The following exercise is chapter one from the famous book, Dvoretsky's Analytical Manual.
This is what Dvoretsky had to say about this position:
Not long ago, I reexamined an exercise from my notebook, in which White won thanks to a beautiful attack. I ran it past "Fritz", which told me the attack was refuted - several different ways. I could not believe that - Black's king position looked much too dangerous. And in fact, soon the computer went from dour to smiley-face, showing equality in lines that it had previously considered won for Black. This was more believable; but I am not a fan of situations in which there are a number of equivalent ways to play (even though, in practice, such situations occur all the time). I continued searching; and finally,a ll the efensive tries went down, save one - now this result, could live with. And in the end, we found a difficult win here, as well.
The course of analysis uncovered more and more subteties, some of them utterly fantastic, "non-human." People cannot play chess at this level yet - in fact, I hardly think they will ever be capable of learning how, especially taking into account the current tendency toward faster time-controlls and the ever-growing predominance of sporting (or should I say, "financial") aspects over the creative side of chess life.
For the top-class player, it might prove an even more effective form of training to play out the starting position against a strong opponent. Play several such games, alternating white and black, allowing yourself and your opponent enough thinking time (not less than an hour apiece), and only afterwards, begin your acquaintance with my analysis.