An imaginary voltmeter displays 2.4 volts. Its display is exactly right for the voltage between its terminals, yet the same circuit had 4 volts between those points before the meter was present. There is no contradiction. The instrument became part of the circuit and changed the quantity being observed. Checking the display against the voltage that exists now answers a different question from asking how much the measurement arrangement changed the earlier voltage.

OpenStax explains that a voltmeter is placed across the two points whose potential difference is of interest, and that its resistance matters to the disturbance it introduces [1]. NI and NJIT likewise describe the effect of a finite input resistance [3][4]. We will follow one invented, steady direct-current circuit on paper, with exact resistances and an ideal display. The example is a calculation, not an assembly activity or a test of a commercial meter. Its deliberately conspicuous loading makes three quantities easy to separate: the voltage before the meter, the voltage after it, and the indicated value.

A voltage measurement has an intended measurand, a measurement configuration and an indication. In the original model here, the unconnected output is 4 V, the output after adding the meter input is 2.4 V, and an ideal indication reports that loaded terminal voltage exactly. Indication error relative to the loaded state is therefore zero by construction, while the change relative to the unloaded state is nonzero. The reference state must accompany any statement that the reading is correct.

OpenStax places a voltmeter in parallel with the potential difference being measured [1]. NI and NJIT discuss its finite input resistance and the resulting influence on the circuit [3][4]. Our construction uses an ideal 12 V source, positive linear resistors, steady DC conditions and an exact indicator with a specified parallel input resistance. There is no physical meter, simulation package, calibration or uncertainty estimate. Within those boundaries, elementary network equations let us separate loading, indication and inference about the original circuit without assigning a product an unsupported accuracy claim.

Start with a 12 V source and two resistors in one series path. The upper resistor is 20 kilo-ohms and the lower resistor is 10 kilo-ohms. A kilo-ohm is one thousand ohms. The output is the junction between the resistors, measured relative to the lower end of the 10 kilo-ohm resistor. Treat that lower end as zero volts. No other path leaves the junction in this first state. Naming both points matters because voltage describes a difference between points, not a quantity belonging to a loose wire alone.

Series resistances add, and the same current passes through both components in this arrangement [2]. The total is 30 kilo-ohms, so the current is 12 divided by 30,000, or 0.0004 ampere. That is 0.4 milliampere. Across the lower 10,000 ohms, this current produces 4 V. The upper resistor accounts for the remaining 8 V, giving the source's full 12 V together. These are assigned ideal values. They establish the undisturbed reference against which we will compare a second, explicitly changed circuit.

Let the source be Vs = 12 V, the upper resistance Ru = 20 kΩ and the lower resistance Rd = 10 kΩ. Define the lower endpoint as the reference node and V as the junction potential relative to it. With no external output branch, the two resistors are in series. Ohm's law and the series-current relation give I0 = Vs/(Ru+Rd) = 0.4 mA and V0 = I0 Rd = 4 V [2]. Here V0 denotes the unloaded output of this particular divider, not the source voltage.

The corresponding upper drop is I0 Ru = 8 V, so the two drops sum to 12 V. This baseline uses an ideal source with no separately assigned source resistance; Ru is a real element of the hypothetical divider rather than a hidden instrument parameter. The output terminal is open only with respect to an added external branch: current still flows through the original divider. Keeping that distinction explicit prevents “unloaded” from being misread as zero current everywhere in the network.

Now give the imaginary voltmeter an input resistance of 10 kilo-ohms. Its ideal display observes the voltage without introducing any additional effect beyond that assigned resistance. When represented across the output and the reference point, the meter input shares both endpoints with the lower resistor. It is therefore a parallel branch, not another resistor in the original series chain [1][2]. This 10 kilo-ohm value is chosen to make the arithmetic visible. It is not a claim about the usual resistance of a digital multimeter.

The original lower resistor has not vanished, and its resistance has not changed. Instead, current arriving at the junction can now take either of two equal 10 kilo-ohm paths. Two equal parallel resistances have an equivalent resistance half as large: 5 kilo-ohms. The source consequently sees the upper 20 kilo-ohms followed by this 5 kilo-ohm equivalent. The network has changed even though the source and original resistors have retained their values. We must solve this new network before deciding what voltage the meter will display.

Model the meter as an ideal voltage indicator with a finite resistance Rm connected between the output node and reference. Set Rm = 10 kΩ for the first case. Since Rd and Rm share both nodes, their conductances add: 1/Rp = 1/Rd + 1/Rm. Thus Rp = Rd Rm/(Rd+Rm) = 5 kΩ [2]. The ideal indicator adds no current beyond the stated Rm. NJIT uses a real-instrument resistance alongside an ideal instrument to explain this distinction [4].

The upper resistor is in series with the parallel combination, so the total resistance seen by the ideal source becomes Ru+Rp = 25 kΩ. It would be incorrect to place Rm in series with Rd or to replace Rd by Rm while discarding the existing lower branch. The meter changes connectivity at the output. The assigned resistance is an illustrative input parameter, not a commercial specification, and the calculation intentionally excludes capacitance, leakage variation and range-dependent behavior. Those effects would require a different input model rather than a correction to this parallel-resistance formula.

Instead of recalculating the lower parallel pair for every meter, we can summarize how this divider responds at its output. Two numbers are needed: the unloaded 4 V and an effective output resistance. For this particular ideal divider, that resistance is the upper resistance times the lower resistance, divided by their sum. It is 20 × 10 ÷ 30, or 6⅔ kilo-ohms. The next section uses that quantity to compare different meter inputs without changing the original divider.

The resulting voltage is the unloaded voltage multiplied by the meter resistance divided by the sum of the meter resistance and the effective output resistance. With a 10 kilo-ohm input, that is 4 × 10 ÷ (10 + 6⅔) = 2.4 V. This is an algebraic summary of the terminal relationship, derived from the same current balance used above [2]. It does not say the original 12 V source has become a 4 V source inside the circuit. The simplified description preserves what the added load sees; it does not reproduce every internal current.

Write the output-node balance as (Vs−V)/Ru = V/Rd + V/Rm. Rearranging gives V = Vs Rd Rm/(Ru Rd + Ru Rm + Rd Rm). Define V0 = Vs Rd/(Ru+Rd) and Ro = Ru Rd/(Ru+Rd). Substituting these definitions yields V = V0 Rm/(Ro+Rm). This source-equivalent form follows directly from the specified resistor relations [2]; NJIT separately identifies output resistance as a source property relevant to loading [4]. For our divider, Ro = 20/3 kΩ.

The equivalent describes the output terminal voltage-current relation for this linear model. It does not identify Ro with the lower 10 kΩ resistor alone, nor does it preserve the divider's internal source-current accounting. For example, the equivalent source's series current is the external load current, while the original 12 V source also supplies the lower divider branch. Thus terminal equivalence cannot be used to substitute one network's internal source power for the other's. The reduction is useful precisely when its preserved boundary is stated.

Suppose our paper requirement is to keep the loading reduction at or below one percent of the original voltage. The denominator is the unloaded 4 V, so the allowed drop is 0.04 V and the connected output must be at least 3.96 V. Solving the model gives a minimum meter resistance of 660 kilo-ohms. At exactly that value, the voltage is 3.96 V and the drop is exactly one percent. This condition covers only the change caused by the assigned input resistance.

The ratio behind the result is instructive. The meter resistance must be at least 99 times the divider's effective output resistance of 6⅔ kilo-ohms. Being ten times that output resistance is not enough for a one-percent limit: it leaves a drop of one part in eleven, about 9.09 percent. The comparison is with the effective output resistance, not automatically with either physical resistor. An overall measurement uncertainty would also need other information that this model has not supplied. Our exact display assumption does not turn the loading calculation into a complete instrument specification.

Define the relative loading drop e = (V0−V)/V0 for V0>0. Substituting the terminal relation gives e = Ro/(Rm+Ro). For a required bound 0<ε<1, e≤ε is equivalent to Rm≥Ro(1/ε−1). With ε = 0.01 and Ro = 20/3 kΩ, the threshold is 99Ro = 660 kΩ. The boundary voltage is V0(1−ε) = 3.96 V. All percentages therefore have an explicit unloaded-voltage denominator.

At Rm = 10Ro, e = 1/11 ≈ 9.0909 percent and V = 40/11 V ≈ 3.63636 V. The familiar-looking ten-to-one ratio cannot be substituted for the derived 99-to-one requirement. This inequality is exact only for the declared positive, purely resistive DC model. It budgets loading, not source variation, resistance tolerances, offsets, noise, gain error or a real meter's calibration uncertainty. The distinction is substantive: a model can satisfy its loading inequality while additional unmodeled error terms remain unknown rather than zero.

If the input resistance and the effective output resistance are both known within our model, the loaded reading can be used to calculate the unloaded voltage. Multiply the reading by one plus the output resistance divided by the meter resistance. For the original 10 kilo-ohm input and 2.4 V reading, the factor is one plus 6⅔ divided by 10. The result is 4 V. This is a conditional reconstruction from known parameters, not something the displayed number reveals on its own.

To see the missing information, consider another hypothetical source with an unloaded voltage of 3 V and an output resistance of 2.5 kilo-ohms. The same 10 kilo-ohm meter would again show 2.4 V: 3 × 10 ÷ 12.5. Thus one identical reading is compatible with at least two different unloaded voltages. A correction formula cannot decide which source description applies unless that description is established separately. The issue is not insufficient decimal places. It is that the observation contains less information than the pair of unknown source quantities.

Rearranging the terminal relation gives V0 = V(1+Ro/Rm). This recovers V0 only conditionally on the model and known Ro and Rm. Case A uses V0 = 4 V, Ro = 20/3 kΩ and Rm = 10 kΩ, producing V = 2.4 V. Case B uses V0 = 3 V, Ro = 2.5 kΩ and the same Rm, also producing V = 3×10/12.5 = 2.4 V. The two cases are an original algebraic counterexample, not observations from two physical sources.

One loaded voltage with known Rm therefore does not uniquely identify both V0 and Ro. The terminal equation constrains their relationship but leaves more than one compatible pair. This is separate from display accuracy: the indication can be exact in both cases while the inferred unloaded voltage remains ambiguous. Additional justified information would be needed to identify the relevant source model. No two-load procedure is carried out here, and no correction is claimed for a nonlinear or time-varying source to which the derived linear relation may not apply.

A useful interpretation records which two points define the voltage, what circuit was present, and what the instrument input adds between those points. It also states whether the desired answer concerns the connected state or the earlier unloaded state. In our example the first answer is 2.4 V and the second is 4 V. Both are correct for their named configurations. Calling either number simply the true voltage would remove the condition that explains why they differ.

The paper model treats the input as a fixed resistance and waits for no changing signal. NJIT notes that instrument impedance can involve capacitance or inductance as well as resistance, while NI discusses different input configurations [3][4]. Those are reasons to retain the model boundary when interpreting real specifications. Our result establishes neither dynamic behavior nor a universally best measurement range. It establishes a narrower, useful point: an accurate indication of the connected circuit can coexist with a substantial change from the circuit you meant to observe. The input's effect and the display's accuracy need separate evidence.

An inspectable loading calculation specifies the reference nodes, unloaded network, source assumptions, instrument input model and intended measurand. It distinguishes V0, the loaded V and the indication. For the original 10 kΩ input, the derived values are V0 = 4 V, V = 2.4 V and indication = 2.4 V. The loading fraction is 40 percent relative to V0, while indication error relative to V is stipulated zero. These statements remain compatible because their reference quantities differ.

The resistive DC reduction is not a general instrument-impedance model. NJIT explicitly allows capacitive and inductive components, and NI describes finite and selectable input behavior [3][4]. Frequency dependence, settling, noise, component uncertainty, protection and actual range specifications are not resolved by the present numbers. Nor does increasing Rm in this fixed model establish the best choice for every task. What survives the example is the reasoning sequence: define the desired state, include the observation branch, solve the changed network, and state which comparison the resulting voltage supports.