Manifolds Unfolded

Differentiation is one of the first landmarks you meet in a calculus journey. It gives us the slope of the tangent — a local, linear approximation of a function. But what happens when that tangent doesn’t behave nicely? What if the slope from the left and the slope from the right disagree? Or what if the function has a cusp or kink?

In this post, we’ll explore symmetric derivatives, a subtle generalization of the usual derivative that can “exist” even where the ordinary derivative does not. We’ll also touch on ideas like generalized pseudo-derivatives and understand why these objects are mathematically interesting.

---

The Usual Derivative: A Quick Recap

When we define the derivative of a real function $f$ at a point $x$, we mean:$$f'(x) \;=\; \lim_{h\to 0} \frac{f(x+h)-f(x)}{h},$$provided this limit exists. This captures the instantaneous rate of change — geometrically, the slope of the tangent line. But this requires the left-hand and right-hand rates of change to agree.

In many simple functions (polynomial, exponential, trigonometric), this works beautifully. But for some functions, this limit fails to exist, yet there’s still intrinsic structure worth capturing.

---

Enter the Symmetric Derivative

The symmetric derivative softens the definition by averaging information from both sides of a point. Instead of a one-sided increment, we use a central difference:

$f_s'(x):=\lim\limits_{h\to 0}\frac{f(x+h)-f(x-h)}{2h}$

This is called the symmetric differential quotient or symmetric derivative of $f$ at $x$.. If this limit exists, we say $f$ is symmetrically differentiable at $x$.

What’s the Intuition?

The symmetric difference quotient:

(a) looks at the slope of the secant line between $(x-h, f(x-h))$ and $(x+h, f(x+h))$(b) centrally balances both sides,
(c) and measures the average local change around $x$.

Unlike the usual derivative, this symmetric limit can exist even when the one-sided limits do not agree. Even when the left hand derivative and right hand derivatives fail to exist, the symmetric derivative can still exist!

---

A Classic Example: $f(x) = |x|$

We all know that $f(x)=|x|$ has a sharp corner at $x=0$. The usual derivative doesn’t exist there because from the right, the slope is +1, whereas from the left, the slope is -1.

These disagree, so $f'(0)$ doesn’t exist.

But the symmetric derivative does exist — because the symmetric differential quotient is:$$\lim_{h\to 0} \frac{|h|-|{-h}|}{2h} = \lim_{h\to 0} \frac{|h|-|h|}{2h} = 0.$$.

Thus, $f’_s(0)=0$ even though $f'(0)$ doesn’t exist!

This simple calculation signals something deep: symmetric derivatives can capture average slopes across singularities.

---

Relationship Between Ordinary and Symmetric Derivatives

A few important facts tie these together:

1. If the regular derivative $f'(x)$ exists, it can be easily shown that the symmetric derivative $f’_s(x)$ also exists, and they are equal.
2. But the converse is not true: symmetric differentiability does not imply usual differentiability, as the above example of $f(x)=|x|$ just showed.

This means the symmetric derivative is a strictly weaker notion — it exists in some cases where the usual derivative doesn’t.

---

More Exotic Examples

$f(x)=\begin{cases}\frac{1}{x^2}\text{, if }x\neq 0\\0,\text{ if }x=0\end{cases}$ ; at $x=0$

Classically, $f$ isn’t even defined at 0 — and nearby it shoots off to infinity. It has an infinite discontinuity at 0. Yet the symmetric quotient yields:$$\frac{f(h)-f(-h)}{2h} = \frac{1/h^2 – 1/h^2}{2h} = 0 \quad (\text{for } h\neq 0).$$So the symmetric derivative exists and is 0 — a strong example of how symmetric derivatives can “exist” even where the function has an infinite discontinuity, let alone being differentiable.

$f(x)=\begin{cases}\cos\big(\frac{1}{x}\big),\text{ if }x\neq 0\\0\text{ if }x=0\end{cases}$ ; at $x=0$

$f$ isn’t even defined at 0 — and it has an essential discontinuity / 2nd kind non-removable discontinuity at 0, as the function oscillates without even the one sides limits. Yet the symmetric quotient yields: $\frac{f(h)-f(-h)}{2h}=\frac{\cos\big(\frac{1}{h}\big)-\cos\big(-\frac{1}{h}\big)}{2h}=0;$since the cosine function is an even function! So the symmetric derivative exists and is 0 — a strong example of how symmetric derivatives can “exist” even when the function may be extremely ill-behaved near the point of consideration.

In fact, observe that if a function is symmetric about x, in that case, the symmetric derivative always exists and is 0, as the left and right side imbalances cancel out; irrespective of how bad behaved the function is.

---

Does This Generalize?

Yes! There are higher-order symmetric derivatives, defined with central differences. The second symmetric derivative is: $f”_{s}(x)=\lim\limits_{h\to 0}\frac{f(x+h)-2f(x)+f(x-h)}{h^2};$which again can exist even if the classical second derivative doesn’t.

---

So What Are Pseudo-derivatives?

In the context of generalizations of differentiability, terms like pseudo-derivative or generalized derivative refer to any extension of the derivative concept that remains useful beyond the classical scope — such as:

• Symmetric derivatives,
• Peano derivatives,
• Generalized Riemann derivatives,
• Weak derivatives from distribution or Sobolev space theory.

We may call symmetric derivatives “pseudo-derivatives” to emphasize that they behave like derivatives in some ways but relax classical requirements. Officially in analysis literature, the symmetric derivative is a well-studied generalized derivative that often appears in real analysis and approximation theory.

Asymmetric Derivatives

In fact, one also studies asymmetric derivatives such as:

$f’_{a,b}(x):=\lim\limits_{h\to 0}\frac{f(x+ah)-f(x+bh)}{(a-b)h};\text{ with }a,b\in \mathbb R\setminus\{0\}\text{ and }a\neq \pm b$

In fact, there are results, on how differentiability is connected with the existence of different asymmetric derivatives.

Theorem (Characterization of Differentiability via Asymmetric Derivatives):

Let $f:\mathbb R\to \mathbb R$ and $x\in \mathbb R$. Then $f$ is differentiable at $x$ if and only if:

1. $f$ is continuous at $x$, and
2. For all $a,b\in \mathbb R\setminus\{0\}$ with $a \neq \pm b$, the limit

$\lim\limits_{h\to 0}\frac{f(x+ah)-f(x+bh)}{(a-b)h}$ exists.

Moreover, when $f$ is differentiable at $x$, all such limits coincide with the usual derivative, i.e., $f’_{a,b}(x)=f'(x)\text{ for all }a,b\in \mathbb R\setminus \{0\} \text{ with }a\neq b.$

---

What Does This Buy Us?

Symmetric derivatives are more than just curiosities:

Better Numerical Behavior

In numerical differentiation, central differences like $\frac{f(x+h)-f(x-h)}{2h}$ often yield higher accuracy than forward or backward differences because error terms cancel symmetrically.

Capturing Local Symmetry

Functions with balanced left and right behaviors at a point — like $|x|$ near 0 — have meaningful symmetric derivatives.

Functional Analysis and Generalized Theory

Symmetric derivatives show up in refined studies — for instance:

1. time-scale calculus, where one defines derivatives on discontinuous domains, and,
2. generalized Riemann derivatives that interpolate between discrete and continuous differentiability.

---

Final Thoughts

The symmetric derivative invites us to rethink what it means for a function to be “locally linear.” Rather than demanding left and right limits match perfectly, it asks whether the balanced difference across a point converges as we zoom in. Sometimes it does — even when the classic derivative refuses to exist.

This is a beautiful example of how mathematics often takes a familiar concept, relaxes a condition, and discovers new structure hidden just beneath the surface.