• Aucun résultat trouvé

Existence of solutions for semi-linear equations involving the $p$-Laplacian : the non coercive case

N/A
N/A
Protected

Academic year: 2021

Partager "Existence of solutions for semi-linear equations involving the $p$-Laplacian : the non coercive case"

Copied!
33
0
0

Texte intégral

(1)

HAL Id: hal-00842589

https://hal.archives-ouvertes.fr/hal-00842589

Submitted on 8 Jul 2013

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires

Existence of solutions for semi-linear equations involving the p-Laplacian : the non coercive case

Isabeau Birindelli, Françoise Demengel

To cite this version:

Isabeau Birindelli, Françoise Demengel. Existence of solutions for semi-linear equations involving the p-Laplacian : the non coercive case. Calculue of Variations and partial differential equations, 2004, 20 (4), pp.343-366. �hal-00842589�

(2)

Existence of solutions for semi-linear equations involving the p-Laplacian : the non coercive

case

Isabeau Birindelli

Universit`a di Roma “La Sapienza”

Fran¸coise Demengel University of Cergy-Pontoise

1 Introduction

In this paper we give necessary and sufficient conditions for the existence of solutions of the following equation

( −div(|∇u|p−2∇u) + (gλ)up−1 =f uq−1, u0 in Ω

u= 0 on ∂Ω, (1.1)

where Ω is a bounded smooth domain of IRN, 1< p < N,p < q N−ppN :=p?, f and g belong toL, andλIR. By solution of (1.1), we mean a function uW01,p(Ω) satisfying (1.1) in the weak usual sense.

In particular we shall study (1.1) considering the position ofλwith respect to the principal eigenvalue. Precisely, it is well known that the concept of

”eigenvalue” and ”eigenfunction” has been generalized by many authors to the quasi-linear setting of thep-Laplacian ∆p := div(|∇.|p−2∇.), in particular let us recall the works of Allegretto and Huang in [2], Anane in [3] and Lindqvist in [19]. We shall now state their definitions and the principal properties obtained in the works cited above.

(3)

Definition 1.1 λ1 the first ”eigenvalue” of −div(|∇.|p−2∇.) +g in W01,p(Ω) is defined by

λ1 := inf

{ψ∈W01,p(Ω),|ψ|p=1}

{

Z

|∇ψ|p+

Z

g|ψ|p}.

It is by now a classical result that there exists φ, positive in Ω for which this infimum is achieved. φ is called the ”eigenfunction” corresponding to λ1.

In particular φ satisfies

( −div(|∇φ|p−2∇φ) + (gλ1p−1 = 0 in Ω

φ = 0 on∂Ω. (1.2)

Furthermore φ is simple, i.e. any solution of (1.2) satisfies v = for some k IR. In the sequel we will normalizeφ in the Lp(Ω) norm.

Clearly for any λ < λ1 the only nonnegative solution of

( −div(|∇u|p−2∇u) + (gλ)up−1 = 0 in Ω

u= 0 on (1.3)

is u0.

On the other hand λ1 is isolated, i.e. there existsδ >0 such that for any λ in (λ1, λ1+δ) the only solution of (1.3) is u0 as well.

Our first results concern some necessary conditions for the existence of solutions.

Theorem 1.2 Suppose that there exists a nonnegative solution u 6≡ 0 of equation (1.1). Then

1) For λ < λ1, the set + defined as

+:={xΩ, f(x)>0}

is nonempty.

2) For λ > λ1, :={xΩ, f(x)<0} 6= and Rf φq <0.

3) For λ=λ1, + 6=∅, 6= and Rf φq<0.

Theorem 1.3 There exists λ0 > λ1 such that there are no non trivial non negative solutions of equation (1.1) for λ > λ0.

(4)

Theorem 1.4 Suppose that there exists λ > λ¯ 1 for which (1.1) possesses a solution. Then, (1.1) has a solution for λ∈]λ1,λ].¯

Our next result concerns the existence of solutions of equation (1.1) in the sub-critical case:

Theorem 1.5 Suppose that + and are nonempty, that p < q < p?, and

R

f φq <0. Then there exists δ > 0 such that for λ 1, λ1+δ) equation (1.1) has at least two non zero and nonnegative solutions of equation (1.1).

For λ = λ1 there exists at least one solution of (1.1) nonnegative and not identically zero.

Remark 1: The solutions are obtained as minima of the two variational problems:

αλ,q = inf

{u∈Wo1,p(Ω),R

f|u|q=−1}

{

Z

|∇u|p+

Z

(gλ)|u|p} and

µλ,q = inf

{u∈Wo1,p(Ω),R

f|u|q=1}

{

Z

|∇u|p+

Z

(gλ)|u|p}.

Indeed, if u Wo1,p(Ω) realizes αλ,q (respectively µλ,q), so does |u|, and it is easy to see that usatisfies:

−div(|∇u|p−2∇u) + (gλ)up−1 =−αλ,qf uq−1 (respectively

−div(|∇u|p−2∇u) + (gλ)up−1 =µλ,qf uq−1).

By a standard scaling argument one obtains two nonnegative solutions of equation (1.1), one being such that Rf uq > 0 and the other such that

R

f uq<0.

For simplicity of notation let αλ :=αλ,p? and µλ :=µλ,p?.

Theorem 1.6 Suppose that q = p? and that +, 6= ∅, that λ > λ1 and that Rf φp? <0. Then there exists δ >0 such that if λ 1, λ1 +δ) there exists at least one solution of equation (1.1). If moreover,

µλ < K(N, p)−psup|f|−pp?,

then, there exist at least two non zero solutions of equation (1.1).

(5)

Remark 2: As in the subcritical case, the solutions are obtained as minima of αλ and µλ.

Remark 3: According to Theorems 1.4 and 1.5 the solutions of equation (1.1) exist for an interval, (λ1,λ). On the other hand for some¯ λ ∈]λ1,λ[,¯ there may be only one solution, because forλ not close toλ1 nothing can be said about the sign ofRf uqλ whenuλ is a solution obtained by Theorem 1.4.

For p = 2 i.e. the classical Laplacian and 2 < q < 2n

n2 problem (1.1) has been extensively studied when f > 0. Since we are concerned with the case where f changes sign, let us recall the main results in that case.

Necessary and sufficient conditions for the existence of solutions for (1.1) have been given by Alama and Tarantello [1], Berestycki, Capuzzo Dolcetta and Nirenberg [5] and Ouyang [20] in the non coercive case.

Alama and Tarantello in [1] and the authors of the present paper in [6]

have studied the critical case i.e. q = n−22n . Let us also mention the very interesting work of Chen and Li in [7].

It is well known that the p-Laplacian appears in many contexts : Non- Newtonian fluids, nonlinear elasticity and reaction diffusion problems just to name a few. Indeed equation (1.1) has been extensively studied for generalp and q; in particular for q critical, existence of solutions of problem (1.1) was studied by Guedda and Veron in [14] for f 1, g(x) λ = 0. Demengel and Hebey in [10] gave existence of variational solutions whenf changes sign and the functional R|∇u|p+R(gλ)|u|p is coercive i.e. λ < λ1.

In [12], the authors study a similar problem with (gλ)up−1 replaced by cteuk−1 with k 6=p.

Always for general p but q subcritical the non coercive case was also studied by Drabek and Pohozaev in [11]; they use the fibering method to obtain some existence results forλ close toλ1. See also Pohozaev and Veron [21] for the Neumann problem.

Finally for qcritical, Drabek and Huang studied the problem in IRN [10], while Arioli and Gazzola in [4] proved existence for solutions changing sign through a linking method.

The above Theorems are the natural extension to the p-Laplacian of the results obtained in [6]. Nonetheless the proofs differ from the case p = 2.

In particular the proofs of Theorems 1.5 and 1.6 follow the approach taken by Ouyang in [20]. Although we should mention that Ouyang treats the

(6)

sub-critical case and he uses bifurcation technic that don’t hold for p6= 2.

The outline of the paper is the following. In the next section we prove the necessary conditions (i.e. Theorem 1.2 and 1.3) using among other things Picone’s identity for the p-Laplacian (cf Allegretto and Huang [2]). In the third section we prove the existence results first for the sub-critical case and then for the critical case. Finally in the last section we construct some test functions to show that the condition on µλ of Theorem 1.6 can be satisfied and easily verified.

2 Proofs of Theorem 1.2, 1.3, 1.4.

Let us recall Picone’s identity for thep-Laplacian as formulated by Allegretto and Huang in [2]. Suppose that v and w belong to W1,p(Ω) with v 0 and w >0, then

|∇v|p − ∇

vp wp−1

·σ(w)0 everywhere in Ω, for σ(w) := |∇w|p−2∇w.

Moreover if equality holds then w=kv for some constantk IR.

Proof of Theorem 1.2

Since in the case λ < λ1 the functional Iλ(u) :=

Z

|∇u|p+

Z

(gλ)|u|p is coercive the first assertion is obvious.

Let us prove 2. Suppose that λ > λ1, and let ube a nonnegative solution of (1.1) . Adapting the strict maximum principle of Vasquez, one has u >0 inside Ω . In addition, from regularity results of [13], [23], [17], [9], u is C1,α( ¯Ω), for every α [0,1[. Using once more the strict maximum principle inspired from Hopf’s lemma, as given in [24], one has the existence of some real > 0 such that φ u on ¯Ω. As a consequence, one is allowed to multiply the equation (1.1) by (u)1−qφq. Integrating by parts on Ω, one obtains

Z

f φq =

Z

σ(u).∇(u1−qφq) +

Z

(gλ)up−1u1−qφq

(7)

= (1q)

Z

|∇u|p φ u

!q

+q

Z

(σ(u).∇φ) φ u

!q−1

+

Z

(g λ)up−qφq. (2.4)

Now we multiply equation (1.2) by φq−p+1up−q and integrate over Ω;

Z

σ(φ).∇(φq−p+1up−q) +

Z

(gλ1qup−q= 0 and then

(qp+ 1)

Z

|∇φ|p φ u

!q−p

+ (pq)

Z

σ(φ).∇u φ u

!q−p+1

+ +

Z

(gλ1qup−q = 0. (2.5) Subtracting (2.4) to (2.5) , one gets

(qp+ 1)

Z

|∇φ|p φ u

!q−p

+ (pq)

Z

σ(φ).∇u φ u

!q−p+1

−q

Z

φ u

!q−1

∇φ.σ(u) + (q1)

Z

φ u

!q

|∇u|p+ λ1)

Z

φqup−q =

Z

f φq. (2.6)

Now apply Picone’s identity as follows

|∇u|p− ∇ up φp−1

!

·σ(φ)0.

Multiplying it by φuq and integrating over Ω it becomes

Z

|∇u|p φ u

!q

p

Z

∇u·σ(φ)up−q−1φq−p+1+ + (p1)

Z

|∇φ|pup−qφq−p 0. (2.7)

(8)

Similarly, exchanging the role of u and φ i.e. considering

|∇φ|p− ∇( φp

up−1)·σ(u)0 and multiplying by φuq−p one gets

Z

|∇φ|p φ u

!q−p

p

Z

φ u

!q−1

∇φ.σ(u) + (p1)

Z

φ u

!q

|∇u|p 0. (2.8) Multiply (2.8) by qp and (2.7) by qp 1 their sum gives

(qp+ 1)

Z

|∇φ|p φ u

!q−p

+ (pq)

Z

∇u·σ(φ) φ u

!q−p+1

+

−q

Z

φ u

!q−1

∇φ·σ(u) + (q1)

Z

|∇u|p φ u

!q

0. (2.9) Substracting (2.9) from (2.6) we obtain

Z

f φq+ (λλ1)

Z

φqup−q 0. (2.10) When λ > λ1, this implies that Rf φq <0 and 2) is proved.

For the proof of 3), let λ = λ1 and let u be a nonnegative solution of equation (1.1). Multiplying it by u one obtains

Z

|∇u|p+

Z

(gλ1)up =

Z

f uq.

Since the functionalIλ1 is non negative, one hasRf uq 0. Suppose that it is zero. Thenuwould be an eigenfunction for the eigenvalueλ1, which would imply that f uq−1 = 0. Then u must be zero on a set of positive measure, which contradicts the fact that u is parallel to φ >0 in Ω. We have proved that Rf uq>0, this implies that Ω+ 6=∅.

We shall now prove that Rf φq < 0, this of course implies also that 6=∅.

(9)

From the previous computations in the proof of 2), and precisely from (2.6) with λ=λ1 and from (2.9), we obtain that

(qp+ 1)

Z

|∇φ|p φ u

!q−p

+ (pq)

Z

∇u·σ(φ) φ u

!q−p+1

−q

Z

φ u

!q−1

∇φ·σ(u) + (q1)

Z

|∇u|p φ u

!q

+

=

Z

f φq. (2.11)

As a consequence Rf φq 0. Suppose by contradiction that Rf φq = 0, then the left hand side of the previous identity is zero. Recalling (2.8) and (2.9)the left hand side is a sum of two nonnegative quantities, hence they must be both null. Therefore we have obtained that

Z

|∇φ|p φ u

!q−p

p

Z

φ u

!q−1

∇φ.σ(u) + (p1)

Z

φ u

!q

|∇u|p = 0 (2.12) and

Z

|∇u|p φ u

!q

−p

Z

∇u·σ(φ)up−q−1φq−p+1+ + (p1)

Z

|∇φ|pup−qφq−p = 0. (2.13) Clearly (2.12) and (2.13) imply that

|∇u|p− ∇ up φp−1

!

·σ(φ) = 0 and

|∇φ|p− ∇ φp up−1

!

·σ(u) = 0.

Each of these identities implies that φ is parallel to u. Then u is an eigen- function. This implies thatf uq−1 is identically zero which is a contradiction.

(10)

Proof of Theorem 1.3.

Let B be a ball on which f >0,B ⊂⊂+. Let then (ψ, µ?) be the non zero and non negative normalized solution, of

( −∆pψ+ (−µ?p−1 = 0 in B

ψ = 0 on ∂B.

Suppose that a solution of equation (1.1) exists forλsuch that|g|? <

λ, u 0 and non identically zero. On B, by the strict maximum principle of Vasquez, u >0. Using Picone’s identity, one has

|∇ψ|p− ∇ ψp up−1

!

.σ(u)0 in B, hence, integrating over B

0

Z

B

?p+

Z

B

(gλ)ψp (2.14)

here, we have used the fact that ψ = 0 on ∂B and the equation verified by u, since

−∆pu+ (gλ)up−1 =f uq−1 0 onB. (2.14) of course contradicts the choice of λ.

Proof of Theorem 1.4.

Let ¯λ be such that λ1 < λ¯ and take λ ∈]λ1,¯λ[. Let ¯u be a solution of (1.1) for ¯λ. Then ¯uis a supersolution of (1.1) for λ. Indeed

−∆pu¯+ (gλ)¯up−1 =fu¯q−1 + (¯λλ)¯up−1 fu¯q−1

and ¯u = 0 on the boundary. On another hand, taking small enough, φ is a subsolution, since

−∆p(φ) + (gλ)(φ)p−1 = (λ1λ)p−1φp−1 f q−1φq−1,

(using p < q and (λ1 λ)p−1φp−1 <0). Moreover, using strong maximum principle of Vasquez and regularity results, one can choose small enough in order to have ¯uφ. Finally we use the following Proposition, whose proof can be found in the appendix and is a mere adaptation of the classical sub and super solution for p= 2. (see e.g. [15], see also [22]):

(11)

Proposition 2.1 Suppose that f(x, t) = a(x)|t|q−2t +b(x)|t|p−2t with 1 <

p < q with a and b two continuous and bounded functions on Suppose that

¯

u is a weak supersolution for −∆pu+f(x, u), u¯= 0 on ∂Ω, and that u is a weak subsolution with u= 0 on Ω. Suppose that there exists some constant c and C such that

−∞< c uu¯C < +∞

Then, there exists a solution u between u and u¯

Using this Proposition with f(x, u) = (gλ)up−1f uq−1, and u=φ, one obtains that there exists a solution which is such that

φ uu.¯

3 Existence of solutions

Proof of Theorem 1.5

This proof is inspired by the arguments used in [20]. We begin with the subcritical case. Suppose thatq < p?. Let us recall the following notations:

λ?q = inf

{u∈W01,p(Ω), |u|pp=1,R

f uq=0}

{

Z

|∇u|p+

Z

(gλ1)|u|p}

αλ,q = inf

{u, R

f|u|q=−1}

{

Z

|∇u|p+

Z

(gλ)|u|p} (3.15) and

µλ,q = inf

{u, R

f|u|q=1}

{

Z

|∇u|p+

Z

(gλ)|u|p}. (3.16) LetIλ(u) := R|∇u|p +R(gλ)|u|p.

We will prove the following facts 1. λ?q >0.

2. For λ ∈]λ1, λ1+λ?q[, αλ,q<0 and it is achieved; αλ1,q = 0.

3. For λ ∈]λ1, λ1+λ?q[, µλ,q >0 and it is achieved. Moreoverµλ1,q >0.

(12)

Proof of 1.

By the definition of λ1, λ?q 0. Suppose by contradiction that λ?q = 0.

Let (un) be a minimizing sequence. Since |∇|un|| = |∇un|, one can assume thatun0. Since|un|p = 1 andR|∇un|p+R(g−λ1)upn 0, thenR|∇un|p is bounded; hence (un) is bounded inW01,p. Extracting from it a subsequence and passing to the limit, one gets that there exists some u 0, weak limit of (un) in W1,p(Ω), such that

Z

|∇u|p+

Z

(g λ1)up 0. (3.17) Clearly (3.17) implies that

Z

|∇u|p+

Z

(gλ1)up = 0.

and then u is an eigenfunction for λ1 and then it is parallel to φ. Moreover u W01,p, R|u|p = 1 and Rf uq = 0, which contradicts the assumption

R

f φq <0. Finally λ?q >0.

Proof of 2.

In order to prove that αλ,q < 0 for λ > λ1, let us take, as an admissible function, v = φ

(−Rf φq)1q

. We then have αλ,q Iλ(v) = 1

(−Rf φq)pqIλ(φ) = 1

(−Rf φq)pq1λ)<0.

Now we will check that

αλ,q >−∞.

If not, there would exist a subsequence (ui), ui 0 for all i, such that

R

f uqi =−1 and Iλ(ui)→ −∞. Clearly |ui|p +∞since lim

Z

(gλ)upi αλ,q. Let wi = ui

|ui|p. One has Rf wqi 0, and (wi) is bounded in W01,p(Ω), since

|wi|p = 1 and

Z

|∇wi|p+

Z

(gλ)wip = Iλ(ui)

|ui|pp

0 implies

Z

|∇wi|p ≤ |gλ|.

(13)

Then, there exists a subsequence still denoted (wi), such thatwi * w weakly in W1,p(Ω). Observe that

Z

|w|p = 1 andIλ(w)0.

This contradicts the definition of λ, since Rf wq = 0 and λ ∈]λ1, λ1+λ?q[.

We have proved that αλ,q >−∞.

We shall now see that αλ,q is achieved. Let (un),un 0 be a minimizing sequence for αλ,q i.e.

Z

|∇un|p+

Z

(gλ)upnαλ,q,

Z

f uqn =−1.

Let us prove first that |un|p is bounded. If not, one can argue as previously by considering wn = un

|un|p. It is easy to see that (wn) converges weakly in W1,p(Ω), up to a subsequence, towards some function w 0 which satisfies

R

f wq= 0, |w|p = 1 and

Z

|∇w|p+

Z

(gλ)wp = 0.

This contradicts the definition of λ. Hence R|un|p is bounded, and so is

R

|∇un|p. By extracting from (un) a subsequence, one obtains that there exists uW01,p, u 0, such thatRf uq =−1 and by lower semi-continuity of the semi-norm |∇u|p with respect to the weak topology,

Z

|∇u|p+

Z

(gλ)up αλ,q.

Finally using the definition of αλ,q, u is a minimizer for αλ,q, hence it is a nonzero solution of

−div(|∇u|p−2∇u) + (gλ)up−1 =−αλ,qf uq−1. Proof of 3.

Acting as we did for αλ,q one can prove that µλ,q > −∞. We are now going to check that µλ,q is achieved.

(14)

Indeed, let un be a sequence such that un0,

Z

|∇un|p+

Z

(gλ)upnµλ,q,

Z

f uqn = 1.

Suppose that |un|p → ∞. Then consideringwn= un

|un|p one gets, by passing to the limit that there exists w 0, a weak limit of (wn) in W1,p(Ω), such

that Z

|∇w|p+

Z

(g λ)wp 0

and Rf wq = 0, which contradicts the assumption λ ∈]λ1, λ1 +λ?q[. Then (un) is bounded and we pass to the limit to obtain

Z

|∇u|p+

Z

(gλ)up =µλ,q and Rf uq = 1. Hence µλ,q is achieved.

Forλ=λ1, µλ1,q 0, but since it is achieved, ifµλ1,q = 0, we would have an eigenfunction usuch that Rf uq= 1, which contradicts the assumptions.

Then µλ1,q >0.

For λ > λ1 letuq 0 which realizes the minimum in µλ,q. Then :

−∆puq+ (gλ)up−1q =µλ,qf uq−1q .

Using the procedure of the proof of Theorem 1.2 for uq, inequality (2.10) becomes

µλ,q

Z

f φq+ (λλ1)

Z

φqup−qq 0.

UsingRf φq <0 and λλ1 >0, one gets µλ,q >0.

Let us now state and prove some results concerning αλ,q and µλ,q. Lemma 3.1 The following convergences hold:

λ→λlim1

αλ,q =αλ1,q = 0, (3.18)

λ→λlim1

µλ,q =µλ1,q (3.19)

(15)

Lemma 3.2 1. λ?p? limq→p?λ?q limq→p?λ?q :=λ? >0.

2. For λ1 λ < λ1 +λ?, then 0 limq→p?µλ,q limq→p?µλ,q µλ(=

µλ,p?).

3. For λ close to λ1, αλ(=αλ,p?)>−∞ and limq→p?αλ,q αλ. Proof of Lemma 3.1

Suppose by contradiction that (3.18) does not hold, then there exist some number α < 0 and a sequence of λ IR, λ λ1, and (uλ) Wo1,p(Ω) such

that Z

|∇uλ|p+

Z

(gλ)|uλ|p α.

Moreover one can assume that uλ 0. If (uλ) is bounded, we may extract from it a subsequence weakly convergent to some uW01,p, such that

Z

|∇u|p+

Z

(gλ1)up α <0, which is absurd.

On the other hand if (uλ) diverges we can normalize it and then we obtain a sequence (wλ) such that R|wλ|p = 1. By extracting a subsequence, there exists w0, such that R|w|p = 1, Rf wq = 0 and

Z

|∇w|p+

Z

(g λ1)wp 0.

This would imply that w is parallel toφ which is absurd since Rf φq <0.

Let us now prove (3.19). Let us define ¯µq := limλ→λ1µλ,q. One already has ¯µq µλ1,q. Let uλ which satisfies uλ 0 and

−∆puλ + (gλ)up−1λ =µλ,qf uq−1λ (3.20)

Z

f uqλ = 1.

As we did above , one can prove that (uλ) is bounded in the W1,p norm. By extracting a subsequence, one gets by passing to the limit when λλ1

Z

|∇u|p+

Z

(gλ1)up µ¯q

and u 0, Rf uq = 1. This clearly implies that ¯µq µλ1,q and gives the required result.

Références

Documents relatifs

Linear elliptic equations in unbounded domains possess the Fredholm property if and only if, in addition to the conditions mentioned above, limiting operators are invertible

Elliptic problems in bounded domains with a sufficiently smooth boundary satisfy the Fredholm property if the ellipticity condition, proper ellipticity and Lopatinskii conditions

We note that if the pairs of parameters (b i , c i ) lie on a line going through the origin, then the functional equation can be reduced to the form ∑ n i=1 a i f (d i x) = 0 which

If f is solution of a linear differential equation with polynomial coefficients, then the Chebyshev coefficients are cancelled by a linear recurrence with polynomial

For a general class of autonomous quasi-linear elliptic equations on R n we prove the existence of a least energy solution and show that all least energy solutions do not change

NIRENBERG, Positive Solutions of Nonlinear Elliptic Equations Involving Critical Sobolev Exponents, Comm. SPRUCK, Variational Problems with Critical Growth and Positive

The purpose of this section is to introduce some preliminaries and prove an existence theorem for blow-up solutions assuming the existence of ordered super-solution and

In this paper we investigate existence and characterization of non-radial pseudo-radial (or separable) solutions of some semi-linear elliptic equa- tions on symmetric