triangular_billiards_ergodicity_v2.txt
Ergodicity of triangular billiards with an irrational angle
===========================================================
A complete, self-contained, high-school-friendly walkthrough.
The paper explained: "Ergodicity of triangular billiards with an
irrational angle" (the LaTeX source, and all references cited below,
live in triangular_billiards_ergodicity_original.txt in this same
directory). The theorem, in one sentence: every nondegenerate Euclidean
triangle that has at least one irrational angle (in units of pi) has an
ergodic billiard flow. No genericity assumption, no Diophantine speed
condition, no exceptions.
You need to be comfortable with:
- basic triangle geometry and angles;
- the fact that any nice periodic function of one variable can be
written as a sum of sines and cosines (a Fourier series; reviewed
when needed);
- the idea that a "state" of a moving particle is (position, direction),
not just position.
The main measure-theoretic words (measure zero, invariant set, ergodic)
are defined where they first show up.
The format is a sequence of "Exercises". Each is a question you should be
able to answer after reading it, then the answer. Facts proved in the
original paper are marked "proof idea"; standard background is marked
"standard".
Exercise 1. What is a billiard?
A flat polygonal table. A small ball rolls at constant unit speed. When
it hits a side, it reflects specularly: the angle it makes with the wall
on the way in is the angle it makes on the way out; the speed is
unchanged. Between hits the path is a straight line. That is the
billiard flow. For this writeup the table is a triangle Q. No pockets,
no friction, no gravity. States are (position in Q, direction in the
circle S^1 = [0, 2pi)).
Exercise 2. What does "ergodic" mean concretely?
Let A be a subset of state space. Call A invariant if a trajectory that
starts inside A stays inside A forever (and a trajectory that starts
outside stays outside). The flow is ergodic if every measurable
invariant set has measure either 0 or equal to the whole measure. A
quantity you would expect to be conserved "by accident" cannot exist.
Equivalently (Birkhoff's ergodic theorem, 1931): for any continuous F on
state space, for almost every starting state, the long-run time average
of F along the trajectory equals the space average of F.
time average = space average for a typical trajectory.
Caution: ergodicity is NOT the same as "every trajectory is dense". It
is a statistical statement about almost every trajectory. Individual
trajectories can be unusual (e.g. a trajectory that stays on a straight
line parallel to a median forever, in a symmetric triangle). Ergodicity
says the set of such special initial states has measure zero.
Exercise 3. Why is this interesting?
Billiards are the simplest mechanical systems that are genuinely
chaotic. An ergodic billiard is the paradigm of "kinetic theory". In
mathematics, ergodicity of the billiard flow implies strong statements
about the eigenfunctions of the Laplacian of the table (the quantum
billiard): the wavefunctions "see" the whole geometry. This is the
subject called quantum chaos. And for triangular tables, ergodicity is
exactly the part of the story that no soft / generic / category argument
can resolve; the exception cases are famous.
Exercise 4. State the theorem precisely.
Let Q be the interior of a nondegenerate Euclidean triangle with angles
alpha, beta, gamma (in radians, so alpha + beta + gamma = pi). Suppose
that at least one of alpha/pi, beta/pi, gamma/pi is irrational. Let
Phi_t be the unit-speed billiard flow on Q x S^1 with specular reflection
at the three open sides. Discard the measure-zero set of states whose
trajectory hits a vertex in either time direction. Then Phi_t is ergodic
with respect to the probability measure
dmu(p, theta) = dArea(p) dtheta / (2 pi Area(Q)).
Equivalently: if a measurable A subset of Q x S^1 satisfies
mu( (Phi_t)^{-1} A symmetric-difference A ) = 0 for every real t, then
mu(A) is 0 or 1.
Reference: main paper, Theorem "main", triangular_billiards_ergodicity_original.txt.
Exercise 5. Which triangles does the theorem cover, and which does it not?
Because alpha/pi + beta/pi + gamma/pi = 1, EXACTLY ONE of the ratios being
irrational is impossible (a rational plus 1 irrational would equal 1, but
1 is rational). So the possible count of irrational ratios is 0, 2, or 3.
The theorem covers:
- 3 irrational ratios: e.g. alpha/pi = 1/sqrt(10) ~ 0.316,
beta/pi = 1/sqrt(3) - 1/sqrt(5) ~ 0.130, gamma/pi ~ 0.554. Sum = 1.
- 2 irrational ratios + 1 rational: e.g. alpha/pi = 1/4, beta/pi = 1/e,
gamma/pi = 3/4 - 1/e ~ 0.382. Sum = 1.
The excluded case is 3 rational ratios: equilateral, 45-45-90,
30-60-90, and any rational-multiple-of-pi triangle.
Exercise 6. Why is the 3-rational case special?
Then the triangle is a "rational polygon". Reflecting the triangle across
its sides (see Exercise 7) generates a finite group of planar isometries;
the trajectory, unfolded, becomes a straight line on a compact flat
surface. There Kerckhoff, Masur and Smillie (1986) proved ergodicity in
almost every direction. For the irrational case, older arguments gave:
- a dense G_delta subset of triangles (Baire category, KMS 1986);
- a quantitative sufficient condition on how fast the angles can be
simultaneously approximated by rationals (Vorobets 1997), with
explicit irrational right-triangle examples;
- dense G_delta weak mixing of the flow for each fixed n-gon (Chaika-
Forni, Annals 2026).
The phrase "dense G_delta" means: a residual subset. NOT one triangle at
a time. The current theorem is pointwise: it names every triangle with
irrational angle. The gap between "residual" and "all" was the open
problem.
Exercise 7. Why can we not just unfold triangles the way the rational case
unfolds polygons?
Unfolding: when the trajectory meets a side, instead of reflecting,
continue it into a mirror-image copy of the triangle. In a rational
polygon the group generated by reflections is finite (or has a
finite-index translation lattice), so finitely many reflected copies
suffice and the trajectory becomes a straight line in a compact flat
surface. If any angle is irrational, the reflection group is infinite
and dense in the similarity rotations; there is no finite family of
copies in which the trajectory is a straight line. Every piece of KMS
machinery (translation surfaces) is lost.
Exercise 8. What replaces unfolding for an irrational triangle?
Only TWO copies are needed: the triangle glued to its mirror image along
the boundary. The result, M, is a compact flat surface with three cone
points, and the billiard trajectory on Q lifts to a straight-line
geodesic on M. This is the "double" of the triangle. It is always
compact, whatever the angles. This is the central technical device.
Exercise 9. How is the double built?
Take two copies of the closed triangle Q with OPPOSITE orientations (one
with the usual orientation, one mirrored). Glue copy 1 to copy 2 side by
side: each side of copy 1 glued to the corresponding side of copy 2 with
the opposite orientation. Result: a closed, connected, flat surface M
without boundary except at three "cone points" located above the three
vertices of Q. Away from those three points the metric is flat, meaning
locally M looks exactly like a piece of the plane. At the vertex where
the triangle has angle alpha, the cone angle on M is 2 alpha: two
triangle-corners of angle alpha back to back.
Topological footnote: M is homeomorphic to the 2-sphere for a triangle,
with three marked cone points.
Exercise 10. How does the billiard trajectory on Q become a flow on M?
When a trajectory hits an edge of Q, instead of reflecting, let it
continue STRAIGHT into the other copy of the triangle. On M this is a
smooth geodesic: no reflection, no kink. The unit-speed billiard flow
on Q is thus, one-to-one (outside the vertex-hit null set), the
unit-speed geodesic flow on M. What was "the reflection rule" becomes
"the gluing rule" on the two sheets of M.
Exercise 11. What does "cone angle 2 alpha" mean geometrically?
Cut out a small disk around the cone point p on M. Unroll that disk onto
a plane. The result is a circular sector with total angle 2 alpha at the
apex (rather than 2 pi for an ordinary flat disk). Around p the surface
looks like the tip of a paper cone. All three cone angles 2 alpha,
2 beta, 2 gamma lie strictly between 0 and 2 pi, so all three are genuine
"pointed" cones. Gauss-Bonnet check (for the curious): the total
"curvature defect" at the three cones is 3 * 2pi - (2 alpha + 2 beta +
2 gamma) = 6 pi - 2 pi = 4 pi, matching the 4 pi you need on a sphere.
Exercise 12. What is the phase space and the invariant measure?
States live in X = M x S^1: a state is (p, theta), p on M, theta in
[0, 2 pi) the direction of motion. The flow is
Phi_t(p, theta) = (point at distance t from p in direction theta, same theta),
continued across the cone points by the flat geodesic motion (undefined
only on the null set of vertex hits). The measure is dArea(p) dtheta,
normalized so total measure = 1. This is the mu of the main theorem.
Exercise 13. Why is the set of trajectories that hit a vertex of measure
zero?
A trajectory hits a vertex iff, for some finite sequence of sides
crossed, an unfolded straight segment ends exactly at a point (the
vertex). For fixed sequence, this is ONE scalar condition on
(position, direction) (three scalars total, one constraint), so the
affected set is codimension 1 in X. The countable union over all
sequences is therefore null. Removing a null set does not change any
measure-theoretic question, so we may simply exclude vertex hits.
Exercise 14. Rephrase the theorem in the form we will actually prove.
Write X for the geodesic-flow vector field on X = M x S^1 (the vector
field whose flow is Phi_t). Then the flow is ergodic IF AND ONLY IF every
measurable f in L^inf(X) with
X f = 0 (in the distributional sense on M x S^1)
satisfies f = constant mu-almost everywhere. Reason: for any measurable
invariant set A, the indicator f = 1_A has Xf = 0 tautologically; and if
every Xf = 0 bounded function is constant, the only invariant
indicators are 0 and 1. Conversely, if f = 1_A were nonconstant with
Xf = 0, the level sets would split A into an invariant partition.
So the whole theorem reduces to: "no nontrivial bounded invariant
functions".
Exercise 15. What is the "Y" vector field, and why introduce it?
Fix a state (p, theta) and look at the phase space. X = "move in the
direction theta" is the flow direction. Define Y = "rotate the motion
direction by a quarter turn" - still a horizontal vector field on the
flat surface, still perpendicular to the flow direction in the Euclidean
metric of the tangent plane. In flat coordinates z = x_1 + i x_2 and
direction angle theta, both X and Y can be written in terms of the
complex derivatives
d/dz = (1/2)(d/dx_1 - i d/dx_2), d/d zbar = (1/2)(d/dx_1 + i d/dx_2),
as
X = e^{i theta} d/dz + e^{-i theta} d/d zbar,
Y = i e^{i theta} d/dz - i e^{-i theta} d/d zbar.
Both are real vector fields on M (the theta-dependent complex
coefficients combine). They are perpendicular in the flat metric and
they commute: [X, Y] = 0. Introducing Y is useful because it "rotates"
the directions of the flow into each other, giving an extra degree of
freedom when we integrate by parts.
Reference: Forni-Moll 2025 (arXiv:2510.18128), Sections 3.2-3.5, for
the flat-surface geodesic-flow framework. Also triangular_billiards_ergodicity_original.txt.
Exercise 16. Why look at the angle theta with a Fourier series?
Because the flow does not change theta: particles move at a fixed
direction until they hit a cone (they never do in the interior). So for
the purposes of differentiating along the flow, theta is a parameter. On
the other hand, theta is periodic (circle S^1), so any function
f(p, theta) in L^2(X) has a Fourier expansion in theta:
f(p, theta) = sum_{j in Z} f_j(p) e^{i j theta}.
The sum is in L^2, coefficients are complex-valued functions f_j on M,
and Parseval reads ||f||_{L^2(X)}^2 = sum_j ||f_j||_{L^2(M)}^2. This is
just the usual decomposition of a periodic function into waves.
The coefficient f_j(p) = (1/2 pi) int_0^{2 pi} f(p, theta) e^{-i j theta} dtheta.
Standard. (Any analysis book, or just: sin/cos Fourier series.)
Exercise 17. What is the geometric meaning of the twist of the coefficients?
The angle theta is not defined globally on M: at every cone point, the
"flat frame" rotates by the cone angle when you go around. Concretely,
if a local chart rotates by an angle phi when comparing two overlapping
charts, then the same point has coordinates (p, theta) and (p, theta +
phi), so the coefficient f_j transforms as
f_j -> e^{-i j phi} f_j.
In other words, f_j is a SECTION of a flat unitary line bundle (the line
bundle L^{-j}, tensor power of the inverse of the tautological bundle of
directions). It behaves like a function LOCALLY, but jumps by the phase
e^{-i j phi} across chart changes. Only f_0 is an honest function on M.
Practical consequence: differentiating f_j by d/dz or d/d zbar produces
sections of slightly different bundles (L^{-j+1} and L^{-j-1}); that is
the source of the mode-shifting identities in Exercise 18.
Reference: triangular_billiards_ergodicity_original.txt, section "The
double and angular coefficients"; Forni-Moll 2025, sections 3.2-3.5.
Exercise 18. What are the mode-by-mode formulas for X and Y?
Plug the Fourier expansion into the definitions. For u = u_j(p) e^{i j theta}:
X u = (d u_j) e^{i (j+1) theta} + (dbar u_j) e^{i (j-1) theta},
Y u = i (d u_j) e^{i (j+1) theta} - i (dbar u_j) e^{i (j-1) theta}.
So the n-th Fourier coefficient of Xu is
(Xu)_n = d u_{n-1} + dbar u_{n+1},
and the n-th Fourier coefficient of Yu is
(Yu)_n = i d u_{n-1} - i dbar u_{n+1}.
Here d = d/dz and dbar = d/d zbar are the complex derivatives of
Exercise 15. Each of X and Y shifts the Fourier index by plus or minus
1. Adjacent coefficients are coupled by derivatives.
Reference: triangular_billiards_ergodicity_original.txt, Equation (2.4) / (eq:modes).
Exercise 19. What does Xf = 0 translate to in the Fourier expansion?
For every integer m, the m-th Fourier coefficient of Xf vanishes:
d f_{m-1} + dbar f_{m+1} = 0 (as distributions on M).
Equivalently, for every integer k:
d f_k = -dbar f_{k+2}.
Every derivative d of a coefficient equals minus the dbar of the
coefficient two steps to the right. This single identity encodes
"invariance" completely. It is a LINEAR RECURRENCE connecting adjacent
angular modes through derivatives.
Reference: triangular_billiards_ergodicity_original.txt, discussion around (eq:modes).
Exercise 20. Why is Xf = 0 not enough on its own? What exactly is the
obstacle?
Suppose we tried to conclude d f_m = 0 immediately. The recurrence says
d f_m = -dbar f_{m+2}, so we would need to know what dbar f_{m+2}
"really is". Differentiating again might loop (because d and dbar do not
commute with themselves across modes: applying the recurrence at shifted
indices gives relations between dbar f_{m+2} and d f_{m+4}, etc.) On the
level of formal series there is a free "initial" coefficient in each
class of modes mod 2. To kill that free coefficient, we must integrate
by parts - i.e. use that f is a fixed function on M, not a formal
sequence.
The obstacle: f is only in L^inf. Its Fourier coefficients f_j are in
L^2(M) but have no a priori spatial derivatives. Any argument that
integrates by parts requires at least SOME regularity. We must supply
it.
Exercise 21. Where are the derivatives supposed to come from?
The flow direction and the perpendicular direction are both "horizontal"
and commute. If f were smooth, Xf = 0 and Yf (also horizontal)
would provide extra identities between neighbouring coefficients, and
integration-by-parts in the transverse direction would give gradient
estimates. The proof DOES run that argument, but only after a two-step
approximation in which the function is (i) smoothed in position, (ii)
truncated to a long time interval, and then (iii) restricted to a region
away from the cone points. The estimates are designed so that neither (i)
nor (ii) nor (iii) degrades them, and the final limit takes a
regularized quantity to an exact one.
Exercise 22. Define the "tail" U and the "endpoint" B.
Fix an integer m. Let P_j denote projection onto the j-th Fourier mode.
Define
U := sum_{k >= 0} P_{m+2k} f,
B := e^{i (m-1) theta} dbar f_m.
So U is f restricted to a HALF-INFINITE arithmetic progression of modes
(m, m+2, m+4, ...) with fixed parity m mod 2 and j >= m; B is a single-
mode field with coefficient dbar f_m.
Reference: triangular_billiards_ergodicity_original.txt, (eq:tail).
Exercise 23. What is XU?
Compute XU mode by mode. For the n-th Fourier coefficient:
(XU)_n = d U_{n-1} + dbar U_{n+1}.
U_{n-1} is nonzero only for n in {m+1, m+3, m+5, ...} (from d-term);
U_{n+1} is nonzero only for n in {m-1, m+1, m+3, ...} (from dbar-term).
So:
- At n = m-1: only the dbar-term contributes, dbar U_m = dbar f_m.
- At n = m+1: (XU)_{m+1} = d f_m + dbar f_{m+2}.
- At n = m+3: (XU)_{m+3} = d f_{m+2} + dbar f_{m+4}.
- etc.
But Xf = 0 says exactly d f_{m+2k} + dbar f_{m+2k+2} = 0 for every k >= 0
(Exercise 19 with index 2k+1). So all terms at n >= m+1 vanish.
Conclusion:
XU = B = e^{i (m-1) theta} dbar f_m (distributionally).
The endpoint contribution at mode m-1 is the ONLY surviving term: the
rest of the arithmetic progression cancels pairwise using Xf = 0.
Reference: triangular_billiards_ergodicity_original.txt, (eq:endpoint).
Exercise 24. Why does proving B = 0 close the whole proof?
By Exercise 19 at index m-1: d f_{m-2} + dbar f_m = 0, so d f_{m-2} =
-dbar f_m. Now compute (Yf)_{m-1} using Exercise 18 at n = m-1:
(Yf)_{m-1} = i d f_{m-2} - i dbar f_m
= i (-dbar f_m) - i dbar f_m
= -2 i dbar f_m.
Hence the single-mode field P_{m-1} Y f satisfies
P_{m-1} Y f = -2 i B.
If B = 0, then dbar f_m = 0. Since m was arbitrary, dbar f_m = 0 for
every integer m; and then d f_k = -dbar f_{k+2} = 0 for every k. So
every f_k has vanishing d and dbar everywhere. Because M is connected,
each f_k is CONSTANT on M. The rest of the proof is: use the angle to
show f_k = 0 for all k != 0, and f_0 = overall constant.
Exercise 25. State the target proposition (the rigidity step).
If f in L^inf(X) satisfies Xf = 0 in distributions, then for every j:
d f_j = 0 and dbar f_j = 0 (as distributions on M).
This is the Proposition in the paper's section "From flow invariance to
parallel coefficients". The proof occupies Exercises 26-43. The
conclusion is exactly what Exercise 24 needs.
Reference: triangular_billiards_ergodicity_original.txt, Proposition.
Exercise 26. What is the smoothing operator S_eps?
For eps > 0 small and a small smooth nonnegative radial "bump" rho on
the unit disk of R^2 with integral 1, set
rho_eps(y) = eps^{-2} rho(y/eps).
If x in M and the ball of radius 4 eps around x is embedded in M (flat
patch, away from cones), define
(S_eps u)(x, theta) := int_{R^2} rho_eps(y) u(x - y, theta) dy.
In words: SPATIAL average of u over a small disk of radius eps, keeping
the DIRECTION theta frozen. It is NOT an average over theta.
Notation: r(x) := distance from x to the nearest cone point (in M). Let
K be a fixed constant such that the ball of radius 4 eps around x is
embedded whenever r(x) > K eps. Such a K exists because near a cone
point the embedded radius is c r(x) for some c > 0, and away from cones
compactness gives a uniform lower bound.
Reference: triangular_billiards_ergodicity_original.txt, section
"Smoothing away from vertices".
Exercise 27. State the smoothing lemma.
The operator S_eps has the following properties (all for small eps):
(a) SAFE-REGION L^2 BOUND: ||S_eps u||_{L^2(r > K eps)} <= ||u||_{L^2}
and ||eps * del S_eps u||_{L^2(r > K eps)} <= C ||u||_{L^2}.
(b) SMALL SCALED DERIVATIVE: for each fixed u in L^2(X),
||eps * Y S_eps u||_{L^2(r > K eps)} -> 0 as eps -> 0.
(c) APPROXIMATE IDENTITY: on compact regular subsets of M,
S_eps u -> u in L^2.
(d) SUPPORT AND ADJOINT: fix L > K + 4. If g in L^2(X) vanishes on
{r < L eps}, then S_eps g is supported in {r >= (L-1) eps} and
extends by zero to a field with compact spatial support. For u in
L^2(X),
<Y S_eps u, g> = -<u, Y S_eps g>,
with both sides well-defined (left side uses only safe centers).
(e) COMMUTATION WITH X: if Xg in L^2 distributionally, then
X (S_eps g) = S_eps (Xg), and ||X S_eps g||_2 <= C ||Xg||_2, using
the same compact-support extension.
Reference: triangular_billiards_ergodicity_original.txt, Lemma "smoothing".
Exercise 28. Why does the smoothing lemma work? (Intuition and one-line
proof idea.)
In a flat chart containing the convolution disk, S_eps is ordinary
spatial convolution with the bump. Its kernel has integral 1, and the
kernel of eps * Y S_eps has integral O(1). Standard convolution
inequalities give (a). For (b), on smooth compactly supported inputs the
integrand is differentiably nicely, but the scaled derivative at scale
eps tends to 0 as eps -> 0 (roughly: |del rho_eps| * eps is O(1) but the
support is shrinking); density of smooth inputs then handles general u
in L^2. For (c), ordinary approximate-identity. For (d), a contributing
kernel pair sits within eps; if the center is in the support of g, both
sits strictly inside the safe region; reversing the short segment gives
the antisymmetry Y_x K(x,y) = -Y_y K(x,y), and Fubini does the rest.
For (e), S_eps commutes with local vector fields in a flat chart, and the
L^2 bound follows from (a).
No fancy Sobolev spaces are involved. The whole smoothing happens on the
SAFE REGION {r > K eps}, which shrinks to the complement of the cone
points as eps -> 0, but each smoothing step touches only truly flat
geometry.
Exercise 29. State the localized energy lemma (the "integration by
parts" core).
Let w in L^2(X) have compact spatial support in M (away from cone points,
in the sense that the support is contained in M \ V minus a
neighbourhood of V). Suppose Xw and Yw are in L^2(X) (distributionally),
so d w_j and dbar w_j are in L^2 for every j. Let E subset of M be any
measurable set such that Xw = 0 on M \ E. Then for every integer j:
||del w_j||_2^2 <= 4 * ||Xw||_2 * ||1_E del w||_2,
where del is the flat gradient on M, and 1_E del w means the L^2
function equal to del w on E and 0 elsewhere.
Reference: triangular_billiards_ergodicity_original.txt, Lemma "energy".
Exercise 30. Sketch the proof of the energy lemma.
Two ingredients.
1) CR equal norms: for each j, the L^2 norms of d w_j and dbar w_j
agree and equal (1/4) ||del w_j||_2^2. (Check: for a compactly
supported complex function w_j, the 2-form d w_j wedge dbar w_j is
exact, so it integrates to zero over M; algebra converts that to
equality of the two norms.) Define D_j := ||d w_j||_2^2 =
||dbar w_j||_2^2 = (1/4)||del w_j||_2^2.
2) Telescoping recurrence. From the mode identity, for every j:
D_j - D_{j+2} = Re <(Xw)_{j+1}, d w_j - dbar w_{j+2}>.
Since d w_j and dbar w_{j+2} are both in L^2, and (Xw)_{j+1} is in
L^2 (because Xw is), Cauchy-Schwarz applies. Sum over k >= 0 and
telescope:
D_j = sum_{k>=0} (D_{j+2k} - D_{j+2k+2}) <= ||Xw|| * ( sum_k ||1_E(d w_{j+2k} - dbar w_{j+2k+2})||^2 )^{1/2}.
The term D_{j+2N} tends to 0 as N -> infinity (Cauchy criterion for
sum_j D_j = (1/4)||del w||_2^2 < infinity). Inside the E-restriction,
use |a - b|^2 <= 2|a|^2 + 2|b|^2 and Parseval in theta, together
with the fact that Xw is supported on E (so the (Xw)_{j+1}
coefficients are automatically supported on E). That bounds the
remaining factor by ||1_E del w||. Substituting D_j = (1/4)||del
w_j||_2^2 gives the claimed constant 4 (or slightly better; the
constant does not matter).
There is nothing about individual coefficient derivatives beyond CR
norms and splitting into half-infinite sequences. Same idea as getting
L^2 gradient bounds from a single directional derivative, only applied
mode-by-mode.
Reference: triangular_billiards_ergodicity_original.txt, proof of the
energy lemma.
Exercise 31. State the bounded-modes lemma.
Let h in L^inf(X) satisfy Xh = 0 distributionally, and let H :=
||h||_{L^inf}. Then EVERY Fourier coefficient h_j has a global L^2
gradient, and
||del h_j||_2 <= C H (uniform in j),
with C depending only on the fixed triangle and the choice of bump.
This is the KEY robustness statement: the SIZE of the coefficient
gradient is controlled by the L^inf norm alone, with no dependence on j
and no reliance on the number of derivatives of h.
Reference: triangular_billiards_ergodicity_original.txt, Lemma
"bounded-modes".
Exercise 32. Sketch the proof of the bounded-modes lemma.
Fix a smooth radial cutoff chi_eps with
chi_eps = 0 on {r <= (K+1) eps},
chi_eps = 1 on {r >= (K+2) eps},
|del chi_eps| <= C / eps.
Define w := chi_eps S_eps h, extended by 0 near the cone points. Then:
- w has compact spatial support in M \ V.
- Xw = (X chi_eps) * (S_eps h), supported in the transition annuli
E_eps = {(K+1) eps < r < (K+2) eps}. Area(E_eps) = O(eps^2).
- On E_eps, |Xw| <= C H / eps and |del w| <= C H / eps (since
||S_eps h||_inf <= H, |X chi_eps|, |del chi_eps| <= C / eps).
- Applying the energy lemma (Exercise 29) with E = E_eps:
||del w_j||_2^2 <= 4 * ||Xw||_2 * ||1_E del w||_2
<= 4 * C H * C H = C' H^2.
Hence ||del w_j||_2 <= C H, UNIFORMLY in eps and j.
- Let eps -> 0. Along a subsequence, w -> h in L^2(M x S^1) on compact
subsets; w_j -> h_j in L^2 on compact regular sets. The gradients
del w_j are uniformly bounded in L^2, so they have weakly convergent
subsequences; testing against smooth compactly supported fields
identifies the weak limit with del h_j. By lower semicontinuity of
norms under weak limits, ||del h_j||_2 <= C H.
This repeats for any fixed j. Since the bound is uniform in j, it applies
to any coefficient we like.
Reference: triangular_billiards_ergodicity_original.txt, proof of the
bounded-modes lemma.
Exercise 33. Set up the rigidity proof.
Fix m in Z. Everything that follows concerns the endpoint field B and the
tail U of Exercise 22:
U = sum_{k >= 0} P_{m+2k} f, B = e^{i (m-1) theta} dbar f_m.
The goal is to show <B, P_{m-1} Y f> = 0. By Exercise 24, that suffices
to prove B = 0, and hence to conclude dbar f_m = 0 (and by taking m
arbitrary, all derivatives vanish).
The proof runs in THREE stages:
1. Set up a "long-segment average" g_eps of f, regular enough to
differentiate and close to f in L^2, but supported away from cones.
2. Integrate the smoothed equation X (Y S_eps U) = Y S_eps B along
such segments, obtaining <B, Y (S_eps g_eps)> = o(1).
3. Take two limits: first eps -> 0 at fixed s, then s -> 0.
Exercise 34. What is the long-segment average g_eps?
Fix s > 0 (a small scale to be shrunk later). Put
T := s / eps (the "long time", much longer than eps),
and choose L > K + 4 as in Exercise 27(d). Define the "good" set of
starting states
G_eps := {z in X : r(Phi_t z) > L eps for every 0 <= t <= T},
i.e. the trajectory over the next T seconds stays at least L eps away
from the three cone points. On its complement the trajectory gets close
to a cone. Finally, average f backward over the segment:
a_eps := f * 1_{G_eps},
g_eps := (1/T) int_0^T (a_eps o Phi_{-t}) dt.
So g_eps is an ordinary spatial function of the state z, equal to the
average of f over backwards flow for time T, but only on states whose
trajectory stays clear of cones. The support of g_eps is away from
cones: if g_eps(z) != 0, then for some t in [0, T] we have
Phi_{-t} z in G_eps, i.e. r(z) > L eps at z = Phi_t(Phi_{-t} z). So
g_eps vanishes on {r < L eps}.
Reference: triangular_billiards_ergodicity_original.txt, (eq:average).
Exercise 35. What are the quantitative bounds on g_eps?
Write H := ||f||_{L^inf}. Then:
(a) ||g_eps||_{L^inf} <= H (obvious: it is an average of numbers in
[-H, H]).
(b) ||g_eps - f||_{L^2} <= C H sqrt(s + eps^2) (small when s and eps
are small).
(c) X g_eps = (a_eps - a_eps o Phi_{-T}) / T in the weak sense,
hence ||X g_eps||_{L^2} <= 2H / T = 2H eps / s.
Proof of (b): By flow invariance, f = f o Phi_{-t} for all t. So
g_eps - f = (1/T) int_0^T ( (a_eps - f) o Phi_{-t} ) dt.
The integrand is zero on G_eps and equals -f o Phi_{-t} on its
complement, so its absolute value is at most H * (1_{X - G_eps} o
Phi_{-t}). By Minkowski and the L^2 norm of an indicator,
||g_eps - f||_2 <= H * nu(X - G_eps)^{1/2}.
Now the bad set: a state in G_eps fails only when its trajectory over
[0, T] enters the band {r <= (L+1) eps} (an area-O(eps^2) set near the
cones) at some time. Detecting this on a time-mesh of spacing eps gives
at most T/eps + 1 moments, and Phi_t preserves nu, so
nu(X - G_eps) <= C (T eps + eps^2) = C(s + eps^2).
Substituting gives (b).
Proof of (c): Differentiate g_eps under the flow using the fundamental
theorem. For each t,
X (a_eps o Phi_{-t}) = -d/dt (a_eps o Phi_{-t}),
so integrating in t from 0 to T and using the fundamental theorem of
calculus gives (c). This is a WEAK identity: it does not differentiate
the indicator 1_{G_eps}. It is justified by testing against smooth
compactly supported functions and using the antitonicity of X (X is
antisymmetric: <X u, v> = -<u, X v> for smooth compactly supported u,
v).
Reference: triangular_billiards_ergodicity_original.txt, (eq:bad-measure),
(eq:average-bounds).
Exercise 36. How is the smoothed equation integrated along a segment?
On the safe region (r > K eps), X and Y S_eps commute, and X S_eps =
S_eps X. Applied to U, this gives the SMOOTH equation on the safe
region:
X (Y S_eps U) = Y S_eps B.
(Proof: X Y S_eps U = Y X S_eps U = Y S_eps X U = Y S_eps B.)
Pair this equation with the segment average g_eps and integrate by
parts. Change variables in the pairing using the flow, and use the
fundamental theorem of calculus along each good segment of length T.
The result is the segment identity:
<Y S_eps B, g_eps>
= (1/T) int_{G_eps} \bar{f(z)} [ (Y S_eps U)(Phi_T z) - (Y S_eps U)(z) ] dnu(z).
Technical footnote: the differential identity on the safe region requires
differentiability in the direction of the flow, and Y S_eps U might only
be a measurably smooth field. The paper handles this by a countable open-
cover by flat charts and by excluding a null set of "exceptional angular
slices" (the slices on which the smoothed equation fails at any rational
time). This bookkeeping is painful but mechanical; we gloss over it. The
upshot: the segment identity holds almost everywhere in z.
Reference: triangular_billiards_ergodicity_original.txt,
(eq:segment-identity) and the surrounding "Integration of the smoothed
equation" paragraph.
Exercise 37. Why is the right-hand side of the segment identity small?
Bound the integrand by H times the absolute difference of two values of
Y S_eps U. The values are evaluations at points inside the safe region
(by the definition of G_eps: r(Phi_t z) > L eps > K eps). Change
variables using measure preservation of Phi_T. Result:
|RHS| <= (2H / T) * ||Y S_eps U||_{L^2(r > K eps)}.
Now use T = s / eps and the small-scaled-derivative property
(Exercise 27(b)) with u = U:
|RHS| <= (2H / s) * (eps ||Y S_eps U||_{L^2(r > K eps)}) = o(1)
as eps -> 0, at fixed s. This is where the long segment T = s/eps is
crucial: the factor 1/T = eps/s, and eps ||Y S_eps U|| -> 0. The
combination is o(1).
Reference: triangular_billiards_ergodicity_original.txt, the display
after (eq:segment-identity).
Exercise 38. Convert to <B, Y w_eps> and bound it per coefficient.
Set w_eps := S_eps g_eps, with the supported extension of
Exercise 27(d): supported in {r >= (L-1) eps}, which is a compact
regular subset of M. The adjoint identity (Exercise 27(d)) converts
<Y S_eps B, g_eps> to
<B, Y w_eps> = -<Y S_eps B, g_eps> = o(1) (as eps -> 0, fixed s).
We need a bound on w_eps's spatial gradient to pass to the limit. Apply
the energy lemma (Exercise 29) with E = M (no restriction; w_eps is
supported away from cones). Bound:
||X w_eps||_2 = ||X S_eps g_eps||_2 = ||S_eps X g_eps||_2 <= ||X g_eps||_2 <= 2H eps/s,
||del w_eps||_2 <= C H / eps (from the scaled-derivative bound in (a)).
Apply the energy lemma with E = M:
||del (w_eps)_j||_2^2 <= 4 * ||X w_eps||_2 * ||del w_eps||_2
<= 4 * (2H eps/s) * (C H / eps)
= C H^2 / s.
Hence, for EVERY fixed j (in particular j = m-2 and j = m),
||del (w_eps)_j||_2 <= C H / sqrt(s), uniform in eps.
Reference: triangular_billiards_ergodicity_original.txt,
(eq:fixed-s-bound).
Exercise 39. Take the first limit eps -> 0 (s fixed).
Fix s > 0. As eps -> 0:
1. Along a subsequence, w_eps converges weakly in L^2(X) to a limit
g^(s). The set {|u| <= H} is weakly closed, so ||g^(s)||_inf <= H.
2. The smoothed inputs themselves approximate g_eps, and g_eps -> g^(s)
weakly in L^2. So g^(s) is also the weak limit of g_eps.
3. From (b) of Exercise 35 and weak lower-semicontinuity of the L^2
norm, ||g^(s) - f||_2 <= C H sqrt(s).
4. From (c) of Exercise 35 and bound ||X g_eps|| <= 2H/T = C eps/s,
together with X w_eps = S_eps X g_eps, X g^(s) = 0 in the
distributional sense.
5. The relevant coefficient derivatives d(w_eps)_{m-2} and
dbar(w_eps)_m are uniformly bounded in L^2 (Exercise 38, only
m-2 and m matter because B lives in mode m-1). Passing to a
further subsequence, they converge weakly to d(g^(s))_{m-2} and
dbar(g^(s))_m (identified by testing against smooth compactly
supported fields).
6. Therefore:
<B, P_{m-1} Y g^(s)> = 0 for every s > 0.
This is the "first limit" of the rigidity proof.
Reference: triangular_billiards_ergodicity_original.txt, "The first
limit: fixed s."
Exercise 40. Take the second limit s -> 0.
We CANNOT pass to the limit in the bound ||del(w_eps)_j|| <= C H / sqrt(s)
because that blows up as s -> 0. Instead use the bounded-modes lemma
(Exercise 31) directly on g^(s):
- g^(s) is in L^inf(X), with ||g^(s)||_inf <= H;
- X g^(s) = 0 (distributional, from the first limit);
so bounded-modes gives ||del (g^(s))_j||_2 <= C H, UNIFORMLY in s and j.
Meanwhile, from ||g^(s) - f||_2 <= C H sqrt(s) and s -> 0,
g^(s) -> f strongly in L^2(X).
Pick a sequence s_n -> 0 along which the relevant coefficient gradients
del(g^(s_n))_{m-2} and del(g^(s_n))_m have weak L^2 limits. Because
g^(s_n) -> f strongly in L^2, those weak limits are exactly the
distributional gradients del f_{m-2} and del f_m (which therefore exist
as L^2 elements). Passing to the limit in
<B, P_{m-1} Y g^(s_n)> = 0
gives
<B, P_{m-1} Y f> = 0.
This is the "second limit" of the rigidity proof.
Reference: triangular_billiards_ergodicity_original.txt, "The second
limit: s -> 0."
Exercise 41. Close the argument.
Combine with Exercise 24: P_{m-1} Y f = -2 i B. So
0 = <B, P_{m-1} Y f> = <B, -2 i B> = -2 i ||B||_2^2.
Therefore ||B||_2 = 0, i.e. B = 0, i.e. dbar f_m = 0. Since m was
arbitrary:
dbar f_m = 0 for every m in Z.
Then d f_k = -dbar f_{k+2} = 0 for every k. So every coefficient f_k has
vanishing d and dbar as distributions. Each f_k is locally constant; M is
connected, so each f_k is a global constant on M.
We have proved the rigidity proposition (Exercise 25): the coefficients
of f are constant functions on M. Nothing about the angles has been used
yet. The whole analytic part of the theorem is finished.
Reference: triangular_billiards_ergodicity_original.txt, end of the
Proposition proof.
Exercise 42. Why does the proof have TWO limits, and what is the role of
each estimate?
Every step in the proof forces one limit, because no finite-eps
approximation is exactly invariant. The two limits handle two different
obstructions:
FIRST LIMIT (eps -> 0, s fixed).
The smoothed equation X Y S_eps U = Y S_eps B is EXACT on the safe
region, but S_eps g_eps is only approximately invariant (residual
O(eps/s) in L^2). Actually passing to the limit kills the smoothing
error O(eps). The result g^(s) is exactly X-invariant, but g^(s)
is only CLOSE to f: the difference is O(sqrt(s)).
THE ESTIMATE HERE IS THE ENERGY LEMMA (Exercise 29). It produces the
per-coefficient gradient bound C H / sqrt(s), uniform in eps and j.
The blowup 1/sqrt(s) as s -> 0 is not (yet) a problem because s is
held fixed.
SECOND LIMIT (s -> 0).
The intermediate function g^(s) is exactly X-invariant, so
bounded-modes applies and produces the bound C H (UNIFORMLY in s).
The 1/sqrt(s) bound is NOT used; it would blow up. Instead, we
START FRESH with the exact invariance of g^(s). This is the
reason the "second passage" that the introduction calls out is a
genuine theorem, and not just a "pass to a limit" step.
So: energy lemma handles smoothing error; bounded-modes handles closeness
error. Two limits, two estimates, in a strict order. Mixing them up (say,
passing both eps and s to 0 at once) would lose the uniform control.
This is the point of the paper's careful "first x, then y" structure.
Reference: triangular_billiards_ergodicity_original.txt, "The first
limit" and "The second limit" subsections; also the introduction's
paragraph on "the low-regularity obstacle" and "the second passage".
Exercise 43. Where does the angle hypothesis enter? The holonomy
argument.
By Exercise 41, each Fourier coefficient f_j is a constant section on M.
For j = 0, "constant section" just means a constant ordinary function. For
j != 0, "constant section" means a covariantly-constant section of the
flat line bundle L^{-j}, and such a section exists only if L^{-j} is
trivial.
Fix the cone point whose cone angle is 2 alpha, at the vertex of the
triangle with angle alpha. A small loop around it rotates the flat frame
by 2 alpha. The number f_j transforms as a section:
f_j -> e^{-i j (2 alpha)} f_j = e^{-2 i j alpha} f_j.
For f_j to be single-valued on M, the phase e^{-2 i j alpha} must be
equal to 1:
e^{-2 i j alpha} = 1
<=> 2 j alpha = 2 pi k for some integer k
<=> alpha / pi = k / j (rational).
If alpha / pi is irrational, NO nonzero j works. Therefore, at that cone
alone, all nonzero f_j must be zero.
Only f_0 survives. Because f_0 is an ordinary function on M with
d f_0 = 0 and M is connected, f_0 is a constant.
Reference: triangular_billiards_ergodicity_original.txt, section
"Holonomy and the conclusion".
Exercise 44. Why is ONE irrational angle enough?
Because the argument of Exercise 43 needs only ONE cone point to have
the irrational cone angle / pi. If any one of alpha/pi, beta/pi,
gamma/pi is irrational, then at that vertex the monodromy e^{-2 i j
alpha} is not equal to 1 for any nonzero j, and the argument goes
through using that one cone. The other cones are irrelevant.
Concretely: if alpha/pi is irrational, use the alpha-cone; if beta/pi is
irrational (but alpha/pi is rational), use the beta-cone; if only
gamma/pi is irrational, use the gamma-cone. At least one of them is
available by the hypothesis of the theorem.
Reference: triangular_billiards_ergodicity_original.txt, the last
paragraphs of the theorem proof; also the Remark after the theorem.
(Note: exactly one of alpha/pi, beta/pi, gamma/pi being irrational is
impossible since their sum is 1, so "at least one" means "at least two."
But the argument itself only uses one cone.)
Exercise 45. Complete the proof of the main theorem.
Let A subset X be a measurable invariant set. Define f = 1_A in L^inf(X).
Then Xf = 0 distributionally (invariance). By the rigidity proposition
(Exercise 25, proved in Exercises 26-42), each f_j is a constant
parallel section on M. By the holonomy argument (Exercises 43-44),
f_j = 0 for every j != 0, because at least one cone angle/pi is
irrational. Hence f = f_0 (constant in theta), and f_0 is a constant on
the connected surface M.
Because f is a 0/1 indicator, f_0 equals 0 or 1 almost everywhere. If
f_0 = 0 then mu(A) = 0; if f_0 = 1 then mu(A) = 1. This is exactly the
definition of ergodicity. The theorem is proved.
Reference: triangular_billiards_ergodicity_original.txt, proof of
Theorem "main".
Exercise 46. Sanity check: does the theorem really need the irrationality
hypothesis?
Yes. For the equilateral triangle (alpha = beta = gamma = pi/3), KMS
1986 gives ergodicity only in almost every direction, and there are
genuinely non-ergodic directions (the "cylinder" directions of the
unfolded torus, along which the flow decomposes into parallel closed
geodesics). Those directions form invariant subsets of intermediate
measure, which means the full billiard flow on Q x S^1 is NOT ergodic.
So excluding the all-rational case is necessary, not just a proof
convenience. The same phenomenon occurs for 45-45-90 and 30-60-90
triangles.
Reference: triangular_billiards_ergodicity_original.txt; KMS 1986
(the original paper, listed in the references.bib within the LaTeX source).
Exercise 47. Spectral consequence 1: INTERIOR quantum ergodicity.
Let Q satisfy the theorem's hypothesis. Fix either Dirichlet or Neumann
boundary conditions. Let (u_j)_{j >= 1} be ANY complete orthonormal
eigenbasis of L^2(Q) for the corresponding Laplacian, with eigenvalues
lambda_j^2 and 0 <= lambda_1 <= lambda_2 <= ... (frequencies listed with
multiplicity). Then there is a density-one subsequence S_int of {1, 2,
3, ...} such that for every classical order-zero pseudodifferential
operator A with compactly supported Schwartz kernel in Q x Q,
lim_{j -> infty, j in S_int} <A u_j, u_j>_{L^2(Q)}
= int_{Q x S^1} sigma_0(A) dmu.
"sigma_0(A)" is the principal symbol of A, restricted to the unit cosphere
bundle of Q (identified with Q x S^1 by the Euclidean metric). "Density
one" means #(S_int intersect {1,...,N}) / N -> 1 as N -> infty.
Interpretation: the eigenfunctions "see" the whole geometry - their
position and momentum distributions converge to the uniform measure on
phase space, along a typical subsequence. Not every j, but almost every
j.
Reference: Zelditch-Zworski 1996, Commun. Math. Phys. 175 (this is the
"ZZ" citation in the original). See also Burq 2003 and the discussion in
triangular_billiards_ergodicity_original.txt, section "Quantum
ergodicity and nodal domains".
Exercise 48. Spectral consequence 2: BOUNDARY quantum ergodicity.
Let Gamma be the union of the three open sides (the regular boundary).
For lambda_j > 0, put h_j := 1/lambda_j, and define the boundary Cauchy
data:
u_j^b := h_j * dn u_j on Gamma (Dirichlet boundary condition),
u_j^b := u_j on Gamma (Neumann boundary condition),
where nu is the outward unit normal.
Let B*Gamma := ((s, eta) in T*Gamma : |eta| < 1) where s is arc length
along Gamma and eta is the signed tangential covector.
There is a density-one subsequence S_partial of {j : lambda_j > 0} such
that, for every a in C_c^infty(T*Gamma) with a properly supported
order-zero semiclassical quantization A_h = Op_h(a),
lim_{j -> infty, j in S_partial} <A_{h_j} u_j^b, u_j^b>_{L^2(Gamma)}
= (2 / (pi Area(Q))) int_{B*Gamma} a(s, eta) w_B(eta) ds deta,
with weights
w_D(eta) = sqrt(1 - eta^2) (Dirichlet),
w_N(eta) = (1 - eta^2)^{-1/2} (Neumann).
The boundary data are NOT normalized to have unit L^2(Gamma) norm.
Interpretation: the eigenfunctions' boundary traces also equidistribute,
but with a nontrivial profile w_B(eta) along eta.
Reference: Hassell-Zelditch 2004, Commun. Math. Phys. 248 ("HZ"). See
also Burq 2003 and section "Quantum ergodicity and nodal domains" of
triangular_billiards_ergodicity_original.txt.
Exercise 49. Spectral consequence 3: NODAL DOMAINS GROW.
Assume the eigenbasis (u_j) is real-valued. Then there is a density-one
subsequence S_nod with the property: the number of NODAL DOMAINS of u_j
(the connected components of Q minus {u_j = 0}) tends to infinity as
lambda_j -> infty with j in S_nod.
Interpretation: the number of "oscillating lobes" of the eigenfunctions
grows unboundedly. This matches the physical intuition that high-
frequency eigenfunctions of a chaotic table should have very many sign
changes. Yau's conjecture predicts a bound proportional to lambda_j on
the number of nodal domains; here we only claim the count grows without
bound, along a density-one subsequence.
Reference: Hezari 2016, arXiv:1606.08066 ("Hezari"). See also Yau's
nodal conjecture background and section "Quantum ergodicity and nodal
domains" of triangular_billiards_ergodicity_original.txt.
Exercise 50. Compressed proof at a glance (for experts).
Let Q be a nondegenerate triangle with at least one angle/pi irrational.
Its double M is a compact flat surface with three cone points of cone
angles 2 alpha, 2 beta, 2 gamma. Let X = SM = M x S^1, mu = normalized
dArea dtheta, X = geodesic-flow field, Y = perpendicular horizontal field
(these commute).
Goal: any f in L^inf(X) with Xf = 0 is constant.
Fourier in theta: f = sum_j f_j e^{i j theta}, with the mode identity
(Xu)_n = d u_{n-1} + dbar u_{n+1}. So Xf = 0 gives d f_{m-1} + dbar
f_{m+1} = 0 for every m. The tail U = sum_{k >= 0} P_{m+2k} f satisfies
XU = B = e^{i(m-1)theta} dbar f_m. It suffices to show B = 0: then
d f_{m-2} = -dbar f_m, so (Yf)_{m-1} = -2iB, and applying the same
argument to all m gives dbar f_m = d f_{m-2} = 0 for all m.
For eps, s > 0 let T = s/eps and G_eps = {r(Phi_t z) > L eps, 0 <= t <=
T}. Let a_eps = f 1_{G_eps} and g_eps = (1/T) int_0^T a_eps o Phi_{-t}
dt. Then ||g_eps - f||_2 <= CH sqrt(s + eps^2), ||X g_eps||_2 <= CH/T,
and g_eps vanishes on {r < L eps}. Apply Y S_eps to XU = B (S_eps =
spatial convolution on the safe region, path-independent, commutes with
X, satisfies <Y S_eps u, g> = -<u, Y S_eps g>), integrate the resulting
identity X(Y S_eps U) = Y S_eps B along good segments, and use the
small-scaled-derivative property eps ||Y S_eps U||_{r > K eps} -> 0 to
get <B, Y (S_eps g_eps)> = o(1) as eps -> 0.
Set w_eps = S_eps g_eps. Energy lemma with E = M:
||del (w_eps)_j||^2 <= 4 ||X w_eps|| ||del w_eps|| <= C H^2 / s.
FIRST LIMIT eps -> 0 at fixed s: along a subsequence, w_eps -> g^(s)
weakly, X g^(s) = 0 in distributions, ||g^(s) - f||_2 <= C H sqrt(s), and
the relevant coefficient gradients del(g^(s))_{m-2}, del(g^(s))_m are
weak limits, so <B, P_{m-1} Y g^(s)> = 0 for every s.
SECOND LIMIT s -> 0: bounded-modes lemma applied to g^(s) (which is
in L^inf with X g^(s) = 0) gives ||del(g^(s))_j|| <= C H, uniform in s;
g^(s) -> f strongly in L^2 as s -> 0; hence <B, P_{m-1} Y f> = 0.
Combine with P_{m-1} Y f = -2 i B to get -2 i ||B||_2^2 = 0, so B = 0.
Therefore dbar f_m = 0 for all m, and d f_j = 0 for all j.
Every f_j is a parallel section of L^{-j} on the connected surface M. At
the alpha-cone, monodromy rotates the frame by 2 alpha, so single-
valuedness of f_j requires e^{-2 i j alpha} = 1. Irrationality of at
least one angle/pi forces f_j = 0 for every j != 0. The zeroth
coefficient f_0 is a constant on M. So f is constant. Ergodicity. Then
interior quantum ergodicity (ZZ 1996), boundary quantum ergodicity (HZ
2004), and growth of the nodal count (Hezari 2016) follow from the flow
ergodicity by the cited theorems.
Reference: triangular_billiards_ergodicity_original.txt (the whole
paper); this is the complete proof in one place.
Exercise 51. What does the theorem NOT say? (Caveats.)
- Not unique ergodicity: the theorem says the Liouville measure mu is
ergodic (it cannot be decomposed into two invariant pieces of positive
measure). It does NOT say mu is the unique invariant probability
measure. Rational polygons, for example, have many invariant
probability measures.
- Not mixing: mixing is stronger (asymptotic independence of past and
future, decaying time correlations). Not claimed. (Chaika-Forni 2026
proved weak mixing for a dense G_delta of n-gons, but for any specific
irrational triangle, mixing is open.)
- Not about every orbit: individual trajectories can be non-dense, or
recurrent, or weird. The theorem is about almost every starting state.
A positive-measure set of unusual orbits is what ergodicity rules out.
- No quantitative rate: the theorem says "time average converges to
space average", without saying how fast. Quantitative equidistribution
would be a separate hard problem.
- Density-one subsequences are not canonical: the sets S_int, S_partial,
S_nod of Exercises 47-49 are allowed to depend on the triangle Q, on
the boundary condition (Dirichlet vs Neumann), and on the particular
eigenbasis (u_j). The theorem does not identify the zero-density
exceptional set.
- No growth rate on nodal domains: only that the count tends to
infinity along S_nod, not how fast.
- The theorem is compatible with "scarring": individual eigenfunctions
might concentrate on special sets. The theorem only rules out scars
occurring with density one.
Reference: triangular_billiards_ergodicity_original.txt, the final
paragraph of section "Quantum ergodicity and nodal domains" and the
remark after the main theorem.
Exercise 52. Notation glossary.
All symbols are defined in the text; collected here for quick lookup.
- alpha, beta, gamma: angles of Q in radians, alpha+beta+gamma = pi.
- pi = 3.14159...: circle constant.
- S^1: the unit circle, identified with [0, 2 pi).
- Q: the interior of the triangle (a table).
- M: the double of Q (two glued copies), a compact flat surface with
three cone points.
- cone angle at the alpha-vertex of M: 2 alpha (similarly for beta,
gamma).
- X: the unit tangent bundle M x S^1, phase space. (Also used for the
flow vector field; context clarifies.)
- mu: invariant probability measure dArea dtheta / (2 pi Area(Q)).
- Phi_t: the geodesic-flow time-t map on X.
- theta: direction in S^1.
- z = x_1 + i x_2: flat coordinate on M.
- d, dbar: complex spatial derivatives (1/2)(d_1 - i d_2) and
(1/2)(d_1 + i d_2) on M.
- X, Y: horizontal vector fields, flow direction and perpendicular.
- del: flat spatial gradient on M.
- f: bounded measurable function on X, often an indicator 1_A.
- f_j: the j-th angular Fourier coefficient of f.
- P_j: projection onto the j-th Fourier mode.
- U: the tail sum_{k >= 0} P_{m+2k} f.
- B: the endpoint field e^{i (m-1) theta} dbar f_m.
- r(x): distance from x to nearest cone point in M.
- K: constant making embedded eps-disks available when r > K eps.
- rho_eps: smoothing bump eps^{-2} rho(y/eps); rho radial, mass 1.
- S_eps: spatial smoothing operator (Exercise 26).
- L: constant with L > K + 4, chosen to keep trajectories clear.
- s, eps: small parameters with T = s/eps.
- H: ||f||_{L^inf}.
- G_eps: good set where trajectory stays > L eps from cones for the
next T seconds.
- a_eps: f * 1_{G_eps}.
- g_eps: (1/T) int_0^T a_eps o Phi_{-t} dt (long-segment average).
- w_eps: S_eps g_eps (smoothed, supported field).
- g^(s): weak limit of w_eps as eps -> 0 at fixed s.
- A: a measurable invariant set.
- Gamma: regular boundary of Q (three open sides).
- B*Gamma: cosphere bundle of Gamma = {(s, eta) : |eta| < 1}.
- w_D(eta) = sqrt(1 - eta^2), w_N(eta) = (1 - eta^2)^{-1/2}: boundary
weights for Dirichlet and Neumann.
- lambda_j: eigenvalues of the Laplacian, listed with multiplicity.
- u_j: the j-th eigenfunction of an orthonormal eigenbasis.
- h_j: lambda_j^{-1}, the semiclassical frequency.
- A_h: semiclassical pseudodifferential operator with symbol a.
- sigma_0(A): principal symbol of the pseudodifferential operator A.
- Nodal domain: connected component of Q minus {u_j = 0}.
Exercise 53. Review questions for a high-school reader.
Answers follow each question in quotes; use them to test yourself.
1. In one sentence, what is ergodicity? "Every measurable invariant
set has measure 0 or 1."
2. Why isn't "every orbit is dense" the same as ergodicity? "Ergodicity
is statistical; a measure-zero set of unusual orbits is allowed."
3. Which triangles does the theorem cover? "Those with at least one of
alpha/pi, beta/pi, gamma/pi irrational."
4. Why is exactly one irrational angle impossible? "Because
alpha/pi + beta/pi + gamma/pi = 1: if two ratios are rational the
third is rational."
5. What is the double of a triangle? "Two copies of the triangle glued
along the boundary, a compact flat surface with three cone points."
6. Why is the double useful? "It converts the billiard into a
straight-line flow without needing to unfold into infinitely many
copies."
7. What is a cone point, and what is its cone angle at the alpha-vertex?
"A point where the flat surface has total angle different from
2 pi. At the alpha-vertex, the cone angle is 2 alpha."
8. What does Xf = 0 mean? "The function f is invariant along the
flow."
9. Why isn't Xf = 0 on its own enough? "Because f could be an
invariant staircase (e.g. an angular-block function). We need
regularity to prove it is constant."
10. What is the mode identity? "(Xu)_n = d u_{n-1} + dbar u_{n+1}; X
shifts the angular index by plus or minus 1."
11. What is the tail U and why does it matter? "U = sum_{k >= 0}
P_{m+2k} f. It satisfies XU = B, a single-mode field, isolating the
endpoint of the arithmetic progression."
12. What does S_eps do? "Spatial-average f over a small disk, keeping
the direction theta frozen; smooth without touching cones."
13. Describe the two limits. "First eps -> 0 at fixed s: kills the
smoothing error; produce g^(s) with X g^(s) = 0 exactly. Second
s -> 0: use bounded-modes to get uniform control; pass to f."
14. Why does the angle argument only need ONE cone? "The monodromy
e^{-2 i j alpha} must equal 1 for a parallel section; if any single
angle/pi is irrational, no nonzero j works."
15. True or false: the theorem says every trajectory is dense. "False.
Ergodicity is a statistical statement."
16. Difference between interior QE, boundary QE, nodal domains?
"Interior QE: eigenfunctions equidistribute in phase space along a
density-one subsequence. Boundary QE: their boundary traces also
equidistribute, with weight w_B(eta). Nodal domains: the count of
sign-change lobes grows without bound along a density-one
subsequence."
Exercise 54. References (plain text; full bibliography inside
triangular_billiards_ergodicity_original.txt, section references.bib).
The bibliography of the paper contains 12 items. Reproduced here in the
short labels used above:
ZK Zemlyakov, A.N. and Katok, A.B., 1975. Topological transitivity
of billiards in polygons. Math. Notes 18.
doi:10.1007/BF01818045.
KMS Kerckhoff, S., Masur, H. and Smillie, J., 1986. Ergodicity of
billiard flows and quadratic differentials. Ann. Math. 124.
doi:10.2307/1971280.
MT Masur, H. and Tabachnikov, S., 2002. Rational billiards and
flat structures. Handbook of Dynamical Systems 1A.
doi:10.1016/S1874-575X(02)80015-7.
Vorobets Vorobets, Ya.B., 1997. Ergodicity of billiards in polygons.
Sb. Math. 188. doi:10.1070/SM1997v188n03ABEH000211.
CF Chaika, J. and Forni, G., 2026. Weakly mixing polygonal
billiards. Ann. Math. 203.
doi:10.4007/annals.2026.203.3.1. arXiv:2003.00890.
FM Forni, G. and Moll, N., 2025. Cohomological equation for
geodesic flows on flat surfaces. arXiv:2510.18128v1.
ZSBJ Zahradova, K., Slipantschuk, J., Bandtlow, O.F. and Just, W.,
2022. Impact of symmetry on ergodic properties of triangular
billiards. Phys. Rev. E 105.
doi:10.1103/PhysRevE.105.L012201.
WCP Wang, J., Casati, G. and Prosen, T., 2014. Nonergodicity and
localization of invariant measure for two colliding masses.
Phys. Rev. E 89. doi:10.1103/PhysRevE.89.042918.
ZZ Zelditch, S. and Zworski, M., 1996. Ergodicity of eigenfunctions
for ergodic billiards. Commun. Math. Phys. 175.
doi:10.1007/BF02099513.
Burq Burq, N., 2003. Quantum ergodicity of boundary values of
eigenfunctions: A control theory approach. arXiv:math/0301349.
HZ Hassell, A. and Zelditch, S., 2004. Quantum ergodicity of
boundary values of eigenfunctions. Commun. Math. Phys. 248.
doi:10.1007/s00220-004-1070-2.
Hezari Hezari, H., 2016. Quantum ergodicity and L^p norms of
restrictions of eigenfunctions. arXiv:1606.08066v1.
For the main paper itself, see triangular_billiards_ergodicity_original.txt.
Exercise 55. Try it yourself: pick a random triangle and check the
hypothesis.
Recipe. Pick alpha/pi = any irrational number in (0, 1). Pick beta/pi =
any positive number with alpha/pi + beta/pi < 1. Set gamma/pi = 1 -
alpha/pi - beta/pi. Then alpha + beta + gamma = pi, and alpha/pi is
irrational, so the theorem applies: this triangle has an ergodic
billiard.
Examples (ratios alpha/pi, beta/pi, gamma/pi listed explicitly).
Excluded (all three ratios rational, no theorem):
0.7, 0.2, 0.1 (sum = 1.0)
0.5, 0.4, 0.1 (sum = 1.0)
0.5, 0.1, 0.4 (sum = 1.0)
0.5, 0.25, 0.25 (sum = 1.0)
equilateral, 45-45-90, 30-60-90 triangles: all excluded.
Covered (at least one ratio irrational):
1/sqrt(2) = 0.70710678..., 0.1, 0.9 - 1/sqrt(2) = 0.19289321...
(sum = 1.0). One irrational ratio, two rational? No: actually
0.1 is rational, the other two are irrational. Covered.
1/sqrt(10) = 0.31622776..., 1/sqrt(3) - 1/sqrt(5) = 0.13020...,
1 - 1/sqrt(10) - 1/sqrt(3) + 1/sqrt(5) = 0.55357...
(sum = 1.0). All three ratios irrational. Covered.
0.5, 0.3, 0.2 (all rational): excluded, not covered.
0.6, 0.4, 0.0 (gamma = 0): degenerate, not a triangle.
Self-check:
Q. Does the theorem say "time average equals space average for every
initial state"? A. No, only almost every.
Q. Does the theorem rule out invariant sets of intermediate measure?
A. Yes, that is exactly what ergodicity means.
Q. Does the theorem prove mixing? A. No, mixing is strictly stronger.
Q. Does the rigidity proposition conclude "each f_j is a parallel
section on M"? A. Yes.
Q. Does holonomy kill every nonzero mode if any single angle/pi is
irrational? A. Yes.
If you can answer all these confidently, you have followed the whole
argument. Congratulations: you have read the proof of a 2025/2026
research theorem.
=================================================================
END OF DOCUMENT
=================================================================