PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of jmathneurSpringerOpen.comSubmit OnlineRegisterThis journalThis article
 
J Math Neurosci. 2017; 7: 6.
Published online 2017 July 14. doi:  10.1186/s13408-017-0049-1
PMCID: PMC5509800

Regularization of Ill-Posed Point Neuron Models

Abstract

Point neuron models with a Heaviside firing rate function can be ill-posed. That is, the initial-condition-to-solution map might become discontinuous in finite time. If a Lipschitz continuous but steep firing rate function is employed, then standard ODE theory implies that such models are well-posed and can thus, approximately, be solved with finite precision arithmetic. We investigate whether the solution of this well-posed model converges to a solution of the ill-posed limit problem as the steepness parameter of the firing rate function tends to infinity. Our argument employs the Arzelà–Ascoli theorem and also yields the existence of a solution of the limit problem. However, we only obtain convergence of a subsequence of the regularized solutions. This is consistent with the fact that models with a Heaviside firing rate function can have several solutions, as we show. Our analysis assumes that the vector-valued limit function v, provided by the Arzelà–Ascoli theorem, is threshold simple: That is, the set containing the times when one or more of the component functions of v equal the threshold value for firing, has zero Lebesgue measure. If this assumption does not hold, we argue that the regularized solutions may not converge to a solution of the limit problem with a Heaviside firing function.

Keywords: Point neuron models, Ill-posed, Regularization, Existence

Introduction

In this paper we analyze some mathematical properties of the following classical point neuron model:

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ1.gif
1

for i = 1, 2, …, N, where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equa.gif

Here, ui(t) represents the unknown electrical potential of the ith unit in a network of N units. The nonlinear function Sβ is called the firing rate function, β is the steepness parameter of Sβ, uθ is the threshold value for firing, {ωij} are the connectivities, {τi} are membrane time constants and {qi(t)} model the external drive/external sources, see, e.g., [13] for further details. The system of ODEs (1) is also referred to as a voltage-based model or Hopfield model (due to Hopfield [4]).

By employing electrophysiological properties one can argue that it is appropriate to use a steep sigmoid firing rate function Sβ. But due to mathematical convenience the Heaviside function is also often employed, see, e.g., [58]. Unfortunately, when β = ∞ the initial-condition-to-solution map for (1) can become discontinuous in finite time [9]. Such models are thus virtually impossible to solve with finite precision arithmetic [10, 11]. Also, in the steep but Lipschitz continuous firing rate regime, the error amplification can be extreme, even though a minor perturbation of the initial condition does not change which neurons that fire. It is important to note that this ill-posed nature of the model is a fundamentally different mathematical property from the possible existence of unstable equilibria, which typically also occur if a firing rate function with moderate steepness is used. See [9] for further details.

The solution of (1) depends on the steepness parameter β. That is,

ui(t) = uβ,i(t),  i = 1, 2, …, N

and the purpose of this paper is to analyze the limit process β → ∞. This investigation is motivated by the fact that the stable numerical solution of an ill-posed problem is very difficult, if not to say impossible, see, e.g., [10, 11]. Consequently, such models must be regularized to obtain a sequence of well-posed equations which, at least in principle, can be approximately solved by a computer. Also, steep firing rate functions, or even the Heaviside function, are often used in simulations. It is thus important to explore the limit process β → ∞ rigorously.

In Sects. 3 and 4 we use the Arzelà–Ascoli theorem [1214] to analyze the properties of the sequence {uβ}, where

uβ(t) = (uβ,1(t),uβ,2(t),…,uβ,N(t))T.
2

More specifically, we prove that this sequence has at least one subsequence which converges uniformly to a limit

v(t) = (v1(t),v2(t),…,vN(t))T

and that this limit satisfies the integral/Volterra version of (1) with SβS, provided that the following set has zero Lebesgue measure:

{s ∈ [0, T]∣∃i ∈ {1, 2, …, N} such that vi(s) = uθ}.

It is thus sufficient that this set is finite or countable; see, e.g., [13]. Furthermore, in Sect. 7 we argue that, if v does not satisfy this threshold property, then this function will not necessarily solve the limit problem.

According to the Picard–Lindelöf theorem [1517], (1) has a unique solution, provided that β < ∞, and that the assumptions presented in the next section hold. In Sect. 5 we show that this uniqueness feature is not necessarily inherited by the limit problem obtained by employing a Heaviside firing rate function. It actually turns out that a different subsequence of {uβ} can converge to different solutions of (1) with SβS. This is explained in Sect. 6, which also contains a result addressing the convergence of the entire sequence {uβ}.

The limit process β → ∞, using different techniques, is studied in [18, 19] for the stationary solutions of neural field equations. It has also been observed [20] for the Wilson–Cowan model that this transition is a subtle matter: Using a steep sigmoid firing rate function instead of the Heaviside mapping can lead to significant changes in a Hopf bifurcation point. ‘the limiting value of the Hopf depends on the choice of the firing rate function’.

If one uses a Heaviside firing rate function in (1) the right-hand-sides of these ODEs become discontinuous. A rather general theory for such equations has been developed [21]. In this theory the system of ODEs is replaced by a differential inclusion, in which the right-hand side of the ODE system is substituted by a set-valued function. The construction of this set-valued operator can be accomplished by invoking Filippov regularization/convexification. But this methodology serves a different purpose than the smoothing processes considered in this paper. More specifically, it makes it possible to prove that generalized solutions (Filippov solutions) to the problem exist but do not provide a family of well-posed equations suitable for numerical solution.

Smoothening techniques for discontinuous vector fields, which are similar to the regularization method considered in this paper, have been proposed and analyzed for rather general phase spaces [2224]. Nevertheless, these studies consider qualitative properties of large classes of problems, whereas we focus on a quantitative analysis of a very special system of ODEs.

For the sake of easy notation, we will sometimes write (1) in the form

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ3.gif
3

where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ4.gif
4

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ5.gif
5

Note that we, for the sake of simplicity, use the same threshold value uθ for all the units in the network; see (4).

Assumptions

Throughout this text we use the standard notation

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Eque.gif

Concerning the sequence {Sβ} of finite steepness firing rate functions, we make the following assumption.

Assumption A

We assume that

  1. Sβ, β ∈ ℕ, is Lipschitz continuous,
  2. 0 ≤ Sβ(x) ≤ 1, x ∈ ℝ,  β ∈ ℕ,
  3. for every pair of positive numbers (ϵδ) there exists Q ∈ ℕ such that
    |Sβ(x)| < ϵ for x < −δ and β > Q
    6
    |1 − Sβ(x)| < ϵ for x > δ and β > Q.
    7

There are many continuous sigmoidal functions which approximate the Heaviside step function and satisfy Assumption A. For example,

S(x) = ½(1 + tanh(x)), 
8

Sβ[x] = S(βx).
9

More generally, if Sβ is nondecreasing (for every β ∈ ℕ), a) and b) hold and {Sβ} converges pointwise to the Heaviside function, then Assumption A holds. Also, if Assumption A is satisfied and limβ→∞Sβ(0) = S(0) = H(0), then {Sβ} converges pointwise to the Heaviside function.

We will consider a slightly more general version of the model than (3). More specifically, we allow the source term to depend on the steepness parameter, qqβ, but in such a way that the following assumption holds.

Assumption B

We assume that qβ(t), t ∈ [0, T], β ∈ ℕ ∪ {∞} is continuous and that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ10.gif
10

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ11.gif
11

Allowing the external drive to depend on the steepness parameter makes it easier to construct illuminating examples. However, we note that our theory will also hold for the simpler case when q does not change as β increases.

In this paper we will assume that Assumptions A and B are satisfied.

Uniformly Bounded and Equicontinuous

In order to apply the Arzelà–Ascoli theorem we must show that {uβ} constitutes a family of uniformly bounded and equicontinuous functions. (For the sake of simple notation, we will write ui and qi, instead of uβ,i and qβ,i, for the component functions of uβ and qβ, respectively.) Multiplying

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equf.gif

with es/τi yields that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equg.gif

and by integrating

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equh.gif

Hence, since Sβ[x] ∈ [0, 1] and we assume that τi > 0 for i = 1, 2, …, N,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equi.gif

where the last inequality follows from (10). This implies that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ12.gif
12

Since the right-hand side of (12) is independent of β and t we conclude that the sequence {uβ} is uniformly bounded.

Next, from the model (3), the triangle inequality, the assumption that Sβ[x] ∈ [0, 1] and assumption (10) we find that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equj.gif

where B is defined in (12). Since τ is a diagonal matrix with positive entries on the diagonal, this yields that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equk.gif

Here the constant K is independent of both β and t ∈ (0, T].

Let i ∈ {1, 2, …, N} and β ∈ ℕ be arbitrary. Then, for any time instances t1t2 ∈ [0, T], with t1 < t2, the mean value theorem implies that there exists t ∈ (t1t2) such that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equl.gif

and hence

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equm.gif

This inequality holds for any i ∈ {1, 2, …, N}, β ∈ ℕ. It therefore follows that

uβ(t2)−uβ(t1)∥ ≤ K|(t2 − t1)|,  t1t2 ∈ [0, T] and β ∈ ℕ, 

from which we conclude that {uβ} is a set of equicontinuous functions

The Arzelà–Ascoli theorem now asserts that there is a uniformly convergent subsequence {uβk}:

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ13.gif
13

According to standard ODE theory, uβk is continuous for each k ∈ ℕ. Hence the uniform convergence implies that v is also continuous.

Threshold Terminology

As we will see in subsequent sections it depends on v’s threshold properties whether we can prove that v actually solves the limit problem with a Heaviside firing rate function. The following concepts turn out to be useful.

For a vector-valued function z = (z1,z2,…,zN)T:[0, T] → ℝN we define

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ14.gif
14

Definition

Threshold simple

A measurable vector-valued function z:[0, T] → ℝN is threshold simple if the Lebesgue measure of the set

Z(z) = {s ∈ [0, T]∣m(sz) = 0}
15

is zero, i.e. |Z(z)| = 0.

Definition

Extra threshold simple

A measurable vector-valued function z:[0, T] → ℝN is extra threshold simple if there exist open intervals

Il = (alal+1),  l = 1, 2, …, L

such that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equp.gif

In words, z is extra threshold simple if there is a finite number of threshold crossings on the time interval [0, T].

The Limit of the Subsequence

Preparations

We will prove that the limit v in (13) solves the integral form of (3) with SH, the Heaviside function, provided that v is threshold simple. The inhomogeneous nonlinear Volterra equation associated with (3) reads

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ16.gif
16

where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equq.gif

etc.; see also (2) and (5). Note that we consider the equations satisfied by the subsequence {uβk}, see (13). We will analyze the convergence of the entire sequence in Sect. 6.

The uniform convergence of {uβk} to v implies that the left-hand-side and the first integral on the right-hand side of (16) converge to τv(t)−τuinit and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq71.gif
, respectively, as k → ∞. Also, due to assumption (11), the third integral on the right-hand side does not require any extra attention. We will thus focus on the second integral on the right-hand side of (16).

For t ∈ [0, T] and δ > 0, define the sets

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ17.gif
17

where m(sv) is defined in (14) and v is the limit in (13). Since v is continuous it follows that m(sv), s ∈ [0, T], is continuous. Hence, the sets p(δt) and r(δt) are Lebesgue measurable. We note that, provided that δ > 0 is small, the set r(δt) contains the times where at least one of the components of v is close to the threshold value uθ for firing. The following lemma turns out to be crucial for our analysis of the second integral on the right-hand side of (16).

Lemma 4.1

If the limit function v in (13) is threshold simple, then

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ18.gif
18

where |r(δt)| denotes the Lebesgue measure of the set r(δt).

Proof

Since v is the uniform limit of a sequence of continuous functions, v is continuous and hence measurable. If v is threshold simple, then

|Z(v)| = 0, 
19

see (15).

Let t ∈ [0, T] be arbitrary. Assume that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equr.gif

or that this limit does not exist. Then

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq86.gif
such that there is a sequence {δn} satisfying

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equs.gif

By construction,

r(δ1t) ⊃ r(δ2t) ⊃  ⋯  ⊃ r(δnt) ⊃ …, 

and |r(δ1t)| ≤ T < ∞. Hence,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equu.gif

see, e.g., [13] (page 62). Since the sequence {|r(δnt)|} is nonincreasing and bounded below, limn→∞|r(δnt)| exists.

Next,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equv.gif

i.e.

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equw.gif

Hence,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equx.gif

which contradicts (19).

Convergence of the Integral

Lemma 4.2

If the limit v in (13) is threshold simple, then

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ20.gif
20

Proof

Let t ∈ [0, T] and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq92.gif
be arbitrary and define

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equy.gif

From (18) we know that there exists Δ > 0 such that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ21.gif
21

Choose a δ which satisfies 0 < δ < Δ. By part (c) of Assumption A, for this δ and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ22.gif
22

there exists Q ∈ ℕ such that (6) and (7) hold.

Recall that β1β2, …, βk, … are the values for the steepness parameter associated with the convergent subsequence {uβk} in (13). By the uniform convergence of {uβk} to v, there is a K ∈ ℕ so that

βK > Q
23

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ24.gif
24

From the definition of the set p(2δt), see (17) and (14),

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ25.gif
25

and from (24) and the triangle inequality it follows that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ26.gif
26

From (24)–(26) we find that

(vj(s)−uθ) ⋅ (uβk,j(s)−uθ) > 0,  s ∈ p(2δt), j ∈ {1, 2, …, N}, k > K.

Also, because of the properties of the Heaviside function,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equaa.gif

j ∈ {1, 2, …, N}. Consequently, due to (23) and part (c) of Assumption A, see (6) and (7), we find that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equab.gif

Hence,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equac.gif

for all k > K, where the second last inequality follows from (22), the fact that |p(2δt)| ≤ T for t ∈ [0, T] and (21). Since

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq105.gif
and t ∈ [0, T] were arbitrary, we conclude that (20) must hold.

Limit Problem

By employing the uniform convergence (13), the convergence of the integral (20) and assumption (11), we conclude from (16) that the limit function v satisfies

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ27.gif
27

provided that v is threshold simple. Recall that v is continuous. Consequently, if v is extra threshold simple, then it follows from the fundamental theorem of calculus that v also satisfies the ODEs, except at time instances when one or more of the component functions equal the threshold value for firing:

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ28.gif
28

where Z(v) is defined in (15).

The existence of a solution matter for point neuron models with a Heaviside firing rate function is summarized in the following theorem.

Theorem 4.3

If the limit v in (13) is threshold simple, then v solves (27). In the case that v is extra threshold simple v also satisfies (28).

In [25] the existence issue for neural field equations with a Heaviside activation function is studied but the analysis is different because a continuum model is considered. We would also like to mention that Theorem 4.3 cannot be regarded as a simple consequence of Carathéodory’s existence theorem [21, 26, 27] because the right-hand-side of (28) is discontinuous with respect to v.

Uniqueness

If β < ∞, then standard ODE theory [1517] implies that (3) has a unique solution. Unfortunately, as will be demonstrated below, this desirable property is not necessarily inherited by the infinite steepness limit problem.

We will first explain why the uniqueness question is a subtle issue for point neuron models with a Heaviside firing rate function. Thereafter, additional requirements are introduced which ensure the uniqueness of an extra threshold simple solution.

Example: Several Solutions

Let us study the problem

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ29.gif
29

where we assume that

ω > uθ ≥ 0.

Note that the ODE in (29) is not required to hold for t = 0. Consider the functions

v1(t) = ω + (uθ − ω)etuθet + (1 − et)ω
30

v2(t) = uθet.
31

Since

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equae.gif

it follows that both v1 and v2 solves (29).

Furthermore, with

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equaf.gif

we actually obtain a third solution of (29). More specifically, the stationary solution

v3(t) = uθt ∈ [0, T].
32

We conclude that models with a Heaviside firing rate function can have several solutions – such problems can thus become ill-posed. (In [9] we showed that the initial-condition-to-solution map is not necessarily continuous for such problems and that the error amplification ratio can become very large in the steep but Lipschitz continuous firing rate regime.) Note that switching to the integral form (27) will not resolve the lack of uniqueness issue for the toy example considered in this subsection.

We also remark that

  • If we define S(0) = 1/2, then neither v1 nor v2 satisfies the ODE in (29) for t = 0. (In the case ω = 2uθ, v3 satisfies the ODE in (29) for t = 0.)
  • If we define S(0) = 1, then v1, but not v2, satisfies the ODE in (29) also for t = 0.
  • If we define S(0) = 0, then v2, but not v1, satisfies the ODE in (29) also for t = 0.

Enforcing Uniqueness

In order to enforce uniqueness we need to impose further restrictions. It turns out that it is sufficient to require that the derivative is continuous from the right and that the ODEs also must be satisfied whenever one, or more, of the component functions equals the threshold value for firing

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ33.gif
33

Note that the ODEs in (33) also must be satisfied for t = 0, in case one of the components of uinit equals uθ.

Definition 1

Right smooth

A vector-valued function z:[0, T] → ℝN is right smooth if z is continuous from the right for all t ∈ [0, T).

Theorem 5.1

The initial value problem (33) can have at most one solution which is both extra threshold simple and right smooth.

Proof

Let v and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq133.gif
be two solutions of (33) which are both right smooth and extra threshold simple:

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equag.gif

and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equah.gif

where I1I2, …, IL and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq135.gif
are disjoint open intervals; see (14) and the definition of extra threshold simple in Sect. 3.1.

Then there exist disjoint open intervals

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq136.gif
such that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ34.gif
34

Let us focus on one of these intervals,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq137.gif
. Define

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equai.gif

and assume that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ35.gif
35

which obviously holds for l = 1. Then

τd(t) = −d(t) + ωγ(t),  t ∈ [alal+1], 
36

d(al) = 0, 
37

where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equaj.gif

Note that, due to (34), γ(t) equals a constant vector c, with components −1, 0 or 1, except possibly at talal+1:

γ(t) = ct ∈ (alal+1).
38

Furthermore, from (35) we find that

γ(al) = 0.
39

Putting tal in (36) and invoking (37) and (39) yield

d(al) = 0, 

and from the right continuity of d and d, (36), (37) and (38) we find that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equal.gif

Since ωγ(t) = ωc = 0, t ∈ (alal+1), and ωγ(al) = 0, see (39), we conclude from (36)–(37) that d satisfies

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equam.gif

which has the unique solution d(t) = 0, t ∈ [alal+1). Both v(t) and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq150.gif
are differentiable on [0, T] and hence continuous. It follows that, by employing the continuity of v and
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq152.gif
at time tal+1,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equan.gif

Since

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq154.gif
we can repeat the argument on the next interval [al+1al+2]. It follows by induction that
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq156.gif
.

We would like to comment the findings presented in the bullet-points at the end of Sect. 5.1 in view of Theorem 5.1: In order to enforce uniqueness for the solution of (29) we can require that the ODE in (29) also should be satisfied for t = 0. Nevertheless, this might force us to define S(0) ≠ ½, which differs from the standard definition of the Heaviside function H.

More generally, if one has accomplished to compute an extra threshold simple and right smooth function v which satisfies (27), one can attempt to redefine S[v(t)−uθ], t ∈ {a1a2, …, aL+1}, such that (33) holds and v is the only solution to this problem. This may imply that S[v(t)−uθ] cannot be generated by using the composition H ∘ [v(t)−uθ]. Instead one must determine zj,kS[vj(ak)−uθ], j = 1, 2, …, N, k = 1, 2, …, L + 1. More precisely, for each k ∈ {1, 2, …, L + 1} one gets a linear system of algebraic equations

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equao.gif

which will have a unique solution (z1,k,z2,k,…,zN,k)T if the connectivity matrix ω = [ωi,j] is nonsingular. (In this paragraph, {0 = a1a2, …, aL+1T} are the time instances employed in the definition of extra threshold simple; see Sect. 3.1.)

Convergence of the Entire Sequence

We have seen that point neuron models with a Heaviside firing rate function can have several solutions. One therefore might wonder if different subsequences of {uβ} can converge to different solutions of the limit problem. In this section we present an example which shows that this can happen, even though the involved sigmoid functions satisfy Assumption A.

Example: Different Subsequences Can Converge to Different Solutions

Let us again consider the initial value problem (29), which we discussed in Sect. 5.1. A finite steepness approximation of this problem, using the notation u(t) = uβ(t), reads:

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ40.gif
40

where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equap.gif

and Sβ is, e.g., either the hyperbolic tangent sigmoid function (8)–(9) or

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ41.gif
41

Note that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq173.gif
converges pointwise, except for x = 0, to the Heaviside function H as β → ∞. In fact,
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq176.gif
satisfies Assumption A.

We consider the case ω = 2uθ. Therefore (29) has three solutions v1v2 and v3, see (30), (31) and (32) in Sect. 5.1. Note that

u(t) = uβ(t)

has the property

  • An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq180.gif
    if β is even,
  • An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq181.gif
    if β is odd,

where c > 0 is a constant which is independent of β. It therefore follows that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equar.gif

and no subsequence converges to the third solution v3. Figure 1 shows numerical solutions of (40) with steepness parameter

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq184.gif
, using the firing rate function (41) to define
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq185.gif
. (If one instead employs (8)–(9) in the implementation of
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq186.gif
, the plots, which are not included, are virtually unchanged.)

Fig. 1
Numerical solutions of (40) computed with Matlab’s ode45 software. In these simulations we used uθ = 0.6 and ω = 1.2. The functions v1 and v2, see (30) and (31), are the solutions of the associated limit problem (29)

We would like to mention that we have not been able to construct an example of this kind for Lipschitz continuous firing rate functions which converge pointwise to the Heaviside function also for x = 0.

Entire Sequence

We have seen that almost everywhere convergence of the sequence of firing rate functions to the Heaviside limit is not sufficient to guarantee that the entire sequence {uβ} converges to the same solution of the limit problem. Nevertheless, one has the following result.

Theorem 6.1

Let v be the limit function in (13). If the limit of every convergent subsequence of {uβ} is extra threshold simple, right smooth and satisfies (33), then the entire sequence {uβ} converges uniformly to v.

Proof

Suppose that the entire sequence {uβ} does not converge uniformly to v. Then there is an ϵ > 0 such that, for every positive integer M, there must exist uβl, βl > M, satisfying

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ42.gif
42

Thus, the subsequence {uβl} does not converge uniformly to v, but constitutes a set of uniformly bounded and equicontinuous functions, see Sect. 3. According to the Arzelà–Ascoli theorem, {uβl} therefore possesses a uniformly convergent subsequence {uβln},

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equas.gif

Due to (42),

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ43.gif
43

On the other hand, both v and

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq202.gif
are limits of subsequences of {uβ} and are by assumption extra threshold simple, right smooth, and they satisfy (33). Hence, Theorem 5.1 implies that
An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq204.gif
, which contradicts (43). We conclude that the entire sequence {uβ} must converge uniformly to v.

Example: Threshold Advanced Limits

We will now show that threshold advanced limits, i.e. limits which are not threshold simple, may possess some peculiar properties. More precisely, such limits can potentially occur in (13). They do not necessarily satisfy the limit problem obtained by using a Heaviside firing rate function.

With source terms which do not depend on the steepness parameter β we have not managed to construct an example with a threshold advanced limit v. If we allow qqβ, this can, however, be accomplished as follows. Let

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equat.gif

where we, for the sake of simplicity, work with the firing rate function (41). Then

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equau.gif

and we find that

uβ(t) = zβ(t)

solves

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equaw.gif

where

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equ44.gif
44

It follows that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equax.gif

and since, for any β ∈ ℕ,

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equay.gif

we conclude that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equaz.gif

Note that

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equba.gif

but

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_IEq208.gif
does not solve the limit problem

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equbb.gif

because

An external file that holds a picture, illustration, etc.
Object name is 13408_2017_49_Article_Equbc.gif

This argument assumes that S[0] = 1/2. If one instead defines S[0] = 1, then v would solve the limit problem.

Due to the properties of the firing rate function (41) the source term qβ in (44) becomes discontinuous. This can be avoided by instead using the smooth version (8)–(9) but then the analysis of this example becomes much more involved.

Discussion and Conclusions

If a Heaviside firing rate function is used, the model (1) may not only have several solutions, but the initial-condition-to-solution map for this problem can become discontinuous [9]. It is thus virtually impossible to develop reliable numerical methods which employ finite precision arithmetic for such problems. One can try to overcome this issue by

  1. Attempting to solve the ill-posed equation with symbolic computations.
  2. Regularize the problem.

To the best of our knowledge, present symbolic techniques are not able to handle strongly nonlinear equations of the kind (1), even when β < ∞. We therefore analyzed the approach b), using the straightforward regularization technique obtained by replacing the Heaviside firing rate function by a Lipschitz continuous mapping. This yields an equation which is within the scope of the Picard–Lindelöf theorem and standard stability estimates for ODEs. That is well-posed and, at least in principle, approximately solvable by numerical methods.

Our results show that the sequence {uβ} of regularized solutions will have at least one convergent subsequence. The limit, v, of this subsequence will satisfy the integral/Volterra form (27) of the limit problem, provided that the set Z(v), see (15), has zero Lebesgue measure. Unfortunately, it seems to be very difficult to impose restrictions which would guarantee that v obeys this threshold property, which we refer to as threshold simple. Also, the example presented in Sect. 7 shows that, if the limit v is not threshold simple, then this function may not solve the associated equation with a Heaviside firing rate function.

One could propose to overcome the difficulties arising when β = ∞ by always working with finite slope firing rate functions. This would potentially yield a rather robust approach, provided that the entire sequence {uβ} converges, because increasing a large β would still guarantee that uβ is close to the unique limit v. However, the fact that different convergent subsequences of {uβ} can converge to different solutions of the limit problem, as discussed in Sect. 6, suggests that this approach must be applied with great care. In addition, the error amplification in the steep firing rate regime can become extreme [9] and the accurate numerical solution of such models is thus challenging.

What are the practical consequences of our findings? As long as there does not exist very reliable biological information about the size of the steepness parameter β and the shape of the firing rate function Sβ, it seems that we have to be content with simulating with various β < ∞. If one observes that uβ approaches a threshold advanced limit, as β increases, or that the entire sequence does not converge, the alarm bell should ring. All simulations with large β must use error control methods which guarantee the accuracy of the numerical solution—we must keep in mind that we are trying to solve an almost ill-posed problem.

In neural field equations one employs a continuous variable, e.g., x ∈ ℝ, instead of a discrete index i ∈ {1, 2, …, N}. The sum in (1) is replaced by an integral; see [1, 2, 6]. For each time instance t ∈ [0, T] one therefore does not get a vector uβ(t) ∈ ℝN, as for the point neural models analyzed in this paper, but a function uβ(xt), x ∈ ℝ. That is, in neural field equations the object associated with each fixed t ∈ [0, T] belongs to an infinite dimensional space. It is often a subtle task to generalize concepts and proofs from a finite to an infinite dimensional setting: It is thus an open problem whether the techniques and results presented in this paper can be adapted to neural field models.

Acknowledgements

This work was supported by the The Research Council of Norway, project number 239070. The author would like to thank the reviewers and Prof. Wyller for a number of interesting comments, which significantly improved this paper.

Footnotes

Competing interests

The author declares that he has no competing interests.

Author’s contributions

All authors read and approved the final manuscript.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Bressloff P. Spatiotemporal dynamics of continuum neural fields. J Phys A, Math Theor. 2012;45 doi: 10.1088/1751-8113/45/3/033001. [Cross Ref]
2. Ermentrout GB. Neural networks as spatio-temporal pattern-forming systems. Rep Prog Phys. 1998;61:353–430. doi: 10.1088/0034-4885/61/4/002. [Cross Ref]
3. Faugeras O, Veltz R, Grimbert F. Persistent neural states: Stationary localized activity patterns in nonlinear continuous n-population, q-dimensional neural networks. Neural Comput. 2009;21:147–187. doi: 10.1162/neco.2009.12-07-660. [PubMed] [Cross Ref]
4. Hopfield JJ. Neurons with graded response have collective computational properties like those of two-state neurons. Proc Natl Acad Sci USA. 1984;81:3088–3092. doi: 10.1073/pnas.81.10.3088. [PubMed] [Cross Ref]
5. Amari S. Dynamics of pattern formation in lateral-inhibition type neural fields. Biol Cybern. 1977;27:77–87. doi: 10.1007/BF00337259. [PubMed] [Cross Ref]
6. Coombes S. Waves, bumps, and patterns in neural field theories. Biol Cybern. 2005;93:91–108. doi: 10.1007/s00422-005-0574-y. [PubMed] [Cross Ref]
7. Pinto DJ, Ermentrout GB. Spatially structured activity in synaptically coupled neuronal networks: I. Traveling fronts and pulses. SIAM J Appl Math. 2001;62:206–225. doi: 10.1137/S0036139900346453. [Cross Ref]
8. Pinto DJ, Ermentrout GB. Spatially structured activity in synaptically coupled neuronal networks: II. Lateral inhibition and standing pulses. SIAM J Appl Math. 2001;62:226–243. doi: 10.1137/S0036139900346465. [Cross Ref]
9. Nielsen BF, Wyller J. Ill-posed point neuron models. J Math Neurosci. 2016;6 doi: 10.1186/s13408-016-0039-8. [PMC free article] [PubMed] [Cross Ref]
10. Engl HW, Hanke M, Neubauer A. Regularization of inverse problems. Dordrecht: Kluwer Academic; 1996.
11. Wikipedia. Well-posed problem. https://en.wikipedia.org/wiki/Well-posed_problem (2017).
12. Griffel DH. Applied functional analysis. Chichester: Ellis Horwood; 1981.
13. Royden HL. Real analysis. 3. New York: Macmillan Co.; 1989.
14. Wikipedia. Arzelà–Ascoli theorem. https://en.wikipedia.org/wiki/Arzel%C3%A0%E2%80%93Ascoli_theorem (2017).
15. Hirsch MW, Smale S. Differential equations, dynamical systems and linear algebra. San Diego: Academic Press; 1974.
16. Teschl G. Ordinary differential equations and dynamical systems. Providence: Am. Math. Soc.; 2012.
17. Wikipedia. Picard–Lindelöf theorem. https://en.wikipedia.org/wiki/Picard%E2%80%93Lindel%C3%B6f_theorem (2017).
18. Oleynik A, Ponosov A, Wyller J. On the properties of nonlinear nonlocal operators arising in neural field models. J Math Anal Appl. 2013;398:335–351. doi: 10.1016/j.jmaa.2012.08.063. [Cross Ref]
19. Oleynik A, Ponosov A, Kostrykin V, Sobolev AV. Spatially localized solutions of the Hammerstein equation with sigmoid type of nonlinearity. J Differ Equ. 2016;261(10):5844–5874. doi: 10.1016/j.jde.2016.08.026. [Cross Ref]
20. Harris J, Ermentrout GB. Bifurcations in the Wilson–Cowan equations with nonsmooth firing rate. SIAM J Appl Dyn Syst. 2015;14:43–72. doi: 10.1137/140977953. [Cross Ref]
21. Filippov AF. Differential equations with discontinuous righthand sides. Dordrecht: Kluwer Academic; 1988.
22. Llibre J, da Silva PR, Teixeira MA. Regularization of discontinuous vector fields on equation M655R3 via singular perturbation. J Dyn Differ Equ. 2007;19:309–331. doi: 10.1007/s10884-006-9057-7. [Cross Ref]
23. Sotomayor J, Teixeira MA. Equadiff 95 proceedings of the international conference on differential equations. Singapore: World Scientific; 1996. Regularization of discontinuous vector fields; pp. 207–223.
24. Teixeira MA, da Silva PR. Regularization and singular perturbation techniques for non-smooth systems. Physica D. 2012;241:1948–1955. doi: 10.1016/j.physd.2011.06.022. [Cross Ref]
25. Potthast R, Beim Graben P. Existence and properties of solutions for neural field equations. Math Methods Appl Sci. 2010;33:935–949.
26. Coddington EA, Levinson N. Theory of ordinary differential equations. New York: McGraw-Hill; 1955.
27. Wikipedia. Carathéodory’s existence theorem. https://en.wikipedia.org/wiki/Carath%C3%A9odory%27s_existence_theorem (2017).

Articles from Journal of Mathematical Neuroscience are provided here courtesy of Springer-Verlag