The last few weeks I’ve been implementing an improved version of my noise function. I knew from the start what the final result should be, but I still needed to actually implement the algorithm in a GPU shader.
I made a lot of false starts, trying one approach to the implementation after another, and kept failing. Finally yesterday I stumbled upon the key insight I needed, and managed to get it all working properly.
The reason I stuck with it, despite all my failed attempts, was that my strategy was to start from the answer. I already knew where I wanted to go, I just needed to figure out how to get there.
I think a lot of research is like that. You see the mountaintop in the distance, and keep your eyes firmly fixed on it. Then you try lots of different paths to get there, learning a little bit from each successive failed attempt.
Eventually you get to the mountaintop, because you always knew it was there. After all, you never took your eyes off of it.