Further Examples: NIntegrate

One Dimension

NIntegrate will often give the numerical value of a definite integral without parameters faster than finding the definite integral and then using N.

In[1]:=

Out[1]=

In[2]:=

Out[2]=

In[3]:=

Out[3]=

Functions that cannot be integrated using algebraic routines can nonetheless be integrated numerically.

In[4]:=

Out[4]=

In[5]:=

Out[5]=

NIntegrate's default option values sometimes lead to a wrong answer.

In[6]:=

Out[6]=

Here are the options of NIntegrate. The default value of Automatic corresponds to GaussKronrod in one dimension and to MultiDimensional in several dimensions.

In[7]:=

Out[7]=

Increasing the number of recursive bisections by setting the option MaxRecursion gives a more reliable result.

In[8]:=

Out[8]=

The previous result agrees with that obtained by Integrate.

In[9]:=

Out[9]=

In[10]:=

Out[10]=

Here is a function that NIntegrate suspects has a singularity.

In[11]:=

Here is the plot of ff.

In[12]:=

NIntegrate does the integration correctly.

In[13]:=

Out[13]=

In[14]:=

Out[14]=

In[15]:=

Out[15]=

This calculates the integral numerically to different precision goals, and accumulates the sampling points used for the integral approximations with the option EvaluationMonitor.

In[16]:=

This plots the points that were accumulated in samplePoints. The function is sampled at the x coordinates in the order of the y coordinates.

In[17]:=

Here is a function that oscillates rapidly.

In[18]:=

Here is its plot.

In[19]:=

Using NIntegrate with its default options settings gives an incorrect result.

In[20]:=

Out[20]=

NIntegrate gives the correct result using .

In[21]:=

Out[21]=

It happens that an exact result is possible and taking N verifies the previous result.

In[22]:=

Out[22]=

In[23]:=

Out[23]=

In[24]:=

Multidimensional

Low-dimensional integration can be done either with genuine space integration rules or with a Cartesian product of one dimensional rules. We compare the two using the very simple two-dimensional function .

Here are the sample points that come from using MultiDimensional, the default for NIntegrate.

In[25]:=

Here are the sample points from using the Cartesian product setting GaussKronrod.

In[26]:=

Here we will use NIntegrate on a more complicated function. Here is its plot.

In[27]:=

Here is a plot of the sample points for the numerical integration using .

In[28]:=

For comparison, here is a plot of the sample points for the numerical integration using .

In[29]:=

In[30]:=

MonteCarlo

is occasionally advantageous in dimensions greater than .

In[31]:=

Out[31]=

Setting the option Method to MonteCarlo[24] is equivalent to using SeedRandom[24] to reset the random number generator.

In[32]:=

Out[32]=

You can specify how many function evaluations are used in the methods MonteCarlo and QuasiMonteCarlo by giving a value to MaxPoints (the default being ).

In[33]:=

Out[33]=

If the AccuracyGoal and PrecisionGoal are set to Automatic, they will be set to in the methods MonteCarlo and QuasiMonteCarlo, and to in all other methods.