scieee Open visual document viewer

A two-phase problem with Robin conditions on the free boundary

Guarino Lo Bianco, Serena,La Manna, Domenico Angelo,Velichkov, Bozhidar

Full text

This is a sel -a chi ed e sion o an o iginal a icle. This e sion may di e om he o iginal in pagina ion and ypog aphic de ails. Au ho (s): Ti le: Yea : Ve sion: Copy igh : Righ s: Righ s u l: Please ci e he o iginal e sion: CC BY 4.0 h ps://c ea i ecommons.o g/licenses/by/4.0/ A wo-phase p oblem wi h Robin condi ions on he ee bounda y © 2021 he Au ho s Published e sion Gua ino Lo Bianco, Se ena; La Manna, Domenico Angelo; Velichko , Bozhida Gua ino Lo Bianco, S., La Manna, D. A., & Velichko , B. (2021). A wo-phase p oblem wi h Robin condi ions on he ee bounda y. Jou nal de l'École poly echnique : Ma héma iques, 8, 1-25. h ps://doi.o g/10.5802/jep.139 2021 Se ena Gua ino Lo Bianco, Domenico Angelo La Manna, &Bozhida Velichko A wo-phase p oblem wi h Robin condi ions on he ee bounda y Tome 8 (2021), p. 1-25. <h p://jep.cen e-me senne.o g/i em/JEP_2021__8__1_0> © Les au eu s, 2021. Ce ains d oi s ése és. Ce a icle es mis à disposi ion selon les e mes de la licence LICENCE INTERNATIONALE D’ATTRIBUTION CREATIVE COMMONS BY 4.0. h ps://c ea i ecommons.o g/licenses/by/4.0/ L’accès aux a icles de la e ue « Jou nal de l’École poly echnique — Ma héma iques » (h p://jep.cen e-me senne.o g/), implique l’acco d a ec les condi ions géné ales d’u ilisa ion (h p://jep.cen e-me senne.o g/legal/). Publié a ec le sou ien du Cen e Na ional de la Reche che Scien i ique Publica ion memb e du Cen e Me senne pou l’édi ion scien i ique ou e e www.cen e-me senne.o g Tome 8, 2021, p.1–25 DOI: 10.5802/jep.139 A TWO-PHASE PROBLEM WITH ROBIN CONDITIONS ON THE FREE BOUNDARY by Se ena Gua ino Lo Bianco, Domenico Angelo La Manna & Bozhida Velichko Abs ac . — We s udy o he i s ime a wo-phase ee bounda y p oblem in which he solu ion sa is ies a Robin bounda y condi ion. We conside he case in which he solu ion is con inuous ac oss he ee bounda y and we p o e an exis ence and a egula i y esul o minimize s o he associa ed a ia ional p oblem. Finally, in he appendix, we gi e an example o a class o S eine symme ic minimize s. Résumé (Un p oblème à on iè e lib e à deux phases a ec condi ions au bo d de Robin) Nous é udions pou la p emiè e ois un p oblème à on iè e lib e à deux phases pou lequel la solu ion sa is ai à une condi ion de Robin au bo d. Nous considé ons le cas où la solu ion es con inue au bo d e nous mon ons un ésul a d’exis ence e de égula i é pou les minimiseu s du p oblème a ia ionnel associé. En in, nous donnons dans l’appendice un exemple d’une classe de minimiseu s a ec une symé ie de S eine . Con en s 1. In oduc ion.................................................................. 2 2. P elimina ies.................................................................. 7 3. A amily o app oxima ing p oblems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4. Exis ence o an op imal se . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . 13 5. Regula i y o he ee bounda y. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Appendix. Examples o minimize s.............................................. 24 Re e ences....................................................................... 25 2020 Ma hema ics Subjec Classi ica ion. — 35R35, 49Q10. Keywo ds. — F ee bounda y p oblems, wo-phase, Robin bounda y condi ions, egula i y. The i s au ho was pa ially suppo ed by PRIN 2017 Nonlinea Di e en ial P oblems ia Va i- a ional, Topological and Se - alued Me hods (G an 2017AYM8XW) and he INdAM-GNAMPA p ojec 2020 “P oblemi di o imizzazione con incoli ia aspo o o imo e ince ezza”. The second au ho was pa ially suppo ed by he Academy o Finland g an 314227. The hi d au ho has been pa ially suppo ed by he Eu opean Resea ch Council (ERC) unde he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme (g an ag eemen VAREG, No. 853404). e-ISSN: 2270-518X h p://jep.cen e-me senne.o g/ 2 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko 1. In oduc ion Fo a ixed a cons an β > 0and a smoo h bounded open se D⊂Rd,d⩾2, we conside he unc ional Jβ(u, Ω) = ZD |∇u|2dx +βZ∂∗Ω u2dHd−1, de ined on he pai s (u, Ω), whe e u∈H1(D),Ω⊂Rdis a se o ini e pe ime e in he sense o De Gio gi (see Sec ion 2) and ∂∗Ωdeno es he educed bounda y o Ω (see Sec ion 2); when Ωis smoo h, ∂∗Ωis he opological bounda y o Ω. In his pape we s udy he exis ence and he egula i y o minimize s o he unc- ional Jβamong all pai s (u, Ω), which a e ixed ou side he domain D. P ecisely, h oughou he pape , we ix a se E⊂Rdo ini e pe ime e , a cons an s m > 0and a unc ion ∈H1 loc(Rd)such ha ⩾min Rdand Z∂∗E 2dHd−1<+∞; we de ine he admissible se s V=u∈H1 loc(Rd) : u− ∈H1 0(D), E=Ω⊂Rd: Pe (Ω) <+∞and Ω = Ein Rd D, and we conside he a ia ional minimiza ion p oblem (1.1) min Jβ(u, Ω) : u∈ V,Ω∈ E. Ou main esul is he ollowing. Theo em 1.1 (Exis ence and egula i y o minimize s). — Le β >0,D⊂Rd, ,E,V and Ebe as abo e. Then he ollowing holds. (i) The e exis s a solu ion (u, Ω) ∈ V × E o he a ia ional p oblem (1.1). (ii) Fo e e y solu ion (u, Ω) o (1.1),uis Hölde con inuous and bounded om below by a s ic ly posi i e cons an in D. (iii) I (u, Ω) is a solu ion o (1.1), hen he ee bounda y ∂Ω∩Dcan be decomposed as he disjoin union o a egula pa Reg(∂Ω) and a singula pa Sing(∂Ω), whe e: –Reg(∂Ω) is a C∞hype su ace and a ela i ely open subse o ∂Ω, and he unc ion uis C∞smoo h on Reg(∂Ω); –Sing(∂Ω) is a closed se , which is emp y i d⩽7, disc e e i d= 8, and o Hausdo dimension d−8, i d > 8. Rema k 1.2. — We no ice ha i (u, Ω) is a solu ion o (1.1), hen uis ha monic in he in e io o Ωand D Ω. Thus, as a consequence o Theo em 1.1(iii), in a neighbo hood o a egula poin x0∈Reg(∂Ω), he unc ions u: Ω →Rand u:D Ω→Ra e C∞ up o he ee bounda y ∂Ω. J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 3 1.1. Ou line o he p oo and o ganiza ion o he pape . — The main di icul y in he p oo o Theo em 1.1 is o p o e he exis ence o a minimizing pai s (u, Ω) and o show ha he unc ion uis Hölde con inuous and bounded om below by a s ic ly posi i e cons an in D. The almos -minimali y o he solu ions is p o ed in Theo em 5.1. Finally, in he Appendix, we gi e examples o minimize s in domains D symme ic wi h espec o he hype plane {xd= 0}. 1.1.1. Exis ence. — The exis ence o a solu ion (u, Ω) and he egula i y o u(Hölde egula i y and non-degene acy) a e ea ed simul aneously. The eason is ha i (un,Ωn)is a minimizing sequence o (1.1), hen in o de o ge he compac ness o Ωn, we need a uni o m bound ( om abo e) on he pe ime e Pe (Ωn), o which we need he unc ions un o be bounded om below by a s ic ly posi i e cons an . Now, no ice ha we canno simply eplace unby un∨ε, o some ε > 0; his is due o he ac ha he second e m in Jβis inc easing in u: Z∂∗Ωn u2 ndHd−1⩽Z∂∗Ωn (ε∨un)2dHd−1. Thus, we selec a minimizing sequence which is in some sense op imal. P ecisely, we ake (un,Ωn) o be solu ion o he auxilia y p oblem (1.2) min Jβ(u, Ω) : u∈ V,Ω∈ E, u ⩾1/n in D, o which he exis ence o an op imal se is much easie (see Sec ion 3, P oposi ion 3.1). S ill, we do no ha e a uni o m (independen om n) bound om below o he unc ions un, so we s ill miss he uni o m bound on he pe ime e o Ωn. On he o he hand, we a e able o p o e ha he sequence unis uni o mly Hölde con inuous in D(see Sec ion 3, Lemma 3.5). This enables us o ex ac a subse- quence un ha con e ges locally uni o mly in D o a non-nega i e Hölde con inuous unc ion u∞:D→R(see Sec ion 4). Now, on each o he se s {u∞> }, > 0, he sequence Ωnhas uni o mly bounded pe ime e . This enables us o ex ac a sub- sequence Ωn ha con e ges poin wise almos -e e ywhe e on {u∞>0} o some Ω∞. Thus, we ha e cons uc ed ou candida e o a solu ion: (u∞,Ω∞). In o de o p o e ha (u∞,Ω∞)is an admissible compe i o in (1.1), we need o show ha Ω∞has ini e pe ime e . We do his in Sec ion 4. We i s use he op imali y o (un,Ωn) o p o e ha (u∞,Ω∞)is op imal when compa ed o a special class o compe i o s. This op imali y condi ion can be w i en as (we e e o Lemma 4.1 o he p ecise s a emen ): (1.3) Jβ(u∞,Ω∞)⩽Jβ(u ,Ω ),whe e u =u∞∨ and Ω = Ω∞∪ {u∞⩽ }, o any > 0. Nex , om his special op imali y condi ion we deduce ha he unc- ion u∞is bounded om below by a s ic ly posi i e cons an (see P oposi ion 4.2). F om his, in Sec ion 4, we deduce ha Ω∞has ini e pe ime e in Rdand ha he pai s (u∞,Ω∞)is a solu ion o (1.1). J.É.P.—M., 2021, ome8 4 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko 1.1.2. Hölde con inui y and non-degene acy o u. — Le now (u, Ω) be any solu ion o (1.1). In o de o p o e he Hölde con inui y and he non-degene acy o ui is su icien o exploi some o he es ima es ha we al eady used o p o e he exis ence. Indeed, we can es he op imali y o (u, Ω) wi h he compe i o s om (1.3). Thus, o > 0small enough, we ha e (1.4) Jβ(u, Ω) ⩽Jβ(u ,Ω )whe e u =u∨ and Ω = Ω ∪ {u⩽ }. In pa icula , ZD |∇u|2dx +βZ∂∗Ω u2⩽ZD |∇(u∨ )|2dx +βZ∂∗(Ω∪{u< }) u2 ⩽ZD |∇(u∨ )|2dx +β 2Pe ({u < }) + βZ{u> }∩∂∗Ω u2, which p o es ha usa is ies he op imali y condi ion (4.1) om Lemma 4.1: (1.5) Z{u< } |∇u|2dx ⩽β 2Pe {u< }. Now, applying P oposi ion 4.2, we ge ha uis bounded om below by a s ic ly posi i e cons an in D. Finally, P oposi ion 3.5 gi es ha uis Hölde con inuous in D. This p o es Theo em 1.1(iii). 1.1.3. Regula i y o he ee bounda y. — In o de o p o e he egula i y o he ee bounda y (Theo em 1.1(iii)), we use he Hölde con inui y and he non-degene acy o u o show ha a solu ion Ωis an almos -minimize o he pe ime e . We do his in Theo em 5.1. Now, om he classical egula i y heo y o almos -minimize s o he pe ime e (see [8]), we ob ain ha (inside D) he ee bounda y ∂Ωcan be decomposed in o a C1,α- egula pa Reg(∂Ω) and a (possibly emp y) singula pa o Hausdo dimension smalle han d−8. Finally, in Theo em 5.2, we p o e he C∞ egula i y o Reg(∂Ω). In o de o do so, we i s show (see Lemma 5.3) ha in a neighbo hood o a egula poin x0, he es ic ions u+and u−o uon Ωand D Ωa e solu ions o he ollowing ansmission p oblem:              ∆u+= 0 in Ω, ∆u−= 0 in D Ω, u+=u−=uon ∂Ω, ∂u+ ∂νΩ −∂u− ∂νΩ + 2βu = 0 on ∂Ω, whe e νΩis he no mal de i a i e o ∂Ω. Now, using he ecen esul s [4] and [5], we ge ha u+and u−a e as egula as he ee bounda y ∂Ω(see Lemma 5.4). On he o he hand, using a ia ions o ualong smoo h ec o ields, we ob ain ha Reg(∂Ω) sol es an equa ion o he o m “Mean cu a u e o ∂Ω"=F(∇u+,∇u−, u±)on ∂Ω, J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 5 whe e Fis an explici ( a ional) unc ion o ∇u±and u. In pa icula , his implies ha ∂Ωgains one mo e de i a i e wi h espec o u, ha is, u∈Ck,α ⇒∂Ω∈Ck+1,α. Thus, by a boo s ap a gumen , he egula pa o he ee bounda y is C∞. 1.2. On he non-degene acy o he solu ions. — We no ice ha he compe i o s (u ,Ω )in (1.3) a e he wo-phase analogue o he ones used by Ca a elli and K i en so in [3], whe e he au ho s s udy a one-phase e sion o (1.1). Ne e heless, he unc ional in [3] in ol es he measu e o Ω, which means ha he op imali y condi ion he e co esponds o Jβ(u, Ω) + C|Ω∩ {u⩽ }| ⩽Jβ(u ,Ω ),whe e u =u∨ and Ω = Ω {u⩽ }, whe e C > 0. The p esence o he cons an Cenables us o p o e he bound om below by using a di e en ial inequali y o a sui ably chosen unc ion ( ), which is gi en in e ms o uand {u< }(see P oposi ion 4.2 and [3, Th. 3.2]). In P oposi ion 4.2, we exploi he same idea, bu since we do no ha e he cons an C, we can only conclude ha ( )⩾ε (which is no in con adic ion wi h he ac ha ( )is de ined o e e y > 0). So, we con inue, and we use his lowe bound o ob ain a bound o he o m (1.6) c⩽β1/2Pe ({u< })1/2|{u < }|1/2 o e e y > 0, whe e u:= u∞and cis a cons an depending on βand d. Then, we no ice ha his en ails c⩽β3/4Pe ({u< })1/4|{u < }|3/4 o e e y > 0. and we use an i e a ion p ocedu e o ge ha c⩽β1−1/2nPe ({u < })1/2n|{u < }|1−1/2n o e e y > 0. Passing o he limi as n→ ∞, we ge ha i uis no bounded away om ze o, hen (1.7) c⩽β|{u < }| ⩽β|D| o e e y > 0. Now, his means ha he measu e o he ze o-se |{u= 0}| is bounded om below. Thus, using again he op imali y o u, we ge ha (1.6) holds wi h an a bi a y small ε > 0in place o β, we ge ha c⩽ε|{u < }| o e e y > 0, which is impossible. A simila non-degene acy esul was p o ed by Bucu and Giacomini in [1] by a De Gio gi i e a ion scheme(1). P ecisely, one can p o e ha any solu ion o (1.1) sa is ies he op imali y condi ion om [1, Rem. 3.7]. Thus, [1, Th. 3.5] also applies o he solu ions o (1.1). Con e sely, he a gumen om 4.2 can be applied o he minimize s o [1] o ob ain he bound om below o [1, Th. 3.5]. (1)We a e g a e ul o he anonymous e e ee o b inging o ou a en ion he e e ence [1]. J.É.P.—M., 2021, ome8 6 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko 1.3. One-phase and wo-phase p oblems wi h Robin bounda y condi ions The p oblem (1.1) is he i s ins ance o a wo-phase ee bounda y p oblem wi h Robin bounda y condi ions. P ecisely, we no ice ha i Ωis a ixed se wi h smoo h bounda y and i uminimizes he unc ional Jβ(·,Ω) in H1(D), hen he unc ions u+:= uon Ωand u−:= uon D Ω, a e ha monic in Ωand D Ω, and sa is y he ollowing condi ions: (1.8) u+=u−and ∂u+ ∂ν+ +β 2u++∂u− ∂ν− +β 2u−= 0 on ∂Ω∩D, whe e ν+and ν−a e he ex e io and he in e io no mals o ∂Ω. No ice ha (1.8) is a wo-phase coun e pa o he one-phase p oblem (1.9) ∆u= 0 in Ω,∂u ∂ν +βu = 0 on ∂Ω∩D, which was s udied by Bucu -Luckhaus in [2] and Ca a elli-K i en so in [3]. As explained in [3], he Robin condi ion in (1.9) na u ally a ises in he physical si ua ion in which he hea di uses eely in Ω, he empe a u e is se o be ze o on he su ace ∂Ω, which is sepa a ed om he in e io o Ωby an in ini esimal insula o . The wo-phase p oblem (1.8) also may be in e p e ed in his way, in his case he hea di uses eely bo h inside Ωand ou side, in D Ω; he empe a u e is se o be ze o on he su ace ∂Ω, which is insula ed om bo h sides; he con inui y o he empe a u e means ha he hea ans e is allowed also ac oss ∂Ω, which happens o ins ance i he su ace ∂Ωis eplaced by a e y hin (in ini esimal) ne . E en i he p oblems in [2, 3] and in he p esen pape lead o he ee bounda y condi ions o he same ype, he echniques a e comple ely di e en . Fo ins ance, he p oblem s udied in [2, 3] is a ee discon inui y p oblem as he unc ion ujumps om posi i e in Ω o ze o in D Ω. Thus, he co esponding a ia ional minimiza ion p oblem can be na u ally s a ed in he class o SBV unc ions, which clea ly in luences bo h he exis ence and he egula i y echniques; oughly speaking, he exis ence is ob ained h ough a compac ness heo em in he SBV class, while he egula i y elies on echniques ela ed o he Mum o d-Shah unc ional. In ou case, he p oblem can be s a ed o he unc ions (u1, u2)wi h disjoin suppo s (u1u2= 0 almos -e e ywhe e in D) which sa is y he ollowing cons ain s: he sum u1+u2should be a Sobole unc ion ( his co esponds o he con inui y condi ion in (1.8)); u2 1and u2 2a e SBV unc ions whose jump se s a e con ained in he bounda y o he posi i i y se s {u1>0}and {u2>0}. Now, i is easonable o expec ha an exis ence esul can be p o ed also in his class, bu hen, in o de o p o e ha a solu ion o (1.1) exis s, one should show ha u1and u2a e o he o m u1=u 1 Ωand u2=u 1 D Ω o a se o ini e pe ime e Ω⊂Rd,ubeing he sum u1+u2. Summa izing, wo king in he class o SBV unc ions would allow o s a e (1.1) in a weake o m, bu i doesn’ seem o be a sho cu o he exis ence o a solu ion (o (1.1)) as i will equi e he analysis o he jump se s o he op imal pai s J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 7 in he SBV class. Thus, we p e e no o ely on he ad anced compac ness esul s o SBV unc ions, bu o p o e he exis ence o a solu ion om sc a ch. Finally, as explained in Sec ion 1.1, once we know ha an op imal pai s (u, Ω) exis s, and ha uis non-degene a e and Hölde con inuous, he egula i y o he ee bounda y ∂Ω ollows immedia ely since he se Ωbecomes an almos -minimize o he pe ime e . 2. P elimina ies 2.1. Se s o ini e pe ime e . — Le A⊂Rdbe a an open se in Rd. We ecall ha he se E⊂Rdis said o ha e a ini e pe ime e in Ai (2.1) Pe (E, A) = sup nZA di ξ(x)dx :ξ∈C1 c(A;Rd),sup x∈Rd |ξ(x)|⩽1o is ini e. We say ha Ehas a locally ini e pe ime e in A, i o e e y open se B⊂Rd such ha B⊂A, we ha e ha Pe (E, B)<∞. We say ha Eis o ini e pe ime e i Pe (E) := Pe (E, Rd)<+∞. By he De Gio gi s uc u e heo em (see o ins ance [7, Th. II.4.9]), i he se E⊂Rd has locally ini e pe ime e in A, hen he e is a se ∂∗E⊂A∩∂E called educed bounda y such ha Pe (E, B) = Hd−1(B∩∂∗E) o e e y se BbA, whe e Hd−1is he (d−1)-dimensional Hausdo measu e in Rd. Mo eo e , he e is aHd−1-measu able unc ion νE:∂∗E→Rd, called gene alized no mal such ha |νE|= 1 and ZE di ξ(x)dx =Z∂∗E νE·ξ dHd−1 o e e y ξ∈C1 c(A;Rd). 2.2. Capaci y and aces o Sobole unc ions. — We de ine he capaci y (o he 2-capaci y) o a se E⊂Rdas cap(E) = in kuk2 H1(Rd):u∈H1(Rd), u ⩾1in a neighbo hood o E. Suppose now ha d⩾3. I is well-known ha he se s o ze o capaci y ha e ze o d−1dimensional Hausdo measu e (see o ins ance [6, §4.7.2, Th. 4]): I cap(E)=0, hen Hd−1(E)=0. The Sobole unc ions a e de ined up o a se o ze o capaci y (i.e., quasi-e e ywhe e), ha is, i A⊂Rdis an open se and u∈H1(A), hen he e is a se Nu⊂Rdsuch ha cap (Nu)=0and u(x0) = lim →0 1 |B |ZB (x0) u(x)dx o e e y x0∈A Nu. Mo eo e , o e e y unc ion u∈H1(A) he e is a sequence un∈C∞(A)∩H1(A) and a se N ⊂ Ao ze o capaci y such ha : –uncon e ges o us ongly in H1(A); –u(x) = limn→∞ un(x) o e e y x∈A (N ∪ Nu). J.É.P.—M., 2021, ome8 14 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko The cons uc ion o Ω0is mo e delica e. Fi s , we ix > 0and δ > 0and we no ice ha he pe ime e o Ωεnis bounded on he open se {u0> } ∩ Dδ. Indeed, he uni o m con e gence o uεn o u0implies ha , o nla ge enough (n⩾N ,δ, o some ixed N ,δ ∈N), uεn⩾ 2on Dδ∩ {u0> }. Thus, we ha e Jβ( , E)⩾βZDδ∩{u0> }∩∂∗Ωεn u2 εndHd−1⩾β 2 2Pe Ωεn;Dδ∩ {u0> }. Now, i we choose such ha Pe ({u0> })<∞(which, by he co-a ea o mula, is ue o almos -e e y > 0), hen we ha e ha Pe Ωεn∩ {u0> } ∩ Dδ⩽C ,δ o e e y n⩾N ,δ, o some cons an C ,δ >0. Now, since all he se s Ωεn∩ {u0> } ∩ Dδa e con ained in Dand ha e uni o mly bounded pe ime e , we can ind a se Ω0and a subsequence o which 1 Ωεn∩{u0> }∩Dδ(x)−→ 1 Ω0∩{u0> }∩Dδ(x) o almos -e e y x∈D. Thus, by a diagonal sequence a gumen , we can ex ac a subsequence o εn(s ill deno ed by εn) and we can de ine he se Ω0⊂Rdas he poin wise limi 1 Ω0(x) = lim n→∞ 1 Ωεn∩{u0>0}(x) o almos -e e y x∈ {u0>0}, and we no ice ha , by cons uc ion, Ω0⊂ {u0>0}. No ice ha , we do no know a p io i ha Ω0has ini e pe ime e . We only know ha Pe (Ω0∩ {u0> } ∩ Dδ)<∞ o e e y δ > 0and almos -e e y > 0. which means ha Ω0∩ {u0> }has locally ini e pe ime e in D o a.e. > 0. 4.2. An op imali y condi ion. — As poin ed ou abo e, we do no know i he pai s (u0,Ω0)is e en an admissible compe i o o (1.1) (we need o show ha Ω0∈ E). Ne e heless, we can s ill p o e ha i sa is ies a sui able op imali y condi ion. Lemma 4.1 (The op imali y condi ion a he limi ). — Le u0and Ω0be as in Sec- ion 4.1. Then, o almos -e e y > 0, we ha e (4.1) Z{u0< } |∇u0|2dx ⩽β 2Pe {u0< }. P oo . — Le now > 0be ixed and such ha he se {u0< }has ini e pe ime e . Then, o nla ge enough, we can use he pai s (u0∨ , Ω0∪ {u0< }) o es he op imali y o (uεn,Ωεn). No ice ha he se Ω0∪ {u0< }has ini e pe ime e o J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 15 a.e. ∈(0, m), as obse ed in he p e ious sec ion. Fo he sake o simplici y, we w i e uεn=un,Ωεn= Ωn,u0=uand Ω0= Ω. Thus, we ha e ZD |∇un|2dx+βZ{u> }∩∂∗Ωn u2 ndHd−1 ⩽ZD |∇un|2dx +βZ∂∗Ωn u2 ndHd−1 ⩽ZD |∇(u∨ )|2dx +βZ∂∗(Ω∪{u< }) u2dHd−1 (4.2) ⩽ZD |∇(u∨ )|2dx +β 2Pe ({u < }) + βZ{u> }∩∂∗Ω u2dHd−1. Now, by he weak con e gence o un o u, we ge ha ZD |∇u|2dx ⩽lim in n→∞ ZD |∇un|2dx. On he o he hand, se ing U ,δ o be he open se U ,δ =Rd Dδ∩ {u⩽ }, o some ixed δ > 0, and applying Lemma 2.4, we ha e ha ZU ,δ∩∂∗Ω u2dHd−1⩽lim in n→∞ ZU ,δ∩∂∗Ωn u2 ndHd−1⩽lim in n→∞ Z{u> }∩∂∗Ωn u2 ndHd−1. Taking he limi as δ→0, by he mono one con e gence heo em, we ge ha lim δ→0ZU ,δ∩∂∗Ω u2dHd−1=ZRd (D∩{u⩽ })∩∂∗Ω u2dHd−1 Now, since u(x) = h(x) o quasi-e e y x∈Rd Dand o Hd−1-almos -e e y x∈Rd D, and since h⩾m > on ∂D, we ha e ha (4.3) ZRd (D∩{u⩽ })∩∂∗Ω u2dHd−1=Z{u> }∩∂∗Ω u2dHd−1. Thus, we ge ha (4.4) Z{u> }∩∂∗Ω u2dHd−1⩽lim in n→∞ ZD∩{u> }∩∂∗Ωn u2 ndHd−1. Now, using (4.4) and (4.2), we ob ain ZD |∇u|2dx+βZ{u> }∩∂∗Ω u2dHd−1 ⩽lim in n→∞ ZD |∇un|2dx +βZ{u> }∩∂∗Ωn u2 ndHd−1 ⩽ZD |∇(u∨ )|2dx +β 2Pe ({u < }) + βZ{u> }∩∂∗Ω u2dHd−1, which gi es (4.1).  J.É.P.—M., 2021, ome8 16 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko 4.3. Non-degene acy. — The c ucial obse a ion in his sec ion is ha he unc- ions usa is ying he op imali y condi ion (4.1) a e non-degene a e in he sense o he ollowing p oposi ion. P oposi ion 4.2 (Non-degene acy). — Le β > 0,m > 0,Dbe a bounded open se o Rdand u∈H1(D)be a non-nega i e unc ion in Dsuch ha u⩾mon ∂D. Le Ω⊂Dbe a se o ini e pe ime e in D. Suppose ha uand Ωsa is y he op imali y condi ion (4.5) ZΩ |∇u|2dx ⩽β 2Pe (Ω )whe e Ω ={u⩽ }, o almos -e e y ∈(0, m). Then, |Ω |= 0 o some > 0. P oo . — By con adic ion, suppose ha |Ω |>0 o e e y > 0. Le ∈(0, m)be ixed. By he co-a ea o mula, he Cauchy-Schwa z inequali y and he op imali y condi ion (4.5), we ge (4.6) ZΩ |∇u|=Z 0 Pe (Ωs)ds ⩽ZΩ |∇u|21/2 |Ω |1/2⩽ β1/2Pe (Ω )1/2|Ω |1/2. We now se ( ) := Z 0 Pe (Ωs)ds =ZΩ |∇u|dx. Using (4.6), we will es ima e ( ) om below. S ep 1. Non-degene acy o . — By he isope ime ic inequali y and he es ima e (4.6), he e is a dimensional cons an Cdsuch ha Z 0 Pe (Ωs)ds ⩽ β1/2CdPe (Ω )(2d−1)/(2d−2). Using he de ini ion o , we can e-w i e his inequali y as ( )(2d−2)/(2d−1) ⩽ (2d−2)/(2d−1)β1/2Cd(2d−2)/(2d−1) 0( ). A e ea anging he e ms and in eg a ing om 0 o , we ob ain ( )1/(2d−1) − (0)1/(2d−1) ⩾ 1/(2d−1) β1/2Cd(2d−2)/(2d−1) . Now, since uis non-nega i e in D, we ha e ha (0) = 0. Thus ( )⩾ β1/2Cd2d−2. Se ing (4.7) C=βCd1−d, we ob ain he lowe bound ( )⩾C . J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 17 In pa icula , as a consequence o (4.6), we ge ha (4.8) C⩽β1/2Pe (Ω )1/2|Ω |1/2. S ep 2. Non-degene acy o |Ω |. — Le α∈(0,1) be ixed. Then, we ha e ha Z 0 Pe (Ωs)α|Ωs|1−αds ⩽Z 0 Pe (Ωs)dsαZ 0 |Ωs|ds1−α ⩽ β1/2Pe (Ω )1/2|Ω |1/2α |Ω |1−α = βα/2Pe (Ω )α/2|Ω |1−α/2. Thus, we ob ain ha o ixed T∈(0, m)and C > 0, he ollowing implica ion holds: (4.9) (I C⩽Pe (Ω )α|Ω |1−α o e e y ∈(0, T ), hen C⩽βα/2Pe (Ω )α/2|Ω |1−α/2 o e e y ∈(0, T ). We claim ha , o e e y n⩾1and e e y ∈(0, m), we ha e he inequali y (4.10) C⩽β1−1/2nPe (Ω )1/2n|Ω |1−1/2n. In o de o p o e (4.10), we a gue by induc ion on n. When n= 1, (4.10) is p e- cisely (4.8). In o de o p o e ha he claim (4.10) o n∈Nimplies he same claim o n+ 1, we apply (4.9) o α= 2−n,n∈N, which gi es p ecisely (4.10) wi h n+ 1. This concludes he p oo o (4.10). Nex , passing o he limi as n→ ∞, we ob ain ha C⩽β|Ω | o e e y ∈(0, T), whe e Cis gi en by (4.7). Thus, he e is a dimensional cons an Cd>0such ha (4.11) β−dCd⩽|Ω | o e e y ∈[0, m). S ep 3. Conclusion. — We now no ice ha lim →0|Ω |=|Ω0|>0. Thus, o e e y ε > 0, he e is Tεsuch ha o all ∈(0, Tε)we ha e ZΩ |∇u|=Z 0 Pe (Ωs)ds ⩽ZΩ |∇u|21/2 |Ω Ω0|1/2 ⩽ ε1/2Pe (Ω )1/2|Ω |1/2. (4.12) Now, epea ing he a gumen o S ep 1 and S ep 2, we ge ha (4.11) should hold wi h εin place o β. Since ε > 0is a bi a y, his is a con adic ion.  J.É.P.—M., 2021, ome8 18 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko 4.4. Exis ence o a solu ion. — We a e now in posi ion o p o e ha he pai s (u0,Ω0), cons uc ed in Sec ion 4.1, is a solu ion o (1.1). P oposi ion 4.3 (Exis ence o a solu ion). — The e is a dimensional cons an Cd>0 such ha i Dis a bounded open se o Rdand β > 0is a gi en posi i e cons an , hen he ollowing holds. Fo e e y se E⊂Rdo ini e pe ime e and e e y ∈H1(Rd) sa is ying ⩾mon D o some cons an m > 0, he e is a solu ion (u, Ω) o he p oblem (1.1). P oo . — Le (u0,Ω0)be as in Sec ion 4.1. Then, by Lemma 4.1, (u0,Ω0)sa is ies he op imali y condi ion (4.5). Now, by P oposi ion 4.2 we ge ha u0⩾ in D, o some > 0. In pa icula , Ω0has ini e pe ime e in D. P ecisely, o e e y δ > 0, we ha e Pe (Ω0;Dδ)⩽lim in n→∞ Pe (Ωεn;Dδ)⩽4 2lim in n→∞ ZDδ∩∂∗Ωεn u2 εndHd−1 ⩽4 β 2lim in n→∞ Jβuεn,Ωεn⩽4 β 2Jβ( , E). Passing o he limi as δ→0, we ge Pe (Ω0;D)⩽4 β 2Jβ( , E). In pa icula , his implies ha Ω0is a se o ini e pe ime e in Rd. Indeed, Pe (Ω0)⩽Pe (Ω0;D) + 2Pe (D) + Pe (Ω0;Rd D) ⩽4 β 2Jβ( , E) + 2Pe (D) + Pe (E;Rd D). Thus, he pai s (u0,Ω0)is admissible in (1.1); i now emains o p o e ha i is op imal. Le eu∈H1(D)be non-nega i e on Dand such ha u− ∈H1 0(D). Le e Ω⊂Rdbe a se o ini e pe ime e such ha e Ω = Eon Rd D. I is su icien o p o e ha Jβ(u0,Ω0)⩽Jβ(eu, e Ω). Le ε > 0be ixed. We now use he pai s (eu∨ε, e Ω) o es he op imali y o uεn,Ωεn: Jβuεn,Ωεn⩽Jβ(eu∨ε, e Ω). Passing o he limi as ε→0, we ge Jβuεn,Ωεn⩽Jβ(eu, e Ω). Now, Lemma 2.4 and he semicon inui y o he H1no m gi es ha Jβ(u0,Ω0)⩽ Jβ(eu, e Ω), which concludes he p oo .  J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 19 5. Regula i y o he ee bounda y In his sec ion, we p o e he egula i y o he ee bounda y. In Theo em 5.1, we p o e ha he solu ions o (1.1) a e almos -minimize s o he pe ime e in D. As a consequence, ∂Ωcan be decomposed in o a egula and a singula pa and ha he egula pa is C1,α mani old. Then, in Theo em 5.2, we p o e ha he egula pa o he ee bounda y is C∞smoo h. Theo em 5.1. — Le (u, Ω) be a solu ion o (1.1). he e is a cons an C > 0such ha Ωis an almos -minimize o he pe ime e in he ollowing sense: Pe Ω ; B (x0)⩽1 + C 1/3Pe Ω0;B (x0), o e e y ball B (x0)⊂Dand e e y se Ω0⊂Rdsuch ha Ω=Ω0ou side B (x0). In pa icula , he ee bounda y ∂Ω∩Dcan be decomposed as he disjoin union o a egula pa Reg(∂Ω) and a singula pa Sing(∂Ω), whe e (i) Reg(∂Ω) is a ela i ely open subse o ∂Ωand is a C1,α smoo h mani old; (ii) Sing(∂Ω) is a closed se , which is emp y i d⩽7, disc e e i d= 8, and o Hausdo dimension d−8, i d > 8. P oo . — We i s no ice ha by Lemma 3.4, u∈C0,1/3(D). Le δ > 0,x0∈Dδand < δ/2. We conside a se Ω0⊂Rdsuch ha Ω0∆Ω bB (x0). Tes ing he op imali y o (u, Ω) agains (u, Ω0)we ge ha ZB (x0)∩∂∗Ω u2dHn−1⩽ZB (x0)∩∂∗Ω0 u2dHn−1, which implies ha min B (x0)u2Pe Ω ; B (x0)⩽max B (x0)u2Pe Ω0;B (x0). By egula i y o u, we ha e ha max B (x0)u2⩽min B (x0)u2+C 1/3⩽min B (x0)u21 + C 1/3, whe e in he second inequali y, we used ha u⩾ > 0. Thus, we ob ain Pe Ω ; B (x0)⩽1 + C 1/3Pe Ω0;B (x0), which p o es ha Ωis an almos -minimize o he pe ime e in D. We nex p o e ha egula pa he ee bounda y Reg(∂Ω) is C∞. Theo em 5.2. — Le (u, Ω) be a solu ion o (1.1). Le D∩∂Ω = Reg(∂Ω) ∪Sing(∂Ω) be he decomposi ion o he ee bounda y om Theo em 5.1. Then, in a neighbo hood o any poin x0∈Reg(∂Ω),∂Ωis C∞- egula and he unc ion uis C∞on ∂Ω. P oo . — We ix a poin x0∈Reg(∂Ω). Wi hou loss o gene ali y, we assume x0= 0. J.É.P.—M., 2021, ome8 20 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko S ep 1. No a ion. — Fo any x∈Rd, we use he no a ion x= (x0, xd), whe e x0∈Rd−1 and xd∈R. By he C1,α egula i y o Reg(∂Ω), in B0×(−ε, ε)⊂Rd−1×R,∂Ωis he g aph o a C1,α egula unc ion η:B0→R, whe e B0is a ball in Rd−1; he se Ω coincides wi h he subg aph o ηin a neighbo hood o he o igin: B0×(−ε, ε)∩Ω = (x0, xd)∈B0×(−ε;ε) : xd< η(x0). and he ex e io no mal νΩis gi en by (5.1) νΩ=(−∇x0η, 1) p1 + |∇x0η|2, whe e ∇x0ηis he g adien o ηin he i s d−1 a iables. Le u+and u−be he es ic ions o uon he se s Ωand D Ω; since uis con inuous ac oss ∂Ω, we ha e u+=u−on ∂Ω. Mo eo e , we w i e he g adien s o u+and u−as ∇u±=∇x0u±, ∂xdu±∈Rd−1×R. S ep 2. T ansmission condi ion and C1,α egula i y o u. — In Lemma 5.3, we keep ixed he ee bounda y ∂Ωand we use e ical pe u ba ions o he unc ion u o ob ain a Robin- ype ansmission condi ion on ∂Ω. We no ice ha he ecen esul s [4, 5] imply he C1,α- egula i y o u+and u−, up o he bounda y ∂Ω. Thus, he g adien is well-de ined and he ansmission condi ions (5.2) hold in he classical sense. S ep 3. Op imali y condi ion and C2,α egula i y o Reg(∂Ω). — In Lemma 5.5 we pe - o m a ia ions o he op imal se o ind he geome ic equa ion sol ed by ∂Ω. P e- cisely, we ind ha he cu a u e o he op imal se sol es an equa ion o he o m “Mean cu a u e o ∂Ω"=F(∇u+,∇u−, u±)on ∂Ω. In pa icula , his implies ha i uis Ck,α, o some k⩾1, hen ∂Ωis Ck+1,α. S ep 4. Boo s ap. — In Lemma 5.4 we use he ecen esul s o [5] o show ha i he bounda y ∂Ωis Ck,α o some k⩾2, hen he solu ions u+and u−a e also Ck,α egula up o he bounda y ∂Ω. Finally, applying his esul (Lemma 5.4) and he esul om he p e ious s ep (Lemma 5.5), we ge ha ∂Ωis C∞. Lemma 5.3 (Robin and con inui y condi ions on ∂Ω). — Suppose ha ∂Ωis C1,α egula in he neighbo hood o he o igin. Le η:B0→R,u+and u−be as abo e. Then, o e e y x0∈B0we ha e (5.2) (∇x0η· ∇x0u+− ∇x0η· ∇x0u−=−∂xdu+−∂xdu−|∇x0η|2 p1 + |∇x0η|2∂xdu+−∂xdu−+βu = 0, whe e u+,u−and hei pa ial de i a i es a e calcula ed in (x0, η(x0)) ∈∂Ω. J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 21 P oo . — Le φ∈C∞ c(D)be a smoo h unc ion suppo ed in B0×(−ε, ε). Then, he op imali y o ugi es ha 0 = ∂ ∂  =0Jβ(u+ φ, Ω) = ZD ∂Ω 2∇u· ∇φ dx +βZ∂Ω 2uφ dHd−1 =Z∂Ω 2νΩ· ∇u+−νΩ· ∇u−+βuφ dHd−1, whe e in he las inequali y we in eg a ed by pa s u+in Ωand u−in D Ω. Since φ is a bi a y we ge ha usa is ies he Robin- ype condi ion on ∂Ω (5.3) νΩ· ∇u+−νΩ· ∇u−+βu on ∂Ω. Now, using (5.1), we can e-w i e his as (5.4) − ∇x0η· ∇x0u++∂xdu+−− ∇x0η· ∇x0u−+∂xdu−+βup1 + |∇x0η|2= 0. On he o he hand uis con inuous ac oss ∂Ω. This means ha ∇x0u+(x0, η(x0)) + ∂xdu+(x0, η(x0))∇x0η=∇x0u−(x0, η(x0)) + ∂xdu−(x0, η(x0))∇x0η. Mul iplying by ∇x0η, we ge (5.5) ∇x0η· ∇x0u++∂xdu+|∇x0η|2=∇x0η· ∇x0u−+∂xdu−|∇x0η|2, whe e u+,u−and hei pa ial de i a i es a e calcula ed in (x0, η(x0)). Pu ing o- ge he (5.4) and (5.5), we ge (5.2).  Lemma 5.4 (Smoo h bounda y ⇒smoo h unc ion). — Le (u, Ω) be a solu ion o (1.1). Suppose ha , in a neighbo hood o ze o, ∂Ωis Ck,α- egula o some k⩾1. Then, in a neighbo hood o he o igin, he unc ions u+and u−a e Ck,α up o he bounda y ∂Ω. P oo . — We a gue by induc ion. The case k= 1 ollows by [5]. We suppose ha k⩾2and ha he claim holds o k−1. Suppose ha ∂Ωis he g aph o η:B0→R, η∈Ck,α(B0), and conside he unc ions +(x0, xd) := u+(x0, xd+η(x0)) and −(x0, xd) := u−(x0, xd+η(x0)), de ined on he hal -space {xd⩾0}. We se Aη=Nd−1−(∇x0η) −∇x0η|∇x0η|2, whe e Nd−1is he null (d−1) ×(d−1) ma ix and we no ice ha Aηhas Ck−1,α egula coe icien s. Now, since u+and u−a e ha monic in Ωand D Ω, we ha e ha +and −a e solu ions o he ansmission p oblem                −di ((Id + Aη)∇ +)=0 in {xd>0} −di ((Id + Aη)∇ −)=0 in {xd<0} += −on {xd= 0} ∂xd +−∂xd −+β 2p1 + |∇x0η|2( ++ −)=0 on {xd= 0}. J.É.P.—M., 2021, ome8 22 S. Gua ino Lo Bianco, D. A. La Manna & B. Velichko We now ix k−1di ec ions i1, . . . , ik−1,ij6=d o e e y j, and we conside he unc ions w+:= ∂i1∂i2. . . ∂ik−1 +and w−:= ∂i1∂i2. . . ∂ik−1 −. We no ice ha , in {xd>0}and {xd<0} he unc ions w+and w−a e solu ions o −di (Id + Aη)∇w±+X I,J di ∂IAη∂J∇u±= 0, whe e he sum is o e all mul iindices Iand Jsuch ha he se s Iand Ja e disjoin subse s o {i1, i2, . . . , ik−1},I∪J={i1, i2, . . . , ik−1}and Iis non-emp y. In pa icula , using ha Aη∈Ck−1,α and ∇u∈Ck−2,α (since by hypo hesis u±∈Ck−1,α), we ge ha w±sol e −di (Id + Aη∇w±) + di (F±)=0 in {±xd>0}, whe e F+and F−a e C0,α con inuous unc ions (depending on i1, . . . , ik). On he o he hand, on he bounda y {xd= 0}we ha e ha w+=w−and ∂xdw+−∂xdw−+∂i1∂i2. . . ∂ik−1β(u++u−) 2p1 + |∇x0η|2= 0 on {xd= 0}. Reasoning as abo e, we no ice ha his condi ion can be w i en as ∂xdw+−∂xdw−=gon {xd= 0}, whe e gis a C0,α unc ion. Now, applying [5, Th. 1.2], we ge ha w+and w−a e C1,α egula up o he bounda y {xd= 0}. Thus, he ace u+=u−is Ck,α smoo h on {xd= 0}. Finally, he classical Schaude es ima es gi e ha u+and u−a e Ck,α on {xd⩾0}and {xd⩽0}, espec i ely.  Lemma 5.5 (Smoo h unc ion ⇒smoo h bounda y). — Le (u, Ω) be a solu ion o (1.1). Suppose ha , in a neighbo hood o ze o, ∂Ωis C1,α- egula and ha he unc ions u+and u−a e Ck,α up o he bounda y ∂Ω, o some k⩾1. Then, ∂Ωis Ck+1,α- egula in a neighbo hood o ze o. P oo . — Le ξ∈C∞ c(D;Rd)be a gi en ec o ield wi h compac suppo in Dand le Ψ be he unc ion Ψ (x) = x+ ξ(x) o e e y x∈D. Then, o small enough, Ψ :D→Dis a di eomo phism and se ing Φ := Ψ−1 , he unc ion u := u◦Φ is well-de ined and belongs o H1(D); he unc ion 7−→ ZD |∇u |2dx is di e en iable a = 0 and ∂ ∂  =0 ZD |∇u |2dx =ZD−2∇u Dξ · ∇u+|∇u|2di ξdx. I is immedia e o check ha −2∇u Dξ · ∇u+|∇u|2di ξ=di |∇u|2ξ−2(ξ· ∇u)∇uin D ∂Ω. J.É.P.—M., 2021, ome8 A wo-phase p oblem wi h Robin condi ions on he ee bounda y 23 We now ake ξ o be smoo h ou side ∂Ωand such ha ξ=φνΩon ∂Ω, whe e νΩis he ex e io no mal o ∂Ωand φ:∂Ω→Ris con inuous and wi h compac suppo . In eg a ing by pa s, we ge ∂ ∂  =0 ZD |∇u |2dx =Z∂Ω|∇u+|2(ξ·νΩ)−2(ξ· ∇u+)(νΩ· ∇u+)dHd−1 −Z∂Ω|∇u−|2(ξ·νΩ)−2(ξ· ∇u−)(νΩ· ∇u−)dHd−1, whe e u+:= uon Ω, and u−:= uon D Ω. Now, i ξ=φedand νΩ=(−∇x0η, 1) p1 + |∇x0η|2, hen p1 + |∇x0η|2|∇u+|2(ξ·νΩ)−2(ξ· ∇u+)(νΩ· ∇u+) −p1 + |∇x0η|2|∇u−|2(ξ·νΩ)−2(ξ· ∇u−)(νΩ· ∇u−) =φ|∇u+|2− |∇u−|2 −2φ∂xdu+− ∇x0η· ∇x0u++∂xdu+−∂xdu−− ∇x0η· ∇x0u−+∂xdu−. We now suppose ha x0∈Reg(∂Ω) and ha ∂Ωis he g aph o he (C1,α) unc ion η:B0→R, whe e B0is a ball in Rd−1. Taking ξ=edφand Ω = Φ (Ω), we ha e ∂ ∂  =0 Z∂Ω u2 dHd−1=∂ ∂  =0 ZB0 u2x0, η(x0)p1 + |∇x0η+ ∇x0φ|2dx0 =ZB0 u2x0, η(x0) p1 + |∇x0η|2∇x0η· ∇x0φ dx0 =ZB0 u2x0, η(x0)Hx0, η(x0)φ(x0)dx0 −2ZB0 φ(x0)ux0, η(x0)∇x0u+∂xdu∇x0η· ∇x0η p1 + |∇x0η|2dx0. In pa icula , combining hese wo compu a ions and using he op imali y o (u, Ω), we ge 0 = ∂ ∂  =0Jβ(u ,Ω ) = ZB0 βu2H(x0)φ(x0)dx0+ZB0|∇u+|2− |∇u−|2φ(x0)dx0 −ZB0 21 + |∇x0η|2(∂xdu+)2−(∂xdu−)2φ(x0)dx0 Since φis a bi a y, we ob ain ha ηis a solu ion o he p oblem −di x0∇x0η p1 + |∇x0η|2= (x0)in B0, J.É.P.—M., 2021, ome8