Additional Tricks

This page is a collection and summary of some generic improvements without any claim to completeness. These are essentially tricks which are relatively independent of the actual time-evolution method and can mostly be implemented in all of the methods we just reviewed. They are not strictly necessary to implement in conjunction with any method and as such will not be benchmarked in detail later, but they are useful to keep in mind in case of particularly hard or challenging problems.

Combining Heisenberg and Schrödinger picture time evolutionn

In the context of MPS methods, combining Schrödinger and Heisenberg picture[1][2] time evolution was first proposed in Ref. 3. Considering a time-dependent observable

$$\begin{equation} \langle \phi | \hat O(t) | \psi \rangle \end{equation}$$

between two arbitrary states, in the Schrödinger picture we would evaluate

$$\begin{equation} \langle \phi | \hat O(t) | \psi \rangle = \left( \langle \phi | e^{\mathrm{i} t \hat H} \right) \hat O \left( e^{-\mathrm{i} t \hat H} | \psi\rangle \right) \quad. \end{equation}$$

That is, we apply time-evolution operators to the states $|\phi\rangle$ and $|\psi\rangle$ to obtain time-evolved states $|\phi(t)\rangle$ and $|\psi(t)\rangle$ and then evaluate the time-independent observable between them. The maximum time $t$ obtainable is then typically limited by the entanglement growth in the states, resulting in larger and larger bond dimensions or errors.

In comparison, the Heisenberg picture would see us time evolve the operator $\hat O$ as

$$\begin{equation} \langle \phi | \hat O(t) | \psi \rangle = \langle \phi | \left( e^{\mathrm{i} t \hat H} \hat O e^{-\mathrm{i}t \hat H} \right) | \psi\rangle \end{equation}$$

while keeping the states $|\phi\rangle$, $|\psi\rangle$ static. Again, the maximal obtainable time t is limited by the entanglement growth in the operator $\hat O$ and the maximal bond dimension we can use to represent it. (Note that it is difficult to compare errors between MPS and MPO truncations, as the error of the MPO truncation is given by the operator norm whereas during an MPO compression, we only control the 2-norm of the operator[4].) If we now combine the two evolutions as

$$\begin{equation} \langle \phi | \hat O(t_1+t_2) | \psi \rangle = \left( \langle \phi | e^{\mathrm{i} t_1 \hat H} \right) \left( e^{\mathrm{i} t_2 \hat H} \hat O e^{-\mathrm{i}t_2 \hat H} \right) \left( e^{-\mathrm{i} t_1 \hat H} | \psi\rangle \right) \end{equation}$$

we can obtain times $t_1 + t_2$ while only requiring MPO and MPS bond dimensions typical of times $t_1$ and $t_2$ respectively. Note that in this case, the computationally limiting operation is no longer the time evolution itself but the evaluation of observables given as the tensor network of a large-$w$ MPO between two large-$m$ MPS[3].

Complex time steps

Let us assume that we have a time-evolution operator $\hat U^\prime(\delta) = \mathbf{\hat 1} - \mathrm{i} \delta \hat H$ which is exact to first order. Applying this operator to a state will result in an error $O(\delta^2)$ compared to the exact evolution with the operator $\hat U(\delta) = e^{-\mathrm{i} \delta \hat H}$. Repeating the process $T / \delta$ times to obtain a state at final time $T$, we incur an error $O(\delta^2) \frac{T}{\delta} = O(\delta)$. However, if we allow complex intermediate steps $\delta_1$ and $\delta_2$, we can solve

$$\begin{equation} \hat U^\prime(\delta_1) \hat U^\prime(\delta_2) = \mathbf{\hat 1} - \mathrm{i} \delta \hat H - \frac{\delta^2}{2} \hat H^2 \end{equation}$$

for $\delta_1$ and $\delta_2$ by expanding the left-hand side:

$$\begin{align} & \quad \hat U^\prime(\delta_1) \hat U^\prime(\delta_2) \\ = & \quad \left( \mathbf{\hat 1} - \mathrm{i} \delta_1 \hat H \right) \left( \mathbf{\hat 1} - \mathrm{i} \delta_2 \hat H \right) \\ = & \quad \mathbf{\hat 1} - \mathrm{i} ( \delta_1 + \delta_2 ) \hat H - \delta_1 \delta_2 \hat H^2 \\ \Rightarrow & \quad \delta_1 + \delta_2 = \delta \quad \land \quad \delta_1 \delta_2 = \frac{\delta^2}{2}\;. \end{align}$$

Two solutions are possible, one of them is

$$\begin{align} \delta_1 = \frac{1 - \mathrm{i}}{2} \delta \quad \land \quad \delta_2 = \frac{1 + \mathrm{i}}{2} \delta \quad. \end{align}$$

Choosing these values for $\delta_{1,2}$ then results in a third-order error per time step and a second-order error overall. This choice of time steps is suggested in particular in combination with the MPO $\W^{II}$ method to obtain a better error per time step. The cost of the method only grows linearly with the number of evolution operators and, e.g., four operators $\hat U^\prime(\delta_{1,2,3,4})$ are required for a third-order error [5]. Four steps for the third-order approximation have an algebraic form

$$\begin{align} \delta_1 = \frac{1}{4}\left( -\frac{1+i}{\sqrt{3}}+1-i \right)\delta \quad \land \quad \delta_2 = i\delta_1 \quad \land \quad \delta_3 = -i \bar \delta_1 \quad \land \quad \delta_4 = \bar \delta_1, \end{align}$$

a fourth-order approximation requires 7 steps[6].

The drawback is the loss of unitarity at each individual time step which may be disadvantageous. Furthermore, if the time evolution is purely imaginary (e.g., for finite-temperature calculations) and the Hamiltonian does not contain complex coefficients, one may avoid complex arithmetic entirely and only use real floating-point scalars for 50% less memory usage and an approximately four-fold speed-up on matrix multiplications. Unfortunately, it is then impossible to use this trick to reduce the time-step error.

Green’s functions 1: Removal of low-lying states

This trick was first proposed in Ref. 7 and is relatively straightforward to implement. Assume a ground state $|0\rangle$ as an MPS obtained via DMRG and let us shift the Hamiltonian such that this state has energy $E_0 = 0$, $\hat H |0\rangle = 0$. We are then interested in the time-dependent observable

$$\begin{equation} x(t) = \langle 0 | \hat A(t) \hat B | 0 \rangle \end{equation}$$

where $\hat A$ and $\hat B$ are typically local operators such as creators or annihilators. The evolution of $\hat B | 0 \rangle$ is generically non-trivial and if we want to capture all frequency contributions in $x(t)$, we need to evolve until at least times $t^\prime = \frac{1}{E_1}$ where $E_1$ is the energy of the lowest eigenstate with non-zero energy $|1\rangle$ contained in $\hat B |0 \rangle$. In contrast, to capture contributions of higher-energy states $|n\rangle$ with energies $E_n > E_1$, we only need to evolve to shorter times $t^{\prime\prime} = \frac{1}{E_n} < t^\prime$.

However, a few low-lying eigenstates can often be calculated also with DMRG by orthogonalizing against previously-found eigenstates. Hence if we run DMRG multiple times, we can obtain not just the ground state $|0\rangle$ but also further eigenstates $|1\rangle$, $|2\rangle$ etc. If we use quantum numbers and $\hat B$ changes the quantum number of the state, these additional eigenstates should be calculated in the quantum number sector of $\hat B |0\rangle$. If we then orthogonalize $\hat B |0\rangle$ against $|1\rangle$, $|2\rangle$ etc., we remove the contributions which rotate (in real-time) or decay (in imaginary-time evolutions) the slowest and hence require the longest time evolutions. The evolution of the removed states can then be done exactly as we know both their energy and initial weight in $\hat B |0\rangle$. Even missing one of the eigenstates due to convergence problems with DMRG does not introduce an error but merely decreases the effectivity of the method.

Green’s functions 2: Linear prediction

Calculating dynamical structure factors or more generally spectral functions from time-dependent data requires two Fourier transformations: first, one needs to transform from real-space to momentum-space and second from real-time to frequency. The former transformation is typically unproblematic, but the latter transformation suffers either from overdamping or strong spectral leakage if the available maximal time $t$ is insufficient. Linear prediction[8][9][10] assumes that the real-time momentum-space Green’s function $G(k,t)$ is composed of multiple distinct exponentially decaying and oscillating contributions arising from a distinct pole structure of $G(k,\omega)$. If this is true for a time series $x_1, x_2, x_3, \ldots$, an additional data point $\tilde{x}_n$ can be approximated well by the form

$$\begin{equation} \tilde{x}_n = - \sum_{i=1}^p a_i x_{n-i-1} \end{equation}$$

with suitably chosen coefficients $a_i$ independent of $n$. We hence first compute a finite time series which is still as long as we can manage with time-dependent MPS methods. Subsequently, we need to find coefficients $a_i$ such that the above holds for the data we computed exactly. Using those $a_i$, we can then extend the time series to arbitrarily long times to generate a sufficiently long time series that a subsequent Fourier transform only requires minimal damping and hence provides for clear features.

It is useful to divide the available calculated data into three segments: first, one should discard an interval $[x_0, \ldots, x_{D-1}]$ at the beginning which captures short-time physics irrelevant and untypical of the longer-time behavior. Second, a fitting interval $[x_{D}, \ldots, x_{D+N-1}]$ of N points is selected over which the coefficients $a_i$ are minimized. Third, trust in the prediction is increased if it coincides with additional calculated data $[x_{D+N}, \ldots, x_{\mathrm{max}}]$ outside the fitting interval.

To select the $a_i$, we want to minimize the error

$$\begin{equation} \epsilon = \sum_{k=D}^{D+N-1} | \tilde{x}_k - x_k |^2 \quad. \end{equation}$$

Note that to evaluate $\tilde{x}_k$, $D$ must be larger than the number of coefficients $p$. The coefficient vector $\underline{a}$ is obtained as

$$\begin{equation} \underline{a} = - \underline{R}^{-1} \underline{r} \end{equation}$$

where the matrix $\underline{R}$ and vector $\underline{r}$ have entries

$$\begin{align} R_{i,j} = & \sum_{k=D}^{D+N-1} x_{k-i}^\star x_{k-j} \\ r_i = & \sum_{k=D}^{D+N-1} x_{k-i}^\star x_k \end{align}$$

respectively. Once the $a_i$ are obtained, data ideally can be generated initially for the interval $[x_{D+N}, \ldots, x_{\mathrm{max}}]$ and, once verified to coincide with the calculated data, extended to arbitrary times.

Several numerical pitfalls need to be considered here: First, the matrix $\underline{R}$ may be singular. Two possible remedies include addition of a small shift $\varepsilon$ or reduction of the number of parameters $p$. Ideally the latter should be considered, but may lead to problems finding the optimal non-singular $p$. Second, if we construct the vector

$$\begin{equation} \underline{x}_n = [x_{n-1}, \ldots, x_{n-p}]^T \end{equation}$$ we can move it forward one step as $$\begin{equation} \underline{\tilde{x}}_{n+1} = \underline{A} {\hspace{0.1cm}} \underline{x}_n \end{equation}$$ where the matrix $\underline{A}$ is of the form $$\begin{equation} \underline{A} = \begin{pmatrix} -a_1 & -a_2 & -a_3 & \cdots & -a_p \\ 1 & 0 & 0 & \cdots & 0 \\ 0 & 1 & 0 & \cdots & 0 \\ \vdots & \ddots & \ddots & \ddots & \vdots \\ 0 & \cdots & 0 & 1 & 0 \end{pmatrix} \end{equation}$$

and its eigenvalues $\alpha_i$ contain the frequencies and dampings of the aforementioned oscillations and exponential decays. As such, $\alpha_i > 1$ are unphysical and need to be dealt with, it appears [10] that setting those contributions to zero works best.

Purification insertion point (PIP)

---

(Above) Graphical representation of the expectation value of an out-of-time-ordered correlator within the purification framework of finite-temperature MPS calculations.

---

Calculating out-of-time-ordered correlators (OTOC) allows us to measure the scrambling of quantum information and finds many interesting and current applications. In general an OTOC of operators $\hat{W},\hat{V}$ is given as an ensemble average

$$\begin{align} C^{\hat V, \hat W}_{\beta}(t) &= \frac{1}{2} \mathrm{Tr} \left\{\hat{\rho}(\beta) \left[\hat{V}(t),\hat{W}\right]^{\dagger} \left[\hat{V}(t),\hat{W}\right] \right\} \notag \\ &= \underbrace{\mathcal{Re} \left[ \mathrm{Tr} \left\{\hat{\rho}(\beta) \hat{V}^{\dagger}(t)\hat{W}^{\dagger} \hat{V}(t)\hat{W} \right\} \right]}_{\equiv F^{\hat V, \hat W}_{\beta}(t)} + \text{time ordered} \end{align}$$

wherein we have suppressed the time-ordered terms and define the OTOC as the out-of-time ordered part $F^{\hat V, \hat W}_{\beta}(t)$. At finite temperature, we have to use a purification to evaluate this quantity. If we would calculate the time evolutions in $F^{\hat V, \hat W}_{\beta}(t)$ naively by direct evolution only in the physical degrees of freedom we would require $\mathcal{O}(N^{2})$ time steps to obtain the OTOC at time $t=N\delta$.

Clearly, the growing numerical expenses forbid to reach both large system sizes and long time scales $t$. Graphically representing this process (see above) immediately suggests to transform the operators in the OTOC in some way as to evenly distribute the required time evolutions leading to only linear scaling of effort in time $t$. In the following, we will explain how to transform these operators in the purification picture and alter the purification insertion point (PIP). For related work in the framework of matrix-product operators, cf. Ref. 11.

Consider the ensemble average $F_{\hat X, \hat Y, \hat Z,\beta} \equiv \mathrm{Tr} \left\{\hat{\rho}(\beta)\hat{Z}\hat{Y}\hat{X}\right\}$ for some global operators $\hat{X},\hat{Y},\hat{Z}$ at inverse temperature $\beta$. Using the cyclic property of the trace the ensemble average can now be written as expectation value in the enlarged Hilbert space

$$\begin{equation} F_{\hat X, \hat Y, \hat Z,\beta} = \mathrm{Tr} \left\{\hat{\rho}(\beta)\hat{Z}\hat{Y}\hat{X}\right\} = \bra{0}\hat{\rho}(\frac{\beta}{2})\hat{Z}\hat{Y}\hat{X}\hat{\rho}(\frac{\beta}{2})\ket{0} \equiv \bra{\frac{\beta}{2}}\hat{Z}\hat{Y}\hat{X}\ket{\frac{\beta}{2}} \;, \end{equation}$$

where we have introduced the purified finite temperature state $\ket{\frac{\beta}{2}} \equiv \hat{\rho}(\frac{\beta}{2})\ket{0}$ based on the infinite temperature state $\ket{0}$. A graphical representation of recasting the trace into an expectation value is given by the two networks (a) and (b) in the figure below with out-going indices representing row vectors and in-going indices column vectors.

---

(Above) Different choices of operator purifications for ensemble average $F_{\hat X, \hat Y, \hat Z,\beta} = \mathrm{Tr} \left\{\hat{\rho}(\beta)\hat{Z}\hat{Y}\hat{X}\right\}$.

---

From the pictographical representation we motivate the infinite temperature state $\ket{0}$ to be represented by a rank $(2,0)$ tensor $\ket{0} \equiv \sum_{a,\bar{b}} D^{a,\bar{b}}\ket{a}\ket{\bar{b}}$ and correspondingly $\bra{0}$ by a rank $(0,2)$ tensor $\bra{0} \equiv \sum_{a,\bar{b}}D_{a,\bar{b}}\bra{a}\bra{\bar{b}}$, where we have placed a bar over those indices labeling ancilla degrees of freedom.

These tensors have to fulfill the orthogonality conditions $$\begin{equation} \sum_{\bar{b}} D^{a,\bar{b}}D_{c,\bar{b}} = \delta^{a}_{c}, \quad \sum_{a} D^{a,\bar{b}}D_{a,\bar{c}} = \delta^{\bar{b}}_{\bar{c}} \end{equation}$$ so that the tensor elements can be choosen to be $D^{a,\bar{b}} \equiv D^{a,\bar{b}}\delta^{\bar{a}}_{\bar{b}}$ and $D_{a,\bar{b}} \equiv D_{a,\bar{b}}\delta^{\bar{b}}_{\bar{a}}$. When contracted over physical degrees of freedom, the action of these tensors is to convert row vectors into column vectors and vice versa $$\begin{equation} D\hat O D^{\dagger} = \sum_{a,c}\sum_{\bar{b},\bar{d}}D^{a,\bar{b}}\hat O^{c}_{a}D_{c,\bar{d}}\ket{\bar{b}}\bra{\bar{d}} = \sum_{\bar{b},\bar{d}}\hat O^{\bar{a}}_{\bar{c}}\ket{\bar{a}}\bra{\bar{c}} = \hat O^{t} \;. \end{equation}$$ If we now interpret indices carrying a bar as maps between ancilla degrees of freedom we can reformulate the purification in terms of the $D$ tensors $$\begin{equation} F_{\hat X, \hat Y, \hat Z,\beta} = \sum_{a,c,\ldots,g,\bar{b}} D_{a,\bar{b}}\hat \rho^{a}_{c}(\frac{\beta}{2})\hat Z^{c}_{d} \hat Y^{d}_{e} \hat X^{e}_{f}\hat \rho^{f}_{g}(\frac{\beta}{2})D^{g,\bar{b}}\;. \end{equation}$$ Inserting identities on the physical Hilbert space between $\hat{\rho}$ and $\hat{X}$ as well as $\hat{X}$ and $\hat{Y}$ and making explicit use of the representation of $\hat{D}$ we obtain $$\begin{align} F_{\hat X, \hat Y, \hat Z,\beta} &= \sum_{\substack{a,c,d,g,\bar{b}, \\ \bar{e},\bar{f},e_l,e_r,f_l,f_r}} D_{a,\bar{b}}\rho^{a}_{c}(\frac{\beta}{2}) \hat Z^{c}_{d} \hat Y^{d}_{e_{l}} \underbrace{D^{e_{l},\bar{e}}D_{e_{r},\bar{e}}}_{\delta^{e_{l}}_{e_{r}}} \hat X^{e_{r}}_{f_{l}} \underbrace{D^{f_{l},\bar{f}}D_{f_{r},\bar{f}}}_{\delta^{f_{l}}_{f_{r}}} \rho^{f_{r}}_{g}(\frac{\beta}{2})D^{g,\bar{b}} \notag \\ &= \sum_{\substack{a,c,d,\bar{b}, \\ \bar{e},\bar{f},e_l}} D_{a,\bar{b}} \rho^{a}_{c}(\frac{\beta}{2}) \hat Z^{c}_{d} \hat Y^{d}_{e_{l}} D^{e_{l},\bar{e}} \underbrace{\hat X^{\bar{f}}_{\bar{e}} \rho^{\bar{b}}_{\bar{f}}(\frac{\beta}{2})}_{\text{act on $\mathcal{H}_{A}$}}\\ & = \bra{0}\left(\hat{\rho}^{t}(\frac{\beta}{2}) \hat{X}^{t} \right)_{A}\otimes \left(\hat{\rho}(\frac{\beta}{2}) \hat{Z} \hat{Y} \right)_{P} \ket{0} \end{align}$$ so that now $\sum_{\bar{f}} \hat X^{\bar{f}}_{\bar{e}} \hat \rho^{\bar{b}}_{\bar{f}}(\frac{\beta}{2}) \equiv \hat{\rho}^{t}(\frac{\beta}{2})\hat{X}^{t}$ are acting on the ancilla space $\mathcal{H}_{A}$, i.e., we have shifted the purification insertion point. Again these manipulations can be represented efficiently in a graphical notation and are given as (c) in the figure above.

Using this procedure, we can rewrite the OTOC $F^{\hat V, \hat W}_{\beta}(t)$ as $$\begin{align} F^{\hat V, \hat W}_{\beta}(t) &= \mathcal{Re}\left[ \mathrm{Tr} \left\{\hat{\rho}(\frac{\beta}{2}) \hat{U}^{\dagger}(t) \hat{V}^{\dagger} \hat{U}(t) \hat{W}^{\dagger} \hat{U}^{\dagger}(t) \hat{V} \hat{U}(t) \hat{W} \hat{\rho}(\frac{\beta}{2})\right\} \right] \notag \\ &= \mathcal{Re} \left[ \bra{0} \left( \hat{U}^{\dagger}(t) \hat{V} \hat{U}(t) \hat{W} \hat{\rho}(\frac{\beta}{2}) \right)^{t}_{A} \otimes \left(\hat{\rho}(\frac{\beta}{2}) \hat{U}^{\dagger}(t) \hat{V}^{\dagger} \hat{U}(t) \hat{W}^{\dagger} \right)_{P} \ket{0} \right] \;. \end{align}$$ Defining the initial states $$\begin{align} \ket{W} &\equiv \hat{W}^{\dagger}_{P}\otimes \mathbf{\hat{1}}_{A}\ket{0}\;, \\ \ket{W_{\beta}} &\equiv \hat{\rho}_{P}(\frac{\beta}{2}) \otimes \left(\hat{W}^{*}\hat{\rho}_{A}(\frac{\beta}{2})\right) \ket{0} \;, \\ \end{align}$$ and their purified time evolutions $$\begin{align} \ket{W(t)} &\equiv \hat{U}^{\vphantom{*}}_{P}(t) \otimes \hat{U}^{*}_{A}(t)\ket{W} \;,\\ \ket{W_{\beta}(t)} &\equiv \hat{U}^{\vphantom{*}}_{P}(t) \otimes \hat{U}^{*}_{A}(t)\ket{W_{\beta}} \end{align}$$ the OTOC can be obtained by calculating the expectation value $$\begin{align} F^{\hat V, \hat W}_{\beta}(t) &= \mathcal{Re} \left[ \bra{W_{\beta}(t)}\hat{V}^{\dagger}_{P} \otimes \hat{V}^{t}_{A}\ket{W(t)} \right]\;. \end{align}$$ We hence only need $N$ steps to evaluate all expectation values for times $t = N \delta$.

From a more general point of view shifting the purification insertion point in the OTOCs reformulates the multiple Schrödinger time evolutions of the physical system in the canonical choice of the PIP into a Heisenberg time evolution on both the physical and ancilla system of a generalized observable $\hat{V}^{\dagger}_{P}\otimes \hat{V}^{\vphantom{\dagger}t}_{A}$.

Local basis optimization

---

(Above) Local basis optimization matrices $U_j$ are inserted on the physical legs of the MPS to transform a large physical basis (indicated by doubled lines) into a smaller effective basis of the MPS tensors $M_j$.

---

While the dimension of local Hilbert spaces is typically very limited in spin and electronic systems, bosonic systems potentially require a large local dimension $\sigma = \mathcal{O}(100)$. As this local dimension typically enters at least quadratically in some operations on matrix-product states, some way to dynamically select the most relevant subspace of the local Hilbert space is potentially extremely helpful. The local basis transformation[12][13][14] method provides for just this: by inserting an additional matrix $U_j$ on each physical leg of the MPS, the large physical dimension is transformed into a smaller effective basis. The rank-3 MPS tensor then only has to work with the smaller effective basis. The method was adapted for TEBD time evolution in Ref. 14 but it is also straightforward to use in the other time-evolution methods presented here. For the MPO \wiii and global Krylov methods, only the MPO-MPS product has to be adapted to generate additionally a new optimal local basis after each step. The DMRG, TDVP and local Krylov method translate[15][16][17] directly in much the same way as DMRG.

---

The content of this page is based on Time-evolution methods for matrix-product states by S. Paeckel, T. Köhler, A. Swoboda, S. R. Manmana, U. Schollwöck and C. Hubig and is licensed under the CC-BY 4.0 license.

References

1. Density Matrix Renormalization Group in the Heisenberg Picture, Michael J. Hartmann, Javier Prior, Stephen R. Clark, Martin B. Plenio, Phys. Rev. Lett. 102, 057202 (2009)
2. Matrix product simulations of non-equilibrium steady states of quantum spin chains, Tomaz Prosen, Marko Znidaric, Journal of Statistical Mechanics: Theory and Experiment 2009, P02035 (2009)
3. Extending the range of real time density matrix renormalization group simulations, D.M. Kennes, C. Karrasch, Computer Physics Communications 200, 37 - 43 (2016)
4. Matrix product purifications for canonical ensembles and quantum number distributions, Thomas Barthel, Phys. Rev. B 94, 115157 (2016)
5. Time-evolving a matrix product state with long-ranged interactions, Michael P. Zaletel, Roger S. K. Mong, Christoph Karrasch, Joel E. Moore, Frank Pollmann, Phys. Rev. B 91, 165112 (2015)
6. Out-of-equilibrium dynamics in a quantum impurity model: Numerics for particle transport and entanglement entropy, Kemal Bidzhiev, Gr'egoire Misguich, Phys. Rev. B 96, 195117 (2017)
7. Nonequilibrium electron transport using the density matrix renormalization group method, Peter Schmitteckert, Phys. Rev. B 70, 121302 (2004)
8. The Theory of Linear Prediction, P.P. Vaidyanathan (2008)
9. Spectral function for the $S=1$ Heisenberg antiferromagetic chain, Steven R. White, Ian Affleck, Phys. Rev. B 77, 134437 (2008)
10. Spectral functions in one-dimensional quantum systems at finite temperature using the density matrix renormalization group, Thomas Barthel, Ulrich Schollwoeck, Steven R. White, Phys. Rev. B 79, 245101 (2009)
11. Scrambling and thermalization in a diffusive quantum many-body system, A Bohrdt, C B Mendl, M Endres, M Knap, New Journal of Physics 19, 063001 (2017)
12. Density Matrix Approach to Local Hilbert Space Reduction, Chunli Zhang, Eric Jeckelmann, Steven R. White, Phys. Rev. Lett. 80, 2661–2664 (1998)
13. Critical and Strong-Coupling Phases in One- and Two-Bath Spin-Boson Models, Cheng Guo, Andreas Weichselbaum, Jan Delft, Matthias Vojta, Phys. Rev. Lett. 108, 160401 (2012)
14. Matrix-product-state method with a dynamical local basis optimization for bosonic systems out of equilibrium, C. Brockt, F. Dorfner, L. Vidmar, F. Meisner, E. Jeckelmann, Phys. Rev. B 92, 241106 (2015)
15. Simulating open quantum dynamics with time-dependent variational matrix product states: Towards microscopic correlation of environment dynamics and reduced system evolution, Florian A. Y. N. Schroeder, Alex W. Chin, Phys. Rev. B 93, 075105 (2016)
16. Florian Dorfner: Numerical methods for strongly correlated many-body systems with bosonic degrees of freedom (Doctoral Thesis), LMU Munich, 2017
17. Claudius Hubig: Symmetry-Protected Tensor Networks (Doctoral Thesis), LMU Muenchen, 2017

Edit This Page