The Morse inequalities #
Let f be a Morse function on a boundaryless finite-dimensional manifold M, let V be a
finite-dimensional module over a division ring k, and write M_a = {f ≤ a}. Let a ≤ b be
values of f at no critical point, with compact band f⁻¹[a, b]. Write
bᵢ = dim Hᵢ(M_b, M_a; V) for the dimensions of relative singular homology, and cᵢ for the
number of critical points of index i with value in (a, b]. Then, for every n:
Hₙ(M_b, M_a; V)is finite-dimensional, andbₙ ≤ cₙ dim V(the weak Morse inequalities);bₙ - bₙ₋₁ + ⋯ ± b₀ ≤ (cₙ - cₙ₋₁ + ⋯ ± c₀) dim V(the strong Morse inequalities);- if
n ≥ dim M, then∑ᵢ₌₀ⁿ (-1)ⁱ bᵢ = (∑ᵢ₌₀ⁿ (-1)ⁱ cᵢ) dim V(Euler's equality); - if
n ≥ dim M, then∑ᵢ₌₀ⁿ cᵢ dim V · tⁱ = ∑ᵢ₌₀ⁿ bᵢ tⁱ + (1 + t) Q(t)for a polynomialQwith natural number coefficients, the coefficient oftⁱbeing the gap in thei-th strong inequality (the Morse polynomial identity); b₀ + ⋯ + bₙis at mostdim Vtimes the number of critical points with values in(a, b];- if no two of these critical points have indices differing by one, then
bₙ = cₙ dim V(the lacunary principle).
The band W = f⁻¹[a, b] is then a compact cobordism from the level f⁻¹(a) to the level
f⁻¹(b), and the inclusion of pairs (W, f⁻¹(a)) ⟶ (M_b, M_a) induces isomorphisms on relative
homology (TauCeti.isIso_singularHomologyMap_bandPairInclusion). So the finiteness, the weak and
strong inequalities, Euler's equality and the polynomial identity hold as well for the numbers
bᵢ = dim Hᵢ(W, f⁻¹(a); V) of the cobordism relative to its incoming boundary.
Read from the other end, W is the band of -f between -b and -a, relative to its incoming
boundary f⁻¹(b) (TauCeti.bandPair_neg). A critical point of index λ of f has index
dim M - λ for -f. So the same statements hold for the numbers bᵢ = dim Hᵢ(W, f⁻¹(b); V) of
the cobordism relative to its outgoing boundary, with cᵢ the number of critical points of index
dim M - i with values in [a, b). Comparing the two Euler equalities, the Euler characteristics
of (W, f⁻¹(b)) and (W, f⁻¹(a)) agree up to the sign (-1)^(dim M).
For compact M all of M is such a band, with M_a = ∅, and the same statements hold for the
singular homology Hᵢ(M; V) and the numbers cᵢ of critical points of index i on M. In
particular the singular homology of a compact manifold admitting a Morse function is
finite-dimensional in every degree.
The proof is Milnor's. Across a single critical level the sublevel pair has the relative homology
of a finite CW pair with one i-cell for each critical point of index i
(TauCeti.IsMorse.exists_relCWComplex_isIso_singularHomologyMap_sublevelPair), where the
three statements are the Morse inequalities for CW pairs. Across a band without critical points
the relative homology vanishes, by the deformation retraction of M_b onto M_a
(TauCeti.isZero_singularHomology_sublevelPair_of_mfderiv_ne_zero).
A band is cut at the critical values it contains into finitely many such pieces
(TauCeti.IsMorse.band_induction), and the long exact sequence of the triple
M_a ⊆ M_m ⊆ M_b shows that dimensions and the alternating sums Sₙ are subadditive
(TauCeti.TopTriple.finrank_singularHomology_totalPair_le and
TauCeti.TopTriple.sum_range_neg_one_pow_mul_finrank_singularHomology_totalPair_le), while the
critical counts are additive. Above the dimension of M the connecting map in degree n + 1
vanishes, since its source is zero by the weak inequalities, which turns the strong inequality
into the equality. The polynomial identity and the lacunary principle are then the numerical
consequences TauCeti.exists_sum_range_monomial_eq_add_mul and
TauCeti.eq_of_forall_sum_range_neg_one_pow_mul_le of the strong inequalities and Euler's
equality.
The coefficients of Q are ranks of boundary maps. Cut the band at regular values
a = s 0 ≤ s 1 ≤ ⋯ ≤ s (N + 1) = b so that each piece f⁻¹[s j, s (j + 1)] contains at most one
critical value; on a compact manifold such values exist
(TauCeti.IsMorse.exists_regular_sublevel_partition). Each piece then has bᵢ = cᵢ dim V, and
summing the defects of subadditivity of Sₙ over the triples
(M_{s (j + 2)}, M_{s (j + 1)}, M_{s 0}) (TauCeti.sum_range_finrank_range_singularHomologyδ_eq)
shows that the coefficient of tⁿ in Q is the sum over j < N of the ranks of the connecting
maps Hₙ₊₁(M_{s (j + 2)}, M_{s (j + 1)}) ⟶ Hₙ(M_{s (j + 1)}, M_{s 0})
(TauCeti.IsMorse.coeff_eq_sum_range_finrank_range_singularHomologyδ_sublevelPair).
Main results #
TauCeti.IsMorse.finrank_singularHomology_sublevelPair_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_sublevelPair_leandTauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_sublevelPair_eq: the weak and strong Morse inequalities and Euler's equality for a compact band between regular values.TauCeti.IsMorse.exists_sum_range_monomial_sublevelPair_eq_add_mul,TauCeti.IsMorse.sum_range_finrank_singularHomology_sublevelPair_leandTauCeti.IsMorse.finrank_singularHomology_sublevelPair_eq_of_forall_ne_add_one: the Morse polynomial identity, the bound on the total Betti number and the lacunary principle for such a band.TauCeti.IsMorse.finite_singularHomology_bandPair,TauCeti.IsMorse.finrank_singularHomology_bandPair_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_bandPair_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_bandPair_eqandTauCeti.IsMorse.exists_sum_range_monomial_bandPair_eq_add_mul: the same statements for the cobordismf⁻¹[a, b]relative to its incoming boundaryf⁻¹(a).TauCeti.IsMorse.finite_singularHomology_upperBandPair,TauCeti.IsMorse.finrank_singularHomology_upperBandPair_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_upperBandPair_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_upperBandPair_eqandTauCeti.IsMorse.exists_sum_range_monomial_upperBandPair_eq_add_mul: the same statements for the cobordism relative to its outgoing boundaryf⁻¹(b), with critical points counted by their complementary index.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_upperBandPair_eq_neg_one_pow_mul: the Euler characteristics of the cobordism relative to its two boundaries agree up to the sign(-1)^(dim M).TauCeti.IsMorse.sum_range_finrank_range_singularHomologyδ_sublevelPair_eqandTauCeti.IsMorse.coeff_eq_sum_range_finrank_range_singularHomologyδ_sublevelPair: for a band cut at regular values into pieces with at most one critical value each, the gap in then-th strong inequality, which is the coefficient oftⁿinQ, is a sum of ranks of connecting maps of triples of sublevels.TauCeti.IsMorse.finrank_singularHomology_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_le,TauCeti.IsMorse.sum_range_neg_one_pow_mul_finrank_singularHomology_eq,TauCeti.IsMorse.exists_sum_range_monomial_eq_add_mul,TauCeti.IsMorse.sum_range_finrank_singularHomology_leandTauCeti.IsMorse.finrank_singularHomology_eq_of_forall_ne_add_one: the same for a compact manifold, with the numbersTauCeti.criticalCountof critical points of each index and their totalTauCeti.totalCriticalCount.TauCeti.IsMorse.sum_range_finrank_range_singularHomologyδ_eqandTauCeti.IsMorse.coeff_eq_sum_range_finrank_range_singularHomologyδ: on a compact manifold, the coefficients ofQas sums of ranks of connecting maps along a cut of the values offinto pieces with at most one critical value each, which exists byTauCeti.IsMorse.exists_regular_sublevel_partition.
References #
- J. Milnor, Morse Theory, Annals of Mathematics Studies 51, Princeton University Press, 1963, Section 5.
- R. Bott, Lectures on Morse theory, old and new, Bulletin of the American Mathematical Society 7 (1982), 331--358, Section 1.
Finiteness of relative homology for a band. Let f be a Morse function, and a ≤ b
regular values with compact band f⁻¹[a, b]. Then the relative singular homology
Hₙ({f ≤ b}, {f ≤ a}; V) is finite-dimensional.
The weak Morse inequalities for a band. Let f be a Morse function, and a ≤ b regular
values with compact band f⁻¹[a, b]. Then Hₙ({f ≤ b}, {f ≤ a}; V) has dimension at most dim V
times the number of critical points of index n with values in (a, b].
The strong Morse inequalities for a band. Let f be a Morse function, and a ≤ b regular
values with compact band f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ({f ≤ b}, {f ≤ a}; V) and
cᵢ for the number of critical points of index i with values in (a, b]. Then
bₙ - bₙ₋₁ + ⋯ ± b₀ ≤ (cₙ - cₙ₋₁ + ⋯ ± c₀) dim V.
Euler's equality for a band. Let f be a Morse function, and a ≤ b regular values with
compact band f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ({f ≤ b}, {f ≤ a}; V) and cᵢ for
the number of critical points of index i with values in (a, b]. Then
∑ᵢ₌₀ⁿ (-1)ⁱ bᵢ = (∑ᵢ₌₀ⁿ (-1)ⁱ cᵢ) dim V for every n ≥ dim M.
The Morse polynomial identity for a band. Let f be a Morse function, and a ≤ b regular
values with compact band f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ({f ≤ b}, {f ≤ a}; V)
and cᵢ for the number of critical points of index i with values in (a, b]. Then for every
n ≥ dim M, ∑ᵢ₌₀ⁿ cᵢ dim V · tⁱ = ∑ᵢ₌₀ⁿ bᵢ tⁱ + (1 + t) Q(t) for a polynomial Q with natural
number coefficients. The coefficient of tⁱ in Q is the gap
(cᵢ - cᵢ₋₁ + ⋯ ± c₀) dim V - (bᵢ - bᵢ₋₁ + ⋯ ± b₀) in the i-th strong Morse inequality for
i < n, and zero for i ≥ n.
The total Betti number of a band is at most dim V times its number of critical points.
Let f be a Morse function, and a ≤ b regular values with compact band f⁻¹[a, b]. Then for
every n, ∑ᵢ₌₀ⁿ dim Hᵢ({f ≤ b}, {f ≤ a}; V) is at most dim V times the number of critical
points with values in (a, b].
The lacunary principle for a band. Let f be a Morse function, and a ≤ b regular values
with compact band f⁻¹[a, b]. If no two critical points with values in (a, b] have indices
differing by one, then for every n the dimension of Hₙ({f ≤ b}, {f ≤ a}; V) is dim V times the
number of critical points of index n with values in (a, b].
The quotient polynomial as a sum of ranks #
The gaps in the strong Morse inequalities are ranks of connecting maps. Let f be a Morse
function and s 0 ≤ s 1 ≤ ⋯ ≤ s (N + 1) regular values with compact band f⁻¹[s 0, s (N + 1)],
such that each band f⁻¹[s j, s (j + 1)] contains at most one critical value. Write M_t for the
sublevel {f ≤ t}, bᵢ = dim Hᵢ(M_{s (N + 1)}, M_{s 0}; V) and cᵢ for the number of critical
points of index i with values in (s 0, s (N + 1)]. Then the gap
(cₙ - cₙ₋₁ + ⋯ ± c₀) dim V - (bₙ - bₙ₋₁ + ⋯ ± b₀) in the n-th strong Morse inequality is the
sum over j < N of the ranks of the connecting maps
Hₙ₊₁(M_{s (j + 2)}, M_{s (j + 1)}) ⟶ Hₙ(M_{s (j + 1)}, M_{s 0}) of the triples of sublevels
(M_{s (j + 2)}, M_{s (j + 1)}, M_{s 0}).
The coefficients of the quotient polynomial are ranks of connecting maps. Let f be a
Morse function and s 0 ≤ s 1 ≤ ⋯ ≤ s (N + 1) regular values with compact band
f⁻¹[s 0, s (N + 1)], such that each band f⁻¹[s j, s (j + 1)] contains at most one critical
value. Write M_t for the sublevel {f ≤ t}, bᵢ = dim Hᵢ(M_{s (N + 1)}, M_{s 0}; V) and cᵢ for
the number of critical points of index i with values in (s 0, s (N + 1)]. If m ≥ dim M and
∑ᵢ₌₀ᵐ cᵢ dim V · tⁱ = ∑ᵢ₌₀ᵐ bᵢ tⁱ + (1 + t) Q(t), then the coefficient of tⁿ in Q is the sum
over j < N of the ranks of the connecting maps
Hₙ₊₁(M_{s (j + 2)}, M_{s (j + 1)}) ⟶ Hₙ(M_{s (j + 1)}, M_{s 0}). Such a Q exists by
TauCeti.IsMorse.exists_sum_range_monomial_sublevelPair_eq_add_mul.
Finiteness of relative homology for a cobordism. Let f be a Morse function, and a ≤ b
regular values with compact band f⁻¹[a, b]. Then the relative singular homology
Hₙ(f⁻¹[a, b], f⁻¹(a); V) of the band relative to its lower level is finite-dimensional.
The weak Morse inequalities for a cobordism. Let f be a Morse function, and a ≤ b
regular values with compact band f⁻¹[a, b]. Then Hₙ(f⁻¹[a, b], f⁻¹(a); V) has dimension at
most dim V times the number of critical points of index n with values in (a, b].
The strong Morse inequalities for a cobordism. Let f be a Morse function, and a ≤ b
regular values with compact band f⁻¹[a, b]. Write bᵢ for the dimension of
Hᵢ(f⁻¹[a, b], f⁻¹(a); V) and cᵢ for the number of critical points of index i with values in
(a, b]. Then bₙ - bₙ₋₁ + ⋯ ± b₀ ≤ (cₙ - cₙ₋₁ + ⋯ ± c₀) dim V.
Euler's equality for a cobordism. Let f be a Morse function, and a ≤ b regular values
with compact band f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ(f⁻¹[a, b], f⁻¹(a); V) and cᵢ
for the number of critical points of index i with values in (a, b]. Then
∑ᵢ₌₀ⁿ (-1)ⁱ bᵢ = (∑ᵢ₌₀ⁿ (-1)ⁱ cᵢ) dim V for every n ≥ dim M.
The Morse polynomial identity for a cobordism. Let f be a Morse function, and a ≤ b
regular values with compact band f⁻¹[a, b]. Write bᵢ for the dimension of
Hᵢ(f⁻¹[a, b], f⁻¹(a); V) and cᵢ for the number of critical points of index i with values in
(a, b]. Then for every n ≥ dim M, ∑ᵢ₌₀ⁿ cᵢ dim V · tⁱ = ∑ᵢ₌₀ⁿ bᵢ tⁱ + (1 + t) Q(t) for a
polynomial Q with natural number coefficients. The coefficient of tⁱ in Q is the gap
(cᵢ - cᵢ₋₁ + ⋯ ± c₀) dim V - (bᵢ - bᵢ₋₁ + ⋯ ± b₀) in the i-th strong Morse inequality for
i < n, and zero for i ≥ n.
The cobordism relative to its outgoing boundary #
The band of -f between -b and -a is the band of f relative to its upper level
(TauCeti.bandPair_neg), and the critical points of index i of -f are those of index
dim M - i of f. So the statements for the band pair of -f give the Morse inequalities for the
upper band pair (f⁻¹[a, b], f⁻¹(b)), with the critical points counted by their complementary
index.
Finiteness of relative homology for a cobordism relative to its outgoing boundary. Let f
be a Morse function, and a ≤ b regular values with compact band f⁻¹[a, b]. Then the relative
singular homology Hₙ(f⁻¹[a, b], f⁻¹(b); V) of the band relative to its upper level is
finite-dimensional.
The weak Morse inequalities for a cobordism relative to its outgoing boundary. Let f be
a Morse function on a manifold of dimension m, and a ≤ b regular values with compact band
f⁻¹[a, b]. Then Hₙ(f⁻¹[a, b], f⁻¹(b); V) has dimension at most dim V times the number of
critical points of index m - n with values in [a, b).
The strong Morse inequalities for a cobordism relative to its outgoing boundary. Let f
be a Morse function on a manifold of dimension m, and a ≤ b regular values with compact band
f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ(f⁻¹[a, b], f⁻¹(b); V) and cᵢ for the number of
critical points of index m - i with values in [a, b). Then
bₙ - bₙ₋₁ + ⋯ ± b₀ ≤ (cₙ - cₙ₋₁ + ⋯ ± c₀) dim V.
Euler's equality for a cobordism relative to its outgoing boundary. Let f be a Morse
function on a manifold of dimension m, and a ≤ b regular values with compact band f⁻¹[a, b].
Write bᵢ for the dimension of Hᵢ(f⁻¹[a, b], f⁻¹(b); V) and cᵢ for the number of critical
points of index m - i with values in [a, b). Then ∑ᵢ₌₀ⁿ (-1)ⁱ bᵢ = (∑ᵢ₌₀ⁿ (-1)ⁱ cᵢ) dim V for
every n ≥ m.
The Morse polynomial identity for a cobordism relative to its outgoing boundary. Let f
be a Morse function on a manifold of dimension m, and a ≤ b regular values with compact band
f⁻¹[a, b]. Write bᵢ for the dimension of Hᵢ(f⁻¹[a, b], f⁻¹(b); V) and cᵢ for the number of
critical points of index m - i with values in [a, b). Then for every n ≥ m,
∑ᵢ₌₀ⁿ cᵢ dim V · tⁱ = ∑ᵢ₌₀ⁿ bᵢ tⁱ + (1 + t) Q(t) for a polynomial Q with natural number
coefficients. The coefficient of tⁱ in Q is the gap
(cᵢ - cᵢ₋₁ + ⋯ ± c₀) dim V - (bᵢ - bᵢ₋₁ + ⋯ ± b₀) in the i-th strong Morse inequality for
i < n, and zero for i ≥ n.
Euler characteristics of a cobordism relative to its two boundaries. Let f be a Morse
function on a manifold of dimension m, and a ≤ b regular values with compact band
W = f⁻¹[a, b], with incoming boundary ∂₋W = f⁻¹(a) and outgoing boundary ∂₊W = f⁻¹(b). Then
for every n ≥ m, ∑ᵢ₌₀ⁿ (-1)ⁱ dim Hᵢ(W, ∂₊W; V) = (-1)ᵐ ∑ᵢ₌₀ⁿ (-1)ⁱ dim Hᵢ(W, ∂₋W; V): the
Euler characteristics of (W, ∂₊W) and (W, ∂₋W) agree up to the sign (-1)ᵐ.
Finiteness of homology. The singular homology of a compact boundaryless manifold carrying
a Morse function, with coefficients in a finite-dimensional module V, is finite-dimensional in
every degree.
On a compact boundaryless manifold carrying a Morse function, the singular homology with
coefficients in a finite-dimensional module V has rank in ℕ∞ equal to its finrank.
The weak Morse inequalities. For a Morse function f on a compact boundaryless manifold
M, the singular homology Hₙ(M; V) has dimension at most dim V times the number of critical
points of index n.
The strong Morse inequalities. For a Morse function f on a compact boundaryless manifold
M, write bᵢ for the dimension of Hᵢ(M; V) and cᵢ for the number of critical points of
index i. Then bₙ - bₙ₋₁ + ⋯ ± b₀ ≤ (cₙ - cₙ₋₁ + ⋯ ± c₀) dim V.
Euler's equality. For a Morse function f on a compact boundaryless manifold M, write
bᵢ for the dimension of Hᵢ(M; V) and cᵢ for the number of critical points of index i.
Then ∑ᵢ₌₀ⁿ (-1)ⁱ bᵢ = (∑ᵢ₌₀ⁿ (-1)ⁱ cᵢ) dim V for every n ≥ dim M.
The Morse polynomial identity. For a Morse function f on a compact boundaryless manifold
M, write bᵢ for the dimension of Hᵢ(M; V) and cᵢ for the number of critical points of index
i. Then for every n ≥ dim M, ∑ᵢ₌₀ⁿ cᵢ dim V · tⁱ = ∑ᵢ₌₀ⁿ bᵢ tⁱ + (1 + t) Q(t) for a polynomial
Q with natural number coefficients. The coefficient of tⁱ in Q is the gap
(cᵢ - cᵢ₋₁ + ⋯ ± c₀) dim V - (bᵢ - bᵢ₋₁ + ⋯ ± b₀) in the i-th strong Morse inequality for
i < n, and zero for i ≥ n.
The gaps in the strong Morse inequalities are ranks of connecting maps. Let f be a Morse
function on a compact boundaryless manifold M, and s 0 ≤ s 1 ≤ ⋯ ≤ s (N + 1) regular values with
s 0 < f ≤ s (N + 1) on M, such that each band f⁻¹[s j, s (j + 1)] contains at most one
critical value. Write M_t for the sublevel {f ≤ t}, so that M_{s 0} is empty, bᵢ for the
dimension of Hᵢ(M; V) and cᵢ for the number of critical points of index i. Then the gap
(cₙ - cₙ₋₁ + ⋯ ± c₀) dim V - (bₙ - bₙ₋₁ + ⋯ ± b₀) in the n-th strong Morse inequality is the
sum over j < N of the ranks of the connecting maps
Hₙ₊₁(M_{s (j + 2)}, M_{s (j + 1)}) ⟶ Hₙ(M_{s (j + 1)}, M_{s 0}).
The coefficients of the quotient polynomial are ranks of connecting maps. Let f be a
Morse function on a compact boundaryless manifold M, and s 0 ≤ s 1 ≤ ⋯ ≤ s (N + 1) regular
values with s 0 < f ≤ s (N + 1) on M, such that each band f⁻¹[s j, s (j + 1)] contains at
most one critical value. Write M_t for the sublevel {f ≤ t}, so that M_{s 0} is empty, bᵢ
for the dimension of Hᵢ(M; V) and cᵢ for the number of critical points of index i. If
m ≥ dim M and ∑ᵢ₌₀ᵐ cᵢ dim V · tⁱ = ∑ᵢ₌₀ᵐ bᵢ tⁱ + (1 + t) Q(t), then the coefficient of tⁿ
in Q is the sum over j < N of the ranks of the connecting maps
Hₙ₊₁(M_{s (j + 2)}, M_{s (j + 1)}) ⟶ Hₙ(M_{s (j + 1)}, M_{s 0}). Such a Q exists by
TauCeti.IsMorse.exists_sum_range_monomial_eq_add_mul.
The total Betti number is at most dim V times the number of critical points. For a Morse
function f on a compact boundaryless manifold M and every n, ∑ᵢ₌₀ⁿ dim Hᵢ(M; V) is at most
dim V times the number of critical points of f.
The lacunary principle. Let f be a Morse function on a compact boundaryless manifold M
such that no two critical points have indices differing by one. Then for every n the dimension
of Hₙ(M; V) is dim V times the number of critical points of index n.