Journal of Physics G: Nuclear and Particle Physics

PAPER • OPEN ACCESS You may also like

- Novel Janus group III chalcogenide
Wave-packet continuum discretisation for monolayers Al2XY2 (X/Y= S, Se, Te):
First-principles insight onto the structural,
nucleon–nucleon scattering predictions electronic, and transport properties
Tuan V. Vu and Nguyen N. Hieu

- Toward establishing the low-lying $P$-

To cite this article: Sean B S Miller et al 2022 J. Phys. G: Nucl. Part. Phys. 49 024001 wave excited $\Sigma_c$ baryon states
Kai-Lei Wang and Xian-Hui Zhong

- Modelling third harmonic ion cyclotron

acceleration of deuterium beams for JET
fusion product studies experiments
View the article online for updates and enhancements. M. Schneider, T. Johnson, R. Dumont et
al.

This content was downloaded from IP address 14.139.221.183 on 27/12/2021 at 12:28

 Journal of Physics G: Nuclear and Particle Physics
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 (28pp) https://doi.org/10.1088/1361-6471/ac3cfd

Wave-packet continuum discretisation for

nucleon–nucleon scattering predictions
Sean B S Miller ∗ , Andreas Ekström and
Christian Forssén
Department of Physics, Chalmers University of Technology, SE-412 96 Gothenburg,
Sweden

E-mail: [email protected] and [email protected]

Received 8 September 2021, revised 23 November 2021

Accepted for publication 24 November 2021
Published 23 December 2021

Abstract
In this paper we analyse the efficiency, precision, and accuracy of comput-
ing elastic nucleon–nucleon (NN) scattering amplitudes with the wave-packet
continuum discretisation method (WPCD). This method provides approximate
scattering solutions at multiple scattering energies simultaneously. We there-
fore utilise a graphics processing unit to explore the benefits of this inher-
ent parallelism. From a theoretical perspective, the WPCD method promises
a speedup compared to a standard matrix-inversion method. We use the chi-
ral NNLOopt interaction to demonstrate that WPCD enables efficient com-
putation of NN scattering amplitudes provided one can tolerate an averaged
method error of 1–5 mb in the total cross section at scattering energies
0–350 MeV in the laboratory frame of reference. Considering only scatter-
ing energies ∼40–350 MeV, we find a smaller method error of  1–2 mb.
By increasing the number of wave-packets we can further reduce the overall
method error. However, the parallel leverage of the WPCD method will be
offset by the increased size of the resulting discretisation mesh. In practice,
a GPU-implementation is mainly advantageous for matrices that fit in the fast
on-chip shared memory. We find that WPCD is a promising method for com-
putationally efficient, statistical analyses of nuclear interactions from effective
field theory, where we can utilise Bayesian inference methods to incorporate
relevant uncertainties.

∗
Author to whom any correspondence should be addressed.
Original content from this work may be used under the terms of the Creative Commons
Attribution 4.0 licence. Any further distribution of this work must maintain attribution
to the author(s) and the title of the work, journal citation and DOI.

0954-3899/21/024001+28$33.00 © 2021 The Author(s). Published by IOP Publishing Ltd Printed in the UK 1
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Keywords: nucleon–nucleon scattering, wave-packet continuum discretization,

chiral effective field theory

(Some figures may appear in colour only in the online journal)

1. Introduction

A large portion of interaction-potential models currently applied in ab initio many-nucleon

calculations are constructed using ideas from chiral effective field theory (χEFT) [1–4]. Such
potentials typically contain 10–30 low-energy constants (LECs), acting as physical calibra-
tion parameters, that must be inferred from data. Bayesian methods for parameter estimation
offer several advantages in this regard, in particular in conjunction with EFTs, see e.g. refer-
ence [5]. However, making reliable inferences typically incur a high total computational cost
due to the large amount of posterior samples. It is therefore important to establish an efficient
computational framework for generating model predictions of physical observables.
In this paper we look to low-energy nucleon–nucleon (NN) scattering cross sections as it
constitutes the bulk part of the standard dataset for inferring the most probable values of the
LECs, see e.g. references [6–9]. The most recent, and statistically consistent, database [10]
of NN scattering cross sections contains data for thousands of measured proton–proton (pp)
and neutron–proton (np) cross sections at hundreds of different laboratory scattering energies,
mostly below the pion-production threshold at T lab ≈ 290 MeV. In most cases, the computa-
tional bottleneck when predicting NN scattering amplitudes for a given interaction potential
comes from obtaining numerical solutions to the Lippmann–Schwinger (LS) equation.
There are essentially three approaches for improving computational efficiency and speed of
a computational procedure: (i) develop improved numerical methods and algorithms tailored
to the physical model at hand and its application, (ii) use specific hardware, e.g. a faster CPU,
increased memory bandwidth, or parallel architectures such as in a graphics processing unit
(GPU) to better handle some of the most dominant computational procedures of the model, or
(iii) replace any computationally expensive model evaluations with a fast, as well as sufficiently
accurate and precise, surrogate model, i.e. an emulator, which mimics the original model out-
put. This latter approach is very interesting and in particular eigenvector continuation (EC)
[11, 12] applied to emulate NN-scattering amplitudes [13, 14] shows great promise, although
uncertainty quantification is yet to be explored. Note, however, that EC emulation attains most
of its speedup when applied to potential models that exhibit linear parameter dependencies. In
cases where such dependencies are not present one might resort to other methods to handle the
non-linear response, such as EC combined with Gaussian process emulation [15]. However,
one should note that Gaussian process emulation [16] and other standard machine learning
methods exhibit poor scaling with increasing dimensionality of the input parameter domain.
Here, we will focus on approaches (i) and (ii) by exploring both the standard matrix-
inversion (MI) method, see e.g. reference [17], and the wave-packet continuum discretisation
(WPCD) method [18] for solving the LS equation. The WPCD method basically corresponds
to a bound-state approach that uses eigenfunctions of the full NN Hamiltonian to approxi-
mate scattering solutions at any on-shell energy. This method is particularly interesting since
it provides approximate scattering amplitudes at multiple scattering energies simultaneously.
We have therefore implemented this inherently parallel method on a GPU. We also note that
WPCD places no constraint on the analytical form of the potential or its parametric depen-
dence. As such, WPCD acceleration for NN scattering complements the EC approaches for
emulation [13–15] mentioned above.
2
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Speeding up computations very often comes at the expense of accuracy and/or precision,
and the WPCD method is no exception to this principle. The magnitude of experimental errors
in the calibration data and the estimated theoretical model-discrepancies [19] provide natural
tolerances for the level of method error that is acceptable. Indeed it is undesirable to have a
method error that dominates the error budget such that it obscures or even hampers the infer-
ence of useful information. We therefore analyse the WPCD method in detail and quantify
realistic computational speedups and compare the method errors to recent estimates of the
model discrepancy in χEFT [20]. We also analyse and compare the numerical complexities of
the MI and WPCD methods.

2. NN scattering

The LS equation, in operator form, for the transition-matrix operator T̂ at some scattering
energy E is given by

T̂(E) = v̂ + v̂ĝ0 (E)T̂. (1)

This is an inhomogeneous Fredholm integral equation where v̂ is some NN-potential operator,

ĝ0 (E) = (E − ĥ0 + i)−1 is the free Green’s function, and ĥ0 is the free Hamiltonian, i.e. the
kinetic energy operator, and i → 0 is the positive imaginary part of the complex energy. Most
realistic NN potentials in nuclear ab initio calculations do not furnish analytical solutions for
the T-matrix. It is however straightforward to numerically solve for the T-matrix for an on-shell
energy E. There exists e.g. variational methods [21–24], as well as Neumann or Born series
expansions for sufficiently weak potentials. One can also try Padé extrapolants [25] in cases
where the integral kernel is not sufficiently perturbative to converge the resulting Neumann
series. In this work we employ the standard MI method [17] which amounts to inverting a
relatively small complex-valued matrix at each scattering energy. This is a trivial operation
that can be straightforwardly carried out within milliseconds on a modern CPU. However,
solving at multiple scattering energies to obtain all cross-sections present in the NN database
amounts to at least a few seconds of computation. In a Bayesian analysis, where one repeatedly
evaluates a likelihood function across a multi-dimensional parameter domain, any speedup in
the solution of the LS equation will directly impact the total computation time.
The momentum-space partial-wave representation of the LS equation for the NN T-matrix,
equation (1), is given by

sJ  sJ  2 ∞
Tl l (q , q; E) = vl l (q , q) + dk k2 vlsJ l (q , k)g0 (k; E)TlsJ l (k, q; E), (2)
π  0
l

where q, q , and k are relative momenta, E = p2 /mN is the centre-of-mass (c.m.) energy with
momentum p, mN is the nucleon mass, and g0 (k; E) ≡ (E − k2 /mN + i)−1 . In this work we
only consider the canonical NN interaction potential, and therefore use the shorthand notation
TlsJ l (q , q) ≡ q , l , s, J|T̂|q, l, s, J for partial-wave amplitudes. Furthermore we use the follow-

ing normalisation of momentum states, k |k = δ(kk−k) k . We suppress isospin notation since the
total isospin T is defined uniquely by the Pauli principle given the spin and angular momentum
quantum numbers s and l, respectively, while the projected isospin T z is defined at the outset
of, and conserved throughout, a scattering observable calculation. Methods for obtaining the
partial-wave-projected potential vlsJ l (q , q) can be found in e.g. reference [26]. Integrated and
differential scattering cross sections at a specific scattering energy E can be straightforwardly
evaluated given the partial-wave T-matrix using the expressions presented in appendix A.
3
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Throughout this work we will compare the efficiency and accuracy of the WPCD method,
presented in section 3.2, to a set of numerically exact results obtained using the MI method
presented below.

2.1. MI method for solving the LS equation

For the resolvent ĝ0 (E), the kernel in the LS equation (2) has a pole singularity at k = p. This
can be handled via a principal-value decomposition1, i.e.
 
1 1
lim =P ∓ iπδ(x), x ∈ R, (3)
→0 x ± i x

such that the remaining integral can be evaluated using e.g. Gauss–Legendre quadrature on
N
some grid {ki }i=Q1 of momenta ki = p with corresponding weights wi . Following [17], the com-
plex T-matrix in equation (2) can be solved via√ the inversion of a finite-dimensional matrix
equation for the on-shell momentum q = p = mN E,

2 
NQ

TlsJ l (q, q; E) = vlsJ l (q, q) + wi vlsJ l (q, ki)g0 (ki ; E)TlsJ l (ki , q). (4)
π  i=1
l

We can introduce a basis qi ∈ {k1 , k2 , . . . , kNQ , p} such that the operators can be written in
matrix form with row i and column j corresponding to qi and q j respectively, i.e. (VlsJ l )i j ≡
qi |vlsJ l |q j . The on-shell T-matrix element is then given by (TlsJ l )NQ +1,NQ +1 . Introducing a vector
D with elements defined as
⎧
⎪
⎪ 2wi ki2 mN
⎪
⎪ if i  NQ ,
⎨ π(ki2 − p2 )
Di ≡ 
NQ (5)
⎪
⎪ 2wi p2 mN iπ pmN
⎪
⎩−
⎪
π(k2 − p2 )
+
2
if i = NQ + 1.
i =1 i

Equation (4) can be rewritten as a linear system of equations,


FlsJ l TlsJ l = VlsJ l , (6)
l

where we have introduced the wave matrix FlsJ l ,

(FlsJ l )i j ≡ δi jδl l − (VlsJ l )i jD j. (7)

Direct inversion of FlsJ l in equation (6) is usually discouraged in scientific computing due to
the instability of MI algorithms [27]. A more advisable practice is to use LU-decomposition.
Additionally, it is numerically more stable to first introduce the K-matrix2 as the principal value
part of the LS equation. Since the potential v is real, the K-matrix is also real. This leads to a
set of purely real matrix-equations [17] based on a nonsingular integral.
The on-shell momentum dependence in FlsJ l demands the solution of an entire linear system
of equations for every energy of interest. Formally, the T-matrix is defined by the potential

1 Kramers–Konig relation, dispersion relation, or the Sokhotski– Plemelj identity.

2 Also referred to as the reactance R-matrix.

4
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

operator via [28]

TlsJ l (q , q) ≡ q |v̂ sJ

l l |ψq ,
+
(8)

where |ψq+ are eigenstates (outbound scattering states) of the full Hamiltonian ĥ with momen-
tum q. With this, we can instead express the LS equation using the full resolvent ĝ(E) =
(E − ĥ + i)−1,

2 ∞
TlsJ l (q , q; E) = vlsJ l (q , q) + dk k2 vlsJ l (q , ψk+ )g(k; E)vlsJl (ψk+ , q), (9)
π  0
l

where vlsJ l (q , ψk+ ) = q |v̂ sJ

l l |ψk
+
and g(k; E) = (E − k2 /mN + i)−1 . This is significantly eas-
ier to evaluate numerically using quadrature since it only amounts to matrix multiplica-
tions. However, the scattering states |ψq+ are not available at the outset. This is where the
WPCD method enters to effectively approximate the scattering states using square-integrable
eigenstates of the NN Hamiltonian.

3. Wave-packet continuum discretisation

The WPCD method [18] effectively eliminates the requirement to explicitly solve the LS
equation at several scattering energies E using MI. Instead, one diagonalises the full Hamil-
tonian in a finite basis, and uses the resulting discrete set of eigenstates to approximate all
scattering states of interest. Equipped with these states, this approach enables straightforward
evaluation of the full resolvent at any value of the on-shell scattering energy E, which makes
the WPCD method intrinsically parallel with respect to obtaining scattering solutions at differ-
ent energies. This is one of several known bound-state techniques to solve the multi-particle
scattering problem [29]. To provide a self-contained presentation, we devote this section to
introduce the WPCD method for describing elastic NN scattering, starting with a definition of
a finite wave-packet basis.

3.1. Scattering observables in a finite basis

Generally we can project some Hamiltonian state |Ψ(E) with positive energy E onto a com-
plete basis (including both bound and free basis states). In this case, the expectation value of an
operator Ô(ĥ) depending purely on the full Hamiltonian ĥ can be represented in the following
form,
nb  ∞
Ψ(E)|Ô|Ψ(E) = u(i )| Ψ|ψi | +
b b 2
dE u(E )| Ψ|ψ(E ) |2 , (10)
i =1 0

where {|ψib }i=b 1 are bound states with energies bi and |ψ(E ) are free states with energy
n

E , both of which are eigenstates of ĥ, and we have defined u(bi ) = ψib |Ô|ψib and u(E) =
ψ(E )|Ô|ψ(E ) . Naturally, we have |Ψ(E) = |ψ(E) (for E > 0) as ĥ is the Hamiltonian
and Ψ the eigenstate, collapsing the expansion above. However, this identity is not useful in
numerical approaches since we do not know ψ(E), leaving us to solve the integral in a finite,
approximative basis.
A computational routine for evaluating integrals, such as quadrature, uses a finite mesh of
points where the integrand is evaluated. In scattering, this approach requires a finite basis of
states |ψ i with corresponding positive energies Ei . These states do not span the whole contin-
uous momentum-space and are thus referred to as pseudostates. Below, we will demonstrate
5
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

how to construct the pseudostates in a wave-packet basis. The pseudostates form a basis for a
finite quadrature-prescription to express an operator

nb

n
Ψ|Ô(ĥ)|Ψ ≈ u(bi )| Ψ|ψib |2 + u(Ei )| Ψ|ψ i |2 , (11)
i =1 i =1

where we introduce quadrature weights w i such that

| Ψ|ψ i |2 = wi | Ψ|ψ(Ei ) |2 . (12)

To determine the weights w i we can use an equivalent quadrature (EQ) technique [30–34] in
which it can be shown that the weights do not depend on the state |Ψ . The weights repre-
sent a type of transformation coefficient between pseudostates |ψ i and the states |ψ(E ) . An
approximate relation based on equation (12) can then be introduced,

ψ i |Ô|ψ i
ψ(Ei )|Ô|ψ(Ei ) ≈ √ . (13)
wi
In the WPCD method we discretise the continuum of free states. A wave packet is defined as
the energy integral over some energy ‘bin’ of width ΔE with Hamiltonian eigenstates |ψ(E)
with positive energy E as the integrand,
 E+ΔE
|ψ(E, ΔE) ≡ dE |ψ(E ) . (14)
E

It follows that limΔE→0 |ψ(E, ΔE) = |ψ(E) . We can define an orthogonal wave-packet basis
by letting the bin boundaries E and widths ΔE lie on a mesh such that the bins do not
N
overlap: {Di | Di ∩ D j = ∅ ∀ i = j}i=WP
1 , where Di ≡ [Ei , Ei + ΔEi ] defines the integral bound-
aries. Note that Ei+1 = Ei + ΔEi . It is straightforward to normalise this basis and show that
the wave packets have eigenenergies ei = Ei + 12 ΔEi .
The expectation value of an operator Ô in some energy bin can now be approximately
represented in a wave-packet basis

Ψ(E)|Ô|ψ(Ei , ΔEi )
Ψ(E)|Ô|ψ(E ) ≈ √ , (15)
ΔEi
where E ∈ Di . As expected, the quality of this approximation is subject to the bin widths
ΔEi . It can be reasoned [35–38], on behalf of equations (13) and (15), that the EQ weights are
approximately given by

wi ≈ ΔEi . (16)

In short, we have a method for approximating the EQ weights in a wave-packet basis, with
which we can express the spectrum of a scattering operator and thereby effectively solve
scattering problems.
We now proceed to approximately represent the pseudostates as Hamiltonian eigenstates in
a wave-packet representation. To this end, we setup a wave-packet equivalent of a partial wave
by generalising equation (14) and thus define a normalised free wave-packet (FWP) as

1
|x i ≡ √ kdk f (k)|k , (17)
Ni Di

6
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

where f (k) is a weighting function and N i is a normalisation constant. We obtain energy or

momentum wave-packets using weighting functions f (k) = 1 or f (k) = μk , respectively,
mN
where μ = 2
is the reduced mass. As above, these two types of wave packets have eigenvalues
 
1
ĥ0 |x i = Ei + ΔEi |x i , (18)
2
 
1
p̂|x i = ki + Δki |x i , (19)
2

where p̂ is the momentum operator. The normalisation of the momentum and energy wave-
packets is given by the bin widths, i.e. N i = Δki or N i = ΔEi , respectively. Operators repre-
sented in a basis of partial-wave plane-wave states are related to the FWP representation via

f (q) f (q) 1
q |Ô|q ≈ x i |Ô|x j , (20)
Ni N j q q
where q ∈ Di , q ∈ D j. In this work we have implemented energy as well as momentum wave-
packets, and have found no significant difference or advantage of either choice.
The eigenstate wave-packets |zi of the full Hamiltonian with positive energy, or scattering
wave packets (SWPs), are given by the eigenvalue equation ĥ|zi = i |zi for energies i > 0.
Similarly, the bound-state3 wave-packets |zbi have energies bi < 0. We obtain them straight-
N
forwardly via diagonalisation in a finite basis {|x i }i=WP
1 of FWPs, where N WP = n + nb . This
operation is the most time-consuming part in the WPCD method to solve the LS equation. It
provides us with a matrix of transformation coefficients C i j ≡ x i |z j such that

H = CDCT , (21)

where D is a diagonal matrix of energies bi and i , and Hi j = x i |ĥ|x j . In order to solve the
LS equation on the form of equation (9), we must define a wave-packet representation of the
outbound scattering states. To define SWPs, we must construct the bin boundaries [Ei , Ei +
ΔEi ] = [Ei , Ei+1 ] of |zi such that the positive eigenenergies are given by

1
i = Ei + ΔEi . (22)
2
This is no different than the wave-packet eigenvalues in equations (18) and (19). This shift
of the FWP energies are indicated in the left panel in figure 1. However, it is not possible to
construct the bin-boundaries Ei exactly since equation (22) only provides n equations while
there are n + 1 boundaries. Therefore, the following scheme [39] can be used to approximate
the bin boundaries of |zi ,

E1 ≡ 0,
1
Ei ≡ (i−1 + i ), (23)
2
1
E n +1 ≡ n + (En − En−1 ),
2

3 Here we change bound-state notation from equation (11) such that |ψ b → |zb . This seems natural since |zb is
k k k
expressed in terms of FWPs from the diagonalisation.

7
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 1. Left: shift in FWP energies ei ≡ Ei + 12 ΔEi to SWP energies i . Right: shift
and splitting of degenerate FWP energies ei into energies i,1 and i,2 shown by solid and
dashed lines, respectively. Note that a solid and a dashed line from two different energies
ei and e j will not cross [18, 40].

such that they yield approximate eigenvalues ¯i ,

1
¯i ≡ Ei + ΔEi ≈ i . (24)
2

In the case of K coupled channels, the FWP energies are degenerate and will be split4 in
the full Hamiltonian eigendecomposition. A FWP with energy ei = Ei + 12 ΔEi will give rise
to K SWPs |zi,κ with energies i,κ , corresponding to each coupled state κ = 1, 2, . . . , K. The
energies are typically ordered such that i,κ < i,κ +1 ∀i, κ [41]. Furthermore, this prevents
mixing between levels with different energies [18], i.e. i,κ <  j,κ +1 ∀i < j. Therefore, we use
the boundary construction scheme in equation (23) such that |zi,κ have boundaries given by
the energies i,κ . Note it is very important to construct the Hamiltonian matrix with degenerate
FWP bases representing each coupled state, i.e. |x i,1 = |x i,2 , see the right panel of figure 1.
The approach presented so far works fine for short-range NN potentials. Likewise, the eigen-
spectrum of the long-range Coulomb Hamiltonian can be straightforwardly described using
wave packets, but these must then be constructed from Coulomb wave functions instead of
free states |q [40] as in equation (17). In WPCD, we automatically ‘smooth’ out the typical
low-momentum singularities presented by the Coulomb Hamiltonian. This means that the for-
malism of WPCD works well for both the short- and long-range parts of the interaction, but it
is necessary to treat them separately. In this work we have not studied Coulomb wave packets
as we only consider neutron–proton scattering.

4
Naturally, K  2 for NN scattering.

8
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

3.2. WPCD-method for solving the LS equation

Following equation (20), we can relate elements of the T-matrix in a continuous partial-wave
basis and a FWP basis via

f (q) f (q) x i |T̂ sJ

l l (E)|x j
TlsJ l (q , q; E) = q |T̂ sJ
l l (E)|q ≈ . (25)
Ni N jqq

If we use the full resolvent ĝ(E), defined in connection with equation (9), we can write

T̂(E) = v̂ + v̂ĝ(E)v̂, (26)

such that in a partial-wave-projected wave-packet basis we obtain

 nb
x i |v̂ sJ
l l |zk zk |v̂ l l |x j
b b sJ
x i |T̂ sJ
l l (E)|x j = x i |v̂ l l |x j +
sJ


E − bk
l k
(27)

n
+ x i |v̂ sJ
l l |zk zk |ĝ(E)|zk zk |v̂ sJ
l l |x j .
l k=1

Note that for a realistic potential we should only have nb = 1 (the deuteron). The full resolvent
is given by the full Hamiltonian, of which |zk are eigenstates. In such a basis we can derive a
closed form expression for ĝ(E), see appendix B. In the WPCD representation of equation (27)
all energy-dependence is straightforwardly evaluated via the resolvent. We can therefore find
an on-shell T-matrix element via simple summation of the scattering wave-packets. This is an
important advantage of using the WPCD method for simulating scattering processes.
Note, however, that the resolvent has a logarithmic singularity for E = Ek or E = Ek+1 . We
handle this by averaging with respect to the on-shell energy E,

1
gi ≡
k
zi |g(E)|zi dE. (28)
ΔEk Dk

The nuclear potential, represented in a FWP basis as

 
1
x i |v̂ l l |x j =
sJ
k kdk dk f (k ) f (k) k |v̂ sJ
l l |k , (29)
Ni N j Di D j

usually vary mildly across a typical momentum-bin Di . It is therefore often sufficient to use a
midpoint approximation to evaluate the integral. This offers a significant reduction in com-
putational cost. In a momentum wave-packet basis, the midpoint approximation is simply
given by

x i |v̂ sJ
l l |x j ≈ k̄i k̄ j Δki Δk j k̄i |v̂ sJ
l l |k̄ j , (30)
k +k
where k̄i = i 2i+1 are the bin midpoints.
Here we summarise the necessary steps to implement the WPCD method for NN-scattering
calculations.
(a) Distribute the bin boundaries for the free wave-packets and choose a weighting func-
tion f (k), see equation (17). Recommended options are e.g. uniform (equidistant),
Gauss–Legendre, or Chebyshev distributions. The results in this work are based on a
9
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

N
Chebyshev distribution. The Chebyshev distribution for N WP points, {y j} j=WP1 , is defined
by [18]
 
2j − 1
y j = α tant π , j = 1, . . . , NWP , (31)
4NWP
where t is a ‘sparseness degree’ and α is a scaling parameter. Let yk be either the momen-
tum or energy bin boundaries, and we then set the initial boundary y0 = 0. For our
simulations we have used momentum wave packets ( f (k) = 1), t = 2, and α = 100 MeV.
(b) Diagonalise the Hamiltonian in a FWP basis. This yields a set of energy eigenvalues i
and accompanying matrix of eigenvectors C as according to equation (21).
(c) Construct bin-boundaries Ei for the pseudostate wave-packets using the set of energy
eigenvalues and equation (23).
(d) Express the full resolvent g(E) in the scattering wave-packet basis, see appendix B. In this
work we use the energy-averaged resolvent from equation (28) to avoid singularities.
(e) Obtain approximate T-matrix elements in the FWP basis by evaluating the LS-equation

Ti j(E) = Vi j + (VC)ik gkk (E)(VC) jk , (32)

where (VC)ik ≡ V i jC jk , Vi j ≡ x i |v̂|x j , gkk ≡ zk |ĝ(E)|zk , and C is given by

equation (21).
The T-matrix can be transformed to a continuous plane-wave basis using equation (20).
However, it is important to note that with energy averaging there is ambiguity in which value
q , q ∈ Di to choose, as the wave-packet representation gives the same value for all choices.
While we discuss this further in section 5.1, we find it more beneficial to choose q , q such that
they match the wave-packet eigenvalues, given in equations (18) and (19).

4. Numerical complexities of the MI and WPCD methods

Here we present the minimum number of floating-point operations (FLOP) required for com-
puting on-shell q|TlsJ l |q matrix elements at nE different values of the scattering energy. We
focus on the MI and WPCD methods, presented in sections 2.1 and 3.2, respectively. We also
assume that the potential matrix p |VlsJ l |p is pre-computed and available in memory at the
outset. Typically, in both methods we use matrices of sizes n × n for n < 100. To avoid con-
fusion, we let N Q denote the number of quadrature points in the case of the MI method, while
N WP denotes the number of wave packets in the case of the WPCD method. We retain n to
symbolise a basis size in general, regardless of method.

4.1. MI complexity

Given a quadrature grid with N Q points, the MI method first requires setting up the wave matrix
(7) at each scattering energy. Naively, the complexity of the matrix construction is dominated
by the matrix–matrix product of the potential V with the resolvent g0 . Since the resolvent
is diagonal, it is more efficient to multiply each row of V with the diagonal element of g0 ,
yielding n + 1 scalar–vector multiplications in a single F-matrix construction with numerical
complexity according to

O(F) = 2(NQ + 1)2 + 2(NQ + 1), (33)

where the factor of 2 is due to g0 being complex.

10
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

The on-shell T-matrix element is obtained from the last element of the T-matrix, i.e.
NQ +1

T(q, q; E) = TNQ +1,NQ +1 = FN−1Q +1,i Vi,NQ +1 , (34)
i =0

where we have expressed it using MI. MI typically requires O(n3 ) operations for an n × n
matrix. However, solving the LS linear system,

FT = V, (35)

is more advisable, and can be done very efficiently in two steps: firstly, we perform a lower-
upper (LU) decomposition, where a square matrix A is expressed as the product of a lower-
triangular matrix L with an upper-triangular matrix U,

A = LU, (36)

which requires O 23 n3 operations. The decomposition allows for AX = B to be solved for
each column of X using O(2n2 ) operations. For complex matrices these numbers increase by a
factor 4 for both routines.
In summary, the MI-method has a total computational complexity of (note the linear scaling
with the number of on-shell energies nE )
 
4 3 2
OMI (T) = 2nE (NQ + 1) + 5(NQ + 1) + NQ + 1 , (37)
3
 
2
OMI (K) = nE (NQ + 1)3 + 5(NQ + 1)2 + NQ + 1 , (38)
3

where OMI (K) shows the cost of using a K-matrix approach instead. The difference is simply
replacing the cost of doing complex calculations with real calculations. We see there is roughly
a factor 4 speedup in using a K-matrix representation instead of a T-matrix representation.

4.2. WPCD complexity

The complexity of the WPCD-method is dominated by three kinds of linear algebra tasks (i)
the Hamiltonian matrix diagonalisation, (ii) two matrix-matrix products, and (iii) one matrix
addition (see steps (a)–(e) in section 3.2). We let N WP be the number of wave packets and nE
the number of scattering energies as before. By ‘parallelism’ we refer to the fact that WPCD
can solve the scattering problem for several energies at once, following a single Hamiltonian
diagonalisation.
For this reason we have fully implemented the WPCD method on a GPU utilising the CUDA
interface with the cuBLAS [42] and cuSOLVER [43] libraries for linear operations. A more
detailed outline of the GPU code is provided in appendix C. Sequential algorithms are usually
easier to handle, and not all hardware allow for optimal parallelisation due to, for example,
slow computer memory transfer. Thus, we present complexity models for both the sequential
and parallel energy-evaluations of the WPCD method. Note, however, that this is simply a
comparison of FLOP models that do not account for memory transfer times or detailed proces-
sor architecture. We emphasise that all WPCD results presented in this paper were calculated
using our parallel GPU-code, and that the sequential complexity is presented purely to provide
further insight.
11
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

The MI method can also be implemented efficiently on a GPU. This approach will not
exhibit the same inherent parallelism with respect to multiple on-shell calculations. However,
there does exist efficient parallel routines for solving linear systems on the form of equation (6).

4.2.1. Sequential complexity. The sequential mode for the WPCD method means solving the
LS equation for each on-shell energy in sequence rather than simultaneously. This suggests
using a CPU, rather than a GPU, since CPUs typically have a faster clock frequency and can
thus iterate through the on-shell energies faster than a GPU. Of course, CPUs today are mul-
ticore and with several processing units, but note that they might not have a number of cores
equal to or greater than nE . The analysis below assumes a scenario where a processing unit is
working with a single computational core.
(a) A real-valued NN Hamiltonian is efficiently diagonalised using QR factorisation or a
divide-and-conquer algorithm, both of which are usually used together with House-
holder transformations. The divide-and-conquer algorithm has a complexity of O 83 n3
for getting both eigenvectors and eigenvalues of an n × n matrix.
(b) Next, we evaluate the matrix-matrix product of the potential matrix V and the coeffi-
cient matrix C in the rightmost term in equation (32). Square matrix-matrix multiplication
requires 2n3 − n2 FLOP in a straightforward, sequential approach.
(c) The LS equation for Tii (E ∈ Di ) is evaluated via simple summation and multiplication.
The sum,

NWP
(VC)i jg j j(E ∈ Di )(VC) ji , (39)
j=1

involves two multiplications per term, and the sum will require N WP − 1 additions. There
is also the addition of the first term in equation (32). These operations must be done for
every on-shell energy, resulting in a complexity of 6N WP × nE FLOP. Note that if we use
energy averaging we only require a single evaluation for each on-shell wave packet, giving
an upper limit nE  N WP .
The WPCD sequential complexity is then given by
3
OWPCD,seq. (T) = 6NWP − NWP
2
+ 6NWP × nE . (40)

To summarise, the overall complexity of the MI method scales linearly with the number of
scattering energies nE , see equation (37). For the WPCD method we get all the on-shell energy
scattering-solutions from a single Hamiltonian diagonalisation, giving a very ‘cheap’ scaling
with nE .

4.2.2. Parallel complexity. The parallel WPCD approach is based on simultaneous solutions
of the LS equation for all on-shell energies. Contrary to the sequential approach, we assume a
sufficiently large set of processors (like in a GPU) to handle all on-shell energies at once. This
will remove the factor nE in the complexity model (40) such that the cubically-scaling term will
clearly become the dominating one. It can be combatted by making use of hardware-specific,
massively parallel algorithms for matrix diagonalisation and matrix–matrix multiplications.
Here we present the parallel approach to the same steps as above, and in the same ordering.
We assume we have p compute processors, or threads, available.
(a) The parallel cyclic-order Jacobi method is a parallel approach to the Jacobi eigenvalue
algorithm: a method based on finding a similarity transformation of a matrix to its diago-
nal form by repeated Jacobi rotations (see e.g. reference [44]). This method has held major
12
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

appeal for parallel computing due to its inherent parallelism compared to other eigenvalue-
finding routines. However, parallel optimisation of the basic algorithm depends greatly on
how a set of processors is organised with regards to communication and memory. There-
fore, there is no general complexity model for the method—any algorithm should be
written with a specific hardware inmind. One  approach can be found in reference [45]
n3
with a demonstrated complexity O p log n for convergence for a symmetric and real
n × n matrix.
(b) Parallel matrix–matrix multiplication algorithms is an ongoing field of research (see e.g.
reference [46]). While there exist very efficient methods, such as Cannon’s algorithm
[47], they depend highly on hardware and matrix characteristics (sparsity, symmetry, etc).
The divide-and-conquer approach is a very general and straightforward way to parallelise
2
matrix multiplications. In essence, we divide the matrices into M = np submatrices
of sizes m × m = p. If combined with on-chip shared memory between the processors,
this permits all processors to simultaneously work on calculating the product while min-
imising processor-to-processor communication time. The method is explained excellently
in literature such as [44]. This simple method has a theoretical minimum complexity of
O (n × M) when performed in parallel5 .
(c) There is limited parallel optimisation to be gained in equation (39). We can multiply each
term of the summation in parallel, and do the whole summation sequentially for the sake
of simplicity. The result is a complexity of O(2(N WP + 2)) FLOP. Note, however, that the
inherent parallelism of the WPCD method allows us to do the summation simultaneously
for all E and thereby effectively removing the factor nE seen so far in the complexity
models.
In conclusion, assuming p > N WP and adequately large shared memory (see point 2 above),
a somewhat optimal and parallel WPCD complexity model is given by
 3

NWP
OWPCD,par. (T) = log(NWP )
p
 2  (41)
NWP
+ NWP + 2(NWP + 2).
p

We see from figure 2 that the efficiency of the parallel WPCD approach scales very well with
the number of available processors. We also see that for a single processor (p = 1) the parallel
and sequential approaches are roughly equal (while not taking into account any overhead in
memory transfers), while for a realistic value p = 1024 the complexity model for the parallel
approach demonstrates a clear advantage. The value p = 1024 corresponds to the current limit
on threads with shared memory for typical Nvidia GPUs.

5. Continuum-discretised neutron–proton scattering computations

In this section we present a detailed analysis of the precision and accuracy of the WPCD method
for computing neutron–proton (np) scattering observables and phase shifts. For all calculations,

5 This is not the full picture. Often, a set of processors is divided in groups of processors with limited shared memory

that cannot fit all p elements from each matrix in the matrix–matrix product, and one has to define more submatrices
than given by M. This introduces yet another complication to the parallel complexity model that has to do with shared
memory size, which we will not account for here.

13
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 2. Complexity, as measured in FLOP, for the MI and WPCD methods as a func-
tion of n, where n represents the number of quadrature points (N Q ) for MI and the number
of wave packets (N WP ) for WPCD. Furthermore, nE is the number of on-shell T-matrix
evaluations (note nE = n = N WP for WPCD).

we employed the optimised next-to-next-to-leading-order chiral potential N2LOopt [48]. The

primary goal is to analyse the trade-off between minimum computational cost and maximum
method accuracy in the WPCD method, and contrast this to the conventional MI method. Note
that there is no problem in obtaining highly accurate results from either method. We simply
focus on the performance and accuracy of the WPCD method as we reduce the number of wave-
packets such that all objects fit in the fast on-chip shared memory on the GPU. In our analysis
we consider the MI method with N Q = 96 Gauss–Legendre points, see equation (4), to yield
virtually exact results. For brevity, we will refer to such numerically converged calculations as
exact.

5.1. Computing phase shifts

In figure 3 we compare np scattering phase shifts in the 1 S0 , 3 P0 , and 3 S1 partial waves as

well as the 1 mixing angle, as obtained from the WPCD method using N WP = 32 and N WP =
64 momentum wave-packets. The FWP bin boundaries follow a Chebyshev distribution with
scaling factor α = 100 MeV and sparseness degree t = 2, see equation (31), and the resolvent
was energy-averaged according to equation (B.7).
For the WPCD-result, one immediately observes a discretised, step-like version of the oth-
erwise smoothly varying exact description of the scattering phase shifts. This is entirely due
to the momentum discretisation and finite number of wave packets. Indeed, due to the energy-
averaging we obtain only one T-matrix element per momentum bin. In the limit N WP → ∞,
and everything else equal, we will recover the exact result obtained with the MI method. We
observe this convergence trend already when going from N WP = 32 to N WP = 64 wave packets.

14
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 3. Phase shifts calculated using the WPCD-method using N WP = 32 (blue) and
N WP = 64 (orange) wave packets. Numerically exact results (black) were obtained using
the MI-method.

For instance, the mixing angle moves closer to the exact results when increasing N WP . Over-
all, the WPCD method is performing as expected, but there are two features we would like to
point out:
(a) The values of the low-energy 1 S0 phase shift are overestimated near the peak where the
phase shift turns over. This is likely due to the momentum-averaging of operators. The
potential matrix in the 1 S0 channel ( q |V1 S |q ) is shown in figure 4 for both a continuous
0
momentum-basis and a N WP = 32 wave-packet basis with bins distributed according to
equation (31). The potential is constant within each wave-packet bin as expected according
to equation (29). This makes it challenging to reproduce finer details of the interaction.
There is also a discrepancy between the continuous and wave-packet values when the
chosen q -momentum is not near any bra-state bin midpoint. This is most distinctive in
the green curves at q = 205.622 MeV, which is very close to a bin boundary. We see
in equation (29) that we average over momenta within two bins, and when a potential
varies strongly within a bin such that the matrix elements at the bin boundaries are quite
different from the bin mid-point values, this averaging is too coarse to mimic the potential
accurately. This effect becomes less significant with increasing N WP since the grid will
grow denser.
(b) The WPCD-results for 3 S1 show a distinctive drop at 90 degrees. This trend is consis-
tent for all basis sizes and is simply due to the treatment of the deuteron bound state
in the WPCD framework. We extract a phase shift δ via an inverse trigonometric func-
tion with values for δ ∈ [−90, 90] degrees. A bound state is characterised by a transition
15
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 4. The 1 S0 potential of NNLOopt [48] as a function of q, at several q values.

The dashed lines represent the potential calculated in a continuous basis, while the solid
lines represent a wave-packet projection. The wave-packet projection was made with an
N WP = 32 basis in a Chebyshev distribution (with α = 100 MeV). In the wave-packet
representation, the bra-state bin contains q . Note that the units for continuous represen-
tation of the potential is in units of MeV−2 while the wave-packet projected potential is
in units of MeV, as follows from the definition in equation (17).

δ(E + ) − δ(E) = 180 degrees at some energy E for an infinitesimal step  > 0, as dic-
tated by Levinson’s theorem [49]. It is apparent that this transition across 90 degrees is
more difficult to reproduce for the WPCD method. The same effect also gives rise to
inaccuracies in the 1 mixing angle. However, this is of no concern when computing a
scattering observable.

5.1.1. Linearly interpolating phase shifts. Although the resolution of the WPCD method is
limited by the energy-averaging across each momentum-bin—i.e., yields a step-like character
of the predictions—the mid-points q̄i = 12 (qi−1 + qi ) of each bin in figure 3 are typically closest
to the exact results, as expected from the wave-packet eigenvalues, see equations (18) and (19).
We can therefore linearly interpolate the phase shifts δi ≡ δ(q̄i ) across several bins, i.e. across
scattering energies, via
   
δi − δi−1 δi − δi−1
δ(q) = q + δi−1 − q̄i−1 , (42)
q̄i − q̄i−1 q̄i − q̄i−1
for c.m. momenta q ∈ [qi−1 , qi ] where qi are the FWP bin boundaries. This simple and straight-
forward approach offers a rather precise prediction for any on-shell scattering energy. Of
course, the phase shifts can be linearly interpolated using other points, i.e. we can let
n
q̄i = qi−1 + (qi − qi−1 ), (43)
m

16
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 5. Phase shifts for the 1 S0 partial wave calculated using WPCD. The colour cod-
ing is the same as in figure 3, i.e. N WP = 32 (light blue) and N WP = 64 (light orange).
The bands indicate the maximum discrepancy due to the choice of interpolation point n
according to equation (43).

for some n ∈ [0, m], such that n = m2 is the bin mid-point again. Figure 5 shows the resulting
predictions using linear (mid-point) interpolation of the 1 S0 phases presented in figure 3. The
bands estimate the effect of varying the interpolation point. The bands span the resulting δ(q)
calculated with m = 10 and n = [0.1, 1, 2, . . . , 8, 9, 9.9] in equation (43). As expected, mid-
point interpolation yields results that are very close to the exact calculation. In the following
we will therefore only use this interpolation choice.

5.2. Computing cross sections

We compute scattering cross sections from scattering amplitudes according to the method out-
lined in appendix A. Figure 6 shows the total cross section obtained using the WPCD-method
using momentum-space bins with N WP = 32 and N WP = 64. As can be seen in the figure, the
linearly-interpolated results reproduce the exact result rather well. On a larger scale it is nearly
impossible to tell any difference between results obtained using N WP = 32 bins and N WP = 64
bins. To emphasise the monotonically increasing accuracy of the WPCD method as we increase
the number of momentum-space bins, we calculate the absolute value of the difference between
the exact and the WPCD prediction for the total cross section for a range of bin resolutions,
see figure 7. From this it is apparent that the WPCD method converges, although slowly, for a
bin partition following a Chebyshev distribution.

5.3. WPCD accuracy

There are primarily two approximations that impact the accuracy of the WPCD method: the
number of bins N WP and the distribution thereof. Also, the averaging of continuous states into
wave packets means we use momentum-averaged matrix representations of operators in the LS
equation. This averaging should improve with reduced bin widths. We can easily control N WP ,
and in this section we analyse the performance of WPCD with respect to different choices of
this method parameter.
To better quantify the convergence of the WPCD method we study the root-mean-square
error (RMSE) with respect to the exact result for the total cross section across a range of
17
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 6. Total cross section calculated with the WPCD method. The solid black line
corresponds to the exact result as obtained using the MI method. The inset demon-
strates the high accuracy of (mid-point) interpolated WPCD results. On this scale, the
two separate WPCD predictions appear to overlap completely.

Figure 7. Absolute values of the relative difference between exact results and WPCD
method for the total cross section, shown as a function of wave-packet basis size N WP
and laboratory kinetic energy.

scattering energies. We use the standard RMSE measure

 2
nE
i=1 σexact,i − σmethod,i
RMSE = , (44)
nE

where σ exact,i denotes the exact results and σ method,i denotes either the WPCD- or MI-calculated
total cross sections at some scattering energy Ei for i = 1, . . . , nE .
We find that the RMSE for WPCD with scattering energies corresponding to labora-
tory kinetic energies 40  T lab  350 MeV remains fairly constant at ∼2.0 mb when using
N WP = 16. This is interesting for three reasons:
• The WPCD coupled channel Hamiltonian will be of size 2N WP × 2N WP , i.e. we diago-
nalise 32 × 32 Hamiltonian matrices. Most GPUs today, including the Nvidia Tesla V100
18
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure 8. Measured wall times for solving the LS equation for different values of total
cross section RMSE for T lab ∈ (0, 350] MeV (top panel) and T lab ∈ (40, 350] MeV (bot-
tom panel). nE is the number of on-shell energies (interpolation points) at WP bin mid-
points, according to a Chebyshev distribution. The numeric label attached to each data
point indicates the number of Gauss–Legendre grid points N Q for the MI method and
the NWP N WP for the WPCD method. The number of energies used for interpolating the
MI results are indicated in the legend. See the main text for hardware specifications.

we have used here, have 64 kB shared memory (also called ‘on-chip’ memory) which is
significantly faster to access than the GPU’s main memory. These matrices fit entirely
on the GPU shared memory, allowing for a strong reduction in GPU memory read/write
demand while performing the diagonalisation.
• A recent Bayesian analysis of chiral interactions suggests that the χEFT truncation error
for scattering cross sections at next-to-next-to-leading-order is at least 2 mb, at 68%
degree-of-belief (DoB), and at least 5 mb at 95% DoB [20].
• Regarding experimental uncertainties, the combined statistical and systematic uncertain-
ties in measured np total cross sections are at the 1% level [50, 51]. In absolute terms, this
amounts to uncertainties of the order 1 mb at laboratory scattering energies 40 MeV.
At lower energies the cross section increases of course, leading to slightly larger absolute
experimental errors.
In summary, we find that the WPCD method error, when using very few wave-packets, is
slightly larger than typical experimental uncertainties but smaller than the estimated model
discrepancy up to next-to-next-to-leading-order in χEFT.
One can clearly reduce the WPCD method error by increasing N WP . However, this will
also increase the computational cost and we recommend studying the actual wall time cost in
relation to, e.g. the RMSE measure defined above. We perform such an analysis in section 6.
Before ending this section on the method accuracy, we would like to emphasise that the
maximum total angular momentum J max in the partial-wave expansion of the potential also
impacts the accuracy of the description of the scattering amplitude. For the linearly-interpolated
WPCD method with N WP = 16, where we observe a method error of ∼2 mb in the total cross
section at T Lab  40 MeV, and find it unnecessary to go beyond J max = 6.
19
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Table 1. Measured wall times, corresponding to the results in figure 8, for solving the
LS equation and predicting total cross sections using the WPCD and MI methods, at
different RMSE levels in the T lab ∈ (40, 350] MeV energy region. The nE parameter
indicates the number of interpolation energies used for the MI method.
MI WPCD
RMSE (mb) NQ nE Time (ms) N WP Time (ms)
3.0 8 16 6 — —
2.5 10 16 7 — —
2.0 12 16 8 16 0.5
1.5 14 16 10 24 2.5
1.0 16 24 15 32 6
0.5 16 32 20 96 35

6. Method performance

With an account of the precision and accuracy of the WPCD method we now profile its time
performance. Typically, when using the MI method, we calculate every on-shell phase shift
of interest by explicitly solving the LS equation, while in the WPCD approach we interpolate
in-between bin mid-points, as discussed above. This will, unsurprisingly, induce a substantial
time-performance penalty in using the MI method, due to the linear scaling with the number
of energy solutions, as shown in equation (37). However, we can of course linearly interpolate
the phase shifts calculated with the MI method, rather than invoking an explicit calculation at
every on-shell energy. Therefore, to facilitate a balanced comparison between the two methods,
we also employed linear interpolation to extract phase shifts when using the MI method. In our
studies, this has turned out to be a highly efficient way to speed up the calculation with the MI
method while maintaining precision and accuracy of the results.
To facilitate a comparison, when using the MI method we solve the LS equation at
nE on-shell energies also following a Chebyshev distribution. We then linearly interpo-
late the phase shifts using these energies to calculate neutron–proton total cross sections
in the T lab energy ranges (0, 350] MeV and (40, 350] MeV for the calculation of RMSE
values.
For WPCD we can only vary the number of wave packets N WP while for MI we can vary
both N Q and nE , i.e. the number quadrature points and the number of interpolation points or
on-shell energies (at bin mid-points), respectively. Figure 8 shows the wall times for solving
the LS equation to obtain cross sections, at different levels of method accuracy measured by the
RMSE value. In table 1 we show a few interpretations of the figure for a handful of relevant
method parameters. As mentioned, the WPCD method is implemented on a GPU while the
MI results were obtained using an optimised CPU implementation [7]. The time profiles were
obtained using an Nvidia Tesla V100 32 GB SMX2 and an Intel Xeon Gold 6130 for the WPCD
GPU and MI CPU results, respectively.
From our analysis we conclude that the WPCD method is faster than the MI method if one
can tolerate ∼1 − 5 mb method RMSE in the prediction of total scattering cross sections. For
such applications it would then be advisable to use N WP  48 bins, on the basis of figure 8. It is
worth noting that the RMSE is dominated by contributions from scattering cross sections below
laboratory kinetic energies T lab ∼ 40 MeV. Indeed, with N WP = 48 we obtain an RMSE value
of ∼4 mb across an interval T lab ∈ (0, 350] MeV. The RMSE drops to ∼0.8 mb when consid-

20
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

ering only cross sections with T lab > 40 MeV. Irrespective of how the WP bins are distributed,
it is therefore necessary to ensure a sufficiently high density of bins below T lab = 40 MeV to
accurately reproduce the 1 S0 phase-shift peak and the 3 S1 bound state. With the WPCD method
we can obtain increasingly faster solutions to the LS equation as we reduce the number of
wave-packets.

7. Summary and outlook

In this study we have compared the WPCD method with the standard MI method, with an
emphasis on their respective efficiency, precision, and accuracy when solving NN scattering
problems. This study is done with an interest in reducing the computational costs of Bayesian
inference analyses of nuclear interaction models.
We find that the WPCD method, with GPU-acceleration, is capable of providing scattering
solutions to the NN LS equation much faster than a conventional MI method implemented
on a CPU. This is largely due to the WPCD method being capable of delivering scattering
amplitudes at several on-shell energies in an inherently parallel fashion, and that linear inter-
polation across several WP bins offers straightforward predictions for any scattering energy.
However, compared to solutions obtained using the standard MI method, the WPCD method
is less accurate, in particular in regions where the scattering amplitudes vary strongly with
energy. This is also expected due to the momentum-space discretisation and approximation
of the scattering states. Nevertheless, in applications where a certain method error can be tol-
erated—as for example in low-order EFT predictions of NN scattering cross sections—the
GPU-implemented WPCD method presents a computational advantage. This finding makes
it particularly promising for computational statistics analyses of EFT nuclear interactions
utilising Bayesian inference methods.
We also find that the computational gain of using GPU hardware is limited by the amount
of shared memory that is available. This constraint basically corresponds to an upper limit
on the number of WP bins that can be used for efficient computations. We note, however,
that GPU hardware is continuously improved, and that a four-fold increase in the size of the
fast shared memory on the GPU would enable a two-fold increase of both the WP-basis and
method accuracy while incurring a very mild additional computational cost. The GPU global
memory is typically not saturated in WPCD calculations of NN scattering calculations. The
WPCD method can also be applied to compute three-nucleon scattering amplitudes [37, 52].
However, the limited amount of global memory might become a constraining factor for such
applications that involve a more complicated Hilbert space basis.
In this study we have focussed on the inherent parallelism of the WPCD method and the
opportunities that it offers to benefit from the use of GPU hardware. Of course, one could
also explore parallelisation of the WPCD method on a CPU and make parallel use of the fast
cores available in modern CPUs. The efficiency of the GPU WPCD approach is almost fully
determined by the time needed to diagonalize the channel Hamiltonian and could therefore
benefit from the faster CPU clock speed. Naturally, one could also consider a GPU imple-
mentation of the MI method wherein the LS equation is solved in a parallel fashion for mul-
tiple scattering energies simultaneously. Still, this would require multiple matrix inversions
(one for each interpolation energy). A simple phase-shift interpolation applied to the MI
method facilitates sufficiently accurate results for NN scattering observables, at a significantly
lower computational cost compared to explicitly solving the LS equation at every scattering
energy of interest.

21
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Acknowledgments

This work was supported by the European Research Council (ERC) under the European
Unions Horizon 2020 research and innovation programme (Grant Agreement No. 758027), and
the Swedish Research Council (Grant No. 2017-04234). The computations were enabled by
resources provided by the Swedish National Infrastructure for Computing (SNIC) at Chalmers
Centre for Computational Science and Engineering (C3SE) and the National Supercomputer
Centre (NSC) partially funded by the Swedish Research Council.

Data availability statement

The data that support the findings of this study are available upon reasonable request from the
authors.

Appendix A. Scattering cross sections and the spin-scattering matrix

A scattering observable of some operator Ô can generally be written as a trace [28],

Tr{Mρ̂i M † Ô}
Ô = , (A.1)
Tr{Mρ̂i M † }

where ρ̂ is the spin-density matrix, usually represented in a helicity basis, and M is the spin-
scattering matrix.
The helicity-basis projection of the spin-scattering matrix is written as Mms s ,ms , where s is
the conserved total spin and ms is the corresponding projected spin. These matrices are usually
expressed in partial-wave expansions,

2π  2l + 1
Mms s ,ms = [1 − (−1)l+s+TZ ]
ip  4π
J,l,l

× Yml s −ms (θ, φ)
(A.2)
× l , s, ms − ms , ms |l , s , j, ms
× l, s, ms , 0|l, s, j, ms
 
× Sl1J l (p , p) − δl,l ,

where the third and fourth rows are Clebsch–Gordan coefficients, J is the total angular momen-
tum, l and l are the orbital angular momenta of the inbound and outbound states respec-
tively, T Z is the azimuthal isospin projection, Yml (θ, φ) is an azimuthal spherical harmonic,
and Sl1J l (p , p) = p , l , s, J|S|p, l, s, J is the usual scattering matrix.
The on-shell partial-wave-projected S-matrix is related to the on-shell T(p, p; E)-matrix
element via

SlsJ l (p, p; E) = 1 − 2πiTlsJ l (p, p; E). (A.3)

22
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Due to the conservation of probability, the S-matrix is unitary, and can therefore be
parametrised by real parameters like phase shifts. In the Stapp convention [53], the coupled-
channel S-matrix Sl1J l is then given by
 
cos(2J )e2iδ−,J i sin(2J )ei(δ−,J +δ+,J )
S= , (A.4)
i sin(2J )ei(δ−,J +δ+,J ) cos(2J )e2iδ+,J

where δ ±,J refers to δ l,J for l = J ± 1 and J is the mixing angle. The uncoupled-channel S-
matrix Sll0J is given by (for l = J)

S = e2iδJ,J . (A.5)

The angle θ denotes the scattering angle between the c.m. inbound and outbound relative
momenta p and p respectively, while φ is the rotation angle of p around the inbound momen-
tum p, but cylindrical symmetry allows us to set φ = 0. Using equation (A.2) we can calculate
scattering observables such as the differential cross section,
dσ 1
= Tr(MM † ). (A.6)
dΩ 4
However, this approach does not utilise symmetries to reduce the number of non-contributing
terms of Mms s ,ms . Instead, the M-matrix can be expressed in terms of non-vanishing spin-
momentum products after some consideration of parity conservation, isospin and time-reversal
symmetries, and the Pauli principle. We use the Saclay convention [54] for these terms,
1
M= [(a + b) + (a − b)(σ1 · n)(σ2 · n)
2
+ (c + d)(σ1 · m)(σ 2 · m) (A.7)
+ (c − d)(σ1 · l)(σ2 · l)
+ e((σ1 + σ 2 ) · m)] ,

where a, b, c, d, and e are the Saclay amplitudes, σ i are the Pauli spin matrices acting on nucleon
i = 1, 2, and where we define the following unit vectors
p + p p − p p × p
l≡ , m≡ , n≡ . (A.8)
|p + p | |p − p | |p × p |
The Saclay amplitudes are given by the spin-projected matrices as,
1 1 1 1
a= (M + M0,0 + M1,−1 ),
2 11
1 1 0 1
b = (M11 + M0,0 + M1,−1 ),
2
1 1
c = (M11 − M0,0
0 1
+ M1,−1 ), (A.9)
2
1
d = −√ 1
(M1,0 1
+ M0,1 ),
2 sin(θ)
i
e = √ (M1,0
1
− M0,1
1
).
2 2

23
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

In this parametrisation, the differential cross-section is given by

dσ 1 
= |a|2 + |b|2 + |c|2 + |d|2 + |e|2 . (A.10)
dΩ 2
For the results in this paper we used the following expression for the total cross section:
2π
σtot = Im(a + b). (A.11)
p
See reference [54] for a complete account of scattering observables in the Saclay
parametrisation.

Appendix B. The resolvent in a wave-packet basis

The resolvent ĝ(E) for the full Hamiltonian ĥ ≡ ĥ0 + v̂,

1
ĝ(E) ≡ , (B.1)
E − ĥ ± i
can be calculated analytically in a pseudostate wave-packet basis {|zi }ni=1 of the full Hamil-
tonian. The resolvent is represented in the basis by (see equation (17))
1
zi |ĝ(E)|z j = zi | |z j
E − ĥ ± i
  √ (B.2)
1  k k k k ψk(+ ) (+)
 |ψk
= dk dk 2 ,
μ Ni N j Di D j E − k2μ ± i

mN k
where μ = 2
. Note that we set the weight function f (k) = μ
and normalisation Ni = ΔEi

 |ψk
as these are energy wave-packets (from the diagonalisation of ĥ). Using ψk(+ ) (+)
= δ(kk−k)
k
,
this becomes

δi j k
zi |ĝ(E)|z j = dk k2
, (B.3)
μNi Di E − 2μ ± i

where we have introduced the Kronecker delta δ i j since ψk(+ ) (+)

 |ψk = 0 ∀ Di = D j . For
p2
positive energies, where E = 2 μ and where p is the on-shell c.m. momentum, we get

2δi j k
zi |ĝ(E)|z j = dk . (B.4)
Di Di p2 − k2 ± i

If E ∈
/ Di , we take the limit  → 0 and solve the integral to find

2δi j k
zi |ĝ(E)|z j = dk
Ni Di p2 − k2
  2 ki+1
δi j k 
= − ln  2 − 1
 (B.5)
Ni p ki
 
δi j  E + Ei+1 
= ln ,
Ni  E + Ei 

24
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Figure B1. Top: total time used in calculating on-shell T-matrix elements and extracting
phase shifts and observables, shown for three different Nvidia GPUs of the Tesla line;
T4 (orange), V100 (blue), and K40 (green). Bottom: time decomposition of V100 total
time, with focus on key parts of calculating NN scattering using the WPCD method. The
green line includes the time spent solving the LS equation.

where p ∈/ Di , and Ei and Ei+1 is the lower and upper boundary of Di expressed in energy,
respectively. If E ∈ Di , then we have a simple pole at p = k. The pole-integration is done
using the infinitesimal complex rotation ±i together with the residue theorem, giving
   
δi j  E + E i +1 
zi |ĝ(E)|z j = 
ln   − iπ(θ(E − Ei ) − θ(E − Ei+1 )) , (B.6)
Ni E + Ei 

where θ is the Heaviside step-function. The derivation of the resolvent expressed in a momen-
tum wave-packet representation follows a similar procedure. In that case, it is possible to use
momentum wave-packets where f (p) = 1 and Ni = ki+1 − ki , in which case the derivation
above changes a little, see [18].
Energy averaging of the resolvent is done by integrating the resolvent with respect to E, in
the bin E ∈ Dk , divided by the bin width ΔEk . We introduce the denotation ḡkij to reflect this.
The derivation is straightforward:

1
ḡi j ≡
k
dE zi |ĝ(E)|z j
ΔEk Dk
    
δi j  E + E i +1 
= dE ln    − iπδik (B.7)
ΔEk Ni Dk E + Ei 
δi j iπ
= Wki − δik ,
ΔEk ΔEi ΔEk

where we used Ni = ΔEi , and


k +1 
i +1
 
Wki ≡ (−1)k−k +i−i [Ek − Ei ] ln |Ek − Ei | , (B.8)
k  =k i  =i

as presented in [18].
25
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

Appendix C. GPU code

The code for GPU-utilisation was written to make use of the CUDA interface, which is devel-
oped and maintained by the Nvidia Corporation. CUDA allows for efficient utilisation of a
Nvidia GPU using high-level programing languages such as C and C++, Python, or Fortran.
For numerically demanding linear algebra operations we made use of the CUDA-libraries
cuBLAS [42] and cuSOLVER [43]. These libraries are very similar in use to BLAS and
LAPACK—two standard libraries for linear algebra on the CPU.
Diagonalising the full Hamiltonian and solving the LS equation are the most numerically
demanding parts of the steps presented in section 3.2. These steps are therefore solved on the
GPU. The ability to work on several channels simultaneously—this being the advantage of the
GPU—is made possible by using CUDA’s batched-routines. These are pre-written routines
that maximise the efficiency of the GPU to solve several linear algebra problems simulta-
neously for small matrix-sizes (typically less than 1000 × 1000 in matrix dimension). Such
efficient parallelism is generally difficult to achieve ‘by hand’ due to the massive load on
GPU-memory read-write accesses.
The importance of efficient memory use is made apparent in figure B1 where we show the
computation time spent by three different Nvidia GPUs (top panel) and the decomposition of
time used for the Nvidia V100 GPU (bottom panel). We see that the majority of the total time is
spent on the Hamiltonian diagonalisation. A large fraction of the diagonalisation task is spent
on memory accesses as part of the Jacobi method. The three GPUs: V100, T4, and K40, have
differences in memory technology [55] which is the main reason for the observed differences
in the performance.
We used cublas<t>gemmStridedBatched to calculate the VC-matrix product in
equation (32). This calculates a matrix-product on the form C ← αAB + βC, where C is over-
written by the right-hand side. This setup is standard for BLAS gemm-routines. Here, α and β
are scalars, while A, B, and C are sets of matrices stored congruently in three arrays, i.e. they
are ‘batched’.
To diagonalise the Hamiltonians we used cusolverDn<t>syevjBatched. This rou-
tine utilises the parallel cyclic-order Jacobi method, as was briefly discussed in section 4.2, to
diagonalise batches of matrices simultaneously.
Lastly, solving equation (39) was done using a custom-written function, referred to as a
‘kernel’ in CUDA. The advantage of using the GPU for this task comes with the energy-
dependence in the resolvent. We can calculate all on-shell T-matrix elements simultaneously
following a single batched Hamiltonian diagonalisation.

ORCID iDs

Sean B S Miller https://orcid.org/0000-0003-4782-8326

Andreas Ekström https://orcid.org/0000-0002-4917-4293
Christian Forssén https://orcid.org/0000-0003-3458-0480

References

[1] Bedaque P F and van Kolck U 2002 Effective field theory for few-nucleon systems Annu. Rev. Nucl.
Part. Sci. 52 339–96
[2] Epelbaum E, Hammer H-W and Meißner U-G 2009 Modern theory of nuclear forces Rev. Mod.
Phys. 81 1773–825

26
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

[3] Machleidt R and Entem D R 2011 Chiral effective field theory and nuclear forces Phys. Rep. 503
1–75
[4] Hammer H-W, König S and van Kolck U 2020 Nuclear effective field theory: status and perspectives
Rev. Mod. Phys. 92 025004
[5] Wesolowski S, Klco N, Furnstahl R J, Phillips D R and Thapaliya A 2016 Bayesian parameter
estimation for effective field theories J. Phys. G: Nucl. Part. Phys. 43 074001
[6] Entem D R and Machleidt R 2003 Accurate charge-dependent nucleon–nucleon potential at fourth
order of chiral perturbation theory Phys. Rev. C 68 041001
[7] Carlsson B D, Ekström A, Forssén C, Fahlin Strömberg D, Jansen G R, Lilja O, Lindby M, Mattsson
B A and Wendt K A 2016 Uncertainty analysis and order-by-order optimization of chiral nuclear
interactions Phys. Rev. X 6 011019
[8] Reinert P, Krebs H and Epelbaum E 2018 Semilocal momentum-space regularized chiral two-
nucleon potentials up to fifth order Eur. Phys. J. A 54 86
[9] Piarulli M, Girlanda L, Schiavilla R, Navarro Pérez R, Amaro J E and Ruiz Arriola E 2015
Minimally nonlocal nucleon–nucleon potentials with chiral two-pion exchange including Δ
resonances Phys. Rev. C 91 024003
[10] Navarro Pérez R, Amaro J E and Ruiz Arriola E 2013 Coarse-grained potential analysis of neu-
tron–proton and proton–proton scattering below the pion production threshold Phys. Rev. C 88
064002
[11] Frame D, He R, Ipsen I, Lee D, Lee D and Rrapaj E 2018 Eigenvector continuation with subspace
learning Phys. Rev. Lett. 121 032501
[12] König S, Ekström A, Hebeler K, Lee D and Schwenk A 2020 Eigenvector continuation as an efficient
and accurate emulator for uncertainty quantification Phys. Lett. B 810 135814
[13] Furnstahl R J, Garcia A J, Millican P J and Zhang X 2020 Efficient emulators for scattering using
eigenvector continuation Phys. Lett. B 809 135719
[14] Melendez J A, Drischler C, Garcia A J, Furnstahl R J and Zhang X 2021 Fast & accurate emulation
of two-body scattering observables without wave functions Phys. Lett. B 821 136608
[15] Zhang X and Furnstahl R J 2021 Fast emulation of quantum three-body Scattering
(arXiv:2110.04269 [nucl-th])
[16] Rasmussen C E and Williams C K I 2005 Gaussian Processes for Machine Learning (Adaptive
Computation and Machine Learning) (Cambridge, MA: MIT Press)
[17] Haftel M I and Tabakin F 1970 Nuclear saturation and the smoothness of nucleon–nucleon potentials
Nucl. Phys. A 158 1–42
[18] Rubtsova O A, Kukulin V I and Pomerantsev V N 2015 Wave-packet continuum discretization for
quantum scattering Ann. Phys., NY 360 613–54
[19] Brynjarsdóttir J and O’Hagan A 2014 Learning about physical parameters: the importance of model
discrepancy Inverse Problems 30 114007
[20] Melendez J A, Wesolowski S and Furnstahl R J 2017 Bayesian truncation errors in chiral effective
field theory: nucleon–nucleon observables Phys. Rev. C 96 024003
[21] Hulthén L 1944 Variational problem for the continuous spectrum of a Schrödinger equation Kungl.
Fysiogr. Sällsk. i Lund Förhandl. 14 13
[22] Hulthén L 1948 On the Sturm–Liouville problem connected with a continuous spectrum Arkiv Mat.
Astron. Fysik 35A 25
[23] Kohn W 1948 Variational methods in nuclear collision problems Phys. Rev. 74 1763–72
[24] Schwartz C 1966 Application of the schwinger variational principle for scattering Phys. Rev. 141
1468–70
[25] Basdevant J L 1972 The Padé approximation and its physical applications Fortschr. Phys. 20
283–331
[26] Erkelenz K, Alzetta R and Holinde K 1971 Momentum space calculations and helicity formalism
in nuclear physics Nucl. Phys. A 176 413–32
[27] du Croz J J and Higham N J 1992 Stability of methods for matrix inversion IMA J. Numer. Anal. 12
1–19
[28] Glöckle W 2012 The Quantum Mechanical Few-Body Problem (Berlin: Springer)
[29] Carbonell J, Deltuva A, Fonseca A C and Lazauskas R 2014 Bound state techniques to solve the
multiparticle scattering problem Prog. Part. Nucl. Phys. 74 55–80
[30] Heller E J 1975 Theory of J-matrix Green’s functions with applications to atomic polarizability and
phase-shift error bounds Phys. Rev. A 12 1222–31

27
J. Phys. G: Nucl. Part. Phys. 49 (2022) 024001 S B S Miller et al

[31] Winick J R and Reinhardt W P 1978 Moment T-matrix approach to e+ -h scattering: I. Angular
distribution and total cross section for energies below the pickup threshold Phys. Rev. A 18
910–24
[32] Winick J R and Reinhardt W P 1978 Moment T-matrix approach to e+ -h scattering: II. Elastic
scattering and total cross section at intermediate energies Phys. Rev. A 18 925–34
[33] Heller E J, Rescigno T N and Reinhardt W P 1973 Extraction of scattering information from Fred-
holm determinants calculated in an L2 basis: a chebyschev discretization of the continuum Phys.
Rev. A 8 2946–51
[34] Corcoran C T and Langhoff P W 1977 Moment-theory approximations for nonnegative spectral
densities J. Math. Phys. 18 651–7
[35] Rubtsova O A and Kukulin V I 2007 Wave-packet discretization of a continuum: path toward
practically solving few-body scattering problems Phys. Atom. Nuclei 70 2025–45
[36] Kukulin V I, Pomerantsev V N and Rubtsova O A 2007 Wave-packet continuum discretization
method for solving the three-body scattering problem Theor. Math. Phys. 150 403–24
[37] Rubtsova O A, Pomerantsev V N and Kukulin V I 2009 Quantum scattering theory on the
momentum lattice Phys. Rev. C 79 064602
[38] Kukulin V I and Rubtsova O A 2003 Discrete quantum scattering theory Theor. Math. Phys. 134
404–26
[39] Müther H, Rubtsova O A, Kukulin V I and Pomerantsev V N 2016 Discrete wave-packet represen-
tation in nuclear matter calculations Phys. Rev. C 94 024328
[40] Kukulin V I and Rubtsova O A 2005 Solving the charged-particle scattering problem by wave packet
continuum discretization Theor. Math. Phys. 145 1711–26
[41] Rubtsova O A, Kukulin V I, Pomerantsev V N and Faessler A 2010 New approach toward a direct
evaluation of the multichannel multienergy s matrix without solving the scattering equations
Phys. Rev. C 81 064003
[42] NVIDIA Corporation 2021 Basic linear algebra on NVIDIA GPUs https://developer.nvidia.com/
cublas (accessed 22 March 2021)
[43] NVIDIA Corporation 2021 Cuda toolkit documentation https://docs.nvidia.com/cuda/cusolver/
index.html (accessed 22 March 2021)
[44] Golub G H and Van Loan C F 1996 Matrix Computations 3rd edn (Baltimore, MD, USA: Johns
Hopkins University Press)
[45] Pourzandi M and Tourancheau B 1994 A parallel performance study of Jacobi-like eigenvalue
solution Technical Report
[46] Abdelfattah A, Tomov S and Dongarra J 2020 Matrix multiplication on batches of small matrices
in half and half-complex precisions J. Parallel Distrib. Comput. 145 188–201
[47] Bae S E, Shinn T-W and Takaoka T 2014 A faster parallel algorithm for matrix multiplication
on a mesh array Procedia Computer Science 29 2230–40 2014 International Conference on
Computational Science
[48] Ekström A et al 2013 Optimized chiral nucleon–nucleon interaction at next-to-next-to-leading order
Phys. Rev. Lett. 110 192502
[49] Levinson N 1949 On the uniqueness of the potential in a schrodinger equation for a given asymptotic
phase Kgl. Danske Videnskab Selskab. Mat. Fys. Medd. 25 9
[50] Lisowski P W, Shamu R E, Auchampaugh G F, King N S P, Moore M S, Morgan G L and Singleton
T S 1982 Search for resonance structure in thenpTotal cross section below 800 MeV Phys. Rev.
Lett. 49 255–9
[51] Abfalterer W P, Bateman F B, Dietrich F S, Finlay R W, Haight R C and Morgan G L 2001
Measurement of neutron total cross sections up to 560 MeV Phys. Rev. C 63 044608
[52] Pomerantsev V N, Kukulin V I and Rubtsova O A 2014 New general approach in few-body scatter-
ing calculations: solving discretized Faddeev equations on a graphics processing unit Phys. Rev.
C 89 064008
[53] Stapp H P, Ypsilantis T J and Metropolis N 1957 Phase-shift analysis of 310 MeV proton–proton
scattering experiments Phys. Rev. 105 302–10
[54] Bystricky J, Lehar F and Winternitz P 1978 Formalism of nucleon–nucleon elastic scattering
experiments J. Phys. France 39 1–32
[55] Jia Z, Maggioni M, Smith J and Scarpazza D P 2019 Dissecting the NVidia turing T4 GPU via
microbenchmarking (arXiv:1903.07486v1 [cs.DC])

28