Question

What is the difference between cell (i.e. with { }) and matrix (i.e. with [ ]) in Matlab?

Answer 1

There are several differences between a cell array and a matrix in MATLAB:

• A cell array may contain any arbitrary type of element in each cell; while a matrix requires the types of its elements to be homogeneous i.e. of the same type.

• As far as memory layout goes, all elements of a matrix are laid out contiguously in memory, while a cell array contains pointers to each element of the array. This can be important when considering things like cache locality for high performance code.

• The flip side of point 2 is that when you resize a matrix every element in the matrix must be copied over to the newly allocated memory area, but in case of a cell array only a list of pointers needs to copied over. Depending on the size and type of elements you’re storing, this might mean cell arrays are much faster to resize.

---

To illustrate the differences in memory layout, let’s consider a simple example:

A = [10 20 30 40];

Here MATLAB creates a new matrix variable named A, allocates enough memory to hold 4 doubles (32 bytes, assuming 8 byte doubles) and assigns this memory to a pointer that points to the real part of A. (If you create a matrix of complex numbers, memory is allocated for the imaginary part also, and a separate pointer points to this memory area).

Now let’s create a cell array that holds these elements:

B = cell(1, 4);
B{1,1} = 10;
B{1,2} = 20;
B{1,3} = 30;
B{1,4} = 40;

When MATLAB executes the first statement, it creates a cell array that contains 4 pointers, each of which can point to an arbitrary type. So B is already using 16 bytes (assuming 32-bit pointers).
The next line creates a 1×1 matrix containing the value 10 and assigns it to the first cell array element. The process here is similar to the one I described above for creation of a 1×4 matrix, except that the memory allocated is only large enough to hold one double (8 bytes). This is repeated for each of the remaining 3 statements. So, at the bare minimum, the second example uses 16 + 8 x 4 = 48 bytes.

Note that each variable in MATLAB also includes memory overhead for a structure called an mxArray that stores information such as dimension, data type and a lot more about that variable. I’ve ignored this overhead for the sake of simplicity.