---
title: Direct Forcing in CFD
url: https://www.emergentmind.com/topics/direct-forcing
type: topic
---

# Direct Forcing in CFD

Searching arXiv for recent and foundational papers on “direct forcing” across CFD contexts.
In computational fluid dynamics, “direct forcing” denotes several related formulations in which forcing is introduced explicitly into the governing equations, or boundary constraints are enforced algebraically through an explicitly computed force. In incompressible Navier–Stokes simulations, the term can mean an added body-force term used to drive a statistically stationary flow, such as a streamwise pressure-gradient surrogate in turbulent channel flow [1509.04877]. In rotating turbulence and magnetohydrodynamics, it can denote prescribed large-scale mechanical or spectral forcing whose temporal coherence, geometry, or field channel controls helicity, anisotropy, and cascade structure [1601.04310], [2112.06286]. In resolvent analysis, the nonlinear term itself is treated as the “direct” forcing of the linearized dynamics [2004.01565]. In immersed-boundary methods, direct forcing refers to computing the fluid–solid interaction force explicitly and algebraically so that the post-step fluid velocity at the immersed interface matches the desired boundary velocity, without artificial spring–damper dynamics [1809.08170].

## 1. Meanings and formal structure

A common formal template is the incompressible momentum equation with an added forcing term,
\[
\frac{\partial u_i}{\partial t} + u_j \frac{\partial u_i}{\partial x_j}
=
-\frac{\partial p}{\partial x_i}
+
\frac{1}{Re}\frac{\partial^2 u_i}{\partial x_j \partial x_j}
+
F_i,
\]
or its dimensional counterpart with a body-force density \(\mathbf{f}\) [1509.04877]. In channel-flow DNS, only the streamwise component is non-zero, \(F_1\equiv F\neq 0\), and the forcing plays the role of the streamwise pressure gradient averaged over the homogeneous directions [1509.04877]. In rotating turbulence, the forcing is an imposed mechanical body force \(\mathbf{f}(\mathbf{x},t)\) in the rotating-frame Navier–Stokes equations, built from large-scale sinusoidal modes whose phases are randomized with a prescribed memory time \(\tau_m\) [1601.04310]. In incompressible MHD, “direct injection on the velocity field” means low-wavenumber stochastic forcing applied to the velocity equation, while “stirring acts on the magnetic field only” means forcing applied to the magnetic induction equation [2112.06286].

A distinct but related usage appears in resolvent analysis. There, the fluctuation equations are written as a stable linear operator driven by the nonlinear term
\[
\mathbf{f}=-(\mathbf{u}\cdot\nabla)\mathbf{u},
\]
which is the nonlinear or direct forcing of the linearized dynamics about the mean flow [2004.01565]. In immersed-boundary methods, by contrast, direct forcing is an algebraic enforcement mechanism: the force at Lagrangian surface points is chosen from a one-step relation so that the fluid velocity after the time step equals the prescribed rigid-body or boundary velocity [1809.08170]. The literature therefore uses the same term for several mathematically adjacent operations: explicit body-force driving, explicit injection into selected equations or scales, nonlinear forcing of linearized models, and algebraic constraint enforcement.

## 2. Explicit body-force driving in DNS

In turbulent channel flow, direct forcing is the explicit addition of a streamwise body force to maintain a statistically stationary state. Three canonical prescriptions are compared: constant pressure gradient (CPG), constant flow rate (CFR), and constant power input (CPI) [1509.04877]. CPG fixes \(F=\text{const}\), so the bulk flow rate adjusts. CFR fixes the bulk velocity \(U_b\) and computes a time-dependent \(F(t)\) at each time step from the discrete momentum equation so that the prescribed mass flux is satisfied exactly. CPI fixes the pumping power per unit wetted area and updates \(F^{n+1}=P_p/U_b^n\), so that \(F(t)\) and \(U_b(t)\) both fluctuate while their product remains constant on average [1509.04877].

At \(Re_\tau\approx 200\), with identical numerics and nearly identical measured \(Re_\tau\), the mean velocity, RMS fluctuations, and one-dimensional energy spectra are indistinguishable within line thickness across CFR, CPG, and CPI [1509.04877]. The main forcing sensitivity appears only in the far tail of the PDF of local streamwise wall friction. For \(\sigma\gtrsim 12\), CFR has lower probability for extreme positive wall-friction events; for \(\sigma>12\), the probability in CFR is roughly \(2\)–\(3\) times lower than in CPG and CPI, with absolute probabilities \(P\sim 10^{-8}\)–\(10^{-9}\) [1509.04877]. No analogous difference is seen in the negative tail or in the spanwise wall-shear PDF. This establishes that the forcing term is not entirely uninfluential, even though its detectable effect is confined to very rare events.

In homogeneous isotropic turbulence with immersed finite-size particles and gravity, direct forcing enters through the Eswaran–Pope random forcing scheme acting on low-wavenumber modes, while particle surfaces are enforced by immersed-boundary forcing [1511.02638]. The forcing is defined in Fourier space on a band \(0<|\boldsymbol{\kappa}|\le\kappa_f\), with each forced mode advanced by an Ornstein–Uhlenbeck process and projected onto the divergence-free subspace [1511.02638]. Because only a limited number of Fourier modes are forced, the forcing field can be evaluated directly in physical space, thereby avoiding full-size transforms [1511.02638]. In vertically elongated boxes, the same forcing parameters maintain the background turbulence while improving decorrelation of Lagrangian signals in the direction of gravity; in the finite-gravity case, the modified Reynolds number \(Re_{\tilde\lambda}\) of the background turbulence remains essentially equal to the single-phase value even though the total dissipation is much larger because of particle–gravity coupling [1511.02638].

## 3. Forcing-dependent dynamics and multiscale transfer

In rotating turbulence, direct mechanical forcing is not a neutral energy source but a control parameter for helicity generation, anisotropy, and cascade direction [1601.04310]. The forcing is imposed in physical space as a superposition of large-scale sinusoidal modes at \(k_f=2,3,4\), with phases \(\phi_x,\phi_y,\phi_z\) randomized after a memory time \(\tau_m\) [1601.04310]. Although the forcing is non-helical on average, \(\langle \mathbf{f}\cdot(\nabla\times\mathbf{f})\rangle=0\), a time-independent forcing with a persistent orientation relative to the rotation axis can generate substantial net helicity once rotation is strong enough. At fixed \(Re_f=333\), time-independent forcing exhibits a jump from \(\rho_H\approx 0\) to \(|\rho_H|\gtrsim 0.3\) at a critical Rossby number \(Ro_f^{\mathrm{crit}}\simeq 0.2\), whereas random-in-time forcing keeps mean \(\rho_H\approx 0\) for all \(Ro_f\) considered [1601.04310]. The same forcing memory controls whether the spectrum is closer to \(k^{-5/2}\), \(k^{-2}\), or \(k^{-5/3}\), and whether inverse transfer and quasi-2Dization dominate or a stronger forward cascade persists.

In incompressible MHD, direct forcing is parameterized by \(a\in[0,1]\), interpolating between direct injection on the velocity field (\(a=1\)) and forcing on the magnetic field only (\(a=0\)) while keeping the total mean injection rate fixed [2112.06286]. The total energy flux is decomposed into four channels: kinetic advection \(\Pi_I^\ell\), Lorentz channel \(\Pi_M^\ell\), magnetic advection \(\Pi_A^\ell\), and magnetic stretching \(\Pi_D^\ell\), together with homochiral and heterochiral subchannels [2112.06286]. The central result is a quasi-singular role of the purely mechanical case: as soon as \(a<1\), the mean kinetic advective flux \(\Pi_I^\ell\) becomes almost vanishing at all scales, not because the nonlinearity is weak but because heterochiral forward and homochiral inverse transfers nearly cancel in a flux-loop balance [2112.06286]. By contrast, the Lorentz, magnetic advection, and magnetic stretching channels remain forward in both homochiral and heterochiral sectors. The resolved kinetic–magnetic conversion reverses around \(a\approx 0.4\): for \(a\lesssim 0.4\), net conversion is magnetic \(\rightarrow\) kinetic, whereas for \(a\gtrsim 0.4\), it is kinetic \(\rightarrow\) magnetic [2112.06286]. This suggests that forcing composition can qualitatively reorganize the multiscale dynamics even when total injected power is fixed.

## 4. Direct forcing in resolvent analysis

In mean-flow resolvent analysis of turbulent channel flow, direct forcing is the nonlinear term treated as an input to the linearized Orr–Sommerfeld/Squire system [2004.01565]. For each \((\alpha,\beta,\omega)\), the Fourier-transformed fluctuation velocity satisfies
\[
\tilde{\mathbf{u}}(\omega)=\mathbf{R}(\omega)\,\tilde{\mathbf{f}}(\omega),
\]
with \(\mathbf{f}=-(\mathbf{u}\cdot\nabla)\mathbf{u}\) obtained directly from DNS, not modeled in the primary analysis [2004.01565]. The relevant statistical object is the forcing cross-spectral density \(\mathbf{P}(\omega)\), linked to the response CSD by
\[
\mathbf{S}(\omega)=\mathbf{R}(\omega)\,\mathbf{P}(\omega)\,\mathbf{R}(\omega)^H.
\]

The forcing is shown not to be uncorrelated, or white, in space. Its CSD has strong wall-normal structure, significant cross-component correlations, and non-negligible projections onto sub-optimal resolvent forcing modes [2004.01565]. Since the nonlinear forcing is non-solenoidal by construction and incompressible velocity responds only to the solenoidal part, the solenoidal forcing is evaluated through the projection operator \(\mathbf{L}\), giving \(\mathbf{P}^{(s)}=\mathbf{L}\mathbf{S}\mathbf{L}^H\) [2004.01565]. The solenoidal part is the combination of oblique streamwise vortices and a streamwise component which counteract each other, as in a destructive interference. A rank-2 approximation of the forcing, retaining only the two most energetic SPOD modes, reproduces the bulk of the response [2004.01565]. At the same time, the projections \(b_i\) of the DNS forcing onto the right-singular vectors remain substantial for higher-index modes, so the actual modal weights \(a_i=\sigma_i b_i\) are not determined by singular values alone. An eddy-viscosity-modified resolvent improves the prediction because the implied effective forcing color is closer to that measured from DNS [2004.01565].

## 5. Direct-forcing immersed boundary methods

In immersed-boundary methods for rigid particles and rigid surfaces, direct forcing means that the force enforcing no-slip is computed explicitly and algebraically from the discretized momentum equation, without artificial stiffness or damping parameters [1809.08170]. The fluid is solved on a fixed Cartesian Eulerian grid, each rigid surface is represented by Lagrangian force points, velocity is interpolated from Eulerian to Lagrangian locations through a regularized delta function, and the force is then spread back to the Eulerian grid [1809.08170]. In Uhlmann’s formulation, the core update at a force point is
\[
\mathbf{F}(\mathbf{X}_l^{(m)})=
\frac{\mathbf{U}^{(d)}(\mathbf{X}_l^{(m)})-\tilde{\mathbf{U}}(\mathbf{X}_l^{(m)})}{\Delta t},
\]
where \(\mathbf{U}^{(d)}\) is the rigid-body velocity and \(\tilde{\mathbf{U}}\) is the interpolated unforced fluid velocity [1809.08170]. This provides second-order accurate no-slip enforcement in the context of the underlying scheme, with no additional time-step restriction beyond the base fluid solver.

The same algebraic viewpoint extends to more strongly coupled and more structured settings. In the semi-implicit direct forcing method for incompressible viscous thermal flows, unknown Lagrangian force densities \(F^k(X^k)\) and power densities \(Q^k(X^k)\) are introduced as Lagrange multipliers and solved through a Schur complement system built from the interpolation and regularization operators, \(S=IH^{-1}R\), so that the interpolated velocity and temperature satisfy the boundary constraints on immersed surfaces [1710.09643]. This formulation accurately meets the thermal and no-slip boundary conditions on the surfaces of immersed bodies for \(10^3\le Ra\le 10^6\) while remaining compatible with a segregated pressure–velocity solver [1710.09643].

Other immersed-boundary variants modify how the direct forcing is transferred between Eulerian and Lagrangian representations. A one-sided direct forcing immersed boundary method constructs kernel functions via moving least squares so that coupling occurs only on one side of the interface, reducing spurious feedback forcing and internal flows associated with isotropic kernels that act on both sides [2104.07738]. Because one-sided MLS kernels can become non-monotone with large over- and undershoots, the method introduces constant-vector-shift and non-constant-vector-shift strategies to construct generating functions that are positive and monotone; the latter yields better boundary layers and improved stability in direct-forcing simulations [2104.07738]. In fluid–rigid-body interaction, the direct-forcing immersed boundary method has also been embedded in an implicit partitioned algorithm that couples Navier–Stokes and Newton–Euler equations through interpolation, direct forcing, spreading, and a fixed relaxation on rigid-body kinematics, thereby handling strongly coupled interface conditions and low density-ratio regimes within a PISO-based solver [2604.24439].

## 6. Accuracy, stability, and emerging extensions

A major line of development addresses the fact that conventional direct forcing does not, in general, satisfy the discrete partition-of-unity condition exactly. The boundary thickening-based direct forcing method starts from the implicit direct forcing relation and replaces the standard solid forcing-shell thickness \(d_{rs}=h\) by a slightly larger optimal thickness \(d_{rs}^{opt}\), chosen so that the local communication between Lagrangian boundary points and neighboring Eulerian nodes better satisfies the partition-of-unity condition [1806.09403]. With this modification, the explicit direct-forcing formula is retained, but numerical accuracy becomes comparable with multi-direct forcing (MDF), implicit velocity correction (IVC), and the reproducing kernel particle method (RKPM), while the computation cost remains much lower and nearly equivalent to the conventional DF scheme [1806.09403]. For the tested kernels, the optimum thickness values are approximately \(1.4h\), \(1.9h\), and \(2.6h\) for the 2-point, 3-point, and 4-point delta functions, respectively [1806.09403].

A complementary analysis treats the accelerated multi-direct-forcing method as a Richardson iteration for the discrete coupling system \(\mathcal{A}\mathcal{G}=\mathcal{U}\) [2507.04986]. The relaxed update
\[
\mathcal{G}^{\ell+1}=(\mathcal{I}-\omega\mathcal{A})\mathcal{G}^{\ell}+\omega\mathcal{U}
\]
introduces an acceleration parameter \(\omega\) and an effective forcing-amplification factor \(\eta\) [2507.04986]. For moving rigid bodies, numerical stability is controlled by a single lumped parameter,
\[
\eta A = \eta\,\omega\,\frac{\rho_f}{\rho_c}\,\frac{S\Delta x}{V},
\]
and stable simulations require \(\eta A \lesssim 1.0\) in the tested regimes [2507.04986]. The optimal acceleration parameter is nearly kernel-universal: \(\omega_{\mathrm{opt}}\approx C_s^{-1}\), giving \(8/3\) for Peskin’s 4-point kernel and \(2\) for the 3-point kernel, largely independent of boundary discretization, boundary shape, and spatial dimensionality in the reported tests [2507.04986]. This clarifies the tradeoff between improved no-slip accuracy and added-mass-type instability in explicit moving-boundary direct forcing.

Direct forcing is also expanding beyond momentum-only problems. In multicomponent miscible-mixture LBM, the proposed forcing mass-transfer approach is not a “direct forcing” scheme in the usual sense of simply adding a source term to the discrete Boltzmann equation; instead, it is an explicit velocity-difference forcing formulation in which each species has a forcing term \(S_\alpha^i\) satisfying \(\sum_\alpha S_\alpha^i=0\) and \(\sum_\alpha \mathbf{e}_\alpha^i S_\alpha^i=\mathbf{F}_i\), so that mixture momentum recovers Navier–Stokes with the total body force and species momentum balances recover a Maxwell–Stefan-type description under external fields [2504.07059]. The method recovers the macroscopic mass conservation equations, the Navier–Stokes equation with external forcing term, and the full Maxwell–Stefan equation for ideal mixtures at low Knudsen numbers, together with a boundary scheme for impermeable solid walls [2504.07059]. Validation on ultracentrifuge separation of uranium isotopes and the Loschmidt tube with gravitational effects indicates that direct forcing can be generalized from boundary enforcement and turbulence driving to species-level transport closures under gravity or centrifugal fields [2504.07059].

Taken together, these developments indicate that direct forcing is best understood not as a single algorithm but as a class of explicit forcing constructions whose role is to impose a target constraint—statistical stationarity, prescribed energy input, rigid-body no-slip, or species-momentum balance—through a force computed at the level of the discretized equations. The literature also makes clear that such forcing is not generically neutral: it can leave rare-event signatures in wall-bounded turbulence [1509.04877], generate or suppress helicity and inverse transfer in rotating turbulence [1601.04310], reorganize channel-wise fluxes in MHD [2112.06286], alter modal weights in resolvent models [2004.01565], or destabilize moving-body simulations unless the coupling strength is controlled [2507.04986].

Source: https://www.emergentmind.com/topics/direct-forcing