You are on page 1of 227

LS and IEs from its analytical and unitarity requirements.

J. A. Oller
Departamento de Fı́sica. Universidad de Murcia.
E-30071 Murcia, Spain.
oller@um.es

Contents
1 Introduction to the Lippmann-Schwinger equation 1
1.1 Stationary formulation of the scattering problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 The Lippmann-Schwinger equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Characteristic properties. Off-shell unitarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 LS equation in partial waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Direct and inverse Lippmann-Schwinger equation 13

3 Analytical properties of the potential 15


3.1 Calculation of the spectral function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2 Criterion for singular potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Partial-wave projected potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4 Right-hand cut 30

5 Partial wave projection 31


5.1 Schwarz reflection principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

6 Dynamical cuts in the half-off-shell amplitude 33

7 Calculation of the on-shell discontinuity from the LS equation 42


7.1 S waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.2 General case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.3 One-step process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.4 Condensed and shorter derivation of the IE for f(ν) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.5 Coupled-channel case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

8 Potential with arbitrary spectral decomposition in coupled channels 76


8.1 Relationships between partial waves with different arguments . . . . . . . . . . . . . . . . . . . . . . . . . 77
8.2 Calculation of the on-shell discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

9 Nonlinear DR for the on-shell partial wave 92


9.1 Imposing the threshold behavior for higher partial waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
′2
10 Dispersion relations for N (k, k′ ; km ). The half-off-shell case. 99
10.1 Sketch of the situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.2 DRs for N (k, k′ ; k′ 2 /m) in the variables k and k′ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
10.2.1 The subtraction functions ai (k) and bi (k′ ) are entire functions . . . . . . . . . . . . . . . . . . . . 105
10.3 DRs for N (k, k′ ; k′ 2 /m) in the variables k2 and k′ 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

11 Calculation of the discontinuities needed in the DRS for half-off-shell scattering 110

12 Different types of DRs within the N/D method 118


12.1 N/D11 DRs, ℓ = 0: Fixing the subtraction constant in terms of as . . . . . . . . . . . . . . . . . . . . . . 118
12.2 N/D12 DRs, ℓ = 0: Fixing the subtraction constants in terms of as and rs . . . . . . . . . . . . . . . . . 118
12.3 N/D22 DRs, ℓ = 0: Fixing the subtraction constants in terms of as , rs and v2 . . . . . . . . . . . . . . . 121
12.4 N/D12 DRs, ℓ = 1: Fixing the subtraction constant δ2 in terms of aV . . . . . . . . . . . . . . . . . . . . 122
12.5 N/D22 DRs, ℓ = 1: Fixing the subtraction constants ν2 and δ2 in terms of aV and rV . . . . . . . . . . . 123
12.6 N/D23 DRs, ℓ = 1: Fixing the subtraction constants ν2 , δ2 and δ3 in terms of aV , rV and ν2 . . . . . . . 124
12.7 N/D13 DRs, ℓ = 1: Fixing δ1 and δ2 in terms of aV and rV . . . . . . . . . . . . . . . . . . . . . . . . . 124
12.8 Imposing ERE for a partial wave with orbital angular momentum ℓ . . . . . . . . . . . . . . . . . . . . . . 126
12.9 (0,1)+m-CDD’s as (m,m+1) DR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
12.10Non-equivalence of (0,2) and (1,1) DRs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
12.11(ℓ, ℓ) and (ℓ, ℓ + 1) DRs. Threshold behavior for singular potentials . . . . . . . . . . . . . . . . . . . . . 129
12.12 DRs with more subtractions in N (A) than in D(A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
12.13 Isolated and relevant conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
12.14 Attractive/repulsive singular potential V (r) for r → 0 and asymptotic behavior of ∆(A) . . . . . . . . . . 135

13 New method to solve the N/D IE: Expansion in Legendre Polynomials 136

14 Application of the formalism to N N 1 S0 140


14.1 Next-to-leading order (NLO) study for the 1 S0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

15 Application of the formalism to N N 1 P1 148

16 Application of the formalism to N N 3 P0 161


16.1 OPE approximation to the potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

17 Application of the formalism to N N −V (3 P0 ) 163


17.1 ∆(A) for this PW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
17.2 N/D01 , N/D11 and N/D12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
17.3 N/D23 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
17.4 N/D22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193

18 Application of the formalism to N N 3 S1 -3 D1 200


18.1 OPE potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
18.2 Calculation of ∆ij (A) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
18.3 Lippmann-Schwinger equation for 3 S1 -3 D1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
18.4 N/D with coupled channels. The simplest case: N/D11 . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
18.5 N/D with coupled channels. General formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
18.6 Numerical procedure used to put in practice the coupled IES of Eqa. (18.99)-(18.100) . . . . . . . . . . . 216
18.7 N D12 case: fixing as and rs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

2
18.8 N D22 case: fixing as and rs with one extra parameter called v2 . . . . . . . . . . . . . . . . . . . . . . . 220

A Basis for the tensor operators in the N N potential: [4] EGM’s & [5] KBW’s forms 221

B Imaginary part of the log in v 221

C List of abbreviations 222

1 Introduction to the Lippmann-Schwinger equation


Let us consider a quantum-mechanical system of two particles with wave function ψ(x1 , x2 ), with xi ∈ R. The Hamiltonian
H has the standard from
 
1 1
Hψ(x1 , x2 ) = − ∇21 − ∇2 + v(x1 − x2 ) ψ(x1 , x2 ) . (1.1)
2m1 2m2 2

Here, we have considered a local potential v(x1 − x2 ) which arises because the particle exchanged (the force carrier)
between two interacting particles arises from a point-like vertex and instantaneous propagation is assumed (accordingly
with nonrelativistic dynamics). This is represented in Fig. 1 where the dashed double line corresponds to the force carrier
and the filled circles the point-like interaction. Next we introduce the relative and center of mass (CM) coordinates, x and

2’
1’

2
1

X, respectively:

x = x1 − x2 ,
1
X= (m1 x1 + m2 x2 ) ,
! M ! !
x I3 −I3 x1
= m1 m2 , (1.2)
X M I3 M I3 x2

where M is the total mass, M = m1 + m2 . The inverse transformation reads


! ! !
m2
x1 M I3 I3 x
= , (1.3)
x2 −mM I3
1
I3 X

3
Now making use of the chain rule one easily obtained for the transformation of the momenta:
1
p = −i∇x = (m2 p1 − m1 p2 ) ,
M
P = −i∇X = p1 + p2 ,
! ! !
m1
p1 I3 M I3 p
= m2 . (1.4)
p2 −I3 M I3 P
Rewriting the Hamiltonian and wave function in terms of the new variables
 
1 1 2
Hψ(X, x) = − ∇2x − ∇ + v(x) ψ(X, x) ,
2µ 2M X
1 2
=− ∇ ψ(X, x) + hψ(X, x) , (1.5)
2M X
where h is the Hamiltonian in the relative variables and the reduced mass of the system, µ, is
m1 m2
µ= . (1.6)
M
In Eq. (1.5) the CM and relative variables are decoupled. We now consider the previous equation in momentum space.
The Fourier transform of the wave function is
Z Z
Φ(P, p) = d3 x d3 Xe−ipx e−iP X ψ(X, x) ,

(1.7)
and for the action of the Hamiltonian, Eq. (1.5), one has
Z Z  
1 2 1 2
d3 x d3 Xe−ipx e−iP X − ∇ − ∇ + v(x) ψ(X, x)
2µ x 2M X
! Z Z
p2 P2
= + φ(P, p) + d3 x d3 Xe−ipx e−iP X v(x)ψ(X, x) . (1.8)
2µ 2M
Let us perform in detail the last Fourier transform:
Z Z Z Z Z
d3 q ′ iq′ x ′ d3 q d3 Q iqx iQX
d3 x d3 Xe−ipx e−iP X e v(q ) e e Φ(Q, q)
(2π)3 (2π)3 (2π)3
Z Z Z
d3 q ′ 3
= d q d3 Qδ(P − Q)δ(p − q − q ′ )v(q ′ )Φ(Q, q) , (1.9)
(2π)3
after doing the integrations in the variables x and X. We end up with
Z
d3 q
v(p − q)Φ(P, q) . (1.10)
(2π)3
From Eqs. (1.8) and (1.10) the action of the Hamiltonian in momentum space then reads
P2
HΦ(P, p) = Φ(P, p) + hΦ(P, p) ,
2M
Z
p2 d3 q
hφ(p) = φ(p) + v(p − q)φ(q) . (1.11)
2µ (2π)3
The action of the operator Hamiltonian h is on the space of the relative variable momentum is given in momentum space.
It is decoupled from the CM variable P .

4
1.1 Stationary formulation of the scattering problem
k2
Stationary Schrödinger equation with energy Ek = 2µ (continuum spectrum of h) and incident three momentum k:
 
1 k2
hψ(x, k) = − ∇2 + v(x) ψ(x, k) = ψ(x, k) . (1.12)
2µ 2µ
We look for solutions of the form

ψ(x, k) = eikx + w(x, k) (1.13)

which is an incident plane wave plus a scattered wave ω(x, k) that satisfies the radiation condition for |x| → ∞:
 
ei|k||x| 1 x
ω(x, k) = f (n, k) +o , n= , (1.14)
|x| |x| |x|

with f (n, k) the scattering amplitude (in a chosen normalization).


Let us rewrite Eq. (1.12) as
!
1 k2
− ∇2 − ψ(x, k) = −v(x)ψ(x, k) . (1.15)
2µ 2µ

The incident wave eikx satisfies the homogeneous free equation:


!
1 k2
h0 ψ(x, k) = − ∇2 − ψ(x, k) = 0 . (1.16)
2µ 2µ

We can then explicit take into account the presence of the incident wave by writing Eq. (1.5) as
!
1 k2
− ∇2 − (ψ(x, k) − eikx ) = −v(x)ψ(x, k) . (1.17)
2µ 2µ

In terms of the resolvent of h0 , denoted by r0 (z),

r0 (z) = (h0 − z)−1 (1.18)

with Imz 6= 0, so that the inverse is defined (if z is real and negative this is also the case). We can write the solution of
Eq. (1.17) as
!−1
ikx k2
ψ(x, k) = e − h0 − − iǫ v(x)ψ(x, k) ,

k2
= eikx − r0 ( + iǫ)v(x)ψ(x, k) , (1.19)

with ǫ → 0+ . The right radiation condition of an outgoing spherical wave is accomplished by the +iǫ.
We also introduce the resolvent of the full Hamiltonian h, r(z):

r(z) = (h − z)−1 , Imz 6= 0 . (1.20)

5
Here even for real and negative z is important to require that Imz 6= 0 because there could exist bound states, that have
negative energy . We can also write two equations for r(z) in terms of r0 (z) and the interaction v:

r(z) = (h0 − z + v)−1 = ((I + vr0 )(h0 − z))−1 = r0 (z)(I + vr0 )−1 = r0 (z)(I + vr0 − vr0 )(I + vr0 )−1
= r0 (z) − r0 (z)vr(z) . (1.21)

Analogously,

r(z) = (h0 − z + v)−1 = ((h0 − z)(I + r0 v))−1 = (I + r0 v)−1 r0 (z) = (I + r0 v)−1 (I + r0 v − r0 v)r0 (z)
= r0 (z) − r(z)vr0 (z) . (1.22)

Let us now prove Povzner’s result for ψ(x, k):


Z
k2
ψ(x, k) = lim −iǫ d3 y r(x, y, + iǫ)eiky (1.23)
ǫ→0+ 2µ

with r(x, y, z) the kernel of r(z) in configuration space.


First we study the action of r0 (k2 /2µ − iǫ) on eikx ,
Z Z
k2 k2
(−iǫ) d3 y r0 (x, y,
+ iǫ)eiky = (−iǫ) d3 y hx|(h0 − − iǫ)−1 |yihy|ki
2µ 2µ
Z
k2
= (−iǫ) d3 y hx|(h0 − − iǫ)−1 |ki = (−iǫ)hx|(−iǫ)−1 |ki = eikx . (1.24)

Next, we take Eq. (1.23) and check that it satisfies the Schrödinger equation Eq. (1.19). We use Eq. (1.21) for r(z),
Z
k2
ψ(x, k) = lim+ (−iǫ) + iǫ)eiky
d3 y r(x, y,
ǫ→0 2µ
Z
k2 k2 k2
= lim (−iǫ) d3 y r0 (x, y, + iǫ)eiky − lim (−iǫ)hx|(h0 − − iǫ)−1 vr( + iǫ)|ki
ǫ→0+ 2µ ǫ→0+ 2µ 2µ
k2
= eikx − lim (h0 − − iǫ)−1 vψ(x, k) . (1.25)
ǫ→0+ 2µ

In the last step we have taken into account again the definition of ψ(x, k) according to the first line of the previous
equation. We can write the last line of Eq. (1.25) as
Z
ikx k2
ψ(x, k) = e − lim d3 y r0 (x, y, + iǫ)v(y)ψ(y, k) . (1.26)
ǫ→0+ 2µ

1.2 The Lippmann-Schwinger equation


We introduce the T -matrix operator t(z) by rewriting Eq. (1.21) for r(z) as

r(z) = r0 (z) − r0 (z)vr(z)


= r0 (z) − r0 (z)t(z)r0 (z) . (1.27)

From the last equation it is clear that t(z) satisfies:

t(z)r0 (z) = vr(z) . (1.28)

6
In the same way, by comparing Eq. (1.28) with Eq. (1.22) one also obtains the following property for t(z)

r0 (z)t(z) = r(z)v . (1.29)

These last two equations imply that the free propagation of a plane wave followed by the a full interaction is equivalent to
the full propagation of the plane wave followed by the interaction with just the potential (and also for reversing the order
between propagation and interaction).
Multiplying Eq. (1.28) to the right by r0 (z)−1 = h0 (z) − z we obtain the Lippmann-Schwinger (LS) equation:

t(z) = vr(z)r0 (z)−1 = v(r0 (z) − r0 (z)vr(z))r0 (z)−1


= v − vr0 (z)t(z) . (1.30)

We can also obtain an analogous equation after multiplying Eq. (1.29) to the left by r0 (z)−1 ,

t(z) = r0 (z)−1 r(z)v = r0 (z)−1 (r0 (z) − r(z)vr0 (z))v


= v − t(z)r0 (z)v . (1.31)

We now take Eq. (1.30) into Povzner’s result Eq. (1.23), by taking into account the free propagation onto a plane
wave, Eq. (1.24), it results

k2 k2 k2 k2
ψ(x, k) = lim (−iǫ)hx|r0 ( + iǫ) − r0 ( + iǫ)t( + iǫ)r0 ( + iǫ)|ki
ǫ→0+ 2µ 2µ 2µ 2µ
k2 k2
= eikx − lim r0 ( + iǫ)t( + iǫ)eikx
ǫ→0+ 2µ 2µ
Z Z
ikx k2 k2
= e − lim d y1 d3 y2 r0 (x, y1 ,
3
+ iǫ)t(y1 , y2 , + iǫ)eiky2 . (1.32)
ǫ→0+ 2µ 2µ
This equation also implies that

k2
lim t( + iǫ)eikx = vψ(x, k) , (1.33)
ǫ→0+ 2µ

as it is clear by comparing Eq. (1.32) with Eq. (1.25). We can also see by employing the LS equation,

k2 k2 k2 k2 k2
t( + iǫ)eikx = (v − vr0 ( + iǫ)t( + iǫ))eikx = v(eikx − r0 ( + iǫ)t( + iǫ)eikx ) = vψ(x, k) , (1.34)
2µ 2µ 2µ 2µ 2µ

in the derivation the understanding that ǫ → 0+ is implicit.


Now let us write the LS equation, Eq. (1.30), both in configuration and momentum space. In these lectures we use the
LS in the latter form.
i) Configuration space: We assume a local potential.

hx′ |xi = δ(x′ − x) ,


hx′ |v|xi = v(x)δ(x′ − x) ,
Z
t(x′ , x, z) = v(x)δ(x′ − x) − d3 y1 v(x)r0 (x, y1 , z)t(y1 , x, z) . (1.35)

ii) Momentum space.

7
We can take the Fourier transform of the result in Eq. (1.35) or take the matrix element of t(z) between plane waves.
In the former case one would proceed analogously to the derivation of Eq. (1.9), taking into account that
Z Z
′ ′
t(k′ , k, z) = d3 x′ d3 x e−ik x t(x′ , x, z)eikx . (1.36)

Taking the matrix element between plane waves the LS equation becomes:

hk′ |ki = (2π)3 δ(k′ − k) , (1.37)


′ ′ ′ ′
t(k , k, z) = hk |t(z)|ki = hk |v|ki − hk |vr0 (z)t(z)|ki . (1.38)

Introducing a resolution of the identity in terms of plane waves between v and t(z), take into account that r0 (z) is diagonal
in momentum space, one has
Z
d3 q 1
t(k′ , k, z) = hk′ |v|ki − 3
hk′ |v|qi q2 hq|t(z)|ki . (1.39)
(2π) −z 2µ

In this form the LS is also valid for nonlocal potentials. Let us now connect the Fourier transform of v(x) introduced in
Eq. (1.9) with the kernel of v as introduced in Eq. (1.35),
Z Z Z Z
′ ′ ′ ′
v(k′ , k) = d3 x′ d3 x e−ik x v(x′ , x)eikx = d3 x′ d3 x e−ik x v(x)δ(x′ − x)eikx
Z

= d3 x e−i(k −k)x v(x) = v(k′ − k) . (1.40)

We insert this result into Eq. (1.39) and the LS in momentum space reads
Z
d3 q v(k′ − q)t(q, k, z)
t(k′ , k, z) = v(k′ − k) − . (1.41)
(2π)3 q2
−z 2µ

Notation:
Full-off-shell scattering: |k′ | =
6 |k|, z 6= Ek , z 6= Ek′ .

Half-off-shell scattering: |k | = 6 |k|, z = Ek or z = Ek′ .
On-shell scattering: |k′ | = |k|, z = Ek = Ek′ .
For a nonlocal potential one can still apply Eq. (1.39), that is usually expressed more commonly as
Z
′ ′ d3 q v(k′ , q)hq|t(z)|ki
t(k , k, z) = v(k , k) − . (1.42)
(2π)3 q2
−z 2µ

Notice that in order to solve Eq. (1.42) for on-shell scattering one needs the half off-shell scattering amplitude.

1.3 Characteristic properties. Off-shell unitarity.


Since both h0 and h are Hermitian (self-adjoint) operators it follows that

r0 (z) = (h0 − z)−1 → r0 (z)† = r0 (z ∗ ) ,


r(z) = (h − z)−1 → r(z)† = r(z ∗ ) . (1.43)

These properties imply for t(z) that

v† = v ,
t(z) = v − vr(z)v → t(z)† = t(z ∗ ) . (1.44)

8
Hilbert identity:
r(z1 ) − r(z2 ) = (z1 − z2 )r(z1 )r(z2 ) . (1.45)
To demonstrate it we take
r(z2 )−1 − r(z1 )−1 = (h − z2 ) − (h − z1 ) = z1 − z2 . (1.46)
Next, we multiply to the left by r(z1 ) and to the right by r(z2 ) and the Hilbert identity is obtained. Proceeding in the
same way, but multiplying to the left by r(z2 ) and to the right by r(z1 ) we would have obtained
r(z1 ) − r(z2 ) = (z1 − z2 )r(z2 )r(z1 ) = (z1 − z2 )r(z1 )r(z2 ) . (1.47)
Note that
r(z1 )r(z2 ) = r(z2 )r(z1 ) , (1.48)
which is clear because a basis that diagonalizes h then diagonalizes simultaneously both r(z1 ) and r(z2 ).
We can rewrite the Hilbert identity in terms of t(z) instead of r(z) by multiplying by v both sides of Eq. (1.45)
vr(z1 )v − vr(z2 )v = (z1 − z2 )vr(z1 )r(z2 )v . (1.49)
The left-hand side (lhs) of this equations is in virtue of Eq. (1.27) t(z1 ) − t(z2 ) while the right-hand side (rhs) is (z1 −
z2 )t(z1 )r0 (z1 )r0 (z2 )t(z2 ) by Eqs. (1.28) and (1.29). We can write from the Hilbert identity
t(z1 ) − t(z2 ) = (z2 − z1 )t(z1 )r0 (z1 )r0 (z2 )t(z2 )
= (z2 − z1 )t(z2 )r0 (z2 )r0 (z1 )t(z1 ) . (1.50)
where the last line follows from the 2nd form of the Hilbert identity, Eq. (1.47). In momentum space we have
Z
′ ′ d3 q 1
t(k , k, z1 ) − t(k , k, z2 ) = (z2 − z1 ) 3
t(k′ , q, z1 ) q2 q2
t(q, k, z2 )
(2π) ( 2µ − z1 )( 2µ − z2 )
Z
d3 q 1
= (z2 − z1 ) 3
t(k′ , q, z2 ) q2 q2
t(q, k, z1 ) . (1.51)
(2π) ( 2µ − z2 )( 2µ − z1 )

Let us consider the case in which z2 = z1∗ , then from the previous expressions we are driven to
Z
′ ′ d3 q t(k′ , q, z1 )t(q, k, z1∗ )
t(k , k, z1 ) − t(k , k, z1∗ ) = −2iImz1
(2π)3 ( q − Rez1 )2 + (Imz1 )2
2

Z
d3 q t(k′ , q, z1∗ )t(q, k, z1 )
= −2iImz1 . (1.52)
(2π)3 ( q2
− Rez1 )2 + (Imz1 )2

Next, we take the limit Imz1 → 0+ (note that if Imz1 → 0− then Imz2 → 0+ and we would repeat the same analysis
exchanging z1 by z2 ). This case is nonzero only for Rez1 > 0, because we then pick up a pole in the denominator of
Eq. (1.52). For any other interval of values of the integration variable |q| the result vanishes as Imz1 = ε → 0+ . Then we
can rewrite Eq. (1.52) as
Z Z ∞
dΩq 1
t(k′ , k, z1 ) − t(k′ , k, z1∗ ) = −2iε2µRez1 t(k′ , |q|q̂, z1 )t(|q|q̂, k, z1∗ ) d|q|
(2π)3 0
q2
( 2µ − Rez1 )2 + ε2
Z Z ∞
dΩq 1
= −2iε2µRez1 t(k′ , |q|q̂, z1∗ )t(|q|q̂, k, z1 ) d|q| . (1.53)
(2π)3 0
q2
( 2µ − Rez1 )2 + ε2

9
Next, we make use of the result
Z +∞ r
d|q|ε µ
I = lim q2
=π . (1.54)
ε→0+ 0 ( 2µ − Rez1 )2 + ε2 2Rez1

This integral can be calculated algebraically by using the Cauchy’s theorem for complex integration. First because the
integrand is an even function in |q| we can write
Z +∞ Z +∞
1 d|q|ε d|q|ε
I = lim = lim µ (1.55)
ε→0+ 2 −∞
q2
( 2µ − Rez1 )2 + ε2 ε→0+ −∞ (|q|2 − 2µRez1 )2 + ε2

with the replacement ε → 2µε. Next we extend the previous integral in the complex plane of |q| and close the integration
contour with an infinite semicircle in the upper half plane of the |q|-complex and apply Cauchy’s theorem for integration.
Making use of Eq. (1.54) we can write Eq. (1.53) as
p Z
′ ′ dΩq
t(k , k, z1 ) − t(k , k, z1∗ ) = −iµ 2µRez1 t(k′ , |q̄|q̂, z1 )t(|q̄|q̂, k, z1∗ )
(2π)2
p Z
dΩq
= −iµ 2µRez1 t(k′ , |q̄|q̂, z1∗ )t(|q̄|q̂, k, z1 ) , (1.56)
(2π)2

with |q̄| = 2µRez1 .
Eq. (1.56) can be written in an operational form. For that let us rewrite it as
Z
′ † d3 q q2
hk |t(z1 ) − t(z1 ) |ki = −2πi δ( − z1 )hk′ |t(z1 )|qihq|t(z1 )† |ki
(2π)3 2µ
Z
d3 q
= −2πi hk′ |t(z1 )δ(h0 − z1 )|qihq|t(z1 )† |ki
(2π)3
= −2πihk′ |t(z1 )δ(h0 − z1 )t(z1 )† |ki
= −2πihk′ |t(z1 )† δ(h0 − z1 )t(z1 )|ki . (1.57)

Where we have taken into account that t(z ∗ ) = t(z)† , Eq. (1.44). Thus we end with the following relation called off-shell
unitarity valid for real z:
t(z) − t(z)† = −2πit(z)δ(h0 − z)t(z)†
= −2πit(z)† δ(h0 − z)t(z) . (1.58)
Another perspective on the off-shell unitarity, Eq. (1.58), results by noticing that
t(k′ , k, z ∗ ) = hk′ |t(z)† |ki = hk|t(z)|k′ i∗ = t(k, k′ , z)∗ . (1.59)
Applying it to Eq. (1.56) we have for half-off-shell unitarity, z = Ek + i0+ ,
Z
′ + ′ + ∗ dΩq
t(k , k, Ek + i0 ) − t(k, k , Ek + i0 ) = −iµ|k| t(k′ , |k|q̂, Ek + i0+ )t(k, |k|q̂, Ek + i0+ )∗
(2π)2
Z
dΩq
= −iµ|k| t(|k|q̂, k′ , Ek + i0+ )∗ t(|k|q̂, k, Ek + i0+ ) , (1.60)
(2π)2
For on-shell unitarity, z = Ek = Ek′ , the same expression as the previous equations results. For on-shell scattering one
also defines the S-matrix operator, whose kernel in momentum space is
S(k′ , k, Ek ) = (2π)3 δ(k − k′ ) − 2πiδ(Ek′ − Ek )t(k′ , k, Ek + i0) , (1.61)

10
On-shell unitarity for the T -matrix implies that S(E) is a unitary operator:

S(E)S(E)† = S(E)† S(E) = I . (1.62)

Let us show it explicitly:


Z Z
′ † d3 q ′ † d3 q
hk |S(Ek )S(Ek ) |ki = hk |S(E k )|qihq|S(E k ) |ki = hk′ |S(E)|qihk|S(E)|qi∗
(2π)3 (2π)3
Z
d3 q n 3 ′ ′
on
3 ∗
o
= (2π) δ(k − q) − 2πiδ(E k ′ − Eq )t(k , q, Ek ′ ) (2π) δ(k − q) + 2πiδ(Ek − Eq )t(k, q, Ek )
(2π)3
= (2π)3 δ(k′ − k) − 2πiδ(Ek′ − Ek )t(k′ , k, Ek ) + 2πiδ(Ek′ − Ek )t(k, k′ , Ek )∗
Z
d3 q
+ (2π)2 δ(Ek′ − Ek ) t(k′ , q, Ek′ )δ(Eq − Ek )t(k, q, Ek )∗
(2π)3
= (2π)3 δ(k − k′ ) , (1.63)

where the last step is a consequence of unitarity, Eq. (1.57). We also note that an infinitesimal positive imaginary part
has been attached to a real energy in Eq. (1.63), even if not explicitly shown, unless the opposite is stated or shown.
Proceeding in the same way we would have for the reversed product

hk′ |S(Ek )† S(Ek )|ki = (2π)3 δ(k′ − k) . (1.64)

The previous unitarity relations simplify in the case of interactions that are invariant under time reversal (θ) and parity
(π) transformations.
θ
t(k′ , k, z) −
→ t(−k, −k′ , z) , t(k′ , k, z) = t(−k, −k′ , z) , (1.65)
′ π ′ ′ ′
t(k , k, z) −
→ t(−k , −k, z) , t(k , k, z) = t(−k , −k, z) , (1.66)
θπ
t(k′ , k, z) −−→ t(k, k′ , z) , t(k′ , k, z) = (k, k′ , z) . (1.67)

Then, Eq. (1.52) simplifies to


Z
′ d3 q t(k′ , q, z1 )t(k, q, z1 )∗
Imt(k , k, z1 ) = −Imz1
(2π)3 ( q2 − Rez1 )2 + (Imz1 )2

Z
d3 q t(k′ , q, z1 )∗ t(k, q, z1 )
= −Imz1 . (1.68)
(2π)3 ( q2 − Rez1 )2 + (Imz1 )2

For the case on-shell and half-off-shell case, Eq. (1.60) reads
Z
µ|k|
′ +
Imt(k , k, Ek + i0 ) = − 2 dΩq t(k′ , |k|q̂, Ek + i0+ )t(k, |k|q̂, Ek + i0+ )∗

Z
µ|k|
= − 2 dΩq t(k′ , |k|q̂, Ek + i0+ )∗ t(k, |k|q̂, Ek + i0+ ) , (1.69)

1.4 LS equation in partial waves


Let |ki be a two-particle state in the relative variable space normalized as in Eq. (1.37). The states with well-defined
orbital angular momentum are:
Z
1
|ℓm, |k|i = √ dk̂Yℓm (k̂)∗ |ki . (1.70)

11
The previous expression can be inverted by making use of the completeness relation of spherical harmonics
∞ X
X ℓ
Yℓm (k̂′ )∗ Yℓm (k̂) = δ(k̂′ − k̂) . (1.71)
ℓ=0 m=−ℓ

This relation can also be seen as


∞ X
X ℓ
hk̂′ |ℓmihℓm|k̂i = hk̂′ |k̂i . (1.72)
ℓ=0 m=−ℓ

The inverse of Eq. (1.70) can be then easily obtained by multiplying its lhs by Yℓm (k̂′ ), summing over all possible values
of ℓ and m and taking into account Eq. (1.71). In this way one has

√ ∞ m=ℓ
X X
|ki = 4π Yℓm (k̂)|ℓm, |k|i . (1.73)
ℓ=0 m=−ℓ

Regarding the normalization of the states |ℓm, |k|i:


ZZ ZZ
′ ′ ′ 1 ′ ′ (2π)3 ′
hℓ m , |k ||ℓm, |k|i = dk̂ dk̂Yℓm (k̂′ )Yℓm (k̂)∗ hk′ |ki = dk̂′ dk̂Yℓm (k̂′ )Yℓm (k̂)∗ δ(k′ − k) . (1.74)
4π 4π
We can write the Dirac-delta function in spherical coordinates as
δ(|k′ | − |k|)
δ(k′ − k) = δ(k̂′ − k̂) , (1.75)
|k|2

and insert this expression back to Eq. (1.74) with the result
′| Z
′ ′ ′ 2 δ(|k − |k|) ′ δ(|k′ | − |k|)
hℓ m , |k ||ℓm, |k|i = 2π dk̂Yℓm m ∗
′ (k̂)Yℓ (k̂) = 2π
2
δℓℓ′ δmm′ . (1.76)
|k|2 |k|2

We now take the matrix element of the LS equation, Eq. (1.30), between states |ℓm, |k|i:

hℓ′ m′ , |k′ ||t(z)|ℓm, |k|i = hℓ′ m′ , |k′ ||v|ℓm, |k|i − hℓ′ m′ , |k′ ||vr0 (z)t(z)|ℓm, |k|i . (1.77)

Because of rotational invariance for particle without spin the orbital angular momentum indices do not mix, ℓ′ = ℓ and
m′ = m. We denote by vℓ (|k′ |, |k|) and tℓ (|k′ |, |k|, z) the matrix elements

hℓ′ m′ , |k′ ||v|ℓm, |k|i = vℓ (|k′ |, |k|)δℓ′ ℓ δm′ m , (1.78)


′ ′ ′ ′
hℓ m , |k ||t(z)|ℓm, |k|i = tℓ (|k |, |k|, z)δℓ′ ℓ δm′ m . (1.79)

Since the states |ℓm, |k|i are also eigenstates of the free Hamiltonian h0 we have for the matrix elements of r0 (z)

1 2π 2 δ(|k′ | − |k|)
hℓ′ m′ , |k′ ||r0 (z)|ℓm, |k|i = δℓ′ ℓ δm′ m . (1.80)
k2
2µ −z |k|2

Inserting a resolution of the identity in between the operators vr0 (z)t(z) we can rewrite Eq. (1.77) as
XZ ∞ d|q||q|2 vℓ (|k′ |, |q|)tℓ (|q|, |k|, z)
′ ′
tℓ (|k |, |k|, z) = vℓ (|k |, |k|) − δℓℓ ′ δmm′ . (1.81)
ℓ′ ,m′ 0
2π 2 |q|2
−z

12
We then arrive to the LS equation in partial waves
Z
1 ∞ vℓ (|k′ |, |q|)tℓ (|q|, |k|, z)
tℓ (|k′ |, |k|, z) = vℓ (|k′ |, |k|) − d|q||q|2 . (1.82)
2π 2 0
|q|2
−z

Let us now follow the same steps as in Sec. 1.3. First we take the matrix elements of the Hilbert identity between states
|ℓm, |k|i (introducing a resolution of the identity making use of the same sates) with the result
Z ∞ dqq 2 tℓ (k′ , q, z1 )tℓ (q, k, z2 )
tℓ (k′ , k, z1 ) − tℓ (k′ , k, z2 ) = (z2 − z1 ) ,
0 2π 2 ( q2 − z1 )( q2 − z2 )
2µ 2ν
Z ∞ dqq 2 tℓ (k′ , q, z2 )tℓ (q, k, z1 )
= (z2 − z1 ) . (1.83)
0 2π 2 ( q2 − z1 )( q2 − z2 )
2µ 2ν

From the fact that t(z ∗ ) = t(z)† , it follows the analogous property to Eq. (1.59) in the spherical basis
tℓ (k′ , k, z1∗ ) = tℓ (k, k′ , z1 )∗ . (1.84)
We now take z2 = z1∗ in Eq. (1.83) which, together with Eq. (1.84), implies
Z ∞ dqq 2 tℓ (k′ , q, z1 )tℓ (k, q, z1 )∗
tℓ (k′ , k, z1 ) − tℓ (k, k′ , z1 )∗ = −2iImz1
0 2π 2 ( q2 − Rez1 )2 + (Imz1 )2

Z ∞ dqq 2 tℓ (q, k′ , z1 )∗ tℓ (q, k, z1 )
= −2iImz1 . (1.85)
0 2π 2 ( q2 − Rez1 )2 + (Imz1 )2

Next we consider Imz1 = ε → 0+ , z1 → z1 + iε (showing explicitly the imaginary part), with z1 ∈ R+ (if z1 ∈ R− the
result of this limit is zero). The same integral as in Eq. (1.54) appears and one has

µ 2µz1 p p
tℓ (k′ , k, z1 + i0+ ) − tℓ (k, k′ , z1 + i0+ )∗ = −i tℓ (k′ , 2µz1 , z1 )tℓ (k, 2µz1 , z1 )∗
√π
µ 2µz1 p p
= −i tℓ ( 2µz1 , k′ , z1 )∗ tℓ ( 2µz1 , k, z1 ) . (1.86)
π
If there is invariance under time reversal it follows that tℓ (k′ , k, z) = tℓ (k, k′ , z). Then, the lhs of the previous equation
can be written as tℓ (k′ , k, z1 + i0+ ) − tℓ (k′ , k, z1 − i0+ ) and it is clear that there is a discontinuity in the partial wave
amplitude tℓ (k′ , k, z1 ) for z1 > 0. This is the extend of the right-hand √ cut (RHC).
Half-off-shell unitarity. We particularize Eq. (1.86) for z1 = Ek , 2µEk = |k|:
µ|k|
tℓ (k′ , k, Ek + i0+ ) − tℓ (k, k′ , Ek + i0+ )∗ = −i tℓ (k′ , k, Ek + i0+ )tℓ (k, k, Ek + i0+ )∗
π
µ|k|
= −i tℓ (k, k′ , Ek + i0+ )∗ tℓ (k, k, Ek + i0+ ) . (1.87)
π
If there is invariance under time reversal (tℓ (k′ , k, z) = tℓ (k, k′ , z)) one concludes from Eq. (1.87) that along the physical
values of the energy, Ek > 0, the half-off-shell partial wave tℓ (k′ , k, Ek ) has the same phase (modulo π) as the on-shell
partial wave tℓ (k, k, Ek ). This is an analogous situation to the Watson’s final-state interaction theorem.
On-shell unitarity, z1 = Ek = Ek′ :
µ|k|
tℓ (k, k, Ek + i0+ ) − tℓ (k, k, Ek + i0+ )∗ = −i tℓ (k, k, Ek + i0+ )tℓ (k, k, Ek + i0+ )∗ ,
π
µ|k|
Imtℓ (k, k, Ek + i0+ ) = − |tℓ (k, k, Ek )|2 . (1.88)

13
The last equation can also be written conveniently (when tℓ (k, k, Ek + i0+ ) 6= 0) as

1 Imtℓ (k, k, Ek + i0+ ) µ|k|


Im +
= − 2
= . (1.89)
tℓ (k, k, Ek + i0 ) |tℓ (k, k, Ek )| 2π
For the on-shell case we can also introduce the partial wave decomposition of the S matrix. From the kernel of the
S-matrix in momentum space we have that its matrix elements between states |ℓm, |k|i is
δ(k′ − k)
Ŝℓ (Ek ) = 2π 2 − 2πiδ(Ek′ − Ek )tℓ (k, k, Ek + i0+ )
|k|2
′  
2 δ(k − k) µ|k| +
= 2π 1−i tℓ (k, k, Ek + i0 ) . (1.90)
|k|2 π

The global (infinite) factor 2π 2 δ(k − k′ )/k2 , attached to the continuous character of the energy, can be extracted out. We
are then considering the subspace of states |ℓmi with the same energy Ek . Within this subspace we define the standard
S-matrix operator in partial waves
iµ|k|
Sℓ (Ek ) = 1 − tℓ (k, k, Ek + i0+ ) . (1.91)
π
This is a unitary operator because it satisfies

Sℓ (Ek )Sℓ (Ek )∗ = Sℓ (Ek )∗ Sℓ (Ek ) = 1 (1.92)

as an immediate consequence of on-shell unitarity in partial waves, Eq. (1.88).

Let us now derive the expression connecting t(k′ , k, z) and tℓ (|k′ |, |k|, z). From the relation between the states |ℓm, |k|i
and |ki, Eqs. (1.70) and (1.73), one has:
ZZ
′ ′ 1
tℓ (|k |, |k|, z) = hℓm, |k ||t(z)|ℓm, |k|i = dk̂′ dk̂Yℓm (k̂′ )Yℓm (k̂)∗ t(k′ , k, z) . (1.93)

Because of the Wigner-Eckart theorem the partial wave amplitude is independent of m. We then sum over m by keeping
in mind the addition theorem of spherical harmonics

X

Pℓ (cos θ ′′ ) = Y m (k̂′ )Yℓm (k̂)∗ , cos θ ′′ = k̂′ · k̂ . (1.94)
2ℓ + 1 m=−ℓ ℓ

Then from Eq. (1.93)



X

hℓm, |k′ ||t(z)|ℓm, |k|i = 4πtℓ (|k′ |, |k|, z)
2ℓ + 1 m=−ℓ
ZZ ℓ
X ZZ
1 4π 1
= dk̂′ dk̂t(k′ , k, z) Y m (k̂′ )Yℓm (k̂)∗ = dk̂′ dk̂Pℓ (k̂′ · k̂)t(k′ , k, z) . (1.95)
4π 2ℓ + 1 m=−ℓ ℓ 4π

Because of rotational symmetry the scattering amplitude t(k′ , k, z) depends only on k2 , k′ 2 and kk′ (we are not considering
spin). Then when performing the angular integrations in Eq. (1.95) for any value of k̂ the angular integration over k̂′ gives
always the same. Thus, the angular integration over k̂ gives rise to a factor 4π and Eq. (1.95) can be rewritten as
Z

4πtℓ (|k |, |k|, z) = dk̂′ Pℓ (k̂′ )t(k′ , |k|z, z) , (1.96)

14
with z a fixed unitary vector that is usually taken along the zed axis. Then, we end with the following more convenient
expression for the partial wave amplitude
Z +1
1
tℓ (|k′ |, |k|, z) = d cos θ ′′ Pℓ (cos θ ′′ )t(k′ , k, z) , (1.97)
2 −1

independently of the chosen direction for the incident three-momentum k, with cos θ ′′ defined in the rhs of Eq. (1.94).
The inversion of the previous relation is straightforward and it is given by (as it can be checked a posteriori)

X
t(k′ , k, z) = (2ℓ + 1)Pℓ (cos θ ′′ )tℓ (|k′ |, |k|, z) , (1.98)
ℓ=0

taking into account the orthogonality property of Legendre polynomials


Z +1 2
Pℓ (x)Pℓ′ (x)dx = δℓℓ′ . (1.99)
−1 2ℓ + 1

Equation (1.98) can also be obtained directly from the expression of the plane wave states |ki in terms of the spherical
basis |ℓmi, Eq. (1.73),

∞ ℓ ′ ℓ
X X X ′

hk |t(z)|ki = 4π Yℓm ′ m ∗ ′
′ (k̂ )Yℓ (k̂) δℓ′ ℓ δm′ m tℓ (|k |, |k|, z)

ℓ,ℓ′ =0 m′ =−ℓ′ m=−ℓ


∞ X
X ℓ ∞
X
= 4π Yℓm (k̂′ )Yℓm (k̂)∗ tℓ (|k′ |, |k|, z) = (2ℓ + 1)Pℓ (cos θ ′′ )tℓ (|k′ |, |k|, z) . (1.100)
ℓ=0 m=−ℓ ℓ=0

2 Direct and inverse Lippmann-Schwinger equation


For on-shell scattering the LS equation in partial waves reads
Z
k2 m ∞ dpp2 k2
t(k, k; ) = v(k, k) + 2 v(k, p)t(p, k; ), (2.1)
m 2π 0 p2 − k 2 m

so that its solution requires the knowledge of the half-off-shell (hfs) partial wave. In order to solve it we proceed by
discretizing the LS equation for half-off-shell scattering. We employ the partition pi ∈ (0, ∞) , i is a natural number,
N
p2k X  m t(pj , pk ; p2k /m + iε) 
t(pi , pk ; + iε) = v(pi , pj ) δjk + 2 p2j ∆j ,
m j=1
2π p2j − p2k − iε
m 2 t(pj , pk ; p2k /m + iε)
Mjk = δjk + p ∆j , (2.2)
2π 2 j p2j − p2k − iε

with ∆j the weight used by the corresponding numerical method employed for integration. The problem with this straight-
forward procedure is that the matrix M cannot be inverted because it is ill-defined for j = k. The method proposed to
me by D. Entem is to take into account that
Z ∞ dp
− = 0 , k2 > 0 (2.3)
0 p2 − k2

15
R
where the symbol − indicates that the Cauchy principal value of the integral is taken. To make use of this fact, let us split
the singular integral in Eq. (2.2) as a sum of the Cauchy principal value plus its residue (k and k′ are real and positive),
Z
′ k′ 2 m ∞ dpp2 ′ k
′2 m ′ ′ ′ k
′2
t(k, k ; + iε) = v(k, k′ ) + 2 − v(k, p)t(p, k ; + iε) + i k v(k, k ′
)t(k , k ; + iε) . (2.4)
m 2π 0 p2 − k′ 2 m 4π m

Now, because of Eq. (2.3), we can get rid of taking the Cauchy principal value by subtracting to the integral above the
zero
Z ′2
m ∞ dp ′2 ′ ′ ′ k
− k v(k, k )t(k , k ; + iε) . (2.5)
2π 2 0 p2 − k′ 2 m

In this way, Eq. (2.4) can also be written as


Z
k′ 2 m ∞ dp  2 ′ k
′2
′2 ′ ′ k
′2 
t(k, k′ ; + iε) = v(k, k′ ) + 2 2 ′ 2 p v(k, p)t(p, k ; + iε) − k v(k, k ′
)t(k , k ; + iε)
m 2π 0 p −k m m
m ′ k′ 2
+i k v(k, k′ )t(k′ , k′ ; + iε) , (2.6)
4π m
which is more suitable for numerical manipulations. Indeed, we can proceed similarly as in Eq. (2.2) and discretize Eq. (2.6)
but, in order to end with a square matrix, we have also to include an extra point corresponding to on-shell scattering,
t(k′ , k′ ; k′ 2 /m + iε). As a result we end with the system of N + 1 equations (with N the number of points in the partition)

k′ 2 m XN
∆j  2 ′ k
′2
′2 ′ ′ k
′2 
t(pi , k′ ; + iε) = v(pi , k′ ) + 2 2 ′ 2 p j v(p ,
i jp )t(p j , k ; + iε) − k v(p i , k ′
)t(k , k ; + iε)
m 2π j=1 pj − k m m
m ′ k′ 2
+i k v(pi , k′ )t(k′ , k′ ; + iε) ,
4π m
k′ 2 m X N
∆j  2 ′ ′ k
′2
′2 ′ ′ k
′2 
t(k′ , k′ ; + iε) = v(k′ , k′ ) + 2 2 ′ 2 p j v(k , p j )t(p j , k ; + iε) − k v(k ′ ′
, k )t(k , k ; + iε)
m 2π j=1 pj − k m m
m ′ ′ ′ k′ 2
+i k v(k , k )t(k′ , k′ ; + iε) , (2.7)
4π m
with k′ 6= pi for all i.
Next, we show that once the hfs T -matrix is calculated we can also obtain the corresponding potential function v. For
that let us establish another partition of N elements in the second argument k′ in t(k, k′ ; k′ 2 /m), kj′ , j = 1, . . . , N , such
that pi−1 < ki′ < pi and 0 < k1′ < p1 . We then have
N
ki′ 2 m X ∆j  2 ′2
′ ki ′2 ′ ′ k i
′2 
t(p, ki′ ; + iε) = v(p, ki′ ) + 2 2 ′ 2 p j v(p, p j )t(p j , ki ; + iε) − k i v(p, ki

)t(ki , ki ; + iε)
m 2π j=1 pj − k i m m
m ′ k′ 2
+i ki v(p, ki′ )t(ki′ , ki′ ; i + iε) . (2.8)
4π m
We assume that the partition is dense enough such that we can consider as a “good approximation” to take

v(p, ki′ ) = v(p, pi ) (2.9)

16
in the rhs of the previous equation, which now reads

ki′ 2 m XN
∆j  2 ′2
′ ki ′2 ′ ′ ki
′2 
t(p, ki′ ; + iε) = v(p, pi ) + 2 2 ′ 2 p j v(p, p j )t(p ,
j ik ; + iε) − k i v(p, p i )t(k ,
i i k ; + iε)
m 2π j=1 pj − k i m m
m ′ k′ 2
+i ki v(p, pi )t(ki′ , ki′ ; i + iε) . (2.10)
4π m
′2
In this way, we have N equations which allow us to determine v(p, pi ) in terms of the hfs partial wave t(p, ki′ ; kmi + iε).
One can check the results that follow from the continuity assumption of Eq. (2.9) by employing other partitions for the
argument k′ and demanding that the resulting potential function is stable, either by increasing N or by changing the
position of the points ki′ with respect to the pi . In a more compact notation, Eq. (2.10) can be written as

ki′ 2 XN
t(p, ki′ ; + iε) = Wij v(p, pj ) ,
m j=1

(2.11)

with the square matrix W given by


 m ′ ′ ′ k′ 2i  m ∆j  2 ′ ki
′2
′2 ′ ′ ki
′2 
Wij = δij 1 + i ki t(ki , ki ; + iε) + 2 2 ′ 2 p j t(p j , k i ; + iε) − δij k i t(k ,
i i k ; + iε) . (2.12)
4π m 2π pj − k i m m

In this way the problem reduces to invert the matrix W. Notice that since the potential is free of RHC we should obtain
the same v(p, pi ) either giving a small positive or negative imaginary part to the energy argument.
• It might be that Eq. (2.8) diverges. This could be cured by introducing a cut-off function in p2j , such that pj < Λ.
The resulting potential is then a function of the cut-off as well, v(p, k′ ; Λ), which would allow us to study the flow of the
potential with the cut-off Λ. Let us note that by its own definition v(p, k′ ; Λ) gives rise to the input T -matrix when the
LS equation is cut with the cut-off Λ. This is complementary to the usual standard in the literature, where the potential
is assumed to be given and then one studies the dependence of the physical results with the cut-off. We find more natural
the opposite situation, the one explored here, that is, that the potential evolves with the cut-off since physical results
should be cut-off independent. E.g. as an application we could compare for low cut-off’s with the Vlow k potentials.

3 Analytical properties of the potential


Let us consider a local potential potential v(r) and write it in terms of its Fourier transform ṽ(q),1 with q = |q|,
Z Z Z Z
d3 q iqr ∞ dqq 2
iqr cos θ 1 ∞ dq q iqr
v(r) = e v
e (q) = dΩq e ve(q) = (e − e−iqr )ve(q)
(2π)3 0 (2π)3 (2π)2 0 ir
Z ∞
1
= 2 Im dq qeiqr ve(q) . (3.1)
2π r 0

Next, we perform the Wick rotation in the variable q, as shown in Fig. 1, and the integration variable along the imaginary
axis is q = iy. Then applying the Cauchy’s integration theorem to the closed integration contour given by the infinity
semicircle and the positive real and imaginary semi-axes, as indicated in Fig. 1, it results
I
dq qeiqr ve(q) = 0 . (3.2)
1
To avoid confusion we use different symbols for the potential in configuration and momentum spaces meanwhile they are simultaneously
used. Later we will just employ the potential in momentum space and we will drop the tilde on e
v (q).

17
Figure 1: Wick rotation applied to the calculation of the inverse Fourier transform of Eq. (3.1).

Note that the integration contour is closed with positive imaginary part so that eiqr = e−rImq eirReq and there is an
exponential dumping along the semicircle at infinity as r > 0. We can then rewrite Eq. (3.1) as
Z ∞ Z ∞
1 −yr 1
v(r) = − 2 Im dy e y ve(iy + ε) = − 2 dy e−yr yImve(iy + ε) . (3.3)
2π r 0 2π r 0

This expression shows that we can write the finite range part of v(r) as a superposition of Yukawa potentials. The
function −Imve(iµ + ε) is called the spectral function, η(µ2 ),
η(µ2 ) = −Imve(iµ + ε) . (3.4)
In terms of it Eq. (3.3) becomes
Z
1 ∞ e−µr
v(r) = dµ2 η(µ2 ) . (3.5)
(2π)2 0 r
The previous equation is suitable to express ve(q) in terms of a dispersion relation (DR),
Z Z Z
3 −iqr 1 ∞ e−yr
ve(q) = d re v(r) = − 2 dy yImve(iy + ε) d3 re−iqr . (3.6)
2π 0 r
The Fourier transform of a Yukawa potential is
Z
e−yr 4π
d3 re−iqr = 2 . (3.7)
r q + y2
This is inserted in Eq. (3.6), which becomes
Z ∞
2 Imve(iy + ε)
ve(q) = − dy y . (3.8)
π 0 y2 + q2
This expression is an unsubtracted DR for the potential ve(q) in momentum space. It explicitly shows that ve(q) only
depends on q 2 . We can also directly check from Eq. (3.8) that
Imve(iµ ± ε) = ±Imve(iµ + ε) . (3.9)
To show it let us substitute q = iµ ± ε, µ > 0, in Eq. (3.8),
Z Z Z
2 ∞ Imve(iy + ε) 2 ∞ Imve(iy + ε) ∞
ve(iµ ± ε) = − dy y 2 2
= − − dy y ± 2i dy yImve(iy + ε)δ(y 2 − µ2 )
π 0 y − µ ± iε π 0 y 2 − µ2 0
Z
2 ∞ Imve(iy + ε)
= − − dy y ± iImve(iµ + ε) . (3.10)
π 0 y 2 − µ2

18
From Eq. (3.9) and the definition of spectral function, Eq. (3.4), one also concludes that

η(µ2 ) = Imve(iµ − ε) = −Imve(iµ + ε) , µ > 0 . (3.11)

An important consequence of Eq. (3.8) is that ve(q) is a function in the q 2 complex plane that only has a left-hand cut,
along which it is necessary that q 2 < 0. For the exchange of massive force carriers there is a lower threshold µ20 above
which the spectral function is zero, so that the LHC extends for q 2 ∈] − ∞, −µ20 ]. Another important property that follows
from Eq. (3.8) is that ve(q 2 ) satisfies the Schwarz reflection principle

ve(z) = ve(z ∗ )∗ . (3.12)

Any analytic function that is real along an open interval of the real axis fulfills Eq. (3.12) in its domain of analyticity. Of
course, the potential is real for q 2 > 0, which is the physical region.

−µ20

Figure 2: Contour of integration C used in the complex integration in Eq. (3.13) to derive the DR for the potential ve(q 2 ).

If ve(q 2 ) does not vanish for q 2 → ∞ one needs to take more subtractions in the DR of Eq. (3.8). This is easily
accomplished by considering the application of the Cauchy’s integration theorem to the integral of the function ve(z)/(z n (z−
q 2 )) along the closed integration contour C shown in Fig. 2 in the q 2 complex plane. This integration contour is a circle
at infinity that engulfs the LHC.
I
ve(z) ve(q 2 ) 1 dn−1 ve(z)
dz n = 2πi + 2πi , (3.13)
z (z − q 2 ) (q 2 )n (n − 1)! dz n−1 z − q 2
z=0
I Z −µ20 Z −µ2
ve(z) ve(k2 + iε) − ve(k2 − iε) 0 Imve(k2 + iε)
dz n = dk2 = 2i dk2 . (3.14)
z (z − q 2 ) −∞ (k2 )n (k2 − q 2 ) −∞ (k2 )n (k2 − q 2 )

Equating Eqs. (3.13) and (3.14) one has


Z
2 (q 2 )n dn−1 ve(z) (q 2 )n −µ20 Imve(k2 + iε)
ve(q ) = − + dk2 . (3.15)
(n − 1)! dz n−1 z − q 2 π −∞ (k2 )n (k2 − q 2 )
z=0

The first term on the rhs of the previous equation is a polynomial in q 2 . Evaluating explicitly the derivative at z = 0 by

19
applying the binomial theorem,
m
!
dm ve(z) X m (m−j) dj 1
m 2
= v
e (z)
dz z − q j=0
j dz z − q 2
j

m
!
X m (−1)j j!
= ve(m−j) (z) . (3.16)
j=0
j (z − q 2 )j+1

One obtains
!
(q 2 )n dn−1 ve(z) (q 2 )n n−1
X n−1 (−1)m m!
− =− ve(n−1−m) (0)
(n − 1)! dz n−1 z − q 2 (n − 1)! m=0 m (−q 2 )m+1
z=0
n−1
X 1
= ve(n−1−m) (0)(q 2 )n−1−m . (3.17)
m=0
(n − 1 − m)!

Introducing the new index for the sum


p=n−1−m , (3.18)
we finally rewrite Eq. (3.15) as
n−1 Z −µ20
2
X ve(p) (0) 2 p (q 2 )n Imve(k2 + iε)
ve(q ) = (q ) + dk2 . (3.19)
p=0
p! π −∞ (k2 )n (k2 − q 2 )

We now perform the inverse Fourier transform of the previous expression to calculate v(r) in configuration space.
 
Z Z Z
d3 q iqr 2 d3 q iqr n−1
X ve(p) (0)
2 p (q 2 )n −µ20 Imve(k2 + iε) 
v(r) = e ve(q ) = e (q ) + dk2 2 n 2
(2π)3 (2π)3 
p=0
p! π −∞ (k ) (k − q 2 ) 
n−1 Z Z −µ20 Z
X v (p) (0) d3 q iqr (−∇2 )n Imve(k2 + iε) d3 q eiqr
= (−∇2 )p e + dk2 . (3.20)
p=0
p! (2π)3 π −∞ (k2 )n (2π)3 k2 − q 2

The two integrations in q in the last line are


Z
eiqr
d3 q = δ(r) ,
(2π)3

Z 2r
d3 q eiqr e− −k
3 2 2
=− , (3.21)
(2π) k − q 4πr
and change of integration variable in the dispersive integral k2 = −µ2 . The resulting expression is
n−1 Z
X v (p) (0) ∞ η(µ2 ) 2 n e−µr
v(r) = (−∇2 )p δ(r) + dµ2 (∇ ) . (3.22)
p=0
p! µ20 (µ2 )n 4π 2 r

From this expression it is clear that the subtractive polynomial in the DR of Eq. (3.19) in momentum space becomes a
polynomial of local terms comprising a Dirac-delta function and its higher order derivatives in configuration space. We
can simplify the dispersive integration in Eq. (3.22) by noticing that
n−1
e−µr e−µr X
(∇2 )n = (µ2 )n − 4π (µ2 )n−1−m (∇2 )m δ(r) . (3.23)
r r m=0

20
Employing this result in Eq. (3.22) it becomes
n−1 Z ∞ Z ∞
X v (p) (0) 2 p 1 n−1
X
2 p
2
2 η(µ ) 2 2 e
−µr
v(r) = (−∇ ) δ(r) − (∇ ) δ(r) dµ + dµ η(µ ) . (3.24)
p=0
p! π p=0 µ20 (µ2 )p+1 µ20 4π 2 r

Depending on the degree of divergence of the spectral function η(µ2 ) in the limit µ2 → ∞ some (or all or any) of the
integrations in the second sum of the rhs of the previous equations will actually diverge. But all these terms in this second
sum can be reabsorbed in the coefficients of the first one (this is called a renormalization procedure) so that we can simply
write
n−1 Z
X v (p) (0) ∞ e−µr
v(r) = (−∇2 )p δ(r) + dµ2 η(µ2 ) . (3.25)
p=0
p! µ20 4π 2 r

This expression implies that any potential can be written as the superposition of Yukawa potentials (weighted in terms of
the spectral function) plus a zero-range part which is a polynomial in δ(r) and its higher order derivatives (according to
the action of the Laplacian operator). To give sense to this polynomial is not a trivial matter and one needs to develop
new techniques which are the main aim of these notes.
The process of acting with ∇2 inside the integrand in Eq. (3.22) is equivalent to manipulate the factor (q 2 )n in the DR
in momentum space, Eq. (3.19), in order to reduce the number of subtractions:
Z −µ20 Z −µ20
(q 2 )n Imve(k2 + iε)
2 (q 2 )n−1 (q 2 − k2 + k2 )Imve(k2 + iε)
dk = dk2
π −∞ (k2 )n (k2 − q 2 ) π −∞ (k2 )n (k2 − q 2 )
Z −µ20 Z −µ20
(q )n−1
2
2 Imve(k2 + iε) (q )n−1
2 Imve(k2 + iε)
=− dk + dk2 (3.26)
π −∞ (k )n
2 π −∞ (k2 )n−1 (k2 − q 2 )

We could iterate further this process for the last integral in the previous equation until the resulting integrals are convergent.
The coefficient in Eq. (3.19) that multiplies (q 2 )n−1 is

ven−1 (0)
(3.27)
(n − 1)!

that cancels with the first term on the rhs of the last line of Eq. (3.26), which is an explicit expression for v (n−1) (0) that
results from a lower than n-times subtracted DR, e.g. take n − 1 derivatives in Eq. (3.8) or in an (n − 1)-times subtracted
DR. Namely,
Z
(−1)p p! ∞ η(µ2 )
ve(p) (0) = dµ2 , (3.28)
π µ20 (µ2 )p+1

an expression that is valid for an m-times subtracted DR for ve(q 2 ) as along as m < p.
The degree of divergence of η(µ2 ) does not only determine the number of subtractions needed to be taken in the DR of
ve(q 2 ) but also how v(r) diverges for r → 0 in configuration space. To see that let us assume that η(µ2 ) −−−→ µn , with
µ→∞
n > 0, and insert this asymptotic behavior for µ > Λ >> µ0 in the integration of Eq. (3.25) (that gives v(r) for r 6= 0).
We have the following contribution to v(r)
Z Z Z
∞ e−µr 1 ∞ (−1)n+1 ∂ n+1 ∞ (n + 1)!
dµ2 η(µ2 ) 2
−→ 2 dµ µn+1 e−µr = dµ e−µr −−−→ , (3.29)
µ20 4π r 2π r Λ 2π 2 r ∂r n+1 Λ r→0 2π 2 r n+3

which gives rise to a singular potential.

21
3.1 Calculation of the spectral function
• In quantum field theory (typically applied in particle physics and in the intersection between this discipline and nuclear
physics) one usually calculates directly the potential in momentum space, from which one can calculate by proper analytical
extrapolation its spectral function by the knowledge of the imaginary part of ve(q 2 ) for negative values of q 2 ± iǫ. However,
in other cases what one knows is the potential in configuration space v(r). Let us determine how to calculate then η(µ2 )
in a way appropriate to singular potentials as well.
We first multiply by the convergent factor e−µr so that there is no problem with the convergence of the Fourier transform
integral in the limit r → ∞, and at the end of the calculations one takes the limit µ → 0+ . In this way, all the potentials
are by construction of short range and are suited for a treatment within S-matrix theory.
Z Z
3 −iqr −µr 4π
ve(q) = d re v(r)e = Im dr re(iq−µ)r v(r) . (3.30)
q
The previous Fourier transform does not exist for singular potentials at the origin, diverging faster or equal than 1/r 3 . For
such cases we can apply the following iteration rule if the potential behaves as
r→0
v(r) −−−→ αr −n , n ≥ 3 , (3.31)
with α a real constant (the constant is real if the potential is Hermitian, as we assume). The method could not be applied
if α were actually a function of r that cannot be expanded in power series of r for r → 0. We do not consider this case
any further because it is not required for the applications that we devise.
Returning to the case of Eq. (3.31) with a constant α we can conveniently rewrite Eq. (3.30) as
Z   Z
4π ∞
(iq−µ)r α 4πα ∞ e(iq−µ)r
ve(q) = Im dr e r v(r) − n + Im dr . (3.32)
q 0 r q 0 r n−1
The point is that the degree of divergence of v(r) − α/r n is reduced compared with v(r) in the limit r → 0. Let us
now discuss the last integral in the previous equation, keeping in mind that i) we are interested in calculating the spectral
function of ve(q) and ii) we proceed by analytical continuation in q starting in the region of positive real values. Performing
an integration by parts and keeping only those terms relevant for the purpose ii) when µ → 0+ one has
Z ∞ Z

(iq−µ)r −n+1 e(iq−µ)r r −n+2 iq ∞
dr e r = + dr e(iq−µ)r r −n+2 . (3.33)
0 −n + 2 n−2 0
0

The first term on the rhs of the last line is zero in the limit r → ∞, and diverges when r → 0 for n ≥ 3. However, this
divergence does not depend on q and, therefore, it does not affect the calculation of the spectral function of ve(q).
Thus, we only keep the last integral term in the previous equation and transform Eq. (3.32) into
Z ∞   Z ∞
4π α 4πα
ve(q) −→ Im dr e(iq−µ)r r v(r) − + Im i dr e(iq−µ)r r −n+2 . (3.34)
q 0 rn n−2 0

The process could be further iterated for the first integral if v(r) − rαn diverges equal or faster than 1/r 3 at short distances,
and for the second one if n ≥ 4.
• Let us now work out explicitly the case of a potential
α
v(r) = n , (3.35)
r
with integer n ≥ 3. Performing the iteration process of Eq. (3.34) n − 2 times one ends with
Z
4παq n−3 ∞ in−2 e(iq−µ)r
ve(q) → Im dr . (3.36)
(n − 2)! 0 r

22
We distinguish between even and odd n because of the factor in−2 :
Even-n:
n−2 Z
4πα(−1) 2 q n−3
even n
∞ sin(qr)e−µr
ve(q) −−−−→ dr
(n − 2)! 0 r
n−2
4πα(−1) q n−3
2 q
= arctan
(n − 2)! µ
n  
2πiα(−1) 2 q n−3 q q
= log(1 + i ) − log(1 − i ) . (3.37)
(n − 2)! µ µ
Here we have expressed
1
arctan x = {log(1 + ix) − log(1 − ix)} . (3.38)
2i
Odd-n:
Z
odd n 4παin−1 q n−3 ∞ e(iq−µ)r
ve(q) −−−→ Im dr i
(n − 2)! 0 r
n+1 n−3
2πα(−1) 2 q q2
= log(1 + 2 )
(n − 2)! µ
n+1 n−3  
2πα(−1) 2 q q q
= log(1 + i ) + log(1 − i ) . (3.39)
(n − 2)! µ µ
When considering complex values of q we take as final expressions those in the last lines of Eqs. (3.37) and (3.39) because
they have the simplest analytical extrapolation with the branch points located at

q = ±iµ , (3.40)

and cuts extending for negative values of the arguments of the log’s,
q
q = ±i µ2 + x2 , x ∈ R , (3.41)

for log(1 ± iq/µ), in order.


• As an example let us work out the DR for v2 (r), corresponding to n = 2 in Eq. (3.35). In this case its Fourier
transform corresponds exactly with the rhs of Eq. (3.37) and gives
 
4π q 2πi q q
ve2 (q) = arctan = − log(1 + i ) − log(1 − i ) . (3.42)
q µ q µ µ
The previous expression has the cuts given in Eq. (3.41) and in order to write down a DR we consider the integration
contour plotted in Fig. 3. The function ve2 (q) vanishes for q → ∞ and then it is not necessary to take any subtraction to
remove the integration along the circle at infinity. We then have
I
1 ve2 (z)
ve2 (q) = dz , (3.43)
2πi C z − q
Z µ Z µ
1 ve2 (−iν − ε) − ve2 (−iν + ε) 1 ve2 (iν + ε) − ve2 (iν − ε)
ve2 (q) = d(−iν) + d(iν)
2πi ∞ −iν − q 2πi ∞ iν − q
Z Z
1 ∞ ve2 (−iν + ε) − ve2 (−iν − ε) 1 ∞ ve2 (iν − ε) − ve2 (iν + ε)
= dν + dν . (3.44)
2π µ iν + q 2π µ iν − q

23
Figure 3: Contour of integration C used in the DR of ve2 (q), Eq. (13.20).

From Eq. (13.18) we directly obtain


 
2π ν ν 4π 2 i
ve2 (iν + ε) − ve2 (iν − ε) = − log(1 − + iε) − log(1 − − iε) = − θ(ν − µ) , (3.45)
ν µ µ ν
4π 2 i
ve2 (−iν + ε) − ve2 (−iν − ε) = θ(ν − µ) . (3.46)
ν
This is substituted in Eq. (13.16) with the result
Z ∞   Z ∞
1 1 1 dν 4π q
ve2 (q) = 2πi dν + = 4π = arctan . (3.47)
µ ν iν − q iν + q µ ν 2 + q2 q µ
• As a second example, also used in later applications, let us consider the calculation of the spectral function for the
short-range potential
7
−1
vs (r) = αe−βr r 2β , β>0. (3.48)

A direct calculation gives for its Fourier transform


2πα 7 n 7 7 o
ves (q) = Γ(1 + ) (−iq + β)−1− 2β − (iq + β)−1− 2β . (3.49)
iq 2β
Now, in complex variable
7 7
(±iq + β)−1− 2β = e−(1+ 2β ) log(±iq+β) , (3.50)

which shows that this function has cuts of the same type as those already discussed for Eq. (3.39)
q
q = ±i β 2 + x2 , x ∈ R . (3.51)

To simplify the writing we introduce the notation


7
γ =1+ . (3.52)

24
The discontinuities of ve2 (q) across the pure imaginary axis is
2πα 2πα
ve2 (iν + ε) − ve2 (iν − ε) = Γ(γ)(e−γ log(−ν+β+iε) − e−γ log(−ν+β−iε) ) = Γ(γ)e−γ log(ν−β) (e−iγπ − eiγπ )θ(ν − β)
ν ν
4πiα
=− Γ(γ)(ν − β)−γ sin(γπ)θ(ν − β) ,
ν
4πiα
ve2 (−iν + ε) − ve2 (−iν − ε) = Γ(γ)(ν − β)−γ sin(γπ)θ(ν − β) , (3.53)
ν

Imve2 (iν − ε) = Γ(γ)(ν − β)−γ sin(γπ)θ(ν − β) = −Imve2 (−iν − ε) . (3.54)
ν

3.2 Criterion for singular potentials


We now consider some remarks concerning the singular potentials, which ultimately will allow us to define a criterion for
their characterization.
i) Let us consider the lowest ℓ = 0 bound state of size a with binding energy −E. Because of the Heisenberg uncertainty
principle the momentum is of order p ≃ 1/a and then we have
1
−E = T + v ≃ +v . (3.55)
2µa2

It is clear that if v(r) is an attractive singular potential with v(a) ∼ −α/an and n > 2 then the system is not stable since
there is no lowest energy bond for a → 0.
ii) For such singular potentials there is not solution of the reduced wave function Schrödinger equation as a power series
expansion around r = 0.

φ(r) = ψ(r)Yℓm (θ, φ) ,


u(r) = rψ(r) , (3.56)
α
v(r) −−−→ γ ,
r→0 r
d2 u(r) α ℓ(ℓ + 1)
− 2
+ γ u(r) + u(r) = k2 u(r) . (3.57)
dr r r2
We try a solution of the form

u(r) ∼ r ν , ν > 0. (3.58)

Substituting it into Eq. (3.57) one obtains

−r ν−2 ν(ν − 1) + αr ν−γ + ℓ(ℓ + 1)r ν−2 = k2 r ν . (3.59)

For the case of a regular potential one has γ < 2 and the kinetic and centrifugal barrier terms dominate over the
potential one for r → 0. We end with the standard equation for ν,

ν(ν − 1) − ℓ(ℓ + 1) = 0 , (3.60)

with the solutions

ν = ℓ + 1 , −ℓ . (3.61)

25
Since u(0) = 0 then only the first solution is acceptable and this is the standard regular solution for the wave function
vanishing as

u(r) −−−→ r ℓ+1 . (3.62)


r→0

When γ > 2 there is no solution for ν since there is no way to compensate the stronger divergence of the potential
when r → 0. The wave function u(r) does not admit a series expansion in powers of r for r → 0.
For the particular case of γ = 2 all the terms in Eq. (3.59) have to be kept. One has the equation

ν(ν − 1) − α − ℓ(ℓ + 1) = 0 ,
ᾱ ≡α + ℓ(ℓ + 1) ,
ν(ν − 1) − ᾱ = 0 ,
(3.63)

The last equation has two solutions


1 1√
ν= ± 1 + 4ᾱ . (3.64)
2 2
We distinguish several cases:

1. 1 < 1 + 4ᾱ. Repulsive effective potential. There is only one acceptable solution behaving as
1 1√
u(r) −−−→ r 2 + 2 1+4ᾱ
. (3.65)
r→0

because the other solutions for ν in Eq. (3.64) gives rise to a divergent wave function u(r).

2. 1 + 4ᾱ = 1 → ᾱ = 0. In this case the differential equation, Eq. (3.59), reduces to

−r ν−2 ν(ν − 1) = k2 r ν → ν = 0 , 1. (3.66)

In the series expansion of u(r) in powers of r one has two linearly independent solutions, one behaves as a constant
and the other as r for r → 0. The subdominant terms are obtained by applying the recurrence relation that follows
from Eq. (3.66) which reads

−cn+2 (n + 2)(n + 1) = k2 cn . (3.67)

In this way, one independent solution only involves even powers of r and the other only has odd powers of r.

3. 0 < 1 + 4ᾱ < 1. Weakly attractive effective potential. In this case both solutions are acceptable and real, so that
there are two independent solutions. The wave function u(r) behaves as
1 1

u(r) −−−→ r 2 ± 2 1+4ᾱ
. (3.68)
r→0

4. 1 + 4ᾱ < 0. Attractive effective potential. There are two linearly independent complex solutions behaving as
1 1√
u(r) −−−→ r 2 ±i 2 −1−4ᾱ
. (3.69)
r→0

26
5. 1 + 4ᾱ = 0 → ᾱ = −1/4. The differential equation Eq. (3.59) becomes
1
−r ν−2 (ν − )2 = k2 r ν , (3.70)
2
and ν = 1/2 is a double root. So that only one linearly independent solution is found in this way. The other
independent solution is of the form
∞ n
X o
u2 (r) = cn r ν+n log βr + c′n r ν+n . (3.71)
n=0

Substituting this expression in the original differential equation one has the condition for ν
1
−(ν − )2 c0 = 0 ,
 2 
1 5
(ν − ) ( − ν)c0 + c′0 = 0 , (3.72)
2 2

which implies ν = 1/2. One can also deduce the following recurrence relations, the first one stems from the coefficient
multiplying the log term and the second from the rest of terms without logarithm,

−(n + 2)2 cn+2 = k2 cn , (3.73)


k2 ′
2cn+2 + (n + 2)c′n+2 = c . (3.74)
n+2 n
One can impose the condition c′0 = 0. Then Eq. (3.74) allows one to fix all the c′n in terms of the cn , which can be
determined from Eq. (3.73) in terms of the free parameter c0 . Any shift in c′0 is reabsorbed in the linear superposition
with the other linearly independent solution u1 (r). The recurrence relations of Eqs. (3.73) and (3.74) only involve
even powers of n because it can be easily shown that c1 and c′1 must vanish. This can be deduced by considering
the coefficients of the terms proportional to r ν−1 , similarly as we obtained Eq. (3.72) by isolating the coefficients
of the terms with r ν−2 . This is also the case for all the other recurrence relations that follows for γ = 2. Notice
that the parameter β has dimensions of inverse of distance, which cannot be determined by the knowledge of the
potential because α is dimensionless in this case with γ = 2. One needs some extra information e.g. by imposing
the presence of a bound state with a certain binding energy.
One extra condition beyond the knowledge of the potential is then necessary to fix the solutions in the cases 2–5 .
>
Thus, a singular potential occurs for u(r) −−−→ αr γ when γ > 2 and for γ = 2 if 1 + 4ᾱ ≤ 1.
r→0
The conditions a) and b) have a clear meaning in terms of the classical turning point for γ = 2. Because we have the
classical condition L2z + α = 2mr 2 E > 0. If we write L2z = ℓ(ℓ + 1) then it is required that ℓ(ℓ + 1) + α > 0 which
corresponds to the condition a), 1 + 4ᾱ > 1.

3.3 Partial-wave projected potential


In this section we discuss the analytical properties of the potential projected in a given partial wave as a function of two
complex arguments, p1 , p2 . The projected potential as a function of one of the two complex variables is an analytical
function in the corresponding cut-complex plane, with the position of the cuts depending on the other argument. The cuts
that appear in the potential are called dynamical cuts (DCs). The same cuts also appear in the on-shell and half-off-shell
amplitudes, with no generation of extra DCs, as discussed in Sec. 6.

27
• The results that follow are explicitly shown for the case of the 1 S0 OPE potential, though they can be straightforwardly
generalized to any local potential by making use of its spectral decomposition
Z ∞ 2µη(µ2 )
v(p1 , p2 ) = dµ , (3.75)
0 q 2 + µ2

with q = p2 − p1 , the momentum transfer and η(µ2 ) the spectral function. Since the analytical properties of the potential
are encoded in the kinematical configurations that put one-shell the pions exchanged, our results could be applied to
non-local potentials, which fulfilled Eq. (3.75) but with a more involved spectral function that could be expressed in the
P
form i ηi (µ2 , p1 , p2 ), with these spectral functions being entire functions of its arguments, typically polynomials in p21 ,
p22 and p1 · p2 . The latter dependence is required by rotational invariance. This is the case for potentials calculated in
chiral EFT. By the same token, although we consider in the following the simpler S-wave projection, which is denoted as
v(p1 , p2 ), with pi = |pi |, our results are also valid for any other partial wave.
• Partial-wave expansion of the potential:
States with well-defined orbital angular momentum:
Z
1
|ℓm, |p|i = √ dp̂Yℓm (p̂)|pi . (3.76)

This relation can also be inverted so that
√ X ∞ m=ℓ
X
|pi = 4π Yℓm (p̂)|ℓm, |p|i . (3.77)
ℓ=0 m=−ℓ

X
v(q 2 ) = (2ℓ + 1)vℓ (p1 , p2 )Pℓ (p̂1 · p̂2 ) . (3.78)

For the 1 S0 N N potential in S-wave one has,


Z +1  −1
v(p1 , p2 ) = g2 p21 + p22 − 2p1 p2 t + m2π dt , (3.79)
−1

where the coupling g is


!2
gA mπ
g= √ . (3.80)
8fπ

This integration has a cut of logarithmic nature for


1  2 
t= p1 + p22 + m2π , t ∈ [−1, +1] . (3.81)
2p1 p2
Solving for p2 ,
q
p2 = p1 t ± i p21 (1 − t2 ) + m2π , (3.82)

which gives the finite cut in the p2 variable as a function of p1 , with the branch points obtained by substituting in the
previous equation t → ±1. In the following we will refer as dynamical cut (DC), any cut that owes its origin to putting
on-shell (particles) pions exchanged.

28
Figure 4: Imaginary part in the p2 -complex plane of the S-wave projection of the potential in Eq. (3.83). The left panel
is the algebraic expression and the right one corresponds to the numerical calculation. One clearly observes the presence
of the finite curved cuts by the discontinuity in the imaginary part. There are two of them in complex conjugate positions
because p1 is real and then v(p1 , p2 ) satisfies in the variable p2 the Schwarz reflection principle. This also follows from
Eq. (3.82).

• Indeed, Eq. (3.79) can be integrated algebraically, although for complex p1 and p2 one has to be careful because the
imaginary part of the denominator can change sign. The right process is to extract first 2p1 p2 as a common factor in the
denominator and then perform the integration in t,
Z +1
g dt
v(p1 , p2 ) = −
2p1 p2 −1 t − (p21 + p22 + m2π )/(2p1 p2 )
g   p2 + p22 + m2π   p2 + p22 + m2π 
=− log 1 − 1 − log − 1 − 1
2p1 p2 2p1 p2 2p1 p2
g 2
 (p1 + p2 ) + mπ 2
= log . (3.83)
2p1 p2 (p1 − p2 )2 + m2π

Notice that since t is real the imaginary part of the denominator in the integral of the equation above does not change
sign and we do not cross the logarithmic cut in the result. We can combine together the two logs in the second line in
just one because their arguments have the same imaginary part. On the other hand, I have also checked that Eq. (3.83)
is in agreement with numerical computations. One can also plot Eq. (3.83) in the p2 -complex plane and observe explicitly
the emergence of the curved cut of Eq. (3.82). These last two facts are shown in Fig. 4 for g = 1, mπ = 1 and p1 = 1,
where we show the imaginary part of Eq. (3.83), both by employing the algebraic expression (left panel) and performing
the result numerically (right panel).
Next, had we considered n pion exchanges this would have implied that the momentum transfer squared should be
equal to −(nmπ )2 , giving rise to the curved cut in the p2 -complex plane:
q
p2 = p1 t ± i p21 (1 − t2 ) + n2 m2π , t ∈ [−1, +1] . (3.84)

E.g. when iterating the LS equation the value of n rises with the order of the iteration and this originates an infinite
sequence of finite curved cuts. This certainly would make difficult the calculation of the discontinuity across the DCs.
• Indeed, when calculating ∆(A) from the iteration of the LS equation we had to find the position of the cuts in the
off-shell S-wave-projected potential and the situation seemed much easier than what has been discussed up to now. Let

29
us understand next the origin of differences and how one can end with a simpler picture, similar to that already used in
the iteration of the LS for calculating ∆(A).
Let us proceed in a different way by introducing a spectral representation for the potential, Eq. (3.75), so that now
Eq. (3.79) reads
Z Z
+1 ∞ 2µη(µ2 )
v(p1 , p2 ) = g dt dµ . (3.85)
−1 0 p21 + p22 − 2p1 p2 t + µ2
We also exchange the order of integration, which is justified (Fubini’s theorem) when the denominator does not vanish in
the integration domain [−1, 1] ⊗ [mπ , ∞], as it happens for real values of p1 and p2 . We are then left with
Z ∞
g     
v(p1 , p2 ) = dµ2 η(µ2 ) log µ2 + (p1 + p2 )2 − log µ2 + (p1 − p2 )2 . (3.86)
2p1 p2 0

This result, for η(µ2 ) = δ(µ2 − m2π ), is not the same as the expression in Eq. (3.83) because we cannot, in general,
combine the difference of the two logs in Eq. (3.86) as the log of the ratio of the arguments, because the latter could have
imaginary parts of different sign.
• However, we argue that we can legitimately use Eq. (3.86), instead of Eq. (3.83), for the analytical extrapolation of
the S-wave projection of the potential. The argument is the following. When solving the LS equation in the physical case
the same result is obtained by employing Eqs. (3.83) and (3.86), as they both agree on the physical axis, and also around
it. Then, to go into the complex plane of the moduli of the momenta we can proceed by analytical continuation of the
projected potential as given in Eq. (3.86), because it has simpler and more convenient analytical properties. This clearly
exemplify that although the physical limit is the same one can consider different analytical extrapolations in the complex
plane.
The cuts in the p2 -complex plane stemming from Eq. (3.86) are
q
p2 = p1 ± i m2π + x2 ,
q
p2 = −p1 ± i m2π + x2 , (3.87)

with x ∈ R . These cuts are vertical lines that extend up to ∞ starting from the branch points that are given by the
previous equation with x = 0.
• Let us take under consideration the function defined by the difference between the potential functions of Eqs. (3.86)
and (3.83). This function vanishes along the real axis of the p2 -complex plane, and then it should be zero in all its
analytical domain containing the real axis. However, this analytical domain is not the whole p2 -complex plane because it is
cut in such a way that there are two strips, delimited by the curved cuts in Eq. (3.83) and the vertical ones of Eq. (3.86),
so that they are isolated from the real axis. This is explicitly shown in Fig. 5 for the same values of the parameters and p1
as in Fig. 4. In these strips the N N partial wave that would result by solving the LS equation would be different regarding
which form for the projected potential, either Eq. (3.86) or Eq. (3.83), is used. As discussed above, we use in the following
the form given by Eq. (3.86) because it has simpler analytical properties.
• We now discuss the technical point regarding the appropriate form for the analytical continuation of the partial wave
projection of an amplitude where sin θ appears. Then, instead of Eq. (3.79) one has
Z +1 √ Z +1 √
1 − x2 dx 1 1 − x2
I= 2 2 2
= − dx , (3.88)
−1 p1 + p2 − 2p1 p2 x + mπ 2p1 p2 −1 x − ξ
with
p21 + p22 + m2π
ξ= . (3.89)
2p1 p2

30
Figure 5: Real part in the p2 -complex plane of the difference between the two forms for the S-wave projection of the
potential given by Eqs. (3.86) and (3.83). One can clearly see the two strips in which the two forms of the S-wave
projected potential differ.

Making use of Eq. (2.282) of Ref. [1] we can write


Z Z  Z +1
1 +1 xdx ξ +1 dx 1  dx
I= √ + √ − 1 − ξ2 √ . (3.90)
2p1 p2 −1 1−x 2 2p1 p2 −1 1−x 2 2p1 p2 −1 (x − ξ) 1 − x2

For the last integral it is convenient to make the changes of variable t = 1/(−x + ξ) > 0, for real and positive p1 and p2
(in which case ξ > 1). The resulting integral can be integrated following Eq. (2.261) [c < 0, ∆ < 0] of Ref. [1]. The final
result can be written in the form
πξ π q q
I= − (p + p )2 + m2 (p − p )2 + m2 . (3.91)
1 2 π 1 2 π
2p1 p2 (2p1 p2 )2

Here, the cut in the square roots is taken at the left, argz ∈] − π, π] ( −1 ± iε = ±i). In this way the cuts in Eq. (3.91)
are located at
q q
p1 = −ip2 ± i m2π + x2 , (for (p1 + p2 )2 + m2π )
q q
p2 = +ip2 ± i m2π + x2 , (for (p1 − p2 )2 + m2π ) , (3.92)
which are the same vertical lines already found in Eq. (3.87) stemming from the log’s in Eq. (3.86). Thus, Eq. (3.91) is
readily written in the appropriate from to be used in the analytical extrapolation of the LS equation, in the same lines as
already discussed above regarding Eq. (3.86).
Finally, when performing the partial wave projection we will find integrals that by iteration can be easily reduce to a
from corresponding either to Eq. (3.79) or (3.88). E.g.:
Z Z Z Z Z Z
x2 dx
+1 +1 +1 xdx +1 +1
2
+1 dx
= x+ξ = dxx + ξ dx + ξ ,
−1 x − ξ −1 −1 x − ξ −1 −1 −1 x−ξ
Z √ Z +1 p Z +1 √
+1 x 1 − x2 dx 1 − x2 dx
= 2
1 − x dx + ξ , (3.93)
−1 x−ξ −1 −1 x−ξ
and so on.

31
4 Right-hand cut
The kernel of the T -matrix operator is t(k, k′ ; z), where k is the outgoing CM three-momentum, k′ is the incoming one
and z is the energy variable. In general z is taken as a complex variable. For on-shell scattering k′ = k and z = k2 /2µ (µ
here is the reduced mass of the two-particle system. Later we will set it as m/2 when studying nucleon-nucleon scattering,
with m the nucleon mass, but this will be announced. In addition, once we introduced partial-wave amplitudes, we will
shift the meaning of k, k′ so that they just refer to the corresponding moduli.)
The kernel for the S-matrix operator is

S(k, k′ , E) = δ(k − k′ ) − i 2πδ(E − E ′ )t(k, k′ , E + i0) , (4.1)

E = k2 /2µ and E ′ = k′ 2 /2µ.


From basic properties of the LS equation and the hermiticity of the potential operator it follows that t(k, k′ ; z) satisfies
this generalized unitarity relation,
Z
1
t(k, k′ ; z1 ) − t(k, k′ ; z2 ) = (z2 − z1 ) t(k, k′′ ; z1 ) t(k′′ , k′ ; z2 ) dk′′ . (4.2)
(k′′ 2 /2µ − z1 )(k′′ 2 /2µ − z2 )

The on-shell scattering unitarity relation is a particular case of the previous equation by taking z1 = k2 /2µ + iε and
z2 = k2 /2µ − iε, ε → 0+ , and it reads
Z
k2 k2 k2 k2
t(k, k′ , + i0+ ) − t(k, k′ , − i0+ ) = −i2πµ|k| dΩk′′ t(k, |k|k̂′′ , + i0+ ) t(k′ , |k|k̂′′ , + i0+ )∗ . (4.3)
2µ 2µ 2µ 2µ

This equation is consistent with the unitarity character of the S-matrix operator, Eq. (4.1).
Let us now consider hfs T -matrix elements and take z1 = k′ 2 /2µ + iε and z2 = k′ 2 /2µ − iε in Eq. (4.2). We end with
the result,
Z
k′ 2 k′ 2 k′ 2 k′ 2
t(k, k′ ; + i0+ ) − t(k, k′ ; − i0+ ) = −2iµπ|k′ | dΩk′′ t(k, |k′ |k̂′′ ; + i0+ ) t(k′ , |k′ |k̂′′ ; + i0+ )∗ . (4.4)
2µ 2µ 2µ 2µ
Note that the second T -matrix elements on the rhs of the equation is on-shell.
• The Eqs. (4.3) and (4.4) can be seen as a result of stating that the discontinuity along the RHC of a partial wave
amplitude stems from the inclusion of a resolution of the identity between the product (2π)2 tt† , with the available energy
for the intermediate states given by the external energy z, as
Z Z
′′ ′′ ′ ′′ ∗k′′ 2
dk t(k, k ; z)t(k , k ; z) δ( − z) = k̄µ dΩk′′ t(k, k̄ k̂′′ ; z)t(k′ , k̄ k̂′′ ; z)∗ , (4.5)


with k̄ = 2µz, z > 0. This expression is then multiplied by −i2π to end with the appropriate numerical factors. This is
then a straightforward continuation of the result for the on-shell case driven by unitarity of the S-matrix,

t − t† = −i2πtt† , z > 0 . (4.6)

In other words, during this work we are using the LS equation to fix the analytical properties of a partial wave amplitude
and evaluate its discontinuities across the cuts required.

32
5 Partial wave projection
Because time-reversal and parity invariance we have that

t(k, k′ ; z) = t(k′ , k; z) . (5.1)

NOTATION: From here on k and k′ denote the moduli of the related three-momenta.
We can now perform a partial-wave expansion for the hfs T -matrix elements,2

k′ 2 X 2ℓ + 1 k′ 2
t(k, k′ ; )= tℓ (k, k′ ; )Pℓ (k̂ · k̂′ ) . (5.2)
2µ ℓ
2 2µ

Here for simplicity we are disregarding spin. But the final conclusions are the same as if we had kept full account of the
spin of the particles. The symmetry relation of Eq. (5.1) at the level of the partial-wave amplitudes implies that

k′ 2 k′ 2
tℓ (k, k′ ; ) = tℓ (k′ , k; ). (5.3)
2µ 2µ

By introducing the partial-wave decomposition Eq. (5.2) into the hfs unitarity relation Eq. (4.4), we end up with

k′ 2 k′ 2 k′ 2 k′ 2
tℓ (k, k′ ; + iε) − tℓ (k, k′ ; − iε) = −4iπ 2 µk′ tℓ (k, k′ ; + i0) tℓ (k′ , k′ ; + i0)∗ . (5.4)
2µ 2µ 2µ 2µ
′2
Note that with respect to the hand-written notes I have redefined the partial-wave amplitude tℓ (k, k′ ; k2µ ) in the notes
′2
by (2ℓ + 1)tℓ (k, k′ ; k2µ ), so that the annoying factor (2ℓ + 1) disappears now in the unitarity relation Eq. (5.4), as it is
customary in N N scattering.
• In the works on N N scattering by applying the N/D method I have used other normalization for the unitarity relation,
the one that is used in Sec. 9. The relationship is given by the factor 1/4π 3 , t(k, k′ ; k′ 2 /m) → t(k, k′ ; k′ 2 /m)/4π 3 .

5.1 Schwarz reflection principle


NOTATION: From here on m refers to the nucleon mass and µ = m/2.
′2
• Let us show that tℓ (k, k′ ; k2µ )/(kℓ k′ ℓ ) is indeed a function of k and k′ squared. This is clear for the potential
Z +1
′ dxPℓ (x)
vℓ (k, k ) = λ 2 . (5.5)
−1 k ′ + − 2kk′ x + m2π
k2

If we exchange k or k′ to −k or −k′ (but not both simultaneously, which trivially leaves vℓ invariant) the potential changes
by a factor (−1)ℓ , because we can change √ the sign √ of the integration variable x. Hence, vℓ (k, k′ )/(kℓ k′ ℓ ) is left invariant
and the squared root ambiguity in k = k2 (k′ = k′ 2 ) does not play a role. As a result, vℓ (k, k′ )/(kℓ k′ ℓ ) is a function
of the moduli squared of k and k′ . √ √
• In our analytical extension of the scattering amplitudes, k2 and k′ 2 have their cut on the left, for negative values

of their arguments. However, when taking the square root of the energy, z, one has the cut on the right, for positive
values of the argument of the square root. This is convenient to have a clear separation between RHC and LHC, which
plays a role e.g. in order to arrive to the fulfillment of the Schwarz reflection principle of our partial waves.
2
The same can be done for full-off-shell scattering amplitudes.

33
• Now the LS equation
Z ∞
′ m ′ vℓ (k, p1 )
tℓ (k, k ; z) = vℓ (k, k ) + 2 dp1 p21 tℓ (p1 , k′ ; z) . (5.6)
2π 0 p21 − z

It is clear that if we exchange the sign of k the rhs of the equation does change by a factor (−1)ℓ and then

tℓ (−k, k′ ; z) = (−1)ℓ tℓ (k, k′ ; z) . (5.7)

Had we changed the sign of k′ the resulting T -matrix element, tℓ (k, −k′ ; z) satisfies the same LS equation as (−1)ℓ tℓ (k, k′ ; z),
and since the solution is supposed to be unique, the T -matrix element is also the same. Hence, we conclude that
tℓ (k, k′ ; z)/(kℓ k′ ℓ ) is a function of the moduli squared of the CM three-momenta. This can also be easily seen by consid-
ering the Neumann series of the LS equation.
The derivation of Eq. (5.7) for complex k and k′ , and the analogous one for the change of sign in k′ , is given below
starting in Eq. (7.7).
• The RHC, Eq. (5.4), extends for k′ 2 > 0 and since for some interval of negative values of k′ 2 and real ones for k,
the reduced potential vℓ (k, k′ )/(kℓ k′ ℓ ) is real then the very same reduced partial wave t(k, k′ ; k′ 2 /m)/(kℓ k′ ℓ ) would be
also real for the same pvalues of the√ arguments. Thus, the Schwarz reflection principle would be satisfied for both reduced
functions and since k ±√iε = k′ 2 (the cut in the square root is on the left), we have for such conditions that along
′ 2

the RHC (k′ 2 > 0 → k′ = k′ 2 > 0),


 ∗
′2 ′2
tℓ (k, k′ ; km + iε) t (k, k′ ; km − iε)
=  ℓ  . (5.8)
kkℓ ′ℓ ℓ ′ℓ
kk
√ √
As ( k2 k′ 2 )ℓ is real for positive k and k′ we can factor out it in the previous relation and simply write

k′ 2 k′ 2
tℓ (k, k′ ; + iε) = tℓ (k, k′ ; − iε)∗ , (5.9)
m m
for positive k′ and k.3 In this way, we can rewrite the hfs unitarity relation Eq. (5.4) as

k′ 2 k′ 2 k′ 2
Imtℓ (k, k′ ; + iε) = −2π 2 mk′ tℓ (k, k′ ; + i0) tℓ (k′ , k′ ; + i0)∗ . (5.10)
m m m
As a result, we have an analogous situation to the Watson’s final-state interaction theorem and then from Eq. (5.10) it
follows that tℓ (k, k′ ; k′ 2 /m + iε) has the same phase as the on-shell partial wave tℓ (k′ , k′ ; k′ 2 /m + iε). Since in an N/D
representation of the latter the phase along the RHC is determined by the inverse of the function Dℓ (k′ 2 + iε) we then
have that

k′ 2 2
tℓ (k, k′ ; + iε) Dℓ (k′ + iε) (5.11)
m
has no RHC. As a result we can write

k′ 2 Nℓ (k, k′ )
tℓ (k, k′ ; )= , (5.12)
m Dℓ (k′ 2 )
3
The previous equation also follows from the definition of the partial wave amplitudes, Eq. (5.2), and the basic fact that t(k, k′ ; z ∗ ) =
t(k , k; z)∗ (lemma 2.3 of [2]), together with time-reversal invariance t(k, k′ ; z) = t(k′ , k; z), Eq. (5.1).

34
with k, k′ > 0 and Dℓ (k′ 2 ) determined from the study of on-shell scattering. Let us recall that by the procedure established
to determine the potential from the hfs partial-wave amplitude only the latter is required to be known for k, k′ > 0.
It is indeed not necessary to determine Dℓ (k′ 2 ) to be used in Eq. (5.11) by solving the N/D method, which then
allows us to isolate Nℓ (k, k′ ) that has no RHC. In this way one can write down a a simpler DR for this function than for
′2
tℓ (k, k′ ; km ), that would indeed require to solve a linear IE to determine the latter. Instead of Dℓ (k′ 2 )−1 one could also
use an Omnès function Ωℓ (k′ 2 ),
Z !
′2 k′ 2 ∞
2 δℓ (q 2 )
Ωℓ (k ) = exp dq 2 2 , (5.13)
π 0 q (q − k′ 2 )
′2
that only has RHC and which phase along it, δℓ (k′ 2 ), is the same as that of the on-shell T -matrix element tℓ (k′ , k′ ; km ).
In this way, once the latter is known by some method, we can define

k′ 2 2
NΩ (k, k′ ) = tℓ (k, k′ ; )/Ωℓ (k′ ) (5.14)
m
that has no RHC either and could be used instead of Nℓ (k, k′ ) above.

6 Dynamical cuts in the half-off-shell amplitude


• We start by considering the half-off-shell (hfs) T -matrix t(k, k′ ; k′ 2 /m) as a function k and take the LS equation in
partial waves as our guide principle to discuss its analytical properties,
Z
′ k′ 2 m dp1 p21 ′ k
′2
t(k, k ; ) = v(k, k′ ) + 2 v(k, p 1 )t(p 1 , k ; ), (6.1)
m 2π C p21 − k′ 2 m

with C the integration contour, that now we discuss how to fix it. For real and positive k and k′ , the contour C consists
of the positive real semi axis, that we can denote as its original extension. This is the starting point for discussing the
analytical extrapolation of the LS equation, Eq. (6.1). Taking into account that the cuts of v(k, p1 ) are located at
q
p1 = +k ± i m2π + x2 ,
q
p1 = −k ± i m2π + x2 , (6.2)

according to Eq. (3.87), it follows then that for complex k with |Imk| < mπ the cuts in Eq. (6.2) do not cross the original
integration contour and hence do not induce any deformation in the contour. Contrarily, for |Imk| > mπ one needs to
deform the integration contour by circumventing the vertical line from |Rek| up to |Rek| ± i(|Imk| − mπ ), namely,

p1 = |Rek| ± iλ(|Imk| − mπ ) , |Imk′ | > mπ , λ ∈ [0, 1] . (6.3)

The sign ± in the previous equation is determined according to the possibilities:

|Imk| > mπ
Rek > 0, ± ↔ +sign(Imk)
Rek < 0, ± ↔ −sign(Imk) (6.4)

35
Regarding the deformation in the contour induced by t(p1 , k′ ; k′ 2 /m) in Eq. (6.1) we proceed by iterating the LS equation
in the Neumann series,
Z   Z Z
k′ 2 m dp1 p21 m 2 dp1 p21 dp2 p22
t(k, k′ ; ) = v(k, k′ ) + 2 v(k, p 1 )v(p 1 , k ′
) + v(k, p 1 ) v(p1 , p2 )v(p2 , k′ )
m 2π p21 − k′ 2 2π 2 p21 − k′ 2 p22 − k′ 2
  Z Z Z
m 3 dp1 p21 dp2 p22 dp3 p23
+ v(k, p 1 ) v(p ,
1 2p ) v(p2 , p3 )v(p3 , k′ ) + . . . (6.5)
2π 2 p21 − k′ 2 p22 − k′ 2 p23 − k′ 2
The induced deformations in the integration contour can be easily realized by reading the terms in the previous equation
from left to right (the ordering in which the integration limits are fixed for every integral in a multiple integral). In this
way we have for |Imk| > mπ and p1 , . . . , pn a set of vertical segments in the deformed contours for each pi of decreasing
extent |Impi−1 | − mπ (according to the rules Eq. (6.4) applied to pi ) until (n + 1)mπ > |Imk|. Then, the next v factor,
v(p1 , p2 ), would induce deformation in the integration contour for p2 if |Imp1 | > mπ , and so on. In addition the rightmost
v factor in Eq. (6.5), v(pm , k′ ) for every m-time iterated contribution, induces for |Imk′ | > mπ an additional vertical
segment in the last integration variable pm , according to Eq. (6.4). For illustration we plot in Fig. 6 a sketch of the set
of deformations to be considered for k and k′ in the first quadrant of their respective complex planes, with δ1 > δ2 > 0,
etc. All the finite set of vertical additions in the original contour due to v(k, p1 )t(p1 , k′ ; k′ 2 /m) coalesces just in one at
|Rek| ± i(|Imk| − mπ ) and another one at |Rek′ | ± i(|Imk′ | − mπ ), according to the rules in Eq. (6.4). This discussion
fixes the contour C of integration in Eq. (6.1).
• The first term in Eq. (6.1), v(k, k′ ), establishes the cuts in the k-complex plane at (3.87)
q
k = k′ ± i m2π + x2 ,
q
k = −k′ ± i m2π + x2 . (6.6)

Let us show now that the partial wave t(k, k′ ; k′ 2 /m) is a regular function in the cut k complex plane with cuts located
according to Eq. (6.6). Precisely the intervals of values in Eq. (6.6) determine the deformation of the integration contour
due to the factor t(p1 , k′ ; k′ 2 /m) and those in Eq. (6.2) the possible deformation because of v(k, p1 ). As a result the
integrand in Eq. (6.1) is regular in k.
No more cuts than those in Eq. (6.6) arise by iterating the LS equation, Eq. (6.5). We could have the following four
cases:
i) Both |Imk| and |Imk′ | are less than mπ , so that the integration contour is the original one. In this case the arguments
of the log’s present in v(pi , pi+1 ), with p0 = k and pn+1 = k′ , have always positive real part and there is no discontinuity.
ii) |Imk| > mπ and |Imk′ | < mπ : The simplest way to proceed is to impose also that |Imk′ | > 0. Then, the energy
denominator does not vanish along the original integration contour, but it would vanish along the vertical addition at |Rek|
if ±k′ places along it (this is analyzed below). In the following the n-times iteration of the potential v(k, k′ ) is denoted
′2
by tn (k, k′ ; km ) and correspond to, Eq. (6.5),
 n Z Z Z Z
k′ 2 m dp1 p21 dp2 p22 dpn p2n
tn (k, k′ ; )= v(k, p1 ) v(p1 , p2 ) ... v(pn−1 , pn )v(pn , k′ ) . (6.7)
m 2π 2 p21 − k′ 2 p22 − k′ 2 p2n − k′ 2

The integration contour is the original one plus one vertical addition at |Rek|, which corresponds to the situation in
Fig. 7. We first show that for n = 1 there are no other cuts in the variable k beyond those given in Eq. (6.6). The
factor v(k, p1 ) induces the added vertical contour at |Rek|. Inpthe integral of p1 the integration contour goes around
the vertical addition but the variable p1 has cuts at (±k′ ) ± i m2π + x2 (the relevant ones indicated by dashed lines
in Fig. 7) because the next factor v(p1 , k′ ). When k varies the added integration contour moves correspondingly so
that only when it overlaps with one of these vertical cuts in p1 at |Rek′ | we have discontinuity in k. If the added

36
Re k’ Re k C

Once iterated

Re k’ Re k −δ1 Re k +δ1 C

Twice iterated

Re k’ Re k −δ1−δ2 Re k −δ1+δ2 Re k +δ1−δ 2 Re k +δ1+δ 2 C

Three-time
iterated

Figure 6: Deformed integration contours in the pi -complex planes for the Neumann series of Eq. (6.5). It is assumed that k and k ′
belong to the first quadrant of their corresponding complex planes and that Imk ′ > mπ and Imk > 2mπ . In addition, δ1 > δ2 .

37
|Im k|−m π |Im k|−m π

k’+i mπ k’+i m π

k’

|Re k| C k’ |Re k| C
k’−i mπ
k’−i mπ

k’+i mπ k’+i m π

k’ |Re k|
|Re k|
C k’ C
k’−i mπ

k’−i m π

−|Im k|+m π −|Im k|+m π

Figure 7: Interplay between the added vertical contours and cuts associated with external variables for case ii) that corresponds to
|Imk| > mπ and |Imk ′ | < mπ . The different possibilities for the signs of Imk are explicitly plotted.

contour at |Rek| extends from 0 up to |Imk| − mπ , first two panels in Fig. 7, the condition for the referred overlapping
implies that sgn(Rek′ )Imk′ + p mπ < |Imk| − mπ (apart from the obvious one |Rek| = |Rek′ |). Notice that since
′ ′
|Imk | < mπ the cut |Imk | − m2π + x2 involves always negative values and does not overlap with the added positive
vertical contour. Then, |Imk| > sgn(Rek)Imk′ + 2mπ and this set of possible values for k belongs to the cuts indicated in
Eq. (6.6). If the added contour extends from −|Imk| + mπ up to 0, we proceed analogously and the condition now is that
sgn(Rek′ )Imk′ − mπ > −|Imk| + mπ so that |Imk| > −sgn(Rek′ )Imk′ + 2mπ and again this set of values for k belongs
to the cuts in Eq. (6.6). All the different possibilities are plotted in Fig. 7, where the added contour at |Rek| follows the
rules in Eqs. (6.3) and (6.4). There is no discontinuity
p in k stemming from the original integration contour because its
intersection with the pertinent cut from (±k) ± i mπ + x2 reduces to a point, and the discontinuity in the v ′ s is just
2

of logarithmic nature. XXX One should stress that this intersection is already avoided by the deformation of the
contour and this region is part of the added vertical piece.
′2
We now consider the generic n-times iterated amplitude tn (k, k′ ; km ), that can also be written as
Z
k′ 2 m dpn p2n k′ 2
tn (k, k′ ; )= 2 t n−1 (k, p n ; )v(pn , k′ ) , (6.8)
m 2π C p2n − k′ 2 m
where the integration contour is the one already fixed in Eq. (6.7). Then, when pn goes around the vertical integration
contour at |Rek| there is a discontinuity in k when this addition intersects the cuts in the next factor v(pn , k′ ), in the
same way as discussed for n = 1. As a result, the cuts in k are contained in Eq. (6.6).
The case of real k′ can be seen as a particular case of the previous discussion by taking the limit Imk′ → 0± because
this limits exists (cf. Sec. 4). Then no new cuts in k arise that those already discussed for Imk′ 6= 0.

38
Nonetheless, we can also discuss explicitly the case of real k′ . The energy denominator in Eq. (6.7) for n = 1 gives
rise to a Dirac-delta function δ(p1 2 − k′ 2 ) that is multiplied by v(k, k′ )v(k′ , k′ ) and the first factor v(k, k′ ) originates the
cuts in Eq. (6.6). We can also apply the idea of the intersection of the cuts in k with the original integration contour to
conclude that in this case the only point of intersection along the real axis at |Rek| gives rise to a discontinuity because
the integrand contains a Dirac-delta function located at k′ 2 . For the nth iteration the vanishing of the energy denominator
in Eq. (6.8) would imply the analogous factor tn−1 (k, k′ ; k′ 2 /m)v(k′ , k′ ) with cuts in k according to Eq. (6.6).

k’+i mπ |Im k|−m π |Im k|−mπ

k’

k’−i mπ

|Re k| C k’+i mπ |Re k| C

k’

k’−i mπ

k’+i mπ

k’

k’−i mπ |Re k|
|Re k|
C k’+i mπ C

k’

k’−i mπ
−|Im k|+m π −|Im k|+m π

Figure 8: Interplay between the added vertical contours and cuts associated with external variables for case iv) that corresponds to
|Imk| > mπ and |Imk ′ | > mπ . The different possibilities for the signs of Imk are explicitly plotted.

iii) |Imk| < mπ and |Imk′ | > mπ : In order to follow an argument in close analogy with ii) we make use of Eq. (5.1) and
write t(k, k′ ; k′ 2 /m) = t(k′ , k; k′ 2 /m). In this way we can apply the same argument as in ii) by just exchanging k ↔ k′ .
As a result, when considering the n-times iterated amplitude tn (k′ , k; k′ 2 /m), that we write it now as
Z
′ k′ 2 ′ k′ 2 m dpn p2n ′ k′ 2
tn (k, k ; ) = tn (k , k; )= 2 t n−1 (k , p n ; )v(pn , k) , (6.9)
m m 2π C p2n − k′ 2 m
p
there is a cut in k when the vertical addition at |Rek′ | overlaps with the cuts in pn located at (±)k ± i m2π + x2 . When
k varies these cuts move correspondingly so that only when one of them overlaps with the added integration contour we
have discontinuity in k.
We can then follow the same arguments as in ii) with the roles of k and k′ exchanged, in particular in Fig. 7, to
arrive to this conclusion: If the added contour at |Rek′ | extends between 0 up to |Imk′ | − mπ , first two panels in Fig. 7,

39
the condition for the referred overlapping implies that sgn(Rek)Imk p + mπ < |Imk′ | − mπ (apart from the obvious one
|Rek| = |Rek′ |). Notice that since |Imk| < mπ the cut |Imk| − m2π + x2 involves always negative values and does not
overlap with the added positive vertical contour. Then, sgn(Rek)Imk < |Imk′ | − 2mπ and all this set of possible values
for k belongs to the cuts indicated in Eq. (6.6). If it extends from −|Imk′ | + mπ up to 0, we proceed analogously and the
condition now is that sgn(Rek)Imk − mπ > −|Imk′ | + mπ so that sgn(Rek)Imk| − |Imk′ | + 2mπ and again this set of
values for k belongs to the cuts in Eq. (6.6). All the different possibilities are plotted in Fig. 7, where the added contour
at |Rek′ | follows the rules in Eqs. (6.3) and (6.4). There is no discontinuity
p in k stemming from the original integration
contour because its intersection and the pertinent cut from (±k′ ) ± i m2π + x2 reduces to a point, and the discontinuity
in the v ′ s is just of logarithmic nature.
iv) |Imk| > mπ and |Imk′ | > mπ : There are vertical contours added with bases at |Rek| and |Rek′ |, as shown in Fig. 8,
when calculating tn (k, k′ ; k′ 2 /m) given in Eq. (6.8), with the factor tn−1 (k, pn ; k′ 2 /m)v(pn , k′ ) mixing both additions.
Since Imk′ 6= 0 the energy denominator does not vanish along the original integration contour, but it would vanish along
the vertical addition at |Rek| if ±k′ places along it (this is analyzed below). We can follow similar steps as in case iii), and
when pn goes around the vertical integration contour at |Rek| there is a discontinuity in k when this addition intersects the
cuts in the next factor v(pn , k′ ). However, as a novelty here compared to iii), the intersection referred should involve only
those cuts that are not avoided by the deformation of the integration contour at |Rek′ |. This gives rise to an interesting
analysis by requiring that the part of the cut in pn that is not part of the added integration contour at |Rek′ | overlaps with
the added integration contour at |Rek|. By definiteness let us assume that |Imk| > |Imk′ | which corresponds to Fig. 8
(if this is not the case we would repeat the analysis exchanging k ↔ k′ ). If the contour at |Rek| extends from 0 up to
|Imk| − mπ (two upper panels in Fig. 8), then the requirement is

sgn(Rek′ )Imk′ + mπ < |Imk| − mπ , (6.10)

so that

|Imk| > sgn(Rek′ )Imk′ + 2mπ , (6.11)

and then this set of possible values for k belongs to the cuts indicated in Eq. (6.6). Similarly, for the two lower panels in
Fig. 8 that correspond to an added vertical cut at |Rek| from −|Imk| + mπ up to 0 the required condition is

sgn(Rek′ )Imk′ − mπ > −|Imk| + mπ , (6.12)

which translates into

|Imk| > −sgn(Rek′ )Imk′ + 2mπ , (6.13)

and the same conclusion follows.


Along the added vertical segment up to ±i(|Imk′ | − mπ ), with |Imk′ | > mπ , the energy denominator does not vanish.
However, further discussion is needed along the added vertical contour at |Rek| when |Imk| > mπ , because then it is
not forbidden that p21 − k′ 2 could vanish (which requires |Imk| − mπ > |Imk′ | and |Rek| = |Rek′ |). Let us assume that
this is the case, and to fix ideas we take that k and k′ are in their respective first quadrant, since the discussion can be
straightforwardly extended to any other location of k and k′ in their complex planes. Then

p1 = Rek + iν1 ± δ , δ → 0+
0 < ν1 < Imk − mπ ,
(6.14)

where δ is included to circumvent the vertical cut, which also avoids the mentioned zero along the cut at
2
(Rek + iν1 )2 − k′ = (Rek + iν1 − k′ )(Rek + iν1 + k′ ) , (6.15)

40
a) b)

Figure 9: Integration contours in Eq. (6.1) around the pole from the energy denominator.

that requires

k′ = Rek + iν1 , (6.16)

because k and k′ have positive real and imaginary parts by assumption. This point can be reached from the analytical
continuation in k′ from just above the real axis at Rek′ by increasing the Imk′ up to its final value. The previous equation
can also be rewritten as
q

k = k − i m2π + y 2 , (6.17)

for real y such that Imk′ > 0. This implies that


q
k = k′ + i m2π + y 2 (6.18)

and this is a value of k that belongs to the cuts of v(k, k′ ) in the k-complex plane according to Eq. (6.6).
Let us show that in this case the integral for t(k, k′ ; k′ 2 /m) is well defined, so that this extra contribution will not spoil
the standard treatment of a cut in DRs. Around the pole position we have that the vertical segments of integration on the
left and right of the pole are equivalent to semicircles of infinitesimal radius run in clockwise direction. This is plotted in
Fig. 9. The integration around both infinitesimal semicircles can be done straightforwardly with the change of integration
variable

z ≡ reiφ = p1 − k′ . (6.19)

Then, for the semicircle to the left of the pole, Fig. 6.5a), φ ∈ [− π2 , π2 ] and we have the result
Z +π ′2
m 2 (k′ + z)2 ′ ′ ′ k
Ia = 2 idφz v(k, z + k )t(z + k , k ; )
2π −π
2
z(z + 2k′ ) m
m k′ ′2
′ ′ ′ k
= iπ v(k, k − r)t(k − r, k ; ) (6.20)
2π 2 2 m
and for the semicircle to the right, Fig. 6.5b), the phase φ ∈ [π/2, −π/2] and the integral reads

m k′ ′2
′ ′ ′ k
Ib = −iπ v(k, k + r)t(k + r, k ; ). (6.21)
2π 2 2 m

41
Summing both

m k′  ′2 ′2 
′ ′ ′ k ′ ′ ′ k
Ia + Ib = iπ v(k, k − r)t(k − r, k ; ) − v(k, k + r)t(k + r, k ; ) . (6.22)
2π 2 2 m m
This equation explicitly shows that the result is regular in k except for cuts located according to Eq. (6.6).
Now, if k′ is such that we do not need to deform the integration contour it is clear that t(k′ ± r, k′ ; k′ 2 /m) is continuous
in r → 0+ , since we are just changing v(k′ ± r, p1 ) in Eq. (6.1), while the energy denominator stays fixed, and this function
is regular in r for p1 real and positive, because we did not need to modify the integration contour.
However, if the value of k′ requires to modify the integration contour then it is not longer true that v(k′ ± r, p1 ) is
regular in r → 0+ , but we can take into account explicitly the new additions in the integration contour, which consists of
two vertical segments of height Imk′ − mπ , located at Rek′ and Rek′ ± r for t(k′ ± r, k′ ; k′ 2 /m), in order, similarly as
those shown in Fig. 6. Notice that along such vertical segments the energy denominator does not vanish in this case for
t(k′ ± r, k′ ; k′ 2 /m), and then its evaluation follows the standard procedure without pole from the energy denominator. We
also know, because of the analysis already performed in this section, that this function of r has no extra cuts or singularities
than those required by v(k′ ± r, k′ ).
• We can work out easily the discontinuity of t(k′ ± r, k′ ; k′ 2 /m). Circumventing the vertical segments by parallel lines
displaced a distance 0 < δ′ < r to the left and right of the vertical segments, we have (notice that Rek′ = 0 because we
are considering the discontinuity for an on-shell PWA, that is, along the LHC)

k′ 2 k′ 2
t(k′ + r, k′ ; ) − t(k′ − r, k′ ; ) = v(k′ + r, k′ ) − v(k′ − r, k′ )
m m
Z Imk′ −mπ ′2 ′2 
dν1 ν12  ′ ′  ′ ′ ′ k ′ ′ ′ k
−i v(k + r, Rek + iν 1 ) t(Rek − δ + iν 1 , k ; ) − t(Rek + δ + iν 1 , k ; )
0 (Rek′ + iν1 )2 − k′ 2 m m
  k′ 2
+ v(k′ + r, Rek′ + r − δ′ + iν1 ) − v(k′ + r, Rek′ + r + δ′ + iν1 ) t(Rek′ + r + iν1 , k′ ; )
m
Z Imk ′ −mπ ′2 ′2 
dν1 ν12  ′ ′  ′ ′ ′ k ′ ′ ′ k
+i v(k − r, Rek + iν 1 ) t(Rek − δ + iν 1 , k ; ) − t(Rek + δ + iν 1 , k ; )
0 (Rek′ + iν1 )2 − k′ 2 m m
  k′ 2
+ v(k′ − r, Rek′ − r − δ′ + iν1 ) − v(k′ − r, Rek′ − r + δ′ + iν1 ) t(Rek′ − r + iν1 , k′ ; ) , (6.23)
m
where we have taken into account that in the first and third integral terms v(k′ ± r, Rek′ + iν1 ) is located to the right
and left of the added contour at Rek′ , respectively, and the same applies to t(Rek′ ± r + iν1 , k′ ; k′ 2 /m) at the second and
fourth integral terms. The two integrals in the previous equation can be grouped together by taking into account that the
discontinuity of the function v is itself continuous in r,

lim+ v(k′ + r, Rek′ + r − δ′ + iν1 ) − v(k′ + r, Rek′ + r + δ′ + iν1 )
r→0

= lim v(k′ − r, Rek′ − r − δ′ + iν1 ) − v(k′ − r, Rek′ − r + δ′ + iν1 ) . (6.24)
r→0+

42
Then we are left with
k′ 2 k′ 2
t(k′ + r, k′ ; ) − t(k′ − r, k′ ; ) = v(k′ + r, k′ ) − v(k′ − r, k′ )
m m
Z Imk′ −mπ
dν1 ν12  ′ ′ ′ ′ 
−i ′ ′ 2 v(k + r, Rek + iν1 ) − v(k − r, Rek + iν1 )
0 (Rek + iν1 ) − k
 k′ 2 k′ 2 
× t(Rek′ − δ′ + iν1 , k′ ; ) − t(Rek′ + δ′ + iν1 , k′ ; )

m m 
+ v(k′ + r, Rek′ + r − δ′ + iν1 ) − v(k′ + r, Rek′ + r + δ′ + iν1 )
 k′ 2 k′ 2 
) − t(Rek′ − r + iν1 , k′ ;
× t(Rek′ + r + iν1 , k′ ; ) , (6.25)
m m
and the terms in the integrand sum up to zero, so that
k′ 2 k′ 2
t(k′ + r, k′ ; ) − t(k′ − r, k′ ; ) = v(k′ + r, k′ ) − v(k′ − r, k′ ) . (6.26)
m m
• The dynamical cuts of t(k, k′ ; k′ 2 /m) as a function of k′ can be described analogously as done above for k but
exchanging k ↔ k′ since t(k, k′ ; k′ 2 /m) = t(k′ , k; k′ 2 /m), Eq. (5.3).4 Of course, the energy is always k′ 2 /m and varies
with k′ but the energy denominator in the LS has not played any important role in determining the position of the dynamical
cuts in the k-complex plane, so that it will not play it either when consider such cuts in the k′ -complex plane (exchange k
by k′ in the discussion above on regards the possible cancellation of the energy denominator in the added vertical segment
in p1 at Rek). p
As a function of k′ the cuts in Eq. (6.22) correspond p to those expected because k ′ = k − i m2 + y 2 , cf. Eq. (6.17).
π
Explicitly, in t(k′ , k′ ; k′ 2 /m) the cuts occur at ± 2i m2πp+ x2 , with the minus sign excluded
p because we have assumed
that Imk′ > 0.pThose in v(k,pk′ ) occur at k′ = ±(k ± i m2π + x2 ), but since k′ = k − i m2π + y 2 it follows then that
k′ = ±(k′ + i m2π + y 2 ± i m 2 + x2 ). We have to select the minus sign in front of k ′ (to end with an equation to
π p p
′ ′ i 
find k ) and obtain that k = 2 − m2π + y 2 ∓ m2π + x2 . The minus p sign in front
p of the second

square root can be
′ ′ i 2 2 2 2
excluded because Imk > 0 was assumed, so that we have, k = 2 −p mπ + y + mπ + x with x > y. Note that
this set of values for k′ comprises the cut for the on-shell scattering, 2i m2π + x2 .
In summary, no extra DCs appear beyond those already present in v(k, k′ ) as a function of k′ for a given k. Of course,
in addition to the DCs, one also has in this case the RHC that happens for k′ real (k′ 2 > 0), as considered above and
discussed in Sec. 1.4.
Comments for dummies
1. We have first determined the integration contour C for given external k and k′ .
2. Then we have studied the analytical properties of t(k, k′ ; k′ 2 /m) with k as a variable (k′ only enters parametrically).
3. As a function of k the energy pole at k′2 could happen if it lies along the added vertical contour at |ℜk|. It contributes
to the discontinuity, for an explicit analysis of the contribution of the energy pole cf. Sec. 11. But it does for values
of k along the standard position of the cuts in this variable.
4. We have then derived the analytical properties of t(k, k′ ; k′ 2 /m) as a function k′ making use of the relation
t(k, k′ ; k′ 2 /m) = t(k′ , k; k′ 2 /m).
5. In order to make this last derivation more clear one has to keep in mind that when k′ changes so it does the added
integration contour at |Rek′ | (in case that |Imk′ | > mπ ), while the original integration contour and the one at |Rek|
(if |Imk| > mπ ) do not change.
2 2
4
For coupled channels the symmetry relation is tij (k, k′ ; k′ /m) = tji (k′ , k; k′ /m).

43
6. In the process of changing k′ we have the standard analysis (as done for k as variable) to conclude the position of
the standard cuts in k′ (analogous to those determined for k).

7. Regarding the energy pole, along the original contour it gives rise the RHC and along the added contour at |Rek′ |
it never happens.

8. Along the added integration contour at |Rek| the energy pole gives its own contribution to the discontinuity but its
position follows the standard rule for the location of the cuts in k′ .

• We have then shown that if k is not located along the cuts of v(k, k′ ) the kernel for the integral equation corresponding
to the LS equation, Eq. (6.1), is regular in k. These values are then connected in a continuous and smooth way to the real
positive axis, so that one would expect that the solution of the LS were unique for real and positive k′ . In that case we
know from Ref. [2] that t(k, k′ ; z) has only bound states for negative values of z in the case of smooth square-integrable
local potentials v(q) (Hölder function).

7 Calculation of the on-shell discontinuity from the LS equation


We distinguish here between S and higher waves, though the formalism developed in Sec. 7.2 can be equally applied to S
waves as well.

7.1 S waves
• The on-shell LS equation for partial waves reads
Z
p2 m ∞ v(p, p1 )t(p1 , p)
t(p, p; ) = v(p, p) + 2 dp1 p21 , (7.1)
m 2π 0 p21 − p2

where, for simplicity, we have omitted the subscript ℓ in the partial waves, as we will do in the following. We next extend
the variable p to its complex plane. We are interested in those values of p corresponding to the dynamical cuts (DCs), so
that we take

p = i k ± ε → p2 = −k2 ± i0+ . (7.2)

For such values one has to deform the original integration contour in Eq. (7.1) so as to avoid the crossing with the DCs.
The latter stem from the potential and happen for the values
q
p1 = ±(ik ± ε) + i m2π + x2 , x ∈ R , (7.3)

with the square root also affected by an extra and unrelated ± sign. Since p1 ∈ [0, +∞] in Eq. (7.1) the crossing with
the DCs of the original integration contour only happens when k > mπ and corresponds to + or − in front of Eq. (7.3),
depending whether one has +ε or −ε, respectively. As a result we have the contours shown in Fig. 10, C + (left panel)
and C − (right panel), respectively. Notice that C − is the complex conjugate of C + .5
• From the spectral decomposition of the potential it is clear the following property

v(p∗1 , p∗2 ) = v(p1 , p2 )∗ . (7.4)


5
Here I have in mind that the DCs originate by the exchange of particles and, when iterating the potential in a LS, we are just augmenting
the threshold of the corresponding contribution to the DCs (this is clear by looking at the iteration as a crossed channel process in which the
particles exchanged are put on-shell.)

44
ε
ε C+ C−

a) b)

Figure 10: Deformed integration contours in the p1 -complex plane for an on-shell partial wave amplitude evaluated at
p = ik + ε (left panel) and p = ik − ε (right panel) with k > mπ .

Let us see that this is also the case for the full partial wave amplitude

p∗2 2 p2
t(p∗1 , p∗2 ; ) = t(p1 , p2 ; 2 )∗ . (7.5)
m m
To obtain this result let us take into account that the integration deformation that is needed for calculating t(p∗1 , p∗2 ; p∗2 2 /m),
indicated by C ∗ , is the complex conjugate of the one used for t(p1 , p2 ; p22 /m), denoted by C. This statement is clear because
the DCs in v(p∗1 , p∗2 ) occur at complex conjugate positions of those in v(p1 , p2 ). Then we have
Z
p∗2 2 m dp∗3 p∗3 2 ∗ ∗ p2
∗2
t(p∗1 , p∗2 ; ) = v(p∗1 , p∗2 ) + 2 v(p ∗ ∗
,
1 3p )t(p ,
3 2p ; )
m 2π C∗ p∗3 2 − p∗2 2 m
( Z )∗
m dp3 p23 ∗2
∗ ∗ p2 ∗
= v(p1 , p2 ) + 2 v(p ,
1 3p )t(p ,
3 2p ; ) . (7.6)
2π C p23 − p22 m

Hence, by taking the complex conjugate of the previous equation it is clear that both t(p∗1 , p∗2 ; p∗2 2 /m)∗ and t(p1 , p2 , p22 /m)
satisfy the same LS equation, and as a result they must be the same.6 Notice that the Schwarz reflection principle Eq. (5.9)
is a particular case of Eq. (7.5) when restricted to real p1 .
• Now we consider the generalization of Eq. (5.7) to complex p1 and p2 . The LS equation is
Z
p22 m dp3 p23 p2 2
t(p1 , p2 ; ) = v(p1 , p2 ) + 2 2 2 v(p1 , p3 )t(p3 , p2 ; ), (7.7)
m 2π C p3 − p2 m

where the integration contour in the IE is denoted as C. One important point to realize is that the same integration
p2 p2
contour is also the one to be used both for t(−p1 , p2 ; m2 ) and t(p1 , −p2 ; m2 ), because of the symmetry around the origin of
p2
the branch points in p3 for v(p1 , p3 ) and t(p3 , p2 ; m2 ). E.g. if p1 = qr + iqi with positive qr and qi > mπ the corresponding
vertical addition at qr in the integration contour C extends from 0 up to i(qi − mπ ). However, both the real and imaginary
parts of −p1 are negative and then, according to the rules in Eq. (6.4), the vertical addition takes place at −Rep1 = qr
6
We assume that the LS equation for a scattering process has a unique solution. Otherwise, we can always choose the solution that satisfies
the property Eq. (7.5), as shown.

45
and extends from 0 up to −i(−Im(p1 ) + mπ ) = i(qi − mπ ). As a result the same integration contour results. Once this
point is clear then one can proceed in completely analogous way as done above around Eq. (5.7) to conclude that

p22 p2
t(−p1 , p2 ; ) = (−1)ℓ t(p1 , p2 ; 2 ) ,
m m
p22 p 2
t(p1 , −p2 ; ) = (−1)ℓ t(p1 , p2 ; 2 ) , (7.8)
m m
also hold for complex arguments.
• Taking into account the properties in Eqs. (7.4) and (7.5) we prove next that it follows the properties

v(ik − ε, ik − ε) = v(ik + ε, ik + ε)∗ ,


−k2 −k2 ∗
t(ik − ε, ik − ε; ) = t(ik + ε, ik + ε; ) . (7.9)
m m
Let us show the first of the relations since the other can be obtained analogously.

v(ik − ε, ik − ε) = v((−ik − ε)∗ , (−ik − ε)∗ ) = v(−ik − ε, −ik − ε)∗ = v(ik + ε, ik + ε)∗ . (7.10)

In the last step we have taken into account that v(−p1 , −p2 ) = v(p1 , p2 ). In the same way, we can demonstrate that
t(ik − ε, ik − ε; −k2 /m) = t(ik + ε, ik + ε; −k2 /m)∗ . Note that it is also clear from Eq. (7.8) that t(−p1 , −p2 ; p22 /m) =
t(p1 , p2 ; p22 /m).
• Because of the properties in Eq. (7.9) we take the difference:

t(ik + ε, ik + ε) − t(ik − ε, ik − ε) = v(ik + ε, ik + ε) − v(ik − ε, ik − ε)


Z
m v(ik + ε, p1 )t(p1 , ik + ε)
+ 2 dp1 p21
2π C + p21 + k2
Z
m v(ik − ε, p1 )t(p1 , ik − ε)
− 2 dp1 p21 . (7.11)
2π C − p21 + k2
In the following, unless the opposite is explicitly stated, the energy at which the partial wave is evaluated corresponds to
the second momentum in t(p1 , p2 ), i.e. E = p22 /m, and it is not explicitly shown in the list of arguments of t.
The contribution of the unperturbed original contours to the difference in Eq. (7.11) vanishes because v and t in the
rhs of Eq. (7.11) are continuous there,7 due to the location of the DCs as given in Eq. (7.3).
Next, we make the following change of integration variable

Imp1 = ν1 . (7.12)

and introduce 0 < δ < ε. Then, Eq. (7.11) is rewritten as

t(ik + ε, ik + ε) − t(ik − ε, ik − ε) = v(ik + ε, ik + ε) − v(ik − ε, ik − ε)


Z
iθ(k − mπ )m k−mπ dν1 ν12
− [v(ik + ε, iν1 + ε − δ)t(iν1 + ε − δ, ik + ε)
2π 2 0 k2 − ν12
−v(ik + ε, iν1 + ε + δ)t(iν1 + ε + δ, ik + ε)]
Z
iθ(k − mπ )m −k+mπ dν1 ν12
+ [v(ik − ε, iν1 + ε − δ)t(iν1 + ε − δ, ik − ε)
2π 2 0 k2 − ν12
−v(ik − ε, iν1 + ε + δ)t(iν1 + ε + δ, ik − ε)] . (7.13)
7
The point in which the positive real axis of p1 crosses with the DC is excluded of the integration contour.

46
We perform the change of integration variable ν1 → −ν1 in the second integral and take into account Eq. (7.9), so that
Eq. (7.13) can be expressed as

t(ik + ε, ik + ε) − t(ik + ε, ik + ε)∗ = v(ik + ε, ik + ε) − v(ik + ε, ik + ε)∗


Z
θ(k − mπ )im k−mπ dν1 ν12 
− v(ik + ε, iν1 + ε − δ)t(iν1 + ε − δ, ik + ε)
2π 2 0 k2 − ν12
− v(ik + ε, iν1 + ε + δ)t(iν1 + ε + δ, ik + ε) + v(−ik − ε, iν1 + ε − δ)∗ t(iν1 + ε − δ, −ik − ε)∗

− v(−ik − ε, iν1 + ε + δ)∗ t(iν1 + ε + δ, −ik − ε)∗ . (7.14)

Finally, since the product vt in the previous equation is an even function of the modulus of the shared three-momenta we
end with

Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12 
− dν1 2 Re v(ik + ε, iν1 + ε − δ)t(iν1 + ε − δ, ik + ε)
2π 2 0
2
k − ν1

− v(ik + ε, iν1 + ε + δ)t(iν1 + ε + δ, ik + ε)
= Imv(ik + ε, ik + ε) (7.15)
Z
θ(k − mπ )m k−mπ dν1 ν 2 
− dν1 2 1 2 Rev(ik + ε, iν1 + ε − δ)Ret(iν1 + ε − δ, ik + ε)
2π 2 0 k − ν1
− Rev(ik + ε, iν1 + ε + δ)Ret(iν1 + ε + δ, ik + ε)
− Imv(ik + ε, iν1 + ε − δ)Imt(iν1 + ε − δ, ik + ε)

+ Imv(ik + ε, iν1 + ε + δ)Imt(iν1 + ε + δ, ik + ε) . (7.16)

• The previous equation can be simplified by taking into account the spectral decomposition of a local potential. For
the projection in S-wave (the conclusions would be the same for any other N N partial wave),
Z Z
∞ +1 η(µ2 )
v(p1 , p2 ) = dµ µ dt .
µ0 −1 p21 + p22 − 2p1 p2 t + µ2
Z ∞
1  
v(ik + ε, iν1 + ε − δ) = dµ µη(µ2 ) log{−(k − ν1 )2 + µ2 + i0+ } − log{−(k + ν1 )2 + µ2 + i0+ } . (7.17)
2kν1 µ0

Of course, µ0 = mπ for N N . It then follows that for k > mπ , 0 < ν1 < k − mπ

Rev(ik + ε, iν1 + ε − δ) = Rev(ik + ε, iν1 + ε + δ) (7.18)



Z ∞

0 µ 0 < µ < k − ν1

π 2
Imv(ik + ε, iν1 + ε − δ) = dµµη(µ ) × −1 k − ν1 < µ < k + ν1 (7.19)
2kν1 µ0 
 0 k+ν <µ
1

Z ∞

 −2 µ0 < µ < k − ν1
π
Imv(ik + ε, iν1 + ε + δ) = dµµη(µ2 ) × −1 k − ν1 < µ < k + ν1 (7.20)
2kν1 µ0 
 0 k + ν1 < µ
(7.21)

• We now assume that η(µ2 ) ∝ δ(µ2 − m2π ). The general case can be worked out straightforwardly, but we first
concentrate on this simplest example to settle the formalism. As a result, the only condition that is fulfilled is mπ < k − ν1

47
a) C b) C ′

Figure 11: Deformed integration contours in the p-complex plane that are employed in Eq. (7.25).

so that we can set

Imv(ik + ε, iν1 + ε − δ) = 0 ,
Z k−ν1
π
Imv(ik + ε, iν1 + ε + δ) = − dµµη(µ2 ) . (7.22)
kν1 µ0

Then Eq. (7.16) becomes


Z
θ(k − mπ )m k−mπ dν1 ν12 
Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε) − Rev(ik + ε, iν1 + ε − δ)
2π 2 0 k2 − ν12
× {Ret(iν1 + ε − δ, ik + ε) − Ret(iν1 + ε + δ, ik + ε)}

+ Imv(ik + ε, iν1 + ε + δ)Imt(iν1 + ε + δ, ik + ε) . (7.23)

• Let us demonstrate that

Ret(iν1 + ε + δ, ik + ε) = Ret(iν1 + ε − δ, ik + ε) , (7.24)

in the limit δ → 0. For that we employ mathematical induction in the process of iterating the LS equation. The amplitudes
involved in Eq. (7.24) have real parts that are continuous at the level of the Born approximation, which satisfies Eq. (7.24).
We assume that this is the case for the nth iteration and write down the expression for the (n + 1)th iteration. In terms
of the integration contours C and C ′ , schematically drawn in Fig. 11, one has
Z
(n+1) m v(iν1 + ε + δ, p)t(n) (p, ik + ε)
t (iν1 + ε + δ, ik + ε) = v(iν1 + ε + δ, ik + ε) + 2 dpp2 ,
2π C p2 + k 2
Z
m v(iν1 + ε − δ, p)t(n) (p, ik + ε)
t(n+1) (iν1 + ε − δ, ik + ε) = v(iν1 + ε − δ, ik + ε) + dpp2 . (7.25)
2π C ′ p2 + k 2
Now, along the unperturbed integration contours the functions involved in the integrand in the previous equation are
continuous in the limit δ → 0, because the DCs cuts are not located along the original contour (note that the two

48
points of intersection along the positive real axis with the DCs are excluded in the deformed contours). As a result, the
contribution from the original unperturbed contours vanishes in the difference of Eq. (7.25). Next, we consider separately
the contributions from the two finite vertical lines in the contours C and C ′ , along which p is purely imaginary (except
for an infinitesimal real part that is always kept). First we consider t(n+1) (iν1 + ε + δ, ik + ε) and the piece of contour
corresponding to the vertical line extending up to i(k − mπ ) (we assume that k > mπ ). With the exchange of integration
variable Imp = κ we have the following contribution from Eq. (7.25)
Z
im k−mπ dκκ2 
− v(iν1 + ε + δ, iκ + ε − δ′ )t(n) (iκ + ε − δ′ , ik + ε)
2π 2 0 k2 − κ2

− v(iν1 + ε + δ, iκ + ε + δ′ )t(n) (iκ + ε + δ′ , ik + ε) , (7.26)

where 0 < δ′ < δ < ε. Here, the argument iκ + ε ± δ′ lies always to the left of the argument iν1 + ε + δ in v, so that this
function is continuous in δ′ and can be extracted as common factor. In addition Ret(n) is continuous by assumption, so
that the previous equation simplifies to
Z
m k−mπ dκκ2  (n) ′ (n) ′ 
v(iν 1 + ε + δ, iκ + ε) Imt (iκ + ε − δ , ik + ε) − Imt (iκ + ε + δ , ik + ε) , (7.27)
2π 2 0 k2 − κ2

Proceeding in the same way for t(n+1) (iν1 + ε − δ, ik + ε), subtracting it to the previous calculation, while keeping in mind
that Rev(iν1 + ε ± δ, iκ + ε) is continuous, we are then left with
Z
im k−mπ dκκ2  
Imv(iν1 + ε + δ, iκ + ε) − Imv(iν1 + ε − δ, ik + ε)
2π 2 0
2
k −κ 2
 
× Imt(n) (iκ + ε − δ′ , ik + ε) − Imt(n) (iκ + ε + δ′ , ik + ε) , (7.28)

that is purely imaginary and does not contribute to Ret(n+1) .


Now, we consider the vertical lines in the contours C and C ′ that extend up to i(ν1 − mπ ) (they only contribute for
ν1 > mπ , which is assumed to be the case). Again we take first under consideration t(n+1) (iν1 + ε + δ, ik + ε). The
corresponding expression is
Z
im ν1 −mπ dκκ2 
− v(iν1 + ε + δ, iκ + ε + δ − δ′ )t(n) (iκ + ε + δ − δ′ , ik + ε)
2π 2 0 k2 − κ2

−v(iν1 + ε + δ, iκ + ε + δ + δ′ )t(n) (iκ + ε + δ + δ′ , ik + ε) . (7.29)

In this case iκ + ε + δ ± δ′ is always at the right of ik + ε so that t(n) (iκ + ε + δ ± δ′ , ik + ε) is continuous for δ′ → 0
and we extract it as common factor. The previous equation then reads (as always Rev is continuous)
Z
m ν1 −mπ dκκ2  
Imv(iν1 + ε + δ, iκ + ε + δ − δ′ ) − Imv(iν1 + ε + δ, iκ + ε + δ + δ′ )
2π 2 0
2
k −κ 2

×t(n) (iκ + ε + δ, ik + ε) . (7.30)

Because of Eq. (7.22) Imv(iν1 + ε + δ, iκ + ε + δ − δ′ ) = 0, indeed. We proceed analogously for t(n+1) (iν1 + ε − δ, ik + ε)
and subtract the result to the previous expression. One obtains
Z
im ν1 −mπ dκκ2  
− Imv(iν1 + ε, iκ + ε + δ′ ) Imt(n) (iκ + ε + δ, ik + ε) − Imt(n) (iκ + ε − δ, ik + ε) , (7.31)
2π 2 0
2
k −κ 2

which again is a purely imaginary contribution.

49
• We now come back to Eq. (7.23) so that now it reads
Z
θ(k − mπ )m k−mπ dν1 ν12
Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε) − Imv(ik + ε, iν1 + ε + δ)
2π 2 0 k2 − ν12
× Imt(iν1 + ε + δ, ik + ε) . (7.32)

• According to this equation in order to determine ∆(A) = Imt(ik + ε, ik + ε) we need to know as well Imt(iν + ε +
e ik + ε) and 0 < ν < k − mπ . We further introduce the notation ε+ = ε + δ,
δ, e 0 < δe < ε. We can proceed similarly as
done above for Imt(ik + ε, ik + ε), and instead of Eq. (7.11) we now consider8

t(iν + ε+ , ik + ε) − t(iν − ε+ , ik − ε) = 2iImt(iν + ε+ , ik + ε) = 2iImv(iν + ε+ , ik + ε)


Z Z
m dp1 p21 m dp1 p21
+ 2 2 v(iν + ε+ , p1 )t(p1 , ik + ε) − 2 v(iν − ε+ , p1 )t(p1 , ik − ε) , (7.33)
2π C+
2
k + p1 2π C− k2 + p21

where C + is an integration contour of the same type as the one in the left panel of Fig. 11, while C − is its complex
conjugate. Proceeding similarly as done previously to obtain Eq. (7.16) we end with

Imt(iν + ε+ , ik + ε) = Imv(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12 
− 2 Re v(iν + ε+ , iν1 + ε − δ)t(iν1 + ε − δ, ik + ε)
2π 2 0
2
k − ν1

−v(iν + ε+ , iν1 + ε + δ)t(iν1 + ε + δ, ik + ε)
Z
θ(ν − mπ )m ν−mπ dν1 ν12 
− 2 Re v(iν + ε+ , iν1 + ε+ − δ)t(iν1 + ε+ − δ, ik + ε)
2π 2 0
2
k − ν1

−v(iν + ε+ , iν1 + ε+ + δ)t(ν1 + ε+ + δ, ik + ε) , (7.34)

with 0 < δ < δ. e Taking now into account that iν1 + ε ± δ is located to the left of iν + ε+ , while iν1 + ε+ ± δ is to
the right of ik + ε, we can extract v and t as common factors in the first and second integrals, respectively, on the rhs
of the previous equation. We also established that Ret(iν1 + ε ± δ, ik + ε), 0 < ν1 < k − mπ , is continuous for δ → 0,
Eq. (7.24), and of course the same can be said for Rev(iν + ε+ , iν1 + ε+ ± δ). Taking these points into account we can
rewrite Eq. (7.34) as

Imt(iν + ε+ , ik + ε) = Imv(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  
+ 2 Imv(iν + ε+ , iν1 + ε) Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε)
2π 2 0
2
k − ν1
Z ν−mπ
θ(ν − mπ )m dν1 ν12  + + + + 
+ 2 Imv(iν + ε , iν 1 + ε − δ) − Imv(iν + ε , iν 1 + ε + δ)
2π 2 0 k 2 − ν1
× Imt(iν1 + ε+ , ik + ε) . (7.35)

In virtue of Eq. (7.22) it follows that Imv(iν + ε+ , iν1 + ε+ − δ) = 0 in the second integral in Eq. (7.35). However, note
that in the first integral Imv(iν + ε+ , iν1 + ε) is not zero, in general, because there are other regions of values in the
8
Some details are omitted in the following derivations since they are of similar type to those already offered in this Section. For more details
see the handwritten notes.

50
variables beyond those considered in Eq. (7.22). We end with,

Imt(iν + ε+ , ik + ε) = Imv(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  
+ Imv(iν + ε+ , iν1 + ε) Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε)
2π 2 0 k2 − ν12
Z
θ(ν − mπ )m ν−mπ dν1 ν12
− Imv(iν + ε+ , iν1 + ε+ + δ)Imt(iν1 + ε+ , ik + ε) . (7.36)
2π 2 0 k2 − ν12

• It is clear from the first integral in this equation that we also need to know Imt(iν + ε − δ, ik + ε). We can proceed
similarly as just done to obtain Imt(iν + ε+ , ik + ε), and obtain for 0 < ν < k − mπ (this is the range of values needed
to solve the integral equation),
Z
θ(k − mπ )m k−mπ dν1 ν12
Imt(iν + ε, ik + ε+ ) = Imv(iν + ε, iν1 + ε+ )
2π 2 0 k2 − ν12
 
× Imt(iν1 + ε+ − δ, ik + ε+ ) − Imt(iν1 + ε+ + δ, ik + ε+ )
Z
θ(ν − mπ )m ν−mπ dν1 ν12
− Imv(iν + ε, iν1 + ε + δ)Imt(ν1 + ε + δ, ik + ε+ ) . (7.37)
2π 2 0 k2 − ν12

In this equation we have again made used of Eq. (7.22) to put Imv(iν + ε, ik + ε − δ) = 0 in the second integral. Now,
by taking the difference between Eqs. (7.37) and (7.36), we can easily obtain an integral equation for the difference of
imaginary parts of t present in Eq. (7.36),

Imt(iν + ε − δ, ik + ε) − Imt(iν + ε + δ, ik + ε) = −Imv(iν + ε + δ, ik + ε)


Z
θ(k − mπ )m dν1 ν12 
k−mπ 
+ 2 Imv(iν + ε, iν1 + ε + δ) − Imv(iν + ε, iν1 + ε − δ)
2π 2 0
2
k − ν1
Z
  θ(ν − mπ )m ν−mπ dν1 ν12
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) −
2π 2 0 k2 − ν12
 
× Imv(iν + ε, iν1 + ε + δ) Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) . (7.38)

There is a partial cancellation between both integrals from which the following expression is obtained

Imt(iν + ε − δ, ik + ε) − Imt(iν + ε + δ, ik + ε) = −Imv(iν + ε + δ, ik + ε)


Z
θ(k − mπ )m k−mπ ν12  
+ dν1 2 θ(ν1 − [ν − mπ ]) Imv(iν + ε, iν1 + ε + δ) − Imv(iν + ε, iν1 + ε − δ)
2π 2 0 k2 − ν1
 
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) , (7.39)

where θ(ν1 − [ν − mπ ]) can be extracted as a common factor because Imv(iν 


+ ε, iν1 + ε − δ) = 0 for 0 < ν1 < ν − mπ.
Now, by taking into account Tables 3 and 4, the difference θ(ν1 −[ν−mπ ]) Imv(iν+ε, iν1 +ε+δ)−Imv(iν+ε, iν1 +ε−δ)
implies that the integral in Eq. (7.39) only contributes for k − mπ > ν1 > mπ + ν. As a result we rewrite this equation as

Imt(iν + ε − δ, ik + ε) − Imt(iν + ε + δ, ik + ε) = −Imv(iν + ε + δ, ik + ε)


Z
θ(k − 2mπ − ν)m k−mπ ν12  
+ dν1 2 Imv(iν + ε, iν 1 + ε + δ) − Imv(iν + ε, iν 1 + ε − δ)
2π 2 mπ +ν k 2 − ν1
 
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) , (7.40)

51
This IE can also be solved in steps of mπ in the variable ν because for ν in between k − 2mπ and k − mπ the only
contribution stems from the independent term. Then, for ν between k − 3mπ and k − 2mπ one can use the values
determined in the previous step to evaluate the integral and obtain the required function in this new interval of values.
This process is further iterated n times until k − (n + 1)mπ < 0, determining Imt(iν + ε − δ, ik + ε) − Imt(iν + ε + δ, ik + ε)
for ν ∈ [0, k − nmπ ]. This result is next employed in the first integral of the rhs of Eq. (7.35), which allows us to calculate
Imt(iν + ε + δ, ik + ε) by solving the resulting integral equation. The solution of it can be straightforwardly implemented
in the integral of Eq. (7.32) to obtained the sought ∆(A).
It is clear that the IE of Eq. (7.40) does not pose any problem for its numerical solution by taking into account that
the integrand and the integration dominate are both finite. Because of the same reasons we conclude that the resulting IE
from Eq. (7.35), once Eq. (7.40) is substituted there, can also be solved straightforwardly in a meaningful way. The result
is then used in the definite integral of Eq. (7.32). Therefore, no infinities stem from the LS equation for the calculation of
∆(A). This is, of course, a desirable feature due to the absence of local counterterms in the chiral EFT contributing to
the discontinuities.

7.2 General case


The application of Eq. (7.32) is not straightforward for waves with ℓ > 0 because Imv(ik + ε, iν1 + ε + δ) and Imt(iν1 +
ε + δ, ik + ε) diverge like 1/ν1ℓ+1 for ν1 → 0, so that its product does like 1/ν 2(ℓ+1) . This behavior at the origin for
the discontinuity is clear if we think of the integral in Eq. (3.85) together with a factor tℓ inside the integrand that will
come from the term with the highest degree in the Legendre polynomial Pℓ (t).9 That this divergent factor also appears in
Imtij (iν1 + ε + δ, ik + ε) is clear from the Neumann series of the LS, cf. Eq. (6.5). In the following to simplify the writing
we indicate by n the previous exponent,

n ≡ 2(ℓ + 1) . (7.41)

For the subsequent derivations we make use of the fact that n is even.
• We take now as our starting point Eq. (7.14) and in order to explicitly keep track of the previous divergence in the
origin we define
ℓ+1
v̂(k, k′ ) = kℓ+1 k′ v(k, k′ ) ,
ℓ+1
t̂(k, k′ ) = kℓ+1 k′ t(k, k′ ) . (7.42)

In terms of this notation Eq. (7.14) reads


Z
θ(k − mπ )m k−mπ dν1 ν12
Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε) − dν1
4π 2 0 k2 − ν12
 v̂(ik + ε, iν1 + ε − δ)t̂(iν1 + ε − δ, ik + ε) − v̂(ik + ε, iν1 + ε + δ)t̂(iν1 + ε + δ, ik + ε)
×
(ik + ε)n (iν1 + ε)n
v̂(ik + ε, iν1 + ε − δ)∗ t̂(iν1 + ε − δ, ik + ε)∗ − v̂(ik + ε, iν1 + ε + δ)∗ t̂(iν1 + ε + δ, ik + ε)∗ 
+ . (7.43)
(ik − ε)n (iν1 − ε)n

Since each integral term is complex conjugate of the other we then combine them according to their common real parts
9
I have checked it explicitly from the expressions given in Ref. [4] for projecting the potential in partial waves.

52
and opposite in sign imaginary parts,
Z
θ(k − mπ )m k−mπ dν1 ν12 
Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε) − dν1
4π 2 (ik)n 0 k2 − ν12
 
Re v̂(ik + ε, iν1 + ε − δ)t̂(iν1 + ε − δ, ik + ε) − v̂(ik + ε, iν1 + ε + δ)t̂(iν1 + ε + δ, ik + ε)
 1 1 
× n
+ n
(iν1 + ε) (iν1 − ε)
 
+ iIm v̂(ik + ε, iν1 + ε − δ)t̂(iν1 + ε − δ, ik + ε) − v̂(ik + ε, iν1 + ε + δ)t̂(iν1 + ε + δ, ik + ε)
 1 1 
× n
− n
. (7.44)
(iν1 + ε) (iν1 − ε)
The previous expression can be simplified further by noticing that for 0 < ν1 < k − mπ Imv(ik + ε, iν1 + ε − δ) = 0,
Eq. (7.22), and that both Rev(ik + ε, iν1 + ε ± δ) and Ret(iν1 + ε ± δ, ik + ν) are continuous in δ → 0. The latter result
is demonstrated in Eq. (7.24). Notice that the factor (ik)n in the product vt → v̂ t̂ is purely real, so that it is extracted
as a common factor when distinguishing between the real and imaginary parts of this product. As a result, we rewrite
Eq. (7.44) as
Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12
− Imv̂(ik + ε, iν1 + ε + δ)Imt̂(iν1 + ε + δ, ik + ε)
4π 2 kn 0 k2 − ν12
 1 1 
× n
+ n
(ν − iε) (ν + iε)
Z
θ(k − mπ )im k−mπ dν1 ν12  
− 2 n 2 2 Rev̂(ik + ε, iν1 + ε + δ) Imt̂(iν1 + ε − δ, ik + ε)
4π k 0 k − ν1
 
− Imt̂(iν1 + ε + δ, ik + ε) − Imv̂(ik + ε, iν1 + ε + δ)Ret̂(iν1 + ε + δ, ik + ε)
 1 1 
× n
− n
, (7.45)
(ν1 − iε) (ν1 + iε)
with i2n = +1. It is interesting to note here that
Z k−mπ  1 1 
dν1 n
+ n
<∞ (7.46)
0 (ν1 − iε) (ν1 + iε)
because n is even, while
Z k−mπ  1 1 
dν1 n
− n
=∞ (7.47)
0 (ν1 − iε) (ν1 + iε)
but notice that the integrand is zero in the finite region since ε → 0+ .
• It is quite intuitive that the last integral in Eq. (7.45) does not develop an infrared singularity because of the near
threshold behavior of Rev̂(ik + ε, iν1 + ε + δ) and of Ret̂(iν1 + ε + δ, ik + ε) that vanish as ν12ℓ+1 . Thus, this factor
together the ν12 from the measure implies that the numerator vanishes as ν1n+1 for ν1 → 0 which compensates the factor
1/ν1n from the diverging factor and still there is an extra ν1 left. We show explicitly below in this section that indeed
Ret̂(iν1 + ε + δ, ik + ε) ∝ ν12ℓ+1 for ν1 → 0. Accepting this result, and the remark above that there is no contribution in
the finite region from the second integral in this equation, Eq. (7.45) simplifies to

Imt(ik + ε, ik + ε) = Imv(ik + ε, ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  1 1 
− 2 + Imv̂(ik + ε, iν1 + ε + δ)
4π 2 kn 0 k2 − ν1 (ν1 − iε)n (ν1 + iε)n
× Imt̂(iν1 + ε + δ, ik + ε) . (7.48)

53
• However, in order that Eqs. (7.46) cures the infrared divergence problem in Eq. (7.48) it is necessary that the series
expansion in ν1 of Imv̂(ik + ε, iν1 + ε + δ)Imt̂(iν1 + ε + δ, ik + ε) contains only even powers for ν1 → 0+ . This is indeed
the case as we show next. For that we derive an expression for Imt̂(iν + ε+ , ik + ε).
′2
Notice first that the LS equation for t̂(k, k′ ; km ) is obtained by diving Eq. (6.1) by (kk′ )ℓ+1 and rewriting accordingly
the integrand, then it reads
Z
k′ 2 m dp1 p21 ′ k
′2
t̂(k, k′ ; ) = v̂(k, k′ ) + 2 v̂(k, p 1 )t̂(p 1 , k ; ). (7.49)
m 2π C (p21 − k′ 2 )pn1 m

This is the same kind of IE as the original LS, Eq. (6.1), but with an extra factor pn1 in the denominator of the integrand.
We now proceed to calculate Imt̂(iν + ε+ , ik + ε), for 0 < ν < k − mπ , as in Eq. (7.33), and instead of Eq. (7.34) we
now have

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z  
θ(k − mπ )m k−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε − δ) t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)

4π 2 0 k2 − ν12 (iν1 + ε)n
Z  
θ(k − mπ )m k−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε − δ)∗ t̂(iν1 + ε − δ, ik + ε)∗ − t̂(iν1 + ε + δ, ik + ε)∗

4π 2 0 k2 − ν12 (iν1 − ε)n
Z  
θ(ν − mπ )m ν−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε+ − δ) − v̂(iν + ε+ , iν1 + ε+ + δ) t̂(iν1 + ε+ − δ, ik + ε)

4π 2 0 k2 − ν12 (iν1 + ε)n
Z  
θ(ν − mπ )m ν−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε+ − δ)∗ − v̂(iν + ε+ , iν1 + ε+ + δ)∗ t̂(iν1 + ε+ − δ, ik + ε)∗
− .
4π 2 0 k2 − ν12 (iν1 − ε)n
(7.50)

Keeping in mind the continuity in δ → 0 of Rev̂(iν + ε+ , iν1 + ε+ ± δ) and Ret̂(iν1 + ε ± δ, ik + ε), the latter shown in
Eq. (7.24), we can rewrite the previous equation as

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z  
θ(k − mπ )mi k−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε − δ) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)

4π 2 0 k2 − ν12 (iν1 + ε)n
Z  
θ(k − mπ )mi k−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε − δ)∗ Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
+
4π 2 0 k2 − ν12 (iν1 − ε)n
Z  
θ(ν − mπ )mi ν−mπ dν1 ν12 Imv̂(iν + ε+ , iν1 + ε+ − δ) − Imv̂(iν + ε+ , iν1 + ε+ + δ) t̂(iν1 + ε+ − δ, ik + ε)

4π 2 0 k2 − ν12 (iν1 + ε)n
Z  
θ(ν − mπ )mi ν−mπ dν1 ν12 Imv̂(iν + ε+ , iν1 + ε+ − δ) − Imv̂(iν + ε+ , iν1 + ε+ + δ) t̂(iν1 + ε+ − δ, ik + ε)∗
+ .
4π 2 0 k2 − ν12 (iν1 − ε)n
(7.51)

Expressing v̂ above in terms of its real and imaginary parts we can group together the first and second integrals, and

54
separately the third and fourth ones, similarly as done in Eq. (7.43), so that the Eq. (7.51) becomes

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12 +  
+ 2 Imv̂(iν + ε , iν 1 + ε − δ) Im t̂(iν 1 + ε − δ, ik + ε) − Im t̂(iν 1 + ε + δ, ik + ε)
4π 2 0 k 2 − ν1
 1 1 
× n
+ n
(iν1 + ε) (iν1 − ε)
Z k−mπ
θ(k − mπ )mi dν1 ν12  
− 2 2 2 Rev̂(iν + ε+ , iν1 + ε − δ) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
4π 0 k − ν1
 1 1 
× −
(iν1 + ε)n (iν1 − ε)n
Z
θ(ν − mπ )m ν−mπ dν1 ν12  
+ 2 2 2 Imv̂(iν + ε+ , iν1 + ε+ − δ) − Imv̂(iν + ε+ , iν1 + ε+ + δ) Imt̂(iν1 + ε+ − δ, ik + ε)
4π 0 k − ν1
 1 1 
× +
(iν1 + ε)n (iν1 − ε)n
Z
θ(ν − mπ )mi ν−mπ dν1 ν12  
− 2 2 2 Imv̂(iν + ε+ , iν1 + ε+ − δ) − Imv̂(iν + ε+ , iν1 + ε+ + δ) Ret̂(iν1 + ε+ − δ, ik + ε)
4π 0 k − ν1
 1 1 
× n
− n
. (7.52)
(iν1 + ε) (iν1 − ε)

The second and fourth integral terms in this equation do not give contribution for ε → 0 in the finite region and because
the near threshold behavior for ν1 → 0+ of Rev̂(iν + ε+ , iν1 + ε − δ) and Ret̂(iν1 + ε+ − δ, ik + ε), vanishing as ν12ℓ+1 ,
as remarked above concerning Eq. (7.45), there is not either infrared singularity.10 Then, we are lead to the result

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  1 1 
+ 2 + Imv̂(iν + ε+ , iν1 + ε)
4π 2 0 k2 − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
Z
θ(ν − mπ )m ν−mπ dν1 ν12  1 1 
− 2 + Imv̂(iν + ε+ , iν1 + ε+ + δ)Imt̂(iν1 + ε+ , ik + ε) , (7.53)
4π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n

where we have taken into account in the last integral term that Imv̂(iν + ε+ , iν1 + ε+ − δ) = 0 as ν − mπ > ν1 , Eq. (7.22).
From the previous equation it is clear that we also need to know Imt̂(iν + ε, ik + ε+ ), which by proceeding in the same
way as for Imt̂(iν + ε+ , ik + ε) is given by the expression
Z
θ(k − mπ )m
+
k−mπ dν1 ν12  1 1 
Imt̂(iν + ε, ik + ε ) = 2 +
4π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imv̂(iν + ε, iν1 + ε ) Imt̂(iν1 + ε − δ, ik + ε+ ) − Imt̂(iν1 + ε+ + δ, ik + ε+ )
+ +
Z
θ(ν − mπ )m ν−mπ dν1 ν12  1 1 
+ 2 +
4π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imv̂(iν + ε, iν1 + ε − δ) − Imv̂(iν + ε, iν1 + ε + δ) Imt̂(iν1 + ε, ik + ε+ ) , (7.54)
10
There is a subtlety affecting the 2nd integral term when ν = mπ because then (iν)2 = −m2π and this spoils the power expansion in ν1 of
Rev̂(iν + ε+ , iν1 + ε) for ν1 → 0. Nevertheless, we can think that the value at ν = mπ is obtained by taking the limit ν → mπ when sending
ε → 0. This justifies to neglect this term for all ν considered.

55
where we have taken into account that Imv̂(iν + ε, ik + ε+ ) = 0, 0 < ν < k − mπ .
• It is straightforward to show that for ν → 0+ there are only even powers in the power expansion in ν of Imt(iν +
ε , ik + ε), as we show next. For ν → 0+ the second integral term in Eq. (7.53) does not contribute, while reading from
+

Table 3 the expression for Imv̂(iν + ε+ , iν1 + ε),11 we then have

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z Z
θ(k − mπ )m  mπ +ν k−mπ dν1 ν12  1 1 
+ +2 +
4π 2 mπ −ν mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n
2 2  
× ρ(ν , ν1 ) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
(7.55)

where ρ(ν 2 , ν12 ) is a polynomial such that

χρ(ν 2 , ν12 ) (7.56)

corresponds to Imv̂(iν1 , iν2 ) according to Tables 3 and 4. This polynomial stems from the kinematical expressions for the
partial wave projection of NN amplitudes, from which we also learn that Imv̂(iν + ε+ , ik + ε) is a polynomial with only
even powers of ν.12 The two integrals in the previous expression can be combined as
Z mπ +ν Z k−mπ Z k−mπ Z k−mπ
 
+2 dν1 → + dν1 . (7.57)
mπ −ν mπ +ν mπ −ν mπ +ν

Thus, Imt̂(iν + ε+ , ik + ε) in the limit ν → 0+ reduces to

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z Z
θ(k − mπ )m  k−mπ k−mπ dν1 ν12  1 1 
+ + +
4π 2 mπ −ν mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n
2 2  
× ρ(ν , ν1 ) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε) ,
(7.58)

and this expression is clearly even in ν. Then its power expansion in this variable has only even powers.
Similar steps can be followed for ν → 0+ in Imt̂(iν + ε, ik + ε+ ), Eq. (7.54). In this limit, Imv̂(iν + ε, iν1 + ε+ ) can
be read from Table (4), so that Eq. (7.54) can be expressed as
Z
θ(k − mπ )m
+
mπ +ν dν1 ν12  1 1 
Imt̂(iν + ε, ik + ε ) = 2 +
4π 2 mπ −ν
2
k − ν1 (iν1 + ε)n (iν1 − ε)n
2  
× ρ(ν , ν12 ) Imt̂(iν1 + ε − δ, ik + ε ) − Imt̂(iν1 + ε+ + δ, ik + ε+ ) ,
+ +
(7.59)

which is not necessarily even in ν. We need show that this does not cause any trouble in the IE for Imt̂(iν + ε+ , ik + ε).
We show this below.
11
Notice that the factor (iνiν1 )n is real.
12
Because v(−p1 , p2 ) = (−1)ℓ v(p1 , p2 ) (an analogously for p2 ) it follows that v̂(−p1 , p2 ) = −v̂(p1 , p2 ), which determines that is an odd
function of p1 , so that only odd powers of ν1 enter in Rev̂(ik, iν1 ± δ). Here one should notice that this real part is insensitive of the sign of δ.
On the other hand, Imv̂(ik, iν1 ± δ) only contains even powers of ν1 in its power expansion, because the imaginary part only stems from the log
factor in the partial wave projection, whose imaginary part is just a constant for δ → 0+ , while its real part only contains odd powers of ν1 . For
explicit expression one can consider Eq. (3.86), or any other contribution to a partial wave projected potential.

56
• We can decouple the IEs of Eqs. (7.54) and (7.53) by taking the difference between them (analogously as already
done in Eq. (7.40)), we then obtain

Imt̂(iν + ε − δ, ik + ε) − Imt̂(iν + ε + δ, ik + ε) = −Imv̂(iν + ε + δ, ik + ε)


Z
θ(k − mπ )m k−mπ dν1 ν12  1 1 
+ 2 + θ(ν1 − (ν − mπ ))
4π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imv̂(iν + ε, iν1 + ε + δ) − Imv̂(iν + ε, iν1 + ε − δ)
 
× Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε) . (7.60)
 
Now, by taking into account Tables 3 and 4, θ(ν1 − (ν − mπ )) Imv̂(iν + ε, iν1 + ε + δ) − Imv̂(iν + ε, iν1 + ε − δ)
adopts a simple expression that allows us to rewrite the previous IE as

f(ν) ≡ Imt̂(iν + ε − δ, ik + ε) − Imt̂(iν + ε + δ, ik + ε) ,


Z
θ(k − 2mπ − ν)m k−mπ dν1 ν12 ρ(ν 2 , ν12 )
f(ν) = −Imv̂(iν + ε + δ, ik + ε) − f(ν1 ) , (7.61)
π2 mπ +ν k2 − ν12 (iν1 )n

so that no infrared divergence arises in this case for finite pion mass and it is well defined. Note also that the form of
the previous IE is the same for all ν ∈ [0, k − mπ ] as ρ(ν 2 , ν12 ) is always the same function in its arguments. The IE in
Eq. (7.61) can also be solved in steps of mπ in the variable ν because for ν in between k − 2mπ and k − mπ the only
contribution to f(ν) stems from the independent term. Then, for ν between k − 3mπ and k − 2mπ one can use the values
determined in the previous step to evaluate the integral and obtain f(ν) in this new interval of values. This process is
further iterated n times until k − (n + 1)mπ < 0, determining f(ν) for ν ∈ [0, k − nmπ ]. Next, we insert the solution of
Eq. (7.61) both in Eqs. (7.53) and (7.54) to evaluate Imt̂(iν + ε+ , ik + ε) and Imt̂(iν + ε, ik + ε+ ), respectively. Let us
rewrite further the previous equations to see that no problems with the infrared region of integration arises.
From Tables 3 and 4 we can rewrite Eq. (7.53) as

Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε) (7.62)


Z
θ(k − 2mπ )θ(k − 2mπ − ν)m k−mπ dν1 ν12 ρ(ν 2 , ν12 )
+ 2 f(ν1 )
2π 2 2
mπ +ν k − ν1 (iν1 )
n
Z
θ(k − 3mπ /2)θ(ν − (2mπ − k))m k−mπ dν1 ν12  1 1 
+ 2 2 2 n
+ n
ρ(ν 2 , ν12 )f(ν1 )
4π |mπ −ν| k − ν1 (iν1 + ε) (iν1 − ε)
Z ν−mπ
θ(k − 2mπ )θ(ν − mπ )m dν1 ν12  1 1 
− 2 2 n
+ ρ(ν 2 , ν12 )Imt̂(iν1 + ε+ , ik + ε) .
2π 0
2
k − ν1 (iν1 + ε) (iν1 − ε)n

The third term is infrared divergent but the integral is finite because we have shown above in Eq. (7.58) that Imt̂(iν +
ε+ , ik + ε) involves only even powers of ν in its power expansion. The second integral term could give rise to infrared
divergences in the limit ν → mπ . To see that the integral is finally finite, let us substitute f(ν1 ), as given by Eq. (7.61),
in this integral so that we are left with
Z
θ(k − 3mπ /2)θ(ν − (2mπ − k))m dν1 ν12  k−mπ
1 1 
− 2 + ρ(ν 2 , ν12 )ρ(ν12 , k2 )
2π 2 |mπ −ν| k 2−ν
1 (iν 1 + ε)n (iν 1 − ε)n
Z
θ(k − 2mπ )θ(ν − (3mπ − k))m2 k−2mπ dν1 ν12  1 1 
− 4 2 2 n
+ n
ρ(ν 2 , ν12 )
4π |mπ −ν| k − ν1 (iν1 + ε) (iν1 − ε)
Z k−mπ dν2 ν22 ρ(ν12 , ν22 )
× f(ν2 ) . (7.63)
mπ +ν1 k2 − ν22 (iν2 )n

57
ν2

k−mπ

ν2

|m π− ν | m π ν2−mπ k−2mπ ν1

Figure 12: Change of integration variables in Eq. (7.63).

The first integral here is finite for ν = mπ because any factor ρ is a polynomial in ν12 . For the last integral we exchange
the order of integration, see Fig. 12, so that it turns
Z
θ(k − 2mπ )θ(ν − (3mπ − k))m2 k−mπ dν2 ν22 f(ν2 )

4π 4 mπ +|mπ −ν| k2 − ν22 (iν2 )n
Z ν2 −mπ dν1 ν12  1 1 
× 2 2 n
+ n
ρ(ν 2 , ν12 )ρ(ν12 , ν22 ) . (7.64)
|mπ −ν| k − ν1 (iν1 + ε) (iν1 − ε)

Here, the infrared divergent part of the integral is clearly isolated in the second one, which gives a finite result for ν = mπ ,
due to the polynomial character in ν12 of both ρ factors.
For numerical purposes seems to be more appropriate to have a form for the IE of Imt̂(iν + ε+ , ik + ε) in which an
infrared divergent expression disappears. In this way, instead of the second integral term in Eq. (7.62) we write Eq. (7.63)
(with the form obtained in Eq. (7.64)) as
Z
+ + θ(k − 2mπ )θ(k − 2mπ − ν)m k−mπ dν1 ν12 ρ(ν 2 , ν12 )f(ν1 )
Imt̂(iν + ε , ik + ε) = Imv̂(iν + ε , ik + ε) +
2π 2 mπ +ν (k2 − ν12 )(iν1 )n
Z Z
θ(k − 2mπ )θ(ν − (3mπ − k))m2 k−mπ dν2 ν22 f(ν2 ) ν2 −mπ dν ν 2 ρ(ν 2 , ν 2 )ρ(ν 2 , ν 2 )
1 1 1 1 2
− 2 2 − ν2
4π 4 mπ +|mπ −ν| (k 2 − ν )(iν )n
2 2 |mπ −ν| k 1
 1 1 
× +
(iν1 + ε)n (iν1 − ε)n
Z
θ(k − 3mπ /2)θ(ν − (2mπ − k))m k−mπ dν1 ν12 ρ(ν 2 , ν12 )ρ(ν12 , k2 )  1 1 
− 2 2 2 n
+ n
2π |mπ −ν| k − ν1 (iν1 + ε) (iν1 − ε)
Z
θ(k − 2mπ )θ(ν − mπ )m ν−mπ dν1 ν12  1 1 
− 2 + ρ(ν 2 , ν12 )Imt̂(iν1 + ε+ , ik + ε) (7.65)
2π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n

58
From this equation we then conclude that one can evaluate safely Imt(ik + ε, ik + ε) in Eq. (7.48) without having an
infrared ill-defined integral.
• The resulting IE in the previous equation can be made explicitly smoother around the origin in ν1 if we add and
subtract a few terms in the Taylor expansion of the function to be solved for ν1 → 0+ , until the integrand is manifestly
convergent. E.g. let us take f (ν) as a generic function that has a power expansion with only even powers in ν. In this
way,
n/2−1 n/2−1
X 1 X 1
(2m) 2m
f (ν1 ) = f (ν1 ) − f (0)ν1 + f (2m) (0)ν12m (7.66)
m=0
(2m)! m=0
(2m)!

and the new function


n/2−1
X 1
g(ν1 ) = f (ν1 ) − f (2m) (0)ν12m , (7.67)
m=0
(2m)!

will have an IE with a kernel free of the infrared divergence.


Since the IE originates from the last integral term in Eq. (7.65), that only contributes for ν > mπ , the procedure outlined
is particularly simple in this case because f 2m (0), m = 0, . . . , n/2 − 1, are calculated directly from the independent term
in that equation. Indeed, the solution of the IE of Eq. (7.65) (which indeed has a simpler form in Eq. (7.53)) can be
obtained in succession in intervals of mπ , because for ν1 ∈ [0, mπ ] the function is given by the independent term. For
ν1 ∈ [mπ , 2mπ ] the values of the function inside the last integral term correspond to those already calculated from the
independent term. For ν1 ∈ [2mπ , 3mπ ] the values required in the integration are those already determined in the previous
step. This process is repeated r times until (r + 1)mπ > k − mπ , since ν < k − mπ .
Finally, since both Imt̂(iν + ε+ , ik + ε) − Imt̂(iν + ε, ik + ε+ ) and Imt̂(iν + ε+ , ik + ε) are free of infrared divergences,
as explicitly shown in Eqs. (7.61) and (7.65), respectively, the same must be true for Imt̂(iν + ε, ik + ε+ ), that results by
taking the difference between both. This can also be shown explicitly by inserting the IE for f(ν) into the expression for
Imt̂(iν + ε, ik + ε+ ), Eq. (7.54), and then exchanging the order of integration for the double-integral terms, similarly as
done above to deduce Eq. (7.65).
• Let us now demonstrate that Ret̂(iν + ε′ , ik + ε) vanishes as ν 2ℓ+1 for ν → 0+ , something left aside above. Let us
proceed analogously as done in Eq. (7.33), but now adding instead of subtracting the involved amplitudes,
Z
′ ′ m dp1 p21
t̂(iν + ε , ik + ε) = v̂(iν + ε , ik + ε) + 2 v̂(iν + ε′ , p1 )t̂(p1 , ik + ε) . (7.68)
2π C+ (k2 + p21 )pn1

Taking into account that t̂(iν − ε′ , ik − ε) = t̂(iν + ε′ , ik + ε)∗ and v̂(iν − ε′ , ik − ε) = v̂(iν + ε′ , ik + ε)∗ , we have
Z
m dp∗1 p∗1 2
t̂(iν − ε′ , ik − ε) = v̂(iν + ε′ , ik + ε)∗ + v̂(iν + ε′ , p1 )∗ t̂(p1 , ik + ε)∗ . (7.69)
2π 2 C+ (k2 + p∗1 2 )p∗1 n

Along the real-p1 axis v̂(iν + ε′ , p1 ) is purely real for ν → 0 and adding Eqs. (7.68) and (7.69) we have
Z
′ ′ m ∞ dp1 p21
Ret̂(iν + ε , ik + ε) = Rev̂(iν + ε , ik + ε) + 2 v̂(iν + ε′ , p1 )Ret̂(p1 , ik + ε)
2π 0 (k2 + p21 )pn1
Z
θ(k − mπ )mi k−mπ dν1 ν12  
− 2 2 2 n
v̂(iν + ε′ , iν1 + ε) t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
4π 0 (k − ν1 )(iν1 + ε)
Z k−mπ
θ(k − mπ )mi dν1 ν12 ′ ∗ ∗ ∗
+ 2 v̂(iν + ε , iν 1 + ε) t̂(iν 1 + ε − δ, ik + ε) − t̂(iν 1 + ε + δ, ik + ε) .
4π 2 0 (k2 − ν1 )(iν1 − ε)n
(7.70)

59
Here we have taken into account that for ν < mπ (which is our case now since ν → 0) there is only one vertical addition at
ε in Fig. 11. Then, by splitting v̂ in terms of its real and imaginary parts, and taking into account that Ret̂(iν1 +ε±δ, ik+ε)
is continuous in δ → 0, Eq. (7.24), we are left with
Z
′ ′ m ∞ dp1 p21
Ret̂(iν + ε , ik + ε) = Rev̂(iν + ε , ik + ε) + 2 v̂(iν + ε′ , p1 )Ret̂(p1 , ik + ε)
2π 0 (k2 + p21 )pn1
Z
θ(k − mπ )m k−mπ dν1 ν12  
+ 2 2 2 Rev̂(iν + ε′ , iν1 + ε) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
4π 0 k − ν1
 1 1 
× +
(iν1 + ε)n (iν1 − ε)n
Z
θ(k − mπ )m k−mπ dν1 ν12  
+i 2 2 2 Imv̂(iν + ε′ , iν1 + ε) Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + δ, ik + ε)
4π 0 k − ν1
 1 1 
× n
− n
. (7.71)
(iν1 + ε) (iν1 − ε)
The last term in this equation must vanish because it would give rise to a purely imaginary contribution, while the total
result should be real and all the other terms are real. The point is that for ν → 0 it is required that ν1 > mπ − ν so that
Imv̂(iν + ε′ , iν1 + ε) 6= 0, but then the next factor cancels in the finite ν1 region.
Next, Rev̂(iν + ε′ , p1 ) in the first integral term, with unperturbed integration contour, has a straightforward power
expansion in ν. The same can be said for Rev̂(iν + ε′ , iν1 + ε) in the second integral term, and we have the desired result
that Ret̂(iν + ε′ , ik + ε) ∝ ν 2ℓ+1 for ν → 0+ .
Nonetheless, there is a subtlety in the second integral term of Eq. (7.71) because for ν1 = mπ ±ν +O(ε, ε′ ) the argument
of the log in v̂ is zero and the expansion in ν cannot be done in this narrow region of width of O(ε, ε′ ). Nonetheless, if
we split the integral in three terms
Z k−mπ Z mπ −∆ Z mπ +∆ Z k−mπ
dν1 = dν1 + dν1 + dν1 , (7.72)
0 0 mπ −∆ mπ +∆

all of them are analytic in iν + ε′ , since the singularities are avoided by the deformation of the integration contour, and
for the first and third integrals one can demonstrate in the same way as above the required power expansion in ν (∆ ≫ ν
to guarantee that the modulus of the variable ξ = (p21 + p22 + m2π )/2p1 p2 , the one that appears in the power expansion of
the log in v̂, is larger than 1 in the integral domains of the first and third integrals, here p1 → iν + ε′ , p2 → iν1 + ε,).
For the second integral in Eq. (7.72) we have that ν1 ≃ mπ , so that we can extract all the factors in the second integral
term of Eq. (7.71) (in the limit ν → 0) except Rev̂(iν + ε′ , iν1 + ε) that crosses a logarithmic singularity, and then this
contribution can be written as
Z mπ +∆
θ(k − mπ )m  
2 n−2 2 2
Imt̂(imπ + ε − δ, ik + ε) − Imt̂(imπ + ε + δ, ik + ε) dν1 Rev̂(iν + ε′ , iν1 + ε) . (7.73)
2π mπ (k − mπ ) mπ −∆

The last integral can be done algebraically in terms of a primitive which has the same expression as in the case with
|ν1 − mπ | ≫ |ν|. But this primitive function is evaluated at the limits of integration for which indeed |ν1 − mπ | = ∆ ≫ |ν|
and then its power expansion in ν around ν = 0 starts at ν 2ℓ+1 as for the first and third integrals in Eq. (7.72).
Another less direct argument is based on the observation that if the second integral in Eq. (7.72) has a power expansion
starting at (iν + ε′ )m , cf. Eq. (7.73), then by a change of O(1) in ∆ (e.g. ∆ → ∆/2) we will have

A1 (iν + ε′ )2ℓ+1 + A∆ (iν + ε′ )m = A′1 (iν + ε′ )2ℓ+1 + A′∆ (iν + ε′ )m , (7.74)

Here A1 refers to the first and third integrals while A∆ corresponds to the second one of the splitting in Eq. (7.72), being
constant at lowest order. It is clear that Eq. (7.74) can only be realized for m = 2ℓ + 1. By doing ν > ε′ → 0+ in the

60
previous expression we obtain the desired result. Nevertheless this argument, as it is given, does not exclude the possibility
that in the difference ν m (A∆ − A′∆ ) those terms involving smaller powers of ν actually cancel between them.
• Although we do not distinguish explicitly here the mixing partial wave, the formalism developed in this section can
be straightforwardly generalized to the case of different initial and final orbital angular momenta, since they differ in two
units.13

7.3 One-step process


• The aim of this section is to derive a new procedure to determine ∆(A), which only requires to solve one IE.
Instead of taking the difference in Eq. (7.11),

t(ik + ε, ik + ε) − t(ik − ε, ik − ε) , (7.75)

we now consider

t(−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) . (7.76)

To show the connection between both discontinuities let us exchange the sign in the first argument of the latter. By taking
into account Eq. (7.8) it results that
 
t(−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) = (−1)ℓ t(ik − ε− , ik + ε) − t(ik − ε+ , ik + ε) . (7.77)

The first arguments in the difference on the lhs of the previous equation are shown by the black full points in Fig. 13,
while the red full points represent the first arguments of the rhs discontinuity. Now, inspection of Fig. 13 reveals that for
ε → 0+ the partial wave amplitudes t(ik + ε, ik + ε) and t(ik − ε− , ik + ε) can be continued analytically one to each other
in that limit.14 A Taylor expansion of t(z1 , ik + ε) can be performed at the point z1 = ik + ε (strictly) within a circle
of radius 2ε which incorporates the point ik − ε− , where t(ik − ε− , ik + ε) is evaluated. We can also proceed similarly
for t(ik − ε, z1 ) and, as a function of z1 , take a Taylor expansion centered at ik − ε of radius ε + ε′ , with ε− < ε′ < ε,
that includes −
p the point t(ik − ε, ik + ε ). The second argument is +just at the left of the cut running along the values
−ik + ε + i m2π + x2 . The latter amplitude is also equal to t(ik − ε , ik + ε) because the second argument is located in
the same relative position with respect to the cuts of t(ik − ε+ , z1 ). As a result it follows that for ε → 0+

t(ik + ε, ik + ε) − t(ik − ε, ik − ε) = t(ik − ε− , ik + ε) − t(ik − ε+ , ik + ε)


 
= (−1)ℓ t(−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) , (7.78)

where in the last equality we have used Eq. (7.77). Notice that the first discontinuity is, as we now, 2i∆(A), while the
latter one is (−1)ℓ if(−k)/k2ℓ+2 . It is then enough to calculate f(ν) for ν = −k to obtain ∆(A),

f(−k)
∆(A) = (−1)ℓ . (7.79)
2k2ℓ+2
• Aside considerations and manipulations. This point can be omitted in a first reading. p
The 2 2
p DCs of t(z1 , ik + ε) as a function of z1±are located, as we know, cf. Sec. 6, at ik + ε ± i mπ + x and −ik − ε ±
− ) and right
i m2π + x2 , with real x. But z
p 1 = −ik + ε , which implies that we are just slightly displaced to the left (ε
(ε+ ) of the cut at ik + ε − m2π + x2 , as shown in Fig. 13 by the corresponding black full points.15 We can make a similar
13
We would have then ℓ′ + 1 for the exponent of k′ in Eq. (7.42).
14
As remarked in the caption of Fig. 13, this figure is not drawn to scale so that the vertical gap, in the middle of which one has the point
ik + ε, is 2mπ wide while the horizontal distance between the vertical cuts is 2ε → 0+ .
15
Of course k > mπ /2, otherwise the two points would lay on the vertical gap of extent 2mπ .

61
analysis of the difference in Eq. (7.75),
p as shown in Fig. 14. For t(ik + ε, ik + ε), panel a), the first argument pik + ε is on
2 2
the right of the cut at −ik − ε + i mπ + x . Note also that ik + ε is p located out of the cuts at ik + ε ± i m2π + x2 .
While for t(ik − ε, ik − ε), ik − ε is to the left of the cut at −ik + ε + i m2π + x2 , panel b). We can move to the upper
part of the k-complex plane the points in the difference of Eq. (7.76), so that the comparison with Eq. (7.75), shown in
Fig. 14, will be more straightforward. This is accomplished by exchanging the sign in the first arguments and Eq. (7.77)
results. Note that in the difference on the rhs of this equation we first take the scattering amplitude at the right of the
cut and subtract its value at the left, like in Eq. (7.75).

+
ik− ε ik− ε− ik+ ε

−ik−ε
−ik+ε− −ik+ε
+

Figure 13: Location of the cuts of t(z1 , ik + ε), with z1 taking the values −ik + ε∓ in the difference t(−ik + ε− , ik + ε) − t(−ik +
ε+ , ik + ε). Note that the figure is not drawn to scale because the gap between two vertical cuts at ±ε is 2mπ , while the horizontal
separation between vertical cuts is 2ε → 0+ .

ik+ ε ik−ε

−ik−ε −ik+ ε

a) b)

Figure 14: Location of the cuts of t(z1 , ik ± ε) in the difference t(ik + ε, ik + ε) − t(ik − ε, ik − ε).

We offer now another demonstration to show that both discontinuities, Eqs. (7.75) and (7.76) are the same. However,
this demonstration cannot be applied to the coupled-channel case because it employs the symmetric relation t(p1 , p2 ; z) =
t(p2 , p1 ; z), while in coupled channels it reads tij (p1 , p2 ; z) = tji (p2 , p1 ; z) and the exchange of subscripts would spoil the

62
derivations for i 6= j. We already know that

t(ik + ε, ik + ε) − t(ik − ε, ik − ε) = 2iImt(ik + ε, ik + ε) , (7.80)

from Eqs. (7.5) and (7.10) (and the discussion that follows then). Next, we make use of Eq. (7.77) and exchange the
order of the arguments in t(ik − ε+ , ik + ε), so that
 
t(−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) = (−1)ℓ t(ik − ε− , ik + ε) − t(ik + ε, ik − ε+ )
   
= (−1)ℓ t(ik − ε− , ik + ε) − t(−ik − ε, −ik + ε+ ) = (−1)ℓ t(ik − ε− , ik + ε) − t(ik − ε, ik + ε+ )∗
= (−1)ℓ 2iImt(ik − ε− , ik + ε) . (7.81)

where we have used time-reversal invariance, Eq. (5.3), and the relations between the T -matrix elements under a change
of sign of one momentum argument, Eq. (7.8), and when all arguments are complex conjugated, Eq. (7.5).
Now, inspection of Fig. 13 reveals that for ε → 0+ the partial wave amplitudes t(ik + ε, ik + ε) and t(ik − ε− , ik + ε)
can be continued analytically one to each other in that limit.16 A Taylor expansion of t(z1 , ik + ε) can be performed at
the point z1 = ik + ε (strictly) within a circle of radius 2ε which incorporates the point ik − ε− , where t(ik − ε− , ik + ε)
is evaluated. As a result their imaginary parts in Eqs. (7.80) and (7.81) are the same.
One can save some steps in deducing Eq. (7.81)
 
t(−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) = (−1)ℓ t(ik − ε− , ik + ε) − t(−ik − ε, −ik + ε+ )
= (−1)ℓ 2iImt(ik − ε− , ik + ε) .

• We can calculate Imt(iν + ε+ , ik + ε) for ν < 0 in the same way as done in Sec. 7.1, Eq. (7.35), for positive ν.
We give below in Sec. 7.4 a more straightforward derivation of the resulting IE, Eq. (7.89). The only point we are going
generalize in detail is that

Ret(iν + ε + δ, ik + ε) = Ret(iν + ε − δ, ik + ε) , δ → 0+ , (7.82)

also holds for ν < 0 with |ν| < k + mπ (this upper limit is taken to avoid the zero of the energy denominator 1/(k2 − ν12 )
for ν < −mπ with ν1 ∈ [ν + mπ , 0]). We have, as usual,
Z
m dpp2
t(iν + ε + δ, ik + ε) = v(iν + ε + δ, ik + ε) + v(iν + ε + δ, p)t(p, ik + ε) ,
2π 2 C p2 + k2
Z
m dpp2
t(iν + ε − δ, ik + ε) = v(iν + ε − δ, ik + ε) + 2 v(iν + ε − δ, p)t(p, ik + ε) , (7.83)
2π C ′ p2 + k2
where the integration contours are analogous to those shown in Fig. 11, but now the vertical segments at ε ± δ are
downwards up to ν + mπ (for ν < −mπ ). Taking the difference between the two scattering amplitudes in Eq. (7.83),
recalling that the unperturbed parts of the integration contours do not contribute to the discontinuity, one has

t(iν + ε + δ, ik + ε) − t(iν + ε − δ, ik + ε) = v(iν + ε + δ, ik + ε) − v(iν + ε − δ, ik + ε)


Z
mi k−mπ dν1 ν12   
− 2 2 2 v(iν + ε + δ, iν1 + ε) − v(iν + ε − δ, iν1 + ε) t(iν1 + ε − δ′ , ik + ε) − t(iν1 + ε + δ′ , ik + ε)
2π 0 k − ν1
Z
mi ν+mπ dν1 ν12   
− 2 2 2 v(iν + ε, iν1 + ε − δ′ ) − v(iν + ε, iν1 + ε + δ′ ) t(iν1 + ε + δ, ik + ε) − t(iν1 + ε − δ, ik + ε) .
2π 0 k − ν1
(7.84)
16
As remarked in the caption of Fig. 13, this figure is not drawn to scale so that the vertical gap, in the middle of which one has the point
ik + ε, is 2mπ wide while the horizontal distance between the vertical cuts is 2ε → 0+ .

63
We already know that the discontinuity in v only affects its imaginary part, and the same conclusion was also reached for t
in the case of ν1 > 0, Eq. (7.24). As a result, the first two lines on the rhs of Eq. (7.84) are purely imaginary. The kernel
multiplying the discontinuity of t in the last integral is purely real. In this way, we are driven to an IE for the discontinuity
of t whose independent term is purely imaginary and whose kernel is real. Its solution is then a purely imaginary function,

t(iν + ε + δ, ik + ε) − t(iν + ε − δ, ik + ε) = iη(ν) , (7.85)

satisfying an explicitly real IE,


Z ν+mπ
m
η(ν) = f (ν) + dν1 K(ν, ν1 )η(ν1 ) ,
2π 2 0
−iν12  ′ ′ 
K(ν, ν1 ) = 2 v(iν + ε, iν 1 + ε − δ ) − v(iν + ε, iν 1 + ε + δ ) . (7.86)
k 2 − ν1

This proves Eq. (7.82) .


• We first calculate Imt(iν + ε+ , ik + ε) as in Eq. (7.33) and proceed in an analogous from as in Eqs. (7.34)-(7.36)

Imt(iν + ε+ , ik + ε) = Imv(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  
+ 2 2 2 Imv(iν + ε+ , iν1 + ε) Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε)
2π 0 k − ν1
Z ν+mπ
θ(−ν − mπ )m dν1 ν12
− 2 − ν2
Imv(iν + ε+ , iν1 + ε+ + δ)Imt(iν1 + ε+ , ik + ε) . (7.87)
2π 2 0 k 1

In the difference of Eq. (7.87) we also need Imt(iν + ε, ik + ε+ ) which can be calculated similarly with the result

Imt(iν + ε, ik + ε+ ) = Imv(iν + ε, ik + ε+ )
Z
θ(k − mπ )m k−mπ dν1 ν12  
+ 2 2 2 Imv(iν + ε, iν1 + ε+ ) Imt(iν1 + ε+ − δ, ik + ε+ ) − Imt(iν1 + ε+ + δ, ik + ε+ )
2π 0 k − ν1
Z ν+mπ
θ(−ν − mπ )m dν1 ν12
− 2 − ν2
Imv(iν + ε, iν1 + ε + δ)Imt(iν1 + ε + δ, ik + ε+ ) . (7.88)
2π 2 0 k 1

Compared with Eq. (7.37), we have kept the first term Imv(iν + ε, ik + ε+ ) since |ν| could be larger than k − mπ (e.g.
apply the relationships of Eq. (B.2)).
Now, taking the difference between Eqs. (7.88) and (7.87) one has in a straightforward manner

Imt(iν + ε− , ik + ε) − Imt(iν + ε+ , ik + ε) = Imv(iν + ε− , ik + ε) − Imv(iν + ε+ , ik + ε)


Z
θ(k − 2mπ − ν)m k−mπ ν12  
+ dν1 2 Imv(iν + ε− , iν1 + ε) − Imv(iν + ε+ , iν1 + ε)
2π 2 ν+mπ
2
k − ν1
 
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) . (7.89)

In the previous equation we have dropped the factor θ(−ν − mπ ) in the lower limit of integration because i) for ν < −mπ
it is equal to 1 and ii) this is implicitly included in discontinuity of the potential for −mπ < ν, because it is then necessary
that ν1 > ν + mπ to generate imaginary part from the potential, v(iν + ε± , iν1 + ε) (Appendix B). When this equation is
particularized to ν = −k we then have the required discontinuity of Eq. (7.76).

64
• However, Eq. (7.89) is suitable only for S waves, while for higher partial waves we should proceed instead as in
Sec. (7.2) (which is also a procedure applicable for S waves as well.) Then, instead of Eq. (7.50) we would have
Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z (  )
m k−mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε) t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
− 2 Re
2π 0 k2 − ν12 (iν1 + ε)n
Z (  )
m ν+mπ dν1 ν12 v̂(iν + ε+ , iν1 + ε+ − δ) − v̂(iν + ε+ , iν1 + ε+ + δ) t̂(iν1 + ε+ , ik + ε)
− 2 Re (7.90)
2π 0 k2 − m2π (iν1 + ε)n
Considering the continuity of both Rev̂ and Ret̂ when δ → 0+ , and the fact that Rev(iν + ε+ , iν1 + ε) and Ret̂(iν1 +
ε+ , ik + ε) vanishes as ν12ℓ+1 as ν1 → 0,17 we arrive to the analogous of Eq. (7.53) for ν < 0,
Imt̂(iν + ε+ , ik + ε) = Imv̂(iν + ε+ , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  1 1 
+ 2 + Imv̂(iν + ε+ , iν1 + ε)
4π 2 0 k2 − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε + δ, ik + ε)
Z
θ(−ν − mπ )m 0 dν1 ν12  1 1 
+ 2 + Imv̂(iν + ε+ , iν1 + ε+ + δ)Imt̂(iν1 + ε+ , ik + ε) . (7.91)
4π 2 ν+mπ
2
k − ν1 (iν1 + ε)n (iν1 − ε)n

Note that in the latter equation we have exchanged the order of the limits of integration in the last integral and have set
Imv̂(iν + ε+ , iν1 + ε+ − δ) = 0 in the corresponding domain of integration (Appendix B). We also have the analogous
equation for Imt̂(iν + ε− , ik + ε),
Imt̂(iν + ε− , ik + ε) = Imv̂(iν + ε− , ik + ε)
Z
θ(k − mπ )m k−mπ dν1 ν12  1 1 
+ 2 + Imv̂(iν + ε− , iν1 + ε)
4π 2 0
2
k − ν1 (iν1 + ε)n (iν1 − ε)n
 
× Imt̂(iν1 + ε − δ, ik + ε) − Imt̂(iν1 + ε− + δ, ik + ε)

Z
θ(−ν − mπ )m 0 dν1 ν12  1 1 
+ 2 + Imv̂(iν + ε− , iν1 + ε− + δ)Imt̂(iν1 + ε− , ik + ε) . (7.92)
4π 2 ν+mπ
2
k − ν1 (iν1 + ε)n (iν1 − ε)n

Let us recall the definition of f(ν), Eq. (7.61), and introduce


∆v̂(ν, ν1 ) = Imv̂(iν + ε− , iν1 + ε) − Imv̂(iν + ε+ , iν1 + ε) . (7.93)
Then, the difference between Eqs. (7.92) and (7.91) can be rewritten as
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = ∆v̂(ν, k) + + ∆v̂(ν, ν1 )f(ν1 ) . (7.94)
4π 2 mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n

Where we have dropped the explicit presence of θ(−ν − mπ ) for the same reasons as in Eq. (7.89), a result that also
follows here from the immediately following discussion. This IE is the analogous to Eq. (7.61) for ν > 0. As in the latter
case one can show that the IE Eq. (7.94) keeps the same form for all values of ν under consideration. This is because
∆v̂(ν, ν1 ) stems from the imaginary part in the difference of log-terms in Eq. (B.1), namely,
   
∆v̂(ν, ν1 ) ∝ log − (ν + ν1 )2 + m2π + i0+ − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+
n    o
− log − (ν + ν1 )2 + m2π + i0+ − log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+
   
= − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+ + log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+ . (7.95)
17
One can apply the demonstration developed in Sec. 7.2, Eqs. (7.68)-(7.74), for negative ν.

65
where the proportionality coefficient is ρ(ν 2 , ν12 )/π, as introduce in Eq. (7.56). The first two log-terms correspond to
Imv̂(iν + ε− , iν1 + ε) and the last two (shown between curly brackets) to Imv̂(iν + ε+ , ik + ε). It follows from Eq. (7.95)
that in order to end with a nonvanishing imaginary part it is required that (ν1 − ν)2 > m2π or, in other terms, that
q
ν1 = ν ± m2π + x2 , x ∈ R . (7.96)
In the integral of Eq. (7.94) only the plus sign matters as ν1 > ν + mπ , and in such circumstances the imaginary part of
Eq. (7.95) is −2π, so that for the whole integration interval in Eq. (7.94) it holds that
∆v̂(ν, ν1 ) = −2ρ(ν 2 , ν12 ) , ν1 ∈ [ν + mπ , k − mπ ] . (7.97)
The same result could be applied also to the independent term in Eq. (7.94) since k > ν + mπ because k − mπ > ν > −k
and k > mπ . Then Eq. (7.94) can be rewritten such that it conserves the same form for all the values of ν of interest,
and it can be written as
Z  
2 2θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = −2ρ(ν , k ) − + ρ(ν 2 , ν12 )f(ν1 ) . (7.98)
2π 2 mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n

The fact that the integrand in Eq. (7.94) is nonzero only for ν1 > ν + mπ also implies that θ(−ν − mπ ) can be neglected
when writing this integral. The reason is because if −ν < mπ the last integrals in Eqs. (7.91) and (7.92) are not present
and for the remaining integrals it is required for a nonvanishing integrand that k − mπ > ν1 > ν + mπ . In this way the
only restriction is that k > ν + 2mπ , which is accounted for by the presence of θ(k − 2mπ − ν) in Eqs. (7.94) and (9.18).
• For ν + mπ < 0, we can rewrite Eq. (7.94) as
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = ∆v̂(ν, k) + 2 − ν2
+ ∆v̂(ν, ν1 )f(ν1 )
4π 2 −mπ −ν k 1 (iν 1 + ε) n (iν 1 − ε)
n
Z  
θ(k − 2mπ − ν)m −mπ −ν dν1 ν12 1 1 1 
+ 2 2 2 n
+ n
∆v̂(ν, ν1 ) f(ν1 ) + f(−ν1 ) . (7.99)
4π mπ +ν k − ν1 (iν1 + ε) (iν1 − ε) 2
where we have taken into account that for ν1 ∈ [mπ + ν, −mπ − ν] one has that ∆v̂(ν, ν1 ) = Imv̂(iν + ε− , iν1 + ε) =
∆v̂(ν, −ν1 ). This follows trivially from the relationships in Eq. (B.2) and Tables 3 and 4 in Appendix B. We can restrict
the last integration in Eq. (7.99) to positive values of ν1 with the result
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = ∆v̂(ν, k) + 2 2 2 n
+ ∆v̂(ν, ν1 )f(ν1 )
4π −mπ −ν k − ν1 (iν1 + ε) (iν1 − ε)n
Z  
θ(k − 2mπ − ν)m −mπ −ν dν1 ν12 1 1  
+ 2 2 2 n
+ n
Imv̂(iν + ε− , iν1 + ε) f(ν1 ) + f(−ν1 ) . (7.100)
4π 0 k − ν1 (iν1 + ε) (iν1 − ε)
The problem arises from the first integral term when ν → −mπ , because of the infrared divergence in the numerator and
because f(ν1 ) has not only even powers in its Taylor expansion around ν1 = 0. This was already shown above in Eq. (7.59)
because of Imt̂(iν + ε− , ik + ε) contributing to f(ν). On the other hand, the integrand of the second integral term in
Eq. (7.100) has only even powers of ν1 in its Taylor expansion around zero and then the infrared divergence can be cured.
An analogous problematic situation occurs for ν +mπ > 0, in which case the lower limit in the integral term of Eq. (7.94)
tends to zero for ν → −mπ .
• This problem for finding an appropriate expression to calculate f(ν) in the limit ν → −mπ can be solved by giving
the IE for f(ν) that is obtained after one interaction of Eq. (7.94).18 To simply the writing we denote by Sn (ν1 ) the sum
1 1
Sn (ν1 ) = n
+ . (7.101)
(iν1 + ε) (iν1 − ε)n
18
Similarly as we did to show that Eq. (7.62) was well defined in the limit ν → mπ .

66
The direct iteration of Eq. (7.94) gives
Z
mdν1 ν12 k−mπ
f(ν) = ∆v̂(ν, k) + θ(k − 2mπ − ν) 2 Sn (ν1 )∆v̂(ν, ν1 )∆v̂(ν1 , k)
4π 2
2
mπ +ν k − ν1
  Z Z k−mπ
m 2 k−2mπ dν1 ν12 dν2 ν22
+ θ(k − 3mπ − ν) S n (ν 1 )∆v̂(ν, ν 1 ) 2 Sn (ν2 )∆v̂(ν1 , ν2 )f(ν2 ) . (7.102)
4π 2 mπ +ν k2 − ν12 2
mπ +ν1 k − ν2

Next, we exchange the order of integration in the last integral of the previous expression (which is legitimate for any
finite ε) and obtain
Z
k−mπ dν ν 2
m 1 1
f(ν) = ∆v̂(ν, k) + θ(k − 2mπ − ν) 2 Sn (ν1 )∆v̂(ν, ν1 )∆v̂(ν1 , k)
4π mπ +ν k − ν12
2
  Z Z ν2 −mπ
m 2 k−mπ dν2 ν22 dν1 ν12
+ θ(k − 3mπ − ν) 2 S n (ν 2 )f(ν 2 ) Sn (ν1 )∆v̂(ν, ν1 )∆v̂(ν1 , ν2 ) . (7.103)
4π 2 2
2mπ +ν k − ν2 mπ +ν k2 − ν12

Note that this expression is well-defined for ν → −mπ because ∆v̂(ν, ν1 )∆v̂(ν1 , ν2 ) only contains even powers of ν1 .
However, the calculation of the last integral for ν = −2mπ is not explicitly obvious, which is not a problem if we use
Eq. (7.100).
Then, we have at our disposal several expressions that could be used in different intervals of ν in order to avoid the
infrared singularities


 Eq.(7.98) −mπ /2 < ν
f(ν) = Eq. (7.103) −3mπ /2 < ν < −mπ /2

 Eq. (7.100) −3m /2 > ν
π

(7.104)

Of course, the limits in the intervals within Eq. (7.104) could be changed. The important point is that ν > −mπ (1 − x)
when using Eq. (7.94), ν ∈ [−mπ (1 − x), −mπ (1 + y)] for Eq. (7.103) and −mπ (1 + y) > ν in Eq. (7.100), with x, y
real and positive numbers in ]0, 1[.
• In order to calculate ∆(A) what we really need is f(−k). We make the following steps:
i) Exchange the integration variable ν1 → −ν1 in the interval from 0 to −k + mπ in Eq. (7.94).
ii) For ν1 ∈ [0, k − mπ ] it is valid that ∆v̂(−k, ν1 ) = Imv̂(−ik + iε− , iν1 + ε). Because of the relations in Eq. (B.2),
it is also true that ∆v̂(−k, −ν1 ) = Imv̂(−ik + iε− , iν1 + ε), so that both ∆v̂ are the same.
We are then left with
Z   h i
m k−mπ dν1 ν12 1 1
f(−k) = ∆v̂(−k, k) + + Imv̂(−ik + ε− , iν1 + ε) f(ν1 ) + f(−ν1 ) .
4π 2 0 k2 − ν12 (iν1 + ε)n (iν1 − ε)n
(7.105)

The infrared divergence in this integral can be cured because the integrand is even in ν1 .
• Indeed, we show next that for S- (which is trivial) and P -waves f(−mπ ) could be directly calculated from Eq. (7.94).
For that let us consider the factor ∆v̂(ν, ν1 ) in the integrand of the problematic first integral term in Eq. (7.100) for
ν = −mπ . This factor is i) a polynomial in ν12 and ii) fulfills that (as demonstrated below)

∆v̂(0, −mπ ) = 0 . (7.106)

These two conditions are enough to guarantee that f(−mπ ) is finite for P waves.

67
The point to see the validity of the requirement in Eq. (7.106) is by noticing that in the partial-projection process we
have integrations of the form
Z +1 Pj (x)
dx , (7.107)
−1 x−ξ

with (−1)j = (−1)ℓ , where we have reintroduced the variable

p21 + p22 + m2π


ξ= . (7.108)
2p1 p2
Pj (x) is a polynomial in x of degree j and consecutively we can reduce this power by x → (x − ξ) + ξ and integrating in
Eq. (7.107), so that at the end the log terms (the ones that originate the imaginary part in v̂) are accompanied by powers
of ξ (in this way we generate at most a term of degree j, ξ j , and powers of smaller degree). In addition, once we extract
(p1 p2 )ℓ+1 as common factor to calculate v̂ from v, Eq. (7.42), the variable ξ → (p21 + p22 + m2π ) and we also generate
powers of (p1 p2 ) multiplying to powers of (p21 + p22 + m2π ) (including possibly a term with ξ 0 ). The degree of the power
of (p1 p2 ) in a monomial increases as the power of ξ decreases. From the definition of partial-wave projection in a wave
one could also have isolated factors of p21 or p22 . At the end only even powers of (p1 p2 ) arise because Imv̂ is a polynomial
in p21 and p22 , footnote 12. Then, for p1 = 0 and p2 = −imπ one has that (p21 + p22 + m2π ) = 0 and (p1 p2 ) = 0 so that
∆v̂(0, −mπ ) = 0. The next term in the power expansion of ∆v̂(ν, ν1 ) in ν1 is quadratic in ν1 .
• In order to apply numerically Eq. (7.100) for −3mπ /2 > ν, according to Eq. (7.104), one could follow a similar
procedure as explained in Sec. 7.2, by removing the first ℓ terms in the Taylor series of f(ν1 ) + f(−ν1 ), which makes the
last integral of Eq. (7.100) be explicitly convergent (see Eq. (7.66)). The extra terms resulting from this process, for which
the dependence on ε of Sn (ν1 ) is crucial and would give problems if evaluated numerically in the limit ε → 0, can then be
integrated algebraically. E.g. for the D waves, one can substitute in the integrand
 d2 f(0) 2  d2 f(0) 2
f(ν1 ) + f(−ν1 ) → f(ν1 ) + f(−ν1 ) − 2f(0) − ν + 2f(0) + ν . (7.109)
dν12 1 dν12 1
So that the first term between brackets is free of infrared divergence and for the rest the integration is done algebraically.
For evaluating the first ℓ coefficients of the Taylor of the symmetric part of f(ν) and ν = 0, f(0), d2 f(0)/dν 2 , . . .,
d2(ℓ−1) f(0)/dν 2(ℓ−1) one can readily use Eq. (7.94). This IE can be solved from top to bottom, because for ν between
k − 2mπ and k − mπ the solution is given by the independent term. Then, for ν in the interval between k − 3mπ to
k − 2mπ we employ in the integrand the values previously determined, and so on. To evaluate f(ν) up to ν = 0 we need
to know f(ν) up to ν = mπ in the integration.
This very same process could be also applied to take care of the infrared singularity in Eq. (7.105) so as to calculate
∆(A) in an accurate manner.

7.4 Condensed and shorter derivation of the IE for f(ν)


Let us recall the relation between ∆(A) and f(−k) given in Eq. (7.79), and which derivation is developed between
Eqs. (7.77) and (7.79) in Sec. 7.3. Our aim here is to offer a more compact and shorter derivation for the IE of f(ν),
ν ∈ [−k, k − mπ ] valid for both S and higher partial waves, that also collects in just one place all the necessary steps to
arrive to the final IE.
The half-off-shell LS equation for partial waves reads
Z
k22 m dp1 p21 k2
t(k1 , k2 ; ) = v(k1 , k2 ) + 2 2 2 v(k1 , p1 )t(p1 , k2 ; 2 ) . (7.110)
m 2π C p1 − k2 m
where the (deformed) integration contour C depends on the external variables k1 and k2 , as discussed in Sec. 6.

68
ε− ε+
ε ε

Figure 15: Vertical contours used in Eq. (7.111) corresponding to the case ν < 0. Left panel applies to t(iν + ε− , ik + ε)
and right panel to t(iν + ε+ , ik + ε).

• Let us discuss first the case ν < 0. We take directly the difference t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) and consider
the added vertical contours shown in Fig. 15, which are the only ones that give contribution to the difference. The
contribution of the unperturbed original contours (with real and positive p1 ∈ [0, ∞]) vanishes because then v(iν + ε± , p1 )
and t(p1 , ik + ε) in the rhs of Eq. (7.110) are continuous there,19 due to the location of the DCs as given in Eq. (7.3).
Then it results

t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) = v(iν + ε− , ik + ε) − v(iν + ε+ , ik + ε)


Z
im ν+mπ dν1 ν12  − − − − 
− θ(−ν − mπ ) 2 v(iν + ε , iν 1 + ε − δ) − v(iν + ε , iν 1 + ε + δ) t(iν1 + ε− , ik + ε)
2π 0 k2 − ν12
Z
im k−mπ dν1 ν12  
− θ(k − mπ ) 2 2 2 v(iν + ε− , iν1 + ε) t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε)
2π 0 k − ν1
Z ν+mπ
im dν1 ν12  
+ θ(−ν − mπ ) 2 2 2 v(iν + ε+ , iν1 + ε+ − δ) − v(iν + ε+ , iν1 + ε+ + δ) t(iν1 + ε+ , ik + ε)
2π 0 k − ν1
Z k−mπ
im dν1 ν12  
+ θ(k − mπ ) 2 2 2 v(iν + ε+ , iν1 + ε) t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) , (7.111)
2π 0 k − ν1
where we have taken into account that v(iν + ε± , iν1 ) is continuous around the upwards vertical cut of extension k − mπ
and so it is t(iν1 , ik + ε) around the downwards vertical cut of extension −ν − mπ (with |ν| > mπ ). Combining the
integrals up to ν + mπ and k − mπ separately one can rewrite the previous equation as
Z
− + − im+
ν+mπ dν1 ν12
t(iν + ε , ik + ε) − t(iν + ε , ik + ε) = v(iν + ε , ik + ε) − v(iν + ε , ik + ε) + θ(−ν − mπ ) 2
2π 0 k2 − ν12
 − − − −  − + 
× v(iν + ε , iν1 + ε + δ) − v(iν + ε , iν1 + ε − δ) t(iν1 + ε , ik + ε) − t(iν1 + ε , ik + ε) (7.112)
Z
im k−mπ dν1 ν12   
− θ(k − mπ ) v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε) t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) .
2π 2 0 k2 − ν12
Exchanging the order of the integration limits in the first integral we can combine the two integrals in one as

t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) = v(iν + ε− , ik + ε) − v(iν + ε+ , ik + ε)


Z
mi k−mπ dν1 ν12  
− θ(k − mπ ) 2 v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε)
2π 2 (ν+mπ )θ(|ν|−mπ )
2
k − ν1
 
× t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) . (7.113)

19
The denominator in the integrand of Eq. (7.110) does not vanish neither along the original integration contour nor the added vertical ones.

69
ε− ε ε ε+

Figure 16: Vertical contours used in Eq. (7.114) corresponding to the case ν > 0. Left panel applies to t(iν + ε− , ik + ε)
and right panel to t(iν + ε+ , ik + ε).

• For the case ν > 0 we have to consider the vertical contours shown in Fig. 16, which are present for ν > mπ and
extend from 0 up to ν − mπ . These are the only ones that give contribution to the difference of partial waves under study.
Then, following the same steps that drove us to Eqs. (7.111) and (7.112), we have now
Z
− + − im +
ν−mπ dν1 ν12
t(iν + ε , ik + ε) − t(iν + ε , ik + ε) = v(iν + ε , ik + ε) − v(iν + ε , ik + ε) + θ(ν − mπ ) 2
2π 0 k2 − ν12
 − − − −  − + 
× v(iν + ε , iν1 + ε + δ) − v(iν + ε , iν1 + ε − δ) t(iν1 + ε , ik + ε) − t(iν1 + ε , ik + ε) (7.114)
Z
im k−mπ dν1 ν12   
− θ(k − mπ ) v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε) t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) .
2π 2 0 k2 − ν12

Let us notice that now both integrals run through positive values of ν1 so that they subtract each other, because of the
different sign in front of them, and we have an analogous expression to Eq. (7.113),

t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) = v(iν + ε− , ik + ε) − v(iν + ε+ , ik + ε)


Z
mi k−mπ dν1 ν12  
− θ(k − mπ ) 2 2 2 v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε)
2π (ν−mπ )θ(ν−mπ ) k − ν1
 
× t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) . (7.115)

The sign of ν is denoted as ǫ(ν), namely,


ν
ǫ(ν) = . (7.116)
|ν|

To shorten notation we indicate in the subsequent the lower limit of integration in Eqs. (7.113) and (7.115) as

ϑ(ν) = ǫ(ν)(|ν| − mπ )θ(|ν| − mπ ) . (7.117)

We can then express both Eqs. (7.113) and (7.115) in the same manner as

t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) = v(iν + ε− , ik + ε) − v(iν + ε+ , ik + ε)


Z
mi k−mπ dν1 ν12  
− θ(k − mπ ) 2 v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε)
2π ϑ(ν) k2 − ν12
 
× t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) . (7.118)

70
• The previous equation allows us to conclude the continuity of Ret(iν + ε ± δ, ik + ε), ν ∈ [−k, k − mπ ], for δ → 0
in a more straightforward manner than the one used to demonstrate it in Eqs. (7.24) and (7.82). This continuity is clear
for Rev(iν + ε± , ik + ε) and Rev(iν + ε± , iν1 + ε) with ε± → ε, see e.g. Eq. (3.86). As a result, the discontinuities in
the potential in Eq. (7.118) are purely imaginary, which allows to rewrite this equation as
 
t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) = i Imv(iν + ε− , ik + ε) − Imv(iν + ε+ , ik + ε)
Z
m k−mπ dν1 ν12  
+ θ(k − mπ ) 2 2 2 Imv(iν + ε− , iν1 + ε) − Imv(iν + ε+ , iν1 + ε)
2π ϑ(ν) k − ν1
 
× t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) . (7.119)

Because the independent term in this IE is purely imaginary while the kernel is purely real it is clear that the solution of
this IE, that corresponds to the discontinuity of t̂(iν + ε ± δ, ik + ε) with δ → 0, is purely imaginary.20
Taking now into account this continuity of Ret(iν1 + ε ± δ, ik + ε) for δ → 0, the previous IE becomes

Imt(iν + ε− , ik + ε) − Imt(iν + ε+ , ik + ε) = Imv(iν + ε− , ik + ε) − Imv(iν + ε+ , ik + ε)


Z
m k−mπ dν1 ν12  
+ θ(k − ν − 2mπ ) 2 Imv(iν + ε− , iν1 + ε) − Imv(iν + ε+ , iν1 + ε)
2π 2 ϑ(ν)
2
k − ν1
 
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) . (7.120)

The function ϑ(ν) in the previous expression can be further simplified by noticing that the discontinuity of the potentials
inside the integrand stems from the difference of the log factors,
   
log − (ν + ν1 )2 + m2π + i0+ (ν + ν1 ) − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+
n    o
− log − (ν + ν1 )2 + m2π + i0+ (ν + ν1 ) − log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+
   
= − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+ + log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+ . (7.121)

This is not zero only for (ν − ν1 )2 > m2π which requires that
q
ν1 = ν ± m2π + x2 , x ∈ R . (7.122)

But ϑ(ν) already implies that ν1 ≥ ν − mπ , Eq. (7.117). Then, of the two branches in the possible values for ν1 in
Eq. (7.122) only the one with the plus sign matters. As a result, we can rewrite Eq. (7.120) as

Imt(iν + ε− , ik + ε) − Imt(iν + ε+ , ik + ε) = Imv(iν + ε− , ik + ε) − Imv(iν + ε+ , ik + ε)


Z
m k−mπ dν1 ν12  − + 
+ θ(k − ν − 2mπ ) 2 Imv(iν + ε , iν 1 + ε) − Imv(iν + ε , iν 1 + ε)
2π 2 ν+mπ k 2 − ν1
 
× Imt(iν1 + ε − δ, ik + ε) − Imt(iν1 + ε + δ, ik + ε) . (7.123)

• From Eq. (7.120) let us derive a new equation free of the integration problems that could arise because of the infrared
divergences for ℓ ≥ 1, which stem from the fact that Imv(iν + ε± , iν1 + ε) and Imt(iν1 + ε ± δ, ik + ε) diverge like 1/ν1ℓ+1
2(ℓ+1)
for ν1 → 0, so that its product does like 1/ν1 , see e.g. the discussion here below Eq. (7.126).
20
It is important to note that the use of this property of continuity for Rev̂ and Ret̂ is applied for finite ε. That is, the integrations are done
with ε 6= 0 but with ε± → ε and δ → 0.

71
To accomplish that we first rewrite Eq. (7.120) as
Imt(iν + ε− , ik + ε) − Imt(iν + ε+ , ik + ε) = Imv(iν + ε− , ik + ε) − Imv(iν + ε+ , ik + ε)
Z n
θ(k − mπ )m k−mπ dν1 ν12 
− 2 Re v(iν + ε− , iν1 + ε) − v(iν + ε+ , iν1 + ε)
2π 2 ϑ(ν)
2
k − ν1
 o
× t(iν1 + ε − δ, ik + ε) − t(iν1 + ε + δ, ik + ε) , (7.124)

where we have used again the continuity of Rev(iν + ε± , iν1 + ε) and Ret(iν1 + ε ± δ, ik + ε) for δ → 0, Eq. (7.119),
that allows us to neglect the contribution
Z
θ(k − mπ )m k−mπ dν1 ν12  
− 2 Rev(iν + ε− , iν1 + ε) − Rev(iν + ε+ , iν1 + ε)
2π 2 ϑ(ν)
2
k − ν1
 
× Ret(iν1 + ε − δ, ik + ε) − Ret(iν1 + ε + δ, ik + ε) −→ 0 . (7.125)
δ→0

We remark here again, as in the footnote 20, that the continuity referred is in ε± → ε and δ → 0. This is the situation
in Eq. (7.125) because one first takes these limits while keeping ε 6= 0. In this way the integrals remain finite and there is
then no ambiguous limit of the type 0 × ∞, that could appear because of the infrared divergences if simultaneously with
ε± → ε and δ → 0 one also took the limit ε → 0 in the integrations.
Next, we employ the functions v̂(k, k′ ) and t̂(k, k′ ; z), already introduced in Eq. (7.42), whose definition we reproduce
here for completeness
ℓ+1
v̂(k, k′ ) = kℓ+1 k′ v(k, k′ ) ,
ℓ+1
t̂(k, k′ ; z) = kℓ+1 k′ t(k, k′ ; z) . (7.126)

In this way the divergent behavior of Imv(iν + ε± , iν1 + ε) as 1/ν ℓ+1 (1/ν1 ℓ+1 ) for ν(ν1 ) → 0 is not longer present
in v̂(iν + ε± , iν1 + ε). This is clear if one thinks in terms of a spectral decomposition for each of the components of
v(iν + ε± , iν1 + ε), see e.g. Appendix B of Ref. [4], and for each of them one keeps only the contribution that gives rise
to the log-term after the angular projection (as also done for the uncoupled case, cf. Eq. (3.86)). The same can be said
for t(iν1 + ε + ±δ, ik + ε) and t̂(iν + ε± , ik + ε) as it is clear by iterating the LS equation. Furthermore, the function
Imv̂(iν + ε± , iν1 + ε) is a polynomial in ν 2 and ν12 , as already discussed above, cf. Eq. (7.56).
Then, we can extract 1/[(iν)ℓ+1 (ik)ℓ+1 ] as a common factor in Eq. (7.124) because i2ℓ+2 = (−1)ℓ+1 is real. To simplify
the notation we also utilize the functions ∆v̂(ν, ν1 ) and f(ν1 ), cf. Eqs. (7.93) and (7.61), respectively, which definition is
also reproduced here
∆v̂(ν, ν1 ) = Imv̂(iν + ε− , iν1 + ε) − Imv̂(iν + ε+ , iν1 + ε) , (7.127)
f(ν) = Imt̂(iν + ε − δ, ik + ε) − Imt̂(iν + ε + δ, ik + ε) , (7.128)
Expanding the real part in the integrand of Eq. (7.124) we can rewrite it as
f(ν) = ∆v̂(ν, k)
Z n
θ(k − mπ )m k−mπ dν1 ν12 1 
− 2 Re Re v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε)
2π 2 ϑ(ν)
2
k − ν1 (iν1 + ε)n
 o
× t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
Z n
θ(k − mπ )m k−mπ dν1 ν12 1 
+ 2 Im Im v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε)
2π 2 ϑ(ν)
2
k − ν1 (iν1 + ε)n
 o
× t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε) , (7.129)

72
with

n = 2ℓ + 2 , (7.130)

as already introduced in Eq. (7.41). The last integral in Eq. (7.129) does not contribute because of two combined facts.
On the one hand,
 
1 1 1 1
Im = − (7.131)
(iν1 + ε)n 2i (iν1 + ε)n (−iν1 + ε)n
is null for finite ν1 . On the other hand,
n  o
Im v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε) t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
   
= Re v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε) Im t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
   
+ Im v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε) Re t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε) (7.132)

is zero in the limit ε± → ε and δ → 0 because of the continuity of both Rev̂(iν + ε± , iν1 + ε) and Ret̂(iν1 + ε ± δ, ik + ε),
in order. The same remark as already done twice above, in footnote 20 and just after Eq. (7.125), is in order here.21
Then, Eq. (7.129) can be finally written as
Z  
θ(k − mπ )m k−mπ dν1 ν12 1 1
f(ν) = ∆v̂(ν, k) + + ∆v̂(ν, ν1 )f(ν1 ) , (7.133)
4π 2 ϑ(ν) k2 − ν12 (iν1 + ε)n (iν1 − ε)n

where we have used again the continuity of Rev̂(iν + ε, iν1 + ε+ ± δ) and Ret̂(iν1 + ε ± δ, ik + ε) for δ → 0 to express
n  o
Re v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε) t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε)
   
→ −Im v̂(iν + ε− , iν1 + ε) − v̂(iν + ε+ , iν1 + ε) Im t̂(iν1 + ε − δ, ik + ε) − t̂(iν1 + ε + δ, ik + ε) . (7.134)

• Eq. (7.133) can be furthered simplified by taking into account that ∆v̂(ν, ν1 ) stems from the imaginary part in the
difference of log-terms in Eq. (B.1), namely,
   
∆v̂(ν, ν1 ) ∝ log − (ν + ν1 )2 + m2π + i0+ (ν + ν1 ) − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+
n    o
− log − (ν + ν1 )2 + m2π + i0+ (ν + ν1 ) − log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+
   
= − log − (ν − ν1 )2 + m2π − i(ν − ν1 )0+ + log − (ν − ν1 )2 + m2π + i(ν − ν1 )0+ . (7.135)

where the proportionality coefficient is ρ(ν 2 , ν12 )/π, as introduce in Eq. (7.56). The first two log-terms in Eq. (7.135)
correspond to Imv̂(iν + ε− , iν1 + ε) and the last two (shown between curly brackets) to Imv̂(iν + ε+ , ik + ε). It follows
from this equation that in order to end with a nonvanishing imaginary part it is required that (ν1 − ν)2 > m2π or, in other
terms, that
q
ν1 = ν ± m2π + x2 , x ∈ R . (7.136)
21
It is worth pointing out that the derivation given here is simpler than the one offered in Secs. 7.2 and 7.3 because we have directly considered
the difference of the partial waves t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε). However, in those sections we started by considering separately each
partial wave in this difference and then it was necessary to make use of extra properties to make sense of the resulting expressions. Namely,
for ν > 0 we have Eq. (7.53) for t(iν + ε+ , ik + ε) and Eq. (7.54) that applies to t(iν + ε− , ik + ε) (as well as their counterparts for ν < 0,
Eqs. (7.87) and (7.88), respectively). It was then used in those equations that Rev̂(iν + ε′ , iν1 + ε) (ν 6= mπ ) and Ret̂(iν1 + ε± , ik + ε) vanish
in the limit ν1 → 0 as ν12ℓ+1 . This factor times the ν12 from the measure vanishes fast enough to cancel the divergence in Eq. (7.131) when
ν1 → 0.

73
Due to ϑ(ν) in the integral of Eq. (7.133) only the plus sign matters as ν1 > ν − mπ . Thus, ν1 > ν + mπ and in such
circumstances the imaginary part of the rhs of Eq. (7.135) is −2π. This also implies that for the whole integration interval
in Eq. (7.133) it holds that

∆v̂(ν, ν1 ) = −2ρ(ν 2 , ν12 ) , ν1 ∈ [ν + mπ , k − mπ ] . (7.137)

The same result could be applied also to the independent term in Eq. (7.133) since k > ν + mπ because k − mπ > ν > −k
and k > mπ .
The fact that the integrand in Eq. (7.133) is nonzero only for ν1 > ν + mπ also implies that one can replace ϑ(ν) by
ν + mπ since

ν + mπ ≥ ǫ(ν)(|ν| − mπ )θ(|ν| − mπ ) . (7.138)

Of course, it is then required that k − mπ > ν + mπ because otherwise ∆v̂(ν, ν1 ) is zero, a condition that can be accounted
for by the presence of θ(k − 2mπ − ν) in front of the integral in Eq. (7.133). As a result we can rewrite this equation as
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = −2ρ(ν 2 , k2 ) − + ρ(ν 2 , ν12 )f(ν1 ) . (7.139)
2π 2 mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n

Notice also that this form of the IE is valid for all values of ν of interest.
• Now, let us discuss how to handle the previous IE in an appropriate form ready to be used to cure numerically the
problem of the infrared divergences for ℓ ≥ 1. We follow here closely the derivations performed at the end of Sec. 7.3,
give a summary of the main steps there and slightly modify some formulas to adapt them to the form of Eq. (7.139).
The problem with the infrared divergences originates when the integration in Eq. (7.139) approaches or crosses the
value ν = −mπ , which happens for ν ≤ −mπ + ∆, with ∆ > 0 but small so that numerically one would get problems in
the evaluation of the integral to solve the IE. The key point is to make use of the symmetry under the exchange ν1 ↔ −ν1
in the function ρ(ν 2 , ν12 ) and symmetrize partially the integral in Eq. (7.139) for ν + mπ < 0, which can then be written
as
Z  
2 2 θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = −2ρ(ν , k ) − 2 2 n
+ ρ(ν 2 , ν12 )f(ν1 )
2π 2
−mπ −ν k − ν1 (iν1 + ε) (iν1 − ε)n
Z  
θ(k − 2mπ − ν)m −ν−mπ dν1 ν12 1 1  
− 2 2 2 n
+ n
ρ(ν 2 , ν12 ) f(ν1 ) + f(−ν1 ) . (7.140)
4π mπ +ν k − ν1 (iν1 + ε) (iν1 − ε)
The last integral can be restricted to positive values of ν1 because the integrand is even and then we have
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = −2ρ(ν 2 , k2 ) − 2 2 2 n
+ ρ(ν 2 , ν12 )f(ν1 )
2π −mπ −ν k − ν1 (iν1 + ε) (iν1 − ε)n
Z  
θ(k − 2mπ − ν)m −ν−mπ dν1 ν12 1 1  
− 2 2 2 n
+ n
ρ(ν 2 , ν12 ) f(ν1 ) + f(−ν1 ) . (7.141)
2π 0 k − ν1 (iν1 + ε) (iν1 − ε)
Still the first integral presents numerical problems for ℓ > 0 when ν → −mπ which can be solved by expressing the IE
after a first iteration. The process is straightforward and explicit steps are worked out in Eqs. (7.102) and (7.103), as well
as in the explanation of the 3rd method in Appendix 15 for the explicit example of the 1 P1 uncoupled wave. The result is
Z
m k−mπ dν1 ν12
f(ν) = −2ρ(ν 2 , ν12 ) + θ(k − 2mπ − ν) 2 2 2 2
2 Sn (ν1 )ρ(ν , ν1 )ρ(ν1 , k )
π2 2
mπ +ν k − ν1
 2 Z Z ν2 −mπ
m k−mπ dν2 ν22 dν1 ν12
+ θ(k − 3mπ − ν) S n (ν 2 )f(ν 2 ) Sn (ν1 )ρ(ν 2 , ν12 )ρ(ν12 , ν22 ) . (7.142)
2π 2 2mπ +ν k2 − ν22 mπ +ν k2 − ν12

74
There is an exchange of variables in the double integral which is clearly represented in Fig. 36. In the previous equation
we have used the function Sn (ν1 ) defined in Eq. (7.101) as
1 1
Sn (ν1 ) = n
+ . (7.143)
(iν1 + ε) (iν1 − ε)n

Then, we have at our disposal several expressions that are used in different intervals of ν in order to avoid the infrared
singularities for ℓ > 0


 Eq. (7.139) −mπ /2 < ν
f(ν) = Eq. (7.142) −3m /2 < ν < −mπ /2
π

 Eq. (7.141) −3m /2 > ν
π

(7.144)

• In order to apply numerically Eq. (7.141) for −3mπ /2 > ν one could follow a similar procedure as explained

in Sec. 7.2,

2
by adding and subtracting the first ℓ terms in the Taylor series in powers of ν1 of the even function f(ν1 ) + f(−ν1 ) .
In this way the difference between this function and the first ℓ terms from its Taylor series expansion around zero makes
the last integral of Eq. (7.141) be explicitly convergent. Note that for ν1 → 0 this difference times ν12 from the measure
vanishes as ν1n which cancels the divergence in Sn (ν1 ) → ν1−n . The added terms from the Taylor series expansion give
ν1 →0
rise to an integral that can be done algebraically and for which convergence the presence of Sn (ν1 ) is crucial. E.g. for the
D waves, one can substitute in the integrand

 d2 f(0) 2  d2 f(0) 2
f(ν1 ) + f(−ν1 ) → f(ν1 ) + f(−ν1 ) − 2f(0) − ν1 + 2f(0) + ν . (7.145)
dν1 2 dν12 1

So that the first term between squared brackets is free of infrared divergence and for the rest the integration is done
algebraically. For evaluating the first ℓ coefficients of the Taylor expansion in powers of ν 2 of the even function (f(ν) +
f(−ν))/2 at ν = 0, f(0), d2 f(0)/dν 2 , . . ., d2(ℓ−1) f(0)/dν 2(ℓ−1) one can readily use Eq. (7.139). This IE can be solved from
top to bottom, because for ν between k − 2mπ and k − mπ the solution is given by the independent term. Then, for ν
in the interval between k − 3mπ to k − 2mπ we employ in the integrand the values previously determined, and so on. To
evaluate f(ν) up to ν = 0 we need to know f(ν) up to ν = mπ in the integration, a process that does not involve any
infrared divergence.

7.5 Coupled-channel case


Note: A more detailed account of the derivation of the discontinuity of an on-shell scattering amplitude across the LHC,
valid also for a potential expressed in terms of a general spectral mass representation, is given in Sec. 8.
Here we discuss the main points involved in the straightforward generalization of Eq. (7.139) to the coupled-channel
case. We restrict to N N scattering for which the particles involved in all channels are the same (analogous consideration
could be applied to any other nonrelativistic two-body scattering that shares this simplifying property of N N ). This is our
key assumption since in this way the kinematics of the different channels is the same as well as the nature of the particles
exchanged that give rise to the LHC (pions here).
In this section we introduce subscripts to indicate the different channels. Because of parity and rotational invariance
only the channels with orbital angular momenta J ±1 are coupled, with J the total angular momentum. Then we denote by
vij (k, k′ ) the partial-wave projected potential and by tij (k, k′ ; z) the partial wave amplitude. The orbital angular momenta
associated to the channels i and j are indicated by ℓi and ℓj , in this order.
The analytical properties of vij (k, k′ ) are the same as those already discussed in Sec. 3 (for this to be true it is of course
essential that all particles are the same in the different coupled-channels). These properties in turn fix the dynamical

75
cuts in tij (k, k′ ; k′ 2 /m) as discussed in Sec. 6. Other properties that we have also used in one-channel scattering, and
that hold as well in the coupled-channel case, are: i) vij (k∗ , k′ ∗ ) = vij (k, k′ )∗ and tij (k∗ , k′ ∗ ; z ∗ ) = t(k, k′ ; z)∗ , as
follows by performing the same steps as in Eqs. (7.4) and (7.6), respectively; ii) the Schwarz reflection principle, which is a
consequence of i) for real k and z = k′ 2 /m, and iii) that tij (−k, k′ ; k′ 2 /m) = (−1)ℓi tij (k, k′ ; k′ 2 /m), tij (k, −k′ ; k′ 2 /m) =
(−1)ℓj tij (k, k′ ; k′ 2 /m) = (−1)ℓi tij (k, k′ ; k′ 2 /m), because the coupled channels have the same parity (ℓi − ℓj = mod 2).
This fact is important as we can then use with respect to these parity transformation properties either ℓi or ℓj , without
distinguishing between the different orbital angular momenta coupled.
We can then repeat readily the analysis at the beginning of Sec. 7.3 and conclude that Imtij (ik − ε− , ik + ε) and
Imtij (ik + ε, ik + ε) are the same so that in order to calculate ∆(A) it is enough to calculate fij (ν) and particularize it at
ν = −k, cf. Eq. (7.79). In order to obtain the IE for fij (ν) we can reproduce the process in Sec. 7.4, presented there for
ν < 0, but the same steps can be applied for ν > 0 (so that now the the integration along the vertical segment originating
from v̂iℓ (iν + ε± , iν1 + ε± ± δ) extends up to ν − mπ for ν > mπ ). Let us perform first the derivation for ν < 0 and later
we derive a common IE for negative and positive ν. Instead of Eq. (7.113) we have,

tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)


Z
θ(k − mπ )mi X k−mπ dν1 ν12  − + 
− 2 vik (iν + ε , iν 1 + ε) − vik (iν + ε , iν 1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ ) k 2 − ν1
 
× tkj (iν1 + ε − δ, ik + ε) − tkj (iν1 + ε + δ, ik + ε) . (7.146)

Now we introduce the continuity of Retij (iν1 + ε ± δ, ik + ε) when δ → 0, that is,

lim Retij (iν1 + ε + δ, ik + ε) = lim Retij (iν1 + ε − δ, ik + ε) . (7.147)


δ→0 δ→0

This result was proved for the uncoupled case in Eq. (7.24) for ν > 0 and in Eq. (7.82) for ν < 0. Analogous steps could
be also repeated here for the coupled-channel case by just including subscripts explicitly. Taking this result into account
Eq. (7.146) simplifies to

Imtij (iν + ε− , ik + ε) − Imtij (iν + ε+ , ik + ε) = Imvij (iν + ε− , ik + ε) − Imvij (iν + ε+ , ik + ε)


Z
θ(k − mπ )mi X k−mπ dν1 ν12  
+ 2 Imvik (iν + ε− , iν1 + ε) − Imvik (iν + ε+ , iν1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ )
2
k − ν1
 
× Imtkj (iν1 + ε − δ, ik + ε) − Imtkj (iν1 + ε + δ, ik + ε) . (7.148)

In order to end with an IE that cures the problem of infrared singularities we rewrite the previous equation as

Imtij (iν + ε− , ik + ε) − Imtij (iν + ε+ , ik + ε) = Imvij (iν + ε− , ik + ε) − Imvij (iν + ε+ , ik + ε)


Z n
θ(k − mπ )m X k−mπ dν1 ν12 
− 2 Re vik (iν + ε− , iν1 + ε) − vik (iν + ε+ , iν1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ )
2
k − ν1
 o
× tkj (iν1 + ε − δ, ik + ε) − tkj (iν1 + ε + δ, ik + ε) , (7.149)

where we have used again the continuity of Retij (iν1 + ε± δ, ik + ε), Eq. (7.147), that allows us to neglect the contribution
Z
θ(k − mπ )mi X k−mπ dν1 ν12  − + 
− 2 Revik (iν + ε , iν 1 + ε) − Revik (iν + ε , iν 1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ ) k 2 − ν1
 
× Retkj (iν1 + ε − δ, ik + ε) − Retkj (iν1 + ε + δ, ik + ε) → 0 . (7.150)
δ→0

76
Next, we introduce the new functions v̂ij (k, k′ ) and t̂ij (k, k′ ; z), analogous to those in Eq. (7.42) for the uncoupled case,
and define them as
ℓj +1
v̂ij (k, k′ ) = kℓi +1 k′ vij (k, k′ ) ,
ℓj +1
t̂ij (k, k′ ; z) = kℓi +1 k′ tij (k, k′ ; z) . (7.151)

In this way the divergent behavior of Imvij (iν, iν1 ) as 1/ν ℓi (1/ν1 ℓj ) for ν(ν1 ) → 0 is not longer present when using
v̂ij (iν, iν1 ). This is clear if one thinks in terms of a spectral decomposition for each of the components of vij (iν, iν1 ), see
e.g. Appendix B of Ref. [4], and for each of them one keeps only the contribution that gives rise to the log-term after
the angular projection (as also done for the uncoupled case, cf. Eq. (3.86)). The same can be said for tij (iν, iν1 ; z) and
t̂ij (iν, iν1 ; z) as it is clear by iterating the LS equation. Furthermore, the function Imv̂ij (iν, iν1 ) is a polynomial in ν 2 and
ν12 , as in the uncoupled case, cf. Eq. (7.56).
Then, we can extract 1/[(iν)ℓi +1 (ik)ℓj +1 ] as a common factor in Eq. (7.149) because iℓi +ℓj +2 = (−1)ℓi +ℓj = 1. To
simplify the notation we include the analogous functions ∆v̂ij (ν, ν1 ) and fij (ν1 ) for the coupled channel case,

∆v̂ij (ν, ν1 ) = Imv̂ij (iν + ε− , iν1 + ε) − Imv̂ij (iν + ε+ , iν1 + ε) , (7.152)


fij (ν) = Imt̂ij (iν + ε − δ, ik + ε) − Imt̂ij (iν + ε + δ, ik + ε) , (7.153)

see Eqs. (7.93) and (7.61), respectively, for the uncoupled waves. By expanding the real part in the integrand we can write
Eq. (7.149) as

fij (ν) = ∆v̂ij (ν, k)


Z n
θ(k − mπ )m X k−mπ dν1 ν12 1 − + 
− 2 Re Re v̂ik (iν + ε , iν 1 + ε) − v̂ik (iν + ε , iν 1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ ) k 2 − ν1 (iν1 + ε)nk
 o
× t̂kj (iν1 + ε − δ, ik + ε) − t̂kj (iν1 + ε + δ, ik + ε)
Z n
θ(k − mπ )m X k−mπ dν1 ν12 1 
+ 2 Im Im v̂ik (iν + ε− , iν1 + ε) − v̂ik (iν + ε+ , iν1 + ε)
2π 2 k=1,2 (ν+mπ )θ(|ν|−mπ )
2
k − ν1 n
(iν1 + ε) k
 o
× t̂kj (iν1 + ε − δ, ik + ε) − t̂kj (iν1 + ε + δ, ik + ε) , (7.154)

with

nk = 2ℓk + 2 . (7.155)

The last integral in Eq. (7.154) does not contribute because of two combined facts. On the one hand,
 
1 1 1 1
Im n
= n
− (7.156)
(iν1 + ε) k 2i (iν1 + ε) k (−iν1 + ε)nk
is null for finite ν1 . On the other hand,
n  o
Im v̂ik (iν + ε− , iν1 + ε) − v̂ik (iν + ε+ , iν1 + ε) t̂kj (iν1 + ε − δ, ik + ε) − t̂kj (iν1 + ε + δ, ik + ε)
   
= Re v̂ik (iν + ε− , iν1 + ε) − v̂ik (iν + ε+ , iν1 + ε) Im t̂kj (iν1 + ε − δ, ik + ε) − t̂kj (iν1 + ε + δ, ik + ε)
   
+ Im v̂ik (iν + ε− , iν1 + ε) − v̂ik (iν + ε+ , iν1 + ε) Re t̂kj (iν1 + ε − δ, ik + ε) − t̂kj (iν1 + ε + δ, ik + ε) (7.157)

vanishes for ν1 → 0 at least as ν12ℓk +1 because both Rev̂ik (iν + ε± , iν1 + ε) and Ret̂kj (iν1 + ε ± δ, ik + ε) tend to ν12ℓk +1
as ν1 → 0. This property was demonstrated at the end of Sec. 7.2 for the uncoupled case. Here one could follow the same

77
steps in the demonstration for the coupled-channel scattering. As a result the product of Eqs. (7.156) and (7.157) times
ν12 from the measure vanishes at least as ν1 for ν1 → 0. Then, Eq. (7.154) can be finally written as
Z  
θ(k − 2mπ − ν)m X k−mπ dν1 ν12 1 1
fij (ν) = ∆v̂ij (ν, k) + + ∆v̂ik (iν, iν1 )fkj (ν1 ) .
4π 2 k=1,2 ν+mπ k2 − ν12 (ν1 + iε)nk (ν1 − iε)nk
(7.158)

Notice that we have dropped the factor θ(|ν| − mπ ) for ν < 0 because this is implicitly included in the discontinuity of
the potential, v̂(ν, ν1 ), as explained after Eq. (7.89).
For ν > 0 we could follow the same steps as performed here but instead the last integral in the previous equation would
extend from (ν − mπ )θ(ν − mπ ). In this way we will directly end with the analogous IE to Eq. (7.60). Next, making use of
the remark after this equation, that gives rise to Eq. (7.61), one ends with the final form of the IE for the coupled-channel
case (valid for both negative and positive ν)
Z  
θ(k − 2mπ − ν)m X k−mπ dν1 ν12 1 1
fij (ν) = ∆v̂ij (ν, k) + + ∆v̂ik (ν, ν1 )fkj (ν1 ) .
4π 2 k=1,2 ν+mπ k2 − ν12 (ν1 + iε)nk (ν1 − iε)nk

(7.159)

To end with a meaningful procedure to calculate fij (ν) one should follow the same steps as explained in between
Eqs. (7.99)–(7.104) to handle with the infrared divergences in the numerator of Eq. (7.159). They can also be used here
since we can symmetrize the integration interval between ν + mπ and −(ν + mπ ) for ν < −mπ , as this is based in the
log-term of v̂ij (iν, iν1 ), which is the same as in the uncoupled case. Then, we can also use the IE in the form resulting
after one iteration, cf. Eq. (7.103), to calculate fij (ν) around ν ≃ −mπ . Of course, to handle efficiently the numerical
integrations one should isolate those integrals with infrared divergent integrands and doing them algebraically, as explain
in Eq. (7.109). All these steps are exemplified for the partial wave 1 P1 in Sec. (15).

8 Potential with arbitrary spectral decomposition in coupled channels


In this section we take from the beginning a potential with an arbitrary spectral mass decomposition. We also consider
several two-body channels coupled, all of them made up of the same particle. This is our key assumption since in this way
the kinematics of the different channels is the same as well as the nature of the particles exchanged that give rise to the
LHC. In this regard our prototype example is N N scattering, as usual.22
We introduce latin subscripts to indicate the different channels. Then, we denote the partial-wave projected potential
and scattering amplitude for the transition j → i by vij (k, k′ ) and by tij (k, k′ ; z), respectively. Because of rotational
invariance all the coupled channels have the same total angular momentum J. The orbital angular momenta associated
to the channels i and j are indicated by ℓi and ℓj , in this order, and because of parity invariance (−1)ℓi = (−1)ℓj , since
one has the same intrinsic parities for the initial and final states.23
We assume a spectral representation for vij (k, k′ ), cf. Eq. (3.75) for the uncoupled case,
Z
1 ∞ 2µηij (µ2 )
vij (p1 , p2 ) = dµ . (8.1)
π 0 q 2 + µ2
22
Nevertheless, our considerations could be easily extended for a two-body channel involving different particles.
23
For the case of N N scattering (in general for any nonrelativistic 1/2 two-particle scattering) parity and rotational invariance altogether
require that only the channels with orbital angular momenta J ± 1 are coupled.

78
In the following we denote by mπ the lowest value of µ above which the spectral functions ηij (µ) start deviating from
zero. Then, the cuts of the partial-wave projected potential vij (k, k′ ) as a function of k in its complex plane are
q
k = (±)k′ ± i m2π + x2 , (8.2)

as follows from Sec. (3.3), where the analytical properties of the partial-wave projected potential v(k, k′ ) were studied.24
The extent of the cuts in the k′ variable also corresponds to Eq. (8.2) by exchanging k and k′ .
Let us notice that the contributions that stem from µ > mπ in Eq. (8.1) give rise to cuts in the partial-wave projected
potential vij (k, k′ ) that are comprised in Eq. (8.2) as they extend along the values
q
k = (±)k′ ± i µ2 + x2 . (8.3)

One should keep in mind that time-reversal invariance requires not only the exchange between k and k′ but also between
the indices i and j, namely,

vij (k, k′ ) = vji (k′ , k) . (8.4)

8.1 Relationships between partial waves with different arguments


• We are interested in the study of the partial-wave projected LS equation in coupled channels which reads
Z
m X dp3 p23
tij (p1 , p2 , z) = vij (p1 , p2 ) + 2 viℓ (p1 , p3 )tℓj (p3 , p2 , z) , (8.5)
2π ℓ C p23 − 2mz

where the sum in ℓ extends over the coupled channels. This equation can also be written as
Z
m X dp3 p23
tij (p1 , p2 , z) = vij (p1 , p2 ) + 2 2 tiℓ (p1 , p3 )vℓj (p3 , p2 , z) , (8.6)
2π ℓ C p3 − 2mz

The integration contour C has to be fixed according to the external arguments p1 , p2 and z from the analytical continuation
of the original contour p3 ∈ [0, ∞] for real p1 , p2 and z.
• For the on-shell LS equation Eq. (8.5) reads
Z
p2 m X ∞ dp1 p21
t(p, p; ) = v(p, p) + 2 viℓ (p, p1 )tℓj (p1 , p) , (8.7)
m 2π ℓ 0 p21 − p2

with p ∈ R. We next extend the variable p to its complex plane. We are interested in those values ofpp corresponding to
the DCs that result from Eq. (8.2) (when taking k = k′ the on-shell cuts originate from p = −p ± i m2π + x2 , x ∈ R),
so that we take

p = i k ± ε → p2 = −k2 ± i0+ . (8.8)

For such values one has to deform the original integration contour in Eq. (8.7) so as to avoid the crossing with the DCs.
The latter stem from the potential and happen for the values
q
p1 = ±(ik ± ε)(±)i m2π + x2 , x ∈ R , (8.9)
24
If the two particles involved in the two-body scattering channel were different the threshold value mπ would correspond to the lowest one of
those associated with the different potentials vij (k, k′ ).

79
Since p1 ∈ [0, +∞] in Eq. (8.7) the crossing with the DCs of the original integration contour only happens when k > mπ
and corresponds to the choice of the pair of signs +(−) or −(+) in Eq. (8.9), depending whether one has +ε or −ε,
respectively. As a result we have the contours shown in Fig. 10, C + (left panel) and C − (right panel), respectively. Notice
that C − is the complex conjugate of C + . The vertical cuts are added at ε and extends from 0 up to ±(|Imk| − mπ ) for
C ±.
• For clarification, let us stress that the contribution in the spectral representation of the potential with mπ < µ < |Imk|
would give rise to the same type of vertical additions but with shorter extension, up to ±(|Imk| − |µ|). These deformations
are then incorporated in the vertical addition attached to the lowest value of µ, µ = mπ .
• From the spectral decomposition of the potential it is clear the following property

vij (p∗1 , p∗2 ) = vij (p1 , p2 )∗ . (8.10)

Let us see that this is also the case for the full partial wave amplitude

p∗2 2 p2
tij (p∗1 , p∗2 ; ) = tij (p1 , p2 ; 2 )∗ , (8.11)
m m
extending the demonstration in Sec. (7.1), given between Eqs. (7.4) and (7.6). To obtain this result let us take into
account that the integration deformation that is needed for calculating tij (p∗1 , p∗2 ; p∗2 2 /m), indicated by C ∗ , is the complex
conjugate of the one used for tij (p1 , p2 ; p22 /m), denoted by C. This statement is clear because the DCs in vij (p∗1 , p∗2 ) occur
at complex conjugate positions of those in vij (p1 , p2 ). Then we have
Z
p∗2 2 m X dp∗3 p∗3 2 ∗ ∗ p2
∗2
tij (p∗1 , p∗2 ; ) = vij (p∗1 , p∗2 ) + 2 v (p ∗ ∗
, p )t
iℓ 1 3 ℓj 3 2(p , p ; )
m 2π ℓ C ∗ p∗3 2 − p∗2 2 m
( Z )∗
m X dp3 p23 ∗2
∗ ∗ p2 ∗
= vij (p1 , p2 ) + 2 v (p , p )t (p
iℓ 1 3 ℓj 3 2 , p ; ) . (8.12)
2π ℓ C p23 − p22 m

Hence, by taking the complex conjugate of the previous equation it is clear that both t(p∗1 , p∗2 ; p∗2 2 /m)∗ and t(p1 , p2 , p22 /m)
satisfy the same LS equation in coupled channels, and as a result they must be the same.25
Notice that the Schwarz reflection principle is a particular case of Eq. (8.11) when restricted to real p1 , so that

p∗2 2 p2
tij (p1 , p∗2 ; ) = tij (p1 , p2 ; 2 )∗ , p1 ∈ R , (8.13)
m m
or when considering on-shell scattering p1 = p2 = p,

p∗2 p2
tij (p∗ , p∗ ; ) = tij (p, p; )∗ . (8.14)
m m
• Now, for real p1 and p2 the half-off-shell LS equation is
Z
p22 m X ∞ dp3 p23 p22
tij (p1 , p2 ; ) = vij (p1 , p2 ) + 2 viℓ (p 1 , p 3 )t ℓj (p 3 , p 2 ; ). (8.15)
m 2π ℓ 0 p23 − p22 m

It is clear that if we exchange the sign of p1 the rhs of the equation does change by a factor (−1)ℓi and then

p22 p2
tij (−p1 , p2 ; ) = (−1)ℓi tij (p1 , p2 ; 2 ) . (8.16)
m m
25
We assume that the LS equation for a scattering process has a unique solution. Otherwise, we can always choose the solution that satisfies
the property Eq. (8.11).

80
p22
Had we changed the sign of p2 the resulting T -matrix element, tij (p1 , −p2 ; m) satisfies the same LS equation as
p22
(−1)ℓj t ij (p1 , p2 ; m ), and since the solution is supposed to be unique, the T -matrix element is also the same. Explic-
itly,
Z
p22 m X ∞ dp3 p23 p22
tij (p1 , −p2 ; ) = vij (p1 , −p2 ) + 2 viℓ (p 1 , p 3 )t ℓj (p 3 , −p 2 ; ),
m 2π ℓ 0 p23 − p22 m
Z
m X ∞ dp3 p23 p2
= (−1)ℓj vij (p1 , p2 ) + 2 2 2 viℓ (p1 , p3 )tℓj (p3 , −p2 ; 2 ) , (8.17)
2π ℓ 0 p3 − p2 m
p22
which is the same IE as the one satisfied by (−1)ℓj tij (p1 , p2 ; m ),
Z
p22 m X ∞ dp3 p23 p22
(−1)ℓj tij (p1 , p2 ; ) = (−1)ℓj vij (p1 , p2 ) + 2 v (p ,
iℓ 1 3 p )(−1)ℓj
t (p ,
ℓj 3 2 p ; ). (8.18)
m 2π ℓ 0 p23 − p22 m
Now we consider the generalization of Eq. (8.16) to complex p1 and p2 . The LS equation is in this case
Z
p22 m X dp3 p23 p2 2
tij (p1 , p2 ; ) = vij (p1 , p2 ) + 2 viℓ (p 1 , p 3 )t ℓj (p 3 , p 2 ; ), (8.19)
m 2π ℓ C p23 − p22 m
where the integration contour in the IE is denoted as C. One important point to realize is that the same integration contour
p2 p2 p2
applies to all tij (−p1 , p2 ; m2 ), tij (p1 , −p2 ; m2 ) and tij (p1 , p2 ; m2 ), because of the symmetry around the origin of the branch
p2
points in p3 for v(p1 , p3 ) and t(p3 , p2 ; m2 ). E.g. if p1 = qr + iqi with positive qr and qi > mπ the corresponding vertical
addition at qr in the integration contour C extends from 0 up to i(qi − mπ ). However, both the real and imaginary parts
of −p1 are negative and then, according to the rules in Eq. (6.4), the vertical addition takes place at −Rep1 = qr and
extends from 0 up to i(−Im(−p1 ) − mπ ) = i(qi − mπ ), and the same integration contour results. Once this point is clear
then one can proceed in completely analogous way as done above for real p1 and p2 to conclude that
p22 p2
) = (−1)ℓi t(p1 , p2 ; 2 ) ,
t(−p1 , p2 ;
m m
p22 p2
t(p1 , −p2 ; ) = (−1)ℓj t(p1 , p2 ; 2 ) , (8.20)
m m
also hold for complex arguments. Let us stress that because of parity conservation
(−1)ℓi = (−1)ℓj . (8.21)
p2 p2 p2 ℓ
As a corollary of Eq. (8.20), we also conclude that tij (p1 , p2 ; m2 )/(p1 p2 )ℓi , tij (p1 , p2 ; m2 )/(p1 p2 )ℓj and tij (p1 , p2 ; m2 )/(pℓ1i p2j )
are functions of the moduli squared of the CM three-momenta because these combinations are invariant under the change
of sign of p1 or p2 .
p2
• Now, we discuss the consequences of time-reversal invariance for tij (p1 , p2 , m2 ), although it is not explicitly used in
the derivation of the on-shell discontinuity across the DCs worked out in Sec. 8.2.
At the level of the potential the time-reversal symmetry implies that
vij (p1 , p2 ) = vji (p2 , p1 ) . (8.22)
Then tij (p1 , p2 ; p22 /m) satisfies the LS equation, Eq. (8.19), which is equivalent to
Z
p22 m X dp3 p23 p22
tij (p1 , p2 ; ) = vji (p2 , p1 ) + 2 vℓi (p 3 , p 1 )t ℓj (p 3 , p 2 ; )
m 2π ℓ C p23 − p22 m
Z
m X dp3 p23 p22
= vji (p2 , p1 ) + t (p ,
ℓj 3 2 p ; )vℓi (p3 , p1 ) (8.23)
2π 2 ℓ C p23 − p22 m

81
which is the same IE as the one satisfied by tji (p2 , p1 ; p22 /m)
Z
p2 m dp3 p23 p22
tji (p2 , p1 ; 2 ) = vji (p2 , p1 ) + 2 t jℓ (p 2 , p 3 ; )vℓi (p3 , p1 ) . (8.24)
m 2π C p23 − p22 m
Then
p22 p2
tij (p1 , p2 ; ) = tji (p2 , p1 ; 2 ) . (8.25)
m m

8.2 Calculation of the on-shell discontinuity


Here, we generalize to coupled channels and to an arbitrary spectral representation of the potential the steps undergone
in Sec. 7.4 in order to calculate Imtij (k, k; k2 /m) along the LHC.
Notation: If the energy argument is not shown in t(k, k′ , z) it is then assumed that the half-off-shell scattering case
is considered, with z = k′ 2 /m. Namely, t(k, k′ ) = t(k, k′ , k′ 2 /m).
• We want to calculate ∆ij (A) = Imtij (ik + ε, ik + ε), with A = −k2 + iε and k ∈ R+ . For that we take the
discontinuity

tij (ik + ε, ik + ε) − tij (ik − ε, ik − ε) = 2i∆ij (A) . (8.26)

This equality follows by applying Eq. (8.20) and (8.11), in this order, to tij (ik − ε, ik − ε) because

tij (ik − ε, ik − ε) = (−1)ℓi +ℓj tij (−ik + ε, −ik + ε) = tij (ik + ε, ik + ε)∗ . (8.27)

+
ik− ε ik− ε− ik+ ε

−ik−ε
−ik+ε− −ik+ε
+

Figure 17: Location of the cuts of tij (z1 , ik + ε), with z1 taking the values −ik + ε∓ in the difference tij (−ik + ε− , ik + ε) −
tij (−ik + ε+ , ik + ε). Note that the figure is not drawn to scale because the gap between two vertical cuts at ±ε is 2mπ , while the
horizontal separation between vertical cuts is 2ε → 0+ .

Instead of taking the difference in Eq. (8.26),

tij (ik + ε, ik + ε) − tij (ik − ε, ik − ε) , (8.28)

we now consider

tij (−ik + ε− , ik + ε) − tij (−ik + ε+ , ik + ε) , (8.29)

82
with ε+ > ε > ε− . To show the connection between both discontinuities let us exchange the sign in the first argument in
Eq. (8.29). By taking into account Eqs. (8.20) and (8.21) it results that
 
tij (−ik + ε− , ik + ε) − tij (−ik + ε+ , ik + ε) = (−1)ℓi tij (ik − ε− , ik + ε) − tij (ik − ε+ , ik + ε) . (8.30)

The first arguments in the difference on the lhs of the previous equation are shown by the black points in Fig. 17, while
the red points represent the first arguments of the rhs discontinuity. Now, inspection of Fig. 17 reveals that for ε → 0+
the partial wave amplitudes tij (ik + ε, ik + ε) and tij (ik − ε− , ik + ε) can be continued analytically one to each other
in that limit.26 A Taylor expansion of tij (z1 , ik + ε) can be performed at the point z1 = ik + ε (strictly) within a circle
of radius 2ε which incorporates the point ik − ε− , where tij (ik − ε− , ik + ε) is evaluated. In this way, we prove that
tij (ik − ε− , ik + ε) − tij (ik + ε, ik + ε) = O(ε). We can also proceed similarly for tij (ik − ε, z1 ) and, as a function of z1 ,
take a Taylor expansion centered at ik − ε of radius ε + ε′ , with ε− < ε′ < ε, that includes −
p the point tij (ik − ε, ik + ε ).
The second argument is just at the left of the cut running along the values −ik + ε + i m2π + x2 . The latter amplitude
is also equal to tij (ik − ε+ , ik + ε) because the second argument is located in the same relative position with respect to
the cuts of tij (ik − ε+ , z1 ). As a result it follows that for ε → 0+

tij (ik + ε, ik + ε) − tij (ik − ε, ik − ε) = tij (ik − ε− , ik + ε) − tij (ik − ε+ , ik + ε)


 
= (−1)ℓi tij (−ik + ε− , ik + ε) − t(−ik + ε+ , ik + ε) , (8.31)

where in the last equality we have used Eq. (8.20). Notice that the first discontinuity is, as we know, 2i∆ij (A), while the
latter one is (−1)ℓi i1+ℓi −ℓj fij (−k)/kℓi +ℓj +2 ,27 where fij (ν) is defined as

ifij (ν) = t̂ij (iν + ε− , ik + ε) − t̂ij (iν + ε+ , ik + ε) ,


ℓj +1
t̂ij (k, k′ ; z) = kℓi +1 k′ tij (k, k′ ; z) . (8.32)

It is then enough to calculate fij (ν) for ν = −k to obtain ∆ij (A),

fij (−k)
∆ij (A) = (−1)ℓi iℓi −ℓj . (8.33)
2kℓi +ℓj +2
Our aim here is to offer a more compact and shorter derivation for the IE of fij (ν), ν ∈ [−k, k − mπ ] valid for both S
and higher partial waves, that also collects in just one place all the necessary steps to arrive to the final IE.
The half-off-shell LS equation for partial waves in coupled channels reads
Z
k′ 2 m X dp1 p21
tij (k, k′ ; ) = vij (k, k′ ) + 2 vin (k, p1 )tnj (p1 , k′ ) , (8.34)
m 2π n C p21 − k′ 2

where the (deformed) integration contour C depends on the external variables k and k′ , as discussed in Sec. 6.
• Let us discuss first the case ν < 0. We take directly the difference t(iν + ε− , ik + ε) − t(iν + ε+ , ik + ε) and
consider the added vertical contours shown in Fig. 18, which are the only ones that give contribution to the difference.
The contribution of the unperturbed original contours (with positive real p1 ∈ [0, ∞]) vanishes because then v(iν + ε± , p1 )
and t(p1 , ik + ε) in the rhs of Eq. (8.34) are continuous there,28 due to the location of the DCs as given in Eq. (7.3).
26
As remarked in the caption of Fig. 17, this figure is not drawn to scale so that the vertical gap, in the middle of which one has the point
ik + ε, is 2mπ wide while the horizontal distance between the vertical cuts is 2ε → 0+ .
27
Notice that for i 6= j then (−i)ℓi +1 iℓj +1 = (−i)ℓi iℓi ±2 = −1.
28
The points in which the positive real axis of p1 crosses with the DCs are part of the vertical additions shown in Fig. 18.

83
ε− ε+
ε ε

Figure 18: Vertical contours used in Eq. (8.35) corresponding to the case ν < 0. Left panel applies to tij (iν + ε− , ik + ε)
and right panel to tij (iν + ε+ , ik + ε).

Then it results
tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)
Z
im X ν+mπ dν1 ν12  
− θ(−ν − mπ ) 2 2 2 vin (iν + ε− , iν1 + ε− − δ) − vin (iν + ε− , iν1 + ε− + δ) tnj (iν1 + ε− , ik + ε)
2π n 0 k − ν1
Z
im X k−mπ dν1 ν12  
− θ(k − mπ ) 2 2 2 vin (iν + ε− , iν1 + ε) tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε)
2π n 0 k − ν1
Z
im X ν+mπ dν1 ν12  + + + + 
+ θ(−ν − mπ ) 2 2 vin (iν + ε , iν 1 + ε − δ) − vin (iν + ε , iν 1 + ε + δ) tnj (iν1 + ε+ , ik + ε)
2π n 0 k 2 − ν1
Z
im X k−mπ dν1 ν12 +  
+ θ(k − mπ ) 2 vin (iν + ε , iν 1 + ε) t nj (iν 1 + ε − δ, ik + ε) − t nj (iν 1 + ε + δ, ik + ε) , (8.35)
2π n 0 k2 − ν12

where we have taken into account that vin (iν + ε± , iν1 ) is continuous around the upwards vertical cut of extension k − mπ
and so it is tnj (iν1 , ik + ε) around the downwards vertical cuts of extension −ν − mπ (with |ν| > mπ ). Combining the
integrals up to ν + mπ and k − mπ separately one can rewrite the previous equation as
tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)
Z
im X ν+mπ dν1 ν12  − − − − 
+ θ(−ν − mπ ) 2 vin (iν + ε , iν 1 + ε + δ) − vin (iν + ε , iν 1 + ε − δ)
2π 2 n 0 k 2 − ν1
 
× tnj (iν1 + ε− , ik + ε) − tnj (iν1 + ε+ , ik + ε) (8.36)
im XZ k−mπ dν1 ν12  
− θ(k − mπ ) vin (iν + ε− , iν1 + ε) − vin (iν + ε+ , iν1 + ε)
2π 2 n 0 k2 − ν12
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) .
Exchanging the order of the integration limits in the first integral we can combine the two integrals in one as
tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)
Z
mi X k−mπ dν1 ν12  − + 
− θ(k − mπ ) 2 vin (iν + ε , iν 1 + ε) − vin (iν + ε , iν 1 + ε)
2π 2 n (ν+mπ )θ(|ν|−mπ ) k2 − ν1
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) . (8.37)
• For the case ν > 0 we have to consider the vertical contours shown in Fig. 19, which are present for ν > mπ and
extend from 0 up to ν − mπ . These are the only ones that give contribution to the difference of partial waves under study.

84
ε− ε ε ε+

Figure 19: Vertical contours used in Eq. (8.38) corresponding to the case ν > 0. Left panel applies to tij (iν + ε− , ik + ε)
and right panel to tij (iν + ε+ , ik + ε).

Then, following the same steps used to derive Eqs. (8.35) and (8.36), we have now

tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)


Z
im X ν−mπ dν1 ν12  
+ θ(ν − mπ ) 2 2 2 vin (iν + ε− , iν1 + ε− + δ) − vin (iν + ε− , iν1 + ε− − δ)
2π n 0 k − ν1
 
× tnj (iν1 + ε− , ik + ε) − tnj (iν1 + ε+ , ik + ε) (8.38)
im XZ k−mπ dν1 ν12  
− +
− θ(k − mπ ) 2 vin (iν + ε , iν 1 + ε) − vin (iν + ε , iν 1 + ε)
2π 2 n 0 k 2 − ν1
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) .

Let us notice that now both integrals run through positive values of ν1 so that they subtract each other, because of the
different sign in front of them, and we have an analogous expression to Eq. (8.37),

tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)


Z
mi X k−mπ dν1 ν12  
− θ(k − mπ ) 2 2 2 vin (iν + ε− , iν1 + ε) − vin (iν + ε+ , iν1 + ε)
2π n (ν−mπ )θ(ν−mπ ) k − ν1
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) . (8.39)

The sign of ν is denoted as ǫ(ν), namely,


ν
ǫ(ν) = . (8.40)
|ν|
To shorten notation we indicate in the subsequent the lower limit of integration in Eqs. (8.37) and (8.39) as

ϑ(ν) = ǫ(ν)(|ν| − mπ )θ(|ν| − mπ ) . (8.41)

We can then express both Eqs. (8.37) and (8.39) in the same manner as

tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = vij (iν + ε− , ik + ε) − vij (iν + ε+ , ik + ε)


Z
mi X k−mπ dν1 ν12  
− θ(k − mπ ) 2 2 2 vin (iν + ε− , iν1 + ε) − vin (iν + ε+ , iν1 + ε)
2π n ϑ(ν) k − ν1
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) . (8.42)

85
• The previous equation allows us to conclude the continuity of Retij (iν + ε ± δ, ik + ε), ν ∈ [−k, k − mπ ], for δ → 0.
This continuity is clear for Revij (iν + ε± , ik + ε) and Revij (iν + ε± , iν1 + ε) with ε± → ε, see e.g. Eq. (3.86). As a
result, the discontinuities in the potential in Eq. (8.42) are purely imaginary, which allows to rewrite this equation as
 
tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = i Imvij (iν + ε− , ik + ε) − Imvij (iν + ε+ , ik + ε)
Z
m X k−mπ dν1 ν12  − + 
+ θ(k − mπ ) 2 Imvin (iν + ε , iν 1 + ε) − Imvin (iν + ε , iν 1 + ε)
2π 2 n ϑ(ν) k2 − ν1
 
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) . (8.43)

Because the independent term in this IE is purely imaginary while the kernel is purely real it is clear that the solution of
this IE, that corresponds to the discontinuity of tij (iν + ε ± δ, ik + ε) with δ → 0, is purely imaginary.29
As a consequence of the continuity of Retij (iν + ε ± δ, ik + ε) for δ → 0 it follows that we can rewrite the definition
of fij (ν), Eq. (8.32), simply as

fij (ν) = Imt̂ij (iν + ε − δ, ik + ε) − Imt̂ij (iν + ε + δ, ik + ε) , (8.44)

Taking now into account this property of continuity of Retij (iν1 + ε ± δ, ik + ε) for δ → 0, the IE in Eq. (8.43) becomes

Imtij (iν + ε− , ik + ε) − Imtij (iν + ε+ , ik + ε) = Imvij (iν + ε− , ik + ε) − Imvij (iν + ε+ , ik + ε)


Z
m X k−mπ dν1 ν12  
+ θ(k − ν − 2mπ ) 2 2 2 Imvin (iν + ε− , iν1 + ε) − Imvin (iν + ε+ , iν1 + ε)
2π n ϑ(ν) k − ν1
 
× Imtnj (iν1 + ε − δ, ik + ε) − Imtnj (iν1 + ε + δ, ik + ε) . (8.45)

• From Eq. (8.45) let us derive a new equation free of the integration problems that arise because of the infrared
divergences for ℓ ≥ 1, which stem from the fact that Imvin (iν + ε± , iν1 + ε) and Imtnj (iν1 + ε ± δ, ik + ε) diverge like
2(ℓ +1)
1/ν1ℓn +1 for ν1 → 0, so that its product does like 1/ν1 n . This behavior of the discontinuity at the origin is clear if
we think of the integral in Eq. (3.85) together with a factor tℓ inside the integrand that will come from the term with the
highest degree in the Legendre polynomial Pℓ (t). That this divergent factor also appears in Imtnj (iν1 + ε + δ, ik + ε) is
clear from the Neumann series of the LS, cf. Eq. (6.5).
To derive this new IE we first rewrite Eq. (8.45) as

Imtij (iν + ε− , ik + ε) − Imtij (iν + ε+ , ik + ε) = Imvij (iν + ε− , ik + ε) − Imvij (iν + ε+ , ik + ε)


Z n
θ(k − mπ )m X k−mπ dν1 ν12 
− 2 Re vin (iν + ε− , iν1 + ε) − vin (iν + ε+ , iν1 + ε)
2π 2 n ϑ(ν)
2
k − ν1
 o
× tnj (iν1 + ε − δ, ik + ε) − tnj (iν1 + ε + δ, ik + ε) , (8.46)

where we have used again the continuity of Revin (iν + ε± , iν1 + ε) and Retnj (iν1 + ε ± δ, ik + ε) for δ → 0, Eq. (8.43),
that allows us to neglect the contribution
Z
θ(k − mπ )m X k−mπ dν1 ν12  
− 2 Revin (iν + ε− , iν1 + ε) − Revin (iν + ε+ , iν1 + ε)
2π 2 n ϑ(ν)
2
k − ν1
 
× Retnj (iν1 + ε − δ, ik + ε) − Retnj (iν1 + ε + δ, ik + ε) −→ 0 . (8.47)
δ→0
29
It is important to note that the use of this property of continuity for Rev̂ij and Ret̂ij is applied for finite ε. That is, the integrations are
done with ε 6= 0 but with ε± → ε and δ → 0.

86
We remark here again, as in the footnote 29, that the continuity referred is in ε± → ε and δ → 0. This is the situation
in Eq. (8.47) because one first takes these limits while keeping ε 6= 0. In this way the integrals remain finite and there is
then no ambiguous limit of the type 0 × ∞, that could appear because of the infrared divergences if simultaneously with
ε± → ε and δ → 0 one also took the limit ε → 0 in the integrations.
Next, we employ the functions v̂ij (k, k′ ) and t̂ij (k, k′ ; z), defined as,
ℓj +1
v̂ij (k, k′ ) = kℓi +1 k′ vij (k, k′ ) ,
ℓj +1
t̂ij (k, k′ ; z) = kℓi +1 k′ tij (k, k′ ; z) . (8.48)

In this way the divergent behavior of Imvij (iν + ε± , iν1 + ε) as 1/ν ℓi +1 (1/ν1 ℓj +1 ) for ν(ν1 ) → 0 is not longer present
in v̂ij (iν + ε± , iν1 + ε). This is clear if one thinks in terms of a spectral decomposition for each of the components of
vij (iν + ε± , iν1 + ε), and for each of them one keeps only the contribution that gives rise to the log-term after the angular
projection, cf. Eq. (3.86). The same can be said for tij (iν1 + ε + ±δ, ik + ε) and t̂ij (iν + ε± , ik + ε), as it is clear by
iterating the LS equation.
Then, we extract 1/[(iν)ℓi +1 (ik)ℓj +1 ] as a common factor in Eq. (8.46) because iℓi +ℓj +2 = (−1)ℓi +ℓj = +1. To
simplify the notation we also utilize the function ∆v̂ij (ν, ν1 ), defined as,

∆v̂ij (ν, ν1 ) = Imv̂ij (iν + ε− , iν1 + ε) − Imv̂ij (iν + ε+ , iν1 + ε) , (8.49)

and fij (ν1 ), Eq. (8.44). Expanding the real part in the integrand of Eq. (8.46) we can rewrite it as

fij (ν) = ∆v̂ij (ν, k)


Z n
θ(k − mπ )m X k−mπ dν1 ν12 1 
− 2 Re Re v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε)
2π 2 n ϑ(ν)
2
k − ν1 (iν1 + ε)n

 o
× t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε)
Z n
θ(k − mπ )m X k−mπ dν1 ν12 1 
+ 2 Im Im v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε)
2π 2 n ϑ(ν)
2
k − ν1 (iν1 + ε)n

 o
× t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε) , (8.50)

with

n = 2ℓn + 2 . (8.51)

The last integral in Eq. (8.50) does not contribute because


n  o
Im v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε) t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε)
   
= Re v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε) Im t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε)
   
+ Im v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε) Re t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε) (8.52)

is zero in the limit ε± → ε and δ → 0 because of the continuity of both Rev̂(iν + ε± , iν1 + ε) and Ret̂(iν1 + ε ± δ, ik + ε),
in order. The same remark as already done twice above, in footnote 29 and just after Eq. (8.47), is in order here.30
30
Note also that
 
1 1 1 1
Im = − (8.53)
(iν1 + ε)n 2i (iν1 + ε)n (−iν1 + ε)n
is null for finite ν1 .

87
Then, Eq. (8.50) can be finally written as
Z  
θ(k − mπ )m X k−mπ dν1 ν12 1 1
fij (ν) = ∆v̂ij (ν, k) + + ∆v̂in (ν, ν1 )fnj (ν1 ) , (8.54)
4π 2 n ϑ(ν) k2 − ν12 (iν1 + ε)n (iν1 − ε)n

where we have used again the continuity of Rev̂in (iν + ε, iν1 + ε+ ± δ) and Ret̂nj (iν1 + ε ± δ, ik + ε) for δ → 0 to express
n  o
Re v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε) t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε)
   
→ −Im v̂in (iν + ε− , iν1 + ε) − v̂in (iν + ε+ , iν1 + ε) Im t̂nj (iν1 + ε − δ, ik + ε) − t̂nj (iν1 + ε + δ, ik + ε) . (8.55)

• Eq. (8.54) can be furthered simplified by taking into account that ∆v̂ij (ν, ν1 ) stems from the imaginary part in the
difference of log-terms in Eq. (3.86), namely,
Z ∞
1
i∆v̂ij (ν, ν1 ) = dµ2 ηij (µ2 )ρij (ν 2 , ν12 ; µ2 ) (8.56)
π m2π
    
× log − (ν + ν1 )2 + µ2 + i(ν + ν1 )0+ − log − (ν − ν1 )2 + µ2 − i(ν − ν1 )0+
n    o
− log − (ν + ν1 )2 + µ2 + i(ν + ν1 )0+ − log − (ν − ν1 )2 + µ2 + i(ν − ν1 )0+
Z 
1 ∞    
= dµ2 ηij (µ2 )ρij (ν 2 , ν12 ; µ2 ) − log − (ν − ν1 )2 + µ2 − i(ν − ν1 )0+ + log − (ν − ν1 )2 + µ2 + i(ν − ν1 )0+ ,
π m2π

where the first two log terms correspond to Imv̂ij (iν + ε− , iν1 + ε) and the last two (shown between curly brackets) to
Imv̂ij (iν + ε+ , ik + ε). The integrand is proportional to the function ρij (ν 2 , ν12 ; µ2 ), which is a polynomial such that

χρij (ν 2 , ν12 ; m2π ) (8.57)

corresponds to Imv̂ij (iν1 , iν2 ) according to Tables 3 and 4.


It follows from Eq. (8.56) that in order to end with a non vanishing imaginary part it is required that (ν1 − ν)2 > µ2
or, in other terms, that
q
ν1 = ν ± µ 2 + x2 , x ∈ R . (8.58)

Due to ϑ(ν) in the integral of Eq. (8.54) only the plus sign matters as ν1 > ν − mπ > ν − µ. Therefore, Eq. (8.56) for
the range of values of ν1 relevant in Eq. (8.54) simplifies as
Z ∞
1
∆v̂ij (ν, ν1 ) = 2µdµηij (µ2 )(−2)ρij (ν 2 , ν12 ; µ2 )θ(ν1 − ν − µ) (8.59)
π mπ
Z ν1 −ν
2
=− 2µdµηij (µ2 )ρij (ν 2 , ν12 ; µ2 ) .
π mπ

The fact that the integrand in Eq. (8.54) is nonzero only for ν1 > ν + mπ also implies that one can replace ϑ(ν) by
ν + mπ since

ν + mπ ≥ ǫ(ν)(|ν| − mπ )θ(|ν| − mπ ) . (8.60)

Of course, it is then required that k − mπ > ν + mπ , because otherwise ∆v̂ij (ν, ν1 ) is zero and ν1 extends up to k − mπ ,
a condition that can be accounted for by the presence of θ(k − 2mπ − ν) in front of the integral in Eq. (8.54). As a result
we can rewrite this equation as
Z  
θ(k − 2mπ − ν)m X k−mπ dν1 ν12 1 1
fij (ν) = ∆v̂ij (ν, k) + + ∆v̂in (ν, ν1 )fnj (ν1 ) (8.61)
4π 2 n mπ +ν k2 − ν12 (iν1 + ε)n (iν1 − ε)n

88
It is convenient to express this equation explicitly in terms of the spectral representation of the partial-wave projected
potential as follows from Eq. (8.59). It results
Z
2 k−ν
fij (ν) = − 2µdµηij (µ2 )ρij (ν 2 , k2 ; µ2 ) (8.62)
π mπ
Z Z ν1 −ν
θ(k − 2mπ − ν)m X k−mπ dν1 ν12
− S (ν )f
2 − ν 2 n 1 nj 1
(ν ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) ,
2π 3 n ν+mπ k 1 mπ

where
1 1
Sn (ν1 ) = 2ℓ +2
+ , (8.63)
(iν1 + ε) n (iν1 − ε)2ℓn +2
a function already introduced in Eq. (7.101).
• Now, let us discuss how to handle the previous IE in an appropriate way ready to be used to cure numerically the
problem of the infrared divergences for ℓ ≥ 1.
The problem with the infrared divergences originates when the integration in Eq. (8.62) approaches or crosses the
value ν = −mπ , which starts happening for ν ≤ −mπ + ∆, with ∆ > 0 but small so that numerically one would
get problems in the evaluation of the integral to solve the IE. The key point is to make use of the symmetry under
the exchange ν1 ↔ −ν1 in the function ρin (ν 2 , ν12 ; µ2 ). This symmetry property is a consequence of the property
t̂(−p1 , p2 ; z) = t̂(p1 , −p2 ; z) = −t̂(p1 , p2 ; z) together with the fact that the difference of log’s in the first two lines of
Eq. (8.56) is odd under the same operation.
• We symmetrize partially the integral in Eq. (8.62) for ν + mπ < 0, which can then be expressed also as
Z Z ν1 −ν n o
−ν−mπ dν1 ν12 2 2 2 2
θ(k − 2mπ − ν) 2 S n (ν 1 ) 2µdµηin (µ )ρin (ν , ν 1 ; µ )f nj (ν 1 ) = ν 1 → −ν 1 (8.64)
ν+mπ k 2 − ν1 mπ
Z −ν−mπ Z −ν1 −ν
dν1 ν12
=θ(k − 2mπ − ν) Sn (ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 )fnj (−ν1 ) .
ν+mπ k2 − ν12 mπ

Notice also that here k − mπ > −ν − mπ because k > −ν > mπ . Taking this result into account we rewrite Eq. (8.62) as
Z Z
2 k−ν
2 2 θ(k − 2mπ − ν)m X
2 2
−ν−mπ dν1 ν12
fij (ν) = − 2µdµηij (µ )ρij (ν , k ; µ ) − Sn (ν1 ) (8.65)
π mπ 4π 3 n ν+mπ k2 − ν12
h Z ν1 −ν Z −ν1 −ν i
× fnj (ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) + fnj (−ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 )
mπ mπ
Z Z
θ(k − 2mπ − ν)m X k−mπ dν1 ν12 ν1 −ν
− Sn (ν1 )fnj (ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) ,
2π 3 n −ν−mπ k2 − ν12 mπ

The terms between square brackets are even in ν1 so that the integration over this variable is well defined even if ν1 crosses
zero. The first integral in ν1 can be restricted to positive values of this variable because the integrand is even and then
we have
Z Z
2 k−ν
2 2 θ(k − 2mπ − ν)m X
2 2
−ν−mπ dν1 ν12
fij (ν) = − 2µdµηij (µ )ρij (ν , k ; µ ) − Sn (ν1 ) (8.66)
π mπ 2π 3 n 0 k2 − ν12
h Z ν1 −ν Z −ν1 −ν i
2 2
× fnj (ν1 ) 2µdµηin (µ )ρin (ν , ν12 ; µ2 ) + fnj (−ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 )
mπ mπ
Z Z
θ(k − 2mπ − ν)m X k−mπ dν1 ν12 ν1 −ν
− Sn (ν1 )fnj (ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) ,
2π 3 n −ν−mπ k2 − ν12 mπ

89
• However, Eq. (8.66) is still troublesome for its numerical implementation because for P and higher partial waves the
variable ν1 in the last line can be arbitrarily close to zero when ν → −mπ . We then proceed to express the first iterated
form of Eq. (8.62), similarly as we did in Sec. 7.4.

µ ν1−ν ν1=µ+ν
k−mπ −ν


ν1
ν+ m π k−mπ

Figure 20: Integration region for the double integral of ν1 , µ in Eq. (8.62). For µ ∈ [mπ , k−mπ −ν] then ν1 ∈ [µ+ν, k−mπ ].

In order to end with integrals in which we can apply the algebraic trick associated with having even powers of ν1
multiplying Sn (ν1 ) when ν1 crosses zero, we first exchange the order of the integrals in the last term of Eq. (8.62). This
is straightforward, particularly taking into account the integration region explicitly shown in Fig. 20. Then we have the
following equivalent form of Eq. (8.62)
Z
2 k−ν
fij (ν) = − 2µdµηij (µ2 )ρij (ν 2 , k2 ; µ2 ) (8.67)
π mπ
Z Z k−mπ
θ(k − 2mπ − ν)m X k−mπ −ν 2 dν1 ν12
− 2µdµη in (µ ) Sn (ν1 )ρin (ν 2 , ν12 ; µ2 )fnj (ν1 ) .
2π 3 n mπ µ+ν k2 − ν12
The contribution of only the first iteration of the previous equation reads
Z Z Z
θ(k − 2mπ − ν)m X k−mπ −ν
2
k−mπ dν1 ν12 k−ν1
2 2
2µdµηin (µ ) Sn (ν1 )ρin (ν 2 , ν12 ; µ2 ) 2µ′ dµ′ ηnj (µ′ )ρnj (ν12 , k2 ; µ′ ) .
π4 n mπ µ+ν k2 − ν12 mπ
(8.68)

We have to change again the order of the last two integrals in the variables ν1 and µ′ . This is clear from the integration
region given in Fig. 21, and we have the equivalent expression for Eq. (8.68)
Z Z Z k−µ′
θ(k − 2mπ − ν)m X k−mπ −ν
2
k−µ−ν
′ ′ ′2 dν1 ν12 2
2µdµηin (µ ) 2µ dµ ηnj (µ ) 2 Sn (ν1 )ρin (ν 2 , ν12 ; µ2 )ρnj (ν12 , k2 ; µ′ ) .
π4 n mπ mπ µ+ν
2
k − ν1
(8.69)
This expression is adequate because the last integration in ν1 is well defined since it only involves even powers of ν1 when
ν1 crosses zero.
Now, we consider the term that comprises all the higher order iterations that result from Eq. (8.67), it reads
Z Z k−mπ
θ(k − 2mπ − ν)m2 X k−mπ −ν 2 dν1 ν12
2µdµηin (µ ) S (ν )ρ (ν 2 , ν12 ; µ2 )
2 − ν 2 n1 1 in1
(8.70)
(2π 3 )2 n1 ,n2 mπ
1
µ+ν k 1
Z Z
k−mπ −ν1
2
k−mπ dν2 ν22 2
×θ(k − 2mπ − ν1 ) 2µ′ dµ′ ηn1 n2 (µ′ ) Sn (ν2 )ρn1 n2 (ν12 , ν22 ; µ′ )fn2 j (ν2 ) .
mπ µ′ +ν1 k2 − ν22 2

90
µ’
k −µ−ν

−ν1+k ν1=−µ ’+k


ν1
µ+ν k−mπ

Figure 21: Integration region for the double integral of ν1 , µ′ in Eq. (8.68). For µ′ ∈ [mπ , k−µ−ν] then ν1 ∈ [µ+ν, k−µ′ ].

An important point is to take into account the step function θ(k − 2mπ − ν1 ) present in the second line of Eq. (8.70). This
implies that ν1 < k − 2mπ and then this also limits the range of the integration in the variable µ, so that µ < k − 2mπ − ν
to guarantee that µ + ν < k − 2mπ . Then, one also must require that k − 2mπ − ν > mπ . As a result, Eq. (8.70) becomes
Z Z k−2mπ
θ(k − 3mπ − ν)m2 X k−2mπ −ν 2 dν1 ν12
2µdµηin (µ ) S (ν )ρ (ν 2 , ν12 ; µ2 )
2 − ν 2 n1 1 in1
(8.71)
(2π 3 )2 n1 ,n2 mπ
1
µ+ν k 1
Z Z
k−mπ −ν1
2
k−mπ dν2 ν22 2
× 2µ′ dµ′ ηn1 n2 (µ′ ) Sn (ν2 )ρn1 n2 (ν12 , ν22 ; µ′ )fn2 j (ν2 ) .
mπ µ′ +ν1 k2 − ν22 2

µ’
k−mπ −µ−ν

−ν1+k−mπ ν1=−µ ’+k−mπ


ν1
µ+ν k−2mπ

Figure 22: Integration region for the double integral of ν1 , µ′ in Eq. (8.71). For µ′ ∈ [mπ , k − mπ − µ − ν] then
ν1 ∈ [µ + ν, k − mπ − µ′ ].

Let us first exchange the order of the integration in the variables ν1 and µ′ in Eq. (8.71), which is analogous to the
one already done to obtain Eq. (8.69) with k − mπ instead of k in the limits of integration. Nonetheless, attending to the
region of integration shown in Fig. 22 the result is also clear. One obtains
Z Z k−mπ −µ−ν
θ(k − 3mπ − ν)m2 X k−2mπ −ν 2 2
3 2
2µdµηin1 (µ ) 2µ′ dµ′ ηn1 n2 (µ′ ) (8.72)
(2π ) n1 ,n2 mπ mπ
Z k−mπ −µ′ Z
dν1 ν12 k−mπ dν2 ν22 2
× Sn (ν1 )ρin1 (ν 2 , ν12 ; µ2 ) Sn (ν2 )ρn1 n2 (ν12 , ν22 ; µ′ )fn2 j (ν2 ) .
µ+ν k2 − ν12 1 µ′ +ν1 k2 − ν22 2

Next, we change the order of integration in the variables ν1 and ν2 , which is also straightforward by considering the

91
ν2 ν1+µ’ ν1=ν2−µ ’
k−mπ

ν+µ+µ’
ν1
µ+ν
+ k−mπ −µ’

Figure 23: Integration region for the double integral of ν1 , ν2 in Eq. (8.72). For ν2 ∈ [µ + µ′ + ν, k − mπ ] then
ν1 ∈ [µ + ν, ν2 − µ′ ].

integration region shown in Fig. 23. In terms of this change Eq. (8.72) becomes
Z Z k−mπ −µ−ν Z k−mπ
θ(k − 3mπ − ν)m2 X k−2mπ −ν 2 ′ ′ ′2 dν2 ν22
2µdµηin (µ ) 2µ dµ ηn n (µ ) 2 Sn2 (ν2 )fn2 j (ν2 )
(2π 3 )2 n1 ,n2 mπ
1

1 2 2
µ+µ′ +ν k − ν2
(8.73)
Z ν2 −µ′ dν1 ν12 2
× Sn (ν1 )ρin1 (ν 2 , ν12 ; µ2 )ρn1 n2 (ν12 , ν22 ; µ′ ) .
µ+ν k2 − ν12 1
The last integral is well-defined because it only involves even powers of ν1 when ν1 crosses zero. Summing the intermediate
expressions obtained in Eqs. (8.69) and (8.73), we have for the first-iterated form of Eq. (8.67)
Z Z
2 k−ν
2 2 θ(k − 2mπ − ν)m X
2 2
k−mπ −ν
fij (ν) = − 2µdµηij (µ )ρij (ν , k ; µ ) + 2µdµηin (µ2 ) (8.74)
π mπ π4 n mπ
Z Z k−µ′
k−µ−ν
′ ′ ′2 dν1 ν12 2
× 2µ dµ ηnj (µ ) 2 2 Sn (ν1 )ρin (ν 2 , ν12 ; µ2 )ρnj (ν12 , k2 ; µ′ )
mπ µ+ν k − ν1
θ(k − 3mπ − ν)m2 X Z k−2mπ −ν Z k−mπ −µ−ν
2
2
+ 2µdµηin1 (µ ) 2µ′ dµ′ ηn1 n2 (µ′ )
(2π 3 )2 n1 ,n2 mπ mπ
Zk−mπ dν ν 2 Z ν2 −µ′
2 2 dν1 ν12 2
× S (ν )f (ν
2 n2 2 n2 j 2 ) Sn (ν1 )ρin1 (ν 2 , ν12 ; µ2 )ρn1 n2 (ν12 , ν22 ; µ′ ) .
2
µ+µ′ +ν k − ν2 µ+ν k2 − ν12 1

Then, we have at our disposal several expressions that are used in different intervals of ν so as to avoid the infrared
singularities for ℓ > 0


 Eq. (8.62) or (8.67) −mπ /2 < ν
fij (ν) = Eq. (8.74) −3mπ /2 < ν < −mπ /2

 Eq. (8.66) −3mπ /2 > ν
(8.75)
• In order to apply numerically Eq. (8.66) for −3mπ /2 > ν one could add and subtract the first ℓn terms in the Taylor
series in powers of ν12 of the even function between brackets in this equation, namely,
Z ν1 −ν Z −ν1 −ν
h(ν12 ) = fnj (ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) + fnj (−ν1 ) 2µdµηin (µ2 )ρin (ν 2 , ν12 ; µ2 ) . (8.76)
mπ mπ

92
Let us denote by Qℓn (ν12 ) the 2(ℓn − 1)-degree polynomial in the variable ν12 resulting from the first ℓn terms in the Taylor
expansion of the h(ν12 ) around 0, namely,
ℓX
n −1
∂ i h(0) 2i
Qℓn (ν12 ) =h(0) + ν , (8.77)
i=1
∂ν12 1

with the understanding that for ℓn = 0 then Qℓn (ν12 ) = 0. Due to the complicated expression of h(ν12 ) it seems that the
most appropriate strategy is to evaluate numerically the different order derivatives of h(ν12 ) at ν1 = 0 needed in Eq. (8.77)
by making use of the already known values of f(ν1 ) for ν1 ≥ mπ . In this way, the integral
Z −ν−mπ dν1 ν12
Sn (ν1 )Qℓn (ν12 ) (8.78)
0 k2 − ν12
can be done algebraically because the polynomial Qℓn (ν12 ) only involves even powers of ν1 . Furthermore, because for
ν1 → 0 the difference h(ν12 ) − Qℓn (ν12 ) times the extra factor of ν12 from the measure vanishes as ν12l+2 one has the
cancellation of the divergence in Sℓn (ν1 ) → ν1−2ℓn −2 . Therefore, the integral
ν1 →0
Z −ν−mπ dν1 ν12  
2 2 Sn (ν1 ) h(ν12 ) − Qℓn (ν12 ) (8.79)
0 k − ν1
can be done numerically.
• The fact that ∆ij (A) = ∆ji (A), which is equivalent to the statement that fij (−k) = fji (−k), cf. Eq. (8.33), does
not imply that fij (ν) = fji (ν). This is clear because time reversal symmetry requires
tij (iν + ε− , ik + ε) − tij (iν + ε+ , ik + ε) = tji (ik + ε, iν + ε− ) − tji (ik + ε, iν + ε+ )
6= tji (iν + ε− , ik + ε) − tji (iν + ε+ , ik + ε) , ν ∈ [−k, k − mπ ] . (8.80)
Then we have to solve separately fij (ν) and fji (ν). For the 2×2 coupled-channel case the linear system of IEs of Eq. (8.61)
explicitly decouples in two systems of two IEs each (because every IE only involves the first subscript i while j is kept
fixed),
Z
m k−mπ dν1 ν12
f11 (ν) = ∆v̂11 (ν, k) + θ(k − 2mπ − ν) [∆v̂11 (ν, ν1 )S1 (ν1 )f11 (ν1 ) + ∆v̂12 S2 (ν1 )f21 (ν1 )] ,
4π 2 mπ +ν k2 − ν12
Z k−mπ
m dν1 ν12
f21 (ν) = ∆v̂21 (ν, k) + θ(k − 2mπ − ν) 2 [∆v̂21 (ν, ν1 )S1 (ν1 )f11 (ν1 ) + ∆v̂22 S2 (ν1 )f21 (ν1 )] , (8.81)
4π mπ +ν k2 − ν12
and
Z
m k−mπ dν1 ν12
f22 (ν) = ∆v̂22 (ν, k) + θ(k − 2mπ − ν) [∆v̂21 (ν, ν1 )S1 (ν1 )f12 (ν1 ) + ∆v̂22 S2 (ν1 )f22 (ν1 )] ,
2
4π mπ +ν k2 − ν12
Z k−mπ
m dν1 ν12
f12 (ν) = ∆v̂12 (ν, k) + θ(k − 2mπ − ν) 2 [∆v̂11 (ν, ν1 )S1 (ν1 )f12 (ν1 ) + ∆v̂12 S2 (ν1 )f22 (ν1 )] . (8.82)
4π mπ +ν k2 − ν12
Finally, let us see that fij (−k) = fji (−k),
ifij (−k) = t̂ij (−ik + ε− , ik + ε) − t̂ij (−ik + ε+ , ik + ε) = t̂ji (ik + ε, −ik + ε− ) − t̂ji (ik + ε, −ik + ε+ )
= t̂ji (−ik + ε, ik + ε− )∗ − t̂ji (−ik + ε, ik + ε+ )∗ (8.83)
The discontinuity only affects the imaginary part of tij so that we can remove the complex conjugation in the last equation
and instead write a global minus sign,
 
ifij (−k) = − t̂ji (−ik + ε, ik + ε− ) − t̂ji (−ik + ε, ik + ε+ ) = t̂ji (−ik + ε, ik + ε+ ) − t̂ji (−ik + ε, ik + ε− ) = ifji (−k) .
(8.84)

93
9 Nonlinear DR for the on-shell partial wave

XXX For singular potentials I would not expect that these DRs converge because ∆(A) diverges faster than a
power of A for A → −∞
Once we know the discontinuity ∆(A) along the LHC of an on-shell partial wave amplitude,

k2
T (A) = t(k, k; ) , k2 = A , (9.1)
m
we can write down a nonlinear DR for T (A) taking advantage of elastic unitarity which allows us to calculate the discon-
tinuity of T (A) along the RHC31

ImT (A) = ρ(A)|T (A)|2 , A > 0 , (9.2)

with ρ(A) the nonrelativistic phase space factor



m A
ρ(A) = . (9.3)

For completeness, we also assume that there is a bound state at A = B (if needed more bound states could be added
straightforwardly). Then, we consider the closed contour C in the A-complex plane, which is a circle that engulfs the
LHC and RHC, Fig. 24, and apply Cauchy integration formula. We distinguish between two cases: i) L < B < 0 and ii)
−∞ < B < L where

L = −m2π /4 (9.4)

corresponds to the onset of LHC.


Case i) L < B < 0
I Z Z
T (z) L
2 ∆(k2 ) ∞ ρ(k2 )|T (k2 )|2
dz = 2i dk + 2i dk2 (9.5)
C (z − C)m (z − A) −∞ (k2 − A)(k2 − C)m 0 (k2 − A)(k2 − C)m
 
 T (A) β m
X cp 
= 2πi + + , (9.6)
 (A − C)m (B − C)m (B − A) p=1 (A − C)p 

where we have used in the integral along the RHC the relation in Eq. (9.2). On the other hand, if there were no bound
states we simply set β = 0 in Eq. (9.6). Now by equating the right-hand sides of Eqs. (9.5) and (9.6) we end with the
following m-time subtracted DR for T (A),
m−1 Z Z
β X (A − C)m L ∆(k2 ) (A − C)m ∞ ρ(k2 )|T (k2 )|2
T (A) = + λi (A − C)i + dk2 2 2 m
+ dk2 .
A−B i=0
π −∞ (k − A)(k − C) π 0 (k2− A)(k2 − C)m
(9.7)

Here, we have reabsorbed in the subtractive polynomial the contribution

(A − C)m 1
−1 . (9.8)
(B − C)m (B − A) B−A
31
We assume elastic unitarity all along the RHC.

94
L

Figure 24: Integration contour C for the on-shell DR of T (A), Eq. (9.6).

Case ii) −∞ < B < L. In this case the bound state of T (A) lies on the LHC so that we consider the function (A−B)T (A),
which has no bound state, and perform the same analysis as in case i). Instead of the integral in the left-hand side (lhs)
of Eq. (9.5) we now have
I m
(z − B)T (z) A−B X λ′p
dz = 2πi T (A) + . (9.9)
C (z − C)m (z − A) (A − C)m p=1
(A − C)p

Notice that for the same asymptotic behavior of T (z) for z → ∞ one should take one more subtraction in the previous
equation as compared to Eq. (9.5) because of the extra factor (z − B) in the numerator. Writing the lhs of Eq. (9.9) in
terms of integrations along the LHC and RHC, as done in rhs of Eq. (9.5), we have
Z Z
(A − C)m L ∆(k2 )(k2 − B) (A − C)m ∞ ρ(q 2 )|T (q 2 )|2 (q 2 − B)
T (A) = dk2 + dq 2
(A − B)π −∞ (k2 − A)(k2 − C)m (A − B)π 0 (q 2 − A)(q 2 − C)m
m
X (A − C)m−p
+ λ′p . (9.10)
p=1
(A − B)

Now by adding and subtracting A we can write


k2 − B A−B
=1+ ,
k2 − A k2 − A
q2 − B A−B
=1+ . (9.11)
q2 − A q2 − A
We also reexpress the last term in Eq. (9.10) as
m
X (A − C)m−p m−1
X (A − C)p m−1
X (A − C)p
λ′p = λ′m−p = λ′′p , (9.12)
p=1
(A − B) p=0
(A − B) p=0
(A − B)

with the understanding that the last sum is not present if m − 1 < 0. Then, Eq. (9.10) can be written as
(Z Z )
(A − C)m L
2∆(k2 ) ∞
2 ρ(q 2 )|T (q 2 )|2
T (A) = dk + dq 2
π −∞ (k2 − A)(k2 − C)m 0 (q − A)(q 2 − C)m
m−1
(Z Z )
X (A− C)p (A − C)m L ∆(k2 ) ∞
2 ρ(q
2 )|T (q 2 )|2
+ λ′′p + dk2
+ dq . (9.13)
p=0
(A − B) (A − B)π −∞ (k2 − C)m 0 (q 2 − C)m

95
Notice that the last two integrals are just numbers independent of A. By repeatedly adding and subtracting B we can
express
m−1 Pm−1
X − C)p
(A p=0 λ′′p (B − C)p m−2
X
λ′′p = + λ̄′′p (A − C)p . (9.14)
p=0
A−B A−B p=0

A result that can also be applied to the coefficient (A − C)m /(A − B) multiplying the last two integrals in Eq. (9.13).
In particular we can extract explicitly the coefficient multiplying (A − C)m−1 , which is the monomial of higher degree in
Eq. (9.13) and that is not present in the sum of Eq. (9.14). Then, the latter equation can also be written as
(Z Z )
(A − C)m L
2 ∆(k2 ) ∞ ρ(q 2 )|T (q 2 )|2
2
T (A) = dk + dq 2
π −∞ (k2 − A)(k2 − C)m 0 (q − A)(q 2 − C)m
m−2
(Z Z )
λ X
p (A − C)m−1 L ∆(k2 )
2

2 ρ(q
2 )|T (q 2 )|2
+ + λ̄p (A − C) + dk + dq . (9.15)
A−B p=0
π −∞ (k2 − C)m 0 (q 2 − C)m

Because of the remark after Eq. (9.9) the integration along the LHC converges with just m − 1 subtractions. Indeed, we
can make use of this to show that several terms in Eq. (9.15) cancel each other because
Z ( Z Z )
(A − C)m L ∆(k2 ) (A − C)m−1 ∆(k2 )
L L ∆(k2 )
dk2 2 = − dk2 2 + dk2 2 ,
π −∞ (k − A)(k2 − C)m π −∞ (k − C)m −∞ (k − C)m−1 (k2 − A)
(9.16)

where we have added and subtracted k2 once in (A − C)m . The term before the last one in this equation cancels the
integration along the LHC multiplying (A − C)m−1 in Eq. (9.15). A similar result holds also for the integration along the
RHC since
Z ( Z Z )
(A − C)m ∞
2 ρ(q 2 )|T (q 2 )|2 (A − C)m−1 ∞ ρ(q 2 )|T (q 2 )|2 ∞ ρ(q 2 )|T (q 2 )|2
dq 2 = − dq 2 2 + dq 22
.
π 0 (q − C)m (q 2 − A) π 0 (q − C)m−1 0 (q − C)m−1 (q 2 − A)
(9.17)

By substituting Eqs. (9.16) and (9.17) into Eq. (9.15) the latter simplifies to
m−2
(Z Z )
λ X (A − C)m−1 L ∆(k2 ) ∞ ρ(q 2 )|T (q 2 )|2
T (A) = + λ̄p (A − C)p + dk 2
+ dq 2 2
,
A−B p=0
π −∞ (k2 − A)(k2 − C)m 0 (q − A)(q 2 − C)m
(9.18)

which is the standard from for a (m − 1)-times subtracted DR, cf. Eq. (9.7). This process can be further iterated in the
case that there are more bound states by the substituting T (A) → (A − B ′ )T (A) in Eq. (9.18), with B ′ the pole position
of the new bound state.
This DR is also appropriate in the case that there is no pole at A = B. In such case there is an extra constraint to be
fulfilled, namely,

lim (A − B)T (A) = λ = 0 . (9.19)


A→B

• The integration along the RHC requires just one subtraction because of the relation between the T - and the S-matrix
in partial waves,

S(A) = 1 + 2iρ(A)T , A > 0 . (9.20)

96
Since S(A) = ηe2iδ (η = 1 for the uncoupled case) then

M
|T (A)| < , (9.21)
ρ(A)

with M a positive constant. In turn, the DR along the LHC converges for vanishing ∆(A) in the limit A → −∞, while
for |∆(A)| < M (−A)m+α (with M a positive constant and 0 ≤ α < 1) one needs at least m + 1 subtractions. So that,as
soon as ∆(A) is not vanishing for A → −∞ the integration along the LHC is the one that determines the number of
subtraction constants.

9.1 Imposing the threshold behavior for higher partial waves


For partial waves with angular momentum ℓ ≥ 1 the amplitude T (A) vanishes as Aℓ for A → 0. In order to impose this
behavior in Eq. (9.7): i) We rewrite the bound state contribution as
ℓ−1  i  ℓ
1 1 1 1 X A A 1
=− =− + . (9.22)
A−B B 1 − A/B B i=0 B B A−B

ii) We take, at least, ℓ subtractions at C = 0.


iii) We impose that the polynomial contribution, stemming from the subtractive polynomial and the expansion of the
bound state contribution, Eq. (9.22), vanishes at least at Aℓ .
As a result, Eq. (9.7) becomes (m ≥ ℓ)

 ℓ m−1 Z
A β X (A − C)m−ℓ Aℓ L ∆(k2 )
T (A) = + λi Ai + dk2
B A−B i=ℓ
π −∞ (k2 − A)(k2 )ℓ (k2 − C)m−ℓ
Z
(A − C)m−ℓ Aℓ ∞ ρ(k2 )|T (k2 )|2
+ dk2 (9.23)
π 0 (k2 − A)(k2 )ℓ (k2 − C)m−ℓ

There is only contribution from the sum on the rhs of the previous equation for m > ℓ, so that m − 1 ≥ ℓ.
Eq. (9.23) is an non-linear IE for T (A) along the RHC. Once it is solved we can determine T (A) for any complex value
of T (A).
• Digression. If we consider the extreme case in which there is no LHC contribution (e.g. when pions are integrated
out at around threshold) then, because of the unitarity bound (9.21), T (A) satisfies
Z
A ∞ ρ(k2 )|T (k2 )|2
T (A) = ν0 δℓ0 + dk2 . (9.24)
π 0 (k2 − A)k2

We could even argue about the feasibility of an unsubtracted DR because since T (A) has only RHC in this case and for
A → ∞ one has Eq. (9.21), then because of the Sugawara-Kanazawa theorem T (A) (admiting that T (A) vanishes faster
than some finite power of s in the complex plane away from the real s axis) should vanish in any other direction of the
A-complex plane. The point to notice is that for D and higher waves one must have
Z ∞ ρ(k2 )|T (k2 )|2
dk2 = 0 , m = 2, . . . , ℓ (9.25)
0 (k2 )m

which comprises ℓ − 1 equations which cannot be satisfied because the integrand is definite positive. Then, one is driven
to add in Eq. (9.24) the contribution from (deeply) bound states, each of them bringing two free parameters (its pole
position and residue).

97
• We propose the following method to solve Eq. (9.23) because:
i) The method keeps T (A) unitary.
ii) Bound states originate dynamically from a given interaction input, which in turn stems from the subtraction constants
and the LHC.
The 0th input is
h i−1
T0 (A) = V0 (A)−1 + G(A; −µ2 ) , (9.26)
Z Z
A ∞ ρ(k2 ) µ2 ∞ ρ(k2 )
G(A; −µ2 ) = − dk2 − dk2
π 0 k2 (k2 − A) π 0 k2 (k2 + µ2 )

im A + i0+

=− , − (9.27)
4π 4π
m−1 Z
X (A − C)m−ℓ Aℓ L ∆(k2 )
V0 (A) = λi Ai + dk2 2 . (9.28)
i=ℓ
π −∞ (k − A)(k2 )ℓ (k2 − C)m−ℓ

We have chosen that

G(−µ2 ; −µ2 ) = 0 . (9.29)

For a given µ, the parameters β, B and the subtraction constants λi , i = ℓ, . . . , m − 1 are determined by the requirement
of minimizing the difference between T0 (A) and the rhs of Eq. (9.23). For that we could consider a χ2 function such that


X mMπ2 |Ai + C|m−ℓ Aℓi
Z ∞ k2 |T (k2 )|2 − |Tr (k2 )|2
χ2 = dk2 , i = 1, . . . , N (9.30)
i
4π 2 0 (k2 + Ai )(k2 )ℓ (k2 − C)m−ℓ

with −Ai a set of points in the low-energy region of the LHC (belonging to the expected region of validity of the EFT)
and r = 0 for the 0th order approach.
Once the minimum is reached we then take the output as the 1th iterative solution, T1 (A) which is then implemented
in the rhs of the DR in Eq. (9.23), and again the parameters β, B and the subtraction constants are varied to minimize
the χ2 in Eq. (9.30) but now with r = 1, and so on, until convergence is reached.
It could happen, when taking more than one subtraction, that the obtained T1 (A) diverges for A → +∞ because of
the subtractive polynomial of degree one or higher. In this case, the integral along the RHC and involving |T1 (A)|2 could
be divergent. As a result we could not further iterate the solution. This can be remedied by enforcing unitarity since the
inverse of a partial wave should satisfy

ImT −1 (A) = −ρ(A) . (9.31)

Then, the imaginary part of the inverse of the partial wave is fixed by unitarity. As a result if from the the minimization
process following the first iteration we have calculated

θ1 (A) ≡ ReT1 (A)−1 . (9.32)

We combine Eqs. (9.31) and (9.32) and redefine T1 (A), which now reads,
h i−1
T1 (A) = θ1 (A) − iρ(A) . (9.33)

In this way,T1 (A) is bounded by unitarity and we can use it again for a second iteration-minimization process in Eq. (9.23).
This process can be further iterated until convergence is reached.

98
• In order to take initial values for the fit it could be an option to match T0 (A) with the effective-range-expansion
(ERE). In this way, one could then determine the initial values of the subtraction constants λi in terms of the ERE shape
parameters. E.g. for S wave and imposing the scattering length (a) and effective range (r) in T0 (A), Eqs.(9.26)-(9.28),
one would have (we take C = 0 in Eq. (9.28))
" ! #−1
4π 4π 1 λ′2 2 A √
T0 (A) = − A + O(A ) − − i A . (9.34)
m m λ′1 λ′1 2 µ

Here we have employed a G(A; −µ2 ) function with two subtractions because V0 (A)−1 is fitted up to O(A) and, in this
way, the subtraction constants that would appear in G(A; −µ2 ) can be absorbed in the λ′i . Namely, we have now
Z √
2 A(A + µ2 ) ∞ 2 ρ(q 2 ) mA im A
G(A; −µ ) = − dq 2 2 = − . (9.35)
π 0 q (q + µ2 )(q 2 − A) 4πµ 4π
In Eq. (9.34) we have used primes in the λi to differentiate those in T0 (A) of the ones on the rhs of Eq. (9.23). The latter
are the ones that together with B and β should be fitted in order to minimize Eq. (9.30). It follows then
!
4π 1 λ′2 A 1 1
′ − ′ 2 A − = − + rA ,
m λ1 λ1 µ a 2
4πa
λ′1 = − ,
m
 
′ 4πa2 1 1
λ2 = − r+ . (9.36)
m 2 µ
• In harmony with the previous discussion we would also impose the exact reproduction of some of the ERE parameters
by expressing conveniently the subtraction constants λi in Eq. (9.23) in terms of these ERE parameters.
For instance, let us take ℓ = 0 with only one subtraction in Eq. (9.23) whose subtraction constant can be fixed in terms
of the scattering length (a) from the relation
β 4πa
λ1 − =− . (9.37)
B m
With this choice Eq. (9.23) becomes then
Z Z
A β 4πa A L
2 ∆(k2 ) A ∞ ρ(k2 )|T (k2 )|2
T (A) = − + dk 2 2
+ dk2 (9.38)
BA−B m π −∞ (k − A)k π 0 (k2 − A)k2
We continue with ℓ = 0 and discuss the constraint of imposing the effective range (r),
4π h 1 1 √ i−1 4πa h 1 √ i
T (A) = − + rA + O(A2 ) − i A =− 1 + raA − a2 A + ia A + O(A3/2 ) . (9.39)
m a 2 m 2
We have to compare it with the low-momentum expansion of the twice-subtracted DR
 2 Z Z
A β A2 L
2 ∆(k2 ) A(A − C) ∞ ρ(k2 )|T (k2 )|2
T (A) = + λ1 + λ2 A + dk 2 − A)(k 2 )2
+ dk2 , (9.40)
B A−B π −∞ (k π 0 (k2 − A)k2 (k2 − C)
where we have already reabsorbed in λ1,2 the first two terms in the low-momentum expansion of the bound-state contri-
bution. Matching the scattering length (independent term) we have the analogous result to Eq. (9.37)
4πa
λ1 = − . (9.41)
m

99
In the expansion up to and including O(A) one has to proceed carefully in the evaluation of the principal part of the
integration along the RHC (the last one in Eq. (9.40)). Its imaginary part gives

iρ(A)|T (A)|2 = iρ(A)λ21 + O(A3/2 ) , (9.42)

where we have taken into account from Eq. (9.39) that


h i
|T (A)|2 = λ21 1 + O(A) . (9.43)

This also implies for the real part of the same integral that
Z Z 
A(A − C) ∞ 2 ρ(k2 )|T (k2 )|2 AC ∞ 2 ρ(k2 ) |T (k2 )|2 − λ21
− dk 2 2 2
=− − dk
π 0 (k − A)k (k − C) π 0 (k2 − A)k2 (k2 − C)
Z
ACλ21 ∞ 2 ρ(k2 )
− − dk + O(A2 ) . (9.44)
π 0 (k2 − A)k2 (k2 − C)
Up to the same order in the expansion we can set A = 0 in the integrand of the first integral on the rhs, while the last
one can be done algebraically.
A simple way to proceed is to use the Cauchy’s integration formula by writing the principal part as
Z (Z
A(A − C) ∞ 2 ρ(k2 ) A(A − C) ∞ ρ(k2 )
− dk 2 2 2
= dk2
π 0 (k − A)k (k − C) 2π 0 (k2 − A − iε)k2 (k2 − C)
Z )

2 ρ(k2 )
+ dk . (9.45)
0 (k2 − A + iε)k2 (k2 − C)

Employing as contour of integration a circle of infinite radius that engulfs the RHC one has
I Z
1 ρ(z) ∞ ρ(k2 )
dz = dk2 , (9.46)
2 (z − A)z(z − C) 0 (k2 − A)k2 (k2 − C)
where we have employed that

ρ(k2 + iε) = −ρ(k2 − iε) , for k2 > 0 . (9.47)

In order to evaluate Eq. (9.46) one can apply the residue formula. Then, the integral in Eq. (9.45) becomes
I I 
A(A − C) ρ(z) ρ(z) m A m A
dz + dz = −i √ =− √ , (9.48)
4π (z − A − iε)z(z − C) (z − A + iε)z(z − C) 4π C 4π −C
with the understanding that C < 0 and that the cut of the square root is on the right. Matching the real parts of
Eqs. (9.39) and (9.40) up to O(A) we end with the result
  Z 2)

1 C ∞
2 ρ(k |T (k2 )|2 − λ21 mλ21
λ2 = λ1 ra − a2 + dk + √ . (9.49)
2 π 0 (k2 )2 (k2 − C) 4π −C
For this case the natural choice for T0 (A) is the one with its parameters fixed also to satisfy the ERE, given in Eq. (9.36).
If this is the case then we only have as possible free parameters B and β which, if present, should be fitted to minimize
Eq. (9.30). Were there no bound state then we just insert T0 (A) on the rhs of Eq. (9.23), calculate θ1 (A), Eq. (9.32),
and in terms of it T1 (A), Eq. (9.33). This is again substituted in the rhs of Eq. (9.23) and the process is further iterated
until convergence is reached.

100
′2
10 Dispersion relations for N (k, k ′; km ). The half-off-shell case.
In the following for simplicity of writing we remove the subscript ℓ in the function N , D and t.

10.1 Sketch of the situation


′2
• In the k′ -complex plane N (k, k′ ; km ) satisfies a DR of the form
Z
k′ 2

X i 1 ∆ℓ (k, p1 ) D(p21 )
N (k, k ; )= ai (k)k′ + dp1 . (10.1)
m i=0
π DC p1 − k ′

This is just a sketch where in the DR integral we have not explicitly written its precise form, which depends on the number
of subtractions taken. In this integral DC refers to the cuts analyzed in Secs. 3, 6, and ∆(k, p1 ) should be calculated
following similar techniques to those developed in Sec. 7.1 (this is explicitly performed in Sec. 11).
The knowledge of the hfs partial wave amplitude allows us to determine the potential in terms of it, Sec. 2, and the full-
off-shell partial wave (that we will use to determine bound state wave functions in momentum space, e.g. of the deuteron),
Sec. XXX. For these purposes we need the hfs partial wave in a partition of the RHC, that is, we need N (ki , kj ; kj2 /m + iε),
with ki,j real and positive and with a small positive imaginary part, Sec. 2. The DCs in the p1 -complex plane for real ki ,
the one of interest for the DR in Eq. (10.1), are specially simple and correspond to the vertical lines given in Eq. (3.87),
with p1 substituted by ki . The calculation of the related ∆(ki , p1 ) is also simplified by applying the fact that the partial
wave amplitude is an even/odd function of the momenta and satisfies the Schwarz reflection principle, Sec. 5.1. Then,
we only need to evaluate it in the first quadrant of the p1 -complex plane. This so because t(ki , −p1 ) = (−1)ℓ t(ki , p1 ),
t(ki , p∗1 ) = t(ki , p1 )∗ , which also imply t(ki , −p∗1 ) = (−1)ℓ t(ki , p1 )∗ .
• It is important to realize that the subtraction ‘constants’ ai (k2 ) in Eq. (10.1) are actually functions of k2 . By equating
Eq. (10.1) to the on-shell case we are left with the constraint
Z
X
′ ′i ′ 1 ′ ∆ℓ (k′ , p1 ) D(p21 )
ai (k )k = N (k , k ) − dp1 . (10.2)
i=0
π DC p1 − k ′

This is enough to fix the subtraction functions only in the case when one subtraction is taken, a0 (k2 ). Otherwise, we cannot
fix the subtraction functions from this equation alone, neither it can be done by taking derivatives with respect to k′ 2 on
both sides of the equality because then new unknown functions would emerge, namely, the derivatives of the subtraction
functions. Thus, we also need to write DRs in the k-complex plane to disentangle the k dependence of the subtraction
functions in Eq. (10.1). An interesting question to elucidate is whether by proceeding in this way the knowledge of on-shell
scattering is enough to determine the k2 dependence of ai (k2 ). These aspects are discussed in detail in Sec. 10.2.

10.2 DRs for N(k, k ′ ; k ′ 2 /m) in the variables k and k ′ .


• Let us now consider the detailed writing of DRs for the function N (k, k′ ; k′ 2 /m). It will be understood in the following
that the energy is given by the last momentum argument of N according to the standard nonrelativistic relation between
energy and momentum, employing the physical mass of the nucleon. With this aim we consider the function

N (k, z)
, (10.3)
(z − k′ )(z − k0 )m′

where m′ subtractions have been taken at k0 , and apply the Cauchy’s theorem of complex integration to the contour C
shown in Fig. 25, with the understanding that m′ is large enough to guarantee that no contribution results from the part

101
Figure 25: Deformed integration contour in the k′ -complex plane that is employed to derive the DR for N (k, k′ ; k′ 2 /m)
in Eq. (10.5). The circulation is counterclockwise. The basic form of the contour is independent of whether k is complex
or not, though the figure is done for real k (for these values the discontinuity across the cuts are evaluated in Sec. 11).
The vertical lines have branch points at ±(k ± imπ ), with unrelated ± signs.

of the contour at infinity. As a result we have


I m −1′
N (k, z) N (k, k′ ) X ′
dz ′ m′ = 2πi ′ m ′ − 2πi ai (k)(k′ − k0 )−m +i .
(z − k )(z − k0 ) (k − k0 ) i=0
I X4 Z
N (k, z) N (k, z)
dz ′ = dz , (10.4)

(z − k )(z − k0 )m
s=1 DCs
(z − k′ )(z − k0 )m′

where the last integrals extend over the four vertical DCs with branch points at ±(k ± imπ ). Equating both expressions
in the equation above it follows that
m′ −1 ′ 4 Z

X
′ i(k′ − k0 )m X N (k, z)
N (k, k ) = ai (k)(k − k0 ) + dz ′ )(z − k )m′
. (10.5)
i=0
2πi s=1 DCs
(z − k 0

We change of integration variable in these integrals to


Im(z − k) = ν1 . (10.6)
so that ν1 extends from ±mπ up to ±∞, respectively. The four integrals in the last sum of Eq. (10.5) can then be
rewritten as
Z ∞N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
i dν1
mπ (k + iν1 − k′ )(k + iν1 − k0 )m′
Z ∞
N (k, −k − ε + iν1 ) − N (k, −k + ε + iν1 )
+i dν1
mπ (−k + iν1 − k′ )(−k + iν1 − k0 )m′
Z ∞
N (k, −k − ε − iν1 ) − N (k, −k + ε − iν1 )
+i dν1
mπ (−k − iν1 − k′ )(−k − iν1 − k0 )m′
Z ∞
N (k, k − ε − iν1 ) − N (k, k + ε − iν1 )
+i dν1 . (10.7)
mπ (k − iν1 − k′ )(k − iν1 − k0 )m′

102
−mπ R R
In the last two integrals we have redefined the integration variable ν1 → −ν1 so that −∞ idν1 → m∞π idν1 . We now
take into account that N (k, −z) = (−1)ℓ N (k, z),32 which allows us to relate the first and second lines in Eq. (10.7) with
the third and fourth ones, in order, so that Eq. (10.7) becomes
Z ∞  
i dν1 N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )


 1 (−1)ℓ+m
× ′ m′ + ′ m′
(k + iν1 − k )(k + iν1 − k0 ) (k + iν1 + k )(k + iν1 + k0 )
Z ∞  
+i dν1 N (k, k − ε − iν1 ) − N (k, k + ε − iν1 )


 (−1)ℓ+m 1
× ′ m′ + ′ m′
(k − iν1 + k )(k − iν1 + k0 ) (k − iν1 − k )(k − iν1 − k0 )
Z ∞  
=i dν1 N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
−∞

 1 (−1)ℓ+m
× ′ m′ + ′ m′ . (10.8)
(k + iν1 − k )(k + iν1 − k0 ) (k + iν1 + k )(k + iν1 + k0 )
For the latter equation one has to take into account that the discontinuity of N in the integrand is non-zero only for
ν12 > m2π . With this result we can express Eq. (10.5) as

m′ −1 ′ Z

X
′ (k′ − k0 )m
i
∞  
N (k, k ) = ai (k)(k − k0 ) + dν1 N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
i=0
2π −∞
( ′
)
1 (−1)ℓ+m
× ′ m′ + ′ . (10.9)
(k + iν1 − k )(k + iν1 − k0 ) (k + iν1 + k′ )(k + iν1 + k0 )m

The previous DR is valid for complex k and k′ .


• Let us consider the DR for N (k, k′ ) in the k-complex plane. We take now the function

N (z, k′ )
, (10.10)
(z − k)(z − k0 )m

and the integration contour is analogous to that in Fig. 25, with vertical cuts starting at the branch points k = ±(k′ ±imπ ),
with unrelated ± signs. Of course, m is large enough so that no contribution arises from the parts of the contour at infinity.
Instead of Eq. (10.5) we now have
m−1 4 Z

X
′ i (k − k0 )m X N (z, k′ )
N (k, k ) = bi (k )(k − k0 ) + dz . (10.11)
i=0
2πi s=1 DCs
(z − k)(z − k0 )m

And analogous manipulations to those used above to arrive at the last expression in Eq. (10.8) allow us to rewrite Eq. (10.11)
as
m−1 Z
X (k − k0 )m ∞  
N (k, k′ ) = bi (k′ )(k − k0 )i + dν1 N (k′ − ε + iν1 , k′ ) − N (k′ + ε + iν1 , k′ )
i=0
2π −∞

 1 (−1)ℓ+m
× ′ ′ m
+ ′ ′ m
. (10.12)
(k + iν1 − k)(k + iν1 − k0 ) (k + iν1 + k)(k + iν1 + k0 )

32
If N (k, z) corresponds to a mixing partial wave note that (−1)ℓ = (−1)ℓ .

103
This form for the DR is simpler than the one in Eq. (10.9) because the energy argument is fixed to k′ 2 /m (not to confuse
the nucleon mass m with the index above). It can also be used for complex k and k′ .
• Let us now study the constraint on the subtraction functions that arises by the properties

N (k, −k′ ) = (−1)ℓ N (k, k′ ) ,


N (−k, k′ ) = (−1)ℓ N (k, k′ ) . (10.13)

For simplicity we take k0 = 0 because in other case the derivation is longer and more involved. In addition, in Sec. 10.3 we
derive the corresponding DRs for N (k, k′ ) in the k2 - or k′ 2 -complex planes where one can take k0 6= 0 straightforwardly.
Applying Eq. (10.9) for N (k, −k′ ) one has

m′ −1 ′ Z

X (−k′ )m ∞  N (k, k − ε + iν1 ) − N (k, k + ε + iν1 ) 
N (k, −k ) = ai (k)(−k′ )i + dν1
i=0
2π −∞ (k + iν1 )m′
( ′
)
1 (−1)ℓ+m
× +
k + iν1 + k′ k + iν1 − k′
m′ −1 ′ Z
X (k′ )m ∞  N (k, k − ε + iν1 ) − N (k, k + ε + iν1 ) 
= ai (k)(k′ )i (−1)i + (−1)ℓ dν1 ′
i=0
2π −∞ (k + iν1 )m
( ′
)
1 (−1)ℓ+m
× + . (10.14)
k + iν1 − k′ k + iν1 + k′

Since this expression must be (−1)ℓ N (k, k′ ) it then follows that

(−1)i = (−1)ℓ , (10.15)

and then only even(odd) powers of k′ are allowed in the k′ -polynomial of the DR for even(odd) ℓ. The same conclusion
applies as well for bi (k′ ) in the DR for the function N (k, k′ ) in the variable k, Eq. (10.12).
• For real values of k′ the function N (k, k′ ) vanishes as kℓ for k → 0 and, similarly, for real values of k it vanishes as

k′ as k′ → 0. The point is that the DC in the variable ′ ′
q k, when k is real, extends along ±(k ± iν) with |ν| ≥ mπ , so
that there is a finite radius of convergence given by k′ 2 + m2π for the series of N (k, k′ ) in powers of k, and viceversa.
This implies that the polynomial in k′ , k in the DRs of Eqs. (10.9) and (10.12) can be chosen to contain only nonzero
subtraction functions for i ≥ ℓ by taking ℓ or more subtractions. As a result we can write down these DRs as
p1 Z
′ ′ ℓ
X
′ 2j (k′ )m1 ∞  N (k, k − ε + iν1 ) − N (k, k + ε + iν1 ) 
N (k, k ) = (k ) aℓ+2j (k)(k ) + dν1
j=0
2π −∞ (k + iν1 )m1
( )
1 (−1)ℓ+m1
× + , (10.16)
k + iν1 − k′ k + iν1 + k′
p2 Z
X km2 ∞  N (k ′ − ε + iν1 , k ′ ) − N (k ′ + ε + iν1 , k ′ ) 
N (k, k′ ) = kℓ bℓ+2j (k′ )k2j + dν1
j=0
2π −∞ (k′ + iν1 )m2
 1 (−1)ℓ+m2
× + , (10.17)
k′ + iν1 − k k′ + iν1 + k
 
mi − 1 − ℓ
pi = . (10.18)
2

104
Though there is no relationship between mi and ℓ, apart from the fact that mi ≥ ℓ, nothing is won by taking an mi such
that (−1)ℓ+mi = −1. E.g. let us assume that this is the case for m1 then Eq. (10.16) becomes
p1 Z  
′ ′ℓ
X
′ 2j (k′ )m1 ∞  N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )  1 1
N (k, k ) = k aℓ+2j (k)(k ) + dν1 − ,
j=0
2π −∞ (k + iν1 )m1 k + iν1 − k ′ k + iν1 + k′
p1 Z
ℓ X (k′ )m1 +1 ∞ N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
= k′ aℓ+2j (k)(k′ )2j + dν1 . (10.19)
j=0
π −∞ (k + iν1 )m1 ((k + iν1 )2 − k′ 2 )

The resulting DR for the case (−1)ℓ+m1 = +1 can also be worked out with the result,
p1 Z  
′ ′ℓ
X
′ 2j (k′ )m1 ∞  N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )  1 1
N (k, k ) = k aℓ+2j (k)(k ) + dν1 + ,
j=0
2π −∞ (k + iν1 )m1 k + iν1 − k′ k + iν1 + k′
p1 Z
′ℓ
X
′ 2j (k′ )m1 ∞ N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
=k aℓ+2j (k)(k ) + dν1 , (10.20)
j=0
π −∞ (k + iν1 )m1 −1 ((k + iν1 )2 − k′ 2 )

which is the same as Eq. (10.19) with the replacement in the latter of m1 + 1 → m1 (note that m1 in Eq. (10.19) is such
that (−1)ℓ+m1 +1 = +1).
• We now consider how to fix the subtraction functions bi (k′ ) in terms of a few constants. For that we equate Eqs. (10.9)
and (10.12). In this way one has:
m−1 m′ −1
X X
′ i
bi (k )(k − k0 ) − ai (k)(k′ − k0 )i = f (k, k′ , k0 ) , (10.21)
i=0 i=0

with
Z (
1 ∞   1

f (k, k , k0 ) = dν1 N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
2π −∞ (k + iν1 − k′ )(k + iν1 − k0 )m′

(−1)ℓ+m  ′  
+ (k′ − k0 )m − N (k′ − ε + iν1 , k′ ) − N (k′ + ε + iν1 , k′ )
(k + iν1 + k′ )(k + iν1 + k0 )m′
)
 1 (−1)ℓ+m 
× + (k − k0 )m . (10.22)
(k′ + iν1 − k)(k′ + iν1 − k0 )m (k′ + iν1 + k)(k′ + iν1 + k0 )m

This is a known function once the discontinuities of N in the integrand are calculated (the needed IEs are derived below
in Sec. 11). To make clear how to proceed let us take k = k0 in Eq. (10.21), it then results
m′ −1
X

b0 (k ) = ai (k0 )(k′ − k0 )i + f (k0 , k′ , k0 ) . (10.23)
i=0

Next, substitute it into Eq. (10.21) which then becomes


m−1 m′ −1
X X  
′ i
bi (k )(k − k0 ) = ai (k) − ai (k0 ) (k′ − k0 )i + f (k, k′ , k0 ) − f (k0 , k′ , k0 ) . (10.24)
i=1 i=0

We can isolate b1 (k′ ) by taking one derivative with respect to k at k = k0 , with the result
m′ −1
X (1)
b1 (k′ ) = ai (k0 )(k′ − k0 )i + f (1) (k0 , k′ , k0 ) , (10.25)
i=0

105
where the superscript indicates the number of derivatives with respect to k. Again substitute this expression for b1 (k′ ) in
Eq. (10.24) with the result,
m−1 m′ −1
X X  (1) 
′ i
bi (k )(k − k0 ) = ai (k) − ai (k0 ) − ai (k0 )(k − k0 ) (k′ − k0 )i
i=2 i=0
+ f (k, k′ , k0 ) − f (k0 , k′ , k0 ) − f (1) (k0 , k′ , k0 )(k − k0 ) , (10.26)

take two derivatives with respect to k at k = k0 and then b2 (k′ ) is determined. This process gives in general

′ 1 mX −1
(j) 1 ∂j
bj (k ) = ai (k0 )(k′ − k0 )i + f (k0 , k′ , k0 ) . (10.27)
j! i=0 j! ∂kj

(j)
In this way we can determine the subtraction functions bi (k′ ) in terms of m × m′ constants ai (k0 )

a0 (k0 ) ··· a1 (k0 ) am′ −1 (k0 )


(1) (1) (1)
a0 (k0 ) ··· a1 (k0 ) am′ −1 (k0 )
(10.28)
···
(m−1) (m−1) (m−1)
a0 (k0 ) a1 (k0 ) · · · am′ −1 (k0 )
(j)
We have chosen this way of proceeding, and not the way around of expressing the ai (k) in terms of the bi (k0 ), because,
as mentioned above, the DR for N (k, k′ ) in the k-complex plane, Eq. (10.12), is simpler, and nothing is won by proceeding
in this other way because we would have the same number of constants to fix.
Notice that we have not implemented the requirement that (−1)ℓ+j = +1 in obtaining Eq. (10.27), discussed in
Eq. (10.15). This is not further done here because when writing the DR for N (k, k′ ) in terms of k2 and k′ 2 it will come
out automatically.
The DR in Eq. (10.12), by taking into account Eq. (10.27), can be finally written as

m−1 ′
X (k − k0 )j mX −1
(j)
m−1
X (k − k0 )j
N (k, k′ ) = ai (k0 )(k′ − k0 )i + f (j) (k0 , k′ , k0 )
j=0
j! i=0 j=0
j!
Z
(k − k0 )m ∞  
+ dν1 N (k′ − ε + iν1 , k′ ) − N (k′ + ε + iν1 , k′ )
2π −∞
 1 (−1)ℓ+m
× ′ ′ m
+ ′ ′ m
(10.29)
(k + iν1 − k)(k + iν1 − k0 ) (k + iν1 + k)(k + iν1 + k0 )

Once this equation is particularized for on-shell scattering, it results


m−1 ′ −1
X mX 1 (j) m−1
X 1
′ ′
N (k , k ) = ai (k0 )(k′ − k0 )i+j + f (j) (k0 , k′ , k0 )(k′ − k0 )j
j=0 i=0
j! j=0
j!
Z
(k′ − k0 )m ∞  
+ dν1 N (k′ − ε + iν1 , k′ ) − N (k′ + ε + iν1 , k′ )
2π −∞
 1 (−1)ℓ+m
× ′ m
+ ′ ′ m
(10.30)
iν1 (k + iν1 − k0 ) (2k + iν1 )(k + iν1 + k0 )
(j)
This equation clearly establishes that only the sum of constants ai (k0 ) with i + j=fixed, can be resolved by studying
on-shell scattering. That is, among the m × m′ constants in Eq. (10.28) only m + m′ − 1 can be determined from on-shell

106
scattering. This two numbers coincide only when m or m′ are equal to 1. For m = 0 or m′ = 0 there would be no
subtraction functions in Eqs. (10.12) or (10.9), respectively, and everything will be given just in terms of ±f (k, k′ , k0 ) and
its derivatives (+ when expressing the bi (k′ ) and − when doing the corresponding for the ai (k).) In any other case, more
information than available from on-shell scattering would be needed. This would be expected to be the case for higher
orders calculations in the chiral EFT because the chiral expansion would involve increasingly divergent discontinuities of
the N function in the DRs of Eqs. (10.9) and (10.12).

10.2.1 The subtraction functions ai (k) and bi (k′ ) are entire functions
Let us that the subtraction functions ai (k) and bi (k′ ) have no DCs for real k and k′ . This is indeed an expected result
because the DCs in the variable k(k′ ) in the functions ai (k)(bi (k′ )) would have branch points dependent on k′ (k), which
cannot be because all the k′ (k) dependence is encoded explicitly in the DR of Eq. (10.9)(Eq. (10.12)). Thus, no further
implicit dependence in the subtraction functions is allowed.
In this section we show that the subtraction functions ai (k) and bi (k′ ) have no DCs for real k′ , k, respectively, so that
they are entire or holomorphic functions in the range of values we are interested in. Let us first discuss the subtraction
functions ai (k), Eq. (10.9), and consider the discontinuity of both sides of the equation between the values k′ + iν2 ± ε
for the argument k, given the fact that the DC would be located at k′ + iν2 , |ν2 | > mπ (the discontinuity for the cut at
−k′ + iν2 can be derived from the former). We do not need to deform the integration contour for the DR in this equation
because we have already taken into account the actual value of the argument k to avoid the resulting cuts in the complex
k′ -plane (in this case we have just to particularize for k → k′ + iν2 ± ε). Let us concentrate first in the discontinuity of
the DR integral in Eq. (10.9), that is given by:
Z ∞  
dν1 N (k′ − ε + iν2 , k′ − ε − δ + iν2 + iν1 ) − N (k′ − ε + iν2 , k′ − ε + δ + iν1 + iν2 )
−∞
 1 (−1)ℓ+m 
× ′ m
+ ′ ′ m
(−ε + iν1 + iν2 )(k − ε + iν1 + iν2 − k0 ) (2k − ε + iν1 + iν2 )(k − ε + iν1 + iν2 + k0 )
Z ∞  
− dν1 N (k′ + ε + iν2 , k′ + ε − δ + iν1 + iν2 ) − N (k′ + ε + iν2 , k′ + ε + δ + iν1 + iν2 )
−∞
 1 (−1)ℓ+m 
× ′ m
+ ′ ′ m
, (10.31)
(ε + iν1 + iν2 )(k + ε + iν1 + iν2 − k0 ) (2k + ε + iν1 + iν2 )(k + ε + iν1 + iν2 + k0 )

where as usual 0 < δ < ε. The previous equation simplifies given the fact that the discontinuity of N itself is continuous
in ε, which also implies that the terms proportional to (−1)ℓ+m cancel between each other. We are then left with
Z ∞  
dν1 N (k′ + iν2 , k′ − δ + iν1 + iν2 ) − N (k′ + iν2 , k′ + δ + iν1 + iν2 )
−∞
 1 1  1
× − . (10.32)
−ε + iν1 + iν2 ε + iν1 + iν2 (k + iν1 + iν2 − k0 )m

Taking into account that


1 1
− = −2πδ(ν1 + ν2 ) , (10.33)
−ε + iν1 + iν2 ε + iν1 + iν2
the ν1 integration in Eq. (10.32) can be trivially done with the result
  1
−2π N (k′ + iν2 , k′ − δ) − N (k′ + iν2 , k′ + δ) (10.34)
(k′ − k0 )m

107
20

10

-30 -25 -20 -15 -10 -5

-10

-20

Figure 26: Deformed integration contour in the k′ 2 -complex plane that is employed to derive the DR for τ (k2 , k′ 2 ) in
Eq. (10.40). The circulation is counterclockwise. The branch points are located at k2 − m2π ± 2ikmπ .

It then follows that the discontinuity in both sides of Eq. (10.9) reduces to
m−1
X  
N (k′ − ε + iν2 , k′ ) − N (k′ + ε + iν2 , k′ ) = ai (k′ − ε + iν2 ) − ai (k′ + ε + iν2 ) (k′ − k0 )i
i=0
(k′ − k0 )m −(2π)  
+ ′ m
N (k′ + ε + iν2 , k′ ) − N (k′ − ε + iν2 , k′ ) , (10.35)
2π (k − k0 )

and hence,

ai (k′ − ε + iν2 ) − ai (k′ + ε + iν2 ) = 0 . (10.36)

For the cut at −k′ + iν2 we make use of the property


 
N (k′ − ε + iν2 , k′ ) − N (k′ + ε + iν2 , k′ ) = N (−k′ + ε − iν2 , k′ ) − N (−k′ − iν2 − ε, k′ ) (−1)ℓ , (10.37)

which implies, from Eq. (10.36), that the discontinuity also vanishes in this case. As a result the function ai (k) are entire
function in the k-complex plane, since there are no complex poles in the first Riemann sheet (the physical bound states
appear are zeros of the D(A) function).33
We can also proceed similarly for the subtraction functions bi (k′ ) in Eq. (10.12) and show that the discontinuity across
the DCs located at ±k + iν2 , |ν2 | > mπ , is zero. These subtraction functions are then entire functions in the k′ -complex
plane.

10.3 DRs for N(k, k ′ ; k ′ 2 /m) in the variables k 2 and k ′ 2 .


• As discussed above to justify Eqs. (10.17) and (10.18) the function N (k, z), for real k, has a power expansion in the
variable z starting at z ℓ and, analogously, for real k′ the function N (z, k′ ) has a power expansion starting at z ℓ . Indeed, this
is the case even for complex k and k′ as long as their moduli are smaller than mπ . The only change is in the value of the
33
We can think of a DR for an analytical function so that if there is no discontinuity across the corresponding cut, because the discontinuity
is zero, then function is C∞ in this region. Note also that there are entire functions which grow faster than any polynomial and would not admit
any DR.

108
radius of convergence. In addition we also have the properties N (k, −z) = (−1)ℓ N (k, z) and N (−z, k′ ) = (−1)ℓ N (z, k′ ),
Eq. (7.8) and subsequent discussion. As a result N (k, k′ )/(kk′ )ℓ is function of k2 and k′ 2 and it does not have a pole at
the origin as k or k′ → 0. In the following we denote this function by τ (k2 , k′ 2 ), namely,

2 N (k, k′ )
τ (k2 , k′ ) = . (10.38)
(kk′ )ℓ

• We first consider the DR of τ (k2 , k′ 2 ) in the variable k′ 2 for real k and follow similar steps to those in Sec. 10.2,
although now in the k′ 2 -complex plane. The DCs are located at
2
k′ = (k + iν)2 = k2 − ν 2 + 2ikν , (10.39)

with |ν| ≥ mπ , as shown in Fig. 26. Then, we take as starting point the integral along the integration contour that
circumvents such cuts as shown in Fig. 26,
I
τ (k2 , z)
dz , (10.40)
(z − k′ 2 )(z − k02 )m′

where m′ subtractions are taken at k02 . Applying the Cauchy theorem for complex integration we then have
m′ −1 Z
(k′ 2 − k02 )m

2 ′2
X
2 ′2 τ (k2 , q 2 − n̂ε) − τ (k2 , q 2 + n̂ε)
τ (k , k ) = ai (k )(k − k02 )i + dq 2
i=0
2πi up (q 2 − k′ 2 )(q 2 − k02 )m′
Z
(k′ 2 − k02 ) m′ τ (k2 , q 2 − n̂ε) − τ (k2 , q 2 + n̂ε)
+ dq 2 , (10.41)
2πi down (q 2 − k′ 2 )(q 2 − k02 )m′
where up is the integration contour with positive imaginary part and down the one with negative imaginary part, which
indeed is the complex conjugate of the contour up. In the previous equation n̂ is a unit vector perpendicular to the contour
pointing downwards.
Now we change of integration variable to
q
q= q2 , (10.42)

with the square-root cut along the negative axis, −1 ± iε = ±i (we keep in this way that the new integration contours
are complex conjugate of each other as well). We also rewrite
 
1 1 1 1
2 ′ 2 = 2q +
q + k′ q − k′
. (10.43)
q −k
Then, the first DR integral on the rhs of Eq. (10.41) transforms to
Z
(k′ 2 − k02 )m

k+i∞ τ (k2 , (q − ε)2 ) − τ (k2 , (q + ε)2 )
dq
2πi k+imπ (q − k′ )(q 2 − k02 )m′
Z
(k′ 2

− k02 )m k+i∞ τ (k2 , (q − ε)2 ) − τ (k2 , (q + ε)2 )
+ dq . (10.44)
2πi k+imπ (q + k′ )(q 2 − k02 )m′

It is easy to convince oneself that the sense of circulation in the variable q, Eq. (10.42), corresponds to that in the variable
q 2 , Eq. (10.41).

(k ± ε + iν)2 = k2 − ν 2 + 2i(k ± ε)ν , (10.45)

109
with a slightly larger/lower imaginary part for ±ε, respectively. One can proceed analogously for the second integral on
the rhs of Eq. (10.41), with the result
Z
(k′ 2 − k02 )m

k−imπ τ (k2 , (q − ε)2 ) − τ (k2 , (q + ε)2 )
dq ′
2πi k−i∞ (q − k′ )(q 2 − k02 )m
Z
(k′ 2

− k02 )m k−imπ τ (k2 , (q − ε)2 ) − τ (k2 , (q + ε)2 )
+ dq ′ . (10.46)
2πi k−i∞ (q + k′ )(q 2 − k02 )m

We display the total result of both integrals in terms of the variable ν, with

Imq = ν1 , (10.47)

and ν1 was already introduced in the DR for N (k, k′ ) in the momentum variables, Eq. (10.6). Eq. (10.41) can then be
written as
m′ −1 Z
(k′ 2 − k02 )m

2 ′2
X 2
∞ τ (k2 , (k − ε + iν1 )2 ) − τ (k2 , (k + ε + iν1 )2 )
τ (k , k ) = ai (k2 )(k′ − k02 ) + dν1
i=0
2π −∞ ((k + iν1 )2 − k02 )m′
 
1 1
× ′
+ . (10.48)
k + iν1 − k k + iν1 + k′

Although ν1 extends from −∞ to +∞ the integrand is nonzero only for |ν1 | > mπ .
Reexpressing Eq. (10.48) in terms of the original function N (k, k′ ), Eq. (10.38), one has

m′ −1
′2 ′ Z
′ ℓ ′ℓ
X 2 ℓ (k − k02 )m +∞ N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
N (k, k ) = k k ai (k2 )(k′ − k02 )i + k′ dν1 .
i=0
π −∞ (k + iν1 )ℓ−1 ((k + iν1 )2 − k′ 2 )((k + iν1 )2 − k02 )m′
(10.49)

This result is analogous to Eqs. (10.19) and (10.20), although now the formalism is equally simple for any chosen value of
the subtraction point k02 and we also learn here that kℓ aℓ+2j (k) in those equations is a function of k2 and corresponds to
k2ℓ aj (k2 ), with aj (k2 ) present in Eq. (10.49).
We can simplify further the IE of Eq. (10.49) by noticing that, cf. Eq. (7.5),

N (k, k − ε − iν1 ) − N (k, k + ε − iν1 ) = N (k, k − ε + iν1 )∗ − N (k, k + ε + iν1 )∗ (10.50)

Then Eq. (10.49) reads

m′ −1
X
′ ℓ ′ℓ 2
N (k, k ) = k k ai (k2 )(k′ − k02 )i
i=0
′ 2 2 m′ Z +∞
′ ℓ (k − k0 ) N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
+k 2Re dν1 . (10.51)
π mπ (k + iν1 )ℓ−1 ((k + iν1 )2 − k′ 2 )((k + iν1 )2 − k02 )m′

We can also proceed analogously in the variable k2 and write a DR in this variable taking as starting point the integral,
I
τ (z, k′ 2 )
dz . (10.52)
(z − k2 )(z − k02 )m

110
One has the result
m−1 Z
ℓ X 2 (k2 − k02 )m +∞ N (k′ − ε + iν1 , k′ ) − N (k′ + ε + iν1 , k′ )
N (k, k′ ) = kℓ k′ bi (k′ )(k2 − k02 )i + kℓ dν1 .
i=0
π −∞ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m
(10.53)
For later convenience, in order to evaluate explicitly the discontinuity in the integral of the previous equation, we have to
specify an infinitesimal imaginary part in the second argument of N (z, k′ ). Since for real k the function N (k, k′ ) has no
RHC in k′ we can then take the symmetric combination
1 
N (k, k′ ) = N (k, k′ + iǫ) + N (k, k′ − iǫ) . (10.54)
2
For this combination Eq. (10.53) becomes
m−1 Z
ℓ X 2 (k2 − k02 )m +∞ 1
N (k, k′ ) = kℓ k′ bi (k′ )(k2 − k02 )i + kℓ dν1
i=0
2π −∞ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m
 
× N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) + N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ) .
(10.55)
Now, we can split the previous integral in two parts, one for positive values of ν1 and the other for negative ones. For the
later contribution we perform the change of integration variable ν1 → −ν1 and then Eq. (10.55) reads
m−1 Z 
ℓ X 2 (k2 − k02 )m +∞ 1
N (k, k′ ) = kℓ k′ bi (k′ )(k2 − k02 )i + kℓ dν1
i=0
2π mπ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m
 
× N (k − ε + iν1 , k + iǫ) − N (k + ε + iν1 , k + iǫ) + N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ)
′ ′ ′ ′

1
+ ′ (10.56)
(k − iν1 ) ((k − iν1 ) − k2 )((k′ − iν1 )2 − k02 )m
ℓ−1 ′ 2
 
× N (k′ − ε − iν1 , k′ + iǫ) − N (k′ + ε − iν1 , k′ + iǫ) + N (k′ − ε − iν1 , k′ − iǫ) − N (k′ + ε − iν1 , k′ − iǫ) .
Next, because of Eq. (7.5) we have that
N (k′ − ε − iν1 , k′ + iǫ) − N (k′ + ε − iν1 , k′ + iǫ) + N (k′ − ε − iν1 , k′ − iǫ) − N (k′ + ε − iν1 , k′ − iǫ)
 ∗
= N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) + N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ) (10.57)
and then Eq. (10.56) simplifies to
m−1 2 Z 
′ ℓ ′ℓ
X
′2 2 ℓ (k − k02 )m +∞ 1
N (k, k ) = k k bi (k )(k − k02 )i +k Re dν1
i=0
π mπ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m
 
× N (k − ε + iν1 , k + iǫ) − N (k + ε + iν1 , k + iǫ) + N (k − ε + iν1 , k − iǫ) − N (k′ + ε + iν1 , k′ − iǫ) .
′ ′ ′ ′ ′ ′

(10.58)
As shown below in Sec. 11 the discontinuities N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) and N (k′ − ε + iν1 , k′ −
iǫ) − N (k′ + ε + iν1 , k′ − iǫ) with ν1 ≥ mπ are the same, cf. Eq. (11.21). Taking this into account Eq. (10.58) becomes
m−1
X
ℓ 2
N (k, k′ ) = kℓ k′ bi (k′ )(k2 − k02 )i
i=0
Z
ℓ (k2 − k02 )m +∞ N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ)
+k 2Re dν1 . (10.59)
π mπ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m

111
• We can also proceed similarly as done above in Sec. 10.2 to express the subtraction functions in terms of a few
constants and a function known. By equating Eqs. (10.51) and (10.59) we arrive to the requirement
m−1 m′ −1
X X
′2 2 2 2
bi (k )(k − k02 )i − ai (k2 )(k′ − k02 )i = ω(k2 , k′ , k02 ) ,
i=0 i=0
Z ∞
(k′ 2 2 m′
2 − k0 ) N (k, k − ε + iν1 ) − N (k, k + ε + iν1 )
ω(k2 , k′ , k02 ) = 2Re dν1
πk ℓ
mπ (k + iν1 )ℓ−1 ((k + iν1 )2 − k′ 2 )((k + iν1 )2 − k02 )m′
Z
(k2 − k02 )m ∞ N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ)
− 2Re dν 1 . (10.60)
πk′ ℓ mπ (k′ + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k + iν1 )2 − k02 )m

Proceeding similarly as in Sec. 10.2, instead of Eq. (10.27), we have now



′2 1 mX −1
(j) 2 1 ∂j 2
bj (k ) = ai (k02 )(k′ − k0 2 )i + w(k02 , k′ , k02 ) . (10.61)
j! i=0 j! ∂k2 j

Notice that when taking the derivatives with respect to k2 of the function ω(k2 , k′ 2 , k02 ) at k02 , only the first integral in
the definition of this function in Eq. (10.60) plays a role. In terms of Eq. (10.61) we can rewrite Eq. (10.59) as

m−1 ′ −1
X mX (k 2 − k02 )j (k′ 2 − k02 )i m−1
X 1 ∂ j w(k 2 , k ′ 2 , k 2 )
′ ℓ ′ℓ (j) ℓ ′ℓ
N (k, k ) = k k ai (k0 2 ) +k k 0
j
0
(k2 − k02 )j
j=0 i=0
j! j=0
j! ∂k 2
Z
(k2 − k02 )m +∞ N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ)
+ kℓ 2Re dν1 . (10.62)
π mπ (k + iν1 )ℓ−1 ((k′ + iν1 )2 − k2 )((k′ + iν1 )2 − k02 )m

(j)
The same discussion on determining the constants ai (k02 ) from on-shell scattering done at the end of Sec. 10.2 can be
repeated here as well.

11 Calculation of the discontinuities needed in the DRS for half-off-shell scattering


Let us recall that N (k, k′ ; k′ 2 /m) = t(k, k′ ; k′ 2 /m)D(k′ 2 ), Eq. (5.12), so that the discontinuity of the function N (k, k′ ; k′ 2 /m)
across a DC is straightforwardly related to the discontinuity of the function t(k, k′ ; k′ 2 /m) through the function D(k′ 2 ).
Hence, we concentrate in the following in the evaluation of the discontinuities of t(k, k′ ; k′ 2 /m) itself.
• We first consider the DR of Eq. (10.59) in the variable k2 . There we need

k′ 2 k′ 2
t(k′ − ε + iν1 , k′ + iǫ; + iǫ) − t(k′ + ε + iν1 , k′ + iǫ; + iǫ) , (11.1)
m m
with k′ > 0, which is what we need for the calculation of the potential (that we assume to be the case).
A ±iǫ is needed in Eq. (11.1) in order to give sense to the integral along the original integration contour that runs
along the RHC. It is not necessary to keep it in the first argument of t(k, k′ ) because |ν1 | > mπ , cf. Eq. (10.12).
We first consider the case of positive ν1 (so that ν1 > mπ ). We will show below that the case with ν1 < −mπ can
be obtained directly from our results for positive ν1 . As in Sec. 7.1 we make use of the LS equation to evaluate the
discontinuity in Eq. (11.1). One has,
Z
′ k′ 2
′ m p21 ′ k
′2
t(k ± ε + iν1 , k ; ) = v(k′ ± ε + iν1 , k′ ) + 2 dp1 v(k ′
± ε + iν 1 , p 1 )t(p 1 , k ; ), (11.2)
m 2π C p21 − k′ 2 m

112
Figure 27: Deformed integration contour in the p1 -complex plane that is employed for the LS equations to calculate the
discontinuity of Eq. (11.1). The branch point is located at k′ ± ε + i(ν1 − mπ ). The deformation takes place only for
ν1 > m π .

where the contour C in the p1 -complex plane is plotted in Fig. 27, with the vertical addition extending from k′ ± ε up to
the branch point k′ ± ε + i(ν1 − mπ ).
Along the vertical part of the integration contour we change the integration variable

Imp1 = ν2 , (11.3)

so that

p1 = k′ ± ε − δ + iν2 ,
p1 = k′ ± ε + δ + iν2 , (11.4)

along the left and right vertical lines that run parallel to the vertical addition, in this order, with 0 < δ < ε and ε → 0+ .
In terms of ν2 the energy denominator in Eq. (11.2) becomes along the vertical addition

p21 − (k′ + iǫ)2 = (k′ ± ε + iν2 )2 − (k′ + iǫ)2 = −(ν2 + ǫ − i2k′ ∓ iε)(ν2 − ǫ ∓ iε) . (11.5)

Here we have not taking into account the ±δ along the left and right sides of the vertical addition, respectively, because
ε > δ, and then ε ± δ is equivalent to ε in the considerations done in the previous equation.

113
Now, the difference in Eq. (11.1) in terms of the LS equations, Eq. (11.2), can be expressed as

t(k′ − ε + iν1 + iǫ, k′ + iǫ) − t(k′ + ε + iν1 + iǫ, k′ + iǫ) = v(k′ − ε + iν1 , k′ + iǫ) − v(k′ + ε + iν1 , k′ + iǫ)
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2

2π 2 0 (ν2 − ǫ + iε)(ν2 + ǫ − 2ik′ + iε)
 
× v(k′ − ε + iν1 , k′ − ε − δ + iν2 ) − v(k′ − ε + iν1 , k′ − ε + δ + iν2 ) t(k′ − ε + iν2 , k′ + iǫ)
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2
+ 2
2π 0 (ν2 − ǫ − iε)(ν2 + ǫ − 2ik′ − iε)
 
× v(k′ + ε + iν1 , k′ + ε − δ + iν2 ) − v(k′ + ε + iν1 , k′ + ε + δ + iν2 ) t(k′ + ε + iν2 , k′ + iǫ)
Z
m ∞ dp1 p21  ′ ′ 
+ 2
− 2 2 v(k + iν 1 − ε, p 1 ) − v(k + iν 1 + ε, p 1 ) t(p1 , k′ + iǫ) . (11.6)
2π 0 p1 − k′ − iǫ

In the first two previous integrals we have extracted t(k′ ± ε + iν2 , k′ + iǫ)pas a common factor in the differences in δ of
potential functions because the nearest DCs in t(p1 , k′ ) are located at k′ ± i m2π + x2 , that are to the right of k′ − ε + iν2
(first integral), and to the left to k′ + ε + iν2 (second integral).
The the last term in Eq. (11.6), corresponding to the integral along the original integration contour, does not give
contribution to the discontinuity. The reason is because the integrand is not zero only around p1 = k′ , which is a region
already endorsed in the added integration contours, corresponding to the first two integrals in Eq. (11.6). Notice that the
basis of the vertical contour (the region between the real axis and below +iǫ) can be extended analytically to the left and
right from k′ ± ε and embrace p1 = k′ . We then have,
Z
m ∞ dp1 p21  
2
− 2 ′ 2 v(k′ + iν1 − ε, p1 ) − v(k′ + iν1 + ε, p1 ) t(p1 , k′ + iǫ) = 0 . (11.7)
2π 0 p1 − k − iǫ

Regarding the integrations along the vertical additions in Eq. (11.6) we consider k′ 6= 0. The value k′ = 0 can be
studied as the limit k′ → 0+ , whose existence is shown below (the proof of its existence is given before Eq. (11.11)). We
can then write the first two integrals in Eq. (11.6) as
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2

2π 2 0 (ν2 − ǫ + iε)(ν2 + ǫ − 2ik′ + iε)
 
× v(k − ε + iν1 , k − ε − δ + iν2 ) − v(k′ − ε + iν1 , k′ − ε + δ + iν2 ) t(k′ − ε + iν2 , k′ + iǫ)
′ ′
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2
+
2π 2 0 (ν2 − ǫ − iε)(ν2 + ǫ − 2ik′ − iε)
 
× v(k + ε + iν1 , k + ε − δ + iν2 ) − v(k′ + ε + iν1 , k′ + ε + δ + iν2 ) t(k′ + ε + iν2 , k′ + iǫ) .
′ ′
(11.8)

Let us recall that ν1 > mπ and this is why we have neglected ǫ in front of ν1 , such that i(ν1 + iε) → iν1 . For a given
ǫ > 0 one has to take the limit ε → 0+ . We can then solve for the pole at ν2 = ǫ in Eq. (11.8) by rewriting the integrals
there as
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2

2π 2 0 (ν2 − ǫ)(ν2 − 2ik′ )
  
× v(k′ − δ + iν1 , k′ + iν2 ) − v(k′ + δ + iν1 , k′ + iν2 ) t(k′ − ε + iν2 , k′ + iǫ) − t(k′ + ε + iν2 , k′ + iǫ)
imk′ θ(ν1 − mπ )  ′ 
+ v(k − δ + iν1 , k′ + iǫ) − v(k′ + δ + iν1 , k′ + iǫ) t(k′ , k′ + iǫ) , (11.9)

where we have taken into account that the discontinuity v(k′ + iν1 + iǫ, k′ − δ + iν2 ) − v(k′ + iν1 + iǫ, k′ + δ + iν2 ) is
continuous in k′ , which is also the case for t(k′ ± ε, k′ + iǫ).

114
We split the integration interval in the first integral of Eq. (11.9) in two segments, ν2 ∈ [0, mπ ] and ν2 ∈ [mπ , ν1 − mπ ].
The point now is that the integration along the first segment vanishes because t(k′ −ε+iν2 , kp ′ +iǫ)−t(k ′ +ε+iν , k ′ +iǫ) = 0
2
+ ′ ′
for ν2 < mπ and ǫ → 0 . Recall that the DCs of t(z1 , k + iǫ) extend along ±(k + iǫ ± i m2π + x2 ).
As a result of Eqs. (11.7) and (11.9) we have the following neat contribution from Eq. (11.6),
t(k′ − ε + iν1 + iǫ, k′ + iǫ) − t(k′ + ε + iν1 + iǫ, k′ + iǫ)
 imk′ ′ ′  
= 1+ t(k , k + iǫ) v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ )

Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ 2 ′
v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π mπ ν2 (ν2 − 2ik )
 ′ 
× t(k − ε + iν2 , k + iǫ) − t(k′ + ε + iν2 , k′ + iǫ) .

(11.10)
Notice the step function in front of the integral taking into account the fact that t(k′ −ε+iν2 , k′ +iǫ)−t(k′ +ε+iν2 , k′ +iǫ) =
0 for 0 < ν2 < mπ . Due to the absence of infrared singularities in the previous integrals because ν2 ≥ mπ it is clear that
the limit k′ → 0+ exists. Multiplying Eq. (11.10) by D(k′ + iǫ) we directly obtain the discontinuity in ε for the function
N (k ± ε + ν1 , k′ ), which reads
N (k′ − ε + iν1 + iǫ, k′ + iǫ) − N (k′ + ε + iν1 + iǫ, k′ + iǫ)
  imk′ 
= v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ ) D(k′ + iǫ) + N (k′ , k′ + iǫ)

Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ 2 ′
v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π mπ ν2 (ν2 − 2ik )
 
× N (k − ε + iν2 , k + iǫ) − N (k′ + ε + iν2 , k′ + iǫ) .
′ ′
(11.11)
The independent term in the previous integral equation can be simplified by noticing that
imk′
ImD(k′ + iǫ) = − N (k′ , k′ + iǫ) . (11.12)

Thus, the combination
imk′
D(k′ + iǫ) + N (k′ , k′ + iǫ) = D(k′ + iǫ)∗ (11.13)

Taking into account this result and the fact that v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ ) in the independent term of
Eq. (11.11) is not zero only for ν1 > mπ , allow us to rewrite Eq. (11.11) as
 
N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) = v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ ) D(k′ + iǫ)∗
Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ 2 ′
v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π mπ ν2 (ν2 − 2ik )
 
× N (k − ε + iν2 , k + iǫ) − N (k′ + ε + iν2 , k′ + iǫ) ,
′ ′
(11.14)

where we have dropped +iǫ in front of ν1 and ν2 since both must be larger than mπ . Dividing the previous IE by D(k′ +iǫ)∗
we can eliminate the explicit presence of this factor in the IE. It then becomes
N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ)
= v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ )
D(k′ + iǫ)∗
Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ 2 ′
v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π mπ ν2 (ν2 − 2ik )
N (k′ − ε + iν2 , k′ + iǫ) − N (k′ + ε + iν2 , k′ + iǫ)
× , (11.15)
D(k′ + iǫ)∗

115
 
in terms of which we can solve for the function N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) /D(k′ + iǫ)∗ with
ν1 > mπ . Once this solved we multiply the result by D(k′ + iǫ)∗ and obtain the sought discontinuity of the half-off-shell
N in Eq. (11.14). In the numerical solution of Eq, (11.14), which is a Volterra IE of the 2nd type, one can proceed
similarly as discussed above regarding Eq. (7.65). For nmπ < ν1 < (n + 1)mπ only those values of the function with
mπ < ν2 < nmπ enter in the integral, which are already known if one proceed step by step by amounts of mπ , because
for mπ < ν1 < 2mπ the solution is entirely given by the independent term.
Comments for dummies:
The basic steps are:
p
1. First we need the discontinuity of N (k, k′ + iǫ; k′ 2 /m + iǫ) which occurs for k′ ± i m2π + x2 = k′ + iν1
2. Since ν1 > mπ then we have to deform the integration contour from 0 up to k′ + i(ν1 − mπ )
3. The subtle point in the treatment of the vertical additions is the energy pole.
4. Since for k′ − ε and k′ + ε we have crossed the discontinuity associated with v(k′ + iν1 , p1 ) there is a discontinuity
from the zero contribution in the denominator.
5. Since we are circumventing the vertical additions the angular path is π and not 2π, so that we have the standard
rule for the ±iπδ(p12 − k′ 2 )
6. Notice that the zero in the denominator is located with a fixed real part at k′ and from this location the sense of the
way in which the vertical cuts are circumvented looks different whether the vertical addition is to the left or right of
k′ . This is why we have an addition of those contributions and not a cancellation.
We also note that in Eq. (11.14) the only contribution to the difference v(k′ + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 ),
as well as that in the independent term, stems from the imaginary part of the log term present in the potential function
v(k, k′ ). However, the coefficient in front of this log term is typically complex in this case and we cannot write simply the
difference between the imaginary parts of the factors v involved, as we have done in the on-shell case. Evaluating explicitly
the difference of the log’s appearing in v(k′ − δ + iν1 , k′ + iν2 ) − v(k′ + δ + iν1 , k′ + iν2 ) for the simplest (and basic) case
of OPE, one has
log((2k′ + i(ν1 + ν2 ))2 + m2π ) − log((i(ν1 − ν2 ) − δ)2 + m2π ) − log((2k′ + i(ν1 + ν2 ))2 + m2π )
+ log((i(ν1 − ν2 ) + δ)2 + m2π ) = log(−(ν1 − ν2 )2 + 2iδ(ν1 − ν2 ) + m2π ) − log(−(ν1 − ν2 )2 − 2iδ(ν1 − ν2 ) + m2π )
= +i2π . (11.16)
The last step follows because ν1 − ν2 > mπ . Proceeding analogously for the independent term it results:
v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ ) → log(−ν12 + 2iεν1 + m2π ) − log(−ν12 − 2iεν1 + m2π )
= +i2π , for ν1 > mπ . (11.17)
In order to evaluate N (k, k′ )from the DR in the variable k of Eq. (10.56) one still needs to evaluate the discontinuity
with −iǫ, namely, N (k − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ). We can follow the same procedure as for the evaluation

of N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) exchanging the sign of ǫ in Eqs. (11.6,11.7,11.8). However, there is
an important difference for the last equation, which now reads
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2

2π 2 0 (ν2 + ǫ + iε)(ν2 − ǫ − 2ik′ + iε)
 
× v(k′ − ε + iν1 , k′ − ε − δ + iν2 ) − v(k′ − ε + iν1 , k′ − ε + δ + iν2 ) t(k′ − ε + iν2 , k′ − iǫ)
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k′ + iν2 )2
+
2π 2 0 (ν2 + ǫ − iε)(ν2 − ǫ − 2ik′ − iε)
 
× v(k′ + ε + iν1 , k′ + ε − δ + iν2 ) − v(k′ + ε + iν1 , k′ + ε + δ + iν2 ) t(k′ + ε + iν2 , k′ − iǫ) . (11.18)

116
The point is that now there is no pole contribution because ν2 + ǫ does not vanish, since ǫ > 0. Then we can directly
combine the two integrations in this equation and obtain the analogous expression to Eq. (11.10) for this case, that reads

t(k′ − ε + iν1 , k′ − iǫ) − t(k′ + ε + iν1 , k′ − iǫ)


 
= v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ )
Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ 2 ′
v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π mπ ν2 (ν2 − 2ik )
 ′ 
× t(k − ε + iν2 , k − iǫ) − t(k′ + ε + iν2 , k′ − iǫ) .

(11.19)

Multiplying the previous equation by D(k′ − iǫ) we directly obtain the discontinuity N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε +
iν1 , k′ − iǫ). Furthermore, noticing that D(k′ − iǫ) = D(k′ + iǫ)∗ , the corresponding expression is the same as that in
Eq. (11.14). Explicitly, we deduce
 
N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ) = v(k′ − ε + iν1 , k′ ) − v(k′ + ε + iν1 , k′ ) D(k′ + iǫ)∗
Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k′ + iν2 )2  ′ 
+ v(k + iν1 − δ, k′ + iν2 ) − v(k′ + iν1 + δ, k′ + iν2 )
2π 2 mπ ν (ν
2 2 − 2ik ′)
 
× N (k − ε + iν2 , k − iǫ) − N (k′ + ε + iν2 , k′ − iǫ) ,
′ ′
(11.20)

Then, we can conclude from the equality of the previous IE with that in Eq. (11.14) that

N (k′ − ε + iν1 , k′ + iǫ) − N (k′ + ε + iν1 , k′ + iǫ) = N (k′ − ε + iν1 , k′ − iǫ) − N (k′ + ε + iν1 , k′ − iǫ) (11.21)

for ν1 ≥ mπ .
• In order to calculate bj (k′ ) in Eq. (10.27) we need to know the function f (k, k′ , k0 ), Eq. (10.22). For that, in addition
to the discontinuity just given by Eqs. (11.14) and (11.20) we also need the discontinuity

t(k, k + iν1 − ε) − t(k, k + iν1 + ε) , (11.22)

that we calculate for k > 0 (since this is the range needed to calculate the resulting potential). The previous discontinuity
originally appeared in the complex k′ -plane DR for the function N (k, k′ ; k′ 2 /m) of Eq. (10.51). We also consider only
ν1 > 0 because the discontinuity for ν1 < 0 is the complex conjugate one of the former, cf. Eq. (10.50). We proceed here
in terms of the LS equation
Z
m p21
t(k, k ± ε + iν1 ) = v(k, k ± ε + iν1 ) + dp1 v(k, p1 )t(p1 , k ± ε + iν1 ) , (11.23)
2π 2 C p21 − (k ± ε + iν1 )2

where the integration contour C is the same as depicted in Fig. 27 but replacing k′ by k, so that now the branch point is
located at k ± ε + i(ν1 − mπ ). The unperturbed part of the integration contour does not contribute to the discontinuity
in Eq. (11.22) (the energy denominator does not vanish), which from Eq. (11.23) reads

t(k, k − ε + iν1 ) − t(k, k + ε + iν1 ) = v(k, k − ε + iν1 ) − v(k, k + ε + iν1 )


Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k + iν2 )2
+ v(k, k − ε + iν2 )
2π 2 0 (ν1 − ν2 )(ν1 + ν2 − 2ik)
 
× t(k − ε − δ + iν2 , k − ε + iν1 ) − t(k − ε + δ + iν2 , k − ε + iν1 )
Z
imθ(ν1 − mπ ) ν1 −mπ dν2 (k + iν2 )2
− v(k, k + ε + iν2 )
2π 2 0 (ν1 − ν2 )(ν1 + ν2 − 2ik)
 
× t(k + ε − δ + iν2 , k + ε + iν1 ) − t(k + ε + δ + iν2 , k + ε + iν1 ) , (11.24)

117
Figure 28: Deformed integration contour in the p1 -complex plane that is employed for the LS equations to calculate the
discontinuity of Eq. (11.26). The deformation, as shown, takes place for ν1 > mπ and ν2 > mπ (0 < ν2 < ν1 − mπ ).

where we have extracted the function v as a common factor in the discontinuities within the integrand, in the same way
as we extracted t in Eq. (11.6). The discontinuity in t is itself continuous in ε, then we can write

t(k, k − ε + iν1 ) − t(k, k + ε + iν1 ) = v(k, k − ε + iν1 ) − v(k, k + ε + iν1 )


Z
imθ(ν1 − 2mπ ) ν1 −mπ dν2 (k + iν2 )2  
+ 2
v(k, k − ε + iν2 ) − v(k, k + ε + iν2 )
2π mπ (ν1 − ν2 )(ν1 + ν2 − 2ik)
 
× t(k − δ + iν2 , k + iν1 ) − t(k + δ + iν2 , k + iν1 ) . (11.25)

Here, we have taken into account that there is only contribution from the integral term for ν1 > 2mπ , because for mπ > ν2
one has that v(k, k − ε + iν2 ) − v(k, k + ε + iν2 ) = 0.
In order to calculate t(k, k − ε + iν1 ) − t(k, k + ε + iν1 ) from Eq. (11.25) we need to know additionally

t(k − δ + iν2 , k + iν1 ) − t(k + δ + iν2 , k + iν1 ) , (11.26)

for k > 0, ν1 > 2mπ and ν1 − mπ > ν2 > mπ . The LS equation for t(k − δ + iν2 , k + iν1 ) involves the deformed contour
plotted in Fig. 28, where 0 < δ < ε, with the analogous one for t(k + δ + iν2 , k + iν1 ). We have the following LS equation
Z
imθ(ν2 − mπ ) ν2 −mπ dν3 (k + iν3 )2
t(k + iν2 ± ε, k + iν1 ) = v(k + iν2 ± ε, k + iν1 ) +
2π 2 0 (ν1 − ν3 )(ν1 + ν3 − 2ik)
 
× v(k + iν2 ± ε, k + iν3 ± ε − δ) − v(k + iν2 ± ε, k + iν3 ± ε + δ) t(k + iν3 ± ε, k + iν1 )
Z
imθ(ν1 − mπ ) ν1 −mπ dν3 (k + iν3 )2 
+ 2
v(k + iν2 ± ε, k + iν3 ) t(k + iν3 − δ, k + iν1 )
2π 0 (ν1 − ν3 )(ν1 + ν3 − 2ik)

− t(k + iν3 + δ, k + iν1 ) + . . . , (11.27)

118
where the ellipses refer to the contribution from the non-deformed part of the integration contour in Fig. 28, that do not
contribute to the discontinuity. In the previous equation the functions t and v are extracted as common factors in the
differences within the integrand of the first and second integrals, respectively. We then have for the difference of functions
t in the integrand of Eq. (11.25),

t(k + iν2 − ε, k + iν1 ) − t(k + iν2 + ε, k + iν1 ) = v(k + iν2 − ε, k + iν1 ) − v(k + iν2 + ε, k + iν1 )
Z
imθ(ν2 − mπ ) ν2 −mπ dν3 (k + iν3 )2  
+ 2
v(k + iν2 + δ, k + iν3 ) − v(k + iν2 − δ, k + iν3 )
2π 0 (ν1 − ν3 )(ν1 + ν3 − 2ik)
 
× t(k + iν3 − ε, k + iν1 ) − t(k + iν3 + ε, k + iν1 )
Z
imθ(ν1 − mπ ) ν1 −mπ dν3 (k + iν3 )2  
+ 2
v(k + iν2 − ε, k + iν3 ) − v(k + iν2 + ε, k + iν3 )
2π 0 (ν1 − ν3 )(ν1 + ν3 − 2ik)
 
× t(k + iν3 − δ, k + iν1 ) − t(k + iν3 + δ, k + iν1 ) . (11.28)

There is a partial cancellation between both terms in the previous equation. We then arrive to the final expression

t(k + iν2 − ε, k + iν1 ) − t(k + iν2 + ε, k + iν1 ) = v(k + iν2 − ε, k + iν1 ) − v(k + iν2 + ε, k + iν1 )
Z
imθ(ν1 − mπ ) ν1 −mπ dν3 (k + iν3 )2 θ(ν3 − (ν2 − mπ ))  
+ 2
v(k + iν2 − δ, k + iν3 ) − v(k + iν2 + δ, k + iν3 )
2π ν2 −mπ (ν1 − ν3 )(ν1 + ν3 − 2ik)
 
× t(k + iν3 − ε, k + iν1 ) − t(k + iν3 + ε, k + iν1 ) . (11.29)

One can easily see that no infrared divergences arise from this equation when taking k → 0 and ν3 → 0, the latter would
occur for ν2 → mπ . This is due to the fact that a nonzero value for the difference v(k+iν2 −δ, k+iν3 )−v(k+iν2 +δ, k+iν3 )
stems only from log(−(ν3 − ν2 )2 + m2π + 2iδ(ν3 − ν2 )) − log(−(ν3 − ν2 )2 + m2π − 2iδ(ν3 − ν2 )). This difference does not
vanish only when ν3 > mπ + ν2 and ν3 < ν2 − mπ . However, the latter option is excluded because of the step function
θ(ν3 − (ν2 − mπ )) and, as a result, ν3 > mπ + ν2 . In this way, we can rewrite Eq. (11.29) in simpler form as

t(k + iν2 − ε, k + iν1 ) − t(k + iν2 + ε, k + iν1 ) = v(k + iν2 − ε, k + iν1 ) − v(k + iν2 + ε, k + iν1 )
Z
imθ(ν1 − 2mπ − ν2 ) ν1 −mπ dν3 (k + iν3 )2  
+ 2
v(k + iν2 − δ, k + iν3 ) − v(k + iν2 + δ, k + iν3 )
2π mπ +ν2 (ν1 − ν3 )(ν1 + ν3 − 2ik)
 
× t(k + iν3 − ε, k + iν1 ) − t(k + iν3 + ε, k + iν1 ) . (11.30)

The solution of Eq. (11.30) is used in Eq. (11.25). However, it is more straightforward to realize that we can also apply
Eq. (11.30) for ν2 = 0 and its solution provides us with

t(k − ε, k + iν1 ) − t(k + ε, k + iν1 ) = t(k, k + ε + iν1 ) − t(k, k − ε + iν1 ) , (11.31)

which is minus the sought solution of Eq. (11.25). We have then all the necessary functions to calculate f (k, k′ , k0 ),
Eq. (10.22).
Eq. (11.30) for k → ε+ and ν1 → k coincides with the IE in Eq. (7.120) for calculating the on-shell discontinuity by
extending ν2 up to −k, that is, including negative values. Thus, as a by product of solving Eq. (11.30) for k → 0+ and
ν2 ∈ [−k, k − mπ ] is the calculation of the on-shell discontinuity.

119
12 Different types of DRs within the N/D method
We use the notation (n, m) to designate n-times subtracted DR in the function N (A) and m-times subtracted DR for the
function D(A).

12.1 N/D11 DRs, ℓ = 0: Fixing the subtraction constant in terms of as


The DRs for the functions N (A) and D(A) in the N/D11 case are given by
Z
4πa A L ∆(k2 )D(k2 )
N (A) = − + dk2 2 2 , (12.1)
m π −∞ k (k − A)
√ Z 2 2
imA L 2 ∆(k )D(k )
D(A) = 1 + ias A + dk √ √ .
4π 2 −∞ k2 ( k2 + A)

The N/D11 IE along the LHC reads:


√ Z
mA L ∆(k2 )D(k2 )
D(A) = 1 − as −A + 2 dk2 √ √ . (12.2)
4π −∞ k2 ( −k2 + −A)

12.2 N/D12 DRs, ℓ = 0: Fixing the subtraction constants in terms of as and rs


We consider a (1,2) DR and fix ν1 and δ2 so as to reproduce the S-wave scattering length as and effective range rs . We
have the following DRs:
Z
∆(k2 )D(k2 )
A L
N (A) = ν1 + , dk2
−∞ k2 (k2 − A)
π
Z
A(A − C) ∞ 2 ρ(q 2 )N (q 2 )
D(A) = 1 + δ2 A − dq 2 2 , (12.3)
π 0 q (q − C)(q 2 − A)

where C < 0 and the phase space factor ρ(A) is



m A
ρ(A) = . (12.4)

As usual, we substitute the DR for N (A) into the one for D(A) and end with the IE
Z
A(A − C) ∞ 2 ρ(q 2 )
D(A) = 1 + δ2 A − ν1 dq 2 2
π 0 q (q − C)(q 2 − A)
Z L 2 2 Z ∞
A(A − C) 2 ∆(k )D(k ) 2 ρ(q 2 )
+ dk dq . (12.5)
π2 −∞ k2 0 (q 2 − C)(q 2 − k2 )(q 2 − A)

The last integration along the RHC, that we denote as h(A, C, k2 ), can be calculated algebraically with the result
Z h i
1 ∞ ρ(q 2 ) 1
h(A, C, k2 ) = dq 2 = g(A, C) − g(A, k 2
) , (12.6)
π 0 (q 2 − C)(q 2 − k2 )(q 2 − A) C − k2

with g(A, B) the basic unitarity integral along the RHC


Z
1 ∞ ρ(q 2 ) im/4π
g(A, B) = dq 2 2 2
=√ √ , (12.7)
π 0 (q − A)(q − B) A + i0+ + B + i0+

120

with the addition of the positive infinitesimal 0+ , which is relevant for negative A or B (with −1 ± i0+ = ±i). Note
also that the integration along the RHC in the first line of Eq. (12.5) corresponds to h(A, C, 0).
Now, let us impose that the solution reproduces given values of as and rs . In our normalization the effective range
expansion (ERE) reads

N (A) 4π 1
T (A) = = √ √ . (12.8)
D(A) m A cot δ − i A

Then,
m √
N (A) A cot δ = D(A) + iρ(A)N (A) . (12.9)

The rhs of the previous equation can be conveniently expressed by writing D(A) as in Eq. (12.5) and N (A) as in Eq. (12.3),
so that
" Z #
A(A − C) ∞
2 ρ(q 2 )
D(A) + iρ(A)N (A) = 1 + δ2 A − ν1 dq 2 2 − iρ(A)
π 0 q (q − C)(q 2 − A)
Z " Z #
1 L
2 ∆(k2 )D(k2 ) A(A − C) ∞
2 ρ(q 2 ) A
+ dk dq 2 + iρ(A) 2 .
π −∞ k2 π 0
2 2 2
(q − C)(q − k )(q − A) k −A
(12.10)

In order to fix as and rs one has to match both sides of Eq. (12.9) up to O(A). In the integrand along the RHC in the
2nd line of Eq. (12.10) we can directly set A = 0 up to the order considered. However, this cannot be done directly in the
integrand of the 1st line because an infrared divergence would arise then. Nonetheless, from the exact algebraic expression
of this integral, h(A, C, 0), we can straightforwardly work out the linear behavior in A of the sum of the two terms between
square brackets in the 1st line of Eq. (12.10). We can then express Eq. (12.10) as
Z
mν1 CA L ∆(k2 )D(k2 )
D(A) + iρ(A)N (A) = 1 + δ2 A + A p − dk2 h(0, C, k2 ) + O(A2 ) . (12.11)
4π |C| π −∞ k2

The S-wave ERE up to O(A) reads


√ 1 1
A cot δ = − + rs A + O(A2 ) (12.12)
as 2
As a result, the lhs of Eq. (12.9) becomes
Z ! 
m A L 2 )D(k 2 )
2 ∆(k 1 1
ν1 + dk − + rs A + O(A2 )
4π π −∞ (k2 )2 as 2
Z !
m ν1 1 A L ∆(k2 )D(k2 )
= − + ν1 rs A − dk2 + O(A2 ) . (12.13)
4π as 2 as π −∞ (k2 )2

The equality of Eqs. (12.11) and (12.13) implies that


4πas
ν1 = − , (12.14)
m
Z  
as 1 1 L ∆(k2 )D(k2 ) m
δ2 = p − as rs + dk2 C h(0, C, k 2
) − . (12.15)
|C| 2 π −∞ k2 4πas k2

121
Substituting these expressions for ν1 and δ2 in Eq. (12.5) we have the following IE
!
as 1 4πas
D(A) = 1 + p − as rs A + A(A − C)h(A, C, 0)
|C| 2 m
Z L 2 )D(k 2 )  
A 2 ∆(k 2 m 2
+ dk (A − C)h(A, C, k ) + C h(0, C, k ) − . (12.16)
π −∞ k2 4πas k2

Taking into account the explicit expressions for h(A, B, C) and g(A, B), cf. Eqs. (12.6) and (12.7), respectively, one can
further simplify Eq. (12.16) as
Z " √ #
1 √ imA L 2 2
2 ∆(k )D(k ) A i
D(A) = 1 − as rs A + ias A − dk √ √ √ − , (12.17)
2 4π 2 −∞ k2 ( A + k2 ) k2 as k2

which is explicitly independent of the subtraction point C.


Asymptotic behavior of the IE Eq. (12.17) We follow an analogous reasoning to that in Sec. 3.1 of Ref. [3] and write
the asymptotic behavior of ∆(A) for A → −∞ as

λ
∆(A) −−−−−→ . (12.18)
A→−∞ (−A)α

The asymptotic behavior of ∆(A) for the 1 S0 corresponds to α = 1, cf. Eq. (14.11).
Next, we rewrite the IE of Eq. (12.17) in terms of the dimensionless variables

x = L/k2 ,
y = L/A , (12.19)

and

L = −m2π /4 . (12.20)

We obtain
Z " #
Las rs −1 q mλ|L| 2 −α
1
1 1 1
D(y) = 1 − y − as |L|y −1/2 − α −1
dx D(x)x y √ √ − p . (12.21)
2 4π 2 0 x + y as |L|

We now introduce the related function D̂(y) defined as

D̂(y) = y ν D(y) . (12.22)

In terms of it Eq. (12.21) reads


q 1 Z " #
1 1 mλ|L| 2 −α 1 1 1
ν
D̂(y) = y − as rs Ly ν−1 − as |L|y ν− 2 − dx D̂(x)x α−ν −1+ν
y √ √ − p . (12.23)
2 4π 2 0 x + y as |L|

For α = 1 and 1/2 < ν < 1 both the independent term and the kernel are square integrable functions. Then, according
to the Fredholm theorem the solution exists and is unique.

122
12.3 N/D22 DRs, ℓ = 0: Fixing the subtraction constants in terms of as , rs and v2
Let us deduce the expressions for the DRs corresponding to a N/D22 IE. As a first step we have
Z
A2 L ∆(k2 )D(k2 )
N (A) = ν1 + ν2 A + dk2 2 2 2 , (12.24)
π −∞ (k ) (k − A)
Z
A(A − C) ∞ 2 ρ(q 2 )N (q 2 )
D(A) = 1 + δ2 A − dq 2 2
π 0 q (q − C)(q 2 − A)
√ √ √
im A( A − C) imA √ √
= 1 + δ2 A + ν1 √ − ν2 ( A − C)
4π C 4π
Z L √ √ √ √ √
A(A − C)im 2 2 A( C + k2 ) + k2 C
2 ∆(k )D(k ) √ √
+ dk √ √ √ √ .
4π 2 −∞ (k2 )2 ( C + k2 )( C + A)( A + k2 )
Now we have to match order by order with the ERE up to and including O(A2 ), cf. Eqs. (12.8) and (12.9):
4π D(A) √ 1 1
+ i A = − + rA + v2 A2 + O(A3 ) . (12.25)
m N (A) a 2
The expansion is easier if multiplying the previous equation by N (A),
√  
4π 1 1 2 3
D(A) + i AN (A) = N (A) − + rA + v2 A + O(A ) . (12.26)
m a 2
Once this is done we obtain the following expressions for ν1 , ν2 and δ2 in terms of the ERE parameters a, r and v2 :
4πa
ν1 = − , (12.27)
m
Z  
8πav2 2 L ∆(k2 )D(k2 ) i 1
ν2 = + dk2 √ + ,
mr πr −∞ (k2 )2 k2 ak2
  √  
2v2 1 i r
δ2 = a − +i C + √ −
r a C 2
Z L 2 2    √ 
m ∆(k )D(k ) iC 2 i 1 1
+ 2 dk2 √ √ − √ + i C + .
4π −∞ (k2 )2 C + k2 r k2 ak2 a
Taking these expressions to N (A) and D(A) in Eq. (12.24) we have the following DRs without any C dependence:
Z  
2v2 A A L ∆(k2 )D(k2 ) A 2i 2
N (A) = ν1 (1 − )+ dk2 2 2
+ √ + , (12.28)
r π −∞ (k ) k − A r k2 ark2
2
√ !
2v2 A √ Ar 2v2 A A
D(A) = 1 − +a i A− −i
r 2 r
Z L ( √ ! √ √ )
mA 2
2 ∆(k )D(k )
2 √ i A 2 2i( A + k2 ) 2
+ 2 dk A √ √ + √ − − 2 2 .
4π −∞ (k2 )2 A + k2 r k2 ark2 ra k
In the last expression we have grouped terms according to the power of a. We can also rewrite it in another more compact
form
 
2v2 A  √  arA
D(A) = 1 − 1 + ia A − (12.29)
r 2
Z ( )
imA L 2
2 ∆(k )D(k )
2 Ak2 2i  √  √ 
+ dk √ √ + 1 + ia A 1 + ia k2 .
4π 2 −∞ (k2 )3 A + k2 ra2

123
This last expression gives rise to the following IE along the LHC:
 
2v2 A  √  arA
D(A) = 1 − 1 − a −A − (12.30)
r 2
Z L ( )
mA 2
2 ∆(k )D(k )
2 Ak2 2  √  p 
+ 2 dk √ √ − 1 − a −A 1 − a −k2 .
4π −∞ (k2 )3 −A + −k2 ra2

12.4 N/D12 DRs, ℓ = 1: Fixing the subtraction constant δ2 in terms of aV


We change the usual normalization by dividing simultaneously N (A) and D(A) by δ2 so that now D(0) = 1/δ2 . The (1,2)
DR for these functions read
Z
A L ∆(k2 )D(k2 )
N (A) = dk2 2 2 ,
π −∞ k (k − A)
Z Z
1 A2 L ∆(k2 )D(k2 ) ∞ 2 ρ(q 2 )
D(A) = +A+ 2 dk2 dq
δ2 π −∞ k2 0 q 2 (q 2 − k2 )(q 2 − A)
Z
1 imA2 L 2 2
2 ∆(k )D(k ) 1
= +A− 2
dk 2
√ √ √ √ . (12.31)
δ2 4π −∞ k ( A + k2 ) A k2
Let us impose the scattering volume, aV , which implies:
Z
δ2 L ∆(k2 )D(k2 )
−aV = dk2 , (12.32)
π −∞ (k2 )2
so that
Z
1 L ∆(k2 )D(k2 )
δ2−1 = − dk2 . (12.33)
aV π −∞ (k2 )2
We substitute this expression in Eq. (12.31) and it results
Z " #
1 L 2 )D(k 2 ) imA2
2 ∆(k 1 1
D(A) = A − dk √ √ √ √ + . (12.34)
π −∞ k2 4π ( A + k2 ) k2 A aV k2
• This IE has been used in the case of a repulsive singular potential for a P wave (with −∆(A)3 P0 from OPE) and it
is not useful to impose aV because N (0) = 0 after solving the (0,1) DR. We come back again to the original solution.
Indeed what happens is that the integral on the rhs of Eq. (12.33) vanishes for the solution obtained and then 1/δ2 = 0
and it just makes nothing in the IE of Eq. (12.31). Then, dividing simultaneously N (A) and D(A) by A we end up with
the standard N/D01 solution without any free parameter.
Indeed, if we renormalized again D(A) in the standard way, with D(0) = 1, by multiplying N (A) and D(A) in
Eq. (12.31) by δ2 , we would have a standard (1,2) DR with ν1 = N (0) = 0. In such a case, imposing aV gives rise to the
implicit equation for δ2 ,
Z
1 L ∆(k2 )D(k2 )
aV = − dk2 . (12.35)
π −∞ (k2 )2
We can then rewrite N (A) as
Z Z Z
A ∆(k2 )D(k2 ) A L
L 2
2 ∆(k )D(k )
2 A L ∆(k2 )D(k2 )
N (A) = dk2
2 (k 2 − A)
+ dk 2 )2
− dk2
π−∞ k π −∞ (k π −∞ (k2 )2
2 Z L 2 2
A ∆(k )D(k )
= −aV A + dk2 2 2 2 . (12.36)
π −∞ (k ) (k − A)

124
Then, by imposing the scattering volume we have indeed included an extra subtraction in N (A).
The fact that the implicit equation for δ2 , Eq. (12.35), is not operative when N (0) = 0, can be seen by noticing first
that this implicit equation for δ2 also arises for the case in which N (0) 6= 0. Here, we apply the (0,1)+CDD DR,
Z Z
γ A L ∞ ρ(q 2 )
D(A) = 1 + + dk2 ∆(k2 )D(k2 ) dq 2 . (12.37)
A π −∞ 0 q 2 (q 2 − A)(q 2 − k2 )
Imposing a given value for aV we then have for γ
Z
1 L ∆(k2 )D(k2 )
γ=− dk2 . (12.38)
aV π −∞ k2
In the case that N (0) = 0 the rhs of this equation vanishes and we do not have any contribution from the CDD. Now,
we transform the (0,1)+CDD into a (1,2) by multiplying N (A) and D(A) by A/γ. We denote by D2 (A) the resulting
function,
Z Z
A A2 L
2 ∆(k
2 )D (k 2 )
2
∞ ρ(q 2 )
D2 (A) = 1 + + dk 2
dq 2 . (12.39)
γ π −∞ k 0 q 2 (q 2 − k2 )(q 2 − A)
We end again with the implicit equation for δ2 , Eq. (12.35), after imposing aV .34 But for N (0) = 0 it should not be
operative because this procedure is equivalent to include a CDD in the (0,1), which we have just shown that drives to null
γ, cf. Eq. (12.38).

12.5 N/D22 DRs, ℓ = 1: Fixing the subtraction constants ν2 and δ2 in terms of aV and rV
• The ℓ = 1 N/D22 DRs:
Z
A2 L ∆(k2 )D(k2 )
N (A) = ν2 A + dk2 2 2 2 ,
π −∞ (k ) (k − A)
√ Z
imA A imA2 L 2 ∆(k2 )D(k2 )
D(A) = 1 + δ2 A − ν2 + dk √ √ . (12.40)
4π 4π 2 −∞ (k2 )2 ( k2 + A)
In order to match with the ERE we make use of Eq. (12.62) for ℓ = 1,
√  
4πA 1 1 2
D(A) + iN (A) A A = N (A) + rV A + O(A ) . (12.41)
m aV 2
Matching both sides of the previous equation up to and including O(A2 ):
4πaV
ν2 = ,
m
Z L
m ∆(k2 )D(k2 ) 1
δ2 = 2 dk2 + aV rV . (12.42)
4π aV −∞ (k2 )3 2
The following N/D22 DR for D(A) arises:
√ Z  
1 mA L ∆(k2 )D(k2 ) 1 iA
D(A) = 1 + aV rV A − iaV A A + 2 dk2 + √ √ . (12.43)
2 4π −∞ (k2 )2 aV k2 A + k2
This gives rise to the N/D22 IE for D(A) along the LHC (A < L):
Z  
1 √ mA L ∆(k2 )D(k2 ) 1 A
D(A) = 1 + aV rV A + aV A −A + 2 dk2 + √ √ . (12.44)
2 4π −∞ (k2 )2 aV k2 −A + −k2
34
The same equation is obtained by performing the transformation D(k2 ) → D2 (k2 )γ/k2 in Eq. (12.38).

125
12.6 N/D23 DRs, ℓ = 1: Fixing the subtraction constants ν2 , δ2 and δ3 in terms of aV , rV and ν2
• The ℓ = 1 N/D23 DRs:
Z
A2∆(k2 )D(k2 )
L
N (A) = ν2 A + dk2 ,
−∞ π(k2 )2 (k2 − A)
mν2 A2 (A − C)
D(A) = 1 + δ2 A + δ3 A2 + i √ √ √ √
4π A C( C + A)
Z
mA2 (A − C) L 2 ∆(k2 )D(k2 )
−i dk √ √ √ √ √ √ . (12.45)
4π 2 −∞ (k2 )2 ( C + A)( C + k2 )( k2 + A)
To avoid infrared singularities we have taken a subtraction at C < 0. To match with the ERE we make use of Eq. (12.41)
for ℓ = 1 matching both sides of the equation up to and including O(A3 ), we find the same expressions as given in
Eq. (12.42) for ν2 and δ2 . We also find now additionally the following result for δ3 :
√ Z 2 2 Z L 2 2
aV m C L 2 ∆(k )D(k ) √ 1 m 2 ∆(k )D(k )
δ3 = aV ν2 − i √ − i dk √ √ + dk
C 4π 2 −∞ (k2 )2 k2 ( k2 + C) 4π 2 aV −∞ (k2 )4
Z
mrV L ∆(k2 )D(k2 )
+ dk2 . (12.46)
8π 2 −∞ (k2 )3
Substituting the resulting expressions for ν2 , δ2 and δ3 into Eq. (12.45) we remove all the dependence of D(A) in C.
It results:
aV rV √
D(A) = 1 + A + aV ν2 A2 − iaV A A
2 √ !
Z L 2 )D(k 2 ) 2 ) + a r k2 A
Am ∆(k 2(A + k V V A A
+ 2 dk2 − i√ √ √ . (12.47)
4π −∞ (k2 )2 2aV (k2 )2 k2 ( k2 + A)
The IE for D(A) with A < L reads
aV rV √
D(A) = 1 + A + aV ν2 A2 + aV A −A
2 !
Z 2 2

Am L 2 ∆(k )D(k ) 2(A + k2 ) + aV rV k2 A A −A
+ 2 dk −√ √ √ . (12.48)
4π −∞ (k2 )2 2aV (k2 )2 −k2 ( −k2 + −A)

12.7 N/D13 DRs, ℓ = 1: Fixing δ1 and δ2 in terms of aV and rV


The DRs for the ℓ = 1 N/D13 case normalized such that D(0) = 1 are
Z
A L ∆(k2 )D(k2 )
N (A) = dk2 2 2 ,
π −∞ k (k − A)
D(A) = 1 + δ2 A + δ3 A2
Z √ √ √
mA2 (A − C) L
2 ∆(k
2 )D(k 2 ) C + k2 + A
+i dk √ √ √ √ √ √ √ √ √ . (12.49)
4π 2 −∞ k2 C k2 A( C + k2 )( C + A)( k2 + A)
The IE for D(A) along the LHC (A < L) is
D(A) = 1 + δ2 A + δ3 A2
Z √ √ √
mA2 (A − C) L
2 ∆(k
2 )D(k 2 ) −C + −k2 + −A
+ dk 1 √ √ √ √ √ √ . (12.50)
4π 2 −∞ k2 (−Ck2 A) 2 ( −C + −k2 )( −C + −A)( −k2 + −A)

126
However, there are problems when trying to fix δ2 and δ3 in Eq. (12.49) because from Eq. (12.41) we would have
Z
m L ∆(k2 )D(k2 )
aV = dk2 , (12.51)
4π 2 −∞ (k2 )2

and then aV cannot be fixed, but it is an output of the approach. This also implies another complication because if we fix
δ2 and δ3 from ERE up to O(A2 ) one finds
Z Z
m L ∆(k2 )D(k2 ) mrV L 2
2 ∆(k )D(k )
2
δ2 = 2 dk2 + dk ,
4π aV −∞ (k2 )3 8π 2 −∞ (k2 )2
Z L Z
m ∆(k2 )D(k2 ) mrV L 2
2 ∆(k )D(k )
2
δ3 = 2 dk2 + dk
4π aV −∞ (k2 )4 8π 2 −∞ (k2 )3
Z Z √ √
mν2 L 2 )D(k 2 ) L 2 )D(k 2 ) k2 + C + k2 C
2 ∆(k m 2 ∆(k
+ dk −i 2 dk √ √ √ √ . (12.52)
4π 2 −∞ (k2 )2 4π −∞ (k2 )2 C k2 ( C + k2 )

Inserting these expressions back into Eq. (12.49) we have:


Z   √ √ √ !
mA L 2 )D(k 2 ) A k2 + A + k2 A
2 ∆(k 1 rV 1 rV
D(A) = 1 + 2 dk + +A + + ν2 − i √ √ √ .
4π −∞ (k2 )2 aV k2 2 aV (k2 )2 2k2 k2 k2 + A
(12.53)

However, this IE is not a linear IE because aV is not an external input but given by Eq. (12.51). This is why it is not
suitable for calculating D(A).
We can change the normalization of D(A) and then fix aV and rV . For that we divide N (A) and D(A) in Eq. (12.49)
by δ3 . We can then write the new DR:

D(A) = δ1 + δ2 A + A2
Z √ √ √
mA2 (A − C) L
2 ∆(k
2 )D(k 2 ) C + k2 + A
+i dk √ √ √ √ √ √ √ √ √ . (12.54)
4π 2 −∞ k2 C k2 A( C + k2 )( C + A)( k2 + A)

Matching the ERE up to O(A) we have


Z
m L ∆(k2 )D(k2 )
δ1 = 2 dk2 ,
4π aV −∞ (k2 )2
Z Z !
m rV L 2 )D(k 2 ) L 2 )D(k 2 )
2 ∆(k 1 2 ∆(k
δ2 = 2 dk + dk . (12.55)
4π 2 −∞ (k2 )2 aV −∞ (k2 )3

Then, we have the following DR for D(A)


Z   √ √ √ !
m L 2 )D(k 2 ) k2 A2 (A − C)( C + k2 + A)
2 2 ∆(k 1 A rV A
D(A) = A + 2 dk 1+ 2 + + i√ √ √ √ √ √ √ √ √ .
4π −∞ (k2 )2 aV k 2 C k2 A( C + k2 )( C + A)( k2 + A)
(12.56)

This DR is explicitly dependent on the value of the subtraction point C, but the physical results that follow from it
are independent of C. This is because a change of the subtraction point would imply only a change in the coefficient
multiplying A2 , so that it is not longer 1 but some function of C. This can be more clearly seen if we rewrite the DR for
D(A) of Eq. (12.49) so that the RHC integration is explicitly kept, before giving its algebraic expression. However, we can

127
redefine the normalization of D(A) and N (A) and divide them simultaneously by this new coefficient multiplying A2 in
D(A), so that we end with the analogous expression as in Eq. (12.56).
The IE of Eq. (12.56) is the one to be explored in the practical applications. Along the LHC it reads
Z L 2 )D(k 2 )   
m2 1 2 ∆(k A rV A
D(A) = A + 2 dk 2 2
1+ 2 +
4π −∞ (k ) aV k 2
√ √ √ √ !
2 2
A(A − C) k A( −C + −k + −A)
+√ √ √ √ √ √ √ . (12.57)
−C( −C + −k2 )( −C + −A)( −k2 + −A)

To apply here the method based on the expansion in LP of Sec. 13 we express the previous equation as
Z 1
2
D(A) = A + dx∆(x)D(x)G(k2 , A) ,
0
  √ √ √ √ !
2 m 1 A rV A A(A − C) k2 A( −C + −k2 + −A)
G(k , A) = − 2 1+ 2 + +√ √ √ √ √ √ √ . (12.58)
4π L aV k 2 −C( −C + −k2 )( −C + −A)( −k2 + −A)

Along the RHC we have


Z   √ √ √ √ √ !
m 1 1 A rV A A A(A − C)( A + i −C + i −k2 ) −k2
2
D(A) = A − 2 dx∆(x)D(x) 1+ 2 + +√ √ √ √ √ √ √ .
4π L 0 aV k 2 −C( −C + −k2 )(i −C + A)(i −k2 + A)
(12.59)

The expression for N (A) is


Z
A 1 k2
N (A) = − dx∆(k2 )D(k2 ) . (12.60)
πL 0 k2 − A

12.8 Imposing ERE for a partial wave with orbital angular momentum ℓ

The ERE for arbitrary ℓ is an expansion of Aℓ A cot δ in powers of A that reads
√ 1 1
Aℓ A cot δ = − + rℓ A + . . . (12.61)
aℓ 2

Its connection with the N/D method is the same as in Eq. (12.8), but multiplying and dividing cot δ by Aℓ . Then we
have,
√  
4π 1 1 N (A)
D(A) + iN (A) A = − + rℓ A + . . . . (12.62)
m aℓ 2 Aℓ
Thus, if

N (A) −−−→ Am , (12.63)


A→0

with 0 ≤ m < ℓ, D(A) diverges as A−ℓ+m for A → 0, with a series of power divergences with smaller degree. This
behavior of N (A), not vanishing as fast as expected, seems to happen for attractive singular potentials to provide an
external input to end with an acceptable T (A), cf. discussion at the end of Sec. 12.11. The coefficient for each diverging
power of A contains a new shape parameter; A−ℓ+m involves −1/aℓ only, A−ℓ+m+1 additionally involves rℓ , A−ℓ+m+2
does so with v2 and so on until A−1 with vℓ−m−1 , cf. Table 1. A shape parameter appearing first at order A−ℓ+m+i also
contributes in the coefficients in front of A−ℓ+m+j with ℓ − m − 1 ≥ j ≥ i because of the expansion of N (A) in power of

128
A starting at Am . Of course, one can also account of terms with zero and positive powers of A by including a subtractive
polynomial, whose coefficients we typically denote by δi .

A−ℓ+m A−ℓ+m+1 A−ℓ+m+2 ... A−1


− a1ℓ rℓ v2 ... vℓ−m−1

Table 1: New shape parameter appearing in connection with a diverging power of A as a result of the ERE with N (A) −−−→
A→0
Am and ℓ > m ≥ 0.

12.9 (0,1)+m-CDD’s as (m,m+1) DR


We show first that a (0,1)+CDD DR is a particular (1,2) DR. A (0,1)+CDD DR, normalized such that D(0) = 1, reads
Z Z
γA A L ∞ ρ(q 2 )
D(A) = 1 + + 2 dk2 ∆(k2 )D(k2 ) dq 2 ,
A − B π −∞ 0 q 2 (q 2 − k2 )(q 2 − A)
Z
1 L ∆(k2 )D(k2 )
N (A) = dk2 . (12.64)
π −∞ k2 − A
Now we multiply both D(A) and N (A) by (B − A)/B and call the resulting functions D1 (A) and N1 (A), in order. It
follows then,
Z Z
(B − A)D(A) B−A γ A(A − B) L 2 2 2

2 ρ(q 2 )
D1 (A) = = − A− dk ∆(k )D(k ) dq
B B B Bπ 2 −∞ 0 q 2 (q 2 − k2 )(q 2 − A)
Z L 2 2 Z ∞ 2
A A(A − B) ∆(k )D1 (k ) ρ(q )
= 1 − (1 + γ) + 2
dk2 2
dq 2 2 2 ,
B π −∞ k −B 0 q (q − k2 )(q 2 − A)
Z
(B − A)N (A) A−B L ∆(k2 )D1 (k2 )
N1 (A) = = dk2 2 . (12.65)
B π −∞ (k − A)(k2 − B)
Eq. (12.65) is a particular case of (1,2) DR, in which ν1 = 0. E.g. just take N1 (A) and substitute it in the corresponding
twice-subtracted DR for D(A), that we call D2 (A),
Z Z
A(A − B) L ∆(k2 )D(k2 ) ∞ ρ(q 2 )(q 2 − B)
D2 (A) = 1 + Aδ2 + dk2 dq 2 , (12.66)
π2 −∞ k2 − B 0 q 2 (q 2 − B)(q 2 − A)
and we have the same expression as D1 (A) with δ2 = −(1 + γ)/B, cf. Eq. (12.65).
This process can be generalized straightforwardly to the addition of m CDDs,
m Z Z
X γi A A L ∞ ρ(q 2 )
D(A) = 1 + + 2 dk2 ∆(k2 )D(k2 ) dq 2 , (12.67)
i=1
A − Bi π −∞ 0 q 2 (q 2 − k2 )(q 2 − A)
with N (A) as in Eq. (12.64). Then, D1 (A) results after removing the m CDD poles, and it is given by
! Qm  
m
Y m 
Y  m
X m A 1− A
Bi − A A γi Y A i=1 Bi
D1 (A) = D(A) = 1− − A 1− +
i=1
Bi i=1
Bi i=1
Bi j6=i
Bj π2
Z Z
L
2 ∆(k2 )D1 (k2 ) ∞ ρ(q 2 )
× dk Q   dq 2
−∞ m
1− k2
0 q 2 (q 2 − k2 )(q 2 − A)
i=1 Bi
m   m m
! Qm Z Z
Y A X γi Y A A i=1 (A − Bi ) L ∆(k2 )D1 (k2 ) ∞ ρ(q 2 )
= 1− − A 1− + dk2 Qm dq 2 .
i=1
Bi i=1
Bi j6=i Bj π2 −∞
2
i=1 (k − Bi ) 0 q 2 (q 2 − k2 )(q 2 − A)
(12.68)

129
For N1 (A) we have
n   Qm Z
Y A i=1 (A − Bi ) L ∆(k2 )D1 (k2 )
N1 (A) = N (A) 1− = dk2 Q . (12.69)
i=1
Bi π −∞ (k2 − A) m 2
i=1 (k − Bi )

Eqs. (12.68) and (12.69) are particular cases of (m,m+1) DR for the N/D method.

12.10 Non-equivalence of (0,2) and (1,1) DRs


The (0,2)-case reads
Z Z
A2 L ∞ ρ(q 2 )
D(A) = 1 + δ2 A + 2 dk2 ∆(k2 )D(k2 ) dq 2 2 2 2 , (12.70)
π −∞ 0 (q ) (q − k2 )(q 2 − A)
Z
1 L ∆(k2 )D(k2 )
N (A) = dk2 (12.71)
π −∞ k2 − A
Z L Z
1 ∆(k2 )D(k2 ) A L 2
2 ∆(k )D(k )
2
= dk2 + dk .
π −∞ k2 π −∞ k2 (k2 − A)

This last form for N (A) is the corresponding one to a (1,1) DR. For D(A) we have still to perform some manipulations.
First, we write inside the last integral in Eq. (12.70),

k2 − q 2 + q 2
1= , (12.72)
k2
which drives to
Z Z Z Z
A2 L ∆(k2 )D(k2 ) ∞ ρ(q 2 ) A2 L ∆(k2 )D(k2 ) ∞ ρ(q 2 )
D(A) = 1 + Aδ2 − dk2 dq 2 + dk2 dq 2 .
π2 −∞ k2 0 (q 2 )2 (q 2 − A) π2 −∞ k2 0 q 2 (q 2 − k2 )(q 2 − A)
(12.73)

Next, we can remove A2 by writing A2 = A(A − q 2 + q 2 ),


( Z Z )
1 L
2 2 2

2 ρ(q 2 )
D(A) = 1 + A δ2 − 2 dk ∆(k )D(k ) dq 2 2 2 (12.74)
π −∞ 0 (q ) (q − k2 )
Z Z Z Z
A L ∆(k2 )D(k2 ) ∞ ρ(q 2 ) A L ∆(k2 )D(k2 ) ∞ ρ(q 2 )
− dk2 dq 2 + 2 dk2 dq 2 .
π2 −∞ k2 0
2 2
q (q − A) π −∞ k2 0 (q 2 − k2 )(q 2 − A)

From the last lines in Eqs. (12.71) and (12.74) it follows that the equivalence of the starting (0,2) and a resulting (1,1)
requires the identification
Z
1 L ∆(k2 )D(k2 )
ν1 = dk2 , (12.75)
π −∞ k2
Z Z ∞
1 L ρ(q 2 )
δ2 − 2 dk2 ∆(k2 )D(k2 ) dq 2 2 2 2 =0. (12.76)
π −∞ 0 (q ) (q − k2 )

While Eq. (12.75) can always be met because it is just a definition of ν1 , Eq. (12.76) is not expected to hold since δ2 is
adjusted to some experimental input and D(A) is obtained then by solving Eq. (12.70). In the solution for D(A) there
is not any extra constraint like that in Eq. (12.76), so as to end with an equivalent (1,1) DR. Furthermore, this latter
equation contains an infrared singularity because of the factor (q 2 )2 in the denominator of the integration along the RHC,

130
while δ2 is a finite number. However, this last reason can be avoided because the original Eq. (12.70) for D(A) is already
infrared divergent. One should have taken a subtraction point at C < 0 and avoid this divergence in an explicit maner.
Had we started with a (1,1) DR and followed similar steps as above we would have ended with the same conditions
as in Eqs. (12.75) and (12.76). But now Eq. (12.76) can be always met because it is just a definition of δ2 (which can
reabsorb the infrared divergence), while Eq. (12.75) is not expected to hold because ν1 is fixed to some experimental input
and the resulting solution D(A) is not required to fulfill any constraint. In particular, it is not required that ν1 be given by
the same expression as it were a prediction from a DR with less subtractions in N (A) (which indeed is not even required
to convergence). One can find in Ref. [3] explicit examples that show that ν1 fixed to some experimental number does
not correspond to Eq. (12.76).

12.11 (ℓ, ℓ) and (ℓ, ℓ + 1) DRs. Threshold behavior for singular potentials
• (ℓ, ℓ) DR with νi = 0, i = 1, . . . , ℓ. We study the IE for A ≤ L < 0. The case ℓ = 0 is excluded because it gives rise to
a homogeneous IE. The DRs read
Z
Aℓ L ∆(k2 )D(k2 )
N (A) = dk2 , (12.77)
π −∞ (k2 )ℓ (k2 − A)
ℓ Z Z
X Aℓ L ∆(k2 )D(k2 ) ∞ ρ(q 2 )
D(A) = 1 + δi Ai−1 + dk2 dq 2 . (12.78)
i=2
π2 −∞ (k2 )ℓ 0 (q 2 − k2 )(q 2 − A)

The sum in the last equation is present when ℓ ≥ 2. The integral along the RHC reads
Z
1 ∞ ρ(q 2 ) m/4π
dq 2 2 2 2
=√ √ (12.79)
π 0 (q − k )(q − A) −k2 + −A
We divide Eqs. (12.77) and (12.78) by Aℓ and define the new function
D(A)
D̂(A) = . (12.80)
Aℓ
We also introduce the dimensionless variables x and y defined by
L
k2 = ,
x
L
A= . (12.81)
y

Then D̂(A) fulfills the IE



X
1 Z 1
1
ℓ −ℓ i−1−ℓ ℓ+1−i m(−L) 2 ∆(x)D̂(x) (xy) 2
D̂(y) = y L + δi L y + dx √ √ . (12.82)
i=2
4π 2 0 x2 x+ y

Now, we consider the asymptotic behavior of ∆(x) as x → 0 in the form

∆(x) −−−→ λ(−k2 )α = λ(−L)α x−α . (12.83)


x→0

and replace it in Eq. (12.82) with the result


ℓ 1 Z 1
X mλ(−L) 2 +α 1 x−2−α (xy) 2
D̂(y) = y ℓ L−ℓ + δi Li−1−ℓ y ℓ+1−i + dxD̂(x) √ √ . (12.84)
i=2
4π 2 0 x+ y

131
We symmetrize the kernel of the IE by considering the IE for the function
α
e
D(y) = y −1− 2 D̂(y) . (12.85)

This new IE reads


ℓ 1 Z α+1
α X α mλ(−L)α+ 2 1 (xy)− 2
e
D(y) = y ℓ−1− 2 L−ℓ + δi Li−1−ℓ y ℓ−i− 2 + e
dxD(x) √ √ . (12.86)
i=2
4π 2 0 x+ y

We can remove from the IE the constant in front of the most divergent term in the independent term by introducing the
new function
Le
D̄(y) = D(y) , (12.87)
δℓ

with δ1 = 1. The IE for D̄(y) reads


ℓ−1 1 Z α+1
−α
X α mλ(−L)α+ 2 1 (xy)− 2
D̄(y) = y 2 + δi′ Li−ℓ y ℓ−i− 2 + dx √ √ D̄(x) . (12.88)
i=1
4π 2 0 x+ y

with δi′ = δi /δℓ . It is interesting to remark that the kernel of the IE is the same independently of ℓ.
In the case of ℓ = 1, namely for a (1, 1) DR, one can get the following strong result:
Strong result: For α > 1/2 the IE requires D̄(x) to vanish for x → 0 but this is not possible for λ > 0 because the
resolvent kernel is then positive definite (think e.g. in the Neumann series) and the independent term is also positive and
diverges for y → 0 with fixed positive sign. The only possibility for having a convergent IE is that λ < 0.
For ℓ ≥ 2 the situation is more involved because of the appearance of non-fixed coefficients δi′ that could have any sign.
We can also generalize Eq. (12.88) by further considering subtraction constants in the function N (A). Along these
lines let us first take the addition of a term νℓ Aℓ , its contribution to D(A) in Eq. (12.78) is
Z
Aℓ ∞ ρ(q 2 )
−νℓ dq 2 . (12.89)
π 0 q 2 (q 2 − A)

The new IE for D̄(y) is


ℓ−1 1 1 Z α+1
α X α νℓ − 1+α m(−L) 2 mλ(−L)α+ 2 1 (xy)− 2
D̄(y) = y − 2 + δi′ Li−ℓ y ℓ−i− 2 + y 2 + dx √ √ D̄(x) . (12.90)
i=1
δℓ 4π 4π 2 0 x+ y

In this case for ℓ = 1 we can conclude that this IE is not convergent for λ > 0 when ν1 ≥ 0. If ν1 < 0 it could be
convergent even if λ > 0. This is indeed what happens for the 3 P0 partial wave for which λ > 0 in the iterated-OPE case.
The (0,1) DR is convergent so that we can rewrite it as a (1,1) DR and it converges then for λ > 0 and ν1 < 0.
We can also include further subtractions in N (A), νi with i = 1, . . . , ℓ. If all of them are taken at A = 0 we have to
face the problem of infrared divergences for i = 1, . . . , ℓ − 1. These divergences are easily removed because they can be
reabsorbed in a redefinition of the δi′ . Let us take νℓ−i , i = 0, . . . , ℓ − 1, whose contribution to D(A) reads
Z
Aℓ ∞ ρ(q 2 )
−νℓ−i dq 2 , (12.91)
π 0 (q 2 )i+1 (q 2 − A)

132
and the infrared divergence for q 2 → 0 is manifest for i ≥ 1. We can reduce progressively the degree of the infrared
divergence in the term depending on A by repeatedly rewrite A → A − q 2 + q 2 :
Z Z
Aℓ−1 ∞ ρ(q 2 ) Aℓ−1 ∞ ρ(q 2 )
νℓ−i dq 2 − ν ℓ−i dq 2
π 0 (q 2 )i+1 π 0 (q 2 )i (q 2 − A)
i Z Z
X Aℓ−j ∞ ρ(q 2 ) Aℓ−i ∞ ρ(q 2 )
= −νℓ−i (−1)j dq 2 2 i+2−j
− νℓ−i dq 2 . (12.92)
j=1
π 0 (q ) π 0 q 2 (q 2
− A)

It is clear that the terms in the sum of the last line, although they are divergent, can be reabsorbed in the δi , cf. Eq. (12.78).
Note also that the last term, that has dependence in A, is finite and for A → −∞ diverges less strongly for increasing i,
as Aℓ−i−1/2 .
• (ℓ, ℓ + 1) DR with νi = 0, i = 1, . . . , ℓ. Here the case ℓ = 0 is allowed. We proceed similarly as for the (ℓ, ℓ) DR.
After dividing by Aℓ we have the DRs
Z
N (A) 1 L ∆(k2 )D̂(k2 )

= dk2 ,
A π −∞ k2 − A
ℓ+1 Z Z
1 X A L ∞ ρ(q 2 )
D̂(A) = ℓ
+ δi Ai−1−ℓ + 2 dk2 ∆(k2 )D̂(k2 ) dq 2 , (12.93)
A i=2
π −∞ 0 q 2 (q 2 − k2 )(q 2 − A)

with D̂(A) introduced in Eq. (12.80). The RHC integration in Eq. (12.93) for A, k2 ≤ L reads
Z √
A ∞ 2 ρ(q 2 ) m −A
dq 2 2 2 2
=− √ √ √ , (12.94)
π 0 q (q − k )(q − A) 4π −k ( −k2 + −A)
2

which is substituted in Eq. (12.93). We then have the following IE making use of the dimensionless variables x and y, cf.
Eq. (12.81),
ℓ+1 1 Z
ℓ −ℓ
X
i−1−ℓ ℓ+1−i m(−L) 2 1 x−1
D̂(y) = y L + δi L y − dx∆(x)D̂(x) √ √ . (12.95)
i=2
4π 2 0 x+ y

Now, we consider the asymptotic behavior of ∆(x) for x → 0 as in Eq. (12.83) and replace it in Eq. (12.95). The latter
equation becomes
ℓ+1 1 Z
ℓ −ℓ
X
i−1−ℓ ℓ+1−i mλ(−L) 2 +α 1 x−α−1
D̂(y) = y L + δi L y − dxD̂(x) √ √ . (12.96)
i=2
4π 2 0 x+ y

We symmetrize the IE by introducing the function


α+1
e
D(y) = y− 2 D̂(y) . (12.97)

Which fulfills the IE


ℓ+1 1 Z α+1
1+α X
i−1−ℓ ℓ+ 1−α mλ(−L) 2 +α 1 (xy)− 2 e
e
D(y) = y ℓ− 2 L−ℓ + δi L y2
−i
− dx √ √ D(x) . (12.98)
i=2
4π 2 0 x+ y

We can also remove δℓ+1 in the IE, since this is the coefficient in front of the term with highest divergence, by considering
the function

D̄(y) = D̂(y)/δℓ+1 . (12.99)

133
We have the IE
ℓ 1 Z 1+α
− α+1
X 1−α mλ(−L) 2 +α 1 (xy)− 2
D̄(y) = y 2 + δi′ Li−1−ℓ y ℓ+ 2 −i − dx √ √ D̄(x) , (12.100)
i=1
4π 2 0 x+ y

with δi′ = δi /δℓ+1 and δ1 = 1. Again the kernel of the IE is independent of the ℓ.
In the case of ℓ = 0, namely for a (0, 1) DR, one can get the following strong result:
Strong result: For α > 1/2 the IE requires D̄(x) to vanish for x → 0 but this is not possible for λ < 0 because the
resolvent kernel is then positive definite (think e.g. in the Neumann series) and the independent term is also positive and
diverges for y → 0 with fixed positive sign. The only possibility for having a convergent IE is to have λ > 0.
We could also generalize the IE of Eq. (12.100) by including subtraction constants in N (A), that is, with νi 6= 0.
The process is analogous to the one described for the (ℓ, ℓ)-DR case but now, even for νℓ 6= 0, we would have infrared
divergences in the form
Z
Aℓ+1 ∞ ρ(q 2 )
−νℓ−i dq 2 , (12.101)
π 0 (q 2 )i+2 (q 2 − A)

cf. Eq. (12.91). Thus, instead of Eq. (12.92) we have now


Z Z
Aℓ ∞ ρ(q 2 )
2 Aℓ ∞ ρ(q 2 )
νℓ−i dq 2 i+2 − νℓ−i dq 2
π 0 (q ) π 0 (q 2 )i+1 (q 2 − A)
i+1 Z Z
X
j Aℓ+1−j ∞ ρ(q 2 )
2 Aℓ−i ∞ ρ(q 2 )
= −νℓ−i (−1) dq 2 i+3−j − νℓ−i dq 2 . (12.102)
j=1
π 0 (q ) π 0 q 2 (q 2 − A)

In this case, even for νℓ the resulting divergence is less strong than the one in the first term on the rhs of Eq. (12.100).
This is clear because it is originally multiplied by Aℓ+1 but then the removal of the infrared divergence reduces this power
to Aℓ and the integration along the RHC gives another A−1/2 , while δℓ is multiplied by Aℓ . In terms of the function D̄(y),
α+1
after multiplying by A−ℓ and y − 2 /δℓ+1 one obtains the contribution to the IE given by
1
νℓ m(−L)− 2 − α
− y 2 . (12.103)
δℓ+1 4π

• Empirical rule: We assume that ∆(A) diverges for A → −∞ at least as in Eq. (12.83).
As far as we have checked the (ℓ, ℓ) DR, with or without νi , is convergent for λ < 0.
In the same way, the (ℓ, ℓ + 1) DR is convergent for λ > 0 in all the cases worked out until now.

• Discussion for the attractive-singular (AS) potentials and connection with the N/D method. Here we
assume that the empirical rule just stated above is fulfilled.

Odd ℓ: As shown in Sec. 12.14 an AS corresponds to λ > 0. According to the empirical rule the (0,1) DR is assumed
to convergence. If we rewrite the (0,1) as (1,1) with λ > 0 then ν1 < 0 (not zero), cf. discussion after Eq. (12.90). As
a result, the matching with the ERE implies a divergence as A−ℓ multiplied by −1/aV , apply discussion after Eq. (12.63)
with m = 0. This is equivalent to an (ℓ, ℓ + 1) DR, with ℓ subtraction constants, the δi , as free parameters. Of course,
we could include less or more free parameters, according to the order up to which the ERE is reproduced. But at least one
free parameter is needed to reproduce the right threshold behavior of the partial wave with orbital angular momentum ℓ.
Even ℓ: As shown in Sec. 12.14 an AS corresponds to λ < 0. According to the empirical rule the (1,1) DR is assumed
to converge. For ℓ = 0, S wave, ν1 is already a free parameter, and this is compatible with the situation when solving
the LS with AS. For higher partial waves, ℓ > 0, the ν1 in the (1,1) DR must vanish because of the threshold behavior.

134
We should keep all the νi = 0 because of the same reason for i = 1 up to ℓ reaching the (ℓ, ℓ) DR. In this way one could
think that the ℓ − 1 δi , i = 2, . . . , ℓ would be free parameters. But this is not necessarily the case because the (1,1) DR
converges and one could think that for the solution described all the νi and the δi correspond also to the predicted values
from the (1,1) DR. In this way there would not be any free parameter if the predicted νi , i = 1, . . . , ℓ, are zero. It is then
not obvious to discard that this last scenario does not indeed occur. Our only criterion to discard this situation would be
based on the fact that there is no renormalized free-parameter solution for the LS equation in the case of AS. Recall that
the N/D method with iterated ∆(A) and the LS equation are compatible.

• Discussion for the repulsive-singular (RS) potentials and connection with the N/D method. Here we assume
that the empirical rule just stated above is fulfilled.

Odd ℓ: As shown in Sec. 12.14 an RS corresponds to λ < 0. According to the empirical rule the (1,1) DR is assumed
to convergence. For ℓ = 1 (P wave) this is a free-parameter solution because the threshold behavior forces that ν1 = 0.
However, for higher partial waves one ends with a similar confusing situation as in AS with even ℓ. Namely, by imposing
the right threshold behavior we have that νi = 0 until we end with a (ℓ, ℓ) DR. Thus, ℓ − 1 δi appear, i = 2, . . . , ℓ, and it
is not clear that they are not indeed free parameters. One could think of a situation in which they would be fixed because
of the convergence of the (1,1) DR. For this to be the case it is necessary that all the predicted νi , i = 1, . . . , ℓ, from the
(1,1) DR vanish. Then, all of the δi would correspond to the predicted values from the (1,1) DR and this solution would
not have any free parameter. However, if all the νi , i = 1, . . . , ℓ, do not vanish then we have to impose the threshold
behavior and there are free parameters among the δi , i = 2, . . . , ℓ (this solution is not any longer the one from the (1,1)
DR). Again, our only criterion to discard this last scenario is based on the equivalence between the N/D method with
iterated ∆(A) and the LS, which requires the existence of a renormalized free-parameter solution.
Even ℓ: As shown in Sec. 12.14 an RS corresponds to λ > 0. According to the empirical rule the (0,1) DR is assumed
to converge. For ℓ = 0, S wave, this is compatible with the LS knowledge because there is no free parameter. However, for
higher partial waves we enter into a contradiction because for ℓ ≥ 2 we can rewrite the (0,1) as (1,1) and its convergence
necessarily requires that ν1 < 0 (not zero), cf. discussion after Eq. (12.90). This would imply the need of imposing the
threshold behavior and then at least one free parameter is required. To exit this paradoxical situation either the empirical
rule fails or ∆(A) for an RS with even ℓ does not diverge asymptotically for k2 → −∞ as or more strongly than λ(−k2 )α
with α ≥ 1/2. E.g. ∆(A) could be less divergent or even convergent or it could be an oscillatory function, etc.

12.12 DRs with more subtractions in N(A) than in D(A)


Let us consider a (2,1) DR with ν1 = 0 because of the threshold behavior for a P wave,
Z
A2 L ∆(k2 )D(k2 )
N (A) = ν2 A + dk2 2 2 2 ,
π −∞ (k ) (k − A)
Z Z 2 Z ∞
A ∞ 2 ρ(q 2 ) A L 2
2 ∆(k )D(k ) 2 ρ(q 2 )q 2
D(A) = 1 − ν2 dq 2 + 2 dk dq . (12.104)
π 0 q − A π −∞ (k2 )2 0 (q 2 − A)(q 2 − k2 )

To regularize the integrals along the RHC we introduce a monopole form factor −Λ2 /(q 2 − Λ2 ) with −Λ2 → +∞. We
then have
Z Z √
1 ∞ 2 ρ(q 2 ) −Λ2 ∞ 2 ρ(q 2 ) m|Λ| im A
dq 2 → dq 2 = + + O(Λ−1 ) . (12.105)
π 0 q −A π 0 (q − A)(q 2 − Λ2 ) 4π 4π
Z Z √
1 ∞ 2 ρ(q 2 )q 2 −Λ2 ∞ 2 ρ(q 2 ) m|Λ| m −k2 im A
dq 2 2 2
→ dq 2 2 2 2 2
= − + √ √ + O(Λ−1 ) .
π 0 (q − A)(q − k ) π 0 (q − A)(q − k )(q − Λ ) 4π 4π 4π k2 + A

135
Taking these expressions into the IE for D(A) in Eq. (12.104) we are lead to
Z 2 2
! √ Z 2 2  p 
m|Λ|A 1 L 2 ∆(k )D(k ) mA A mA L 2 ∆(k )D(k ) 2+ √
iA
D(A) = 1 − ν2 − dk − i ν 2 + dk − −k √ .
4π π −∞ (k2 )2 4π 4π 2 −∞ (k2 )2 k2 + A
(12.106)

In order to remove the explicit divergences in the previous equation we impose the relation
Z
1 L ∆(k2 )D(k2 )
ν2 = dk2 + O(|Λ|−1 ) . (12.107)
π −∞ (k2 )2

The first term on the rhs of this equation is the prediction for ν2 stemming from a (1,1) DR, while the remnant vanishes
as |Λ| → ∞ although gives a finite contribution to Eq. (12.106). It is then worth rewriting Eq. (12.107) as
Z 2 2

1 L 2 ∆(k )D(k ) −L
ν2 = dk 2 2
+ ν̃2 , (12.108)
π −∞ (k ) |Λ|

and then Eq. (12.106) becomes in the limit |Λ| → ∞


√ Z 2  p 
Am −L Am L 2
2 ∆(k )D(k )
√ A
D(A) = 1 − ν̃2 + 2 dk 2
− −k − i A + i √ √ . (12.109)
4π 4π −∞ (k2 )2 A + k2
√ √ √ √
By rewriting A → A( A + k2 − k2 ) in the last term we can express the previous equation in a simpler form as
Z ! √ Z
Am √ 1 L 2
2 ∆(k )D(k )
2 A Am L 2 ∆(k2 )D(k2 )
D(A) = 1 − ν̃2 −L + dk √ + dk √ √ √ . (12.110)
4π π −∞ ( −k2 )3 4π 2 −∞ ( −k2 )3 ( A + k2 )

The corresponding DR for N (A) taking into account Eq. (12.108) is


Z Z
A L ∆(k2 )D(k2 ) A2 L ∆(k2 )D(k2 )
N (A) = dk2 + dk2
π −∞ (k2 )2 π −∞ (k2 )2 (k2 − A)
Z L 2 2
A ∆(k )D(k )
= dk2 2 2 . (12.111)
π −∞ k (k − A)

Thus, Eqs. (12.110) and (12.111) correspond to a (1,2) DR, with ν1 = 0, cf. Eq. (12.31).
• The removal of the |Λ| → ∞ divergence in Eq. (12.106) can also proceed from cancellations with terms stemming
from the integral on the rhs of the equation. We cannot exclude this and then (12.107) and its consequences, Eqs. (12.110)
and (12.111), are not the only possibility.

12.13 Isolated and relevant conclusions


• General impossibility of having a CDD pole along the LHC. Had we included a CDD along the LHC located at A = B
then T (B) = 0, which would in turn imply that ∆(B) = 0, a fact typically incompatible with the free-parameter calculation
of ∆(A) form the LS equation.

• Recall: one subtraction at least in N N S waves because OPE behaves like a constant at infinity, though the discontinuity
vanishes.

• For a repulsive singular potential (RS) and ℓ > 0 it is required that the resulting partial wave after solving the N/D
method has N (0) = 0, since there is no free parameter in its solution within the LS equation. Let us recall that the

136
solutions provided by the LS equation are a subset of those provided by the N/D method and N (0) = 0 is a requirement
that any solution must fulfill. From these solutions resulting with the N/D method we can elaborate others that include
free parameters.
• On the question of the existence of a DR representation for the N (A) and D(A) functions. It is not clear a priory
because for singular potentials ∆(A) grows exponentially.
R
But in the DRs associated to the N/D method, the ∆(A) does
not appear in the one for D(A), D(A) = 1 − π1 0∞ dq 2 ρ(q 2 )N (q 2 )/(q 2 − A) (for definiteness I have just written the DR
RL
for the case N/D01 ). It appears, however, in the DR for N (A), N (A) = π1 −∞ dk2 ∆(k2 )D(k2 )/(k2 − A). Nonetheless,
we already know that these equations indeed have solutions that reproduce the standard solutions obtained within the
Lippmann-Schwinger equation. Any of these solutions could always be rewritten in terms of DRs with more subtractions,
with the values of the extra subtraction constants predicted by the standard solution obtained with less subtractions. We
can also use other values for these subtraction constants and then obtain new solutions whenever there is convergence.
The only thing we have to admit is that any solution for potential scattering is not lost by this procedure. We know
that there are N/D equations that do not converge in the case of singular potentials, but this is also exemplified by the
standard solutions which e.g. require one free parameter for the case of attractive singular potentials.

12.14 Attractive/repulsive singular potential V (r) for r → 0 and asymptotic behavior of ∆(A)
XXX This is not right, because in Eq. (12.113) Pℓ (1 − q 2 /2p2 ) could have zeros at finite position for which q 2 is around
p2 .
Let us denote by vc (r) a central potential whose partial-wave projection corresponds to the actual partial-wave projected
potential v(r). In the case of coupled channels one should first diagonalize the potential matrix in the limit r → 0. Let
∆c (A) be the spectral function of vc (r), denoted in Sec. 3 as −η(µ2 ) with µ2 = −A. As studied in Sec. 3 the behavior
of a diverging spectral function when µ → +∞ determines the degree of divergence of the singular potential for r → 0,
cf. Eq. (3.29). From this equation it is also clear that the sign of η(µ2 ) for µ → +∞ determines also the attractive or
character of v(r) for r → 0, such that if η is positive (negative) in the asymptotic ultraviolet region then the potential is
attractive (repulsive).
Now, the partial-wave projection of the equivalent central potential gives rise to our partial-wave projected ∆(A)
function. Thus,
Z +1
1
∆(A) = Pℓ (x)∆c (q 2 )dx ,
2 −1
(12.112)

where q 2 = 2A(1 − x), and x is the cosine of the relative angle between the final and initial CM three-momenta. We
change the integration variable to q 2 ≡ q 2 and then Eq. (12.112) is rewritten as
Z
1 L q2
∆(A) = − dq 2 Pℓ (1 − )∆c (q 2 ) . (12.113)
4A 4A 2A

The integration for a diverging ∆c (q 2 ) in the asymptotic limit A → −∞ is dominated by q 2 → 4A (lower limit of
integration) and Pℓ (1 − q 2 /2A) → Pℓ (−1) = (−1)ℓ .
• As a result ∆(A) and ∆c (A) have the relative sign (−1)ℓ (notice that −1/4p2 is positive).
Thus, for ∆(A) → λ(−A)α when A → −∞ with α > 0, one has an attractive or repulsive singular potential for
(−1)ℓ λ < 0 or > 0, respectively (recall that ∆c (A) is −η(−A)). A variation in the sign of λ takes place as ℓ moves from
one partial wave to another. This is summarized in Table 12.14.

137
Attractive Repulsive
Even ℓ λ<0 λ>0
Odd ℓ λ>0 λ<0

Table 2: Attractive or repulsive character of V (r) for r → 0 corresponding to the asymptotic behavior ∆(p2 ) → λ(−p2 )α ,
α > 0, when p2 → −∞.

13 New method to solve the N/D IE: Expansion in Legendre Polynomials


When writing an N/D IE for D(A) with A ≤ L there is a non-factorizable part, that we denote by G(x, y), with x = L/k2 ,
y = L/A and x, y ∈ [0, 1]. The function G(x, y) can always be chosen to be finite and continuous for x ∈ [0, 1]. If needed
we can multiply it by a factor xα , α > 0. Of course, if this is done we have also to divide by xα the factor multiplying
G(x, y) in the integrand. We can then make an expansion in Legendre polynomials (LP) of the function G(x, y) in the
variable x and the coefficients will be a function of y alone. In this form the kernel of the IE becomes a sum of IEs with
separable kernels and its solution gives us the function D(A) as a sum over the coefficients determined above. The method
is simpler to explain with an example.
Let us consider the N/D01 for a P -wave:
√ Z 2 )D(k 2 )
m A L 2 √ ∆(k
D(A) = 1 − i dk √ √ ,
4π 2 −∞ k2 ( k2 + A)
Z
1 L 2 ∆(k2 )D(k2 )
N (A) = dk . (13.1)
π 0 k2 − A
We rewrite the IE for D(A) in the variables x and y, and then it reads
1 Z 1 √
m(−L) 2 ∆(x)D(x) xy
D(y) = 1 − 1 dx 3 √ √ . (13.2)
4π 2 y 2 0 x2 x+ y
A good check about the correctness of the last equation is to reverse the change of variables in it and check whether
Eq. (13.1) is obtained again.
Then, we make the LP expansion of the function
1 √ 1
m(−L) 2 xy m(−L) 2 1
G(x, y) = − 1 √ √ =− 1
q q . (13.3)
4π 2 y 2 x+ y 4π 2 y 2 1
+ 1
x y

Since the function is only known for x ∈ [0, 1] we make an expansion involving only even LP, and write35
NL
X
G(x, y) = cj (y)P2j (x) , (13.4)
j=0

with NL + 1 the maximum number of LP included. The coefficients cj (y) are given by the standard projection formula
into a LP:
Z 1
1 Z 1
m(−L) 2 1
cj (y) = (4j − +) dxG(x, y)P2j (x) = −(4j − 1) 1 dxP2j (x) q q . (13.5)
0 4π 2 y 2 0 1
x + 1
y
35
One could also try an expansion involving only odd LP. However, a preliminary study drove to me to consider it worse behaved since the
expansion in LP is more oscillatory. This is due to the mismatch between the linear behavior of P1 (x) and the actual derivative of the function
G(x, y) near x = 0. Furthermore, in the general case G(0, y) 6= 0 as required by an expansion in LPs with odd indeces.

138
Inserting the expansion of G(x, y) in Eq. (13.4) back to Eq. (13.2) we are lead with
NL
X Z 1 ∆(x)D(x)
D(y) = 1 + cj (y) dx 3 P2j (x) . (13.6)
j=0 0 x2

In the following the last integral is denoted by ξj , namely,


Z 1 ∆(x)D(x)
ξj = dx 3 P2j (x) . (13.7)
0 x2

Next, we multiply Eq. (13.6) by ∆(y)P2i (y)/y 3/2 and integrate it in the variable y in the interval [0, 1]. In this form the
IE transforms into an algebraic matrix equation for the ”moments” ξi :
NL
X
ξi = ω i + aij ξj . (13.8)
j=0

In this equation,
Z 1 ∆(x)P2i (x)
ωi = dx 3 , (13.9)
0 x2
and the matrix elements aij are given by
Z 1 ∆(x)P2i (x)
aij = dx 3 cj (x) . (13.10)
0 x2
We could also write aij as a double integral by inserting in the previous equation the expression for cj (y) given by Eq. (13.5).
Numerically, I have first created an array with the coefficient function cj (y) of dimension (NL + 1) × ML , with ML the
number of points considered along the LHC. In terms of the cj (y) stored I then calculate the integration to obtain the aij
as an (NL + 1) × (NL + 1) array.
Eq. (13.8) can be easily solved and we can express D(A) along the LHC as, cf. Eq. (13.6),
NL
X
D(A) = 1 + cj (y)ξj . (13.11)
j=0

For the calculation of D(A) along the RHC we come back to the expression of D(y) in Eq. (13.2) and substitute y for
L/A. However, this process requires some care when fractional powers of y appears. Now, for A + iε, A > 0, one has:

L
y= ,
A
s s 1
1 L L (−L) 2
y =
2 = + iε = i √ , (13.12)
A + iε A A

where we have employed that L < 0. Taking this result into account we rewrite Eq. (13.2) as
√ 1 Z
m A(−L) 2 1 ∆(x)D(x) 1
D(A) = 1 − 2
dx 3 √ √ . (13.13)
4π 0 x2 A + i −k2

139
From this expression and the definition of the ξj in Eq. (13.7), we easily identify the function to be expanded along the
RHC, denoted as R(k2 , A), and given by
√ 1
2 m A(−L) 2 1
R(k , A) = − √ √
4π 2 A + i −k2
NL
X
= dj (A)P2j (x) . (13.14)
j=0

The coefficients dj (A) correspond to


1 Z
L
dj (A) = (4j − 1) dxR( , A)P2j (x)
0 x
√ 1 Z 1
m A(−L) 2 P2j (x)
=− 2
(4j − 1) dx √ √ . (13.15)
4π 0 A + i −k2
In this way, D(A) for A ≥ 0 can be rewritten as
NL
X
D(A) = 1 + dj (A)ξj , (13.16)
j=0

with the ξj already determined by studying D(A) along the LHC and given by the solution of Eq. (13.8).
Equations (13.13) and (13.16) can also be obtained in another less direct away. To accomplish that we express D(A)
for A ≥ 0 in the original expression given in Eq. (13.1) as
√ Z
m A L 2 ∆(k2 )D(k2 )
D(A) = 1 − dk √ √ √ . (13.17)
4π 2 −∞ −k2 ( A + i −k2 )
Next, one rewrites the integration in k2 as an integral in x and from the expression obtained one can identify the part of
the integrand giving rise to the ξj , while the remaining factor is the one that should be expanded in LP.
The process to calculate N (A) is also similar as the one followed to obtained D(A) for A ≥ 0. We have the starting
expression given in the second line of Eq. (13.1) and one rewrites it in terms of the variable x identifying the part of the
integrand giving rise to the ξj . Working it out explicitly one has:
Z 1
L ∆(x)D(x) 1
N (A) = − dx 3 1
π 0 x (k2 − A)x 2 2

Z 1 1 √
(−L) ∆(x)D(x) −k2
2
= dx . (13.18)
π 0
3
x2 k2 − A
Then, we write N (A) as
NL
X
N (A) = nj (A)ξj ,
j=0
1 Z √
(−L) 2 1 −k2
nj (A) = (4j − 1) dxP2j (x) 2 . (13.19)
π 0 k −A
Along the RHC unitarity implies that

ImD(A) = −ρ(A)N (A) . (13.20)

140
Taking the imaginary part of the expression for D(A) with A ≥ 0, Eq. (13.16), we have
NL
X
ImD(A) = Imdj (A) ξj ,
j=0
√ 1 Z 1 √
m A(−L) 2 P2j (x) −k2
Imdj (A) = − (4j − 1) dx . (13.21)
4π 2 0 k2 − A

We see that Imdj (A) corresponds to nj (A) times −m A/4π and the unitarity relation is then fulfilled.
• Criterion of numerical stability: The function G(x, y) defined in Eq. (13.3) is amenable to a Legendre expansion in
both the variables x and y. As a stability criterion for our numerical calculations we could also compare our results with

a LP expansion of the new function xG(x, y). This would require to divide the integrands of ωj and aij in Eqs. (13.9)
√ √
and (13.10), respectively, by x (it is like transforming ∆(x) → ∆(x)/ x in their expressions). This would make that
the corresponding integrations given these quantities are more divergent for singular potentials, but at least for low values
of A we should demand stability. E.g. if this is done for the N/D22 m3 P0 wave we immediately realize that it is unstable,
while this is not the case for the N/D23 IE.
After working out several cases we have realized of the following rule which indeed is quite intuitive. Given the function
G(x, y) the best convergence properties to solve the IE are obtained when expanding the function G(x, y)/xα , with α
taking the maximum value such that

G(x, y)
lim <∞. (13.22)
x→0 xα
In this way we would have to multiply ∆(x) by xα which somewhat favors the convergence of the integrals involving the
(divergent) function ∆(x).
• To generate numerically the LP we have used the standard recurrence relation:

ℓPℓ−1 (x) − (2ℓ + 1)xPℓ (x) + (ℓ + 1)Pℓ+1 = 0 , (13.23)

together with P0 (x) = 1 and P1 (x) = x. The use this recurrence relation is also recommended by the Fortran Recipes
book. In this form we have generated numerically a matrix of LP of dimension 2NL × ML .
• The orthogonality properties calculated numerically for the LP generated get clearly worse typically for NL & 100.
• The calculation of the G(x, y) can be simplified in the sense that it is not really needed to express it in terms of the
variables x and y. We could just write Eq. (13.1) in terms of the variable x along the LHC as
√ Z
m AL 1 ∆(x)D(x) 1
D(A) = 1 − 2
dx 3/2
√ √ , (13.24)
4π 0 x 2
−k + −A
with

2 m AL 1
G(k , A) = − 2
√ √ . (13.25)
4π 2
−k + −A

141
14 Application of the formalism to N N 1S0
We want to apply in this section the formalism developed for the N N partial wave 1 S0 .
• First we discuss the one-pion-exchange (OPE) case and start by calculating ∆(A). The potential is given in this case
by,
g     
v(p1 , p2 ) = log (p1 + p2 )2 + m2π − log (p1 − p2 )2 + m2π ,
2p1 p2
!2
gA mπ 2 m2
gA π
g= √ = 2
. (14.1)
8fπ 8fπ

We show next that in the limit k ≫ mπ one can deduce an algebraic expression for ∆(A) and that the same result is
obtained either by using the three-step process of Sec. 7.2 or the one-step formalism of Sec. 7.3.
• In both cases one needs the function f(ν), defined in Eq. (7.61), that obeys the IE of Eq. (7.94). It can be
straightforwardly deduced that this IE in our present case reads (recall that ν < k − mπ )
Z k−mπ dν1
f(ν) = −πg + θ(k − 2mπ − ν)λ f(ν1 ) , (14.2)
mπ +ν k2− ν12
where we have introduced the constant λ,
mg
λ= . (14.3)

We transform this IE into an differential equation (DE) by differentiating it with respect to ν. The resulting DE is:
f′ (ν) θ(k − 2mπ − ν)
= −λ 2 (14.4)
f(ν + mπ ) k − (mπ + ν)2
Notice that the term with the derivative of the Heaviside function θ(k − 2mπ − ν) from Eq. (14.2), that results in a
Dirac-delta function, does not contribute because its coefficient, an integral of zero extent, is zero since.
The DE of Eq. (14.4) is not easy to solve due to the shift in mπ in the argument of the denominator in the lhs. We are
then tempted to neglect this shift because we are assuming that mπ is very small compared with k. However, the variable
ν could take values around k − mπ or −k in which case the factor 1/(k2 − (mπ + ν)2 ) develops an infrared divergence
∝ 1/kmπ . That is, one is also sensitive to k − |ν| which could be also very small.
Let us first assume that we can transform the DE of Eq. (14.4) into the much simpler form
f′ (ν) θ(k − 2mπ − ν)
= −λ 2 (14.5)
f(ν) k − (mπ + ν)2
which keeps the possible infrared divergent behavior of the rhs, as in the original DE. The solution of Eq. (14.5) is
straightforward and is given by
  
λ ν + mπ
f(ν) = f(k − 2mπ ) exp − arctanh θ(k − 2mπ − ν) . (14.6)
k k
Now, f(ν) = −πg for ν ≥ k − 2mπ , as follows from the IE obeyed by f(ν), Eq. (14.2). This allows us to fix the integration
constant in Eq. (14.6) and hence, it results
     
λ ν + mπ λ k − mπ
f(ν) = −πg exp − arctanh exp arctanh θ(k − 2mπ − ν) − πgθ(ν − k + 2mπ )
k k k k
 −λ/2k
k + ν + mπ mπ
= −πg θ(k − 2mπ − ν) − πgθ(ν − k + 2mπ ) . (14.7)
k − ν − mπ 2k − mπ

142
We can check this solution by reinserting it back in the original DE Eq. (14.4):
    
f′ (ν) λ λ λ ν + mπ λ ν + 2mπ
+ 2 2
= 2 1 − exp − arctanh exp arctanh . (14.8)
f(ν + mπ ) k − (ν + mπ ) k − (ν + mπ )2 k k k k

If our solution Eq. (14.7) were exact the rhs of this equation, that we call error function, would be zero. It is not so but
for k >> mπ , |ν| the error function is

λ2 m π 1
− 4
+ × O(ε5 ) , (14.9)
k k
with ε = ν/k or mπ /k. For |ν| ∼ k then error function is

λ2
− × log const , (14.10)
4mπ k2
where the constant depends exactly on the value of ν. It only goes to zero for ν → k − 2mπ (where the scale for how
far/close ν is to this value is set by mπ ), but in this limit the solution is exact because it is fixed by the boundary condition
guaranteed by the first Heaviside function in Eq. (14.7).
• In the case of the one-step process ∆(A) can be expressed directly in terms of f(−k). From Eq. (7.79) if follows that
    −λ/k
f(−k) λπ 2 2λ k − mπ λπ 2 mπ
∆(A) = 2
= − 2
exp arctanh =− . (14.11)
2k mk k k mk2 2k − mπ
• Now, let us probe that the same result is obtained from the longer three-step process. In this case, the discontinuity
∆(A) is given by Eq. (7.48) (3rd step), that in our present case reads
Z k−mπ
πg λ dν1
∆(A) = − − θ(k − mπ ) 2 Imt̂(iν1 + ε+ , ik + ε) . (14.12)
2k 2 k 0 k2 − ν12

This integral first requires the knowledge of the function Imt̂(iν + ε+ , ik + ε), ν ≥ 0, that we denote in the following as
h(ν),

h(ν) = Imt̂(iν + ε+ , ik + ε) . (14.13)

This function obeys the IE in Eq. (7.53) (2nd step), which now becomes
Z k−mπ Z ν−mπ
θ(k − mπ )m dν1 dν1
h(ν) = πg − Imv̂(iν + ε+ , iν1 + ε)f(ν1 ) + θ(ν − mπ )λ h(ν1 ) . (14.14)
2π 2 0 k − ν12
2
0 k2− ν12

The independent term of this IE requires to know the function f(ν) for ν ≥ 0, that we already calculated (1st step),
Eq. (14.7). Next, from Table 3 the first integral (k ≫ mπ ) can be written as
Z ν+mπ Z k−mπ
dν1 1 dν1
−λ θ(k − 2mπ − ν) f(ν1 ) + θ(k − 2mπ − ν) f(ν1 )
|ν−mπ | k − ν12 2
2
ν+mπ k2− ν12
Z ! Z Z !
k−mπ dν1 1 λ k−mπ dν1 k−mπ dν1
+ θ(ν − k + 2mπ ) 2 f(ν1 ) = − f(ν1 ) + θ(k − 2mπ − ν) f(ν1 ) .
ν−mπ
2
k − ν1 2 2 |ν−mπ | k − ν12
2
ν+mπ k − ν12
2

(14.15)

143
Combining Eqs. (14.14) and (14.15) we have the following IE for h(ν),
Z k−mπ Z k−mπ Z ν−mπ
λ dν1 λ dν1 dν1
h(ν) = πg − 2 f(ν1 ) − θ(k − 2mπ − ν) f(ν1 ) + θ(ν − mπ )λ h(ν1 ) .
2 |ν−mπ |
2
k − ν1 2 ν+mπ k − ν12
2
0 k2 − ν12
(14.16)

We now transform this IE into a DE, as done above regarding Eq. (14.2), by taking the derivative with respect to ν in
both sides of the expression (keeping in mind that ν < k − mπ ), one ends with
h(ν − mπ ) λ f(ν − mπ ) λ f(ν + mπ )
h′ (ν) − λ = θ(k − mπ − ν) 2 + θ(k − 2mπ − ν) 2 . (14.17)
k2 − (ν − mπ )2 2 k − (ν − mπ )2 2 k − (ν + mπ )2
We substitute in the previous equation the expression for f(ν) obtained in Eq. (14.7), which then reads
(  − λ
′ h(ν − mπ ) λπg θ(k − 2mπ − ν)θ(ν − k + 3mπ ) θ(k − mπ − ν) k + ν mπ 2k
h (ν) − λ 2 2
=− 2 2
+ 2
k − (ν − mπ ) 2 k − (ν + mπ ) k − (ν − mπ )2 k − ν 2k − mπ
 − λ )
θ(k − 3mπ − ν) k + 2mπ + ν mπ 2k
+ 2 . (14.18)
k − (ν + mπ )2 k − 2mπ − ν 2k − mπ

Now, the first term on the rhs of the previous equation only gives contribution in the narrow interval k − 3mπ < ν <
k − 2mπ . Its contribution to h(ν) goes like 1/(2mπ k), which is then suppressed by one power of 1/k. The presence of
the infrared enhancement 1/mπ is compensated by the narrow region in the integration that drives to ∆(A), Eq. (14.12).
We disregard this term in the following. The second and third terms on the rhs of Eq. (14.18) only differ by a shift in ν
of ∓mπ , respectively, in ν, which is only relevant for the region ν → k − mπ that, as we have shown, has a suppressed
contribution to ∆(A). We then combine both terms by reabsorbing in ν the mentioned shift, and similarly we rewrite h(ν)
instead of h(ν − mπ ) in the second term on the lhs. Then Eq. (14.18) simplifies to
 − λ
′ h(ν) θ(k − 2mπ − ν) k + mπ + ν mπ 2k
h (ν) − λ 2 2
= −λπg (14.19)
k − (ν − mπ ) k2 − ν 2 k − mπ − ν 2k − mπ
The solution to the homogeneous DE associated with the previous equation can be readily obtained (it is of the same type
as Eq. (14.4)),
 
λ ν − mπ
hH (ν) = h0 exp arctanh . (14.20)
k k
For the particular solution we neglect all the factors of mπ in front of k or ν (recall that the resulting h(ν) is going to be
used within an integral in ν of an extent k − mπ ≫ mπ ) so that Eq. (14.19) simplifies to
2  − λ
h(ν) 2π 1 k + ν mπ 2k

h (ν) − λ 2 2
= −2λ , (14.21)
k −ν m k − ν2
2 k − ν 2k
where we have used that g = 2πλ/m, cf. Eq. (14.3). A particular solution of this DE is
   
λπ 2 λ ν λ k − mπ
hP (ν) = exp − arctanh exp arctanh , (14.22)
m k k k k
The sum of Eqs. (14.21) and (14.22) gives
     
λ ν − mπ λπ 2 λ ν λ k − mπ
h(ν) = h0 exp arctanh + exp − arctanh exp arctanh . (14.23)
k k m k k k k

144
To determine the integration constant h0 we consider the original IE obeyed by h(ν), Eq. (14.14), whose independent term
is the solution for ν < mπ . We take ν = 0 and then
Z
2π 2 λ k−mπ dν1
h(0) = −λ f(ν1 ) , (14.24)
m mπ k2 − ν12

where we have used Table 3 to express Imv̂(ε+ , iν1 + ε). The integration can be done algebraically by neglecting the shift
in ν in the expression for f(ν), Eq. (14.7), which we take as
     
2π 2 λ λ ν λ k − mπ
f(ν) → − exp − arctanh exp arctanh . (14.25)
m k k k k
Inserting this expression back to Eq. (14.24) one obtains
       
2π 2 λ λ mπ λ mπ 2π 2 λ λ mπ
h(0) = exp − arctanh exp arctanh 1 − → exp arctanh 1 − . (14.26)
m k k k k m k k
We can now fix h0 by equation h(0) above with the solution Eq. (14.23) evaluated at ν = 0, if follows that
  
λπ 2 λ mπ
h0 = exp arctanh 1 − , (14.27)
m k k
2        
λπ λ mπ λ ν − mπ λ ν
h(ν) = exp arctanh 1 − exp arctanh + exp − arctanh . (14.28)
m k k k k k k
We are now ready to calculate ∆(A) in the three-step process by performing the integration in Eq. (14.12). This integral
can be done algebraically if we neglect the shift in ν by mπ in the first exponential between curly brackets in Eq. (14.28),
that is, by using
       
λπ 2 λ mπ λ ν λ ν
h(ν) → exp arctanh 1 − exp arctanh + exp − arctanh . (14.29)
m k k k k k k
This would be the expression obtained if in the homogeneous DE had we neglected from the start the shift in ν in the
denominator of lhs of Eq. (14.19), i.e. k2 − (ν − mπ )2 → k2 − ν 2 . The same result for ∆(A) as in Eq. (14.11) is then
obtained after performing the integration in Eq. (14.12).

14.1 Next-to-leading order (NLO) study for the 1 S0


We take the calculation of Ref. [5] for the N N potential at NLO in BChPT. The result is expressed in terms of the function
L(x) which has only LHC for x = q 2 /4m2π < −1. To accomplish this property from an algebraic expression of L(x) that
coincides with the original one in an open domain around the physical region for q 2 , namely q 2 > 0, the function L(x) has
to be written properly. For this aim we define the auxiliary function sq(x) = x1/2 with the cut along the RHC, so that
argx ∈ [0, 2π] contrary to the standard Fortran or Mathematica conventions for the square root function, that we denote

by x. In this way,
√  
1+x √
L(x) = log sq(x) + 1 + x . (14.30)
sq(x)
Its discontinuity along the LHC is given by

1−x
lim {L(x + iǫ) − L(x − iǫ)} = 2iImL(x) = iπ √ , x < −1 . (14.31)
ǫ→0+ −x

145
As a check of this result we can write down the following once-subtracted DR for L(x)
Z
x −1 ImL(y + iǫ)
L(x) = L(0) + dy (14.32)
π −∞ y(y − x)
Z √
x −1 1−y 1
= L(0) + dy √ , (14.33)
2 −∞ −y y(y − x)

which is very well fulfilled. In its numerical calculation we have performed the change of integration variable t = −1/y
and have the simpler expression
Z √
x 1 1−t
L(x) = L(0) + dt . (14.34)
2 0 1 − tx
We follow the procedure explained in App. A to determine the N N partial-wave projected N N potential from Refs. [5]
and [4]. Ref. [5] gives all the NLO terms in terms of the functions WC (q 2 ), VT (q 2 ) = −VS (q 2 )/q 2 with q 2 = q2 . The
resulting formula for the NLO contribution to the 1 S0 potential is
Z
1 +1 n o
V (1 S0 ) = dx WC (q 2 ) + 2q 2 VT (q 2 ) , (14.35)
2 −1
q = 2p2 (1 − x) .
2

In order to proceed we need the spectral decomposition of the integrand in the previous equation. For that we need
in turn the imaginary parts of the potential functions WC (q 2 ) and VT (q 2 ) in terms of the imaginary part of the function
L(x) at q 2 + iǫ for q 2 < −4m2π . The most delicate structure for this purpose is the following one appearing in WC (q 2 ),

L(x)
. (14.36)
x+1
It is important to stress for the following discussion that L(−1) = 0, as it is clear from Eq. (14.30). In this way the
contribution from the imaginary part of 1/(x + 1) for x + iǫ and x < −1, which is −iπδ(x + 1), finally vanishes when used
within the integral of the spectral representation of a potential, cf. Eq. (3.20), because it is multiplied by L(−1) = 0. We
have checked the calculated imaginary parts for the potentials WC and VT by the agreement between these functions and
the corresponding DRs. The detailed expressions for the imaginary parts and DRs can be found in the Mathematica code
/Numerics/HigherOrders.nb.
Let us also remind that for the calculation of the discontinuity of a DR one can use the unsubtracted form of the DR.
Additionally, we are interested in ∆v̂(ν, ν1 ), which requires integrating the spectral decomposition for the potential up to
ν1 − ν, cf. Eq. (8.59). Thus,
Z h i Z
′ pp′ ν1 −ν +1 dx
∆v̂(ν, ν1 ) = pp ∆V (ν, ν1 ) = 2µdµ ηC (µ2 ) − 2µ2 ηT (µ2 ) ∆ (14.37)
2π 0 −1 µ2 + q 2
Z ν1 −ν n o
1
=− 2µdµ ηC (µ2 ) − 2µ2 ηT .
2 0
Notice that for the calculation of the discontinuity we have replaced the factor q 2 in front of VT (q 2 ) in Eq. (A.6) by −µ2
because the difference q 2 + µ2 cancels the denominator and no discontinuity would result then.
We have checked that Eq. (14.37) reproduces the perturbative NLO contribution to ∆(A), already used in Ref. [3].
The expression is simply
Z −4A n o
∆v̂(−k, k) 1
∆(A) = = dµ2 ηC (µ2 ) − 2µ2 ηT . (14.38)
2k2 4A 0

146
For the present case one can perform algebraically the integration in Eq. (14.37) and obtain a close expression for
∆v̂(ν, ν1 ) that is explicitly given in the Mathematica code /Numerics/HigherOrders.nb. In terms of it we can then apply
the standard procedure to calculate f(ν), e.g. by employing Eqs. (7.137), (7.139) and (7.79).
We show in Fig. 29 the function ∆(A) calculated following Entem’s (MD) and Oller’s (MO) methods. It is clear from
this figure that the calculation of ∆(A) is not an easy task for high energies, where it becomes huge, as it is clear by
the lack of convergence within the different lines calculated with MO as k is getting closer to 103 mπ . There is also
persistent differences between the results with MO and MD methods in the same range of large momenta even when using
600 · 20 = 1200 points in MO (we do not go further by including more points in MO because as we can see in Fig. 32 the
phase shifts have already converged and it is enough to point out this difficulty). The green points in the far right of the
right panel of Fig. 29 are plotted at zero by gnuplot because their size is larger than 10100 .
100 100
10 10
MD.L5e3.npar2.mt100.mp4 MD.L5e3.npar2.mt100.mp4
90 MO.L5e3.npar2.mt100.mp50 90
MD.L5e3.npar2.mt100.mp6
10 MO.L5e3.npar2.mt100.mp100 10 MO.L5e3.npar2.mt50.mp50
MO.L5e3.npar2.mt100.mp200 MO.L5e3.npar2.mt100.mp100
80 MO.L5e3.npar2.mt100.mp400 80 MO.L5e3.npar2.mt100.mp200
10 10
MO.L5e3.npar2.mt100.mp600 MO.L5e3.npar2.mt100.mp400
70 MO.L5e3.npar2.mt100.mp600
10 70
10

60
10 10
60

50
|∆| (mπ-2)

10 10
50

|∆|(|L| )
-1
40
10 40
10
30
10 30
10
20
10 20
10
10
10 10
10
0
10 0
10
-10
10 -10
1 10 100 1000 10000 10
6 7 8
-A (mπ)
√ 0.1 1 10 100 1000 10000 100000 1x10 1x10 1x10
A(|L|)


Figure 29: ∆(A) is plotted as a function of −A (mπ ) (left panel) and A (L) (right panel). In the left panel ∆(A) is given in units
of m−2
π while in the right one the units are L−1 .

Another way to realize of these difficulties in the calculation of ∆(A) at high |A| is by plotting the relative difference
between calculations using different number of points in the discretization process or by computing the relative difference
between results with different methods. This relative difference is defined as
2(∆(A) − ∆(A)′ )
ε∆ = . (14.39)
∆(A) + ∆(A)′

We plot ε∆ in Fig. 30 where we include a solution with MD in the left panel. It is obvious from the figure that we do
not (indeed we cannot with the present methods) calculate ∆(A) accurately for k larger than some value below 103 mπ .
We also point out that the straightforward application of MD drives to problems in the threshold region because we are
using in its application a smooth extrapolation in ∆v̂(ν, ν1 ) in the variable ν1 , which is problematic in the onset of the 2π
threshold, which stems from the NLO ChPT calculation of Ref. [5].
Next, we apply the function ∆(A) to calculate the N N 1 S0 partial wave by solving the N/D11 IE deduced in Sec. 12.1.
We need here as input the 1 S0 N N scattering length that we fix to the value as = −23.75 fm. The straightforward
solution is shown in the left panel of Fig. 31. A good numerical convergence in the calculation of the phase shifts is
observed, which implies that the discussed difficulties for higher k in the calculation of ∆(A) only gives rise to very small
uncertainties in this observable. In the right panel we also show the phase shifts obtained from the N/D11 IE but cutting

147
2 1
10 10
MO.mp600 MD.mp6 MO.mp600 MO.mp400
MO.mp600 MO.mp400 0 MO.mp400 MO.mp200
MO.mp400 MO.mp200 10 MO.mp200 MO.mp100
0
10 MO.mp200 MO.mp100
-1
10
-2
10 -2
10

-3
10
-4 10
2|∆O-∆D|/|∆O+∆D|

2|∆O-∆D|/|∆O+∆D|
-4
10
-6
10
-5
10

-8
10 10
-6

-7
-10 10
10
-8
10
-12
10
-9
10

-14 -10
10 10
0.1 1 10 100 1000 10000 0.1 1 10 100 1000 10000
-A (mπ)
√ -A (mπ)

Figure 30: Relative difference in ∆(A) between calculations using different number of points and/or different methods (MO and
MD).

off the DR IE of Eq. (12.2) along the LHC. Namely, we nave now the cutoff IE
√ Z
mA L ∆(k2 )D(k2 )
D(A) = 1 − as −A + 2 dk2 √ √ . (14.40)
4π −Λ2 k2 ( −k2 + −A)

We use in practice Λ = 103 and 102 mπ and the resulting phase shifts are shown in the right panel of Fig. 31. The former
value of Λ is indicated in the legend of each calculation by the letter “c” and the latter by “cc”. One can see that there
is a good convergence in Λ so that in the scale of the figure one cannot basically distinguish between the results for these
values of Λ and Λ → ∞ (also shown). This is why the onset of numerical problems in the calculation of ∆(A) do not
finally affect the calculation of the phase shifts since their calculation has already converged.

We also show in Fig. 33 the product |∆(A)D(A)| (left panel) and |∆(A)D(A)/ A (right panel). The cloud of points
indicates the region where there are strong numerical problems in its calculation. It is clear that handling the large values
of |∆(A)| that results at high values of k gives rise to numerical problems in the solution of the IE of Eq. (12.2) also for
large k.
Another way to solve the IE of Eq. (12.2) is multiplying it by ∆(A) and directly obtain an IE for the product ∆(A)D(A).
This IE is
√ i mA∆(A) Z L
h ∆(k2 )D(k2 )
2
∆(A)D(A) = ∆(A) 1 − as −A + dk √ √ . (14.41)
4π 2 −Λ2 k2 ( −k2 + −A)

However, the resulting IE is numerically more unstable than the one for D(A) because for the former the independent
term is also affected by the numerically problematic ∆(A) (while for Eq. (12.2) this is not the case). As a result the phase
shifts are not stable when calculated from Eq. (14.41) with Λ → ∞, as shown in the left panel of Fig. 33. When using the
values Λ = 103 and 102 mπ we obtain the results shown in the right panel of the same figure. For the smaller value of Λ
the phase shifts are convergent to the results obtained already from Eq. (12.2), while for Λ = 103 mπ they are not under
numerical control. This again insists on the numerical problems for calculating precisely ∆(A) when k is getting closer to
103 mπ . It is also important to notice that the good convergence towards the previous calculated values for Λ = 102 mπ
is achieved because we are using ∆(A) calculated with the same cutoff as already employed (5 · 103 mπ ), which is also
the largest one at our disposition. Namely, we are not using as a cutoff to calculate ∆(A) the value of Λ appearing in
Eq. (14.41).

148
70 70
MD.mt100.mp6 MO.mt100.mp400 MD.mt100.mp6 MOc.mt100.mp100
MD.mt100.mp4 MO.mt100.mp600 MDc.mt100.mp4 MOcc.mt100.mp100
MO.mt100.mp100 NLO-paper MDcc.mt100.mp4 NLO-paper
60 MO.mt100.mp200 60 MO.mt100.mp100
24

50 20 50
16

12
40 40
8
δ

δ
4
320 360 400
30 30

20 20

10 10

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
p p

Figure 31: The 1 S0 phase shifts calculated from the N/D11 IE of Eq. (??) (left panel) and the cutoff one of Eq. (14.40). The
inset in the left panel shows the phase shifts for p ≥ 300 MeV obtained by the calculations with the highest number of points in the
determination of ∆(A).

14 12
1x10 10
MD.mt100.mp4 MD.mt100.mp6
MDC.mt100.mp6 MO.mt100.mp200
12 MO.mt100.mp100 10 MO.mt100.mp400
1x10 10
MO.mt100.mp200 MO.mt100.mp600
MO.mt100.mp400
MO.mt100.mp600
10 8
1x10 10
-A| (mπ-3)

8 6
1x10 10
|D(A)∆(A)|

|D(A)∆(A)/√

6 4
1x10 10

2
10000 10

0
100 10

-2
1 10
6 7 8 6 7 8
0.1 1 10 100 1000 10000 100000 1x10 1x10 1x10 0.1 1 10 100 1000 10000 100000 1x10 1x10 1x10
2
|A| (mπ ) |A| (mπ2)


Figure 32: The product |∆(A)D(A)| (left panel) and |∆(A)D(A)/ A| (right panel) calculated from the N/D11 IE of Eq. (12.2).

149
100 100

80 80

60 60

40 40

20 20

0 0
δ

δ
-20 -20

-40 -40

-60 -60

-80 -80

-100 -100
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
p p
MD.mt100.mp6 MO.mt100.mp600 MOD.mt100.mp200 MD.mt100.mp6 MO.mt100.mp600 MODc.mt100.mp400
MDD.mt100.mp6 MOD.mt100.mp600 MOD.mt100.mp100 MDDc.mt100.mp6 MODc.mt100.mp600 MODcc.mt100.mp400
MDD.mt100.mp4 MOD.mt100.mp400 NLO-paper MDDcc.mt100.mp4 MODcc.mt100.mp600 NLO-paper

Figure 33: The phase shifts calculated from the IE of Eq. (14.41). In the left panel Λ → ∞ while in the right one Λ = 103 and
102 mπ . The numerical results are convergent for Λ = 102 mπ , which is indicated by the letter “c” in the legends of the results, while
there is no convergence for Λ = 103 mπ , for which we use the letters “cc”.

We also show the product |∆(A)D(A)| obtained from the solution of the IE of Eq. (14.41) in Fig. 34. There are also
numerical instabilities for small k because the independent term is affected by ∆(A) in this IE. Only when the IE is cutoff
with Λ . 102 mπ we obtain numerically stable results, as shown by the dots in the right panel of the same figure. In this
case the phase shifts agree very well with those obtained by solving the standard N/D11 IE of Eq. (12.2), as just shown
in Fig. 33.
We finally show the results that are obtained by employing the method based on the expansion in Legendre polynomials
of the non-separable part in the integrand of the IE of Eq. (12.2), a method explained in Sec. 13. Again we obtain unstable
results when directly solving this IE without cutting it off. However, if we use the IE of Eq. (14.41) and take Λ = 102 mπ
the results are numerically stable and convergent (although a very large number of Legendre polynomials are needed to
really obtain the same results as with the other two methods.) This corresponds to the phase shifts shown in the left panel
of Fig. 35. However, by taking larger values of Λ, the results are not numerically under control. This is why the phase
shifts shown in the right panel of the Fig. 35 vary so much from one calculation to the other. In all the cases Λ = 103 mπ
or 5 · 103 mπ . The huge size of |∆(A)| at high energies make problematic the inversion of the system of linear equation
in Eq. (13.7). E.g. one can check that the solution obtained fails to satisfy the system of linear equations when Λ is too
large, in particular, within the range of values shown in the right panel of Fig. 35.

15 Application of the formalism to N N 1P1


• The OPE partial-wave projected potential in the 1 P1 wave is
Z +1
′ t
v(p, p ) = −3g dt t
−1 q2 + m2π
3g 3g h    i
2 ′2 2 ′ 2 2 ′ 2 2
= ′− (p + p + m π ) log (p + p ) + m π − log (p − p ) + m π . (15.1)
pp 4(pp′ )2

Only the second term on the rhs of the previous equation contributes to the discontinuity of the potential.
• The IE we have to solve is the particularization of Eq. (7.94) to the present case. Similarly as in the 1 S0 one obtains

150
8 8
10 10
MDDcc.mt100.mp6
7 MODc.mt100.mp600
10 10
7
MODcc.mt100.mp600
6
10 6
10
5
10
5
|D(A)∆(A)| (mπ-3)

10
4
10

|D(A)∆(A)| (mπ-3)
4
10
3
10
3
2
10
10

2
10
1 10

0 1
10 10

-1
10 10
0
6 7 8
0.1 1 10 100 1000 10000 100000 1x10 1x10 1x10
|A| (mπ2)
-1
MDD.mt100.mp6 MOD.mt100.mp600 MODc.mt100.mp600 10 6
0.1 1 10 100 1000 10000 100000 1x10
MDD.mt100.mp4 MOD.mt100.mp400
|A| (mπ2)
MDDc.mt100.mp6 MOD.mt100.mp200

Figure 34: The product |∆(A)D(A)| from the solution of the IE of Eq. (14.41). In the left panel we take Λ → ∞ and two cases with
Λ = 103 mπ . In the right panel we show the convergent situation with Λ = 102 mπ by the dots while the line is for Λ = 103 mπ ,
which has strong numerical problems.

70 100

80
60
60

50 40

20
40
0
δ
δ

30 -20

-40
20
-60

10 -80

-100
0 0 50 100 150 200 250 300 350 400
0 50 100 150 200 250 300 350 400 p
p PG.mt100.mp200.np200.ffx0p5 PGc.mt100.mp600.np800.ffx0p5
mt100.mp600.np200.ffx0p5 PGcc.mt100.mp600.np1200.ffx0p5 PGcc.mt100.mp600.np800.ffx1p0
PG.mt100.mp200.np400 PGc.mt100.mp600.np1000.ffx0p5
mt100.mp600.np400.ffx0p5 PGcc.mt100.mp600.np1400.ffx0p5 MODcc.L5e3.npar2.mt100.mp600
mt100.mp600.np600.ffx0p5 PGcc.mt100.mp600.np1600.ffx0p5 MO.L5e3.npar2.mt100.mp600 PG.mt100.mp200.np300 MO.L5e3.npar2.mt100.mp600
mt100.mp600.np800.ffx0p5 PGcc.mt100.mp600.np1800.ffx0p5 MD.L5e3.npar2.mt100.mp8 PGc.mt100.mp600.np400.ffx0p5 MD.L5e3.npar2.mt100.mp8
mt100.mp600.np1000.ffx0p5 PGcc.mt100.mp600.np2000.ffx0p5 PGc.mt100.mp600.np600.ffx0p5

Figure 35: The phase shifts calculated by the method of expanding in Legendre polynomials the non-separable part of the N/D11
IE, Eqs. (12.2) and (14.41). The left panel take Λ = 102 mπ and it is numerically stable. However, the results become in the right
panel because the Λ used is too large, Λ = 103 or Λ = 5 · 103 mπ .

151
the same IE for f(ν) in the whole ν interval, −k ≤ ν ≤ k − mπ , which can be written as:
Z  
3πg 2 3gm k−mπ 1 1 ν 2 + ν12 − m2π
f(ν) = − (k + ν 2 − m2π ) − θ(k − 2mπ − ν) dν1 + f(ν1 ) .
2 8π mπ +ν (ν1 + iε)2 (ν1 − iε)2 k2 − ν12
(15.2)

We tried to transform this IE into a DE but because of the ν 2 dependence in the integrand one needs to consider at
least a third order DE, which makes that one cannot find a simple algebraic solution even in the limit k ≫ mπ as done in
Sec. 14 for the 1 S0 wave.
• Then we consider the numerical solution of the IE in Eq. (15.2). In the rest of this section we take mπ as the unit of
energy. Eq. (15.2) is then rewritten as
Z
3πg 2 3gm k−1 ν 2 + ν12 − 1
f(ν) = − (k + ν 2 − 1) − θ(k − 2 − ν) dν1 S2 (ν1 )f(ν1 ) . (15.3)
2 8π 1+ν k2 − ν12
where we have used the function S2 (ν1 ), cf. Eq. (7.101), defined as
1 1
S2 (ν1 ) = 2
+ . (15.4)
(ν1 + iε) (ν1 − iε)2
• For k < 1 there is no contribution from the integral because θ(k − 2 − ν) = 0, since
k − 2 − ν > 0 −→ k − 2 > ν > −k −→ k > 1 . (15.5)
As a result
3πg 2
f(ν) = − (k + ν 2 − 1) , k < 1 . (15.6)
2
• Let us discuss now the case k > 1. For ν = −k the lower limit of integration in Eq. (15.3) is always negative and
corresponds to −k + 1 = −(k − 1), which implies that the integration interval for f(−k) is symmetric around ν1 = 0.
In order to discretize the IE in Eq. (15.3) we perform a change of integration variable to end with integration limits
that remain finite even if k → ∞.
1
x= ,
k − ν1
1
ν1 = k − . (15.7)
x
We discuss the tree cases solved to calculate f(ν) with increasing degree of accurateness to treat the issue of the infrared
divergences.
1
• 1st method: IE of Eq. (15.3) (which is the version of Eq. (7.94) for 1 P1 ). Here we create a partition xi ∈ [ 2k−1 , 1],
i = 1, . . . , N (xi > xi+1 ) for the maximum integration interval ν1 ∈ [−k + 1, k − 1]. We use the Gauss method to calculate
numerically the integral with weights ωi . For a given xi we first determine whether the condition k − 2 − νi > 0 is fulfilled,
with νi = k − x1i . If it is not fulfilled then
3πg 2
f(νi ) = − (k + νi2 − m2π ) . (15.8)
2
For smaller values of νi this condition will be satisfied and then we determine which is the last νj that verifies that
νj > νi + mπ and then
j
3πg 2 3gm X ωk νi2 + νk2 − 1
f(νi ) = − (k + νi2 − m2π ) − S2 (νk )f(νk ) , (15.9)
2 8π k=1 x2k k2 − νk2

152
where the discretized values of the integration variable, according to the same partition, are indicated by νk .
Nevertheless, this first method is not suited for taking numerically the limit ε → 0 in S2 (νk ). We show explicitly below
this conclusion in Fig. 37.
• 2nd method: For ν + 1 < 0 we symmetrize part of the integration interval in Eq. (15.3), from ν + 1 to −(ν + 1),
and leave the rest of the integration untouched, from −(ν + 1) up to k − 1. This can be done because the kernel in the IE
Eq. (15.3) is symmetric under the exchange ν ↔ −ν. Similar steps were taken when obtaining Eq. (7.99) from Eq. (7.94).
Z
3πg 2 3gm −1−ν ν 2 + ν12 − 1
f(ν) = − (k + ν 2 − 1) − θ(k − 2 − ν) dν1 S2 (ν1 )f(ν1 )
2 8π 1+ν k2 − ν12
Z
3gm k−1 ν 2 + ν12 − 1
− θ(k − 2 − ν) dν1 S2 (ν1 )f(ν1 )
8π −1−ν k2 − ν12
Z
3πg 2 3gm −1−ν ν 2 + ν12 − 1
=− (k + ν 2 − 1) − θ(k − 2 − ν) dν1 S2 (ν1 )(f(ν1 ) + f(−ν1 ))
2 8π 0 k2 − ν12
Z
3gm k−1 ν 2 + ν12 − 1
− θ(k − 2 − ν) dν1 S2 (ν1 )f(ν1 ) . (15.10)
8π −1−ν k2 − ν12

In order to perform the numerical integration at ν1 = 0 we make use of the identity in Eq. (7.109), although now for a
P wave we only need to add and subtract f(0). Then Eq. (15.10) becomes
Z
3πg 2 3gm −1−ν ν 2 + ν12 − 1
f(ν) = − (k + ν 2 − 1) − θ(k − 2 − ν) dν1 S2 (ν1 )(f(ν1 ) + f(−ν1 ) − 2f(0))
2 8π 0 k2 − ν12
Z −1−ν Z
3gm ν 2 + ν12 − 1 3gm k−1 ν 2 + ν12 − 1
− θ(k − 2 − ν) f(0) dν1 2 S 2 (ν 1 ) − θ(k − 2 − ν) dν 1 S2 (ν1 )f(ν1 ) .
4π 0 k 2 − ν1 8π −1−ν k2 − ν12
(15.11)

In this way the integrand in the first integral is finite for ν1 → 0 and poses no problems for its numerical evaluation, while
the integral multiplied by f(0) can be done algebraically with the result
Z  
−1−ν ν 2 + ν12 − 1 1 k+1+ν
dν1 2 2 S2 (ν1 ) = 3 −2k + 2kν + (1 − k2 − ν 2 ) log . (15.12)
0 k − ν1 k k−1−ν

Inserting back this expression into Eq. (15.11) we obtain the final IE for the 2nd method:
Z
3πg 2 3gm −1−ν ν 2 + ν12 − 1
f(ν) = − (k + ν 2 − 1) − θ(k − 2 − ν) dν1 S2 (ν1 )(f(ν1 ) + f(−ν1 ) − 2f(0))
2 8π 0 k2 − ν12
 
3gmf(0) 2 2 k+1+ν
− θ(k − 2 − ν) −2k + 2kν + (1 − k − ν ) log
4πk3 k−1−ν
Z k−1 2 2
3gm ν + ν1 − 1
− θ(k − 2 − ν) dν1 S2 (ν1 )f(ν1 ) . (15.13)
8π −1−ν k2 − ν12

The calculation of f(0) is straightforward, since it only involves values of ν1 ≥ 1 for which there is no infrared divergences,
and it can be calculated directly making use of Eq. (15.3)
Z
3πg 2 3gm k−1 ν12 − 1
f(0) = − (k − 1) − θ(k − 2) dν1 S2 (ν1 )f(ν1 ) . (15.14)
2 8π 1 k2 − ν12

For the numerical evaluation of Eqs. (15.13) and (15.14) we use again the change of variables of Eq. (15.7). We also
proceed in a way so that the partition around ν = 0 is symmetric in the maximal integration domain that occurs for

153
ν2

k−1

ν2

2+ ν

1+ ν k−2 ν1

Figure 36: Change of integration variables in Eq. (15.17). The integration region is the triangle delimited by the wide
lines.

ν1 ∈ [−k + 1, k − 1]. Our reference partition is taken for ν1 ∈ [0, k − 1]:


1
x ∈ [ , 1] . (15.15)
k
For ν1 ∈ [−k + 1, 0], we just take the negative values of the partition selected for ν1 > 0. This would correspond to the
change of variable
1 1
x= =− ,
−k − ν1 k − |ν1 |
1 1
ν1 = −k − = −(k − ). (15.16)
x |x|

And x ∈ [−1, −1 k ] for the negative integration range from 0 up to −k + 1.


We use the points in the partition from Eq. (15.15) to evaluate f(ν), ν ≥ 0, employing the IE of Eq. (15.3). For
negative ν but ν + 1 > 0 we still use the latter equation while for ν + 1 < 0 we then employ Eq. (15.13), which requires
f(0) evaluated according to Eq. (15.14).
The drawback for the 2nd method stems from the last integral of Eq. (15.13) in the limit ν → −1 because then the
lower limit of integration vanishes as −(1 + ν) → 0.
• 3rd method: We then proceed similarly as done to obtain Eq. (7.103) and in the domain −3/2 < ν < −1/2 we
rewrite Eq. (15.3) in the form that follows after a first iteration
  Z
3πg 2 3g 2 m k−1 dν1 (ν 2 + ν12 − 1)(k2 + ν12 − 1)
f(ν) = − (k + ν 2 − 1) + θ(k − 2 − ν) S(ν1 )
2 2 4 ν+1 k2 − ν12
  Z Z k−1
3gm 2 k−1 dν1 (ν 2 + ν12 − 1) dν2 (ν12 + ν22 − 1)
+ θ(k − 2 − ν) S(ν 1 )θ(k − 2 − ν 1 ) S(ν2 )f(ν2 ) . (15.17)
8π ν+1 k2 − ν12 ν1 +1 k2 − ν22

154
We now exchange the order of the integration variables ν1 ↔ ν2 in the last integral (see Fig. 36). Let us notice also
that in this integral one has to fulfill k − 2 > ν1 and ν1 > ν + 1 which in turn implies that k − 3 > ν. It then results
  Z
3πg 2 3g 2 m k−1 dν1 (ν 2 + ν12 − 1)(k2 + ν12 − 1)
f(ν) = − (k + ν 2 − 1) + θ(k − 2 − ν) S(ν1 )
2 2 4 ν+1 k2 − ν12
  Z Z ν2 −1
3gm 2 k−1 dν2 S(ν2 ) dν1 (ν 2 + ν12 − 1)(ν12 + ν22 − 1)S(ν1 )
+ θ(k − 3 − ν) 2 f(ν 2 ) . (15.18)
8π 2
ν+2 k − ν2 ν+1 k2 − ν12
The last integral can be done algebraically with the result
Z
dν1 (ν 2 + ν12 − 1)(ν12 + ν22 − 1)S(ν1 )
ν2 −1 2 h
F(ν2 ) = = k(2 + ν − ν2 )(k2 − (ν − 1)(ν2 + 1))
ν+1 k2 − ν12 k3
 
ν+1 ν2 − 1
− (k2 + ν 2 − 1)(k2 + ν22 − 1) arctanh − arctanh . (15.19)
k k
The particularization of this expression for ν2 = k corresponds to the first integral in Eq. (15.18). In terms of the function
F(ν2 ) we can rewrite Eq. (15.18) as
 2  2 Z k−1
3πg 2 3g m 3gm dν2 S(ν2 )
f(ν) = − (k + ν 2 − 1) + θ(k − 2 − ν) F(k) + θ(k − 3 − ν) F(ν2 )f(ν2 ) .
2 2 4 8π ν+2 k2 − ν22
(15.20)

For the other intervals of ν we proceed in the same way as in the 2nd method. That is, for ν > −1/2 we use Eq. (15.3)
and for ν < −3/2 we employ Eq. (15.13).
• Once f(ν) is obtained we calculate ∆(A) according to

f(−k)
∆(A) = − , (15.21)
2k4
cf. Eq. (7.79).
We now show in Figs. 37 and 38 the resulting ∆(A) for several values of gA , namely, gA =1.32, 2, 5 and 10. The red
crosses are the 3rd method, the green diagonal crosses the 2nd method and the blue bursts the first one. The latter is only
shown for the smallest gA = 1.32, since with increasing gA its associated numerical instabilities become more severe. We
also show by the magenta dashed lines the once-iterated result for ∆(A) (which stems from the first two terms on the rhs
of Eq. (15.20)). This is exact up to k = 3/2, but the larger gA and k are, the larger the disagreement with the full result
is. The 3rd and 2nd method show a rapid convergence with ε → 0 and the results shown correspond to ε = 10−6 , while
the 1st method is shown for ε = 10−2 and we were not able to go to smaller values of ε in this case. Even for such values
of ε the 1st method becomes numerically unstable for larger k, as shown in Fig. 37. The 2nd and 3rd methods agree very
well between each other in the calculation of ∆(A), with only some visible differences in the scale of the figures in the last
panel for gA = 5. The number of Gauss points used to calculate ∆(A) for each value of A is 200 × 20, though the results
for the 2nd and 3rd method converge fast with increasing number of Gauss points.
The difference between the 2nd and 3rd method is clear if we plot f(ν) because, as discussed above, the 2nd method
presents instabilities for ν around −1. To show it we plot in Figs. 39 and 40 f(ν) for k = 2.1886 and ν ∈ [−k + 1, k − 1].
The instability of the 2nd method at around ν = −1 is clearly seen in these figures, but it is also notorious that in average
these instabilities tend to cancel and this is why in the evaluation of ∆(A) no significant differences are seen between the
2nd and 3rd methods. Another remarkable fact is the presence of a kink in f(ν). The reason is clear from our analysis in
Sec. 14, which shows that the derivative of f(ν) is discontinuous at ν = k − 2, cf. Eq. (14.4), and this is reason for this
kink at k − 2 = 0.1886. The fact that for the larger values of gA the function ∆(A) becomes oscillating has a reflection
in f(ν) as well, since it develops a more complicated ν dependence with increasingly larger changes, even of sign.

155
gA=1.32 gA=1.32
2 0.5
3rd
0.45 2nd
0 1st
0.4
Once
-2 0.35
0.3
-4
0.25
-6 0.2
3rd
2nd 0.15
-8 1st
Once
0.1
-10 0.05
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2 2.5 3 3.5 4 4.5 5

gA=1.32 gA=1.32
1.1 0.1
3rd
1 0.05
2nd
0.9 1st 0
0.8 Once
-0.05
0.7
-0.1
0.6
-0.15
0.5
-0.2
0.4
0.3 -0.25 3rd

0.2 -0.3 2nd


1st
0.1 -0.35 Once

0 -0.4
5 6 7 8 9 10 10 15 20 25 30 35 40

gA=2 gA=2
5 1.2
3rd
0 1 2nd
Once
-5 0.8

-10 0.6

-15 0.4

-20
3rd 0.2
2nd
Once
-25 0
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2 2.5 3 3.5 4 4.5 5

gA=2 gA=2
0.065 0.024
0.06 3rd 0.022 3rd
2nd 2nd
0.055 Once 0.02 Once
0.05 0.018
0.016
0.045
0.014
0.04
0.012
0.035
0.01
0.03 0.008
0.025 0.006
0.02 0.004
0.015 0.002
5 6 7 8 9 10 10 15 20 25 30 35 40

Figure 37: ∆(A) for 1 P1 as a function of k in units of mπ = 1, for gA = 1.32 (top) and gA = 2 (bottom). In all cases the
red crosses correspond to the 3rd method, the green diagonal crosses is the 2nd method and the blue bursts (only shown
for gA = 1.32) the 1st method. The dashed line is the once-iterated perturbative result.
156
gA=5 gA=5
150 40
20
100
0
50
-20
0 -40
-60
-50
3rd -80
3rd
-100 2nd 2nd
-100
Once Once
-150 -120
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2 2.5 3 3.5 4 4.5 5

gA=5 gA=5
0.05 0.003

0 0.0025

0.002
-0.05
0.0015
-0.1
0.001
-0.15
0.0005
3rd 3rd
-0.2 2nd 0
2nd

-0.25 -0.0005
5 6 7 8 9 10 10 15 20 25 30 35 40

gA=10 gA=10
2000 20000
1000 15000
0 10000
-1000
5000
-2000
0
-3000
-5000
-4000
-5000 -10000
3rd 3rd
-6000 2nd -15000 2nd
Once Once
-7000 -20000
0.6 0.8 1 1.2 1.4 1.6 1.8 2 2 2.5 3 3.5 4 4.5 5

gA=10 gA=10
1000 0.12
3rd
0 0.1 2nd
-1000 0.08

-2000 0.06

-3000 0.04

-4000 0.02

-5000 3rd 0
2nd
-6000 -0.02
5 6 7 8 9 10 20 40 60 80 100 120 140

Figure 38: ∆(A) as a function of k in units of mπ , for gA = 5 (top) and gA = 10 (bottom). For the meaning of the
symbols see Fig. 37.

157
Detail f(ν), gA=1.32, k=2.1886 Mπ Detail f(ν), gA=2, k=2.1886 Mπ
-8.5 -18

-9 -20

-9.5
-22

-10
-24

-10.5
-26
-11

-28
-11.5

-30
-12

-12.5 -32

2nd 2nd
3rd 3rd
-13 -34
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5

Figure 39: f(ν) for 1 P1 as a function of ν in units of mπ = 1 at k = 2.1886 for gA = 1.32 (left) and gA = 2 (right). The
red solid line is the 3rd method and the green dashed line corresponds to the 2nd method.

Detail f(ν), gA=5, k=2.1886 Mπ Detail f(ν), gA=10, k=2.1886 Mπ


0 16000

14000
-50

12000
-100
10000

-150
8000

-200 6000

4000
-250

2000
-300
0

-350
-2000
2nd 2nd
3rd 3rd
-400 -4000
-1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5

Figure 40: f(ν) for 1 P1 as a function of ν in units of mπ = 1 at k = 2.1886, for gA = 5 (top) and gA = 10 (bottom). For
the meaning of the symbols see Fig. 39.

158
• Let us now compare the phase shifts obtained by the solving the LS equation and N/D method. The phase shifts
should be the same because the 1 P1 potential is an ordinary one.
LS equation:
The potential v(p, p′ ) and LS equation are given in Eqs. (15.1) and (2.1), respectively. We rewrite the latter as in Eq. (2.4)
to make it available for a numerical treatment. To have a finite interval of integration we use the change of variable
1
p = −αmπ + ,
x
1
x= , (15.22)
αmπ + p
In this way

x ∈ [0, α] for p ∈ [0, ∞] . (15.23)

For our present analysis we take in the following α = 1 (recall that mπ = 1 too, but we have shown it explicitly in the
previous equation for clarity). We introduce a partition xi , i = 1, . . . , N , whose weights ωi correspond to the application
of the Gauss method for integration. Theses value of xi correspond to pi = −1 + 1/xi in the variable p. For each of these
2
N points we have the T matrix element t(pi , k; km + iε), and additionally we also need the on-shell T matrix element
2
t(k, k; km + iε) that enters explicitly in Eq. (2.4). In this way, we have a linear system of N + 1 equations that we discretize
in the form
N
( )
X m ωj p2j k2 k2
δij − 2 2 2 v(p i , p j ) t(p j , k; + iε) + v(p i , k)σ k t(k, k; + iε) = v(pi , k) ,
j=1
2π xj pj − k2 m m
N
k2 m X ωj p2j k2
[1 + σk v(k, k)] t(k, k; + iε) − 2 v(k, p j )t(p j , k; + iε) = v(k, k) , (15.24)
m 2π j=1 x2j p2j − k2 m

with
N
mk2 X ωj 1 imk
σk = 2 2 2 2
− . (15.25)
2π j=1 xj pj − k 4π

N/D method: We follow the notation in Ref. [3]


Z
A L ∆(q 2 )D(q 2 )
D(A) = 1 + dq 2 g(A, q 2 ) , (15.26)
π −∞ q2
Z
1 ∞ 2 ρ(r 2 ) im/4π
g(A, q 2 ) = dr 2 2 2
=√ p ,
π 0 (r − A)(r − q ) A + i0+ + q 2 + i0+
Z
A L ∆(q 2 )D(q 2 )
N (A) = dq 2 , (15.27)
π −∞ q 2 (q 2 − A)
with L introduced in Eq. (12.20). Notice that there is no free parameter because of the vanishing of the T matrix at
k = 0, which fixes ν1 = 0. To solve numerically the linear IE for D(A), Eq. (15.26), we make the change of variable
1
x= , (15.28)
q2
so that in the domain of integration q 2 ∈ [−∞, L]

x ∈ [1/L, 0] . (15.29)

159
A partition is taken with N Gauss points, xi , i = 1, . . . , N and weights ωi corresponding to the Gauss method for
integration. We then have the linear system of N equations
N
!
X q 2 wj ∆(qj2 )g(qi2 , qj2 )
δij − i 2 D(qj2 ) = 1 . (15.30)
j=1
π xj qj2

To increase the accurateness of the calculation it is convenient numerically to consider M intervals [hℓ−1 , hℓ ], ℓ =
1, . . . , M , with hℓ < hℓ+1 , h0 = L and h(M ) = 0 and an independent partition for each of them. The point is that
by taking just one partition according to the Gauss method for integration one tends to accumulate points mostly in the
lowest part of the domain. This could be remedied by including several intervals whose union is the original integration
domain.

160
Phase Shifts, gA=1.32 Phase Shifts, gA=1.32
0 -7
LS LS
-5 N/D,mt20 -8 N/D,mt20
Once Once
-10 PWA93 -9
δ( P1)

δ( P1)
-15 -10
1

1
-20 -11

-25 -12

-30 -13
0 0.1 0.2 0.3 0.4 0.5 0.5 0.6 0.7 0.8 0.9 1
GeV GeV

Phase Shifts, gA=1.32 Phase Shifts, gA=1.32


15 35
LS LS
N/D,mt20 30 N/D,mt20
10
Once 25 Once
5
20
δ( P1)

δ( P1)
0 15
1

1
10
-5
5
-10
0
-15 -5
1 1.5 2 2.5 3 3.5 4 4.5 5 5 6 7 8 9 10
GeV GeV

Phase Shifts, gA=2 Phase Shifts, gA=2


0 60
LS 50 LS
-5 N/D,mt20 N/D,mt20
Once 40 Once
-10 PWA93 30 N/D,mt300
N/D,mt300
δ( P1)

δ( P1)

20
-15
1

10
-20 0
-10
-25
-20
-30 -30
0 0.1 0.2 0.3 0.4 0.5 0.5 0.6 0.7 0.8 0.9 1
GeV GeV

Phase Shifts, gA=2 Phase Shifts, gA=2


-8 -5.5
LS LS
-10 N/D,mt20 -6 N/D,mt20
-12 N/D,mt300 N/D,mt300
-6.5
-14
δ( P1)

δ( P1)

-7
-16
1

-7.5
-18
-20 -8

-22 -8.5
-24 -9
1 1.5 2 2.5 3 3.5 4 4.5 5 5 6 7 8 9 10
GeV GeV

Figure 41: Phase shifts δ(1 P1 ) for gA = 1.3 (4 top panels) and gA = 2 (4 bottom panels): LS (red solid line), N/D (green
dashed line), perturbative once-iterated result for ∆(A) (blue short-dashed line) and PWA93 (magenta dotted line), which
is shown only in the lowest-energy region panel. In the last two panels we do not show the perturbative once-subtracted
phase shifts because they are so different that would not allow
161 to compare between the N/D and LS calculations since it
would involve a different scale.
Phase Shifts, gA=5 Phase Shifts, gA=5
100 70
80 60
60 50
40 40
20 LS
δ( P1)

δ( P1)
30 N/D,mt200
0
1

1
-20
20 Once
LS 10 N/D,mt20
-40 N/D,mt200
-60 Once 0
-80 N/D,mt20 -10
-100 -20
0 0.1 0.2 0.3 0.4 0.5 0.5 0.6 0.7 0.8 0.9 1
GeV GeV

Phase Shifts, gA=5 Phase Shifts, gA=5


100 100
80 80
60 60
40 40
LS
20 LS 20 N/D,mt200
δ( P1)

δ( P1)
0 N/D,mt200 0 Once
Once
1

1
-20 -20 N/D,mt20
N/D,mt20
-40 -40
-60 -60
-80 -80
-100 -100
1 1.5 2 2.5 3 3.5 4 4.5 5 5 6 7 8 9 10
GeV GeV

Phase Shifts, gA=10 Phase Shifts, gA=10


100 90
80 LS,mt200 80 LS,mt200
N/D,n19 N/D,n19,mt20
60 Once 70 Once
40 LS,mt20 60 LS,mt0
20 N/D,mt200 N/D,mt200
δ( P1)

δ( P1)

50
0
1

40
-20
-40 30
-60 20
-80 10
-100 0
0 0.1 0.2 0.3 0.4 0.5 0.5 0.6 0.7 0.8 0.9 1
GeV GeV

Phase Shifts, gA=10 Phase Shifts, gA=10


100 100
80 80
60
60
40
20 40
δ( P1)

δ( P1)

0 20
1

-20 LS,mt200 0 LS,mt200


-40 N/D,n19,mt20 N/D,n19,mt20
Once -20 Once
-60
LS,mt20 -40 LS,mt20
-80 N/D,mt200 N/D,mt200
-100 -60
1 1.5 2 2.5 3 3.5 4 4.5 5 5 6 7 8 9 10
GeV GeV

Figure 42: Phase shifts δ(1 P1 ) for gA = 5 (4 top panels) and gA = 10 (4 bottom panels). LS with 200 × 20 Gaussian
points (red solid line), N/D with 19 partitions with 20 × 20 Gaussian points each partition (green dashed line), perturbative
once-iterated result for ∆(A) (blue short-dashed line), LS with 20 × 20 Gaussian points (magenta dotted line), N/D with
200 × 20 Gaussian points (cyan dot-dashed line). 162
We show in Figs. 41 and 42 the resulting phase shifts for gA = 1.32, 2, 5 and 10. For gA = 1.32 and 2 we observe
that both methods give the same result in the whole momentum range shown. In the later case one has to increase the
number of Gaussian points, so that the cyan dot-dashed lines result by taking 300 × 20 Gaussian points and then there
is good agreement with the LS result in the highest momentum region as well. For gA = 10 we also compare several
calculations for LS and N/D with different numerical precision. In the highest momentum region all the results in this
case depend on the numerical precision, but for k . 1 GeV all the methods agree. We think that the discrepancy between
the N/D and LS calculations in the higher momentum region for gA = 10 and solving the N/D method with 19 partitions
and 20 × 20 Gaussian points each is a consequence of the limitation in the calculation of ∆(A) and not due to the lack
of numerical precision in the solution of the N/D method in this case. We have also checked that the result for gA = 5
with 200 × 20 Gauss points is stable if we increase to 300 × 20 points. To improve computation of ∆(A) an extrapolation
method between discretized points should be explored.

16 Application of the formalism to N N 3P0


• Here we follow the derivation in Sec. 7.4 employing the standard definitions for the functions v̂(k, k′ ) and t̂(k, k′ ; z),
Eq. (7.126), that we reproduce here for completeness:
ℓ+1
v̂(k, k′ ) = kℓ+1 k′ v(k, k′ ) ,
ℓ+1
t̂(k, k′ ; z) = kℓ+1 k′ t(k, k′ ; z) . (16.1)
We also use the functions ∆v̂(ν, ν1 ) and f(ν1 ), cf. Eqs. (7.127) and (7.128), respectively, which definitions are also
reproduced here
∆v̂(ν, ν1 ) = Imv̂(iν + ε− , iν1 + ε) − Imv̂(iν + ε+ , iν1 + ε) , (16.2)
f(ν) = Imt̂(iν + ε − δ, ik + ε) − Imt̂(iν + ε + δ, ik + ε) , (16.3)
In terms of them we have the general IE Eq. (7.94),
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = ∆v̂(ν, k) + + ∆v̂(ν, ν1 )f(ν1 ) . (16.4)
4π 2 ν+mπ k2 − ν12 (iν1 + ε)n (iν1 − ε)n

16.1 OPE approximation to the potential


In this Section we consider only the OPE approximation to the potential v(k, k′ ).
• The OPE partial-wave potential in the 3 P0 is
Z +1 h i
′ 1 2
v(k, k ) = dt (k2 + k′ )t − 2kk′ Vσq (q2 ) ,
2 −1
 2
gA 1
Vσq (q2 ) = − . (16.5)
2fπ q 2 + m2π
where we have used the equations for the partial-wave projection of off-shell potentials given in Ref. [4]. Notice that
Epelbaum et al. introduce a factor 4π for the partial-wave projected potential and a minus sign in the potential (e.g.
compare Vσq (q2 ) above with WT (q2 ) given by Kaiser et al. [5]). This is already taken into account in the expression for
v(k, k′ ) in Eq. (16.5). Performing the angular integration we end with
gA 2 h ih    i
′2 2 ′2
v(k, k′ ) = − (k 2
− k ) + (k 2
+ k )m 2
π log [k + k ′ 2
] + m 2
π − log [k − k ′ 2
] + m 2
π ,
32fπ2 (kk′ )2
g2 h 2 2
ih    i
v̂(k, k′ ) = − A 2 (k2 − k′ )2 + (k2 + k′ )m2π log [k + k′ ]2 + m2π − log [k − k′ ]2 + m2π . (16.6)
32fπ

163
In the OPE case because ν < kπ − mπ and ν < ν1 − mπ we have that

∆v̂(ν, ν1 ) = −2ρ(ν 2 , ν12 ) , ν1 ∈ [ν + mπ , k − mπ ] .


∆v̂(ν, k) = −2ρ(ν 2 , k2 ) , ν ∈ [−k, k − mπ ] . (16.7)

where ρ(ν 2 , ν12 ), cf. Eq. (7.56), is a polynomial such that

χρ(ν 2 , ν12 ) . (16.8)

It corresponds to Imv̂(iν1 , iν2 ) according to Tables 3 and 4. We can then rewrite Eq. (16.4) as
Z  
θ(k − 2mπ − ν)m k−mπ dν1 ν12 1 1
f(ν) = −2ρ(ν 2 , k2 ) − + ρ(ν 2 , ν12 )f(ν1 ) . (16.9)
2π 2 ν+mπ k2 − ν12 (iν1 + ε)n (iν1 − ε)n

In the present case it follows from Eq. (16.6) it follows that

πgA2 h i
ρ(ν 2 , ν12 ) = − (ν − ν 1 )2
− (ν 2
+ ν 2
1 )m 2
π . (16.10)
32fπ2

• We then have the IE XXXaquiXXX

164
1e+25
MD.NonPer,NP=362,Nt=16000
MD.NonPer,NP=82,Nt=400
MD.NonPer,NP=202,Nt=400
1e+20 MO.NonPer,NP=1000,Nt=4000
MO.NonPer,NP=80000,Nt=1600
MO.NonPer,NP=80000,Nt=1600

1e+15
∆(p2)

1e+10

100000

1e-05
1 10 100 1000 10000
|p|(Mπ)

Figure 43: Comparison between nonperturbative solutions based on MD and MO for −V (3 P0 ). Both methods agree if
enough number of points are used within the MO.

17 Application of the formalism to N N −V (3P0 )


In this section we discuss several N/Dm1 m2 IEs applied to −V (3 P0 ) whose subtraction constants (if any) are fixed in terms
of the ERE for an uncoupled P wave.

17.1 ∆(A) for this PW


Let us first discuss the discontinuity ∆(A) calculated for this case and shown in Fig. 43. The two methods MO and MD
agree very well in the range shown up to 104 mπ . Nonetheless, for large values of |p| one has to increase enough the
number of points (Np ) used in MO to evaluate ∆(A) for every A ≤ L, as shown by the difference between the magenta
points and the rest of points.

17.2 N/D01 , N/D11 and N/D12


The N/D12 was discussed in Sec. 12.4, and a specific discussions regarding the OPE −V (3 P0 ) were given. All the three
cases give rise to the same phase shifts, as shown in Fig. 44, which perfectly agree with those obtained from the solution of
the LS equation. All the curves overlap each other. The cases N/D01 and N/D11 are clear because of the P -wave nature
of the problem. For the N/D12 the reason is understood as explained in Sec. 12.4, see discussion around Eq. (12.38).
We also compare for the N/D01 the method to obtain D(A) point by point by discrtizing the IE and the one based in

165
0
LS
N/D01
-5 N/D11
N/D12

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 44: Phase shifts for N/D01 , N/D11 and N/D12 as well as from the LS equation. All the curves perfectly agree.

the expansion in LP, discussed in Sec. 13. In this case the relevant function to be expanded along the LHC is
1
g(x, y) = q q , (17.1)
1 1
x + y

which coincides with G(x, y) in Eq. (13.3) except for a global factor independent of x. For a given number of orthogonal
polynomials in the expasion,
n
X
gn (x, y) = P2i (x)ci , (17.2)
i=0

we compute the relative mean square error for several values of y,


R1
0 dx(g(x, y) − gn (x, y))2
Mn (y) = R1 . (17.3)
2
0 dxg(x, y)

In the expansion indicated in Eq. (17.2) we use both Legendre (LP) and Chebyshev (CP) polynomials. We also make
√ √
the expansion for the variants g(x, y) x and g(x, y)/ x. The result of this exercise is shown in Fig. 45 which is a log-log
plot. To distinguish between these three cases we introduce a label “fac” equal to 1 for the original function, x1/2 when

multiplied by x and x−1/2 when divided. The panels on the top row corresonds to fac=x−1/2 , that is, to the expansion

of g(x, y)/ x. In the panel on the top left we observe that for the two points closer to 1 the error diminishes up to
n = 500, for n = 550 it is still very small but then it increases much more for n ≥ 600 [in the figure n corresponds to
npg (Legendre) or npc (Chebyshev)]. The same is done for the expansion in Chebyshev polynomials in the top left panel.

166
npg=50 npg=250 npg=450 npg=650 npg=850 npc=50 npc=250 npc=450 npc=650 npc=850
npg=100 npg=300 npg=500 npg=700 npg=900 npc=100 npc=300 npc=500 npc=700 npc=900
npg=150 npg=350 npg=550 npg=750 npg=950 npc=150 npc=350 npc=550 npc=750 npc=950
npg=200 npg=400 npg=600 npg=800 npg=1000 npc=200 npc=400 npc=600 npc=800 npc=1000

fac=x-1/2 ; LP fac=x-1/2 ; CP
1 1

0.1
0.1
Relative Mean Square Error

Relative Mean Square Error


0.01
0.01

0.001
0.001
0.0001

0.0001
1e-05

1e-05
1e-06

1e-07 1e-06
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x x

npg=50 npg=250 npg=550 npg=50 npg=300 npg=600


npg=100 npg=300 npg=600 npg=100 npg=400 npg=700
npg=150 npg=400 npg=700 npg=150 npg=500
npg=200 npg=500 npg=800 npg=200 npg=550

fac=1 ; LP fac=x+1/2 ; LP
0.01 0.01

0.001
0.001
Relative Mean Square Error

Relative Mean Square Error

0.0001
0.0001

1e-05
1e-05
1e-06

1e-06
1e-07

1e-07
1e-08

1e-08 1e-09
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x x

Figure 45: Relative mean square error Mn (y) defined in Eq. (17.3) for the expansion of g(x, y) of Eq. (17.1).

167
The same qualitative behavior is observed, though the error is larger than in the Legendre polynomial expansion. It is also
interesting to note that for the other three points the error do not get worse with increasing n but, in any case, they are
not either (very) small, particularly for the two smallest points y → 0.
Now let us move to the expansion of the original function g(x, y). In this case, the same behaviour as already discussed
is observed for the two points closer to 1. The intermediate point now reaches a small error for n around 500 (getting
larger for n ≥ 600). Similarly, the last two points near zero now have very small errors for n up to 550 (particularly for
n ≤ 500) and then they become quickly much larger for higher values of n. A similar discussion as the previous one holds

for fac=x1/2 , or g(x, y) x. All the errors are small for n up to 550 (specially for n ≤ 500), and then they increase fast.
We also show the actual expansion gn (x, y) compared to the exact g(x, y) as a function of x ∈ [0, 1] in Fig. 46 for the
same five values of y as before. This is done for n = 500 and n = 700, although inspectioning by eye the different curves
it is not easy to dilucidate the worsening of the reproduction of g(x, y) by the expansion gn (x, y), as we have been able to
qualify numerically by using Mn (y).
From these discussions on regards Fig. 45 we learn that we should trust our method based on the expansion on Legendre
or Chebyshev polynomials up to including a maximum of around 550 terms in the expansion. However, this is not what
is observed in the solution of D(A) because convergent and improving results are obtained for larger values of n
as well. A possible explanation to be taken with a grain of salt is that the failure of the LP or CP expansion to faithfully
reproduce g(x, y) starts from x = 1 and then towards lower values. This happens in a way such that gn (x, y) oscillates
quickly around the true g(x, y) (that it would be the average value of such oscillation), with an increasing amplitude of
the oscillations as n increases. However, if gn (x, y) is present in an integration with a more slowly oscillating function (or
not oscillatory at all) then its integral is the same as the one obtained by employing its mean value, the true g(x, y). This
is probably the reason why we can get the same or improved D(A) while using very high values of n, e.g n = 1000 ≫ 500.
However, this fact is not reflected when calculating the relative mean square error, Eq. (17.3), because in this quantity
always the squared of the difference appears, which is then the integration of positive definite function.

168
1:Exact 1:npg=500 1:npg=700 1:Exact 1:npc=500 1:npc=700
2:Exact 2:npg=500 2:npg=700 2:Exact 2:npc=500 2:npc=700
3:Exact 3:npg=500 3:npg=700 3:Exact 3:npc=500 3:npc=700
4:Exact 4:npg=500 4:npg=700 4:Exact 4:npc=500 4:npc=700
5:Exact 5:npg=500 5:npg=700 5:Exact 5:npc=500 5:npc=700

fac=1 ; LP fac=1 ; CP
10 10

1 1
Relative Mean Square Error

Relative Mean Square Error


0.1 0.1

0.01 0.01

0.001 0.001

0.0001 0.0001

1e-05 1e-05
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x x

1:Exact 1:npg=500 1:npg=700 1:Exact 1:npg=500 1:npg=700


2:Exact 2:npg=500 2:npg=700 2:Exact 2:npg=500 2:npg=700
3:Exact 3:npg=500 3:npg=700 3:Exact 3:npg=500 3:npg=700
4:Exact 4:npg=500 4:npg=700 4:Exact 4:npg=500 4:npg=700
5:Exact 5:npg=500 5:npg=700 5:Exact 5:npg=500 5:npg=700

-1/2 1/2
fac=x ; LP fac=x ; LP
10 10

1
1
0.1
Relative Mean Square Error

Relative Mean Square Error

0.01
0.1
0.001

0.01 0.0001

1e-05
0.001
1e-06

1e-07
0.0001
1e-08

1e-05 1e-09
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x x

Figure 46: Comparison between g(x, y) and its expansion gn (x, y) in LP and CP as a function of x for five values of y.
The value of y decreases in the curves from top to bottom.

Next we start with the solution obtained by applying the method of Sec. 13. We consider various values of the cutoff
and for each of them we vary the number of LP or CP in the expansion and study the convergence as we increase them
for both the phase shifts and the function D(A) along the LHC. The phase shifts are shown in Figs. 47-49, while D(A) is
discussed in Figs. 50-52.

169
n=10 n=200 n=400 n=600 n=800
n=50 n=250 n=450 n=650 Full N/D01
n=100 n=300 n=500 n=700
n=150 n=350 n=550 n=750

L=1e1 ; mt=4000
5

-5

-10

-15

-20
δ(p)

-25

-30

-35

-40

-45

-50
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 47: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 10mπ .

170
n=100 n=400 n=500 , L=1e1
n=200 n=500 Full N/D01
n=300 n=600

L=1e2 , mt=4000
5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 48: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 102 mπ .

171
n=100 n=400 n=700 n=500 , L=1e2
n=200 n=500 n=800 Full N/D01
n=300 n=600 n=500 , L=1e1

L=1e3 , mt=4000
5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 49: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 103 mπ .

172
n=10 n=200 n=400 n=600 n=800
n=50 n=250 n=450 n=650 Full N/D01
n=100 n=300 n=500 n=700
n=150 n=350 n=550 n=750

L=1e1 ; mt=4000
1

0.9

0.8

0.7

0.6
∆(p2)

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

Figure 50: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 10mπ .

173
n=100 n=400 n=500 , L=1e1
n=200 n=500 Full N/D01
n=300 n=600

L=1e2 , mt=4000
1

0.9

0.8

0.7

0.6
∆(p2)

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

Figure 51: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 102 mπ .

174
n=100 n=400 n=700 n=500 , L=1e2
n=200 n=500 n=800 Full N/D01
n=300 n=600 n=500 , L=1e1

L=1e3 , mt=4000
1

0.9

0.8

0.7

0.6
∆(p2)

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

Figure 52: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 103 mπ .

We do not get convergence when ∆(A) is calculated with a cutoff of 104 mπ and with 82 Gaussian points for each
point in which ∆(A) is calculated. This comes out independently of the number of points that we used along the LHC
(4, 8 or 16×103 ). However, we do get convergence when using the nearby Λ = 7 × 103 mπ but with 322 Gaussian points
for each point in which ∆(A) is calculated. This is shown in Fig. 53 for phase shifts and in Fig. 54 for the function
D(A) and using 16 × 103 Gaussian points along the LHC. We also show in these figures the results obtained by making
the expansion with CP. They perfectly agree with the ones obtained by making the expansion in LP. The same happens

when using g(x, y)/ x instead of g(x, y), as shown in the figures. This result indicates that when moving to values of
the cutoff around 104 mπ one needs to know ∆(A) with enough precision by using a large enough number of Gaussian
points at every point in which ∆(A) is calculated. In fact, we have also evaluated the phase shifts and D(A), A < L, for
Λ = 7 × 103 mπ but now with many fewer points along the LHC, namely, for 4 × 103 points. The results perfectly agree
with our calculation with 16 × 103 , as shown in Figs. 55 and 56. We notice that a slight disagreement between the phase
shifts obtained by the straight solution of the N/D01 IE or the LS equation and the solution of the former by expanding in
orthogonal polynomials persists up to including around 500-600 LP. However, we should also notice that the agreement for

175
D(A) is perfect. Indeed, if we increase n to 800 and 1000, also shown in Figs. 57-59 in more detail, we observe that the
phase shifts for n = 800 overlap with the straight solution for N/D01 , and this is almost the same for n = 1000, although
one can appreciate for lower energies a slight worsening compared with n = 800, see Fig. 57. However, the function D(A)
along the LHC gets a bit worse than for n . 600, as expected from the discussion in relation with the relative mean
quadratic error Mn (y), cf. Fig. 45. Indeed this is also the case if we consider the expansion along the RHC of the function
1
r(x, A) = √ q (17.4)
A + i − Lx

and evaluate the associated relative mean square radius, plotted in Fig. 60. From this figure we also see that the expansion
along the RHC of r(x, y) has the smallest relative mean square radius for n . 550. Thus, it is not clear why the phase
shifts for n = 800 indeed overlap the straight solution of N/D01 or the LS equation and it does not happen as occurred
for D(A) along the LHC for which the reproduction was indeed a bit poorer. Probably, it is just an accident.

LP, n=100 LP, n=500 LP,g x-1/2,n=560 LP,n=500,L=1e3


LP, n=200 LP, n=600 CP,g x-1/2,n=560 Full N/D01
LP, n=300 LP, n=800 LP,n=500,L=1e1
LP, n=400 LP, n=1000 LP,n=500,L=1e2

L=7e3 , mt=16000
5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 53: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP and CP, Λ = 7 × 103 mπ .

176
LP, n=100 LP, n=500 LP,g x-1/2,n=560 LP,n=500,L=1e3
LP, n=200 LP, n=600 CP,g x-1/2,n=560 Full N/D01
LP, n=300 LP, n=800 LP,n=500,L=1e1
LP, n=400 LP, n=1000 LP,n=500,L=1e2

L=7e3 , mt=16000
1

0.9

0.8

0.7

0.6
∆(p2)

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

Figure 54: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP and CP, Λ = 7 × 103 mπ .

177
LP, n=100 LP, n=400 LP,n=500,L=1e1 LP, n=500
LP, n=200 LP, n=500 LP,n=500,L=1e2 Full N/D01
LP, n=300 LP, n=600 LP,n=500,L=1e3

L=7e3 , mt=4000
5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 55: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 7 × 103 mπ with 4 × 103
points along the LHC.

178
LP, n=100 LP, n=400 LP,n=500,L=1e1 LP, n=500
LP, n=200 LP, n=500 LP,n=500,L=1e2 Full N/D01
LP, n=300 LP, n=600 LP,n=500,L=1e3

L=7e3 , mt=4000
1

0.9

0.8

0.7

0.6
∆(p2)

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

Figure 56: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 7 × 103 mπ with 4 × 103 points
along the LHC.

179
LP,n=500,L=1e1 LP, np=600 Full N/D01
LP,n=500,L=1e2 LP, np=800
LP,n=500,L=1e3 LP, np=1000

L=7e3 , mt=16000. Higher n


1

-1

-2
δ(p)

-3

-4

-5

-6

-7
0 20 40 60 80 100
|p| (MeV)

Figure 57: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 7 × 103 mπ . The curves
correspond to the higher values of n.

180
LP,n=500,L=1e1 LP, np=600 Full N/D01
LP,n=500,L=1e2 LP, np=800
LP,n=500,L=1e3 LP, np=1000

L=7e3 , mt=16000. Higher n


5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 58: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 7 × 103 mπ . The curves
correspond to the higher values of n.

181
LP,n=500,L=1e1 LP, np=600 Full N/D01
LP,n=500,L=1e2 LP, np=800
LP,n=500,L=1e3 LP, np=1000

L=7e3 , mt=16000. Higher n


0.95

0.9

0.85

0.8
∆(p2)

0.75

0.7

0.65

0.6

0.55
-10 -8 -6 -4 -2 0
2 2
p (mπ )

Figure 59: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, Λ = 7 × 103 mπ . The curves correspond
to the higher values of n


There is no convergence when multiply the function g(x, y), Eq. (17.1), by x or higher powers. This is shown in
Figs. 61 and 62. For definiteness we use a cutoff Λ = 103 mπ with 4 × 103 points along the LHC.

182
npg=50 npg=200 npg=350 npg=500 npg=650 npg=800 npg=950
npg=100 npg=250 npg=400 npg=550 npg=700 npg=850 npg=1000
npg=150 npg=300 npg=450 npg=600 npg=750 npg=900

fac=1 ; LP ; RHC
0.01

0.001

0.0001
Relative Mean Square Error

1e-05

1e-06

1e-07

1e-08
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x

Figure 60: Relative mean square error Mn (A) for the expansion of r(x, A) of Eq. (17.3).

183
g x1/2,n=100 g x1/2,n=400 n=500 , L=1e2
g x1/2, n=200 g x1/2,n=500 n=600 , L=1e3
g x1/2,n=300 n=500 , L=1e1 Full N/D01

L=1e3 , mt=4000 , g x1/2


5

-5

-10

-15
δ(p)

-20

-25

-30

-35

-40
0 50 100 150 200 250 300 350 400
|p| (MeV)
1
Figure 61: Phase shifts for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, g(x, y) → x 2 g(x, y) and
Λ = 103 mπ .

184
g x1/2,n=100 g x1/2,n=400 n=500 , L=1e2
g x1/2, n=200 g x1/2,n=500 n=600 , L=1e3
g x1/2,n=300 n=500 , L=1e1 Full N/D01

L=1e3 , mt=4000 , g x1/2


1.1

0.9

0.8

0.7
∆(p2)

0.6

0.5

0.4

0.3

0.2

0.1
-100 -80 -60 -40 -20 0
2 2
p (mπ )

1
Figure 62: D(A) for −V (3 P0 ) from the solution of the N/D01 in expansion of LP, g(x, y) → x 2 g(x, y) and Λ = 103 mπ .

17.3 N/D23
We continue from the results derived in Sec. 12.6. The IE for D(A) (A < L) using the variables x and y is
3 1 Z √   √ !
aV rV L aV ν2 L2 aV (−L) 2 m(−L) 2 1 x y x + y rV x y
D(y) = 1 + + 2
− 3/2
− dx∆(x)D(x) √ + +√ √ .
2y y y 4π 2 y 5/2 0 −L aV L 2 x+ y
(17.5)

The function G(x, y) is now


√   √
x y x + y rV x y
G(x, y) = √ + +√ √ , (17.6)
−L aV L 2 x+ y

that vanishes as x for x → 0. This implies that we can also expand G(x, y) divided at most up to x,

185
The independent-term function f (y) is
3
aV rV L aV ν2 L2 aV (−L) 2
f (y) = 1 + + − . (17.7)
2y y2 y 3/2
The coefficients ξj , ωj and aij are calculated as
Z 1
ξj = dxD(x)∆(x)P2j (x) ,
0
Z 1
ωj = dxf (x)∆(x)P2j (x) ,
0
Z 1
aij = dx∆(x)D(x)P2i (x)cj (x) . (17.8)
0

pFor the calculation of D(A) along the RHC (A > 0) we have that, from Eq. (17.5) doing the transformation y →
i −L/A, cf. Eq. (13.12),
aV rV 3
D(A) = 1 + A + aV ν2 A2 − iaV A 2
2
Z 1    
mA2 x L/A + x rV x
− 3 dx∆(x)D(x) + + √ ,
4π 2 (−L) 2 0 (−L)1/2 aV L 2 x + i(−L/A)1/2
Z 1    
mA2 x L/A + x rV x
dj (A) = −(4j − 1) 3 dxP2j (x) + + √ . (17.9)
4π 2 (−L) 2 0 (−L)1/2 aV L 2 x + i(−L/A)1/2

The calculation of N (A) we follow the method discussed in Sec. 13 and is also exactly the same as discussed for the
N/D22 case, cf. Eq. (17.19).
Z
4πaV A2 1 1
N (A) = A− dx∆(x)D(x) 2 ,
m Lπ 0 k −A
Z
A2 1 P2j (x)
nj (A) = −(4j − 1) dx 2 . (17.10)
Lπ 0 k −A
Unitarity is also fulfilled, because

m A
Imdj (A) = − nj (A) . (17.11)

• The phase shifts obtained for different values of the cutoff Λ from 10 up to 7 × 103 mπ and using g(x, y) (gx0p0),

g(x, y)/ x (gxm0p5) or g(x, y)/x (gxm1p0) are shown in Fig. 63. The number of Legendre polynomials (LP’s) is indicated
by “n=” and the value of Λ is given in the title of each panel as “L=”, e.g. L=1e1 would correspond to Λ = 10mπ .
Typically in some panels we also introduce results corresponding to other panels which are then shown by small dots. In
their corresponding legends we indicate Λ as e.g. “1e1”, then the number of points, as e.g. “4e3”, and then the number
of LP’s, e.g. “n=600”. Putting all together it would be “1e1,4e3,n=600”. We observe that a convergent result is quickly
reached.
Nonetheless, the strongest convergence properties are obtained as expected for g(x, y)/x as shown in Fig. 64 where we
plot the phase shifts for Λ = 7 × 103 mπ with g(x, y)/x (top panel) and for g(x, y) (bottom panel). For the former we
reach convergence but no convergence results for the latter panel.
The same exercise is performed for the D(A) function along the LHC, A < L, in Fig. 65 for different values of Λ and
in Fig. 66 for the specific value of Λ = 7 × 103 mπ . In all these figures we use 4 × 103 points along the LHC. The same
comments as in Figs. 63 and 64 for the phase shifts are in order.

186
n=50 n=300 n=600 n=900 n=50 n=300 n=600 n=900 ERE
n=100 n=400 n=700 n=1000 n=100 n=400 n=700 n=1000
n=200 n=500 n=800 ERE n=200 n=500 n=800 gxm1p0,n=600

L=1e1 ; mt=4000 ; gxm1p0 L=1e1 ; mt=4000 ; gxm0p5


160 160

140 140

120 120

100 100
δ(p)

δ(p)
80 80

60 60

40 40

20 20

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
|p| (MeV) |p| (MeV)

n=50 n=400 n=800 gxm1p0,n=600 n=50 n=400 n=800 7e3,4e3,n=600


n=100 n=500 n=900 ERE n=100 n=500 n=900 7e3,16e3,n=600
n=200 n=600 n=1000 n=200 n=600 n=1000 ERE
n=300 n=700 gxm0p5,n=600 n=300 n=700 1e1,4e3,n=600

L=1e1 ; mt=4000 ; gx0p0 L=1e2 ; mt=4000 ; gxm1p0


160 160

140 140

120 120

100 100
δ(p)

δ(p)

80 80

60 60

40 40

20 20

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
|p| (MeV) |p| (MeV)

n=50 n=500 n=900 7e3,4e3,n=600 n=50 n=400 n=800 7e3,16e3,n=600


n=200 n=600 n=1000 7e3,16e3,n=600 n=100 n=500 n=900 ERE
n=300 n=700 1e1,4e3,n=600 ERE n=200 n=600 n=1000
n=400 n=800 1e2,4e3,n=600 n=300 n=700 1e1,4e3,n=600

L=1e3 ; mt=4000 ; gxm1p0 L=7e3 ; mt=4000 ; gxm1p0


160 160

140 140

120 120

100 100
δ(p)

δ(p)

80 80

60 60

40 40

20 20

0 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
|p| (MeV) |p| (MeV)

Figure 63: Phase shifts for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 4 × 103 points along the

LHC and different Λ’s. As indicated in the legends one uses either g(x, y) (gx0p0), g(x, y)/ x (gxm0p5) or g(x, y)/x
(gxm1p0). 187
n=50 n=400 n=800 7e3,16e3,n=600
n=100 n=500 n=900 ERE
n=200 n=600 n=1000
n=300 n=700 1e1,4e3,n=600

L=7e3 ; mt=4000 ; gxm1p0


160

140

120

100
δ(p)

80

60

40

20

0
0 50 100 150 200 250 300 350 400
|p| (MeV)

n=50 n=300 n=600 1e1,4e3,n=600


n=100 n=400 n=700 7e3,16e3,n=600
n=200 n=500 n=800 ERE

L=7e3 ; mt=4000 ; gx0p0


180

160

140

120

100
δ(p)

80

60

40

20

0
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 64: Phase shifts for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 4 × 103 points along the
LHC and Λ = 7 × 103 mπ . Top panel g(x, y)/x (gxm1p0), bottom panel g(x, y) (gx0p0).

188
n=50 n=300 n=600 n=900 n=50 n=300 n=600 n=900
n=100 n=400 n=700 n=1000 n=100 n=400 n=700 n=1000
n=200 n=500 n=800 n=200 n=500 n=800 gxm1p0,n=600

L=1e1 ; mt=4000 ; gxm1p0 L=1e1 ; mt=4000 ; gxm0p5


500 500

0 0

-500 -500

-1000 -1000

-1500 -1500
∆(p2)

∆(p2)
-2000 -2000

-2500 -2500

-3000 -3000

-3500 -3500

-4000 -4000
-100 -80 -60 -40 -20 0 -100 -80 -60 -40 -20 0
2
p (mπ2) p2 (mπ2)

n=50 n=400 n=800 gxm1p0,n=600 n=50 n=400 n=800 7e3,4e3,n=600


n=100 n=500 n=900 n=100 n=500 n=900 7e3,16e3,n=600
n=200 n=600 n=1000 n=200 n=600 n=1000
n=300 n=700 gxm0p5,n=600 n=300 n=700 1e1,4e3,n=600

L=1e1 ; mt=4000 ; gx0p0 L=1e2 ; mt=4000 ; gxm1p0


500 1000

0
0
-1000
-500
-2000
-1000
-3000
∆(p2)

∆(p2)

-1500 -4000

-5000
-2000
-6000
-2500
-7000
-3000
-8000

-3500 -9000
-100 -80 -60 -40 -20 0 -100 -80 -60 -40 -20 0
2
p (mπ2) p2 (mπ2)

n=50 n=500 n=900 7e3,4e3,n=600 n=50 n=400 n=800 7e3,16e3,n=600


n=200 n=600 n=1000 7e3,16e3,n=600 n=100 n=500 n=900
n=300 n=700 1e1,4e3,n=600 n=200 n=600 n=1000
n=400 n=800 1e2,4e3,n=600 n=300 n=700 1e1,4e3,n=600

L=1e3 ; mt=4000 ; gxm1p0 L=7e3 ; mt=4000 ; gxm1p0


2000 2000

0 0

-2000 -2000

-4000 -4000
∆(p2)

∆(p2)

-6000 -6000

-8000 -8000

-10000 -10000

-12000 -12000
-100 -80 -60 -40 -20 0 -100 -80 -60 -40 -20 0
p2 (mπ2) p2 (mπ2)

Figure 65: D(A), A ≤ L for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 4 × 103 points along

the LHC and different of the Λ. As indicated in the legends one uses either g(x, y) (gx0p0), g(x, y)/ x (gxm0p5) or
g(x, y)/x (gxm1p0). 189
n=50 n=400 n=800 7e3,16e3,n=600 n=50 n=300 n=600 1e1,4e3,n=600
n=100 n=500 n=900 n=100 n=400 n=700 7e3,16e3,n=600
n=200 n=600 n=1000 n=200 n=500 n=800
n=300 n=700 1e1,4e3,n=600
L=7e3 ; mt=4000 ; gx0p0
L=7e3 ; mt=4000 ; gxm1p0 4000
2000
2000
0
0

-2000 -2000

-4000 -4000

∆(p2)
∆(p2)

-6000
-6000
-8000
-8000
-10000

-10000
-12000

-12000 -14000
-100 -80 -60 -40 -20 0 -100 -80 -60 -40 -20 0
2
p (mπ2) p2 (mπ2)

Figure 66: D(A), A ≤ L for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 4 × 103 points along the
LHC and Λ = 7 × 103 . Left panel (gxm1p0), right panel (gx0p0). The case with gx0p0 does not converge, it corresponds
to the bottom panel of Fig. 64.

In some of the panels we have already included the case that employs Λ = 7 × 103 mπ with 16 × 103 points along the
LHC. This is shown separately in Fig. 67 for the phase shifts and in Fig. 68 for D(A), A ≤ L. For the top panel we focus
in the lower energy region by showing the function D(A) with restricted values of A in the interval −10m2π ≥ A.

n=50 n=400 n=800 1e2,4e3,n=600


n=100 n=500 n=900 7e3,4e3,n=600
n=200 n=600 n=1000 ERE
n=300 n=700 1e1,4e3,n=600

L=7e3 ; mt=16000 ; gxm1p0


160

140

120

100
δ(p)

80

60

40

20

0
0 50 100 150 200 250 300 350 400
|p| (MeV)

Figure 67: Phase shifts for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 16 × 103 points along the
LHC and Λ = 7 × 103 mπ .

190
n=50 n=400 n=800 1e2,4e3,n=600
n=100 n=500 n=900 7e3,4e3,n=600
n=200 n=600 n=1000
n=300 n=700 1e1,4e3,n=600

L=7e3 ; mt=16000 ; gxm1p0


50

-50

-100
∆(p2)

-150

-200

-250

-300
-10 -8 -6 -4 -2 0
p2 (mπ2)

n=50 n=400 n=800 1e2,4e3,n=600


n=100 n=500 n=900 7e3,4e3,n=600
n=200 n=600 n=1000
n=300 n=700 1e1,4e3,n=600

L=7e3 ; mt=16000 ; gxm1p0


2000

-2000

-4000
∆(p2)

-6000

-8000

-10000

-12000

-14000
-100 -80 -60 -40 -20 0
p2 (mπ2)

Figure 68: D(A), A ≤ L for −V (3 P0 ) from the solution of the N/D23 in expansion of LP with 16 × 103 points along the
LHC and Λ = 7 × 103 mπ .

191
• It is also interesting to compare the D(A) for A ≤ L obtained by solving the IE of Eq. (17.5) expanding in LP’s with
the result found by plugging the D(A) solved on the r.h.s of the same IE. This is shown in Fig. 69. In every line we indicate
the type of g(x, y) function (by which factor (if any) is multiplied), the value of the cutoff Λ, number of points along the
LHC in which ∆(A) has been calculated and whether the resulting D(A) is the solution or its output (in/out), in order.
E.g. the legend “m1p0,1e1,4e3,in” refers to the D(A) solved (in) with g(x, y)/x, Λ = 10mπ and 4 × 103 points along
the LHC. We can see that both D(A) match for the top panel, showing the interval −10m2π ≤ A ≤ L, except for the
highest Λ = 17 × 103 mπ . From the bottom panel we deduce that this good agreement persists up to around −20m2π < A,
starting to diverge for lower values of A.
• The relative quadratic error for five values of y = L/A along the LHC is shown in Fig. 70. The values of Λ used
are 7e3, 4e3 and 4e3 from top to bottom and left to right, in order. In the title of each panel we indicate the number
of points along the LHC for which ∆(A) is calculated and the type of g(x, y) function used by the label “fac=xα ”. We
observe that the smallest values of the relative mean quadratic error occur when 16e3 points are used along the LHC. This
is also the case for which the addition of LP’s improve the expansion up to higher values of n (referred in the panels as
“npg”). Nonetheless, it is worth stressing that there is no a clear connection between the relative mean quadratic error
and the numerical convergence of D(A). This is clear because we already got the same convergent solution for values of
n between 100 and 1000, while there is a large change in the values obtained for the relative mean quadratic error.
• In summary, we observe a very nice and convincing convergence of the results worked out numerically for the different
values of Λ employed, namely, from 10 up to 7 × 103 mπ . This is checked both for the phase shifts and D(A) along the
LHC. It is important to look also at D(A) separately so as to avoid any possible diverting factor that could be factorized
out when calculating the phase shifts (“peratization”).

192
m1p0,1e1,4e3,in m1p0,1e2,4e3,in m1p0,7e3,16e3,in
out out out
m0p5,1e1,4e3,in m1p0,1e3,4e3,in
out out
0p0,1e1,4e3,in m1p0,7e3,4e3,in
out out

D(A) in/out
100

50

-50
∆(p2)

-100

-150

-200

-250

-300
-10 -8 -6 -4 -2 0
p2 (mπ2)

m1p0,1e1,4e3,in m1p0,1e2,4e3,in m1p0,7e3,16e3,in


out out out
m0p5,1e1,4e3,in m1p0,1e3,4e3,in
out out
0p0,1e1,4e3,in m1p0,7e3,4e3,in
out out

D(A) in/out

-1000

-2000
∆(p2)

-3000

-4000

-5000
-100 -80 -60 -40 -20 0
p2 (mπ2)

Figure 69: D(A), A ≤ L for −V (3 P0 ) from the solution of the N/D23 in expansion of LP for several values of Λ, number
of points along the LHC in the calculation of ∆(A) and factors diving g(x, y). Every legend follows the scheme: type of
g, Λ, number of points, in/out. Here “in” means the outcome directly obtained by solving the IE expanding in LP’s while
“out” is the D(A) function calculated by substituting the D(A) solved in the r.h.s of the IE.

193
npg=50 npg=200 npg=400 npg=600 npg=800 npg=1000 npg=50 npg=200 npg=400 npg=600 npg=800 npg=1000
npg=100 npg=300 npg=500 npg=700 npg=900 npg=100 npg=300 npg=500 npg=700 npg=900

mt=16000 ; fac=x-1 ; LP ; LHC mt=4000 ; fac=x-1 ; LP ; LHC


1 10

0.1 1

0.01
Relative Mean Square Error

Relative Mean Square Error


0.1

0.001
0.01
0.0001
0.001
1e-05

0.0001
1e-06

1e-07 1e-05

1e-08 1e-06
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1 1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x x

npg=50 npg=200 npg=400 npg=600 npg=800 npg=1000


npg=100 npg=300 npg=500 npg=700 npg=900

mt=4000 ; fac=x0 ; LP ; LHC


10

0.1
Relative Mean Square Error

0.01

0.001

0.0001

1e-05

1e-06

1e-07

1e-08

1e-09
1e-09 1e-08 1e-07 1e-06 1e-05 0.0001 0.001 0.01 0.1 1
x

Figure 70: Relative quadratic error for five values of y = L/A along the LHC. In the legend of each panel we indicate the
number of points along the LHC for which ∆(A) is calculated and the type of g(x, y) function used by the label “fac=xα”.

FORGOTTEN: In the Fig. 71 I show the results with the LP method employing Λ = 7 × 103 mπ with 16 × 103 Gaussian
points along the LHC with the phase shifts obtained by iterating the Neumann series up to 20 iterations in even steps.

194
140

120

100

80

60

40
"fort.22" u 1:2
"fort.24" u 1:2
"fort.26" u 1:2
"fort.28" u 1:2
"fort.30" u 1:2
20 "fort.32" u 1:2
"fort.34" u 1:2
"fort.36" u 1:2
"fort.38" u 1:2
"fort.40" u 1:2
"gxm1p0/des.npg500.mt16000.7e3.a2p89.r1p0.vm1p0.out" u 1:2
0
0 50 100 150 200 250 300 350 400

Figure 71: Phase shifts obtained by iterating OPE (dots) compared to those calculated with the LP method with Λ =
7 × 103 mπ , g(x, y)/x, and 16 × 103 points along the LHC.

17.4 N/D22
The N/D22 general setup was discussed in Sec. 12.5. In order to apply the method of Sec. 13 based on the expansion
in LP’s we rewrite the IE of D(A) along the LHC, Eq. (12.44), in terms of the variables x = L/k2 , y = L/A with x,
y ∈ [0, 1]:
 
3 1 Z 1
aV (−L) 2 aV rV L m(−L) 2  xy 1 
D(y) = 1 − + + dx∆(x)D(x)  +q q  . (17.12)
y 3/2 2y 4π 2 y 2 0 aV (−L)
3
2 1
+ 1
x y

The function G(x, y) is


 
1
m(−L)  2 xy 1 
G(x, y) = 2 2  3 +
q q 
4π y aV (−L) 2 1 1
x + y
X
= cj (y)P2j (x) ,
j
Z 1
cj (y) = (4j − 1) dxG(x, y)P2j (x) . (17.13)
0
√ √
The function G(x, y) vanishes as x for x → 0, so that we can expand this function divided at most by x (referred as
“m0p5” in many legends in the figures that follow, while if the original G(x, y) is used, Eq. (17.13), we employ the label

195
“0p0”). The other needed terms analogously to Eq. (13.8) are
Z 1
ξj = dxD(x)∆(x)P2j (x) ,
0
Z 1
ωj = dxf (x)∆(x)P2j (x) , (17.14)
0

with f (x) the independent term in the IE of Eq. (17.12)


3
aV (−L) 2 aV rV L
f (x) = 1 − 3/2
+ . (17.15)
x 2x
The matrix elements aij are given by
Z 1
aij = dx∆(x)D(x)P2i (x)cj (x) . (17.16)
0

For the calculation of D(A) along the RHC (A > 0), we have from Eq. (12.41),
√ Z  
1 mA 1 1 A
D(A) = 1 − iaV A A + aV rV A − 2 dx∆(x)D(x) + √ √ . (17.17)
2 4π L 0 aV k2 −k2 − i A

Thus, the corresponding R(k2 , A) function in this case, cf. Eq. (13.17), is
 
1
mA A
R(k2 , A) = − +√ √
4π 2 L
aV k2 2
−k − i A
X
= dj (A)P2j (x)
j
Z 1 L
dj (A) = (4j − 1) dxR( , A)P2j (x) . (17.18)
0 x
Although for the calculation of the phase shifts one does not need the function N (A) its calculation is interesting as a
check of unitarity or for the calculation of the shape parameters. The process to calculate N (A) is elaborated in Sec. 13.
We rewrite Eq. (12.41) as
Z
4πaV A2 1 1
N (A) = A− dx∆(x)D(x) 2 ,
m Lπ 0 k −A
Z
A2 1 P2j (x)
nj (A) = −(4j − 1) dx 2 . (17.19)
Lπ 0 k −A
Along the RHC unitarity is fulfilled because

m A
Imdj (A) = − nj (A) . (17.20)


• We plot in Fig. 72 the results (phase shifts and D(A), A < L) obtained with Λ = 10mπ , g(x, y)/ x and 4 × 103
points along the LHC. We obtain convergence when increasing n but this occurs in a much slower way than for the N/D01
and N/D23 IE’s. After some undecided behavior for n ≤ 200 the tendency towards convergence stars which is reached
for n & 500. However, the converged values for the phase shifts depart very strongly from the ERE for values of the
three-momentum much smaller than expected in N N scattering (e.g much smaller than the pion mass). This is clearly

196
indicating that the N/D22 IE is not adequate. We do similarly in Fig. 73 but now using the original function g(x, y). The
conclusions are the same as the ones with respect to Fig. 72 and the results agree between both figures.
• In order to show the typical behavior with increasing n for higher values of Λ we show the phase shifts and D(A),
−10m2π < A < L obtained with Λ = 7 × 103 mπ and with 16 × 103 points along the LHC in Fig. 74. In the same figure
we also show the results obtained with lower values of Λ, 10, 102 and 103 mπ by dots. First, we notice a very weird
pattern of convergence towards the result obtained for the highest values of n (somewhat n ≥ 800), so that the results
with these large indices are disconnected from the ones obtained with the smaller values of n. The latter do resemble to
the ERE result while the former depart strongly even for |p| ≪ mπ . We also show in the figure by “Standard” the solution
of the N/D01 case. Curiously enough we observe a cutoff dependence of the “converged” results but it seems that a slow
convergence occurs with increasing cutoff, so that the solutions agree for Λ = 103 mπ and 7 × 103 mπ and the one for
Λ = 102 mπ is closer to the previous one than to the result for Λ = 10mπ . Again, this pattern of behavior is strongly at
odds with the one observed for the convergent numerical results for N/D01 and N/D23 .

We also do the same in Fig. 75 but using now g(x, y)/ x (which is expected to have better convergent properties),
Λ = 102 mπ and 4 × 103 points along the LHC. The results for the higher n are rather close to the results obtained before
for Λ & 103 mπ and shown in Fig. 74. But again there is a clear disconnection between n . 500 and greater values of
this index, which should not be accepted and that it is contrary to the clear convergent behavior observed for N/D01 and
N/D23 .
• Finally, we compare the function D(A) solution of the IE of Eq. (17.12) by the method of the expansion in LP’s with
the one that results once the former is used in the r.h.s of the IE. This is shown in Fig. 76. In every legend we indicate
first the type of g(x, y) used, Λ and the number of points along the LHC. We observe that only for Λ up to 10mπ there is
agreement between the D(A) solved and the one resulting from the former. This is in contrast with the N/D23 case where
we could reach up to Λ = 103 mπ with matching functions D(A), cf. Fig. 69. This fact indicates the fallacious nature of
the D(A) found for very high values of n by solving the IE in expansion of LP’s, which is not smoothly connected with
the one resulting for n . 500.
• We conclude that the D(A) obtained for n → 103 is not acceptable because of three reasons. The main one is the
lack of smooth connection of this solution found for the higher n with its behavior for the lower n values, being the value
of transition n ≃ 500. A second important reason is that only for the lowest value of Λ (10mπ ) the D(A) solved coincides
with the D(A) deduced. In addition, on more physical grounds, this solutions departs from the ERE for |p| ≪ mπ , which
is extremely strange in N N scattering, a behavior that is not observed in the good solution found for N/D23 .

197
n=50 n=400 n=800
n=100 n=500 n=900
n=200 n=600 n=1000
n=300 n=700 ERE

L=1e1 ; mt=4000 ; gxm0p5


80

70

60

50

40
δ(p)

30

20

10

-10
0 50 100 150 200 250 300 350 400
|p| (MeV)

n=50 n=400 n=800


n=100 n=500 n=900
n=200 n=600 n=1000
n=300 n=700

L=1e1 ; mt=4000 ; gxm0p5


9000

8000

7000

6000

5000
∆(p2)

4000

3000

2000

1000

-1000
-100 -80 -60 -40 -20 0
p2 (mπ2)

Figure 72: Phase shifts (top) and D(A), A < L (bottom panel) for −V (3 P0 ) from the solution of the N/D22 in expansion
1
of LP’s, g(x, y) → x− 2 g(x, y) (gxm0p5), Λ = 10mπ (L=1e1) and 4 × 103 points along the LHC in which ∆(A) is
calculated (mt=4000).

198
n=50 n=400 n=800 ERE
n=100 n=500 n=900
n=200 n=600 n=1000
n=300 n=700 xm0p5,n=1000

L=1e1 ; mt=4000 ; gx0p0


80

70

60

50

40
δ(p)

30

20

10

-10
0 50 100 150 200 250 300 350 400
|p| (MeV)

n=50 n=400 n=800


n=100 n=500 n=900
n=200 n=600 n=1000
n=300 n=700 xm0p5,n=1000

L=1e1 ; mt=4000 ; gx0p0


8000

7000

6000

5000

4000
∆(p2)

3000

2000

1000

-1000
-100 -80 -60 -40 -20 0
p2 (mπ2)

Figure 73: Phase shifts (top) and D(A), A < L (bottom panel) for −V (3 P0 ) from the solution of the N/D22 in expansion
of LP’s, g(x, y), Λ = 10mπ and 4 × 103 points along the LHC in which ∆(A) is calculated.

199
n=50 n=600 1e2,x0p0,4e3,n=1000
n=100 n=700 1e2,x0p0,8e3,n=1000
n=200 n=800 1e3,x0p0,4e3,n=1000
n=300 n=900 Standard
n=400 n=1000 ERE
n=500 1e1,xm0p5,4e3,n=1000

L=7e3 ; mt=16000 ; gx0p0


100

80

60

40

20
δ(p)

-20

-40

-60

-80

-100
0 50 100 150 200 250 300 350 400
|p| (MeV)

n=50 n=600 1e2,x0p0,4e3,n=1000


n=100 n=700 1e2,x0p0,8e3,n=1000
n=200 n=800 1e3,x0p0,4e3,n=1000
n=300 n=900 Standard
n=400 n=1000
n=500 1e1,xm0p5,4e3,n=1000

L=7e3 ; mt=16000 ; gx0p0

400

200
∆(p2)

-200

-400

-10 -8 -6 -4 -2 0
p2 (mπ2)

Figure 74: Phase shifts (top) and D(A), A < L (bottom panel) for −V (3 P0 ) from the solution of the N/D22 in expansion
of LP’s, g(x, y), Λ = 7 × 103 mπ and 16 × 103 points along the LHC in which ∆(A) is calculated. Results obtained with
other values for these parameters are also shown by the dots.

200
n=50 n=400 n=800 x0p0,1e2,n=1000
n=100 n=500 n=900 x0p0,1e3,n=1000
n=200 n=600 n=1000 ERE
n=300 n=700 xm0p5,1e1,n=1000

L=1e2 ; mt=4000 ; gxm0p5


100

80

60

40

20
δ(p)

-20

-40

-60

-80

-100
0 50 100 150 200 250 300 350 400
|p| (MeV)

n=50 n=400 n=800 x0p0,1e2,n=1000


n=100 n=500 n=900 x0p0,1e3,n=1000
n=200 n=600 n=1000 x0p0,7e3,n=1000
n=300 n=700 xm0p5,n=1000

L=1e2 ; mt=4000 ; gxm0p5

400

200
∆(p2)

-200

-400

-10 -8 -6 -4 -2 0
p2 (mπ2)


Figure 75: Phase shifts and D(A) for −V (3 P0 ) from the solution of the N/D22 in expansion of LP’s, g(x, y)/ x,
Λ = 102 mπ and 4 × 103 points along the LHC in which ∆(A) is calculated. Results obtained with other values for these
parameters are also shown by the dots.

201
m0p5,1e2,4e3 0p0,1e2,4e3 0p0,1e3,4e3
out out out
m0p5,1e1,4e3 0p0,1e2,8e3 Standard
out out

L=1e2 ; mt=4000 ; gxm0p5


500

400

300
∆(p2)

200

100

-100
-10 -8 -6 -4 -2 0
p2 (mπ2)

Figure 76: D(A), A < L for −V (3 P0 ) from the solution of the N/D22 in expansion of LP’s for several values of Λ, number
of points along the LHC in the calculation of ∆(A) and factors diving g(x, y). Every legend follows the scheme: type of
g, Λ, number of points, in/out. Here “in” means the outcome directly obtained by solving the IE expanding in LP’s while
“out” is the D(A) function calculated by substituting the D(A) solved in the l.h.s of the IE.

18 Application of the formalism to N N 3S1 -3 D1


18.1 OPE potential
The OPE potential for the coupled waves 3 S1 -3 D1 is calculated from Refs. [4, 5]. Specifically, we employ the expressions
for the off-shell partial-wave decomposition of Ref. [4], removing any reference to the component Vσk (which is already
accounted for by the other structures kept, as explained in the Appendix A). The expressions of Ref. [4] for the partial-wave
decomposition reproduce those of Ref. [5] when considered on-shell, except for a global minus sign in the mixing partial
wave.
We also clarify that we use for the OPE amplitude the expression given in Ref. [5]
2
gA
WT = , (18.21)
4fπ2 (q 2+ m2π )
which differs by a minus sign compared with the OPE given in Ref. [4]. The latter reference uses the ”Hamiltonian”
convention for a potential, while Ref. [5] follows a ”Lagrangian” convention, employing Li in the application of the
Feynman rules. 36
36
The calculation of the slope at threshold of T12 does not correspond with the slope of the mixing angle ε1 , they differ by a minus sign.
The reason is that the signs of ∆12 (A) and ∆21 (A) were changed by hand by myself. I did that because when using the partial-wave projection
formulas of Ref. [5] I am aware of the minus sign of difference in the partial-wave-projection formula for the mixing partial waves. However, in
this case, I am using the expressions of Ref. [4] for the partial-wave projections because for calculating fij (ν) one needs off-shell scattering.

202
We now give the expressions for the potentials vij (p, k) and their discontinuity across the LHC:

p2 + k2 + m2π
ξ= , (18.22)
2kp
ρ = log (ξ + 1) − log (ξ − 1) ,
 2
gA
λ= ,
2fπ
λm2π ρ
v11 (p, k) = −λ + , (18.23)
4kp
 ! !
√ k 3ξ 2 − 1 p 3ξp
v12 (p, k) = 2λ ρ ξ − − + −2 ,
2p 4k 2k
 ! !
√ p 3ξ 2 − 1 k 3ξk
v21 (p, k) = 2λ ρ ξ − − + −2 ,
2k 4p 2p
   !!
3ξ k2 + p2 3ξ 2 − 1 k2 + p2 ξ
v22 (p, k) = λ 1 − +ρ − .
4kp 8kp 2

Regarding the discontinuities we present them for the intervals of the arguments needed to calculate ∆ij (A), A = −k2 :

k > mπ , ν ∈ [−k, k − 2mπ ] , ν1 ∈ [k − mπ , ν + mπ ] ,


π
∆v̂11 (ν, ν1 ) = − λm2π , (18.24)
2
2 
πλ 3 ν 2 − ν12 + 3m4π + 2m2π ν12 − 3ν 2
∆v̂12 (ν, ν1 ) = √ ,
4 2
 2 
πλ 3 ν 2 − ν12 + 3m4π + 2m2π ν 2 − 3ν12
∆v̂21 (ν, ν1 ) = √ ,
4 2
 2       
πλ
∆v̂22 (ν, ν1 ) = 3 ν 2 − ν12 ν 2 + ν12 + 3m4π ν 2 + ν12 − 2m2π 3(ν 4 + ν14 ) + 2ν 2 ν12 .
16

We also need ∆v̂ij (ν, k) for ν ∈ [−k, k − mπ ]. Since k > ν + mπ for k > mπ /2 (the starting point of the LHC) we can
particularize the expressions in Eq. (18.24) with ν1 = k to obtain ∆v̂ij (ν, k).

18.2 Calculation of ∆ij (A)


We use the formalism developed in Sec. 8.2 to calculate ∆ij (A) for the 3 S1 -3 D1 system, with ℓ1 = 0 and ℓ2 = 2. We then
proceeded to solve the IEs of Eq. 8.75, according to the value of ν. The IEs couple fij (ν) with different i but with the same
j. Therefore with end with two separated coupled IEs, one for the set {f11 (ν), f21 (ν)} and another one for {f12 (ν), f12 (ν)}.
For simplicity in the presentation, we rewrite them here since they adopt a simpler form in the OPE case than the
general ones in Eq. (8.62):
Z k−mπ
m dν1
fij (ν) = ∆v̂ij (ν, k) − θ(k − 2mπ − ν) ∆v̂i1 (ν, ν1 )f1j (ν1 )
2π mπ +ν k − ν12
2 2
Z k−mπ  
m dν1 1 1
− θ(k − 2mπ − ν) 2 + ∆v̂i2 (ν, ν1 )f2j (ν1 ) . (18.25)
4π mπ +ν k2 − ν12 (ν1 + iε)4 (ν1 − iε)4

203
In order to avoid the infrared singularity for ν1 in the last term of the previous equation for the intermediate state
with ℓ = 2, as explained after Eq. (8.75), we need the Taylor expansion around ν = 0 of the symmetric combination
f2i (ν) + f2i (−ν) up to O(ν 2 ) by employing Eq. (18.25), which for ν > −mπ /2 simplifies further as
k−mπ Z
m dν1
fij (ν) = ∆v̂ij (ν, k) − θ(k − 2mπ − ν) 2 ∆v̂i1 (ν, ν1 )f1j (ν1 ) (18.26)
2π mπ +ν k2 − ν12
Z k−mπ
m dν1
− θ(k − 2mπ − ν) 2 ∆v̂i2 (ν, ν1 )f2j (ν1 ) .
2π mπ +ν (k − ν12 )ν14
2

From here we directly get


Z
k−mπ
m dν1
fij (0) = ∆v̂ij (0, k) − θ(k − 2mπ ) 2 ∆v̂i1 (0, ν1 )f1j (ν1 ) (18.27)
2π mπ k2 − ν12
Z k−mπ
m dν1
− θ(k − 2mπ ) 2 ∆v̂i2 (0, ν1 )f2j (ν1 ) .
2π mπ (k − ν12 )ν14
2

For the following steps to obtain the derivatives of f2i (ν)+f2i (−ν) at ν = 0 it is important to keep in mind that ∆v̂ij (ν, ν1 )
depends only quadratically in the arguments, as explicitly shown in Eq. (18.24).
Another result that we are going to use is that

∆v̂ij (0, mπ ) = 0 for ℓi 6= 0 . (18.28)

The reason is because ξ2kp = 0 for p = 0 and k = imπ , cf. Eq. (18.22) with p = iν and k = iν1 . As a result all the terms
ℓ +1
in ∆v̂ij vanish when we multiply by ν ℓi +1 ν1j in order to pass from vij to ∆v̂ij (this is why it is needed that ℓi 6= 0,
without any constraint in ℓj , because if ℓi = 0 we do not multiply by pℓi that is zero for ν = 0.)
The first derivative of f2j (ν) from (18.26) is

f′2j (ν) = ∆v̂2j



(ν, k) (18.29)
m ∆v̂21 (ν, mπ + ν)f1j (mπ + ν) m ∆v̂22 (ν, mπ + ν)f2j (mπ + ν)
+ θ(k − 2mπ − ν) 2 2 2
+ θ(k − 2mπ − ν) 2 2
2π k − (mπ + ν) 2π (k − (mπ + ν)2 )(mπ + ν)4
Z k−mπ Z k−mπ
m dν1 ′ m dν1
− θ(k − 2mπ − ν) 2 ∆v̂21 (ν, ν1 )f1j (ν1 ) − θ(k − 2mπ − ν) 2 ∆v̂ ′ (ν, ν1 )f2j (ν1 ) .
2
2π mπ +ν k − ν1 2 2π mπ +ν (k − ν12 )ν14 22
2

The terms involving the derivative of the Heaviside function θ(k − 2mπ − ν) in Eq. (18.26) do not contribute to the
derivative because they give rise to a Dirac delta function, −δ(k − 2mπ − ν), times an integral evaluated at ν = k − 2mπ .
The latter vanishes because its integration interval shrinks to zero
We now proceed to evaluate the second derivative of Eq. (18.26) at ν = 0. For this particular value many terms in the
derivative of Eq. (18.29) do vanish. The derivatives of the Heaviside functions in Eq. (18.29) do not give contribution.
Those multiplied by an integral do not contribute because of the reason already explained in the previous paragraph.
Regarding the others without integrals they also vanish because they are multiplied by ∆v̂2i (0, mπ ) = 0.
The second line in Eq. (18.29) does not give contribution to this derivative because the terms that do not involve the
′ (ν, m ± ν)|
derivative of ∆v̂2i (ν, mπ ± ν) are then multiplied by ∆v̂2i (0, mπ ) = 0. Those terms involving ∆v̂2i π ν=0 also
vanish. The reason is the following

∂∆v̂2i (ν, mπ + ν) ∂∆v̂2i (ν ′ , mπ ) ∂∆v̂2i (0, ν ′ )
= ′ + ′ . (18.30)
∂ν ν=0 ∂ν ′ ν =0 ∂ν ′ ν =mπ

The first term on the rhs of the previous equation is zero because ∆v̂ij (ν, mπ ) for OPE is a polynomial in ν 2 and
∆v̂2j (0, mπ ) = 0, so that ∆v̂2i (ν, mπ ) ∝ ν 2 . The last term is also zero because for calculating ∆v̂ij (p, k) we have to

204
multiply by pℓi +1 kℓj +1 . Therefore, only the term with the highest power of ξ survives, in our case ξ 2 because ℓi = 2, but
then ∂(p2 + k2 + m2π )2 /∂k = 2k(p2 + k2 + m2π ) being zero for p = 0 and k = imπ .
We are then left with the following expression for f′′2i (0),
f′′2j (0) = ∆v̂2j
′′
(0, k) (18.31)
Z k−mπ Z k−mπ
m dν1 m dν1
− θ(k − 2mπ ) ∆v̂ ′′ (0, ν1 )f1j (ν1 ) − θ(k − 2mπ ) 2 ∆v̂ ′′ (0, ν1 )f2j (ν1 ) .
2π 2 mπ k2 − ν12 21 2π mπ (k2 − ν12 )ν14 22
Of course, this second derivative of f2j (ν) at ν = 0 could be also evaluated numerically from Eq. (18.26). We have checked
that both results agree within numerical precision. We also discuss below another method applicable for k > 3mπ /2 by
fitting f2j (ν) + f2j (−ν) in the region ν ∈ [−mπ /2, mπ /2] with a polynomial of fourth degree in ν, cf. Eq. (18.44).
We put together the IEs to get fij (ν) in the OPE case distinguishing the three ranges of values of ν, particularizing
Eq. (8.75) to this case:
ν > −mπ /2 :
Z k−mπ
m dν1
fij (ν) = ∆v̂ij (ν, k) − θ(k − 2mπ − ν) ∆v̂i1 (ν, ν1 )f1j (ν1 ) (18.32)
2π mπ +ν k − ν12
2 2
Z k−mπ
m dν1
− θ(k − 2mπ − ν) ∆v̂i2 (ν, ν1 )f2j (ν1 ) .
2π 2 mπ +ν (k − ν12 )ν14
2

ν < −3mπ /2 :
fij (ν) = ∆v̂ij (ν, k) (18.33)
Z k−mπ Z k−mπ
m dν1 m dν1
− θ(k − 2mπ − ν) 2 ∆v̂i1 f1j (ν1 ) − θ(k − 2mπ − ν) 2 ∆v̂i2 (ν, ν1 )f2j (ν1 )
2 2
2π mπ +ν k − ν1 2π −(mπ +ν) (k − ν12 )ν14
2
Z −mπ −ν
m dν1  
− θ(k − 2mπ − ν) 2 2 2 4 ∆v̂i2 (ν, ν1 ) f2j (ν1 ) + f2j (−ν1 ) − 2f2j (0) − ν12 f′′2j (0)
2π 0 (k − ν1 )ν1
Z  
m  2 ′′  −mπ −ν dν1 1 1
− θ(k − 2mπ − ν) 2 2f2j (0) + ν1 f2j (0) + ∆v̂i2 (ν, ν1 ) ,
4π 0 k2 − ν12 (ν1 + iε)4 (ν1 − iε)4
and the last integral is done algebraically for ∆v̂i2 (ν, ν1 ) as given in Eq. (18.24), being free of infrared divergences.
For the last interval to be considered one has
−3mπ /2 < ν < −mπ /2 :
Z
k−mπ
m dν1
fij (ν) = ∆v̂ij (ν, k) − θ(k − 2mπ − ν) ∆v̂i1 (ν, ν1 )f1j (ν1 ) (18.34)
2π mπ +ν k2 − ν12
2
Z k−mπ
m dν1
− θ(k − 2mπ − ν) ∆v̂i2 (ν, ν1 )∆v̂2j (ν1 , k)S4 (ν1 )
4π 2 mπ +ν k − ν12
2
!2 Z Z
m2 k−mπ dν1 ν1 −mπ dν2
+ 2θ(k − 3mπ − ν) f2j (ν1 ) ∆v̂i2 (ν, ν1 )∆v̂22 (ν2 , ν1 )S4 (ν1 )
4π 2 2mπ +ν (k2 − ν12 )ν14 mπ +ν k2 − ν22
!2 Z Z
m2 k−mπ dν1 ν1 −mπ dν2
+ 2θ(k − 3mπ − ν) f1j (ν1 ) ∆v̂i2 (ν, ν2 )∆v̂21 (ν2 , ν1 )S4 (ν2 ) ,
4π 2 2mπ +ν k2 − ν12 mπ +ν k2 − ν22
with the function Sn (ν) defined e.g. in Eq. (8.63), that we reproduce here
1 1
S2n (ν) = 2n
+ . (18.35)
(iν + ε) (iν − ε)2n

205
The integrations in the variable ν2 in Eq. (18.34) are done algebraically and in this way the infrared singularities are avoided.
• A trick to keep in mind is that the primitive of any integral of the form
Z
dν1
S2n (ν1 )ν12m = Fm (ν1 ) , (18.36)
k2 − ν12

can be chosen such that Fm (0) = 0, without any additive integration constant.37 Let us prove it in detail:
For m = n the result is straightforward
Z
dν1 1 ν1
2 = arctanh . (18.37)
k2 − ν1 k k

For m > n the result is then clear because we can always reduce in steps of one the power of ν12 in the numerator, e.g.
for m = 1 we have
Z Z
dν1 ν12 dν1 (ν12 − k2 + k2 ) ν1
= = −ν1 + k arctanh . (18.38)
k2 − ν12 2
k − ν1 2 k

In more general terms for m > n the numerator is a polynomial Q(ν12 ) of degree m − n in ν12 . Thus,
Z Z   Z Z
dν1 Q(ν12 ) dν1 Q(ν12 ) − Q(k2 ) + Q(k2 ) dν1
= =− dν1 Q′ (ν1 )2 + Q(k2 ) , (18.39)
k2 − ν12 k2 − ν12 k2 − ν12

with Q′ (ν12 ) = dQ(ν12 )/dν12 . The first integral in the last term is zero at ν1 = 0 because it is the integration of a polynomial
of degree m − n − 1 in ν12 (no additive subtraction constant is included).
Let us now consider m < n. We reach the same conclusion by applying the following trick, p = n − m,
Z Z
dν1  1 1  1 dν1 (k2 − ν12 + ν12 )  1 1 
2 2 2p
+ 2p
= 2 2 2 2p
+ 2p
(18.40)
k − ν1 (ν1 + iε) (ν1 − iε) k k − ν1 (ν1 + iε) (ν1 − iε)
Z Z
1  1 1  2 dν1  1 1 
= 2 dν1 2p
+ 2p
+ 2 2 2 2(p−1)
+ 2(p−1)
.
k (ν1 + iε) (ν1 − iε) k k − ν1 (ν1 + iε) (ν1 − iε)
The integral before the last one is
Z
1  1 1  1  1 1 
dν1 + =− 2 + (18.41)
k2 (ν1 + iε)2p (ν1 − iε)2p k (2p − 1) (ν1 + iε)2p−1 (ν1 − iε)2p−1

which is zero for ν1 = 0. Finally, the last term in Eq. (18.40) is of the same type as the starting point and can be treated
in the same way, giving rise again to a finite amount of terms of the form in Eq. (18.41).
• In order to improve the efficiency of the numerical solution we finally present the results for three numerical strategies.
The point is that the Taylor expansion of f21 (ν) and of f22 (ν) for ν around 0 is numerically demanding and it gives rise
to numerical instabilities for too small values of ν. Namely, for ν → 0 one observes that the subtracted combination
(appearing in the intermediate states with ℓ = 2 in Eq. (8.64) and similar ones)
′′ (0)ν 2
f2j (ν) + f2j (−ν) − 2f2j (0) − f2j
(18.42)
ν4
is stable and reach a constant value. But if we proceed towards smaller values of ν then at some point this limit turns out
to be numerically unstable.
37
This is nothing but an exemplification of the fact that the primitive of an even function is an odd function, if we do not include any additive
integration constant. As a result the primitive function vanishes in the origin.

206
1) The first strategy is the standard one. We use a partition in the variable x ∈ [mπ /k, 1] such that ν1 = k − mπ /x,
which is reflected at the origin to create the negative values of ν1 so that finally the partition expands the whole interval
ν1 ∈ [−k + mπ , k − mπ ] symmetrically. We then try to improve the output by increasing the number of Gausßian points.
2) For the second strategy we take a different partition, with the aim of effectively reducing the number of Gaussian
points around ν1 = 0, which is a conflictive numerical region particularly for increasing k (there are less and less points
available as k increases to calculate properly fij (ν1 ) with ν1 around zero). We then take the partition
πxα
ν1 = (k − mπ ) tan , x ∈ [0, 1] , (18.43)
4
and reflect it around ν1 = 0. It is clear that as k increases the weight in the integral in ν1 of Eq. (8.62) of the region with
small values of ν1 decreases in the calculation of fij (ν). For a given set of x points between 0 and 1 there are less points
with small ν1 as k increases from Eq. (18.43). Additionally, for smaller values of α the concentration of points for a given
k with larger values of ν1 increases. As a compromise we have used α = 0.25 that favors the smoothness of the results
for intermediate and large values of k (k & 10 mπ ).
3) A typically more accurate strategy consists of making a polynomial fit to f2i (ν) + f2i (−ν) in the region around ν = 0,
namely for ν ∈ [−mπ /2, mπ /2], when used in Eq. (18.33) for k > 5mπ /2. As discussed above we need at least to work
out the second derivative at the origin of f2i (ν), so that we use a fourth degree polynomial to fit the previous symmetric
combination:

Qi (ν 2 ) = αi + βi ν 2 + γi ν 4 ,
X
χ2 = (f2i (νj ) + f2i (−νj ) − Qi (νj ))2 . (18.44)
j

The minimization process implies the following three equations

∂χ2 X 
= −2 f2i (νi ) + f2i (−νj ) − αi − βi νj2 − γi νj4 = 0 , (18.45)
∂αi j
∂χ2 X 
= −2 f2i (νi ) + f2i (−νj ) − αi − βi νj2 − γi νj4 νj2 = 0 ,
∂βi j
∂χ2 X 
= −2 f2i (νi ) + f2i (−νj ) − αi − βi νj2 − γi νj4 νj4 = 0 .
∂γi j

These equations can be expressed in a simpler form that also makes clear its generalization to even higher-degree polynomials
X  X X X
f2i (νj ) + f2i (−νj ) = αi +βi νj2 + γi νj4 , (18.46)
j j i j
X   X X X
νj2 f2i (νj ) + f2i (−νj ) = αi νj2 + βi νj4 + γi νj6 ,
j j j j
X   X X X
νj4 f2i (νj ) + f2i (−νj ) = αi νj4 + βi νj6 + γi νj8 .
j j j j

Its algebraic solution is implemented in the Fortran codes to evaluate ∆ij (A), so that we use Qi (ν1 ) for −mπ /2 < ν1 <
mπ /2, instead of f2i (ν1 ) in the integrands of Eq. (18.33) for ν < −3mπ /2. In connection with this we also employ αi /2 and
βi instead of f2i (0) and f′′2i (0), respectively, so that the subtracted combination Qi (ν1 )−2f2i (0)−ν12 f′′2i (0) = γi νi4 = O(ν14 ),
by construction.
We show in Fig. 77 the |∆ij (A)| calculated following the three described strategies as a log-log plot. In all the cases we
use 1000 × 20 points in the calculation of ∆ij for a given A ≥ m2π . Within the scale of the figure one can only distinguish

207
1x1010 1x1010
Strategy 1: mp=1000 Strategy 1: ∆12, mp=1000
Strategy 2: mp=1000, α=0.25 Strategy 1: ∆21, mp=1000
Strategy 3: mp=1000, α=1.0 Strategy 2: ∆12, mp=1000, α=0.25
1x108 Strategy 2: ∆21, mp=1000, α=0.25
1x108 Strategy 3: ∆12, mp=1000, α=1.0
Strategy 3: ∆21, mp=1000, α=1.0

1x106
1x106

|∆12|, |∆21|
10000
|∆11|

10000

100

100
1

1
0.01

0.01 0.0001
0.1 1 10 100 1000 0.1 1 10 100 1000
p(MeV) p(MeV)
1x1010
Strategy 1: mp=1000
Strategy 2: mp=1000, α=0.25
Strategy 3: mp=1000, α=1.0
1x108

1x106

10000
|∆22|

100

0.01

0.0001
0.1 1 10 100 1000
p(MeV)


Figure 77: From top to bottom and left to right, ∆11 (A), ∆12,21 (A) and ∆22 (A) are plotted as a function of p = A
in units of mπ .

208
between the first strategy and the other two. Notice that as discussed above the first strategy is the less stable one and it
needs an improvement, which is achieved in the other two cases. We also observe that ∆12 (A) = ∆21 (A) for the strategies
2) and 3) within the numerical precision that can be distinguished in the figure.

18.3 Lippmann-Schwinger equation for 3 S1 -3 D1


Now, we develop the formalism for the calculation of the 3 S1 -3 D1 scattering amplitudes by employing the LS equation
in coupled channels with the restriction of reproducing a given value of the 3 S1 scattering length, as . We apply here the
well-known criterion [7] that when sending the cut-off to infinity the wave functions in coupled channels for a given energy
require as many free parameters (short-distance phases) as attractive eigenvalues had the potential Vij (r) in coordinate
space in the limit r → 0. In addition, one has to impose the orthogonality of the wave functions with different energy.
In the case of the OPE potential there is only one negative eigenvalue so that only one free parameter is required for the
calculation of the renormalized scattering amplitudes [8, 9]. Here we choose to fix the S-wave scattering length.
We employ the same trick as in Eq. (2.6) to remove the principal value when considering the IE for obtaining the
2
solution of the LS equation. As there, we consider the half-off-shell T -matrix element tαβ (k′ , k; km + iε), where α, β = 1, 2
refer to the coupled partial waves (1 for 3 S1 and 2 for 3 D1 ). In this way, we write for the coupled-channel case

2 Z ∞
k2 m X dp  2 k2 k2 
tαβ (k′ , k; + iε) = vαβ (k′ , k) + 2 p vαγ (k ′
, p)t γβ (p, k; + iε) − k 2
vαγ (k ′
, k)t γβ (k, k; + iε)
m 2π γ=1 0 p2 − k2 m m
2
mk X k2
+i vαγ (k′ , k)tγβ (k, k; + iε) . (18.47)
4π γ=1 m

After discretization {pi ∈ [0, ∞], i = 1, . . . , N } we have the 4 × N equations for tαβ (pi , k; k2 /2m + iε) supplemented
with four more equations to calculate the on-shell matrix elements tαβ (k, k; k2 /2m + iε)}:
2 X N
k2 m X ∆pj  2 k2 2 k2 
tαβ (pi , k; + iε) = vαβ (pi , k) + 2 2 p j vαγ (p i , p j )t γβ (p j , k; + iε) − k vαγ (p i , k)t γβ (k, k; + iε)
m 2π γ=1 j=1 pj − k2 m m
2
mk X k2
+i vαγ (pi , k)tγβ (k, k; + iε) ,
4π γ=1 m
2 X N
k2 m X ∆pj  2 k2 2 k2 
tαβ (k, k; + iε) = vαβ (k, k) + 2 2 2
p v
j αγ (k, p )t (p
j γβ j , k; + iε) − k vαγ (k, k)t γβ (k, k; + iε)
m 2π γ=1 j=1 pj − k m m
2
mk X k2
+i vαγ (k, k)tγβ (k, k; + iε) . (18.48)
4π γ=1 m

It is important to stress that in these equations the index β plays only a parametric role.
For the numerical implementation of the solution it is convenient to duplicate the partition such that pi+N = pi ,
p2N +1 = p2N +2 = k and assign the first N entries with the coupled-channel index α = 1 and the N last ones to α = 2.
We also associate 2N + 1 and 2N + 2 with the on-shell three-momentum k and α = 1, 2, respectively. In this way we
introduce primed indices that run from 1 to 2N + 2 and define the vector ti′ β (k) and the matrix vi′ j ′ as:

 1 ≤ i′ ≤ N
 t1β (pi′ , k)
ti′ β (k) = t (p ′ , k) N + 1 ≤ i′ ≤ 2N (18.49)
 2β i
 tαβ (k, k) i′ = 2N + α

209
 
1 ≤ j′ ≤ N N + 1 ≤ j ′ ≤ 2N j ′ = 2N + β
 
 1 ≤ i′ ≤ N v11 (pi′ , pj ′ ) v12 (pi′ , pj ′ ) v1β (pi′ , k) 
vi′ j ′ =   . (18.50)
 N + 1 ≤ i′ ≤ 2N v21 (pi′ , pj ′ ) v22 (pi′ , pj ′ ) v2β (pi′ , k) 
i′ = 2N + α vα1 (k, pj ′ ) vα2 (k, pj ′ ) vαβ (k, k)

With this notation the whole Eq. (18.48) comprises into


2N 2 2
m X ∆pj ′  2 k2 X  imk X
ti′ β (k) = vi′ 2N +β + p j ′ vi′ j ′ tj ′ β (k) − vi′ 2N +γ t2N +γ β (k) + vi′ 2N +γ t2N +γ β (k) .
2π 2 j ′ =1 p2j ′ − k2 2 γ=1 4π γ=1
(18.51)

Notice the factor 1/2 in front of the sum over γ within the square brackets because the sum over j ′ is from 1 to 2N, while
in Eq. (18.48) is from 1 to N . This expression is ready for its numerical implementation.
Now let us discuss how to impose a given value of the scattering length as . For that we add a counterterm v0 to
v11 (p, k) and re-define it as

v11 (p, k) = v0 + v̂11 (p, k) (18.52)


λm2π ρ
v̂11 (p, k) = −λ + ,
4kp
Let us note that
 λm2π ρ 
v̂11 (0, 0) = lim −λ+ =0. (18.53)
p,k→0 4kp
In order to fix v0 we particularize Eq. (18.51) for k = 0 and β = 1, which simplifies to
2N
m X
ti′ β (0) = vi′ 2N +β + ∆pj ′ vi′ j ′ tj ′ β (0) . (18.54)
2π 2 j ′ =1

In the following we denote t11 (0, 0) by ν1 , which in terms of as reads


4πas
ν1 = − . (18.55)
m
Taking into account the splitting in Eq. (18.52) and the relation in Eq. (18.55) we have the following expression to isolate
v0 from Eq. (18.54) with i′ = 2N + 1 (the matrix v̂i′ j ′ is defined analogously to vi′ j ′ ),
 
 m X2N  m X2N
v0 1 + 2 ∆pj ′ tj ′ 2N +1 (0) = ν1 − 2 ∆pj ′ v̂2N +1 j ′ tj ′ 2N +1 (0) (18.56)
 2π j ′ =1  2π j ′ =1

38 We then need to calculate from Eq. (18.54) the half-off shell T -matrix ti′ 2N +1 (0) with 1 ≤ i′ ≤ 2N :
 
 m X2N  m X2N
ti′ 1 (0) = v0 θ(N + 1 − i′ ) 1 + 2 ∆pj ′ tj ′ 2N +1 (0) + v̂i′ 2N +1 + 2 ∆pj ′ v̂i′ j ′ tj ′ 2N +1 (0) , (18.57)
 2π j ′ =1  2π j ′ =1

where θ(x) = 1 for x > 0, and 0 otherwise. Implementing Eq. (18.56) in the previous equation, one has
2N
m X 
ti′ 1 = ν1 θ(N + 1 − i′ ) + v̂i′ 2N +1 + 2
∆pj ′ v̂i′ j ′ − θ(N + 1 − i′ )v̂2N +1 j ′ tj ′ 2N +1 (0) . (18.58)
2π j ′ =1

210
100 2.5
LS: δ11 LS
LS: δ22 ND11
80 ND11: δ11
ND11: δ22

60 2

40

20 1.5
δ11,22 (deg)

ε1 (deg)
0

-20 1

-40

-60 0.5

-80

-100 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
p(MeV) p(MeV)


Figure 78: From left to right, we plot the phase shifts δ11,22 and the mixing angle ε1 as a function of p = A in units of
MeV. The dots indicate the results obtained by the LS equation and the lines those obtained from the N D11 DRs.

Thus, once this IE is solved we can calculate v0 by using Eq. (18.56) and then we are ready to obtain the half-off-shell
T -matrix from Eq. (18.51).
We show in Fig. 78 the results obtained for the 3 S1 -3 D1 coupled partial waves. In the panel on the left we give the 3 S1
(δ11 ) and 3 D1 (δ22 ) phase shifts, while in the right one the mixing angle ǫ1 is depicted. The points are the results obtained
by applying the LS equation as explained in this section. The results converge very well by varying the cut-off from 10
mπ up to more than 103 mπ . These results also agree very well with the ones obtained by applying the once-subtracted
N/D method (case N D11 ), elaborated in the next section, and that correspond to the solid lines. In all the cases we take
as = 5.424 fm.

18.4 N/D with coupled channels. The simplest case: N/D11


We use a matrix language for the T , N and D functions, so that they are n × n square matrices (with n the number of
coupled channels; here n = 2). We write the T matrix as

T = D −1 N , (18.59)
−1
D=NT ,
N = DT .

Unitarity implies that:

ImD(A) = N (A)ImT (A)−1 = −N ρ(A) , A > 0 , (18.60)

with ρ(A) the usual phase space factor. In matrix language we define a diagonal matrix ρ(A) whose matrix elements are
also ρ(A).
38
For numerical purposes it is convenient to introduce a smooth cut-off function gj ′ = exp(−p2j ′ /Λ2 ) and multiply tj ′ 2N+1 (0) in all the sums
by gj ′ . One should study the stability of the results when increasing Λ. In our case we obtain stable results up to very high values of Λ.

211
The LHC discontinuity ∆(A) implies that:

ImN (A) = D(A)ImT (A) = D(A)∆(A) , A < L . (18.61)

Since we want to impose a given value for the 3 S1 scattering length as , we consider a once-subtracted DR for N (A),
as well as the standard once-subtracted DR for D(A). It results:
Z
A L D(k2 )∆(k2 )
N (A) = n
e+ dk2 2 2 ,
π −∞ k (k − A)
Z
A ∞ 2 N (q 2 )ρ(q 2 )
D(A) = de − dq 2 2 , (18.62)
π 0 q (q − A)

where ne and de are constant matrices, with D(0) = d.e Since D(A) is assumed to have inverse so it must have d.
e Therefore,
e−1
we can multiply simultaneously the original matrices D and N by d , an operation that leaves invariant T (A) according
to Eq. (18.59) . In this way, we can always finally take de = I in Eq. (18.63) and rewrite the DRs for N (A) and D(A) as
Z
A L D(k2 )∆(k2 )
N (A) = n + dk2 2 2 , (18.63)
π −∞ k (k − A)
Z
A ∞ 2 N (q 2 )ρ(q 2 )
D(A) = I − dq 2 2 . (18.64)
π 0 q (q − A)

Replacing N (A) in the last equation as given by its DR one has


Z Z Z
A ∞ ρ(q 2 ) A L D(k2 )∆(k2 ) ∞ ρ(q 2 )
D(A) = I − n dq 2 2 2
+ 2 dk2 dq 2 . (18.65)
π 0 q (q − A) π −∞ k2 0 (q 2 − A)(q 2 − k2 )

Recalling the definition and algebraic computation of the basic unitarity integral g(A, B) along the RHC as given in
Eq. (12.7), we rewrite the previous equation as:
√ Z 2 2
im A imA L 2 D(k )∆(k ) 1
D(A) = I − n+ 2
dk 2
√ √ . (18.66)
4π 4π −∞ k A + i −k2
This representation provides us with the following IE for D(A) with A along the LHC,
√ Z
m −A mA L D(k2 )∆(k2 ) 1
D(A) = I + n+ 2 dk2 2
√ √ , A<L. (18.67)
4π 4π −∞ k −A + −k2
The imposition of the scattering length as is easily achieved by taking
!
ν1 0
n= . (18.68)
0 0

Here we take into account that Tαβ (0) = 0 except for α = β = 1, and ν1 is given in Eq. (18.55).
It is worth stressing that in the IE for Dαβ (A), the row index α only plays a parametric role and it is held fixed.
As in the LS equation, we duplicate the partition along the LHC {Ai , 1 ≤ i ≤ N } and used primed indices that run
from 1 to 2N . Define the vector Dαi′ as
(
Dα1 (Ai′ ) 1 ≤ i′ ≤ N ,
Dαi′ = (18.69)
Dα2 (Ai′ ) N + 1 ≤ i′ ≤ 2N ,

212
100 2.5
δ11, Λ=4 102 Λ=4 102
δ22, Λ=4 102 Λ=3 1022
80 δ11, Λ=3 1022 Λ=2 10
δ22, Λ=3 102 Λ=1 102
δ11, Λ=2 102
60 δ22, Λ=2 10 2
δ11, Λ=1 1022
δ22, Λ=1 10
40

20 1.5
δ11,22 (deg)

ε1 (deg)
0

-20 1

-40

-60 0.5

-80

-100 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
p(MeV) p(MeV)


Figure 79: From left to right, we plot the phase shifts δ11,22 and the mixing angle ε1 as a function of p = A in units of
MeV. The different curves are obtained with the N/D11 DRs up to a maximum three-momentum along the LHC indicated
by the value of Λ in the legend, from around 102 up to 4 102 mπ . The convergence is clear.

e ′ ′ as
and the matrix ∆ ij
 
1 ≤ j′ ≤ N N + 1 ≤ j ′ ≤ 2N

e ′ ′ =  1 ≤ i′ ≤ N 
∆ ij ∆11 (Aj ′ ) ∆21 (Aj ′ )  . (18.70)
N + 1 ≤ i′ ≤ 2N ∆12 (Aj ′ ) ∆22 (Aj ′ )

We also denote by [i′ ] the integer part of i′ /N plus 1, which is 1 if 1 ≤ i′ ≤ N and 2 for N + 1 ≤ i′ ≤ 2N . In this
notation Eq. (18.67) becomes
√ ! 2N
m −Ai′ m X e ′ ′
δA ′ A ′ ∆
Dαi′ = δα[i′ ] 1+ ν1 + 2 √ j i p ij
Dαj ′ . (18.71)
4π 4π j ′ =1 Aj ′ ( −Ai′ + −Aj ′ )

The matrix to be inverted numerically in the previous equation is also independent of α. The difference between α = 1
and 2 stems from the independent term.
The results obtained with the N/D11 DRs are shown in Figs. 78 and 79 by the different lines. In the former one they
are compared with the phase shifts and mixing angle obtained by solving the LS equation in coupled channels and shown
by the dots. The agreement between the two methods is very accurate. In Fig. 79 we study the numerical convergence of
the results by varying the maximum momentum allowed along the LHC, which modulus is denoted by Λ and ranging in
the figure from 90 mπ up to 400 mπ . The convergence is quite clear.

18.5 N/D with coupled channels. General formalism


The N/D method in coupled channels making use of matrix notation as a direct generalization of the uncoupled case [10]
does not typically provide a symmetric T -matrix. However, this property of partial-wave amplitudes is a basic consequence
of time reversal.

213
To understand better the problem let us take the combination T − T T , with T = D −1 N , and multiply it by D to the
left and by DT to the right, it results:

D(T − T T )D T = N D T − DN T . (18.72)

This combination has no cuts because along the RHC

Im(D(T − T T )D T ) = N Im(DT ) − Im(D)N T = −N ρT N T − N ρN T = 0 , A > 0 , (18.73)

where we have used Eq. (18.60). It also has no cut along the LHC by taking into account Eq. (18.61) and that ∆(A) is a
symmetric function. Namely,

Im(D(T − T T )D T ) = Im(N )DT − DIm(N T ) = D∆DT − D∆D T = 0, A < sL . (18.74)

As a result D(T − T T )D T is a function analytic everywhere in the physical Riemann sheet of the A complex plane.
However, for general N/D DRs we cannot conclude that this function is zero. In the simplest applications of the N/D
method this was assumed to be typically the case as found in Ref. [11]. Indeed we have used a similar argument as the one
originally employed in this reference to show that D(T − T T )D T is an entire function. However, the limiting assumption
of this reference is to take N/D DRs of the form
Z
1 ImN (k2 )
L
N (A) = dk2, (18.75)
π
−∞ k2 − A
Z
A ∞ 2 ImD(q 2 )
D(A) = I + dq 2 2 .
π 0 q (q − A)

If one assumes that Im(N (A)) → 0 for A → −∞ at least as fast as 1/A then one can conclude easily from the previous
DRs that D(T − T T )DT → 0 for A → ∞. This is enough to guarantee the vanishing of this function for all A because
of the Liouville theorem in complex analysis, which states that an entire function is only bounded at infinity when it is a
constant. In our present discussion in connection with Eq. (18.75) this constant would be just zero.
However, these are too simplistic arguments to be applied to our problem where ∆(A) is not bounded, and it can even
grow faster than any power for A → −∞, cf. Fig. 77. At the practical level this makes very cumbersome the application
of the standard N/D equations in coupled channels since they typically produce wrong results from the start, which are
not even symmetric by construction.
As an example that the resulting T matrix is not symmetric from the standard N/D DRs in coupled channels we
present here the case in which we also fix the value of the effective range of the 3 S1 wave, rs , in addition to the scattering
length. This can be done quite straightforwardly because the mixing between partial waves is an O(A2 ) effect and it can
be disregarded up to this order in a power expansion around A = 0 of the T11 partial wave . Namely, from T = D−1 A we
have
1 D11 D22 − D12 D21
= . (18.76)
T11 D22 N11 − D12 N21

Taking into account that D12,21 = O(A), N21 = O(A) and that D11,22 → 1 + O(A) we are left with the announced result
that
1 D11
= + O(A2 ) . (18.77)
T11 N11

214
Thus, when writing down a once-subtracted DR for N11 and a twice-subtracted DR for D11 the subtraction constants can
be fixed in terms of as and rs , similarly as we did for in Sec. 12.2.39 Instead of Eqs. (12.11) and (12.13) we have now
m  1 1 
D11 + iρN11 = N11 (A) − + rs A + O(A2 ) , (18.78)
4π a 2
Z
Amν1 CA L [D(k2 )∆(k2 )]11
D11 + iρN11 = 1 + δ2 A + p − dk2 h(0, C, k2 ) + O(A2 )
4π |C| π −∞ k2
Z L
m  ν1 1 A [D(k2 )∆(k2 )]11 
= − + ν1 rs A − dk2 + O(A2 ) . (18.79)
4π as 2 as π −∞ (k2 )2

Here [D(k2 )∆(k2 )]11 is the product of the matrices D(A) and ∆(A) out of which we take the matrix element 11. This is
the difference with respect to Eqs. (12.11) and (12.13), all the other contributions are the same. Therefore, keeping this
in mind, we can immediately adapt Eq. (12.17) to our present case and have
Z √
1 √ mA L 2 2
2 [D(k )∆(k )]11  √ A 1 
D11 (A) = 1 − as rs A + ias A − 2 dk 2
√ √ + . (18.80)
2 4π −∞ k 2
( A + i −k ) −k 2 as k2

For the other matrix elements of D(A) as well as for N (A) we employ Eqs. (18.66) and (18.63), in order.
Taking these expressions along the LHC we have the IEs of Eq. (18.67) and the new one for D11 (A),
Z √
1 √ mA L 2 2
2 [D(k )∆(k )]11  −A 1 
D11 (A) = 1 − as rs A − as −A − 2 dk 2
√ √ √ + . (18.81)
2 4π −∞ k 2
( −A + −k ) −k 2 as k2

As a check, when we use the value of rs corresponding to the one predicted by the solution of the N D11 DRs,

rs;11 = 1.615 fm (18.82)

we reproduce the results previously discussed in Sec. 18.4 for the N D11 case. However, for other values of rs the resulting
T -matrix is not symmetric. For example, this is the case when using the experimental value of rs

rs;ex = 1.753 fm (18.83)

obtained from the Nijmegen phase shifts analysis.


Our aim now is to obtain a modified version of the coupled-channel N/D IEs compatible with the expression T = D −1 N ,
with D having only RHC and N only LHC, but that by construction the resulting T -matrix be symmetric.
• We take the basic coupled-channel N/D expressions of Eq. (18.59). From here the equation for the requirement that
T12 = T21 is

D22 N12 − D12 N22 = D11 N21 − D21 N11 . (18.84)

This equation allows to obtain N21 in terms of the others matrix elements. So we have to find a way to obtain such matrix
elements.
We recall the the N/D equations for Dij and Nij do not change the subscript i. These equations just depend on the
product D∆, so that the row index is held fixed. In this way we make use of them to calculate the matrix elements with
i = 1, namely, D11 (A), D12 (A), N11 (A) and N12 (A).
In order to determine the functions D2j (A) we impose on them to have only RHC, which discontinuity given by

Im(D2j (A)) = −N2j (A)ρ(A) , A > 0 , (18.85)


39
We refer here to the value corresponding to the 3 S1 eigenphase

215
as in Eq. (18.60). Next, we impose that N22 (A) has only LHC according to Eq. (18.61),
2
X
Im(N22 (A)) = D2j (A)∆j2 (A) , A < L . (18.86)
j=1

Making use of the imaginary parts by now established along the LHC and RHC we are now in disposition to show that
N21 (A), as resulting from Eq. (18.84), has only LHC. First, we show that it is real along the RHC. In the following we
indicate by the superscript r the real part of a function and by the superscript i its imaginary part. In this way, we rewrite
Eq. (18.84) along the RHC as
r i r i r i r i
(D22 + iD22 )N12 − (D12 + iD12 )N22 + (D21 + iD21 )N11 = (D11 + iD11 )N21 . (18.87)
Now the unitarity relations for Im(Dij (A)) for A > 0 imply that Im(Dij (A)) = −Nij (A)ρ(A). When this is incorporated
in Eq. (18.87) all the terms multiplied by i cancel each other and we are left with
r r r r
D22 N12 − D12 N22 + D21 N11 = D11 N21 , (18.88)
which explicitly shows that N21 (A) is real for A > 0 (recall that N1j (A) and N22 (A) are real for A > 0 by construction).
Next we consider the Eq. (18.84) along the LHC, and take into account that we already have that Im(Nij (A)) =
P
l Dil (A)Nlj (A), A < L, for i = 1, j = 1, 2 and i = 2, j = 2, together with the fact that the Dij (A) are real for A < 0.
After simplifying the straightforward cancellations between terms we are left with
r r r r i
D22 N12 − D12 N22 + D21 N11 + iD22 D11 ∆12 + iD21 D11 ∆11 = D11 (N21 + iN21 ). (18.89)
Equating the real and imaginary parts we have for the latter that
i
N21 = D22 ∆12 + D21 ∆11 (18.90)
= D22 ∆21 + D21 ∆11 ,
where in the last step we have taken into account that ∆ij (A) is a symmetric matrix. We then end with the right expression
for Im(N21 (A)) according to the coupled-channel N/D equations, cf. Eq. (18.61). Regarding the real part of Eq. (18.89)
we have the equation for N21 r

r 1  r r r 
N21 = D22 N12 + D21 N11 − D12 N22 , A<L, (18.91)
D11
which is completely analogous to Eq. (18.88) with obvious replacements. Then, we see that N21 (A) obtained by imposing
Eq. (18.88) has the right analytical cut structure as required by the N/D method. That this should be fulfilled it is clear
because we already obtained that the N D11 case gave rise to a symmetric T matrix, the same as the simple case of
Eq. (18.75) already discussed. Thus, the general structure of the N/D method in coupled channels is not incompatible
with having a symmetric T matrix, the problem arises by taking generic DRs.
However, we can go further and realize that D21 (A) satisfies a DR in terms of N21 (A) along the RHC. The latter can
be calculated from Eq. (18.88) in terms of D22 , N22 and D21 itself, while N22 is calculated in terms of D2j along the
LHC. As a result all these functions are coupled and one can calculate them through a suitable coupled IE of the D2j .
For definiteness we take that these functions satisfy once-subtracted DRs, the generalization to higher-subtracted DRs is
straightforward. For N22 (A) and D22 (A) we have the same DRs as already written for the N D11 case in Eqs. (18.63) and
(18.64):
Z
A L [D(k2 )∆(k2 )]22
N22 (A) = dk2 , (18.92)
π −∞ k2 (k2 − A)
Z
A ∞ 2 N22 (q 2 )ρ(q 2 )
D22 (A) = 1 − dq
π 0 q 2 (q 2 − A)
Z L 2 2
imA 2 [D(k )∆(k )]22 √ 1
=1+ 2
dk 2
√ . (18.93)
4π −∞ k A + i −k2

216
r (A),
When particularized for A > 0 it results for D22
Z
r mA L [D(k2 )∆(k2 )]22
D22 (A) = 1 + dk2 √ ,A > 0 , (18.94)
4π 2 −∞ −k2 (k2 − A)

while for A < L we have


Z
mA L [D(k2 )∆(k2 )]22
D22 (A) = 1 + 2 dk2 √ √ , A<L. (18.95)
4π −∞ k2 ( −A + −k2 )

The function D21 (A) satisfies


Z
A ∞ N21 (q 2 )
D21 (A) = − dq 2 . (18.96)
π 0 q (q 2 − A)
2

The function N21 (A) along the RHC is calculated from Eq. (18.88). Implementing in the latter Eqs. (18.92) and (18.94),
one can write for A > 0
N12 (A) r r (A)
D12 N11 (A) r
N21 (A) = D22 (A) − N22 (A) + r D (A) (18.97)
r
D11 (A) r
D11 (A) D11 (A) 21
Z ! Z
N12 (A) mA L 2 )∆(k 2 )] r (A) A L [D(k2 )∆(k2 )]22
2 [D(k 22 D12 N11 (A) r
= r 1+ dk √ − dk2 + r D (A) .
D11 (A) 4π −∞ −k2 (k2 − A) r (A) π
D11 −∞
2 2
k (k − A) D11 (A) 21

Inserting the latter equation into Eq. (18.96) we are left with the following representation for D21 (A):
Z Z
A ∞ 2 ρ(q 2 ) N12 (q 2 ) A ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2
D21 (A) = − dq 2 2 r (q 2 ) − π dq 2 2 r (q 2 ) D21 (q ) (18.98)
π 0 q (q − A) D11 0 q (q − A) D11
Z 2 2 Z ∞
A L 2 [D(k )∆(k )]22 ρ(q 2 ) r (q 2 )
 D12 m 
+ 2 dk √ dq 2 2 2 2 r 2
√ + N12 (q 2 ) .
π −∞ −k 2 0 (q − k )(q − A)D11 (q ) −k 2 4π

This equation together with Eq. (18.94) allows us to determine these functions and close the problem. Indeed, what we
r (A) for A > 0 and the combination [D(A)∆(A)]
really need to find is the functions D21 22 along the LHC (A < L). The
latter can be obtained directly by combining Eqs. (18.94) and (18.98). Therefore, the coupled IEs that we have to solve
to obtain D2j (A) can be finally written as

A <L
Z
∆12 (A)A ∞ 2 ρ(q 2 ) N12 (q 2 )
[D(A)∆(A)]22 =∆22 (A) − dq 2 2 r (q 2 ) (18.99)
π 0 q (q − A) D11
Z 2 2 Z
∆22 (A)mA L 2 [D(k )∆(k )]22 ∆12 (A)A ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2
+ dk √ √ − dq 2 2 r (q 2 ) D21 (q )
4π 2 −∞ k2 ( −A + −k2 ) π 0 q (q − A) D11
Z Z
∆12 (A)A L [D(k2 )∆(k2 )]22 ∞ ρ(q 2 ) r (q 2 )
 D12 m 
+ dk2 √ dq 2 r
√ + N12 (q 2 ) ,
π2 −∞ −k2 0
2 2 2 2
(q − k )(q − A)D11 (q ) −k2 4π
A >0
Z Z
r A ∞ 2 ρ(q 2 ) N12 (q 2 ) A ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2
D21 (A) = − − dq 2 2 r (q 2 ) − π − dq q 2 (q 2 − A) D r (q 2 ) D21 (q ) (18.100)
π 0 q (q − A) D11 0 11
Z Z
A L [D(k2 )∆(k2 )]22 ∞ 2 ρ(q 2 ) r (q 2 )
 D12 m 
+ 2 dk2 √ − dq 2 2 2 r 2
√ + N12 (q 2 ) .
π −∞ −k 2 0 (q − k )(q − A)D11 (q ) −k 2 4π

217
• By solving this IE we can determine D2j (A) in the whole complex-A plane by employing Eqs. (18.93) and (18.98).
• The calculation of N22 (A) in the complex-A plane can be performed by integrating Eq. (18.92).
• Recall that D1j (A) and N1j (A) are known functions obtained by solving the straightforward coupled-channel N/D
DRs, that decouple into D2j (A) and D1j (A).
• The function N21 (A) is obtained from Eq. (18.84) for any complex value of A once the other functions Nij (A) and
Dij (A) are already determined (as discussed).
• We can then calculate the T -matrix as T = D−1 N in the whole complex A-plane which is a symmetric matrix by
construction.
• One should keep in mind that the DRs used to end with Eqs. (18.99)-(18.100) for D2j and N22 are once-subtracted.
Following an analogous procedure one could straightforwardly obtained the expressions for the general several-times sub-
tracted DRs.
• When this procedure is used for the case in which we take once-subtracted DRs for all the Dij (A), fixing as as the
only unknown subtraction constant in N11 (A), the results agree with those discussed of the N D11 method in Sec. 18.4.
This is also a good numerical check for the codes to put in practice the new method.

18.6 Numerical procedure used to put in practice the coupled IES of Eqa. (18.99)-(18.100)
• The coupled IEs in Eqs. (18.99)-(18.100) are singular IEs because they involve the unknown function D21 (A) in RHC
integrals with a vanishing denominator at the point q 2 = A, and the principal value of the integral should be implemented.
We have followed a simplistic procedure that in practice produces accurate numerical results that might be improved
by increasing the number of integration points. When employed it drives to a method of the same degree of complexity
as solving non-singular IEs, as those for the D1j (A) functions.
Let us take by definiteness the RHC integration
Z
A ∞ ρ(q 2 ) N11 (q 2 ) r 2
I2 (A) = − − dq 2 2 2 r (q 2 ) D21 (q ) (18.101)
π 0 q (q − A) D11

that appears in Eq. (18.100) for D21 (A) (we proceed in the same way for the other RHC integrations involving a principal-
value integration). We take a partition of points {qi2 ∈ [0, ∞], 1 ≤ i ≤ N } along the RHC to evaluate this integral
numerically. Of course, the problem arises when A is one of the points in the partition, let us say, A = qi2 . Then we rewrite
I2 (A) by adding and subtracting the same term that removes the pole in the numerical integration. It becomes
Z Z
q2 ∞ ρ(q 2 )  N11 (q 2 ) r 2 N11 (qi2 ) r 2  qi2 ∞ 2 ρ(q 2 ) N11 (qi2 ) r 2 
I2 (A) = − i − dq 2 2 2 r (q 2 ) D21 (q ) − r (q 2 ) D21 (qi ) − π − dq q 2 (q 2 − A) D r (q 2 ) D21 (qi ) .
π 0 q (q − qi2 ) D11 D11 i 0 11 i
(18.102)

The last integration in the previous equation can be done algebraically and it is one function that we already evaluated,
cf. Eq. (12.7),
q
Z im qi2 Z
qi2 ∞ ρ(q 2 ) q2 ∞ ρ(q 2 )
dq 2 2 2 = → i − dq 2 2 2 =0, (18.103)
π 0 q (q − A) 4π π 0 q (q − A)
since the principal value is a real number and the full integration is purely imaginary. This is the same as the vanishing
result in Eq. (2.3). Indeed, this equation can be generalized to the result
Z ∞ ρ(q 2 )
− dq 2 = 0 , for all p2i > 0 . (18.104)
0 (q 2 − p21 )(q 2 − p22 ) . . . (q 2 − p2n )

This result follows by applying reiteratively Eq. (12.7) and reducing Eq. (18.113) to simpler fractions.

218
The first integration in Eq. (18.102) is done numerically and is given by
Z
qi2 ∞ 2 ρ(q 2 )  N11 (q 2 ) r 2 N11 (qi2 ) r 2 
− − dq 2 2 D
r (q 2 ) 21 (q ) − r (q 2 ) D21 (qi ) (18.105)
π 0 q (q − qi2 ) D11 D11 i
q 2 X 2 ρ(qj2 )  N11 (qj2 ) r 2 r (q 2 ) 
N11 (qi2 ) r 2  qi2 2 ρ(qi2 ) d  N11 (q 2 )D21
=− i δqj 2 2 D (q
r (q 2 ) 21 j ) − D
r (q 2 ) 21 i(q ) − δq .
π j6=i qj (qj − qi2 ) D11 j D11 i π i qi2 dqi2 D11r (q 2 ) q 2 =qi2

We simplify the numerical calculation by neglecting the last term. The point is that it is just one more point among many
others of the same magnitude. We assume here that the number of integration points is large enough, the expected error
in this approximation is O(N −1 ) (although this estimation could be too naive in some parts of the integration region,
depending on the precise way the integration points are distributed). A simple numerical integration that can be tested
with this method is
Z 2
∞ ρ(q 2 )e−λq X 1  2 2
− 2
dq 2 2 2 → δqj2 2 2 2 ρ(qj2 )e−λqj − ρ(qi2 )e−λqi , λ > 0 , (18.106)
0 q (q − qi ) j6=i
qj (qj − qi )

by comparing with the principal value integral calculated with Mathematica. The convergence is rapidly reached with
increasing number of Gaussian points.
Therefore we numerically approximate I2 (qi2 ) as
Z
qi2 ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2 qi2 X 2 ρ(qj2 )  N11 (qj2 ) r 2 N11 (qi2 ) r 2 
− − dq 2 2 2 r D21 (q ) → − δq j 2 2 2 r 2 D21 (q j ) − r (q 2 ) D21 (qi ) . (18.107)
π 0 q (q − qi ) D11 (q 2 ) π j6=i qj (qj − qi ) D11 (qj ) D11 i

Proceeding in an analogous way we have the following numerical approximation for the other two RHC integrations in
Eq. (18.100) for D21 (A):
Z
qi2 ∞ 2 ρ(q 2 ) N12 (q 2 ) X
2
ρ(qj2 )  N12 (qj2 ) N12 (qi2 ) 
− − dq 2 2 r (q 2 ) → δqj 2 2 r (q 2 ) − D r (q 2 ) , (18.108)
π 0 q (q − qi2 ) D11 j6=i
qj (qj − qi2 ) D11 j 11 i
Z √
qi2 ∞ 2 ρ(q 2 )  1 D12 r (q 2 ) m N12 (q 2 )  mqi2 −k2  1 D12 r (q 2 )
i m N12 (qi2 ) 
− dq 2 √ r (q 2 ) + r (q 2 ) → √ +
π 0 (q − k2 )(q 2 − qi2 ) −k2 D11 4π D11 4π qi2 − k2 −k2 D11 r (q 2 )
i
r (q 2 )
4π D11 i
2 X 2
ρ(qj ) r 2 2 r 2 2
q  1 D12 (qj ) m N12 (qj ) 1 D12 (qi ) m N12 (qi ) 
+ i 2 2 2 2
√ r 2 + r 2 −√ r 2 − r (q 2 ) . (18.109)
π j6=i (qj − k )(qj − qi ) −k D11 (qj ) 4π D11 (qj )
2 −k D11 (qi ) 4π D11
2
i

• The final numerical complication is due to the possible presence of zeroes of D11 r (A) for A > 0 that could affect
r
Eqs. (18.99)-(18.100). However, the presence of D21 (A) in the denominator stems from the expression of N21 (A) that
results from Eq. (18.84). By construction this function has no poles40 Afterwards, the resulting expression is employed in
the integrand of Eq. (18.96). This implies that actually there are no such poles due to the zeros of D11 r (A), A > 0, in

the integrand and that they are cancelled by zeroes in the numerator (we have explicitly checked that for the N D11 case).
However, when splitting the integral in several contributions (some of them even involving the unknown functions in the
IE) these zeros appear as a technical complication, but they qualify as fake singularities.
One can avoid this problem by taking into account that they are simple zeroes and that we can give meaning to the
integrals where they appear by taking their principal value. This procedure reduces to a standard integration when the
pole is cancelled by a zero in the numerator, as it is finally the case here. These zeroes are denoted by q0ℓ2 > 0 (recall
r 2
that D11 (0) = 1 by construction), with ℓ = 1, . . . , M . We typically find a few zeroes in the q range in which the RHC
40
They can always be removed by multiplying N and D by the same polynomial having a zero at this pole position.

219
extends after taking the corresponding partition along the RHC. The position of these zeroes is determined numerically in
a straightforward manner because D11 r (A) is already a known function which can be easily calculated at any value of A

wished.
To evaluate the principal value integrals around the zeroes of D11 r (A), A > 0, we can apply the same trick as above for
2
the evaluation of the same integrals around the pole at A = qi . We do not repeat the process here but first we illustrate
it with a particular integral and then give the final result for the coupled IEs.
The integral in question is
Z
qi2 ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2
− − dq 2 2 r (q 2 ) D21 (q ) (18.110)
π 0 q (q − qi2 ) D11
r (A) and we take A = q 2 , one of the point in the partition along the RHC. To
which is present in the calculation of D21 i
2 of D r (A) we add and subtract a term involving Ḋ r (q 2 )(q 2 − q 2 )
remove the contribution of the pole around the zero q0ℓ 11 11 0ℓ 0ℓ
with

r (q 2 )
dD11
r 2
Ḋ11 (q0ℓ ) = . (18.111)
dq 2 q2 =q2
0ℓ

This derivative can be evaluated numerically in an accurate way as a by product of the search of the zeroes of D11r (A),

A > 0. It can also be calculated by making use of the DR for D11r (A), as we will show below for specific types of N/D

DRs. These two methods perfectly agree.


After this preamble, the integral in Eq. (18.110) becomes
Z
qi2 ∞ 2 ρ(q 2 ) N11 (q 2 ) r 2 qi2 X 2 ρ(qj2 ) r (q 2 )
N11 (qj2 )D21 j
r (q 2 )
N11 (qi2 )D21 i
− − dq 2 2 D21 (q ) = − δq − (18.112)
π 0 r (q 2 )
q (q − qi2 ) D11 π j6=i j qj2 (qj2 − qi2 )D11
r (q 2 )
j (qj2 − qi2 )D11
r (q 2 )
i
M
!
X N11 (q0ℓ2 )D r (q 2 )
21 0ℓ
− 2 − q 2 )Ḋ r (q 2 )(q 2 − q 2 )
,
ℓ=1
(q j 0ℓ 11 0ℓ 0ℓ i

here we have also used Eq. (18.103) with A → q0ℓ 2 > 0.


The final discretized version of Eqs. (18.99) and (18.100), writing the terms involving the unknown function on the lhs

220
and the independent ones on the rhs, is
q
r
ki2 X h m∆22 (ki2 )/4 ∆12 (ki2 ) X  D12 (qj2 )/ −kj2′ + mN12 (qj2 )/4π
[D(ki2 )∆(ki2 )]22 − δkj2′ [D(kj2′ )∆(kj2′ )]22 p q + q δqj2 ρ(qj2 ) r (q 2 )
π2 kj2′ ( −ki2 + −kj2′ ) −kj2′ j
(qj2 − ki2 )(qj2 − kj2′ )D11 j
j′
q
r 2 2
M
X q2 D12 (q0ℓ )/ −kj2′ + mN12 (q0ℓ )/4π i ∆12 (ki2 )ki2 X 2 ρ(qj2 ) h N11 (qj2 )D21
r
(qj2 ) M
X 2
N11 (q0ℓ r
)D21 2
(q0ℓ ) i
0ℓ
− + δqj 2 r (q 2 ) −
qj2 2 )Ḋ r (q 2 )(q 2 − k 2 )(q 2 − k 2 )
(qj2 − q0ℓ 11 0ℓ 0ℓ i 0ℓ j′
π j
qj (qj2 − ki2 )D11 j
2 )Ḋ r (q 2 )(q 2 − k 2 )
(qj2 − q0ℓ 11 0ℓ 0ℓ i
ℓ=1 ℓ=1

∆12 (ki2 )ki2 X 2 ρ(qj2 ) h i


M
X
N12 (qj2 ) 2
N12 (q0ℓ )
=∆22 (ki2 ) − δqj 2 r (q 2 ) − , (18.113)
π j
qj (qj2 − ki2 )D11 j (qj2 − 2 r 2 2
q0ℓ )Ḋ11 (q0ℓ )(q0ℓ − ki2 )
ℓ=1

qi2 X 2 ρ(qj2 ) h N11 (qj2 )D21 i


r XM
r
(qj2 ) r
N11 (qi2 )D21 (qi2 ) N11 (q0ℓ2 r
)D21 (q0ℓ2
)
D21 (qi2 ) + δqj 2 r (q 2 ) − r (q 2 ) −
π qj (qj2 − qi2 )D11 j (qj2 − qi2 )D11 i
2 )Ḋ r (q 2 )(q 2 − q 2 )
(qj2 − q0ℓ 11 0ℓ 0ℓ i
j6=i ℓ=1
q q
r 2 2 2 r 2 2 2
q 2 X 2 [D(kj2′ )∆(kj2′ )]22 X 2 2  D12 (qj )/ −kj ′ + mN12 (qj )/4π qi2 D12 (qi )/ −kj ′ + mN12 (qi )/4π
− i2 δkj ′ q δqj ρ(qj ) r (q 2 ) − r (q 2 )
π −kj2′ (qj2 − kj2′ )(qj2 − qi2 )D11 j qj2 (qj2 − qi2 )(qi2 − kj2′ )D11 i
j′ j6=i
q
r 2 2
XM 2
q0ℓ D12 (q0ℓ )/ −kj2′ + mN12 (q0ℓ )/4π i qi2 X 2 ρ(qj2 ) h N12 (qj2 ) N12 (qi2 )
− 2 = − δq j 2 r − 2 r (q 2 )
2 2 r
qj (qj − q0ℓ )Ḋ11 (q0ℓ )(q0ℓ − qi )(q0ℓ − kj ′ )
2 2 2 2 2 π qj (qj − qi )D11 (qj ) (qj − qi2 )D11
2 2 2
i
ℓ=1 j6=i
M
X 2
N12 (q0ℓ ) i
− r
. (18.114)
ℓ=1
(qj2 − 2 2 )(q 2
q0ℓ )Ḋ11 (q0ℓ 0ℓ − qi2 )

Here two partitions are taken, one along the LHC, {ki2 ∈ [−∞, L], i = 1, . . . , NL } and another along the RHC {qi2 ∈ [0, ∞], i =
1, . . . , NR }.
r r 2
Instead of adding M extra points corresponding to the zeros of D11 to the previous system of equations to evaluate D21 (q0ℓ ), we
2
simplify the numerical procedure by approximate this value by a linear extrapolation between the nearest points qi to the left and to
2 2 2
the right of q0ℓ that we denote by qℓL and qℓR , in order. In this way we implement in Eq. (18.114) that
2 2
r 2 r 2 qℓR − q0ℓ r 2 q 2 − qℓL
2
D21 (q0ℓ ) =D21 (qℓL ) 2 2 + D21 (qℓR ) 20ℓ 2 . (18.115)
qℓR − qℓL qℓR − qℓL

The numerical implementation of the coupled IEs of Eqs. (18.113) and (18.114) is now straightforward.
When the N D11 DRs are used for the D1j (A) functions and the IEs of Eq. (18.113) and (18.114) for D2j we perfectly reproduce
r 2
the results for the once-subtracted N/D case (N D11 ) discussed in Sec. 18.4. To apply these equations we need the derivative Ḋ11 (q0ℓ )
which from Eq. (18.67) is
Z L 2 2

m 2
r 2 2 [D(k )∆(k )]11 −k
Ḋ11 (q ) = − 2 dk . (18.116)
4π −∞ (k 2 − q 2 )2

We discuss next the case with other DRs (higher substracted ones) used to calculate the functions D1j (A). These N/D DRs do not
give rise to a symmetric T -matrix when implenting in the standard coupled-channel N/D method, but they do when employing the
new Eqs. (18.113)-(18.114) for the D2j (A).

18.7 ND12 case: fixing as and rs


We make use of the DR for the function D11 (A) already derived in Eq. (18.80) to guarantee the reproduction of the effective range
rs , together with Eq. (18.67) for the function D12 (A). The DRs for N1j (A) are those of Eq. (18.63). Once the functions D1j (A)
are calculated we employ the IEs of Eqs. (18.99)-(18.100) to calculate D2j (A), according to the method of Sec. 18.5. The derivative

221
r
of D11 (q 2 ) is
r Z L Z L
dD11 (q 2 ) 1 mq 2 [D(k 2 )∆(k 2 )]11  m [D(k 2 )∆(k 2 )]11
2
= − as rs + dk 2 √ 2k 2 − q 2 − 2 dk 2 . (18.117)
dq 2 4π 2 −∞
2 2 2
k −k (k − q ) 2 2 4π as −∞ (k 2 )2
The results are not numerically convergent. We find large variation in the results under changes of the partition along the LHC
and RHC.

18.8 ND22 case: fixing as and rs with one extra parameter called v2
In this section we impose given values for the slope of the mixing partial wave (aε ), as and rs . We use the same DRs for the
functions Nij (A) and Dij (A) as in the previous section except for N12 (A) and D12 (A), that are modified to reproduce aε . The latter
requirement implies that
T12 (A) 4π
−−−→ − aε . (18.118)
A A→0 m
From the basic equation T = D−1 N we have that
D22 N12 − D12 N22
T12 = . (18.119)
D11 D22 − D12 D21
An expansion in powers of A around the origin gives the leading contribution
T12 −−−→ N12 + O(A2 ) = ν12 A + O(A2 ) . (18.120)
A→0

Comparing between Eqs. (18.118) and (18.119) we have that



ν12 = − aε . (18.121)
m
We write down a twice-subtracted DR for N12 (A) that implements Eq. (18.121) and another twice-subtracted DR for D12 (A) in
terms of an extra free parameter δ12 ,
Z
4π A2 L [D(k 2 )∆(k 2 )]12
N12 (A) = − aε A + dk 2 2 2 2 , (18.122)
m π −∞ (k ) (k − A)
√ Z
imA2 L [D(k 2 )∆(k 2 )]12
D12 (A) =δ12 A + iaε A A + 2
dk 2 √ √ . (18.123)
4π −∞ (k 2 )2 ( A + k 2 )
We need an extra subtraction as well for D12 (A) so as to avoid the IR divergent integrals that would appear when exchanging the
order of the integrations along the RHC and LHC, respectively, as already discussed other times along these notes. The values of aε
and δ12 from the N D11 case are
aε = − 1.72 fm3 , (18.124)
δ12 = − 0.24 , (18.125)
while the value of aε corresponding to the Nijmegen PWA is
aε =1.75 fm3 . (18.126)
In all our practical applications we have fixed δ12 to the value in Eq. (18.124), although changes of this parameter keeping its order
of magnitude does not modify the conclusions that follow:
• We have tried to find a solution of the resulting IEs that reproduces the experimental values for as , rs and aε . But the phase
shifts and mixing angle is not smoothly convergent as we increase the cutoff value along the LHC nor when increase the number of
points along the RHC for some values of the LHC cutoff. It is also the case that the function D11 (A) along the RHC is not convergent.
We finally disregard this case.
• We have also tried to impose only the experimental value of rs , while aε takes the value from the N D11 case in Eq. (18.124).
Similar conclusions as in the previous point arise.

222
A Basis for the tensor operators in the N N potential: [4] EGM’s & [5] KBW’s forms
Ref. [4] gives the N N potential as
i E
V = VCE + VσE ~σ1 · ~σ2 + VSL E
(~σ1 + ~σ2 ) · (k × q) + VσL E
~σ1 · (q × k)~σ2 · (q × k) + Vσq E
~σ1 · q~σ2 · q + Vσk ~σ1 · k~σ2 · k , (A.1)
2
with q = p′ − p and k = (p′ + p)/2. However, in this equation there are three tensor operators involved, while two are just enough.
The idea is to use an orthogonal tryad made from the vectors q and k as
q
u1 = (A.2)
|q|
q×k
u2 = ,
|q × k|
q × (q × k)
u3 = .
|q × (q × k)|
Then we have the identity,

~σ1 · ~σ2 = ~σ1 · u1~σ2 · u1 + ~σ1 · u2~σ2 · u2 + ~σ1 · u3~σ2 · u3 , (A.3)


E
which allows to express one of the tensor structures in terms of the other two. In practice we remove the structure proportional to Vσk
E E
in favour of Vσq and VσL . Notice that this result does not depend on whether the scattering is on-the-energy shell or not. Related to
this, one has the relation
2
q × (q × k) = −p(p′ − p · p′ ) − p′ (p2 − p · p′ ) . (A.4)

Therefore, for on-the-energy-shell scattering q × (q × k) is just proportinal to k, but in general this is not the case and one should
deliver the argument taking the tryad in Eq. (A.2).
In this way, we can use the decomposition of the N N potential introduced in Ref. [5], that only involves two tensor structures and
five structures in total,

V = VCK + VSK ~σ1 · ~σ2 + VSO


K
i(~σ1 + ~σ2 ) · (q × p) + VTK ~σ1 · q~σ2 · q + VQK ~σ1 · (q × p)~σ2 · (q × p) . (A.5)

Comparing with the structures in Eq. (A.1) (once the term proportional to Vσk is removed), we have a one to one correspondence
between them, with the Vi related as

VCE = VCK , (A.6)


VσE = VSK ,
E
Vσq = VTK ,
E
VσL = VQK ,
E K
VSL = −2VSO .

For the partial-wave projections we will use the five structures given in Ref. [5] but within the formulas for the more general
off-the-energy shell case given in Ref. [4], taking into account the relations given in Eq. (A.6).

B Imaginary part of the log in v


The imaginary part of
   
ŵ(p1 , p2 ) = log (p1 + p2 )2 + m2π − log (p1 − p2 )2 + m2π (B.1)

for p1 = iν1 + ε+ , p2 = iν2 + ε is given in Table 3, and for p1 = iν1 + ε, p2 = iν2 + ε+ in Table 4, with 0 < ν1,2 < k − mπ and
0 < ε < ε+ (ε+ → 0) in all the cases. For the rest of cases not contemplated in the Tables 3 and 4 the imaginary part is zero. Of

223
Table 3: Imŵ(iν1 + ε+ , iν2 + ε) = πχ, the factor χ is given in the table.
3mπ /2 < k < 2mπ 2mπ − k < ν1 < k − mπ m π − ν1 < ν2 < k − m π
χ = +1
2mπ < k < 3mπ 0 < ν1 < k − 2mπ m π − ν1 < ν2 < m π + ν1 m π + ν1 < ν2 < k − m π
χ = +1 χ = +2
k − 2mπ < ν1 < k − mπ |mπ − ν1 | < ν2 < k − mπ
χ = +1
3mπ < k 0 < ν1 < k − 2mπ |mπ − ν1 | < ν2 < mπ + ν1 mπ + ν1 < ν2 < k − mπ
χ = +1 χ = +2
k − 2mπ < ν1 < k − mπ ν1 − m π < ν2 < k − m π
χ = +1

Table 4: Imŵ(iν1 + ε, iν2 + ε+ ) = πχ, the factor χ is given in the table.


3mπ /2 < k < 2mπ 2mπ − k < ν1 < k − mπ m π − ν 1 < ν2 < k − m π
χ = +1
2mπ < k m π < ν1 < k − m π 0 < ν2 < ν1 − m π
χ = +2
0 < ν1 < k − 2mπ |ν1 − mπ | < ν2 < ν1 + mπ
χ = +1
k − 2mπ < ν1 < k − mπ |ν1 − mπ | < ν2 < k − mπ
χ = +1

course, since ŵ(p1 , p2 ) is symmetric in its two arguments one can derive one table from the other, which is a check that we have
done.
To apply Tables 3 and 4 for negative values of the arguments, one can work out straightforwardly the following relationships

Imŵ(−iν + ε− , iν1 + ε) = −Imŵ(iν + ε− , iν1 + ε) Imŵ(iν + ε− , −iν1 + ε) = +Imŵ(iν + ε− , iν1 + ε)


(B.2)
Imŵ(−iν + ε+ , iν1 + ε) = +Imŵ(iν + ε+ , iν1 + ε) Imŵ(iν + ε+ , −iν1 + ε) = −Imŵ(iν + ε+ , iν1 + ε) ,

by deriving them for the imaginary part of the function ω̂(p1 , p2 ), (B.1), and taking into account that this imaginary part is multiplied
by a polynomial in ν 2 and ν12 , as discussed in the footnote 12.

C List of abbreviations
Direct and inverse Lippmann-Schwinger equation
Lippmann-Schwinger: LS
Lippmann-Schwinger equation: LS equation
On-shell LS equation: (2.1)
Half-off-shell: hfs
Half-off-shell LS equation: (2.2), (2.6)
Inverse Ls equation: (2.11), (2.12).

Analytical properties of the potential


Dynamics cuts: (DCs)
Dynamical cut: (DC)

224
Potential: v(p1 , p2 )
Spectral decomposition: Eq. (3.75)
Spectral function η(µ2 ) Eq. (3.75)
Momentum transfer: q = p2 − p1
Finite cuts (3.84)
Vertical cuts (3.87)
Algebraic discussion of analytical properties of partial wave projection of v(p1 , p2 ) for 1 S0 OPE potential
Analytical extrapolation
Technical: When sin θ is present in the projection
ξ Eq. (3.89)

Right-hand cut
S(k, k ′ , E) (4.1)
Generalized unitarity relation, (4.2)
On-shell and hfs unitarity relation
Relation between new and standard normalization

Partial wave projection


Parity and time-reversal invariance (5.1)
Partial-wave amplitudes (5.2)
hfs unitarity in partial-wave amplitudes (5.4)
t(k, k ′ ; k ′ 2 /m)/(kk
√ ) is a function of k 2 and k ′ 2 .
′ ℓ
√ √
LHC in k 2 , k ′ 2 , RHC in z.
2
Dℓ (k ′ ) (5.11)
Integral Equation IE
Nℓ (k, k ′ ) for hfs scattering (5.12)
NΩ (k, k ′ ) for hfs scattering (5.14)

Dynamical cuts in the half-off-shell amplitude


Deformed integration contour
Vertical cuts from left to right (6.4)
2
Analytical properties of t(k, k ′ ; k ′ /m) in the variable k
Pole of the energy denominator
2
Analytical properties of t(k, k ′ ; k ′ /m) in the variable k ′
Argument to conclude that the t(k, k ′ ; k ′ 2 /m) is unique for complex arguments if it is unique for real and positive k, k ′ and viceversa.

Calculation of the on-shell discontinuity from the LS equation


S waves
Values for LHC, p = ik k > Mπ
Integration contour for p = ik ± ε
Proposition: v(p∗1 , p∗2 ) = v(p1 , p2 )∗ Eq. (7.4)
p∗ ∗
p2
Theorem: t(p∗1 , p∗2 ; m2
) = t(p1 , p2 ; m2 )∗ Eq. (7.5)
Proposition: v(ik − ε, ik − ε) = v(ik + ε, ik + ε)∗ Eq. (7.9)
2
−k2 ∗
Theorem: t(ik − ε, ik − ε; −k m ) = t(ik + ε, ik + ε; m ) Eq. (7.9)
Digression: Properties of v(ik + ε, iν1 + ε ± δ)
Theorem: Ret(iν1 + ε + δ, ik + ε) = Ret(iν1 + ε − δ, ik + ε)
Result: Imt(ik + ε, ik + ε) Eq. (7.32)
Result: Imt(iν + ε+ , ik + ε) , 0 < ν < k − mπ , Eq. (7.36)
Result: Imt(iν + ε, ik + ε+ ) , 0 < ν < k − mπ , Eq. (7.37)

225
Result: IE for Imt(iν + ε − δ, ik + ε) − Imt(iν + ε + δ, ik + ε)
Observation: IE for ∆(A) is finite. Indeed, local counterterms cannot contribute to ∆(A)

General case
n = 2(ℓ + 1) Eq. (7.41)
v̂(k, k ′ ), t̂(k, k ′ ) Eq. (7.42)
Keep the explicit dependence on the infrared divergent factor 1/ν1n
Combinations 1/(ν1 − iε)n ± 1/(ν1 + iε)n Result: Imt(ik + ε, ik + ε) Eq. (7.48)
′2
LS equation for t̂(k, k ′ ; km ) Eq. (7.49)
Result: Imt̂(iν + ε+ , ik + ε) Eq. (7.53)
Result: Imt̂(iν + ε, ik + ε+ ) Eq. (7.53)
ρ(ν 2 , ν12 ) Eq. (7.56)
Even powers in ν for ν → 0 in Imt̂(iν + ε+ , ik + ε) Eq. (7.58)
Not only even powers for ν → 0 in Imt̂(iν + ε, ik + ε+ ) Eq. (7.59)
f(ν) Eq. (7.61)
Result: IE for f(ν)
Absence of infrared singularity in the expression for Imt̂(iν + ε+ , ik + ε) when ν → mπ
Procedure for the numerical solution of the IE for Imt̂(iν + ε+ , ik + ε) removing the infrared singularity in the numerical computation
of integrals
Theorem: Ret̂(iν + ε′ , ik + ε) → ν 2ℓ+1 for ν → 0

One-step process
Result: t(ik − ε− , ik + ε) − t(ik + ε+ , ik + ε) = 2iImt(ik − ε− , ik + ε)
Theorem: Imt(ik + ε, ik + ε) = Imt(ik − ε− , ik + ε)
Theorem: Ret(iν + ε + δ, ik + ε) = Ret(iν + ε − δ, ik + ε) , ν < 0 , δ → 0+
Result: Imt(iν + ε+ , ik + ε) Eq. (7.87)
Result: Imt(iν + ε, ik + ε+ ) Eq. (7.88)
Result: IE for f(ν) Eq. (7.94)
Making sense of f(ν) for mπ + ν < 0
S(ν1 ) Eq. (7.101)
Different equations used to calculate f(ν) for negative ν. Intervals given in Eq. (7.104)
Specific expression to calculate f(−k) Eq. (7.105)
∆v̂(0, −mπ ) = 0 Eq. (7.106)
Accurate numerical procedure by adding and subtracting the ℓ first terms in the Taylor series of f(ν) + f(−ν)

Nonlinear DR for the on-shell partial wave Elastic unitarity, Eq. (9.2)
Nonrelativistic phase space, ρ(A), Eq. (9.3)
Bound state residue, β, Eq. (9.6)
Pole position of bound state, B, Eq. (9.6)
Threshold of LHC, L, Eq. (9.4)
Nonlinear DR for T (A), Eq. (9.7)
Number of subtractions
DR with threshold behavior imposed, Eq. (9.23)
Deeply bound states
Unitary method to solve the nonlinear IE for T (A)
T0 (A), Eq. (9.26)
G(A; −µ2 ), Eq. (9.27)
V0 (A), Eq. (9.28)
Minimizing χ2 for the difference between iterations, Eq. (9.30)
Effective range expansion, ERE
Scattering length, a, Eqs. (9.36), (9.39)

226
Effective range, r, Eqs. (9.36), (9.39)
Determining V0 (A) in terms of ERE parameters
Imposing a and r in the nonlinear IE

Dispersion relations for N (k 2 , k‘2 ). The half-off-shell case


Integration contour Fig. 25
DR in the argument k ′ , Eq. (10.9)
DR in the argument k, Eq. (10.12)
Imposing N (−k, k ′ ) = (−1)ℓ N (k, k ′ ), N (k, −k ′ ) = (−1)ℓ N (k, k ′ ), Eq. (10.15)
Threshold behavior, (−1)ℓ + m = +1, Eqs. (10.16), (10.17)
Fixing the subtraction functions in terms of a few constants, Eq. (10.28)
ai (k) and bi (k ′ ) are entire functions, Sec. (10.2.1)

2
DRs in k 2 and k ′
Integration contour Fig. 26
DR in the variable k ′ 2 , Eq. (10.49)
DR in the variable k 2 , Eq. (10.53)
Fixing the subtraction constants in terms of a few constants, Eq. (10.61)

Calculation of the discontinuity for DRs in the hfs case


Discontinuity t(k ′ − ε + iν1 , k ′ ) − t(k ′ + ε + iν1 ), Eq. (??)
Discontinuity t(k, k − ε + iν1 ) − t(k, k + ε + iν1 ), Eq. (11.25)
Master Equation, Eq. (11.30)
Discontinuity t(k + iν2 − ε, k + iν1 ) − t(k + iν2 + ε, k + iν1 ), Eq. (11.30)

N N 1 S0
One-pion-exchange, OPE

References
[1] I. S. Gradshteyn and I. M. Ryzhik, “Table of Integrals, Series and Products”. Academic Press, 1980.
[2] L.D. Faddeev, “Mathematical Aspects of the Three-Body Problem in the Quantum Scattering Theory”. Academy of Sciences of
the U.S.S.R. Works of the Steklov Mathematical Institute, Vol.69, 1963.
[3] Z.-H. Guo, J.A. Oller and G. Rı́os, Phys.Rev.C 89 (2014) 014002.
[4] E. Epelbaum, W. Glöckle and U.-G. Meißner, hep-ph/0405048.
[5] N. Kaiser, R. Brockmann and W. Weise, Nucl. Phys. A 625 (1997) 758.
[6] D. R. Entem and J. A. Oller, arXiv:1610.01040 [nucl-th].
[7] M. Pavón Valderrama and E. Ruiz Arriola, Phys. Rev. C74, 054001 (2006).
[8] M. Pavón Valderrama and E. Ruiz Arriola, Phys. Rev. C72, 054002 (2005).
[9] A. Nogga, R. G. E. Timmermans, and U. van Kolck, Phys. Rev. C72, 054006 (2005).
[10] J. D. Bjorken, Phys. Rev. Lett. 4, 473 (1960).
[11] J. D. Bjorken and M. Nauenberg, Phys. Rev. 121, 1250 (1960).

227

You might also like