If the model is linear in coefficients, for example, the temperature dependence of specific heat is traditionally modeled as c p = a + bT + cT 2 + dT −1 , and then linear regression can be used to...

If the model is linear in coefficients, for example, the temperature dependence of specific heat is traditionally modeled as c_p
= a + bT + cT²
+ dT
⁻¹, and then linear regression can be used to determine model coefficient values. First, measure c_p
at a variety of temperatures to generate the data. Then set the derivative of the SSD with respect to each model coefficient to zero, and you have the four normal equations. Linear algebra will provide the exact values of the coefficients for that particular experimental data set. Alternately, if you choose a numerical optimization procedure (Newton–Raphson, ISD, LM, HJ, LF, etc.), it will approach the optimum DV values and stop in the proximity of the true DV∗ values when the convergence criterion is satisfied. Which method is better? Briefly state three (or more) issues.