The Fundamental Theorem for Gradients

The fundamental theorem of single-variable calculus says that integrating a derivative recovers the original function up to its boundary values: $\int_a^b (\mathrm{d}f/\mathrm{d}x)\,\mathrm{d}x = f(b)-f(a)$ . In several dimensions a scalar field $T(\mathbf{r})$ has no single derivative but a gradient $\nabla T$ , a vector whose components are the partial derivatives. The fundamental theorem for gradients is the exact analogue: summing the directional change $(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}$ along any path from $\mathbf{a}$ to $\mathbf{b}$ reproduces the total change $T(\mathbf{b})-T(\mathbf{a})$ . The geometric content is that $(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}$ is precisely the infinitesimal change $\mathrm{d}T$ experienced as one steps by $\mathrm{d}\boldsymbol{\ell}$ . Adding up these infinitesimal changes is a telescoping sum: all the intermediate values cancel and only the endpoints survive. This is why the integral cannot depend on the route taken, only on where it begins and ends. A field that can be written as a gradient is called conservative, and path-independence is its defining property. The closed-loop corollary, $\oint(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=0$ , is the seed of two deeper facts: a gradient field has zero curl, and — by Stokes' theorem, on simply connected domains — the converse holds too. In electromagnetism the theorem is indispensable. The electrostatic field is a gradient, $\mathbf{E}=-\nabla V$ , so the work per unit charge moved between two points is path-independent and equals a potential difference. The same structure underlies magnetic scalar potentials in current-free regions and the definition of voltage in circuit theory. Understanding this theorem is the cleanest way to see why 'potential' is a meaningful, single-valued concept — and it is the first of the three great integral theorems of this chapter, the pattern the divergence and Stokes theorems will repeat in higher dimension.

Key takeaways

The fundamental theorem for gradients states that $\int_{\mathbf{a}}^{\mathbf{b}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell} = T(\mathbf{b})-T(\mathbf{a})$ , the net change in the scalar field between the endpoints.
It is the multivariable generalization of the fundamental theorem of calculus, with $\nabla T$ playing the role of the derivative and the line integral the role of the antiderivative.
Because the result depends only on the endpoints, the line integral of a gradient is path-independent, and around any closed loop it vanishes: $\oint (\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=0$ .
This theorem is the mathematical foundation of scalar potentials: in electrostatics $\mathbf{E}=-\nabla V$ implies $V(\mathbf{b})-V(\mathbf{a})=-\int_{\mathbf{a}}^{\mathbf{b}}\mathbf{E}\cdot \mathrm{d}\boldsymbol{\ell}$ .
A field that is a gradient is curl-free; the closed-loop statement here is the bridge to Stokes' theorem, which makes the equivalence precise.
Together with the divergence theorem and the curl theorem, it forms the trio of integral theorems that vector calculus uses to convert between derivatives and integrals over boundaries.
Key idea: The line integral of a gradient telescopes: $(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=\mathrm{d}T$ , so summing along a path gives $T(\mathbf{b})-T(\mathbf{a})$ — depending only on endpoints, never on the path.

Where this lives today

Every height on Earth is this theorem in bureaucratic form. "Elevation above sea level" is really a statement about the gravitational potential: geodesists define the geoid as an equipotential surface and measure heights as potential differences — path-independent by exactly this theorem, which is why two survey crews taking different routes up a mountain must agree. Modern GNSS positioning converts geometric satellite heights to physical heights using precisely these potential models, and the new definition of the second is being tied to it too: optical clocks tick measurably slower at lower gravitational potential, so clock networks now measure potential differences (relativistic geodesy) at the centimetre-equivalent level. In computational chemistry, reactions are paths on a potential-energy surface: the energy released by a reaction is the endpoint difference, path-independent, while the barrier along the way depends on the route — the whole art of catalysis is rerouting the path without touching the endpoints. And in your pocket, the battery's labelled voltage is an endpoint statement guaranteed meaningful by the vanishing-loop corollary; the moment fast-changing magnetic fields enter (wireless chargers, switching converters), engineers must reckon with the non-conservative loop integrals that the warning box flagged.

The physical picture

Picture walking across a hilly landscape where $T$ is your altitude. Each small step $\mathrm{d}\boldsymbol{\ell}$ changes your altitude by $\mathrm{d}T=\nabla T\cdot \mathrm{d}\boldsymbol{\ell}$ — uphill when you move along the gradient, downhill against it, no change when you walk along a level contour. The total altitude gained between start and finish is fixed by the two points; the meandering route you took is irrelevant.

Figure

Two hikers, one mountain — every route from a to b gains exactly the same altitude

Think of the integral as a telescoping sum. Slice the path into tiny segments; the integral is $\sum \mathrm{d}T = (T_1-T_0)+(T_2-T_1)+\cdots+(T_N-T_{N-1})$ . Every interior value appears once with each sign and cancels, leaving only $T_N - T_0$ . Path-independence is just the statement that this cancellation does not care how the intermediate points were chosen. The closed-loop version is intuitive too: if you return to your starting point, your net altitude change is zero. You cannot keep gaining height forever by walking in circles on a fixed landscape — that would let you extract energy from nothing, which is exactly why such fields are called conservative. The E&M payoff is the concept of voltage itself: because the electrostatic field is a gradient, "the potential difference between these two terminals" is a single number, not a number-per-route — and every circuit diagram you will ever draw quietly relies on this theorem.

The theorem and its immediate family

Here is the statement, deserving of its name:

\int_{\mathbf{a}}^{\mathbf{b}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=T(\mathbf{b})-T(\mathbf{a}) \qquad (1)

where: $T(\mathbf{r})$ — differentiable scalar field; $\nabla T$ — its gradient; $\mathrm{d}\boldsymbol{\ell}$ — infinitesimal displacement along any path from $\mathbf{a}$ to $\mathbf{b}$ . A derivative-like object integrated over a region (the path) equals the original object evaluated on the region's boundary (the two endpoints) — the pattern the divergence and Stokes theorems will replay with volumes and surfaces. The engine inside is the gradient's master relation, inherited from its own article:

\mathrm{d}T=(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell} \qquad (2)

where: $\mathrm{d}T$ — the change of $T$ under the step $\mathrm{d}\boldsymbol{\ell}$ ; this is the multivariable chain rule, $\mathrm{d}T = \partial_x T\,\mathrm{d}x + \partial_y T\,\mathrm{d}y + \partial_z T\,\mathrm{d}z$ , recognized as a dot product. Every ounce of the theorem's power is here: the integrand is an exact differential, so integrating it is bookkeeping. Two corollaries follow by choosing endpoints cleverly. Close the path and the difference dies:

\oint_{\mathcal{C}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=0 \qquad (3)

where: $\mathcal{C}$ — any closed curve in the domain of a single-valued $T$ . And compare two paths sharing endpoints — their difference is a closed loop, so:

\int_{\mathcal{P}_1}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=\int_{\mathcal{P}_2}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell} \qquad (4)

where: $\mathcal{P}_1,\mathcal{P}_2$ — any two paths from $\mathbf{a}$ to $\mathbf{b}$ . Path-independence, closed loops vanishing, and being a gradient are three faces of one property. The differential-side shadow of the same fact is the identity proved in the gradient article:

\nabla\times(\nabla T)=\mathbf{0} \qquad (5)

where: each curl component is a difference of mixed partials that commute. Stokes' theorem (two articles ahead) will close the circle, showing that on simply connected domains zero curl is also sufficient for a field to be a gradient.

Derivation AProof of the fundamental theorem for gradients

We are proving equation (1) — the statement that powers the entire potential formalism of electrostatics. The strategy is to reduce the multivariable claim to the one-dimensional fundamental theorem of calculus: parametrize the path, recognize $(\nabla T)\cdot(\mathrm{d}\mathbf{r}/\mathrm{d}s)$ as the ordinary derivative of $T$ along the curve via the chain rule, and let 1D calculus finish. The step to savour is the middle one — it is the same $\mathrm{d}T = \nabla T\cdot\mathrm{d}\boldsymbol{\ell}$ identity that defined the gradient, now cashing its promised dividend. Note what is never used: any property of the path beyond smoothness. That silence is the theorem's content.

Step 1. Parametrize the path by a scalar $s$ :

\mathbf{r}(s),\quad s\in[s_a,s_b],\quad \mathbf{r}(s_a)=\mathbf{a},\ \ \mathbf{r}(s_b)=\mathbf{b}, \qquad \mathrm{d}\boldsymbol{\ell}=\frac{\mathrm{d}\mathbf{r}}{\mathrm{d}s}\,\mathrm{d}s

Why: any smooth traversal works — the line-integrals article proved the value is reparametrization-invariant, so we lose nothing by picking one.

Step 2. Apply the multivariable chain rule to the composite function $T(\mathbf{r}(s))$ :

(\nabla T)\cdot\frac{\mathrm{d}\mathbf{r}}{\mathrm{d}s} = \frac{\partial T}{\partial x}\frac{\mathrm{d}x}{\mathrm{d}s}+\frac{\partial T}{\partial y}\frac{\mathrm{d}y}{\mathrm{d}s}+\frac{\partial T}{\partial z}\frac{\mathrm{d}z}{\mathrm{d}s} = \frac{\mathrm{d}}{\mathrm{d}s}\,T(\mathbf{r}(s))

Why: along the curve, $T$ is a function of the single variable $s$ , and its total derivative routes through all three coordinates — the integrand is revealed as an ordinary derivative in disguise.

Step 3. The line integral is now one-dimensional:

\int_{\mathbf{a}}^{\mathbf{b}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=\int_{s_a}^{s_b}\frac{\mathrm{d}}{\mathrm{d}s}\,T(\mathbf{r}(s))\,\mathrm{d}s

Why: substituting Step 1 and Step 2 leaves the integral of a total derivative of a single-variable function — exactly the object the 1D fundamental theorem devours.

Step 4. Apply the one-dimensional fundamental theorem of calculus:

= T(\mathbf{r}(s_b))-T(\mathbf{r}(s_a)) = T(\mathbf{b})-T(\mathbf{a})

The answer references only the endpoints — the parametrization and the route have vanished from the result, which is the theorem. Verified: matches Griffiths §1.3.3; the worked example below checks it numerically along two different routes for $T=xy^2$ .

Go deeper · Derivation B · 1 minThe closed-loop corollary and path independence

Derivation BThe closed-loop corollary and path independence

We are extracting the two corollaries (3) and (4) that carry all the physical weight: closed loops of gradient fields vanish, and open-path integrals depend only on endpoints. The strategy is endpoint gymnastics on the boxed theorem — no new calculus, just the observation that a closed loop is a path whose endpoints coincide, and that two paths with shared endpoints differ by a closed loop. This little argument is the template for every "conservative force" statement in mechanics and every well-defined voltage in circuits.

Step 1. For a closed curve, set $\mathbf{b}=\mathbf{a}$ in the theorem:

\oint_{\mathcal{C}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=T(\mathbf{a})-T(\mathbf{a})=0

Why: the theorem never cared about the route, so even a wildly meandering loop must return the endpoint difference — which is now zero. This assumes $T$ is single-valued: after the round trip, $T$ must come back to the same number.

Step 2. Split any loop through $\mathbf{a}$ and $\mathbf{b}$ into an outbound path $\mathcal{P}_1$ and a return path $\mathcal{P}_2$ reversed:

0 = \oint_{\mathcal{C}} = \int_{\mathcal{P}_1} - \int_{\mathcal{P}_2}

Why: traversing $\mathcal{P}_2$ backwards flips the sign of its integral (orientation reversal from the line-integrals article), so the loop is the difference of the two open-path integrals.

Step 3. Conclude path independence:

\int_{\mathcal{P}_1}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=\int_{\mathcal{P}_2}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}

for any two paths sharing endpoints. The three properties — gradient structure, vanishing loops, path independence — are logically interlocked: each implies the others (on simply connected domains, via Stokes for the converse direction). Verified: standard corollary chain, Purcell §2.3; consistent with Stokes' theorem applied to $\nabla\times\nabla T=\mathbf{0}$ .

The electrostatic dividend

Attach the theorem to $\mathbf{E}=-\nabla V$ and voltage is born:

\mathbf{E}=-\nabla V\ \Longrightarrow\ V(\mathbf{b})-V(\mathbf{a})=-\int_{\mathbf{a}}^{\mathbf{b}}\mathbf{E}\cdot \mathrm{d}\boldsymbol{\ell} \qquad (6)

where: $V$ — electrostatic scalar potential; $\mathbf{E}$ — electrostatic field. The right side is a line integral that could, in principle, depend on the route between the terminals; the theorem guarantees it does not. That guarantee is why a battery can be labelled "1.5 V" without a footnote specifying the path. The energy bookkeeping follows immediately:

W=q\,[V(\mathbf{b})-V(\mathbf{a})] \qquad (7)

where: $W$ — work done by an external agent moving charge $q$ slowly from $\mathbf{a}$ to $\mathbf{b}$ against the field. One number per pair of points, whatever the route — the whole energetics of circuits and electron optics rests on equations (6) and (7). There is also a constructive reading, which the boundary-value chapter will use: given any curl-free field $\mathbf{F}$ on a simply connected region, define $T(\mathbf{r})=\int_{\mathbf{r}_0}^{\mathbf{r}}\mathbf{F}\cdot\mathrm{d}\boldsymbol{\ell}$ from a reference point $\mathbf{r}_0$ along any convenient path. Path-independence makes the definition unambiguous, and differentiating recovers $\nabla T = \mathbf{F}$ . This is exactly how the potential $V$ is manufactured from $\mathbf{E}$ in the electrostatics chapter — reference point at infinity, path of your choosing.

Figure

One pattern, three theorems — integrate a derivative inside, read the answer on the boundary

Limiting cases

Special configurations sharpen the meaning of the theorem:

Case	Result	Meaning
Closed loop	$\oint_{\mathcal{C}}(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=0$	When $\mathbf{a}=\mathbf{b}$ the right side vanishes for any loop, the defining property of a conservative field.
Path along a contour	$\nabla T\perp \mathrm{d}\boldsymbol{\ell}\Rightarrow \int(\nabla T)\cdot \mathrm{d}\boldsymbol{\ell}=0$	Moving along a level surface of $T$ contributes nothing, since the gradient is normal to level surfaces.
Constant field	$T=\text{const}\Rightarrow \nabla T=\mathbf{0},\ \int=0$	A flat scalar field has no gradient; every line integral is trivially zero.
Electrostatic potential	$V(\mathbf{b})-V(\mathbf{a})=-\int_{\mathbf{a}}^{\mathbf{b}}\mathbf{E}\cdot \mathrm{d}\boldsymbol{\ell}$	With $\mathbf{E}=-\nabla V$ the theorem defines potential difference independent of path.
One dimension	$\int_a^b f'(x)\,\mathrm{d}x = f(b)-f(a)$	The theorem collapses to the ordinary fundamental theorem of calculus — its direct ancestor.

Worked examples

Example 1 — the theorem checked along two routes (confidence builder). Take $T = xy^2$ and go from $\mathbf{a}=(0,0,0)$ to $\mathbf{b}=(2,1,0)$ . The theorem promises $T(\mathbf{b})-T(\mathbf{a}) = 2\cdot 1^2 - 0 = 2$ for every path. The gradient is $\nabla T = y^2\,\hat{\mathbf{x}} + 2xy\,\hat{\mathbf{y}}$ .

Step 1. Route A — along the $x$ -axis, then up. Leg 1, $(0,0)\to(2,0)$ with $y=0$ : the integrand $y^2\,\mathrm{d}x = 0$ , contributing $0$ . Leg 2, $(2,0)\to(2,1)$ with $x=2$ , $\mathrm{d}x=0$ :

\int_0^1 2xy\,\mathrm{d}y = \int_0^1 4y\,\mathrm{d}y = 2

Total: $0 + 2 = \mathbf{2}$ .

Step 2. Route B — the straight line $x = 2t$ , $y = t$ , $t: 0\to 1$ , so $\mathrm{d}x = 2\,\mathrm{d}t$ , $\mathrm{d}y = \mathrm{d}t$ :

\int_0^1\big(y^2\cdot 2 + 2xy\cdot 1\big)\,\mathrm{d}t = \int_0^1\big(2t^2 + 4t^2\big)\,\mathrm{d}t = \int_0^1 6t^2\,\mathrm{d}t = 2

Both routes deliver exactly 2 — the endpoint difference $T(\mathbf{b})-T(\mathbf{a})$ , as the theorem demands. Compare the line-integrals article, where a non-gradient field gave 11, 9 and 10 along three routes: the contrast is the whole distinction between conservative and non-conservative fields, computed by hand.

Example 2 — from voltage to electron speed (exam level). In an electron gun, an electron ( $q = -1.60\times10^{-19}\ \mathrm{C}$ , $m = 9.11\times10^{-31}\ \mathrm{kg}$ ) is released at rest from the cathode and reaches the anode, held $\Delta V = V_{\text{anode}} - V_{\text{cathode}} = +2000\ \mathrm{V}$ higher. How fast does it arrive — and why does the crooked route through the gun not matter?

Step 1. By the theorem via equation (6), the potential difference between the electrodes is one number, path-independent. The field does work $W_{\text{field}} = -q\,\Delta V$ on the charge along any trajectory:

W_{\text{field}} = -(-1.60\times10^{-19})(2000) = +3.20\times10^{-16}\ \mathrm{J}

Step 2. Work-energy theorem, starting from rest:

\tfrac12 mv^2 = 3.20\times10^{-16}\ \mathrm{J} \quad\Longrightarrow\quad v = \sqrt{\frac{2(3.20\times10^{-16})}{9.11\times10^{-31}}} = \sqrt{7.02\times10^{14}} = 2.65\times10^{7}\ \mathrm{m/s}

The electron arrives at 26,500 km/s (about 9% of light speed) — computed without knowing anything about its actual winding path through the fields. That is the theorem working as engineering: only the endpoint potentials enter. (At this speed the relativistic correction is under 1%; at tens of kilovolts it would no longer be.)

Step 3. Sanity-check the sign machinery: the electron (negative $q$ ) accelerates toward higher potential, gaining kinetic energy as electrostatic potential energy $qV$ drops — signs consistent with equation (7) read for the field rather than the external agent.

Applications

Electrostatic potential and voltage. Because the electrostatic field obeys $\mathbf{E}=-\nabla V$ , the theorem gives $V(\mathbf{b})-V(\mathbf{a})=-\int_{\mathbf{a}}^{\mathbf{b}}\mathbf{E}\cdot \mathrm{d}\boldsymbol{\ell}$ . This makes potential difference — the 'voltage' between two points — a well-defined quantity independent of the test charge's route. The energy delivered to a charge $q$ moving through the field is $q\,[V(\mathbf{a})-V(\mathbf{b})]$ , the basis of all circuit energetics.

Work-energy and conservative forces. For any force derivable from a potential energy, $\mathbf{F}=-\nabla U$ , the work done is $W=\int\mathbf{F}\cdot \mathrm{d}\boldsymbol{\ell}=U(\mathbf{a})-U(\mathbf{b})$ , depending only on endpoints. This is precisely why gravitational and electrostatic forces conserve mechanical energy and why a particle cannot gain net kinetic energy by traversing a closed orbit in such a field.

Detecting non-conservative fields. The contrapositive is a powerful diagnostic: if you can find a closed loop with $\oint\mathbf{F}\cdot \mathrm{d}\boldsymbol{\ell}\neq 0$ , then $\mathbf{F}$ is not a gradient and no single-valued potential exists. This is exactly what happens for the induced electric field around a changing magnetic flux (Faraday's law) and for the magnetic field circulating a wire — signalling that those fields require vector potentials or have topological obstructions.

Building potentials by integration. Conversely, once a field is known to be curl-free on a simply connected region, the theorem provides a constructive recipe: pick a reference point $\mathbf{r}_0$ , set $T(\mathbf{r})=\int_{\mathbf{r}_0}^{\mathbf{r}}\mathbf{F}\cdot \mathrm{d}\boldsymbol{\ell}$ along any convenient path, and the result is a valid potential with $\nabla T=\mathbf{F}$ . Path-independence guarantees the construction is unambiguous up to the additive constant set by $\mathbf{r}_0$ .

Origins · 1 minGeorge Green and how this was discovered

George Green

1793–1841 · Nottingham and Cambridge

Green is the unlikeliest founder in mathematical physics: a miller's son with one year of schooling, he taught himself analysis from journals while running the family windmill outside Nottingham — grinding grain below, working through Laplace above. In 1828 he self-published, by subscription, an Essay on the Application of Mathematical Analysis to the Theories of Electricity and Magnetism; fifty-one neighbours bought copies few could read. The Essay coined the word potential for the function whose gradient gives the force, and built the machinery — potential theory, boundary integrals, the functions now bearing his name — on exactly the endpoint-difference logic of this article. It sank without trace. Green entered Cambridge as an undergraduate at forty, died at forty-seven, and only then did William Thomson (Kelvin) rediscover the Essay and circulate it to an astonished Europe. Maxwell built on it directly. The windmill still stands, restored as a science centre — the only physics landmark that once paid for itself in flour.

History

1687

— Newton's Principia implicitly uses path-independence of gravitational work, the physics awaiting the theorem

1788

— Lagrange computes gravitational attraction from a single scalar function, the potential in all but name

1828

— Green's self-published Essay names the potential and founds potential theory on boundary-value logic

1833

— Hamilton's optics and dynamics elevate endpoint-difference principles into a general method

1847

— Thomson (Kelvin) rediscovers Green's Essay and republishes it, injecting potential theory into mainstream physics

1873

— Maxwell's Treatise makes the potential difference of the electrostatic field a cornerstone of electromagnetic theory

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §1.3.3 · Jackson, Classical Electrodynamics, 3rd ed., §1.5 and front-cover identities · Purcell & Morin, Electricity and Magnetism, 3rd ed., §2.3 · Zangwill, Modern Electrodynamics, §1.4

Provenance. Equations are standard vector-calculus identities. A verifier should: (1) confirm the chain-rule step $\nabla T\cdot(\mathrm{d}\mathbf{r}/\mathrm{d}s)=\mathrm{d}T/\mathrm{d}s$ and the reduction to a 1D fundamental-theorem-of-calculus integral (Griffiths §1.3.3); (2) verify the identity $\nabla\times(\nabla T)=\mathbf{0}$ via equality of mixed partials (Jackson front-cover identities); (3) check the electrostatic relation $V(\mathbf{b})-V(\mathbf{a})=-\int\mathbf{E}\cdot \mathrm{d}\boldsymbol{\ell}$ against Purcell §2.3 and Griffiths Ch. 2. No empirical constants appear.

Builds on: The Gradient Operator · Line Integrals · Scalar and Vector Fields · The Del Operator (Nabla) Leads to: The Curl (Stokes) Theorem · The Curl of a Vector Field · The Divergence (Gauss) Theorem

Is this really a different theorem from the fundamental theorem of calculus, or the same one dressed up?

It is the same idea promoted one honest step: reduce to 1D and it is the classic theorem (the last row of the limiting-cases table). What is genuinely new in three dimensions is the freedom of route — in 1D there is only one way from $a$ to $b$ , so path-independence is invisible; in 3D it becomes the theorem's main content and the definition of "conservative". The proof strategy reflects this: parametrize to borrow the 1D result, then observe that nothing about the route survived. The divergence and Stokes theorems repeat the same promotion with surfaces and volumes.

If the gradient of the azimuthal angle is a legitimate gradient, why does its loop integral around the axis give 2π instead of 0?

Because $\phi$ is not a single-valued function on the full ring around the axis — circle once and it has grown by $2\pi$ , so "the endpoint difference" is genuinely $2\pi$ , not $0$ . Locally, on any patch avoiding a full wrap, $\nabla\phi = \hat{\boldsymbol\phi}/s$ is a perfectly good gradient and the theorem holds. The failure is topological: the domain (space minus the axis) is not simply connected, so local gradients need not integrate to a global potential. This is not pathology — it is precisely the structure of the magnetic field around a wire and of the Aharonov–Bohm phase.

Where exactly does the minus sign convention of potentials enter, since the theorem itself has no minus sign?

The theorem is sign-free mathematics: gradient in, endpoint difference out. Physics inserts the minus when defining potentials so that energy bookkeeping feels right: $\mathbf{E}=-\nabla V$ and $\mathbf{F}=-\nabla U$ make fields point downhill and stable equilibria sit at potential minima. Feed $-\nabla V$ into the theorem and the minus rides along: $\int\mathbf{E}\cdot\mathrm{d}\boldsymbol{\ell} = V(\mathbf{a})-V(\mathbf{b})$ . Every voltage sign headache traces to this one convention interacting with limit order — anchor on "moving against the field raises the potential" and re-derive rather than recall.

Does zero curl guarantee a field is a gradient?

On a simply connected region, yes — that is the converse direction, and it needs Stokes' theorem: zero curl makes every loop integral vanish (any loop bounds a surface through which no curl flux passes), which makes open-path integrals path-independent, which lets you build the potential by integrating from a reference point. Puncture the domain and the chain breaks at the first link: some loops bound no legal surface, circulation can survive, and only multivalued potentials exist. So the honest hierarchy is: gradient ⇒ curl-free always; curl-free ⇒ gradient only where the topology cooperates.

The potential is only defined up to a constant — does the arbitrariness ever matter?

Never for physics, always for bookkeeping. The theorem only ever delivers differences, so shifting $T$ (or $V$ ) by a constant changes nothing measurable — forces, fields, and energies are untouched. The freedom is spent by convention: circuit theory grounds a node at $0\ \mathrm{V}$ ; electrostatics of localized charges sets $V(\infty) = 0$ ; and some idealized problems (the infinite charged plane) refuse the infinity convention, forcing a finite reference point. Two solutions differing by a constant are the same physics — a first, tame glimpse of gauge freedom.

Is the voltage a voltmeter reads always the potential difference of this theorem?

Only when the fields are (quasi-)static. A voltmeter reads the line integral of the electric field along its own leads; the theorem promises this equals $V(\mathbf{a})-V(\mathbf{b})$ because the electrostatic field is a gradient. Introduce a changing magnetic flux (a transformer core, an induction cooktop) and $\mathbf{E}$ acquires curl: loop integrals no longer vanish, and two voltmeters spanning the same two points along different lead routes genuinely read different values. This staple demonstration puzzle is not a paradox — it is the boundary of this theorem's jurisdiction, patrolled by Faraday's law.

How does this theorem relate to the other two integral theorems of the chapter?

All three are one pattern — the boundary principle: the integral of a derivative over a region equals the original quantity evaluated on the region's boundary. Here the region is a curve (1D) and its boundary is two points; in Stokes' theorem the region is a surface (2D) and the boundary is its rim; in the divergence theorem the region is a volume (3D) and the boundary is its skin. Each dimension has its own derivative (gradient, curl, divergence) matched to its own region. Mathematics unifies them as the generalized Stokes theorem for differential forms — but the physical instinct "derivatives inside live on the boundary" is already the whole idea.