$$ \nonumber \newcommand{\br}{\mathbf{r}} \newcommand{\bR}{\mathbf{R}} \newcommand{\bp}{\mathbf{p}} \newcommand{\bk}{\mathbf{k}} \newcommand{\bq}{\mathbf{q}} \newcommand{\bv}{\mathbf{v}} \newcommand{\bx}{\mathbf{x}} \newcommand{\bz}{\mathbf{z}} \DeclareMathOperator*{\E}{\mathbb{E}} $$
Sarang Gopalakrishnan and AL 2019
Pieter Claeys and AL 2020, and arXiv:2009.03791
Slides at austen.uk/slides/dual-unitaries-pcts
Sarang Gopalakrishnan and AL (2019) interpreted earlier results of Bertini, Kos, Prosen PRX (2019) on the kicked Ising model in terms of dual unitarity
Simultaneously, Bertini, Kos, Prosen PRL (2019) showed how to compute correlations for general dual unitarities
$$ \begin{aligned} H_{\text{KIM}}(t) = H_\text{I}[\mathbf{h}] + \sum_{m}\delta(t-n)H_\text{K}\\ H_\text{I}[\mathbf{h}]=\sum_{j=1}^L\left[J Z_j Z_{j+1} + h_j Z_j\right],\qquad H_\text{K} &= b\sum_{j=1}^L X_j, \end{aligned} $$
$$ \begin{aligned} U(n_+) = \left[U(1_+)\right]^n,\qquad U(1_-) = K I_\mathbf{h}\\ I_\mathbf{h} = e^{-iH_\text{I}[\mathbf{h}]}, \qquad K &= e^{-iH_\text{K}}, \end{aligned} $$
$$ \lim_{L\to\infty} S^{(n)}_A(t) =\min(2t-2,N)\log 2, $$
$$ S^{(n)}_A = \frac{1}{1-n}\log \text{tr}\left[\rho^n\right]=\frac{1}{1-n}\sum_\alpha \lambda_\alpha^n $$
$$ \lambda_\alpha = \left(\frac{1}{2}\right)^{\min(2t-2,N)} $$
$$ \begin{aligned} \mathcal{K} &= \exp\left[-i b X\right]\\ \mathcal{I} &= \exp\left[-iJ Z_1 Z_2 -i \left(h_1 Z_1 + h_2 Z_2\right)/2\right]. \end{aligned} $$
When $|J|=|b|=\pi/4$ KIM gate is dual unitary
This allows for simple proof of entanglement dynamics!
$$ \overbrace{\frac{\mathbb{1}}{2}\otimes \frac{\mathbb{1}}{2} \cdots }^{t-1} \otimes\overbrace{|Z_1\rangle\langle Z_1|\otimes |Z_2\rangle\langle Z_2| \cdots }^{N-2t+2 } \otimes \overbrace{\frac{\mathbb{1}}{2}\otimes \frac{\mathbb{1}}{2} \cdots }^{t-1} $$
$$ \overbrace{\frac{\mathbb{1}}{2}\otimes \frac{\mathbb{1}}{2} \cdots }^{t-1} \otimes\overbrace{|Z_1\rangle\langle Z_1|\otimes |Z_2\rangle\langle Z_2| \cdots }^{N-2t-2 } \otimes \overbrace{\frac{\mathbb{1}}{2}\otimes \frac{\mathbb{1}}{2} \cdots }^{t-1} $$
Piroli, et al., (2020) considered 2-site MPS initial conditions that allow for solution in similar way
For $2t>N $ result is the same: $\infty$-temperature RDM
At earlier times structure of initial state important
$$ c(x,y,t)=\mathop{\text{tr}}\left[U(t)^\dagger O(x)U(t) O(y)\right] $$
$$ \begin{align} c_\nu^{\alpha\beta}(\nu t,t) = \frac{1}{q} {\rm tr}\left[\mathcal M_{\nu}^{2t}(a^\beta)a^\alpha\right]\\ \mathcal M_{+}(a) = \frac{1}{q} {\rm tr}_1\left[U^\dagger (a\otimes\mathbb{1}) U\right] \end{align} $$
$$ C_{\alpha \beta}(x,t) = \langle \sigma_{\alpha}(0,t) \sigma_{\beta}(x,0) \sigma_{\alpha}(0,t) \sigma_{\beta}(x,0) \rangle. $$
$C_{\alpha \beta}(|x|>t,t)=1$ since $\left[\sigma_{\alpha}(0,t),\sigma_{\beta}(x,0)\right]=0$ for $|x|>t$
For smaller $|x|$ OTOC deviates from 1.
Limiting value of $|x|/t$ where this occurs defines butterfly velocity
Example: random unitary circuits with local dimension $q$ have $$ v_\text{B} = \frac{q^2-1}{q^2+1} $$ $v_\text{B}\to 1$ as $q\to\infty$
(and no broadening)
$$ \begin{align} C^{+}_{\alpha \beta }(x,t) = \left(L(\sigma_{\alpha})\right|\left(T_{n_-}\right)^{n_+}\left|R^{-}(\sigma_{\beta})\right), \nonumber\\ \end{align} $$
Using only unitarity
This is the value outside light cone, so $v_\text{B}<1$ generically
$v_\text{B}=1$ requires additional eigenstates with eigenvalue one
$C^{+}_{\alpha \beta}(x,t)$
vanishes inside light cone
$$ \begin{align} \lim_{(x+t) \to \infty }C^{+}_{\alpha \beta}(x,t) = \begin{cases} -\frac{1}{q^2-1} \qquad &\text{if} \quad x=t,\\ 0 \qquad &\text{if} \quad x \neq t. \end{cases} \end{align} $$
For odd $n$, $C^{-}_{\alpha \beta}(x,t)$
can be found in terms of channel $\mathcal M_{\pm}(\cdot)$
describing light cone correlator
Range of behaviour moving inside light cone
Integrable SDKIM: Exponentially many eigenvectors with eigenvalue one. OTOC immediately saturates to constant value inside light cone.
Are all maximal velocity circuits dual unitary? No! Found a kicked XY model with $v_\text{B}=1$
Dual unitary circuits are a big solvable family with a diverse phenomenology that includes integrable and chaotic behaviour…
… but maybe not that diverse for such a big family! $v_\text{B}=1$ always, for example
Still more to do: effect of measurements, coding, computational complexity…