betaFeedback

WorksPhysics and Astronomy

Finite-time spectral enhancement of transient quantum thermometry

Abstract

Transient quantum thermometry extracts temperature information before a probe reaches thermal equilibrium. We study how finite interaction times affect thermometric precision for a harmonic-oscillator probe prepared in a number state and weakly coupled to a thermal bosonic reservoir. A second-order time-convolutionless master equation retains the finite-time reservoir response, allowing modes away from the probe frequency to shape the populations and their temperature dependence. For the baseline parameters, this response raises the quantum Fisher information about fivefold above the Markovian prediction, in agreement with an exact solution of the model. When the probability of leaving the initial state is small, we derive an analytical condition for this enhancement. To assess its operational significance, we examine the temperature information acquired per unit time. Under the exact dynamics, this rate is largest within the reservoir memory time. With negligible preparation and readout overhead, the optimum is about five times the best Markovian rate and 30 times an upper bound on the equilibrium rate. The advantage survives smooth switching, although overhead and longer coupling ramps reduce it.

Introduction

Quantum thermometry infers temperature from measurable changes in quantum probes, offering access to microscopic systems where conventional thermometers are difficult to implement [1, 2, 3, 4]. Experiments with diamond spins and optomechanical systems demonstrate this approach [5, 6, 7, 8]. The quantum Fisher information bounds local estimation precision by quantifying the distinguishability of probe states at nearby temperatures [9, 10]. At equilibrium, sensitivity depends on energy fluctuations, spectral gaps, correlations, and the accessible part of the system [11, 12, 13, 14, 15, 16]. Measurement restrictions [17, 18, 19, 20] and backaction [21] further constrain the information extracted. Global and Bayesian approaches address estimation over broader temperature ranges [22, 23], including adaptive strategies that use prior information [24, 25]. For an interacting probe, a further practical question is how much temperature information can be acquired before equilibrium is reached.

In this transient regime, thermometric performance depends on the probe preparation and interrogation time, as explored in qubit models and photonic simulations [26, 27, 28, 29]. Ancilla-assisted schemes can encode temperature in coherences [30, 31, 32], while initial and bath-induced correlations can improve sensitivity [33, 34].

Collisional protocols [35, 36] and sequential measurements [37] also exploit information acquired through successive interactions or observations. The time required to acquire information is addressed by finite-time precision bounds [38], prethermal probes [39], and nonequilibrium single-atom sensing [40]. For oscillator probes, nonequilibrium operations do not necessarily improve sensitivity [41], although suitable nonclassical Gaussian preparations offer advantages [42]. Fock states and other non-Gaussian preparations can further improve transient performance under dissipative dynamics [43, 44]. Exact treatments of Gaussian oscillator probes have investigated strong-coupling finite-time thermometry [45]. Alongside these preparation-dependent advantages, the reservoir response determines how temperature information is encoded during the interaction.

The reservoir influences thermometric performance through the frequencies and coupling strengths of its modes. Impurity probes access temperature information in Bose and Fermi gases [46, 47, 48, 49, 50], while dephasing schemes connect sensitivity to the reservoir spectrum [51, 52]. Strong coupling can improve thermometry, but its benefits depend on the operating regime [53, 54, 55]. Reservoir effects have also been investigated in non-Markovian models [56, 57, 58, 59], and periodic driving can control the frequencies involved in sensing [60, 61]. These studies motivate examining how reservoir modes contribute to temperature sensing during a finite probe–reservoir interaction. For the oscillator model considered here, the Markovian rates sample the reservoir only at the probe frequency. During a finite interaction, however, modes away from resonance can also influence the temperature dependence of the probe populations. The non-Markovian studies cited above show that reservoir memory can change thermometric precision. Our aim is different. We identify which reservoir frequencies carry the extra information, give a closed condition for when it raises the quantum Fisher information (QFI), and test whether the gain survives as an operational advantage.

We address these questions with a harmonic-oscillator probe prepared in a number (Fock) state and weakly coupled to a bosonic reservoir. We describe its dynamics with a second-order time-local master equation that retains the finite interaction interval in the reservoir response. We also solve the model exactly and use the exact solution as a benchmark. Modes away from the probe frequency affect both the excitation probabilities and how strongly they change with temperature. These modes can raise the QFI above the Markovian prediction at the same interrogation time. When the probability of leaving the initial state is small, an analytical QFI ratio identifies when this happens. Because the finite-time and Markovian QFIs describe the same experiment, their ratio measures what the Markovian description misses. We therefore ask a second, operational question. We compare the information acquired per unit interrogation time under the exact dynamics with the best rate available after the reservoir memory has decayed. We find that the optimal interrogation lies inside the memory time. We then show how preparation overhead and a finite switching time limit this advantage, and we map its dependence on temperature, initial number state, and reservoir spectrum.

The remainder of this work is organized as follows. Section II introduces the model, its finite-time and Markovian dynamics, and its exact solution. Section III defines the QFI comparison, develops the enhancement mechanism, and tests it numerically. It then examines the information rate, finite switching times, and the dependence on temperature and initial state. Section IV concludes the work. Appendices A–E give the master-equation derivation, the analytical population and QFI calculations, the numerical validation and spectral scans, the exact solution, and the switching filte

Model and finite-time reservoir response

We consider a harmonic-oscillator probe coupled to a thermal bosonic reservoir for a controlled interaction time. The reservoir temperature is inferred from the probe’s number-state populations. We introduce the physical model and its finite-time dynamics, together with the Markovian benchmark used for comparison.

Physical model

The probe, reservoir, and their interaction are described by

H=HS+HB+HI,(1)H = H_{S} + H_{B} + H_{I}, \tag*{(1)}
HS=ℏω0a†a,(2)H_{S} = \hbar\omega_{0}a^{\dagger}a, \tag*{(2)}
HB=∑kℏωkbk†bk,(3)H_{B} = \sum_{k}\hbar\omega_{k}b_{k}^{\dagger}b_{k}, \tag*{(3)}
HI=ℏλ∑k(gka†bk+gk∗abk†).(4)H_{I} = \hbar\lambda\sum_{k}\left(g_{k}a^{\dagger}b_{k} + g_{k}^{*}ab_{k}^{\dagger}\right). \tag*{(4)}

Here, HH is the total Hamiltonian, while HSH_{S}, HBH_{B}, and HIH_{I} describe the probe, reservoir, and interaction, respectively. The operators aa and bkb_{k} annihilate excitations of the probe and reservoir mode kk, respectively, and satisfy [a,a†]=1[a,a^{\dagger}]=1 and [bk,bℓ†]=δkℓ[b_{k},b_{\ell}^{\dagger}]=\delta_{k\ell}, where δkℓ\delta_{k\ell} is the Kronecker delta. The probe and reservoir modes have positive angular frequencies ω0\omega_{0} and ωk\omega_{k}, respectively. The coupling amplitude gkg_{k} has units of angular frequency, and λ≪1\lambda\ll1 is a dimensionless coupling parameter. Each interrogation begins from the factorized state

ρtot(0)=ρ(0)⊗ρB(T),(5)\rho_{\mathrm{tot}}(0)=\rho(0)\otimes\rho_{B}(T), \tag*{(5)}
ρ(0)=∣n0⟩⟨n0∣,ρB(T)=e−HB/(kBT)ZB.(6)\rho(0)=|n_{0}\rangle\langle n_{0}|,\qquad\rho_{B}(T)=\frac{e^{-H_{B}/(k_{B}T)}}{Z_{B}}. \tag*{(6)}

Here, ρtot\rho_{\mathrm{tot}} is the joint state, ρ\rho is the reduced probe state, and ρB(T)\rho_{B}(T) is the thermal reservoir state at temperature TT. The Fock state ∣n0⟩|n_{0}\rangle has a prescribed nonnegative integer occupation n0n_{0}. The constant kBk_{B} is Boltzmann’s constant, and ZB=Tr⁡B[e−HB/(kBT)]Z_{B}=\operatorname{Tr}_{B}[e^{-H_{B}/(k_{B}T)}] is the reservoir partition function. Equation (5) assumes that the probe and reservoir are initially uncorrelated.

After preparation, the probe interacts with the reservoir for a time tt, and its excitation number is measured. Until Section III we take the coupling to switch on at time zero and off at time tt instantaneously. Section III replaces this step with a smoothly switching profile. Independent repetitions under the same initial conditions sample the number-state populations used for temperature inference. The finite-time and Markovian evolutions are compared using the same initial preparation, interrogation time, and physical parameters. Throughout the analysis, temperature derivatives are taken at fixed interrogation time, initial occupation, probe frequency, and reservoir spectral and coupling parameters.

Dynamics and Markovian benchmark

For the Hamiltonian in Eq. (1) and the initial state in Eq. (5), we obtain the reduced dynamics from a second-order time-convolutionless (TCL2) expansion that retains the reservoir response over the finite interaction interval [62]. The derivation in Appendix A gives the Schrödinger-picture master equation

dρ(T,t)dt=−iℏ[HS+HLS(T,t),ρ(T,t)]+γ−,FT(T,t)D[a]ρ(T,t)+γ+,FT(T,t)D[a†]ρ(T,t).(7)\frac{d\rho(T,t)}{dt}=-\frac{i}{\hbar}[H_{\mathrm{S}}+H_{\mathrm{LS}}(T,t),\rho(T,t)]+\gamma_{-,\mathrm{FT}}(T,t)\mathcal{D}[a]\rho(T,t)+\gamma_{+, \mathrm{FT}}(T,t)\mathcal{D}[a^\dagger]\rho(T,t). \tag*{(7)}

Here, the Lindblad dissipator is

D[L]ρ=LρL†−12{L†L,ρ}.(8)\mathcal{D}[L]\rho=L\rho L^\dagger-\frac{1}{2}\{L^\dagger L,\rho\}. \tag*{(8)}

where LL is the jump operator and braces denote an anticommutator. The reservoir-induced Lamb-shift Hamiltonian HLS(T,t)H_{\mathrm{LS}}(T,t) is given in Eq. (52). The time-local coefficients γ−,FT\gamma_{-,\mathrm{FT}} and γ+,FT\gamma_{+,\mathrm{FT}} describe downward and upward transitions, respectively, with FT denoting the finite-time response. Their expressions in Eq. (54) can be written in the equivalent sinc form

γ−,FT(T,t)=2λ2t∫0∞dω J(ω)[nT(ω)+1]×sinc⁡[(ω0−ω)t],(9)\gamma_{-,\mathrm{FT}}(T,t)=2\lambda^2t\int_{0}^{\infty}d\omega\,J(\omega)[n_T(\omega)+1]\times\operatorname{sinc}[(\omega_0-\omega)t], \tag*{(9)}
γ+,FT(T,t)=2λ2t∫0∞dω J(ω)nT(ω)×sinc⁡[(ω0−ω)t].(10)\gamma_{+,\mathrm{FT}}(T,t)=2\lambda^2t\int_{0}^{\infty}d\omega\,J(\omega)n_T(\omega)\times\operatorname{sinc}[(\omega_0-\omega)t]. \tag*{(10)}

Here, J(ω)J(\omega) is the reservoir spectral density, nT(ω)n_T(\omega) is the thermal occupation at angular frequency ω\omega, and sinc⁡x=sin⁡x/x\operatorname{sinc}x=\sin x/x, with sinc⁡0=1\operatorname{sinc}0=1. The spectral density and thermal occupation are

J(ω)=∑k∣gk∣2δ(ω−ωk),(11)J(\omega)=\sum_k|g_k|^2\delta(\omega-\omega_k), \tag*{(11)}
nT(ω)=1exp⁡[ℏω/(kBT)]−1,(12)n_T(\omega)=\frac{1}{\exp[\hbar\omega/(k_{\mathrm{B}}T)]-1}, \tag*{(12)}

where δ\delta is the Dirac delta distribution. At finite tt, the sinc kernels in Eqs. (9) and (10) allow off-resonant reservoir modes to contribute to the transition coefficients. To specify the reservoir coupling spectrum associated with the interaction Hamiltonian in Eq. (4), we choose the continuum form [63]

J(ω)=κΓΩ2ω(ω2−Ω2)2+(Γω)2.(13)J(\omega)=\kappa\frac{\Gamma\Omega^2\omega}{(\omega^2-\Omega^2)^2+(\Gamma\omega)^2}. \tag*{(13)}

Here, Ω\Omega sets the characteristic reservoir resonance frequency, Γ\Gamma controls its spectral width, and κ\kappa scales the spectral coupling weight. This form is motivated by a harmonic resonance broadened through coupling to an Ohmic background [63], where Γ\Gamma describes the damping of the collective resonance. Through Eq. (11), it specifies how the squared microscopic couplings are distributed over the reservoir normal-mode frequencies. The spectral maximum depends on both Ω\Omega and Γ\Gamma, and need not coincide with Ω\Omega. This choice allows the spectral position, width, and coupling scale to be varied relative to the fixed probe frequency.

The reservoir correlations decay on the memory time τB≃2/Γ\tau_{\mathrm{B}}\simeq2/\Gamma, which equals about 17/ω017/\omega_0 for the baseline width Γ/ω0=0.12\Gamma/\omega_0=0.12.

The Markovian benchmark follows from the long-time limit of the kernel tsinc⁡[(ω0−ω)t]t\operatorname{sinc}[(\omega_0-\omega)t] appearing in Eqs. (9) and (10). For fixed ω0\omega_0,

sin⁡[(ω0−ω)t]ω0−ω→t→∞Dπδ(ω−ω0),(14)\frac{\sin[(\omega_0-\omega)t]}{\omega_0-\omega}\xrightarrow[t\to\infty]{\mathcal{D}}\pi\delta(\omega-\omega_0), \tag*{(14)}

where the arrow denotes distributional convergence under integration against suitable smooth frequency weights. Applying this limit to Eqs. (9) and (10) gives

γ−,M(T)=γ0[nˉ(T)+1],γ+,M(T)=γ0nˉ(T),(15)\gamma_{-,\mathrm{M}}(T)=\gamma_0[\bar{n}(T)+1],\qquad\gamma_{+,\mathrm{M}}(T)=\gamma_0\bar{n}(T), \tag*{(15)}

where M denotes the Markovian benchmark. The damping scale and resonant thermal occupation are

γ0=2πλ2J(ω0),nˉ(T)=nT(ω0).(16)\gamma_0=2\pi\lambda^2J(\omega_0),\qquad\bar{n}(T)=n_T(\omega_0). \tag*{(16)}

Here, γ0\gamma_0 is temperature independent, and nˉ(T)\bar{n}(T) is the reservoir occupation at the probe frequency. The Markovian rates sample the reservoir spectrum at ω0\omega_0, whereas the finite-time coefficients γ±,FT\gamma_{\pm,\mathrm{FT}} include off-resonant contributions.

Exact solution

The model in Eqs. (1)–(4) conserves the total excitation number and is quadratic. Its Heisenberg equations are therefore linear and can be solved exactly. Appendix D shows that the probe operator evolves as

a(t)=G(t)a+∑kfk(t)bk,(17)a(t)=G(t)a+\sum_k f_k(t)b_k, \tag*{(17)}

where G(t)G(t) is the probe propagator and fk(t)f_k(t) are reservoir amplitudes. Two functions then fix the reduced dynamics. The first is the transmissivity η(t)=∣G(t)∣2\eta(t)=|G(t)|^2. The second is the thermal excitation transferred into the probe, N(T,t)=∑k∣fk(t)∣2nT(ωk)N(T,t)=\sum_k|f_k(t)|^2n_T(\omega_k). Only NN depends on temperature. The exact populations obey Eq. (55) with the exact time-local rates

γ+,ex=N˙+ΓexN,γ−,ex=Γex+γ+,ex,(18)\gamma_{+,\mathrm{ex}}=\dot{N}+\Gamma_{\mathrm{ex}}N,\qquad\gamma_{-,\mathrm{ex}}=\Gamma_{\mathrm{ex}}+\gamma_{+,\mathrm{ex}}, \tag*{(18)}

where Γex(t)=−η˙/η\Gamma_{\mathrm{ex}}(t)=-\dot{\eta}/\eta and dots denote time derivatives. We use this solution as a benchmark for the TCL2 results in Sec. III.

Finite-time thermometric enhancement

We now examine how these off-resonant contributions affect the temperature information carried by the probe.

We first introduce the QFI and show that excitation-number measurements attain it. We then relate the QFI to the reservoir-induced population changes and their temperature dependence. Finally, we compare the analytical prediction of QFI enhancement over the Markovian benchmark with numerical results.

Thermometric figure of merit

We quantify thermometric performance through the precision attainable from repeated measurements of the probe. For mm independent, identically prepared interrogations of duration tt, the quantum Cramér–Rao bound gives [9, 64]

Δ2T≥1mFX(T,t),(19)\Delta^2 T \ge\frac{1}{m\mathcal{F}_X(T,t)}, \tag*{(19)}

where Δ2T\Delta^2 T is the variance of a locally unbiased temperature estimator and FX(T,t)\mathcal{F}_X(T,t) is the QFI per interrogation for temperature estimation. The index X∈{FT,M}X \in\{\mathrm{FT},\mathrm{M}\} distinguishes the finite-time and Markovian evolutions. At fixed mm, a larger QFI lowers the corresponding variance bound. To evaluate the QFI from the probe state, we introduce the symmetric logarithmic derivative,

∂TρX(T,t)=12[ρX(T,t)LT,X+LT,XρX(T,t)],\partial_T \rho_X(T,t) = \frac{1}{2}\left[\rho_X(T,t)L_{T,X} + L_{T,X}\rho_X(T,t)\right],
FX(T,t)=Tr⁡[ρX(T,t)LT,X2],(20)\mathcal{F}_X(T,t) = \operatorname{Tr}\left[\rho_X(T,t)L_{T,X}^2\right], \tag*{(20)}

where ρX(T,t)\rho_X(T,t) is the reduced probe state under evolution XX, and LT,XL_{T,X} is the Hermitian symmetric logarithmic derivative, with its dependence on TT and tt left implicit. Temperature derivatives follow the fixed-parameter convention specified in Sec. II A.

The Hamiltonian terms in Eq. (7) are diagonal in the number basis, and the dissipators in Eq. (7) preserve diagonal states. The initial Fock state therefore evolves into

ρX(T,t)=∑n=0∞pnX(T,t)∣n⟩⟨n∣,(21)\rho_X(T,t) = \sum_{n=0}^{\infty}p_n^X(T,t)|n\rangle\langle n|, \tag*{(21)}

where pnX(T,t)p_n^X(T,t) is the probability of observing nn probe excitations. Because the number basis is independent of temperature, differentiating Eq. (21) changes only the populations, and LT,XL_{T,X} can be chosen diagonal. Substituting this state into the first line of Eq. (20) and taking number-state matrix elements gives ⟨n∣LT,X∣n⟩=∂TpnX/pnX\langle n|L_{T,X}|n\rangle=\partial_Tp_n^X/p_n^X. Inserting these matrix elements into the second line of Eq. (20) yields

FX(T,t)=∑n:pnX>0[∂TpnX(T,t)]2pnX(T,t).(22)\mathcal{F}_X(T,t) = \sum_{n:p_n^X>0}\frac{\left[\partial_Tp_n^X(T,t)\right]^2}{p_n^X(T,t)}. \tag*{(22)}

The sum runs over number states with nonzero probability. Ideal number-resolved detection samples the distribution in Eq. (21). Its classical Fisher information is the population sum in Eq. (22), giving

Fnum,X(T,t)=FX(T,t),(23)\mathcal{F}_{\mathrm{num},X}(T,t) = \mathcal{F}_X(T,t), \tag*{(23)}

where Fnum,X\mathcal{F}_{\mathrm{num},X} denotes the classical Fisher information of the number-measurement outcomes. Thus, number measurement attains the QFI in a basis independent of the unknown temperature. Under the usual regularity conditions, an asymptotically efficient estimator can approach the variance bound in Eq. (19) as the number of independent repetitions increases.

Having expressed the QFI in terms of the measured populations, we now compare the finite-time dynamics with the Markovian benchmark. At the same temperature and interrogation time, we define

RQ(T,t)=FFT(T,t)FM(T,t).(24)\mathcal{R}_Q(T,t) = \frac{\mathcal{F}_{\mathrm{FT}}(T,t)}{\mathcal{F}_{\mathrm{M}}(T,t)}. \tag*{(24)}

Here, RQ\mathcal{R}_Q quantifies the effect of the Markovian description on the QFI by comparing the finite-time and Markovian results under the same conditions. Both QFIs use the same initial Fock state, probe frequency, reservoir parameters, temperature, and interrogation time. Values RQ>1\mathcal{R}_Q>1 indicate that the Markovian description predicts a lower QFI, whereas RQ<1\mathcal{R}_Q<1 indicate that it predicts a higher QFI. Both QFIs describe the same protocol, so RQ\mathcal{R}_Q measures the information that the Markovian description misses. Evaluating Eq. (24) requires the populations and their temperature derivatives. We next determine how transitions change these populations during the interrogation and which reservoir frequencies contribute to their temperature dependence.

Spectral mechanism of enhancement

To explain how finite-time reservoir response can enhance the QFI, we first identify how the transition probabilities and their temperature dependence contribute to it. We consider an interrogation during which the probability of leaving the initial Fock state is small. Therefore leading population changes then involve only the neighbouring number states. Expanding the population dynamics obtained from Eq. (7) to leading order gives

pn0−1FT(T,t)≃n0K−,FT(T,t),p_{n_0-1}^{\mathrm{FT}}(T,t) \simeq n_0\mathcal{K}_{-,\mathrm{FT}}(T,t),
pn0FT(T,t)≃1−εFT(T,t),p_{n_0}^{\mathrm{FT}}(T,t) \simeq1-\varepsilon_{\mathrm{FT}}(T,t),
pn0+1FT(T,t)≃(n0+1)K+,FT(T,t).(25)p_{n_0+1}^{\mathrm{FT}}(T,t) \simeq(n_0+1)\mathcal{K}_{+,\mathrm{FT}}(T,t). \tag*{(25)}

Here, the integrated transition coefficients are

K±,FT(T,t)=∫0tγ±,FT(T,s) ds,(26)\mathcal{K}_{\pm,\mathrm{FT}}(T,t) = \int_0^t\gamma_{\pm,\mathrm{FT}}(T,s)\,\mathrm{d}s, \tag*{(26)}

and the leading probability of leaving the initial state is

εFT=n0K−,FT+(n0+1)K+,FT.(27)\varepsilon_{\mathrm{FT}} = n_0\mathcal{K}_{-,\mathrm{FT}}+(n_0+1)\mathcal{K}_{+,\mathrm{FT}}. \tag*{(27)}

For n0=0n_0 = 0, the downward-transition outcome is absent. The expansion requires εFT≪1\varepsilon_{\mathrm{FT}} \ll1, in addition to the weak-coupling approximation used to obtain the master equation (7). Its derivation and validity are discussed in Appendix B.

Differentiating the populations in Eq. (25) with respect to temperature, substituting the results into Eq. (22), and retaining the leading contributions gives

FFTlead=χFT2[n0K−,FT+n0+1K+,FT],(28)\mathcal{F}_{\mathrm{FT}}^{\mathrm{lead}} = \chi_{\mathrm{FT}}^2 \left[\frac{n_0}{\mathcal{K}_{-,\mathrm{FT}}} + \frac{n_0+1}{\mathcal{K}_{+,\mathrm{FT}}}\right], \tag*{(28)}

where lead denotes the leading weak-transition expression and

χFT(T,t)=∂TK−,FT(T,t)=∂TK+,FT(T,t).(29)\chi_{\mathrm{FT}}(T,t) = \partial_T \mathcal{K}_{-,\mathrm{FT}}(T,t) = \partial_T \mathcal{K}_{+,\mathrm{FT}}(T,t). \tag*{(29)}

The response χFT\chi_{\mathrm{FT}} is the temperature derivative of the integrated transition coefficients. Equation (28) omits the higher-order contribution from the initially occupied state. The temperature derivatives of all three retained populations are of the same order, but the initial-state probability remains close to unity. Dividing the squared derivatives by the probabilities in Eq. (22) therefore makes the contributions from the weakly populated neighbouring states dominant, as detailed in Appendix B.

Equation (28) shows the two features that determine the QFI. The factor χFT2\chi_{\mathrm{FT}}^2 describes the temperature dependence of the transitions, while the bracketed terms account for their probabilities through K±,FT\mathcal{K}_{\pm,\mathrm{FT}} and the initial occupation n0n_0. Finite-time effects change both the transition probabilities and their temperature dependence. Therefore, an increase in transition probabilities alone does not guarantee better temperature estimation. We next determine how different reservoir frequencies affect these probabilities and their temperature dependence.

Substituting Eqs. (9) and (10) into Eq. (26) and performing the time integral gives

K−,FT(T,t)=λ2t2∫0∞dω J(ω)[nT(ω)+1]Φt(ω),\mathcal{K}_{-,\mathrm{FT}}(T,t) = \lambda^2 t^2 \int_0^\infty\mathrm{d}\omega\, J(\omega)[n_T(\omega)+1]\Phi_t(\omega),
K+,FT(T,t)=λ2t2∫0∞dω J(ω)nT(ω)Φt(ω),(30)\mathcal{K}_{+,\mathrm{FT}}(T,t) = \lambda^2 t^2 \int_0^\infty\mathrm{d}\omega\, J(\omega)n_T(\omega)\Phi_t(\omega), \tag*{(30)}

where the finite-time filter is

Φt(ω)=sinc⁡2[(ω−ω0)t2].(31)\Phi_t(\omega) = \operatorname{sinc}^2\left[\frac{(\omega-\omega_0)t}{2}\right]. \tag*{(31)}

These integrals resolve the transition coefficients into contributions from different reservoir frequencies. The spectrum J(ω)J(\omega) describes the distribution of coupling strengths, the occupation factors distinguish upward and downward transitions, and Φt(ω)\Phi_t(\omega) weights their contributions during the interrogation. The central lobe of the filter has a width proportional to 1/t1/t, so longer interaction times make the frequency weighting increasingly selective around ω0\omega_0.

To determine the temperature dependence of these transitions, we differentiate K+,FT\mathcal{K}_{+,\mathrm{FT}} in Eq. (30) with respect to TT, holding the interrogation time, probe frequency, and reservoir spectral and coupling parameters fixed. Using Eq. (29) then gives

χFT(T,t)=λ2t2∫0∞dω WT(ω)Φt(ω),(32)\chi_{\mathrm{FT}}(T,t) = \lambda^2 t^2 \int_0^\infty\mathrm{d}\omega\, W_T(\omega)\Phi_t(\omega), \tag*{(32)}

where

WT(ω)=J(ω)∂TnT(ω).(33)W_T(\omega) = J(\omega)\partial_T n_T(\omega). \tag*{(33)}

The thermal spectral weight WT(ω)W_T(\omega) describes how strongly reservoir modes at frequency ω\omega are coupled to the probe and how their occupation changes with temperature. It identifies frequencies that can contribute strongly to the temperature dependence of the probe populations when the filter gives them appreciable weight. The spectrum J(ω)J(\omega) alone does not contain this information about the temperature response.

We compare this response with the Markovian benchmark at the same temperature and interrogation time. Integrating the constant rates in Eq. (15) and differentiating the resulting integrated coefficients with respect to temperature gives

K−,M=γ0t(nˉ+1),K+,M=γ0tnˉ,\mathcal{K}_{-,\mathrm{M}} = \gamma_0 t(\bar{n}+1), \qquad\mathcal{K}_{+,\mathrm{M}} = \gamma_0 t\bar{n},
χM=γ0t∂Tnˉ=2πλ2tWT(ω0).(34)\chi_{\mathrm{M}} = \gamma_0 t\partial_T\bar{n} = 2\pi\lambda^2 t W_T(\omega_0). \tag*{(34)}

Here, χM\chi_{\mathrm{M}} is the temperature derivative of the Markovian integrated transition coefficients. The Markovian response samples the thermal spectral weight only at the probe frequency ω0\omega_0. Dividing Eq. (32) by this response χM\chi_{\mathrm{M}} gives

Rχ(T,t)=χFTχM=t2πWT(ω0)∫0∞dω WT(ω)Φt(ω).(35)\mathcal{R}_{\chi}(T,t) = \frac{\chi_{\mathrm{FT}}}{\chi_{\mathrm{M}}} = \frac{t}{2\pi W_T(\omega_0)}\int_0^\infty\mathrm{d}\omega\, W_T(\omega)\Phi_t(\omega). \tag*{(35)}

The ratio Rχ\mathcal{R}_{\chi} compares the temperature derivatives of the integrated transition coefficients K±,FT(T,t)\mathcal{K}_{\pm,\mathrm{FT}}(T,t). When Rχ>1\mathcal{R}_{\chi} > 1, the leading near-neighbour-state probabilities change more strongly with temperature than in the Markovian benchmark. The ratio does not depend on the initial occupation n0n_0. For the spectrum considered here, the ratio is well defined at T>0T > 0 and t>0t > 0.

To determine whether this stronger response produces a QFI advantage, we must also account for the transition probabilities. Applying Eq. (28) to both evolutions and taking their ratio gives

RQlead=FFTleadFMlead=Rχ2RK,(36)\mathcal{R}_{\mathcal{Q}}^{\mathrm{lead}} = \frac{\mathcal{F}_{\mathrm{FT}}^{\mathrm{lead}}}{\mathcal{F}_{\mathrm{M}}^{\mathrm{lead}}} = \mathcal{R}_{\chi}^2\mathcal{R}_{\mathcal{K}}, \tag*{(36)}

where

RK=n0/K−,FT+(n0+1)/K+,FTn0/K−,M+(n0+1)/K+,M.(37)\mathcal{R}_{\mathcal{K}} = \frac{n_0/\mathcal{K}_{-,\mathrm{FT}}+(n_0+1)/\mathcal{K}_{+,\mathrm{FT}}}{n_0/\mathcal{K}_{-,\mathrm{M}}+(n_0+1)/\mathcal{K}_{+,\mathrm{M}}}. \tag*{(37)}

Finite-time sampling of off-resonant reservoir modes changes both the integrated transition coefficients K±\mathcal{K}_{\pm} and their temperature derivatives. The ratio Rχ\mathcal{R}_{\chi} compares these derivatives with the Markovian benchmark, whereas RK\mathcal{R}_{K} compares the inverse-K±K_{\pm} terms in the leading QFI given in Eq. (28). If both integrated coefficients exceed their Markovian counterparts, RK<1\mathcal{R}_{K} < 1, so their temperature derivatives must increase sufficiently to outweigh this reduction. Equation (36) expresses this balance through the condition Rχ2RK>1\mathcal{R}_{\chi}^{2}\mathcal{R}_{K} > 1, showing why increased transition probabilities alone do not establish a thermometric advantage. This enhancement criterion applies when the leading population expressions and their temperature derivatives accurately describe both evolutions. In particular, the probabilities of leaving the initial state must remain small, εFT≪1\varepsilon_{\mathrm{FT}} \ll1 and εM≪1\varepsilon_{\mathrm{M}} \ll1, where εM=n0K−,M+(n0+1)K+,M\varepsilon_{\mathrm{M}} = n_0K_{-,\mathrm{M}} + (n_0 + 1)K_{+,\mathrm{M}}. Appendix B discusses these validity conditions, and the following subsection tests the analytical prediction against numerical results.

Figure 1 explains how off-resonant reservoir modes can strengthen the temperature dependence of the transition probabilities, increasing the sensitivity factor Rχ2\mathcal{R}_{\chi}^{2} in Eq. (36). Figure 1(a) compares J(ω)J(\omega) and WT(ω)W_T(\omega), defined in Eqs. (13) and (33), respectively, each normalized by its value at the probe frequency. Their different profiles show why coupling strength alone does not determine a mode’s contribution to temperature sensing. The temperature dependence of its occupation also enters through WT(ω)W_T(\omega). For the chosen spectrum and temperature, WT(ω)/WT(ω0)W_T(\omega)/W_T(\omega_0) is substantially larger than unity over a region below ω0\omega_0. Sampling this region can therefore strengthen the temperature dependence of the transitions, whereas the Markovian response samples only the thermal spectral weight at resonance. Figure 1(b) shows how the finite-time filter in Eq. (31) weights these off-resonant modes. At short interrogation times, its broad central lobe includes the region of elevated thermal weight identified in Figure 1(a). These modes consequently

Spectral origin of the enhanced temperature response

Figure 1. Spectral origin of the enhanced temperature response. (a) Reservoir spectrum J(ω)J(\omega) (blue dashed) and thermal spectral weight WT(ω)W_T(\omega) (orange solid), defined in Eqs. (13) and (33), respectively, each normalized by its value at ω0\omega_0. Vertical dotted lines mark Ω\Omega and ω0\omega_0. (b) The finite-time filter in Eq. (31) at ω0t=0.5\omega_0t = 0.5 (gray dashed), 8 (orange solid), and 40 (green dash-dotted). Vertical dotted lines indicate the same frequencies as in (a). (c) The response ratio Rχ\mathcal{R}_{\chi} in Eq. (35) (orange solid). The blue dashed line marks unity, and shading indicates a stronger temperature response than the Markovian benchmark. Gray, orange, and green markers correspond to the respective interrogation times in (b). Parameters are kBT/(ℏω0)=0.40k_{\mathrm{B}}T/(\hbar\omega_0) = 0.40, Ω/ω0=0.70\Omega/\omega_0 = 0.70, Γ/ω0=0.12\Gamma/\omega_0 = 0.12, and κ/ω0=1\kappa/\omega_0 = 1.

contribute to χFT\chi_{\mathrm{FT}} through Eq. (32). As the interaction time increases, the central lobe narrows around ω0\omega_0, making the frequency weighting increasingly selective near resonance. However, the response also contains the prefactor t2t^2. Access to informative modes is therefore insufficient by itself, since their effect on the probe must develop during the interaction.

Figure 1(c) shows the resulting response ratio Rχ\mathcal{R}_{\chi}. At very short times, Φt(ω)→1\Phi_t(\omega) \to1, so Eq. (32) gives χFT∝t2\chi_{\mathrm{FT}} \propto t^2 for the spectrum considered here. The Markovian response grows linearly with time, giving Rχ∝t\mathcal{R}_{\chi} \propto t. Thus, the broad short-time filter does not immediately produce a stronger temperature response. At intermediate times, the population response has developed while the filter still gives appreciable weight to the off-resonant region of elevated thermal weight. Their combined effect yields χFT>χM\chi_{\mathrm{FT}} > \chi_{\mathrm{M}}. This increases the sensitivity factor Rχ2\mathcal{R}_{\chi}^{2}, providing the mechanism through which finite-time sampling can enhance the leading QFI. At longer times, the frequency weighting becomes increasingly concentrated around ω0\omega_0. The filter and integrated-response limits in Eqs. (65) and (66) give Rχ→1\mathcal{R}_{\chi} \to1. The sensitivity gain identified in Figure 1 produces a leading QFI advantage only when it outweighs any opposing change in the population factor, so that Rχ2RK>1\mathcal{R}_{\chi}^{2}\mathcal{R}_{K} > 1 in Eq. (36).

Numerical results and interpretation

We now determine whether the stronger temperature response identified in Figure 1 outweighs the accompanying changes in transition probabilities and produces a QFI advantage. We propagate the populations and their temperature derivatives using Eq. (55) and its Markovian counterpart, then evaluate the QFI from Eq. (22).

This calculation includes successive transitions within the second-order master equation without imposing the additional weak-transition expansion. To assess the accuracy of the TCL2 description, we also propagate the populations and their temperature derivatives using the exact time-local rates in Eq. (18) and evaluate the corresponding QFI. The numerical methods and convergence checks are presented in Appendix C. We express the numerical results in terms of

θ=kBTℏω0,τ=ω0t,(38)\theta= \frac{k_{\mathrm{B}}T}{\hbar\omega_0}, \qquad\tau= \omega_0 t, \tag*{(38)}

where θ\theta and τ\tau are the dimensionless temperature and interrogation time. Applying the chain rule to Eq. (22) gives

FX(T,t)=(kBℏω0)2Fθ,X(θ,τ),(39)\mathcal{F}_X(T,t) = \left(\frac{k_{\mathrm{B}}}{\hbar\omega_0}\right)^2 \mathcal{F}_{\theta,X}(\theta,\tau), \tag*{(39)}

where Fθ,X\mathcal{F}_{\theta,X} is the QFI for estimating θ\theta, with X∈{FT,M,ex}X \in\{\mathrm{FT},\mathrm{M},\mathrm{ex}\}. The common conversion factor cancels from RQR_Q. Both evolutions use the same initial Fock state and physical parameters, specified in Figure 2. The spectrum and temperature match those in Figure 1, connecting the spectral mechanism directly to its numerical test. Figure 2(a) shows that the finite-time QFI is approximately five times the Markovian prediction at the same interrogation time. The leading expression in Eq. (36) closely follows the numerically calculated ratio. The exact solution coincides with the TCL2 result. Their QFIs differ by less than 0.3%0.3\% over 0<τ≤600 < \tau\le60, so the TCL2 truncation does not affect our conclusions. The maximum lies at τ≃7.5\tau\simeq7.5, well inside the reservoir memory time ω0τB≃17\omega_0\tau_B \simeq17. In this interval, the probability of leaving the initial state is small in both evolutions, and the individual leading QFIs agree with their numerical counterparts.

Three panels showing transient QFI enhancement and comparison with the Markovian benchmark

Figure 2. Transient QFI enhancement and comparison with the Markovian benchmark. (a) The QFI ratio RQR_Q in Eq. (24), obtained by numerically integrating Eq. (7) and its Markovian counterpart with rates (15) (orange solid), compared with the analytical prediction in Eq. (36) (gray dashed) and with the exact solution of Sec. II C (black dotted). The marker identifies the maximum relative enhancement. The analytical curve is displayed only where εFT,εM<0.05\varepsilon_{\mathrm{FT}},\varepsilon_{\mathrm{M}} < 0.05. (b) Absolute QFIs Fθ,X\mathcal{F}_{\theta,X} for finite-time (orange solid), exact (black dotted), and Markovian (blue dashed) dynamics. The marker identifies the first local maximum of the finite-time QFI. (c) The numerically calculated QFI ratio over an extended interval, shown on a logarithmic time axis, for the TCL2 (orange solid) and exact (black dotted) dynamics. Blue dotted lines in panels (a) and (c) mark unity. Parameters are θ=0.40\theta= 0.40, Ω/ω0=0.70\Omega/\omega_0 = 0.70, Γ/ω0=0.12\Gamma/\omega_0 = 0.12, κ/ω0=1\kappa/\omega_0 = 1, λ=0.0075\lambda= 0.0075, and n0=4n_0 = 4.

These checks support the weak-transition description of the enhancement explained in Sec. III B. Near the ratio maximum, RX2>1\mathcal{R}_X^2 > 1, while RK<1R_K < 1, and their product exceeds unity. Thus, the gain in temperature sensitivity outweighs the opposing effect of the change in transition probabilities. This balance confirms the analytical mechanism developed in Sec. III B, with the sensitivity gain arising from the off-resonant contributions identified in Figure 1.

Figure 2(b) shows the absolute QFI per interrogation. The finite-time QFI reaches its first local maximum later than the maximum relative enhancement in Figure 2(a). These times differ because RQR_Q compares two evolving QFIs, whereas Fθ,FT\mathcal{F}_{\theta,\mathrm{FT}} measures the information carried by the finite-time probe state itself. The marker identifies the first local maximum and does not establish a global optimum. A large relative enhancement, therefore, need not coincide with the largest absolute information. Figure 2(c) extends the comparison to longer times and shows the QFI ratio approaching unity. This behavior is consistent with the increasingly resonance-selective response discussed in Sec. III B. The curve is calculated from numerically propagated populations throughout the displayed interval. Together, Figure 1 and Figure 2 explain the origin of the increased temperature sensitivity and demonstrate that it produces a transient QFI advantage. The scans in Appendix C show that the advantage persists across the investigated linewidths at the baseline reservoir frequency, while also identifying settings without enhancement within the sampled time interval.

Information per unit time

The ratio RQR_{Q} compares two descriptions of the same experiment at a fixed interrogation time. To assess the operational benefit of the finite-time response, we consider the temperature information acquired within a fixed total experimental time. For independent repetitions of duration t+toht+t_{\mathrm{oh}}, where toht_{\mathrm{oh}} accounts for state preparation and readout, the number of measurements is m≃ttot/(t+toh)m\simeq t_{\mathrm{tot}}/(t+t_{\mathrm{oh}}). Equation (19) then gives Δ2T≥[ttotFX/(t+toh)]−1\Delta^{2}T\geq[t_{\mathrm{tot}}\mathcal{F}_{X}/(t+t_{\mathrm{oh}})]^{-1}. The information rate FX/(t+toh)\mathcal{F}_{X}/(t+t_{\mathrm{oh}}) therefore determines the precision bound for a given time budget.

Figure 3(a) compares the information rates for negligible overhead. Under the exact dynamics, the rate reaches a maximum within the reservoir memory time. At this interrogation time, the probe has acquired appreciable temperature information while the interaction remains short enough to allow frequent repetition. The Markovian description instead predicts a nearly constant rate at short times, reflecting the linear growth of its QFI. The transient peak is about five times this Markovian rate. Because the latter varies little over the weak-transition regime, the rate advantage closely follows the same-time QFI enhancement in Figure 2(a).

Information per unit interrogation time

Figure 3. Information per unit interrogation time. (a) Rate Fθ/τ\mathcal{F}_{\theta}/\tau for the exact dynamics (black solid) and the Markovian description (blue dashed), with toh=0t_{\mathrm{oh}}=0. Beyond τ=400\tau=400 the exact curve uses the Markovian populations, since RQR_{Q} is within 0.4% of unity there Figure 2(c). The green dash-dotted line marks γ0Fθeq\gamma_{0}\mathcal{F}_{\theta}^{\mathrm{eq}}, an upper bound on the equilibrium rate. The vertical dotted line marks the reservoir memory time ω0τB≃17\omega_{0}\tau_{B}\simeq17. (b) Rate advantage A\mathcal{A} in Eq. (40) versus the overhead per interrogation. The blue dotted line marks unity. Parameters are those of Figure 2.

The benefit of short interrogations is also evident when comparing different interaction durations under the exact dynamics. For interrogations longer than three reservoir memory times, the information acquired per unit time remains below the transient peak over the investigated interval. Although longer interactions can increase the information obtained from an individual measurement, they allow fewer repetitions within the total time budget. Equilibrium thermometry incurs the additional cost of waiting for the probe to thermalize. The equilibrium QFI, Fθeq=nˉ(nˉ+1)/θ4\mathcal{F}_{\theta}^{\mathrm{eq}}=\bar{n}(\bar{n}+1)/\theta^{4}, must therefore be considered together with the relaxation timescale 1/γ01/\gamma_{0}. Taking this timescale as an optimistic equilibration-cycle duration gives the dimensionless rate estimate (γ0/ω0)Fθeq(\gamma_{0}/\omega_{0})\mathcal{F}_{\theta}^{\mathrm{eq}} shown in Figure 3(a). The transient optimum is about 30 times higher than this estimate. Preparation and readout overhead reduce the benefit of frequent repetition. To quantify this effect, we compare the optimized rates,

A(toh)=max⁡tFθ,ex(t)/(t+toh)max⁡tFθ,M(t)/(t+toh).(40)\mathcal{A}(t_{\mathrm{oh}})=\frac{\max_{t}\mathcal{F}_{\theta,\mathrm{ex}}(t)/(t+t_{\mathrm{oh}})}{\max_{t}\mathcal{F}_{\theta,\mathrm{M}}(t)/(t+t_{\mathrm{oh}})}. \tag*{(40)}

Figure 3(b) shows that small overhead preserves most of the transient advantage. As the overhead increases, short interrogations devote a larger fraction of each cycle to preparation and readout. Longer interactions then become more competitive, and the rate advantage decreases. For the parameters studied, retaining a substantial improvement requires preparation and readout within a few tens of inverse probe frequencies. This condition connects the predicted advantage to the duration of the complete measurement cycle. Lower-frequency probes allow a longer overhead in physical time for the same dimensionless constraint, provided that their preparation and readout can be completed on that timescale.

Finite switching time

So far the coupling has switched on and off instantaneously. The filter in Eq. (31) is the Fourier transform of this rectangular window, and its slowly decaying side lobes reach far from ω0\omega_{0}. A real coupler switches in a finite time. We therefore replace λ\lambda in Eq. (4) by λf(s)\lambda f(s) for 0≤s≤t0\leq s\leq t. The envelope f(s)f(s) rises as sin⁡2[πs/(2tr)]\sin^{2}[\pi s/(2t_{r})] over a ramp time trt_{r}, stays at unity, and falls symmetrically at the end. To leading order in λ\lambda, the switching enters only through the filter. Appendix E shows that t2Φt(ω)t^{2}\Phi_{t}(\omega) in Eqs. (30) and (32) becomes ∣f~(ω)∣2|\tilde{f}(\omega)|^2, with

f~(ω)=∫0tds f(s)ei(ω0−ω)s.(41)\tilde{f}(\omega) = \int_{0}^{t} \mathrm{d}s\, f(s)e^{i(\omega_0-\omega)s}. \tag*{(41)}

The Markovian benchmark uses the same envelope, which replaces tt by teff=∫0tf2(s) dst_{\mathrm{eff}} = \int_{0}^{t} f^2(s)\,\mathrm{d}s in Eq. (34).

Figure 4(a) shows that ramps suppress the side lobes and lower the central lobe at fixed tt. The relative gain survives. Figure 4(b) shows that RQmax⁡R_Q^{\max} stays between 5.1 and 5.6 for ω0tr≤8\omega_0t_r \le8 and falls to 3.5 at ω0tr=12\omega_0t_r = 12. The informative modes near Ω\Omega lie inside the central lobe of the filter, so the side lobes contribute little. The gain is determined by the finite interaction duration, without requiring abrupt switching.

Effect of a finite switching time

Figure 4. Effect of a finite switching time. (a) Normalized filter ∣f~(ω)∣2/t2|\tilde{f}(\omega)|^2/t^2 from Eq. (41) at ω0t=12\omega_0 t = 12 for ramp times ω0tr=0\omega_0 t_r = 0 (orange solid), 3 (blue dashed), and 6 (green dash-dotted). Vertical dotted lines mark Ω\Omega and ω0\omega_0. (b) Maximum QFI ratio against the Markovian benchmark with the same envelope (orange circles) and rate advantage over the best Markovian rate at toh=0t_{\mathrm{oh}} = 0 (black squares), both maximized over 0<ω0t≤800 < \omega_0 t \le80. The vertical dotted line marks ω0tr=1/∣ω0−Ω∣\omega_0 t_r = 1/|\omega_0-\Omega|, and the horizontal dotted line marks unity. Results use the leading-order expressions. Other parameters are those of Figure 2.

Ramps require additional time, extending the interrogation without providing full-strength coupling and thereby reducing the information rate. With toh=0t_{\mathrm{oh}} = 0, the rate advantage over the best Markovian rate drops from 5.1 for instantaneous switching to 3.6 at ω0tr=3\omega_0t_r = 3, 2.6 at ω0tr=6\omega_0t_r = 6, and 1.3 at ω0tr=12\omega_0t_r = 12. Retaining most of the rate advantage therefore requires ramps shorter than about 1/∣ω0−Ω∣1/|\omega_0-\Omega|, which equals 3.3/ω03.3/\omega_0 here. All results in this subsection satisfy εFT<0.012\varepsilon_{\mathrm{FT}} < 0.012 at the ratio maxima, which keeps them inside the weak-transition regime.

Temperature and initial-state dependence

The gain persists over the temperature range we examine. Figure 5 shows the maximum propagated ratio over 0<τ≤600 < \tau\le60 for 0.15≤θ≤2.40.15 \le\theta\le2.4. At n0=4n_0 = 4 it rises from 4.7 at high temperature to 7.7 at θ=0.2\theta= 0.2 and 11.9 at θ=0.15\theta= 0.15. The enhancement is strongest for cold reservoirs. In this regime ∂TnT(ω)\partial_T n_T(\omega) grows rapidly toward low frequencies, so the modes below ω0\omega_0 carry much more thermal weight than the resonant mode. The maximizing time stays between τ≃6.9\tau\simeq6.9 and 7.7 at all temperatures, which reflects the fixed spectral detuning.

Temperature and initial-state dependence of the transient QFI enhancement

Figure 5. Temperature and initial-state dependence of the transient QFI enhancement. The maximum propagated ratio RQmax⁡R_Q^{\max} over 0<τ≤600 < \tau\le60 is shown versus θ\theta for n0=0n_0 = 0 (blue circles), 1 (green squares), 4 (orange diamonds), and 10 (black triangles). The dotted line marks unity. Other parameters are those of Figure 2.

The initial number state has little effect on the enhancement. At θ=0.4\theta= 0.4, the maximum ratio is 4.8 for the vacuum, 5.0 for n0=1n_0 = 1, 5.1 for n0=4n_0 = 4, and 5.0 for n0=10n_0 = 10. The curves for different n0n_0 nearly coincide at all temperatures. This follows from Eq. (36): the sensitivity factor Rχ\mathcal{R}_{\chi} is independent of n0n_0, while n0n_0 only reweights the two inverse-K±\mathcal{K}_{\pm} terms in RK\mathcal{R}_{\mathcal{K}}. Thus, the finite-time gain is primarily determined by the reservoir response rather than the initial number state. Fock states still change the absolute information, which is the role studied in Refs. [43, 44].

Conclusion

We have shown that finite-time reservoir response can enhance transient thermometry with a harmonic-oscillator probe prepared in a number state. Off-resonant reservoir modes can strengthen the temperature dependence of the probe populations and thereby increase the QFI above the Markovian prediction. Our analysis identifies which reservoir frequencies contribute most strongly to the temperature dependence of the probe populations.

When the probability of leaving the initial state is small, the analytical criterion determines whether this increased sensitivity outweighs the accompanying changes in transition probabilities. Comparison with the exact dynamics confirms the transient enhancement predicted by the second-order master equation.

The operational significance of this enhancement follows from the temperature information acquired per unit time. For the parameters studied, the exact dynamics yield an optimal interrogation within the reservoir memory time, with an information rate exceeding both the Markovian benchmark and that available from longer interrogations. Preparation and readout overhead reduce this advantage, making the duration of the full experimental cycle an essential consideration. The enhancement survives smooth switching because the informative reservoir modes remain within the central lobe of the finite-time filter, although longer ramps reduce the information rate. Across the investigated range, the relative QFI enhancement increases at lower temperatures and depends weakly on the initial occupation.

Ideal excitation-number measurements attain the QFI because the probe remains diagonal in a temperature-independent number basis.

An experimental test would combine repeated number-state preparation, a controlled interaction duration, and measurements of the excitation populations. The reservoir temperature could be inferred by fitting these populations to the predicted dynamics. Measurements at nearby calibrated temperatures would also allow the population derivatives and the corresponding Fisher information to be estimated at different interrogation times.

Trapped-ion experiments have demonstrated heating and decay of motional Fock states in engineered reservoirs [65], while superconducting resonators support deterministic oscillator-state preparation and Wigner tomography [66]. These capabilities provide a starting point for an implementation with a characterized thermal reservoir and suitable off-resonant spectral weight. Observing the predicted information-rate advantage would further require sufficiently rapid coupling control, preparation, and readout. These findings provide a basis for choosing interrogation times that improve the precision of transient thermometry under experimental timing constraints.

Acknowledgments

M. Tahir Naseem acknowledges support from the National Center for Quantum Computing (NCQC), Pakistan. A. U. and Ö. E. M. acknowledge financial support from the Scientific and Technological Research Council of Türkiye (TÜBİTAK) under Project Grant No. 123F150.

Data availability

The data that support the findings of this article are available from the authors upon reasonable request.

Derivation of the finite-time master equation

We derive the finite-time master equation to second order in the probe–reservoir coupling. Transforming the interaction Hamiltonian in Eq. (4) with the free Hamiltonian HS+HBH_S+H_B gives

H~I(t)=ℏλ[a†eiω0tB(t)+ae−iω0tB†(t)],(42)\widetilde{H}_I(t)=\hbar\lambda\left[a^\dagger e^{i\omega_0t}B(t)+ae^{-i\omega_0t}B^\dagger(t)\right], \tag*{(42)}

where the tilde denotes the interaction picture and

B(t)=∑kgkbke−iωkt.(43)B(t)=\sum_k g_kb_ke^{-i\omega_kt}. \tag*{(43)}

For the initially factorized state in Eq. (5), the second-order time-convolutionless expansion gives

dρ~(t)dt=−1ℏ2∫0tds Tr⁡B[H~I(t),[H~I(s),ρ~(t)⊗ρB(T)]].(44)\frac{d\widetilde{\rho}(t)}{dt}=-\frac{1}{\hbar^2}\int_0^t ds\,\operatorname{Tr}_B\left[\widetilde{H}_I(t),\left[\widetilde{H}_I(s),\widetilde{\rho}(t)\otimes\rho_B(T)\right]\right]. \tag*{(44)}

Here, ρ~(t)\widetilde{\rho}(t) is the reduced probe state, and ρB(T)\rho_B(T) is the initial thermal reservoir state. We suppress the temperature argument of the reduced state throughout this Appendix. In Eq. (44), the first-order contribution vanishes because the thermal reservoir has ⟨bk⟩B=⟨bk†⟩B=0\langle b_k\rangle_B=\langle b_k^\dagger\rangle_B=0, which implies Tr⁡B[H~I(t)ρB(T)]=0\operatorname{Tr}_B[\widetilde{H}_I(t)\rho_B(T)]=0. Thus, the leading reservoir-induced change in the reduced state is second order in λ\lambda. Equation (44) specifies the truncated generator without assuming that the exact joint state remains factorized during the evolution. The finite integration interval is retained.

To evaluate the bath trace in Eq. (44), we first expand the double commutator,

[Ht,[Hs,R]]=HtHsR−HtRHs−HsRHt+RHsHt.(45)[H_t,[H_s,R]]=H_tH_sR-H_tRH_s-H_sRH_t+RH_sH_t. \tag*{(45)}

where Ht=H~I(t)H_t=\widetilde{H}_I(t), Hs=H~I(s)H_s=\widetilde{H}_I(s), and R=ρ~(t)⊗ρB(T)R=\widetilde{\rho}(t)\otimes\rho_B(T). Substituting H~I\widetilde{H}_I into these terms produces products containing two reservoir operators. Their thermal traces follow from ⟨bkbℓ†⟩B=δkℓ[nT(ωk)+1]\langle b_kb_\ell^\dagger\rangle_B=\delta_{k\ell}[n_T(\omega_k)+1], ⟨bk†bℓ⟩B=δkℓnT(ωk)\langle b_k^\dagger b_\ell\rangle_B=\delta_{k\ell}n_T(\omega_k), and ⟨bkbℓ⟩B=⟨bk†bℓ†⟩B=0\langle b_kb_\ell\rangle_B=\langle b_k^\dagger b_\ell^\dagger\rangle_B=0. Here, ⟨O⟩B=Tr⁡B[OρB(T)]\langle O\rangle_B=\operatorname{Tr}_B[O\rho_B(T)], and nT(ω)n_T(\omega) is the Bose occupation. Using the bath operator B(t)B(t) and these thermal averages, then converting the mode sums to frequency integrals using Eq. (11), gives the two-time bath correlations

C−(u)=⟨B(u)B†(0)⟩B=∫0∞dω J(ω)[nT(ω)+1]e−iωu,(46)C_{-}(u)=\langle B(u)B^{\dagger}(0)\rangle_{B}=\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)[n_{T}(\omega)+1]e^{-i\omega u}, \tag*{(46)}
C+(u)=⟨B†(u)B(0)⟩B=∫0∞dω J(ω)nT(ω)eiωu.(47)C_{+}(u)=\langle B^{\dagger}(u)B(0)\rangle_{B}=\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)n_{T}(\omega)e^{i\omega u}. \tag*{(47)}

Here, u=t−su=t-s is the time difference. Thermal stationarity makes the correlations depend only on uu and gives C±(−u)=C±∗(u)C_{\pm}(-u)=C_{\pm}^{*}(u).

Using these correlations to evaluate the four bath traces and substituting the result into Eq. (44) yields

dρ~dt=G−(aρ~a†−a†aρ~)+G−∗(aρ~a†−ρ~a†a)+G+(a†ρ~a−aa†ρ~)+G+∗(a†ρ~a−ρ~aa†),(48)\frac{\mathrm{d}\tilde{\rho}}{\mathrm{d}t}=G_{-}\left(a\tilde{\rho}a^{\dagger}-a^{\dagger}a\tilde{\rho}\right)+G_{-}^{*}\left(a\tilde{\rho}a^{\dagger}-\tilde{\rho}a^{\dagger}a\right)+G_{+}\left(a^{\dagger}\tilde{\rho}a-aa^{\dagger}\tilde{\rho}\right)+G_{+}^{*}\left(a^{\dagger}\tilde{\rho}a-\tilde{\rho}aa^{\dagger}\right), \tag*{(48)}

where ρ~=ρ~(t)\tilde{\rho}=\tilde{\rho}(t) and G±=G±,FT(T,t)G_{\pm}=G_{\pm,\mathrm{FT}}(T,t) abbreviate the response coefficients obtained after changing variables from ss to uu,

G−,FT(T,t)=λ2∫0tdu eiω0uC−(u),G_{-,\mathrm{FT}}(T,t)=\lambda^{2}\int_{0}^{t}\mathrm{d}u\,e^{i\omega_{0}u}C_{-}(u),
G+,FT(T,t)=λ2∫0tdu e−iω0uC+(u).(49)G_{+,\mathrm{FT}}(T,t)=\lambda^{2}\int_{0}^{t}\mathrm{d}u\,e^{-i\omega_{0}u}C_{+}(u). \tag*{(49)}

The vanishing anomalous bath correlations eliminate terms containing two probe creation or two probe annihilation operators. For the excitation-exchange interaction in Eq. (4), no additional secular averaging is required. Retaining the upper limit tt in Eq. (49) preserves the transient reservoir response. Extending these correlation-time integrals to infinity gives the Markovian coefficients when the corresponding limits exist.

Substituting the correlations into Eq. (49) and performing the finite-time integration gives

G−,FT(T,t)=λ2∫0∞dω J(ω)[nT(ω)+1]Λ+(Δ,t),G_{-,\mathrm{FT}}(T,t)=\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)[n_{T}(\omega)+1]\Lambda_{+}(\Delta,t),
G+,FT(T,t)=λ2∫0∞dω J(ω)nT(ω)Λ−(Δ,t),(50)G_{+,\mathrm{FT}}(T,t)=\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)n_{T}(\omega)\Lambda_{-}(\Delta,t), \tag*{(50)}

where

Λ±(Δ,t)=sin⁡(Δt)Δ±i1−cos⁡(Δt)Δ.(51)\Lambda_{\pm}(\Delta,t)=\frac{\sin(\Delta t)}{\Delta}\pm i\frac{1-\cos(\Delta t)}{\Delta}. \tag*{(51)}

Here, Δ=ω0−ω\Delta=\omega_{0}-\omega is the probe–mode detuning. At resonance, the continuous limit is Λ±(0,t)=t\Lambda_{\pm}(0,t)=t, so the finite-time kernels are regular. We denote these kernels by Λ±\Lambda_{\pm} to distinguish them from the integrated coefficients K±,FTK_{\pm,\mathrm{FT}} in Eq. (26).

Separating the real and imaginary contributions and returning to the Schrödinger picture gives Eq. (7), with γ±,FT=2Re⁡G±,FT\gamma_{\pm,\mathrm{FT}}=2\operatorname{Re}G_{\pm,\mathrm{FT}}. The Lamb-shift Hamiltonian is

HLS(T,t)=ℏS−,FT(T,t)a†a+ℏS+,FT(T,t)aa†,(52)H_{\mathrm{LS}}(T,t)=\hbar S_{-,\mathrm{FT}}(T,t)a^{\dagger}a+\hbar S_{+,\mathrm{FT}}(T,t)aa^{\dagger}, \tag*{(52)}

where S±,FT=Im⁡G±,FTS_{\pm,\mathrm{FT}}=\operatorname{Im}G_{\pm,\mathrm{FT}} are the frequency-shift coefficients. Taking the imaginary parts, using Eq. (51), gives

S−,FT(T,t)=λ2∫0∞dω J(ω)[nT(ω)+1]1−cos⁡(Δt)Δ,S_{-,\mathrm{FT}}(T,t)=\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)[n_{T}(\omega)+1]\frac{1-\cos(\Delta t)}{\Delta},
S+,FT(T,t)=−λ2∫0∞dω J(ω)nT(ω)1−cos⁡(Δt)Δ.(53)S_{+,\mathrm{FT}}(T,t)=-\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)n_{T}(\omega)\frac{1-\cos(\Delta t)}{\Delta}. \tag*{(53)}

The opposite signs follow from the imaginary parts of Λ+\Lambda_{+} and Λ−\Lambda_{-}. Taking twice the real parts of the same response functions gives

γ−,FT(T,t)=2λ2∫0∞dω J(ω)[nT(ω)+1]sin⁡(Δt)Δ,\gamma_{-,\mathrm{FT}}(T,t)=2\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)[n_{T}(\omega)+1]\frac{\sin(\Delta t)}{\Delta},
γ+,FT(T,t)=2λ2∫0∞dω J(ω)nT(ω)sin⁡(Δt)Δ.(54)\gamma_{+,\mathrm{FT}}(T,t)=2\lambda^{2}\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)n_{T}(\omega)\frac{\sin(\Delta t)}{\Delta}. \tag*{(54)}

These expressions recover Eqs. (9) and (10), completing the derivation of Eq. (7).

Population dynamics and analytical QFI

This appendix derives the population expansion and leading QFI used in Sec. III B. It also establishes the spectral identities and limiting behavior underlying the finite-time thermal response. We take T>0T>0 and hold the interrogation time and all Hamiltonian parameters fixed when differentiating with respect to temperature. Taking diagonal matrix elements of Eq. (7) in the number basis gives

p˙nFT=γ−,FT[(n+1)pn+1FT−npnFT]+γ+,FT[npn−1FT−(n+1)pnFT].(55)\dot{p}_{n}^{\mathrm{FT}}=\gamma_{-,\mathrm{FT}}\left[(n+1)p_{n+1}^{\mathrm{FT}}-np_{n}^{\mathrm{FT}}\right]+\gamma_{+,\mathrm{FT}}\left[np_{n-1}^{\mathrm{FT}}-(n+1)p_{n}^{\mathrm{FT}}\right]. \tag*{(55)}

Here, n≥0n\geq0, p−1FT=0p_{-1}^{\mathrm{FT}}=0, and the common arguments (T,t)(T,t) are suppressed. The factors nn and n+1n+1 follow from the number-state matrix elements of aa and a†a^{\dagger}. The Hamiltonian terms do not change the populations because both the free and Lamb-shift Hamiltonians are diagonal in this basis. The Markovian population equation follows by replacing the finite-time coefficients with Eq. (15).

Integrating Eq. (55) once and inserting the initial populations pn(0)=δn,n0p_{n}(0)=\delta_{n,n_{0}} on its right-hand side gives Eq. (25), with κ±,FT\kappa_{\pm,\mathrm{FT}} and ϵFT\epsilon_{\mathrm{FT}} defined in Eqs. (26) and (27). At fixed TT and tt, the transition coefficients and their integrals are of order λ2\lambda^{2}. A further iteration therefore introduces corrections of order λ4\lambda^{4}, including populations reached through successive transitions. This order counting describes an expansion of the adopted second-order master equation. It does not supply the higher-order terms omitted from its generator. To evaluate the integrated transition coefficients, we substitute Eqs. (9) and (10) into Eq. (26). The required time integral is

∫0t2s sinc⁡(Δs) ds=2[1−cos⁡(Δt)]Δ2=t2sinc⁡2(Δt2)(56)\int_{0}^{t} 2s\,\operatorname{sinc}(\Delta s)\,\mathrm{d}s = \frac{2[1-\cos(\Delta t)]}{\Delta^2} = t^2 \operatorname{sinc}^2\left(\frac{\Delta t}{2}\right) \tag*{(56)}

where Δ=ω0−ω\Delta=\omega_0-\omega, as in Appendix A, and the expression at Δ=0\Delta=0 is understood by continuity. Using this result in the frequency integrals gives Eq. (30). Since J(ω)J(\omega), nT(ω)n_T(\omega), and Φt(ω)\Phi_t(\omega) are nonnegative, the integrated coefficients K±,FT\mathcal{K}_{\pm,\mathrm{FT}} are nonnegative, even when the instantaneous coefficients change sign.

Subtracting the two expressions in Eq. (30) gives

K−,FT−K+,FT=λ2t2∫0∞dω J(ω)Φt(ω).(57)\mathcal{K}_{-,\mathrm{FT}}-\mathcal{K}_{+,\mathrm{FT}}=\lambda^2t^2\int_{0}^{\infty}\mathrm{d}\omega\,J(\omega)\Phi_t(\omega). \tag*{(57)}

For a temperature-independent spectrum, the right-hand side has no temperature dependence. Differentiating therefore establishes the equality of derivatives in Eq. (29). Differentiating the Bose occupation gives

∂TnT(ω)=ℏωkBT2nT(ω)[nT(ω)+1].(58)\partial_T n_T(\omega)=\frac{\hbar\omega}{k_{\mathrm{B}}T^2}n_T(\omega)[n_T(\omega)+1]. \tag*{(58)}

Using this derivative in Eq. (30) yields the thermal-response integral in Eq. (32). The temperature derivatives of the leading populations follow directly from Eqs. (25) and (29),

∂Tpn0−1FT≃n0χFT,∂Tpn0FT≃−(2n0+1)χFT,∂Tpn0+1FT≃(n0+1)χFT.(59)\begin{aligned} \partial_Tp_{n_0-1}^{\mathrm{FT}}&\simeq n_0\chi_{\mathrm{FT}},\\ \partial_Tp_{n_0}^{\mathrm{FT}}&\simeq-(2n_0+1)\chi_{\mathrm{FT}},\\ \partial_Tp_{n_0+1}^{\mathrm{FT}}&\simeq(n_0+1)\chi_{\mathrm{FT}}. \tag*{(59)} \end{aligned}

Their sum vanishes, as required by normalization. Substituting these derivatives and the leading populations into Eq. (22) gives

FFTtr=χFT2[n0K−,FT+n0+1K+,FT]+(2n0+1)2χFT21−εFT.(60)\mathcal{F}_{\mathrm{FT}}^{\mathrm{tr}}=\chi_{\mathrm{FT}}^2\left[\frac{n_0}{\mathcal{K}_{-,\mathrm{FT}}}+\frac{n_0+1}{\mathcal{K}_{+,\mathrm{FT}}}\right]+\frac{(2n_0+1)^2\chi_{\mathrm{FT}}^2}{1-\varepsilon_{\mathrm{FT}}}. \tag*{(60)}

Here, FFTtr\mathcal{F}_{\mathrm{FT}}^{\mathrm{tr}} is the QFI of the normalized distribution retained in Eq. (25), with εFT<1\varepsilon_{\mathrm{FT}}<1. The first contribution comes from the neighbouring populations, and the second comes from the initially occupied state. Terms associated with the downward transition are omitted throughout when n0=0n_0=0. To compare these contributions, the Cauchy–Schwarz inequality gives

(2n0+1)2≤εFT[n0K−,FT+n0+1K+,FT].(61)(2n_0+1)^2\leq\varepsilon_{\mathrm{FT}}\left[\frac{n_0}{\mathcal{K}_{-,\mathrm{FT}}}+\frac{n_0+1}{\mathcal{K}_{+,\mathrm{FT}}}\right]. \tag*{(61)}

Using Eq. (61) in Eq. (60) bounds the initial-state contribution relative to the neighbouring-state contribution by εFT/(1−εFT)\varepsilon_{\mathrm{FT}}/(1-\varepsilon_{\mathrm{FT}}). It is therefore subleading when εFT≪1\varepsilon_{\mathrm{FT}}\ll1. Retaining the leading contribution gives Eq. (28).

At fixed TT and tt, K±,FT\mathcal{K}_{\pm,\mathrm{FT}} and χFT\chi_{\mathrm{FT}} are of order λ2\lambda^2. The leading QFI is consequently of order λ4\lambda^4, whereas the initial-state contribution is of order λ4\lambda^4. Corrections to the neighbouring populations and the populations reached through successive transitions can also contribute at order λ4\lambda^4. Thus, retaining the initial-state term alone does not give the complete QFI at that order. This counting assumes that the population expansion and its temperature derivative remain regular. Small depletion provides a useful validity check, but the accuracy of the QFI also depends on the accuracy of the temperature derivatives. For this reason, Figure 2(a) compares the leading and propagated QFIs directly.

For the Markovian evolution, substituting Eq. (34) into the leading population expression gives

FMlead=γ0t(∂Tnˉ)2[n0nˉ+1+n0+1nˉ].(62)\mathcal{F}_{\mathrm{M}}^{\mathrm{lead}}=\gamma_0t(\partial_T\bar n)^2\left[\frac{n_0}{\bar n+1}+\frac{n_0+1}{\bar n}\right]. \tag*{(62)}

This expression requires

εM=γ0t[n0(nˉ+1)+(n0+1)nˉ]≪1.(63)\varepsilon_{\mathrm{M}}=\gamma_0t[n_0(\bar n+1)+(n_0+1)\bar n]\ll1. \tag*{(63)}

Taking the ratio of Eqs. (28) and (62) then gives the factorization in Eq. (36). Its interpretation applies where both population expansions and their temperature derivatives accurately describe the propagated dynamics. The short-time behavior follows from Φt(ω)→1\Phi_t(\omega)\to1 as t→0t\to0. For the spectrum in Eq. (13), the frequency integrals of J(ω)[nT(ω)+1]J(\omega)[n_T(\omega)+1], J(ω)nT(ω)J(\omega)n_T(\omega), and WT(ω)W_T(\omega) are finite. Equations (30) and (32) therefore give K±,FT∝t2\mathcal{K}_{\pm,\mathrm{FT}}\propto t^2 and χFT∝t2\chi_{\mathrm{FT}}\propto t^2. Using these scalings in Eq. (28) and comparing with Eq. (62) yields

FFTlead∝t2,FMlead∝t,RQlead∝t.(64)\mathcal{F}_{\mathrm{FT}}^{\mathrm{lead}}\propto t^2,\qquad\mathcal{F}_{\mathrm{M}}^{\mathrm{lead}}\propto t,\qquad\mathcal{R}_{Q}^{\mathrm{lead}}\propto t. \tag*{(64)}

Thus, broad spectral access at very short times does not produce an immediate QFI advantage. At long times, the filter satisfies the distributional limit

tΦt(ω)⟶2πδ(ω−ω0).(65)t\Phi_t(\omega)\longrightarrow2\pi\delta(\omega-\omega_0). \tag*{(65)}

For the integrable spectral weights used here, which are continuous at ω0>0\omega_0>0, applying this limit to Eqs. (30) and (32) gives

K−,FTt⟶γ0(nˉ+1),K+,FTt⟶γ0nˉ,χFTt⟶γ0∂Tnˉ.(66)\frac{\mathcal{K}_{-,\mathrm{FT}}}{t}\longrightarrow\gamma_0(\bar n+1),\qquad\frac{\mathcal{K}_{+,\mathrm{FT}}}{t}\longrightarrow\gamma_0\bar n,\qquad\frac{\chi_{\mathrm{FT}}}{t}\longrightarrow\gamma_0\partial_T\bar n. \tag*{(66)}

These limits establish recovery of the Markovian integrated response per unit time. They do not extend the weak-transition population expansion to arbitrarily long interrogations. The longer-time QFI comparison in Figure 2(c) therefore uses the propagated populations.

Numerical validation and spectral robustness

To assess the dependence of the QFI enhancement on the reservoir spectrum, we propagate the populations and their temperature derivatives using Eq. (55), then evaluate the QFI from Eq. (22). The Markovian benchmark is calculated in the same way, using the rates in Eq. (15). We use the adaptive DOP853 integrator with number states from 0 to 40. Frequency integrals are evaluated using composite Gauss–Legendre quadrature, with oscillatory quadrature for the extended evolution in Figure 2(c). The frequency cutoff is ω/ω0=80\omega/\omega_0 = 80 for Figure 2 and 40 for the robustness scan. To check convergence, we increase the frequency cutoff to 80 and the number-state cutoff to 60, and refine the integration and sampling. At the tested low-frequency, near-resonant, and high-frequency settings, these changes alter the peak ratios by less than 10−610^{-6} in relative terms. We also independently validate the propagated temperature derivatives using centered finite differences. These checks establish numerical convergence at the tested settings within the adopted second-order master equation. For the exact solution of Appendix § 2.3.4, we use δω/ω0=0.004\delta\omega/\omega_0 = 0.004 and ωc/ω0=8\omega_c/\omega_0 = 8 for τ≤400\tau\le400, and δω/ω0=0.002\delta\omega/\omega_0 = 0.002 for Figure 2(c). The recurrence time then exceeds the displayed interval. Halving δω\delta\omega and raising ωc\omega_c to 12ω012\omega_0 changes the maximum of RQR_Q by less than 10−610^{-6} in relative terms. For the information-rate benchmark, the switching analysis and the temperature and initial-state scans, frequency integrals are evaluated by trapezoidal quadrature on a nonuniform grid, with spacing 0.0015ω00.0015\omega_0 below 6ω06\omega_0 and logarithmic spacing up to the cutoff. Populations are propagated with the same DOP853 integrator. At the baseline parameters, this procedure reproduces the maximum RQ≃5.08R_Q \simeq5.08 at τ≃7.5\tau\simeq7.5 obtained with Gauss–Legendre quadrature.

We vary the characteristic reservoir frequency Ω\Omega and linewidth parameter Γ\Gamma in Eq. (13), while keeping the temperature, initial state, and remaining coupling parameters fixed. At each spectral setting, both evolutions use the same physical parameters. To compare their relative performance over a common transient interval, we define

RQmax⁡(Ω,Γ)=max⁡0<ω0t≤200RQ(T,t;Ω,Γ).(67)R_Q^{\max}(\Omega,\Gamma) = \max_{0<\omega_0 t\le200} R_Q(T,t;\Omega,\Gamma). \tag*{(67)}

Here, RQmax⁡R_Q^{\max} is the largest propagated QFI ratio within the specified interval, with its dependence on the varied spectral parameters displayed explicitly. A maximum at the upper boundary does not establish an unrestricted optimal interrogation time.

Figure 6 shows that the enhancement persists across the investigated linewidths at the baseline reservoir frequency used in Sec. III. At this frequency, narrower spectra yield larger maximum QFI ratios. The relative enhancement also depends strongly on spectral position. The largest sampled ratios occur at the lower-frequency edge of the scan, while the advantage weakens near resonance for some settings. The ratios increase again at higher reservoir frequencies, showing that enhancement is not restricted to spectra with Ω<ω0\Omega< \omega_0.

Spectral dependence of the transient QFI enhancement

Figure 6. Spectral dependence of the transient QFI enhancement. The maximum propagated ratio in Eq. (67) is shown versus Ω/ω0\Omega/\omega_0 for Γ/ω0=0.08\Gamma/\omega_0 = 0.08 (green dashed), 0.12 (orange solid), and 0.20 (blue dotted). The scan covers 0.45≤Ω/ω0≤1.350.45 \le\Omega/\omega_0 \le1.35 in steps of 0.025, with 0<ω0t≤2000 < \omega_0t \le200. Lines connect the sampled spectral settings. Fixed parameters are θ=0.40\theta= 0.40, κ/ω0=1\kappa/\omega_0 = 1, λ=0.0075\lambda= 0.0075, and n0=4n_0 = 4. The filled orange marker identifies the baseline setting used in Figure 1 and Figure 2. Open markers indicate maxima at the upper time boundary. The horizontal dotted line marks unity, the vertical dotted line marks Ω=ω0\Omega= \omega_0, and the inset resolves the region near unity. All curves are calculated from propagated populations.

These trends can be interpreted through the thermal spectral weight WT(ω)W_T(\omega) in Eq. (33) and the finite-time response in Eq. (35). Changing Ω\Omega and Γ\Gamma changes both the thermal weight available away from resonance and the resonant contribution sampled by the Markovian benchmark. Within the weak-transition regime, Eq. (36) connects the resulting thermal-response gain to the QFI through the accompanying population factor. A mode’s

frequency alone therefore does not determine its thermometric contribution. Some maximizing times in the scan lie outside this analytical regime, so the figure uses propagated QFIs throughout rather than extending the leading factorization beyond its validity. The inset in Figure 6 also identifies spectral settings for which no enhancement is found within the sampled time interval. The spectral modification therefore does not guarantee an advantage. Open markers require particular care since they indicate that the largest sampled ratio occurs at the final interrogation time. Neither the absence of enhancement nor a boundary maximum establishes the behavior outside the specified time window.

Finally, relative enhancement must be distinguished from absolute information. Holding κ\kappa fixed does not keep the resonant damping scale γ0\gamma_0 in Eq. (16) fixed as Ω\Omega and Γ\Gamma vary. For the baseline linewidth, moving from the baseline reservoir frequency to the lower-frequency edge of the scan increases the maximum QFI ratio but reduces the finite-time QFI evaluated at the respective ratio maxima. The largest relative gain, therefore, need not identify the setting with the largest absolute information. The scan supports the persistence and spectral tunability of the transient enhancement over the investigated settings, while showing that its magnitude and occurrence depend on the reservoir parameters.

Exact solution

The Hamiltonian in Eqs. (1)–(4) is quadratic and conserves the total excitation number. The Heisenberg equations

a˙=−iω0a−iλ∑kgkbk,b˙k=−iωkbk−iλgk∗a(68)\dot{a}=-i\omega_0a-i\lambda\sum_k g_kb_k, \qquad \dot{b}_k=-i\omega_kb_k-i\lambda g_k^*a \tag*{(68)}

are therefore linear. Their solution has the form of Eq. (17), with ∣G(t)∣2+∑k∣fk(t)∣2=1|G(t)|^2+\sum_k|f_k(t)|^2=1. Eliminating the reservoir operators gives the integro-differential equation

G˙(t)=−iω0G(t)−λ2∫0tds C0(t−s)G(s),G(0)=1,(69)\dot{G}(t)=-i\omega_0G(t)-\lambda^2\int_0^t ds\,C_0(t-s)G(s), \qquad G(0)=1, \tag*{(69)}

where C0(u)=∫0∞dω J(ω)e−iωuC_0(u)=\int_0^\infty d\omega\,J(\omega)e^{-i\omega u} does not depend on temperature [67].

The initial reservoir state is Gaussian with zero mean, and the combination ∑kfkbk\sum_k f_kb_k is a single bosonic mode in a phase-invariant Gaussian state. The reduced dynamics is therefore a thermal attenuator channel. It mixes the input with an effective thermal mode at transmissivity η=∣G∣2\eta=|G|^2 and adds the mean excitation N(T,t)=∑k∣fk∣2nT(ωk)N(T,t)=\sum_k|f_k|^2n_T(\omega_k). The probe mean occupation is ⟨a†a⟩(t)=ηn0+N\langle a^\dagger a\rangle(t)=\eta n_0+N. A phase-invariant Gaussian channel is generated by a time-local master equation of the form of Eq. (7) [67]. Matching the evolution of the mean occupation and of the transmissivity gives the rates in Eq. (18). The populations then obey Eq. (55), and their temperature derivatives follow from ∂TN\partial_TN alone, because η\eta is temperature independent. We evaluate GG and fkf_k numerically. We discretize the reservoir into equally spaced modes with spacing δω\delta\omega below a cutoff ωc\omega_c, with ∣gk∣2=J(ωk)δω|g_k|^2=J(\omega_k)\delta\omega. We diagonalize the single-excitation Hamiltonian once, which gives G(t)G(t) and fk(t)f_k(t) at all times. The discretization is accurate for times shorter than the recurrence time 2π/δω2\pi/\delta\omega.

Switching envelope

With a time-dependent coupling λf(s)\lambda f(s), the integrated transition coefficients of the second-order theory become

K−(T,t)=λ2∫0∞dω J(ω)[nT(ω)+1]∣f~(ω)∣2,\mathcal{K}_{-}(T,t)=\lambda^2\int_0^\infty d\omega\,J(\omega)[n_T(\omega)+1]|\widetilde{f}(\omega)|^2,
K+(T,t)=λ2∫0∞dω J(ω)nT(ω)∣f~(ω)∣2,(70)\mathcal{K}_{+}(T,t)=\lambda^2\int_0^\infty d\omega\,J(\omega)n_T(\omega)|\widetilde{f}(\omega)|^2, \tag*{(70)}

with f~\widetilde{f} defined in Eq. (41). These expressions follow from first-order time-dependent perturbation theory for each exchange process. For f=1f=1 they reduce to Eq. (30), since ∣f~∣2=t2Φt|\widetilde{f}|^2=t^2\Phi_t. The response χ\chi follows by replacing J(nT+1)J(n_T+1) with WTW_T, and Eq. (28) applies unchanged. In the Markovian limit, ∣f~(ω)∣2→2πteffδ(ω−ω0)|\widetilde{f}(\omega)|^2\to2\pi t_{\mathrm{eff}}\delta(\omega-\omega_0) for envelopes that vary slowly on the scale 1/ω01/\omega_0.

For the envelope used in Sec. III E, the rising ramp contributes

A(Δ)=12[EΔ−12(EΔ+π/tr+EΔ−π/tr)],(71)A(\Delta)=\frac{1}{2}\left[E_\Delta-\frac{1}{2}\left(E_{\Delta+\pi/t_r}+E_{\Delta-\pi/t_r}\right)\right], \tag*{(71)}

where Ek=(eiktr−1)/(ik)E_k=(e^{ikt_r}-1)/(ik) and Δ=ω0−ω\Delta=\omega_0-\omega. The full transform is

f~(ω)=A(Δ)+eiΔtreiΔ(t−2tr)−1iΔ+eiΔtA∗(Δ).(72)\widetilde{f}(\omega)=A(\Delta)+e^{i\Delta t_r}\frac{e^{i\Delta(t-2t_r)}-1}{i\Delta}+e^{i\Delta t}A^*(\Delta). \tag*{(72)}

Each ramp contributes 3tr/83t_r/8 to tefft_{\mathrm{eff}}, so teff=t−5tr/4t_{\mathrm{eff}}=t-5t_r/4.

References

References

  1. [1]T. M. Stace, Quantum limits of thermometry, Phys. Rev. A 82, 011611 (2010).DOI
  2. [2]L. A. Correa, M. Mehboudi, G. Adesso, and A. Sanpera, Individual quantum probes for optimal thermometry, Phys. Rev. Lett. 114, 220405 (2015).pubmed.ncbi.nlm.nih.gov/26196607
  3. [3]M. Brunelli, S. Olivares, and M. G. A. Paris, Qubit thermometry for micromechanical resonators, Phys. Rev. A 84, 032105 (2011).DOI
  4. [4]M. Brunelli, S. Olivares, M. Paternostro, and M. G. A. Paris, Qubit-assisted thermometry of a quantum harmonic oscillator, Phys. Rev. A 86, 012125 (2012).DOI
  5. [5]D. M. Toyli, C. F. de las Casas, D. J. Christle, V. V. Dobrovitski, and D. D. Awschalom, Fluorescence thermometry enhanced by the quantum coherence of single spins in diamond, Proc. Natl. Acad. Sci. U.S.A. 110, 8417 (2013).DOI
  6. [6]G. Kucsko, P. C. Maurer, N. Y. Yao, M. Kubo, H. J. Noh, P. K. Lo, H. Park, and M. D. Lukin, Nanometre-scale thermometry in a living cell, Nature 500, 54 (2013).
  7. [7]P. Neumann, I. Jakobi, F. Dolde, C. Burk, R. Reuter, G. Waldherr, J. Honert, T. Wolf, A. Brunner, J. H. Shim, D. Suter, H. Sumiya, J. Isoya, and J. Wrachtrup, High-Precision Nanoscale Temperature Sensing Using Single Defects in Diamond, Nano Lett. 13, 2738 (2013).pubmed.ncbi.nlm.nih.gov/23721106
  8. [8]T. P. Purdy, K. E. Grutter, K. Srinivasan, and J. M. Taylor, Quantum correlations from a room-temperature optomechanical cavity, Science 356, 1265 (2017).DOI
  9. [9]C. W. Helstrom, Quantum detection and estimation theory, J. Stat. Phys. 1, 231 (1969).DOI
  10. [10]S. L. Braunstein and C. M. Caves, Statistical distance and the geometry of quantum states, Phys. Rev. Lett. 72, 3439 (1994).pubmed.ncbi.nlm.nih.gov/10056200
  11. [11]M. Mehboudi, M. Moreno-Cardoner, G. De Chiara, and A. Sanpera, Thermometry precision in strongly correlated ultracold lattice gases, New J. Phys. 17, 055020 (2015).DOI
  12. [12]M. G. A. Paris, Achieving the landau bound to precision of quantum thermometry in systems with vanishing gap, J. Phys. A: Math. Theor. 49, 03LT02 (2016).DOI
  13. [13]A. De Pasquale, D. Rossini, R. Fazio, and V. Giovannetti, Local quantum thermal susceptibility, Nat. Commun. 7, 12782 (2016).
  14. [14]H. J. D. Miller and J. Anders, Energy-temperature uncertainty relation in quantum thermodynamics, Nat. Commun. 9, 2203 (2018).
  15. [15]S. Campbell, M. G. Genoni, and S. Deffner, Precision thermometry and the quantum speed limit, Quantum Sci. Technol. 3, 025002 (2018).DOI
  16. [16]K. V. Hovhannisyan and L. A. Correa, Measuring the temperature of cold many-body quantum systems, Phys. Rev. B 98, 045101 (2018).DOI
  17. [17]P. P. Potts, J. B. Brask, and N. Brunner, Fundamental limits on low-temperature quantum thermometry with finite resolution, Quantum 3, 161 (2019).
  18. [18]M. R. Jørgensen, P. P. Potts, M. G. A. Paris, and J. B. Brask, Tight bound on finite-resolution quantum thermometry at low temperatures, Phys. Rev. Research 2, 033394 (2020).
  19. [19]K. V. Hovhannisyan, M. R. Jørgensen, G. T. Landi, A. M. Alhambra, J. B. Brask, and M. Perarnau-Llobet, Optimal quantum thermometry with coarse-grained measurements, PRX Quantum 2, 020322 (2021).
  20. [20]M. F. Cenni, L. Lami, A. Acín, and M. Mehboudi, Thermometry of Gaussian quantum systems using Gaussian measurements, Quantum 6, 743 (2022).DOI
  21. [21]L. Seveso and M. G. A. Paris, Trade-off between information and disturbance in qubit thermometry, Phys. Rev. A 97, 032129 (2018).DOI
  22. [22]W.-K. Mok, K. Bharti, L.-C. Kwek, and A. Bayat, Optimal probes for global quantum thermometry, Commun. Phys. 4, 62 (2021).DOI
  23. [23]J. Rubio, J. Anders, and L. A. Correa, Global Quantum Thermometry, Phys. Rev. Lett. 127, 190402 (2021).DOI
  24. [24]M. Mehboudi, M. R. Jørgensen, S. Seah, J. B. Brask, J. Kolodyński, and M. Perarnau-Llobet, Fundamental Limits in Bayesian Thermometry and Attainability via Adaptive Strategies, Phys. Rev. Lett. 128, 130502 (2022).DOI
  25. [25]J. Glatthard, J. Rubio, R. Sawant, T. Hewitt, G. Barontini, and L. A. Correa, Optimal Cold Atom Thermometry Using Adaptive Bayesian Strategies, PRX Quantum 3, 040330 (2022).DOI
  26. [26]J. Sjevic, D. Newman, T. Rudolph, and T. M. Stace, Single-qubit thermometry, Phys. Rev. A 91, 012331 (2015).
  27. [27]L. Mancino, M. Sbroscia, I. Gianani, E. Roccia, and M. Barbieri, Quantum Simulation of Single-Qubit Thermometry Using Linear Optics, Phys. Rev. Lett. 118, 130502 (2017).DOI
  28. [28]V. Cavina, L. Mancino, A. De Pasquale, I. Gianani, M. Sbroscia, R. I. Booth, E. Roccia, R. Raimondi, V. Giovannetti, and M. Barbieri, Bridging thermodynamics and metrology in nonequilibrium quantum thermometry, Phys. Rev. A 98, 050101(R) (2018).DOI
  29. [29]M. M. Feyles, L. Mancino, M. Sbroscia, I. Gianani, and M. Barbieri, Dynamical role of quantum signatures in quantum thermometry, Phys. Rev. A 99, 062114 (2019).DOI
  30. [30]A. H. Kilerich, A. De Pasquale, and V. Giovannetti, Dynamical approach to ancilla-assisted quantum thermometry, Phys. Rev. A 98, 042124 (2018).
  31. [31]A. Ullah, M. T. Naseem, and Ö. E. Müstecaplıoğlu, Low-temperature quantum thermometry boosted by coherence generation, Phys. Rev. Research 5, 043184 (2023).DOI
  32. [32]Y. Aiache, A. El Allati, and K. El Anouz, Harnessing coherence generation for precision single- and two-qubit quantum thermometry, Phys. Rev. A 110, 032605 (2024).DOI
  33. [33]E. Trombetti, M. Malitesta, M. Pezzutto, and S. Gherardini, Nonequilibrium thermometry via an ensemble of initially correlated qubits, Phys. Rev. A 113, 062208 (2026).
  34. [34]G. Planella, M. F. B. Cenni, A. Acín, and M. Mehboudi, Bath-induced correlations enhance thermometry precision at low temperatures, Phys. Rev. Lett. 128, 040502 (2022).DOI
  35. [35]S. Seah, S. Nimmrichter, D. Grimmer, J. P. Santos, V. Scarani, and G. T. Landi, Collisional quantum thermometry, Phys. Rev. Lett. 123, 180602 (2019).DOI
  36. [36]G. O. Alves, M. A. F. Santos, and G. T. Landi, Collisional thermometry for Gaussian systems, Phys. Rev. A 110, 052421 (2024).arxiv.org/abs/2405.12030
  37. [37]Y. Yang, V. Montenegro, and A. Bayat, Sequential-measurement thermometry with quantum many-body probes, Phys. Rev. Applied 22, 024069 (2024).DOI
  38. [38]P. Sekatski and M. Perarnau-Llobet, Optimal nonequilibrium thermometry in Markovian environments, Quantum 6, 869 (2022).DOI
  39. [39]N. Anto-Sztrikacs, H. J. D. Miller, A. Nazir, and D. Segal, Bypassing thermalization timescales in temperature estimation using prethermal probes, Phys. Rev. A 109, L060201 (2024).
  40. [40]Q. Bouton, J. Nettersheim, D. Adam, F. Schmidt, D. Mayer, T. Lausch, E. Tiesinga, and A. Widera, Single-Atom Quantum Probes for Ultracold Gases Boosted by Nonequilibrium Spin Dynamics, Phys. Rev. X 10, 011018 (2020).
  41. [41]L. Mancino, M. G. Genoni, M. Barbieri, and M. Paternostro, Nonequilibrium readiness and precision of gaussian quantum thermometers, Phys. Rev. Research 2, 033498 (2020).
  42. [42]S. Mirkhalaf, M. Mehboudi, Z. Nafari Qaleh, and S. Rahimi-Keshari, Operational significance of nonclassicality in nonequilibrium Gaussian quantum thermometry, New J. Phys. 26, 023046 (2024).arxiv.org/abs/2207.10742
  43. [43]A. Ullah, M. T. Naseem, and Ö. E. Müstecaplıoğlu, Optimal strategies for transient and equilibrium quantum thermometry using gaussian and non-gaussian probes, Phys. Rev. A 112, 062601 (2025).DOI
  44. [44]P. Chattopadhyay, Non-gaussian dissipative quantum thermometry beyond gaussian bounds, Phys. Rev. A 112, 062606 (2025).DOI
  45. [45]M. Winzek, M. Horodecki, and R. R. Rodríguez, Non-equilibrium quantum thermometry with bosonic samples (2026), arXiv:2606.27894 [quant-ph].
  46. [46]D. Hangleiter, M. T. Mitchison, T. H. Johnson, M. Bruderer, M. B. Plenio, and D. Jaksch, Nondestructive selective probing of phononic excitations in a cold Bose gas using impurities, Phys. Rev. A 91, 013611 (2015).DOI
  47. [47]M. Mehboudi, A. Lampo, C. Charalambous, L. A. Correa, M. Á. García-March, and M. Lewenstein, Using polarons for sub-nK quantum nondemolition thermometry in a Bose-Einstein condensate, Phys. Rev. Lett. 122, 030403 (2019).DOI
  48. [48]T. H. Johnson, F. Cosco, M. T. Mitchison, D. Jaksch, and S. R. Clark, Thermometry of ultracold atoms via nonequilibrium work distributions, Phys. Rev. A 93, 053619 (2016).DOI
  49. [49]M. T. Mitchison, T. Fogarty, G. Guarnieri, S. Campbell, T. Busch, and J. Goold, In situ thermometry of a cold fermi gas via dephasing impurities, Phys. Rev. Lett. 125, 080402 (2020).
  50. [50]S. Brattegard and M. T. Mitchison, Thermometry by correlated dephasing of impurities in a one-dimensional fermi gas, Phys. Rev. A 109, 023309 (2024).DOI
  51. [51]S. Razavian, C. Benedetti, M. Bina, Y. Akbari-Kourbolagh, and M. G. A. Paris, Quantum thermometry by single-qubit dephasing, Eur. Phys. J. Plus 134, 284 (2019).DOI
  52. [52]F. Gebbia, C. Benedetti, F. Benatti, R. Floreanini, M. Bina, and M. G. A. Paris, Two-qubit quantum probes for the temperature of an ohmic environment, Phys. Rev. A 101, 032112 (2020).DOI
  53. [53]L. A. Correa, M. P. Peraurnau-Llobet, K. V. Hovhannisyan, S. Hernández-Santana, M. Mehboudi, and A. Sanpera, Enhancement of low-temperature thermometry by strong coupling, Phys. Rev. A 96, 062103 (2017).DOI
  54. [54]R. Ravell Rodríguez, M. Mehboudi, M. Horodecki, and M. Peraurnau-Llobet, Strongly coupled fermionic probe for nonequilibrium thermometry, New J. Phys. 26, 013046 (2024).
  55. [55]Q.-S. Q. Tan, Y. Liu, X. Liu, H. Chen, X. Xiao, and W. Wu, Limitations of strong coupling in non-markovian quantum thermometry, Phys. Rev. A 112, 024612 (2025).DOI
  56. [56]N. Zhang, C. Chen, S.-Y. Bai, W. Wu, and J.-H. An, Non-markovian quantum thermometry, Phys. Rev. Applied 17, 034076 (2022).DOI
  57. [57]Y. Aiache, C. Seida, K. El Anouz, and A. El Allati, Non-markovian enhancement of nonequilibrium quantum thermometry, Phys. Rev. E 110, 024132 (2024).DOI
  58. [58]Y. Aiache, A. Ullah, Ö. E. Müstecaplıoğlu, and A. El-Allati, Quantum metrology of a structured reservoir, Phys. Rev. A 111, 062619 (2025).DOI
  59. [59]J. a. C. P. Porto, C. H. S. Vieira, I. G. da Paz, P. R. Diegue z, and L. S. Marinho, Temperature and non-markovian parameter estimation in quantum brownian motion, Phys. Rev. A 112, 042424 (2025).DOI
  60. [60]V. Mukherjee, A. Zwick, A. Ghosh, X. Chen, and G. Kurzki, Enhanced precision bound of low-temperature quantum thermometry via dynamical control, Commun. Phys. 2, 162 (2019).DOI
  61. [61]J. Glatthard and L. A. Correa, Bending the rules of low-temperature thermometry with periodic driving, Quantum 6, 705 (2022).DOI
  62. [62]H.-P. Breuer and F. Petruccione, The theory of open quantum systems (OUP Oxford, 2002).
  63. [63]J. Reichert, P. Nalbach, and M. Thorwart, Dynamics of a quantum two-state system in a linearly driven quantum bath, Phys. Rev. A 94, 023127 (2016).DOI
  64. [64]M. G. A. Paris, Quantum estimation for quantum technology, Int. J. Quantum Inf. 07, 125 (2009).
  65. [65]Q. A. Turchette, C. J. Myatt, B. E. King, C. A. Sackett, D. Kielpinski, W. M. Itano, C. Monroe, and D. J. Wineland, Decoherence and decay of motional quantum states of a trapped atom coupled to engineered reservoirs, Phys. Rev. A 62, 053807 (2000).DOI
  66. [66]M. Hofheinz, H. Wang, M. Ansmann, R. C. Bialczak, E. Lucero, M. Neeley, A. D. O’Connell, D. Sank, J. Wenner, J. M. Martinis, and A. N. Cleland, Synthesizing arbitrary quantum states in a superconducting resonator, Nature 459, 546 (2009).DOI
  67. [67]J.-H. An and W.-M. Zhang, Non-markovian entanglement dynamics of noisy continuous-variable quantum channels, Phys. Rev. A 76, 042127 (2007).DOI

Paper details

vlnc:W_2wg8hse6
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents