---
title: Discrete Witten–Morse Theory
url: https://www.emergentmind.com/topics/discrete-witten-morse-theory
type: topic
---

# Discrete Witten–Morse Theory

Searching arXiv for recent and foundational papers on discrete Witten–Morse theory and related formulations.
Discrete Witten–Morse theory denotes a family of constructions in which Witten deformation is used to pass from smooth or combinatorial chain-level data to low-energy complexes generated by critical objects. In the semiclassical SYZ setting, it is the small-$\hbar$ regime of Witten’s deformation of de Rham theory, where wedge products of differential forms and Lie brackets on Kodaira–Spencer complexes are transferred to finite counts of gradient flow lines, gradient trees, and scattering diagrams [1811.09042]. On finite graphs and CW-type settings, it is the discrete analogue of Witten’s supersymmetric construction, with deformed boundary operators, Laplacians, low-energy cut-offs, and Morse inequalities [1704.08354]. On digraphs, it is formulated on path complexes: discrete Morse functions are shown to be flat Witten–Morse functions, Witten complexes of transitive digraphs approach Morse complexes, and a critical-path complex computes path homology under explicit invariance assumptions [2108.08004]. Across these settings, the common mechanism is spectral localization toward critical cells, critical paths, or critical loci, together with a transfer from smooth or full chain complexes to combinatorial structures.

## 1. Semiclassical Witten deformation and localization

In the smooth Morse-theoretic formulation, one starts with a Morse function $f$ on a Riemannian manifold $M$ and semiclassical parameter $\hbar>0$. The Witten-twisted differential and Laplacian are
$$
d_{f,\hbar}=e^{-f/\hbar}de^{f/\hbar}=d+\hbar^{-1}df\wedge,
$$
$$
\Delta_{f,\hbar}=d_{f,\hbar}d_{f,\hbar}^*+d_{f,\hbar}^*d_{f,\hbar},
$$
with
$$
d_{f,\hbar}^*=e^{f/\hbar}d^*e^{-f/\hbar}=d^*+\hbar^{-1}\iota_{\nabla f}.
$$
As $\hbar\to 0$, the spectrum of $\Delta_{f,\hbar}$ separates into a cluster of small eigenvalues near $0$ and a spectral gap above. The small-eigenvalue eigenforms concentrate near critical points of $f$ and along the stable and unstable manifolds. The decay is controlled by the Agmon metric, with distance $\rho_f$ governing the exponential estimate
$$
|\psi_q(x)|\lesssim C e^{-\rho_f(q,x)/\hbar},
$$
for normalized small-eigenvalue eigenforms $\psi_q$ associated to a critical point $q$, and similarly for $d_{f,\hbar}\psi_q$ and $d_{f,\hbar}^*\psi_q$ [1811.09042].

The chain-level comparison with Morse theory is expressed by a linear map
$$
\phi:CM^*(f)\to \Omega^*(M)_{sm},
$$
where $CM^*(f)$ is the Morse cochain complex generated by critical points and $\Omega^*(M)_{sm}$ is the small eigenspace of the Witten Laplacian. The quasimode $\phi(q)$ is localized near $q$ and normalized by
$$
\int_{V_q^-} e^{-f/\hbar}\phi(q)=1+O(\hbar).
$$
Under this identification, $d_{f,\hbar}$ recovers the Morse differential. More precisely, the matrix elements have WKB expansions governed by negative gradient trajectories $\gamma:p\to q$:
$$
\langle d_{f,\hbar}\phi(p),\phi(q)\rangle
=
\sum_{\gamma\in \mathcal{M}(p,q)}\epsilon(\gamma)e^{-[f(q)-f(p)]/\hbar}(1+O(\hbar^{1/2})),
$$
yielding
$$
\partial_{Morse}(p)=\sum_q\sum_{\gamma\in \mathcal{M}(p,q)}\epsilon(\gamma)e^{-\mathcal{A}(\gamma)/\hbar}q,
\qquad
\mathcal{A}(\gamma)=f(q)-f(p).
$$
The same construction extends to pairs of functions $(f_i,f_j)$ via $f_{ij}:=f_j-f_i$ and twisted differentials
$$
d_{ij}=e^{-f_{ij}/\hbar} d e^{f_{ij}/\hbar},
$$
whose small eigenspaces $\Omega^*_{ij}(M)_{sm}$ are canonically identified with Morse complexes $CM^*(f_{ij})$ [1811.09042].

This setting is often called “discrete” not because the ambient geometry ceases to be smooth, but because in the small-$\hbar$ regime the operations are determined by finite counts of gradient trajectories and trees rather than by direct manipulation of smooth differential forms. The survey explicitly distinguishes this from Forman’s discrete Morse theory: the constructions remain smooth-analytic and derive combinatorial structures by semiclassical localization rather than by starting from a CW decomposition [1811.09042].

## 2. Transfer to Morse $A_\infty$ and discrete $L_\infty$ operations

A central theme is the transfer of smooth algebraic operations to discrete ones. In the de Rham dg-category $DR_M$, objects are smooth functions $f_i$, morphism complexes are $\Omega^*(M)$ equipped with $d_{ij}$, and composition is the wedge product $\wedge$. The corresponding Morse-theoretic object is an $A_\infty$ pre-category $\operatorname{Morse}(M)$ with the same objects, morphisms $CM^*(f_{ij})$, Morse differential, and higher products $m_k^{Morse}$ defined by counts of gradient flow trees [1811.09042].

The transfer is implemented by homological perturbation using the Witten Laplacian. For each pair $(i,j)$, let $P_{ij}$ be the orthogonal projection onto the small eigenspace and $G^0_{ij}$ the Green operator for $\Delta_{ij}$. Define
$$
G_{ij}:=(I-P_{ij})G^0_{ij},\qquad H_{ij}:=d_{ij}^*G_{ij},
$$
so that
$$
d_{ij}H_{ij}+H_{ij}d_{ij}=I-P_{ij}.
$$
This gives a homotopy retract of the full de Rham complex onto the small eigenspace. The transferred $A_\infty$ structure $\{\mu^k\}$ is obtained by pulling back wedge product along this retract; its coefficients are produced by iterated applications of $H_{ij}$ along internal edges of planar trees and wedge products at internal vertices.

For a directed trivalent planar $k$-tree $T$, the operation $\mu_T^k$ is defined by inclusion of inputs, wedge at each internal vertex, application of $H_{ab}$ along each internal edge labeled by $(a,b)$, and output projection. Summing over trees yields
$$
\mu^k=\sum_{T\in \mathcal{T}_k}\mu_T^k.
$$
The low-order terms are
$$
\mu^1=\text{restriction of } d_{ij} \text{ to }\Omega^*_{ij}(M)_{sm},
$$
$$
\mu^2(\alpha,\beta)=P_{02}\big(\iota_{12}\alpha\wedge \iota_{01}\beta\big),
$$
$$
\mu^3(\alpha,\beta,\gamma)=P_{03}\big(H_{13}(\alpha\wedge\beta)\wedge \gamma\big)+P_{03}\big(\alpha\wedge H_{02}(\beta\wedge\gamma)\big).
$$
The transfer formula is summarized as
$$
\mu^2=\Pi\circ (\wedge)\circ (\iota\otimes \iota)+\text{higher corrections},
$$
where the higher corrections are rectified by $\mu^3,\mu^4,\dots$ [1811.09042].

The principal identification theorem states that, for generic collections of functions $(f_0,\dots,f_k)$ and critical points $q_{01},\dots,q_{(k-1)k}$,
$$
\mu^k\big(\phi(q_{(k-1)k}),\dots,\phi(q_{01})\big)
=
e^{-A/\hbar}\phi\big(m_k^{Morse}(q_{(k-1)k},\dots,q_{01})\big)+O(\hbar^{1/2}),
$$
with
$$
A=f_{0k}(q_{0k})-f_{01}(q_{01})-\cdots-f_{(k-1)k}(q_{(k-1)k}).
$$
Thus the transferred smooth operations coincide, up to the explicit semiclassical weight, with Morse $A_\infty$ operations counting gradient trees [1811.09042].

On the Kodaira–Spencer side, the same transfer principle appears in $L_\infty$ form. On a complex manifold such as the semi-flat mirror $\check{X}_0\cong (\mathbb{C}^*)^n$, the polyvector fields
$$
PV^{*,*}(\check{X}_0)=\Omega^{0,*}(\check{X}_0,\wedge^*T^{1,0})
$$
carry a dgBV structure $(\bar{\partial},\Delta,\wedge)$ and induced dgLa with differential $D=\bar{\partial}+\Delta$ and Schouten bracket. Relative to a holomorphic volume form $\check{\Omega}$,
$$
\Delta \alpha \,\lrcorner\, \check{\Omega}=\partial(\alpha \,\lrcorner\, \check{\Omega}),
$$
and deformations are governed by the Maurer–Cartan equation
$$
D\check{\phi}+\frac12[\check{\phi},\check{\phi}]=0.
$$
Using a homotopy operator $H$, Kuranishi’s method yields tree-level operations $\ell_k$ defined by brackets at internal vertices and $-H$ on internal and outgoing edges. The solution of
$$
\Phi=a-\frac12H[\Phi,\Phi]
$$
is
$$
\Phi=\sum_{k\ge 1}\ell_k(a,\dots,a),
$$
and the general $L_\infty$ Maurer–Cartan equation is
$$
\sum_{k\ge 1}\frac{1}{k!}\ell_k(\alpha,\dots,\alpha)=0.
$$
The survey describes this as the $L_\infty$-transfer analogue of the $A_\infty$ transfer above: wedge is exchanged for bracket, and differential forms are exchanged for polyvector fields [1811.09042].

## 3. SYZ mirror symmetry, Maurer–Cartan theory, and scattering diagrams

In the SYZ framework, discrete Witten–Morse theory provides a bridge from Kodaira–Spencer deformation theory to scattering diagrams. For dual torus fibrations $X_0\leftrightarrow \check{X}_0$, the fibrewise Fourier transform
$$
\mathcal{F}:\Omega^{*,*}_{cf}(\mathcal{L}X_0)\to PV^{*,*}(\check{X}_0)
$$
identifies differential-geometric data on the loop space with polyvector fields on the mirror. Under $\mathcal{F}$, the de Rham differential on $\check{X}_0$ corresponds to a Witten differential on the loop space,
$$
d_W=d+2\pi i(\dot{\gamma}\,\lrcorner\,(\beta-i\omega))\wedge = e^{-f_m/\hbar}de^{f_m/\hbar},
$$
where $f_m$ is the symplectic area functional attached to a fibrewise loop of homology class $m$. This recasts Kodaira–Spencer deformation theory in Witten–Morse form [1811.09042].

Walls in the Gross–Siebert and Kontsevich–Soibelman sense are codimension-one loci in the affine base $B_0$ equipped with gluing automorphisms. A wall with support $\mathfrak{d}$, primitive normal $n_{\mathfrak{d}}$, and attached function $f_{\mathfrak{d}}$ defines
$$
\theta_{\mathfrak{d}}:x^m\mapsto x^m\cdot \exp\big(\langle n_{\mathfrak{d}},m\rangle\cdot \log f_{\mathfrak{d}}\big).
$$
A scattering diagram $\mathcal{D}$ is a set of such walls, and consistency means that the path-ordered product around any loop avoiding walls is the identity. Wall-attached functions have the form
$$
f_{\mathfrak{d}}=\exp\Big(\sum_{\beta}N_\beta z^\beta\Big),
$$
where $N_\beta$ are counts, such as relative Gromov–Witten or holomorphic disk counts. In the Witten–Morse description, these coefficients are computed combinatorially via counts of gradient trees or tropical disks with the relevant Fourier data [1811.09042].

The survey describes the wall-building mechanism through explicit Maurer–Cartan inputs. Given a single wall $(m,P,\Theta)$, one uses a smoothed delta form
$$
\delta_{-m}(u)=\Big(\frac{\lambda}{\pi}\Big)^{1/2} e^{-\lambda (u^2)^2/2}du^2
$$
with a cut-off $\chi$ supported near $P$, and sets
$$
\alpha_{\mathfrak{w}}
=
-\sum_{k>0}\sum_j\sum_{n\perp m}
a_{jk}^n\,\mathcal{F}(\chi\delta_{-m})(w^{-km}\otimes \check{\partial}_n)t^j.
$$
This solves the Kodaira–Spencer Maurer–Cartan equation. Because $\check{X}_0$ has no nontrivial deformations, $\alpha_{\mathfrak{w}}$ is gauge-equivalent to $0$; after gauge fixing with a chosen homotopy $\check{H}$, the gauge element $\check{\xi}$ has a leading asymptotic term given by a step function with jump $\log(\Theta)$ across $P$ [1811.09042].

For two transversally intersecting initial walls $(m_1,P_1,\Theta_1)$ and $(m_2,P_2,\Theta_2)$, the tree-sum solution of the Maurer–Cartan equation for $\alpha=\alpha^{(1)}+\alpha^{(2)}$ decomposes as
$$
\Phi=\alpha+\sum_a \Phi^{(a)},
$$
where each $\Phi^{(a)}$ is supported near a half-plane $P_a$ of rational slope between the initial walls and is gauge-equivalent to $\log(\Theta_a)$ on one side and $0$ on the other, modulo $O(\hbar^{1/2})$. The associated scattering diagram is monodromy-free:
$$
\Theta_1^{-1}\Theta_2\Big(\prod^{\rightharpoonup}\Theta_a\Big)\Theta_1\Theta_2^{-1}=\mathrm{Id}.
$$
This gives an analytic realization of the Kontsevich–Soibelman consistent completion [1811.09042].

An explicit two-dimensional example starts with two initial walls along the coordinate axes, with automorphisms
$$
\Theta_i=\exp\Big(2\log(1+t_i (w^i)^{-1})\otimes \check{\partial}_{n_i}\Big),\qquad i=1,2.
$$
The consistent completion adds infinitely many walls with slopes $(k+1)/k$ and $k/(k+1)$ supporting
$$
\Theta_{k,k+1}
=
\exp\Big(2\log(1+t_1^k t_2^{k+1}(w^1)^{-k}(w^2)^{-(k+1)})\otimes \check{\partial}_{(-(k+1),k)}\Big),
$$
and also
$$
\Theta_{1,1}
=
\exp\Big(-4\log(1-t_1 t_2 (w^1 w^2)^{-1})\otimes \check{\partial}_{(-1,1)}\Big).
$$
These automorphisms arise from the asymptotic gauges of the Maurer–Cartan solution and satisfy the consistency identity above [1811.09042].

## 4. Finite graphs, supersymmetric formulation, and Morse inequalities

On finite graphs, discrete Witten–Morse theory is formulated in the language of cochains, incidence matrices, and supersymmetric quantum mechanics. For a finite graph $G=(V,E)$ with an orientation of each edge, the cochain spaces are
$$
C^0(G)=\mathbb{R}^{V},\qquad C^1(G)=\mathbb{R}^{E},
$$
and with incidence matrix $I$, the undeformed Laplacians are
$$
\Delta_+ := \mathrm{val}-A = I I^\top,\qquad \Delta_- := I^\top I.
$$
With diagonal weights $W_V>0$ and $W_E>0$, the coboundary and adjoint are
$$
d=B,\qquad d^*=W_V^{-1}B^\top W_E,
$$
hence
$$
\Delta^{(0)}=d^*d,\qquad \Delta^{(1)}=dd^*.
$$
The discrete Hodge theorem identifies
$$
H^0(G;\mathbb{R})\cong \ker(I^\top)=\ker(\Delta_+),\qquad
H^1(G;\mathbb{R})\cong \ker(I)=\ker(\Delta_-),
$$
so the kernel dimensions recover the number of connected components and the cyclomatic number [1704.08354].

A discrete Morse function in the sense used for graphs is a function $f:V\cup E\to \mathbb{R}$ on the cell poset of vertices and edges satisfying Forman-type inequalities. Critical cells are vertices or edges for which both relevant cardinalities vanish. Noncritical cells are paired by a discrete gradient vector field $V_f$, and gradient curves are alternating vertex–edge–vertex sequences descending in $f$ [1704.08354].

The Witten deformation is defined by diagonal matrices
$$
E_V=\operatorname{diag}(e^{sf(v)}),\qquad E_E=\operatorname{diag}(e^{sf(e)}),
$$
and
$$
d_s=E_V d E_E^{-1},\qquad d_s^\dagger=E_E^{-1}d^*E_V.
$$
Equivalently,
$$
d_s=\exp(sf)\,d\,\exp(-sf),\qquad d_s^\dagger=\exp(-sf)\,d^*\,\exp(sf).
$$
The deformed Laplacian is
$$
\Delta_s=d_s^\dagger d_s+d_s d_s^\dagger,
$$
with even and odd blocks
$$
\Delta_{+,s}=d_s d_s^\dagger,\qquad \Delta_{-,s}=d_s^\dagger d_s.
$$
In the unweighted convention,
$$
\Delta_{+,s}=E_V I E_E^{-2} I^\top E_V,\qquad
\Delta_{-,s}=E_E^{-1} I^\top E_V^{2} I E_E^{-1}.
$$
The associated supercharges are $Q=d_s$ and $Q^*=d_s^\dagger$, the Dirac-type operator is $D_s=Q+Q^*$, and the Hamiltonian is
$$
H_s=D_s^2=\Delta_s.
$$
Supersymmetry pairs nonzero-energy eigenstates across degrees, while zero-energy states are harmonic representatives of cohomology [1704.08354].

The graph formulation also gives explicit deformed weighted Laplacians. On vertices,
$$
\big(\Delta_{+,s}u\big)(i)=\sum_{j\sim i}w_{ij}(s)\big(u(i)-u(j)\big),
\qquad
w_{ij}(s)=e^{s(f(i)+f(j)-2f(e_{ij}))},
$$
and the small-$s$ expansion is
$$
w_{ij}(s)=1+s\big(f(i)+f(j)-2f(e_{ij})\big)+\frac{s^2}{2}\big(f(i)+f(j)-2f(e_{ij})\big)^2+\cdots.
$$
On edges,
$$
\big(\Delta_{-,s}w\big)(e)
=
\sum_{v\in \partial e}\sum_{e'\sim_v e}
e^{2sf(v)-sf(e)-sf(e')}I(v,e)I(v,e')w(e').
$$
The paper interprets these as weighted Laplacians on vertices and on the line graph of $G$ [1704.08354].

For any energy cut-off $a\ge 0$, the low-energy subcomplex obtained from eigenspaces with eigenvalues $\le a$ computes the same cohomology as the full complex. In the deformed setting the same remains true for every $s$. As $s\to \infty$, the low-energy spectrum concentrates on critical cells: the matrices $\Delta_{+,\infty}$ and $\Delta_{-,\infty}$ have entries $0$ and $1$, and the number of zero columns equals the number of corresponding critical cells. This yields the discrete Morse inequalities. If $b_k$ are Betti numbers and $c_k$ the numbers of critical $k$-cells, then
$$
c_k\ge b_k,
$$
and the strong inequalities are
$$
\sum_{i=0}^k (-1)^{k-i} c_i \ge \sum_{i=0}^k (-1)^{k-i} b_i,
$$
with equality in top degree giving
$$
\sum_k (-1)^k c_k=\chi(G)=\sum_k (-1)^k b_k.
$$
The paper presents this as the graph-theoretic version of Witten’s proof of Morse inequalities via low-energy spectral localization and SUSY pairing [1704.08354].

## 5. Digraphs, path homology, and flat Witten–Morse functions

For digraphs, the relevant chain model is not the ordinary graph cochain complex but the path complex. A digraph $G=(V,E)$ consists of a finite vertex set and a non-empty edge set $E\subset V\times V\setminus \{\mathrm{diag}\}$; $u\to v$ denotes $(u,v)\in E$. The digraph is transitive if $u\to v$ and $v\to w$ imply $u\to w$, and its transitive closure $G^T$ is the smallest transitive digraph containing it [2108.08004].

An elementary $n$-path is a sequence $v_0v_1\cdots v_n$ of vertices, and $\Lambda_n(V)$ is the $\mathbb{R}$-vector space of formal linear combinations of elementary $n$-paths. The face maps are
$$
d_i(v_0v_1\cdots v_n)=v_0\cdots v_{i-1}v_{i+1}\cdots v_n,
$$
and the standard boundary is
$$
\partial_n=\sum_{i=0}^n (-1)^i d_i.
$$
Allowed elementary $n$-paths are those with $v_{i-1}\to v_i$ an edge for all $i$, and $P_n(G)$ is the vector space they span. Since $\partial$ need not preserve allowed paths, one defines $\Omega_n(G)$ as the subspace of allowed $n$-paths whose boundary remains in $P_{n-1}(G)$. The path homology is
$$
H_m(G;\mathbb{R})=H_m(\{\Omega_n(G),\partial_n\}_{n\ge 0}).
$$
For transitive digraphs, $\Omega_n(G)=P_n(G)$ for all $n$ [2108.08004].

A function $f:V(G)\to [0,+\infty)$ is extended to allowed elementary paths by summing over the vertices:
$$
f(v_0v_1\cdots v_n)=\sum_{i=0}^n f(v_i).
$$
It is a discrete Morse function if for every allowed elementary $n$-path $\alpha$,
$$
\#\{\gamma\in P_{n+1}(G): \gamma>\alpha,\ f(\gamma)=f(\alpha)\}\le 1,
$$
$$
\#\{\beta\in P_{n-1}(G): \beta<\alpha,\ f(\beta)=f(\alpha)\}\le 1.
$$
Critical paths are those for which both cardinalities are zero. A key lemma states that for any allowed $\alpha$, the two noncritical equalities cannot both hold simultaneously: there cannot exist both a face $\beta<\alpha$ and a coface $\gamma>\alpha$ with the same $f$-value [2108.08004].

The paper introduces Witten–Morse and flat Witten–Morse functions on digraphs. A Witten–Morse function satisfies two average inequalities over distinct cofaces and faces. A flat Witten–Morse function satisfies the stronger min/max inequalities
$$
f(\alpha)\le \min\{f(\gamma_1),f(\gamma_2)\}
$$
for two distinct cofaces $\gamma_i>\alpha$, and
$$
f(\alpha)\ge \max\{f(\beta_1),f(\beta_2)\}
$$
for two distinct faces $\beta_i<\alpha$, whenever they exist. The main structural statement is that every discrete Morse function is a flat Witten–Morse function [2108.08004].

The discrete gradient vector field is defined on allowed paths by
$$
(\operatorname{grad}_f)(\alpha)
=
-\sum_{\gamma>\alpha,\ f(\gamma)=f(\alpha)} \langle \partial \gamma,\alpha\rangle \gamma,
$$
and is zero when no such $\gamma$ exists. The discrete gradient flow is
$$
\Phi=\operatorname{Id}+\partial V+V\partial,\qquad V=\operatorname{grad}_f.
$$
This is directly analogous to Forman’s algebraic formula, but here it acts on path complexes of digraphs rather than on cell complexes [2108.08004].

For transitive digraphs, the Witten deformation is defined by
$$
e^{tf}(\alpha)=e^{tf(\alpha)}\alpha,\qquad \partial_t=e^{tf}\partial e^{-tf},
$$
so
$$
\partial_t(\alpha)=\sum_{\beta<\alpha,\ \beta\in \Omega_{n-1}(G)} e^{t[f(\beta)-f(\alpha)]}\beta.
$$
The Witten Laplacian is
$$
\Delta_n(t)=\partial_t\partial_t^*+\partial_t^*\partial_t,
$$
and if $W_n(t)$ denotes the span of eigenvectors whose eigenvalues tend to $0$ as $t\to \infty$, then for transitive $G$ and discrete Morse $f$,
$$
\lim_{t\to \infty} W_n(t)=\operatorname{Crit}_n(G).
$$
The asymptotic formula
$$
\Delta_n(t)\alpha=
\Big[
\sum_{\beta<\alpha}\langle \partial \alpha,\beta\rangle e^{2t(f(\beta)-f(\alpha))}
+
\sum_{\gamma>\alpha}\langle \partial \gamma,\alpha\rangle e^{2t(f(\alpha)-f(\gamma))}
\Big]\alpha
+O(e^{-tc})
$$
shows that noncritical contributions are exponentially suppressed. The paper concludes that the Witten complex approaches the critical-path Morse complex and that their homologies agree with path homology [2108.08004].

## 6. Critical-path complexes, examples, and limitations

For general, possibly non-transitive digraphs, the deformed boundary need not preserve $\Omega_*(G)$. This is one of the main differences from both the smooth setting and the graph/cell-complex setting. The paper therefore constructs a corrected complex using the transitive closure $\overline{G}$ and the $\Phi$-invariant module. On a transitive digraph with discrete Morse function $f$,
$$
P_*^\Phi(G)=\mathbb{R}\text{-span}\{\alpha+V\partial(\alpha): \alpha\in \operatorname{Crit}_*(G)\}.
$$
More generally, considering critical paths in the transitive closure that are still allowed in $G$,
$$
\operatorname{Crit}_n(\overline{G})\cap P_n(G),
$$
one obtains an isomorphism of graded modules
$$
\Phi^\infty|_{\operatorname{Crit}_n(\overline{G})\cap P_n(G)}:
\operatorname{Crit}_n(\overline{G})\cap P_n(G)\to P_n^\Phi(G)\cap \Omega_n(G),
$$
under the hypotheses stated in the paper. The corrected boundary is
$$
\widetilde{\partial}=(\Phi^\infty)^{-1}\circ \partial \circ \Phi^\infty,
$$
and, assuming $V$-invariance of $\Omega_*(G)$ and $\Phi(\alpha)\in \Omega(G)$ for $\alpha\in \operatorname{Crit}(\overline{G})\cap P(G)$,
$$
H_m(\{\operatorname{Crit}_n(\overline{G})\cap P_n(G),\widetilde{\partial}\}_{n\ge 0})
\cong H_m(G;\mathbb{R}).
$$
This gives a critical-path complex computing path homology [2108.08004].

The same paper derives Morse inequalities for digraphs. Writing
$$
b_m=\dim H_m(G;\mathbb{R}),\qquad
l_m=\dim(\operatorname{Crit}_m(\overline{G})\cap P_m(G)),\qquad
L_m=\dim \operatorname{Crit}_m(\overline{G}),
$$
the weak inequalities are
$$
L_m\ge l_m,\qquad l_m\ge b_m,
$$
and the strong inequalities are
$$
l_m-l_{m-1}+\cdots \pm l_0 \ge b_m-b_{m-1}+\cdots \pm b_0.
$$
The Euler characteristic bound is
$$
l_0-l_1+\cdots \pm l_{\dim G}\ge b_0-b_1+\cdots \pm b_{\dim G}.
$$
An equivalent polynomial form is also recorded:
$$
\sum_{k\ge 0} l_k t^k-(1+t)Q(t)=\sum_{k\ge 0} b_k t^k
$$
for some polynomial $Q(t)$ with nonnegative coefficients [2108.08004].

The examples in the literature display how these abstractions are computed. In the square digraph example with
$$
V=\{v_0,v_1,v_2,v_3\},\qquad
E=\{v_0\to v_1,\ v_0\to v_2,\ v_1\to v_3,\ v_2\to v_3\},
$$
the transitive closure adds $v_0\to v_3$. The path complex has
$$
\Omega_2(G)=\operatorname{span}\{v_0 1 3-v_0 2 3\},
$$
and the explicit gradient flow satisfies, among other formulas,
$$
\Phi(v_0)=v_1,\qquad
\Phi(v_0 2)=v_0 2-v_0 1,\qquad
\Phi(v_2 3)=v_2 3-v_1 3.
$$
The resulting corrected critical-path complex has
$$
H_0\cong \mathbb{R},\qquad H_m=0\ \text{for }m\ge 1,
$$
in agreement with path homology. A second example with six vertices yields
$$
H_0\cong \mathbb{R},\qquad H_1\cong \mathbb{R},\qquad H_m=0\ \text{for }m\ge 2,
$$
again matching path homology [2108.08004].

The literature also records explicit finite-graph examples for the Witten deformation itself. For the graph $K_2$ with one edge and a Morse function satisfying $f(v_1)=0$, $f(v_2)=1$, $f(e)=0$, the deformed Laplacians satisfy
$$
\Delta_{+,\infty}=
\begin{bmatrix}
0&0\\
0&1
\end{bmatrix},
\qquad
\Delta_{-,\infty}=[1],
$$
so the kernel of the even Laplacian is spanned by the unique critical vertex and the odd kernel vanishes. For the graph $K_3$, one choice of Morse function yields one critical vertex and one critical edge, and the limiting kernels are exactly those spans; another height-type function makes all cells critical and produces $\Delta_{\pm,\infty}=0$ [1704.08354].

Several limitations recur across the three settings. In the SYZ semiclassical theory, the construction requires Morse genericity, transversality of gradient trees, orientation choices, small-eigenvalue separation, Agmon-type decay estimates, and control of inhomogeneous Witten equations, with errors typically of order $O(\hbar^{1/2})$ [1811.09042]. In digraphs, transitivity is crucial for $\partial_t$ to define a Witten complex on $\Omega_*(G)$, and the paper gives an example showing that on a non-transitive digraph the Witten deformation may fail to be a chain complex [2108.08004]. In finite graphs and cell complexes, the identification of the low-energy sector with the Morse complex is asymptotic in the large deformation limit, even though the deformed Hodge theorem holds for every deformation parameter $s$ [1704.08354].

Taken together, these formulations show that “discrete Witten–Morse theory” is not a single definition but a coherent pattern. In the smooth SYZ setting it means semiclassical transfer of $\wedge$, brackets, and Maurer–Cartan theory to combinatorial trees and scattering diagrams [1811.09042]. In finite graphs and cell complexes it means Witten deformation of boundary operators, supersymmetric Laplacians, low-energy cut-offs, and Morse inequalities [1704.08354]. In digraph path homology it means flat Witten–Morse functions, Witten complexes converging to critical-path complexes on transitive digraphs, and a corrected critical-path model for general digraphs [2108.08004]. A plausible implication is that the unifying content of the subject is spectral localization: analytic or algebraic deformation reorganizes a full complex so that topology and higher operations are recovered from critical generators and their discrete incidence data.

Source: https://www.emergentmind.com/topics/discrete-witten-morse-theory