Modified DeTurck Trick in Geometric Flows

• Modified DeTurck trick is a family of gauge-fixing and reparametrization strategies that restore strong parabolicity in geometric flows while preserving the underlying evolution.
• It improves numerical stability and mesh redistribution in applications like curve shortening, mean curvature, and moving-boundary problems through controlled tangential motion.
• The approach extends across Ricci flows, rough metrics, and G₂-structures, unifying various techniques to overcome degenerate behavior induced by diffeomorphism invariance.

The modified DeTurck trick is a family of gauge-fixing and reparametrization procedures derived from the classical DeTurck idea, in which a geometric evolution is altered by a carefully chosen tangential, Lie-derivative, or harmonic-map-driven term so that the modified system becomes more tractable analytically or numerically while preserving the underlying geometric evolution. In the literature represented here, the term encompasses several distinct but structurally related constructions: reparametrizations of curve shortening flow, mean curvature flow, and moving-boundary problems by harmonic map heat flow; Ricci–DeTurck formulations adapted to rough, noncompact, Euclidean, or hyperbolic backgrounds; DeTurck-type uniqueness arguments using harmonicity of the identity map; and modified gauge choices for heat-type flows of $G_2$-structures (Elliott et al., 2016).

1. Classical DeTurck framework and the meaning of “modified”

The classical DeTurck trick addresses the failure of strong parabolicity caused by diffeomorphism invariance. For Ricci flow,

$\partial_t g=-2\,Rc_g,$

the invariance $Rc_{F^*g}=F^*(Rc_g)$ implies that the linearized principal symbol has a kernel generated by infinitesimal diffeomorphisms. In the presentation surveyed in the BRIDGES lectures, one introduces a reference metric $g_0$ and the DeTurck vector field

$\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$

then studies the Ricci–DeTurck flow

$\partial_t g = -2Rc_g + L_{\W(g)}g.$

Its linearization has principal symbol equal to the identity on symmetric tensors, so it is strongly parabolic; Ricci flow is then recovered by pulling back along diffeomorphisms solving

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$

The same lectures emphasize that the auxiliary diffeomorphism flow can be identified with a harmonic-map heat flow, $\partial_t F_t=\Delta_{g(t),g_0}F_t$, which already indicates the close relation between DeTurck gauge-fixing and harmonic-map-based reparametrization (Karigiannis, 15 Aug 2025).

In later work, “modified” refers not to a single canonical alteration but to several setting-dependent changes of this basic mechanism. In extrinsic geometric flows, the modification is a reparametrization by harmonic map heat flow with an inverse diffusion parameter $\alpha>0$; in moving-boundary problems it is the use of harmonic map heat flow on manifolds with boundary together with mixed boundary conditions; in $G_2$-flows it is a DeTurck vector field augmented by a torsion correction term; in asymptotically hyperbolic geometry it is a normalized Ricci–DeTurck flow with an additional $\partial_t g=-2\,Rc_g,$0 term; and in rough-metric theory it is the deployment of Ricci–DeTurck gauge relative to a fixed smooth background metric $\partial_t g=-2\,Rc_g,$1 together with quantitative a priori estimates suited to low regularity (Elliott et al., 2016).

A common misconception is that the DeTurck correction changes the geometric motion itself. The cited papers repeatedly distinguish the geometric part from the gauge part: the added term changes parametrization, gauge, or coordinate representation, but is introduced precisely so that the evolving geometric object as a set, or the underlying Ricci flow modulo diffeomorphism, is preserved (Duan, 5 Jul 2026).

2. Harmonic-map reparametrization for curves, surfaces, and moving boundaries

For extrinsic flows, the modified DeTurck trick takes the form of a reparametrization driven by harmonic map heat flow. In the 2016 work on approximations of curve shortening flow and mean curvature flow, the authors introduce a time-dependent diffeomorphism $\partial_t g=-2\,Rc_g,$2 satisfying

$\partial_t g=-2\,Rc_g,$3

where $\partial_t g=-2\,Rc_g,$4 is an inverse diffusion constant. The resulting family of reparametrized equations adds tangential motion while keeping the normal geometric evolution unchanged. For curve shortening flow, the reparametrized family is

$\partial_t g=-2\,Rc_g,$5

and for mean curvature flow one obtains

$\partial_t g=-2\,Rc_g,$6

which is strongly parabolic for $\partial_t g=-2\,Rc_g,$7 (Elliott et al., 2016).

The moving-boundary formulation in "On algorithms with good mesh properties for problems with moving boundaries based on the Harmonic Map Heat Flow and the DeTurck trick" specializes this principle to compact submanifolds with boundary,

$\partial_t g=-2\,Rc_g,$8

with reference manifold $\partial_t g=-2\,Rc_g,$9 and induced metric $Rc_{F^*g}=F^*(Rc_g)$0. The original motion law is $Rc_{F^*g}=F^*(Rc_g)$1. To avoid mesh degeneration, the embedding is reparametrized by

$Rc_{F^*g}=F^*(Rc_g)$2

where $Rc_{F^*g}=F^*(Rc_g)$3 solves the harmonic map heat flow on manifolds with boundary,

$Rc_{F^*g}=F^*(Rc_g)$4

subject to

$Rc_{F^*g}=F^*(Rc_g)$5

This is the “modified” step relative to the classical DeTurck trick: the harmonic map heat flow is used on manifolds with boundary and with mixed boundary conditions designed to preserve boundary geometry while allowing tangential redistribution (Elliott et al., 2016).

A central proposition in that work states that if $Rc_{F^*g}=F^*(Rc_g)$6 solves this harmonic map heat flow, then

$Rc_{F^*g}=F^*(Rc_g)$7

where

$Rc_{F^*g}=F^*(Rc_g)$8

or equivalently $Rc_{F^*g}=F^*(Rc_g)$9. The harmonic map heat flow is therefore eliminated in favor of an explicit reparametrized velocity field. When the inverse map $g_0$0 is used, the transformed motion law becomes

$g_0$1

which is the DeTurck-type transformed motion law used in the numerics (Elliott et al., 2016).

The role of the boundary conditions is especially explicit. Pure Dirichlet conditions would freeze the boundary parametrization and prevent redistribution along $g_0$2; pure Neumann conditions would not keep the boundary mapped into itself. The chosen mixed conditions imply

$g_0$3

and a key lemma shows that the induced correction term is tangential on the boundary. Boundary vertices therefore slide along $g_0$4 without moving normal to it, so the geometric boundary shape is not altered (Elliott et al., 2016).

3. Mesh redistribution, weak formulations, and discrete algorithms

The numerical motivation for the modified DeTurck trick is that directly moving mesh vertices with the physical velocity $g_0$5 usually degenerates the mesh. In the moving-boundary setting, mesh quality is measured by

$g_0$6

where $g_0$7 is simplex diameter and $g_0$8 is the inradius. Small $g_0$9 means well-shaped simplices. The DeTurck reparametrization redistributes points so that the computational mesh becomes closer to a harmonic parametrization of the reference manifold. The stated effects are that interior mesh quality improves because parametrization distortions are regularized, and boundary mesh quality improves because the mixed boundary conditions induce tangential redistribution along $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$0, avoiding clustering or stretching at the boundary (Elliott et al., 2016).

This mechanism is encoded in weak formulations. For the moving-boundary problem, the weak form for the identity map $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$1 and the auxiliary field $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$2 is

$\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$3

together with

$\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$4

A global operator $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$5 is then introduced so that

$\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$6

allowing a tangential-gradient formulation convenient for finite elements (Elliott et al., 2016).

The corresponding discrete DeTurck algorithm on a simplicial mesh $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$7 and moving discrete submanifold $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$8 has two principal steps at each time level. First, solve for $\W(g)^k = g^{ij}\big((\Gamma^g)^k_{ij}-(\Gamma^{g_0})^k_{ij}\big),$9 from

$\partial_t g = -2Rc_g + L_{\W(g)}g.$0

Second, update the mesh by the reparametrized velocity, obtaining $\partial_t g = -2Rc_g + L_{\W(g)}g.$1 from a linear system that discretizes

$\partial_t g = -2Rc_g + L_{\W(g)}g.$2

then set

$\partial_t g = -2Rc_g + L_{\W(g)}g.$3

A notable computational advantage is that $\partial_t g = -2Rc_g + L_{\W(g)}g.$4 need not be explicitly computed, since $\partial_t g = -2Rc_g + L_{\W(g)}g.$5 is represented by fixed vertex data. At boundary vertices, the discrete correction is projected onto the tangent space of the discrete boundary using $\partial_t g = -2Rc_g + L_{\W(g)}g.$6, mirroring the continuous fact that the DeTurck correction on $\partial_t g = -2Rc_g + L_{\W(g)}g.$7 is tangential only (Elliott et al., 2016).

In the earlier 2016 extrinsic-flow paper, the same philosophy appears in families of semidiscrete and fully discrete schemes. For the curve shortening flow, the weak form is

$\partial_t g = -2Rc_g + L_{\W(g)}g.$8

and the authors obtain an $\partial_t g = -2Rc_g + L_{\W(g)}g.$9-type estimate for the semi-discrete scheme,

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$0

That estimate is proved only for $\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$1, and the case $\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$2 remains open analytically (Elliott et al., 2016).

The same paper relates its $\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$3-family to earlier algorithms. For curve shortening flow, $\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$4 corresponds to the Deckelnick–Dziuk form, while the formal limit $\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$5 is connected to the Barrett–Garcke–Nürnberg scheme. This suggests that the modified DeTurck trick supplies a unifying PDE-level interpretation of previously distinct mesh-redistribution strategies (Elliott et al., 2016).

4. Mean curvature flow and curve shortening flow: strict parabolicity, redistribution, and error analysis

The modified DeTurck trick has been developed in detail for parametric mean curvature flow and revisited for curve shortening flow. In the semidiscrete finite element analysis of mean curvature flow proposed by Elliott and Fritz, the geometric motion

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$6

is written parametrically as

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$7

Because this formulation is degenerate parabolic, the flow is reparametrized by a diffeomorphism generated by

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$8

yielding the strictly parabolic system

$\partial_t F_t(p) = -\W(g(t))_{F_t(p)}, \qquad F_0=\mathrm{id}_M.$9

The first term is the geometric mean curvature flow, and the second is an artificial tangential diffusion coming from the DeTurck reparametrization. It does not change the evolving surface as a set, but changes the parametrization, regularizes the PDE, and improves mesh point distribution (Deckelnick et al., 20 May 2026).

That work also rewrites the reparametrized PDE using a global metric tensor

$\partial_t F_t=\Delta_{g(t),g_0}F_t$0

and derives a weak formulation

$\partial_t F_t=\Delta_{g(t),g_0}F_t$1

where

$\partial_t F_t=\Delta_{g(t),g_0}F_t$2

Since

$\partial_t F_t=\Delta_{g(t),g_0}F_t$3

coercivity becomes a central stability mechanism. For finite element spaces $\partial_t F_t=\Delta_{g(t),g_0}F_t$4 of order $\partial_t F_t=\Delta_{g(t),g_0}F_t$5, the paper proves the optimal $\partial_t F_t=\Delta_{g(t),g_0}F_t$6-error estimate

$\partial_t F_t=\Delta_{g(t),g_0}F_t$7

equivalently

$\partial_t F_t=\Delta_{g(t),g_0}F_t$8

The restriction $\partial_t F_t=\Delta_{g(t),g_0}F_t$9 is stated to be technical rather than geometric (Deckelnick et al., 20 May 2026).

The same paper’s numerical experiments on a shrinking sphere and a dumbbell-shaped surface confirm both convergence and mesh-quality improvements. For the dumbbell surface, smaller $\alpha>0$0 gives better distributed meshes, and mesh quality is quantified by

$\alpha>0$1

with smaller values indicating better-shaped triangles. The experiments show that $\alpha>0$2 improves as $\alpha>0$3 decreases (Deckelnick et al., 20 May 2026).

For curve shortening flow, a 2026 revisit emphasizes the extrinsic derivation of the curve shortening–DeTurck flow and establishes optimal $\alpha>0$4 error estimates. Starting from

$\alpha>0$5

the DeTurck-coupled system leads, for the choice $\alpha>0$6, to

$\alpha>0$7

equivalently

$\alpha>0$8

The factor

$\alpha>0$9

changes the speed in the tangential direction relative to the normal direction, introducing controlled tangential redistribution while leaving the geometric evolution of the curve unchanged. The tangential quantity is explicitly

$G_2$0

which the paper identifies as the generator of tangential motion and mesh regularization (Duan, 5 Jul 2026).

Analytically, that paper proves dissipation of both perimeter and harmonic energy: $G_2$1 For fully discrete Euler and Crank–Nicolson finite element schemes of degree $G_2$2, it proves

$G_2$3

for Euler, and

$G_2$4

for Crank–Nicolson. The paper states that the optimal $G_2$5 estimates had remained open because of technical difficulties in handling the tangential terms and the nonlinear coefficient $G_2$6, and that the crucial ingredient is the superconvergence estimate

$G_2$7

with $G_2$8 for Euler and $G_2$9 for Crank–Nicolson (Duan, 5 Jul 2026).

5. Ricci–DeTurck modifications for rough metrics, scalar curvature, and mass

In Ricci-flow-based analysis, the modified DeTurck trick usually means the standard Ricci–DeTurck gauge deployed in a setting where the initial metric is rough, continuous, Euclidean-near, or asymptotically hyperbolic, with the background metric used as a fixed geometric reference. In "Ricci-Deturck flow from rough metrics and applications", the metric $\partial_t g=-2\,Rc_g,$00 evolves by the Ricci–DeTurck $\partial_t g=-2\,Rc_g,$01-flow

$\partial_t g=-2\,Rc_g,$02

and in local coordinates becomes a strictly parabolic quasilinear system with principal part the rough Laplacian with respect to $\partial_t g=-2\,Rc_g,$03. The initial metric is only assumed to be bi-Lipschitz to $\partial_t g=-2\,Rc_g,$04,

$\partial_t g=-2\,Rc_g,$05

and to satisfy small local scaling-invariant gradient concentration

$\partial_t g=-2\,Rc_g,$06

The main short-time existence theorem gives a smooth solution on $\partial_t g=-2\,Rc_g,$07, with $\partial_t g=-2\,Rc_g,$08, satisfying smoothing estimates

$\partial_t g=-2\,Rc_g,$09

and

$\partial_t g=-2\,Rc_g,$10

together with $\partial_t g=-2\,Rc_g,$11 convergence to the initial metric as $\partial_t g=-2\,Rc_g,$12 (Chu et al., 2022).

A related modification based on Morrey-type control appears in "Ricci-DeTurck Flow from Initial Metric with Morrey-type Integrability Condition". There the background $\partial_t g=-2\,Rc_g,$13 is smooth complete with uniformly bounded curvature and all covariant derivatives, and the DeTurck vector field is

$\partial_t g=-2\,Rc_g,$14

The Ricci–DeTurck $\partial_t g=-2\,Rc_g,$15-flow

$\partial_t g=-2\,Rc_g,$16

is used for initial metrics $\partial_t g=-2\,Rc_g,$17, smooth away from a compact singular set $\partial_t g=-2\,Rc_g,$18, globally bi-Lipschitz to $\partial_t g=-2\,Rc_g,$19, and satisfying the Morrey-type condition

$\partial_t g=-2\,Rc_g,$20

The short-time existence theorem gives a unique smooth solution on $\partial_t g=-2\,Rc_g,$21 with derivative bounds

$\partial_t g=-2\,Rc_g,$22

and convergence to $\partial_t g=-2\,Rc_g,$23 in $\partial_t g=-2\,Rc_g,$24 and in $\partial_t g=-2\,Rc_g,$25 as $\partial_t g=-2\,Rc_g,$26. That flow is then used to promote a distributional scalar curvature lower bound to a classical lower bound $\partial_t g=-2\,Rc_g,$27 for every $\partial_t g=-2\,Rc_g,$28 (Lee et al., 2024).

Euclidean-background formulations make the perturbative character of Ricci–DeTurck especially explicit. In the study of perturbations of Euclidean space,

$\partial_t g=-2\,Rc_g,$29

the flow is written as

$\partial_t g=-2\,Rc_g,$30

with background metric $\partial_t g=-2\,Rc_g,$31. Writing $\partial_t g=-2\,Rc_g,$32, the equation takes the schematic form

$\partial_t g=-2\,Rc_g,$33

or

$\partial_t g=-2\,Rc_g,$34

Near $\partial_t g=-2\,Rc_g,$35, the linearization is the heat equation, which underlies the long-time $\partial_t g=-2\,Rc_g,$36 and decay estimates. The scalar curvature obeys

$\partial_t g=-2\,Rc_g,$37

and the paper uses the resulting decay and regularity theory to derive a rigidity statement for non-negative scalar curvature perturbations of Euclidean space (Appleton, 2016).

The same gauge-fixed flow also underlies recent mass constructions for continuous metrics. In the asymptotically flat $\partial_t g=-2\,Rc_g,$38 setting, the Ricci–DeTurck flow

$\partial_t g=-2\,Rc_g,$39

is coupled to a backward radial heat-type equation for the cutoff,

$\partial_t g=-2\,Rc_g,$40

so that a $\partial_t g=-2\,Rc_g,$41 local mass has controlled distortion in time. The distortion estimate bounds

$\partial_t g=-2\,Rc_g,$42

by a quadratic error in the $\partial_t g=-2\,Rc_g,$43-size of the perturbation and an exponentially small cutoff error; this leads to existence and coordinate independence of the $\partial_t g=-2\,Rc_g,$44 mass at infinity under weak nonnegative scalar curvature assumptions (Burkhardt-Guim, 2022).

In the asymptotically hyperbolic case, the modification is explicit normalization. The normalized Ricci–DeTurck flow is

$\partial_t g=-2\,Rc_g,$45

with hyperbolic background $\partial_t g=-2\,Rc_g,$46. The extra term $\partial_t g=-2\,Rc_g,$47 is added because hyperbolic space expands under the unnormalized Ricci flow. Writing $\partial_t g=-2\,Rc_g,$48, the PDE becomes

$\partial_t g=-2\,Rc_g,$49

which is the normalized $\partial_t g=-2\,Rc_g,$50-flow formulation used to define a mass function for $\partial_t g=-2\,Rc_g,$51-asymptotically hyperbolic manifolds and a $\partial_t g=-2\,Rc_g,$52-weak scalar curvature lower bound for continuous metrics (Li, 21 Dec 2025).

6. Extensions beyond classical flows: metric invariants, $\partial_t g=-2\,Rc_g,$53-structures, and uniqueness arguments

The modified DeTurck trick extends beyond standard Ricci flow in at least three directions represented here. First, for more general second-order metric flows, the BRIDGES lectures discuss flows of the form

$\partial_t g=-2\,Rc_g,$54

For the Ricci–Bourguignon flow

$\partial_t g=-2\,Rc_g,$55

the same DeTurck vector field is used to form

$\partial_t g=-2\,Rc_g,$56

The symbol analysis yields the strong parabolicity criterion

$\partial_t g=-2\,Rc_g,$57

The lectures also note that a previously claimed stronger condition in the literature is not sufficient, because the relevant symbol operator is not self-adjoint (Karigiannis, 15 Aug 2025).

Second, for heat-type flows of $\partial_t g=-2\,Rc_g,$58-structures, the modification involves both metric and torsion data. On a compact oriented $\partial_t g=-2\,Rc_g,$59-manifold with $\partial_t g=-2\,Rc_g,$60-structure $\partial_t g=-2\,Rc_g,$61, one has the decomposition

$\partial_t g=-2\,Rc_g,$62

and the torsion tensor decomposes as

$\partial_t g=-2\,Rc_g,$63

A classification theorem states that a basis for the independent second-order differential invariants of a $\partial_t g=-2\,Rc_g,$64-structure $\partial_t g=-2\,Rc_g,$65 are the symmetric $\partial_t g=-2\,Rc_g,$66-tensors $\partial_t g=-2\,Rc_g,$67 and the vector fields $\partial_t g=-2\,Rc_g,$68, where

$\partial_t g=-2\,Rc_g,$69

Accordingly, second-order quasilinear $\partial_t g=-2\,Rc_g,$70-flows can be written schematically as

$\partial_t g=-2\,Rc_g,$71

The sufficient condition for short-time existence and uniqueness is that if

$\partial_t g=-2\,Rc_g,$72

then the flow

$\partial_t g=-2\,Rc_g,$73

has short-time existence and uniqueness by a slight modification of the DeTurck trick. The corresponding modified gauge vector field is

$\partial_t g=-2\,Rc_g,$74

The added torsion correction $\partial_t g=-2\,Rc_g,$75 is the specifically $\partial_t g=-2\,Rc_g,$76-geometric modification (Karigiannis, 15 Aug 2025).

Third, there is a DeTurck-type uniqueness method that does not produce a parabolic evolution but instead compares connections by harmonicity of the identity map. In "On DeTurck uniqueness theorems for Ricci tensor", the identity map

$\partial_t g=-2\,Rc_g,$77

is harmonic precisely when the deformation tensor

$\partial_t g=-2\,Rc_g,$78

satisfies $\partial_t g=-2\,Rc_g,$79. The Weitzenböck formula

$\partial_t g=-2\,Rc_g,$80

for harmonic maps then leads, under nonnegative sectional curvature and the hypothesis $\partial_t g=-2\,Rc_g,$81, to rigidity of the Levi-Civita connection. In the compact case, if $\partial_t g=-2\,Rc_g,$82, then $\partial_t g=-2\,Rc_g,$83; if moreover $\partial_t g=-2\,Rc_g,$84 is irreducible, then $\partial_t g=-2\,Rc_g,$85 for some $\partial_t g=-2\,Rc_g,$86. This is described there as a modified DeTurck-type ingredient because the comparison is carried out through harmonicity of the identity map rather than direct metric comparison (Stepanov, 2015).

Across these settings, the recurrent structure is the same: one isolates the degeneracy caused by diffeomorphism invariance, tangential freedom, or torsion-sensitive gauge directions, then introduces a tailored correction that restores strong parabolicity, effective coercivity, or rigidity. A plausible implication is that “modified DeTurck trick” is best understood not as a single formula but as a design principle: preserve the geometric content, change the parametrization or gauge so that analysis or computation becomes feasible.