We propose a quantum algorithm which uses the number of qubits in an optimal way and efficiently simulates a physical model with rich and complex dynamics described by the quantum sawtooth map. The numerical study of the effect of static imperfections in the quantum computer hardware shows that the main elements of the phase space structures are accurately reproduced up to a time scale which is polynomial in the number of qubits. The errors generated by these imperfections are more dangerous than the errors of random noise in gate operations.PACS numbers: 03.67. Lx, 05.45.Mt, 24.10.Cn When applied to computation, quantum mechanics opens completely new perspectives: a quantum computer, if constructed, could perform certain computations faster than classical computers, exploiting quantum mechanical features such as entanglement or superposition [1]. Shor discovered a quantum algorithm [2] which factorizes large integers exponentially faster than any known classical algorithm. It was also shown by Grover [3] that the search of an item in an unstructured database can be done with a square root speed up over any classical algorithm. However, at present only a few quantum algorithms are known which simulate physical systems with exponential efficiency. Such systems include certain many-body problems [4], spin lattices [5], and models of quantum chaos [6,7].In this Letter, we present a quantum algorithm which computes the time evolution of the quantum sawtooth map, exponentially faster than any classical computation. This model has rich and complex dynamics and finds various applications, e.g. for dynamical localization in billiards [8][9][10]. Our algorithm, based on the Quantum Fourier Transform (QFT) [11], simulates the dynamics of a system with N levels in O((log 2 N )2 ) operations per map iteration, while a classical computer, which performs Fast Fourier Transforms (FFT), requires O(N log 2 N ) operations. A further striking advantage of the algorithm is the optimum utilization of qubits: one needs only n q = log 2 N qubits (without any extra work space). We demonstrate that complex phase space structures can be simulated with less that 10 qubits, while about 40 qubits would allow one to make computations inaccessible to present-day supercomputers. This is particularly important, since experiments with few qubits are being performed at present [12,13], in particular the QFT was implemented on a three qubit nuclear magnetic resonance quantum computer [14]. For this reason the investigation of this interesting physical system will be accessible to first quantum computers, operating with few qubits and for which large-scale computations like integer factoring are not possible.The classical sawtooth map is given bywhere (n, θ) are conjugated action-angle variables (0 ≤ θ < 2π), and the bars denote the variables after one map iteration. Introducing the rescaled variable p = T n, one can see that the classical dynamics depends only on the single parameter K = kT , so that the motion is stable for −4 < K < 0 and com...