Focus Article

Low entanglement wavefunctions
Garnet Kin-Lic Chan∗
We review a class of efficient wavefunction approximations that are based around the limit of low entanglement. These wavefunctions, which go by such names as matrix product states and tensor network states, occupy a different region of Hilbert space from wavefunctions built around the Hartree–Fock limit. The best known class of low entanglement wavefunctions, the matrix product states, forms the variational space of the density matrix renormalization group algorithm. Because of their different structure to many other quantum chemistry wavefunctions, low entanglement approximations hold promise for problems conventionally considered hard in quantum chemistry, and in particular problems which have a multireference or strong correlation nature. In this review, we describe low entanglement wavefunctions at an introductory level, focusing on the main theoretical ideas. Topics covered include the theory of efficient wavefunction approximations, entanglement, matrix product states, and tensor network states including the tree tensor network, projected entangled pair states, and the multiscale entanglement renormalization ansatz. C 2012 John Wiley &
Sons, Ltd.

How to cite this article:

WIREs Comput Mol Sci 2012, 2: 907–920 doi: 10.1002/wcms.1095

n wavefunction-based quantum chemistry, we wish to determine the stationary states of the electronic Schrodinger equation. This is a difficult task ¨ because the dimension of the Hilbert space grows exponentially with the number of electrons. Thus, any practical calculation must use some simplifying approximations, or ansatz, for the state we wish to study. Figure 1 illustrates schematically the full Hilbert space of wavefunctions. Complete expansions, for example, via full configuration interaction or full resonance valence bond theory, span the entire Hilbert space at exponential cost. Practical wavefunction approximations are of polynomial cost, and are necessarily biased toward a given portion of the Hilbert space. In opposite corners of the rectangle in Figure 1, we show two families of approximations salient to our discussion. One is biased toward the Hartree– Fock mean-field or independent particle limit. Such wavefunctions are well developed in quantum chemistry and can be systematically made more flexible
Correspondence to: Department of Chemistry and Chemical Biology, Cornell University, Ithaca, NY 14853, USA DOI: 10.1002/wcms.1095


by increasing the excitation level in the wavefunction. The other family of wavefunctions, which are the main subject of this review, are biased toward the independent local subsystem, or zero entanglement, limit. These wavefunctions are systematically made more flexible by increasing the amount of encoded entanglement. Low entanglement wavefunctions first became widely applied with the advent of the density matrix renormalization group (DMRG).1–8 Within a few years, it was understood that the DMRG algorithm is an energy minimization algorithm within a class of low entanglement wavefunctions known as matrix product states (MPS).5–7,9–15 Decimation in the renormalization group (RG) algorithm is equivalent to encoding limited entanglement in the state, linking the RG picture with quantum information perspectives.6,13,15–17 In the last decade, MPSs also have been used in quantum chemistry via the DMRG algorithm.8,17–50 Quantum chemical DMRG has significantly widened the scope of ab initio calculations with respect to strongly correlated electrons. In particular, it has allowed multireference active space calculations with more than 30, and in some cases up to 100, active orbitals (see Refs 8,22,23,25– 30,42,43,47–49 and Ref 7 for a summary of recent calculations).

Volume 2, November/December 2012

c 2012 John Wiley & Sons, Ltd.


with a small error. The practical success of quantum chemistry. have been introduced. or add in the interactions between the electrons. they can be made to span the full Hilbert space. In two-dimensional system.Focus Article wires. This is not essentially changed if include antisymmetry. We finish with our conclusions in section Conclusions. which remove some fundamental limitations. consider a set of N noninteracting electrons. in approximations built around these functions around the Hartree–Fock Limit and Low Entanglement Wavefunctions. Indeed. We begin with a brief discussion of wavefunctions built around the independent particle limit. (2) The overlap of the approximate and exact wavefunctions is a product of the overlaps of the orbitals | = φ 1 | φ1 φ2 | φ2 . (1) F I G U R E 2 | Area laws express locality in physical systems. The independent particle and independent local subsystem limits lie in different regions of the Hilbert space.51–64 Such tensor network wavefunctions are still in their infancy. Then we turn to our main focus—low entanglement wavefunctions. and the multiscale entanglement renormalization ansatz (MERA). the discussion in Ref 66). (3) We see that the overlap of the true and approximate determinants decreases exponentially with the number of particles. Volume 2. e. for example.15. 60. thus entanglement entropy in a physical ground state is expected to be independent of system size. We next describe more general tensor network wavefunctions. it is possible to faithfully approximate a quantum state. we first consider why. except for critical and topological corrections. Indeed. despite the exponential size of the Hilbert space. We describe them also in this review. φ |φ = 1 − . respectively. This suggests that any approximate wavefunction is a very poor representation of the quantum state in a large systems. 65 for early work in this direction). F I G U R E 1 | The Hilbert space of many-electron wavefunctions. 908 c 2012 John Wiley & Sons. generalizations of MPSs. In section Wavefunctions and the Orthogonality Catastrophe. Nonetheless. This might even suggest that a wavefunction is not a valid way to describe large quantum systems (see. φ N ( r N ) . we describe the two families of wavefunctions presented above. We introduce the mathematical definition of entanglement. Then the ground-state wavefunction is a product of N orbitals = φ1 ( r 1 ) φ2 ( r 2 ) . the boundary area is independent of the overall system size.wiley. The structure of the discussion is as follows. and the entanglement entropy scales similarly. More recently.g. . see Refs 50. unless at a quantum critical point. In a one-dimensional system. with respect to system dimensionality. there appears to be a fundamental problem. . By increasing the excitation level or entanglement. and efficient computational techniques to work with them remain under development. as the orthogonality catastrophe shows that in a large system. November/December 2012 . To illustrate the catastrophe. they have yet to be applied to quantum chemistry (also. In sections Wave- We imagine representing each orbital φ by an approximate φ . projected entangled pair states. For simplicity. . Ltd. WAVEFUNCTIONS AND THE ORTHOGONALITY CATASTROPHE How do we construct a many-electron wavefunction to represent a quantum state? At first glance. . φ N | φ N = (1 − ) N ∼ exp(− N). any approximate wavefunction is orthogonal to the state of interest. they hold considerable promise due to their increased flexibility over MPSs. we neglect antisymmetry. indicates that approximate many-electron wavefunctions do contain useful information. however. including tree tensor networks (TTNs). the boundary area can scale as the system width. and proceed to describe in detail the theory of MPS.

known as Jastrow factors (sometimes known as correlation factors). .67–70 Thus. † (8) There are other ways to construct wavefunctions around the Hartree–Fock limit aside from an excitation expansion. We first describe wavefunctions that are built around the Hartree–Fock limit. Thus. For the scheme to be practical. Compact wavefunction approximations are constructed by restricting the generality of the excitation expansion.. b. ator T ˆ = T ia † tia aa ai + i > ja >b † tiab j aa ab ai a j + .72 . Now that we have established that approximate wavefunctions can in principle provide meaningful descriptions of N electron systems. j. Further examination of the original argument also shows that the catastrophe is not actually as severe as it first appears. the Jastrow determinant wavefunction can also cover the full quantum Hilbert space. ji j n † (9) ˆ i is a number operator ai ai in some basis.. that is. .. Indeed..g. . . Ltd. Starting from a determinant allows a separation of orbitals into an occupied space (labeled by i.. we still have the problem of determining appropriate approximations for physical problems. with expectation single-particle operator Q iq ˆ| . expectation values of operators can be very good. to the determinant. In both the excitation and Jastrow expansions. thus defining excita- As successively more terms are included in the Jastrow factor.WIREs Computational Molecular Science Low entanglement wavefunctions even with a poor overlap. φi |q (4) + ia cia | ciab j | a i ab ij We see that the error in the expectation value per particle is O( ).71. | =| + i > j . we must be able to determine an orbital to a given error of in a time polynomial in −1 . Symmetry-broken solutions can often be associated with different kinds of electronic phases and can thus be chosen to reflect the Volume 2. even using the total wavefunction overlap as a measure of accuracy. 909 . the thermodynamic limit. The Hartree–Fock wavefunction is the determinant which minimizes the energy E = |H| / | 71. However. (6) By mixing in successively higher excitations. . . the approximate wavefunction can retain unit overlap with the true state. wavefunctions with poor overlap can yield meaningful expectation values even in the limit of large particle number. one expands a general wavefunction as a linear combination of excitations around the Hartree–Fock reference.. . (5) N In this case. In truncated configuration interaction and perturbation theories. . The Jastrow determinant wavefunction is then | = Jˆ | . Some solutions of the energy minimization do not preserve the symmetry of the original Hamiltonian. . value Q = | Q ˆ| |Q = i tions relative to the determinant. We assumed that we compute each orbital to within an error of . where n not necessarily the molecular orbital basis of the determinant. the Hartree–Fock energy minimization has many local minima. by Gaussian basis set ¨ expansion) satisfy this criterion. independent of the number of particles.75–77 Jastrow factors take the general form Jˆ = i ˆi + ji n i> j ˆin ˆ j + . we could try to represent each orbital more accurately as the number of electrons increases. φ |φ ∼ 1 − + .72 we restrict the maximum excitation level in the expansion (6). Current methods to solve the single-particle molecular Schrodinger equation (e. In the standard excitation hierarchy. November/December 2012 c 2012 John Wiley & Sons.) and a virtual space (labeled by a. For example. one covers the full quantum Hilbert space. to defeat the exponential loss of overlap we need the error in each orbital to decrease such as .). regardless of how large the system is. whereas truncated coupled cluster theory rewrites the expansion first in cluster form71–74 | ˆ )| = exp(T 0 . the choice of starting determinant is not unique. (7) and truncates the excitation level in the cluster operˆ. it is possible in some cases to faithfully represent the state of a large system with a wavefunction. then move to our main focus of low entanglement wavefunctions.. In the above case. Another class of compact wavefunctions is obtained by applying diagonal operators.a >b 0 ˆ i |φi = Q + NO( ). (10) WAVEFUNCTIONS AROUND THE HARTREE–FOCK LIMIT Wavefunctions built around the Hartree–Fock meanfield limit are widely used in quantum chemistry. each of which may form a suitable reference. for a ˆ = ˆ i .

For example. strong correlation and entanglement are thus not synonyms. where H ˆ is the full electronic HamiltoV nian of the problem. rather than strong correlation. If only a single-resonance structure is used. it seems more natural to build the wavefunction from the atomic states. The quality of the starting reference is governed by the magnitude of the perturbation ˆ =H ˆ −h ˆ . transforming to the valence bond basis does not by itself yield a compact wavefunction.) In molecular systems. November/December 2012 .78 (We refer to the BCS wavefunction also as a determinant for simplicity. strong correlation is commonly defined with respect to a perturbative partitioning of the Hamiltonian. we obtain a method known as generalized valence bond theory. Breaking spin symmetry yields unrestricted and general Hartree–Fock determinants. the wavefunction at large separation may be written as | = 2−1/2 (A[σg (1)ασg (2)β ] − A[σu (1)ασu (2)β ]). however. Equation (13) is also the starting point for valence bond (VB) theory. and the second is the pairing field associated with breaking particle number symmetry in the BCS wavefunction.89 Moving beyond a single-resonance structure theory in a scalable. the corresponding spin measurement on atom B will yield spin β . there are only spin couplings of the above form along specific chemical bonds. Determinant wavefunctions are eigenfunctions of one-particle effective Hamiltonians ˆ= h ij lustrated by the stretched two-electron bond in the hydrogen molecule. fully general.81 Despite the flexibility in the choice of starting reference. and the projected states are known as the antisymmetrized geminal power79. although it is a determinant of quasiparticles. ti j ai a j + † i j (ai a j + a j ai ). (12) where σ g and σ u are the bonding and antibonding molecular orbitals. Volume 2.82–86 as can multireference Jastrow wavefunctions. for three hydrogen atoms. in a molecule. This is not the case for large numbers of strongly correlated nature of the final state we wish to capture. there are three neutral doublet resonance structures. This suggests that compact wavefunction approximations may be constructed by encoding only the entanglement between local states to reflect the bonding network of the molecule.72. (6). the wavefunction is expanded in a valence bond basis. however. multireference approximations are compact only so long as the number of starting references remains polynomial in the system size. For small numbers of strongly correlated electrons. is characterized by ˆ relative to the energy spacing a large magnitude of V ˆ of h. as the basis is exponentially large. In VB theory. This type of strong correlation is known as entanglement. Using an excitation expansion such as Eq. commonly found in transition metal chemistry and excited electronic states.80 or (if they break spin symmetry also) Pfaffian wavefunctions. † † (11) (13) where the first term is a number conserving one particle term.88. Strongly correlated electronic structure. that is. not all atoms are bonded to all other atoms. if we measure spin α on atom A. The stretched bond wavefunction requires a multireference approximation as there is no dominant determinant. Perturbative. not electrons. and is associated with the chemical bond. However. proves quite difficult.wiley. Ltd. and compact way. For a compact parametrization. as illustrated in the next section). while breaking number symmetry leads to the Bardeen–Cooper–Schrieffer (BCS) wavefunction. one can augment the starting reference determinant by a linear combination of starting determinants. For example.Focus Article wires. we write | = 1 [1sa (1)1sb(2) + 1sa (2)1sb(1)] 2 × [αa (1)βb(2) − βb(1)αa (2)]. One of the attractions of entanglement as a descriptor. This gives rise to multireference wavefunction approximations. the BCS wavefunction is typically projected to restore the particle number symmetry. At large separation. LOW ENTANGLEMENT WAVEFUNCTIONS Valence Bond Theory The motivations for constructing wavefunctions by encoding entanglement between local objects are il- The coupling between the atomic spins reflects strong correlations. is the ability to define precise mathematical measures of entanglement. we have to limit the number of resonance structures in the expansion. Denoting the atomic orbitals as 1sa and 1sb . However. configuration interaction. We therefore turn our attention to a different class of wavefunction approximations that seems well suited to strong correlation problems: the low entanglement wavefunctions. corresponding to the classical resonance structures of chemistry. and coupled cluster theories from a multireference starting point can be formulated.90–92 Low entanglement wavefunctions change the focus from 910 c 2012 John Wiley & Sons. (Note that as discussed above. there are some situations when no good starting determinant can be found.87 However.

i2 of the matrices. We can now try to generalize the above to more than two local systems. 2 be spanned by sets of (orthonormal) states {|n1 }. Consider three systems 1. this gives | = n1 n2 . Vn2 i |n2 . there is one pair of entangled spins. yields the wavefunction coefficient n1 n2 n3 associated with the configuration |n1 n2 n3 . i2 . (22) σi2 log2 σi2 . We imagine entangling system 1 with system 2. as we can think of the index i (which we call an auxiliary index) as entangling the original states {|n1 }. An3 . We now turn to a mathematical discussion of entanglement. The natural generalization of Eq. {|n2 }. respectively. in the two systems. through |ri = n1 Un1 i |n1 . i (17) = (23) where {|ri } and {|li } denote the basis in which the entangling is performed. the matrices U and V define a transformation of the original bases.i (Here we work in a second quantized formulation. Relating this to the original basis |n1 n2 . 2 is expanded in the product space {|n1 n2 } as | = n1 n2 |li = In terms of the transformed basis. If we fix this at some dimension M. we say that the MPS is of dimension M. all singular values are the same. (21) ψn1 n2 |n1 n2 . and the entanglement entropy is log2 M. that is. {|n2 }. Any state in the combined system 1. respectively). The hydrogen molecule at infinite separation is in a maximally entangled state. (22) is to use two entangling auxiliary indices i1 . The maximally entangled state is thus written as | 1 =√ M M where we have for notational purposes absorbed the singular values in Eq. n1 n2 n3 |ri li .9–15 The accuracy and complexity of the MPS is controlled by the dimension of the auxiliary indices i1 . 2. is defined as follows: let systems 1. n2 (19) (20) Entanglement Bipartite entanglement. We define {|li } and {|ri } | spin = where in the last line we have switched to a matrix notation. (18) 2 we find that the entanglement entropy is log2 2 = 1. If the wavefunction is a product wavefunction. 911 . where fermion antisymmetry is expressed in the commutation relations between operators. Ltd. Since the wavefunction coefficient is obtained as a matrix product. and An1 and An3 are thin matrices (a row vector and a column vector. (14) Matrix Product States When constructing a wavefunction from the states of two local systems.93 or the entanglement between two systems. and the entanglement entropy is zero. This form is very suggestive. (15). The bipartite entanglement entropy is defined from a singular value decomposition of the coefficients ψn1 n2 . the wavefunction then assumes a diagonal form | = i σi |ri li . An2 . Considering the spin part of the wavefunction 1 [α (1)β (2) − β (1)α (2)]. From Eq. the wavefunction can always be written in the diagonal form | = i σ i |ri li . The entanglement entropy is S=− i Ain1 Ain2 |n1 n2 . and to write the wavefunction as | = n1 n2 n3 i 1 i 2 Ain11 Ain12i2 Ain23 |n1 n2 n3 An1 An2 An3 |n1 n2 n3 . and 3. (16) The entanglement entropy quantifies the correlations between the systems.WIREs Computational Molecular Science Low entanglement wavefunctions the bonds to the local states and provide a more practical framework for compact approximations. and then system 2 with system 3. Any wavefunction can be written in a diagonal form similar to the maximally entangled state. (15) into the definitions of the A matrices. In the maximally entangled state. there is only a single nonzero singular value. Writing this as a matrix ψ . then ψ = Uσ VT . this wavefunction approximation is known as a matrix product state (MPS). Volume 2.5–7. rather than by antisymmetrizing products of states). November/December 2012 c 2012 John Wiley & Sons. The multiplication of the matrices An1 . (15) where σ is the diagonal matrix of singular values.

e. . where we entangle them in a sequential manner. This is the terminology primarily used in the MPS and DMRG literature. however. |α . From Eq. However. the boundary of a division of the system can scale like the width L of the system. if we imagine dividing our k sites into two sets. irrespective of the number of sites in the problem. .5. (24) is thus an MPS wavefunction for k sites. thus entanglement entropy in a one-dimensional ground state is independent of system size. This structure means that correlation 912 c 2012 John Wiley & Sons. that is.Focus Article wires. However. however. . It is indeed possible to devise states with more general entanglement structures than the MPS.and three-dimensional systems is a bad idea. using the fact that the E matrices have an outer product structure. one finds that expectation values of local operators can also be evaluated in O( M3 k) time. Note that the MPS construction is general and does not depend on the definition of the local system or the nature of the constituent states |n . This is a clear improvement over full configuration interaction. crucially. . spanned by the four states. (25) is reminiscent of a transfer matrix expression in statistical mechanics. This may suggest that using an MPS to simulate two. (24) where n denotes the configuration |n1 n2 . M ∼ e L. ni where the E matrices are of dimension M2 × M2 . we may take each system to be an atom (in which case the local states span the Fock space of each atom. this expresses locality. therefore. more naturally expect the true entropy to capture in a physical problem to scale with the size of the problem. If the dimension of the auxiliary indices is M. We might. MPS offers significant computational advantages when simulating systems even with two. Ei = (26) The evaluation of the norm is itself a matrix product. in a two. the maximum entanglement entropy log2 M is independent of system size. .94–96 Whereas thermodynamic entropy is an extensive quantity (i. For example. Ltd.or three-dimensional connectivities. where by Fock space we mean the full Hilbert space across all particle number sectors). | = n An1 An2 . For example. and thus the exponential cost is associated only with the remaining dimensions. If we apply an MPS to describe a two or higher-dimensional system. (25). which naturally capture the area law in higher dimensions and such states are discussed in section General Tensor Network States. we now introduce the term site as a synonym for local system. entanglement entropy is believed to scale only with the size (i. One way to analyze the accuracy of the MPS approximation is to consider how much (bipartite) entanglement entropy can be described by such a state. . and dimensionality is given rigorously by so-called area laws. area) of the boundary between the two systems used to define the entropy.. As we argued above. At an intuitive level. |β . the matrix dimensions must grow exponentially with the boundary area.or higher-dimensional system.15 We illustrate this with the computation of the norm. and thus the entanglement entropy scales as the system width (see Figure 2). In practice. it is simply proportional to volume). |αβ . Ek. .e. that is.6. .com/wcms We can generalize the MPS to k systems. November/December 2012 . Ani † ⊗ Ani . where the E matrices are the transfer operators. to capture the growing entanglement entropy. | − . Given local interactions in the Hamiltonian. the MPS captures the correct structure of entanglement for one of the spatial dimensions. then in the ground state.6 The expectation value structure in Eq..97 We now discuss a few miscellaneous topics related to MPS: algorithms. MPS Algorithms MPSs are compatible with many efficient computational algorithms. however. Volume 2. Or. any division of the system yields a boundary size independent of system size. an area. and we see that this is clearly appropriate to a one-dimensional system ground state. overlaps and expectation values can be obtained efficiently. The link between entanglement entropy. Ank † ) ⊗ (An1 An2 . it would appear to take O( M4 k) time. it requires only O( M3 k) time. we could take each system to be as small as a single orbital.wiley. The wavefunction approximation in Eq. where the cost scales exponentially with the system volume. The boundary size depends on the overall system size in a dimensionally dependent way. Ank ) (25) = E1 E2 . In one dimension. The norm of a matrix product state is | = n (An1 † An2 † . their interpretation as projected wavefunctions. then the maximum entanglement expressed by the MPS for any cut between the sites is log2 M. . for an MPS of dimension M. Ank |n . and thus the entropy scales with boundary size. To emphasize this generality. and finally their graphical representation. only quantum degrees of freedom along the boundary can be entangled with each other.nk . By a similar construction. system size. .

Graphically. we obtain a Schrodinger equation for each component. we see the DMRG as an algorithm which works within the space of MPS to variationally minimize the energy with respect to a single component An at a time.12 As M (and thus the dimension of E) is increased. we can introduce auxiliary states |ri li . .6. thus capturing longer range correlations. which is a mean-field theory for independent electrons. the product over all the matrices yields a scalar. where H is an effective Hamiltonian and a denotes An flattened into a vector form.14 Unlike Hartree–Fock theory. (32) × i2 |ri2 li2 The view of the MPS as a set of commuting operators acting on a simple product reference function formally connects the MPS construction to other kinds of wavefunctions studied in quantum chemistry such as the coupled cluster wavefunction (which is a product of a set of commuting excitation operators on a single-determinant reference) although this analogy has yet to be much explored. the wavefunction coefficient n1 n2 .15 In other contexts. A ˆ2 where we have defined projection operators A as ˆ1 = A in1 Ain1 |n1 ri |. it is possible to reproduce smaller and smaller λ. MPS arose originally in conjunction with the DMRG.14 ˆ 1. In this language. and the sweep algorithm of the DMRG is then analogous to the traditional self-consistent field algorithm of Hartree. λ is related to the smallest eigenvalue of the transfer operator.. ¨ H · a = E a. In this way.i ⎞ |rik−1 lik−1 ⎠ . (27) F I G U R E 3 | Graphical representation of matrix product states. ip−1 and ip . A matrix An is represented by an inverted T-shaped object with lines going left and right (the two auxiliary indices).. The indices of the matrix. Formally.. We then find that the MPS can be written as a set of projections | ˆ2 ˆ1 A =A i |ri li . November/December 2012 c 2012 John Wiley & Sons. where each orbital is determined sequentially. 913 . Ain1 Ain2 |n1 n2 . MPSs represent a mean-field theory for each of the local systems or sites. we contract over all the auxiliary indices in the product An1 An2 Ank . the contraction is represented by joining together the auxiliary index lines. By minimizing the variational MPS energy with respect to the individual matrix components An . Q ˆ j act on sites ˆ j (where Q functions such as Q i and j. there is a matrix An p . row and column vectors.98 Consider. and the effective Fock operator is updated one orbital at a time until self-consistency is reached. The (matrix) product structure of MPS means that they are a generalization of a mean-field theory. As described in Introduction. This is analogous to the orbital equation in Hartree–Fock theory.WIREs Computational Molecular Science Low entanglement wavefunctions ˆiQ ˆ i. (29) Volume 2. .10.. in2 (30) (31) ˆ2 = A The state that is being projected from is a maximally entangled state of auxiliary particles i |ri li . the idea of introducing fictitious particles to describe a state is sometimes known as using a shadow wavefunction. the MPS arises as a projection from a reference function. but are fictitious or auxiliary particles.nk . the reference for an MPS for k sites can be viewed as a simple product of entangled states (one set for each auxiliary index) operated on by a set of commuting projection operators | ˆ1 A ˆk ˆ2 . the MPS for two sites | = n1 n2 . Ltd. From the viewpoint of the current discussion.. (Right) To obtain the matrix product state approximation to the wavefunction coefficient n1 n2 . The particles appearing in the reference are not the actual electrons. are referred to as auxiliary indices and are of dimension M. Note that the leftmost and rightmost matrices are.nk .13. (28) Graphical Representation The proliferation of indices in the MPS means that it is often more convenient to use a graphical rather than algebraic representation. A =A i1 |ri1 li1 ⎛ . Ain1 |n2 li |.. In general.⎝ i k−1 Projected Wavefunctions There is another way to view MPS that is often useful and which connects to some traditional quantum chemistry ideas. respectively) decay in MPS exponentially like exp ( − λ|i − j|). The equations solved at each step of the DMRG are the effective Schrodinger equations ¨ (27).6. (Left) For every state of a site |np .6 The graphical representation of an MPS is shown in Figure 3. in practice.

and the MERA. expectation values can be efficiently obtained. The graphical representation clearly shows the nature of the index contractions. A second advantage of the TTN is that.54. This becomes even more essential when considering states with a more general entanglement structure than the MPSs. as opposed to exponentially in the case of MPS.60 which extend MPSs and maintain a similar computational efficiency. which allows entangling to the left and right neighbors. Ltd. denoting summation over all the state indices. row and column vectors. to which we now F I G U R E 5 | (Left) For every local state |np . The graphical representation is useful not only for representing the states. which decay algebraically with respect to site separation. a tree tensor network F I G U R E 4 | (Bottom) A norm can be represented graphically by fusing the vertical indices of the matrix product state (MPS) bra (pointing downward) with the vertical indices of the MPS ket (pointing upward).nk . and many questions remain to be answered. shown in Figure 5. the ability of TTN to better represent long-range 914 c 2012 John Wiley & Sons.Focus Article wires. in practice. The TTN state can be written as | = n z i i An zi |n .55 which are a very general way to encode a network of entanglement. The computation of an overlap is shown in Figure 4. and z denotes the contraction with the indices of neighboring tensors according to the connectivity of the tree. One advantage of the TTN over the MPS is that the maximum distance between two sites in a tree is logarithmic in the total number of sites. n2 . as described in section MPS Algorithms. (Top) Performing a state summation at each site leads to an intermedate matrix E of dimension M2 . and the norm evaluation is a product of the E matrices. In the generalization to a tree structure. This allows the TTN to describe states with correlation functions. November/December 2012 . n1 . more general encodings can be explored. a local system is represented by a matrix An with two auxiliary indices. but which still restrict the structure of the entanglement.. (Note that the leftmost and rightmost matrices are. projected entangled pair states (PEPS). but also operators and expectation values. While the scaling of a z > 2 TTN computation is higher than that for a corresponding MPS (z = 2). This is a rapidly expanding area of investigation.wiley. .61–64 which combines some of the practical strengths of the TTN while achieving some of the formal advantages of PEPS. which allows couplings to z neighbors. Lines that are connected represent indices that are summed over.57.52. (33) General Tensor Network States The limitation of MPS is that they can only efficiently encode a sequential structure of entanglement. just as in the MPS.) and a vertical line (the site index n).nk is obtained by contracting the auxiliary indices according to the tree connectivity Tree Tensor Networks In an MPS.. The accuracy of the TTN is controlled by the dimensions of the auxiliary indices z.58. . as opposed to linear in the case of the MPS.15. a local system is represented by a tensor with z auxiliary indices. (TTN) associates a tensor with z auxiliary indices. we see that the cost for calculating a norm is O(M)z+1 k. (Right) The wavefunction coefficient n1 n2 . For arbitrary systems. Here we discuss three promising kinds of tensor networks: TTN.. Volume 2. where zi denotes the set of z indices on each tensor. which we can denote by M. By contracting the tensors from out to in. Here shown is a TTN where each tensor has three auxiliary indices.

Ltd. mally entangled states of auxiliary particles. The PEPS approximation for the wavefunction coefficient n1 n2 . we form intermediates where the number of dangling bonds is proportional to the system width. described in section Projected Wavefunctions for MPSs. r.WIREs Computational Molecular Science Low entanglement wavefunctions F I G U R E 6 | Projected entangled pair states (PEPS) are associated with arbitrary networks of entanglement. (Top) Each contraction of the local states |np results in an E tensor with four auxiliary indices. correlations means that we expect a TTN will reach the accuracy of a given MPS calculation with smaller auxiliary dimension of the tensors. it appears to be possible to approximate these intermediates by simpler forms. a two column contracted intermediate can be accurately approximated by a single-column intermediate with slightly larger auxiliary dimension (Figure 8). In the PEPS. where Ezz = n An z Az . To illustrate the PEPS construction. The name PEPS arises from the projected wavefunction interpretation of the state. F I G U R E 7 | (Bottom) Projected entangled pair states (PEPS) norm evaluation can be represented by fusing a PEPS bra lattice (pointing downward) with a PEPS ket lattice (pointing upward). where z denotes the four auxiliary indices l. d that point along the bonds of the square lattice. Every site is surrounded by four “bonds” on which we place maxi¯ . However. and down.. r. one proceeds through intermediates where the number of dangling bonds is independent of system size. In this notation. However. d that point toward the neighbors in the directions left. The cost of forming these intermediates. except the auxiliary indices are contracted according to the ‘bonds’ of the square lattice (see Figure 6). Shown here is a representation of the PEPS for a square lattice of sites. As we contract column by column. Preliminary investigations appear to confirm this. This is why these states are compatible with very simple and efficient algorithms for calculating norms and expectation values. States with arbitrary networks of entanglement on the other hand. shown in Figure 4 for the MPS. and thus the cost of the contraction. for physical systems. they are expected to provide compact descriptions of the ground states of almost arbitrary physical systems. For the square lattice PEPS above.60 physical particles. m |ηη and each site is associated with a projector operator which projects from the auxiliary particles onto the (34) The whole PEPS wavefunction is the product of such commuting projection operators (associated with the sites) on a reference. right. with no cycles between the tensors in the wavefunctions. which is a product of entangled pairs on the bonds. grows exponentially with system width. As one contracts the expectation value network. Consequently. as shown n in Figure 7.nk is obtained by contracting all the auxiliary indices. u. the evaluation of an expectation value requires contracting a square lattice network of E-type tensors. the PEPS wavefunction takes exactly the same form as the TTN (33). each site is associated with a tensor An z . the ability to represent the wavefunction compactly does not guarantee that computations of observables are efficient. we consider an appropriate state to describe a square lattice of sites. ˆ= A nlr ud n Alr ud |n ηl ηr ηu ηd |. Each local state |np is then associated with a tensor with four auxiliary indices l. are known as PEPS. This means that an iterative approximation algorithm can be constructed where the evaluation of an Volume 2. Projected Entangled Pair States The TTN and the MPS possess a special connectivity. November/December 2012 c 2012 John Wiley & Sons. 915 . The result is a square lattice which must be contracted over the auxiliary indices. u. up. For example. PEPS wavefunctions can be constructed to satisfy the correct entanglement area laws in two and three dimensions.. including cycles. each of dimension M2 .

and thus neglects some entanglement that may be present in the ground state.n1 n2 n3 even more coarse-grained site {|n1 n2 n3 } → {|n1 }. we take each block to contain three sites. The final top-to-bottom contraction involves only a single column and is of the same complexity as an MPS norm evaluation. which perform a unitary rotation of the basis of two or more sites. observable reduces finally to a single-column contraction. which first handle the entanglement between the boundaries of the blocks. MERAs consist of a set of operators known as isometries. W ˆ l) ˆ 1U = (W n |n . F I G U R E 8 | Naive contraction of the square lattice of E tensors that arises from the projected entangled pair states (PEPS) norm evaluation is of exponential cost. which perform coarse graining. Volume 2. in the above. the coarse graining can be iterated. we obtain a final effective site with an M-dimensional Hilbert space. we would use a unitary operator which acts on the product space of sites 3 and 4. For concreteness. In a similar way. r2 . we can contract the full PEPS from left to right efficiently. alternating with layers of isometries. (36) |n1 Wn11n2 n3 n1 n2 n3 |.61 For example. thus giving a lattice with k/3 effective sites. The MERA consists of a layer of disentanglers. and disentanglers. here r1 . For simplicity. in which the ground state of the entire problem is expressed. we can use another isometry to map a block of three coarse-grained sites to a single. After log3 k levels of coarse graining. the number of layers in the MERA. We divide the k sites under consideration into blocks. r4 ) as the local state indices of a fictitious wavefunction F r 1 r 2 r 3 r 4 . Second. and disentanglers U. The operator which performs this projection is an isometry. Ltd. which handle the entanglement within the blocks. we describe the MERA in one dimension. Once all the blocks have been coarse grained. ˆ = W n1 . | ˆ1 W ˆ 2U ˆ2 . This effectively approximates the two column structure by a single-column structure. The MERA possesses several unique strengths. November/December 2012 . The MERA structure is illustrated in Figure 10. although analogous constructions can be considered in two and three dimensions. which are unitary rotations. The iterated isometries are connected in a tree structure (of different nature to the TTN deˆ scribed above) and the coefficient of the isometries W are the variational parameters of the state. but do not between sites 3 and 4. we consider first a contraction of two E columns. The idea behind the MERA is that one can incorporate this missing entanglement between the block boundaries. The first block spans the space {|n1 n2 n3 }. We can imagine approximating such a wavefunction by a matrix product state (MPS) of a small auxiliary dimension.wiley. the close relationship between the MERA and a tree structure means that unlike the 916 c 2012 John Wiley & Sons. each with an Mdimensional Hilbert space. Instead.Focus Article wires. where l is log3 k. We first discuss the isometries. We then view the dangling bonds (pointing toward the right. . {n2 } build in entanglement within the first and second blocks of three sites in the ground state. the effective coarse-grained states in the first two blocks {|n1 }.56 F I G U R E 9 | The multiscale entanglement renormalization ansatz is constructed from two types of tensors: isometries W. n (35) as illustrated in Figure 9. which map the states of a block of sites to a single effective sites. starting from the left edge. which are on the boundaries of the blocks. in the above example. Ar 1 Ar 2 Ar 3 Ar 4 . the presence of disentanglers allows the MERA to correctly describe the entanglement area laws in any dimension. even including the corrections to area laws that are present in gapless one-dimensional systems. Multiscale Entanglement Renormalization Ansatz The MERA is a flexible entanglement structure that is closely related to the original real space renormalization group used in statistical problems. Repeating this for the remaining columns of the E-square lattice. The problem with using only isometries in a tree structure is that the coarse graining does not treat the sites uniformly. r3 . For example. we coarse grain all the three site blocks onto effective sites.55. We now coarse grain from the Hilbert space of the three site block {|n1 n2 n3 } to a single effective site with a reduced Hilbert space {|n1 } of specified dimension M. For example. . as in the case of an MPS observable evaluation. by applying a unitary operator to these sites.

Density matrix renormalization group: introduction from Volume 2. 6. The density-matrix renormalization ¨ group in the age of matrix product states. REFERENCES 1. in addition to MPSs. 2. Dukelsky J. 43:457–462. we have also described more general low entanglement approximations. Hachmann J. Maruani J. The density-matrix renormalization ¨ group. Chan GK-L. Frontiers in Quantum Systems in Chemistry and Physics. 4. Equivalence of the variational matrix product method and the density matrix renormalization group applied to spin chains. Sierra G. we have described the basic theory behind low entanglement wavefunction approximations. 64:4084– 4087. Schollwock U. Sharma S. These states are considerably more complex than MPSs and their properties are not yet completely understood. In: Wilson S. Piecuch P. This has been used to describe large multireference. Martin-Delgado MA. Density-matrix algorithms for quantum renormalization groups. (Right) This is rectified within the MERA by first performing a disentangling operation U between the block boundaries before coarse graining. 917 . White SR. expectation values and observables can be exactly computed in polynomial time. In recent years. Progress in Theoretical Chemistry and Physics. This is an area where we can expect the wealth of knowledge in quantum chemistry in computational algorithms for complex wavefunctions to play a major part. 326:96–213. A key remaining challenge is the development of practical numerical algorithms to work with these flexible states. PEPS. Hieida Y. An introduction to the density matrix renormalization group ansatz in quantum chemistry. Schollwock U. Marti KH. Product wave function renormalization group. 9. Neuscamman E. they offer the intriguing possibility to efficiently model chemical problems that challenge standard quantum chemistry methods. Ghosh D. 3. and MERA. Vol. Okunishi K. 55:477–526. eds. Europhys Lett 1998. the coarse grainings within the blocks neglect to take into account entanglement at the boundaries of the blocks. The density matrix renormalization group algorithm in quantum chemistry. 18. Because low entanglement wavefunctions occupy a region of Hilbert space far from the Hartree– Fock mean-field limit. practical evidence of this has been provided by the DMRG algorithm (based on low entanglement wavefunctions known as MPSs). 69:2863– 2866. Hallberg KA. Numerical computation with general low entanglement approximations is thus positioned to be an exciting possibility for quantum chemistry to explore in this coming decade. Phys Rev Lett 1992. Nonetheless. 8. it is clear at the mathematical level that they overcome many of the formal limitations of MPSs. White SR. the computational effort of using the MERA remains high. Yanai T. ACKNOWLEDGMENTS The author acknowledges support from the NSF programs CHE-0645380 and CHE-1004603. Okunishi K. Dorando JJ. Nishino T. Springer Netherlands. such as TTNs. particularly in relation to the encoding of entanglement as a function of dimensionality. Ann Phys 2010. 224:583–599. 5. Wang H. 49– 65. However. Adv Phys 2006. Delgado-Barrio G. New trends in density matrix renormalization. Ltd. Z Phys Chem 2010. W). Annu Rev Phys Chem 2011. In our discussion. Density matrix formulation for quantum renormalization groups. Nishino T. November/December 2012 c 2012 John Wiley & Sons. Rev Mod Phys 2005. strong correlation problems involving transition metals and excited states. Phys Rev B 1993. Dordrecht. 11. The Netherlands. 48:10345– 10356. 77:259–315. Reiher M. Grout PJ. where a layer of sites is subjected to successive coarse grainings (isometries. J Phys Soc Jpn 1995. The density matrix renormalization group in quantum chemistry. 2008. Nishino T. and applications remain at a very early stage of development. CONCLUSIONS In this review. 10. 7. Hikihara T. 62:465–481.WIREs Computational Molecular Science Low entanglement wavefunctions F I G U R E 1 0 | (Left) The original real space RG is associated with a tree structure for the wavefunction.61 Nonetheless. Chan GK-L. PEPS.

Porras D. Xiang T. 18. Optimizing the density-matrix ´ renormalization group method using quantum information entropy. 130:234114. ¨ Solyom Legeza O. 121:6110–6116. State-of-the-art den´ sity matrix renormalization group and coupled cluster theory studies of the nitrogen binding curve. 37. 122:204101. 26. Phys Chem Chem Phys 2008. J. J Chem Phys 2003. 15. 28. Luo H-G. J Chem Phys 2006. 13:1–26. Highly correlated calculations with a polynomial cost algorithm: a study of the density matrix renormalization group. Adv Phys 2008. Vidal G. ¨ Rommer S. Ciofini I. 36. Martin RL. J Chem Phys 2002. Marti KH. Yanai T. 128:144117. J Chem Phys 2007. 19. ¨ Solyom 33. J Chem Phys 2005. J Chem Phys 2007. Palmieri P. Chan GK-L. 55:2164–2181. Int J Mod Phys B 1999. 118:8551–8554. Legeza O. 23. Fano G. Density matrix renormalization group and periodic boundary conditions: a quantum information perspective. Relativistic DMRG calculations on the curve crossing of cesium hydride. 20. Quantum chemistry using the density matrix renormalization 12. 93:40502. Ghosh D. Head-Gordon M. Hachmann J. 13. J Chem Phys 2007. Density matrix renormalisation group Lagrangians. with applications to polyenes and β -carotene. Palmieri P. 29.wiley. ¨ Roder 32. Hachmann J. Phys Rev B 2003. Phys Rev B 2004. 101:2019–2028. J. Legeza O. Moritz G. 110:4127–4130. 128:014104. Moritz G. Chan GK-L. 41. Reiher M. projected entangled pair states. Chan GK-L. Chan GK-L. Density matrix renormalization group calculations on relative energies of transition metal complexes and clusters. J Chem Phys 2009. Avil´ es M. Orbital optimization in the density matrix renormalization group. 81:235129. 93:227205. 127: 084109. Chan GK-L. An algorithm for large scale density matrix renormalization group calculations. Dorando JJ. Gauss J. 39. 34. Reiher M. Murg V. 124:034103. Kurashige Y. Exact solution (within a triple-zeta. J Chem Phys 2004. J Chem Phys 2009. 17. 115:6815–6821. Phys Rev B 2010. 27. High-performance ab initio density matrix renormalization group method: applicability to large-scale multireference problems for metal compounds. Mitrushenkov AO. Verstraete F. double polarization basis set) of the electronic Schrodinger equation for water. 123:184105. Hess BA. Chan GK-L. Convergence behavior of the density-matrix renormalization group algorithm for optimized orbital orderings. Ond´ ık IM. 119: 4148–4158. Analytic theory of response in the density matrix renormalization group. Moritz G. Class of ansatz wave functions for one-dimensional spin systems and their relation to the density matrix renormalization group. Daul C. Linguerri R. Fano G. Phys Rev B 1997. Chan GK-L. 40. Linguerri R. ¨ Roder 31. Volume 2. Chan GK-L. 24. 120:3172–3178. Controlling the accuracy ¨ of the density-matrix renormalization-group method: the dynamical block state selection approach. J. Density-matrix renormalization-group algorithms with nonorthogonal orbitals and non-Hermitian operators. Cardoen W. and the densitymatrix renormalization-group method. 67:125114. J Chem Phys 1999. Hachmann J. Int J Quantum Chem 2000. Mol Phys 2003. 116:4462–4476. Van Voorhis T. Reiher M. Ghosh D. Verstraete F. J Chem Phys 2008. Mitrushenkov AO. Chan GK-L. 42. 122:024107. 57:143–224. Dorando JJ. Hachmann J. Phys Rev Lett 2004. Ab initio quantum chemistry using the density matrix renormalization group. Reiher M. Ltd. Daul S. Wolf A. Yanai T. White SR. Reiher M. a variational point of view. ´ quantum information generation. Accelerating convergence in iterative solutions of large 918 c 2012 John Wiley & Sons. 126:244109. The radical character of the acenes: a density matrix renormalization group study. Hess BA. 79:331–342. II. Kurashige Y. Yanai T. 130:184111. Qin M-P. Phys Rev Lett 2004. Ostlund S. Full-CI quantum chemistry using the density matrix renormalization group. Hachmann J. 125:144101. J Chem Phys ¨ 2003. Dorando JJ. 25. QC-DMRG study of the ¨ ionic-neutral curve crossing of LiF. J Chem Phys 2001. Targeted excited state algorithms. J. J Chem Phys 2004. Multireference correlation in long molecules with the quadratic scaling density matrix renormalization group. Optimizing Hartree– Fock orbitals by the density-matrix renormalization group. Moritz G. Chan GK-L. 127:134309. 14. 35. 10:3454– 3459. Cirac JI. 16. Ortolani F. White SR. Chan GK-L. 70:205118. 38. November/December 2012 . Quantum chemistry using the density matrix renormalization group. Hess BA. Quantum data compression. Chan GK-L. 30. Construction of environment states in quantum-chemical density-matrix renormalization group calculations. Head-Gordon M. Legeza O. J Chem Phys 2005. Matrix product states. J Chem Phys 2005. Cirac JI. 21. Efficient simulation of one-dimensional quantum many-body systems. 68:195116. Moritz G. and variational renormalization group methods for quantum spin systems. Phys Rev B 2003. Kallay M. J Chem Phys 2006. J Chem Phys 2008. Decomposition of density matrix renormalization group states into a slater determinant basis.Focus Article wires. and applications to polyenes. 22.

J Chem Phys 2008. Optimized trial wave functions for quantum Monte Carlo calculations. Marti KH. Harrison RJ. 57. Kraus CV. Phys Rev Lett 2007. J Chem Phys 2008. Neuscamman E. 60:1719–1722. 66. Strongly contracted canonical transformation theory. 79:291–352. 102:180406. Fann GI. 73. 45. 44. Modern Quantum Chemistry. Schmidt MW. 12:103008. Coupled-cluster theory in quantum chemistry. 71:1253–1266. Kutzelnigg W. UK: John Wiley & Sons. 72. Maeshima N. Annu Rev Phys Chem 1981. 81:052338. Yanai T. Int J Quantum Chem.1651. 61. Gendiar A. 58. Zgid D. Bauer B. Phys. Vidal G. Entanglement renormalization. Marti KH.WIREs Computational Molecular Science Low entanglement wavefunctions 43. Bartlett RJ. Phys Chem Chem Phys 2011. Beylkin G. Nishino T. Self-consistent tensor product variational approximation for 3D classical models. Zgid D. 75. Wilkins JW. NY: Dover Publications. Reiher M. 51:447–463. Obtaining the two-body density matrix in the density matrix renormalization group method. Schuch N. Arxiv preprint arXiv:0912. Hieida Y. Noack RM. J Chem Phys 1979. Neuscamman E. 575:504–512. Molecular Electronic Structure Theory. Nooijen M. Vidal G. Pahl FA. Troyer M. 80:235127. Phys Rev B 2009. 1955. Hieida Y. Vidal G. Szabo A. Nucl Phys B 2000. Cirac JI. information analysis of electronic states of different molecular structures. Kurashige Y. 128:144115. Akutsu Y. 101:110501. ¨ Legeza. 99:220405. Chan GK-L. 76. 919 . Bartlett RJ. 68. 1994. 71:3951–3962. 63. Entanglement renormalization in two spatial dimensions. 75:033605. 2009. Verstraete F. 65. Theory of the expansion of wavefunctions in a gaussian basis. Vidal G. J Chem Phys 2010. 69. Yanai T. Okunishi K. Nooijen M. J Chem Phys 2010. 100:3199– 3224. Evenbly G. 64:016705. New J Phys 2010. Classical simulation of quantum many-body systems with a tree tensor network. Okunishi K. Reiher M. Vidal G. Kohn W. Mol Phys 2002. Tagliacozzo L. Duan L. Simulation of twodimensional quantum systems using a tree tensor network that exploits the entropic area law. Many-body perturbation theory and coupled cluster theory for electron correlation in molecules. Chichester. Umrigar CJ. Vidal G. Two-dimensional tensor product variational formulation. Cirac JI. Ruedenberg KE. Nishino T. QuantumBarcza G. Rev Mod Phys 1999. 98:1479–1484. 53. Phys Rev Lett 1988. Phys Rev B 2010. Phys Rev Lett 2009. 128:014107. Vertical density matrix algorithm: a higher-dimensional numerical renormalization scheme based on the tensor product state ansatz. 1996. O. Nobel lecture: Electronic structure of matter wave functions and density functionals. 54. Phys Rev A 2007. Handy NC. Shi Y. Nishino T. Simu60. J Chem Phys 2008. November/December 2012 c 2012 John Wiley & Sons. Complete-graph tensor network states: a new fermionic wave function ansatz for molecules. Rev Mod Phys 2007. Murg V. Vidal G. Musial M. Ltd. Phys Rev B 2009. 132:024105. Reiher M. J Chem Phys 2004. Entanglement renormalization: an introduction. 50. 70. Evenbly G. 80:165129. 47. 105:409–417. 56. Prog Theor Phys 2001. Akutsu Y. Okunishi K. 59. Cirac JI. Phys Rev E 2001. 74:22320. Variational study of hard-core bosons in a two-dimensional optical lattice using projected entangled pair states. Kurashige Y. Yanai T. Helgaker T. Verstraete F. 132:024106. Fermionic projected entangled pair states. Multiresolution quantum chemistry: basic theory and initial applications. Verstraete F. On the spin and symmetry adaptation of the density matrix renormalization group method. 74. active-space self-consistent field calculations. Volume 2. Jorgensen P. Mizukami W. 71. Phys Rev Lett 2008. New electron correlation theories for transition metal chemistry. 109:2178–2190. Olsen J. The density matrix renormalization group self-consistent field method: orbital optimization with the density matrix renormalization group method in the active space. Jastrow R. Gan Z. Nooijen M. 82:205105. J Chem Phys 2010. Int J Quantum Chem 2009. Verstraete F. Phys Rev A 2010. Effective convergence to complete orbital bases and to the atomic Hartree– Fock limit through systematic sequences of gaussian primitives. 52. Verstraete F. 32:359–401. Rev. Zgid D. 62. Ostlund NS. Legeza O. ¨ Marti KH. 128:144116. Chan GK-L. 49. Maeshima N. 2000. Novel quantum states of electron spins in polycarbenes from ab initio density matrix renormalization group calculations. Hieida Y. 67. Akutsu Y. 83:012508. Phys Rev A 2011. Plane waves and radial polynomials: a new mixed basis. Mineola. Corboz P. Murg V. Arxiv preprint cond-mat/0407066. 2004. Multireference quantum chemistry through a joint density matrix renormalization group and canonical transformation theory. 133:091101. Wilson KG. Phys Rev A 2006. 48. 55. 46. lating strongly correlated quantum systems with tree tensor networks. Fermionic multiscale entanglement renormalization ansatz. Many-body problem with strong forces. Renormalization algorithms for quantum-many body systems in two and higher dimensions. 121:11587– 11598. Yanai T. Maeshima N. 13:6750–6759. Class of quantum many-body states that can be efficiently simulated. 51. 64.

85. A survey of recent developments in ab initio valence bond theory. Voter AF. Bajdich M. Goddard III WA. A method for describing resonance between generalized valence bond wavefunctions. Shadow wave function for many-boson systems. 123:084102. Proc R Soc London A: Math Phys Sci 1969. Luchow A. Drobn G. Calculation of optimal generalized antisymmetrized geminal-power functions and their associated excitation spectrum. 92. Mukherjee D. Stoudenmire EM. Schmidt KE. Masserini GL. An exponential multireference wavefunction ansatz. Weiner B. Adv Quantum Chem 1998. 90:2714–2731. 115:7814–7821. Structure of fermion density matrices. Antisymmetrized geminal powers. Chem Rev 1991. 89. The determination of energies and wavefunctions with full electronic correlation. Reatto L. Schaefer III HF. November/December 2012 . Plenio MB. Ripka G. An area law for one-dimensional quantum systems. 57:253–259. Audretsch J. 96:130201. Coleman AJ. Jeziorski B. 90. Colloquium: area laws for the entanglement entropy. 30:163–193. 98.Focus Article wires. 83. 91:929–964. Connections between coupled cluster and generalized valence bond theories. Goscinski O. J Math Phys 1965. 28:137–151. Handy NC. Entropy and entanglement in quantum ground states. Rev Mod Phys 2010. State-specific multi-reference coupled cluster formulations: two paradigms. Hiberty PC. 76:035114. Volume 2. Hunt WJ. J Stat Mech 2007. White SR. Head-Gordon M. 79. 2007:08024. Evangelista FA. Bandyopadhyay B. 22:2374–2391. Blaizot JP. Raimondi M. John Wiley & Sons. Quantum Theory of Finite Systems. 310:43–61. 128:124104. Ann Rev Cond Mat Phys 2011. 6:1425–1431. 80. Hastings MB. 95. 6:368–376. Generalized valence bond description of bonding in low-lying states of molecules. 920 c 2012 John Wiley & Sons. 84. Phys Rev Lett 2006. Comput Mol Sci 2011. An orbital-invariant internally contracted multireference coupled cluster approach. J Chem Phys 2008. 134:114102. Hay PJ. 82. Van Voorhis T. J Comput Chem 2007. 94. 87. 81. Allen WD. 82:277–306. Cambridge. 1:388–402. Evangelista FA. 38: 4516–4522. Gerratt J. 3:111– 128. Dunning Jr TH. 96. Shaik S. Studying two dimensional systems with the density matrix renormalization group. Ltd. 2007. Gauss J.wiley. Phys Rev B 2007. Simmonett AC. 97. MA: MIT 77. Cramer M. Acc Chem Res 1973. J Chem Phys 1989. J Chem Phys 2001. 1986. Gauss J. J Chem Phys 2011. Paldus J. Boys SF. Phys Rev B 1988. 93. 86. 88. Chem Phys 1981. Hastings MB. Cooper DL. 91. Triple excitations in state-specific multireference coupled cluster theory: Application of mkmrcc sdt and mk-mrccsdt-n methods to model systems. Applications of spin-coupled valence bond theory. Mahapatra US. Wagner LK. Phys Rev A 1980. Eisert J. Entangled Systems: New Directions in Quantum Physics. J Chem Phys 2005. Mitas L. Valence universal exponential ansatz and the cluster structure of multireference configuration interaction wave function. Goddard III WA. II. Pfaffian pairing wave functions in electronicstructure quantum monte carlo simulations. 78. Hanrath M. Datta B. Quantum monte carlo methods.

Sign up to vote on this title
UsefulNot useful