Stability of the Inviscid Power-Law Vortex in

Self-Similar Coordinates
arXiv:2411.13397v1 [math.AP] 20 Nov 2024

Matei P. Coiculescu
November 21, 2024

Abstract
We prove that the stationary power-law vortex ω(x) = β|x|−α , which ex-
plicitly solves the incompressible Euler equations in R2 , is linearly stable in
self-similar coordinates with the natural scaling.

1 Introduction
Consider the Cauchy problem of the two-dimensional Euler equations in vorticity
form:
∂t ω + (v · ∇)ω = f
K2 ∗ ω(·, t) = v(·, t) (1)
ω(·, t = 0) = ω0 (·)
Here, f (x, t) is a real-valued forcing term and the vorticity ω(x, t) is a real-valued
function defined on R2 × [0, T ). We consider solutions in the sense of distribution.
In particular, we say ω is a solution of Equation (1) if the following integral identity
holds for every φ ∈ Cc∞ (R2 × [0, T )):
Z T Z


Z
ω(∂t φ + (K2 ∗ ω) · ∇φ) + f φ dxdt = − φ(x, 0)ω0 (x)dx. (2)
0 R2 R2

Here and in the sequel we denote the standard two-dimensional Biot-Savart kernel
by K2 . Although the operator ω → K2 ∗ ω defined on Schwartz functions cannot
be continuously extended to L2 , as shown in [2], it is well behaved on some closed
linear subspaces of L2 . More generally, we shall show that for every 2 ≤ q < ∞ there
exist some closed linear subspaces of Lq , which we denote Lqm , to which the operator

1
can be continuously extended. Here m ≥ 2 is an integer and Lqm will essentially be
the space of m-fold symmetric functions in Lq . A detailed description appears in
subsequent sections.
We shall restrict ourselves to solutions of Equation (1) in a particular class of
integrability. First, for any 1 ≤ q ≤ ∞ and 2 < p ≤ ∞, we define:
Definition 1. The function ω(x, t) is in the class ΥTq,p if and only if

ω ∈ Lqt ([0, T ], (L1 ∩ Lp )x )

K2 ∗ ω ∈ Lqt ([0, T ], L2x )

Definition 2. The function ω(x) is in the class Υ0p if and only if

ω ∈ (L1 ∩ Lp )x

K2 ∗ ω ∈ L2x
T
Definition 3. The function ω(x, t) is in the class Υ∞
q,p if ω ∈ Υq,p for all T ≥ 0.

We also use a slightly different notation if the domain of times does not include
the time t = 0:
[a,b]
Definition 4. Let a, b ∈ R. The function ω(x, t) is in the class Υq,p if and only if

ω ∈ Lqt ([a, b], (L1 ∩ Lp )x )

K2 ∗ ω ∈ Lqt ([a, b], L2x ).

[a,∞) [a,b]
We also say that ω ∈ Υq,p if and only if ω ∈ Υq,p for every b > a.
0
One generally searches for solutions in Υ∞
∞,p with initial data in Υp . The famous
theorem of Yudovich, proven in [10], is:
Theorem 1.1. Let ω0 (x) ∈ Υ0∞ and let f be some forcing term such that f ∈ Υ∞ 1,∞ .
∞
Then there exists a unique solution ω(x, t) to Equation (1) in the class Υ∞,∞ with
initial data ω0 .
In a remarkable couplet of papers [7] and [8], Vishik provides the first evidence
that the so-called “Yudovich Class” L1 ∩ L∞ is sharp, proving:
Theorem 1.2. For every 2 < p < ∞, there exists ω0 (x) ∈ Υ0p and a force f ∈
Υ∞ ∞
1,p with the property that there are uncountably many solutions ω(x, t) ∈ Υ∞,p to
Equation (1) with initial data ω0 .

2
 The monograph [2] provides an alternative proof of Vishik’s theorem while follow-
ing a similar approach, and we shall generally use the notation and terminology from
[2]. Succinctly, one may describe Vishik’s general strategy as constructing an un-
stable radial vortex in self-similar coordinates that generates non-uniqueness while
breaking radial symmetry. We now discuss the interpretation of Vishik’s proof in
terms of dynamical systems proposed by the authors of the monograph [2].
We first observe that Equation (1) admits many stationary solutions in the form
of “radial vortices” or vortex profiles of the form:

ω(x) = g(|x|), v(x) = ζ(|x|)x⊥ ,

where x⊥ = (−x2 , x1 ) and K2 ∗ ω = v. Suppose one could find a vortex profile ω that
is linearly unstable, for instance that one finds a real and strictly positive eigenvalue λ
of the linearized Euler equations and a trajectory on the unstable manifold associated
to λ and ω of the form
ω = ω + ωlin + o(eλ t).
Here ωlin = eλt η(x) is a solution of the linearized Euler equations. We would then
expect “non-uniqueness at time t = −∞” because of the instability of the vortex.
The outline proposed in [2] is to instead consider a choice of self-similar coordinates
for which self-similar solutions are sufficiently integrable and to find an unstable
stationary vortex profile in the self-similar coordinates. Given a positive parameter
α > 0, one such choice of coordinates is given by

ξ = xt−1/α , τ = log(t)

v(x, t) = t1/α−1 V (ξ, τ ), ω(x, t) = t−1 Ω(ξ, τ ).

The Euler equations in these similarity variables, without force, are given by
ξ
∂τ Ω − (1 + · ∇ξ )Ω + V · ∇ξ Ω = 0
α
V = K2 ∗ Ω.
If a self-similar profile Ω satisfies kΩkLp = O(1) as τ → −∞, then we also have
2
that kωkLp = O(t−1+ pα ) as t → 0+ . In particular, choosing p = 2/α ensures that
the Lebesgue norms are O(1) in both coordinate systems, which is one reason why
we consider this a “natural” choice of self-similar coordinates. To prove Vishik’s
theorem, one should take 0 < α ≤ 2/p, which ensures the desired integrability. If
one can find an unstable stationary solution Ω of the self-similar equations, then

3
one can hope to prove non-uniqueness at time τ = −∞, which corresponds to non-
uniqueness at physical time t = 0.
It is not difficult to see that the only stationary radial vortices solving the self-
similar Euler equations are precisely the power-law vortices of the form

Ω(ξ) = β(2 − α)|ξ|−α ,

and we note that these correspond exactly to radial power-law vortices solving the
stationary Euler equations in the original coordinates. The exponent α of the sta-
tionary profile is exactly determined by the choice of scaling for the self-similar
coordinates, and the prefactor β is an arbitrary real number. The natural question
arises: are the power-law vortices unstable in the self-similar coordinates? This ques-
tion was also posed by the authors of [2]. An affirmative answer would suggest that
non-uniqueness can arise from the (simple and explicit) power-law vortex, while a
negative answer shows that a more complex stationary profile that necessarily de-
pends on the angular variable would have to be found if there is any hope to complete
the program proposed in [2]. We prove that the power-law vortex is linearly stable,
in a way we shall now make more precise.
We can write the linearization of (3) around Ω as

(∂τ − Lss )Ω = 0.

The domain of Lss as on operator into Lqm is denoted

Dm (Lss ) = {Ω ∈ Lqm : Lss (Ω) ∈ Lqm }.

Thus, Lss is a closed, densely defined operator. We define the resolvent set of an
operator L on Lqm to be the open set of z ∈ C for which L − z has a bounded inverse
from Lqm → Lqm . We define the spectrum of L, which we denote by specm,q (L), to be
the closed set which is the complement of the resolvent set.
Theorem 1.3. Let β ∈ R and α ∈ (0, 1). Let q be such that 2 ≤ q ≤ 2/α and let
m ≥ 2 be an integer. If Ω(ξ) = β(2 − α)|ξ|−α is the radial power-law vortex that
solves the Euler equations in self-similar coordinates, then for any λ ∈ specm,q (Lss ),
2
we have Re(λ) ≤ a0 := 1 − αq ≤ 0.
Using the Hille-Yosida theorem, we have the following related theorem:
Theorem 1.4. Let β ∈ R and α ∈ (0, 1). Let q be such that 2 ≤ q ≤ 2/α and
let m ≥ 2 be an integer. The operator Lss is the generator of a strongly continuous
semigroup on Lqm , which we denote eτ Lss , and the growth bound G(Lss ) of eτ Lss is

4
less than or equal to a0 . As a consequence, for any ǫ there exists a number M(α, ǫ)
depending only on α, ǫ such that

keτ Lss ΩkLq ≤ M(α, ǫ)e(a0 +ǫ)τ kΩkLq ∀τ, ∀Ω ∈ Lqm .

The choice of Lebesgue space Lq for 2 ≤ q ≤ 2/α is justified since one expects
non-uniqueness to emerge from an (integrable) singularity at the spatial origin, and
the power-law vortex Ω is in Lqloc if and only if q < 2/α. Our theorems prove that
the power-law vortex is linearly stable up to the borderline case: 2 ≤ q ≤ 2/α, and
in the case when 2 ≤ q < 2/α, we prove that the power-law vortex is exponentially
stable. We would also like to remark that our result, and indeed all work done
on the non-uniqueness of the incompressible Euler equations with vorticity in the
class Υ∞∞,p , has implications for potential non-uniqueness of Leray-Hopf solutions of
the incompressible Navier-Stokes equations. As shown by the authors of [1] in the
case with a force, if one can construct an unstable vortex for the Euler equations in
self-similar coordinates, one has essentially proven that Leray solutions of the Navier-
Stokes equations admit non-unique solutions. Our result confirms the statement in
[1] that finding such an unstable vortex is “far from elementary”.
In addition, we may view our result in the context of classical results on hy-
drodynamics stability. In the comprehensive work [4] by Chandrasekhar, various
stability and instability results for fluid motion are described, including the well-
known Rayleigh’s criterion for stability of steady inviscid flow between two co-axial
cylinders. Rayleigh’s criterion would suggest, but not rigorously prove in the case
of our infinite energy power-law vortices on the whole plane, that a vorticity profile
|x|−α is stable whenever α < 2. We consider our Theorem 1.3 as mathematical proof
of the expected stability in this case. In fact, the numerology of our theorem (we get
linear stability and local integrability if αq ≤ 2 and we need q ≥ 1 to get a complete
Lebesgue norm), suggests that Rayleigh’s criterion holds exactly true. We also men-
tion the recent work [12] of Zelati and Zillinger, in which the authors consider the
linear stability of vorticity profiles with singularities of power-law type. The work in
[12], however, only considers the stability at sufficiently high frequency modes.
We now remark the relationship our work has with the stability theorems proven
by Arnold in [3] and the related questions proposed by Yudovich in [11]. Let ψ be
the stream function of a stationary solution of the incompressible Euler equations.
Arnold proves in [3] that the stationary flow is stable if the velocity profile is convex,
or ∇ψ/∇∆ψ > 0. When the velocity profile is concave, or ∇ψ/∇∆ψ < 0, then
there are finitely many unstable eigenvalues of the corresponding linear problem.
The velocity profile corresponding to the power-law vortex Ω is concave, so we expect
finitely many unstable eigenvalues. Our work improves this to a statement that there

5
are no unstable eigenvalues for the linearization around Ω. Yudovich in [11] proposes
that understanding the stability (or instability) of ideal fluid flows is an important
problem in mathematical hydrodynamics, and we consider our work as progress in
that direction.
Besides the application to fluid dynamics, we consider an intriguing aspect of
our work to be the novel techniques we use in our analysis of the Euler equations
linearized around the singular power-law vortex. The singularity of the background
vortex leads to an unbounded operator Lss with unbounded coefficients, which, to
our knowledge, cannot be handled by any previously known method. For example,
Lss cannot be thought of as the compact perturbation of a skew-adjoint operator,
which would be case if the background vortex were smooth.
We thank our advisor, Professor Camillo De Lellis, for his constant encouragement
and support during our graduate studies. We would also like to thank him for sharing
his deep insight into this and many other problems in mathematics. We acknowledge
the support of the National Science Foundation in the form of an NSF Graduate
Research Fellowship.

2 Exponential Self-Similar Coordinates

We now more precisely discuss the change of coordinates used by Vishik in [7] and
[8] as well as by the authors of [2]. Given a solution ω(x, t) of Equation (1) on
R2 × [T0 , T1 ], we introduce a function Ω on R2 × [log T0 , log T1 ] given by the following
transformation. We set τ = log t, ξ = xt−1/α , and let
Ω(ξ, τ ) = eτ ω(eτ /α ξ, eτ ).
The reverse transformation is given by:
ω(x, t) = t−1 Ω(t−1/α x, log t).
If the vector field v(x, t) is given by (K2 ∗ ω)(x, t) and the vector field V (ξ, τ ) is given
by (K2 ∗ Ω)(ξ, τ ), then we have the following transformation rules:
V (ξ, τ ) = eτ (1−1/α) v(eτ /α ξ, eτ ),
v(x, t) = t−1+1/α V (t−1/α x, log t).
Vishik in [7] and [8] as well as the authors of [2] show that if ω(x, t) satisfies Equation
(1), then the function Ω(ξ, τ ) satisfies
 
ξ
∂τ Ω − 1 + · ∇ Ω + (V · ∇)Ω = 0
α (3)
K2 ∗ Ω(·, τ ) = V (·, τ )

6
We usually refer to this system of coordinates as the exponential self-similar coordi-
nates or simply the self-similar coordinates.
Now we introduce exponential self-similar polar coordinates. In particular, if
(r, θ, t) are the usual polar coordinates on R2 ×[T1 , T2 ], we let ρ = rt−1/α , θ unchanged,
and τ = log t to get a new (ρ, θ, τ ) system of coordinates. We consider the velocity
form of the Euler Equations:

∂t v + (v · ∇)v = −∇p
(4)
div v = 0

which we write in terms of exponential self-similar polar coordinates as:

1 ρ Vθ V2
− 1 Vρ − ∂ρ Vρ + Vρ ∂ρ Vρ + ∂θ Vρ − θ = −∂ρ P


∂τ Vρ +
α α ρ ρ
1
ρ Vθ Vρ Vθ 1 (5)
∂τ Vθ + − 1 Vθ − ∂ρ Vθ + Vρ ∂ρ Vθ + ∂θ Vθ + = − ∂θ P
α α ρ ρ ρ
∂ρ (ρVρ ) + ∂θ Vθ = 0

One can see that, up to a constant prefactor, the only stationary radial vortex
V = Vθ (ρ)eθ satisfying the exponential self-similar equations is precisely the power-
law vortex with Vθ (ρ) = βρ1−α , where β is any real number.

Proposition 2.1. The unique solution of Equation (5) of the form V = Vθ (ρ)eθ is
given by the profile Vθ (ρ) = βρ1−α , where β is any real number.

Proof. The divergence-free condition is clearly satisfied. The second listed equation
in Equation (5) simplifies to (1/α−1)Vθ (ρ)−(p/α)Vθ′ (ρ) = 0, a first-order differential
equation whose unique solution is Vθ (ρ) = βρ1−α . For an appropriate choice of the
pressure P , the equation listed first in Equation (5) will also be satisfied.
We are thus led to define the velocity profile:

V (ρ) = βρ1−α eθ (6)

and its associated vorticity profile:

Ω(ρ) = β(2 − α)ρ−α . (7)

Note that Ω corresponds exactly by the transformation back to physical coordi-

nates to the stationary radial vortex with vorticity profile β(2 − α)r −α .

7
3 Stability of the Power-Law Vortex
The goal of this section is to prove Theorem 1.3 (which consequently will also prove
Theorem 1.4). Theorem 1.3 is the final result of a technical analysis of the lin-
earization of the Euler equations around a singular power-law vortex. We write the
linearization as
(∂τ − Lss )Ω = 0.
The main technical issue, for which no standard method can be used, is the highly
singular behavior of the coefficients of Lss . For example, we cannot simply say that
Lss is the compact perturbation of a skew-adjoint operator, which, on the contrary,
would be the case if the background vortex profile Ω were smooth.
We now introduce some useful notation. We denote Lqm to be the closed linear
subspace of Lq of elements that are m-fold symmetric. In other words, if Rθ : R2 →
R2 is the counterclockwise rotation of angle θ around the origin, then a function
f ∈ Lqm satisfies
f = f ◦ R2π/m .
It will be convenient to define the following closed linear subspaces of Lqm :

Uk := {f (r)eikθ : f ∈ Lq (R+ , rdr)}

so we have the direct sum: M

Lqm = Ukm .
k∈Z

We denote the Schwartz space by S and its dual by S ∗ . We begin by stating a

slight improvement of a lemma from [2]. The improvement allows for a continuous
extension of K2 ∗ to Lqm whenever 2 ≤ q < ∞.

Lemma 3.1. For every m ≥ 2 there exists a unique continuous operator T : Lqm →
S ∗ satisfying:

1. If ϕ ∈ S, then T ϕ = K2 ∗ ϕ in the sense of distribution.

2. There exists a constant C > 0 such that for every ϕ ∈ Lqm there exists v(ϕ) :=
1,q
v ∈ Wloc such that

• R−1 kvkLq (BR ) + kDvkLq (BR ) ≤ CkϕkLq (R2 ) for all R > 0
• div v = 0 and T (ϕ) = v(ϕ) in the sense of distribution.

8
 We shall denote T interchangeably with ∇⊥ ∆−1 or K2 ∗, although we shall mostly
use the latter notation. Recall that we wrote the linearization of (3) around Ω as
(∂τ − Lss )Ω = 0
where the linear operator Lss is given by:
ξ
− V · ∇Ω − ∇⊥ ∆−1 Ω · ∇Ω.


Lss Ω = Ω +
α
The domain of Lss as on operator into Lqm is denoted
Dm (Lss ) = {Ω ∈ Lqm : Lss (Ω) ∈ Lqm }.
Thus, Lss is a closed, densely defined operator. For a general operator L, we define
the resolvent set of L on Lqm to be the open set of z ∈ C for which L − z has a
bounded inverse from Lqm → Lqm . We define the spectrum of L, which we denote by
specm,q (L), to be the closed set which is the complement of the resolvent set.
We take the opportunity to restate the theorems that we shall prove. Our first
theorem states that the real part of any element of the spectrum of Lss is nonpositive:
a linear stability result.
Theorem 3.2. Let β ∈ R and α ∈ (0, 1). Let q be such that 2 ≤ q ≤ 2/α and
let m ≥ 2 be an integer. If Ω(ξ) = β(2 − α)|ξ|−α is a radial power-law vortex that
solves the Euler equations in self-similar coordinates, then for any λ ∈ specm,q (Lss ),
2
we have Re(λ) ≤ a0 := 1 − αq ≤ 0.
The value of the parameter m will not change the proof or the statement of the
theorem in any way. Functionally, in the proof that follows, one may change any
appearance of “2k” with “mk” and get an identical proof for the general case. For
this reason, and to simplify our presentation, we may assume that m = 2.
Recall the Hille-Yosida Theorem, stated, for example, as Corollary 3.6 of Chapter
2 in [5]:
Theorem 3.3 (Hille-Yosida). Let w ∈ R. Consider the linear operator A : D(A) ⊂
X → X on the Banach space X. Then A generates a strongly continuous semigroup
T (t) satisfying
kT (t)k ≤ Cewt ∀t ≥ 0
if and only if A is a closed operator, D(A) is dense in X, and for any λ ∈ C with
Re λ > w, λ is in the resolvent of A and for a constant C > 0 independent of λ:
C
k(A − λI)−1 k ≤ .
Re λ − w

9
 We would like to use the Hille-Yosida theorem to also conclude our Theorem 1.4,
which we restate:

Theorem 3.4. Let β ∈ R and α ∈ (0, 1). Let q be such that 2 ≤ q ≤ 2/α and
let m ≥ 2 be an integer. The operator Lss is the generator of a strongly continuous
semigroup on Lqm , which we denote eτ Lss , and the growth bound G(Lss ) of eτ Lss is
less than or equal to a0 . As a consequence, for any ǫ there exists a number M(α, ǫ)
depending only on α, ǫ such that

keτ Lss ΩkLq ≤ M(α, ǫ)e(a0 +ǫ)τ kΩkLq ∀τ, ∀Ω ∈ Lqm .

By the Hille-Yosida theorem, Theorem 3.4 will follow as a consequence of Theorem

3.2 and the following proposition:

Proposition 3.5. Let β ∈ R and α ∈ (0, 1). For any m ≥ 2, for any 2 ≤ q ≤ 2/α,
2
and for any λ ∈ C with Re λ > 1 − qα ,

Kα
k(Lss − λI)−1 k ≤ 2
Re λ − (1 − qα
)

for a constant Kα > 0 depending only on α.

We now proceed to prove these claims.

3.1 Proof of Lemma 3.1

Let f ∈ Lqm ∩ S and let v = K2 ∗ f . We first claim that
Z
v = 0 for every R > 0. (8)
BR

Indeed, we have v = ∇⊥ h, where h is the unique classical solution of ∆h = f given by

1
K ∗ f where K(x) = 2π log |x|. Since the kernel K is invariant under all rotations Rθ
and since f is m-fold symmetric, we conclude that h is m-fold symmetric. Therefore,

R−2π/m ∇h(R2π/m x) = ∇h(x).

Thus, integrating in x we conclude

Z Z
∇h = R2π/m ∇h,
BR BR

10
so
m−1
1 X
Z Z
∇h = R2kπ/m ∇h.
BR m k=0 BR
R
However, since m ≥ 2, the sum is zero, which shows that BR ∇h = 0. Finally, with
the property just shown, we may use the Poincaré inequality to conclude:

R−1 kvkLq (BR ) + kDvkLq (BR ) ≤ Ckf kLq (BR )

since kDvkLq ≤ kf kLq by the Calderón-Zygmund theorem.

3.2 Fourier Expansion of the Linearized Equations

In what follows, we shall abuse notation and denote the (exponential self-similar)
radial coordinate by r instead of ρ. As previously mentioned, we may without loss
of generality examine only the case when the parameter m = 2.
We recall that the inverse Laplacian ∆−1 preserves rotational symmetries. Thus,
for any Ω ∈ Lq2 ∩ S, ∆−1 Ω ∈ Lq2 ∩ C ∞ . Now we decompose
X
∆−1 Ω = fk (r)e2ikθ (9)
k∈Z

where fk (r) ∈ Lq (R+ , rdr) for all k ∈ Z. Let λ = λ1 + iλ2 be our putative element
2
of the spectrum, where λ1 , λ2 ∈ R and λ1 > 1 − qα . We consider the eigenvalue
equation Lss Ω − λΩ = 0, or equivalently
ξ
− V · ∇Ω − ∇⊥ ∆−1 Ω · ∇Ω = 0.


(1 − (λ1 + iλ2 ))Ω +
α
Using the decomposition (9), the above linear partial differential equation becomes
equivalent to an infinite family of ordinary differential equations. The ordinary dif-
ferential equation corresponding to the parameter k is:

∂r ∂θ2
(1 − (λ1 + iλ2 ))(∂r2 + + 2 )(fk (r)e2ikθ )+
r r
r ∂r ∂ 2 ∂θ ∂r ∂ 2
er −βr 1−α eθ ·(∂r (∂r2 + + θ2 )(fk (r)e2ikθ ) er + (∂r2 + + θ2 )(fk (r)e2ikθ )eθ +




+
α r r r r r
∂θ
(fk (r)e2ikθ )er − ∂r (fk (r)e2ikθ )eθ · − α(2 − α)βr −1−α er = 0.



+
r

11
Simplifying, we get
r ′′ f ′ (r) 4k 2 fk (r)
′ fk′ (r) 4k 2 fk (r)

fk (r) + k − + (1 − (λ 1 + iλ 2 )) fk
′′
(r) + −
α r r2 r r2
−α ′′ fk′ (r) 4k 2 fk (r)

−2ikβr fk (r) + − 2
− 2ikr −2−α α(2 − α)βfk (r) = 0 (10)
r r

3.3 Characterization of the Spectrum

2
We want to show that the intersection of the spectrum of Lss with {z : Re z > 1− qα }
2
is in fact empty. In particular, for any Re λ > 1 − qα , λ is in the resolvent of Lss .
Our proof will also directly show the bound required on the resolvent to apply the
Hille-Yosida theorem, so this section will finish the proof of both Theorem 3.2 and
Proposition 3.5.
Proving that Lss −λI is surjective with bounded resolvent is equivalent to proving
that the operator Rλ (g) := (Lss − λI)−1 (g) exists and is bounded as an operator into
D2 (Lss ). We thus assume that g is an arbitrary function in Lq2 . We shall now reduce
our problem to showing that the unique solution of an ordinary differential equation
boundary value problem is in some Lebesgue space with quantified Lq bound.
In particular, that Lss − λI is surjective is equivalent to stating that the inhomo-
geneous ordinary differential equation
r ′′ f ′ (r) 4k 2 fk (r)
′ fk′ (r) 4k 2 fk (r)

fk (r) + k − + (1 − (λ 1 + iλ 2 )) fk
′′
(r) + −
α r r2 r r2
f ′ (r) 4k 2 fk (r)

−2ikβr −α fk′′ (r) + k − − 2ikr −2−α α(2 − α)βfk (r) = gk (r) (11)
r r2
f ′ (r) 2
has a solution fk (r) with fk′′ (r) + kr − 4k rf2k (r) ∈ Lq (R, rdr) for any given function
gk (r) ∈ Lq (R, rdr). That Lss − λI is injective is equivalent to stating that the
homogeneous ordinary differential equation Equation (10) has no non-trivial solution
or, equivalently, that Lss − λI = g has a unique solution for every choice of the
function g.
Henceforth, we drop the subscript notation showing dependence on k and simply
write f (r) and g(r). Moreover, we define:
f ′ (r) 4k 2 f (r)
u(r) := f ′′ (r) + − ,
r r2
which is the quantity we wish to control in Lq (rdr). We now perform the change of
variables et = r and define:
ψ(t) := f (et )e(2/q−2)t G(t) := g(et )e2t/q U(t) := u(et )e2t/q .

12
These functions are chosen so that

g(r) ∈ Lq (rdr) ⇐⇒ G(t) ∈ Lq (dt)

u(r) ∈ Lq (rdr) ⇐⇒ U(t) ∈ Lq (dt)

f (r)/r 2 ∈ Lq (rdr) ⇐⇒ ψ(t) ∈ Lq (dt).
Given G(t) ∈ Lq (dt), our new goal is to solve for U(t) ∈ Lq (dt) (or alternatively
ψ(t) ∈ W 2,2 (dt)). The ordinary differential equations in terms of the new variable
and functions are
   
′′ 4 ′ 2 4 8
U(t) = ψ (t) + 4 − ψ (t) + 4 − 4k + 2 − ψ(t)
q q q
and
1 ′ 2
U (t) + (1 − − λ1 − iλ2 )U(t) − 2ikβe−αt U(t) − 2ikα(2 − α)βe−αt ψ(t) = G(t).
α αq
In the case when k = 0, we can integrate the equation of first order and get:
Z t
t(2/q+α(λ−1)) t(2/q+α(λ−1))
U(t) = ce + αe e−s(2/q+α(λ−1)) G(s)ds.
0

The unique choice of constant that will make U integrable is precisely

Z ∞
c = −α e−s(1+α(λ−1)) G(s)ds,
0

with which we have

Z
U(t) = −α χ(−∞,0) (t − s)e(t−s)(2/q+α(λ−1)) G(s)ds.
R

By Young’s convolution inequality, it follows that U(t) ∈ Lq (dt) is a solution of the

ordinary differential equation with
αkGkq
kUkq ≤ .
2/q + α(λ − 1)
This is the unique solution, since in the case when k = 0, the homogeneous problem
reduces to :
1 ′ 2
U (t) + (1 − − λ1 − iλ2 )U(t) = 0,
α αq

13
whose only solution is given by U(t) = c1 et(2/q+α(λ−1)) , which is not in any Lq space,
unless identically zero.
We may henceforth without loss of generality assume that k ≥ 1. Now, integrat-
ing the differential equation of first order yields

U(t) = c1 exp − 2ikβe−αt + t(2/q + α(λ − 1)) +

t
Z
−αt
exp 2ikβe−αs −s(2/q+α(λ−1)) G(s)ds+


+α exp −2ikβe +t(2/q+α(λ−1))
0
2
+2ikβα (2 − α) exp − 2ikβe−αt + t(2/q + α(λ − 1)) ×


Z t
exp 2ikβe−αs − s(2/q + αλ) ψ(s)ds.


× (12)
0
The constant c1 depends on the choice of initial condition. Also, integrating the
differential equation of second order gets us:
e−(2k+2−2/q)t t (2k+2−2/q)s
Z
−(2k+2−2/q)t
ψ(t) = c2 e − e U(s)ds+
4k 0

e(2k−2+2/q)t t −(2k−2+2/q)s
Z
(2k−2+2/q)t
+c3 e + e U(s)ds. (13)
4k 0
Finally to have ψ(t) ∈ Lq (dt) we use Equation (13), with the unique choice of
constants: Z 0
1
c2 = − e(2k+2−2/q)s U(s)ds
4k −∞
and ∞
1
Z
c3 = − e−(2k−2+2/q)s U(s)ds.
4k 0
These bounded linear functionals of U are uniquely chosen so that ψ ∈ Lq (dt) if
U ∈ Lq (dt). In addition, for smooth functions u, f , Equation (13) implies that
∆−1 u = f in the classical sense, while the unique choice of c2 , c3 above guarantee the
desired integrability of f . We may thus write ψ in terms of the convolution operator:
Z  
1 −(2k+2−2/q)(t−s) (2k−2+2/q)(t−s)
ψ(t) = − e χ(0,∞) (t − s) + e χ(−∞,0) (t − s) U(s)ds
4k R
We can denote the kernel above by
 
−(2k+2−2/q)(t−s) (2k−2+2/q)(t−s)
K1 (t, s) := e χ(0,∞) (t − s) + e χ(−∞,0) (t − s) .

14
We remark that given a function g, there is also at most one unique choice of the
parameter c1 such that the function U(t) lies in Lq (dt) and satisfies Equation (12).
In fact, the bounded functional c1 (U) will be given by
Z ∞
exp 2ikβe−αs − s(2/q + α(λ − 1)) G(s)ds−


c1 = −α
0
Z ∞
2
exp 2ikβe−αs − s(2/q + αλ) ψ(s)ds


2ikβα (2 − α)
0
and Equation (12) becomes
Z
U(t) = −α exp −2ikβe−αt +2ikβe−αs +(t−s)(2/q+α(λ−1)) χ(−∞,0) (t−s)G(s)ds−


R

2ikβα2 (2 − α)×
Z
e−αs exp − 2ikβe−αt + 2ikβe−αs + (t − s)(2/q + α(λ − 1)) χ(−∞,0) (t − s)ψ(s)ds.


×
R
We denote the kernel:

K2 (t, s) := exp − 2ikβe−αt + 2ikβe−αs + (t − s)(2/q + α(λ − 1)) χ(−∞,0) (t − s).

We thus have the following Fredholm integral equation for U:

iα2 (2 − α)β
Z Z Z
U(t) = −α K2 (t, s)G(s)ds + K2 (t, s)e−αs K1 (s, r)U(r)drds.
R 2 R R

However, we can integrate the kernels in the s variable first:

Z
K2 (t, s)e−αs K1 (s, r)ds =
R
Z
exp 2ikβe−αs − αs − (2/q + α(λ − 1) + 2k + 2 − 2/q)s χ(max(t,r),∞) (s)ds+


= F1 (t, r)
R
Z
exp 2ikβe−αs − αs − (2/q + α(λ − 1) − (2k − 2 + 2/q))s χ(t,r) (s)ds.


+F2 (t, r)
R
Where

F1 (t, r) = exp − 2ikβe−αt + (2/q + α(λ − 1))t + r(2k + 2 − 2/q)

and
F2 (t, r) = exp − 2ikβe−αt + (2/q + α(λ − 1))t − r(2k − 2 + 2/q) .

15
 We remark that  
d i 2ikβe−αs −αs
e = e2ikβe −αs .
ds 2αkβ
Thus integrating by parts yields (for some generic m > 0):
Z ∞ Z ∞
2ikβe−αs −αs −ms i 2ikβe−αt −mt i −αs
e e ds = − e e − e2ikβe e−ms ds =
t 2αkβ 2αkβ t
  Z ∞
−i 1 2ikβe−αt −mt 1 −αs
= 1+ e e − e2ikβe −αas e−ms ds.
2αkβ m m t
Simplifying we get (for m > 0):
Z ∞
−αs 1 −αt
e2ikβe −αs e−ms ds = e2ikβe e−mt . (14)
t 2ikβα
Likewise, integrating by parts yields (for some generic m ∈ R not necessarily posi-
tive): Z r  
2ikβe−αs −αs −ms 1 2ikβe−αt −mt 2ikβe−αr −mr
e e ds = e e −e e . (15)
t 2ikβα
Applying Equation (14) and Equation (15) to the original integral under considera-
2
tion (and using that λ > 1 − qα ) gets us
Z
K2 (t, s)e−αs K1 (s, r)ds =
R
1 −αt
= F1 (t, r)χ(0,∞) (t − r)e2ikβe e−(2/q+α(λ−1)+2k+2−2/q)t +
2ikβα
1 −αr
+ F1 (t, r)χ(−∞,0) (t − r)e2ikβe e−(2/q+α(λ−1)+2k+2−2/q)r +
2ikβα

1 −αt
+ F2 (t, r)χ(−∞,0) (t − r) e2ikβe e−(2/q+α(λ−1)−(2k−2+2/q))t −
2ikβα

2ikβe−αr −(2/q+α(λ−1)−(2k−2+2/q))r
e e .

Simplifying further (there is a remarkable cancellation that occurs) we achieve:

Z
2ikβα K2 (t, s)e−αs K1 (s, r)ds =
R

e−(2k+2−2/q)(t−r)
χ(0,∞) (t − r) + χ(−∞,0) (t − r)e(2k−2+2/q)(t−r) = K1 (t, r).
We conclude that
α(2 − α)
Z Z
U(t) = −α K2 (t, s)G(s)ds + K1 (t, r)U(r)dr.
R 4k R

16
3.3.1 The Integral Transforms
We have a convolution operator:
Z
Φ1 (U)(t) := K1 (t, s)U(s)ds.
R

Now, by Young’s convolution inequality, we have

Z ∞ Z 0 
−(2k+2−2/q)y (2k−2+2/q)y 2
kΦ1 (U)(t)kq ≤ kUkq e dy+ e dy ≤ kUkq .
0 −∞ (2k − 2 + 2/q)
In a similar manner, we denote
Z
Φ2 (U) := K2 (t, s)U(s)ds
R

and observe that Z

|Φ2 (U)| ≤ K̂2 (t, s)U(s)ds
R
where
K̂2 (t, s) := e(t−s)(2/q+α(λ−1)) χ(−∞,0) (t − s)
is the kernel of a convolution operator. Thus, arguing as before, we apply the Young
convolution inequality and get
Z ∞ 
−(2/q+α(λ−1))y 1
kΦ2 (U)(t)kq ≤ kUkq e dy ≤ kUkq .
0 2/q + α(λ − 1)

3.3.2 Conclusion
We recall what we have heretofore shown. We found two bounded integral transforms
from Lq → Lq , given by Φ1 and Φ2 . Showing that the operator we are working on is
surjective then becomes equivalent to showing (for every G ∈ Lq ) the existence of a
fixed point of the map
α(2 − α)
U → −αΦ2 (G) + Φ1 (U).
4k
This, by the Banach fixed point theorem, is equivalent to showing that the map
α(2−α)
4k
Φ1 is contractive. However, by our work in the previous subsection, we know
that this is true, since:
α(2 − α) 2α(2 − α) α(2 − α)
k Φ1 (U)kLq ≤ ≤ <1
4k 4k(2k − 2 + 2/q) 4/q

17
since qα ≤ 2. Thus the unique solubility of the ordinary differential equation, hence
the surjectivity and injectivity of the map, is proven.
Moreover, by the bound we found on Φ2 , the solution U certainly satisfies

Mα kGkLq
kUkLq ≤ 2
Re λ − (1 − qα )
2
for all λ with Re λ > 1 − qα , for some choice of constant Mα large enough (depending
only on α). This shows that the solution map satisfies the hypotheses of the Hille-
Yosida theorem, which finishes our work.

4 Appendix
4.1 Explicit Solutions of Equation (10)
We present an alternative proof that the point spectrum of Lss is empty. For sim-
plicity of notation, we assume β = 1, but the proof is entirely the same in the general
case. We recall Equation (10) :

r ′′ f ′ (r) 4k 2 fk (r)
′ fk′ (r) 4k 2 fk (r)

fk (r) + k − + (1 − (λ 1 + iλ 2 )) fk
′′
(r) + −
α r r2 r r2
fk′ (r) 4k 2 fk (r)

−2ikr −α fk′′ (r) + − − 2ikr −2−α α(2 − α)fk (r) = 0.
r r2
We consider the above homogeneous ordinary differential equation and perform a
change of variables z = −2ikr −α , since this will put the ordinary differential equation
in a form that is already well studied. More precisely, we define a new function wk (z)
that is given by

wk (z) = fk ((−2ik)1/α z −1/α ) · (−2ik)−2k/α z 2k/α .

The function is defined so that if fk (r) = w(−2ikr −α )r 2k satisfies the homogeneous

ordinary differential equation, we get that w satisfies the following ordinary differ-
ential equation:


0 = 2kr −3α+2k−2
i(α − 4k)r α (r α (αλ − 2k + 2) + 2iαk)w ′ (−2ikr −α )+

18
 +2αk((r α (α(λ + 2) − 6k + 2) + 2iαk)w ′′(−2ikr −α ) − 2iαkw (3) (−2ikr −α ))+

2α −α
+i(α − 2)αr w(−2ikr )

Letting r = (−2ik)1/α z −1/α and simplifying we finally have:

2 − 6k ′′
0 = z 2 w (3) (z) + z(1 − z + λ + 1 + )w (z)+
α
2 − 2k −1
+((λ + )α (α − 4k) − z(−1 + α−1(α − 4k) + 1))w ′(z) + α−1 (α − 2)w(z). (16)
α
We put the equation in this form because Equation 07.25.13.0004.01 from [9]
states that the general solution of the ordinary differential equation:

z 2 w (3) (z) + z(1 − z + b1 + b2 )w ′′(z) + (b1 b2 − z(a1 + a2 + 1))w ′(z) − a1 a2 w(z) = 0

is given exactly by
w(z) = c1 · 2 F̃2 (a1 , a2 ; b1 , b2 ; z)+
 
G2,2 1 − a1 , 1 − a2; 0, 1 − b1 , 1 − b2 + G2,2



+c2 2,3 z 2,3 z 1 − a1 , 1 − a2 ; 0, 1 − b2, 1 − b1 +

+c3 G3,2


2,3 − z 1 − a1 , 1 − a2 ; 0, 1 − b1 , 1 − b2 .
Here 2 F̃2 denotes the regularized hypergeometric function and Gp,q m,n denotes the
Meijer-G function. The definition of both of these classes of special functions can be
found in [6] or [9].
We choose
√ √
−2k − α2 − 2α + 4k 2 −2k + α2 − 2α + 4k 2
a1 = , a2 =
α α
α − 4k 2 − 2k + αλ
b1 =
, b2 =
α α
and denote √
α2 − 2α + 4k 2
qα,k := .
α
Then the solution of Equation (16) is exactly (for any choice of constants c1 , c2 , c3 ):

2k 2k 4k 2 − 2k
w(z) = c1 · 2 F̃2 (− − qα,k , − + qα,k ; 1 − , λ + ; z)+
α α α α

19
 
2k 2k 4k 2k − 2

+c2 G2,2 2,3 z 1 + + qα,k , 1 + − qα,k ; 0, , 1 − λ + +
α α α α

2,2 2k 2k 2k − 2 4k

+G2,3 z 1 + + qα,k , 1 + − qα,k ; 0, 1 − λ + , +
α α α α
3,2 2k 2k 4k 2k − 2

+c3 G2,3 −z 1+ + qα,k , 1 + − qα,k ; 0, , 1 − λ + .
α α α α
Using the transformation back to fk (r), we see that we have found the solutions of the
homogeneous differential equation (10). Then, by using the asymptotic expansions
for the special functions 2 F̃2 and Gp,qm,n described in Chapter 5 of [6], one can see that
the solution cannot lie in any Lebesgue space unless it is identically zero. We leave
the details to the interested reader.

References
[1] Albritton, D., Brue, E., Colombo, M. Non-Uniqueness of Leray Solutions of the
Forced Navier-Stokes Equations. Annals of Math. Vol. 196. pp. 415-455. (2022).

[2] Albritton, D., Brue, E., Colombo, M., De Lellis, C., Giri, V., Janisch, M., Kwon,
H. Instability and Non-uniqueness for the 2D Euler Equations, after M. Vishik.
Annals of Mathematics Studies. 219. Princeton, N.J., (2023).

[3] Arnold, V.I. Conditions for Non-Linear Stability of Stationary Plane Curvilinear
Flows of an Ideal Fluid. Sov. Math. Dokl. 162, No. 5. pp. 773-777. (1965).

[4] Chandrasekhar, S. Hydrodynamic and Hydromagnetic Stability. Oxford Press.

(1961).

[5] Engel, K-J., Nagel, R. One-Parameter Semigroups. for Linear Evolution Equa-
tions. Graduate Texts in Mathematics. Springer-Verlag. (2000).

[6] Luke, Y.L. Mathematical Functions and their Approximations. Academic Press
Inc. (1975).

[7] Vishik, M. Instability and non-uniqueness in the Cauchy problem for the Euler
equations of an ideal incompressible fluid. Part I. arXiv: 1805.09426 (2018).

[8] Vishik, M. Instability and non-uniqueness in the Cauchy problem for the Euler
equations of an ideal incompressible fluid. Part II. arXiv: 1805.09440 (2018).

20
 [9] Wolfram Research, Inc. https://functions.wolfram.com. Mathematical Func-
tions Site. (2024).

[10] Yudovich, V.I. Non-stationary flow of an ideal incompressible liquid. USSR Com-
putational Mathematics and Mathematical Physics. (1963).

[11] Yudovich, V.I. Eleven Great Problems of Mathematical Hydrodynamics. Moscow

Math. J. Vol 3. No. 2. pp. 711-737. (2003).

[12] Zelati, M.C., Zillinger, C. On degenerate circular and shear flows: the point
vortex and power law circular flows. Comm. PDE. Vol. 44. pp. 110-155. (2019).

21