Jump to content

DOW Experimental Design: Difference between revisions

From mintOC
 
(70 intermediate revisions by 2 users not shown)
Line 1: Line 1:
The '''DOW Experimental Design problem''' models the OED problem for the parameter estimation problem formulated by the DOW Chemical Co. in 1981. The following formulation is taken from "Nonlinear Parameter Estimation:
{{Dimensions
a Case Study Comparison" by L. T. Biegler and J. J. Damiano <span style="color:red">add quote</span>.
|nd        = 1
|nx        = 11
|nw        = 2
|nc        = 4
|nre      = 11
}}


The chemical species are disguised for proprietary reasons and the desired reaction is given by <math>HA + 2BM \rightarrow AB + HBMH</math>, where AB is the desired product. The reactions are described as follows:
The '''DOW Experimental Design problem''' models the OED problem for the parameter estimation problem formulated by the DOW Chemical Co. in 1981. The following formulation is taken from [[#Biegler1986|[1]]].
 
== Chemical background ==
 
The chemical species are disguised for proprietary reasons and the desired reaction is given by <math>HA + 2BM \rightarrow AB + HBMH</math>, where <math>AB</math> is the desired product. The reactions are described as follows:


Slow Kinetic Reactions:
Slow Kinetic Reactions:
Line 21: Line 30:
  MBMH \leftarrow K_1 \rightarrow MBM^- + H^+ \\
  MBMH \leftarrow K_1 \rightarrow MBM^- + H^+ \\
  HA \leftarrow K_2 \rightarrow A^- + H^+ \\
  HA \leftarrow K_2 \rightarrow A^- + H^+ \\
  HABM \leftarrow K_2 \rightarrow ABM^- + H^+  
  HABM \leftarrow K_3 \rightarrow ABM^- + H^+  
  \end{array}
  \end{array}
</math>
</math>
Line 27: Line 36:
In order to devise a model to account for these reactions, it is
In order to devise a model to account for these reactions, it is
first necessary to distinguish between the overall concentration of
first necessary to distinguish between the overall concentration of
a species and the concentration of its neutral form. Overall con-
a species and the concentration of its neutral form. Overall concentrations are defined for three components based on neutral
centrations are defined for three components based on neutral
and ionic species
and ionic species
<p>
<p>
<math>
<math>
  \begin{array}{c}  
  \begin{align}
   \left[HBMH\right] = \left[ (MBMH)_N \right] + \left[ MBM^- \right] \\
   \left[HBMH\right] &= \left[ (MBMH)_N \right] + \left[ MBM^- \right] \\
   \left[HA\right] = \left[ (HA)_N \right] + \left[ A^- \right] \\
   \left[HA\right] &= \left[ (HA)_N \right] + \left[ A^- \right] \\
   \left[HABM\right] = \left[ (HABM)_N \right] + \left[ ABM^- \right]
   \left[HABM\right] &= \left[ (HABM)_N \right] + \left[ ABM^- \right]
  \end{array}  
  \end{align}  
</math>
</math>
</p>
</p>
Here <math>[ \ ]</math> denotes the concentration of the species in <math>gmol/kg</math>.
Here <math>[ \ ]</math> denotes the concentration of the species in <math>mol/kg</math>.
By assuming the rapid acid-base reactions are at equilibrium, the equilibrium constants <math>K_1,K_2,K_3</math> can be defined as
By assuming the rapid acid-base reactions are at equilibrium, the equilibrium constants <math>K_1,K_2,K_3</math> can be defined as
<p>
<p>
<math>
<math>
\begin{array}{c}
\begin{align}
  K_1 = \frac{[MBM^-][H^+]}{[(MBMH)_N]} \\
  K_1 &= \frac{[MBM^-][H^+]}{[(MBMH)_N]} \\
  K_1 = \frac{[A^-][H^+]}{[(HA)_N]} \\
  K_2 &= \frac{[A^-][H^+]}{[(HA)_N]} \\
  K_1 = \frac{[ABM^-][H^+]}{[(HABM)_N]}  
  K_3 &= \frac{[ABM^-][H^+]}{[(HABM)_N]}  
\end{array}
\end{align}
</math>
</math>
</p>
</p>
Line 54: Line 62:
<p>
<p>
<math>
<math>
\begin{array}{cr}
\begin{align}
  \left[MBM^-\right] = \frac{K_1[MBMH]}{K_1 + [H^+]} & \quad (a) \\
  \left[MBM^-\right] &= \frac{K_1[MBMH]}{K_1 + [H^+]} && \quad (a) \\
  \left[A^-\right] = \frac{K_2[HA]}{K_2 + [H^+]} & \quad (b)\\
  \left[A^-\right] &= \frac{K_2[HA]}{K_2 + [H^+]} && \quad (b)\\
  \left[ABM^-\right] = \frac{K_3[HABM]}{K_3 + [H^+]}& \quad (c)
  \left[ABM^-\right] &= \frac{K_3[HABM]}{K_3 + [H^+]} && \quad (c)
\end{array}
\end{align}
</math>
</math>
</p>
</p>
Line 65: Line 73:
<p>
<p>
<math>
<math>
\begin{array}{cr}
\begin{align}
  \frac{d[M^-]}{dt} = -k_1 \left[ M^- \right] \left[ BM \right] + k_{-1} \left[ MBM^- \right] - k_3 \left[ M^- \right]\left[ AB \right] + k_{-1} \left[ ABM^- \right] &  \quad (d)\\
  \frac{d[M^-]}{dt} &= -k_1 \left[ M^- \right] \left[ BM \right] + k_{-1} \left[ MBM^- \right] - k_3 \left[ M^- \right]\left[ AB \right] + k_{-1} \left[ ABM^- \right] &&  \quad (d)\\
  \frac{d[BM]}{dt} = -k_1 \left[ M^- \right] \left[ BM \right] + k_{-1} \left[ MBM^- \right] - k_2 \left[ A^- \right]\left[ BM \right] &  \quad (e)\\  
  \frac{d[BM]}{dt} &= -k_1 \left[ M^- \right] \left[ BM \right] + k_{-1} \left[ MBM^- \right] - k_2 \left[ A^- \right]\left[ BM \right] &&  \quad (e)\\  
  \frac{d[AB]}{dt} = -k_3 \left[ M^- \right] \left[ AB \right] + k_{-3} \left[ ABM^- \right]  &  \quad (f)  
  \frac{d[AB]}{dt} &= -k_3 \left[ M^- \right] \left[ AB \right] + k_{-3} \left[ ABM^- \right]  &&  \quad (f)  
\end{array}
\end{align}
</math>
</math>
</p>
</p>
Line 76: Line 84:
<p>
<p>
<math>
<math>
\begin{array}{cr}
\begin{align}
  \frac{d[MBMH]}{dt} = k_1 \left[ M^- \right] \left[ BM \right] + k_{-1} \left[ MBM^- \right]  &  \quad (g)\\
  \frac{d[MBMH]}{dt} &= k_1 \left[ M^- \right] \left[ BM \right] - k_{-1} \left[ MBM^- \right]  &&  \quad (g)\\
  \frac{d[HA]}{dt} = k_2 \left[ A^- \right] \left[ BM \right] &  \quad (h)\\
  \frac{d[HA]}{dt} &= k_2 \left[ A^- \right] \left[ BM \right] &&  \quad (h)\\
  \frac{d[HABM]}{dt} = k_2 \left[ A^- \right] \left[ BM \right] + k_3 \left[ M^- \right] \left[ AB \right] - k_{-3} \left[ ABM^- \right]&  \quad (i)
  \frac{d[HABM]}{dt} &= k_2 \left[ A^- \right] \left[ BM \right] + k_3 \left[ M^- \right] \left[ AB \right] - k_{-3} \left[ ABM^- \right] &&  \quad (i)
\end{array}
\end{align}
</math>
</math>
</p>
</p>
An electroneutrality constraint gives the hydrogen ion con-
An electroneutrality constraint gives the hydrogen ion concentration <math>\left[ H^+ \right]</math> as
centration <math>\left[ H^+ \right]</math> as
<p>
<p>
  <math>
  <math>
Line 95: Line 102:
  <math> k_3 = k_1, \quad k_{-3} = \frac{1}{2} k_{-1} </math>
  <math> k_3 = k_1, \quad k_{-3} = \frac{1}{2} k_{-1} </math>
</p>
</p>
With these assumptions, the three rate constants <math>k_1,k_2</math> and <math>k_3</math> must be estimated. Each of these can be fitted with two adjustable model parameters, assuming an Arrhenius temperature dependence. That is
With these assumptions, the three rate constants <math>k_1,k_2</math> and <math>k_{-1}</math> must be estimated. Each of these can be fitted with two adjustable model parameters, assuming an Arrhenius temperature dependence. That is
<p>
<p>
  <math>  
  <math>  
Line 103: Line 110:
  </math>
  </math>
</p>
</p>
Here <math>R</math> is the gas constant, <math>T</math> is reaction temperature in Kelvins and the parameters <math>\alpha_i,E_i</math> represent the pre-exponential factor and activation energy, respectively, for the appropriate rate constant.
Here <math>R \approx 1.98720425864083 \ cal/(K \cdot mol) </math> is the gas constant and <math>T</math> is the reaction temperature in Kelvins. The parameters <math>\alpha_i</math>, given in <math>mol/( kg \cdot h)</math>, represent the pre-exponential factors and the <math>E_i</math>, with unit <math>cal/mol</math>, are the activation energies.


== Mathematical formulation ==
== Mathematical formulation ==


The chemical processes <math>(a)-(j)</math> can be expressed mathematically as six differential and four algebraic equations:  
The chemical processes <math>(a)-(j)</math> can be expressed mathematically as six differential equations and four algebraic equations:  
<p>
<p>
  <math>
  <math>
   \begin{array}{lr}
   \begin{align}
   \frac{d y_1}{dt} = -k_2 y_8 y_2 & \quad (1),(h) \\
   \dot{y}_1 &= -k_2 y_8 y_2 && \quad (1),(h) \\
   \frac{d y_2}{dt} = -k_1 y_6 y_2 + k_{-1} y_{10} - k_2 y_8 y_2 & \quad (2),(e) \\
   \dot{y}_2 &= -k_1 y_6 y_2 + k_{-1} y_{10} - k_2 y_8 y_2 && \quad (2),(e) \\
   \frac{d y_3}{dt} = -k_2 y_8 y_2 + k_1 y_6 y_4 - \frac{1}{2} k_{-1} y_9 & \quad (3),(i) \\
   \dot{y}_3 &= k_2 y_8 y_2 + k_1 y_6 y_4 - \frac{1}{2} k_{-1} y_9 && \quad (3),(i) \\
   \frac{d y_4}{dt} = -k_1 y_6 y_4  + \frac{1}{2} k_{-1} y_9 & \quad (4),(f) \\
   \dot{y}_4 &= -k_1 y_6 y_4  + \frac{1}{2} k_{-1} y_9 && \quad (4),(f) \\
   \frac{d y_5}{dt} = -k_1 y_6 y_2 + k_{-1} y_{10} & \quad (5),(g) \\
   \dot{y}_5 &= k_1 y_6 y_2 - k_{-1} y_{10} && \quad (5),(g) \\
   \frac{d y_6}{dt} = -k_1 (y_6 y_2 + y_6 y_4) + k_{-1} (y_{10} + \frac{1}{2} y_9) & \quad (6),(d) \\
   \dot{y}_6 &= -k_1 (y_6 y_2 + y_6 y_4) + k_{-1} (y_{10} + \frac{1}{2} y_9) && \quad (6),(d) \\
   y_7 = -\left[ Q^+ \right] + y_6 + y_8 + y_9 + y_{10} & \quad (7),(j)\\
   y_7 &= -\left[ Q^+ \right] + y_6 + y_8 + y_9 + y_{10} && \quad (7),(j)\\
   y_8 = \frac{\theta_8 y_1}{\theta_8 + y_7} & \quad (8),(b)\\
   y_8 &= \frac{\theta_8 y_1}{\theta_8 + y_7} && \quad (8),(b)\\
   y_9 = \frac{\theta_9 y_3}{\theta_9 + y_7} & \quad (9),(c)\\
   y_9 &= \frac{\theta_9 y_3}{\theta_9 + y_7} && \quad (9),(c)\\
   y_{10} = \frac{\theta_7 y_5}{\theta_7 + y_7} & \quad (10),(a)
   y_{10} &= \frac{\theta_7 y_5}{\theta_7 + y_7} && \quad (10),(a)
   \end{array}
   \end{align}
  </math>
  </math>
</p>
</p>
Here the letter stands for the corresponding chemical process and the quantity <math>\left[Q^+\right]</math> is a concentration which is assumed to be constant during the reactions.
Here the letters in parentheses stand for the corresponding chemical process and the quantity <math>\left[Q^+\right]=0.0131</math> is a constant during the reaction.
The nine parameters form the vector  
The nine parameters form the vector  
<p>
<p>
Line 139: Line 146:


Let <math>f_k(\cdot)</math> denote the right hand side of equation <math>(k)</math> for <math>k=1,\ldots,6</math>.  
Let <math>f_k(\cdot)</math> denote the right hand side of equation <math>(k)</math> for <math>k=1,\ldots,6</math>.  
We reformulate the last four algebraic equations as differential ones:
 
The right hand sides of <math>(1)-(10)</math> are summarized as the vector-valued function <math>f</math>.
Moreover, let
<p>
<math>
  \begin{align}
  f_{y,m,n}(\cdot) &= \frac{\partial f_m(\cdot)}{\partial y_n}, \quad m,n \in \{1,\ldots,10\} \\
  f_{\theta,m,n}(\cdot) &= \frac{\partial f_m(\cdot)}{\partial \theta_n}, \quad m \in \{1,\ldots,10\}; \ n\in \{1,\ldots,9\}
  \end{align}
</math>
</p>
 
== Parameters ==
 
The initial parameter estimates are:
<table style="border-collapse: collapse;" border="1.">
            <tr>
                <td style="text-align: center; padding:5pt"> <math> \alpha_1 </math> </td>
                <td style="text-align: center; padding:5pt"> <math>2.0 \cdot 10^{13} \ mol / (kg \cdot h)</math> </td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\alpha_2</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{13} \ mol / (kg \cdot h)</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\alpha_{-1}</math></td>
                <td style="text-align: center; padding:5pt"><math>4.3 \cdot 10^{15} \ mol / (kg \cdot h)</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_1</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4} \ cal / mol</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_2</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4} \ cal / mol </math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_{-1}</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4} \ cal / mol</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>K_1</math></td>
                <td style="text-align: center; padding:5pt"><math>1.0\cdot 10^{-17} \ mol / kg</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>K_2</math></td>
                <td style="text-align: center; padding:5pt"><math>1.0\cdot 10^{-11} \ mol / kg</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>K_3</math></td>
                <td style="text-align: center; padding:5pt"><math>1.0\cdot 10^{-17} \ mol / kg</math></td>
            </tr>
    </table>
Note that for the calculations all temperatures given in<math>{\ }^{\circ}C</math> have to be rescaled to <math>K</math> by adding <math>273.15</math>.
 
There are three datasets for different temperatures <math>T</math>, with corresponding starting values
<table style="border-collapse: collapse;" border="1.">
            <tr style="border-bottom: 2pt solid black;">
                <th style="border-right: 2pt solid black;"></th>
                <th style="text-align: center; padding:5pt"> <math>40^\circ C </math></th>
                <th style="text-align: center; padding:5pt"> <math>67^\circ C </math></th>
                <th style="text-align: center; padding:5pt"> <math>100^\circ C </math></th>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt; border-right: 2pt solid black;"> <math> y_1(0) </math> </td>
                <td style="text-align: center; padding:5pt"> <math> 1.7066 </math> </td>
                <td style="text-align: center; padding:5pt"> <math> 1.6749 </math> </td>
                <td style="text-align: center; padding:5pt"> <math> 1.5608 </math> </td>           
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt; border-right: 2pt solid black"><math>y_2(0)</math></td>
                <td style="text-align: center; padding:5pt"><math>8.32</math></td>
                <td style="text-align: center; padding:5pt"><math>8.2262</math></td>
                <td style="text-align: center; padding:5pt"><math>8.3546</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt; border-right: 2pt solid black"><math>y_3(0)</math></td>
                <td style="text-align: center; padding:5pt"><math>0.01</math></td>
                <td style="text-align: center; padding:5pt"><math>0.0104</math></td>
                <td style="text-align: center; padding:5pt"><math>0.0082</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt; border-right: 2pt solid black"><math>y_4(0)</math></td>
                <td style="text-align: center; padding:5pt"><math>0</math></td>
                <td style="text-align: center; padding:5pt"><math>0.0017</math></td>
                <td style="text-align: center; padding:5pt"><math>0.0086</math></td>
            </tr>
    </table>
<!--
<p>
<p>
  <math>
  <math>
   \begin{array}{l r}
   \begin{array}{lccc}
   \frac{d y_7}{dt} = f_7 = f_6 + f_8 + f_9 + f_{10} & \quad (7') \\
   & 40^\circ C & 67^\circ C & 100^\circ C \\
  \frac{d y_8}{dt} = f_8 = \frac{\theta_8 f_1 \cdot (\theta_8 + f_7) - \theta_8 y_1 f_7}{(\theta_8 + y_7)^2} & \quad (8') \\
  y_1(0) = & 1.7066 & 1.6749 & 1.5608 \\
  \frac{d y_9}{dt} = f_9 = \frac{\theta_9 f_3 \cdot (\theta_9 + f_7) - \theta_9 y_3 f_7}{(\theta_9 + y_7)^2}  & \quad (9')\\
  y_2(0) = & 8.32 & 8.2262 & 8.3546 \\
  \frac{d y_{10}}{dt} = f_{10} = \frac{\theta_7 f_5 \cdot (\theta_7 + f_7) - \theta_7 y_5 f_7}{(\theta_7 + y_7)^2} & \quad (10')
  y_3(0) = & 0.01 & 0.0104 & 0.0082 \\
  y_4 (0) = & 0 & 0.0017 & 0.0086
   \end{array}
   \end{array}
  </math>
  </math>
</p>
</p>
The right hand sides of <math>(1)-(6)</math> and <math>(7')-(9')</math> are summarized as the vector-valued function <math>f</math>.
-->
Moreover, let
The initial model conditions in addition to those given in the
data sets are:
<p>
<p>
  <math>
  <math>
   \begin{array}{l}
   \begin{array}{l}
  f_{y,m,n}(\cdot) = \frac{\partial f_m(\cdot)}{\partial y_n}, \quad m,n \in \{1,\ldots,6\} \\
  y_5 = 0 \\
  f_{\theta,m,n}(\cdot) = \frac{\partial f_m(\cdot)}{\partial y_n}
  y_6 = [Q^+] \\
  y_7 = \frac{1}{2} \cdot \left( -K_2 + \sqrt{K_2^2 + 4K_2 y_1(0)} \right) \\
  y_8 = y_7 \\
  y_9 = 0 \\
  y_{10} = 0
   \end{array}
   \end{array}
  </math>
  </math>
</p>  
</p>
 
To reduce the intercorrelation between the parameters in the rate constants, we apply the following reparametrization (cf. [[#Kristensen2004|[4]]].):
<p>
<math>
  \begin{align}
  k_i &= \alpha_i \cdot \exp \left( - \frac{E_i}{RT} \right) \\
      &= k_{0,i} \cdot \exp \left( - \frac{E_i}{R} \cdot \left( \frac{1}{T} - \frac{1}{T_0} \right) \right), \quad i=1,2,-1
\end{align}
</math>
</p>
in which <math>k_{0,i} = \alpha_i \cdot \exp\left( - \frac{E_i}{RT_0} \right)</math>. The reference temperature in <math>T_0</math> is chosen as the average over all performed experiments, i.e., <math>T_0 = 69^\circ C</math>. Additionally, we add a logarithmic transformation, which gives rise to the following transformed starting values:
 
<table style="border-collapse: collapse;" border="1.">
            <tr>
                <td style="text-align: center; padding:5pt"> <math> \ln k_{0,1} </math> </td>
                <td style="text-align: center; padding:5pt"> <math>1.194</math> </td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\ln k_{0,2}</math></td>
                <td style="text-align: center; padding:5pt"><math>1.194</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\ln k_{0,-1}</math></td>
                <td style="text-align: center; padding:5pt"><math>6.565</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_1</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4}</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_2</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4} </math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>E_{-1}</math></td>
                <td style="text-align: center; padding:5pt"><math>2.0 \cdot 10^{4}</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\ln K_1</math></td>
                <td style="text-align: center; padding:5pt"><math>-34.54</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\ln K_2</math></td>
                <td style="text-align: center; padding:5pt"><math>-25.33</math></td>
            </tr>
            <tr>
                <td style="text-align: center; padding:5pt"><math>\ln K_3</math></td>
                <td style="text-align: center; padding:5pt"><math>-39.14</math></td>
            </tr>
    </table>
 
== Optimal Experimental Design Problem ==
To be specified.


<!--
<!--
Line 165: Line 319:
<p>
<p>
  <math>
  <math>
   \begin{array}{c}
   \begin{align}
   k_{i,\alpha} := \frac{\partial k_i}{\partial \alpha_i} = \exp(-E_i/(RT)) \\
   k_{i,\alpha} &:= \frac{\partial k_i}{\partial \alpha_i} = \exp(-E_i/(RT)) \\
   k_{i,E} := \frac{\partial k_i}{\partial E_i} = -\frac{\alpha_i}{RT} \exp(-E_i/(RT))
   k_{i,E} &:= \frac{\partial k_i}{\partial E_i} = -\frac{\alpha_i}{RT} \exp(-E_i/(RT))
   \end{array}
   \end{align}
  </math>
  </math>
</p>
</p>
Line 205: Line 359:
<p>
<p>
  <math>
  <math>
\begin{array}{l}
\begin{align}
   f_y(\cdot) \in \mathbb{R}^{10 \times 10} \quad \text{ with } (f_y)_{i,j} = f_{y,i,j}, \\
   f_y(\cdot) &\in \mathbb{R}^{10 \times 10} \quad &&\text{ with } (f_y)_{i,j} = f_{y,i,j}, \\
   f_\theta(\cdot) \in \mathbb{R}^{10 \times 9} \quad \text{ with } (f_\theta)_{i,j} = \frac{\partial f_i}{\partial \theta_j}
   f_\theta(\cdot) &\in \mathbb{R}^{10 \times 9} \quad &&\text{ with } (f_\theta)_{i,j} = f_{\theta,i,j}
  \end{array}
  \end{align}
  </math>
  </math>
</p>  
</p>  
Line 218: Line 372:
</p>
</p>


Now we formulate the OED problem as described in (Optimal Experimental Design for Universal Differential Equations <span style="color:red">add quote</span>)
Now we formulate the OED problem as described in [[#OEDUDE | [2]]].
<p>
<p>
<math>
<math>
Line 243: Line 397:
are the (binary) sampling decisions, where <math> w_i (t) = 1 </math> denotes the decision to perform a measurement at
are the (binary) sampling decisions, where <math> w_i (t) = 1 </math> denotes the decision to perform a measurement at
time <math>t</math>.
time <math>t</math>.
== Parameters ==
The initial parameter estimates are:
<p>
<math>
  \begin{array}{c}
  \alpha_1 = 2.0 \times 10^{13} \\
  E_1 = 2.0 \times 10^{4} \\
  \alpha_2 = 2.0 \times 10^{13} \\
  E_2 = 2.0 \times 10^{4} \\
  \alpha_{-1} = 4.3 \times 10^{15} \\
  E_{-1} = 2.0 \times 10^{4} \\
  K_1 = 1.0 \times 10^{-17} \\
  K_2 = 1.0 \times 10^{-11} \\
  K_3 = 1.0 \times 10^{-17} \\
  \end{array}
</math>
</p>
The initial model conditions in addition to those given in the
data sets are:
<p>
<math>
  \begin{array}{l}
  y_5 = 0 \\
  y_6 = 0.0131 \\
  y_7 = \frac{1}{2} \cdot \left( -K_2 + \sqrt{K_2^2 + 4K_2 y_1(0)} \right) \\
  y_8 = y_7 \\
  y_9 = 0 \\
  y_{10} = 0
  \end{array}
</math>
</p>


== Miscellaneous and Further Reading ==
== Miscellaneous and Further Reading ==
The Lotka Volterra fishing problem was introduced by Sebastian Sager in a proceedings paper <bib id="Sager2006" /> and revisited in his PhD thesis <bib id="Sager2005" />. These are also the references to look for more details. The experimental design problem has been described in the habilitation thesis of Sager, <bib id="Sager2011d" />.
To be specified.


== References ==
== References ==
<biblist />
<span id="Biegler1986">[1]</span> "Nonlinear Parameter Estimation: a Case Study Comparison" by L. T. Biegler and J. J. Damiano  <br>
<span id="OEDUDE">[2]</span> "Optimal Experimental Design for Universal Differential Equations" by C. Plate, C.J. Martensen and S. Sager <br>
<span id="Stortelder1998">[3]</span> "Parameter estimation in nonlinear systems" by W.J.H. Stortelder <br>
<span id="Kristensen2004">[4]</span> "Parameter Estimation in Nonlinear Dynamical Systems" by Morten Rode Kristensen <br>


<!--List of all categories this page is part of. List characterization of solution behavior, model properties, ore presence of implementation details (e.g., AMPL for AMPL model) here -->
<!--List of all categories this page is part of. List characterization of solution behavior, model properties, ore presence of implementation details (e.g., AMPL for AMPL model) here -->

Latest revision as of 06:19, 26 August 2025

DOW Experimental Design
State dimension: 1
Differential states: 11
Discrete control functions: 2
Path constraints: 4
Interior point equalities: 11


The DOW Experimental Design problem models the OED problem for the parameter estimation problem formulated by the DOW Chemical Co. in 1981. The following formulation is taken from [1].

Chemical background

The chemical species are disguised for proprietary reasons and the desired reaction is given by HA+2BMAB+HBMH, where AB is the desired product. The reactions are described as follows:

Slow Kinetic Reactions:

M+BMk1k1MBMA+BMk2ABMM+ABk3k3ABM

Acid-Base Reactions:

MBMHK1MBM+H+HAK2A+H+HABMK3ABM+H+

In order to devise a model to account for these reactions, it is first necessary to distinguish between the overall concentration of a species and the concentration of its neutral form. Overall concentrations are defined for three components based on neutral and ionic species

[HBMH]=[(MBMH)N]+[MBM][HA]=[(HA)N]+[A][HABM]=[(HABM)N]+[ABM]

Here [ ] denotes the concentration of the species in mol/kg. By assuming the rapid acid-base reactions are at equilibrium, the equilibrium constants K1,K2,K3 can be defined as

K1=[MBM][H+][(MBMH)N]K2=[A][H+][(HA)N]K3=[ABM][H+][(HABM)N]

The anionic species may then be represented by

[MBM]=K1[MBMH]K1+[H+](a)[A]=K2[HA]K2+[H+](b)[ABM]=K3[HABM]K3+[H+](c)

Material balance equations for the three reactants in the slow kinetic reactions yield:

d[M]dt=k1[M][BM]+k1[MBM]k3[M][AB]+k1[ABM](d)d[BM]dt=k1[M][BM]+k1[MBM]k2[A][BM](e)d[AB]dt=k3[M][AB]+k3[ABM](f)

From stoichiometry, rate expressions can also be written for the total species:

d[MBMH]dt=k1[M][BM]k1[MBM](g)d[HA]dt=k2[A][BM](h)d[HABM]dt=k2[A][BM]+k3[M][AB]k3[ABM](i)

An electroneutrality constraint gives the hydrogen ion concentration [H+] as

[H+]+[Q+]=[M]+[MBM]+[A]+[ABM](j)

Based on similarities of reacting species, we assume

k3=k1,k3=12k1

With these assumptions, the three rate constants k1,k2 and k1 must be estimated. Each of these can be fitted with two adjustable model parameters, assuming an Arrhenius temperature dependence. That is

ki=αiexp(Ei/(RT)),i{1,1,2}

Here R1.98720425864083 cal/(Kmol) is the gas constant and T is the reaction temperature in Kelvins. The parameters αi, given in mol/(kgh), represent the pre-exponential factors and the Ei, with unit cal/mol, are the activation energies.

Mathematical formulation

The chemical processes (a)(j) can be expressed mathematically as six differential equations and four algebraic equations:

y˙1=k2y8y2(1),(h)y˙2=k1y6y2+k1y10k2y8y2(2),(e)y˙3=k2y8y2+k1y6y412k1y9(3),(i)y˙4=k1y6y4+12k1y9(4),(f)y˙5=k1y6y2k1y10(5),(g)y˙6=k1(y6y2+y6y4)+k1(y10+12y9)(6),(d)y7=[Q+]+y6+y8+y9+y10(7),(j)y8=θ8y1θ8+y7(8),(b)y9=θ9y3θ9+y7(9),(c)y10=θ7y5θ7+y7(10),(a)

Here the letters in parentheses stand for the corresponding chemical process and the quantity [Q+]=0.0131 is a constant during the reaction. The nine parameters form the vector

θ=(α1,E1,α2,E2,α1,E1,K1,K2,K3)

The predicted concentrations form the vector

y=(HA,BM,HABM,AB,MBMH,M,H+,A,ABM,MBM)

Let fk() denote the right hand side of equation (k) for k=1,,6.

The right hand sides of (1)(10) are summarized as the vector-valued function f. Moreover, let

fy,m,n()=fm()yn,m,n{1,,10}fθ,m,n()=fm()θn,m{1,,10}; n{1,,9}

Parameters

The initial parameter estimates are:

α1 2.01013 mol/(kgh)
α2 2.01013 mol/(kgh)
α1 4.31015 mol/(kgh)
E1 2.0104 cal/mol
E2 2.0104 cal/mol
E1 2.0104 cal/mol
K1 1.01017 mol/kg
K2 1.01011 mol/kg
K3 1.01017 mol/kg

Note that for the calculations all temperatures given in C have to be rescaled to K by adding 273.15.

There are three datasets for different temperatures T, with corresponding starting values

40C 67C 100C
y1(0) 1.7066 1.6749 1.5608
y2(0) 8.32 8.2262 8.3546
y3(0) 0.01 0.0104 0.0082
y4(0) 0 0.0017 0.0086

The initial model conditions in addition to those given in the data sets are:

y5=0y6=[Q+]y7=12(K2+K22+4K2y1(0))y8=y7y9=0y10=0

To reduce the intercorrelation between the parameters in the rate constants, we apply the following reparametrization (cf. [4].):

ki=αiexp(EiRT)=k0,iexp(EiR(1T1T0)),i=1,2,1

in which k0,i=αiexp(EiRT0). The reference temperature in T0 is chosen as the average over all performed experiments, i.e., T0=69C. Additionally, we add a logarithmic transformation, which gives rise to the following transformed starting values:

lnk0,1 1.194
lnk0,2 1.194
lnk0,1 6.565
E1 2.0104
E2 2.0104
E1 2.0104
lnK1 34.54
lnK2 25.33
lnK3 39.14

Optimal Experimental Design Problem

To be specified.

We are interested in when to measure (with an upper bound Mi on the measuring time for each observable). We define

fy()10×10 with (fy)i,j=fy,i,j,fθ()10×9 with (fθ)i,j=fθ,i,j

In this approach, we add the so-called sensitivities G=dy/dθ. For the differential equations this means

G˙(t)=fy(y(t),θ)G(t)+fθ(y(t),θ),G(0)=y(0)θ

Now we formulate the OED problem as described in [2].

miny,G,F,z,wtrace(F1(tf))subject toy˙(t)=f(y(t),θ)G˙(t)=fy(y(t),θ)G(t)+fθ(y(t),θ)F˙(t)=i=1nowi(t)(hyi(y(t))G(t))T(hyi(y(t))G(t))z˙(t)=w(t),y(0)=y0G(0)=y(0)θF(0)=0,z(0)=0w(t)𝒲zi(tf)Mi

Here h:10no is the observed function. The evolution of the symmetric matrix F:[0,tf]9×9 is given by the weighted sum of observability Gramians hyi(y(t))G(t), i=1,,no for each observed function of states. The weights wi(t){0,1}, i=1,,no are the (binary) sampling decisions, where wi(t)=1 denotes the decision to perform a measurement at time t.

Miscellaneous and Further Reading

To be specified.

References

[1] "Nonlinear Parameter Estimation: a Case Study Comparison" by L. T. Biegler and J. J. Damiano
[2] "Optimal Experimental Design for Universal Differential Equations" by C. Plate, C.J. Martensen and S. Sager
[3] "Parameter estimation in nonlinear systems" by W.J.H. Stortelder
[4] "Parameter Estimation in Nonlinear Dynamical Systems" by Morten Rode Kristensen