In mathematics, Bochner's theorem (named for Salomon Bochner) characterizes the Fourier-Stieltjes transform of a positive finite Borel measure on the real line.[1] More generally in harmonic analysis, Bochner's theorem asserts that under Fourier transform a continuous positive-definite function on a locally compact abelian group corresponds to a finite positive measure on the Pontryagin dual group. The case of sequences was first established by Gustav Herglotz[2][3]
The theorem for locally compact abelian groups
Bochner's theorem for a locally compact abelian group
G
{\displaystyle G}
, with dual group
G
^
{\displaystyle {\widehat {G}}}
, says the following:[4]
Theorem For any normalized continuous positive-definite function
f
:
G
→
C
{\displaystyle f:G\to \mathbb {C} }
(normalization here means that
f
{\displaystyle f}
is 1 at the unit of
G
{\displaystyle G}
), there exists a unique probability measure
μ
{\displaystyle \mu }
on
G
^
{\displaystyle {\widehat {G}}}
such that
f
(
g
)
=
∫
G
^
ξ
(
g
)
d
μ
(
ξ
)
,
{\displaystyle f(g)=\int _{\widehat {G}}\xi (g)\,d\mu (\xi ),}
i.e.
f
{\displaystyle f}
is the Fourier transform of a unique probability measure
μ
{\displaystyle \mu }
on
G
^
{\displaystyle {\widehat {G}}}
. Conversely, the Fourier transform of a probability measure on
G
^
{\displaystyle {\widehat {G}}}
is necessarily a normalized continuous positive-definite function
f
{\displaystyle f}
on
G
{\displaystyle G}
. This is in fact a one-to-one correspondence.
The Gelfand–Fourier transform is an isomorphism between the group C*-algebra
C
∗
(
G
)
{\displaystyle C^{*}(G)}
and
C
0
(
G
^
)
{\displaystyle C_{0}({\widehat {G}})}
. The theorem is essentially the dual statement for states of the two abelian C*-algebras.
The proof of the theorem passes through vector states on strongly continuous unitary representations of
G
{\displaystyle G}
(the proof in fact shows that every normalized continuous positive-definite function must be of this form).
Given a normalized continuous positive-definite function
f
{\displaystyle f}
on
G
{\displaystyle G}
, one can construct a strongly continuous unitary representation of
G
{\displaystyle G}
in a natural way: Let
F
0
(
G
)
{\displaystyle F_{0}(G)}
be the family of complex-valued functions on
G
{\displaystyle G}
with finite support, i.e.
h
(
g
)
=
0
{\displaystyle h(g)=0}
for all but finitely many
g
{\displaystyle g}
. The positive-definite kernel
K
(
g
1
,
g
2
)
=
f
(
g
1
−
g
2
)
{\displaystyle K(g_{1},g_{2})=f(g_{1}-g_{2})}
induces a (possibly degenerate) inner product on
F
0
(
G
)
{\displaystyle F_{0}(G)}
. Quotienting out degeneracy and taking the completion gives a Hilbert space
(
H
,
⟨
⋅
,
⋅
⟩
f
)
,
{\displaystyle ({\mathcal {H}},\langle \cdot ,\cdot \rangle _{f}),}
whose typical element is an equivalence class
[
h
]
{\displaystyle [h]}
. For a fixed
g
{\displaystyle g}
in
G
{\displaystyle G}
, the "shift operator"
U
g
{\displaystyle U_{g}}
defined by
(
U
g
h
)
(
g
′
)
=
h
(
g
′
−
g
)
{\displaystyle (U_{g}h)(g')=h(g'-g)}
, for a representative of
[
h
]
{\displaystyle [h]}
, is unitary. So the map
g
↦
U
g
{\displaystyle g\mapsto U_{g}}
is a unitary representations of
G
{\displaystyle G}
on
(
H
,
⟨
⋅
,
⋅
⟩
f
)
{\displaystyle ({\mathcal {H}},\langle \cdot ,\cdot \rangle _{f})}
. By continuity of
f
{\displaystyle f}
, it is weakly continuous, therefore strongly continuous. By construction, we have
⟨
U
g
[
e
]
,
[
e
]
⟩
f
=
f
(
g
)
,
{\displaystyle \langle U_{g}[e],[e]\rangle _{f}=f(g),}
where
[
e
]
{\displaystyle [e]}
is the class of the function that is 1 on the identity of
G
{\displaystyle G}
and zero elsewhere. But by Gelfand–Fourier isomorphism, the vector state
⟨
⋅
[
e
]
,
[
e
]
⟩
f
{\displaystyle \langle \cdot [e],[e]\rangle _{f}}
on
C
∗
(
G
)
{\displaystyle C^{*}(G)}
is the pullback of a state on
C
0
(
G
^
)
{\displaystyle C_{0}({\widehat {G}})}
, which is necessarily integration against a probability measure
μ
{\displaystyle \mu }
. Chasing through the isomorphisms then gives
⟨
U
g
[
e
]
,
[
e
]
⟩
f
=
∫
G
^
ξ
(
g
)
d
μ
(
ξ
)
.
{\displaystyle \langle U_{g}[e],[e]\rangle _{f}=\int _{\widehat {G}}\xi (g)\,d\mu (\xi ).}
On the other hand, given a probability measure
μ
{\displaystyle \mu }
on
G
^
{\displaystyle {\widehat {G}}}
, the function
f
(
g
)
=
∫
G
^
ξ
(
g
)
d
μ
(
ξ
)
{\displaystyle f(g)=\int _{\widehat {G}}\xi (g)\,d\mu (\xi )}
is a normalized continuous positive-definite function. Continuity of
f
{\displaystyle f}
follows from the dominated convergence theorem. For positive-definiteness, take a nondegenerate representation of
C
0
(
G
^
)
{\displaystyle C_{0}({\widehat {G}})}
. This extends uniquely to a representation of its multiplier algebra
C
b
(
G
^
)
{\displaystyle C_{b}({\widehat {G}})}
and therefore a strongly continuous unitary representation
U
g
{\displaystyle U_{g}}
. As above we have
f
{\displaystyle f}
given by some vector state on
U
g
{\displaystyle U_{g}}
f
(
g
)
=
⟨
U
g
v
,
v
⟩
,
{\displaystyle f(g)=\langle U_{g}v,v\rangle ,}
therefore positive-definite.
The two constructions are mutual inverses.
Special cases
Bochner's theorem in the special case of the discrete group
Z
{\displaystyle \mathbb {Z} }
is often referred to as Herglotz's theorem and says that a function
f
{\displaystyle f}
on
Z
{\displaystyle \mathbb {Z} }
with
f
(
0
)
=
1
{\displaystyle f(0)=1}
is positive-definite if and only if there exists a probability measure
μ
{\displaystyle \mu }
on the circle
T
{\displaystyle \mathbb {T} }
such that
f
(
k
)
=
∫
T
e
−
2
π
i
k
x
d
μ
(
x
)
,
{\displaystyle f(k)=\int _{\mathbb {T} }e^{-2\pi ikx}\,d\mu (x),}
are the coefficients of a Fourier-Stieltjes series.[5][6]
Similarly, a continuous function
f
:
R
d
→
C
{\displaystyle f:\mathbb {R} ^{d}\to \mathbb {C} }
with
f
(
0
)
=
1
{\displaystyle f(0)=1}
is positive-definite if and only if there exists a probability measure
μ
{\displaystyle \mu }
on
R
d
{\displaystyle \mathbb {R} ^{d}}
such that
f
(
t
)
=
∫
R
d
e
−
2
π
i
ξ
⋅
t
d
μ
(
ξ
)
.
{\displaystyle f(t)=\int _{\mathbb {R} ^{d}}e^{-2\pi i\xi \cdot t}\,d\mu (\xi ).}
Here,
f
{\displaystyle f}
is positive definite if for any finite set of points
α
1
,
⋯
,
α
N
∈
R
d
{\displaystyle \alpha _{1},\cdots ,\alpha _{N}\in \mathbb {R} ^{d}}
, and any complex numbers
ρ
1
,
⋯
,
ρ
N
∈
C
{\displaystyle \rho _{1},\cdots ,\rho _{N}\in \mathbb {C} }
, there holds
∑
p
,
q
=
1
N
f
(
α
p
−
α
q
)
ρ
p
ρ
¯
q
⩾
0.
{\displaystyle \sum _{p,q=1}^{N}f(\alpha _{p}-\alpha _{q})\rho _{p}{\bar {\rho }}_{q}\geqslant 0.}
Applications
In statistics, Bochner's theorem can be used to describe the serial correlation of certain type of time series. A sequence of random variables
{
f
n
}
{\displaystyle \{f_{n}\}}
of mean 0 is a (wide-sense) stationary time series if the covariance
Cov
(
f
n
,
f
m
)
{\displaystyle \operatorname {Cov} (f_{n},f_{m})}
only depends on
n
−
m
{\displaystyle n-m}
. The function
g
(
n
−
m
)
=
Cov
(
f
n
,
f
m
)
{\displaystyle g(n-m)=\operatorname {Cov} (f_{n},f_{m})}
is called the autocovariance function of the time series. By the mean zero assumption,
g
(
n
−
m
)
=
⟨
f
n
,
f
m
⟩
,
{\displaystyle g(n-m)=\langle f_{n},f_{m}\rangle ,}
where
⟨
⋅
,
⋅
⟩
{\displaystyle \langle \cdot ,\cdot \rangle }
denotes the inner product on the Hilbert space of random variables with finite second moments. It is then immediate that
g
{\displaystyle g}
is a positive-definite function on the integers
Z
{\displaystyle \mathbb {Z} }
. By Bochner's theorem, there exists a unique positive measure
μ
{\displaystyle \mu }
on
[
0
,
1
]
{\displaystyle [0,1]}
such that
g
(
k
)
=
∫
e
−
2
π
i
k
x
d
μ
(
x
)
.
{\displaystyle g(k)=\int e^{-2\pi ikx}\,d\mu (x).}
This measure
μ
{\displaystyle \mu }
is called the spectral measure of the time series. It yields information about the "seasonal trends" of the series.
For example, let
z
{\displaystyle z}
be an
m
{\displaystyle m}
-th root of unity (with the current identification, this is
1
/
m
∈
[
0
,
1
]
{\displaystyle 1/m\in [0,1]}
) and
f
{\displaystyle f}
be a random variable of mean 0 and variance 1. Consider the time series
{
z
n
f
}
{\displaystyle \{z^{n}f\}}
. The autocovariance function is
g
(
k
)
=
z
k
.
{\displaystyle g(k)=z^{k}.}
Evidently, the corresponding spectral measure is the Dirac point mass centered at
z
{\displaystyle z}
. This is related to the fact that the time series repeats itself every
m
{\displaystyle m}
periods.
When
g
{\displaystyle g}
has sufficiently fast decay, the measure
μ
{\displaystyle \mu }
is absolutely continuous with respect to the Lebesgue measure, and its Radon–Nikodym derivative
f
{\displaystyle f}
is called the spectral density of the time series. When
g
{\displaystyle g}
lies in
ℓ
1
(
Z
)
{\displaystyle \ell ^{1}(\mathbb {Z} )}
,
f
{\displaystyle f}
is the Fourier transform of
g
{\displaystyle g}
.
See also
Notes
- Katznelson 2004, p. 170.
- William Feller, Introduction to probability theory and its applications, volume 2, Wiley, p. 634
- Rudin 1990, p. 19.
- Reiter & Stegeman 2000, p. 149.
- Maruyama 2017, p. 130.
- Helson 2010, p. 40.
References
- Bochner, S. (1955), Harmonic analysis and the theory of probability, University of California Press, ISBN 978-0-520-34529-4
{{citation}}: ISBN / Date incompatibility (help) - Helson, Henry (2010), Harmonic Analysis, vol. 7, Gurgaon: Hindustan Book Agency, doi:10.1007/978-93-86279-47-7, ISBN 978-93-80250-05-2
- Katznelson, Yitzhak (2004), An Introduction to Harmonic Analysis, Cambridge University Press, doi:10.1017/cbo9781139165372, ISBN 978-0-521-83829-0
- Loomis, L. H. (1953), An introduction to abstract harmonic analysis, Van Nostrand
- Maruyama, Toru (2017), "Herglotz-Bochner representation theorem via theory of distributions", Journal of the Operations Research Society of Japan, 60 (2): 122–135, doi:10.15807/jorsj.60.122, ISSN 0453-4514
- Reed, Michael; Simon, Barry (1975), II: Fourier Analysis, Self-Adjointness, San Diego New York Berkeley [etc.]: Elsevier, ISBN 0-12-585002-6
- Reiter, Hans; Stegeman, Jan Derk (2000), Classical Harmonic Analysis and Locally Compact Groups, Oxford: Oxford University Press on Demand, ISBN 0-19-851189-2
- Rudin, W. (1990), Fourier analysis on groups, Wiley-Interscience, ISBN 0-471-52364-X