.Modelica.Media.Common.smoothStep

Information

This function is used to approximate the equation

y = if x > 0 then y1 else y2;

by a smooth characteristic, so that the expression is continuous and differentiable:

y = smooth(1, if x >  x_small then y1 else
              if x < -x_small then y2 else f(y1, y2));

In the region -x_small < x < x_small a 2nd order polynomial is used for a smooth transition from y1 to y2.

If mass fractions X[:] are approximated with this function then this can be performed for all nX mass fractions, instead of applying it for nX-1 mass fractions and computing the last one by the mass fraction constraint sum(X)=1. The reason is that the approximating function has the property that sum(X) = 1, provided sum(X_a) = sum(X_b) = 1 (and y1=X_a[i], y2=X_b[i]). This can be shown by evaluating the approximating function in the abs(x) < x_small region (otherwise X is either X_a or X_b):

X[1]  = smoothStep(x, X_a[1] , X_b[1] , x_small);
X[2]  = smoothStep(x, X_a[2] , X_b[2] , x_small);
   ...
X[nX] = smoothStep(x, X_a[nX], X_b[nX], x_small);

or

X[1]  = c*(X_a[1]  - X_b[1])  + (X_a[1]  + X_b[1])/2
X[2]  = c*(X_a[2]  - X_b[2])  + (X_a[2]  + X_b[2])/2;
   ...
X[nX] = c*(X_a[nX] - X_b[nX]) + (X_a[nX] + X_b[nX])/2;
c     = (x/x_small)*((x/x_small)^2 - 3)/4

Summing all mass fractions together results in

sum(X) = c*(sum(X_a) - sum(X_b)) + (sum(X_a) + sum(X_b))/2
       = c*(1 - 1) + (1 + 1)/2
       = 1

Interface

function smoothStep
  extends Modelica.Icons.Function;
  input Real x "Abscissa value";
  input Real y1 "Ordinate value for x > 0";
  input Real y2 "Ordinate value for x < 0";
  input Real x_small(min = 0) = 1e-5 "Approximation of step for -x_small <= x <= x_small; x_small > 0 required";
  output Real y "Ordinate value to approximate y = if x > 0 then y1 else y2";
end smoothStep;

Revisions

• April 29, 2008 by Martin Otter:
  Designed and implemented.
• August 12, 2008 by Michael Sielemann:
  Minor modification to cover the limit case x_small -> 0 without division by zero.