How is the division of two numbers carried out in Matlab?

The behavior is as expected when both the denominator and the numerator are in 'double' precision. If either of the two are integer, as in the example below, the rounding error yields unexpected division results.

double(475904) / double(512) = 929.5

int32(475904) / int32(512) = 930

int32(475904) / double(512) = 930

double(475904) / int32(512) = 930

[David Fletcher]

In addition to John D'Errico's answer below this is from the Matlab documentation:

Arithmetic operations that involve both integers and floating-point always result in an integer data type. MATLAB rounds the result, when necessary, according to the default rounding algorithm. The example below yields an exact answer of 1426.75 which MATLAB then rounds to the next highest integer:

int16(325) * 4.39
ans =

    int16

     1427

Also if you are finding that unexpected, what about this:

[int16(5) int32(3) double(5.52) logical(1) pi]

ans =

    1×5 int16 row vector

     5   3   6   1   3

but again, from the documentation:

If you combine integers with double, single, or logical classes, all elements of the resulting matrix are given the data type of the left-most integer.

[PE]

Thanks! Well explained.

[Stephen23]

The documentation that David Fletcher refers to are these:

https://www.mathworks.com/help/matlab/matlab_prog/integers.html

https://www.mathworks.com/help/matlab/numeric-types.html

[PE]

Thanks. Here are two not so obvious learnings in Matlab- 1) Data operations that involve both integers result in an integer data type. 2) The division of two integers is rounded to nearest integer by default. If at all one would have to do this operation, idivide( int32(475904), int32(512) ) would be used.

Answer 1

[John D'Errico]

Think about what the class of the result is. If you don't know how to use the class function to check that, then learn to read the output of the whos command. (In fact, in newer MATLAB releases, just displaying a variable at the command line tells you the class, if it is not a double.)

Then consider if the result is stored in an integer variable, then why would you expect a non-integer result? That is, an int32 cannot represent non-integers. So MATLAB stores only the integer part, essentially rounding to the nearest integer.

double(4) / int32(3)
ans =
  int32
   1

double(5) / int32(3)
ans =
  int32
   2

[PE]

Thanks! This wasn't intuitive given the experience of other languages. But it seems it is clearly documented in Matlab Fundamentals. Thanks for the reference link.