betaFeedback

WorksPhysics and Astronomy

A four-state Markovian Mpemba effect in which every triplet test fails

Abstract

Avitan, Factor and Gelbwaser-Klimovsky derived necessary conditions on the transition rates of a three-level system for the Markovian Mpemba effect, and proposed applying them to every triplet of levels of an NN-level system. For a zero-temperature bath and non-degenerate spectrum they proved that a strong violation of the conditions in all triplets excludes the effect; for non-zero temperature they conjectured, on the basis of numerical sampling, that violation in all triplets excludes it. We give an explicit reversible four-state chain, with energies U(0,1,2,64)U(0,1,2,64) and a bath at inverse temperature ln⁡2/U\ln 2/U, in which all four triplet conditions fail strictly and yet a hotter thermal initial state has a strictly smaller slow-mode amplitude than a colder one. This is a direct Mpemba effect, so the finite-temperature conjecture fails for N=4N=4. The example is marginal: it uses Boltzmann weights of order 2−642^{-64}, initial temperatures of order 102 U10^2\,U, a relative amplitude gap of about 0.07%0.07\%, and three of the four triplets fail by margins of order 10−610^{-6}. We then give a second, less marginal example with energies (0,130,0.434,5)(0,\frac{1}{30},0.434,5), bath inverse temperature 33 and initial inverse temperatures 0.060.06 and 16\frac{1}{6}: every triplet fails by at least 1.08×10−31.08\times10^{-3} relative to its largest rate, and the relative amplitude gap is 1.22%1.22\%. A variant of the second example with rational Boltzmann factors has the same properties and is certified in exact rational arithmetic without exponential enclosures. All claims are reduced to finite exact rational or rational-interval computations, which we describe. In both examples the ground level escapes more slowly than the slowest mode relaxes; an exact elimination identity shows that the slow mode then sees effective couplings between the excited levels that depend on the slowest eigenvalue, and these are not visible to tests built from the original rates.

Mathematics Subject Classification (2020). Primary 82C05; Secondary 60J27.

Keywords. Mpemba effect; continuous-time Markov chains; detailed balance; relaxation to equilibrium; slowest relaxation mode; counterexample; interval arithmetic.

Setting

Consider NN levels with energies E1<⋯<ENE_1 < \cdots< E_N and a continuous-time Markov chain p˙=Mp\dot{p}=Mp on column probability vectors, with Mij=aij≥0M_{ij}=a_{ij}\geq0 the rate j→ij\to i for i≠ji\neq j and Mjj=−∑i≠jaijM_{jj}=-\sum_{i\neq j}a_{ij}, as in [1 Eq. (1)]. The rates satisfy detailed balance with respect to the bath, aije−βbEj=ajie−βbEia_{ij}e^{-\beta_b E_j}=a_{ji}e^{-\beta_b E_i}, so that π(βb)\pi(\beta_b) is stationary, where π(β)i=e−βEi/Z(β)\pi(\beta)_i=e^{-\beta E_i}/Z(\beta).

If the slowest nonzero eigenvalue λ2\lambda_2 of MM is simple, with right eigenvector V2V_2 and left eigenvector ℓ\ell normalized by ℓTV2=1\ell^{\mathsf{T}}V_2=1, then for the initial state p(0)=π(β)p(0)=\pi(\beta),

p(t)−π(βb)=b(β)eλ2tV2+O(eλ3t),b(β)=ℓTπ(β),(1)p(t)-\pi(\beta_b)=b(\beta)e^{\lambda_2 t}V_2+O(e^{\lambda_3 t}), \qquad b(\beta)=\ell^{\mathsf{T}}\pi(\beta), \tag*{(1)}

where λ3<λ2\lambda_3<\lambda_2 is the next eigenvalue; bb is the coefficient b2b_2 of [1 Eq. (2)], going back to [3]. For two initial temperatures above the bath temperature, βh<βc<βb\beta_h<\beta_c<\beta_b, a direct Mpemba effect occurs when ∣b(βh)∣<∣b(βc)∣|b(\beta_h)|<|b(\beta_c)|: the hotter state is asymptotically closer to equilibrium. This is the criterion of [3] and [4 Sec. II]; a strong effect in the sense of [4 Sec. III] would require in addition b=0b=0 at some initial temperature. For a review of the Mpemba effect and related anomalous relaxation phenomena see [5].

The triplet tests. For three levels i<j<ki < j < k write −- = ii, 0=j0 = j, +=k+ = k, Δ0−=Ej−Ei\Delta_{0-} = E_j - E_i, and C+−=1+12e−βb(Ek−Ei)C_{+-} = 1 + \frac{1}{2}e^{-\beta_b(E_k-E_i)}, C+0=1+12e−βb(Ek−Ej)C_{+0} = 1 + \frac{1}{2}e^{-\beta_b(E_k-E_j)}. With the original rates of MM, set

κ=a−++a0+a−0(1+e−βbΔ0−)\kappa= \frac{a_{-+}+a_{0+}}{a_{-0}(1+e^{-\beta_b\Delta_{0-}})}

The necessary conditions of [1 Eq. (3)] are

κ≥1:a−+C+−−a0+C+0>a−0(1−e−βbΔ0−),(2)\kappa\ge1:\quad a_{-+}C_{+-}-a_{0+}C_{+0}>a_{-0}(1-e^{-\beta_b\Delta_{0-}}), \tag*{(2)}
κ≤1:a−+e−βbΔ0−>a0+.(3)\kappa\le1:\quad a_{-+}e^{-\beta_b\Delta_{0-}}>a_{0+}. \tag*{(3)}

(The letter kk is used for κ\kappa in [1].) A triplet fails the test if the inequality of its branch does not hold. In all triplets below κ≠1\kappa\ne1, so the convention at κ=1\kappa= 1 plays no role.

What is proved in [1]. For NN levels one applies (1)–(2) to every triplet with the original rates. For a zero-temperature bath (triangular MM), with non-degenerate eigenvalues and the strong violation a−0>a−++a0+a_{-0}>a_{-+}+a_{0+} in every triplet, [1] proves that the NN-level system shows no Mpemba effect. For a non-zero-temperature bath, [1] conjectures that if all triplets violate (1)–(2) then there is no Mpemba effect, supported by random sampling of 4- to 7-level systems (section “Numerics for non-zero temperature baths”). For N=3N = 3 the only triplet is the system itself and (1)–(2) are the three-level necessary conditions derived in [1], so the conjecture concerns N≥4N \ge4. A protocol for producing the effect in NN-level systems is developed in [2]; its Section II restates the triplet criterion, without the zero-temperature restriction, as a result of [1]: if no triplet satisfies the conditions, the NN-level system shows no Markovian Mpemba effect. The examples below are counterexamples, with N=4N = 4, to the finite-temperature conjecture and hence to this restatement; they are not in conflict with the zero-temperature theorem.

The example

Fix an energy unit U>0U > 0, a time unit τ>0\tau> 0 and 0<ε≤10−60 < \varepsilon\le10^{-6}. Let

E=U(0,1,2,64),βb=ln⁡2U,δ=2−64,q=(1,12,14,δ),E = U(0,1,2,64),\qquad\beta_b = \frac{\ln2}{U},\qquad\delta= 2^{-64},\qquad q = \left(1,\frac{1}{2},\frac{1}{4},\delta\right),

so π(βb)=q/(74+δ)\pi(\beta_b)=q/(\frac{7}{4}+\delta). With the symmetric conductances

(d12,d13,d14,d23,d24,d34)=(1+ε,53,2,102,3,3+ε),(d_{12},d_{13},d_{14},d_{23},d_{24},d_{34})=\left(1+\varepsilon,\frac{5}{3},2,102,3,3+\varepsilon\right),

set aij=qidij/τa_{ij}=q_i d_{ij}/\tau for i≠ji\ne j. Then aijqj=qiqjdij/τ=ajiqia_{ij}q_j=q_iq_jd_{ij}/\tau=a_{ji}q_i, which is detailed balance, and every rate is positive (the smallest is a41=2δ/τa_{41}=2\delta/\tau). Explicitly,

τMε=(−1112−ε2−2δ1+ε5321+ε2−532−ε−3δ5132512512−1583−(3+ε)δ3+ε42δ3δ(3+ε)δ−174−ε4).\tau M_\varepsilon= \begin{pmatrix} -\frac{11}{12}-\frac{\varepsilon}{2}-2\delta& 1+\varepsilon& \frac{5}{3} & 2 \\ \frac{1+\varepsilon}{2} & -\frac{53}{2}-\varepsilon-3\delta& 51 & \frac{3}{2} \\ \frac{5}{12} & \frac{51}{2} & -\frac{158}{3}-(3+\varepsilon)\delta& \frac{3+\varepsilon}{4} \\ 2\delta& 3\delta& (3+\varepsilon)\delta& -\frac{17}{4}-\frac{\varepsilon}{4} \end{pmatrix}.

The two initial inverse temperatures are

βh=ln⁡(256/255)U<βc=ln⁡(128/127)U<βb,\beta_h=\frac{\ln(256/255)}{U}<\beta_c=\frac{\ln(128/127)}{U}<\beta_b,

which holds because 1128<127128<255256<1\frac{1}{128}<\frac{127}{128}<\frac{255}{256}<1. Writing t=e−βUt=e^{-\beta U}, one has π(β)=(1,t,t2,t64)/(1+t+t2+t64)\pi(\beta)=(1,t,t^2,t^{64})/(1+t+t^2+t^{64}), and th=255256t_h=\frac{255}{256}, tc=127128t_c=\frac{127}{128} are rational.

Theorem 1. For every U,τ>0U,\tau> 0 and 0<ε≤10−60 < \varepsilon\le10^{-6}.

  1. all four triplets {1,2,3}\{1,2,3\}, {1,2,4}\{1,2,4\}, {1,3,4}\{1,3,4\}, {2,3,4}\{2,3,4\} fail the tests (1)–(2), each with a strict inequality;

  1. the nonzero eigenvalues of MεM_{\varepsilon} are simple, and the slowest one is λ2=μ2/τ\lambda_2=\mu_2/\tau with

μ2∈[−2.13661042001,−2.13660892000];\mu_2\in[-2.13661042001,-2.13660892000];
  1. with the slow-mode amplitude bb of Section 1,

0.398692971890≤b(βh)≤0.398695686013,0.398692971890\le b(\beta_h)\le0.398695686013,
0.398971230884≤b(βc)≤0.398973922273,0.398971230884\le b(\beta_c)\le0.398973922273,
0.000278140493≤b(βc)−b(βh)≤0.000278354761.0.000278140493\le b(\beta_c)-b(\beta_h)\le0.000278354761.

In particular 0<b(βh)<b(βc)0<b(\beta_h)<b(\beta_c), so ∣b(βh)∣<∣b(βc)∣|b(\beta_h)|<|b(\beta_c)|: the chain shows a direct Mpemba effect although every triplet fails the test.

Changing UU rescales all energies and inverse temperatures together and leaves βE\beta E unchanged; changing τ\tau multiplies all rates and eigenvalues by a common factor. Neither affects any statement, so below U=τ=1U=\tau=1.

Proof

(i) Triplets. For the triplet i<j<ki<j<k one has a−0=qidija_{-0}=q_i d_{ij}, a−+=qidika_{-+}=q_i d_{ik}, a0+=qjdjka_{0+}=q_j d_{jk} and e−βbΔ0−=qj/qie^{-\beta_b\Delta_{0-}}=q_j/q_i, C+−=1+qk/(2qi)C_{+-}=1+q_k/(2q_i), C+0=1+qk/(2qj)C_{+0}=1+q_k/(2q_j). Direct substitution gives the following, where “residual” is left side minus right side of the applicable inequality (1) or (2):

tripletκ\kappabranchresidual
1233169(1+ε)>1\frac{316}{9(1+\varepsilon)}>1(1)−4998−ε2-\frac{499}{8}-\frac{\varepsilon}{2}
12473(1+ε)>1\frac{7}{3(1+\varepsilon)}>1(1)−δ+ε2-\frac{\delta+\varepsilon}{2}
1343(11+ε)25>1\frac{3(11+\varepsilon)}{25}>1(1)−δ2−(14+δ2)ε-\frac{\delta}{2}-\left(\frac{1}{4}+\frac{\delta}{2}\right)\varepsilon
2349+ε306<1\frac{9+\varepsilon}{306}<1(2)−ε4-\frac{\varepsilon}{4}

Table 1.

For instance, for 234: a−+e−βbΔ0−−a0+=12⋅3⋅12−14(3+ε)=−ε4a_{-+}e^{-\beta_b\Delta_{0-}}-a_{0+}=\frac{1}{2}\cdot3\cdot\frac{1}{2}-\frac{1}{4}(3+\varepsilon)=-\frac{\varepsilon}{4}. The κ\kappa values lie strictly on one side of 1 for all 0≤ε≤10−60\le\varepsilon\le10^{-6}, and every residual is strictly negative for ε>0\varepsilon>0. Thus each triplet fails its test with a strict inequality.

(ii) Spectrum. At ε=0\varepsilon=0, expanding the determinant gives det⁡(zI−M0)=zf(z)\det(zI-M_0)=zf(z) with

f(z)=z3+24δ+2533z2+1008δ2+20380δ+2433148z+(4δ+7)(432δ2+11598δ+9707)96.f(z)=z^3+\frac{24\delta+253}{3}z^2+\frac{1008\delta^2+20380\delta+24331}{48}z+\frac{(4\delta+7)(432\delta^2+11598\delta+9707)}{96}.

Evaluating ff exactly at rational endpoints shows sign changes on the disjoint intervals

[−2.13660892001,−2.13660892000],[−4.25000000001,−4.24999999999],[-2.13660892001,-2.13660892000],\qquad[-4.25000000001,-4.24999999999],
[−77.94672441334,−77.94672441332].[-77.94672441334,-77.94672441332].

A cubic has at most three roots, so these contain all roots of ff, each simple. For ε>0\varepsilon>0 let D=diag⁡(q)D=\operatorname{diag}(q) and Sε=D−1/2MεD1/2S_{\varepsilon}=D^{-1/2}M_{\varepsilon}D^{1/2}, a real symmetric matrix with off-diagonal entries qiqj dij\sqrt{q_iq_j}\,d_{ij} and the same spectrum as MεM_{\varepsilon}. Only d12d_{12} and d34d_{34} depend on ε\varepsilon, and

Sε−S0=−ε(w12w12T+w34w34T),wij=qj ei−qi ej.S_{\varepsilon}-S_0=-\varepsilon(w_{12}w_{12}^{\mathsf T}+w_{34}w_{34}^{\mathsf T}),\qquad w_{ij}=\sqrt{q_j}\,e_i-\sqrt{q_i}\,e_j.

The vectors w12,w34w_{12},w_{34} have disjoint supports and squared norms 32\frac{3}{2} and 14+δ\frac{1}{4}+\delta, so −32εI⪯Sε−S0⪯0-\frac{3}{2}\varepsilon I\preceq S_{\varepsilon}-S_0\preceq0. By Weyl’s inequality each ordered eigenvalue moves down by at most 1.5×10−61.5\times10^{-6}; zero remains an eigenvalue because π(βb)\pi(\beta_b) is stationary. The shifted intervals

μ2∈[−2.13661042001,−2.13660892000],μ3∈[−4.25000150001,−4.24999999999],\mu_2\in[-2.13661042001,-2.13660892000],\qquad\mu_3\in[-4.25000150001,-4.24999999999],
μ4∈[−77.94672591334,−77.94672441332]\mu_4\in[-77.94672591334,-77.94672441332]

are disjoint and negative, which proves (ii).

(iii) Slow left eigenvector. Let B=MεT−μ2IB=M_{\varepsilon}^{\mathsf{T}}-\mu_{2}I and J={1,3,4}J=\{1,3,4\}. Fix ℓ2=1\ell_{2}=1 and solve BJJℓJ=−BJ2B_{JJ}\ell_{J}=-B_{J2} by Cramer’s rule in outward-rounded rational interval arithmetic, with μ2\mu_{2} ranging over its enclosure and ε\varepsilon over [0,10−6][0,10^{-6}] independently. This gives det⁡BJJ∈[131.7448386,131.7451652]\det B_{JJ}\in[131.7448386,131.7451652], so BJJB_{JJ} is invertible, and

ℓ1∈[−0.746173536399,−0.746170651756],ℓ3∈[0.984686309510,0.984691237270],\ell_{1}\in[-0.746173536399,-0.746170651756],\qquad\ell_{3}\in[0.984686309510,0.984691237270],
ℓ4∈[0.353067360670,0.353070507243].\ell_{4}\in[0.353067360670,0.353070507243].

The omitted equation (row 2) also holds: permuting index 2 last, 0=det⁡B=det⁡BJJ(B22−B2JBJJ−1BJ2)0=\det B=\det B_{JJ}(B_{22}-B_{2J}B_{JJ}^{-1}B_{J2}), and the second factor is exactly row 2 evaluated at ℓ\ell. Hence ℓ\ell is a left eigenvector for λ2\lambda_{2}. By reversibility V2=diag⁡(π(βb))ℓ/(ℓTdiag⁡(π(βb))ℓ)V_{2}=\operatorname{diag}(\pi(\beta_{b}))\ell/(\ell^{\mathsf{T}}\operatorname{diag}(\pi(\beta_{b}))\ell) is a right eigenvector with ℓTV2=1\ell^{\mathsf{T}}V_{2}=1, so b(β)=ℓTπ(β)b(\beta)=\ell^{\mathsf{T}}\pi(\beta) is the amplitude of § 1.

(iv) Amplitudes. Since π(βh)\pi(\beta_{h}) and π(βc)\pi(\beta_{c}) have rational entries, evaluating ℓTπ(βh)\ell^{\mathsf{T}}\pi(\beta_{h}), ℓTπ(βc)\ell^{\mathsf{T}}\pi(\beta_{c}) and ℓT(π(βc)−π(βh))\ell^{\mathsf{T}}(\pi(\beta_{c})-\pi(\beta_{h})) in interval arithmetic with the enclosures of (iii) gives the bounds of (iii) in Theorem 1. The comparison ∣b(βh)∣<∣b(βc)∣|b(\beta_{h})|<|b(\beta_{c})| does not depend on the normalization of ℓ\ell, since both amplitudes use the same eigenvector.

Remark 2 (Sign of bb). The enclosures give ℓ1<0<ℓ4<ℓ3<ℓ2=1\ell_{1}<0<\ell_{4}<\ell_{3}<\ell_{2}=1. Hence N(t)=ℓ1+t+ℓ3t2+ℓ4t64N(t)=\ell_{1}+t+\ell_{3}t^{2}+\ell_{4}t^{64} is strictly increasing for t>0t>0; since b(βb)=0b(\beta_{b})=0, i.e. N(12)=0N(\frac{1}{2})=0, b(β)>0b(\beta)>0 for all 0≤β<βb0\leq\beta<\beta_{b}. As b(βh)<b(βc)b(\beta_{h})<b(\beta_{c}) and b(βb)=0b(\beta_{b})=0, ∣b∣|b| has an interior maximum on (βh,βb)(\beta_{h},\beta_{b}), which is the criterion for the effect quoted in [1]. Since bb has no zero on [0,βb)[0,\beta_{b}), the effect is not strong in the sense of [4] (its direct Mpemba index is zero). Note also that ℓ\ell is not monotone in the energy (ℓ4<ℓ3)(\ell_{4}<\ell_{3}).

How marginal the first example is

The example of Theorem 1 is a counterexample, not a typical case.

  • Extreme weights. Level 4 sits at 64U64U with βbU=ln⁡2\beta_{b}U=\ln2, so its bath weight is δ/(74+δ)≈3×10−20\delta/(\frac{7}{4}+\delta)\approx3\times10^{-20} and all rates into it are of order δ=2−64\delta=2^{-64}. At the initial temperatures, by contrast, level 4 is well populated: th64≈0.78t_{h}^{64}\approx0.78.

  • Near-infinite initial temperatures. kBTh=U/ln⁡(256/255)≈255.5Uk_{\mathrm{B}}T_{h}=U/\ln(256/255)\approx255.5U and kBTc≈127.5Uk_{\mathrm{B}}T_{c}\approx127.5U, compared with kBTb≈1.44Uk_{\mathrm{B}}T_{b}\approx1.44U.

  • Small gap. b(βc)−b(βh)≈2.78×10−4b(\beta_{c})-b(\beta_{h})\approx2.78\times10^{-4}, about 0.07%0.07\% of b(βh)≈0.3987b(\beta_{h})\approx0.3987.

  • Nearly passing triplets. Triplet 123 fails by a wide margin, but the residuals of 124, 134, 234 are −(δ+ε)/2-(\delta+\varepsilon)/2, −δ/2−(14+δ2)ε-\delta/2-(\frac{1}{4}+\frac{\delta}{2})\varepsilon and −ε/4-\varepsilon/4, i.e. of order ε≤10−6\varepsilon\leq10^{-6} in units of 1/τ1/\tau. At ε=0\varepsilon=0 the triplet 234 meets (2) with equality, which is still a failure of the strict test.

Since for ε>0\varepsilon>0 every inequality in Theorem 1 is strict and every quantity is continuous in the rates, small perturbations of the rates (keeping detailed balance) and of the energies preserve the conclusions; the example is therefore not an isolated point. Section 5 gives an example with much larger margins.

A robust example

For a triplet that fails its test, define the relative margin as −(residual)/max⁡(a−0,a−+,a0+)-(\text{residual})/\max(a_{-0},a_{-+},a_{0+}), with the residual of § 3; it is invariant under rescaling all rates. Take U=τ=1U=\tau=1 and

E=(0,130,217500,5),βb=3,βh=350,βc=16,E=\left(0,\frac{1}{30},\frac{217}{500},5\right),\qquad\beta_{b}=3,\qquad\beta_{h}=\frac{3}{50},\qquad\beta_{c}=\frac{1}{6},
(d12,d13,d14,d23,d24,d34)=(0.01646, 2.504, 2.648, 29.12, 2.928, 3.044),(d_{12},d_{13},d_{14},d_{23},d_{24},d_{34})=(0.01646,\ 2.504,\ 2.648,\ 29.12,\ 2.928,\ 3.044),

and aij=e−βbEidija_{ij}=e^{-\beta_{b}E_{i}}d_{ij} for i≠ji\ne j, so detailed balance holds exactly. The bath weights relative to the ground level are 11, e−0.1≈0.905e^{-0.1}\approx0.905, e−1.302≈0.272e^{-1.302}\approx0.272 and e−15≈3.1×10−7e^{-15}\approx3.1\times10^{-7}.

Theorem 3. For this chain MM:

  1. all four triplets fail the tests (1)–(2), with κ\kappa, branch and relative margin

tripletκ\kappabranchrelative margin
123920.240 . . .(1)≥1.0423\ge1.0423
124168.955 . . .(1)≥1.1060×10−3\ge1.1060 \times10^{-3}
1341.0913 . . .(1)≥1.0850×10−3\ge1.0850 \times10^{-3}
2340.10147 . . .(2)≥1.1974×10−3\ge1.1974 \times10^{-3}

Table 1 (PDF p. 5).

  1. the nonzero eigenvalues λ2>λ3>λ4\lambda_2 > \lambda_3 > \lambda_4 of MM are simple, with

λ2∈[−1.218899696,−1.218899693],λ3∈[−6.125294088,−6.125294085],\lambda_2 \in[-1.218899696,-1.218899693], \qquad\lambda_3 \in[-6.125294088,-6.125294085],
λ4∈[−36.266646875,−36.266646872];\lambda_4 \in[-36.266646875,-36.266646872];
  1. with the left eigenvector normalized by ℓ1=1\ell_1 = 1,

ℓ2∈[−0.8801465839,−0.8801465701],ℓ3∈[−0.7486027690,−0.7486027569],\ell_2 \in[-0.8801465839,-0.8801465701], \qquad\ell_3 \in[-0.7486027690,-0.7486027569],
ℓ4∈[−0.0618822687,−0.0618822652],\ell_4 \in[-0.0618822687,-0.0618822652],
b(βh)∈[−0.1760225842,−0.1760225766],b(βc)∈[−0.1781722318,−0.1781722239],b(\beta_h) \in[-0.1760225842,-0.1760225766], \qquad b(\beta_c) \in[-0.1781722318,-0.1781722239],
∣b(βc)∣−∣b(βh)∣∈[0.0021496470,0.0021496478],∣b(βc)∣−∣b(βh)∣∣b(βh)∣∈[0.0122123,0.0122124].|b(\beta_c)|-|b(\beta_h)| \in[0.0021496470,0.0021496478], \qquad\frac{|b(\beta_c)|-|b(\beta_h)|}{|b(\beta_h)|} \in[0.0122123,0.0122124].

In particular ∣b(βh)∣<∣b(βc)∣|b(\beta_h)| < |b(\beta_c)| with βh<βc<βb\beta_h < \beta_c < \beta_b: a direct Mpemba effect with a relative amplitude gap of 1.22%, while every triplet fails its test by a relative margin of at least 1.08×10−31.08 \times10^{-3}.

Proof. The argument is that of Section 3, with exact rationals replaced by rational intervals because the Boltzmann factors are irrational. Each e−ye^{-y} with rational y≥0y \ge0 is enclosed by writing y=2kzy = 2^k z with 0≤z≤180 \le z \le\frac{1}{8}, bracketing e−ze^{-z} between two consecutive partial sums of its alternating Taylor series (the terms decrease), and squaring kk times; all endpoints are rounded outward to multiples of 2−3202^{-320}. This gives interval enclosures of all rates, of the diagonal of MM, of κ\kappa and of both residuals of every triplet, and (i) follows because the computed κ\kappa intervals exclude 1 and the relevant residual intervals are negative. Since the columns of MM sum to zero, det⁡(zI−M)=zf(z)\det(zI-M)=zf(z) with ff a monic cubic; expanding the determinant over the 24 permutations in interval arithmetic encloses the coefficients of ff, and evaluating ff at the six rational bracket endpoints gives strict sign changes. The three brackets are disjoint, so they contain all roots of ff, each simple, which is (ii). For (iii), solve (MT−μI)ℓ=0(M^{T}-\mu I)\ell=0 with ℓ1=1\ell_1=1 by Cramer’s rule on the rows and columns {2,3,4}\{2,3,4\}, with μ\mu ranging over the bracket of λ2\lambda_2; the denominator lies in [113.0915273,113.0915291][113.0915273,113.0915291], so it is nonzero, and the omitted row holds by the Schur determinant argument of Section 3(iii). The amplitudes are then evaluated with interval enclosures of π(βh)\pi(\beta_h) and π(βc)\pi(\beta_c); their signs agree, so ∣b(βc)∣−∣b(βh)∣=−(b(βc)−b(βh))|b(\beta_c)|-|b(\beta_h)|=-(b(\beta_c)-b(\beta_h)), which is evaluated as −ℓT(π(βc)−π(βh))-\ell^{T}(\pi(\beta_c)-\pi(\beta_h)).

What is and is not robust. Every residual in (1)–(2) is a combination of the three rates with coefficients of absolute value at most 32\frac{3}{2}, so multiplying each dijd_{ij} by a factor in [1−η,1+η][1-\eta,1+\eta] with η≤2.5×10−4\eta\le2.5 \times10^{-4} keeps all four failures (and all branches) of Theorem 3(i); the Mpemba inequality is an open condition, but we have not quantified its neighbourhood. The energies lie in [0,5][0,5] and the bath weight of the top level is e−15e^{-15} instead of 2−642^{-64}. Three features remain special: the ground pair is close in energy (βb(E2−E1)=0.1\beta_b(E_2-E_1)=0.1) and weakly coupled directly (d12d_{12} is the smallest conductance); the rates span about seven orders of magnitude because of the Boltzmann factor e−15e^{-15}; and the initial temperatures, κBTh=503≈16.7\kappa_B T_h=\frac{50}{3}\approx16.7 and κBTc=6\kappa_B T_c=6, are moderate compared with E4=5E_4=5 but large compared with κBTb=13\kappa_B T_b=\frac{1}{3}. The numerical search for this example is summarized in Appendix A; in it we found no example meeting the same margins with βh>0.1βb\beta_h>0.1\beta_b.

An exactly rational version. The proof of Theorem 3 needs interval enclosures of exponentials only because its Boltzmann factors are irrational. As in Section 2, this can be avoided by taking integer energies and rational values of e−βUe^{-\beta U}.

Proposition 4. Let U>0U>0 and τ=1\tau=1, and define βb,βh,βc\beta_b,\beta_h,\beta_c by e−βbU=1921e^{-\beta_b U}=\frac{19}{21}, e−βhU=499500e^{-\beta_h U}=\frac{499}{500}, e−βcU=179180e^{-\beta_c U}=\frac{179}{180}, so that βh<βc<βb\beta_h<\beta_c<\beta_b. Let E=U(0,1,13,150)E=U(0,1,13,150), let dd be the conductances of Theorem 3, and aij=(1921)nidija_{ij}=(\frac{19}{21})^{n_i}d_{ij} for i≠ji\ne j, where n=(0,1,13,150)n=(0,1,13,150). Then:

  1. all four triplets fail the tests (1)–(2), on the same branches as in Theorem 3, with relative margins at least 1.04251.0425 (123), 1.0231×10−31.0231\times10^{-3} (124), 1.1358×10−31.1358\times10^{-3} (134) and 1.1986×10−31.1986\times10^{-3} (234);

  1. the nonzero eigenvalues of MM are simple, with

λ2∈[−1.2199282300,−1.2199282299],λ3∈[−6.1258309069,−6.1258309068],\lambda_2\in[-1.2199282300,-1.2199282299],\qquad\lambda_3\in[-6.1258309069,-6.1258309068],
λ4∈[−36.2712922746,−36.2712922745];\lambda_4\in[-36.2712922746,-36.2712922745];
  1. with the left eigenvector normalized by ℓ1=1\ell_1=1,

b(βh)∈[−0.1759821412,−0.1759821409],b(βc)∈[−0.1781358067,−0.1781358064],b(\beta_h)\in[-0.1759821412,-0.1759821409],\qquad b(\beta_c)\in[-0.1781358067,-0.1781358064],
∣b(βc)∣−∣b(βh)∣∣b(βh)∣∈[0.0122379,0.0122380].\frac{|b(\beta_c)|-|b(\beta_h)|}{|b(\beta_h)|}\in[0.0122379,0.0122380].

In units of kBTbk_BT_b the energies are βbE=ln⁡2119(0,1,13,150)≈(0,0.1001,1.3011,15.013)\beta_bE=\ln\frac{21}{19}(0,1,13,150)\approx(0,0.1001,1.3011,15.013), and βh/βb≈0.0200\beta_h/\beta_b\approx0.0200, βc/βb≈0.0557\beta_c/\beta_b\approx0.0557; all of these are within 0.2%0.2\% of the corresponding values (0,0.1,1.302,15)(0,0.1,1.302,15), 0.020.02, 0.05560.0556 of Theorem 3.

Proof. All Boltzmann weights (1921)ni(\frac{19}{21})^{n_i}, (499500)ni(\frac{499}{500})^{n_i}, (179180)ni(\frac{179}{180})^{n_i} are rational, hence so are all entries of MM, of π(βh)\pi(\beta_h) and π(βc)\pi(\beta_c), every κ\kappa, every residual and every relative margin; (i) is an exact rational computation. As in § 3(ii), det⁡(zI−M)=zf(z)\det(zI-M)=zf(z) with ff a cubic with rational coefficients and f(0)≠0f(0)\ne0; ff has strict sign changes at the rational endpoints of the three disjoint brackets, which gives (ii). For (iii), solve (MT−μI)ℓ=0(M^{\mathsf T}-\mu I)\ell=0 with ℓ1=1\ell_1=1 by Cramer’s rule on the rows and columns {2,3,4}\{2,3,4\}, with μ\mu ranging over the bracket of λ2\lambda_2, in rational interval arithmetic with exact endpoints; the denominator lies in [113.1948667,113.1948669][113.1948667,113.1948669], and the omitted row holds by the Schur determinant argument of § 3(iii). The amplitudes are then evaluated in the same arithmetic; both are negative, and ∣b(βc)∣−∣b(βh)∣=−ℓT(π(βc)−π(βh))∈[0.0021536654,0.0021536656]|b(\beta_c)|-|b(\beta_h)|=-\ell^{\mathsf T}(\pi(\beta_c)-\pi(\beta_h))\in[0.0021536654,0.0021536656]. No exponential is enclosed and no endpoint is rounded.

Why the triplet tests do not see the effect

The following two facts are elementary and hold for any reversible NN-level chain with λ2\lambda_2 simple; we include their proofs.

Lemma 5. If ℓ\ell is monotone in the energy (ℓ1≤⋯≤ℓN\ell_1\le\cdots\le\ell_N or ℓ1≥⋯≥ℓN\ell_1\ge\cdots\ge\ell_N), then ∣b∣|b| is strictly decreasing on [0,βb][0,\beta_b]; in particular there is no direct Mpemba effect.

Proof. ddβπ(β)i=π(β)i(⟨E⟩β−Ei)\frac{d}{d\beta}\pi(\beta)_i=\pi(\beta)_i(\langle E\rangle_\beta-E_i), so b′(β)=−Cov⁡β(ℓ,E)=−12∑i,jπ(β)iπ(β)j(ℓi−ℓj)(Ei−Ej)b'(\beta)=-\operatorname{Cov}_\beta(\ell,E)=-\frac{1}{2}\sum_{i,j}\pi(\beta)_i\pi(\beta)_j(\ell_i-\ell_j)(E_i-E_j). All terms have the same sign, and not all vanish because ℓ\ell is not constant (ℓTπ(βb)=0\ell^{\mathsf T}\pi(\beta_b)=0 and ℓ≠0\ell\ne0). So bb is strictly monotone, and b(βb)=0b(\beta_b)=0 gives the claim.

Lemma 5 is the discrete-state form of an argument already used for overdamped Langevin dynamics. There the derivative of the slow-mode overlap with respect to the initial temperature is an equilibrium covariance of the slow left eigenfunction with the potential [6 Eq. (7)], [7 Eq. (15)], and [7 Sec. III.D] applies the Fortuin–Kasteleyn–Ginibre inequality to a monotone eigenfunction and a monotone potential to exclude the effect in symmetric single wells. A decomposition of the effect into contributions of pairs of states, with factors given by differences of the second eigenfunction and of the energy, appears in the Supplemental Material of [8]. We did not find the statement of Lemma 5 for finite-state chains in these works.

Both examples have non-monotone ℓ\ell: ℓ1<0<ℓ4<ℓ3<ℓ2\ell_1<0<\ell_4<\ell_3<\ell_2 in Theorem 1 and ℓ2<ℓ3<ℓ4<0<ℓ1\ell_2<\ell_3<\ell_4<0<\ell_1 in Theorem 3. The ordering of ℓ\ell is a property of the slow eigenvector of the whole chain. The next identity shows how it depends on rates outside a given triplet. Write γm=∑j≠majm\gamma_m=\sum_{j\ne m}a_{jm} for the escape rate of level $m.

Lemma 6. Let MTℓ=λℓM^{\mathsf T}\ell=\lambda\ell and let mm be a level with γm+λ≠0\gamma_m+\lambda\ne0. Then for every i≠mi\ne m,

∑j≠i,ma~ji(ℓj−ℓi)=λ(1+amiγm+λ)ℓi,a~ji=aji+ajmamiγm+λ.\sum_{j\ne i,m}\tilde{a}_{ji}(\ell_j-\ell_i)=\lambda\left(1+\frac{a_{mi}}{\gamma_m+\lambda}\right)\ell_i,\qquad\tilde{a}_{ji}=a_{ji}+\frac{a_{jm}a_{mi}}{\gamma_m+\lambda}.

The a~\tilde{a} satisfy detailed balance with respect to π(βb)\pi(\beta_b) restricted to the levels ≠m\ne m.

Proof. Row ii of MTℓ=λℓM^{\mathsf T}\ell=\lambda\ell reads ∑j≠iaji(ℓj−ℓi)=λℓi\sum_{j\ne i}a_{ji}(\ell_j-\ell_i)=\lambda\ell_i. Row mm gives (γm+λ)ℓm=∑j≠majmℓj(\gamma_m+\lambda)\ell_m=\sum_{j\ne m}a_{jm}\ell_j, hence ℓm−ℓi=(∑j≠majm(ℓj−ℓi)−λℓi)/(γm+λ)\ell_m-\ell_i=(\sum_{j\ne m}a_{jm}(\ell_j-\ell_i)-\lambda\ell_i)/(\gamma_m+\lambda). Substituting this into row ii gives the identity. Detailed balance follows from ajmamiπi=ajmaimπm=amjaimπja_{jm}a_{mi}\pi_i=a_{jm}a_{im}\pi_m=a_{mj}a_{im}\pi_j.

For a 4-level chain and mm the level outside a triplet, Lemma 6 is a three-level eigen-equation for ℓ\ell on the triplet, but with the couplings a~\tilde{a} and a level-dependent weight on the right-hand side. The tests (1)–(2) are applied to the original rates aa, i.e. they ignore the correction ajmami/(γm+λ2)a_{jm}a_{mi}/(\gamma_m+\lambda_2). This correction depends on the slowest eigenvalue λ2\lambda_2 of the whole chain. It is large when γm\gamma_m is close to ∣λ2∣\lvert\lambda_2\rvert, and it is negative when γm<∣λ2∣\gamma_m<\lvert\lambda_2\rvert, in which case some a~\tilde{a} can be negative and the effective three-level problem need not come from a Markov generator (in Theorem 3, with m=1m=1 and λ=λ2\lambda=\lambda_2, a~34≈−2.62\tilde{a}_{34}\approx-2.62 and a~43≈−2.9×10−6\tilde{a}_{43}\approx-2.9\times10^{-6}; in Theorem 1 all six a~\tilde{a} remain positive, floating-point evaluation). At λ=0\lambda=0 the a~\tilde{a} are the rates of the chain observed only on the levels ≠m\ne m, which is again an ordinary reversible chain.

Observation (both examples). In both examples the ground level is a bottleneck: its escape rate is below the spectral gap. The test function 1{1}\mathbf{1}_{\{1\}} in the Rayleigh quotient of the reversible generator gives ∣λ2∣≤γ1/(1−π(βb)1)\lvert\lambda_2\rvert\le\gamma_1/(1-\pi(\beta_b)_1), and

  • in Theorem 1: γ1=1112+ε2+2δ<0.92<2.13≤∣λ2∣≤γ1/(1−π(βb)1)≈2.139\gamma_1=\frac{11}{12}+\frac{\varepsilon}{2}+2\delta<0.92<2.13\le\lvert\lambda_2\rvert\le\gamma_1/(1-\pi(\beta_b)_1)\approx2.139;

  • in Theorem 3: γ1∈[0.6959505683,0.6959505684]\gamma_1\in[0.6959505683,0.6959505684], γ1+λ2∈[−0.5229491277,−0.5229491246]\gamma_1+\lambda_2\in[-0.5229491277,-0.5229491246] (certified), and γ1/(1−π(βb)1)≈1.287\gamma_1/(1-\pi(\beta_b)_1)\approx1.287.

So λ2\lambda_2 nearly saturates the bound, and the slow mode is close to the mode that changes only the ground-state population: ℓ\ell is nearly constant on the excited levels. Whether ∣b∣\lvert b\rvert rises at high temperature is then decided by the small variation of ℓ\ell over {2,3,4}\{2,3,4\}, which by Lemma 6 with m=1m=1 is governed by the couplings a~\tilde{a} with γ1+λ2<0\gamma_1+\lambda_2<0. Consistently, the triplet {2,3,4}\{2,3,4\} fails its test (2) with the original rates and also with the λ=0\lambda=0 rates a~\tilde{a} (floating-point evaluation), but satisfies (2) when the rates a~\tilde{a} at λ=λ2\lambda=\lambda_2 are inserted: for Theorem 3 the formal κ\kappa is −0.00137…<1-0.00137\ldots<1 and the residual of (2) lies in [3.39436930,3.39436933]>0[3.39436930,3.39436933]>0 (certified); for Theorem 1 the residual is about +5.4×10−3+5.4\times10^{-3} relative to the largest rate (floating-point only).

This is a heuristic explanation, not a theorem: the tests of [1] are derived for genuine three-level generators, and we do not claim that the tests applied to a~\tilde{a} are necessary or sufficient for anything. What the identity does show is the quantity the triplet tests miss: the slowest eigenvalue λ2\lambda_2 of the full chain, through the factor 1/(γm+λ2)1/(\gamma_m+\lambda_2), which compares it with the escape rate of the level outside the triplet.

Verification procedure

For Theorem 1, every step above is a finite exact computation over Q\mathbb{Q}; no floating-point value is used in any decision.

  1. Build τMε\tau M_{\varepsilon} with δ=2−64\delta=2^{-64} as an exact rational matrix at ε=0\varepsilon=0 and ε=10−6\varepsilon=10^{-6}. Check column sums zero, Mq=0Mq=0, detailed balance Mijqj=MjiqiM_{ij}q_j=M_{ji}q_i and positivity of off-diagonal entries.

  1. For each triplet compute κ\kappa and both residuals exactly at the two endpoints; they are affine in ε\varepsilon, so endpoint signs cover [0,10−6][0,10^{-6}]. Require that no branch passes, with strict failure at ε=10−6\varepsilon=10^{-6}.

  1. Expand det⁡(zI−τM0)\det(zI-\tau M_0) over the 24 permutations and compare with zf(z)zf(z) at the actual δ\delta; also compare at δ=0,1,2,3,4\delta=0,1,2,3,4, which proves the identity as polynomials in δ\delta (each coefficient has degree at most 4 in δ\delta).

  1. Evaluate ff at the six rational interval endpoints and require sign changes; shift the intervals by −32⋅10−6-\frac{3}{2}\cdot10^{-6} and require disjointness.

  1. With μ2\mu_2 and ε\varepsilon as rational intervals, compute det⁡BJJ\det B_{JJ} and the Cramer quotients by outward interval arithmetic; require 0∉det⁡BJJ0\notin\det B_{JJ} and the sign pattern of ℓ\ell.

  1. Form π\pi at t=255256t=\frac{255}{256} and t=127128t=\frac{127}{128} exactly; require b(βh)>0b(\beta_h)>0, b(βc)>0b(\beta_c)>0 and b(βc)−b(βh)>0b(\beta_c)-b(\beta_h)>0 for the interval lower bounds.

The ancillary file check_four_state.py (Python standard library only, fractions.Fraction) performs steps 1–6 and writes all rational endpoints to a JSON file; its checks raise exceptions and remain active under python -O. A run with Python 3.14 returned PASS in under 0.01 s of arithmetic time and reproduced every interval stated above. The script certifies the finite arithmetic; the use of Weyl’s inequality, Cramer’s rule, the Schur determinant identity and the spectral expansion is the mathematical argument given in § 3.

For Theorem 3, the ancillary file check_robust_four_state.py (Python standard library only) reads the data from robust_witness.json and performs the interval computation described in the proof: the exponential enclosures, the triplet tests with certified relative margins, the interval characteristic polynomial and the sign changes at the given brackets, the interval Cramer solve, and the amplitudes. It also certifies the two diagnostics of § 6 (γ1+λ2<0\gamma_1+\lambda_2<0 and the sign of the residual of (2) for the triplet {2,3,4}\{2,3,4\} with the rates a~\tilde{a} at λ=λ2\lambda=\lambda_2). All decisions raise exceptions and remain active under python -O. A run with Python 3.14 returned PASS in under 0.1 s and reproduced every interval stated in Theorem 3. As an independent check, a 50-digit floating-point computation with mpmath gave values inside all these intervals; it is not part of the proof.

For Proposition 4, the ancillary file check_rational_four_state.py (Python standard library only, fractions.Fraction, no rounding) reads the integers nin_i, the three rational Boltzmann factors, the conductances and the eigenvalue brackets from rational_witness.json. It checks the generator (column sums, stationarity of π(βb)\pi(\beta_b), detailed balance, positivity), computes every κ\kappa, residual and relative margin exactly, expands det⁡(zI−M)\det(zI-M) over the 24 permutations and checks the sign changes of ff, and performs the interval Cramer solve and the amplitude evaluation of the proof. All decisions raise exceptions and remain active under python -O. A run with Python 3.14 returned PASS in under 0.1 s and reproduced every number stated in Proposition 4; a 60-digit mpmath computation gave values inside these intervals (not part of the proof). The same script, given the data of Theorem 1 at ε=10−6\varepsilon=10^{-6} with the brackets of § 3, also returns PASS; this checks a single value of ε\varepsilon only.

AI assistance disclosure

OpenAI GPT-5.6 Sol, accessed through ChatGPT and Codex, was used to find and audit the first counterexample and its exact verification, and for drafting. Anthropic Claude Opus 5.5, used through Claude Code, was used to write the note, find and certify the robust witness of Theorem 3, and for the mechanism section. Anthropic Claude Opus 5.5 (Claude Code) produced the exact rational variant of Proposition 4 and its checker check_rational_four_state.py. The author reviewed and takes responsibility for every statement and proof. AI systems are not authors of this work, and the work has not been peer reviewed.

The numerical search (numerical evidence only)

Nothing in this appendix is proved; it records how the example of Theorem 3 was found and what the search did not find. Units are βb=1\beta_b = 1, so energies are measured in units of kBTbk_B T_b; write ϱ=βh/βb\varrho= \beta_h/\beta_b and s=βb(E2−E1)s = \beta_b(E_2 - E_1). The score was the minimum of (relative margin of the worst triplet)/10−3/10^{-3} and (relative amplitude gap)/10−2/10^{-2}, where the gap is max⁡ϱβb≤βh<βc≤βb(∣b(βc)∣−∣b(βh)∣)/∣b(βc)∣\max_{\varrho\beta_b\leq\beta_h<\beta_c\leq\beta_b} (|b(\beta_c)|-|b(\beta_h)|)/|b(\beta_c)| on a grid; a branch with κ\kappa within the margin of 1 was required to fail in both lines. Eigenvectors were computed in double precision from the symmetrized generator.

  • Global search. Differential evolution over E∈[0,5]4E \in[0,5]^4 with spacings ≥0.05\geq0.05, βb∈(0.1,6]\beta_b \in(0.1,6], βh≥0.1\beta_h \geq0.1, ∣ln⁡dij∣≤8|\ln d_{ij}| \leq8 and smallest bath weight ≥10−6\geq10^{-6}, several seeds: no point was found with all four triplets failing and any Mpemba gap. Without the triplet constraints, gaps of 100% (a sign change of bb) were found at once.

  • Continuation from the example of Theorem 1 (Nelder–Mead, step-wise tightening of βbE4\beta_b E_4, ϱ\varrho and ss). Best scores found, as (worst relative triplet margin, relative gap):

βbE4≤\beta_b E_4 \leqϱ\varrhossmargingap
603×10−43 \times10^{-4}9.3×10−49.3 \times10^{-4}0.93%
600.033.8×10−53.8 \times10^{-5}0.04%
1000.071.1×10−31.1 \times10^{-3}1.1%
100.020.021.4×10−31.4 \times10^{-3}1.4%
150.020.11.2×10−31.2 \times10^{-3}1.2%
200.020.28.7×10−48.7 \times10^{-4}0.87%
200.050.0031.3×10−31.3 \times10^{-3}1.3%
200.050.11.3×10−41.3 \times10^{-4}0.13%
300.070.0013.7×10−43.7 \times10^{-4}0.37%
300.10.001none found

Table 2.

All 62 optimized continuation points with worst relative margin ≥10−4\geq10^{-4} have a ground pair that is close in energy and weakly coupled directly (d12d_{12} at most 0.05 times every other conductance), triplets 124, 134, 234 failing by margins within a factor 2 of each other, and γ1<∣λ2∣\gamma_1 < |\lambda_2|, the structure discussed in § 6. Theorem 3 is the βbE4≤15\beta_b E_4 \leq15, ϱ=0.02\varrho= 0.02, s=0.1s = 0.1 point with rounded data, rescaled to E4=5E_4 = 5.

The search is local and not exhaustive. It suggests a trade-off between the margins, the size of βbE4\beta_b E_4, the closeness of the ground pair and the hot temperature, and in particular that examples with βbEmax⁡≲6\beta_b E_{\max} \lesssim6 or βh≳0.1βb\beta_h \gtrsim0.1\beta_b that meet the margins of Theorem 3 are rare or absent; we have not proved either statement.

References

  1. [1]I. Avitan, R. Factor, and D. Gelbwaser-Klimovsky, Necessary conditions for the Markovian Mpemba effect, arXiv:2603.04567v1 [cond-mat.stat-mech] (2026). https://arxiv.org/abs/2603.04567
  2. [2]D. Gelbwaser-Klimovsky, Efficient protocol for the Markovian Mpemba effect in N-level systems, arXiv:2609.09988v1 [cond-mat.stat-mech] (2026). https://arxiv.org/abs/2609.09988
  3. [3]Z. Lu and O. Raz, Nonequilibrium thermodynamics of the Markovian Mpemba effect and its inverse. Proc. Natl. Acad. Sci. USA 114(20) (2017), 5083–5088, doi:10.1073/pnas.1701264114; preprint arXiv:1609.05271 (2016), titled Anomalous cooling and heating – the Mpemba effect and its inverse.DOI
  4. [4]I. Klich, O. Raz, O. Hirschberg, and M. Vucelja, Mpemba index and anomalous relaxation, Phys. Rev. X 9 (2019), 021060, doi:10.1103/PhysRevX.9.021060; arXiv:1711.05829.DOI
  5. [5]G. Teza, J. Bechhoefer, A. Lasanta, O. Raz, and M. Vucelja, Speedups in nonequilibrium thermal relaxation: Mpemba and related effects, Phys. Rep. 1164 (2026), 1–97, doi:10.1016/j.physrep.2025.10.009; arXiv:2502.01758.DOI
  6. [6]M. R. Walker and M. Vucelja, Mpemba effect in terms of mean first passage time, arXiv:2212.07496v4 [cond-mat.stat-mech] (2023). https://arxiv.org/abs/2212.07496
  7. [7]Y. Liu, T. Van Vu, R. Chétrite, F. van Wijland, and H. Hayakawa, Predicting the conditions for observing the Mpemba effect, arXiv:2606.03445v4 [cond-mat.stat-mech] (2026). https://arxiv.org/abs/2606.03445
  8. [8]N. Ohga, H. Hayakawa, and S. Ito, Microscopic theory of Mpemba effects and a no-Mpemba theorem for monotone many-body systems, arXiv:2410.06623v2 [cond-mat.stat-mech] (2024). https://arxiv.org/abs/2410.06623

Paper details

vlnc:W_3tp9p98x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

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

  • Physics and Astronomy

OpenAI ChatGPT GPT-5.6 Sol, OpenAI Codex GPT-5.6 Sol, Anthropic Claude Code Claude Opus 5.5

View statement
Looking for related papers…
Contents