Gradient — Physica

Equivalent forms

\nabla f = \sum_i \frac{\partial f}{\partial x_i}\,\hat{\mathbf{e}}_i

df = \nabla f \cdot d\mathbf{r}

One symbol, ∇, eats a scalar landscape and hands back the direction you'd climb — and it is always perpendicular to the level sets.

Symbol	Name	SI unit	Dimension	Range
$\|\nabla f\|$	Gradient magnitudeoutput Steepest slope of the field at the point	1/m	L^-1	0 – 10
$a$	Field amplitude Height scale of the Gaussian hill f = a·exp(−(x²+y²)/2σ²)	1	1	0.1 – 5
$x$	x position Horizontal coordinate of the sample point	m	L	-3 – 3
$\sigma$	Hill width Standard deviation of the Gaussian hill	m	L	0.3 – 3

Unit systems

Symbol	SI	CGS	Imperial
$x$	m	cm	ft
$\sigma$	m	cm	ft
$\|\nabla f\|$	1/m	1/cm	1/ft

Where it holds

Any differentiable scalar field f on R^n.

Dimensional analysis

\nabla f \to [f]/[L]

\partial f/\partial x \to [f]/[L]

Each component divides the field's units by a length. For a dimensionless f, |

\nabla f

| has units of 1/length.

\checkmark

Discovery

William Rowan Hamilton / Josiah Willard Gibbs · 1881

Hamilton introduced the ∇ operator ("nabla", named for an ancient harp of the same shape) in the 1840s. Gibbs and Heaviside forged it into the vector calculus physicists use today, turning a page of partial derivatives into a single geometric arrow.

Try this

Drop a marble on a hill in the dark. Which way does it roll? The gradient already knows — it always points straight uphill, steepest first.

Sample a scalar field f(x,y) and watch ∇f spring out as the arrow of fastest increase, perpendicular to every contour line.

Research status: stable

Real-world applications

Gradient descent — the workhorse of machine-learning optimization
Electric field from potential, $E = -\nabla V$
Heat and diffusion fluxes, $q = -k\nabla T$
Image edge detection (intensity gradients)

Common misconceptions

The gradient is a scalar — no, it is a vector field built from scalar derivatives.
$\nabla f$ points along a contour — it points perpendicular to contours.
Bigger f means bigger gradient — the gradient depends on how fast f changes, not on f itself.

Experimental verification

Operationally verified every time a topographic map's steepest-descent water flow, an electric field

E = -\nabla V

, or a heat-flux

q = -k\nabla T

matches measurement.

Derivation

From the multivariable chain rule,

df = (\partial f/\partial x)dx + (\partial f/\partial y)dy + (\partial f/\partial z)dz = \nabla f \cdot dr

.

Among all unit directions û, the rate of change

\nabla f\cdot

û is maximized when û is parallel

to \nabla f

, with value |

\nabla f

| —

so \nabla f

is by construction the direction of steepest ascent.

Limiting cases

At a peak or valley⟶

\nabla f = 0

Extrema and saddles are exactly the points where every partial derivative vanishes.

f constant⟶

\nabla f = 0

everywhereA flat landscape has no preferred uphill direction.

Far from the hill⟶ |

\nabla f| \to 0

A localized bump becomes flat at infinity.

What if…

What if I move perpendicular

to \nabla f

?

f doesn't change at all — you're walking along a contour line.

What if the field has a sharp ridge?

The gradient is undefined on the crease; smoothness is required $for \nabla f$ to exist.

1

Slope of a paraboloid

Given ·

f:: $x^{2} + y^{2}$
point:: (3, 4)

Find ·

\nabla f

and |

\nabla f

|

Steps

$\partial f/\partial x = 2x = 6$
$\partial f/\partial y = 2y = 8$
| $\nabla f| = \sqrt{6^{2} + 8^{2}} = \sqrt{100} = 10$

Result ·

\nabla f = (2x

,

2y) = (6

, 8); |

\nabla f| = 10

2

Direction to climb fastest

Given ·

f:: $x^{2} + y^{2}$
point:: (3, 4)

Find · unit uphill direction

Steps

Uphill direction $is \nabla f/|\nabla f$ |
(6, $8)/10 = (0.6$ , 0.8)
Pointing radially outward, as expected for a bowl

Result · (6,

8)/10 = (0.6

, 0.8)

Divergence→Curl→Laplacian→directional derivative