Reducing thermal noises by quantum refrigerators Han-Jia Bi, Sheng-Wen Li Abstract Reducing the thermal noises in microwave (MW) resonators can bring about significant progress in many research fields. In this study, we consider using three-level or four-level systems as “quantum refrigerators” to cool down MW resonators so as to reduce the thermal noises, and investigate their possible cooling limits. In such a quantum refrigerator system, the MW resonator is coupled with many three-level or four-level systems. Proper light pump makes the multilevel systems concentrated into their ground states, which continuously absorb the thermal photons in the MW resonator. By adiabatic elimination, we give a more precise description for this cooling process. For three level systems, though the laser driving can cool down the multilevel systems efficiently, a too strong driving strength also significantly perturbs their energy levels, breaking the resonant interaction between the atom and the resonator, which weakens the cooling effect, and that sets a finite region for cooling parameters. In four level systems, by adopting an indirect pumping approach, such a finite cooling region can be further released. In both cases, we obtain analytical results for the cooling limit of the MW resonator. Based on practical parameters, our estimation shows the cooling limit could reach lower than the liquid helium temperature, without resorting to the traditional cryogenic systems. I Introduction Microwave (MW) resonators are essential electronic devices widely used in many research areas, such as the signal radiation and detection in communication systems [ 1 ] , cosmology radio telescopes [ 2 ] , and electron/nuclear spin resonance spectrometers [ 3 , 4 ] . But at room temperature ( T\sim 300\,\text{K} ), the thermal noises in MW resonators are generally quite strong. For the instance of an MW resonator with \omega_{\text{{r}}}/2\pi=1\,\text{GHz} , the thermal photon number in the MW resonator is \sim 6\times 10^{3} . The signals weaker than this intensity level would be buried in the fluctuating noise background. Thus, generally a complicated cryogenic system is needed to cooled down the system to the liquid helium temperature ( T\simeq 4\,\text{K} , thermal photon number \apprle 10^{2} ) [ 5 , 6 , 7 , 8 ] . Alternatively, a bench-top cooling method by “quantum refrigerators” is adopted to cool down the MW resonators [ 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 ] , where the MW resonator is coupled with an ensemble of multilevel systems. By proper light pump [ 12 , 14 , 13 , 16 , 15 ] , or MW radiation [ 9 , 11 , 10 ] , the ensemble populations can be concentrated into the ground states, then the thermal photons in the MW resonator can be continuously absorbed by the ensemble, where the heat is dumped away through light radiation. It is reported that the liquid nitrogen temperature can be reached ( \sim 66\,\text{K} [ 16 ] ). However, because of the complicated noise environments in solid defects systems, some theoretical analysis also shows this is almost the cooling limit of such system [ 17 ] . To get a better cooling effect, some more improvements are still needed. It is worth noting that such systems are quite similar as the Scovil–Schulz-DuBois–Geusic (SSDG) quantum refrigerator [ 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 ] . Here, we make a thorough analysis on the possible cooling limit of the MW resonator, by adopting more simplified three-level or four-level structures as the quantum refrigerator. The lowest two atom levels are resonantly coupled with the resonator mode, and a driving laser is applied on the atom, which could make the atom population fully concentrated into the ground state, effectively giving a zero temperature. By adiabatic elimination [ 28 , 29 , 30 ] , we derive a master equation for the resonator mode, which gives a more precise description. We find that, for three level systems, the atom can be cooled down quite efficiently, but a too strong driving strength also brings in significant perturbation to the atom levels. Such perturbation could break the resonant interaction between the atom and the resonator, which prevents the heat transport from the MW resonator to the refrigerator, and weakens the cooling effect. That sets a finite region for cooling parameters. By utilizing an indirect pumping approach in a four-level refrigerator [ 25 , 26 , 31 ] , such a constraint condition can be released. In this case, the driving laser is applied on the upper two levels, thus no longer perturb the resonant coupling between the lower two levels and the resonator, and the heat can be indirectly absorbed away like the “siphonic” effect. In both cases, we obtain analytical results for the cooling limit of the MW resonator. Based on some practical parameters in current experiments, our estimation shows the cooling limit of the MW resonator could reach the liquid helium temperature. II Quantum refrigerator setup In analog to the SSDG quantum refrigerator [ 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 ] , here we consider using a three-level atom as a quantum refrigerator to cool down the MW resonator ( \hat{H}_{\text{{r}}}=\omega_{\text{{r}}}\,\hat{a}^{{\dagger}}\hat{a} ). The self Hamiltonian of the atom is described by \hat{H}_{\text{{a}}}=\sum_{\alpha}\,\varepsilon_{\alpha}|\alpha\rangle\langle\alpha| [ \alpha=\mathsf{a,b,e} , see Fig. 1 (a)]. The energy gap ( \Omega_{\mathsf{ab}}:=\varepsilon_{\mathsf{a}}-\varepsilon_{\mathsf{b}} ) between the lowest two levels |\mathsf{a}\rangle , |\mathsf{b}\rangle is in resonance with the MW resonator \omega_{\text{{r}}} , and their interaction is described by \hat{V}_{\text{{ar}}}=g\,(\hat{\sigma}^{+}\hat{a}+\hat{\sigma}^{-}\hat{a}^{{\dagger}}) . Here, we denote 1 1 1 In this paper, generally we denote \hat{\tau}_{\alpha\beta}^{+}:=|\alpha\rangle\langle\beta|:=(\hat{\tau}_{\alpha\beta}^{-})^{{\dagger}} as the transition operator between |\alpha\rangle , |\beta\rangle (for \varepsilon_{\alpha}>\varepsilon_{\beta} ), and \Omega_{\alpha\beta}:=\varepsilon_{\alpha}-\varepsilon_{\beta} as the energy gap. The MW transition operator \hat{\sigma}^{\pm} is equivalent with \hat{\tau}_{\mathsf{ab}}^{\pm} . \hat{\sigma}^{+}:=|\mathsf{a}\rangle\langle\mathsf{b}|=(\hat{\sigma}^{-})^{{\dagger}} , and g is the atom-resonator coupling strength. The energy gaps \Omega_{\mathsf{ea}} , \Omega_{\mathsf{eb}} are in the optical regime. In addition, a driving laser with frequency \omega_{L}=\Omega_{\mathsf{ea}} is resonantly applied to the transition |\mathsf{e}\rangle\leftrightarrow|\mathsf{a}\rangle , which is described by \hat{V}_{\text{d}}(t)=\tilde{\Omega}_{\text{d}}\,(\hat{\tau}_{\mathsf{ea}}^{+}+\hat{\tau}_{\mathsf{ea}}^{-})\cos\omega_{L}t\simeq\frac{1}{2}\tilde{\Omega}_{\text{d}}\,(\hat{\tau}_{\mathsf{ea}}^{+}e^{-i\omega_{L}t}+\mathrm{H.c.}) . Here \tilde{\Omega}_{\text{d}} is the Rabi frequency, which characterizes the driving light intensity. Figure 1: Demonstration for the interaction between a MW resonator mode and (a) a three-level atom, (b) a four-level atom. Here \omega_{L} is the frequency of the driving laser, and \omega_{\text{{r}}} is the frequency of the MW mode. The transition pathways are indicated by the dashed lines. In the three-level atom, the laser driving could bring in perturbation to |\mathsf{a}\rangle , which could break the exchanging resonance between |\mathsf{a,b}\rangle and the MW resonator. The dynamics of the composite atom-resonator state \rho(t) is described by the following master equation (interaction picture) \displaystyle\partial_{t}\rho \displaystyle=i[\rho,\,\tilde{V}_{\text{d}}+\tilde{V}_{\text{{ar}}}]+\mathcal{D}_{\text{{a}}}[\rho]+\mathcal{D}_{\text{dep}}[\rho]+\mathcal{D}_{\text{{r}}}[\rho], (1) \displaystyle\mathcal{D}_{\text{{r}}}[\rho] \displaystyle=\kappa\bar{\mathsf{n}}_{\text{{r}}}\big(\hat{a}^{{\dagger}}\rho\hat{a}-\tfrac{1}{2}\{\hat{a}\hat{a}^{{\dagger}},\,\rho\}\big) \displaystyle+\kappa(\bar{\mathsf{n}}_{\text{{r}}}+1)\big(\hat{a}\rho\hat{a}^{{\dagger}}-\tfrac{1}{2}\{\hat{a}^{{\dagger}}\hat{a},\,\rho\}\big), \displaystyle\mathcal{D}_{\text{{a}}}[\rho] \displaystyle=\sum_{\alpha,\beta}^{\varepsilon_{\alpha}>\varepsilon_{\beta}}\Gamma_{\alpha\beta}^{+}\big(\hat{\tau}_{\alpha\beta}^{+}\rho\hat{\tau}_{\alpha\beta}^{-}-\tfrac{1}{2}\{\hat{\tau}_{\alpha\beta}^{-}\hat{\tau}_{\alpha\beta}^{+},\rho\}\big) \displaystyle+\Gamma_{\alpha\beta}^{-}\big(\hat{\tau}_{\alpha\beta}^{-}\rho\hat{\tau}_{\alpha\beta}^{+}-\tfrac{1}{2}\{\hat{\tau}_{\alpha\beta}^{+}\hat{\tau}_{\alpha\beta}^{-},\rho\}\big), \displaystyle\mathcal{D}_{\text{dep}}[\rho] \displaystyle=\sum_{\alpha}\tilde{\gamma}_{\text{p}}^{\alpha}\big(\hat{\text{{n}}}_{\alpha}\rho\hat{\text{{n}}}_{\alpha}-\tfrac{1}{2}\{\hat{\text{{n}}}_{\alpha},\rho\}\big). Here, \mathcal{D}_{\text{{r}}}[\rho] and \mathcal{D}_{\text{{a}}}[\rho] describe the dissipation effect of the MW mode and the atom respectively. We denote \kappa as the resonator decay rate, \Gamma_{\alpha\beta}^{+}:=\gamma_{\alpha\beta}\bar{\mathsf{n}}_{\alpha\beta} , \Gamma_{\alpha\beta}^{-}:=\gamma_{\alpha\beta}(\bar{\mathsf{n}}_{\alpha\beta}+1) as the dissipation rates of the atom, where \gamma_{\alpha\beta} are the spontaneous decay rates, and \bar{\mathsf{n}}_{\text{{r}}}:=(e^{\omega_{\text{{r}}}/T}-1)^{-1} , \bar{\mathsf{n}}_{\alpha\beta}:=(e^{\Omega_{\alpha\beta}/T}-1)^{-1} are the Planck functions with temperature T . These dissipation rates satisfy the Boltzmann ratio \Gamma_{\alpha\beta}^{+}/\Gamma_{\alpha\beta}^{-}=e^{-\Omega_{\alpha\beta}/T} . In the atom dissipation term \mathcal{D}_{\text{{a}}}[\rho] we only consider the optical transitions |\mathsf{e}\rangle\leftrightarrow|\mathsf{a}\rangle , |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle [the dashed paths in Fig. 1 (a)]. Since \Omega_{\mathsf{ab}} is in the MW regime, generally the spontaneous decay rate between |\mathsf{a}\rangle , |\mathsf{b}\rangle is negligibly small. In addition, the pure dephasing effects of the atom levels are also taken into account, which is described by \mathcal{D}_{\text{dep}}[\rho] (here, \hat{\text{{n}}}_{\alpha}:=|\alpha\rangle\langle\alpha| , and \tilde{\gamma}_{\text{p}}^{\alpha} are the pure dephasing rates). Such a three level system can be regarded as an SSDG quantum refrigerator [ 18 , 19 , 25 , 27 ] . A driving laser is applied on |\mathsf{a}\rangle\leftrightarrow|\mathsf{e}\rangle , making the population on |\mathsf{a}\rangle greatly reduced and approach zero, and then the “heat” is dumped away through the optical emission |\mathsf{e}\rangle\rightarrow|\mathsf{b}\rangle . Effectively, that makes the three-level atom become a system with zero temperature, which could absorb “heat” from the MW resonator. By this way the thermal photons in the MW resonator can be continuously reduced by the quantum refrigerator. III Photon dynamics of the MW mode To give a more precise description for the above cooling process, we need a dynamical equation solely for the resonator state \varrho_{\text{{r}}}\equiv\mathrm{tr}_{\text{{a}}}\rho . Generally speaking, the atom could achieve its steady state much faster than the resonator mode. That enables us to apply the adiabatic elimination [ 28 , 29 , 30 ] , which finally gives an equation for the MW mode alone, that is (see Appendix A), \displaystyle\partial_{t}\varrho_{\text{{r}}}= \displaystyle(A_{+}+\kappa\bar{\mathsf{n}}_{\text{{r}}})\big(\hat{a}^{\dagger}\varrho_{\text{{r}}}\hat{a}-\tfrac{1}{2}\{\varrho_{\text{{r}}},\,\hat{a}^{\dagger}\hat{a}\}\big) \displaystyle+[A_{-}+\kappa(\bar{\mathsf{n}}_{\text{{r}}}+1)]\big(\hat{a}\varrho_{\text{{r}}}\hat{a}^{{\dagger}}-\tfrac{1}{2}\{\varrho_{\text{{r}}},\,\hat{a}\hat{a}^{\dagger}\}\big). (2) Here, A_{-(+)} can be regarded as the cooling (heating) rate induced by the atom, and they are given by A_{\pm}:=2g^{2}\,\mathrm{Re}\int_{0}^{\infty}ds\,\langle\hat{\sigma}^{\pm}(s)\hat{\sigma}^{\mp}(0)\rangle_{\text{ss}}. (3) Here \langle\hat{\sigma}^{\pm}(s)\hat{\sigma}^{\mp}(0)\rangle_{\text{ss}} is the time correlation function of the atom in the steady state when it is not coupled with the MW resonator, which is described by ( \varrho_{\text{{a}}}\equiv\mathrm{tr}_{\text{{r}}}\rho is the atom state) \partial_{t}\varrho_{\text{{a}}}=i[\varrho_{\text{{a}}},\,\tilde{V}_{\text{d}}]+\mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}]+\mathcal{D}_{\text{dep}}[\varrho_{\text{{a}}}]. (4) In the steady state t\rightarrow\infty , the above resonator equation ( 2 ) gives the MW photon number as \langle\hat{n}\rangle_{\text{ss}}=\frac{A_{+}+\kappa\bar{\mathsf{n}}_{\text{{r}}}}{A_{-}-A_{+}+\kappa}. (5) If the cooling rate is fast enough ( A_{-}\gg A_{+},\,\kappa\bar{\mathsf{n}}_{\text{{r}}} ), we obtain \langle\hat{n}\rangle_{\text{ss}}\rightarrow 0 , which means the thermal noise in the MW resonator is greatly suppressed. The time correlation functions \langle\hat{\sigma}^{\pm}(s)\hat{\sigma}^{\mp}(0)\rangle_{\text{ss}} in A_{\pm} [Eq. ( 3 )] can be calculated with the help of the quantum regression theorem from the atom equation ( 4 ) [ 29 , 30 , 32 , 33 ] . For the three level system [Fig. 1 (a)], that gives the heating and cooling rates ( 3 ) as (the full results are presented in Appendix B) \displaystyle A_{+} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\mathrm{Re}\big[\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}+\frac{i\tilde{\Omega}_{\text{d}}}{2\tilde{\Upsilon}_{\mathsf{eb}}}\langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle_{\text{ss}}\big], \displaystyle A_{-} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}. (6) Here, \tilde{\Upsilon}_{\mathsf{ab}}:=\frac{1}{2}(\Gamma_{\mathsf{ea}}^{+}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) and \tilde{\Upsilon}_{\mathsf{eb}}:=\frac{1}{2}(\Gamma_{\mathsf{ea}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) are the total dephasing rates for the oscillations between the levels |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle and |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle respectively 2 2 2 When there is no the driving field ( \tilde{\Omega}_{\text{d}}=0 ), the atom equation ( 4 ) gives \partial_{t}\langle\hat{\tau}_{\mathsf{eb}}^{-}\rangle=-\tilde{\Upsilon}_{\mathsf{eb}}\langle\hat{\tau}_{\mathsf{eb}}^{-}\rangle , and \partial_{t}\langle\hat{\sigma}^{-}\rangle=-\tilde{\Upsilon}_{\mathsf{ab}}\langle\hat{\sigma}^{-}\rangle (see also the derivations in Appendix B). . \langle\hat{\text{{n}}}_{\mathsf{a(b)}}\rangle_{\text{ss}} is the steady state population on |\mathsf{a(b)}\rangle , \langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle_{\text{ss}} is the coherence term, which can be solved by the atom equation ( 4 ), and they give \displaystyle\frac{\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}} \displaystyle=\frac{\Gamma_{\mathsf{eb}}^{+}}{\Gamma_{\mathsf{eb}}^{-}}\cdot\frac{\Gamma_{\mathsf{ea}}^{-}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}}{\Gamma_{\mathsf{ea}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}},\quad\frac{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}}=e^{-\Omega_{\mathsf{eb}}/T}, \displaystyle\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle_{\text{ss}} \displaystyle=\frac{i\tilde{\Omega}_{\text{d}}}{2\tilde{\Upsilon}_{\mathsf{ea}}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle\big). (7) When the driving strength \tilde{\Omega}_{\text{d}} is strong enough, the above ratios give \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\rightarrow e^{-\Omega_{\mathsf{eb}}/T}\simeq 0 (for the optical frequency \Omega_{\mathsf{eb}}\gg T\simeq 300\,\text{K} ), namely, \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\simeq 1 and \langle\hat{\text{{n}}}_{\mathsf{a,e}}\rangle_{\text{ss}}\simeq\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle_{\text{ss}}\simeq 0 (see details in Appendix B). That means, the population are fully concentrated in the ground state |\mathsf{b}\rangle . As mentioned above, such a population distribution effectively gives a zero temperature T_{\mathsf{ab}}^{\text{eff}}\rightarrow 0 . But it is also worth noting that the driving strength \tilde{\Omega}_{\text{d}} also appears in the correction factor 2g^{2}/(\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}) in the cooling/heating rates A_{\pm} [Eq. ( 6 )]. As a result, when the driving strength \tilde{\Omega}_{\text{d}} is too large, both the cooling and heating rates decrease towards zero A_{\pm}\rightarrow 0 , and that weakens the cooling effect. The reason can be understood by the following picture. For the two levels |\mathsf{e}\rangle and |\mathsf{a}\rangle under the laser driving, effectively the driving field also perturbs these two levels, which makes them shift upwards and downwards. As a result, the energy gap of the lowest two levels would also be changed correspondingly, and that makes the energy gap \Omega_{\mathsf{ab}} no longer exactly resonant with the resonator frequency \omega_{\text{{r}}} . Because of such an off-resonant coupling, the energy cannot be efficiently transported from the MW resonator to the atom refrigerator, and thus the cooling and heating rates are both weakened. In Fig. 2 (a), we show the steady state photon number \langle\hat{n}\rangle_{\text{ss}} in the MW resonator changing with the driving strength \tilde{\Omega}_{\text{d}} . With the increase of the driving strength \tilde{\Omega}_{\text{d}} , the photon number \langle\hat{n}\rangle_{\text{ss}} first decreases rapidly, and finally increases again, which roots from the correction factor in the cooling rate ( 6 ) discussed above. Thus, the driving strength \tilde{\Omega}_{\text{d}} should be controlled within a finite working region so as to reach the optimum cooling effect. Now we make an estimation of the cooling limit of \langle\hat{n}\rangle_{\text{ss}} within the working region. From Fig. 2 (a), we can see indeed a small driving strength \tilde{\Omega}_{\text{d}} is enough to cool down the atoms to their ground states, which gives \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}\simeq\langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle_{\text{ss}}\simeq 0 , and \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\simeq 1 ; meanwhile, such a driving strength is still not large enough to bring in significant change to the correction factor 2g^{2}/(\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}) in A_{\pm} , thus that gives A_{-}\simeq 2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}} , A_{+}\simeq 0 . Therefore, in the working region, the cooling limit of \langle\hat{n}\rangle_{\text{ss}} is estimated as [from Eq. ( 5 )] \text{min}\langle\hat{n}\rangle_{\text{ss}}\simeq\frac{\kappa\bar{\mathsf{n}}_{\text{{r}}}}{2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}}+\kappa}\simeq\bar{\mathsf{n}}_{\text{{r}}}\big/\frac{2g^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}}. (8) The upper constraint for the working region of \tilde{\Omega}_{\text{d}} can be roughly estimated from the denominator form of the correction factor [Eq. ( 6 )], which gives significant decrease when \tilde{\Upsilon}_{\mathsf{ab}}\simeq\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}} , and that gives \tilde{\Omega}_{\text{d}}\simeq 2\sqrt{\tilde{\Upsilon}_{\mathsf{eb}}\tilde{\Upsilon}_{\mathsf{ab}}} . It turns out such a upper constraint estimation well fits the numerical result [see the position of the increasing part in Fig. 2 (a)]. Figure 2: The steady photon number \langle\hat{n}\rangle_{\text{ss}} [Eq. ( 5 )] changing with the driving strength \tilde{\Omega}_{\text{d}} for the (a) three-level and (b) four-level system. The parameters are set as: resonator frequency \omega_{\text{{r}}}/2\pi=1\,\text{GHz} , atom-resonator coupling strength g_{N}/2\pi=1.5\,\text{MHz} , total dephasing rate for |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle \tilde{\Upsilon}_{\mathsf{ab}}=1\,\text{MHz} , spontaneous decay rate \gamma_{\alpha\beta}\equiv\gamma=1\,\text{GHz} (for all the optical transitions), initial photon number \bar{\mathsf{n}}_{\text{{r}}}=6200 (corresponding to temperature T\simeq 300\,\text{K} ). The frequencies of the optical transitions are set as \Omega_{\mathsf{ea}}/2\pi=400\,\text{THz} (for the three-level system), \Omega_{\mathsf{em}}/2\pi=300\,\text{THz} and \Omega_{\mathsf{ma}}/2\pi=400\,\text{THz} (for the four-level system). The numerical minimums in these two cases are both \text{min}\langle\hat{n}\rangle_{\text{ss}}\simeq 68.1 . IV Four-level refrigerator Now we further consider using a four-level system as the refrigerator, and adopting an indirect pumping approach [ 25 , 26 , 31 ] . In this setup [Fig. 1 (b)], still the lowest two levels |\mathsf{a,b}\rangle are resonantly coupled with the MW resonator, while a mediated level |\mathsf{m}\rangle is added between |\mathsf{e}\rangle and |\mathsf{a}\rangle (assuming \Omega_{\mathsf{em}},\,\Omega_{\mathsf{ma}} are in the optical frequency regime), and now the driving laser is applied to the transition |\mathsf{e}\rangle\leftrightarrow|\mathsf{m}\rangle , which would no longer perturb |\mathsf{a,b}\rangle directly. Similarly as the above three-level case, only the optical transitions |\mathsf{e}\rangle\leftrightarrow|\mathsf{m}\rangle , |\mathsf{m}\rangle\leftrightarrow|\mathsf{a}\rangle , and |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle are considered [the dashed paths in Fig. 1 (b)], while the transition between |\mathsf{a}\rangle and |\mathsf{b}\rangle is neglected. The dephasing effects of these levels are also taken into account. In this case, the driving laser moves the population from |\mathsf{m}\rangle to |\mathsf{e}\rangle , and the “heat” could be dumped out through the optical emission |\mathsf{e}\rangle\rightarrow|\mathsf{b}\rangle ; meanwhile, the population decrease in |\mathsf{m}\rangle would be complemented from |\mathsf{a}\rangle through the thermal excitation |\mathsf{m}\rangle\leftrightarrow|\mathsf{a}\rangle , until their populations satisfy the Boltzmann distribution. This is similar as a thermally “siphonic” effect. As a result, under a strong enough driving intensity, the atom population could be fully concentrated into the ground state |\mathsf{b}\rangle , meanwhile, the energy gap \Omega_{\mathsf{ab}} would no longer be perturbed and still keep resonant with the MW resonator ( \Omega_{\mathsf{ab}}=\omega_{\text{{r}}} ). Intuitively, the population in |\mathsf{m}\rangle is almost zero, which makes the driving laser seem being applied on “nothing”. But in a finite temperature T , there still remains a nonzero population, though quite small. It turns out this is enough to achieve the above “siphonic” cooling process. For this four-level system, we still apply the adiabatic elimination to derive the equation for the MW mode alone, and it turns out to have the same form as the above resonator equation ( 2 ) for the three-level case. And the heating and cooling rates A_{\pm}^{\prime} are still defined by Eq. ( 3 ), except now the correlation functions \langle\hat{\sigma}^{\pm}(s)\hat{\sigma}^{\mp}(0)\rangle_{\text{ss}} in A_{\pm}^{\prime} [Eq. ( 3 )] should be calculated from the four-level system under the laser driving. With the help of the quantum regression theorem, for this four-level setup, the heating and cooling rates are obtained as (the full results are presented in Appendix C) A_{+}^{\prime}=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}},\qquad A_{-}^{\prime}=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}. (9) Here, \tilde{\Upsilon}_{\mathsf{ab}}^{\prime}:=\frac{1}{2}(\Gamma_{\mathsf{eb}}^{+}+\Gamma_{\mathsf{ma}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) is the total dephasing rate for the oscillation between the levels |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle (see from the atom equation \partial_{t}\langle\hat{\sigma}^{-}\rangle , similarly as the above three level case). In the steady state, the population ratios in this four-level system satisfy (see Appendix C) \displaystyle\frac{\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}} \displaystyle=e^{-\Omega_{\mathsf{ma}}/T},\hskip 10.00002pt\frac{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}}=e^{-\Omega_{\mathsf{eb}}/T}, \displaystyle\frac{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle_{\text{ss}}} \displaystyle=\frac{\Gamma_{\mathsf{em}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}{\Gamma_{\mathsf{em}}^{-}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}. (10) When the driving strength \tilde{\Omega}_{\text{d}}\rightarrow\infty , the ratios ( 10 ) give \langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle_{\text{ss}}\simeq 1 , and \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\simeq e^{-(\Omega_{\mathsf{eb}}-\Omega_{\mathsf{ma}})/T}\rightarrow 0 . That indicates the populations could be fully concentrated into the ground state, i.e., \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\rightarrow 1 , \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}\rightarrow 0 , which also gives T_{\mathsf{ab}}^{\text{eff}}\rightarrow 0 . Unlike the above three-level case [Eq. ( 6 )], now the driving strength \tilde{\Omega}_{\text{d}} no longer appears in the correction factor 2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}}^{\prime} in A_{\pm}^{\prime} [Eq. ( 9 )]. Thus, with the increase of the driving light intensity, here the cooling (heating) rate increases (decreases) monotonically. Therefore, when the driving strength \tilde{\Omega}_{\text{d}}\rightarrow\infty , the cooling performance could achieve the optimum, and that gives A_{+}^{\prime}\rightarrow 0 and A_{-}^{\prime}\rightarrow 2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}}^{\prime} . In Fig. 2 (b), we see the steady photon number \langle\hat{n}\rangle_{\text{ss}} in the MW resonator decreases monotonically with the increase of the driving strength \tilde{\Omega}_{\text{d}} , and now there is no upper constraint for \tilde{\Omega}_{\text{d}} . Based on the above results, for this four level system, the cooling limit of the photon number in the MW resonator [Eq. ( 5 )] is obtained as \displaystyle\langle\hat{n}\rangle_{\text{ss}} \displaystyle= \displaystyle\frac{\frac{2g^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle+\bar{\mathsf{n}}_{\text{{r}}}}{\frac{2g^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{b}}-\hat{\text{{n}}}_{\mathsf{a}}\rangle+1} (11) \displaystyle\stackrel{{\scriptstyle\tilde{\Omega}_{\text{d}}\rightarrow\infty}}{{\longrightarrow}} \displaystyle\frac{\kappa\bar{\mathsf{n}}_{\text{{r}}}}{2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}+\kappa}\simeq\bar{\mathsf{n}}_{\text{{r}}}\big/\frac{2g^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}, which just has the same form as the above three level case [Eq. ( 8 )]. Here, \tilde{\Upsilon}_{\mathsf{ab}}^{\prime} is the total dephasing rate for the oscillation between the MW levels |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle , which can be measured from the practical decay time of the atom coherence. From its definition under Eq. ( 9 ), \Gamma_{\mathsf{eb}}^{+}\equiv\gamma_{\mathsf{eb}}\,\bar{\mathsf{n}}_{\mathsf{eb}}\simeq 0 , \Gamma_{\mathsf{ma}}^{+}\equiv\gamma_{\mathsf{ma}}\,\bar{\mathsf{n}}_{\mathsf{ma}}\simeq 0 (since \Omega_{\mathsf{eb}},\Omega_{\mathsf{ma}}\gg T ), we obtain \tilde{\Upsilon}_{\mathsf{ab}}^{\prime}\simeq\frac{1}{2}(\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) . All the above discussions are based on the interaction between the MW resonator and a single multi-level system. Generally, the coupling strength g between an MW resonator mode and a single atom is quite small. This can be improved by adopting N atom refrigerators to couple with the resonator. Correspondingly, the above cooling and heating rates A_{\pm} obtained from a single refrigerator can be enlarged by N times. Effectively, this also can be regarded as the enlargement in the coupling strength, g\mapsto g_{N}\equiv\sqrt{N}\,g , which is similar as the treatment in lasing problems [ 29 , 33 , 32 ] . Therefore, to achieve a better cooling effect, we need Ng^{2}/\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}\gg 1 . V Experiment estimation Now we make an estimation for the possible cooling limit [Eq. ( 11 )] in realistic experiments. For the example of an MW resonator with \omega_{\text{{r}}}/2\pi=1\,\text{GHz} , at room temperature T=300\,\text{K} , the thermal photon number from the surrounding reservoir is \bar{\mathsf{n}}_{\text{{r}}}\simeq 6.2\times 10^{3} . A quality factor Q\simeq 10^{4} is an achievable estimation, which gives the resonator decay rate as \kappa\simeq 0.1\,\text{MHz} [ 34 , 9 , 16 ] . The multi-level systems can be implemented by the defect structures in solid crystals (e.g., NV or SiV centers in diamonds or silicon carbide [ 35 , 14 , 36 ] , pentacene molecules doped in the p -terphenyl crystal [ 9 , 8 ] ). It is reported that the effective coupling strength between the MW resonator and NV ensemble could achieve g_{N}/2\pi\simeq 1.5\,\text{MHz} [ 16 ] . The typical dephasing time of the MW levels is around T_{2}^{*}\simeq 2\,\mu\text{s} [ 37 , 38 , 34 ] , and that gives the total dephasing rate as \tilde{\Upsilon}_{\mathsf{ab}}\simeq 0.5\,\text{MHz} . Based on the above experimental parameters, the cooling limit ( 11 ) gives the steady photon number in the MW resonator as \langle\hat{n}\rangle_{\text{ss}}\simeq 68.9 , which well fits the numerical results in Fig. 2 . That corresponds to an effective temperature T_{\text{{r}}}^{\text{eff}}\equiv\hbar\omega_{\text{{r}}}\big/k_{\text{{b}}}\ln(1+1/\langle\hat{n}\rangle_{\text{ss}})\simeq 3.3\,\text{K} (starting from room temperature). Such a cooling effect is well comparable with the liquid helium temperature. Alternatively, another more specific platform is to use atom gas as the multilevel systems. For the instance of {}^{23}\mathrm{Na} atom, the interaction with the nuclear makes the ground state 3\text{S}_{1/2} splitted into two hyperfine levels, whose energy gap is \Delta\nu\simeq 1.77\,\text{GHz} . Both these two states can be excited to the state 3\text{P}_{1/2} via electric dipole transition, thus such a structure well satisfies the above three level refrigerator. The lowest two atom levels can be coupled with the MW resonator via the magnetic dipole transition interaction. The transition magnetic dipole moment of these two hyperfine levels is mainly contributed from the electron spin, which is similar as the situation of the NV center. Thus, the transition magnetic dipole moment of these two atom levels has the similar magnitude as the above NV center. Therefore, the effective coupling strength between an ensemble of {}^{23}\mathrm{Na} atoms and the MW resonator could achieve at least the similar strength as the NV ensemble [ 16 ] . Moreover, indeed generally the volume of the atom gas is much larger than the diamond crystal, thus the total number N of the atoms could be much larger than that of NV centers. Further, it is worth noting that the cooling limit ( 8 ) does not depend on the dephasing rate of the two optical transitions, but only on that of the MW transition. For dilute atom gas, such a dephasing rate \tilde{\Upsilon}_{\mathsf{ab}} is mainly contributed from the Doppler broadening. For {}^{23}\mathrm{Na} atom gas at room temperature ( \sim 300\,\text{K} ), the Doppler broadening for the MW transition ( \Delta\nu\simeq 1.77\,\text{GHz} ) between these two hyperfine levels, the Doppler broadening 3 3 3 The Doppler broadening here is estimated by \Delta\nu_{\text{D}}\simeq(\nu_{0}/c)\sqrt{8\ln 2\,k_{\text{{b}}}T/m} , where the central frequency is taken as \nu_{0}=1.77\,\text{GHz} . is just \sim 4.6\,\text{kHz} , which gives a much smaller dephasing rate \tilde{\Upsilon}_{\mathsf{ab}} than the above solid defect system. Based on these conditions, the photon number in MW resonator could be cooled down even below the single photon level, without resorting to traditional cryogenic systems. Figure 3: Demonstration for atom gas in MW resonator. For the instance of {}^{23}\mathrm{Na} atom, the interaction with the nuclear makes the ground state 3\text{S}_{1/2} splitted into two hyperfine levels ( \Delta\nu\simeq 1.77\,\text{GHz} ). Together with the optical transitions to the excited state, such a system well satisfies the structure of our three level refrigerator. VI Discussions In this paper, we consider using three-level or four-level atoms as quantum refrigerators to cool down an MW resonator, and investigate the possible cooling limits. Under proper transition structures, a laser pump drives the atom to work as an SSDG quantum refrigerator, and the atom population are fully concentrated into the ground state, effectively giving a zero temperature. Then the thermal photons in the MW resonator can be continuously absorbed away by the atom. By adopting the adiabatic elimination, we obtain a master equation for the resonator mode, which gives a more precise description for this cooling system. We find that, for three level systems, though the atom can be cooled down quite efficiently, a too strong driving strength also brings in significant perturbation to the atom levels, which causes the atom-resonator interaction to become off-resonant, and that weakens the cooling effect. By utilizing an indirect pumping approach in a four-level refrigerator, such a constraint condition can be released. We obtain analytical results for the cooling limits in both cases, which share the same form. Based on some practical parameters, our estimation shows the cooling limit of the MW resonator could reach the liquid helium temperature. Our results highlight the potential of quantum refrigerators as practical, high-performance solutions for suppressing thermal noise in MW devices without cryogenic complexity [ 39 , 40 , 41 , 23 , 24 ] . Acknowledgments - SWL appreciates quite much for the helpful discussion with H. Wu, B. Zhang, and D. Xu in BIT. This study is supported by NSF of China (Grant No. 12475030). Appendix A The master equation for the MW resonator Here we show the derivation of the master equation which describes the resonator mode alone. In the interaction picture (applied by \hat{H}_{\text{{a}}}+\hat{H}_{\text{{r}}} ), we rewrite the equation for the atom-resonator system as \partial_{t}\rho=(\mathcal{L}_{\text{{a}}}+\mathcal{K}_{\text{{ar}}}+\mathcal{D}_{\text{{r}}})[\rho] , where \displaystyle\mathcal{L}_{\text{{a}}}[\rho] \displaystyle=i[\rho,\,\tilde{V}_{\text{d}}]+\mathcal{D}_{\text{{a}}}[\rho]+\mathcal{D}_{\text{dep}}[\rho], \displaystyle\mathcal{K}_{\text{{ar}}}[\rho] \displaystyle=i[\rho,\,\tilde{V}_{\text{{ar}}}]=i\,[\rho,\,g(\hat{\sigma}^{+}\hat{a}+\hat{\sigma}^{-}\hat{a}^{{\dagger}})]. (12) Here \mathcal{D}_{\text{{r}}}[\rho] describes the dissipation of the resonator mode, \mathcal{L}_{\text{{a}}}[\rho] describes the atom dissipation together with the laser driving \tilde{V}_{\text{d}} , and \mathcal{K}_{\text{{ar}}}[\rho] describes the interaction between the atom and the resonator. Here we need a dynamical equation solely for the resonator state \varrho_{\text{{r}}}=\mathrm{tr}_{\text{{a}}}\rho . Generally, the dissipation rate of the MW resonator is much slower than that of the atom, and \mathcal{D}_{\text{{r}}}[\rho] simply gives \mathrm{tr}_{\text{{a}}}\big\{\mathcal{D}_{\text{{r}}}[\rho]\big\}=\mathcal{D}_{\text{{r}}}[\varrho_{\text{{r}}}] , therefore, in the following discussions we first omit this term and then take it back in the final step. The master equation \partial_{t}\rho=(\mathcal{L}_{\text{{a}}}+\mathcal{K}_{\text{{ar}}})[\rho] has a similar linear structure to the Schrödinger equation, thus we treat \mathcal{K}_{\text{{ar}}} as a perturbation based on \mathcal{L}_{\text{{a}}} . The steady states of \mathcal{L}_{\text{{a}}}[\rho]=0 form a degenerated subspace, i.e., \mathcal{L}_{\text{{a}}}\left[\ |n\rangle\langle n|\otimes\varrho_{\text{{a}}}^{\text{ss}}\ \right]=0 , where |n\rangle are the Fock states for the MW mode, and \varrho_{\text{{a}}}^{\text{ss}} is the steady state of the atom ( \mathcal{L}_{\text{{a}}}[\varrho_{\text{{a}}}^{\text{ss}}]=0 ). Thus, the degenerated perturbation can be applied based on this subspace [ 42 , 43 ] . Here we introduce some projection operators \mathcal{P}[\rho]:=\mathcal{P}_{\text{{r}}}\cdot\mathcal{P}_{\text{{a}}}[\rho] , where \displaystyle\mathcal{P}_{\text{{r}}}[\rho] \displaystyle:=\sum_{n=0}^{\infty}\langle n|\rho|n\rangle\ |n\rangle\langle n|,\qquad\mathcal{P}_{\text{{a}}}[\rho]:=\lim_{t\rightarrow\infty}e^{t\mathcal{L}_{\text{{a}}}}[\rho]=\varrho_{\text{{a}}}^{\text{ss}}\otimes\mathrm{tr}_{\text{{a}}}[\rho]. (13) Such a projection gives \mathcal{P}[\rho(t)]=\varrho_{\text{{a}}}^{\text{ss}}\otimes\mu(t) , where \mu(t)\equiv\sum p_{n}(t)\,|n\rangle\langle n| is the diagonal part of the resonator state \varrho_{\text{{r}}}(t) . Then effectively the above master equation can be described by [ 42 , 43 , 28 , 30 ] \displaystyle\partial_{t}\mathcal{P}[\rho] \displaystyle=\mathcal{P}\mathcal{K}_{\text{{ar}}}(-\mathcal{L}_{\text{{a}}})^{-1}\mathcal{K}_{\text{{ar}}}\mathcal{P}\,[\rho], \displaystyle\Rightarrow\quad\partial_{t}\mu \displaystyle=\mathcal{P}_{\text{{r}}}\ \mathrm{tr}_{\text{{a}}}\Big\{\mathcal{P}_{\text{{a}}}\ \mathcal{K}_{\text{{ar}}}(-\mathcal{L}_{\text{{a}}})^{-1}\mathcal{K}_{\text{{ar}}}\mathcal{P}[\rho]\Big\}=\mathcal{P}_{\text{{r}}}\int_{0}^{\infty}ds\,\mathrm{tr}_{\text{{a}}}\Big\{\mathcal{K}_{\text{{ar}}}\,e^{s\mathcal{L}_{\text{{a}}}}\mathcal{K}_{\text{{ar}}}\,\mathcal{P}[\rho(t)]\Big\}. (14) In the last equation, the super operator \mathcal{L}_{\text{{a}}}^{-1} is formally replaced by its Laplace integral, and the integration term has been simplified by \mathrm{tr}_{\text{{a}}}\big\{\mathcal{P}_{\text{{a}}}\ \mathcal{L}[\rho]\big\}=\mathrm{tr}_{\text{{a}}}\big\{\,\varrho_{\text{{a}}}^{\text{ss}}\otimes\mathrm{tr}_{\text{{a}}}(\mathcal{L}[\rho])\,\big\}=\mathrm{tr}_{\text{{a}}}\big\{\mathcal{L}[\rho]\big\} . Further, denoting \Theta_{s}:=e^{s\mathcal{L}_{\text{{a}}}}\mathcal{K}_{\text{{ar}}}\mathcal{P}[\rho] for short, the above integration term becomes \mathrm{tr}_{\text{{a}}}\big\{\mathcal{K}_{\text{{ar}}}[\Theta_{s}]\big\}=ig\ [\,\mathrm{tr}_{\text{{a}}}(\hat{\sigma}^{+}\Theta_{s}),\,\hat{a}\,]+ig\ [\,\mathrm{tr}_{\text{{a}}}(\hat{\sigma}^{-}\Theta_{s}),\,\hat{a}^{{\dagger}}\,] (using the relation \mathrm{tr}_{\text{{a}}}\big\{\,[\Theta_{s},\,\hat{X}_{\text{{a}}}\cdot\hat{Y}_{\text{{r}}}]\,\big\}=[\,\mathrm{tr}_{\text{{a}}}(\hat{X}_{\text{{a}}}\Theta_{s}),\,\hat{Y}_{\text{{r}}}\,] ). That further gives \displaystyle\mathrm{tr}_{\text{{a}}}(\hat{\sigma}^{+}\Theta_{s}) \displaystyle=\mathrm{tr}_{\text{{a}}}\big\{\hat{\sigma}^{+}\cdot e^{s\mathcal{L}_{\text{{a}}}}\ \mathcal{K}_{\text{{ar}}}[\mathcal{P}\rho]\big\}=ig\ \mathrm{tr}_{\text{{a}}}\Big\{\hat{\sigma}^{+}(s)\cdot[\varrho_{\text{{a}}}^{\text{ss}}\cdot\mu_{t},\,\hat{\sigma}^{+}\hat{a}+\hat{\sigma}^{-}\hat{a}^{{\dagger}}]\Big\} \displaystyle=ig\ \Big\{\langle\hat{\sigma}^{+}(0)\hat{\sigma}^{+}(s)\rangle_{\text{ss}}\>\mu_{t}\hat{a}+\langle\hat{\sigma}^{-}(0)\hat{\sigma}^{+}(s)\rangle_{\text{ss}}\>\mu_{t}\hat{a}^{{\dagger}}-\langle\hat{\sigma}^{+}(s)\hat{\sigma}^{+}(0)\rangle_{\text{ss}}\>\hat{a}\mu_{t}-\langle\hat{\sigma}^{+}(s)\hat{\sigma}^{-}(0)\rangle_{\text{ss}}\>\hat{a}^{{\dagger}}\mu_{t}\Big\}, (15) \displaystyle\mathrm{tr}_{\text{{a}}}(\hat{\sigma}^{-}\Theta_{s}) \displaystyle=ig\ \Big\{\langle\hat{\sigma}^{+}(0)\hat{\sigma}^{-}(s)\rangle_{\text{ss}}\>\mu_{t}\hat{a}+\langle\hat{\sigma}^{-}(0)\hat{\sigma}^{-}(s)\rangle_{\text{ss}}\>\mu_{t}\hat{a}^{{\dagger}}-\langle\hat{\sigma}^{-}(s)\hat{\sigma}^{+}(0)\rangle_{\text{ss}}\>\hat{a}\mu_{t}-\langle\hat{\sigma}^{-}(s)\hat{\sigma}^{-}(0)\rangle_{\text{ss}}\>\hat{a}^{{\dagger}}\mu_{t}\Big\}. Taking these results back to the above Eq. ( 14 ), we obtain \displaystyle\partial_{t}\mu= \displaystyle ig\,\mathcal{P}_{\text{{r}}}\int_{0}^{\infty}ds\ \Big\{[\mathrm{tr}_{\text{a}}(\hat{\sigma}^{+}\Theta_{s}),\,\hat{a}]+[\mathrm{tr}_{\text{a}}(\hat{\sigma}^{-}\Theta_{s}),\,\hat{a}^{{\dagger}}]\Big\} \displaystyle= \displaystyle-g^{2}\int_{0}^{\infty}ds\ \Big\{\langle\hat{\sigma}^{-}(0)\hat{\sigma}^{+}(s)\rangle_{\text{ss}}\,[\mu_{t}\hat{a}^{{\dagger}},\,\hat{a}]-\langle\hat{\sigma}^{+}(s)\hat{\sigma}^{-}(0)\rangle_{\text{ss}}\>[\hat{a}^{{\dagger}}\mu_{t},\,\hat{a}] \displaystyle\quad\qquad\qquad+\langle\hat{\sigma}^{+}(0)\hat{\sigma}^{-}(s)\rangle_{\text{ss}}\>[\mu_{t}\hat{a},\,\hat{a}^{{\dagger}}]-\langle\hat{\sigma}^{-}(s)\hat{\sigma}^{+}(0)\rangle_{\text{ss}}\>[\hat{a}\mu_{t},\,\hat{a}^{{\dagger}}]\Big\}, (16) where the double annihilation/creation terms are dropped due to the projection operation \mathcal{P}_{\text{{r}}} . Since \left(\langle\hat{\sigma}^{\pm}(0)\hat{\sigma}^{\mp}(s)\rangle_{\text{ss}}\right)^{*}=\langle\hat{\sigma}^{\pm}(s)\hat{\sigma}^{\mp}(0)\rangle_{\text{ss}} , the above equation also can be written as \displaystyle\partial_{t}\mu \displaystyle=\frac{1}{2}A_{+}\big([\hat{a}^{{\dagger}}\mu,\hat{a}]+\mathrm{H.c.}\big)+\frac{1}{2}A_{-}\big([\hat{a}\mu,\hat{a}^{{\dagger}}]+\mathrm{H.c.}\big), \displaystyle A_{+} \displaystyle:=2g^{2}\,\mathrm{Re}\int_{0}^{\infty}ds\,\langle\hat{\sigma}^{+}(s)\hat{\sigma}^{-}(0)\rangle_{\text{ss}},\qquad A_{-}:=2g^{2}\,\mathrm{Re}\int_{0}^{\infty}ds\,\langle\hat{\sigma}^{-}(s)\hat{\sigma}^{+}(0)\rangle_{\text{ss}}. (17) Now taking back the original dissipation term \mathcal{D}_{\text{{r}}}[\rho] , we obtain the master equation solely for the MW resonator, i.e., \displaystyle\partial_{t}\mu \displaystyle=\Gamma_{+}\big(\hat{a}^{\dagger}\mu\hat{a}-\frac{1}{2}\{\mu,\,\hat{a}^{\dagger}\hat{a}\}\big)+\Gamma_{-}\big(\hat{a}\mu\hat{a}^{{\dagger}}-\frac{1}{2}\{\mu,\,\hat{a}\hat{a}^{\dagger}\}\big), (18) \displaystyle\Gamma_{+} \displaystyle=A_{+}+\kappa\bar{\mathsf{n}}_{\text{{r}}},\qquad\Gamma_{-}=A_{-}+\kappa(\bar{\mathsf{n}}_{\text{{r}}}+1). Here \Gamma_{+\,(-)} indicates the increasing (decreasing) rate of the photon number of the resonator, and A_{+\,(-)} can be regarded as the heating (cooling) rate induced by the atom. In the steady state t\rightarrow\infty , that gives the MW photon number as \langle\hat{n}\rangle_{\text{ss}}=\frac{\Gamma_{+}}{\Gamma_{-}-\Gamma_{+}}=\frac{A_{+}+\kappa\bar{\mathsf{n}}_{\text{{r}}}}{A_{-}-A_{+}+\kappa}. (19) If the cooling rate is fast enough ( A_{-}\gg A_{+},\kappa\bar{\mathsf{n}}_{\text{{r}}} ), the MW photon number becomes \langle\hat{n}\rangle_{\text{ss}}\rightarrow 0 , which means the thermal noise in the resonator can be greatly suppressed. The above derivations for the cooling and heating rates are based on the interaction between the MW mode and a single atom. A large number of N atoms can be placed in the MW resonator and used to absorb the thermal photons together. In this case, the cooling and heating rates can be enlarged by N times ( A_{\pm}\mapsto NA_{\pm} ), or equivalently, the atom-resonator coupling strength g can be regarded as enlarged by \sqrt{N} times ( g\mapsto g_{N}:=\sqrt{N}\,g ). Appendix B The three level system under driving B.1 Steady state expectations Here we study the behavior of the three level system when there is no interaction with the resonator. A driving laser is applied to the transition path |\mathsf{e}\rangle\leftrightarrow|\mathsf{a}\rangle , and the atom dynamics is described by the master equation (interaction picture) \displaystyle\partial_{t}\varrho_{\text{{a}}} \displaystyle=i[\varrho_{\text{{a}}},\,\frac{1}{2}\tilde{\Omega}_{\text{d}}(\hat{\tau}_{\mathsf{ea}}^{+}+\hat{\tau}_{\mathsf{ea}}^{-})]+\mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}]+\mathcal{D}_{\text{dep}}[\varrho_{\text{{a}}}], (20) \displaystyle\mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}] \displaystyle=\sum_{\alpha,\beta}^{\varepsilon_{\alpha}>\varepsilon_{\beta}}\Gamma_{\alpha\beta}^{+}\big(\hat{\tau}_{\alpha\beta}^{+}\varrho_{\text{{a}}}\hat{\tau}_{\alpha\beta}^{-}-\frac{1}{2}\{\hat{\tau}_{\alpha\beta}^{-}\hat{\tau}_{\alpha\beta}^{+},\,\varrho_{\text{{a}}}\}\big)+\Gamma_{\alpha\beta}^{-}\big(\hat{\tau}_{\alpha\beta}^{-}\varrho_{\text{{a}}}\hat{\tau}_{\alpha\beta}^{+}-\frac{1}{2}\{\hat{\tau}_{\alpha\beta}^{+}\hat{\tau}_{\alpha\beta}^{-},\,\varrho_{\text{{a}}}\}\big), \displaystyle\mathcal{D}_{\text{dep}}[\varrho_{\text{{a}}}] \displaystyle=\sum_{\alpha}\tilde{\gamma}_{\text{p}}^{\alpha}\big(\hat{\text{{n}}}_{\alpha}\varrho_{\text{{a}}}\hat{\text{{n}}}_{\alpha}-\tfrac{1}{2}\{\hat{\text{{n}}}_{\alpha},\varrho_{\text{{a}}}\}\big). The transition structure of \mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}] is demonstrated in Fig. 1(a) in the main text (for |\mathsf{e}\rangle\leftrightarrow|\mathsf{a}\rangle and |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle ). For the transition |\alpha\rangle\leftrightarrow|\beta\rangle ( \varepsilon_{\alpha}>\varepsilon_{\beta} ), we denote \hat{\tau}_{\alpha\beta}^{+}:=|\alpha\rangle\langle\beta|=(\hat{\tau}_{\alpha\beta}^{-})^{{\dagger}} as the transition operators; \Gamma_{\alpha\beta}^{+}=\gamma_{\alpha\beta}\bar{\mathsf{n}}_{\alpha\beta} , \Gamma_{\alpha\beta}^{-}=\gamma_{\alpha\beta}(\bar{\mathsf{n}}_{\alpha\beta}+1) are the dissipation rates, with \bar{\mathsf{n}}_{\alpha\beta}\equiv(e^{\Omega_{\alpha\beta}/T}-1)^{-1} and \Omega_{\alpha\beta}\equiv\varepsilon_{\alpha}-\varepsilon_{\beta} , which satisfy \Gamma_{\alpha\beta}^{+}/\Gamma_{\alpha\beta}^{-}=e^{-\Omega_{\alpha\beta}/T} . The energy gap between |\mathsf{a}\rangle and |\mathsf{b}\rangle is in the MW regime, thus the spontaneous decay rate between these two levels is generally negligibly small. And \mathcal{D}_{\text{dep}}[\varrho_{\text{{a}}}] describes the pure dephasing effects of the atom levels, where \hat{\text{{n}}}_{\alpha}:=|\alpha\rangle\langle\alpha| , and \tilde{\gamma}_{\text{p}}^{\alpha} are the pure dephasing rates. From the above master equation ( 20 ) it turns out the equations of \langle\hat{\tau}_{\mathsf{ea}}^{\pm}\rangle , \langle\hat{\text{{n}}}_{\mathsf{e,a,b}}\rangle ( \hat{\text{{n}}}_{\alpha}:=|\alpha\rangle\langle\alpha| ) form a closed set, i.e., \displaystyle\partial_{t}\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle \displaystyle=+\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle\big)-\frac{1}{2}(\Gamma_{\mathsf{ea}}^{+}+\Gamma_{\mathsf{ea}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}})\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle:=\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle\big)-\tilde{\Upsilon}_{\mathsf{ea}}\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle, \displaystyle\partial_{t}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle \displaystyle=\big[\Gamma_{\mathsf{eb}}^{+}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle-\Gamma_{\mathsf{eb}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle\big]+\big[\Gamma_{\mathsf{ea}}^{+}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle-\Gamma_{\mathsf{ea}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle\big]+\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle-\langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle\big), (21) \displaystyle\partial_{t}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle \displaystyle=-\big[\Gamma_{\mathsf{eb}}^{+}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle-\Gamma_{\mathsf{eb}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle\big]. Here we denote \tilde{\Upsilon}_{\mathsf{ea}}:=\frac{1}{2}(\Gamma_{\mathsf{ea}}^{+}+\Gamma_{\mathsf{ea}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}) . The above equation \partial_{t}\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle indicates \tilde{\Upsilon}_{\mathsf{ea}} is just the total dephasing rate for the oscillation between the levels |\mathsf{e}\rangle and |\mathsf{a}\rangle , which sums up the contributions from \Gamma_{\alpha\beta}^{\pm} and \tilde{\gamma}_{\text{p}}^{\alpha} . In the steady state, the time derivatives all give zero, and their steady states give \frac{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}}=\frac{\Gamma_{\mathsf{eb}}^{+}}{\Gamma_{\mathsf{eb}}^{-}}=e^{-\Omega_{\mathsf{eb}}/T},\qquad\frac{\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}}=\frac{\Gamma_{\mathsf{eb}}^{+}}{\Gamma_{\mathsf{eb}}^{-}}\cdot\frac{\Gamma_{\mathsf{ea}}^{-}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}}{\Gamma_{\mathsf{ea}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}},\qquad\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle_{\text{ss}}=\frac{i\tilde{\Omega}_{\text{d}}}{2\tilde{\Upsilon}_{\mathsf{ea}}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}-\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}\big). (22) Together with the condition \langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}+\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}+\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}=1 , the above steady state values can be well obtained. When there is no driving light ( \tilde{\Omega}_{\text{d}}\rightarrow 0 ), the population ratio \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}=\Gamma_{\mathsf{eb}}^{+}\Gamma_{\mathsf{ea}}^{-}/\Gamma_{\mathsf{eb}}^{-}\Gamma_{\mathsf{ea}}^{+}=e^{-\omega_{\text{{r}}}/T} well returns the Boltzmann distribution. When the driving strength is quite strong ( \tilde{\Omega}_{\text{d}}\rightarrow\infty ), this population ratio becomes \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\rightarrow e^{-\Omega_{\mathsf{eb}}/T} . That means, under room temperature ( T\simeq 300\,\text{K} ), if \Omega_{\mathsf{eb}} is in the optical frequency regime ( \Omega_{\mathsf{eb}}/T\gg 1 ), we have \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\simeq 1 and \langle\hat{\text{{n}}}_{\mathsf{b,e}}\rangle_{\text{ss}}\simeq\langle\hat{\tau}_{\mathsf{ea}}^{-}\rangle_{\text{ss}}\simeq 0 [see Eq. ( 22 )], which means the atom population is fully concentrated in the ground state |\mathsf{b}\rangle . B.2 Time correlation functions To calculate the cooling and heating rates A_{\pm} from the correlation functions [Eq. ( 17 )], we need to study the equations of \langle\hat{\sigma}^{\pm}(t)\rangle [denoting \hat{\sigma}^{+}:=|\mathsf{a}\rangle\langle\mathsf{b}|=(\hat{\sigma}^{-})^{{\dagger}} ], and that gives \displaystyle\partial_{t}\langle\hat{\sigma}^{+}\rangle \displaystyle=\frac{i}{2}\tilde{\Omega}_{\text{d}}\langle\hat{\tau}_{\mathsf{eb}}^{+}\rangle-\frac{1}{2}(\Gamma_{\mathsf{ea}}^{+}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}})\langle\hat{\sigma}^{+}\rangle:=\frac{i}{2}\tilde{\Omega}_{\text{d}}\langle\hat{\tau}_{\mathsf{eb}}^{+}\rangle-\tilde{\Upsilon}_{\mathsf{ab}}\langle\hat{\sigma}^{+}\rangle, \displaystyle\partial_{t}\langle\hat{\tau}_{\mathsf{eb}}^{+}\rangle \displaystyle=\frac{i}{2}\tilde{\Omega}_{\text{d}}\langle\hat{\sigma}^{+}\rangle-\frac{1}{2}(\Gamma_{\mathsf{ea}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}})\langle\hat{\tau}_{\mathsf{eb}}^{+}\rangle:=\frac{i}{2}\tilde{\Omega}_{\text{d}}\langle\hat{\sigma}^{+}\rangle-\tilde{\Upsilon}_{\mathsf{eb}}\langle\hat{\tau}_{\mathsf{eb}}^{+}\rangle. (23) Here we denote \tilde{\Upsilon}_{\mathsf{ab}}:=\frac{1}{2}(\Gamma_{\mathsf{ea}}^{+}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) and \tilde{\Upsilon}_{\mathsf{eb}}:=\frac{1}{2}(\Gamma_{\mathsf{ea}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\Gamma_{\mathsf{eb}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) . These two equations indicate that \tilde{\Upsilon}_{\mathsf{ab}} and \tilde{\Upsilon}_{\mathsf{eb}} are just the total dephasing rates for the oscillations |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle and |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle respectively. Denoting \mathbf{v}_{t}:=\big(\,\langle\hat{\sigma}_{\mathsf{ab}}^{+}(t)\rangle,\,\langle\hat{\tau}_{\mathsf{eb}}^{+}(t)\rangle\,\big)^{T} , these two equations also can be written as \partial_{t}\mathbf{v}_{t}=\mathbf{G}\cdot\mathbf{v}_{t},\qquad\mathbf{G}:=\left[\begin{array}[]{cc}-\tilde{\Upsilon}_{\mathsf{ab}}&i\tilde{\Omega}_{\text{d}}/2\\ i\tilde{\Omega}_{\text{d}}/2&-\tilde{\Upsilon}_{\mathsf{eb}}\end{array}\right]. (24) Then the correlation function \langle\hat{\sigma}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}} can be calculated with the help of the quantum regression theorem [ 29 , 32 , 33 ] . Denoting \mathbf{V}_{t}:=\big(\,\langle\hat{\sigma}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}},\,\langle\hat{\tau}_{\mathsf{eb}}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}}\,\big)^{T} , which satisfies \mathbf{V}_{t\rightarrow\infty}=(0,0)^{T} , the quantum regression theorem states that \mathbf{V}_{t} has the same equation form as that of \mathbf{v}_{t} [Eq. ( 23 )], i.e., \partial_{t}\mathbf{V}_{t}=\mathbf{G}\cdot\mathbf{V}_{t} . Thus, the correlation function \langle\hat{\sigma}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}} can be obtained as the first component of \mathbf{V}_{t}=e^{\mathbf{G}t}\cdot\mathbf{V}_{0} , where \mathbf{V}_{0}=\big(\,\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}},\,\langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle_{\text{ss}}\,\big)^{T} . Then the time integration of \langle\hat{\sigma}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}} can be directly obtained as the first component of \int_{0}^{\infty}dt\,\mathbf{V}_{t}=\int_{0}^{\infty}dt\,e^{\mathbf{G}t}\cdot\mathbf{V}_{0}=-\mathbf{G}^{-1}\cdot\mathbf{V}_{0}. (25) As a result, the heating rate [Eq. ( 17 )] is obtained as \displaystyle A_{+} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\mathrm{Re}\big[\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}+\frac{i\tilde{\Omega}_{\text{d}}}{2\tilde{\Upsilon}_{\mathsf{eb}}}\langle\hat{\tau}_{\mathsf{ea}}^{+}\rangle_{\text{ss}}\big] \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\cdot\frac{\Gamma_{\mathsf{ea}}^{-}+(1-\gamma_{\mathsf{ea}}/2\tilde{\Upsilon}_{\mathsf{eb}})\,\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}}{(1+e^{\frac{\Omega_{\mathsf{ea}}}{T}}+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\Gamma_{\mathsf{ea}}^{+}+(2+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\,\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}}. (26) Similarly, the correlation function \langle\hat{\sigma}^{-}(t)\hat{\sigma}^{+}(0)\rangle_{\text{ss}} is calculated in the same way, where the above vectors \mathbf{V}_{t,\,0} and matrix \mathbf{G} should be changed to be \mathbf{V}_{t}:=\big(\,\langle\hat{\sigma}_{\mathsf{ab}}^{-}(t)\hat{\sigma}^{+}(0)\rangle_{\text{ss}},\,\langle\hat{\tau}_{\mathsf{eb}}^{-}(t)\hat{\sigma}^{+}(0)\rangle_{\text{ss}}\,\big)^{T},\qquad\mathbf{V}_{0}=\big(\,\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}},\,0\,\big)^{T},\qquad\mathbf{G}=\left[\begin{array}[]{cc}-\tilde{\Upsilon}_{\mathsf{ab}}&-i\tilde{\Omega}_{\text{d}}/2\\ -i\tilde{\Omega}_{\text{d}}/2&-\tilde{\Upsilon}_{\mathsf{eb}}\end{array}\right]. (27) And that gives the cooling rate [Eq. ( 17 )] as \displaystyle A_{-} \displaystyle=\frac{2g^{2}}{\bar{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}}\cdot\frac{e^{\frac{\Omega_{\mathsf{eb}}}{T}}(\Gamma_{\mathsf{ea}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}})}{(1+e^{\frac{\Omega_{\mathsf{ea}}}{T}}+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\Gamma_{\mathsf{ea}}^{+}+(2+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\,\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{ea}}}. (28) When there is no driving light ( \tilde{\Omega}_{\text{d}}\rightarrow 0 ), the cooling and heating rates give A_{-}/A_{+}=\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}=e^{\omega_{\text{{r}}}/T} , which naturally returns to the Boltzmann ratio, and that indicates the cooling and heating effect to the MW resonator is the same with the contribution of the surrounding bath with temperature T , which keeps the photon number in the resonator as \langle\hat{n}\rangle_{\text{ss}}=\bar{\mathsf{n}}_{\text{{r}}} . With the increase of the driving strength, the cooling rate A_{-} firstly increases, but then decreases towards zero due to the correction factor 2g^{2}/(\tilde{\Upsilon}_{\mathsf{ab}}+\tilde{\Omega}_{\text{d}}^{2}/4\tilde{\Upsilon}_{\mathsf{eb}}) , and that weakens the cooling effect. Appendix C The four level system under driving Here we study the behavior of the four level system when there is no interaction with the resonator. A driving laser is applied to the transition path |\mathsf{e}\rangle\leftrightarrow|\mathsf{m}\rangle , and the atom dynamics is described by the master equation (interaction picture) \partial_{t}\varrho_{\text{{a}}}=i[\varrho_{\text{{a}}},\,\frac{1}{2}\tilde{\Omega}_{\text{d}}(\hat{\tau}_{\mathsf{em}}^{+}+\hat{\tau}_{\mathsf{em}}^{-})]+\mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}]+\mathcal{D}_{\text{dep}}[\varrho_{\text{{a}}}]. (29) \mathcal{D}_{\text{{a}}}[\varrho_{\text{{a}}}] describes the transitions for |\mathsf{e}\rangle\leftrightarrow|\mathsf{b}\rangle , |\mathsf{e}\rangle\leftrightarrow|\mathsf{m}\rangle and |\mathsf{m}\rangle\leftrightarrow|\mathsf{a}\rangle [see Fig. 1(b) in the main text]. Then we obtain the equations of \langle\hat{\tau}_{\mathsf{em}}^{\pm}\rangle , \langle\hat{\text{{n}}}_{\mathsf{e,m,a,b}}\rangle , i.e., \displaystyle\partial_{t}\langle\hat{\tau}_{\mathsf{em}}^{-}\rangle \displaystyle=+\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle\big)-\frac{1}{2}(\Gamma_{\mathsf{em}}^{+}+\Gamma_{\mathsf{em}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\Gamma_{\mathsf{ma}}^{-}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{m}})\langle\hat{\tau}_{\mathsf{em}}^{-}\rangle:=\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle\big)-\tilde{\Upsilon}_{\mathsf{em}}^{\prime}\,\langle\hat{\tau}_{\mathsf{em}}^{-}\rangle, \displaystyle\partial_{t}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle \displaystyle=\big[\Gamma_{\mathsf{eb}}^{+}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle-\Gamma_{\mathsf{eb}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle\big]+\big[\Gamma_{\mathsf{em}}^{+}\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle-\Gamma_{\mathsf{em}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle\big]+\frac{i}{2}\tilde{\Omega}_{\text{d}}\big(\langle\hat{\tau}_{\mathsf{em}}^{-}\rangle-\langle\hat{\tau}_{\mathsf{em}}^{+}\rangle\big), \displaystyle\partial_{t}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle \displaystyle=\Gamma_{\mathsf{ma}}^{-}\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle-\Gamma_{\mathsf{ma}}^{+}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle, \displaystyle\partial_{t}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle \displaystyle=\Gamma_{\mathsf{eb}}^{-}\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle-\Gamma_{\mathsf{eb}}^{+}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle. (30) Here, we denote \tilde{\Upsilon}_{\mathsf{em}}^{\prime}=\frac{1}{2}(\Gamma_{\mathsf{em}}^{+}+\Gamma_{\mathsf{em}}^{-}+\Gamma_{\mathsf{eb}}^{-}+\Gamma_{\mathsf{ma}}^{-}+\tilde{\gamma}_{\text{p}}^{\mathsf{e}}+\tilde{\gamma}_{\text{p}}^{\mathsf{m}}) , and the above equation \partial_{t}\langle\hat{\tau}_{\mathsf{em}}^{-}\rangle indicates \tilde{\Upsilon}_{\mathsf{em}}^{\prime} is just the total dephasing rate for the oscillation |\mathsf{e}\rangle\leftrightarrow|\mathsf{m}\rangle . In the steady state, their steady state values give \frac{\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle_{\text{ss}}}=\frac{\Gamma_{\mathsf{ma}}^{-}}{\Gamma_{\mathsf{ma}}^{+}}=e^{\Omega_{\mathsf{ma}}/T},\qquad\frac{\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}=\frac{\Gamma_{\mathsf{eb}}^{-}}{\Gamma_{\mathsf{eb}}^{+}}=e^{\Omega_{\mathsf{eb}}/T},\qquad\frac{\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle_{\text{ss}}}{\langle\hat{\text{{n}}}_{\mathsf{e}}\rangle_{\text{ss}}}=\frac{\Gamma_{\mathsf{em}}^{-}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime}}{\Gamma_{\mathsf{em}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime}}. (31) Together with \langle\hat{\text{{n}}}_{\mathsf{e}}\rangle+\langle\hat{\text{{n}}}_{\mathsf{m}}\rangle+\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle+\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle=1 , their specific values can be obtained. When \tilde{\Omega}_{\text{d}}\rightarrow\infty , we have \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\rightarrow e^{(\Omega_{\mathsf{ma}}-\Omega_{\mathsf{eb}})/T} . Thus, if \Omega_{\mathsf{em}} is in the optical frequency regime, under room temperature, the population ratio \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}} is almost zero, namely, \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\simeq 1 , \langle\hat{\text{{n}}}_{\mathsf{a,m,e}}\rangle_{\text{ss}}\simeq 0 . To calculate the cooling and heating rates A_{\pm} [Eq. ( 17 )], we need the equations of \langle\hat{\sigma}^{\pm}(t)\rangle , i.e., \partial_{t}\langle\hat{\sigma}^{+}\rangle=-\frac{1}{2}(\Gamma_{\mathsf{eb}}^{+}+\Gamma_{\mathsf{ma}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}})\langle\hat{\sigma}^{+}\rangle:=-\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}\langle\hat{\sigma}^{+}\rangle. (32) Clearly, here \tilde{\Upsilon}_{\mathsf{ab}}^{\prime}:=\frac{1}{2}(\Gamma_{\mathsf{eb}}^{+}+\Gamma_{\mathsf{ma}}^{+}+\tilde{\gamma}_{\text{p}}^{\mathsf{a}}+\tilde{\gamma}_{\text{p}}^{\mathsf{b}}) is the total dephasing rate for the oscillation |\mathsf{a}\rangle\leftrightarrow|\mathsf{b}\rangle . It is worth noting that, unlike the three level system situation [Eq. ( 23 )], here the equation of \langle\hat{\sigma}^{+}\rangle is no longer coupled with the other dynamical variables. According to the quantum regression theorem, the correlation function \langle\hat{\sigma}^{+}(t)\hat{\sigma}^{-}(0)\rangle_{\text{ss}} follows the same equation form as that of \langle\hat{\sigma}^{+}(t)\rangle [Eq. ( 32 )]. As a result, similarly as the discussions around Eq. ( 25 ), here the heating and cooling rates are obtained as \displaystyle A_{+}^{\prime} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\frac{e^{\frac{\Omega_{\mathsf{ma}}}{T}}(\Gamma_{\mathsf{em}}^{-}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime})}{(1+e^{\frac{\Omega_{\mathsf{ma}}}{T}})\Gamma_{\mathsf{em}}^{-}+(1+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\Gamma_{\mathsf{em}}^{+}+(2+e^{\frac{\Omega_{\mathsf{eb}}}{T}}+e^{\frac{\Omega_{\mathsf{ma}}}{T}})\,\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime}}, \displaystyle A_{-}^{\prime} \displaystyle=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}=\frac{2g^{2}}{\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}\frac{e^{\frac{\Omega_{\mathsf{eb}}}{T}}(\Gamma_{\mathsf{em}}^{+}+\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime})}{(1+e^{\frac{\Omega_{\mathsf{ma}}}{T}})\Gamma_{\mathsf{em}}^{-}+(1+e^{\frac{\Omega_{\mathsf{eb}}}{T}})\Gamma_{\mathsf{em}}^{+}+(2+e^{\frac{\Omega_{\mathsf{eb}}}{T}}+e^{\frac{\Omega_{\mathsf{ma}}}{T}})\,\tilde{\Omega}_{\text{d}}^{2}/2\tilde{\Upsilon}_{\mathsf{em}}^{\prime}}. (33) Here the driving strength \tilde{\Omega}_{\text{d}} no longer appears in the correction factor 2g^{2}/\tilde{\Upsilon}_{\mathsf{ab}}^{\prime} as the three level system situation [Eqs. ( 26 , 28 )]. Therefore, with the increase of the driving light intensity, the cooling (heating) rate here increases (decreases) monotonically. When there is no driving light, the ratio between the cooling and heating rates gives A_{-}^{\prime}/A_{+}^{\prime}=\langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}/\langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}=e^{\omega_{\text{{r}}}/T} [see from Eq. ( 31 )], which naturally returns to the Boltzmann ratio. When the driving strength \tilde{\Omega}_{\text{d}}\rightarrow\infty , the populations could be fully concentrated in the ground state, i.e., \langle\hat{\text{{n}}}_{\mathsf{b}}\rangle_{\text{ss}}\rightarrow 1 , \langle\hat{\text{{n}}}_{\mathsf{a}}\rangle_{\text{ss}}\rightarrow 0 . In this case, the steady state photon number becomes \langle\hat{n}\rangle_{\text{ss}}=\frac{A_{+}^{\prime}+\kappa\bar{\mathsf{n}}_{\text{{r}}}}{A_{-}^{\prime}-A_{+}^{\prime}+\kappa}\stackrel{{\scriptstyle\tilde{\Omega}_{\text{d}}\rightarrow\infty}}{{\longrightarrow}}\bar{\mathsf{n}}_{\text{{r}}}\big/\big(\frac{2g_{N}^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}+1\big)\simeq\bar{\mathsf{n}}_{\text{{r}}}\big/\frac{2g_{N}^{2}}{\kappa\tilde{\Upsilon}_{\mathsf{ab}}^{\prime}}. (34) Here the atom-resonator coupling strength has been modified as g_{N}=\sqrt{N}g for the situation of many atoms.