Team:Evry/LogisticFunctions

From 2013.igem.org

(Difference between revisions)
 
(16 intermediate revisions not shown)
Line 5: Line 5:
<div id="mainTextcontainer">
<div id="mainTextcontainer">
-
<h1>Logistic functions :</h1>
+
<h1>Logistic functions</h1>
<p>When we started to model biological behaviors, we realised very soon that we were going to need a function that simulates a non-exponential evolution, that would include a simple speed control and a maximum value. A smooth step function.</p>
<p>When we started to model biological behaviors, we realised very soon that we were going to need a function that simulates a non-exponential evolution, that would include a simple speed control and a maximum value. A smooth step function.</p>
Line 18: Line 18:
<li><b>Q</b> : Magnitude.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>The limit of g as x approaches infinity is Q.</i></li>
<li><b>Q</b> : Magnitude.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>The limit of g as x approaches infinity is Q.</i></li>
<li><b>d</b> : Threshold.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>The value of x from which we consider the start of the phenomenon.</i></li>
<li><b>d</b> : Threshold.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>The value of x from which we consider the start of the phenomenon.</i></li>
-
<li><b>p</b> : Precision.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>g(d)=Q*p Since the function never reaches 0 nor Q, we have to set an approximation for 0 or Q.</i></li>
+
<li><b>p</b> : Precision.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i><img src="https://static.igem.org/mediawiki/2013/3/3b/Gdp.jpg"/> Since the function never reaches 0 nor Q, we have to set an approximation for 0 or Q.</i></li>
<li><b>k</b> : Efficiency.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>This parameter influences the length of the phenomenon.</i></li>
<li><b>k</b> : Efficiency.<br/>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;<i>This parameter influences the length of the phenomenon.</i></li>
</ul>
</ul>
-
<p><img src="https://static.igem.org/mediawiki/2013/0/05/CourbeLogistique.png"/></p>
+
<p><img width="70%" src="https://static.igem.org/mediawiki/2013/0/05/CourbeLogistique.png"/></p>
-
<div id="citation_box">
+
<h2>Differential form:</h2>
-
<p id="references">References:</p>
+
<p>Let the following be a Cauchy problem:<br/>
-
<ol>
+
<img src="https://static.igem.org/mediawiki/2013/b/b2/Logistic_cauchy.jpg"/>
-
  <li></li>
+
</p>
-
</ol>
+
 
-
</div>
+
<p>The solution of this Cauchy problem is as below:<br/>
 +
<img src="https://static.igem.org/mediawiki/2013/3/3d/Solution_cauchy.jpg"/>
 +
</p>
 +
 
 +
<p>Here is our logistic function. Yet, differential equations are not always time-related.<br/>
 +
Let x be a temporal function, and y be a x-related logistic function. In order to integrate y into a temporal ODE, we need to write it differently:<br/>
 +
<img src="https://static.igem.org/mediawiki/2013/4/40/Logistic_calcul1.jpg"/><br/>
 +
 
 +
But this equation can't be integrated in a temporal system like other equations. Because y depend on x. In our model, x is a state variable of the system. To implement this equation, we solve it before the entire system.
 +
 
 +
<!--
 +
If <img src="https://static.igem.org/mediawiki/2013/a/a0/Txdet.jpg"/> is a continuous real function, then:<br/>
 +
<img src="https://static.igem.org/mediawiki/2013/5/5e/Yxtegalyt.jpg"/><br/>
 +
Finally,<br/>
 +
<img src="https://static.igem.org/mediawiki/2013/b/bd/Logistic_calcul2.jpg"/>
 +
-->
 +
</p>
</div>
</div>

Latest revision as of 02:42, 29 October 2013

Iron coli project

Logistic functions

When we started to model biological behaviors, we realised very soon that we were going to need a function that simulates a non-exponential evolution, that would include a simple speed control and a maximum value. A smooth step function.

Such functions, named logistic functions were introduced around 1840 by M. Verhulst.

These functions looked perfect, but we needed more control : we needed to set a starting value and a precision.

Parameters:

  • Q : Magnitude.
             The limit of g as x approaches infinity is Q.
  • d : Threshold.
             The value of x from which we consider the start of the phenomenon.
  • p : Precision.
           Since the function never reaches 0 nor Q, we have to set an approximation for 0 or Q.
  • k : Efficiency.
             This parameter influences the length of the phenomenon.

Differential form:

Let the following be a Cauchy problem:

The solution of this Cauchy problem is as below:

Here is our logistic function. Yet, differential equations are not always time-related.
Let x be a temporal function, and y be a x-related logistic function. In order to integrate y into a temporal ODE, we need to write it differently:

But this equation can't be integrated in a temporal system like other equations. Because y depend on x. In our model, x is a state variable of the system. To implement this equation, we solve it before the entire system.