10.34, Numerical Methods Applied to Chemical Engineering Professor William H. Green Lecture #25: Conclude Models vs. Data Parameter Estimation 1) Model definition/Formulation: choosing θ 2) Compile/Assess what you already knew before adjusting θ a. estimate parameters, error bars b. initial guess θ 3) Adjust θ a. Determine if Model is Consistent with Data: if inconsistent, you have learned something important b. θbestfit at θlocalminima of χ2(θ) 4) Refine p(θ): narrow range for the parameters a. summarize what we have learned 4-STEP PROCESS IS ALSO CALLED: “LEAST-SQUARES FITTING” 1) Repeat measurement “i” Nreplicates times Yi(j) j = 1,Nreplicates N rep Yi data = χ = 2 ∑ Yi j =1 N rep N data ⎛ ∑ i =1 N rep ( j) Y data − Yi model (θ ) ⎞ ⎜ i ⎟ ⎜ ⎟ σ i ⎝ ⎠ σi = ∑ (Yi( j ) − Yi data ) 2 j =1 N rep − 1 2 Quantitative Definition of “Consistent” calc χ2(θ; Ydata) Prob(measure Ydata with χ2 ≥ χ2expt) Î ∫ ∞ 2 xexpt 2 ⎛ v xexpt Γ⎜ , ⎜2 2 t e dt = ⎝ N ⎛v⎞ ⎛v⎞ 2 2 Γ⎜ ⎟ Γ⎜ ⎟ ⎝2⎠ ⎝2⎠ v t −1 − 2 2 {if small, unlikely that our model is right} ⎞ ⎟ ⎟ ⎠ ν = Ndata - Nparams_adjusted GAMMA FUNCTION Cite as: William Green, Jr., course materials for 10.34 Numerical Methods Applied to Chemical Engineering, Fall 2006. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY]. ⎛v χ2 Γ⎜⎜ , ⎝2 2 If ⎛v⎞ Γ⎜ ⎟ ⎝2⎠ ⎞ ⎟⎟ ⎠ < 0.01 Æ very unlikely model is consistent with data χ2(θ) < χ2max for consistency If you have more data, you have more confidence. Need lots more data than number of θ’s. If >> : inconsistent If = : consistent 2 P(χ ) χ2expt Ndata [χ2] Figure 1. Chi-squared distribution tests. MatLab: chi2cdf.m inconsistent ∆H1 ⎛ ⎛ k1 ⎞ ⎞ ⎟⎟, Tdata ⎟⎟ ⎝ ⎝ ΔH 1 ⎠ ⎠ x 2 ⎜ χ exp ⎜ ⎜⎜ ∆H consistent k1 k θthat is inconsistent with data neglect Figure 2. An example of two parameter fitting. χ2(θ) = χ2(θbestfit) + ½(θ – θbest)T*H(θbest)*(θ – θbest) + O(∆θ3) H ≈ JTJ Jin = ∂Yi model ∂θ n θ best 10.34, Numerical Methods Applied to Chemical Engineering Prof. William Green Lecture 25 Page 2 of 3 Cite as: William Green, Jr., course materials for 10.34 Numerical Methods Applied to Chemical Engineering, Fall 2006. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY]. χ2 = χ2best + 2 χ2 = χ2best + 1 ∆H δ k χ2(θ) = χ2m=x \Figure 3. Contours around best fit. People want these contours to be circles Range of parameters that are acceptable: ∆H ± δ(∆H) k = kbest + δk Covariance Matrix Equilibrium Concentration kinetics experiments give the ellipses ∆Hdist. P(θ) Experiment ∆H gives the values of k the horizontal Figure 4. Each experiment tells you about different “cuts” or ellipses and cuts. where they all intersect is the answer. BAYESIAN: store p(θ) STORE ALL THE DATA: Thermochemistry Active Tables, PrIMe 10.34, Numerical Methods Applied to Chemical Engineering Prof. William Green Lecture 25 Page 3 of 3 Cite as: William Green, Jr., course materials for 10.34 Numerical Methods Applied to Chemical Engineering, Fall 2006. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].