7.2.2.2 Executable DIM Statements
  Executable DIM statements have at least one variable bound.
  Bounds can be constants or simple variables, but at least one
  bound must be a variable. Executable DIM statements let you
  redimension an array at run time. The bounds of the array
  can become larger or smaller, but the number of dimen-
  sions cannot change. For example, you cannot redimension a
  four-dimensional array to be five-dimensional.

  The executable DIM statement cannot be used on arrays in
  COMMON, MAP, DECLARE, or declarative DIM state-
  ments, nor on virtual arrays or arrays received as formal
  parameters.

  Whenever an executable DIM statement executes, it reini-
  tializes the array. If you change the values of an executable
  DIM statement, the initial values are reset each time the
  DIM statement is executed.

  In the following example, the second DIM statement reini-
  tializes the array real_array ; therefore, real_array(1%)
  equals zero in the second PRINT statement:
  X% = 10%
  Y% = 20%
  DIM real_array(X%)
  real_array(1%) = 100
  PRINT real_array(1%)
  DIM real_array(Y%)
  PRINT real_array(1%)
  END

  Output
  100
  0
  You cannot reference an array named in an executable
  DIM statement until after the DIM statement executes. If
  you reference an array element declared in an executable
  DIM statement whose subscripts are larger than the bounds
  specified in the last execution of the DIM statement, BASIC
  signals the run-time error ``Subscript out of range'' (ERR =
  55), provided subscript checking is enabled.