Discrete Green's Identity

• Discrete Green's Identity is a set of formulas in discrete analysis that relate operators such as graph Laplacians or difference operators to kernel or pseudo-inverse representations encoding boundary and point-evaluation data.
• It appears in various settings including random walks on graphs, planar quad-graphs, lattice difference equations, and finite-element discretizations, each adapting the classical integration-by-parts notion.
• The framework provides practical insights into operator inversion, optimal stopping rules, and numerical challenges like positivity preservation in discretized simulations.

Discrete Green’s identity denotes a family of formulas in discrete analysis that relate a discrete operator, its inverse or fundamental solution, and boundary, source, or point-evaluation data. Depending on the setting, the operator may be a graph Laplacian, a constant-coefficient difference operator on $\mathbb{Z}^n$, a finite-element discretization of an elliptic or convection–diffusion problem, or a fractional nabla-difference operator. Correspondingly, the Green object may be a pseudo-inverse, a boundary-value kernel, or a variational representer of a point source. On planar quad-graphs, the notion appears as literal discrete analogues of Green’s first and second identities; on graphs and in numerical discretizations, it often appears instead as an operator equation or bilinear-form representation carrying the same integration-by-parts content (Bobenko et al., 2015, Beveridge, 2015).

1. Range of meanings in discrete settings

The term does not refer to a single universally fixed formula. In the literature, it covers several structurally related constructions: summation-by-parts identities, representation formulas for point evaluation, pseudo-inverse identities for discrete Laplacians, and existence statements for discrete fundamental solutions. What unifies them is the passage from a discrete operator to a kernel or functional that encodes inversion, boundary behavior, or source localization.

The following correspondences are explicit in the cited works.

Setting	Operator	Green relation
Random walk on a graph	$\mathbb{L}=I-P$	$$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$
Planar quad-graph	$\triangle$	Green’s first and second identities
Lattice difference equations	$P$	$Pf=\delta$
Standard finite elements	$-\Delta_h$	$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$
SDFEM on Shishkin meshes	$a_{SD}$	$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$
Nabla fractional BVPs	$\mathbb{L}=I-P$0	$\mathbb{L}=I-P$1

A recurrent feature is normalization. In graph Laplacians on finite state spaces, constants lie in the kernel, so the Green object is a pseudo-inverse rather than an ordinary inverse. In boundary-value formulations, the kernel is instead determined by boundary conditions and jump or forcing constraints. In variational discretizations, the Green object is the Riesz representer of evaluation under the discrete bilinear form.

2. Graph Laplacians, hitting times, and exact stopping rules

For a graph $\mathbb{L}=I-P$2, the discrete Laplace operator can be written as

$\mathbb{L}=I-P$3

where $\mathbb{L}=I-P$4 is the random-walk transition matrix. In the undirected case,

$\mathbb{L}=I-P$5

and the stationary distribution is

$\mathbb{L}=I-P$6

The discrete Green’s function $\mathbb{L}=I-P$7 is defined as the pseudo-inverse characterized by

$\mathbb{L}=I-P$8

These conditions say that $\mathbb{L}=I-P$9 inverts the Laplacian on the subspace orthogonal to constants and is uniquely normalized by zero row sums (Beveridge, 2015).

The central formula expresses $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$0 directly through random-walk hitting times: $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$1 Here $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$2 is the expected time for a walk started at $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$3 to hit $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$4, and

$$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$5

is the stationary-average hitting time to $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$6. This formula gives a discrete Green-type representation in which the inverse of the Laplacian is encoded by expectations of stopping times rather than by purely algebraic inversion.

The same work identifies a second layer of structure through optimal stopping rules. For a stopping rule $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$7, the access time is

$$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$8

and the exit frequencies $$\mathbb{G}\mathbb{L}=I-\mathbf{1}\pi^\top,\ \mathbb{G}\mathbf{1}=0$$9 satisfy the conservation law

$\triangle$0

An optimal rule exists precisely when some vertex is never exited,

$\triangle$1

and the exit frequencies admit the explicit formula

$\triangle$2

For the stationary target $\triangle$3, the Green matrix is

$\triangle$4

or entrywise,

$\triangle$5

This identifies the discrete Green’s function as a matrix of exit frequencies plus a rank-one correction enforcing the normalization $\triangle$6.

The graph-theoretic viewpoint also extends to arbitrary target distributions $\triangle$7: $\triangle$8 with

$\triangle$9

In this formulation, discrete Green’s identity is not a boundary integral formula but an operator-and-expectation identity linking Laplacian inversion, hitting times, and exact stopping behavior.

3. Green’s first and second identities on planar quad-graphs

A literal discrete analogue of Green’s identities is developed for planar bipartite quad-graphs. The framework uses the medial graph $P$0, discrete exterior calculus, and a discrete Laplacian

$P$1

For functions $P$2, the Laplacian factorizes as

$P$3

This provides the structural analogue of the factorization of the continuous Laplacian by Cauchy–Riemann operators (Bobenko et al., 2015).

Let $P$4 be finite, and let $P$5. The discrete Green’s first identity is

$P$6

The discrete Green’s second identity is

$P$7

The boundary term is a contour integral over the boundary of the medial graph. In this way, the continuous boundary pairing $P$8 is replaced by a discrete contour integral built from the Hodge star and exterior derivative.

The proof mirrors the smooth case. The decisive ingredients are the derivation property

$P$9

and the discrete Stokes theorem

$Pf=\delta$0

Choosing $Pf=\delta$1 yields the first identity, and the second follows by skew-symmetrization. The work explicitly states that this is the first formulation of a discrete Green’s first identity in this quad-graph setting.

These identities underpin several further results. The discrete Dirichlet energy is

$Pf=\delta$2

and the Green identities support variational characterizations of discrete harmonicity, uniqueness of the discrete Dirichlet problem, Weyl’s lemma, and harmonic-conjugate theory. Here, discrete Green’s identity is a full discrete exterior-calculus theorem rather than only an operator representation.

4. Fundamental solutions on $Pf=\delta$3 and discrete operator inversion

For constant-coefficient difference operators on the lattice, the Green object appears as a fundamental solution. If

$Pf=\delta$4

with finite $Pf=\delta$5, then the symbol is the Laurent polynomial

$Pf=\delta$6

The discrete analogue of the Malgrange–Ehrenpreis theorem states that for every non-zero such operator there exists a function $Pf=\delta$7 on $Pf=\delta$8 such that

$Pf=\delta$9

where $-\Delta_h$0 is the discrete delta at the origin (Zeilberger, 2011).

This result does not present Green’s first or second identity in the classical PDE format. Instead, its first proof uses a duality mechanism that plays the same structural role. A lattice function $-\Delta_h$1 determines a linear functional $-\Delta_h$2 on Laurent polynomials by

$-\Delta_h$3

and one seeks $-\Delta_h$4 satisfying

$-\Delta_h$5

for every Laurent polynomial $-\Delta_h$6. Multiplication by the symbol on the polynomial side corresponds to application of the difference operator on the lattice side. This is the discrete analogue of moving an operator across a pairing, which in continuous analysis is the mechanism behind integration by parts and Green’s formulas.

The second proof is constructive. Associating to $-\Delta_h$7 the bilateral Laurent series

$-\Delta_h$8

the equation $-\Delta_h$9 becomes

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$0

Hence

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$1

The right-hand side has $(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$2 possible formal Laurent expansions, each producing a discrete Green function. The accompanying Maple package LEON, through the procedure FS, implements this constructive expansion. In this lattice setting, discrete Green’s identity is best understood as the existence and construction of $(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$3, together with the duality relation that replaces explicit summation-by-parts formulas.

5. Variational identities in finite-element and stabilized discretizations

In finite-element analysis, discrete Green’s identity is often the variational statement that point evaluation is represented by the bilinear form. For a fixed $(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$4, the discrete Green’s function $(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$5 is defined by

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$6

The discrete Laplacian is

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$7

so the Green function satisfies

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$8

This is the finite-element analogue of integration by parts and point-source representation (Miller, 2023).

The same framework yields quantitative information. In three dimensions,

$(\nabla G_h^z,\nabla v_h)_\Omega=v_h(z)$9

so the discrete Green’s function is not uniformly bounded in $a_{SD}$0 at the singularity. The paper also proves exponential decay away from the singularity,

$a_{SD}$1

and positivity away from a neighborhood of size $a_{SD}$2: if

$a_{SD}$3

then

$a_{SD}$4

At the same time, numerically persistent negative values are reported on unstructured three-dimensional Delaunay meshes. The stated implication is that a discrete Harnack inequality cannot be established for unstructured finite-element discretizations without stronger mesh restrictions, such as an $a_{SD}$5-matrix condition.

For the streamline diffusion finite element method on Shishkin triangular meshes, the same representer idea is adapted to a convection–diffusion bilinear form. The discrete Green’s function $a_{SD}$6 is defined by

$a_{SD}$7

The analysis then uses weighted identities obtained by integration by parts, for example

$a_{SD}$8

where $a_{SD}$9 is the nodal interpolant. For problems with only exponential layers, this leads to the estimate

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$0

while for singularly perturbed problems with characteristic layers the final weighted estimate is region-dependent and takes the form

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$1

These SDFEM papers do not isolate a standalone theorem named “discrete Green’s identity,” but their analysis is built on the exact adjoint representation of nodal evaluation and on weighted integration-by-parts formulas (Zhang, 2017, Liu et al., 2017).

6. Fractional-difference boundary value problems and interpretive issues

For higher-order boundary value problems involving the nabla Caputo fractional difference, the operator

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$2

serves as a discrete fractional analogue of a self-adjoint differential operator. The associated Green theorem states that if the homogeneous boundary value problem has only the trivial solution, then the unique solution of the nonhomogeneous problem with zero boundary data is

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$3

The Green kernel $a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$4 is defined by combining a homogeneous solution $a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$5 satisfying the left boundary conditions with the Cauchy function $a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$6, via

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$7

and a two-branch definition of $a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$8 on the relevant index ranges (Ahrendt et al., 2018).

The operational substitute for classical integration by parts is the discrete Leibniz rule

$a_{SD}(v^N,G)=v^N(\mathbf{x}^*)$9

together with its shifted form

$\mathbb{L}=I-P$00

These identities are used iteratively in the proof of the variation-of-constants formula and play the role ordinarily served by Green-type integration-by-parts arguments.

Across these settings, two misconceptions recur. First, discrete Green’s identity need not always mean a literal discrete version of Green’s first identity with a boundary term; in several important frameworks it appears instead as a pseudo-inverse relation, a duality formula, or a variational representer of point evaluation. Second, properties familiar from symmetric elliptic theory do not automatically persist under discretization. On unstructured three-dimensional finite-element meshes, positivity is delicate and may fail numerically away from the singularity (Miller, 2023). In the fractional nabla setting, the operator is motivated as a self-adjoint analogue, but the cited work does not prove a full adjointness theorem with a bilinear-form identity (Ahrendt et al., 2018).

Taken together, the literature presents discrete Green’s identity not as a single formula but as a coherent principle: discrete operators admit Green objects that convert local forcing, point evaluation, hitting data, or boundary constraints into global representations. The exact syntax varies—from

$\mathbb{L}=I-P$01

to

$\mathbb{L}=I-P$02

but the underlying content is stable: a discrete inverse or adjoint relation is expressed in a form that exposes conservation, normalization, and representation.