A Two-Grid Finite Element Method for the Nonlinear Poisson-Boltzmann Equation Based on a Solution Decomposition

• The PBE’s numerical challenges include point-charge singularities, a dielectric jump at the molecular surface, and an exponentially dependent ionic charge density.Source: §1 · p. 1
• The decomposition assigns point-charge singularities to G, boundary and interface conditions to Ψ, and the ionic nonlinearity to bΦ.The bΦ problem permits any number of ionic species.Source: §1 · p. 2
• Repeated fine-mesh linear solves in Newton-type methods motivate the coarse nonlinear solve followed by a fine-grid correction.Earlier analysis for regularized PBE required a discrete maximum principle, a restriction avoided here.Source: §1 · p. 2
• The method applies a two-grid finite element scheme to Xie’s decomposition, solving the nonlinear problem on a coarse mesh and using one fine-mesh Newton correction.Only two symmetric positive definite linear systems are solved on the fine mesh.Source: §1 · pp. 2–3
• A C2 truncation of the exponential matches the original problem above an explicit threshold, so the error analysis requires no discrete maximum principle.The truncation grows quadratically above τ; for τ above the stated bound, the truncated and original problems have the same solution.Source: §1 · p. 2
• The choice H ∼ h^((s−1)/(2(s−1)+t/2)) keeps the accuracy of the nonlinear fine-mesh solution; when σ ≥ s, in particular s = σ, this is H ∼ h^(2/5).Source: §1 · p. 3

2.1 The dimensionless PBE model

• The dimensionless model partitions a bounded domain into solute and solvent regions with piecewise constant dielectric permittivity and ionic species in the solvent.The stated setting assumes 0 < ϵp ≤ ϵs and electrically neutral bulk solution.Source: §2.1 · pp. 3–4

2.2 Solution decomposition

• The decomposition writes u as G + Ψ + bΦ: G is the Coulomb potential, Ψ handles boundary and interface conditions, and bΦ carries ionic charges.The nonlinearity acts only in the solvent, and the decomposition permits any number of ionic species.Source: §2.2 · pp. 4–5

2.3 Notation and weak formulations

• The weak formulations define a linear interface problem for Ψ and a semilinear problem for bΦ; the latter has a unique solution under the stated assumptions.The linear problem has a unique solution when the boundary data belongs to H^(1/2)(∂Ω).Source: §2.3 · p. 5

3 A C2 truncation of the nonlinearity and a priori bounds

• The exponential-growth nonlinearity complicates finite element error analysis because an H1 bound does not control it globally.The approach replaces the exponential with a C2 function of quadratic growth and proves the solution is unchanged.Source: §3 · p. 6, §3.1 · p. 6

3.1 The truncated exponential

• Eτ equals e^y up to τ and continues quadratically beyond τ, with positive bounded growth estimates for Eτ and its derivative.The truncated ionic density is strictly increasing, and its primitive is convex.Source: §3.1 · pp. 6–7

3.2 The truncated problem and its well-posedness

• For every τ > 0, the truncated problem has a unique H1_0(Ω) solution that uniquely minimizes its associated functional.Source: §3.2 · p. 8

3.3 A priori L∞bounds and equivalence with the original problem

• When τ ≥ 2Zmax MU, the truncated and original problems have the same unique solution, with the solution bounded in magnitude by MU.The threshold makes the truncation inactive at the solution; the practical overflow safeguard therefore preserves the target problem.Source: §3.3 · p. 9

4.1 Meshes, spaces and assumptions

• Fitted, shape-regular tetrahedral meshes and piecewise regularity assumptions provide the framework for the approximation estimates.The meshes conform to the molecular interface; quasi-uniformity is used only to relate mesh sizes to numbers of unknowns.Source: §4.1 · p. 10

4.2 The linear component

• The linear component Ψ is approximated with energy- and L2-error estimates governed by the piecewise regularity exponents s and σ.Its discretization imposes interpolated nonhomogeneous Dirichlet data; the estimates assume smooth boundary data.Source: §4.2 · pp. 11–12

4.3 The nonlinear component: discrete problem and stability

• The discrete nonlinear problem has a unique solution with a bound independent of mesh size.The analysis applies to the truncated nonlinearity and permits the approximate input U* + Φh to be unbounded pointwise.Source: §4.3 · p. 13

4.4 Energy-norm error estimate

• For sufficiently small h, the nonlinear-component energy error is controlled under the stated regularity assumptions, without pointwise bounds on Φh or mesh-angle conditions.The quadratic growth of the truncated nonlinearity controls the quadratic error term, which can be absorbed for small h; unlike [19, Theorem 3.5], the proof needs no pointwise bound on Φh.Source: §4.4 · pp. 13, 15

4.5 L2 and L4 error estimates

• The nonlinear-component L2 and L4 estimates follow under an additional data-perturbation condition, using the bounded second derivative of the truncated nonlinearity.With the exponential itself, the corresponding L2-estimate step would require pointwise bounds on Φh, as in [19, Theorem 3.7].Source: §4.5 · p. 16
• Corollary 4.10 gives an error estimate for the fine-mesh solution uh = G + Ψh + Φh under its stated assumptions.Its proof decomposes u − uh into the linear-component error Ψ − Ψh and the nonlinear-component error bΦ − Φh.Source: §4.5 · p. 16
• For continuous piecewise polynomials of degree k ≥ 1, piecewise Hs regularity with 1 < s ≤ k + 1 permits the section’s results to extend with r = s.The same extension applies to the results of Section 5.Source: §4.5 · p. 16
• For molecular surfaces, limited regularity of bΦ means higher-order elements pay off only when the interface is also approximated to higher order.Source: §4.5 · p. 16

5.1 The algorithm

• The nonlinear coarse problem is followed by one fine-mesh Newton correction, leaving only two symmetric positive definite fine-mesh systems to solve.The two systems are for Ψh and the correction dh; both can be solved by preconditioned conjugate gradients with multilevel preconditioners robust to the dielectric jump.Source: §5.1 · pp. 17–18

5.2 Error analysis

• The two-grid solution differs from the fine-mesh nonlinear solution by a quantity quadratic in the coarse-mesh error.The coarse problem may use either the fine-mesh or coarse-mesh datum; both variants are covered by the analysis.Source: §5.2 · pp. 18–19
• When σ ≥ s, H ∼ h^2/5 gives the two-grid solution the same asymptotic accuracy as the fine-mesh solution.For σ < s, the coarse mesh must be somewhat finer relative to the fine mesh.Source: §5.2 · p. 19
• The coarse mesh must still resolve the molecular surface, and geometry-dependent constants can make a somewhat finer coarse mesh preferable.For quasi-uniform meshes with γ = 2/5, NH ∼ h^-6/5 ∼ N^2/5.Source: §5.2 · p. 19

5.3 The symmetric 1:1 electrolyte

• For a symmetric 1:1 electrolyte, τ ≥ 2MU gives the two-grid error conclusions, and the regular part satisfies |bΦ| ≤ MU almost everywhere.Here the ionic species have opposite unit charges and equal positive bulk concentrations.Source: §5.3 · p. 20
• The error-estimate constants depend exponentially on the truncation level and on MU, which can be large for highly charged solutes.The convergence-rate exponents do not depend on the choice of τ ≥ τ0.Source: §5.3 · p. 20

6.1 Setting

• The test uses a smooth manufactured nonlinear component with s = 2, prescribes U analytically, and leaves the linear component Ψ uncomputed.The cubic interface and dielectric ratio mean σ = 2 cannot be expected; the stated L2-rate comparison is consistent with t = 1.Source: §6 · p. 21, §6.1 · p. 21

6.2 Fine-mesh finite element solution

• For the fine-mesh finite-element solution Φh of (41) with U*=IhU, the energy-norm error is O(h) and L2 error O(h^2); at h=1/32, relative errors are about 8.7% and 0.36%.The L4 error also converges as O(h^2), faster than the h^(5/4) rate guaranteed by (55); the passage says this rate is not sharp for smooth solutions.Source: §6.2 · p. 22

6.3 Two-grid solution with H ∼h2/5

• For h ≤ 1/16, the two-grid and fine-mesh energy errors agree to three significant digits, while the two-grid correction error is two to three orders of magnitude below the discretization error.At h = 1/32, the two-grid solve takes 2.6 s versus 8.2 s for the full fine-mesh Newton iteration; the gain is moderate because that iteration needs only three steps in this test.Source: §6.3 · p. 22, Table 2 · p. 22

6.4 Dependence on the coarse mesh for fixed h

• With h = 1/32 fixed, the two-grid correction error has observed order about 3.7 as H varies, while the coarse L4 error converges approximately as O(H^2).The correction-to-coarse-error ratio stays between 0.27 and 0.52 in the asymptotic range, and the observed correction order exceeds the guaranteed H^(5/2) bound.Source: §6.4 · p. 23
• At H = 1/4, 343 interior coarse degrees of freedom yield a two-grid energy error only about 0.01% above the fine-mesh solution's error.The fine mesh has 250 047 interior degrees of freedom.Source: §6.4 · p. 23

6.5 Dependence on the truncation level

• For τ ≥ 3, the coarse-mesh solutions are bitwise identical to the τ = 85 solution, whereas truncation changes the computed solution when τ falls below the 2.508 threshold.At τ = 2.5 the maximum-norm difference is 3 × 10^-9; at τ = 2 and 1.5 it is 6.4 × 10^-4 and 3.5 × 10^-3, respectively.Source: §6.5 · p. 24

7 Concluding remarks

• A C2 truncation growing quadratically above τ preserves the original problem's solution for τ ≥ 2Zmax∥G + Ψ∥L∞(Ds), yielding error analysis without a discrete maximum principle or mesh-angle condition.The authors also state that this provides a new well-posedness proof and shows the practical overflow safeguard does not change the solution.Source: §7 · p. 24
• The two-grid method solves the nonlinear problem for bΦ on a coarse mesh and applies one Newton correction on the fine mesh, requiring only two fine-mesh symmetric positive definite linear systems.The systems are for Ψh and the Newton correction.Source: §7 · p. 24
• Under piecewise H^s primal regularity and H^σ dual regularity, the two-grid error is bounded by C(h^(s−1) + H^(2(s−1)+t/2)), with t = min{s, σ} − 1; H ∼h^(2/5) suffices when σ ≥ s.The estimates account for approximating U by its finite-element approximation, allow coarse-mesh data, and hold for any number of ionic species.Source: §7 · p. 24
• A three-dimensional manufactured-solution test agrees with the stated error estimates.Source: §7 · p. 24
• The authors identify higher-order elements, approximate curved interfaces, other regularized PBE forms, size-modified PBE, and Poisson–Nernst–Planck equations as possible extensions.Protein experiments using the cited PBE finite-element package are planned future work.Source: §7 · p. 24