FUNCTIONAL INEQUALITIES FOR BROWNIAN MOTION ON

MANIFOLDS WITH STICKY-REFLECTING BOUNDARY DIFFUSION

MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Abstract. We prove geometric upper bounds for the Poincaré and Logarithmic Sobolev
constants for Brownian motion on manifolds with sticky reflecting boundary diffusion i.e.
extended Wentzell-type boundary condition under general curvature assumptions on the
manifold and its boundary. The method is based on an interpolation involving energy
interactions between the boundary and the interior of the manifold. As side results
we obtain explicit geometric bounds on the first nontrivial Steklov eigenvalue, for the
arXiv:2401.00206v2 [math.PR] 3 Apr 2024

norm of the boundary trace operator on Sobolev functions, and on the boundary trace
logarithmic Sobolev constant. The case of Brownian motion with pure sticky reflection
is also treated.

1. Introduction and description of main results

Let Ω be a smooth compact connected Riemannian manifold of dimension d ≥ 2 with
smooth connected boundary ∂Ω. We consider the semigroup on C(Ω) induced from the
Feller generator (D(A), A) given by
D(A) = {f ∈ C(Ω) ∣ Af ∈ C(Ω)}
∂f
Af = ∆f 1Ω + (β∆τ f − γ ) 1∂Ω ,
∂N
∂f
where ∂N is the outer normal derivative, ∆τ is the Laplace-Beltrami operator on ∂Ω,
β ≥ 0 and γ > 0. The induced Markov process is a diffusion on Ω which performs Brownian
motion in the interior and sticky reflected Brownian motion along the boundary. The case
of pure sticky reflection but no diffusion along the boundary corresponds to the regime
β = 0. First rigorous constructions of such processes on special domains Ω date back to
[26, 46, 51] and were later extended to jump-diffusion processes on general domains cf.
[7], [45]. Renewed interest for models with β > 0 emerges from applications in interacting
particle systems with singular boundary or zero-range pair interaction [1, 16, 23, 31, 40].
An efficient construction in the diffusion case via Dirichlet forms was given recently in
[21, 25], the associated JKO-Wasserstein gradient flow structure is investigated in [10].
In this work we come back to the problem raised in [30] of estimating the speed of
convergence of this family of processes to their equilibrium state. Any such equilibrium
is a composition of the two mutually singular uniform measures on the interior and the
boundary and thus has infinite negative curvature even if Ω and ∂Ω are positively curved.
As a consequence, standard arguments for e.g. log-concave measures are not applicable.
However, as shown in [30] one can use a simple interpolation together with the Reilly
formula to obtain explicit estimates for the L2 -spectral gap in the positive curvature case.
We aim to extend the interpolation approach in two ways. First we go beyond the strictly
positive curvature assumptions on Ω and ∂Ω by circumventing the Reilly formula via an
Date: April 4, 2024.
SUPPORTED IN PART BY THE NATIONAL KEY R&D PROGRAM OF CHINA (NO.
2022YFA1006000, 2020YFA0712900) AND NNSFC (11921001).
1
2 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

integration by parts against a carefully chosen test function to obtain explicit estimates
for the crucial bulk-boundary interaction terms also under negative lower curvature bound
assumptions. The second main extension is concerned with an analogous approach to the
derivation of an explicit logarithmic Sobolev inequality. We develop the interpolation for
this case and show how to obtain the needed boundary-bulk interaction estimates, again
via integration by parts against properly chosen test functions. As side results we obtain
new explicit geometric estimates of independent interest for the first Steklov eigenvalue
(Corollary 2.1), the boundary-trace operator for Sobolev functions (Proposition 6.2) and
the boundary trace logarithmic Sobolev constant (Proposition 6.5), which in special cases
was considered before in e.g. [4, 39]. As we go along we also discuss how to treat the case
β = 0 of purely sticky reflection without boundary diffusion.

2. Poincaré Inequality
For simplicity throughout this paper we restrict the analysis to the case β = 1 (and
β = 0 for the case of pure sticky reflection) and γ > 0, where we assume that Ω and ∂Ω
have finite (Hausdorff) measure ∣Ω∣ < ∞ and ∣∂Ω∣ < ∞. Furthermore by λΩ resp. λ∂Ω we
denote the normalised volume measure on Ω resp. normalised Hausdorff measure on ∂Ω
and choose α ∈ (0, 1), such that
α ∣∂Ω∣
= γ.
1 − α ∣Ω∣
Moreover we set
λα ∶= αλΩ + (1 − α)λ∂Ω .
and find that −A is λα -symmetric.

Our aim is to estimate the Poincaré – and further down below – the logarithmic Sobolev
constants for such processes in terms of the geometry of Ω and its boundary ∂Ω.

We say that a Poincaré Inequality is fulfilled if there is a constant Cα such that ∀f ∈ C 1 (Ω)
V arλα (f ) ≤ Cα Eα (f ),
where for f ∈ C 1 (Ω)
2
V arλα (f ) = ∫ f 2 dλα − (∫ f dλα )
Ω Ω

Eα (f ) = α ∫ ∣∇f ∣2 dλΩ + (1 − α) ∫ ∣∇τ f ∣2 dλ∂Ω

Ω ∂Ω
and ∇τ denotes the tangential derivative operator on ∂Ω. In the following we denote by
Cα the optimal such constant. By CΩ and C∂Ω we denote the usual (Neumann) Poincaré
constants of Ω and ∂Ω respectively. We assume that CΩ and C∂Ω (or respective upper
bounds for them) are known. In [30, Proposition 2.1] the following statement was proved
in the setting introduced above using an interpolation approach:
Proposition 2.1. Assume there exist constants K∂Ω,Ω , K1 , K2 such that for any f ∈ C 1 (Ω)
(1) V arλ∂Ω (f ) ≤ K∂Ω,Ω ∫ ∣∇f ∣2 dλΩ
Ω
and
2
(2) (∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ + K2 ∫ ∣∇τ f ∣2 dλ∂Ω ,
Ω ∂Ω Ω ∂Ω
 3

then it holds for any α ∈ (0, 1)

Cα ≤ max(CΩ + (1 − α)K1 , αK2 ,
(1 − α)K∂Ω,Ω C∂Ω + αCΩ C∂Ω + α(1 − α)(K∂Ω,Ω K2 + C∂Ω K1 )
).
(1 − α)K∂Ω,Ω + αC∂Ω
In [30, section 3.2] constants K∂Ω,Ω , K1 , K2 were found under the assumption of a posi-
tive lower bound on Ricci curvature and a positive lower bound on the second fundamental
form on the boundary ∂Ω (i.e. a convex boundary). Our aim is to find K∂Ω,Ω , K1 , K2 and
thus an upper bound on Cα assuming any upper bound on Sectional curvature and lower
bound on Ricci curvature and any upper and lower bound for the second fundamental
form on the boundary and to thereby generalise section 3.2 in [30].
The constants K1 and K2 in (2) are not unique. In fact we could add another optimisation
over the pairs of values K1 , K2 fulfilling (2) and thus improve the upper bound on Cα .
In order to find suitable constants K1 , K2 fulfilling (2) we start with the following
proposition where we do not yet make any assumptions on the geometry of the manifold.
Later it will be combined with assumptions on curvature and second fundamental form
to obtain explicit admissible choices for K1 , K2 .
∂φ
Proposition 2.2. For any φ ∈ C 1 (Ω) such that ∂N ∣∂Ω = ±1 and ∇φ is Lipschitz continuous
on Ω inequality (2) in Proposition 2.1 is fulfilled with K2 = 0 and
2
∣Ω∣
K1 = ( ) inf [(1 + ε)∣∇φ∣22 + (1 + ε−1 )CΩ ∣∆φ∣22 ] ,
∣∂Ω∣ ε∈(0,∞)

where ∣ ⋅ ∣2 denotes the L2 -norm on Ω with respect to λΩ .

Proof. Let f ∈ C 1 (Ω). Without loss of generality we can assume that ∫Ω f dλΩ = 0. Now
∀ε > 0
2
(∫ f dλΩ − ∫ f dλ∂Ω )
Ω ∂Ω
2 2
∂φ
= (∫ f dλ∂Ω ) = (∫ f dλ∂Ω )
∂Ω ∂Ω ∂N
2
∣Ω∣ ∣Ω∣
=( ∫ ∇f ⋅ ∇φdλΩ + f ∆φdλΩ )
∣∂Ω∣ Ω ∣∂Ω∣ ∫Ω
2 2 2
∣Ω∣
≤( ) [(1 + ε) (∫ ∇f ⋅ ∇φdλΩ ) + (1 + ε−1 ) (∫ f ∆φdλΩ ) ]
∣∂Ω∣ Ω Ω
2
∣Ω∣
≤( ) [(1 + ε) ∫ ∣∇f ∣2 dλΩ ∫ ∣∇φ∣2 dλΩ + (1 + ε−1 ) ∫ f 2 dλΩ ∫ (∆φ)2 dλΩ ]
∣∂Ω∣ Ω Ω Ω Ω
2
∣Ω∣
≤( ) [(1 + ε) ∫ ∣∇f ∣2 dλΩ ∫ ∣∇φ∣2 dλΩ + (1 + ε−1 )CΩ ∫ ∣∇f ∣2 dλΩ ∫ (∆φ)2 dλΩ ]
∣∂Ω∣ Ω Ω Ω Ω
2
∣Ω∣
=( ) [(1 + ε) ∫ ∣∇φ∣2 dλΩ + (1 + ε−1 )CΩ ∫ (∆φ)2 dλΩ ] ∫ ∣∇f ∣2 dλΩ .
∣∂Ω∣ Ω Ω Ω

□
We next find K∂Ω,Ω such that inequality (1) is fulfilled by proceeding similarly as in the
proof of Proposition 2.2:
4 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

∂ρ
Proposition 2.3. For any ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is Lipschitz continuous
on Ω inequality (1) in Proposition 2.1 is fulfilled with
∣Ω∣ 1/2
K∂Ω,Ω = (2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ ) ,
∣∂Ω∣
where (⋅)− denotes the negative part of a function and ∣ ⋅ ∣∞ denotes the L∞ -norm on Ω
with respect to λΩ .
Proof. Let f ∈ C 1 (Ω). Without loss of generality we can assume that ∫Ω f dλΩ = 0. Now
we can calculate similarly as in the previous result
2
∂ρ
V arλ∂Ω (f ) = ∫ f 2 dλ∂Ω − (∫ f dλ∂Ω ) ≤ ∫ f 2 dλ∂Ω = − ∫ f2 dλ∂Ω
∂Ω ∂Ω ∂Ω ∂Ω ∂N
∣Ω∣ ∣Ω∣
=− ∫ 2f ∇f ⋅ ∇ρdλΩ − ∫ f 2 ∆ρdλΩ
∣∂Ω∣ Ω ∣∂Ω∣ Ω
∣Ω∣ ∣Ω∣
≤2 ∫ ∣f ∣∣∇f ∣∣∇ρ∣dλΩ + f 2 (∆ρ)− dλΩ
∣∂Ω∣ Ω ∣∂Ω∣ ∫Ω
∣Ω∣ ∣Ω∣
≤2 ∣∇ρ∣∞ ∫ ∣f ∣∣∇f ∣dλΩ + ∣(∆ρ)− ∣∞ ∫ f 2 dλΩ
∣∂Ω∣ Ω ∣∂Ω∣ Ω
1/2
∣Ω∣ ∣Ω∣
≤2 ∣∇ρ∣∞ (∫ f 2 dλΩ ∫ ∣∇f ∣2 dλΩ ) + ∣(∆ρ)− ∣∞ ∫ f 2 dλΩ
∣∂Ω∣ Ω Ω ∣∂Ω∣ Ω

∣Ω∣ 1/2 ∣Ω∣

≤2 ∣∇ρ∣∞ CΩ ∫ ∣∇f ∣2 dλΩ + ∣(∆ρ)− ∣∞ CΩ ∫ ∣∇f ∣2 dλΩ
∣∂Ω∣ Ω ∣∂Ω∣ Ω

∣Ω∣ 1/2
= (2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ ) ∫ ∣∇f ∣2 dλΩ .
∣∂Ω∣ Ω

□
Remark 2.1. Denote by σ the first nontrivial eigenvalue of the Steklov eigenvalue problem
∆f = 0, in Ω
{ ∂f
∂N = σf, on ∂Ω,
which is characterised (using normalised measures) by
2
∣Ω∣ ∫Ω ∣∇f ∣ dλΩ
σ= inf 2
.
∣∂Ω∣ 1
f ∈C (Ω) ∫∂Ω f dλ∂Ω
∫∂Ω f dλ∂Ω =0

Thus we have for the optimal constant K∂Ω,Ω in inequality (1) in Proposition 2.1
2
V arλ∂Ω (f ) ∫∂Ω f dλ∂Ω ∣Ω∣ −1
K∂Ω,Ω = sup 2
= sup 2
= σ .
f ∈C 1 (Ω) ∫Ω ∣∇f ∣ dλΩ f ∈C 1 (Ω) ∫Ω ∣∇f ∣ dλΩ ∣∂Ω∣
∫∂Ω f dλ∂Ω =0

Therefore by finding upper bounds for the optimal K∂Ω,Ω we find lower bounds for the first
nontrivial Steklov eigenvalue. We use this connection in the computations for Example
2.1.2 below.
To obtain an explicit constant it now remains to specify functions φ and ρ with the
desired properties. Note that despite fulfilling the same assumptions, φ and ρ may be
chosen independently in order to optimise the estimates. It seems natural to define both
functions of the form ψ ○ ρ∂Ω for some appropriate function ψ, where ρ∂Ω denotes the
 5

distance to the boundary function.

We use for k, γ ∈ R the function
⎧ √ √
⎪cos( √
⎪ kt) − √γk sin( kt), k≥0
(3) h ∶ [0, ∞) → R, h(t) ∶= ⎨ γ
√
⎪
⎪ cosh( −kt) − √−k sinh( −kt), k < 0.
⎩
Let h−1 (0) ∶= inf{t ≥ 0 ∶ h(t) = 0}, where h−1 (0) = ∞ if h(t) > 0 for all t ≥ 0. We denote by
Ric and sect the Ricci and sectional curvatures of Ω, and by Π the second fundamental
form on the boundary ∂Ω, i.e.
Π(X, Y ) ∶= ⟨∇X N, Y ⟩, X, Y ∈ Tx ∂Ω, x ∈ ∂Ω,
where N is the outward pointing unit normal vector field of ∂Ω.
Lemma 2.1. Let k1 , k2 ∈ R such that Ric ≥ k1 (d − 1), sect ≤ k2 and γ1 , γ2 ∈ R such that
∂φ
γ1 id ≤ Π ≤ γ2 id. We construct a function φ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇φ is
Lipschitz continuous on Ω and for t0 ∈ (0, h2 (0))
−1

1 t0 t 2
∣∇φ∣22 ≤ Hd−1 ({ρ ∂Ω = t}) (1 − ) dt,
∣Ω∣ ∫0 t0
2
1 t0 h′2 t 1 −
∣∆φ∣22 ≤ Hd−1 ({ρ ∂Ω = t})((((d − 1) (t) (1 − ) − ) )
∣Ω∣ ∫0 h2 t0 t0
2
h′ t 1 +
+ (((d − 1) 1 (t) (1 − ) − ) ) )dt,
h1 t0 t0
where hi , i = 1, 2 are as defined above in (3) with k = ki and γ = γi .
Proof. It is easy to see that h−1
2 (0) ≤ h1 (0). Let ρ∂Ω be the distance function to the
−1

boundary. By the Laplacian comparison theorem, we have

(d − 1)h′1
(4) ∆ρ∂Ω ≤ (ρ∂Ω ) on {ρ∂Ω < h−1
1 (0)},
h1
(d − 1)h′2
(5) ∆ρ∂Ω ≥ (ρ∂Ω ) on {ρ∂Ω < h−1
2 (0)}.
h2
Indeed, (5) follows from [49, Theorem 3.1] for the Laplacian comparison theorem due to
[27], and by [49, Corollary 3.2] which says that the injectivity radius of ∂Ω is larger than
h−1
2 (0). Next, for x ∈ Ω with ρ∂Ω (x) < h1 (0), let p ∈ ∂Ω be the projection such that
−1

γ(s) ∶= exp[−sN (p)], s ∈ [0, ρ∂Ω (x)] be the minimal geodesic from p to x. Let {Xi }1≤i≤d−1
be orthonormal vector fields around x orthogonal to ∇ρ∂Ω (x). Let Ji (s))s∈[0,ρ∂Ω (x)] be the
Jacobi fields along the geodesic γ such that Ji (ρ∂Ω (x)) = Xi (x) and
⟨J˙i (0), v⟩ = −Π(Ji (0), v), v ∈ Tp ∂Ω.
Let R be the Riemannian curvature tensor. By the second variational formula (see page
321 in [13]) we have
ρ∂Ω (x)
Hessρ∂Ω (Xi , Xi )(x) = −Π(Ji (0), Ji (0)) + ∫ (∣J˙i (s)∣2 − ⟨R(γ̇(s), Ji (s))γ̇, Ji (s)⟩)ds.
0

Let (Xi (s))s∈[0,ρ∂Ω (x)] be the parallel displacement of (Xi (0) ∶= Xi (x))s∈[0,ρ∂Ω (x)] , and de-
note
h1 (s)
J˜i (s) = Xi (s), 1 ≤ i ≤ d − 1, s ∈ [0, ρ∂Ω (x)].
h1 (ρ∂Ω (x))
6 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Then the index lemma yields

ρ∂Ω (x)

Hessρ∂Ω (Xi , Xi )(x) ≤ −Π(J˜i (0), J˜i (0)) + ∫ (∣J˜˙i (s)∣2 − ⟨R(γ̇(s), J˜i (s))γ̇, J˜i (s)⟩)ds.
0

Noting that h′′1 (s) = −k1 h1 (s), this implies (4).

Now for t0 ∈ (0, h−12 (0)) we define
ρ∂Ω s +
φ=∫ (1 − ) ds.
0 t0
We have
⎧ ρ (x)
⎪∇ρ∂Ω (x) ⋅ (1 − ∂Ωt0 ) , ρ∂Ω (x) ≤ t0
⎪
∇φ(x) = ⎨
⎩0, else,
⎪
⎪
∂φ
and thus ∂N ∣∂Ω = −1 and ∇φ is Lipschitz continuous. Furthermore
⎧ ρ (x)
⎪∆ρ∂Ω (x) (1 − ∂Ωt0 ) − t10 , ρ∂Ω (x) ≤ t0 ,
⎪
∆φ(x) = ⎨
⎩0 else.
⎪
⎪
Using the Coarea formula we get
2 ρ∂Ω 2 ρ∂Ω 2
∫Ω ∣∇φ∣ dλΩ = ∫{ρ (1 − ) ∣∇ρ∂Ω ∣2 dλΩ = ∫ (1 − ) dλΩ
∂Ω ≤t0 } t0 {ρ∂Ω ≤t0 } t0
1 t0 t 2
= (1 − ) dHd−1 dt
∣Ω∣ ∫0 ∫{ρ∂Ω =t} t0
1 t0 t 2
= ∫ Hd−1 ({ρ∂Ω = t}) (1 − ) dt,
∣Ω∣ 0 t0
where Hd−1 denotes the (d − 1)-dimensional Hausdorff measure. Furthermore
2 2 2 2
∫Ω (∆φ) dλΩ = ∫Ω ((∆φ) − (∆φ) ) dλΩ = ∫Ω ((∆φ) ) dλΩ + ∫Ω ((∆φ) ) dλΩ .
+ − + −

We see that by inequalities (4) and (5) on {ρ∂Ω ≤ t0 }

h′2 ρ∂Ω 1 h′ ρ∂Ω 1 −
∆φ ≥ (d − 1) (ρ∂Ω ) (1 − ) − ⇒ (∆φ)− ≤ ((d − 1) 2 (ρ∂Ω ) (1 − )− ) ,
h2 t0 t0 h2 t0 t0
h′
ρ∂Ω 1 h′
ρ∂Ω 1 +
∆φ ≤ (d − 1) 1 (ρ∂Ω ) (1 − ) − ⇒ (∆φ)+ ≤ ((d − 1) 1 (ρ∂Ω ) (1 − )− ) .
h1 t0 t0 h1 t0 t0
Thus
2
2 h′2 ρ∂Ω 1 −
∫Ω (∆φ) dλ ≤
Ω ∫ (((d − 1) (ρ ∂Ω ) (1 − ) − ) )
{ρ∂Ω ≤t0 } h2 t0 t0
2
h′1 ρ∂Ω 1 +
+ (((d − 1) (ρ∂Ω ) (1 − ) − ) ) dλΩ
h1 t0 t0
2
1 t0 h′2 t 1 −
= ∫ Hd−1 ({ρ∂Ω = t})((((d − 1) (t) (1 − ) − ) )
∣Ω∣ 0 h2 t0 t0
2
h′ t 1 +
+ (((d − 1) 1 (t) (1 − ) − ) ) )dt.
h1 t0 t0
□
 7

2 2
In the previous Lemma t0 ∈ (0, h−1
2 (0)) may be chosen to either optimise ∣∇φ∣2 or ∣∆φ∣2 .

Lemma 2.2. Let k2 ∈ R such that sect ≤ k2 and γ2 ∈ R such that Π ≤ γ2 id. Then k2 > −γ22 ,
∂ρ
and for all ε > 0 there exists a function ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is Lipschitz
continuous on Ω and

∣∇ρ∣∞ ≤ 1,
h′2 t 1 −
∣(∆ρ)− ∣∞ ≤ inf−1 sup ((d − 1) (t) (1 − ) − ) + ε.
t1 ∈(0,h2 (0)) t∈(0,t1 ) h2 t1 t1

Proof. Let h2 be the function defined as in (3) with k = k2 and γ = γ2 . If k2 ≤ −γ22 , then
h−1
2 (0) = ∞, so that [49][Corollary 3.2] implies that the cut locus of ∂Ω is empty, which
is contradictive to the fact that the maximum point of ρ∂Ω is in the cut locus. Hence,
k2 > −γ22 .
Let t1 ∈ (0, h−1
2 (0)) to be chosen later. By (5)

h′2
(6) ∆ρ∂Ω ≥ (d − 1) (ρ∂Ω ) on {ρ∂Ω ≤ t1 }.
h2
Now define
ρ∂Ω s +
ρ=∫ (1 − ) ds.
0 t1
We have
⎧ ρ (x)
⎪∇ρ∂Ω (x) ⋅ (1 − ∂Ωt1 ) , ρ∂Ω (x) ≤ t1
⎪
∇ρ(x) = ⎨
⎩0, else,
⎪
⎪
∂ρ
and thus ∂N ∣∂Ω = −1, ∣∇ρ∣∞ ≤ 1 and ∇ρ is Lipschitz continuous. Furthermore
⎧ ρ (x)
⎪∆ρ∂Ω (x) (1 − ∂Ωt1 ) − t11 , ρ∂Ω (x) ≤ t1
⎪
∆ρ(x) = ⎨
⎩0, else,
⎪
⎪

and thus by inequality (6) on {ρ∂Ω ≤ t1 }

h′2 ρ∂Ω 1 h′ ρ∂Ω 1 −

∆ρ ≥ (d − 1) (ρ∂Ω ) (1 − ) − ⇒ (∆ρ)− ≤ ((d − 1) 2 (ρ∂Ω ) (1 − )− ) .
h2 t1 t1 h2 t1 t1
We can still choose t1 ∈ (0, h−1
2 (0)) to obtain for arbitrary ε > 0

h′2 t 1 −
∣(∆ρ)− ∣∞ ≤ inf−1 sup ((d − 1) (t) (1 − ) − ) + ε.
t1 ∈(0,h2 (0)) t∈(0,t1 ) h2 t1 t1

Inserting φ and ρ as defined in Lemma 2.1 and Lemma 2.2 in Proposition 2.2 and
Proposition 2.3 we now get explicit constants K1 , K2 and K∂Ω,Ω in terms of bounds on
sectional and Ricci curvature and second fundamental form on the boundary. We state
these in the following Proposition.
8 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Proposition 2.4. Let k1 , k2 ∈ R such that Ric ≥ (d − 1)k1 , sect ≤ k2 and γ1 , γ2 ∈ R such
that γ1 id ≤ Π ≤ γ2 id. Then the assumptions in Proposition 2.1 are fulfilled with
∣Ω∣ t0 t 2
K1 = inf inf [∫ (1 + ε)Hd−1 ({ρ∂Ω = t}) (1 − )
∣∂Ω∣2 t0 ∈(0,h−1
2 (0))
ε∈(0,∞) 0 t0
2
1 h′ t 1 −
+ (1 + ) CΩ Hd−1 ({ρ∂Ω = t})((((d − 1) 2 (t) (1 − ) − ) )
ε h2 t0 t0
2
h′ t 1 +
+ (((d − 1) 1 (t) (1 − ) − ) ) )dt],
h1 t0 t0

K2 = 0,
∣Ω∣ 1/2 h′ t 1 −
K∂Ω,Ω = (2CΩ + CΩ inf−1 sup ((d − 1) 2 (t) (1 − ) − ) ) .
∣∂Ω∣ t1 ∈(0,h2 (0)) t∈(0,t1 ) h2 t1 t1

As explained in Remark 2.1, upper bounds for the optimal K∂Ω,Ω correspond to lower
bounds for the first nontrivial Steklov eigenvalue. Thus we now also get a lower bound
on the first nontrivial Steklov eigenvalue σ that is explicit in terms of upper bounds on
sectional curvature and second fundamental form on the boundary:
Corollary 2.1. Let k2 ∈ R such that sect ≤ k2 and γ2 ∈ R such that Π ≤ γ2 id. Then for
the first non-trivial Steklov eigenalue σ of Ω it holds that
−1
1/2 h′2 t 1 −
σ≥ (2CΩ + CΩ inf sup ((d − 1) (t) (1 − ) − ) ) .
t1 ∈(0,h−1
2 (0)) t∈(0,t1 ) h2 t1 t1

Example 2.1. As an example one may consider a ball of radius one around the origin in
the hyperbolic plane. Then CΩ ≈ 0.3377 and h2 (t) = cosh(t) − coth(1)sinh(t) (for more
details on this see also Example 2.1.2 below). This results in the lower bound σ ≥ 0.5145
while as stated in [20] σ = coth(1) − tanh(1/2) ≈ 0.8509 for this example.
By inserting the set of constants stated in Proposition 2.4 into Proposition 2.1 we get
an explicit upper bound on the Poincaré constant again in terms of bounds on sectional
and Ricci curvature and second fundamental form. For comparison we state the following
obvious upper bound for the optimal Poincaré constant without the interpolation, which
can be derived from Proposition 2.1 by considering the limit as K∂Ω,Ω tends to infinity.
Proposition 2.5. Assume there exist constants K1 , K2 such that for any f ∈ C 1 (Ω)
2
(∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ + K2 ∫ ∣∇τ f ∣2 dλ∂Ω ,
Ω ∂Ω Ω ∂Ω

then it holds for any α ∈ (0, 1)

Cα ≤ max (CΩ + (1 − α)K1 , C∂Ω + αK2 ) .
2.1. Examples. In the following we consider as examples balls in Euclidean plane and in
hyperbolic plane and compare the results obtained above with or without the interpolation
approach. While the former example has already been treated in [30], the latter was not
included there due to negative curvature.
 9

2.1.1. Ball in R2 . Let Ω ∶= B1 be the unit ball in R2 . In this case the sectional curvature
equals k = 0, and for the second fundamental form on the boundary we have γ = 1. The
constants K1 , K2 and K∂Ω,Ω are now to be computed for this specific example. The general
results in [30, section 3.2] concerning the values for these constants are not applicable.
However the same paper also contains a computation adapted to this specific example. In
the following we will consider the values for the three constants obtained by computations
adapted to this specific example as well as the values obtained by the above general results.
For both sets of constants we will compare the upper bounds on Poincaré constants
obtained by the interpolation approach with the bounds obtained without interpolation
as in Proposition 2.5 and the exact values for the Poincaré constants.
We first recall the results of the computations adapted to the ball example made in [30,
section 3.1]:
1 3 1
CΩ ≈ , C∂Ω = 1, K1 = , K2 = 0, K∂Ω,Ω = .
3.39 16 2
The upper bound obtained from Proposition 2.1 is
8(1 − α) + 16αCΩ + 3α(1 − α)
Cα ≤ ,
8(1 + α)
while the upper bound obtained for the same values of K1 , K2 and K∂Ω,Ω from Proposition
2.5 is Cα ≤ 1. Moreover we also refer to [30] for the procedure to calculate the exact values
for Cα , α ∈ (0, 1). Using the results from the previous pages instead, we find different
constants: We have h1 (t) = h2 (t) = 1 − t from which follows by Lemma 2.1 and Lemma
2.2 that
∀ε > 0 ∃φ ∶ ∣∇φ∣22 ≤ ε, ∣∆φ∣22 ≤ 1 + ε
and
∀ε > 0 ∃ρ ∶ ∣∇ρ∣∞ ≤ 1, ∣(∆ρ)− ∣∞ ≤ 2 + ε.
Inserting this in Proposition 2.2 and Proposition 2.3 we get
CΩ
K1′ = , K2′ = 0, K∂Ω,Ω
′
≈ 0.8381.
4
Inserting this in Proposition 2.5 results in Cα ≤ 1 while inserting in Proposition 2.1 we
get
(1 − α)K∂Ω,Ω
′
+ αCΩ + α(1 − α)K1′
Cα ≤ max (CΩ + (1 − α)K1 ,′
).
(1 − α)K∂Ω,Ω
′
+α
We depict these results in Figure 1. Note that the green and purple curves overlap.
From this we see that the upper bounds obtained from the above general results are only
slightly worse than the upper bounds obtained by computing with the specific example in
mind. Furthermore it is obvious from the proof of Proposition 2.1, that results obtained
from the interpolation approach must be at least as good as results obtained without
interpolation. However the figure shows for both sets of constants that the interpolation
results clearly differ from the no interpolation results and give significantly better bounds
than the approach without interpolation.
We remark that using the interpolation approach with K1′ , K2′ and K∂Ω,Ω (i.e. combining
the respective best constants from above) would result in an even better approximation
of the curve of exact values.
10 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Figure 1. Exact Poincaré constants (blue), interpolation (yellow) and no

interpolation (green) results using computations specifically adapted to the
example, interpolation (red) and no interpolation (purple) results using
computations not specifically adapted to the example.

2.1.2. Ball in hyperbolic plane. We consider the unit metric ball in the hyperbolic plane
and compute the exact Poincaré constants Cα , α ∈ (0, 1) numerically. As in the previous
example we compare these with general/specific example and interpolation/no interpoal-
tion results. In more detail we consider the unit ball B1 (0) ⊂ R2 with the hyperbolic
metric
4
(7) gh = g,
(1 − ∣x∣2 )2
where g = (dx1 )2 + (dx2 )2 is the standard metric in R2 , resulting in the space form of
constant sectional curvature K = −1. Ω will be a unit ball in this hyperbolic plane. We
will start by computing the exact values for Cα , α ∈ (0, 1). Using that ∣Ω∣ = 2π(cosh(1)−1)
and ∣∂Ω∣ = 2πsinh(1) the operator A = Aα associated with the Dirichlet form Eα then
becomes
α sinh(1) ∂f
Aα f = ∆f 1Ω + (∆τ f − ) 1∂Ω .
1 − α cosh(1) − 1 ∂N
An eigenvector of −Aα for eigenvalue λ ≥ 0 is then a function f ∈ D(Aα ) such that the
following system of partial differential equations is fulfilled
⎧
⎪∆f = −λf
⎪ in Ω
⎨ τ α sinh(1) ∂f .
⎪
⎪ ∆ f − 1−α cosh(1)−1 ∂N = −λf on ∂Ω
⎩
Since f and Aα f are continuous on Ω, this is equivalent to
⎧
⎪∆f = −λf
⎪ in Ω
(8) ⎨ τ α sinh(1) ∂f .
⎪
⎪ ∆ f − 1−α cosh(1)−1 ∂N = ∆f on ∂Ω
⎩
Following the well-known procedure for the Laplacian with Neumann boundary condi-
tions, see e.g. [12, Chapter 2.5], we introduce spherical coordinates about x = 0 by
x = rξ, r = tanh(t/2),
 11

where r ∈ [0, 1], t ∈ [0, ∞), ξ ∈ S1 . Ω is then characterised by restriction of t to [0, 1]. In
these coordinates (7) reads
gh = (dt)2 + sinh2 (t)∣dξ∣2 .
We then separate variables, i.e. f (t, ξ) = T (t)G(ξ). Furthermore we denote by ◻ the
Laplacian on S1 and by ′ differentiation with respect to t. Using the Laplacian in spherical
coordinates, the first equation of (8) becomes
sinh(t)−1 (sinh(t)T ′ (t))′ G(ξ) + sinh(t)−2 T (t) ◻ G(ξ) = −λT (t)G(ξ).
and in terms of G and T
◻G(ξ) + γG(ξ) = 0
{
(sinh(t)T ′ (t))′ + (λ − γsinh(t)−2 )sinh(t)T (t) = 0,
where γ = l2 , l ∈ N are the negatives of the eigenvalues of ◻ on S1 . According to [12,
Chapter 12.5] the solution T is given via
T (t) = Pνµ (cosh(t)),
where Pνµ (⋅) is the associated Legendre function of first kind with µ and ν given via
√
1 1
µ = l, ν = − ± −λ + .
2 4
We thus obtain a two parameter family of eigenfunctions fn,l (t, ξ) = Pnl (cosh(t))Gl (ξ)
where Gl is the eigenfunction for eigenvalue −l2 and n respectively λn is constrained via
the boundary condition as follows. Using that ∆τ f = sinh12 (1) T ◻ G, the second equation
of (8) which holds on the boundary, i.e. for t = 1, becomes
1 sinh(1) α
∆(T G)(1, ξ) = 2
T (1) ◻ G(ξ) − T ′ (1)G(ξ)
sinh (1) cosh(1) − 1 1 − α
cosh(1) sinh(1) α
⇔T ′′ (1) + T ′ (1) ( + )=0
sinh(1) cosh(1) − 1 1 − α
′′ ′ sinh2 (1) α
⇔Pνl (cosh(1))sinh2 (1) + Pνl (cosh(1)) (2cosh(1) + ) = 0.
cosh(1) − 1 1 − α
We obtain a two-parameter family of values λl,n satisfying this equation. For α ∈ (0, 1)
λα ∶= minl,n λl,n is the desired spectral gap and Cα = 1/λα .

As we found no explicit account of CΩ and C∂Ω we also compute these here. Following
the same procedure as above including spherical coordinates and separation of variables
we see that eigenfunctions f of the Laplacian on Ω with Neumann boundary conditions
on ∂Ω are again of the form
√
1 1
f (t, ξ) = T (t)G(ξ), with T (t) = Pνµ (cosh(t)), µ = l, ν = − ± −λ + ,
2 4
where G are eigenfunctions of the Laplacian on S1 for the eigenvalues −l2 , l ∈ N.
Now the boundary condition amounts to
∂f
= 0 on ∂Ω ⇔ Pνµ ′ (cosh(1)) = 0.
∂N
We again obtain a two-parameter family of values λl,n satisfying this equation. Again
λ1 ∶= minl,n λl,n ≈ 2.9614 is the desired spectral gap and CΩ = 1/λ1 ≈ 0.3377.
12 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Furthermore by scaling we can deduce that since the eigenvalues of ∆SR1 are known to
2
be λk ∶= −k 2 , k ∈ N0 (see e.g. [12, Chapter 2.4]), those of ∆SH1 are λ̃k = − sinhk 2 (1) , k ∈ N0
and thus C∂Ω = sinh2 (1).

We now compute K1 and K2 in a fashion adapted to the specific example. The following
computation is similar to the one referenced above in Example 2.1.1. Using that for
f ∈ L1 (∂Ω):

∫∂Ω f (y)λ∂Ω (dy) = ∫Ω f (x/∣x∣)λΩ (dx),

we get
2
(∫ f dλΩ − ∫ f dλ∂Ω )
Ω ∂Ω

≤ ∫ (f (x) − f (x/ ∣x∣)2 λΩ (dx)

Ω
1 1
2
= ∫ ∫ (f (tanh(t/2)ξ) − f (tanh(1/2)ξ)) sinh(t)dtdξ
∣Ω∣ ∂Ω 0
1 1 d 2
1
= ( f (tanh(s/2)ξ)ds) sinh(t)dtdξ
∣Ω∣ ∫∂Ω ∫0 ∫t ds
1 1 2
1 d
≤ (1 − t) ∫ ( f (tanh(s/2)ξ)) ds sinh(t)dtdξ
∣Ω∣ ∫∂Ω ∫0 t ds
1 s 2
1 d
= (1 − t)sinh(t)dt ( f (tanh(s/2)ξ)) dsdξ
∣Ω∣ ∫∂Ω ∫0 ∫0 ds
1 2
1 d
= (sinh(s) − (s − 1)cosh(s) − 1) (⟨∇f (tanh(s/2)ξ), tanh(s/2)ξ⟩) dsdξ
∣Ω∣ ∫∂Ω ∫0 ds
1 1
≤ ∫ ∫ (sinh(s) − (s − 1)cosh(s) − 1) ∣∇f (tanh(s/2)ξ)∣2 dsdξ
∣Ω∣ ∂Ω 0
1 1
≤ K1 ∫ ∫ ∣∇f (tanh(s/2)ξ)∣2 sinh(s)dsdξ
∣Ω∣ ∂Ω 0
= K1 ∫ ∣∇f (x)∣2 λΩ (dx),
Ω

where
sinh(s) − (s − 1)cosh(s) − 1
K1 ∶= max ≈ 0.1782.
s∈[0,1] sinh(s)
In particular, with K1 as above setting K2 = 0 is an admissible choice.
As explained in Remark 2.1 we may obtain the optimal constant K∂Ω,Ω in Proposi-
∣Ω∣ −1
tion 2.1 as ∣∂Ω∣ σ , where σ denotes the first nontrivial Steklov eigenvalue. In the present
example the first Steklov eigenvalue is coth(1) − tanh(1/2), cf. [20] and thus
cosh(1) − 1
K∂Ω,Ω = (coth(1) − tanh(1/2))−1 ≈ 0.5431.
sinh(1)
Inserting this in Proposition 2.5 results in Cα ≤ C∂Ω ≈ 1.3811. Furthermore we insert the
same set of constants in Proposition 2.1.
Using the general results from the previous section instead, we find different constants:
 13

Figure 2. Exact Poincaré constants (blue), interpolation (yellow) and no

interpolation (green) results using computations specifically adapted to the
example, interpolation (red) and no interpolation (purple) results using
computations not specifically adapted to the example.

We have h1 (t) = h2 (t) = cosh(t) − coth(1)sinh(t) and H1 ({ρ∂Ω = t}) = 2πsinh(t) from
which follows by Lemma 2.1 (with t0 → 0) and Lemma 2.2 (with t1 → 1) that
1
∀ε > 0 ∃φ ∶ ∣∇φ∣22 ≤ ε, ∣∆φ∣22 ≤ + ε,
2(cosh(1) − 1)
and
∃ρ ∶ ∣∇ρ∣∞ ≤ 1, ∣(∆ρ)− ∣∞ ≤ 2.3131.
Inserting this in Proposition 2.2 and Proposition 2.3 we get
sinh2 (1/2)
K1′ = ⋅ CΩ ≈ 0.0664, K2′ = 0, K∂Ω,Ω
′
≈ 0.8981.
sinh2 (1)
Inserting this in Proposition 2.5 results in Cα ≤ C∂Ω . Furthermore we insert the same set
of constants in Proposition 2.1.

The curves for actual Poincaré constants and respective upper bounds via Proposi-
tion 2.1 as well as via Proposition 2.5 obtained by plugging in the quantities collected
above are depicted in Figure 2. Note that the green and purple curves overlap. Again from
the figure we may see that our general results are only slightly worse than the ones ob-
tained from computations specifically adapted to the example, and that the interpolation
approach results in a significant improvement compared to no interpolation.

3. Purely sticky reflection case (β = 0)

The above results may as well be used to give upper bounds for Brownian motion with
sticky reflection from the boundary (but without boundary diffusion). I.e. under the same
assumptions on Ω as above we consider a diffusion on Ω with Feller generator (D(Â), Â)
D(Â) = {f ∈ C(Ω) ∣ Âf ∈ C(Ω)}
∂f
Âf = ∆f 1Ω − γ 1∂Ω ,
∂N
14 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

∂f
where ∂N is the outer normal derivative and γ > 0, which corresponds to inward sticky
reflection at ∂Ω. A construction was given again in [25] using Dirichlet forms. We choose
α ∈ (0, 1), such that
α ∣∂Ω∣
= γ.
1 − α ∣Ω∣
and set
λα ∶= αλΩ + (1 − α)λ∂Ω .
We find that −Â is λα -symmetric with spectral gap σ̂α characterised by the Rayleigh
quotient resp. Poincaré constant Ĉα
Eˆα (f ) V arλα (f )
σ̂α = inf , Ĉ α ∶= σ̂α
−1
= sup
f ∈C 1 (Ω) V arλα (f ) f ∈C 1 (Ω) Eˆα (f )
V arλα (f )>0 Eˆα (f )>0

where
2
V arλα (f ) = ∫ f 2 dλα − (∫ f dλα ) , Eˆα (f ) = α ∫ ∣∇f ∣2 dλΩ , f ∈ C 1 (Ω)
Ω Ω Ω

By CΩ and C∂Ω we still denote the usual (Neumann) Poincaré constants of Ω and ∂Ω
respectively, which we assume to be known.

The eigenvalue problem corresponding to the Poincaré constant may be stated as

∆f = −λf in Ω,
{ ∂f
−γ ∂N = −λf on ∂Ω.
This type of eigenvalue problems with eigenvalue featured in the boundary condition has
been of separate interest, see e.g. [5], [48], [22], [3], [9]. For Brownian motion with sticky
reflection spectral asymptotics have been examined e.g. in [48], however we are not aware
of results on the spectral gap.

Proposition 3.1. Assume there exist constants K∂Ω,Ω , K1 such that for any f ∈ C 1 (Ω)
V arλ∂Ω (f ) ≤ K∂Ω,Ω ∫ ∣∇f ∣2 dλΩ
Ω
and
2
(∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ ,
Ω ∂Ω Ω
then it holds for any α ∈ (0, 1)
(1 − α)
Ĉα ≤ CΩ + K∂Ω,Ω + (1 − α)K1 .
α
Proof. Let f ∈ C 1 (Ω)
2
V arλα (f ) = αV arλΩ (f ) + (1 − α)V arλ∂Ω (f ) + α(1 − α) (∫ f dλΩ − ∫ f dλ∂Ω )
Ω ∂Ω

≤ αCΩ ∫ ∣∇f ∣2 dλΩ + (1 − α)K∂Ω,Ω ∫ ∣∇f ∣2 dλΩ + α(1 − α)K1 ∫ ∣∇f ∣2 dλΩ
Ω Ω Ω
(1 − α)
= (CΩ + K∂Ω,Ω + (1 − α)K1 ) α ∫ ∣∇f ∣2 dλΩ .
α Ω
□
 15

For f ∈ C 1 (Ω) with V arλ∂Ω (f ) > 0 the term V arλα (f ) stays positive as α tends to zero
while Eˆα (f ) vanishes. Thus Ĉα blows up as α tends to zero and accordingly so does the
bound on Ĉα proven in Proposition 3.1. Of course the interpolation approach from above
is not of any use anymore in this setting. As we have previously shown that inequality (2)
can be fulfilled with K2 = 0, we may now insert K1 and K∂Ω,Ω as computed above. Thus
in sum the results obtained above for Brownian motion with sticky reflecting boundary
diffusion may also be used for the case without boundary diffusion precisely for the reason
that we were able to show that inequality (2) is fulfilled with K2 = 0. If we make the more
strict assumption of respective positive lower bounds kR on Ricci curvature on Ω and γ
on second fundamental form on ∂Ω as in [30, section 3.2] we may as well insert the set of
d−1 ∣Ω∣ 2
constants K1 = dk R
, K2 = 0, K∂Ω,Ω = ∣∂Ω∣ γ obtained there. Note that this is possible as we
again have K2 = 0.

3.1. Examples. We again consider as examples a unit ball in R2 and a unit metric ball
in the hyperbolic plane.

3.1.1. Ball in R2 . As in Example 2.1.1 we consider a unit ball in R2 . In order to compute

the exact values for Ĉα , α ∈ (0, 1) we again proceed as described in [30, section 3.1] and
only need to adapt the boundary condition. I.e. an eigenfunction f fulfills
∆f = −λf in Ω
{ 2α ∂f
− 1−α ∂N = ∆f on ∂Ω

and by passing to polar coordinates and seperating variables we obtain a family λ̂m,l , m, l ∈
N0 characterised by
√ √ √ 1+α √ m
λJm
′′
( λ) + Jm
′
( λ) − Jm ( λ) = 0,
1−α λ
where Jm , m ∈ N0 are the Bessel functions of the first kind. We then get λ̂α = minm,l∈N0 λ̂m,l
and Ĉα = λ̂1 .
α

In order to calculate the explicit values for the upper bound stated in Proposition 3.1 we
need CΩ , K∂Ω,Ω and K1 . All of these have been computed in Example 2.1.1, in particular
K∂Ω,Ω and K1 have been computed once in a manner adapted to the specific example and
once from the previously stated general results (the latter marked by ′ ).
The curves for actual Poincaré constants and respective upper bounds via Proposition 3.1
obtained by plugging in the quantities collected above are depicted in Figure 3. As
mentioned before Ĉα blows up as α tends to zero. We therefore only consider α ≥ 0.2
for the plot. Figure 3 suggests that Proposition 3.1 offers a precise upper bound for Ĉα
that is (in particular for small α) highly depended on how close the values of K1 and
K∂Ω,Ω are to the optimal constants in inequality (1) and inequality (2). More precisely
for small values of α it is mainly the precision of the value for K∂Ω,Ω that is relevant. Note
that the value K∂Ω,Ω
′
obtained from our general results is worse than K∂Ω,Ω obtained from
computations adapted to the specific example, while the opposite is true for the values of
K1′ and K1 .

3.1.2. Ball in hyperbolic plane. As in Example 2.1.2 we consider a unit ball in the hyper-
bolic plane and use the notation introduced above. To calculate the exact values of the
16 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Figure 3. Exact Poincaré constants (blue), estimates using computations

specifically adapted to the example (yellow), estimates using computations
not specifically adapted to the example (green).

Poincaré constants we proceed as explained in Example 2.1.2 and only need to adapt the
boundary condition (see the second equation in (8)) to
−α sinh(1) ∂f
= ∆f on ∂Ω.
1 − α cosh(1) − 1 ∂N
Inserting f (t, ξ) = T (t)G(ξ) and then using T (t) = Pνµ (cosh(t)) as before, this results in
−α sinh(1) cosh(1) ′ T (1)
T ′ (1)G(ξ) = ( T (1) + T ′′ (1)) G(ξ) + ◻ G(ξ)
1 − α cosh(1) − 1 sinh(1) sinh2 (1)
α sinh(1) cosh(1) T (1)
⇔ T ′ (1) ( + ) + T ′′ (1) − γ =0⇔
1 − α cosh(1) − 1 sinh(1) sinh2 (1)
α sinh2 (1) Pνµ (cosh(1))
Pνµ ′′ (cosh(1))sinh2 (1) + Pνµ ′ (cosh(1)) ( + 2cosh(1)) − γ = 0.
1 − α cosh(1) − 1 sinh2 (1)
We obtain a two-parameter family of values λ̂l,n satisfying this equation. For α ∈ (0, 1)
λ̂α ∶= minl,n λ̂l,n is the desired spectral gap and Ĉα = 1/λ̂α .
In order to calculate the explicit values for the upper bound stated in Proposition 3.1 we
may again use the values for CΩ , K∂Ω,Ω and K1 as computed in Example 2.1.2 in a manner
adapted to the specific example or from the previously stated general results (the latter
marked by ′ ). The curves for actual Poincaré constants and respective upper bounds via
Proposition 3.1 obtained by plugging in these quantities are depicted in Figure 4. Again
we only consider α ≥ 0.2 for the plot, as Ĉα blows up as α tends to zero. From Figure
4 we may again see that the precision of the upper bound for Ĉα offered in Proposition
3.1 depends particularly for small α highly on how close the values of K1 and K∂Ω,Ω are
to the optimal constants in inequality (1) and inequality (2). Note that in this example
again the value K∂Ω,Ω ′
is worse than K∂Ω,Ω , while the opposite is true for K1′ and K1 .

4. Logarithmic Sobolev Inequality

Using the notation from above we say that a (tight) logarithmic Sobolev inequality is
fulfilled if
∃Lα ≥ 0 s.t. Entλα (f 2 ) ≤ Lα ⋅ Eα (f ) ∀f ∈ C 1 (Ω).
 17

Figure 4. Exact Poincaré constants (blue), estimates using computations

specifically adapted to the example (yellow) and estimates using computa-
tions not specifically adapted to the example (green).

where

Entλα (f 2 ) = ∫ f 2 log(f 2 )dλα − (∫ f 2 dλα ) log (∫ f 2 dλα )

Ω Ω Ω

Eα (f ) = α ∫ ∣∇f ∣2 dλΩ + (1 − α) ∫ ∣∇τ f ∣2 dλ∂Ω , f ∈ C 1 (Ω)

Ω ∂Ω

and ∇τ denotes the tangential derivative operator on ∂Ω.

In the following by Lα we will denote the optimal such constant. By LΩ respectively
L∂Ω we will denote the optimal logarithmic Sobolev constant associated to the Laplace
operator on Ω with Neumann boundary conditions and the logarithmic Sobolev constant
associated to the Laplace-Beltrami operator on ∂Ω. We assume that LΩ and L∂Ω (or
upper bounds for them) are known. We aim at bounding Lα for α ∈ (0, 1) from above.
We consider here the entropy with respect to λα which is a mixture or more specifically
a convex combination of the two measures λΩ and λ∂Ω . The entropy with respect to
mixtures of two measures such as λα has been considered previously e.g. in [11], [43], [37].
We first show an analogue of [30, Proposition 2.1] for the entropy with respect to λα :

Proposition 4.1. Assume there are constants K∂Ω,Ω , L∂Ω,Ω , K1 , K2 such that ∀f ∈ C 1 (Ω):

(9) V arλ∂Ω (f ) ≤ K∂Ω,Ω ∫ ∣∇f ∣2 dλΩ

Ω

(10) Entλ∂Ω (f 2 ) ≤ L∂Ω,Ω ∫ ∣∇f ∣2 dλΩ

Ω
2
(11) (∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ + K2 ∫ ∣∇τ f ∣2 dλ∂Ω .
Ω ∂Ω Ω ∂Ω

Then it holds for any α ∈ (0, 1)

(1 − α) (1 − α)(log(α) − log(1 − α))
Lα ≤ inf max{LΩ + sL∂Ω,Ω + (CΩ + tK∂Ω,Ω + K1 ) ,
s,t∈[0,1] α 2α − 1
α(log(α) − log(1 − α))
(1 − s)L∂Ω + ((1 − t)C∂Ω + K2 )}.
2α − 1
18 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Proof. Applying (9) and (10) we can estimate for any f ∈ C 1 (Ω):

V arλ∂Ω (f ) ≤ tK∂Ω,Ω ∫ ∣∇f ∣2 dλΩ + (1 − t)C∂Ω ∫ ∣∇τ f ∣2 dλ∂Ω

Ω ∂Ω

Entλ∂Ω (f 2 ) ≤ sL∂Ω,Ω ∫ ∣∇f ∣2 dλΩ + (1 − s)L∂Ω ∫ ∣∇τ f ∣2 dλ∂Ω

Ω ∂Ω

for any s, t ∈ [0, 1]. We apply this in the following after first using a decomposition of the
entropy with respect to the mixture of two measures as well as an optimal logarithmic
Sobolev inequality for Bernoulli measures as described in [11, section 4]
Entλα (f 2 ) ≤ αEntλΩ (f 2 ) + (1 − α)Entλ∂Ω (f 2 )
α(1 − α)(log(α) − log(1 − α))
+ (V arλΩ (f ) + V arλ∂Ω (f ) + (EλΩ (f ) − Eλ∂Ω (f ))2 )
2α − 1
α(1 − α)(log(α) − log(1 − α))
≤ ∫ ∣∇f ∣2 dλΩ (αLΩ + (1 − α)sL∂Ω,Ω + (CΩ + tK∂Ω,Ω + K1 ))
Ω 2α − 1
α(1 − α)(log(α) − log(1 − α))
+ ∫ ∣∇τ f ∣2 dλ∂Ω ((1 − α)(1 − s)L∂Ω + ((1 − t)C∂Ω + K2 )) .
∂Ω 2α − 1
And thus
(1 − α) (1 − α)(log(α) − log(1 − α))
Lα ≤ inf max{LΩ + sL∂Ω,Ω + (CΩ + tK∂Ω,Ω + K1 ) ,
s,t∈[0,1] α 2α − 1
α(log(α) − log(1 − α))
(1 − s)L∂Ω + ((1 − t)C∂Ω + K2 )}.
2α − 1
□
As mentioned previously one may again add another optimisation over the pairs of
values K1 , K2 fulfilling (11) in order to improve the upper bound on Lα .
Remark 4.1. Using that for any a, b, c, d, e, θ ∈ R≥0 it holds
inf max(a + sb + tc, d − se − tθ)
s,t∈[0,1]
⎧
⎪ a, a>d if
⎪
⎪
⎪
⎪
⎪
⎪ d − e − θ, if
d−e−θ >a+b+c
⎪
⎪
⎪
⎪a + c + b ⋅ ( d−a−c−θ
⎪
e+b
), if e+b
a ≤ d, d − e − θ ≤ a + b + c, b − c ⋅ c+θ ≥ 0, d−a−c−θ
e+b ≥0
=⎨ d−a e+b d−a−c−θ
⎪
⎪
⎪ a + c ⋅ ( c+θ ) , if
a ≤ d, d − e − θ ≤ a + b + c, b − c ⋅ c+θ ≥ 0, e+b < 0
⎪
⎪
⎪
⎪
⎪
⎪ a + b ⋅ ( d−a
e+b ,
) if e+b
a ≤ d, d − e − θ ≤ a + b + c, b − c ⋅ c+θ < 0, d−a
e+b ≤ 1
⎪
⎪ d−a−e−b e+b d−a
⎩a + b + c ⋅ ( c+θ ) ,
⎪ if
a ≤ d, d − e − θ ≤ a + b + c, b − c ⋅ c+θ < 0, e+b > 1
d−a−c−θ
= max{a, d − e − θ, min[a + c + b ( ),
e+b
d−a d−a d−a−e−b
a + c( ),a + b( ),a + b + c( )]},
c+θ e+b c+θ
we may rewrite the result of Proposition 4.1 in analogy with Proposition 2.1 as follows:
For any α ∈ (0, 1)
Lα ≤ max{a, d − e − θ,
d−a−c−θ d−a d−a d−a−e−b
min[a + c + b ( ),a + c( ),a + b( ),a + b + c( )]},
e+b c+θ e+b c+θ
 19

where
(1 − α)(log(α) − log(1 − α)) 1−α
a = LΩ + (CΩ + K1 ), b= L∂Ω,Ω ,
2α − 1 α
(1 − α)(log(α) − log(1 − α)) α(log(α) − log(1 − α))
c= K∂Ω,Ω , d = L∂Ω + (C∂Ω + K2 )
2α − 1 2α − 1
α(log(α) − log(1 − α))
e = L∂Ω , θ= C∂Ω .
2α − 1
For comparison we also state the simpler result one can obtain based on [11, section 4]
without interpolation.
Proposition 4.2. Assume there exist constants K1 , K2 such that ∀f ∈ C 1 (Ω):
2
(∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ + K2 ∫ ∣∇τ f ∣2 dλ∂Ω .
Ω ∂Ω Ω ∂Ω

Then it holds for any α ∈ (0, 1)

(1 − α)(log(α) − log(1 − α))
Lα ≤ max{LΩ + (CΩ + K1 ),
2α − 1
α(log(α) − log(1 − α))
L∂Ω + (C∂Ω + K2 )}.
2α − 1
Proof. As in the proof of Proposition 4.1 we use a decomposition of the entropy with
respect to the mixture of two measures as well as an optimal logarithmic Sobolev inequality
for Bernoulli measures as described in [11, section 4]:
α(1 − α)(log(α) − log(1 − α))
Entλα (f 2 ) ≤αEntλΩ (f 2 ) + (1 − α)Entλ∂Ω (f 2 ) +
2α − 1
⋅ (V arλΩ (f ) + V arλ∂Ω (f ) + (EλΩ (f ) − Eλ∂Ω (f ))2 )
α(1 − α)(log(α) − log(1 − α))
≤αLΩ ∫ ∣∇f ∣2 dλΩ + (1 − α)L∂Ω ∫ ∣∇τ f ∣2 dλ∂Ω +
Ω ∂Ω 2α − 1
⋅ ((CΩ + K1 ) ∫ ∣∇f ∣2 dλΩ + (C∂Ω + K2 ) ∫ ∣∇τ f ∣2 dλ∂Ω )
Ω ∂Ω
(1 − α)(log(α) − log(1 − α))
= (LΩ + (CΩ + K1 )) α ∫ ∣∇f ∣2 dλΩ
2α − 1 Ω
α(log(α) − log(1 − α))
+ (L∂Ω + (C∂Ω + K2 )) (1 − α) ∫ ∣∇τ f ∣2 dλ∂Ω
2α − 1 ∂Ω
(1 − α)(log(α) − log(1 − α))
≤ max(LΩ + (CΩ + K1 ),
2α − 1
α(log(α) − log(1 − α))
L∂Ω + (C∂Ω + K2 )) ⋅ Eα (f ).
2α − 1
□
In general the logarithmic Sobolev constant of a mixture of two measures may blow up
as the mixture proportion goes to 0 or 1, c.f. [11]. Accordingly so may our bounds for the
logarithmic Sobolev constant as α approaches 0 or 1. The upper bound in Proposition 4.2
always blows up as α approaches 0 or 1. The same is true for the bound in Proposition
4.1 as α tends to 0 but not necessarily as α tends to 1. On the contrary the Poincaré
constant as well as the upper bound for it in Proposition 2.1 does not blow up as α tends
to 0 or 1.
20 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

Figure 5. Lower bound via exact Poincaré constants (blue), and upper
bound via interpolation (yellow) and no interpolation (green) for the Log-
arithmic Sobolev constant

4.1. Example: Ball in R2 . To show the feasibility of the approach consider again Ex-
ample 2.1.1: We consider Ω = B1 , the unit ball in R2 . We evaluate the upper bounds for
Lα , α ∈ (0, 1) obtained via Proposition 4.2 and Proposition 4.1. Furthermore we use the
following quantities collected in Example 2.1.1:
1 CΩ 1
CΩ = , C∂Ω = 1, K1′ = , K2 = 0, K∂Ω,Ω = .
3.39 4 2
Added to that we need values for LΩ , L∂Ω and L∂Ω,Ω . It follows by the Bakry-Émery
criterion that L∂Ω ≤ 2, cf. [2]. Furthermore from [4] we obtain L∂Ω,Ω = 1 and from [14] we
obtain LΩ ≤ 1.1799
Since we have not computed the precise value of Lα we compare our estimate with the
corresponding spectral gap, using that a logarithmic Sobolev inequality with constant C
implies a Poincaré inequality with constant C/2, cf. [2, Proposition 5.1.3]. Thus a lower
bound for the optimal logarithmic Sobolev constants is given via
Lα ≥ 2 ⋅ Cα , ∀α ∈ (0, 1),
with Cα as computed in Example 2.1.1.
We insert the collected quantities into Proposition 4.2 and Proposition 4.1 and depict the
results in Figure 5. Note that the yellow and green curves partly overlap. The figure
shows that the interpolation results clearly differ from the no interpolation results and
give significantly better bounds than the approach without interpolation. In particular
the interpolation approach gives an upper bound that does not blow up as α tends to 1.

4.2. Example: Ball in hyperbolic plane. We revisit Example 2.1.2: We consider the
unit metric ball in the hyperbolic plane and evaluate the upper bounds for Lα , α ∈ (0, 1)
obtained via Proposition 4.2 and Proposition 4.1. We recall the quantities collected above
in Example 2.1.2:
sinh2 (1/2)
CΩ = 0.3377, C∂Ω = sinh2 (1), K1′ = ⋅ CΩ ,
sinh2 (1)
cosh(1) − 1
K2 = 0, K∂Ω,Ω = .
sinh(1)(coth(1) − tanh(1/2))
 21

Figure 6. Lower bound via exact Poincaré constants (blue), and upper
bound via interpolation (yellow) and no interpolation (green) for the Log-
arithmic Sobolev constant

Furthermore it follows by the Bakry-Émery criterion that L∂Ω ≤ 2 ⋅ sinh2 (1), cf. [2] and
from [14] we obtain LΩ ≤ 3.5088.
To obtain an upper bound on L∂Ω,Ω we proceed as follows: Rewriting integrals over ∂Ω as
integrals over the Euclidean unit sphere and using the dual-spectral form of the Hardy-
Littlewood-Sobolev inequality (for the Euclidean unit sphere) and bound of its right-hand
side as presented in [4][section 2] we obtain for q ∈ [2, ∞)
2/q
(∫ ∣f ∣q dλ∂Ω ) ≤ ∑ a2k (1 + (q − 2)k) ∫ ∣yk ∣2 dλ∂Ω .
∂Ω k ∂Ω

Here ∑k ak yk is a representation of f in terms of spherical harmonics, i.e. yk is an eigen-

function of the spherical hyperbolic Laplacian for the eigenvalue −k 2 /sinh2 (1). Using
separation of variables we may extend each yk to a harmonic function on Ω with normal
derivative on ∂Ω equal to k/sinh(1). We thus get as an analogue of [4][equation (36)] for
our unit ball in the hyperbolic plane:
2/q
(∫ ∣f ∣q dλ∂Ω ) ≤∫ ∣f ∣2 dλ∂Ω + (q − 2)(cosh(1) − 1) ∫ ∣∇u∣2 dλΩ ,
∂Ω ∂Ω Ω

where u is the harmonic extension of f to Ω. Proceeding as in [2, Proposition 6.2.3] (see

also [2, Proposition 5.1.8] for details), this results in

Entλ∂Ω (f 2 ) ≤ 2(cosh(1) − 1) ∫ ∣∇f ∣2 dλΩ ,

Ω

i.e. L∂Ω,Ω ≤ 2(cosh(1) − 1).

We do not compute the precise value of Lα , but use the lower bound for the optimal
logarithmic Sobolev constants given via
Lα ≥ 2 ⋅ Cα , ∀α ∈ (0, 1),
with Cα as computed in Example 2.1.2.
We insert the collected quantities into Proposition 4.2 and Proposition 4.1 and depict
the results in Figure 6. Note that the yellow and green curves partly overlap. Again
this figure shows the advantage of the interpolation results as compared to the approach
without interpolation.
22 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

5. Purely sticky reflection case (β = 0)

As in section 3 instead of Brownian motion with sticky reflecting boundary diffusion
the above results may as well be used to give upper bounds for Brownian motion with
sticky reflection from the boundary (but without boundary diffusion). The Logarithmic
Sobolev inequality in this setting is

Entλα (f 2 ) ≤ L̂α Eˆα (f ) ∀f ∈ C 1 (Ω),

where
Eˆα (f ) = α ∫ ∣∇f ∣2 dλΩ , f ∈ C 1 (Ω).
Ω

Proposition 5.1. Assume there are constants K∂Ω,Ω , L∂Ω,Ω , K1 such that for f ∈ C 1 (Ω)

V arλ∂Ω f ≤ K∂Ω,Ω ∫ ∣∇f ∣2 dλΩ ,

Ω

Entλ∂Ω (f ) ≤ L∂Ω,Ω ∫ ∣∇f ∣2 dλΩ ,

2
Ω
2
(∫ f dλΩ − ∫ f dλ∂Ω ) ≤ K1 ∫ ∣∇f ∣2 dλΩ ,
Ω ∂Ω Ω

then it holds for any α ∈ (0, 1)

(1 − α) (1 − α)(log(α) − log(1 − α))
L̂α ≤ (LΩ + L∂Ω,Ω + (CΩ + K∂Ω,Ω + K1 )) .
α 2α − 1
Proof. As above we use a decomposition of the entropy with respect to the mixture of
two measures as well as an optimal logarithmic Sobolev inequality for Bernoulli measures
as described in [11, section 4]:

Entλα (f 2 ) ≤ αEntλΩ (f 2 ) + (1 − α)Entλ∂Ω (f 2 )

α(1 − α)(log(α) − log(1 − α))
+ (V arλΩ (f ) + V arλ∂Ω (f ) + (EλΩ (f ) − Eλ∂Ω (f ))2 )
2α − 1
(1 − α) (1 − α)(log(α) − log(1 − α))
≤ (LΩ + L∂Ω,Ω + (CΩ + K∂Ω,Ω + K1 )) α ∫ ∣∇f ∣2 dλΩ .
α 2α − 1 Ω

6. Boundary-Interior Inequalities
To obtain explicit bounds on Lα in the general setting we can use the results from
section 2 for estimate (9) and estimate (11) and so it remains to find L∂Ω,Ω such that
inequality (10) is fulfilled. In this final section we present a general approach to establish
explicit geometric estimates for this quantity under general curvature assumptions.

6.1. Sobolev-Poincaré-Trace Inequalities. In the following we present some boundary-

interior inequalities that can be proved in a similar fashion as Proposition 2.2. They may
be seen as alternatives for Proposition 2.2 but might also be of independent interest. We
start from the following statement obtained in the proof of Proposition 2.3.
 23

∂ρ
Proposition 6.1. For any ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is Lipschitz continuous
on Ω it holds
1/2
∣Ω∣
∫∂Ω f 2 dλ∂Ω ≤ {2∣∇ρ∣∞ (∫ f 2 dλΩ ⋅ ∫ ∣∇f ∣2 dλΩ ) + ∣(∆ρ)− ∣∞ ∫ f 2 dλΩ }
∣∂Ω∣ Ω Ω Ω

1/2 ∣Ω∣
≤ {2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ } ∣∇f ∣2 dλΩ
∣∂Ω∣ ∫Ω
∀f ∈ C 1 (Ω) with ∫Ω f dλΩ = 0.
The statement in Proposition 6.1 is stronger than needed for inequality (2) in Propo-
sition 2.1 because we bound from above the integral of the squared function as opposed
to the square of the integral. Nevertheless the proof of the next corollary follows directly
as we may assume for inequality (2) without loss of generality that ∫Ω f dλΩ = 0. We thus
get an upper bound for K1 in inequality (2) that is alternative to Proposition 2.2.
∂ρ
Corollary 6.1. For any ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is Lipschitz continuous
on Ω inequality (2) in Proposition 2.1 is fulfilled with K2 = 0 and
1/2 ∣Ω∣
K1 = {2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ } .
∣∂Ω∣
Additionally we think that this computation is of independent interest for the following
reason:
Remark 6.1. We may calculate as in the proof of Proposition 6.1 to obtain for f ∈ C 1 (Ω)
(but not necessarily centered on Ω):
1/2
∣Ω∣
∫∂Ω f 2 dλ∂Ω ≤ {2∣∇ρ∣∞ (∫ f 2 dλΩ ⋅ ∫ ∣∇f ∣2 dλΩ ) + ∣(∆ρ)− ∣∞ ∫ f 2 dλΩ }
∣∂Ω∣ Ω Ω Ω

∣Ω∣
≤ {∣∇ρ∣∞ ∫ ∣∇f ∣2 dλΩ + (∣∇ρ∣∞ + ∣(∆ρ)− ∣∞ ) ∫ f 2 dλΩ } .
∣∂Ω∣ Ω Ω

∣Ω∣
From this it follows that for K3 ∶= ∣∂Ω∣ (∣∇ρ∣∞ + ∣(∆ρ)− ∣∞ )
√
∣f ∣2L2 (∂Ω,λ∂Ω ) ≤ K3 ∣f ∣2W 1,2 (Ω,λΩ ) ⇔ ∣f ∣L2 (∂Ω,λ∂Ω ) ≤ K3 ∣f ∣W 1,2 (Ω,λΩ ) .
As W 1,2 (Ω, λΩ ) is the completion of smooth functions whose derivatives up to degree 1 are
in L2 (Ω, λΩ ), the inequality also holds for all functions in W 1,2 (Ω, λΩ ). Thus via stating a
specific constant K3 , as can be obtained from Lemma 2.2, we also give an upper bound for
the norm of the Trace operator ∣∂Ω ∶ W 1,2 (Ω, λΩ ) → L2 (∂Ω, λ∂Ω ) that is explicit in terms
of upper bounds on sectional curvature and second fundamental form on the boundary.
(An optimal upper bound in terms of the geometry of Ω seems to be unknown in this
form as of yet.)
Proposition 6.2. Let k2 ∈ R such that sect ≤ k2 and γ2 ∈ R such that Π ≤ γ2 id. Then the
norm of the Trace operator ∣∂Ω ∶ W 1,2 (Ω, λΩ ) → L2 (∂Ω, λ∂Ω ) is bounded from above by
1/2
∣Ω∣ h′ t 1 −
( (1 + inf−1 sup ((d − 1) 2 (t) (1 − ) − ) )) .
∣∂Ω∣ t1 ∈(0,h2 (0)) t∈(0,t1 ) h2 t1 t1
It is known that on a smooth, compact d-dimensional Riemannian manifold (Ω, g) for
qd
q ∈ [1, d) and p1 = 1q − d1 (and thus for all p ∈ [1, d−q ]) there is a constant Cp,q such that for
24 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

f ∈ H 1,q (Ω):
1/p 1/q
(∫ ∣f − f¯∣p dλΩ ) ≤ Cp,q (∫ ∣∇f ∣q dλΩ ) ,
Ω Ω

where f¯ = ∫Ω f dλΩ .
In terms of these Sobolev-Poincaré constants we may also show a generalisation of
Proposition 6.1:
Proposition 6.3. Let (Ω, g) be a smooth, compact Riemannian manifold of dimension
∂ρ
d ≥ 3, with a connected boundary. For any ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is
Lipschitz continuous on Ω it holds ∀f ∈ C 1 (Ω) with ∫Ω f dλΩ = 0 and p ∈ [2, 2d−2
d−2 ]

2/p 2/p 2/p

p
⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ 2
⎞ 2
(∫ ∣f ∣ dλ∂Ω ) ≤ ( ∣∇ρ∣∞ p) C2(p−1),2 +( ∣(∆ρ)− ∣∞ ) Cp,2 ∫Ω ∣∇f ∣ dλΩ .
∂Ω ⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠
Proof. We may calculate as in the previous proofs to obtain
2/p
2/p ∣Ω∣ ∣Ω∣
(∫ ∣f ∣p dλ∂Ω ) ≤( ∣∇ρ∣∞ p ∫ ∣f ∣p−1 ∣∇f ∣dλΩ + ∣(∆ρ)− ∣∞ ∫ ∣f ∣p dλΩ )
∂Ω ∣∂Ω∣ Ω ∣∂Ω∣ Ω

1/2 1/2 2/p

∣Ω∣ ∣Ω∣
≤( ∣∇ρ∣∞ p (∫ ∣f ∣2(p−1) dλΩ ) (∫ ∣∇f ∣2 dλΩ ) + ∣(∆ρ)− ∣∞ ∫ ∣f ∣p dλΩ )
∣∂Ω∣ Ω Ω ∣∂Ω∣ Ω

1/2 (p−1) 1/2

∣Ω∣ 2 2
≤( ∣∇ρ∣∞ p (C2(p−1),2 (∫ ∣∇f ∣ dλΩ ) ) (∫ ∣∇f ∣ dλΩ )
∣∂Ω∣ Ω Ω

1/2 p 2/p
∣Ω∣
+ ∣(∆ρ)− ∣∞ (Cp,2 (∫ ∣∇f ∣2 dλΩ ) ))
∣∂Ω∣ Ω

(p−1) 2/p
⎛ ∣Ω∣ 1/2 1/2 ⎞
≤ ∣∇ρ∣∞ p (C2(p−1),2 (∫ ∣∇f ∣2 dλΩ ) ) 2
(∫ ∣∇f ∣ dλΩ )
⎝ ∣∂Ω∣ Ω Ω ⎠
2/p
∣Ω∣ 2 2
+( ∣(∆ρ)− ∣∞ ) Cp,2 ∫ ∣∇f ∣ dλΩ
∣∂Ω∣ Ω
2/p (p−1)/p 1/p
∣Ω∣ 2 2
=( ∣∇ρ∣∞ p) (C2(p−1),2 ∫ ∣∇f ∣ dλΩ ) (∫ ∣∇f ∣2 dλΩ )
∣∂Ω∣ Ω Ω
2/p
∣Ω∣ 2 2
+( ∣(∆ρ)− ∣∞ ) Cp,2 ∫ ∣∇f ∣ dλΩ
∣∂Ω∣ Ω
2/p 2/p
∣Ω∣ 2(p−1)/p 2 ∣Ω∣ 2 2
=( ∣∇ρ∣∞ p) C2(p−1),2 ∫ ∣∇f ∣ dλΩ + ( ∣(∆ρ)− ∣∞ ) Cp,2 ∫ ∣∇f ∣ dλΩ
∣∂Ω∣ Ω ∣∂Ω∣ Ω
2/p 2/p
⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ 2
⎞ 2
= ( ∣∇ρ∣∞ p) C2(p−1),2 +( ∣(∆ρ)− ∣∞ ) Cp,2 ∫Ω ∣∇f ∣ dλΩ .
⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠
Here we have used the Sobolev-Poincaré inequalities associated with C2(p−1),2 and Cp,2 .
Note therefor that for p ∈ [2, 2d−2 2d
d−2 ] it holds p, 2(p − 1) ∈ [1, d−2 ]. □

Again explicit constants may be obtained from Lemma 2.2 in terms of upper bounds
on sectional curvature and second fundamental form on the boundary.
 25

6.2. Boundary Trace Logarithmic Sobolev Inequalities. As point of departure we

recall the following lemma, cf. [42].
Lemma 6.1 (Rothaus’ Lemma). Let f ∶ ∂Ω → R be measurable and assume that
2 2
∫∂Ω f log(1 + f )dλ∂Ω < ∞. For every a ∈ R

Entλ∂Ω ((f + a)2 ) ≤ Entλ∂Ω (f 2 ) + 2 ∫ f 2 dλ∂Ω .

∂Ω

Lemma 6.2. If f ∈ C 1 (Ω) fulfills ∫Ω f dλΩ = 0 and if there are constants C̃p,2 such that
2/p
2d − 2
(12) (∫ ∣f ∣p dλ∂Ω ) ≤ C̃p,2 ∫ ∣∇f ∣2 dλΩ , ∀p ∈ [2, ],
∂Ω Ω d−2
then it holds
p C̃p,2
Entλ∂Ω (f 2 ) ≤ inf 2
∫Ω ∣∇f ∣ dλΩ .
2d−2
p∈[2, d−2 ] p − 2 e

The proof of this Lemma is adapted from [2, Proposition 6.2.3], see also [2, Proposition
5.1.8] for details.

Proof. Without loss of generality we may assume ∫∂Ω f 2 dλ∂Ω = 1 and define
r
ϕ ∶ (0, 1] → R, ϕ(r) ∶= log ((∫ ∣f ∣1/r dλ∂Ω ) ) .
∂Ω

ϕ is convex and ϕ′ ( 21 ) = −Entλ∂Ω (f 2 ). Now for p ∈ [2, 2d−2

d−2 via the convexity of ϕ
]
1 1 1/2 1 1 1
d (ϕ ( ) − ϕ ( )) = d ∫ ϕ′ (s)ds ≤ dϕ′ ( ) ( − )
2 p 1/p 2 2 p
2p 1 1
⇔ − Entλ∂Ω (f 2 ) ≥ (ϕ ( ) − ϕ ( ))
p−2 2 p
2/p
p
⇔Entλ∂Ω (f 2 ) ≤ log ((∫ ∣f ∣p dλ∂Ω ) ) .
p−2 ∂Ω

Inserting inequality (12) we obtain

p
Entλ∂Ω (f 2 ) ≤ log (C̃p,2 ∫ ∣∇f ∣2 dλΩ ) .
p−2 Ω

p
We define ϕ̃ ∶ (0, ∞) → R, ϕ̃(r) ∶= p−2 log(C̃p,2 r). ϕ̃ is concave and we may thus compute

Entλ∂Ω (f 2 ) ≤ ϕ̃ (∫ ∣∇f ∣2 dλΩ ) ≤ ϕ̃(r) + ϕ̃′ (r) (∫ ∣∇f ∣2 dλΩ − r)

Ω Ω

= ϕ̃′ (r) ∫ ∣∇f ∣2 dλΩ + (ϕ̃(r) − rϕ̃′ (r)) .

Ω
e
Choosing r = C̃p,2
the last term vanishes and we obtain

e p C̃p,2
Entλ∂Ω (f 2 ) ≤ ϕ̃′ ( ) ∫ ∣∇f ∣2 dλΩ = ∣∇f ∣2 dλΩ .
C̃p,2 Ω p − 2 e ∫Ω
□
26 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

∂ρ
Proposition 6.4. Assume that d ≥ 3. For any ρ ∈ C 1 (Ω) such that ∂N ∣∂Ω = −1 and ∇ρ is
Lipschitz continuous on Ω inequality (10) in Proposition 4.1 is fulfilled with

2/p 2/p
p 1 ⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ 2
⎞
L∂Ω,Ω = inf ( ∣∇ρ∣∞ p) C2(p−1),2 +( ∣(∆ρ)− ∣∞ ) Cp,2
p∈[2, 2d−2
d−2
] p − 2 e ⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠
1/2 ∣Ω∣
+ 2 {2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ } ,
∣∂Ω∣

where (⋅)− denotes the negative part of a function.

Proof. Let f ∈ C 1 (Ω) then for a ∶= ∫Ω f dλΩ we define f˜ ∶= f − a and by Lemma 6.1 it holds

Entλ∂Ω (f 2 ) = Entλ∂Ω ((f˜ + a)2 ) ≤ Entλ∂Ω (f˜2 ) + 2 ∫ f˜2 dλ∂Ω .

∂Ω

As f˜ is centered on Ω the assumptions of Lemma 6.2 are fulfilled due to Proposition 6.3
and we obtain
2/p 2/p
p 1 ⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ ⎞
Entλ∂Ω (f˜2 ) ≤ inf ( ∣∇ρ∣∞ p) C2(p−1),2 +( ∣(∆ρ)− ∣∞ ) 2
Cp,2
p∈[2, 2d−2
d−2
] p − 2 e ⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠

⋅ ∫ ∣∇f˜∣2 dλΩ .
Ω

Furthermore by Proposition 6.1

1/2 ∣Ω∣
∫∂Ω f˜ dλ∂Ω ≤ {2∣∇ρ∣∞ CΩ + ∣(∆ρ) ∣∞ CΩ } ∣∂Ω∣ ∫Ω ∣∇f˜∣ dλΩ .
2 − 2

Thus we have
2/p 2/p
⎛ p 1 ⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ ⎞
Entλ∂Ω (f 2 ) ≤ inf ( ∣∇ρ∣∞ p) C2(p−1),2 + ( ∣(∆ρ)− ∣∞ ) 2
Cp,2
⎝ p∈[2, 2d−2
d−2
] p − 2 e ⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠

1/2 ∣Ω∣ ⎞
+2 {2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ } ∫ ∣∇f˜∣2 dλΩ
∣∂Ω∣ ⎠ Ω
2/p 2/p
⎛ p 1 ⎛ ∣Ω∣ 2(p−1)/p ∣Ω∣ 2
⎞
= inf ( ∣∇ρ∣∞ p) C2(p−1),2 +( ∣(∆ρ)− ∣∞ ) Cp,2
⎝ p∈[2, 2d−2
d−2
] p − 2 e ⎝ ∣∂Ω∣ ∣∂Ω∣ ⎠

1/2 ∣Ω∣ ⎞
+2 {2∣∇ρ∣∞ CΩ + ∣(∆ρ)− ∣∞ CΩ } ∣∇f ∣2 dλΩ .
∣∂Ω∣ ⎠ ∫Ω

Combining this with Lemma 2.2 again results in an explicit upper bound for L∂Ω,Ω .
 27

Proposition 6.5. Assume that d ≥ 3. Let k2 ∈ R such that sect ≤ k2 and γ2 ∈ R such that
Π ≤ γ2 id. Then inequality (10) in Proposition 4.1 is fulfilled with
2/p
p 1 ∣Ω∣ 2(p−1)/p
L∂Ω,Ω = inf 2d−2 p − 2 e
[( p) C2(p−1),2
p∈[2, d−2 ] ∣∂Ω∣
2/p
∣Ω∣ h′2 t 1 − 2
+( inf sup ((d − 1) (t) (1 − ) − ) ) Cp,2 ]
∣∂Ω∣ t1 ∈(0,h−1
2 (0)) t∈(0,t1 ) h2 t1 t1
1/2 h′2 t 1 − ∣Ω∣
+ 2 {2CΩ + inf−1 sup ((d − 1) (t) (1 − ) − ) CΩ } .
t1 ∈(0,h2 (0)) t∈(0,t1 ) h2 t1 t1 ∣∂Ω∣
6.3. Semigroup Approach. Lastly we present an alternative approach via semigroup
theory to bound L∂Ω,Ω and Lα :
Let PtN be the Neumann semigroup on Ω. Then
d
(13) ∣PtN ∣1→∞ ≤ 1 + ct− 2 , t > 0
holds for some constant c > 0, where ∣ ⋅ ∣1→∞ is the operator norm from L1 (λΩ ) to L∞ (λΩ ).
Next, let D be the set of positive functions 1 ≤ ϕ ∈ Cb2 (Ω) such that
∂log(ϕ)
∣ . Π≥
∂N ∂M
See Lemma 3.5.8 in [50] for concrete examples of ϕ. Let
Kϕ ∶= inf {Ric(X, X) − (ϕ∆ϕ−1 )(x) ∶ x ∈ Ω, X ∈ Tx M, ∣X∣ = 1}.
Let L′∂Ω,Ω be the smallest constant such that ∀f ∈ C 1 (Ω)
2 2 2 2
∫∂Ω (f log(f ) + 1 − f ) dλ∂Ω ≤ L∂Ω,Ω ∫Ω ∣∇f ∣ dλΩ .
′

Proposition 6.6. (1) We have L∂Ω,Ω ≤ L′∂Ω,Ω and

αLΩ + (1 − α)L′∂Ω,Ω
Lα ≤ .
α
(2) Let ϕ ∈ D and ρ ∈ Cb2 (Ω) with ∂ρ
∣
∂N ∂M = 1. If Kϕ > 0, then
1/2
∣∂Ω∣ ′ √ ∞
L∂Ω,Ω ≤ LΩ ∣(∆ρ)+ ∣∞ + 32∣∇ρ∣∞ ∣ϕ∣3∞ ( ∫ e−2Kϕ t (1 + ∣log(∣PtN ∣1→∞ )∣)dt) .
∣Ω∣ 0

In particular, when Ric ≥ K for some constant K > 0 and Π ≥ 0, the estimate
holds for ∣ϕ∣∞ = 1 and Kϕ = K.
(3) In general,
∣∂Ω∣ ′
L∂Ω,Ω ≤ LΩ ∣(∆ρ)+ ∣∞
∣Ω∣
1/2
√ 1 ∞
+ 32∣∇ρ∣∞ ∣ϕ∣3∞ ( ∫ e −2Kϕ t
(1 + ∣log(∣PtN ∣1→∞ )∣)dt + ∫ e e
−2Kϕ −(t−1)/CΩ N
∣P1 ∣1→∞ dt) .
0 1

Proof. (1) By Remark 1.20 in [17], we have

(14) Entλ∂Ω (f 2 ) = inf ∫ (f 2 log(f 2 ) − f 2 log(t) + t − f 2 ) dλ∂Ω .

t>0 ∂Ω
28 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

By taking t = 1 we obtain

Entλ∂Ω (f 2 ) ≤ ∫ (f 2 log(f 2 ) + 1 − f 2 ) dλ∂Ω ≤ L′∂Ω,Ω ∫ ∣∇f ∣2 dλΩ ,

∂Ω Ω

so that L∂Ω,Ω ≤ L′∂Ω,Ω .

Applying (14) to λα replacing λ∂Ω , and assuming that t = ∫Ω f 2 dλΩ = 1, we obtain

Entλα (f 2 ) ≤ αEntλΩ (f 2 ) + (1 − α) ∫ (f 2 log(f 2 ) + 1 − f 2 ) dλ∂Ω

∂Ω

≤ (αLΩ + (1 − α)L′∂Ω,Ω ) ∫ ∣∇f ∣2 dλΩ .

Ω

(2) Letting ∫Ω f 2 dλΩ = 1 and noting that

f 2 log(f 2 ) + 1 − f 2 ≥ 0,

and using integration by parts, we obtain

∣∂Ω∣
(f 2 log(f 2 ) + 1 − f 2 ) dλ∂Ω = ∫ (f 2 log(f 2 ) + 1 − f 2 )∆ρ + log(f 2 )∇f 2 ⋅ ∇ρdλΩ
∣Ω∣ ∫∂Ω Ω
1/2
2 2 2 2 2
(15) ≤ ∣(∆ρ) ∣∞ EntλΩ (f ) + 2∣∇ρ∣∞ (∫ ∣∇f ∣ dλΩ ∫ f (log(f )) dλΩ )
+
.
Ω Ω

Moreover, since ∣PtN f 2 − ∫Ω f 2 dλΩ ∣2 → 0 as t → ∞, we have

2 2 2
∞d
∫Ω f (log(f )) dλΩ = − ∫0 [ (P N f 2 )(log(PtN f 2 ))2 dλΩ ]dt
dt ∫Ω t
∞ ∣∇PtN f 2 ∣2
(16) = 2∫ ∫ N 2
(1 + log(PtN f 2 ))dλΩ dt.
0 Ω P t f
By Theorem 3.6.1(2) in [50], for the reflecting diffusion process Xt on Ω generated by ∆,
t
and a martingale Mt with quadratic variation ⟨M ⟩t = ∫0 ∣∇ log ϕ(Xs )∣2 ds, we have
√ 2
∣∇PtN f 2 ∣2 ≤ 4∣ϕ∣2∞ (E[∣f ∇f ∣(Xt )e−Kϕ t+ 2Mt −2⟨M ⟩t
])
√
≤ 4∣ϕ∣2∞ e−2Kϕ t (PtN f 2 )E[∣∇f ∣2 (Xt )e2 2Mt −4⟨M ⟩t
] = 4∣ϕ∣2∞ e−2Kϕ t (PtN f 2 )P̄tN ∣∇f ∣2 ,

where, by Girsanov’s theorem, P̄tN is the Neumann semigroup on Ω generated by

∆ + 4∇ log ϕ,

which is symmetric in L2 (ϕ4 λΩ ) so that

N 2 N 2 4 2 4 4 2
∫Ω P̄t ∣∇f ∣ dλΩ ≤ ∫Ω P̄t ∣∇f ∣ ϕ dλΩ = ∫Ω ∣∇f ∣ ϕ dλΩ ≤ ∣ϕ∣∞ ∫Ω ∣∇f ∣ dλΩ .

Thus,
∣∇PtN f 2 ∣2 N 2 6 −2K t N 2
(17) ∫Ω P N f 2 (1 + log(Pt f ))dλΩ ≤ 4∣ϕ∣∞ e ϕ (1 + ∣log(∣Pt ∣1→∞ )∣) ∫Ω ∣∇f ∣ dλΩ .
t

Combining this with (15) and (16) we obtain the desired estimate. When Π ≥ 0 and
Ric ≥ K > 0, we may take ϕ ≡ 1 so that the estimate holds for ∣ϕ∣∞ = 1 and Kϕ = K.
 29

(3) In general, by the semigroup property and Poincaré inequality, when t ≥ 1 we have
N 22 N 22
∫Ω ∣∇Pt f ∣ dλΩ ≤ e ∫Ω ∣∇P1 f ∣ dλΩ
−(t−1)/CΩ

≤ 4∣ϕ∣2∞ e−2Kϕ e−(t−1)/CΩ ∫ (P̄1N ∣∇f ∣2 )(P1N f 2 )dλΩ

Ω

≤ 4∣ϕ∣6∞ e−2Kϕ e−(t−1)/CΩ ∣P1N ∣1→∞ ∫ ∣∇f ∣2 dλΩ .

Ω
1+logr
Combining this with (17) and the fact that r ≤ 1 for r > 0, we obtain
∣∇PtN f 2 ∣2 N 2 6 2 e−2Kϕ e−(t−1)/CΩ ∣P1N ∣1→∞ , t ≥ 1
∫Ω (1+log(P t f ))dλΩ ≤ 4∣ϕ∣ ∫ ∣∇f ∣ dλΩ {
PtN f 2 ∞
Ω e−2Kϕ t (1 + log(∣PtN ∣1→∞ )), t ∈ (0, 1).
Therefore, the desired estimate follows from (15) and (16). □

References
[1] A. Aurell and B. Djehiche. Behavior near walls in the mean-field approach to crowd dynamics. SIAM
J. Appl. Math., 80(3):1153–1174, 2020.
[2] D. Bakry, I. Gentil, and M. Ledoux. Analysis and geometry of Markov diffusion operators, volume
348 of Grundlehren Math. Wiss. Cham: Springer, 2014.
[3] L. Barbu, A. Burlacu, and G. Moroşanu. An eigenvalue problem involving the (p, q)-Laplacian with
a parametric boundary condition. Mediterr. J. Math., 20(4):18, 2023. Id/No 232.
[4] W. Beckner. Sharp Sobolev inequalities on the sphere and the Moser-Trudinger inequality. Ann.
Math. (2), 138(1):213–242, 1993.
[5] P. Binding, R. Hryniv, H. Langer, and B. Najman. Elliptic eigenvalue problems with eigenparameter
dependent boundary conditions. J. Differential Equations, 174(1):30–54, 2001.
[6] J. F. Bonder, J. D. Rossi, and R. Ferreira. Uniform bounds for the best Sobolev trace constant. Adv.
Nonlinear Stud., 3(2):181–192, 2003.
[7] J.-M. Bony, P. Courrège, and P. Priouret. Semi-groupes de Feller sur une variété à bord compacte et
problèmes aux limites intégro-différentiels du second ordre donnant lieu au principe du maximum.
Ann. Inst. Fourier (Grenoble), 18(fasc. 2):369–521 (1969), 1968.
[8] N. Bou-Rabee and M. C. Holmes-Cerfon. Sticky Brownian motion and its numerical solution. SIAM
Rev., 62(1):164–195, 2020.
[9] D. Buoso, C. Falcó, M. d. M. González, and M. Miranda. Bulk-boundary eigenvalues for bilaplacian
problems. Discrete Contin. Dyn. Syst., 43(3-4):1175–1200, 2023.
[10] J.-B. Casteras, L. Monsaingeon, and F. Santambrogio. Sticky diffusion as a Wasserstein gradient
flow. 2024. in preparation.
[11] D. Chafaı̈ and F. Malrieu. On fine properties of mixtures with respect to concentration of measure
and Sobolev type inequalities. Ann. Inst. Henri Poincaré Probab. Stat., 46(1):72–96, 2010.
[12] I. Chavel. Eigenvalues in Riemannian geometry. With a chapter by Burton Randol. With an appendix
by Jozef Dodziuk, volume 115 of Pure Appl. Math., Academic Press. Academic Press, New York, NY,
1984.
[13] I. Chavel. Riemannian Geometry: A Modern Introduction. Cambridge University Press, 1995.
[14] M.-F. Chen and F.-Y. Wang. Estimates of logarithmic Sobolev constant: an improvement of Bakry-
Emery criterion. J. Funct. Anal., 144(2):287–300, 1997.
[15] B. Colbois, A. Girouard, C. Gordon, and D. Sher. Some recent developments on the steklov eigenvalue
problem. Revista Matemática Complutense, pages 1–161, 09 2023.
[16] J.-D. Deuschel, G. Giacomin, and L. Zambotti. Scaling limits of equilibrium wetting models in
(1 + 1)-dimension. Probab. Theory Related Fields, 132(4):471–500, 2005.
[17] J.-D. Deuschel and D. W. Stroock. Hypercontractivity and spectral gap of symmetric diffusions with
applications to the stochastic Ising models. J. Funct. Anal., 92(1):30–48, 1990.
[18] K.-J. Engel. The Laplacian on C(Ω) with generalized Wentzell boundary conditions. Arch. Math.
(Basel), 81(5):548–558, 2003.
[19] H.-J. Engelbert and G. Peskir. Stochastic differential equations for sticky Brownian motion. Stochas-
tics, 86(6):993–1021, 2014.
30 MARIE BORMANN, MAX VON RENESSE, AND FENG-YU WANG

[20] J. F. Escobar. The geometry of the first non-zero Stekloff eigenvalue. J. Funct. Anal., 150(2):544–556,
1997.
[21] T. Fattler, M. Grothaus, and R. Voßhall. Construction and analysis of a sticky reflected distorted
Brownian motion. Ann. Inst. Henri Poincaré Probab. Stat., 52(2):735–762, 2016.
[22] G. François. Spectral asymptotics stemming from parabolic equations under dynamical boundary
conditions. Asymptot. Anal., 46(1):43–52, 2006.
[23] G. R. Goldstein. Derivation and physical interpretation of general boundary conditions. Adv. Dif-
ferential Equations, 11(4):457–480, 2006.
[24] G. R. Goldstein, J. A. Goldstein, D. Guidetti, and S. Romanelli. Maximal regularity, analytic semi-
groups, and dynamic and general Wentzell boundary conditions with a diffusion term on the bound-
ary. Ann. Mat. Pura Appl. (4), 199(1):127–146, 2020.
[25] M. Grothaus and R. Voßhall. Stochastic differential equations with sticky reflection and boundary
diffusion. Electron. J. Probab., 22:Paper No. 7, 37, 2017.
[26] N. Ikeda. On the construction of two-dimensional diffusion processes satisfying Wentzell’s boundary
conditions and its application to boundary value problems. Mem. Coll. Sci. Univ. Kyoto Ser. A.
Math., 33:367–427, 1960/61.
[27] A. Kasue. Applications of Laplacian and Hessian comparison theorems. In Geometry of geodesics
and related topics (Tokyo, 1982), volume 3 of Adv. Stud. Pure Math., pages 333–386. North-Holland,
Amsterdam, 1984.
[28] J. Kennedy. An isoperimetric inequality for the second eigenvalue of the Laplacian with Robin
boundary conditions. Proc. Amer. Math. Soc., 137(2):627–633, 2009.
[29] A. V. Kolesnikov and E. Milman. Brascamp-Lieb-type inequalities on weighted Riemannian mani-
folds with boundary. J. Geom. Anal., 27(2):1680–1702, 2017.
[30] V. Konarovskyi, V. Marx, and M. von Renesse. Spectral gap estimates for brownian motion on
domains with sticky-reflecting boundary diffusion. 2021. arxiv: 2106.00080.
[31] V. Konarovskyi and M. von Renesse. Reversible coalescing-fragmentating Wasserstein dynamics on
the real line, 2017. arXiv:1709.02839,.
[32] N. Kuznetsov and A. Nazarov. Sharp constants in the Poincaré, Steklov and related inequalities (a
survey). Mathematika, 61(2):328–344, 2015.
[33] Y. Li and M. Zhu. Sharp Sobolev trace inequalities on Riemannian manifolds with boundaries.
Comm. Pure Appl. Math., 50(5):449–487, 1997.
[34] F. Maggi and C. Villani. Balls have the worst best Sobolev inequalities. J. Geom. Anal., 15(1):83–121,
2005.
[35] F. Maggi and C. Villani. Balls have the worst best Sobolev inequalities. II: Variants and extensions.
Calc. Var. Partial Differ. Equ., 31(1):47–74, 2008.
[36] S. Matculevich and S. Repin. Explicit constants in Poincaré-type inequalities for simplicial domains
and application to a posteriori estimates. Comput. Methods Appl. Math., 16(2):277–298, 2016.
[37] G. Menz and A. Schlichting. Poincaré and logarithmic Sobolev inequalities by decomposition of the
energy landscape. Ann. Probab., 42(5):1809–1884, 2014.
[38] D. Mugnolo and S. Romanelli. Dirichlet forms for general Wentzell boundary conditions, analytic
semigroups, and cosine operator functions. Electron. J. Differential Equations, pages No. 118, 20,
2006.
[39] B. Nazaret. Best constant in Sobolev trace inequalities on the half-space. Nonlinear Anal., Theory
Methods Appl., Ser. A, Theory Methods, 65(10):1977–1985, 2006.
[40] A. Nonnenmacher and M. Grothaus. Overdamped limit of generalized stochastic hamiltonian systems
for singular interaction potentials, 2018.
[41] R. C. Reilly. Applications of the Hessian operator in a Riemannian manifold. Indiana Univ. Math.
J., 26(3):459–472, 1977.
[42] O. S. Rothaus. Analytic inequalities, isoperimetric inequalities and logarithmic Sobolev inequalities.
J. Funct. Anal., 64(2):296–313, 1985.
[43] A. Schlichting. Poincaré and log-sobolev inequalities for mixtures. Entropy, 21(1), 2019.
[44] A. Shouman. Generalization of Philippin’s results for the first Robin eigenvalue and estimates for
eigenvalues of the bi-drifting Laplacian. Ann. Global Anal. Geom., 55(4):805–817, 2019.
[45] K. Taira. Boundary value problems and Markov processes, volume 1499 of Lecture Notes in Math-
ematics. Springer, Cham, third edition, [2020] ©2020. Functional analysis methods for Markov
processes.
 31

[46] S. Takanobu and S. Watanabe. On the existence and uniqueness of diffusion processes with Wentzell’s
boundary conditions. J. Math. Kyoto Univ., 28(1):71–80, 1988.
[47] A. D. Ventcel. On boundary conditions for multi-dimensional diffusion processes. Theor. Probability
Appl., 4:164–177, 1959.
[48] J. von Below and G. François. Spectral asymptotics for the Laplacian under an eigenvalue dependent
boundary condition. Bull. Belg. Math. Soc. Simon Stevin, 12(4):505–519, 2005.
[49] F.-Y. Wang. Estimates of the first Neumann eigenvalue and the log-Sobolev constant on non-convex
manifolds. Math. Nachr., 280(12):1431–1439, 2007.
[50] F.-Y. Wang. Analysis for diffusion processes on Riemannian manifolds, volume 18 of Advanced Series
on Statistical Science & Applied Probability. World Scientific Publishing Co. Pte. Ltd., Hackensack,
NJ, 2014.
[51] S. Watanabe. On stochastic differential equations for multi-dimensional diffusion processes with
boundary conditions. II. J. Math. Kyoto Univ., 11:545–551, 1971.

Universität Leipzig, Fakultät für Mathematik und Informatik, Augustusplatz 10, 04109
Leipzig, Germany and Max Planck Institute for Mathematics in the Sciences, 04103
Leipzig, Germany
Email address: [email protected]

Universität Leipzig, Fakultät für Mathematik und Informatik, Augustusplatz 10, 04109
Leipzig, Germany
Email address: [email protected]

Center for Applied Mathematics, Tianjin University, Tianjin 300072, China

Email address: [email protected]