A quick introduction to -convergence and its applications Luigi - - PowerPoint PPT Presentation

a quick introduction to convergence and its applications
SMART_READER_LITE
LIVE PREVIEW

A quick introduction to -convergence and its applications Luigi - - PowerPoint PPT Presentation

A quick introduction to -convergence and its applications Luigi Ambrosio Scuola Normale Superiore, Pisa http://cvgmt.sns.it luigi.ambrosio@sns.it Luigi Ambrosio (SNS) Toronto, October 2014 1 / 11 Outline Basic abstract theory A


slide-1
SLIDE 1

A quick introduction to Γ-convergence and its applications

Luigi Ambrosio

Scuola Normale Superiore, Pisa http://cvgmt.sns.it luigi.ambrosio@sns.it

Luigi Ambrosio (SNS) Toronto, October 2014 1 / 11

slide-2
SLIDE 2

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-3
SLIDE 3

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-4
SLIDE 4

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-5
SLIDE 5

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-6
SLIDE 6

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-7
SLIDE 7

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-8
SLIDE 8

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-9
SLIDE 9

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-10
SLIDE 10

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-11
SLIDE 11

Outline

  • Basic abstract theory
  • A model case with no derivatives
  • Discrete to continuum and viceversa
  • Elliptic operators in divergence form
  • Expansions by Γ-convergence
  • Phase transitions and image segmentation
  • Problems with multiple scales
  • Dimension reduction
  • From convergence of minimizers to evolution problems

Luigi Ambrosio (SNS) Toronto, October 2014 2 / 11

slide-12
SLIDE 12

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-13
SLIDE 13

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-14
SLIDE 14

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-15
SLIDE 15

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-16
SLIDE 16

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-17
SLIDE 17

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-18
SLIDE 18

Introduction

The theory of Γ-convergence was invented in the ’70 by E.De Giorgi. Among the precursors of the theory, one should mention:

  • the Mosco convergence (for convex functions and their duals);
  • the G-convergence of Spagnolo for elliptic operators in divergence

form;

  • the epi-convergence, namely the Hausdorff convergence of the

epigraphs. But, it is only with De Giorgi and with the examples worked out by his school that the theory reached a mature stage.

Luigi Ambrosio (SNS) Toronto, October 2014 3 / 11

slide-19
SLIDE 19

References

G.ALBERTI, S.MÜLLER: A new approach to variational problems with multiple scales. Comm. Pure Appl. Math., 54 (2001), 764-825. L.AMBROSIO, V.M.TORTORELLI: Approximation of functionals depending on jumps by elliptic functionals via Γ-convergence. Comm. Pure Appl. Math., 43 (1990), 999-1036. L.AMBROSIO, N.GIGLI, G.SAVARÉ: Gradient flows in metric spaces and in the space of probability measures. Birkhäuser, 2005 (Second edition, 2008). A.BRAIDES: Γ-convergence for beginners. Oxford University Press, 2002. A.BRADES: Local minimization, variational evolution and Γ-convergence. Springer, 2014. G.DAL MASO: An introduction to Γ-convergence. Birkhäuser, 1993. E.DE GIORGI, T.FRANZONI: Su un tipo di convergenza variazionale. Accad. Naz. Lincei Rend.

  • Cl. Sci. Fis. Mat. Natur., 58 (1975), 842-850.

G.FRIESECKE, R.D.JAMES, S.MÜLLER: A hierarchy of plate models derived from nonlinear elasticity by Γ-convergence. Archive for Rational Mechanics & Analysis, 180 (2006), 183-236. L.MODICA: The gradient theory of phase transitions and the minimal interface criterion. Arch. Rational Mech. Anal., 98 (1987), 123-142. L.MODICA, S.MORTOLA: Un esempio di Γ-convergenza. Boll. Un. Mat. Ital., 14 (1977), 285-299. D.MUMFORD, J.SHAH: Optimal approximation by piecewise smooth functions and associated variational problems. Comm. Pure Appl. Math., 17 (1989), 577-685. E.SANDIER, S.SERFATY: Γ-convergence of gradient flows and applications to Gingzburg-Landau vortex dynamics. Comm. Pure Appl. Math., 57 (2004), 1627-1672. S.SPAGNOLO: Sulla convergenza di soluzioni di equazioni paraboliche ed ellittiche. Ann. Scuola

  • Norm. Sup. Pisa, Cl. Sci., 22 (1968), 577-597.

Luigi Ambrosio (SNS) Toronto, October 2014 4 / 11

slide-20
SLIDE 20

Introduction

Γ-convergence is a “variational” convergence, somehow the most the natural one to pass to the limit in variational problems. More specifically we shall deal with the Γ− convergence, the one designed to pass to the limit in minimum problems. The most general definition of Γ− upper and lower limits, for F : I × X → [−∞, +∞]:          Γ−,+ lim F(x) := sup

U∋x

inf

i∈I sup j≥i

inf

y∈U F(j, y),

Γ−,− lim F(x) := sup

U∋x

sup

i∈I

inf

j≥i inf y∈U F(j, y).

From now on, our index set I will be N and we work in a metric space (X, d), dropping the − from Γ−.

Luigi Ambrosio (SNS) Toronto, October 2014 5 / 11

slide-21
SLIDE 21

Introduction

Γ-convergence is a “variational” convergence, somehow the most the natural one to pass to the limit in variational problems. More specifically we shall deal with the Γ− convergence, the one designed to pass to the limit in minimum problems. The most general definition of Γ− upper and lower limits, for F : I × X → [−∞, +∞]:          Γ−,+ lim F(x) := sup

U∋x

inf

i∈I sup j≥i

inf

y∈U F(j, y),

Γ−,− lim F(x) := sup

U∋x

sup

i∈I

inf

j≥i inf y∈U F(j, y).

From now on, our index set I will be N and we work in a metric space (X, d), dropping the − from Γ−.

Luigi Ambrosio (SNS) Toronto, October 2014 5 / 11

slide-22
SLIDE 22

Introduction

Γ-convergence is a “variational” convergence, somehow the most the natural one to pass to the limit in variational problems. More specifically we shall deal with the Γ− convergence, the one designed to pass to the limit in minimum problems. The most general definition of Γ− upper and lower limits, for F : I × X → [−∞, +∞]:          Γ−,+ lim F(x) := sup

U∋x

inf

i∈I sup j≥i

inf

y∈U F(j, y),

Γ−,− lim F(x) := sup

U∋x

sup

i∈I

inf

j≥i inf y∈U F(j, y).

From now on, our index set I will be N and we work in a metric space (X, d), dropping the − from Γ−.

Luigi Ambrosio (SNS) Toronto, October 2014 5 / 11

slide-23
SLIDE 23

Introduction

Γ-convergence is a “variational” convergence, somehow the most the natural one to pass to the limit in variational problems. More specifically we shall deal with the Γ− convergence, the one designed to pass to the limit in minimum problems. The most general definition of Γ− upper and lower limits, for F : I × X → [−∞, +∞]:          Γ−,+ lim F(x) := sup

U∋x

inf

i∈I sup j≥i

inf

y∈U F(j, y),

Γ−,− lim F(x) := sup

U∋x

sup

i∈I

inf

j≥i inf y∈U F(j, y).

From now on, our index set I will be N and we work in a metric space (X, d), dropping the − from Γ−.

Luigi Ambrosio (SNS) Toronto, October 2014 5 / 11

slide-24
SLIDE 24

Sequential definition of Γ-convergence

Let (X, d) be a metric space, Fn : X → [−∞, +∞] lower semicontinu-

  • us. As in many other cases, to define convergence we pass through

the intermediate notions of upper and lower limits: Γ − lim sup

n→∞

Fn(x) := inf

  • lim sup

n→∞

Fn(xn) : xn → x

  • ,

Γ − lim inf

n→∞ Fn(x) := inf

  • lim inf

n→∞ Fn(xn) : xn → x

  • .

It is obvious that Γ − lim infn Fn ≤ Γ − lim supn Fn, and it is not too difficult to check that they are both lower semicontinuous. We say that Fn Γ converge if Γ − lim sup

n→∞

Fn(x) ≤ Γ − lim inf

n→∞ Fn(x)

∀x ∈ X and we denote the common value of the upper and lower Γ limits by Γ − lim

n→∞ Fn.

Luigi Ambrosio (SNS) Toronto, October 2014 6 / 11

slide-25
SLIDE 25

Sequential definition of Γ-convergence

Let (X, d) be a metric space, Fn : X → [−∞, +∞] lower semicontinu-

  • us. As in many other cases, to define convergence we pass through

the intermediate notions of upper and lower limits: Γ − lim sup

n→∞

Fn(x) := inf

  • lim sup

n→∞

Fn(xn) : xn → x

  • ,

Γ − lim inf

n→∞ Fn(x) := inf

  • lim inf

n→∞ Fn(xn) : xn → x

  • .

It is obvious that Γ − lim infn Fn ≤ Γ − lim supn Fn, and it is not too difficult to check that they are both lower semicontinuous. We say that Fn Γ converge if Γ − lim sup

n→∞

Fn(x) ≤ Γ − lim inf

n→∞ Fn(x)

∀x ∈ X and we denote the common value of the upper and lower Γ limits by Γ − lim

n→∞ Fn.

Luigi Ambrosio (SNS) Toronto, October 2014 6 / 11

slide-26
SLIDE 26

Sequential definition of Γ-convergence

Let (X, d) be a metric space, Fn : X → [−∞, +∞] lower semicontinu-

  • us. As in many other cases, to define convergence we pass through

the intermediate notions of upper and lower limits: Γ − lim sup

n→∞

Fn(x) := inf

  • lim sup

n→∞

Fn(xn) : xn → x

  • ,

Γ − lim inf

n→∞ Fn(x) := inf

  • lim inf

n→∞ Fn(xn) : xn → x

  • .

It is obvious that Γ − lim infn Fn ≤ Γ − lim supn Fn, and it is not too difficult to check that they are both lower semicontinuous. We say that Fn Γ converge if Γ − lim sup

n→∞

Fn(x) ≤ Γ − lim inf

n→∞ Fn(x)

∀x ∈ X and we denote the common value of the upper and lower Γ limits by Γ − lim

n→∞ Fn.

Luigi Ambrosio (SNS) Toronto, October 2014 6 / 11

slide-27
SLIDE 27

Sequential definition of Γ-convergence

Let (X, d) be a metric space, Fn : X → [−∞, +∞] lower semicontinu-

  • us. As in many other cases, to define convergence we pass through

the intermediate notions of upper and lower limits: Γ − lim sup

n→∞

Fn(x) := inf

  • lim sup

n→∞

Fn(xn) : xn → x

  • ,

Γ − lim inf

n→∞ Fn(x) := inf

  • lim inf

n→∞ Fn(xn) : xn → x

  • .

It is obvious that Γ − lim infn Fn ≤ Γ − lim supn Fn, and it is not too difficult to check that they are both lower semicontinuous. We say that Fn Γ converge if Γ − lim sup

n→∞

Fn(x) ≤ Γ − lim inf

n→∞ Fn(x)

∀x ∈ X and we denote the common value of the upper and lower Γ limits by Γ − lim

n→∞ Fn.

Luigi Ambrosio (SNS) Toronto, October 2014 6 / 11

slide-28
SLIDE 28

Sequential definition of Γ-convergence

Let (X, d) be a metric space, Fn : X → [−∞, +∞] lower semicontinu-

  • us. As in many other cases, to define convergence we pass through

the intermediate notions of upper and lower limits: Γ − lim sup

n→∞

Fn(x) := inf

  • lim sup

n→∞

Fn(xn) : xn → x

  • ,

Γ − lim inf

n→∞ Fn(x) := inf

  • lim inf

n→∞ Fn(xn) : xn → x

  • .

It is obvious that Γ − lim infn Fn ≤ Γ − lim supn Fn, and it is not too difficult to check that they are both lower semicontinuous. We say that Fn Γ converge if Γ − lim sup

n→∞

Fn(x) ≤ Γ − lim inf

n→∞ Fn(x)

∀x ∈ X and we denote the common value of the upper and lower Γ limits by Γ − lim

n→∞ Fn.

Luigi Ambrosio (SNS) Toronto, October 2014 6 / 11

slide-29
SLIDE 29

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-30
SLIDE 30

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-31
SLIDE 31

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-32
SLIDE 32

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-33
SLIDE 33

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-34
SLIDE 34

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-35
SLIDE 35

How one proves Γ-convergence

As soon as we have a guess F for the Γ-limit, we have to prove that Γ − lim sup

n→∞

Fn(x) ≤ F(x) and F(x) ≤ Γ − lim inf

n→∞ Fn(x).

The first inequality means that we should be able to find (xn) ⊂ X convergent to x with lim supn Fn(xn) ≤ F(x). Any sequence (xn) with this property is called recovery sequence. The second inequality means that we should be able to prove, for any (xn) ⊂ X convergent to x, the lower bound for the liminf, namely lim infn Fn(xn) ≥ F(x). Warning!! In general pointwise convergence has nothing to do with Γ-convergence, for instance Fn(x) = sin(nx) Γ-converge to −1. In this case xn = − π 2n + 2[nx/2]π n is a recovery sequence.

Luigi Ambrosio (SNS) Toronto, October 2014 7 / 11

slide-36
SLIDE 36

The two basic theorems of Γ-convergence

The first result clarifies the meaning of variational convergence: limits

  • f (asymptotic) minimizers are minimizers and we have convergence of

minimum values. Theorem 1. If Γ − lim

n→∞ Fn = F and (xn) ⊂ X is asymptotically

minimizing for Fn, i.e. Fn(xn) ≤ inf

X Fn + ǫn

with ǫn → 0, then any limit point x of (xn) minimizes F. In addition, under the equi-coercitivity assumption inf

X Fn = inf K Fn

for some compact set K ⊂ X independent of n,

  • ne has that Fn attain their minimum value, and

lim

n→∞ min X Fn = min X F.

Luigi Ambrosio (SNS) Toronto, October 2014 8 / 11

slide-37
SLIDE 37

The two basic theorems of Γ-convergence

The first result clarifies the meaning of variational convergence: limits

  • f (asymptotic) minimizers are minimizers and we have convergence of

minimum values. Theorem 1. If Γ − lim

n→∞ Fn = F and (xn) ⊂ X is asymptotically

minimizing for Fn, i.e. Fn(xn) ≤ inf

X Fn + ǫn

with ǫn → 0, then any limit point x of (xn) minimizes F. In addition, under the equi-coercitivity assumption inf

X Fn = inf K Fn

for some compact set K ⊂ X independent of n,

  • ne has that Fn attain their minimum value, and

lim

n→∞ min X Fn = min X F.

Luigi Ambrosio (SNS) Toronto, October 2014 8 / 11

slide-38
SLIDE 38

The two basic theorems of Γ-convergence

The first result clarifies the meaning of variational convergence: limits

  • f (asymptotic) minimizers are minimizers and we have convergence of

minimum values. Theorem 1. If Γ − lim

n→∞ Fn = F and (xn) ⊂ X is asymptotically

minimizing for Fn, i.e. Fn(xn) ≤ inf

X Fn + ǫn

with ǫn → 0, then any limit point x of (xn) minimizes F. In addition, under the equi-coercitivity assumption inf

X Fn = inf K Fn

for some compact set K ⊂ X independent of n,

  • ne has that Fn attain their minimum value, and

lim

n→∞ min X Fn = min X F.

Luigi Ambrosio (SNS) Toronto, October 2014 8 / 11

slide-39
SLIDE 39

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-40
SLIDE 40

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-41
SLIDE 41

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-42
SLIDE 42

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-43
SLIDE 43

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-44
SLIDE 44

The two basic theorems of Γ-convergence

Proof of the first part. Let x = lim

k→∞ xn(k) be a limit point of (xn).

Obviously we still have F = Γ − lim

k→∞ Fn(k), so that

inf

X F ≤ F(x) ≤ lim inf k→∞ Fn(k)(xn(k)) = lim inf k→∞ inf X Fn(k).

On the other hand, if (yn(k)) is a recovery sequence relative to y, then lim sup

k→∞

inf

X Fn(k) ≤ lim sup k→∞

Fn(k)(yn(k)) ≤ F(y). By taking the infimum w.r.t. y we can obtain infX F in the right hand side. Now, combining these two inequalities we obtain that x minimizes F and that infX Fn(k) converge to minX F.

Luigi Ambrosio (SNS) Toronto, October 2014 9 / 11

slide-45
SLIDE 45

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-46
SLIDE 46

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-47
SLIDE 47

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-48
SLIDE 48

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-49
SLIDE 49

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-50
SLIDE 50

The two basic theorems of Γ-convergence

Theorem 2. If (X, d) is separable, then Γ-convergence is sequentially compact.

  • Proof. Let (Ui)i∈N be a countable basis for the open sets of X, stable

under finite intersections. If Fn are given, we may extract by a diagonal argument a subsequence n(k) such that ℓi := lim

k→∞ inf Ui

Fn(k) exists for all i ∈ N. Then, define F(x) := sup

Ui∋x

ℓi, x ∈ X. The Γ-liminf inequality follows by lim inf

k→∞ Fn(k)(xk) ≥ lim inf k→∞ inf Ui

Fn(k) = ℓi for all i s.t. x ∈ Ui. The proof of Γ-limsup inequality is left as an exercise.

Luigi Ambrosio (SNS) Toronto, October 2014 10 / 11

slide-51
SLIDE 51

Other easy properties

  • When the convergence is monotone, i.e. Fn ≤ Fn+1, the monotone

(or pointwise) limit is F(x) = supn Fn(x) (in this case the recovery sequence is constant). This happens, for instance for the Lp norms

  • |f|p dµ

1/p in a probability space, whose limit and Γ-limit as p ↑ ∞ is the L∞ norm.

  • Γ-convergence is invariant under additive continuous perturbations

and left compositions with non-decreasing maps: F = Γ − lim

n→∞ Fn

= ⇒ F + g = Γ − lim

n→∞(Fn + g) ∀g ∈ C(X, R),

F = Γ− lim

n→∞ Fn

= ⇒ φ◦F = Γ− lim

n→∞ φ◦Fn

φ non-decreasing.

Luigi Ambrosio (SNS) Toronto, October 2014 11 / 11

slide-52
SLIDE 52

Other easy properties

  • When the convergence is monotone, i.e. Fn ≤ Fn+1, the monotone

(or pointwise) limit is F(x) = supn Fn(x) (in this case the recovery sequence is constant). This happens, for instance for the Lp norms

  • |f|p dµ

1/p in a probability space, whose limit and Γ-limit as p ↑ ∞ is the L∞ norm.

  • Γ-convergence is invariant under additive continuous perturbations

and left compositions with non-decreasing maps: F = Γ − lim

n→∞ Fn

= ⇒ F + g = Γ − lim

n→∞(Fn + g) ∀g ∈ C(X, R),

F = Γ− lim

n→∞ Fn

= ⇒ φ◦F = Γ− lim

n→∞ φ◦Fn

φ non-decreasing.

Luigi Ambrosio (SNS) Toronto, October 2014 11 / 11

slide-53
SLIDE 53

Other easy properties

  • When the convergence is monotone, i.e. Fn ≤ Fn+1, the monotone

(or pointwise) limit is F(x) = supn Fn(x) (in this case the recovery sequence is constant). This happens, for instance for the Lp norms

  • |f|p dµ

1/p in a probability space, whose limit and Γ-limit as p ↑ ∞ is the L∞ norm.

  • Γ-convergence is invariant under additive continuous perturbations

and left compositions with non-decreasing maps: F = Γ − lim

n→∞ Fn

= ⇒ F + g = Γ − lim

n→∞(Fn + g) ∀g ∈ C(X, R),

F = Γ− lim

n→∞ Fn

= ⇒ φ◦F = Γ− lim

n→∞ φ◦Fn

φ non-decreasing.

Luigi Ambrosio (SNS) Toronto, October 2014 11 / 11