wim.
qxp 3/16/99 3:06 PM Page 32
A Celebration of
Women in
Mathematics
I
n March 1994 I organized a two-day conference at
the Massachusetts Institute of Technology in which
nine distinguished women mathematicians gave
colloquium lectures about their current mathe-
matical research. There are many exciting mathe-
matical meetings where the speakers are all men: this
was a meeting where exciting mathematics was pre-
sented that was unusual only in that all the speakers
were women. From the comments that I heard after the
meeting, the conference was clearly very successful. The
audience contained a cross-section of the mathemati-
cal community ranging from MIT Institute Professors
to undergraduates. It was pleasing to hear enthusiasm
on all sides for the mathematics that was discussed. I
have been told that the meeting was particularly in-
spiring for many young women in the audience who are
in the early stages of their mathematical careers.
The conference was funded by MIT and the Visiting
Professorsip for Women Program of the National Sci-
ence Foundation. The Mathematics Department at MIT
was a most supportive host for the conference and I
thank the many people in the department who con-
tributed to the success of the conference, particulary
David Benney and Haynes Miller and the excellent ad-
ministrative staff of the department. To our pleasure,
the conference was opened by Mark Wrighton, the MIT
Provost. Mildred Dresselhaus, a distinguished Institute
Professor, introduced the second day with entertaining
and illuminating remarks about her career at MIT. She
made the very pertinent observation that for a woman
32 NOTICES OF THE
to be really helpful to the cause of women in sci-
ence, it is important that she work from a base
of solid scientific achievement. The talks at the
meeting were excellent illustrations of such
achievements.
This article contains short summaries of the
mathematics presented by the speakers:
Joan S. Birman, Studying Surfaces in Knot Com-
plements
Ingrid Daubechies, Wavelets and Applications
Dusa McDuff, The Geometry of Symplectic Energy
Jill P. Mesirov, Mathematical Theory in Parallel
Algorithms
Cathleen S. Morawetz, The Wave Equation Re-
visited
Jean E. Taylor, Surface Motion Due to Surface En-
ergy Reduction
Chuu-Lian Terng, Soliton Equations and Differ-
ential Geometry
Karen K. Uhlenbeck, Moduli Spaces and Adia-
batic Limits
—Susan Friedlander
University of Illinois-Chicago
(e-mail address: susan@math.nwu.edu)
AMS VOLUME 42, NUMBER 1
wim.qxp 3/16/99 3:06 PM Page 33

Studying Surfaces bedded in 3-space, and so also an orientable sur-

in Knot
Complements
Joan S. Birman
My talk at the Celebra-
tion of Women in Mathe-
matics was about appli-
cations of the theory of
braids to the study of
knots in 3-space.
A knot K is a topolog-
ical circle which is
smoothly embedded in 3-
space R 3 . Its knot type is
its equivalence class
under isotopies of the
pair (R 3 , K) . Our interest
is in the knot problem, i.e.,the problem of how
to decide whether two knots in 3-space determine
the same knot type. My current research (joint
with William Menasco) is directed toward solv-
ing that problem algorithmically. The ingredients
face S of minimum genus. For example, every
representative U of the unknot bounds a disc D,
our standard representative bounding the unit
disc D0 in the plane z = 0 . This disc is foliated
radially by the arcs {D0 ∩ Hθ , 0 ≤ θ < 2π } .
These arcs also define a flow on D0 as one
pushes the Hθ ’s forward through the fibration
of R 3 − A. See Figure 1a. What happens to this
foliation and the associated flow when we replace
U0 with a nonstandard closed-braid represen-
tative U of the unknot, say one which passes n
times around the axis A before closing up to a
circle? Clearly U will still bound a disc, say D,
which will in general admit a singular foliation,
the leaves of the foliation being the components
of {D ∩ Hθ , 0 ≤ θ < 2π }. It can be shown that
the disc D can be assumed to have a particularly
nice foliation: the leaves are all arcs, the singu-
larities (if any) are all saddle points, and near each
point of A ∩ D the foliation is radial. Figure 1b
shows the foliation in a neighborhood of two
points in A ∩ D . It is immediately clear that if
this foliation is to be extended to all of D, then
singularities must occur.
O2@@@@@@@@6K
O2@@@@@0M?e@??I4@@@@@6K
O2@@@0M?h@?g?I4@@@6K

needed for our proposed algorithm are (i) a con-

O2@0M?hf@?he?I4@6K
O2@0M?
O2@@X? @?hf?I4@6K
@? ?I'@6K
O2@0?'1?
O2@0M??N@? @?
@? ?@@@6K
7(MI4@6X
W2@0M?f@L @? ?J@He?I')K?
?W&(M?g31 @? ?75?fV4@6X?
W&(YhN@L?
?W&(Y?h?31? @?
@? ?@H?gI')X
J@h?V')X?
O&(Yhe?N@?
W2@0Y?hf@L @?
@?hf?J@Hhe?V')X?75heV')X
?W&(M? 31 @?hf?75?hfV')X
W&(Y
?W&(Y? N@L?hf@?hfJ@H?hf?V')X?
?31?hf@?hf75 V')K
W&@? ?N@Lhf@?hf@H ?V'@6X

cept of a ‘simplest’ representative K0 of a knot,

?W&@@)X?
W&(MI')X 31hf@?he?J@?
N@hf@?he?75? ?@@)X?
?W&(?'1?
?W&(Y??V')X?
?7(YfV')X ?@L?he@?heJ@H?
?31?he@?he75 W&(Y?N@L
?W&(Y?e3)X?
J@H?f?V')X?
?W&5hV')X ?N@Lhe@?h?J@H
31he@?h?75? W&(YfV')X
?W&(Y?f?V'1
?7(Yh?V')X?
J@H?heV')X N@L?h@?h?@H?
?31?h@?hJ@ W&(YhN@L?
?W&(Y?h?3)X
?W&5hf?V')X?hf?N@?h@?h75
W&(Y V')X @Lh@?g?J@Hhf?W&(Y?hfN@L? W&(Yhe?V'1
7(Y? ?V')X?hf31h@?g?75?hfW&(Y ?3)X
?J@H V')XhfN@L?g@?gJ@H?he?W&(Y? ?V')X?
W&5? ?V')X?he?31?g@?g75hfW&(Y V'1?
7(Y?
?J@Y V')Xhe?N@?g@?g@Hhe?W&(Y?
?V')X?he@Lg@?f?J@?heW&(Y ?N@L
?J@)X?
W&@@6K V')Xhe31g@?f?75?he7(Y? O&@@1?

chosen so that each knot K has at least one and

7(MI4@6K
?J@He?I4@6K ?V')X?hN@L?f@?fJ@H?h?J@H
V')Xh?31?f@?f75heW&5? O2@0M?@L
O2@0M?e31
?75?f?I4@6K
J@H?g?I4@6K ?V')X?g?N@Lf@?f@Hh?W&(Y?
V')Xh31f@?e?J@?hW&(Y ?O2@@0M?fN@L?
?O2@0Mh?31?
75he?I4@@6K? ?V')X?gN@f@?e?75?g?W&(Y?hf?O2@0Mhe?N@L
?J@H I4@6K? V')Xg?@L?e@?eJ@H?gW&(Yhf?O2@0M 31
?75? I4@6K?hf?V')X?f?31?e@??O&@g?W&(Y?he?O2@0M N@L?
J@H? I4@6K?hfV')Xf?N@?e@W2@@5gW&(Yhe?O2@0M ?31?
75
?J@H I4@6K?he?V')X?e?J@@@@@@@@(Yf?W&(Y?hO2@@0M
I4@6K?heV')XeO&@@e@??@H?fW&(YhO2@0M? ?N@L
31
?75?
J@H? I4@@6Kh?V')?2@0Y@L?@?J@f?W&(Y?gO2@0M?
?I4@6Kh?@@@e?31?@?75fW&(YgO2@0M? N@L?
?31?
75
@H ?I4@6Kg7@@@L??N@?@?@He?W&(Y?fO2@0M?
?I4@6Ke?J@?I')Xe@@@@5?eW&(YfO2@0M? ?N@L
31
?J@? ?I4@6KO&5??V')K?@@@@e?W&(Y??O2@@0M? N@L?
?75? ?I4@@@Y?eV'@@(MI')KO&(Y?O2@0M
?I4@@@Y?eV'@@@@@@)KO&(Y?O2@0M ?31?

at most finitely many distinct simplest repre-

O2@@6K J@H?
75 @@@6K??N@(Y??V'@@@UO2@0M
@@@6K??N@@@@@@@@@@UO2@0M
?J@?I4@6K?@HfN@@V@@0M ?N@L
?O2@@@@@@@@@@V4@@@@@@@@6K? ?J@Y ?J@?I4@6K?@@@@@@@@@V@@0M
O&@?e?@@@@?f?@@@@?
O&@?e?@@@@@@@@@@@@@@?
31
V@
?O2@@@0Mh@?hI4@@@6K? ?7@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?f?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@L?
?7@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@L?
O2@@0Mhf@?hfI4@@6K ?@ ?@h@?f?@
?@h@@@@@@@@ I'1?
?O2@@0M?
?O2@0M @?
@? ?I4@@6K?
I4@6K? ?O2@@@@@@@@@@@@@@@@@@6K?
O2@@0M I4@@@@6K J@
75 ?@h@LfJ@
?@h@@@@@@@@
?@h3)X??W&5 ?N@?
?W2@0M @? I4@6X? O2@@@0M? ?I4@@6X? @H ?@h3@@@@@@5
?@L?gV@)KO&@Y
?@L?gV@@@@@@Y
@?
@L
O&(M @? I')K ?O2@@@@@@0M? I')K @? ?3)KeO2@@@@@@@@@@@@6K 31
O2@0Y? @? ?V4@6K ?W2@0M ?V4@@? ?J@? ?V@@@@@0M??W@@@@X??I4@@@6K N@
W2@0M?
?W&(M? @?
@? ?I4@6X
?I')X? ?7(M
?@H? @L
31 ?75?
?@H? O2@@@@@@gW&(Y@@)Xf?I4@@@@@6K
O2@@@@@@@@@0M?e?@L?e?W&(Y?@V')X?h?I4@@@@@@@@@6K ?@L?
?31?
O&(Y
W2@0Y? @?
@? V')K
?V4@6X ?@
?@ N@
?@ J@g?O2@@@@@@@@@@@@@@@@@@@@@@@@0M?hf?3)XeO&(Ye@?V')K
7@@@@@@@@@0M ?V')?2@0Y?e@??V4@6K ?I4@@@@@@@@@@@@@@@@@@@@@@@@6K?f?V@L
I4@@@@@@@@@1
?W&(M? @? ?I')X? ?@ ?@ @? V@@@g@?e?W@@6K ?I'@

sentatives; (ii) a complexity function c(K) which

W&(Y
?W&(Y? @?
@? V')X
?N@)X? J@
75 ?@
?@ ?J@?
?75? O2@0?4@6K?e@@@@@@Y?@@6K
O2@0M?eI4@6K?@@@@@@@@@@@@6K N@
?@L?
W&@H
?W&@@L @?
@? ?J@@)X
W&(?')X? @H
@? ?@
?@@6K? ?@H?
J@ ?O2@@0M?gI4@@@@@@@@@@@@@@@@@6K?
?O2@@@0M @?@@@@@@@@@0M?I4@@@6K? ?31?
?N@?
W&(?')X? @? ?W&(Y?V')X @? I4@@6X 75 ?O2@@@0M @@@@@@@0M?hI4@@@6K? @L
?W&(Y?V')X
?7(Ye?V')X? @?
@? W&(Ye?V')X?
?W&(Y?fV'1? @?
@? ?I'1
N@ @H
@? ?O2@@@0M
O2@@@@0M @@@@0M
@(M? I4@@@6K?
I4@@@@6K 31
N@
J@H?fV')X
?W&5g?V')X? @?
@? W&(Yg?N@L
?W&(Y?h3)X? @?
?O2@@? ?@
?@ ?@
?@L? ?J@?
?7@@@@@@@@@@@@@@@@@@@@@0M? O2@@@@@@@@@0M? @H
@? ?I4@@@@@@@@@6K
?I4@@@@@@@@@@@@@@@@@@@@@L? ?@
W&(YhV')X @? W&(YheV')X ?W2@0M ?)X?he?@he?W2(he?@@? ?@(M ?J@? ?@1?
7(Y?h?V')X?
?J@HhfV')X @?
@? ?W&(Y?he?V'1
W&(Y N@L? ?7(M
J@H? ?3)Xhe?@heW&(Yhe?@H?
?V')X?h?@h?W&(Y?he?@ ?@H?
?@ W&@L
?O&@@)K? ?3@?
?V@?
?75?hf?V')X?
J@H? V')X @?
@? ?W&(Y?
W&(Y ?31?
?N@L 75
@H V')XhJ@hW&(Yhf?@
?V')X?f?O&@@@f?W&(Y?hf?@ J@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
7(M? ?I'@@(M? @?
75
?J@H ?V')X?
V')X @?
@? ?W&(Y?
W&(Y 31
N@L? @?
@? V')K?O2@@@@@gW&(Y
?V'@@@0Me?@f?W&(Y? J@
75 @Y
@@@@@@@@@6K? V'@H
?N@? ?O2@@@@@@@@1 @L

measures how far K is from the same K0; (iii) a

?75?
J@H? ?V')X?
V')X @?
@? ?W&(Y?
W&(Y ?31?
?N@L @?
?J@? ?@@Xf?@fW&(Y
7@@)X?e?@e?W&(Y? @H
@? @?gI4@@@@@@@@6K
@? ?I4@@6K? @?
@? O2@@@@@@@@0Mg?@
O2@@@0M? ?@
75 ?V')X? @? ?W&(Y? 31 ?7@? ?J@?I')XeJ@L?W&(Y @? ?J@? I4@@@6K? @? O2@@@0M? ?@
?J@H
?75? V')X
?V')X? @?
@? W&(Y
?W&(Y? N@L?
?31? ?3@?
?N@? ?75??V')KO&@)?&@H?
?@H?e?@@@0?4@@@ @? ?75? I4@@@6K? @? ?O2@@0M? ?@
?@H? V')X @? W&(Y ?N@? @? ?@H?e?@@@@@@@@@
?@f7(M?f?@L?
?@f7@@@@@@@@@L?
@?
@? ?@H?
?@ I4@@@6K?
I4@@@6K? @?
@? ?O2@@@0M
?O2@@@0M ?@L?
?31?
?@
J@ ?V')X?
V')X @?hf?W&(Y?
@?hfW&(Y @?
@L @?
@? ?@f@Hg?3)K
?@f@@@@@@@@@@)K
@@@@@@@@@@@@@?h@@@@@@@@@@@@h@? @? ?@ I4@6K? @? ?O2@0M ?N@?
75 ?V')X?hf@?he?W&(Y? 31 @L @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@h@?
?@f@Lg?7(M
J@ I4@6K? @? O2@@0M @?
@H V')Xhf@?heW&(Y N@ 31 ?@f@@@@@@@@@@(M
?@L?e3)K?f?@H?
?@L?e3@@@@@@@@@H?
@L
3)X? 75
@H I4@6K?
I4@@6X @?
@? O2@0M?
O2@0M? @?
@L
?J@?
?75? ?V')X?he@?h?W&(Y?
V')Xhe@?hW&(Y ?@L?
?31? N@L?
?31? ?3)Xe?@@6K?O2@@
?3)Xe?@@@@@@@@@
?V')KO&(?4@@@@@@L? V')X
?V')X? @?
@@@@@@@@@@@@@@@@@@6K ?I')K?
V4@6X? @?
@?hf?O&(M? W2@0M? ?O2@@@@@@@@@@@@@@@@@@@ @1
?@H? ?V')X?h@?g?W&(Y? ?N@? ?N@L V'@@(YeW@eI')X V'1? @? ?I4@@@@@@@@6K? I')K @?he?W2@0Y ?O2@@@@@@@0M ?@
?@
?@ V')Xh@?gW&(Y
?V')X?g@?f?W&@ @?
@? 3)X?
V'1? @@Y?e7@e?V@)X?
W&@@@@6X@@@@@@@@)X ?N@L
3)X? @?
@? I4@@@@@@6K ?V4@6Xhf@?heW&(M
?I4@@6K?hf?I')K?he@?h?O&(Y?hf?O2@@0M? O2@@@@@@0M ?@
?@
?@ V')Xg@?fW&@@1? @? ?N@L ?W&(M??I4@@@@@@@@@@)X? V'1? @? I4@@6KhfV4@@he@?g?W2@0YhfO2@@0M ?@

method for reducing c(K) ; and (iv) an under-

?@
J@ ?V')X?e?J@?e?W&(Y@@?
V')XeO&@?eW&(YJ@@? @?
@L 3)X?
V')X W&(Yg@@@@@@0MI')X
?W&(Y?g3@@0M?e?V')X? ?N@?
@? ?J@?
?7@? ?I4@@6K?he?@L?h@?gW&(Mhe?O2@@0M?
I4@@6Kh?3)Xh@?f?W&(Y?hO2@@0M ?@
?@
75 ?V')?2@@@@6?&(Y?7@@L 31 ?V'1 W&(YhN@hV')X @? ?@@@@6K? ?I4@@6K?f?V')X?g@@@@@6T&(Yg?O2@@0M? ?O2@@@
@H
@? V'@@e?I'@(Ye@@@1
V'@@@@@@@@(Ye@@@1
@??N@5fN@f@@@@ N@ N@L?he?W&(Y?h?@h?V')X? @? ?@eI4@@6K I4@@6XfV')KO2@@@@@?e?B@(Y?fO2@@0M ?O2@@@0M?@
@? @??N@@@@@@@@f@@@@
@?e@Hf?31?e@@@@L?
@?e@@@@@@@@1?e@@@@L?
?@
?@ ?31?heW&(Yhe?@heV')X
?N@?he&0Y?he?@he?V4) @?
@? ?@f?I4@@@6K
?@h?I4@@6K? ?I')K?f@@@(M?e@?e?J@HfW2@0M?
V4@6K?O&@@@?f@?eW&5?e?O&(M? ?O2@@@0Mf?@
O2@@0Mh?@L?
@? @?e@?f?N@?e@@@@)K
@?e@@@@@@@@@?e@@@@)K
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?f?J@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ ?@ @L ?@ @? ?@hfI4@@@6K? I'@@@(MI'1f@??W&(Y??W2@0Y O2@@@0M?he?@1?
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@? @?e@Lf?7(Me?@
@1 ?@ @? ?@ I4@@@6K? @@@?eN@L?e@??7(YeO&(M ?O2@@0M? ?@@?
@? @?e@@@@@@@@(Me?@
@Le@1fJ@H?e?@
@Le@@@@@@@@H?e?@
?@
?@ @5
@H @?
@? ?@
?@ I4@@6K
?I4@@@6K W&(?')X??31?e@?J@H?W2@0Y?
?W&(Y?V')X?N@Le@?75?W&(M? ?O2@@@0M
O2@@0M ?@@L
?@@1
@?
@L 31?J@@e?O&@L?eJ@
31?J@@@@@@@@L?eJ@
N@W&@@@@@@@@)Xe75 ?@
J@ @?
@? @L
31 ?@
?@ ?I4@@6K?
I4@@@6K? ?7(Ye?V')X?31e@?@H?7(Y
J@H?fV')XN@L?@@5?J@H? O2@@@0M?
?O2@@0M? ?@@@
?@@@L?hg
31 ?@@(M?I'@??I')K?@H 75 @? N@ ?@ I4@@6Khf75g?V')T@1?@@HW&5 ?O2@@@0M J@@@1?hg
N@
?@ J@@?e?N@?eV'@@5?
?W&@@)X?e@?e?N@@ @H
@? @L
31 ?@
?@ ?@
?@ ?I4@@@6Kg?J@HhV'@Y@?@@T&(Y
?I4@@@6Ke?75?h?N@W@@@V@(Y? ?O2@@@0M
O2@@0M 7@@@@?hg
@@@@@?hg

standing of how the finitely many simplest rep-

?@
?@ W&(MI')Xe@?e?J@@)X
?W&(Y??V')K?@?eO&(?')X? @?
@? N@
?@L? ?@
?@ ?@
J@ ?I4@@6X@Y?he3@Y@@@@HhfO2@@@0M?
I4@@@6K?hV@@@@@@?h?O2@@0M? @@@@@Lhg
@@@@@1hg
?@ W&(YfV4@@@W2@@0Y?V')X @? ?31? ?@ 75 ?@V4@@6Kf?@@0M??I4@@?e?O2@@@0M
?@V4@@6Kf?@@@@@@@@@@?e?O2@@@0M ?J@@@@@@hg
?@L?
?31? ?W&(Y?h@@0Mf?V')X?
W&(Yhe@?hV')X ?J@?
?75? ?N@?
@? J@
75 @H
@? ?@e?I4@@@6K?@h@?O2@@0M
?@e?I4@@@6K?@@@@@@@@@@?O2@@0M
?@h@@@@h@@@@ ?7@@@V4@hg
?N@? ?W&(Y?he@?h?V')X? ?@H? @? ?J@H ?@h@@@@@@@@@@@@@@@@
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@h@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
?@e@?
@L
31 W&(Yhf@?heV')X
?W&(Y?hf@?he?V')X? J@
75 ?J@?
O&5? ?75?
J@H? @X
31 ?@f@@@@0M?@L?f?J@?I4@@@?
?@f@@@@0M?@@@@@@@@@@?I4@@@?
?@W2@@0Mf?3)KfO&5?e?I4@@@6K @?g@?h
N@ W&(Y @?hfV')X @H O2@0Y? 75 N@ ?@W2@@0Mf?3@@@@@@@@5?e?I4@@@6K
O2@@@(M?g?N@@@@@@@(Y?g?I4@@6K?
@?g@?h
@?g@W.?g
?@L?
?31? ?W&(Y?
W&(Y @?hf?V')X?
@? V')X ?J@?
?75? O2@0M?
W2@0M? @H
@? ?@
?@ ?O2@@0M?@Hhe@@X@@@@?hfI4@@@6K?
?O2@@@0Mf@Lh?J@V@@@W@)X? I4@@6K @?f?J@@U?g
@?f?7Y@,?g
?N@? ?W&(Y? @? ?V')X? ?@H? 7(M? ?O2@@? ?@ ?O2@@@0Mh3)X?gW&@X@?@@R')X ?I4@@@6K @?f?@?(Y?g
@L W&(Y @? V')X J@ @H ?@ ?W2@@@0M ?@ O2@@0MhfV')Xf?W&(R@5?@@LV'1 ?I4@@@6K @?
31 ?W&(Y? @? ?V')X? 75 @? ?@ ?7@? ?@ O2@@@0M? ?V')X?eW&(YJ@H?@@1?N@L? ?I4@@6K? @?
N@L?
?31? W&(Y
?W&(Y? @?
@? V')X
?V')X? ?J@H
?75? @?
@? ?)X?he?@hfW&g?3@?
?3)Xhe?@he?W&5g?N@L ?@
?@ ?O2@@0M?
?O2@@@0M V')X?W&(Y?75e@?@L?3)X
?V')?&(Y?J@He@?31?V')K? I4@@@6K?
I4@@6K @?
@?

resentatives are related to one another. Braid

?N@L W&(Y @? V')X J@H? @? ?V')X?h?@heW&(Yh31 ?@ O2@@0M ?@@@Y??75?e@LN@L?V4@6X? ?I4@@@6K @?
31 ?W&(Y? @? ?V')X? 75 @? V')Xh?@h?W&(Y?hN@L? ?@ O2@@@0M? ?W&@@@@6X@H?e@)X@)KeI')K ?I4@@6K? ?J@?
N@L? W&(Y @? V')X ?J@H ?J@? ?V')X?g?@hW&(Yhe?31? ?@hfO2@@@0M? O&(Me?@@@f@@@@@@6K?V@@6X I4@@@6K?he?75?
?3)Xhf?W&(Y?
?V'1hfW&(Y @?
@? ?V')X?
V')X W&5?
7(Y? ?75?
?@H? V')Xe?O2@@@@@f?W&(Y?he?N@?
?V')KO2@0M?@gW&(Y @? ?@h?O2@@0M?
?@f?O2@@@0M W2@0Y?eJ@@@@@@@@@@@@@@@@@@@@)K?
?O&(M?e?W&(M?f@?@@@@@@@@@@@@@6K? I4@@6Kh?@H?
?I4@@@6Kf?@
N@L?h?W&(Y? @? ?V')X?he?J@H ?@ N@@@(Me?@f?W&(Y? @? ?@eO2@@0M O2@@0YfW&(Yg@@@@@@@@@@0MeI4@@6K ?I4@@@6K?@
?3)XhW&(Y @? V')XheW&5? ?@ J@@@f?@fW&(Y @? ?@@@@0M? ?O2@@0M?f?W&(Y?g@@@@@@@@@Xg?I4@@6K? ?I4@@@
?V')X?f?W&(Y? @? ?V')X?g?W&(Y? ?@ ?W&(?')Xe?@e?W&(Y? @? ?@X? O2@@0Mh?7(Yh@@@0M??I')X?hI4@@6K ?@
V'1?fW&(Y
?N@Le?W&(Y? @?
@? V')Xg?7(Y
?V')X?fJ@H? ?@
J@ ?7(Y?V')K??@eO&(Y
?@H?eN@@@@@@@@@H? @?
@? ?31?
?N@? ?O2@@0M?he?@H?h@?gV')Khe?I4@@6K?
O2@@0MhfW2@@he@?g?V4@6XhfI4@@6K ?@
?@
3)X?W&(Y @? V')Xe?W&5 75 ?@fJ@@0M?I4@@
?@fJ@@@@@@@@@ @? @? ?O2@@0M?hf?O&(M?he@?h?I')X?hf?I4@@6K? ?@
V')?&(Y? @? ?V')X?W&(Y @H ?@f7(M?f?@L?
?@f7@@@@@@@@@L? @? @? O2@@@@@@0M ?W2@0Yhf@?heV')K I4@@@@@@6K ?@
?V'@@H @? V')?&(Y? @? ?@f@Hg?3)K
?@f@@@@@@@@@@)K @? @? ?O2@@@@@@@@0M? O&(M @?he?V4@6X ?I4@@@@@@@6K ?@
V'@L
?V')X? @?
@? ?N@@@H
?J@@@? @?
@? @@@@@@@@@@@@@?h@@@@@@@@@@@@h@?
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@h@?
?@f@Lg?7(M @@@@@@@@@@@@@@@@@@0M W2@0Y? @?hf?I')K? ?I4@@@@@@@@@@@@@@@@@@@
V')K @? O&(M @? ?@f@@@@@@@@@@(M
?@L?e3)K?f?@H? @? @? ?O&(M? @? V4@6K? W@
?@L?e3@@@@@@@@@H? @? @L O2@@0Y @? I4@6K? 75

structures provide the tools for attacking these

?V4@6X @? W2@0Y? @? ?3)Xe?@@@6?2@@@L?
?3)Xe?@@@@@@@@@L? @? 31 O2@0M? @? I4@6K? @H
?I')X? @? ?W&(M? @? ?V')KO&(MI'@(MI')X @? N@ O2@0M? @? I4@@6K @?
V')X
?V')K? @?
@? W&(Y
?O&(Y? @?
@? V'@@@He?@Y??V')X?
@@@??O&@@6K?V@)X @?
@? ?@
?@ O2@0M?
O2@@@0M? @?
@? ?I4@6K
?I4@@@6K @?
?J@?
V4@6K? @? ?O2@0Y @? W&(?4@@@@@@@@@@@@)X? @? ?@ O2@@@0M? @? ?I4@@@6K ?75?
I4@6K? @? ?O2@0M @? ?W&(Y?eI'@@@@@@@@@@)X @? ?@L? O2@@@0M? @? ?I4@@6K? ?@H?
I4@6K? @? ?O2@0M @? W&(Yf?N@@@@@0M??I')X?hf@? ?31? O2@@@0M? @? I4@@@6K? ?@
I4@6K?
I4@6K? @?
@? ?O2@0M
?O2@0M @?
@? ?W&(Y?g3@hV')Xhf@?
W&(YhN@h?V')X?he@? ?N@?
@?gO2@@@@@@@@0M ?O2@@0M? ?J@?
W&@L I4@@@6K?
I4@@@@@@@@6Kg?@ ?@
I4@@6K @? O2@@0M @? ?W&(Y?h?@heV')Xhe@? @@@@@@@@@@ ?O&@@)K? @@@@@@@@@@
?I4@@6K?he@?g@@@@@@@@@(M? @L W&(Yhe?@he?V'1he@? @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@5
I4@@@@6Kg@?gV@@@@@@@0Y 31 &0Y?he?@hfV'he@? 3@ ?I'@@(M? @H
?I4@@@@@@@@@@@@@@@@@@@@@0M
?@@@0Mhe?@ N@eW2@@@6X?
?@?O&(M?I'1? ?@
?@ ?J@?
?75? N@
?@ V'@H
?N@? @?
@?
?@M?hf?@ ?@@@0Ye?N@L ?@H? ?@ @? @?

problems.
?@W. 31 ?@ ?@@@@@@@@@@@@@@@6K @? O2@@@@@@@@@@@@@@5?
J@@U N@L? ?@ ?@X?he?I4@@@@@@@@@6K @L O2@@@@@@@@@0M?he?@H?
7Y@,
@?(Y ?31?
?N@? J@
75 ?31?
?N@? ?I4@@@@@@@@6K?
I4@@@6K? @@
?O2@@@@H ?O2@@@@@@@@0M?
?O2@@@0M ?@
?@
@? ?J@H @? I4@@@6K?hfO2@@@@@@@? ?O2@@@0M J@
@? O&5? @L I4@@@6K?f?O2@@@@@@@@@@? ?O2@@@0M 75
@? O2@0Y? 31 I4@@@6KO2@@@@@@@@@@@@@@@@6K?f?O2@@@0M @H
@?
@?hf?W2@@@6K @@@0M?
?J@? N@
?@L? I4@@@??I4@@@@@@@@XeI4@@6KO2@@0M
?I4@6KfI4@@@@f?W@@@@X? ?J@?
?75?
@?hfO&(M?I4@6X W&5? ?31? ?I4@6Kg@HfO&@@@@)X ?@H?
@LheO2@0Y?e?I')K?he?W&(Y? ?N@? ?I4@6Xf@?eW2@0M??I')X? ?@
31hO2@0M?gV4@6X?hO&(Y @@@@@@@@@@@@@@@@@@@@@6K? ?I')X?e@??W&(M?fV')X ?O2@@@@@@@@@@@@@@@@@@@@5
W.eW&g?/X?
7He*@g?N1? N@gW2@0M?heI')KgW2@0Y?
?@f?O&(M?hf?V4@6Ke?O&(M? 3@
N@ I4@@@@@@@@@@@@@@@@6K V')Xe@?W&(Yh@)K?fO2@@@@@@@@@@@@@@@@0M
?I4@@@@@@@@@6Kg?V')K?@W&(Y?gO&@@@@@@@@@0M? @H
?J@?

Choose polar coordinates (z, r , θ) in R 3 . Let

@?eN@eW-X?e@? ?@L??W2@0Y ?I4@@@@@0Y ?@L? ?I4@@@@6K?eN@@@@@@He?O2@@@@0MI'@ ?75?
@?e?@e*>1?e@? ?3)KO&(M ?31? I4@@@6X@(MI'@W2@@@0MgN@
I4@@@6X@@@@@@W2@@@0MgN@ ?@H?
@?e?@eS@@?e@? ?V'@@0Y? ?N@? I'@(Y??V'@(Mhe?@L?
I'@@@@@@@@(Mhe?@L? J@
@?eJ@L?7>@Le@?
@?e@@@?@0R/e@? V+M? @L
3)K? ?N@HfN@H?he?31?
?N@@@@@@@@H?he?31?
@?f?@hf?V@? 75
3Lhe?J5? @@@@@@@@hf?V@?
N@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?f?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@? @H
V/he?.Y? N@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
?@X? ?W@?f?@X?hf@X
?31? ?W@@@@@@@@X?hf@X
O&@LfJ@)Khf@1
O&@@@@@@@@)Khf@1
?@5?
?@H?
?N@?
@L O2@@@)X??W&@@@6Khe@5
O2@@@@@@@@@@@@6Khe@5
O2@0M?W@)KO&@@@W@@6Kh@H J@
31 O2@0M?W@@@@@@@@W@@6Kh@H
O2@0M??W&@@@@@?I'@UI4@6Kg@? 75
?J@H
N@L? ?O2@@0M?eW&(M?@@@eV')X?I4@6Ke?J@? ?75?
?31? ?O2@0Mf?W&(Y?J@@@L??V')X??I4@@6?&5? J@H?
?N@L
31 ?O2@0MgW&(Ye7@@@1?eV')XfI4@@Y?
?O2@0Mg?W&(Y??J@?@?@Le?V')X?fW@@6K? 75
@H

A be the z -axis. Then R 3 − A is fibered by half-

N@L? ?O2@0MhW&(Ye?75?@?31fV')Xe?W&(?4@6K? ?J@?
?31? ?O2@0Mh?W&(Y?e?@H?@?N@f?V')X?W&(YeI4@6K? ?75?
?N@L O2@@0MheW&(YfJ@e@??@L?fV')?&(Y?fI4@6K? J@H?
?W.??W&??@@?e/X 31
N@L? O2@0M?he?W&(Y?f75e@??31?f?N@@@HhI4@@6K
O2@0M?hfW&(Yf?J@He@??N@Lf?J@@@Lhe?I4@6K 75
?J@H
?7H??*@??N@LeN1 ?31? O2@0M?hf?W&(Y?f?75?e@?e31fO&(?')X?he?I4@6K ?75?
?@e?N@?e@)X??@ ?N@L O2@0M? W&(Yg?@H?e@?eN@eO2@0Y?V')Xhf?I4@6K J@H?
?@f@?e@V1??@ 31he?O2@@0M? ?W&(Y?gJ@f@?e?@@@@0M?e?V')X?hf?I4@6Khf75
?@f@?e@?@??@
?@e?J@L?J@W5??@ N@L?g?O2@0M
?31?f?O2@0M W&(Yh7@=?e@W2@@@@(M?gV')X
?W&(Y?g?J@V4@@@@@0Me@?h?V'1 ?I4@6Kh?J@H
?I4@@6K?f?75?
?@e?@@@?@@0Y??@ ?N@?e?O2@0M W&(Yh?75?f@?f31heN@L? I4@6K?eJ@H?
?3L?heJ5 @??O2@0M ?W&(Y?hJ@H?f@?fN@L?h?3)X I4@6KO&5
?V/?he.Y 3@@@0M W&(Yhe75g@?f?31?h?V')X? I4@@(Y
V'@X
?V'1 ?W&(Y?he@Hg@?f?N@?heV')X
W&(Yhe?J@?g@?g@Lhe?V')X? W@H?
?W&5
N@L? ?W&(Y?he?75?g@?g31hfV')X ?7(Y

planes Hθ = {(z, r , θ) ∈ R 3 /θ = constant} . Our

?3)X W&(YhfJ@H?g@?gN@L?he?V')X? J@H?
?V'1 ?W&(Y?hf75h@?g?31?hfV')X ?W&5
N@L? W&(Y
?3)Xhf?W&(Y?hf?J@?h@?h31 @Hh@?g?N@Lhf?V')X?hfW&(Y V')Xhf7(Y?
?V')X?heW&(Y ?75?h@?hN@ ?V')X?h?J@H
V'1?h?W&(Y? J@H?h@?h?@L? V')XhW&5?
?N@LhW&(Y 75he@?h?31? ?V')X?g7(Y?
3)X?f?W&(Y?
V'1?fW&(Y ?J@Hhe@?h?N@L
?75?he@?he31 V')Xf?J@H
?V')X?eW&5?
?N@Le?W&(Y? ?@H?he@?heN@L? V')X?W&(Y?
3)X?W&(Y J@hf@?he?31? ?V')?&(Y
V')?&(Y? 75hf@?he?N@? N@@(Y?
?V4@@?
?I')X? ?J@Hhf@?hf@L
?75?hf@?hf31 J@(Y
?W&(Y?
V')X
?V')X? J@H?hf@?hfN@L?
75 @?hf?31?hf?W&(Y? W&(Y
V')X @H @?hf?N@LhfW&(Y
?V')X?he?J@? @? 31he?W&(Y?

knot K is represented as a closed braid with re-

V')Xhe?75? @? N@heW&(Y
?V')X?hJ@H?
V')Kh75 @?
@? ?@L?g?O&(Y?
?31?f?W2@0Y
?V4@6Kg@H @? ?N@LfO&(M
?I4@6Ke?J@?
?I4@6KO&5? @?
@? 31eO2@0Y?
N@W2@0M?
?I4@@@H?
?I'@ @?
@? ?@@(M?
O2@@0Y
V4@6K? @?hfO2@0M?
I4@6K?hf@?heO2@0M?
I4@@@@@6K?g@?f?O2@@@@0M?
I4@@@@@6K?@W2@@@@@0M
I4@@@@0M

spect to the z -axis A if K meets every Hθ trans- W.eW&g?/X?

7He*@g?N1?
@?eN@eW2@?e@?
@?e?@e7R'?e@?
@?e?@e@?f@?
@?eJ@L?3T.?e@?
@?e@@@?V+Y?e@?
3Lhe?J5?
V/he?.Y?

versally, i.e. K is nowhere tangent to an Hθ . For

example, if K is the ‘unknot’, an especially sim-
ple representative would be the unit circle
U0 = {(0, 1, θ) ∈ R 3 ; 0 ≤ θ < 2π } in the (r , θ)
plane. This is our simplest representative, and
it is unique up to isotopy in the complement of
the braid axis.
Every smooth representative of every knot
type bounds an orientable surface which is em-
Supported in part by NSF grant DMS 91-06584.
Joan Birman is professor of mathematics at Columbia
University, New York, New York. Her e-mail address is
jb@math.columbia.edu.
Figure 1
The foliations which we study can be shown
to decompose the surface S into a union of tiles,
each of which is a 2-cell which contains exactly
one singularity of the foliation. In particular, S
is a union of three types of foliated tiles, depicted
in Figures 2a, 2b, and 2c, each of which has a
canonical embedding in 3-space relative to our
cylindrical coordinates. Two of the tile types in-
tersect K nontrivially, but the third is entirely in
the interior of S . These tiles are all depicted as
they appear when viewed on the positive side of

JANUARY 1995 NOTICES OF THE AMS 33

wim.qxp 3/16/99 3:06 PM Page 34

?@
?@
?@W.
J@@U
7Y@,
@?(Y
O2@6K?g@@@1
?)X?
?@1?
J@@?
7@@?
@@@L
?@
?@
?@W.
J@@U
7Y@,
@?(Y ?)X?
?@1?
J@@?
?
?
?
?
?
?
?
?
?
?
?
?
?
?

??
Wavelets and
Applications
?@g?@@0?4@@g@@@@
?@g?@@@@@@@g@@@@ 7@@?f?W2@@@@6X? ?
?@gJ@f?@L?e?J@@@@L?
?@gJ@@@@@@@L?e?J@@@@L? @@@?f?7(MeI'1?
@@@?f?7@@@@@@1? ?
?@f?O&5f?3)KeO&@@@@)K
?@f?O&@@@@@@@)KeO&@@@@)K
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?g@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@? @@@Lf?@H?e?N@?
@@@Lf?@@@@@@@@? ?
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
?@ ?@X?e?I'1f?7(Me?@@@V40M
?J@@@1f?@g@L
?J@@@1f?@@@@@@@@L ?
?@ ?@X?e?I'@@@@@@@(Me?@@@V40M
?31?fN@f?@H?e?@Y@
@? ?7@@@@f?@g@@f?@
?7@@@@f?@@@@@@@@@f?@ ?
?@ ?31?fN@@@@@@@H?e?@Y@
?N@?f?@@6?2@@g?@
?N@?f?@@@@@@@g?@
@?
@? ?@@@@@f?@L?e?J@Hf?@
?@@@@@f?@@@@@@@@Hf?@
J@@@@@fJ@)KeO&@Lf?@
J@@@@@fJ@@@@@@@@Lf?@
?
?
?@
?@ @LfJ@V'@(Y@L?fJ@
31e?W&5e@??31?f75 @?
@? 7@@@@@e?O&@@@@@@@@)K?e?@
@0Y@V@?W2@0Y@@@@@?I4@6X??@ ?
?
?@ N@L??7(Y?7@1?N@Le?J@H @? ?@L?W&(Me@?@@@?eI')X?@ ?
?@
?@ ?31?J@H??@@@e3)X??75?
?N@W&5e?@@@L?V'1?J@H? @?
@? ?3)?&(Y??J@@@@@Le?V')X@
@@(Ye?7@@?@@1fV'@@ ??
?@
?@ 3@@HeJ@@@1??N@W&5
?@@?e7@@?@??J@@(Y @?
@? W&@@H?eJ@?@?@?@g@@
?W&(Y@f75?@?@@@f?7@@@6X? ?
?
?@ 7@@@6X@?@?@?O&@@ @? W&(Y?@L?e@H?@?@@@L?eJ@eI')X ?
?@ ?J@??I'@@?@?@@@0?'1? @? ?O&(Y??3)X?J@?J@?@@@1??W&5e?V')X? @@@@@@@6K? O2@@@@@@@?g?
?@
?@ ?75?eN@@@@@@@e?N@L
J@H?e?@e@??@L?e3)X? @?
@? ?W2@0Ye?V')?&5?75?@@?@?W&(YfV')X
W&(MgV'@@e@H?@@?@W&(Y?f?V')K? W@@6K?
?O&@@@@6X? O2@0M?hf?
W2@0M?O2@6K?h?
?@
?@ ?W&5f?@e@??31?eV')X
O&(YfJ@e@??N@?e?V')X? @?
@? ?W&(Y?g?N@@)X@??@@?@@(YhV4@6X?
W&(Yhe@@@@@@@@@@@@H?heI')X @@0?4@@@)X
@@@@@@@@)X
?J@?f@W@)X? ?W&(M??@@0?4@@h?
?W&(M??@@@@@@@h?
?@ W2@0Y?f75e@?e@LfV')X @? ?O&(Y?h?J@?e@??@@??@hf?V')X? ?J@@@@@@@W@)X?
?75?f3@Y@)K
?7@@@@@@@@Y@)K
W&(YeJ@f?@L?g?
W&(YeJ@@@@@@@L?g?
?O&@Y??O&5f?31?g?
?O&@Y??O&@@@@@@@1?g?
?@ ?W&(M?g@He@?e31f?V')K? @? ?W2@0Yhe?75??J@??@@??@ V')K ?@g?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?g@?g?
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?g?
?@
?@ O&(Yg?J@?e@?eN@gV4@6X?
W2@0Y?g?75?e@?e?@L?gI')X @?
@? W&(MhfJ@H??75??@@L?@L?hf?V4@6X
?W&(Y?hf75e?@H??@@1?31? ?I')X? ?31?f7(M??@
?3@@@@@@@(M??@
?N@?f@Ye?@L? ?@(Mf@1f?75?g?
?@(Mf@@@@@@@@5?g?
?@ ?W&(M?h?@H?e@?e?31?g?V')X? @? O&(Yhf?J@He?@e?@?@?N@L V')X ?N@@@@@@@Ye?@L?
@@6?2@@@6KC@1?
@@@@@@@@6KC@1?
?@H??W2@@@f?@H?g?
?@H??W2@@@@@@@@@H?g?
?@eW&(M?@@6?2@@h?
?@eW&(M?@@@@@@@h?
?@ O&(YheJ@f@?e?N@?hV')K @? W2@0Y?hfW&5?e?@e?@?@e31 ?V')X? ?I'@@@@W@@@Y@? ?@?O&(Y??@@@@@he?
?@ ?O2@@0Y?he75f@?f@Lh?V4@@6K? @? ?W&(M? 7(Y?eJ@e?@?@eN@L? V')K N@?@@@<I'@@L J@@@0YeJ@@??@he?
?@
?@he?O2@@0M? O2@@0Mhf?J@Hf@?f31hfI4@6K?
?75?f@?fN@ I4@@6Khf@? @? W&(Y
?O&(Y? ?J@Hf75e?@?@L??31?
?75?f@He?@?31??N@L ?V4@6X
?I')X? ?@?3@@L?N@@)K?
?@?N@@)X?@@@@@6K ?O&@(Me?W&@5??@he?
?O2@@@H?eW&@@H??@he?
?@hO2@@0M ?@H?f@?f?@ ?I4@@6K?h@? ?W2@0Y J@H?e?J@?e?@?N@?e31 V')X ?@e3@@)X@e?I4@6K ?O2@0M?@L?e7@@5e?@he?
?@@@@@@@@@@0M?
?@ J@g@?f?@L?
75g@?f?31? I4@@@@@@@@@@@?
I'@? W&(M
?W&(Y? 75f?75?e?@e@?eN@L?
?J@Hf?@H?e?@e@?e?31? ?V')X?
V')K ?@eN@W@@@f?I4@@6K?
?@e?3@Y@@hI4@@6K ?O2@0Me?31??J@?@He?@he?
O2@@0Mf?N@?O&@@5?e?@he?
?@
?@ ?J@Hg@?f?N@?
?75?g@?g@L ?N@?
@? W&(Y
?O&(Y? W&5?fJ@f?@e@Le?N@L
?W&(Y?f75f?@e31f3)X? ?V4@6X
?I')X? ?@e?N@@@@L?h?I4@@6K?
?@e?J@@V')XhfI4@@@@6K ?O2@@0M?h3@@@X@H?e?@he?
?O2@@@0MhfN@@V@5f?@he?
?@
?@ J@H?g@?g3)X?
75h@?gV'1? @?
@? ?W2@0Y
W&(M W&(Yg@Hf?@eN@L?eV')X
?W&(Y?f?J@?f?@e?31?e?V'1 V')X
?V')K? ?@eO&@@LV'1
?O2@@@@6X@W2@0?'1?N@L? ?I4@@@6K
?I4@@@@@@@@@6Kg?O2@@@@@@@@@0M ?O2@@@0M J@@@@YfJ@he?
?W&(Y@@@6Ke@@@@@6K?f?
?@
?@ ?J@Hh@?g?N@L
?75?h@?h31 @?
@? ?W&(Y?
O&(Y ?7(Yg?75?f?@e?N@LfN@L?
J@H?gJ@H?f?@f31f?3)X V4@6X?
I')X W2@@@@@@@@@V@@@0M??N@L?3)X
&@@@@@@@@@@@@@g31?V')X? ?I4@@@@@@@@@0M W&(Y?@?I4@6KV@@@@@@@@@6X?
7(Y?J@e?I4@@@@@@@@@@@@)?
?@ J@H?h@?hN@L? @? W2@0Y? ?W&5h75g?@fN@L?e?V')X? ?V')X? I4@@@@@@@??@gN@eV'1? ?J@He75gW@@@@@@@0Me?
?@ ?W&5he@?h?3)X @? ?W&(M? W&(Yg?J@Hg?@f?31?fV')X V')X ?I4@@@@??@g?@e?N@L W&5?e@Hg7@@@0Mg?
?@ ?7(Yhe@?h?V'1 @? W&(Y ?O&(Y?gW&5?g?@f?N@Lf?V')X? ?V')K? ?I@??@g?@L?e3)X? ?W&(Y?e@?g3@he?
?@
?@ J@H?he@?heN@L?
75hf@?he?3)X @?
@? ?O&(Y?
?W2@0Y ?W2@0Yh7(Y?g?@g31gV')X
W&(Mh?J@Hh?@gN@L?f?V')X? V4@6X?
I')X ?@g?31?eV')X
?@g?N@?e?V')X? ?7(Ye?J@?gN@he?
J@H?e?75?g?@he?

Ingrid Daubechies
?@
?@ ?J@Hhf@?he?V'1
W&5?hf@?hfN@L? @?
@? W&(M
?W&(Y? ?W&(Y?h?75?h?@g?3)XgV')X
O&(Yhe?@H?h?@g?V')X?f?V')X? ?V')X?
V')X ?@h@?fV'1?
?@h@Lf?N@L ?W&5f?@H?g?@he?
W&(Yf?@h?@he?
?@ ?W&(Y?hf@?hf?3)X @? W&(Y W2@0Y?hW2@@he?@hV')XgV')K ?V')K? ?@h31g3)X? 7(Y?fJ@h?@he?
?@
?@ W&(Y
7(Y? @?hf?V'1
@? N@L? @?
@? ?O&(Y?
?W2@0Y ?O&(M?h?W&(M?he?@h?V')X?f?V4@6X
?O2@0YheW&(Yhf?@heV')Xg?I')K? V4@6X?
I')X ?@hN@gV')X
?@h?@g?V'1 ?J@Hg75h?@he?
W&5?g@Hh?@he?
?@ ?J@H @? ?3)X @? W&(M ?O2@0Mhe?W&(Y?hf?@he?V')K?gV4@6K? ?V')X? ?@h?@L?gN@L? ?W&(Y?g@?h?@he?
?@
?@ W&5?
?O&(Y? @?
@? ?V')X?
V')X @?
@? ?O2@@?gO&(Y ?W&(Y? ?O2@0MhfW&(Y
?O2@0Mhf?O&(Y? ?@hfV4@6K?gI4@6K?
?@ I4@6K?gI4@6K? V')K
?V4@6X ?@h?31?g?3)X
?@h?N@?g?V')X? ?7(Yg?J@?h?@he?
J@H?g?75?h?@he?
?@ ?W2@0Y @? ?V')X? @? O2@@@@5?fW2@0Y?hf?O2@@@0Mhf?O2@0Y ?@ I4@6K?gI4@6K? ?I')X? ?@he@?hV'1? ?W&5h?@H?h?@he?
?@
?@ O&(M
W2@0Y? @?
@? V')K
?V4@6X @?
@? W2@@@@@@@@Y?e?O&(M?hfO2@@0M
&@@@@@@@@@@@@@@@@HheO2@@@0M? ?O2@0M
?O2@0M ?@
?@ I4@6K?gI4@6K?
I4@6K?gI4@@6K V')XeO2@@@@@@@?
?V')?2@0M? ?@he@?h?N@L
?@he@Lhe3)X? W&(Yh?@he?@he?
7(Y?h?@he?@he?
?@hf?O&(M?
?@he?W2@0Y @?
@? ?I')K?hf@?
V4@6X?he@? I4@@@@@@@?e@@@?e?O2@@@@@@0M?
@@@@@@@@@@@@@@0M ?O2@0M
O2@@0M ?@
?@ I4@6K?g?I4@@6K?
I4@@6KhI4@@@6K?he?W&@@@)K ?@@@X? ?@he31heV')X
?@heN@he?V'1 ?J@Hhe?@he?@he?
W&5?he?@he?@he?
?@heO&(M @? I')Khe@? W2@@@@@@@@@0MI'@X? ?O2@@0M? ?@ ?I4@@@@6K?gI4@@@6K?gW&(M?I'@@@@@6X ?@he?@hfN@L? ?W&(Y?heJ@he?@he?
?@hO2@0Y?
?@e?O2@@@@0M? @?
@? ?V4@6Kh@?
?I4@@@@6K?e@? 7(M??I'@gV'1?
7@@@@@@@gV'1?
@HfN@g?N@Lhe?O2@@@@0M? O2@@@@0M J@ I4@@@@6KgI4@@@@@6KO&@Y?eV@(M?I')X?
I4@@@@6KgI4@@@@@6KO&@Y?eV@@@@@@)X? ?@he?@hf?3)X ?7(Yhf75he?@he?
?@@@@@0M @? I4@@@@@? @@@@@@@@g?N@Lhe?O2@@@@0M?
@?f?@h@)?2@@@@@@@@@@@?
@@@@@@@@h@)?2@@@@@@@@@@@?
?W&@L?
O&@@)K ?I4@@@6Kh?@@@@@@@@@@(Y?eV'1?
?I4@@@6Kh?@@@@@@@@@@@@@@@@@1?
@@@@@@@@@@@@@@@@@@@@@Hf?N@?
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
?@he?@hf?V')X?
?@he?@ V'1? J@H?hf@Hhe?@he?
?W&5 @?he?@he?
?@
?@ @?
@? I'@?
?N@? @?f?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@Lf?J@?
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
@LfJ@hI'@? ?@he?@ ?N@L W&(Y @?he?@he?
?@ @? @? @@@@@@@@hI'@?
3)K??O&@@6K?f?N@?
?I'@ @(M?e?@@)X?eW&5?
@(M?e?@@@@@@@@@5? ?@he?@ 3)X? 7(Y? @?he?@he?
?@ @? @? 3@@@@@@@@6K?f?N@?
V4@@@@@XI4@6K?f@?
N@
?@
@HeO2@@?@)K?O&(Y?
@HeO2@@?@@@@@@(Y?
@W2@@@X@@@@@@@(Y
?@he?@
?@he?@L?
V')X
?V')X?
?J@H
W&5?
@?he?@he?
@?he?@he?
?@ @? @? I'@V')X?I4@6K?e@? ?@ ?J@@(M?S@(M??I'@H? ?@he?31? V'1? ?W&(Y? @?he?@he?
?@ @? @? ?N@?V')XeI4@6K?@? ?@ W&@@H??7(YfN@ ?@he?N@? ?N@L ?7(Y @?he?@he?
?@
?@ @?
@? @?
@? @??V'1fI4@@@?
@?eN@L?fW@@? ?@
?@ ?W&(Y@eJ@H?f?@
W&(Y?@?O&5g?@ ?@hf@?
?@hf@? 3)X?
V')X J@H?
?W&5 @?he?@he?
@?he?@he?
?@
?@ @?
@? @?
@? @?e?31?f7@@@6X
@?e?N@Le?J@??I')X? ?@
?@ ?O&(Y??3@@(Yg?@
?W2@0Ye?N@@h?@ ?@hf@?
?@hf@? ?V'1
N@L? W&(Y
7(Y? ?J@?he?@he?
?75?he?@he?
?@ @? @? @?f31eW&5?eV')X ?@ W&(Mg@@)Xg?@ ?@hf@? ?3)X ?J@H ?@H?he?@he?

The relatively new de-

?@ ?J@L @? @?fN@?O&(Y?e?V')X? ?@ 7(Y?f?J@V')K?f?@ ?@hf@? ?V')X? W&5? ?@hf?@he?
?@ O&@)K?
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@? @? @?fJ@@@0YgV')X
@?e?O&@@Xh?V')X? ?@
?@ ?J@Hg?75?V4@6K?O2@@@@6K
W&5?gJ@H?eI4@@@Y@@@@@@@@@6X? ?@hf@?
?@hf@? V'1?
?N@L ?W&(Y?
?7(Y ?@hf?@he?
?@hf?@he?
?@ I'@(M? @? @?O2@@0?'1heV')X ?@ ?W&(Y?g75g?I4@@@@@@@@@@@)? ?@hf@? 3)X?hfJ@H? ?@hf?@he?
?@
?@ ?N@H
@? @?
@? @@@@@0M?eN@L?h?V'1
@?g?31?heN@L? ?@
?@ ?7(Yh@Hhe@@@@@@@0M?
J@H?h@?he@@@@0M ?@hf@?
?@hf@? V')Xhe?W&5
?V'1heW&(Y ?@hf?@he?
?@L?he?@he?
?@ @? @? @?g?N@?he?3)X ?@ ?W&5h?J@?he3@ ?@hf@? N@L?h7(Y? ?31?he?@he?
?@ @? @? @?h@?he?V'1 ?@ W&(Yh?75?heN@ ?@hf@? ?3)Xg?J@H ?N@?he?@he?
?@
?@ @?
@? @?
@? @?h@LhfN@L?
@?h31hf?3)X ?@
?@ 7(Y?h?@H?he?@
?J@HheJ@hf?@ ?@hf@?
?@hf@? ?V')X?fW&5?
V'1?e?W&(Y? @?he?@he?
@?he?@he?
?@ @? @? @?hN@hf?V'1 ?@ ?75?he75hf?@ ?@hf@? ?N@Le?7(Y @?he?@he?
?@
?@ @?
@? @?
@? @?h?@
@?h?@L?hf?31? N@L? ?@
?@ J@H?he@Hhf?@
75hf@?hf?@ ?@hf@?
?@hf@? 3)X?J@H?
V')?&5 @?he?@he?
@?he?@he?
?@ @? @? @?h?31?hf?N@L ?@ ?J@Hhe?J@?hf?@ ?@hf@? ?N@@@H @?he?@he?
?@
?@ @?
@? @?
@? @?h?N@?
@?he@? 31
N@L? ?@
?@ ?75?he?75?hf?@
J@H?he?@H?hf?@ ?@hf@?
?@hf@? @@@?
?J@?@L @?he?@he?
@?he?@he?

velopment of wavelet
?@@6K? @? O2@@@@@? @?he@L ?31? ?@ 75hfJ@ ?@ ?@hf@? W&5?3)X? @?he?@he?
?@V4@@6K
?@e?I4@6K @?
@? O2@0M?e@?
O2@0M?f@? @?he31
@?heN@ ?N@?
@L ?@
?@ ?J@Hhf75
?75?hf@H ?@
?@ ?@hf@?
?@hf@? ?W&(Y?V')X
?7(Ye?V'1 @?he?@he?
@?he?@he?
?@f?I4@@6X?
?@hI')X @?
@? W2@0M?g@?
7(M?h@? @?he?@
@?he?@L? 31
N@L? ?@
?@ J@H?hf@?
75hf?J@? ?@
?@ ?@hf@?
?@hf@? J@H?fN@L?
?W&5g?3)X @?he?@he?
@?he?@he?
?@h?V')X? @? ?J@Hhe@? @?he?31? ?31? ?@ @Hhf?75? ?@ ?@he?J@? W&(Yg?V')X? @?he?@he?
?@heV')X
@6X@@@h?V')X? @?
@? W&5?he@?
?W&(Y?he@? @?he?N@?
@?hf@? ?N@?
@L ?@
?@ @?hf?@H?
?J@?hf?@ ?@
?@ ?@he?75?
?@he?@H? 7(Y?hV'1?
?J@Hhe?N@L @Lhe?@he?
31he?@he?
3@@@@@heV')X @? W&(Yhf@? @?hf@L 31 ?@ ?75?hfJ@ ?@ ?@he?@ W&5?hf3)X? N@he?@he?
N@@@@@he?V')X? @? ?W&(Y?hf@? @?hf31 N@ ?@ ?@H?hf75 ?@ ?@he?@ ?W&(Y?hfV')X ?@he?@he?
?@@@@5hfV'1?
?@@@@Hhf?N@L @?
@? W&(Y
?W&(Y? @?
@? @?hfN@
@?hf?@ ?@
?@L? ?@
?@ ?@
J@ @H
@? ?@
?@ ?@he?@
?@he?@ ?7(Y
J@H? ?V'1
N@L? ?@he?@he?
?@he?@he?
?@@@@? 3)X? @? W&(Y @? @?hf?@ ?31? ?@ 75 @? ?@ ?@he?@ ?W&5 ?3)X ?@he?@he?
?3@@@?
?N@@5? V')X
?V')X? @?
@? 7(Y?
?J@H @?
@? @?hf?@
@?hf?@L? ?N@?
@? ?@
?@ @H
?J@? @?
@? ?@
?@ ?@heJ@
?@he75 W&(Y
7(Y? ?V')X?
V'1? ?@L?h?@he?
?31?h?@he?
@@H?
@@ V')X
?V'1 @?
@? W&5?
7(Y? ?J@?
?7@? @?hf?31?
@?hf?N@? @L
31 ?@
?@ ?75?hf?J@?
?@H?hf?75? ?@
?@ ?@he@H
?@he@? ?J@H
W&5? ?N@L
3)X? ?N@?h?@he?
@?h?@he?

theory in the last ten

3@
N@ N@L?
?31? @?
@? ?J@H
?75? ?@@L
?@@1 @?
@? @?
@? N@
?@ ?@
?@ ?@
J@ ?@H?
?@ ?@
?@ ?@he@?
?@he@? ?W&(Y?
?7(Y V')X
?V')X? @?h?@he?
@?h?@he?
?@ ?N@L @? J@H? J@@@ @? @? ?@L? ?@ 75 ?@ ?@ ?@he@? J@H? V'1? @?h?@he?
?@
?@ 31
N@L? @?
@?hf?J@H 75 7@@@
@@@@ @?
@? @?
@? ?31?
?N@? ?@
?@ @H
@? ?@
?@ ?@
?@ ?@he@?
?@h?J@? ?W&5
W&(Y ?N@L
3)X? @Lh?@he?
31h?@he?
?@ ?31? @?hf?75? @@@@L? @? @? @? ?@ ?J@? ?@ ?@ ?@h?75? 7(Y? V')X N@h?@he?
?@ ?N@L @?hfJ@H? ?J@@@@1? @? @L @? ?@ ?75? J@ ?@ ?@h?@H? ?J@H ?V'1 ?@h?@he?
?@
?@ 31
N@L?hf@?he?J@H @?hf75 ?7@@@@@?
?@0Y@@@? @?
@? 31
N@ @?
@? ?@
?@ ?@H?
?@ 75
@H ?@
?@ ?@h?@
?@h?@ W&5?
?W&(Y? N@L?
?3)X ?@h?@he?
?@h?@he?
?@ ?31?hf@?he?75? @V@? @? ?@ @? ?@ J@ @? ?@ ?@h?@ ?7(Y ?V')X?hf?@h?@he?
?@
?@ ?N@Lhf@?heJ@H?
31hf@?he75 @?
@? @?
@? ?@
?@ @?
@L ?@
?@ 75
@H @?
@? ?@
?@ ?@h?@
?@hJ@hf?W&5 J@H? V'1?hf?@L?g?@he?
?N@Lhf?31?g?@he?
?@@@6K
?@?I4@@@6K N@L?he@?h?J@H
?31?he@?h?75? ?O2@@@@?
?O2@@@0Me@? @?
@? ?@
?@ 31
N@ ?@
?@ @?
?J@? @?
@? ?@
?@ ?@h75hfW&(Y
?@h@Hhf7(Y? 3)X?he?N@?g?@he?
V')Xhf@?g?@he?
?@f?I4@@6X?
?@hI')K ?N@Lhe@?hJ@H?
31he@?h75 W2@@0Mg@?
?O&(M?h@? @?
@? ?@
?@L? ?@
?@ ?@
?@ ?75?
?@H? @?
?J@? ?@
?@ ?@h@?he?J@H
?@h@?heW&5? ?V'1hf@?g?@he?
N@L?he@?g?@he?

years or so represents a
?@h?V4@6X N@L?h@?g?J@Hhf?W2@0Yhe@? @? ?31? ?@ ?@ ?@ ?75? ?@ ?@h@?h?W&(Y? ?3)Xhe@?g?@he?
?@he?I')K?hf?31?h@?g?75?hfW&(Mhf@?
?@hfV4@6X?he?N@Lh@?gJ@H?he?O&(Y?hf@? @?
@? ?N@?
@? ?@
?@ ?@
?@ ?@
?@ ?@H?
?@ ?@
?@ ?@h@?h?7(Y
?@g?J@?hJ@H? ?V')X?h@Lg?@he?
V'1?h31g?@he?
?@
?@ I')Xhf31h@?g75he?W2@0Y
?V')K?heN@h@?g@HheW&(M @?
@? @?
@? @?
@? ?@
?@L? ?@
?@ ?@
?@ ?@
?@ ?@
?@ ?@g?75?g?W&5
?@g?@H?gW&(Y ?N@LhN@g?@he?
3)X?g?@g?@he?
?@ V4@6X?h?@L?g@?f?J@?h?O&(Y? @? @? @? ?31? ?@ ?@ ?@ ?@ ?@g?@h7(Y? V')Xg?@L?f?@he?
?@
?@ I')Xh?31?g@?f?75?h@@0Y
?V')X?g?N@?g@?f?@H?g?J@? @?
@? @?
@? @?
@? ?N@?
@? ?@
?@ ?@
?@ ?@
?@ ?@
?@ ?@gJ@g?J@H
?@g75gW&5? ?V')X?f?31?f?@he?
V'1?f?N@?fJ@he?
?@ V')Xh@Lg@?f?@hW&5? @? @? @? @? ?@ ?@ ?@ ?@ ?O@??@g@Hf?W&(Y? ?N@Lg@Lf7@he?
?@
?@ ?V'1h31g@?fJ@g?W&(Y?
N@L?gN@g@?f75gW&(Y @?
@? @?
@? @?
@? @?
@? ?@
?@ J@
75 ?@
J@ ?@
?@ ?O2@@@@??@f?J@?f?7(Y
O2@@@@@@@??@f?75?fJ@H? O2@@@@@@@@@6K? 3)X?f31f3@@@6Kg?
V')XfN@fV@@@@@@@6Ke?
?@ ?3)Xg?@g@?f@Hf?W&(Y? @? @? @? @? ?@ @H 75 ?@ '@@@@@@@@@@@@@f?@H?e?W&5 O2@@@@@@@@@0M?gI4@@@@@@@@@6K? ?V'1f?@?O2@@@@@@@@@@@@(?
?@
?@ ?V')X?f?@L?f@?e?J@?f?7(Y
V')Xf?31?f@?e?75?fJ@H? @?
@? @?
@? @?
@? @?
@L ?@
?@ @?
@? @H
@? ?@
?@ V4@@@@@@@@@W@@@6K??@fW&(YhfO2@@@0M?
?I4@@@@0Y@V4@@@5f7(Y?hO2@@@0M? I4@@@6K? N@L?e?@@@0MW@@@@@@@@@0Y?
I4@@@@6Khe?3)KO2@@@?e@@@@@0M?f?
?@ ?V')X?e?N@?f@?e?@H?e?W&5 @? @? @? 31 ?@ @? @? ?@ ?@eI'@?e?J@Hg?O2@@0M? ?I4@@6K?g?V'@@(M?@LeN@he?
?@ V')Xf@Lf@?e?@fW&(Y @? @? @? N@ ?@ @? @? ?@ ?@e?N@)X?W&5?fO2@@0M I4@@6Kg?@@He31e?@he?

synthesis of ideas from

?@
?@ ?V'1f31f@?eJ@e?W&(Y?
N@L?eN@f@?e75e?7(Y @?
@? @?
@? @?
@? ?@
?@ ?@
?@ @?
@? @?
@? ?@
?@ ?@e?J@@)?&(Y?eO2@0M?
?@e?7(?'@(YeO2@0M? ?I4@@6K?e7@@LeN@e?@he?
I4@6K?@V')X??@e?@he?
?@
?@ ?3)Xe?@L?e@?e@HeJ@H?
?V'1e?31?e@??J@?e75 @?
@? ?@he@? @? @?
@? ?@
?@ ?@
?@ @?
@? @?
@? ?@
?@ ?@e?@H??@H?O2@0M?
?@eJ@?W&@W2@0M? I4@@@?S@)X?@L??@he?
?@@@@>@1?31??@he?
?@
?@ N@L??N@?e@??75??J@H
?31?e@?e@??@H??75? @?
@? ?@he@?
?@W.h@? @?
@? ?@
?@ ?@
?@ @?
@? @?
@?f?@6Khf?@ ?@ ?@e7@T&@@@(M?
?@e@V@@@@@H ?@?I'@Y@LN@??@he?
?@eV4@@)X@L?@he?
?@ ?N@Le@L?J@??@eJ@H? @? J@@Uh@? @L ?@ ?@ @? @?f?3@@@@6Kh?@ ?@e@?@@(Y@? ?@fI4@@@)X@he?
?@
?@ 31e@)?&@@@@?O&5
N@e@@@@@?@@@@@? @?
@? 7Y@,h@? @)K? ?@ ?@ @?
@?(Yh@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?f?V@@@@@@@@6Kf?@ W2@@@@@@@0YJ@?
?W&(M?I'@@e?75? ?@gI'@@@@@6X?g?
?@ ?3@@@@X?@?@??@@)X? @? W@@@@@@@0M ?W&@@@@@@@e?75?
?7(Ye?V'@e?@H?
?7@@@@@@@@e?@H?
?@h@(M?I')Xg?
?@h@@@@@@)Xg?
?@L?f?7(Ye?V'1g?
?@L?f?7@@@@@@@1g?
?@
?@ @@V'1?@?@?J@V')X
W&@@?N@?@@5?75?V')X? @?
@? @@@0M? ?@H?fN@e?@
?@@@@@@@@@e?@ ?@)Kf?@H?fN@g?
?@)Kf?@@@@@@@@@g?
?@L?fJ@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@L?fJ@g?
?@ 7(Y@L?@?@@HJ@HeV'1? @? ?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@g?
?3)Xe?W&@X@(M
?3@@@@@@@@X@(M ?I'@X?eI')Xe?W&5g?
?I'@X?eI'@@@@@@@5g?
?@ ?J@H?31?3@@@?75?e?N@L @? ?V')K?O&@>@(Y?
?V'@@@@@@>@(Y? V')Xe?V')K?O&(Yg?
V')Xe?V'@@@@@(Yg?
?@
?@ ?75??N@LN@@@X@H?f31
?@H?e31?@@V@5gN@L? @?
@? V4@@@@V@(Y
?@?@@0Y? ?V')K?eV4@@@0Y?g?
V4@6K?eI@M?h?
?@ J@fN@?@@@@?g?31? @? ?@@0M? I4@6K?hf?

mathematics, physics,
?@
?@ 7@L?e?@@0?4@1g?N@?
7@L?e?@@@@@@1g?N@?
@V1?eJ@f?@L?g@? @? @@@@@@@0M? I4@@@@@@@?g?
?@ @V1?eJ@@@@@@@L?g@?
O2@@@@@??O&5f?3)Kg@?
O2@@@@@??O&@@@@@@@)Kg@?
@?
@? ?
?
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?g@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@?
?@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@? ?
@@@@@??I'1f?7(Mg@?
@@@@@??I'@@@@@@@(Mg@?
@@@@@?eN@f?@H?g@? ?
@@@@@?eN@@@@@@@H?g@?
3@@@5?e?@@6?2@@h@?
3@@@5?e?@@@@@@@h@?
?
?
N@@@H?fI4@0M? ?
?@@@
?3@@ ?
?
?N@@ ?
@5
(Y ?
?
?
?
?

engineering, and com-

?
?
?
?
?
?
?W.?W26Xg/X
?7H?.MB1gN1 ?W.??W26X?@@e?/X?
?7H??.MB1?N@L??N1? W.eW26Xf?/X?
7He.MB1f?N1? ?
?
?@fJ5?W-Xe?@
?@e?W.Y?*>1e?@ ?@f?J5??@)Xe@?
?@fW.Y??@V1e@? @?fJ5?W2@e@?
@?e?W.Y?7R'e@? ?
?
?@eW.Y??S@@e?@
?@e7YO@?7>@L??@ ?@e?W.Ye?@?@e@?
?@e?7YO@?J@W5e@? @?eW.Y??@f@?
@?e7YO@?3T.e@? ?
?
?@e@@@@?@0R/??@ ?@e?@@@@?@@0Ye@? @?e@@@@?V+Ye@? ?
?3L?heJ5
?V/?he.Y ?3L?he?J5?
?V/?he?.Y? 3Lhe?J5?
V/he?.Y? ?
?
?

puter science, some with

?
?
?
?
?

roots going back to the

Figure 2 start of the century. The
goal of the talk was to
introduce wavelets, to ex-
S . The black dots show the points where A plain both their mathe-
pierces S . There is a well-defined sense of the matical and algorithmi-
flow about each pierce-point, clockwise or coun- cal aspects, and to illustrate these by several
terclockwise. The orientation of K is consistent applications. In 1 dimension, wavelets are typi-
with the sense of the flow around the vertices. cally defined as the dilates and translates of one
Notice that counterclockwise flows are preferred, function,
 
reflecting the fact that the linking number of K x−b
a,b
(x) = |a|−1/2 ,
with A is n > 0 . a
Returning to our example and pulling D back where is often required to have some smooth-
(with its foliation) to its preimage D under the ness and some decay, say | (x)|≤C(1+ |x|)−1−ε ,
embedding we are able to replace D, which is | ˆ (ξ)|
R ≤ C(1 + |ξ|)
−1−ε. Moreover, should sat-
difficult to visualize because of its complicated isfy (x)dx = 0 . It then turns out that any
embedding in 3-space, by the foliated model square integrable function f (and, in fact, many
disc D . A simple example is given in Figure 1c. other functions as well) can be “decomposed”
The disc is pierced 3 times by the braid axis and into such wavelets,
Z Z
∞ ∞
is foliated by two tiles of the type illustrated in
f (x) = C hf , a,b
i a,b
(x)a−2 dadb,
Figure 2a. We leave it as an exercise for the −∞ −∞
reader to reconstruct the embedded disc and the a fact long known to harmonic analysts as
closed braid representative of its boundary from Calderón’s formula and to quantum physicists
the foliated model disc of Figure 1c. as the resolution of the identity for the ax + b -
The kind of data which we can see in the fo- group. The interest of such a decomposition is
liation includes: (i) properties of the graph and that it views f as a superposition of “building
subgraphs formed by the singular leaves; (ii) the blocks” a,b , each of which is well localized in
number and types of tiles which meet at each ver- “time and frequency”; a typical example is
2
tex; (iii) the sense of the flow near the vertices (x) = (1 − x2 )e−x /2 , where | a,b (x)| is mostly
(clockwise or counterclockwise); (iv) the sign of concentrated in the region |x − b| ≤ 2|a| , and
the outward-drawn normal to S near a singularity | ˆ a,b (ξ)| mostly in |a|/4 ≤ |ξ| ≤ 4|a| . By cut-
of the foliation. It turns out that this data (and ting the domain of integration in a, b into dif-
other combinatorial data) tells us how to mod- ferent parts, one effectively carves f into dif-
ify K so as to reduce c(K) . Similar methods also ferent pieces, which may have very different
apply to the study of other surfaces which have properties. This is done in (e.g.) Littlewood-Paley
a special position relative to that of K, e.g., closed type arguments.
incompressible tori in the complement of K. There are also wavelet families where the di-
This work is still in progress. At this writing lation parameter a and the translation parame-
we have learned a great deal about the closed ter b are assigned discrete values only. One
braid representatives of knots and have an al-
gorithm in a very special case: knots of braid
Ingrid Daubechies is a professor in the Program for Ap-
index at most 3, which are represented by a
plied and Computational Mathematics and the Math-
closed 3-braid. We have partial results for closed ematics Department at Princeton University, Prince-
braid representatives of the unlink and other spe- ton, New Jersey. Her e-mail address is ingrid@
cial cases. math.princeton.edu.

34 NOTICES OF THE AMS VOLUME 42, NUMBER 1

wim.qxp 3/16/99 3:06 PM Page 35
widely used choice is a = 2−j , b = 2−j k, with
j, k ∈ Z . It is then also possible to choose in
−j −j
such a way that the j,k = 2 ,2 k constitute
an orthonormal basis for L2 (R). Instead of the
integral representation above, we now have
X
f = hf , j,k i j,k ,
j,k
which can again be understood in L2-sense, or
in many other functional spaces. Harmonic analy-
sis methods can be used to show that such j,k
provide unconditional bases for a variety of Ba-
nach spaces, such as Lp (R), 1 < p < ∞, the Hardy
space H 1 (R) and its dual BMO, the Sobolev
q
spaces W s (R) , the Besov spaces Bp,s (R) , the
s
Hölder spaces C (R) , etc. That is, for each of
these spaces, there exists a criterion that de-
cides whether f lies in the space or not, solely
by looking at the behavior of the absolute val-
ues |hf , j,k i| of the wavelet coefficients. This
unconditionality of wavelet bases for many func-
tional spaces makes them a special and power-
ful tool.
Wavelet bases are also linked with fast algo-
rithms. Every wavelet base is derived from a
multiresolution analysis (provided has some
decay and some smoothness). This means that
there exists an auxiliary function φ, the scaling
function, such that
X X
φ(x) = cn φ(2x − n), (x) = dn φ(2x − n);
n n
R Cambridge Univ. Press.
moreover, φ(x)dx = 1 . Because of this latter
property, φ can be used to write a resolution of
the identity, i.e., Z
f (x) = lim 2J f (y)φ(2J (y − x))dy
J→∞
forR continuous f . In particular, the integrals
2J f (y)φ(2J y − k)dy can be approximated by
the sampled values f (2−J k) . (In practice, this
approximation may have to be refined.) By the
recurrence relations above, one can then, by
means of a fast algorithm, compute for suc-
c e s sive j < JR the wavelet coefficients
hf , j,k i = 2j/2 f (y) (2j x − k)dx and the scal-
ing coefficients hf , φj,k i. In fact, algorithms of
this type were already known to electrical engi-
neers as subband filtering (with exact recon-
struction).
A combination of these fast algorithms on
one hand, and a thorough and deep under-
standing of the mathematical properties of the
underlying functional analytic tool on the other
hand, has led to a variety of applications. A few
examples discussed in the talk were compression
of large, dense matrices (Beylkin, Coifman, and
Rokhlin); theorems showing that wavelets give
asymptotically optimal denoising and estima-
tion techniques for many different types of noisy
JANUARY 1995
signals (Donoho—this application exploits that
wavelets are an unconditional basis in many
spaces); image compression and manipulation
from edge information encoded in wavelet co-
efficients (Mallat).
References
[1] J. Benedetto, and M. Frazier, eds. Wavelets: Math-
ematics and Applications. CRC Press, 1993.
[2] G. Beylkin, R. Coifman, and V. Rokhlin, Fast
wavelet transforms and numerical algorithms, Comm.
Pure Appl. Math. 44 (1991), 141–183.
[3] D. Donoho, Unconditional bases are optimal
bases for data compression and for statistical estima-
tion, Applied and Computational Harmonic Analysis,
1 (1992) , 100–115.
[4] C. Chui, ed., Wavelets: A tutorial in theory and
applications, Academic Press, 1992.
[5] I. Daubechies, Ten lectures on wavelets. CBMS-
NSF Regional Conf. Ser. in Appl. Math., vol. 61, SIAM,
Philadelphia, PA, 1992.
[6] ———, ed., Different perspectives on wavelets,
Proc. Sympos. Appl. Math., vol. 46, Amer. Math. Soc.,
Providence, RI, 1993.
[7] D. Donoho,Unconditional bases are optimal bases
for data compression and for statistical estimation,
Appl. Comput. Harmonic Anal., 1 (1992), 100–115.
[8] S. Mallat and S. Zhong, Characterization of sig-
nals from multiscale edges, IEEE Trans. PAMI 14 (1992).
[9] Y. Meyer,Ondelettes ét opérateurs, I: Ondelettes;
II: Opérateurs de Calderón-Zygmund; III: Opérateurs
multilinéaires, Hermann, Paris, 1990; English transl.
[10] ———, Wavelets: Algorithms & applications. SIAM,
Philadelphia, PA, 1993.
[11] M. B. Ruskai et al., eds., Wavelets: Theory and
Application, Jones and Bartlett, Boston, 1992.
The Geometry of
Symplectic Energy
Dusa McDuff
The talk was divided into three parts.
The first part was an elementary introduction
to symplectic geometry—the geometry intro-
duced on a smooth 2n -manifold M by a closed
2 -form ω which is nondegenerate (that is, its top
exterior power ωn is a volume form). The basic
example of a symplectic manifold is Euclidean
space R2n with its standard form
ω0 = dx1 ∧ dx2 + · · · + dx2n−1 dx2n .
Dusa McDuff, F.R.S., is professor of mathematics at the
State University of New York at Stony Brook. Her e-mail
address is dusa@math.sunysb.edu.
NOTICES OF THE AMS 35
wim.qxp 3/16/99 3:06 PM Page 36
One important concept is that of the sym-
plectic gradient XH of a function H : M → R .
This vector field is characterized by the equation
ι(XH )ω = ω(XH , ·) = dH,
and, in contrast to the gradient with respect to
a Riemannian metric, points tangent to the level
surfaces of H . Moreover, the flow which it gen-
erates (called the Hamiltonian flow) preserves the
symplectic form ω. Thus, if M is the unit 2 -
sphere S 2 ⊂ R3 with its usual area form, and if
H is the height function (given by the third co-
ordinate function), the symplectic gradient XH
is tangent to the horizontal circles H = const and
generates a rotation about the north-south axis.
Here, one can let the Hamiltonian depend on
t
time t ∈ [0, 1]. One then gets a family XH of vec-
tor fields which integrates up to a family
φt , 0 ≤ t ≤ 1, of symplectomorphisms. The map
φ1 is called the time- 1 map of Ht . Essentially,
any symplectomorphism (i.e., diffeomorphism
which preserves ω) is the time 1 -map of such
a flow.
Symplectomorphisms have important geo-
metric properties, which are just beginning to be
understood. One striking result is Gromov’s non-
squeezing theorem (1985), which states that if
there is a symplectomorphism which embeds a
2n -ball of radius r into a cylinder of radius R ,
then r ≤ R. Here, a cylinder is a product of a 2-
dimensional disc of radius R with another sym-
plectic manifold (M, ω) of dimension 2n − 2 .
Gromov proved this result when M is Euclidean
space R2n−2 with its usual symplectic form, and
it has been proved for all manifolds M by
Lalonde–McDuff.
This leads to the notion of a capacity. We de-
fine the capacity c(B) of a ball of radius r to be
π r 2 , and, for an arbitrary subset U in a sym-
plectic 2n -manifold, define c(U) to be the supre-
mum of the capacities of the symplectic 2n -
balls which embed symplectically in U. Thus, the
nonsqueezing theorem implies that the capac-
ity of the cylinder B 2 (r ) × R2n−2 is π r 2 .
The second part introduced Hofer’s idea of the
energy of a symplectomorphism φ. We measure
the size of a Hamiltonian function H by
Totvar H = max H(x) − min H(x),
x∈M x∈M
and define the length of the path φt , 0 ≤ t ≤ 1,
generated by Ht to be
Z1
L(φt ) = Totvar Ht dt.
0
The energy e(φ) is then defined to be the infi-
mum of the lengths of the paths φt which have
time- 1 map equal to φ. Note that, although one
uses the first derivative of H in order to gener-
ate the flow, the size of H depends only on its
36 NOTICES OF THE AMS
values, not its derivative. Therefore, it is a pri-
ori not clear that the energy of a nontrivial map
is always positive. That this is the case follows
from Hofer’s energy-capacity inequality, which
states that if U is an open set which is disjoined
by φ, i.e.,
φ(U) ∩ U = ∅,
then e(φ) ≥ c(U) . As an example, consider the
function H = y on R2 with ω0 = dx ∧ dy . The
flow generated by H is translation in the x di-
rection:
(x, y) 7→ (x + t, y),
and it is easy to check that the energy required
to disjoin the rectangle [0, 1] × [0, a] is precisely
the area a.
The third part of the talk outlined a geomet-
ric proof of the energy-capacity inequality, which
is due to Lalonde–McDuff. The idea is to use a
flow φt whose time-1 map disjoins a ball in M
of radius r to construct an embedded ball of ra-
dius r in a cylinder B 2 (R) × M where
π R 2 = L (φt ) + π r 2 /2 + .
The nonsqueezing theorem then implies that
π r 2 ≤ L (φt ) + π r 2 /2 + ,
and hence, taking the infimum over all such
paths φt that
c(U) = sup π r 2 ≤ 2e(φ).
A refinement of this argument gives (a slightly
modified version of) the full energy-capacity in-
equality.
Reference
F. Lalonde and D. McDuff, The geometry of symplectic
energy, Ann. of Math. (to appear).
Mathematical
Theory in Parallel
Algorithms
Jill P. Mesirov
Suppose we would like to understand a physi-
cal phenomenon, for example, how air flows
past an airplane wing, or how a protein folds. We
might perform real experiments and get data
from measurements of actual phenomena in the
Jill Mesirov is Director of Research at Thinking Ma-
chines Corporation in Cambridge, Massachusetts. Her
e-mail address is mesirov@think.com.
VOLUME 42, NUMBER 1
wim.qxp 3/16/99 3:06 PM Page 37
laboratory, e.g., via a
wind tunnel. However, it
would be difficult infer-
ring the behavior in un-
known cases from this
data. On the other hand,
we might also try to build
a mathematical model,
perhaps adding some
simplifying assumptions,
which characterizes the
physical phenomenon.
Using the model we can
Jill P. Mesirov
then do simulations. His-
torically, these calculations were done by hand,
but now we use computers to do them. We can
test the model’s predictions by comparing the
results of simulations against real experimental
data obtained for stereotypic examples. If the pre-
dictions match the experimental data, then we
gain confidence in our models and may use them
to predict behavior in cases where it is unknown.
The mathematical model/simulation para-
digm described above is used to help design the
cars we drive, the medicines we take, and the air-
planes in which we fly. This approach often rep-
resents a more cost-effective way of doing some
“experiments” via computer simulation. Its suc-
cess has created an ever-increasing demand for
higher-performance computers to allow for real
time, interactive design, and graphics. Mathe-
matics plays an important role in this process,
both in modeling physical phenomena, and in de-
signing efficient algorithms for high-perfor-
mance parallel computers.
For the purpose of our discussion, a parallel
computer consists of a collection of processors,
each with its own local memory, connected by a
network. A processor can only operate on data
in its own local memory. If it has to use data that
is located in another processor’s memory, then
that data must be sent over the network. This
type of action is referred to as “communica-
tions”. When designing algorithms for such dis-
tributed- memory parallel computers, we seek to
minimize the time spent in communications.
This makes the algorithms more efficient in
using the processing power of the machine. In
this lecture we described an interesting interplay
between mathematics and the design of effi-
cient parallel algorithms in the context of N-
body algorithms for both fluid flow and molec-
ular dynamics applications. We also showed
some videos of our simulations which the labo-
ratories of the future may produce in real time.
The two cases we considered require differ-
ent communication patterns. For the fluid flow
problem, we showed how rotated and translated
Gray codes can be used to construct timewise
edge independent Hamiltonian paths on hyper-
JANUARY 1995
cubes, yielding an optimal global communication
pattern for machines with this kind of network.
In molecular dynamics codes for protein dy-
namics, the force law is a local one. Thus, one is
led to solving a local, i.e., on subcubes of side
2a + 1 , gossiping problem in multidimensional
grids. We showed that by restricting to the case
where the data distribution paths are transla-
tionally and rotationally symmetric, the problem
corresponds to finding Hamiltonian paths in
orbit graphs. Solutions were found for certain di-
mensions and side lengths by finding solutions
in the case a = 1, and then giving a method of
extending those solutions to the cases a > 1 .
Bibliography
[1] W. Taylor, J. Mesirov, and B. Boghosian, Optimized
local data distribution on lattice graphs (to appear).
[2] J-Ph. Brunet, A. Edelman, and J. Mesirov, Hy-
percube algorithms for direct N-body solvers, SIAM J.
Sci. Statist. Comput. 14 (1993), 1143–1158.
[3] J. Sethian, J-Ph. Brunet, A. Greenberg, and J.
Mesirov, Two-dimensional, viscous, incompressible flow
in complex geometries, J. Comput. Phys. 101 (1992),
185–206.
[4] P. Tamayo, J. Mesirov, and B. Boghosian, Paral-
lel approaches to short range molecular dynamics sim-
ulations, Proceedings of Supercomputing ‘91, IEEE
Computer Society Press, 462–470.
The Wave Equation
Revisited
Cathleen S. Morawetz
It is a surprising fact that
there is still much to
learn about the wave
equation. Beginning with
representing the solution
of the initial value prob-
lem in terms of a funda-
mental solution, it is an
easy step to see that in
3D+T Huyghens’s princi-
ple holds, i.e., if the sur-
face of the light cone
from a point backwards
Cathleen Morawetz in time does not cut the
support of the initial val-
ues, then the solution at that point is identically
zero. This will be true even if there is a pertur-
bation of the wave equation confined to the in-
side of the cone.
Cathleen Morawetz is professor emeritus at the Courant
Institute of Mathematical Sciences, New York Univer-
sity. Her e-mail address is morawetz@acfl.nyu.edu.
NOTICES OF THE AMS 37
wim.qxp 3/16/99 3:06 PM Page 38

This has led to many investigations, includ-

ing scattering theory for perturbations, and with Surface Motion Due
Huyghens’s principle gives neater results in-
cluding the exponential decay of the local energy
if the wave equation is perturbed locally in cer-
to Surface Energy
tain ways, e.g., the presence of a star-shaped re-
flecting obstacle. See (LPM) and (M).
Reduction
More recent questions are about finding Jean E. Taylor
“transparency conditions”. These are conditions Suppose one has an ir-
on a boundary which guarantee that the solution regular blob of some-
could be continued outside the boundary as a so- thing, without any exter-
lution of the wave equation. See (EM), (BT), and nal forces such as gravity
(KM) for approximate transparency conditions. acting on it. Because it is
H. Warchall (W) has shown under more general not a round ball, it does
but analytic conditions that if, and we think here not have the smallest sur-
only of the wave equation, the initial data is sup- face area for its volume.
ported in a ball of radius b , then the solution in- If it evolves so as to re-
side that ball for t > t1 > 0 depends only on the duce its surface area,
data inside the ball at t = t1. how does it evolve? What
In fact, for the simple case of the wave equa- different evolution laws
tion the transparency condition can be written are reasonable, and how
down explicitly by giving the outward charac- do the shapes differ under these different types
teristic derivative in terms of an integral over the of motion? Can you tell, by some not-too-deli-
ball of support at the same time. Conversely, if cate features of the shapes, which law is, in fact,
the solution satisfies this integral condition, governing the motion?
then it can be continued outside as a solution of For a crystalline material, such as a single
the wave equation. Cauchy integrals are involved, crystal of ice, the shape of least surface energy
and thus the solution must have three deriva- for a given volume is not a round ball, but some-
tives. thing else, like a hexagonal prism. How does a
Finally, this circle of ideas leads one to a way snowflake evolve towards this?
of avoiding transparency conditions but still The claim is that physically interesting motion
solving in a finite domain by repeated use of problems can be formulated in a variational set-
ting, as gradient flows under various natural
Huyghens’s principle.
inner products. This enables existence proofs and
Bibliography a unifying framework for different motion laws
[BT] A. Bayliss and E. Turkel, Boundary conditions
and different surface free energy functions. It
for exterior acoustic problems, Report No. 79-7, ICASE, also underlies the whole crystalline approach,
NASA Langley Research Center, 1979. which is interesting both as a model in its own
[KM] G. A. Kriegsmann and C. S. Morawetz, Nu- right and as a possible means of approximation
merical solutions of exterior problems with the reduced for other surface free energy functions, and
wave equation, J. Comput. Phys. 28, (1978), 181–197. which has been implemented by computer pro-
[LPM] P. D. Lax, C. S. Morawetz, and R. S. Phillips, grams in many cases.
The exponential decay of solutions of the wave equa-
tion in the exterior of a star-shaped obstacle, Comm. Surface Energy and Weighted Mean
Pure Appl. Math. 16 (1963), 477–486. Curvature
[M] C. S. Morawetz, Exponential decay of solutions Why does mean curvature H arise for variational
of the wave equation, Comm. Pure Appl. Math. 19 problems involving surface area? Because it is the
(1966), 439–444. rate of decrease of surface area with volume
[ME] B. Engquist and A. Majda, Absorbing boundary
under deformations. More precisely, the first
conditions for the numerical simulation of waves, Math.
variation of area of a smooth oriented surface
Comp. 31 (1977), 629–652.
S is the bounded linear operator on continuous
[W] H. A. Warchall, Wave propagation at computa-
tional domain boundaries (to appear).
vector fields g defined by
Z
d
δS(g) = 1dH 2 x,
dt x∈ht S

Jean Taylor is professor of mathematics at Rutgers

University in New Brunswick, NJ. Her e-mail address
is taylor@math.rutgers.edu.

38 NOTICES OF THE AMS VOLUME 42, NUMBER 1

wim.qxp 3/16/99 3:06 PM Page 39
where ht S = {y : y = x + tg(x) for some x ∈ S} .1
One can also compute that with n(x) denoting
the oriented normal at x and Hn the mean cur-
vature vector,
Z weighted mean curvature, R by selecting
δS(g) = g · (−Hn (x))dH 2 x.
x∈S
For solids, with crystal lattice orientations
fixed in space, the surface free energy per unit
area is a function γ from the unit sphere to the
real numbers, and it is convenient to extend the
domain of γ to all vectors by γ(p) = |p|γ(p/|p|).
If γ is C 2 and convex and S smooth,
Z and crystalline γ [AT].
d Are there other reasonable inner products
δγ S(g) = γ(n(x))dH 2 x
dt x∈ht S
Z
= g · (−κγ n)dH 2 x
x∈S
∂2 γ ∂2 γ
where κγ = κ
∂p12 1
+ κ .
∂p22 2
Here κi is the i th
principal curvature and pi the corresponding
principal direction. We call κγ the
R weighted R
mean curvature. Note that if ∆E = ht S γ − S γ
and ∆V denotes the signed volume from S to
ht S , then for g with small support around
x0 , Et ≈ δS(g) ≈ −κγ (x0 ) Vt , so κγ (x0 ) ≈ − VE .
Given any γ there is associated the Wulff
shape
Wγ = {x : x · p ≤ γ(p)∀p}.
If Wγ is polyhedral, γ is crystalline, and the first
variation is neither bounded nor linear. Instead,
the crystalline weighted mean curvature (ab-
breviated crystalline curvature) is nonlocal: one
uses the rate of decrease of surface free energy
with volume, with the deformations moving en-
tire segments. Edges and corners must be energy-
minimizing to avoid infinite contributions to
the analog of first variation. For polygonal curves,
the formula reduces to something quite simple:
κγ (segment i) = −σi Λi /`i , where `i is the length
of the segment, Λi is the length of the segment
of the Wulff shape boundary with the same nor-
mal direction, and σi is 1 if the curve makes left
turns at each end (when traveling in the direc-
tion of its orientation) is −1 if both ends make
right turns, and is 0 otherwise [T].
Inner Products and Gradient Flows
Observe that the action of the linear operator δS
on a function g is given by the L2 inner prod-
uct of g with −Hn . Thus, flow by mean curva-
ture is gradient flow for surface area in the L2
inner product. It is an unusual type of gradient
1There are better definitions that allow for singulari-
ties in S[A] .
JANUARY 1995
flow, in that the inner product space keeps
changing, as S itself keeps changing.
One can use this fastest decrease property to
construct a variational approach to motion by
a time
1
step ∆t and minimizing h#S γ + 2t h • h starting
with S0 to get St, then with St to get S2t, etc. Con-
vergence of these piecewise constant flows to a
continuous flow as ∆t goes to zero has been
shown [ATW] (with an approximately equal vol-
ume integral replacing the L2 inner product). The
resulting flow is the PDE solution to motion by
weighted mean curvature where that exists, and
is motion by crystalline curvature for curves
than the L2 inner product? Yes. In particular, mo-
tion by the negative Laplacian of weighted mean
curvature arises if one does gradient flow in the
H −1 inner product. Mullins (who also intro-
duced motion by curvature) derived that motion
by the negative Laplacian of curvature models
surface diffusion: H is a potential, D∇H gives
the flux of atoms down the gradient of the po-
tential, and −D∆H is then the accumulation of
atoms, which translates into a velocity with the
appropriate constants. For a survey of this field,
see [CT].
This type of motion is difficult to model, how-
ever. There is no maximal principle, and no level-
set (viscosity solution) formulation. However, a
crystalline formulation of this motion for curves
has been devised and implemented in a com-
puter program [CRCT].
We have also derived from physical principles
a whole family of motion laws [CT]. At one ex-
treme, where attachment kinetics are rapid and
surface diffusion is slow, there is motion by
−∆κγ . At the other, where attachment kinetics
are slow and surface diffusion is rapid, there is
motion by κγ − κav where κav is an overall av-
erage; this is L2 gradient flow restricted to de-
formations with integral 0. In between is a whole
family of motion laws, which turn out to be flow
by the inner product, which is the appropriate
linear combination of the L2 and H −1 inner
products!
Comparing Shapes
We return to the original question of whether
there are gross features of shapes moving under
various laws that can be used to distinguish one
law from another. We have made videos of com-
puter simulations of several initial shapes mov-
ing under each of the two extreme-case motion
laws discussed above. These videos can be
viewed using Mosaic [M]. We comment on the fol-
lowing observable features:
(1) In motion by κγ − κav , zigzag portions of
the curve translate as a whole, with the same ve-
NOTICES OF THE AMS 39
wim.qxp 3/16/99 3:06 PM Page 40
locity, whereas in surface diffusion, there is a
proximity effect, and each segment can have a
different velocity.
(2) Curves that are convex (or, more generally,
which have κγ nonpositive) remain so under
motion by κγ − κav , whereas long skinny por-
tions tend to develop bulbs at the end under mo-
tion by surface diffusion.
(3) Topology changes can happen under either
type of motion, but tend to be more frequent and
dramatic under motion by surface diffusion. In
particular, a simple closed curve with κγ non-
positive everywhere can break into two con-
nected components under motion by surface
diffusion.
Summary
The importance of regarding flows as being gra-
dient flows under different inner products has
several advantages. First, it shows there is noth-
ing magic about “decreasing energy fastest”—
each decreases energy fastest in its own metric.
Second, it provides a unifying approach, putting
different surface energies, different motion laws,
and even sharp versus diffuse interface ap-
proaches under one umbrella [TC]. Third, it al-
lows variational techniques to be used for mo-
tion problems. Often both elliptic and highly
nonelliptic problems can be handled in the same
proofs. And, in particular, in the crystalline con-
text the gradient flow approach provides a good
criterion for when a facet must be
stepped [CRCT].
There are also advantages to considering the
crystalline approach. Some physical materials
are faceted. One can see what is special about
area, by comparing area-reducing motions to
corresponding surface-energy reducing motions.
It is flexible; in both theory and computation, one
can alter the motion law without affecting the
basic underlying structure. It is computable: it
uses a natural parametrization, by the Gauss
map. It is a good way to measure curvature nu-
merically, and one can fairly easily detect and
make topological changes. One can use it for ap-
proximations (convergence has been proved in
some cases [G][S]). And it has a nice theory in its
own right, which is sometimes similar to that for
area-related problems and sometimes different.
A longer version of this lecture summary is
expected to appear [T2].
References
[A] W. K. Allard, On the first variation of a varifold,
Ann. of Math. 95 (1972), 417–491.
[AT] F. J. Almgren and J. E. Taylor, Flat flow is mo-
tion by crystalline curvature for curves with crystalline
energies, J. Diff. Geom. (to appear).
[ATW] F. J. Almgren, J. E. Taylor, and L. Wang, Cur-
vature driven flows: A variational approach, SIAM J.
Control Optim. 31 (1992), 387–437.
40 NOTICES OF THE AMS
[CT] J. W. Cahn and J. E. Taylor, Surface motion by
surface diffusion, Acta Metal. Mater. 42 (1994),
1045–1063.
[CRCT] W. C. Carter, A. Roosen, J. W. Cahn, and J. E.
Taylor, Shape evolution by surface diffusion and sur-
face attachment limited kinetics on completely faceted
surfaces, preprint.
[G] P. Girao, preprint, and P. Girao and R. Kohn,
preprint.
[M] In Mosaic, Open URL to http://jeeves.
nist.gov and follow the prompts.
[S] M. Soner, private communication.
[T] J. E. Taylor, Mean curvature and weighted mean
curvature, Acta Metal. Mater. 40 (1992), 1475–1485.
[T2] J. E. Taylor, Surface motion due to surface en-
ergy reduction, in elliptic and parabolic methods in
geometry (to appear).
[TC] J. E. Taylor and J. W. Cahn, Linking anisotropic
sharp and diffuse surface motion laws via gradient
flows, J. Stat. Phys. (1994) (to appear).
Soliton Equations
and Differential
Geometry
Chuu-Lian Terng
In this talk, I explained
some symplectic geo-
metric properties of har-
monic maps from the
Lorentze space R 1,1 to a
symmetric space: the ex-
istence of a sequence of
compatible sympletic
structures that make the
harmonic map equation
Hamiltonian, a hierarchy
of commuting flows, and
an action of the affine
Kac-Moody group on the space of harmonic
maps.
It is well known that most finite dimensional,
completely integrable, Hamiltonian systems can
be obtained by applying the Adler-Kostant-Symes
(AKS) theorem to some Lie algebra G equipped
with an ad-invariant, nondegenerate bilinear
form, and a decomposition G = K + N. The sym-
plectic manifold is some coadjoint N -orbit
M ⊂ K ⊥ ' N ∗ , and the equation is the Hamil-
tonian equation of the restriction of some Ad(G)-
Chuu-Lian Terng is professor of mathematics at North-
eastern University, Boston, Massachusetts. Her e-mail
address is terng@neu.edu.
VOLUME 42, NUMBER 1
wim.qxp 3/16/99 3:06 PM Page 41
invariant function f on G to M . Applying the AKS
theorem to the loop algebra of an affine Kac-
Moody algebra, we obtain many well-known soli-
ton equations and equations from differential
geometry. In fact, let G/K be a symmetric space,
G = K + P the Cartan decomposition, A a max-
imal abelian subaglebra in P , P 1 = A⊥ ∩ P ,
a ∈ A a fixed regular element, and b ∈ A . We
obtain a sequence of symplectic structures
{ωr | r ∈ Z} on the space of Schwartz functions
M = S(R, P 1 ) , functionals {Fb,n : M→R| n ≥ 1},
and evolution equations, the n-th flow associated
to G/K with respect to a, b, ut = [Qb,n (u), a] with
n = −1 or n ≥ 2 such that
(1) all ωr are compatible. That is, {, }r + µ{, }s
is again a Poisson structure on M for any
µ ∈ R and r , s ∈ Z , where {, }r is the Poisson
structure corresponding to ωr ,
(2) the Hamiltonian equation for Fb,n with re-
spect to ωr is ut = [Qb,n+r +1 (u), a] ,
(3) Fb,n are commuting Hamiltonians with re-
spect to ωr for all n, r .
These flows contain many interesting equa-
tions. For example, the equation for harmonic
maps from R 1,1 to G/K is the −1 -flow associ-
ated to G/K, the nonlinear Schrödinger equation
is the third flow associated to SU(2), the Modi-
fied KdV equation is the fourth flow associated
to SU(2)/SO(2) , and the Beals-Coifman evolution
equations are the n-th flows associated to a
compact Lie group G with n ≥ 2 .
There is also a system of partial differential
equations of n variables naturally arising from
the second flow associated to a rank n sym-
metric space. For example, the natural PDE sys-
tem associated to
SO(2n + 1)/SO(n) × SO(n + 1),
SO(2n)/SO(n) × SO(n)
and SO(2n, 1)/SO(n) × SO(n, 1)
is the equation for isometric immersions from
N n (c) into N 2n (c) with flat normal bundle and
independent curvature normals with c = 1, 0, −1
respectively, where N n (c) is an n-dimensional
space form with sectional curvature c .
For these sequence of flows we also obtain
analogues of K. Uhlenbeck’s loop group action
on the space of harmonic maps from R 1,1 into
Lie group and the classical Bäcklund transfor-
mations for the Sine-Gorden equation.
JANUARY 1995
Moduli Spaces and
Adiabatic Limits
Karen K. Uhlenbeck
The last ten years of dif-
ferential geometry have
been dominated by the
study of very special
geometric partial differ-
ential equations. The
equations I have in mind
include the minimal and
constant mean curvature
equations, the Yamabe
equation, the Kähler-Ein-
stein equation, harmonic
maps, and Yang-Mills
with all its variations.
The origins of the equations vary tremendously,
as they come from both very classical geometry
and modern high-energy physics, but the equa-
tions themselves are clearly very special. One of
the interesting and useful features of most of the
equations is the “large”, by which we mean multi-
dimensional, solution spaces which are regu-
larly found for these equations in special cir-
cumstances.
By and large the origin of the equations of
geometry is intimately connected with varia-
tional principles. The classical equations are
usually posed as minimization problems. It is a
hard and fast rule of physics that interesting
equations are the “equations of motion” for
some integral with a classical Lagrangian. In
these cases, the Euler-Lagrange equation, which
is generated by the variational principle, is nearly
always second order. However, for technical rea-
sons, the second variation is self-adjoint, and
hence has zero index if it can be represented as
a Fredholm operator between Banach spaces.
This means that formally speaking, the dimen-
sion of a critical set, or the solution space of a
geometric equation, tends to be zero dimen-
sional.
There are a multitude of origins for the phe-
nomena of “nonaccidental” solution manifolds
of large dimension. By “nonaccidental” we mean
that a small perturbation which does not change
the nature of the problem will not destroy them.
The most common cause is the existence of a
continuous family of symmetries which leaves
the equation invariant. This is closely allied to
the phenomena of “integrals” which is important
Karen Uhlenbeck is Regents’ Chair in Mathematics at
the University of Texas at Austin. Her e-mail address is
uhlen@math.utexas.edu.
NOTICES OF THE AMS 41
wim.qxp 3/16/99 3:06 PM Page 42
for mean curvature surfaces. A second philo-
sophically separate cause is the reduction of the
original variational problem, with its second-
order geometric equation, to a first-order equa-
tion for a topological minima. This mechanism
reduces the second-order harmonic map equa-
tion from a Riemann surface into an almost com-
plex symplectic manifold to a pseudo-holomor-
phic curve equation, and Yang-Mills to self-dual
Yang-Mills. The index of the first-order equa-
tion obtained in this fashion is topologically de-
termined and typically not necessarily zero. A
final and not-so-well-known origin for multi-di-
mensional parameter spaces is the introduction
of point singularities, which changes the index.
The Yamabe equation, which has zero index on
a compact manifold, has index k on a manifold
with k punctures.
An optimum strategy for progress includes
the search for new equations, as well as techni-
cal progress on the well-studied equations. An-
other angle is to search for new uses of these al-
ready-studied moduli spaces of solutions to
geometric partial differential equations. We de-
scribe how a reduction technique can result in
the moduli space of solutions to one problem as
the target manifold for more geometry. Credit
for this idea, at least in its modern form in an
application to monopoles, is due to the physi-
cist Nick Manton. Here is a simplified version.
Consider the question of locating long, slow,
closed orbits of the classical mechanics problem
of a point mass moving in Rn with potential en-
ergy V (x) and kinetic energy 12 |ẋ|2. To make the
problem interesting we must have
M = {x ∈ Rn : dV (x) = 0}
a nontrivial manifold, and technically speaking
we assume that M is a nondegenerate critical
manifold in the sense of Bott. The action for the
problem is ZT 

1 2
J(x) = 2 |ẋ| − V (x) dt
0 Lett. B 154 (1985).
with x(0) = x(T ). In our search for long, slow,
closed orbits we renormalize t = T τ to get
As T → ∞, ε → 0 and we hypothesize that there
is a smooth family of solutions x(ε, τ) which for
each ε solves (*), and which has a limit as ε → 0.
We call this orbit at ε = 0 and adiabatic limit. The
following theorem is cute, if nonobvious, and the
proof can be done with advanced calculus.
Theorem. A necessary condition for x(τ) to be
an adiabatic limit for the original mechanics
problem is that x is a closed geodesic on M .
By replacing Rn by a function manifold and
V : Rn → R by one of the geometric integrals of
the type with a moduli space of critical points,
one finds adiabatic approximations for hyper-
bolic equations. It is also possible to replace t
by static parameters to obtain “adiabatic” lim-
its of static problems. Proving that the adiabatic
limit has some physical or geometric reality is
much more difficult.
References
[1] M. Atiyah and N. Hitchen, Geometry and dy-
namics of magnetic monopoles, Princeton Univ. Press,
1988.
[2] S. Dostoglou and D. Salamon, Self-dual instan-
tons and holomorphic curves, Ann. of Math., University
of Warwick, preprint, 1992.
[3] ———, Cauchy-Riemann operators, self-duality,
and the spectral flow, Proceedings of the First European
Congress of Mathematics, Paris, 1992, preprint, Uni-
versity of Warwick, 1992.
[4] R. A. Leese, Low-energy scattering of solitons in
the CP 1 model, Nuclear Phys. B 344 (1990), 33.
[5] Leese, Peyrard, and Zakrzewski, Soliton scatter-
ing in some relativistic models in (2 + 1) dimensions,
Nonlinearity 3 (1990), 387.
[6] Leese, Peyrard and Zakrzewski, Soliton stability
in the O(3) sigma model in (2 + 1) dimensions, Non-
linearity 3 (1990), 773.
[7] N. Manton, A remark on the scattering of BPS
monopoles, Phys. Lett. B 110 (1982), 54–56.
[8] ——— , Monopole interactions at long range, Phys.
[9] R. Montgomery and L. Bates, Closed geodesics on
the stable two-monopoles, Comm. Math. Phys. 118
Z1  2 ! (1988), 635–640.
1  
−2  dx  [10] D. Stuart, Dynamics of abelian Higgs vortices
J(x) = 2 T  dτ  − V (x) T dτ. in the near Bogomolny range, Comm. Math. Phys.
0
[11] ———, The geodesic approximation for the Yang-
Let ε = T −2 . In renormalizing time τ, the Euler- Mills-Higgs Equation, preprint, University of California,
Lagrange equations are Davis, 1993.
    [12] R.S. Ward, Slowly-moving lumps in the CP 1
d 2
(*) ε x(τ) + dV x(τ) = 0. model in (2 + 1) dimensions, Phys. Lett. 185 (1985),
dτ 424–428.

42 NOTICES OF THE AMS VOLUME 42, NUMBER 1