Web Quantum Computer Simulator

Heraklit writes "As reported on Heise News, the Frauenhofer Institute of Computer Architecture and Software Technology has made available the first online quantum computer simulator - it will be simulating up to 31 quantum bits, for testing new advanced quantum algorithms. Behind the scenes, it is a 32 node Athlon 3200 Myrinet Linux Cluster with 56GByte RAM! Now imagine the computing power of a few hundred qubits, if ever constructed..."

Comments

Re:wow!!!

[zeath]

Unfortunately, quantum computers aren't as powerful to the giddy consumer as that cluster describes. They're capable of doing repetitive, simple mathemetical tasks simultaneously on a large number of values. It's extremely complicated how that works, but I have it written in this paper (pdf) that I wrote a few years ago. The paper was focused primarily on quantum physics for the first half (also interesting, and related to the story ran a few weeks ago on the red laser and the parallel universe theory), while the second half deals with explaining how the quantum registers work. It starts in the second paragraph of page 3, though a few terms reference previous topics from the paper. It's only a few pages long and it'll explain a lot of things (some things more technical than others) that none of the articles explained. Especially pay attention to the first full paragraph on page four, which I'll quote here:

Richard Feynman was one of the first to see the potential in quantum superposition for solving such exponentially complicated problems much faster. For example, a system of 500 qubits, which is impossible to simulate with any computer today, represents a quantum superposition of as many as 2^500 states. Each of these states would be equivalent to a single list of 500 1's and 0's in a classical computer. A single quantum operation on such a system would simultaneously operate on all 2^500 states; with a single tick of the quantum computer's clock, the operation would compute not just on one machine state, as our serial computers do, but on all 2^500 machine states at once. Eventually, observing the system would cause it to reduce into a single state corresponding to a single answer, a single list of 500 1's and 0's, as measured by an axiom of quantum mechanics. A classical super computer would take approximately 10^150 separate processors to accomplish this task in the same amount of time (which is, of course, impossible).

What I can explain without too much trouble is that the cluster is merely emulating the abilities of a quantum computer. A quantum computer, conversely, would be incapable of matching the performance of, say, seti@home on all of those machines. Emulation is taxing on any system - just ask the people who are using PearPC on their brand spankin' new computers only to get sub-G3 performance out of OS X.

Re:For the quantumly challenged amoung us

[NonSequor]

Basically this stuff can't be done in polynomial time. For all quantum algorithms you start by setting a bunch of qubits into a uniform superposition of states (e.g. if you do this to 8 qubits and then measure them, you will be equally likely to get any number between 0 and 255 as your result). Then you can use these qubits as input into a function and effectively calculate the value of that function over every possible value of the input. The trouble is that you don't get 2^n different values of the function, you get a superposition of 2^n states. When you measure the output, you'll only find out one of the values of the function. So in order to get a working quantum algorithm, you have to manipulate the quantum state until you have a high probability of measuring the state you want.

Quantum computing has other complexities. Every function must output as many qubits as it has for input. It's also impossible to make a copy of a qubit without altering the original qubit. This means that in any quantum programming langauge, all funciton parameters must be passed by reference. All functions must be invertible. This can be generally accomplished by leaving the inputs unaltered and writing the output to some scratch qubits which are set to 0 beforehand.

If you want to learn more about quantum algorithms, I suggest you read up on Grover's search algorithm. It's much simpler than many quantum algorithms and it's also proven very adaptible to other situations.