Comptes Rendus Mathématique Volume 354, n° 11 pages 10921095 (novembre 2016)
Doi : 10.1016/j.crma.2016.09.013
Received : 15 February 2016 ;
accepted : 3 October 2016    A converse to Fortin's Lemma in Banach spaces Une réciproque au lemme de Fortin dans les espaces de Banach  
Alexandre Ern ^{a} , JeanLuc Guermond ^{b}
^{ a} Université ParisEst, CERMICS (ENPC), 77455 MarnelaVallée cedex 2, France ^{ b} Department of Mathematics, Texas A&M University 3368 TAMU, College Station, TX 77843, USA
We establish the converse of Fortin's Lemma in Banach spaces. This result is useful to assert the existence of a Fortin operator once a discrete inf–sup condition has been proved. The proof uses a specific construction of a rightinverse of a surjective operator in Banach spaces. The key issue is the sharp determination of the stability constants. The full text of this article is available in PDF format.On montre une réciproque au lemme de Fortin dans les espaces de Banach. Ce résultat est utile afin d'affirmer l'existence d'un opérateur de Fortin une fois qu'une condition inf–sup discrète a été prouvée. La preuve utilise une construction spécifique d'un inverse à droite d'un opérateur surjectif dans les espaces de Banach. Le point crucial est la détermination précise des constantes de stabilité. The full text of this article is available in PDF format.   Let V and W be two complex Banach spaces equipped with the norms and , respectively. We adopt the convention that dual spaces are denoted with primes and are composed of antilinear forms; complex conjugates are denoted by an overline. Let a be a sesquilinear form on (linear w.r.t. its first argument and antilinear w.r.t. its second argument). We assume that a is bounded, i.e. (1)‖a‖:=supv∈Vsupw∈Wa(v,w)‖v‖V‖w‖W<∞, and that the following inf–sup condition holds:(2)α:=infv∈Vsupw∈Wa(v,w)‖v‖V‖w‖W>0. Here and in what follows, arguments in infima and suprema are implicitly assumed to be nonzero. Assume that we have at hand two sequences of finitedimensional subspaces and with and for all , where the parameter h typically refers to a family of underlying meshes. The spaces and are equipped with the norms of V and W , respectively. A question of fundamental importance is to assert the following discrete inf–sup condition: (3)αˆh:=infvh∈Vhsupwh∈Wha(vh,wh)‖vh‖V‖wh‖W>0. The aim of this Note is to prove the following result.
Under the above assumptions, consider the following two statements: (i)  there exists a map and a real number such that , for all , and for all ;  (ii)  the discrete inf–sup condition ( (3)) holds.  Then, with . Conversely, with , and can be constructed to be idempotent. Moreover, can be made linear if W is a Hilbert space. The statement (i) ⇒ (ii) in Theorem 1 is classical and is known in the literature as Fortin's Lemma, see [[5]] and [[1]]. It provides an effective tool to prove the discrete inf–sup condition ((3)) by constructing explicitly a Fortin operator . We briefly outline a proof that (i) ⇒ (ii) for completeness. Assuming (i), we have supwh∈Wha(vh,wh)‖wh‖W≥supw∈Wa(vh,Πhw)‖Πhw‖W=supw∈Wa(vh,w)‖Πhw‖W≥γΠhsupw∈Wa(vh,w)‖w‖W≥γΠhα‖vh‖V, since a satisfies ((2)) and . This proves (ii) with . The proof of the converse (ii) ⇒ (i) is the main object of this Note. This property is useful when it is easier to prove the discrete inf–sup condition directly rather than constructing a Fortin operator. Another application of current interest is the analysis framework for discontinuous Petrov–Galerkin methods (dPG) recently proposed in [[3]], which includes the existence of a Fortin operator among its key assumptions. The proof of the converse is not so straightforward if one wishes to establish a sharp stability bound for , i.e. that indeed one can take . Incidentally, we observe that there is a gap in the stability constant between the direct and the converse statements, since the ratio of the two is equal to (which is independent of the discrete setting). This Note is organized as follows. In Section 2, we establish a sharp bound on the stability of the rightinverse of surjective operators in Banach spaces. Since this result can be of independent theoretical interest, we present it in the infinitedimensional setting. Then in Section 3, we prove the converse of Fortin's Lemma. The proof is relatively simple once the sharp stability estimate from Section 2 is available.


Rightinverse of surjective Banach operators  Let Y and Z be two complex Banach spaces equipped with the norms and , respectively. Let be a bounded linear map. The following result is a wellknown consequence of Banach's Open Mapping and Closed Range Theorems, see, e.g., [[2]] or [[4]].
The following three statements are equivalent: (i)  is surjective;  (ii)  is injective and is closed in ;  (iii)  the following holds: (4)infz′∈Z′‖B⁎z′‖Y′‖z′‖Z′=infz′∈Z′supy∈Y〈B⁎z′,y〉Y′,Y‖z′‖Z′‖y‖Y=:β>0.
 Let us now turn to the main result of this section. To motivate the result, assume that ((4)) holds; then B is surjective and thus admits a bounded rightinverse. The crucial question is whether the stability of this rightinverse can be formulated using precisely the constant from ((4)).
Assume that ((4)) holds and that Y is reflexive. Then there is a rightinverse map such that (5)∀z∈Z,(B∘B†)(z)=zandβ‖B†z‖Y≤‖z‖Z. Moreover, this rightinverse map is linear if Y is a Hilbert space.
Parts of this result can be found in [[4]]; for completeness, we present a proof. Owing to Lemma 2, is injective and is closed in . Since the operator is injective, it admits a leftinverse linear map such that for all . Moreover, the inf–sup condition ((4)) implies that for all . Consider now the adjoint . Let be the Hahn–Banach extension operator that extends antilinear forms over into antilinear forms over (see [[2]]); maps from to . Let (resp., ) be the canonical isometry from Y to (resp., Z to ), and observe that is an isomorphism since Y is assumed to be reflexive. Let us set (6)B†:=JY−1∘ER′Y″hb∘B⁎‡⁎∘JZ:Z→Y, and let us verify that satisfies the expected properties. We have, for all ,〈z′,B(B†(z))〉Z′,Z=〈B⁎z′,B†(z)〉Y′,Y=〈JY(B†(z)),B⁎z′〉Y″,Y′‾=〈ER′Y″hb(B⁎‡⁎(JZz)),B⁎z′〉Y″,Y′‾=〈B⁎‡⁎(JZz),B⁎z′〉R′,R‾=〈JZz,B⁎‡B⁎z′〉Z″,Z′‾=〈JZz,z′〉Z″,Z′‾=〈z′,z〉Z′,Z, where we have used that to pass from the first to the second line. This shows that . Moreover, since is an isometry and the extension operator preserves the norm, we observe that, for all ,‖B†z‖Y=‖B⁎‡⁎(JZz)‖R′=supz′∈Z′〈B⁎‡⁎(JZz),B⁎z′〉R′,R‖B⁎z′‖Y′=supz′∈Z′〈JZz,z′〉Z″,Z′‖B⁎z′‖Y′≤supz′∈Z′‖z′‖Z′‖B⁎z′‖Y′‖z‖Z. We conclude from ((4)) that . Finally, if Y is a Hilbert space, we can consider the orthogonal complement of R in (recall that R is a closed subspace of ) and write . Then, the Hahn–Banach extension operator in ((6)) can be replaced by the linear map such that, for all , for all with , , . □
  Let be the operator defined by for all . We identify with and with (since these spaces are finitedimensional). We consider the adjoint operator , and identify with . We apply Lemma 3 to , , and . Owing to the discrete inf–sup condition ((3)), we infer that ((4)) holds with . Therefore, there exists a rightinverse map such that, for all , and . Let us now set (7)Πh:=Ah⁎†∘Θ:W→Wh, with the linear map such that, for all , for all . We then infer thata(vh,Πh(w))=〈Ahvh,Ah⁎†(Θ(w))〉Wh′,Wh=〈Ah⁎(Ah⁎†(Θ(w))),vh〉Vh′,Vh‾=〈Θ(w),vh〉Vh′,Vh‾=a(vh,w), which establishes that for all . Moreover,αˆh‖Πh(w)‖W=αˆh‖Ah⁎†(Θ(w))‖W≤‖Θ(w)‖Vh′≤‖a‖‖w‖W, which proves that . In addition, we observe that〈Θ(Ah⁎†(θh)),vh〉Vh′,Vh=〈Ahvh,Ah⁎†(θh)〉Wh′,Wh‾=〈Ah⁎(Ah⁎†(θh)),vh〉Vh′,Vh=〈θh,vh〉Vh′,Vh, for all , which proves that for all . As a result, , i.e., is idempotent. Finally, if W is a Hilbert space, the rightinverse map is linear by Lemma 3, and so is the operator defined from ((7)).
Without the use of Lemma 3, one only knows that has a stable rightinverse, but a stability bound for this rightinverse is not available. Here, we obtain that, provided the discrete inf–sup condition ((3)) holds uniformly with respect to h , i.e. if there is such that for all , then a uniform stability bound holds for since for all .
Even in the case of Banach spaces, the linearity of the map can be asserted if one has at hand a stable decomposition such that there is such that the induced projector satisfies for all (this property holds in the Hilbertian setting with ). Then, one can adapt the reasoning at the end of the proof of Lemma 3 to build a stable, linear rightinverse map . The mild price to be paid is that the stability constant of now becomes .
Whether Lemma 3 holds true when Y is not reflexive seems to be an open question.    
 Boffi D., Brezzi F., Fortin M. Mixed Finite Element Methods and Applications Heidelberg, Germany: Springer (2013).    Brezis H. Functional Analysis, Sobolev Spaces and Partial Differential Equations New York: Springer (2011).    Carstensen C., Demkowicz L., Gopalakrishnan J. A posteriori error control for DPG methods SIAM J. Numer. Anal. 2014 ; 52 (3) : 13351353 [crossref]    Ern A., Guermond J.L. Theory and Practice of Finite Elements New York: SpringerVerlag (2004).    Fortin M. An analysis of the convergence of mixed finite element methods RAIRO Anal. Numér. 1977 ; 11 : 341354  
     
© 2016
Académie des sciences@@#104156@@     
