View
6
Download
0
Category
Preview:
Citation preview
Mediterr. J. Math. (2020) 17:142https://doi.org/10.1007/s00009-020-01579-31660-5446/20/050001-19published onlineAugust 20, 2020c© The Author(s) 2020
On Some Numerical Integration Formulason the d-Dimensional Simplex
Filomena Di Tommaso and Benaissa Zerroudi
Abstract. In this paper, we consider the problem of the approximationof the integral of a function f over a d-dimensional simplex S of Rd bysome quadrature formulas which use only the functional and derivativevalues of f on the boundary of the simplex S or function data at thevertices of S, at points on its facets and at its center of gravity. Thequadrature formulas are computed by integrating over S a polynomialapproximant of f which uses functional and derivative values at thevertices of S.
Mathematics Subject Classification. Primary 65D32, Secondary 65D30.
Keywords. Numerical integration, Cubature, Simplex.
1. Introduction
The problem of the determination of quadrature rules for triangles, tetra-hedra and, in general, for d-dimensional simplicial domains has reached theattention of a number of scholars starting from the middle of the nineteenthcentury up today (see [13] and the references therein). Although many papersfocus on quadrature rules for triangles [3,10,12,16], only a limited literatureis available on the integration in three or higher dimensions [4,14,17]. In thispaper, we approach the problem of integration over general d -dimensionalsimplices by special type integration formulas which use functional and de-rivative values of the integrand function f mainly on points on the boundaryof the d-dimensional simplex S. When the nodes lie only on the boundary ofS these formulas are called boundary type quadrature formulas and are usedwhen the values of f and its derivatives inside the simplex are not given orare not easily determinable. Applications of these formulas can be realized
This research has been accomplished within the RITA “Research ITalian network on Ap-proximation”. This research was supported by GNCS–INdAM 2019 project. The authorswould like to thank Professor F. Dell’Accio, Professor O. Nouisser and the anonymousreferee for their valuable comments.
142 Page 2 of 19 F. Di Tommaso, B. Zerroudi MJOM
in the framework of the numerical solution of boundary value problems ofpartial differential equations (see [9] and the references therein).
To reach our goal, we follow the approach proposed in Refs. [1,2]. Moreprecisely, we approximate the integrand function f with a polynomial inter-polant LS
r [f ](x) which uses functional and derivative data values up to afixed order r ∈ N at the vertices of S, i.e.
f(x) = LSr [f ](x) + RS
r [f ](x), x ∈ S, (1.1)
and then, we integrate both sides of (1.1) over the d-dimensional simplex Sto get the quadrature formula∫
S
f(x)dx = QSr [f ] + ES
r [f ] (1.2)
where
QSr [f ] =
∫S
LSr [f ](x)dx and ES
r [f ] =∫
S
RSr [f ](x)dx.
The obtained quadrature formula (1.2) uses function and derivative dataup to the order r at the vertices of S and has degree of exactness 1 + r, i.e.ES
r [f ] = 0 whenever f is a polynomial in d variables of total degree at most1 + r. The main feature of the quadrature formula QS
r [f ] is that it reliesonly on function and derivative data up to the order r at the vertices of S.This motivates us to look for approximations of those derivatives to obtainquadrature formulas which do not use any derivative data. To this end, werestrict to the case r = 1 and, by following the technique described in Ref.[6], we approximate the first order derivative data by three-point finite dif-ferences approximation. According to the choice of the approximation of thederivative data, we get different quadrature formulas with degree of exact-ness 2 and, to increase the degree of exactness of such formulas, we considerthe convex combination of two of them to get a quadrature formula with de-gree of exactness 3 (see Sect. 3). Finally, we restrict to the two-dimensionalcase (see Sect. 4) and we provide numerical results to test the approximationaccuracies of the proposed formulas (see Sect. 5).
2. A Quadrature Formula on the d-Dimensional Simplex
2.1. Preliminaries and Notations
Let S ⊂ Rd be a not degenerated d-dimensional simplex with vertices v0, . . . ,
vd ∈ Rd and
A(S) =
∣∣∣∣∣∣∣∣∣
1 v0
1 v1
1...
1 vd
∣∣∣∣∣∣∣∣∣the signed volume of the hyperparallelepiped with vertices v0, . . . ,vd. For apoint x ∈ S and for each l = 0, 1, . . . , d we denote by Sl(x) the d-dimensional
MJOM On Some Numerical Integration Formulas on the. . . Page 3 of 19 142
simplex of vertices v0, . . . ,vl−1,x,vl+1, . . . ,vd. The barycentric coordinatesof x with respect to the simplex S are then defined by
λl(x) =A(Sl)A(S)
, l = 0, 1, . . . , d. (2.1)
For each α = (α1, α2, . . . , αd) ∈ Nd and x = (x1, . . . , xd) ∈ R
d, as usual, wedenote by
|α| = α1 + α2 + · · · + αd, α! = α1! . . . αd!, xα = xα11 xα2
2 . . . xαd
d .
We also set λ = (λ0, λ1, λ2, . . . , λd) and λl = (λ0, . . . , λl−1, λl+1, . . . , λd) foreach l = 0, . . . , d. Moreover, we denote by Dαf = ∂|α|f
∂xα11 ∂x
α22 ...∂x
αdd
and by
Dkx−vf =
∑|α|=k
α∈Nd
k!α!
(x − v)αDαf (2.2)
the k-th order directional derivative of f along the line segment between xand v. Finally, we use the notations
Dijf = (vi − vj) · ∇f, i, j = 0, 1, . . . , d (2.3)
for the derivative of f along the directed line segment from xj to xi (as usual,· denotes the dot product) and
Dαl = Dα1
0,lDα21,l . . . Dαl
l−1,lDαl+1l+1,l . . . D
αd
d,l , l = 0, 1, . . . , d, (2.4)
for the composition of derivatives along the directed sides of the simplex.Under these assumptions, we get the following result.
Lemma 2.1. Let f ∈ Cr(S), then
Dkx−v l
f(vl) =∑
|α|=k
α∈Nd
k!α!
Dαl f(vl)λα
l (x), l = 0, 1, . . . , d, (2.5)
for any k ∈ N, k ≤ r and x ∈ Rd.
Proof. Due to the properties satisfied by the barycentric coordinates, for anyl = 0, 1, . . . , d, we have
x − vl = vlλl(x) − vl +d∑
j=0j �=l
vjλj(x)
= vl (λl(x) − 1) +d∑
j=0j �=l
vjλj(x)
= −vl
⎛⎜⎜⎝
d∑j=0j �=l
λj(x)
⎞⎟⎟⎠ +
d∑j=0j �=l
vjλj(x)
142 Page 4 of 19 F. Di Tommaso, B. Zerroudi MJOM
=d∑
j=0j �=l
(vj − vl)λj(x)
= (v0 − vl, . . . ,vl−1 − vl,vl+1 − vl, . . . ,vd − vl) · λl(x). (2.6)
By substituting (2.6) in (2.2) and by definition (2.4), we have
Dkx −vl
f(vl) =∑
|α|=k
α∈Nd
k!
α!((v0 − vl, . . . , v l−1 − v l, v l+1 − v l, . . . , vd − vl) · λl(x))
αD
αf(vl)
=∑
|α|=k
α∈Nd
k!
α!((v1 − vl, . . . , vl−1 − v l, v l+1 − v l, . . . , vd − vl) Df(vl))
αλl(x)
α
=∑
|α|=k
α∈Nd
k!
α!D
αl f(v l)
αλl(x)
α.
�
Proposition 2.2. Let S ⊂ Rd be a not degenerated d-dimensional simplex with
vertices v0, . . . ,vd then∫
S
λα(x)dx =A(S)α!
(d + |α|)! , α ∈ Nd+1. (2.7)
Proof. See [15, Theorem 2.2]. �
Proposition 2.3. Let S ⊂ Rd be a not degenerated d-dimensional simplex with
vertices v0,v1, . . . ,vd. For any x ∈ S and r ∈ N, we have
∫S
d∑i=0
‖x − vi‖r+2λi(x)dx ≤ A(S)(d + 2)!
d∑i=0
d∑j=0
‖vi − vj‖r+2. (2.8)
Proof. By the equality (2.6) and by recalling that 0 ≤ λj(x) ≤ 1 for eachx ∈ S, we easily obtain
‖x − vl‖r+2 ≤⎛⎝ d∑
j=0
‖vj − vl‖λj(x)
⎞⎠
r+2
≤d∑
j=0
‖vj − vl‖r+2λj(x)
for any l = 0, . . . , d. Consequently,
∫S
d∑i=0
‖x − vi‖r+2λi(x)dx ≤∫
S
d∑i=0
d∑j=0
‖vj − vi‖r+2λj(x)λi(x)dx
≤d∑
i=0
d∑j=0
‖vj − vi‖r+2
∫S
λj(x)λi(x)dx
and, by Proposition 2.2, we easily get the inequality (2.8). �
MJOM On Some Numerical Integration Formulas on the. . . Page 5 of 19 142
2.2. Construction of the Quadrature Formula
The multivariate Lagrange interpolation polynomial on the simplex S inbarycentric coordinates is
LS [f ](x) =d∑
l=0
λl(x)f(vl). (2.9)
The operator LS reproduces polynomials up to the degree 1 and interpolatesthe values of f at the vertices vl of the simplex S. If the function f belongsto Cr(S), we can replace the values f(vl) by the modified Taylor polynomialof degree r at v� proposed in [6], the resulting polynomial operator is
LSr [f ](x) =
d∑l=0
(r∑
k=0
ark
k!Dk
x−v lf(vl)
)λl(x), x ∈ R
d, (2.10)
where ark = (1+r−k)!r!(1+r)!(r−k)! . As specified in [6], the operator LS
r [f ](x) reproducespolynomials up to the degree max {1, 1 + r} = 1+r. Moreover, for each x ∈ Sand f ∈ Cr+2 (S), its remainder term RS
r [f ](x) = f(x) − LSr [f ](x) can be
explicitly represented as
RSr [f ](x) =
d∑l=0
λt(x)∫ 1
0
−t(1 − t)r
(1 + r)!Dr+2
x−v lf (vl + t (x − vl)) dt. (2.11)
Remark 2.4. Since S is a compact convex domain and LS is a linear boundedoperator, in line with [8], LS
r can be interpreted as
LSr [f ](x) = LS
[r∑
k=0
ark
k!Dk
x−·f
](x)
and, from [8, Proposition 3.4], it follows that LSr inherits the interpolation
properties of the Lagrange operator (2.9).
To obtain the desired quadrature formula, we rearrange polynomial(2.10) by taking into account Lemma 2.1. More precisely,
LSr [f ](x) =
d∑l=0
⎛⎜⎜⎝
r∑k=0
ark
k!
∑|α|=k
α∈Nd
k!α!
Dαl f(vl)λα
l (x)
⎞⎟⎟⎠ λl(x)
=d∑
l=0
⎛⎜⎜⎝
r∑k=0
(1 + r − k)!r!(1 + r)!(r − k)!
∑|α|=k
α∈Nd
1α!
Dαl f(vl)λα
l (x)
⎞⎟⎟⎠ λl(x)
=d∑
l=0
⎛⎜⎜⎝
r∑k=0
1 + r − k
1 + r
∑|α|=k
α∈Nd
1α!
Dαl f(vl)λα
l (x)
⎞⎟⎟⎠ λl(x) (2.12)
142 Page 6 of 19 F. Di Tommaso, B. Zerroudi MJOM
and, by the change of dummy index, we get
LSr [f ](x) =
r∑k=0
1 + r − k
1 + r
⎛⎜⎜⎝
d∑l=0
∑|α|=k
α∈Nd
1α!
Dαl f(vl)λα
l (x)λl(x)
⎞⎟⎟⎠ . (2.13)
The quadrature formula is then computed by integrating the right hand sideof (2.13) on the simplex S.
Theorem 2.5. Let f ∈ Cr+2(S). Then∫S
f(x)dx = QSr [f ] + ES
r [f ],
where
QSr [f ] =
A(S)(1 + r)
r∑k=0
1 + r − k
(d + 1 + k)!
⎛⎜⎜⎝
d∑l=0
∑|α|=k
α∈Nd
Dαl f(vl)
⎞⎟⎟⎠ (2.14)
and
ESr [f ] =
∫S
RSd [f ](x)
=∫
S
d∑l=0
(∫ 1
0
−t(1 − t)r
(1 + r)!Dr+2
x−v lf (vl + t (x − vl)) dt
)λl(x)dx.
(2.15)
Moreover, the quadrature formula QSr [f ] has degree of exactness 1 + r .
Proof. By integrating the right hand side of equality (2.13), we get
QSr [f ](x) =
∫S
LSd [f ](x)dx
=r∑
k=0
1 + r − k
(1 + r)
⎛⎜⎜⎝
d∑l=0
∑|α|=k
α∈Nd
1α!
Dαl f(vl)
∫S
λαl (x)λl(x)dx
⎞⎟⎟⎠ .
(2.16)
By Proposition (2.2)∫S
λαl (x)λl(x)dx =
A(S)α!(d + 1 + |α|)!
and then (2.16) becomes
QSr [f ] =
r∑k=0
1 + r − k
(1 + r)
⎛⎜⎜⎝
d∑l=0
∑|α|=k
α∈Nd
1α!
Dαl f(vl)
A(S)α!(d + 1 + |α|)!
⎞⎟⎟⎠ . (2.17)
MJOM On Some Numerical Integration Formulas on the. . . Page 7 of 19 142
The expression of ESr [f ] is obtained by integrating on the simplex S the
remainder term RSr [f ](x) in formula (2.11). Since RS
r [f ](x) vanishes wheneverf is a polynomial in d variables of total degree at most 1 + r, ES
r [f ] inheritsthis property and the quadrature formula has degree of exactness 1 + r. �2.3. Error Bounds
To give a bound for the remainder term ESr [f ] of the quadrature formula in
Theorem 2.5, we need some additional notations. More precisely, for a k-timescontinuous differentiable function f : S → R, we introduce the norm∣∣Dkf
∣∣S
:= supx∈S
sup{∣∣Dkyf(x)
∣∣ : y ∈ Rd, ‖y‖ = 1}. (2.18)
where ‖ · ‖ denotes the Euclidean norm in Rd, and y is assumed to be a
column vector. Consequently, for any x ∈ S and y ∈ Rd, we have
|Dkyf(x)| ≤ |Dkf |S · ‖y‖r+2. (2.19)
Proposition 2.6. Let S ⊂ Rd be a not degenerated d-dimensional simplex with
vertices v0, . . . ,vd and f ∈ Cr+2(S). Then
∣∣ESr [f ]
∣∣ ≤∣∣Dr+2f
∣∣S
(r + 2)!(r + 1)A(S)
(d + 2)!
d∑l=0
d∑j=0
‖vl − vj‖r+2. (2.20)
Proof. By taking the modulus of both sides of equality (2.15), by applyingthe triangular inequality and by bounding the directional derivative of f oforder r + 2 by (2.19), we have
∣∣ESr [f ]
∣∣ ≤∣∣∣∣∣∫
S
d∑l=0
(∫ 1
0
−t(1 − t)r
(1 + r)!Dr+2
x−v lf (vl + t (x − vl)) dt
)λl(x)dx
∣∣∣∣∣≤ ∣∣Dr+2f
∣∣S
∫S
d∑l=0
(∫ 1
0
t(1 − t)r
(1 + r)!‖x − vl‖r+2dt
)λl(x)dx
≤∣∣Dr+2f
∣∣S
(1 + r)!
∫S
d∑l=0
‖x − vl‖r+2
(∫ 1
0
t(1 − t)rdt
)λl(x)dx.
Using the inequality in Proposition (2.3), and by the fact that∫ 1
0
t(1 − t)rdt =1
(r + 2)(r + 1),
we have∣∣ES
r [f ]∣∣ ≤
∣∣Dr+2f∣∣S
(1 + r)!(r + 1)(r + 2)A(S)
(d + 2)!
d∑l=0
d∑j=0
‖vl − vj‖r+2
and then (2.20). �Remark 2.7. It is worth noting that Theorem 2.5 gives a quadrature formulaobtained by integrating both sides of the expression in [6, Theorem 1] andthe bound in Proposition 2.6 is nothing but the integral of the bound givenin [6, Theorem 2], where Ω = S, m = 1 and φi (x) = λi(x). Consequently,the bound (2.20) is the best possible estimation for each r ∈ N0.
142 Page 8 of 19 F. Di Tommaso, B. Zerroudi MJOM
3. Integration Formulas on the Simplex with Only FunctionData
The main feature of the quadrature formula (2.14) is that it uses only deriva-tives of f along the edges of S; this motivates to consider approximations ofthose derivatives to obtain quadrature formulas which use the function dataat the vertices of the simplex S, at points on its facets or at its center ofgravity. To this aim, we focus on the case r = 1 and we consider differentkinds of approximation of the derivatives in (2.14). For r = 1, the quadratureformula (2.14) in Theorem 2.5 becomes
QS1 [f ] =
A(S)(d + 1)!
d∑l=0
f(vl) +A(S)
2(d + 2)!
d∑l=0
∑|α|=1
α∈Nd
Dαl f(vl). (3.1)
Proposition 3.1. Let f : S → R be a 3-times continuous differentiable func-tion on S, then
QS1 [f ] =
A(S)(d + 2)!
((3d + 2)
d∑l=0
f(vl) − 4d−1∑l=0
d∑r=l+1
f
(vr + vl
2
))
+A(S)
2(d + 2)!
d−1∑l=0
d∑r=l+1
εl,r. (3.2)
where
εl,r[f ] = D3vl−vr
f(vr + ξ1(vl − vr))
−D3vl−vr
f(vr + ξ2(vl − vr)), ξ1, ξ2 ∈ [0, 1].
Moreover, the quadrature formula (3.2) has degree of exactness 2.
Proof. By definition (2.4), the sum of first-order derivatives in the secondterm of QS
1 [f ] can be rewritten as
d∑l=0
∑|α|=1
α∈Nd
Dαl f(vl) =
d∑l=0
d∑r=0r �=l
Drlf(vl) =d−1∑l=0
d∑r=l+1
(Dlrf(vr) − Dlrf(vl)) ,
(3.3)
where the differences of directional derivatives along the edges of the simplexS can be replaced by a three-point finite difference approximation. To dothis, let us recall that for a univariate function g, it is possible to considerthe derivation formula
g′(a − h) =1h
(−1
2g(a + h) + 2g(a) − 3
2g(a − h)
)+
h2
3g′′′(ξ) (3.4)
for some ξ ∈ [a − h, a + h]. Using this formula with h = ±1/2 and a = 1/2we get a three-point approximation for g′(0) − g′(1) with a remainder term
MJOM On Some Numerical Integration Formulas on the. . . Page 9 of 19 142
which is expressed in terms of the modulus of continuity of g′′′ [6, Section5.1]. By applying the formula (3.4) along the edges of S we get
Dlrf(vr) − Dlrf(vl) = 4(
f(vr) − 2f
(vr + vl
2
)+ f(vl)
)+ εl,r (3.5)
with
|εl,r| ≤ 112
ω(D3
l f ((1 − t)vl + tvr) , 1),
where ω denotes the modulus of continuity with respect to t ∈ [0, 1]. Bysubstituting expression (3.5) in (3.3) and by rearranging, we get
d∑l=0
∑|α|=1
α∈Nd
Dαl f(vl) = 4d
d∑l=0
f(vl) − 8d−1∑l=0
d∑r=l+1
f
(vr + vl
2
)+
d−1∑l=0
d∑r=l+1
εl,r.
(3.6)
Finally, by substituting (3.6) in (3.1), we get
QS1 [f ] =
A(S)(d + 2)!
((3d + 2)
d∑l=0
f(vl) − 4d−1∑l=0
d∑r=l+1
f
(vr + vl
2
))
+A(S)
2(d + 2)!
d−1∑l=0
d∑r=l+1
εl,r.
εl,r[f ] = 0 whenever f is a polynomial in d variables of total degree 2 andthis implies that QS
1 [f ] has degree of exactness 2. �
Proposition 3.2. Let S ⊂ Rd be a not degenerated d-dimensional simplex with
vertices (vl)l=0,1,...,d. Let us denote by sd−1l , l = 0, 1, . . . , d the facet opposite
to the vertex vl and by gl the barycenter of sd−1l . For all α ∈ (0, 1) we have
∫S
f(x)dx =A(S)
2(d + 2)!
(α(d + 4) − d
α
d∑l=0
f(vl) +d
α − α2
d∑l=0
f(yl(α))
+αd
α − 1
d∑l=0
f(gl)
)+ R(d, α)[f ],
(3.7)
with yl(α) = vl + α(gl − vl) and
R(d, α)[f ] =dA(S)
4(d + 2)!(α − α2)
d∑l=0
(α2
∫ 1
0
(1 − t)2D3g l−v l
f(v l + t(g l − v l))dt
−∫ α
0
(α − t)2D3g l−v l
f(v l + t(g l − v l))dt
). (3.8)
The quadrature formula (3.7) has degree of exactness 2.
142 Page 10 of 19 F. Di Tommaso, B. Zerroudi MJOM
Proof. Let gl be the barycenter of sd−1l , by Lemma (2.1) the sum of first-order
derivatives along the edges of S in (3.1) can be rewritten asd∑
l=0
∑|α|=1
α∈Nd
Dαl f(vl)λα(gl) =
d∑l=0
Dg l−v lf(vl)
and, since λk(gl) = 1d , for each l, k = 0, . . . , d, then
d∑l=0
∑|α|=1
α∈Nd
Dαl f(vl) = d
d∑l=0
Dg l−v lf(vl). (3.9)
By substituting (3.9) in (3.1), we get
QS1 [f ] =
A(S)(d + 1)!
d∑l=0
f(vl) +dA(S)
2(d + 2)!
d∑l=0
Dg l−v lf(vl). (3.10)
To have a three point finite difference approximation of the directional deriva-tives in (3.10), for each l = 0, . . . , d, let us introduce the univariate function
hl :{
[0, 1] → R,t �→ hl(t) = f(vl + t(gl − vl)).
(3.11)
For t = 1 and t = α ∈ (0, 1), the second-order Taylor expansion of hl(1) andhl(α) centered at 0 with integral remainder are
hl(1) = hl(0) + h′l(0) +
12h
′′l (0) +
12
∫ 1
0
(1 − t)2h′′′l (t)dt (3.12)
and
hl(α) = hl(0) + αh′l(0) +
12α2h
′′l (0) +
12
∫ α
0
(α − t)2h′′′l (t)dt. (3.13)
Then, by (3.12) and (3.13), we get
hl(α) − α2hl(1) =(1 − α2
)hl(0) + (α − α2)h′
l(0) + Rl[f ](α) (3.14)
where
Rl[f ](α) =12
∫ α
0
(α − t)2h′′′l (t)dt − α2 1
2
∫ 1
0
(1 − t)2h′′′l (t)dt.
Therefore, by (3.14) it follows that
h′l(0) =
1α − α2
hl(α) − α
1 − αhl(1) − 1 + α
αhl(0) +
1α − α2
Rl(α)[f ].
(3.15)
By rewriting equality (3.15) in terms of f we get
Dg l−v lf(vl) =
1α − α2
f (αgl + (1 − α)vl) − α
1 − αf(gl) − 1 + α
αf(vl)
+1
α − α2Rl(α)[f ], (3.16)
MJOM On Some Numerical Integration Formulas on the. . . Page 11 of 19 142
where
Rl(α)[f ] =12
(α2
∫ 1
0
(1 − t)2D3g l−v l
f(vl + t(gl − vl))dt
−∫ α
0
(α − t)2D3g l−v l
f(vl + t(gl − vl))dt
). (3.17)
Finally, by substituting (3.16) in (3.1), we get (3.7). R(α)[f ] = 0 whenever fis a polynomial in d variables of total degree 2 and, therefore, the quadratureformula (3.7) has degree of exactness 2. �
Remark 3.3. 1. For α =d
d + 4the formula (3.7) becomes
∫S
f(x)dx =A(S)
2(d + 2)!
((d + 4)2
4
d∑l=0
f
(dgl + 4vl
d + 4
)− d2
4
d∑l=0
f(gl)
)
+R
(d,
d
d + 4
)[f ], (3.18)
that is a quadrature formula which uses only the function data at thepoints gl and dg l+4v l
d+4 , l = 0, . . . , d, and is exact for all polynomial ofdegree less than or equal to 2.
2. For α = dd+1 , the quadrature formula (3.7) becomes,
∫S
f(x)dx =A(S)
2(d + 2)!
(3
d∑l=0
f (vl) + (d + 1)3f(x∗) − d2d∑
l=0
f(gl)
)
+R
(d,
d
d + 1
)[f ] (3.19)
where x∗ is the center of gravity of S.
To improve the approximation accuracy of the quadrature formula (3.19),let us consider a convex combination of this formula with a multivariate Simp-son rule for a simplex proposed in [7, Theorem 5.1]. For a particular valueof the parameter of the linear combination, we are able to get a quadratureformula with an higher degree of exactness.
Corollary 3.4. Let f : S → R be a 3-times continuous differentiable functionon S. Let us denote by x∗ the center of gravity of S, by sd−1
l , l = 0, 1, . . . , d
the facets of S opposite to the vertex vl and by gl the barycenter of sd−1l .
Then, ∫S
f(x)dx = τF1[f ] + (1 − τ)F2[f ] + R(τ)[f ], τ ∈ R, (3.20)
where
F1[f ] =A(S)
d!
(d + 1d + 2
f(x∗) +1
(d + 1)(d + 2)
d∑l=0
f(vl)
)
142 Page 12 of 19 F. Di Tommaso, B. Zerroudi MJOM
is the multivariate Simpson rule for a simplex [7, Theorem 5.1],
F2[f ] =A(S)
2(d + 2)!
(3
d∑l=0
f (vl) + (d + 1)3f(x∗) − d2d∑
l=0
f(gl)
)
is given by (3.7) for α = dd+1 and
R(τ)[f ] = τRSid [f ] + (1 − τ)R
(d,
d
d + 1
)[f ] (3.21)
with RSid [f ] denoting the remainder term in the multivariate Simpson rule for
a simplex. For all τ ∈ R we have R(τ)[f ] = 0, whenever f is a polynomial ind variables of total degree at most 2.
Proof. Since RSid [f ] = 0 and R
(d, d
d+1
)[f ] = 0 whenever f is a polynomial
in d variables of total degree less than or equal to 2, it easily follows thatR(τ)[f ] vanishes for each polynomial of degree at most 2. �
For τ = 3(d+1)d+3 , the family of quadrature formulas (3.20) yields to a
formula which has degree of exactness 3.
Theorem 3.5. Let f : S → R be a 3-times continuous differentiable functionon S and let us denote by x∗ the center of gravity of S, by sd−1
l , l = 0, 1, . . . , d
the facet opposite to the vertex vl and by gl the barycenter of sd−1l . The
quadrature formula∫
S
f(x)dx =A(S)
(d + 3)!
(3
d∑l=0
f(vl) + d3d∑
l=0
f(gl) + (d + 1)3(3 − d)f(x∗)
)
+R
(3(d + 1)d + 3
)[f ], (3.22)
with R(
3(d+1)d+3
)[f ] defined in (3.21), has degree of exactness 3.
Proof. For τ = 3(d+1)d+3 , the quadrature formula (3.20) reduces to
∫S
f(x)dx ≈ F3[f ]
=A(S)
(d + 3)!
(3
d∑l=0
f(vl) + d3d∑
l=0
f(gl) + (d + 1)3(3 − d)f(x∗)
)
(3.23)
and, by Corollary 3.4, it follows that F3[f ] has degree of exactness 2. LetP3(x) be a polynomial in d variables of degree 3; we can write P3(x) as
P3(x) = P2(x) +d∑
i=0
cix3i +
d∑i=0
d∑j=0j �=i
bijx2i xj +
d∑i=0
d∑j=0j �=i
d∑k=0
k �=i,j
dijkxixjxk
MJOM On Some Numerical Integration Formulas on the. . . Page 13 of 19 142
where P2(x) is a polynomial of degree 2. Therefore, it is sufficient to provethat F3[·] is exact for the monomials
M1 = x3i ; M2 = x2
i xj (j = i); M3 = xixjxk (k = i, j) i, j, k = 0, . . . , d.
Thanks to the linear isomorphism which maps the standard simplex Δd ofR
d to a generic simplex S, without loss of generality, we can restrict to thecase of the simplex Δd of vertices v0 = (0, . . . , 0); v1 = (1, 0, . . . , 0); vl =(0, 0, . . . , 1, . . . , 0); vd = (0, . . . , 0, 1). The center of gravity of Δd is x∗ =(
1d+1 , . . . , 1
d+1
)and the barycenter of the facets are
g0 =(
1d , 1
d , . . . , 1d
), g1 =
(0, 1
d , . . . , 1d
), . . . , gd =
(1d , 1
d , . . . , 0).
Let us now consider M1 = x3i , i = 0, . . . , d. The exact integral of M1 over the
simplex Δd is [11] ∫Δd
M1dx =6
(d + 3)!,
and, in addition,d∑
l=0
f(vl) = 1,
d∑l=0
f(gl) =d
d3,
f(x∗) = 1(d+1)3 .
(3.24)
By substituting equalities (3.24) in the quadrature formula (3.23) we have
F3[M1] =A(Δd)(d + 3)!
(3 + d3 d
d3+ (d + 1)3(3 − d)
1(d + 1)3
)
=6
(d + 3)!=
∫Δd
M1dx.
Let us consider M2 = x3i xj , i, j = 0, . . . , d and j = i. The exact integral of
M2 over the simplex Δd is [11]∫Δd
M2dx =2
(d + 3)!,
and, in addition,d∑
l=0
f(vl) = 0,
d∑l=0
f(gl) =d − 1d3
,
f(x∗) = 1(d+1)3 .
(3.25)
By substituting equalities (3.25) in the quadrature formula (3.23) we have
F3[M2] =A(Δd)(d + 3)!
(0 + d3 d − 1
d3+ (d + 1)3(3 − d)
1(d + 1)3
)
142 Page 14 of 19 F. Di Tommaso, B. Zerroudi MJOM
=2
(d + 3)!=
∫Δd
M2dx.
Finally, let us consider M3 = xixjxk, i, j, k = 0, . . . , d and k = i, j for whichthe exact integral over the simplex Δd is [11]∫
Δd
M3dx =1
(d + 3)!
andd∑
l=0
f(vl) = 0,
d∑l=0
f(gl) =d − 2d3
,
f(x∗) = 1(d+1)3 .
(3.26)
By substituting equalities (3.26) in the quadrature formula (3.23), we have
F3[M3] =A(S)
(d + 3)!
(0 + d3 d − 2
d3+ (d + 1)3(3 − d)
1(d + 1)3
)
=1
(d + 3)!=
∫Δd
M3dx.
Then ∫Δd
P3(x)dx = F3[P3]
and this shows that the degree of exactness of the quadrature formula (3.22)is 3. �
4. The 2-Simplex Case
In this section, we restrict to the case d = 2 in which S is a triangle of verticesv0, v1, v2. In this particular case, the quadrature formula (2.14) reduces to
QSr [f ] =
1(1 + r)
r∑k=0
1 + r − k
(3 + k)!
k∑j=0
(Dj
v1−v0Dk−j
v2−v0f(v0)
+Djv0−v1
Dk−jv2−v1
f(v1)
+Djv0−v2
Dk−jv1−v2
f(v2))
(4.1)
and, by easy computations, the bound for the approximation error becomes∣∣EΔ2
r [f ]∣∣ ≤
∣∣Dr+2f∣∣S
12(r + 2)!(1 + r)
(1 +
√2r
). (4.2)
The quadrature formula (3.19), which has degree of exactness 2 and usesonly function data at the vertices of S, at the midpoints of its sides and atits center of gravity, becomes∫
S
f(x)dx = QS1 [f ] + R
(2,
23
)[f ]
MJOM On Some Numerical Integration Formulas on the. . . Page 15 of 19 142
where
QS2 [f ] =
A(S)16
(2∑
l=0
f(vl) − 43
1∑l=0
2∑r=l+1
f (gl) + 9f(x∗)
)(4.3)
and
R
(2,
23
)[f ] =
3A(S)32
2∑l=0
(49
∫ 1
0
(1 − t)2D3g l−v l
f(vl + t(gl − vl))dt
−∫ 2
3
0
(23
− t
)2
D3g l−v l
f(vl + t(gl − vl))dt
).
The quadrature formula (3.22), which has degree of exactness 3 and usesonly function data at the vertices of S, at the midpoints of its sides and atits center of gravity, becomes
QS3 [f ] =
A(S)40
(d∑
l=0
f(vl) +83
d∑l=0
f(gl) + 9f(x∗)
), (4.4)
where gl =vl + vr
2, i.e. gl is nothing but the midpoint of the side of the
triangle S opposite to vl. For d = 2 and α = 1/3 the formula (3.7), which hasdegree of exactness 2 and uses function data at the midpoints of the sides ofS and at the points g l+2v l
3 , l = 0, . . . , d, becomes∫S
f(x)dx = QS2 [f ] + R
(2,
13
)[f ] (4.5)
with
QS2 [f ] =
A(S)48
(9
2∑l=0
f
(gl + 2vl
3
)−
2∑l=0
f(gl)
)(4.6)
and R(2, 1
3
)[f ] defined in (3.8).
To enhance the degree of exactness of the quadrature formula (4.6), let usconsider the midpoint formula for the 2-dimensional simplex [5]
∫S
f(x)dx =A(S)
6
2∑r=0
f(gr) + E[f ]
where E[f ] = 0 whenever f is a polynomial in 2 variables of total degree atmost 2. We set
Qmid2 [f ] =
A(S)6
2∑l=0
f(gl)
and define the convex combination∫S
f(x)dx = αQS2 [f ] + (1 − α)Qmid
2 [f ] + Eα[f ], α ∈ R. (4.7)
where
Eα[f ] = αR
(2,
13
)[f ] + (1 − α)E[f ]. (4.8)
142 Page 16 of 19 F. Di Tommaso, B. Zerroudi MJOM
Since, for all α ∈ R, Eα[f ] = 0 whenever f is a polynomial in 2 variables oftotal degree at least 2, the quadrature formula (4.7) has at least degree ofexactness 2.
Theorem 4.1. Let f : S ⊂ R2 → R be a 3-times continuous differentiable
function on S. Then, the quadrature formula∫S
f(x)dx = MQS3 [f ] + E 4
5[f ] (4.9)
with
MQS3 [f ] =
A(S)20
(2∑
l=0
f
(gl + 2vl
3
)+
13
3∑l=1
f(gl)
)
(4.10)
has degree of exactness 3.
Proof. Equality (4.10) follows by setting α = 45 in (4.7) and by rearranging.
To prove the degree of exactness of the formula (4.9), it is sufficient to followthe same arguments used in the proof of Theorem 3.5 for d = 2. �
Finally, to obtain a quadrature formula over S with degree of exactness4, we consider the convex combination of the quadrature formulas (4.4) and(4.10) ∫
S
f(x)dx = αQS3 [f ] + (1 − α)MQS
3 [f ] + E′α[f ], α ∈ R, (4.11)
where
E′α[f ] = αR
(95
)[f ] + (1 − α)E 4
5[f ],
with R(
95
)[f ] given by the Eq. (3.21) and E 4
5[f ] by the equation (4.8). Since,
for all α ∈ R , E′α[f ] = 0 whenever f is a polynomial in 2 variables of
total degree at least 3, the quadrature formula (4.11) has at least degree ofexactness 3.
Theorem 4.2. Let f : S ⊂ R2 → R be a 3-times continuous differentiable
function on S and x∗ the center of gravity of S. Then, the quadrature formula∫S
f(x)dx = MQS4 [f ] + E′
13[f ] (4.12)
where
MQS4 [f ] =
A(S)120
(2∑
l=0
f(vl) + 42∑
l=0
f(gl) + 12f(
gl + 2vl
3
)+ 9f(x∗)
)
(4.13)
and
E′13[f ] =
13R
(95
)[f ] +
(1 − 1
3
)E 4
5[f ],
has degree of exactness 4.
MJOM On Some Numerical Integration Formulas on the. . . Page 17 of 19 142
Table 1. Absolute value of the remainder terms EΔ2r [fi] =
QΔ2r [fi] − ∫
Δ2fi(x)dx, i = 1, . . . , 4; r = 1, . . . , 10
r∣∣EΔ2
r [f1]∣∣ ∣∣EΔ2
r [f2]∣∣ ∣∣EΔ2
r [f3]∣∣ ∣∣EΔ2
r [f4]∣∣
1 2.67e−3 4.39e−2 3.02e−4 1.91e−42 1.06e−3 1.03e−2 3.54e−5 3.87e−53 7.03e−5 1.86e−2 2.55e−6 1.68e−64 1.73e−5 4.35e−3 1.75e−7 1.93e−75 1.18e−6 5.51e−3 1.36e−8 8.93e−96 2.15e−7 4.01e−3 6.76e−10 7.55e−107 1.35e−8 6.67e−4 5.11e−11 3.30e−118 2.60e−9 1.85e−3 2.02e−12 2.28e−129 6.48e−10 3.82e−4 1.43e−13 9.02e−1410 5.07e−10 5.86e−4 5.16e−15 4.94e−15
Table 2. Absolute value of the remainder terms EΔ22 [fi] ,
EΔ23 [fi], EΔ2
MQ3[fi], EΔ2
MQ4[fi] i = 1, . . . , 4
fi
∣∣∣EΔ22 [fi]
∣∣∣∣∣∣EΔ2
3 [fi]∣∣∣
∣∣∣EΔ2MQ3
[fi]∣∣∣
∣∣∣EΔ2MQ4
[fi]∣∣∣
f1 1.55e−3 2.86e−4 1.46e−4 1.12e−6f2 3.85e−3 8.64e−3 4.59e−3 1.83e−4f3 3.71e−4 1.01e−5 5.17e−6 7.89e−8f4 3.76e−4 1.08e−5 5.51e−6 7.91e−8
Proof. Equality (4.13) follows by setting α = 13 in (4.11). To prove the degree
of exactness of the formula (4.12), we proceed by verifying the exactness ofthe quadrature formula for the monomials x4, x3y, xy3, x2y2, y4, similarlyto the proof of Theorem 3.5 for d = 2. �
5. Numerical Results in d = 2
To test the approximation accuracies of the proposed formulas, we considerthe case d = 2 and the standard triangle S = Δ2 of vertices v0 = (0, 0), v1 =(1, 0), v2 = (0, 1). The numerical experiments are conducted by consideringthe following set of test functions [1]
f1(x, y) = cos(√
1 + x2 + y2)
,
f2(x, y) = exp(
− (3x − 2)2 + (3y − 2)2
4
),
f3(x, y) = sin(π
4x +
π
6y)
,
f4(x, y) = sinh(π
4x +
π
6y)
.
142 Page 18 of 19 F. Di Tommaso, B. Zerroudi MJOM
In all the experiments, the exact value of the integrals for functions f1 andf2 are computed by assuming as exact the numerical integration performedby Mathematica. In Table 1, we report the absolute value of the remainderterms EΔ2
r [fi] = QΔ2r [fi] − ∫
Δ2fi(x)dx, i = 1, . . . , 4, r = 1, . . . , 10, and, in
Table 2, we display the absolute value of the remainder terms
EΔ22 [fi] = QΔ2
2 [fi] −∫
Δ2
fi(x)dx, i = 1, . . . , 4,
EΔ23 [fi] = QΔ2
3 [fi] −∫
Δ2
fi(x)dx, i = 1, . . . , 4,
EΔ2MQ3
[fi] = MQΔ23 [fi] −
∫Δ2
fi(x)dx, i = 1, . . . , 4,
EΔ2MQ4
[fi] = MQΔ24 [fi] −
∫Δ2
fi(x)dx, i = 1, . . . , 4.
Open Access. This article is licensed under a Creative Commons Attribution 4.0International License, which permits use, sharing, adaptation, distribution and re-production in any medium or format, as long as you give appropriate credit tothe original author(s) and the source, provide a link to the Creative Commonslicence, and indicate if changes were made. The images or other third party ma-terial in this article are included in the article’s Creative Commons licence, unlessindicated otherwise in a credit line to the material. If material is not included inthe article’s Creative Commons licence and your intended use is not permitted bystatutory regulation or exceeds the permitted use, you will need to obtain permis-sion directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher’s Note Springer Nature remains neutral with regard to jurisdic-tional claims in published maps and institutional affiliations.
References
[1] Costabile, F.A., Dell’Accio, F.: New embedded boundary-type quadrature for-mulas for the simplex. Numer. Algorithms 45(1–4), 253–267 (2007)
[2] Costabile, F.A., Dell’Accio, F., Guzzardi, L.: New bivariate polynomial expan-sion with boundary data on the simplex. Calcolo 45(3), 177–192 (2008)
[3] Dunavant, D.A.: High degree efficient symmetrical Gaussian quadrature rulesfor the triangle. Int. J. Numer. Methods Eng. 21(6), 1129–1148 (1985)
[4] Grundmann, A., Moller, H.M.: Invariant integration formulas for the n-simplexby combinatorial methods. SIAM J. Numer. Anal. 15(2), 282–290 (1978)
[5] Guenther, R.B., Roetman, E.L.: Newton–Cotes formulae in n-dimension. Nu-mer. Math. 14, 330–345 (1970)
[6] Guessab, A., Nouisser, O., Schmeisser, G.: Multivariate approximation by acombination of modified Taylor polynomials. J. Comput. Appl. Math. 196(1),162–179 (2006)
[7] Guessab, A., Schmeisser, G.: Convexity results and sharp error estimates inapproximate multivariate integration. Math. Comput. 73, 1365–1384 (2003)
MJOM On Some Numerical Integration Formulas on the. . . Page 19 of 19 142
[8] Guessab, A., Nouisser, O., Schmeisser, G.: Enhancement of the algebraic pre-cision of a linear operator and consequences under positivity. Positivity 13(4),693–707 (2009)
[9] He, T.X.: Dimensionality Reducing Expansion of Multivariate Integration, pp.1–23. Birkhauser, Boston (2001)
[10] Hesthaven, J.S.: From electrostatics to almost optimal nodal sets for polyno-mial interpolation in a simplex. SIAM J. Numer. Anal. 35(2), 655–676 (1998)
[11] Konerth, N.: Exact multivariate integration on simplices: an explanation ofthe Lasserre–Avrachenkov theorem. Technical report no. 2014-05 (AU-Cas-Mathstats) (2014)
[12] Lyness, J.N., Cools, R.: A survey of numerical cubature over triangles. In: Pro-ceedings of Symposia in Applied Mathematics, vol. 48. American MathematicalSociety, pp. 127–150 (1994)
[13] Shunn, L., Ham, F.: Symmetric quadrature rules for tetrahedra based on acubic close-packed lattice arrangement. J. Comput. Appl. Math. 236(17), 4348–4364 (2012)
[14] Silvester, P.: Symmetric quadrature formulae for simplexes. Math. Comput.24(109), 95–100 (1970)
[15] Vermolen, F.J., Segal, A.: On an integration rule for products of barycentriccoordinates over simplexes in R
n. J. Comput. Appl. Math. 330, 289–294 (2018)
[16] Wandzurat, S., Xiao, H.: Symmetric quadrature rules on a triangle. Comput.Math. Appl. 45(12), 1829–1840 (2003)
[17] Zhang, L., Cui, T., Liu, H.: A set of symmetric quadrature rules on trianglesand tetrahedra. J. Comput. Math. 27, 89–96 (2009)
Filomena Di TommasoDepartment of Mathematics and Computer ScienceUniversity of CalabriaVia P. Bucci, cubo 30A87036 RendeItalye-mail: filomena.ditommaso@unical.it
Benaissa ZerroudiUniversity Ibn TofailKenitraMoroccoe-mail: zerroudi@gmail.com
Received: September 25, 2019.
Revised: March 11, 2020.
Accepted: August 2, 2020.
Recommended