Heat pumps move heat uphill — in both directions

Table of Contents

Thermodynamics: Vapour-compression cycle

A heat pump does not make heat the way a resistance heater or a furnace does. It spends a small amount of work W (in kilowatts, kW) to transport a larger amount of heat Q (also in kW) from a cold place to a warm place. Reverse the plumbing with one valve and the same machine becomes an air conditioner. This page follows the refrigerant around the loop in both directions, puts real numbers (temperatures in °C and kelvin (K), pressures in bar, enthalpies in kJ/kg) on every step, and then looks at the part most diagrams leave out: hysteresis, the reason the real machine never quite retraces its own path.

The one-line energy balance. Everything in this explainer is bookkeeping on a single statement of the first law of thermodynamics for one full cycle:

QH=QC+W

Heat delivered at the hot side (QH) equals heat absorbed at the cold side (QC) plus the work you paid for (W). If W = 1.0 kW of electricity moves QC = 2.4 kW of outdoor heat indoors, you receive QH = 3.4 kW of heating. Nothing is created — the outdoor air is left slightly colder.

1 — The four-part cycle

Four components, one sealed loop of refrigerant (commonly R32 or R410A in residential units, increasingly R290/propane in newer European models):

  1. Evaporator: cold, low-pressure liquid refrigerant absorbs heat and boils. Boiling at low pressure happens at low temperature, which is why the coil can pull heat out of air at −10 °C.
  2. Compressor: an electric motor (typically 0.5–3 kW in a home unit) squeezes the vapour. Squeezing raises both its pressure (roughly 3–4×) and its temperature to 60–90 °C at the discharge.
  3. Condenser: the hot, high-pressure vapour gives up its heat and condenses back to liquid.
  4. Expansion valve: the liquid is throttled through a narrow orifice. Pressure and temperature crash, and a cold liquid–vapour mixture re-enters the evaporator. This step is deliberately irreversible — remember it when we reach hysteresis.

A reversing valve (a sliding four-way valve in the refrigerant line) swaps which coil plays which role. The compressor, valve and physics do not change — only the direction of heat flow does. That is the whole difference between “heat pump” and “air conditioner”.

2 — One lab, five figures: both directions at once

The control panel below drives Figures 1–5 together, so you can watch a cold snap propagate through the schematics, the efficiency chart, the energy balance and the pressure–enthalpy loop at the same time. Figure titles update with the slider state. Realistic ranges are used throughout: outdoor air from −20 °C (a hard winter night) to +40 °C (a heatwave), indoor setpoints from 18 °C to 26 °C.

Figure 1: Heating mode at outdoor −5.0 °C

INDOORS21.0 °COUTDOORS−5.0 °CEvaporator (outdoor coil)boils at ≈ −12 °C, ≈ 5.1 barCondenser (indoor coil)condenses at ≈ 33 °C, ≈ 20 barCOMP.W inExpansion valveQCabsorbedQHdelivered to room ↘

Heating direction. The outdoor coil is the evaporator: refrigerant boils a few kelvin below outdoor air temperature so heat flows into it. The indoor coil is the condenser, running a few kelvin above room temperature so heat flows out of it. Those two approach differences (≈ 7 K outside, ≈ 12 K inside, including the air-to-refrigerant gap) are why the compressor must lift the refrigerant further than the room-to-outdoor difference alone.

Figure 2: Cooling mode at outdoor −5.0 °C

INDOORS21.0 °COUTDOORS−5.0 °CCondenser (outdoor coil)condenses at ≈ 7 °C, ≈ 9 barEvaporator (indoor coil)boils at ≈ 9 °C, ≈ 10 barCOMP.W inExpansion valveQHrejected outside④ vapour returns to compressorQCremoved from room ↘Reversing valve: flow through both coils flipped

Cooling direction: the identical hardware. The reversing valve flips refrigerant flow: now the indoor coil evaporates (absorbing room heat, QC) and the outdoor coil condenses (rejecting QH=QC+W to the summer air). Cooling labels here use the same approach rules: evaporation ≈ 12 K below indoor air, condensation ≈ 12 K above outdoor air. In a real heatwave (drag Tout to 35–40 °C) watch the condensing pressure climb past 30 bar: that rising lift is exactly why air conditioners lose efficiency on the hottest afternoons.

Cycle lab: shared controls for Figures 1–5

−5.0 °C
21.0 °C
26.0 KTemperature lift ΔT
…COP heating (realistic)
…COP cooling (realistic)
…Carnot limit (heating)

How to read the two COP readouts: the heating COP applies when it is colder outside than in (the usual winter case); the cooling COP applies in summer. Both are computed at the current slider settings from the reservoir temperatures, using a second-law efficiency of 45% for heating and 40% for cooling, typical of a good modern unit (see §3). When Tout crosses Tin, the “lift” passes through zero and both Carnot COPs blow up — correctly: moving heat across almost no temperature difference costs almost no work.

Figure 3: COP versus outdoor temperature at indoor 21.0 °C

Solid curve: realistic heating COP (45% of the Carnot limit). Dashed curve: the Carnot ceiling itself, COPCarnot,heat=THTH−TC with temperatures in kelvin. The dot tracks the shared outdoor-temperature slider; the indoor setpoint sets TH. Two lessons are visible at a glance: COP collapses as the lift grows (at −20 °C a real unit may only reach COP ≈ 2), and COP climbs steeply in mild weather (COP 4–6 near +10 °C), which is why seasonal averages (SCOP) look so much better than worst-case ratings.

Figure 4: Energy balance at outdoor −5.0 °C

Heating at a fixed delivered load of 3.5 kW (a typical single-room / small-home demand on a cold day). The work slice W is what appears on the electricity bill; QC is harvested free from outdoor air. Check the arithmetic on the bars: QH=QC+W at every slider position.

Figure 5: Pressure–enthalpy loop at outdoor −5.0 °C

The cycle as engineers draw it: specific enthalpy h (kJ/kg, R410A-like values) horizontally, pressure (bar, logarithmic axis) vertically. 1→2 compression (work in, h rises by W per kg), 2→3 condensation (QH out at constant pressure), 3→4 throttling (vertical drop, pressure falls, h unchanged: no work recovered, pure irreversibility), 4→1 evaporation (QC in). Point labels ride with the loop as the sliders move. The enclosed shape fattens as the lift grows: more work per kilogram circulated.

3 — Efficiency done properly: COP and the Carnot limit

“Efficiency” above 100% is not a violation of anything, because a heat pump is not converting work into heat — it is moving heat. The correct figure of merit is the coefficient of performance, and its theoretical ceiling comes from the reversible (Carnot) refrigerator run between the same two reservoirs. The derivation below keeps every step visible before any special case is taken.

Step 1: First law, full cycleQH=QC+W   Energy in = energy out over one complete cycle (the refrigerant returns to its starting state, so its net energy change is zero). All quantities in kW (or kJ per cycle).
Step 2: Definitions, both directionsCOPheating=QHW   (what you want ÷ what you pay, in heating)    COPcooling=QCW   (in cooling). Subtracting the two definitions using Step 1 gives COPheating=COPcooling+1 for the same machine at the same conditions: heating always scores exactly 1 higher, because the compressor’s work itself ends up as useful heat indoors.
Step 3: Carnot (reversible) limit, generalCOPCarnot,heat=THTH−TC   and   COPCarnot,cool=TCTH−TC, where TH and TC are the hot- and cold-reservoir temperatures in kelvin (K=°C+273.15). Using Celsius here is the single most common COP error: a 26 K lift is a small denominator either way, but TH = 294 K, not 21.
Step 4: Worked example (slider defaults)Tout = −5 °C = 268.15 K, Tin = 21 °C = 294.15 K, lift = 26.0 K. Carnot heating COP = 294.1526.0=11.3. A real unit at 45% of Carnot: COP≈5.1 on reservoir temperatures, before the coil approach differences of Figure 1 are charged against it, which is why nameplate COP at −5 °C is more often quoted near 3–3.5. Both numbers are honest; they answer different questions (reservoir limit vs. delivered rating).

Cooling ratings you will meet in the wild. In cooling mode the same COP is often repackaged: EER=COPcooling×3.412 in BTU/(h·W), and SEER/SEER2 average it over a season of outdoor temperatures. A 19 SEER2 heat pump and a 19 SEER2 air conditioner cool identically — they are the same cycle; the heat pump merely owns a reversing valve, which costs a few percent of pressure drop in cooling and buys the entire heating season.

ConditionToutLift (to 21 °C)Typical real COPheatNote
Mild autumn day+10 °C11 K4.5 to 6.0Best-case operating point
Freezing day0 °C21 K3.2 to 4.2Common rating point (EN 14511 A2/W35 ≈ here)
Cold snap−10 °C31 K2.2 to 3.0Defrost losses begin to bite (see §4)
Severe cold−20 °C41 K1.6 to 2.2Still beats resistance heat (COP = 1.0), but capacity also falls
Summer cooling, 35 °C day+35 °C14 K (to 21 °C)3.0 to 4.0 (cooling)Condensing pressure ≈ 31 bar for R410A

Ranges are representative of modern inverter air-source units delivering low-temperature heating water or warm air; ground-source units sit higher because their “outdoor” reservoir (soil at ≈ 8–12 °C year-round) never sees the cold snap.

4 — Hysteresis: why the machine never retraces its path

Hysteresis means the state of a system depends on its history, not just on its current inputs: the path out and the path back are different, and the loop between them encloses an area that measures energy dissipated or behaviour delayed. Heat pumps exhibit it in three distinct places — and confusing them is common, so they are separated here.

(a) Control hysteresis: the thermostat deadband

If a thermostat switched the compressor on at 20.99 °C and off at 21.01 °C, the compressor would start and stop every few seconds (“short-cycling”), destroying efficiency and the compressor itself — most wear happens at start-up, and a cycle needs minutes to reach steady pressures. So controllers deliberately build in a deadband: switch on only when the room falls to (setpoint−deadband2), switch off only when it rises to (setpoint+deadband2). Between the thresholds, the machine keeps doing whatever it was already doing — that history-dependence is hysteresis. The price is a temperature swing; the reward is long, efficient cycles. Manufacturers typically use 0.5–1.0 °C; inverter units soften the effect by modulating speed instead of switching fully off.

(b) Cycle hysteresis: irreversibility fattens the loop

On the pressure–enthalpy diagram (Figure 5), a reversible Carnot machine would trace a rectangle and return along its own path. The real loop does not close on itself reversibly: throttling (3→4) throws away the work a turbine could have recovered, heat only crosses the coils because of finite temperature gaps (≈ 7–12 K), and the compressor’s friction turns extra work into discharge heat. Each irreversibility widens the gap between the real loop and the reversible one — the thermodynamic sense in which the cycle has “memory” of having been driven. Defrost is the most visible episode: on humid days near 0 °C, frost chokes the outdoor coil, so the machine briefly reverses into cooling mode, melting the frost with indoor heat, then flips back. Whether it is “in defrost” at, say, −1 °C depends on how long it has been running and how much frost has accumulated — history again, not just temperature.

(c) What hysteresis is not

The reversing valve’s few-percent cooling penalty, and the capacity drop at low temperature, are ordinary losses, not hysteresis — they are the same on the way up and on the way down. Reserve the word for genuine path-dependence: deadbands, defrost state, and the irreversible widening of the cycle loop.

Figure 6: Room temperature over 3 hours (deadband 1.0 °C)

Indoor temperature saw-toothing between the two switching thresholds (dashed lines at setpoint±deadband2). The compressor state is shaded along the top: it turns on only at the lower threshold and off only at the upper one — between them, the recent past decides.

Hysteresis lab: thermostat deadband 1.0 °C at outdoor −5 °C

1.0 °C
−5.0 °C
…Compressor starts / hour
…Room temperature swing
…Compressor duty cycle
…Assessment

Simulation model (stated so you can check it): setpoint 21.0 °C; house loses heat at 0.055 °C/min per 26 K of indoor–outdoor difference, scaled linearly; with the compressor ON the room gains a net +0.16 °C/min before losses. One simulated run of 3 hours, 6-second steps. Drag the deadband down toward 0.2 °C and watch starts/hour explode — that is short-cycling. Drag it up to 3 °C and starts collapse, but the room now wanders through a 3-degree swing.

Figure 7: The hysteresis loop itself (deadband 1.0 °C)

The same behaviour drawn as a loop: room temperature (horizontal) against compressor state (vertical). Rising along the bottom, the machine stays OFF until the lower threshold; falling along the top, it stays ON until the upper threshold. The loop is traversed counter-clockwise forever, and its width is the deadband. A zero-deadband controller would collapse this loop to a single vertical line — and short-cycle itself to death on it.

5 — Putting it together: a worked winter evening

GivenOutside −5 °C (268.15 K), inside 21 °C (294.15 K), house needs 3.5 kW to hold temperature.
Step 1: Carnot ceilingCOPCarnot=294.15294.15−268.15=294.1526.0=11.3
Step 2: Realistic COPAt ≈ 45% of Carnot on reservoir temperatures, COP ≈ 5.1; charging the coil approach gaps (evaporating at −12 °C, condensing at 33 °C, a 45 K refrigerant lift) gives the nameplate-style figure: COP≈3.3. Use 3.3 for bill estimates.
Step 3: Electricity drawnW=QHCOP=3.5 kW3.3≈1.06 kW
Step 4: Heat harvestedQC=QH−W=3.5−1.06=2.44 kW pulled from the −5 °C air, about 70% of the delivered heat was already outdoors, free.
Step 5: The comparisonA resistance heater delivering the same 3.5 kW draws 3.5 kW (COP=1.0 by definition). The heat pump draws ≈ 1.06 kW, roughly a 70% saving on this evening, before defrost and deadband losses (a few percent) are counted.

And in July the same box, valve flipped, removes 3.5 kW from the room at a cooling COP near 3.5 on a 35 °C day, drawing ≈ 1.0 kW and rejecting ≈ 4.5 kW outside: QH=QC+W, again, with the labels rotated. One machine, one equation, two directions.

Units used throughout: temperature in °C and K (differences are identical in both), power and heat flow in kW, specific enthalpy in kJ/kg, pressure in bar absolute (1 bar = 100 kPa). Saturation pressures are R410A-like approximations for illustration; a specific unit’s service manual is authoritative for its refrigerant. COP model: second-law efficiency 45% (heating) / 40% (cooling) applied to reservoir temperatures, a teaching model, not a rating.