Fluid Mechanics, HVAC & Refrigeration

Density is mass per unit volume, $\rho = m/V$ — how much material is packed into a given space. It is an intensive property, meaning it does not depend on how much of the substance you have: a drop of water and a lake full of it have the same density, because doubling the mass doubles the volume and the ratio does not move. That is what makes it useful as a fingerprint of a material rather than a measure of a sample.

Water is the reference point everyone carries: 1000 kg/m³, which is the same number as 1.000 g/mL and 1.00 kg/L. So a 20 L pail of 12% sodium hypochlorite with a specific gravity of 1.20 holds $m = 1.20 \times 20 = 24$ kg of solution, not 20 kg. Mild steel runs about 7850 kg/m³, aluminium 2700, concrete 2400, and dry air a mere 1.2 — which is why a cubic metre of air weighs about as much as a large apple and is easy to forget entirely.

Rearranged for volume it is how a dosing calculation gets from a required mass of chemical to a number of litres to pump, and it is the density term in hydrostatic pressure, $P = \rho g h$, and in buoyancy. The temperature dependence has consequences well beyond the arithmetic: water is at its densest not when frozen but at 3.98 °C, at 999.97 kg/m³, so ice floats and lakes freeze from the surface downward instead of solidifying from the bottom up.

The mistake that ruins dosing calculations is confusing the density of a solution with the concentration of what is dissolved in it. A 50% sodium hydroxide solution at 1.53 kg/L delivers 1.53 kg of solution per litre and only 0.765 kg of actual NaOH; treating the litre as though it were all caustic doubles your dose. A related slip is confusing density with specific gravity — specific gravity is a dimensionless ratio against water, so it has no units and 1.20 SG is 1200 kg/m³, not 1.20 of anything. Two more: density shifts with temperature, so a chemical metered by volume while hot delivers less mass than the same volume cold, which matters for fuel and for concentrated solutions; and "pounds per gallon" is ambiguous unless you say which gallon, because the imperial gallon is about 20% larger than the US one.

Specific gravity strips the units off density by dividing by water's 1000 kg/m³. What is left is a pure ratio, and that is the whole reason the quantity has survived: the number is the same in every unit system. A fluid of SG 1.19 is 1.19 times denser than water whether you work in kilograms per cubic metre, pounds per cubic foot, or anything else. It also reads as a float-or-sink test at a glance — below 1 floats on water, above 1 sinks. Gasoline sits near 0.74, sea water at 1.025, concrete around 2.4, steel at 7.85.

The practical use is converting a label into a weight. A drum of concentrated hydrochloric acid is marked SG 1.19, so its density is 1190 kg/m³, and a 205 L drum holds $205 \times 1.19 = 244$ kg of liquid — worth knowing before it goes on a hand truck. Same arithmetic in a mechanical room: a 1000 L loop charged with 40% propylene glycol at SG 1.045 holds 1045 kg, and the extra mass shows up in the pump's power draw.

The instrument that reads it is Archimedes' principle made into a tool. A hydrometer is a weighted float that sinks until it displaces its own weight, so the depth it settles to is a direct readout of the surrounding fluid's density — no calculation, just a scale on a stem. Brewers watch wort fall from about 1.050 to 1.010 as sugar becomes alcohol, and the drop estimates the strength without opening the vessel. Battery technicians read the same instrument against a different scale: a healthy lead-acid cell shows about 1.265 charged and 1.120 discharged, because the sulfuric acid is genuinely consumed as the cell delivers current. Petroleum has its own derived scales, Baumé and API gravity, which are specific gravity rearranged so the numbers run the other way.

The error the name invites is treating specific gravity as a density. It is not one, and it has no units. Converting to kg/m³ by multiplying by 1000 works only because water happens to be 1000 kg/m³; in pounds per cubic foot the multiplier is 62.4. Feed an SG into an equation that wants ρ — hydraulic power, buoyancy, pressure head — and the answer is off by a factor of a thousand. The two quantities are also often quoted in the same breath on a product sheet, which does nothing to help.

Then the fine print that separates a careful figure from a rough one. A specific gravity is meaningless without two temperatures, because both the sample and the water reference expand. Standards state it as a basis: 60/60 °F means sample and reference both at 60 °F, while a 20/4 basis compares a 20 °C sample against water at its 4 °C density maximum, and the two differ by a few tenths of a percent — trivial for a drum count, not trivial in custody transfer. A hydrometer reading therefore needs a temperature correction, and a warm sample always reads low. Gases are a separate convention entirely: the specific gravity of natural gas, about 0.6, is referenced to air, not water, and reading it against water is an 800-fold error. Finally, using SG to infer a concentration requires the right table for that specific solute, and for a few solutions the relationship is not even single-valued — sulfuric acid reaches peak density near 98% and gets lighter above it, so one reading can correspond to two very different strengths.

Stand a column of fluid up and it presses down with its own weight. The pressure at depth $h$ is $P = \rho g h$ — density times gravity times depth — and the remarkable thing about it is what is not in the formula. There is no area, no volume, no mention of the shape of the container. Pressure at a given depth depends only on how far down you are and what the fluid is. A litre of water in a narrow tube 3 m tall produces exactly the same pressure at its base as a swimming pool 3 m deep.

Three metres down in fresh water: $P = 1000 \times 9.80665 \times 3 \approx 29.4$ kPa. The useful number to carry is that every metre of water is about 9.81 kPa, so every 10 m of water adds roughly one atmosphere — which is why divers count depth in atmospheres and why your ears complain at the deep end of a pool. Mercury, at 13 546 kg/m³, does the same job in 760 mm, which is where that famous barometric height comes from.

Simon Stevin worked this out in the 1580s, and Pascal is said to have demonstrated it by fixing a long thin tube into the top of a sealed barrel and bursting the barrel with a few cups of water poured down the tube. Whether or not the barrel story is literally true, the point stands and is still called the hydrostatic paradox. In practice this equation is mostly used as a translator: divide a pressure by $\rho g$ and you get head in metres, which is the language pump curves are written in.

Three things go wrong, and the density one costs the most money. Head in metres is fluid-specific: a pump rated for 30 m of head delivers $1000 \times 9.80665 \times 30 \approx 294$ kPa on water, but on a 1.20 SG brine the same 30 m of head is 353 kPa, and sizing the system as though a metre were a fixed pressure will have you short. Second, $h$ is vertical depth, measured straight down from the free surface — not the length of pipe, not the run along a sloping hose, and not the distance around a bend. A hundred metres of hose lying flat on the ground develops no static head at all. Third, this is gauge pressure, the amount by which the fluid exceeds the atmosphere above it. For absolute pressure, add about 101 kPa; at 3 m depth in a pool the absolute pressure is roughly 131 kPa, not 29.4.

A column of fluid of height $h$ presses on its base with $\rho g h$. Turn that around and any pressure can be quoted as the height of the column that would produce it: $h = P/\rho g$. Hydraulic engineers prefer to work this way because head translates straight into the geometry of a building — a number in metres can be compared against the distance from the mechanical room to the top floor, which a number in pascals cannot.

One atmosphere is $101\,325/(1000\times9.80665) = 10.33$ m of water. That single figure explains why no suction pump anywhere on Earth can draw water up more than about 10 m: atmospheric pressure is doing the pushing, and it has only 10.33 m of head to spend. A practical case from a mechanical room: filling a hydronic system whose highest point is 30 m above the fill valve requires $P = 1000\times9.80665\times30 = 294$ kPa (42.7 psi) merely to reach the top, and standard practice adds another 35 kPa or so to keep the high point positively pressurised — so the fill regulator gets set near 330 kPa, about 48 psi. Get that number wrong and the top-floor coils fill with air rather than water.

The same arithmetic with mercury gives the barometer: $101\,325/(13\,595\times9.80665) = 0.760$ m, which is where 760 mmHg comes from. It is also why water towers exist. Elevation is stored pressure, and every 10 m of tank height banks another atmosphere available to the distribution network with no pump running at all.

Head is only a pressure once you say what fluid it is in. Ten metres of water is 98 kPa; ten metres of 40% propylene glycol, at about 1045 kg/m³, is 102 kPa; ten metres of mercury is 1330 kPa. This has a consequence that catches people constantly: a centrifugal pump produces head, not pressure. Put a denser fluid through the same pump at the same speed and it delivers the same metres of head, but a higher pressure and a proportionally higher motor load. A pump curve is drawn in metres for exactly this reason, and reading it as though the vertical axis were kPa is a fair way to undersize a motor.

Three further cautions. Head computed from a gauge reading is head above atmospheric, so a calculation that mixes it with an absolute pressure is out by 10.33 m of water. Density moves with temperature — water at 80 °C is about 972 kg/m³, so the same pressure buys 6% more head than it does at 20 °C, which is small but not nothing when the margins are thin. And do not confuse this static head with what a pump must actually produce. In an open system the pump lifts the fluid and the elevation is a real cost, but in a closed loop the down leg returns everything the up leg spent, so the static height cancels completely and the pump is sized for friction alone. The building's height sets the fill pressure, not the pump head — two different questions that use the same equation and are routinely answered with each other's number.

A tire gauge reads zero in open air, yet a barometer standing beside it says 101 kPa. Both instruments are right, because they use different zeros. A gauge has atmosphere on the back of its diaphragm, so it can only ever report the difference between what it is connected to and the air around it. Absolute pressure is measured from a sealed vacuum reference and counts everything. $P_{abs} = P_{gauge} + P_{atm}$ is the bridge, and neither reading is more correct than the other — they answer different questions.

A tire inflated to 220 kPa gauge holds 321 kPa absolute. The relation matters just as much in the other direction: a "vacuum" rated at −80 kPa gauge is 21 kPa absolute, still holding a fifth of an atmosphere. And there is a floor. The most negative a gauge can ever read is −101.3 kPa at sea level, because a perfect vacuum on one side and one atmosphere on the other is the entire range available. Anyone quoting a −150 kPa vacuum is quoting something that does not exist.

The distinction dates to Torricelli's barometer in 1643, which settled two arguments at once: that the atmosphere has weight, and that a vacuum can exist. Before that there was no absolute zero of pressure to measure from. The unit conventions still carry the split — psia and psig, bara and barg, and in North American vacuum work a third convention entirely, inches of mercury below atmosphere, so that "28 inHg of vacuum" means a gauge reading of −94.8 kPa and an absolute pressure of 6.5 kPa.

Every gas law demands absolute pressure, and forgetting that is the classic error on this site. Boyle's law, Gay-Lussac's law, the combined gas law and $PV = nRT$ all count molecular impacts, and a gauge has quietly subtracted an atmosphere from the count. Compressing a tire from 220 to 440 kPa gauge is not doubling the pressure — in absolute terms it goes from 321 to 541, a factor of 1.69, and a calculation done on the gauge figures is wrong by nearly 20%. The rule is simple enough to make automatic: convert to absolute before any gas law, and convert back only at the end if a gauge reading is what somebody wants.

Then the subtler trap, which is $P_{atm}$ itself. It is not 101.325 kPa where you are. That figure is the standard atmosphere at sea level, and elevation takes it away quickly — Calgary at 1045 m sits near 89 kPa, Denver near 84, and weather moves any of them by ±3 kPa on its own. Converting a gauge reading with a textbook 101.325 in a mountain city introduces a 12 kPa error, which is 12% of an atmosphere. Worse, the barometric pressure a weather service reports is usually sea-level corrected: it has been adjusted upward to what the pressure would be if the station were at sea level, specifically so that maps are comparable. It is not the local absolute pressure and must not be used as one. If you need the real figure, read a station barometer or compute it from elevation. This is not academic — it is why a pump's available NPSH is stated absolute and why a suction lift that works at sea level can cavitate at altitude, with the equation and the elevation together explaining exactly how much margin was lost.

Picture a plane cut across the pipe. In one second, every particle within a distance $v$ upstream of that plane will have crossed it, so what passes is a cylinder of length $v$ and cross-section $A$. Volume per second is $Av$, and there is nothing more to the relation than that geometry. It is worth noticing that this is really the definition of the average velocity rather than a discovery about it: real flow moves at different speeds at different points across the bore, and $v = Q/A$ is how we agree to summarise that profile with one number.

A 100 mm pipe has a bore area of $\pi(0.05)^2 = 7.854\times10^{-3}$ m². At 1.5 m/s that carries $Q = 0.0118$ m³/s, which is 11.8 L/s or about 707 L/min. Those three numbers are the same quantity in the three units different trades habitually speak in, and being fluent in all three saves a great deal of grief on site.

Almost every other calculation in fluids takes its velocity from here. The Reynolds number needs $v$ to decide whether the flow is laminar or turbulent; the velocity head needs $v^2$; friction loss correlations are written in terms of $v$ or $Q$ throughout. It is also the relation behind the design velocities that govern piping layout — closed hydronic loops are usually kept between roughly 1 and 3 m/s, fast enough to carry air and dirt along and slow enough to stay quiet, with copper held lower still because erosion-corrosion begins to strip the protective oxide film somewhere above about 1.2 m/s in hot water.

The error that costs the most is using the nominal size as the bore. A pipe's name is not a dimension. Nominal 2 in. steel has an outside diameter of 60.3 mm and a Schedule 40 inside diameter of 52.5 mm — nowhere near 50.8. Type L copper described as 1 in. actually runs about 26.8 mm inside. Because area goes as the square of the diameter, a 5% error in bore becomes a 10% error in flow, and it compounds silently through everything downstream. Look up the schedule and the wall thickness; do not assume the label.

Two further cautions. $v$ is the average velocity across the section, not the reading at the centre. In fully turbulent pipe flow the centreline runs about 1.2 times the mean, and in laminar flow it is exactly twice the mean — so a pitot or an insertion probe held in the middle of the pipe overstates the flow unless it is corrected or traversed. And $A$ is the area actually available to flow, which is not the clean-bore area in a scaled or fouled line, and not the full section in a partly filled gravity drain.

Water does not pile up inside a pipe and it does not vanish, so whatever volume enters one end each second must leave the other. That is the whole argument. If the pipe is narrower at the second point, the same volume has to get through a smaller opening in the same second, and the only way it can is by moving faster — in exact inverse proportion to the area. $A_1v_1 = A_2v_2$ is conservation of mass for a fluid whose density does not change, dressed in the units a pipe fitter would use.

Work a reducer. A 100 mm main carrying 1.5 m/s steps down to 50 mm. The diameter halves, so the area falls by a factor of four, and the velocity must rise by four: $v_2 = 6$ m/s. Now look at what that does to the energy carried as motion. Velocity head goes as $v^2$, so it climbs from $1.5^2/19.61 = 0.115$ m to $6^2/19.61 = 1.84$ m — a sixteenfold increase, and that energy has to come from somewhere. It comes out of the pressure, which is why a gauge downstream of a sudden contraction reads lower than one upstream even before any friction is counted.

That trade is Bernoulli's equation, and continuity is the half of it that most people find intuitive. Together they explain the venturi meter, where a deliberate constriction converts a measurable pressure drop into a flow reading; the carburettor, where the same drop pulls fuel into an air stream; and the bruit a stethoscope picks up over a narrowed artery, which is the sound of blood forced through a reduced bore at a speed that has tipped it into turbulence. Rivers do it too, running slow and broad across a plain and fast through a gorge with the same discharge in both places.

The dominant error is working in diameters instead of areas. Halving the diameter does not double the velocity, it quadruples it, because area carries the square. Halving the area doubles the velocity. The two sentences sound alike and differ by a factor of two, and the mistake usually survives because the answer still looks plausible. Convert to areas first, every time.

Three conditions on when the equation applies. It is written for incompressible flow — fine for liquids and for gases well below about Mach 0.3, but a compressible flow obeys $\rho_1A_1v_1 = \rho_2A_2v_2$ instead, and above Mach 1 the behaviour inverts so that a converging duct slows the flow rather than speeding it. It is a two-point statement along a single stream: at a tee, the inlet flow equals the sum of the branches, and applying the two-term form across a branch fitting quietly loses whatever went the other way. And unlike Bernoulli, continuity does not care about friction at all. It is pure kinematics, so it holds through a filthy pipe, across a valve, and through a pump — friction and pumps change the pressure, never the volume balance.

This is continuity for a round pipe with the area written out, and the American shorthand version — v (ft/s) = 0.4085 Q (gpm) / d² (in) — is memorised by every pipefitter. Push 100 gpm through 3 in pipe and you get 0.4085 × 100/9 ≈ 4.54 ft/s, right in the sweet spot. Solve it for D instead and it becomes the sizing tool: choose your target velocity, and the pipe size follows.

Velocity limits, not pressure drop, usually govern pipe selection. Below about 2 ft/s (0.6 m/s) closed-loop systems fail to sweep air out of high points and gather sediment; above about 8–10 ft/s (2.5–3 m/s) copper erodes at the elbows, noise becomes audible in occupied spaces, and water hammer forces climb. Hot domestic water is capped lower still — 4–5 ft/s — because erosion-corrosion accelerates sharply with temperature, and this single number is why so many recirculation lines fail at the first fitting downstream of the pump.

Dynamic pressure is $\tfrac{1}{2}mv^2$ written per unit volume instead of per unit mass. Replace the mass with the density and you have $q = \tfrac{1}{2}\rho v^2$, which has units of pressure and means something physical: it is the extra pressure a moving fluid develops when it is brought completely to rest, its kinetic energy converted into a push. In Bernoulli's accounting, static pressure plus dynamic pressure equals stagnation pressure, and $q$ is the part of the total that exists only because the fluid is moving.

A 30 m/s wind — about 108 km/h, a serious storm — in air at 1.225 kg/m³ carries $q = \tfrac{1}{2}\times1.225\times30^2 = 551$ Pa. On a billboard 3 m by 6 m, that is 551 Pa over 18 m², and with a drag coefficient around 1.2 for a flat plate broadside to the flow the total is roughly 11.9 kN, a load of about 1.2 tonnes trying to fold the sign. Note where the drag coefficient entered, because that is the next paragraph but one.

Run the relation backwards and it becomes an instrument. A pitot tube faces into the flow so the air stagnates in its mouth, while static ports on the side read the undisturbed pressure; the difference is $q$, and $v = \sqrt{2q/\rho}$ turns it into a speed. Every airspeed indicator, most air-balancing hoods, and the pitot-array flow stations in large ducts work on exactly this. The same $\tfrac{1}{2}\rho v^2$ is the heart of the drag equation and of the lift equation, which are both written as $q$ times an area times a dimensionless coefficient.

Dynamic pressure is not the force per unit area on an object, and treating it as one is the standard error. Only a surface that brings the flow fully to rest, facing it squarely, sees the whole of $q$. Everything else sees some fraction or multiple of it, captured by a drag coefficient — about 1.2 for a flat plate, 0.47 for a sphere, and as little as 0.04 for a good aerofoil. The equation gives you the scale of the loading; the coefficient tells you what the shape does with it.

Two further traps, both about ρ. Air density is not a constant, and 1.225 kg/m³ is specifically dry air at sea level and 15 °C. At 2000 m elevation, or on a hot day, it can be 15 to 20% lower — which drops $q$ by the same fraction and is why aircraft need longer runways in Denver in July. This is also why an airspeed indicator reads indicated rather than true airspeed: it is calibrated for sea-level density, so at altitude the true speed exceeds the reading, and pilots learn the correction rather than a new instrument. Finally, $v$ is always the relative velocity between fluid and object, and the square means it is unforgiving. Doubling the wind speed quadruples the pressure, which is the reason storm ratings escalate so steeply and why the difference between a 100 km/h and a 140 km/h gust is a doubling of load, not a 40% increase.

Velocity head is Torricelli's law read backwards. Instead of asking how fast a fluid moves after falling a height $h$, it asks how high a moving stream could climb if all its speed were traded back for elevation. The answer is the same expression rearranged: $h_v = v^2/2g$. It is kinetic energy per unit weight of fluid, and expressing it as a length is what lets it be added directly to elevation and pressure head in one budget.

Water at 3 m/s carries $3^2/19.613 = 0.459$ m of velocity head. That is a useful benchmark, because it shows how small this term usually is. At the 1.5 m/s a hydronic loop typically runs, velocity head is 0.115 m — about a tenth of a metre against a pump head of perhaps 15 m, so under 1% of the energy budget. In ordinary piping, velocity head is nearly always negligible; in nozzles, orifices and jets it is the entire story. Knowing which situation you are in saves a great deal of unnecessary arithmetic.

The three heads together are Bernoulli's equation: elevation head plus pressure head plus velocity head, constant along a streamline. Where a pipe narrows, continuity forces the velocity up, velocity head rises as $v^2$, and the pressure head falls to pay for it — which is the venturi effect and the reason a pitot tube works, since the difference between a stagnation tap and a static tap is exactly $h_v$. Multiply velocity head by ρg and you have the dynamic pressure $q$ from the neighbouring page; they are the same quantity in different currency.

Here the honest correction matters, because textbooks state Bernoulli far too broadly. It is a statement of energy conservation along a single streamline in steady, inviscid flow — not a general law of fluids. It does not hold across a pump, which adds energy; it does not hold across a turbine, which removes it; and it does not hold through any region where friction is significant, because friction converts head irreversibly into heat. Writing "total head is constant" across 100 m of pipe is simply false — that is precisely where the head goes. The usable form carries explicit terms for pump head added and friction head lost, and the pure three-term version applies only over short, smooth, unobstructed runs.

Two smaller points. The $v$ here is the average velocity, but kinetic energy depends on $v^2$, and squaring an average is not the same as averaging the squares. The true kinetic energy of a real velocity profile is always higher, which engineers handle with a correction factor α — about 1.05 in turbulent flow, where the profile is fairly flat, but exactly 2.0 in laminar flow, where the parabolic profile means the simple formula understates the kinetic energy by half. And velocity head, like pressure head, is a length and not a pressure until it is multiplied by ρg, so it can only be added to other heads of the same fluid.

Daniel Bernoulli published this in 1738, and every one of its three terms is an energy per unit volume: static pressure, the kinetic term $\tfrac{1}{2}\rho v^2$, and the potential term $\rho g z$. Their sum is fixed along a streamline, so any one of them can only grow at another's expense. Push water from 2 m/s to 8 m/s in a level pipe and the pressure must fall by $\tfrac{1}{2}\times 1000 \times (64 - 4) = 30\ \text{kPa}$, which is the entire working principle of a Venturi, a carburettor, a laboratory aspirator and an aircraft wing.

The famous field mistake is applying it where its assumptions have quietly died. Bernoulli assumes no friction, no pump, no heat and constant density. Real pipe runs bleed head to friction, so a designer writes the extended form with an $h_f$ term on the downstream side and gets that number from Darcy-Weisbach. Gases obey it only while Mach number stays under about 0.3, above which density stops being constant and compressible relations take over.

Here is the part that catches people out: the equation holds along a streamline, not across a flow field, unless the flow is irrotational. Two points in the same pipe cross-section can carry genuinely different totals, which is why a pitot traverse across a duct reads a curve rather than a plateau. And note that velocity enters squared, so this page's velocity brains return the positive root only. If your flow actually runs from 2 back to 1, swap the two ends rather than expecting a negative answer.

Water leaving a hole in the side of a tank comes out exactly as fast as if it had fallen freely from the surface down to the hole. The argument is energy conservation along a streamline: a parcel starts at the free surface, where the pressure is atmospheric and it is barely moving, and arrives at the opening, where the pressure is atmospheric again and it is moving fast. The pressure terms are equal and cancel, leaving $mgh = \tfrac{1}{2}mv^2$, and the mass cancels too. What survives is $v = \sqrt{2gh}$ — the free-fall speed.

A tap 5 m below the water line of a standpipe jets out at $\sqrt{2\times9.80665\times5} = 9.90$ m/s. Through a 25 mm opening (area $4.91\times10^{-4}$ m²) that would be $Q = 4.86\times10^{-3}$ m³/s, or 292 L/min — if the opening passed the full theoretical flow, which it does not. Hold that number; the last paragraph collects on it.

Evangelista Torricelli published this in 1643. He was Galileo's secretary in the old man's final months and inherited both his notes and his instinct for stripping a problem to its mechanics, and it is the same Torricelli who invented the barometer a year earlier. The two results belong together: one says a column of fluid produces a pressure, the other says that pressure spends itself as speed. Notice what is absent from the formula — the fluid's density. Mercury and water pour from the same depth at the same speed, for the same reason a bowling ball and a marble fall alike. Read alongside the pressure-head and velocity-head pages, this law is simply the statement that one head converts entirely into the other.

The real discharge is far below the ideal, and this is where the equation misleads people. Fluid approaching a sharp-edged hole arrives from all directions and cannot turn the corner instantly, so the jet keeps contracting after it leaves — the vena contracta — narrowing to about 62% of the hole area a short distance out. Add a small velocity loss and the discharge coefficient for a sharp-edged orifice lands near 0.61. That 292 L/min is really about 180. Round the entry into a bellmouth and $C_d$ rises to roughly 0.98, which is why a well-formed nozzle passes half again as much water as a drilled plate of the same diameter.

Three more conditions. $h$ is measured from the free surface down to the opening, not from the top of the tank and not from the bottom — and it shrinks as the tank drains, so this gives the speed at one instant, not an average. Working out how long a tank takes to empty means integrating, which produces the pleasant result that draining time scales with the square root of the starting depth. The tank must also be vented: seal it and the flow chokes as a partial vacuum forms above the water, which is why a full jerry can glugs instead of pouring. And if the space above the liquid is deliberately pressurised, that pressure adds its own head on top of $h$, and the simple form no longer applies on its own.

Osborne Reynolds injected dye into pipe flow in 1883 and watched it either glide in a smooth filament or erupt into eddies — and found one dimensionless group predicted which. Below about Re = 2300 pipe flow is laminar; above roughly 4000 it is turbulent. Water at 1 m/s in a 5 cm pipe gives Re = 1000 × 1 × 0.05 / 0.001 = 50 000: solidly turbulent, like nearly all industrial water flow.

Because only the combination ρvD/μ matters, a small model in a wind tunnel can faithfully stand in for a full-size aircraft as long as the Reynolds numbers match — the principle that makes scale testing legitimate.

This is one of the few exact results in pipe hydraulics: integrate the parabolic Hagen–Poiseuille velocity profile and the Darcy friction factor falls out as precisely 64/Re, with no empirical fitting anywhere. At Re = 1600 that is f = 0.04. Because it is exact, it also anchors the left-hand edge of every Moody diagram ever drawn — the straight line of slope −1 on log-log paper that all the roughness curves eventually peel away from.

The striking implication is that roughness is irrelevant in laminar flow. A rusty pipe and a glass pipe give identical pressure drop, because the fluid nearest the wall is not moving and the flow never sees the bumps. Above Re ≈ 2300 the assumption collapses; in the 2300–4000 transition band no correlation is reliable and prudent engineers simply design out of it, usually by choosing a diameter that puts velocity solidly into turbulent territory.

The reason buoyancy exists is that pressure in a fluid increases with depth. The bottom of a submerged body sits deeper than its top, so it is pushed up harder than the top is pushed down, and the net of that imbalance — added over the whole surface — comes out exactly equal to the weight of the fluid the body displaced. That is Archimedes' principle, and the derivation makes clear why the material of the body is irrelevant. The fluid only ever touches the outside; it has no way of knowing what is inside the shape.

Displace 1 m³ of fresh water and you get $1000 \times 1 \times 9.80665 = 9807$ N of lift — about a tonne-force — whether the displacer is steel, styrofoam or a sealed void. This is why a ship's size is quoted as its displacement: a 10 000 tonne vessel is one that settles until it has pushed aside 10 000 tonnes of water, roughly 10 000 m³ of it. A submarine hovers by adjusting ballast until the two figures match, and a hot-air balloon does the identical arithmetic in air, where a 2500 m³ envelope displaces about 3 tonnes of atmosphere.

The story of Archimedes leaping from the bath comes from Vitruvius, writing two centuries after the fact, and the crown test as popularly told — measuring the overflow — would have been far too crude to catch the adulteration. The method that actually works, and that Archimedes' own writing supports, is to weigh the crown in air and again suspended in water: the difference is the buoyant force, which gives the volume, which gives the density. That comparison is still the standard way to measure the density of an irregular solid, and it is the principle inside every hydrometer.

Two substitutions account for most wrong answers, and they are easy to state. First, $\rho$ is the density of the fluid, never of the object — the object's density decides whether it floats, but it plays no part in this equation. Second, $V$ is the displaced volume, which equals the object's volume only when the object is fully submerged. A floating body displaces just the part below the waterline, and it settles until the fluid it has pushed aside weighs exactly what the body weighs. Fully submerged: displaced volume equals object volume. Floating: displaced weight equals object weight. Reaching for the wrong one of those two is the classic slip.

Then the details that matter in practice. Fluid density is not one number: sea water at 1025 kg/m³ gives 2.5% more lift than fresh at 1000, which sounds trivial until a loaded ship moves from ocean into a river and sinks noticeably lower — the reason Plimsoll load lines carry separate marks for fresh and salt, summer and winter. What a scale reads for a submerged object is the apparent weight, true weight minus $F_b$, not the buoyant force itself. And air is a fluid too: everything weighed on a bench is buoyed by about 1.2 kg per cubic metre of its volume, which is negligible for a steel block and a genuine correction when calibrating precision masses.

Julius Weisbach published this in 1845 and Henry Darcy supplied the experimental friction factors; nearly two centuries on it is still the only pipe-friction equation with no fluid, temperature or material restrictions. The structure is transparent: velocity head v²/2g is the currency, L/D counts how many diameters of pipe you are traversing, and f is the price per diameter. Water at 2 m/s through 100 m of 100 mm pipe with f = 0.02 loses 0.02 × 1000 × (4/19.613) ≈ 4.08 m.

Two traps. The Darcy friction factor is four times the Fanning friction factor used in chemical engineering — mixing them up produces a 4× error, and the giveaway is that Fanning f for laminar flow is 16/Re rather than 64/Re. And D must be the true inside diameter, not the nominal size: 4 in Schedule 40 steel is actually 4.026 in, while 4 in Schedule 80 is 3.826 in, and since head loss scales as roughly D⁻⁵ at constant flow that 5% difference is a 28% error in pressure drop.

Colebrook and White's 1939 equation is the accepted description of turbulent friction, but it has f on both sides and must be iterated — a genuine nuisance in the slide-rule era and still an irritation in a spreadsheet. In 1976 Prabhata Swamee and Akalank Jain published this explicit fit that lands within about 1% of Colebrook across the whole practical range, and it has been the default in hydraulic software ever since. Commercial steel (ε = 0.045 mm) at 100 mm bore and Re = 100 000 gives f ≈ 0.0202, matching a Moody chart read to the width of a pencil line.

Watch the roughness values: ε is absolute, in the same length units as D, and it varies enormously — 0.0015 mm for drawn tubing, 0.045 mm for new commercial steel, 0.15 mm for galvanised, 0.26 mm for cast iron, 3 mm for riveted steel. New-pipe roughness is also optimistic; a domestic-water steel line ten years into service can have several times the design ε from tuberculation, which is exactly why plant hydraulic models drift from reality and why Hazen–Williams C values are quietly downgraded as systems age.

Cyril Colebrook and Cedric White published this in 1937 after fitting Nikuradse's sand-grain roughness experiments to commercial pipe, and it is still the reference description of turbulent friction. The Moody diagram, which nearly every engineer has read off at some point, is nothing more than this equation plotted. Take a relative roughness of 0.0001 at $Re = 10^5$: iterate $x = -2\log_{10}(\varepsilon/3.7D + 2.51x/Re)$ from any sensible start and it settles on $x = 7.3494$ within five passes, giving $f = 1/x^2 = 0.01851$. That is a Moody chart read to the width of a pencil line.

The awkwardness is that $f$ appears on both sides. Before calculators this meant either a chart or a slide-rule iteration, which is why Swamee and Jain's explicit fit and half a dozen rivals exist. This page does the iteration properly rather than approximating it, so it agrees with the reference equation to machine precision instead of to about 1%. Going the other way is easier than it looks: once $f$ is known, $1/\sqrt f$ is known, the logarithm can be undone, and roughness, diameter or Reynolds number each fall out in closed form. That is why this page can solve for all four.

The surprise is at the two ends of the curve. At very high Reynolds number the $2.51/(Re\sqrt f)$ term vanishes and friction stops depending on velocity at all, which is the fully rough regime where the Moody curves go flat. Ask this page for a Reynolds number in that region and it will tell you honestly that none can be recovered. At the other end, below about $Re = 4000$, the equation simply does not apply; use $f = 64/Re$ instead, and treat the 2300 to 4000 transition band as territory no correlation describes and no prudent designer operates in.

Allen Hazen and Gardner Williams fitted this in 1905 to thousands of waterworks measurements, and its great virtue is that all the pipe's character collapses into one number, C, that a utility can measure by fire-flow test and track over decades. New plastic runs C = 150, cement-lined ductile iron 140, new steel 130, twenty-year-old unlined cast iron 100 or worse. Push 50 L/s through 100 m of 200 mm pipe at C = 130 and you lose about 1.28 m — comfortably inside the 3 m per 100 m that most distribution standards allow.

The formula is emphatically not general. It is calibrated for water near 15 °C at velocities under about 3 m/s in pipes 50 mm and larger; use it on glycol, on oil, on steam or on hot water and the answer is wrong by an amount nobody can bound, because viscosity appears nowhere in it. That is also its charm — fire-protection codes such as NFPA 13 mandate Hazen–Williams precisely because it needs no fluid properties and no iteration, and every sprinkler hydraulic calculation in North America runs on it. For anything but cool water, use Darcy–Weisbach.

This is Hazen–Williams in its original velocity form, and it is the version that handles part-full sewers and open channels as well as pressure pipes, because everything geometric enters through the hydraulic radius R = flow area ÷ wetted perimeter. For a full circular pipe R is simply D/4, so a 200 mm main is R = 0.05 m; at C = 130 on a 1% gradient that gives $v = 0.849 \times 130 \times 0.05^{0.63} \times 0.01^{0.54} \approx 1.39$ m/s.

The constant carries the units, and this is where careless work goes wrong. In SI, with R in metres and v in metres per second, the coefficient is 0.849; the familiar American form $v = 1.318\,C\,R^{0.63} S^{0.54}$ wants R in feet and returns feet per second, and the two are related by $1.318 \times 0.3048^{0.37} \approx 0.849$. S is dimensionless in both systems — head loss per unit length, so 12 ft per 1000 ft is S = 0.012, not 12. Enter it as a plain ratio here.

Every disturbance a fitting imposes on the flow — the sudden turn of an elbow, the contraction through a gate seat, the wake behind a tee — dissipates some multiple of the velocity head. Crane Technical Paper 410 tabulates those multiples as K: about 0.9 for a standard threaded 90° elbow, 0.2 for a fully open gate valve, 10 for a globe valve, 0.5 for a sharp-edged pipe entrance, 1.0 for the exit into a tank. Water at 8 ft/s through that elbow loses 0.9 × 64/64.35 ≈ 0.90 ft.

Calling these "minor" losses is a historical joke that has misled generations. In a long transmission main they truly are minor, but in a chiller room or a pump skid — twenty elbows, four valves and a strainer in thirty feet of pipe — they dominate, often carrying three-quarters of the total loss. The other subtlety is the velocity: K is referenced to the velocity in the pipe of the fitting's nominal size, so for a reducing fitting you must be clear which of the two velocities the tabulated K assumes.

Set Darcy–Weisbach equal to the minor-loss equation — f(L/D)(v²/2g) = K(v²/2g) — and the velocity heads cancel, leaving $L_{\mathrm{eq}} = KD/f$. The point is bookkeeping: rather than tracking two different loss equations, a designer converts every valve and fitting into a phantom length of pipe, adds it to the measured run, and computes one friction loss for the whole circuit. A 90° elbow (K = 0.9) on 4 in pipe with f = 0.02 is worth 0.9 × 0.333/0.02 = 15 ft of straight pipe.

Because f cancels out of nothing here, the equivalent length depends on the pipe's friction factor — which is why the old rule "an elbow equals 30 diameters" is only true near f = 0.03. On smooth plastic with f = 0.018 the same elbow is worth 50 diameters. Fire-protection and refrigeration codes finesse this by publishing fixed equivalent-length tables for a nominated schedule and material; those tables are fine inside their intended context and misleading outside it.

Cv is a valve's capacity expressed as a single number: the US gallons per minute of 60 °F water that flow through it, wide open, at exactly 1 psi of pressure drop. Because loss goes as velocity squared, flow scales as the square root of the drop, so a Cv 25 valve passes 25 × √4 = 50 gpm at 4 psi. Run the equation the other way to size a valve — the required Cv for 100 gpm of 1.1 SG glycol at a 1 bar (14.5 psi) design drop is 100/√(14.5/1.1) ≈ 27.5, so you pick the next standard size up.

The single biggest control-valve error is sizing for full flow at minimum drop. A control valve only controls if it owns a meaningful share of the circuit's pressure — typically 25–50% of the variable loss at design flow. Size it for the tiny drop the pump has left over and the valve becomes a hole in the pipe: it does its entire job in the first 10% of travel, hunts, and wears out its trim in a season. Note also that Cv assumes non-flashing liquid; at high drops choked flow caps the capacity no matter how much more ΔP you apply.

Kv is the European twin of Cv: cubic metres per hour of water at 1 bar of pressure drop instead of gallons per minute at 1 psi. A Kv 18 valve passes 18 m³/h — 300 L/min — at 1 bar. The two coefficients differ only by unit conversion, Cv = 1.156 Kv, so a valve stamped Kv 10 is a Cv 11.6 valve; European balancing-valve charts and North American control-valve catalogues describe exactly the same hardware in these two dialects.

The practical warning is that manufacturers publish both, and mixing them silently under-sizes a valve by about 15%. Check the datasheet units before you trust a number, and beware a third variant, Av, defined in SI base units (m² of equivalent orifice) with Av = 2.4 × 10⁻⁵ Kv. On balancing valves the published Kv is also a function of handwheel position, so the catalogue table has a Kv per turn — always read at the setting you will actually commission.

An orifice plate is a disc with a hole, bolted between flanges, and it has survived a century of better ideas because it is cheap, has no moving parts, and is described by an international standard anyone can audit. Drop the flow through a restriction and the pressure falls by an amount proportional to the square of velocity, so the flow comes back out as a square root. A 50 mm bore in a 100 mm line with 20 kPa across it passes about 469 L/min of water. Double the flow and you need four times the differential, which is exactly why an orifice meter's usable turndown is only about 3 or 4 to 1: at a third of full flow the differential has fallen to a ninth and the transmitter is reading noise.

The discharge coefficient near 0.61 is not a fudge factor, it is the vena contracta. The jet keeps converging for a diameter or so downstream of the plate, so the true minimum flow area is smaller than the drilled hole. ISO 5167 gives the Reader-Harris/Gallagher equation for $C_d$ as a function of $\beta$ and Reynolds number, and it stays within a few tenths of a percent when the installation follows the standard. The velocity-of-approach factor $1/\sqrt{1-\beta^4}$ is a separate correction, accounting for the fact that the upstream fluid was already moving.

What surprises people is how much of the accuracy lives in the pipe rather than the plate. The standard demands long straight runs, typically 10 to 40 diameters upstream depending on what disturbance sits there, because a swirl left over from two out-of-plane elbows will bias the reading by several percent and no calibration will find it. The other quiet killer is a nicked or rounded upstream edge: an orifice plate installed backwards, or one that has been in erosive service for a few years, reads high by a margin that nothing on the transmitter reveals.

Clemens Herschel patented the Venturi tube in 1887, naming it after Giovanni Venturi, and the arithmetic is identical to the orifice plate's. Only the coefficient changes. Because the convergent cone guides the flow instead of shearing it, there is no vena contracta and $C$ sits near 0.98 rather than 0.61. A 200/100 mm Venturi on 30 kPa passes about 3695 L/min of water.

The real argument for a Venturi is not accuracy but permanent pressure loss. An orifice plate throws away most of the differential it creates, typically 60 to 80% of it, forever, as heat and turbulence. A classical Venturi with its long 7 to 8 degree divergent recovery cone gives back most of what it borrowed and loses perhaps 10 to 15%. On a large pumped main running continuously, that difference is a genuine and permanent pump-energy line item, and it is what justifies the Venturi's much higher purchase price and its much greater length.

Two things worth knowing. The Venturi tolerates dirty and slurry service that would erode an orifice edge into uselessness, which is why they are standard on raw water, sewage and mineral slurries. And because the coefficient is so close to 1 and so flat with Reynolds number, a Venturi can often be used uncalibrated to within a percent, whereas an orifice plate near the low end of its range genuinely needs its $C_d$ computed rather than assumed.

Cut a pressurised pipe lengthwise in your imagination: the pressure acting on the projected area D per unit length must be resisted by two wall thicknesses in tension, which gives P = 2St/D directly. Peter Barlow published it in 1836 for cast-iron water mains and it has outlived nearly every rival. A 6.625 in OD line pipe with a 0.25 in wall in 35 000 psi material bursts, in theory, at 2 × 35 000 × 0.25/6.625 ≈ 2642 psi.

That theoretical number is never the rating. Codes such as ASME B31.3 and B31.8 multiply it by a design factor (0.72 for most gas transmission), a longitudinal-joint factor and a temperature derating, then subtract mill tolerance and a corrosion allowance from t — so the stamped working pressure is often less than half the Barlow result. Note too that D is the outside diameter here; some references write the same formula with mean or inside diameter and a correspondingly different answer, and being conservative means using OD.

Nikolai Joukowsky derived this in 1898 after investigating burst mains in the Moscow water system, and the result is brutally simple: stop a moving column of liquid instantly and the pressure spike is ρaΔv, independent of pipe length or line pressure. Water at 2 m/s in steel pipe (a ≈ 1200 m/s) surges 1000 × 1200 × 2 = 2.4 MPa — about 350 psi on top of whatever the system already carried. That is why a slammed solenoid valve makes the pipes bang and why a 150 psi-rated fitting can fail on a 60 psi system.

The word "instantly" is doing real work. The surge only reaches full Joukowsky value if the valve closes faster than the wave's round trip, 2L/a — for a 300 m run in steel that is half a second, so most quarter-turn valves qualify. Close slower and the surge falls roughly in proportion. The engineering answers are all about slowing that change: geared or motorised valve actuators, soft-start and soft-stop pump drives, air chambers and bladder arrestors near quick-closing fixtures, and surge tanks on long transmission mains. And note the flip side — a pump tripping on power loss produces a downsurge that can pull the line below vapour pressure and cause column separation, whose rejoining slam is often worse than the original event.

Knowing the system volume is what turns a chemical dose rate into a number of litres to pour in. A 100 ft run of 4 in Schedule 40 steel holds π/4 × (4 in)² × 1200 in ≈ 15 080 in³ ≈ 65.3 US gallons, which is the origin of the plumber's constant "0.653 gallons per foot for 4-inch". Water treaters carry a whole table of these — 0.0408 gal/ft for ½ in, 0.163 for 1 in, 1.47 for 6 in — and estimate a building's loop volume by tallying pipe lengths from the riser diagram.

Use the true inside diameter, not the nominal size, or you will be out by 10–15% before you start: nominal 4 in Schedule 40 is 4.026 in ID, Schedule 80 is 3.826 in, and type L copper is 3.905 in. And remember the pipe is usually the small half of a system — an air handler coil, a buffer tank and a boiler can hold several times what the distribution piping does, which matters enormously when you are calculating how long a glycol charge or a biocide slug takes to circulate.

A dipstick in a horizontal tank does not read linearly, and the reason is geometry: the wetted cross-section is a circular segment whose area grows fastest near the centreline and slowest at top and bottom. Half-full is the one easy case — a 1 m radius, 3 m long tank at h = r holds 3 × (π/2) ≈ 4.712 m³ — but at a quarter depth the tank holds only about 19.6% of its capacity, not 25%. Every fuel-oil, propane and chemical bulk tank on site comes with a strapping chart that is nothing more than this formula tabulated.

The formula covers the straight shell only. Real tanks have dished, elliptical or hemispherical heads that add 5–15% more volume, and a slight tilt (deliberate, for drainage) skews the reading further. Note also the arccosine: enter a depth greater than the full diameter 2r and the expression has no real value, which is the calculator's way of telling you the tank is already overflowing.

Total dynamic head is the pump's job description in a single number. Lift the water 85 ft, lose 22.5 ft pushing it through the pipe, and hand it over with 1.5 ft of velocity head still on board, and the pump must produce 109 ft. Every pump selection begins here: TDH sets the y-coordinate of the duty point, design flow sets the x-coordinate, and the curve that passes closest to that point through its best-efficiency region wins the job.

The recurring mistake is measuring static head from the pump instead of between the two liquid surfaces. If the supply tank is flooded above the pump, that submergence subtracts from static head — a fact worth real money on a long transfer line. And do not double-count: many engineers fold velocity head into the friction allowance, which is fine for water systems where it is a fraction of a foot, but sloppy on high-velocity or short-run systems where it can be several percent of the total.

Each second, a pump moves $\rho Q$ kilograms of fluid and raises them $h$ metres. The work done on that mass is $\rho Q g h$ joules, and since it happened in one second, that is the power. There is nothing more to the derivation than $mgh$ with the mass supplied at a steady rate, which is why the equation is exact rather than empirical. What it gives is the water power — the energy actually landing in the fluid, before any machine is asked to deliver it.

Pumping 100 L/s up 20 m: $P = 1000 \times 9.80665 \times 0.1 \times 20 = 19.6$ kW at the water. Now walk it back through the machinery. A pump running at 75% efficiency needs 26.2 kW at the shaft, and a motor at 92% draws about 28.5 kW from the panel. Run it 4000 hours a year at ten cents a kilowatt-hour and that pump costs roughly $11 400 annually in electricity — a figure that usually dwarfs the capital cost and is the real reason efficiency points are worth arguing over.

Run the equation backwards and it is hydroelectricity. The same $Q$ and $h$ falling through a turbine yield the same $\rho gQh$, less the turbine's own efficiency, which is why a run-of-river plant's output can be estimated from a flow gauge and a contour map before anything is built. It is also the honest reply to most perpetual-motion schemes involving pumped water: the energy to lift it is exactly the energy available when it falls, and every real component takes a cut in both directions.

The mistake that undersizes more pumps than any other is putting the physical lift into $h$. The $h$ this equation wants is the total dynamic head: the static lift, plus the friction loss through every metre of pipe, elbow, valve and heat exchanger, plus any residual pressure the discharge has to be delivered against. On a long or restricted run, friction can exceed the geometric lift several times over. Sizing on height alone gives a pump that produces its rated flow into an open pipe and much less into the real system.

The reverse error appears in closed loops and is just as common. A hydronic circuit's return leg gives back everything the supply leg spent lifting, so the static height cancels entirely and $h$ is friction alone — a pump for a thirty-storey closed loop may need only a few metres of head. Height sets the fill pressure; friction sets the pump. Three more points of arithmetic. $Q$ belongs in cubic metres per second, and the usual slips are a factor of 1000 from litres per second or 60 000 from litres per minute. Efficiency is not a fixed property of a pump but a curve, and a unit operating far from its best efficiency point can fall to 40% or worse, which is why throttling a valve to reduce flow is such an expensive way to control a system. And ρ is the density of what you are actually pumping — 40% glycol at 1045 kg/m³ costs about 4.5% more power for the same head and flow, before its higher viscosity adds friction on top.

Water horsepower is the honest output of a pump: the weight of liquid lifted per second times the height it is lifted. In US customary units that reduces to gpm × ft × SG ÷ 3960, where the 3960 is just 33 000 ft·lbf/min per horsepower divided by 8.33 lb per gallon of water. Pump 500 gpm to 100 ft of head on water and you are delivering 500 × 100 × 1.0 / 3960 ≈ 12.6 hp to the liquid — this calculator works in exact SI (ρgQH with g = 9.80665) and returns 12.64 hp, the small difference being the rounding baked into 3960.

The specific-gravity term is where people come unstuck. Head in feet is independent of density, but power is not: the same pump moving 1.2 SG glycol brine to the same 100 ft of head needs 20% more horsepower even though the gauge pressure and the curve point are unchanged. Size the motor for the coldest, densest fluid the system will ever see, including a winter glycol charge that maintenance may add years later.

Brake horsepower is water horsepower divided by the pump's efficiency — the power that actually shows up on the coupling, and therefore the number that sizes the motor. Take 500 gpm at 100 ft on water through a 70%-efficient pump: 500 × 100 × 1.0 / (3960 × 0.70) ≈ 18.0 hp, so a 20 hp motor. The 3960 is the same US-customary constant as in water horsepower, and this page evaluates the exact SI equivalent ρgQH/η so you may enter L/s, metres and kilowatts freely.

Efficiency is not a property of the pump; it is a property of the operating point. A pump rated 78% at best efficiency may run at 55% out on the far right of its curve, and the shaft power can then exceed the nameplate even though the flow looks reasonable. Read η off the curve at your actual duty, not off the datasheet headline, and add the motor's own efficiency (typically 90–94%) if you want electrical input rather than shaft power.

Everything a pump fails to put into the water it puts into heat, noise and vibration. A unit delivering 15 kW of hydraulic power while absorbing 20 kW at the shaft is 75% efficient, and the missing 5 kW is warming the casing and the liquid. Large water-supply pumps reach 85–90%; small close-coupled circulators struggle to pass 40%; and any positive-displacement chemical metering pump is efficient in a different sense entirely.

The trap is comparing efficiencies measured at different boundaries. Pump efficiency is hydraulic over shaft; wire-to-water efficiency multiplies that by the motor and drive efficiencies and is the number that matters on the electricity bill. A 75% pump on a 92% motor behind a 97% VFD is 67% wire-to-water — so if you are calculating operating cost from a "75% efficient" pump, you will underestimate the annual energy by about 12%.

Cavitation is boiling by pressure drop rather than by heating: if the absolute pressure at the impeller eye falls to the liquid's vapour pressure, bubbles form and then collapse violently downstream, pitting the impeller and making the pump sound like it is passing gravel. NPSHa is the margin you have; NPSHr, from the manufacturer's curve, is the margin the pump demands. Keep NPSHa at least 2–3 ft above NPSHr, more on hot or volatile service. A typical open-tank job: 33.9 ft of atmosphere, 10 ft of flooded suction, 2.5 ft of suction friction, 0.6 ft of vapour pressure gives 40.8 ft available — enormous margin.

The killer variable is temperature. Water at 20 °C has 0.8 ft of vapour-pressure head; at 90 °C it has 24 ft, and at boiling it swallows the whole atmospheric term. Condensate and boiler-feed pumps are therefore designed with flooded suctions and generous suction piping, and a clogged suction strainer — which quietly grows $h_f$ — is the most common single cause of a pump that "suddenly started cavitating" after years of good service. Enter $h_s$ as a negative number when the pump sits above the liquid level.

Put the cavitation number and the Euler number side by side. One is $\Delta p/\rho v^2$. The other is $(p - p_v)/\tfrac{1}{2}\rho v^2$. Strip the units off both and they are the same group: a pressure difference divided by a dynamic pressure. Dimensional analysis cannot tell them apart, and that is worth sitting with for a moment.

Buckingham's theorem hands you a pressure group. It does not tell you to put $p - p_v$ on top rather than $p_1 - p_2$, because it knows nothing about vapour pressure, or boiling, or the fact that a liquid has a floor below which it stops being a liquid. That knowledge comes from physics, and the choice of numerator encodes a question: Euler asks how much pressure this component costs me. Cavitation asks how close this flow is to boiling. Same algebra, different engineering. Dimensional analysis organises understanding; it does not supply it.

The mechanism is straightforward and the damage is not. Where a flow accelerates — around a blade leading edge, through a valve seat, at a pump impeller eye, over a hydrofoil — the static pressure falls. If it falls to the vapour pressure of the liquid, vapour cavities form. They are carried downstream into a region of higher pressure and collapse, and the collapse is violent: a bubble imploding near a solid surface does so asymmetrically, driving a microjet at the wall at speeds measured in hundreds of metres per second. Repeat that a few million times and steel pits, bronze erodes, and a pump impeller that looks sandblasted comes out of service.

There is no universal critical value, and anyone who quotes one is quoting a geometry. Incipient cavitation occurs when the cavitation number equals the magnitude of the geometry's minimum pressure coefficient — roughly 0.2 for a well-designed hydrofoil, 1 to 2 for a blunt body, several for a sharp-edged throttling valve. The threshold is a property of the shape. Worse, it is not even a single number for a given shape: nuclei content, dissolved gas, surface finish and the residence time at low pressure all move it, which is why the number at which cavitation appears as you speed up differs from the number at which it disappears as you slow down. Incipient and desinent cavitation numbers bracket a hysteresis loop.

Two input traps. $p$ is an absolute pressure. A gauge reading entered here makes the margin look about 101 kPa smaller than it is, which is a conservative error and still an error. And $p_v$ moves fast with temperature: water goes from 2.34 kPa at 20 °C to 7.38 kPa at 40 °C to 101.3 kPa at 100 °C. A system that never cavitates in winter can cavitate reliably in August, and a hot-water pump has a fraction of the margin the same pump has on cold service.

The pump trade expresses the identical margin as NPSH available, a height in metres rather than a ratio, and the two are the same statement. The dimensionless form travels better between geometries; the head form is easier to compare against a manufacturer's curve. Use whichever the person you are talking to uses, and know that they are the same calculation.

A centrifugal impeller is a volumetric scoop: every revolution flings out roughly the same slug of liquid, so turn it faster and the capacity climbs in lockstep. Slow a 1750 rpm pump to 1150 rpm and a 500 gpm duty becomes 500 × 1150/1750 ≈ 329 gpm. This is the whole economic case for variable-frequency drives — but the flow law is the gentlest of the three; head follows the square and power the cube, which is where the savings actually come from.

The trap is applying it across a duty the pump cannot reach. Affinity laws slide a point along a parabola through the origin, not along the system curve; if your system has 40 ft of static lift, dropping speed eventually drives the pump head below that lift and flow collapses to zero rather than scaling smoothly. Always plot the new curve against the real system curve before promising a customer a linear turndown.