Parameter Id HB

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Parameter identification in the Herschel-Bullkey model∗

Sergio González-Andrade1 and Alejandro Quiroga1


1
Research Center on Mathematical Modeling (MODEMAT) and Departamento de Matemática - Escuela
Politécnica Nacional, Quito 170413, Ecuador.
{sergio.gonzalez,alejandro.quiroga}@epn.edu.ec

November 16, 2023

Abstract
In this paper we propose a

Keywords: Variational inequalities of the second kind, viscoplastic fluids.


AMS Subject Classification: .

1 Introduction

2 Problem statement
Let Ω ⊂ Rd , d = 2, 3, be a bounded domain with Lipschitz boundary ∂Ω. We are concerned with
the problem of identifying parameters µ ∈ Uad and g ∈ Gad in the following variational inequality
of the second kind
Z Z Z
µ(x)|∇u|p−2 ∇u · ∇(v − u) dx + g(x) (|∇v| − |∇u|) dx ≥ f (v − u) dx, ∀v ∈ W01,p (Ω). (1)
Ω Ω Ω
0
Here, 1 < p < ∞ and f ∈ Lp (Ω), while Uad and Gad stand for suitable admissible sets for µ and g,
respectively.
The scalar variational inequality (VI) encapsulates the variational formulation of the Herschel-
Bulkley model, designed for a specific class of viscoplastic materials characterized by the fluid index
given by p. In this context, our primary objective is the determination of the viscosity parameter
(µ) and the yield stress parameter (g) governing the behavior of the fluid.
By following the methodology outlined in [4], we undertake the analysis of this problem through
a discretize then optimize approach. This strategic choice is made to capitalize on the inherent
regularity exhibited in the solution operator as detailed in the aforementioned work.

Supported in part by the Escuela Politécnica Nacional del Ecuador, under the project PIM 20-01.

1
3 Discretize then optimize approach
In this section, we analyze the first order finite element discretization of (1). In this aim, let T h
be a regular triangulation of Ω in the sense of Ciarlet (see [3, p. 38]). Next, let Ωh be a polygonal
approximation of Ω, given by [
Ωh = T,
T ∈T h
where all the open triangles T are disjoint one to one and whose diameter is bounded by h. Further,
for any pair of triangles, their closures are either disjoint or have a common vertex or a common
edge. Finally, let {Pj }N h
j=1 be the vertices (nodes) associated to the triangulation T . Hereafter, we
assume that Pj ∈ ∂Ωh implies that Pj ∈ ∂Ω and that Ωh ⊂ Ω.
Taking all this into account, we introduce the following finite-dimensional space associated to
the triangulation T h

W0h := {v ∈ C(Ωh ) : v|τ ∈ P1 , ∀τ ∈ T h and v = 0 on ∂Ωh },

where P1 is the space of polynomials with degree less than or equal to 1.


Together with these considerations, it is possible to define the following finite element approxi-
mation of problem (1):
Z Z Z
p−2
µh (x)|∇uh | ∇uh ·∇(vh −uh ) dx+ gh (x) (|∇vh | − |∇uh |) dx ≥ fh (vh −uh ) dx, ∀vh ∈ W0h .
Ω Ω Ω
(2)
Consequently, we have the following proposition.
Proposition 3.1. Problem (2) has a unique solution ūh ∈ W0h .
Proof. The result is a direct consequence of the fact that W0h is a closed subspace of W01,p (Ω).

PnWe now discuss in detail this discrete version of the VI. In this aim, we hhave that uh (x) :=
Pn n ≥ 1 and {ϕ1 (x), . . . , ϕn (x)} is a base for the space W0 . Further, we have
i=1 ui ϕi (x), where
that ∇uh (x) = i=1 ui ∇ϕi (x). Thus, if m > 1 is the cardinality of Th , i.e., the number of elements
in the triangulation, we use the following discrete version of the gradient
!
∂ h
1
∇h := ∈ R2m×n , (3)
∂2h
∂ϕ (x) ∂ϕ (x)
where ∂1h := ∂xj 1 |τk y ∂2h := ∂xj 2 |τk , for j = 1, . . . , n and Tk ∈ T h . Finally, let us define the
function ηh : R2m → Rm given by

η(w)k = |(wk , wk+m )|T , k = 1, . . . , m.

Therefore, given u ∈ Rn , this function will allow us to approximate the value of the norm of the
vector ∇h u at each triangle T ∈ Th .
Next, let us discuss the approximation for the parameters. Since this functions are supposed
to be bounded in the geometry, we propose to approximate them as piecewise constant at each
Tk ∈ T h , i.e., µk ≈ µh (x)|Tk and gk ≈ gh (x)|Tk , for all k = 1, . . . , m.
Taking into account the discussion above, we have the following discretization for the nonsmooth
term in (2). R Pm R
Ω gh (x)|∇uh (x)| dx ≈ k=1 Tk gh (x)|∇uh (x)| dx
Pm (4)
h u) .

≈ k=1 |Tk |gk η h (∇ k

2
Now, let us focus on the p-Laplacian term in (2). In this case, by taking vh = ϕj , j = 1, . . . , n,
we have that
R p−2 ∇u · ∇ϕ dx ≈
Pm R h p−2 ( m u ∇ϕ ) · ∇ϕ dx
P
Ω µh (x)|∇uh | h j k=1 Tk µh (x)(ηh (∇ u)k ) i=1 i i j
Pm Pm R h p−2 (∇ϕ · ∇ϕ ) .
≈ i=1 ui k=1 µk Tk (ηh (∇ u)k ) i j

We now define the operator Ahu (·, ·) : Rm × Rn × Rn → Rn×n as follows


Pn Pm
hAhu (µ, u), vi := h p−2 (∇ϕ · ∇ϕ ) dx
R
i,j=1 ui vj k=1 µk Tk (ηh (∇ u)k ) i j
Pm R h p−2 ( m u ∇ϕ ) · ( m v ∇ϕ ) dx
P P
= k=1 µk Tk (ηh (∇ u)k ) i=1 i i j=1 j j (5)
Pm R h p−2 (∇h u · ∇h v) dx, ∀v ∈ Rn .
= k=1 µk Tk (ηh (∇ u)k )

Finally, by considering (4) and (5), we have the following finite dimensional variational inequality
of the second kind
m
X m
X
hAhu (µ, u), v − ui + |Tk |gk (ηh (∇h v))k − |Tk |gk (ηh (∇h u))k ≥ hf, v − ui, ∀v ∈ Rn (6)
k=1 k=1

4 Gradient-type algorithms

5 Numerical results

6 Conclusions and outlook

References
[1] R. A. Adams and S.J.F. Fournier, Sobolev Spaces, Academic Press, Elsevier, 4 ed. (2003)

[2] E. Casas, P. I. Kogut and G. Leugering, Approximation of optimal control problems in


the coefficient for the p-Laplace equation. I. convergence result. SIAM J. Control Optim., 54
(2016) 1406 - 1422.

[3] P. G. Ciarlet. The Finite Element Method for Elliptic Problems. Society for Industrial and
Applied Mathematics, Jan. 2002.

[4] C. Christof, Gradient-based solution algorithms for a class of bilevel optimization and opti-
mal control problems with a nonsmooth lower level. SIAM J. Optim., 30 (2020) 290-318.

[5] J. Gwinner, An optimization approach to parameter identification in variational inequalities


of second kind. Optim. Letters, 12 (2018) 1141-1154.

[6] M. Hintermüller, Inverse coefficient problems for variational inequalities: optimality con-
ditions and numerical realization. ESAIM: M2AN, 35 (2001) 129-152.

You might also like