Interacting non-Hermitian edge and cluster bursts on a digital quantum processor Authors: Jin Ming Koh, Wen-Tan Xue, Tommy Tai, Dax Enshan Koh, Ching Hua Lee Published: 2026 ABSTRACT Dissipation can drive striking dynamics in quantum systems. A recent discovery is the non-Hermitian edge burst, where energy density transiently accumulates near boundaries rather than decaying uniformly. Previously observed in single-particle settings, this effect remains unexplored in interacting multi-particle systems, as tunable interactions are difficult to realize in conventional platforms, and non-Hermitian simulation is demanding on quantum processors. Here, we develop a framework for digital non-Hermitian Hamiltonian simulation that composes a linear-combination-of-unitaries scheme and product formulae. The method is efficient in classical preprocessing costs and circuit sizes. Implementing an interacting quantum ladder model on a superconducting quantum processor, we observe edge-burst signatures in up to 64 unit cells, directly probe dissipative-gap closing, and identify interaction-driven phenomena: spatially extended and patterned edge bursts, and cluster bursts in the bulk. Our study establishes digital quantum processors as a versatile platform for non-Hermitian physics. Introduction While closed quantum-mechanical systems are described by Hermitian Hamiltonians, a large range of open quantum systems can be effectively described by non-Hermitian Hamiltonians1–5 that model coupling to an external environment. This description captures a variety of physical platforms, including loss-engineered optical structures6,7, superconducting and circuit-QED components, and other open systems that emulate dissipative quantum dynamics8–10. Beyond the rich physics conferred by non-Hermiticity, non-Hermitian quantum systems have also been harnessed for their distinctive properties in the development of ultra-sensitive sensors and detectors11–13, among other technologies. Recently, the non-Hermitian edge burst 14–18 has emerged as a particularly striking manifestation of dissipative dynamics, in which dissipation becomes transiently and anomalously concentrated at system boundaries over time. While its static counterpart, the non-Hermitian skin effect (NHSE)19–23, is now well-established even in interacting many-body settings24–28, the edge burst remains less explored and its behavior under many-body dynamics is unknown. Extending it to the many-body regime raises fundamental questions: how interactions reshape the collective flow of dissipation, and whether the interplay between boundary bursts and particle correlations can lead to qualitatively different phenomenology. Digital quantum simulation provides a compelling opportunity to explore this regime, as they enable the effective implementation of tunable many-body interactions that are otherwise difficult to realize in engineered analog systems29,30. While digital quantum processors are increasingly utilized for quantum dynamics31–38 and various areas of (Hermitian) condensed-matter simulation39–49, their use in studying non-Hermitian physics is relatively nascent owing to difficulties in scalably realizing non-unitary time-evolution on unitary gate-based hardware, particularly on near-term noisy intermediate-scale quantum (NISQ) devices. Demonstrations to-date are restricted to small system sizes and single-particle models or rely on resource-intensive protocols50–55. In this work, we report realizations of both single-body and multi-particle interacting generalizations of the non-Hermitian edge burst on superconducting transmon-based quantum hardware. We overcome the aforementioned limitations by developing an efficient and general digital quantum simulation framework for non-unitary quantum dynamics. Our approach leverages a linear combination of unitaries (LCU) circuit construction within a stepwise time-evolution sequence. To accommodate anti-Hermiticity in the Hamiltonian, we utilize an ancilla-assisted circuit block that performs controlled forward and backward Hermitian time-evolution, and which is reused in each timestep via mid-circuit resets. Our simulation method uses a small constant number of ancillary qubits, does not require exponentially expensive classical preprocessing nor large numbers of intermediary measurements or variational iterations, and accommodates arbitrary interactions in multi-particle settings—marking an advance over previous approaches limited to small-sized or single-particle systems. We apply our method to probe a minimal quantum ladder supporting the non-Hermitian edge burst. Beyond observing signatures of the conventional single-body non-interacting edge burst on up to 64 unit cells, we show on quantum hardware that introducing sequences of density-density interactions leads to the formation of spatially extended and patterned versions of the edge burst, and in certain regimes can also lead to bursts occurring in the bulk of the system far from boundaries, which we dub cluster bursts. These results extend edge burst physics into the multi-particle setting, in particular demonstrating how particle interactions and correlations modify dissipative transport and localization in non-Hermitian systems. In the near term, our work establishes digital quantum simulation as a promising platform for studying correlated non-Hermitian dynamics. Results Minimal interacting non-Hermitian quantum ladder The non-Hermitian edge burst model we investigate is built atop a minimal bosonic one-dimensional quantum ladder14, characterized by a two-band Bloch Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}(k)$$\end{document}H0eb(k) together with a sequence of density-density interactions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}$$\end{document}Hinteb, namely 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_0^{{{{\rm{eb}}}}}(k) =\left({v}_{1}+{v}_{2}\cos k\right){\sigma }^{x}+\left({v}_{2}\sin k+\frac{i\gamma }{2}\right){\sigma }^{z}-\frac{i\gamma }{2}{\mathbb{I}},\\ {{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}} ={\sum}_{r\ge 1}{U}_{r}{\sum}_{z=1}^{2N}{n}_{z}{n}_{z+r},$$\end{document}H0eb(k)=v1+v2coskσx+v2sink+iγ2σz−iγ2I,Hinteb=∑r≥1Ur∑z=12Nnznz+r, where v1, v2 ≥ 0 are tight-binding hopping coefficients, γ > 0 is an on-site loss rate and is responsible for the non-Hermiticity of the system, and σx, σz are Pauli operators acting on two pseudospin degrees of freedom which we associate with a and b sublattices within a unit cell; and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{r}\in {\mathbb{R}}$$\end{document}Ur∈R is an interaction potential that depends on the occupation ns at the site s on the flattened ladder, separated by a distance r. Specifically, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{z}={n}_{x\ell }={c}_{x\ell }^{{{\dagger}} }{c}_{x\ell }$$\end{document}nz=nxℓ=cxℓ†cxℓ for flattened index z = 2x − δℓa where the particle operator cxℓ acts on the unit cell x ∈ [N] and sublattice ℓ ∈ {a, b}. We consider hardcore bosons on the ladder, which emerge in a regime of strong on-site repulsive interactions on top of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}$$\end{document}Hinteb (see Methods). As we will demonstrate, this model supports interaction-stabilized dissipative phenomena such as spatially extended edge bursts and cluster formation, revealing how correlations modify the canonical picture of the non-Hermitian edge burst. The Schrödinger equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i({{{\rm{d}}}}/{{{\rm{d}}}}t){\omega }_{t}={{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}{\omega}_{t} - {\omega}_{t} ({{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}})^{\dag}$$\end{document}i(d/dt)ωt=Hebωt−ωt(Heb)†] prescribes that from an initial normalized quantum state (density matrix) ω0 of the ladder, the state at time t is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{t}={V}_{t}^{{{{\rm{eb}}}}}{\omega }_{0}{({V}_{t}^{{{{\rm{eb}}}}})}^{{{\dagger}} }$$\end{document}ωt=Vtebω0(Vteb)† where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{t}^{{{{\rm{eb}}}}}=\exp (-i{{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}t)$$\end{document}Vteb=exp(−iHebt) is the time-evolution propagator. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is non-Hermitian, the propagator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{t}^{{{{\rm{eb}}}}}$$\end{document}Vteb is non-unitary and state normalization is not generically preserved. In particular, the quantum state norm decays as 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{{\rm{d}}}}}{{{{\rm{d}}}}t}{{{\rm{tr}}}}({\omega }_{t})=i{\left\langle {{{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}}^{{{\dagger}} }-{{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}\right\rangle }_{{\omega }_{t}}=-2\gamma {\sum }_{x=1}^{N}{\left\langle {n}_{xb}\right\rangle }_{{\omega }_{t}}\le 0.$$\end{document}ddttr(ωt)=iHeb†−Hebωt=−2γ∑x=1Nnxbωt≤0. The norm of the quantum state describes the probability of the walker (i.e., boson) remaining on the ladder. Accordingly, the escape probability of the walker at unit cell x by time t is given by the time integral 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{x}(t)=2\gamma \int_{0}^{t}{\left\langle {n}_{xb}\right\rangle }_{{\omega }_{\tau }}\,{{{\rm{d}}}}\tau,$$\end{document}Px(t)=2γ∫0tnxbωτdτ, and the final cell-resolved escape probabilities are given by the long-time limit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}={{\lim }_{t\to \infty }}{P}_{x}(t)$$\end{document}Px=limt→∞Px(t). Restricted to pure single-particle quantum states, Eqs. (2), (3) reduces to expressions that are consistent with expressions in Ref. 14. Our central quantities of interest are the final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px which enable direct probing of the presence and properties of the non-Hermitian edge burst. Accordingly, we seek to measure the site-resolved occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt in our realizations, from which the escape probabilities Px(t) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px can be recovered. Despite the apparent simplicity of the model [Eq. (1)], realizing such dynamics in a controlled experimental setting is highly nontrivial. The system features interactions of varying range and strength as well as non-Hermitian loss, a combination that imposes difficult constraints on analog implementations (e.g., photonic lattices, cold atoms) due to decoherence and limitations in control over the form of couplings and interactions in the implemented evolution. Digital quantum simulation provides a distinctive versatility advantage in accommodating arbitrary coupling and interaction terms, but classical preprocessing costs in the construction of the quantum circuit and the sheer overhead in circuit size have been major challenges in approaches to-date50–55. Realizing non-Hermitian quantum dynamics on quantum hardware We implement the dynamics of the non-Hermitian quantum ladder [Eq. (1)] on a superconducting quantum processor that executes a quantum circuit encoding its time evolution. As the quantum simulation approach we develop for non-Hermitian Hamiltonians \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H is general, we herein introduce the methodology for an arbitrary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H; the particular setting of the quantum ladder fixes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}={{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}H=Heb. At a high level, we perform time evolution under \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H in time steps, where each time step naturally splits into unitary and non-unitary components arising respectively from the Hermitian and anti-Hermitian parts of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H. Namely, on the quantum ladder \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb hosting the edge burst, these comprise the coherent hoppings and density-density interactions, and on-site loss, respectively. The former is implemented through standard techniques, but the latter is nontrivial and is performed via a linear combination of unitaries (LCU) circuit block, as we describe shortly. Complete circuit details and experiment procedures are described in Methods. To start, as quantum states on a quantum device are always normalized, we perform a normalized version of the non-Hermitian time-evolution and subsequently reconstruct the corresponding non-normalized physical state from observable data. The normalized evolution is 4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{t}=\frac{{V}_{t}{\rho }_{0}{V}_{t}^{{{\dagger}} }}{{{{\rm{tr}}}}({V}_{t}{\rho }_{0}{V}_{t}^{{{\dagger}} })},\qquad {\rho }_{0}={\omega }_{0},$$\end{document}ρt=Vtρ0Vt†tr(Vtρ0Vt†),ρ0=ω0,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{t}={e}^{-i{{{\mathcal{H}}}}t}$$\end{document}Vt=e−iHt is the (non-unitary) time-evolution propagator over time t. The non-normalized physical state ωt and ρt are related by the rescaling \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{t}={A}_{t}^{2}{\rho }_{t}$$\end{document}ωt=At2ρt for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{t}^{2}={{{\rm{tr}}}}({\omega }_{t})$$\end{document}At2=tr(ωt), and accordingly \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle O\right\rangle }_{{\omega }_{t}}={A}_{t}^{2}{\left\langle O\right\rangle }_{{\rho }_{t}}$$\end{document}Oωt=At2Oρt for any observable O. Thus, in addition to implementing the normalized time-evolution, we develop a method to recover the normalization factor At from measurements. We employ a time-stepping algorithm to perform time-evolution from ρ0 to ρt (see Fig. 1c) that splits the evolution into m steps, each approximately realizing time-evolution over an interval Δt = t/m to an error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(\Delta {t}^{2}\right)$$\end{document}OΔt2. Thus, the error accumulated over all steps scales as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(1/m\right)$$\end{document}O1/m and arbitrarily high simulation precision can be reached by increasing m. Writing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}={{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}H=HH−iHA where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}−iHA are the Hermitian and anti-Hermitian parts of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H respectively, each time step comprises the quantum maps \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA in sequence, designed to perform normalized time-evolution on arbitrary incident states by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}−iHA respectively over time Δt to error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(\Delta {t}^{2}\right)$$\end{document}OΔt2.Fig. 1Non-Hermitian interacting quantum ladder model and quantum simulation methodology.a Schematic of quantum ladder built from sublattice chains a and b, with reciprocal hopping v1, v2 between sites and on-site loss γ on the b chain. Leftward drift of a wavepacket is induced by the non-Hermitian skin effect (NHSE), accompanied by dampening of wavepacket amplitude by the non-Hermitian loss. b A sequence of range-r density-density interactions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{U}_{r}\}}_{r}$$\end{document}{Ur}r between sites on the ladder, giving rise to multi-particle effects enriching the non-Hermitian edge burst phenomenology such as spatial patterning and cluster bursts in the bulk explored in this work. c Structure of time-stepping quantum algorithm simulating time-evolution by a general non-Hermitian Hamiltonian on a quantum processor. The quantum maps \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA perform time-evolution by the Hermitian and anti-Hermitian parts of the Hamiltonian, respectively. d The non-unitary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA is implemented through a linear combination of unitaries (LCU) circuit primitive, with the aid of a small ancillary register. e Structure of time-evolution circuits executed on the quantum processor. The LCU implementing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA superposes forward and backward time-evolution, coherently controlled by a single ancillary qubit that is re-used through mid-circuit qubit reset. Observables such as site-resolved occupancies are measured on the time-evolved state at the end of the circuit. f Time-evolution by Hermitian Hamiltonian components by trotterization. The R gates denote varying multi-qubit Pauli rotations. Non-Hermitian interacting quantum ladder model and quantum simulation methodology. a Schematic of quantum ladder built from sublattice chains a and b, with reciprocal hopping v1, v2 between sites and on-site loss γ on the b chain. Leftward drift of a wavepacket is induced by the non-Hermitian skin effect (NHSE), accompanied by dampening of wavepacket amplitude by the non-Hermitian loss. b A sequence of range-r density-density interactions \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{U}_{r}\}}_{r}$$\end{document}{Ur}r between sites on the ladder, giving rise to multi-particle effects enriching the non-Hermitian edge burst phenomenology such as spatial patterning and cluster bursts in the bulk explored in this work. c Structure of time-stepping quantum algorithm simulating time-evolution by a general non-Hermitian Hamiltonian on a quantum processor. The quantum maps \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA perform time-evolution by the Hermitian and anti-Hermitian parts of the Hamiltonian, respectively. d The non-unitary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA is implemented through a linear combination of unitaries (LCU) circuit primitive, with the aid of a small ancillary register. e Structure of time-evolution circuits executed on the quantum processor. The LCU implementing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA superposes forward and backward time-evolution, coherently controlled by a single ancillary qubit that is re-used through mid-circuit qubit reset. Observables such as site-resolved occupancies are measured on the time-evolved state at the end of the circuit. f Time-evolution by Hermitian Hamiltonian components by trotterization. The R gates denote varying multi-qubit Pauli rotations. The unitary channel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH can be implemented by standard methods. In the present work, we use the first-order Trotter-Lie product formula56, which rewrites \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH in the Pauli basis and performs time-evolution through a sequence of multi-qubit Pauli rotations (see Fig. 1f). In general, higher-order product formulae or alternative unitary time-evolution circuit construction methods57,58 can also be used. More details on the implementation of trotterized circuits on hardware are given in Methods. In contrast the map \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA, realizing action \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \mapsto {V}_{\Delta t}^{{{{\rm{A}}}}}\rho {V}_{\Delta t}^{{{{\rm{A}}}}{{\dagger}} }+{{{\mathcal{O}}}}\left(\Delta {t}^{2}\right)$$\end{document}ρ↦VΔtAρVΔtA†+OΔt2 up to normalization for time-evolution propagator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{\Delta t}^{{{{\rm{A}}}}}={e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}VΔtA=e−HAΔt and arbitrary incident states ρ, is non-unitary and is a source of difficulty. To perform \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA, we make use of the linear combination of unitaries (LCU) circuit primitive, which enables an effective action defined by sums of unitaries rather than products on a system register. In general, to effect an action (μ1U1 + ⋯ + μdUd) for coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{\mu }_{k}\ > \ 0\}}_{k}$$\end{document}{μk>0}k and unitaries \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{U}_{k}\}}_{k}$$\end{document}{Uk}k up to normalization, a register of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lceil {\log }_{2}d\rceil$$\end{document}⌈log2d⌉ ancillary qubits and the ability to perform Uk coherently controlled by the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert k\right\rangle$$\end{document}k state of the ancillary register suffice (see Fig. 1d). For our purpose, we approximate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{\Delta t}^{{{{\rm{A}}}}}$$\end{document}VΔtA by a linear combination of forward and backward unitary time-evolution to error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(\Delta {t}^{2}\right)$$\end{document}OΔt2. In fact, in general, for any d ≥ 2, applicable mixtures of time-evolution operations can be efficiently found through power series expansions in Δt, and solutions of higher-order error are available (see Methods). To minimize the hardware resources required, we used d = 2, which necessitated only a single ancillary qubit. The pair of forward and backward time-evolution slices in the LCU are then defined by μ+ = μ− and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{\pm }={e}^{\pm i{{{\mathcal{R}}}}\Delta \tau }$$\end{document}U±=e±iRΔτ, for a rescaled time step \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tau=\sqrt{2\Delta t}$$\end{document}Δτ=2Δt and an auxiliary Hermitian Hamiltonian satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{R}}}}}^{2}={{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}R2=HA. In the present edge burst setting, while choices of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{R}}}}}^{{{{\rm{eb}}}}}$$\end{document}Reb satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${({{{{\mathcal{R}}}}}^{{{{\rm{eb}}}}})}^{2}={{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}(Reb)2=HAeb are readily found, further leveraging the structure of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}HAeb allowed refinements to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{R}}}}}^{{{{\rm{eb}}}}}$$\end{document}Reb such that the LCU achieves the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{\Delta t}^{{{{\rm{A}}}}}$$\end{document}VΔtA time-evolution exactly instead of to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(\Delta {t}^{2}\right)$$\end{document}OΔt2 error (see Methods). We implemented the U± unitary time-evolution slices via first-order trotterization, with a low-overhead scheme in effecting the coherent controls by the ancillary qubit (see Methods). The structure of the overall circuit we executed in experiments is shown in Fig. 1e. Through mid-circuit qubit reset, we re-initialized the ancillary qubit used for the LCU primitive between time steps—thus a single ancilla sufficed for the entire evolution from ρ0 to ρt. Unlike other methods that can be adapted for non-Hermitian dynamical evolution59–62, this protocol does not assume an ansatz for the time-evolved state, nor does it require iterative variational optimization or step-wise circuit construction based on intermediate measurements, which are often large in number. In Methods, we provide a generalized description of the time-stepping quantum simulation procedure outlined here. For cases where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{R}}}}$$\end{document}R is difficult to obtain, we also describe a method based on a similar LCU expansion but requiring only access to terms in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}HA. Thus, in totality, our framework accommodates non-Hermitian lattice Hamiltonians of arbitrary structure. Upon completion of evolution to ρt, we simultaneously measured site-resolved occupancies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt for all unit cells x ∈ [N] and sublattices ℓ ∈ {a, b} on the quantum ladder. We then recovered the quantum state normalization factor At via Eq. (2) as the time integral 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{t}=\exp \left(-\gamma \int_{0}^{t}{\sum }_{x=1}^{N}{\left\langle {n}_{xb}\right\rangle }_{{\rho }_{\tau }}\,{{{\rm{d}}}}\tau \right),$$\end{document}At=exp−γ∫0t∑x=1Nnxbρτdτ, which then enabled the rescaling of the measured \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt. Lastly, a second time integration produces the escape probabilities Px(t) as in Eq. (3). We run experiments to large times t to access the long-time escape behavior \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px (see Methods). We utilized IBM superconducting quantum processors in our experiments, which host up to 133 transmon qubits connected in a heavy-hexagon topology with decoherence times T1, T2 up to ~200 μs. These devices support mid-circuit qubit readout used in our quantum circuits. Further details on the quantum hardware are provided in Methods, and device specifications such as gate and readout error rates are summarized in Supplementary Table 4. To address noise, which is non-negligible on present-day quantum hardware, we integrate several error suppression and mitigation methods such as dynamical decoupling, zero noise extrapolation with randomized gate twirling, and readout error mitigation (see Methods). Observing signatures of the non-Hermitian edge burst A key benchmark of our approach is that the non-Hermitian edge-burst dynamics can be implemented faithfully on a quantum processor at negligible classical preprocessing cost through trotterized time-evolution, without further sophisticated quantum circuit optimization (e.g., circuit recompilation) or variational circuit tuning. This indicates that, even at extended lattice sizes, the same LCU-based quantum simulation method can be executed directly on quantum hardware in a scalable fashion to observe the physical phenomenology of interest (assuming progress in quantum hardware continues to drive noise rates down). To start, we realized the non-Hermitian quantum ladder \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb with N = 8 unit cells without interactions (Ur = 0 uniformly) and with open boundary conditions—natural on an open chain of qubits on the quantum processor—such that the ladder terminated in spatial boundaries past x = 1 and x = 8. The ladder is initialized with a particle on the a-sublattice of the x = 6 unit cell. We constructed our time-evolution circuits via trotterization as described earlier with standard transpilation onto hardware (see Methods), without employing more sophisticated circuit optimization techniques. We present a breakdown of experimental results in Fig. 2, illustrating the major steps in the measurement and data processing pipeline in addition to the final results. First, we show in Fig. 2a site-resolved occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt on the normalized time-evolved quantum state ρt as measured on the hardware. These occupancy densities were acquired at several noise amplification factors λ ≥ 1, where λ = 1 corresponds to the circuits without modification and λ > 1 are on folded circuits of artificially increased gate count and depth (see Methods) but achieving, in the absence of noise, the same evolution. The increased circuit size amplifies noise suffered during time evolution, arising from gate errors that are averaged by randomized gate twirling (see Methods) and decoherence of the qubits over time. The impact of the amplified noise is clear in the measured data: at higher λ, the amplitude of oscillations in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt are suppressed and small-scale fluctuations become more prominent.Fig. 2Observing signatures of the non-Hermitian edge burst on a quantum processor.a Site-resolved occupancy densities after normalized time-evolution on a 8-unit cell quantum ladder, comparing data measured on quantum hardware at various noise amplification factors λ, after post-processing with zero-noise extrapolation (ZNE), and exact numerics. Solid thin lines are with readout error mitigation (RO) applied; translucent thin lines behind are without. To avoid clutter, only the a sublattice of unit cells x ∈ {1, 3, 5, 6} are shown. b Unit-cell-resolved escape probabilities Px(t) obtained from the data in (a). Time-integration of the occupancy densities first recovers proper wavefunction norm, and a second time-integration produces Px(t). c Total escape probability P(t) summed over all unit cells, which approaches unity as time progresses. d, e Final unit-cell-resolved escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px in the long-time limit. Data obtained on quantum hardware with and without ZNE and RO, and from exact numerics, are shown. In a parameter regime supporting the edge burst (d), anomalously high escape probability on the x = 1 edge of the ladder is detected; whereas in the trivial regime (e) escape probability is concentrated only near the initial location of the particle (at x0 = 6). Error bars are standard deviations across 8 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Observing signatures of the non-Hermitian edge burst on a quantum processor. a Site-resolved occupancy densities after normalized time-evolution on a 8-unit cell quantum ladder, comparing data measured on quantum hardware at various noise amplification factors λ, after post-processing with zero-noise extrapolation (ZNE), and exact numerics. Solid thin lines are with readout error mitigation (RO) applied; translucent thin lines behind are without. To avoid clutter, only the a sublattice of unit cells x ∈ {1, 3, 5, 6} are shown. b Unit-cell-resolved escape probabilities Px(t) obtained from the data in (a). Time-integration of the occupancy densities first recovers proper wavefunction norm, and a second time-integration produces Px(t). c Total escape probability P(t) summed over all unit cells, which approaches unity as time progresses. d, e Final unit-cell-resolved escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px in the long-time limit. Data obtained on quantum hardware with and without ZNE and RO, and from exact numerics, are shown. In a parameter regime supporting the edge burst (d), anomalously high escape probability on the x = 1 edge of the ladder is detected; whereas in the trivial regime (e) escape probability is concentrated only near the initial location of the particle (at x0 = 6). Error bars are standard deviations across 8 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Employing a form of zero noise extrapolation (ZNE), we regressed the acquired data into the λ = 0 noiseless limit, taking into account physicality constraints such as particle number conservation and non-negativity of occupancy densities. As evident in Fig. 2a, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt data after ZNE closely matched theoretical expectations calculated via exact diagonalization (ED) and presented a marked improvement over the λ = 1 data before ZNE. In addition to ZNE, we also employed readout error mitigation (RO), which approximately corrects bit-flip errors in measurement outcomes on the quantum processor (see Methods). Results without RO are inferior in accuracy and are shown in Fig. 2a as translucent lines. By performing time-integration on the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt data, as described in Eq. (5), we recovered the normalization factor of the quantum state and thereby rescaled \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt into occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt on the physical state ωt, the norm of which decreases over time (as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is lossy). A second time-integration on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt yielded the escape probabilities Px(t) on unit cells x. We present these cell-resolved Px(t) in Fig. 2b. The short-time increase of Px(t) at x = 6 is expectedly due to the initial localization of the particle on that unit cell, but as time progresses and the particle diffuses asymmetrically leftward under the NHSE, accumulation of Px(t) at cells of smaller x occurs. We observe the impact of ZNE on the accuracy of the data, which brought the experiment Px(t) into close agreement with ED predictions. In Fig. 2c, we report the total escape probability summed over all cells of the ladder, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(t)={\sum }_{x=1}^{N}{P}_{x}(t)$$\end{document}P(t)=∑x=1NPx(t), which monotonically approached unity; the experiment was terminated when P(t) reached sufficiently close to unity. Finally in Fig. 2d we present cell-resolved final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px along the ladder, showing raw experiment data, data with RO only, with ZNE only, and with both ZNE and RO applied. We had tuned \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb to be in a regime supporting the edge burst, here clearly manifesting as a prominent spike in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px at the x = 1 boundary—larger, in fact, than at the x0 = 6 initial localization of the particle. These results highlight the relevance of error mitigation: the signature of the edge burst is washed out without ZNE and RO, whereas with mitigation the escape probabilities closely agree with theory. Indeed, we emphasize the inherent sensitivity of the experimental measurements of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px: small inaccuracies in the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt data, such as the suppressed oscillations in Fig. 2a, can accumulate over the time-integration into large deviations in Px(t) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px at long times. This sensitivity places considerable demand on the robustness of the quantum simulation methodology and hardware, and underscores the difficulty of all experiments in our study. In Fig. 2e we report the counterpart of Fig. 2d on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb in the trivial regime, which does not exhibit the edge burst. The only peak in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px occurs at the initial location of the particle, reflecting the decay of the particle in-situ with no physically significant dynamics over time. We describe additional experiments on the same N = 8 ladder but with initial particle localization at x0 = 4, and on a smaller N = 4 ladder, in Supplementary Note 4A. The conclusions from those results are qualitatively identical to those drawn here, and the real-space signature of the edge burst in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px were likewise clearly observed. Demonstration of edge burst on larger lattices We next investigated models with larger unit-cell counts. Aside from demonstrating the versatility of our quantum simulation approach, experiments on larger ladders also minimize finite-size effects and produce clearer evidence of the non-Hermitian edge burst. Here, we realize an N = 64 unit-cell (128-site) ladder, which is 8 times the system size examined in the previous section. At these larger lattice sizes, trotterization and standard circuit transpilation produce circuits far too deep for present NISQ hardware to feasibly accommodate—naïve execution of these circuits would result in near-complete decoherence of the quantum state and vanishing signal-to-noise ratios when measuring observables. To overcome this present limitation, we employed an additional tensor-network-aided circuit recompilation technique45–47,60 for circuit compression, which replaces components of the circuits with approximate lower-depth parametrized ansatzes that are variationally optimized (see Methods). In this process, we exploited symmetries of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb, such as number conservation, to enhance circuit construction performance and quality. We emphasize that circuit recompilation was only required here to remain within current NISQ hardware limitations. As hardware progress continues to drive qubit and gate error rates down, our time-stepping LCU-based approach can revert to time-evolution circuit construction schemes of negligible classical cost (e.g., product formulae), thus retaining scalability at classical and quantum levels. The low overhead and operational simplicity set our method apart from prior methods employed in quantum simulation, such as quantum imaginary-time evolution schemes59,60,62 which demand tomographically complete measurements within a correlation domain at every time step to construct the next, hardware-efficient block-encodings53,54 that require exponentially growing classical resources, or variational methods in specific settings50,55,61,63 entailing measurements and classical optimization run iteratively. In the left panel of Fig. 3a, we report site-resolved occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt on the N = 64 ladder in the edge burst regime, as obtained on hardware with ZNE and RO error mitigation applied. Underlying this data were \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\rho }_{t}}$$\end{document}nxℓρt measurements that had been time-integrated. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is lossy, the occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt decay with time, reflecting particle escape from the ladder. In the right panel we report the cell-resolved escape probabilities Px(t) time-integrated from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt.Fig. 3Quantum simulation of edge-burst dynamics on a 64-unit-cell lattice.a Site-resolved occupancy densities and unit-cell-resolved escape probabilities Px(t) over time on a 64-unit cell quantum ladder in the edge burst regime measured on quantum hardware, with zero-noise extrapolation (ZNE) and readout error mitigation (RO) applied. The particle was initially localized at x0 = 51. Proper quantum state normalization was recovered via time-integration of measured occupancy densities and applied to both panels. The total escape probability P(t) approaches unity as time progresses. b Same as (a) but in the trivial regime. c Final unit-cell-resolved escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px in the long-time limit, comparing results in the edge burst regime and in the trivial regime. Data obtained on quantum hardware with ZNE and RO applied, RO only, and exact numerics are shown. Error bars are standard deviations across 16 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Quantum simulation of edge-burst dynamics on a 64-unit-cell lattice. a Site-resolved occupancy densities and unit-cell-resolved escape probabilities Px(t) over time on a 64-unit cell quantum ladder in the edge burst regime measured on quantum hardware, with zero-noise extrapolation (ZNE) and readout error mitigation (RO) applied. The particle was initially localized at x0 = 51. Proper quantum state normalization was recovered via time-integration of measured occupancy densities and applied to both panels. The total escape probability P(t) approaches unity as time progresses. b Same as (a) but in the trivial regime. c Final unit-cell-resolved escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px in the long-time limit, comparing results in the edge burst regime and in the trivial regime. Data obtained on quantum hardware with ZNE and RO applied, RO only, and exact numerics are shown. Error bars are standard deviations across 16 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. For comparison, we show in Fig. 3b the counterpart to Fig. 3a for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb in the trivial regime. In both set-ups, asymmetrical leftwards drift of the particle along the ladder driven by the NHSE is clearly observed; but in the edge burst regime the decay of the particle is significantly slower, allowing the particle to reach the x = 1 boundary with non-negligible amplitude (i.e. survival probability). Both experiments were terminated when the total escape probability P(t) reached close to unity. Finally, we present the final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px in Fig. 3c for both regimes. As before, the edge burst manifests as a prominent spike in escape probability at the x = 1 boundary, which is absent in the trivial regime. This demanding setting at large lattice size reinforces the relevance of a robust quantum simulation methodology integrated with effective error mitigation: the experimental \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px data closely agrees with theory with mitigation fully employed. Spectral information of the edge burst Time-evolution on the non-Hermitian quantum ladder Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb comprises conceptually of two intermixed components—real time-evolution on the Hermitian part of the Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}HHeb and imaginary time-evolution on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}HAeb. Unlike real-time evolution, imaginary time-evolution on a Hermitian Hamiltonian is not energy-conserving and, in fact, yields quantum states of extremal energy at long times. Thus, by evolving to long times under \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb, we can purify a starting state into an eigenstate of extremal imaginary eigenenergy. An illustration of this type of quantum circuit we executed is shown in Fig. 4a. To ensure nonzero overlap with the extremal eigenstates, we initiated time-evolution with the infinite-temperature Gibbs state (i.e., maximally mixed state). This is prepared through a completely depolarizing channel implemented in constant depth through a single round of mid-circuit measurements whose outcomes are discarded (see Methods). The time-evolution quantum algorithm for non-Hermitian Hamiltonians, as described earlier (drawn in Fig. 1c), is invoked following state preparation. After evolution for a sufficiently long time, we measured the imaginary energy of the quantum state through Hamiltonian averaging on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}HAeb (see Methods).Fig. 4Probing spectral properties of the edge burst system on a quantum processor.a Schematic of the circuits used to measure extremal imaginary energy. The system is prepared in a Gibbs state and time-evolved under the non-Hermitian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb for long times. The expectation value of the anti-Hermitian component of the Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}−iHAeb is measured on the time-evolved state. b Maximum and minimum imaginary energy versus hopping amplitudes v1/v2, showing hardware data measured on an N = 16 quantum ladder with and without ZNE and RO, and exact numerics. The v1/v2 ≤ 1 regime, wherein the dissipative gap Δ closes, supports the edge burst; the v1 > v2 regime is trivial. Error bars are standard deviations across 25 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. c Complex energy spectra of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb along indicated values of v1/v2 from numerics. The maxima and minima of these spectra are traced in (b). The horizontal and vertical reflection symmetries of the spectra are protected by symmetries of the Hamiltonian. Probing spectral properties of the edge burst system on a quantum processor. a Schematic of the circuits used to measure extremal imaginary energy. The system is prepared in a Gibbs state and time-evolved under the non-Hermitian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb for long times. The expectation value of the anti-Hermitian component of the Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}−iHAeb is measured on the time-evolved state. b Maximum and minimum imaginary energy versus hopping amplitudes v1/v2, showing hardware data measured on an N = 16 quantum ladder with and without ZNE and RO, and exact numerics. The v1/v2 ≤ 1 regime, wherein the dissipative gap Δ closes, supports the edge burst; the v1 > v2 regime is trivial. Error bars are standard deviations across 25 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. c Complex energy spectra of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb along indicated values of v1/v2 from numerics. The maxima and minima of these spectra are traced in (b). The horizontal and vertical reflection symmetries of the spectra are protected by symmetries of the Hamiltonian. We report the extremal imaginary eigenenergies ℑ EℑE measured on hardware as a function of hopping amplitudes v1/v2 in Fig. 4b. Of particular relevance is the largest imaginary eigenenergy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\max {{\Im}} E$$\end{document}maxℑE, which is zero for v1 ≤ v2 but strictly negative for v1 > v2. That is, the imaginary gap \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta=-\max {{\Im}} E$$\end{document}Δ=−maxℑE, also known as the dissipative gap, of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb closes for v1 ≤ v2. In Fig. 4c we illustrate the complex energy spectrum of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb at various v1/v2, which makes clear the recession of the spectrum into the negative imaginary half-plane as v1 exceeds v2. We remark that the horizontal and vertical reflection symmetries of the complex spectra, about ℜ E = 0ℜE=0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \Im E = - \gamma/2$$\end{document}ℑE=γ/2, are a consequence of chiral and time-reversal symmetries of the quantum ladder model (see Supplementary Note 1A). Prior theoretical analysis14 indicated that the closing of the dissipative gap Δ Δ is a necessary condition for the edge burst to occur in addition to the presence of the NHSE. Intuitively, the zero imaginary eigenenergy modes suffer no decay during time evolution and are long-lived, thus enabling an initial wavepacket driven by the NHSE to reach the boundary of the system. The particle is then trapped against the boundary by the NHSE and decays, giving rise to the edge burst. In contrast, upon opening of the dissipative gap, any wavepacket suffers exponential loss during propagation and cannot reach the boundary before decay. This qualitative difference was clearly observed, for example, in our experimental results in Fig. 3. Indeed, in all our experiments, the edge burst regime—where edge burst signatures are present—occur when v1 ≤ v2, and the trivial regimes occur when v1 > v2. Indeed, our results demonstrate the closure of the dissipative gap and the change in edge burst phenomenology across the transition boundary directly on a quantum processor. Spatially extended and patterned edge bursts with multiple interacting particles We now turn to our key set of results: edge burst phenomena observed with multiple interacting particles. Systems hosting strongly interacting quantum particles are difficult to realize on classical or effectively single-body simulators, such as electrical (i.e., topolectrical) circuits30,64–66 or waveguide photonic systems29,67,68, that have thus far been used to study non-Hermitian physics—let alone achieving tunability of interaction strengths, ranges and types. In this respect, digital quantum simulation on quantum hardware, making use of the full many-body Hilbert space hosted by the device and programmable quantum operations in that space, presents a clear versatility advantage. The quantum simulation approach we developed easily accommodates arbitrary interactions in the Hamiltonian (see Methods), and here we exploit this versatility to explore non-Hermitian edge burst phenomenology enriched by interactions. To start, we examined two particles (hardcore bosons) on an N = 12 ladder in the edge burst regime with U1 > 0 interactions switched on—that is, repulsive density-density interactions acting between sites on the ladder a range r = 1 apart (see Fig. 1b for illustration). The energy scale of the interaction was comparable to those of the non-interacting parts of the Hamiltonian (hoppings v1, v2 and loss γ). We present experiment results in Fig. 5a, showing the measured site-resolved occupancy densities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{x\ell }\right\rangle }_{{\omega }_{t}}$$\end{document}nxℓωt, recovered cell-resolved escape probabilities Px(t), and final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px. As in our prior experiment on the vanilla edge burst at large lattice size, we employed circuit recompilation to compress circuit depth and utilized ZNE and RO error mitigation to address hardware noise.Fig. 5Spatially extended edge bursts with multiple interacting particles.a Site-resolved occupancy densities and unit-cell-resolved escape probabilities Px(t) measured on a 12-unit cell quantum ladder in the edge burst regime hosting two interacting particles. Final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px obtained on hardware with and without zero-noise extrapolation (ZNE) and readout error mitigation (RO), and exact numerics, are shown. An edge burst spatially extended across two unit cells is observed, highlighted as black shading in the cartoon inset. b Same as (a) but in the trivial regime, which does not give rise to an edge burst. c, d By increasing the range of interactions on the quantum ladder, spatially extended edge bursts separated by one or more unit cells arise. Gray arrows denote the initial localization of the particles. Error bars are standard deviations across 10 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Spatially extended edge bursts with multiple interacting particles. a Site-resolved occupancy densities and unit-cell-resolved escape probabilities Px(t) measured on a 12-unit cell quantum ladder in the edge burst regime hosting two interacting particles. Final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px obtained on hardware with and without zero-noise extrapolation (ZNE) and readout error mitigation (RO), and exact numerics, are shown. An edge burst spatially extended across two unit cells is observed, highlighted as black shading in the cartoon inset. b Same as (a) but in the trivial regime, which does not give rise to an edge burst. c, d By increasing the range of interactions on the quantum ladder, spatially extended edge bursts separated by one or more unit cells arise. Gray arrows denote the initial localization of the particles. Error bars are standard deviations across 10 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Here, interestingly, from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px we observe the manifestation of a spatially extended version of the edge burst, spanning two unit cells (x ∈ {1, 2}) instead of the typical single unit cell in vanilla edge bursts. The mechanism underlying this spatial extension stems from a Pauli exclusion-like effect arising from the interactions. At short times, both particles expectedly diffuse to the left under the NHSE, but as the first impacts on the x = 1 boundary, the hardcore nature of the bosons and the U1 interactions inhibit the second from also occupying the first unit cell—effectively, the second boson encounters a virtual boundary at x = 2. This is akin to the formation of the so-called Fermi skin in fermionic non-Hermitian systems25,69, wherein fermions are pushed up against a boundary by the NHSE and thereby exhibit a Fermi surface in real space. Here, the second boson is trapped at the x = 2 unit cell and decays, giving rise to a spatially extended edge burst occurring over two unit cells. In the trivial regime, the edge burst does not manifest, as shown in Fig. 5b. Moreover, we observed that by tuning the range of interactions, spatially patterned manifestations of the edge burst can arise. In Fig. 5c we present experimental results on an N = 14 ladder hosting two bosons with Ur > 0 switched on for r ≤ 3, and in Fig. 5d we present results with Ur > 0 for r ≤ 5. In the former, we witness a spatially extended edge burst occupying alternate unit cells, and in the latter the edge burst is separated by two unit cells of small escape probability. The cartoon insets in Figs. 5c and d highlight these patterns. We refer to these as δ = 2 and δ = 3 patterned edge bursts, respectively, where δ records the unit-cell separation of the edge burst (i.e., escape probability) peaks; the spatially extended edge burst in Fig. 5a then corresponds to δ = 1, and the conventional single-unit-cell edge burst (e.g., Figs. 2 and 3) has δ = 0. The mechanism underlying the δ > 1 patterned bursts here is similar to the δ = 1 setting, but here the longer-ranged interactions induce virtual boundaries located farther from the first boson at x = 1. The same phenomenon occurs when p > 2 particles are present, which gives rise to spatially extended edge bursts occurring over p unit cells. Generally, the presence of Ur > 0 interactions for r ≤ 2d − 1 can produce δ = d patterned edge bursts; with p bosons, the same δ separation in peaks occurs p times starting from the left boundary of the ladder (the drive direction of the NHSE). In Supplementary Note 4B, we describe additional experiments conducted with three bosons yielding a three-unit-cell spatially extended edge burst. We also observed qualitatively similar δ = 2 and δ = 3 patterning under longer-range interactions on the three-boson setup. Likewise, we verified that the edge bursts do not arise in the trivial regime. In the settings examined here (in Fig. 5), the initial conditions were such that the bosons do not significantly infringe into the spatial ranges of Ur interactions under the dynamics of the system—i.e. they do not approach too closely to one another—until they impact on the left boundary. This allows the discussed physical mechanism to carry through unimpeded. As the leftward drift induced by the NHSE is approximately uniform, this condition is simple to achieve and boils down to the initial localization of the bosons. Indeed, in Figs. 5a, c, d, with Ur > 0 for r ≤ 2d − 1, the bosons were initially separated by at least 2d + 1 sites on the ladder or equivalently ~d unit cells. When this condition is not satisfied, a different type of interaction-driven burst phenomenology can occur, as we describe next. Interaction-induced cluster bursts Finally, we demonstrate that with appropriately configured interactions, anomalously high escape probabilities can be induced at chosen locations on the quantum ladder away from boundaries, in fact regardless of whether the Hamiltonian is in the (canonical) edge burst or trivial regimes. We term this phenomenon cluster bursts. Unlike the canonical edge burst, cluster bursts need not occur at the edges of the system and arise intrinsically from interactions; there is no analogous single-body non-interacting counterpart known at present. We show in Fig. 6a experimental results on an N = 14 ladder in the canonical edge burst regime (v1 ≤ v2) hosting two bosons. The bosons were initially localized on the a-sublattices of unit cells x0 = 8, 9, and we enable density-density interaction U1 of energy scale comparable to the non-interacting parts of the Hamiltonian (v1, v2, γ). Thus, the initial conditions are such that the particle separation condition discussed in the previous section is not clearly satisfied—in particular, a single hop of either boson toward the other, via either v1 or v2 channels, places them within the range of interactions and causes a significant energy change. Strikingly, we observe in experimentally measured \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px that both bosons decay essentially in-situ, with an overwhelming proportion of escape probability concentrated precisely at x = 8, 9. The same cluster burst arises when the Hamiltonian is tuned into the trivial regime, as we show in Fig. 6b, with qualitatively similar features in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px.Fig. 6Cluster bursts away from boundaries induced by interactions.a Site-resolved occupancy densities, unit-cell-resolved escape probabilities Px(t), and total escape probability P(t) summed over all unit cells, measured on a 14-unit cell quantum ladder in the edge burst regime hosting two interacting particles. Final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px obtained on hardware with zero-noise extrapolation (ZNE) and readout error mitigation (RO), and exact numerics, are shown. Initial localizations of the two particles (gray triangles) are barely outside the range of interactions (red arrow). b Same as (a) but with the quantum ladder in the trivial regime. c, d Measurements on the same quantum ladder as a, b but with the initial localizations of the particles (gray triangles) inside the range of interactions (red arrow), with the ladder in (c) the edge burst regime and (d) the trivial regime. In all cases, a cluster burst is observed in the vicinity of the initial locations of the particles, far from boundaries. Error bars are standard deviations across 10 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. Cluster bursts away from boundaries induced by interactions. a Site-resolved occupancy densities, unit-cell-resolved escape probabilities Px(t), and total escape probability P(t) summed over all unit cells, measured on a 14-unit cell quantum ladder in the edge burst regime hosting two interacting particles. Final escape probabilities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}$$\end{document}Px obtained on hardware with zero-noise extrapolation (ZNE) and readout error mitigation (RO), and exact numerics, are shown. Initial localizations of the two particles (gray triangles) are barely outside the range of interactions (red arrow). b Same as (a) but with the quantum ladder in the trivial regime. c, d Measurements on the same quantum ladder as a, b but with the initial localizations of the particles (gray triangles) inside the range of interactions (red arrow), with the ladder in (c) the edge burst regime and (d) the trivial regime. In all cases, a cluster burst is observed in the vicinity of the initial locations of the particles, far from boundaries. Error bars are standard deviations across 10 experiment runs. See Supplementary Tables 2 and 3 for Hamiltonian parameter values and superconducting quantum devices used. This same phenomenon can also arise when the initial localizations of the bosons are inside the range of interactions, such that outward hopping puts them outside their interaction range and causes an energy change. In this case, the bosons decay and a cluster burst emerges within the spatial confines of the interaction range. We show an example on the same N = 14 ladder in Figs. 6c and d tuned into the edge burst and trivial regimes, respectively. As before, the bosons were initialized on unit cells x0 = 8, 9, but here interactions Ur > 0 were switched on for r ≤ 3. In both edge burst and trivial regimes, we observe clearly a cluster burst in the vicinity of the initial boson locations, far from boundaries. The mechanism underlying these cluster bursts stems from a combination of the spectral structure of the system and initialization of the bosons on the ladder. First, the strong density-density interactions Ur of finite range produce a clear separation in the energy scale of eigenstates dependent on particle separation, and the initialization of the system is such that only one of the two spectral components carries weight (i.e., overlap with the initial state). Under interaction-induced deformation of the spectrum, this component need not exhibit an imaginary gap closing in both v1/v2 < 1 and v1/v2 > 1 regimes, and the purely negative imaginary eigenenergies result in rapid wavefunction decay before the bosons can reach the boundary. Thus the edge burst does not emerge regardless of the v1/v2 identification of nontrivial and trivial regimes. (See Supplementary Note 4C for further details.) Second, the energy separation in the spectral components induces a slowdown of dynamics due to (partially) forbidden state transitions. Recall that energy is conserved in time-translation-symmetric closed quantum systems—i.e., the energy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{tr}}}}({\omega }_{t}{{{\mathcal{H}}}})$$\end{document}tr(ωtH) of a quantum state ωt is fixed under time evolution by a Hermitian Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H. Strong interactions can then cause energy differences between states connected by the Hamiltonian so large as to inhibit transitions, as there is no available linear combination of accessible states after the transition that maintains the conserved energy. Indeed, this mechanism underlies a doublon decay phenomenon observed on Hubbard models70,71, where pairs of excitations in proximity exhibit exponentially long lifetimes, and is also responsible for the well-studied Stark many-body localization72,73. A similar understanding holds in a non-Hermitian context, with the modification that energy is not exactly conserved but there exists nonetheless a speed limit on how fast energy can change in the system, imposed by the Hamiltonian (see Supplementary Note 1B). On the quantum ladder, particle movement becomes constrained, and this slowdown in dynamics freezes the bosons long enough that they decay largely in their initial locations, generating the cluster burst observed. In the scenarios demonstrated in Fig. 6, for example, two simultaneous hoppings of the bosons must occur for them to move along the NHSE drive direction without changing their interaction energy, which is a second-order process and happens with low amplitude. In the unlikely event that these higher-order hoppings occur and the bosons reach the boundary of the system, spatially extended edge bursts of small amplitude can additionally manifest, as can be seen in Fig. 6a. (Further technical elaboration of the cluster burst mechanism is provided in Supplementary Note 4C.) Discussion While the experimental frontier of non-Hermitian condensed-matter physics has to-date enjoyed a spectacular period of progress utilizing broad palettes of custom-designed analog optical, metamaterial, and classical simulator platforms, the direct realization of important classes of non-Hermitian systems on near-term (i.e., NISQ) quantum platforms has remained limited. The advantages offered by digital quantum simulation are clear and tantalizing: the natural ability to access the many-body Hilbert space of arbitrary Hamiltonians, and in association, liberal versatility in tuning interactions of different ranges, strengths, and types. But leveraging quantum processors to probe non-Hermitian systems is challenging, as quantum gate sets are unitary and thus sophisticated methods must be developed to achieve the non-unitary evolution associated with non-Hermitian Hamiltonians. Moreover, the overhead incurred in doing so invariably runs up against qubit error and lifetime limitations. Nevertheless, recent studies of various condensed-matter phenomena on digital quantum computers, ranging from discrete time crystals39,74,75 to topological phases39,43–47,76–78, illustrate promising capabilities despite hardware constraints. It is therefore timely to investigate the prospects of studying non-Hermitian quantum physics on quantum hardware. Here, we probed the recently discovered phenomenon of non-Hermitian edge bursts on transmon-based superconducting quantum processors, and observed at high fidelity the signatures of the edge burst on up to 64 unit cells of a lossy quantum ladder model. Furthermore, by incorporating sequences of density-density interactions, we unveiled the possibility of engineering spatially extended and patterned variants of the edge burst, as well as bursts occurring away from boundaries, which we dubbed cluster bursts. Our present realization of the non-Hermitian edge burst on an intrinsically quantum platform, together with our experimental study of its interacting variants, complements prior experimental efforts on the canonical non-interacting version of the edge burst79,80. To enable this advance, we developed a general methodology for efficient non-Hermitian Hamiltonian simulation on digital quantum processors, leveraging a linear combination of unitaries circuit construction technique with ancillary qubit re-use, which is applicable to generic non-Hermitian models beyond the scope of the edge burst. We comment on several avenues for further work. The first is the prospect of exploring finite-size scaling relations of edge burst phenomena on quantum hardware. The canonical single-particle edge burst is associated with a scaling relation between the boundary loss peak and representative bulk loss. Concretely, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{\rm{edge}}} \sim d^{-\alpha} $$\end{document}Pedge~d−α and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{{\rm{bulk}}} \sim d^{-(\alpha + 1)} $$\end{document}Pbulk~d−(α+1), for escape probability P and initial particle distance to the boundary d, are expected14. By running copies of our experiments at various N and d, this scaling relation could be directly measured. But as the single-particle edge burst is increasingly well-understood, a more interesting question lies in mapping out the possibility of analogous bulk-edge scaling relations in the multi-particle interacting setting. The quantum simulation approach developed here is particularly well-suited for exploring such interacting physics. A second direction regards integrating more sophisticated readout strategies. At present, computational-basis readout is well-matched to the primary observable of the edge burst, namely the site-resolved escape probabilities reconstructed from occupation densities. However, the enriched phenomenology of the edge burst with interactions motivates a broader set of diagnostics than density profiles alone. Randomized local measurements, as in shadow tomography81–84, could allow the same family of evolved states to be post-processed for many observables, such as connected or higher-order density correlators, subsystem entanglement, and structure factors resolving spatial patterning. The ability to measure diverse diagnostics, albeit at higher overhead, can enable finer-grained probes of the physics involved15. Lastly, we remark on the prospect of quantum advantage. The present experiments demonstrate our methods for non-Hermitian Hamiltonian simulation but are not beyond classical capabilities. A limitation to the classical hardness of simulation is that, at fixed particle number p, the dimensionality of the Hilbert space accessed under dynamics grows polynomially with lattice size N as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left({N}^{p}\right)$$\end{document}ONp, not exponentially. Moreover, the physics of the edge burst and its interacting analogs studied here require neither very large N nor p, or breaking number conservation, to manifest. However, theoretical opportunities exist for quantum advantage even in few-body simulation85–88. For example, on condensed-matter lattice Hamiltonians, an advantage in the dimensionality of the lattice manifests45. Thus, with sufficiently good hardware and suitably choosing a model to study, there remains exciting potential for quantum simulation to present meaningful utility. The present work opens the door for future experimental investigation of quantum non-Hermitian condensed-matter physics. As quantum hardware continues to advance, and digital quantum computation moves toward early fault-tolerance, we anticipate that similar low-overhead methods will enable the direct study of more sophisticated systems exhibiting rich intertwined physics, such as critical versions of non-Hermitian pumping26 and the interplay between non-Hermiticity and entanglement phase transitions89. Methods Quantum hardware We used IBM transmon-based superconducting quantum devices in our experiments. These included 27-qubit devices ibm_hanoi and ibm_mumbai hosting the Falcon processor, 127-qubit devices ibm_sherbrooke, ibm_osaka, ibm_kyoto and ibm_nazca hosting the Eagle processor, and a 133-qubit device ibm_torino hosting the Heron processor. The basis gate sets of all devices comprise 1-qubit gates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{{{{\rm{X}}}},\sqrt{{{{\rm{X}}}}},{{{\rm{RZ}}}}\}$$\end{document}{X,X,RZ}, with RZ implemented virtually via framechanges90, and 2-qubit gate CX for Falcon processors, echoed cross resonance (ECR) for Eagle processors91, and CZ for the Heron processor. The CX, ECR, and CZ gates are equivalent up to 1-qubit rotations. We construct all experiment circuits using CX gates, including the application of protocols for error suppression and mitigation such as Pauli twirling and randomized gate folding for zero-noise extrapolation (see below), and transpile to the native 2-qubit gate of each device before execution. Typical performance metrics of the devices, such as relaxation T1 and dephasing T2 times, gate and readout error rates, and gate times, are provided in Supplementary Table 4. Model We examined the bosonic quantum ladder Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}={{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}+{{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb=H0eb+Hinteb, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}$$\end{document}H0eb and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}$$\end{document}Hinteb comprise non-interacting and interacting terms respectively, given by 6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}= \mathop{\sum }_{x=1}^{N}\left\{-i\gamma {c}_{xb}^{{{\dagger}} }{c}_{xb}+\left({v}_{1}{c}_{xa}^{{{\dagger}} }{c}_{xb}\right.\right.\\ +\frac{{v}_{2}}{2}\left[{c}_{(x+1)a}^{{{\dagger}} }{c}_{xb}+{c}_{(x+1)b}^{{{\dagger}} }{c}_{xa}\right]\\ \left.\left.+\frac{i{v}_{2}}{2}\left[{c}_{(x+1)a}^{{{\dagger}} }{c}_{xa}-{c}_{(x+1)b}^{{{\dagger}} }{c}_{xb}\right]+{{{\rm{h.c.}}}}\right)\right\},\\ {{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}= {U}_{0}{\sum }_{z=1}^{2N}{n}_{z}\left({n}_{z}-1\right)+{\sum}_{r\ge 1}{U}_{r}\left({\sum }_{z=1}^{2N}{n}_{z}{n}_{z+r}\right),$$\end{document}H0eb=∑x=1N−iγcxb†cxb+v1cxa†cxb+v22c(x+1)a†cxb+c(x+1)b†cxa+iv22c(x+1)a†cxa−c(x+1)b†cxb+h.c.,Hinteb=U0∑z=12Nnznz−1+∑r≥1Ur∑z=12Nnznz+r,over N unit cells. Above, v1, v2 ≥ 0 are intra- and inter-cell hopping coefficients, γ > 0 is a loss rate on b-sublattices, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{0}\in {\mathbb{R}}$$\end{document}U0∈R is an on-site interaction strength, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{r}\in {\mathbb{R}}$$\end{document}Ur∈R are range-r density-density interaction strengths between sites on the flattened ladder. The operator cxℓ annihilates a particle at sublattice ℓ ∈ {a, b} of unit cell x ∈ [N], and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{z}={c}_{x\ell }^{{{\dagger}} }{c}_{x\ell }$$\end{document}nz=cxℓ†cxℓ is the number operator with index z = 2x − δℓa ∈ [2N] labeling sites on the ladder with sublattice structure flattened. Open boundary conditions (OBC) are imposed by zeroing cxℓ = 0 for x ∉ [N]; periodic boundary conditions (PBC) correspond to setting cxℓ = c(x+N)ℓ. The Hamiltonian is manifestly non-Hermitian (as γ > 0). Writing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}={{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}^{{{{\rm{eb}}}}}-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb=HHeb−iHAeb for Hermitian and anti-Hermitian components \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}HHeb and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}−iHAeb, we have 7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}=\gamma {\sum }_{x=1}^{N}{c}_{xb}^{{{\dagger}} }{c}_{xb},$$\end{document}HAeb=γ∑x=1Ncxb†cxb, containing the non-Hermitian loss. The non-interacting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}$$\end{document}H0eb is identical to the paradigmatic lossy quantum walk Hamiltonian studied in Refs. 14,17. Notably, the non-Hermitian skin effect (NHSE) is present for all v1 ≠ 0, as implied by an equivalence of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{0}^{{{{\rm{eb}}}}}$$\end{document}H0eb to the non-Hermitian Su-Schrieffer-Heeger (SSH) model19 with left-right asymmetric hoppings under a unitary basis transformation. Alternatively, the NHSE can be understood as arising from the suppression of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$+\hat{{{{\bf{x}}}}}$$\end{document}+x^-direction portion of flux-induced rotational motion by the γ loss. The imaginary (i.e. dissipative) gap closes14 for v1 ≤ v2, which is the canonical regime where the non-Hermitian edge burst manifests. In \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{int}}}}}^{{{{\rm{eb}}}}}$$\end{document}Hinteb, we take the U0 → ∞ limit for simplicity, such that hardcore bosonic statistics arise92. We allow Ur to be independently tuned for each r to access a variety of multi-particle interacting edge burst phenomena. Note that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb possesses a U(1) number conservation symmetry in the c bosons. Qubit encoding To perform simulation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb on quantum hardware, a mapping between basis states of the system and qubit states must be fixed, which also induces a mapping between the operators of the system (cxℓ) and qubit operators. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is number-conserving, its Hilbert (Fock) space is the direct sum of disconnected Fock sectors identified by particle numbers. We leverage this structure in state encoding. In the single-particle sector, we identified 8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\vert (x,\ell )\right\rangle }_{{{{\rm{sys}}}}}={\left\vert 2x-{\delta }_{\ell a}-1\right\rangle }_{{{{\rm{q}}}}},$$\end{document}(x,ℓ)sys=2x−δℓa−1q, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\vert (x,\ell )\right\rangle }_{{{{\rm{sys}}}}}$$\end{document}(x,ℓ)sys is the system state with a particle at unit cell x ∈ [1, N] and sublattice l ∈ {a, b}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\vert 2x-{\delta }_{\ell a}-1\right\rangle }_{{{{\rm{q}}}}}$$\end{document}2x−δℓa−1q is the associated computational basis state on the qubits. (Throughout our work an implicit conversion of integer labels for qubit states to binary representation is assumed, that is, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert z\right\rangle=\left\vert \overline{z}\right\rangle=\left\vert {\overline{z}}_{1}\right\rangle \otimes \left\vert {\overline{z}}_{2}\right\rangle \otimes \cdots$$\end{document}z=z¯=z¯1⊗z¯2⊗⋯ for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z\in {\mathbb{N}}$$\end{document}z∈N and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{z}$$\end{document}z¯ the binary representation (i.e. bitstring) of z, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overline{z}}_{1},{\overline{z}}_{2},\ldots$$\end{document}z¯1,z¯2,… are its bits.) Thus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{{{{\rm{q}}}}}=\lceil {\log }_{2}2N\rceil$$\end{document}Nq=⌈log22N⌉ system qubits are required to represent the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb quantum ladder in this Fock sector. More generally, in the p-particle sector, we used a bijection between system states \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\vert \{({x}_{1},{\ell }_{1}),\ldots,({x}_{p},{\ell }_{p})\}\right.\rangle }_{{{{\rm{sys}}}}}$$\end{document}{(x1,ℓ1),…,(xp,ℓp)}⟩sys, which denote particle occupation at distinct sites {(x1, ℓ1), …, (xp, ℓp)} for xk ∈ [1, N], ℓk ∈ {a, b}, and qubit states in lexicographic ascending order. Thus minimally \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{{{{\rm{q}}}}}=\lceil {\log }_{2}(2N)!/(2N-p)!\rceil \in {{{\mathcal{O}}}}\left(p{\log }_{2}N\right)$$\end{document}Nq=⌈log2(2N)!/(2N−p)!⌉∈Oplog2N qubits are required to represent the quantum ladder. This mapping is similar to first-quantization encoding used in quantum chemistry88,93,94 as opposed to second-quantization mapping95, which requires a number of qubits independent of p but linear in N. Fock-sector error detection In our experiments, the number of available qubit states may exceed the number of system states—therefore some qubit states may be unused in state encoding. These qubit states have no physical meaning and should not be involved in time-evolution of any physical system state. Nonetheless, hardware noise may cause these states to be erroneously populated. We discard circuit shots that measured occupation of these states, as these results are unphysical and indicate hardware error. Time-evolution on non-Hermitian Hamiltonians Given a local Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H, which need not be Hermitian, an initial state ρ0, and an evolution time t ≥ 0, we describe an algorithm to perform normalized time-evolution, 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{0}\to {\rho }_{t}={{{\mathcal{N}}}}\left[{e}^{-i{{{\mathcal{H}}}}t}{\rho }_{0}{e}^{+i{{{{\mathcal{H}}}}}^{{{\dagger}} }t}\right],$$\end{document}ρ0→ρt=Ne−iHtρ0e+iH†t,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{N}}}}[\rho ]=\rho /{{{\rm{tr}}}}(\rho )$$\end{document}N[ρ]=ρ/tr(ρ) normalizes a quantum state. Writing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}={{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}H=HH−iHA such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-i{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}−iHA are respectively the Hermitian and anti-Hermitian components of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H, we split (i.e. trotterize) the evolution into m time steps, such that 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left({{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}\circ {{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}\right)}^{m}({\rho }_{0})={\rho }_{t}+{{{\mathcal{O}}}}\left(\frac{1}{m}\right),$$\end{document}MA∘MHm(ρ0)=ρt+O1m, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA are quantum maps that we seek to implement, satisfying 11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}(\rho ) ={e}^{-i{{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}\Delta t}\rho {e}^{+i{{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}\Delta t}+{{{\mathcal{O}}}}\left(\frac{1}{{m}^{2}}\right),\\ {{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}(\rho ) ={{{\mathcal{N}}}}\left[{e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}\rho {e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}\right]+{{{\mathcal{O}}}}\left(\frac{1}{{m}^{2}}\right),$$\end{document}MH(ρ)=e−iHHΔtρe+iHHΔt+O1m2,MA(ρ)=Ne−HAΔtρe−HAΔt+O1m2, for the time interval Δt = t/m and any arbitrary ρ. The error scaling in Eq. (10) implies that arbitrary precision in time evolution can be achieved by increasing m. The main difficulty is in realizing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA as the transformation is non-unitary. The core idea is to utilize a dilated Hilbert space with an ancillary register, which can be as small as a single qubit, and to implement a joint unitary such that the desired transformation is applied on the system register upon measurement of the ancillae. We detail a systematic non-variational approach for constructing such a circuit with classical processing costs polynomial in the size of the system (see below). For future use, we denote the Pauli decompositions 12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}={\sum }_{k=1}^{{K}_{{{{\rm{H}}}}}}{\alpha }_{k}{\sigma }^{k},\,\,\,\,{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}={\sum }_{k=1}^{{K}_{{{{\rm{A}}}}}}{\beta }_{k}{\chi }^{k},$$\end{document}HH=∑k=1KHαkσk,HA=∑k=1KAβkχk, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\alpha }_{k},{\beta }_{k}\in {\mathbb{R}}$$\end{document}αk,βk∈R and Pauli strings σk, χk assumed to be known from the given \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H is local, the number of Pauli terms KH, KA ∈ poly(N). Implementation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH The unitary channel \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH performs time-evolution by the Hermitian Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH and can be implemented through any product formula. In our work, we used the first-order Lie-Trotter formula to set 13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}(\rho )={U}_{{{{\rm{H}}}}}\rho {U}_{{{{\rm{H}}}}}^{{{\dagger}} },\,\,\,\,{U}_{{{{\rm{H}}}}}={\prod }_{k=1}^{{K}_{{{{\rm{H}}}}}}{e}^{-i{\alpha }_{k}{\sigma }^{k}\Delta t},$$\end{document}MH(ρ)=UHρUH†,UH=∏k=1KHe−iαkσkΔt, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{({\alpha }_{k},{\sigma }^{k})\}}_{k}$$\end{document}{(αk,σk)}k is the Pauli decomposition of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{H}}}}}$$\end{document}HH as in Eq. (12). Thus, the circuit for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH consists of consecutive layers of exponentiated Pauli strings, each implementable with a standard construction of an Rz rotation on a single qubit sandwiched by CX gates and single-qubit basis changes spanning the support of the Pauli string (see Supplementary Note 2A for further details). On n qubits with nearest-neighbor qubit connectivity, this standard implementation of an exponentiated Pauli string is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{O}}}}\left(n\right)$$\end{document}On depth in the worst case. We utilized several techniques to achieve circuits with fewer CX gates and lower depth than naïve construction. For isolated subsets of Pauli strings of weight ≤2 sharing the same support, we invoked KAK decomposition96, which produces circuit components for U(4) unitaries comprising 1-qubit rotations and at most 3 CX gates. For subsets of Pauli strings of weight ≤3 sharing the same support, we used the QSEARCH algorithm97 in the public BQSKit toolkit98 to produce circuit components, which promises optimal-depth synthesis for unitaries up to 4 qubits. For remaining Pauli strings of larger weight, we used a circuit construction procedure that takes advantage of simultaneous diagonalization99. We partitioned the Pauli strings into commuting groups by finding a clique cover on their commutation graph. (Determining the minimum clique cover, which is an NP-complete problem in the worst case, is not necessary for this method to work, though generally the fewer the number of cliques, the smaller the resultant circuit.) Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{P}_{k}\}}_{k}$$\end{document}{Pk}k be one such commuting group on n qubits; then there exists a Clifford unitary C that simultaneously diagonalizes the group, that is, CPkC† = Qk for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Q}_{k}\in {\{{\mathbb{I}},{\sigma }^{z}\}}^{\otimes n}$$\end{document}Qk∈{I,σz}⊗n, for all k. We solved for C and its circuit implementation, and the diagonal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{Q}_{k}\}}_{k}$$\end{document}{Qk}k, through a classically efficient stabilizer-based method99,100. The circuit for the exponentiated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{P}_{k}\}}_{k}$$\end{document}{Pk}k then follows the structure \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}^{{{\dagger}} }\left[{\prod}_{k}\exp (-i{Q}_{k}\theta )\right]C$$\end{document}C†∏kexp(−iQkθ)C, and suitable orderings of the Qk terms according to their support allow cascading cancellation of CX gates between neighboring terms99. This approach provides reductions in circuit depth and number of CX gates compared to the naïve concatenation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\exp (-i{P}_{k}\theta )$$\end{document}exp(−iPkθ) circuit components in practice. Implementation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA We devised a scheme based on the linear combination of unitaries (LCU) circuit pattern to implement the non-unitary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}e−HAΔt, which is equivalent to imaginary-time evolution under \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}HA. In general, equipped with unitaries U0, …, Ud−1 controlled respectively by the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 0\right\rangle,\ldots,\left\vert d-1\right\rangle$$\end{document}0,…,d−1 states of an ancillary qubit register, a compact circuit primitive is known101,102 that probabilistically implements the action (γ0U0 + … + γd−1Ud−1) for coefficients γk > 0 on a system register up to normalization of the resultant quantum state (see Fig. 1d). This LCU primitive requires \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lceil {\log }_{2}d\rceil$$\end{document}⌈log2d⌉ ancillary qubits, and comprises an initialization unitary V on the ancillary register dependent on the coefficients, application of the controlled Uk unitaries, and V† followed by measurements on the ancillary register. The primitive succeeds when the ancillae report \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 0\right\rangle$$\end{document}0 upon measurement and fails otherwise. We first establish a general method for an arbitrary Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}HA. The central idea is to approximate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}e−HAΔt using a linear combination of forward and backward real time-evolution. Specifically, we examine expansions of the type 14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{0}{\mathbb{I}}+{\sum }_{b=1}^{B}{A}_{b}\left({e}^{+i{{{\mathcal{R}}}}\Delta {\tau }_{b}}+{e}^{-i{{{\mathcal{R}}}}\Delta {\tau }_{b}}\right)={e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}+{{{\mathcal{O}}}}\left(\frac{1}{{m}^{\kappa }}\right),$$\end{document}A0I+∑b=1BAbe+iRΔτb+e−iRΔτb=e−HAΔt+O1mκ, for coefficients A0, …, AB ≥ 0, rescaled simulation times \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{\Delta {\tau }_{b}\}}_{b}$$\end{document}{Δτb}b, and a Hermitian auxiliary Hamiltonian \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{R}}}}$$\end{document}R satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{R}}}}}^{2}={{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}R2=HA. To satisfy error requirements of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA as declared in Eq. (11), we require an approximation order κ ≥ 2. The identity \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb{I}}$$\end{document}I and time-evolution propagators \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{\pm i{{{\mathcal{R}}}}\Delta {\tau }_{b}}$$\end{document}e±iRΔτb are unitary, thus their linear combination can be implemented by the LCU primitive above. In particular, the time-evolution propagators \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{\pm i{{{\mathcal{R}}}}\Delta {\tau }_{b}}$$\end{document}e±iRΔτb can be implemented via Lie-Trotter product formulae, in the same way as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{H}}}}}$$\end{document}MH, and the ancillae controls for the LCU need only be added to the central Rz rotation for each exponentiated Pauli string. We refer readers to Supplementary Note 2 for elaboration on the lower-level implementation details. For any κ ≥ 2, solutions for the expansion can be identified by matching Taylor expansion terms by order on both sides of Eq. (14). In broad families of condensed-matter lattice models, either choices of Hermitian roots \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{R}}}}$$\end{document}R are known or analytical properties of the Hamiltonians enable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{R}}}}$$\end{document}R to be efficiently computed. For example, the roots of topological insulators and superconductors in one and higher dimensions103–106, and generalized classes of symmetry-protected topological or topologically ordered systems107–112 are well-studied. In cases where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{R}}}}$$\end{document}R is difficult or inefficient to determine, we also provide an alternative method relying only on knowledge of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}HA—see Supplementary Note 2B. Minimally, choosing d = 2 terms in the LCU necessitates only a single ancillary qubit. This corresponds to a cosine approximation to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}e−HAΔt with a single pair B = 1 in Eq. (14), with coefficients A0 = 0 and A1 = 1/2, and rescaled simulation times \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \tau=\sqrt{2\Delta t}$$\end{document}Δτ=2Δt, achieving approximation order κ = 2. Solutions with larger number of terms d achieving higher approximation order κ can readily be found—see Supplementary Table 7 for examples. In the present edge burst context, Hermitian auxiliary Hamiltonians \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{R}}}}}^{{{{\rm{eb}}}}}$$\end{document}Reb satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${({{{{\mathcal{R}}}}}^{{{{\rm{eb}}}}})}^{2}={{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}$$\end{document}(Reb)2=HA can be found analytically. However, more carefully exploiting the structure of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}HAeb in fact allows a solution to Eq. (14) with zero approximation error (effectively κ → ∞). As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}$$\end{document}HAeb comprises only on-site terms, we find 15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\Delta t}={\sum }_{x=1}^{N}\left({n}_{xa}+{e}^{-\gamma \Delta t}{n}_{xb}\right)=\frac{1}{2\eta }\left({U}_{+}^{{{{\rm{eb}}}}}+{U}_{-}^{{{{\rm{eb}}}}}\right),$$\end{document}e−HAebΔt=∑x=1Nnxa+e−γΔtnxb=12ηU+eb+U−eb, for forward and backward time-evolution unitaries \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{\pm }^{{{{\rm{eb}}}}}={e}^{\pm i{{{{\mathcal{H}}}}}_{{{{\rm{aux}}}}}^{{{{\rm{eb}}}}}}$$\end{document}U±eb=e±iHauxeb on the effective Hermitian Hamiltonian 16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}_{{{{\rm{aux}}}}}^{{{{\rm{eb}}}}}=\arccos ({e}^{-\gamma \left\vert \Delta t\right\vert })\left\{\begin{array}{ll}{\sum }_{x=1}^{N}{n}_{xb}\quad &\Delta t\ge 0,\\ {\sum }_{x=1}^{N}{n}_{xa}\quad &\Delta t < 0,\end{array}\right.$$\end{document}Hauxeb=arccos(e−γΔt)∑x=1NnxbΔt≥0,∑x=1NnxaΔt<0, and scaling factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta=\min (1,{e}^{\gamma \Delta t})$$\end{document}η=min(1,eγΔt). Here we present solutions for both Δt ≥ 0 and Δt < 0 cases for completeness. Thus an exact LCU construction to implement \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\Delta t}$$\end{document}e−HAebΔt is a priori known. We use this solution in our experiments, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${U}_{\pm }^{{{{\rm{eb}}}}}$$\end{document}U±eb implemented via the first-order Trotter-Lie product formula. Qubit reset and re-use We performed mid-circuit reset of the ancillary qubit, used in the LCU circuit implementation of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA map, to the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 0\right\rangle$$\end{document}0 state after each time step such that the same qubit is re-used over the entire time-evolution experiment. Each reset is performed by executing an X gate on the qubit conditioned on the outcome of the measurement in the preceding \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{M}}}}}_{{{{\rm{A}}}}}$$\end{document}MA circuit component. The mid-circuit readout and feedforward capability of the quantum processors113–115 enable this resource-saving measure. Quantum state normalization Quantum states on a quantum computer are normalized by the nature of the platform. Above, we described the implementation of normalized time-evolution of an initial state [Eq. (9)]. However, in some simulation settings, as in the present edge burst context, the process of interest is time-evolution without normalization, 17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{0}\to {\omega }_{t}={e}^{-i{{{\mathcal{H}}}}t}{\omega }_{0}{e}^{+i{{{{\mathcal{H}}}}}^{{{\dagger}} }t},$$\end{document}ω0→ωt=e−iHtω0e+iH†t, where ω0 = ρ0 is a normalized initial state but ωt is not normalized, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\rm{tr}}}}({\omega }_{t})\ne 1$$\end{document}tr(ωt)≠1. Enforcing normalization \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{N}}}}({\omega }_{t})={\rho }_{t}$$\end{document}N(ωt)=ρt produces the normalized time-evolution in Eq. (9), which is realized on the quantum platform. Here we examine the recovery of ωt from ρt. In particular, we describe two methods of recovering the quantum state normalization factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{t}=\sqrt{{{{\rm{tr}}}}({\omega }_{t})} > 0$$\end{document}At=tr(ωt)>0 such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\omega }_{t}={A}_{t}^{2}{\rho }_{t}$$\end{document}ωt=At2ρt. The first method is general for any lossy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H and t ≥ 0 such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\Vert {e}^{-i{{{\mathcal{H}}}}t}\right\Vert \le 1$$\end{document}e−iHt≤1, and works by examining the success probability St of the normalized time-evolution algorithm (as detailed above). In each time step, failure of the algorithm can occur, which is detected by a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 1\right\rangle$$\end{document}1 measurement outcome on the ancillary qubit, and indicates an incorrect projection of the quantum state that realizes a complementary action \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb{I}}-{e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}I−e−HAΔt instead of the desired \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}^{-{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\Delta t}$$\end{document}e−HAΔt. The decrease of quantum state norm over time directly manifests in the probability of success, in particular, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{t}^{2}\approx {S}_{t}$$\end{document}At2≈St. Thus St measured in experiments directly allow recovery of At, up to errors in circuit construction (e.g. trotterization and LCU approximation) and hardware noise that distort the observed St. See Supplementary Note 4D for technical elaboration and experimental investigation. The second method is through time-integration of site-resolved occupancy data, as described in the main text. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is lossy on b-sublattices, we obtain 18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{{\rm{d}}}}}{{{{\rm{d}}}}t}{{{\rm{tr}}}}({\omega }_{t})=-2{\left\langle {{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\right\rangle }_{{\omega }_{t}}=-2\gamma {A}_{t}^{2}{\sum }_{x=1}^{N}{\left\langle {n}_{xb}\right\rangle }_{{\rho }_{t}},$$\end{document}ddttr(ωt)=−2HAebωt=−2γAt2∑x=1Nnxbρt, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{xb}\right\rangle }_{{\rho }_{t}}={{{\rm{tr}}}}\left({\rho }_{t}{n}_{xb}\right)$$\end{document}nxbρt=trρtnxb is the occupancy of sublattice b of unit cell x as measured on the ρt state on the quantum platform. This is a restatement of Eq. (2) of the main text. At the same time, by definition of At, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({{{\rm{d}}}}/{{{\rm{d}}}}t){{{\rm{tr}}}}({\omega }_{t})=({{{\rm{d}}}}/{{{\rm{d}}}}t){A}_{t}^{2}{{{\rm{tr}}}}({\rho }_{t})=2{A}_{t}({{{\rm{d}}}}{A}_{t}/{{{\rm{d}}}}t)$$\end{document}(d/dt)tr(ωt)=(d/dt)At2tr(ρt)=2At(dAt/dt). Thus we have 19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{{{\rm{d}}}}{A}_{t}}{{{{\rm{d}}}}t}=-\gamma {A}_{t}{\sum }_{x=1}^{N}{\left\langle {n}_{xb}\right\rangle }_{{\rho }_{t}},$$\end{document}dAtdt=−γAt∑x=1Nnxbρt,with the solution for At reported in Eq. (5). Thus At can be recovered through numerical integration of b-sublattice occupancies measured on normalized time-evolved states in experiments, sampled over the time domain [0, t]. This is the method used in all our experiments discussed in the main text (Figs. 2–6) as well as Supplementary Figs. 2 to 5. Escape probabilities As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is lossy on b-sublattices, as reported in Eq. (3), the probability of escape from unit cell x by time t is given by 20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{x}(t)=2\gamma \int_{0}^{t}{\left\langle {n}_{xb}\right\rangle }_{{\omega }_{\tau }}\,{{{\rm{d}}}}\tau=2\gamma \int_{0}^{t}{A}_{\tau }^{2}{\left\langle {n}_{xb}\right\rangle }_{{\rho }_{\tau }}\,{{{\rm{d}}}}\tau,$$\end{document}Px(t)=2γ∫0tnxbωτdτ=2γ∫0tAτ2nxbρτdτ, where Aτ are quantum state normalization factors recovered from experiment data and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle {n}_{xb}\right\rangle }_{{\rho }_{\tau }}$$\end{document}nxbρτ are b-sublattice occupancies on normalized time-evolved states measured in experiments. The total escape probability across all unit cells is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(t)={\sum }_{x=1}^{N}{P}_{x}(t)$$\end{document}P(t)=∑x=1NPx(t) accordingly. Note Px(0) = P(0) = 0, and that Px(t) and P(t) are non-decreasing with t. In the long-time limit, each Px(t) approaches an asymptotic (i.e. steady-state) value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{P}}}}}_{x}={{\lim }_{t\to \infty }}{P}_{x}(t)$$\end{document}Px=limt→∞Px(t), and P(∞) = 1 as expected of a properly normalized probability distribution, consistent with Ref. 14. The dynamics of the system tends to completion as P(t) → 1, and there is no further purpose in assessing dynamics at larger times which contribute negligibly to escape probabilities. We terminate time-evolution at t sufficiently large to achieve P(t) ≳ 0.995 in our experiments, more than sufficient to clearly observe edge burst phenomenology as demonstrated. Measuring imaginary energy gap Time-evolution by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H as written in Eq. (9) encodes an effective imaginary-time evolution with respect to the anti-Hermitian component of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H in the following sense. Suppose \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert \phi \right\rangle$$\end{document}ϕ is a (right) eigenstate of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H with eigenenergy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E\in {\mathbb{C}}$$\end{document}E∈C, that is, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}\left| \phi \right\rangle=E\left| \phi \right\rangle$$\end{document}Hϕ=Eϕ. Then 21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {e}^{-i{{{\mathcal{H}}}}t}\left| \phi \right\rangle ={e}^{-iEt}\left| \phi \right\rangle={e}^{-i\Re (E)t}{e}^{{{\Im}}(E)t}\left| \phi \right\rangle,\\ \frac{\left\Vert {e}^{-i{{{\mathcal{H}}}}t}\left| \phi \right\rangle \right\Vert }{\left\Vert \left| \phi \right\rangle \right\Vert } ={e}^{{{\Im}}(E)t}.$$\end{document}e−iHtϕ=e−iEtϕ=e−iℜ(E)teℑ(E)tϕ,e−iHtϕϕ=eℑ(E)t. Let us consider the time-evolution of an initial state ρ0 which comprises a mixture of eigenstates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{H}}}}$$\end{document}H. Then for t > 0 (t < 0), the amplitudes of all other eigenstates become exponentially damped compared to the eigenstate with the largest (smallest) ℑ(E)ℑ(E). Thus, for sufficiently large \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert t\right\vert$$\end{document}t, the time-evolved state ρt purifies to an eigenstate of extremal ℑ(E)ℑ(E); the required \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert t\right\vert$$\end{document}t can be estimated by demanding that P(t) converges close to unity (see above). After time-evolution, the eigenenergy E can be measured through standard techniques. We used Hamiltonian averaging116–118 in our experiments, 22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\Im}}(E)=-{\left\langle {{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}\right\rangle }_{{\rho }_{t}}=-{\sum }_{k=1}^{{K}_{{{{\rm{A}}}}}}{\beta }_{k}{\left\langle {\chi }^{(k)}\right\rangle }_{{\rho }_{t}},$$\end{document}ℑ(E)=−HAρt=−∑k=1KAβkχ(k)ρt, where as before \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\left\langle O\right\rangle }_{{\rho }_{t}}={{{\rm{tr}}}}({\rho }_{t}O)$$\end{document}Oρt=tr(ρtO) is the measured expectation value of an observable O on ρt on the quantum device. As χ(k) are Pauli strings, measuring their expectation values on a quantum processor involves only Clifford basis changes before computational-basis readouts. To reduce the number of circuits required, commuting subsets of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{\chi }^{(k)}\}}_{k}$$\end{document}{χ(k)}k can be measured simultaneously on the same circuit, with the joint Clifford basis change given by an efficient stabilizer-based method100. We are interested in the imaginary energy (i.e., dissipative) gap in the present work, which is open when the largest ℑ(E) < 0ℑ(E)<0 and closed otherwise. We took the initial state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{0}={{{\mathcal{N}}}}({\mathbb{I}})$$\end{document}ρ0=N(I) to be the maximally mixed state, equivalently the infinite-temperature Gibbs state, prepared through a single round of mid-circuit computational-basis measurements on system qubits initialized in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert+\right\rangle$$\end{document}+ states, whose outcomes are discarded (see Supplementary Fig. 1 for circuit schematic). Circuit recompilation We follow the formulation established in prior works, in particular Refs. 45–47,60,119–121. Circuit recompilation is performed using a circuit ansatz whose parameters are dynamically optimized. We use an ansatz comprising an initial layer of single-qubit rotations (U3 gates) on all qubits followed by K ≤ 20 ansatz layers, each comprising a layer of CX gates entangling adjacent qubits and a layer of U3 rotations (see Supplementary Fig. 1). Each U3 gate in the ansatz is associated with rotation angles (θ, ϕ, λ); we collate them into a parameter vector ϑ. Then, given a target circuit component unitary V and an initial state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert {\psi }_{0}\right\rangle $$\end{document}ψ0, the optimization problem 23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\arg {\max}_{\vartheta }{{{\mathcal{F}}}} \left({V}_{\vartheta }\left\vert {\psi }_{0}\right\rangle,V\left\vert {\psi }_{0}\right\rangle \right)=\arg {\max} _{\vartheta }{\left| \left\langle {\psi }_{0}\right\vert {V}_{\vartheta }^{{{\dagger}} }V{\psi }_{0}\right|}^{2},$$\end{document}argmaxϑFVϑψ0,Vψ0=argmaxϑψ0Vϑ†Vψ02, is numerically treated, where Vϑ is the circuit ansatz unitary with parameters ϑ. The recompiled circuit component is then the ansatz with optimal parameters fixed. As \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is number-conserving, to enhance recompilation performance and the quality of recompiled circuits, we focus optimization on the Fock sectors relevant to the time-evolution simulation45. In our implementation, we estimate Vθ through auto-differentiable tensor network-based ansatz simulation122, and use L-BFGS-B with basin-hopping to perform the optimization123. Recompilation is set to never exceed a few minutes per circuit. Readout error mitigation The quantum devices used in our experiments present non-negligible probabilities (~1–3%) of reporting a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 1\right\rangle$$\end{document}1 outcome when the measured qubit is in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert 0\right\rangle$$\end{document}0 and vice versa, typical for current-era quantum hardware. This is referred to as readout error and is caused, for example, by ambiguity in microwave signal characteristics (i.e., IQ data) that inhibit classification of measurement outcomes. We mitigated the effect of these errors by characterizing the measurement bit-flip probabilities of the qubits used, then performing linear inversion to approximately recover the true measurement counts from observed measurement counts124–126. Unlike prior works45–47,124–126, our experiment circuits contain mid-circuit measurements—in particular a qubit can be measured multiple times throughout the circuit. We thus employed a modified scheme as described below. Consider an experiment circuit containing measurements on n qubits, labeled by [n], containing L layers of measurements. The set of qubits measured in layer l is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{I}}}}}_{l}\subseteq [n]$$\end{document}Il⊆[n], of size \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{l}=\left\vert {{{{\mathcal{I}}}}}_{l}\right\vert$$\end{document}pl=Il, and the total number of measurements over all layers is p ≥ n. The measurement outcomes of each shot of the circuit form a bitstring x ∈ {0, 1}p, where entry xj records the outcome of the j-th measurement in the circuit. Upon completion of a specified number of shots, a raw counts vector c is returned, where entry cx records the number of times the bitstring x was observed. We assume that there are no correlations in readout errors for measurements in different layers, as these measurements are separated in time, consistent with standard Markovian assumptions on noise channels127,128; but within the same layer we accommodate arbitrary correlations. Let M be the readout calibration matrix on the n qubits, where each entry Mst records the probability of obtaining outcome s ∈ {0, 1}n upon measurement of the qubits when the true outcome is t ∈ {0, 1}n. Then c and the ideal counts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{{{{\bf{c}}}}}$$\end{document}c~ that would hypothetically be observed without readout errors are related by 24\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bf{c}}}=\left({\otimes }_{l=1}^{L}M[{{{\mathcal{I}}}}_{l}]\right)\widetilde{{{\bf{c}}}}\iff \widetilde{{{\bf{c}}}}=\left({\otimes }_{l=1}^{L}M{[{{{\mathcal{I}}}}_{l}]}^{-1}\right){{\bf{c}}},$$\end{document}c=⊗l=1LM[Il]c~⇔c~=⊗l=1LM[Il]−1c, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M[{{{{\mathcal{I}}}}}_{l}]$$\end{document}M[Il] is M with support retained over the qubits \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{I}}}}}_{l}$$\end{document}Il of layer l but marginalized over the remaining qubits, that is, 25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M{[{{{{\mathcal{I}}}}}_{l}]}_{st}=\frac{1}{{2}^{n-{p}_{l}}} {\sum}_{\begin{array}{c}s' \in {\{0,1\}}^{n}\\ s' [{{{{\mathcal{I}}}}}_{l}]=s\end{array}}\,\, {\sum}_{\begin{array}{c}t' \in {\{0,1\}}^{n}\\ t' [{{{{\mathcal{I}}}}}_{l}]=t\end{array}}\,\,{M}_{s' t' },$$\end{document}M[Il]st=12n−pl∑s′∈{0,1}ns′[Il]=s∑t′∈{0,1}nt′[Il]=tMs′t′, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s[{{{{\mathcal{I}}}}}_{l}]$$\end{document}s[Il] denotes the substring formed by entries of a bitstring s at qubit indices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{I}}}}}_{l}$$\end{document}Il. Thus, Eq. (24) enables the recovery of ideal counts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{{{{\bf{c}}}}}$$\end{document}c~ from raw counts c. To ensure that the mitigated counts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{{{{\bf{c}}}}}$$\end{document}c~ are physical, we employed an additional step of projecting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{{{{\bf{c}}}}}$$\end{document}c~ onto the nearest non-negative probability distribution (in l2-norm)129. The matrix M can be characterized by running calibration circuits, which prepare the n qubits in state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\vert t\right\rangle$$\end{document}t for every t ∈ {0, 1}n and record probabilities of observing each outcome s ∈ {0, 1}n upon measurement. However, this direct method requires 2n calibration circuits and is prohibitively costly for large n. We instead employed a tensored approach45–47, which splits the n qubits into sub-registers containing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{n}_{g}\}}_{g}$$\end{document}{ng}g qubits and assumes that readout errors are uncorrelated between sub-registers. The calibration is performed for each sub-register, producing matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{M}^{(g)}\}}_{g}$$\end{document}{M(g)}g, and the overall \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M={\bigotimes}_{g}{M}^{(g)}\iff {M}^{-1}={\bigotimes}_{g}{{M}^{(g)}}^{-1}$$\end{document}M=⨂gM(g)⇔M−1=⨂gM(g)−1. Then, performing linear inversion as in Eq. (24), each \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{M}^{(g)}}^{-1}$$\end{document}M(g)−1 can act sector-wise on c, avoiding the cost of assembling the entire \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ M^{-1}$$\end{document}M1. Moreover, the calibration circuits for different sub-registers can be merged as they act on disjoint sets of qubits; thus this approach requires \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2}^{{{\max }_{g}}{n}_{g}}\ll {2}^{n}$$\end{document}2maxgng≪2n calibration circuits. We chose sub-registers such that ng ≤ 5 in our experiments. While the employed readout error mitigation addresses errors in the measurement outcome statistics, it does not address errors occurring in feedforward—for example, a faulty mid-circuit measurement outcome on an ancillary qubit in our circuits can cause an incorrect reset. A recent work by several of the authors presented a protocol that overcomes this limitation130 but some experiments in this study predate this development; we do not use this protocol here. Dynamical decoupling Idle periods on qubits are invariably present in our experiment circuits, as a qubit may need to wait for gates to complete on other qubits before being involved in 2-qubit operations with them. During idle periods, qubits are still subject to decoherence (e.g., dephasing noise and thermal relaxation). Dynamical decoupling131, which runs through alternating basis changes that multiply to the identity, can be used to suppress the effects of such decoherence. At a basic level, dynamical decoupling refocuses the dephasing experienced by the qubits such that the accumulated noise contributions cancel to a certain order. We employed the common XY4 pulse sequence131,132 for dynamical decoupling in our experiments. The sequence is inserted for idle periods not exceeding the equivalent of 2 CX gate durations (~ 800–1000 ns); for longer idle durations, repetitions of the sequence are inserted. While there exist more sophisticated dynamical decoupling sequences that cancel noise to higher orders133,134, empirical results135 suggest that simple sequences (e.g. XY4) can be more performant in practice, due to the presence of errors on the gates introduced by the sequence and differences between hardware noise backgrounds and theoretical assumptions. Pauli twirling We applied randomized twirling of CXs on each experiment circuit. Each CX gate in the circuit is replaced by P1(CX)P2, where (P1, P2) are randomly drawn Paulis such that the action of the twirled CX remains invariant. See Supplementary Table 6 for a complete list of (P1, P2) pairs used. In general, the twirling of gates in circuits, also referred to as randomized compiling in literature136,137, has the effect of converting coherent and non-Markovian error sources into stochastic errors which can be averaged over in an experiment. Pauli twirling, in particular, converts error processes into diagonal Pauli channels138. We used Pauli twirling in conjunction with zero-noise extrapolation. Zero-noise extrapolation A general method for mitigating inaccuracy of expectation values due to hardware noise is zero-noise extrapolation (ZNE), which functions by collecting additional data at amplified noise levels and then extrapolating into the zero-noise limit42,138–141. We employ a gate-level implementation of ZNE in our experiments; alternative pulse-level methods have been explored in the literature125,142. Our procedure was as follows. For an experiment measuring a set of observables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{O}_{k}\}}_{k}$$\end{document}{Ok}k, we perform repetitions at noise amplification factors 1 = λ1 < λ2 < … < λr, where λ = 1 corresponds to the unmodified experiment. Noise in an experiment is amplified by randomized local gate folding140, that is, each gate G in each circuit is probabilistically replaced by GG†G, such that the total number of 1- and 2-qubit gates in the circuit is each increased by factor λ > 1. Note that G† is known for all basis gates (e.g., the Paulis and CX are involutory), so compilation of the folded circuits is straightforward. To illustrate, λ = 3 folds every gate in each circuit, 1 < λ < 3 contains a mixture of folded and unfolded gates, and λ > 3 can be reached by repeated folding. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {O}_{k}^{(\lambda )}\rangle$$\end{document}⟨Ok(λ)⟩ denote \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {O}_{k}\rangle$$\end{document}⟨Ok⟩ measured on hardware at noise amplification factor λ. We assume an affine relation between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {O}_{k}^{(\lambda )}\rangle$$\end{document}⟨Ok(λ)⟩ and λ, that is, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {O}_{k}^{(\lambda )}\rangle \approx {m}_{k}\lambda+{c}_{k}$$\end{document}⟨Ok(λ)⟩≈mkλ+ck for gradient mk and zero-noise intercept ck, which enables a simple formulation of the extrapolation problem. This linearization is valid for λ not too large, such that hardware noise does not saturate measured observables. We treat the least-squares regression 26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({{{{\bf{m}}}}}^{*},{{{{\bf{c}}}}}^{*})={{{\rm{argmin}}}}_{{{{\bf{m}}}},{{{\bf{c}}}}} {\sum}_{k} {\sum }_{\gamma=1}^{r}{\left({m}_{k}{\lambda }_{\gamma }+{c}_{k}-\left\langle {O}_{k}^{({\lambda }_{\gamma })}\right\rangle \right)}^{2},$$\end{document}(m*,c*)=argminm,c∑k∑γ=1rmkλγ+ck−Ok(λγ)2, and the zero-noise estimates \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{\langle {O}_{k}^{(0)}\rangle \}}_{k}$$\end{document}{⟨Ok(0)⟩}k are then given by the intercepts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{c}_{k}^{*}\}}_{k}$$\end{document}{ck*}k. We additionally impose constraints on (m, c) based on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{O}_{k}\}}_{k}$$\end{document}{Ok}k to ensure physicality of regression results. In particular, for experiments in the p-particle Fock sector measuring site-resolved occupancies, that is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{O}_{k}\}}_{k}={\{{n}_{x\ell }\}}_{x\ell }$$\end{document}{Ok}k={nxℓ}xℓ, we imposed number conservation and hardcore bosonic statistics, such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0\le \left\langle {n}_{x\ell }\right\rangle \le 1$$\end{document}0≤nxℓ≤1 for all sites and the sum of occupancies equaled p, at all noise levels on the regression lines. For experiments measuring the imaginary energy in the p-particle sector, that is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\{{O}_{k}\}}_{k}=\{{{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\}$$\end{document}{Ok}k={HAeb}, we imposed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\rangle \le 0$$\end{document}⟨HAeb⟩≤0 as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{H}}}}}^{{{{\rm{eb}}}}}$$\end{document}Heb is lossy and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$-p\gamma \le \langle {{{{\mathcal{H}}}}}_{{{{\rm{A}}}}}^{{{{\rm{eb}}}}}\rangle$$\end{document}−pγ≤⟨HAeb⟩ by the variational principle. These constraints translate into conditions on (m, c)—see Supplementary Note 3. The constrained regression is convex and can be solved using standard techniques with guaranteed optimal solutions143. In our experiments, we chose (λ1,…, λr) = (1, 1.25, 1.5, 1.75, 2) and used the general-purpose operator-splitting quadratic program (OSQP) solver144 to perform the constrained regressions. For each experiment circuit at each λ, we produced 16 gate-folded circuits, each further compiled with an independent instance of randomized Pauli twirling. Each set of 16 circuits were executed on hardware, and their results averaged to produce the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle {O}_{k}^{(\lambda )}\rangle$$\end{document}⟨Ok(λ)⟩ expectation value for regression. See Supplementary Note 3 for further technical details and discussion. Qubit selection The quantum devices we utilized have more qubits than needed for our experiments, and there is variation in qubit error rates on each device (see Supplementary Table 4 for device performance characteristics). We hence used a search procedure to select qubits of the lowest estimated error to use in our simulations, following the demonstrated methods in Refs. 33,45–47,145,146. Given an experiment circuit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{C}}}}$$\end{document}C requiring n qubits and the connectivity graph of 2-qubit couplings (CX, ECR, or CZ) on the device, we identified all distinct qubit chains \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{X}}}}$$\end{document}X of length n on the device, and for each chain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x\in {{{\mathcal{X}}}}$$\end{document}x∈X computed an estimated circuit error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{E}}}}}_{x}[{{{\mathcal{C}}}}]$$\end{document}Ex[C] by summing the calibrated gate errors of each gate in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\mathcal{C}}}}$$\end{document}C. We selected the qubit chain x that minimized \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{E}}}}}_{x}[{{{\mathcal{C}}}}]$$\end{document}Ex[C]. This selection procedure is approximate, as the gate error rates are obtained from routine (~daily) calibration of the quantum devices and are subject to drift over time, and the circuit error estimation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\mathcal{E}}}}}_{x}[{{{\mathcal{C}}}}]$$\end{document}Ex[C] is not exact. Supplementary information Supplementary Information Transparent Peer Review file Supplementary Information Transparent Peer Review file