Tangential Subdifferentials

• Tangential subdifferentials are generalized first-order objects that decompose subdifferential information into tangential directions, which guide motion along feasible sets, and normal components, which capture deviations.
• Multiple formulations—including epigraph-based, directional, and projection methods—provide equivalent characterizations under regularity conditions in both smooth and convex settings.
• They play a crucial role in nonsmooth optimization and mathematical programming, enabling precise stationarity conditions and effective analysis in MPEC and variational frameworks.

Tangential subdifferentials are generalized first-order objects that encode nonsmooth behavior through tangent geometry. In recent variational-analysis literature, the term appears in several closely related senses: an epigraph-based subdifferential built from the contingent cone to $\operatorname{epi} f$, a directional limiting subdifferential that filters generalized gradients by the actual direction of approach, a support-function-based subdifferential for tangentially convex functions, and, under partial smoothness, the projection of the limiting subdifferential onto the tangent space of an active manifold (Pang, 2011, Benko et al., 2017, Qin et al., 26 Jan 2025, Mishra et al., 3 Sep 2025). The common theme is a decomposition of first-order information into tangential and normal parts, with the tangential part governing motion along feasible sets or active manifolds and the normal part encoding deviation away from them.

1. Foundational formulations

A basic tangential construction begins with the epigraph of an extended-real-valued function. For $f:X\to \mathbb{R}\cup\{+\infty\}$, with epigraphical mapping $E(x)=[f(x),+\infty)$, the tangent cone to $\operatorname{epi} f$ induces the contingent epigraphical directional derivative

$$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$

The corresponding tangential subdifferential is

$$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$

Equivalently, in terms of the graphical derivative of $E$,

$$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$

and $\partial_T f(\bar x)$ is the set of dual elements dominated by $\inf E'(h)$ in every direction (Pang, 2011).

In finite dimensions, the same paper shows that tangential and normal-cone approaches are equivalent under its generalized-derivative framework: graphical derivatives of the epigraph mapping and coderivatives of the epigraphical map yield equivalent characterizations of tangential subdifferentials. When $f:X\to \mathbb{R}\cup\{+\infty\}$0 is Clarke regular at $f:X\to \mathbb{R}\cup\{+\infty\}$1, the tangential subdifferential coincides with the Clarke subdifferential, and for convex $f:X\to \mathbb{R}\cup\{+\infty\}$2 it agrees with the usual convex subdifferential (Pang, 2011).

A second foundational formulation is used for tangentially convex functions. If $f:X\to \mathbb{R}\cup\{+\infty\}$3 is directionally differentiable at $f:X\to \mathbb{R}\cup\{+\infty\}$4 and the map $f:X\to \mathbb{R}\cup\{+\infty\}$5 is convex, then $f:X\to \mathbb{R}\cup\{+\infty\}$6 is tangentially convex at $f:X\to \mathbb{R}\cup\{+\infty\}$7, and its tangential subdifferential is

$f:X\to \mathbb{R}\cup\{+\infty\}$8

This set is nonempty, compact, and convex, and the directional derivative is its support function: $f:X\to \mathbb{R}\cup\{+\infty\}$9 In the smooth case, $E(x)=[f(x),+\infty)$0; for convex $E(x)=[f(x),+\infty)$1 on an open domain, $E(x)=[f(x),+\infty)$2; and under Clarke regularity it coincides with the Clarke subdifferential (Mishra et al., 3 Sep 2025).

2. Directional and constrained tangential subdifferentials

A directional refinement replaces non-directional limiting objects by approach-direction-dependent ones. For a closed set $E(x)=[f(x),+\infty)$3, the directional limiting normal cone is

$E(x)=[f(x),+\infty)$4

It satisfies $E(x)=[f(x),+\infty)$5, $E(x)=[f(x),+\infty)$6, and $E(x)=[f(x),+\infty)$7 if $E(x)=[f(x),+\infty)$8 (Benko et al., 2017).

For functions, the directional limiting subdifferential is defined on epigraph directions: $E(x)=[f(x),+\infty)$9 The paper also uses an analytic directional subdifferential $\operatorname{epi} f$0 and proves that, under calmness of $\operatorname{epi} f$1 in direction $\operatorname{epi} f$2,

$\operatorname{epi} f$3

If $\operatorname{epi} f$4 is directionally differentiable at $\operatorname{epi} f$5 along $\operatorname{epi} f$6, then $\operatorname{epi} f$7 (Benko et al., 2017).

This framework yields a constrained or tangential subdifferential relative to a set $\operatorname{epi} f$8. Since

$\operatorname{epi} f$9

one may regard

$$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$0

as the tangential subdifferential along feasible direction $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$1. The directional sum rule gives

$$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$2

and, under calmness and weak directional qualification conditions, equality often holds (Benko et al., 2017).

The directional viewpoint sharpens classical subdifferential calculus because constraint contributions vanish in interior-pointing directions. For the half-space $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$3 with $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$4, one has

$$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$5

Thus, for strictly interior feasible directions, the tangential subdifferential of $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$6 reduces to the unconstrained directional subdifferential of $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$7 (Benko et al., 2017).

3. Partial smoothness and manifold-projected tangential components

Within partial smoothness, tangential subdifferentials arise from an active manifold. A function $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$8 is partly smooth at $$f^\circ(\bar x;h):=\inf\{r\in\mathbb{R}:(h,r)\in T_{\operatorname{epi} f}(\bar x,f(\bar x))\}.$$9 relative to a set $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$0 containing $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$1 if $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$2 is a $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$3-smooth manifold around $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$4 and four conditions hold: smoothness of $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$5, prox-regularity and regularity near $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$6 with nonempty subdifferentials, sharpness, and continuity of $$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$7. The sharpness condition is

$$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$8

equivalently

$$\partial_T f(\bar x):=\{v\in X^*:\langle v,h\rangle \le f^\circ(\bar x;h)\ \text{for all }h\in X\}.$$9

For set-valued operators $E$0, the operator version of partial smoothness replaces $E$1 by $E$2 and requires

$E$3

together with continuity of $E$4 along $E$5 (Qin et al., 26 Jan 2025).

The operator framework provides a canonical “smooth representative”

$E$6

which is single-valued and continuous along $E$7 near $E$8. For functions, with $E$9, this motivates the tangential subdifferential

$$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$0

Under partial smoothness, $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$1 is single-valued and equals the Riemannian gradient of the restriction: $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$2 The normal component is generated by the subdifferential span, yielding

$$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$3

with equality at $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$4 and locally along $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$5 under the sharpness and continuity assumptions (Qin et al., 26 Jan 2025).

This tangential-normal decomposition drives identifiability. The paper introduces the local union

$$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$6

where $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$7 is an $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$8-localization around $$f^\circ(\bar x;h)=\inf E'(h),\qquad E'(h):=DE(\bar x\mid f(\bar x))(h),$$9 and $\partial_T f(\bar x)$0. It proves that $\partial_T f(\bar x)$1 and

$\partial_T f(\bar x)$2

and if $\partial_T f(\bar x)$3 with $\partial_T f(\bar x)$4, then any sequence $\partial_T f(\bar x)$5 satisfying

$\partial_T f(\bar x)$6

eventually lies in $\partial_T f(\bar x)$7 (Qin et al., 26 Jan 2025). This removes the need for exact dual convergence and allows identification under non-vanishing errors and degeneracy.

The examples are explicit. For $\partial_T f(\bar x)$8 and support $\partial_T f(\bar x)$9, the active manifold is

$\inf E'(h)$0

the tangent subspace is the coordinate subspace on $\inf E'(h)$1, and

$\inf E'(h)$2

equals $\inf E'(h)$3 on $\inf E'(h)$4 and $\inf E'(h)$5 on $\inf E'(h)$6, which is exactly $\inf E'(h)$7. For $\inf E'(h)$8, the tangential component vanishes while the normal component is $\inf E'(h)$9 (Qin et al., 26 Jan 2025).

4. Tangential transversality and Clarke subdifferential calculus

Not all uses of “tangential” introduce a new subdifferential. In the Banach-space transversality framework of Bivas, Krastanov, and Ribarska, the central objects are uniform tangent sets and strong tangential transversality, and the subdifferentials that appear are Clarke’s $f:X\to \mathbb{R}\cup\{+\infty\}$00 and the singular $f:X\to \mathbb{R}\cup\{+\infty\}$01 (Bivas et al., 2018).

For a closed set $f:X\to \mathbb{R}\cup\{+\infty\}$02 and $f:X\to \mathbb{R}\cup\{+\infty\}$03, a bounded set $f:X\to \mathbb{R}\cup\{+\infty\}$04 is a uniform tangent set if, for each $f:X\to \mathbb{R}\cup\{+\infty\}$05, there exists $f:X\to \mathbb{R}\cup\{+\infty\}$06 such that for every $f:X\to \mathbb{R}\cup\{+\infty\}$07 and every $f:X\to \mathbb{R}\cup\{+\infty\}$08 one can find $f:X\to \mathbb{R}\cup\{+\infty\}$09 with

$f:X\to \mathbb{R}\cup\{+\infty\}$10

Such sets are contained in the Clarke tangent cone $f:X\to \mathbb{R}\cup\{+\infty\}$11, and under separability or hypertangent assumptions they generate that cone (Bivas et al., 2018).

For two closed sets $f:X\to \mathbb{R}\cup\{+\infty\}$12, strong tangential transversality at $f:X\to \mathbb{R}\cup\{+\infty\}$13 means that there exist uniform tangent sets $f:X\to \mathbb{R}\cup\{+\infty\}$14 and $f:X\to \mathbb{R}\cup\{+\infty\}$15 and some $f:X\to \mathbb{R}\cup\{+\infty\}$16 such that

$f:X\to \mathbb{R}\cup\{+\infty\}$17

This implies tangential transversality and transversality of the Clarke tangent cones. The payoff is the Clarke normal intersection property

$f:X\to \mathbb{R}\cup\{+\infty\}$18

with the sum on the right weak* closed (Bivas et al., 2018).

The same paper applies this to split epigraphs

$f:X\to \mathbb{R}\cup\{+\infty\}$19

and derives a Clarke sum rule: $f:X\to \mathbb{R}\cup\{+\infty\}$20 whenever $f:X\to \mathbb{R}\cup\{+\infty\}$21 and $f:X\to \mathbb{R}\cup\{+\infty\}$22 satisfy the required normal intersection property; strong tangential transversality is a sufficient condition. The paper also shows that cone transversality alone may fail to guarantee weak* closedness of $f:X\to \mathbb{R}\cup\{+\infty\}$23, as illustrated by Example 3.19 in $f:X\to \mathbb{R}\cup\{+\infty\}$24 (Bivas et al., 2018).

A recurrent misconception is therefore that “tangential subdifferential” must always denote a distinct subdifferential construction. The transversality paper shows a different usage: “tangential” can instead designate primal tangent-set hypotheses that enable Clarke normal-cone and Clarke subdifferential calculus.

5. Tangential subdifferentials in nonsmooth mathematical programming

The most explicit optimization-oriented use of the term appears in nonsmooth mathematical programs with equilibrium constraints. In the MPEC setting

$f:X\to \mathbb{R}\cup\{+\infty\}$25

the paper works with functions that are tangentially convex at the reference point and builds all constraint qualifications and stationarity concepts from tangential subdifferentials (Mishra et al., 3 Sep 2025).

The fundamental calculus is support-function based. If $f:X\to \mathbb{R}\cup\{+\infty\}$26 and $f:X\to \mathbb{R}\cup\{+\infty\}$27 are tangentially convex at $f:X\to \mathbb{R}\cup\{+\infty\}$28, then

$f:X\to \mathbb{R}\cup\{+\infty\}$29

for all $f:X\to \mathbb{R}\cup\{+\infty\}$30, and consequently

$f:X\to \mathbb{R}\cup\{+\infty\}$31

Positive scaling and affine perturbation behave analogously. The same paper also defines $f:X\to \mathbb{R}\cup\{+\infty\}$32-pseudoconvexity and $f:X\to \mathbb{R}\cup\{+\infty\}$33-quasiconvexity through sign conditions on $f:X\to \mathbb{R}\cup\{+\infty\}$34 for all $f:X\to \mathbb{R}\cup\{+\infty\}$35 (Mishra et al., 3 Sep 2025).

Constraint qualifications are expressed through dual cones built from unions of tangential subdifferentials of active constraint functions. With $f:X\to \mathbb{R}\cup\{+\infty\}$36 and $f:X\to \mathbb{R}\cup\{+\infty\}$37 defined in this way, the generalized standard Abadie CQ is

$f:X\to \mathbb{R}\cup\{+\infty\}$38

the MPEC Abadie CQ is

$f:X\to \mathbb{R}\cup\{+\infty\}$39

and the MPEC Zangwill CQ is

$f:X\to \mathbb{R}\cup\{+\infty\}$40

Because $f:X\to \mathbb{R}\cup\{+\infty\}$41 and $f:X\to \mathbb{R}\cup\{+\infty\}$42 is closed, MPEC Zangwill CQ implies MPEC-ACQ. The paper further proves that its MPEC weak reverse convex CQ implies the MPEC Zangwill CQ (Mishra et al., 3 Sep 2025).

Stationarity is likewise formulated entirely in terms of tangential subdifferentials. GA-stationarity requires multipliers $f:X\to \mathbb{R}\cup\{+\infty\}$43 and $f:X\to \mathbb{R}\cup\{+\infty\}$44 satisfying an inclusion of the form

$f:X\to \mathbb{R}\cup\{+\infty\}$45

together with nonnegativity and complementarity-index zeroing conditions. GS-stationarity strengthens this by imposing $f:X\to \mathbb{R}\cup\{+\infty\}$46 for all degenerate indices $f:X\to \mathbb{R}\cup\{+\infty\}$47 (Mishra et al., 3 Sep 2025).

Necessary and sufficient optimality results follow. If $f:X\to \mathbb{R}\cup\{+\infty\}$48 is a local minimizer, $f:X\to \mathbb{R}\cup\{+\infty\}$49 is tangentially convex and locally Lipschitz near $f:X\to \mathbb{R}\cup\{+\infty\}$50, GS-ACQ holds, and the cone

$f:X\to \mathbb{R}\cup\{+\infty\}$51

is closed, then $f:X\to \mathbb{R}\cup\{+\infty\}$52 is GS-stationary. Under MPEC-ACQ and tangential convexity of the effective data, $f:X\to \mathbb{R}\cup\{+\infty\}$53 is GA-stationary. Conversely, if $f:X\to \mathbb{R}\cup\{+\infty\}$54 is GA-stationary, $f:X\to \mathbb{R}\cup\{+\infty\}$55 is $f:X\to \mathbb{R}\cup\{+\infty\}$56-pseudoconvex, the active constraints satisfy the stated $f:X\to \mathbb{R}\cup\{+\infty\}$57-quasiconvexity assumptions, and the multiplier-index sets obey

$f:X\to \mathbb{R}\cup\{+\infty\}$58

then $f:X\to \mathbb{R}\cup\{+\infty\}$59 is a global minimizer (Mishra et al., 3 Sep 2025).

6. Relations, examples, and recurring distinctions

Several standard examples clarify how the different frameworks relate. For $f:X\to \mathbb{R}\cup\{+\infty\}$60 at $f:X\to \mathbb{R}\cup\{+\infty\}$61, the epigraph tangent cone is

$f:X\to \mathbb{R}\cup\{+\infty\}$62

hence $f:X\to \mathbb{R}\cup\{+\infty\}$63 and

$f:X\to \mathbb{R}\cup\{+\infty\}$64

For $f:X\to \mathbb{R}\cup\{+\infty\}$65 at the origin, one obtains $f:X\to \mathbb{R}\cup\{+\infty\}$66; for $f:X\to \mathbb{R}\cup\{+\infty\}$67 with $f:X\to \mathbb{R}\cup\{+\infty\}$68, one gets $f:X\to \mathbb{R}\cup\{+\infty\}$69 (Pang, 2011).

In the support-function framework, $f:X\to \mathbb{R}\cup\{+\infty\}$70 at $f:X\to \mathbb{R}\cup\{+\infty\}$71 satisfies $f:X\to \mathbb{R}\cup\{+\infty\}$72, and the paper records

$f:X\to \mathbb{R}\cup\{+\infty\}$73

In the MPEC example with objective $f:X\to \mathbb{R}\cup\{+\infty\}$74 and feasible set $f:X\to \mathbb{R}\cup\{+\infty\}$75, the tangential subdifferentials

$f:X\to \mathbb{R}\cup\{+\infty\}$76

support a direct verification of GS-stationarity (Mishra et al., 3 Sep 2025).

In the partial-smoothness framework, the representative example is the $f:X\to \mathbb{R}\cup\{+\infty\}$77 norm. There the tangential subdifferential is not a support-function dual object but the tangent-space projection of the limiting subdifferential: $f:X\to \mathbb{R}\cup\{+\infty\}$78 which equals the Riemannian gradient of the restriction to the active manifold. A different phenomenon occurs for $f:X\to \mathbb{R}\cup\{+\infty\}$79, where the tangential component vanishes and the normal component carries the structure (Qin et al., 26 Jan 2025).

The directional framework adds a further distinction: tangential information depends on the chosen feasible direction $f:X\to \mathbb{R}\cup\{+\infty\}$80. For indicators,

$f:X\to \mathbb{R}\cup\{+\infty\}$81

so the tangential contribution of the constraint can be zero in interior-pointing directions and nontrivial in boundary-preserving directions. This directional selectivity is precisely what makes the directional limiting approach sharper than non-directional limiting calculus (Benko et al., 2017).

Across these formulations, several equivalences recur under regularity. In the smooth case, the tangential subdifferential collapses to the gradient; for convex functions it coincides with the convex subdifferential; and under Clarke regularity it coincides with the Clarke subdifferential (Pang, 2011, Mishra et al., 3 Sep 2025). What changes from one framework to another is not the objective of encoding first-order nonsmooth information, but the geometric mechanism used to extract the tangential part: contingent epigraph geometry, direction-filtered limiting normals, projection onto an active tangent manifold, or support-function dualization of a sublinear directional derivative.