You noted that there is no efficiency when you have iterations a[i]=a[i-1]+c, loop dependency. That is correct, but still vectors could be used successfully if you have more than one array. For example: vector instruction SSE allows me to calculate 4 iterations with one _m128 register, Mandelbrot set for example. First iteration at first 32 bits field of that package, second iteration at second field of xmm register, third at third 32 bits package and 4th at the last one. So we are able to calculate that with xmm. Instead of using xmm for one iteration, I can use it for four iterations. Still we are able to use it for a large scale of complex number iterations...
