TOPICS
Search

Discrete Harmonic Function


A discrete harmonic function on a finite connected simple graph G=(V,E) with at least two vertices is a real-valued function whose value at each specified graph vertex is the arithmetic mean of its values at the adjacent vertices. Writing w∼v for adjacent vertices, this condition is

 f(v)=1/(deg(v))sum_(w∼v)f(w),
(1)

where deg(v) is the vertex degree. This averaging condition is a discrete analog of the condition satisfied by a harmonic function.

More generally, assign a positive real number c_(vw)=c_(wv) to each graph edge. The weighted harmonicity condition replaces the arithmetic mean by

 f(v)=(sum_(w∼v)c_(vw)f(w))/(sum_(w∼v)c_(vw)).
(2)

The unweighted condition is recovered when all c_(vw)=1 (Doyle and Snell 2006, p. 36).

For a nonempty subset B of the vertex set V, arbitrary real values prescribed on B have a unique extension to V that is harmonic at every graph vertex in V\B. Its values lie between the smallest and largest prescribed values. To prove this bound, suppose a global largest value is attained outside B. Positivity in the averaging equation forces every graph vertex adjacent to it to have the same value. Repeating this argument along a graph path to B shows that the largest value is attained on B. Applying the same argument to -f gives the lower bound.

The difference of two extensions with the same boundary values is harmonic and zero on B, so the bounds force it to vanish everywhere. This proves uniqueness. Existence is immediate if B=V. Otherwise the interior averaging equations form a linear system of equations with as many equations as unknown interior values. With zero boundary values, the uniqueness argument gives only the zero solution, so its coefficient matrix is nonsingular and every choice of boundary values has a solution. This extends the averaging argument for finite grids and two-boundary networks to any nonempty B (Doyle and Snell 2006, pp. 13-14 and 36; Nachmias 2020).

The definition also applies to infinite locally finite graphs. On the integer lattice Z^d, with edges between points differing by one standard basis vector, harmonicity is the condition that each value is the average of its 2d nearest-neighbor values. Hu (2026) reports the following one-sided Liouville and sparse-support results for this unweighted lattice.

There is an absolute epsilon>0 such that a real harmonic function on Z^2 is constant if it is nonnegative at at least a proportion 1-epsilon of the points in every sufficiently large centered square. More strongly, for every nonconstant such function and every fixed real a<=b, both sets {u<a} and {u>b} occupy an absolute positive proportion of the same arbitrarily large centered squares. The conclusion need not hold at every large scale: a single function can be positive on a proportion tending to 1 along one sequence of squares and negative on a proportion tending to 1 along another. Each fixed strict level set nevertheless contains at least cN^2/lnN points in every sufficiently large square [-N,N]^2 intersection Z^2, for an absolute c>0, and the logarithmic order is attained along a sequence (Hu 2026).

In dimension three, put Q_R=[-R,R]^3 intersection Z^3. For each A>0 there is a constant K_A such that, for every integer R>=1, a complex-valued function harmonic in Q_(30R) with at most AR^2 nonzero values there has its support in Q_R contained in at most K_A planes x_i+/-x_j=b, where 1<=i<j<=3 and b is an integer. For a fixed nonzero whole-lattice harmonic function, the following are equivalent: its support has O(N^2) points in Q_N along an unbounded sequence; the support lies in finitely many such planes; and its complex closure in the Zariski topology is a proper subset of C^3. There is then a unique irredundant family of k planes, and the support count is k(2N+1)^2+O_u(N) (Hu 2026).

If m_3(N) is the minimum support count in Q_N over whole-lattice harmonic functions with u(0)!=0, then

 m_3(N)=(1+o(1))(2N+1)^2.
(3)

The upper bound is attained by the example u(x,y,z)=(-1)^x(3+sqrt(8))^z when x=y and u(x,y,z)=0 otherwise. Just above the quadratic scale, the support can be much less rigid: for every positive H(N)->infty, there is a real harmonic function whose translates are dense in the space of real whole-lattice harmonic functions for pointwise convergence and for which |supp(u-1) intersection Q_N|/(2N+1)^2 tends to infinity while remaining at most H(N) for all sufficiently large N (Hu 2026).

In every dimension, if a real lattice harmonic function is nonnegative outside a proper real algebraic set, it agrees there with a unique nonnegative constant c. The vanishing ideals of {u<0} and supp(u-c) coincide. For every nonconstant real harmonic function on Z^d, d>=2, the complex Zariski closure of each fixed strict level set has dimension at least |_d/2_|+1, and this bound is sharp (Hu 2026). Hu (2026) reports that AI agents developed the characteristic-plane inverse argument and the contact-set argument for the planar theorem.


See also

Graph, Harmonic Function, Laplacian Matrix

Explore with Wolfram|Alpha

References

Doyle, P. G. and Snell, J. L. Random Walks and Electric Networks. Version of 5 Jul 2006. https://math.dartmouth.edu/~doyle/docs/walks/walks.pdf.Hu, X. "A One-Sided Liouville Theorem and a Critical Sparse-Support Inverse Theorem for Discrete Harmonic Functions." 19 Nov 2026. https://arxiv.org/abs/2611.25952v1.Nachmias, A. "Random Walks and Electric Networks." In Planar Maps, Random Walks and Circle Packing. Cham, Switzerland: Springer, pp. 11-31, 2020. https://doi.org/10.1007/978-3-030-27968-4_2.Numberphile. "Squares and Tilings." Featuring A. Okounkov. Aug. 12, 2020. https://www.youtube.com/watch?v=Rpjab-XQ0U.

Cite this as:

Weisstein, Eric W. "Discrete Harmonic Function." From MathWorld--A Wolfram Resource. https://mathworld.wolfram.com/DiscreteHarmonicFunction.html

Subject classifications