betaFeedback

WorksPhysics and Astronomy

Optimal relaxation for Witten Lindbladians

Abstract

We prove optimal trace-norm relaxation for the one-dimensional Lindbladian associated with the Witten differential a=h∂x+V′a=h\partial_x+V' which annihilates the classical Gibbs density: if λ1(h)λ_1(h) is the first positive eigenvalue of H=a∗aH=a^*a, then the relaxation rate is γh=λ1(h)/(2h)γ_h=λ_1(h)/(2h). It applies to initial operators whose Schwartz kernels, after a Gibbs conjugation, satisfy L2 L^2 estimates for the restriction to the diagonal and for the normal derivative. An appendix by Chat GPT 6 presents a stronger result specialised to positive initial data.

Statement of the result and motivation

With motivation and references provided below, we study the evolution governed by the following superoperator:

LT=1h(aTa∗−12(a∗aT+Ta∗a)),a:=h∂x+V′(x).(1)\mathcal{L}T=\frac{1}{h}\left(aTa^*-\frac{1}{2}\left(a^*aT+Ta^*a\right)\right),\qquad a:=h\partial_x+V'(x). \tag*{(1)}

Since the jump operator, aa, is given by a Witten differential [25] (and H:=a∗aH:=a^*a is the scalar Witten Laplacian), we refer to L\mathcal{L} as the Witten Lindbladian. The assumptions on VV are given in (1.8). By adding an irrelevant hh-dependent constant we assume that

∫Re−2V(x)/h dx=1.(2)\int_{\mathbb{R}}e^{-2V(x)/h}\,dx=1. \tag*{(2)}

For initial conditions for the Lindblad evolution, we consider T∈L1(L2(R))T\in L_1(L^2(\mathbb{R})), the space of trace-class operators on L2(R)L^2(\mathbb{R}),

Tu(x)=∫Re−(V(x)+V(y))/hr(x,y)u(y) dy,r∈C1,Tu(x)=\int_{\mathbb{R}}e^{-(V(x)+V(y))/h}r(x,y)u(y)\,dy,\qquad r\in C^1,

where, with dμ(y):=e−2V(y)/h dyd\mu(y):=e^{-2V(y)/h}\,dy, rr satisfies

r∣{(x,x)}∈L2(dμ),∣∂yr(x,y)∣≤G0(y)∈L2(dμ).(3)r|_{\{(x,x)\}}\in L^2(d\mu),\qquad|\partial_y r(x,y)|\leq G_0(y)\in L^2(d\mu). \tag*{(3)}

Theorem 1. *There is a constant Ch,V>0C_{h,V}>0 such that for TT above with tr⁡T=1\operatorname{tr}T=1, we have for t≥0t\geq0 and Π:=∣e−V/h⟩⟨e−V/h∣\Pi:=|e^{-V/h}\rangle\langle e^{-V/h}|,

∥etLT−Π∥L1≤Ch,Ve−γht(∥T∥L1+∥r∣{(x,x)}−1∥L2(μ)+∥G0∥L2(μ)),(4)\left\|e^{t\mathcal{L}}T-\Pi\right\|_{L_1}\leq C_{h,V}e^{-\gamma_h t}\left(\|T\|_{L_1}+\|r|_{\{(x,x)\}}-1\|_{L^2(\mu)}+\|G_0\|_{L^2(\mu)}\right), \tag*{(4)}

where

γh:=λ1(h)2h>0,λ1(h):=inf⁡(Spec⁡(H)∖{0}),(5)\gamma_h:=\frac{\lambda_1(h)}{2h}>0,\qquad\lambda_1(h):=\inf\left(\operatorname{Spec}(H)\setminus\{0\}\right), \tag*{(5)}

and this exponent is optimal.*

Remarks. 1. The regularity assumptions on rr are chosen to keep the proof elementary. The pointwise differentiability in yy is likely to be replaceable by an appropriate absolute-continuity condition. We do not pursue this endpoint question here, but some restrictions on TT are always needed, as L\mathcal{L} does not have a spectral gap on L1L_1 (easily seen in the case of V(x)=x2V(x)=x^2. The more substantial problem is the extension to higher dimensions: our methods are very much one dimensional.

  1. Tracing the constants in the proof, one may take

Ch,V≤max⁡(2eγh,1,2Ah,V(1+γh)1/2)V(V),C_{h,V} \le\max\left(2e^{\gamma_h},1,2A_{h,V}(1+\gamma_h)^{1/2}\right)\mathcal{V}(V),

where (with SS defined in (2.3))

Ah,V=∥⟨x⟩(1−S)−1/2∥L2(μ)→L2(μ),A_{h,V}=\left\|\langle x\rangle(1-S)^{-1/2}\right\|_{L^2(\mu)\to L^2(\mu)},

and V(V)\mathcal{V}(V) was defined in (3.2). For 0<h≤h00<h\le h_0, the assumptions on VV give Ah,V=OV(1)A_{h,V}=\mathcal{O}_V(1) and γh=OV(1)\gamma_h=\mathcal{O}_V(1). Thus the hh-dependence provided by the proof is contained in V(V)\mathcal{V}(V). It is bounded as h→0h\to0 in the one-well case, while for a nondegenerate double well of barrier height S0S_0 the proof gives

Ch,V=OV(heS0/h)=OV(hγh).C_{h,V}=\mathcal{O}_V\left(\sqrt{h}e^{S_0/h}\right)=\mathcal{O}_V\left(\sqrt{\frac{h}{\gamma_h}}\right).

We do not claim that this is the optimal common prefactor in the estimate.

  1. When, in the theorem, T≥0T\ge0, we can replace the hypotheses by r∣{(x,x)}∈L∞r|_{\{(x,x)\}}\in L^\infty and the assumption that tr⁡HT=∥aT1/2∥22<∞\operatorname{tr}HT=\|aT^{1/2}\|_2^2<\infty. The result then follows from the following inequality proved at the of this introduction:

∣∂yr(x,y)∣≤G0(y),∥G0∥L2(μ)≤h−1∥r∣{(x,x)}∥L∞∥aT1/2∥L2.(6)|\partial_y r(x,y)|\le G_0(y),\qquad\|G_0\|_{L^2(\mu)}\le h^{-1}\|r|_{\{(x,x)\}}\|_{L^\infty}\|aT^{1/2}\|_{L^2}. \tag*{(6)}

At our request, Chat GPT 6 upgraded this by weakening the hypothesis to r∣{(x,x)}∈Lq(dμ)r|_{\{(x,x)\}}\in L^q(d\mu), q>1q>1. That result, Theorem 2, is presented in the appendix.

When V(x)=x2/2V(x)=x^2/2, (1.1) is, after the standard normalization of the annihilation operator, the one-mode pure-loss channel, with transmissivity e−2te^{-2t}; see [15, 24, 6, 7]. The general potential replaces the harmonic-oscillator annihilation operator by the Witten differential h∂x+V′h\partial_x+V'. Related first-order factorizations for anharmonic oscillators appear in [9, 21, 22].

Lindblad generators [13, 19] provide the standard Markovian framework for dissipative quantum dynamics. Recently, they have also appeared in quantum Gibbs sampling; see [4, 8] for finite-dimensional constructions and [11, 2] for related infinite-dimensional work. A model based on the simpler equation considered here appeared in Leng–Ding–Chen–Lin [17], where Witten differentials were used as Lindblad jump operators. The stationary state Π\Pi is obtained from the square root e−V/he^{-V/h} of the Gibbs density. The trace norm is natural here since

∣tr⁡((T−Π)O)∣≤∥T−Π∥L1∥O∥\left|\operatorname{tr}((T-\Pi)O)\right|\le\|T-\Pi\|_{L^1}\|O\|

for every bounded observable OO.

The diagonal evolution is governed by the Witten Laplacian HH. The theorem says that the full Lindblad evolution, including the off-diagonal part of the kernel, has the same optimal relaxation rate γh\gamma_h. For a self-contained one-dimensional account of the semiclassical asymptotics of λ1(h)\lambda_1(h), and references, see [20]. If VV has exactly one critical point, a unique nondegenerate minimum x0x_0, harmonic approximation gives

γh=V′′(x0)+O(h).(7)\gamma_h = V''(x_0) + \mathcal{O}(h). \tag*{(7)}

For a multiwell potential, γh\gamma_h is exponentially small and is governed by the Eyring–Kramers law. Thus the result includes the metastable regime. Related semiclassical Lindblad–Fokker–Planck constructions are discussed in [16, 10, 18, 23].

We now impose, in addition to the normalization (1.2),

V(k)(x)=O(⟨x⟩),k≥1,V′′(x)≥−C0,xV′(x)≥x2/C1−C1.(8)V^{(k)}(x) = \mathcal{O}(\langle x\rangle),\quad k \ge1,\qquad V''(x) \ge-C_0,\qquad xV'(x) \ge x^2/C_1 - C_1. \tag*{(8)}

These assumptions are chosen for a short proof rather than for maximal generality. Since −H/(2h)-H/(2h) generates a contraction semigroup on L2(R)L^2(\mathbb{R}), the Davies construction gives the trace-class contraction semigroup used below; see [10], §1, Remark. The remaining bounds are used in the maximum-principle, weighted, and tail estimates. We write

H:=a∗a=−h2∂x2+∣V′∣2−hV′′,ν:=e−V/h,Π:=∣ν⟩⟨ν∣.(9)H := a^*a = -h^2\partial_x^2 + |V'|^2 - hV'',\qquad\nu:= e^{-V/h},\qquad\Pi:= |\nu\rangle\langle\nu|. \tag*{(9)}

Then, aν=0a\nu= 0, ker⁡H=Cν\ker H = \mathbb{C}\nu, and, from (1.2), ∥ν∥L2=1\|\nu\|_{L^2} = 1. The assumptions (1.8) imply that HH has compact resolvent and hence that λ1(h)>0\lambda_1(h) > 0.

Proof of (1.6). By the spectral theorem, T=∑j≥0λj∣φj⟩⟨φj∣T = \sum_{j\ge0}\lambda_j|\varphi_j\rangle\langle\varphi_j|, λj≥0\lambda_j \ge0, ∑j≥0λj=1\sum_{j\ge0}\lambda_j = 1, where the φj\varphi_j are orthonormal in L2(R)L^2(\mathbb{R}). Writing φj=νfj\varphi_j = \nu f_j we introduce the ℓ2\ell^2-valued function F(x):={λjfj(x)}j∈N∈L2(R;ℓ2(N),dμ)F(x) := \{\sqrt{\lambda_j}f_j(x)\}_{j\in\mathbb{N}} \in L^2(\mathbb{R};\ell^2(\mathbb{N}),d\mu). It follows that

r(x,y)=∑j≥0λjfj(x)fj(y)‾=⟨F(x),F(y)⟩ℓ2,r(x,x)=∥F(x)∥ℓ22.(10)r(x,y) = \sum_{j\ge0}\lambda_j f_j(x)\overline{f_j(y)} = \langle F(x),F(y)\rangle_{\ell^2},\qquad r(x,x) = \|F(x)\|_{\ell^2}^2. \tag*{(10)}

The definition (1.9), the spectral decomposition above, and the fact that a(νf)=hνf′a(\nu f)=h\nu f' give

tr⁡HT=∥aT12∥L22=∑j=0∞λj∥aφj∥L2(R)2=∑j=0∞λj∥hνfj′∥L2(R)2=h2∫R∥F′(x)∥ℓ2(N)2 dμ(x).(11)\begin{aligned} \operatorname{tr} HT = \|aT^{\frac{1}{2}}\|_{L^2}^2 = \sum_{j=0}^{\infty}\lambda_j\|a\varphi_j\|_{L^2(\mathbb{R})}^2 \\ &= \sum_{j=0}^{\infty}\lambda_j\|h\nu f'_j\|_{L^2(\mathbb{R})}^2 = h^2\int_{\mathbb{R}}\|F'(x)\|_{\ell^2(\mathbb{N})}^2\,d\mu(x). \tag*{(11)} \end{aligned}

Differentiating (1.10) in the weak sense gives ∂yr(x,y)=⟨F(x),F′(y)⟩ℓ2\partial_y r(x,y) = \langle F(x),F'(y)\rangle_{\ell^2}. Hence, by the Cauchy–Schwarz inequality and (1.10),

∣∂yr(x,y)∣2≤r(x,x)∥F′(y)∥ℓ22≤∥r∣{(x,x)}∥L∞∥F′(y)∥ℓ22.(12)|\partial_y r(x,y)|^2 \le r(x,x)\|F'(y)\|_{\ell^2}^2 \le\|r|_{\{(x,x)\}}\|_{L^\infty}\|F'(y)\|_{\ell^2}^2. \tag*{(12)}

This and (1.11) proves (1.6). □

Acknowledgements. This note grew out of discussions between the authors and ChatGPT 5.6 about improving an earlier probabilistic argument with a non-optimal decay rate γh/2\gamma_h/2 [23]. In particular, the key Volterra factorisation argument was suggested by ChatGPT and developed to a substantial extent through these discussions. MZ gratefully acknowledges partial support from the Simons Foundation through Targeted Grant Award No. 896630, “Moiré Materials Magic”.

The conjugated kernel and the maximum principle

In the language of the theorem in §1, let T0=TT_0=T and Tt=etLT0T_t=e^{tL}T_0. Using (1.9), write

Kt(x,y)=ν(x)ν(y)rt(x,y).(13)K_t(x,y)=\nu(x)\nu(y)r_t(x,y). \tag*{(13)}

The identities

a(νf)=hνf′,a∗a(νf)=hν(2V′f′−hf′′)a(\nu f)=h\nu f',\qquad a^*a(\nu f)=h\nu(2V'f'-hf'')

give the following simple

Lemma 1. The conjugated kernel satisfies

∂trt=Mrt,M=12h(∂x+∂y)2−V′(x)∂x−V′(y)∂y.(14)\partial_t r_t=Mr_t,\qquad M=\tfrac{1}{2}h(\partial_x+\partial_y)^2-V'(x)\partial_x-V'(y)\partial_y. \tag*{(14)}

If Rr(x):=r(x,x)Rr(x):=r(x,x) and

Q=12h∂x2−V′∂x,S=Q−V′′,(15)Q=\tfrac{1}{2}h\partial_x^2-V'\partial_x,\qquad S=Q-V'', \tag*{(15)}

then

RM=QR,∂yM=(M−V′′(y))∂y.(16)RM=QR,\qquad\partial_yM=(M-V''(y))\partial_y. \tag*{(16)}

One striking consequence of (2.4) is

rt(x,x)=etQr0(x,x).(17)r_t(x,x)=e^{tQ}r_0(x,x). \tag*{(17)}

Multiplication by ν\nu gives the unitary equivalences of (unbounded) operators on L2(dμ)L^2(d\mu) (on the left) with operators on L2(dx)L^2(dx) (on the right):

Q≃−a∗a2h,S≃−aa∗2h.(18)Q\simeq-\frac{a^*a}{2h},\qquad S\simeq-\frac{aa^*}{2h}. \tag*{(18)}

The operator aa∗aa^* has no zero mode, and its spectrum agrees with the nonzero spectrum of H=a∗aH=a^*a. Hence

∥etQf−μ(f)∥L2(μ)≤e−γht∥f−μ(f)∥L2(μ),(19)\lVert e^{tQ}f-\mu(f)\rVert_{L^2(\mu)}\le e^{-\gamma_h t}\lVert f-\mu(f)\rVert_{L^2(\mu)}, \tag*{(19)}
∥etSg∥L2(μ)≤e−γht∥g∥L2(μ).(20)\lVert e^{tS}g\rVert_{L^2(\mu)}\le e^{-\gamma_h t}\lVert g\rVert_{L^2(\mu)}. \tag*{(20)}

Since ∫r0(x,x)dμ(x)=tr⁡T0=1\int r_0(x,x)d\mu(x)=\operatorname{tr}T_0=1, this gives

∥rt(x,x)−1∥L2(μ)≤e−γht∥r0(x,x)−1∥L2(μ).(21)\lVert r_t(x,x)-1\rVert_{L^2(\mu)}\le e^{-\gamma_h t}\lVert r_0(x,x)-1\rVert_{L^2(\mu)}. \tag*{(21)}

For the whole-space comparison we will use the second condition in (1.3) and the maximum principle for degenerate heat equations. Thus we start with

Lemma 2. Let w∈C2([0,T]×R2;R)w \in C^2([0,T]\times\mathbb{R}^2;\mathbb{R}) have at most polynomial growth, and satisfy

(∂t−M+V′′(y))w≥0,w(0,x,y)≥0.(22)(\partial_t-M+V^{\prime\prime}(y))w \ge0,\qquad w(0,x,y)\ge0. \tag*{(22)}

Then w(t,x,y)≥0w(t,x,y)\ge0 for (t,x,y)∈[0,T]×R2(t,x,y)\in[0,T]\times\mathbb{R}^2.

Proof. For C0C_0 in (1.8) put

z=e−C0tw,q(y)=−V′′(y)−C0≤0,P=∂t−M−q(y).z=e^{-C_0t}w,\qquad q(y)=-V^{\prime\prime}(y)-C_0\le0,\qquad P=\partial_t-M-q(y).

Then (2.10) gives Pz≥0Pz\ge0. Let ρ(x,y)=1+x2+y2\rho(x,y)=1+x^2+y^2. For every integer k≥1k\ge1, (1.8) and direct differentiation give Mρk≤CkρkM\rho^k\le C_k\rho^k (see (2.2) for the definition of MM).

Choose mm so that ∣z(t,x,y)∣≤CTρ(x,y)m|z(t,x,y)|\le C_T\rho(x,y)^m on [0,T]×R2[0,T]\times\mathbb{R}^2 and put

φ(t,x,y)=eAtρ(x,y)m+1.\varphi(t,x,y)=e^{At}\rho(x,y)^{m+1}.

Since q≤0q\le0, choosing A>Cm+1A>C_{m+1} gives Pφ>0P\varphi>0.

For ε>0\varepsilon>0, set v=z+εφv=z+\varepsilon\varphi. It is positive at t=0t=0, and it tends to +∞+\infty as x2+y2→∞x^2+y^2\to\infty, uniformly for 0≤t≤T0\le t\le T. If it first vanished at (t0,x0,y0)(t_0,x_0,y_0), then

∂tv≤0,∂xv=∂yv=0,∇x,y2v≥0.\partial_t v\le0,\qquad\partial_xv=\partial_yv=0,\qquad\nabla^2_{x,y}v\ge0.

In particular, (∂x+∂y)2v≥0(\partial_x+\partial_y)^2v\ge0. At the contact point the drift and zeroth-order terms vanish, and therefore

Pv=∂tv−12h(∂x+∂y)2v≤0.Pv=\partial_tv-\frac{1}{2}h(\partial_x+\partial_y)^2v\le0.

But Pv=Pz+εPφ>0Pv=Pz+\varepsilon P\varphi>0, a contradiction. Thus z+εφ>0z+\varepsilon\varphi>0 and letting ε→0+\varepsilon\to0+ proves the claim. □

The next proposition propagates the estimate (1.3) on the initial condition:

Proposition 3. For all t≥0t\ge0, and for SS defined in (2.3),

∣∂yrt(x,y)∣≤(etSG0)(y),x,y∈R.(23)|\partial_yr_t(x,y)|\le(e^{tS}G_0)(y),\qquad x,y\in\mathbb{R}. \tag*{(23)}

Proof. Put ut=∂yrtu_t=\partial_yr_t. By (2.4),

∂tut=(M−V′′(y))ut.\partial_tu_t=(M-V^{\prime\prime}(y))u_t.

Let gt=etSG0g_t=e^{tS}G_0 and regard gt(y)g_t(y) as a function of (x,y)(x,y). Since MM reduces to QQ on functions of yy alone, it satisfies the same equation. For bounded smooth data, Lemma 2, applied for every θ∈R\theta\in\mathbb{R} to

gt(y)−Re⁡(e−iθut(x,y)),g_t(y)-\operatorname{Re}(e^{-i\theta}u_t(x,y)),

gives (2.11). The same lemma shows that the closed semigroup generated by M−V′′(y)M-V^{\prime\prime}(y) is positivity preserving. Since u0u_0 and G0(y)G_0(y) belong to L2(R2;dμ(x)dμ(y))L^2(\mathbb{R}^2;d\mu(x)d\mu(y)), the order inequalities

−G0(y)≤Re⁡(e−iθu0(x,y))≤G0(y)-G_0(y)\le\operatorname{Re}(e^{-i\theta}u_0(x,y))\le G_0(y)

pass to the L2L^2 closure. This proves the result for the stated G0∈L2(μ)G_0\in L^2(\mu). □

For quadratic confinement the Volterra estimate below requires one spatial weight. It is obtained using smoothing for positive times:

Lemma 4. There is Ch,V>0C_{h,V} > 0 such that, for g∈L2(μ)g \in L^2(\mu) and t>0t > 0,

∥⟨x⟩etSg∥L2(μ)≤Ch,V(1+t−1/2)e−γht∥g∥L2(μ).(24)\left\|\langle x\rangle e^{tS}g\right\|_{L^2(\mu)} \le C_{h,V}(1+t^{-1/2})e^{-\gamma_h t}\|g\|_{L^2(\mu)}. \tag*{(24)}

Proof. As stated in (2.6), 1−S1-S is unitarily equivalent to 1+aa∗/(2h)1+aa^*/(2h). The assumptions (1.8) imply ∣V′(x)∣2≥cx2−C|V'(x)|^2 \ge cx^2-C, while

aa∗=−h2∂x2+∣V′∣2+hV′′.aa^*=-h^2\partial_x^2+|V'|^2+hV''.

Hence its quadratic form gives

∥⟨x⟩f∥L2(μ)≤Ch,V∥(1−S)1/2f∥L2(μ).(25)\|\langle x\rangle f\|_{L^2(\mu)} \le C_{h,V}\|(1-S)^{1/2}f\|_{L^2(\mu)}. \tag*{(25)}

The spectral theorem and the fact that Spec⁡(S)⊂(−∞,−γh]\operatorname{Spec}(S)\subset(-\infty,-\gamma_h] give

∥(1−S)1/2etS∥L2(μ)→L2(μ)≤Ch,V(1+t−1/2)e−γht.\|(1-S)^{1/2}e^{tS}\|_{L^2(\mu)\to L^2(\mu)}\le C_{h,V}(1+t^{-1/2})e^{-\gamma_h t}.

Combining the two estimates proves the lemma. □

The Volterra factorisation and proof of Theorem 1

Define

F−(s)=∫−∞se−2V(x)/h dx,F+(s)=∫s∞e−2V(x)/h dx,(26)F_-(s)=\int_{-\infty}^{s}e^{-2V(x)/h}\,dx,\qquad F_+(s)=\int_s^\infty e^{-2V(x)/h}\,dx, \tag*{(26)}

and

V(V)2=∫RF−(s)F+(s)e2V(s)/h⟨s⟩−2 ds.(27)\mathcal{V}(V)^2=\int_{\mathbb{R}}F_-(s)F_+(s)e^{2V(s)/h}\langle s\rangle^{-2}\,ds. \tag*{(27)}

Lemma 5. For VV satisfying (1.8), V(V)<∞\mathcal{V}(V)<\infty.

Proof. For ss sufficiently large, (1.8) gives V′(x)≥cxV'(x)\ge cx for x≥sx\ge s. Therefore

F+(s)≤Ch,Vs−1e−2V(s)/h.F_+(s)\le C_{h,V}s^{-1}e^{-2V(s)/h}.

Since F−(s)≤1F_-(s)\le1, the integrand in (3.2) is O(s−3)O(s^{-3}) as s→+∞s\to+\infty. The argument for s→−∞s\to-\infty is identical. □

The next lemma provides the crucial Volterra trace factorisation inspired by [12]:

Lemma 6. Suppose that ∣u(x,s)∣≤v(s)|u(x,s)|\le v(s) and ⟨s⟩v∈L2(μ)\langle s\rangle v\in L^2(\mu). Then the operator with kernel

R(x,y):=ν(x)ν(y)∫xyu(x,s) ds(28)R(x,y):=\nu(x)\nu(y)\int_x^y u(x,s)\,ds \tag*{(28)}

is of trace class and

∥R∥L1≤2V(V)∥⟨s⟩v∥L2(μ).(29)\|R\|_{\mathcal{L}_1}\le2\mathcal{V}(V)\|\langle s\rangle v\|_{L^2(\mu)}. \tag*{(29)}

Proof. We write

R=R++R−,R±(x,y):=R(x,y)1±x<±y,R=R_++R_-,\qquad R_\pm(x,y):=R(x,y)1_{\pm x<\pm y},

and for the factorisation of R+R_+, we define

w+(s)=F−(s)e2V(s)/h⟨s⟩−2>0.w_+(s)=F_-(s)e^{2V(s)/h}\langle s\rangle^{-2}>0.

We then put

A+(x,s):=ν(x)u(x,s)1{x<s}w+(s)−1/2,B+(s,y):=w+(s)1/2ν(y)1{s<y},A_{+}(x,s) := \nu(x)u(x,s)1_{\{x<s\}}w_{+}(s)^{-1/2}, \qquad B_{+}(s,y) := w_{+}(s)^{1/2}\nu(y)1_{\{s<y\}},

so that (identifying integral kernels with operators), R+=A+B+R_{+}=A_{+}B_{+}. We now have

∥A+∥L22≤∫w+(s)−1v(s)2F−(s) ds=∥⟨s⟩v∥L2(μ)2.(30)\lVert A_{+}\rVert_{L^{2}}^{2} \le\int w_{+}(s)^{-1}v(s)^{2}F_{-}(s)\,ds = \lVert\langle s\rangle v\rVert_{L^{2}(\mu)}^{2}. \tag*{(30)}
∥B+∥L22=∫w+(s)F+(s) ds=V(V)2.(31)\lVert B_{+}\rVert_{L^{2}}^{2} = \int w_{+}(s)F_{+}(s)\,ds = \mathcal{V}(V)^{2}. \tag*{(31)}

This gives ∥R+∥L1≤V(V)∥⟨s⟩v∥L2(μ)\lVert R_{+}\rVert_{L^{1}} \le\mathcal{V}(V)\lVert\langle s\rangle v\rVert_{L^{2}(\mu)}. For R−R_{-}, put w−(s)=F+(s)e2V(s)/h⟨s⟩−2w_{-}(s)=F_{+}(s)e^{2\mathcal{V}(s)/h}\langle s\rangle^{-2} and

A−(x,s):=ν(x)u(x,s)1{s<x}w−(s)−1/2,B−(s,y):=w−(s)1/2ν(y)1{y<s}.A_{-}(x,s) := \nu(x)u(x,s)1_{\{s<x\}}w_{-}(s)^{-1/2}, \qquad B_{-}(s,y) := w_{-}(s)^{1/2}\nu(y)1_{\{y<s\}}.

Then R−=−A−B−R_{-}=-A_{-}B_{-}, and the same estimates give ∥R−∥L1≤V(V)∥⟨s⟩v∥L2(μ)\lVert R_{-}\rVert_{L^{1}} \le\mathcal{V}(V)\lVert\langle s\rangle v\rVert_{L^{2}(\mu)}. Together with the estimate for R+R_{+} this gives (29).

Proof of the Theorem 1. Put Bt=Tt−ΠB_t=T_t-\Pi. For t>0t>0, split its kernel as

Bt(x,y)=ν(x)ν(y)(rt(x,x)−1)+ν(x)ν(y)(rt(x,y)−rt(x,x))=:Dt(x,y)+Rt(x,y).(32)\begin{aligned} B_t(x,y) &= \nu(x)\nu(y)(r_t(x,x)-1)+\nu(x)\nu(y)(r_t(x,y)-r_t(x,x)) \\ &=:D_t(x,y)+R_t(x,y). \tag*{(32)} \end{aligned}

The first term has rank one, and (21) gives

∥Dt∥L1≤e−γht∥r0(x,x)−1∥L2(μ).(33)\lVert D_t\rVert_{L^{1}} \le e^{-\gamma h t}\lVert r_0(x,x)-1\rVert_{L^{2}(\mu)}. \tag*{(33)}

For the second term,

rt(x,y)−rt(x,x)=∫xy∂yrt(x,s) ds.r_t(x,y)-r_t(x,x)=\int_x^y \partial_y r_t(x,s)\,ds.

By Proposition 3, the integrand is dominated by vt=etSG0v_t=e^{tS}G_0. For t≥1t\ge1, Lemmas 4 and 6 give

∥Rt∥L1≤Ch,Ve−γht∥G0∥L2(μ).(34)\lVert R_t\rVert_{L^{1}} \le C_{h,V}e^{-\gamma h t}\lVert G_0\rVert_{L^{2}(\mu)}. \tag*{(34)}

This proves (1.4) for t≥1t\ge1. For 0≤t≤10\le t\le1, stationarity of Π\Pi and trace-norm contractivity give

∥Bt∥L1≤∥T0−Π∥L1≤∥T0∥L1+1,\lVert B_t\rVert_{L^{1}} \le\lVert T_0-\Pi\rVert_{L^{1}} \le\lVert T_0\rVert_{L^{1}}+1,

and the short interval is absorbed by enlarging Ch,VC_{h,V}.

For optimality, let Hφ1=λ1(h)φ1H\varphi_1=\lambda_1(h)\varphi_1, with φ1⊥ν\varphi_1\perp\nu, and write φ1=νf1\varphi_1=\nu f_1. Then f1∈L2(μ)f_1\in L^{2}(\mu) and f1′∈L2(μ)f_1'\in L^{2}(\mu). To see the second claim we note that

h2∥f1′∥L2(μ)2=∥a(νf1)∥L2(dx)2=⟨Hφ1,φ1⟩L2(dx)=λ1(h)∥φ1∥L2(dx)2<∞.h^{2}\lVert f_1'\rVert_{L^{2}(\mu)}^{2}=\lVert a(\nu f_1)\rVert_{L^{2}(dx)}^{2}=\langle H\varphi_1,\varphi_1\rangle_{L^{2}(dx)}=\lambda_1(h)\lVert\varphi_1\rVert_{L^{2}(dx)}^{2}<\infty.

This shows that operator

T0=Π+∣ν⟩⟨φ1∣,tr⁡T0=1+⟨ν,φ1⟩L2(R)=1,T_0=\Pi+|\nu\rangle\langle\varphi_1|,\qquad\operatorname{tr}T_0=1+\langle\nu,\varphi_1\rangle_{L^{2}(\mathbb{R})}=1,

has conjugated kernel r0(x,y)=1+f1(y)r_0(x,y)=1+f_1(y), so that r0(x,x)−1=f1(x)∈L2(μ)r_0(x,x)-1=f_1(x)\in L^{2}(\mu) and

∣∂yr0(x,y)∣=∣f1′(y)∣.|\partial_y r_0(x,y)|=|f_1'(y)|.

Hence it satisfies the hypotheses of the theorem, and

L(∣ν⟩⟨φ1∣)=−γh∣ν⟩⟨φ1∣.\mathcal{L}(|\nu\rangle\langle\varphi_1|)=-\gamma_h|\nu\rangle\langle\varphi_1|.

Thus no larger uniform exponent is possible. □

Appendix by ChatGPT 6

The following strengthens Remark 3 in §1, under the same assumptions (1.2) and (1.8). Function norms ∥⋅∥p\|\cdot\|_p are in Lp(μ)L^p(\mu), and inner products are linear in the first factor.

Theorem 2. Let T≥0T \ge0, tr⁡T=1\operatorname{tr} T = 1, and let d(x)=r(x,x)d(x) = r(x,x) be its conjugated diagonal, as in (1.10). If

E(T):=tr⁡HT=∥aT1/2∥L22<∞andd∈Lq(dμ),1<q<∞,E(T) := \operatorname{tr} HT = \|aT^{1/2}\|_{L^2}^2 < \infty\quad\text{and}\quad d \in L^q(d\mu), \quad1 < q < \infty,

then

∥etLT−Π∥L1≤Ch,V,qe−γht(∥d∥qE(T)λ1(h))1/2,t≥0.(35)\|e^{t\mathcal{L}}T-\Pi\|_{L^1} \le C_{h,V,q}e^{-\gamma_h t}\left(\frac{\|d\|_q E(T)}{\lambda_1(h)}\right)^{1/2}, \qquad t \ge0. \tag*{(35)}

For d∈L∞(μ)d \in L^\infty(\mu), replace ∥d∥q\|d\|_q by ∥d∥∞\|d\|_\infty; the estimate then holds with an absolute constant.

Proof. Let F(x)={λjfj(x)}jF(x)=\{\sqrt{\lambda_j}f_j(x)\}_j be the ℓ2\ell^2-valued function introduced in the proof of (1.6). Equations (1.10)–(1.11) give

r(x,y)=⟨F(x),F(y)⟩,d(x)=∥F(x)∥2,h2∥F′∥22=E(T).r(x,y)=\langle F(x),F(y)\rangle,\qquad d(x)=\|F(x)\|^2,\qquad h^2\|F'\|_2^2=E(T).

We first suppose that F∈C∞(R;CN)F\in C^\infty(\mathbb{R};\mathbb{C}^N) and ∥F∥2=1\|F\|_2=1; the approximation is given below.

For any positive trace-one operator AA, the operator A−ΠA-\Pi has at most one negative eigenvalue, since its quadratic form is nonnegative on ν⊥\nu^\perp. Its trace is zero, so

∥A−Π∥L1=2∥A−Π∥≤2∥A−Π∥L2;(36)\|A-\Pi\|_{L^1}=2\|A-\Pi\|\le2\|A-\Pi\|_{\mathcal{L}_2}; \tag*{(36)}

see also [5]. It therefore suffices to estimate the Hilbert–Schmidt norm. To this end, put

b(x)=∫0x2∥F(s)∥∥F′(s)∥ ds,k(x,y)=r(x,y)−12(d(x)+d(y)).(37)b(x)=\int_0^x 2\|F(s)\|\|F'(s)\|\,ds,\qquad k(x,y)=r(x,y)-\frac{1}{2}(d(x)+d(y)). \tag*{(37)}

Then bb is bounded and nondecreasing, and

∣k(x,y)∣≤∣b(x)−b(y)∣,∣d(x)−d(y)∣≤∣b(x)−b(y)∣.(38)|k(x,y)|\le|b(x)-b(y)|,\qquad|d(x)-d(y)|\le|b(x)-b(y)|. \tag*{(38)}

Indeed, for complex Hilbert-space vectors u,vu,v, with α=∥u∥\alpha=\|u\|, β=∥v∥\beta=\|v\|,

∣⟨u,v⟩−12(α2+β2)∣≤12(α+β)∥u−v∥≤∥αu−βv∥.\left|\langle u,v\rangle-\frac{1}{2}(\alpha^2+\beta^2)\right|\le\frac{1}{2}(\alpha+\beta)\|u-v\|\le\|\alpha u-\beta v\|.

The first inequality follows by Cauchy–Schwarz, the second by expanding squared norms. Apply them to u=F(x)u=F(x), v=F(y)v=F(y) and use ∥(∥F∥F′)′∥≤2∥F∥∥F′∥=b′\|(\|F\|F')'\|\le2\|F\|\|F'\|=b'; also ∣d′∣≤b′|d'|\le b'.

For p=2q/(q+1)∈(1,2)p=2q/(q+1)\in(1,2), Hölder also gives

∥b′∥p≤2∥F∥2q∥F′∥2=2h∥d∥qE(T).(39)\|b'\|_p\le2\|F\|_{2q}\|F'\|_2=\frac{2}{h}\sqrt{\|d\|_qE(T)}. \tag*{(39)}

To propagate (A.4), use MM, QQ, SS from (2.2)–(2.3) and put Pt=etQP_t=e^{tQ}, dt=Ptdd_t=P_td, bt=Ptbb_t=P_tb, kt=rt−12(dt(x)+dt(y))k_t=r_t-\frac{1}{2}(d_t(x)+d_t(y)); dt(x)=rt(x,x)d_t(x)=r_t(x,x) by (2.5). On x<yx<y, both ktk_t and bt(y)−bt(x)b_t(y)-b_t(x) solve ∂tu=Mu\partial_t u = Mu and vanish on x=yx=y. The barrier proof of Lemma 2, with the zeroth-order term omitted, applied to bt(y)−bt(x)−Re⁡(eiθkt)b_t(y)-b_t(x)-\operatorname{Re}(e^{i\theta}k_t) for every θ\theta, propagates the first bound. Apply the same argument to dt(y)−dt(x)d_t(y)-d_t(x) for the second bound, and reflect for x>yx>y. Smooth majorants converging uniformly to b′b' justify comparison when b′b' is not smooth.

Set σt=∥bt−μ(bt)∥2\sigma_t=\lVert b_t-\mu(b_t)\rVert_2. Using 2∥f−μ(f)∥22=∬∣f(x)−f(y)∣2 dμ(x)dμ(y)2\lVert f-\mu(f)\rVert_2^2=\iint\lvert f(x)-f(y)\rvert^2\,d\mu(x)d\mu(y) gives

∥dt−1∥2≤σt,∥kt∥L2(μ⊗μ)≤2 σt.\lVert d_t-1\rVert_2\leq\sigma_t,\qquad\lVert k_t\rVert_{L^2(\mu\otimes\mu)}\leq\sqrt{2}\,\sigma_t.

The L2(μ⊗μ)L^2(\mu\otimes\mu) norm of 12(dt(x)+dt(y))−1\frac{1}{2}(d_t(x)+d_t(y))-1 is ∥dt−1∥2/2\lVert d_t-1\rVert_2/\sqrt{2}. Since ∂xQ=S∂x\partial_xQ=S\partial_x, we have bt′=etSb′b'_t=e^{tS}b'. Thus (A.2) and the Poincaré inequality associated with (2.6)–(2.7) give

∥etLT−Π∥L1≤32 σt≤32hλ1(h)∥etSb′∥2.(40)\lVert e^{t\mathcal{L}}T-\Pi\rVert_{\mathcal{L}_1}\leq3\sqrt{2}\,\sigma_t\leq\frac{3\sqrt{2}h}{\sqrt{\lambda_1(h)}}\lVert e^{tS}b'\rVert_2. \tag*{(40)}

The confinement assumptions (1.8) give c∣x∣≤∣V′(x)∣≤C∣x∣c\lvert x\rvert\leq\lvert V'(x)\rvert\leq C\lvert x\rvert, xV′(x)>0xV'(x)>0, near infinity. Writing ρ=e−2V/h\rho=e^{-2V/h} and defining mm to the median of μ\mu (∫−∞mρ(s) ds=12\int_{-\infty}^{m}\rho(s)\,ds=\frac{1}{2}), these bounds give, for large x>0x>0,

F+(x)≳h,Vρ(x)x,∫mxdsρ(s)≲h,V1xρ(x),log⁡1F+(x)≲h,Vx2.F_+(x)\gtrsim_{h,V}\frac{\rho(x)}{x},\qquad\int_m^x\frac{ds}{\rho(s)}\lesssim_{h,V}\frac{1}{x\rho(x)},\qquad\log\frac{1}{F_+(x)}\lesssim_{h,V}x^2.

Hence F+(x)log⁡(1/F+(x))∫mxρ(s)−1 dsF_+(x)\log(1/F_+(x))\int_m^x\rho(s)^{-1}\,ds is bounded with the similar argument for F−F_-. Thus μ\mu satisfies a logarithmic Sobolev inequality [1] [BR03, Theorem 1], and [Gr75, Theorem 6] gives ∥Ptp∥Lp→L2≤1\lVert P_{t_p}\rVert_{L^p\to L^2}\leq1 for some tp<∞t_p<\infty; take t2=0t_2=0. Since V′′≥−C0V''\geq-C_0, comparison gives 0≤etSG≤eC0tPtG0\leq e^{tS}G\leq e^{C_0t}P_tG for G≥0G\geq0. The spectral bound for SS in (2.8) therefore yields

∥etSb′∥2≤e−γh(t−tp)eC0tp∥b′∥p,t≥tp.(41)\lVert e^{tS}b'\rVert_2\leq e^{-\gamma h(t-t_p)}e^{C_0t_p}\lVert b'\rVert_p,\qquad t\geq t_p. \tag*{(41)}

Equations (A.5)–(A.7) prove (A.1) for t≥tpt\geq t_p. For t≤tpt\leq t_p, contractivity, the pure-state trace-distance formula and Cauchy–Schwarz give

∥etLT−Π∥L1≤21−tr⁡(ΠT)≤2E(T)λ1(h),\lVert e^{t\mathcal{L}}T-\Pi\rVert_{\mathcal{L}_1}\leq2\sqrt{1-\operatorname{tr}(\Pi T)}\leq2\sqrt{\frac{E(T)}{\lambda_1(h)}},

using H≥λ1(h)(I−Π)H\geq\lambda_1(h)(I-\Pi). Since ∥d∥q≥1\lVert d\rVert_q\geq1, this covers short times. For q=∞q=\infty, take p=2p=2, tp=0t_p=0; the constant is 626\sqrt{2}.

Finally, (1.11) and d∈Lq(μ)d\in L^q(\mu) show that F∈H1(μ;ℓ2)∩L2q(μ;ℓ2)F\in H^1(\mu;\ell^2)\cap L^{2q}(\mu;\ell^2). For finite qq, component and spatial cutoffs, then mollification and normalization, give smooth finite-dimensional FnF_n converging in both spaces, with ∥Fn∥2=1\lVert F_n\rVert_2=1. Thus ∥Tn−T∥L1≤2∥Fn−F∥2→0\lVert T_n-T\rVert_{\mathcal{L}_1}\leq2\lVert F_n-F\rVert_2\to0, and the energies and diagonal norms converge. Contractivity passes (A.1) to TT. For q=∞q=\infty, the same approximation gives ∥Fn∥∞≤(1+o(1))∥F∥∞\lVert F_n\rVert_\infty\leq(1+o(1))\lVert F\rVert_\infty. □

References

  1. [1][BR03] F. Barthe and C. Roberto, Sobolev inequalities for probability measures on the real line, Studia Math. 159 (2003), no. 3, 481–497.DOI
  2. [2][BRS26] S. Becker, C. Rouzé and R. Salzmann, Quantum Gibbs sampling in infinite dimensions: generation, mixing times and circuit implementation, arXiv:2604.01192.arxiv.org/abs/2604.01192
  3. [4]C-F. Chen, M. J. Kastoryano, and A. Gilyén, An efficient and exact noncommutative quantum Gibbs sampler, arXiv:2311.09207.arxiv.org/abs/2311.09207
  4. [5]P. J. Coles, M. Cerezo, and L. Cincio, Strong bound between trace distance and Hilbert–Schmidt distance for low-rank states, Phys. Rev. A 100 (2019), 022103.DOI
  5. [6]G. De Palma, D. Trevisan, and V. Giovannetti, Gaussian states minimize the output entropy of the one-mode quantum attenuator, IEEE Trans. Inform. Theory 63 (2017), no. 1, 728–737.DOI
  6. [7]G. De Palma, D. Trevisan, and V. Giovannetti, The one-mode quantum-limited Gaussian attenuator and amplifier have Gaussian maximizers, Ann. Henri Poincaré 19 (2018), no. 10, 2919–2953.DOI
  7. [8]Z. Ding, B. Li, and L. Lin, Efficient quantum Gibbs samplers with Kubo–Martin–Schwinger detailed balance condition, Comm. Math. Phys. 406 (2025), Paper No. 67.
  8. [9]S. Gao, Dissipative quantum dynamics at surfaces: A nonlinear-coupling model, Phys. Rev. B 60 (1999), 15609.DOI
  9. [10]J. Galkowski and M. Zworski, with an appendix by Z. Huang and M. Zworski, Classical–quantum correspondence in Lindblad evolution, J. Math. Phys. 66 (2025), Paper No. 091503.
  10. [11]J. Galkowski and M. Zworski, Localised Davies generators for (pseudo)differential operators, arXiv:2604.00306.arxiv.org/abs/2604.00306
  11. [12]I. C. Gohberg and M. G. Krein, Theory and Applications of Volterra Operators in Hilbert Space, Translations of Mathematical Monographs 24, American Mathematical Society, 1970.
  12. [13]V. Gorini, A. Kossakowski, and E. C. G. Sudarshan, Completely positive dynamical semigroups of N-level systems, J. Math. Phys. 17 (1976), 821–825.DOI
  13. [15]A. S. Holevo and R. F. Werner, Evaluating capacities of bosonic Gaussian channels, Phys. Rev. A 63 (2001), 032312.
  14. [16]F. Hernández, D. Ranard, and C. J. Riedel, Classical correspondence beyond the Ehrenfest time for open quantum systems with general Lindbladians, Comm. Math. Phys. 406 (2025), Paper No. 4.
  15. [17]J. Leng, Z. Ding, Z. Chen, and L. Lin, Operator-level quantum acceleration of non-logconcave sampling, Proc. Natl. Acad. Sci. USA 123 (2026), e2512789123.arxiv.org/abs/2505.05301
  16. [18]Z. Li, Long time quantum-classical correspondence for open systems in trace norm, Ann. Henri Poincaré (2025). https://doi.org/10.1007/s00023-025-01613-8
  17. [19]G. Lindblad, On the generators of quantum dynamical semigroups, Comm. Math. Phys. 48 (1976), 119–130.DOI
  18. [20]L. Michel and M. Zworski, A semiclassical approach to the Kramers–Smoluchowski equation, SIAM J. Math. Anal. 50 (2018), no. 5, 5362–5379.
  19. [21]M. Nest and P. Saalfrank, Open-system quantum dynamics for gas-surface scattering: Nonlinear dissipation and mapped Fourier grid methods, J. Chem. Phys. 113 (2000), no. 19, 8753–8761.DOI
  20. [22]M. Nest, Model for dissipative quantum dynamics and nonlinear coupling: Lennard–Jones potential, Phys. Rev. A 65 (2002), 052117.DOI
  21. [23]H. Smith, Lindblad evolution with subelliptic diffusion, Commun. Math. Phys. 407, 217 (2026). https://doi.org/10.1007/s00220-026-05750-x
  22. [24]C. Weedbrook, S. Pirandola, R. García-Patrón, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, Gaussian quantum information, Rev. Mod. Phys. 84 (2012), 621–669.
  23. [25]E. Witten, Supersymmetry and Morse theory, J. Differential Geom. 17 (1982), no. 4, 661–692.DOI

Paper details

vlnc:W_asnrs9m5
Version
v1 · latest
Published
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

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

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents