Note: This page has been translated by MathWorks. Please click here

To view all translated materals including this page, select Japan from the country navigator on the bottom of this page.

To view all translated materals including this page, select Japan from the country navigator on the bottom of this page.

The **Expansion factor** and **Contraction
factor** options, in **Mesh** options, control
how much the mesh size is expanded or contracted at each iteration.
With the default **Expansion factor** value of `2`

,
the pattern search multiplies the mesh size by `2`

after
each successful poll. With the default **Contraction factor** value
of `0.5`

, the pattern search multiplies the mesh
size by `0.5`

after each unsuccessful poll.

You can view the expansion and contraction of the mesh size
during the pattern search by selecting **Mesh size** in
the **Plot functions** pane. To also display
the values of the mesh size and objective function at the command
line, set **Level of display** to `Iterative`

in
the **Display to command window** options.

For example, set up the problem described in Linearly Constrained Problem as follows:

Enter the following at the command line:

x0 = [2 1 0 9 1 0]; Aineq = [-8 7 3 -4 9 0]; bineq = 7; Aeq = [7 1 8 3 3 3; 5 0 -5 1 -5 8; -2 -6 7 1 1 9; 1 -1 2 -2 3 -3]; beq = [84 62 65 1];

Set up your problem in the Optimization app to match the following figures.

Run the optimization.

The Optimization app displays the following plot.

To see the changes in mesh size more clearly, change the *y*-axis
to logarithmic scaling as follows:

Select

**Axes Properties**from the**Edit**menu in the plot window.In the Properties Editor, select the

**Y Axis**tab.Set

**Scale**to**Log**.

Updating these settings in the MATLAB^{®} Property Editor shows
the plot in the following figure.

The first 5 iterations result in successful polls, so the mesh sizes increase steadily during this time. You can see that the first unsuccessful poll occurs at iteration 6 by looking at the command-line display:

Iter f-count f(x) MeshSize Method 0 1 2273.76 1 1 2 2251.69 2 Successful Poll 2 3 2209.86 4 Successful Poll 3 4 2135.43 8 Successful Poll 4 5 2023.48 16 Successful Poll 5 6 1947.23 32 Successful Poll 6 15 1947.23 16 Refine Mesh

Note that at iteration 5, which is successful, the mesh size
doubles for the next iteration. But at iteration 6, which is unsuccessful,
the mesh size is multiplied `0.5`

.

To see how **Expansion factor** and **Contraction
factor** affect the pattern search, make the following changes:

Set

**Expansion factor**to`3.0`

.Set

**Contraction factor**to`2/3`

.

Then click **Start**. The **Run
solver and view results** pane shows that the final point
is approximately the same as with the default settings of **Expansion
factor** and **Contraction factor**, but
that the pattern search takes longer to reach that point.

When you change the scaling of the *y*-axis
to logarithmic, the mesh size plot appears as shown in the following
figure.

Note that the mesh size increases faster with **Expansion
factor** set to `3.0`

, as compared with
the default value of `2.0`

, and decreases more slowly
with **Contraction factor** set to `0.75`

,
as compared with the default value of `0.5`

.

The mesh accelerator can make a pattern search converge faster
to an optimal point by reducing the number of iterations required
to reach the mesh tolerance. When the mesh size is below a certain
value, the pattern search contracts the mesh size by a factor smaller
than the **Contraction factor** factor. Mesh accelerator
applies only to the GPS and GSS algorithms.

For best results, use the mesh accelerator for problems in which the objective function is not too steep near the optimal point, or you might lose some accuracy. For differentiable problems, this means that the absolute value of the derivative is not too large near the solution.

To use the mesh accelerator, set **Accelerator** to `On`

in
the **Mesh** options. Or, at the command line, set
the `AccelerateMesh`

option to `true`

.

For example, set up the problem described in Linearly Constrained Problem as follows:

Enter the following at the command line:

x0 = [2 1 0 9 1 0]; Aineq = [-8 7 3 -4 9 0]; bineq = 7; Aeq = [7 1 8 3 3 3; 5 0 -5 1 -5 8; -2 -6 7 1 1 9; 1 -1 2 -2 3 -3]; beq = [84 62 65 1];

Set up your problem in the Optimization app to match the following figures.

Run the optimization.

The number of iterations required to reach the mesh tolerance
is 78, as compared with 84 when **Accelerator** is
set to `Off`

.

You can see the effect of the mesh accelerator by setting **Level
of display** to `Iterative`

in **Display
to command window**. Run the example with **Accelerator** set
to `On`

, and then run it again with **Accelerator** set
to `Off`

. The mesh sizes are the same until iteration
70, but differ at iteration 71. The MATLAB Command Window displays
the following lines for iterations 70 and 71 with **Accelerator** set
to `Off`

.

Iter f-count f(x) MeshSize Method 70 618 1919.54 6.104e-05 Refine Mesh 71 630 1919.54 3.052e-05 Refine Mesh

Note that the mesh size is multiplied by `0.5`

,
the default value of **Contraction factor**.

For comparison, the Command Window displays the following lines
for the same iteration numbers with **Accelerator** set
to `On`

.

Iter f-count f(x) MeshSize Method 70 618 1919.54 6.104e-05 Refine Mesh 71 630 1919.54 1.526e-05 Refine Mesh

In this case the mesh size is multiplied by `0.25`

.

Was this topic helpful?