# Nodal solutions of weighted indefinite problems

This paper analyzes the structure of the set of nodal solutions of a class of one-dimensional superlinear indefinite boundary values problems with an indefinite weight functions in front of the spectral parameter. Quite astonishingly, the associated high order eigenvalues might not be concave as it is the lowest one. As a consequence, in many circumstances the nodal solutions can bifurcate from three or even four bifurcation points from the trivial solution. This paper combines analytical and numerical tools. The analysis carried over on it is a paradigm of how mathematical analysis aids the numerical study of a problem, whereas simultaneously the numerical study confirms and illuminate the analysis.

## Authors

• 3 publications
• 2 publications
05/19/2020

### Global bifurcation diagrams of positive solutions for a class of 1-D superlinear indefinite problems

This paper analyzes the structure of the set of positive solutions of a ...
11/26/2020

### Accurate Spectral Collocation Computation of High Order Eigenvalues for Singular Schrödinger Equations

We are concerned with the study of some classical spectral collocation m...
10/24/2021

### Spurious solutions for high order curl problems

We investigate numerical solutions of high order curl problems with vari...
05/27/2020

### Oligopoly Dynamics

The present notes summarise the oligopoly dynamics lectures professor Lu...
11/12/2019

### Numerical solutions for a class of singular boundary value problems arising in the theory of epitaxial growth

The existence of numerical solutions to a fourth order singular boundary...
12/08/2021

### Sparse Representations of Solutions to a class of Random Boundary Value Problems

In the present study, we consider sparse representations of solutions to...
08/15/2019

### On boundedness and growth of unsteady solutions under the double porosity/permeability model

There is a recent surge in research activities on modeling the flow of f...
##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1. Introduction

In this paper we analyze the nodal solutions of the one-dimensional nonlinear weighted boundary value problem

 {−u′′−μu=λm(x)u−a(x)u2in(0,1),u(0)=u(1)=0, (1.1)

where are functions that change sign in and are regarded as bifurcation parameters. More precisely, is the primary parameter, and the secondary one. All the numerical experiments carried out in this paper have been implemented in the special case when

 (1.2)

because this is the weight function considered by López-Gómez and Molina-Meyer in [28] to compute the global bifurcation diagrams of positive solutions there in. In this paper we pay a very special attention to the particular, but very interesting, case when

 m(x)=sin(nπx)

for some integer .

Up to the best of our knowledge, this is the first paper where the problem of the existence and the structure of the nodal solutions of a weighted superlinear indefinite problem is addressed when changes of sign. The existence results of large solutions of Mawhin, Papini and Zanolin [39] required , as well as the results of López-Gómez, Tellini and Zanolin [35], where the attention was focused on the problem of ascertaining the structure of the set of positive solutions. Most of the available results on nodal solutions dealt with the special cases when , and is a positive function with (see Rabinowitz [40, 41, 42]), or with the degenerate case when is a continuous positive function such that (see López-Gómez and Rabinowitz [36, 37, 38], and López-Gómez, Molina-Meyer and Rabinowitz [33]). In strong contrast with the classical cases when , in the degenerate case when with the set of nodal solutions might consist of two, or even more, components, depending on the nature of the weight function (see [33] and [38] for any further required details). Nevertheless, as for the special choice given by (1.2), is negative in the intervals and , while it is positive in the central interval , this is the first time that the problem of analyzing the structure of the nodal solutions in this type of superlinear indefinite problems is addressed.

A natural strategy for constructing the nodal solutions of (1.1) with interior zeroes, or nodes, consists in linearizing (1.1) at the trivial solution,

, and then searching for the eigenvalues of the linearization having an associated eigenfunction with exactly

interior nodes in , for as these values of the parameters will provide us, through the local bifurcation theorem of Crandall and Rabinowitz [12], with all the small nodal solutions of (1.1) bifurcating from . This strategy provides us in a rather natural way with the linear weighted eigenvalue problem

 {−φ′′−μφ−λm(x)φ=σφin(0,1),φ(0)=φ(1)=0. (1.3)

By the Sturm–Liouville theory, the problem (1.3) has a sequence of eigenvalues

 Σn(λ,μ):=σn[−D2−μ−λm(x);(0,1)],n≥1,

which are algebraically simple. Moreover, associated with each of them there is an eigenfunction, , with , unique up to a multiplicative constant, with exactly interior nodes, necessarily simple, in . By uniqueness,

 Σn(λ,μ):=σn[−D2−λm(x);(0,1)]−μ,n≥1. (1.4)

It turns out that the set of all the possible bifurcation points from to solutions of (1.1) with interior zeroes are provided by the values of and for which

 Σn(λ,μ)=0.

So, the huge interest in analyzing them. Throughout this paper, we will denote

 Σn(λ):=Σ(λ,0)=σn[−D2−λm(x);(0,1)],n≥1. (1.5)

Then,

 Σn(λ,μ)=Σn(λ)−μ

and for all . Based on a classical result of Kato [21] on perturbation from simple eigenvalues, for every , is analytic in . A proof of this can be easily accomplished from [26, Ch. 9] and Section 5 of Antón and López-Gómez [2], where the result was established when . An extremely important property of is its strict concavity with respect to the parameter (see Berestycki, Nirenberg and Varadhan [5], Cano-Casanova and López-Gómez [10] and Chapter 9 of [26]). According to it, for all , , for all , and

 Σ′′1(λ)<0for allλ∈R. (1.6)

Since , this property entails that, for every , consists of two values of ,

 λ−≡λ−(μ)<0<λ+≡λ+(μ),

which are the unique bifurcation values to positive solutions from of (1.1) (see López-Gómez and Molina-Meyer [28]). Even dealing with general second order elliptic operators under general mixed boundary conditions of non-classical type, the strict concavity of relies on the strong ellipticity of the elliptic operator (see, e.g., Chapter 8 of [26]).

For analytic semigroups the spectral mapping theorem holds (see, e.g., [3, 4]), i.e,

Thus, the spectral radius of the associated semigroup is given through the formula

 ϱ(λ):=spr(eD2+λm)=e−σ1(−D2−λm;(0,1))=e−Σ1(λ),λ∈R.

Hence, is logarithmically convex, which is a classical property going back to Kato [20], because is convex. Rather astonishingly, there are examples of weight functions for which none of the remaining eigenvalues , , is concave with respect to . Figure 1 shows one of these examples for the special choice .

In this case, is the unique eigencurve which is concave, for as the remaining ones, , , are far from concave. Indeed, all of them are symmetric functions of , with a quadratic local minimum at , as illustrated by Figure 1. This fact has dramatic implications from the point of view of the structure of the set of nodal solutions of the problem (1.1). Indeed, setting

 μn=maxλ∈RΣn(λ),n≥1, (1.7)

it becomes apparent that for all and, hence, for every and any , consists of two negative eigenvalues, , and two positive eigenvalues such that

 0<λ+[2,n](μ)=−λ−[2,n](μ)<λ+[1,n](μ)=−λ−[1,n](μ).

Therefore, for this range of ’s we expect that the solutions with interior nodes of (1.1) will bifurcate from the trivial solution at each of the four values

 λ=λ±[i,n],i=1,2.

By simply having a look at Figure 1, it is easily realized that

 λ±[2,n]((nπ)2)=0.

Moreover,

 λ−[1,n](μn)=λ−[2,n](μn)<0<λ+[2,n](μn)=λ+[1,n](μn),

at least for .

As illustrated by Figure 2, the number of eigencurves, , , which are concave in might vary with the weight function . Indeed, when , it turns out that not only but also is strictly concave, while the remaining eigencurves, , with , are not concave. Similarly, when , then are concave for , while they are not concave for .

Quite astonishingly, as suggested by our numerical computations, the more wiggled is the higher number of modes is concave. This astonishing feature might have some important implications in quantum mechanics.

The distribution of this paper is as follows. Section 2 studies some global properties of the eigencurves for all and analyzes their concavities in the special case when, for some ,

 m(x)=sin(2kπx),x∈[0,1]. (1.8)

Section 3 provides us some global bifurcation diagramas of nodal solutions of (1.1) with one and two interior nodes, which are superimposed to the global bifurcation diagrams of positive solutions of López-Gómez and Molina-Meyer [28]. Finally, in Section 4 we describe, very shortly, the numericical schemes used to get the global bifurcation diagrams of Section 3.

## 2. Some global properties of the nodal eigencurves Σn(λ)

Throughout this paper, for any given with and every continuous function , we denote by , , the -th eigenvalue of the eigenvalue problem

 {−φ′′+q(x)φ=σφin(r,s),φ(r)=φ(s)=0. (2.1)

The next properties are well known (see, e.g., [9]):

1. Monotonicity of with respect to : If satisfy , then

 σn[−D2+q(x);(r,s)]<σn[−D2+~q;(r,s)]for% alln≥1.
2. Monotonicity of with respect to the interval: If , then

 σn[−D2+q;(r,s)]<σn[−D2+q;(α,β)]for % alln≥1.

Based on these properties, as suggested by Figures 1 and 2, the next result holds.

###### Proposition 2.1.

Suppose that there exist such that , i.e., changes the sign in . Then, for every ,

 limλ↓−∞Σn(λ)=−∞,limλ↑∞Σn(λ)=−∞. (2.2)
###### Proof.

Consider a sufficiently small such that

 Jε:=[x+−ε,x++ε]⊂(0,1),minJεm=mL>0.

Then, by the monotonicity properties of , for every and , we have that

 Σn(λ) =σn[−D2−λm(x);(0,1)]<σn[−D2−λm(x);Jε] <σn[−D2−λmL;Jε]=σn[−D2;Jε]−λmL=(nπ2ε)2−λmL.

Thus, letting , the second relation of (2.2) holds. The first one follows by applying this result to the weight function . This ends the proof. ∎

The fact that all the eigencurves plotted in Figures 1 and 2 are symmetric about the ordinate axis is a direct consequence of the next general result, because

 sin(2kπ(1−x))=−sin(2kπx),

for all integer and .

###### Proposition 2.2.

Suppose that is a continuous function in such that

 m(1−x)=−m(x)for allx∈[0,1]; (2.3)

this holds under condition . Then, for all and any integer . In particular,

 ˙Σn(0)=0for alln≥1, (2.4)

where we are denoting .

###### Proof.

Since , either there exists such that , or for some . Suppose the first alternative occurs. Then, by (2.3), we also have that

 m(1−x+)=−m(x+)<0

Pick an integer , a real number , and let be an eigenfunction associated to . Then, possesses zeros in , , and

 −ϕ′′n(x)=λm(x)ϕn(x)+Σn(λ)ϕn(x)

for all . Thus, setting

 ψn(x):=ϕn(1−x),x∈[0,1],

it is easily seen that

 ψ′n(x):=−ϕ′n(1−x),ψ′′n(x)=ϕ′′n(1−x),x∈[0,1],

and hence, for every ,

 −ψ′′n(x)=−ϕ′′n(1−x) =λm(1−x)ϕn(1−x)+Σn(λ)ϕn(1−x) =λm(1−x)ψn(x)+Σn(λ)ψn(x) =−λm(x)ψn(x)+Σn(λ)ψn(x).

Consequently, is an eigenfunction associated to with interior zeros. Therefore, by the uniqueness of , it becomes apparent that

 Σn(−λ)=Σn(λ)for allλ∈R.

Since is an analytic function of , necessarily . This ends the proof. ∎

By having a glance at Figure 3, it is easily realized that the function might not be an even function of if condition (2.3) fails.

The next result establishes that, as already suggested by Figures 1 and 2, the nodal eigencurves, , cannot be concave for the choice (1.8) if . We conjecture that, in general, for that particular choice, is concave if . Therefore, should be concave if, and only if, . But the analysis of the concavity when for the choice (1.8) remains outside the general scope of this paper.

###### Theorem 2.1.

Assume (1.8) for some integer . Then, as soon as ,

 ¨Σn(0)>0for alln≥k+1. (2.5)

Therefore, by (2.4), is a local minimum of and, in particular, cannot be concave.

###### Proof.

Since is algebraically simple for all , we already know that is analytic, by some well known perturbation results of Kato [20]. Moreover, the eigenfunction associated to , denoted by , can be chosen to be analytic in by normalizing it so that

 ∫10φ2[n,λ](x)dx=12. (2.6)

By definition, and

 −φ′′[n,λ](x)=λm(x)φ[n,λ](x)+Σn(λ)φ[n,λ](x)for allx∈(0,1). (2.7)

Thus, since , particularizing (2.7) at and taking into account (2.6), it becomes apparent that actually is an analytic perturbation of the eigenfunction

 φ[n,0](x)=sin(nπx),x∈[0,1].

Moreover, differentiating (2.7) with respect to yields

 −˙φ′′[n,λ](x)=λm˙φ[n,λ]+mφ[n,λ]+˙Σn(λ)φ[n,λ]+Σn(λ)˙φ[n,λ]in(0,1). (2.8)

Thus, since and , particularizing (2.8) at shows that solves the problem

 {[−D2−(nπ)2]u=mφ[n,0]in(0,1),u(0)=u(1)=0. (2.9)

In order to find out , we first determine the general solution of the linear inhomogeneous equation

 [−D2−(nπ)2]u=m(x)sin(nπx). (2.10)

To get it, we will set in order to vary coefficients in the first order system associated to (2.10),

 (u′v′)=(01−(nπ)20)(uv)+(0−m(x)sin(nπx)). (2.11)

Since

 W(x):=(cos(nπx)sin(nπx)−nπsin(nπx)nπcos(nπx))

is a fundamental matrix of solutions for the homogeneous linear system associated to (2.11), the change of variable

 (uv)=W(x)(c1(x)c2(x))

transforms (2.11) into the equivalent system

 W(x)(c′1(x)c′2(x))=(0−m(x)sin(nπx)),

whose solution, according to Cramer rule, is given through

 c′1(x)=1nπm(x)sin2(nπx),c′2(x)=−1nπm(x)sin(nπx)cos(nπx).

Thus,

 c1(x) =1nπ∫x0m(s)sin2(nπs)ds+A, c2(x) =−1nπ∫x0m(s)sin(nπs)cos(nπs)ds+B,

for some constants . Therefore, the general solution of (2.10) is given by

 u(x) =cos(nπx)c1(x)+sin(nπx)c2(x) =cos(nπx)(A+1nπ∫x0m(s)sin2(nπs)ds) +sin(nπx)(B−1nπ∫x0m(s)sin(nπs)cos(nπs)ds) =Acos(nπx)+Bsin(nπx)+p(x),

where

 p(x):=1nπ∫x0m(s)sin(nπs)sin[nπ(s−x)]ds,x∈[0,1], (2.12)

is a particular solution of (2.10). It is the solution obtained by making the choice . Obviously, . Moreover, by (1.8),

 p(1) =∫10m(s)sin(nπs)sin[nπ(s−1)]ds =(−1)n∫10sin(2kπs)sin2(nπs)ds=0,

because the integrand,

 θ(s):=sin(2kπs)sin2(nπs),s∈[0,1],

satisfies for all

and hence, it is odd about

. As we are interested in solving (2.9), we should make the choice

 0=u(0)=A+p(0)=A.

Thus,

 ˙φ[n,0](x)=Bsin(nπx)+p(x),x∈[0,1],

for some constant . To determine , we can proceed as follows. Differentiating (2.6) with respect to and particularizing the resulting identity at yields

 0=∫10φ[n,0](x)˙φ[n,0](x)dx=B∫10sin2(nπx)dx+∫10sin(nπx)p(x)dx.

Consequently,

 B=−2∫10sin(nπx)p(x)dx

and therefore,

 ˙φ[n,0](x)=−2(∫10sin(nπs)p(s)ds)sin(nπx)+p(x),x∈[0,1]. (2.13)

To find out , we can differentiate with respect to the identity (2.8). After rearranging terms, this provides us with the identity

 [−D2−λm−Σn(λ)]¨φ[n,λ]=2m˙φ[n,λ]+2˙Σn(λ)˙φ[n,λ]+¨Σn(λ)φ[n,λ].

Thus, particularizing at yields

 [−D2−(nπ)2]¨φ[n,0]=2m˙φ[n,0]+¨Σn(0)φ[n,0] (2.14)

and hence, multiplying (2.14) by and integrating in it is apparent that

 ¨Σn(0)=−4∫10m(x)˙φ[n,0](x)φ[n,0](x)dx. (2.15)

Therefore, substituting (2.13) into (2.15) and using (1.8) yields

 ¨Σn(0) =−4∫10m(x)φ[n,0](x)p(x)dx =−4∫10sin(2kπx)sin(nπx)[1nπ∫x0sin(2kπs)sin(nπs)sin(nπ(s−x))ds]dx.

Finally, we need the trigonometric formulas

 sinxsiny = 12[cos(x−y)−cos(x+y)], (2.16) sinxcosy = 12[sin(x−y)+sin(x+y)], (2.17)

to simplify the integrands arising in integrals of . First, we will ascertain the function . For this, we use the formula (2.16) on and then the formula (2.17) to simplify the integrand in . Then, integrating yields

 p(x)=−18π2[cos(πx(2k−n))k(n−k)+cos(πx(2k+n))k(n+k)−ncos(nπx)k(n2−k2)]. (2.18)

After substituting (2.18) into the formula for , we can again use the formulas (2.16) and (2.17) to simplify the underlying integrands, which can then be directly integrated. The result can be simplified to get the final formula

 ¨Σn(0)=14π2(n2−k2).

Obviously, if , and therefore . Hence, the eigencurves for are convex in a neighborhood of and thus they cannot be globally concave. ∎

## 3. Global bifurcation of nodal solutions

Since , for every the set consists of two points,

 λ−(μ)<0<λ+(μ),

such that

 limμ↑π2λ±(μ)=0.

Moreover, owing to Theorem 9.4 of [26],

 ˙Σ1(λ−(μ))>0and˙Σ1(λ+(μ))<0.

Thus, by the main theorem of Crandall and Rabinowitz [12] (one can see also Chapter 2 of [25]), are the unique bifurcation values of to positive solutions of (1.1) from . The first plot of Figure 1 of López-Gómez and Molina-Meyer [28] shows one of those bifurcation diagrams for the special choice (1.2) of with

 m(x)=sin(2πx),x∈[0,1]. (3.1)

Trying to complement the numerical experiments of [28] with our new findings here, all the numerical experiments of this section has been carried out for this special choice of . As grows up to reach the critical value , the set of positive solutions of (1.1) bifurcating from consists of one single closed loop bifurcating from at the single point . These loops, separated away from , are persistent for a large range of values of , until they shrink to a single point before disappearing at some critical value of the parameter (see [28, Fig. 1]).

According to Theorem 2.1, is not concave if (3.1) holds, which is clearly illustrated by simply looking at the plot of superimposed in Figure 1. This feature has important implications concerning the structure of the set of 1-node solutions of (1.1). Indeed, according to the plot of , for every , the set consists of two single values with and . Thus, according to [26, Th. 9.4], the transversality condition of Crandall and Rabinowitz [12] holds at . Thus, an analytic curve of 1-node solutions of (1.1) emanates from at each of these values of , . Figure 3(a) shows the plots of these two curves for the value of the parameter . Our numerical experiments suggest that they are separated away from each other. In this bifurcation diagram, as well as in all the remaining ones, we are representing the values of the parameter , in abscisas, versus the -norm of the computed solutions, in ordinates. So, each point on the curves of the bifurcation diagrams, , represent a value of and a nodal solution of (1.1) for that particular value of .

When grows up to reach the critical value , the two previous components become closer and closer until they meet at at , where the set of bifurcation points to 1-node solutions from consists of the points plus . This is the situation sketched by Figure 3(b), where we have plotted the global bifurcation diagram computed for

 μ=39.6>39.4786∼(2π)2.

When , where is given by (1.7), the set consists of four values: two negative, , plus two positive, . Moreover, by Proposition 2.2, it is apparent that

 0<λ+[2,2](μ)=−λ−[2,2](μ)<λ+[1,2](μ)=−λ−[1,2](μ).

Furthermore, as suggested by our numerical experiments,

 ˙Σ2(λ−[1,2](μ))>0,˙Σ2(λ−[2,2](μ))<0,˙Σ2(λ+[2,2](μ))>0,˙Σ2(λ+[1,2](μ))<0.

Thus, again the transversality condition of [12] holds at each of these critical values of the parameter . Therefore, (1.1) should possess four analytic curves filled in by 1-node solutions bifurcating from at each of these critical values of the parameter . Figure 3(c) shows the global bifurcation diagram of 1-node solutions bifurcating from these four bifurcation points that we have computed for the choice . Once again, the set of 1-node solutions consists of two components.

Actually, as soon as the transversality condition of Crandall and Rabinowitz [12] holds, the generalized algebraic multiplicity of Esquinas and López-Gómez [15, 25], , equals 1 and hence, thanks to Theorem 5.6.2 of López-Gómez [25], the Leray–Schauder index of , as a solution of (1.1), changes as crosses each of these values. Therefore, each of the components of the set of non-trivial solutions of (1.1) emanating from at each of these critical values of the primary parameter satisfies the global alternative of Rabinowitz [40], i.e., either it is unbounded in

, or it meets the trivial solution in, at least, two of these singular values.

Each of the two components plotted in Figure 3(c) bifurcates from two different points of and, according to our numerical experiments, both seem to be unbounded. The problem of ascertaining their precise global behavior remains open in this paper. As increases and crosses the critical value , these two components abandone the trivial curve and stay separated away from the trivial solution. So, they became isolas. Figure 3(d) shows the plots of these components for the choice . In Figure 5 we have plotted some distinguished solutions with 1-node along some of the pieces of the global bifurcation diagrams already plotted in Figure 4. Precisely, Figure 4(b) shows a series of solutions with one node along the bifurcation diagram plotted on Figure 4(a), which is a magnification of a piece of the left component of Figure 3(a), and Figure 4(d) shows a series of solutions with one node along the bifurcation diagram plotted in Figure 4(c), which is a magnification of a piece of the left component plotted in Figure 3(d). The colors of each of these 1-node solutions corresponds with the color of the piece of the bifurcation diagram on the left where they are coming from.

Similarly, according to Theorem 2.1, for the special choice (3.1), the third eigencurve, , is far from concave if (3.1) holds. This becomes apparent by simply having a look at the plot of superimposed in Figure 1. According to it, for every , the set consists of two negative eigenvalues, , plus two positive eigenvalues, . Moreover, by Proposition 2.2,

 0<λ+[2,3](μ)=−λ−[2,3](μ)<λ+[1,3](μ)=−λ−[1,3](μ)

and, according to our numerical experiments,

 ˙Σ2(λ−[1,3](μ))>0,˙Σ2(λ−[2,3](μ))<0,˙Σ2(λ+[2,3](μ))>0,˙Σ2(λ+[1,3](μ))<0.

Thus, the transversality condition of [12] holds at each of these critical values. Therefore, owing to the local bifurcation theorem of [12], an analytic curve of 2-node solutions emanates from at each of these four singular values of . The first three plots of Figure 6 show these curves for three different values of the secondary parameter . Namely: , and , respectively. All these values of are bellow . The last plot of Figure 6 has been computed for and shows three components of 2-node solutions separated away from . For this value of no solution with 2 interior nodes can bifurcate from .

More precisely, at the problem (1.1) possesses three components of solutions with two interior nodes. Two of them bifurcating from at and , respectively, and the third one linking with . According to our numerical experiments these components are unbounded in , and are persistent for all further value of bellow some critical value, , where the three components meet. Thus, for there is a component of the set of non-trivial solutions of (1.1) bifurcating from at four different values of : and . The plot in Figure 5(b) shows the corresponding global bifurcation diagram for , a value of slightly greater than , where the three components of set of non-trivial solutions are very close. By comparison with the global bifurcation diagram for , it becomes apparent that a global imperfect bifurcation phenomenon has happened at the critical value . As a consequence of this imperfect bifurcation one of the components bifurcating from links with , another links with , while the third one remains separated away from . Actually, the latest one is separated away from zero for any further value of . Therefore, there have occurred a sort of reorganization in components of the set of 2-node solutions of (1.1) as the parameter crossed the critical value . The pictures in Figures 5(c), 5(d) show the plots of the corresponding components for and , where the previous bifurcations from of these components are lost. For larger values of the solutions along these three components become larger and larger and it remains an open problem to ascertain whether, or not, (1.1) can admit some 2-node solution for sufficiently large . Figure 7 shows the plots of some distinguished 2-node solutions of (1.1) along some of the curves of the bifurcation diagrams plotted in Figure 6.

Finally, Figure 8 superimposes the global bifurcation diagrams of positive solutions found in [28] (in blue) with the global bifurcation diagrams of nodal solutions with one node (in red) and two nodes (in black) computed in this paper for four different values of : , , and . Although all the components of nodal solutions persist for these values of , the component of positive solutions shrinks to a single point and disappear at a value of above but very close to it. In Figure 7(b) one can still see an small piece of blue trace component shortly before disappearing for an slightly grater value of .

## 4. Numeric of bifurcation problems

To discretize (1.1) we have used two methods. To compute the small solutions bifurcating from we implemented a pseudo-spectral method combining a trigonometric spectral method with collocation at equidistant points, as in most of our previous numerical experiments (see, e.g., [17, 18, 27, 29, 30, 31, 32]). This gives high accuracy (see, e.g., Canuto, Hussaini, Quarteroni and Zang [11]). However, to compute the large solutions we have used a centered finite differences scheme, which gives high accuracy at a lower computational cost, for as it provides us with a much faster code to compute large pieces of curves of the global bifurcation diagrams.

The pseudo-spectral method is easier to use and more efficient for choosing the shot direction from the trivial solution in order to compute the small nodal solutions of (1.1), as well as to detect bifurcation points along the bifurcation diagrams. Its main advantage for accomplishing this task relies on the fact that it provides us with the true bifurcation values from the trivial solution, while the differences scheme only provides with an approximation to these bifurcation values.

For general Galerkin approximations, the local convergence of the solution paths at regular, turning and simple bifurcation points was proven by Brezzi, Rappaz and Raviart in [6, 7, 8] and by López-Gómez et al. in [27, 34] at codimension two singularities in the context of systems. In these situations, the local structure of the solution sets for the continuous and the discrete models are known to be equivalent. The global continuation solvers used to compute the solution curves of this papers, as well as the dimensions of the unstable manifolds of all the solutions along them, have been built from the theory on continuation methods of Allgower and Georg [1], Crouzeix and Rappaz [13], Eilbeck [16], Keller [22], López-Gómez [24] and López-Gómez, Eilbeck, Duncan and Molina-Meyer [27].

The complexity of the bifurcation diagrams, as well as their quantitative features, required an extremely careful control of all the steps in subroutines. This explains why the available commercial bifurcation packages, such as AUTO-07P are un-useful to deal with differential equations with heterogeneous coefficients. As a matter of fact, Doedel and Oldeman admitted in [14, p.18] that

“Note that, given the non-adaptive spatial discretization, the computational procedure here is not appropriate for PDEs with solutions that rapidly vary in space, and care must be taken to recognize spurious solutions and bifurcations.”

This is just one of the main problems that we found in our numerical experiments, as the number of critical points of the solutions increases according to the dimensions of unstable manifolds, and the turning and bifurcation points might be very close.

## References

• [1] E. L. Allgower and K. Georg, Introduction to Numerical Continuation Methods, SIAM Classics in Applied Mathematics 45, SIAM, Philadelphia, 2003.
• [2] I. Antón and J. López-Gómez, Principal eigenvalues of weighted periodic-parabolic problems, Rend. Istit. Mat. Univ. Trieste 49 (2017), 287–318.
• [3] W. Arendt, C. J. K. Batty, M. Hieber and F. Neubrander, Vector-valued Laplace transforms and Cauchy problems, Monographs in Mathematics vol. 96, Birkhäuser/Springer, Basel, 2011.
• [4] W. Arendt, A. Grabosch, G. Greiner, U. Groh, H. P. Lotz, U. Moustakas, R. Nagel, F. Neubrander and U. Schlotterbeck, One Parameter Semigroups of Positive Operators, Lectures Notes in Mathematics 1184, Berlin, Springer, 1986.
• [5] H. Berestycki, L. Nirenberg and S. R. S. Varadhan, The principal eigenvalue and maximum principle for second order elliptic operator sin general domains, Comm. Pure Appl. Math. 47 (1994), 47–92.
• [6] F. Brezzi, J. Rappaz and P. A. Raviart, Finite dimensional approximation of nonlinear problems, part I: Branches of nonsingular solutions, Numer. Math. 36 (1980), 1–25.
• [7] F. Brezzi, J. Rappaz and P. A. Raviart, Finite dimensional approximation of nonlinear problems, part II: Limit points, Numer. Math. 37 (1981), 1–28.
• [8] F. Brezzi, J. Rappaz and P. A. Raviart, Finite dimensional approximation of nonlinear problems, part III: Simple bifurcation points, Numer. Math. 38 (1981), 1–30.
• [9] G. Buttazzo, M. Giaquinta and S. Hildebrandt, One-dimensional Variational Problems, Clarendon Press, Oxford, 1998.
• [10] S. Cano-Casanova and J. López-Gómez, Properties of the principal eigenvalues of a general class of nonclassical mixed boundary value problems, J. Dif. Eqns. 178 (2002), 123–211.
• [11] C. Canuto, M. Y. Hussaini, A. Quarteroni and T. A. Zang, Spectral Methods in Fluid Mechanics, Springer, Berlin, Germany, 1988.
• [12] M. G. Crandall and P. H. Rabinowitz, Bifurcation from simple eigenvalues, J. Funct. Anal. 8 (1971), 321–340.
• [13] M. Crouzeix and J. Rappaz, On Numerical Approximation in Bifurcation Theory, Recherches en Mathématiques Appliquées 13, Masson, Paris, 1990.
• [14] E. J. Doedel and B. E. Oldeman, AUTO-07P: Continuation and bifurcation software for ODEs, 2012, http://www.dam.brown.edu/people/sandsted/auto/auto07p.pdf.
• [15] J. Esquinas and J. López-Gómez, Optimal multiplicity in local bifurcation theory: Generalized generic eigenvalues, J. Diff. Eqns. 71 (1988), 72–92.
• [16] J. C. Eilbeck, The pseudo-spectral method and path-following in reaction-diffusion bifurcation studies, SIAM J. Sci. Stat. Comput. 7 (1986), 599–610.
• [17] R. Gómez-Reñasco and J. López-Gómez, The effect of varying coefficients on the dynamics of a class of superlinear indefinite reaction diffusion equations, J. Diff. Eqns. 167 (2000), 36–72.
• [18] R. Gómez-Reñasco and J. López-Gómez, On the existence and numerical computation of classical and non-classical solutions for a family of elliptic boundary value problems, Nonl. Anal. TMA 48 (2002), 567–605.
• [19] P. Hess and T. Kato, On some linear and nonlinear eigenvalue problems with an indefinite weight function, Comm. Part. Dif. Eqns. 5 (1980), 999–1030.
• [20] T. Kato, Superconvexity of the spectral radius and convexity of the spectral bound and the type, Math. Z. 180 (1982), 265–273.
• [21] T. Kato, Perturbation Theory for Linear Operators, Springer, 1995.
• [22] H. B. Keller, Lectures on Numerical Methods in Bifurcation Problems, Tata Insitute of Fundamental Research, Springer, Berlin, Germany, 1986.
• [23] H. B. Keller and Z. H. Yang, A direct method for computing higher order folds, SIAM J. Sci. Stat. 7 (1986), 351–361.
• [24] J. López-Gómez Estabilidad y Bifurcación Estática. Aplicaciones y Métodos Numéericos, Cuadernos de Matemática y Mecánica, Serie Cursos y Seminarios 4, Santa Fe, R. Argentina, 1988.
• [25] J. López-Gómez, Spectral Theory and Nonlinear Functional Analysis, CRC Press, Boca Raton, 2001.
• [26] J. López-Gómez, Linear Second Order Elliptic Operators, World Scientific Publishing, 2013.
• [27] J. López-Gómez, J. C. Eilbeck, K. Duncan and M. Molina-Meyer, Structure of solution manifolds in a strongly coupled elliptic system, IMA J. Numer. Anal. 12 (1992), 405–428.
• [28] J. López-Gómez and M. Molina-Meyer, Bounded components of positive solutions of abstract fized point equations: mushrooms, loops and isolas, J. Diff. Eqns. 209 (2005), 416–441.
• [29] J. López-Gómez and M. Molina-Meyer, Superlinear indefinite systems: Beyond Lotka Volterra models, J. Differ. Eqns. 221 (2006), 343–411.
• [30] J. López-Gómez and M. Molina-Meyer, The competitive exclusion principle versus biodiversity through segregation and further adaptation to spatial heterogeneities, Theor. Popul. Biol. 69 (2006), 94–109.
• [31] J. López-Gómez and M. Molina-Meyer, Modeling coopetition, Math. Comput. Simul. 76 (2007), 132–140.
• [32] J. López-Gómez, M. Molina-Meyer and A. Tellini, Intricate dynamics caused by facilitation in competitive environments within polluted habitat patches, Eur. J. Appl. Maths. doi:10.1017/S0956792513000429 (2014), 1–17.
• [33] J. López-Gómez, M. Molina-Meyer and P. H. Rabinowitz, Global bifurcation diagrams of one node solutions in a class of degenerate bundary value problems, Disc. Cont. Dyn. Sys. B 22 (2017), 923–946.
• [34] J. López-Gómez, M. Molina-Meyer and M. Villareal, Numerical coexistence of coexistence states, SIAM J. Numer. Anal. 29 (1992), 1074–1092.
• [35] J. López-Gómez, A. Tellini and F. Zanolin, High multiplicity and complexity of the bifurcation diagrams of large solutions for a class of superlinear indefinite problems, Commun. Pure Appl. Anal. 13 (2014), 1–73.
• [36] J. López-Gómez and P. H. Rabinowitz, Nodal solutions for a class of degenerate boundary value problems, Adv. Nonl. Studies 15 (2015), 253–288.
• [37] J. López-Gómez and P. H. Rabinowitz, Nodal solutions for a class of degenerate one dimensional BVPs, Top. Meth. Nonl. Anal. 49 (2017), 359–376.
• [38] J. López-Gómez and P. H. Rabinowitz, The structure of the set of 1-node solutions of a clss of degenerate BVP’s, J. Diff. Eqns. 268 (2020), 4691–4732.
• [39] J. Mawhin, D. Papini and F. Zanolin, Boundary blow-up for differential equations with indefinite weight, J. Diff. Eqns. 188 (2003), 33–51.
• [40] P. H. Rabinowitz, Some global results for nonlinear eigenvalue problems, J. Funct. Anal. 7 (1971), 487–513.
• [41] P. H. Rabinowitz, A note on a nonlinear eigenvalue problem for a class of differential equations, J. Diff. Eqns. 9 (1971), 536–548.
• [42] P. H. Rabinowitz, A note on a pair of solutions of a nonlinear Sturm–Liouville problem, Manuscr. Math. 11 (1974), 273–282.