Fokas Method (unied transform)

January 25, 2017

1 Brief History
There exists a particular class of nonlinear PDEs called integrable. By the
mid eighty’s the initial value problem of integrable evolution equations in
one and two space variables was solved via the so called inverse scatter-
ing transform. Following this development, the outstanding open problem
in the analysis of these equations became the solution of initial–boundary–
value problems. A general approach for solving such problems was nally
announced in 1997 [1] and was developed further in the works of almost
100 researchers. It is remarkable that these results have motivated the dis-
covery of a new transform method for solving boundary–value problems
for linear evolution PDEs with x–derivatives of arbitrary order, as well as
for linear elliptic PDEs in two dimensions [2]. This has led to the emer-
gence of a new method in mathematical physics, which is usually referred
to as the Fokas method or the unied transform. Several hundred pa-
pers have been written using this method, some of which can be found in
[http://www.uniedmethod.azurewebsites.net]. The Fokas method has had
a signicant impact, from the analysis of boundary value problems for inte-
grable nonlinear PDEs [3] and the introduction of a new method for studying
the well posedness of arbitrary nonlinear evolution PDEs [4], to a novel for-
mulation of the classical problem of water waves [5]. This method, which
is based on the synthesis, as opposed to the separation of variables [6], uni-
es and extends several classical branches of mathematics, form the usual
transforms to the formulation of Ehrenpreis type integral representations.
It is important to note that the solution of any inhomogeneous boundary
value problem solved by the usual transforms,suers from lack of uniform
convergence at the boundaries. This serious disadvantage, which renders
these representations unsuitable for numerical computations, has not been

1
emphasised in the literature. In contrast, the unied transform yields rep-
resentations which are uniformly convergent. Thus, it gives new formulae
even for such basic problems as the heat equation on the half line, and on
a nite interval (see below). Furthermore, it yields eective analytical for-
mulae for several problems for which there do not exist usual transforms
[6]. Also, it has given rise to new numerical techniques: for evolution PDEs
see, for example, the book The computation of spectral representations for
evolution PDE by S. Vetra–Carvalh [7], where it is noted that for linear
evolutionary PDEs the numerical implementation of the Fokas method is
faster and more accurate than a pseudospectral method; for elliptic PDEs
see, for example, [8]; for other applications see, for example [9].
A pedagogical introduction of the Fokas method is presented in [10]. In
an accompanying editorial, the editor of SIAM Review wrote: Similar to
the Fosbury Flop the method of Fokas approached familiar problems from
a new direction, providing students and instructors with new insights into
linear PDEs.

2 Separation of Variables and Transform Pairs

Until the development of the Fokas method, the most important method
for the analytic solution of linear PDEs was the method of separation of
variables, and the use of an appropriate transform pair. Consider for
example the heat equation

∂u ∂2u
=  (2.1)
∂t ∂x2
Seeking a separable solution in the form,

u(x, t) = X(x)T (t), (2.2)

we nd
XT ′ = X ′′ T,
where prime denotes dierentiation.
Thus,
X ′′ T′
=  (2.3)
X T
Since the LHS of the above equation is a function of x, whereas the RHS is
a function of t, it follows that each of the above ratios is a constant, which

2
for convenience we write as −λ2 , λ arbitrary complex number. Thus, (2.3)
yields the two ODE’s
X ′′ (x) + λ2 X(x) = 0, (2.4)
and
T ′ (t) + λ2 T (t) = 0 (2.5)
Clearly the representation (2.2) is very limited, however, the intuitive idea
is that if we can solve the ODE’s (2.4), (2.5), and if we can sum up
appropriate solutions over λ, then perhaps we can obtain the general solution
of the heat equation.
2
For example, the exponentials eiλx and e−λ t are particular solutions of
equations (2.4) and (2.5) respectively. Hence, equation (2.2) implies that a
particular solution of the heat equation is given by
2
U (λ)eiλx−λ t ,

where λ is an arbitrary complex constant, and U (λ) is an arbitrary function.

Clearly, the following expression is also a solution of the heat equation:
∫
2
u(x, t) = U (λ)eiλx−λ t dλ, (2.6)

where we assume that the above integral makes sense.

It turns out that their exists a general, deep result in analysis known
as the Ehrenpreis Principle, which when applied to the particular case of
the heat equation, shows that for a well posed problem formulated in a
bounded, smooth, convex, domain, the solution can always be written in
the form (2.6). However, this result does not provide a systematic way for
choosing the above contour, as well as for determining the function U (λ).
For the initial value problem of the heat equation, using the Fourier
transform pair, it is straighforward to obtain both the relevant contour and
the function U (λ) :

∫ ∞
1 2
u(x, t) = eiλx−λ t û0 (λ)dλ, − ∞ < x < ∞, t > 0, (2.7)
2π −∞

where ûo (λ) is the Fourier transform of u0 (x),

∫ ∞
û0 (λ) = e−iλx u0 (x)dx, −∞ < λ < ∞
−∞

3
 Suppose that u(x, t) satises the heat equation (2.1) on the half line,

∂u ∂2u
= , 0 < x < ∞, t > 0, (2.8)
∂t ∂x2
together with the following initial and boundary conditions:

u(x, 0) = u0 (x), 0 < x < ∞; u(0, t) = g0 (t), t > 0 (2.9)

Usually, this problem is solved via the sine transform pair:

∫ ∞
ˆ
fs (λ) = sin(λx)f (x)dx, 0 < λ < ∞, (2.10)
0
∫ ∞
2
f (x) = sin(λx)fˆs (λ)dλ, 0 < x < ∞ (2.11)
π 0
By employing the above pair it is straightforward to show that
∫ ∫ ∞ ∫ t 
2 ∞ −λ2 t λ2 τ
u(x, t) = sin(λx)e sin(λξ)u0 (ξ)dξ + λ e g0 (τ )dτ dλ,
π 0 0 0

0 < x < ∞, t > 0 (2.12)

We rst note that this represented is not of the Ehrenpreis form. Second,
it is not straightforward to verify that u(0, t) = g0 (t): if we attempt to verify
this condition by letting x = 0 in the RHS of (2.12) we fail since sin 0 = 0.
This implies that we cannot exchange the integral with the limit x → 0. In
other words, the representation (2.12) is not uniformly convergent at x = 0
unless g0 (t) = 0. This lack of uniform convergence makes the representation
(2.12) unsuitable for the numerical evaluation of the solution.
It should be emphasised that the above pathology, namely the lack of
uniform convergence at the boundary, is a characteristic of any solution
obtained via the usual transform methods. Indeed, these transforms are
dened by considering the homogeneous version of the given inhomogeneous
problem (see the discussion below). Thus, they construct solutions that are
uniformly convergent only for homogeneous data.
In addition to the above major disadvantage of the usual transform meth-
ods, we also note that in this particular case we were able to guess the
correct transform. The good news is that there does exist a systematic,
albeit complicated, way for deriving the appropriate transform pair for a
given IBVP. For example, for the case of equations (2.8) and (2.9) one rst

4
computes the associated Green’s function, namely one solves the following
ODE:
∂2
G(x, ξ, λ) + λ2 G(x, ξ, λ) = δ(x − ξ), 0 < x < ∞, 0 < ξ < ∞,
∂x2
G(0, ξ, λ) = 0, lim G(x, ξ, λ) = 0
x→∞

Then, one computes the integral of G around an appropriate contour in

the complex λ-plane, and this yields the sine transform pair.
The bad news is that for many important IBVPs there does not exist
an x-transform. For example, there does not exist an x-transform for the
so-called Stokes equation on the half-line, namely for the equation

∂u ∂u ∂ 3 u
+ + = 0, 0 < x < ∞, t > 0 (2.13)
∂t ∂x ∂x3
Indeed, this equation, supplemented with the initial and boundary con-
ditions (2.9), denes an x-spectral problem which is non-self adjoint, for
which there does not exist an appropriate transform.
It should be noted that an evolution PDE in one space variable can
also be analysed via a transform in t, which turns out to be the Laplace
transform.
Denoting by ûL (s, x) the Laplace transform of u(t, x) we nd

∂ ûL (s, x) ∂ 3 ûL (s, x)

+ + sûL (s, x) = u0 (x), 0 < x < ∞ (2.14)
∂x ∂x3
One must now solve this third order ODE supplemented with the boundary
condition ∫ ∞
ûL (s, 0) = e−st g0 (t)dt (2.15)
0
The starting point for solving equation (2.14) is to seek an exponential
solution of the homogeneous version of (2.14). Letting uL (s, x) = exp(Ω(s)x)
we nd that Ω(s) satises the cubic equation

Ω(s)3 + Ω(s) + s = 0

In summary, the sine transform pair, in contrast to the Fourier transform

pair, has a very limited applicability. Furthermore, the representation (2.12)
obtained via this transform is not uniformly convergent at x = 0, and is not
of the form (2.6). The Laplace transform could in principle be applied to

5
PDEs involve higher derivatives, but, it has the disadvantage that involves
0 < t < ∞, and also it requires the analysis of high order nonlinear algebraic
equations.
It turns out that there does exist the proper analogue of the Fourier
transform pair for solving evolution PDEs: in the next section we will de-
ne a representation which is both uniformly convergent at x = 0 and it
is of the form (2.6). Furthermore, analogous representations exist for the
solution of a general evolution PDE.

6
3 The Heat Equation on the Half-Line via the
Fokas Method
The new method involves three steps. The rst step is identical with the
procedure used for the implementation of the usual transforms, whereas the
third step involves only algebraic manipulations; the second step requires
the use of Cauchy’s theorem.
1a. Given a domain, derive the Global Relation (GR), which is an equa-
tion coupling the function and its derivatives on the boundary of the domain.

Figure 3.1

For the domain

Ω = 0 < x < ∞, t > 0, (3.1)
the GR is
2
eλ t û(−iλ, t) = uˆ0 (−iλ) − g̃1 (λ2 , t) − iλg̃0 (λ2 , t), =λ ≤ 0, (3.2)

where
∫ ∞
û(−iλ, t) = e−iλx u(x, t)dx, t > 0, =λ ≤ 0, (3.3)
0
∫ ∞
uˆ0 (−iλ) = e−iλx u0 (x)dx, t > 0, =λ ≤ 0, (3.4)
0
∫ t ∫ t
λτ
g̃0 (λ, t) = e g0 (τ )dτ, g̃1 (λ, t) = eλτ g1 (τ )dτ, t > 0, λ ∈ C, (3.5)
0 0
with
g1 (t) = ux (0, t), g0 (t) = u(0, t), t > 0 (3.6)
Regarding equations (3.3) and (3.4) we note that

e−iλx  = e−iλR x+λI x  = eλI x ,

7
thus, this term is bounded as x → ∞, for λI < 0.
The functions g̃0 and g̃1 are dened for all complex values of λ, whereas
û and û0 are dened for =λ ≤ 0, thus the global relation (3.2) is valid for
=λ ≤ 0.
Conceptually, the simplest way to derive the global relation is to use the
half–Fourier transform, and to follow the same procedure used with the sine
transform. Indeed, let the half–Fourier transform of u(x, t) be dened by
(3.3).

Then, ∫ ∫
∞ ∞
−iλx
ût = e ut dx = e−iλx uxx dx
0 0
= ux e−iλx ∞
0 + iλue −iλx ∞
0 − λ2 û
Thus,
ût + λ2 û = −g1 (t) − iλg0 (t)
Hence,
2 2
(ûeλ t )t = −eλ t (g1 (t) + iλg0 (t)),
or ∫ t
λ2 t 2
ûe = û0 − eλ τ [g1 (τ ) + iλg0 (τ )]dτ,
0
which is the GR.
2. Express the solution as an integral in the complex λ-plane involving
û0 (−iλ), as well as the t-transforms of all the relevant boundary values.
For the heat equation formulated on the half-line, we nd
∫ ∞ ∫
1 iλx−λ2 t 1 2  
u(x, t) = e û0 (−iλ)dλ− eiλx−λ t g̃1 (λ2 , t) + iλg̃0 (λ2 , t) dλ,
2π −∞ 2π ∂D+
(3.7)
+
where the contour ∂D is the boundary of the domain D dened by+

 
D + = =λ ≥ 0, <λ2 < 0 , (3.8)
see gure 3.2.
Indeed, solving the global relation (3.2) for û(−iλ, t) and then using the
inverse Fourier transform formula, we nd an expression similar to (3.7) but
with the contour of integration along the real line instead of ∂D + . In order
to deform from the real line to ∂D + we use Cauchy’s theorem and Jordan’s
Lemma. We rst consider the function
∫ t
iλx−λ2 t 2
e 2
g̃1 (λ , t) = e iλx
e−λ (t−τ ) g1 (τ )dτ,
0

8
 Figure 3.2

which is an analytic function of λ. This function involves the two exponen-

tials
2 2 2
eiλx = eiλR x−λI x , e−λ (t−τ ) = e−<(λ )(t−τ )−i=(λ )(t−τ ) ,
thus since x ≥ 0 and t − τ ≥ 0, the above exponentials are bounded as
λ → ∞ if λ satises =λ ≥ 0 and <λ2 ≥ 0. Furthermore, integration by
parts implies that the above function is of O(1λ2 ) as λ → ∞:
∫ t
2t 2 g1 (t)
e−λ eλ τ g1 (τ )dτ ∼ , λ → ∞
0 λ2

Thus, Cauchy’s theorem in the domain bounded by the real line and ∂D +
implies that the integral of the above function can be deformed from R to
∂D + .
The situation is similar with the term iλ exp[iλx − λ2 t]g̃0 (λ2 , t), but now
because of the λ factor this function is of O(1λ) as λ → ∞, thus we need
to supplement Cauchy’s theorem with Jordan’s lemma.
3. For given boundary conditions, by employing the global relation as
well as certain invariant transformations, eliminate from the integral repre-
sentation obtained in step 2 the transforms of the unknown boundary values.
Consider for example the Dirichlet problem of the heat equation formu-
lated on the half line, i.e., equation (2.8) supplemented with the initial and
boundary conditions (2.9). In this case, the functions û0 and g̃0 appearing in
the global relation (3.2) are known but the functions û and g̃1 are unknown.
The global relation is valid for =λ ≤ 0, whereas we need g˜1 for λ ∈ ∂D + ,
thus we need to compute g˜1 for =λ ≥ 0. We note that the transformation

9
λ → −λ has two crucial properties: rst, it maps the domain =λ ≤ 0 to the
domain =λ ≥ 0, and also leaves g˜0 (λ2 , t) and g˜1 (λ2 , t) invariant. Using this
transformation, the GR yields
2
eλ t û(iλ, t) = û0 (iλ) − g˜1 (λ2 , t) + iλg˜0 (λ2 , t), =λ ≥ 0 (3.9)

Our strategy will be to use equation (3.9) to eliminate g̃1 ; in this procedure
we ignore the fact that û(iλ, t) is unknown since it will turn out that its
contribution to u(x, t) vanishes. Solving (3.9) for g̃1 (λ2 , t) we nd
2
g̃1 = iλg̃0 + û0 (iλ) + eλ t û(iλ, t), =λ ≥ 0 (3.10)

Replacing in equation (3.7) g̃1 with the RHS of (3.10) we nd

u(x, t) =
∫ ∞ ∫
1 iλx−λ2 t 1 2  
e û0 (−iλ)dλ − eiλx−λ t 2iλg̃0 (λ2 , t) + û0 (iλ) dλ
2π −∞ 2π ∂D+
(3.11)
The term exp(λ2 t)û(iλ, t) gives rise to the term
∫
1
− eiλx û(iλ, t)dλ, 0 < x < ∞, t > 0,
2π ∂D+

which vanishes, since both exp(iλx) and û(iλ, t) are bounded and analytic in
the upper half of the complex λ plane, and furthermore û(iλ, t) is of O(1λ)
as λ → ∞:
∫ ∞
u(0, t)
û(iλ, t) = eiλx u(x, t)dx ∼ − , λ → ∞
0 iλ

Thus, Cauchy’s theorem supplemented with Jordan Lemma in the do-

main D + imply the desired result.

Remarks
(a) Suppose that the heat equation is valid for 0 < t < T . Let

g̃0 (λ) = g̃0 (λ, T ), g̃1 (λ) = g̃1 (λ, T ) (3.12)

Then, equation (3.7) is equivalent to the equation

u(x, t) =

10
 ∫ ∞ ∫
1 iλx−λ2 t 1 2t  
e û0 (−iλ)dλ − eiλx−λ g̃1 (λ2 ) + iλg̃0 (λ2 ) dλ
2π −∞ 2π ∂D +
(3.13)
Indeed, the RHS of equation (3.7) and the RHS of equation (3.13) dier by
the term
∫ ∫ T ∫ T 
1 iλx λ2 (τ −t) λ2 (τ −t)
e e g1 (τ )dτ + iλ e g0 (τ )dτ dλ,
2π ∂D+ t t

and Cauchy’s theorem supplemented with Jordan’s lemma imply that the
above term vanishes.
Similarly, equation (3.11) is equivalent to the equation

u(x, t) =
∫ ∞ ∫
1 iλx−λ2 t 1 2  
e û0 (−iλ)dλ − eiλx−λ t 2iλg̃0 (λ2 ) + û0 (iλ) dλ
2π −∞ 2π ∂D+
(3.14)
This equation is of the Ehrenpreis form (2.6). Actually, the Fokas method
always yields representations of this form. The advantage of (3.14) is that
2
the only (x, t) dependence of the RHS of this equation is of the form eiλx−λ t ,
thus it immediately follows that the function u dened in (3.14) satises the
heat equation. On the other hand, (3.7) is consistent with causality, since
the function u(x, t) cannot depend on the values of g0 (τ ) for τ > t.

(b) In deriving (3.7), the real line was deformed to ∂D + . This defor-
mation is always possible before using the global relation. However, after
using the global relation we introduce û0 and then it is not always possible
to return to the real axis. Actually, the cases where there do exist usual
transforms, are precisely the cases where this return is possible.
In the particular case of (3.7), we note that û0 (iλ) is bounded and ana-
lytic in the upper half of the complex λ plane, thus it is possible to return
to the real axis:
∫ ∞ ∫
1 iλx−λ2 t i ∞ iλx−λ2 t
u(x, t) = e [û0 (−iλ) − û0 (iλ)] dλ− λe g̃0 (λ2 , t)dλ
2π −∞ π −∞

Splitting the integral along R to an integral from −∞ to 0 plus an integral

from 0 to ∞, and letting λ → −λ in the former integral we obtain the
representation obtaned in section 2 via the sine transform. An easier way to
obtain this representation is to recall that the global relation together with

11
the equation obtained from the global relation after replacing λ with −λ are
the following equations:
2
eλ t û(−iλ, t) = û0 (−iλ) − g̃1 − iλg̃0 , =λ ≤ 0,
2
eλ t û(iλ, t) = û0 (iλ) − g̃1 + iλg̃0 , =λ ≥ 0 (3.15)
If λ is real, then both these equations are valid. Hence if g0 is given, we
subtract equations (3.15) and we obtain the equation for the sine transform
of u(x, t). Similarly, if ux (0, t) is given, we add equations (3.15) and we
obtain
2
eλ t ûc (λ, t) = û0c (−iλ) − g̃1 (λ2 , t), λ ∈ R,
where ûc and û0c denote the cosine transform of u(x, t) and u0 (x) respec-
tively, namely:
∫ ∞ ∫ ∞
ûc (λ, t) = cos(λx)u(x, t)dx, û0c (λ) = cos(λx)u0 (x)dx
0 0

(c) Equation (3.14) immediately implies that u(x, t) satises the heat
equation. Furthermore, evaluating (3.14) at t = 0 we nd
∫ ∞ ∫
1 iλx 1
u(x, 0) = e û0 (−iλ)dλ − eiλx û0 (iλ)dλ, x > 0
2π −∞ 2π ∂D+

Jordan’s lemma implies that the second integral in the above expression
vanishes and hence by recalling the denition of û0 (−iλ) and employing the
inverse Fourier transform formula we nd u(x, 0) = u0 (x).
Evaluating (3.14) at x = 0 we nd

u(0, t) =
∫ ∞ ∫  ∫
1 −λ2 t −λ2 t 1 2
e û0 (−iλ)dλ − e û0 (iλ)dλ − 2iλe−λ t g̃0 (λ2 )dλ
2π −∞ ∂D + 2π ∂D +
(3.16)
By deforming the second integral to the real axis and then replacing λ with
−λ we nd that the rst two terms in the RHS of (3.16) cancel. Furthermore,
letting iλ2 = l in the last integral in the RHS of (3.16) we nd
∫ ∞ ∫ T
1  
u(0, t) = eilt e−ilτ g0 (τ )dτ dl = g0 (t)
2π −∞ 0

12
 Numerical Evaluations[11]
For the simple cases when the transforms of the given data can be com-
puted explicitly, the numerical evaluation of the solution obtained by the
Fokas method reduces to the computation of a single integral in the com-
plex λ-plane. Using simple contour deformations, it is possible to obtain an
integrand which decays exponentially as λ → ∞.

Example Consider the heat equation on the half line with

2
u(x, 0) = e−a x , u(0, t) = cos(bt), a, b real constants

Then, ∫ ∞
2 1
û0 (−iλ) = e−iλx−a x dλ = ,
0 iλ + a2
∫  
t (λ+ib)t − 1 (λ−ib) t − 1
1 e e
g̃0 (λ, t) = eλτ cos(bτ )dτ = + 
0 2 λ + ib λ − ib

Hence (3.7) becomes

∫ ∞ 2
eiλx−λ t
2πu(x, t) = dλ
−∞ iλ + a2
∫  
iλx−λ2 t 1 iλ  (λ2 +ib)t  iλ  (λ2 −ib)t 
− e + e −1 + e − 1 dλ
∂D + −iλ + a2 λ + ib λ − ib
The term exp(iλx) in the integrand of the second integral decays as λ → ∞,
but the term exp(−λ2 t) oscillates. However, if we deform ∂D + to a contour
L between the real line and ∂D + , then we achieve exponential decay in both
exp(iλx) and exp(−λ2 t):
∫    
ibt − e−λ2 t −ibt − e−λ2 t

2 1 1 e e
2πu(x, t) = eiλx−λ t + + iλeiλx + dλ,
L iλ + a2 iλ − a2 λ + ib λ − ib

where L depicted in Figure 3.3.

The above integral can be computed using the demand of MATLAB.
With the command we have stated in the homework

13
 Figure 3.3

4 The Heat Equation on the Finite-Interval via

the Fokas Method
We now solve the heat equation formulated on the nite interval

0 < x < L (4.1)

In what follows, we implement steps 1, 2, and 3, of section 3.

Figure 4.1

Step 1. In analogy with equation (3.2) we now have

2
[ ]
eλ t û(−iλ, t) = û0 (−iλ)−g̃1 (λ2 , t)−iλg̃0 (λ2 , t)+e−iλL h̃1 (λ2 , t) + iλh̃0 (λ2 , t) ,

λ ∈ C, (4.2)
where û and û0 are the nite Fourier transforms of u(x, t) and u0 (x), dened
by
∫ L ∫ L
−iλx
û(−iλ, t) = e u(x, t)dx, û0 (−iλ) = e−iλx u0 (x)dx, λ ∈ C,
0 0
(4.3)

14
g̃1 , g̃0 are dened in (3.5) and h̃1 , h̃0 are dened by
∫ t ∫ t
λτ
h̃0 (λ, t) = e h0 (τ )dτ, h̃1 (λ, t) = eλτ h1 (τ )dτ, t > 0, λ ∈ C,
0 0
(4.4)
with h0 (t) = u(L, t), h1 (t) = ux (L, t), t > 0
In order to derive (4.2) we consider the nite Fourier Transform of u(x, t)
dend in (4.3). Then,
∫ L ∣L ∣L
∣ ∣
ût = e−iλx uxx dx = ux e−iλx ∣ + iλue−iλx ∣ − λ2 û
0 0 0

Thus,
ût + λ2 û = −g1 (t) − iλg0 (t) + e−iλL (h1 (t) + λh0 (t))
Hence
2 2 2 t−iλL
(ûeλ t )t = −eλ t (g1 (t) + iλg0 (t)) + eλ (h1 (t) + iλh0 (t)),

which upon integration implies (4.2)

Figure 4.2

Step 2. Solving (4.2) for û(−iλ, t), using the inverse Fourier transform
formula, well as deforming from R to ∂D + in the integral involving g˜1 , g˜0 ,
and from R to ∂D − in the integral involving h˜1 , h˜0 , we nd
∫ ∞ ∫
1 iλx−λ2 t 1 2
u(x, t) = e û0 (−iλ)dλ − eiλx−λ t [g̃1 + iλg̃0 ] dλ
2π −∞ 2π ∂D+

15
 ∫ [ ]
1 2t
− e−iλ(L−x)−λ h̃1 + iλh̃0 dλ, (4.5)
2π ∂D −

where D − is the reection of D + with respect to the real axis and D − is to

the left of the increasing direction of ∂D − , see gure 4.2.

Step 3. The transformation λ → −λ together with the global relation

(4.2) yields two equations. Since there exist four unknown boundary val-
ues (two at each end of the domain), we require two boundary conditions.
However, we cannot assign these conditions in an arbitrary manner. It can
be shown that the terms arising from û(±iλ, t) are bounded as λ → ∞ in
the relevant domains D + and D − , if and only if one boundary condition is
prescribed at each end of the domain.

Example

u(0, t) = g0 (t), u(L, t) = h0 (t)

The global relation (4.2) can be written in the form
2
eλ t û(−iλ, t) = G(λ, t) − g̃1 + e−iλL h̃1 , (4.6)

where the known function G is dened by

G(λ, t) = û0 (−iλ) − iλg̃0 (λ2 , t) + iλe−iλL h̃0 (λ2 , t) (4.7)

Letting λ 7→ −λ in (4.6), and recalling that g̃1 and h̃1 are invariant with
respect to λ 7→ −λ, we obtain
2
eλ t û(iλ, t) = G(−λ, t) − g̃1 + eiλL h̃1  (4.8)

Solving equations (4.6) and (4.8) for g̃1 and h̃1 , we nd

g̃1 =
1 { [ ]}
iλL −iλL λ2 t −iλL iλL
e G(λ, t) − e G(−λ, t) + e e û(iλ, t) − e û(−iλ, t) ,
eiλL − e−iλL
(4.9)
1 { 2
}
h̃1 = iλL G(λ, t) − G(−λ, t) + eλ t [û(iλ, t) − û(−iλ, t)] 
e − e−iλL
(4.10)
We next substitute g̃1 and h̃1 in (4.5). We claim that the terms involving
û(±iλ, t) yield a zero contribution. Indeed, since this is a well-posed BVP,

16
the relevalt terms are bounded as λ → ∞. Let us verify this explicitly; the
term in g̃1 involves the following contribution from û(±iλ, t):

e−iλL û(iλ, t) − eiλL û(−iλ, t)


eiλL − e−iλL

Since =λ ≥ 0, e−iλL grows, and then the above expression, as λ → ∞,

becomes ∫ L
−û(iλ, t) + eiλL eiλ(L−x) u(x, t)dx,
0

which is clearly bounded as λ → ∞ with =λ ≥ 0. Similarly the term in h̃1

involves the following contribution from û(±iλ, t):

û(−iλ, t) − û(iλ, t)
,
eiλL − e−iλL
which as λ → ∞, =λ ≤ 0, simplies to the expression
∫ L
e−iλ(L−x) u(x, t)dx − e−iλL û(iλ, t),
0

which is clearly bounded as λ → ∞, =λ ≤ 0.

We also note that the zeros of exp(iλL) − exp(−iλL) occur on the real
axis, and hence are outside D except for λ = 0 which is a removable singu-
larity, since
[ ]
e−iλL û(−iλ, t) − eiλL û(iλ, t) = [û(−iλ, t) − û(iλ, t)]λ=0 = 0
λ=0

Thus, (4.5) becomes

∫ ∞
1 2
u(x, t) = eiλx−λ t û0 (λ)dλ
2π −∞
∫  
1 2t eiλL G(λ, t) − e−iλL G(−λ, t)
− eiλx−λ iλg̃0 (λ2 , t) + dλ
2π∂D + eiλL − e−iλL
∫  
1 iλx−λ2 t 2 G(λ, t) − G(−λ, t)
− e iλh̃0 (λ , t) + dλ, (4.11)
2π ∂D− eiλL − e−iλL
where ∂D + and ∂D − are detected in Figure 4.2.

Remarks 4.1.

17
1. It is possible to deform ∂D + and ∂D − back to the real axis and then
using the residue theorem the usual sine-sine solution can be rederived.
A simpler way to obtain the usual solution representation is to subtract
(4.6), (4.8):
∫ L
λ2 t
2ie sin(λx)u(x, t)dx = (eiλL −e−iλL )h̃1 (λ2 , t)+G(−λ, t)−G(λ, t)
0
(4.12)
The unknown function h̃1 can be eliminated by evaluating the above
equation at those values of λ for which the coecient of h̃1 vanishes:
nπ
eiλL − e−iλL = 0, λ = , n = 0, 1, 2,    
L
Hence (4.12) becomes
 2 ∫ L 
nπ nπx   nπ   nπ 
2ie L t sin u(x, t)dx = G − , t − G ,t ,
0 L L L
and then the usual representation follows using the following transform
pair: ∫
2 L  nπx 
fn = sin f (x)dx, n = 1, 2,   
L 0 L
∞
∑  nπx 
f (x) = fn sin 
L
n=1

2. The function eiλL − e−iλL , which appears in (4.11), has simple poles at
the points nπL which occur on the real axis. Thus, the classical rep-
resentational is formulated on the worst part of the complex plane.
Perhaps this is related with the fact that this classical representation
is not uniformly convergent at x = 0 and x = L.

3. The numerical implementation of the Fokas method for evolution PDEs

on the nite interval is discussed in [12].
Example

ux (0, t) − γu(0, t) = gR (t), ux (L, t) = 0, γ > 0

The classical representational involves a series area, over λ∞

1 , where
are the real zeros of the transcendental equation
∆(λ) = (iλ − γ)e−iλL − (iλ + γ)eiλL  (4.13)

18
This series is not uniformly convergent at x = 0 and x = L.
On the other hand, the Fokas method yields a solution which is similar
to (4.11):
∫ ∞ ∫
1 iλx−λ2 t 1 2 g̃(λ)
u(x, t) = e uˆ0 (−iλ)dλ − eiλx−λ t dλ
2π −∞ 2π ∂D+ ∆(λ)

∫
1 2t h̃(λ)
− eiλx−λ dλ, (4.14)
2π ∂D − ∆(λ)
where û0 (−iλ) is the Fourier transform of u0 (x), ∆(λ) is dened by
(4.13), ∂D + and ∂D − are dened as in (4.5) and the transforms g̃, h̃
are explicitly given in and terms of uˆ0 (±iλ), and or gR which is the
t–transforms of gR :

g̃(λ) = 2iλe−iλL g̃R (λ2 ) − (iλ + γ)(eiλL û0 (−iλ) + e−iλL û0 (iλ),

and

h̃(λ) = 2iλg̃R (λ2 ) − (iλ − γ)û0 (−iλ) − (iλ + γ)û0 (iλ)

19
5 Elliptic equation in the Interior of a Convex
Polygon
The most important elliptic PDEs are the Laplace, the modied Helmholtz
and the Helmholtz equations. The Laplace equation is:

uxx + uyy = 0 (5.1)

If u satises the Laplace equation (5.1), then u is called a harmonic

function. Traditionally, harmonic functions are associated with the real and
imaginary parts of an analytic function. However, there is an alternative
direct way to associate harmonic and analytic functions:
the function u(x, y), which may be complex, is harmonic if and only if
uz is analytic.
Indeed, if uz is analytic then uz z̄ = 0, i.e., u is harmonic; the inverse is
also true.

The Global Relation

Recall that the rst step of the Fokas method consists of deriving the
global relation.
For elliptic PDEs involving second order derivatives, we need two global
relations. However, if we assume that u is real, then the second global
relation can be obtained from the rst via complex conjugation.
The simplest way to derive a global relation is to consider the formal
adjoint of the Laplace equation, which is itself,

vxx + vyy = 0 (5.2)

Multiplying equations (5.1) and (5.2) by v and u respectively, and then

subtracting the resulting equations we nd

(vux − uvx )x + (vuy − uvy )y = 0

Letting v = exp(−iλx + λy), which is a particular solution of (5.2) for any

complex constant λ, we nd the family of conservation laws
[ ] [ ]
−iλx+λy −iλx+λy
e (ux + iλu) + e (uy − λu) = 0, λ ∈ C (5.3)
x y

The exponential exp(iλx+λy) provides an other particular solution of (5.2),

and this yields
[ ] [ ]
eiλx+λy (ux − iλu) + eiλx+λy (uy − λu) = 0, λ ∈ C (5.4)
x y

20
We note that if u is real, then equation (5.4) can be obtained from (5.3) by
taking the complex conjugate and then replacing in the resulting equation
λ by λ̄. This procedure is called Schwartz conjugation.
Suppose that the Laplace equation is valid in the domain Ω. Then,
equations (5.3) and (5.4) together with Green’s theorem, imply the following
global relations:
∫
e−iλx+λy [(ux + iλu)dy − (uy − λu)dx] = 0, λ ∈ C, (5.5)
∂Ω

and ∫
eiλx+λy [(ux − iλu)dy − (uy − λu)dx] = 0, λ ∈ C, (5.6)
∂Ω
where ∂Ω denotes the boundary of Ω.
The most well known boundary value problems for elliptic PDEs are
either the Dirichlet problem where u is prescribed on the boundary, or the
Neumann problem where the normal derivative, denoted by uω , is prescribed
on the boundary.
In order to rewrite the global relations in terms of u and uω , we pa-
rameterize the boundary ∂Ω in terms of its arclength which we denote by
s. Then, if uT denotes the derivative of u along the tangent to ∂Ω, and
uω denotes the derivative of u normal to uT in the outward direction, then
dierentiating u(x(s), y(s)) we nd

ux dx + uy dy = uT ds (5.7)

Since the innitesimal vector (dy, −dx) is normal to the innitesimal vector
(dx, dy), we nd
ux dy − uy dx = uω ds (5.8)
Thus, we can rewrite equations (5.5) and (5.6) in terms of u and uω :

(ux + iλu)dy − (uy − λu)dx = uω ds + λu(dx + idy)

Hence, the global relation (5.5) becomes

∫  
−iλx+λy
 dx dy 
e uω + λu +i ds = 0 (5.9)
∂Ω ds ds

Similarly ∫  
iλx+λy
 dx dy 
e uω + λu −i ds = 0 (5.10)
∂Ω ds ds

21
Letting
z = x + iy, z̄ = x − iy, (5.11)
equations (5.9) and (5.10) become
∫
 dz 
e−iλz uω + λu ds = 0, (5.12)
∂Ω ds

and ∫
 dz̄ 
eiλz̄ uω + λu ds = 0 (5.13)
∂Ω ds
A Polygonal Domain
Let Ω be the interior of the polygonal domain specied by the complex
numbers z1 , z2 ,    ,zn , zn+1 = z1 , see gure 5.1.

Figure 5.1

Let Lj denote the side (zj , zj+1 ).

Then, the global relation (5.12) becomes
n
∑ n
∑
Ŵj + λ D̂j = 0, λ ∈ C, (5.14)
j=1 j=1

where Wj n1 denote the transforms of the Neumann boundary values and
Dj n1 denote the transforms of the Dirichlet boundary values:
∫ zj+1
Ŵj = e−iλz uwj ds, j = 1, 2,    , n, λ ∈ C (5.15)
zj

22
and ∫ zj+1
dz
D̂j = e−iλz uj ds, j = 1, 2,    , n, λ ∈ C (5.16)
zj ds
If u is real, then instead of analysing the global relation (5.13), we can
analyse the complex conjugate of equation (5.12). Thus, for real u, equation
(5.12) and its complex conjugate provide two equations for n unknown func-
tions, since for a well posed problem only one boundary condition is given
on each side. This situation appears ominus, however in equation (5.12) the
complex constant λ is arbitrary, thus in this sense equation (5.12) contains
innitely many equations. It turns out that this observation provides a
most ecient way for the numerical integration of this problem.

Approximate Global Relations

The numerical solution of the global relations for determining the un-
known boundary values involves the following two steps [13]-[15]:
[ ]u
∂u
1. Expand the function [uj ]n1 and ∂wj in terms of N basis functions
1
−1
denoted by Sl (t)N
0 :

N
∑ −1 N
∑ −1
∂uj (t)
uj (t) ≈ ajl Sl (t), ≈ bjl Sl (t), j = 1, 2,    , n
∂ω
l=0 l=0

A convenient such basis is given by the Legendre polynomials of order l,

denote by Pl .
Let Ŝl (λ) denote the Fourier transform of Sl (t), namely
∫ 1
Ŝl (λ) = e−iλt Sl (t)dt, λ ∈ C (5.17)
−1

For the Legendre polynomials the relevant Fourier transform can be com-
puted explicitly,
∫ 1 l
∑  
−iλt (l + k)! (−1)l+k eiλ − e−iλ
e Pl (t)dt = i  (5.18)
−1 (l − k)!k! (2iλ)k+1
k=0

Then, the global relation and its complex conjugate yield two equations
involving the constants ajl and bjl . By evaluating these equations at appro-
priately chosen values of λ called collocation points, we can solve for the
unknown coecients.

23
 Example
Consider the Laplace equation in the interior of the square with corners

z1 = −1 + i, z2 = −1 − i, z3 = 1 − i, z4 = 1 + i

Then, the global relation (5.12) involves the following terms:

∫ −1 [ ]
iλ
û1 (λ) = e eλy u(1)
x + iλu
(1)
dy,
1
∫ 1 [ ]
û2 (λ) = e−λ e−iλx −u(2)
y + λu (2)
dx,
−1
∫ 1 [ ]
−iλ
û3 (λ) = e eλy u(3)
x + iλu (3)
dy,
−1
∫ −1 [ ]
û4 (λ) = eλ e−iλx −u(4)
y + λu
(4)
dx (5.19)
1
Let
∫ 1 ∫ 1
λt
Ŵ (λ) = e W (t)dt, D̂(λ) = eλt D(t)dt, λ ∈ C, (5.20)
−1 −1

where W (t) and D(t) denote Neumann and Dirichlet boundary values re-
spectively. Then,
[ ]
û1 (λ) = −eiλ Ŵ1 (λ) + iλD̂1 (λ) ,
[ ]
û2 (λ) = e−λ Ŵ2 (−iλ) + λD̂2 (−iλ) ,
[ ]
û3 (λ) = e−iλ Ŵ3 (λ) + iλD̂3 (λ) ,
[ ]
û4 (λ) = eλ Ŵ4 (−iλ) − λD̂4 (−iλ)  (5.21)

The approximate global relation yields

û1 (λ) + û2 (λ) + û3 (λ) + û4 (λ) = 0, λ ∈ C, (5.22)

where
N
∑ −1 [ ]
û1 (λ) ≈ −eiλ iλal1 P̂l (λ) + bl1 P̂l (λ) ,
l=0

24
 N
∑ −1 [ ]
û2 (λ) ≈ e−λ λal2 P̂l (−iλ) + bl2 P̂l (−iλ) ,
l=0
N
∑ −1 [ ]
−iλ
û3 (λ) ≈ e iλal3 P̂l (λ) + bl3 P̂l (λ) ,
l=0
N
∑ −1 [ ]
û4 (λ) ≈ eλ −λal4 P̂l (−iλ) + bl4 P̂l (−iλ)  (5.23)
l=0

2. For a given side, choose λ in such a way that for the given side we
obtain the usual Fourier transform (FT) of the Legendre functions, whereas
the contribution from the remaining sides vanishes as λ → ∞ It turns out
that for a convex polygon such a choice is always possible).

• side 1. Multiply (5.22) by e−iλ and then let λ = −iρ, ρ > 0.

[ ]4
We nd the following forms for Ŵj (and similarly for D̂j :
1

Ŵ1 (−iρ), eiρ e−ρ Ŵ2 (−ρ), e−2ρ Ŵ3 (−iρ), e−iρ e−ρ Ŵ4 (−ρ)

The rst terms involve the FT, whereas the remaining terms vanish as ρ →
∞. This is obvious for the third term, whereas the second and the fourth
terms involve the integral
∫ 1
e−ρ(1+t) w(t)dt;
−1

since −1 < t < 1, it follows that 1 + t > 0, thus exp[−ρ(1 + t)] vanishes as
ρ → ∞.

• side 2. Multiply by eλ and then let λ = −ρ, ρ > 0.

• side 3. Multiply by eiλ and then let λ = iρ, ρ > 0.

• side 4. Multiply by e−λ and then let λ = ρ, ρ > 0.

R
For ρ we can use the discrete values ρ = M m, m = 1, 2,    , M , R > 0,
where RM determines how close are the collocation points.

25
 It is found numerically [15] that the following rules for low condition
number:
R
≥ 2, M ≥ N n
m
The above numerical technique can be viewed as the counterpart in the
complex Fourier plane of the boundary integral method (which is formulated
in the physical plane).

26
6 Modied Helmholtz and Helmholtz Equations
For the other two basic elliptic equations the situational is similar. In par-
ticular, for the modied Helmholtz equation,

uxx + uyy − k 2 u = 0, (x, y) ∈ Ω; k > 0, (6.1)

the global relation is given by

∫ [ ] ( )
ik z̄(t)
−λz(t) ku dz(t) 1 dz̄(t)
e 2 λ
uω + λ + dt = 0, k > 0, λ ∈ C\0
∂Ω 2 dt λ dt
(6.2)
For the case that Ω is the interior of the polygon with corners at z1 , z2 ,    , zn ,
equation (6.2) becomes
n
∑
ûj (λ) = 0, λ ∈ C\0, (6.3)
j=1

where ûj (λ) is dened by

∫ zj+1 ( ) 
−i k2 (λz− λ
z̄
) k k
ûj (λ) = e uz + i λu dz − (uz̄ + u)dz̄ , j = 1, · · · , n
zj 2 2iλ
(6.4)
A second global relation is obtained from equation (6.2) by replacing λ with
1λ. If u is real, we can obtain the second global relation by taking the
Schwarz conjugate of equation (6.2).
Remark Recall that for evolution PDE’s the second step of the Fokas
method involves the derivation of an integral representation, dened in the
complex plane which depends a all boundary values. This step can also be
implemented for the three basic elliptic PDEs dened in the interior of a
convex polygonal. For example, by employing either the classical Green’s
representation formula, or by performing the spectral analysis of the associ-
ated Lax pair [16], we nd the following novel integral representation:
n ∫
1 ∑ k z̄ dλ
u(z, z̄) = ei 2 (λz− λ ) ûj (λ) , z ∈ Ω, (6.5a)
4iπ lj λ
j=1

when ûj n1 are dened in (6.4) in terms of all boundary values and lj n1
are the rays in the complex λ-plane oriented towards innity and dened by

lj = λ ∈ C : arg λ = − arg(zj+1 −zj ), j = 1, · · · , n, zn+1 = z1  (6.5b)

27
 For simple domains, it is possible, to impement step 3 of the Fokas
method: using the global relations and their invariant properties, it is pos-
sible to express all transforms in terms of the given boundary data, using
only algebraic manipulations. This has led to the analytic solution of several
BVPs for which the usual approaches apparently fail [16].
For more complicated domains, the global relations suggest the novel
numerical technique for the determination of the unknown boundary values,
discussed earlier.

Further development

The rigorous foundation of the new method for linear forced evolution
PDEs in Sobolev spaces is presented in [4], [17]. These results actually lead
to a new approach for proving well posedness for nonlinear IBVPs. The
crucial ingredient of this approach is to use for the linear version of the
given nonlinear PDE, the formulae obtained via the new method. Earlier
authors have been able to prove well posedness for IBVPs using ideas similar
to those used in the treatment of initial-value problems. In particular, one
rst obtains a solution formula for the linear IBVP with forcing and then
uses this formula to derive appropriate linear estimates. Subsequently, one
replaces the forcing in the linear formula by the nonlinearity and uses the
linear estimates together with a contraction mapping argument to deduce
well-posedness of the nonlinear IBVP.
It is often the case, however, that even the derivation of the linear so-
lution formula is somewhat technical and unintuitive, not to mention the
derivation of the relevant linear estimates. The main advantage of the new
method is that it yields explicit formulae for forced linear evolution equa-
tions with arbitrary number of derivatives. Thus, it is not surprising that
these naturally emerging linear formulae can be used to establish local
well-posedness of nonlinear evolution IBVPs through a contraction mapping
approach.
Anthony Ashton employing the new method has developed a remarkable
formalism for the rigorous analysis of elliptic PDEs, see for example [18].
For recent results regarding the characterization of the Dirichlet to Neu-
mann map for integrable nonlinear evolution PDEs, see for example [19]-[22].
Linear evolution PDEs with either non-separable or other complicated
boundary conditions are analyzed in [23]–[27].
The new method can be extended to three dimensions, see for example
[5], [28]–[29]
Reviews of the Fokas method for linear and for integrable nonlinear PDEs

28
are presented in [30] and [31], respectively.

29
References
[1] A. S. Fokas, A unied transform method for solving linear and certain nonlinear
PDEs, Proc. Roy. Soc. London Ser. A 453 (1997), 1411–1443.
[2] A. S. Fokas, Two-dimensional linear partial dierential equations in a convex polygon,
R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 457 (2001), no. 2006, 371–393, DOI
10.1098/rspa.2000.0671. MR1848093 (2002j:35084)
[3] A. S. Fokas, A. R. Its, and L.-Y. Sung, The nonlinear Schrödinger equation on the half-
line, Nonlinearity 18 (2005), no. 4, 1771–1822, DOI 10.1088/0951-7715/18/4/019.
MR2150354 (2006c:37074)
[4] A. S. Fokas, A. A. Himonas, and D. Mantzavinos, The nonlinear Schrödinger equation
on the half-line, Trans. Amer. Math. Soc. (2015), DOI 10.1090/tran/6734.
[5] M. J. Ablowitz, A. S. Fokas, and Z. H. Musslimani, On a new non-local formulation of
water waves, J. Fluid Mech. 562 (2006), 313–343, DOI 10.1017/S0022112006001091.
MR2263547 (2007k:76013)
[6] A. S. Fokas and E. A. Spence, Synthesis, as opposed to separation, of variables, SIAM
Rev. 54 (2012), no. 2, 291–324, DOI 10.1137/100809647. MR2916309
[7] S. Vetra-Carvalho and A. S. Fokas, Computation of Spectral Representations for Evo-
lution PDEs, 2010.
[8] B. Fornberg and N. Flyer, A numerical implementation of Fokas boundary integral
approach: Laplaces equation on a polygonal domain, Proc. R. Soc. Lond. Ser. A
Math. Phys. Eng. Sci., posted on 2011, DOI 10.1098/rspa.2011.0032. MR2659495
(2011f:35008)
[9] David M. Ambrose and David P. Nicholls, Fokas integral equations for three di-
mensional layered-media scattering, J. Comput. Phys. 276 (2014), 1–25, DOI
10.1016/j.jcp.2014.07.018. MR3252567
[10] B. Deconinck, T. Trogdon, and V. Vasan, The Method of Fokas for Solving Linear
Partial Dierential Equations, Society for Industrial and Applied Mathematics 56
(2014), no. 2095, 159–186, DOI 10.1137/110821871.
[11] N. Flyer and A. S. Fokas, A hybrid analyticalnumerical method for solving evolution
partial dierential equations. I. The half-line, R. Soc. Lond. Proc. Ser. A Math. Phys.
Eng. Sci. 2095 (2008), DOI 10.1098/rspa.2008.0041.
[12] E. Kesici, B. Pelloni, T. Pryer, and D. Smith, A numerical implementation of the
unied Fokas transform for evolution problems on a nite interval(submitted).
[13] A. S. Fokas, N. Flyer, S. A. Smitheman, and E. A. Spence, A semi-analytical numerical
method for solving evolution and elliptic partial dierential equations, J. Comput.
Appl. Math. 227 (2009), no. 1, 59–74, DOI 10.1016/j.cam.2008.07.036. MR2512760
(2010d:65276)
[14] A. G. Sifalakis, A. S. Fokas, S. R. Fulton, and Y. G. Saridakis, The generalized
Dirichlet-Neumann map for linear elliptic PDEs and its numerical implementation,
J. Comput. Appl. Math. 219 (2008), no. 1, 9–34, DOI 10.1016/j.cam.2007.07.012.
MR2437692 (2009g:35042)
[15] P. Hashemzadeh, A. S. Fokas and S. A. Smitheman, A Numerical Technique for Linear
Elliptic PDEs in Polygonal Domains, Proc. R. Soc. A 471 (2015), no. 2175.

30
[16] A. S. Fokas, A Unied Approach to Boundary Value Problems, SIAM, 78 (2008).
[17] A. S. Fokas, A. A. Himonas, and D. Mantzavinos, The Korteweg-de Vries equation
on the half-line (2015, submitted).
[18] A. C. L. Ashton, On the rigorous foundations of the Fokas method for linear elliptic
partial dierential equations, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 468
(2012), no. 2141, 1325–1331, DOI 10.1098/rspa.2011.0478. MR2910351
[19] Anne Boutet de Monvel, Alexander Its, and Vladimir Kotlyarov, Long-time asymp-
totics for the focusing NLS equation with time-periodic boundary condition on the
half-line, Comm. Math. Phys. 290 (2009), no. 2, 479–522, DOI 10.1007/s00220-009-
0848-7. MR2525628 (2010i:37169)
[20] D. C. Antonopoulou and S. Kamvissis, On the Dirichlet to Neumann problem for the
1D cubic NLS equation, (preprint).
[21] J. Lenells and A. S. Fokas, The nonlinear Schrodinger equation with t-periodic data:
I. Exact results, Proc. R. Soc. A 471 (2015), no. 2181.
[22] J. Lenells and A. S. Fokas, The nonlinear Schrodinger equation with t-periodic data:
II. Perturbative results, Proc. R. Soc. A 471 (2015), no. 2181.
[23] Dionyssios Mantzavinos and Athanassios S. Fokas, The unied method for the
heat equation: I. Non-separable boundary conditions and non-local constraints
in one dimension, European J. Appl. Math. 24 (2013), no. 6, 857–886, DOI
10.1017/S0956792513000223. MR3181485
[24] B. Deconinck, B. Pelloni and N. E. Sheils, Non-steady-state heat conduction in com-
posite walls, Proc. Roy. Soc. A 470, 2165: 22pp., 2014.
[25] Natalie E. Sheils and Bernard Deconinck, Heat conduction on the ring: interface
problems with periodic boundary conditions, Appl. Math. Lett. 37 (2014), 107–111,
DOI 10.1016/j.aml.2014.06.006. MR3231736
[26] M. Asvestas, A. G. Sifalakis, E. P. Papadopoulou and Y. G. Saridakis, Fokas
method for a multi-domain linear reaction-diusion equation with discontinuous
diusivity, Journal of Physics: Conference Series 490, 012143, doi:10.1088/1742-
6596/490/1/012143, 2014.
[27] A. Its, E. Its and J. Kaplunov, Riemann-Hilbert Approach to the Elastodynamical
Equation, Letters in Mathematical
[28] D. P. Nicholls, A High-Order Perturbation of Surfaces (HOPS) Approach to Fokas In-
tegral Equations: Three Dimensional Layered-Media Scattering, Quarterly of Applied
Mathematics, (to appear).
[29] G. Dassios and A. S. Fokas, Methods for solving elliptic PDEs in spherical coordi-
nates, SIAM J. Appl. Math. 68 (2008), no. 4, 1080–1096, DOI 10.1137/070679223.
MR2390980 (2009h:35038)
[30] A. S. Fokas, Lax pairs: a novel type of separability, Inverse Problems 25 (2009), no. 12,
123007, 44, DOI 10.1088/0266-5611/25/12/123007. MR2565573 (2010k:37112)
[31] Beatrice Pelloni, Advances in the study of boundary value problems for nonlinear
integrable PDEs, Nonlinearity 28 (2015), no. 2, R1–R38. MR3303170

31