03.15.12 · modern-geometry / morse-homology

Witten's supersymmetric Morse theory (survey/pointer)

shipped3 tiersLean: none

Anchor (Master): Witten 1982 (*J. Diff. Geom.* 17); Helffer-Sjöstrand 1985 (*Comm. PDE* 10); Bismut 1986 (*Comm. Math. Phys.* 103); Zhang *Lectures on Chern-Weil Theory and Witten Deformations*

Intuition Beginner

Picture a landscape of hills and valleys, and imagine a particle that wants to settle into the lowest place it can find. In quantum physics a particle does not sit perfectly still: it has a faint quantum buzz that spreads it over a small region. Near a valley bottom this buzz produces a tidy ground state, a little cloud hovering over the dip. Witten's idea was to take the height function of a landscape, treat it as the potential energy of such a quantum particle, and then turn up a knob that makes the buzz sharper and sharper. As the knob turns, each cloud shrinks toward exactly one resting point of the landscape.

Here is the payoff. The resting points of a landscape are its bottoms, its saddles, and its tops, and you already know from earlier units that these assemble into a bookkeeping machine that counts the shape's holes. Witten found the same machine hiding inside the quantum problem. Count the sharp clouds and you count the resting points. Watch how clouds at one shelf leak, very slightly, toward clouds one shelf below, and you recover the boundary rule of that machine. The faint leakage between two clouds happens along the single downhill stream that joins their resting points.

That leakage is the quantum effect called tunnelling: a particle slipping through a barrier it could not climb. The size of the leak is governed by the downhill path of steepest descent between the two resting points. So a fact about the topology of a shape, the number and connection of its holes, turns out to be the same as a fact about how a quantum particle tunnels through the barriers of a potential. This unit is a guided tour of that bridge, pointing back to the rigorous construction and out to the physics.

Visual Beginner

Alt text: A two-panel figure. Left panel: a one-dimensional potential with two wells and a barrier; a localised bell-shaped ground state sits in each well, a dashed tunnelling arrow connects them through the barrier, and a knob marked t sharpens the states as it increases. Right panel: the same wells and barrier redrawn as boxes of a Morse complex with a downward boundary arrow, so the tunnelling arrow and the boundary arrow are the same arrow. The picture conveys that the localised quantum ground states are the generators of the Morse complex and that tunnelling between them is the boundary operator.

Worked example Beginner

Take the simplest interesting landscape: a circle standing upright, with a single bottom $b$ and a single top $t$ , and one stream of steepest descent on each side running from the top down to the bottom. The bottom has no downhill directions, so it sits on shelf $0$ ; the top has one downhill direction along the circle, so it sits on shelf $1$ .

Now run Witten's recipe by hand in spirit. Turn up the knob $t$ . A sharp quantum cloud forms over the bottom $b$ and another over the top. The cloud over the bottom is a genuine ground state with no buzz left in it. The cloud over the top is almost a ground state, but it can leak downhill toward the bottom along the two descent streams. The two streams arrive at the bottom from opposite sides, and their two leakage contributions come with opposite signs.

Add the two contributions: one stream gives a leak of $+ 1$ unit, the other gives $- 1$ unit, and the total is $1 - 1 = 0$ . So the boundary of the top is zero. The boundary of the bottom is also zero, since there is nothing below shelf $0$ .

What this tells us: the surviving count is one cloud on shelf $0$ and one cloud on shelf $1$ , with no leakage connecting them. The numbers $1, 1$ are exactly the two holes of the circle, one piece and one loop. The quantum tunnelling computation has recovered the topology of the circle, and the cancellation of the two opposite-sign streams is the same cancellation you met when building the boundary rule directly.

Check your understanding Beginner

Question 1 (easy, multiple-choice). In Witten's picture, what plays the role of the potential energy felt by the quantum particle?

The volume of the manifold.
The Morse function (the height of the landscape).
The number of critical points.
The temperature of the system.

Hint

The landscape's height is what the particle wants to minimise.

Answer

B. The Morse function is read as the potential energy, so its valleys, saddles, and peaks become the features the quantum particle responds to. Feedback if correct: right, the height function becomes the potential, and turning up the deformation knob sharpens the ground states around its critical points. Feedback if wrong: the height function is the potential; the quantum particle localises near the places where that height has a flat tangent, which are exactly the critical points.

Formal definition Intermediate+

Let $(M^{n}, g)$ be a closed oriented Riemannian manifold and $f \in C^{\infty} (M)$ a Morse function. On the de Rham complex $(Ω^{*} (M), d)$ with its exterior derivative $d$ , Witten's deformation is the conjugated differential $$ d_t = e^{-tf}, d, e^{tf} = d + t, df\wedge,,\qquad t \in \mathbb{R}, $$ acting on $Ω^{*} (M)$ , where $df \land$ is exterior multiplication by the one-form $df$ . Because $e^{t f}$ is an invertible bundle automorphism, $d_{t}^{2} = e^{- t f} d^{2} e^{t f} = 0$ , so $(Ω^{*} (M), d_{t})$ is a cochain complex, and the map $α \mapsto e^{- t f} α$ is a cochain isomorphism $(Ω^{*}, d) \to (Ω^{*}, d_{t})$ . Hence the cohomology $H^{*} (Ω^{*}, d_{t})$ is the de Rham cohomology $H_{dR}^{*} (M)$ for every $t$ ^{[Witten 1982]}.

The formal adjoint with respect to the $L^{2}$ inner product on forms is $d_{t}^{*} = e^{t f} d^{*} e^{- t f} = d^{*} + t ι_{\nabla f}$ , where $ι_{\nabla f}$ is contraction by the gradient vector field. The Witten Laplacian is $$ \Delta_t = d_t d_t^* + d_t^* d_t . $$ A Bochner-type computation gives the pointwise identity $$ \Delta_t = \Delta + t^2 |\nabla f|^2 + t, \mathcal{L}f,\qquad \mathcal{L}f = \sum{i,j}\frac{\partial^2 f}{\partial x_i,\partial x_j},\big[ dx^i\wedge,\ \iota{\partial_j}\big], $$ where $Δ = d d^{*} + d^{*} d$ is the Hodge Laplacian 03.04.15 and $L_{f}$ is a zeroth-order operator built from the Hessian of $f$ acting on the form degrees through the commutator $[d x^{i} \land, ι_{\partial_{j}}]$ . The potential term $t^{2} ∣\nabla f ∣^{2}$ vanishes precisely on $Crit (f)$ and grows like $t^{2}$ away from it; this is the harmonic-confinement mechanism that localises the low-lying spectrum.

The supersymmetric quantum mechanics reading interprets $d_{t}$ and $d_{t}^{*}$ as the two supercharges $Q = d_{t} + d_{t}^{*}$ split into raising and lowering parts, $Δ_{t} = Q^{2}$ as the Hamiltonian, and the $Z_{2}$ -grading by form parity as fermion number. Harmonic forms (the kernel of $Δ_{t}$ ) are the supersymmetric ground states; their count by degree is fixed by $H_{dR}^{*} (M)$ , while the number of all critical points of index $k$ counts the would-be ground states before tunnelling lifts the false ones — the spectral content of the Morse inequalities.

Counterexamples to common slips

The deformation $d_{t}$ changes the operator but not the cohomology: conjugation by an invertible operator never changes cohomology, so the harmonic-form count is $b_{k} (M)$ for every $t$ , including $t = 0$ . What varies with $t$ is the geometry of the eigenforms, not their cohomology classes.
The number of small eigenvalues of $Δ_{t}$ acting on $k$ -forms tends, as $t \to \infty$ , to the number $c_{k}$ of index- $k$ critical points, not to the Betti number $b_{k}$ . The Betti number is the number of exactly zero eigenvalues; the remaining $c_{k} - b_{k}$ small eigenvalues are positive but exponentially small in $t$ , and it is their lifting that encodes the boundary operator.
The localising potential is $t^{2} ∣\nabla f ∣^{2}$ , which vanishes on all of $Crit (f)$ regardless of index. Index enters through the zeroth-order Hessian term $t L_{f}$ , whose ground-state energy at a critical point of index $λ$ selects the $λ$ -form sector. Omitting the $t L_{f}$ term loses the grading entirely.

Key theorem with proof Intermediate+

The analytic core of Witten's argument is the localisation of the low-lying spectrum and the resulting count, which already yields the Morse inequalities before any tunnelling computation. The originator statement is Witten's ^{[Witten 1982]}; the rigorous semiclassical form is due to Helffer-Sjöstrand ^{[Helffer-Sjostrand 1985]}.

Theorem (spectral localisation and the Morse inequalities). Let $(M^{n}, g)$ be closed and $f$ a Morse function. For each form degree $k$ there is a gap $δ (t) \to \infty$ and a constant $C$ such that, for $t$ large, the spectrum of $Δ_{t}$ on $Ω^{k} (M)$ splits into a small part contained in $[0, C]$ and a large part contained in $[δ (t), \infty)$ , and $$ \dim\Big(\text{small-eigenvalue space of }\Delta_t\big|_{\Omega^k}\Big) = c_k := #{x\in\mathrm{Crit}(f): \mu(x)=k}. $$ Consequently $c_{k} \geq b_{k} (M)$ for all $k$ (the weak Morse inequalities), and $\sum_{k} (- 1)^{k} c_{k} = \sum_{k} (- 1)^{k} b_{k} = χ (M)$ .

Proof. Work near a critical point $x$ of index $λ$ . By the Morse lemma choose coordinates with $f (y) = f (x) - \frac{1}{2} \sum_{i \leq λ} y_{i}^{2} + \frac{1}{2} \sum_{i > λ} y_{i}^{2}$ and $g$ Euclidean to leading order. In these coordinates $∣\nabla f ∣^{2} = \sum_{i} y_{i}^{2}$ and the Hessian term is constant, so the leading operator is a sum of one-dimensional harmonic oscillators, $$ \Delta_t \approx \sum_{i=1}^n \Big(-\partial_{y_i}^2 + t^2 y_i^2\Big) + t\sum_{i=1}^n \varepsilon_i,\big[dy^i\wedge,\ \iota_{\partial_i}\big],\qquad \varepsilon_i = \pm 1, $$ with $ε_{i} = - 1$ for $i \leq λ$ and $+ 1$ for $i > λ$ . Each scalar oscillator $- \partial_{y_{i}}^{2} + t^{2} y_{i}^{2}$ has spectrum ${t (2 m + 1) : m \geq 0}$ 12.04.02, so its ground energy is $t$ . The Clifford term contributes $\pm t$ per direction depending on whether the form occupies $d y^{i}$ . The total ground energy on a monomial $d y^{I}$ is $$ E_I = t\sum_{i=1}^n 1 + t\sum_{i\in I}\varepsilon_i - t\sum_{i\notin I}\varepsilon_i\quad\text{(schematically)}, $$ and a direct check shows $E_{I} = 0$ exactly when $I = {1, \dots, λ}$ , i.e. for the single $λ$ -form $d y^{1} \land \dots \land d y^{λ}$ aligned with the unstable directions, and $E_{I} \geq 2 t$ otherwise. Thus each index- $λ$ critical point contributes exactly one approximate zero mode, a Gaussian $λ$ -form $\sim e^{- t ∣ y ∣^{2} /2} d y^{1} \land \dots \land d y^{λ}$ concentrated at $x$ , and contributes nothing in other degrees.

These local model ground states are approximately orthonormal and approximately annihilated by $Δ_{t}$ with error exponentially small in $t$ (the genuine eigenforms differ from the Gaussians by $O (e^{- c t})$ tails). A standard min-max / Agmon-estimate argument ^{[Helffer-Sjostrand 1985]} upgrades "approximately $c_{k}$ states with energy $o (t)$ and a gap to the next band at order $t$ " to the exact count: the small-eigenvalue space of $Δ_{t} ∣_{Ω^{k}}$ has dimension $c_{k}$ . Since the genuine kernel of $Δ_{t}$ has dimension $b_{k} (M)$ by Hodge theory applied to the conjugated complex, and the kernel sits inside the small-eigenvalue space, $b_{k} \leq c_{k}$ . The Euler characteristic identity follows because the supersymmetric pairing $d_{t} : Ω_{small}^{k} \to Ω_{small}^{k + 1}$ makes the small-eigenvalue spaces a finite complex with cohomology $H_{dR}^{*} (M)$ , so $\sum (- 1)^{k} c_{k} = \sum (- 1)^{k} b_{k} = χ (M)$ . $□$

Bridge. This theorem builds toward the rigorous Morse complex of 03.15.06: the finite-dimensional small-eigenvalue complex $(Ω_{small}^{*}, d_{t})$ produced here is a Morse complex, with one generator per critical point graded by index, and the foundational reason its cohomology is $H_{dR}^{*} (M)$ is exactly that $d_{t}$ is conjugate to $d$ . What localisation leaves open is the boundary operator: the theorem counts generators but does not yet evaluate the matrix elements $⟨ d_{t} ϕ_{x}, ϕ_{y} ⟩$ between adjacent critical points. The central insight is that those matrix elements are tunnelling amplitudes, and this is exactly the content recovered analytically by Helffer-Sjöstrand and geometrically by the trajectory count of 03.15.06; putting these together identifies the analytic small-eigenvalue complex with the geometric flow-line complex. This pattern appears again in 03.07.23 and in the symplectic Floer chapter, where the same spectral-versus-geometric duality of the differential recurs in infinite dimensions; the bridge is that a deformed Laplacian's low-lying spectrum and a compactified moduli of gradient lines compute the same complex.

Exercises Intermediate+

Exercise 4 (medium, short-answer). Explain why $\dim\ker\Delta_t = b_k(M)$ for all $t$, while the number of small (but possibly nonzero) eigenvalues is $c_k$. What distinguishes the two counts?

Hint

Conjugation invariance fixes the kernel; localisation fixes the small band.

Answer

Rubric. Full credit: the kernel of $Δ_{t}$ is the harmonic forms of the conjugated complex, and since $(Ω^{*}, d_{t}) ≅ (Ω^{*}, d)$ has cohomology $H_{dR}^{*} (M)$ , Hodge theory gives $dim ker Δ_{t} ∣_{Ω^{k}} = b_{k} (M)$ for every $t$ . The small-eigenvalue band, by contrast, has dimension $c_{k}$ for large $t$ because each critical point of index $k$ contributes one localised approximate ground state. The $c_{k} - b_{k}$ extra small eigenvalues are strictly positive but exponentially small in $t$ ; they are the false ground states lifted by tunnelling, and their count is the gap in the weak Morse inequality $c_{k} \geq b_{k}$ .

Exercise 6 (hard, short-answer). State, at the level of leading-order asymptotics, why the off-diagonal matrix element of $d_t$ between the localised states at an index-$k$ point $x$ and an index-$(k-1)$ point $y$ scales like $e^{-t(f(x)-f(y))}$ times a polynomial, and how this exponential is tied to the gradient trajectory from $x$ to $y$.

Hint

The conjugating factor $e^{- t f}$ weights the overlap integral; the dominant contribution is along the steepest-descent path.

Answer

Rubric. Full credit: the localised state $ϕ_{x} \sim e^{- t ∣ y - x ∣^{2} /2}$ carries an implicit weight from the conjugation, and the matrix element $⟨ d_{t} ϕ_{x}, ϕ_{y} ⟩$ is an overlap integral whose integrand is exponentially peaked along the path of steepest descent of $f$ connecting $x$ to $y$ . A Laplace / WKB (steepest-descent) evaluation gives a leading factor $e^{- t (f (x) - f (y))}$ — the exponential of the action $\int ∣\nabla f ∣$ along the gradient line, which equals the height drop $f (x) - f (y)$ — multiplied by a power of $t$ from the Gaussian fluctuation determinant transverse to the path. The instanton of supersymmetric quantum mechanics is precisely this gradient trajectory; summing the signed contributions of all such trajectories between $x$ and $y$ reproduces $n (x, y)$ , the Morse boundary coefficient. Full marks require naming the steepest-descent path as the gradient flow line and identifying the exponent as the height drop.

Exercise 7 (hard, short-answer). The strong Morse inequalities in polynomial form read $M_t - P_t = (1+s)\,Q(s)$ with $Q \ge 0$ (here $s$ is the polynomial variable). Explain how the supersymmetric small-eigenvalue complex $(\Omega^*_{\mathrm{small}}, d_t)$ yields this statement, and what $Q$ measures.

Hint

Apply the standard $(1 + s)$ -divisibility for a finite complex of free modules to the small-eigenvalue complex.

Answer

Rubric. Full credit: the small-eigenvalue spaces form a finite complex $(Ω_{small}^{*}, d_{t})$ with $dim Ω_{small}^{k} = c_{k}$ and cohomology $H_{dR}^{*} (M)$ , so $dim H^{k} = b_{k}$ . For any finite complex of finite-dimensional vector spaces, the counting polynomial $M_{s} = \sum c_{k} s^{k}$ and the Poincaré polynomial $P_{s} = \sum b_{k} s^{k}$ satisfy $M_{s} - P_{s} = (1 + s) Q (s)$ with $Q (s) = \sum q_{k} s^{k}$ , $q_{k} = rank (d_{t} : Ω_{small}^{k} \to Ω_{small}^{k + 1}) \geq 0$ . Thus $Q$ measures the ranks of the boundary maps of the Witten complex, i.e. how many of the $c_{k}$ would-be ground states are lifted by tunnelling at each degree. The inequalities are equalities precisely when $d_{t} = 0$ on the small complex, the perfect-Morse-function case.

Lean formalization Intermediate+

Mathlib does not carry the de Rham complex of a smooth manifold as an unbounded operator, the Hodge Laplacian on $L^{2}$ forms, the Witten conjugation $d_{t} = e^{- t f} d e^{t f}$ , or any semiclassical spectral asymptotics, so this survey unit ships at lean_status: none. The sketch below records the intended statements; it does not compile, because deRhamComplex, hodgeLaplacian, wittenLaplacian, and the small-eigenvalue projection are not defined in Mathlib.

-- Illustrative only; not wired into the Lean build (lean_status: none).

-- The Witten-deformed differential d_t = d + t·(df ∧ ·).
noncomputable def wittenDifferential (M) [RiemannianManifold M] (f : M → ℝ)
    (t : ℝ) : DeRhamForm M →ₗ[ℝ] DeRhamForm M :=
  deRhamDifferential M + t • exteriorMul (deRhamDifferential M f)

-- Conjugation invariance of the cohomology: H*(Ω, d_t) ≅ H*(Ω, d).
theorem witten_cohomology_iso (M) [RiemannianManifold M] (f : M → ℝ) (t : ℝ) :
    Cohomology (wittenDifferential M f t) ≃ₗ[ℝ] deRhamCohomology M := sorry

-- The Witten Laplacian Δ_t = Δ + t²|∇f|² + t·Hess(f)-term.
noncomputable def wittenLaplacian (M) [RiemannianManifold M] (f : M → ℝ)
    (t : ℝ) : DeRhamForm M →ₗ[ℝ] DeRhamForm M := sorry

-- Spectral localisation: for large t the dimension of the small-eigenvalue
-- space of Δ_t on k-forms equals the number of index-k critical points.
theorem witten_small_eigenspace_dim (M) [RiemannianManifold M] (f : M → ℝ)
    (hf : Morse f) (k : ℕ) :
    ∀ᶠ t in Filter.atTop,
      Module.rank ℝ (smallEigenspace (wittenLaplacian M f t) k)
        = (critPointsOfIndex f k).card := sorry

Once deRhamComplex, the Hodge star, and an elliptic-spectral-theory layer for manifolds exist in Mathlib, the conjugation-invariance statement is the accessible first target; the localisation and tunnelling statements require WKB / Agmon-distance machinery entirely absent from the library and are long-range formalisation goals shared with the trajectory-moduli gap of 03.15.06.

Advanced results Master

Beyond the localisation count, Witten's framework gives a complete analytic route to the integral Morse complex and feeds two distinct rigorous developments: the semiclassical tunnelling analysis of Helffer-Sjöstrand and the heat-kernel approach of Bismut.

The tunnelling matrix elements and the differential. After localisation, the small-eigenvalue complex $(Ω_{small}^{*}, d_{t})$ has one generator $ϕ_{x}$ per critical point, graded by index, and its differential is encoded by the matrix elements $⟨ d_{t} ϕ_{x}, ϕ_{y} ⟩$ between generators of adjacent index. Witten argued, and Helffer-Sjöstrand proved ^{[Helffer-Sjostrand 1985]}, that for a Morse-Smale pair these matrix elements are, to leading order as $t \to \infty$ , $$ \langle d_t,\phi_x,\ \phi_y\rangle ;\sim; \Big(\sum_{\gamma\in\widehat{\mathcal{M}}(x,y)} \varepsilon(\gamma)\Big), t^{1/2}, e^{-t,(f(x)-f(y))},\big(1 + O(1/t)\big), $$ a sum over the gradient trajectories from $x$ to $y$ , each weighted by a sign $ε (γ) \in {\pm 1}$ from the relative orientation of the unstable manifolds. The exponential $e^{- t (f (x) - f (y))}$ is the tunnelling suppression set by the height drop along the instanton; the prefactor's coefficient is exactly the signed count $n (x, y)$ of 03.15.06. Rescaling the generators $ϕ_{x} \mapsto e^{t f (x)} ϕ_{x}$ removes the exponential and leaves the integer-valued Morse differential. The semiclassical eigenvalue splittings are correspondingly $\sim e^{- 2 t (f (x) - f (y))}$ , the Agmon-distance asymptotics rigorously established by Helffer-Sjöstrand.

Bismut's heat-kernel proof of the degenerate inequalities. Bismut ^{[Bismut 1986]} reproved the Morse inequalities, including their Morse-Bott (degenerate) generalisation, by analysing the heat kernel $e^{- u Δ_{t}}$ in the coupled limit $t \to \infty$ , $u \to 0$ with $u t$ fixed, localising the supertrace $Tr_{s} e^{- u Δ_{t}}$ near $Crit (f)$ . The supertrace is $t$ - and $u$ -independent and equals $χ (M)$ ; its localisation gives the equivariant version of the local index contributions, and the resulting bound on the off-diagonal heat flow yields the strong inequalities. This is the analytic ancestor of the Bismut-Zhang theorem comparing Ray-Singer and Reidemeister torsions, where the Witten deformation interpolates between the de Rham complex and the Thom-Smale (Morse) complex.

Conjugation invariance as a deformation-retract of complexes. The cochain isomorphism $α \mapsto e^{- t f} α$ shows $(Ω^{*}, d_{t})$ and $(Ω^{*}, d)$ have the same cohomology for all $t$ , but the deeper statement, made precise by the small-eigenvalue projection, is that $(Ω^{*}, d_{t})$ deformation-retracts onto the finite-dimensional Witten complex as $t \to \infty$ , and that finite complex is the Thom-Smale complex of the unstable cells 03.02.31. The Witten deformation is thus a one-parameter family interpolating between the infinite-dimensional analytic complex and the finite-dimensional combinatorial one, with cohomology constant along the path.

Synthesis. The Witten Laplacian is the foundational object of this survey: its $t^{2} ∣\nabla f ∣^{2}$ potential is exactly what localises the spectrum at $Crit (f)$ , and the entire bridge to topology is the statement that the small-eigenvalue complex is the Morse complex. Putting these together, the analytic count of small eigenvalues is the chain group of 03.15.06, the tunnelling matrix elements are its boundary operator, and the conjugation invariance of $d_{t}$ is the reason both compute $H_{*} (M)$ — this is exactly the spectral shadow of the geometric boundary-of-a-boundary identity, and the bridge is that a deformed Laplacian's low-lying spectrum and a compactified moduli of gradient lines define the same differential. The Helffer-Sjöstrand tunnelling asymptotics and Bismut's heat-kernel localisation are dual analytic realisations of one geometric fact, and the central insight is that the rigorous trajectory count of Schwarz ^{[Schwarz Part II]} and Floer ^{[Floer 1989]} is the limit, not a competitor, of Witten's spectral picture. This pattern appears again in the Floer chapters, where the action functional's deformed gradient flow and its formal Laplacian play the same two roles in infinite dimensions.

Full proof set Master

The localisation theorem and the Morse inequalities are proved in the Key theorem section. The structural propositions underlying the survey are recorded here.

Proposition (conjugation invariance). For every $t \in R$ , the map $C_{t} : α \mapsto e^{- t f} α$ is an isomorphism of cochain complexes $(Ω^{*} (M), d) \to (Ω^{*} (M), d_{t})$ , and hence $H^{*} (Ω^{*}, d_{t}) ≅ H_{dR}^{*} (M)$ for all $t$ , with $dim ker Δ_{t} ∣_{Ω^{k}} = b_{k} (M)$ .

Proof. The map $C_{t}$ is a vector-space isomorphism with inverse $α \mapsto e^{t f} α$ . It intertwines the differentials: $d_{t} C_{t} = e^{- t f} d e^{t f} \cdot e^{- t f} = e^{- t f} d = C_{t} d$ , so $C_{t}$ is a cochain map and an isomorphism of complexes. An isomorphism of complexes induces an isomorphism on cohomology, giving $H^{*} (Ω^{*}, d_{t}) ≅ H^{*} (Ω^{*}, d) = H_{dR}^{*} (M)$ . Since $M$ is closed, the conjugated complex is again elliptic and self-adjoint for the same $L^{2}$ structure (the adjoint $d_{t}^{*} = e^{t f} d^{*} e^{- t f}$ is the genuine formal adjoint), so Hodge theory applies: $ker Δ_{t} ∣_{Ω^{k}}$ represents $H^{k} (Ω^{*}, d_{t}) ≅ H_{dR}^{k} (M)$ , whose dimension is $b_{k} (M)$ . $■$

Proposition (harmonic-oscillator ground state at a critical point). At a nondegenerate critical point $x$ of index $λ$ , in Morse coordinates with Euclidean metric, the leading Witten Laplacian on $Ω^{*}$ has a unique normalisable ground state of energy $0$ , occurring in form degree $λ$ , given to leading order by $ψ_{x} = (t / π)^{n /4} e^{- t ∣ y ∣^{2} /2} d y^{1} \land \dots \land d y^{λ}$ ; all other local ground energies are $\geq 2 t$ .

Proof. The leading operator splits over coordinate directions as $\sum_{i} H_{i}$ with $H_{i} = - \partial_{y_{i}}^{2} + t^{2} y_{i}^{2} + t ε_{i} K_{i}$ , where $K_{i} = [d y^{i} \land, ι_{\partial_{i}}]$ acts as $+ 1$ on monomials containing $d y^{i}$ and $- 1$ on those without, and $ε_{i} = - 1$ for $i \leq λ$ , $+ 1$ for $i > λ$ . The scalar part $- \partial_{y_{i}}^{2} + t^{2} y_{i}^{2}$ has ground energy $t$ with ground state $(t / π)^{1/4} e^{- t y_{i}^{2} /2}$ 12.04.02. On a monomial $d y^{I}$ , $H_{i}$ contributes ground energy $t + t ε_{i} σ_{i}$ where $σ_{i} = + 1$ if $i \in I$ and $- 1$ if $i \in / I$ . The total $\sum_{i} (t + t ε_{i} σ_{i})$ is minimised, and equals $0$ , exactly when $ε_{i} σ_{i} = - 1$ for every $i$ , i.e. $σ_{i} = - ε_{i}$ : this forces $i \in I$ for $i \leq λ$ (where $ε_{i} = - 1$ ) and $i \in / I$ for $i > λ$ (where $ε_{i} = + 1$ ), so $I = {1, \dots, λ}$ . Any other $I$ flips at least one sign and raises the energy by $2 t$ . The corresponding eigenfunction is the product Gaussian times $d y^{1} \land \dots \land d y^{λ}$ , normalised as stated. $■$

Proposition (deformation-independence of the Euler characteristic). For any Morse function $f$ on a closed $M$ , $\sum_{k} (- 1)^{k} c_{k} = χ (M)$ , independent of $t$ and of $f$ .

Proof. The supertrace $Tr_{s} e^{- u Δ_{t}} = \sum_{k} (- 1)^{k} Tr e^{- u Δ_{t} ∣_{Ω^{k}}}$ is independent of $u$ and $t$ : independence of $u$ holds because nonzero eigenvalues cancel in pairs between adjacent degrees under the supersymmetry $d_{t}$ (a nonzero eigenform of $Δ_{t}$ in degree $k$ is paired with one in degree $k \pm 1$ of equal eigenvalue), so only the kernel survives the $u \to \infty$ limit, giving $\sum_{k} (- 1)^{k} b_{k} = χ (M)$ ; independence of $t$ holds because $Δ_{t}$ and $Δ_{0}$ are conjugate. Taking instead $t \to \infty$ with the small-eigenvalue localisation, the surviving small band has $c_{k}$ states in degree $k$ , and the same supersymmetric pairing cancels the strictly positive small eigenvalues in adjacent-degree pairs, leaving $\sum_{k} (- 1)^{k} c_{k}$ . Equating the two evaluations of the same $t$ - and $u$ -independent supertrace gives $\sum_{k} (- 1)^{k} c_{k} = χ (M)$ . $■$

The three propositions are proved here in full; the tunnelling asymptotics of the matrix elements are stated without proof in the Advanced results section — see Helffer-Sjöstrand ^{[Helffer-Sjostrand 1985]} for the semiclassical proof and Schwarz ^{[Schwarz Part II]} for the equivalent moduli-counting realisation.

Connections Master

03.15.06 (the Morse complex and $\partial^{2} = 0$ ) is the rigorous target this unit surveys. The small-eigenvalue complex $(Ω_{small}^{*}, d_{t})$ Witten constructs analytically is isomorphic to the Morse-Smale-Witten complex built there by counting gradient trajectories: generators correspond to critical points graded by index, and the tunnelling matrix elements of $d_{t}$ reproduce the signed counts $n (x, y)$ . This unit is the physics-originator layer for that construction; 03.15.06 is its differential-topological realisation.

03.04.15 (Hodge Laplacian on a Riemannian manifold) supplies the operator that gets deformed. The Witten Laplacian $Δ_{t} = Δ + t^{2} ∣\nabla f ∣^{2} + t L_{f}$ is a zeroth-order perturbation of the Hodge Laplacian $Δ$ , and the harmonic-form interpretation of cohomology used throughout this unit is exactly Hodge theory applied to the conjugated complex. Without the Hodge decomposition the identity $dim ker Δ_{t} = b_{k} (M)$ would have no meaning.

12.04.02 (quantum harmonic oscillator) is the local model at every critical point. The leading Witten Laplacian near a critical point is a sum of one-dimensional harmonic oscillators with a Clifford shift, and the ground-state energy $t (2 m + 1)$ of the oscillator is what produces a single zero mode in the unstable-form degree and a gap of order $t$ to all other states. The entire localisation mechanism is the spectral gap of the harmonic oscillator made $t$ -dependent.

03.02.31 (handle attachment, CW homotopy type, and the Morse inequalities) is the classical route to the same inequalities. Where Milnor derives $c_{k} \geq b_{k}$ by attaching cells across critical levels, Witten derives them from the count of small eigenvalues of $Δ_{t}$ ; the finite Witten complex is the Thom-Smale complex of the unstable cells, so the two derivations describe the same finite complex from analytic and combinatorial sides.

12.10.01 (path-integral formulation) is the physical setting in which Witten's argument is naturally phrased: the tunnelling amplitudes are instanton contributions to a supersymmetric-quantum-mechanics path integral, with the gradient trajectories as the instantons and the height drop as the Euclidean action. The semiclassical (large- $t$ ) limit of that path integral is the WKB expansion whose leading term gives the Morse differential.

12.07.04 (WKB and Bohr-Sommerfeld) supplies the semiclassical method behind the tunnelling estimate. The exponential suppression $e^{- t (f (x) - f (y))}$ of the off-diagonal matrix elements is a WKB / steepest-descent evaluation of an overlap integral peaked along the gradient line, and the Agmon-distance asymptotics of Helffer-Sjöstrand are the rigorous multidimensional form of the same Bohr-Sommerfeld-style tunnelling analysis.

Historical & philosophical context Master

Edward Witten introduced the deformation $d_{t} = e^{- t f} d e^{t f}$ and the supersymmetric reading of Morse theory in Supersymmetry and Morse theory (Journal of Differential Geometry 17, 1982, 661–692) ^{[Witten 1982]}, written while he was developing the broader programme that would connect quantum field theory to geometry and topology. The paper observes that the conjugated de Rham complex has the same cohomology as the original for every $t$ , computes the Witten Laplacian and its harmonic-oscillator localisation as $t \to \infty$ , and predicts that the off-diagonal matrix elements between adjacent critical points are tunnelling amplitudes along the connecting instantons — the gradient trajectories — thereby recovering the Morse-Smale-Witten complex as the large- $t$ limit. Witten presented the instanton count as a physical prediction; the analytic justification of the tunnelling asymptotics was left as a programme.

That programme was carried out rigorously by Bernard Helffer and Johannes Sjöstrand in their series on multiple wells in semiclassical mechanics, notably Puits multiples en mécanique semi-classique IV (Communications in Partial Differential Equations 10, 1985, 245–340) ^{[Helffer-Sjostrand 1985]}, who established the exponentially small eigenvalue splittings and the leading tunnelling matrix elements via Agmon estimates. Jean-Michel Bismut gave an independent heat-kernel proof of the Morse inequalities, including the degenerate Morse-Bott case, in The Witten complex and the degenerate Morse inequalities (Journal of Differential Geometry 23, 1986, 207–240) ^{[Bismut 1986]}, localising the supertrace of $e^{- u Δ_{t}}$ near the critical set. The fully geometric moduli-counting realisation of Witten's differential — defining the boundary operator by signed trajectory counts and proving $\partial^{2} = 0$ from the compactified moduli — is due to Andreas Floer, Witten's complex and infinite-dimensional Morse theory (Journal of Differential Geometry 30, 1989, 207–221) ^{[Floer 1989]}, in the infinite-dimensional symplectic setting, and to Matthias Schwarz's finite-dimensional account ^{[Schwarz Part II]}, which is the rigorous construction surveyed in 03.15.06.

Bibliography Master

@article{Witten1982,
  author  = {Witten, Edward},
  title   = {Supersymmetry and {M}orse theory},
  journal = {Journal of Differential Geometry},
  volume  = {17},
  number  = {4},
  pages   = {661--692},
  year    = {1982}
}

@article{HelfferSjostrand1985,
  author  = {Helffer, Bernard and Sj\"ostrand, Johannes},
  title   = {Puits multiples en m\'ecanique semi-classique. {IV}. \'Etude du complexe de {W}itten},
  journal = {Communications in Partial Differential Equations},
  volume  = {10},
  number  = {3},
  pages   = {245--340},
  year    = {1985}
}

@article{Bismut1986,
  author  = {Bismut, Jean-Michel},
  title   = {The {W}itten complex and the degenerate {M}orse inequalities},
  journal = {Journal of Differential Geometry},
  volume  = {23},
  number  = {3},
  pages   = {207--240},
  year    = {1986}
}

@article{Floer1989witten,
  author  = {Floer, Andreas},
  title   = {Witten's complex and infinite-dimensional {M}orse theory},
  journal = {Journal of Differential Geometry},
  volume  = {30},
  number  = {1},
  pages   = {207--221},
  year    = {1989}
}

@book{Schwarz1993,
  author    = {Schwarz, Matthias},
  title     = {Morse Homology},
  series    = {Progress in Mathematics},
  volume    = {111},
  publisher = {Birkh\"auser Verlag, Basel},
  year      = {1993}
}

@book{Zhang2001,
  author    = {Zhang, Weiping},
  title     = {Lectures on {C}hern-{W}eil Theory and {W}itten Deformations},
  series    = {Nankai Tracts in Mathematics},
  volume    = {4},
  publisher = {World Scientific, River Edge, NJ},
  year      = {2001}
}

@incollection{CFKS1987,
  author    = {Cycon, Hans L. and Froese, Richard G. and Kirsch, Werner and Simon, Barry},
  title     = {Schr\"odinger Operators, with Application to Quantum Mechanics and Global Geometry},
  chapter   = {11 (Witten's deformation)},
  series    = {Texts and Monographs in Physics},
  publisher = {Springer-Verlag, Berlin},
  year      = {1987}
}

@article{BismutZhang1992,
  author  = {Bismut, Jean-Michel and Zhang, Weiping},
  title   = {An extension of a theorem by {C}heeger and {M}\"uller},
  journal = {Ast\'erisque},
  volume  = {205},
  year    = {1992}
}

Prerequisites

03.15.06
03.04.15
12.04.02

Tier anchors

beginner: Witten 1982 §1 (the heuristic picture); Hori et al. *Mirror Symmetry* Ch. 10 (supersymmetric quantum mechanics, informal)
intermediate: Cycon-Froese-Kirsch-Simon *Schrödinger Operators* Ch. 11 (Witten's deformation and the Witten Laplacian); Roe *Elliptic Operators, Topology and Asymptotic Methods* Ch. 14
master: Witten 1982 (*J. Diff. Geom.* 17); Helffer-Sjöstrand 1985 (*Comm. PDE* 10); Bismut 1986 (*Comm. Math. Phys.* 103); Zhang *Lectures on Chern-Weil Theory and Witten Deformations*

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Morse functions, gradient flows, and the Morse inequalities (classical background)
Witten — Supersymmetry and Morse theory · Journal of Differential Geometry 17 (1982), pp. 661-692 (the deformed differential $d_t = e^{-tf} d e^{tf}$, the Witten Laplacian, the instanton-tunnelling prediction of the Morse complex)
Helffer-Sjöstrand — Puits multiples en mécanique semi-classique IV · Communications in Partial Differential Equations 10 (1985), pp. 245-340 (rigorous semiclassical tunnelling and the leading exponential asymptotics of the small eigenvalues)
Bismut — The Witten complex and the degenerate Morse inequalities · Journal of Differential Geometry 23 (1986), pp. 207-240 (heat-kernel proof of the Morse inequalities via the Witten deformation)
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part II (the rigorous moduli-counting construction that realises Witten's heuristic differential)
Floer — Witten's complex and infinite-dimensional Morse theory · Journal of Differential Geometry 30 (1989), pp. 207-221 (the moduli-counting realisation of Witten's complex)

Estimated time

beginner: 14m
intermediate: 34m
master: 70m