; 6,1,1 1,2,6; 1,3,5; 1,4,4; 1,5,3; 9 25 0.115741 0.118605 1,6,2; 2,1,6; 2,2,5; 2,3,4; 0.375000 .2251 .2391 2,4,3; 2,5,2; 2,6,1; 3,1,5; 3,2,4; 3,3,3; 3,4,2; 3,5,1; 4,1,4; 4,2,3; 4,3,2; 4,4,1; 5,1,3; 5,2,2; 5,3,1; 6,1,2; 6,2,1 1,3,6; 1,4,5; 1,5,4; 1,6,3; 10 27 0.125000 0.132967 2,2,6; 2,3,5; 2,4,4; 2,5,3; 0.500000 0.0000 0.0000 2,6,2; 3,1,6; 3,2,5; 3,3,4; 3,4,3; 3,5,2; 3,6,1; 4,1,5; 4,2,4; 4,3,3; 4,4,2; 4,5,1; 5,1,4; 5,2,3; 5,3,2; 5,4,1; 6,1,3; 6,2,2; 6,3,1 1,4,6; 1,5,5; 1,6,4; 2,3,6; 11 27 0.125000 0.132967 2,4,5; 2,5,4; 2,6,3; 3,2,6; 0.625000 +0.2251 +0.2391 3,3,5; 3,4,4; 3,5,3; 3,6,2; 100

PAGE 104

Allen R. Overman Nonlinear Models 4,1,6; 4,2,5; 4,3,4; 4,4,3; 4,5,2; 4,6,1; 5,1,5; 5,2,4; 5,3,3; 5,4,2; 5,5,1; 6,1,4; 6,2,3; 6,3,2; 6,4,1 1,5,6; 1,6,5; 2,4,6; 2,5,5; 12 25 0.115741 0.118605 2,6,4; 3,3,6; 3,4,5; 3,5,4; 0.740741 +0.4563 +0.4781 3,6,3; 4,2,6; 4,3,5; 4,4,4; 4,5,3; 4,6,2; 5,1,6; 5,2,5; 5,3,4; 5,4,3; 5,5,2; 5,6,1; 6,1,5; 6,2,4; 6,3,3; 6,4,2; 6,5,1 1,6,6; 2,5,6; 2,6,5; 3,4,6; 13 21 0.097222 0.094367 3,5,5; 3,6,4; 4,3,6; 4,4,5; 0.837963 +0.6979 +0.7172 4,5,4; 4,6,3; 5,2,6; 5,3,5; 5,4,4; 5,5,3; 5,6,2; 6,1,6; 6,2,5; 6,3,4; 6,4,3; 6,5,2; 6,6,1 2,6,6; 3,5,6; 3,6,5; 4,4,6; 14 15 0.069444 0.066973 4,5,5; 4,6,4; 5,3,6; 5,4,5; 0.907407 +0.9372 +0.9563 5,5,4; 5,6,3; 6,2,6; 6,3,5; 6,4,4; 6,5,3; 6,6,2 3,6,6; 4,5,6; 4,6,5; 5,4,6; 15 10 0.046296 0.042397 5,5,5; 5,6,4; 6,3,6; 6,4,5; 0.953703 +1.1878 +1.1953 6,5,4; 6,6,3; 4,6,6; 5,5,6; 5,6,5; 6,4,6; 16 6 0.027778 0.023940 6,5,5; 6,6,4 0.981481 +1.4726 +1.4344 5,6,6; 6,5,6; 6,6,5 17 3 0.013889 0.012058 0.995370 6,6,6 18 1 0.004630 0.005418 1.000000 C 216 = 63 S S F Z 23906.05102.2 2 1 2 12erf 1 r = 0.999719 5001.10,1830.42 101

PAGE 105

Allen R. Overman Nonlinear Models 2 21830.4 500.10 exp13488.0 2 exp 2 1 S S f f f 0752.100590.0 r = 0.99800 For three dice the discrete di stribution is closely approximated by the continuous Gaussian distribution. I conclude that the peg board is a much simpler illustration of frequency distributions than dice. 102

PAGE 106

Allen R. Overman Nonlinear Models Table A30. Frequency distri bution for four dice x S c f 1,1,1,1 4 1 0.000772 1,1,1,2; 1,1,2,1; 1,2,1,1; 5 4 0.003086 2,1,1,1 1,1,1,3; 1,1,2,2; 1,1,3,1; 6 10 0.007716 1,2,1,2; 1,2,2,1; 1,3,1,1; 2,1,1,2; 2,1,2,1; 2,2,1,1; 3,1,1,1 1,1,1,4; 1,1,2,3; 1,1,3,2; 7 20 0.015432 1,1,4,1; 1,2,1,3; 1,2,2,2; 1,2,3,1; 1,3,1,2; 1,3,2,1; 1,4,1,1; 2,1,1,3; 2,1,2,2; 2,1,3,1; 2,2,1,2; 2,2,2,1; 2,3,1,1; 3,1,1,2; 3,1,2,1; 3,2,1,1; 4,1,1,1 1,1,1,5; 1,1,2,4; 1,1,3,3; 8 35 0.027006 1,1,4,2; 1,1,5,1; 1,2,1,4; 1,2,2,3; 1,2,3,2; 1,2,4,1; 1,3,1,3; 1,3,2,2; 1,3,3,1; 1,4,1,2; 1,4,2,1; 1,5,1,1; 2,1,1,4; 2,1,2,3; 2,1,3,2; 2,1,4,1; 2,2,1,3; 2,2,2,2; 2,2,3,1; 2,3,1,2; 2,3,2,1; 2,4,1,1; 3,1,1,3; 3,1,2,2; 3,1,3,1; 3,2,1,2; 3,2,2,1; 3,3,1,1; 4,1,1,2; 4,1,2,1; 4,2,1,1; 5,1,1,1 1,1,1,6; 1,1,2,5; 1,1,3,4; 9 56 0.043210 1,1,4,3; 1,1,5,2; 1,1,6,1; 1,2,1,5; 1,2,2,4; 1,2,3,3; 1,2,4,2; 1,2,5,1; 1,3,1,4; 1,3,2,3; 1,3,3,2; 1,3,4,1; 1,4,1,3; 1,4,2,2; 1,4,3,1; 1,5,1,2; 1,5,2,1; 1,6,1,1; 2,1,1,5; 2,1,2,4; 2,1,3,3; 2,1,4,2; 2,1,5,1; 2,2,1,4; 2,2,2,3; 2,2,3,2; 2,2,4,1; 2,3,1,3; 2,3,2,2; 2,3,3,1; 103

PAGE 107

Allen R. Overman Nonlinear Models 2,4,1,2; 2,4,2,1; 2,5,1,1; 3,1,1,4; 3,1,2,3; 3,1,3,2; 3,1,4,1; 3,2,1,3; 3,2,2,2; 3,2,3,1; 3,3,1,2; 3,3,2,1; 3,4,1,1; 4,1,1,3; 4,1,2,2; 4,1,3,1; 4,2,1,2; 4,2,2,1; 4,3,1,1; 5,1,1,2; 5,1,2,1; 5,2,1,1; 6,1,1,1 1,1,2,6; 1,1,3,5; 1,1,4,4; 10 80 0.061728 1,1,5,3; 1,1,6,2; 1,2,1,6; 1,2,2,5; 1,2,3,4; 1,2,4,3; 1,2,5,2; 1,2,6,1; 1,3,1,5; 1,3,2,4; 1,3,3,3; 1,3,4,2; 1,3,5,1; 1,4,1,4; 1,4,2,3; 1,4,3,2; 1,4,4,1; 1,5,1,3; 1,5,2,2; 1,5,3,1; 1,6,1,2; 1,6,2,1; 2,1,1,6; 2,1,2,5; 2,1,3,4; 2,1,4,3; 2,1,5,2; 2,1,6,1; 2,2,1,5; 2,2,2,4; 2,2,3,3; 2,2,4,2; 2,2,5,1; 2,3,1,4; 2,3,2,3; 2,3,3,2; 2,3,4,1; 2,4,1,3; 2,4,2,2; 2,4,3,1; 2,5,1,2; 2,5,2,1; 2,6,1,1; 3,1,1,5; 3,1,2,4; 3,1,3,3; 3,1,4,2; 3,1,5,1; 3,2,1,4; 3,2,2,3; 3,2,3,2; 3,2,4,1; 3,3,1,3; 3,3,2,2; 3,3,3,1; 3,4,1,2; 3,4,2,1; 3,5,1,1; 4,1,1,4; 4,1,2,3; 4,1,3,2; 4,1,4,1; 4,2,1,3; 4,2,2,2; 4,2,3,1; 4,3,1,2; 4,3,2,1; 4,4,1,1; 5,1,1,3; 5,1,2,2; 5,1,3,1; 5,2,1,2; 5,2,2,1; 5,3,1,1; 6,1,1,2; 6,1,2,1; 6,2,1,1 1,1,3,6; 1,1,4,5; 1,1,5,4; 11 104 0.080247 1,1,6,3; 1,2,2,6; 1,2,3,5; 1,2,4,4; 1,2,5,3; 1,2,6,2; 1,3,1,6; 1,3,2,5; 1,3,3,4; 1,3,4,3; 1,3,5,2; 1,3,6,1; 1,4,1,5; 1,4,2,4; 1,4,3,3; 1,4,4,2; 1,4,5,1; 1,5,1,4; 1,5,2,3; 1,5,3,2; 1,5,4,1; 1,6,1,3; 1,6,2,2; 1,6,3,1; 104

PAGE 108

Allen R. Overman Nonlinear Models 2,1,2,6; 2,1,3,5; 2,1,4,4; 2,1,5,3; 2,1,6,2; 2,2,1,6; 2,2,2,5; 2,2,3,4; 2,2,4,3; 2,2,5,2; 2,2,6,1; 2,3,1,5; 2,3,2,4; 2,3,3,3; 2,3,4,2; 2,3,5,1; 2,4,1,4; 2,4,2,3; 2,4,3,2; 2,4,4,1; 2,5,1,3; 2,5,2,2; 2,5,3,1; 2,6,1,2; 2,6,2,1; 3,1,1,6; 3,1,2,5; 3,1,3,4; 3,1,4,3; 3,1,5,2; 3,1,6,1; 3,2,1,5; 3,2,2,4; 3,2,3,3; 3,2,4,2; 3,2,5,1; 3,3,1,4; 3,3,2,3; 3,3,3,2; 3,3,4,1; 3,4,1,3; 3,4,2,2; 3,4,3,1; 3,5,1,2; 3,5,2,1; 3,6,1,1; 4,1,1,5; 4,1,2,4; 4,1,3,3; 4,1,4,2; 4,1,5,1; 4,2,1,4; 4,2,2,3; 4,2,3,2; 4,2,4,1; 4,3,1,3; 4,3,2,2; 4,3,3,1; 4,4,1,2; 4,4,2,1; 4,5,1,1; 5,1,1,4; 5,1,2,3; 5,1,3,2; 5,1,4,1; 5,2,1,3; 5,2,2,2; 5,2,3,1; 5,3,1,2; 5,3,2,1; 5,4,1,1; 6,1,1,3; 6,1,2,2; 6,1,3,1; 6,2,1,2; 6,2,2,1; 6,3,1,1 1,1,4,6; 1,1,5,5; 1,1,6,4; 12 125 0.096451 1,2,3,6; 1,2,4,5; 1,2,5,4; 1,2,6,3; 1,3,2,6; 1,3,3,5; 1,3,4,4; 1,3,5,3; 1,3,6,2; 1,4,1,6; 1,4,2,5; 1,4,3,4; 1,4,4,3; 1,4,5,2; 1,4,6,1; 1,5,1,5; 1,5,2,4; 1,5,3,3; 1,5,4,2; 1,5,5,1; 1,6,1,4; 1,6,2,3; 1,6,3,2; 1,6,4,1; 2,1,3,6; 2,1,4,5; 2,1,5,4; 2,1,6,3; 2,2,2,6; 2,2,3,5; 2,2,4,4; 2,2,5,3; 2,2,6,2; 2,3,1,6; 2,3,2,5; 2,3,3,4; 2,3,4,3; 2,3,5,2; 2,3,6,1; 2,4,1,5; 2,4,2,4; 2,4,3,3; 2,4,4,2; 2,4,5,1; 2,5,1,4; 2,5,2,3; 2,5,3,2; 2,5,4,1; 2,6,1,3; 2,6,2,2; 2,6,3,1; 3,1,2,6; 3,1,3,5; 3,1,4,4; 105

PAGE 109

Allen R. Overman Nonlinear Models 3,1,5,3; 3,1,6,2; 3,2,1,6; 3,2,2,5; 3,2,3,4; 3,2,4,3; 3,2,5,2; 3,2,6,1; 3,3,1,5; 3,3,2,4; 3,3,3,3; 3,3,4,2; 3,3,5,1; 3,4,1,4; 3,4,2,3; 3,4,3,2; 3,4,4,1; 3,5,1,3; 3,5,2,2; 3,5,3,1; 3,6,1,2; 3,6,2,1; 4,1,1,6; 4,1,2,5; 4,1,3,4; 4,1,4,3; 4,1,5,2; 4,1,6,1; 4,2,1,5; 4,2,2,4; 4,2,3,3; 4,2,4,2; 4,2,5,1; 4,3,1,4; 4,3,2,3; 4,3,3,2; 4,3,4,1; 4,4,1,3; 4,4,2,2; 4,4,3,1; 4,5,1,2; 4,5,2,1; 4,6,1,1; 5,1,1,5; 5,1,2,4; 5,1,3,3; 5,1,4,2; 5,1,5,1; 5,2,1,4; 5,2,2,3; 5,2,3,2; 5,2,4,1; 5,3,1,3; 5,3,2,2; 5,3,3,1; 5,4,1,2; 5,4,2,1; 5,5,1,1; 6,1,1,4; 6,1,2,3; 6,1,3,2; 6,1,4,1; 6,2,1,3; 6,2,2,2; 6,2,3,1; 6,3,1,2; 6,3,2,1; 6,4,1,1 1,1,5,6; 1,1,6,5; 1,2,4,6; 13 140 0.108025 1,2,5,5; 1,2,6,4; 1,3,3,6; 1,3,4,5; 1,3,5,4; 1,3,6,3; 1,4,2,6; 1,4,3,5; 1,4,4,4; 1,4,5,3; 1,4,6,2; 1,5,1,6; 1,5,2,5; 1,5,3,4; 1,5,4,3; 1,5,5,2; 1,5,6,1; 1,6,1,5; 1,6,2,4; 1,6,3,3; 1,6,4,2; 1,6,5,1; 2,1,4,6; 2,1,5,5; 2,1,6,4; 2,2,3,6; 2,2,4,5; 2,2,5,4; 2,2,6,3; 2,3,2,6; 2,3,3,5; 2,3,4,4; 2,3,5,3; 2,3,6,2; 2,4,1,6; 2,4,2,5; 2,4,3,4; 2,4,4,3; 2,4,5,2; 2,4,6,1; 2,5,1,5; 2,5,2,4; 2,5,3,3; 2,5,4,2; 2,5,5,1; 2,6,1,4; 2,6,2,3; 2,6,3,2; 2,6,4,1; 3,1,3,6; 3,1,4,5; 3,1,5,4; 3,1,6,3; 3,2,2,6; 3,2,3,5; 3,2,4,4; 3,2,5,3; 3,2,6,2; 3,3,1,6; 3,3,2,5; 3,3,3,4; 3,3,4,3; 3,3,5,2; 106

PAGE 110

Allen R. Overman Nonlinear Models 3,3,6,1; 3,4,1,5; 3,4,2,4; 3,4,3,3; 3,4,4,2; 3,4,5,1; 3,5,1,4; 3,5,2,3; 3,5,3,2; 3,5,4,1; 3,6,1,3; 3,6,2,2; 3,6,3,1; 4,1,2,6; 4,1,3,5; 4,1,4,4; 4,1,5,3; 4,1,6,2; 4,2,1,6; 4,2,2,5; 4,2,3,4; 4,2,4,3; 4,2,5,2; 4,2,6,1; 4,3,1,5; 4,3,2,4; 4,3,3,3; 4,3,4,2; 4,3,5,1; 4,4,1,4; 4,4,2,3; 4,4,3,2; 4,4,4,1; 4,5,1,3; 4,5,2,2; 4,5,3,1; 4,6,1,2; 4,6,2,1; 5,1,1,6; 5,1,2,5; 5,1,3,4; 5,1,4,3; 5,1,5,2; 5,1,6,1; 5,2,1,5; 5,2,2,4; 5,2,3,3; 5,2,4,2; 5,2,5,1; 5,3,1,4; 5,3,2,3; 5,3,3,2; 5,3,4,1; 5,4,1,3; 5,4,2,2; 5,4,3,1; 5,5,1,2; 5,5,2,1; 5,6,1,1; 6,1,1,5; 6,1,2,4; 6,1,3,3; 6,1,4,2; 6,1,5,1; 6,2,1,4; 6,2,2,3; 6,2,3,2; 6,2,4,1; 6,3,1,3; 6,3,2,2; 6,3,3,1; 6,4,1,2; 6,4,2,1; 6,5,1,1 1,1,6,6; 1,2,5,6; 1,2,6,5; 14 146 0.112654 1,3,4,6; 1,3,5,5; 1,3,6,4; 1,4,3,6; 1,4,4,5; 1,4,5,4; 1,4,6,3; 1,5,2,6; 1,5,3,5; 1,5,4,4; 1,5,5,3; 1,5,6,2; 1,6,1,6; 1,6,2,5; 1,6,3,4; 1,6,4,3; 1,6,5,2; 1,6,6,1; 2,1,5,6; 2,1,6,5; 2,2,4,6; 2,2,5,5; 2,2,6,4; 2,3,3,6; 2,3,4,5; 2,3,5,4; 2,3,6,3; 2,4,2,6; 2,4,3,5; 2,4,4,4; 2,4,5,3; 2,4,6,2; 2,5,1,6; 2,5,2,5; 2,5,3,4; 2,5,4,3; 2,5,5,2; 2,5,6,1; 2,6,1,5; 2,6,2,4; 2,6,3,3; 2,6,4,2; 2,6,5,1; 3,1,4,6; 3,1,5,5; 3,1,6,4; 3,2,3,6; 3,2,4,5; 3,2,5,4; 3,2,6,3; 3,3,2,6; 3,3,3,5; 3,3,4,4; 3,3,5,3; 3,3,6,2; 3,4,1,6; 3,4,2,5; 107

PAGE 111

Allen R. Overman Nonlinear Models 3,4,3,4; 3,4,4,3; 3,4,5,2; 3,4,6,1; 3,5,1,5; 3,5,2,4; 3,5,3,3; 3,5,4,2; 3,5,5,1; 3,6,1,4; 3,6,2,3; 3,6,3,2; 3,6,4,1; 4,1,3,6; 4,1,4,5; 4,1,5,4; 4,1,6,3; 4,2,2,6; 4,2,3,5; 4,2,4,4; 4,2,5,3; 4,2,6,2; 4,3,1,6; 4,3,2,5; 4,3,3,4; 4,3,4,3; 4,3,5,2; 4,3,6,1; 4,4,1,5; 4,4,2,4; 4,4,3,3; 4,4,4,2; 4,4,5,1; 4,5,1,4; 4,5,2,3; 4,5,3,2; 4,5,4,1; 4,6,1,3; 4,6,2,2; 4,6,3,1; 5,1,2,6; 5,1,3,5; 5,1,4,4; 5,1,5,3; 5,1,6,2; 5,2,1,6; 5,2,2,5; 5,2,3,4; 5,2,4,3; 5,2,5,2; 5,2,6,1; 5,3,1,5; 5,3,2,4; 5,3,3,3; 5,3,4,2; 5,3,5,1; 5,4,1,4; 5,4,2,3; 5,4,3,2; 5,4,4,1; 5,5,1,3; 5,5,2,2; 5,5,3,1; 5,6,1,2; 5,6,2,1; 6,1,1,6; 6,1,2,5; 6,1,3,4; 6,1,4,3; 6,1,5,2; 6,1,6,1; 6,2,1,5; 6,2,2,4; 6,2,3,3; 6,2,4,2; 6,2,5,1; 6,3,1,4; 6,3,2,3; 6,3,3,2; 6,3,4,1; 6,4,1,3; 6,4,2,2; 6,4,3,1; 6,5,1,2; 6,5,2,1; 6,6,1,1 1,2,6,6; 1,3,5,6; 1,3,6,5; 15 140 0.108025 1,4,4,6; 1,4,5,5; 1,4,6,4; 1,5,3,6; 1,5,4,5; 1,5,5,4; 1,5,6,3; 1,6,2,6; 1,6,3,5; 1,6,4,4; 1,6,5,3; 1,6,6,2; 2,1,6,6; 2,2,5,6; 2,2,6,5; 2,3,4,6; 2,3,5,5; 2,3,6,4; 2,4,3,6; 2,4,4,5; 2,4,5,4; 2,4,6,3; 2,5,2,6; 2,5,3,5; 2,5,4,4; 2,5,5,3; 2,5,6,2; 2,6,1,6; 2,6,2,5; 2,6,3,4; 2,6,4,3; 2,6,5,2; 2,6,6,1; 3,1,5,6; 3,1,6,5; 3,2,4,6; 3,2,5,5; 3,2,6,4; 3,3,3,6; 3,3 4,5; 3,3,5,4; 3,3,6,3; 3,4,2,6; 3,4,3,5; 3,4,4,4; 108

PAGE 112

Allen R. Overman Nonlinear Models 3,4,5,3; 3,4,6,2; 3,5,1,6; 3,5,2,5; 3,5,3,4; 3,5,4,3; 3,5,5,2; 3,5,6,1; 3,6,1,5; 3,6,2,4; 3,6,3,3; 3,6,4,2; 3,6,5,1; 4,1,4,6; 4,1,5,5; 4,1,6,4; 4,2,3,6; 4,2,4,5; 4,2,5,4; 4,2,6,3; 4,3,2,6; 4,3,3,5; 4,3,4,4; 4,3,5,3; 4,3,6,2; 4,4,1,6; 4,4,2,5; 4,4,3,4; 4,4,4,3; 4,4,5,2; 4,4,6,1; 4,5,1,5; 4,5,2,4; 4,5,3,3; 4,5,4,2; 4,5,5,1; 4,6,1,4; 4,6,2,3; 4,6,3,2; 4,6,4,1; 5,1,3,6; 5,1,4,5; 5,1,5,4; 5,1,6,3; 5,2,2,6; 5,2,3,5; 5,2,4,4; 5,2,5,3; 5,2,6,2; 5,3,1,6; 5,3,2,5; 5,3,3,4; 5,3,4,3; 5,3,5,2; 5,3,6,1; 5,4,1,5; 5,4,2,4; 5,4,3,3; 5,4,4,2; 5,4,5,1; 5,5,1,4; 5,5,2,3; 5,5,3,2; 5,5,4,1; 5,6,1,3; 5,6,2,2; 5,6,3,1; 6,1,2,6; 6,1,3,5; 6,1,4,4; 6,1,5,3; 6,1,6,2; 6,2,1,6; 6,2,2,5; 6,2,3,4; 6,2,4,3; 6,2,5,2; 6,2,6,1; 6,3,1,5; 6,3,2,4; 6,3,3,3; 6,3,4,2; 6,3,5,1; 6,4,1,4; 6,4,2,3; 6,4,3,2; 6,4,4,1; 6,5,1,3; 6,5,2,2; 6,5,3,1; 6,6,1,2; 6,6,2,1 1,3,6,6; 1,4,5,6; 1,4,6,5; 16 125 0.096451 1,5,4,6; 1,5,5,5; 1,5,6,4; 1,6,3,6; 1,6,4,5; 1,6,5,4; 1,6,6,3; 2,2,6,6; 2,3,5,6; 2,3,6,5; 2,4,4,6; 2,4,5,5; 2,4,6,4; 2,5,3,6; 2,5,4,5; 2,5,5,4; 2,5,6,3; 2,6,2,6; 2,6,3,5; 2,6,4,4; 2,6,5,3; 2,6,6,2; 3,1,6,6; 3,2,5,6; 3,2,6,5; 3,3,4,6; 3,3,5,5; 3,3,6,4; 3,4,3,6; 3,4,4,5; 3,4,5,4; 3,4,6,3; 3,5,2,6; 3,5,3,5; 3,5,4,4; 3,5,5,3; 3,5,6,2; 3,6,1,6; 3,6,2,5; 109

PAGE 113

Allen R. Overman Nonlinear Models 3,6,3,4; 3,6,4,3; 3,6,5,2; 3,6,6,1; 4,1,5,6; 4,1,6,5; 4,2,4,6; 4,2,5,5; 4,2,6,4; 4,3,3,6; 4,3,4,5; 4,3,5,4; 4,3,6,3; 4,4,2,6; 4,4,3,5; 4,4,4,4; 4,4,5,3; 4,4,6,2; 4,5,1,6; 4,5,2,5; 4,5,3,4; 4,5,4,3; 4,5,5,2; 4,5,6,1; 4,6,1,5; 4,6,2,4; 4,6,3,3; 4,6,4,2; 4,6,5,1; 5,1,4,6; 5,1,5,5; 5,1,6,4; 5,2,3,6; 5,2,4,5; 5,2,5,4; 5,2,6,3; 5,3,2,6; 5,3,3,5; 5,3,4,4; 5,3,5,3; 5,3,6,2; 5,4,1,6; 5,4,2,5; 5,4,3,4; 5,4,4,3; 5,4,5,2; 5,4,6,1; 5,5,1,5; 5,5,2,4; 5,5,3,3; 5,5,4,2; 5,5,5,1; 5,6,1,4; 5,6,2,3; 5,6,3,2; 5,6,4,1; 6,1,3,6; 6,1,4,5; 6,1,5,4; 6,1,6,3; 6,2,2,6; 6,2,3,5; 6,2,4,4; 6,2,5,3; 6,2,6,2; 6,3,1,6; 6,3,2,5; 6,3,3,4; 6,3,4,3; 6,3,5,2; 6,3,6,1; 6,4,1,5; 6,4,2,4; 6,4,3,3; 6,4,4,2; 6,4,5,1; 6,5,1,4; 6,5,2,3; 6,5,3,2; 6,5,4,1; 6,6,1,3; 6,6,2,2; 6,6,3,1 1,4,6,6; 1,5,5,6; 1,5,6,5; 17 104 0.080247 1,6,4,6; 1,6,5,5; 1,6,6,4; 2,3,6,6; 2,4,5,6; 2,4,6,5; 2,5,4,6; 2,5,5,5; 2,5,6,4; 2,6,3,6; 2,6,4,5; 2,6,5,4; 2,6,6,3; 3,2,6,6; 3,3,5,6; 3,3,6,5; 3,4,4,6; 3,4,5,5; 3,4,6,4; 3,5,3,6; 3,5,4,5; 3,5,5,4; 3,5,6,3; 3,6,2,6; 3,6,3,5; 3,6,4,4; 3,6,5,3; 3,6,6,2; 4,1,6,6; 4,2,5,6; 4,2,6,5; 4,3,4,6; 4,3,5,5; 4,3,6,4; 4,4,3,6; 4,4,4,5; 4,4,5,4; 4,4,6,3; 4,5,2,6; 4,5,3,5; 4,5,4,4; 4,5,5,3; 4,5,6,2; 4,6,1,6; 4,6,2,5; 4,6,3,4; 4,6,4,3; 4,6,5,2; 110

PAGE 114

Allen R. Overman Nonlinear Models 4,6,6,1; 5,1,5,6; 5,1,6,5; 5,2,4,6; 5,2,5,5; 5,2,6,4; 5,3,3,6; 5,3,4,5; 5,3,5,4; 5,3,6,3; 5,4,2,6; 5,4,3,5; 5,4,4,4; 5,4,5,3; 5,4,6,2; 5,5,1,6; 5,5,2,5; 5,5,3,4; 5,5,4,3; 5,5,5,2; 5,5,6,1; 5,6,1,5; 5,6,2,4; 5,6,3,3; 5,6,4,2; 5,6,5,1; 6,1,4,6; 6,1,5,5; 6,1,6,4; 6,2,3,6; 6,2,4,5; 6,2,5,4; 6,2,6,3; 6,3,2,6; 6,3,3,5; 6,3,4,4; 6,3,5,3; 6,3,6,2; 6,4,1,6; 6,4,2,5; 6,4,3,4; 6,4,4,3; 6,4,5,2; 6,4,6,1; 6,5,1,5; 6,5,2,4; 6,5,3,3; 6,5,4,2; 6,5,5,1; 6,6,1,4; 6,6,2,3; 6,6,3,2; 6,6,4,1 6,6,1,5; 6,6,2,4; 6,6,3,3; 18 80 0.061728 6,6,4,2; 6,6,5,1; 6,5,1,6; 6,5,2,5; 6,5,3,4; 6,5,4,3; 6,5,5,2; 6,5,6,1; 6,4,2,6; 6,4,3,5; 6,4,4,4; 6,4,5,3; 6,4,6,2; 6,3,3,6; 6,3,4,5; 6,3,5,4; 6,3,6,3; 6,2,4,6; 6,2,5,5; 6,2,6,4; 6,1,5,6; 6,1,6,5; 5,6,1,6; 5,6,2,5; 5,6,3,4; 5,6,4,3; 5,6,5,2; 5,6,6,1; 5,5,2,6; 5,5,3,5; 5,5,4,4; 5,5,5,3; 5,5,6,2; 5,4,3,6; 5,4,4,5; 5,4,5,4; 5,4,6,3; 5,3,4,6; 5,3,5,5; 5,3,6,4; 5,2,5,6; 5,2,6,5; 5,1,6,6; 4,6,2,6; 4,6,3,5; 4,6,4,4; 4,6,5,3; 4,6,6,2; 4,5,3,6; 4,5,4,5; 4,5,5,4; 4,5,6,3; 4,4,4,6; 4,4,5,5; 4,4,6,4; 4,3,5,6; 4,3,6,5; 4,2,6,6; 3,6,3,6; 3,6,4,5; 3,6,5,4; 3,6,6,3; 3,5,4,6; 3,5,5,5; 3,5,6,4; 3,4,5,6; 3,4,6,5; 3,3,6,6; 2,6,4,6; 2,6,5,5; 2,6,6,4; 2,5,5,6; 2,5,6,5; 2,4,6,6; 1,6,5,6; 1,6,6,5; 1,5,6,6 111

PAGE 115

Allen R. Overman Nonlinear Models 6,6,1,6; 6,6,2,5; 6,6,3,4; 19 56 0.043210 6,6,4,3; 6,6,5,2; 6,6,6,1; 6,5,2,6; 6,5,3,5; 6,5,4,4; 6,5,5,3; 6,5,6,2; 6,4,3,6; 6,4,4,5; 6,4,5,4; 6,4,6,3; 6,3,4,6; 6,3,5,5; 6,3,6,4; 6,2,5,6; 6,2,6,5; 6,1,6,6; 5,6,2,6; 5,6,3,5; 5,6,4,4; 5,6,5,3; 5,6,6,2; 5,5,3,6; 5,5,4,5; 5,5,5,4; 5,5,6,3; 5,4,4,6; 5,4,5,5; 5,4,6,4; 5,3,5,6; 5,3,6,5; 5,2,6,6; 4,6,3,6; 4,6,4,5; 4,6,5,4; 4,6,6,3; 4,5,4,6; 4,5,5,5; 4,5,6,4; 4,4,5,6; 4,4,6,5; 4,3,6,6; 3,6,4,6; 3,6,5,5; 3,6,6,4; 3,5,5,6; 3,5,6,5; 3,4,6,6; 2,6,5,6; 2,6,6,5; 2,5,6,6; 1,6,6,6 6,6,2,6; 6,6,3,5; 6,6,4,4; 20 35 0.027006 6,6,5,3; 6,6,6,2; 6,5,3,6; 6,5,4,5; 6,5,5,4; 6,5,6,3; 6,4,4,6; 6,4,5,5; 6,4,6,4; 6,3,5,6; 6,3,6,5; 6,2,6,6; 5,6,3,6; 5,6,4,5; 5,6,5,4; 5,6,6,3; 5,5,4,6; 5,5,5,5; 5,5,6,4; 5,4,5,6; 5,4,6,5; 5,3,6,6; 4,6,4,6; 4,6,5,5; 4,6,6,4; 4,5,5,6; 4,5,6,5; 4,4,6,6; 3,6,5,6; 3,6,6,5; 3,5,6,6; 2,6,6,6; 6,6,3,6; 6,6,4,5; 6,6,5,4; 21 20 0.015432 6,6,6,3; 6,5,4,6; 6,5,5,5; 6,5,6,4; 6,4,5,6; 6,4,6,5; 6,3,6,6; 5,6,4,6; 5,6,5,5; 5,6,6,4; 5,5,5,6; 5,5,6,5; 5,4,6,6; 4,6,5,6; 4,6,6,5; 4,5,6,6; 3,6,6,6 6,6,4,6; 6,6,5,5; 6,6,6,4; 22 10 0.007716 6,5,5,6; 6,5,6,5; 6,4,6,6; 5,6,5,6; 5,6,6,5; 5,5,6,6; 112

PAGE 116

Allen R. Overman Nonlinear Models 4,6,6,6 6,6,6,5; 6,6,5,6; 6,5,6,6; 23 4 0.003086 5,6,6,6 6,6,6,6 24 1 0.000772 113

PAGE 117

Allen R. Overman Nonlinear Models Table A31. Summ ary of frequency distribution for four dice. S c f F Z Z f 0.000000 4 1 0.000772 0.001510 0.000772 5 4 0.003086 0.003455 0.003858 6 10 0.007716 0.007246 0.011574 .605467 .565423 7 20 0.015432 0.013929 0.027006 .359960 .356700 8 35 0.027006 0.024540 0.054012 .135382 .147977 9 56 0.043210 0.039627 0.097222 .917975 .939254 10 80 0.061728 0.058651 0.158950 .706891 .730531 11 104 0.080247 0.079564 0.239197 .501122 .521808 12 125 0.096451 0.098928 0.335648 .299664 .313085 13 140 0.108025 0.112740 0.443673 .100551 .104362 14 146 0.112654 0.117760 0556327 +0.100551 +0.104362 15 140 0.108025 0.112740 0.664352 +0.299664 +0.313085 16 125 0.096451 0.098928 0.760803 +0.501122 +0.521808 17 104 0.080247 0.079564 0.841050 +0.706891 +0.521808 18 80 0.061728 0.058651 0.902778 +0.917975 +0.730531 19 56 0.043210 0.039627 0.945988 +1.135382 +0.939254 20 35 0.027006 0.024540 0.972994 +1.359960 +1.147977 21 20 0.015432 0.013929 0.988426 +1.605467 +1.565423 22 10 0.007716 0.007246 0.996142 23 4 0.003086 0.003455 0.999228 24 1 0.000772 0.001510 1.000000 114

PAGE 118

Allen R. Overman Nonlinear Models C 1296 = 64 S S F Z 208723.09221.2 2 1 2 12erf 1 r = 0.999770 0000.14,79104.42 2 279104.4 000.14 exp11776.0 2 exp 2 1 S S f f f 07548.100512.0 r = 0.99883 The frequency distribution for a set of dice more closely confor ms to the continuous Gaussian distribution as the number of dice increases. Complexity of computing the discrete distribution increases dramatically with the number of dice, and becomes unwieldy beyond four dice. For five dice the total number of combinations is 65 = 7776! The two dice problem illustrates how well the co ntinuous Gaussian distribution approximates the discrete distribution (triangular). Even though the approximation is not exact, it does bring in an analytic function which we have us ed in the model for plant growth. The peg board offers a simpler model of the fr equency distribution than does a set of dice. Reference: Speyer, E. 1994. Six Roads from Newton: Great Discoveries in Physics. John Wiley & Sons. New York. NY. 115

PAGE 119

Allen R. Overman Nonlinear Models Derivatives for power functi ons (as defined by Cauchy) Function Derivative 0xy 0 11)(00 xxxy 0 0 xx y 0lim0 x y dx dyx 1xy xxxxxxxy 11) ( 1 x x x y 1lim0 x y dx dyx 2xy 2 22 222)(2)(2 )( xxxxxxxxxxxy xx x xxx x y 2 )(22 x x y dx dyx2lim0 3xy ( xy 3 2 233 2 2333)()(33)()(33 ) xxxxxxxxxxxxxx 2 2)(33 xxxx x y 2 03limx x y dx dyx 4 xy 3 22 3 43 22 3444)(4)(64 )(4)(64 )( xxxxxx xxxxxxxxxxxy 2 23)(464 xxxxx x y 3 04limx x y dx dyx nxy 1nnx dx dy 116

PAGE 120

Allen R. Overman Nonlinear Models The reader m ay note that the binomial expansion with n = 0, 1, 2, leads to finite power series. Newton was the first to prove this. Then he showed that for n either negative or a fraction the expansion leads to an infinite power series (Berlinski, 2000, p. 30). This led to Newtons first memoir in 1669 entitled On Analysis by Infinite Series which preceded his development of calculus. It is common to write the general solution as an infinite power series 3 0 4 4 3 3 2 210 i i ixa xaxaxaxaay where ai are the expansion coefficients. Now the coefficients can be evaluated using the derivatives from calculus and the boundary values at x = 0. This leads to the following relationships 00)0( ayxy 1 0a dx dyx 0 2 2 2 2 0 2 2!2 1 12 x xdx yd aa dx yd 0 3 3 3 3 0 3 3!3 1 123 x xdx yd aa dx yd 0 4 4 4 4 0 4 4!4 1 1234 x xdx yd aa dx yd 0 0! 1 x n n n n x n ndx yd n aan dx yd This leads finally to the infinite series n x n n x xx dx yd n x dx yd x dx dy yy0 2 0 2 2 0 0! 1 !2 1 !1 1 117

PAGE 121

Allen R. Overman Nonlinear Models m ay be recognized as the Taylor series, which was discovered in 1715 by Brook Taylor. This approach assumes that the derivatives exist. It also assumes that the series converges to a finite value for any value of x or at least for a limited domain of x Physics usually enters into the process by way of a differential equation w ith initial conditions at x = 0. Berlinski, D. 2000. Newtons Gift: How Sir Isaac Newton Unlocked the System of the World. Simon & Schuster. New York, NY. Example: Consider the first order differential equation k dx dy with the initial condition 0 at0 xyy where k is a constant. Solution: k dx dyx 0 00 2 2 xdx yd 00 0 4 4 0 3 3 x n n x xdx yd dx yd dx yd The solution becomes kxy x dx yd n x dx yd x dx dy yyn x n n x x 0 0 2 0 2 2 0 0! 1 !2 1 !1 1 which is the equation of a straight line. 118

PAGE 122

Allen R. Overman Nonlinear Models 119 Example: Consider the first order differential equation ky dx dy with the initial condition 0 at0 xyy where k is a constant. Solution: 0 0ky dx dyx 0 2 0 0 2 2yk dx dy k dx ydx x 0 3 0 3 3yk dx ydx 0 0yk dx ydn x n n The solution becomes n x n n x xx dx yd n x dx yd x dx dy yy0 2 0 2 2 0 0! 1 !2 1 !1 1 nnxyk n xykxkyy0 2 0 2 00! 1 !2 1 nkx n kxkxy 1 !2 1 !1 1 12 1 0 kxy exp0 where nkx n kxkx kx 1 !2 1 !1 1 1exp2 as defined by Euler.