Differential calculations are an efficient way to reduce the time it takes to do complex calculations on old computers. On modern computers this doesn't matter much, and perhaps not in BASIC either where all the calculations are floating point, but in an assembly programs this can make a lot of difference.
For the first example, let's say we need to calculate x squared from 1 to 10. An obvious way to do this is:
10 for x=1 to 10
20 print x*x
30 next x
But we don't like the multiplication since this takes much more time than addition or subtraction. Let's investigate how the value of x*x changes when x changes from x to x+1:
dx = (x+1)*(x+1) - x*x = x*x + x + x + 1 - x*x = 2*x + 1
That's not bad since 2*x can be done as a binary shift in assembly, and +1 is an INC instruction, which together are faster than a multiplication.
In BASIC it looks like this:
20 for n=1 to 10
30 print x
50 next n
But we can go further to reduce the complexity of the calculations. Let's investigate what happens to dx=2*x+1 as x changes from x to x+1:
ddx = 2*(x+1)+1 - (2*x+1) = 2*x + 2 + 1 - 2*x - 1 = 2
We like that
: no more multiplications. For each loop we just need to add 2 to dx, and dx to x:
30 for n=1 to 10
40 print x
70 next n
So we have ended up calculating a series of squares using additions only. In assembly dx=dx+2 can simply be written as INCT @DX which is even faster than a general addition.
This was a very simple example. Let's take another where differential calculations really matter. We want to rotate all the pixels on the screen by a certain angle (a). The formulas for rotating a single point (x, y) are:
x1 = x * cos(a) - y * sin(a)
y1 = y * cos(a) + x * sin(a)
That's pretty complex, but as we move through a scanline x always changes from x to x+1. Let's investigate what happens to x1 and y1 in this case:
dx1 = (x+1) * cos(a) - y * sin(a) - (x * cos(a) - y * sin(a)) = (x+1) * cos(a) - y * sin(a) - x * cos(a) + y * sin(a) = (x+1) * cos(a) - x * cos(a) = cos(a)
dy1 = y * cos(a) + (x+1) * sin(a) - (y * cos(a) + x * sin(a)) = y * cos(a) + (x+1) * sin(a) - y * cos(a) - x * sin(a) = (x+1) * sin(a) - x * sin(a) = sin(a)
So as we move a pixel to the right, dx1=dx1+cos(a) and dy1=dy1+sin(a). Since we want to rotate the entire image by the same angle a, cos(a) and sin(a) are constants, so again we have reduced the complex calculations to simple additions.
If we do the same math when we move one scanline down (y -> y + 1) we get dx1=dx1-sin(a) and dy1=dy1+cos(a), so we can rotate an entire image using just additions and subtractions.
I hope this shows how differential calculations can be useful. If you have other examples, please post them here.