In most ores the valuable mineral occupies a small fraction of the particles. When the particles are coarse, even a heavy sample contains relatively few grains of the mineral, and the grade varies widely between samples taken from the same material. The error is not a consequence of careless sampling but of the number of particles: too few grains of a rare mineral are present for the sample grade to be reliable. Crushing the ore finer produces many more particles in the same mass, so that a smaller sample becomes representative. Gy's formula quantifies this relationship, giving the sample mass required at a given particle size to achieve a given precision.
Sampling error is often the largest error in a metallurgical figure, larger than the assay error, and it cannot be corrected afterwards by better analysis. Pierre Gy's theory of sampling provides the means to design sampling and preparation so that the error is known and acceptable. This article covers the formula, its parameters, a worked protocol and the errors outside the scope of the formula.
Gy's formula
where
- relative standard deviation of the fundamental sampling error (fraction; 0.05 means ±5%)
- ,
- mass of the sample and of the lot it is taken from (g); the second form applies when the lot is much larger
- shape factor (about 0.5 for most broken ore)
- size range factor (about 0.25 for unsized broken ore)
- mineralogical factor (g/cm³), from the grade and the densities
- liberation factor (between 0 and 1)
- top size, the size that 95% passes (cm)
The parameters describe the material:
where
- mass fraction of the valuable mineral in the ore (not the metal grade)
- ,
- density of the valuable mineral and of the gangue (g/cm³)
- liberation size of the valuable mineral (cm)
- exponent, commonly taken as 0.5 (calibrated for the ore where possible)
The shape factor f is about 0.5 for most broken ore; it is lower for flaky minerals such as mica and much lower for gold flakes. The size range factor g is about 0.25 for unsized material and higher for closely sized material. The mineralogical factor c is large when the mineral is rare and dense, because a single grain then carries a large amount of metal.
The liberation factor l is 1 when the mineral is fully liberated (particles finer than the liberation size) and falls for coarser particles, in which most grains are composites of mineral and gangue. An exponent b of 0.5 is a common default, but François-Bongarçon and Gy showed that it should be calibrated for each ore where possible, because an incorrect form of the liberation factor is the most common error in applying the formula.
Worked example for a copper ore
A copper ore grades 1.0% Cu, carried by chalcopyrite (34.6% Cu, density 4.2 g/cm³) in gangue of density 2.7 g/cm³, and chalcopyrite liberates at about 150 µm (0.015 cm). The mass fraction of chalcopyrite is a = 1.0 ÷ 34.6 = 0.0289, and the mineralogical factor is c = (0.971 ÷ 0.0289) × (0.971 × 4.2 + 0.0289 × 2.7) = 140 g/cm³.
At a top size of 1 cm, l = (0.015 ÷ 1)0.5 = 0.122. For a relative standard deviation of 5% (σ = 0.05), MS = 0.5 × 0.25 × 140 × 0.122 × 1³ ÷ 0.05² = 855 g.
Crushed to 2 mm (0.2 cm), the liberation factor rises to 0.27 but d³ falls 125-fold, and the minimum mass for the same precision falls to about 15 g. Pulverised to 75 µm, it is a few milligrams. Particle size is the dominant variable.
Sample preparation protocol
A sample is rarely taken and assayed in a single step. It is reduced in stages: crushed, split, ground, split, pulverised, and an aliquot taken. Each split adds its own fundamental error, and the variances (the squares of the relative standard deviations) are additive:
σtotal² = σ1² + σ2² + σ3² + …
| Stage | Top size | Mass taken from | Mass kept | Minimum mass for ±5% | Relative SD of this stage |
|---|---|---|---|---|---|
| Primary sample from the lot | 25 mm | large lot | 25 kg | 8.5 kg | 2.9% |
| Crush, split | 10 mm | 25 kg | 2 kg | 855 g | 3.1% |
| Grind, split | 2 mm | 2 kg | 250 g | 15 g | 1.2% |
| Pulverise, weigh aliquot | 75 µm | 250 g | 0.5 g | < 0.01 g | 0.4% |
| Total | 4.4% |
The relative standard deviation of each stage is the square root of f·g·c·l·d³ × (1/MS − 1/ML). For the second stage, 0.5 × 0.25 × 140 × 0.122 × 1 × (1/2,000 − 1/25,000) = 0.00098, a relative SD of 3.1%. The total, √(0.0291² + 0.0314² + 0.0116² + 0.0038²) = 4.4%, meets the ±5% target. The two coarse stages carry almost all of the error. Increasing the analytical aliquot would achieve almost nothing, whereas crushing finer before the first split would have the greatest effect.
The nomogram presents the same protocol graphically. Each vertical line is a size reduction at constant mass, and each downward line is a split at constant size. The protocol is satisfactory while every point remains above the line for the target precision.
Gold ores
Gold is rare and very dense (19.3 g/cm³), so c is very large, and gold often occurs as particles that do not break during crushing. A single coarse gold grain can contain more gold than the remainder of the sample. For gold ores, the calculation should be based on the size of the gold particles themselves, measured by a gravity or screened-metallic test, rather than on an assumed liberation size, and much larger sample masses should be expected. Screen fire assays, larger fire assay charges or whole-sample leaching are used when coarse gold is present. The measurement of the gold size distribution is covered in gravity recoverable gold.
Sampling errors outside the formula
Gy's formula covers only the fundamental error, which remains when sampling is carried out correctly. The other errors in Gy's theory arise from incorrect practice, and they are often larger. Delimitation and extraction errors arise from a cutter that does not cross the whole stream, is too narrow for the largest particles, or overflows. Grouping and segregation errors arise from taking a few large increments instead of many small ones from a segregated pile or stream.
Preparation errors include loss of fines as dust, contamination from the crusher or pulveriser, and oxidation of sulphides. Analytical error arises from the assay itself.
These errors are controlled by correct sampler design (cross-stream cutters, cutter openings at least three times the top size, and constant cutter speed), by taking many increments, and by quality control with duplicates at each stage.
Application
- Specify the required precision, for example ±5% relative at one standard deviation for a plant feed sample used in metal accounting.
- Estimate the parameters from mineralogy. Where liberation data are not available, a heterogeneity test, in which many samples of the same mass are assayed, calibrates the constant directly.
- Check each splitting stage, and modify the protocol where the error is concentrated.
- Verify the protocol with duplicate samples at each stage. The scatter between duplicates measures the actual error, including the components that the formula omits.
Glossary
- Fundamental sampling error
- The random error caused by the particle-to-particle variation in grade, present even when sampling is done correctly.
- Lot
- The whole body of material being sampled, such as a stockpile, a shift's plant feed or a drill hole interval.
- Top size, d
- The screen size that 95% of the material passes.
- Liberation size
- The particle size below which the valuable mineral exists mostly as separate grains.
- Split
- Dividing a sample into a smaller, representative part with a riffle splitter or rotary divider.
- Aliquot
- The small portion of a pulverised sample weighed out for assay.
- Relative standard deviation
- The standard deviation divided by the mean, often given in per cent.
- Heterogeneity test
- An experiment that assays many equal-mass samples of the same material to measure its sampling constant directly.
Frequently asked questions
What does Gy's formula calculate?
The variance of the fundamental sampling error for a given sample mass, particle top size and material. Rearranged, it gives the minimum sample mass for a chosen precision.
Why does particle size matter so much in sampling?
The error depends on the cube of the top size. Halving the particle size reduces the error variance about eightfold, partly offset by a lower liberation factor, which is why samples are crushed before they are split.
What values of f and g should be used?
About 0.5 for f and 0.25 for g suit most unsized broken ore. Gold flakes and closely sized material require different values.
Is Gy's formula valid for gold ores?
Yes, but the parameters must reflect the gold particle size, which is often much coarser than an assumed liberation size. Coarse gold ores require much larger samples, and a heterogeneity test is the best calibration.
Discuss your plant with a metallurgist
Tiegel is a Perth metallurgy company. We help with testwork, plant optimisation, metal accounting and investigations, and we build the Tiegel Suite software.
Talk to a metallurgistReferences and further reading
- Gy, P.M. (1979). Sampling of Particulate Materials: Theory and Practice. Elsevier.
- Pitard, F.F. (1993). Pierre Gy's Sampling Theory and Sampling Practice, 2nd edition. CRC Press.
- François-Bongarçon, D. and Gy, P. (2002). The most common error in applying 'Gy's Formula' in the theory of mineral sampling, and the history of the liberation factor. Journal of the South African Institute of Mining and Metallurgy, 102. Link
Worked examples and charts marked illustrative use invented figures, not data from any operation.