mathcamp2013 fa slides
TRANSCRIPT
-
8/12/2019 Mathcamp2013 Fa Slides
1/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Functional Analysis Review
Lorenzo Rosascoslides courtesy of Andre Wibisono
9.520: Statistical Learning Theory and Applications
September 9, 2013
L. Rosasco Functional Analysis Review
http://find/http://goback/ -
8/12/2019 Mathcamp2013 Fa Slides
2/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
1 Vector Spaces
2
Hilbert Spaces
3 Functionals and Operators (Matrices)
4 Linear Operators
L. Rosasco Functional Analysis Review
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
3/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Vector Space
A vector space is a set Vwith binary operations
+ : V VV and : R VV
such that for all a, b R and v,w, xV:1 v +w= w +v2 (v +w) + x= v + (w + x)3 There exists 0 Vsuch that v +0=v for all v V4 For every v Vthere exists v Vsuch that v + (v) =0
5 a(bv) = (ab)v6 1v=v7 (a + b)v= av + bv8 a(v +w) =av + aw
L. Rosasco Functional Analysis Review
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
4/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Vector Space
A vector space is a set Vwith binary operations
+ : V VV and : R VV
such that for all a, b R and v,w, xV:1 v +w= w +v2 (v +w) + x= v + (w + x)3 There exists 0 Vsuch that v +0=v for all v V4 For every v Vthere exists v Vsuch that v + (v) =0
5 a(bv) = (ab)v6 1v=v7 (a + b)v= av + bv8 a(v +w) =av + aw
Example: Rn, space of polynomials, space of functions.
L. Rosasco Functional Analysis Review
O li
http://find/http://goback/ -
8/12/2019 Mathcamp2013 Fa Slides
5/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Inner Product
An inner product is a function , : V V R suchthat for all a, b R and v,w, xV:
L. Rosasco Functional Analysis Review
O tli
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
6/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Inner Product
An inner product is a function , : V V R suchthat for all a, b R and v,w, xV:
1 v,w=w,v
2 av + bw, x=av, x + bw, x
3 v,v 0 and v,v=0 if and only ifv=0.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
7/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Inner Product
An inner product is a function , : V V R suchthat for all a, b R and v,w, xV:
1 v,w=w,v
2 av + bw, x=av, x + bw, x
3 v,v 0 and v,v=0 if and only ifv=0.
v,w Vare orthogonal ifv,w=0.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
8/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Inner Product
An inner product is a function , : V V R suchthat for all a, b R and v,w, xV:
1 v,w=w,v
2 av + bw, x=av, x + bw, x
3 v,v 0 and v,v=0 if and only ifv=0.
v,w Vare orthogonal ifv,w=0.
Given WV, we have V=W W, whereW ={vV |v,w=0 for all w W}.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
9/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Inner Product
An inner product is a function , : V V R suchthat for all a, b R and v,w, xV:
1 v,w=w,v
2 av + bw, x=av, x + bw, x
3 v,v 0 and v,v=0 if and only ifv=0.
v,w Vare orthogonal ifv,w=0.
Given WV, we have V=W W, whereW ={vV |v,w=0 for all w W}.
Cauchy-Schwarz inequality: v,w v,v1/2w,w1/2.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
10/53
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Norm
Can define norm from inner product: v=v,v1/2.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
11/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Norm
A norm is a function : V R such that for all a Rand v,w V:
1 v 0, and v=0 if and only ifv=0
2 av= |a| v
3 v +w v + w
Can define norm from inner product: v=v,v1/2.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
12/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Metric
Can define metric from norm: d(v,w) =v w.
L. Rosasco Functional Analysis Review
Outline
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
13/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Metric
A metric is a function d : V V R such that for allv,w, x V:
1 d(v,w) 0, and d(v,w) =0 if and only ifv=w
2 d(v,w) =d(w,v)
3 d(v,w) d(v, x) + d(x,w)
Can define metric from norm: d(v,w) =v w.
L. Rosasco Functional Analysis Review
OutlineV S
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
14/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Basis
B={v1, . . . ,vn} is a basis ofV if every v Vcan beuniquely decomposed as
v=a1v1+ + anvn
for some a1, . . . , an R.
L. Rosasco Functional Analysis Review
OutlineV t S
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
15/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Basis
B={v1, . . . ,vn} is a basis ofV if every v Vcan beuniquely decomposed as
v=a1v1+ + anvn
for some a1, . . . , an R.
An orthonormal basis is a basis that is orthogonal
(vi,vj=0 for i=j) and normalized (vi=1).
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
16/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
1 Vector Spaces
2 Hilbert Spaces
3 Functionals and Operators (Matrices)
4 Linear Operators
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
17/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Hilbert Space, overview
Goal: to understand Hilbert spaces (complete innerproduct spaces) and to make sense of the expression
f=
i=1
f, ii, f H
Need to talk about:
1 Cauchy sequence
2 Completeness
3 Density
4 Separability
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
18/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Cauchy Sequence
Recall: limn
xn =x if for every >0 there exists
N N such that x xn< whenever n N.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
19/53
Vector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Cauchy Sequence
Recall: limn
xn =x if for every >0 there exists
N N such that x xn< whenever n N.
(xn)nN is a Cauchy sequence if for every >0 thereexists N Nsuch that xm xn< whenever m, n N.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
20/53
pHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Cauchy Sequence
Recall: limn
xn =x if for every >0 there exists
N N such that x xn< whenever n N.
(xn)nN is a Cauchy sequence if for every >0 thereexists N Nsuch that xm xn< whenever m, n N.
Every convergent sequence is a Cauchy sequence (why?)
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
21/53
pHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
Completeness
A normed vector space V is complete if every Cauchysequence converges.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
22/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Completeness
A normed vector space V is complete if every Cauchysequence converges.
Examples:1 Q is not complete.
2 R is complete (axiom).
3 Rn is complete.
4 Every finite dimensional normed vector space (over R) iscomplete.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilb S
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
23/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Hilbert Space
A Hilbert space is a complete inner product space.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilb t S
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
24/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Hilbert Space
A Hilbert space is a complete inner product space.
Examples:
1 Rn
2 Every finite dimensional inner product space.
3 2 ={(an)n=1 |an R,
n=1 a
2n
-
8/12/2019 Mathcamp2013 Fa Slides
25/53
OutlineVector Spaces
Hilbert Spaces
-
8/12/2019 Mathcamp2013 Fa Slides
26/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Density
Y is dense in X ifY=X.
Examples:
1 Q is dense in R.
2 Qn is dense in Rn.
3 Weierstrass approximation theorem: polynomials are densein continuous functions (with the supremum norm, on
compact domains).
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
27/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Separability
X is separable if it has a countable dense subset.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
28/53
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
Separability
X is separable if it has a countable dense subset.
Examples:1 R is separable.
2 Rn is separable.
3 2, L2([0, 1]) are separable.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
29/53
pFunctionals and Operators (Matrices)
Linear Operators
Orthonormal Basis
A Hilbert space has a countable orthonormal basis if andonly if it is separable.
Can write:
f=
i=1
f, ii for all fH.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
30/53
pFunctionals and Operators (Matrices)
Linear Operators
Orthonormal Basis
A Hilbert space has a countable orthonormal basis if andonly if it is separable.
Can write:
f=
i=1
f, ii for all fH.
Examples:1 Basis of2 is (1 , 0 , . . . , ), (0 , 1 , 0 , . . . ), (0 , 0 , 1 , 0 , . . . ), . . .
2 Basis ofL2([0, 1]) is 1, 2 sin 2nx,2cos2nx for n N
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
31/53
Functionals and Operators (Matrices)Linear Operators
1 Vector Spaces
2 Hilbert Spaces
3 Functionals and Operators (Matrices)
4 Linear Operators
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
F i l d O (M i )
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
32/53
Functionals and Operators (Matrices)Linear Operators
Maps
Next we are going to review basic properties of maps on a
Hilbert space.functionals: :H R
linear operators A:HH, such thatA(af + bg) =aAf + bAg, with a, b R and f, gH.
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
F ti l d O t (M t i )
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
33/53
Functionals and Operators (Matrices)Linear Operators
Representation of Continuous Functionals
Let H be a Hilbert space and gH, then
g(f) =f, g , fH
is a continuous linear functional.Riesz representation theorem
The theorem states that every continuous linear functional can be written uniquely in the form,
(f) =f, g
for some appropriate element g H.
L. Rosasco Functional Analysis Review
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
34/53
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)
-
8/12/2019 Mathcamp2013 Fa Slides
35/53
Functionals and Operators (Matrices)Linear Operators
Matrix
Every linear operator L : Rm Rn can be represented byan m n matrix A.
IfA Rmn, the transpose ofAis A Rnm satisfying
Ax,yRm = (Ax)y=xAy=x, AyRn
for every x Rn and y Rm.
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
36/53
Functionals and Operators (Matrices)Linear Operators
Matrix
Every linear operator L : Rm Rn can be represented byan m n matrix A.
IfA Rmn, the transpose ofAis A Rnm satisfying
Ax,yRm = (Ax)y=xAy=x, AyRn
for every x Rn and y Rm.
A is symmetric ifA =A.
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
37/53
Functionals and Operators (Matrices)Linear Operators
Eigenvalues and Eigenvectors
Let A Rnn. A nonzero vector v Rn is an eigenvectorofA with corresponding eigenvalue R ifAv=v.
L. Rosasco Functional Analysis Review
http://find/http://goback/ -
8/12/2019 Mathcamp2013 Fa Slides
38/53
-
8/12/2019 Mathcamp2013 Fa Slides
39/53
-
8/12/2019 Mathcamp2013 Fa Slides
40/53
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Li O
-
8/12/2019 Mathcamp2013 Fa Slides
41/53
Linear Operators
Singular Value Decomposition
Every A Rmn can be written as
A=UV,
where U Rmm is orthogonal, Rmn is diagonal,and V Rnn is orthogonal.
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Li O t
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
42/53
Linear Operators
Singular Value Decomposition
Every A Rmn can be written as
A=UV,
where U Rmm is orthogonal, Rm
n is diagonal,
and V Rnn is orthogonal.
Singular system:
Avi =iui AAui =
2
i
ui
Aui =ivi AAvi =
2ivi
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linea O e ato s
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
43/53
Linear Operators
Matrix Norm
The spectral norm ofA Rmn is
Aspec =max(A) =
max(AA) =
max(AA).
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
44/53
Linear Operators
Matrix Norm
The spectral norm ofA Rmn is
Aspec =max(A) =
max(AA) =
max(AA).
The Frobenius norm ofA Rmn is
AF =
m
i=1
n
j=1
a2ij =min{m,n}
i=1
2i .
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
45/53
Linear Operators
Positive Definite Matrix
A real symmetric matrix A Rmm is positive definite if
xtAx >0, x Rm.
A positive definite matrix has positive eigenvalues.
Note: for positive semi-definite matrices > is replaced by .
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
46/53
Linear Operators
1 Vector Spaces
2 Hilbert Spaces
3 Functionals and Operators (Matrices)
4 Linear Operators
L. Rosasco Functional Analysis Review
OutlineVector SpacesHilbert Spaces
Functionals and Operators (Matrices)Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
47/53
Op
Linear Operator
An operator L : H1H2 is linear if it preserves the linearstructure.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
48/53
p
Linear Operator
An operator L : H1H2 is linear if it preserves the linearstructure.
A linear operator L : H1H2 is bounded if there existsC >0 such that
LfH2 CfH1 for all fH1.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://goforward/http://find/http://goback/ -
8/12/2019 Mathcamp2013 Fa Slides
49/53
Linear Operator
An operator L : H1H2 is linear if it preserves the linearstructure.
A linear operator L : H1H2 is bounded if there existsC >0 such that
LfH2 CfH1 for all fH1.
A linear operator is continuous if and only if it is bounded.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
50/53
Adjoint and Compactness
The adjoint of a bounded linear operator L : H1H2 is abounded linear operator L : H2H1 satisfying
Lf, gH2 =f, LgH1 for all fH1, gH2.
L is self-adjoint ifL =L. Self-adjoint operators have realeigenvalues.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
51/53
Adjoint and Compactness
The adjoint of a bounded linear operator L : H1H2 is abounded linear operator L : H2H1 satisfying
Lf, gH2 =f, LgH1 for all fH1, gH2.
L is self-adjoint ifL =L. Self-adjoint operators have realeigenvalues.
A bounded linear operator L : H1H2 is compact if the
image of the unit ball in H1 has compact closure in H2.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
52/53
Spectral Theorem for Compact Self-Adjoint Operator
Let L : HH be a compact self-adjoint operator. Thenthere exists an orthonormal basis ofH consisting of theeigenfunctions ofL,
Li =ii
and the only possible limit point ofi as i is 0.
L. Rosasco Functional Analysis Review
OutlineVector Spaces
Hilbert SpacesFunctionals and Operators (Matrices)
Linear Operators
http://find/ -
8/12/2019 Mathcamp2013 Fa Slides
53/53
Spectral Theorem for Compact Self-Adjoint Operator
Let L : HH be a compact self-adjoint operator. Thenthere exists an orthonormal basis ofH consisting of theeigenfunctions ofL,
Li =ii
and the only possible limit point ofi as i is 0.Eigendecomposition:
L=i=1
ii, i.
L. Rosasco Functional Analysis Review
http://find/