Arxiv:2102.07429V2 [Cond-Mat.Stat-Mech] 24 Jul 2021

Toward conjecture: Heating is faster than cooling

Tan Van Vu1, 2, ∗ and Yoshihiko Hasegawa1, † 1Department of Information and Communication Engineering, Graduate School of Information Science and Technology, The University of Tokyo, Tokyo 113-8656, Japan 2Current Address: Department of Physics, Keio University, 3-14-1 Hiyoshi, Kohoku-ku, Yokohama 223-8522, Japan (Dated: July 27, 2021) An asymmetry in thermal relaxation toward equilibrium has been uncovered for Langevin systems near stable minima [Phys. Rev. Lett. 125, 110602 (2020)]. It is conjectured that, given the same degree of nonequilibrium of the initial distributions, relaxation from a lower temperature state (heating) is faster than that from a higher temperature state (cooling). In this study, we elucidate the conjecture for discrete-state Markovian systems described by the master equation. We rigorously prove that heating is faster than cooling for arbitrary two-state systems, whereas for systems with more than two distinct energy levels, the conjecture is, in general, invalid. Furthermore, for systems whose energy levels degenerate into two energy states, we ﬁnd that there exist critical thresholds of the energy gap. Depending on the magnitude of this energy gap, heating can be faster or slower than cooling, irrespective of the transition rates between states. Our results clarify the conjecture for discrete-state systems and reveal several hidden features inherent in thermal relaxation.

I. INTRODUCTION via the convergence rate of the system state toward the equilibrium state [18]. Given two identical systems initi- Systems attached to thermal reservoirs will relax to- ated in thermal states, one at a lower and the other at a ward a stationary state. Such thermal relaxation pro- higher temperature than the given temperature, a natu- cesses are ubiquitous in nature and possess rich prop- ral question arises: which one relaxes faster? Recently, erties from both dynamic and thermodynamic perspec- this question has been addressed by Lapolla and Godec tives. One of the counterintuitive behaviors is the [19] for continuous-state Langevin systems. By consider- Mpemba effect [1], where cooling a hot system is faster ing a pair of thermodynamically equidistant temperature than cooling a cold system. Such nonmonotonic relax- quenches (which have the same nonequilibrium free en- ation phenomena have been observed in various systems ergy difference), they unveiled an unforeseen asymmetry [2–6] and theoretically analyzed for microscopic dynamics in thermal relaxation; i.e., relaxation from a lower tem- [7–9]. In addition, it was found that cooling a system be- perature is faster than that from a higher temperature. fore heating it could lead to exponentially fast relaxation Roughly speaking, it implies that heating up cold ob- [10]. From the perspective of thermodynamics, thermal jects is faster than cooling down hot objects. This phe- relaxation processes exhibit universal relations regard- nomenon has been proven for quenches of dynamics near ing irreversibility, which is quantified by irreversible en- stable minima, whereas the case of general overdamped tropy production [11]. Notably, it has been shown that dynamics remains a conjecture. Considering the average irreversible entropy production during thermal relaxation energy, it can be observed that the energy of a thermal is lower-bounded by information-theoretical [12–14] and state at a higher temperature is more than that at a lower geometrical [15] distances between the initial and final temperature. This allows us to say that, from the ener- states in both classical and quantum regimes. These re- getic perspective, uphill relaxation is faster than downhill lations imply stronger inequalities than the conventional relaxation, which is counterintuitive to an extent. More- second law of thermodynamics and impose geometrical over, relaxation speed cannot be characterized solely by constraints on the possible relaxation path. Since ther- thermodynamic quantities such as dissipation or frenesy mal relaxation is important in condensed matter [16] and [20]. Therefore, it is highly nontrivial that free energy heat engines [17], deepening our understanding of ther- plays an essential role as a quantifier of nonequilibrium mal relaxation would benefit research in these areas. degree in equidistant temperature quenches. Consider preparing a thermal state corresponding to a In this study, we elucidate the conjecture for discrete- arXiv:2102.07429v2 [cond-mat.stat-mech] 24 Jul 2021 given temperature via thermal relaxation; i.e., attaching state systems modeled by Markov jump processes, thus a system to a single reservoir and allowing it to relax to- improving our understanding of thermal relaxation. ward equilibrium. In this setting, the relaxation time is a First, we prove that for arbitrary two-state systems, heat- quantity of interest and can be approximately estimated ing is faster than cooling, which partially validates the conjecture. However, we find that it is not the case for general systems with at least three distinct energy levels. By analytically constructing counterexamples, we ∗ [email protected] demonstrate that heating can be faster or slower than † [email protected] cooling, depending on the transition rates, which implies 2 that the conjecture does not universally hold for discrete- time-evolution distribution corresponding to the initial state systems. Nevertheless, restricted to a particular state |πii, then heating is said to be faster (slower) than c f h f class of systems, some universal results on relaxation cooling if D(pt kπ ) < (>)D(pt kπ ) in the long time. In asymmetry are obtained. We show that when the en- what follows, we explain in detail how to determine the ergy levels of the system are two-state degenerate to two relaxation speed in the long-time regime. energy states, there exist two critical energy gap thresh- Let 0 = λ1 > λ2 > ··· > λN be the eigenvalues of the olds. Depending on whether an energy gap is larger or transition matrix and {|vni}n be the set of corresponding smaller than these thresholds, it can be concluded with eigenvectors, certainty that heating is faster or slower than cooling. R|vni = λn|vni. (2) Notably, all eigenvalues are real numbers since the ma- II. SETUP trix R satisfies the detailed balance condition [22]. The N eigenvectors {|vni} form a basis for the space R with f We consider the thermal relaxation process of an open |v1i = |π i, and |vni is a traceless vector for all n ≥ 2 system with N states. The system is coupled to a ther- (i.e., h1|vni = 0 since h1|R = h0|). Here, |0i (|1i) denotes −1 mal reservoir at the inverse temperature βf = (kBTf ) , the N-dimensional vector with all zero (one) elements. i where kB is the Boltzmann constant. Owing to inter- Therefore, the initial distribution |π i can be expressed action with the thermal reservoir, stochastic transitions as a linear combination of {|vni} as follows: between states are induced. The dynamics of the system N are governed by the master equation, i f X i |π i = |π i + γn|vni, (3) n=2 ∂t|pti = R|pti, (1) where γi ’s are real numbers. Consequently, the proba- > n where |pti := [p1(t), . . . , pN (t)] denotes the probabil- bility distribution at time t can be analytically written ity distribution of the system at time t; the matrix R = in the following form: N×N [Rmn] ∈ R is time-independent with Rmn ≥ 0 de- noting the transition rate from state n to state m (6= n), N i f X i λnt P |pti = |π i + γne |vni. (4) and m Rmn = 0. Without loss of generality, we assume n=2 that E1 ≥ · · · ≥ EN , where En denotes the energy of state n. The transition rates satisfy the detailed balance i In the long-time limit, the probability distribution |pti −β En −β Em condition, Rmne f = Rnme f , which is the suf- can be approximated up to the second-order term as ficient condition such that the system always relaxes to f i f i λ2t the thermal Gibbs state |π i after a sufficiently long time, |p (t)i ' |π i + γ2e |v2i. (5) irrespective of the initial state. Here, πf = e−βf En /Z n βf Thus, the relaxation speed can be quantified via the value PN −βf En and Zβ = e . The transition rates can be i f n=1 of |γ2| [7]. Accordingly, heating is faster (slower) than −βf (Bmn−En) c h expressed as Rmn = Γe for m 6= n, where cooling if |γ2| < (>)|γ2 | (see AppendixA for proof). i Bmn = Bnm are the barrier coefficients, and Γ is a posi- A closed form of γ2 can be obtained analytically [8]. tive constant. The transition rate matrix R can be transformed to a Now, let us formulate the problem. We consider re- symmetric matrix R as follows: laxation that initiates from a thermal state |πii associ- 1/2 −1/2 −1 R = F RF , (6) ated with the inverse temperature βi = (kBTi) . This can be regarded as a temperature quench Ti → Tf at N×N βf En − where F = [Fmn] ∈ R with Fmn = e δmn. time t = 0 . Given a pair of cold and hot tempera- Notably, the matrix R has the same eigenvalues as R, tures, Tc and Th, satisfying Tc < Tf < Th, we investigate and its eigenvectors {|fni} are related to those of R as the relaxation speed depending on the quench direction, 1/2 |fni = F |vni. Moreover, these eigenvectors are mu- Ti = Tc ↑ Tf (heating) and Ti = Th ↓ Tf (cooling). tually orthogonal, hfm|fni = hvn|F|vmiδmn. Multiplying The degree of nonequilibrium (or free energy) of each hf |F1/2 on both sides of Eq. (3), we can show that γi initial state is the same, i.e., D(πckπf ) = D(πhkπf ), 2 2 P is proportional to the inner product between the initial where D(pkq) := n pn ln(pn/qn) is the relative entropy distribution |πii and the vector |f i, given by between two distributions |pi and |qi. We aim to an- 2 1/2 i i swer the question of which quench direction fastens the i hf2|F |π i hf 2|π i relaxation. To this end, we first need to quantify the re- γ2 = = , (7) hf2|f2i hf2|f2i laxation speed, which can be evaluated by the distance 1/2 f between the system state and the thermal state. Analo- where |f 2i := F |f2i. Note that hf 2|π i = 0 since 1/2 f gous to Refs. [7, 19], the relative entropy is used to mea- |f2i and |f1i = F |π i are orthogonal. Because the i sure the distance between states. In thermal relaxation, sign of γn can be absorbed by changing the eigenvectors h the relative entropy is closely related to the free energy |vni → −|vni, hereafter, we assume γ2 ≤ 0, which implies i h and irreversible entropy production [21]. Let |pti be the that hf 2|π i ≤ 0. 3

P III. MAIN RESULTS where S(p) := − n pn ln pn is the Shannon entropy of the distribution |pi. Note that xc < xh < 1/2 and i Given the above setup, we now present our main re- g(xi) := dS(π )/dxi = ln[(1 − xi)/xi] is a strictly con- sults, which are embodied in three theorems on relaxation vex function over xi ∈ [0, 1/2]. Applying the Hermite– asymmetry. Hadamard inequality for g(xi), we obtain the following:

Z xh Theorem 1. For two-state systems, heating is faster 1 xc + xh than cooling. g(xi)dxi > g , (9) xh − xc xc 2 This result positively supports the conjecture. Even or equivalently, for two-state systems, it is highly nontrivial that heating is faster than cooling. For discrete-state Markovian dy- x + x S(πh) − S(πc) > (x − x )g c h . (10) namics, the speed of state transformation is constrained h c 2 by time-antisymmetric dissipation and time-symmetric frenesy (or dynamical activity) [23, 24]. Relaxation from Combining Eqs. (8) and (10) results in the following in- a thermal state at a higher temperature has a higher dy- equality: namical activity; in other words, there are more jumps between states than relaxation from a thermal state at x + x g(x ) > g c h . (11) a lower temperature. Consequently, one may intuitively f 2 expect that cooling, which has higher dynamical activity, is faster than heating. However, the result is counterintu- Since g(xi) is a strictly decreasing function, we have (xc+ itive, implying that the dynamical activity alone cannot xh)/2 > xf , which completes the proof. account for this relaxation asymmetry. Next, we consider more general systems that have at c Proof of Theorem1. It suffices to prove that |γ2| < least three distinct energy levels; i.e., there exist three h h |γ2 | = −γ2 . For N = 2, an arbitrary probability dis- indices 1 ≤ i < j < k ≤ N such that Ei > Ej > Ek. For i tribution |π i can be expressed as a point (xi, 1 − xi) such systems, we obtain the following result. in the two-dimensional space. Since E1 > E2, all thermal states, |πci, |πf i, and |πhi, lie on the segment with Theorem 2. For systems with more than two distinct (0, 1) and (1/2, 1/2) as endpoints. These probability dis- energy levels, heating can be faster or slower than cooling, tributions are geometrically illustrated in Fig.1. In the depending on the transition rates. following, we employ a geometrical approach to prove f Theorem2 implies that the conjecture is invalid in the Theorem1. From the conditions, hf 2|π i = 0 and h c general case. With an appropriate choice of barrier co- hf 2|π i ≤ 0, we can conclude that hf 2|π i ≥ 0 or c c h efficients, we can construct a discrete-state system with γ2 ≥ 0. Thus, it is sufficient to show that γ2 + γ2 < 0 c h c h or hf |(πc + πh)/2i < 0, which is equivalent to prov- |γ2| < |γ2 | or |γ2| > |γ2 | as desired. This difference be- 2 tween the continuous- and discrete-state systems can be ing that (xc + xh)/2 > xf . We can rewrite the equality D(πckπf ) = D(πhkπf ) as follows: explained as follows. For simplicity, we consider a single- particle system described by the overdamped Langevin h c 1 − xf equation. In this continuous-state system, the particle S(π ) − S(π ) = (xh − xc) ln , (8) xf tends to transit to a close place at any instant time. By contrast, in discrete-state systems, the particle, in prin- ciple, can jump to anywhere, provided the transition rate between these states is positive. This degree of freedom could lead to a complicated relaxation compared with the continuous-state system. As shown below, the construction of the transition matrix leading to the violation of the conjecture is somewhat artificial; therefore, a realistic system with such a transition matrix may not exist. We anticipate that the conjecture may be valid if appropriate constraints are imposed on the transition rates.

Proof of Theorem2. We prove the theorem by analytically constructing a transition rate matrix such that heating is slower than cooling. A transition rate matrix for the opposite case can also be analogously constructed. First, one can prove that |πci, |πhi, and |πf i are linearly independent (see AppendixB). Let {|e1i, |e2i} be an or- FIG. 1. Geometrical illustration of probability distributions. thogonal basis of the space S spanned by |πf i and |πhi, Two vectors |f i and |πf i are orthogonal, hf |πf i = 0. 2 2 and P := |e1ihe1|+|e2ihe2| be the projection matrix to the 4

c c space S. Define |f 2i := |π i − P|π i, then |f 2i= 6 |0i be- ∆E is large, the jump from energy state En+1 to En is cause |πci, |πhi, and |πf i are linearly independent. Triv- less likely to occur compared with the opposite jump. f h ially, hf 2|e1i = hf 2|e2i = 0; thus, hf 2|π i = hf 2|π i = 0. Thus, heating is expected to be slower than cooling. c Moreover, hf 2|π i = hf 2|f 2i > 0 since |f 2i= 6 |0i. However, counterintuitively, heating is faster than cool- Next, we construct a transition rate matrix R that re- ing as ∆E is sufficiently large. In addition, provided c h sults in |γ2| > |γ2 |. Set Bmn = Em + En, one can explic- n ≤ N/2, heating is faster than cooling, regardless of the itly calculate that the matrix R has a single zero eigenvalue of ∆E. This implies that the number of excited 1/2 f value associated with the eigenvector |f1i = F |π i, states also plays a determinant role in relaxation speed. and the remaining eigenvalues are all −Zβf [8]. Let U = {|xi ∈ N | hx|f i = 0} be a subspace orthog- Proof of Theorem3. We employ the same strategy used R 1 in proving Theorem1. We prove the former case first, i.e., onal to |f1i. Then, there exists an orthogonal basis −1/2 βh∆E ≥ ln[n/(N − n)] leads to a faster heating. It can {|f2i, |f3i,..., |fN i} of U, where |f2i := F |f 2i, since hf |f i = 0 (i.e., |f i ∈ U). Obviously, |f i (n ≥ 2) be observed that all points lying on the segment ` with 2 1 2 n |πci and |πhi as endpoints are thermal states. It is also is an eigenvector of R with the corresponding eigenvalue c f h −Z . Following the idea in Ref. [8], we slightly modify evident that |π i, |π i, and |π i are linearly dependent, βf i.e., there exists a real number a ∈ (0, 1) such that |πf i = R as follows: c h f h a|π i + (1 − a)|π i. Since hf 2|π i = 0 and hf 2|π i ≤ 0, c c N hf 2|π i ≥ 0 or γ2 ≥ 0 follows from the continuity of X |fnihfn| R → R + n . (12) the inner product hf 2|pi for |pi ∈ `. Thus, it suffices to hfn|fni c h c h n=2 show that γ2 + γ2 < 0 or hf 2|(π + π )/2i < 0, which is equivalent to proving that (xc + xh)/2 > xf , where Here, Zβf > 2 > ··· > N ≥ 0 are small numbers that i i xi = πn completely characterizes the thermal state |π i. ensure the positivity of Rmn (m 6= n). It is easy to check The condition D(πckπf ) = D(πhkπf ) can be rewritten that R|f1i = |0i and R|fni = (−Zβf + n)|fni for all as follows: n ≥ 2. Now, the matrix R has N different eigenvalues, h c 1 − nxf and |f2i is precisely the eigenvector corresponding to the S(π ) − S(π ) = (xh − xc)n ln . (14) (N − n)xf second-largest eigenvalue λ2 = −Zβf +2. The transition rate matrix can be recovered as R = F−1/2RF1/2, and the Note that x < x ≤ 1/(2n) since β ∆E ≥ ln[n/(N −n)], detailed balance condition is satisfied due to the symme- c h h and g(x ) := dS(πi)/dx = n ln[(1 − nx )/(N − n)x ] is a try of R. With this construction, the relation between i i i i c h strictly convex function over xi ∈ [0, 1/(2n)]. Applying |γ2| and |γ2 | can be clarified as the Hermite–Hadamard inequality for g(xi) and following c h the same steps as in Eqs. (9) and (10), we obtain (xc + c hf 2|π i hf 2|π i h |γ2| = > 0 = = |γ2 |, (13) xh)/2 > xf , which proves the former case. hf2|f2i hf2|f2i When βc∆E ≤ ln[n/(N − n)], one can derive that which completes the proof. 1/(2n) ≤ xc < xh, and g(xi) is a strictly concave function over xi ∈ [1/(2n), 1]. Applying the Hermite–Hadamard Finally, we consider the remaining case, wherein the inequality for g(xi), we obtain the following: energy levels are degenerate to two energy states. In Z xh other words, there exists an index 1 ≤ n < N such that 1 xc + xh g(x ) = g(x )dx < g , (15) E = ··· = E > E = ··· = E . Such degener- f i i 1 n n+1 N xh − xc xc 2 ated two-level systems are seen in atoms [25] and have h c been used to enhance quantum-annealing performance or xf > (xc + xh)/2, which implies |γ2 | < |γ2|. [26] and dissipation-less heat current [27]. For convenience, we define the energy gap ∆E := En − En+1 > 0. Remarkably, we find that, depending on the magnitude of this energy gap, heating can be faster or slower than IV. SUMMARY cooling, regardless of the transition rates. Details are summarized in the following theorem. In this study, we elucidated the conjecture on the re- Theorem 3. If β ∆E ≥ ln[n/(N − n)], then heating is laxation asymmetry for discrete-state systems described h by Markov jump processes. We proved that the conjec- faster than cooling. Moreover, if βc∆E ≤ ln[n/(N − n)], then heating is slower than cooling. ture holds for any two-state systems but fails for general systems with more than two distinct energy levels. For Theorem3 indicates that there are two critical thresh- systems with two degenerate energy levels, we obtained olds of the energy gap. As the energy gap is above or some universal results indicating that the asymmetry in below these thresholds, a universal conclusion on asym- relaxation depends on the energy gap and the number of metry in thermal relaxation can be drawn. It is also excited states. highly nontrivial that the energy gap affects the relax- Although the conjecture is shown to fail in the general ation speeds of heating and cooling in this way. When discrete-state case, the conjecture may be valid under 5

> some constraints on transition rates. Such a question Here, |vni = [vn1, . . . , vnN ] . Consequently, we have requires further investigation and will be addressed in future work. It would also be interesting to study the re- N v2 h c h 2 c 2 X 2k 2λ2t laxation asymmetry for non-Markovian systems [28] and D(pt ) − D(pt ) = (γ2 ) − (γ2) e πf open quantum systems [29, 30] where coherence emerges k=1 k X (λm+λn)t 3λ2t in the initial state. + amne + O(e ), m,n≥2 Note added. After the completion of this work, we no- max{m,n}>2 ticed a related work [31], which studied the relaxation (A3) asymmetry via a numerical analysis of a two-state sys- where amn are constants. The ﬁrst term on the right- c h tem. hand side is positive since |γ2| < |γ2 |. The remaining terms may be negative; however, they are negligible compared with the ﬁrst term in the long-time limit. There- c h fore, D(pt ) < D(pt ) as t → ∞.

Appendix B: Proof of linear independence

ACKNOWLEDGMENTS To prove the linear independence of |πci, |πf i, and |πhi, it is suﬃcient to show that the determinant of the following matrix is negative. We thank Keiji Saito for insightful discussions. This work was supported by Ministry of Education, Culture, e−βcE1 e−βcE2 e−βcE3  Sports, Science and Technology (MEXT) KAKENHI X = e−βf E1 e−βf E2 e−βf E3  . (B1) Grant No. JP19K12153. e−βhE1 e−βhE2 e−βhE3

Here, E1 > E2 > E3 and βc > βf > βh. Without loss of generality, we assume that E3 = 0 and βh = 0. In this case, the determinant can be calculated as follows:

|X| = (1−e−βcE1 )(1−e−βf E2 )−(1−e−βcE2 )(1−e−βf E1 ). (B2) Appendix A: Quantiﬁcation of relaxation speed Therefore, |X| < 0 is equivalent to

1 − e−βcE1 1 − e−βcE2 f < . (B3) For convenience, we deﬁne D(p) := D(pkπ ). Given 1 − e−βf E1 1 − e−βf E2 |πii = |πf i + PN γi |v i and |γc| < |γh|, we will n=2 n n 2 2 −β x −β x c h Set h(x) := (1−e c )/(1−e f ), Eq. (B3) is equivalent show that D(pt ) < D(pt ) as t → ∞. Note that N to h(E ) < h(E ). We need only prove that h(x) is |pii = |πf i + P γi eλnt|v i. In the long-time limit, 1 2 t n=2 n n a strictly decreasing function over x > 0. Taking the PN i λnt the term n=2 γne |vni vanishes. Since derivative of h(x) with respect to x, we have

(β −βc)x β x βcx dh(x) e f [βc(e f − 1) − βf (e − 1)] = β x 2 . (B4) N dx (e f − 1) X dp2 D(pkp + dp) = k + O(∆3), (A1) pk Since k=1 ∞ X xn β (eβf x − 1) − β (eβcx − 1) = (β βn − β βn) c f n! c f f c n=1 N where ∆ = P |dp |, we can approximate ∞ k=1 k X xn = β β (βn−1 − βn−1) n! c f f c n=1 2 < 0 (∵ 0 < βf < βc), N PN i λnt n=2 γne vnk i X 3λ2t D(pt) = f + O(e ). (A2) we have dh(x)/dx < 0, which completes the proof. k=1 πk

[1] E. B. Mpemba and D. G. Osborne, Cool?, Phys. Educ. [2] D. Auerbach, Supercooling and the Mpemba eﬀect: 4, 172 (1969). When hot water freezes quicker than cold, Am. J. Phys. 6

63, 882 (1995). [17] G. Benenti, G. Casati, K. Saito, and R. S. Whitney, Fun- [3] A. Lasanta, F. Vega Reyes, A. Prados, and A. Santos, damental aspects of steady-state conversion of heat to When the hotter cools more quickly: Mpemba effect in work at the nanoscale, Phys. Rep. 694, 1 (2017). granular fluids, Phys. Rev. Lett. 119, 148001 (2017). [18] K. Temme, M. J. Kastoryano, M. B. Ruskai, M. M. Wolf, [4] M. Baity-Jesi, E. Calore, A. Cruz, L. A. Fernandez, and F. Verstraete, The χ2-divergence and mixing times J. M. Gil-Narvión, A. Gordillo-Guerrero, D. Iñiguez, of quantum Markov processes, J. Math. Phys. 51, 122201 A. Lasanta, A. Maiorano, E. Marinari, V. Martin-Mayor, (2010). J. Moreno-Gordo, A. M. Sudupe, D. Navarro, G. Parisi, [19] A. Lapolla and A. c. v. Godec, Faster uphill relaxation S. Perez-Gaviro, F. Ricci-Tersenghi, J. J. Ruiz-Lorenzo, in thermodynamically equidistant temperature quenches, S. F. Schifano, B. Seoane, A. Tarancón,R. Tripiccione, Phys. Rev. Lett. 125, 110602 (2020). and D. Yllanes, The Mpemba effect in spin glasses is a [20] C. Maes, Frenesy: Time-symmetric dynamical activity in persistent memory effect, Proc. Natl. Acad. Sci. U.S.A. nonequilibria, Phys. Rep. 850, 1 (2020). 116, 15350 (2019). [21] U. Seifert, Stochastic thermodynamics, fluctuation the- [5] A. Kumar and J. Bechhoefer, Exponentially faster cool- orems and molecular machines, Rep. Prog. Phys. 75, ing in a colloidal system, Nature 584, 64 (2020). 126001 (2012). [6] A. Biswas, V. V. Prasad, O. Raz, and R. Rajesh, [22] J. Schnakenberg, Network theory of microscopic and Mpemba effect in driven granular Maxwell gases, Phys. macroscopic behavior of master equation systems, Rev. Rev. E 102, 012906 (2020). Mod. Phys. 48, 571 (1976). [7] Z. Lu and O. Raz, Nonequilibrium thermodynamics of [23] N. Shiraishi, K. Funo, and K. Saito, Speed limit for clas- the Markovian Mpemba effect and its inverse, Proc. Natl. sical stochastic processes, Phys. Rev. Lett. 121, 070601 Acad. Sci. U.S.A. 114, 5083 (2017). (2018). [8] I. Klich, O. Raz, O. Hirschberg, and M. Vucelja, Mpemba [24] V. T. Vo, T. Van Vu, and Y. Hasegawa, Unified approach index and anomalous relaxation, Phys. Rev. X 9, 021060 to classical speed limit and thermodynamic uncertainty (2019). relation, Phys. Rev. E 102, 062132 (2020). [9] S. Takada, H. Hayakawa, and A. Santos, Mpemba effect [25] L. Margalit, M. Rosenbluh, and A. D. Wilson-Gordon, in inertial suspensions, Phys. Rev. E 103, 032901 (2021). Degenerate two-level system in the presence of a trans- [10] A. Gal and O. Raz, Precooling strategy allows exponen- verse magnetic field, Phys. Rev. A 87, 033808 (2013). tially faster heating, Phys. Rev. Lett. 124, 060602 (2020). [26] S. Watabe, Y. Seki, and S. Kawabata, Enhancing quan- [11] G. T. Landi and M. Paternostro, Irreversible entropy tum annealing performance by a degenerate two-level sys- production, from quantum to classical, arXiv preprint tem, Sci. Rep. 10 (2020). arXiv:2009.07668 (2020). [27] H. Tajima and K. Funo, Superconducting-like heat [12] A. M. Alhambra and M. P. Woods, Dynamical maps, current: Effective cancellation of current-dissipation quantum detailed balance, and the Petz recovery map, trade off by quantum coherence, arXiv preprint Phys. Rev. A 96, 022118 (2017). arXiv:2004.13412 (2020). [13] N. Shiraishi and K. Saito, Information-theoretical bound [28] Z.-Y. Yang and J.-X. Hou, Non-Markovian Mpemba ef- of the irreversibility in thermal relaxation processes, fect in mean-field systems, Phys. Rev. E 101, 052106 Phys. Rev. Lett. 123, 110603 (2019). (2020). [14] T. Van Vu and Y. Hasegawa, Lower bound on irreversibil- [29] H.-P. Breuer and F. Petruccione, The Theory of Open ity in thermal relaxation of open quantum systems, arXiv Quantum Systems (Oxford University Press, New York, preprint arXiv:2102.07348 (2021). 2002). [15] T. Van Vu and Y. Hasegawa, Geometrical bounds of the [30] A. Nava and M. Fabrizio, Lindblad dissipative dynamics irreversibility in Markovian systems, Phys. Rev. Lett. in the presence of phase coexistence, Phys. Rev. B 100, 126, 010601 (2021). 125102 (2019). [16] S. Dattagupta, Relaxation Phenomena in Condensed [31] S. K. Manikandan, Faster uphill relaxation of a two- Matter Physics (Elsevier, New York, 2012). level quantum system, arXiv preprint arXiv:2102.06161 (2021).