Arps Decline Curve Analysis — a practical guide

● Production Engineering · May 17, 2026 · 10 min read

Arps decline curve analysis has been the workhorse of production forecasting since 1945. Every reservoir engineer has fit one. Most have over-trusted at least one. This guide is about the practical decisions — which form to use, what window to fit, and how to read the b-factor without fooling yourself.

What Arps actually solved

In 1945, Jan Arps published three empirical equations that fit nearly every conventional oil and gas well decline anyone had observed. The equations were not derived from reservoir physics — they were fit to data. Decades later, when researchers showed which reservoir flow regimes produce which Arps form, the equations gained physical legitimacy. But the original insight remains: most wells, most of the time, behave like one of three curves.

The three Arps forms differ only in one parameter: b. Set b = 0 and you get exponential decline. Set b = 1 and you get harmonic decline. Anywhere in between gives hyperbolic decline. That's the whole framework. The hard part is choosing b correctly and knowing when the framework breaks.

The three equations

All three forms describe how production rate changes with time:

General hyperbolic form q(t) = q_i / (1 + b × D_i × t)^(1/b)

where:
  q_i = initial rate at t = 0
  D_i = initial nominal decline rate (1/time)
  b = decline exponent (0 to 1, sometimes higher)
  t = time since q_i

The two limiting cases simplify nicely. Exponential decline (b = 0) is what you get when you apply L'Hôpital's rule:

Exponential — b = 0 q(t) = q_i × exp(−D × t)

Decline rate D is constant over time. Each year, production drops by the same fraction.

Harmonic decline (b = 1) is the other extreme:

Harmonic — b = 1 q(t) = q_i / (1 + D_i × t)

Effective decline rate drops continuously. The well "decelerates" — it produces longer than exponential predicts.

Hyperbolic decline (0 < b < 1) is everything between. In practice, most real wells fit somewhere in this band.

Figure 1Three decline curves starting from the same qi = 1000 BOPD with Di = 0.20/yr. On a semilog plot, exponential is a straight line. Hyperbolic and harmonic curve upward — slower decline at late times. By year 10, harmonic gives roughly 2.5× the residual rate of exponential. Most of the EUR difference between methods comes from this late-time divergence.

How to read the b-factor

The b-factor is more than a fitting parameter — it carries physical meaning:

b-factor	Reservoir behavior	Typical case
b = 0	Single-phase liquid, constant productivity	Undersaturated oil, late-life depletion
b = 0.2 – 0.4	Solution-gas drive	Most conventional oil reservoirs
b = 0.4 – 0.6	Gas reservoir, single-layer	Conventional dry gas
b = 0.6 – 0.9	Layered or stratified reservoir	Multi-layer gas, commingled
b = 1.0	Strong water drive, gravity dominated	Edge water drive, gas cap expansion
b > 1.0	Transient flow, unconventional	Tight gas, shale — early time only

The b-factor is reservoir physics in disguise. A b-fit far from your expected range is a signal — usually that the wrong decline regime is being fit.

Which form to use, and when

The choice is not arbitrary. Each form fits a specific reservoir condition well — and poorly outside it.

Use exponential (b = 0) when:

The well is in late-life pseudosteady-state decline
Reservoir is single-layer with constant productivity index
You need a conservative estimate (exponential under-predicts most reservoirs)
Bank-grade reserves estimates that require defensibility over flexibility

Use hyperbolic (0 < b < 1) when:

Multi-phase flow is occurring (oil with gas, gas with condensate)
Reservoir has layered behavior or natural fractures
You have enough data to fit b reliably (≥ 12-18 months of stable decline)
Conventional gas and most conventional oil wells

Use harmonic (b = 1) when:

Strong water-drive or gas-cap expansion confirmed
You see the actual production curve "flattening" beyond hyperbolic
Combination drive reservoirs in middle-life

The window selection problem

Where you start your fit matters more than which equation you choose. The same well can yield three completely different forecasts depending on which 6 months of historical data is fit. This is the most common source of bad reserves estimates.

Rule of thumb: skip the transient

Conventional wells often produce above expectation for the first 3-6 months due to transient flow effects. Fitting Arps to this early-time data over-predicts long-term rate. Wait until production stabilizes into pseudosteady-state, then fit. For most conventional wells, this means using data from month 6 onward.

Recommended workflow

Plot rate vs time on semi-log paper. If the data plots as a straight line, exponential decline is the right form. If it curves, hyperbolic or harmonic.
Identify the transient period. Discard early-time data where rate declines abnormally fast (often the first 3-6 months).
Fit the stable region. Use the past 12-24 months of consistent decline as the fit window.
Check the residual. If R² < 0.85, the fit is unreliable — try a different window or form.
Sanity-check b. Compare against the reservoir-physics table above. A b-factor far from expected range means something is wrong with the fit, not with reservoir.

Estimating EUR — and where it goes wrong

Estimated ultimate recovery (EUR) requires integrating the decline curve from now to the economic limit rate (q_eco). For exponential:

EUR — exponential decline EUR = N_p + (q_i − q_eco) / D

where N_p is cumulative production to date and q_eco is economic limit.

For hyperbolic, the integration is messier but tractable:

EUR — hyperbolic decline EUR = N_p + q_i^b × (q_i^1−b − q_eco^1−b) / ((1−b) × D_i)

Valid for b ≠ 1.

For harmonic (b = 1), EUR formally diverges — the integral of harmonic decline is infinite. This is why pure harmonic fits are dangerous: they imply infinite recoverable reserves. Practical EUR for harmonic always requires a hard economic limit.

The terminal decline correction

Hyperbolic decline is unrealistic over long horizons. A well fitting b = 0.7 today cannot physically maintain that decline behavior forever — eventually it transitions to exponential as the reservoir depletes. Ignoring this transition leads to systematically inflated EUR.

Standard practice: apply a terminal decline switch. When effective decline rate D_e drops below a threshold (typically 5-10% per year), switch from hyperbolic to exponential at that level. The forecast becomes piecewise:

Modified hyperbolic with terminal switch Hyperbolic until D_e(t) ≤ D_terminal
Exponential with D = D_terminal thereafter

This is what most reserves software actually computes — even when the user thinks they're running pure hyperbolic. It's the difference between a defensible reserves number and a fantasy.

Figure 2The terminal decline correction. Pure hyperbolic forecast (red dashed) keeps decelerating and never reaches abandonment rate — EUR grows without bound. Switching to exponential at the minimum economic decline rate (green solid) caps the late-life rate and produces a defensible EUR. The gap between the two curves at year 15-20 is where the optimism lives.

Common pitfalls

1. Fitting too short a window. Anything less than 12 months of stable decline is noise-fitting. Be honest about whether you have enough data.

2. Including transient data. Early-life production above the long-term trend will pull the fit upward and make the well look better than it is.

3. Trusting b > 1. Outside unconventional wells in early time, b values above 1 indicate a fitting problem, not a physical reality. Cap b at 1 unless you have strong reason otherwise.

4. Ignoring intermittent production. Wells that cycle on/off, or have long shut-in periods within the fit window, give garbage fits. Either clean the data or pick a window where production is continuous.

5. Forgetting the economic limit. EUR without q_eco is meaningless. Always specify the rate at which the well stops paying for itself.

6. Single-form religion. Some engineers always use exponential ("conservative"), others always use hyperbolic ("realistic"). Both are wrong — fit the form the data supports, not the form you prefer.

Field practice: building the workflow

A practical DCA workflow for a mature field looks like this:

Per-well screening: Plot every producing well's monthly rate vs time. Flag wells with stable decline patterns (good DCA candidates) versus erratic wells (need cleanup first).
Fit window selection: Identify the last 18-24 months of clean decline for each well.
Multi-b fit: Fit Arps with multiple b values (0, 0.2, 0.4, 0.6, 0.8, 1.0) and pick the b that gives best R² without exceeding physical bounds.
Apply terminal decline: Switch hyperbolic fits to exponential when D_e drops below 6-10% annual.
Forecast to q_eco: Integrate forward to estimate remaining EUR per well.
Aggregate to field level: Sum per-well EUR with appropriate uncertainty bands.

When DCA stops working

Arps is empirical. It works when reservoir behavior is consistent over the fit window. It breaks when the well changes behavior — and most mature wells do, eventually. Watch for:

Liquid loading in gas wells (Coleman-Turner critical rate crossed)
Water breakthrough in oil wells (water cut rising sharply)
Workover or recompletion events (different completion = different well)
Reservoir pressure support changes (waterflood started/stopped)
Reservoir compartmentalization (fault or barrier encountered)

Once any of these occurs, the prior Arps fit no longer applies. Refit from the event date forward, or switch to a different forecasting framework entirely (material balance, physics-based, machine learning).

Three takeaways

The b-factor is reservoir physics. Read it for what it tells you about flow behavior, not just as a fitting knob.
Window selection beats form selection. The wrong window with the right equation is worse than a defensible window with a slightly less-optimal equation.
Always apply terminal decline. Pure hyperbolic forecasts to infinity will eventually embarrass whoever signed off on the reserves.

References & further reading:
Arps, J. J. (1945). Analysis of Decline Curves. Trans. AIME, 160, 228–247.
Fetkovich, M. J. (1980). Decline Curve Analysis Using Type Curves. JPT, 32(6), 1065–1077.
Ilk, D., Rushing, J. A., Perego, A. D., Blasingame, T. A. (2008). Exponential vs. Hyperbolic Decline in Tight Gas Sands. SPE 116731.
Robertson, S. (1988). Generalized Hyperbolic Equation. SPE 18731.
Cronquist, C. (2001). Estimation and Classification of Reserves of Crude Oil, Natural Gas, and Condensate. SPE.
Poston, S. W., Poe, B. D. (2008). Analysis of Production Decline Curves. SPE.

Try Arps DCA on your wells

Fit decline curves with multi-b auto-selection

RFour Energy ships free tools for production engineers: a downloadable Excel template with multi-b Arps fitting and terminal decline switch, plus GOWIS — a live dashboard that auto-fits decline curves on every well and forecasts EUR with engineering-judgment manual override.

Open GOWIS demo → Get the Excel tool