
Physics I course sequence · Interactive lesson
Derivatives & Integrals in Physical Models
You will spend this course turning rates into amounts and amounts into rates. Here you practise both moves on graphs, on formulas and on tables of readings, and meet the one extra fact every model needs before it can predict anything: where it started.
UPI-01.07 · Lesson 7 of 9 in this unit. Position is not a completion score.Build the model: derivatives as rates
A derivative gives the instantaneous rate of change—the slope of the tangent.
v(t) = dx/dt
a(t) = dv/dt = d²x/dt²
A rate at an instant is a limit of chord slopes
A rate answers “how fast is Q changing right now?”, and “right now” needs a limit. Take the chord slope ΔQ/Δt across an interval, then shrink the interval: the chord slopes settle on a single number, dQ/dt, the slope of the tangent at that instant. A leaking bucket holding m(t) = (12.0 kg)exp(−t/τ), with τ = 40 s, shows the limit arriving. From t = 20 s, chords 10 s, 1.0 s and 0.10 s wide give −0.161, −0.180 and −0.1817 kg s⁻¹, closing on the exact slope −(12.0 kg ÷ 40 s)exp(−0.5) = −0.182 kg s⁻¹. The widest chord is about 12% too shallow, because the leak slows across those 10 s. Two checks come free with every derivative. Its unit is the unit of Q divided by the unit of t, and its sign says whether Q is rising or falling at that instant, independent of how large Q is.
The full physical model
Physical models are usually written as rates. Newton's second law fixes dv/dt, a leak fixes dm/dt, a motor fixes dE/dt, because what nature settles locally is how a quantity changes in the next instant, not what its value is. The derivative dQ/dt is that local statement: the limit of ΔQ/Δt as the interval shrinks, drawn as the slope of the Q–t graph and carrying the unit of Q per unit of time. The definite integral runs the argument backwards, adding r dt over thin slices until the net change appears as the signed area under the rate graph. The two operations undo each other with one gap: a rate fixes the shape of Q(t) but not its level, so an absolute prediction needs one measured value, the initial condition, to pin that level down. With data instead of formulas, the limits become finite differences and sums whose truncation errors typically shrink as h or h² for sufficiently smooth functions. At an idealised jump in a finite rate, the accumulated quantity may have a corner: its derivative need not exist there, but piecewise integration still works. And measured values carry noise, which a difference quotient divides by h: shrink the step far enough and the derivative gets worse, not better.
Definition & motor-power example
A derivative dQ/dt is how fast a quantity changes at one instant, the slope of its graph; a definite integral of that rate over an interval is the net change it builds up, the signed area under the rate graph. A motor feeds a flywheel power P(t) = (40 W s⁻¹)t. At t = 5.0 s the stored energy is rising at dE/dt = 200 W, and over those 5.0 s it has gained ΔE = ∫ (40 W s⁻¹)t dt, taken from t = 0 to 5.0 s, which is (20 W s⁻¹)(5.0 s)² = 500 J.
Integrals as signed accumulation
The area under the rate curve gives a change, not the starting amount.
Δx = ∫ v(t) dt, from t₁ to t₂
x(t) = x₀ + ∫₀ᵗ v(t′) dt′
An integral adds rate × time, and the area is signed
Run the argument backwards. Across a thin slice dt the quantity changes by r dt, so across a whole interval the net change is the sum of every slice, which in the limit of thin slices is the definite integral ΔQ = ∫ r dt. On the rate graph each slice is a strip of height r and width dt, so ΔQ is the area under the curve, carrying the unit of r × t: L s⁻¹ × s = L, W × s = J, N × s = kg m s⁻¹. The area is signed. A net force F(t) = (6.0 N) − (1.2 N s⁻¹)t on a trolley delivers +15 N s of momentum in the first 5.0 s and −15 N s in the next 5.0 s, so over 10 s the integral is zero: the trolley leaves with the momentum it arrived with, though it was pushed hard both ways. ∫F dt answers “how much did p change?” The total push, ∫|F| dt = 30 N s, is a different question with a different integral.
Initial conditions
Integration introduces a constant. One measured value fixes the level.
Q(t) = ∫ r(t) dt + C
Q(t₀) = Q₀
Constant acceleration
Integrate a twice, using v(0) = v₀ and x(0) = x₀:
v(t) = v₀ + at
x(t) = x₀ + v₀t + ½at²
Power is an energy rate: P = dE/dt. Net force is a momentum rate: F = dp/dt. In each case, the rate’s integral gives a change in the stored quantity.
The rate fixes the shape, a measurement fixes the level
Differentiation throws information away. Q(t) and Q(t) + C have the same slope at every instant, so a rate on its own pins Q down only up to a constant. One measured value restores it. Write Q(t) = Q₀ + ∫ r(t′) dt′ from t₀ to t, with t′ a dummy variable and Q₀ = Q(t₀): differentiating returns r(t), and setting t = t₀ returns Q₀, so both halves of the model hold. The measurement need not be at t = 0. A tank filling at r(t) = (0.60 L s⁻¹) − (0.020 L s⁻²)t has V(t) = C + 0.60t − 0.010t², and a reading of 30.0 L at t = 10 s gives 30.0 = C + 6.0 − 1.0, so C = 25.0 L. Each integration costs one condition. Newton's second law fixes acceleration, so position needs two: a = dv/dt leaves one constant and v = dx/dt leaves another, which is why a launch needs both a starting position and a starting velocity.
Reading graphs: slope and area
Distinguish the largest amount from the fastest increase.
- r > 0: Q is increasing.
- r < 0: Q is decreasing.
- Positive-to-negative rate crossing: a local maximum of Q.
- r increasing: Q curves upward.
- A rate that only touches zero need not create a maximum.
Average rate = ΔQ / Δt
For a maximum over a closed interval, compare both endpoints and every interior stationary point. At a discontinuity, also check the relevant one-sided behaviour.
Read the slope and the area straight off the graphs
The pairing works on graphs before any algebra. The height of the rate graph at a time is the slope of the Q graph at that time, and the area under the rate graph between two times is the rise of the Q graph between them. Three readings follow: Q has a peak or a trough wherever r crosses zero, Q rises most steeply wherever r is largest, and Q curves upward wherever r is increasing. Grain pours into a hopper at a rate that climbs steadily from zero to 2.0 kg s⁻¹ at t = 4.0 s, then falls steadily back to zero at 10 s. The mass never decreases, because the rate is never negative. It gains the triangle's area, ½ × 10 s × 2.0 kg s⁻¹ = 10 kg. It grows fastest at 4.0 s, yet only 4.0 kg has arrived by then, and the halfway mass of 5.0 kg arrives at 10 − √30 = 4.5 s. The fastest instant and the halfway instant are different, and only the area locates the second.
Finite differences from data
Use changes over finite intervals to estimate a derivative.
Forward difference
f′(t) ≈ [f(t + h) − f(t)] / h
Central difference
f′(t) ≈ [f(t + h) − f(t − h)] / (2h)
For sufficiently smooth, noise-free data, their leading truncation errors are O(h) and O(h²). A smaller step does not always improve a noisy estimate.
From a table, a derivative becomes a finite difference
Measured data arrive at a spacing h, so the limit has to stop at a finite step, and where the chord sits decides how wrong it is. The forward difference [Q(t + h) − Q(t)]/h starts its chord at t. Taylor's theorem gives Q(t + h) = Q + hQ′ + ½h²Q″ + …, so its error is about (h/2)Q″, proportional to h. The central difference [Q(t + h) − Q(t − h)]/(2h) straddles t, the h² terms cancel, and its error is about (h²/6)Q‴. For the bucket at t = 20 s, readings 4.0 s apart give a forward estimate of −0.1732 kg s⁻¹, off by 0.0088, and a central estimate of −0.1823 kg s⁻¹, off by 0.0003. At 2.0 s spacing the forward error halves to 0.0045 and the central error quarters to 0.00008. Centring the chord makes the estimate about thirty times more accurate at 4.0 s spacing, and about sixty times at 2.0 s.
Numerical integration
Approximate accumulation by joining neighbouring readings.
∫ₐᵇ f(t) dt ≈ h[½f₀ + f₁ + … + fₙ₋₁ + ½fₙ]
Exact for the piecewise-linear interpolant; O(h²) convergence needs appropriate smoothness.
Sums from a table, and where the calculus model breaks
An integral from readings uses the same idea in reverse. A left sum h(r₀ + r₁ + … + rₙ₋₁) treats the rate as constant across each step, with error proportional to h; the trapezoid rule joins neighbouring readings with straight lines, with error proportional to h². The bucket loses 12.0(1 − e⁻¹) = 7.585 kg in its first 40 s. Four trapezoids estimate 7.625 kg and eight give 7.595 kg, so halving h cut the error from 0.040 kg to 0.010 kg, while the left sums, 8.573 kg and 8.069 kg, only halve theirs. Two limits need respect. Reading noise is divided by h in a difference but multiplied by h in a sum, so differentiating data can amplify rapidly varying noise while integration can reduce its effect; a constant rate bias still accumulates with time, and below some spacing a smaller h makes a derivative worse. A finite jump in the rate can leave a corner in the accumulated quantity. Its derivative may be undefined at that instant, while its integral is still finite.
Smoothness, noise & limits
Numerical precision does not guarantee a valid physical model.
Differentiation amplifies rapid variation
A 0.05 m oscillatory reading error with angular frequency 20 s⁻¹ produces a 1 m/s derivative-error amplitude.
- For reading error bounded by ±ε, a forward or backward difference has noise error at most 2ε/h.
- The corresponding central-difference bound is ε/h.
- Integration can reduce rapidly alternating noise, but a constant rate bias b builds an error bΔt.
- Integrate piecewise across a finite rate jump; do not require a tangent at the jump.
Check the units, sampling interval, instrument behaviour and physical limits before interpreting a result.
Interactive model: water tank
Change the net flow, then compare its signed area with the change in volume.
r(t) = r₀ + A sin(2πt/T)
ΔV = r₀t + AT/(2π) [1 − cos(2πt/T)]
V(t) = V₀ + ΔV
Compare exact accumulation with trapezoids
We sample the chosen rate evenly from 0 to 30.00 min, stopping at the model limit when necessary.
Trapezoids: 1,070.079 L. Exact integral: 1,071.595 L. Signed error: -1.516 L.
More intervals usually reduce discretisation error for a smooth rate. This model has no measurement noise; real readings need an uncertainty analysis.
Idealised net flow, no capacity limit or evaporation. The maximum compares endpoints and all stationary times over the stated valid interval, not just the time selected. Results use unrounded calculations.
Explore the original seconds-based tank model
This model uses litres and seconds, with a rate decreasing by 0.05 L/s each second. Its settings and results are independent of the minutes-based lab above.
Drag t past the moment the rate line crosses zero: the shading starts to grow below the axis, V passes its peak and the tangent tips from rising to falling. Then move only V₀. The whole volume curve lifts, yet every slope and every shaded area stays exactly as it was.
RATE r = SLOPE OF V0.40 L s⁻¹
SHADED AREA = ΔV4.80 L
VOLUME V = V₀ + ΔV10.80 L
FULLEST WHERE r = 0, AT t16.0 s
Live interpretationRATE r = SLOPE OF V: 0.40 L s⁻¹. SHADED AREA = ΔV: 4.80 L. VOLUME V = V₀ + ΔV: 10.80 L. FULLEST WHERE r = 0, AT t: 16.0 s
Common traps & quick check
A large rate means a steep slope—not necessarily a large amount.
Explain the rate-versus-amount misconception
The commonest error is reading the height of a rate graph as the height of the quantity: the bath must be fullest when the tap runs hardest, the flywheel must hold most energy when the motor pushes hardest. A rate graph carries no levels at all. Its height at an instant is the slope of Q there, so a tall rate means Q is climbing steeply, not that Q is large. Q keeps climbing for as long as r stays positive, however small r has become, so Q peaks when r falls through zero, often long after r's own maximum. Fill a bath while slowly closing the tap and the water reaches its highest level only when the tap shuts, with the inflow at zero, not at the start when it was fastest. A level comes from an area plus a starting value; a single height on the rate graph supplies neither.
- Keep signed area separate from total absolute area.
- Include the initial condition.
- Do not confuse average and instantaneous rates.
- Check endpoints as well as zero crossings for global maxima.
Quick check
Five questions · practice for this visit, not saved course completion
Review or reset practice
Refreshing clears these answers. Existing saved progress is not changed.
Worked examples
Apply integration, data-based estimates and a conservative error budget.
Easy · Rain-gauge accumulation
A rain gauge already holds 4.0 mm of water when a storm starts. For the next 6.0 h rain falls at R(t) = (6.0 mm h⁻²)t − (1.0 mm h⁻³)t². Find the depth D(t), the depth when the storm ends, and compare the storm's average rainfall rate with its peak rate.
Show worked solution →
- The model is a rate plus a starting value: dD/dt = R(t) with D(0) = 4.0 mm. Work in millimetres and hours throughout.
- Integrate term by term: D(t) = C + 3.0t² − t³/3. The condition D(0) = 4.0 mm gives C = 4.0 mm, so D(t) = 4.0 + 3.0t² − t³/3. Check by differentiating: dD/dt = 6.0t − t², which is R(t).
- At t = 6.0 h: D = 4.0 + 3.0 × 36 − 216/3 = 4.0 + 108 − 72 = 40 mm, so the storm itself added 36 mm.
- Average rate: ΔD/Δt = 36 mm ÷ 6.0 h = 6.0 mm h⁻¹. The peak is where dR/dt = 6.0 − 2.0t = 0, at t = 3.0 h, giving R = 18 − 9.0 = 9.0 mm h⁻¹.
D(t) = 4.0 + 3.0t² − t³/3 mm with t in hours; D = 40 mm at 6.0 h, 36 mm of it from the storm. Average rate 6.0 mm h⁻¹, against a peak of 9.0 mm h⁻¹ at 3.0 h.
Medium · Flywheel energy from power data
A bus's braking-energy flywheel logs its net power every 10 s: 42, 30, 12, −6 and −20 kW at t = 0, 10, 20, 30 and 40 s. It stores 1.20 MJ at t = 0. Estimate the energy stored at 40 s, explain why a left-endpoint sum would mislead here, and estimate when the stored energy peaked and its value then.
Show worked solution →
- Power is the rate of change of stored energy, so ΔE = ∫P dt, the signed area under the power readings. With h = 10 s, kW × s gives kJ.
- Trapezoid rule: ΔE ≈ 10 s × [½(42) + 30 + 12 + (−6) + ½(−20)] kW = 10 × 47 = 470 kJ. So E(40 s) ≈ 1.20 MJ + 0.47 MJ = 1.67 MJ.
- A left sum, 10 × (42 + 30 + 12 − 6) = 780 kJ, prices every step at its opening reading. Power falls throughout, so each opening reading is the largest in its step and the sum overshoots; the right sum, 160 kJ, undershoots. The trapezoid is their average: (780 + 160)/2 = 470 kJ.
- E peaks where P crosses zero, between 20 s (+12 kW) and 30 s (−6 kW). Interpolating linearly: t = 20 + 10 × 12/18 = 26.7 s.
- Energy gained up to the peak: the trapezoids to 20 s, 10 × (21 + 30 + 6) = 570 kJ, plus the triangle from 20 s to 26.7 s, ½ × 6.67 s × 12 kW = 40 kJ, so 610 kJ, and E ≈ 1.20 + 0.61 = 1.81 MJ. Check: the remaining area, −140 kJ, brings it back to 1.67 MJ.
E(40 s) ≈ 1.67 MJ. The left sum, 780 kJ, exceeds the trapezoid estimate by 310 kJ because power falls throughout. The store peaks near 27 s, at about 1.81 MJ.
Hard · Noisy draining bottle
A bottle drains onto a balance that reads to the nearest 0.1 g, and its mass follows m(t) = m₀(1 − t/T)² with m₀ = 500 g and T = 100 s. A controller must estimate the flow rate at the newest reading, so it uses the backward difference [m(t) − m(t − h)]/h. Find the sampling interval h that minimises the conservative worst-case error bound at t = 20 s, and compare it with h = 0.1 s and h = 10 s.
Show worked solution →
- Exact values at t = 20 s: m = 500 × 0.80² = 320 g, dm/dt = −2m₀(1 − t/T)/T = −8.0 g s⁻¹, and d²m/dt² = 2m₀/T² = 0.10 g s⁻², constant because m is quadratic in t.
- Curvature error: for a quadratic, m(t − h) = m − hm′ + ½h²m″ exactly, so [m(t) − m(t − h)]/h = m′ − (h/2)m″. The truncation error is (h/2)(0.10 g s⁻²) = (0.050 g s⁻²)h, growing with h.
- Rounding error: each reading lies within ±0.05 g of the true mass, so the difference of two readings is off by at most 0.1 g, and dividing by h gives (0.1 g)/h, growing as h shrinks.
- Conservative error bound: E(h) = 0.050h + 0.1/h, in g s⁻¹ with h in s. Setting dE/dh = 0.050 − 0.1/h² = 0 gives h² = 2.0 s², so h = 1.4 s. There the two errors are equal at 0.071 g s⁻¹ each, and E = 0.14 g s⁻¹.
- Compare: h = 0.1 s gives E = 0.005 + 1.0 ≈ 1.0 g s⁻¹, 13% of the 8.0 g s⁻¹ flow and almost all rounding; h = 10 s gives E = 0.50 + 0.01 = 0.51 g s⁻¹, 6.4% and almost all curvature. The best interval sits between them, at 1.8%.
h ≈ 1.4 s, where the conservative worst-case error bound is 0.14 g s⁻¹, 1.8% of the −8.0 g s⁻¹ flow. Sampling every 0.1 s allows up to 1.0 g s⁻¹ (13%); every 10 s, up to 0.51 g s⁻¹ (6.4%).
The result minimises the conservative sum of truncation and two-reading rounding-error bounds. It does not identify the exact error for a particular sequence of rounded readings.
Summary & full formula reference
A derivative gives an instantaneous rate. An integral gives signed accumulation. Add the initial value to recover the quantity, and test numerical estimates against measurement uncertainty and physical limits.
Open all six formulae and their conditions
Convergence statements below describe smooth-function truncation error, not total measurement error. If each reading has error bounded by ±ε, forward/backward noise error is at most 2ε/h and central noise error at most ε/h.
The derivative is a limit of chord slopes
dQ/dt = lim(Δt → 0) [Q(t + Δt) − Q(t)] / Δt
The slope of the tangent to the Q–t graph. Its sign says whether Q is rising or falling at that instant, whatever Q's size.
unit of Q per unit of t: m s⁻¹, L s⁻¹, kg s⁻¹, J s⁻¹ = WNet change is the integral of the rate
Q(t₂) − Q(t₁) = ∫ r dt, taken from t₁ to t₂, with r = dQ/dt
The signed area under the rate graph. Area below the axis is a loss; ∫|r| dt, which counts losses as gains, answers a different question.
unit of r × unit of t = unit of Q: L s⁻¹ × s = L, N × s = kg m s⁻¹Initial-value form
Q(t) = Q₀ + ∫ r(t′) dt′, taken from t₀ to t
The rate fixes the shape of Q(t); one measured value fixes its level. Differentiate and r(t) returns; set t = t₀ and Q₀ returns.
Q₀ = Q(t₀) is the initial condition, in the unit of Q; t′ is a dummy time variableAverage rate over an interval
r̄ = ΔQ/Δt = (1/Δt) ∫ r dt, over Δt = t₂ − t₁
The height of the rectangle with the same area as the region under r. It lies between the smallest and largest r, and usually equals neither.
same unit as r; it is the chord slope of the Q–t graphRates from a table
forward: [Q(t + h) − Q(t)]/h, error ≈ (h/2)Q″; central: [Q(t + h) − Q(t − h)]/(2h), error ≈ (h²/6)Q‴
Halving h halves a forward error and quarters a central one, until reading noise, growing as 1/h, takes over.
h is the positive sample spacing; separate truncation error from reading uncertainty.Accumulation from a table
trapezoid: ∫ r dt ≈ h[½r₀ + r₁ + … + rₙ₋₁ + ½rₙ]; left sum: ∫ r dt ≈ h[r₀ + r₁ + … + rₙ₋₁]
Trapezoid error shrinks as h², left-sum error only as h. The trapezoid is exact whenever r runs straight across each step.
n equal steps of width h = (t₂ − t₁)/n; rₖ is the reading at t₁ + khOptional supporting lessons
These chapter resources can help with earlier ideas; they are not additional required Physics I completion tasks.