Lee-Yang Zeros of a Bosonic system associated with a single trapped ion

The Lee-Yang zeros (LYZ’s) are points on the complex plane where the partition function vanishes[1]. Considering an Ising model under a complex magnetic field, Lee and Yang showed that the partition function is analytic and all physical properties are determined by its zeros. If the LYZ’s approach the real axis, a phase transition is expected with diverging free energy, since the zeros of the partition function are precisely the singularities of the free energy.

The LYZ’s has been extended to many interesting spin models[2, 3] to study phase transition physics, such as dynamical phase transition[4] and critical exponents[5]. Despite the intractability for complex physical parameters, Wei and Liu[6] discovered that the coherence of a probe spin coupled to a many-body system can be mapped directly to the partition function. They proposed a method to observe zeros by measuring the time when such coherence vanishes. Subsequent experiments used liquid-state trimethylphosphite molecules[7] to first observe the LYZ’s. A recent experiment[8] measured LYZ’s with trapped-ion quantum circuits, providing a scalable method for studying partition functions with near-term devices. Although LYZ’s of spin systems are widely studied, they remain elusive in Boson systems[9]. Gnatenko et al[10] demonstrated the relationship between the two-time correlation function and LYZ’s. However, measuring such correlation is not straightforward in real experiments.

Phonon modes in trapped-ions are suitable tools to study quantum simulations for Boson systems[11, 12]. Moreover, recent experiments[13, 14] that harness trapped-ion motion beyond the Lamb-Dicke(LD) regime—where the spatial wavefunction is much smaller than coupling electromagnetic wavelength—provide new possibilities for quantum simulations, in which case the higher-order sidebands cannot be neglected. Moreover, the carrier coupling strength has a non-linear dependence on phonon number $n$, which imprints an $n$-dependent phase factor on spin evolution. As for experimental progress, laser cooling with an LD parameter up to $\eta=1.3$ is demonstrated[15], allowing for quantum simulations with deeper coupling between spin and phonon modes.

In this work, we propose an experiment scheme to observe LYZ’s of a Boson system using a single trapped ion. We use one phonon mode to construct the Boson system of interest and a pair of Zeeman states as the probe spin. Beyond the LD regime, the coupling strength varies non-linearly with respect to phonon number $n$. We first prepare thermal states with such dependence, and apply a carrier transition outside the LD regime. To generate the target thermal state pertaining to the strong coupling interaction, we extend the averaging method[16] to the Boson case and simulate such states with an ensemble of coherent states. This method shows a low infidelity that does not interfere with the pattern of zeros. Finally, we demonstrate that the scheme is robust against experimental noise such as heating and possible spin decoherence.

The work is organized as follows. In Sec. II we present our model and implementation of the thermal states. In Sec. III, we describe the spin coherence and partition function measurements with a practical scenario. In Sec. IV, we demonstrate the robustness of zeros with known experimental imperfections. The scheme is summarized and discussed in Sec. V.

Here we consider a Bosonic Hamiltonian $H_{S}$ in the presence of an external field given by $H_{B}=hH_{I}$, where $h=h_{R}+ih_{I}$ is complex. The partition function can be written as

where $\beta=1/kT$ is the inverse temperature. The exponential term resembles a thermal state of $H_{S}+h_{R}H_{I}$ under evolution $H_{I}$ for a duration of $\beta h_{I}$ when $H_{S}$ and $H_{I}$ commute. We could further extract the partition function by coupling the system with a probe spin. We consider a diagonal system Hamiltonian, e.g. $H_{S}=\omega_{m}a^{\dagger}a$, where $a(a^{\dagger})$ are annihilation(creation) operators.The commuting criteria suggests a diagonal form $H_{I}=\sum\xi_{n}\ket{n}\bra{n}$, which gives the necessary coupling Hamiltonian

We use the internal state of a single trapped ion as the probe and a phonon mode with frequency $\omega_{m}$ as the Bosonic system. To achieve the coupling Hamiltonian, we apply optical transitions on resonance with the spin. The interaction Hamiltonian can be reduced to the following form[17] with a proper rotating frame,

where $\eta=kz_{0}$ is the Lamb-Dicke parameter, i.e. the ratio between wave function and coupling wavelength. In the sideband-resolved limit where $\omega_{m}$ is much larger than the Rabi frequency $\Omega$, we can further neglect the oscillating terms and get a Hamiltonian diagonal in the motion subspace,

where $\Omega_{n}=\Omega e^{-\eta^{2}/2}L_{n}(\eta^{2})$ is the Rabi frequency between $|\downarrow,n\rangle\leftrightarrow|\uparrow,n\rangle$ and $L_{n}$ is the Laguerre polynomial. This yields the coupling interaction 2 with $\xi_{n}=\Omega_{n}/\Omega$. We initialize our system to the product of a superposition state of $|+\rangle$ and $|-\rangle$, i.e. $|\uparrow\rangle$ and a thermal state,

where $Z_{0}=\text{Tr}\exp(-\beta(H_{s}+h_{R}H_{I}))$ is the partition function at zero imaginary field. After the evolution stated above, the normalized partition function $Z/Z_{0}$ is the off-diagonal element of the probe spin density matrix under $|\pm\rangle$ basis.

The probability distribution of our thermal state deviates from any exponential distribution, nor can it be created by any simple operation, such as displacement and squeezing. To generate this density matrix, we prepare a weighted mixture of $N$ coherent states,

where $\alpha_{i}$ are displacement parameters and their probablities $p_{i}$ satisfy $\sum p_{i}=1$. Since the motion subspace is traced out as we measure the spin components, only the diagonal elements of $\rho_{N}$ will contribute to the spin measurements. Off-diagonal elements of coherent states do not contribute to any observable effect for the probe spin. Therefore we only consider diagonal terms when analyzing the fidelity. Having $N$ fixed, we optimize $p_{i}$ and $\alpha_{i}$ to maximize the fidelity $f$. Figure 1 shows the infidelity after optimization. By using 3 coherent states the fidelity should reach $99.9\%$, and the maximum displacement in these ensembles is about 3.3, within the range of typical experiments[18]. As one increases $h_{R}/\omega_{m}$, $H_{I}$, whose ground-state is a Fock state $\ket{n}$, gradually dominates the thermal distribution. Therefore, our method fails at very large $h_{R}$ when the target distribution becomes sub-Poissonian. An example of this scenario is shown in the inset of figure 1(a).
To examine the robustness of this method against experiment noise, we calculate the fidelity by fixing $\alpha_{1}$ and vary $\alpha_{2}$ and $\alpha_{3}$ for the ensemble shown in figure 1(b). We take a typical parameter $h_{R}/\omega_{m}=7$ and $\beta\omega_{m}=0.5$ that exhibit multiple zeros. As shown by figure 1(c), a fidelity threshold of $99\%$ tolerates fluctuation of $15\%$ on the displacement.

Diagonal elements of $H_{I}$, which depend on $\eta$, will be nearly uniform for $n\leq 20$ if the Lamb-Dicke criterion, i.e. $\eta\ll 1$, is satisfied. This will lead to uniform Rabi oscillation and therefore no spin decoherence. In order to observe Lee-Yang zeros, the initial Gibbs state needs to cover a larger range of Fock states that exhibit variation on Rabi frequencies $\Omega_{n}$, which is impossible for a precise preparation. Therefore, our experiment scheme favors a larger $\eta$. Figure 2 compares Lee-Yang zeros of different $\eta$ at the same temperature $\beta\omega_{m}=0.5$ and typical initial state probability distributions. When the LD criterion $\eta\ll 1$ is satisfied, one needs to control distributions at $n\simeq 100$ to observe zeros, which is experimentally intractable. Beyond the LD regime, the initial distribution is restricted to $n\leq 20$. This is easier to control for trapped-ion experiments [19].

We consider a practical scenario with a single ${}^{40}\text{Ca}^{+}$ ion trapped in the harmonic potential similar to the setup in[20]. Figure 3 shows the levels engaged in this scheme. The ground states $|\uparrow\equiv m_{s}=+1/2\rangle$ and $|\downarrow\equiv m_{s}=-1/2\rangle$ of ${}^{2}S_{1/2}$ manifold form the probe spin, and the Boson system we consider is encoded into the phonon mode of the ion. The Zeeman splitting between two states is $2\pi\times 20~{}\text{MHz}$ at a magnetic field of $0.68~{}\text{mT}$, and the trapping frequency of axial mode is $2\pi\times 600~{}\text{kHz}$. By using a pair of counter-propagating laser beams at $397~{}\text{nm}$ to drive Raman transitions, we can reach a Lamb-Dicke parameter up to $\eta=0.47$. Similar configurations with large $\eta$ are available with other types of ions such as ${}^{9}\text{Be}^{+}$[19, 15]. A large $\eta$ restricts the motion states of interest to $n\leq 20$. We apply laser pulses at $729~{}\text{nm}$ to couple the ${}^{2}S_{1/2}$ and ${}^{2}D_{5/2}$ manifolds. By applying a carrier and a red-sideband(RSB) transition between $|\uparrow\rangle$ and $|^{2}D_{5/2},m_{J}=1/2\rangle$, the eigenstates are engineered to displaced Fock states[18]. In this case, optical pumping from $|^{2}D_{5/2},m_{J}=1/2\rangle$ to $|\uparrow\rangle$ extracts entropy from the system and eventually cool the ion motion to a displaced ground-state $|\alpha\rangle=D(\alpha)|0\rangle$. By tuning the Rabi frequency ratio between carrier and RSB transition, we can prepare different coherent states. Subsequent summation of the results gives an effective mixed state from different $\alpha$’s

In our scheme, the imaginary axis for LYZ’s is mapped to the time of unitary evolution under the interacting Hamiltonian 4. After applying this carrier transition for time $t$, we analyze the spin components perpendicular to $\sigma_{x}$, i.e. $\sigma_{z}$ and $\sigma_{y}$, the latter requires a pair of co-propagating Raman beams to achieve motion-independent rotation $R^{x}_{\pi/2}$

We plot in figure 4 the partition function and typical spin coherence data that exhibit LYZ’s at certain evolution times. By assuming a Rabi frequency of $2\pi\times 50~{}\text{kHz}$, we found that the spacing between the zeros is above $10~{}\mu s$, which can be readily resolved in experiments.

We also study the partition function at various temperatures to test the generality of our methods. The partition function and fidelity of motional states are plotted in figure 5. For lower temperatures(high $\beta$), the sub-Poissonian low entropy thermal states are more difficult to simulate with coherent states, as shown by the decaying fidelity in figure 5(a) and (b). For higher temperatures(low $\beta$), the fidelity remains high for the states we consider. However, spin measurement contrasts near zeros are lower as shown in figure 5(c). Such a scenario requires more experiment runs for higher resolution.

In this section, we discuss the effect of experimental imperfections on our scheme, and show that our proposal is robust against common imperfections in trapped ion experiments. We first consider the anomalous heating as a noise in the motional subspace. this could be modeled by the Lindblad form master equation, with jump operators $\sqrt{\gamma}a$ and $\sqrt{\gamma}a^{\dagger}$[21],

where $\gamma$ is the heating rate. Since we always start with a product state, the effect of heating is restricted to the motional subspace. In other words, the initial distribution $\rho_{N}$ we prepare becomes $\rho_{N}+\Delta\rho$ after heating. We calculate $\Delta\rho$ with the master equation above, and modify the partition function according to equation 7

We also discuss the influence of spin decoherence. Coherence time of similar systems reach hundreds of microseconds[22], about 3 orders of magnitude higer than the time we consider. Besides, noise models that can be described by quantum channels, such as spin dephasing, decay the density matrix exponentially; this should not affect the position of zeros, where the coherence completely vanishes. Therefore, we focus on shot-to-shot frequency fluctuation $V=\frac{1}{2}\Delta\sigma_{z}$, and calculate the average of experimental runs. A static detuning mixes spin components at $x$ and $z$ direction, and modifies the Rabi frequency by $X_{n}=\sqrt{(\Delta^{2}+\Omega_{n}^{2})}$, and the spin measurement outcomes are varied accordingly.

By tuning $\eta$, we can maximize the smallest frequency $\Omega_{n}$ within the range we consider, thus minimizing the term $\Delta^{2}/\Omega_{n}^{2}$. This yields $\eta=0.455$.

We calculate the deviance caused by heating and frequency detuning individually according to equation 9 and 10 in figure 6(a) and (b) by assuming a heating rate of $300~{}\text{Quanta/s}$ and frequency detuning of $1~{}\text{kHz}$. Deviations do not exceed $0.02$ at the region of zeros. In figure 6(c) we calculate the partition function with non-zero heating rate and shot-to-shot frequency fluctuations among 100 experiment runs. Compared with figure 4(a), the patterns of zeros are not disturbed.

To summarize, we propose a scheme to simulate LYZ’s of a Boson system using a single trapped ion. By applying a carrier interaction beyond the LD regime, we modeled the interaction Hamiltonian that lead to partition function zeros. We showed that the thermal state of a non-linear Hamiltonian can be simulated using an ensemble of coherent states. This method could also be extended to other types of thermal states that cannot be generated with single operations. Furthermore, we proved that our scheme is robust against experimental imperfections.

For future studies, it is possible to exploit other properties, such as quasi-probability distributions, to explain the pattern of the zeros. Moreover, our method can be scaled up to multiple modes with phase transition, such as the Rabi-Hubbard model.