What does the circuit look like from here?
Most of the time you do not care about everything inside a circuit. You care about what it does at two terminals. A microcontroller's analogue input sees a sensor circuit through two wires. A phone sees its battery through two contacts. An amplifier sees a guitar pickup, a microphone or an antenna through a cable. The question is always the same: if I connect something here, what voltage and current will it get?
In 1883 the French telegraph engineer Léon Charles Thévenin gave a startling answer. However complicated a circuit of resistors and sources is, as seen from two terminals it behaves exactly like one ideal voltage source in series with one resistor. Two numbers describe it completely. In 1926 Edward Norton at Bell Labs, and independently Hans Ferdinand Mayer at Siemens, gave the mirror image: one current source in parallel with one resistor. (Hermann von Helmholtz had stated a version as early as 1853.)
These equivalents rest on one property, linearity, which also gives us superposition: the response to several sources is the sum of the responses to each one alone. This lesson develops both, then turns to two practical questions: how to get the most power out of a source, and how to measure the Thévenin equivalent of a real one, a battery.
Linearity and superposition
Definition— Linear circuit
A circuit is linear if it is built from resistors, independent sources, and dependent sources whose values are proportional to a voltage or current (the four kinds of Nodal and mesh analysis). Diodes and transistors operated over large ranges are not linear; their small-signal models are.
In the previous lesson every such circuit led to a matrix equation , where holds the unknown node voltages (and voltage-source currents). The matrix is built from the resistors and the dependent sources only. The independent sources appear only in , each in its own entries and in proportion to its value. Everything in this lesson follows from that structure.
Theorem— Superposition
In a linear circuit with a unique solution, every node voltage and branch current is a sum of terms, one per independent source, each proportional to that source's value. The term for source is the response with source acting alone and every other independent source killed: voltage sources replaced by short circuits (0 V) and current sources by open circuits (0 A). Dependent sources are never killed.
Proof
Write , where holds only source 's contribution and is proportional to its value . Then , and each term is proportional to . The term is the solution with every other source's value set to zero, and a voltage source of 0 V is a short circuit (no voltage, any current) while a current source of 0 A is an open circuit (no current, any voltage). Dependent sources live in , not in , so they stay.
∎
Example— Superposition in the two-source network
Take the network of Kirchhoff's laws: V behind kΩ and V behind kΩ, both feeding node b, with kΩ from b to ground. We found V.
alone ( shorted, so goes straight to ground and sits in parallel with ): kΩ, and the divider gives V.
alone ( shorted): kΩ, and V.
Sum: V, as before.
Superposition is a method, but more importantly it is a way of thinking. It says that each source's effect is independent of the others: double and its contribution doubles, whatever is doing. That is why an amplifier can be analysed separately for its bias (DC) and for its signal (AC), and why a noisy circuit's noise sources can be studied one at a time.
Source transformation
Definition— Source transformation
A voltage source in series with a resistor , and a current source in parallel with the same resistor , are equivalent at their two terminals: connected to any load, they give it the same voltage and current.
To see why, write each one's terminal law, the relation between the voltage across the terminals and the current drawn out of them. For the voltage source and series resistor, KVL gives . For the current source and parallel resistor, KCL says the resistor carries , so . With the two laws are identical, and two circuits with the same terminal law cannot be told apart from outside.
Source transformation is a quick tool for hand analysis: in circuit A of the previous lesson, the 10 V source behind 2 kΩ is the same, as far as node 2 can tell, as a 5 mA source in parallel with 2 kΩ, which is exactly the mA that appeared on the right-hand side of the nodal equations. It is also the simplest case of the theorem that follows.
Thévenin and Norton equivalents
Theorem— Thévenin and Norton
Let a linear circuit have two terminals, a and b. If the circuit has a unique solution whatever current is drawn from the terminals, then, as seen from a and b, it is equivalent to
- (Thévenin) an ideal voltage source in series with a resistance , where is the open-circuit voltage between a and b; and, when ,
- (Norton) an ideal current source in parallel with , where is the short-circuit current from a to b.
Proof
Connect any load to the terminals. Whatever the load is, it draws some current out of terminal a, so we can replace it, for the circuit's purposes, by a current source of value . Now apply superposition to the voltage between a and b. With the test source at 0 A (an open circuit), the internal sources produce . With the internal independent sources killed and only the test source acting, is proportional to ; call the constant , since drawing current out of a normally lowers the voltage. So
which is exactly the terminal law of a source in series with . Solving for gives , the terminal law of a current source in parallel with ; setting shows that this current is .
∎
Three ways to find the equivalent
- Open-circuit voltage and short-circuit current. Solve the circuit twice: once with nothing connected (giving ) and once with a and b shorted (giving ). Then . This is also how you measure an equivalent on the bench, when the short is safe.
- Kill the sources and combine resistors. With every independent source killed, the circuit is a network of resistors, and is its resistance between a and b, found by series and parallel combinations. Fast, but valid only when there are no dependent sources.
- Test source. Kill the independent sources, push a test current into terminal a (or apply a test voltage), and compute the resulting voltage ; then . This always works, including with dependent sources.
Example— The two-source network, seen by R3
Remove from the two-source network and look into node b and ground.
Way 1. With b open, the only loop is , , , , carrying mA from towards , so V. With b shorted to ground, each source drives its own resistor: mA. So kΩ.
Way 2. Short both sources: and both run from b to ground, in parallel: kΩ.
Reconnect to the equivalent: V, the answer of the first lesson, now from a one-loop circuit. And if you want for ten different loads, you solve the network once and each load in one line.
Example— An amplifier's output, where killing sources fails
Take the feedback amplifier of the previous lesson ( mV, kΩ, kΩ, mS, kΩ) with the load removed, and find its Thévenin equivalent at the output node 2.
Open circuit. Solving the nodal equations without gives V: an open-circuit gain of .
Way 2, done wrongly. Kill and the dependent source, and the output sees in series with : kΩ.
Way 3, done right. Kill only , keep the dependent source, and inject a test current at node 2. The computation in the next section gives kΩ, thirty times smaller than the wrong answer. The dependent source, driven by the feedback through , fights any change of output voltage, which is what makes the output stiff.
Check. With the 5 kΩ load reconnected, the equivalent predicts a gain of , which is what the full nodal solution gave in the previous lesson.
Checking the equivalents
The first script finds the two-source network's equivalent at node b all three ways, then confirms the defining property: for every load, the full network and the equivalent give the same voltage.
# The two-source network of "Kirchhoff's laws", seen from node b and ground with R3 removed.
R1, R2 = 2e3, 1e3
def v_b(V1, V2, i_in=0.0, G_load=0.0):
"""Voltage of node b by KCL: sources V1 and V2 behind R1 and R2, a current i_in
injected into b from outside, and an optional load conductance to ground."""
return (V1 / R1 + V2 / R2 + i_in) / (1 / R1 + 1 / R2 + G_load)
# Way 1: open-circuit voltage and short-circuit current.
Voc = v_b(12, 5)
Isc = 12 / R1 + 5 / R2 # with b shorted to ground, each source drives its own resistor
print(f"Way 1: Voc = {Voc:.4f} V, Isc = {Isc*1e3:.3f} mA, Rth = Voc/Isc = {Voc/Isc:.2f} Ω")
# Way 2: kill the sources (short them) and combine: R1 in parallel with R2.
print(f"Way 2: R1 || R2 = {R1 * R2 / (R1 + R2):.2f} Ω")
# Way 3: kill the sources and push a 1 mA test current into b.
v_test = v_b(0, 0, i_in=1e-3)
print(f"Way 3: the test current gives {v_test:.4f} V, so Rth = {v_test / 1e-3:.2f} Ω")
# The equivalent must agree with the full network for EVERY load.
Vth, Rth = Voc, Voc / Isc
print(f"\n{'load':>10} {'full network':>13} {'Thévenin':>10}")
for RL in [100, 666.7, 3e3, 10e3, 1e6]:
full = v_b(12, 5, G_load=1 / RL)
equiv = Vth * RL / (Rth + RL)
print(f"{RL:>8g} Ω {full:11.4f} V {equiv:8.4f} V")For the amplifier, a simulator does the work. SPICE's .tf (transfer function) analysis computes the small-signal ratio of an output to an input source and, along the way, the resistance seen at the output: precisely , found with the dependent source alive, as the test-source method requires.
Thevenin equivalent of an amplifier output (load removed)
Vs in 0 DC 10m
Rs in 1 1k
Rpi 1 0 2.5k
Rf 1 2 100k
* voltage-controlled current source: 40 mS times v(1), from node 2 to ground
Gm 2 0 1 0 40m
.control
op
print v(2)
* tf finds the small-signal ratio v(2)/Vs and the resistance seen at node 2
tf v(2) Vs
print transfer_function output_impedance_at_v(2)
.endc
.endngspice prints v(2) = -9.65942e-01, transfer_function = -9.65942e+01 and output_impedance_at_v(2) = 3.405797e+03: an open-circuit voltage of V and Ω, as found by hand. (SPICE says "impedance" because the same analysis works at any frequency; for a circuit of resistors, impedance is resistance.)
Maximum power transfer, and why it is not maximum efficiency
A source with Thévenin equivalent , drives a load . The current is , so the load receives
A very small load gets plenty of current but has almost no voltage across it; a very large load has almost all of but almost no current. In between there is a best value.
Theorem— Maximum power transfer
For fixed and , the power delivered to a resistive load is greatest when , and that greatest power is . At that point the efficiency, the fraction of the source's power that reaches the load, is exactly 50 %.
Proof
Without calculus: , and
with equality exactly when . So , with equality at the match. (With calculus, which this level assumes you have met: , positive below the match and negative above it.) The same current flows through and , so the power divides between them in the ratio of their resistances; the efficiency is , which is at the match.
∎
# A source with Vth = 10 V and Rth = 50 Ω drives a load RL: power to the load, and efficiency.
Vth, Rth = 10.0, 50.0
print(f"{'RL (Ω)':>7} {'P_load (W)':>11} {'efficiency':>11}")
for RL in [5, 10, 25, 50, 100, 200, 500, 5000]:
i = Vth / (Rth + RL)
p_load = i**2 * RL
p_total = Vth * i # everything the ideal source inside delivers
bar = "#" * round(p_load / 0.5 * 30)
print(f"{RL:7d} {p_load:11.4f} {p_load / p_total:10.1%} {bar}")
print(f"maximum possible: Vth^2 / (4 Rth) = {Vth**2 / (4 * Rth):.4f} W")The power peaks at 0.5 W for a 50 Ω load, and the curve is gentle: half or double the matched value (25 Ω or 100 Ω) still delivers 0.4444 W, 89 % of the maximum. The efficiency column tells a different story. It rises steadily with the load, to 90.9 % at 500 Ω and 99.0 % at 5 kΩ, while the power falls.
Which matters depends on what is scarce.
- When the source is weak and its energy is cheap or free, and the signal is what counts, engineers match. A radio receiver's antenna delivers picowatts; wasting half of them inside the antenna is a price worth paying to get the other half. Radio-frequency systems are built around a standard resistance, usually 50 Ω, so that everything can be matched. Transmission lines and Antennas and radio links take this further.
- When the energy is what counts, engineers do the opposite and make as small as possible compared with the load. A power station matched to its customers would burn as much power in its own generators and lines as it delivered, and a phone battery matched to the phone would waste half its charge as heat inside the cell. Power systems run at efficiencies far above 50 %, which means the load is many times the source resistance. The power grid returns to this.
- Audio amplifiers are not matched either: their output resistance is a small fraction of an ohm, against loudspeakers of 4 or 8 Ω, so that the amplifier controls the speaker's voltage tightly.
Measuring a battery's Thévenin equivalent
A battery is not an ideal voltage source. Its chemistry produces an electromotive force (EMF), and its electrolyte, electrodes and contacts add an internal resistance in series. That is its Thévenin equivalent: is the open-circuit voltage, and . You have seen the effect: a car's headlights dim while the starter motor turns, because the starter's huge current drops a large voltage across the battery's internal resistance.
The method is Way 1 with the short circuit replaced by a known load, because shorting a battery is never acceptable. Measure the open-circuit voltage with a voltmeter (its 10 MΩ input draws almost nothing). Connect a load for a few seconds and read the terminal voltage . The load current is , the drop inside is , so
Example— An AA cell
Suppose a fresh AA alkaline cell reads V, and 1.547 V with a 10.0 Ω load. Then mA, the internal drop is 33 mV, and Ω. The cell's equivalent is about 1.58 V in series with 0.2 Ω.
import numpy as np
# Simulated bench data for one AA cell: terminal voltage under several load resistors.
# (A made-up but realistic cell, Voc = 1.580 V and r = 0.22 Ω, read on a meter with 1 mV steps.)
rng = np.random.default_rng(2026)
true_voc, true_r = 1.580, 0.22
loads = np.array([100.0, 47.0, 22.0, 10.0, 4.7]) # ohms
v_ideal = true_voc * loads / (loads + true_r)
v_meas = np.round(v_ideal + rng.normal(0, 0.0007, loads.size), 3) # noise, then 1 mV rounding
i_meas = v_meas / loads # Ohm's law on the known load
print(f"{'load':>7} {'V (V)':>7} {'I (mA)':>8}")
for R, v, i in zip(loads, v_meas, i_meas):
print(f"{R:6.1f}Ω {v:7.3f} {i*1e3:8.2f}")
# Thévenin says V = Voc - r I: a straight line. Fit it by least squares.
slope, intercept = np.polyfit(i_meas, v_meas, 1)
print(f"\nfit: Voc = {intercept:.4f} V, r = {-slope:.3f} Ω")
# The quick two-point estimate, from the lightest and heaviest loads only.
r_two = (v_meas[0] - v_meas[-1]) / (i_meas[-1] - i_meas[0])
print(f"two-point estimate: r = {r_two:.3f} Ω")The fit recovers V and Ω from data generated with 1.580 V and 0.22 Ω. A real cell adds complications that the linear model hides, and a careful engineer states them:
- It is a linearisation. A battery is a chemical system, not a resistor. Its voltage sags further the longer a load is applied, and recovers slowly afterwards, so the "resistance" depends on how long you wait before reading. Read quickly and consistently.
- It drifts. Internal resistance rises as a cell discharges, as it ages, and sharply in the cold, which is why batteries struggle on winter mornings.
- Industry measures it in standard ways. Battery makers commonly quote an AC internal resistance measured with a small 1 kHz current, and a DC resistance from a timed current pulse; the two differ. Battery management systems in electric vehicles and laptops estimate internal resistance continuously, because its growth is one of the best signs of a cell's ageing. Research labs go further with impedance spectroscopy, measuring the cell's response over a range of frequencies. Renewables and storage picks this up.
Note
Doing this safely. Use one or two AA or AAA alkaline or NiMH cells, a load of at least 4.7 Ω, and a resistor rated for the power: W, so use a 1 W part, which may still get warm. Connect the load for a few seconds at a time. Never short-circuit a cell to measure its current, and never try this with lithium-ion packs, car batteries or other large batteries: their internal resistance is so low that a short or a small load can deliver enough current to start a fire or cause burns.
The professional view
Thévenin and Norton equivalents are how engineers talk about interfaces. A datasheet rarely describes what is inside a part; it gives output impedance and input impedance, the Thévenin resistance looking out of an output and the resistance looking into an input, and from those two numbers you know how parts will load each other. The rule of thumb from the divider table in Kirchhoff's laws becomes general: a stage should drive inputs whose resistance is much larger than its own output resistance.
- Data converters. An analogue-to-digital converter samples its input onto a small capacitor; the source's Thévenin resistance limits how fast that capacitor can charge, so converter datasheets specify a maximum source resistance or recommend a driver amplifier. Sensors and data conversion works through an example.
- Transistor models. The small-signal models of transistors are Norton equivalents: a dependent current source with a parallel output resistance. Transistor amplifiers uses them constantly.
- Power grids. Seen from any point, the grid has a Thévenin equivalent too, and its short-circuit current is the Norton current. Grid operators use this "strength" to decide how much solar and wind generation, or how large a motor, can be connected there.
- Solar panels. A panel is not linear, but the maximum-power idea survives: its best operating point moves with sunlight and temperature, and the inverter's maximum power point tracking controller keeps adjusting the load the panel sees to stay on it.
Common misconceptions
Misconception
"To kill a current source, short it." Killing a source sets its value to zero. A current source of 0 A carries no current, whatever the voltage: an open circuit. A voltage source of 0 V has no voltage, whatever the current: a short. Mixing them up changes the circuit.
Misconception
"Kill every source, including the dependent ones, to find ." Dependent sources are part of the circuit's structure, not inputs to it. Killing them in the amplifier example gave 100.7 kΩ instead of the correct 3.406 kΩ.
Misconception
"The Thévenin equivalent tells you what happens inside the circuit." It is equivalent only at the two terminals. The power dissipated in is generally not the power dissipated inside the original circuit: with no load at all, the Thévenin equivalent dissipates nothing, while the two-source network still burns mW in and .
Misconception
"Matching the load gives the best efficiency." Matching gives the most power from a fixed source, at exactly 50 % efficiency. Efficiency keeps rising as the load resistance grows.
Misconception
"A Thévenin resistance is always positive." For a network of resistors, yes. With dependent sources, can be zero, negative or undefined. A negative resistance means the circuit can supply power to a resistor connected to it, and it is the basis of some oscillators.
Exercises
1. Superposition, Thévenin and Norton
Warm-upHumanYou are using superposition, and need the response to one source with all the others killed. A circuit contains an independent voltage source, an independent current source, and a current source whose value is , where is a current elsewhere in the circuit. Which replacement is correct?
2. Superposition, Thévenin and Norton
CoreHumanA 9 V source (negative terminal on b) connects through kΩ to terminal a. A resistor kΩ joins a to b, and a 1 mA current source pushes current from b into a. What is the Thévenin voltage between a and b, in volts?
3. Superposition, Thévenin and Norton
CoreHumanFor the same circuit (9 V through 3 kΩ to a, 6 kΩ from a to b, 1 mA pushed from b into a), what is the Norton current , the short-circuit current from a to b, in mA? Find it directly by shorting a to b, then check it against .
4. Superposition, Thévenin and Norton
StretchHumanA 6 V source drives terminal a through kΩ, and kΩ joins a to b (b is the source's negative terminal and the reference). A voltage-controlled current source pushes from b into a, with mS, where is the voltage of a. What is between a and b, in ohms?
5. Superposition, Thévenin and Norton
Warm-upHumanA source has V and Ω. What is the greatest power, in milliwatts, that it can deliver to a resistive load?
6. Superposition, Thévenin and Norton
CoreComputationalWrite fit_cell(loads, volts) that estimates a cell's Thévenin equivalent from measurements. loads lists load resistances in ohms and volts the terminal voltage measured with each. Compute each load current, fit the straight line by least squares, and return (voc, r).
7. Superposition, Thévenin and Norton
StretchComputationalA circuit's nodal equations are , with the load removed. Write thevenin_at(G, i_s, k) returning (Vth, Rth) seen between node k (a 0-based index into ) and ground.
Use the lesson's ideas: is the open-circuit voltage at node , and is the voltage at node when the independent sources are killed () and a test current of 1 A is injected into node . Your function must work for non-symmetric (dependent sources).
Further reading and resources
- MIT OpenCourseWare 6.002: Circuits and ElectronicsFree lectures and notes; its early weeks treat superposition and the Thévenin and Norton methods.
- ngspice documentationThe manual describes the .tf analysis used below, which reports a circuit's gain and its input and output resistances.