Solow growth model, convergence, and endogenous growth

Intuition Beginner

Why do some countries grow rich while others stay poor? The Solow model answers by tracking capital — the machines, factories, and roads a nation builds. Each year a country saves part of its income and turns it into new capital. But two forces pull capital back down: old equipment wears out, and new workers arrive who must each be equipped. The economy settles where new investment exactly offsets wear-out and dilution. That resting point is the steady state.

A richer steady state needs a higher saving rate, but here is the catch. Once a country reaches its steady state, growth per worker stops — unless technology keeps improving. The model treats technology as a gift arriving from outside. A deeper question follows: what sets the pace of technology itself? Endogenous growth theory makes innovation a deliberate activity, driven by the rewards to inventing, so the growth rate is chosen inside the model rather than handed down from outside.

The takeaway: piling up capital explains the level of prosperity, but only better ideas and invention lift the growth rate of living standards in the long run.

Visual Beginner

The Solow diagram plots two curves against capital per effective worker, $k$. The curved line is actual investment $s\cdot f(k)$, rising but flattening because each extra unit of capital yields less. The straight line is break-even investment $(\delta +n+g)\cdot k$, the amount needed to cover wear-out and equip new workers. Where they cross, the economy is at rest.

Curve	What it shows	Shape
Investment $s\cdot f(k)$	New capital added each year	Rising, flattening out
Break-even $(\delta +n+g)\cdot k$	Capital lost to wear-out and dilution	Straight line through origin
Crossing point $k^{\ast }$	The steady state	Investment equals break-even

Left of $k^{\ast }$ investment exceeds break-even, so $k$ rises; right of $k^{\ast }$ break-even exceeds investment, so $k$ falls. Either way the economy is pulled toward $k^{\ast }$.

Worked example Beginner

A country saves $20\%$ of income, so $s=0.2$. Capital wears out at $\delta =0.05$ per year. Population grows at $n=0.02$ and technology at $g=0.02$. Capital's share of income is $\alpha =0.3$. Find the steady state and the long-run growth rate.

Step 1. Add the three shrinkers: $\delta +n+g=0.05+0.02+0.02=0.09$. This is the break-even rate — the investment per unit of capital needed to hold $k$ steady.

Step 2. Divide saving by the break-even rate: $0.2/0.09\approx 2.22$.

Step 3. Raise to the power $1/(1-\alpha )=1/0.7\approx 1.43$. The steady-state capital per effective worker is $k^{\ast }\approx 2.22^{1.43}\approx 3.1$.

Step 4. Output per effective worker settles at $y^{\ast }=(k^{\ast }{)}^{\alpha }\approx 3.1^{0.3}\approx 1.4$, and consumption per effective worker is $(1-s)\cdot y^{\ast }\approx 0.8\times 1.4\approx 1.13$.

What this tells us: at the steady state, output per worker grows at the technology rate $g=2\%$ per year, while total output grows at $n+g=4\%$. Raising the saving rate would lift the level of $k^{\ast }$ but leave the long-run growth rate at $2\%$.

Check your understanding Beginner

Formal definition Intermediate+

Let output be produced from physical capital $K_t$ and effective labour $A_t{L}_t$ by a Cobb-Douglas production function $Y_t=K_t^{\alpha }(A_t{L}_t{)}^{1-\alpha }$ with $0<\alpha <1$ ^{[Solow 1956]}. Labour grows at rate $n$ and labour-augmenting technology at rate $g$, so $L_t=L_0{e}^{nt}$ and $A_t=A_0{e}^{gt}$. A constant fraction $s$ of output is saved and invested, and capital depreciates at rate $\delta$, giving the accumulation equation

$${\dot{K}}_t=s{Y}_t-\delta K_t.$$

Writing variables in per-effective-worker units, $k_t=K_t/(A_t{L}_t)$ and $y_t=f(k_t)=k_t^{\alpha }$, the law of motion collapses to the one-dimensional system

$${\dot{k}}_t=sf(k_t)-(\delta +n+g)k_t=s{k}_t^{\alpha }-(\delta +n+g)k_t.$$

Steady state. A steady state is a value $k^{\ast }>0$ with $\dot{k}=0$. Setting $s(k^{\ast }{)}^{\alpha }=(\delta +n+g)k^{\ast }$ gives the unique positive solution

$$k^{\ast }={\left(\frac{s}{\delta +n+g}\right)}^{1/(1-\alpha )}.$$

Golden rule. Steady-state consumption per effective worker is $c^{\ast }=f(k^{\ast })-(\delta +n+g)k^{\ast }$. The golden-rule capital stock $k_{GR}$ maximises $c^{\ast }$ and satisfies the first-order condition $f^{\prime }(k_{GR})=\delta +n+g$: the marginal product of capital equals the break-even rate. For Cobb-Douglas this gives a golden-rule saving rate $s_{GR}=\alpha$.

Augmented Solow (Mankiw-Romer-Weil). Introducing human capital $H$ with share $\varphi$ gives $Y=K^{\alpha }{H}^{\varphi }(AL{)}^{1-\alpha -\varphi }$, where $\alpha +\varphi <1$ ^{[Mankiw-Romer-Weil 1992]}. If fractions $s_K$ and $s_H$ of income are accumulated as physical and human capital, the model retains the same convergence dynamics but delivers a higher implied total-capital share $\alpha +\varphi$, which fits the cross-country income distribution far better than the unaugmented model.

Endogenous growth (AK). Replacing diminishing returns with the linear technology $Y=AK$ breaks convergence: the law of motion $\dot{k}/k=sA-(\delta +n)$ is constant, so the long-run growth rate becomes $g=sA-\delta -n$, a function of the saving rate ^{[Barro-Sala-i-Martin 2003]}.

Counterexamples to common slips

• Capital accumulation is not the source of long-run per-worker growth. In the basic Solow model it raises the level of the steady state; the per-worker growth rate is $g$, set by exogenous technology.
• Convergence is conditional, not absolute. Economies converge toward their own steady states, determined by their own $s$, $n$, $g$, and human-capital accumulation; identical steady states are not guaranteed across countries.
• A higher saving rate can reduce consumption. Above the golden rule, dynasties would be better off saving less: the extra capital is not worth the forgone consumption, a state called dynamic inefficiency.

Economic theory Intermediate+

The central theoretical claim of the Solow model ^{[Solow 1956]} is global convergence to a unique steady state under the assumption of a strictly concave production function.

Theorem (Solow steady state, uniqueness, and global convergence). Consider $\dot{k}=s{k}^{\alpha }-(\delta +n+g)k$ with $0<\alpha <1$ and $s,\delta ,n,g>0$. There is a unique positive steady state $k^{\ast }=(s/(\delta +n+g){)}^{1/(1-\alpha )}$, and from any initial $k_0>0$ the trajectory converges to $k^{\ast }$. The local rate of convergence is $(1-\alpha )(\delta +n+g)$, and at the steady state output per worker grows at rate $g$ while total output grows at rate $n+g$.

Proof. The steady-state equation $s{k}^{\alpha }=(\delta +n+g)k$ rearranges to $k^{1-\alpha }=s/(\delta +n+g)$, giving the displayed $k^{\ast }$ as the unique positive solution since $1-\alpha >0$. For stability, examine the growth rate of $k$,

$$\frac{\dot{k}}{k}=s{k}^{\alpha -1}-(\delta +n+g).$$

Because $\alpha -1<0$, the map $k\mapsto s{k}^{\alpha -1}$ is strictly decreasing from $+\infty$ as $k\to 0^{+}$ to $0$ as $k\to \infty$, so it crosses the constant $\delta +n+g$ exactly once, at $k^{\ast }$. Below the crossing $\dot{k}>0$, above it $\dot{k}<0$, so the vector field points toward $k^{\ast }$ on both sides and every positive trajectory converges to $k^{\ast }$. Linearising,

$${\frac{\partial \dot{k}}{\partial k}|}_{k^{\ast }}=s\alpha (k^{\ast }{)}^{\alpha -1}-(\delta +n+g)=-(1-\alpha )(\delta +n+g)<0,$$

using the steady-state identity $s(k^{\ast }{)}^{\alpha -1}=\delta +n+g$; hence the local convergence rate is $(1-\alpha )(\delta +n+g)$. At $k^{\ast }$, per-effective-worker variables are constant, so $Y/L=A(k^{\ast }{)}^{\alpha }$ grows at rate $g$ and $Y=AL(k^{\ast }{)}^{\alpha }$ grows at rate $n+g$. $\square$

Absolute versus conditional convergence. The model predicts absolute convergence — poor economies catching up to rich ones regardless of initial conditions — only under the assumption that economies share the same fundamentals $(s,\delta ,n,g)$ and therefore the same steady state. Across heterogeneous economies the prediction is conditional convergence: each economy converges to its own $k^{\ast }$, so the regression evidence should detect catch-up only after controlling for the determinants of the steady state. Barro and Sala-i-Martin's cross-regional estimates find a convergence rate near $2\%$ per year, close to the model's prediction $(1-\alpha )(\delta +n+g)$ once human capital is admitted.

The Mankiw-Romer-Weil augmentation. The unaugmented Solow model with $\alpha \approx 0.3$ predicts far less cross-country income dispersion than the data show. Mankiw, Romer, and Weil restored the fit by adding human capital: with $\alpha \approx 0.3$ and $\varphi \approx 0.28$, the implied total-capital share $\alpha +\varphi \approx 0.58$ matches the international distribution of income while preserving the convergence property, because the production function remains strictly concave in the broadened capital inputs.

Bridge. This convergence result builds toward 44.08.03, whose fixed-point and linearisation analysis provides the dynamical-systems machinery the proof invokes, and appears again in 52.03.01, where the conditional-convergence prediction is tested in cross-country growth regressions. The foundational reason the steady state organises long-run outcomes is that it separates the level of prosperity, pinned down by saving, depreciation, population, and technology, from the growth rate, pinned down by technology alone; this is exactly the structure every modern growth model reuses, and putting these together, the bridge is that augmented Solow, Ramsey, and endogenous-growth models are each a law of motion plus a steady-state characterisation, a pattern the central insight of this unit carries forward when technology is made endogenous.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none. The Solow model and its endogenous-growth descendants are calibrated economic objects whose correctness gate is internal consistency and econometric fit, not formal proof. The convergence theorem is an application of one-dimensional dynamical-systems theory — fixed points, concavity, linearisation, stability — covered in 44.08.03, and the welfare and golden-rule arguments lean on the constrained-optimisation machinery of 44.02.01. Those mathematical layers are the natural targets for formalisation; Mathlib has no Cobb-Douglas production type, no growth-accounting infrastructure, and no formalisation of the Ramsey or R&D idea-production problems.

Advanced results Master

Golden rule of capital accumulation. The welfare question within Solow is which steady state a benevolent planner would choose. Because steady-state consumption is $c^{\ast }=f(k^{\ast })-(\delta +n+g)k^{\ast }$, the optimum solves $f^{\prime }(k_{GR})=\delta +n+g$, equating the marginal product of capital to the break-even rate. Steady states with $s>s_{GR}$ are dynamically inefficient: they hold more capital than is needed to maximise consumption, and every generation could be made better off by saving less. For Cobb-Douglas the clean result $s_{GR}=\alpha$ gives a sharp empirical benchmark against which measured saving rates can be read.

The Mankiw-Romer-Weil augmentation. The striking empirical finding of Mankiw, Romer, and Weil (1992) is that the Solow model fits the cross-country data well once human capital is admitted as an accumulable factor ^{[Mankiw-Romer-Weil 1992]}. With shares $\alpha \approx 0.3$ and $\varphi \approx 0.28$, the model reproduces the wide dispersion of international incomes while leaving the convergence prediction intact, because the production function remains strictly concave in the broadened capital aggregate. The implication is that cross-country income differences reflect differences in physical and human-capital accumulation more than differences in technology levels.

Absolute versus conditional convergence. The convergence rate $(1-\alpha )(\delta +n+g)$ gives roughly $6\%$ per year at the unaugmented calibration, faster than the $2\%$ found empirically; augmenting with human capital and admitting measurement error reconciles the gap. Barro and Sala-i-Martin's cross-regional evidence — the US states, the Japanese prefectures, the European regions — shows steady conditional convergence at about $2\%$ per year, the regularity that makes the Solow model the empirical benchmark for long-run growth.

The AK model. The simplest departure from Solow replaces diminishing returns with a linear technology $Y=AK$. The growth rate becomes $g=sA-\delta -n$, a function of the saving rate, so the model predicts no convergence and persistent effects of policy on growth. The AK framework captures the intuition that if broad capital (physical plus human plus knowledge) accumulates without diminishing returns, growth is endogenous — but it does so by assumption rather than by modelling the source of the non-rivalry.

R&D-driven endogenous growth. Romer (1990) endogenises the growth rate by modelling research as a deliberate activity that produces non-rival ideas ^{[Romer 1990]}. With an idea-production function $\dot{A}=\theta L_{A}A$ — where $L_A$ is labour allocated to research — the technology growth rate is $\theta L_A$, and the long-run growth rate depends on the size of the research sector, the interest rate, and the strength of intellectual-property protection. Growth becomes a choice variable inside the model.

Scale effects and policy. The Romer model predicts that larger economies (more researchers) grow faster — a scale effect that the data do not cleanly support, motivating the Schumpeterian quality-ladder models of Aghion and Howitt (1992), where growth depends on the research intensity rather than the research level. Across these frameworks, policy towards saving, education, and research shifts the long-run growth rate itself, not merely the income level.

Synthesis. The foundational reason the Solow model became the backbone of growth theory is that one law of motion plus a concave production function delivers existence, uniqueness, and stability of the steady state at once; this is exactly the structure that pins long-run income to saving, depreciation, population, and technology. Putting these together with the Mankiw-Romer-Weil human-capital extension, the central insight is that broadening capital deepens the steady state without overturning convergence. The bridge is the move from diminishing to constant returns: once knowledge enters as a non-rival input, concavity gives way and the growth rate becomes a choice variable driven by incentives to innovate. The pattern generalises from the AK model to Romer's R&D-driven growth and the Schumpeterian quality ladder, each of which makes policy toward saving, education, or research shift the long-run growth rate rather than only the income level.

Full proof set Master

Proposition (Golden-rule saving rate for Cobb-Douglas). In the Solow model with $f(k)=k^{\alpha }$, $0<\alpha <1$, the golden-rule saving rate equals $\alpha$, and the economy is dynamically efficient if and only if $s\le \alpha$.

Proof. Steady-state consumption per effective worker is $c^{\ast }=f(k^{\ast })-(\delta +n+g)k^{\ast }$, where $k^{\ast }$ depends monotonically on $s$. Maximising over $k^{\ast }$ gives $f^{\prime }(k_{GR})=\delta +n+g$, i.e. $\alpha k_{GR}^{\alpha -1}=\delta +n+g$. The saving rate that implements any steady state $k^{\ast }$ satisfies $s=(\delta +n+g)k^{\ast }/f(k^{\ast })=(\delta +n+g)(k^{\ast }{)}^{1-\alpha }$. Evaluating at $k_{GR}$ and substituting $\delta +n+g=\alpha k_{GR}^{\alpha -1}$ gives $s_{GR}=\alpha k_{GR}^{\alpha -1}\cdot k_{GR}^{1-\alpha }=\alpha$. Because $c^{\ast }$ is strictly concave in $k^{\ast }$ (since $f^{\prime \prime }<0$), consumption is maximised at $k_{GR}$ and falls for $k^{\ast }>k_{GR}$; equivalently $s>s_{GR}$ is dynamically inefficient. $\square$

Proposition (Mankiw-Romer-Weil steady state). In the augmented Solow model with $y=k^{\alpha }{h}^{\varphi }$, $\alpha +\varphi <1$, and accumulation $\dot{k}=s_{K}y-(\delta +n+g)k$, $\dot{h}=s_{H}y-(\delta +n+g)h$, the steady-state income level is

$$y^{\ast }={\left[\frac{s_K^{\alpha }{s}_H^{\varphi }}{(\delta +n+g{)}^{\alpha +\varphi }}\right]}^{1/(1-\alpha -\varphi )}.$$

Proof. At the steady state $\dot{k}=\dot{h}=0$, so $k^{\ast }=s_K{y}^{\ast }/(\delta +n+g)$ and $h^{\ast }=s_H{y}^{\ast }/(\delta +n+g)$. Substituting into $y^{\ast }=(k^{\ast }{)}^{\alpha }(h^{\ast }{)}^{\varphi }$,

$$y^{\ast }={\left(\frac{s_K{y}^{\ast }}{\delta +n+g}\right)}^{\alpha }{\left(\frac{s_H{y}^{\ast }}{\delta +n+g}\right)}^{\varphi }=\frac{s_K^{\alpha }{s}_H^{\varphi }}{(\delta +n+g{)}^{\alpha +\varphi }}(y^{\ast }{)}^{\alpha +\varphi }.$$

Dividing by $(y^{\ast }{)}^{\alpha +\varphi }$ gives $(y^{\ast }{)}^{1-\alpha -\varphi }=s_K^{\alpha }{s}_H^{\varphi }/(\delta +n+g{)}^{\alpha +\varphi }$, and raising to the power $1/(1-\alpha -\varphi )$ yields the displayed expression. Taking logs,

$$\ln y^{\ast }=\frac{\alpha }{1-\alpha -\varphi }\ln s_K+\frac{\varphi }{1-\alpha -\varphi }\ln s_H-\frac{\alpha +\varphi }{1-\alpha -\varphi }\ln (\delta +n+g),$$

the regression specification that Mankiw, Romer, and Weil estimate across countries. $\square$

Proposition (AK growth rate). In the AK model $Y=AK$ with $\dot{K}=sY-\delta K$ and labour growth $n$, output per worker grows at the constant rate $g=sA-\delta -n$ (when positive), and there is no convergence.

Proof. Writing $k=K/L$ with $\dot{L}/L=n$, the per-worker accumulation is $\dot{k}=sAk-(\delta +n)k$. Dividing by $k$ gives $\dot{k}/k=sA-\delta -n$, which is constant and independent of the level of $k$. Since $y=Ak$, output per worker grows at the same rate. Two economies with identical $s$, $A$, $\delta$, $n$ but different initial $k_0$ grow at the same rate forever, so the ratio $k_1/k_2$ is constant: there is no catch-up. $\square$

Connections Master

• Macroeconomics — aggregates, growth, and cycles 52.02.01. This unit deepens the long-run half of the parent survey: where the overview introduced the Solow steady state and the growth/cycle split, here the convergence theorem, golden rule, and endogenous-growth extensions carry the model to its full depth. The short-run cycle material (IS-LM, AD-AS, Phillips) belongs to the sibling and is deliberately not duplicated.

• Microeconomics 52.01.01. The endogenous-growth models replace the exogenous saving rate with household and firm optimisation — the Ramsey problem of intertemporal utility maximisation and the firm's R&D investment decision — so the constrained-optimisation and equilibrium tools of microeconomics are the load-bearing machinery for the micro-founded descendants of Solow.

• Optimisation 44.02.01 and dynamic systems 44.08.03. The golden-rule argument is a constrained-optimisation problem, the convergence proof is a one-dimensional dynamical-systems result (fixed point, concavity, linearisation, stability), and the Ramsey replacement of the exogenous saving rate is a Bellman-equation dynamic programme. These two units supply the formal machinery on which the growth-model proofs stand.

• Econometrics 52.03.01. The conditional-convergence prediction and the Mankiw-Romer-Weil income regression are tested and quantified with econometric methods: cross-country growth regressions, instrumental variables for the saving and human-capital rates, and panel estimates of the convergence coefficient. Identification of causal effects of policy on growth is the central open problem where growth theory meets econometrics.

Historical & philosophical context Master

Long-run growth theory was formalised independently in 1956 by Robert Solow ^{[Solow 1956]} and Trevor Swan ^{[Swan 1956]}. Solow's paper, published in the Quarterly Journal of Economics, replaced the rigid fixed-proportions (Harrod-Domar) model with a smooth neoclassical production function and showed that capital accumulation alone could not sustain per-worker growth — that the residual, later named total factor productivity, must do the work. Swan's contribution, published the same year in the Economic Record, reached the same steady-state and convergence conclusions through an Australian pedagogical lens; the model is now universally known as Solow-Swan. Solow's 1957 follow-up estimated that the bulk of US historical growth came from the unexplained residual of technical progress, founding modern growth accounting.

The convergence prediction was tested intensively in the late 1980s. Barro and Sala-i-Martin documented conditional convergence at roughly $2\%$ per year across regions, while Mankiw, Romer, and Weil showed in 1992 that augmenting Solow with human capital restored the model's empirical fit ^{[Mankiw-Romer-Weil 1992]}. Their estimated shares, $\alpha \approx 0.3$ and $\varphi \approx 0.28$, implied a total-capital share near $0.6$, rescuing the neoclassical model from the charge that it could not explain cross-country income dispersion.

The endogenous-growth turn opened by Paul Romer's 1990 paper modelled technology as a non-rival input produced by deliberate research effort ^{[Romer 1990]}, making the long-run growth rate a function of incentives rather than an exogenous datum. Aghion and Howitt's 1992 Schumpeterian model recast growth as a sequence of quality improvements, each displacing its predecessor through creative destruction. Barro and Sala-i-Martin's Economic Growth synthesised the neoclassical and endogenous programmes ^{[Barro-Sala-i-Martin 2003]} and remains the canonical graduate reference.

The standing philosophical tension is whether long-run growth is best understood as an equilibrium outcome of accumulating broad capital — the neoclassical view, in which policy affects the level — or as an inherently increasing-returns process driven by the production of knowledge — the endogenous view, in which policy affects the growth rate. The data have not fully adjudicated between them; the difference turns on whether measured total factor productivity reflects genuine non-rivalry or unmeasured factor accumulation, a question that remains open.

Bibliography Master

1. Solow, R. M. (1956). A Contribution to the Theory of Economic Growth. Quarterly Journal of Economics 70(1), 65–94.
2. Swan, T. W. (1956). Economic Growth and Capital Accumulation. Economic Record 32(2), 334–361.
3. Mankiw, N. G., Romer, D., & Weil, D. N. (1992). A Contribution to the Empirics of Economic Growth. Quarterly Journal of Economics 107(2), 407–437.
4. Romer, P. M. (1990). Endogenous Technological Change. Journal of Political Economy 98(5, Part 2), S71–S102.
5. Barro, R. J., & Sala-i-Martin, X. (2003). Economic Growth (2nd ed.). Cambridge, MA: MIT Press.
6. Romer, D. (2018). Advanced Macroeconomics (5th ed.). New York: McGraw-Hill.
7. Aghion, P., & Howitt, P. (1992). A Model of Growth Through Creative Destruction. Econometrica 60(2), 323–351.
8. Solow, R. M. (1957). Technical Change and the Aggregate Production Function. Review of Economics and Statistics 39(3), 312–320.