Skip main navigation
We use cookies to give you a better experience, if that’s ok you can close this message and carry on browsing. For more info read our cookies policy.
We use cookies to give you a better experience. Carry on browsing if you're happy with this, or read our cookies policy for more information.

Skip to 0 minutes and 4 secondsGame-playing programs, even in the field of classical computers, have evolved considerably, as evidenced by the rise of old Deep Blue in the 1990s, several current shogi programs, and recently Alpha Go. These games are hard because the number of possible games grows exponentially with the number of choices at each turn. If we apply quantum computers to these kinds of games, it seems that you'll be able to find the most advantageous move by calculating all the possibilities. In fact, however, there are many unresolved issues. Search and optimization problems pervade our economy and society. How can a factory minimize the amount of raw material it uses?

Skip to 0 minutes and 54 secondsHow can a trip planner such as the mapping applications from major IT companies like Google and Apple pick the best route, given a start and end point and some constraints? In a lot of real-world problems, you know what the answer is, or at least what the answer looks like, but not how to get it. Chess is a good example. In chess, you move your pieces around until it is impossible for your opponent's king to escape being captured. It's pretty easy to see when you have checkmated your opponent, but it's not easy to see the sequence of moves that takes you there.

Skip to 1 minute and 32 secondsThe most straightforward way to figure out how to win is to test the entire set of possible moves you can make, followed by all of your opponent's possible responses, and your replies, and so forth. This is called the brute force approach. How many candidate moves you have to consider at each step is called the branching factor, and each step is called a ply. In chess, white has twenty possible first moves, and black has twenty possible responses, so there are 20×20=400 possible combinations in the first two plies. This number grows exponentially. If there are b branches and you need to check k plies, then you have b^k total possibilities.

Skip to 2 minutes and 17 secondsFor complete chess games, that gives you maybe 10^120 possible sequences to test, far too many for a human or even a supercomputer to check individually. So, humans and computers both use heuristics to take a good guess at a possible solution, dramatically reducing the number of moves they have to test. This makes the computation possible, but it doesn't guarantee that you'll find the optimal solution. Heuristics also require that the problem have a certain amount of structure, so that you can guide your choices toward a good, if not perfect, answer.

Skip to 2 minutes and 53 secondsIn chess, for example, capturing an opponent's piece gives you an advantage, so it makes sense for a program to sense for a program to spend more time looking at cases where it's likely that you will capture a piece, to see if they lead to a win. Grover's algorithm is a quantum algorithm for searching these kinds of spaces. It gives you its best benefit on problems with no visible structure. If testing one possible solution tells you yes or no, but doesn't give you any help in making a good choice for the next candidate to test, then it may be a good candidate for solving using Grover's algorithm.

Skip to 3 minutes and 36 secondsThe speedup using a quantum computer, however, isn't as large as it is for factoring. We only reduce the number of times we need to run our testing routine to the square root of the original number. For chess, then, instead of 10^120, we would have √(10^120 )=10^60, which is still too large a number to actually test. So, we have to be careful when picking problems to solve using Grover's algorithm.

Generalized search

Computer scientist Lov Grover found a quantum algorithm, or perhaps more correctly an algorithmic framework, that allows quantum computing to be applied to almost any problem. It is commonly applied to search problems, where we are looking for a path across town or a trying to identify an input value for a function that will give us a particular output value. Small-data problems with high branching factors, like solving chess, shogi or go, are good candidates.

Share this video:

This video is from the free online course:

Understanding Quantum Computers

Keio University

Contact FutureLearn for Support