Natural Circulation in SMRs: What Drives the Flow?

Natural circulation is flow driven by gravity acting on density differences around a connected fluid circuit. In a water-cooled SMR, a hotter, less-dense rising path and a cooler, denser return path can circulate primary coolant without a primary circulation pump.

But “hot water rises” leaves an important question unanswered: how much water flows? Enough to carry heat away depends on the heating, the geometry, the hydraulic resistance and the available heat sink. Let's change those ingredients and watch the balance move.

Figure 03 / Buoyancy-driven flow

Heat it. Watch the water move.

Unfolded primary loop · single-phase teaching model

Heat input1.00×Q / Q₀ · 0.5× to 2× reference

Effective height1.00×H / H₀ · 0.5× to 2× reference

Hydraulic resistance1.00×K / K₀ · 0.5× to 2× reference

Mass flow1.00×

Core temperature rise1.00×

At the reference state, the driving head exactly balances the hydraulic resistance. All ratios are 1.00×.

See why the curves meet here

Warm curve: driving head · Charcoal curve: losses. Curves are clipped at the top. The shared value is 1.00× the reference pressure difference. Each candidate flow on the warm curve uses the steady energy balance; this is not a stability analysis.

Each slider recalculates an equilibrium with an available heat sink. Constant water properties, fixed flow area, quadratic losses. The moving dots show direction; reduced-motion settings keep them still. This model does not calculate boiling, startup or loss of cooling.

How does natural circulation work in a reactor?

Imagine unfolding the primary flow path into a loop. Heat enters near the bottom, and a cooler removes it higher up. Heating lowers the density of the rising water; cooling makes the returning water denser. Gravity acting on these unequal columns creates a net driving head. At steady flow, that head balances friction and local losses around the loop.

The cooler is a heat exchanger with another system, not a device that makes energy disappear. That second system must ultimately transfer the energy to a heat sink. Duffey's analysis of natural-convection flow in advanced reactor concepts identifies the driving head, flow resistance, heat sink and stability as linked limits on heat removal.

The word thermosiphon is often used for a heat-driven circulation system. This example uses a single liquid phase. Other thermosiphons involve evaporation and condensation; those require a different flow model. Nor is this an ordinary siphon powered by draining one reservoir into another at a lower elevation: here water recirculates through a heated and cooled circuit.

Which SMRs use natural circulation?

The NuScale US600 design uses natural circulation in its primary system, with the core and helical-coil steam generators sharing a vessel. NRC design overview

Argentina's CAREM is another integral water-cooled example listed with natural primary circulation in the IAEA SMR Catalogue 2024. These examples identify a design approach, not operating-performance data for the interactive model.

Other designs use forced circulation during normal operation. A plant can also use different circulation modes in different circuits or operating states. The useful question is therefore: which fluid circuit, doing which job, under which conditions?

Why does the height of the cooler matter?

For an idealised pair of hot and cold columns with effective height $H$ , the driving pressure difference is approximately

\Delta p_{\mathrm{drive}} \simeq gH(\rho_{\mathrm{cold}}-\rho_{\mathrm{hot}}).

The density difference acts over a vertical distance. A larger effective separation between heat addition and removal can produce a larger driving head. In a real circuit, the distribution of density with elevation matters; $H$ is a reduced representation of that distribution, not automatically the full vessel height.

For small density changes, use a positive volumetric thermal-expansion coefficient $\beta$ and a reference density $\rho_0$ :

\rho_{\mathrm{cold}}-\rho_{\mathrm{hot}}\simeq\rho_0\beta\Delta T, \qquad \Delta p_{\mathrm{drive}}\simeq\rho_0g\beta H\Delta T.

Here $\Delta T$ is the representative hot-to-cold water temperature difference. Water properties depend on state; the widget keeps them fixed to isolate the competing effects. This approximation is not appropriate near boiling or wherever property changes are large.

Does doubling the heat double the flow?

Try Double the heat. The flow becomes about 1.26 times its reference value; the temperature rise becomes about 1.59 times its reference value. That result follows from solving two balances together.

First, energy carried by the circulating liquid must match the imposed heat input:

\dot Q=\dot m c_p\Delta T.

Second, choose a quadratic hydraulic-loss law with fixed reference area $A$ and dimensionless total loss coefficient $K$ :

\Delta p_{\mathrm{loss}}=K\frac{\dot m^2}{2\rho_0A^2}.

At steady state, set driving head equal to losses and eliminate $\Delta T$ using the energy balance:

\dot m^3=\frac{2\rho_0^2A^2g\beta H\dot Q}{Kc_p}.

With these assumptions, flow scales as the cube root of heat input, not linearly. More heat creates a larger temperature difference and more buoyancy, but the growing flow also brings larger losses.

A square-root law would apply to flow versus an independently imposed temperature difference under this loss model. Our slider imposes heat input instead: temperature difference must adjust with flow. Confusing those two experiments gives the wrong scaling.

The exact equations used by the sliders

Divide every quantity by its value at one reference equilibrium. Let $q=\dot Q/\dot Q_0$ , $h=H/H_0$ , $k=K/K_0$ , $f=\dot m/\dot m_0$ and $\theta=\Delta T/\Delta T_0$ . The widget solves

f=\left(\frac{hq}{k}\right)^{1/3},\qquad \theta=\frac{q}{f},\qquad \frac{\Delta p_{\mathrm{drive}}}{\Delta p_0}=h\theta=kf^2.

These dimensionless ratios need no invented plant rating. The reference represents a positive, single-phase equilibrium. The flow area, water properties and shape of the temperature distribution are fixed; hydraulic losses are quadratic with a constant coefficient at each selected setting. A laminar friction law or a Reynolds-number-dependent coefficient changes the scaling.

The height slider holds resistance fixed to separate the two effects. In an actual redesign, a longer flow path can add friction while increasing driving head. The sketch is not drawn to scale.

The heat sink is assumed able to remove the imposed input. The model calculates a temperature rise, not absolute fuel or coolant temperatures, and cannot establish a boiling margin or heat-exchanger capacity.

What limits natural-circulation cooling?

Try Add resistance. At the same heat input, the smaller flow needs a larger temperature rise to transport that heat. Try Lift the cooler: holding resistance fixed, more driving height reduces the temperature rise required at equilibrium.

Neither experiment establishes a safe operating limit. A real assessment must include the heat-transfer capability of the fuel and cooler, the actual density field, fluid inventory, flow distribution and operating boundaries. The IAEA's natural-circulation report includes treatment of single- and two-phase systems, stability, stagnation and thermal stratification.

The graph in the widget shows one algebraic intersection. An intersection is not proof of dynamic stability. Fluid inertia and the time it takes a temperature disturbance to travel around a loop can matter. In the real world, a stationary balance may coexist with oscillatory behaviour that a steady calculation cannot predict.

Is natural circulation the same as passive safety?

Natural circulation is a physical mechanism. A passive safety function is a system-level job that must be achieved under specified conditions. Replacing a circulation pump does not remove dependence on a continuous flow path, adequate coolant inventory, usable driving head and an effective heat sink.

It also does not remove decay heat after shutdown. The residual heat source and the heat-removal path must be considered together. Which valves, heat exchangers and sinks participate depends on the design and event. A normal-operation primary-loop model is not automatically a model of emergency cooling. Duffey: flow and heat-removal limits

Question	What to establish
What drives the flow?	The density distribution and its elevation
What resists it?	Friction, local losses and the available flow path
Where does the heat go?	A heat exchanger and a sustained heat sink
Will the response remain acceptable?	A transient model, limits and validation evidence

Where does neutronics enter the picture?

In this widget, heat input is prescribed. In a coupled reactor simulation, neutron behaviour determines fission heating; coolant flow and fuel temperature respond; the resulting material state changes the neutron calculation again.

That adds another connection to the loop: power → temperature → density and flow → temperature feedback → power. Each arrow represents quantities that a model must define and exchange consistently. Stronger physical feedback does not by itself tell us whether a particular numerical iteration will converge.

The interactive neutronics–thermal-hydraulics article isolates that numerical question with fixed coolant flow. This post lets flow respond to thermal driving at steady state. Neither model alone is a full natural-circulation reactor transient.

For the wider plant picture, read How Does a Water-Cooled SMR Work?. To discuss a coupled model, let’s start with its assumptions and verification cases.