Introduction and main results
The core difficulty of the quantum many-body problem is what is known as the curse of dimensionality: the Hilbert space dimension grows exponentially with the number of constituents. A priori, this full complexity is needed to accurately describe the effects of interactions. If one assumes that the interactions are governed by a small parameter, then expansion methods (e.g., rigorous renormalization group or dilute limits) can be used to account for interactions perturbatively. In strongly interacting systems, the key insight to tame the curse of dimensionality is to use locality to effectively break the system up into O ( 1 ) O(1) O ( 1 ) -regions in which the exponential growth is effectively controllable. The mathematical manifestation of locality is in fact dynamical in nature: the Lieb–Robinson bound (LRB) , named after its first appearance in [32 ] , controls the spreading of a local observable under the Heisenberg dynamics.
The “unreasonable effectiveness of Lieb–Robinson bounds” is well-documented and we only mention a few highlights here:
In the 1970s, Robinson’s first motivation to pursue LRBs was to use them to establish the existence and strong continuity of the thermodynamic limit of the dynamics in the sense of operator algebras [32 , 7 ]
In the early 2000s Hastings famously discovered that LRBs combine productively with spectral gaps and time-frequency Fourier transforms which led to a number of breakthroughs on the locality properties of gapped ground states [HK04, NS03,
Has20]. Consequently, LRBs have become central to the description of topological quantum phases via automorphic equivalence [27 , 5 ] and they are the decisive analytical tools to prove stability of the spectral gap [3 , 38 , 43 , 42 , 11 , 47 ] .
(iii) In condensed matter-physics, LRBs are the decisive tool for mathematically rigorous many-body adiabatic theory with applications to quantization of the Hall conductance [2 , 25 , 26 , 8 , 18 , 24 ] and other prethermal phenomena [1 , 49 , 37 ] .
(iv) In quantum information theory, LRBs control the transfer of quantum information and entanglement quite generally [4 ] . On the one hand, this implies upper bounds on the performance of quantum-enhanced messaging protocols [13 ] . On the other hand, LRBs underpin the main algorithms to efficiently simulate the dynamics by suitably reducing the global problem to a local one. This applies both to classical simulation of the Heisenberg dynamics of local observables [10 ] and to the digital quantum simulation of the global dynamics through the state-of-the-art Hastings-Haah-Kothari-Low algorithm [21 , 30 ] .
For further background on LRBs, see the recent review [10 ] . The main problem we face is that all of these far-reaching and physically profound applications of LRBs (i)–(iv) are almost entirely limited to quantum spin systems and lattice fermions . The reason is that the standard proofs of LRBs require the interactions to be bounded in operator norm sense.
In the last few years, the frontier has been pushed to an important new class of lattice bosons models: Bose–Hubbard Hamiltonians [46 , 48 , 29 , 15 , 34 , 30 , 28 , 35 , 33 , 12 , 45 , 31 ] ; see also the earlier work [39 ] on perturbations of harmonic oscillators. In these lattice boson systems, the local interactions are unbounded, but one is nonetheless able to derive LRBs for lattice bosons which have two new features compared to their quantum spin system antecedent:
(b) The effective velocity of information propagation is v ∼ t d − 1 v \sim t^{d-1} v ∼ t d − 1 where d d d is the spatial dimension. Thus, the light cone is “bent” and the velocity is no longer uniformly bounded in time. However, it is bounded uniformly in system size (also called “thermodynamically stable”) and this is sufficient for the applications (i)–(iv) described before. The bent light cone just changes some exponents in the applications.
Accordingly, the applications (i)-(iv) of the new bosonic LRBs are now being actively explored and it is presumably only a matter of time until we reach a quantum information theory of lattice bosons that — subject to the changes implied by points (a) and (b) above — reaches a similar level of generality as its quantum spin precursor.
The true frontier are then continuum systems . These are both more fundamental and more realistic than their lattice counterparts and so it would be extremely desirable to unlock the far-reaching possibilities (i)–(iv) for them. LRBs have been shown to holds for certain systems of continuum fermions: ones with explicit ultraviolet cutoffs in the interaction [44 , 19 , 23 ] , ones acting on localized magnetic frames [BN25] or ones with pseudo-relativistic kinetic operator − Δ x i + m 2 \sqrt{-\Delta_{x_i}+m^2} − Δ x i + m 2 in a semiclassical regime [16 , 17 ] . However, the existing techniques do not apply to a standard many-body Hamiltonian of the form
H N = ∑ i = 1 N ( − Δ x i ) + ∑ 1 ≤ i < j ≤ N W ( x i − x j ) (1) H_N = \sum_{i=1}^{N}(-\Delta_{x_i}) + \sum_{1\leq i<j\leq N}W(x_i-x_j) \tag*{(1)} H N = i = 1 ∑ N ( − Δ x i ) + 1 ≤ i < j ≤ N ∑ W ( x i − x j ) (1)
To explain this further, we note that for a Bose–Hubbard Hamiltonian the kinetic operator of individual bosons is hopping, i.e., the discrete Laplacian which of course has a bounded group velocity. In some sense, it was this fact which could be exploited to treat Bose–Hubbard Hamiltonians in [46 , 48 , 29 , 15 , 34 , 30 , 28 , 35 , 33 , 12 , 45 , 31 ] . Roughly speaking, in a first step, one uses the bounded group velocity to control the accumulation of bosons locally. In a second step, this is used to approximate the bosons with a system of finite local dimension for which one can then derive an LRB using spin-system tools.
However, for the continuum Hamiltonian H N H_N H N , each particle’s kinetic energy is described by the continuum Laplacian − Δ x i -\Delta_{x_i} − Δ x i and therefore has unbounded group velocity. This is the reason why proving an LRB for any such Hamiltonian has remained completely open for a long time. To quote the last sentence of Bratteli-Robinson’s 1997 edition [7 ] : “Finally, the description of continuous system is notably incomplete. But, this is a feature of all methods which have been applied to their analysis and it is doubtful that any complete understanding of these models will be obtained in the near future.” LRBS for continuum systems have been even more sorely lacking since the advent of applications (ii)–(iv) in the meantime.
In the present paper, we break this longstanding barrier and prove the first Lieb–Robinson bounds for a continuum system. Specifically, we consider one-dimensional fermions in a torus geometry. Our LRB mirrors the two features (a) and (b) of the LRBS for lattice bosons recalled above:
(b)’ The effective velocity of information propagation is v ∼ e c t v \sim e^{ct} v ∼ e c t . Thus, the light cone is “exponentially bent” and the velocity is not uniformly bounded in time. However, it is bounded uniformly in system size .
In the present paper, we only aim to derive the first LRB for a standard continuum system. We postpone all applications, in particular those in the direction of (i)–(iv) from above, to future work.
Next, we describe the setup and main result, which is stated as Theorem 1.4 below and extended to general polynomial observables in Theorem 1.5. Afterwards, in Section 1.5, we will describe our proof strategy. In a nutshell, we build on the blueprint provided by the lattice boson strategy, by rewriting the continuum system as a lattice Hamiltonian on “pseudolattices”. The local Hilbert space of a pseudosite is the Fock space over L 2 ( I ) L^2(I) L 2 ( I ) for suitable intervals I I I . It is infinite-dimensional, but the occupation of high-energy modes can be controlled in terms of local energies, in some rough analogy with lattice bosons. This “pseudolattice” picture provides the first systematic conceptual framework for proving LRBS for continuum systems.
Our new conceptual framework makes equal sense for higher-dimensional continuum fermions and bosons and we believe it will allow to cover all of these systems eventually. However, the technical implementation is analytically challenging and, for technical reasons, we currently require a strong form of the Pauli exclusion principle that only holds in 1D; see Remark 1.6 for further details.
Setup
Write N = { 1 , 2 , … } \mathbb{N}=\{1,2,\ldots\} N = { 1 , 2 , … } and N 0 = N ∪ { 0 } \mathbb{N}_0=\mathbb{N}\cup\{0\} N 0 = N ∪ { 0 } . Given an integer L > 0 L>0 L > 0 , we identify T L = R / L Z \mathbb{T}_L=\mathbb{R}/L\mathbb{Z} T L = R / L Z with [ 0 , L ) [0,L) [ 0 , L ) , equipped with Lebesgue measure of mass L L L , and we set
∥ x ∥ = min { x , L − x } , B R ( x ) = { y ∈ T L : ∥ x − y ∥ < R } . (2) \lVert x\rVert=\min\{x,L-x\},\qquad B_R(x)=\{y\in\mathbb{T}_L:\lVert x-y\rVert<R\}. \tag*{(2)} ∥ x ∥ = min { x , L − x } , B R ( x ) = { y ∈ T L : ∥ x − y ∥ < R } . (2)
Because of the identification mentioned above, we can parameterize points on T L \mathbb{T}_L T L as real points taken modulo L L L . Thus ∥ x − y ∥ \|x-y\| ∥ x − y ∥ denotes the periodic distance on the torus. We shall describe many-body fermionic systems on T L \mathbb{T}_L T L in the grand-canonical formalism, so that the associated space of states is the fermionic Fock space F \mathcal{F} F , given by
F = ⨁ N ≥ 0 H N , H N = L 2 ( T L ) ∧ N , (3) \mathcal{F}=\bigoplus_{N\geq0}\mathcal{H}_N,\qquad\mathcal{H}_N=L^2(\mathbb{T}_L)^{\wedge N}, \tag*{(3)} F = N ≥ 0 ⨁ H N , H N = L 2 ( T L ) ∧ N , (3)
with H 0 = C \mathcal{H}_0=\mathbb{C} H 0 = C . For f ∈ L 2 ( T L ) f\in L^2(\mathbb{T}_L) f ∈ L 2 ( T L ) , we define the usual annihilation and creation operators a ( f ) , a ∗ ( f ) ∈ B ( F ) a(f),a^*(f)\in\mathcal{B}(\mathcal{F}) a ( f ) , a ∗ ( f ) ∈ B ( F ) by
( a ( f ) u N ) ( x 1 , … , x N − 1 ) = N ∫ T L f ( y ) ‾ u N ( y , x 1 , … , x N − 1 ) d y , a ∗ ( f ) = a ( f ) ∗ , (4) (a(f)u_N)(x_1,\ldots,x_{N-1})=\sqrt{N}\int_{\mathbb{T}_L}\overline{f(y)}u_N(y,x_1,\ldots,x_{N-1})\,dy,\qquad a^*(f)=a(f)^*, \tag*{(4)} ( a ( f ) u N ) ( x 1 , … , x N − 1 ) = N ∫ T L f ( y ) u N ( y , x 1 , … , x N − 1 ) d y , a ∗ ( f ) = a ( f ) ∗ , (4)
for N ≥ 1 N\geq1 N ≥ 1 , with zero annihilation on the vacuum. These operators satisfy the canonical anticommutation relations (CAR)
{ a ( f ) , a ( g ) } = 0 , { a ( f ) , a ∗ ( g ) } = ⟨ f , g ⟩ 1 , (5) \{a(f),a(g)\}=0,\qquad\{a(f),a^*(g)\}=\langle f,g\rangle\mathbf{1}, \tag*{(5)} { a ( f ) , a ( g )} = 0 , { a ( f ) , a ∗ ( g )} = ⟨ f , g ⟩ 1 , (5)
which imply ∥ a ( f ) ∥ = ∥ a ∗ ( f ) ∥ = ∥ f ∥ 2 \|a(f)\|=\|a^*(f)\|=\|f\|_2 ∥ a ( f ) ∥ = ∥ a ∗ ( f ) ∥ = ∥ f ∥ 2 . We use a x , a x ∗ a_x,a_x^* a x , a x ∗ for the corresponding operator-valued distributions.
Furthermore, recall that any self-adjoint operator A A A on a single-particle Hilbert space g \mathfrak{g} g lifts to its second quantization d Γ ( A ) \mathrm{d}\Gamma(A) d Γ ( A ) on the associated Fock space, whose sector-wise action is given by
d Γ ( A ) ∣ g ∧ N ≡ A N = ∑ i = 1 N A ( i ) . (6) \mathrm{d}\Gamma(A)|_{\mathfrak{g}^{\wedge N}}\equiv A_N=\sum_{i=1}^{N}A^{(i)}. \tag*{(6)} d Γ ( A ) ∣ g ∧ N ≡ A N = i = 1 ∑ N A ( i ) . (6)
In particular, given a measurable set U ⊂ T L U\subset\mathbb{T}_L U ⊂ T L , the local particle number operator is given by N U = d Γ ( 1 U ) \mathcal{N}_U=\mathrm{d}\Gamma(\mathbf{1}_U) N U = d Γ ( 1 U ) , with N ≡ N T L \mathcal{N}\equiv\mathcal{N}_{\mathbb{T}_L} N ≡ N T L being the global particle number operator. We consider Hamiltonians of the form
H = d Γ ( − ∂ 2 ) + 1 2 ∫ T L 2 W ( x − y ) a y ∗ a x ∗ a x a y d x d y . (7) H=\mathrm{d}\Gamma(-\partial^2)+\frac{1}{2}\int_{\mathbb{T}_L^2}W(x-y)a_y^*a_x^*a_xa_y\,dx\,dy. \tag*{(7)} H = d Γ ( − ∂ 2 ) + 2 1 ∫ T L 2 W ( x − y ) a y ∗ a x ∗ a x a y d x d y . (7)
The two-body interaction potential W ∈ C ∞ ( T L ; R ) W\in C^\infty(\mathbb{T}_L;\mathbb{R}) W ∈ C ∞ ( T L ; R ) is an L L L -periodization of a function in C ∞ ( R ; R ) C^\infty(\mathbb{R};\mathbb{R}) C ∞ ( R ; R ) that is even, non-negative, and finite-range, meaning it satisfies
W ( s ) = W ( − s ) ≥ 0 , supp W ⊂ B r 0 ( 0 ) , (8) W(s)=W(-s)\geq0,\qquad\operatorname{supp}W\subset B_{r_0}(0), \tag*{(8)} W ( s ) = W ( − s ) ≥ 0 , supp W ⊂ B r 0 ( 0 ) , (8)
for a fixed r 0 > 0 r_0>0 r 0 > 0 . In particular, r 0 r_0 r 0 is independent of L L L and W W W and all its derivatives are bounded independently of L L L as well. On each N N N -particle sector H H H takes the form
H N = − ∑ i = 1 N ∂ i 2 + ∑ 1 ≤ i < j ≤ N W ( x i − x j ) , (9) H_N=-\sum_{i=1}^{N}\partial_i^2+\sum_{1\leq i<j\leq N}W(x_i-x_j), \tag*{(9)} H N = − i = 1 ∑ N ∂ i 2 + 1 ≤ i < j ≤ N ∑ W ( x i − x j ) , (9)
which, since W W W is bounded and non-negative, is a self-adjoint and non-negative operator with domain D ( H N ) = H 2 ( T L N ) ∩ H N D(H_N)=H^2(\mathbb{T}_L^N)\cap\mathcal{H}_N D ( H N ) = H 2 ( T L N ) ∩ H N . On F \mathcal{F} F , H = ⨁ N ≥ 0 H N H=\bigoplus_{N\geq0}H_N H = ⨁ N ≥ 0 H N , with H 0 = 0 H_0=0 H 0 = 0 , is then self-adjoint on the corresponding direct-sum domain.
In the following sections we denote Heisenberg time evolution of a bounded observable A A A on F \mathcal{F} F by
τ t H ( A ) = e i t H A e − i t H . \tau_t^H(A)=e^{itH}Ae^{-itH}. τ t H ( A ) = e i t H A e − i t H .
Local energy moments
Let η ∈ C ∞ ( T L ; R ) \eta\in C^\infty(\mathbb{T}_L;\mathbb{R}) η ∈ C ∞ ( T L ; R ) be radial, with 0 ≤ η ≤ 1 0 \le\eta\le1 0 ≤ η ≤ 1 and η ∈ C ∞ ( T L ; R ) \sqrt{\eta} \in C^\infty(\mathbb{T}_L;\mathbb{R}) η ∈ C ∞ ( T L ; R ) , such that
η ( x ) = { 1 ∥ x ∥ ≤ 1 , 0 ∥ x ∥ ≥ 2. (10) \eta(x) =
\begin{cases}
1 & \lVert x\rVert\le1,\\
0 & \lVert x\rVert\ge2.
\end{cases} \tag*{(10)} η ( x ) = { 1 0 ∥ x ∥ ≤ 1 , ∥ x ∥ ≥ 2. (10)
For any z ∈ T L z \in\mathbb{T}_L z ∈ T L , we define
η z ( y ) : = η ( y − z ) , B z W = B 2 + r 0 ( z ) , (11) \eta_z(y) := \eta(y-z), \qquad B_z^W = B_{2+r_0}(z), \tag*{(11)} η z ( y ) := η ( y − z ) , B z W = B 2 + r 0 ( z ) , (11)
where the superscript W W W indicates the interaction neighborhood of B 2 ( z ) B_2(z) B 2 ( z ) : B z W B_z^W B z W contains all points that can interact through W W W with points of the interval B 2 ( z ) B_2(z) B 2 ( z ) centered at z z z .
Definition 1.1 (Local energy). On F \mathcal{F} F , E z \mathcal{E}_z E z is the nonnegative self-adjoint operator defined by
E z : = d Γ ( − ∂ η z ∂ ) + 1 4 ∫ T L 2 ( η z ( x ) + η z ( y ) ) W ( x − y ) a x ∗ a y ∗ a y a x d x d y , (12) \mathcal{E}_z := \mathrm{d}\Gamma(-\partial\eta_z\partial) + \frac{1}{4}\int_{\mathbb{T}_L^2}(\eta_z(x)+\eta_z(y))W(x-y)a_x^*a_y^*a_ya_x\,\mathrm{d}x\mathrm{d}y, \tag*{(12)} E z := d Γ ( − ∂ η z ∂ ) + 4 1 ∫ T L 2 ( η z ( x ) + η z ( y )) W ( x − y ) a x ∗ a y ∗ a y a x d x d y , (12) which is associated to the closure of the following energy form on C ∞ ( T L N ) ∩ H N C^\infty(\mathbb{T}_L^N)\cap\mathcal{H}_N C ∞ ( T L N ) ∩ H N :
E z [ u ] = ∑ i = 1 N ∫ T L N η z ( x i ) ∣ ∂ i u ( x ) ∣ 2 d x + 1 2 ∑ 1 ≤ i < j ≤ N ∫ T L N [ η z ( x i ) + η z ( x j ) ] W ( x i − x j ) ∣ u ( x ) ∣ 2 d x , (13) \mathcal{E}_z[u] = \sum_{i=1}^{N}\int_{\mathbb{T}_L^N}\eta_z(x_i)|\partial_i u(x)|^2\,\mathrm{d}x+\frac{1}{2}\sum_{1\le i<j\le N}\int_{\mathbb{T}_L^N}[\eta_z(x_i)+\eta_z(x_j)]W(x_i-x_j)|u(x)|^2\,\mathrm{d}x, \tag*{(13)} E z [ u ] = i = 1 ∑ N ∫ T L N η z ( x i ) ∣ ∂ i u ( x ) ∣ 2 d x + 2 1 1 ≤ i < j ≤ N ∑ ∫ T L N [ η z ( x i ) + η z ( x j )] W ( x i − x j ) ∣ u ( x ) ∣ 2 d x , (13) with x = ( x 1 , … , x N ) ∈ T L N x=(x_1,\ldots,x_N)\in\mathbb{T}_L^N x = ( x 1 , … , x N ) ∈ T L N .
The form is closable because each weighted first derivative is closable and the potential is bounded and nonnegative. With a slight abuse of notation, we denote the form and its closure by the same symbol E z \mathcal{E}_z E z . On F \mathcal{F} F , E z \mathcal{E}_z E z denotes the direct sum, with vacuum value zero.
In what follows, we will only use the local energy and particle number operators centered at integer points j = 0 , … , L − 1 j=0,\ldots,L-1 j = 0 , … , L − 1 , which are the centers of our “pseudosite intervals” introduced later.
Write S 1 ( F ) \mathcal{S}_1(\mathcal{F}) S 1 ( F ) for the trace-class operators on F \mathcal{F} F . We call a 0 ≤ ρ ∈ S 1 ( F ) 0\le\rho\in\mathcal{S}_1(\mathcal{F}) 0 ≤ ρ ∈ S 1 ( F ) with Tr ρ = 1 \operatorname{Tr}\rho=1 Tr ρ = 1 a state. For any nonnegative selfadjoint G G G on F \mathcal{F} F , we use the convention
Tr ρ G = ∥ G 1 / 2 ρ 1 / 2 ∥ 2 2 = ∑ ν = 1 rank ρ λ ν ∥ G 1 / 2 u ν ∥ 2 , ρ = ∑ ν = 1 rank ρ λ ν ∣ u ν ⟩ ⟨ u ν ∣ , λ ν > 0 , (14) \operatorname{Tr}\rho G=\lVert G^{1/2}\rho^{1/2}\rVert_2^2=\sum_{\nu=1}^{\operatorname{rank}\rho}\lambda_\nu\lVert G^{1/2}u_\nu\rVert^2,\qquad\rho=\sum_{\nu=1}^{\operatorname{rank}\rho}\lambda_\nu|u_\nu\rangle\langle u_\nu|,\quad\lambda_\nu>0, \tag*{(14)} Tr ρG = ∥ G 1/2 ρ 1/2 ∥ 2 2 = ν = 1 ∑ rank ρ λ ν ∥ G 1/2 u ν ∥ 2 , ρ = ν = 1 ∑ rank ρ λ ν ∣ u ν ⟩ ⟨ u ν ∣ , λ ν > 0 , (14)
where ∥ ⋅ ∥ 2 \lVert\cdot\rVert_2 ∥ ⋅ ∥ 2 is the Hilbert–Schmidt norm and a term outside the form domain has value + ∞ +\infty + ∞ . The same convention applies to nonnegative closed forms.
Definition 1.2. For any p > 0 p>0 p > 0 , and for any state ρ \rho ρ on F \mathcal{F} F , we define the p p p -th local moment of ρ \rho ρ to be
M p ( ρ ) = 1 + max 0 ≤ j < L Tr ρ ( E j p + N B j W p ) . (15) M_p(\rho)=1+\max_{0\le j<L}\operatorname{Tr}\rho\left(\mathcal{E}_j^p+\mathcal{N}_{B_j^W}^p\right). \tag*{(15)} M p ( ρ ) = 1 + 0 ≤ j < L max Tr ρ ( E j p + N B j W p ) . (15)
We shall assume our initial state ρ \rho ρ to satisfy
M p ( ρ ) < ∞ (16) M_p(\rho)<\infty \tag*{(16)} M p ( ρ ) < ∞ (16)
for all p ∈ N p\in\mathbb{N} p ∈ N . The following example provides physically relevant cases satisfying this local energy and particle-number moment bound.
Example 1.3 (Quasi-free initial states). For any quasi-free state ρ L \rho_L ρ L on F \mathcal{F} F with zero pairing, let γ L ( x , y ) = Tr ( ρ L a x ∗ a y ) \gamma_L(x,y)=\operatorname{Tr}(\rho_L a_x^*a_y) γ L ( x , y ) = Tr ( ρ L a x ∗ a y ) denote the kernel of its one-particle density matrix. A sufficient condition for (16) to hold uniformly in L L L is that γ L \gamma_L γ L is smooth and
sup L sup x , y ∈ T L ∣ ∂ x r ∂ y s γ L ( x , y ) ∣ < ∞ for every r , s ∈ N 0 . (17) \sup_L\sup_{x,y\in\mathbb{T}_L}\left|\partial_x^r\partial_y^s\gamma_L(x,y)\right|<\infty\qquad\text{for every }r,s\in\mathbb{N}_0. \tag*{(17)} L sup x , y ∈ T L sup ∂ x r ∂ y s γ L ( x , y ) < ∞ for every r , s ∈ N 0 . (17)
Indeed, Wick’s rule and the smoothness and finite range of W W W then give sup L M p ( ρ L ) < ∞ \sup_L M_p(\rho_L)<\infty sup L M p ( ρ L ) < ∞ for every p > 0 p>0 p > 0 . A concrete subclass of such states is that of Slater determinants built out of smooth functions.
In particular, for translation-invariant quasi-free states with zero pairing, the normalized plane waves L − 1 / 2 e i k x L^{-1/2}e^{ikx} L − 1/2 e ik x , k ∈ ( 2 π / L ) Z k\in(2\pi/L)\mathbb{Z} k ∈ ( 2 π / L ) Z , are independently occupied with probabilities ν L ( k ) ∈ [ 0 , 1 ] \nu_L(k)\in[0,1] ν L ( k ) ∈ [ 0 , 1 ] , and
γ L ( x , y ) = 1 L ∑ k ∈ ( 2 π / L ) Z ν L ( k ) e i k ( y − x ) . (18) \gamma_L(x,y)=\frac{1}{L}\sum_{k\in(2\pi/L)\mathbb{Z}}\nu_L(k)e^{ik(y-x)}. \tag*{(18)} γ L ( x , y ) = L 1 k ∈ ( 2 π / L ) Z ∑ ν L ( k ) e ik ( y − x ) . (18)
The kernel bounds (17) are then equivalent to
sup L 1 L ∑ k ∈ ( 2 π / L ) Z ( 1 + k 2 ) m ν L ( k ) < ∞ for every m ∈ N 0 . (19) \sup_L\frac{1}{L}\sum_{k\in(2\pi/L)\mathbb{Z}}(1+k^2)^m\nu_L(k)<\infty\qquad\text{for every }m\in\mathbb{N}_0. \tag*{(19)} L sup L 1 k ∈ ( 2 π / L ) Z ∑ ( 1 + k 2 ) m ν L ( k ) < ∞ for every m ∈ N 0 . (19)
A physically relevant example for which condition (19) holds is provided by the free grand-canonical thermal states at inverse temperature β > 0 \beta>0 β > 0 and chemical potential μ ∈ R \mu\in\mathbb{R} μ ∈ R ,
ρ L = e − β d Γ ( − ∂ 2 − μ ) Tr e − β d Γ ( − ∂ 2 − μ ) , ν L ( k ) = 1 1 + e β ( k 2 − μ ) , (20) \rho_L=\frac{e^{-\beta\mathrm{d}\Gamma(-\partial^2-\mu)}}{\operatorname{Tr}e^{-\beta\mathrm{d}\Gamma(-\partial^2-\mu)}},\qquad\nu_L(k)=\frac{1}{1+e^{\beta(k^2-\mu)}}, \tag*{(20)} ρ L = Tr e − β d Γ ( − ∂ 2 − μ ) e − β d Γ ( − ∂ 2 − μ ) , ν L ( k ) = 1 + e β ( k 2 − μ ) 1 , (20)
since their occupation probabilities decay exponentially in k 2 k^2 k 2 . It also holds whenever ν L ( k ) = 0 \nu_L(k)=0 ν L ( k ) = 0 for ∣ k ∣ > K |k|>K ∣ k ∣ > K , with K ≥ 0 K\geq0 K ≥ 0 independent of L L L , including the free Fermi ball ν L ( k ) = 1 { ∣ k ∣ ≤ K } \nu_L(k)=1_{\{|k|\leq K\}} ν L ( k ) = 1 { ∣ k ∣ ≤ K } .
Main result
We bound the evolved graded commutator of two initially separated observables in the Hilbert–Schmidt norm weighted by the initial state.
Theorem 1.4 (Main result). Let L ≥ 16 ( r 0 + 3 ) L\geq16(r_0+3) L ≥ 16 ( r 0 + 3 ) , k ∈ N k\in\mathbb{N} k ∈ N ; further, let W W W satisfy (8) . For a given pair x , y ∈ T L x,y\in\mathbb{T}_L x , y ∈ T L , and let f x , f y ∈ C ∞ ( T L ; C ) f_x,f_y\in C^\infty(\mathbb{T}_L;\mathbb{C}) f x , f y ∈ C ∞ ( T L ; C ) satisfy
supp f x ⊂ B 1 ( x ) , supp f y ⊂ B 1 ( y ) . (21) \operatorname{supp}f_x\subset B_1(x),\qquad\operatorname{supp}f_y\subset B_1(y). \tag*{(21)} supp f x ⊂ B 1 ( x ) , supp f y ⊂ B 1 ( y ) . (21) Then, there exist constants c k , C k > 0 c_k,C_k>0 c k , C k > 0 such that, for all states ρ \rho ρ such that M 2 k + 6 ( ρ ) < ∞ M_{2k+6}(\rho)<\infty M 2 k + 6 ( ρ ) < ∞ , for all σ , ν ∈ { − , + } \sigma,\nu\in\{-,+\} σ , ν ∈ { − , + } , and for all t ∈ R t\in\mathbb{R} t ∈ R we have
∥ { τ t H ( a σ ( f x ) ) , a ν ( f y ) } ρ 1 / 2 ∥ 2 ≤ C k M 2 k + 6 ( ρ ) 1 / 2 e c k ∣ t ∣ ⟨ x − y ⟩ k , (22) \left\|\left\{\tau_t^H\left(a^\sigma(f_x)\right),a^\nu(f_y)\right\}\rho^{1/2}\right\|_2\leq C_kM_{2k+6}(\rho)^{1/2}\frac{e^{c_k|t|}}{\langle x-y\rangle^k}, \tag*{(22)} { τ t H ( a σ ( f x ) ) , a ν ( f y ) } ρ 1/2 2 ≤ C k M 2 k + 6 ( ρ ) 1/2 ⟨ x − y ⟩ k e c k ∣ t ∣ , (22) where ⟨ x ⟩ : = ( 1 + ∥ x ∥ 2 ) 1 / 2 \langle x\rangle:=(1+\|x\|^2)^{1/2} ⟨ x ⟩ := ( 1 + ∥ x ∥ 2 ) 1/2 .
The constants c k c_k c k and C k C_k C k also depend on W W W , r 0 r_0 r 0 , the choice of cutoff function η \sqrt{\eta} η and the H 2 k + 4 H^{2k+4} H 2 k + 4 norms of f x , f y f_x,f_y f x , f y . Importantly, they are independent of L , ρ , t L,\rho,t L , ρ , t .
Generalization to polynomial observables
It is possible to generalize Theorem 1.4 to arbitrary polynomials in creation and annihilation operators. For open U ⊂ T L U \subset\mathbb{T}_L U ⊂ T L and ϵ ∈ { 0 , 1 } \epsilon\in\{0,1\} ϵ ∈ { 0 , 1 } , let
P ϵ ( U ) = span { ∏ i = 1 r a σ i ( f i ) : r ∈ N 0 , r ≡ ϵ ( m o d 2 ) , σ i ∈ { − , + } , f i ∈ C ∞ ( T L ; C ) , supp f i ⊂ U } ⊂ B ( F ) , (23) \mathfrak{P}^{\epsilon}(U)=\operatorname{span}\left\{\prod_{i=1}^{r}a^{\sigma_i}(f_i):\ r\in\mathbb{N}_0,\ r\equiv\epsilon\pmod{2},\ \sigma_i\in\{-,+\},\ f_i\in C^{\infty}(\mathbb{T}_L;\mathbb{C}),\ \operatorname{supp}f_i\subset U\right\}\subset\mathcal{B}(\mathcal{F}), \tag*{(23)} P ϵ ( U ) = span { i = 1 ∏ r a σ i ( f i ) : r ∈ N 0 , r ≡ ϵ ( mod 2 ) , σ i ∈ { − , + } , f i ∈ C ∞ ( T L ; C ) , supp f i ⊂ U } ⊂ B ( F ) , (23)
where a − = a a^{-}=a a − = a , a + = a ∗ a^{+}=a^{*} a + = a ∗ and an empty product is 1. An operator A ∈ B ( F ) A\in\mathcal{B}(\mathcal{F}) A ∈ B ( F ) has parity ϵ \epsilon ϵ if
( − 1 ) N A ( − 1 ) N = ( − 1 ) ϵ A . (24) (-1)^{\mathcal{N}}A(-1)^{\mathcal{N}}=(-1)^{\epsilon}A. \tag*{(24)} ( − 1 ) N A ( − 1 ) N = ( − 1 ) ϵ A . (24)
For A , B ∈ B ( F ) A,B\in\mathcal{B}(\mathcal{F}) A , B ∈ B ( F ) of parities ϵ , δ \epsilon,\delta ϵ , δ , respectively, we introduce the graded commutator
[ A , B ] g r : = A B − ( − 1 ) ϵ δ B A , (25) [A,B]_{\mathrm{gr}}:=AB-(-1)^{\epsilon\delta}BA, \tag*{(25)} [ A , B ] gr := A B − ( − 1 ) ϵδ B A , (25)
which corresponds to the anticommutator in the case in which both A , B A,B A , B are odd, and to the commutator in all other cases.
Theorem 1.5 (Extension to polynomial observables). Under the same assumptions as in Theorem 1.4 , for any ϵ , δ ∈ { 0 , 1 } \epsilon,\delta\in\{0,1\} ϵ , δ ∈ { 0 , 1 } , for any A ∈ P ϵ ( B 1 ( x ) ) A\in\mathfrak{P}^{\epsilon}(B_1(x)) A ∈ P ϵ ( B 1 ( x )) and B ∈ P δ ( B 1 ( y ) ) B\in\mathfrak{P}^{\delta}(B_1(y)) B ∈ P δ ( B 1 ( y )) , there exist constants c k , C k > 0 c_k,C_k>0 c k , C k > 0 such that
∥ [ T t H ( A ) , B ] g r ρ 1 / 2 ∥ 2 ≤ C k M 2 k + 6 ( ρ ) 1 / 2 e c k ∣ t ∣ ⟨ x − y ⟩ k (26) \left\|\left[\mathcal{T}_t^{H}(A),B\right]_{\mathrm{gr}}\rho^{1/2}\right\|_2\leq C_kM_{2k+6}(\rho)^{1/2}\frac{e^{c_k|t|}}{\langle x-y\rangle^k} \tag*{(26)} [ T t H ( A ) , B ] gr ρ 1/2 2 ≤ C k M 2 k + 6 ( ρ ) 1/2 ⟨ x − y ⟩ k e c k ∣ t ∣ (26) for all t ∈ R t\in\mathbb{R} t ∈ R .
The constants c k c_k c k and C k C_k C k now also depend on the finite polynomial expressions for A , B A,B A , B , and the H 2 k + 4 H^{2k+4} H 2 k + 4 norms of their test functions.
The proof of Theorem 1.5 is a simple adaptation of the proof of Theorem 1.4 which we relegate to Appendix § 8 .
Proof strategy
Our main contribution is to develop a conceptual framework for deriving continuum LRBs through a lattice-based picture. Our goal with this lattice-based picture is to link to the blueprint for deriving LRBs for lattice bosons that has recently emerged. Of course, the continuum poses the fundamentally new challenge that the group velocity of individual electrons is unbounded, but nonetheless we have found extrapolating from the bosonic blueprint to be extremely fruitful in developing the proof.
The lattice-based picture is constructed as follows; see Figure 2 for a figure. We decompose space using a quadratic partition of unity. This yields an effective decomposition of space into overlapping integrals, which we think of as “pseudosites”. Using Fock-space localization techniques, we can isometrically map the true continuum Fock space to a pseudolattice picture. Importantly, the effective pseudolattice Hamiltonian is local . For the interaction this is clear, especially since we even assume it has finite range for simplicity: it couples a pseudosite to all of those within the W W W -interaction range. Moreover, the boundary contributions to the kinetic energy are nearest-neighbor interactions between pseudosites and can be calculated by using the IMS localization formula [9 ] (which also saves a derivative).
Figure 2. From the continuum to the pseudolattice. Smooth localization gives a lattice of pseudosite labels, with an infinite dimensional Hilbert space at each pseudosite. (A) Localization on the original line. The black curve is f f f , the dashed curves are the cutoffs χ 1 \chi_1 χ 1 and χ 2 \chi_2 χ 2 , and the solid colored curves are the localized pieces χ 1 f \chi_1 f χ 1 f and χ 2 f \chi_2 f χ 2 f . The intervals I 1 = B 2 ( 1 ) I_1 = B_2(1) I 1 = B 2 ( 1 ) and I 2 = B 2 ( 2 ) I_2 = B_2(2) I 2 = B 2 ( 2 ) are displayed on separate rows below the axis, with their centers marked at 1 and 2. Only 0 ≤ x ≤ 3 0 \le x \le3 0 ≤ x ≤ 3 is shown; the arrows indicate that I 1 I_1 I 1 continues to the left and I 2 I_2 I 2 to the right. (B) Each piece is placed on its own pseudosite, corresponding to L 1 f = ( χ 1 f , χ 2 f ) \mathcal{L}_1 f = (\chi_1 f,\chi_2 f) L 1 f = ( χ 1 f , χ 2 f ) on the two active pseudosites. The operators B 12 B_{12} B 12 and B 21 B_{21} B 21 represent kinetic coupling between neighboring pseudosites. The pair potential also gives couplings of finite range.
Note that a pseudosite interval I I I still hosts the antisymmetric Fock space over L 2 ( I ) L^2(I) L 2 ( I ) . This is of course infinite-dimensional, but the point is that through this Fock-space localization procedure we have arrived at an effective local lattice model with infinitely many local modes . The occupancy of these local modes can be framed in terms of local particle number and kinetic energy, and therefore in terms of M p M_p M p -moments. In particular, initially, the occupancy of high modes is well-controlled by our assumed bound on M p M_p M p and Markov’s inequality. If we are able to control the dynamical growth of the M p M_p M p -moments, then we can now ideas of truncation dynamics developed in the lattice boson context to approximate the effective dynamics in the pseudolattice picture by a kind of lattice fermion system. It turns out, though, that this control poses a major technical obstacle.
Physically, we certainly expect that when starting from an initial state of bounded local energy moments as we do, then under the dynamics the local moment bounds may grow with time, but not with system size. Heuristically, any accumulation of energy in a local region must come from many and/or fast electrons reaching the region, but the creation of many fast electrons far away is in turn controlled by the growth of the same kind of local moment operator. This suggests that, as usual in dynamical proofs, one should aim to close a certain differential inequality within a suitable family of observables. Implementing this physically intuitive picture mathematically requires finding a family of observables that satisfies the following:
It controls all energy scales locally and high-energy scales far-away, with the energy scales controlled far away growing with the distance.
Their dynamics satisfies a differential inequality that closes within the same family of observables.
However, it turns out that many natural candidates of observables, such as the M p ′ M'_p M p ′ ’s themselves or other variants like spatially smeared out local kinetic operators, do not satisfy both of these requirements, either because they do not suitably link the scales, as required by the first bullet point, or because they do not close under taking commutators with the Hamiltonian, as required by the second bullet point.
Our technical contribution is then to introduce a new class of observables that fulfils both of these requirements. We call these new observables Momentum-adapted localization observables (MALOs) ; see Definition 2.4 . First, we prove that MALOs satisfy a Gronwall-type propagation bound Proposition 2.6 and thus grow at most exponentially in time. Second, we show that they are controlled by the M p M_p M p ’s (Proposition 2.7) which means our assumption on the initial state also provides initial control on the MALOs.
At this stage, we can extend the truncation dynamics technique developed for lattice bosons to the pseudolattice picture to derive an LRB-type operator localization estimate in terms of MALOs (Proposition 2.8), again in the pseudolattice picture. Combining this with the propagation bound for MALOs proved before and transferring back to the original Fock space picture, this rather easily implies prove Theorem 1.4 .
Organization of the paper
In § 2 , we formally introduce our new momentum-adapted localization observables (MALOs) and record their key properties, Propositions 2.6, 2.7, and 2.8 mentioned above. They are then combined to prove Theorem 1.4 , alongside a flowchart that summarizes the proof strategy described above (Figure 3 ).
Figure 3. Proof strategy for the continuum locality bound. Under the hypotheses of Theorem 1.4, take s = 2 k + 4 s=2k+4 s = 2 k + 4 and p = 2 k + 6 p=2k+6 p = 2 k + 6 . The arrows follow the order in which the estimates are applied. Proposition 2.7 bounds the initial MALO expectation by the local moments, Proposition 2.6 controls its evolution, and Proposition 2.8, combined with the initial comparison, gives the displayed local approximation for 1 ≤ R ≤ L / 4 1\le R\le L/4 1 ≤ R ≤ L /4 . Together with the operator-insertion estimate, these estimates yield the locality bound of Theorem 1.4 , as proved in § 2 .
Sections 3 and 4 are devoted to the proof of Proposition 2.6. Since time derivatives produce a commutator with the Hamiltonian, we have to control both the commutator of the MALO with the kinetic energy and its commutator with the potential energy. This requires different arguments, which are separated into Sections 3 and 4.
Section 2 contains the proof of Proposition 2.7, which rests on expressing the kinetic energy as the full energy minus the potential energy and controlling the potential energy by particle number.
Section 6 establishes the pseudolattice picture and, inspired by the lattice boson truncation schemes, prove an LRB provided one has sufficiently dynamical control on the MALO.
Appendix A contains a short argument about the operator domain of the kinetic boundary contribution in the pseudolattice picture and Appendix B adapts the proof to polynomial observables in Theorem 1.5 .
Momentum-adapted localization observables (MALOs) and proof of the main result
Definition of momentum-adapted localization observables (MALOs)
The quantity we propagate is a positive one-particle form localized near an integer pseudosite center. We use it to control the error from removing high pseudosite modes. It must satisfy two competing requirements: initial local moments must control its sum over all pseudosites, and its commutator with the interaction must be bounded by the form itself. The weight defined below combines spatial decay at fixed momentum with control of high momenta.
We first localize the one-particle space into bounded intervals. Periodic Fourier modes on these intervals define the weighted form and will later provide the finite-mode approximation used for locality.
For centers j ∈ { 0 , … , L − 1 } j \in\{0,\ldots,L-1\} j ∈ { 0 , … , L − 1 } , we define the j j j th pseudosite centered at j j j as the interval
I j = B 2 ( j ) , (27) I_j = B_2(j), \tag*{(27)} I j = B 2 ( j ) , (27)
and we fix a smooth non-negative quadratic partition of unity { χ j } j = 0 L − 1 \{\chi_j\}_{j=0}^{L-1} { χ j } j = 0 L − 1 on T L \mathbb{T}_L T L with the following properties:
χ j ( y ) = χ 0 ( y − j ) , ∑ j = 0 L − 1 χ j 2 = 1 , supp χ j ⊂ B 7 / 8 ( j ) . (28) \chi_j(y) = \chi_0(y-j), \qquad\sum_{j=0}^{L-1}\chi_j^2 = 1, \qquad\operatorname{supp}\chi_j \subset B_{7/8}(j). \tag*{(28)} χ j ( y ) = χ 0 ( y − j ) , j = 0 ∑ L − 1 χ j 2 = 1 , supp χ j ⊂ B 7/8 ( j ) . (28)
For every m ∈ N 0 m \in\mathbb{N}_0 m ∈ N 0 , we require ∥ ∂ m χ j ∥ ∞ ≤ C m \|\partial^m\chi_j\|_\infty\le C_m ∥ ∂ m χ j ∥ ∞ ≤ C m , uniformly in L L L and independently of all model parameters. We refer to 1 for a depiction of the functions χ j \chi_j χ j .
This partition of unity allows us to decompose a one-body state ψ ∈ L 2 ( T L ) \psi\in L^2(\mathbb{T}_L) ψ ∈ L 2 ( T L ) into the collection ( χ j ψ ) j = 0 L − 1 (\chi_j\psi)_{j=0}^{L-1} ( χ j ψ ) j = 0 L − 1 of its components in each pseudosite. This consideration motivates the introduction of the following definitions.
Definition 2.1 (Pseudolattice picture). We define the 1 1 1 -body pseudosite Hilbert space h ~ \widetilde{\mathfrak h} h and the corresponding pseudolattice Fock space F ~ \widetilde{\mathcal F} F as
h ~ : = ⨁ j = 0 L − 1 L 2 ( I j ) , F ~ : = ⨁ N ≥ 0 h ~ ∧ N . (29) \widetilde{\mathfrak h} := \bigoplus_{j=0}^{L-1} L^2(I_j), \qquad\widetilde{\mathcal F} := \bigoplus_{N\geq0}\widetilde{\mathfrak h}^{\wedge N}. \tag*{(29)} h := j = 0 ⨁ L − 1 L 2 ( I j ) , F := N ≥ 0 ⨁ h ∧ N . (29)
In the following sections, for any object O O O related to F \mathcal F F , we shall denote its equivalent on F ~ \widetilde{\mathcal F} F with O ~ \widetilde O O .
Figure 1. The cutoffs. The solid black curve is the energy cutoff η j \eta_j η j from (11) , and the dashed colored curves are those χ j \chi_j χ j pseudosite cutoff functions (28) whose supports overlap it. The cutoff η j \eta_j η j has support [ j − 2 , j + 2 ] [j-2,j+2] [ j − 2 , j + 2 ] and equals 1 on [ j − 1 , j + 1 ] [j-1,j+1] [ j − 1 , j + 1 ] . The support of χ j \chi_j χ j lies inside this plateau; the supports of χ j ± 1 \chi_{j\pm1} χ j ± 1 lie inside the support of η j \eta_j η j ; and the supports of χ j ± 2 \chi_{j\pm2} χ j ± 2 overlap the support of η j \eta_j η j and extend beyond it.
Definition 2.2 (Smooth localization isometry). The 1-body smooth localization isometry is the map
L 1 : L 2 ( T L ) → h ~ , ψ ↦ L 1 ψ = ( χ j ψ ) j = 0 , … , L − 1 , (30) \mathcal{L}_1:L^2(\mathbb{T}_L)\to\widetilde{\mathfrak{h}},\qquad\psi\mapsto\mathcal{L}_1\psi=(\chi_j\psi)_{j=0,\ldots,L-1}, \tag*{(30)} L 1 : L 2 ( T L ) → h , ψ ↦ L 1 ψ = ( χ j ψ ) j = 0 , … , L − 1 , (30) with adjoint
L 1 ∗ : h ~ → L 2 ( T L ) , ψ ~ ↦ L 1 ∗ ψ ~ = ∑ j = 0 L − 1 χ j ψ ~ j , (31) \mathcal{L}_1^*:\widetilde{\mathfrak{h}}\to L^2(\mathbb{T}_L),\qquad\widetilde{\psi}\mapsto\mathcal{L}_1^*\widetilde{\psi}=\sum_{j=0}^{L-1}\chi_j\widetilde{\psi}_j, \tag*{(31)} L 1 ∗ : h → L 2 ( T L ) , ψ ↦ L 1 ∗ ψ = j = 0 ∑ L − 1 χ j ψ j , (31) where every product is extended by zero to T L \mathbb{T}_L T L .
This in turn, by second quantization, induces a map between the respective Fock spaces, which we call the smooth localization isometry L \mathcal{L} L :
L : F → F ~ , L : = Γ ( L 1 ) ≡ ⨁ N ≥ 0 L 1 ⊗ N (32) \mathcal{L}:\mathcal{F}\to\widetilde{\mathcal{F}},\qquad\mathcal{L}:=\Gamma(\mathcal{L}_1)\equiv\bigoplus_{N\geq0}\mathcal{L}_1^{\otimes N} \tag*{(32)} L : F → F , L := Γ ( L 1 ) ≡ N ≥ 0 ⨁ L 1 ⊗ N (32)
For f ∈ L 2 ( T L ) f \in L^2(\mathbb{T}_L) f ∈ L 2 ( T L ) and σ ∈ { − , + } \sigma\in\{-,+\} σ ∈ { − , + } , the localization isometries give
L ∗ a σ ( L 1 f ) L = a σ ( f ) , ∥ a σ ( L 1 f ) ∥ = ∥ f ∥ 2 . (34) \mathcal{L}^*a^\sigma(\mathcal{L}_1 f)\mathcal{L}=a^\sigma(f), \qquad\left\|a^\sigma(\mathcal{L}_1 f)\right\|=\left\|f\right\|_2. \tag*{(34)} L ∗ a σ ( L 1 f ) L = a σ ( f ) , ∥ a σ ( L 1 f ) ∥ = ∥ f ∥ 2 . (34)
The single-particle kinetic energy operator in each interval is then defined by using the negative Laplacian with periodic boundary conditions, namely
T j = − ∂ 2 , D ( T j ) = H p e r 2 ( I j ) , T = ⨁ j = 0 L − 1 T j . (35) T_j=-\partial^2,\qquad D(T_j)=H^2_{\mathrm{per}}(I_j),\qquad T=\bigoplus_{j=0}^{L-1}T_j. \tag*{(35)} T j = − ∂ 2 , D ( T j ) = H per 2 ( I j ) , T = j = 0 ⨁ L − 1 T j . (35)
Therefore, the eigenvalues of T j T_j T j , counted with multiplicity, are of the form ( π n / 2 ) 2 (\pi n/2)^2 ( π n /2 ) 2 , n ∈ Z n \in\mathbb{Z} n ∈ Z .
Definition 2.4 (Momentum-adapted localization observable (MALO)). For s ∈ N s \in\mathbb{N} s ∈ N and a reference pseudosite j 0 ∈ { 0 , … , L − 1 } j_0 \in\{0,\ldots,L-1\} j 0 ∈ { 0 , … , L − 1 } , we set
g s , j ( λ ) : = ( 1 + λ ) s + 1 ⟨ j ⟩ 2 + λ ( j ∈ T L ∩ Z , λ ≥ 0 ) (36) g_{s,j}(\lambda):=\frac{(1+\lambda)^{s+1}}{\langle j\rangle^2+\lambda}\qquad(j\in\mathbb{T}_L\cap\mathbb{Z},\ \lambda\ge0) \tag*{(36)} g s , j ( λ ) := ⟨ j ⟩ 2 + λ ( 1 + λ ) s + 1 ( j ∈ T L ∩ Z , λ ≥ 0 ) (36) and we define the momentum-adapted localization observable (MALO) of order s s s to be
A s , j 0 : = L ∗ d Γ ( ⨁ j = 0 L − 1 g s , j − j 0 ( T j ) ) L , (37) A_{s,j_0}:=\mathcal{L}^*d\Gamma\left(\bigoplus_{j=0}^{L-1}g_{s,j-j_0}(T_j)\right)\mathcal{L}, \tag*{(37)} A s , j 0 := L ∗ d Γ ( j = 0 ⨁ L − 1 g s , j − j 0 ( T j ) ) L , (37) in the form sense.
We remark that using the periodic Laplacian is slightly more convenient for us than the Dirichlet Laplacian, which would also be possible since we are implementing zero-b.c. truncations. The slight advantage of the periodic Laplacian is that some calculations that use the explicit eigenbasis are easier for exponentials than for sine functions.
We use the same symbol A s , j 0 A_{s,j_0} A s , j 0 also for its associated densely defined, closed, non-negative quadratic form on F \mathcal{F} F , namely
A s , j 0 [ u ] = ∥ d Γ ( ⨁ j = 0 L − 1 g s , j − j 0 ( T j ) ) 1 / 2 L u ∥ 2 (38) A_{s,j_0}[u]=\left\| \mathrm{d}\Gamma\left(\bigoplus_{j=0}^{L-1}g_{s,j-j_0}(T_j)\right)^{1/2}\mathcal{L}u\right\|^2 \tag*{(38)} A s , j 0 [ u ] = d Γ ( j = 0 ⨁ L − 1 g s , j − j 0 ( T j ) ) 1/2 L u 2 (38)
with form domain
D ( A s , j 0 ) = { u ∈ F : A s , j 0 [ u ] < ∞ } = D ‾ ( ∥ ⋅ ∥ 2 + A s , j 0 [ ⋅ ] ) 1 / 2 (39) D(A_{s,j_0})=\{u\in\mathcal{F}:A_{s,j_0}[u]<\infty\}=\overline{\mathcal{D}}^{(\|\cdot\|^2+A_{s,j_0}[\cdot])^{1/2}} \tag*{(39)} D ( A s , j 0 ) = { u ∈ F : A s , j 0 [ u ] < ∞ } = D ( ∥ ⋅ ∥ 2 + A s , j 0 [ ⋅ ] ) 1/2 (39)
with D \mathcal{D} D being a smooth core, namely
D = ⨁ N ≥ 0 a l g ( C ∞ ( T L N ) ∩ H N ) . (40) \mathcal{D}=\bigoplus_{N\geq0}^{\mathrm{alg}}\left(C^\infty(\mathbb{T}_L^N)\cap\mathcal{H}_N\right). \tag*{(40)} D = N ≥ 0 ⨁ alg ( C ∞ ( T L N ) ∩ H N ) . (40)
Constants C C C may change from line to line. Unless stated otherwise, they depend only on the displayed orders and on W W W , r 0 r_0 r 0 , η \sqrt{\eta} η , with the universal localization cutoffs fixed as above.
Creation and annihilation operators on the pseudosites. The local approximation will take values in the algebra of bounded observables over a set of pseudosites. Therefore, we specify that algebra in terms of localized creation and annihilation operators before stating the approximation result. For pseudosite sets Z ⊂ { 0 , … , L − 1 } Z\subset\{0,\ldots,L-1\} Z ⊂ { 0 , … , L − 1 } , we define
h ~ Z : = ⨁ j ∈ Z L 2 ( I j ) , A Z : = { a ( f ~ ) , a ∗ ( f ~ ) : f ~ ∈ h ~ Z } ′ ′ = B ( F ( h ~ Z ) ) ⊗ 1 ⊂ B ( F ~ ) . (42) \widetilde{\mathfrak{h}}_Z:=\bigoplus_{j\in Z}L^2(I_j),\qquad\mathcal{A}_Z:=\{a(\widetilde{f}),a^*(\widetilde{f}):\widetilde{f}\in\widetilde{\mathfrak{h}}_Z\}^{\prime\prime}=\mathcal{B}(\mathcal{F}(\widetilde{\mathfrak{h}}_Z))\otimes1\subset\mathcal{B}(\widetilde{\mathcal{F}}). \tag*{(42)} h Z := j ∈ Z ⨁ L 2 ( I j ) , A Z := { a ( f ) , a ∗ ( f ) : f ∈ h Z } ′′ = B ( F ( h Z )) ⊗ 1 ⊂ B ( F ) . (42)
Here, the auxiliary creation and annihilation operators are defined as above on the Fock space over h ~ \widetilde{\mathfrak{h}} h . We write N ~ = d Γ ( 1 h ~ ) \widetilde{\mathcal{N}}=\mathrm{d}\Gamma(1_{\widetilde{\mathfrak{h}}}) N = d Γ ( 1 h ) for its total particle-number operator. Notice that, if supp f ⊂ B 1 ( x ) \operatorname{supp}f\subset B_1(x) supp f ⊂ B 1 ( x ) , then L 1 f ∈ h ~ X \mathcal{L}_1f\in\widetilde{\mathfrak{h}}_X L 1 f ∈ h X , where
X = { j : supp χ j ∩ B 1 ( x ) ≠ ∅ } ⊂ { j : ∥ j − x ∥ < 2 } , ∣ X ∣ ≤ C . (43) X=\{j:\operatorname{supp}\chi_j\cap B_1(x)\ne\varnothing\}\subset\{j:\lVert j-x\rVert<2\},\qquad|X|\leq C. \tag*{(43)} X = { j : supp χ j ∩ B 1 ( x ) = ∅ } ⊂ { j : ∥ j − x ∥ < 2 } , ∣ X ∣ ≤ C . (43)
Key facts about MALOs
The first proposition propagates each form and controls its change under insertion of a smeared creation or annihilation operator. The second bounds the forms A s , j 0 A_{s,j_0} A s , j 0 by the initial moments at pseudosite centers. The third uses these bounds to approximate its evolution by an operator supported near its initial support. The approximation error yields the locality estimate.
Proposition 2.6 (Propagation). For any s ∈ N s\in\mathbb{N} s ∈ N , any j 0 ∈ { 0 , … , L − 1 } j_0\in\{0,\ldots,L-1\} j 0 ∈ { 0 , … , L − 1 } , and for all t ∈ R t\in\mathbb{R} t ∈ R , there exists C s C_s C s such that
e i t H A s , j 0 e − i t H ≤ e C s ∣ t ∣ A s , j 0 , (44) e^{itH}A_{s,j_0}e^{-itH}\le e^{C_s|t|}A_{s,j_0}, \tag*{(44)} e i t H A s , j 0 e − i t H ≤ e C s ∣ t ∣ A s , j 0 , (44)
in the form sense on D ( A s , j 0 ) D(A_{s,j_0}) D ( A s , j 0 ) .
Proposition 2.7 (Initial comparison). Let s ∈ N s\in\mathbb{N} s ∈ N , and let p p p be the smallest even integer with p ≥ s + 1 p\ge s+1 p ≥ s + 1 . Then, there exists C s > 0 C_s>0 C s > 0 independent of L , ρ L,\rho L , ρ , such that
max 0 ≤ j < L Tr ρ A s , j ≤ C s M p ( ρ ) . (45) \max_{0\le j<L}\operatorname{Tr}\rho A_{s,j}\le C_sM_p(\rho). \tag*{(45)} 0 ≤ j < L max Tr ρ A s , j ≤ C s M p ( ρ ) . (45)
Proposition 2.8 (Approximation by a local observable). Let k ∈ N k\in\mathbb{N} k ∈ N , σ ∈ { − , + } \sigma\in\{-,+\} σ ∈ { − , + } , and f ∈ C ∞ ( T L ; C ) f\in C^\infty(\mathbb{T}_L;\mathbb{C}) f ∈ C ∞ ( T L ; C ) with supp f ⊂ B 1 ( x ) \operatorname{supp}f\subset B_1(x) supp f ⊂ B 1 ( x ) . Let 1 ≤ R ≤ L / 4 1\le R\le L/4 1 ≤ R ≤ L /4 . For every t ∈ R t\in\mathbb{R} t ∈ R , there exists an odd operator a R σ ( t , f ) ∈ A x [ R ] a_R^\sigma(t,f)\in\mathcal{A}_{x[R]} a R σ ( t , f ) ∈ A x [ R ] , where x [ R ] = { j ∈ { 0 , … , L − 1 } : ∥ j − x ∥ ≤ R } x[R]=\{j\in\{0,\ldots,L-1\}:\lVert j-x\rVert\le R\} x [ R ] = { j ∈ { 0 , … , L − 1 } : ∥ j − x ∥ ≤ R } , such that ∥ a R σ ( t , f ) ∥ ≤ ∥ f ∥ 2 \lVert a_R^\sigma(t,f)\rVert\le\lVert f\rVert_2 ∥ a R σ ( t , f )∥ ≤ ∥ f ∥ 2 and, for every state ρ \rho ρ ,
∥ [ L t H ( α σ ( f ) ) − a R σ ( t , f ) , L ] ρ 1 / 2 ∥ 2 ≤ C k ∥ f ∥ H 2 k + 4 e c k ∣ t ∣ ( 1 + R ) k ( 1 + max 0 ≤ j < L Tr ρ A 2 k + 4 , j ) 1 / 2 , (46) \left\lVert\left[\mathcal{L}_t^H(\alpha^\sigma(f))-a_R^\sigma(t,f),\mathcal{L}\right]\rho^{1/2}\right\rVert_2
\le\frac{C_k\lVert f\rVert_{H^{2k+4}}e^{c_k|t|}}{(1+R)^k}
\left(1+\max_{0\le j<L}\operatorname{Tr}\rho A_{2k+4,j}\right)^{1/2}, \tag*{(46)} [ L t H ( α σ ( f )) − a R σ ( t , f ) , L ] ρ 1/2 2 ≤ ( 1 + R ) k C k ∥ f ∥ H 2 k + 4 e c k ∣ t ∣ ( 1 + 0 ≤ j < L max Tr ρ A 2 k + 4 , j ) 1/2 , (46)
for positive constants c k , C k c_k,C_k c k , C k independent of the chosen ρ , f \rho,f ρ , f .
Proof of main result from key MALO properties
We begin by summarizing the proof strategy, which was already described in words in Subsection 1.5 in the following flowchart, 3.
Proof of Theorem 1.4. We apply the local approximation at a radius proportional to the separation of the supports. Set s = 2 k + 4 s=2k+4 s = 2 k + 4 . The initial comparison and the insertion estimate in Propositions 2.6 and 2.7, together with (179) , give
1 + max 0 ≤ j < L Tr ρ A s , j ≤ C M 2 k + 6 ( ρ ) , 1+\max_{0\le j<L}\operatorname{Tr}\rho A_{s,j}\le C M_{2k+6}(\rho), 1 + 0 ≤ j < L max Tr ρ A s , j ≤ C M 2 k + 6 ( ρ ) ,
Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ + max 0 ≤ j < L Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ A s , j ≤ C ∥ f y ∥ H s 2 M 2 k + 6 ( ρ ) . (47) \operatorname{Tr}a^{\nu}(f_y)\rho(a^{\nu}(f_y))^*+\max_{0\le j<L}\operatorname{Tr}a^{\nu}(f_y)\rho(a^{\nu}(f_y))^*A_{s,j}\le C\lVert f_y\rVert_{H^s}^2M_{2k+6}(\rho). \tag*{(47)} Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ + 0 ≤ j < L max Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ A s , j ≤ C ∥ f y ∥ H s 2 M 2 k + 6 ( ρ ) . (47)
For sufficiently large d = ∥ x − y ∥ d=\lVert x-y\rVert d = ∥ x − y ∥ , take R = d / 3 ≤ L / 6 R=d/3\le L/6 R = d /3 ≤ L /6 . Since the supports of a R σ ( t , f x ) a_R^\sigma(t,f_x) a R σ ( t , f x ) and a ν ( L 1 f y ) a^\nu(\mathcal{L}_1f_y) a ν ( L 1 f y ) are disjoint by (43) , their anticommutator vanishes. Put
D = L t H ( a σ ( f x ) ) − a R σ ( t , f x ) L . (48) D=\mathcal{L}_t^H(a^\sigma(f_x))-a_R^\sigma(t,f_x)\mathcal{L}. \tag*{(48)} D = L t H ( a σ ( f x )) − a R σ ( t , f x ) L . (48)
The identities (34) give
L { L t H ( a σ ( f x ) ) , a ν ( f y ) } = D a ν ( f y ) + a ν ( L 1 f y ) D . (49) \mathcal{L}\{\mathcal{L}_t^H(a^\sigma(f_x)),a^\nu(f_y)\}=Da^\nu(f_y)+a^\nu(\mathcal{L}_1f_y)D. \tag*{(49)} L { L t H ( a σ ( f x )) , a ν ( f y )} = D a ν ( f y ) + a ν ( L 1 f y ) D . (49)
Apply (46) with ρ \rho ρ and, when β : = Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ > 0 \beta:=\operatorname{Tr}a^\nu(f_y)\rho(a^\nu(f_y))^*>0 β := Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ > 0 , with the state ρ f y : = β − 1 a ν ( f y ) ρ ( a ν ( f y ) ) ∗ \rho_{f_y}:=\beta^{-1}a^\nu(f_y)\rho(a^\nu(f_y))^* ρ f y := β − 1 a ν ( f y ) ρ ( a ν ( f y ) ) ∗ . The identity
∥ D a ν ( f y ) ρ 1 / 2 ∥ 2 2 = Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ D ∗ D = β ∥ D ρ f y 1 / 2 ∥ 2 2 (50) \lVert Da^\nu(f_y)\rho^{1/2}\rVert_2^2=\operatorname{Tr}a^\nu(f_y)\rho(a^\nu(f_y))^*D^*D=\beta\lVert D\rho_{f_y}^{1/2}\rVert_2^2 \tag*{(50)} ∥ D a ν ( f y ) ρ 1/2 ∥ 2 2 = Tr a ν ( f y ) ρ ( a ν ( f y ) ) ∗ D ∗ D = β ∥ D ρ f y 1/2 ∥ 2 2 (50)
and the preceding bounds yield
∥ { L t H ( a σ ( f x ) ) , a ν ( f y ) } ρ 1 / 2 ∥ 2 ≤ ∥ D a ν ( f y ) ρ 1 / 2 ∥ 2 + ∥ f y ∥ 2 ∥ D ρ 1 / 2 ∥ 2 ≤ C ∥ f x ∥ H s ∥ f y ∥ H s e C ∣ t ∣ ( 1 + R ) k M 2 k + 6 ( ρ ) 1 / 2 . (51) \begin{aligned}
\lVert\{\mathcal{L}_t^H(a^\sigma(f_x)),a^\nu(f_y)\}\rho^{1/2}\rVert_2
&\le\lVert Da^\nu(f_y)\rho^{1/2}\rVert_2+\lVert f_y\rVert_2\lVert D\rho^{1/2}\rVert_2\\
&\le\frac{C\lVert f_x\rVert_{H^s}\lVert f_y\rVert_{H^s}e^{C|t|}}{(1+R)^k}M_{2k+6}(\rho)^{1/2}. \tag*{(51)}
\end{aligned} ∥{ L t H ( a σ ( f x )) , a ν ( f y )} ρ 1/2 ∥ 2 ≤ ∥ D a ν ( f y ) ρ 1/2 ∥ 2 + ∥ f y ∥ 2 ∥ D ρ 1/2 ∥ 2 ≤ ( 1 + R ) k C ∥ f x ∥ H s ∥ f y ∥ H s e C ∣ t ∣ M 2 k + 6 ( ρ ) 1/2 . (51)
Since R = d / 3 R=d/3 R = d /3 , this proves (22) for large d d d . For small d d d , we use the trivial bound
∥ { L t H ( a σ ( f x ) ) , a ν ( f y ) } ∥ ≤ 2 ∥ f x ∥ 2 ∥ f y ∥ 2 \lVert\{\mathcal{L}_t^H(a^\sigma(f_x)),a^\nu(f_y)\}\rVert\le2\lVert f_x\rVert_2\lVert f_y\rVert_2 ∥{ L t H ( a σ ( f x )) , a ν ( f y )}∥ ≤ 2 ∥ f x ∥ 2 ∥ f y ∥ 2
to conclude the proof.
Sections 3–6 prove the three propositions in order. The propagation rests on the kinetic and interaction commutator estimates, and the initial comparison rests on a product-rule comparison of kinetic and local-energy powers. For the local approximation, we first remove high pseudosite modes and then remove boundary couplings. Duhamel’s formula places the unbounded truncation error on the exact physical evolution, where the propagated form A s , j 0 A_{s,j_0} A s , j 0 controls it.
Propagation estimates: kinetic contribution
In this section we prove Proposition 2.6. The propagated form controls the mode-truncation error, and the insertion estimate allows us to apply this control after a creation or annihilation operator acts. Fix s ∈ N s\in\mathbb{N} s ∈ N and a reference pseudosite j 0 ∈ { 0 , … , L − 1 } j_0\in\{0,\ldots,L-1\} j 0 ∈ { 0 , … , L − 1 } .
For u ∈ D u\in\mathcal{D} u ∈ D , we set u t : = e − i t H u u_t:=e^{-itH}u u t := e − i t H u . Since H H H preserves particle number and W W W is smooth and bounded, u t ∈ D u_t\in\mathcal{D} u t ∈ D . On each of the finitely many sectors present in u u u , the map t ↦ u t t\mapsto u_t t ↦ u t is continuously differentiable in every Sobolev norm, with derivative − i H u t -iHu_t − i H u t .
Since, on each fixed particle number sector, A s , j 0 A_{s,j_0} A s , j 0 is bounded by a (sector-dependent) multiple of the squared H s H^s H s norm, t ↦ A s , j 0 [ u t ] t\mapsto A_{s,j_0}[u_t] t ↦ A s , j 0 [ u t ] is differentiable and
d d t A s , j 0 [ u t ] = ⟨ u t , i [ H , A s , j 0 ] u t ⟩ . (52) \frac{\mathrm{d}}{\mathrm{d}t}A_{s,j_0}[u_t]=\langle u_t,i[H,A_{s,j_0}]u_t\rangle. \tag*{(52)} d t d A s , j 0 [ u t ] = ⟨ u t , i [ H , A s , j 0 ] u t ⟩ . (52)
Writing H = K + V H=K+V H = K + V , with
K = d Γ ( − ∂ 2 ) , V = ⨁ n ≥ 0 ∑ 1 ≤ i < j ≤ n W ( x i − x j ) , (53) K=\mathrm{d}\Gamma(-\partial^2),\qquad V=\bigoplus_{n\geq0}\sum_{1\leq i<j\leq n}W(x_i-x_j), \tag*{(53)} K = d Γ ( − ∂ 2 ) , V = n ≥ 0 ⨁ 1 ≤ i < j ≤ n ∑ W ( x i − x j ) , (53)
we obtain
∣ d d t A s , j 0 [ u t ] ∣ ≤ ∣ ⟨ u t , [ A s , j 0 , K ] u t ⟩ ∣ + ∣ ⟨ u t , [ A s , j 0 , V ] u t ⟩ ∣ . (54) \left|\frac{\mathrm{d}}{\mathrm{d}t}A_{s,j_0}[u_t]\right|\leq\left|\langle u_t,[A_{s,j_0},K]u_t\rangle\right|+\left|\langle u_t,[A_{s,j_0},V]u_t\rangle\right|. \tag*{(54)} d t d A s , j 0 [ u t ] ≤ ∣ ⟨ u t , [ A s , j 0 , K ] u t ⟩ ∣ + ∣ ⟨ u t , [ A s , j 0 , V ] u t ⟩ ∣ . (54)
We bound the first term in (54) by C s A s , j 0 [ u t ] C_sA_{s,j_0}[u_t] C s A s , j 0 [ u t ] in Proposition 3.1 . We postpone an analogous bound on the second term, which depends on the two-body interaction, to § 4 .
Proposition 3.1 (Kinetic contribution). For u ∈ D u\in\mathcal{D} u ∈ D , ∣ ⟨ u , [ A s , j 0 , K ] u ⟩ ∣ ≤ C s A s , j 0 [ u ] \left|\langle u,[A_{s,j_0},K]u\rangle\right|\leq C_sA_{s,j_0}[u] ∣ ⟨ u , [ A s , j 0 , K ] u ⟩ ∣ ≤ C s A s , j 0 [ u ] .
Key estimates for the kinetic contribution
This subsection proves two key estimates used in the proof of Proposition 3.1 .
Since all intervals in the pseudosite decomposition have length 4 4 4 , in this subsection we work on I = R / ( 4 Z ) I=\mathbb{R}/(4\mathbb{Z}) I = R / ( 4 Z ) .
Let g \mathfrak{g} g be a complex Hilbert space and let T = − ∂ 2 T=-\partial^2 T = − ∂ 2 be the nonnegative selfadjoint operator on L 2 ( I ; g ) L^2(I;\mathfrak{g}) L 2 ( I ; g ) with D ( T ) = H 2 ( I ; g ) D(T)=H^2(I;\mathfrak{g}) D ( T ) = H 2 ( I ; g ) .
Lemma 3.2 (Multiplication estimate). For any j ∈ R j\in\mathbb{R} j ∈ R , and for any v ∈ C ∞ ( I ; R ) v\in C^\infty(I;\mathbb{R}) v ∈ C ∞ ( I ; R ) and w ∈ C ∞ ( I ; g ) w\in C^\infty(I;\mathfrak{g}) w ∈ C ∞ ( I ; g ) ,
∣ ⟨ w , [ g s , j ( T ) , v ] w ⟩ ∣ ≤ C s ∥ v ∥ C s + 2 ( I ) ∥ g s , j ( T ) 1 / 2 w ∥ ∥ ( g s , j ( T ) 1 + T ) 1 / 2 w ∥ , (55) \left|\langle w,[g_{s,j}(T),v]w\rangle\right|\leq C_s\|v\|_{C^{s+2}(I)}\|g_{s,j}(T)^{1/2}w\|\left\|\left(\frac{g_{s,j}(T)}{1+T}\right)^{1/2}w\right\|, \tag*{(55)} ∣ ⟨ w , [ g s , j ( T ) , v ] w ⟩ ∣ ≤ C s ∥ v ∥ C s + 2 ( I ) ∥ g s , j ( T ) 1/2 w ∥ ( 1 + T g s , j ( T ) ) 1/2 w , (55) for some constant C s C_s C s independent of j , g j,\mathfrak{g} j , g .
Lemma 3.3 (Paired weighted estimate). Given a ∈ C ∞ ( I ; R ) a\in C^\infty(I;\mathbb{R}) a ∈ C ∞ ( I ; R ) , define B : H 1 ( I ; g ) → L 2 ( I ; g ) B:H^1(I;\mathfrak{g})\to L^2(I;\mathfrak{g}) B : H 1 ( I ; g ) → L 2 ( I ; g ) by
B f : = a ∂ f + a ′ 2 f . Bf:=a\partial f+\frac{a'}{2}f. B f := a ∂ f + 2 a ′ f . Let j , k ∈ R j,k\in\mathbb{R} j , k ∈ R be such that ∣ ⟨ j ⟩ − ⟨ k ⟩ ∣ ≤ 1 |\langle j\rangle-\langle k\rangle|\leq1 ∣ ⟨ j ⟩ − ⟨ k ⟩ ∣ ≤ 1 ; then, for all f , g ∈ C ∞ ( I ; g ) f,g\in C^\infty(I;\mathfrak{g}) f , g ∈ C ∞ ( I ; g ) ,
∣ ⟨ f , ( g s , j ( T ) B − B g s , k ( T ) ) g ⟩ ∣ ≤ C ∥ a ∥ C s + 3 ( I ) ∥ g s , j ( T ) 1 / 2 f ∥ ∥ g s , k ( T ) 1 / 2 g ∥ . (56) |\langle f,(g_{s,j}(T)B-Bg_{s,k}(T))g\rangle|\leq C\|a\|_{C^{s+3}(I)}\|g_{s,j}(T)^{1/2}f\|\|g_{s,k}(T)^{1/2}g\|. \tag*{(56)} ∣ ⟨ f , ( g s , j ( T ) B − B g s , k ( T )) g ⟩ ∣ ≤ C ∥ a ∥ C s + 3 ( I ) ∥ g s , j ( T ) 1/2 f ∥∥ g s , k ( T ) 1/2 g ∥. (56) for some constant C s C_s C s independent of j , k j,k j , k , and g \mathfrak{g} g .
Proof of the two lemmas. The main estimate is
∥ [ g s , j ( T ) 1 / 2 , v ] w ∥ ≤ C s ∥ v ∥ C s + 2 ( I ) ∥ ( g s , j ( T ) 1 + T ) 1 / 2 w ∥ , (57) \|[g_{s,j}(T)^{1/2},v]w\|\leq C_s\|v\|_{C^{s+2}(I)}\left\|\left(\frac{g_{s,j}(T)}{1+T}\right)^{1/2}w\right\|, \tag*{(57)} ∥ [ g s , j ( T ) 1/2 , v ] w ∥ ≤ C s ∥ v ∥ C s + 2 ( I ) ( 1 + T g s , j ( T ) ) 1/2 w , (57)
for v ∈ C ∞ ( I ; C ) v\in C^\infty(I;\mathbb{C}) v ∈ C ∞ ( I ; C ) and w ∈ C ∞ ( I ; g ) w\in C^\infty(I;\mathfrak{g}) w ∈ C ∞ ( I ; g ) , uniformly in b b b and g \mathfrak{g} g . We prove it in the first two steps, and in the third step we prove the two lemmas. For simplicity of notation, we employ the shorthand b : = ⟨ j ⟩ b:=\langle j\rangle b := ⟨ j ⟩ and c : = ⟨ k ⟩ c:=\langle k\rangle c := ⟨ k ⟩ .
Step 1: Differences of square roots of the weights. For p , q ∈ R p,q\in\mathbb{R} p , q ∈ R , put t = ( 1 + p 2 ) 1 / 2 t=(1+p^2)^{1/2} t = ( 1 + p 2 ) 1/2 and u = ( 1 + q 2 ) 1 / 2 u=(1+q^2)^{1/2} u = ( 1 + q 2 ) 1/2 . Then ∣ t − u ∣ ≤ ∣ p − q ∣ |t-u|\leq|p-q| ∣ t − u ∣ ≤ ∣ p − q ∣ and t , u ≥ 1 t,u\geq1 t , u ≥ 1 . If t ≥ u t\geq u t ≥ u , then
( t u ) s ≤ ( g s , j ( p 2 ) g s , j ( q 2 ) ) 1 / 2 = ( t u ) s + 1 ( b 2 − 1 + u 2 b 2 − 1 + t 2 ) 1 / 2 ≤ ( t u ) s + 1 . (58) \left(\frac{t}{u}\right)^s\leq\left(\frac{g_{s,j}(p^2)}{g_{s,j}(q^2)}\right)^{1/2}=\left(\frac{t}{u}\right)^{s+1}\left(\frac{b^2-1+u^2}{b^2-1+t^2}\right)^{1/2}\leq\left(\frac{t}{u}\right)^{s+1}. \tag*{(58)} ( u t ) s ≤ ( g s , j ( q 2 ) g s , j ( p 2 ) ) 1/2 = ( u t ) s + 1 ( b 2 − 1 + t 2 b 2 − 1 + u 2 ) 1/2 ≤ ( u t ) s + 1 . (58)
For the lower bound, the last square root is at least u / t u/t u / t , because b 2 − 1 ≥ 0 b^2-1 \ge0 b 2 − 1 ≥ 0 . For the upper bound, this square root is at most one. In particular, the weight increases with t t t . The finite geometric sum therefore gives, when t ≥ u t \ge u t ≥ u ,
∣ g s , j ( p 2 ) 1 / 2 − g s , j ( q 2 ) 1 / 2 ∣ g s , j ( q 2 ) 1 / 2 ≤ ( t u ) s + 1 − 1 = t − u u ∑ ℓ = 0 s ( t u ) ℓ ≤ ( s + 1 ) ∣ p − q ∣ u ( 1 + ∣ p − q ∣ ) s . (59) \begin{aligned}
\frac{\left|g_{s,j}(p^2)^{1/2}-g_{s,j}(q^2)^{1/2}\right|}{g_{s,j}(q^2)^{1/2}} \le\left(\frac{t}{u}\right)^{s+1}-1 \\
&= \frac{t-u}{u}\sum_{\ell=0}^{s}\left(\frac{t}{u}\right)^{\ell} \le(s+1)\frac{|p-q|}{u}(1+|p-q|)^s. \tag*{(59)}
\end{aligned} g s , j ( q 2 ) 1/2 g s , j ( p 2 ) 1/2 − g s , j ( q 2 ) 1/2 ≤ ( u t ) s + 1 − 1 = u t − u ℓ = 0 ∑ s ( u t ) ℓ ≤ ( s + 1 ) u ∣ p − q ∣ ( 1 + ∣ p − q ∣ ) s . (59)
When t ≤ u t \le u t ≤ u , the same ratio bound with the momenta exchanged gives
( t u ) s + 1 ≤ ( g s , j ( p 2 ) g s , j ( q 2 ) ) 1 / 2 ≤ 1. (60) \left(\frac{t}{u}\right)^{s+1} \le\left(\frac{g_{s,j}(p^2)}{g_{s,j}(q^2)}\right)^{1/2} \le1. \tag*{(60)} ( u t ) s + 1 ≤ ( g s , j ( q 2 ) g s , j ( p 2 ) ) 1/2 ≤ 1. (60)
The normalized difference is then at most 1 − ( t / u ) s + 1 ≤ ( s + 1 ) ( u − t ) / u 1-(t/u)^{s+1} \le(s+1)(u-t)/u 1 − ( t / u ) s + 1 ≤ ( s + 1 ) ( u − t ) / u . Thus, in both cases,
∣ g s , j ( p 2 ) 1 / 2 − g s , j ( q 2 ) 1 / 2 ∣ ≤ C ∣ p − q ∣ ( 1 + ∣ p − q ∣ ) s ( g s , j ( q 2 ) 1 + q 2 ) 1 / 2 . (61) \left|g_{s,j}(p^2)^{1/2}-g_{s,j}(q^2)^{1/2}\right| \le C|p-q|(1+|p-q|)^s\left(\frac{g_{s,j}(q^2)}{1+q^2}\right)^{1/2}. \tag*{(61)} g s , j ( p 2 ) 1/2 − g s , j ( q 2 ) 1/2 ≤ C ∣ p − q ∣ ( 1 + ∣ p − q ∣ ) s ( 1 + q 2 g s , j ( q 2 ) ) 1/2 . (61)
Step 2: The square-root commutator. Let p n = π n / 2 p_n=\pi n/2 p n = π n /2 and e n ( x ) = e i p n x / 2 e_n(x)=e^{ip_nx/2} e n ( x ) = e i p n x /2 for n ∈ Z n\in\mathbb{Z} n ∈ Z . These are the orthonormal Fourier modes in L 2 ( I ) L^2(I) L 2 ( I ) . For v ∈ C ∞ ( I ; C ) v\in C^\infty(I;\mathbb{C}) v ∈ C ∞ ( I ; C ) , define
v ^ l = 1 4 ∫ I v ( x ) e − i p l x d x , l ∈ Z . (62) \widehat{v}_l=\frac{1}{4}\int_I v(x)e^{-ip_lx}\,\mathrm{d}x,\qquad l\in\mathbb{Z}. \tag*{(62)} v l = 4 1 ∫ I v ( x ) e − i p l x d x , l ∈ Z . (62)
Multiplication by v v v has matrix v ^ n − m \widehat{v}_{n-m} v n − m in this basis. For every integer m ≥ 0 m\ge0 m ≥ 0 , Cauchy–Schwarz and Parseval give
∑ l ( 1 + ∣ l ∣ ) m ∣ v ^ l ∣ ≤ ( ∑ l ( 1 + ∣ l ∣ ) − 2 ) 1 / 2 ( ∑ l ( 1 + ∣ l ∣ ) 2 m + 2 ∣ v ^ l ∣ 2 ) 1 / 2 ≤ C m ( ∑ r = 0 m + 1 ∥ ∂ r v ∥ L 2 ( I ) 2 ) 1 / 2 ≤ C m ∥ v ∥ C m + 1 ( I ) . (63) \begin{aligned}
\sum_l(1+|l|)^m|\widehat{v}_l|\le\left(\sum_l(1+|l|)^{-2}\right)^{1/2}\left(\sum_l(1+|l|)^{2m+2}|\widehat{v}_l|^2\right)^{1/2} \\
&\le C_m\left(\sum_{r=0}^{m+1}\|\partial^r v\|_{L^2(I)}^2\right)^{1/2}\le C_m\|v\|_{C^{m+1}(I)}. \tag*{(63)}
\end{aligned} l ∑ ( 1 + ∣ l ∣ ) m ∣ v l ∣ ≤ ( l ∑ ( 1 + ∣ l ∣ ) − 2 ) 1/2 ( l ∑ ( 1 + ∣ l ∣ ) 2 m + 2 ∣ v l ∣ 2 ) 1/2 ≤ C m ( r = 0 ∑ m + 1 ∥ ∂ r v ∥ L 2 ( I ) 2 ) 1/2 ≤ C m ∥ v ∥ C m + 1 ( I ) . (63)
For the second line, we use ( 1 + ∣ l ∣ ) 2 m + 2 ≤ C m ∑ r = 0 m + 1 ∣ p l ∣ 2 r (1+|l|)^{2m+2}\le C_m\sum_{r=0}^{m+1}|p_l|^{2r} ( 1 + ∣ l ∣ ) 2 m + 2 ≤ C m ∑ r = 0 m + 1 ∣ p l ∣ 2 r and ∑ l ∣ p l ∣ 2 r ∣ v ^ l ∣ 2 = ∥ ∂ r v ∥ L 2 ( I ) 2 / 4 \sum_l|p_l|^{2r}|\widehat{v}_l|^2=\|\partial^r v\|_{L^2(I)}^2/4 ∑ l ∣ p l ∣ 2 r ∣ v l ∣ 2 = ∥ ∂ r v ∥ L 2 ( I ) 2 /4 .
We first take w w w to be a finite Fourier series, with coefficients w n = ∫ I e n ( x ) w ( x ) d x ∈ g w_n=\int_I e_n(x)w(x)\,\mathrm{d}x\in\mathfrak{g} w n = ∫ I e n ( x ) w ( x ) d x ∈ g . The Fourier coefficients of the commutator are
( [ g s , j ( T ) 1 / 2 , v ] w ) n = ∑ m ( g s , j ( p n 2 ) 1 / 2 − g s , j ( p m 2 ) 1 / 2 ) v ^ n − m w m . (64) ([g_{s,j}(T)^{1/2},v]w)_n=\sum_m\left(g_{s,j}(p_n^2)^{1/2}-g_{s,j}(p_m^2)^{1/2}\right)\widehat{v}_{n-m}w_m. \tag*{(64)} ([ g s , j ( T ) 1/2 , v ] w ) n = m ∑ ( g s , j ( p n 2 ) 1/2 − g s , j ( p m 2 ) 1/2 ) v n − m w m . (64)
Apply (61) and the triangle inequality in ℓ 2 ( Z ; g ) \ell^2(\mathbb{Z};\mathfrak{g}) ℓ 2 ( Z ; g ) , summing over the diagonals n − m = l n-m=l n − m = l . Parseval’s identity and (63) , with m = s + 1 m=s+1 m = s + 1 , give
∥ [ g s , j ( T ) 1 / 2 , v ] w ∥ ≤ C ∑ l ( 1 + ∣ l ∣ ) s + 1 ∣ v ^ l ∣ ( ∑ m g s , j ( p m 2 ) 1 + p m 2 ∥ w m ∥ 2 ) 1 / 2 ≤ C ∥ v ∥ C s + 2 ( I ) ∥ ( g s , j ( T ) 1 + T ) 1 / 2 w ∥ . (65) \begin{aligned}
\|[g_{s,j}(T)^{1/2},v]w\|\le C\sum_l(1+|l|)^{s+1}|\widehat{v}_l|\left(\sum_m\frac{g_{s,j}(p_m^2)}{1+p_m^2}\|w_m\|^2\right)^{1/2} \\
&\le C\|v\|_{C^{s+2}(I)}\left\|\left(\frac{g_{s,j}(T)}{1+T}\right)^{1/2}w\right\|. \tag*{(65)}
\end{aligned} ∥ [ g s , j ( T ) 1/2 , v ] w ∥ ≤ C l ∑ ( 1 + ∣ l ∣ ) s + 1 ∣ v l ∣ ( m ∑ 1 + p m 2 g s , j ( p m 2 ) ∥ w m ∥ 2 ) 1/2 ≤ C ∥ v ∥ C s + 2 ( I ) ( 1 + T g s , j ( T ) ) 1/2 w . (65)
Fourier truncations of a smooth vector converge in every Sobolev norm. Since g s , j ( p 2 ) 1 / 2 ≤ ( 1 + p 2 ) s / 2 g_{s,j}(p^2)^{1/2}\le(1+p^2)^{s/2} g s , j ( p 2 ) 1/2 ≤ ( 1 + p 2 ) s /2 and multiplication by v v v preserves these spaces, passing to the limit proves (57) for all smooth w w w .
Step 3: Deduction of the two estimates. For real v v v , multiplication by v v v is selfadjoint. Hence
∣ ⟨ w , [ g s , j ( T ) , v ] w ⟩ ∣ = ∣ ⟨ g s , j ( T ) 1 / 2 w , [ g s , j ( T ) 1 / 2 , v ] w ⟩ − ⟨ [ g s , j ( T ) 1 / 2 , v ] w , g s , j ( T ) 1 / 2 w ⟩ ∣ ≤ 2 ∥ g s , j ( T ) 1 / 2 w ∥ ∥ [ g s , j ( T ) 1 / 2 , v ] w ∥ . (66) \begin{aligned}
\left|\langle w,[g_{s,j}(T),v]w\rangle\right|
&=\left|\langle g_{s,j}(T)^{1/2}w,[g_{s,j}(T)^{1/2},v]w\rangle-\langle[g_{s,j}(T)^{1/2},v]w,g_{s,j}(T)^{1/2}w\rangle\right|\\
&\le2\left\|g_{s,j}(T)^{1/2}w\right\|\left\|[g_{s,j}(T)^{1/2},v]w\right\|. \tag*{(66)}
\end{aligned} ∣ ⟨ w , [ g s , j ( T ) , v ] w ⟩ ∣ = ⟨ g s , j ( T ) 1/2 w , [ g s , j ( T ) 1/2 , v ] w ⟩ − ⟨[ g s , j ( T ) 1/2 , v ] w , g s , j ( T ) 1/2 w ⟩ ≤ 2 g s , j ( T ) 1/2 w [ g s , j ( T ) 1/2 , v ] w . (66)
Estimate (57) proves Lemma 3.2.
For the paired estimate, assume ∣ b − c ∣ ≤ 1 |b-c|\le1 ∣ b − c ∣ ≤ 1 . Since b , c ≥ 1 b,c\ge1 b , c ≥ 1 , we have b / 2 ≤ c ≤ 2 b b/2\le c\le2b b /2 ≤ c ≤ 2 b , and hence, for p ∈ R p\in\mathbb{R} p ∈ R ,
1 2 ≤ ( g s , j ( p 2 ) g s , k ( p 2 ) ) 1 / 2 ≤ 2 , \frac{1}{2}\le\left(\frac{g_{s,j}(p^2)}{g_{s,k}(p^2)}\right)^{1/2}\le2, 2 1 ≤ ( g s , k ( p 2 ) g s , j ( p 2 ) ) 1/2 ≤ 2 ,
∣ g s , j ( p 2 ) 1 / 2 − g s , k ( p 2 ) 1 / 2 ∣ g s , k ( p 2 ) 1 / 2 = ∣ c 2 + p 2 − b 2 + p 2 ∣ b 2 + p 2 ≤ 1 1 + p 2 . (67) \frac{\left|g_{s,j}(p^2)^{1/2}-g_{s,k}(p^2)^{1/2}\right|}{g_{s,k}(p^2)^{1/2}}
=\frac{\left|\sqrt{c^2+p^2}-\sqrt{b^2+p^2}\right|}{\sqrt{b^2+p^2}}
\le\frac{1}{\sqrt{1+p^2}}. \tag*{(67)} g s , k ( p 2 ) 1/2 g s , j ( p 2 ) 1/2 − g s , k ( p 2 ) 1/2 = b 2 + p 2 c 2 + p 2 − b 2 + p 2 ≤ 1 + p 2 1 . (67)
where in the last inequality we used that b ↦ b 2 + p 2 b\mapsto\sqrt{b^2+p^2} b ↦ b 2 + p 2 is 1-Lipschitz. Therefore, for smooth g g g , we have
∥ g s , j ( T ) 1 / 2 g ∥ ≤ 2 ∥ g s , k ( T ) 1 / 2 g ∥ , \left\|g_{s,j}(T)^{1/2}g\right\|\le2\left\|g_{s,k}(T)^{1/2}g\right\|, g s , j ( T ) 1/2 g ≤ 2 g s , k ( T ) 1/2 g ,
∥ ( 1 + T ) 1 / 2 ( g s , j ( T ) 1 / 2 − g s , k ( T ) 1 / 2 ) g ∥ ≤ ∥ g s , k ( T ) 1 / 2 g ∥ . (68) \left\|(1+T)^{1/2}\left(g_{s,j}(T)^{1/2}-g_{s,k}(T)^{1/2}\right)g\right\|\le\left\|g_{s,k}(T)^{1/2}g\right\|. \tag*{(68)} ( 1 + T ) 1/2 ( g s , j ( T ) 1/2 − g s , k ( T ) 1/2 ) g ≤ g s , k ( T ) 1/2 g . (68)
Since the Fourier multipliers commute with ∂ \partial ∂ , expansion gives
g s , j ( T ) 1 / 2 B − B g s , k ( T ) 1 / 2 = [ g s , j ( T ) 1 / 2 , a ] ∂ + 1 2 [ g s , j ( T ) 1 / 2 , a ′ ] + B ( g s , j ( T ) 1 / 2 − g s , k ( T ) 1 / 2 ) . (69) \begin{aligned}
g_{s,j}(T)^{1/2}B-Bg_{s,k}(T)^{1/2}
={}&[g_{s,j}(T)^{1/2},a]\partial+\frac{1}{2}[g_{s,j}(T)^{1/2},a']\\
&+B\left(g_{s,j}(T)^{1/2}-g_{s,k}(T)^{1/2}\right). \tag*{(69)}
\end{aligned} g s , j ( T ) 1/2 B − B g s , k ( T ) 1/2 = [ g s , j ( T ) 1/2 , a ] ∂ + 2 1 [ g s , j ( T ) 1/2 , a ′ ] + B ( g s , j ( T ) 1/2 − g s , k ( T ) 1/2 ) . (69)
The first two terms are bounded by (57) and (68) , using ∥ ( 1 + T ) − 1 / 2 ∥ ≤ 1 \|(1+T)^{-1/2}\|\le1 ∥ ( 1 + T ) − 1/2 ∥ ≤ 1 and ∥ ∂ ( 1 + T ) − 1 / 2 ∥ ≤ 1 \|\partial(1+T)^{-1/2}\|\le1 ∥ ∂ ( 1 + T ) − 1/2 ∥ ≤ 1 . For the last term, use ∥ B h ∥ ≤ C ∥ a ∥ C 1 ( I ) ∥ ( 1 + T ) 1 / 2 h ∥ \|Bh\|\le C\|a\|_{C^1(I)}\|(1+T)^{1/2}h\| ∥ B h ∥ ≤ C ∥ a ∥ C 1 ( I ) ∥ ( 1 + T ) 1/2 h ∥ and the second bound in (68) . We obtain
∥ ( g s , j ( T ) 1 / 2 B − B g s , k ( T ) 1 / 2 ) g ∥ ≤ C ∥ a ∥ C s + 3 ( I ) ∥ g s , k ( T ) 1 / 2 g ∥ . (70) \left\|\left(g_{s,j}(T)^{1/2}B-Bg_{s,k}(T)^{1/2}\right)g\right\|
\le C\|a\|_{C^{s+3}(I)}\left\|g_{s,k}(T)^{1/2}g\right\|. \tag*{(70)} ( g s , j ( T ) 1/2 B − B g s , k ( T ) 1/2 ) g ≤ C ∥ a ∥ C s + 3 ( I ) g s , k ( T ) 1/2 g . (70)
The same estimate holds with j , k j,k j , k exchanged. All multipliers preserve smoothness, so integration by parts is valid in the following identity:
⟨ f , ( g s , j ( T ) B − B g s , k ( T ) ) g ⟩ = ⟨ g s , j ( T ) 1 / 2 f , ( g s , j ( T ) 1 / 2 B − B g s , k ( T ) 1 / 2 ) g ⟩ + ⟨ ( g s , k ( T ) 1 / 2 B − B g s , j ( T ) 1 / 2 ) f , g s , k ( T ) 1 / 2 g ⟩ . (71) \begin{aligned}
\langle f,(g_{s,j}(T)B-Bg_{s,k}(T))g\rangle
={}&\langle g_{s,j}(T)^{1/2}f,(g_{s,j}(T)^{1/2}B-Bg_{s,k}(T)^{1/2})g\rangle\\
&+\langle(g_{s,k}(T)^{1/2}B-Bg_{s,j}(T)^{1/2})f,g_{s,k}(T)^{1/2}g\rangle. \tag*{(71)}
\end{aligned} ⟨ f , ( g s , j ( T ) B − B g s , k ( T )) g ⟩ = ⟨ g s , j ( T ) 1/2 f , ( g s , j ( T ) 1/2 B − B g s , k ( T ) 1/2 ) g ⟩ + ⟨( g s , k ( T ) 1/2 B − B g s , j ( T ) 1/2 ) f , g s , k ( T ) 1/2 g ⟩ . (71)
Indeed, the two additional terms cancel because ⟨ B h 1 , h 2 ⟩ = − ⟨ h 1 , B h 2 ⟩ \langle Bh_1,h_2\rangle=-\langle h_1,Bh_2\rangle ⟨ B h 1 , h 2 ⟩ = − ⟨ h 1 , B h 2 ⟩ for smooth periodic h 1 , h 2 h_1,h_2 h 1 , h 2 , as a a a is real-valued. Cauchy–Schwarz and (70) now prove Lemma 3.3.
The kinetic contribution
Define the real coefficients
a j k = χ j ′ χ k − χ j χ k ′ , S = ∑ k ∣ χ k ′ ∣ 2 . (72) a_{jk}=\chi'_j\chi_k-\chi_j\chi'_k,\qquad S=\sum_k|\chi'_k|^2. \tag*{(72)} a j k = χ j ′ χ k − χ j χ k ′ , S = k ∑ ∣ χ k ′ ∣ 2 . (72)
Moreover, a k j = − a j k a_{kj}=-a_{jk} a k j = − a j k , and a j k = 0 a_{jk}=0 a j k = 0 unless the supports of χ j \chi_j χ j and χ k \chi_k χ k overlap, which forces ∥ j − k ∥ ≤ 1 \lVert j-k\rVert\le1 ∥ j − k ∥ ≤ 1 . Notice that S S S is 1-periodic on T L T_L T L , and its restriction to each interval is a smooth multiplication operator.
Proof of Proposition 3.1 . Since the kinetic energy K K K is translation invariant, we have
[ A s , j 0 , K ] = e − j 0 ∂ [ A s , 0 , K ] e j 0 ∂ [A_{s,j_0},K]=e^{-j_0\partial}[A_{s,0},K]e^{j_0\partial} [ A s , j 0 , K ] = e − j 0 ∂ [ A s , 0 , K ] e j 0 ∂
Therefore, since D \mathcal{D} D is an invariant set under translations, we may, without loss of generality, set j 0 = 0 j_0=0 j 0 = 0 for the rest of the proof.
Secondly, the operator A s , 0 A_{s,0} A s , 0 does not connect different particle number sectors, therefore it suffices to prove the claim on each sector separately.
We fix N ≥ 1 N\geq1 N ≥ 1 and consider u ∈ C ∞ ( T L N ) ∩ H N u\in C^\infty(\mathbb{T}_L^N)\cap\mathcal{H}_N u ∈ C ∞ ( T L N ) ∩ H N . The definition of A s , 0 A_{s,0} A s , 0 gives
⟨ u , [ A s , 0 , K ] u ⟩ = ∑ i = 1 N ∑ j = 0 L − 1 ⟨ u , [ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , K N ] u ⟩ (73) \langle u,[A_{s,0},K]u\rangle=\sum_{i=1}^{N}\sum_{j=0}^{L-1}\langle u,[\chi_j(x_i)g_{s,j}(T_j)\chi_j(x_i),K_N]u\rangle \tag*{(73)} ⟨ u , [ A s , 0 , K ] u ⟩ = i = 1 ∑ N j = 0 ∑ L − 1 ⟨ u , [ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , K N ] u ⟩ (73)
where g s , j ( T j ) g_{s,j}(T_j) g s , j ( T j ) acts in coordinate x i x_i x i . Since K N = − ∑ m = 1 N ∂ m 2 K_N=-\sum_{m=1}^{N}\partial_m^2 K N = − ∑ m = 1 N ∂ m 2 , and operators in different coordinates commute,
[ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , − ∂ m 2 ] = 0 ( m ≠ i ) . (74) [\chi_j(x_i)g_{s,j}(T_j)\chi_j(x_i),-\partial_m^2]=0\qquad(m\ne i). \tag*{(74)} [ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , − ∂ m 2 ] = 0 ( m = i ) . (74)
We are therefore left with
⟨ u , [ A s , 0 , K ] u ⟩ = ∑ i = 1 N ∑ j ⟨ u , [ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , − ∂ i 2 ] u ⟩ . (75) \langle u,[A_{s,0},K]u\rangle=\sum_{i=1}^{N}\sum_j\langle u,[\chi_j(x_i)g_{s,j}(T_j)\chi_j(x_i),-\partial_i^2]u\rangle. \tag*{(75)} ⟨ u , [ A s , 0 , K ] u ⟩ = i = 1 ∑ N j ∑ ⟨ u , [ χ j ( x i ) g s , j ( T j ) χ j ( x i ) , − ∂ i 2 ] u ⟩ . (75)
Step 1. Fix i i i and regard u u u as a smooth function of x i x_i x i with values in g = L 2 ( T L N − 1 ) \mathfrak{g}=L^2(\mathbb{T}_L^{N-1}) g = L 2 ( T L N − 1 ) . We write primes for derivatives in x i x_i x i and set w j = χ j u ∈ C p e r ∞ ( I j ; g ) w_j=\chi_j u\in C^\infty_{\mathrm{per}}(I_j;\mathfrak{g}) w j = χ j u ∈ C per ∞ ( I j ; g ) .
The product rule and the partition identities ∑ k χ k 2 = 1 \sum_k\chi_k^2=1 ∑ k χ k 2 = 1 , ∑ k χ k χ k ′ = 0 \sum_k\chi_k\chi_k'=0 ∑ k χ k χ k ′ = 0 underlying IMS [[9 ] , § 3.1 ] give
χ j ( − u ′ ′ ) = T j w j + S w j + ∑ k B j k w k , (76) \chi_j(-u'')=T_jw_j+Sw_j+\sum_k B_{jk}w_k, \tag*{(76)} χ j ( − u ′′ ) = T j w j + S w j + k ∑ B j k w k , (76)
with B j k = 2 a j k ∂ + a j k ′ B_{jk}=2a_{jk}\partial+a'_{jk} B j k = 2 a j k ∂ + a j k ′ , as one may check by differentiating ∑ k χ k 2 = 1 \sum_k\chi_k^2=1 ∑ k χ k 2 = 1 twice.
Step 2. We estimate the terms in (3.25) using § 3.1 . The T j T_j T j terms cancel because T j T_j T j commutes with its Fourier multipliers. For the S S S contribution, (3.4) with v = S v=S v = S and ( 1 + T j ) − 1 ≤ 1 (1+T_j)^{-1}\leq1 ( 1 + T j ) − 1 ≤ 1 yields
∣ ⟨ w j , [ g s , j ( T j ) , S ] w j ⟩ ∣ ≤ C s ∥ g s , j ( T j ) 1 / 2 w j ∥ 2 . (77) \left|\langle w_j,[g_{s,j}(T_j),S]w_j\rangle\right|\leq C_s\|g_{s,j}(T_j)^{1/2}w_j\|^2. \tag*{(77)} ∣ ⟨ w j , [ g s , j ( T j ) , S ] w j ⟩ ∣ ≤ C s ∥ g s , j ( T j ) 1/2 w j ∥ 2 . (77)
It remains to estimate the first-order terms1 . Integration by parts gives ⟨ B j k w k , w j ⟩ = ⟨ w k , B k j w j ⟩ \langle B_{jk}w_k,w_j\rangle=\langle w_k,B_{kj}w_j\rangle ⟨ B j k w k , w j ⟩ = ⟨ w k , B k j w j ⟩ . Using (3.25) in both terms of the commutator and reindexing the second first-order sum therefore gives
∑ j ⟨ u , [ χ j g s , j ( T j ) χ j , − ∂ 2 ] u ⟩ = ∑ j ⟨ w j , [ g s , j ( T j ) , S ] w j ⟩ + ∑ j , k ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k ) ) w k ⟩ . (78) \sum_j\langle u,[\chi_jg_{s,j}(T_j)\chi_j,-\partial^2]u\rangle
=\sum_j\langle w_j,[g_{s,j}(T_j),S]w_j\rangle
+\sum_{j,k}\langle w_j,(g_{s,j}(T_j)B_{jk}-B_{jk}g_{s,k}(T_k))w_k\rangle. \tag*{(78)} j ∑ ⟨ u , [ χ j g s , j ( T j ) χ j , − ∂ 2 ] u ⟩ = j ∑ ⟨ w j , [ g s , j ( T j ) , S ] w j ⟩ + j , k ∑ ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k )) w k ⟩ . (78)
f ( r ) ⟼ f ( r − ( k − j ) ) , f ∈ L 2 ( R / ( 4 Z ) ; g ) , (79) f(r)\longmapsto f(r-(k-j)),\qquad f\in L^2(\mathbb{R}/(4\mathbb{Z});\mathfrak{g}), \tag*{(79)} f ( r ) ⟼ f ( r − ( k − j )) , f ∈ L 2 ( R / ( 4 Z ) ; g ) , (79)
expresses a pseudosite-k k k function in the pseudosite-j j j coordinate. It is unitary and commutes with differentiation and every Fourier multiplier used. Both localized supports lie in the common interval interior. All paired pseudosite terms below, including Fourier-multiplier outputs, are understood under this identification.
Since ∣ ⟨ j ⟩ − ⟨ k ⟩ ∣ ≤ 1 |\langle j\rangle-\langle k\rangle|\le1 ∣ ⟨ j ⟩ − ⟨ k ⟩ ∣ ≤ 1 for overlapping intervals, we may apply Lemma 3.3 , with f = w j f=w_j f = w j , g = w k g=w_k g = w k , a = 2 a j k a=2a_{jk} a = 2 a j k , and b = ⟨ j ⟩ b=\langle j\rangle b = ⟨ j ⟩ , c = ⟨ k ⟩ c=\langle k\rangle c = ⟨ k ⟩ . Together with the uniform bounds on the derivatives of a j k a_{jk} a j k , it yields
∣ ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k ) ) w k ⟩ ∣ ≤ C s ∥ g s , j ( T j ) 1 / 2 w j ∥ ∥ g s , k ( T k ) 1 / 2 w k ∥ . (80) \left|\langle w_j,(g_{s,j}(T_j)B_{jk}-B_{jk}g_{s,k}(T_k))w_k\rangle\right|
\le C_s\|g_{s,j}(T_j)^{1/2}w_j\|\|g_{s,k}(T_k)^{1/2}w_k\|. \tag*{(80)} ∣ ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k )) w k ⟩ ∣ ≤ C s ∥ g s , j ( T j ) 1/2 w j ∥∥ g s , k ( T k ) 1/2 w k ∥. (80)
Summing (80) over the ordered pairs, and using that only neighbouring intervals contribute, gives
∣ ∑ j , k ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k ) ) w k ⟩ ∣ ≤ C s ∑ j ∥ g s , j ( T j ) 1 / 2 w j ∥ ∥ g s , j + 1 ( T j + 1 ) 1 / 2 w j + 1 ∥ ≤ C s 2 ∑ j ( ∥ g s , j ( T j ) 1 / 2 w j ∥ 2 + ∥ g s , j + 1 ( T j + 1 ) 1 / 2 w j + 1 ∥ 2 ) = C s ∑ j ∥ g s , j ( T j ) 1 / 2 w j ∥ 2 , (81) \begin{aligned}
\left|\sum_{j,k}\langle w_j,(g_{s,j}(T_j)B_{jk}-B_{jk}g_{s,k}(T_k))w_k\rangle\right|
\le C_s\sum_j\|g_{s,j}(T_j)^{1/2}w_j\|\|g_{s,j+1}(T_{j+1})^{1/2}w_{j+1}\| \\
&\le\frac{C_s}{2}\sum_j\left(\|g_{s,j}(T_j)^{1/2}w_j\|^2+\|g_{s,j+1}(T_{j+1})^{1/2}w_{j+1}\|^2\right) \\
&= C_s\sum_j\|g_{s,j}(T_j)^{1/2}w_j\|^2, \tag*{(81)}
\end{aligned} j , k ∑ ⟨ w j , ( g s , j ( T j ) B j k − B j k g s , k ( T k )) w k ⟩ ≤ C s j ∑ ∥ g s , j ( T j ) 1/2 w j ∥∥ g s , j + 1 ( T j + 1 ) 1/2 w j + 1 ∥ ≤ 2 C s j ∑ ( ∥ g s , j ( T j ) 1/2 w j ∥ 2 + ∥ g s , j + 1 ( T j + 1 ) 1/2 w j + 1 ∥ 2 ) = C s j ∑ ∥ g s , j ( T j ) 1/2 w j ∥ 2 , (81)
with the understanding that the summations are taken modulo L L L . Combining (81) with (77) in (78) , we obtain
∣ ∑ j ⟨ u , [ χ j g s , j ( T j ) χ j , − ∂ 2 ] u ⟩ ∣ ≤ C s ∑ j ∥ g s , j ( T j ) 1 / 2 w j ∥ 2 . (82) \left|\sum_j\langle u,[\chi_jg_{s,j}(T_j)\chi_j,-\partial^2]u\rangle\right|
\le C_s\sum_j\|g_{s,j}(T_j)^{1/2}w_j\|^2. \tag*{(82)} j ∑ ⟨ u , [ χ j g s , j ( T j ) χ j , − ∂ 2 ] u ⟩ ≤ C s j ∑ ∥ g s , j ( T j ) 1/2 w j ∥ 2 . (82)
Finally, summing over i i i produces
∣ ⟨ u , [ A s , 0 , K ] u ⟩ ∣ ≤ C s ∑ i = 1 N ∑ j ∥ g s , j ( T j ) 1 / 2 ( χ j ( x i ) u ) ∥ 2 = C s A s , 0 [ u ] . (83) |\langle u,[A_{s,0},K]u\rangle|
\le C_s\sum_{i=1}^{N}\sum_j\|g_{s,j}(T_j)^{1/2}(\chi_j(x_i)u)\|^2
= C_sA_{s,0}[u]. \tag*{(83)} ∣ ⟨ u , [ A s , 0 , K ] u ⟩ ∣ ≤ C s i = 1 ∑ N j ∑ ∥ g s , j ( T j ) 1/2 ( χ j ( x i ) u ) ∥ 2 = C s A s , 0 [ u ] . (83)
Orthogonal summation proves the assertion for general u ∈ D u\in\mathcal{D} u ∈ D .
Propagation estimates: interaction contribution
In this section we prove the following bound on the contribution to the time derivative of the MALO, coming from the two-body interaction. At the end of this section we also provide the proof of Proposition 2.6.
Proposition 4.1 (Interaction contribution). For u ∈ D u\in\mathcal{D} u ∈ D , ∣ ⟨ u , [ A s , j 0 , V ] u ⟩ ∣ ≤ C s A s , j 0 [ u ] |\langle u,[A_{s,j_0},V]u\rangle|\le C_sA_{s,j_0}[u] ∣ ⟨ u , [ A s , j 0 , V ] u ⟩ ∣ ≤ C s A s , j 0 [ u ] .
Exclusion estimate in the pseudosite modes
Before moving on to prove Proposition 4.1 , we first prove the following lemma, which will play an important role in the proof of Proposition 4.1 .
Lemma 4.2 (Exclusion on pseudosite sets). Let ∅ ≠ C ⊂ { 0 , … , L − 1 } \varnothing\ne C\subset\{0,\ldots,L-1\} ∅ = C ⊂ { 0 , … , L − 1 } be a set of pseudosites, and let N ~ C \widetilde{\mathcal{N}}_C N C denote the number operator on h C \mathfrak{h}_C h C .
For any α ≥ 0 \alpha\ge0 α ≥ 0 and every nonnegative nondecreasing Borel function f : [ 0 , ∞ ) → [ 0 , ∞ ) f:[0,\infty)\to[0,\infty) f : [ 0 , ∞ ) → [ 0 , ∞ ) ,
N ~ C α d Γ ( f ( T C ) ) ≤ C α , ∣ C ∣ d Γ ( ( 1 + T C ) α / 2 f ( T C ) ) . (84) \widetilde{\mathcal{N}}_C^\alpha\mathrm{d}\Gamma(f(T_C))
\le C_{\alpha,|C|}\mathrm{d}\Gamma\left((1+T_C)^{\alpha/2}f(T_C)\right). \tag*{(84)} N C α d Γ ( f ( T C )) ≤ C α , ∣ C ∣ d Γ ( ( 1 + T C ) α /2 f ( T C ) ) . (84) in the sense of spectral forms.
Proof. We first count the one-particle modes, then use their occupation numbers to absorb the particle-number factor into the derivative weight. The number of eigenmodes of T C T_{\mathcal C} T C with eigenvalue at most Λ ≥ 0 \Lambda\ge0 Λ ≥ 0 , counted with multiplicity, is
N C ( Λ ) = ∣ C ∣ ( 1 + 2 ⌊ 2 Λ π ⌋ ) ≤ C ∣ C ∣ ( 1 + Λ ) 1 / 2 . (85) \mathcal{N}_{\mathcal C}(\Lambda)=|\mathcal C|\left(1+2\left\lfloor\frac{2\sqrt{\Lambda}}{\pi}\right\rfloor\right)\le C|\mathcal C|(1+\Lambda)^{1/2}. \tag*{(85)} N C ( Λ ) = ∣ C ∣ ( 1 + 2 ⌊ π 2 Λ ⌋ ) ≤ C ∣ C ∣ ( 1 + Λ ) 1/2 . (85)
In an occupation vector with n ≥ 1 n\ge1 n ≥ 1 particles in these intervals, list their eigenvalues as λ 1 ≤ ⋯ ≤ λ n \lambda_1\le\cdots\le\lambda_n λ 1 ≤ ⋯ ≤ λ n . No eigenmode is repeated, so
i ≤ N C ( λ i ) ≤ C ∣ C ∣ ( 1 + λ i ) 1 / 2 , i\le\mathcal{N}_{\mathcal C}(\lambda_i)\le C|\mathcal C|(1+\lambda_i)^{1/2}, i ≤ N C ( λ i ) ≤ C ∣ C ∣ ( 1 + λ i ) 1/2 ,
1 n ∑ i = 1 n ( 1 + λ i ) α / 2 ≥ c α , ∣ C ∣ n ∑ i = 1 n i α ≥ c α , ∣ C ∣ ′ n α . (86) \frac{1}{n}\sum_{i=1}^{n}(1+\lambda_i)^{\alpha/2}\ge\frac{c_{\alpha,|\mathcal C|}}{n}\sum_{i=1}^{n}i^\alpha\ge c'_{\alpha,|\mathcal C|}n^\alpha. \tag*{(86)} n 1 i = 1 ∑ n ( 1 + λ i ) α /2 ≥ n c α , ∣ C ∣ i = 1 ∑ n i α ≥ c α , ∣ C ∣ ′ n α . (86)
For nondecreasing sequences ( a i ) (a_i) ( a i ) and ( b i ) (b_i) ( b i ) ,
n ∑ i a i b i − ( ∑ i a i ) ( ∑ i b i ) = ∑ i < j ( a j − a i ) ( b j − b i ) ≥ 0. (87) n\sum_i a_i b_i-\left(\sum_i a_i\right)\left(\sum_i b_i\right)=\sum_{i<j}(a_j-a_i)(b_j-b_i)\ge0. \tag*{(87)} n i ∑ a i b i − ( i ∑ a i ) ( i ∑ b i ) = i < j ∑ ( a j − a i ) ( b j − b i ) ≥ 0. (87)
Apply this identity with a i = ( 1 + λ i ) α / 2 a_i=(1+\lambda_i)^{\alpha/2} a i = ( 1 + λ i ) α /2 and b i = f ( λ i ) ≥ 0 b_i=f(\lambda_i)\ge0 b i = f ( λ i ) ≥ 0 . The preceding average bound gives
n α ∑ i f ( λ i ) ≤ C ∑ i ( 1 + λ i ) α / 2 f ( λ i ) . (88) n^\alpha\sum_i f(\lambda_i)\le C\sum_i(1+\lambda_i)^{\alpha/2}f(\lambda_i). \tag*{(88)} n α i ∑ f ( λ i ) ≤ C i ∑ ( 1 + λ i ) α /2 f ( λ i ) . (88)
When n = 0 n=0 n = 0 , both sides of (84) vanish. Expansion in the occupation basis therefore proves the inequality.
Proof of interaction contribution
Proof of Proposition 4.1. Since V V V preserves particle number and A s , j 0 A_{s,j_0} A s , j 0 does not connect different sectors, it suffices to take u ∈ C ∞ ( T L N ) ∩ H N u\in C^\infty(\mathbb{T}_L^N)\cap\mathcal{H}_N u ∈ C ∞ ( T L N ) ∩ H N , for any given N ≥ 1 N\ge1 N ≥ 1 .
Step 1. We first localize the interaction involving a selected coordinate. We set Φ = L u \Phi=\mathcal{L}u Φ = L u and we write
j = ( j 1 , … , j N ) ∈ { 0 , … , L − 1 } N , I j = ∏ k = 1 N I j k . (89) j=(j_1,\ldots,j_N)\in\{0,\ldots,L-1\}^N,\qquad I_j=\prod_{k=1}^{N}I_{j_k}. \tag*{(89)} j = ( j 1 , … , j N ) ∈ { 0 , … , L − 1 } N , I j = k = 1 ∏ N I j k . (89)
The definition of A s , j 0 A_{s,j_0} A s , j 0 gives
⟨ u , [ A s , j 0 , V ] u ⟩ = ∑ i = 1 N ∑ j ⟨ u , [ χ j ( x i ) g s , j − j 0 ( T j ) χ j ( x i ) , V ] u ⟩ , (90) \langle u,[A_{s,j_0},V]u\rangle=\sum_{i=1}^{N}\sum_j\langle u,[\chi_j(x_i)g_{s,j-j_0}(T_j)\chi_j(x_i),V]u\rangle, \tag*{(90)} ⟨ u , [ A s , j 0 , V ] u ⟩ = i = 1 ∑ N j ∑ ⟨ u , [ χ j ( x i ) g s , j − j 0 ( T j ) χ j ( x i ) , V ] u ⟩ , (90)
where g s , j − j 0 ( T j ) g_{s,j-j_0}(T_j) g s , j − j 0 ( T j ) acts in coordinate i i i . Moreover,
( L ( V u ) ) j = ( ∑ 1 ≤ m < n ≤ N W ( x m − x n ) ) Φ j . (91) (\mathcal{L}(Vu))_j=\left(\sum_{1\le m<n\le N}W(x_m-x_n)\right)\Phi_j. \tag*{(91)} ( L ( V u ) ) j = ( 1 ≤ m < n ≤ N ∑ W ( x m − x n ) ) Φ j . (91)
Given an interval label j j j , let
C j = { k : supp χ k ∩ B r 0 + 2 ( j ) ≠ ∅ } (92) \mathcal C_j=\{k:\operatorname{supp}\chi_k\cap B_{r_0+2}(j)\ne\varnothing\} \tag*{(92)} C j = { k : supp χ k ∩ B r 0 + 2 ( j ) = ∅ } (92)
be a neighbourhood containing all interval labels that can contribute to the cutoff interaction. For k ∈ C j k\in\mathcal C_j k ∈ C j , ∥ j − k ∥ ≤ r 0 + 3 \lVert j-k\rVert\le r_0+3 ∥ j − k ∥ ≤ r 0 + 3 ; since the centers have unit spacing, this gives
sup j ∣ C j ∣ ≤ C , sup k # { j : k ∈ C j } ≤ C . (93) \sup_j|\mathcal C_j|\le C,\qquad\sup_k\#\{j:k\in\mathcal C_j\}\le C. \tag*{(93)} j sup ∣ C j ∣ ≤ C , k sup # { j : k ∈ C j } ≤ C . (93)
with the constant C C C being independent of N N N , L L L , j 0 j_0 j 0 . Fix i ∈ { 1 , … , N } i \in\{1,\ldots,N\} i ∈ { 1 , … , N } and j j j , and x m ∈ supp χ j m x_m \in\operatorname{supp}\chi_{j_m} x m ∈ supp χ j m for m ≠ i m \ne i m = i . For x i ∈ supp χ j i ⊂ B 7 / 8 ( j i ) x_i \in\operatorname{supp}\chi_{j_i} \subset B_{7/8}(j_i) x i ∈ supp χ j i ⊂ B 7/8 ( j i ) , finite range gives W ( x i − x m ) = 0 W(x_i-x_m)=0 W ( x i − x m ) = 0 whenever j m ∉ C j i j_m \notin\mathcal{C}_{j_i} j m ∈ / C j i . With the remaining coordinates held fixed, we write
w i , j ( x i ) = Φ j ( x 1 , … , x N ) , v i , j ( x i ) = η j i ( x i ) ∑ m ≠ i W ( x i − x m ) , n i , j = ∑ m ≠ i 1 C j i ( j m ) . (94) \begin{aligned}
w_{i,j}(x_i)&=\Phi_j(x_1,\ldots,x_N),\qquad v_{i,j}(x_i)=\sqrt{\eta_{j_i}(x_i)}\sum_{m\ne i}W(x_i-x_m),\\
n_{i,j}&=\sum_{m\ne i}\mathbf{1}_{\mathcal{C}_{j_i}}(j_m). \tag*{(94)}
\end{aligned} w i , j ( x i ) n i , j = Φ j ( x 1 , … , x N ) , v i , j ( x i ) = η j i ( x i ) m = i ∑ W ( x i − x m ) , = m = i ∑ 1 C j i ( j m ) . (94)
Since the support of w i , j w_{i,j} w i , j is contained in the support of χ j i \chi_{j_i} χ j i , and since η j i \eta_{j_i} η j i is constant on it, we have
( 1 − η j i ) w i , j = 0. (95) (1-\sqrt{\eta_{j_i}})w_{i,j}=0. \tag*{(95)} ( 1 − η j i ) w i , j = 0. (95)
For the rest of the proof, we adopt the following notation:
I j ∖ { i } = ∏ m ≠ i I j m , d x i ^ = ∏ m ≠ i d x m , x m ∈ I j m . (96) I_{j\setminus\{i\}}=\prod_{m\ne i}I_{j_m},\qquad\mathrm{d}\widehat{x_i}=\prod_{m\ne i}\mathrm{d}x_m,\qquad x_m\in I_{j_m}. \tag*{(96)} I j ∖ { i } = m = i ∏ I j m , d x i = m = i ∏ d x m , x m ∈ I j m . (96)
Terms in V V V not involving coordinate x i x_i x i commute with the MALO g s , j i − j 0 ( T j i ) g_{s,j_i-j_0}(T_{j_i}) g s , j i − j 0 ( T j i ) . Combining this with (95) and the evenness of W W W , we can therefore write (90) as follows:
⟨ u , [ A s , j 0 , V ] u ⟩ = ∑ i = 1 N ∑ j ∫ I j ∖ { i } ⟨ w i , j , [ g s , j i − j 0 ( T j i ) , v i , j ] w i , j ⟩ d x i ^ . (97) \langle u,[A_{s,j_0},V]u\rangle=\sum_{i=1}^{N}\sum_j\int_{I_{j\setminus\{i\}}}\langle w_{i,j},[g_{s,j_i-j_0}(T_{j_i}),v_{i,j}]w_{i,j}\rangle\,\mathrm{d}\widehat{x_i}. \tag*{(97)} ⟨ u , [ A s , j 0 , V ] u ⟩ = i = 1 ∑ N j ∑ ∫ I j ∖ { i } ⟨ w i , j , [ g s , j i − j 0 ( T j i ) , v i , j ] w i , j ⟩ d x i . (97)
Step 2. We apply the multiplication estimate. We set f j ( λ ) : = g s , j − j 0 ( λ ) / ( 1 + λ ) f_j(\lambda):=g_{s,j-j_0}(\lambda)/(1+\lambda) f j ( λ ) := g s , j − j 0 ( λ ) / ( 1 + λ ) . Since η j i \eta_{j_i} η j i vanishes outside B 2 ( j i ) B_2(j_i) B 2 ( j i ) , at most n i , j n_{i,j} n i , j summands in v i , j v_{i,j} v i , j can be nonzero. Furthermore, uniform bounds on the derivatives of η j i \sqrt{\eta_{j_i}} η j i and W W W give
∥ v i , j ∥ C s + 2 ( I j i ) ≤ C s n i , j . (98) \|v_{i,j}\|_{C^{s+2}(I_{j_i})}\le C_s n_{i,j}. \tag*{(98)} ∥ v i , j ∥ C s + 2 ( I j i ) ≤ C s n i , j . (98)
Therefore, at fixed i i i , j j j and remaining positions, (55) and the bound on v i , j v_{i,j} v i , j above give
∣ ⟨ w i , j , [ g s , j i − j 0 ( T j i ) , v i , j ] w i , j ⟩ ∣ ≤ C s n i , j ∥ g s , j i − j 0 ( T j i ) 1 / 2 w i , j ∥ ∥ f j i ( T j i ) 1 / 2 w i , j ∥ . (99) \left|\langle w_{i,j},[g_{s,j_i-j_0}(T_{j_i}),v_{i,j}]w_{i,j}\rangle\right|\le C_s n_{i,j}\|g_{s,j_i-j_0}(T_{j_i})^{1/2}w_{i,j}\|\|f_{j_i}(T_{j_i})^{1/2}w_{i,j}\|. \tag*{(99)} ∣ ⟨ w i , j , [ g s , j i − j 0 ( T j i ) , v i , j ] w i , j ⟩ ∣ ≤ C s n i , j ∥ g s , j i − j 0 ( T j i ) 1/2 w i , j ∥∥ f j i ( T j i ) 1/2 w i , j ∥. (99)
By the definition of A s , j 0 A_{s,j_0} A s , j 0 ,
∑ i = 1 N ∑ j ∫ I j ∖ { i } ∥ g s , j i − j 0 ( T j i ) 1 / 2 w i , j ∥ 2 d x i ^ = A s , j 0 [ u ] . (100) \sum_{i=1}^{N}\sum_j\int_{I_{j\setminus\{i\}}}\|g_{s,j_i-j_0}(T_{j_i})^{1/2}w_{i,j}\|^2\,\mathrm{d}\widehat{x_i}=A_{s,j_0}[u]. \tag*{(100)} i = 1 ∑ N j ∑ ∫ I j ∖ { i } ∥ g s , j i − j 0 ( T j i ) 1/2 w i , j ∥ 2 d x i = A s , j 0 [ u ] . (100)
Inserting (99) into (97) and then applying Cauchy–Schwarz over i i i , j j j , and the remaining positions yields
∣ ⟨ u , [ A s , j 0 , V ] u ⟩ ∣ ≤ C s ∑ i = 1 N ∑ j ∫ I j ∖ { i } n i , j ∥ g s , j i − j 0 ( T j i ) 1 / 2 w i , j ∥ ∥ f j i ( T j i ) 1 / 2 w i , j ∥ d x i ^ ≤ C s A s , j 0 [ u ] 1 / 2 ( ∑ i = 1 N ∑ j ∫ I j ∖ { i } n i , j 2 ∥ f j i ( T j i ) 1 / 2 w i , j ∥ 2 d x i ^ ) 1 / 2 . (101) \begin{aligned}
|\langle u,[A_{s,j_0},V]u\rangle|
&\le C_s\sum_{i=1}^{N}\sum_j\int_{I_{j\setminus\{i\}}}n_{i,j}\|g_{s,j_i-j_0}(T_{j_i})^{1/2}w_{i,j}\|\|f_{j_i}(T_{j_i})^{1/2}w_{i,j}\|\,\mathrm{d}\widehat{x_i}\\
&\le C_s A_{s,j_0}[u]^{1/2}\left(\sum_{i=1}^{N}\sum_j\int_{I_{j\setminus\{i\}}}n_{i,j}^{2}\|f_{j_i}(T_{j_i})^{1/2}w_{i,j}\|^{2}\,\mathrm{d}\widehat{x_i}\right)^{1/2}. \tag*{(101)}
\end{aligned} ∣ ⟨ u , [ A s , j 0 , V ] u ⟩ ∣ ≤ C s i = 1 ∑ N j ∑ ∫ I j ∖ { i } n i , j ∥ g s , j i − j 0 ( T j i ) 1/2 w i , j ∥∥ f j i ( T j i ) 1/2 w i , j ∥ d x i ≤ C s A s , j 0 [ u ] 1/2 ( i = 1 ∑ N j ∑ ∫ I j ∖ { i } n i , j 2 ∥ f j i ( T j i ) 1/2 w i , j ∥ 2 d x i ) 1/2 . (101)
Since j i ∈ C j i j_i\in\mathcal{C}_{j_i} j i ∈ C j i , the number operator of Lemma 4.2 satisfies
( N ~ C j i Φ ) j = ( n i , j + 1 ) Φ j . (102) (\widetilde{\mathcal{N}}_{\mathcal{C}_{j_i}}\Phi)_j=(n_{i,j}+1)\Phi_j. \tag*{(102)} ( N C j i Φ ) j = ( n i , j + 1 ) Φ j . (102)
The Fourier multipliers preserve pseudosite labels, so their actions in each coordinate commute with N ~ C j \widetilde{\mathcal N}_{C_j} N C j . Consequently,
∑ i = 1 N ∑ j ∫ J i ∖ { i } ( n i , j + 1 ) 2 ∥ f j ( T j i ) 1 / 2 w i , j ∥ 2 d x ı ^ = ∑ j ∑ i = 1 N ∑ j i : j i = j ∫ J i ∖ { i } ( n i , j + 1 ) 2 ∥ f j ( T j ) 1 / 2 w i , j ∥ 2 d x ı ^ = ∑ j ⟨ Φ , N ~ C j 2 d Γ ( f j ( T j ) ) Φ ⟩ . (103) \sum_{i=1}^{N}\sum_j\int_{J_i\setminus\{i\}}(n_{i,j}+1)^2\lVert f_j(T_{j_i})^{1/2}w_{i,j}\rVert^2\,\mathrm{d}x_{\hat{\imath}}
= \sum_j\sum_{i=1}^{N}\sum_{j_i:j_i=j}\int_{J_i\setminus\{i\}}(n_{i,j}+1)^2\lVert f_j(T_j)^{1/2}w_{i,j}\rVert^2\,\mathrm{d}x_{\hat{\imath}}
= \sum_j\langle\Phi,\widetilde{\mathcal N}_{C_j}^{2}\,\mathrm{d}\Gamma(f_j(T_j))\Phi\rangle. \tag*{(103)} i = 1 ∑ N j ∑ ∫ J i ∖ { i } ( n i , j + 1 ) 2 ∥ f j ( T j i ) 1/2 w i , j ∥ 2 d x ^ = j ∑ i = 1 ∑ N j i : j i = j ∑ ∫ J i ∖ { i } ( n i , j + 1 ) 2 ∥ f j ( T j ) 1/2 w i , j ∥ 2 d x ^ = j ∑ ⟨ Φ , N C j 2 d Γ ( f j ( T j )) Φ ⟩ . (103)
Since n i , j 2 ≤ ( n i , j + 1 ) 2 n_{i,j}^{2}\leq(n_{i,j}+1)^2 n i , j 2 ≤ ( n i , j + 1 ) 2 , inserting this identity into (101) yields
∣ ⟨ u , [ A s , j 0 , V ] u ⟩ ∣ ≤ C s A s , j 0 [ u ] 1 / 2 ( ∑ j ⟨ Φ , N ~ C j 2 d Γ ( f j ( T j ) ) Φ ⟩ ) 1 / 2 . (104) \lvert\langle u,[A_{s,j_0},V]u\rangle\rvert\leq C_sA_{s,j_0}[u]^{1/2}\left(\sum_j\langle\Phi,\widetilde{\mathcal N}_{C_j}^{2}\,\mathrm{d}\Gamma(f_j(T_j))\Phi\rangle\right)^{1/2}. \tag*{(104)} ∣⟨ u , [ A s , j 0 , V ] u ⟩∣ ≤ C s A s , j 0 [ u ] 1/2 ( j ∑ ⟨ Φ , N C j 2 d Γ ( f j ( T j )) Φ ⟩ ) 1/2 . (104)
Step 3. The exclusion estimate controls the last factor in (104) . In f j ( T C j ) f_j(T_{C_j}) f j ( T C j ) , the parameter j j j is fixed on every pseudosite of C j C_j C j . The function f j f_j f j is nondecreasing, because
∂ λ log f j ( λ ) = s 1 + λ − 1 ⟨ j − j 0 ⟩ 2 + λ ≥ s − 1 1 + λ ≥ 0. (105) \partial_\lambda\log f_j(\lambda)=\frac{s}{1+\lambda}-\frac{1}{\langle j-j_0\rangle^2+\lambda}\geq\frac{s-1}{1+\lambda}\geq0. \tag*{(105)} ∂ λ log f j ( λ ) = 1 + λ s − ⟨ j − j 0 ⟩ 2 + λ 1 ≥ 1 + λ s − 1 ≥ 0. (105)
We apply Lemma 4.2 , with the specific choices α = 2 \alpha=2 α = 2 and f = f j f=f_j f = f j , which, together with ( 1 + λ ) f j ( λ ) = g s , j − j 0 ( λ ) (1+\lambda)f_j(\lambda)=g_{s,j-j_0}(\lambda) ( 1 + λ ) f j ( λ ) = g s , j − j 0 ( λ ) , then gives the following form inequalities:
N ~ C j 2 d Γ ( f j ( T j ) ) ≤ N ~ C j 2 d Γ ( f j ( T C j ) ) ≤ C s d Γ ( g s , j − j 0 ( T C j ) ) . (106) \widetilde{\mathcal N}_{C_j}^{2}\,\mathrm{d}\Gamma(f_j(T_j))\leq\widetilde{\mathcal N}_{C_j}^{2}\,\mathrm{d}\Gamma(f_j(T_{C_j}))\leq C_s\mathrm{d}\Gamma(g_{s,j-j_0}(T_{C_j})). \tag*{(106)} N C j 2 d Γ ( f j ( T j )) ≤ N C j 2 d Γ ( f j ( T C j )) ≤ C s d Γ ( g s , j − j 0 ( T C j )) . (106)
The weights on C j C_j C j are uniformly comparable: indeed, since y ↦ ⟨ y − j 0 ⟩ y\mapsto\langle y-j_0\rangle y ↦ ⟨ y − j 0 ⟩ is 1-Lipschitz,
∣ ⟨ j − j 0 ⟩ − ⟨ k − j 0 ⟩ ∣ ≤ ∥ j − k ∥ ≤ r 0 + 3 , k ∈ C j . (107) \left|\langle j-j_0\rangle-\langle k-j_0\rangle\right|\leq\lVert j-k\rVert\leq r_0+3,\qquad k\in C_j. \tag*{(107)} ∣ ⟨ j − j 0 ⟩ − ⟨ k − j 0 ⟩ ∣ ≤ ∥ j − k ∥ ≤ r 0 + 3 , k ∈ C j . (107)
Consequently, we have
g s , j − j 0 ( λ ) g s , k − j 0 ( λ ) = ⟨ k − j 0 ⟩ 2 + λ ⟨ j − j 0 ⟩ 2 + λ ≤ ( r 0 + 4 ) 2 . (108) \frac{g_{s,j-j_0}(\lambda)}{g_{s,k-j_0}(\lambda)}
=\frac{\langle k-j_0\rangle^2+\lambda}{\langle j-j_0\rangle^2+\lambda}\leq(r_0+4)^2. \tag*{(108)} g s , k − j 0 ( λ ) g s , j − j 0 ( λ ) = ⟨ j − j 0 ⟩ 2 + λ ⟨ k − j 0 ⟩ 2 + λ ≤ ( r 0 + 4 ) 2 . (108)
Let us use the shorthand h = ⨁ k g s , k − j 0 ( T k ) h=\bigoplus_k g_{s,k-j_0}(T_k) h = ⨁ k g s , k − j 0 ( T k ) ; since every pseudosite lies in finitely many sets C j C_j C j ,
∑ j g s , j − j 0 ( T C j ) = ⨁ k ∑ j : k ∈ C j g s , j − j 0 ( T k ) ≤ C s ⨁ k g s , k − j 0 ( T k ) = C s h . (109) \sum_jg_{s,j-j_0}(T_{C_j})=\bigoplus_k\sum_{j:k\in C_j}g_{s,j-j_0}(T_k)\leq C_s\bigoplus_kg_{s,k-j_0}(T_k)=C_sh. \tag*{(109)} j ∑ g s , j − j 0 ( T C j ) = k ⨁ j : k ∈ C j ∑ g s , j − j 0 ( T k ) ≤ C s k ⨁ g s , k − j 0 ( T k ) = C s h . (109)
Summing both sides of (106) over j j j , and using (109) we obtain
∑ j N ~ C j 2 d Γ ( f j ( T j ) ) ≤ C s d Γ ( h ) . (110) \sum_j\widetilde{\mathcal N}_{C_j}^{2}\,\mathrm{d}\Gamma(f_j(T_j))\leq C_s\mathrm{d}\Gamma(h). \tag*{(110)} j ∑ N C j 2 d Γ ( f j ( T j )) ≤ C s d Γ ( h ) . (110)
Since ⟨ Φ , d Γ ( h ) Φ ⟩ = A s , j 0 [ u ] \langle\Phi,\mathrm{d}\Gamma(h)\Phi\rangle=A_{s,j_0}[u] ⟨ Φ , d Γ ( h ) Φ ⟩ = A s , j 0 [ u ] , substitution into (104) proves the bound on the N N N -particle sector. For general u ∈ D u\in\mathcal D u ∈ D , it then suffices to decompose it into its sector components.
Proof of main propagation estimate
Proof of Proposition 2.6. We prove (44) by a standard density argument. For u ∈ D u \in\mathcal{D} u ∈ D , combining (54) with Propositions 3.1 and 4.1 and applying Grönwall’s inequality gives
∣ d d t A s , j 0 [ u t ] ∣ = ∣ ⟨ u t , i [ H , A s , j 0 ] u t ⟩ ∣ ≤ C s A s , j 0 [ u t ] , \left|\frac{\mathrm{d}}{\mathrm{d}t}A_{s,j_0}[u_t]\right| = \left|\langle u_t,i[H,A_{s,j_0}]u_t\rangle\right| \le C_s A_{s,j_0}[u_t], d t d A s , j 0 [ u t ] = ∣ ⟨ u t , i [ H , A s , j 0 ] u t ⟩ ∣ ≤ C s A s , j 0 [ u t ] ,
A s , j 0 [ u t ] ≤ e C s ∣ t ∣ A s , j 0 [ u ] . (111) A_{s,j_0}[u_t] \le e^{C_s|t|}A_{s,j_0}[u]. \tag*{(111)} A s , j 0 [ u t ] ≤ e C s ∣ t ∣ A s , j 0 [ u ] . (111)
For u ∈ D ( A s , j 0 ) u \in D(A_{s,j_0}) u ∈ D ( A s , j 0 ) , we pick a sequence ( u n ) n ⊂ D (u_n)_n \subset\mathcal{D} ( u n ) n ⊂ D converging to u u u in the form norm by (39) . Unitarity and lower semicontinuity of the closed form give
A s , j 0 [ e − i t H u ] ≤ lim inf n → ∞ A s , j 0 [ e − i t H u n ] ≤ e C s ∣ t ∣ lim n → ∞ A s , j 0 [ u n ] = e C s ∣ t ∣ A s , j 0 [ u ] . (112) A_{s,j_0}[e^{-itH}u] \le\liminf_{n \to\infty} A_{s,j_0}[e^{-itH}u_n]
\le e^{C_s|t|}\lim_{n \to\infty} A_{s,j_0}[u_n] = e^{C_s|t|}A_{s,j_0}[u]. \tag*{(112)} A s , j 0 [ e − i t H u ] ≤ n → ∞ lim inf A s , j 0 [ e − i t H u n ] ≤ e C s ∣ t ∣ n → ∞ lim A s , j 0 [ u n ] = e C s ∣ t ∣ A s , j 0 [ u ] . (112)
This concludes the proof.
Initial comparison with local moments
We recover the localized derivatives in A s , j 0 A_{s,j_0} A s , j 0 from powers of the local energy and the local particle number. The proof uses the second-order energy estimate, the product rule for powers of the one-coordinate kinetic operators, and their joint spectral calculus with the nonnegative operator in the other coordinates.
Lemma 5.1 (Localized derivative estimates). Let j ∈ { 0 , … , L − 1 } j \in\{0,\ldots,L-1\} j ∈ { 0 , … , L − 1 } . For N , a ∈ N N,a \in\mathbb{N} N , a ∈ N , one has on H N \mathcal{H}_N H N , in the form sense,
d Γ ( χ j ( 1 + T j ) 2 a χ j ) ≤ C a ( E j 2 a + N B j w 2 a ) . (113) \mathrm{d}\Gamma\left(\chi_j(1+T_j)^{2a}\chi_j\right) \le C_a\left(\mathcal{E}_j^{2a}+\mathcal{N}_{B_j^w}^{2a}\right). \tag*{(113)} d Γ ( χ j ( 1 + T j ) 2 a χ j ) ≤ C a ( E j 2 a + N B j w 2 a ) . (113) For u ∈ D ( E j a ) ∩ H N u \in D(\mathcal{E}_j^a)\cap\mathcal{H}_N u ∈ D ( E j a ) ∩ H N , each localized vector χ j ( x i ) u \chi_j(x_i)u χ j ( x i ) u belongs to D ( ( 1 + T j ) α ) D((1+T_j)^\alpha) D (( 1 + T j ) α ) , viewing the function as taking values in the L 2 L^2 L 2 space of the remaining coordinates. The constant is independent of N , L , j N,L,j N , L , j . On Fock space the right-hand form domain is D ( E j a ) ∩ D ( N B j w a ) D(\mathcal{E}_j^a)\cap D(\mathcal{N}_{B_j^w}^a) D ( E j a ) ∩ D ( N B j w a ) .
Proof of Proposition 2.7. We sum the localized derivative bounds with the spatial weights in A s , j 0 A_{s,j_0} A s , j 0 . Fix j 0 ∈ { 0 , … , L − 1 } j_0 \in\{0,\ldots,L-1\} j 0 ∈ { 0 , … , L − 1 } and set a = ⌈ ( s + 1 ) / 2 ⌉ = p / 2 a=\lceil(s+1)/2\rceil=p/2 a = ⌈( s + 1 ) /2 ⌉ = p /2 . The scalar bound g s , j − j 0 ( λ ) ≤ ⟨ j − j 0 ⟩ − 2 ( 1 + λ ) p g_{s,j-j_0}(\lambda)\le\langle j-j_0\rangle^{-2}(1+\lambda)^p g s , j − j 0 ( λ ) ≤ ⟨ j − j 0 ⟩ − 2 ( 1 + λ ) p and Lemma 5.1 , followed by orthogonal summation over particle sectors, give the form inequalities
A s , j 0 ≤ ∑ j = 0 L − 1 ⟨ j − j 0 ⟩ − 2 d Γ ( χ j ( 1 + T j ) p χ j ) ≤ C ∑ j = 0 L − 1 ⟨ j − j 0 ⟩ − 2 ( E j p + N B j w p ) . (114) A_{s,j_0}\le\sum_{j=0}^{L-1}\langle j-j_0\rangle^{-2}\mathrm{d}\Gamma\left(\chi_j(1+T_j)^p\chi_j\right)\le C\sum_{j=0}^{L-1}\langle j-j_0\rangle^{-2}\left(\mathcal{E}_j^p+\mathcal{N}_{B_j^w}^p\right). \tag*{(114)} A s , j 0 ≤ j = 0 ∑ L − 1 ⟨ j − j 0 ⟩ − 2 d Γ ( χ j ( 1 + T j ) p χ j ) ≤ C j = 0 ∑ L − 1 ⟨ j − j 0 ⟩ − 2 ( E j p + N B j w p ) . (114)
Taking traces therefore gives
Tr ρ A s , j 0 ≤ C ∑ j = 0 L − 1 ⟨ j − j 0 ⟩ − 2 Tr ρ ( E j p + N B j w p ) ≤ C s max 0 ≤ j < L Tr ρ ( E j p + N B j w p ) . (115) \begin{aligned}
\operatorname{Tr}\rho A_{s,j_0}\le C\sum_{j=0}^{L-1}\langle j-j_0\rangle^{-2}\operatorname{Tr}\rho\left(\mathcal{E}_j^p+\mathcal{N}_{B_j^w}^p\right) \\
&\le C_s\max_{0\le j<L}\operatorname{Tr}\rho\left(\mathcal{E}_j^p+\mathcal{N}_{B_j^w}^p\right). \tag*{(115)}
\end{aligned} Tr ρ A s , j 0 ≤ C j = 0 ∑ L − 1 ⟨ j − j 0 ⟩ − 2 Tr ρ ( E j p + N B j w p ) ≤ C s 0 ≤ j < L max Tr ρ ( E j p + N B j w p ) . (115)
The last step uses unit spacing of the centers:
∑ j = 0 L − 1 ⟨ j − j 0 ⟩ − 2 ≤ 1 + 2 ∑ n = 1 ∞ ( 1 + n 2 ) − 1 < ∞ . (116) \sum_{j=0}^{L-1}\langle j-j_0\rangle^{-2}\le1+2\sum_{n=1}^{\infty}(1+n^2)^{-1}<\infty. \tag*{(116)} j = 0 ∑ L − 1 ⟨ j − j 0 ⟩ − 2 ≤ 1 + 2 n = 1 ∑ ∞ ( 1 + n 2 ) − 1 < ∞. (116)
This bound is uniform in L , j 0 L,j_0 L , j 0 . The form inequalities include the required domain inclusion. Taking the maximum over j 0 j_0 j 0 proves (45) .
Proof of the initial derivative estimate
For the derivative estimate, we localize to the periodic interval, establish a second-order energy bound, and obtain higher-order Sobolev bounds by induction with constants independent of volume and particle number. The strong commutation used to reduce to a fixed number of coordinates in B j W B_j^W B j W is proved in Lemma 5.2 below.
Proof of Lemma 5.1 . Fix a , N ∈ N a,N\in\mathbb{N} a , N ∈ N and take z = j z=j z = j . The projections in each product in (137) reduce E j \mathcal{E}_j E j . Work in one such subspace with n ≥ 1 n\geq1 n ≥ 1 coordinates in B j W B_j^W B j W , relabelled 1 , … , n 1,\ldots,n 1 , … , n , on L 2 ( ( B j W ) n × ( ( B j W ) c ) N − n ) L^2((B_j^W)^n\times((B_j^W)^c)^{N-n}) L 2 (( B j W ) n × (( B j W ) c ) N − n ) . Coordinates outside B j W B_j^W B j W contribute neither kinetic nor potential terms. Indeed, η j ( y ′ ) = 0 \eta_j(y')=0 η j ( y ′ ) = 0 for y ′ ∉ B j W y'\notin B_j^W y ′ ∈ / B j W , and finite range gives
[ η j ( y ) + η j ( y ′ ) ] W ( y − y ′ ) = 0 ( y ∈ B j W , y ′ ∉ B j W ) . (117) [\eta_j(y)+\eta_j(y')]W(y-y')=0\qquad(y\in B_j^W,\ y'\notin B_j^W). \tag*{(117)} [ η j ( y ) + η j ( y ′ )] W ( y − y ′ ) = 0 ( y ∈ B j W , y ′ ∈ / B j W ) . (117)
Let K j , i ≥ 0 K_{j,i}\geq0 K j , i ≥ 0 be the operator associated with the closed kinetic form in coordinate i i i , and put
K j = ∑ i = 1 n K j , i , K j , i = − ∂ i η j ( x i ) ∂ i , V j = 1 2 ∑ 1 ≤ i < l ≤ n [ η j ( x i ) + η j ( x l ) ] W ( x i − x l ) . (118) K_j=\sum_{i=1}^{n}K_{j,i},\qquad K_{j,i}=-\partial_i\eta_j(x_i)\partial_i,\qquad V_j=\frac{1}{2}\sum_{1\leq i<l\leq n}[\eta_j(x_i)+\eta_j(x_l)]W(x_i-x_l). \tag*{(118)} K j = i = 1 ∑ n K j , i , K j , i = − ∂ i η j ( x i ) ∂ i , V j = 2 1 1 ≤ i < l ≤ n ∑ [ η j ( x i ) + η j ( x l )] W ( x i − x l ) . (118)
The potential is a bounded nonnegative multiplication operator on this fixed sector, with ∥ V j ∥ ≤ C n 2 \lVert V_j\rVert\leq Cn^2 ∥ V j ∥ ≤ C n 2 . Hence E j = K j + V j \mathcal{E}_j=K_j+V_j E j = K j + V j and D ( E j ) = D ( K j ) D(\mathcal{E}_j)=D(K_j) D ( E j ) = D ( K j ) . For each pair 1 ≤ i < ℓ ≤ n 1\leq i<\ell\leq n 1 ≤ i < ℓ ≤ n , the summand 1 2 [ η j ( x i ) + η j ( x ℓ ) ] W ( x i − x ℓ ) \frac{1}{2}[\eta_j(x_i)+\eta_j(x_\ell)]W(x_i-x_\ell) 2 1 [ η j ( x i ) + η j ( x ℓ )] W ( x i − x ℓ ) in V j V_j V j , together with its partial derivatives in x i x_i x i and x ℓ x_\ell x ℓ up to any fixed order, is bounded uniformly in n , N , L , j n,N,L,j n , N , L , j .
Step 1: Localization to the periodic interval. Let w ∈ D ( E j a ) w\in D(\mathcal{E}_j^a) w ∈ D ( E j a ) . The quadratic form associated with χ j ( x i ) ( 1 + T j ) i 2 a χ j ( x i ) \chi_j(x_i)(1+T_j)^{2a}_i\chi_j(x_i) χ j ( x i ) ( 1 + T j ) i 2 a χ j ( x i ) , evaluated at w w w , is the squared Sobolev norm of χ j ( x i ) w \chi_j(x_i)w χ j ( x i ) w , given by
∥ ( 1 + T j ) a ( χ j ( x i ) w ) ∥ 2 = ∥ χ j ( x i ) w ∥ H p e r , x i 2 a 2 . (119) \lVert(1+T_j)^a(\chi_j(x_i)w)\rVert^2=\lVert\chi_j(x_i)w\rVert^2_{H^{2a}_{\mathrm{per},x_i}}. \tag*{(119)} ∥( 1 + T j ) a ( χ j ( x i ) w ) ∥ 2 = ∥ χ j ( x i ) w ∥ H per , x i 2 a 2 . (119)
We establish the required Sobolev domain membership and bound below. Sobolev norms differentiate only coordinate i i i , viewing the function as taking values in the L 2 L^2 L 2 space of the remaining coordinates. It therefore suffices to prove
∑ i = 1 n ∥ χ j ( x i ) w ∥ H p e r , x i 2 a 2 ≤ C a ∥ ( E j + n ) a w ∥ 2 , w ∈ D ( E j a ) . (120) \sum_{i=1}^{n}\lVert\chi_j(x_i)w\rVert^2_{H^{2a}_{\mathrm{per},x_i}}\leq C_a\lVert(\mathcal{E}_j+n)^aw\rVert^2,\qquad w\in D(\mathcal{E}_j^a). \tag*{(120)} i = 1 ∑ n ∥ χ j ( x i ) w ∥ H per , x i 2 a 2 ≤ C a ∥( E j + n ) a w ∥ 2 , w ∈ D ( E j a ) . (120)
Step 2 proves the second-derivative bound, and Step 3 iterates it up to order 2 a 2a 2 a . The support of χ j \chi_j χ j and all its derivatives lies where η j = 1 \eta_j=1 η j = 1 .
Step 2: Induction basis. For w ∈ D ( E j ) = D ( K j ) w\in D(\mathcal{E}_j)=D(K_j) w ∈ D ( E j ) = D ( K j ) , expansion gives
∥ E j w ∥ 2 = ∥ K j w ∥ 2 + ∥ V j w ∥ 2 + 2 Re ⟨ K j w , V j w ⟩ . (121) \lVert\mathcal{E}_jw\rVert^2=\lVert K_jw\rVert^2+\lVert V_jw\rVert^2+2\operatorname{Re}\langle K_jw,V_jw\rangle. \tag*{(121)} ∥ E j w ∥ 2 = ∥ K j w ∥ 2 + ∥ V j w ∥ 2 + 2 Re ⟨ K j w , V j w ⟩ . (121)
Multiplication by V j V_j V j preserves the kinetic form domain, so we can evaluate the mixed term using the kinetic form. Since η j ( x i ) w \sqrt{\eta_j(x_i)}w η j ( x i ) w has an L 2 L^2 L 2 weak derivative in x i x_i x i , we apply the product rule to η j ( x i ) ∣ w ∣ 2 \eta_j(x_i)|w|^2 η j ( x i ) ∣ w ∣ 2 and integrate by parts to obtain
2 Re ⟨ K j w , V j w ⟩ = 2 ∑ i = 1 n ∫ η j ( x i ) V j ∣ ∂ i w ∣ 2 d x − ∫ ∑ i = 1 n ∂ i ( η j ( x i ) ∂ i V j ) ∣ w ∣ 2 d x . (122) 2\operatorname{Re}\langle K_jw,V_jw\rangle
=2\sum_{i=1}^{n}\int\eta_j(x_i)V_j|\partial_iw|^2\,\mathrm{d}x
-\int\sum_{i=1}^{n}\partial_i\bigl(\eta_j(x_i)\partial_iV_j\bigr)|w|^2\,\mathrm{d}x. \tag*{(122)} 2 Re ⟨ K j w , V j w ⟩ = 2 i = 1 ∑ n ∫ η j ( x i ) V j ∣ ∂ i w ∣ 2 d x − ∫ i = 1 ∑ n ∂ i ( η j ( x i ) ∂ i V j ) ∣ w ∣ 2 d x . (122)
There is no boundary term because η j = 0 \eta_j=0 η j = 0 near ∂ B j W \partial B_j^W ∂ B j W . Since V j ≥ 0 V_j\geq0 V j ≥ 0 , the first integral is nonnegative. Each pair term in V j V_j V j contributes derivatives in only its two coordinates, and hence
∥ ∑ i = 1 n ∂ i ( η j ( x i ) ∂ i V j ) ∥ ∞ ≤ C n 2 . (123) \left\lVert\sum_{i=1}^{n}\partial_i\bigl(\eta_j(x_i)\partial_iV_j\bigr)\right\rVert_{\infty}\leq Cn^2. \tag*{(123)} i = 1 ∑ n ∂ i ( η j ( x i ) ∂ i V j ) ∞ ≤ C n 2 . (123)
Since the nonnegative operators K j , i K_{j,i} K j , i strongly commute, joint spectral calculus gives the first inequality below. The second follows by substituting (122) into (121) and applying (123) .
∑ i = 1 n ∥ K j , i w ∥ 2 ≤ ∥ K j w ∥ 2 ≤ ∥ E j w ∥ 2 + C n 2 ∥ w ∥ 2 . (124) \sum_{i=1}^{n}\lVert K_{j,i}w\rVert^{2}\leq\lVert K_jw\rVert^{2}\leq\lVert\mathcal{E}_jw\rVert^{2}+Cn^{2}\lVert w\rVert^{2}. \tag*{(124)} i = 1 ∑ n ∥ K j , i w ∥ 2 ≤ ∥ K j w ∥ 2 ≤ ∥ E j w ∥ 2 + C n 2 ∥ w ∥ 2 . (124)
On a neighborhood of the support of every derivative of χ j \chi_j χ j , we have η j = 1 \eta_j=1 η j = 1 . The kinetic form controls ∂ i w \partial_i w ∂ i w there, and the product rule gives
− ∂ i 2 ( χ j ( x i ) w ) = χ j ( x i ) K j , i w − 2 χ j ′ ( x i ) ∂ i w − χ j ′ ′ ( x i ) w . (125) -\partial_i^2(\chi_j(x_i)w)=\chi_j(x_i)K_{j,i}w-2\chi_j'(x_i)\partial_iw-\chi_j''(x_i)w. \tag*{(125)} − ∂ i 2 ( χ j ( x i ) w ) = χ j ( x i ) K j , i w − 2 χ j ′ ( x i ) ∂ i w − χ j ′′ ( x i ) w . (125)
Since η j = 1 \eta_j=1 η j = 1 on supp χ j ′ \operatorname{supp}\chi_j' supp χ j ′ , the kinetic form gives
∑ i = 1 n ∥ χ j ′ ( x i ) ∂ i w ∥ 2 ≤ ∥ χ j ′ ∥ ∞ 2 ∑ i = 1 n ∫ η j ( x i ) ∣ ∂ i w ∣ 2 d x = ∥ χ j ′ ∥ ∞ 2 ⟨ w , K j w ⟩ . (126) \sum_{i=1}^{n}\lVert\chi_j'(x_i)\partial_iw\rVert^{2}\leq\lVert\chi_j'\rVert_{\infty}^{2}\sum_{i=1}^{n}\int\eta_j(x_i)|\partial_iw|^{2}\,dx=\lVert\chi_j'\rVert_{\infty}^{2}\langle w,K_jw\rangle. \tag*{(126)} i = 1 ∑ n ∥ χ j ′ ( x i ) ∂ i w ∥ 2 ≤ ∥ χ j ′ ∥ ∞ 2 i = 1 ∑ n ∫ η j ( x i ) ∣ ∂ i w ∣ 2 d x = ∥ χ j ′ ∥ ∞ 2 ⟨ w , K j w ⟩ . (126)
Thus the right-hand side of (125) belongs to L 2 L^2 L 2 . Since χ j ( x i ) w \chi_j(x_i)w χ j ( x i ) w is supported strictly inside I j I_j I j , it follows that χ j ( x i ) w ∈ H p e r , x i 2 \chi_j(x_i)w\in H_{\mathrm{per},x_i}^{2} χ j ( x i ) w ∈ H per , x i 2 . Using (125) and (126) , we obtain
∑ i = 1 n ∥ χ j ( x i ) w ∥ H p e r , x i 2 2 ≤ C ( ∑ i = 1 n ∥ K j , i w ∥ 2 + ⟨ w , K j w ⟩ + n ∥ w ∥ 2 ) ≤ C ∥ ( E j + n ) w ∥ 2 . (127) \begin{aligned}
\sum_{i=1}^{n}\lVert\chi_j(x_i)w\rVert_{H_{\mathrm{per},x_i}^{2}}^{2}\leq C\left(\sum_{i=1}^{n}\lVert K_{j,i}w\rVert^{2}+\langle w,K_jw\rangle+n\lVert w\rVert^{2}\right) \\
&\leq C\lVert(\mathcal{E}_j+n)w\rVert^{2}. \tag*{(127)}
\end{aligned} i = 1 ∑ n ∥ χ j ( x i ) w ∥ H per , x i 2 2 ≤ C ( i = 1 ∑ n ∥ K j , i w ∥ 2 + ⟨ w , K j w ⟩ + n ∥ w ∥ 2 ) ≤ C ∥( E j + n ) w ∥ 2 . (127)
The second inequality follows from (124) , the form inequality 0 ≤ K j ≤ E j 0\leq K_j\leq\mathcal{E}_j 0 ≤ K j ≤ E j , and n ≥ 1 n\geq1 n ≥ 1 . This proves (120) for a = 1 a=1 a = 1 . The same argument applies when χ j \chi_j χ j is replaced by any of its derivatives needed in Step 3, since their supports lie in the region where η j = 1 \eta_j=1 η j = 1 .
Step 3: Induction steps and summation. For the fixed coordinate y = x i y=x_i y = x i , put g i = L 2 ( ( B j W ) n − 1 × ( ( B j W ) c ) N − n ) \mathfrak{g}_i=L^2((B_j^W)^{n-1}\times((B_j^W)^c)^{N-n}) g i = L 2 (( B j W ) n − 1 × (( B j W ) c ) N − n ) and let
H j , i = ∑ k ≠ i K j , k + 1 2 ∑ k < l k , l ≠ i [ η j ( x k ) + η j ( x l ) ] W ( x k − x l ) . (128) H_{j,i}=\sum_{k\ne i}K_{j,k}+\frac{1}{2}\sum_{\substack{k<l\\ k,l\ne i}}[\eta_j(x_k)+\eta_j(x_l)]W(x_k-x_l). \tag*{(128)} H j , i = k = i ∑ K j , k + 2 1 k < l k , l = i ∑ [ η j ( x k ) + η j ( x l )] W ( x k − x l ) . (128)
This is a nonnegative selfadjoint operator on g i \mathfrak{g}_i g i , independent of y y y , and strongly commuting with K j , i K_{j,i} K j , i . The remaining interaction v j , i ( y ) v_{j,i}(y) v j , i ( y ) consists of the n − 1 n-1 n − 1 pair terms involving y = x i y=x_i y = x i . Each summand and its y y y -derivatives of any fixed order are uniformly bounded as multiplication operators on g i \mathfrak{g}_i g i :
v j , i ( y ) = 1 2 ∑ k ≠ i [ η j ( y ) + η j ( x k ) ] W ( y − x k ) , sup y ∥ ∂ y ℓ v j , i ( y ) ∥ B ( g i ) ≤ C ℓ n ( ℓ ∈ N 0 ) . (129) v_{j,i}(y)=\frac{1}{2}\sum_{k\ne i}[\eta_j(y)+\eta_j(x_k)]W(y-x_k),\qquad\sup_y\lVert\partial_y^\ell v_{j,i}(y)\rVert_{\mathcal{B}(\mathfrak{g}_i)}\leq C_\ell n\quad(\ell\in\mathbb{N}_0). \tag*{(129)} v j , i ( y ) = 2 1 k = i ∑ [ η j ( y ) + η j ( x k )] W ( y − x k ) , y sup ∥ ∂ y ℓ v j , i ( y ) ∥ B ( g i ) ≤ C ℓ n ( ℓ ∈ N 0 ) . (129)
Bounded perturbation gives
K j , i + H j , i = E j − v j , i , D ( K j , i + H j , i ) = D ( E j ) . (130) K_{j,i}+H_{j,i}=\mathcal{E}_j-v_{j,i},\qquad D(K_{j,i}+H_{j,i})=D(\mathcal{E}_j). \tag*{(130)} K j , i + H j , i = E j − v j , i , D ( K j , i + H j , i ) = D ( E j ) . (130)
The induction only requires the localized vectors χ j ( ℓ ) ( x i ) w \chi_j^{(\ell)}(x_i)w χ j ( ℓ ) ( x i ) w , 0 ≤ ℓ ≤ 2 a 0\leq\ell\leq2a 0 ≤ ℓ ≤ 2 a . Since η j = 1 \eta_j=1 η j = 1 on a neighborhood of supp χ j \operatorname{supp}\chi_j supp χ j , K j , i K_{j,i} K j , i acts as ( T j ) i = − ∂ i 2 (T_j)_i=-\partial_i^2 ( T j ) i = − ∂ i 2 on each of these vectors. We first write the product-rule identity for χ j ( x i ) w \chi_j(x_i)w χ j ( x i ) w ; the same identity holds with χ j \chi_j χ j replaced by any of these derivatives. Thus, in L 2 ( I j ; g i ) L^2(I_j;\mathfrak{g}_i) L 2 ( I j ; g i ) ,
( 1 + ( T j ) i + H j , i ) ( χ j ( x i ) w ) = χ j ( x i ) E j w + ( 1 − v j , i ) χ j ( x i ) w − 2 ∂ i ( χ j ′ ( x i ) w ) + χ j ′ ′ ( x i ) w . (131) (1+(T_j)_i+H_{j,i})(\chi_j(x_i)w)=\chi_j(x_i)\mathcal{E}_jw+(1-v_{j,i})\chi_j(x_i)w-2\partial_i(\chi_j'(x_i)w)+\chi_j''(x_i)w. \tag*{(131)} ( 1 + ( T j ) i + H j , i ) ( χ j ( x i ) w ) = χ j ( x i ) E j w + ( 1 − v j , i ) χ j ( x i ) w − 2 ∂ i ( χ j ′ ( x i ) w ) + χ j ′′ ( x i ) w . (131)
Denote the right-hand side of (131) by r r r . Taking Fourier coefficients in ξ i \xi_i ξ i gives ( 1 + λ m + H j , i ) χ j ( ξ i ) w m = r ^ m (1+\lambda_m+H_{j,i})\chi_j(\xi_i)w_m=\widehat r_m ( 1 + λ m + H j , i ) χ j ( ξ i ) w m = r m , where λ m \lambda_m λ m is the eigenvalue of T j T_j T j for the m m m -th Fourier mode. Since H j , i ≥ 0 H_{j,i}\ge0 H j , i ≥ 0 , each Fourier mode satisfies ( 1 + λ m ) ∥ χ j ( ξ i ) w m ∥ g i ≤ ∥ r ^ m ∥ g i (1+\lambda_m)\lVert\chi_j(\xi_i)w_m\rVert_{g_i}\le\lVert\widehat r_m\rVert_{g_i} ( 1 + λ m ) ∥ χ j ( ξ i ) w m ∥ g i ≤ ∥ r m ∥ g i . For s ∈ N 0 s\in\mathbb{N}_0 s ∈ N 0 , we obtain
∥ χ j ( ξ i ) w ∥ H p e r , ξ i s + 2 2 = ∑ m ( 1 + λ m ) s + 2 ∥ χ j ( ξ i ) w m ∥ g i 2 ≤ ∑ m ( 1 + λ m ) s ∥ r ^ m ∥ g i 2 = ∥ r ∥ H p e r , ξ i s 2 . (132) \lVert\chi_j(\xi_i)w\rVert_{H^{s+2}_{\mathrm{per},\xi_i}}^2=\sum_m(1+\lambda_m)^{s+2}\lVert\chi_j(\xi_i)w_m\rVert_{g_i}^2\le\sum_m(1+\lambda_m)^s\lVert\widehat r_m\rVert_{g_i}^2=\lVert r\rVert_{H^s_{\mathrm{per},\xi_i}}^2. \tag*{(132)} ∥ χ j ( ξ i ) w ∥ H per , ξ i s + 2 2 = m ∑ ( 1 + λ m ) s + 2 ∥ χ j ( ξ i ) w m ∥ g i 2 ≤ m ∑ ( 1 + λ m ) s ∥ r m ∥ g i 2 = ∥ r ∥ H per , ξ i s 2 . (132)
In particular, χ j ( ξ i ) w ∈ H p e r , ξ i s + 2 \chi_j(\xi_i)w\in H^{s+2}_{\mathrm{per},\xi_i} χ j ( ξ i ) w ∈ H per , ξ i s + 2 whenever r ∈ H p e r , ξ i s r\in H^s_{\mathrm{per},\xi_i} r ∈ H per , ξ i s . Applying the product rule and using (129) and (131) , we obtain
∥ χ j ( ξ i ) w ∥ H p e r , ξ i s + 2 ≤ C s ( ∥ χ j ( ξ i ) E j w ∥ H p e r , ξ i s + n ∥ χ j ( ξ i ) w ∥ H p e r , ξ i s + ∥ χ j ′ ( ξ i ) w ∥ H p e r , ξ i s + 1 + ∥ χ j ′ ′ ( ξ i ) w ∥ H p e r , ξ i s ) . (133) \lVert\chi_j(\xi_i)w\rVert_{H^{s+2}_{\mathrm{per},\xi_i}}\le C_s\bigl(\lVert\chi_j(\xi_i)\mathcal{E}_jw\rVert_{H^s_{\mathrm{per},\xi_i}}+n\lVert\chi_j(\xi_i)w\rVert_{H^s_{\mathrm{per},\xi_i}}+\lVert\chi'_j(\xi_i)w\rVert_{H^{s+1}_{\mathrm{per},\xi_i}}+\lVert\chi''_j(\xi_i)w\rVert_{H^s_{\mathrm{per},\xi_i}}\bigr). \tag*{(133)} ∥ χ j ( ξ i ) w ∥ H per , ξ i s + 2 ≤ C s ( ∥ χ j ( ξ i ) E j w ∥ H per , ξ i s + n ∥ χ j ( ξ i ) w ∥ H per , ξ i s + ∥ χ j ′ ( ξ i ) w ∥ H per , ξ i s + 1 + ∥ χ j ′′ ( ξ i ) w ∥ H per , ξ i s ) . (133)
C s C_s C s is independent of H j , i H_{j,i} H j , i , n n n , N N N , L L L , j j j .
The preceding estimate controls the H s + 2 H^{s+2} H s + 2 -norm of χ j w \chi_jw χ j w by the H s H^s H s -norms of χ j E j w \chi_j\mathcal{E}_jw χ j E j w and n χ j w n\chi_jw n χ j w , together with lower-order Sobolev norms involving derivatives of χ j \chi_j χ j . We prove (120) by induction on the power b b b , including the estimates for the derivatives of χ j \chi_j χ j needed at each order. Each step uses lower-order estimates involving at most three additional derivatives of χ j \chi_j χ j . Thus, for fixed a a a , only finitely many derivatives are needed, and the corresponding constants can be absorbed into C a C_a C a . Step 2 gives the base case b = 1 b=1 b = 1 .
Suppose (120) holds at order b − 1 b-1 b − 1 . Let w ∈ D ( E j b ) w\in D(\mathcal{E}_j^b) w ∈ D ( E j b ) , where 2 ≤ b ≤ a 2\le b\le a 2 ≤ b ≤ a . Then w , E j w ∈ D ( E j b − 1 ) w,\mathcal{E}_jw\in D(\mathcal{E}_j^{b-1}) w , E j w ∈ D ( E j b − 1 ) . Applying (133) with s = 2 b − 3 s=2b-3 s = 2 b − 3 and cutoff χ j ′ \chi'_j χ j ′ , then using the embedding H p e r , ξ i 2 b − 2 ⊂ H p e r , ξ i 2 b − 3 H^{2b-2}_{\mathrm{per},\xi_i}\subset H^{2b-3}_{\mathrm{per},\xi_i} H per , ξ i 2 b − 2 ⊂ H per , ξ i 2 b − 3 and the induction hypothesis at order b − 1 b-1 b − 1 , gives
∑ i = 1 n ∥ χ j ′ ( ξ i ) w ∥ H p e r , ξ i 2 b − 1 2 ≤ C a ∑ i = 1 n ( ∥ χ j ′ ( ξ i ) E j w ∥ H p e r , ξ i 2 b − 3 2 + n 2 ∥ χ j ′ ( ξ i ) w ∥ H p e r , ξ i 2 b − 3 2 + ∥ χ j ′ ′ ( ξ i ) w ∥ H p e r , ξ i 2 b − 2 2 + ∥ χ j ′ ′ ′ ( ξ i ) w ∥ H p e r , ξ i 2 b − 3 2 ) ≤ C a ( ∥ ( E j + n ) b − 1 E j w ∥ 2 + n 2 ∥ ( E j + n ) b − 1 w ∥ 2 ) . (134) \begin{aligned}
\sum_{i=1}^{n}\lVert\chi'_j(\xi_i)w\rVert_{H^{2b-1}_{\mathrm{per},\xi_i}}^2\le C_a\sum_{i=1}^{n}\bigl(\lVert\chi'_j(\xi_i)\mathcal{E}_jw\rVert_{H^{2b-3}_{\mathrm{per},\xi_i}}^2+n^2\lVert\chi'_j(\xi_i)w\rVert_{H^{2b-3}_{\mathrm{per},\xi_i}}^2+\lVert\chi''_j(\xi_i)w\rVert_{H^{2b-2}_{\mathrm{per},\xi_i}}^2+\lVert\chi'''_j(\xi_i)w\rVert_{H^{2b-3}_{\mathrm{per},\xi_i}}^2\bigr) \\
&\le C_a\bigl(\lVert(\mathcal{E}_j+n)^{b-1}\mathcal{E}_jw\rVert^2+n^2\lVert(\mathcal{E}_j+n)^{b-1}w\rVert^2\bigr). \tag*{(134)}
\end{aligned} i = 1 ∑ n ∥ χ j ′ ( ξ i ) w ∥ H per , ξ i 2 b − 1 2 ≤ C a i = 1 ∑ n ( ∥ χ j ′ ( ξ i ) E j w ∥ H per , ξ i 2 b − 3 2 + n 2 ∥ χ j ′ ( ξ i ) w ∥ H per , ξ i 2 b − 3 2 + ∥ χ j ′′ ( ξ i ) w ∥ H per , ξ i 2 b − 2 2 + ∥ χ j ′′′ ( ξ i ) w ∥ H per , ξ i 2 b − 3 2 ) ≤ C a ( ∥( E j + n ) b − 1 E j w ∥ 2 + n 2 ∥( E j + n ) b − 1 w ∥ 2 ) . (134)
Taking s = 2 b − 2 s=2b-2 s = 2 b − 2 in (133) and summing the squared norms over i i i , we obtain the following estimate using (134) and the induction hypothesis:
∑ i = 1 n ∥ χ j ( ξ i ) w ∥ H p e r , ξ i 2 b 2 ≤ C a ( ∥ ( E j + n ) b − 1 E j w ∥ 2 + n 2 ∥ ( E j + n ) b − 1 w ∥ 2 ) ≤ C a ∥ ( E j + n ) b w ∥ 2 . (135) \begin{aligned}
\sum_{i=1}^{n}\lVert\chi_j(\xi_i)w\rVert_{H^{2b}_{\mathrm{per},\xi_i}}^2\le C_a\bigl(\lVert(\mathcal{E}_j+n)^{b-1}\mathcal{E}_jw\rVert^2+n^2\lVert(\mathcal{E}_j+n)^{b-1}w\rVert^2\bigr) \\
&\le C_a\lVert(\mathcal{E}_j+n)^bw\rVert^2. \tag*{(135)}
\end{aligned} i = 1 ∑ n ∥ χ j ( ξ i ) w ∥ H per , ξ i 2 b 2 ≤ C a ( ∥( E j + n ) b − 1 E j w ∥ 2 + n 2 ∥( E j + n ) b − 1 w ∥ 2 ) ≤ C a ∥( E j + n ) b w ∥ 2 . (135)
The last step follows from spectral calculus, using t 2 + n 2 ≤ ( t + n ) 2 t^2+n^2\le(t+n)^2 t 2 + n 2 ≤ ( t + n ) 2 for t ≥ 0 t\ge0 t ≥ 0 . Taking b = a b=a b = a proves (120) on D ( E j a ) D(\mathcal{E}_j^a) D ( E j a ) .
Combining (120) with (119) gives
∑ i = 1 n ∥ ( 1 + T j ) a ( χ j ( ξ i ) w ) ∥ 2 = ∑ i = 1 n ∥ χ j ( ξ i ) w ∥ H p e r , ξ i 2 a 2 ≤ C a ∥ ( E j + n ) a w ∥ 2 ≤ C a ( ∥ E j a w ∥ 2 + n 2 a ∥ w ∥ 2 ) . (136) \begin{aligned}
\sum_{i=1}^{n}\lVert(1+T_j)^a(\chi_j(\xi_i)w)\rVert^2=\sum_{i=1}^{n}\lVert\chi_j(\xi_i)w\rVert_{H^{2a}_{\mathrm{per},\xi_i}}^2 \\
&\le C_a\lVert(\mathcal{E}_j+n)^aw\rVert^2\le C_a\bigl(\lVert\mathcal{E}_j^aw\rVert^2+n^{2a}\lVert w\rVert^2\bigr). \tag*{(136)}
\end{aligned} i = 1 ∑ n ∥( 1 + T j ) a ( χ j ( ξ i ) w ) ∥ 2 = i = 1 ∑ n ∥ χ j ( ξ i ) w ∥ H per , ξ i 2 a 2 ≤ C a ∥( E j + n ) a w ∥ 2 ≤ C a ( ∥ E j a w ∥ 2 + n 2 a ∥ w ∥ 2 ) . (136)
The last inequality uses ( t + n ) 2 a ≤ 2 2 a − 1 ( t 2 a + n 2 a ) (t+n)^{2a}\le2^{2a-1}(t^{2a}+n^{2a}) ( t + n ) 2 a ≤ 2 2 a − 1 ( t 2 a + n 2 a ) .
On L 2 ( T L N ) L^2(\mathbb{T}_L^N) L 2 ( T L N ) , define the coordinate projections, for 1 ≤ i ≤ N 1\le i\le N 1 ≤ i ≤ N and 0 ≤ n ≤ N 0\le n\le N 0 ≤ n ≤ N ,
P i = 1 B j W ( x i ) , Π n = ∑ S ⊂ { 1 , … , N } ∣ S ∣ = n ∏ i ∈ S P i ∏ k ∉ S ( 1 − P k ) . (137) P_i=\mathbf{1}_{B_j^W}(x_i),\qquad\Pi_n=\sum_{\substack{S\subset\{1,\ldots,N\}\\ |S|=n}}\prod_{i\in S}P_i\prod_{k\notin S}(1-P_k). \tag*{(137)} P i = 1 B j W ( x i ) , Π n = S ⊂ { 1 , … , N } ∣ S ∣ = n ∑ i ∈ S ∏ P i k ∈ / S ∏ ( 1 − P k ) . (137)
Since N B j W = ∑ i = 1 N P i \mathcal{N}_{B_j^W}=\sum_{i=1}^{N}P_i N B j W = ∑ i = 1 N P i , the Π n \Pi_n Π n are its spectral projections. Each Π n \Pi_n Π n is permutation invariant and therefore preserves H N \mathcal{H}_N H N .
Return to u ∈ D ( E j a ) ∩ H N u\in D(\mathcal{E}_j^a)\cap\mathcal{H}_N u ∈ D ( E j a ) ∩ H N and decompose u = ∑ n = 0 N Π n u u=\sum_{n=0}^{N}\Pi_nu u = ∑ n = 0 N Π n u . On Π 0 u \Pi_0u Π 0 u , the local energy and all localized vectors vanish. For n ≥ 1 n\geq1 n ≥ 1 , apply (136) to each product component of Π n u \Pi_nu Π n u in (137) . For fixed i i i , localization annihilates the components with i ∉ S i\notin S i ∈ / S , while the remaining components stay orthogonal in the other coordinates. Summing squared norms gives
∑ i = 1 N ∥ ( 1 + T j ) a ( χ j ( x i ) u ) ∥ 2 ≤ C a ∑ n = 0 N ( ∥ E j a Π n u ∥ 2 + n 2 a ∥ Π n u ∥ 2 ) = C a ( ∥ E j a u ∥ 2 + ∥ N B j W a u ∥ 2 ) . (138) \sum_{i=1}^{N}\left\|(1+T_j)^a(\chi_j(x_i)u)\right\|^2
\leq C_a\sum_{n=0}^{N}\left(\left\|\mathcal{E}_j^a\Pi_nu\right\|^2+n^{2a}\left\|\Pi_nu\right\|^2\right)
= C_a\left(\left\|\mathcal{E}_j^au\right\|^2+\left\|\mathcal{N}_{B_j^W}^au\right\|^2\right). \tag*{(138)} i = 1 ∑ N ∥ ( 1 + T j ) a ( χ j ( x i ) u ) ∥ 2 ≤ C a n = 0 ∑ N ( E j a Π n u 2 + n 2 a ∥ Π n u ∥ 2 ) = C a ( E j a u 2 + N B j W a u 2 ) . (138)
Here we use the strong commutation property proved in Lemma 5.2 below and N B j W Π n = n Π n \mathcal{N}_{B_j^W}\Pi_n=n\Pi_n N B j W Π n = n Π n . This proves (113) and its domain assertion on H N \mathcal{H}_N H N , with the same constant. Orthogonal summation over N N N , including the vacuum, proves the Fock-space statement, with right-hand form domain D ( E j a ) ∩ D ( N B j W a ) D(\mathcal{E}_j^a)\cap D(\mathcal{N}_{B_j^W}^a) D ( E j a ) ∩ D ( N B j W a ) .
Lemma 5.2 (Strong commutation). The operators E j \mathcal{E}_j E j and N B j W \mathcal{N}_{B_j^W} N B j W strongly commute, meaning that their spectral projections commute.
Proof of Lemma 5.2. Fix N ∈ N N\in\mathbb{N} N ∈ N . Realize the closure of the form (13) on L 2 ( T L N ) L^2(\mathbb{T}_L^N) L 2 ( T L N ) , before restricting to its antisymmetric subspace, and use the coordinate projections P i P_i P i introduced above.
Since supp η j ⊂ B j W \operatorname{supp}\eta_j\subset B_j^W supp η j ⊂ B j W , choose real smooth cutoffs 0 ≤ φ ε ≤ 1 0\leq\varphi_\varepsilon\leq1 0 ≤ φ ε ≤ 1 converging almost everywhere to 1 B j W \mathbf{1}_{B_j^W} 1 B j W , with derivatives supported where η j = 0 \eta_j=0 η j = 0 . The product rule gives
η j ( x k ) 1 / 2 ∂ k ( φ ε ( x i ) u ) = φ ε ( x i ) η j ( x k ) 1 / 2 ∂ k u , 1 ≤ k ≤ N . \eta_j(x_k)^{1/2}\partial_k(\varphi_\varepsilon(x_i)u)=\varphi_\varepsilon(x_i)\eta_j(x_k)^{1/2}\partial_ku,\qquad1\leq k\leq N. η j ( x k ) 1/2 ∂ k ( φ ε ( x i ) u ) = φ ε ( x i ) η j ( x k ) 1/2 ∂ k u , 1 ≤ k ≤ N .
This identity extends to the form domain by approximation. Dominated convergence then gives convergence to P i u P_iu P i u in the form norm, so P i P_i P i preserves that domain. For u ∈ D ( E j ) u\in D(\mathcal{E}_j) u ∈ D ( E j ) and every v v v in the form domain,
E j ( v , P i u ) = E j ( P i v , u ) = ⟨ P i v , E j u ⟩ = ⟨ v , P i E j u ⟩ . \mathcal{E}_j(v,P_iu)=\mathcal{E}_j(P_iv,u)=\langle P_iv,\mathcal{E}_ju\rangle=\langle v,P_i\mathcal{E}_ju\rangle. E j ( v , P i u ) = E j ( P i v , u ) = ⟨ P i v , E j u ⟩ = ⟨ v , P i E j u ⟩ .
The representation theorem therefore yields
P i D ( E j ) ⊂ D ( E j ) , E j P i u = P i E j u ( u ∈ D ( E j ) ) . (139) P_iD(\mathcal{E}_j)\subset D(\mathcal{E}_j),\qquad\mathcal{E}_jP_iu=P_i\mathcal{E}_ju\quad(u\in D(\mathcal{E}_j)). \tag*{(139)} P i D ( E j ) ⊂ D ( E j ) , E j P i u = P i E j u ( u ∈ D ( E j )) . (139)
Thus each P i P_i P i , and hence each Π n \Pi_n Π n , commutes with the spectral projections of E j \mathcal{E}_j E j . Since Π n \Pi_n Π n and E j \mathcal{E}_j E j preserve H N \mathcal{H}_N H N , their restrictions commute. The Π n \Pi_n Π n are the spectral projections of N B j W \mathcal{N}_{B_j^W} N B j W , proving strong commutation on each sector. Orthogonal direct summation proves the Fock-space assertion.
LRB in the pseudolattice picture
In this section we prove Proposition 2.8.
The auxiliary Hamiltonian H ~ \widetilde{H} H constructed below reproduces the physical evolution on Ran L \operatorname{Ran}\mathcal{L} Ran L . Restriction of its local couplings to finitely many modes makes them bounded (Lemma 6.4 ); the error is controlled by the weighted forms of Proposition 2.6. A commutator estimate for the restricted dynamics (Lemma 6.6 ) then bounds the error from removing the boundary couplings (Lemma 6.5 ). The mode and boundary errors are combined in the proof of the proposition at the end of the section.
For later convenience, we introduce the following notation for the pseudosite picture. For a set of pseudosites Z ⊂ { 0 , … , L − 1 } Z \subset\{0,\ldots,L-1\} Z ⊂ { 0 , … , L − 1 } , we write F ~ Z = ⨁ N ≥ 0 h ~ Z ∧ N \widetilde{\mathcal{F}}_Z = \bigoplus_{N\geq0} \widetilde{\mathfrak{h}}_Z^{\wedge N} F Z = ⨁ N ≥ 0 h Z ∧ N for the Fock space over the pseudosites in Z Z Z . Furthermore, we denote by P Z ∈ B ( h ~ ) P_Z \in\mathcal{B}(\widetilde{\mathfrak{h}}) P Z ∈ B ( h )
( P Z f ~ ) j = 1 Z ( j ) f ~ j (140) (P_Z\widetilde{f})_j = \mathbf{1}_Z(j)\widetilde{f}_j \tag*{(140)} ( P Z f ) j = 1 Z ( j ) f j (140)
the one-body projection onto the pseudosites in Z Z Z . Also, given an operator A Z A_Z A Z on h ~ Z \widetilde{\mathfrak{h}}_Z h Z , we will often identify it with its extension A Z ⊕ 0 A_Z \oplus0 A Z ⊕ 0 to h ~ \widetilde{\mathfrak{h}} h .
Auxiliary dynamics and local couplings
On the pseudosite Fock space F ~ \widetilde{\mathcal{F}} F , we introduce the auxiliary Hamiltonian H ~ = ⨁ N ≥ 0 H ~ N \widetilde{H} = \bigoplus_{N\geq0} \widetilde{H}_N H = ⨁ N ≥ 0 H N , defined by
H ~ N = ∑ i = 1 N ( T + C ) ( i ) + V ~ N , (141) \widetilde{H}_N = \sum_{i=1}^{N}(T+C)^{(i)}+\widetilde{V}_N, \tag*{(141)} H N = i = 1 ∑ N ( T + C ) ( i ) + V N , (141)
where T = ⨁ j = 0 L − 1 T j T = \bigoplus_{j=0}^{L-1}T_j T = ⨁ j = 0 L − 1 T j , and the auxiliary kinetic operator C : D ( T ) → h ~ C:D(T)\to\widetilde{\mathfrak{h}} C : D ( T ) → h is constructed from (76) , namely
( C f ) j = S f j + ∑ k B j k f k = S f j + ∑ k ( 2 a j k ∂ + a j k ′ ) f k . (142) (Cf)_j = Sf_j+\sum_k B_{jk}f_k = Sf_j+\sum_k(2a_{jk}\partial+a'_{jk})f_k. \tag*{(142)} ( C f ) j = S f j + k ∑ B j k f k = S f j + k ∑ ( 2 a j k ∂ + a j k ′ ) f k . (142)
and where the two-body interaction V ~ N \widetilde{V}_N V N is given by the following bounded operator:
( V ~ N Φ ) j ( x ) : = ∑ i < m W j i m ( x i , x m ) Φ j ( x ) W j k ( x , y ) : = η j ( x ) η k ( y ) W ( x − y ) . (143) (\widetilde{V}_N\Phi)_j(x):=\sum_{i<m}W_{j_im}(x_i,x_m)\Phi_j(x)
\qquad
W_{jk}(x,y):=\sqrt{\eta_j(x)\eta_k(y)}W(x-y). \tag*{(143)} ( V N Φ ) j ( x ) := i < m ∑ W j i m ( x i , x m ) Φ j ( x ) W j k ( x , y ) := η j ( x ) η k ( y ) W ( x − y ) . (143)
Remark 6.1. Notice that the single-pseudosite kinetic energy T j T_j T j can be interpreted as an (unbounded) on-pseudosite term acting on h ~ { j } ≡ L 2 ( I j ) \widetilde{\mathfrak{h}}_{\{j\}}\equiv L^2(I_j) h { j } ≡ L 2 ( I j ) ; on the other hand C C C (respectively, V ~ N \widetilde{V}_N V N ) can be seen as a sum of nearest-neighbor terms, since B j k = 0 B_{jk}=0 B j k = 0 if ∥ j − k ∥ ≥ 2 \lVert j-k\rVert\geq2 ∥ j − k ∥ ≥ 2 (resp. finite-range two-pseudosite terms, since W j k W_{jk} W j k vanishes if ∥ j − k ∥ ≥ 4 + r 0 \lVert j-k\rVert\geq4+r_0 ∥ j − k ∥ ≥ 4 + r 0 ).
Lemma 6.2. The auxiliary kinetic operator C C C defined in (142) is infinitesimally T T T -bounded. In particular, T + C T+C T + C is self-adjoint on D ( T + C ) = D ( T ) D(T+C)=D(T) D ( T + C ) = D ( T ) .
We defer the proof to Appendix § 7 . As a standard consequence of Lemma 6.2 we then have that H ~ N \widetilde{H}_N H N is self-adjoint on D ( ∑ i = 1 N T ( i ) ) ∩ h ~ ∧ N D(\sum_{i=1}^{N}T^{(i)})\cap\widetilde{\mathfrak{h}}^{\wedge N} D ( ∑ i = 1 N T ( i ) ) ∩ h ∧ N , therefore making H ~ \widetilde{H} H self-adjoint on its natural domain.
We now write the decomposition described in Lemma 6.1 explicitly. We set κ = ⌈ 2 ( r 0 + 5 ) ⌉ \kappa=\lceil2(r_0+5)\rceil κ = ⌈ 2 ( r 0 + 5 )⌉ , and we define the family of pseudosite sets
Z = { { j } : 0 ≤ j < L } ∪ { { j , l } : 1 ≤ ∥ j − l ∥ ≤ κ } . (144) \mathcal{Z}=\bigl\{\{j\}:0\leq j<L\bigr\}\cup\bigl\{\{j,l\}:1\leq\lVert j-l\rVert\leq\kappa\bigr\}. \tag*{(144)} Z = { { j } : 0 ≤ j < L } ∪ { { j , l } : 1 ≤ ∥ j − l ∥ ≤ κ } . (144)
For Z ∈ Z Z\in\mathcal{Z} Z ∈ Z , we define the following first-order operator on D ( T Z ) ⊂ h ~ Z D(T_Z)\subset\widetilde{\mathfrak{h}}_Z D ( T Z ) ⊂ h Z
C Z : = ∑ j , l : { j , l } = Z P { j } C P { l } , (145) C_Z:=\sum_{j,l:\{j,l\}=Z}P_{\{j\}}CP_{\{l\}}, \tag*{(145)} C Z := j , l : { j , l } = Z ∑ P { j } C P { l } , (145)
which then vanishes for pairs Z = { j , l } Z=\{j,l\} Z = { j , l } with ∥ j − l ∥ ≥ 2 \lVert j-l\rVert\geq2 ∥ j − l ∥ ≥ 2 . Analogously, define the bounded two-particle operator W Z ∈ B ( h ~ Z ∧ 2 ) W_Z\in\mathcal{B}(\widetilde{\mathfrak{h}}_Z^{\wedge2}) W Z ∈ B ( h Z ∧ 2 ) by
( W Z Φ ) j j ′ ( y , y ′ ) : = 1 { Z = { j , j ′ } } W j , j ′ ( y , y ′ ) Φ j j ′ ( y , y ′ ) . (146) (W_Z\Phi)_{jj'}(y,y'):=\mathbf{1}_{\{Z=\{j,j'\}\}}W_{j,j'}(y,y')\Phi_{jj'}(y,y'). \tag*{(146)} ( W Z Φ ) j j ′ ( y , y ′ ) := 1 { Z = { j , j ′ }} W j , j ′ ( y , y ′ ) Φ j j ′ ( y , y ′ ) . (146)
For N ∈ N N \in\mathbb{N} N ∈ N , we define the local interaction
Ψ Z ∣ h ^ Z N = ∑ i = 1 N ( C Z ) i + ∑ 1 ≤ i < m ≤ N ( W Z ) i m , (147) \left.\Psi_Z\right|_{\widehat{\mathfrak h}_Z^N}=\sum_{i=1}^{N}(C_Z)_i+\sum_{1\le i<m\le N}(W_Z)_{im}, \tag*{(147)} Ψ Z ∣ h Z N = i = 1 ∑ N ( C Z ) i + 1 ≤ i < m ≤ N ∑ ( W Z ) im , (147)
with Ψ Z = 0 \Psi_Z=0 Ψ Z = 0 on the vacuum sector. Then we have that
H ~ = d Γ ( T ) + ∑ Z ∈ Z Ψ Z . (148) \widetilde H=\mathrm{d}\Gamma(T)+\sum_{Z\in\mathcal{Z}}\Psi_Z. \tag*{(148)} H = d Γ ( T ) + Z ∈ Z ∑ Ψ Z . (148)
Proposition 6.3 (Auxiliary Hamiltonian). H ~ \widetilde H H is the Hamiltonian describing the dynamics in the pseudolattice picture. That is, for every t ∈ R t\in\mathbb{R} t ∈ R , we have
e − i t H = L ∗ e − i t H ~ L . (149) e^{-itH}=\mathcal{L}^{*}e^{-it\widetilde H}\mathcal{L}. \tag*{(149)} e − i t H = L ∗ e − i t H L . (149)
Proof. The direct IMS identity (76) , together with the fact that χ j \chi_j χ j is smooth and supported in the interior of I j I_j I j , yields
L 1 D ( K 1 ) ⊂ D ( T ) , K 1 u = L 1 ∗ ( T + C ) L 1 u (150) \mathcal{L}_1D(K_1)\subset D(T),\qquad K_1u=\mathcal{L}_1^*(T+C)\mathcal{L}_1u \tag*{(150)} L 1 D ( K 1 ) ⊂ D ( T ) , K 1 u = L 1 ∗ ( T + C ) L 1 u (150)
for any u ∈ C ∞ ( T L ) u\in C^\infty(\mathbb{T}_L) u ∈ C ∞ ( T L ) , and by density they extend to D ( K 1 ) = H 2 ( T L ) D(K_1)=H^2(\mathbb{T}_L) D ( K 1 ) = H 2 ( T L ) . Moreover, ( ⨁ N V N ) L = L V (\bigoplus_N V_N)\mathcal{L}=\mathcal{L}V ( ⨁ N V N ) L = L V because η j χ j = χ j \sqrt{\eta_j}\chi_j=\chi_j η j χ j = χ j , therefore L 1 ⊗ N \mathcal{L}^{\otimes N}_1 L 1 ⊗ N takes D ( H N ) D(H_N) D ( H N ) into D ( H ~ N ) D(\widetilde H_N) D ( H N ) , and by (150) we have L H N = H ~ N L \mathcal{L}H_N=\widetilde H_N\mathcal{L} L H N = H N L . Applying to ( H N − z ) − 1 u (H_N-z)^{-1}u ( H N − z ) − 1 u , with z ∈ C ∖ R z\in\mathbb{C}\setminus\mathbb{R} z ∈ C ∖ R , this gives ( H ~ N − z ) − 1 L = L ( H N − z ) − 1 (\widetilde H_N-z)^{-1}\mathcal{L}=\mathcal{L}(H_N-z)^{-1} ( H N − z ) − 1 L = L ( H N − z ) − 1 . The identities for z ‾ \overline z z and z ‾ \overline z z show that Ran L \operatorname{Ran}\mathcal{L} Ran L reduces H ~ N \widetilde H_N H N , and the spectral theorem then yields (6.10). □
Restriction to finitely many modes
For any energy E ≥ 1 E\ge1 E ≥ 1 , and for any set of pseudosites Z Z Z , we introduce the energy cut-off projector
π Z ( E ) = 1 [ 1 , E ] ( 1 + T Z ) ∈ B ( h ~ Z ) . (151) \pi_Z(E)=\mathbf{1}_{[1,E]}(1+T_Z)\in\mathcal{B}(\widetilde{\mathfrak h}_Z). \tag*{(151)} π Z ( E ) = 1 [ 1 , E ] ( 1 + T Z ) ∈ B ( h Z ) . (151)
and, on h ^ Z N \widehat{\mathfrak h}_Z^N h Z N , we then define the truncated couplings
Ψ Z ( E ) = ∑ i = 1 N π i ( C Z ) i π i + ∑ i < m N π i π m ( W Z ) i m π i π m , π i : = π Z ( E ) i . (152) \Psi_Z^{(E)}=\sum_{i=1}^{N}\pi_i(C_Z)_i\pi_i+\sum_{i<m}^{N}\pi_i\pi_m(W_Z)_{im}\pi_i\pi_m,\qquad\pi_i:=\pi_Z(E)_i. \tag*{(152)} Ψ Z ( E ) = i = 1 ∑ N π i ( C Z ) i π i + i < m ∑ N π i π m ( W Z ) im π i π m , π i := π Z ( E ) i . (152)
Lemma 6.4 (Restriction to finitely many modes). Let Z ∈ X Z\in\mathcal{X} Z ∈ X , and E ≥ 1 E\ge1 E ≥ 1 . The operator Ψ Z ( E ) \Psi_Z^{(E)} Ψ Z ( E ) is bounded, even, and self-adjoint; also, there exists C > 0 C>0 C > 0 independent of E E E such that
∥ Ψ Z ( E ) ∥ ≤ C E . (153) \|\Psi_Z^{(E)}\|\le CE. \tag*{(153)} ∥ Ψ Z ( E ) ∥ ≤ C E . (153) Furthermore, for any p , N ∈ N p,N\in\mathbb{N} p , N ∈ N , and for any vector smooth, periodic vector u u u
∥ ( Ψ Z − Ψ Z ( E ) ) u ∥ 2 ≤ C p ∥ d Γ ( ( 1 + T Z ) p + 1 ) u ∥ 2 E 2 p . (154) \|(\Psi_Z-\Psi_Z^{(E)})u\|^2\le C_p\frac{\|\mathrm{d}\Gamma((1+T_Z)^{p+1})u\|^2}{E^{2p}}. \tag*{(154)} ∥ ( Ψ Z − Ψ Z ( E ) ) u ∥ 2 ≤ C p E 2 p ∥ d Γ (( 1 + T Z ) p + 1 ) u ∥ 2 . (154) for a constant C p C_p C p independent of E E E and N N N .
Proof. Step 1: Boundedness of the truncated coupling. Denoting by d E d_E d E the rank of π Z ( E ) \pi_Z(E) π Z ( E ) , eigenvalue counting in one dimension gives
d E ≤ C E , ∥ π Z ( E ) C Z π Z ( E ) ∥ ≤ C E , ∥ π Z ( E ) ⊗ 2 W Z π Z ( E ) ⊗ 2 ∥ ≤ C . (155) d_E\le C\sqrt{E},\qquad\|\pi_Z(E)C_Z\pi_Z(E)\|\le C\sqrt{E},\qquad\|\pi_Z(E)^{\otimes2}W_Z\pi_Z(E)^{\otimes2}\|\le C. \tag*{(155)} d E ≤ C E , ∥ π Z ( E ) C Z π Z ( E ) ∥ ≤ C E , ∥ π Z ( E ) ⊗ 2 W Z π Z ( E ) ⊗ 2 ∥ ≤ C . (155)
For the middle bound, we have used that ∥ ( 1 + T Z ) 1 / 2 π Z ( E ) ∥ ≤ E \lVert(1+T_Z)^{1/2}\pi_Z(E)\rVert\le\sqrt{E} ∥( 1 + T Z ) 1/2 π Z ( E )∥ ≤ E .
On the Fock space generated by these modes, the Pauli exclusion principle implies that particle number is bounded by d E ≤ C E d_E\le C\sqrt{E} d E ≤ C E . Thus the one-body sum in (152) has norm at most d E C E ≤ C E d_EC\sqrt{E}\le CE d E C E ≤ C E , and the pair sum has norm at most ( d E 2 ) ∥ W Z ∥ ≤ C E \binom{d_E}{2}\lVert W_Z\rVert\le CE ( 2 d E ) ∥ W Z ∥ ≤ C E . This proves (153) .
Step 2: Estimates for the spectral tails. Since the operators in the claimed estimate preserve the number of particles in Z Z Z , by orthogonality it is sufficient to consider u u u with exactly n n n particles in the pseudosites contained in Z Z Z , with n ≤ N n\le N n ≤ N . Let us write G i = ( 1 + T Z ) ( i ) G_i=(1+T_Z)^{(i)} G i = ( 1 + T Z ) ( i ) , with i = 1 , … , n i=1,\ldots,n i = 1 , … , n .
By definition, Ψ Z = ∑ i ( C Z ) i + ∑ i < m ( W Z ) i m \Psi_Z=\sum_i(C_Z)_i+\sum_{i<m}(W_Z)_{im} Ψ Z = ∑ i ( C Z ) i + ∑ i < m ( W Z ) im , therefore we may estimate each term separately. First, we decompose the one-body difference as
( C Z ) i − π i ( C Z ) i π i = ( 1 − π i ) ( C Z ) i + π i ( C Z ) i ( 1 − π i ) . (156) (C_Z)_i-\pi_i(C_Z)_i\pi_i=(1-\pi_i)(C_Z)_i+\pi_i(C_Z)_i(1-\pi_i). \tag*{(156)} ( C Z ) i − π i ( C Z ) i π i = ( 1 − π i ) ( C Z ) i + π i ( C Z ) i ( 1 − π i ) . (156)
By spectral calculus, we have the Markov inequalities
∥ ( 1 − π i ) u ∥ ≤ E − p ∥ G i p u ∥ ∥ G i 1 / 2 ( 1 − π i ) u ∥ ≤ E − p ∥ G i p + 1 / 2 u ∥ , (157) \lVert(1-\pi_i)u\rVert\le E^{-p}\lVert G_i^p u\rVert\qquad\lVert G_i^{1/2}(1-\pi_i)u\rVert\le E^{-p}\lVert G_i^{p+1/2}u\rVert, \tag*{(157)} ∥( 1 − π i ) u ∥ ≤ E − p ∥ G i p u ∥ ∥ G i 1/2 ( 1 − π i ) u ∥ ≤ E − p ∥ G i p + 1/2 u ∥ , (157)
which, together with the product rule, then imply ∥ G i p ( C Z ) i u ∥ ≤ C p ∥ G i p + 1 / 2 u ∥ \lVert G_i^p(C_Z)_iu\rVert\le C_p\lVert G_i^{p+1/2}u\rVert ∥ G i p ( C Z ) i u ∥ ≤ C p ∥ G i p + 1/2 u ∥ . Hence
∥ ( ( C Z ) i − π i ( C Z ) i π i ) u ∥ ≤ E − p ∥ G i p ( C Z ) i u ∥ + C ∥ G i 1 / 2 ( 1 − π i ) u ∥ ≤ C p E − p ∥ G i p + 1 / 2 u ∥ . (158) \begin{aligned}
\lVert((C_Z)_i-\pi_i(C_Z)_i\pi_i)u\rVert
&\le E^{-p}\lVert G_i^p(C_Z)_iu\rVert+C\lVert G_i^{1/2}(1-\pi_i)u\rVert\\
&\le C_pE^{-p}\lVert G_i^{p+1/2}u\rVert. \tag*{(158)}
\end{aligned} ∥(( C Z ) i − π i ( C Z ) i π i ) u ∥ ≤ E − p ∥ G i p ( C Z ) i u ∥ + C ∥ G i 1/2 ( 1 − π i ) u ∥ ≤ C p E − p ∥ G i p + 1/2 u ∥ . (158)
For the pair term put Π = π i π m \Pi=\pi_i\pi_m Π = π i π m and use
( W Z ) i m − Π ( W Z ) i m Π = ( 1 − Π ) ( W Z ) i m + Π ( W Z ) i m ( 1 − Π ) , 1 − Π = ( 1 − π i ) + π i ( 1 − π m ) . (159) (W_Z)_{im}-\Pi(W_Z)_{im}\Pi=(1-\Pi)(W_Z)_{im}+\Pi(W_Z)_{im}(1-\Pi),\qquad1-\Pi=(1-\pi_i)+\pi_i(1-\pi_m). \tag*{(159)} ( W Z ) im − Π ( W Z ) im Π = ( 1 − Π ) ( W Z ) im + Π ( W Z ) im ( 1 − Π ) , 1 − Π = ( 1 − π i ) + π i ( 1 − π m ) . (159)
Since the multiplication operator W Z W_Z W Z is bounded and smooth in the i i i -th variable, we have ∥ G i p ( W Z ) i m u ∥ ≤ C p ∥ G i p u ∥ \lVert G_i^p(W_Z)_{im}u\rVert\le C_p\lVert G_i^p u\rVert ∥ G i p ( W Z ) im u ∥ ≤ C p ∥ G i p u ∥ , and similarly for m m m . The spectral-tail inequalities give
∥ ( ( W Z ) i m − π i π m ( W Z ) i m π i π m ) u ∥ 2 ≤ C p E − 2 p ( ∥ G i p u ∥ 2 + ∥ G m p u ∥ 2 ) . (160) \lVert((W_Z)_{im}-\pi_i\pi_m(W_Z)_{im}\pi_i\pi_m)u\rVert^2\le C_pE^{-2p}\left(\lVert G_i^p u\rVert^2+\lVert G_m^p u\rVert^2\right). \tag*{(160)} ∥(( W Z ) im − π i π m ( W Z ) im π i π m ) u ∥ 2 ≤ C p E − 2 p ( ∥ G i p u ∥ 2 + ∥ G m p u ∥ 2 ) . (160)
Step 3: Summation and exclusion principle. Applying Cauchy–Schwarz over the one-body and pair sums gives factors n n n and ( n 2 ) \binom{n}{2} ( 2 n ) , respectively. In particular, for the pair contribution,
( n 2 ) ∑ i < m ( ∥ G i p u ∥ 2 + ∥ G m p u ∥ 2 ) = ( n 2 ) ( n − 1 ) ∑ i ∥ G i p u ∥ 2 ≤ n 3 ∑ i ∥ G i p u ∥ 2 . (161) \binom{n}{2}\sum_{i<m}\left(\lVert G_i^p u\rVert^2+\lVert G_m^p u\rVert^2\right)=\binom{n}{2}(n-1)\sum_i\lVert G_i^p u\rVert^2\le n^3\sum_i\lVert G_i^p u\rVert^2. \tag*{(161)} ( 2 n ) i < m ∑ ( ∥ G i p u ∥ 2 + ∥ G m p u ∥ 2 ) = ( 2 n ) ( n − 1 ) i ∑ ∥ G i p u ∥ 2 ≤ n 3 i ∑ ∥ G i p u ∥ 2 . (161)
Consequently
∥ ( Ψ Z − Ψ Z ( E ) ) u ∥ 2 ≤ C p E − 2 p ( n ∑ i ∥ G i p + 1 / 2 u ∥ 2 + n 3 ∑ i ∥ G i p u ∥ 2 ) ≤ C p E − 2 p ⟨ u , ∑ i G i 2 p + 3 / 2 u ⟩ ≤ C p E − 2 p ⟨ u , ∑ i G i 2 p + 2 u ⟩ . (162) \begin{aligned}
\lVert(\Psi_Z-\Psi_Z^{(E)})u\rVert^2
&\le C_pE^{-2p}\left(n\sum_i\lVert G_i^{p+1/2}u\rVert^2+n^3\sum_i\lVert G_i^p u\rVert^2\right)\\
&\le C_pE^{-2p}\left\langle u,\sum_iG_i^{2p+3/2}u\right\rangle
\le C_pE^{-2p}\left\langle u,\sum_iG_i^{2p+2}u\right\rangle. \tag*{(162)}
\end{aligned} ∥( Ψ Z − Ψ Z ( E ) ) u ∥ 2 ≤ C p E − 2 p ( n i ∑ ∥ G i p + 1/2 u ∥ 2 + n 3 i ∑ ∥ G i p u ∥ 2 ) ≤ C p E − 2 p ⟨ u , i ∑ G i 2 p + 3/2 u ⟩ ≤ C p E − 2 p ⟨ u , i ∑ G i 2 p + 2 u ⟩ . (162)
where in the second passage we applied Lemma 4.2 with the following choices: C = Z \mathcal{C}=Z C = Z , with α = 1 \alpha=1 α = 1 and f ( λ ) = ( 1 + λ ) 2 p + 1 f(\lambda)=(1+\lambda)^{2p+1} f ( λ ) = ( 1 + λ ) 2 p + 1 for the one-body term, and with α = 3 \alpha=3 α = 3 and f ( λ ) = ( 1 + λ ) 2 p f(\lambda)=(1+\lambda)^{2p} f ( λ ) = ( 1 + λ ) 2 p for the pair term. The last inequality follows from G i ≥ 1 G_i\ge1 G i ≥ 1 .
Removal of boundary couplings
Given an observable localized near x ∈ T L x \in\mathbb{T}_L x ∈ T L , we want to approximate its full evolution by dynamics acting only on a pseudosite ball x [ R ] = { j ∈ { 0 , … , L − 1 } : ∥ j − x ∥ ≤ R } x[R] = \{j \in\{0,\ldots,L-1\}: \lVert j-x\rVert\le R\} x [ R ] = { j ∈ { 0 , … , L − 1 } : ∥ j − x ∥ ≤ R } . We pass from the full H ~ \widetilde{H} H to such a Hamiltonian in two stages.
First, we replace every interaction Ψ Z \Psi_Z Ψ Z meeting x [ R ] x[R] x [ R ] by its truncated version Ψ Z ( E ) \Psi_Z^{(E)} Ψ Z ( E ) , producing the intermediate Hamiltonian H ^ R , E \widehat{H}_{R,E} H R , E , which therefore differs from H ~ \widetilde{H} H in O ( R ) O(R) O ( R ) interaction terms, uniformly in L L L .
Next, removing the truncated couplings crossing the boundary we are left with the desired Hamiltonian H R , E H_{R,E} H R , E , which contains the kinetic energy and truncated interactions inside x [ R ] x[R] x [ R ] , up to an exterior Hamiltonian H o u t H_{\mathrm{out}} H out whose dynamics does not affect observables supported inside. In particular, H R , E H_{R,E} H R , E generates the approximating evolution of the observable.
We define
H ^ R , E : = d Γ ( T ) + ∑ Z ∩ x [ R ] ≠ ∅ Ψ Z ( E ) + ∑ Z ∩ x [ R ] = ∅ Ψ Z . (163) \widehat{H}_{R,E} := \mathrm{d}\Gamma(T) + \sum_{Z\cap x[R]\ne\varnothing}\Psi_Z^{(E)} + \sum_{Z\cap x[R]=\varnothing}\Psi_Z . \tag*{(163)} H R , E := d Γ ( T ) + Z ∩ x [ R ] = ∅ ∑ Ψ Z ( E ) + Z ∩ x [ R ] = ∅ ∑ Ψ Z . (163)
H R , E : = d Γ ( T x [ R ] ) + ∑ Z ⊂ x [ R ] Ψ Z ( E ) . (164) H_{R,E} := \mathrm{d}\Gamma(T_{x[R]}) + \sum_{Z\subset x[R]}\Psi_Z^{(E)} . \tag*{(164)} H R , E := d Γ ( T x [ R ] ) + Z ⊂ x [ R ] ∑ Ψ Z ( E ) . (164)
The following lemma shows that the dynamics generated by H ^ R , E \widehat{H}_{R,E} H R , E and H R , E H_{R,E} H R , E are locally close, with an error that decays exponentially in the distance of the observable from the boundary of x [ R ] x[R] x [ R ] .
Lemma 6.5 (Removal of boundary couplings). Let 1 ≤ R ≤ L / 4 1 \le R \le L/4 1 ≤ R ≤ L /4 and E ≥ 1 E \ge1 E ≥ 1 . For nonempty X ⊂ x [ R ] X \subset x[R] X ⊂ x [ R ] , u ~ ∈ h X \widetilde{u} \in\mathfrak{h}_X u ∈ h X , σ ∈ { − , + } \sigma\in\{-,+\} σ ∈ { − , + } , and r X = max j ∈ X ∥ j − x ∥ r_X = \max_{j\in X}\lVert j-x\rVert r X = max j ∈ X ∥ j − x ∥ , one has, for every t ∈ R t \in\mathbb{R} t ∈ R ,
∥ τ t H ^ R , E ( a σ ( u ~ ) ) − τ t H R , E ( a σ ( u ~ ) ) ∥ ≤ C 0 ∥ u ~ ∥ ∣ X ∣ e − c ( R − r X ) ( e c 1 E ∣ t ∣ − 1 ) . (165) \left\lVert\tau_t^{\widehat{H}_{R,E}}\left(a^\sigma(\widetilde{u})\right)-\tau_t^{H_{R,E}}\left(a^\sigma(\widetilde{u})\right)\right\rVert\le C_0\lVert\widetilde{u}\rVert|X|e^{-c(R-r_X)}\left(e^{c_1E|t|}-1\right). \tag*{(165)} τ t H R , E ( a σ ( u ) ) − τ t H R , E ( a σ ( u ) ) ≤ C 0 ∥ u ∥ ∣ X ∣ e − c ( R − r X ) ( e c 1 E ∣ t ∣ − 1 ) . (165) for constants C 0 , c , c 1 > 0 C_0,c,c_1>0 C 0 , c , c 1 > 0 dependent on κ \kappa κ and independent of E E E , R R R , L L L , and t t t .
Proof. The exterior operator
H o u t = d Γ ( T ( x [ R ] ) c ) + ∑ Z ⊂ ( x [ R ] ) c Ψ Z (166) H_{\mathrm{out}} = \mathrm{d}\Gamma\left(T_{(x[R])^c}\right) + \sum_{Z\subset(x[R])^c}\Psi_Z \tag*{(166)} H out = d Γ ( T ( x [ R ] ) c ) + Z ⊂ ( x [ R ] ) c ∑ Ψ Z (166)
Since it is supported on x [ R ] c x[R]^c x [ R ] c , it strongly commutes with both H R , E H_{R,E} H R , E and a σ ( u ~ ) a^\sigma(\widetilde{u}) a σ ( u ) . Moreover,
H ^ R , E = H R , E + H o u t + ∑ Z ∩ x [ R ] ≠ ∅ Z ∩ ( x [ R ] ) c ≠ ∅ Ψ Z ( E ) . (167) \widehat{H}_{R,E}=H_{R,E}+H_{\mathrm{out}}+\sum_{\substack{Z\cap x[R]\ne\varnothing\\ Z\cap(x[R])^c\ne\varnothing}}\Psi_Z^{(E)}. \tag*{(167)} H R , E = H R , E + H out + Z ∩ x [ R ] = ∅ Z ∩ ( x [ R ] ) c = ∅ ∑ Ψ Z ( E ) . (167)
Since R ≤ L / 4 R\le L/4 R ≤ L /4 , x [ R ] x[R] x [ R ] is an arc of consecutive pseudosites, so at most 4 κ 2 4\kappa^2 4 κ 2 pairs cross its boundary. For such a pair Z = { j , l } Z=\{j,l\} Z = { j , l } with l ∉ x [ R ] l\notin x[R] l ∈ / x [ R ] , the triangle inequality and ∥ j − l ∥ ≤ κ \lVert j-l\rVert\le\kappa ∥ j − l ∥ ≤ κ give
dist ( X , Z ) ≥ R − r X − κ . (168) \operatorname{dist}(X,Z)\ge R-r_X-\kappa. \tag*{(168)} dist ( X , Z ) ≥ R − r X − κ . (168)
The boundary sum is bounded, and
τ t H R , E + H o u t ( a σ ( u ~ ) ) = τ t H R , E ( a σ ( u ~ ) ) , (169) \tau_t^{H_{R,E}+H_{\mathrm{out}}}\left(a^\sigma(\widetilde{u})\right)=\tau_t^{H_{R,E}}\left(a^\sigma(\widetilde{u})\right), \tag*{(169)} τ t H R , E + H out ( a σ ( u ) ) = τ t H R , E ( a σ ( u ) ) , (169)
Duhamel’s formula and Lemma 6.6 therefore give, for t ≥ 0 t\ge0 t ≥ 0 ,
∥ τ t H ^ R , E ( a σ ( u ~ ) ) − τ t H R , E ( a σ ( u ~ ) ) ∥ ≤ ∑ Z ∩ x [ R ] ≠ ∅ Z ∩ ( x [ R ] ) c ≠ ∅ ∫ 0 t ∥ [ Ψ Z ( E ) , τ r H R , E ( a σ ( u ~ ) ) ] ∥ d r ≤ C ∥ u ~ ∥ ∣ X ∣ E e − ( R − r X ) / κ ∫ 0 t e c 1 E r d r ≤ C 0 ∥ u ~ ∥ ∣ X ∣ e − c ( R − r X ) ( e c 1 E t − 1 ) , (170) \begin{aligned}
\left\lVert\tau_t^{\widehat{H}_{R,E}}\left(a^\sigma(\widetilde{u})\right)-\tau_t^{H_{R,E}}\left(a^\sigma(\widetilde{u})\right)\right\rVert
&\le\sum_{\substack{Z\cap x[R]\ne\varnothing\\ Z\cap(x[R])^c\ne\varnothing}}\int_0^t \left\lVert\left[\Psi_Z^{(E)},\tau_r^{H_{R,E}}\left(a^\sigma(\widetilde{u})\right)\right]\right\rVert\,\mathrm{d}r\\
&\le C\lVert\widetilde{u}\rVert|X|Ee^{-(R-r_X)/\kappa}\int_0^t e^{c_1Er}\,\mathrm{d}r\\
&\le C_0\lVert\widetilde{u}\rVert|X|e^{-c(R-r_X)}\left(e^{c_1Et}-1\right), \tag*{(170)}
\end{aligned} τ t H R , E ( a σ ( u ) ) − τ t H R , E ( a σ ( u ) ) ≤ Z ∩ x [ R ] = ∅ Z ∩ ( x [ R ] ) c = ∅ ∑ ∫ 0 t [ Ψ Z ( E ) , τ r H R , E ( a σ ( u ) ) ] d r ≤ C ∥ u ∥ ∣ X ∣ E e − ( R − r X ) / κ ∫ 0 t e c 1 E r d r ≤ C 0 ∥ u ∥ ∣ X ∣ e − c ( R − r X ) ( e c 1 E t − 1 ) , (170)
with c = 1 / κ c=1/\kappa c = 1/ κ and C 0 = C / c 1 C_0=C/c_1 C 0 = C / c 1 . Reversing the signs of both Hamiltonians gives the bound for negative times.
For nonempty pseudosite sets X , Y X,Y X , Y , let us indicate with dist ( X , Y ) : = min j ∈ X , l ∈ Y ∥ j − l ∥ \operatorname{dist}(X,Y):=\min_{j\in X,l\in Y}\lVert j-l\rVert dist ( X , Y ) := min j ∈ X , l ∈ Y ∥ j − l ∥ their distance.
Lemma 6.6 (LRB bound). There exist constants C 0 , c 1 > 0 C_0,c_1>0 C 0 , c 1 > 0 dependent on κ \kappa κ such that the following holds. Let 1 ≤ R ≤ L / 4 1\le R\le L/4 1 ≤ R ≤ L /4 and E ≥ 1 E\ge1 E ≥ 1 . For any nonempty X ⊂ x [ R ] X\subset x[R] X ⊂ x [ R ] , u ~ ∈ h X \tilde{u}\in\mathfrak{h}_X u ~ ∈ h X , σ ∈ { − , + } \sigma\in\{-,+\} σ ∈ { − , + } , and for any Y ∈ Z Y\in\mathcal{Z} Y ∈ Z , and for every t ∈ R t\in\mathbb{R} t ∈ R , we have
∥ [ τ t H R , E ( a σ ( u ~ ) ) , Ψ Y ( E ) ] ∥ ≤ C 0 ∥ u ~ ∥ ∥ Ψ Y ( E ) ∥ ∣ X ∣ e c 1 E ∣ t ∣ − dist ( X , Y ) / κ . (171) \left\lVert[\tau_t^{H_{R,E}}(a^\sigma(\tilde{u})),\Psi_Y^{(E)}]\right\rVert\le C_0\lVert\tilde{u}\rVert\lVert\Psi_Y^{(E)}\rVert\lvert X\rvert e^{c_1E\lvert t\rvert-\operatorname{dist}(X,Y)/\kappa}. \tag*{(171)} [ τ t H R , E ( a σ ( u ~ )) , Ψ Y ( E ) ] ≤ C 0 ∥ u ~ ∥ ∥ Ψ Y ( E ) ∥ ∣ X ∣ e c 1 E ∣ t ∣ − dist ( X , Y ) / κ . (171) for constants C 0 , c 1 > 0 C_0,c_1>0 C 0 , c 1 > 0 dependent on κ \kappa κ and independent of E , R , L E,R,L E , R , L , and t t t .
Proof. Without loss of generality, we may assume2 X ∩ Y = ∅ X\cap Y=\varnothing X ∩ Y = ∅ . We may also bring the time evolution on Ψ Y ( E ) \Psi_Y^{(E)} Ψ Y ( E ) .
Step 1. Following [40 Section 3] , we devise an interaction picture to remove the unbounded on-(pseudo)site contributions to H R , E H_{R,E} H R , E . We set K = d Γ ( T x [ R ] ) K=\mathrm{d}\Gamma(T_{x[R]}) K = d Γ ( T x [ R ] ) and define, for r ∈ R r\in\mathbb{R} r ∈ R ,
Ψ Z ( r ) = e i r K Ψ Z ( E ) e − i r K , U ( r ) = e i r K e − i r H R , E . (172) \Psi_Z(r)=e^{irK}\Psi_Z^{(E)}e^{-irK},\qquad U(r)=e^{irK}e^{-irH_{R,E}}. \tag*{(172)} Ψ Z ( r ) = e i r K Ψ Z ( E ) e − i r K , U ( r ) = e i r K e − i r H R , E . (172)
Due to K K K being a sum of on-pseudosite terms, Ψ Z ( r ) \Psi_Z(r) Ψ Z ( r ) is even, supported in Z Z Z , and norm-continuous, with ∥ Ψ Z ( r ) ∥ = ∥ Ψ Z ( E ) ∥ \lVert\Psi_Z(r)\rVert=\lVert\Psi_Z^{(E)}\rVert ∥ Ψ Z ( r )∥ = ∥ Ψ Z ( E ) ∥ . Moreover,
i U ′ ( r ) = ∑ Z ⊂ x [ R ] Ψ Z ( r ) U ( r ) , τ r H R , E ( Ψ Y ( E ) ) = U ( r ) ∗ Ψ Y ( r ) U ( r ) . (173) iU'(r)=\sum_{Z\subset x[R]}\Psi_Z(r)U(r),\qquad\tau_r^{H_{R,E}}(\Psi_Y^{(E)})=U(r)^*\Psi_Y(r)U(r). \tag*{(173)} i U ′ ( r ) = Z ⊂ x [ R ] ∑ Ψ Z ( r ) U ( r ) , τ r H R , E ( Ψ Y ( E ) ) = U ( r ) ∗ Ψ Y ( r ) U ( r ) . (173)
Therefore, by Duhamel interpolation we have
[ τ t H R , E ( Ψ Y ( E ) ) , a σ ( u ~ ) ] = i ∑ Z ⊂ x [ R ] ∫ 0 t [ U ( r ) ∗ [ Ψ Z ( r ) , Ψ Y ( t ) ] U ( r ) , a σ ( u ~ ) ] d r . (174) [\tau_t^{H_{R,E}}(\Psi_Y^{(E)}),a^\sigma(\tilde{u})]=i\sum_{Z\subset x[R]}\int_0^t[U(r)^*[\Psi_Z(r),\Psi_Y(t)]U(r),a^\sigma(\tilde{u})]\,\mathrm{d}r. \tag*{(174)} [ τ t H R , E ( Ψ Y ( E ) ) , a σ ( u ~ )] = i Z ⊂ x [ R ] ∑ ∫ 0 t [ U ( r ) ∗ [ Ψ Z ( r ) , Ψ Y ( t )] U ( r ) , a σ ( u ~ )] d r . (174)
Step 2. With the unbounded K K K handled, we are left with the time-dependent interactions Ψ Z ( r ) \Psi_Z(r) Ψ Z ( r ) , Z ∈ Z Z\in\mathcal{Z} Z ∈ Z , Z ⊂ x [ R ] Z\subset x[R] Z ⊂ x [ R ] . For this, we apply a standard Lieb-Robinson bound on the finite pseudosite lattice CAR algebra CAR ( ⨁ j = 0 L − 1 ( Ran π { j } ( E ) + span { u ~ j } ) ) \operatorname{CAR}\left(\bigoplus_{j=0}^{L-1}(\operatorname{Ran}\pi_{\{j\}}(E)+\operatorname{span}\{\tilde{u}_j\})\right) CAR ( ⨁ j = 0 L − 1 ( Ran π { j } ( E ) + span { u ~ j }) ) . Following [43 ] , we choose the positive symmetric kernel
G ( j , l ) = e − ∥ j − l ∥ / κ 32 ( 1 + ∥ j − l ∥ ) 2 , (175) G(j,l)=\frac{e^{-\lVert j-l\rVert/\kappa}}{32(1+\lVert j-l\rVert)^2}, \tag*{(175)} G ( j , l ) = 32 ( 1 + ∥ j − l ∥ ) 2 e − ∥ j − l ∥ / κ , (175)
which satisfies, uniformly in L L L ,
sup j ∑ l G ( j , l ) ≤ 3 32 , ∑ m G ( j , m ) G ( m , l ) ≤ 3 8 G ( j , l ) . (176) \sup_j\sum_lG(j,l)\le\frac{3}{32},\qquad\sum_mG(j,m)G(m,l)\le\frac{3}{8}G(j,l). \tag*{(176)} j sup l ∑ G ( j , l ) ≤ 32 3 , m ∑ G ( j , m ) G ( m , l ) ≤ 8 3 G ( j , l ) . (176)
By Lemma 6.4 and the finite range of Ψ Z \Psi_Z Ψ Z , we have
∑ Z ⊂ x [ R ] j , l ∈ Z ∥ Ψ Z ( r ) ∥ ≤ C E 1 { ∥ j − l ∥ ≤ κ } ≤ C κ E G ( j , l ) , (177) \sum_{\substack{Z\subset x[R]\\j,l\in Z}}\lVert\Psi_Z(r)\rVert\le CE\mathbf{1}_{\{\lVert j-l\rVert\le\kappa\}}\le C_\kappa EG(j,l), \tag*{(177)} Z ⊂ x [ R ] j , l ∈ Z ∑ ∥ Ψ Z ( r )∥ ≤ C E 1 {∥ j − l ∥ ≤ κ } ≤ C κ E G ( j , l ) , (177)
uniformly in r r r . Thus the hypotheses of [43 ] are verified.
Applying then [43 ] , Theorem 3.1(i) to the integrand in (174) with respect to r r r , and using ∥ Ψ Y ( t ) ∥ = ∥ Ψ Y ( E ) ∥ \lVert\Psi_Y(t)\rVert=\lVert\Psi_Y^{(E)}\rVert ∥ Ψ Y ( t )∥ = ∥ Ψ Y ( E ) ∥ and ∥ a σ ( u ~ ) ∥ = ∥ u ~ ∥ \lVert a^\sigma(\tilde{u})\rVert=\lVert\tilde{u}\rVert ∥ a σ ( u ~ )∥ = ∥ u ~ ∥ , we obtain
∥ [ τ t H R , E ( Ψ Y ( E ) ) , a σ ( u ~ ) ] ∥ ≤ C 0 ∥ Ψ Y ( E ) ∥ ∥ u ~ ∥ ∣ X ∣ e c 1 E t − dist ( X , Y ) / κ . (178) \left\lVert[\tau_t^{H_{R,E}}(\Psi_Y^{(E)}),a^\sigma(\tilde{u})]\right\rVert\le C_0\lVert\Psi_Y^{(E)}\rVert\lVert\tilde{u}\rVert\lvert X\rvert e^{c_1Et-\operatorname{dist}(X,Y)/\kappa}. \tag*{(178)} [ τ t H R , E ( Ψ Y ( E ) ) , a σ ( u ~ )] ≤ C 0 ∥ Ψ Y ( E ) ∥ ∥ u ~ ∥ ∣ X ∣ e c 1 E t − dist ( X , Y ) / κ . (178)
Reversing K K K and all couplings gives the bound for t < 0 t<0 t < 0 .
Remark 6.7. By Lemma 6.4 , increasing E E E improves the mode-cutoff error bound needed to compare H ~ \widetilde{H} H with H ~ R , E \widetilde{H}_{R,E} H R , E , but at the same time it increases the norm bound C E CE C E for the truncated interactions, which was employed in these previous two lemmas; in the the proof of Proposition 2.8 we will balance the two errors by choosing E E E proportional to R / ( 1 + ∣ t ∣ ) R/(1+|t|) R / ( 1 + ∣ t ∣ ) .
Proof of Proposition 2.8
The following lemma gives a form bound for creation and annihilation operators with respect to the MALOs.
Lemma 6.8. For any σ ∈ { − , + } \sigma\in\{-,+\} σ ∈ { − , + } and any f ∈ C ∞ ( T L ; C ) f \in C^\infty(\mathbb{T}_L;\mathbb{C}) f ∈ C ∞ ( T L ; C ) , there exists a constant C s > 0 C_s>0 C s > 0 such that
a − σ ( f ) ( 1 + A s , j 0 ) a σ ( f ) ≤ C s ∥ f ∥ H s ( T L ) 2 ( 1 + A s , j 0 ) (179) a^{-\sigma}(f)(1+A_{s,j_0})a^\sigma(f) \le C_s\lVert f\rVert_{H^s(\mathbb{T}_L)}^2(1+A_{s,j_0}) \tag*{(179)} a − σ ( f ) ( 1 + A s , j 0 ) a σ ( f ) ≤ C s ∥ f ∥ H s ( T L ) 2 ( 1 + A s , j 0 ) (179) in the form sense.
Proof. Let us adopt the shorthands h : = ⨁ j g s , j − j 0 ( T j ) h:=\bigoplus_j g_{s,j-j_0}(T_j) h := ⨁ j g s , j − j 0 ( T j ) and f ~ : = L 1 f \widetilde{f}:=L_1f f := L 1 f . For u ∈ D ( A s , j 0 ) u\in D(A_{s,j_0}) u ∈ D ( A s , j 0 ) , denote L u = : u ~ ∈ D ( d Γ ( h ) 1 / 2 ) \mathcal{L}u=:\widetilde{u}\in D(\mathrm{d}\Gamma(h)^{1/2}) L u =: u ∈ D ( d Γ ( h ) 1/2 ) . We use the localization identity L a σ ( f ) = a σ ( f ~ ) L \mathcal{L}a^\sigma(f)=a^\sigma(\widetilde{f})\mathcal{L} L a σ ( f ) = a σ ( f ) L , to get
A s , j 0 [ a σ ( f ) u ] = ∥ d Γ ( h ) 1 / 2 a σ ( f ~ ) u ~ ∥ 2 . (180) A_{s,j_0}[a^\sigma(f)u]=\lVert\mathrm{d}\Gamma(h)^{1/2}a^\sigma(\widetilde{f})\widetilde{u}\rVert^2. \tag*{(180)} A s , j 0 [ a σ ( f ) u ] = ∥ d Γ ( h ) 1/2 a σ ( f ) u ∥ 2 . (180)
Since the pseudosite Fourier modes form an orthonormal eigenbasis ( e ν ) (e_\nu) ( e ν ) of h h h , with eigenvalues λ ν > 0 \lambda_\nu>0 λ ν > 0 , we obtain the spectral identity
A s , j 0 [ a σ ( f ) u ] = ∥ d Γ ( h ) 1 / 2 a σ ( f ~ ) u ~ ∥ 2 = ∑ ν λ ν ∥ a ( e ν ) a σ ( f ~ ) u ~ ∥ 2 . (181) A_{s,j_0}[a^\sigma(f)u]=\lVert\mathrm{d}\Gamma(h)^{1/2}a^\sigma(\widetilde{f})\widetilde{u}\rVert^2=\sum_\nu\lambda_\nu\lVert a(e_\nu)a^\sigma(\widetilde{f})\widetilde{u}\rVert^2. \tag*{(181)} A s , j 0 [ a σ ( f ) u ] = ∥ d Γ ( h ) 1/2 a σ ( f ) u ∥ 2 = ν ∑ λ ν ∥ a ( e ν ) a σ ( f ) u ∥ 2 . (181)
We now verify its well-posedness. The CAR give
a ( e ν ) a ( f ~ ) = − a ( f ~ ) a ( e ν ) , a ( e ν ) a ∗ ( f ~ ) = ⟨ e ν , f ~ ⟩ − a ∗ ( f ~ ) a ( e ν ) . (182) a(e_\nu)a(\widetilde{f})=-a(\widetilde{f})a(e_\nu),\qquad a(e_\nu)a^*(\widetilde{f})=\langle e_\nu,\widetilde{f}\rangle-a^*(\widetilde{f})a(e_\nu). \tag*{(182)} a ( e ν ) a ( f ) = − a ( f ) a ( e ν ) , a ( e ν ) a ∗ ( f ) = ⟨ e ν , f ⟩ − a ∗ ( f ) a ( e ν ) . (182)
Then, the second identity in (182) and ∥ a ∗ ( f ~ ) ∥ = ∥ f ~ ∥ \lVert a^*(\widetilde{f})\rVert=\lVert\widetilde{f}\rVert ∥ a ∗ ( f )∥ = ∥ f ∥ give
∑ ν = 1 m λ ν ∥ a ( e ν ) a ∗ ( f ~ ) u ~ ∥ 2 ≤ 2 ∥ f ~ ∥ 2 ∑ ν = 1 m λ ν ∥ a ( e ν ) u ~ ∥ 2 + 2 ∑ ν = 1 m λ ν ∣ ⟨ e ν , f ~ ⟩ ∣ 2 ∥ u ~ ∥ 2 ≤ 2 ∥ f ~ ∥ 2 ∥ d Γ ( h ) 1 / 2 u ~ ∥ 2 + 2 ∥ h 1 / 2 f ~ ∥ 2 ∥ u ~ ∥ 2 = 2 ∥ f ∥ 2 A s , j 0 [ u ] + 2 ∥ h 1 / 2 f ~ ∥ 2 ∥ u ∥ 2 . (183) \begin{aligned}
\sum_{\nu=1}^{m}\lambda_\nu\lVert a(e_\nu)a^*(\widetilde{f})\widetilde{u}\rVert^2
&\le2\lVert\widetilde{f}\rVert^2\sum_{\nu=1}^{m}\lambda_\nu\lVert a(e_\nu)\widetilde{u}\rVert^2+2\sum_{\nu=1}^{m}\lambda_\nu|\langle e_\nu,\widetilde{f}\rangle|^2\lVert\widetilde{u}\rVert^2\\
&\le2\lVert\widetilde{f}\rVert^2\lVert\mathrm{d}\Gamma(h)^{1/2}\widetilde{u}\rVert^2+2\lVert h^{1/2}\widetilde{f}\rVert^2\lVert\widetilde{u}\rVert^2\\
&=2\lVert f\rVert^2A_{s,j_0}[u]+2\lVert h^{1/2}\widetilde{f}\rVert^2\lVert u\rVert^2. \tag*{(183)}
\end{aligned} ν = 1 ∑ m λ ν ∥ a ( e ν ) a ∗ ( f ) u ∥ 2 ≤ 2 ∥ f ∥ 2 ν = 1 ∑ m λ ν ∥ a ( e ν ) u ∥ 2 + 2 ν = 1 ∑ m λ ν ∣ ⟨ e ν , f ⟩ ∣ 2 ∥ u ∥ 2 ≤ 2 ∥ f ∥ 2 ∥ d Γ ( h ) 1/2 u ∥ 2 + 2 ∥ h 1/2 f ∥ 2 ∥ u ∥ 2 = 2 ∥ f ∥ 2 A s , j 0 [ u ] + 2 ∥ h 1/2 f ∥ 2 ∥ u ∥ 2 . (183)
Taking m → ∞ m\to\infty m → ∞ , and employing (181) , we prove that a ∗ ( f ) u ∈ D ( A s , j 0 ) a^*(f)u\in D(A_{s,j_0}) a ∗ ( f ) u ∈ D ( A s , j 0 ) . Analogously, the first CAR identity yields
A s , j 0 [ a ( f ) u ] = ∥ d Γ ( h ) 1 / 2 a ( f ~ ) u ~ ∥ 2 = ∑ ν λ ν ∥ a ( f ~ ) a ( e ν ) u ~ ∥ 2 ≤ ∥ f ∥ 2 A s , j 0 [ u ] , (184) A_{s,j_0}[a(f)u]=\lVert\mathrm{d}\Gamma(h)^{1/2}a(\widetilde{f})\widetilde{u}\rVert^2=\sum_\nu\lambda_\nu\lVert a(\widetilde{f})a(e_\nu)\widetilde{u}\rVert^2\le\lVert f\rVert^2A_{s,j_0}[u], \tag*{(184)} A s , j 0 [ a ( f ) u ] = ∥ d Γ ( h ) 1/2 a ( f ) u ∥ 2 = ν ∑ λ ν ∥ a ( f ) a ( e ν ) u ∥ 2 ≤ ∥ f ∥ 2 A s , j 0 [ u ] , (184)
so a ( f ) u a(f)u a ( f ) u also belongs to D ( A s , j 0 ) D(A_{s,j_0}) D ( A s , j 0 ) .
On the other hand, each component of f ~ = ( χ j f ) j \widetilde{f}=(\chi_jf)_j f = ( χ j f ) j is smooth, so we may apply (41) and the product rule to get
∥ h 1 / 2 f ~ ∥ 2 ≤ ∑ j ∥ ( 1 + T j ) s / 2 ( χ j f ) ∥ 2 ≤ C s ∥ f ∥ H s ( T L ) 2 . (185) \lVert h^{1/2}\widetilde{f}\rVert^2\le\sum_j\lVert(1+T_j)^{s/2}(\chi_jf)\rVert^2\le C_s\lVert f\rVert_{H^s(\mathbb{T}_L)}^2. \tag*{(185)} ∥ h 1/2 f ∥ 2 ≤ j ∑ ∥( 1 + T j ) s /2 ( χ j f ) ∥ 2 ≤ C s ∥ f ∥ H s ( T L ) 2 . (185)
Finally it is enough to insert (185) into (183) to obtain
A s , j 0 [ a σ ( f ) u ] ≤ C s ∥ f ∥ H s ( T L ) 2 ( ∥ u ∥ 2 + A s , j 0 [ u ] ) , (186) A_{s,j_0}[a^\sigma(f)u]\le C_s\lVert f\rVert_{H^s(\mathbb{T}_L)}^2(\lVert u\rVert^2+A_{s,j_0}[u]), \tag*{(186)} A s , j 0 [ a σ ( f ) u ] ≤ C s ∥ f ∥ H s ( T L ) 2 (∥ u ∥ 2 + A s , j 0 [ u ]) , (186)
hence proving the claim.
We are now ready to prove Proposition 2.8.
Proof of Proposition 2.8. Here, we shall use the shorthand a = a σ ( L 1 f ) \mathbf{a} = a^{\sigma}(\mathcal{L}_{1}f) a = a σ ( L 1 f ) , and we shall set p = k + 1 p = k + 1 p = k + 1 and s = 2 p + 2 = 2 k + 4 s = 2p + 2 = 2k + 4 s = 2 p + 2 = 2 k + 4 , and let γ \gamma γ be the constant in (44) at order s s s .
First of all, for the case of large R R R , we make the ansatz
a R σ ( t , f ) = τ t H R , E ( a ) = τ t H R , E ( a σ ( L 1 f ) ) , a_{R}^{\sigma}(t,f) = \tau_{t}^{H_{R,E}}(\mathbf{a}) = \tau_{t}^{H_{R,E}}\left(a^{\sigma}(\mathcal{L}_{1}f)\right), a R σ ( t , f ) = τ t H R , E ( a ) = τ t H R , E ( a σ ( L 1 f ) ) ,
with cutoff E E E to be suitably chosen in terms of R R R and t t t (see Step 3). Indeed, it satisfies all of the requirements of the claim: its support lies in A x [ R ] \mathcal{A}_{x[R]} A x [ R ] because both a \mathbf{a} a and H R , E H_{R,E} H R , E are supported on the pseudosite set x [ R ] x[R] x [ R ] ; it is odd and has norm ∥ f ∥ 2 \|f\|_{2} ∥ f ∥ 2 ; it is independent of the initial state ρ \rho ρ .
∥ [ L τ t H ( a σ ( f ) ) − a R σ ( t , f ) L ] ρ 1 / 2 ∥ 2 = ∥ [ τ t H ~ ( a ) − τ t H R , E ( a ) ] L ρ 1 / 2 ∥ 2 . (187) \left\|\left[\mathcal{L}\tau_{t}^{H}\left(a^{\sigma}(f)\right)-a_{R}^{\sigma}(t,f)\mathcal{L}\right]\rho^{1/2}\right\|_{2}
=
\left\|\left[\tau_{t}^{\widetilde{H}}(\mathbf{a})-\tau_{t}^{H_{R,E}}(\mathbf{a})\right]\mathcal{L}\rho^{1/2}\right\|_{2}. \tag*{(187)} [ L τ t H ( a σ ( f ) ) − a R σ ( t , f ) L ] ρ 1/2 2 = [ τ t H ( a ) − τ t H R , E ( a ) ] L ρ 1/2 2 . (187)
Telescoping τ t H ^ R , E \tau_{t}^{\widehat{H}_{R,E}} τ t H R , E and using ∥ L ρ 1 / 2 ∥ 2 = 1 \|\mathcal{L}\rho^{1/2}\|_{2}=1 ∥ L ρ 1/2 ∥ 2 = 1 , we then obtain
∥ [ L τ t H ( a σ ( f ) ) − a R σ ( t , f ) L ] ρ 1 / 2 ∥ 2 ≤ ∥ [ τ t H ~ ( a ) − τ t H ^ R , E ( a ) ] L ρ 1 / 2 ∥ 2 + ∥ τ t H ^ R , E ( a ) − τ t H R , E ( a ) ∥ . (188) \left\|\left[\mathcal{L}\tau_{t}^{H}\left(a^{\sigma}(f)\right)-a_{R}^{\sigma}(t,f)\mathcal{L}\right]\rho^{1/2}\right\|_{2}
\leq
\left\|\left[\tau_{t}^{\widetilde{H}}(\mathbf{a})-\tau_{t}^{\widehat{H}_{R,E}}(\mathbf{a})\right]\mathcal{L}\rho^{1/2}\right\|_{2}
+
\left\|\tau_{t}^{\widehat{H}_{R,E}}(\mathbf{a})-\tau_{t}^{H_{R,E}}(\mathbf{a})\right\|. \tag*{(188)} [ L τ t H ( a σ ( f ) ) − a R σ ( t , f ) L ] ρ 1/2 2 ≤ [ τ t H ( a ) − τ t H R , E ( a ) ] L ρ 1/2 2 + τ t H R , E ( a ) − τ t H R , E ( a ) . (188)
We will dedicate Step 1 (respectively 2) to estimate the first (resp. second) term, and Step 3 to appropriately choose the cutoff.
Step 1: bound for the first term. We put Δ : = H ~ − H ^ R , E \Delta:= \widetilde{H}-\widehat{H}_{R,E} Δ := H − H R , E , and for each Z ∩ x [ R ] ≠ ∅ Z\cap x[R]\neq\varnothing Z ∩ x [ R ] = ∅ , we choose a center j Z ∈ Z j_{Z}\in Z j Z ∈ Z . Since ∥ j − j Z ∥ ≤ κ \|j-j_{Z}\|\leq\kappa ∥ j − j Z ∥ ≤ κ for any j ∈ Z j\in Z j ∈ Z , (41) gives the form bound
L ∗ d Γ ( ( 1 + T Z ) s ) L ≤ C A s , j Z , (189) \mathcal{L}^{*}\mathrm{d}\Gamma\left((1+T_{Z})^{s}\right)\mathcal{L}\leq CA_{s,j_{Z}}, \tag*{(189)} L ∗ d Γ ( ( 1 + T Z ) s ) L ≤ C A s , j Z , (189)
which, together with the stronger bound established in Step 3 of the proof of Lemma 6.4 , yields the norm estimate
∥ Δ L u ∥ ≤ C E − p ∑ Z ∩ x [ R ] ≠ ∅ A s , j Z [ u ] 1 / 2 (190) \|\Delta\mathcal{L}u\|\leq CE^{-p}\sum_{Z\cap x[R]\neq\varnothing} A_{s,j_{Z}}[u]^{1/2} \tag*{(190)} ∥Δ L u ∥ ≤ C E − p Z ∩ x [ R ] = ∅ ∑ A s , j Z [ u ] 1/2 (190)
for any u ∈ D u\in\mathcal{D} u ∈ D . Writing U ~ ( t ) = e − i t H ~ \widetilde{U}(t)=e^{-it\widetilde{H}} U ( t ) = e − i t H , U ^ ( t ) = e − i t H ^ R , E \widehat{U}(t)=e^{-it\widehat{H}_{R,E}} U ( t ) = e − i t H R , E , by a Duhamel iteration we obtain
[ τ t H ~ ( a ) − τ t H ^ R , E ( a ) ] L u = i ∫ 0 t [ U ^ ( r − t ) Δ U ~ ( − r ) a U ~ ( t ) − U ^ ( − t ) a U ^ ( r ) Δ U ~ ( t − r ) ] L u d r . (191) \left[\tau_{t}^{\widetilde{H}}(\mathbf{a})-\tau_{t}^{\widehat{H}_{R,E}}(\mathbf{a})\right]\mathcal{L}u
=
i\int_{0}^{t}
\left[
\widehat{U}(r-t)\Delta\widetilde{U}(-r)\mathbf{a}\widetilde{U}(t)
-
\widehat{U}(-t)\mathbf{a}\widehat{U}(r)\Delta\widetilde{U}(t-r)
\right]\mathcal{L}u\,\mathrm{d}r. \tag*{(191)} [ τ t H ( a ) − τ t H R , E ( a ) ] L u = i ∫ 0 t [ U ( r − t ) Δ U ( − r ) a U ( t ) − U ( − t ) a U ( r ) Δ U ( t − r ) ] L u d r . (191)
We estimate the norm of the integrand in (191) by
∥ Δ U ~ ( − r ) a U ~ ( t ) L u ∥ + ∥ a ∥ ∥ Δ U ~ ( t − r ) L u ∥ ≤ C E − p e γ ∣ r ∣ / 2 ∑ Z A s , j Z [ a σ ( f ) e − i t H u ] 1 / 2 + C ∥ f ∥ 2 E − p e γ ∣ t − r ∣ / 2 ∑ Z A s , j Z [ u ] 1 / 2 ≤ C ∥ f ∥ H s E − p e γ ∣ t ∣ ∑ Z ( ∥ u ∥ 2 + A s , j Z [ u ] ) 1 / 2 . (192) \begin{aligned}
\|\Delta\widetilde{U}(-r)\mathbf{a}\widetilde{U}(t)\mathcal{L}u\|+\|\mathbf{a}\|\|\Delta\widetilde{U}(t-r)\mathcal{L}u\|
&\leq CE^{-p}e^{\gamma|r|/2}\sum_{Z}A_{s,j_{Z}}\left[a^{\sigma}(f)e^{-itH_{u}}\right]^{1/2}\\
&\quad+C\|f\|_{2}E^{-p}e^{\gamma|t-r|/2}\sum_{Z}A_{s,j_{Z}}[u]^{1/2}\\
&\leq C\|f\|_{H^{s}}E^{-p}e^{\gamma|t|}\sum_{Z}\left(\|u\|^{2}+A_{s,j_{Z}}[u]\right)^{1/2}. \tag*{(192)}
\end{aligned} ∥Δ U ( − r ) a U ( t ) L u ∥ + ∥ a ∥∥Δ U ( t − r ) L u ∥ ≤ C E − p e γ ∣ r ∣/2 Z ∑ A s , j Z [ a σ ( f ) e − i t H u ] 1/2 + C ∥ f ∥ 2 E − p e γ ∣ t − r ∣/2 Z ∑ A s , j Z [ u ] 1/2 ≤ C ∥ f ∥ H s E − p e γ ∣ t ∣ Z ∑ ( ∥ u ∥ 2 + A s , j Z [ u ] ) 1/2 . (192)
where in the first passage we used (190) and then Proposition 2.6 regarding time propagation of the MALO; in the last passage we employed Lemma 6.8 followed by Proposition 2.6 again. Integrating over r r r we obtain the same bound multiplied by ∣ t ∣ |t| ∣ t ∣ :
∥ ( τ t H ~ ( a ) − τ t H ^ R , E ( a ) ) L u ∥ ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ ∑ Z ( ∥ u ∥ 2 + A s , j Z [ u ] ) 1 / 2 . (193) \left\|\left(\tau_{t}^{\widetilde{H}}(\mathbf{a})-\tau_{t}^{\widehat{H}_{R,E}}(\mathbf{a})\right)\mathcal{L}u\right\|
\leq
C\|f\|_{H^{s}}|t|E^{-p}e^{\gamma|t|}
\sum_{Z}\left(\|u\|^{2}+A_{s,j_{Z}}[u]\right)^{1/2}. \tag*{(193)} ( τ t H ( a ) − τ t H R , E ( a ) ) L u ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ Z ∑ ( ∥ u ∥ 2 + A s , j Z [ u ] ) 1/2 . (193)
By denseness, the bound (193) extends to all u u u in the joint form domain of the A s , j A_{s,j} A s , j . We may assume max j Tr ρ A s , j < ∞ \max_j \operatorname{Tr}\rho A_{s,j}<\infty max j Tr ρ A s , j < ∞ , since otherwise the claimed estimate is trivial. Then every eigenvector of ρ \rho ρ with positive eigenvalue belongs to this domain, and hence we get
∥ ( τ ~ t H ( a ) − τ ^ t H R , E ( a ) ) L ρ 1 / 2 ∥ 2 ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ ∑ Z ( 1 + Tr ρ A s , j Z ) 1 / 2 ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ R Q s ( ρ ) , (194) \begin{aligned}
\left\|\left(\widetilde{\tau}^{H}_{t}(a)-\widehat{\tau}^{H_{R,E}}_{t}(a)\right)\mathcal{L}\rho^{1/2}\right\|_{2}\leq C\|f\|_{H^{s}}|t|E^{-p}e^{\gamma|t|}\sum_{Z}\left(1+\operatorname{Tr}\rho A_{s,j_Z}\right)^{1/2} \\
&\leq C\|f\|_{H^{s}}|t|E^{-p}e^{\gamma|t|}RQ_s(\rho), \tag*{(194)}
\end{aligned} ( τ t H ( a ) − τ t H R , E ( a ) ) L ρ 1/2 2 ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ Z ∑ ( 1 + Tr ρ A s , j Z ) 1/2 ≤ C ∥ f ∥ H s ∣ t ∣ E − p e γ ∣ t ∣ R Q s ( ρ ) , (194)
where in the last passage we used that the number of terms in the sum is bounded by C R CR C R , and where we introduced Q s ( ρ ) Q_s(\rho) Q s ( ρ ) as
Q s ( ρ ) : = sup j ∈ { 0 , … , L − 1 } ( 1 + Tr ρ A s , j ) 1 / 2 . (195) Q_s(\rho):=\sup_{j\in\{0,\ldots,L-1\}}\left(1+\operatorname{Tr}\rho A_{s,j}\right)^{1/2}. \tag*{(195)} Q s ( ρ ) := j ∈ { 0 , … , L − 1 } sup ( 1 + Tr ρ A s , j ) 1/2 . (195)
Step 2: bound for the second term. Let R ≥ R 0 : = 4 κ + 8 R\geq R_0:=4\kappa+8 R ≥ R 0 := 4 κ + 8 . Then the support X X X of a a a , given in (43) , lies in x [ R ] x[R] x [ R ] , and it has r X < 2 r_X<2 r X < 2 . Applying Lemma 6.5 with u ~ = L 1 f \widetilde{u}=\mathcal{L}_1f u = L 1 f yields
∥ τ ^ t H R , E ( a ) − τ t H R , E ( a ) ∥ ≤ C ∥ f ∥ 2 e − c R ( e c 1 E ∣ t ∣ − 1 ) . (196) \left\|\widehat{\tau}^{H_{R,E}}_{t}(a)-\tau^{H_{R,E}}_{t}(a)\right\|\leq C\|f\|_2e^{-cR(e^{c_1E|t|}-1)}. \tag*{(196)} τ t H R , E ( a ) − τ t H R , E ( a ) ≤ C ∥ f ∥ 2 e − c R ( e c 1 E ∣ t ∣ − 1 ) . (196)
Step 3: choice of cutoff. Inserting the bounds from Steps 1 and 2 into (188) and using Q s ( ρ ) ≥ 1 Q_s(\rho)\geq1 Q s ( ρ ) ≥ 1 finally yields
∥ [ L τ H ( a σ ( f ) ) − a R σ ( t , f ) L ] ρ 1 / 2 ∥ 2 ≤ C ∥ f ∥ H s Q s ( ρ ) [ R ∣ t ∣ E − p e γ ∣ t ∣ + e − c R ( e c 1 E ∣ t ∣ − 1 ) ] . (197) \left\|\left[\mathcal{L}^{H}_{\tau}(a^\sigma(f))-a^\sigma_R(t,f)\mathcal{L}\right]\rho^{1/2}\right\|_2\leq C\|f\|_{H^s}Q_s(\rho)\left[R|t|E^{-p}e^{\gamma|t|}+e^{-cR(e^{c_1E|t|}-1)}\right]. \tag*{(197)} [ L τ H ( a σ ( f )) − a R σ ( t , f ) L ] ρ 1/2 2 ≤ C ∥ f ∥ H s Q s ( ρ ) [ R ∣ t ∣ E − p e γ ∣ t ∣ + e − c R ( e c 1 E ∣ t ∣ − 1 ) ] . (197)
We now pick 0 < ε ≤ 1 0<\varepsilon\leq1 0 < ε ≤ 1 such that c 1 ε < c / 2 c_1\varepsilon<c/2 c 1 ε < c /2 ; notice that it is independent of k k k . For R ≥ max { R 0 , ( 1 + ∣ t ∣ ) / ε } R\geq\max\{R_0,(1+|t|)/\varepsilon\} R ≥ max { R 0 , ( 1 + ∣ t ∣ ) / ε } , choosing the energy cutoff as E : = ε R / ( 1 + ∣ t ∣ ) ≥ 1 E:=\varepsilon R/(1+|t|)\geq1 E := εR / ( 1 + ∣ t ∣ ) ≥ 1 gives
R ∣ t ∣ E − p e γ ∣ t ∣ = ε − p R − k ∣ t ∣ ( 1 + ∣ t ∣ ) p e γ ∣ t ∣ ≤ C k ( 1 + R ) − k e ( γ + k + 2 ) ∣ t ∣ , R|t|E^{-p}e^{\gamma|t|}=\varepsilon^{-p}R^{-k}|t|(1+|t|)^pe^{\gamma|t|}\leq C_k(1+R)^{-k}e^{(\gamma+k+2)|t|}, R ∣ t ∣ E − p e γ ∣ t ∣ = ε − p R − k ∣ t ∣ ( 1 + ∣ t ∣ ) p e γ ∣ t ∣ ≤ C k ( 1 + R ) − k e ( γ + k + 2 ) ∣ t ∣ ,
e − c R ( e c 1 E ∣ t ∣ − 1 ) ≤ e − c R / 2 ≤ C k ( 1 + R ) − k . (198) e^{-cR(e^{c_1E|t|}-1)}\leq e^{-cR/2}\leq C_k(1+R)^{-k}. \tag*{(198)} e − c R ( e c 1 E ∣ t ∣ − 1 ) ≤ e − c R /2 ≤ C k ( 1 + R ) − k . (198)
therefore proving the claim for large R R R . For R < max { R 0 , ( 1 + ∣ t ∣ ) / ε } R<\max\{R_0,(1+|t|)/\varepsilon\} R < max { R 0 , ( 1 + ∣ t ∣ ) / ε } , our pick is a R σ ( t , f ) = 0 a^\sigma_R(t,f)=0 a R σ ( t , f ) = 0 , which yields an error of at most ∥ f ∥ 2 \|f\|_2 ∥ f ∥ 2 for all t ∈ R t\in\mathbb{R} t ∈ R , while 1 + R ≤ ( 1 + R 0 + ε − 1 ) ( 1 + ∣ t ∣ ) 1+R\leq(1+R_0+\varepsilon^{-1})(1+|t|) 1 + R ≤ ( 1 + R 0 + ε − 1 ) ( 1 + ∣ t ∣ ) and ( 1 + ∣ t ∣ ) k ≤ e k ∣ t ∣ (1+|t|)^k\leq e^{k|t|} ( 1 + ∣ t ∣ ) k ≤ e k ∣ t ∣ , so the same bound holds after possibly enlarging the prefactor.
Relative boundedness of the kinetic coupling
Proof of Lemma 6.2. For smooth f , g ∈ D ( T ) f,g\in D(T) f , g ∈ D ( T ) , integration by parts yields
⟨ f j , B j k g k ⟩ L 2 ( I j ) = ⟨ B k j f j , g k ⟩ L 2 ( I k ) , (199) \langle f_j,B_{jk}g_k\rangle_{L^2(I_j)}=\langle B_{kj}f_j,g_k\rangle_{L^2(I_k)}, \tag*{(199)} ⟨ f j , B j k g k ⟩ L 2 ( I j ) = ⟨ B k j f j , g k ⟩ L 2 ( I k ) , (199)
because a k j = − a j k a_{kj}=-a_{jk} a k j = − a j k and the overlap coefficients vanish near both pseudosite boundaries. Summing over j , k j,k j , k and adding S S S then proves symmetry of C C C on smooth vectors.
Next, we obtain the relative bound for smooth f f f : using the fact that a j k = 0 a_{jk}=0 a j k = 0 if ∥ j − k ∥ ≥ 2 \lVert j-k\rVert\geq2 ∥ j − k ∥ ≥ 2 , for any ε > 0 \varepsilon>0 ε > 0 we have
∥ C f ∥ 2 ≤ C ∑ j ( ∥ f j ∥ 2 + ∑ k : ∥ j − k ∥ ≤ 1 ( ∥ f k ′ ∥ 2 + ∥ f k ∥ 2 ) ) ≤ C ∑ k ( ∥ f k ′ ∥ 2 + ∥ f k ∥ 2 ) = C ∥ ( 1 + T ) 1 / 2 f ∥ 2 ≤ ε 2 ∥ T f ∥ 2 + C ε 2 ∥ f ∥ 2 , (200) \begin{aligned}
\|Cf\|^2\leq C\sum_j\left(\|f_j\|^2+\sum_{k:\|j-k\|\leq1}\left(\|f'_k\|^2+\|f_k\|^2\right)\right)
\leq C\sum_k\left(\|f'_k\|^2+\|f_k\|^2\right)=C\|(1+T)^{1/2}f\|^2 \\
&\leq\varepsilon^2\|Tf\|^2+C_\varepsilon^2\|f\|^2, \tag*{(200)}
\end{aligned} ∥ C f ∥ 2 ≤ C j ∑ ∥ f j ∥ 2 + k : ∥ j − k ∥ ≤ 1 ∑ ( ∥ f k ′ ∥ 2 + ∥ f k ∥ 2 ) ≤ C k ∑ ( ∥ f k ′ ∥ 2 + ∥ f k ∥ 2 ) = C ∥ ( 1 + T ) 1/2 f ∥ 2 ≤ ε 2 ∥ T f ∥ 2 + C ε 2 ∥ f ∥ 2 , (200)
where in the last inequality we used C ( 1 + λ ) ≤ ε 2 λ 2 + C ε 2 C(1+\lambda)\leq\varepsilon^2\lambda^2+C_\varepsilon^2 C ( 1 + λ ) ≤ ε 2 λ 2 + C ε 2 for λ ≥ 0 \lambda\geq0 λ ≥ 0 . This bound establishes both that C C C is symmetric on D ( T ) D(T) D ( T ) and that it is infinitesimally T T T -bounded. By the Kato-Rellich theorem, we then conclude that T + C T+C T + C is self-adjoint.
Extension to polynomial observables
In this appendix, we prove Theorem 1.5 .
The polynomial result follows from the initial comparison and the insertion and approximation estimates below. Insertion is controlled by iteration in the form norm; approximation also requires a product expansion for the boundary error.
Replacing each a σ ( f ) a^\sigma(f) a σ ( f ) in a physical polynomial A A A by a σ ( L 1 f ) a^\sigma(\mathcal{L}_1 f) a σ ( L 1 f ) defines A ~ \widetilde{A} A with
A ~ L = L A , A ~ ∗ L = L A ∗ , ∥ A ~ ∥ = ∥ A ∥ . (201) \widetilde{A}\mathcal{L}=\mathcal{L}A,\qquad\widetilde{A}^{*}\mathcal{L}=\mathcal{L}A^{*},\qquad\|\widetilde{A}\|=\|A\|. \tag*{(201)} A L = L A , A ∗ L = L A ∗ , ∥ A ∥ = ∥ A ∥. (201)
The intertwining identities follow from (34) by multiplication and linearity. Fock factorization over Ran L 1 \operatorname{Ran}\mathcal{L}_1 Ran L 1 and its orthogonal complement proves the norm identity and shows that the lift is independent of the chosen polynomial expression. If A ∈ P ϵ ( B 1 ( x ) ) A\in\mathcal{P}^{\epsilon}(B_1(x)) A ∈ P ϵ ( B 1 ( x )) , then its corresponding A ~ \widetilde{A} A has also parity ϵ \epsilon ϵ and its support is contained
X = { j : supp χ j ∩ B 1 ( x ) ≠ ∅ } ⊂ { j : ∥ j − x ∥ < 2 } , ∣ X ∣ ≤ C . (202) X=\{j:\operatorname{supp}\chi_j\cap B_1(x)\ne\varnothing\}\subset\{j:\|j-x\|<2\},\qquad|X|\le C. \tag*{(202)} X = { j : supp χ j ∩ B 1 ( x ) = ∅ } ⊂ { j : ∥ j − x ∥ < 2 } , ∣ X ∣ ≤ C . (202)
Lemma B.1 (Action of polynomial observables). For s ∈ N s\in\mathbb{N} s ∈ N , a reference pseudosite j 0 j_0 j 0 , and every finite polynomial A A A in smooth creation and annihilation operators, the following holds in the form sense on D ( A s , j 0 ) D(A_{s,j_0}) D ( A s , j 0 ) :
A ∗ ( 1 + A s , j 0 ) A ≤ C A ( 1 + A s , j 0 ) . (203) A^{*}(1+A_{s,j_0})A\le C_A(1+A_{s,j_0}). \tag*{(203)} A ∗ ( 1 + A s , j 0 ) A ≤ C A ( 1 + A s , j 0 ) . (203)
for a constant C A C_A C A dependent on the chosen expression for A A A , in particular the H s H^s H s -norms of its test functions, uniformly in L , j 0 L,j_0 L , j 0 .
Lemma B.2 (Approximation of polynomial observables). Let k ∈ N k\in\mathbb{N} k ∈ N , A ∈ P ϵ ( B 1 ( x ) ) A\in\mathcal{P}^{\epsilon}(B_1(x)) A ∈ P ϵ ( B 1 ( x )) , and 1 ≤ R ≤ L / 4 1\le R\le L/4 1 ≤ R ≤ L /4 . For every t ∈ R t\in\mathbb{R} t ∈ R , there is A R ( t ) ∈ A x [ R ] A_R(t)\in\mathcal{A}_{x[R]} A R ( t ) ∈ A x [ R ] , of parity ϵ \epsilon ϵ , such that ∥ A R ( t ) ∥ ≤ ∥ A ∥ \|A_R(t)\|\le\|A\| ∥ A R ( t ) ∥ ≤ ∥ A ∥ and, for every state ρ \rho ρ ,
∥ ( L t H ( A ) − A R ( t ) L ) ρ 1 / 2 ∥ 2 ≤ C A e ( γ + k + 2 ) ∣ t ∣ ( 1 + R ) k ( 1 + max 0 ≤ j < L Tr ρ A 2 k + 4 , j ) 1 / 2 . (204) \|(\mathcal{L}^{H}_t(A)-A_R(t)\mathcal{L})\rho^{1/2}\|_2\le\frac{C_A e^{(\gamma+k+2)|t|}}{(1+R)^k}\left(1+\max_{0\le j<L}\operatorname{Tr}\rho A_{2k+4,j}\right)^{1/2}. \tag*{(204)} ∥ ( L t H ( A ) − A R ( t ) L ) ρ 1/2 ∥ 2 ≤ ( 1 + R ) k C A e ( γ + k + 2 ) ∣ t ∣ ( 1 + 0 ≤ j < L max Tr ρ A 2 k + 4 , j ) 1/2 . (204)
Remark B.3. Here, γ \gamma γ is the constant in (44) at order 2 k + 4 2k+4 2 k + 4 . Moreover, the constant C A C_A C A depends on k k k and on the chosen finite expression for A A A ; in particular, it depends on the H 2 k + 4 H^{2k+4} H 2 k + 4 -norms of its test functions.
Proof of Theorem 1.5 . Set s = 2 k + 4 s=2k+4 s = 2 k + 4 . The initial comparison and (B.3) yield
1 + max j Tr ρ A s , j ≤ C M 2 k + 6 ( ρ ) , 1+\max_j\operatorname{Tr}\rho A_{s,j}\le CM_{2k+6}(\rho), 1 + j max Tr ρ A s , j ≤ C M 2 k + 6 ( ρ ) ,
Tr B ρ B ∗ + max j Tr B ρ B ∗ A s , j ≤ C B ( 1 + max j Tr ρ A s , j ) ≤ C B M 2 k + 6 ( ρ ) . (205) \operatorname{Tr}B\rho B^{*}+\max_j\operatorname{Tr}B\rho B^{*}A_{s,j}\le C_B\left(1+\max_j\operatorname{Tr}\rho A_{s,j}\right)\le C_BM_{2k+6}(\rho). \tag*{(205)} Tr B ρ B ∗ + j max Tr B ρ B ∗ A s , j ≤ C B ( 1 + j max Tr ρ A s , j ) ≤ C B M 2 k + 6 ( ρ ) . (205)
For sufficiently large d = ∥ x − y ∥ d=\|x-y\| d = ∥ x − y ∥ , take R = d / 3 ≤ L / 6 R=d/3\le L/6 R = d /3 ≤ L /6 . The supports of A R ( t ) A_R(t) A R ( t ) and B ~ \widetilde{B} B are disjoint, so [ A R ( t ) , B ~ ] g r = 0 [A_R(t),\widetilde{B}]_{\mathrm{gr}}=0 [ A R ( t ) , B ] gr = 0 . Set
D = L t H ( A ) − A R ( t ) L . (206) D=\mathcal{L}^{H}_t(A)-A_R(t)\mathcal{L}. \tag*{(206)} D = L t H ( A ) − A R ( t ) L . (206)
The intertwining identities give
L [ L t H ( A ) , B ] g r = D B − ( − 1 ) ϵ δ B ~ D , \mathcal{L}[\mathcal{L}^{H}_t(A),B]_{\mathrm{gr}}=DB-(-1)^{\epsilon\delta}\widetilde{B}D, L [ L t H ( A ) , B ] gr = D B − ( − 1 ) ϵδ B D ,
∥ [ L t H ( A ) , B ] g r ρ 1 / 2 ∥ 2 ≤ ∥ D B ρ 1 / 2 ∥ 2 + ∥ B ∥ ∥ D ρ 1 / 2 ∥ 2 . (207) \|[\mathcal{L}^{H}_t(A),B]_{\mathrm{gr}}\rho^{1/2}\|_2\le\|DB\rho^{1/2}\|_2+\|B\|\|D\rho^{1/2}\|_2. \tag*{(207)} ∥ [ L t H ( A ) , B ] gr ρ 1/2 ∥ 2 ≤ ∥ D B ρ 1/2 ∥ 2 + ∥ B ∥∥ D ρ 1/2 ∥ 2 . (207)
Apply (B.4) with ρ \rho ρ and, if β = Tr B ρ B ∗ > 0 \beta=\operatorname{Tr}B\rho B^{*}>0 β = Tr B ρ B ∗ > 0 , with β − 1 B ρ B ∗ \beta^{-1}B\rho B^{*} β − 1 B ρ B ∗ , multiplying the latter estimate by β 1 / 2 \beta^{1/2} β 1/2 . Since
∥ D B ρ 1 / 2 ∥ 2 2 = Tr B ρ B ∗ D ∗ D = ∥ D ( B ρ B ∗ ) 1 / 2 ∥ 2 2 , (208) \|DB\rho^{1/2}\|_2^2=\operatorname{Tr}B\rho B^{*}D^{*}D=\|D(B\rho B^{*})^{1/2}\|_2^2, \tag*{(208)} ∥ D B ρ 1/2 ∥ 2 2 = Tr B ρ B ∗ D ∗ D = ∥ D ( B ρ B ∗ ) 1/2 ∥ 2 2 , (208)
the preceding weighted-form bounds give (26) . If β = 0 \beta=0 β = 0 , then B ρ 1 / 2 = 0 B\rho^{1/2}=0 B ρ 1/2 = 0 . For bounded d d d , the estimate follows from ∥ [ L t H ( A ) , B ] g r ∥ ≤ 2 ∥ A ∥ ∥ B ∥ \|[\mathcal{L}^{H}_t(A),B]_{\mathrm{gr}}\|\le2\|A\|\|B\| ∥ [ L t H ( A ) , B ] gr ∥ ≤ 2∥ A ∥∥ B ∥ .
Adaptation of the insertion estimate
Proof of Lemma B.1. For s ∈ N s \in\mathbb{N} s ∈ N and a reference pseudosite j 0 j_0 j 0 , the single-operator estimate (179) gives boundedness of every smooth creation or annihilation operator on D ( A s , j 0 ) D(A_{s,j_0}) D ( A s , j 0 ) in the norm
u ↦ ( ∥ u ∥ 2 + A s , j 0 [ u ] ) 1 / 2 . (209) u \mapsto\left(\lVert u\rVert^2 + A_{s,j_0}[u]\right)^{1/2}. \tag*{(209)} u ↦ ( ∥ u ∥ 2 + A s , j 0 [ u ] ) 1/2 . (209)
Iterating this, we obtain the bound for every monomial. The triangle inequality in this norm gives (B.3) for finite linear combinations.
Adaptation of the local approximation
Proof of Lemma B.2. The mode comparison follows from (191) and the polynomial insertion estimate. The boundary comparison then follows by expanding the difference of two evolved products.
Set s = 2 k + 4 s=2k+4 s = 2 k + 4 and p = k + 1 p=k+1 p = k + 1 . Sums over Z \mathcal{Z} Z run over the at most C R C R C R members of Z \mathcal{Z} Z meeting x [ R ] x[R] x [ R ] , with j Z ∈ Z j_{\mathcal{Z}}\in\mathcal{Z} j Z ∈ Z chosen as in the proof of Proposition 2.8. For u ∈ D u\in\mathcal{D} u ∈ D , t ≥ 0 t\geq0 t ≥ 0 , and 0 ≤ r ≤ t 0\leq r\leq t 0 ≤ r ≤ t , (B.3) and (44) give
A s , j Z [ e i r H A A e − i t H u ] ≤ C A e γ ( r + t ) ( ∥ u ∥ 2 + A s , j Z [ u ] ) , A_{s,j_{\mathcal{Z}}}\left[e^{irH_A}Ae^{-itH}u\right]\leq C_A e^{\gamma(r+t)}\left(\lVert u\rVert^2+A_{s,j_{\mathcal{Z}}}[u]\right), A s , j Z [ e i r H A A e − i t H u ] ≤ C A e γ ( r + t ) ( ∥ u ∥ 2 + A s , j Z [ u ] ) ,
A s , j Z [ e − i ( t − r ) H u ] ≤ e γ ( t − r ) A s , j Z [ u ] . (210) A_{s,j_{\mathcal{Z}}}\left[e^{-i(t-r)H}u\right]\leq e^{\gamma(t-r)}A_{s,j_{\mathcal{Z}}}[u]. \tag*{(210)} A s , j Z [ e − i ( t − r ) H u ] ≤ e γ ( t − r ) A s , j Z [ u ] . (210)
For the observable A ~ \widetilde{A} A , the intertwining identities (B.1) and (6.10) place the two mode errors in (191) on the vectors above. The estimate (190) takes square roots of these form bounds, so both exponential factors are at most e γ t e^{\gamma t} e γ t . Reversing both generators covers negative times. The same extension from D \mathcal{D} D to the common form domain and triangle inequality over eigenvectors as in the proof of Proposition 2.8 therefore give
∥ ( τ t H ~ ( A ~ ) − τ t H ^ R , E ( A ~ ) ) L ρ 1 / 2 ∥ 2 ≤ C A ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ ∑ Z ( 1 + Tr ρ A s , j Z ) 1 / 2 ≤ C A R ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ ( 1 + max j Tr ρ A s , j ) 1 / 2 . (211) \begin{aligned}
\left\lVert\left(\tau_t^{\widetilde{H}}(\widetilde{A})-\tau_t^{\widehat{H}_{R,E}}(\widetilde{A})\right)\mathcal{L}\rho^{1/2}\right\rVert_2
\leq C_A |t| E^{-(k+1)}e^{\gamma|t|}\sum_{\mathcal{Z}}\left(1+\operatorname{Tr}\rho A_{s,j_{\mathcal{Z}}}\right)^{1/2} \\
&\leq C_A R|t|E^{-(k+1)}e^{\gamma|t|}\left(1+\max_j\operatorname{Tr}\rho A_{s,j}\right)^{1/2}. \tag*{(211)}
\end{aligned} ( τ t H ( A ) − τ t H R , E ( A ) ) L ρ 1/2 2 ≤ C A ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ Z ∑ ( 1 + Tr ρ A s , j Z ) 1/2 ≤ C A R ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ ( 1 + j max Tr ρ A s , j ) 1/2 . (211)
If the right side is finite, all positive-weight eigenvectors lie in the required common form domain; otherwise the estimate is automatic.
For the boundary comparison, first let P = ∏ i = 1 m a σ i ( f i ) P=\prod_{i=1}^{m}a^{\sigma_i}(f_i) P = ∏ i = 1 m a σ i ( f i ) be a monomial with m ≥ 1 m\geq1 m ≥ 1 and supp f i ⊂ B 1 ( x ) \operatorname{supp}f_i\subset B_1(x) supp f i ⊂ B 1 ( x ) . Put c i = a σ i ( L 1 f i ) c_i=a^{\sigma_i}(\mathcal{L}_1f_i) c i = a σ i ( L 1 f i ) , and abbreviate α = τ t H ^ R , E \alpha=\tau_t^{\widehat{H}_{R,E}} α = τ t H R , E and β = τ t H R , E \beta=\tau_t^{H_{R,E}} β = τ t H R , E in the following calculation. Multiplicativity of these evolutions gives the ordered product identity
α ( P ~ ) − β ( P ~ ) = ∑ i = 1 m ( ∏ ℓ < i α ( c ℓ ) ) ( α ( c i ) − β ( c i ) ) ( ∏ ℓ > i β ( c ℓ ) ) . (212) \alpha(\widetilde{P})-\beta(\widetilde{P})=\sum_{i=1}^{m}\left(\prod_{\ell<i}\alpha(c_\ell)\right)\left(\alpha(c_i)-\beta(c_i)\right)\left(\prod_{\ell>i}\beta(c_\ell)\right). \tag*{(212)} α ( P ) − β ( P ) = i = 1 ∑ m ( ℓ < i ∏ α ( c ℓ ) ) ( α ( c i ) − β ( c i ) ) ( ℓ > i ∏ β ( c ℓ ) ) . (212)
All products are in increasing index order, with empty products equal to 1 1 1 . For R ≥ R 0 = 4 κ + 8 R\geq R_0=4\kappa+8 R ≥ R 0 = 4 κ + 8 , every field has interval support in x [ R ] x[R] x [ R ] with r X < 2 r_X<2 r X < 2 . Since ∥ c i ∥ = ∥ f i ∥ 2 \lVert c_i\rVert=\lVert f_i\rVert_2 ∥ c i ∥ = ∥ f i ∥ 2 , the single-operator estimate (165) yields
∥ α ( P ~ ) − β ( P ~ ) ∥ ≤ ∑ i = 1 m ( ∏ ℓ ≠ i ∥ f ℓ ∥ 2 ) ∥ α ( c i ) − β ( c i ) ∥ ≤ C m ( ∏ i = 1 m ∥ f i ∥ 2 ) e − c R ( e c 1 E ∣ t ∣ − 1 ) . (213) \begin{aligned}
\lVert\alpha(\widetilde{P})-\beta(\widetilde{P})\rVert
\leq\sum_{i=1}^{m}\left(\prod_{\ell\neq i}\lVert f_\ell\rVert_2\right)\lVert\alpha(c_i)-\beta(c_i)\rVert\\
&\leq C_m\left(\prod_{i=1}^{m}\lVert f_i\rVert_2\right)e^{-cR}\left(e^{c_1E|t|}-1\right). \tag*{(213)}
\end{aligned} ∥ α ( P ) − β ( P )∥ ≤ i = 1 ∑ m ℓ = i ∏ ∥ f ℓ ∥ 2 ∥ α ( c i ) − β ( c i )∥ ≤ C m ( i = 1 ∏ m ∥ f i ∥ 2 ) e − c R ( e c 1 E ∣ t ∣ − 1 ) . (213)
Summing with the coefficients in the chosen expression for A A A therefore gives
∥ H ^ t R , E ( A ~ ) − τ t H R , E ( A ~ ) ∥ ≤ C A e − c R ( e c 1 E ∣ t ∣ − 1 ) . (214) \left\|\widehat{\mathcal{H}}_{t}^{R,E}(\widetilde{A})-\tau_{t}^{H_{R,E}}(\widetilde{A})\right\|\leq C_{A}e^{-cR}\left(e^{c_{1}E|t|}-1\right). \tag*{(214)} H t R , E ( A ) − τ t H R , E ( A ) ≤ C A e − c R ( e c 1 E ∣ t ∣ − 1 ) . (214)
Combining the two comparisons and using (6.10) gives
∥ ( L t H ( A ) − τ t H R , E ( A ~ ) L ) ρ 1 / 2 ∥ 2 ≤ C A [ R ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ + e − c R ( e c 1 E ∣ t ∣ − 1 ) ] ( 1 + max j Tr ρ A s , j ) 1 / 2 . (215) \left\|\left(\mathcal{L}_{t}^{H}(A)-\tau_{t}^{H_{R,E}}(\widetilde{A})\mathcal{L}\right)\rho^{1/2}\right\|_{2}
\leq C_{A}\left[R|t|E^{-(k+1)}e^{\gamma|t|}+e^{-cR}\left(e^{c_{1}E|t|}-1\right)\right]\left(1+\max_{j}\operatorname{Tr}\rho_{A_{s,j}}\right)^{1/2}. \tag*{(215)} ( L t H ( A ) − τ t H R , E ( A ) L ) ρ 1/2 2 ≤ C A [ R ∣ t ∣ E − ( k + 1 ) e γ ∣ t ∣ + e − c R ( e c 1 E ∣ t ∣ − 1 ) ] ( 1 + j max Tr ρ A s , j ) 1/2 . (215)
Choose 0 < ε ≤ 1 0<\varepsilon\leq1 0 < ε ≤ 1 with c 1 ε < c / 2 c_{1}\varepsilon<c/2 c 1 ε < c /2 . For R ≥ max { R 0 , ( 1 + ∣ t ∣ ) / ε } R\geq\max\{R_{0},(1+|t|)/\varepsilon\} R ≥ max { R 0 , ( 1 + ∣ t ∣ ) / ε } , set E = ε R / ( 1 + ∣ t ∣ ) E=\varepsilon R/(1+|t|) E = εR / ( 1 + ∣ t ∣ ) and A R ( t ) = τ t H R , E ( A ~ ) A_{R}(t)=\tau_{t}^{H_{R,E}}(\widetilde{A}) A R ( t ) = τ t H R , E ( A ) . The cutoff calculation in the proof of Proposition 2.8 bounds the two terms in brackets by C k ( 1 + R ) − k e ( γ + k + 2 ) ∣ t ∣ C_{k}(1+R)^{-k}e^{(\gamma+k+2)|t|} C k ( 1 + R ) − k e ( γ + k + 2 ) ∣ t ∣ . Since H R , E H_{R,E} H R , E acts only in x [ R ] x[R] x [ R ] and is even, A R ( t ) A_{R}(t) A R ( t ) has the required support and parity; unitary conjugation and (B.1) give ∥ A R ( t ) ∥ = ∥ A ∥ \|A_{R}(t)\|=\|A\| ∥ A R ( t ) ∥ = ∥ A ∥ .
For the remaining radii, set A R ( t ) = 0 A_{R}(t)=0 A R ( t ) = 0 . The error is at most ∥ A ∥ \|A\| ∥ A ∥ , and 1 + R ≤ ( 1 + R 0 + ε − 1 ) ( 1 + ∣ t ∣ ) 1+R\leq(1+R_{0}+\varepsilon^{-1})(1+|t|) 1 + R ≤ ( 1 + R 0 + ε − 1 ) ( 1 + ∣ t ∣ ) , so ( 1 + ∣ t ∣ ) k ≤ e k ∣ t ∣ (1+|t|)^{k}\leq e^{k|t|} ( 1 + ∣ t ∣ ) k ≤ e k ∣ t ∣ gives the same bound after enlarging C A C_{A} C A .
Acknowledgments
This is supported by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the grants TRR 352 – Project-ID 470903074 and FOR 5413 – Project-ID 465199066, as well as by the European Union through ERC Starting Grant MathQuantProp, Grant Agreement 101163620.3
AI declaration
The authors acknowledge help from ChatGPT (6 Astra) in finding and checking mathematical arguments and for assistance with parts of the writing. The authors take full responsibility for the complete manuscript.