Legendre transforms connecting them: Visual walkthrough

Step 1 — A curve is just a hill of points

WHAT. Draw a bowl-shaped curve. Call the horizontal axis $x$ and the height $f (x)$ .

WHY. In thermodynamics $x$ will be entropy $S$ and the height will be energy $U$ . But for now forget physics — just a curve. Everything hinges on one fact: a curve can be described two totally different ways, and both hold the same information.

PICTURE. Look at the figure. The curve is drawn as a string of dots $(x, f (x))$ . That is description #1: "tell me your height at every horizontal position."

Figure — Legendre transforms connecting them

Step 2 — The slope: a second name for each point

WHAT. At one point on the curve, draw the tangent line — the straight line that just kisses the curve. Its steepness is the slope, which we name $p$ .

WHY. Sliding along a bowl, the slope changes smoothly: gentle on the left, steeper on the right. Because it never repeats on a one-way-bending curve, the slope alone can act as an address. Instead of "the point at $x = 2$ ," we can say "the point where the slope is $p$ ." This swap — position for slope — is the entire idea.

PICTURE. The red tangent line touches the curve; its rise-over-run triangle is drawn beside it. The green arrow shows that as we walk right, the slope grows. Each slope value points to exactly one spot.

Step 3 — Where does that tangent line hit the wall?

WHAT. Extend the tangent line back to the vertical axis ( $X = 0$ ). It crosses at some height. Call that crossing the intercept $g$ .

WHY. A straight line is completely fixed by two numbers: its slope and its intercept. We already have the slope $p$ . If we also record where the line hits the axis, we have captured the whole tangent line — and a bowl is nothing but the family of all its tangent lines.

PICTURE. The tangent line is continued (dashed) leftward until it strikes the vertical axis. The little dot on that axis is $g$ . Notice: for a bowl the line dips below the curve's own starting height, so $g$ often sits lower than $f (x)$ — that gap is exactly $p \cdot x$ .

Let us read the intercept off the geometry. The tangent line through the point $(x, f (x))$ with slope $p$ is $Y = height at the touch point f (x) + slope p (horizontal distance travelled X - x) .$ Set $X = 0$ to reach the axis: $g = f (x) - p x .$

Step 4 — Check nothing was lost: the mirror slope

WHAT. Ask: if I only kept the intercepts $g (p)$ , could I rebuild the original curve? Differentiate $g = f - p x$ .

WHY. Re-packaging is only useful if it's reversible. We want proof that the tangent-line description hides none of the curve.

PICTURE. The figure shows the curve $f$ (lavender) beside its transform $g$ (coral). An arrow marks that where $f$ has slope $p$ , the new curve $g$ has slope $- x$ at the location $p$ . The two are perfect mirror-partners.

Differentiate carefully, treating $x$ as depending on $p$ : $d g = f^{'} d x = p d x df - product rule p d x - product rule x d p = (p - p) d x - x d p = - x d p .$

Step 5 — The degenerate case: a straight line has no transform

WHAT. Test the machine on a curve that does not bend: a straight line $f (x) = m x + c$ .

WHY. The contract says cover every case. A straight line has the same slope $m$ everywhere — so "address a point by its slope" fails: infinitely many points share slope $m$ , and no other slope value exists at all. The inversion $p = f^{'} (x)$ breaks.

PICTURE. A perfectly straight line with every tangent identical to itself; the "slope axis" collapses to a single point $p = m$ . The transform is defined only there, and there it gives a single number — the whole line's intercept $c$ .

Step 6 — Apply it to energy: born of $U (S, V, N)$

WHAT. Take the real curve: internal energy $U$ plotted against entropy $S$ (holding $V, N$ fixed). Its slope is temperature, $T = \partial U / \partial S$ . Legendre-transform on the $S$ -slot.

WHY. $S$ is not a lab knob; $T$ is. We want a potential whose independent variable is $T$ . This is exactly Step 3, with $x \to S$ , $f \to U$ , $p \to T$ .

PICTURE. The $U$ -vs- $S$ bowl with a tangent of slope $T$ . Its intercept on the energy axis is the new potential $F$ . The vertical drop from $U$ down to that intercept is the term $T S$ .

Copy the intercept formula with the thermodynamic names: $F = height (energy) U - slope T position S = U - T S .$ Differentiate using $d U = T d S - P d V + μ d N$ (the first law, see Internal Energy and the First Law): $d F = T d S - P d V + μ d N d U - d (T S) T d S - S d T = - S d T - P d V + μ d N .$ The $T d S$ cancelled; the driver is now $d T$ .

Step 7 — The sign trap: transforming on volume

WHAT. Now transform on the $V$ -slot to swap $V$ for pressure. But the slope of $U$ in $V$ is $\partial U / \partial V = - P$ — it carries a minus sign (from $d U = \dots - P d V$ ).

WHY. The recipe subtracts (slope) $\times$ (variable). If we blindly subtract $P V$ we get the wrong potential. We must read the slope's sign off the differential first.

PICTURE. The $U$ -vs- $V$ curve slopes downward, so its tangent slope is negative, $= - P$ . Subtracting a negative = adding. The figure highlights the double-minus turning into a plus.

$slope = \frac{\partial U}{\partial V} = - P \Rightarrow H = U - (slope) V = U - (- P) V = U + P V .$

Transform on both slots at once → the lab's favourite, $G = U - T S + P V$ , natural in $(T, P, N)$ — the potential behind Chemical Potential and Phase Equilibrium.

The one-picture summary

WHAT. One diagram stacks the whole story: a convex curve, a tangent, its slope $p$ , its intercept $g = f - p x$ , and the four-potential "square" that results when $x, f$ become $S, U$ .

The map from geometry to physics, in one line:

The last arrow leads straight to Maxwell Relations; the same tangent-line trick reappears in mechanics as Lagrangian → Hamiltonian, and the intercepts show up statistically in Partition Functions.

Recall Feynman retelling (plain words, no symbols)

Imagine a smooth valley drawn on paper. You can describe that valley by listing its height at every step across the ground. Or — here's the trick — you can describe it by listing, for every possible steepness of hillside, the exact tilted line that grazes the valley there, and where that line would hit the far wall.

Those two descriptions hold the identical valley. The second one — "for each steepness, here's the grazing line's wall-height" — is the Legendre transform. Its magic is reversibility: the steepness of the new description hands you back your original horizontal position, so you never lose the valley.

Now rename things: the ground is entropy, the height is energy, the steepness of the energy valley is temperature. A thermostat sets the steepness (temperature), never the entropy directly. So we re-describe energy by steepness — and out pops the free energy $F = U - T S$ . Read its steepness in the temperature direction, you recover the entropy; read it in the volume direction, you recover the pressure. Same information, re-packaged so the lab's knobs are the handles. Do it on volume instead (watching the minus sign hidden in $- P d V$ ) and you get enthalpy; do both and you get Gibbs energy. Every step is the same one picture: a tangent line, its slope, its intercept.

Recall Quick self-test

The transform re-addresses points of a curve by their ::: slope $p$ instead of position $x$ . The transform value $g$ is geometrically the ::: intercept of the tangent line (its height at $X = 0$ ). The proof no info is lost is $d g / d p =$ ::: $- x$ , so transforming twice returns the original. Why $H = U + P V$ not $U - P V$ ? ::: the slope $\partial U / \partial V = - P$ , and subtracting $- P \cdot V$ adds $P V$ . A straight-line $f$ has no full transform because ::: its slope is constant, so slopes can't address distinct points.