Average energy from partition function: Visual walkthrough

Step 0 — What are the pieces? (build the vocabulary before any math)

Before a single symbol, meet the cast.

The knob we will actually turn is not $T$ but its inverse:

Figure — Average energy from partition function

Look at the figure: cold (left) piles the marble in the lowest box; hot (right) spreads it across all boxes. Same boxes, same energies — only the dial $β$ changed.

Prerequisites we lean on live here: Boltzmann distribution and Partition function (canonical ensemble).

Step 1 — How likely is each box? (the Boltzmann weight)

WHAT. Assign each box a raw "unnormalized likeliness" number.

WHY the exponential and not, say, $1/ E_{i}$ ? Because we want two independent systems' weights to multiply (energies add: $E = E_{1} + E_{2} \Rightarrow e^{- β (E_{1} + E_{2})} = e^{- β E_{1}} e^{- β E_{2}}$ ). Only the exponential converts "add the energies" into "multiply the weights." That is the unique property we need.

PICTURE. Each box gets a bar whose height is $w_{i}$ . High-energy boxes have tiny bars; the cold dial makes far bars almost invisible.

Step 2 — Make the weights into honest probabilities (enter $Z$ )

WHAT. The bars $w_{i}$ do not add up to $1$ , so they are not yet probabilities. Divide by their total.

Now the honest probability of landing in box $i$ :

WHY. A probability must sum to $1$ ; dividing by $Z$ is the one rescaling that does it. $Z$ is the normalizer.

PICTURE. Same bars as Step 2's figure, but now stacked into a full-height column of "shares."

Step 3 — Write the average energy honestly

WHAT. The average (mean) energy is each box's energy weighted by how often we visit it.

WHY. This is the only definition of an average — value times probability, summed. Nothing clever yet.

The annoyance: that sum $\sum_{i} E_{i} e^{- β E_{i}}$ has an extra factor $E_{i}$ glued onto each weight. Computing it directly, box by box, is tedious. Step 4 removes the tedium.

PICTURE. Bars again, but each bar now tagged with its energy $E_{i}$ ; the average is the "balance point" (centre of mass) of the energy values weighted by bar height.

Step 4 — The derivative trick: manufacture the missing $E_{i}$

WHAT. Ask: what happens to a single weight if we nudge the dial $β$ ?

$\frac{\partial}{\partial β} e^{- β E_{i}} = - E_{i} e^{- β E_{i}}$

$\frac{\partial}{\partial β}$ means "rate of change as $β$ increases, holding energies fixed."
The chain rule: derivative of $e^{(stuff)}$ is $e^{(stuff)}$ times the derivative of the stuff. Here stuff $= - β E_{i}$ , whose slope in $β$ is $- E_{i}$ .

WHY this tool. We wanted an extra $E_{i}$ multiplying each weight (Step 3's annoyance). Differentiation in $β$ produces exactly that factor $E_{i}$ — for free, with a minus sign attached. That is why the derivative, and not any other operation, is the right key.

PICTURE. Turn the dial by a tiny $δ β$ ; watch each bar shrink. A tall-energy bar (big $E_{i}$ ) shrinks by more — because its shrink rate is $- E_{i} w_{i}$ , proportional to $E_{i}$ . The picture is showing you the factor $E_{i}$ being born.

Step 5 — Replace the whole sum with one derivative of $Z$

WHAT. Add up Step 4 over every box:

$\frac{\partial Z}{\partial β} = \frac{\partial}{\partial β} \sum_{i} e^{- β E_{i}} = \sum_{i} (- E_{i}) e^{- β E_{i}} = - \sum_{i} E_{i} e^{- β E_{i}}$

The derivative of a sum = sum of derivatives, so we differentiate term by term.
The result is minus the very sum that annoyed us in Step 3.

Flip the sign to isolate that sum:

$\sum_{i} E_{i} e^{- β E_{i}} = - \frac{\partial Z}{\partial β}$

WHY. Step 3's tedious sum is now just "the slope of $Z$ ." One quantity, $Z (β)$ , and its slope carry everything.

PICTURE. Two curves of $Z$ vs $β$ (a decreasing curve). The steepness of the slope at a point is the weighted energy sum. Where $Z$ falls steeply, energies being visited are large.

Step 6 — Substitute, then fold into a log

WHAT. Put Step 5 into Step 3:

$⟨ E ⟩ = \frac{1}{Z} (- \frac{\partial Z}{\partial β}) = - \frac{1}{Z} \frac{\partial Z}{\partial β}$

The stray $\frac{1}{Z}$ from the average is still there — that is the normalization we must not forget.

Now the tidy-up. Recall a fact about logarithms and the chain rule:

$\frac{\partial}{\partial β} ln Z = \frac{1}{Z} \frac{\partial Z}{\partial β}$

$ln$ is the natural logarithm; its derivative rule " $\frac{d}{d x} ln f = \frac{f ^{'}}{f}$ " is precisely the " $\frac{1}{Z} \times$ slope of $Z$ " shape we have.

So the two operations — divide by $Z$ and take the slope — collapse into one operation on $ln Z$ :

WHY the $ln$ . It is not decoration — the log is the exact function whose derivative is the messy combination " $\frac{1}{Z} \partial_{β} Z$ ." Using $ln Z$ means we never write the $1/ Z$ by hand again.

PICTURE. Plot $ln Z$ against $β$ (a downward-bending curve). Its downhill slope, negated, reads off directly as $⟨ E ⟩$ at every temperature.

Step 7 — Edge & limiting cases (never leave a gap)

Every derivation must survive its extremes. Test on the two-level system: boxes at energies $0$ and $ε$ , giving $Z = 1 + e^{- β ε}$ and $⟨ E ⟩ = \frac{ε}{e ^{β ε} + 1}$ .

PICTURE. The $⟨ E ⟩$ -vs- $T$ curve for the two-level system: flat at $0$ when cold, rising, then saturating at $ε /2$ when hot. Both plateaus are the limits above.

The same machine, differentiated twice, gives energy fluctuations and links to Heat capacity and energy fluctuations; the continuous quadratic case reproduces the Equipartition theorem; and the discrete ladder gives the Quantum harmonic oscillator — thermal result. One formula, all of them.

The one-picture summary

Boxes → weights → normalize into $Z$ → one derivative of $ln Z$ (with a minus) → the average energy. The whole page in a single arrow chain.

Recall Feynman retelling — the whole walkthrough in plain words

Picture a row of numbered boxes; each box has an energy tag. A temperature dial decides how often a jumpy marble visits each box — cold means it hugs the low boxes, hot means it wanders everywhere. To turn "how often" into real chances, we add up every box's raw likeliness into one grand total called $Z$ , then each box's chance is its slice of $Z$ . Now the average energy is just each box's energy times its chance, summed — but that sum is fiddly because of the extra energy tag on every term. Here's the magic: if you gently turn the dial and watch $Z$ change, the math automatically multiplies each box by its own energy (that's exactly the factor we were missing), with a minus sign tagging along. Divide by $Z$ and you've divided-and-sloped in one move — which is the same as sloping the logarithm of $Z$ . Flip the sign and there it is: the typical energy of the system, from one nudge of a counter. Test it cold (energy freezes to the bottom), test it hot (energy averages out) — it passes both. That's the entire idea.

Visual walkthrough — Average energy from partition function