You are on page 1of 4
Hoofdstuk 2 General definition and properties of CMPS Elk IXTEX document bestaat uit twee delen: een preamble en cen body. In de preamble komen verschillende opties die voor het hele document gelden. In de body komt de eigenlijke tekst van het document. De preamble begint bij het begin van het bestand en eindigt bij het TATEX commando \begin{document}. Dit is ook het begin van de body. De body eindigt met \end{docunent}. Hen minimaal IXTX bestand zict ex dus als volgt uit: \docunentclass{article} \begin{docunent} In theorie zijn theorie en praktijk gelijk, maar in praktijk zijn ze verschillend. \end{document} 2.1 Tensor Networks Why tensor networks? Upon examining the existing literature on numerical methods for tarts to wonder about the wide variety of strongly correlated quantum systems one soon existing methods. One can for example exactly diagonalize the hamiltonian the system but this is of course restricted to small systems and far away from the thermodynamic limit, where the subject of this thesis is situated. Another popular method is mean field approximation, but because for every particle mean field theory averages out the interaction with all other particles quantum correlations are completely lost, Quantum Monte Carlo is not applicable to every system because of the sign problem, but is often a benchmark tool for the systems where it does. ‘Tensor networks isn’t a perfect ansatz cither but it extends the range of accessible med. systems considerably by making use of the entanglement structures of the systems con Hoofdstuk 2. General definition and properties of CMPS 8 2.1.1 Introduction Consider the most general quantum many-body wave function of a system consisting of particles. The wave function of this generic system is then given by 1) = SP Cinta fini stn) (2.1) As} where the each (in| denotes the local basis of particle denoted by n. Assume that the system considered consists of particles of the same specie, each living in its local Hilbert space of dimension d, H = C4, In order to describe the full phase space of this system, one then needs to keep track of the d coefficients Ci,i,.iy. This exponential growth in the number of coefficients with respect to the number of particles renders the exact calculation of ma- croscopic quantities virtually impossible, considering that one needs particle numbers of the order of the constant of Avogadro, 6.02214076 x 10°, for macroscopic systems. However, in the earl Wilson, developed the so-called density matrix renormalisation group method (DMRG), an algorithm constructed to find the ground state of one-dimensional quantum systems. White y nineties of the previous century Steven White, a former PhD student of Kenneth had in fact understood that most of phase space was uscless because the relevant degrees of freedom of the wavefunction were the entanglement degrees of freedom. It took some years before it was recognized that the DMRG algorithm was in essence a variational optimization algorithm over matrix product states (MPS will be des and that the result were finitely correlated states!. Tensor networks, a generalisation of MPS, ibed in detail in the next section) has thus provided an invaluable too! by reducing the phase space exponentially so that is now has become much easier, the complexity growth now scales polynomially in N, to describe ground states and the entanglement structures of various quantum systems? (Note however that critical phenomena usual fall outside this scope because they are often characterized by scale invariance and thus infinite correlation Iengthe *It bas boon argued that only for spatial dimension D = 1 T! 's, in that case MPS, really would fficiently capture the subspace of states satisfying an area-law. When the spatial dimension of the system D > 2, it was proven that tensor networks only describe an exponential small subspace of this the already exponentially small subspace, However, when the condition of vanishing ot- rolation functions is tightened to exponentially decaying correlation functions this remains an open ‘question. ver wijzingenvragenaanbenoit /juthoofdatwelkloptwantmpszijnsowiesoalexponetiallydecayingvoorhuncort Hoofdstuk 2. General definition and properties of CMPS 9 — a eR Figuur 2.1: Various examples of description of Cji..-14- (a) is the visualisation of a matrix product state (MPS) with periodic boundary conditions, (b) that of the 2/)-dimensional extension of MPS, namely PEPS and (c) that of the multi-scale entanglement renormalization ansatz (MERA). ‘One can always divide the system described by Eq.2.1 in two arbitrary subsystems A and B. If the physical state is picked at random from the d¥-dimensional phase space, and one would calculate the entanglement entropy between those two regions, one would find that in gencral the entanglement entropy would scale with the volume of the smallest of these two subregions. Physicists have the incredible Iuck that nature has ensured that physical relevant Hamiltonians all are local. For the ground states and low energy excitations these local Hamiltonians the entanglement entropy tends only to scale with the surface area of the boundary between the two subsystems, $= 8A, The low-energy states of a local Hamiltonian are thus heavily constrained in the sense that they must obey the area law and they cover only a tiny exponentially small subspace of the unmanageable large many-body Hilbert space. Figuur 2.2: The small comer of the many-body Hilbert space where the area-law applies. This is algo the rogion where virtual all physical relevant states reside Furthermore, it has also been proven quite recently that when a random state of HE is Hoofdstuk 2. General definition and properties of CMPS 10 evolved over time that it takes a exponentially large time in the number of particles. Given that the physical ground states of local Hamiltonians already live in a tiny subspace of the vast Hilbert space of all possible states, it is now also shown that it is almost impossible to leave this subspace of finitely-correlated states. Another consequence is that the vast majority of the many-body Hilbert space seems to be unphysical as it can not be reached in a reasonable amount of time, This begs the question if it is sensible to describe physical states as vectors in the Hilbert space of all possible states. 2.2 MPS 2.2.1 Construction Consider a one dimensional system of V spins living on a lattice £. The individual spin states are encoded as |s), where ¢ denotes the site. Pure states of this system are denoted as |sis2--+sy) and any arbitrary state can be described as a superposition of these pure states. IB) = 0 Casaay [st82-- sw) (2.2) © Depending on which Hamiltonian is used, the spins will interact which each other in one or the other way, Because this is a quantum system, entanglement is introduced in the system, ‘The reason why matrix product states became a popular technique to tackle such systems is because they perfectly snited to produce long range entanglement, even when the system has only nearest neighbour interactions, In order to introduce entanglement in an efficient manner, virtual degrees of freedom as, bj, are introduced to every site i. These are called ancillas and live in a generally simple Hilbert space H with dimension D. ‘This is called the bond dimension. ‘The next step is to build maximally entangled states between these virtual degrees of freedom of neighbouring sites. One can choose for example the canonical maximally entangled state and denote this as D IK) = laa, @ lo), (2.3) where |a),, with @ = 1,.--,D is the D-dimensional basis of the aneilla a;. This entangled structure is projected upon the physical spin states through the linear map A(i) given by AG) = Aty..O|3}; @ (ale, ® (Bl, ® (2.4) ‘The phys product of the above elements cal state is of course still a superposition of pure states and is constructed of a direct

You might also like