Relations between functional norms of a non-negative function and its square root on the positive cone of
Besov and Triebel-Lizorkin spaces
Lubomir T. Dechevsky 1 and Niklas Grip 2
Outline
We will derive direct estimates of functionals depending on f by respective functionals depending on √
f and inverse estimates, where the role of f and √
f is reversed.
Organization of the talk
1 Relations between the inner-product wavelet coefficients of f and √
f
Sobolev norms, variable metric index, fixed integer smoothness index
We will derive direct estimates of functionals depending on f by respective functionals depending on √
f and inverse estimates, where the role of f and √
f is reversed.
Organization of the talk
1 Relations between the inner-product wavelet coefficients of f and √ f
2 Sobolev norms, variable metric index, fixed integer smoothness index
3 Besov and Triebel-Lizorkin spaces, fixed metric index, variable (fractional) smoothness index Direct results
Inverse results
4 Interpolation results
Outline
We will derive direct estimates of functionals depending on f by respective functionals depending on √
f and inverse estimates, where the role of f and √
f is reversed.
Organization of the talk
1 Relations between the inner-product wavelet coefficients of f and √
f
Sobolev norms, variable metric index, fixed integer smoothness index
We will derive direct estimates of functionals depending on f by respective functionals depending on √
f and inverse estimates, where the role of f and √
f is reversed.
Organization of the talk
1 Relations between the inner-product wavelet coefficients of f and √ f
2 Sobolev norms, variable metric index, fixed integer smoothness index
3 Besov and Triebel-Lizorkin spaces, fixed metric index, variable (fractional) smoothness index Direct results
Inverse results
4 Interpolation results
Relations between inner-product wavelet coeffs of f and√ f
Relations between the inner-product wavelet coefficients of f and √
f
Background: Different statistical wavelet approximations [PD,PV] of a distribution density f (defined on R), that preserve the crucial properties
f (x) ≥ 0 and Z
R
f (x) dx = 1.
Relations between the inner-product wavelet coefficients of f and √
f
Background: Different statistical wavelet approximations [PD,PV] of a distribution density f (defined on R), that preserve the crucial properties
f (x) ≥ 0 and Z
R
f (x) dx = 1.
It was also proved in [PD] that this approximation achieves certain assymptotycally minimax rates. However, these rates were in terms of functional norms depending on the wavelet coefficients of √
f rather than the ones of f .
[PD] S. Penev, and L. Dechevsky, J. Nonparametr. Statist. 7, 365–394 (1997), ISSN 1048-5252.
[PV] A. Pinheiro, and B. Vidakovic, Comput. Statist. 25, 399–415 (1997), ISSN 0943-4062.
Relations between inner-product wavelet coeffs of f and√ f
Assumptions and notation:
ϕ is real-valued, compactly supported and bounded, ϕ j ,k (x) = 2
j2ϕ(2 j x − k), x ∈ R, j ∈ Z, k ∈ Z.
No assumptions about (bi)orthogonality of the wavelet basis, thus
allowing also for non-biorthogonal wavelet constructions, such as those
Assumptions and notation:
ϕ is real-valued, compactly supported and bounded, ϕ j ,k (x) = 2
j2ϕ(2 j x − k), x ∈ R, j ∈ Z, k ∈ Z.
No assumptions about (bi)orthogonality of the wavelet basis, thus allowing also for non-biorthogonal wavelet constructions, such as those considered, e.g. in [DPa,DPb].
[DPa] L. T. Dechevsky, and S. I. Penev, Stochastic Anal. Appl. 15, 187–215 (1997), ISSN 0736-2994.
[DPb] L. T. Dechevsky, and S. I. Penev, Stochastic Anal. Appl. 16, 423–462 (1998), ISSN 0736-2994.
.
Relations between inner-product wavelet coeffs of f and√ f
Assumptions and notation:
ϕ is real-valued, compactly supported and bounded, ϕ j ,k (x) = 2
j2ϕ(2 j x − k), x ∈ R, j ∈ Z, k ∈ Z.
No assumptions about (bi)orthogonality of the wavelet basis, thus
allowing also for non-biorthogonal wavelet constructions, such as those
Assumptions and notation:
ϕ is real-valued, compactly supported and bounded, ϕ j ,k (x) = 2
j2ϕ(2 j x − k), x ∈ R, j ∈ Z, k ∈ Z.
No assumptions about (bi)orthogonality of the wavelet basis, thus allowing also for non-biorthogonal wavelet constructions, such as those considered, e.g. in [DPa,DPb].
[DPa] L. T. Dechevsky, and S. I. Penev, Stochastic Anal. Appl. 15, 187–215 (1997), ISSN 0736-2994.
[DPb] L. T. Dechevsky, and S. I. Penev, Stochastic Anal. Appl. 16, 423–462 (1998), ISSN 0736-2994.
.
Relations between inner-product wavelet coeffs of f and√ f
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on
kf k ∞ and d.
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on kf k ∞ and d.
Proposition (Inverse results) Under the assumptions (1),
D √ f , ϕ j ,k
E
≤ c 6 2 −
j4|hf , ϕ j ,k i|
12where 0 < c 6 < ∞ and c 6 depends on kf k ∞ , d, kϕk ∞ and on |supp ϕ|.
If, additionally, it is also known that f | supp ϕ
j ,k≥ c f > 0 (c f = c f (j, k)), then
D √ f , ϕ j ,k
E
≤ c 7 ≤ |hf , ϕ j ,k i| holds, where 0 < c 7 < ∞ and c 7
depends on c f .
Relations between inner-product wavelet coeffs of f and√ f
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on
kf k ∞ and d.
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on kf k ∞ and d.
Proposition (Inverse results) Under the assumptions (1),
D √ f , ϕ j ,k
E
≤ c 6 2 −
j4|hf , ϕ j ,k i|
12where 0 < c 6 < ∞ and c 6 depends on kf k ∞ , d, kϕk ∞ and on |supp ϕ|.
If, additionally, it is also known that f | supp ϕ
j ,k≥ c f > 0 (c f = c f (j, k)), then
D √ f , ϕ j ,k
E
≤ c 7 ≤ |hf , ϕ j ,k i| holds, where 0 < c 7 < ∞ and c 7
depends on c f .
Relations between inner-product wavelet coeffs of f and√ f
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on
kf k ∞ and d.
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on kf k ∞ and d.
Proposition (Inverse results) Under the assumptions (1),
D √ f , ϕ j ,k
E
≤ c 6 2 −
j4|hf , ϕ j ,k i|
12where 0 < c 6 < ∞ and c 6 depends on kf k ∞ , d, kϕk ∞ and on |supp ϕ|.
If, additionally, it is also known that f | supp ϕ
j ,k≥ c f > 0 (c f = c f (j, k)), then
D √ f , ϕ j ,k
E
≤ c 7 ≤ |hf , ϕ j ,k i| holds, where 0 < c 7 < ∞ and c 7
depends on c f .
Relations between inner-product wavelet coeffs of f and√ f
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on
kf k ∞ and d.
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on kf k ∞ and d.
Proposition (Inverse results) Under the assumptions (1),
D √ f , ϕ j ,k
E
≤ c 6 2 −
j4|hf , ϕ j ,k i|
12where 0 < c 6 < ∞ and c 6 depends on kf k ∞ , d, kϕk ∞ and on |supp ϕ|.
If, additionally, it is also known that f | supp ϕ
j ,k≥ c f > 0 (c f = c f (j, k)), then
D √ f , ϕ j ,k
E
≤ c 7 ≤ |hf , ϕ j ,k i| holds, where 0 < c 7 < ∞ and c 7
depends on c f .
Relations between inner-product wavelet coeffs of f and√ f
Proposition (Direct results) Assume that
f ∈ L 1 (R) ∩ L ∞ ((R))
|hf , ϕ j ,k i| ≥ d or |hf , ϕ j ,k i| = 0 (i.e. hard thresholded coeffs) (1) Then |hf , ϕ j ,k i| ≤ c 5
D √
f , ϕ j,k E
, where 0 < c 5 < ∞, c 5 depends on
kf k ∞ and d.
Sobolev norms with variable metric indices and fixed constant integer smoothness index
Definition
(Inhomogeneous) Sobolev spaces with metric index 0 < p ≤ ∞ and smoothness index m = 0, 1, 2, . . . can be defined via their (quasi)norms
kf k W
p0def = kf k L
p, 0 < p ≤ ∞,
and
kf k W
pmdef = kf k L
p+ f
(m) L
p
, 0 < p ≤ ∞, m = 1, 2, . . .
Variable metric index, fixed integer smoothness index
Sobolev norms with variable metric indices and fixed constant integer smoothness index
Definition
(Inhomogeneous) Sobolev spaces with metric index 0 < p ≤ ∞ and smoothness index m = 0, 1, 2, . . . can be defined via their (quasi)norms
def
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
√ f p =
√ f
1/2 p/2 . If m = 1, 2, . . ., assume additionally that √ 1
f ∈ W q m −1 for p ≤ q ≤ ∞.
Then
√ f
W
pm≤ c(p, m)
√ 1 f
w
qm−1· kf k W
rmwith 1 r = 1
p − 1
q .
Variable metric index, fixed integer smoothness index
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
√ f p =
√ f
1/2 p/2 . If m = 1, 2, . . ., assume additionally that √ 1
f ∈ W q m −1 for p ≤ q ≤ ∞.
Then
√ f
W
pm≤ c(p, m)
√ 1 f
w
qm−1· kf k W
rmwith 1 r = 1
p − 1
q .
Variable metric index, fixed integer smoothness index
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
√ f p =
√ f
1/2 p/2 . If m = 1, 2, . . ., assume additionally that √ 1
f ∈ W q m −1 for p ≤ q ≤ ∞.
Then
√ f
W
pm≤ c(p, m)
√ 1 f
w
qm−1· kf k W
rmwith 1 r = 1
p − 1
q .
Variable metric index, fixed integer smoothness index
Theorem (Direct results)
For 0 < p ≤ ∞ and m = 0, 1, 2, . . . kf k W
pm≤ c 2 (p, m)
√ f
2 W
2pm. Theorem (Inverse results)
Let 0 < p ≤ ∞. If m = 0, then the above result is also inverse:
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Direct results
Using pointwise multipliers to preserve the metric, we have derived four theorems of the kind
“ √
f ∈ B p,q s (or F p,q s ) + additional assumptions
⇒ f ∈ B p,q s (or F p,q s )”.
Theorem
Let −∞ < s < ∞, 0 < p ≤ ∞, 0 < q ≤ ∞ and σ > max n
s, min{p,1} 1 − 1 − s o . If √
f ∈ B ∞,∞ σ ∩ B p,q s (which is possible only for a density with bounded support), then
f ∈ B p,q s and kf k B
p,qs≤ c
√ f B
σ∞,∞
√ f
B
p,qs.
Fixed metric index, variable (fractional) smoothness index Direct results
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Direct results
Using pointwise multipliers to preserve the metric, we have derived four theorems of the kind
“ √
f ∈ B p,q s (or F p,q s ) + additional assumptions
⇒ f ∈ B p,q s (or F p,q s )”.
Theorem
Let −∞ < s < ∞, 0 < p < ∞, 0 < q < ∞ and σ > max
s, 1
min { p,q,1 } − 1 − s
. If √
f ∈ B ∞,∞ σ ∩ F p,q s , , then
f ∈ F p,q s and kf k F
p,qs≤ c
√ f B
∞,∞σ√ f
F
p,qs.
Underlining denotes what is different from the previous theorem.
Fixed metric index, variable (fractional) smoothness index Direct results
Theorem
Let −∞ < s < ∞, 0 < p < ∞, 0 < q < ∞ and σ > max
s, 1
min { p,q,1 } − 1 − s
. If √
f ∈ B ∞,∞ σ ∩ F p,q s , , then
√ √
Theorem
Let −∞ < s < ∞, 0 < p < ∞, 0 < q < ∞ and σ > max
s, 1
min { p,q,1 } − 1 − s
. If √
f ∈ B ∞,∞ σ ∩ F p,q s , , then
f ∈ F p,q s and kf k F
p,qs≤ c
√ f B
∞,∞σ√ f
F
p,qs.
Underlining denotes what is different from the previous theorem.
Fixed metric index, variable (fractional) smoothness index Direct results
Theorem
Let −∞ < s < ∞, 0 < p < ∞, 0 < q < ∞ and σ > max
s, 1
min { p,q,1 } − 1 − s
. If √
f ∈ B ∞,∞ σ ∩ F p,q s , , then
√ √
The next direct result is related to the continuity of f and √
f in view of the range of s and p for which Sobolev-type embeddings hold true.
Theorem
Let 0 < p ≤ ∞ and either 1 < q ≤ ∞, s > 1/p or 0 < q ≤ 1, s ≥ 1/p. If
√ f ∈ B p,q s , then f ∈ B p,q s and kf k B
p,qs≤ c
√ f
2 B
p,qs. Theorem
Let either 0 < p < q ≤ ∞, s > 1 p or 0 < p ≤ q < ∞, s > 1 2
1 p + 1 q
. If √
f ∈ F p,q s , then f ∈ F p,q s and kf k F
p,qs≤ c
√ f
2 F
p,qs.
Here, again, we point out with underlining what is different from the
previous theorem.
Fixed metric index, variable (fractional) smoothness index Direct results
The next direct result is related to the continuity of f and √
f in view of the range of s and p for which Sobolev-type embeddings hold true.
Theorem
Let 0 < p ≤ ∞ and either 1 < q ≤ ∞, s > 1/p or 0 < q ≤ 1, s ≥ 1/p. If
√ f ∈ B p,q s , then f ∈ B p,q s and kf k B
p,qs≤ c
√ f
2
B
p,qs.
The next direct result is related to the continuity of f and √
f in view of the range of s and p for which Sobolev-type embeddings hold true.
Theorem
Let 0 < p ≤ ∞ and either 1 < q ≤ ∞, s > 1/p or 0 < q ≤ 1, s ≥ 1/p. If
√ f ∈ B p,q s , then f ∈ B p,q s and kf k B
p,qs≤ c
√ f
2 B
p,qs. Theorem
Let either 0 < p < q ≤ ∞, s > 1 p or 0 < p ≤ q < ∞, s > 1 2
1 p + 1 q
. If √
f ∈ F p,q s , then f ∈ F p,q s and kf k F
p,qs≤ c
√ f
2 F
p,qs.
Here, again, we point out with underlining what is different from the
previous theorem.
Fixed metric index, variable (fractional) smoothness index Direct results
The next direct result is related to the continuity of f and √
f in view of the range of s and p for which Sobolev-type embeddings hold true.
Theorem
Let 0 < p ≤ ∞ and either 1 < q ≤ ∞, s > 1/p or 0 < q ≤ 1, s ≥ 1/p. If
√ f ∈ B p,q s , then f ∈ B p,q s and kf k B
p,qs≤ c
√ f
2
B
p,qs.
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Inverse results
Using modifications of the ideas behind the proofs of the direct results, we obtained the following four corresponding inverse results of the kind
“ f ∈ B p,q s (or F p,q s ) + additional assumptions ⇒ √
f ∈ B p,q s (or F p,q s )”.
Note, however, that the inverse estimates depend essentially on the behaviour of √ 1
f : namely, in which Besov and Triebel-Lizorkin spaces it is a pointwise mutliplier. Thus we also get conditions on derivatives of the function 1/pf (x), which here is understood to vanish when f (x) = 0.
Theorem
Let 1 ≤ p ≤ ∞, 0 < q ≤ ∞, m ∈ N 0
def = N ∪ {0}, s > m. If
√ 1 f
(µ)
∈ B ∞,q s −m , µ = 0, 1, . . . , m and if f ∈ B p,q s , then ( √
f ) (m) ∈ B p,q s −m
and (
√ f ) (m) B
p,qs−m≤ c m P m µ=0
√ 1 f
(µ) B
s−m∞,∞
!
· kf k B
p,qsFixed metric index, variable (fractional) smoothness index Inverse results
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Inverse results
Using modifications of the ideas behind the proofs of the direct results, we obtained the following four corresponding inverse results of the kind
“ f ∈ B p,q s (or F p,q s ) + additional assumptions ⇒ √
f ∈ B p,q s (or F p,q s )”.
Note, however, that the inverse estimates depend essentially on the behaviour of √ 1
f : namely, in which Besov and Triebel-Lizorkin spaces it is
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Inverse results
Using modifications of the ideas behind the proofs of the direct results, we obtained the following four corresponding inverse results of the kind
“ f ∈ B p,q s (or F p,q s ) + additional assumptions ⇒ √
f ∈ B p,q s (or F p,q s )”.
Note, however, that the inverse estimates depend essentially on the behaviour of √ 1
f : namely, in which Besov and Triebel-Lizorkin spaces it is a pointwise mutliplier. Thus we also get conditions on derivatives of the function 1/pf (x), which here is understood to vanish when f (x) = 0.
Theorem
Let 1 ≤ p ≤ ∞, 0 < q ≤ ∞, m ∈ N 0
def = N ∪ {0}, s > m. If
√ 1 f
(µ)
∈ B ∞,q s −m , µ = 0, 1, . . . , m and if f ∈ B p,q s , then ( √
f ) (m) ∈ B p,q s −m
and (
√ f ) (m) B
p,qs−m≤ c m P m µ=0
√ 1 f
(µ) B
s−m∞,∞
!
· kf k B
p,qsFixed metric index, variable (fractional) smoothness index Inverse results
Besov and Triebel-Lizorkin spaces with fixed metric index and variable, possibly fractional smoothness index
Inverse results
Using modifications of the ideas behind the proofs of the direct results, we obtained the following four corresponding inverse results of the kind
“ f ∈ B p,q s (or F p,q s ) + additional assumptions ⇒ √
f ∈ B p,q s (or F p,q s )”.
Note, however, that the inverse estimates depend essentially on the behaviour of √ 1
f : namely, in which Besov and Triebel-Lizorkin spaces it is
Theorem
Let 0 < p ≤ ∞, 0 < q ≤ ∞, m ∈ N 0 and σ > s > m − a + min{p,1} 1 , s > m. If
√ 1 f
(µ)
∈ B ∞,∞ σ−m , µ = 0, 1, . . . , m and if f ∈ B p,q s , then ( √
f ) (m) ∈ B p,q s −m and
(
√ f ) (m)
B
p,qs−m≤ c m,p P m µ=0
√ 1 f
(µ) B
∞,∞σ−m!
· kf k B
p,qsFixed metric index, variable (fractional) smoothness index Inverse results
Theorem
Let 0 < p ≤ ∞, 0 < q ≤ ∞, m ∈ N 0 and σ > s > m + max n
1
min{p,q} , min{p,q,1} 1 − 1 o . If
√ 1 f
(µ)
∈ B ∞,∞ s −m , µ = 0, 1, . . . , m and if f ∈ F s , then ( √
f ) (m) ∈ F s −m and
Theorem
Let 0 < p ≤ ∞, m ∈ N 0 and
either 1 < q ≤ ∞, s > m + 1/p or 0 < q ≤ 1, s ≥ m + 1/p. If
√ 1 f
(µ)
∈ B p,q s −m , µ = 0, 1, . . . , m and if f ∈ B p,q s , then ( √
f ) (m) ∈ B p,q s −m
and (
√ f ) (m) B
s−mp,q
≤ c m,p P m µ=0
√ 1 f
(µ) B
p,qσ−m!
· kf k B
p,qs.
Fixed metric index, variable (fractional) smoothness index Inverse results
Theorem
Let m ∈ N 0 , 0 < p ≤ ∞, 0 < q ≤ ∞, s > min{p,q} 1 . If
√ 1 f
(µ)
∈ F p,q s −m , µ = 0, 1, . . . , m and if f ∈ F p,q s , then ( √
f ) (m) ∈ F p,q s −m and
Interpolation results
from real and complex interpolation of the so far presented estimates
Our next aim is to generalize the direct results in the first theorem of this talk for the Besov and Triebel–Lizorkin space scales for a continual (fractional) range of the smoothness parameter s. This results in . . . Theorem (Direct results)
Under the assumptions of Theorem 1, suppose that 1 ≤ p ≤ ∞. Then for m ∈ N, 1 ≤ p ≤ ∞, 0 < q ≤ ∞ and 0 < s < m,
it holds that
kf k B
p,qs≤ c(p, q, s, m) ·
√ f
2 B
2p,qs.
If, moreover, p belongs to the more constrained range 1 < p < ∞, then kf k F
p,2s≤ c(p) 1−θ · c(p, m) θ ·
√ f
2
F
2p,2s.
Interpolation results