92 4 General discussion

4. General discussion

4.1. Overview of the method

For completeness, next we present a short overview of the method, skipping minor details and
omitting the demonstration of some properties. The reader is referred to papers (P1)-(P3) for a more
complete description. In this short overview, we also introduce some new notation with the purpose
of making the explanation more simple and clear.

4.1.1. Weak formulation

For shortness, we write the Euler equations for compressible flows (1)-(3) as

𝜕𝑡 𝑢𝐼 + 𝜕𝑗 (𝑣𝑗 𝑢𝐼 ) = 𝜕𝑗 (𝑓𝐼 𝑗 ) , (5)

with 𝐮 = [𝜌, 𝑚1 , 𝑚2 , ]𝑇 the vector of conservative variables, 𝐟 = 𝐟(𝐮) the pressure flux matrix, given
by 𝑓1,𝑗 = 0, 𝑓1+𝑖,𝑗 = −𝑝𝛿𝑖𝑗 , 𝑓4,𝑗 = −𝑝𝑣𝑗 , and 𝛿𝑖𝑗 the Kronecker delta. Here uppercase and lowercase indices
vary from 1 to 4 and from 1 to 2, respectively.

In (P1), we show that conservation in the numerical scheme can be achieved by means of the
so-called weak Lagrange–Galerkin formulation [49, 59, 60, 74, 75]. In the case of the Euler equations,
this formulation reads
d
𝑢𝐼 𝜓 dΩ = − ∫ 𝑓𝐼 𝑗 𝜕𝑗 𝜓 dΩ + ∮ 𝑓𝐼 𝑗 𝜓 𝑛𝑗 d𝜎, ∀𝜓 ∈ , (6)
d𝑡 ∫Ω𝑓 Ω𝑓 𝜕Ω𝑓

where Ω𝑓 (𝑡) is an arbitrary domain that{moves with the fluid, 𝐧(𝐱, 𝑡) is the outward } normal vec-
0
tor to the boundary 𝜕Ω𝑓 (𝑡), and  ∶= 𝜓 (𝐱, 𝑡) ∈  (Ω𝑓 (𝑡), ℝ) ∶ 𝜕𝑡 𝜓 + 𝑣𝑗 𝜕𝑗 𝜓 = 0 is the space of
continuous functions that remain constant along the trajectories of the fluid particles. As demon-
strated in [59, 60], it is possible to arrive at (6) by multiplying Eq. (5) by 𝜓 , integrating by parts
and using Reynolds theorem. In (P1), however, we offer an alternative demonstration based on the
approach followed in books of Continuum Mechanics and Applied Mathematics for the derivation
of conservation laws.

Remark 1. Eq. (6) can be seen a set of conservation laws for a weighted mass, momentum and total
energy, from which the standard conservation laws can be recovered by making 𝜓 ≡ 1.

4.1.2. Numerical discretization

Hereafter, we assume for simplicity that the flux 𝑓𝐼 𝑗 is weakly imposed at the boundary (see (P3)
for details on the implementation of other boundary conditions). For the numerical discretization,
we consider a moving polygonal domain Ω ̃ ℎ (𝑡) which approximates Ω𝑓 (𝑡). This domain is partitioned
̃
onto a mesh 𝕋ℎ (𝑡) of triangular (curvilinear) isoparametric elements of degree 𝑟 ≥ 1. The nodes, 𝝌 𝑖 (𝑡),
𝑖 = 1, … , 𝑁ℎ , are distributed according to a Chebyshev grid of the second kind [16] to allow for stable
high-order discretizations, and move according to the equation

d𝝌 𝑖 (𝑡) 𝐦ℎ (𝝌 𝑖 (𝑡), 𝑡)
= . (7)
d𝑡 𝜌ℎ (𝝌 𝑖 (𝑡), 𝑡)
 4.1 Overview of the method 93

{ }
Let 𝐾̂ ∶= 𝐱̂ ∈ ℝ2 ∶ 0 ≤ 𝑥̂1 ≤ 1, 0 ≤ 𝑥̂2 ≤ 1 − 𝑥̂1 be the well-known reference element, in which
there are (𝑟 + 1)(𝑟 + 2)/2 Chebyshev nodes of the second kind [16]. With a view to employ multiscale
modelling, we define two spaces on 𝐾̂ . The first one is the so-called large-scale space, and is the space
of polynomials of maximal degree 𝑟 in 𝐾̂ , which we denote by 𝑃̂𝑟 (𝐾̂ ) and has dimension (𝑟 +1)(𝑟 +2)/2.
The Lagrangian basis associated with this space can be computed as shown in [61, Section 3.1.2].
The second one is the fine-scale space, which is defined as
{ }
̂ 𝑝 ∈ 𝑃̂𝑟−1 (𝐾̂ ), Π𝑟 𝜓 = 0 ,
𝐵̂𝑟 (𝐾̂ ) ∶= 𝜓 ∶ 𝜓 = 𝑏𝑝, (8)

where 𝑏(̂̂ 𝐱) = 27̂ 𝑥1 𝑥̂2 (1 − 𝑥̂1 − 𝑥̂2 ) the elementary bubble function — a cubic polynomial that vanishes
at the boundary 𝜕 𝐾̂ — and Π𝑟 is the Lagrangian interpolation operator onto 𝑃̂𝑟 (𝐾̂ ). That is, 𝐵̂𝑟 (𝐾̂ )
is the space of polynomials of maximal degree 𝑟 + 2 that vanish at the boundary of 𝐾̂ — hence the
functions in this space are also called bubbles — and that their interpolation onto the large-scale
space is null — this is done to exclude polynomials that are already in 𝑃̂𝑟 (𝐾̂ ), so the basis for 𝑃̂𝑟 (𝐾̂ )
and 𝐵̂𝑟 (𝐾̂ ) are linearly independent. The basis for 𝐵̂𝑟 (𝐾̂ ) can be constructed as shown in Appendix
A.1. For convenience, we also define the space 𝑃̂𝑟bubble (𝐾̂ ) ∶= 𝑃̂𝑟 (𝐾̂ ) ⊕ 𝐵̂𝑟 (𝐾̂ ).

Now let 𝐅̃ 𝐾 (𝑡) be the standard 𝑟th-degree polynomial application which maps from 𝐾̂ to a given
element 𝐾 (𝑡) ∈ 𝕋 ̃ℎ (𝑡). For a given 𝑡, we define the space
{ }
̃ℎ (𝑡)) ∶= 𝜓 (𝐱) ∈  0 (Ω
𝑉̃ℎ (𝑡) = 𝑃𝑟bubble (𝕋 ̃ ℎ (𝑡)) ∶ 𝜓 || (𝐅̃ 𝐾 (𝑡) (̂𝐱)) = 𝜓̂ (̂𝐱) ∈ 𝑃̂𝑟bubble (𝐾̂ ), ∀𝐾 (𝑡) ∈ 𝕋
̃ ℎ (𝑡) .
|𝐾 (𝑡)
The finite element solution 𝑢𝐼 ℎ (𝐱, 𝑡) is defined in this space. To define the test functions in (6), which
remain constant along the trajectories, we need the more restrictive space
{ }
̃ ℎ (𝑡), ℝ) , 𝑣 || (𝐅̃ 𝐾 (𝑡) (̂𝐱) , 𝑡 ) = 𝑣̂ (̂𝐱) ∈ 𝑃̂𝑟bubble (𝐾̂ ), ∀𝐾 (𝑡) ∈ 𝕋
ℎ ∶= 𝑣(𝐱, 𝑡) ∈  0 (Ω ̃ ℎ (𝑡) .
|𝐾 (𝑡)
Note that, for a given 𝑡, any function 𝑣(𝐱, 𝑡) ∈ ℎ belongs to 𝑉̃ℎ (𝑡); however, a function 𝑣(𝐱, 𝑡) such
that 𝑣(𝐱, 𝑡) ∈ 𝑉̃ℎ (𝑡) for all 𝑡 does not necessarily belong to ℎ . Also note that, for any 𝑡, 𝑣(𝐱, 𝑡) ∈ ℎ
reads the same in the reference element. Equivalently, if 𝑣(𝐱, 𝑡 𝑚 ) is the 𝑖th shape function of 𝑉̃ℎ (𝑡 𝑚 ),
𝑣(𝐱, 𝑡 𝑛 ) is the 𝑖th shape function of 𝑉̃ℎ (𝑡 𝑛 ), for two arbitrary time levels 𝑡 𝑚 and 𝑡 𝑛 . This is the most
important property for the time discretization of Eq. (6).

With the definitions above, Eq. (6) is discretized in space as

d
𝑢𝐼 ℎ 𝜓ℎ dΩ = − ∫ 𝑓𝐼 𝑗 ℎ 𝜕𝑗 𝜓ℎ dΩ + ∮ 𝑓𝐼 𝑗 ℎ 𝜓ℎ 𝑛𝑗 d𝜎 − 𝐼 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) , ∀𝜓ℎ ∈ ℎ , (9)
d𝑡 ∫Ω̃ℎ ̃ℎ
Ω ̃ℎ
𝜕Ω

where 𝑢𝐼 ∗ (𝐱, 𝑡) ∈ 𝑉̃ℎ (𝑡) is a non-stabilized approximation to 𝑢𝐼 (𝐱, 𝑡) and 𝐼 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) is a stabilization
operator. Both 𝑢𝐼 ∗ and 𝐼 are to be introduced later. Only mass, momentum and total energy are
discretized; the rest of variables are computed from these using appropriate relations whenever
required.

Remark 2. As we show in (P1)-(P3), if the total force and the total power on the boundary are null, the
test function 𝜓ℎ (𝐱, 𝑡) ≡ 1 belongs to ℎ , and the stabilization operator is conservative — i.e. 𝐼 (𝐮∗ ; 𝐮ℎ , 1) =
0 —, then the discretized conservation law (9) yields
d
𝑢𝐼 ℎ dΩ = 0.
d𝑡 ∫Ω̃ℎ
Therefore, any time-marching scheme that preserves constant solutions, such as multistep and Runge–
Kutta methods, will also preserve the numerical mass, momentum and total energy in the moving do-
̃ ℎ (𝑡), irrespective of the space and time discretization orders.
main Ω
94 4 General discussion

Now let Ωℎ be a (fixed) reference domain associated with a (fixed) finite element mesh 𝕋ℎ and
a finite element space 𝑉ℎ = 𝑃𝑟bubble (𝕋ℎ ). Let us assume that we aim to compute the solution at a
time level 𝑡 𝑛+1 from the solutions computed at some previous time levels. Normally, in Lagrange–
Galerkin methods, we choose a moving mesh such that 𝕋 ̃ℎ (𝑡 𝑛+1 ) ≡ 𝕋ℎ . Then, we displace the nodes
̃ ℎ backward-in-time to those previous time levels, and discretize (9) with a multistep formula.
of 𝕋
To do so, we need first to extrapolate in time the velocity field from previous solutions; see, e.g.,
[12, 17, 59]. However, as pointed by other authors [37] and also according to our experience, this
procedure is not stable for high-order time-discretizations.

In (P2), we show that stable high-order discretizations can be achieved in a simple way by in-
tegrating (9) forward-in-time in [𝑡 𝑛 , 𝑡 𝑛+1 ] with an implicit–explicit Runge–Kutta (RK) method, fol-
lowing the philosophy of ALE methods. More specifically, we first choose a moving mesh such that
̃ℎ (𝑡 𝑛 ) ≡ 𝕋ℎ (also, 𝑉̃ℎ (𝑡 𝑛 ) ≡ 𝑉ℎ ). The time level 𝑡 𝑛 corresponds to the first stage of the RK method.
𝕋
Then, for the remaining stages 𝑙 = 2, … , 𝑠, we perform the following steps:
(S1) The nodes of 𝕋 ̃ ℎ (𝑡) are displaced by integrating (7) with the explicit part of the method.
(S2) A non-stabilized solution at the current stage, 𝐮[𝑙] ∗ , is computed by integrating (9) with no
stabilization operator and with the explicit part of the method.
(S3) The non-stabilized solution is post-processed to compute the stabilization operator 𝐼 (𝐮[𝑙] ∗ ; 𝐮ℎ ,
𝜓ℎ ).
(S4) Eq. (9) is solved again, but now considering the stabilization operator and the implicit part of
the RK scheme. This yields a stabilized solution 𝐮[𝑙] ℎ .
𝑛+1 𝑛+1
The solution at 𝑡 , 𝐮̃ ℎ , is readily obtained from the solutions at the internal stages, and is defined
on the displaced mesh 𝕋 ̃𝑛+1 2
ℎ . This solution is remapped onto the reference mesh 𝕋ℎ via 𝐿 -projections;
that is, we find 𝑢𝐼 𝑛+1
ℎ ∈ 𝑉ℎ such that

𝑛+1 𝑛+1
∫ 𝑢𝐼 ℎ 𝜓ℎ dΩ = ∫ 𝑢̃𝐼 ℎ 𝜓ℎ dΩ, ∀𝜓ℎ ∈ 𝑉ℎ . (10)
Ωℎ Ωℎ

See also Fig. 2.

Trajectories of the fluid particles

̃ 𝑛+1
Control volume Ω ℎ ,
Control volume Ωℎ , ̃
mesh 𝕋ℎ𝑛+1

mesh 𝕋ℎ

Figure 2: Scheme that illustrates the domains and meshes that appear in the formulation.
The nodes in 𝕋ℎ are convected forward in time with the flow velocity field to form 𝕋 ̃ 𝑛+1 .
ℎ
𝑛+1 𝑛+1
The variables 𝑢𝐼 ℎ and 𝜓ℎ in Eq. (10) are defined piecewise in 𝕋ℎ , whereas 𝑢̃ℎ is defined
̃𝑛+1 .
piecewise in 𝕋 ℎ

To integrate (9) with a RK method, it is essential to recall that, for two arbitrary stages [𝑙] and
[𝑘]:
 4.1 Overview of the method 95

(i) the spaces 𝑉̃ℎ (𝑡 [𝑙] ) and 𝑉̃ℎ (𝑡 [𝑘] ) are standard polynomial + bubble spaces associated with the
meshes 𝕋 ̃ ℎ (𝑡 [𝑙] ) and 𝕋
̃ ℎ (𝑡 [𝑘] ), respectively,
(ii) if 𝜓ℎ (𝐱, 𝑡 [𝑙] ) is the 𝑖th shape function of 𝑉̃ℎ (𝑡 [𝑙] ), then 𝜓ℎ (𝐱, 𝑡 [𝑘] ) is the 𝑖th shape function of
𝑉̃ℎ (𝑡 [𝑘] ) and vice versa.

Remark 3. The mesh 𝕋 ̃ ℎ (𝑡) is moved with the explicit part of the Runge–Kutta method, whereas the
fluid variables are solved with the implicit part. Hence, 𝕋 ̃ℎ (𝑡) and the corresponding domain Ω
̃ ℎ (𝑡) are
fixed when solving (9).

Remark 4. Eq. (10) theoretically preserves mass in the domain Ωℎ . However, the right-hand side is
the scalar product of functions defined piecewise in different meshes. We compute this term with high-
order quadrature rules to maintain stability [89]. Despite that, this term is not exactly computed, and
therefore there are some small mass, momentum and energy losses. To avoid confusions, we shall say
that our method is nearly-conservative, i.e., conservative with high-accuracy. Alternatively, one could
use mesh intersection techniques [5, 52, 113] or other remapping algorithms [6, 121].

Remark 5. Despite the small losses of mass, momentum and energy due to the mesh remap, the numer-
ical experiments show that the method is able to adequately capture and track the discontinuities. Note
that some other Lagrangian methods are not fully conservative either [36, 42], and that fixed-mesh
(theoretically conservative) methods may also suffer from this issue when combined with anisotropic
mesh refinement. Hence, we think that our method could be applied to the same spectrum of problems
than these other methods. We do not know how important these losses would be in the case of very long
simulations, such as in climate modelling.

Remark 6. The present method is very similar in nature to ALE methods of Lagrangian+remap type.
The difference is that, here, we remap onto a fixed reference mesh via 𝐿2 -projections, instead of remap-
ping onto a smoothed Lagrangian mesh using some other techniques [6, 121]. Hence, the method does
not naturally refine the mesh at the discontinuities, although it does not either make the mesh more
coarse at the expansion regions. In any case, the present method could work as well by considering a
smoothed Lagrangian mesh instead of a fixed reference mesh. For this reason, we shall say that the
present forward Lagrange–Galerkin method is another class of ALE methods.

See also Appendix A.2 for a brief discussion on the local conservation property of the present
method.

4.1.3. Stabilization of the scheme

The Lagrangian formulation stabilizes the scheme against the convective terms of the equations.
However, there are some other facts that may cause instabilities:
(i) the inexact computation of the right-hand side in (10) [10, 13],
(ii) the presence of acoustic waves, which propagate through the domain at the sound velocity
with respect to the moving mesh 𝕋 ̃ ℎ (𝑡) [107] — or equivalently, the presence of nonsymmetric
terms in the Galerkin formulation (9) —,
(iii) the presence of discontinuities in the solution.
In this work, stability is achieved with two ingredients: a multiscale model and an appropriate
viscous regularization of the Euler equations.

In the multiscale model, we assume that the finite element space 𝑉̃ℎ (𝑡) associated with the moving
96 4 General discussion

mesh 𝕋̃ ℎ (𝑡) is the direct sum of a large-scale space,

{ }
𝑉̃ℎ (𝑡) = 𝑃𝑟 (𝕋 ̃ ℎ (𝑡)) ∶ 𝜓 || (𝐅̃ 𝐾 (𝑡) (̂𝐱)) = 𝜓̂ (̂𝐱) ∈ 𝑃̂𝑟 (𝐾̂ ), ∀𝐾 (𝑡) ∈ 𝕋
̃ ℎ (𝑡)) ∶= 𝜓 (𝐱) ∈  0 (Ω ̃ℎ (𝑡) ,
|𝐾 (𝑡)

and a fine-scale space,

{ }
𝑉̃ℎ′ (𝑡) = 𝐵𝑟 (𝕋 ̃ ℎ (𝑡)) ∶ 𝜓 || (𝐅̃ 𝐾 (𝑡) (̂𝐱)) = 𝜓̂ (̂𝐱) ∈ 𝐵̂𝑟 (𝐾̂ ), ∀𝐾 (𝑡) ∈ 𝕋
̃ℎ (𝑡)) ∶= 𝜓 (𝐱) ∈  0 (Ω ̃ℎ (𝑡) ,
|𝐾 (𝑡)

i.e., 𝑉̃ℎ (𝑡) ∶= 𝑉̃ℎ (𝑡)⊕ 𝑉̃ℎ′ (𝑡). Hence, any function 𝜓ℎ (𝐱, 𝑡) ∈ 𝑉̃ℎ (𝑡) is expressed as the sum of a large-scale
term, 𝜓 ℎ (𝐱, 𝑡), and a fine-scale term, 𝜓ℎ′ (𝐱, 𝑡), that is,

𝜓ℎ = 𝜓 ℎ + 𝜓ℎ′ , 𝜓 ℎ ∈ 𝑉̃ℎ , 𝜓ℎ′ ∈ 𝑉̃ℎ′ , 𝜓ℎ ∈ 𝑉̃ℎ = 𝑉̃ℎ ⊕ 𝑉̃ℎ′ . (11)

Recall that 𝑉̃ℎ (𝑡) is the standard continuous polynomial space of degree 𝑟, whereas 𝑉̃ℎ′ (𝑡) is the cor-
responding bubble space.

Remark 7. The most important property of the multiscale decomposition is that the large-scale term
captures the behavior of the solution, whereas the fine-scale term captures the error in the large-scale
term.

For the viscous regularization of the Euler equations, we employ Brenner’s model for viscous
and compressible flows [20]. In contrast to the classical Navier–Stokes model, Brenner’s model takes
into account for mass diffusion and for the transport of momentum and energy associated with the
latter. This is achieved by replacing the velocity 𝑣𝑖 in the convective terms of the Navier–Stokes
equations by a mass velocity
𝜕𝑗 𝜌
(𝑣𝑚 )𝑗 ∶= 𝑣𝑗 − 𝜖 ,
𝜌
with 𝜖 the mass-diffusion coefficient. A great advantage of this model is that it makes easier to
remove instabilities in the density, as it introduces diffusion in the continuity equation. Also, it
satisfies entropy inequalities and ensures positivity of density and energy [65]. The existence of
weak solutions to Brenner’s model has been proved in [53] and numerical experiments [90] confirm
that this model yields better numerical results.

The most robust regularization corresponds to the case 𝜅/𝑐𝑣 = 𝜌𝜖, with 𝜅 the conductivity co-
efficient and 𝑐𝑣 the specific heat for constant volume [65]. In that case, Brenner’s model reads (see
(P3) for more details)

𝜕𝑡 𝜌 + 𝜕𝑗 (𝜌𝑣𝑗 ) = 𝜕𝑗 (𝜖𝜕𝑗 𝜌 ) , (12)

𝜕𝑡 𝑚𝑖 + 𝜕𝑗 (𝑚𝑖 𝑣𝑗 ) = −𝜕𝑖 𝑝 + 𝜕𝑗 𝜎𝑖𝑗∗ , (13)
𝜕𝑡  + 𝜕𝑗 (𝑣𝑗 ) = −𝜕𝑗 (𝑝𝑣𝑗 ) + 𝜕𝑗 𝑞𝑗∗ , (14)

with
2
𝜎𝑖𝑗∗ = 𝜎𝑖𝑗∗ (𝜖, 𝜈; 𝐯; ∇𝜌, ∇𝐦) ∶= 𝜈 [𝜕𝑖 𝑚𝑗 − 𝑣𝑗 𝜕𝑖 𝜌 + 𝜕𝑗 𝑚𝑖 − 𝑣𝑖 𝜕𝑗 𝜌 ] − 𝜈 (𝜕𝑘 𝑚𝑘 − 𝑣𝑘 𝜕𝑘 𝜌) 𝛿𝑖𝑗 + 𝜖𝑣𝑖 𝜕𝑗 𝜌,
3
∗ ∗ ∗
𝑞𝑗 = 𝑞𝑗 (𝜖, 𝜈; 𝐯; ∇𝜌, ∇𝐦, ∇) ∶= 𝜖 [𝜕𝑗  − 𝑣𝑖 𝜕𝑗 𝑚𝑖 ] + 𝑣𝑖 𝜎𝑖𝑗 (𝜖, 𝜈; 𝐯; ∇𝜌, ∇𝐦) ,

and 𝜈 the kinematic viscosity.

 4.1 Overview of the method 97

The discontinuity capturing (DC) operator aims to stabilize shocks, shear layers and contact dis-
continuities. It was developed in (P2) for scalar hyperbolic conservation laws and extended in (P3)
for compressible flows. The DC operator is based on post-processing the non-stabilized solution 𝐮∗ .
Since 𝐮∗ is computed without any stabilization, we can expect spurious oscillations at the disconti-
nuities. On the other hand, according to Remark 7 and to Eq. (11), the fine-scale terms 𝐮′∗ provide
an estimation of the error in the large-scale terms 𝐮∗ . Thus, we can expect that the fine-scale terms
be large at the discontinuities and small at the smooth regions. The DC operator adds sufficient
artificial diffusion where 𝐮′∗ is large, and a little amount where 𝐮′∗ is small. This allows to capture the
discontinuities without spoiling the numerical solution at the smooth regions. Briefly speaking, we
𝑇
recall Brenner’s model (12)-(14) and define the discontinuity operator 𝐼DC = [𝜌DC , 𝑚DC1 , 𝑚DC2 , DC ]
as

𝜌DC (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝜖ℎDC 𝜕𝑗 𝜌ℎ 𝜕𝑗 𝜓ℎ dΩ, (15)

̃ℎ
Ω

𝑚DC𝑖 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝜎𝑖𝑗∗ (𝜖ℎDC , 𝜈ℎDC ; 𝐯ℎ ; ∇𝜌ℎ , ∇𝐦ℎ ) 𝜕𝑗 𝜓ℎ dΩ, (16)

̃ℎ
Ω

DC (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝑞𝑗∗ (𝜖ℎDC , 𝜈ℎDC ; 𝐯ℎ ; ∇𝜌ℎ , ∇𝐦ℎ , ∇ℎ ) 𝜕𝑗 𝜓ℎ dΩ, (17)
̃ℎ
Ω

where 𝜖ℎDC and 𝜈ℎDC are piecewise linear viscosities that depend on 𝐮′∗ . As said before, these viscosities
are high — of the order of the mesh size — at the discontinuities and small at the smooth regions.

Apart from discontinuities, instabilities may arise also due to the presence of nonsymmetric
terms in the Galerkin formulation (9) and to the inexact computation of the integrals in (10). These
(initially small) instabilities propagate through acoustic waves, as demonstrated by Scovazzi [107].
Unfortunately, the DC operator above is not able to detect and remove these acoustic instabilities,
and thus it is necessary to provide another stabilization term.

For that purpose, we extend in (P2)-(P3) the subgrid stabilization (SS) technique developed by
Guermond et al. for incompressible flows [62, 63] to the present case. In particular, we define the SS
𝑇
operator 𝐼SS = [𝜌SS , 𝑚SS1 , 𝑚SS2 , SS ] as

𝜌SS (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝜖ℎSS 𝜕𝑗 𝜃̃ℎ 𝜌ℎ 𝜕𝑗 𝜃̃ℎ 𝜓ℎ dΩ, (18)

̃ℎ
Ω

𝑚SS𝑖 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝜎𝑖𝑗∗ (𝜖ℎSS , 𝜈ℎSS ; 𝐯ℎ ; ∇𝜃̃ℎ 𝜌ℎ , ∇𝜃̃ℎ 𝐦ℎ ) 𝜕𝑗 𝜃̃ℎ 𝜓ℎ dΩ, (19)
̃ℎ
Ω

SS (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ∶= ∫ 𝑞𝑗∗ (𝜖ℎSS , 𝜈ℎSS ; 𝐯ℎ ; ∇𝜃̃ℎ 𝜌ℎ , ∇𝜃̃ℎ 𝐦ℎ , ∇𝜃̃ℎ ℎ ) 𝜕𝑗 𝜃̃ℎ 𝜓ℎ dΩ. (20)
̃ℎ
Ω

Here, 𝜖ℎSS and 𝜈ℎSS are piecewise linear viscosities that depend on 𝐮∗ and are always of the order of the
mesh size. In addition, 𝜃̃ℎ ∶= 𝕀 − Π ̃ℎ is the so-called fluctuation operator, with 𝕀 the identity and Π
̃ℎ
3 ̃ ̃ ̃ ′ ′
the Lagrange interpolation operator onto 𝕋ℎ . Note from (8) and (11) that 𝜃ℎ 𝜓ℎ = 𝜃𝜓ℎ = 𝜓ℎ ∀𝜓ℎ ∈ 𝑉ℎ . ̃
Hence, the SS operator is applying a high viscosity in the whole domain, although it acts only in the
fine-scale terms.

The stabilization operator in (9) is hence

𝐼 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) = 𝐼DC (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) + 𝐼SS (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) .

3 Here(Π̃ℎ 𝑢)|𝐾 (𝑡) is not the result of interpolating 𝑢 via the corresponding Lagrange polynomials in the physical
element 𝐾 (𝑡), but in the reference element 𝐾̂ .
98 4 General discussion

Remark 8. In theory, the stabilization operator depends on five parameters — not introduced above
for brevity — that have to be fine-tuned to adjust the amount of diffusion. However, in (P3) we set
fixed values for these parameters and obtain satisfactory results for different test problems, mesh sizes,
discretization orders and time step sizes. Hence, in practice it is not necessary to fine-tune any parameter.

4.1.4. Resolution of Eq. (9)

Now let us look back at (S4), where we have to solve the stage [𝑙] of the RK method. In particular,
we have to find 𝑢𝐼 [𝑙] ̃[𝑙]
ℎ ∈ 𝑉ℎ such that

[𝑙] [𝑙] [𝑙] [𝑙]

∫̃[𝑙] 𝑢𝐼 ℎ 𝜓ℎ dΩ = −𝑔𝐼 (𝐮ℎ , 𝜓ℎ ) + 𝑏𝐼 (𝜓ℎ ), ∀𝜓ℎ ∈ ℎ , (21)
Ωℎ

where
𝑔𝐼[𝑙] (𝐮ℎ , 𝜓ℎ ) ∶= 𝑎𝑙𝑙 Δ𝑡 ∫ 𝑓𝐼 𝑗 [𝑙] [𝑙] [𝑙] [𝑙] [𝑙]
ℎ 𝜕𝑗 𝜓ℎ dΩ + 𝑎𝑙𝑙 Δ𝑡 𝐼 (𝐮∗ ; 𝐮ℎ , 𝜓ℎ ) ,
̃ [𝑙]
Ω ℎ
𝑙−1 𝑙
𝑏𝐼[𝑙] (𝜓ℎ ) ∶= ∫ 𝑢𝐼 [1]
ℎ 𝜓ℎ
[1]
dΩ − Δ𝑡 ∑ 𝑎𝑙𝑘 ∫ 𝑓𝐼 𝑗 ℎ[𝑘] 𝜕𝑗 𝜓ℎ[𝑘] dΩ + Δ𝑡 ∑ 𝑎𝑙𝑘 ∮ 𝑓𝐼 𝑗 ℎ[𝑘] 𝜓ℎ[𝑘] 𝑛𝑗 d𝜎−
̃ [1]
Ω 𝑘=1 ̃ [𝑘]
Ω 𝑘=1 ̃ [𝑘]
𝜕Ω
ℎ ℎ ℎ
𝑙−1
[𝑘] [𝑘]
Δ𝑡 ∑ 𝑎𝑙𝑘 𝐼 (𝐮[𝑘]
∗ , 𝐮ℎ , 𝜓ℎ ) ,
𝑘=1

(⋅)[𝑘] denotes any variable evaluated at 𝑡 [𝑘] , Δ𝑡 is the time step, and 𝑎𝑙𝑘 are the coefficients of the
well-known Butcher tableau of the implicit part of the RK method. Recall that the non-stabilized
solution 𝐮[𝑙]
∗ has already been computed; therefore, all the viscosity coefficients that appear in the
[𝑙] [𝑙] [𝑙]
definition of the stabilization operator 𝐼 (𝐮[𝑙]
∗ ; 𝐮ℎ , 𝜓ℎ ) are fixed and independent of 𝐮ℎ .

For the case 𝐼 = 1, Eq. (21) corresponds to the continuity equation. Recalling that 𝑓1,1 = 𝑓1,2 = 0
and the structure of the operators (15) and (18), we see that (21) reads as a linear equation for the
density. Thus, 𝜌ℎ[𝑙] can be readily computed in one step.

On the contrary, the cases 𝐼 = 2, 3, 4 form a nonlinear system which is coupled due to the right-
hand side term (the left-hand side term only involves 𝑢𝐼 for a given 𝐼 ). To solve this, one could
proceed with a fixed-point iteration: given 𝐮ℎ[𝑙],𝑚 , we evaluate the right-hand side of (21) and obtain
𝐮[𝑙],𝑚+1
ℎ from the left-hand side. However, the right-side is stiff due to the stability terms, and thus
this fixed-point iteration will suffer from convergence problems.

To alleviate this drawback, in (P3) we proceed as follows. First, we accelerate the fixed-point
iteration by adding to both sides of (21) a linear term 𝐺𝐼[𝑙] that approximately cancels the stiffness
of term 𝑔𝐼[𝑙] . In particular, this term represents the contributions of ∇𝑚1ℎ , ∇𝑚2ℎ and ∇ℎ to the
momentum and energy equations, viz.,
[𝑙]
𝐺1+𝑖 (𝐮[𝑙] [𝑙]
ℎ , 𝜓ℎ ) ∶= ∫ 𝜎𝑖𝑗∗∗ (𝜈ℎDC ; ∇𝐦ℎ ) 𝜕𝑗 𝜓ℎ dΩ + ∫ 𝜎𝑖𝑗∗∗ (𝜈ℎSS ; ∇𝜃̃ℎ 𝐦ℎ ) 𝜕𝑗 𝜃̃ℎ 𝜓ℎ dΩ, (22)
̃ [𝑙]
Ω ̃ [𝑙]
Ω
ℎ ℎ

𝐺4[𝑙] (𝐮[𝑙] [𝑙]

ℎ , 𝜓ℎ ) ∶= ∫ 𝑞𝑗∗∗ (𝜖ℎDC ; ∇ℎ ) 𝜕𝑗 𝜓ℎ dΩ + ∫ 𝑞𝑗∗∗ (𝜖ℎSS ; ∇𝜃̃ℎ ℎ ) 𝜕𝑗 𝜃̃ℎ 𝜓ℎ dΩ, (23)
̃ [𝑙]
Ω ̃ [𝑙]
Ω
ℎ ℎ

with
⎡ 4 𝜕1 𝑚1 𝜕2 𝑚1 ⎤
𝝈 ∗∗ (𝜈; ∇𝐦) = 𝜈 ⎢ 3 4
⎥, 𝐪∗∗ (𝜖; ∇) ∶= 𝜖 ∇.
⎢ 𝜕𝑚 𝜕𝑚 ⎥
⎣ 1 2 3 2 2 ⎦
 4.1 Overview of the method 99

Then, we rewrite Eq. (21) as

[𝑙] [𝑙] [𝑙] [𝑙] [𝑙] [𝑙]

∫̃[𝑙] 𝑢𝐼 ℎ 𝜓ℎ dΩ + 𝐺𝐼 (𝐮ℎ , 𝜓ℎ ) = −𝑔𝐼 (𝐮ℎ , 𝜓ℎ ) + 𝐺𝐼 (𝐮ℎ , 𝜓ℎ ) + 𝑏𝐼 (𝜓ℎ ), ∀𝜓ℎ ∈ ℎ , (24)
Ωℎ

and define the following fixed-point iteration: given 𝐮[𝑙],𝑚

ℎ , we evaluate the right-hand side of (24)
and obtain 𝐮[𝑙],𝑚+1
ℎ from the left-hand side. This fixed-point iteration is accelerated via Anderson’s
method [117]. We note that the latter significantly reduces the number of required iterations.

The following remarks are in order:

(i) As said before, the density can be computed in one step. Thus, we solve Eq. (24) iteratively
only for 𝐼 = 2, 3, 4.
(ii) 𝐺𝐼 depends only on 𝑢𝐼 ℎ . Hence, the left-hand side of (24) is uncoupled.
(iii) For a given 𝐼 , the left-hand side of (24) leads to a linear combination of standard mass and stiff-
ness matrices. The resulting matrix is positive definite and remains fixed along the iterations.
To increase the efficiency, static condensation is applied to eliminate the fine-scale degrees of
freedom.
(iv) It is slightly more efficient to solve the fixed-point iterations in a Gauss–Seidel fashion. That
is, we first solve (24) for the momentum equations (𝐼 = 2, 3), and then we solve for the energy
(𝐼 = 4) with the updated momentum.
(v) The main advantage of the present approach over the classical Newton method is that the
matrix that has to be inverted4 (which stems from the left-hand side of (24)) is block diago-
nal. Hence, if there are 𝑁ℎ mesh nodes and each node shares at least one element with other
𝑘 nodes (in average term), then the cost of solving the whole system applying a conjugate
gradients method5 to each block is 3 (𝑘𝑁ℎ2 ). In contrast, in the Newton method, the cor-
responding Jacobian matrix (which stems from Eq. (21) for 𝐼 = 2, 3, 4) is formed by three
coupled blocks. Thus, in this case the Jacobian is of dimension 3𝑁ℎ × 3𝑁ℎ and has an aver-
age of 3𝑘 nonzero values per row, so the cost of solving the corresponding linear system is
 (3𝑘(3𝑁ℎ )2 ) = 27 (𝑘𝑁ℎ2 ), i.e., 9 times more expensive. In addition, it is more complicated to
compute the exact Jacobian than the approximations (22)-(23). One may argue that the New-
ton method requires less iterations to converge; however, it is still more expensive due to the
higher cost of each iteration [87].

4.1.5. Some numerical examples

To check the accuracy of the method, we have solved several benchmark problems using up to
fifth-order elements and a fourth-order RK scheme [78]. As an example, next we show the results
for the single-material triple-point problem [105, 121]; other results are available in (P3).

In this test, the rectangular domain [−1, 6]×[−1.5, 1.5] is filled with a gas of constant 𝛾 = 7/5 at the
three different states shown in Fig. 3. The gas is initially at rest in all the domain. Non-penetration
conditions are imposed on the entire boundary and the final time is 𝑇 = 5.

4 Here, “inverting a matrix” means to solve the corresponding linear system with an appropriate method, not to
explicitly compute the matrix inverse.
5 We have also tried algebraic multigrid, which requires less iterations, but it was more expensive due to the setup
cost.
100 4 General discussion

𝜌 = 0.125, 𝑝 = 0.1 1.5

𝜌 = 1, 𝑝 = 1
𝜌 = 1, 𝑝 = 0.1 1.5

1 6

Figure 3: Initial conditions for the single-material triple-point test.

Fig. 4 shows the internal energy for ℎ/𝑟 ≃ 0.027 (with ℎ the mesh size and 𝑟 the space-discretization
order). The results are nearly identical for first and fifth-order elements. The only differences are
seen at the vortex, which is slightly thinner in the case 𝑟 = 5, and at the Kelvin–Helmholtz instabil-
ities in the horizontal contact discontinuity, which are slightly more visible for 𝑟 = 5. The solution
is very good agreement with [105, 121].

3.40

2.33

1.27

0.20

Figure 4: Internal energy in the single-material, triple-point test at 𝑡 = 5 for ℎ/𝑟 ≃ 0.027.

In addition, Fig. 5 shows the evolution of the conservation errors in mass and total energy
(normalized with the theoretical mass and energy) with respect to time. As can be seen, they increase
very quickly from double precision to single precision during the first time levels. After that, they
remain stable or increase moderately. For linear elements, the conservation errors are worse (of
the order of  (10−4 )) because the employed rules of quadrature are of smaller degree — see Table
1 in Paper (P3) — and thus the remap is less accurate. For higher-order elements, however, the
conservation errors lie in a range from  (10−5 ) to  (10−8 ).

Mass Energy
10-4 10-4

10-6 10-6

10-8 10-8

10-10 10-10
r=1 r=1
r=2 r=2
10-12 r=3
10-12 r=3
r=4 r=4
r=5 r=5

10-14 10-14
0 2.5 5 0 2.5 5
t t

Figure 5: Relative errors in the conservation of mass and total energy for the single-
material triple-point test and for ℎ/𝑟 ≃ 0.027.
 4.2 Specific contribution of each paper 101

Finally, Fig. 6 shows the relation between the 𝐿1 error and the total CPU time for other two tests:
the Taylor–Green vortex, whose solution is smooth, and the Noh test, whose solution is discontin-
uous. Here, the 𝐿1 error is defined as
2
‖𝜌ℎ − 𝜌‖1,Ωℎ ‖𝑚𝑖 ℎ − 𝑚𝑖 ‖1,Ωℎ ‖ℎ − ‖1,Ωℎ
𝑒1 ∶= +∑√ + ,
‖𝜌‖1,Ωℎ 𝑖=1 ‖𝜌‖1,Ωℎ ‖‖1,Ωℎ ‖‖1,Ωℎ
where ‖𝑢‖1,Ωℎ ∶= ∫Ωℎ |𝑢| dΩ denotes the usual 𝐿1 norm. As can be seen, high-order elements perform
much better for smooth solutions, and similarly to low-order elements for discontinuous ones.
-1
10 r=1
0.6 r=2
r=3
r=4
-3
r=5
10

e1 e1
-5
10

r=1
r=2
r=3 0.1
10
-7
r=4
r=5

10
1
10
2
10
3
10
4
102 103 104
tCPU [s] tCPU [s]

Figure 6: 𝐿1 errors versus the total CPU time (in seconds) for the Taylor–Green test (left)
and for the Noh test (right).

4.2. Specific contribution of each paper

4.2.1. Paper 1: the linear convection–diffusion equation

In (P1), we develop a (nearly) conservative Lagrange–Galerkin method to solve the scalar, linear,
convection-diffusion equation. This scheme considers high-order backward differentiation formulas
for the time discretization, as well as high-order continuous space discretizations on unstructured
triangular meshes. The scheme is based on a formulation which is theoretically conservative, al-
though in practice there is a small mass loss — of the order of 10−7 with respect to the total mass —
due to the inexact computation of some terms in the formulation.

We show that the errors of this method are smaller than in the case of the standard, non-
conservative Lagrange–Galerkin method for the same meshes and time-order discretizations. We
also verify that high-order elements yield much smaller errors for the same computational effort
than low-order ones.

4.2.2. Paper 2: the nonlinear convection–diffusion equation

In (P2), we extend the method in (P1) to the case of scalar hyperbolic conservation laws, in
which the velocity field is not known a priori. As main feature, we introduce a novel discontinuity-
capturing operator which is based on post-processing the fine-scale terms of a non-stabilized solu-
tion to add artificial viscosity wherever needed. Also, we displace the Lagrangian mesh forward-in-
time, following the spirit of ALE methods, to allow for an easier implementation of stable high-order
102 4 General discussion

time discretizations. In particular, we use an implicit–explicit RK method, in which the explicit part
is used to displace the mesh and to compute the non-stabilized solution, whereas the implicit part
is used to compute the stabilized solution. Despite we use an implicit–explicit RK method, the re-
sulting systems are linear due to the structure of the equations, and hence nonlinear solvers are not
needed.

The method is tested on several benchmark problems, including the hard test by Kurganov–
Petrova–Popov [82]. The results show that the discontinuity-capturing operator is able to capture
the discontinuities accurately without spoiling the solution at the smooth regions. High-order el-
ements are able to capture the discontinuities within a smaller number of elements and are again
more efficient than low-order ones. As in (P1), mass is preserved with high accuracy.

4.2.3. Paper 3: the Euler equations for compressible flows

In (P3), we extend the method from (P2) to the case of the Euler equations for compressible
flows. In contrast to the latter, in this case the system that results from the implicit–explicit time
discretization is nonlinear, and involves four variables instead of one. We approach the resolution of
this system via Anderson’s method [117]. Furthermore, now there are nonsymmetric terms in the
Lagrange–Galerkin formulation, which makes the subgrid-stabilization even more important. The
discontinuity-capturing operator is also more complicated than in (P2), although its fundamentals
are the same. Both the discontinuity-capturing and the subgrid-stabilization operators are based on
Brenner’s model for viscous flows [20].

The results show that the method is able to accurately follow the discontinuities in the fluid
(shocks, contacts, shear layers, etc.). In addition, we corroborate that subgrid stabilization is neces-
sary even if the discontinuity-capturing operator is active, as pointed in [107]. The simulations are
performed using up to fifth-order triangular elements and a fourth-order time-marching formula.
We show that high-order elements are more efficient for smooth solutions, whereas for discontinu-
ous solutions high and low-order elements perform similarly.

4.3. Conclusions and future works

In this thesis, a novel Lagrange–Galerkin method for the resolution of compressible and inviscid
flows is proposed.

The method is formulated on a set of conservation laws for a weighted mass, momentum and
total energy from which the standard conservation laws can be easily recovered, both in the con-
tinuous and in the discrete levels, irrespective of the order the space and time discretizations. Nu-
merical experiments confirm that the scheme adequately tracks shocks and other discontinuities, as
expected from a conservative method.

In addition, the present scheme considers high-order, continuous space-discretizations on tri-

angular meshes. Numerical experiments — performed with up to fifth order elements — show that
high-order elements perform better in terms of error vs. CPU time for smooth solutions, whereas
for discontinuous solutions they perform similarly to low-order ones. In addition, the conservation
errors are slightly smaller for high-order elements.

High-order time-discretizations are considered as well. However, in contrast to the classical

 4.3 Conclusions and future works 103

Lagrange–Galerkin formulation, here we displace the mesh forward-in-time, in a fashion similar to

arbitrary Lagrangian–Eulerian methods, and not backward-in-time, as it results more simple and
stable. In particular, we employ an implicit–explicit RK method, in which the explicit part is used to
displace the mesh and the implicit part is used to compute the solution. To solve the nonlinear sys-
tem resulting from the implicit discretization, we define a stabilized fixed-point iteration and apply
Anderson’s acceleration method [117]. We note that the latter considerably reduces the number of
required iterations.

The scheme is equipped with a subgrid-stabilization operator and a with discontinuity-capturing

operator. Both rely on a multiscale representation of the solution and on Brenner’s model for viscous
flows [20]. The subgrid-stabilization operator removes acoustic instabilities that originate from the
nonsymmetric terms in the Galerkin formulation and from the inexact computation of some inte-
grals. This accomplished by adding diffusion in the fine-scale terms of the equations. In contrast,
the discontinuity-capturing operator post-processes the fine-scale terms of a non-stabilized solution
to detect the discontinuities. Then, it adds a sufficient amount of viscosity — both in the large and
fine scales — at the discontinuities, and a much smaller amount at the smooth regions. Numerical
experiments show that the subgrid-stabilization operator is essential to remove the acoustic insta-
bilities, whereas the discontinuity-capturing operator is able to detect and track the discontinuities
without spoiling the solution at the smooth regions.

Some related works could be addressed in the future. In particular, we are exploring ℎ𝑝-refinement
techniques [24, 25, 43, 45, 46] to reduce the computational cost while keeping the same numerical
accuracy. Also, we plan to implement Rosenbrock–Wanner [15, 99, 101] and LIRK-W methods [115],
which at each stage require only the solution of an implicit linear system, to avoid the resolution of
nonlinear Eq. (21). Finally, we are interested in extending the present method to the case of reactive
flows [26, 27] and to three-dimensional large-scale problems.
104 A Technical material

A. Technical material

A.1. On the bubble finite element space

To describe the space of high-order bubbles, we follow the ideas in [104]; however, here we
provide more details about the numerical implementation of its basis functions.
{ }
Recall that 𝐾̂ ∶= 𝐱̂ ∈ ℝ2 ∶ 0 ≤ 𝑥̂1 ≤ 1, 0 ≤ 𝑥̂2 ≤ 1 − 𝑥̂1 is the well-known reference element,
and that 𝑃̂𝑟 (𝐾̂ ) is the space of polynomials of maximal degree 𝑟 in 𝐾̂ , which has dimension (𝑟 + 1)(𝑟 +
2)/2. Also, the space of high-order bubbles is
{ }
̂ 𝑝 ∈ 𝑃̂𝑟−1 (𝐾̂ ), Π𝑟 𝜓 = 0 ,
𝐵̂𝑟 (𝐾̂ ) ∶= 𝜓 ∶ 𝜓 = 𝑏𝑝, (25)

̂ 𝐱) = 27̂
where 𝑏(̂ 𝑥1 𝑥̂2 (1− 𝑥̂1 − 𝑥̂2 ) the elementary bubble function and Π𝑟 is the Lagrangian interpolation
operator onto 𝑃𝑟 (𝐾̂ ). Definition (25) is equivalent to
̂
{ }
|
𝐵̂𝑟 (𝐾̂ ) ∶= 𝜓 ∈ 𝑃̂𝑟+2 (𝐾̂ ), 𝜓 | ̂ = 0, Π𝑟 𝜓 = 0 .
|𝜕 𝐾

To construct a numerical basis for 𝐵̂𝑟 (𝐾̂ ), we take advantage of the fact that, to describe polyno-
mial spaces, one normally can work with a basis {𝜙𝑖,𝑗 } such that
{ }
𝑃̂𝑘 (𝐾̂ ) = 𝑃̂𝑘−1 (𝐾̂ ) ∪ span 𝜙𝑖,𝑗 , 𝑖 + 𝑗 = 𝑘, 𝑖, 𝑗 ≥ 0 (26)

for arbitrary 𝑘. That is, 𝜙𝑖,𝑗 is a polynomial of degree 𝑖 + 𝑗 in 𝐾̂ , the polynomials 𝜙𝑖,𝑗 , 𝑖 + 𝑗 = 𝑘, are
linearly independent from polynomials of degree less than 𝑘, and the basis of 𝑃̂𝑘 (𝐾̂ ) is obtained by
adding 𝑘 + 1 functions to the basis of 𝑃̂𝑘−1 (𝐾̂ ). This happens, e.g., for the Proriol–Koornwinder–
Dubiner (PKD) basis [50, 81, 98] and for the Vandermonde basis — in the latter, 𝜙𝑖,𝑗 (̂𝐱) = 𝑥̂1𝑖 𝑥̂2𝑗 . (In our
case, we employ 𝜙𝑖,𝑗 (̂𝐱) = 𝑇𝑖 (̂ 𝑥2 ), with 𝑇𝑖 the 𝑖th Chebyshev polynomial shifted to the interval
𝑥1 )𝑇𝑗 (̂
[0, 1].) So, we can first generate an auxiliary space 𝑆̂𝑟 (𝐾̂ ) by multiplying 𝑏̂ with the basis functions
in 𝑃̂𝑟−1 (𝐾̂ ). Then, we retain only the subspace “orthogonal” to 𝑃̂𝑟 (𝐾̂ ), that is, the set of functions
𝜓 ∈ 𝑆̂𝑟 (𝐾̂ ) such that Π𝑟 𝜓 = 0. For that purpose, we define the fluctuation operator 𝜃𝑟 ∶= 𝕀 − Π𝑟 , with
𝕀 the identity. Note that (see also Fig. 7)
(i) 𝜃𝑟 𝜓 = 𝜓 − Π𝑟 𝜓 = 0 if and only if 𝜓 ∈ 𝑃̂𝑟 (𝐾̂ ),
(ii) Π𝑟 (𝜃𝑟 𝜓 ) = Π𝑟 𝜓 − Π𝑟 (Π𝑟 𝜓 ) = 0 for all 𝜓 ,
| | | |
(iii) 𝜃𝑟 𝜓 | ̂ = 𝜓 | ̂ − (Π𝑟 𝜓 )| ̂ = 0 if 𝜓 | ̂ = 0.
|𝜕 𝐾 |𝜕 𝐾 |𝜕 𝐾 |𝜕 𝐾
Hence,
{ }
𝐵̂𝑟 (𝐾̂ ) = span 𝜃𝑟 (𝑏𝜙 ̂ 𝑖,𝑗 ), 0 ≤ 𝑖 + 𝑗 ≤ 𝑟 − 1, 𝑖, 𝑗 ≥ 0 (27a)
{ }
= span 𝜃𝑟 (𝑏𝜙 ̂ 𝑖,𝑗 ), max{0, 𝑟 − 2} ≤ 𝑖 + 𝑗 ≤ 𝑟 − 1, 𝑖, 𝑗 ≥ 0 . (27b)

It is possible to show that the functions 𝜃𝑟 (𝑏𝜙̂ 𝑖,𝑗 ) in the definition (27b) are linearly independent.
̂ 𝑖,𝑗 ) = 0 (the
In effect, let us assume that there exist non-trivial coefficients 𝛼𝑖,𝑗 such that ∑𝑖,𝑗 𝛼𝑖,𝑗 𝜃𝑟 (𝑏𝜙
sum is done for max{0, 𝑟 − 2} ≤ 𝑖 + 𝑗 ≤ 𝑟 − 1). Since the operator 𝜃𝑟 is linear, this would imply that
𝑏̂ ∑𝑖,𝑗 𝛼𝑖,𝑗 𝜙𝑖,𝑗 is a polynomial of maximal degree 𝑟, which means that 𝑝𝛼 ∶= ∑𝑖,𝑗 𝛼𝑖,𝑗 𝜙𝑖,𝑗 is a polynomial
of maximal degree 𝑟 − 3. For 𝑟 < 3, this is not possible, because a polynomial has non-negative
 A.1 On the bubble finite element space 105

𝜃𝑟 𝜙 𝜙
Fluctuation 𝑃̂𝑟 (𝐾̂ )
Π𝑟 𝜙
Lagrangian interpolation
𝐵̂𝑟 (𝐾̂ )
𝑆̂𝑟 (𝐾̂ ) ∶= 𝑏̂ ⋅ 𝑃̂𝑟−1 (𝐾̂ )

Figure 7: Scheme of the construction of the basis for 𝐵̂𝑟 (𝐾̂ ). The large-scale space 𝑃̂𝑟 (𝐾̂ ) is
represented by the green straight line. The auxiliary space 𝑆̂𝑟 generated by the functions
̂ 𝜙 ∈ 𝑃̂𝑟−1 (𝐾̂ ) is represented by the yellow plane. The yellow plane contains the green
𝑏𝜙,
line to indicate that 𝑆̂𝑟 (𝐾̂ ) contains some functions of the space 𝑃̂𝑟 (𝐾̂ ). 𝐵̂𝑟 (𝐾̂ ) is the space
generated by the fluctuations of the functions in 𝑆̂𝑟 (𝐾̂ ).

degree. For 𝑟 ≥ 3, this is not possible either, because the polynomials 𝜙𝑖,𝑗 , 𝑟 − 2 ≤ 𝑖 + 𝑗 ≤ 𝑟 − 1 are of
degree at least 𝑟 − 2 and are linearly independent from the polynomials of maximal degree 𝑟 − 3, as
stated previously. Hence, such non-trivial coefficients do not exist and therefore the functions in the
basis are linearly independent. From (27b), we also deduce that the dimension of 𝐵̂𝑟 (𝐾̂ ) is 𝑟(𝑟 + 1)/2
if 𝑟 < 3, and 2𝑟 − 1 if 𝑟 ≥ 3. See Fig. 8 for an illustration of the bubble functions for the case 𝑟 = 2.

1.0

0.5

0.0

−0.5

−1.0

Figure 8: Bubble functions (normalized with their maximum value) for the case 𝑟 = 2.

{ }
Remark 9. The space defined by (27b) is unique and does not depend on the basis 𝜙𝑖,𝑗 provided that
the later satisfies Eq. (26).

It should be pointed that the definition (25) is slightly different to the one employed in Paper
(P3), which is 𝐵̂𝑟 (𝐾̂ ) ∶= 𝑏̂𝑃̂𝑟−1 (𝐾̂ ) − 𝑃̂𝑟 (𝐾̂ ). With the latter, we meant that, from the auxiliary space
𝑆̂𝑟 (𝐾̂ ) ∶= 𝑏̂𝑃̂𝑟−1 (𝐾̂ ), we have to remove the functions that are already in 𝑃̂𝑟 (𝐾̂ ). However, looking
at the analogy in Fig. 7, it seems unclear how to define a space such as 𝑆̂𝑟 (𝐾̂ ) − 𝑃̂𝑟 (𝐾̂ ). In addition,
this space is not linear, because two functions may belong to 𝑆̂𝑟 (𝐾̂ ) and not to 𝑃̂𝑟 (𝐾̂ ), but a linear
combination of them may do belong to 𝑃̂𝑟 (𝐾̂ ). With the new definition, 𝐵̂𝑟 (𝐾̂ ) is a linear space and
the removal operation is better defined by the condition Π𝑟 𝜓 = 0. In addition, Eq. (27b) represents
the way in which we have numerically implemented the bubble functions.
106 A Technical material

A.2. On the local conservation property

Local conservation is a fundamental property to adequately capture any discontinuities in the

solution. Also, together with consistency and stability, it guarantees that the method converges to
a weak solution of Brenner’s equations [1–3]. To demonstrate that this property is satisfied by our
scheme, we follow an approach similar to Hughes et al [70].

For simplicity, let us consider the scalar scale, with strong Dirichlet conditions 𝑢 = 𝑢𝐷 on the
boundary Γ𝐷 (𝑡) ⊆ 𝜕 Ω̃ ℎ (𝑡) and weak Neumann conditions 𝑞 ∶= 𝑓𝑗 (𝑢)𝑛𝑗 = 𝑞𝑁 on Γ𝑁 (𝑡) ⊆ 𝜕 Ω
̃ ℎ (𝑡). That
̃𝐷
is, we have to find 𝑢ℎ (𝐱, 𝑡) ∈ 𝑉ℎ (𝑡) such that
d
𝑢ℎ 𝜓ℎ dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ dΩ = ∮ 𝑞𝜓ℎ d𝜎, ∀𝜓ℎ ∈ ℎ0 , (28)
d𝑡 ∫Ω̃ℎ ̃ℎ
Ω ̃ℎ
𝜕Ω

with 𝑓𝑗 ℎ = 𝑓𝑗 (𝑢ℎ ), 𝑉̃ℎ𝐷 (𝑡) ∶= {𝜓 ∈ 𝑉̃ℎ (𝑡) ∶ 𝜓 |Γ𝐷 = 𝑢𝐷 } and ℎ0 ∶= {𝜓 ∈ ℎ ∶ 𝜓 |Γ𝐷 = 0}. Recall that, in
our scheme, there are two types of degrees of freedom: those associated with the mesh nodes and
the corresponding Lagrangian interpolation polynomials6 , and those related to the bubble functions.

In principle, a global mass balance can only be satisfied if Γ𝐷 = ∅. In that case, the function
𝜓ℎ ≡ 1 is contained in the trial space ℎ0 and Eq. (28) yields
d
𝑢ℎ dΩ = ∮ 𝑞 d𝜎.
d𝑡 ∫Ω̃ℎ ̃ℎ
𝜕Ω

In the general case Γ𝐷 ≠ ∅, the function 𝜓ℎ ≡ 1 ∉ ℎ0 and the latter equation does not apply. However,
̃ℎ.
it is still possible to write a global mass balance by defining an appropriate flux at the boundary 𝜕 Ω
𝐵 ̃ 𝐵
In particular, let 𝑞ℎ be a numerical approximation to the flux 𝑞 on 𝜕 Ωℎ . The flux 𝑞ℎ is defined only
at the boundary 𝜕 Ω ̃ ℎ and is of the form
| |
𝑞ℎ𝐵 | ̃ = ∑ 𝜓ℎ,𝛽 | ̃ 𝑄𝛽𝐵 ,
|𝜕 Ωℎ ̃
|𝜕 Ωℎ
𝛽∈𝜕 Ωℎ

where 𝛽 ∈ 𝜕 Ω̃ ℎ denotes that the sum is performed for all the nodes 𝛽 placed at 𝜕 Ω
̃ ℎ , and 𝜓ℎ,𝛽 is the
𝐵
basis function associated with 𝛽. We define the coefficients 𝑄𝛽 in such a way that
d
𝑢ℎ 𝜓ℎ,𝛼 dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ,𝛼 dΩ = ∮ 𝑞ℎ𝐵 𝜓ℎ,𝛼 d𝜎, ∀𝛼. (29)
d𝑡 ∫Ω̃ℎ ̃ℎ
Ω ̃ℎ
𝜕Ω

If 𝛼 refers to a node inner to the domain Ω ̃ ℎ or to a bubble degree of freedom, Eq. (29) yields the
identity 0 = 0, due to Eq. (28) and to the fact that 𝜓ℎ,𝛼 |𝜕 Ω̃ℎ = 0. Otherwise, 𝛼 ∈ 𝜕 Ω ̃ ℎ and Eq. (29)
𝐵
yields a linear system which determines the coefficients 𝑄𝛽 . Now we note that 1 ≡ ∑𝛼∈ (Ω̃ℎ ) 𝜓ℎ,𝛼 ,
where  (Ω ̃ ℎ ) is the set of mesh nodes in Ω̃ ℎ . Hence, by construction, Eq. (29) provides the global
mass balance
d
𝑢ℎ dΩ = ∮ 𝑞ℎ𝐵 d𝜎.
d𝑡 ∫Ω̃ℎ ̃ℎ
𝜕Ω

Now let us divide Ω̃ ℎ in two arbitrary subdomains Ω ̃ −ℎ and Ω

̃ +ℎ formed by connected elements
and separated by a boundary Γ (see Fig. 9). We can also construct numerical fluxes so as to satisfy
balance laws at these subdomains. In particular, we seek fluxes 𝑞ℎ− and 𝑞ℎ+ , defined only on Γ, of the
form
𝑞ℎ− = ∑ 𝜓ℎ,𝛽 𝑄𝛽− , 𝑞ℎ+ = ∑ 𝜓ℎ,𝛽 𝑄𝛽+ ,
𝛽∈Γ 𝛽∈Γ

6 Technically, they are polynomials in the reference element, not in the physical elements.
 A.2 On the local conservation property 107

so that
d
∫ 𝑢ℎ 𝜓ℎ,𝛼 dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ,𝛼 dΩ = ∫ 𝑞ℎ𝐵 𝜓ℎ,𝛼 d𝜎 + ∫ 𝑞ℎ− 𝜓ℎ,𝛼 d𝜎, ∀𝛼, (30)
d𝑡 Ω̃ℎ − ̃
Ω −
ℎ 𝜕Ω̃ ℎ ∩𝜕 Ω
̃ −
ℎ Γ
d
𝑢ℎ 𝜓ℎ,𝛼 dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ,𝛼 dΩ = ∫ 𝑞ℎ𝐵 𝜓ℎ,𝛼 d𝜎 + ∫ 𝑞ℎ+ 𝜓ℎ,𝛼 d𝜎, ∀𝛼. (31)
d𝑡 ∫Ω̃+ℎ ̃+
Ω ℎ 𝜕 ̃ ℎ ∩𝜕 Ω
Ω ̃+
ℎ Γ

Again, the relations yield two systems of equations that determine the coefficients 𝑄𝛽− and 𝑄𝛽+ when
̃ −ℎ and that
𝛼 is a node in Γ. Otherwise, they yield the identity (29). Noting that 1 ≡ ∑𝛼∈ (Ω̃−ℎ ) 𝜓ℎ,𝛼 in Ω
̃ +ℎ , by construction, Eqs. (30)-(31) yield the local mass balances
1 ≡ ∑𝛼∈ (Ω̃+ℎ ) 𝜓ℎ,𝛼 in Ω

d
𝑢ℎ dΩ = ∫ 𝑞ℎ𝐵 d𝜎 + ∫ 𝑞ℎ− d𝜎,
d𝑡 ∫Ω̃−ℎ ̃ ̃
𝜕 Ωℎ ∩𝜕 Ωℎ − Γ
d
𝑢ℎ dΩ = ∫ 𝑞ℎ𝐵 d𝜎 + ∫ 𝑞ℎ+ d𝜎.
d𝑡 ∫Ω̃+ℎ 𝜕Ω̃ ℎ ∩𝜕 Ω
̃ +
ℎ Γ

Finally, if we sum Eqs. (30)-(31) and use Eq. (29), we arrive at

− +
∫ (𝑞ℎ + 𝑞ℎ ) 𝜓ℎ,𝛼 d𝜎 = 0, ∀𝛼,
Γ

which can only be satisfied if 𝑞ℎ− + 𝑞ℎ+ ≡ 0. That is, as expected, the flux that enters one subdomain
is the same that the one that leaves the other subdomain.

̃ −ℎ
Ω
Γ

̃ −ℎ
Ω
Γ

̃ +ℎ
Ω ̃ +ℎ
Ω

̃ ℎ divided into two subdomains Ω

Figure 9: Domain Ω ̃ − and Ω
̃ + by a boundary Γ.
ℎ ℎ

Remark 10. In discontinuous Galerkin methods, we define a flux at each edge of the mesh. Then, if
we consider two arbitrary lines Γ1 and Γ2 (formed by connected edges) that contain a certain edge Γ𝑒 :
(i) the flux at Γ1 or Γ2 is continuous at the edges but discontinuous at the mesh vertices, and (ii) the flux
in Γ𝑒 is the same in both cases. However, in continuous Galerkin methods, it is not clear how to define
a flux at each edge of the mesh. Instead, with the procedure above, (i) the flux at Γ1 or Γ2 is continuous
at the mesh vertices, and (ii) the flux at the edge Γ𝑒 depends on the considered line. See also Fig. 10.

It is also possible and more simple to write a conservation balance in a pointwise sense. Indeed,
for a given element 𝐾 and each degree of freedom 𝛼 related to 𝐾 , we define the generalized flux (or
element nodal flux in [70])
d
𝐾𝛼 ∶= 𝑢ℎ 𝜓ℎ,𝛼 dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ,𝛼 dΩ. (32)
d𝑡 ∫𝐾 𝐾
108 A Technical material

Γ1

Γ𝑒
Γ2

Figure 10: In continuous Galerkin methods, the flux along a line (Γ1 or Γ2 ) is continuous
at the mesh vertices, and the flux at a given edge Γ𝑒 depends on the considered line.

Comparing (32) with (28), we see that 𝐾𝛼 can be interpreted as the contribution of the degree of
freedom 𝛼 to the total flux through 𝜕𝐾 , or as a delta-type flux entering 𝐾 through the node 𝛼. Note
that, if 𝛼 is a node inner to 𝐾 or a bubble degree of freedom, 𝐾𝛼 = 0 due to Eq. (29). Otherwise, 𝛼 is
a node in 𝜕𝐾 . Furthermore, 1 ≡ ∑𝛼∈ (𝐾 ) 𝜓ℎ,𝛼 in 𝐾 . Hence,

d
∑ 𝐾𝛼 = ∑ 𝐾𝛼 = 𝑢ℎ dΩ. (33)
𝛼∈𝜕𝐾 𝛼∈ (𝐾 )
d𝑡 ∫𝐾

On the other hand, Eq. (29) can be written as

d 𝐵
∑
[ d𝑡 ∫ 𝑢ℎ 𝜓ℎ,𝛼 dΩ + ∫ 𝑓𝑗 ℎ 𝜕𝑗 𝜓ℎ,𝛼 dΩ] = ∮ ̃ 𝑞ℎ 𝜓ℎ,𝛼 d𝜎, ∀𝛼,
𝐾 ∋𝛼 𝐾 𝐾 𝜕 Ωℎ

where 𝐾 ∋ 𝛼 denotes that the sum is performed for all the elements containing the node 𝛼. Using
now Eq. (32), we have
∑ 𝐾𝛼 = ∮ 𝑞ℎ𝐵 𝜓ℎ,𝛼 d𝜎. (34)
𝐾 ∋𝛼 ̃ℎ
𝜕Ω

For the nodes inner to the elements and the bubble degrees of freedom, the latter equation is the
identity 0 = 0. For nodes placed at the mesh skeleton — i.e., the set of edges of the triangles —,
but not at the boundary 𝜕 Ω ̃ ℎ , it reads ∑𝐾 ∋𝛼 𝐾𝛼 = 0, that is, the sum of generalized fluxes at each
node is zero. Finally, for the nodes at the boundary 𝜕 Ω ̃ ℎ , the latter equation implies that the sum of
generalized fluxes at each node is related to the (weighted) flux through the boundary. Hence, Eq.
(33) yields a pointwise mass conservation law on each element 𝐾 , whereas Eq. (34) states that the
pointwise fluxes do not create nor destroy mass at the nodes.

It is worth remarking that the pointwise conservation balance above, due to Hughes et al. [70],
is very similar to the approach employed by Abgrall et al. [1, 3]; however, the latter authors define
residuals instead of fluxes.
 References 109

References

[1] Abgrall, R.; Lipnikov, K.; Morgan, N.; Tokareva, S., “Multidimensional Staggered Grid Residual Distri-
bution Scheme for Lagrangian Hydrodynamics,” SIAM Journal on Scientific Computing, vol. 42(1):pp.
A343–A370, 2020, URL http://dx.doi.org/10.1137/18M1223939.
[2] Abgrall, R.; Roe, P.L., “High Order Fluctuation Schemes on Triangular Meshes,” Journal of Scientific
Computing, vol. 19(1):pp. 3–36, 2003, URL http://dx.doi.org/10.1023/A:1025335421202.
[3] Abgrall, R.; Tokareva, S., “Staggered Grid Residual Distribution Scheme for Lagrangian Hydrodynam-
ics,” SIAM Journal on Scientific Computing, vol. 39(5):pp. A2317–A2344, 2017, URL http://dx.doi.
org/10.1137/16M1078781.
[4] Al-Lawatia, M., “A Higher-Order Eulerian-Lagrangian Localized Adjoint Method for Two-Dimensional
Unsteady Advection-Diffusion Problems,” Journal of Computational Mathematics, vol. 30(3):pp. 324–
336, 2012, URL http://dx.doi.org/10.4208/jcm.1110-m3465.
[5] Alauzet, F., “A parallel matrix-free conservative solution interpolation on unstructured tetrahedral
meshes,” Computer Methods in Applied Mechanics and Engineering, vol. 299:pp. 116–142, 2016, URL
http://dx.doi.org/10.1016/j.cma.2015.10.012.
[6] Anderson, R.W.; Dobrev, V.A.; Kolev, T.V.; Rieben, R.N., “Monotonicity in high-order curvilinear finite
element arbitrary Lagrangian-Eulerian remap,” International Journal for Numerical Methods in Fluids,
vol. 77(5):pp. 249–273, 2015, URL http://dx.doi.org/10.1002/fld.3965.
[7] Anderson Jr., J.D., Fundamentals of Aerodynamics, McGraw-Hill Education, sixth ed., 2016, URL https:
//www.mheducation.com/highered/product/fundamentals-aerodynamics-anderson/
M9781259129919.html.
[8] Arnold, D.N.; Brezzi, F.; Cockburn, B.; Marini, L.D., “Unified Analysis of Discontinuous Galerkin Meth-
ods for Elliptic Problems,” SIAM Journal on Numerical Analysis, vol. 39(5):pp. 1749–1779, 2002, URL
http://dx.doi.org/10.1137/S0036142901384162.
[9] Barter, G.E.; Darmofal, D.L., “Shock capturing with PDE-based artificial viscosity for DGFEM: Part I.
Formulation,” Journal of Computational Physics, vol. 229(5):pp. 1810–1827, 2010, URL http://dx.doi.
org/10.1016/j.jcp.2009.11.010.
[10] Bermejo, R.; Galán del Sastre, P.; Saavedra, L., “A Second Order in Time Modified Lagrange–Galerkin
Finite Element Method for the Incompressible Navier–Stokes Equations,” SIAM Journal on Numerical
Analysis, vol. 50(6):pp. 3084–3109, 2012, URL http://dx.doi.org/10.1137/11085548X.
[11] Bermejo, R.; Saavedra, L., “Modified Lagrange–Galerkin Methods to Integrate Time Dependent Incom-
pressible Navier–Stokes Equations,” SIAM Journal on Scientific Computing, vol. 37(6):pp. B779–B803,
2015, URL http://dx.doi.org/10.1137/140973967.
[12] Bermejo, R.; Saavedra, L., “Lagrange–Galerkin methods for the incompressible Navier-Stokes equa-
tions: a review,” Communications in Applied and Industrial Mathematics, vol. 7(3):pp. 26–55, 2016, URL
http://dx.doi.org/10.1515/caim-2016-0021.
[13] Bermejo, R.; Saavedra, L., “A second order in time local projection stabilized Lagrange–Galerkin method
for Navier–Stokes equations at high Reynolds numbers,” Computers & Mathematics with Applications,
vol. 72(4):pp. 820–845, 2016, URL http://dx.doi.org/10.1016/j.camwa.2016.05.012.
[14] Bezanson, J.; Edelman, A.; Karpinski, S.; Shah, V.B., “Julia: A Fresh Approach to Numerical Computing,”
SIAM Review, vol. 59(1):pp. 65–98, 2017, URL http://dx.doi.org/10.1137/141000671.
[15] Blom, D.S.; Birken, P.; Bijl, H.; Kessels, F.; Meister, A.; van Zuijlen, A.H., “A comparison of Rosen-
brock and ESDIRK methods combined with iterative solvers for unsteady compressible flows,” Advances
in Computational Mathematics, vol. 42(6):pp. 1401–1426, 2016, URL http://dx.doi.org/10.1007/
s10444-016-9468-x.
[16] Blyth, M.G.; Luo, H.; Pozrikidis, C., “A comparison of interpolation grids over the triangle or the tetra-
hedron,” Journal of Engineering Mathematics, vol. 56(3):pp. 263–272, 2007, URL http://dx.doi.org/
10.1007/s10665-006-9063-0.
[17] Bonaventura, L., “An introduction to semi - Lagrangian methods for geophysical scale flows,” Tech. rep.,
Politecnico di Milano, 2004, URL https://www.researchgate.net/publication/228927793_
An_introduction_to_semi-Lagrangian_methods_for_geophysical_scale_flows.
110 References

[18] Boscheri, W.; Dumbser, M., “Arbitrary-Lagrangian–Eulerian Discontinuous Galerkin schemes with a
posteriori subcell finite volume limiting on moving unstructured meshes,” Journal of Computational
Physics, vol. 346:pp. 449–479, 2017, URL http://dx.doi.org/10.1016/j.jcp.2017.06.022.
[19] Boyce, M.P., Gas Turbine Engineering Handbook, Elsevier, 2012, URL http://dx.doi.org/10.1016/
C2009-0-64242-2.
[20] Brenner, H., “Fluid mechanics revisited,” Physica A: Statistical Mechanics and its Applications, vol.
370(2):pp. 190–224, 2006, URL http://dx.doi.org/10.1016/j.physa.2006.03.066.
[21] Brezzi, F.; Franca, L.; Hughes, T.; Russo, A., “$b=\int g$,” Computer Methods in Applied Mechanics
and Engineering, vol. 145:pp. 329–339, 1997, URL http://dx.doi.org/10.1016/S0045-7825(96)
01221-2.
[22] Busto, S.; Dumbser, M.; Peshkov, I.; Romenski, E., “On Thermodynamically Compatible Finite Volume
Schemes for Continuum Mechanics,” SIAM Journal on Scientific Computing, vol. 44(3):pp. A1723–A1751,
2022, URL http://dx.doi.org/10.1137/21M1417508.
[23] Cangiani, A.; Chapman, J.; Georgoulis, E.; Jensen, M., “On the Stability of Continuous–Discontinuous
Galerkin Methods for Advection–Diffusion–Reaction Problems,” Journal of Scientific Computing,
vol. 57(2):pp. 313–330, 2013, URL http://dx.doi.org/10.1007/s10915-013-9707-y.
[24] Carpio, J.; Prieto, J.L., “An anisotropic, fully adaptive algorithm for the solution of convection-
dominated equations with semi-Lagrangian schemes,” Computer Methods in Applied Mechanics and
Engineering, vol. 273:pp. 77–99, 2014, URL http://dx.doi.org/10.1016/j.cma.2014.01.025.
[25] Carpio, J.; Prieto, J.L.; Bermejo, R., “Anisotropic “Goal-Oriented” Mesh Adaptivity for Elliptic Problems,”
SIAM Journal on Scientific Computing, vol. 35(2):pp. A861–A885, 2013, URL http://dx.doi.org/10.
1137/120874606.
[26] Carpio, J.; Prieto, J.L.; Galán del Sastre, P., “An anisotropic adaptive, Lagrange–Galerkin numerical
method for spray combustion,” Journal of Computational Physics, vol. 381:pp. 246–274, 2019, URL http:
//dx.doi.org/10.1016/j.jcp.2018.12.022.
[27] Carpio, J.; Prieto, J.L.; Vera, M., “A local anisotropic adaptive algorithm for the solution of low-Mach
transient combustion problems,” Journal of Computational Physics, vol. 306:pp. 19–42, 2016, URL http:
//dx.doi.org/10.1016/j.jcp.2015.11.011.
[28] Chan, J.; Demkowicz, L.; Moser, R., “A DPG method for steady viscous compressible flow,” Computers &
Fluids, vol. 98:pp. 69–90, 2014, URL http://dx.doi.org/10.1016/j.compfluid.2014.02.024.
[29] Cheng, J.; Yang, X.; Liu, X.; Liu, T.; Luo, H., “A direct discontinuous Galerkin method for the com-
pressible Navier–Stokes equations on arbitrary grids,” Journal of Computational Physics, vol. 327:pp.
484–502, 2016, URL http://dx.doi.org/10.1016/j.jcp.2016.09.049.
[30] Clay Mathematics Institute, “Millennium Problems,” 2021, URL http://www.claymath.org/
millennium-problems.
[31] Codina, R.; Badia, S.; Baiges, J.; Principe, J., “Variational Multiscale Methods in Computational Fluid
Dynamics,” in E. Stein; R. de Borst; T.J.R. Hughes (Eds.), Encyclopedia of Computational Mechanics Second
Edition, pp. 1–28, John Wiley & Sons, Ltd, Chichester, UK, 2018, URL http://dx.doi.org/10.1002/
9781119176817.ecm2117.
[32] Colera, M.; Carpio, J.; Bermejo, R., “A nearly-conservative high-order Lagrange–Galerkin method for
the resolution of scalar convection-dominated equations in non-divergence-free velocity fields,” Com-
puter Methods in Applied Mechanics and Engineering, vol. 372:p. 113366, 2020, URL http://dx.doi.
org/10.1016/j.cma.2020.113366.
[33] Colera, M.; Carpio, J.; Bermejo, R., “A nearly-conservative, high-order, forward Lagrange–Galerkin
method for the resolution of scalar hyperbolic conservation laws,” Computer Methods in Applied
Mechanics and Engineering, vol. 376:p. 113654, 2021, URL http://dx.doi.org/10.1016/j.cma.
2020.113654.
[34] Colera, M.; Carpio, J.; Bermejo, R., “A nearly-conservative, high-order, forward Lagrange–Galerkin
method for the resolution of compressible flows on unstructured triangular meshes,” Journal of Com-
putational Physics, vol. 467:p. 111471, 2022, URL http://dx.doi.org/10.1016/j.jcp.2022.
111471.
 References 111

[35] Cormie, D.; Mays, G.; Smith, P. (Eds.), Blast effects on buildings, Thomas Telford Publishing, second
edition ed., 2009, URL http://dx.doi.org/10.1680/beob2e.35218.
[36] Cremonesi, M.; Frangi, A., “A Lagrangian finite element method for 3D compressible flow applications,”
Computer Methods in Applied Mechanics and Engineering, vol. 311:pp. 374–392, 2016, URL http://dx.
doi.org/10.1016/j.cma.2016.08.005.
[37] Crouseilles, N.; Respaud, T.; Sonnendrücker, E., “A forward semi-Lagrangian method for the numerical
solution of the Vlasov equation,” Computer Physics Communications, vol. 180(10):pp. 1730–1745, 2009,
URL http://dx.doi.org/10.1016/j.cpc.2009.04.024.
[38] Dao, T.A.; Nazarov, M., “A High-Order Residual-Based Viscosity Finite Element Method for the Ideal
MHD Equations,” Journal of Scientific Computing, vol. 92(3):p. 77, 2022, URL http://dx.doi.org/
10.1007/s10915-022-01918-4.
[39] Demkowicz, L.; Gopalakrishnan, J., “An overview of the DPG method,” Tech. rep., Institute for Compu-
tational Engineering and Sciences, The University of Texas at Austin, 2013, URL https://www.oden.
utexas.edu/media/reports/2013/1302.pdf.
[40] Diosady, L.T.; Murman, S.M., “Design of a Variational Multiscale Method for Turbulent Compressible
Flows,” in AIAA Computational Fluid Dynamics Conference, San Diego, CA, 2013, URL https://ntrs.
nasa.gov/citations/20140010516.
[41] Dobrev, V.A.; Ellis, T.E.; Kolev, T.V.; Rieben, R.N., “Curvilinear finite elements for Lagrangian hydrody-
namics,” International Journal for Numerical Methods in Fluids, vol. 65(11-12):pp. 1295–1310, 2011, URL
http://dx.doi.org/10.1002/fld.2366.
[42] Dobrev, V.A.; Kolev, T.V.; Rieben, R.N., “High-Order Curvilinear Finite Element Methods for Lagrangian
Hydrodynamics,” SIAM Journal on Scientific Computing, vol. 34(5):pp. B606–B641, 2012, URL http:
//dx.doi.org/10.1137/120864672.
[43] Dolejší, V., “Anisotropic hp-adaptive method based on interpolation error estimates in the Lq-norm,”
Applied Numerical Mathematics, vol. 82:pp. 80–114, 2014, URL http://dx.doi.org/10.1016/j.
apnum.2014.03.003.
[44] Dolejší, V.; Feistauer, M., Discontinuous Galerkin Method, Springer Series in Computational Math-
ematics, Springer Cham, first ed., 2015, URL https://link.springer.com/book/10.1007/
978-3-319-19267-3.
[45] Dolejší, V.; May, G., Anisotropic hp-Mesh Adaptation Methods: Theory, implementation and applications,
Nečas Center Series, Springer International Publishing, Cham, 2022, URL http://dx.doi.org/10.
1007/978-3-031-04279-9.
[46] Dolejší, V.; May, G.; Rangarajan, A., “A continuous hp-mesh model for adaptive discontinuous Galerkin
schemes,” Applied Numerical Mathematics, vol. 124:pp. 1–21, 2018, URL http://dx.doi.org/10.
1016/j.apnum.2017.09.015.
[47] Dolejší, V.; Svärd, M., “Numerical study of two models for viscous compressible fluid flows,” Journal of
Computational Physics, vol. 427:p. 110068, 2021, URL http://dx.doi.org/10.1016/j.jcp.2020.
110068.
[48] Donea, J.; Huerta, A.; Ponthot, J.P.; Rodríguez-Ferrán, A., “Arbitrary Lagrangian–Eulerian methods,”
in E. Stein; R. de Borst; T.J.R. Hughes (Eds.), Encyclopedia of computational mechanics, John Wiley,
Chichester, West Sussex, 2004.
[49] Donea, J.; Quartapelle, L., “An introduction to finite element methods for transient advection problems,”
Computer Methods in Applied Mechanics and Engineering, vol. 95(2):pp. 169–203, 1992, URL http://
dx.doi.org/10.1016/0045-7825(92)90139-B.
[50] Dubiner, M., “Spectral methods on triangles and other domains,” Journal of Scientific Computing,
vol. 6(4):pp. 345–390, 1991, URL http://dx.doi.org/10.1007/BF01060030.
[51] Dumbser, M.; Zanotti, O.; Loubère, R.; Diot, S., “A posteriori subcell limiting of the discontinuous
Galerkin finite element method for hyperbolic conservation laws,” Journal of Computational Physics,
vol. 278:pp. 47–75, 2014, URL http://dx.doi.org/10.1016/j.jcp.2014.08.009.
[52] Farrell, P.; Maddison, J., “Conservative interpolation between volume meshes by local Galerkin pro-
jection,” Computer Methods in Applied Mechanics and Engineering, vol. 200(1-4):pp. 89–100, 2011, URL
http://dx.doi.org/10.1016/j.cma.2010.07.015.
112 References

[53] Feireisl, E.; Vasseur, A., “New Perspectives in Fluid Dynamics: Mathematical Analysis of a Model
Proposed by Howard Brenner,” in A.V. Fursikov; G.P. Galdi; V.V. Pukhnachev (Eds.), New Directions
in Mathematical Fluid Mechanics: The Alexander V. Kazhikhov Memorial Volume, Advances in Mathe-
matical Fluid Mechanics, pp. 153–179, Birkhäuser, Basel, 2010, URL http://dx.doi.org/10.1007/
978-3-0346-0152-8_9.
[54] Fernandez, P.; Nguyen, C.; Peraire, J., “A physics-based shock capturing method for unsteady laminar
and turbulent flows,” in 2018 AIAA Aerospace Sciences Meeting, American Institute of Aeronautics and
Astronautics, Kissimmee, Florida, 2018, URL http://dx.doi.org/10.2514/6.2018-0062.
[55] Freund, J., “The space-continuous–discontinuous Galerkin method,” Computer Methods in Applied
Mechanics and Engineering, vol. 190(26):pp. 3461–3473, 2001, URL http://dx.doi.org/10.1016/
S0045-7825(00)00279-6.
[56] Fries, T.P.; Matthies, H.G., “A Review of Petrov-Galerkin Stabilization Approaches and an Extension
to Meshfree Methods,” Tech. rep., Institute of Scientific Computing - Technische Universität Braun-
schweigh, 2003, URL https://publikationsserver.tu-braunschweig.de/receive/dbbs_
mods_00001549.
[57] Fung, Y.C., An Introduction to the Theory of Aeroelasticity, Dover Publications, Mineola, N.Y., 1st ed.,
2008.
[58] Futai, K.; Kolbe, N.; Notsu, H.; Suzuki, T., “A Mass-Preserving Two-Step Lagrange–Galerkin Scheme
for Convection-Diffusion Problems,” Journal of Scientific Computing, vol. 92(2):p. 37, 2022, URL http:
//dx.doi.org/10.1007/s10915-022-01885-w.
[59] Giraldo, F.X., “The Lagrange–Galerkin method for the two-dimensional shallow water equations on
adaptive grids,” International Journal for Numerical Methods in Fluids, vol. 33(6):pp. 789–832, 2000, URL
http://dx.doi.org/10.1002/1097-0363(20000730)33:6<789::AID-FLD29>3.0.CO;2-1.
[60] Giraldo, F.X., “Strong and weak Lagrange-Galerkin spectral element methods for the shallow water
equations,” Computers & Mathematics with Applications, vol. 45(1-3):pp. 97–121, 2003, URL http://
dx.doi.org/10.1016/S0898-1221(03)80010-X.
[61] Giraldo, F.X.; Warburton, T., “A nodal triangle-based spectral element method for the shallow water
equations on the sphere,” Journal of Computational Physics, vol. 207(1):pp. 129–150, 2005, URL http:
//dx.doi.org/10.1016/j.jcp.2005.01.004.
[62] Guermond, J.L., “Stabilization of Galerkin approximations of transport equations by subgrid modeling,”
ESAIM: Mathematical Modelling and Numerical Analysis, vol. 33(6):pp. 1293–1316, 1999, URL http:
//dx.doi.org/10.1051/m2an:1999145.
[63] Guermond, J.L.; Marra, A.; Quartapelle, L., “Subgrid stabilized projection method for 2D unsteady flows
at high Reynolds numbers,” Computer Methods in Applied Mechanics and Engineering, vol. 195(44-47):pp.
5857–5876, 2006, URL http://dx.doi.org/10.1016/j.cma.2005.08.016.
[64] Guermond, J.L.; Pasquetti, R.; Popov, B., “Entropy viscosity method for nonlinear conservation laws,”
Journal of Computational Physics, vol. 230(11):pp. 4248–4267, 2011, URL http://dx.doi.org/10.
1016/j.jcp.2010.11.043.
[65] Guermond, J.L.; Popov, B., “Viscous Regularization of the Euler Equations and Entropy Principles,”
SIAM Journal on Applied Mathematics, vol. 74(2):pp. 284–305, 2014, URL http://dx.doi.org/10.
1137/120903312.
[66] Guermond, J.L.; Popov, B.; Saavedra, L., “Second-order invariant domain preserving ALE approximation
of hyperbolic systems,” Journal of Computational Physics, vol. 401:p. 108927, 2020, URL http://dx.
doi.org/10.1016/j.jcp.2019.108927.
[67] Guermond, J.L.; Popov, B.; Saavedra, L.; Yang, Y., “Invariant Domains Preserving Arbitrary La-
grangian Eulerian Approximation of Hyperbolic Systems with Continuous Finite Elements,” SIAM Jour-
nal on Scientific Computing, vol. 39(2):pp. A385–A414, 2017, URL http://dx.doi.org/10.1137/
16M1063034.
[68] Guermond, J.L.; Popov, B.; Tomov, V., “Entropy–viscosity method for the single material Euler equations
in Lagrangian frame,” Computer Methods in Applied Mechanics and Engineering, vol. 300:pp. 402–426,
2016, URL http://dx.doi.org/10.1016/j.cma.2015.11.009.
 References 113

[69] Huerta, A.; Angeloski, A.; Roca, X.; Peraire, J., “Efficiency of high-order elements for continuous
and discontinuous Galerkin methods,” International Journal for Numerical Methods in Engineering,
vol. 96(9):pp. 529–560, 2013, URL http://dx.doi.org/10.1002/nme.4547.
[70] Hughes, T.J.; Engel, G.; Mazzei, L.; Larson, M.G., “The Continuous Galerkin Method Is Locally Conser-
vative,” Journal of Computational Physics, vol. 163(2):pp. 467–488, 2000, URL http://dx.doi.org/
10.1006/jcph.2000.6577.
[71] Hughes, T.J.; Feijóo, G.R.; Mazzei, L.; Quincy, J.B., “The variational multiscale method—a paradigm for
computational mechanics,” Computer Methods in Applied Mechanics and Engineering, vol. 166(1-2):pp.
3–24, 1998, URL http://dx.doi.org/10.1016/S0045-7825(98)00079-6.
[72] Hughes, T.J.R.; Scovazzi, G.; Franca, L.P., “Multiscale and Stabilized Methods,” in E. Stein; R. de Borst;
T.J.R. Hughes (Eds.), Encyclopedia of Computational Mechanics, pp. 1–64, John Wiley & Sons, Ltd, Chich-
ester, UK, 2017, URL http://dx.doi.org/10.1002/9781119176817.ecm2051.
[73] Hughes, T.J.R.; Scovazzi, G.; Tezduyar, T.E., “Stabilized Methods for Compressible Flows,” Jour-
nal of Scientific Computing, vol. 43(3):pp. 343–368, 2010, URL http://dx.doi.org/10.1007/
s10915-008-9233-5.
[74] Kaazempur-Mofrad, M.; Ethier, C., “An efficient characteristic Galerkin scheme for the advection equa-
tion in 3-D,” Computer Methods in Applied Mechanics and Engineering, vol. 191(46):pp. 5345–5363, 2002,
URL http://dx.doi.org/10.1016/S0045-7825(02)00461-9.
[75] Kaazempur-Mofrad, M.; Minev, P.; Ethier, C., “A characteristic/finite element algorithm for time-
dependent 3-D advection-dominated transport using unstructured grids,” Computer Methods in Ap-
plied Mechanics and Engineering, vol. 192(11-12):pp. 1281–1298, 2003, URL http://dx.doi.org/10.
1016/S0045-7825(02)00627-8.
[76] Kapferer, W., “Supersonic CFD simulation TYCHO,” 2021, URL https://www.youtube.com/watch?
v=kaVDDm222H8.
[77] Kelly, J.F.; Giraldo, F.X., “Continuous and discontinuous Galerkin methods for a scalable three-
dimensional nonhydrostatic atmospheric model: Limited-area mode,” Journal of Computational Physics,
vol. 231(24):pp. 7988–8008, 2012, URL http://dx.doi.org/10.1016/j.jcp.2012.04.042.
[78] Kennedy, C.A.; Carpenter, M.H., “Higher-order additive Runge–Kutta schemes for ordinary differential
equations,” Applied Numerical Mathematics, vol. 136:pp. 183–205, 2019, URL http://dx.doi.org/
10.1016/j.apnum.2018.10.007.
[79] Kim, M.Y.; Wheeler, M.F., “A multiscale discontinuous Galerkin method for convec-
tion–diffusion–reaction problems,” Computers & Mathematics with Applications, vol. 68(12):pp.
2251–2261, 2014, URL http://dx.doi.org/10.1016/j.camwa.2014.08.007.
[80] Kirby, R.M.; Sherwin, S.J.; Cockburn, B., “To CG or to HDG: A Comparative Study,” Jour-
nal of Scientific Computing, vol. 51(1):pp. 183–212, 2012, URL http://dx.doi.org/10.1007/
s10915-011-9501-7.
[81] Koornwinder, T., “Two-Variable Analogues of the Classical Orthogonal Polynomials,” in Theory and
Application of Special Functions, pp. 435–495, Elsevier, 1975, URL http://dx.doi.org/10.1016/
B978-0-12-064850-4.50015-X.
[82] Kurganov, A.; Petrova, G.; Popov, B., “Adaptive Semidiscrete Central-Upwind Schemes for Nonconvex
Hyperbolic Conservation Laws,” SIAM Journal on Scientific Computing, vol. 29(6):pp. 2381–2401, 2007,
URL http://dx.doi.org/10.1137/040614189.
[83] Lele, S.K., “Compact finite difference schemes with spectral-like resolution,” Journal of Computational
Physics, vol. 103(1):pp. 16–42, 1992, URL http://dx.doi.org/10.1016/0021-9991(92)90324-R.
[84] Lentine, M.; Grétarsson, J.T.; Fedkiw, R., “An unconditionally stable fully conservative semi-Lagrangian
method,” Journal of Computational Physics, vol. 230(8):pp. 2857–2879, 2011, URL http://dx.doi.
org/10.1016/j.jcp.2010.12.036.
[85] Leslie, L.M.; Purser, R.J., “Three-Dimensional Mass-Conserving Semi-Lagrangian Scheme Employing
Forward Trajectories,” Monthly Weather Review, vol. 123(8):pp. 2551–2566, 1995, URL http://dx.
doi.org/10.1175/1520-0493(1995)123<2551:TDMCSL>2.0.CO;2.
[86] LeVeque, R.J., Finite Difference Methods for Ordinary and Partial Differential Equations, Other Titles in
114 References

Applied Mathematics, Society for Industrial and Applied Mathematics, 2007, URL http://dx.doi.
org/10.1137/1.9780898717839.
[87] Martínez, J.M., “Practical quasi-Newton methods for solving nonlinear systems,” Journal of Computa-
tional and Applied Mathematics, vol. 124(1-2):pp. 97–121, 2000, URL http://dx.doi.org/10.1016/
S0377-0427(00)00434-9.
[88] Monaghan, J.J., “Smoothed particle hydrodynamics,” Reports on Progress in Physics, vol. 68(8):pp. 1703–
1759, 2005, URL http://dx.doi.org/10.1088/0034-4885/68/8/R01.
[89] Morton, K.W.; Priestley, A.; Suli, E., “Stability of the Lagrange-Galerkin method with non-exact integra-
tion,” ESAIM: Mathematical Modelling and Numerical Analysis - Modélisation Mathématique et Analyse
Numérique, vol. 22(4):pp. 625–653, 1988, URL https://eudml.org/doc/193544.
[90] Nazarov, M.; Larcher, A., “Numerical investigation of a viscous regularization of the Euler equations by
entropy viscosity,” Computer Methods in Applied Mechanics and Engineering, vol. 317:pp. 128–152, 2017,
URL http://dx.doi.org/10.1016/j.cma.2016.12.010.
[91] Oliveira, A.; Baptista, A.M., “A comparison of integration and interpolation Eulerian-Lagrangian
methods,” International Journal for Numerical Methods in Fluids, vol. 21(3):pp. 183–204, 1995, URL
http://dx.doi.org/10.1002/fld.1650210302.
[92] Paolucci, S.; Zikoski, Z.J.; Wirasaet, D., “WAMR: An adaptive wavelet method for the simulation of
compressible reacting flow. Part I. Accuracy and efficiency of algorithm,” Journal of Computational
Physics, vol. 272:pp. 814–841, 2014, URL http://dx.doi.org/10.1016/j.jcp.2014.01.025.
[93] Peraire, J.; Persson, P.O., High-order discontinuous Galerkin methods for CFD, vol. 2, pp. 119–152,
WORLD SCIENTIFIC, 2011, URL http://dx.doi.org/10.1142/9789814313193_0005.
[94] Peshkov, I.; Romenski, E., “A hyperbolic model for viscous Newtonian flows,” Continuum Me-
chanics and Thermodynamics, vol. 28(1):pp. 85–104, 2016, URL http://dx.doi.org/10.1007/
s00161-014-0401-6.
[95] Pironneau, O., Finite element methods for fluids, Wiley, 1st ed., 1989.
[96] Prieto, J.L., “SLEIPNNIR: A multiscale, particle level set method for Newtonian and non-Newtonian
interface flows,” Computer Methods in Applied Mechanics and Engineering, vol. 307:pp. 164–192, 2016,
URL http://dx.doi.org/10.1016/j.cma.2016.04.019.
[97] Prieto, J.L.; Carpio, J., “A-SLEIPNNIR: A multiscale, anisotropic adaptive, particle level set framework
for moving interfaces. Transport equation applications,” Journal of Computational Physics, vol. 377:pp.
89–116, 2019, URL http://dx.doi.org/10.1016/j.jcp.2018.10.031.
[98] Proriol, J., “Sur une famille de polynomes á deux variables orthogonaux dans un triangle,” Comptus
Rendus de L’Academie des Ciences Paris, vol. 245(26):pp. 2459–2461, 1957.
[99] Rahunanthan, A.; Stanescu, D., “High-order W-methods,” Journal of Computational and Applied Mathe-
matics, vol. 233(8):pp. 1798–1811, 2010, URL http://dx.doi.org/10.1016/j.cam.2009.09.017.
[100] Rajaraman, P.; Vo, G.D.; Hansen, G.; Heys, J.J., “Comparison of continuous and discontinuous finite
element methods for parabolic differential equations employing implicit time integration,” International
Journal for Computational Methods in Engineering Science and Mechanics, vol. 18(2-3):pp. 182–190, 2017,
URL http://dx.doi.org/10.1080/15502287.2017.1310150.
[101] Rang, J., “Improved traditional Rosenbrock–Wanner methods for stiff ODEs and DAEs,” Journal of Com-
putational and Applied Mathematics, vol. 286:pp. 128–144, 2015, URL http://dx.doi.org/10.1016/
j.cam.2015.03.010.
[102] Reis, G.A.; Tasso, I.V.M.; Souza, L.F.; Cuminato, J.A., “A compact finite differences exact projection
method for the Navier–Stokes equations on a staggered grid with fourth-order spatial precision,” Com-
puters & Fluids, vol. 118:pp. 19–31, 2015, URL http://dx.doi.org/10.1016/j.compfluid.2015.
06.015.
[103] Rispoli, F.; Rafael Saavedra, G.Z., “A stabilized finite element method based on SGS models for com-
pressible flows,” Computer Methods in Applied Mechanics and Engineering, vol. 196(1):pp. 652–664, 2006,
URL http://dx.doi.org/10.1016/j.cma.2006.07.006.
[104] Roos, H.G.; Stynes, M.; Tobiska, L., Robust Numerical Methods for Singularly Perturbed Differential Equa-
tions, vol. 24 of Springer Series in Computational Mathematics, Springer, 2008, URL http://dx.doi.
org/10.1007/978-3-540-34467-4.
 References 115

[105] Samulyak, R.; Wang, X.; Chen, H.C., “Lagrangian particle method for compressible fluid dynamics,”
Journal of Computational Physics, vol. 362:pp. 1–19, 2018, URL http://dx.doi.org/10.1016/j.
jcp.2018.02.004.
[106] Saravanamuttoo, H.I.H.; Cohen, H.; Rogers, G.F.C., Gas turbine theory, Pearson Education,
2018, URL https://www.pearson.com/store/p/gas-turbine-theory/P100003100755/
9781292093093.
[107] Scovazzi, G., “Lagrangian shock hydrodynamics on tetrahedral meshes: A stable and accurate varia-
tional multiscale approach,” Journal of Computational Physics, vol. 231(24):pp. 8029–8069, 2012, URL
http://dx.doi.org/10.1016/j.jcp.2012.06.033.
[108] Scovazzi, G.; Christon, M.; Hughes, T.; Shadid, J., “Stabilized shock hydrodynamics: I. A Lagrangian
method,” Computer Methods in Applied Mechanics and Engineering, vol. 196(4-6):pp. 923–966, 2007, URL
http://dx.doi.org/10.1016/j.cma.2006.08.008.
[109] Scovazzi, G.; Love, E.; Shashkov, M., “Multi-scale Lagrangian shock hydrodynamics on Q1/P0 finite
elements: Theoretical framework and two-dimensional computations,” Computer Methods in Applied
Mechanics and Engineering, vol. 197(9-12):pp. 1056–1079, 2008, URL http://dx.doi.org/10.1016/
j.cma.2007.10.002.
[110] Scovazzi, G.; Shadid, J.; Love, E.; Rider, W., “A conservative nodal variational multiscale method for La-
grangian shock hydrodynamics,” Computer Methods in Applied Mechanics and Engineering, vol. 199(49-
52):pp. 3059–3100, 2010, URL http://dx.doi.org/10.1016/j.cma.2010.03.027.
[111] Sevilla, R.; Huerta, A., “Tutorial on Hybridizable Discontinuous Galerkin (HDG) for Second-Order El-
liptic Problems,” in J. Schröder; P. Wriggers (Eds.), Advanced Finite Element Technologies, CISM Inter-
national Centre for Mechanical Sciences, pp. 105–129, Springer International Publishing, Cham, 2016,
URL http://dx.doi.org/10.1007/978-3-319-31925-4_5.
[112] Shu, C.W., “Discontinuous Galerkin Method for Time-Dependent Problems: Survey and Recent Devel-
opments,” in X. Feng; O. Karakashian; Y. Xing (Eds.), Recent Developments in Discontinuous Galerkin
Finite Element Methods for Partial Differential Equations, vol. 157, pp. 25–62, Springer International
Publishing, Cham, 2014, URL http://dx.doi.org/10.1007/978-3-319-01818-8_2.
[113] Tabata, M.; Uchiumi, S., “A genuinely stable Lagrange–Galerkin scheme for convection-diffusion
problems,” Japan Journal of Industrial and Applied Mathematics, vol. 33(1):pp. 121–143, 2016, URL
http://dx.doi.org/10.1007/s13160-015-0196-2.
[114] Toro, E.F., Riemann Solvers and Numerical Methods for Fluid Dynamics: A Practical Introduction,
Springer-Verlag, Berlin Heidelberg, third ed., 2009, URL http://dx.doi.org/10.1007/b79761.
[115] Tranquilli, P.; Sandu, A.; Zhang, H., “LIRK-W: Linearly-implicit Runge-Kutta methods with approx-
imate matrix factorization,” arXiv:1611.07013 [math], 2016, URL http://arxiv.org/abs/1611.
07013.
[116] Vázquez-Cendón, M.E., Solving Hyperbolic Equations with Finite Volume Methods, vol. 90 of
UNITEXT, Springer International Publishing, Cham, 2015, URL http://dx.doi.org/10.1007/
978-3-319-14784-0.
[117] Walker, H.F.; Ni, P., “Anderson Acceleration for Fixed-Point Iterations,” SIAM Journal on Numerical
Analysis, vol. 49(4):pp. 1715–1735, 2011, URL http://dx.doi.org/10.1137/10078356X.
[118] Wang, H.; Dahle., H.K.; Ewing, R.E.; Espedal, M.S.; Sharpley, R.C.; Man, S., “An ELLAM Scheme
for Advection-Diffusion Equations in Two Dimensions,” SIAM Journal on Scientific Computing,
vol. 20(6):pp. 2160–2194, 1999, URL http://dx.doi.org/10.1137/S1064827596309396.
[119] White, F.M., Fluid mechanics, McGraw-Hill Education, seventh ed., 2011.
[120] Zarin, H., “Continuous–discontinuous finite element method for convection-diffusion problems with
characteristic layers,” Journal of Computational and Applied Mathematics, vol. 231(2):pp. 626–636, 2009,
URL http://dx.doi.org/10.1016/j.cam.2009.04.010.
[121] Zeng, X.; Scovazzi, G., “A frame-invariant vector limiter for flux corrected nodal remap in arbitrary
Lagrangian–Eulerian flow computations,” Journal of Computational Physics, vol. 270:pp. 753–783, 2014,
URL http://dx.doi.org/10.1016/j.jcp.2014.03.054.
116 References

[122] Zeng, X.; Scovazzi, G., “A variational multiscale finite element method for monolithic ALE computa-
tions of shock hydrodynamics using nodal elements,” Journal of Computational Physics, vol. 315:pp.
577–608, 2016, URL http://dx.doi.org/10.1016/j.jcp.2016.03.052.
[123] Zienkiewicz, O.C.; Taylor, R.L.; Nithiarasu, P., “The Finite Element Method for Fluid Dynamics,” in The
Finite Element Method for Fluid Dynamics (Seventh Edition), p. i, Butterworth-Heinemann, Oxford, 2014,
URL http://dx.doi.org/10.1016/B978-1-85617-635-4.00016-9.