RADIATIVE CORRECTIONS TO THE DYNAMICS OF A TRACER

PARTICLE COUPLED TO A BOSE SCALAR FIELD

ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Abstract. We consider a tracer particle coupled to a Bose scalar field and study the
arXiv:2405.05251v1 [math-ph] 8 May 2024

regime where the field’s propagation speed approaches infinity. For initial states devoid
of field excitations, we introduce an effective approximation of the time-evolved wave
function and prove its validity in Hilbert space norm. In this approximation, the field
remains in the vacuum state while the tracer particle propagates with a modified dispersion
relation. Physically, the new dispersion relation can be understood as the effect of radiative
corrections due to interactions with virtual bosons. Mathematically, it is defined as the
solution of a self-consistent equation, whose form depends on the relevant time scale.

Contents
1. Introduction 1
2. Main results 4
3. The integration-by-parts formula 12
4. Effective dynamics for massless fields 15
5. Analysis of the generator 18
6. Effective dynamics for massive fields 24
7. Resolvent estimates 30
Appendix A. Effective dynamics for massless fields: Revisited 34
References 37

1. Introduction
The study of particles interacting with quantum fields represents a universal challenge
across various physics disciplines, ranging from condensed matter systems to fundamen-
tal theories like QED. Understanding the behavior of such systems is often approached
through the lens of effective properties of independent particles, including effective disper-
sion relations, mediated interactions and finite lifetimes.
Such emergent properties can be categorized into two classes based on their association
with real or virtual excitations of the field. Real excitations, exemplified by phenomena
like the polaron effect, involve the tangible presence of field excitations surrounding the
particle. In the case of the polaron, the particle carries a cloud of real excitations, such
as phonons in a crystal lattice, resulting in a reduction of its mobility. Conversely, virtual
excitations, as observed in phenomena like the Lamb shift within QED, involve transient
1
2 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

fluctuations in the field without the presence of physical excitations. In the Lamb shift,
the creation and absorption of virtual photons induces subtle but measurable shifts in the
energy levels of an atom. In the physics literature, effects related to virtual excitations are
often referred to as radiative corrections.
In this work, we consider a Nelson type quantum field theory in three dimensions, where a
tracer particle (impurity, electron, etc.) is coupled to a Bose scalar field with large propaga-
tion speed. The purpose of our analysis is to understand the impact of radiative corrections
on the propagation of the particle. To this end, we study the dynamics of initial states with
zero field excitations and employ an effective approximation where the particle evolves in-
dependently but with a modified dispersion relation. The precise form of this effective
dispersion relation depends on the relevant time scale of interest. Our main result provides
a rigorous norm estimate for the difference between the original dynamics and the effective
evolution. We note that our approximation maintains unitarity, thereby precluding any
observed decay effects within the considered time scale.
Let us now turn to the the mathematical model that we consider. The Hilbert space of
the system consists of the tensor product
8 Â n
H :“ L2X b F F :“ C ‘
À
with L2 pR3 q . (1.1)
n“1 sym

Here, L2X ” L2 pR3 , dXq is the state space for the tracer particle, and F denotes the bosonic
Fock space. As usual, we equip F with standard creation and annihilation operators that
satisfy the Canonical Commutation Relations (CCR) in momentum space
rbk , b˚ℓ s “ δpk ´ ℓq and rbk , bℓ s “ rb˚k , b˚ℓ s “ 0 (1.2)
for any k, ℓ P R , and we denote by Ω “ p1, 0q P F the vacuum vector. On the Hilbert
3

space H we consider the dynamics that is generated by the following Hamiltonian

ż ż
1{2
` ˘
H :“ ´∆X ` µ ˚
ωpkqbk bk dk ` µ V pkq eik¨X bk ` e´ik¨X b˚k dk (1.3)
R3 R3
and its dependence with the large parameter
µ " 1. (1.4)
Here, the first two terms correspond to the kinetic energy of the tracer particle and the
field energy, respectively, and the third term corresponds to their interaction. We refer
to ωpkq as the dispersion relation of the field, and V pkq as the interaction potential, also
called the form factor. Both are real-valued and rotationally symmetric.
Let us note that the described model is translation invariant, which will be important for
our analysis. Throughout this article, we use the following notation to denote the operators
of momentum ż
p :“ ´i∇X and P :“ k b˚k bk dk . (1.5)
R3
In particular, translation invariance implies that the total momentum is a conserved quan-
tity, i.e. rH, p ` P s “ 0. Unless confusion arises, we also use the letter p P R3 to denote the
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 3

´3{2
ş ´ip¨X
Fourier variable in momentum
ş representation φ̂ppq ” p2πq R3
e φpXqdX. Finally,
we denote by N :“ R3 b˚k bk dk the number operator associated to the boson field.
We study the dynamics of the wave function Ψptq “ e´itH Ψ0 for initial states of the form
Ψ0 “ φ b Ω, where φ P L2X is suitably localized in momentum space. For concreteness, we
assume that the dispersion is given by
?
ωpkq “ k 2 ` m2 , (1.6)
where m ě 0 represents a possible mass term. We say that the field is massless if m “ 0, and
say it is massive otherwise. As for the interaction, the main example of physical interest that
we keep in mind corresponds to the Nelson model with ultraviolet cut-off. More precisely,
1p|k| ď Λq
V pkq “ a (1.7)
ωpkq
for some positive parameter Λ.
To our knowledge, the described scaling for H with µ " 1 was introduced by Davies [3],
who interprets µ´1 H as a model for a heavy tracer particle weakly coupled to the Bose
field. Hence, this choice of scaling is also referred to as the weak-coupling limit. In [4, 5],
Hiroshima studies the same scaling for the Nelson model, but with the cutoff Λpµq Ñ 8
removed as µ Ñ 8. For a comparison of these works with our results, see Section 2.3. In
[19, Section 6.3], Teufel explains that H represents the canonical quantization of a classical
system for a particle coupled to a scalar field, where the propagation speed of the field
tends towards infinity. Thus, we think of the boson field as the fast subsystem relative to
the tracer particle. We are interested in understanding the effective dynamics that such
system may generate.
1.1. Description of main results. Let us informally describe our main results. These
are classified according to the absence or existence of a mass term. In order to keep the
following exposition simple, we assume here that the form factor is given by the Nelson
model (1.7).
Massless fields. In the present case, we prove in Theorem 2.2 that there is an effective
µ-dependent Hamiltonian hg ppq on L2X such that
Ψptq « e´ithg ppq φ b Ω as µÑ8 (1.8)
for all times |t| ! µ1{2 plog µq´1 . The approximation holds in the norm of H with explicit
error control. The effective Hamiltonian can be written as a perturbation of the free kinetic
energy hg ppq “ p2 ´ gppq, and is defined in terms of a function gppq that solves the following
nonlinear equation
µ |V pkq|2 dk
ż
gppq “ `
2 2
˘ . (1.9)
R3 µ ωpkq ` pp ´ kq ´ p ` gppq

As we will see, gppq is an Op1q correction, caused by the emission and absorption of virtual
bosons.
4 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Massive fields. In this situation, the previous approximation can be extended to longer
time scales by introducing higher-order radiative corrections. For concreteness, assume
m ě 1 for the moment. We prove in Theorem 2.4 that for all n P N odd there is an effective
Hamiltonian hgn ppq such that
Ψptq « e´ithgn ppq φ b Ω as µÑ8 (1.10)
for all times |t| ! µpn`1q{2 plogpµqq´pn`2q{2 . The approximation also holds in the norm of H
with an explicit error. The Hamiltonian is defined analogously hgn ppq “ p2 ´ gn ppq, but
the nonlinear expression that defines gn ppq is more involved since it contains higher-order
corrections. See Definition 2.3.
Let us stress that the addition of higher-order corrections does not reduce the distance
between the original and the effective dynamics, which is always at least of order Opµ´1{2 q.
Rather, it has the effect of extending the time scale for which the approximation is valid.
To the best of our knowledge, the consideration of gppq as an effective dynamics to Nelson
type models of the form (1.3) has not been studied before. We include a brief discussion
of related results for similar problems in Subsection 2.3.
Organization of the article. In Section 2 we state our main results, Theorems 2.2
and 2.4. In Section 3 we introduce a suitable integration-by-parts formula that allows the
comparison of the dynamics. In Section 4 we prove Theorem 2.2 for massless fields, and
in Section 5 we analyze the effective generator in the massless case. In Section 6 we prove
Theorem 2.4 for massive fields, and in Section 7 we establish some necessary resolvent
estimates.

2. Main results
In this section, we give the statements of our main results. These correspond to Theorem
2.2 and 2.4. First, let us give the precise conditions that we consider on the initial datum
φ and the form factor V .
Condition 1. The initial datum is a tensor product Ψ0 “ φ b Ω with }φ}L2 “ 1 and Ω the
Fock space vacuum. We assume that there exists 0 ă ε ă 21 such that
φ “ 1p|p| ď εµqφ . (2.1)
Condition 2. The form factor V : R3 Ñ R is rotationally symmetric, non-zero, and we
1
assume that both }V }L2 and }|k|´ 2 V }L2 are finite.
Convention. We say C ą 0 is a constant if it is a positive number, depending only on ε,
1
}V }L2 and }|k|´ 2 V }L2 . Throughout proofs, its value may change from line to line.
Under these conditions, the Hamiltonian H is self-adjoint in its natural domain. Indeed,
let us denote here and in the rest of the article
ż ż
1{2
` ˘
T “µ ˚
ωpkqbk bk dk and V“µ V pkq eik¨X bk ` e´ik¨X b˚k dk . (2.2)
R3 R3
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 5

Then, it is straightforward to verify that V is relatively bounded with respect to T , with

relative bound less than one. Thus, the Kato-Rellich theorem implies that H is self-adjoint
on Dpp2 q X DpT q. Consequently, the original dynamics
Ψptq “ e´itH Ψ0 with Ψ0 “ φ b Ω (2.3)
is well-defined for all t P R.
On the other hand, given a bounded measurable function g : R3 Ñ R let us define the
effective Hamiltonian as
hg ppq :“ p2 ´ gppq (2.4)
with p “ ´i∇X . We refer to hg ppq as the effective dispersion relation or the effective
generator. Thanks to the boundedness of g, the effective generator is again self-adjoint on
Dpp2 q. Thus, the effective dynamics
Ψeff ptq :“ e´ithg ppq φ b Ω (2.5)
is well-defined for all t P R. As we have already explained in the introductory section,
the function gppq is determined by a nonlinear self-consistent equation. It depends on the
choice of ω and V and, in particular, also on the time scale that one is interested in. The
available time scales depend on the absence or presence of a mass term in the dispersion
relation of the field.
Remark 2.1. Let us explain the heuristics behind Condition 1: For an excitation of mo-
mentum k P R3 to be emitted from an electron with momentum p P R3 , conservation of
kinetic energy reads
µωpkq “ p2 ´ pp ´ kq2 . (2.6)
The collection pp, kq P R6 that satisfy (2.6) are sometimes referred to as resonant sets. In
particular, under Condition 1 these sets are excluded. Thus, we can interpret the assump-
tion |p| ď εµ as an energetic constraint. Namely, that the electron lacks sufficient energy
to create real excitations. On the level of the effective dynamics, this is manifest in the
fact that N Ψeff ptq “ 0. Our analysis will show that real excitations are in fact suppressed
in the original time evolution for large µ, in the sense that Ψptq « Ψeff ptq. On the other
hand, thanks to the Heisenberg uncertainty principle, energy conservation can be violated
by the fluctuations of the field. These are called virtual excitations, and they give rise to
non-trivial modifications of the dispersion of the electron. These effects are described by
the function gppq.
2.1. Massless fields. Throughout this subsection, we assume that the boson field is mass-
less. Namely, that
ωpkq “ |k| (2.7)
for all k P R3 . The main result in the present case is the estimate contained in Theorem
2.2, stated below.
The precise definition of the effective Hamiltonian is as follows.
6 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Definition 2.1. We define the function g : R3 Ñ p0, 8q as the solution of the equation
µ |V pkq|2 dk
ż
gppq “ `
2 2
˘ (2.8)
R3 µ ωpkq ` pp ´ kq ´ p ` gppq

for |p| ă 21 µ and g ” 0 otherwise.

Let us argue that the solution to (2.8) exists, is unique and of class C 1 on t|p| ă 21 µu.
Sketch of proof. Fix 0 ă ε0 ă 1{2. Then, for |p| ď ε0 µ we define the following auxiliary
function
ş ` ˘´1
F pp, xq :“ µ R3 |V pkq|2 µ ωpkq ´ 2p ¨ k ` k 2 ` x dk , xą0. (2.9)
First, observe that µωpkq ´ 2p ¨ k ` k 2 ` x ě p1 ´ 2ε0 qµ|k|. Hence, F pp, xq ď C for a
constant C ą 0 and so F is well-defined. Next, note F pp, 0q ą 0, x ÞÑ F pp, xq is C 1 , and
limxÑ8 F pp, xq “ 0. Thus, the intermediate value theorem implies there exists x P p0, 8q
such that x ´ F pp, xq “ 0. Further, note Bx F ă 0. Thus, x ÞÑ F pp, xq is strictly decreasing
and the fixed point is unique. We then set gppq “ x. Finally, since x ÞÑ F pp, xq is C 1 on
|p| ă ε0 µ the implicit function theorem implies that p ÞÑ gppq is C 1 on t|p| ă ε0 µu. The
claim now follows from taking ε0 arbitrarily close to 1{2. □

To state our main results, we need to introduce the following µ-dependent norm
› ›
›V ›
|||V |||µ :“ ›› ›› with ωµ pkq :“ ωpkq ` µ´1 . (2.10)
ωµ 2
L

As we shall see below, the µ-norm (2.10) appears naturally thanks to the presence of the
generator gppq in certain denominators. It can be understood as an infrared regularization
of the norm }ω ´1 V }L2 . In particular, while for the massless Nelson model (1.7) the latter
norm is infinite, the µ-norm is finite and grows logarithmically with µ:
ż ˆ ˙
2 dk ` ˘ µΛ
|||V |||µ “ ´1 2
“ 4π log µΛ ` 1 ´ . (2.11)
|k|ďΛ |k|p|k| ` µ q µΛ ` 1

We are now ready to state our first main result.

Theorem 2.2. Let ωpkq “ |k| and assume that Ψ0 and V satisfy Condition 1 and 2,
respectively. Then, there exists a constant C ą 0 such that
|||V |||2µ
ˆ ˙
|||V |||µ
}Ψptq ´ Ψeff ptq}H ď C ` |t| 1{2 (2.12)
µ1{2 µ
for all t P R and all µ ą 0 large enough.
Remark 2.2. A few remarks are in order.
(1) The relevant parameters in (2.12) are µ and t. Since Ψptq and Ψeff ptq are both normal-
ized, we are interested in the parameter regime, for which the upper bound is small
compared to one.
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 7

(2) For the the massless Nelson model, the logarithmic growth (2.11) and Theorem 2.2
imply that we have an approximation for µ " 1 and
µ1{2
|t| ! .
log µ
1
We show in Appendix A how to extend this time scale by a factor plog µq 2 .
(3) It is possible to introduce a scale of less singular of form factors: Va pkq “ |k|´a 1p|k| ď Λq
where a P r0, 1{2q. For these models, the norm |||Va |||µ is uniformly bounded in µ ą 0,
hence Theorem 2.2 provides an approximation for |t| ! µ1{2 . In Appendix A we show
how to extend this approximation to times |t| ! µ1´a .
2.1.1. Polynomial generators. In Section 5 we analyze the generator gppq in more detail.
As a first step, we show that for all 0 ď ε ă 1{2 there are constants C1 and C2 such that
C1 ď gppq ď C2 @|p| ď εµ . (2.13)
We then refine the analysis, and show that gppq can be approximately solved in terms of
|p| and the value g0 :“ gp0q. Namely, we consider the function
ˆ ˙
|V pkq|2
ż
µ ´1 2|p||k|
geff ppq :“ tanh dk (2.14)
2|p| R3 |k| µ|k| ` k 2 ` g0
1
and prove in Proposition 5.1 that for every 0 ď ε ă 2
there is a constant C ą 0 such that
C|||V |||2µ
|gppq ´ geff ppq| ď @|p| ď εµ . (2.15)
µ
Thus, we can replace gppq with geff ppq in Theorem 2.2 and keep the same error estimate.
See Corollary 2.1 below.
While geff ppq is explicit, it can still be regarded a complicated function of p. To obtain
a simpler form for the effective dispersion, we use the series expansion of the hyperbolic
tangent function. This motivates the definition of the following polynomial generators of
degree N P 2N0
ÿ
2
hpN
g
q
:“ p ´ αj pµq|p|j (2.16)
jP2N0 :jďN

where αj pµq are positive coefficients, defined explicitly in (5.9). For completeness, we also
set hp8q
g :“ p2 ´ geff ppq. For example, the generator hp2q 2
g “ p1 ´ α2 qp ´ α0 , describes a free
particle with enhanced mass. For N ě 4, the generators encompass non-trivial corrections
to the parabolic shape of the free dispersion relation.
The next corollary shows that for suitable initial conditions, the effective generators
hpN q
g ppq can still be used to approximate the wave function Ψptq “ e
´itH
φ b Ω.
8 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Corollary 2.1. Assume Conditions 1 and 2 and additionally φ P 1p|p| ď P0 qL2X . Then
there is a constant C ą 0 such that
ˆ ˙N `2 ˙
|||V |||2µ
ˆ
pN q
´ithg |||V |||µ P0
}Ψptq ´ e φ b Ω}H ď C ` |t| ` |t| for N P 2N0 , (2.17)
µ1{2 µ1{2 µ
|||V |||2µ
ˆ ˙
p8q
´ithg |||V |||µ
}Ψptq ´ e φ b Ω}H ď C ` |t| 1{2 for N “ 8 (2.18)
µ1{2 µ
for all t P R, P0 ą 0 and all µ ą 0 large enough.
Remark 2.3. Using the truncated version of the effective dynamics leads to the additional
error term |t|pP0 {µqN `2 . We now discuss its consequences for the Nelson model (1.7).
(1) Clearly, the new error term imposes additional constraints on the validity of the ap-
proximation, when compared to Theorem 2.2 or (2.18). Indeed, while the original error
is uniform in the momentum scale P0 , the new error is not. Observe for instance that
at scales P0 „ µ the right hand side of (2.17) is small only for |t| ! 1, as opposed to
|t| ! µ1{2 plog µq´1 .
(2) In Section 5.3 we compare in detail the quality of the different approximations, and
argue that the new error in (2.17) is in fact optimal. For illustration, let N “ 2 and
consider initial data satisfying
φ P 1p 21 P0 ď |p| ď P0 qL2X with P0 “ µ23{24 . (2.19)
Then, the additional error is small only for |t| ! µ1{6 . In Theorem 5.1 we strengthen
this observation by showing that one cannot consider longer time scales and still expect
the approximation to be valid. Namely, we prove that there is a constant τ ą 0 such
that convergence fails for t “ τ µ1{6 :
p2q
lim inf }Ψptq ´ e´ithg ppq
φ b Ω}H ą 0 . (2.20)
µÑ8

On the other hand, Corollary 2.1 implies that the approximation remains effective on
this time scale for N ě 4. This highlights the relevance of the higher-order corrections
in hpN q
g ppq.

2.2. Massive fields. Let us now state our main result regarding massive boson fields with
dispersion relation ?
ωpkq “ k 2 ` m2 (2.21)
for some mass m ą 0. In contrast to the massless case, here one is able to systematically
extend the time scale of validity of the effective approximation. However, one needs to
modify the generator gppq to include higher-order terms.
For the precise definition, we need the following notation for a collection of sequences σ
of length j P N:
Σ0 pjq :“ tσ P t`1, ´1uj : ℓi“1 σpiq ě 1 @ℓ ď j ´ 1 and
ř řj
i“1 σpiq “ 0u . (2.22)
See Figure 1 for a visual representation of an element σ P Σ0 p8q. We will also denote
# #
bk `1 ” b˚k and bk ´1 ” bk .
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 9

P`
3  s(`) = i=1 σ(i)
2   
1   
0 
1 2 3 4 5 6 7 8 `

Figure 1. The partial sums of σ “ p1, 1, 1, ´1, ´1, 1, ´1, ´1q P Σ0 p8q

Definition 2.3. Let n P N. Then, we define the function gn : R3 Ñ p0, 8q as the solution
of the equation
n`1 ż
# #
ÿ ÿ
j{2
gn ppq “ µ dk1 ¨ ¨ ¨ dkj V pk1 q ¨ ¨ ¨ V pkj q xΩ, bkjσpjq . . . bk1σp1q Ωy
j“2 σPΣ0 pjq R3j
j even
j´1 ˆ ÿ
ź ℓ ´ ℓ
ÿ ¯2 ˙´1
2
ˆ µ σpiqωpki q ` p ´ σpiqki ´ p ` gn ppq (2.23)
ℓ“1 i“1 i“1

for |p| ă 21 µ and gn ” 0 otherwise.

Remark 2.4. The generator gn ppq can be regarded as the solution of the fixed point equation
1
gn ppq “ Fn pp, gn ppqq , |p| ă µ (2.24)
2
where Fn pp, xq is determined from the right hand side of (2.23) (see e.g. (6.7)). In par-
ticular, for each σ P Σ0 pjq, the bosonic expectation values can be evaluated in terms of
δ-distributions via Wick’s theorem. After carrying out the contractions, the summability
condition of Σ0 pjq implies that every denominator in (2.23) is manifestly positive and de-
creases with x. Thus, the proof we sketched for the n “ 1 case can be adapted to n ě 2 to
show that the solution to (2.24) exists, is unique and of class C 1 . We leave the details to
the reader.
We now consider the effective dynamics
pnq
Ψeff ptq :“ e´ithgn ppq φ b Ω (2.25)
for all t P R with hgn ppq “ p2 ´ gn ppq.
In the next result, we keep track of the role of the mass term m ą 0 in the approximation.
We are interested in the case m " µ´1 since the threshold m “ µ´1 turns out to be critical
in an appropriate sense. Recall (2.10) for the definition of |||V |||µ .
?
Theorem 2.4. Let ωpkq “ k 2 ` m2 and assume that Ψ0 and V satisfy Condition 1 and
2, respectively. Then, there exists a constant C ą 0 such that for every n P N odd
|||V |||nµ |t| |||V |||n`1
ˆ " *˙
pnq n
? |||V |||µ µ |||V |||µ
}Ψptq ´ Ψeff ptq}H ď C n! 1 ` 1 n´1 ` 1 n´1 min 1, 1
µ2 µ 2 pµmq 2 µ 2 pµmq 2 µ2 m
10 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

for all t P R, m ě µ´1 and all µ ą 0 large enough.

Remark 2.5 (Mass criticality). In Theorem 2.4 the mass value m “ µ´1 is critical. Namely,
the dependence of our estimates with µ ą 0 does not improve when invoking higher itera-
tions, relative to the massless case. That is, the n “ 1 iteration from Theorem 2.2.
Remark 2.6. Additional remarks are in order. For concreteness, we assume that V is given
by (1.7) so that |||V |||2µ “ Oplogpµqq.
(1) Consider n “ 1. Then, for µ´1 ď m ď plog µq1{2 µ´1{2 the statement coincides with the
massless case treated in Theorem 2.2. For m " plog µq1{2 µ´1{2 , the bound improves
compared to Theorem 2.2, as the approximation now holds for all
mµ
|t| ! .
plog µq3{2
(2) For n ě 3 and above the critical mass threshold, the approximation extends to sub-
stantially longer times. For instance, if m ě 1, convergence holds for all
n`1
µ 2
|t| ! n`2 .
plog µq 2

(3) The reason for considering only n P N odd is that gn “ gn´1 for all n ě 2 even. This
follows from the simple fact that xΩ, b# #
kj . . . bk1 Ωy “ 0 for all j P N odd.
(4) Similarly as in the massless case, it is in principle possible to introduce polynomial
generators hpN q
gn that result from the truncation of an appropriateř power series. The
latter is obtained via expanding the denominators in (2.23) around j µωpkj q. Thanks
to the mass term, this expansion is in fact less singular that in the massless situation.
Since this procedure is relatively straightforward, we omit the details.
2.3. Comparison with previous results. Let us compare our analysis with some related
results from the literature.
(1) Davies [3] considers heavy tracerřparticles weakly coupled to a scalar boson field.
paq a řa
He utilizes the Hamiltonian H “ i“1 p´∆Xi q ` T ` i“1 VXi on the Hilbert space
ba L2 pR3 q b F , with T and VXi as defined in (2.2). For a “ 1, this Hamiltonian coincides
with our choice of H. He establishes [3, Theorem 2.2] that for t P R and all φpaq P ba L2 pR3 q,
paq paq
lim e´itH φpaq b Ω “ e´ith φpaq b Ω, (2.26)
µÑ8

where hpaq “ ai“1 p´∆Xi q`K paq is an operator on ba L2 pR3 q, and K paq describes an effective
ř
pair potential between the tracer particles, given by
paq
ÿ ż |V pkq|2
K “ cospkpXi ´ Xj qq dk. (2.27)
1ďi,jďa
ωpkq
For a single tracer particle, Davies’ result is reproduced with an explicit rate of convergence
by Corollary 2.1, when applied for n “ 0. To see this, note that limµÑ8 α0 pµq Ñ K p1q .
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 11

Our analysis can be readily extended to reproduce Davies’ result for a ě 2 as well.
However, our primary focus lies in extending the approximation to longer time scales,
particularly those requiring non-trivial modifications of the dispersion relation. For clarity
and transparency, we limit our investigation to the case where a “ 1, postponing the
analysis of more than one tracer particle for future study. In such cases, we would expect
momentum-dependent corrections to the effective pair potential K paq .
(2) In [4, 5] Hiroshima studies the same model as Davies, but with a µ-dependent UV
cutoff in (1.7). He establishes (2.26) in the simultaneous limit Λpµq Ñ 8 as µ Ñ 8, which
requires the subtraction of a diverging energy. In this limit, the operator K paq in (2.27)
describes a Coulomb pm “ 0q resp. Yukawa (m ą 0) pair potential.
In the present work, we opt to maintain a fixed UV cutoff in (1.7). However, we expect
that our results can be extrapolated to this coupled limit as well. In particular, one can
verify that H ´gp0q, gppq´gp0q, and consequently pe´itH ´e´ithg qΨ0 , have appropriate limits
as Λ Ñ 8. Nevertheless, as our error bounds depend on }V }L2 „ Λ and }|k|´1{2 V } „ Λ1{2 ,
this observation alone is not sufficient for a direct extrapolation. We expect that refinement
error bounds could be attained through the use of Gross’ unitary dressing transformation.
However, to keep the focus on the conceptual ideas, we refrain from introducing additional
technical details. Finally, it is worth noting that we are unaware of any work that establishes
(2.26) for the renormalized Nelson model.
(3) The works by Davies and Hiroshima have been revisited and extended to other particle-
field models as well as different scalings. See for instance [1, 15, 16, 17, 6, 7].
(4) Teufel [19, 20] and Tenuta and Teufel [18] consider the Neslon model in a slightly
different scaling, namely they study the dynamics for large µ generated by
ż ż
´∆X ` ˘
H “ ` ωpkqb˚k bk dk ` V pkq eik¨X bk ` e´ik¨X b˚k dk ; (2.28)
µ R3 R3

see e.g. [18, Eq. (10)] for ε “ µ´1{2 . As opposed to our model, the operator µH has an
additional factor µ1{2 in front of the third term, which makes the interaction comparable
to the field energy T . For a heuristic comparison of the two scalings, see [19, Section 6.3].
Conceptually, the scaling in (2.28) is similar to the one observed in the Born–Oppenheimer
theory for electrons coupled to heavy nuclei. In fact, both can be understood through the
general framework of adiabatic perturbation theory, and we refer to the book of Teufel
[19] for an in-depth exposition of the subject. In this framework, Teufel [20] and Tenuta
and Teufel [18] obtained a norm approximation for the dynamics generated by (2.28);
see e.g. [19, Theorem 6.10]. More concretely, the authors considerş‘ small electron velocity,
and the wave function approximation consists of wave packets R3 φpt, XqΩpXqdX,ş where
ΩpXq is the ground state of the X-dependent Fock space operator HpXq “ ωpkqb˚k bk dk `
apeikX V q ` a˚ peikX V q, and φpt, Xq is driven by an effective Hamiltonian. Depending on
the time scale, the effective Hamiltonian contains an effective pair interaction (including
(2.27), but also the momentum-dependent Darwin term) and corrections to the mass of
12 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

the particle(s). Contrary to our results, the effective states are not vacuous, but rather
represent dressed electron states.
(5) The emergence of an effective dispersion relation for a tracer particle coupled to a quan-
tum field has also been explored in other contexts. Specifically, in [2], Bach, Chen, Faupin,
Fröhlich and Sigal, and previously in [14], Spohn and Teufel investigated the dynamics of
an electron in a slowly varying external potential within the framework of non-relativistic
QED. Their results show that the dynamics of such systems can be described in terms of
dressed electron states whose evolution is governed by an effective dispersion relation Eppq.
This dispersion is characterized, as usual for translation invariant particle-field models, as
the infimum of the energy at fixed total momentum p P R3 . Notably, while Eppq plays a
similar role as our hg ppq, our description of the generator is more explicit. In particular, we
describe the effective dispersion as an explicit function of |p| (up to small errors; see e.g.
(2.14)) and do not refer to the fiber decomposition of the Hamiltonian.

3. The integration-by-parts formula

The main goal of this section is to introduce an integration-by-parts formula that will
be the heart of our analysis. The formula provides an expansion scheme for the difference
between the original and the effective dynamics, and will be used in the proofs of Theorem
2.2 and 2.4.
The heuristics guiding our expansion are as follows: Upon invoking Duhamel’s formula
(see Eq. (3.2)), the full propagator eisH exhibits rapid oscillations as µ Ñ 8, when acting
2
on states
ş in LX b pSpanpΩqqK . This behavior arises from the presence of the large operator
T “ µ dkωpkqb˚k bk and effectively, it suppresses all contributions in this subspace. On the
other hand, contributions in the orthogonal complement L2X b SpanpΩq, do not manifest
these rapid oscillations. Later, we will demonstrate how these terms add up to the effective
generator gppq.
Throughout this section, we denote Ψeff ptq “ e´itHeff Ψ0 where Ψ0 “ φ b Ω P H with
φ P Dpp2 q. The effective Hamiltonian is
Heff “ hg ppq b 1 (3.1)
where the generator gppq is not fixed but arbitrary, and for notational convenience we drop
the subscript, i.e. we write h ” hg “ p2 ´ gppq.
Let us start with the following calculation for the difference between the original and the
effective dynamics. Namely
´ ¯
Ψptq ´ Ψeff ptq “ e´itH 1 ´ eitH e´itHeff Ψ0
żt
´itH 1
´ ¯
“e eisH H ´ Heff e´isHeff Ψ0 ds
i 0
żt
´itH 1
´ ¯
“e eisH V ` gppq e´isHeff Ψ0 ds (3.2)
i 0
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 13

where we used T Ω “ 0. Let us now note that thanks to P Ω “ 0 we have Heff Ψ0 “

hpp ` P qΨ0 . Hence, since rV, p ` P s “ rgppq, p ` P s “ 0, we obtain the identity
żt
´itH 1
´ ¯
isH ´ishpp`P q
Ψptq ´ Ψeff ptq “ e e e ds V ` gppq Ψ0 . (3.3)
i 0
The last expression motivates the following definition.
Definition 3.1. For all t P R we define the operator
1 t isH ´ishpp`P q
ż
Iptq :“ e e ds . (3.4)
i 0
Remark 3.1 (Dynamics difference). Clearly, for all t P R, we can now write
´ ¯
Ψptq ´ Ψeff ptq “ e´itH Iptq V ` gppq Ψ0 . (3.5)

Note that the first term VΨ0 is a state that contains one boson. We are now interested
in studying the action of the operator Iptq on states that are orthogonal to the vacuum,
as they lead to rapidly oscillating phases. To this end, we introduce
PΩ “ 1 b |Ωy xΩ| and QΩ “ 1 ´ P Ω . (3.6)
It will be convenient to define the following two auxiliary operators. The first one we call
the resolvent. The second one we call the boundary term.
Definition 3.2 (Auxiliary operators). We define as operators on H “ L2X b F
(1) The resolvent
´ ¯´1
2
R :“ QΩ hpp ` P q ´ p ´ T QΩ (3.7)
(2) The boundary term, for all t P R
´ ¯
itH ´ithpp`P q
Bptq :“ e e ´1 . (3.8)

Remark 3.2 (Domain of R). The resolvent operator R is an unbounded operator on H .

Here, we define its domain as DpRq :“ tΨ P H : s– limϵÓ0 RpϵqΨ exists in H u, where
´ ¯´1
2
Rpϵq :“ QΩ hpp ` P q ´ p ´ T ´ iϵ QΩ , ϵ ą 0. (3.9)
We then define the strong limit RΨ :“ limεÓ0 RpϵqΨ. In practice, we shall always apply
R to states Ψ P H that are evidently in the domain DpRq–see the estimates contained
in Sections 4 and 7, respectively. In order to keep the exposition simple, we do not refer
anymore to the unbounded nature of R outside of this section.
Let us now relate Iptq, R and Bptq. The following proposition contains the core idea of
our proof.
Proposition 3.1 (Integration-by-parts formula). For all Ψ P DpRq X DpVRq, we have
´ ¯
IptqQΩ Ψ “ BptqR ` IptqV R Ψ . (3.10)
14 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Remark 3.3. Let us comment on the above proposition.

(1) In practice, we only apply the integration-by-parts formula to states Ψ P QΩ H . Thus,
unless confusion arises, we will often omit the projection QΩ .
(2) Let us note that Proposition 3.1 is partially inspired by previous works [9, 10, 13, 12,
11], where similar integration-by-parts formulas were used in different contexts. The
common feature of the considered models is that a slow system is coupled to a fast
system, and that the evolution of the slow system is governed by an effective generator.
However, we want to emphasize that our expansion is novel in that it yields an effective
generator that solves a nonlinear equation; see e.g. (2.8).

Proof. In what follows, we understand the operator identities to hold on QΩ H . The main
idea is to integrate by parts in a convenient way. To this end, we first compare with the
free kinetic energy

1 t isH ´ishpp`P q
ż
Iptq “ e e ds (3.11)
i 0
1 t isH ´ispp2 `T q ispp2 `T q ´ishpp`P q
ż
“ e e e e ds . (3.12)
i 0

Note that rhpp ` P q, p2 ` T s “ 0. Therefore, we can write

żt
1
` ˘
2 `T q hpp`P q´p2 ´T
Iptq “ eisH e´ispp e´is ds . (3.13)
i 0

Next, we use

d
` ˘ ` ˘
´is hpp`P q´p2 ´T hpp`P q´p2 ´T
e “ e´is iR . (3.14)
ds
and integrate by parts
˘ ˇˇt
1 isH ´ispp2 `T q ´is hpp`P q´p2 ´T
`
Iptq “ e e e iRˇˇ
i s“0
1 t isH 1 ´ 2
ż ¯ ` ˘
´ispp `T q ´is hpp`P q´p2 ´T
2
´ e p `T ´H e e iRds . (3.15)
i 0 i

The first term in (3.15) corresponds to the boundary term Bptq. Namely, using again
rhpp ` P q, p2 ` T s “ 0
ˇt
1 isH ´ispp2 `T q ´is
` ˘
hpp`P q´p2 ´T
ˇ
e e e iRˇˇ “ BptqR . (3.16)
i s“0
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 15

For the second term in (3.15) we observe that p2 ` T ´ H “ ´V. Using again rhpp ` P q, p2 `
T s “ 0 and rV, p ` P s “ 0 we find
1 t isH 1 ´ 2
ż ¯ ` ˘
´ispp2 `T q ´is hpp`P q´p2 ´T
´ e p `T ´H e e iRds
i 0 i
1 t isH 1 ´ispp2 `T q ´is hpp`P q´p2 ´T
ż ` ˘
“ e Ve e iRds
i 0 i
1 t isH ´ishpp`P q
ż
“ e Ve Rds
i 0
1 t isH ´ishpp`P q
ż
“ e e dsVR “ IptqVR . (3.17)
i 0
This finishes the proof of the proposition. □

4. Effective dynamics for massless fields

In this section, we apply the integration-by-parts formula
Iptq “ BptqR ` IptqV R (4.1)
provided by Proposition 3.1, in order to prove Theorem 2.2. In what follows, we always
assume that Ψ0 satisfies Condition 1 relative to some fixed ε, and that V satisfies Condition
2. The field is either massless or massive, but the results in this section are mostly relevant
to the massless case.
We choose the function gppq according to Definition 2.1 and unless confusion arises we
drop the following subscript h “ hg . In addition, we use the decomposition V “ V` ` V´
ż ż
1{2 1{2
`
V ”µ V pkqe´ik¨X ˚
bk dk and ´
V ”µ V pkqeik¨X bk dk (4.2)
R3 R3

in terms of creation and annihilation operators.

Let us recall that the difference Ψeff ptq´Ψptq was written in terms of V and gppq in Remark
3.1. In what follows, we integrate-by-parts the V term using (4.1) and VΨ0 P QΩ H , where
PΩ and QΩ “ 1 ´ PΩ were the projections introduced in (3.6). We find that
´ ¯ ´ ¯
eitH Ψptq ´ Ψeff ptq “ Iptq V ` gppq Ψ0
´ ¯
“ Iptq VRV ` gppq Ψ0 ` BptqRVΨ0
´ ¯
“ Iptq PΩ VRV ` gppq Ψ0 ` IptqQΩ VRVΨ0 ` BptqRVΨ0 . (4.3)

The next step is to realize that our choice of gppq forces the first term on the right side to
vanish. The other two terms then need to be estimated. Let us now calculate the projection
onto the vacuum.
16 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Lemma 4.1. The following identity holds

ż ´ ¯´1
PΩ VRVΨ0 “ ´µ |V pkq|2 µ ωpkq ` pp ´ kq2 ´ hppq dk b 1 Ψ0 . (4.4)

An immediate consequence of Lemma 4.1 is that thanks to the choice of gppq given by
Def. 2.1, it follows from (4.3) that
}Ψptq ´ Ψeff ptq}H ď }BptqRVΨ0 } ` }IptqQΩ VRVΨ0 } . (4.5)
We dedicate the rest of this section to the proof of Lemma 4.1 and to provide estimates
on the remainder terms given by the right hand side of (4.5). These estimates are given
in Proposition 4.1 and 4.2. We informally refer to these two terms as the boundary term,
and the projection term, respectively.
Proof of Lemma 4.1. We decompose V and consider only the combination of creation and
annihilation operators that give a non-zero contribution. Namely
ż ´ ˘
PΩ VRVΨ0 “ PΩ µV pk1 qV pk2 q eik2 X bk2 Re´ik1 X b˚k1 Ψ0 dk1 dk2 . (4.6)

Next, we compute using standard commutation relations that

´ ¯´1
Re´ikX b˚k “ e´ikX b˚k hpp ` P q ´ pp ´ kq2 ´ T ´ µ ωpkq . (4.7)
In particular, if we evaluate the above operator on Ψ0 “ φ b Ω we find
´ ¯´1
Re´ikX b˚k Ψ0 “ e´ikX b˚k hppq ´ pp ´ kq2 ´ µ ωpkq Ψ0 , (4.8)

in view of T Ω “ P Ω “ 0. The CCR now imply that for all k1 , k2 P R3

PΩ bk2 b˚k1 “ δpk1 ` k2 qPΩ (4.9)
since PΩ b˚k “ 0. The proof is finished once we put together (4.6), (4.8) and (4.9). □
Before we turn to the analysis of the remainder estimates, let us record here the proof of
Theorem 2.2.
Proof of Theorem 2.2. Starting from the estimate (4.5) we use the results from Proposition
4.1 to estimate the boundary term, and the result from Proposition 4.2 to estimate the
projection term. □
4.1. Error analysis. Recall that the initial datum Ψ0 “ φ b Ω satisfies Condition 1 with
respect to ε P p0, 1{2q, and that ωpkq ě |k|. We also remind the reader of the notation
ωµ pkq “ ωpkq ` µ´1 .
Proposition 4.1 (Boundary term). There exists a constant C ą 0 such that
C
}BptqRVΨ0 }H ď }ω ´1 V }L2 (4.10)
µ1{2 µ
for all t P R and µ ą 0 large enough.
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 17

Remark 4.1. We record here the usual estimates for creation- and annihilation operators,
extended to H “ L2X b F . The following is enough for our purposes: for n P N and
f P L2 pR3n ; L2X q we have that
›ż ›
› ˚ ˚
› ?
› 3n f pk1 , ¨ ¨ ¨ , kn q b bkn ¨ ¨ ¨ bk1 Ω dk1 ¨ ¨ ¨ dkn › ď n!}f }L2 pL2X q . (4.11)
› ›
R H

Remark 4.2. In Section 5 we describe in more detail the generator gppq. In the present
section, we will only need the following bounds, valid for µ large enough:
gppq ą 0 and C1 ď |gppq| ď C2 (4.12)
for |p| ď εµ and for some constants C1 and C2 . See Lemma 5.1 for more details.
Proof of Proposition 4.1. First, we note that }BptqRVΨ0 } ď 2}RV` Ψ0 }. Further, we use the
decomposition (4.2) to write VΨ0 “ V` Ψ0 . Then, thanks to (4.8) we find
ż ´ ¯´1
1{2
`
RV Ψ0 “ µ V pkqe´ikX hppq ´ pp ´ kq2 ´ µωpkq φ b b˚k Ω dk . (4.13)
R3
Next, the number estimates (4.11) imply that in the momentum representation
µ |V pkq|2 |φ̂ppq|2 dpdk
ż
` 2
}RV Ψ0 }H ď 2 2
. (4.14)
R6 |hppq ´ pp ´ kq ´ µωpkq|
The next step is to find an appropriate lower bound of the denominator in the above
integrand. To this end, we use the lower bound |gppq| ě C1 from Remark 4.2, the fact that
|p| ď εµ and ωpkq ě µ|k| to find that
|hppq ´ pp ´ kq2 ´ µωpkq| ě µωpkq ` 2p ¨ k ` k 2 ` gppq
ě p1 ´ 2εqµωpkq ` C1
` ˘
ě Cµ ωpkq ` µ´1 (4.15)
where C “ minp1 ´ 2ε, C1 q. The proof is finished once we combine (4.14) and (4.15). □
Next, we turn to the following proposition, in which we estimate the projection term.
Proposition 4.2 (Projection term). There exists a constant C ą 0 such that
C` ˘
}IptqQΩ VRVΨ0 }H ď 1 ` µ1{2 |t| }ωµ´1 V }2L2 (4.16)
µ
for all t P R and µ ą 0 large enough.
Remark 4.3. Before we turn to the proof, we make the following observation. Let n ě 1
and let Φ P H be an n-particle state, i.e. N Φ “ nΦ. Then
` 1 ˘
}IptqΦ}H ď 2 ` |t| µ1{2 }V }L2 pn ` 1q 2 }RΦ}H . (4.17)
Indeed, starting from (4.1) one uses that }Bptq} ď 2 as well as }Iptq} ď |t| in operator
norm. Recall that rR, N s “ 0. Thus, }VRΦ} ď }VpN ` 1q´1{2 pn ` 1q1{2 RΦ}. Then, it suffices
to use the standard estimate }VpN ` 1q´1{2 } ď µ1{2 }V }L2
18 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Proof of Proposition 4.2. We use the observation (4.17) as well as the decomposition (4.2)
of the interaction term to find that for some constant C ą 0
` ˘
}IptqQΩ VRVΨ0 }H ď C 1 ` µ1{2 |t| }RV` RV` Ψ0 }H . (4.18)
It suffices now to estimate the norm on the right hand side. To this end, we do a two-fold
application of the commutation relation (4.7) to find that in analogy with (4.13)
ż
` `
RV RV Ψ0 “ µ V pk1 qV pk2 qe´ipk1 `k2 qX Rp pk1 qRp pk1 , k2 qφ b b˚k2 b˚k1 Ωdk1 dk2 . (4.19)
R6
Here, Rp pk1 q is the operator that appears from the commutation of R and one of the
b˚k operators. The resolvent Rp pk1 , k2 q appears due to the commutation of R and two b˚k
operators. They are defined via spectral calculus for p “ ´i∇X on the subspace 1p|p| ď
εµqL2X via the formulae:
´ ¯´1
2
Rp pk1 q :“ hg ppq ´ pp ´ k1 q ´ µωpk1 q , (4.20)
´ ¯´1
Rp pk1 , k2 q :“ hg ppq ´ pp ´ k1 ´ k2 q2 ´ µωpk1 q ´ µωpk2 q , (4.21)
and extended by zero to L2X . In particular, for |p| ď εµ we may replicate the lower bounds
(4.15) for the denominators to find that the following operator norm estimates hold, for a
constant C ą 0
C C
}Rp pk1 , k2 q} ď and }Rp pk1 q} ď . (4.22)
µ ωµ pk2 q µ ωµ pk1 q
Finally, we proceed analogously as we did in the proof of Proposition 4.1. That is, we
combine the number estimate (4.11) and the resolvent bounds (4.22) to find that
ż
2 2
` `
}RV RV Ψ0 }H ď 2µ |V pk1 q|2 |V pk2 q|2 }Rp pk1 qRp pk1 , k2 qφ}2L2 dk1 dk2
X
R6
C ´1 4
ď }ω V }L2 . (4.23)
µ2 µ
This finishes the proof. □

5. Analysis of the generator

In this section we analyze the effective generator gppq for massless boson fields, with
dispersion ωpkq “ |k|. That is, the function on Bµ{2 ” t|p| ă 12 µu satisfying
ş ` ˘´1
gppq “ µ R3 |V pkq|2 µ|k| ` pp ´ kq2 ´ p2 ` gppq dk (5.1)
and first introduced in Definition 2.1. We remind the reader that in Theorem 2.2 we proved
the validity of the approximation with the effective Hamiltonian hg “ p2 ´ gppq.
Our analysis contains two parts. First, we prove that gppq can be approximately solved in
terms of an explicit analytic function of p P Bµ{2 . The error in the approximation is uniform
over compact sets of Bµ{2 . Secondly, based on this analytic approximation, we introduce a
sequence of polynomial generators, corresponding to the truncations of the associated power
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 19

series. These polynomials then induce a sequence of effective Hamiltonians hpN q

g . Combined
with Theorem 2.2, we can study their validity as an approximation of the original dynamics.
This will lead to a proof of Corollary 2.1. We then analyze the optimality of our approach
in Theorem 5.1.
Throughout this section, we assume that V satisfies Condition 2.
5.1. Solving for the generator. The first step towards solving for gppq is the following
result that was already used in the last section.
Lemma 5.1. Fix ε P r0, 12 q. Then, for all |p| ď εµ
ˆ ˙ż ˆ ˙ż
1 ` δε,V pµq |V pkq|2 1 |V pkq|2
dk ď gppq ď dk , (5.2)
1 ` 2ε R3 |k| 1 ´ 2ε R3 |k|
where δε,V pµq Ñ 0 as µ Ñ 8.
Proof. First, we prove the upper bound. Recall that gppq ą 0 and therefore we can estimate
the denominator |hg ppq ´ pp ´ kq2 ´ µωpkq| ě p1 ´ 2εqµ|k|. It suffices to ş plug´1this bound
back in (5.1) and use the triangle inequality. We write C2 ” p1 ´ 2εq ´1
dk|k| |V pkq|2 .
Now, we prove the lower bound. To this end, let θ P p0, 1q. Then, we find
|V pkq|2 dk
ż
gppq “ µ `
2
˘
R3 µωpkq ´ 2p ¨ k ` k ` gppq

|V pkq|2 dk
ż
ěµ `
2
˘
R3 p1 ` 2εqµ|k| ` k ` C2

|V pkq|2 dk
ż
ěµ `
2
˘
tθ´1 µ´1 ď|k|ďθµu p1 ` 2εqµ|k| ` k ` C2
ş
tθ´1 µ´1 ď|k|ďθµu
|k|´1 |V pkq|2 dk
ě ` ˘ . (5.3)
1 ` 2ε ` θp1 ` C2 q
where in the last line we used k 2 ď θµ|k| and 1 ď θµ|k|. It suffices now to take θ “ µ´1{2 ,
apply the monotone convergence theorem, and perform elementary manipulations. □
In our next result, we describe the generator gppq in terms of an explicit function of
p P Bµ{2 , plus a small error term. To this end, we denote g0 :“ gp0q ą 0. We also introduce
the notation for the denominator
Dpkq :“ µ|k| ` k 2 ` g0 , k P R3 , (5.4)
which we use extensively in the rest of this section.
Proposition 5.1. Fix ε P r0, 12 q. Then, there is a constant C ą 0 such that for all |p| ď εµ
ˇ ż 2
ˆ ˙ ˇ
ˇ µ |V pkq| 2|p||k| ˇ C
ˇgppq ´ tanh´1
dk ˇˇ ď |||V |||2µ , (5.5)
ˇ 2|p| R3 |k| Dpkq µ
for all µ ą 0 large enough.
20 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS
ş
Proof. Let us denote Dpp, kq ” µ|k|`k 2 ´2p¨k `gppq so that gppq “ µ dk|V pkq|2 {Dpp, kq.
In particular, Lemma 5.1 implies that C1 ď gppq ď C2 for a pair of constants C1 and C2 .
Thus, we find the following lower bounds for the denominators
Dpp, kq ě C3 pµ|k| ` 1q and Dpkq ˘ 2p ¨ k ě C3 pµ|k| ` 1q (5.6)
for an appropriate constant C3 ą 0.
The first step towards (5.5) is to compare gppq and its denominator Dpp, kq with the
simpler one Dpkq ´ 2p ¨ k. We find
|V pkq|2 dk |V pkq|2 dk |V pkq|2 dk
ż ż ż
gppq ` ˘
“ “ ` gppq ´ g0 ` ˘ . (5.7)
µ R3 Dpp, kq R3 Dpkq ´ 2p ¨ k R3 Dpp, kq Dpkq ´ 2p ¨ k

Thus, (5.7) and (5.6) imply

ˇ 2
ˇ
|V pkq|2 dk
ż ż
ˇ gppq |V pkq| dk ˇ 2C2 C
ˇ
ˇ µ ´ ˇ ď 2 ˘2 “ 2 |||V |||2µ ,
R3 Dpkq ´ 2p ¨ k C3 R3 µ|k| ` 1 µ
ˇ `

where C ” 2C2 {C32 and we remind the reader that |||V |||µ is given by (2.10). This gives the
estimate (5.5) in the statement of the proposition.
The second step is to analyze the integral on the left hand side of (5.5). To this end, we
use that V pkq and Dpkq are rotational symmetric, and thus we compute
|V pkq|2 dk |V pkq|2 Dpkqdk |V pkq|2 dk
ż ż ż
“ “ ` 2p¨k 2
˘ . (5.8)
R3 Dpkq ´ 2p ¨ k R3 pDpkq ´ 2p ¨ kqpDpkq ` 2p ¨ kq R3 Dpkq 1 ´ | Dpkq |

In the last integral we can compute the angular integration. Let us denote ρ “ |k| P p0, 8q
and ω “ k{|k| P S2 . Without loss of generality, we may assume inside the integral that
p ¨ k “ |p|k3 and, therefore, in spherical coordinates we compute
2 ż 2π żπ
ˆ ˙
|V pkq|2
ż ż8
1 2 |V pρq| sin θdθ
2p¨k
dk “ dρ ρ dφ ` cos θ 2
˘
R3 Dpkq 1 ´ | Dpkq |2 0 Dpρq 0 0 1 ´ | |p|ρ
Dpρq
|
2|p|ρ
|V pρq|2 Dpρq Dpρq dx
ż8 ż
“ 4π dρ ρ2
0 Dpρq 2|p|ρ 0 1 ´ x2

2π 8
ż ´ 2|p|ρ ¯
“ dρ ρ|V pρq|2 tanh´1 .
|p| 0 Dpρq
This finishes the proof. □

5.2. Polynomial approximations. From Proposition 5.1 we conclude that gppq is ap-
proximately determined by an explicit function of |p|.
Let us now argue that the explicit function gives rise to an analytic approximation. To see
this, recall that the inverse hyperbolic tangent function is analytic in p´1, 1q and its power
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 21
n
series expansion is tanh´1 pxq “ nP2N´1 xn . Thus, the generator gppq is approximately
ř
given by a power series: Fixing 0 ď ε ă 1{2, we have
˙j
|||V |||2µ
ˆ ˙ ˆ
2j µ |V pkq|2
ż
ÿ
j |k|
gppq “ αj pµq|p| ` O , αj pµq :“ dk (5.9)
jP2N
µ pj ` 1q R 3 Dpkq Dpkq
0

with the error being uniform over |p| ď εµ. Note that the coefficients are positive αj pµq ą 0
and satisfy αj pµq „ µ´j . More precisely
2j |V pkq|2
ż
j
lim µ αj pµq “ dk . (5.10)
µÑ8 pj ` 1q R3 |k|

By combining the results from Theorem 2.2 and (5.9) we can now verify Corollary 2.1
about the approximation of the original dynamics with the polynomial generators
ÿ
2
hpN
g
q
ppq “ p ´ αj pµq|p|j N P 2N0 . (5.11)
jP2N0 :jďN

Proof of Corollary 2.1. Let N P 2N0 and ε P r0, 12 q and note that due to Condition 1, we
can assume that P0 ď εµ. First note that the coefficients αj pµq in (5.9) satisfy the upper
bound
p2{µqj |V pkq|2
ż
αj pµq ď dk (5.12)
pj ` 1q R3 |k|
for all j P 2N0 and µ ą 0. Thus, an elementary estimate using the geometric series shows
that (5.12) implies that for any J ě 2
8 ˆ ˙J ˆ ˙J
ÿ j }|k|´1{2 V }2 2P0 2P0
αj pµqP0 ď ” C0 (5.13)
j“J;j even
p1 ´ 2P 0 {µq µ µ

Note that C0 is independent of J. Next, we use the Duhamel formula, the power series
expansion (5.9) with ε “ Pµ0 , and the bound (5.13) with J “ N ` 2 to find that for some
constant C ą 0
pN q
}e´ithg φ ´ e´ithg φ} ď |t| }phg ´ hgpN q qφ}
ˆ ÿ 8 ˙
j ´1 2
ď |t| αj pµq} |p| φ} ` Cµ |||V |||µ
j“N `2
ˆ ˙
´1 N `2
` ˘
ď C |t| 2P0 µ ´1
` µ |||V |||2µ . (5.14)

This finishes the proof for N ă 8, after we invoke Theorem 2.2 and use the triangle
inequality. Using Proposition 5.1, the proof for N “ 8 follows directly from Theorem 2.2
and the triangle inequality. □
22 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

b b b b b
... ...
I0 I2 I4 IN I
a a a a a

Figure 2. The pa, bq-regions IN

5.3. Range of validity. In the remainder of this section, we provide a detailed comparison
of the range of validity of the sequence of polynomial approximations, as stated in Corollary
2.1. The following discussion can be viewed as an elaboration on Remark 2.3.
In order to extend the discussion further, we parametrize momentum and time scales
through two non-negative numbers pa, bq as follows
φ P 1p 21 µa ď |p| ď µa qL2X and t „ µb , (5.15)

where φ P L2X is the initial datum. In this context, our strongest result in the massless
case is Theorem 2.2, which proves limµÑ8 pe´ithg φ b Ω ´ Ψptqq “ 0 for all pa, bq P I :“
r0, 1q ˆ r0, 1{2q.
Certainly, employing the polynomial generators hgpN q ppq as an effective dynamics has
the advantage of being explicit in comparison to hg ppq. However, the truncation of the
series introduces an additional error term in Corollary 2.1, relative to Theorem 2.2. Thus,
convergence is only guaranteed for parameters pa, bq P IN belonging to the smaller regions
ď “ ‰ “ ˘
IN :“ 0, a ˆ 0, minp1{2, pN ` 2qp1 ´ aqq . (5.16)
aPr0,1q

Note that as n Ñ 8, the sets IN start to cover the full window I. For illustration purposes,
some of these regions are sketched in Figure 2.
The next theorem demonstrates that these smaller pa, bq-regions are in fact optimal. More
concretely, we show that the approximation using hpN q
g ppq fails on the diagonal boundary
of IN . That is, for bpaq “ pN ` 2qp1 ´ aq with a P p1 ´ 2pN1`2q , 1q.
Here for simplicity we consider the massless Nelson model (1.7), so that |||V |||2µ “ Oplog µq.
Theorem 5.1. Let N P 2N0 be an even integer, and assume that
ˆ ˙
1 a a 1
φ “ 1p 2 µ ď |p| ď µ qφ with aP 1´ ,1 . (5.17)
2pN ` 1q
Then, there exists a constant τ “ τ pN q ą 0 such that for t “ τ µb with b “ pN ` 2qp1 ´ aq
›´ pN q
¯ ›
lim inf › e´itH ´ e´ithg ppq φ b Ω› ą 0. (5.18)
› ›
µÑ8
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 23

Remark 5.1. To continue the example from Remark 2.3, consider an initial state φ satisfying
(5.17) with a “ 23{24. Remember that Corollary 2.1 proves
pN q ␣ (
lim pΨptq ´ e´ithg ptq φ b Ωq “ 0 for all |t| ! min µ1{2 logpµq´1 , µpN `2q{24 . (5.19)
µÑ8

pN q
By Theorem 5.1, on the other hand, we know that convergence towards e´ithg ppq Ψ0 fails
for t „ µpN `2q{24 . Thus, we can conclude that (5.19) is in fact optimal. Note that in the
considered case, we hit the threshold at N “ 10, for which the time scale does not extend
further by considering larger values of N . The discussion is easily generalized to other
choices of P0 “ µa .
Proof of Theorem 5.1. Let N P 2N0 . Throughout this proof we consider the remainder
term ÿ
RN ppq “ αj pµq|p|j on 12 µa ď |p| ď µa (5.20)
jěN `2:j even

corresponding to the truncation of the series in (5.9) of order n. In view of (5.13) we have
the following upper bound for |p| ď µa with a ă 1 and µ ą 0 large enough
RN ppq ď C1 µpN `2qpa´1q , (5.21)
for a constant C1 “ C1 pN q. On the other hand, under the assumption |p| ě 21 µa , it follows
from (5.10) that the following lower bounds holds for µ ą 0 large enough
RN ppq ě αN `2 pµq|p|N `2 ě C2 µpN `2qpa´1q (5.22)
ş ´1
for a constant C2 ě CpN q |k| |V pkq|2 dk ą 0.
pN q
Next, we use these bounds to derive a lower bound for pe´ithg ppq ´ e´ihg ppq qφ. To this
end, we invoke Duhamel’s formula twice, that is
´ ¯ żt ´ ¯
pN q pN q
ithg ppq ´ithg ppq
i 1´e e φ “ dteishg ppq e´ishg ppq hg ppq ´ hgpN q ppq φ
´0 ¯
“ t hg ppq ´ hgpN q ppq φ
żt żs ´ ¯2
pN q
ipr´sqhg ppq ´irhg ppq pN q
´ ds dr e e hg ppq ´ hg ppq φ. (5.23)
0 0

In view of Proposition 5.1 and 5.9, we have (the ratio ε “ µa {µ ! 1 can be disregarded)
hg ppq ´ hpN q
g ppq “ RN ppq ` Opµ
´1
logpµqq (5.24)
and thus, using (5.21), (5.22) and µ´1 logpµq “ opµpN `2qp1´aq q as µ Ñ 8, we can estimate
›´ ¯2 › ` ˘2
pN q
h h φ› ď DV µpN `2qpa´1q (5.25)
› ›
› g ppq ´ g ppq
›´ ¯ ›
› hg ppq ´ hgpN q ppq φ› ě CV µpN `2qpa´1q (5.26)
› ›
24 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

for suitable constants DV , CV ą 0. Choosing t “ τ µb with b “ pN ` 2qp1 ´ aq and

τ 2 “ CV {DV ą 0, we arrive at
›´ pN q
¯ › ›´ ¯ › t2 ›´ ¯2 ›
› 1 ´ eithg ppq e´ithg ppq φ› ě |t|› hg ppq ´ hpN q
φ h hpN q
φ›
› › › › › ›
g ppq › ´ › g ppq ´ g ppq
2
DV 2 CV
ě CV τ ´ τ “ ą0. (5.27)
2 2
Together with lim supµÑ8 }pe´itH ´ e´ithg ppq qφ b Ω} “ 0 by Theorem 2.2, this proves the
claimed statement. □

6. Effective dynamics for massive fields

The main goal of this section is to prove Theorem 2.4. To this end, we consider a massive
dispersion relation ?
ωpkq “ k 2 ` m2 (6.1)
with mass term m ě µ´1 . In what follows, Ψ0 denotes the initial datum satisfying Condition
1 relative to 0 ă ε ă 1{2, and V is the form factor satisfying Condition 2. We let Ψeff ptq “
e´ithg φ b Ω be the effective dynamics, with Hamiltonian hg “ p2 ´ gppq. The generator
g : R3 Ñ R is assumed to be bounded and measurable, and shall be chosen below.
The first step is to recall from Section 3 that the following identity holds for all t P R
´ ¯ ´ ¯
eitH Ψptq ´ Ψeff ptq “ Iptq V ` gppq Ψ0 . (6.2)

Recalling our notation PΩ and QΩ from (3.6), an application of the integration by parts
identity yields
IptqQΩ V “ BptqRV ` IptqVRV
“ BptqRV ` IptqPΩ VRV ` IptqQΩ VRV . (6.3)
Therefore, we combine (6.2) and (6.3) to find that
´ ¯ ´ ¯
itH
e Ψptq ´ Ψeff ptq “ BptqRVΨ0 ` Iptq PΩ VRVΨ0 ` gppqΨ0 ` IptqQΩ VRVΨ0 . (6.4)
The next step is note that the expansion process can be continued, if we now apply the
identity (6.3) to the last term in (6.4) and expand IptqQΩ RVR again. This process can be
repeated arbitrarily many times. In particular, for n P N we find
´ ¯
eitH Ψptq ´ Ψeff ptq (6.5)
ÿ ˆ ÿ ˙
“ BptqpRVqj Ψ0 ` Iptq PΩ VpRVqj Ψ0 ` gppqΨ0 ` IptqQΩ VpRVqn Ψ0 ,
1ďjďn 1ďjďn
j odd

where we have used the fact that for any even j P N : PΩ VpRVqj PΩ “ 0.
In the remainder of the section, we show that one can choose a generator gppq “ gn ppq
such that the middle term in (6.5) vanishes. Every other term can be regarded as an error
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 25

term, with precise estimates derived in Section 7. In Section 6.3, the results are combined
to prove Theorem 2.4.
6.1. Choosing the generator. Our present goal is to prove that by choosing g “ gn as
in Definition 2.3, the middle term of (6.5) cancels exactly. This is the content of Lemma
6.1 stated below.
To this end, let us recall the following notation for a special class of collection of sequences
σ of length j P N.
Σ0 pjq “ tσ P t`1, ´1uj : ℓi“1 σpiq ě 1 @ℓ ď j ´ 1 and
ř řj
i“1 σpiq “ 0u . (6.6)
Next, we introduce auxiliary functions that will help us navigate the proof of the upcoming
Lemma. More precisely, let n P N and |p| ă 21 µ. We define Fn pp, ¨q : p0, 8q Ñ p0, 8q as
n`1
ÿ ÿ
Fn pp, xq :“ Fjσ pp, xq (6.7)
j“2 σPΣ0 pjq
j even
ż
# #
Fjσ pp, xq :“ µ j{2
dk1 ¨ ¨ ¨ dkj V pk1 q ¨ ¨ ¨ V pkj q xΩ, bkjσpjq . . . bk1σp1q Ωy
R3j
j´1 ˆ ÿ
ź ℓ ´ ℓ
ÿ ¯2 ˙´1
2
ˆ µ σpiqωpki q ` p ´ σpiqki ´ p ` x . (6.8)
ℓ“1 i“1 i“1

In particular, we may re-phrase the definition of gn ppq given in Definition 2.3 in terms as
solutions of the fixed point equation
gn ppq “ Fn pp, gn ppqq (6.9)
for all |p| ă 21 µ.
Lemma 6.1. Let n P N and let g “ gn be as in Definition 2.3. Then, it holds that
ÿ
PΩ VpRVqj Ψ0 “ ´pgppq b 1qΨ0 . (6.10)
1ďjďn
j odd

Proof. Throughout the proof, we treat gppq as an arbitrary real-valued, bounded measur-
able function on R3 . The precise choice given by Definition 2.3 will only be made at the
end. For transparency, we denote the g-dependent resolvent on QΩ H
Rg :“ phg pp ` P q ´ p2 ´ T q´1 (6.11)
We will consider on the Hilbert space H “ L2X b F the following operator, regarded as a
function of g:
n`1
ÿ
An rgs :“ Aj rgs with Aj rgs :“ PΩ VpRg QΩ Vqj´1 PΩ (6.12)
j“2
j even

which coincides with the operator in (6.5) after the change of variables j ÞÑ j ´ 1.
26 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

First, we use the decomposition V “ V` ` V` into creation and annihilation operators, in

each factor of Aj rgs. The expansion can be represented as a sum over sequences σ P Σ0 pjq
as follows
ÿ
Aj rgs “ Aσj rgs, Aσj rgs :“ PΩ Vσpjq Rg Vσpj´1q ¨ ¨ ¨ Rg Vσp1q PΩ . (6.13)
σPΣ0 pjq

Here, we have dropped the projections QΩ in (6.12) at the expense of summing over the
sequences σ P Σ0 pjq. The summability condition ℓi“1 σpiq ě 1 @ℓ ď j ´ 1 guarantees that
ř
Rg acts on states that contain at least one particle.
The next step is to compute the operator Aσj rgs using commutation relations between Rg
and e´ikX and bk and b˚k :
` ˘
Rg eiki X bki “ eiki X bki hg pp ` P q ´ pp ` ki q2 ´ T ` µωpki q (6.14)
´iki X ˚ ´iki X ˚
` 2
˘
Rg e bki “ e bki hg pp ` P q ´ pp ´ ki q ´ T ´ µωpki q . (6.15)

The idea is to move all the factors eikX bk and e´ikX b˚k to the left of all the resolvents,
in the operator (6.13). Once they hit the vacuum projection PΩ on the right, the shifted
resolvents are evaluated using T Ω “ P Ω “ 0. The result is an operator that only acts on
# #
the L2X variables, tensored with the monomial bkjσpjq . . . bk1σp1q . One then takes the vacuum
ş
expectation
ş value over such monomial. This calculation yields (here p “ ´i∇X and “
R3j
dk1 ¨ ¨ ¨ dkj )
ż j´1
# #
ź
Aσj rgs “µ j{2
V pk1 q ¨ ¨ ¨ V pkj q Rgσp1q¨¨¨σpℓq pp; k1 , ¨ ¨ ¨ , kℓ qxΩ, bkjσpjq . . . bk1σp1q Ωy b 1 (6.16)
ℓ“1

where we introduce the following operators on L2X

ˆ ´ ¯2 ˙´1
σp1q...σpℓq
řℓ řℓ 2
Rg pp; k1 , . . . , kℓ q “ p´1q µ i“1 σpiqωpki q ` p ´ i“1 σpiqki ´ p ` gppq
(6.17)

Finally, in terms of the auxiliary functions introduced in (6.7) we identify that Aσj rgsΨ0 “
p´1qj´1 Fjσ pp, gppqqb1Ψ0 where the right hand side is evaluated on the operator p “ ´i∇X
in the spectral subspace 1p|p| ď εµq. We sum over all sequences σ P Σ0 pjq and all even
2 ď j ď n ` 1 to obtain
n`1
ÿ n`1
ÿ
An rgsΨ0 “ Aσj rgsΨ0 “ ´ Fjσ pp, gppqq b 1 Ψ0 “ ´Fn pp, gppqq b 1 Ψ0 . (6.18)
j“2 j“2
j even j even

Therefore, by choosing g “ gn as in Definition 2.3 we see that An rgn sΨ0 “ ´gn ppq b 1Ψ0 .
This finishes the proof. □
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 27

6.2. Estimate for pRVqn Ψ0 . In this subsection, we state some estimates that we will need
in order to control the expansion (6.5). The main challenge is to estimate terms of the form
#
RV ¨ ¨ ¨ RV# pφ b Ωq .
loooooomoooooon (6.19)
n times
#
Here, V can be either a creation V or annihilation V´ operator (see (4.2) for a definition)
`

and R is the resolvent operator, introduced in Definition 3.2. It is crucial to observe that
the presence of the projections QΩ in the definition of R guarantees that the states (6.19)
are either zero, or contain at least one particle.
In this regard, our most relevant result is Proposition 6.1 stated below. In order to
formulate it, we introduce some some notation that encodes the structure of the states
(6.19). Let us consider the set of sequences of length n, taking values on t`1, ´1u, and
whose all partial sums are bounded below by one. That is,
Σpnq :“ tσ P t`1, ´1un : ji“1 σpiq ě 1 @j “ 1, . . . , nu ,
ř
nPN. (6.20)
Contrary to Σ0 pnq, the set of sequences in Σpnq do not drop to zero at the last entry.
The proof of the following proposition is rather involved and will be postponed to Sec-
tion 7. We remind the reader that we assume m ě µ´1 . Here, we employ the notation
V`1 ” V` and V´1 ” V´ , and n˘ ” |ti : σpiq “ ˘1u| whenever σ is known from context.
Proposition 6.1. Let n ě 2. Then, there exists a constant C ą 0 such that for all
sequences σ P Σpnq
› n
›ź › n a
ˆ ˙n ´
› pRVσpiq qpφ b Ωq› ď C 1 n
›
n{2
pn` ´ n´ q! |||V |||nµ` }V }L´2 (6.21)
›
i“1
›
H µ m
for all µ ą 0 large enough.
Remark 6.1. Two comments are in order
(1) The summability conditions implies n` ´ n´ ě 1. On the other hand, n “ n´ ` n` .
Consequently, 2n´ ď n ´ 1 or equivalently n´ ď t n´1 2
u represents the worst possible
scenario, regarding the growth with respect to m. In particular, we obtain the following
bound that is uniform over all σ P Σpnq
› n › n
? ˆ ˙t n´1 u
›ź σpiq
› C n! 1 2
› pRV qpφ b Ωq› ď
n{2
|||V |||nµ . (6.22)
›
i“1
›
H µ m
where we used1 }V }L2 ď C}V }µ . This bound, although not optimal with respect to
n P N, will be useful in the proof of the main theorem.
(2) Consider a µ-dependent mass term m “ µ´δ where δ P p0, 1q and let n be odd. Then,
1
the right hand side of (6.22) is of order µ´ 2 pnp1´δq´δq . In particular, the rate gets better
with the order of iteration n ě 1, but this growth is modulated by δ, i.e. the size
1Indeed, the Monotone Convergence Theorem implies }V }µ ě 2}|k|´1{2 V }L2 for µ large enough. Addi-
tionally, note that C0 :“ }|k|´1{2 V }L2 {}V }L2 ą 0 is itself by definition a constant. Thus, take C “ 2C0 .
28 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

of the mass term. Further, the case is m “ µ´1 is critical : our bound on the rate of
convergence does not improve with n ě 1 but rather stagnates.

6.3. Proof of Theorem 2.4. We now have all the necessary ingredients to prove our
main approximation result for massive fields.

pnq
Proof of Theorem 2.4. Let Ψeff ptq “ e´ithgn φ b Ω where the generator gn is chosen as in
Definition 2.3. It follows immediately from (6.5), Lemma 6.1 and }Bptq} ď 2 that

pnq
ÿ
}Ψeff ptq ´ Ψptq} ď 2 }pRVqj Ψ0 } ` }IptqQΩ VpRVqn Ψ0 } . (6.23)
1ďjďn

Next, let us note that since n is odd, the state Ψn :“ QVpRVqn Ψ0 has a trivial one-
particle sector, and a non-trivial two-particle sector. In particular, PΩ VRΨn “ 0. Thus,
integrating by parts one more time (i.e. apply (6.3)) leads to

IptqΨn “ BptqRΨn ` IptqVRΨn “ BptqRΨn ` IptqQΩ VRΨn (6.24)

Below we will use this identity in the case µ´1 ď m ď µ´1{2 |||V |||µ , which we call the small
mass case. Conversely, in the opposite large mass case m ě µ´1{2 |||V |||µ , it is favourable to
apply the integration by parts formula one more time, that is

IptqΨn “ BptqRΨn ` BptqRVRΨn ` IptqVRVRΨn . (6.25)

For the remainder estimates, note that in the first case, the relevant state is RΨn “
pRVqn`1 Ψ0 which contains at most n ` 1 particles, an even number. In the second case,
the relevant state is RVRΨn “ pRVqn`2 Ψ0 which contains at most n ` 2 particles, an odd
number. This on its turn implies that the least integer bracket in the estimate (6.22) takes
different values and thus makes the dependence on the mass term different.
The small mass case. An application of (6.23), (6.24) and (6.22) yields

n`1
ÿ
pnq 1
}Ψeff ptq ´ Ψptq} ď 2 }pRVqj Ψ0 } ` |t| µ1{2 }V }L2 pn ` 2q 2 }pRVqn`1 Ψ0 } (6.26)
j“1
n`1 ? ˆ ˙t j´1 u ˆ ˙ n´1
ÿ C j j! 1 2
j
a C n`1 1 2
ď |||V |||µ ` |t| pn`2q! n |||V |||µn`1
j“1
µj{2 m µ2 m

where in the last line we have evaluated t n2 u “ n´1

2
for n odd. The above bound records
the dependence of the estimates with respect to n P N. Let us now simplify it. First,
without loss of generality we assume µ´1{2 |||V |||µ ď 1 (otherwise the bound in Theorem
2.4 holds trivially). Next, for the t-independent error we estimate the summands using
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 29
? a
C j j! ď C n`1 pn ` 1q! as well as the following inequalities (recall n is odd):
j´1 ˆ ˙ j´1 j´2
n`1
ÿ |||V |||jµ ˆ 1 ˙t 2 u n n`1
ÿ |||V |||jµ 1 2 ÿ |||V |||jµ ˆ 1 ˙ 2
“ `
j“1
µj{2 m j“1
µj{2 m j“2
µj{2 m
j odd j even
n ˆ ˙ j´1 n ˆ ˙ j´1
ÿ |||V |||jµ 1 2 |||V |||µ ÿ |||V |||jµ 1 2
“ j{2
` 1{2
j“1
µ m µ j“1
µj{2 m
j odd j odd
n ˆ ˙j
|||V |||jµ 1 2
ˆ ˙
|||V |||µ ? ÿ
“ 1` m
µ1{2 j“1
µj{2 m
j odd
ˆ ˙ n´1
|||V |||µ |||V |||nµ 1
ˆ 2
˙
ď 2n ` n{2 . (6.27)
µ1{2 µ m
In the first line, we evaluated even and odd terms; in the second line we shifted the second
sum j Ñ j ` 1 and in third line we grouped terms. In the last line we used µ´1{2 |||V |||µ ď 1
and elementary geometric series estimates. This gives (here we absorb n ď C n )
ˆ ˙ n´1
|||V |||nµ 1 |t| |||V |||µn`1
ˆ„ ȷ ˙
pnq n
? |||V |||µ 2
}Ψeff ptq ´ Ψptq} ď C n! ` n{2 ` 1{2 (6.28)
µ1{2 µ m µ pµmq n´1 2

The large mass case. An application of (6.23), (6.25) and (6.22) yields
pnq
ÿ 1 1
}Ψeff ptq ´ Ψptq} ď 2 }pRVqj Ψ0 } ` |t| µ 2 }V }L2 pn ` 3q 2 }pRVqn`2 Ψ0 } (6.29)
1ďjďn`2
? ˆ ˙t j´1 u n`2
a ˆ ˙ n`1
ÿ C j j! 1 2
j C pn`3q! 1 2
ď j{2
|||V |||µ ` |t| |||V |||n`2
µ
1ďjďn`2
µ m µ pn`1q{2 m

where in the last line we have evaluated?t n`1 2

u “ n`1
2
for n odd. In order to simplify the
j
? n
t-independent term we use C j! ď C n! for the summands. Additionally, we use the
calculation (6.27) and include an additional term to the sum to find
j´1 ˆ ˙ n`1
n`2
ÿ |||V |||jµ ˆ 1 ˙t 2 u |||V |||µ |||V |||n`2
ˆ ˙
µ 1 2 |||V |||µ
j{2
ď 2pn ` 1q 1{2
` n`2 ď Cn 1{2 (6.30)
j“1
µ m µ µ 2 m µ
where in the last inequality we use that for all k P N in the large mass regime:
ˆ ˙ k´1 ˙ k´1 ˆ ˙ k`1
|||V |||kµ 1 2
ˆ
|||V |||µ 2 |||V |||µ 2 |||V |||µ
k “ ď (6.31)
µ2 m µ1{2 m µ1{2 µ1{2
Thus, we arrive at (again, we absorb powers of n in C n ):
|t| |||V |||n`1
ˆ ˙
pnq n
? |||V |||µ µ |||V |||µ
}Ψeff ptq ´ Ψptq} ď C n! 1 ` 1{2 . (6.32)
µ2 µ pµmq n´1 2
1
µ2 m
30 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

The proof of the theorem is finished once we combine (6.32) and (6.28). □

7. Resolvent estimates
The main goal of this section is to give a proof of Proposition 6.1, stated in Section 6.2,
which allows us to control the norm of states of the form (6.19) in the case of massive
fields. Thus, we assume throughout this section that the dispersion relation is given by
?
ωpkq “ k 2 ` m2 (7.1)
with m ě µ´1 . The generator is taken as in Definition 2.23, but the results in this Section
apply to any non-negative bounded measurable function gppq ě 0.
Throughout this section, the parameter 0 ď ε ă 1{2 is fixed, and V is the form factor
satisfying Condition 2.
The organization of this section is as follows: In Subsection 7.1 we introduce relevant
resolvent functions. In Subsection 7.2 we give an example of an estimate for a state (6.19)
consisting of only V` . Finally, in Subsection 7.3 we turn to the proof of Proposition 6.1.

7.1. Resolvents. Let us generalize the resolvents (4.20) and (4.21) that naturally emerged
in the proof of Theorem 2.2. To this end, let n ě 1 and k1 , ¨ ¨ ¨ , kn P R3 . We denote
kn ” pk1 , . . . , kn q P R3n . For p P R3 , we introduce the following notations
ˆ ˙´1
řn řn 2 2
Rp pkn q :“ p´1q µ i“1 ωpki q ` pp ´ i“1 ki q ´ p ` gppq (7.2)
ˆ ˙´1
ˇ řn řn
Rp pkn ˇkq :“ p´1q µ i“1 ωpki q ´ µ ωpkq ` pp ´ i“1 ki ` kq2 ´ p2 ` gppq (7.3)

Note that both Rp pkn q and Rp pkn |kq are symmetric with respect to the permutation of the
variables k1 , ¨ ¨ ¨ , kn . We often refer to Rp pkn |kq as the contracted resolvent. In particular,
we observe that for all 1 ď i ď n there holds
Rp pkn |ki q “ Rp pknpiq q (7.4)
piq
where kn ” pk1 , . . . , ki´1 , ki`1 , . . . , kn q P R3pn´1q .
The following estimates will be crucial in our analysis. We remind the reader of the
notation ωµ pkq “ ωpkq ` µ´1 and |||V |||µ ” }ωµ´1 V }L2 .
Lemma 7.1. Let n ě 1. Then, the following is true for all |p| ď εµ and all 1 ď i ď n:
(1) There exists a constant C ą 0 such that for all kn P R3n
C
|Rp pkn q| ď (7.5)
µωµ pki q
(2) If n ě 2, then also for all kn P R3n
C
|Rp pkn |ki q| ď . (7.6)
pn ´ 1qmµ
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 31

Proof. (1) The denominator on the right hand side of (7.2) can be bounded below using
p ¨ k ď εµ|k| and gppq ě . We find that
µ nj“1 ωpkj q ` pp ´ nj“1 kj q2 ´ p2 ` gppq “ nj“1 µ ωpkj q ´ 2p ¨ kj ` p nj“1 kj q2 ` gppq
ř ř ř ` ˘ ř

ě C nj“1 µωpkj q .
ř
(7.7)
Next, note that we can use the bound m ě µ´1 and obtain that for constants C1 and C2
C1 ωµ pkq ď ωpkq ď C2 ωµ pkq . (7.8)
Note also that (7.7) implies Rp pkq ă 0. This finishes the proof of the first item, after we
use the trivial bound nj“1 ωµ pkj q ě ωµ pki q. The proof of item (2) is analogous. Here we
ř

use the contraction identity (7.4), and use the alternative lower bound ωpkj q ě m. □
Remark 7.1 (Heuristics). We are interested in the regime for which ωµ´1 V P L2 grows much
slower than µ ą 0 (for instance, logarithmically as in the Nelson model). Consequently,
the generic bound that we use for the resolvent function is of the form (here n “ 1)
C
µ1{2 }Rp pkqV pkq}L2 ď 1{2 |||V |||µ
µ
uniformly in |p| ď εµ. Thus, up to the |||V |||µ norm, this estimate extracts a full factor
µ´1{2 . Hence, we refer to resolvent functions as good resolvents.
On the other hand for the contracted resolvents we use the generic bound (here n “ 2)
for some function ψ P L2 pR6 q
ˇ C
µ1{2 }Rp pk1 , k2 ˇk1 qV pk1 qψpk1 , k2 q}L2 ď 1{2 }V }L2 }ψ}L2
µ m
which is only able to extract the possibly larger factor µ´1{2 m´1 , uniformly in |p| ď εµ.
Hence, we refer to these as bad resolvents.
Remark 7.2 (Operator calculus). The resolvent functions Rp pkn q define bounded operators
on the spectral subspace 1p|p| ď εµqL2X via spectral calculus of p “ ´i∇X , which we
extend by zero to L2X . Unless confusion arises we also denote these operators by Rp pkn q.
In particular, the estimates (7.5) are translated into operator norm estimates.
Remark 7.3. The bounds (7.5) also apply in the massless case m “ 0 provided the generator
satisfies the lower bound gppq ě C. This only has the effect of changing the value of the
constant in Lemma 7.1, and we will use this observation in Appendix A.
7.2. An example. Before we turn to the estimates of a general product state (6.19), let
us warm up with an example that only contains good resolvents. This state arises as the
case where only V` operators are considered, as no contractions are present.
For n ě 1 one can calculate the following representation for the n-body state using
commutation relations
ż n
ź
n ř
` ` ´i i ki ¨X
pRV ¨ ¨ ¨ RV qpφ b Ωq “ µ 2 e V pki qRp pki qφ b b˚kn ¨ ¨ ¨ b˚k1 Ωdk1 ¨ ¨ ¨ dkn .
R3k i“1
(7.9)
32 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

In particular, thanks to Lemma 7.1 and the number estimate (4.11) we find that
ż źn
2 n
` `
}pRV ¨ ¨ ¨ RV qpφ b Ωq}H ď n!µ |V pki q|2 }Rp pki qφ}2L2 dk1 ¨ ¨ ¨ dkn
X
R3pn`1q i“1

C n n!
|||V |||2n
ď µ . (7.10)
µn
Here, C ą 0 is a constant independent of n.
7.3. The general case. In the previous example, there were only good resolvents because
only creation operators V` were used. When annihilation operators V´ are present, ˇone
needs to be more careful because bad resolvents show up: these are of the form Rp pkn ˇkq.
For these, we carry on the contractions between creation- and annihilation- operators to
estimate them. We now study this process.
In what follows, we denote B εµ ” tp P R3 : |p| ď εµu. Our first step is to introduce an
appropriate class of functions with a relevant norm. Namely, for n P N we consider
` ˘
M P L2 L8 pRdn ˆ B εµ q ” L2 R3n ; L8 pB εµ q (7.11)
which we equip with the norm
ˆż ´ ¯2 ˙1{2
}M}L2 L8 :“ sup |Mpkn , pq| dk . (7.12)
R3 |p|ďεµ

Analogously as in Remark 7.2, these functions induce bounded operators Mpkn , pq on L2X .
Example. Up to irrelevant factors, these functions should be regarded as the coefficients
of the product states (6.19). For instance, in the previous example we have
ż ř
` `
pRV ¨ ¨ ¨ RV qpφ b Ωq “ e´i i ki ¨X Mn pkn , pqφ b b˚kn ¨ ¨ ¨ b˚k1 Ωdk1 ¨ ¨ ¨ dkn . (7.13)
R3k
n śn
with Mn pkn , pq :“ µ 2
i“1 V pki qRp pki q.
The specific example given above can be readily generalized in the following sense.
Definition 7.1. Let n P N. Then, to any M P L2 L8 pRdn ˆB εµ q we associate the n-particle
state ż ř
ΦM :“ e´i i ki ¨X Mpkn , pqφ b b˚kn ¨ ¨ ¨ b˚k1 Ωdk1 ¨ ¨ ¨ dkn . (7.14)
R3k

Remark 7.4. For any M P L2 L8 pRdn ˆ B εµ q the number estimate (4.11) shows that
?
}ΦM }H ď n! }M}L2 L8 . (7.15)
2 8
The bound (7.15) motivates the introduction of the space L ř
L pRdn ˆ B εµ q. Let us also
mention here that we have included the pre-factor p´1qn e´i ki ¨X only for convenience.
Our next goal is to study the action of RV` and RV´ on states ΦM . This calculation is
recorded in the following lemma, which is a straightforward consequence of the definitions
of V` , V´ , R and the commutation relations.
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 33

Lemma 7.2. Let M P L2 L8 pRdn ˆ B εµ q and ΦM be as in Definition 7.1. Then,

(1) For all n ě 1 it holds that RV` ΦM “ ΦM` where
M` pkn`1 , pq :“ µ1{2 V pkn`1 qRp pkn`1 qMpkn , pq . (7.16)
(2) For all n ě 2 it holds that RV´ ΦM “ ΦM´ where
n ż
ÿ
1{2
M´ pkn´1 , pq :“ µ V pkn qRp pkn´1 qMpπi kn , pqdkn , (7.17)
3
i“1 R

where we denote kn ” pkn´1 , kn q P Rdn and πi P Sn is the transposition that changes ki

and kn , i.e. πi kn “ pk1 ¨ ¨ ¨ kn ¨ ¨ ¨ ki q.
Remark 7.5. Using the notation of Lemma 7.2, we find that thanks to the resolvent esti-
mates of Lemma 7.1 the following upper bounds hold: for n ě 1
C
}M` }L2 L8 ď |||V |||µ }M}L2 L8 (7.18)
µ1{2
and for n ě 2 (here we use n{pn ´ 1q ď 2)
C
}M´ }L2 L8 ď }V }L2 }M}L2 L8 , (7.19)
µ1{2 m
where for the M´ bound we made use of the Cauchy-Schwarz inequality.
We now have all the ingredients that we need to estimate the general product states
(6.19), i.e. we turn to the proof of Proposition 6.1. Let us remind the reader of the notation
Σpnq “ tσ P t`1, ´1un : ji“1 σpiq ě 1 @j “ 1, . . . , nu ,
ř
nPN. (7.20)
as well as V`1 ” V` , V´1 ” V´ and when σ is known from context n˘ ” |ti : σpiq “ ˘1u|.
Proof of Proposition 6.1. Note that for all σ P Σpnq it holds that σp1q “ σp2q “ `1. Then,
let us start the proof by considering the following two-body state (see e.g. Definition 7.1)
RV` RV` pφ b Ωq “ ΦM2 with M2 pk2 , pq “ µV pk1 qV pk2 qRp pk1 , k2 qRp pk1 q. (7.21)
In particular, }M2 }L2 L8 ď Cµ´1 |||V |||2µ thanks to the resolvent estimate (7.5).
Fix now σ P Σpnq. We define inductively a sequence pMℓ qnℓ“2 as follows. Namely, for
ℓ “ 2 we set M2 as above. Assume that Mℓ for ℓ ě 2 has been defined. We now define in
terms of the notation of Lemma 7.2
#
pMℓ q` , σpℓ ` 1q “ `1
Mℓ`1 “ (7.22)
pMℓ q´ , σpℓ ` 1q “ ´1
In particular, the construction of the function Mn is so that
źn
pRVσpiq qpφ b Ωq “ ΦMn . (7.23)
i“1
34 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

Further, with the kernels defined in this way, we realize thanks to estimates (7.18) and
(7.19) that }Mℓ }L2 L8 ď f pσ, ℓ, µq}Mℓ´1 }L2 L8 where we denote
#
C |||V |||µ if σpℓq “ `1
f pσ, ℓ, µq :“ 1{2 ´1
(7.24)
µ m }V }L2 if σpℓq “ ´1
for all ℓ ě 2. Here, C ą 0 is chosen as the maximum constant between the constants that
appear in (7.18) and (7.19). This estimate can be iterated from ℓ “ n to ℓ “ 3 and we
obtain
źn
}Mn }L2 L8 ď f pσ, r, µq}M2 }L2 L8 . (7.25)
r“3
Finally, we notice that ΦMn is a?state of M :“ n` ´n´ particles. Thus, it suffices now to use
the number estimate }ΦMn } ď M !}Mn }L2 L8 , the initial bound }M2 }L2 L8 ď Cµ´1 |||V |||2µ ,
and the formula (7.24). □

Appendix A. Effective dynamics for massless fields: Revisited

In this section, we consider massless models and iterate the algorithm in the proof of
Theorem 2.2 one more time. Our goal is two-fold. First, we show for the Nelson model
that it is possible to improve the time scale given by Theorem 2.2 by a logarithmic factor.
Secondly, we study massless models whose form factors are less singular than the Nelson
model.
The following family of models serves as an archetype for both situations
Va pkq “ |k|´a 1p|k| ď Λq , (A.1)
where a P r0, 1{2s. It will be convenient to introduce the following scale of norms, quanti-
fying infrared behaviour
|||V |||s,µ :“ }ωµ´s V }L2 , są0. (A.2)
Note that our old norm fits into this scale as follows }V }µ “ }V }1,µ . The reader should
keep in mind that the following estimates hold for the models (A.1)
}Va }1,µ ď Cplog µq1{2 and }Va }2,µ ď Cµ for a “ 1{2 (A.3)
1
`a
}Va }1,µ ď C and }Va }2,µ ď Cµ 2 for a P r0, 1{2q . (A.4)
In our next theorem, we expand the dynamics and obtain an approximation that is valid
over longer time scales, in comparison to the results of Theorem 2.2. Here, in order to keep
the next statement to a reasonable length, we have chosen to consider only the relevant
form factors given by (A.1), although our results apply to general interactions if one wishes
to keep track of the norms }V }s,µ .
Theorem A.1. Let ωpkq “ |k| and V “ Va as in (A.1) with a P r0, 1{2s. Let Ψ0 satisfy
Condition 1. Then, there exists a constant C ą 0 such that for all t P R and µ ą 0 large
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 35

enough
ˆ ˙1{2
log µ
}Ψptq ´ Ψeff ptq}H ďC p1 ` |t|q for a “ 1{2 , (A.5)
µ
C
}Ψptq ´ Ψeff ptq}H ď 1´a p1 ` |t|q for a P r0, 1{2q . (A.6)
µ
Remark A.1. Let us now comment on the consequences that Theorem A.1 has for the
observed time scales.
1
(1) For a “ 2
(the massless Nelson model) we have an effective approximation for time
scales
´ µ ¯1{2
|t| !
log µ
1
which improves the result of Theorem 2.2 by a factor plog µq 2 .
(2) For 0 ď a ă 12 we have an effective approximation for time scales

|t| ! µ1´a
which improves the result of Theorem 2.2 by a factor µ1{2´a .
(3) The reader may wonder if it possible to extend the massless approximation to even
longer time scales. Currently, our methods do not allow for such an extension unless
one introduces a suitable infrared cut-off on the interaction potential V . In particular,
we cannot prove a result similarly to Theorem 2.4 for the massless case.
Proof. Similarly as in the proof of Theorem 2.2, we consider
C|||V |||µ
}Ψptq ´ Ψeff ptq} ď }BptqRVΨ0 } ` }IptqQΩ VRVΨ0 } ď ` }IptqQΩ VRVΨ0 }, (A.7)
µ1{2
where we have employed Proposition 4.1 to estimate the first boundary term. Our next
goal is to give a more precise estimate of the error term
Eptq ” IptqQΩ VRVΨ0 “ IptqV` RV` Ψ0 , (A.8)
where in the second line we have kept only the non-zero contributions. Making use of the
integration by parts formula (4.1) we find
Eptq “ IptqV` RV` Ψ0
´ ¯
“ BptqR ` IptqVR V` RV` Ψ0
“ BptqRV` RV` Ψ0 ` IptqVRV` RV` Ψ0
“ BptqRV` RV` Ψ0 ` IptqV` RV` RV` Ψ0 ` IptqV´ RV` RV` Ψ0 . (A.9)
Observe that the last term of the right hand side of (A.9) contains an operator V´ . This has
the effect of introducing contractions between creation and annihilation operator, which
36 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

lead to worse infrared behaviour. However, these are still one-particle states and may be
integrated by parts one more time. Namely, we find
Eptq “ BptqRV` RV` Ψ0 ` IptqV` RV` RV` Ψ0 ` BptqRV´ RV` RV` Ψ0
` IptqVRV´ RV` RV` Ψ0
“ BptqRV` RV` Ψ0 ` IptqV` RV` RV` Ψ0 ` BptqRV´ RV` RV` Ψ0
` IptqV` RV´ RV` RV` Ψ0 ` IptqV´ RV´ RV` RV` Ψ0 (A.10)
where in the last line we decomposed V “ V` ` V´ . Next, we make use of the observation
form Remark 4.3, as well as the bounds }Bptq} ď 2 and }Iptq} ď |t| to find that
}Eptq} ď C}RV` RV` Ψ0 } ` Cp1 ` µ1{2 |t|q}RV` RV` RV` Ψ0 } ` C}RV´ RV` RV` Ψ0 }
` Cp1 ` µ1{2 |t|q}RV` RV´ RV` RV` Ψ0 } ` C |t| }V´ RV´ RV` RV` Ψ0 } (A.11)
for some constant C ą 0. All the terms in the expansion for Eptq are separately estimated
in Proposition A.1. The proof of the theorem is finished once we gather these estimates
back in (A.7) and use (A.3) to collect the leading order terms in the two different regimes
for a P r0, 1{2s. □
Proposition A.1. There is a constant C ą 0 such that the following five estimates hold.
(i) }RV` RV` Ψ0 }H ď Cµ´1 |||V |||21,µ
(ii) }RV` RV` RV` Ψ0 }H ď Cµ´3{2 |||V |||31,µ
(iii) }RV´ RV` RV` Ψ0 }H ď Cµ´3{2 |||V |||2,µ |||V |||21{2,µ
(iv) }RV` RV´ RV` RV` Ψ0 }H ď Cµ´2 |||V |||1,µ |||V |||2,µ |||V |||21{2,µ
(v) }V´ RV´ RV` RV` Ψ0 }H ď Cµ´1 |||V |||21,µ |||V |||21{2,µ .
In the following proof we will make use of the estimates in Section 7 that were developed
for massive fields, but apply here as well as long as no mass terms appear. See Remark
7.3. We also use the notation for resolvents Rp pkq, Rp pk1 , k2 q, Rp pk1 , k2 |k3 q etc. that was
introduced in Section 7.1.
Proof of Proposition A.1. (i) Apply (7.10) with n “ 2.
(ii) Apply (7.10) with n “ 3.
(iii) The state can be calculated explicitly using the commutation relations. We find
RV´ RV` RV` Ψ0 (A.12)
ż
“ µ3{2 V pk1 qV pk2 qV pk3 qe´ipk1 `k2 ´k3 qX Rp pk1 qRp pk1 , k2 qRp pk1 , k2 |k3 qφ b bk3 b˚k2 b˚k1 Ω

where we employ the same notation as in Section 7. We now contract ş ş the field operators
bk3 b˚k2 b˚k1 Ω “ δpk3 ´ k2 qb˚k1 Ω ` δpk3 ´ k1 qb˚k2 Ω to find that (here “ R2d dk1 dk2 )
ż
3{2
´ `
RV RV RV Ψ0 “ µ `
V pk1 q|V pk2 q|2 e´ik1 X Rp pk1 q2 Rp pk1 , k2 qφ b b˚k1 Ω (A.13)
ż
` µ3{2 |V pk1 q|2 V pk2 qe´ik2 X Rp pk1 qRp pk2 qRp pk1 , k2 qφ b b˚k2 Ω “: ΦM1
 TRACER PARTICLE COUPLED TO A SCALAR FIELD 37

where we used Rp pk1 , k2 |k2 q “ Rp pk1 q and Rp pk1 , k2 |k1 q “ Rp pk2 q. Here we have written
the one-particle state RV´ RV` RV` Ψ0 in terms of ΦM1 (see Definition 7.1) with coefficients
ż
M1 pk1 , pq :“ µ V pk1 qRp pk1 q |V pk2 q|2 Rp pk1 , k2 qdk2
3{2 2

ż
` µ V pk1 qRp pk1 q |V pk2 q|2 Rp pk2 qRp pk1 , k2 qdk2 .
3{2
(A.14)

Finally, we borrow estimates from Section 7. An application of (7.15) and (7.5) imply that
´1
}RV´ RV` RV` Ψ0 } ď }M1 }L2 L8 ď Cµ´3{2 }ωµ´2 V }L2 }ωµ 2 V }2L2 . (A.15)
(iv) In the notation of (iii) and Lemma 7.2 we note that RV` RV´ RV` RV` Ψ0 “ RV` ΦM1 “
ΦM`1 . Consequently,
}RV` RV´ RV` RV` Ψ0 } “ }ΦM`1 } ď }M`
1 }L2 L8 ď Cµ
´1{2
|||V |||1,µ }M1 }L2 L8 . (A.16)
It suffices to combine the previous estimate with (A.15) to finish the proof.
ş
(v) We use the representation (A.13) and act on it with V´ “ V pkqeikX bk dk to find
ˆż ˙
1
´ ´ ` `
V RV RV RV Ψ0 “ µ 2 V pkqM1 pk, pqdk φ b Ω . (A.17)
R3

Next, we note that for all |p| ď εµ we may estimate thanks to the explicit expression (A.14)
and the resolvent estimates (7.5):
›ż ›
› 1{2
› C
› µ V pkqM 1 pk, pqdk › ď |||V |||21,µ |||V |||21{2,µ . (A.18)
› 3
R
› µ
The proof is then finished once we combine the last two displayed estimates. □

Acknowledgments. E.C. is deeply grateful to Robert Seiringer for his hospitality at ISTA,
without which this project would not have been possible. E.C. is thankful to Thomas Chen
for valuable comments and for pointing out useful references. E.C gratefully acknowledges
support from the Provost’s Graduate Excellence Fellowship at The University of Texas at
Austin and from the NSF grant DMS- 2009549, and the NSF grant DMS-2009800 through
T. Chen.

References
[1] A. Arai. An asymptotic analysis and its application to the nonrelativistic limit of the Pauli-Fierz
and a spin-boson model. J. Math. Phys. 31, 2653–2663, (1990).
[2] V. Bach, T. Chen, F. Jérémy, J. Fröhlich and I.M. Sigal. Effective Dynamics of an Electron
Coupled to an External Potential in Non-relativistic QED. Ann. Henri Poincaré 14, 1573–1597 (2013).
[3] E. B. Davies. Particle–boson interactions and the weak coupling limit. J. Math. Phys., 1, 20 (3):
345–351, (1979).
[4] F. Hiroshima. Weak coupling limit with a removal of an ultraviolet cutoff for a Hamiltonian of
particles interacting with a massive scalar field. Infinite Dimensional Analysis, Quantum Probability
and Related Topics, 1(03), 407-423 (1998).
38 ESTEBAN CÁRDENAS AND DAVID MITROUSKAS

[5] F. Hiroshima. Weak coupling limit and removing an ultraviolet cutoff for a Hamiltonian of particles
interacting with a quantized scalar field. J. Math. Phys., 40(3), 1215-1236 (1999).
[6] F. Hiroshima. Observable effects and parametrized scaling limits of a model in non-relativistic quan-
tum electrodynamics. J. Math. Phys. 43, 1755–1795 (2002).
[7] F. Hiroshima and I. Sasaki. Enhanced binding of an N-particle system interacting with a scalar
bose field I. Mathematische Zeitschrift, 259, 657–680, (2008).
[8] F. Hiroshima and I. Sasaki. Enhanced binding of an N-particle system interacting with a scalar
field II. Relativistic version. Publ. Res. Inst. Math. Sci., 51 (2015), 655–690, (2015).
[9] M. Jeblick, D. Mitrouskas, S. Petrat and P. Pickl. Free time evolution of a tracer particle
coupled to a Fermi gas in the high-density limit. Commun. Math. Phys., 356, 143–187, (2017).
[10] M. Jeblick, D. Mitrouskas and P. Pickl. Effective dynamics of two tracer particles coupled to a
Fermi gas in the high-density limit. In Macroscopic limits of quantum systems, Springer Proceedings
in Mathematics & Statistics, (2018).
[11] N. Leopold, D. Mitrouskas, S. Rademacher, B. Schlein and R. Seiringer. Landau–Pekar
equations and quantum fluctuations for the dynamics of a strongly coupled polaron. Pure Appl. Anal.,
3, 4, 653–676, (2021).
[12] D. Mitrouskas. A note on the Fröhlich dynamics in the strong coupling limit. Letters in Mathemat-
ical Physics, 45, 111, (2021).
[13] D. Mitrouskas and P. Pickl. Effective pair interaction between impurity particles induced by a
dense Fermi gas. Preprint, https://arxiv.org/abs/2105.02841, (2021).
[14] H. Spohn and S. Teufel. Semiclassical motion of dressed electrons. Rev. Math. Phys. 4, 1–28
(2002).
[15] T. Takaesu. Scaling limit of quantum electrodynamics with spatial cutoffs. J. Math. Phys., 52(2)
(2011).
[16] T. Takaesu. Scaling limits for the system of semi-relativistic particles coupled to a scalar bose field.
Letters in Mathematical Physics, 97, 213-225 (2011).
[17] T. Takaesu. Scaling limits with a removal of ultraviolet cutoffs for semi-relativistic particles system
coupled to a scalar Bose field. Journal of Mathematical Analysis and Applications, 477(1), 657-669
(2019).
[18] L. Tenuta and S. Teufel. Effective Dynamics for Particles Coupled to a Quantized Scalar Field.
Commun. Math. Phys. 280, 751–805 (2008).
[19] S. Teufel. Adiabatic perturbation theory in quantum dynamics. Springer Science & Business
Media, 2003.
[20] S. Teufel. Effective N-body dynamics for the massless Nelson model and adiabatic decoupling
without spectral gap. Annales Henri Poincaré 3, 939–965 (2002).

(Esteban Cárdenas) Department of Mathematics, University of Texas at Austin, 2515

Speedway, Austin TX, 78712, USA
Email address: [email protected]

(David Mitrouskas) Institute of Science and Technology Austria (ISTA), Am Campus 1,

3400 Klosterneuburg, Austria
Email address: [email protected]