s. b. damelin, peter dragnev*, and a. b. j. kuijlaars the support of the equilibrium measure for a...
DESCRIPTION
weighted approximation orthogonal polynomials integrable systems random matrix theory Applications of extremal measures : Remark: The support of the extremal measure is a main ingredient in the solution. If we know that the support consists of N intervals, then we set up a system of equations for the endpoints; once we know the endpoints, we find the extremal measure from a Riemann-Hilbert problem.TRANSCRIPT
S. B. Damelin, Peter Dragnev*, and A. B. J. Kuijlaars
The Support of the Equilibrium Measure for a Class of External Fields on a Finite Interval
• Introduction and Results• Potential Theory Preliminaries• Elasticity Interpretation• Proofs
1. Introduction and ResultsExternal field Let Q C( ), say =[-1,1].
Weighted energy Define IQ():
)()(2)()(||
1log)( tdtQtdsdts
IQ
Extremal measure Q is defined as:
IQ(Q)=min{IQ(): ( )},
where ( ) are all unit Borel measures supported on .
weighted approximation orthogonal polynomials integrable systemsrandom matrix theory
Applications of extremal measures:
Remark: The support of the extremal measure is a main ingredient in the solution. If we know that the support consists of N intervals, then we set up a system of equations for the endpoints; once we know the endpoints, we find the extremal measure from a Riemann-Hilbert problem.
Two results on the nature of the support:
• (Mhaskar-Saff, ‘85) If Q is convex, then supp(Q ) is one interval.
• (Deift-Kriecherbauer-McLaughlin, ‘98) If Q is real analytic, then supp(Q ) consists of a finite number of intervals.
The determination of the number of intervals is a
nontrivial problem.
We shall study nonconvex and non real analytic external fields on = [-1,1] of the type:
Q,c(x)= -c sign(x) |x| (1.1)
• For integer - considered in {DKM} and {DaK};• For = [0,1] - considered in {KDr};
Main result: The support of Q for Q = Q,c consists of at most two intervals.
Definition: Let Q be -Hölder continuous on [-1,1], i.e.
Q C1+([-1,1]) |Q´(x)-Q´(y)|C|x-y|,for some >0 and some positive C.
Theorem 1. Let Q C1+([-1,1]). Suppose there exists a1 [-1,1], such that
a) Q is convex on [-1, a1];
b) for every a [-1, a1], there is t0 [a1 ,1], such that
,
dsa)s)(s(1ts
(s)Qπ1t
1
a
decreases on [a1 , t0] and increases on [t0 ,1].
Then supp(Q ) is the union of at most two intervals.
(1.2)
Theorem 2. For 1 and c>0, let Q,c be given by (1.1).
Then for every a[-1,0], there is t0 [0 ,1], such that
,
, dsa)s)(s(1ts(s)Q
π1t
1
a
c
decreases on [0, t0] and increases on [t0 ,1]. As a result,
the support of Q,c consists of at most two intervals.
(1.3)
Remark: The result for any 1 is harder than the integer cases and required a new approach (Th1). This simplified the proofs in the other cases as well.
2. Potential Theory Preliminaries
• Frostman characterization of Q for Q C1+([-1,1]):
U (x)+Q(x)=F for xsupp()
U (x)+Q(x)F for x[-1,1], (2.1)
where U (x)= - log|x-t| d(t).Let =supp(Q ) and dQ = v(t) dt, then we have
log|x-t| v(t)dt = Q(x)-F, x (2.2)
v(t)dt = 1.
If =k [ak ,bk] we get a singular integral equation
int ),( )( xxQdt
txtv
The solution v(t) of the SIE in (2.3) has N (the number of intervals in ) free parameters, determined from the fact that the constant F is unique for all intervals and the total mass of v is one. In general, the solution is not positive. Denote the signed measure =v(t)dt. For =[a,1]
(2.3)
] 11[))(1(
1
dsa)s)(s(1ts
(s)Qattdt
d1
a
(2.4)
Balayage of a measure:
The balayage onto of a nonnegative measure is a
measure that is supported on and
U (x)= U (x)+c, for q. e. x
We write also =Bal(; ). For signed measure = +--
Bal(;) = Bal(+; )-Bal(- ;).
From the definition of it is clear that if 12 then
1 = Bal( 2
; 1)
Two Lemmas on balayage and convexity:
Lemma 1. Let and n be finite unions of intervals with
limn n= . Then limn n = (in weak* sense).
Lemma 2. Let be a finite union of closed intervals and be the associated signed measure, and let v be its density. Suppose that [a,b] and:
(a) Q is convex on [a,b]; (b) v(a)0 and v(b)0; (c) v(t)0 on \ [a,b];
Then v(t)>0 for all t[a,b];
3. Elasticity interpretation [KV]Let the lower half-plane be elastic and Q be the profile of a rigid punch (up to a constant). Suppose a force f is applied to Q. Let D be the displacement of Q. Denote the region of
contact S and the pressure p(x)dx. For small D the profile of the plane is given by -U= log|x-y|p(y)dy. Thus
log|x-y|p(y)dy= Q(x)+F on S
log|x-y|p(y)dy Q(x)+F on [-1,1]. (3.1)
Then =p(t)dt/f is an extremal measure with external field
Q/f (compare (3.1) with Frostman conditions (2.1)).
Now it is easy to illustrate our result. The rigid punch has a profile -x3. Applying force f corresponds to an external field Qc = -cx3 with c=1/f. Then we expect to have three
critical numbers 0<c1<c2 <c3, such that the support is
[-1,1] for c<c1; for c1<c<c2 it is [a,1] with a<0; for c2
<c<c3 it is [a,p][q,1]; and for c>c3 the support is [a,1]
with a<0.
Theorem 1. Let Q C1+([-1,1]). Suppose there exists a1 [-1,1], such that a) Q is convex on [-1, a1]; b) for every a [-1, a1], there is t0 [a1 ,1], such that
,
dsa)s)(s(1ts
(s)Qπ1t
1
a
decreases on [a1 , t0] and increases on [t0 ,1].
Then supp(Q) is the union of at most two intervals.
(1.2)
4. Proofs
Proof: Let a=min {x: x supp()}.
If supp() [a1 ,1], then the problem reduces to
{KDr}. Therefore, assume a< a1.
For every pair (p,q), with a<pq1, let vp,q be the
density of the signed measure with =[a,p][q,1] (if q=1, then =[a,p]);
Introduce Z, consisting of all (p,q) such that a) a<pq1 and q a1;
b) supp() [a,p][q,1];
c) If q<1, then vp,q(t) increases on (q,1).
d) If p>a1, then vp,q(t) decreases on (a1,1).
))(1( att ))(1( att
First, Z. Indeed by formula (2.4) (recall) and condition b) of the theorem we have that (t0, t0) Z.
] 11[))(1(
1
dsa)s)(s(1ts
(s)Qattdt
d1
a
(2.4)
Next, we show that Z is closed. Let (pn,qn) Z and
pnp, qnq. We have to show that (p,q) Z. It is clear that (p,q) satisfies conditions a) and b). By Lemma 1 we obtain vpn ,qn
vp,q from which we derive c) and d).
Finally, we find a pair (p,q) in Z such that q-p
maximal. For this choice vp,q is positive. Since supp()
[a,p][q,1], it follows supp() = [a,p][q,1]. QED
Theorem 2. For 1 and c>0, let Q,c be given by (1.1). Then for every a[-1,0], there is t0 [0 ,1], such that
,
, dsa)s)(s(1ts(s)Q
π1t
1
a
c
decreases on [0, t0] and increases on [t0 ,1]. As a result, the support of Q,c
consists of at most two intervals.
Proof: Write Q=Q,c. Since Q is convex on [-1,1], it is left to show that
, )(
dsa)s)(s(1ts
(s)Qπc
1tG1
a
(4.1)
has the decreasing/increasing property for t[0,1].
21
1
0
1 || IIds
tsa)s)(s(1s
π1
dsts
a)s)(s(1sπ1tG
a || )(
0 1
where I2 is p.v. integral. We establish the properties
(i) G (0) 0;
(ii) G (1) > 0;
(iii) For every >1, there is t [0 ,1], such that
G´ (t)<0 on [0,t ], G´ (t)>0 on [t ,1], and
G´´ (t) 0 on [ t ,1].
(4.2)
Let
Suppose now 1< <2. Consider
f(z)=z -1[(z-1)(z-a)]1/2
defined for z(-,1]. Then I2 may be written as
R
dztz
zfi
dztz
zfi
I )(2
1)(2
12
where the contours and R are given below.
(4.3)
Properties (i) and (ii) are straight forward. We prove (iii) by induction on k=[]. For =1 we
explicitly find G1 (t)=t-(1+a)/2 and (iii) is true with t1=0.
Using contour integration we get the representation:
RCdz
tzzf
itG )(
21)(
dxtx
a)x)(x(1xπ
a
R ||sin 1
dxtx
a)x)(x(1xa
||cos1 0 1
(4.4)
dxtx
a)x)(x(1xπ
tGa
R
)(||sin2)( 3
1
dxtx
a)x)(x(1xπ a
)(
||)cos1(2 0
3
1
Differentiating (4.4) twice and letting R we obtain:
We now conclude that G´´ (t) > 0 for t (0 ,1), in the case
when 1< <2. Therefore G is strictly convex and (iii)
follows easily from G (0) < 0 and G (1) > 0. Thus we
have established (iii) for []=1.
(4.5)
dxa)x)(x(1tx
xtxtxπ
tGa
)(
||||||1)(1 221
Now let k 2, and suppose (iii) is true for all with [] = k-1.
where
(4.6)
)( )(
||||11
1 21ttGdxa)x)(x(1
txxtx
π a
)()(: 1 ttGtF
dxa)x)(x(1xtxtx
πtF
a ||||1)( 2
1
For the derivatives of F(t) we get for 0<t<1:
0 ||)(
21)( 20
2
dxa)x)(x(1xtxx
πtF
a
0 ||)(
41)( 20
3
dxa)x)(x(1xtxx
πtF
a
Differentiating (4.6) we get
)()()()( 11 tGttGtFtG
)()(2)()( 11 tGttGtFtG
(4.7)
(4.8)
By the inductive hypothesis there exists t -1, such that
G´ -1(t)<0 on (0,t -1), and G´ -1(t)>0, G´´ -1 (t) 0 on
(t -1,1). Consider first t -1 >0 (the other case is similar).
)()()()( 11 tGttGtFtG
)()(2)()( 11 tGttGtFtG
F´(t)<0F´´(t)>0 (4.7) (4.8)
• On (0,t -1); G (0) 0, G´ -1(t) <0 G -1(t)
• On (0,t -1); G -1(t) < 0 G´ (t) < 0 on (0,t -1)
• On (t -1,1); G´ -1(t)>0, G´´ -1(t)>0 G´´ (t)>0 (s.c.)
• Then t >t -1: G´ (t) is (+) on (0,t ) and (-) on (t ,1).
• Of course G´´ (t)>0 on (t ,1). This proves (iii). QED